Introduction
We study the advection–diffusion equation
{ ∂ t θ + b ⋅ ∇ θ − κ Δ θ = 0 in ( t i n , ∞ ) × T 2 , θ ( t i n , ⋅ ) = θ i n on T 2 (1) \begin{cases}
\partial_t\theta+\mathbf{b}\cdot\nabla\theta-\kappa\Delta\theta=0 & \text{in }(t_{\mathrm{in}},\infty)\times\mathbb{T}^2,\\
\theta(t_{\mathrm{in}},\cdot)=\theta_{\mathrm{in}} & \text{on }\mathbb{T}^2 \tag*{(1)}
\end{cases} { ∂ t θ + b ⋅ ∇ θ − κ Δ θ = 0 θ ( t in , ⋅ ) = θ in in ( t in , ∞ ) × T 2 , on T 2 (1)
on the torus T 2 : = R 2 / Z 2 \mathbb{T}^2:=\mathbb{R}^2/\mathbb{Z}^2 T 2 := R 2 / Z 2 , for a divergence-free velocity field b \mathbf{b} b , a diffusivity κ > 0 \kappa>0 κ > 0 , an initial time t i n ∈ R t_{\mathrm{in}}\in\mathbb{R} t in ∈ R and an initial datum θ i n ∈ L 2 ( T 2 ) \theta_{\mathrm{in}}\in L^2(\mathbb{T}^2) θ in ∈ L 2 ( T 2 ) . The solution θ \theta θ is a passive scalar, carried by b \mathbf{b} b and diffused with the molecular diffusivity κ \kappa κ . Since ∇ ⋅ b = 0 \nabla\cdot\mathbf{b}=0 ∇ ⋅ b = 0 , the transport term conserves the L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) norm, and diffusion alone decreases it: for every t > t i n t>t_{\mathrm{in}} t > t in ,
∥ θ ( t , ⋅ ) ∥ L 2 ( T 2 ) 2 + 2 κ ∥ ∇ θ ∥ L 2 ( ( t i n , t ) × T 2 ) 2 = ∥ θ i n ∥ L 2 ( T 2 ) 2 . \lVert\theta(t,\cdot)\rVert_{L^2(\mathbb{T}^2)}^2+2\kappa\lVert\nabla\theta\rVert_{L^2((t_{\mathrm{in}},t)\times\mathbb{T}^2)}^2=\lVert\theta_{\mathrm{in}}\rVert_{L^2(\mathbb{T}^2)}^2. ∥ θ ( t , ⋅ ) ∥ L 2 ( T 2 ) 2 + 2 κ ∥ ∇ θ ∥ L 2 (( t in , t ) × T 2 ) 2 = ∥ θ in ∥ L 2 ( T 2 ) 2 .
We are interested in small κ \kappa κ and in velocity fields which, like the velocity of a turbulent fluid, are continuous but rough: b \mathbf{b} b is Hölder continuous with an exponent α < 1 \alpha<1 α < 1 . The question is how regular the solutions are, uniformly in κ \kappa κ .
For the velocity of a turbulent fluid, whose increments over a distance scale like the power 1 / 3 1/3 1/3 of the distance in Kolmogorov’s theory [40 ] , Obukhov [44 ] and Corrsin [18 ] predicted by a scaling argument that the increments of the scalar scale like the same power 1 / 3 1/3 1/3 , over a range of distances that depends on κ \kappa κ . The same argument predicts the exponent 1 2 ( 1 − α ) \frac{1}{2}(1-\alpha) 2 1 ( 1 − α ) when the increments of the velocity scale like the power α \alpha α [5 ] , Section 1.3. We call 1 2 ( 1 − α ) \frac{1}{2}(1-\alpha) 2 1 ( 1 − α ) the Obukhov–Corrsin exponent of a velocity field which is Hölder continuous with exponent α \alpha α .
In [5 ] we constructed, for every α < 1 / 3 \alpha<1/3 α < 1/3 , a multiscale “fractal” velocity field b \mathbf{b} b which is continuous in time with values in C 0 , α ( T 2 ) C^{0,\alpha}(\mathbb{T}^2) C 0 , α ( T 2 ) . For this field, the solutions θ κ \theta^\kappa θ κ of (1) with diffusivity κ \kappa κ , initial time t i n = 0 t_{\mathrm{in}}=0 t in = 0 and mean-zero initial data θ i n ∈ H 1 ( T 2 ) \theta_{\mathrm{in}}\in H^1(\mathbb{T}^2) θ in ∈ H 1 ( T 2 ) exhibit anomalous dissipation,
inf κ ∈ K A V κ ∥ ∇ θ κ ∥ L 2 ( ( 0 , 1 ) × T 2 ) 2 ≥ c ∥ θ i n ∥ L 2 ( T 2 ) 2 . (2) \inf_{\kappa\in\mathcal{K}_{\mathrm{AV}}}\kappa\lVert\nabla\theta^\kappa\rVert_{L^2((0,1)\times\mathbb{T}^2)}^2\geq c\lVert\theta_{\mathrm{in}}\rVert_{L^2(\mathbb{T}^2)}^2. \tag*{(2)} κ ∈ K AV inf κ ∥ ∇ θ κ ∥ L 2 (( 0 , 1 ) × T 2 ) 2 ≥ c ∥ θ in ∥ L 2 ( T 2 ) 2 . (2)
Here K A V \mathcal{K}_{\mathrm{AV}} K AV is a countable union of bands accumulating at 0 0 0 , the permissible diffusivities of [5 (3.43)] , which is the set K \mathcal{K} K of (2.18) for the value of q q q fixed in [5 ] , and the constant c > 0 c>0 c > 0 depends on the ratio ∥ θ i n ∥ L 2 ( T 2 ) / ∥ ∇ θ i n ∥ L 2 ( T 2 ) \lVert\theta_{\mathrm{in}}\rVert_{L^2(\mathbb{T}^2)}/\lVert\nabla\theta_{\mathrm{in}}\rVert_{L^2(\mathbb{T}^2)} ∥ θ in ∥ L 2 ( T 2 ) / ∥ ∇ θ in ∥ L 2 ( T 2 ) [5 Theorem 1.1 and the discussion after it] . In this paper we prove that, for every exponent s s s below 1 2 ( 1 − α ) \frac{1}{2}(1-\alpha) 2 1 ( 1 − α ) and a velocity field b \mathbf{b} b of the same construction depending on s s s , the solutions are uniformly Hölder continuous with exponent s s s at positive times, for every datum in L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) , uniformly in κ \kappa κ in the corresponding union of bands. Consequently, for this velocity field, an estimate of the form (2) holds for every mean-zero datum in L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) and from every initial time, with a constant c c c independent of the datum (Theorem B below).
Main results
Construction of the velocity field. The velocity field of [5 ] is built level by level. Level m ≥ 1 m\geq1 m ≥ 1 has a wavelength ε m \varepsilon_m ε m , and the wavelengths decrease supergeometrically: ε m − 1 / ε m ≥ Λ \varepsilon_{m-1}/\varepsilon_m\geq\Lambda ε m − 1 / ε m ≥ Λ for a large integer Λ \Lambda Λ , and ε m \varepsilon_m ε m is of the order of ε m − 1 q \varepsilon_{m-1}^q ε m − 1 q for m ≥ 2 m\geq2 m ≥ 2 , for a parameter q > 1 q>1 q > 1 . Level m m m adds to the velocity field b m − 1 \mathbf{b}_{m-1} b m − 1 of the coarser levels short pulses of a shear flow of wavelength ε m \varepsilon_m ε m and speed of order ε m α \varepsilon_m^\alpha ε m α , alternating between the two coordinate directions and written in the Lagrangian coordinates of b m − 1 \mathbf{b}_{m-1} b m − 1 . A fixed cutoff function ζ \zeta ζ of time switches the pulses on and off. The limit b \mathbf{b} b of the velocity fields b m \mathbf{b}_m b m is periodic in space and time, divergence free, and Hölder continuous with exponent α \alpha α in space and time. The velocity field of this paper is that of [5 ] with one change: the parameter q q q , which [5 ] fixes in terms of α \alpha α , is here a free parameter, taken close to 1 depending on α \alpha α and on the exponent s s s of the main result (§ 1.1 ).
With β : = 1 + α \beta:= 1 + \alpha β := 1 + α and γ : = ( q − 1 ) β / ( q + 1 ) \gamma:= (q - 1)\beta/(q + 1) γ := ( q − 1 ) β / ( q + 1 ) , the diffusivities that we allow are the permissible diffusivities of [5 (3.43)] for the present q q q : the union K \mathcal{K} K of the bands
K M = [ 1 2 ε M β − γ , 2 ε M β − γ ] , M ≥ 1 , \mathcal{K}_{M} = \left[\tfrac{1}{2}\varepsilon_{M}^{\beta-\gamma}, 2\varepsilon_{M}^{\beta-\gamma}\right], \qquad M \ge1, K M = [ 2 1 ε M β − γ , 2 ε M β − γ ] , M ≥ 1 ,
defined in (20) , which are disjoint and accumulate at 0. The exponent γ \gamma γ corrects β \beta β in the exponent of the effective diffusivities, described in the strategy below; since q ( β − γ ) = β + γ q(\beta-\gamma)=\beta+\gamma q ( β − γ ) = β + γ , a diffusivity in K M \mathcal{K}_{M} K M , M ≥ 2 M \ge2 M ≥ 2 , is of the order of ε M − 1 β + γ \varepsilon_{M-1}^{\beta+\gamma} ε M − 1 β + γ .
Interior regularity. Our main result is Theorem A, stated precisely in § 1.4 . Let α ∈ ( 0 , 1 / 3 ) \alpha\in(0,1/3) α ∈ ( 0 , 1/3 ) and s ∈ ( 0 , 1 2 ( 1 − α ) ) s \in(0,\frac{1}{2}(1-\alpha)) s ∈ ( 0 , 2 1 ( 1 − α )) , let q > 1 q > 1 q > 1 , close to 1 depending on α \alpha α and s s s , and the cutoff function ζ \zeta ζ be as in § 1.1 , and let Λ \Lambda Λ be an integer at least a threshold Λ 0 ( α , s , q , ζ ) ≥ 128 \Lambda_{0}(\alpha,s,q,\zeta) \ge128 Λ 0 ( α , s , q , ζ ) ≥ 128 . Then the velocity field b \mathbf{b} b of § 1.1 , which depends on s s s through q q q , has the following property: for every 0 < t 0 < T 0 < t_{0} < T 0 < t 0 < T there exists C ( α , s , q , ζ , Λ , t 0 , T ) < ∞ C(\alpha,s,q,\zeta,\Lambda,t_{0},T) < \infty C ( α , s , q , ζ , Λ , t 0 , T ) < ∞ such that, for every t i n ∈ R t_{\mathrm{in}} \in\mathbb{R} t in ∈ R , κ ∈ K \kappa\in\mathcal{K} κ ∈ K and θ i n ∈ L 2 ( T 2 ) \theta_{\mathrm{in}} \in L^{2}(\mathbb{T}^{2}) θ in ∈ L 2 ( T 2 ) , every weak solution of (1.1) on ( t i n , t i n + T ) × T 2 (t_{\mathrm{in}},t_{\mathrm{in}}+T) \times\mathbb{T}^{2} ( t in , t in + T ) × T 2 satisfies
∥ θ ∥ L ∞ ( [ t i n + t 0 , t i n + T ] ; C 0 , s ( T 2 ) ) ≤ C ∥ θ i n ∥ L 2 ( T 2 ) , \|\theta\|_{L^{\infty}([t_{\mathrm{in}}+t_{0},t_{\mathrm{in}}+T];C^{0,s}(\mathbb{T}^{2}))} \le C\|\theta_{\mathrm{in}}\|_{L^{2}(\mathbb{T}^{2})}, ∥ θ ∥ L ∞ ([ t in + t 0 , t in + T ] ; C 0 , s ( T 2 )) ≤ C ∥ θ in ∥ L 2 ( T 2 ) ,
where C 0 , s ( T 2 ) C^{0,s}(\mathbb{T}^{2}) C 0 , s ( T 2 ) is the space of Hölder continuous functions with exponent s s s , normed by (22) . Theorem A is the anomalous regularization of the title: the solution operator maps L 2 ( T 2 ) L^{2}(\mathbb{T}^{2}) L 2 ( T 2 ) into C 0 , s ( T 2 ) C^{0,s}(\mathbb{T}^{2}) C 0 , s ( T 2 ) at every positive time, with a bound that does not degenerate as κ → 0 \kappa\to0 κ → 0 in K \mathcal{K} K . The exponent s s s may be taken as close as desired to the Obukhov–Corrsin exponent, but as s s s increases to 1 2 ( 1 − α ) \frac{1}{2}(1-\alpha) 2 1 ( 1 − α ) , the bounds (3) force q q q down to 1.
Uniform anomalous dissipation. Theorem B of § 1.4 follows from Theorem A alone. In the setting of Theorem A, there exists c ( α , s , q , ζ , Λ ) > 0 c(\alpha,s,q,\zeta,\Lambda) > 0 c ( α , s , q , ζ , Λ ) > 0 such that, for every t i n ∈ R t_{\mathrm{in}} \in\mathbb{R} t in ∈ R , κ ∈ K \kappa\in\mathcal{K} κ ∈ K and θ i n ∈ L 2 ( T 2 ) \theta_{\mathrm{in}} \in L^{2}(\mathbb{T}^{2}) θ in ∈ L 2 ( T 2 ) , with ( θ i n ) T 2 (\theta_{\mathrm{in}})_{\mathbb{T}^{2}} ( θ in ) T 2 the mean of θ i n \theta_{\mathrm{in}} θ in , every weak solution of (1.1) satisfies
κ ∥ ∇ θ ∥ L 2 ( ( t i n , t i n + 1 ) × T 2 ) 2 ≥ c ∥ θ i n − ( θ i n ) T 2 ∥ L 2 ( T 2 ) 2 , \kappa\|\nabla\theta\|_{L^{2}((t_{\mathrm{in}},t_{\mathrm{in}}+1)\times\mathbb{T}^{2})}^{2} \ge c\|\theta_{\mathrm{in}}-(\theta_{\mathrm{in}})_{\mathbb{T}^{2}}\|_{L^{2}(\mathbb{T}^{2})}^{2}, κ ∥∇ θ ∥ L 2 (( t in , t in + 1 ) × T 2 ) 2 ≥ c ∥ θ in − ( θ in ) T 2 ∥ L 2 ( T 2 ) 2 ,
and, for every t ≥ t i n t \ge t_{\mathrm{in}} t ≥ t in ,
∥ θ ( t , ⋅ ) − ( θ i n ) T 2 ∥ L 2 ( T 2 ) 2 ≤ 2 exp ( − c ( t − t i n ) ) ∥ θ i n − ( θ i n ) T 2 ∥ L 2 ( T 2 ) 2 . \|\theta(t,\cdot)-(\theta_{\mathrm{in}})_{\mathbb{T}^{2}}\|_{L^{2}(\mathbb{T}^{2})}^{2} \le2\exp(-c(t-t_{\mathrm{in}}))\|\theta_{\mathrm{in}}-(\theta_{\mathrm{in}})_{\mathbb{T}^{2}}\|_{L^{2}(\mathbb{T}^{2})}^{2}. ∥ θ ( t , ⋅ ) − ( θ in ) T 2 ∥ L 2 ( T 2 ) 2 ≤ 2 exp ( − c ( t − t in )) ∥ θ in − ( θ in ) T 2 ∥ L 2 ( T 2 ) 2 .
Over every unit of time, every solution thus dissipates a fixed fraction of its variance. The proof is short: by Theorem A, the solution operator over a unit of time maps L 2 ( T 2 ) L^{2}(\mathbb{T}^{2}) L 2 ( T 2 ) boundedly into C 0 , s ( T 2 ) C^{0,s}(\mathbb{T}^{2}) C 0 , s ( T 2 ) , and a positive operator with this property which preserves constants and integrals contracts the functions of mean zero, with a fixed contraction constant (Lemma 9.4).
Local regularity. Theorem C, stated and proved in § 5 , localizes Theorem A. Consider a solution of the advection–diffusion equation with κ ∈ K \kappa\in\mathcal{K} κ ∈ K , in the sense of distributions and with finite energy, in a space-time tube that follows the trajectories of the velocity field b m \mathbf{b}_{m} b m of the first m m m levels over the time 8 h m + 1 8h_{m+1} 8 h m + 1 (with h m + 1 h_{m+1} h m + 1 the renewal time of level m + 1 m+1 m + 1 , Subsection 2.2), with no lateral data prescribed. Theorem C bounds its norm in C 0 , s C^{0,s} C 0 , s on the trajectories of a smaller concentric disc, from a waiting time on, by a constant times its root mean square over the tube, and the constant depends neither on κ \kappa κ nor on the final time of the tube. The proof compares the solution with the solution on the whole torus from a cut-off datum, to which Theorem A applies.
Vanishing diffusivity. Corollary D, also in § 5 , passes to the limit κ → 0 \kappa\to0 κ → 0 . Let κ \kappa κ tend to 0 along a sequence K ′ \mathcal{K}' K ′ in K \mathcal{K} K , and let θ κ \theta^\kappa θ κ be the solutions, from a common initial time t i n t_{\mathrm{in}} t in , with data θ i n κ \theta^\kappa_{\mathrm{in}} θ in κ bounded in L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) . A subsequence converges locally uniformly on ( t i n , ∞ ) × T 2 (t_{\mathrm{in}},\infty) \times\mathbb{T}^2 ( t in , ∞ ) × T 2 to a continuous solution θ ∗ \theta_* θ ∗ of the transport equation ∂ t θ ∗ + ∇ ⋅ ( b θ ∗ ) = 0 \partial_t\theta_* + \nabla\cdot(b\theta_*) = 0 ∂ t θ ∗ + ∇ ⋅ ( b θ ∗ ) = 0 , in the sense of distributions. The limit takes the weak limit of the data along the subsequence as its initial value, weakly in L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) , satisfies the bound of Theorem A with sup κ ∈ K ′ ∥ θ i n κ ∥ L 2 ( T 2 ) \sup_{\kappa\in\mathcal{K}'} \|\theta^\kappa_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)} sup κ ∈ K ′ ∥ θ in κ ∥ L 2 ( T 2 ) in place of ∥ θ i n ∥ L 2 ( T 2 ) \|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)} ∥ θ in ∥ L 2 ( T 2 ) , and is Hölder continuous in time with exponent s / ( 1 + s ) s/(1+s) s / ( 1 + s ) on every set [ t i n + t 0 , t i n + T ] × T 2 [t_{\mathrm{in}}+t_0,t_{\mathrm{in}}+T] \times\mathbb{T}^2 [ t in + t 0 , t in + T ] × T 2 . Corollary D(b) is the local form, in the tubes of Theorem C. The conclusion is subsequential: for the velocity field of [5 ] with α ≥ 1 / 67 \alpha\ge1/67 α ≥ 1/67 , vanishing diffusivity does not select a unique limit [5 Proposition 5.5] ; see also [15 , 36 ] .
Prior results on anomalous dissipation and the Obukhov–Corrsin exponent
If b b b is Lipschitz continuous in space, the transport equation ∂ t θ + b ⋅ ∇ θ = 0 \partial_t\theta+b\cdot\nabla\theta=0 ∂ t θ + b ⋅ ∇ θ = 0 is well posed, its solutions conserve the L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) norm, and, by the energy identity, the dissipation κ ∥ ∇ θ κ ∥ L 2 ( ( t i n , t i n + 1 ) × T 2 ) 2 \kappa\|\nabla\theta^\kappa\|^2_{L^2((t_{\mathrm{in}},t_{\mathrm{in}}+1)\times\mathbb{T}^2)} κ ∥∇ θ κ ∥ L 2 (( t in , t in + 1 ) × T 2 ) 2 tends to zero with κ \kappa κ [5 Section 1.1] . Anomalous dissipation , the failure of this limit for rougher velocity fields, was first discussed by Obukhov [44 ] ; it is a basic assumption of the phenomenological theories of scalar turbulence, and it is supported by strong experimental and numerical evidence [47 , 49 , 27 ] . Rigorous examples include the white-in-time stochastic model of [41 , 11 ] , for which a quantitative proof of anomalous dissipation is given in [45 ] , and deterministic constructions. Those of [24 , 15 , 28 ] are built on alternating shear flows or on mixing, up to a singular time at which all the dissipation takes place [15 p. 4] , [28 Section 1.2.2] , [33 Remark 2.1] ; those of [33 , 34 ] iterate a construction of this type and dissipate at countably many times. Related constructions exist for forced fluid equations [9 , 8 , 17 , 35 ] . Autonomous examples were obtained in [38 , 39 ] (see also [7 ] ); the second of these uses the relation between spontaneous stochasticity and scalar dissipation established in [23 ] . The velocity field of [5 ] has infinitely many active scales at almost every time, and the proof of [5 Theorem 1.1] homogenizes these scales one at a time, from the scale at which molecular diffusion acts up to the unit scale, where the effective diffusivity is of order one.
The Obukhov–Corrsin exponent is a threshold for anomalous dissipation. As observed in [5 Section 1.3] , the commutator estimate of [16 ] shows that, for a velocity field which is continuous in time with values in C 0 , α ( T 2 ) C^{0,\alpha}(\mathbb{T}^2) C 0 , α ( T 2 ) , solutions which are bounded uniformly in κ \kappa κ , continuously in time, in a Hölder space of exponent larger than 1 2 ( 1 − α ) \frac{1}{2}(1-\alpha) 2 1 ( 1 − α ) dissipate no variance in the limit κ → 0 \kappa\to0 κ → 0 ; see [24 Theorem 4] for the scalar statement. Below the exponent, Colombo, Crippa and Sorella [15 ] obtained anomalous dissipation together with uniform Hölder bounds in the full range of exponents, for a particular datum and in L 2 L^2 L 2 in time [5 Sections 1.1 and 1.3] . A construction with anomalous dissipation for every nonzero smooth mean-zero datum and sharp time-integrated regularity bounds was subsequently given in [28 Theorem 2] . In contrast, the estimate of Theorem A is in L ∞ L^\infty L ∞ in time at positive times and holds for every datum in L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) , uniformly in κ ∈ K \kappa\in\mathcal{K} κ ∈ K .
Regularization from rough data has also been proved for other velocity fields. In [34 Theorem 1.5] , the deterministic construction mentioned above gives a positive Sobolev gain from data in L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) , on the time interval [ 0 , 1 ] [0,1] [ 0 , 1 ] from the initial time 0, measured in L 2 L^2 L 2 in time and uniformly in the diffusivity. For rough white-in-time Kraichnan velocities, anomalous Sobolev regularization was proved in [30 ] , and expected Sobolev bounds at each positive time for L 2 L^2 L 2 data were obtained in [25 Theorem 1.3] . On the torus, uniform-in-diffusivity estimates for the expected regularity yield bounds on the stationary scalar spectrum in [46 ] . The Sobolev threshold in these white-in-time models is 1 − α 1-\alpha 1 − α , rather than the exponent 1 2 ( 1 − α ) \frac{1}{2}(1-\alpha) 2 1 ( 1 − α ) for time-regular velocities [Row25, Remark 1.4]. For the elliptic, time-independent problem in self-similar random incompressible flows, in a perturbative regime, iterated homogenization gives Hölder estimates uniform in the molecular diffusivity, with a random constant determined by a minimal scale [1 Theorem C] . For regularization in nonlinear models, see [21 , 26 ] .
Burczak, Székelyhidi and Wu [13 ] combined the iterated homogenization of [5 ] with the convex integration methods used in the proof of Onsager’s conjecture [37 , 10 ] , and obtained anomalous dissipation for every nonzero mean-zero datum in H 1 ( T 3 ) H^1(\mathbb{T}^3) H 1 ( T 3 ) by Hölder continuous weak solutions of the unforced incompressible Euler equations on T 3 \mathbb{T}^3 T 3 , with any exponent below 1 / 3 1/3 1/3 [13 Theorem 1.1] . In [14 ] they proved anomalous dissipation for mean-zero data with more than 1 2 ( 1 + α ) \frac{1}{2}(1+\alpha) 2 1 ( 1 + α ) derivatives in L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) , a class larger than H 1 ( T 2 ) H^1(\mathbb{T}^2) H 1 ( T 2 ) , together with uniform Hölder bounds, along a sequence of diffusivities tending to 0. Their velocity fields are divergence free on T 2 \mathbb{T}^2 T 2 , with space-time Hölder exponent α < 1 / 3 \alpha< 1/3 α < 1/3 , and built, as they explain in [14 Section 1.3] , essentially as in [5 , 13 ] . Given s < 1 2 ( 1 − α ) s < \frac{1}{2}(1-\alpha) s < 2 1 ( 1 − α ) , their Hölder bound reads, in our notation, for mean-zero data θ i n ∈ H 1 + s ( T 2 ) \theta_{\mathrm{in}} \in H^{1+s}(\mathbb{T}^2) θ in ∈ H 1 + s ( T 2 ) with s ′ > s s' > s s ′ > s ,
∥ θ ∥ L ∞ ( [ 0 , 1 ] ; C 0 , s ( T 2 ) ) ≤ C ∥ θ i n ∥ H 1 + s ′ ( T 2 ) , \|\theta\|_{L^\infty([0,1];C^{0,s}(\mathbb{T}^2))} \le C\|\theta_{\mathrm{in}}\|_{H^{1+s'}(\mathbb{T}^2)}, ∥ θ ∥ L ∞ ([ 0 , 1 ] ; C 0 , s ( T 2 )) ≤ C ∥ θ in ∥ H 1 + s ′ ( T 2 ) ,
with C C C independent of the datum and of the diffusivity along the sequence [14 Theorem 1.1] .
In [14 Section 1.2.1] , the authors write that they “do not follow the ‘large scale regularity’ approach suggested” in [5 ] . Of a “naive approach to regularity”, they write that its second difficulty, an order-one background transport against a vanishing diffusivity at the larger scales, “seems to be a fundamental feature of our setup, prohibiting the use of ‘large-scale regularity’ techniques” [14 Section 1.3] . Their proof transfers regularity from coarser homogenized equations to finer ones, the mechanism described in [AV26, Section 4.1], by high-order corrector expansions and estimates for systems of forced equations [14 Theorems 6.10 and 7.2] , started separately for the Littlewood–Paley pieces of the datum [14 Section 8.3] .
For smooth data, their bound gives, for a velocity field depending on the exponent, the uniform Hölder regularity below the Obukhov–Corrsin exponent sought in [5 Section 1.3] . The interior regularization announced at the end of [5 Section 1.3] is a bound of a different kind: a bound at positive times of the Hölder norm of the solution by the norm of the datum in L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) , for every datum in L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) , with a constant that tends to infinity as the waiting time tends to 0. The bound of [14 ] propagates regularity: it holds up to the initial time, and its right side requires a datum which is already Hölder continuous.
Theorem A proves this interior regularization for every s < 1 2 ( 1 − α ) s < \frac{1}{2}(1-\alpha) s < 2 1 ( 1 − α ) , with the velocity field constructed here depending on s s s . Its proof carries out the large-scale regularity program announced in [5 Section 1.3] , by the induction described in [AV26, Section 4.1]. The comparison of two levels, in L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) at a high power of ε m − 1 \varepsilon_{m-1} ε m − 1 , is read in C 0 , s ( T 2 ) C^{0,s}(\mathbb{T}^2) C 0 , s ( T 2 ) through representations, which also measure derivatives at each wavelength, and the derivatives that the correctors consume are regenerated at every level (Proposition 5.3); both are explained in the strategy below.
The constants in the lower bounds of [5 Theorem 1.1] , [13 Theorem 1.1] and [14 Theorem 1.1] depend on the datum through a ratio of its L 2 L^2 L 2 norm to a Sobolev seminorm of positive order, while the constant of Theorem B is independent of the datum, of the initial time and of κ ∈ K \kappa\in\mathcal{K} κ ∈ K . Independence from the datum was announced in [5 Section 1.3] and [AV26, Section 4.1] as a consequence of uniform regularity. Among deterministic velocity fields, dissipation uniform over the initial datum was obtained by a different construction in [33 Corollary 1.3] , which gives total dissipation in the limit κ → 0 \kappa\to0 κ → 0 for every mean-zero datum in L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) ; for white-in-time velocity fields, see [45 , 46 ] . A related construction gives accelerating exponential decay for its time-periodic extension [34 Corollary 1.2] . These estimates are stated from the initial time 0, for velocity fields constructed on a unit time interval, whose dissipation takes place at singular times [34 Section 1.1] . Theorem B holds from every initial time t i n ∈ R t_{\mathrm{in}} \in\mathbb{R} t in ∈ R , for a velocity field which is periodic and Hölder continuous in space and time, with infinitely many active scales at almost every time.
Strategy of the proof
The proof of Theorem A rests on nine propositions and one lemma, all stated in Part I; their logical dependence is shown in Figure 3.1 . We next outline the main ideas of the proof, and postpone the technical estimates to Part II.
Figure 3.1. The statements and their uses. A solid arrow from one statement to another means that the proof of the second uses the first. The dashed arrow means that the entry repeats the method of the cycle at the terminal level. The two main steps are framed in bold. The dashed frame collects Proposition 2.1 and the two tools of Section 6, whose uses are not drawn.
Reduction to finitely many scales. Let κ ∈ K M \kappa\in\mathcal{K}_M κ ∈ K M . The levels of b \mathbf{b} b much finer than level M M M do not matter: their stream functions are much smaller than κ \kappa κ , a perturbation of the stream function of size κ ϖ \kappa\varpi κϖ moves the solution by at most ϖ ∥ θ i n ∥ L 2 ( T 2 ) \varpi\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)} ϖ ∥ θ in ∥ L 2 ( T 2 ) in L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) , and a Hölder estimate for small perturbations of the heat equation turns this into a bound in C 0 , s ( T 2 ) C^{0,s}(\mathbb{T}^2) C 0 , s ( T 2 ) . It therefore suffices to prove the bound of Theorem A for the velocity field b M + n \mathbf{b}_{M+n} b M + n , for a fixed number n n n of extra levels and for M M M large; the finitely many remaining bands follow from the same Hölder estimate directly (§ 1.5 ).
Effective diffusivities and three kinds of level. For κ ∈ K M \kappa\in\mathcal{K}_M κ ∈ K M , the effective diffusivities κ M : = κ , κ M − 1 , … , κ 0 \kappa_M := \kappa,\kappa_{M-1},\ldots,\kappa_0 κ M := κ , κ M − 1 , … , κ 0 of [5 ] are defined by a recursion: κ m − 1 \kappa_{m-1} κ m − 1 is the diffusivity that the solution feels once the pulses of level m m m , at diffusivity κ m \kappa_m κ m , have been averaged out (Subsection 2.5). By Taylor dispersion [48 ] (see also [29 , 43 ] ), the pulses of level m m m , whose stream function has amplitude ε m β \varepsilon_m^\beta ε m β , add about 9 80 ε m 2 β κ m \frac{9}{80}\frac{\varepsilon_m^{2\beta}}{\kappa_m} 80 9 κ m ε m 2 β to the diffusivity (Proposition 2.1 (a)). When the enhancement dominates, κ m − 1 κ m ≍ ε m 2 β \kappa_{m-1}\kappa_m \asymp\varepsilon_m^{2\beta} κ m − 1 κ m ≍ ε m 2 β , and the scaling κ m ≍ ε m β + γ \kappa_m \asymp\varepsilon_m^{\beta+\gamma} κ m ≍ ε m β + γ is consistent with this exactly when q ( β − γ ) = β + γ q(\beta-\gamma)=\beta+\gamma q ( β − γ ) = β + γ , which defines γ \gamma γ . For κ ∈ K M \kappa\in\mathcal{K}_M κ ∈ K M this scaling holds at every level 1 ≤ m < M 1 \le m < M 1 ≤ m < M , independently of κ \kappa κ (Proposition 2.1 (b)). The restriction to the bands comes from an instability of the recursion, which, according to [5 Section 3.3] , does not appear to be an artifact of the proof. The Péclet number ε m β / κ m \varepsilon_m^\beta/\kappa_m ε m β / κ m then sorts the levels finer than a fixed top level m 0 m_0 m 0 into three kinds, crossed as described below:
Table 1.
From L 2 L^2 L 2 to L ∞ L^\infty L ∞ , scale by scale. The proof of anomalous dissipation in [5 ] compares the solutions of the equations of the successive levels, with velocity field b m \mathbf{b}_m b m and diffusivity κ m \kappa_m κ m . For 1 ≤ m < M 1 \le m < M 1 ≤ m < M and a datum of unit norm, the energy identity at the diffusivity κ m \kappa_m κ m shows that the gradient of the solution of level m m m is at most of order κ m − 1 / 2 ≲ ε m − ( β + γ ) / 2 \kappa_m^{-1/2} \lesssim\varepsilon_m^{-(\beta+\gamma)/2} κ m − 1/2 ≲ ε m − ( β + γ ) /2 in L 2 ( ( t i n , ∞ ) × T 2 ) L^2((t_{\mathrm{in}},\infty)\times\mathbb{T}^2) L 2 (( t in , ∞ ) × T 2 ) (Proposition 2.1 (b)). If this bound held in L ∞ L^\infty L ∞ at every level 1 ≤ m < M 1 \le m < M 1 ≤ m < M , the solution would be Hölder continuous with the exponent
1 − β + γ 2 = 1 − q α 1 + q ↗ 1 − α 2 as q ↓ 1. 1-\frac{\beta+\gamma}{2}=\frac{1-q\alpha}{1+q}\nearrow\frac{1-\alpha}{2}
\quad\text{as }q\downarrow1. 1 − 2 β + γ = 1 + q 1 − q α ↗ 2 1 − α as q ↓ 1.
This heuristic (Subsection 5.1) is where the Obukhov–Corrsin exponent comes from, and why q q q is taken close to 1 1 1 . The proof turns the L 2 L^2 L 2 information carried by the effective diffusivities into L ∞ L^\infty L ∞ information, one scale at a time. This is the mechanism of large-scale regularity in homogenization: solutions are Lipschitz on scales larger than the microscopic scale, uniformly in the scale separation, for periodic [3 ] and random [4 , 2 , 32 ] coefficients. In the quantitative argument the solution is compared on each scale, as in the classical proof of Schauder estimates, with a solution of the homogenized equation, whose regularity improves the excess , the normalized L 2 L^2 L 2 distance of the solution to the affine functions, on the next smaller scale. For a Hölder estimate with exponent s < 1 s < 1 s < 1 , as here, the L 2 L^2 L 2 oscillation of the solution about its mean, divided by the scale to the power s s s , plays this role. Here this idea must be repeated across the infinitely many scales of the problem, as proposed in [AV26, Section 4.1] (compare [1 ] , where on each scale the solution is compared with a harmonic function): the homogenized equation of level m m m is the equation of level m − 1 m-1 m − 1 , which is again fractal and must be homogenized in its turn. From one level to the next, separation of scales gains ( ε m / ε m − 1 ) 1 − s (\varepsilon_m/\varepsilon_{m-1})^{1-s} ( ε m / ε m − 1 ) 1 − s and the enhancement costs ( κ m − 1 / κ m ) 1 / 2 (\kappa_{m-1}/\kappa_m)^{1/2} ( κ m − 1 / κ m ) 1/2 , the amplitude of the correctors of level m m m relative to ε m \varepsilon_m ε m ; their product is at most C ε m − 1 μ C\varepsilon_{m-1}^{\mu} C ε m − 1 μ for 2 ≤ m ≤ M 2 \le m \le M 2 ≤ m ≤ M , with the exponent μ = ( q − 1 ) ( 1 − β + γ 2 − s ) \mu= (q-1)(1-\frac{\beta+\gamma}{2}-s) μ = ( q − 1 ) ( 1 − 2 β + γ − s ) of (5) and (29) , which is positive exactly when s < 1 − β + γ 2 s < 1-\frac{\beta+\gamma}{2} s < 1 − 2 β + γ .
The main difficulty. At every level, the velocity field b m − 1 \mathbf{b}_{m-1} b m − 1 of the coarser levels is of order one, while the diffusivity is small at the fine levels: κ m ≲ ε m β + γ \kappa_m \lesssim\varepsilon_m^{\beta+\gamma} κ m ≲ ε m β + γ for 1 ≤ m < M 1 \le m < M 1 ≤ m < M . The construction removes this transport from the analysis of a level. The pulses of level m m m are written in the Lagrangian coordinates of b m − 1 \mathbf{b}_{m-1} b m − 1 , renewed on blocks of time of length h m h_m h m , the renewal time of level m m m , which is short compared with the time scale ε m − 1 2 − β \varepsilon_{m-1}^{2-\beta} ε m − 1 2 − β of b m − 1 \mathbf{b}_{m-1} b m − 1 . Within a renewal block the coordinates are close to rigid, and the coarse flow carries the fine shear along instead of sweeping across it (§ 1.1 ). In these coordinates the coarse transport disappears, and the Laplacian becomes a divergence-form operator whose coefficient, the conductivity , is within C ε m − 1 2 δ C\varepsilon_{m-1}^{2\delta} C ε m − 1 2 δ of the identity, where δ > 0 \delta> 0 δ > 0 is the small exponent of (5) . The correctors and the whole analysis of level m m m are written in these coordinates. The other difficulty is that the correctors consume derivatives, while the datum is only in L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) and the induction cannot lose derivatives at each of infinitely many levels: the derivatives are regenerated at every level, within one period of its pulses.
Two dual cascades. Fix M M M and κ ∈ K M \kappa\in\mathcal{K}_M κ ∈ K M . By the definition of the effective diffusivities, homogenizing the pulses of level m ≤ M m \le M m ≤ M turns the operator ∂ t + b m ⋅ ∇ − κ m Δ \partial_t+\mathbf{b}_m\cdot\nabla-\kappa_m\Delta ∂ t + b m ⋅ ∇ − κ m Δ of level m m m into the operator of level m − 1 m-1 m − 1 ; for m > M m>M m > M , the operator of level m m m has κ \kappa κ in place of κ m \kappa_m κ m . The proof is an induction on the levels made of two dual cascades (Subsection 5.2), both run by Proposition 5.2 between the levels M − 1 M-1 M − 1 and m 0 m_0 m 0 . The homogenization cascade goes from fine to coarse, up the levels and forward in time: the entry (Proposition 5.1 ) crosses the levels M + n , … , M M+n,\ldots,M M + n , … , M , and then one cycle at a time (Proposition 5.3), each in a short interval of time, crosses the levels M − 1 , … , m 0 + 1 M-1,\ldots,m_0+1 M − 1 , … , m 0 + 1 , up to a fixed top level m 0 m_0 m 0 , independent of M M M and κ \kappa κ , at which the solution is smooth at a fixed scale. Each cycle compares the solution of a level with one of the homogenized equation, the equation of the next coarser level; this cascade carries the marked amplitude , a weighted L 2 L^2 L 2 norm taken in L ∞ L^\infty L ∞ in time (Subsection 5.2). The regularity cascade goes from coarse to fine, back down the levels at one fixed time: as in large-scale regularity, it transfers a Hölder bound from the homogenized equation to the heterogeneous one, one finer level at a time. Each cascade needs the other: the first produces the homogenized solutions with which the second compares, and the second turns the L 2 L^2 L 2 control of the first into a Hölder bound.
Families of coupled equations. The comparison of a level with the next coarser one leaves error terms that are not small and that carry derivatives of the coarse solution. They become new unknowns, each solving the equation of its level with a source made of smoothed derivatives of earlier unknowns, so that the object passed from one level to the next is a coupled system of equations with the operator of its level, a family (Subsection 5.4). The marked amplitude of a solution of a family, which weights the L 2 L^2 L 2 norms of its components , the unknowns of the system, is multiplied per cycle by at most one plus a constant times a fixed positive power of ε m − 1 \varepsilon_{m-1} ε m − 1 ; since ε m ≤ Λ − m \varepsilon_m \le\Lambda^{-m} ε m ≤ Λ − m , these factors have a bounded product, uniformly in the number of levels.
The entry. At the levels finer than M M M , the enhancement of a level is small compared with κ \kappa κ . A weak passage (Subsection 15.2) therefore keeps the diffusivity κ \kappa κ , carries the enhancement as a source of the coarser family (Subsection 5.4), and regains derivatives on the quiet margins, the pulse-free intervals of length τ m ′ \tau'_m τ m ′ at both ends of each block of time of length h m h_m h m on which the pulses of level m m m are renewed (§ 1.1 ). Over a margin, molecular diffusion reaches the wavelength ε m − 1 \varepsilon_{m-1} ε m − 1 . Level M M M also has a small Péclet number, but its margins are too short for this, κ τ M ′ ≪ ε M − 1 2 \kappa\tau'_M \ll\varepsilon_{M-1}^2 κ τ M ′ ≪ ε M − 1 2 , so it is crossed by a terminal cycle (Subsection 15.3), which passes to κ M − 1 \kappa_{M-1} κ M − 1 and produces the first family of the homogenization cascade (Proposition 5.2 ), the seed (Subsection 5.4). The weak passages and the terminal cycle form the entry (§ 6 ).
One homogenization cycle. A cycle has two parts. The first, the preparation (§ 5.6 ), regenerates derivatives: within one period 4 h m 4h_m 4 h m of the pulses, the solution of level m m m becomes, up to a small error, smooth at the effective diffusion length ρ ( m ) = ( κ m − 1 h m ) 1 / 2 \rho(m)=(\kappa_{m-1}h_m)^{1/2} ρ ( m ) = ( κ m − 1 h m ) 1/2 , which lies between the two wavelengths, ε m ≪ ρ ( m ) ≪ ε m − 1 \varepsilon_m\ll\rho(m)\ll\varepsilon_{m-1} ε m ≪ ρ ( m ) ≪ ε m − 1 (Proposition 2.1 (c) and Lemma 7.1 (c)). The proof uses the Bloch transform, the Fourier transform adapted to periodic coefficients. For the periodic elliptic theory of corrector expansions and Bloch approximation, see [12 , 20 , 19 ] . In the Lagrangian coordinates of a renewal block, with the conductivity frozen at a point, the operator of level m m m is periodic with period ε m \varepsilon_m ε m , and Bloch waves split its evolution into evolutions on the period cell, one for each frequency ξ \xi ξ . The principal branch of these evolutions, the one that continues the constants, has, to leading order, the symbol − ξ ⋅ J m κ ( t ) ξ -\xi\cdot J_m^{\kappa}(t)\xi − ξ ⋅ J m κ ( t ) ξ , where the averaged flux J m κ ( t ) J_m^{\kappa}(t) J m κ ( t ) is the diffusivity matrix produced by the pulses through Taylor dispersion, enhanced along the direction of the active pulses ((18) ). Since every renewal block contains pulses of both directions, the real part of this symbol, integrated over a block, is at most a negative multiple of κ m − 1 h m ∣ ξ ∣ 2 \kappa_{m-1}h_m|\xi|^2 κ m − 1 h m ∣ ξ ∣ 2 , in every direction and uniformly in the level: the principal branch is damped like a heat flow at the effective diffusion length. At the small frequencies, the part of each of these evolutions off the branch decays at the rate κ m / ε m 2 \kappa_m/\varepsilon_m^2 κ m / ε m 2 of the cell problem, and at the remaining frequencies the whole evolution does.
The second part, the comparison (§ 5.8 ), builds from the solution of the next family, the successor (Subsection 5.4), a lift to level m m m , in three steps (Subsection 5.5). It adds each new component of the successor to its parent component (Subsection 5.4), with a small weight. It undoes the oscillation in time of the averaged flux, which alternates between the two directions of the pulses and has the average κ m − 1 I \kappa_{m-1}I κ m − 1 I over a period: it applies the truncated exponential of the second-order operator Γ m ( t ) : ∇ 2 \Gamma_m(t):\nabla^2 Γ m ( t ) : ∇ 2 , where the covariance Γ m ( t ) \Gamma_m(t) Γ m ( t ) is the accumulated difference between J m κ J_m^{\kappa} J m κ and κ m − 1 I \kappa_{m-1}I κ m − 1 I . Since ∣ Γ m ( t ) ∣ ≤ C ε m − 1 δ ρ ( m ) 2 |\Gamma_m(t)|\le C\varepsilon_{m-1}^{\delta}\rho(m)^2 ∣ Γ m ( t ) ∣ ≤ C ε m − 1 δ ρ ( m ) 2 , each power of the covariance gains ε m − 1 δ \varepsilon_{m-1}^{\delta} ε m − 1 δ on functions that are smooth at the scale ρ ( m ) \rho(m) ρ ( m ) , so the exponential can be truncated. Finally, it adds the correctors of the pulses, which oscillate at the wavelength ε m \varepsilon_m ε m . The errors that are not small become the sources of the next family; the rest is a residual of the size of a fixed power of ε m − 1 \varepsilon_{m-1} ε m − 1 , which the energy estimate for families converts into a bound on the difference between the solution and the lift. A cycle that starts at a time t c t_c t c is prepared by a time t p r t_{\rm pr} t pr within 4 h m 4h_m 4 h m of t c t_c t c , and the comparison holds from t p r t_{\rm pr} t pr on (Proposition 5.3).
Representations and the Hölder norm. The regularity cascade carries the Hölder norm by representations : decompositions of a solution of a family into a smooth base and oscillating profiles, each measured at its wavelength (Subsection 5.6); the size of a representation plays the role of the excess. In terms of representations, a cycle is a passage (Proposition 5.4 ): from a representation of the solution of level m − 1 m-1 m − 1 at the wavelength ε m − 1 \varepsilon_{m-1} ε m − 1 , it builds one of the solution of level m m m at the wavelength ε m \varepsilon_m ε m , at the same time. The old profiles are copied and never differentiated again; the new ones are the parts of the lift and the error of the comparison. A passage multiplies the size of a representation by at most one plus a constant times a fixed positive power of ε m − 1 \varepsilon_{m-1} ε m − 1 , and adds at most a constant times ε m − 1 \varepsilon_{m-1} ε m − 1 times the marked amplitude of the solution; the balance ε m − 1 μ \varepsilon_{m-1}^{\mu} ε m − 1 μ above pays for the correctors. Running the two cascades bounds the size of the representation at the finest level, and with it the C 0 , s ( T 2 ) C^{0,s}(\mathbb{T}^2) C 0 , s ( T 2 ) norm, by the L 2 L^2 L 2 norm of the datum, uniformly in the number of levels (Proposition 5.2 ). At the finest level, the lifts through the entry recover θ \theta θ from the solution of the seed (§ 6 ).
Energy estimates and paired smoothing. The families are coupled systems of parabolic equations, and their energy does not grow when the coupling is small (Proposition 6.2 ). The sources that a homogenization cycle creates are smoothed by products of averages over two thin ellipses, whose long axes come from the two orientations of the pulses; such pairs smooth in every direction, uniformly in their eccentricity (Proposition 6.1 ). For the sources of the seed the two ellipses are round discs.
Lean formalization
A formalization of this paper in the Lean 4 proof assistant is available at
https://github.com/scottnarmstrong/AVAnomalousRegularization .
It contains Theorems A, B and C, Corollary D and every statement of Part I; the propositions and lemmas of Part II follow from its formal statements, directly or by elementary steps, or are proved, where they are used, inside its formal proofs. A correspondence table, published with the repository, maps every numbered statement and display of the paper to the formal statements that prove it. Some results are stated here in a weaker form than their formal versions: for example, where the paper bounds a norm in L ∞ L^\infty L ∞ in time, the formal statement gives the bound at every time. Theorems A, B and C and Corollary D are also stated in a self-contained form, whose definitions use only Mathlib and the statement definitions of the formalization of [5 ] , with formal proofs connecting these statements to the development. The formal proofs follow the arguments given here; in some places the paper combines formally proved statements by elementary steps, such as an induction or a sequence of inequalities, which the formal proof combines differently. The formalization uses Lean 4 and Mathlib at version v4.35.0-rc2 and builds on the formalization of [5 ] , available at
https://github.com/scottnarmstrong/AVAnomalousDiffusion .
It contains no unproved placeholders and no custom axioms: the main theorems depend only on the three standard axioms of Lean 4 (propext, Classical.choice and Quot.sound).
Use of artificial intelligence
The main ideas of this work were developed by the authors several years ago; the program was announced in [5 Section 1.3] and described in [AV26, Section 4.1]. Starting from a detailed outline of the proof written by the authors, a very long and unpolished manuscript (612 pages of slop), in which every detail of the proof was filled in, was first produced by the authors together with AI agents (Claude Opus 5.5 and GPT-6.1 Sol). This manuscript was then formalized in Lean by AI agents directed by the authors: Claude Opus 5.5 orchestrated the development throughout, and the Lean code was written by Claude Opus 5.5 and Claude Sonnet 5.5, with parts written by GPT-6.1 Sol and GPT-6 Luna. Only then was the present paper written. Its first draft was written by Claude Opus 5.5 following the formalization, and Claude Sonnet 5.5 and GPT-6-Astra checked its statements against the formal ones and reviewed it. The authors checked this draft and rewrote parts of it themselves; the rest was rewritten by Claude Opus 5.5 agents under the direction of the authors, over several rounds of revision and audit.
Organization of the paper
Part I (§ 1.1 –§ 2.4 ) is a detailed outline of the proof, which can be read by itself. § 1.1 constructs the velocity field and states the properties of its effective diffusivities, § 1.4 states Theorems A and B, and § 1.5 reduces Theorem A to a velocity field with finitely many scales. § 2 explains the two dual cascades, defines families, lifts and representations, and states the entry, the induction on the levels, one cycle and one passage. § 2.4 states the two tools. Part II (§ 4.1 –§ 6 ) states the notions used in Part I in full and proves the statements of Part I that are not proved there. Its sections treat the properties of the construction and of the effective diffusivities, relying on [5 ] , and the flows and renewal charts (§ 4.1 ); the two tools (§ 4.2 ); the reduction and the main results, including the statements of Theorem C and Corollary D (§ 5 ); the cell problems of the pulses, which give the correctors and the averaged flux (§ 5.2 ); admissible families (§ 5.4 ); the two parts of one cycle, the preparation and the comparison (§ 5.6 and § 5.8 ); the passage and the induction on the levels (§ 5.11 ); and the entry (§ 6 ).
A map of the proof
Construction of the velocity field
In this section we fix the parameters, define the scales, the velocity field b \mathbf{b} b and the effective diffusivities, and state their properties. The scales, shear streams, cutoff function and recursion are those of [5 ] , except for the parameter q q q , which [5 (2.2)] fixes as q = 1 2 ( 1 + 2 − β 2 ( β − 1 ) ) q = \frac{1}{2}\left(1+\frac{2-\beta}{2(\beta-1)}\right) q = 2 1 ( 1 + 2 ( β − 1 ) 2 − β ) . Equation and statement numbers of [5 ] refer to its arXiv version, arXiv:2305.05048v3.
Parameters
Throughout we fix
α ∈ ( 0 , 1 / 3 ) \alpha\in(0,1/3) α ∈ ( 0 , 1/3 ) , the Hölder exponent of the velocity field b \mathbf{b} b ; we also set β : = 1 + α \beta:= 1+\alpha β := 1 + α ;
s ∈ ( 0 , 1 2 ( 1 − α ) ) s \in\left(0,\frac{1}{2}(1-\alpha)\right) s ∈ ( 0 , 2 1 ( 1 − α ) ) , the Hölder exponent of the solutions in Theorem A;
q q q satisfying
1 < q < min { 17 16 , 1 − α 2 α , 1 − s α + s } , (3) 1 < q < \min\left\{\frac{17}{16},\frac{1-\alpha}{2\alpha},\frac{1-s}{\alpha+s}\right\}, \tag*{(3)} 1 < q < min { 16 17 , 2 α 1 − α , α + s 1 − s } , (3)
the supergeometric scale separation parameter; such a q q q exists because α < 1 / 3 \alpha< 1/3 α < 1/3 and s < 1 2 ( 1 − α ) s < \frac{1}{2}(1-\alpha) s < 2 1 ( 1 − α ) , and the last two bounds make the exponents δ \delta δ and μ \mu μ below positive;
an even cutoff function ζ ∈ C c ∞ ( ( − 2 3 , 2 3 ) ) \zeta\in C^\infty_c\left(\left(-\frac{2}{3},\frac{2}{3}\right)\right) ζ ∈ C c ∞ ( ( − 3 2 , 3 2 ) ) with
0 ≤ ζ ≤ 1 , ∑ k ∈ Z ζ ( ⋅ − k ) = 1 , ∫ R ζ ( t ) 2 d t = 9 10 , (4) 0 \le\zeta\le1,\qquad\sum_{k\in\mathbb{Z}}\zeta(\cdot-k)=1,\qquad\int_{\mathbb{R}}\zeta(t)^2\,\mathrm{d}t=\frac{9}{10}, \tag*{(4)} 0 ≤ ζ ≤ 1 , k ∈ Z ∑ ζ ( ⋅ − k ) = 1 , ∫ R ζ ( t ) 2 d t = 10 9 , (4)
which switches the pulses on and off; the last condition fixes the constant 9 80 \frac{9}{80} 80 9 of Proposition 2.1 , and such a ζ \zeta ζ exists (Subsection 7.1).
Together with these we fix the exponents
γ : = ( q − 1 ) β q + 1 , δ : = 1 4 ( q − 1 ) ( 1 − ( 2 q + 1 ) β 2 q + 2 ) , \gamma:=\frac{(q-1)\beta}{q+1},\qquad\delta:=\frac{1}{4}(q-1)\left(1-\frac{(2q+1)\beta}{2q+2}\right), γ := q + 1 ( q − 1 ) β , δ := 4 1 ( q − 1 ) ( 1 − 2 q + 2 ( 2 q + 1 ) β ) ,
μ : = ( q − 1 ) ( 1 − s ) − q γ , η : = min { δ 8 , μ 4 } . (5) \mu:=(q-1)(1-s)-q\gamma,\qquad\eta:=\min\left\{\frac{\delta}{8},\frac{\mu}{4}\right\}. \tag*{(5)} μ := ( q − 1 ) ( 1 − s ) − q γ , η := min { 8 δ , 4 μ } . (5)
The exponent γ \gamma γ solves q ( β − γ ) = β + γ q(\beta-\gamma)=\beta+\gamma q ( β − γ ) = β + γ and gives the scaling of the effective diffusivities (Proposition 2.1 (b)), and the small exponent δ > 0 \delta>0 δ > 0 separates the time scales of Subsection 2.2. The exponents μ > 0 \mu>0 μ > 0 and η > 0 \eta>0 η > 0 are used from § 2 on, μ \mu μ as the exponent of the cost {(5.2)} of one level.
We write x ≍ y x\asymp y x ≍ y if C − 1 y ≤ x ≤ C y C^{-1}y\le x\le Cy C − 1 y ≤ x ≤ C y for a constant C C C depending only on α \alpha α , q q q , ζ \zeta ζ , and, for m ≥ 2 m\ge2 m ≥ 2 , we write x ≪ y x\ll y x ≪ y , or y ≫ x y\gg x y ≫ x , if x ≤ C ε m − 1 c y x\le C\varepsilon_{m-1}^{c}y x ≤ C ε m − 1 c y for constants c > 0 c>0 c > 0 and C C C depending only on α \alpha α , q q q , ζ \zeta ζ .
Throughout, ∣ ⋅ ∣ \lvert\cdot\rvert ∣ ⋅ ∣ is the absolute value of a number, the Euclidean norm of a vector, and the operator norm, for the Euclidean norms, of a matrix, or of a multilinear map regarded as a linear map on arrays. We denote by ( g ) E : = 1 ∣ E ∣ ∫ E g (g)_E:=\frac{1}{\lvert E\rvert}\int_E g ( g ) E := ∣ E ∣ 1 ∫ E g the average of a function g g g over a set E E E of finite positive measure. On R 2 \mathbb{R}^2 R 2 and C 2 \mathbb{C}^2 C 2 we also write ∣ ⋅ ∣ ∞ \lvert\cdot\rvert_\infty ∣ ⋅ ∣ ∞ for the largest modulus of a coordinate and, for a 2 × 2 2\times2 2 × 2 matrix A A A , ∣ A ∣ ∞ \lvert A\rvert_\infty ∣ A ∣ ∞ for its operator norm for ∣ ⋅ ∣ ∞ \lvert\cdot\rvert_\infty ∣ ⋅ ∣ ∞ , the largest sum of the moduli of the entries of a row. For a map or a vector field X X X , ∇ X \nabla X ∇ X is its Jacobian matrix, ( ∇ X ) i i ′ = ∂ i X i ′ (\nabla X)_{ii'}=\partial_iX_{i'} ( ∇ X ) i i ′ = ∂ i X i ′ .
Scales
For m ≥ 1 m\ge1 m ≥ 1 , level m m m has the wavelength ε m \varepsilon_m ε m , the renewal time h m h_m h m , the margin time τ m ′ \tau'_m τ m ′ and the pulse time τ m \tau_m τ m , defined by ε 0 : = 1 \varepsilon_0:=1 ε 0 := 1 and [5 (2.8), (2.12)–(2.14)]
ε m : = [ Λ q m / ( q − 1 ) ] − 1 , h m : = 2 − 25 ⌈ ε m − 1 − ( 2 − β + 2 δ ) ⌉ , \varepsilon_m:=\left[\Lambda^{q^m/(q-1)}\right]^{-1},\qquad h_m:=\frac{2^{-25}}{\left\lceil\varepsilon_{m-1}^{-(2-\beta+2\delta)}\right\rceil}, ε m := [ Λ q m / ( q − 1 ) ] − 1 , h m := ⌈ ε m − 1 − ( 2 − β + 2 δ ) ⌉ 2 − 25 ,
τ m ′ : = h m 4 ⌈ ε m − 1 − δ ⌉ + 1 , τ m : = h m ( 4 ⌈ ε m − 1 − δ ⌉ + 1 ) 2 . (6) \tau'_m:=\frac{h_m}{4\left\lceil\varepsilon_{m-1}^{-\delta}\right\rceil+1},\qquad\tau_m:=\frac{h_m}{\left(4\left\lceil\varepsilon_{m-1}^{-\delta}\right\rceil+1\right)^2}. \tag*{(6)} τ m ′ := 4 ⌈ ε m − 1 − δ ⌉ + 1 h m , τ m := ( 4 ⌈ ε m − 1 − δ ⌉ + 1 ) 2 h m . (6)
Then, for every m ≥ 2 m\ge2 m ≥ 2 , ε m ≍ ε m − 1 q \varepsilon_m\asymp\varepsilon_{m-1}^q ε m ≍ ε m − 1 q , and, for every m ≥ 1 m\ge1 m ≥ 1 , ε m − 1 / ε m ≥ Λ \varepsilon_{m-1}/\varepsilon_m\ge\Lambda ε m − 1 / ε m ≥ Λ , 4 ⌈ ε m − 1 − δ ⌉ + 1 ≍ ε m − 1 − δ 4\lceil\varepsilon_{m-1}^{-\delta}\rceil+1\asymp\varepsilon_{m-1}^{-\delta} 4 ⌈ ε m − 1 − δ ⌉ + 1 ≍ ε m − 1 − δ and
h m ≍ ε m − 1 2 − β + 2 δ , τ m ′ ≍ ε m − 1 2 − β + 3 δ , τ m ≍ ε m − 1 2 − β + 4 δ , (7) h_m\asymp\varepsilon_{m-1}^{2-\beta+2\delta},\qquad\tau'_m\asymp\varepsilon_{m-1}^{2-\beta+3\delta},\qquad\tau_m\asymp\varepsilon_{m-1}^{2-\beta+4\delta}, \tag*{(7)} h m ≍ ε m − 1 2 − β + 2 δ , τ m ′ ≍ ε m − 1 2 − β + 3 δ , τ m ≍ ε m − 1 2 − β + 4 δ , (7)
so that τ m ≪ τ m ′ ≪ h m \tau_m\ll\tau'_m\ll h_m τ m ≪ τ m ′ ≪ h m for m ≥ 2 m\ge2 m ≥ 2 ; the ratios τ m ′ / τ m \tau'_m/\tau_m τ m ′ / τ m and h m / τ m ′ h_m/\tau'_m h m / τ m ′ both equal 4 ⌈ ε m − 1 − δ ⌉ + 1 4\lceil\varepsilon_{m-1}^{-\delta}\rceil+1 4 ⌈ ε m − 1 − δ ⌉ + 1 .
Pulses and the slow cutoff
The slots of level m m m are indexed by k ∈ Z k\in\mathbb{Z} k ∈ Z , and slot k k k carries the shear stream [5 (2.19)–(2.20)]
ψ m , k ( y ) : = { ε m β sin ( 2 π y 1 / ε m ) if k ∈ 4 Z + 1 , ε m β sin ( 2 π y 2 / ε m ) if k ∈ 4 Z + 3 , 0 if k ∈ 2 Z , (8) \psi_{m,k}(y):=
\begin{cases}
\varepsilon_m^\beta\sin(2\pi y_1/\varepsilon_m) & \text{if } k\in4\mathbb{Z}+1,\\
\varepsilon_m^\beta\sin(2\pi y_2/\varepsilon_m) & \text{if } k\in4\mathbb{Z}+3,\\
0 & \text{if } k\in2\mathbb{Z},
\end{cases} \tag*{(8)} ψ m , k ( y ) := ⎩ ⎨ ⎧ ε m β sin ( 2 π y 1 / ε m ) ε m β sin ( 2 π y 2 / ε m ) 0 if k ∈ 4 Z + 1 , if k ∈ 4 Z + 3 , if k ∈ 2 Z , (8)
so even slots are empty and the odd ones, the pulses , alternate between the two directions. Slot k k k is active for a time of about τ m \tau_m τ m around k τ m k\tau_m k τ m , through the fast cutoff
ζ m , k ( t ) : = ζ ( t − k τ m τ m ) . (9) \zeta_{m,k}(t) := \zeta\left(\frac{t-k\tau_m}{\tau_m}\right). \tag*{(9)} ζ m , k ( t ) := ζ ( τ m t − k τ m ) . (9)
The renewal blocks are the intervals [ ( l − 1 2 ) h m , ( l + 1 2 ) h m ] [(l-\frac{1}{2})h_m,(l+\frac{1}{2})h_m] [( l − 2 1 ) h m , ( l + 2 1 ) h m ] , l ∈ Z l\in\mathbb{Z} l ∈ Z ; slot k k k lies in the block of
l m , k : = ⌊ k τ m + 1 2 ( h m − τ m ) h m ⌋ . (10) l_{m,k}:=\left\lfloor\frac{k\tau_m+\frac{1}{2}(h_m-\tau_m)}{h_m}\right\rfloor. \tag*{(10)} l m , k := ⌊ h m k τ m + 2 1 ( h m − τ m ) ⌋ . (10)
This is the label for which [5 (2.18)] holds: the slot [ ( k − 1 2 ) τ m , ( k + 1 2 ) τ m ] [(k-\frac{1}{2})\tau_m,(k+\frac{1}{2})\tau_m] [( k − 2 1 ) τ m , ( k + 2 1 ) τ m ] lies in the block of l m , k l_{m,k} l m , k (Lemma 7.1(d)). The slow cutoff of the block of l l l is
ζ ^ m , l ( t ) : = H ( t − ( l − 1 2 ) h m τ m ′ − 1 ) H ( ( l + 1 2 ) h m − t τ m ′ − 1 ) , (11) \widehat{\zeta}_{m,l}(t):=H\left(\frac{t-(l-\frac{1}{2})h_m}{\tau'_m}-1\right)H\left(\frac{(l+\frac{1}{2})h_m-t}{\tau'_m}-1\right), \tag*{(11)} ζ m , l ( t ) := H ( τ m ′ t − ( l − 2 1 ) h m − 1 ) H ( τ m ′ ( l + 2 1 ) h m − t − 1 ) , (11)
where H H H is the smooth step with H = 0 H=0 H = 0 on ( − ∞ , 0 ] (-\infty,0] ( − ∞ , 0 ] , H = 1 H=1 H = 1 on [ 1 , ∞ ) [1,\infty) [ 1 , ∞ ) and
H ( r ) : = exp ( − 1 / r ) exp ( − 1 / r ) + exp ( − 1 / ( 1 − r ) ) for 0 < r < 1. H(r):=\frac{\exp(-1/r)}{\exp(-1/r)+\exp(-1/(1-r))}\quad\text{for }0<r<1. H ( r ) := exp ( − 1/ r ) + exp ( − 1/ ( 1 − r )) exp ( − 1/ r ) for 0 < r < 1.
The slow cutoff vanishes within τ m ′ \tau'_m τ m ′ of the ends of the block and equals one at distance at least 2 τ m ′ 2\tau'_m 2 τ m ′ from them. Each block therefore has a pulse-free quiet margin of length τ m ′ \tau'_m τ m ′ at both ends, and the envelope of slot k k k is
ζ ~ m , k ( t ) : = ζ ^ m , l m , k ( t ) ζ m , k ( t ) . (12) \widetilde{\zeta}_{m,k}(t):=\widehat{\zeta}_{m,l_{m,k}}(t)\zeta_{m,k}(t). \tag*{(12)} ζ m , k ( t ) := ζ m , l m , k ( t ) ζ m , k ( t ) . (12)
Our slow cutoff (11) is one explicit choice among those of [5 (2.29), (2.31)] , since ζ ^ m , l = ζ ^ m , 0 ( ⋅ − l h m ) \widehat{\zeta}_{m,l}=\widehat{\zeta}_{m,0}(\cdot-lh_m) ζ m , l = ζ m , 0 ( ⋅ − l h m ) , it has the support and the plateau stated above, and its derivatives satisfy the bounds of Lemma 7.1(d). Figure 2.1 draws one block.
Figure 2.1. One renewal block of level m m m , in time (not to scale: a block holds ( 4 ⌈ ε m − 1 − δ ⌉ + 1 ) 2 (4\lceil\varepsilon_{m-1}^{-\delta}\rceil+1)^2 ( 4 ⌈ ε m − 1 − δ ⌉ + 1 ) 2 slots). Shaded: the quiet margins, of length τ m ′ \tau'_m τ m ′ . Dashed: the slow cutoff ζ ^ m , l \widehat{\zeta}_{m,l} ζ m , l of (11) . Solid: the envelopes ζ ~ m , k \widetilde{\zeta}_{m,k} ζ m , k of (12) of the pulses k k k , which are 2 τ m 2\tau_m 2 τ m apart; the even slots are empty. The arrow above a pulse is its direction of flow e ( k ) e_{(k)} e ( k ) (Subsection 2.5): vertical for k ∈ 4 Z + 1 k\in4\mathbb{Z}+1 k ∈ 4 Z + 1 , horizontal for k ∈ 4 Z + 3 k\in4\mathbb{Z}+3 k ∈ 4 Z + 3 .
The velocity field
Set ϕ 0 : = 0 \phi_0:=0 ϕ 0 := 0 and ∇ ⊥ : = ( − ∂ 2 , ∂ 1 ) \nabla^\perp:=(-\partial_2,\partial_1) ∇ ⊥ := ( − ∂ 2 , ∂ 1 ) . For m ≥ 1 m\ge1 m ≥ 1 , let X m − 1 ( t , x , t ′ ) X_{m-1}(t,x,t') X m − 1 ( t , x , t ′ ) denote the position at time t t t of the trajectory of b m − 1 = ∇ ⊥ ϕ m − 1 \mathbf{b}_{m-1}=\nabla^\perp\phi_{m-1} b m − 1 = ∇ ⊥ ϕ m − 1 through x x x at time t ′ t' t ′ , that is, the solution of
{ ∂ t X m − 1 ( t , x , t ′ ) = b m − 1 ( t , X m − 1 ( t , x , t ′ ) ) in R , X m − 1 ( t ′ , x , t ′ ) = x . (13) \begin{cases}
\partial_tX_{m-1}(t,x,t')=\mathbf{b}_{m-1}(t,X_{m-1}(t,x,t')) & \text{in }\mathbb{R},\\
X_{m-1}(t',x,t')=x. \tag*{(13)}
\end{cases} { ∂ t X m − 1 ( t , x , t ′ ) = b m − 1 ( t , X m − 1 ( t , x , t ′ )) X m − 1 ( t ′ , x , t ′ ) = x . in R , (13)
and define [5 (2.36)]
ϕ m ( t , x ) : = ϕ m − 1 ( t , x ) + ∑ k ∈ Z ζ ~ m , k ( t ) ψ m , k ( X m − 1 ( l m , k h m , x , t ) ) , b m : = ∇ ⊥ ϕ m . (14) \phi_m(t,x):=\phi_{m-1}(t,x)+\sum_{k\in\mathbb{Z}}\widetilde{\zeta}_{m,k}(t)\psi_{m,k}(X_{m-1}(l_{m,k}h_m,x,t)),\qquad\mathbf{b}_m:=\nabla^\perp\phi_m. \tag*{(14)} ϕ m ( t , x ) := ϕ m − 1 ( t , x ) + k ∈ Z ∑ ζ m , k ( t ) ψ m , k ( X m − 1 ( l m , k h m , x , t )) , b m := ∇ ⊥ ϕ m . (14)
The level-m m m pulses are thus written in the Lagrangian coordinates of the coarser flow, renewed at the centre l h m l h_m l h m of each block: within a block the variable is the position at time l h m l h_m l h m of the trajectory of b m − 1 \mathbf{b}_{m-1} b m − 1 , so the coarse flow carries the fine shear along instead of sweeping across it. The coarse velocity field satisfies ∥ ∇ b m − 1 ∥ L ∞ ( R × T 2 ) ≤ C ε m − 1 β − 2 \|\nabla\mathbf{b}_{m-1}\|_{L^\infty(\mathbb{R}\times\mathbb{T}^2)}\le C\varepsilon_{m-1}^{\beta-2} ∥∇ b m − 1 ∥ L ∞ ( R × T 2 ) ≤ C ε m − 1 β − 2 with an absolute constant C C C , and the renewal time is short on the time scale ε m − 1 β − 2 \varepsilon_{m-1}^{\beta-2} ε m − 1 β − 2 :
ε m − 1 β − 2 h m ≤ 2 − 25 ε m − 1 2 δ . (15) \varepsilon_{m-1}^{\beta-2}h_m\le2^{-25}\varepsilon_{m-1}^{2\delta}. \tag*{(15)} ε m − 1 β − 2 h m ≤ 2 − 25 ε m − 1 2 δ . (15)
Within a block, the gradient of the flow of b m − 1 \mathbf{b}_{m-1} b m − 1 differs from the identity by at most C ε m − 1 2 δ C\varepsilon_{m-1}^{2\delta} C ε m − 1 2 δ , with an absolute constant C C C , so the flow barely deforms the shear [5 Corollary 2.4] . The coordinates change only at the block boundaries, where the slow cutoff has switched the pulses off.
The velocity fields b m \mathbf{b}_m b m are smooth, Z 2 \mathbb{Z}^2 Z 2 -periodic in space and 1 1 1 -periodic in time. The stream functions ϕ m \phi_m ϕ m converge to a stream function ϕ \phi ϕ with ∇ ⊥ ϕ = b \nabla^\perp\phi=\mathbf{b} ∇ ⊥ ϕ = b , where b \mathbf{b} b is the pointwise limit of the b m \mathbf{b}_m b m , and there exists C ( α , q , ζ , Λ ) < ∞ C(\alpha,q,\zeta,\Lambda)<\infty C ( α , q , ζ , Λ ) < ∞ such that, for every m ≥ 0 m\ge0 m ≥ 0 ,
∥ ϕ − ϕ m ∥ L ∞ ( R × T 2 ) ≤ C ε m + 1 β , ∥ b − b m ∥ L ∞ ( R × T 2 ) ≤ C ε m + 1 α . (16) \|\phi-\phi_m\|_{L^\infty(\mathbb{R}\times\mathbb{T}^2)}\le C\varepsilon_{m+1}^{\beta},\qquad\|\mathbf{b}-\mathbf{b}_m\|_{L^\infty(\mathbb{R}\times\mathbb{T}^2)}\le C\varepsilon_{m+1}^{\alpha}. \tag*{(16)} ∥ ϕ − ϕ m ∥ L ∞ ( R × T 2 ) ≤ C ε m + 1 β , ∥ b − b m ∥ L ∞ ( R × T 2 ) ≤ C ε m + 1 α . (16)
The velocity field b \mathbf{b} b is bounded, Z 2 \mathbb{Z}^2 Z 2 -periodic in space, 1 1 1 -periodic in time, divergence free, and Hölder continuous with exponent α \alpha α jointly in time and space: there exists C ( α , q , ζ , Λ ) < ∞ C(\alpha,q,\zeta,\Lambda)<\infty C ( α , q , ζ , Λ ) < ∞ such that, for every t , t ′ ∈ R t,t'\in\mathbb{R} t , t ′ ∈ R and x , y ∈ R 2 x,y\in\mathbb{R}^2 x , y ∈ R 2 ,
∣ b ( t , x ) − b ( t ′ , y ) ∣ ≤ C ( ∣ t − t ′ ∣ + ∣ x − y ∣ ) α . (17) |\mathbf{b}(t,x)-\mathbf{b}(t',y)|\le C(|t-t'|+|x-y|)^\alpha. \tag*{(17)} ∣ b ( t , x ) − b ( t ′ , y ) ∣ ≤ C ( ∣ t − t ′ ∣ + ∣ x − y ∣ ) α . (17)
This is the regularity stated in [5 (1.3)] for the choice of q q q in [5 ] . This b \mathbf{b} b is the velocity field of Theorem A; it depends on α , q , Λ , ζ \alpha,q,\Lambda,\zeta α , q , Λ , ζ .
Effective diffusivities
For ν > 0 \nu>0 ν > 0 , the averaged flux of the level-m m m pulses at diffusivity ν \nu ν is the matrix
J m ν ( t ) : = ν I + 2 π 2 ε m β − 2 ∑ k ∈ 2 Z + 1 ζ ~ m , k ( t ) ( ∫ − ∞ t exp ( − 4 π 2 ν ( t − t ′ ) ε m 2 ) ζ ~ m , k ( t ′ ) d t ′ ) e ( k ) ⊗ e ( k ) . (18) \mathbf{J}_m^\nu(t):=\nu I+2\pi^2\varepsilon_m^{\beta-2}\sum_{k\in2\mathbb{Z}+1}\widetilde{\zeta}_{m,k}(t)\left(\int_{-\infty}^{t}\exp\left(-\frac{4\pi^2\nu(t-t')}{\varepsilon_m^2}\right)\widetilde{\zeta}_{m,k}(t')\,\mathrm{d}t'\right)e_{(k)}\otimes e_{(k)}. \tag*{(18)} J m ν ( t ) := ν I + 2 π 2 ε m β − 2 k ∈ 2 Z + 1 ∑ ζ m , k ( t ) ( ∫ − ∞ t exp ( − ε m 2 4 π 2 ν ( t − t ′ ) ) ζ m , k ( t ′ ) d t ′ ) e ( k ) ⊗ e ( k ) . (18)
where e ( k ) : = e 2 e_{(k)}:=e_2 e ( k ) := e 2 for k ∈ 4 Z + 1 k\in4\mathbb{Z}+1 k ∈ 4 Z + 1 and e ( k ) : = e 1 e_{(k)}:=e_1 e ( k ) := e 1 for k ∈ 4 Z + 3 k\in4\mathbb{Z}+3 k ∈ 4 Z + 3 . This is the spatial average of the flux of the correctors of the level-m m m pulses [5 (3.15)] , written explicitly as in [5 (3.17)–(3.18)] : the shear dispersion of the pulses. Pulse k k k enhances diffusion in the direction e ( k ) e_{(k)} e ( k ) of its flow, and the time integral averages its envelope over the past at the rate 4 π 2 ν / ε m 2 4\pi^2\nu/\varepsilon_m^2 4 π 2 ν / ε m 2 at which the diffusivity ν \nu ν damps the modes sin ( 2 π y i / ε m ) \sin(2\pi y_i/\varepsilon_m) sin ( 2 π y i / ε m ) . For a pulse held on forever, the time integral equals ε m 2 / ( 4 π 2 ν ) \varepsilon_m^2/(4\pi^2\nu) ε m 2 / ( 4 π 2 ν ) and the enhancement is ε m 2 / ( 2 ν ) \varepsilon_m^2/(2\nu) ε m 2 / ( 2 ν ) , Taylor’s dispersion formula for the steady shear ψ m , k \psi_{m,k} ψ m , k . The matrix J m ν ( t ) \mathbf{J}_m^\nu(t) J m ν ( t ) is diagonal, satisfies
ν I ≤ J m ν ( t ) ≤ ( ν + ε m 2 β 2 ν ) I , \nu I\le\mathbf{J}_m^\nu(t)\le\left(\nu+\frac{\varepsilon_m^{2\beta}}{2\nu}\right)I, ν I ≤ J m ν ( t ) ≤ ( ν + 2 ν ε m 2 β ) I ,
and equals ν I \nu I ν I when no pulse is active. Since the number h m / τ m h_m/\tau_m h m / τ m of slots in a block is odd, a shift of time by 2 h m 2h_m 2 h m maps slot k k k to slot k + 2 h m / τ m ≡ k + 2 ( m o d 4 ) k+2h_m/\tau_m\equiv k+2\pmod4 k + 2 h m / τ m ≡ k + 2 ( mod 4 ) and exchanges the two diagonal entries; a shift by 4 h m 4h_m 4 h m leaves the matrix unchanged. Its average over every interval of length 4 h m 4h_m 4 h m is the same scalar matrix [5 (3.16)] : for every t ∈ R t \in\mathbb{R} t ∈ R ,
∫ t t + 4 h m J m ′ ( t ′ ) d t ′ = K ‾ m ( ν ) I , K ‾ m ( ν ) : = 1 2 ∫ 0 4 h m tr J m ′ ( t ′ ) d t ′ . (19) \int_t^{t+4h_m} J_m^\prime(t^\prime)\,\mathrm{d}t^\prime=\overline{K}_m(\nu)\,I,\qquad\overline{K}_m(\nu):=\frac{1}{2}\int_0^{4h_m}\operatorname{tr}J_m^\prime(t^\prime)\,\mathrm{d}t^\prime. \tag*{(19)} ∫ t t + 4 h m J m ′ ( t ′ ) d t ′ = K m ( ν ) I , K m ( ν ) := 2 1 ∫ 0 4 h m tr J m ′ ( t ′ ) d t ′ . (19)
The permissible diffusivities are the union K : = ⋃ M ≥ 1 K M \mathcal{K}:=\bigcup_{M\geq1}\mathcal{K}_M K := ⋃ M ≥ 1 K M of the bands [5 (3.43)]
K M : = [ 1 2 ε M β − γ , 2 ε M β − γ ] , (20) \mathcal{K}_M:=\left[\frac{1}{2}\varepsilon_M^{\beta-\gamma},2\varepsilon_M^{\beta-\gamma}\right], \tag*{(20)} K M := [ 2 1 ε M β − γ , 2 ε M β − γ ] , (20)
which are disjoint and accumulate at 0 0 0 . For κ ∈ K M \kappa\in\mathcal{K}_M κ ∈ K M , we call M M M the terminal level , and the effective diffusivities κ M , κ M − 1 , … , κ 0 \kappa_M,\kappa_{M-1},\ldots,\kappa_0 κ M , κ M − 1 , … , κ 0 are defined by the recursion [5 (3.42)]
κ M : = κ , κ m − 1 : = K ‾ m ( κ m ) , m = M , M − 1 , … , 1. (21) \kappa_M:=\kappa,\qquad\kappa_{m-1}:=\overline{K}_m(\kappa_m),\qquad m=M,M-1,\ldots,1. \tag*{(21)} κ M := κ , κ m − 1 := K m ( κ m ) , m = M , M − 1 , … , 1. (21)
Thus κ m − 1 \kappa_{m-1} κ m − 1 is the diffusivity the solution feels once the pulses of level m m m , at diffusivity κ m \kappa_m κ m , have been averaged out.
Proposition 2.1 (Effective diffusivities and scales). There exist Λ 0 ( α , q , ζ ) ≥ 128 \Lambda_0(\alpha,q,\zeta)\geq128 Λ 0 ( α , q , ζ ) ≥ 128 and C ( α , q , ζ ) < ∞ C(\alpha,q,\zeta)<\infty C ( α , q , ζ ) < ∞ such that, for every Λ ≥ Λ 0 \Lambda\geq\Lambda_0 Λ ≥ Λ 0 , M ≥ 1 M\geq1 M ≥ 1 , κ ∈ K M \kappa\in\mathcal{K}_M κ ∈ K M and 1 ≤ m ≤ M 1\leq m\leq M 1 ≤ m ≤ M , the following hold.
(a) At each level the recursion adds about 9 80 ε m 2 β κ m \frac{9}{80}\frac{\varepsilon_m^{2\beta}}{\kappa_m} 80 9 κ m ε m 2 β :
∣ κ m − 1 − κ m − 9 80 ε m 2 β κ m ∣ ≤ C ε m 2 β κ m ( ε m 2 κ m τ m + ε m δ m ) . \left|\kappa_{m-1}-\kappa_m-\frac{9}{80}\frac{\varepsilon_m^{2\beta}}{\kappa_m}\right|\leq C\frac{\varepsilon_m^{2\beta}}{\kappa_m}\left(\frac{\varepsilon_m^2}{\kappa_m\tau_m}+\varepsilon_m^{\delta_m}\right). κ m − 1 − κ m − 80 9 κ m ε m 2 β ≤ C κ m ε m 2 β ( κ m τ m ε m 2 + ε m δ m ) . (b) If m < M m<M m < M , then κ m ≍ ε m β + γ \kappa_m\asymp\varepsilon_m^{\beta+\gamma} κ m ≍ ε m β + γ .
(c) If m ≥ 2 m\geq2 m ≥ 2 , then
κ m ≤ κ m − 1 , ε m 2 κ m τ m ≤ C ε m − 1 4 δ , ( κ m − 1 h m ) 1 / 2 ≍ ε m − 1 1 + γ / 2 + δ , ( κ m − 1 κ m ) 1 / 2 ≤ C ε m − 1 − q γ . \kappa_m\leq\kappa_{m-1},\qquad\frac{\varepsilon_m^2}{\kappa_m\tau_m}\leq C\varepsilon_{m-1}^{4\delta},\qquad(\kappa_{m-1}h_m)^{1/2}\asymp\varepsilon_{m-1}^{1+\gamma/2+\delta},\qquad\left(\frac{\kappa_{m-1}}{\kappa_m}\right)^{1/2}\leq C\varepsilon_{m-1}^{-q\gamma}. κ m ≤ κ m − 1 , κ m τ m ε m 2 ≤ C ε m − 1 4 δ , ( κ m − 1 h m ) 1/2 ≍ ε m − 1 1 + γ /2 + δ , ( κ m κ m − 1 ) 1/2 ≤ C ε m − 1 − q γ . If moreover m < M m<M m < M , then ( κ m − 1 / κ m ) 1 / 2 ≍ ε m − 1 − q γ (\kappa_{m-1}/\kappa_m)^{1/2}\asymp\varepsilon_{m-1}^{-q\gamma} ( κ m − 1 / κ m ) 1/2 ≍ ε m − 1 − q γ , and, if instead m = M m=M m = M , then κ m − 1 ≤ C κ m \kappa_{m-1}\leq C\kappa_m κ m − 1 ≤ C κ m .
Part (a) is [5 Lemma 3.3] , and part (b) is the bound (3.45) of [5 Lemma 3.4] for 1 ≤ m < M 1\leq m<M 1 ≤ m < M , both for the choice of q q q in [5 ] . Idea of the proof. (a) averages (18) over a period: a quarter of the slots enhances each direction, each by about ∫ R ζ ( t ) 2 d t = 9 10 \int_{\mathbb{R}}\zeta(t)^2\,\mathrm{d}t=\frac{9}{10} ∫ R ζ ( t ) 2 d t = 10 9 times Taylor’s enhancement ε m 2 β / ( 2 κ m ) \varepsilon_m^{2\beta}/(2\kappa_m) ε m 2 β / ( 2 κ m ) , which gives 9 80 \frac{9}{80} 80 9 . Part (b) follows by induction on m m m , going up the levels from κ M ∈ K M \kappa_M\in\mathcal{K}_M κ M ∈ K M . Part (c) follows from (b), (20) , (7) , the inequality K ‾ m ( ν ) ≥ ν \overline{K}_m(\nu)\geq\nu K m ( ν ) ≥ ν of Lemma 7.10(iii) and Lemma 7.1(a), (c).
Let M ≥ 2 M\geq2 M ≥ 2 and κ ∈ K M \kappa\in\mathcal{K}_M κ ∈ K M . Since q ( β − γ ) = β + γ q(\beta-\gamma)=\beta+\gamma q ( β − γ ) = β + γ , we have κ ≍ ε M − 1 β + γ \kappa\asymp\varepsilon_{M-1}^{\beta+\gamma} κ ≍ ε M − 1 β + γ , while κ m ≍ ε m β + γ \kappa_m\asymp\varepsilon_m^{\beta+\gamma} κ m ≍ ε m β + γ for 1 ≤ m < M 1\leq m<M 1 ≤ m < M by (b), independently of κ \kappa κ . By (c), the enhancement is strong below the terminal level, κ m − 1 ≫ κ m \kappa_{m-1}\gg\kappa_m κ m − 1 ≫ κ m for 2 ≤ m < M 2\leq m<M 2 ≤ m < M , and bounded at level M M M , κ M ≤ κ M − 1 ≤ C κ M \kappa_M\leq\kappa_{M-1}\leq C\kappa_M κ M ≤ κ M − 1 ≤ C κ M .
The scales of one level. Let 2 ≤ m < M 2\leq m<M 2 ≤ m < M . On the time scale ε m − 1 2 − β \varepsilon_{m-1}^{2-\beta} ε m − 1 2 − β of the coarse velocity field b m − 1 \mathbf{b}_{m-1} b m − 1 (before (15) ), the cell diffusion time ε m 2 / κ m \varepsilon_m^2/\kappa_m ε m 2 / κ m of level m m m , the pulse time, the margin time and the renewal time are ε m − 1 δ 8 \varepsilon_{m-1}^{\delta_8} ε m − 1 δ 8 , ε m − 1 δ 4 \varepsilon_{m-1}^{\delta_4} ε m − 1 δ 4 , ε m − 1 δ 3 \varepsilon_{m-1}^{\delta_3} ε m − 1 δ 3 and ε m − 1 δ 2 \varepsilon_{m-1}^{\delta_2} ε m − 1 δ 2 times this time scale:
ε m 2 κ m ≍ ε m − 1 2 − β + 8 δ , τ m ≍ ε m − 1 2 − β + 4 δ , τ m ′ ≍ ε m − 1 2 − β + 3 δ , h m ≍ ε m − 1 2 − β + 2 δ , \frac{\varepsilon_m^2}{\kappa_m}\asymp\varepsilon_{m-1}^{2-\beta+8\delta},\qquad\tau_m\asymp\varepsilon_{m-1}^{2-\beta+4\delta},\qquad\tau_m^\prime\asymp\varepsilon_{m-1}^{2-\beta+3\delta},\qquad h_m\asymp\varepsilon_{m-1}^{2-\beta+2\delta}, κ m ε m 2 ≍ ε m − 1 2 − β + 8 δ , τ m ≍ ε m − 1 2 − β + 4 δ , τ m ′ ≍ ε m − 1 2 − β + 3 δ , h m ≍ ε m − 1 2 − β + 2 δ ,
by (7) , Proposition 2.1(b) and the identity q ( 2 − β − γ ) = 2 − β + 8 δ q(2-\beta-\gamma)=2-\beta+8\delta q ( 2 − β − γ ) = 2 − β + 8 δ of Lemma 7.1(c). The effective diffusion length over one renewal time, ρ ( m ) : = ( κ m − 1 h m ) 1 / 2 \rho(m):=(\kappa_{m-1}h_m)^{1/2} ρ ( m ) := ( κ m − 1 h m ) 1/2 , lies between the two wavelengths:
ε m ≍ ε m − 1 q γ + 3 δ ρ ( m ) , ρ ( m ) ≍ ε m − 1 1 + γ / 2 + δ , so ε m ≪ ρ ( m ) ≪ ε m − 1 , \varepsilon_m\asymp\varepsilon_{m-1}^{q\gamma+3\delta}\rho(m),\qquad\rho(m)\asymp\varepsilon_{m-1}^{1+\gamma/2+\delta},\qquad\text{so}\qquad\varepsilon_m\ll\rho(m)\ll\varepsilon_{m-1}, ε m ≍ ε m − 1 q γ + 3 δ ρ ( m ) , ρ ( m ) ≍ ε m − 1 1 + γ /2 + δ , so ε m ≪ ρ ( m ) ≪ ε m − 1 ,
by Proposition 2.1(c) and the identity q = 1 + 4 δ + ( q + 1 2 ) γ q=1+4\delta+(q+\frac{1}{2})\gamma q = 1 + 4 δ + ( q + 2 1 ) γ , which is Lemma 7.1(c) with 2 − β = 1 − α 2-\beta=1-\alpha 2 − β = 1 − α .
Why the time scales have these sizes. Three requirements fix them. The renewal time h m h_m h m is short on the time scale ε m − 1 2 − β \varepsilon_{m-1}^{2-\beta} ε m − 1 2 − β of b m − 1 \mathbf{b}_{m-1} b m − 1 , by (15) , so that the Lagrangian coordinates stay close to rigid. The pulse time τ m \tau_m τ m is long compared with the cell diffusion time ε m 2 / κ m \varepsilon_m^2/\kappa_m ε m 2 / κ m of level m m m , by Proposition 2.1(c), so that each pulse homogenizes. The margins of length τ m ′ \tau'_m τ m ′ , where the coordinates are renewed, are long compared with a pulse and short compared with a block.
In the notation of [5 ] , the renewal time h m h_m h m is τ m ′ ′ \tau''_m τ m ′′ , the number ε m β − 2 \varepsilon_m^{\beta-2} ε m β − 2 is a m a_m a m , the label l m , k l_{m,k} l m , k is l k l_k l k , and the envelope ζ ~ m , k \widetilde{\zeta}_{m,k} ζ m , k is the product ζ ^ m , l k ζ m , k \widehat{\zeta}_{m,l_k}\zeta_{m,k} ζ m , l k ζ m , k of the two cutoffs. The argument X m − 1 ( l h m , x , t ) X_{m-1}(lh_m,x,t) X m − 1 ( l h m , x , t ) of the shear is written X − 1 ( t , x , l h m ) X^{-1}(t,x,lh_m) X − 1 ( t , x , l h m ) there, and the matrix J m ν \mathbf{J}_m^\nu J m ν and the number K ‾ m ( ν ) \overline{K}_m(\nu) K m ( ν ) are written J m κ \mathbf{J}_m^\kappa J m κ and K ‾ m κ \overline{K}_m^\kappa K m κ , with κ = ν \kappa=\nu κ = ν . The effective diffusivities are the renormalized diffusivities of [5 ] , and the exponent β − γ \beta-\gamma β − γ of the bands is written 2 β q + 1 \frac{2\beta}{q+1} q + 1 2 β there. Section 7 proves the properties stated in this section: Lemma 7.1 the relations between the scales, (7) , (15) and the identities between the exponents; Proposition 7.3 the bound on ∇ b m − 1 \nabla b_{m-1} ∇ b m − 1 , (16) and (17) ; Lemma 7.7(a) the bound on the gradient of the flow within a block, for m ≥ 2 m\ge2 m ≥ 2 ; Lemma 7.10 the properties of J m ν \mathbf{J}_m^\nu J m ν ; and Proposition 2.1 is proved in Subsection 7.5.
Main results
For a Z 2 \mathbb{Z}^2 Z 2 -periodic function g g g we write
∥ g ∥ C 0 , s ( T 2 ) : = sup T 2 ∣ g ∣ + [ g ] C 0 , s ( T 2 ) , [ g ] C 0 , s ( T 2 ) : = sup x ≠ y ∣ g ( x ) − g ( y ) ∣ ∣ x − y ∣ T 2 s . (22) \|g\|_{C^{0,s}(\mathbb{T}^2)} := \sup_{\mathbb{T}^2}|g| + [g]_{C^{0,s}(\mathbb{T}^2)}, \qquad[g]_{C^{0,s}(\mathbb{T}^2)} := \sup_{x\ne y}\frac{|g(x)-g(y)|}{|x-y|_{\mathbb{T}^2}^{s}}. \tag*{(22)} ∥ g ∥ C 0 , s ( T 2 ) := T 2 sup ∣ g ∣ + [ g ] C 0 , s ( T 2 ) , [ g ] C 0 , s ( T 2 ) := x = y sup ∣ x − y ∣ T 2 s ∣ g ( x ) − g ( y ) ∣ . (22)
for the Hölder norm and seminorm, where ∣ x − y ∣ T 2 |x-y|_{\mathbb{T}^2} ∣ x − y ∣ T 2 is the distance on the torus. A weak solution of the initial-value problem below is a function θ ∈ L ∞ ( ( t i n , t i n + T ) ; L 2 ( T 2 ) ) \theta\in L^\infty((t_{\mathrm{in}},t_{\mathrm{in}}+T);L^2(\mathbb{T}^2)) θ ∈ L ∞ (( t in , t in + T ) ; L 2 ( T 2 )) with ∇ θ ∈ L 2 ( ( t i n , t i n + T ) × T 2 ) \nabla\theta\in L^2((t_{\mathrm{in}},t_{\mathrm{in}}+T)\times\mathbb{T}^2) ∇ θ ∈ L 2 (( t in , t in + T ) × T 2 ) , weakly continuous from [ t i n , t i n + T ] [t_{\mathrm{in}},t_{\mathrm{in}}+T] [ t in , t in + T ] to L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) , which solves the equation in the sense of distributions and takes the initial value. A weak solution on ( t i n , ∞ ) (t_{\mathrm{in}},\infty) ( t in , ∞ ) is a function that, for every T > 0 T>0 T > 0 , is a weak solution on ( t i n , t i n + T ) (t_{\mathrm{in}},t_{\mathrm{in}}+T) ( t in , t in + T ) .
Theorem A (Interior regularity). Let α ∈ ( 0 , 1 / 3 ) \alpha\in(0,1/3) α ∈ ( 0 , 1/3 ) , s ∈ ( 0 , 1 2 ( 1 − α ) ) s\in(0,\frac{1}{2}(1-\alpha)) s ∈ ( 0 , 2 1 ( 1 − α )) , q q q satisfy (3) , and let ζ \zeta ζ be an even function in C ∞ ( ( − 2 3 , 2 3 ) ) C^\infty((-\frac{2}{3},\frac{2}{3})) C ∞ (( − 3 2 , 3 2 )) satisfying (4) . There exists Λ 0 ( α , s , q , ζ ) ≥ 128 \Lambda_0(\alpha,s,q,\zeta)\ge128 Λ 0 ( α , s , q , ζ ) ≥ 128 such that, for every integer Λ ≥ Λ 0 \Lambda\ge\Lambda_0 Λ ≥ Λ 0 , the velocity field b \mathbf{b} b of Section 2 has the following property. For every 0 < t 0 < T 0<t_0<T 0 < t 0 < T there exists C ( α , s , q , ζ , Λ , t 0 , T ) < ∞ C(\alpha,s,q,\zeta,\Lambda,t_0,T)<\infty C ( α , s , q , ζ , Λ , t 0 , T ) < ∞ such that, for every t i n ∈ R t_{\mathrm{in}}\in\mathbb{R} t in ∈ R , κ ∈ K \kappa\in\mathcal{K} κ ∈ K and θ i n ∈ L 2 ( T 2 ) \theta_{\mathrm{in}}\in L^2(\mathbb{T}^2) θ in ∈ L 2 ( T 2 ) , every weak solution θ \theta θ of the initial-value problem
∂ t θ + b ⋅ ∇ θ − κ Δ θ = 0 in ( t i n , t i n + T ) × T 2 , θ ( t i n , ⋅ ) = θ i n on T 2 (23) \begin{aligned}
\partial_t\theta+\mathbf{b}\cdot\nabla\theta-\kappa\Delta\theta&=0 &&\text{in }(t_{\mathrm{in}},t_{\mathrm{in}}+T)\times\mathbb{T}^2,\\
\theta(t_{\mathrm{in}},\cdot)&=\theta_{\mathrm{in}} &&\text{on }\mathbb{T}^2 \tag*{(23)}
\end{aligned} ∂ t θ + b ⋅ ∇ θ − κ Δ θ θ ( t in , ⋅ ) = 0 = θ in in ( t in , t in + T ) × T 2 , on T 2 (23) satisfies
∥ θ ∥ L ∞ ( [ t i n + t 0 , t i n + T ] ; C 0 , s ( T 2 ) ) ≤ C ∥ θ i n ∥ L 2 ( T 2 ) . (24) \|\theta\|_{L^\infty([t_{\mathrm{in}}+t_0,t_{\mathrm{in}}+T];C^{0,s}(\mathbb{T}^2))}\le C\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)}. \tag*{(24)} ∥ θ ∥ L ∞ ([ t in + t 0 , t in + T ] ; C 0 , s ( T 2 )) ≤ C ∥ θ in ∥ L 2 ( T 2 ) . (24) The constant C C C is independent of κ \kappa κ , and 1 2 ( 1 − α ) \frac{1}{2}(1-\alpha) 2 1 ( 1 − α ) is the Obukhov–Corrsin exponent of a C 0 , α C^{0,\alpha} C 0 , α velocity field [44 , 18 ] ; Subsection 5.1 explains where it enters the proof.
Theorem A also implies that the variance is dissipated at a rate independent of κ ∈ K \kappa\in\mathcal{K} κ ∈ K .
Theorem B (Uniform anomalous dissipation). Let α \alpha α , s s s , q q q , ζ \zeta ζ and Λ 0 \Lambda_0 Λ 0 be as in Theorem A, and let Λ ≥ Λ 0 \Lambda\ge\Lambda_0 Λ ≥ Λ 0 be an integer. For the velocity field b \mathbf{b} b of Section 2, there exists c ( α , s , q , ζ , Λ ) > 0 c(\alpha,s,q,\zeta,\Lambda)>0 c ( α , s , q , ζ , Λ ) > 0 such that, for every t i n ∈ R t_{\mathrm{in}}\in\mathbb{R} t in ∈ R , κ ∈ K \kappa\in\mathcal{K} κ ∈ K and θ i n ∈ L 2 ( T 2 ) \theta_{\mathrm{in}}\in L^2(\mathbb{T}^2) θ in ∈ L 2 ( T 2 ) , every weak solution θ \theta θ of the initial-value problem
∂ t θ + b ⋅ ∇ θ − κ Δ θ = 0 in ( t i n , ∞ ) × T 2 , θ ( t i n , ⋅ ) = θ i n on T 2 (25) \begin{aligned}
\partial_t\theta+\mathbf{b}\cdot\nabla\theta-\kappa\Delta\theta&=0 &&\text{in }(t_{\mathrm{in}},\infty)\times\mathbb{T}^2,\\
\theta(t_{\mathrm{in}},\cdot)&=\theta_{\mathrm{in}} &&\text{on }\mathbb{T}^2 \tag*{(25)}
\end{aligned} ∂ t θ + b ⋅ ∇ θ − κ Δ θ θ ( t in , ⋅ ) = 0 = θ in in ( t in , ∞ ) × T 2 , on T 2 (25)
satisfies
κ ∥ ∇ θ ∥ L 2 ( ( t i n , t i n + 1 ) × T 2 ) 2 ≥ c ∥ θ i n − ( θ i n ) T 2 ∥ L 2 ( T 2 ) 2 (26) \kappa\lVert\nabla\theta\rVert_{L^2((t_{\mathrm{in}},t_{\mathrm{in}}+1)\times\mathbb{T}^2)}^2 \ge c \lVert\theta_{\mathrm{in}}-(\theta_{\mathrm{in}})_{\mathbb{T}^2}\rVert_{L^2(\mathbb{T}^2)}^2 \tag*{(26)} κ ∥ ∇ θ ∥ L 2 (( t in , t in + 1 ) × T 2 ) 2 ≥ c ∥ θ in − ( θ in ) T 2 ∥ L 2 ( T 2 ) 2 (26)
and, for every t ≥ t i n t \ge t_{\mathrm{in}} t ≥ t in ,
∥ θ ( t , ⋅ ) − ( θ i n ) T 2 ∥ L 2 ( T 2 ) ≤ 2 exp ( − c ( t − t i n ) ) ∥ θ i n − ( θ i n ) T 2 ∥ L 2 ( T 2 ) . (27) \lVert\theta(t,\cdot)-(\theta_{\mathrm{in}})_{\mathbb{T}^2}\rVert_{L^2(\mathbb{T}^2)} \le2\exp\bigl(-c(t-t_{\mathrm{in}})\bigr)\lVert\theta_{\mathrm{in}}-(\theta_{\mathrm{in}})_{\mathbb{T}^2}\rVert_{L^2(\mathbb{T}^2)}. \tag*{(27)} ∥ θ ( t , ⋅ ) − ( θ in ) T 2 ∥ L 2 ( T 2 ) ≤ 2 exp ( − c ( t − t in ) ) ∥ θ in − ( θ in ) T 2 ∥ L 2 ( T 2 ) . (27)
We compare Theorem B with [5 Theorem 1.1] , recalled in Subsection 5.1, which concerns the velocity field of [5 ] and gives a lower bound for lim sup κ → 0 κ ∥ ∇ θ κ ∥ L 2 ( ( 0 , 1 ) × T 2 ) 2 \limsup_{\kappa\to0}\kappa\lVert\nabla\theta^\kappa\rVert_{L^2((0,1)\times\mathbb{T}^2)}^2 lim sup κ → 0 κ ∥ ∇ θ κ ∥ L 2 (( 0 , 1 ) × T 2 ) 2 for mean-zero data in H 1 H^1 H 1 . The restriction to the bands and a lower bound for every κ \kappa κ in them are already in [5 ] : the discussion after [5 Theorem 1.1] states the lower bound for every κ \kappa κ in an explicit union of bands, the permissible diffusivities [5 (3.43)] , with a constant depending on the datum. For the velocity field of this paper, Theorem B gains three things: the datum is only in L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) ; the constant c c c is independent of the datum and of the initial time t i n t_{\mathrm{in}} t in , whereas the constant in [5 ] depends on the ratio ∥ θ i n ∥ L 2 ( T 2 ) / ∥ ∇ θ i n ∥ L 2 ( T 2 ) \lVert\theta_{\mathrm{in}}\rVert_{L^2(\mathbb{T}^2)}/\lVert\nabla\theta_{\mathrm{in}}\rVert_{L^2(\mathbb{T}^2)} ∥ θ in ∥ L 2 ( T 2 ) / ∥ ∇ θ in ∥ L 2 ( T 2 ) ; and the variance decays exponentially, at a rate independent of κ \kappa κ . As a consequence, since the bands accumulate at 0 0 0 , for every mean-zero θ i n ∈ L 2 ( T 2 ) \theta_{\mathrm{in}}\in L^2(\mathbb{T}^2) θ in ∈ L 2 ( T 2 ) the weak solutions θ κ \theta^\kappa θ κ of (25) with t i n = 0 t_{\mathrm{in}}=0 t in = 0 , one for each κ ∈ K \kappa\in\mathcal{K} κ ∈ K , which are unique (§ 5 ), satisfy
lim inf κ → 0 , κ ∈ K κ ∥ ∇ θ κ ∥ L 2 ( ( 0 , 1 ) × T 2 ) 2 ≥ inf κ ∈ K κ ∥ ∇ θ κ ∥ L 2 ( ( 0 , 1 ) × T 2 ) 2 ≥ c ∥ θ i n ∥ L 2 ( T 2 ) 2 . \liminf_{\kappa\to0,\ \kappa\in\mathcal{K}}\kappa\lVert\nabla\theta^\kappa\rVert_{L^2((0,1)\times\mathbb{T}^2)}^2
\ge
\inf_{\kappa\in\mathcal{K}}\kappa\lVert\nabla\theta^\kappa\rVert_{L^2((0,1)\times\mathbb{T}^2)}^2
\ge
c\lVert\theta_{\mathrm{in}}\rVert_{L^2(\mathbb{T}^2)}^2. κ → 0 , κ ∈ K lim inf κ ∥ ∇ θ κ ∥ L 2 (( 0 , 1 ) × T 2 ) 2 ≥ κ ∈ K inf κ ∥ ∇ θ κ ∥ L 2 (( 0 , 1 ) × T 2 ) 2 ≥ c ∥ θ in ∥ L 2 ( T 2 ) 2 .
with the constant c c c of Theorem B: a lower bound on the limit inferior, where [5 Theorem 1.1] bounds the limit superior.
Idea of the proof of Theorem B. Fix t i n t_{\mathrm{in}} t in and κ \kappa κ . The solution operator S S S , which maps θ i n \theta_{\mathrm{in}} θ in to θ ( t i n + 1 , ⋅ ) \theta(t_{\mathrm{in}}+1,\cdot) θ ( t in + 1 , ⋅ ) , is linear, positive by the maximum principle, and preserves constants and integrals. Theorem A on ( t i n , t i n + 1 ) (t_{\mathrm{in}},t_{\mathrm{in}}+1) ( t in , t in + 1 ) , with t 0 = 1 2 t_0=\frac{1}{2} t 0 = 2 1 , bounds it from L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) into C 0 , s ( T 2 ) C^{0,s}(\mathbb{T}^2) C 0 , s ( T 2 ) by a constant C ≥ 1 C\geq1 C ≥ 1 independent of t i n t_{\mathrm{in}} t in and κ ∈ K \kappa\in\mathcal{K} κ ∈ K . With N N N the least integer such that N s ≥ C N^s\geq C N s ≥ C , Lemma 9.4 gives
∥ S θ i n ∥ L 2 ( T 2 ) 2 ≤ ( 1 − 1 2 N 4 ) ∥ θ i n ∥ L 2 ( T 2 ) 2 \|S\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)}^2\leq\left(1-\frac{1}{2N^4}\right)\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)}^2 ∥ S θ in ∥ L 2 ( T 2 ) 2 ≤ ( 1 − 2 N 4 1 ) ∥ θ in ∥ L 2 ( T 2 ) 2
for every mean-zero θ i n ∈ L 2 ( T 2 ) \theta_{\mathrm{in}}\in L^2(\mathbb{T}^2) θ in ∈ L 2 ( T 2 ) , and the energy identity turns this drop into (26) , with c = 1 4 N 4 c=\frac{1}{4N^4} c = 4 N 4 1 , and into a contraction by 1 − 2 c 1-2c 1 − 2 c of ∥ θ ( t , ⋅ ) − ( θ i n ) T 2 ∥ L 2 ( T 2 ) 2 \|\theta(t,\cdot)-(\theta_{\mathrm{in}})_{\mathbb{T}^2}\|_{L^2(\mathbb{T}^2)}^2 ∥ θ ( t , ⋅ ) − ( θ in ) T 2 ∥ L 2 ( T 2 ) 2 over each unit of time, which gives (27) . The proof is in Subsection 9.6.
Reduction to a velocity field with finitely many scales
For κ ∈ K M \kappa\in\mathcal{K}_M κ ∈ K M , the levels of b \mathbf{b} b much finer than level M M M are irrelevant to the solution: their stream functions are much smaller than κ \kappa κ , and molecular diffusion averages them out. In divergence form, a perturbation of the stream function of size κ ϖ \kappa\varpi κϖ is a skew-symmetric perturbation of relative size ϖ \varpi ϖ of the diffusion matrix κ I \kappa I κ I , and it moves the solution by at most ϖ ∥ θ i n ∥ L 2 ( T 2 ) \varpi\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)} ϖ ∥ θ in ∥ L 2 ( T 2 ) in L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) (Lemma 9.2, as in the comparison of θ \theta θ with θ M \theta_M θ M in [5 Section 5.4] ). Proposition 4.2 turns this into a bound in C 0 , s ( T 2 ) C^{0,s}(\mathbb{T}^2) C 0 , s ( T 2 ) , at a cost which is bounded once the levels finer than M + n M+n M + n are removed, for a fixed number n n n of extra levels. We therefore prove the estimate for the smooth velocity field b M + n \mathbf{b}_{M+n} b M + n , which has finitely many scales (Proposition 4.1 , proved in Section 5), and transfer it to b \mathbf{b} b .
Proposition 4.1 (Regularity for the truncated velocity field). There exists Λ 0 ( α , s , q , ζ ) ≥ 128 \Lambda_0(\alpha,s,q,\zeta)\geq128 Λ 0 ( α , s , q , ζ ) ≥ 128 such that, for every integer Λ ≥ Λ 0 \Lambda\geq\Lambda_0 Λ ≥ Λ 0 , integer n ≥ 1 n\geq1 n ≥ 1 and 0 < t 0 < T 0<t_0<T 0 < t 0 < T , there exist M 1 ( α , s , q , ζ , Λ , n , t 0 , T ) ∈ N M_1(\alpha,s,q,\zeta,\Lambda,n,t_0,T)\in\mathbb{N} M 1 ( α , s , q , ζ , Λ , n , t 0 , T ) ∈ N and C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ C(\alpha,s,q,\zeta,\Lambda,n,t_0,T)<\infty C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ with the following property. For every M ≥ M 1 M\geq M_1 M ≥ M 1 , κ ∈ K M \kappa\in\mathcal{K}_M κ ∈ K M , t i n ∈ R t_{\mathrm{in}}\in\mathbb{R} t in ∈ R and θ i n ∈ L 2 ( T 2 ) \theta_{\mathrm{in}}\in L^2(\mathbb{T}^2) θ in ∈ L 2 ( T 2 ) , every weak solution θ \theta θ of
{ ∂ t θ + b M + n ⋅ ∇ θ − κ Δ θ = 0 in ( t i n , t i n + T ) × T 2 , θ ( t i n , ⋅ ) = θ i n on T 2 (28) \begin{cases}
\partial_t\theta+\mathbf{b}_{M+n}\cdot\nabla\theta-\kappa\Delta\theta=0 & \text{in }(t_{\mathrm{in}},t_{\mathrm{in}}+T)\times\mathbb{T}^2,\\
\theta(t_{\mathrm{in}},\cdot)=\theta_{\mathrm{in}} & \text{on }\mathbb{T}^2 \tag*{(28)}
\end{cases} { ∂ t θ + b M + n ⋅ ∇ θ − κ Δ θ = 0 θ ( t in , ⋅ ) = θ in in ( t in , t in + T ) × T 2 , on T 2 (28) satisfies
∥ θ ∥ L ∞ ( [ t i n + t 0 , t i n + T ] ; C 0 , s ( T 2 ) ) ≤ C ∥ θ i n ∥ L 2 ( T 2 ) . \|\theta\|_{L^\infty([t_{\mathrm{in}}+t_0,t_{\mathrm{in}}+T];C^{0,s}(\mathbb{T}^2))}\leq C\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)}. ∥ θ ∥ L ∞ ([ t in + t 0 , t in + T ] ; C 0 , s ( T 2 )) ≤ C ∥ θ in ∥ L 2 ( T 2 ) .
Proposition 4.2 (Transfer to the full velocity field). There exist ϖ 0 ( s ) > 0 \varpi_0(s)>0 ϖ 0 ( s ) > 0 and C ( s ) < ∞ C(s)<\infty C ( s ) < ∞ with the following property. Let Φ , Φ ~ : R × R 2 → R \Phi,\widetilde{\Phi}:\mathbb{R}\times\mathbb{R}^2\to\mathbb{R} Φ , Φ : R × R 2 → R be Z 2 \mathbb{Z}^2 Z 2 -periodic in x x x , with Φ \Phi Φ continuous and C 1 C^1 C 1 in x x x , ∇ ⊥ Φ \nabla^\perp\Phi ∇ ⊥ Φ bounded and continuous, and Φ ~ \widetilde{\Phi} Φ smooth. Let κ > 0 \kappa>0 κ > 0 , 0 < r ≤ 1 8 0<r\leq\frac18 0 < r ≤ 8 1 and ϖ ≥ 0 \varpi\geq0 ϖ ≥ 0 satisfy
∥ Φ − Φ ~ ∥ L ∞ ( R × T 2 ) ≤ κ ϖ , ϖ + r 2 κ ∥ ∇ 2 Φ ~ ∥ L ∞ ( R × T 2 ) ≤ ϖ 0 . \|\Phi-\widetilde{\Phi}\|_{L^\infty(\mathbb{R}\times\mathbb{T}^2)}\leq\kappa\varpi,\qquad
\varpi+\frac{r^2}{\kappa}\|\nabla^2\widetilde{\Phi}\|_{L^\infty(\mathbb{R}\times\mathbb{T}^2)}\leq\varpi_0. ∥Φ − Φ ∥ L ∞ ( R × T 2 ) ≤ κϖ , ϖ + κ r 2 ∥ ∇ 2 Φ ∥ L ∞ ( R × T 2 ) ≤ ϖ 0 . Let t i n , t 1 , t 2 ∈ R t_{\mathrm{in}},t_1,t_2\in\mathbb{R} t in , t 1 , t 2 ∈ R with t i n ≤ t 1 − r 2 / κ t_{\mathrm{in}}\leq t_1-r^2/\kappa t in ≤ t 1 − r 2 / κ and t 1 < t 2 t_1<t_2 t 1 < t 2 , let θ i n ∈ L 2 ( T 2 ) \theta_{\mathrm{in}}\in L^2(\mathbb{T}^2) θ in ∈ L 2 ( T 2 ) , and let θ \theta θ and θ ~ \widetilde{\theta} θ be weak solutions of
{ ∂ t θ + ∇ ⊥ Φ ⋅ ∇ θ − κ Δ θ = 0 , ∂ t θ ~ + ∇ ⊥ Φ ~ ⋅ ∇ θ ~ − κ Δ θ ~ = 0 in ( t i n , t 2 + r 2 / κ ) × T 2 , θ ( t i n , ⋅ ) = θ ~ ( t i n , ⋅ ) = θ i n on T 2 . \begin{cases}
\partial_t\theta+\nabla^\perp\Phi\cdot\nabla\theta-\kappa\Delta\theta=0,\quad
\partial_t\widetilde{\theta}+\nabla^\perp\widetilde{\Phi}\cdot\nabla\widetilde{\theta}-\kappa\Delta\widetilde{\theta}=0
& \text{in }(t_{\mathrm{in}},t_2+r^2/\kappa)\times\mathbb{T}^2,\\
\theta(t_{\mathrm{in}},\cdot)=\widetilde{\theta}(t_{\mathrm{in}},\cdot)=\theta_{\mathrm{in}}
& \text{on }\mathbb{T}^2.
\end{cases} { ∂ t θ + ∇ ⊥ Φ ⋅ ∇ θ − κ Δ θ = 0 , ∂ t θ + ∇ ⊥ Φ ⋅ ∇ θ − κ Δ θ = 0 θ ( t in , ⋅ ) = θ ( t in , ⋅ ) = θ in in ( t in , t 2 + r 2 / κ ) × T 2 , on T 2 . Let Ξ ≥ 0 \Xi\geq0 Ξ ≥ 0 be such that
∥ θ ~ ∥ L ∞ ( [ t 1 − r 2 / κ , t 2 + r 2 / κ ] ; C 0 , s ( T 2 ) ) ≤ Ξ . \|\widetilde{\theta}\|_{L^\infty([t_1-r^2/\kappa,t_2+r^2/\kappa];C^{0,s}(\mathbb{T}^2))}\leq\Xi. ∥ θ ∥ L ∞ ([ t 1 − r 2 / κ , t 2 + r 2 / κ ] ; C 0 , s ( T 2 )) ≤ Ξ. Then
∥ θ ∥ L ∞ ( [ t 1 , t 2 ] ; C 0 , s ( T 2 ) ) ≤ C ( Ξ + ϖ r − 1 − s ∥ θ i n ∥ L 2 ( T 2 ) ) . \|\theta\|_{L^\infty([t_1,t_2];C^{0,s}(\mathbb{T}^2))}\leq C\left(\Xi+\varpi r^{-1-s}\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)}\right). ∥ θ ∥ L ∞ ([ t 1 , t 2 ] ; C 0 , s ( T 2 )) ≤ C ( Ξ + ϖ r − 1 − s ∥ θ in ∥ L 2 ( T 2 ) ) .
The power r − 1 − s r^{-1-s} r − 1 − s converts the L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) deviation ϖ ∥ θ i n ∥ L 2 ( T 2 ) \varpi\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)} ϖ ∥ θ in ∥ L 2 ( T 2 ) into a deviation in C 0 , s ( T 2 ) C^{0,s}(\mathbb{T}^2) C 0 , s ( T 2 ) . Idea of the proof. In a cylinder of radius r r r moving with ∇ ⊥ Φ ~ \nabla^\perp\widetilde{\Phi} ∇ ⊥ Φ , after rescaling, θ \theta θ solves the equation of Lemma 4.3: Φ − Φ ~ \Phi-\widetilde{\Phi} Φ − Φ gives the skew part, of size at most ϖ \varpi ϖ , and the variation of ∇ ⊥ Φ ~ \nabla^\perp\widetilde{\Phi} ∇ ⊥ Φ gives the drift, of size at most r 2 ∥ ∇ 2 Φ ~ ∥ L ∞ ( R × T 2 ) / κ r^2\|\nabla^2\widetilde{\Phi}\|_{L^\infty(\mathbb{R}\times\mathbb{T}^2)}/\kappa r 2 ∥ ∇ 2 Φ ∥ L ∞ ( R × T 2 ) / κ . On the rescaled cylinder, the L 2 L^2 L 2 norm of θ − θ ~ \theta-\widetilde{\theta} θ − θ is at most C ϖ r − 1 ∥ θ i n ∥ L 2 ( T 2 ) C\varpi r^{-1}\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)} C ϖ r − 1 ∥ θ in ∥ L 2 ( T 2 ) , and θ ~ \widetilde{\theta} θ stays within C Ξ r s C\Xi r^s C Ξ r s of a constant, by the bound on θ ~ \widetilde{\theta} θ and its equation tested with a fixed bump in the moving frame. So Lemma 4.3 places θ ( t , ⋅ ) \theta(t,\cdot) θ ( t , ⋅ ) within C ( ϖ r − 1 ∥ θ i n ∥ L 2 ( T 2 ) + Ξ r s ) C(\varpi r^{-1}\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)}+\Xi r^s) C ( ϖ r − 1 ∥ θ in ∥ L 2 ( T 2 ) + Ξ r s ) of θ ~ ( t , ⋅ ) \widetilde{\theta}(t,\cdot) θ ( t , ⋅ ) , and bounds its increments over distances ρ < r 4 \rho<\frac{r}{4} ρ < 4 r by this quantity times ( ρ / r ) ( 1 + s ) / 2 (\rho/r)^{(1+s)/2} ( ρ / r ) ( 1 + s ) /2 ; since 1 2 ( 1 + s ) > s \frac{1}{2}(1+s)>s 2 1 ( 1 + s ) > s , the two bounds give the bound in C 0 , s ( T 2 ) C^{0,s}(\mathbb{T}^2) C 0 , s ( T 2 ) at all distances (Step 4 of the proof in Subsection 9.2).
Lemma 4.3 (Hölder estimate for perturbed heat equations). There exist ϖ 1 ( s ) > 0 \varpi_1(s)>0 ϖ 1 ( s ) > 0 and C ( s ) < ∞ C(s)<\infty C ( s ) < ∞ with the following property. Let D 1 : = { x ∈ R 2 : ∣ x ∣ < 1 } D_1:=\{x\in\mathbb{R}^2:|x|<1\} D 1 := { x ∈ R 2 : ∣ x ∣ < 1 } and D 1 / 2 : = { x ∈ R 2 : ∣ x ∣ < 1 2 } D_{1/2}:=\{x\in\mathbb{R}^2:|x|<\frac{1}{2}\} D 1/2 := { x ∈ R 2 : ∣ x ∣ < 2 1 } , and let [ ⋅ ] C p a r ( ( − 1 4 , 0 ] × D 1 / 2 ) 0 , ( 1 + s ) / 2 [\cdot]_{C^{0,(1+s)/2}_{\mathrm{par}((-\frac{1}{4},0]\times D_{1/2})}} [ ⋅ ] C par (( − 4 1 , 0 ] × D 1/2 ) 0 , ( 1 + s ) /2 be the Hölder seminorm on ( − 1 4 , 0 ] × D 1 / 2 (-\frac{1}{4},0]\times D_{1/2} ( − 4 1 , 0 ] × D 1/2 for the parabolic distance ∣ x − y ∣ + ∣ t − t ′ ∣ 1 / 2 |x-y|+|t-t'|^{1/2} ∣ x − y ∣ + ∣ t − t ′ ∣ 1/2 . Let a a a and b ∗ b_* b ∗ be measurable functions on ( − 1 , 0 ) × D 1 (-1,0)\times D_1 ( − 1 , 0 ) × D 1 with a − I a-I a − I skew-symmetric, b ∗ ( t , ⋅ ) b_*(t,\cdot) b ∗ ( t , ⋅ ) divergence free in the sense of distributions for almost every t t t , and ∣ a − I ∣ + ∣ b ∗ ∣ ≤ ϖ 1 |a-I|+|b_*|\leq\varpi_1 ∣ a − I ∣ + ∣ b ∗ ∣ ≤ ϖ 1 almost everywhere, and let U ∈ L ∞ ( ( − 1 , 0 ) ; L 2 ( D 1 ) ) U\in L^\infty((-1,0);L^2(D_1)) U ∈ L ∞ (( − 1 , 0 ) ; L 2 ( D 1 )) with ∇ U ∈ L 2 ( ( − 1 , 0 ) × D 1 ) \nabla U\in L^2((-1,0)\times D_1) ∇ U ∈ L 2 (( − 1 , 0 ) × D 1 ) be a weak solution of
∂ t U − ∇ ⋅ ( a ∇ U ) + b ∗ ⋅ ∇ U = 0 in ( − 1 , 0 ) × D 1 . \partial_t U-\nabla\cdot(a\nabla U)+b_*\cdot\nabla U=0\quad\text{in }(-1,0)\times D_1. ∂ t U − ∇ ⋅ ( a ∇ U ) + b ∗ ⋅ ∇ U = 0 in ( − 1 , 0 ) × D 1 . Then U ( t , ⋅ ) U(t,\cdot) U ( t , ⋅ ) converges in L 2 ( D 1 / 2 ) L^2(D_{1/2}) L 2 ( D 1/2 ) as t ↑ 0 t\uparrow0 t ↑ 0 , and, with this limit as U ( 0 , ⋅ ) U(0,\cdot) U ( 0 , ⋅ ) ,
[U]_{C^{0,(1+s)/2}_{\mathrm{par}((-\frac{1}{4},0]\times D_{1/2})}}+\|U-(U)_{(-1,0)\times D_1}\|_{L^\infty((-\frac{1}{4},0]\times D_{1/2})}
\leq C\left(\fint_{(-1,0)\times D_1}|U(t,x)-(U)_{(-1,0)\times D_1}|^2\,\mathrm{d}x\,\mathrm{d}t\right)^{1/2}.
Idea of the proof. Since a − I a-I a − I and b ∗ b_* b ∗ are small, U U U is close in L 2 L^2 L 2 to a caloric function, a solution of the heat equation, on every smaller cylinder, and Campanato’s characterization of Hölder continuity by the decay of these L 2 L^2 L 2 distances gives the bound. The proof is in Subsection 9.1.
Idea of the proof of Theorem A. For κ ∈ K M \kappa\in\mathcal{K}_M κ ∈ K M with M M M large, we continue θ \theta θ to ( t i n , t i n + T + 1 ) (t_{\mathrm{in}},t_{\mathrm{in}}+T+1) ( t in , t in + T + 1 ) , as Proposition 4.2 uses the solutions after t 2 t_2 t 2 , and compare it with the solution θ ~ \widetilde{\theta} θ of (28) , with Φ = ϕ \Phi=\phi Φ = ϕ and Φ ~ = ϕ M + n \widetilde{\Phi}=\phi_{M+n} Φ = ϕ M + n ; Proposition 4.1 bounds θ ~ \widetilde{\theta} θ in L ∞ ( [ t i n + 1 2 t 0 , t i n + T + 1 ] ; C 0 , s ( T 2 ) ) L^\infty([t_{\mathrm{in}}+\frac{1}{2}t_0,t_{\mathrm{in}}+T+1];C^{0,s}(\mathbb{T}^2)) L ∞ ([ t in + 2 1 t 0 , t in + T + 1 ] ; C 0 , s ( T 2 )) . By (16) and Proposition 7.3 (ii), we may take ϖ \varpi ϖ a multiple of ε M + n + 1 β / κ \varepsilon^{\beta}_{M+n+1}/\kappa ε M + n + 1 β / κ and r 2 r^2 r 2 a small multiple of κ ε M 2 − β \kappa\varepsilon^{2-\beta}_M κ ε M 2 − β , and a choice of n n n depending only on α , s , q \alpha,s,q α , s , q makes ϖ r − 1 − s \varpi r^{-1-s} ϖ r − 1 − s bounded (Step 2 of the proof in Subsection 9.3). The remaining finitely many bands lie in a compact subinterval of ( 0 , 2 ] (0,2] ( 0 , 2 ] ; there the Hölder continuity of b b b makes the drift small at the scale κ 1 / β \kappa^{1/\beta} κ 1/ β along its trajectories, and Lemma 4.3 applies in a frame moving along a trajectory of $b ({xref}Lemma 9.1`).
Iterated homogenization
This section reduces Proposition 4.1 to four propositions, proved in Part II, by an induction on the levels of b M + n b_{M+n} b M + n , which are homogenized one at a time, from the finest to a fixed coarse level.
From L2 to L-infinity, scale by scale
This subsection is heuristic, except for the definition of the corrector amplitude and the inequality (30) , which is proved here and used in the proof; the statements of the proof begin in Subsection 5.2.
Overview: two dual cascades
Anomalous dissipation in [5 ] . For the velocity field of [5 ] , that is, for the choice of q q q made there, [5 Theorem 1.1] states that the scalar variance is dissipated anomalously: for every mean-zero θ i n ∈ H 1 ( T 2 ) \theta_{\mathrm{in}} \in H^1(\mathbb{T}^2) θ in ∈ H 1 ( T 2 ) with θ i n ≠ 0 \theta_{\mathrm{in}} \ne0 θ in = 0 , the solutions θ κ \theta^\kappa θ κ of
{ ∂ t θ κ + b ⋅ ∇ θ κ − κ Δ θ κ = 0 in ( 0 , 1 ) × T 2 , θ κ ( 0 , ⋅ ) = θ i n on T 2 \begin{cases}
\partial_t\theta^\kappa+\mathbf{b}\cdot\nabla\theta^\kappa-\kappa\Delta\theta^\kappa=0 & \text{in }(0,1)\times\mathbb{T}^2,\\
\theta^\kappa(0,\cdot)=\theta_{\mathrm{in}} & \text{on }\mathbb{T}^2
\end{cases} { ∂ t θ κ + b ⋅ ∇ θ κ − κ Δ θ κ = 0 θ κ ( 0 , ⋅ ) = θ in in ( 0 , 1 ) × T 2 , on T 2
satisfy
lim sup κ → 0 κ ∥ ∇ θ κ ∥ L 2 ( ( 0 , 1 ) × T 2 ) 2 > 0 , \limsup_{\kappa\to0}\kappa\|\nabla\theta^\kappa\|_{L^2((0,1)\times\mathbb{T}^2)}^2>0, κ → 0 lim sup κ ∥∇ θ κ ∥ L 2 (( 0 , 1 ) × T 2 ) 2 > 0 ,
with a lower bound depending only on ∥ θ i n ∥ L 2 ( T 2 ) \|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)} ∥ θ in ∥ L 2 ( T 2 ) and the ratio ∥ θ i n ∥ L 2 ( T 2 ) / ∥ ∇ θ i n ∥ L 2 ( T 2 ) \|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)}/\|\nabla\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)} ∥ θ in ∥ L 2 ( T 2 ) /∥∇ θ in ∥ L 2 ( T 2 ) . The proof compares, for κ ∈ K M \kappa\in K_M κ ∈ K M , the solutions θ m \theta_m θ m of the equations of level m m m , with velocity field b m \mathbf{b}_m b m , diffusivity κ m \kappa_m κ m and the same initial datum, and shows that the dissipation κ m ∥ ∇ θ m ∥ L 2 ( ( 0 , 1 ) × T 2 ) 2 \kappa_m\|\nabla\theta_m\|_{L^2((0,1)\times\mathbb{T}^2)}^2 κ m ∥∇ θ m ∥ L 2 (( 0 , 1 ) × T 2 ) 2 is multiplied only by 1 + O ( ε m − 1 δ ) 1+O(\varepsilon_{m-1}^{\delta}) 1 + O ( ε m − 1 δ ) from one level to the next, at all levels finer than a threshold depending on the initial datum [5 Proposition 5.2] .
The expected regularity. By the energy identity, the dissipation of each level satisfies κ m ∥ ∇ θ m ∥ L 2 ( ( 0 , ∞ ) × T 2 ) 2 ≤ 1 2 ∥ θ i n ∥ L 2 ( T 2 ) 2 \kappa_m\|\nabla\theta_m\|_{L^2((0,\infty)\times\mathbb{T}^2)}^2\le\frac{1}{2}\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)}^2 κ m ∥∇ θ m ∥ L 2 (( 0 , ∞ ) × T 2 ) 2 ≤ 2 1 ∥ θ in ∥ L 2 ( T 2 ) 2 . Heuristically, θ m \theta_m θ m is the solution seen at the scale ε m \varepsilon_m ε m , with the oscillations of smaller wavelength averaged out, so at the scale ε m \varepsilon_m ε m the gradient of the solution is of size at most κ m − 1 / 2 ≃ ε m − ( β + γ ) / 2 \kappa_m^{-1/2}\simeq\varepsilon_m^{-(\beta+\gamma)/2} κ m − 1/2 ≃ ε m − ( β + γ ) /2 in L 2 ( ( 0 , ∞ ) × T 2 ) L^2((0,\infty)\times\mathbb{T}^2) L 2 (( 0 , ∞ ) × T 2 ) , for 1 ≤ m < M 1\le m<M 1 ≤ m < M , by Proposition 2.1(b). If the gradient of θ m \theta_m θ m were of this size pointwise in ( 0 , ∞ ) × T 2 (0,\infty)\times\mathbb{T}^2 ( 0 , ∞ ) × T 2 , the solution would oscillate by at most ε m ⋅ ε m − ( β + γ ) / 2 \varepsilon_m\cdot\varepsilon_m^{-(\beta+\gamma)/2} ε m ⋅ ε m − ( β + γ ) /2 over a distance ε m \varepsilon_m ε m , at every level 1 ≤ m < M 1\le m<M 1 ≤ m < M : it would be Hölder continuous with exponent 1 − 1 2 ( β + γ ) 1-\frac{1}{2}(\beta+\gamma) 1 − 2 1 ( β + γ ) . Since β + γ = 2 q β / ( q + 1 ) \beta+\gamma=2q\beta/(q+1) β + γ = 2 q β / ( q + 1 ) ,
1 − β + γ 2 = 1 − q α 1 + q > 1 − α 2 as q ↓ 1 , μ = ( q − 1 ) ( 1 − s ) − q γ = ( q − 1 ) ( 1 − β + γ 2 − s ) . (29) 1-\frac{\beta+\gamma}{2}=\frac{1-q\alpha}{1+q}>\frac{1-\alpha}{2}\quad\text{as }q\downarrow1,\qquad\mu=(q-1)(1-s)-q\gamma=(q-1)\left(1-\frac{\beta+\gamma}{2}-s\right). \tag*{(29)} 1 − 2 β + γ = 1 + q 1 − q α > 2 1 − α as q ↓ 1 , μ = ( q − 1 ) ( 1 − s ) − q γ = ( q − 1 ) ( 1 − 2 β + γ − s ) . (29)
The limit is the Obukhov–Corrsin exponent, and μ \mu μ is q − 1 q-1 q − 1 times the margin by which s s s lies below the heuristic exponent. Theorem A makes this heuristic rigorous, for q q q close to 1. The cost of one level is read off from its correctors. Let a function be bounded in C 0 , s ( T 2 ) C^{0,s}(\mathbb{T}^2) C 0 , s ( T 2 ) and smooth at the wavelength ε m − 1 \varepsilon_{m-1} ε m − 1 , so that its gradient is of size ε m − 1 s − 1 \varepsilon_{m-1}^{s-1} ε m − 1 s − 1 . The correctors of level m m m add to it a function that oscillates at the wavelength ε m \varepsilon_m ε m , of size the corrector amplitude ε ˉ m : = ε m ( κ m − 1 / κ m ) 1 / 2 \bar{\varepsilon}_m:=\varepsilon_m(\kappa_{m-1}/\kappa_m)^{1/2} ε ˉ m := ε m ( κ m − 1 / κ m ) 1/2 times this gradient; for 2 ≤ m < M 2\le m<M 2 ≤ m < M the corrector amplitude is, up to a constant, ε m \varepsilon_m ε m times the Péclet number ε m q / κ m \varepsilon_m^q/\kappa_m ε m q / κ m of level m m m , by Proposition 2.1(b) and ε m ≃ ε m − 1 q \varepsilon_m\simeq\varepsilon_{m-1}^q ε m ≃ ε m − 1 q . A function that oscillates at the wavelength ε m \varepsilon_m ε m has a C 0 , s ( T 2 ) C^{0,s}(\mathbb{T}^2) C 0 , s ( T 2 ) norm of order ε m − s \varepsilon_m^{-s} ε m − s times its supremum. So, for 2 ≤ m ≤ M 2\le m\le M 2 ≤ m ≤ M , the correctors of level m m m cost, relative to the C 0 , s ( T 2 ) C^{0,s}(\mathbb{T}^2) C 0 , s ( T 2 ) norm,
ε m − s ε ˉ m ε m − 1 s − 1 = ( ε m ε m − 1 ) 1 − s ( κ m − 1 κ m ) 1 / 2 ≤ C ε m − 1 ( q − 1 ) ( 1 − s ) − q γ = C ε m − 1 μ (30) \varepsilon_m^{-s}\bar{\varepsilon}_m\varepsilon_{m-1}^{s-1}
=\left(\frac{\varepsilon_m}{\varepsilon_{m-1}}\right)^{1-s}\left(\frac{\kappa_{m-1}}{\kappa_m}\right)^{1/2}
\le C\varepsilon_{m-1}^{(q-1)(1-s)-q\gamma}
=C\varepsilon_{m-1}^{\mu} \tag*{(30)} ε m − s ε ˉ m ε m − 1 s − 1 = ( ε m − 1 ε m ) 1 − s ( κ m κ m − 1 ) 1/2 ≤ C ε m − 1 ( q − 1 ) ( 1 − s ) − q γ = C ε m − 1 μ (30)
by Proposition 2.1(c), the terminal level included, and by ε m ≃ ε m − 1 q \varepsilon_m\simeq\varepsilon_{m-1}^q ε m ≃ ε m − 1 q . Separation of scales gains ( ε m / ε m − 1 ) 1 − s (\varepsilon_m/\varepsilon_{m-1})^{1-s} ( ε m / ε m − 1 ) 1 − s , enhancement costs ( κ m − 1 / κ m ) 1 / 2 (\kappa_{m-1}/\kappa_m)^{1/2} ( κ m − 1 / κ m ) 1/2 , and their balance μ > 0 \mu>0 μ > 0 is the condition that s s s lies below the exponent in (29) .
Large-scale regularity. In the quantitative argument of large-scale regularity recalled in Subsection 1.3, the homogenization errors are summed over the scales in an iteration of Campanato type, in L 2 L^2 L 2 in [2 ] and in L ∞ L^\infty L ∞ in [4 ] . Here the comparison is made once per level, on the whole torus. The seminorm \lambda^{-s}[\cdot]_{C^{N_{\mathrm{rep}}(\mathbb{T}^2)}^{\lambda}} (see (40) ) of a solution, bounded by the size of a representation (Subsection 5.6) at the wavelength λ \lambda λ , plays the role of the excess at the scale λ \lambda λ , and (30) is the cost of one level.
Overview: two dual cascades
Fix M M M , κ ∈ K M \kappa\in\mathcal{K}_M κ ∈ K M and a weak solution θ \theta θ of (28) . For each level m ≤ M m \le M m ≤ M the effective diffusivity κ m \kappa_m κ m was defined so that, once the pulses of level m m m are homogenized, the operator ∂ t + b m ⋅ ∇ − κ m Δ \partial_t + \mathbf{b}_m \cdot\nabla- \kappa_m \Delta ∂ t + b m ⋅ ∇ − κ m Δ of level m m m becomes the operator of level m − 1 m-1 m − 1 , with b m − 1 \mathbf{b}_{m-1} b m − 1 and κ m − 1 \kappa_{m-1} κ m − 1 . The proof runs two dual cascades on the levels (Figure 5.1 ). The homogenization cascade goes up the levels one at a time, from M M M to a fixed top level m 0 m_0 m 0 , toward coarser levels and forward in time; the regularity cascade then goes back down, toward finer levels, at one fixed time. The proof has four stages, and Proposition 5.2 runs both cascades between the seed and the top level.
Figure 5.1. The induction of Subsection 5.2, levels against time t ′ t' t ′ (not to scale), coarser levels higher. The homogenization cascade, up the levels and forward in time: the entry (Proposition 5.1 ) takes θ \theta θ , at level M + n M+n M + n and time t i n t_{\mathrm{in}} t in , to the seed, at level M − 1 M-1 M − 1 and time t e t_{\mathrm{e}} t e ; each up arrow is then a cycle (Proposition 5.3), with the factor of (31) , and the cycles end less than 1 2 t 0 \frac{1}{2}t_0 2 1 t 0 after t e t_{\mathrm{e}} t e . Grey: the solution U m U_m U m of each family, on [ t ( m ) , t f ] [t^{(m)},t_{\mathrm{f}}] [ t ( m ) , t f ] . The regularity cascade, back down the levels at one time t ∈ [ t i n + t 0 , t i n + T ] t \in[t_{\mathrm{in}}+t_0,t_{\mathrm{in}}+T] t ∈ [ t in + t 0 , t in + T ] : each down arrow is a passage (Proposition 5.4 ), with the bound (5.4), and the dashed arrow is the lift through the entry. Dotted: levels not drawn.
The entry (Proposition 5.1 ) disposes of the levels M + n , … , M M+n,\ldots,M M + n , … , M . At the levels finer than M M M , the Péclet number ε m / κ \varepsilon^m/\kappa ε m / κ is small and molecular diffusion over the quiet margins smooths at the wavelength of the next coarser level, so these levels are crossed at the same diffusivity κ \kappa κ . At level M M M a full homogenization cycle passes to κ M − 1 \kappa_{M-1} κ M − 1 . The output is a system at level M − 1 M-1 M − 1 , the seed, whose solution, lifted back to level M + n M+n M + n , reproduces θ \theta θ up to a small error in C 0 , s ( T 2 ) C^{0,s}(\mathbb{T}^2) C 0 , s ( T 2 ) .
The homogenization cascade (Proposition 5.2 ) then homogenizes the levels M − 1 , … , m 0 + 1 M-1,\ldots,m_0+1 M − 1 , … , m 0 + 1 one at a time. Each step up is a homogenization cycle (Proposition 5.3).
At the top level m 0 m_0 m 0 , a fixed level independent of M M M and κ \kappa κ , the solution is smooth at a fixed scale.
The regularity cascade (Proposition 5.2 ) then goes back down the levels m 0 + 1 , … , M − 1 m_0+1,\ldots,M-1 m 0 + 1 , … , M − 1 at one fixed time. Each step down is a passage (Proposition 5.4 ), and the lifts through the entry (Proposition 5.1 ) carry the Hölder bound from the seed to θ \theta θ .
What the homogenization cascade carries up the levels. A cycle compares the solution at level m m m with a solution at level m − 1 m-1 m − 1 , regenerating derivatives and keeping the marked amplitude, defined below, up to the factor in (31) . The object passed from level m m m to level m − 1 m-1 m − 1 is a coupled system of equations with the operator of its level, a family (Subsection 5.4). Starting from the seed, the cycles produce families G m G_m G m , with solutions U m U_m U m on time windows [ t ( m ) , t f ] [t_{(m)},t_f] [ t ( m ) , t f ] that end at a common final time t f t_f t f , where t ( M − 1 ) = t e t_{(M-1)}=t_e t ( M − 1 ) = t e and, for m < M − 1 m<M-1 m < M − 1 , t ( m ) t_{(m)} t ( m ) is the preparation time t p r t_{\mathrm{pr}} t pr of the cycle at level m + 1 m+1 m + 1 (Proposition 5.3). The marked amplitude of U m U_m U m is its norm A m : = ∥ U m ∥ L ∞ ( [ t ( m ) , t f ] ; L G m 2 ( T 2 ) ) A_m := \lVert U_m\rVert_{L^\infty([t_{(m)},t_f];L^2_{G_m}(\mathbb{T}^2))} A m := ∥ U m ∥ L ∞ ([ t ( m ) , t f ] ; L G m 2 ( T 2 )) , where ∥ ⋅ ∥ L G m 2 ( T 2 ) \lVert\cdot\rVert_{L^2_{G_m}(\mathbb{T}^2)} ∥ ⋅ ∥ L G m 2 ( T 2 ) is the marked norm of G m G_m G m (Subsection 5.4). One cycle costs
A m − 1 ≤ [ 1 + C ( ε m − 1 Q + ε m − 1 δ ) ] A m , (31) A_{m-1} \le\left[1+C\left(\varepsilon_{m-1}^{Q}+\varepsilon_{m-1}^{\delta}\right)\right]A_m, \tag*{(31)} A m − 1 ≤ [ 1 + C ( ε m − 1 Q + ε m − 1 δ ) ] A m , (31)
where ε m − 1 Q \varepsilon_{m-1}^{Q} ε m − 1 Q is the accuracy of the comparison and ε m − 1 δ \varepsilon_{m-1}^{\delta} ε m − 1 δ is the cost of the retained trace, from which the solution of the successor starts (Subsection 5.5); Proposition 6.2 prevents growth along a family. The exponent Q Q Q , the order N r e t N_{\mathrm{ret}} N ret below and the other parameters of the proof are fixed in Subsection 5.3.
What the regularity cascade carries back down. The C 0 , s ( T 2 ) C^{0,s}(\mathbb{T}^2) C 0 , s ( T 2 ) norm is carried by a representation : a smooth base plus oscillating profiles, each measured at its wavelength (Subsection 5.6). A representation is a Hölder bound together with bounds on derivatives at the scale of the current wavelength, which the correctors of the next finer level use. Read on representations, a cycle is a passage : it carries a representation of U m − 1 ( t ) U_{m-1}(t) U m − 1 ( t ) at wavelength ε m − 1 \varepsilon_{m-1} ε m − 1 to a representation of U m ( t ) U_m(t) U m ( t ) at wavelength ε m \varepsilon_m ε m , and the profiles of the representation of U m − 1 ( t ) U_{m-1}(t) U m − 1 ( t ) are copied into that of U m ( t ) U_m(t) U m ( t ) by a constant map (Proposition 5.4). With R m ( t ) R_m(t) R m ( t ) a representation of U m ( t ) U_m(t) U m ( t ) , one passage costs
size ε m ( R m ( t ) ) ≤ ( 1 + C ε m − 1 η ) size ε m − 1 ( R m − 1 ( t ) ) + C ε m − 1 A m , (32) \operatorname{size}_{\varepsilon_m}(R_m(t)) \le(1+C\varepsilon_{m-1}^{\eta})\operatorname{size}_{\varepsilon_{m-1}}(R_{m-1}(t))+C\varepsilon_{m-1}A_m, \tag*{(32)} size ε m ( R m ( t )) ≤ ( 1 + C ε m − 1 η ) size ε m − 1 ( R m − 1 ( t )) + C ε m − 1 A m , (32)
where η \eta η is the exponent of (5) . The factor collects the readout (Subsection 5.5), the exponential part of the lift and the correctors of the passage, which cost (30) , with η ≤ μ / 4 \eta\le\mu/4 η ≤ μ /4 ; the additive term collects the comparison error of Proposition 5.3 and what the lift adds to the high frequencies of U m − 1 ( t ) U_{m-1}(t) U m − 1 ( t ) (the idea of the proof of Proposition 5.4; Table 14.1 ). At the top level, the size of a representation of U m 0 ( t ) U_{m_0}(t) U m 0 ( t ) is at most a constant, depending on m 0 m_0 m 0 , times the marked amplitude A m 0 + 1 A_{m_0+1} A m 0 + 1 : the third bound of Proposition 5.3(b) at level m 0 + 1 m_0+1 m 0 + 1 gives N r e t N_{\mathrm{ret}} N ret derivatives of U m 0 ( t ) U_{m_0}(t) U m 0 ( t ) at the scale ρ ( m 0 + 1 ) \rho(m_0+1) ρ ( m 0 + 1 ) , and these bound the size at the wavelength ε m 0 \varepsilon_{m_0} ε m 0 , by Sobolev embedding, since N r e t ≥ N r e p + 2 N_{\mathrm{ret}}\ge N_{\mathrm{rep}}+2 N ret ≥ N rep + 2 by (180) .
Table 14.1. The five profiles that a passage at level m m m appends to the readout of a representation R ′ R' R ′ of Y ( t ) Y(t) Y ( t ) : the wavelength λ P \lambda_P λ P of each profile P P P , the bound on λ P − s ∥ P ∥ C λ P N r e p ( T 2 ) \lambda_P^{-s}\lVert P\rVert_{C_{\lambda_P}^{N_{\mathrm{rep}}}(\mathbb{T}^{2})} λ P − s ∥ P ∥ C λ P N rep ( T 2 ) proved in Steps 3–6 of the proof of Proposition 5.4, and the origin of its smallness. With the readout of Proposition 14.1(i), the first two give the coefficient 1 + C ε m − 1 η 1+C\varepsilon_{m-1}^{\eta} 1 + C ε m − 1 η in Proposition 5.4; the last three give its additive term.
Why the products converge. Since ε 0 = 1 \varepsilon_0=1 ε 0 = 1 and ε m − 1 / ε m ≥ Λ ≥ 128 \varepsilon_{m-1}/\varepsilon_m\ge\Lambda\ge128 ε m − 1 / ε m ≥ Λ ≥ 128 for m ≥ 1 m\ge1 m ≥ 1 , we have ε m ≤ Λ − m \varepsilon_m\le\Lambda^{-m} ε m ≤ Λ − m , so the sums ∑ m ε m − 1 Q \sum_m\varepsilon_{m-1}^{Q} ∑ m ε m − 1 Q , ∑ m ε m − 1 δ \sum_m\varepsilon_{m-1}^{\delta} ∑ m ε m − 1 δ and ∑ m ε m − 1 η \sum_m\varepsilon_{m-1}^{\eta} ∑ m ε m − 1 η are bounded independently of the number of levels, and so are the products ∏ m ( 1 + C ε m − 1 Q + C ε m − 1 δ ) \prod_m(1+C\varepsilon_{m-1}^{Q}+C\varepsilon_{m-1}^{\delta}) ∏ m ( 1 + C ε m − 1 Q + C ε m − 1 δ ) and ∏ m ( 1 + C ε m − 1 η ) \prod_m(1+C\varepsilon_{m-1}^{\eta}) ∏ m ( 1 + C ε m − 1 η ) . Iterating (31) along the homogenization cascade and (5.4) along the regularity cascade therefore bounds the size of the representation at the finest level by a constant times the norm ∥ g ∥ L G 2 ( T 2 ) \lVert g\rVert_{L^2_{G}(\mathbb{T}^2)} ∥ g ∥ L G 2 ( T 2 ) of the seed’s datum g g g , uniformly in the number of levels, hence in M M M and κ \kappa κ (Proposition 5.2); this norm is the marked norm, since the markers of a seed are 1 1 1 .
Parameters
Fix Λ ≥ Λ 0 \Lambda\ge\Lambda_0 Λ ≥ Λ 0 , with Λ 0 \Lambda_0 Λ 0 as in Proposition 4.1, an integer n ≥ 1 n\ge1 n ≥ 1 and 0 < t 0 < T 0<t_0<T 0 < t 0 < T . The proof fixes the following parameters, all depending only on α , s , q , ζ , λ , n , t 0 , T \alpha,s,q,\zeta,\lambda,n,t_0,T α , s , q , ζ , λ , n , t 0 , T :
the truncation order p p p of the exponential in a lift;
the order N r e t ≥ N c o r + 2 p N_{\mathrm{ret}} \ge N_{\mathrm{cor}} + 2p N ret ≥ N cor + 2 p up to which a cycle controls the derivatives of the solution of a successor, and the order N o u t : = N r e t + 4 p N_{\mathrm{out}} := N_{\mathrm{ret}} + 4p N out := N ret + 4 p of the preparations;
the order N r e p N_{\mathrm{rep}} N rep of the representations, which is fixed by the entry (§ 6 );
the order N s r c ≥ 2 N_{\mathrm{src}} \ge2 N src ≥ 2 of the source terms of the families;
an accuracy exponent Q > 0 Q > 0 Q > 0 ;
the bounds on generation and word length of the successors (Subsection 5.4);
the top level m 0 ≥ 2 m_0 \ge2 m 0 ≥ 2 , chosen after the others: the largest of the thresholds on the level in § 5.4 –§ 5.11 (item (4) of Subsection 11.1).
Table 11.1 lists them, with the parameters of Part II, in the order in which they are chosen. The preparations of level m m m are smooth at the effective diffusion length over one renewal time (§ 1.1 ),
ρ ( m ) = ( κ m − 1 h m ) 1 / 2 . (33) \rho(m) = \left(\kappa_{m-1}h_m\right)^{1/2}. \tag*{(33)} ρ ( m ) = ( κ m − 1 h m ) 1/2 . (33)
The notions of family, admissible family, lift and representation below use these choices, families use the effective diffusivities of κ ∈ K M \kappa\in\mathcal{K}_M κ ∈ K M , and time windows have length at most T T T .
Families
A comparison between levels m m m and m − 1 m-1 m − 1 leaves error terms carrying derivatives of the coarse solution. Estimating them would cost derivatives at every level, which the induction cannot afford, so they become new unknowns, each solving the equation of its level forced by smoothed derivatives of earlier unknowns. These new sources are the higher-order terms of the averaged equation that the lift of Subsection 5.5 produces: the defect of the averaged flux, the means of the lift along the paths of sources of the family, and their conjugates by the truncated exponential. They, and the old sources carried over through their cell averages, are listed in Definition 11.3 .
Families and their solutions. A family is the data of a system of linear equations of level m m m , one for each of its components, coupled through smoothed derivatives; a solution is a vector of functions on [ t 1 , t 2 ] × T 2 [t_1,t_2] \times\mathbb{T}^2 [ t 1 , t 2 ] × T 2 , one for each component. Precisely, a family G G G of level m m m consists of
a finite index set V V V , whose elements are the components (labels, which are words in the families of the proof, as described below), with masses w v > 0 w_v > 0 w v > 0 and markers Ψ v ≥ 1 \Psi_v \ge1 Ψ v ≥ 1 , v ∈ V v \in V v ∈ V ;
finitely many sources σ \sigma σ , each from a component v ′ v' v ′ to a component v v v (written σ : v ′ → v \sigma: v' \to v σ : v ′ → v ; there may be several sources, or none, from v ′ v' v ′ to v v v ), with a linear smoothing operator S σ ( t ) S_\sigma(t) S σ ( t ) of width r σ > 0 r_\sigma> 0 r σ > 0 , and coefficients A σ , d A_{\sigma,d} A σ , d , 2 ≤ d ≤ N s r c 2 \le d \le N_{\mathrm{src}} 2 ≤ d ≤ N src ;
constants c j ≥ 0 c_j \ge0 c j ≥ 0 such that the operators S σ ( t ) S_\sigma(t) S σ ( t ) , which commute with derivatives, satisfy, for every source σ \sigma σ , time t t t , f ∈ L 2 ( T 2 ) f \in L^2(\mathbb{T}^2) f ∈ L 2 ( T 2 ) and j ≤ N s r c − 2 j \le N_{\mathrm{src}} - 2 j ≤ N src − 2 ,
∥ ∇ j S σ ( t ) f ∥ L 2 ( T 2 ) ≤ c j r σ − j ∥ f ∥ L 2 ( T 2 ) . \left\|\nabla^j S_\sigma(t)f\right\|_{L^2(\mathbb{T}^2)} \le c_j r_\sigma^{-j}\|f\|_{L^2(\mathbb{T}^2)}. ∇ j S σ ( t ) f L 2 ( T 2 ) ≤ c j r σ − j ∥ f ∥ L 2 ( T 2 ) .
A datum of G G G is g = ( g v ) v ∈ V ∈ L 2 ( T 2 ) V g=(g_v)_{v\in V}\in L^2(\mathbb{T}^2)^V g = ( g v ) v ∈ V ∈ L 2 ( T 2 ) V , and the solution of G G G from g g g at time t 1 t_1 t 1 , on a time window [ t 1 , t 2 ] [t_1,t_2] [ t 1 , t 2 ] on which the hypotheses of Proposition 6.2 hold, is the solution U = ( U v ) v ∈ V U=(U_v)_{v\in V} U = ( U v ) v ∈ V of
{ ∂ t U v + b m ⋅ ∇ U v − κ m Δ U v = ∇ ⋅ ∑ v ′ ∈ V ∑ σ : v ′ → v ∑ d = 2 N s r c A σ , d ∇ d − 1 ( S σ U v ′ ) in ( t 1 , t 2 ) × T 2 , v ∈ V , U v ( t 1 , ⋅ ) = g v on T 2 , (34) \left\{
\begin{aligned}
\partial_t U_v+\mathbf{b}_m\cdot\nabla U_v-\kappa_m\Delta U_v
&=\nabla\cdot\sum_{v'\in V}\sum_{\sigma:v'\to v}\sum_{d=2}^{N_{\mathrm{src}}}A_{\sigma,d}\nabla^{d-1}(S_\sigma U_{v'}) &&\text{in }(t_1,t_2)\times\mathbb{T}^2,\qquad v\in V,\\
U_v(t_1,\cdot)&=g_v &&\text{on }\mathbb{T}^2,
\end{aligned}
\right. \tag*{(34)} ⎩ ⎨ ⎧ ∂ t U v + b m ⋅ ∇ U v − κ m Δ U v U v ( t 1 , ⋅ ) = ∇ ⋅ v ′ ∈ V ∑ σ : v ′ → v ∑ d = 2 ∑ N src A σ , d ∇ d − 1 ( S σ U v ′ ) = g v in ( t 1 , t 2 ) × T 2 , v ∈ V , on T 2 , (34)
in the class of Proposition 6.2 , where the inner sum runs over the sources from v ′ v' v ′ to v v v , so that the right side collects all the sources into v v v ; this is (45) with ϕ m \phi_m ϕ m and κ m \kappa_m κ m in place of Φ \Phi Φ and ν \nu ν . The smoothing operators are products of two smoothings along ellipses (Proposition 6.1 ). For the sources that a cycle creates at level m m m the ellipses are thin. The averaged flux of a pulse is enhanced only along its direction of flow, and the two ellipses follow the averaged flux of a horizontal and of a vertical pulse, so that their axes are in a ratio of order ( κ m / κ m − 1 ) 1 / 2 (\kappa_m/\kappa_{m-1})^{1/2} ( κ m / κ m − 1 ) 1/2 , which is small below the terminal level M M M (Proposition 2.1 (c)); together they smooth in every direction.
Norms. Fields of G G G are measured in the norms
∥ g ∥ L ω 2 ( T 2 ) 2 : = ∑ v ∈ V ∥ g v ∥ L 2 ( T 2 ) 2 ω v , \lVert g\rVert_{L^2_\omega(\mathbb{T}^2)}^2:=\sum_{v\in V}\frac{\lVert g_v\rVert_{L^2(\mathbb{T}^2)}^2}{\omega_v}, ∥ g ∥ L ω 2 ( T 2 ) 2 := v ∈ V ∑ ω v ∥ g v ∥ L 2 ( T 2 ) 2 ,
∥ g ∥ L G 2 ( T 2 ) 2 : = ∑ v ∈ V Ψ v 2 ∥ g v ∥ L 2 ( T 2 ) 2 ω v , (35) \lVert g\rVert_{L^2_G(\mathbb{T}^2)}^2:=\sum_{v\in V}\frac{\Psi_v^2\lVert g_v\rVert_{L^2(\mathbb{T}^2)}^2}{\omega_v}, \tag*{(35)} ∥ g ∥ L G 2 ( T 2 ) 2 := v ∈ V ∑ ω v Ψ v 2 ∥ g v ∥ L 2 ( T 2 ) 2 , (35)
∥ g ∥ H ρ , G N ( T 2 ) 2 : = ∑ v ∈ V Ψ v 2 ∥ g v ∥ H ρ N ( T 2 ) 2 ω v , \lVert g\rVert_{H^N_{\rho,G}(\mathbb{T}^2)}^2:=\sum_{v\in V}\frac{\Psi_v^2\lVert g_v\rVert_{H^N_\rho(\mathbb{T}^2)}^2}{\omega_v}, ∥ g ∥ H ρ , G N ( T 2 ) 2 := v ∈ V ∑ ω v Ψ v 2 ∥ g v ∥ H ρ N ( T 2 ) 2 ,
where, for ρ > 0 \rho>0 ρ > 0 and an integer N ≥ 0 N\geq0 N ≥ 0 ,
∥ f ∥ H ρ N ( T 2 ) 2 : = ∑ j ≤ N ρ 2 j ( j ! ) 2 ∥ ∇ j f ∥ L 2 ( T 2 ) 2 (36) \lVert f\rVert_{H^N_\rho(\mathbb{T}^2)}^2:=\sum_{j\leq N}\frac{\rho^{2j}}{(j!)^2}\lVert\nabla^j f\rVert_{L^2(\mathbb{T}^2)}^2 \tag*{(36)} ∥ f ∥ H ρ N ( T 2 ) 2 := j ≤ N ∑ ( j ! ) 2 ρ 2 j ∥ ∇ j f ∥ L 2 ( T 2 ) 2 (36)
measures N N N derivatives at the length scale ρ \rho ρ . The norm ∥ ⋅ ∥ L G 2 ( T 2 ) \lVert\cdot\rVert_{L^2_G(\mathbb{T}^2)} ∥ ⋅ ∥ L G 2 ( T 2 ) is the marked norm .
The coupling matrix. The coupling of the components is measured by the coupling matrix of G G G on a time window [ t 1 , t 2 ] [t_1,t_2] [ t 1 , t 2 ] ,
B v v ′ : = ( ω v ′ ω v ) 1 / 2 ∑ σ : v ′ → v ∑ d = 2 N s r c c d − 2 sup [ t 1 , t 2 ] × T 2 ∣ A σ , d ∣ κ m r σ d − 2 , v , v ′ ∈ V , (37) B_{vv'}:=\left(\frac{\omega_{v'}}{\omega_v}\right)^{1/2}\sum_{\sigma:v'\to v}\sum_{d=2}^{N_{\mathrm{src}}}c_{d-2}\frac{\sup_{[t_1,t_2]\times\mathbb{T}^2}|A_{\sigma,d}|}{\kappa_m r_\sigma^{d-2}},\qquad v,v'\in V, \tag*{(37)} B v v ′ := ( ω v ω v ′ ) 1/2 σ : v ′ → v ∑ d = 2 ∑ N src c d − 2 κ m r σ d − 2 sup [ t 1 , t 2 ] × T 2 ∣ A σ , d ∣ , v , v ′ ∈ V , (37)
with the constants c j c_j c j of G G G . By Schur’s test, the operator norm of ( B v v ′ ) (B_{vv'}) ( B v v ′ ) on R V \mathbb{R}^V R V with the Euclidean norm is at most the geometric mean of its largest row sum and its largest column sum. By Proposition 6.2 , if all the row and column sums of the coupling matrix are at most 1 8 \frac{1}{8} 8 1 , then the energy ∥ U ( t ) ∥ L ω 2 ( T 2 ) \lVert U(t)\rVert_{L^2_\omega(\mathbb{T}^2)} ∥ U ( t ) ∥ L ω 2 ( T 2 ) does not grow. Indeed, testing the equation for U v U_v U v with U v / ω v U_v/\omega_v U v / ω v and writing ∇ d − 1 S σ U v ′ = ∇ d − 2 S σ ∇ U v ′ \nabla^{d-1}S_\sigma U_{v'}=\nabla^{d-2}S_\sigma\nabla U_{v'} ∇ d − 1 S σ U v ′ = ∇ d − 2 S σ ∇ U v ′ , we bound the contribution of the sources from v ′ v' v ′ to v v v by κ m B v v ′ x v x v ′ \kappa_m B_{vv'}x_vx_{v'} κ m B v v ′ x v x v ′ , where x v : = ω v − 1 / 2 ∥ ∇ U v ∥ L 2 ( T 2 ) x_v:=\omega_v^{-1/2}\lVert\nabla U_v\rVert_{L^2(\mathbb{T}^2)} x v := ω v − 1/2 ∥ ∇ U v ∥ L 2 ( T 2 ) . By Schur’s test the coupling is then at most one eighth of the dissipation κ m ∑ v ∈ V x v 2 \kappa_m\sum_{v\in V}x_v^2 κ m ∑ v ∈ V x v 2 . The coupling matrix for the masses ω v / Ψ v 2 \omega_v/\Psi_v^2 ω v / Ψ v 2 is (37) with ω v / Ψ v 2 \omega_v/\Psi_v^2 ω v / Ψ v 2 in place of ω v \omega_v ω v ; with these masses the norm ∥ ⋅ ∥ L ω 2 ( T 2 ) \lVert\cdot\rVert_{L^2_\omega(\mathbb{T}^2)} ∥ ⋅ ∥ L ω 2 ( T 2 ) is the marked norm.
Successors. A cycle turns a family G G G of level m m m into a successor G ′ G' G ′ of level m − 1 m-1 m − 1 . Each component v v v of G G G is kept and becomes the parent of new components, its children , labelled by words whose letters name the source terms the comparison creates. A new component has mass ω v \omega_v ω v times a positive power of ε m − 1 \varepsilon_{m-1} ε m − 1 and marker Ψ v \Psi_v Ψ v times a negative power of ε m − 1 \varepsilon_{m-1} ε m − 1 , so the markers grow with the generation (the number of creations). A successor is a family of Definition 11.3, which gives its letters, the words of its components and its sources; it keeps the components of bounded generation and bounded word length. The solution is controlled in the marked norm, so the components of high generation are small in ∥ ⋅ ∥ L ω 2 ( T 2 ) \lVert\cdot\rVert_{L^2_\omega(\mathbb{T}^2)} ∥ ⋅ ∥ L ω 2 ( T 2 ) and the flux into the discarded ones is negligible ((Lemma 13.9)); components with longer words are small because each letter carries a positive power of ε m − 1 \varepsilon_{m-1} ε m − 1 .
Seeds and admissible families. A seed is a family whose components have generation 0 0 0 and marker 1 1 1 . A family G G G of level m m m is admissible on a time window if it descends by cycles from an admissible seed , a seed whose coefficients obey the bounds of Part II (Definition 11.2), each cycle, at a level m ′ m' m ′ , moving the start of the window forward by 3 τ m ′ 3\tau_{m'} 3 τ m ′ (Definition 11.4). Its coefficients then obey the bounds of Part II (Proposition 11.5), and the row and column sums of its coupling matrices, for the masses ω v \omega_v ω v and ω v / Ψ v 2 \omega_v/\Psi_v^2 ω v / Ψ v 2 , are at most a constant times a fixed positive power of ε m − 1 \varepsilon_{m-1} ε m − 1 .
Lifts
Let G ′ G' G ′ be a successor of a family G G G of level m m m . The lift Y ♯ Y^\sharp Y ♯ of a field Y Y Y of G ′ G' G ′ to the components of G G G is built in three steps:
( Υ m Y ) v : = Y v + ∑ v ′ c h i l d o f v ε m − 1 η # v ′ Y v ′ , Y ♯ ( t ) : = Corr m [ exp p ( Γ m ( t ) : ∇ 2 ) Υ m Y ( t ) ] (38) (\Upsilon_mY)_v := Y_v+\sum_{v'\ \mathrm{child\ of}\ v}\varepsilon_{m-1}^{\eta\#v'}Y_{v'},\qquad Y^\sharp(t):=\operatorname{Corr}_m\left[\operatorname{exp}_p\left(\Gamma_m(t):\nabla^2\right)\Upsilon_mY(t)\right] \tag*{(38)} ( Υ m Y ) v := Y v + v ′ child of v ∑ ε m − 1 η # v ′ Y v ′ , Y ♯ ( t ) := Corr m [ exp p ( Γ m ( t ) : ∇ 2 ) Υ m Y ( t ) ] (38)
where the sum runs over the children of v v v (Subsection 5.4), # v ′ \#v' # v ′ is the number of letters of the word of v ′ v' v ′ , η \eta η is the exponent of (5) , and Corr m \operatorname{Corr}_m Corr m is the corrector operator of Definition 10.8, described below. First, the readout Υ m \Upsilon_m Υ m adds each new component to its parent with a weight that is a positive power of ε m − 1 \varepsilon_{m-1} ε m − 1 . Second, the truncated exponential of Γ m ( t ) : ∇ 2 : = ∑ i , i ′ ( Γ m ) i i ′ ( t ) ∂ i ∂ i ′ \Gamma_m(t):\nabla^2:=\sum_{i,i'}(\Gamma_m)_{ii'}(t)\partial_i\partial_{i'} Γ m ( t ) : ∇ 2 := ∑ i , i ′ ( Γ m ) i i ′ ( t ) ∂ i ∂ i ′ is applied, where
exp p ( ± Γ m ( t ) : ∇ 2 ) : = ∑ l = 0 p 1 l ! ( ± Γ m ( t ) : ∇ 2 ) l , Γ m ( t ) : = ∫ 0 t ( J m κ ( t ′ ) − κ m − 1 I ) d t ′ (39) \operatorname{exp}_p\left(\pm\Gamma_m(t):\nabla^2\right):=\sum_{l=0}^{p}\frac{1}{l!}\left(\pm\Gamma_m(t):\nabla^2\right)^l,\qquad\Gamma_m(t):=\int_0^t\left(J_m^{\kappa}(t')-\kappa_{m-1}I\right)\,\mathrm{d}t' \tag*{(39)} exp p ( ± Γ m ( t ) : ∇ 2 ) := l = 0 ∑ p l ! 1 ( ± Γ m ( t ) : ∇ 2 ) l , Γ m ( t ) := ∫ 0 t ( J m κ ( t ′ ) − κ m − 1 I ) d t ′ (39)
The covariance Γ m ( t ) \Gamma_m(t) Γ m ( t ) of the level-m m m pulses is the accumulated surplus of their averaged flux J m κ J_m^{\kappa} J m κ over its average κ m − 1 I \kappa_{m-1}I κ m − 1 I . Since ∂ t Γ m ( t ) = J m κ ( t ) − κ m − 1 I \partial_t\Gamma_m(t)=J_m^{\kappa}(t)-\kappa_{m-1}I ∂ t Γ m ( t ) = J m κ ( t ) − κ m − 1 I and operators with constant coefficients in space commute, the full exponential exp ( Γ m ( t ) : ∇ 2 ) \operatorname{exp}(\Gamma_m(t):\nabla^2) exp ( Γ m ( t ) : ∇ 2 ) would turn solutions of ∂ t − κ m − 1 Δ \partial_t-\kappa_{m-1}\Delta ∂ t − κ m − 1 Δ into solutions of ∂ t − J m κ ( t ) : ∇ 2 \partial_t-J_m^{\kappa}(t):\nabla^2 ∂ t − J m κ ( t ) : ∇ 2 ; as Γ m ( t ) \Gamma_m(t) Γ m ( t ) is symmetric and in general indefinite, this operator is unbounded on L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) , so the lift truncates it. The covariance is small at the scale ρ ( m ) \rho(m) ρ ( m ) : by Lemmas 10.3(b) and 13.7(iii), for every t t t ,
∣ Γ m ( t ) ∣ ≤ C κ m − 1 τ m ′ ≤ C ε m − 1 δ ρ ( m ) 2 , \lvert\Gamma_m(t)\rvert\le C\kappa_{m-1}\tau'_m\le C\varepsilon_{m-1}^{\delta}\rho(m)^2, ∣ Γ m ( t )∣ ≤ C κ m − 1 τ m ′ ≤ C ε m − 1 δ ρ ( m ) 2 ,
∥ ( exp p ( ± Γ m ( t ) : ∇ 2 ) − I ) F ∥ H ρ ( m ) N ( T 2 ) ≤ C ε m − 1 δ ∥ F ∥ H ρ ( m ) N + 2 p ( T 2 ) ; \left\lVert\left(\operatorname{exp}_p\left(\pm\Gamma_m(t):\nabla^2\right)-I\right)F\right\rVert_{H^N_{\rho(m)}(\mathbb{T}^2)}
\le C\varepsilon_{m-1}^{\delta}\lVert F\rVert_{H^{N+2p}_{\rho(m)}(\mathbb{T}^2)}; ( exp p ( ± Γ m ( t ) : ∇ 2 ) − I ) F H ρ ( m ) N ( T 2 ) ≤ C ε m − 1 δ ∥ F ∥ H ρ ( m ) N + 2 p ( T 2 ) ;
this is the origin of the term ε m − 1 δ \varepsilon_{m-1}^{\delta} ε m − 1 δ in the first bound of Proposition 5.3(b). Third, Corr m \operatorname{Corr}_m Corr m (Definition 10.8) adds on each pulse the correctors of degrees up to N c o r N_{\mathrm{cor}} N cor , which oscillate at the wavelength ε m \varepsilon_m ε m in the Lagrangian coordinates of the pulse; they change a field by the corrector amplitude ε m \varepsilon_m ε m times its derivatives. The retained trace at time t p r t_{\mathrm{pr}} t pr of a datum g g g of G G G is the datum of G ′ G' G ′ given by the truncated inverse of the second step, exact up to terms of degree at least p + 1 p+1 p + 1 in Γ m ( t p r ) : ∇ 2 \Gamma_m(t_{\mathrm{pr}}):\nabla^2 Γ m ( t pr ) : ∇ 2 : it is exp p ( − Γ m ( t p r ) : ∇ 2 ) g v \operatorname{exp}_p(-\Gamma_m(t_{\mathrm{pr}}):\nabla^2)g_v exp p ( − Γ m ( t pr ) : ∇ 2 ) g v on each component v v v of G G G and 0 0 0 on the new components.
Representations
Fields U U U of a family take values in R V \mathbb{R}^{V} R V , normed pointwise by ∣ U ( x ) ∣ ω : = ( ∑ v ω v − 1 U v ( x ) 2 ) 1 / 2 |U(x)|_{\omega}:=\left(\sum_{v}\omega_{v}^{-1}U_{v}(x)^{2}\right)^{1/2} ∣ U ( x ) ∣ ω := ( ∑ v ω v − 1 U v ( x ) 2 ) 1/2 , and ∥ ⋅ ∥ L ∞ ( T 2 ) \|\cdot\|_{L^{\infty}(\mathbb{T}^{2})} ∥ ⋅ ∥ L ∞ ( T 2 ) denotes the supremum of this pointwise norm; the Hölder norm and seminorm (22) of such fields use it in place of ∣ ⋅ ∣ |\cdot| ∣ ⋅ ∣ . For an integer N ≥ 1 N\geq1 N ≥ 1 , 0 < λ ≤ 1 0<\lambda\leq1 0 < λ ≤ 1 and F ∈ C N ( T 2 ) V F\in C^{N}(\mathbb{T}^{2})^{V} F ∈ C N ( T 2 ) V we use the norm and seminorm at the scale λ \lambda λ
[ F ] C λ N ( T 2 ) : = ∑ 1 ≤ ∣ j ∣ ≤ N λ ∣ j ∣ ∥ ∂ j F ∥ L ∞ ( T 2 ) j ! , ∥ F ∥ C λ N ( T 2 ) : = ∥ F ∥ L ∞ ( T 2 ) + [ F ] C λ N ( T 2 ) , (40) [F]_{C^{N}_{\lambda}(\mathbb{T}^{2})}:=\sum_{1\leq|j|\leq N}\frac{\lambda^{|j|}\|\partial^{j}F\|_{L^{\infty}(\mathbb{T}^{2})}}{j!},\qquad\|F\|_{C^{N}_{\lambda}(\mathbb{T}^{2})}:=\|F\|_{L^{\infty}(\mathbb{T}^{2})}+[F]_{C^{N}_{\lambda}(\mathbb{T}^{2})}, \tag*{(40)} [ F ] C λ N ( T 2 ) := 1 ≤ ∣ j ∣ ≤ N ∑ j ! λ ∣ j ∣ ∥ ∂ j F ∥ L ∞ ( T 2 ) , ∥ F ∥ C λ N ( T 2 ) := ∥ F ∥ L ∞ ( T 2 ) + [ F ] C λ N ( T 2 ) , (40)
where j = ( j 1 , j 2 ) j=(j_{1},j_{2}) j = ( j 1 , j 2 ) runs over multi-indices, ∣ j ∣ : = j 1 + j 2 |j|:=j_{1}+j_{2} ∣ j ∣ := j 1 + j 2 and j ! : = j 1 ! j 2 ! j!:=j_{1}!j_{2}! j ! := j 1 ! j 2 ! . A representation R R R of U U U at wavelength λ ∗ \lambda_{*} λ ∗ is a decomposition into functions in C N r e p ( T 2 ) V C^{N_{\mathrm{rep}}}(\mathbb{T}^{2})^{V} C N rep ( T 2 ) V , a base E E E and a finite collection P \mathcal{P} P of profiles P P P (repetitions allowed), each with a wavelength λ P ∈ [ λ ∗ , 1 ] \lambda_{P}\in[\lambda_{*},1] λ P ∈ [ λ ∗ , 1 ] , such that U = E + ∑ P ∈ P P U=E+\sum_{P\in\mathcal{P}}P U = E + ∑ P ∈ P P . Its size is
size λ ∗ ( R ) : = ∥ E ∥ L ∞ ( T 2 ) + [ E ] C 0 , s ( T 2 ) + λ ∗ − s [ E ] C λ ∗ N r e p ( T 2 ) + ∑ P ∈ P λ P − s ∥ P ∥ C λ P N r e p ( T 2 ) . (41) \operatorname{size}_{\lambda_{*}}(R):=\|E\|_{L^{\infty}(\mathbb{T}^{2})}+[E]_{C^{0,s}(\mathbb{T}^{2})}+\lambda_{*}^{-s}[E]_{C^{N_{\mathrm{rep}}}_{\lambda_{*}}(\mathbb{T}^{2})}+\sum_{P\in\mathcal{P}}\lambda_{P}^{-s}\|P\|_{C^{N_{\mathrm{rep}}}_{\lambda_{P}}(\mathbb{T}^{2})}. \tag*{(41)} size λ ∗ ( R ) := ∥ E ∥ L ∞ ( T 2 ) + [ E ] C 0 , s ( T 2 ) + λ ∗ − s [ E ] C λ ∗ N rep ( T 2 ) + P ∈ P ∑ λ P − s ∥ P ∥ C λ P N rep ( T 2 ) . (41)
Then ∥ U ∥ C 0 , s ( T 2 ) ≤ 3 size λ ∗ ( R ) \|U\|_{C^{0,s}(\mathbb{T}^{2})}\leq3\operatorname{size}_{\lambda_{*}}(R) ∥ U ∥ C 0 , s ( T 2 ) ≤ 3 size λ ∗ ( R ) , by Lemma 14.2(a). The readout Υ m \Upsilon_{m} Υ m , like every constant linear map, maps a representation R R R of a field Y Y Y of G ′ G' G ′ to a representation of Υ m Y \Upsilon_{m}Y Υ m Y of size at most ∣ Υ m ∣ size λ ∗ ( R ) |\Upsilon_{m}|\operatorname{size}_{\lambda_{*}}(R) ∣ Υ m ∣ size λ ∗ ( R ) , where ∣ Υ m ∣ |\Upsilon_{m}| ∣ Υ m ∣ is its operator norm for the norms ∣ ⋅ ∣ ω |\cdot|_{\omega} ∣ ⋅ ∣ ω . Each profile keeps its wavelength: measured at λ ∗ < λ P \lambda_{*}<\lambda_{P} λ ∗ < λ P , its term λ P − s ∥ P ∥ L ∞ ( T 2 ) \lambda_{P}^{-s}\|P\|_{L^{\infty}(\mathbb{T}^{2})} λ P − s ∥ P ∥ L ∞ ( T 2 ) would grow by the factor ( λ P / λ ∗ ) s (\lambda_{P}/\lambda_{*})^{s} ( λ P / λ ∗ ) s .
Initialization and induction
Proposition 5.1 (Initialization from rough data). There exist M ∗ ( α , s , q , ζ , Λ , n , t 0 , T ) ∈ N M_{*}(\alpha,s,q,\zeta,\Lambda,n,t_{0},T)\in\mathbb{N} M ∗ ( α , s , q , ζ , Λ , n , t 0 , T ) ∈ N and C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ C(\alpha,s,q,\zeta,\Lambda,n,t_{0},T)<\infty C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ with the following property. For every M ≥ M ∗ M\geq M_{*} M ≥ M ∗ , κ ∈ K M \kappa\in\mathcal{K}_{M} κ ∈ K M and t i n ∈ R t_{\mathrm{in}}\in\mathbb{R} t in ∈ R , with t f : = t i n + T t_{\mathrm{f}}:=t_{\mathrm{in}}+T t f := t in + T , there exist t e ∈ [ t i n , t i n + 1 4 t 0 ] t_{\mathrm{e}}\in[t_{\mathrm{in}},t_{\mathrm{in}}+\frac{1}{4}t_{0}] t e ∈ [ t in , t in + 4 1 t 0 ] and an admissible seed G G G of level M − 1 M-1 M − 1 on [ t e , t f ] [t_{\mathrm{e}},t_{\mathrm{f}}] [ t e , t f ] such that the following holds. For every θ i n ∈ L 2 ( T 2 ) \theta_{\mathrm{in}}\in L^{2}(\mathbb{T}^{2}) θ in ∈ L 2 ( T 2 ) and every weak solution θ \theta θ of (4.1), there exists a datum g g g of G G G with ∥ g ∥ L ω 2 ( T 2 ) ≤ C ∥ θ i n ∥ L 2 ( T 2 ) \|g\|_{L^{2}_{\omega}(\mathbb{T}^{2})}\leq C\|\theta_{\mathrm{in}}\|_{L^{2}(\mathbb{T}^{2})} ∥ g ∥ L ω 2 ( T 2 ) ≤ C ∥ θ in ∥ L 2 ( T 2 ) such that the solution U U U of G G G from g g g at time t e t_{\mathrm{e}} t e has the following property: for every interval [ t 1 , t 2 ] ⊂ [ t i n + 1 4 t 0 , t f ] [t_{1},t_{2}]\subset[t_{\mathrm{in}}+\frac{1}{4}t_{0},t_{\mathrm{f}}] [ t 1 , t 2 ] ⊂ [ t in + 4 1 t 0 , t f ] and every Ξ ≥ 0 \Xi\geq0 Ξ ≥ 0 , if, for almost every t ∈ [ t 1 , t 2 ] t\in[t_{1},t_{2}] t ∈ [ t 1 , t 2 ] , there exists a representation R ( t ) R(t) R ( t ) of U ( t ) U(t) U ( t ) at wavelength ε M − 1 \varepsilon_{M-1} ε M − 1 with size ε M − 1 ( R ( t ) ) ≤ Ξ \operatorname{size}_{\varepsilon_{M-1}}(R(t))\leq\Xi size ε M − 1 ( R ( t )) ≤ Ξ , then
∥ θ ∥ L ∞ ( [ t 1 , t 2 ] ; C 0 , s ( T 2 ) ) ≤ C ( Ξ + ∥ θ i n ∥ L 2 ( T 2 ) ) . \|\theta\|_{L^{\infty}([t_{1},t_{2}];C^{0,s}(\mathbb{T}^{2}))}\leq C\left(\Xi+\|\theta_{\mathrm{in}}\|_{L^{2}(\mathbb{T}^{2})}\right). ∥ θ ∥ L ∞ ([ t 1 , t 2 ] ; C 0 , s ( T 2 )) ≤ C ( Ξ + ∥ θ in ∥ L 2 ( T 2 ) ) .
Idea of the proof. At the levels m ∈ ( M , M + n ] m\in(M,M+n] m ∈ ( M , M + n ] the Péclet number ε m β / κ \varepsilon_{m}^{\beta}/\kappa ε m β / κ is at most C ε m − 1 ( q + 2 ) γ C\varepsilon_{m-1}^{(q+2)\gamma} C ε m − 1 ( q + 2 ) γ , and molecular diffusion over a quiet margin reaches the wavelength ε m − 1 \varepsilon_{m-1} ε m − 1 of the next coarser level: ε m − 1 2 / ( κ τ m ′ ) ≤ C ε m − 1 γ − 3 δ \varepsilon_{m-1}^{2}/(\kappa\tau_{m}')\leq C\varepsilon_{m-1}^{\gamma-3\delta} ε m − 1 2 / ( κ τ m ′ ) ≤ C ε m − 1 γ − 3 δ , with γ > 3 δ \gamma>3\delta γ > 3 δ . So a weak passage compares the solution with a family of level m − 1 m-1 m − 1 at the same diffusivity, which keeps the enhancement of level m m m as a source instead of adding it to the diffusivity, and regains derivatives on the margins. After n n n of them the sources are smoothed once, by isotropic paired convolutions of width slightly below ε M \varepsilon_{M} ε M . At level M M M a margin is too short, κ τ M ′ ≫ ε M − 1 2 + γ + 3 δ ≪ ε M − 1 2 \kappa\tau_{M}'\gg\varepsilon_{M-1}^{2+\gamma+3\delta}\ll\varepsilon_{M-1}^{2} κ τ M ′ ≫ ε M − 1 2 + γ + 3 δ ≪ ε M − 1 2 , and a terminal cycle , prepared over a time of order h M h_{M} h M by the method of Proposition 5.3, passes to κ M − 1 ∈ [ κ , C κ ] \kappa_{M-1}\in[\kappa,C\kappa] κ M − 1 ∈ [ κ , C κ ] and yields the seed. For Proposition 4.1, θ \theta θ is, up to C ε M − 1 2 ∥ θ i n ∥ L 2 ( T 2 ) C\varepsilon_{M-1}^{2}\|\theta_{\mathrm{in}}\|_{L^{2}(\mathbb{T}^{2})} C ε M − 1 2 ∥ θ in ∥ L 2 ( T 2 ) in L ∞ ( [ t i n + 1 4 t 0 , t i n + T ] ; C 0 , s ( T 2 ) ) L^{\infty}([t_{\mathrm{in}}+\frac{1}{4}t_{0},t_{\mathrm{in}}+T];C^{0,s}(\mathbb{T}^{2})) L ∞ ([ t in + 4 1 t 0 , t in + T ] ; C 0 , s ( T 2 )) , the lift of U U U through the terminal cycle and the n n n weak passages (variants of (38) , defined in § 6 ), and these lifts carry a representation at ε M − 1 \varepsilon_{M-1} ε M − 1 to one at ε M + n \varepsilon_{M+n} ε M + n with fewer than N r e p N_{\mathrm{rep}} N rep derivatives, at a constant cost.
Proposition 5.2 (Induction on the levels). There exists C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ C(\alpha,s,q,\zeta,\Lambda,n,t_{0},T)<\infty C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ with the following property. Let m 0 < m < M m_{0}<m<M m 0 < m < M and κ ∈ K M \kappa\in\mathcal{K}_{M} κ ∈ K M , let t s , t f ∈ R t_{\mathrm{s}},t_{\mathrm{f}}\in\mathbb{R} t s , t f ∈ R with t s + 1 2 t 0 < t f ≤ t s + T t_{\mathrm{s}}+\frac{1}{2}t_{0}<t_{\mathrm{f}}\leq t_{\mathrm{s}}+T t s + 2 1 t 0 < t f ≤ t s + T , let G G G be an admissible seed of level m m m on [ t s , t f ] [t_{\mathrm{s}},t_{\mathrm{f}}] [ t s , t f ] , let g g g be a datum of G G G , and let U U U be the solution of G G G
from g g g at time t S t_{\mathrm{S}} t S . Then, for almost every t ∈ [ t S + 1 2 t 0 , t f ] t \in[t_{\mathrm{S}} + \frac{1}{2}t_0,t_{\mathrm{f}}] t ∈ [ t S + 2 1 t 0 , t f ] , there exists a representation R ( t ) R(t) R ( t ) of U ( t ) U(t) U ( t ) at wavelength ε m \varepsilon_m ε m such that
size ε m ( R ( t ) ) ≤ C ∥ g ∥ L ω 2 ( T 2 ) . \operatorname{size}_{\varepsilon_m}(R(t)) \le C\lVert g\rVert_{L^2_\omega(\mathbb{T}^2)}. size ε m ( R ( t )) ≤ C ∥ g ∥ L ω 2 ( T 2 ) .
The constant is independent of m m m , that is, of the number of cycles. Idea of the proof. This is the induction of Subsection 5.2: we apply Propositions 5.3 and 5.4 to G G G and its iterated successors, which gives {(xref)}(5.3) and {(xref)}(5.4), with G = G m G=G_m G = G m , G ′ = G m − 1 G'=G_{m-1} G ′ = G m − 1 , U = U m U=U_m U = U m , Y = U m − 1 Y=U_{m-1} Y = U m − 1 , t p r = t ( m − 1 ) t_{\mathrm{pr}}=t_{(m-1)} t pr = t ( m − 1 ) , R = R m ( t ) R=R_m(t) R = R m ( t ) and R ′ = R m − 1 ( t ) R'=R_{m-1}(t) R ′ = R m − 1 ( t ) . The times of the cycles are chosen recursively (Step 2 of the proof in Subsection 14.4), and they are less than t S + 1 2 t 0 t_{\mathrm{S}}+\frac{1}{2}t_0 t S + 2 1 t 0 , since m 0 m_0 m 0 is chosen so that
∑ j = m 0 + 1 m ( 3 τ j + 4 h j ) < 1 2 t 0 . \sum_{j=m_0+1}^{m}(3\tau_j+4h_j)<\frac{1}{2}t_0. j = m 0 + 1 ∑ m ( 3 τ j + 4 h j ) < 2 1 t 0 .
Proof of Proposition 4.1 . Let Λ 0 ( α , s , q , ζ ) ≥ 128 \Lambda_0(\alpha,s,q,\zeta)\ge128 Λ 0 ( α , s , q , ζ ) ≥ 128 be at least the thresholds of Proposition 2.1 and of the statements of Part II; this choice is the first row of the table of the order of the parameters in Subsection 11.1. Let Λ ≥ Λ 0 \Lambda\ge\Lambda_0 Λ ≥ Λ 0 , n ≥ 1 n\ge1 n ≥ 1 and 0 < t 0 < T 0<t_0<T 0 < t 0 < T be given. Let M ∗ M_* M ∗ be as in Proposition 5.1 , with m 0 m_0 m 0 the top level of Subsection 5.3, and let M ≥ M 1 : = max { M ∗ , m 0 + 2 } M\ge M_1:=\max\{M_*,m_0+2\} M ≥ M 1 := max { M ∗ , m 0 + 2 } . Let G G G be the seed of Proposition 5.1 and g g g its datum. Proposition 5.2 applies to G G G and g g g at level M − 1 M-1 M − 1 , with t S = t e t_{\mathrm{S}}=t_{\mathrm{e}} t S = t e and t f = t i n + T t_{\mathrm{f}}=t_{\mathrm{in}}+T t f = t in + T , since m 0 < M − 1 < M m_0<M-1<M m 0 < M − 1 < M , t e + 1 2 t 0 ≤ t i n + 1 4 t 0 < t i n + T t_{\mathrm{e}}+\frac{1}{2}t_0\le t_{\mathrm{in}}+\frac{1}{4}t_0<t_{\mathrm{in}}+T t e + 2 1 t 0 ≤ t in + 4 1 t 0 < t in + T and t i n + T ≤ t e + T t_{\mathrm{in}}+T\le t_{\mathrm{e}}+T t in + T ≤ t e + T ; it gives for almost every t ∈ [ t e + 1 2 t 0 , t i n + T ] ⊃ [ t i n + t 0 , t i n + T ] t\in[t_{\mathrm{e}}+\frac{1}{2}t_0,t_{\mathrm{in}}+T]\supset[t_{\mathrm{in}}+t_0,t_{\mathrm{in}}+T] t ∈ [ t e + 2 1 t 0 , t in + T ] ⊃ [ t in + t 0 , t in + T ] a representation R ( t ) R(t) R ( t ) of U ( t ) U(t) U ( t ) at wavelength ε M − 1 \varepsilon_{M-1} ε M − 1 with size ε M − 1 ( R ( t ) ) ≤ C ∥ g ∥ L 2 ( T 2 ) ≤ C ∥ θ i n ∥ L 2 ( T 2 ) \operatorname{size}_{\varepsilon_{M-1}}(R(t))\le C\lVert g\rVert_{L^2(\mathbb{T}^2)}\le C\lVert\theta_{\mathrm{in}}\rVert_{L^2(\mathbb{T}^2)} size ε M − 1 ( R ( t )) ≤ C ∥ g ∥ L 2 ( T 2 ) ≤ C ∥ θ in ∥ L 2 ( T 2 ) . Proposition 5.1 then applies with [ t 1 , t 2 ] = [ t i n + t 0 , t i n + T ] [t_1,t_2]=[t_{\mathrm{in}}+t_0,t_{\mathrm{in}}+T] [ t 1 , t 2 ] = [ t in + t 0 , t in + T ] and Ξ = C ∥ θ i n ∥ L 2 ( T 2 ) \Xi=C\lVert\theta_{\mathrm{in}}\rVert_{L^2(\mathbb{T}^2)} Ξ = C ∥ θ in ∥ L 2 ( T 2 ) , and bounds ∥ θ ∥ L ∞ ( [ t i n + t 0 , t i n + T ] ; C 0 , s ( T 2 ) ) \lVert\theta\rVert_{L^\infty([t_{\mathrm{in}}+t_0,t_{\mathrm{in}}+T];C^{0,s}(\mathbb{T}^2))} ∥ θ ∥ L ∞ ([ t in + t 0 , t in + T ] ; C 0 , s ( T 2 )) by C ∥ θ i n ∥ L 2 ( T 2 ) C\lVert\theta_{\mathrm{in}}\rVert_{L^2(\mathbb{T}^2)} C ∥ θ in ∥ L 2 ( T 2 ) .
One homogenization cycle and the passage to the next level
A cycle at level m m m has two parts (Subsection 5.2): the preparation, part (a) below, regenerates derivatives within one period 4 h m 4h_m 4 h m , and the comparison, part (b), bounds the difference between the solution and the lift of the solution of a successor. Proposition 5.4 reads the cycle on representations.
Proposition 5.3 (One homogenization cycle). There exists C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ C(\alpha,s,q,\zeta,\Lambda,n,t_0,T)<\infty C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ with the following property. Let M ≥ 1 M\ge1 M ≥ 1 , κ ∈ K M \kappa\in\mathcal{K}_M κ ∈ K M and m 0 < m < M m_0<m<M m 0 < m < M , let t A , t c , t f ∈ R t_{\mathrm{A}},t_{\mathrm{c}},t_{\mathrm{f}}\in\mathbb{R} t A , t c , t f ∈ R with t A + 3 τ m ≤ t c t_{\mathrm{A}}+3\tau_m\le t_{\mathrm{c}} t A + 3 τ m ≤ t c and t c + 4 h m ≤ t f t_{\mathrm{c}}+4h_m\le t_{\mathrm{f}} t c + 4 h m ≤ t f , let G G G be an admissible family of level m m m on [ t A , t f ] [t_{\mathrm{A}},t_{\mathrm{f}}] [ t A , t f ] , and let G ′ G' G ′ be a successor of G G G . There exist t p r ∈ [ t c , t c + 4 h m ] t_{\mathrm{pr}}\in[t_{\mathrm{c}},t_{\mathrm{c}}+4h_m] t pr ∈ [ t c , t c + 4 h m ] and a linear map Π : L 2 ( T 2 ) V → H N o u t ( T 2 ) V \Pi:L^2(\mathbb{T}^2)^V\to H^{N_{\mathrm{out}}}(\mathbb{T}^2)^V Π : L 2 ( T 2 ) V → H N out ( T 2 ) V such that, for every datum g g g of G G G , the solution U U U of G G G from g g g at time t c t_{\mathrm{c}} t c and the solution Y Y Y of G ′ G' G ′ from the retained trace of Π g \Pi g Π g at time t p r t_{\mathrm{pr}} t pr satisfy the following.
(a) The preparation Π g \Pi g Π g approximates U ( t p r ) U(t_{\mathrm{pr}}) U ( t pr ) , is almost no larger than the datum, and has N o u t N_{\mathrm{out}} N out derivatives at the scale ρ ( m ) \rho(m) ρ ( m ) :
∥ Π g − U ( t p r ) ∥ L G 2 ( T 2 ) ≤ C ε m − 1 Q ∥ g ∥ L G 2 ( T 2 ) , ∥ Π g ∥ L G 2 ( T 2 ) ≤ ( 1 + C ε m − 1 Q ) ∥ g ∥ L G 2 ( T 2 ) , \lVert\Pi g-U(t_{\mathrm{pr}})\rVert_{L^2_G(\mathbb{T}^2)}
\le C\varepsilon_{m-1}^{Q}\lVert g\rVert_{L^2_G(\mathbb{T}^2)},\qquad
\lVert\Pi g\rVert_{L^2_G(\mathbb{T}^2)}
\le(1+C\varepsilon_{m-1}^{Q})\lVert g\rVert_{L^2_G(\mathbb{T}^2)}, ∥ Π g − U ( t pr ) ∥ L G 2 ( T 2 ) ≤ C ε m − 1 Q ∥ g ∥ L G 2 ( T 2 ) , ∥ Π g ∥ L G 2 ( T 2 ) ≤ ( 1 + C ε m − 1 Q ) ∥ g ∥ L G 2 ( T 2 ) , ∥ Π g ∥ H ρ ( m ) , G N o u t ( T 2 ) ≤ C ∥ g ∥ L G 2 ( T 2 ) . \lVert\Pi g\rVert_{H^{N_{\mathrm{out}}}_{\rho(m),G}(\mathbb{T}^2)}
\le C\lVert g\rVert_{L^2_G(\mathbb{T}^2)}. ∥ Π g ∥ H ρ ( m ) , G N out ( T 2 ) ≤ C ∥ g ∥ L G 2 ( T 2 ) . (b) The successor G ′ G' G ′ is admissible on [ t A + 3 τ m , t f ] [t_{\mathrm{A}}+3\tau_m,t_{\mathrm{f}}] [ t A + 3 τ m , t f ] , and
∥ Y ∥ L ∞ ( [ t p r , t f ] ; L G ′ 2 ( T 2 ) ) ≤ ( 1 + C ( ε m − 1 Q + ε m − 1 δ ) ) ∥ g ∥ L G 2 ( T 2 ) , \lVert Y\rVert_{L^\infty([t_{\mathrm{pr}},t_{\mathrm{f}}];L^2_{G'}(\mathbb{T}^2))}
\le(1+C(\varepsilon_{m-1}^{Q}+\varepsilon_{m-1}^{\delta}))\lVert g\rVert_{L^2_G(\mathbb{T}^2)}, ∥ Y ∥ L ∞ ([ t pr , t f ] ; L G ′ 2 ( T 2 )) ≤ ( 1 + C ( ε m − 1 Q + ε m − 1 δ )) ∥ g ∥ L G 2 ( T 2 ) , ∥ U − Y ♯ ∥ L ∞ ( [ t p r , t f ] ; L ω 2 ( T 2 ) ) + κ m 1 / 2 ∥ ∇ ( U − Y ♯ ) ∥ L 2 ( ( t p r , t f ) ; L ω 2 ( T 2 ) ) ≤ C ε m − 1 Q ∥ g ∥ L G 2 ( T 2 ) , \lVert U-Y^{\sharp}\rVert_{L^\infty([t_{\mathrm{pr}},t_{\mathrm{f}}];L^2_\omega(\mathbb{T}^2))}
+\kappa_m^{1/2}\lVert\nabla(U-Y^{\sharp})\rVert_{L^2((t_{\mathrm{pr}},t_{\mathrm{f}});L^2_\omega(\mathbb{T}^2))}
\le C\varepsilon_{m-1}^{Q}\lVert g\rVert_{L^2_G(\mathbb{T}^2)}, ∥ U − Y ♯ ∥ L ∞ ([ t pr , t f ] ; L ω 2 ( T 2 )) + κ m 1/2 ∥ ∇ ( U − Y ♯ ) ∥ L 2 (( t pr , t f ) ; L ω 2 ( T 2 )) ≤ C ε m − 1 Q ∥ g ∥ L G 2 ( T 2 ) , ∥ Y ∥ L ∞ ( [ t p r , t f ] ; H ρ ( m ) , G ′ N r e t ( T 2 ) ) ≤ C ∥ g ∥ L G 2 ( T 2 ) . \lVert Y\rVert_{L^\infty([t_{\mathrm{pr}},t_{\mathrm{f}}];H^{N_{\mathrm{ret}}}_{\rho(m),G'}(\mathbb{T}^2))}
\le C\lVert g\rVert_{L^2_G(\mathbb{T}^2)}. ∥ Y ∥ L ∞ ([ t pr , t f ] ; H ρ ( m ) , G ′ N ret ( T 2 )) ≤ C ∥ g ∥ L G 2 ( T 2 ) .
In (a), the preparation time t p r t_{\mathrm{pr}} t pr is chosen so that the velocity added at level m m m vanishes near it: b m = b m − 1 b_m=b_{m-1} b m = b m − 1 on [ t p r − τ m ′ , t p r + τ m ′ ] × T 2 [t_{\mathrm{pr}}-\tau'_m,t_{\mathrm{pr}}+\tau'_m]\times\mathbb{T}^2 [ t pr − τ m ′ , t pr + τ m ′ ] × T 2 (Proposition 12.1 (i)), so that Corr m \operatorname{Corr}_m Corr m adds no corrector at t p r t_{\mathrm{pr}} t pr and Y ♯ ( t p r ) = exp p ( Γ m ( t p r ) : ∇ 2 ) exp p ( − Γ m ( t p r ) : ∇ 2 ) Π g Y^{\sharp}(t_{\mathrm{pr}})=\exp_{\mathrm{p}}(\Gamma_m(t_{\mathrm{pr}}):\nabla^2)\exp_{\mathrm{p}}(-\Gamma_m(t_{\mathrm{pr}}):\nabla^2)\Pi g Y ♯ ( t pr ) = exp p ( Γ m ( t pr ) : ∇ 2 ) exp p ( − Γ m ( t pr ) : ∇ 2 ) Π g (Lemma 13.1 ). Since Π g \Pi g Π g has N o u t = N r e t + 4 p N_{\mathrm{out}}=N_{\mathrm{ret}}+4p N out = N ret + 4 p derivatives and the retained trace uses 2 p 2p 2 p of them, N r e t + 2 p N_{\mathrm{ret}}+2p N ret + 2 p remain; on [ t p r , t f ] [t_{\mathrm{pr}},t_{\mathrm{f}}] [ t pr , t f ] the third bound of (b) gives N r e t N_{\mathrm{ret}} N ret derivatives of Y Y Y , at least the 2 p + N c o r 2p+N_{\mathrm{cor}} 2 p + N cor that its lift uses (2 p 2p 2 p for the truncated exponential and N c o r N_{\mathrm{cor}} N cor for the correctors).
Idea of the proof. (a): in the Duhamel expansion of U U U in the sources, each term alternates free evolutions under the operator of level m m m with single applications of a source. The Bloch transform splits a free evolution, in the Lagrangian coordinates of a renewal block and with the conductivity frozen at a point, into evolutions on the period cell, one for each frequency. The principal branch of these cell evolutions dissipates over a complete renewal block inside [ t c , t c + 4 h m ] [t_{\mathrm{c}},t_{\mathrm{c}}+4h_m] [ t c , t c + 4 h m ] at a rate proportional to the square of the frequency, in every direction and uniformly in the level, and the two orientations of the pulses damp the high frequencies. (b): the correctors cancel the leading terms of ( ∂ t + b m ⋅ ∇ − κ m Δ ) Y ♯ (\partial_t+b_m\cdot\nabla-\kappa_m\Delta)Y^\sharp ( ∂ t + b m ⋅ ∇ − κ m Δ ) Y ♯ . The errors from switching the correctors on and off in the intervals between consecutive pulses are exponentially small, the errors of the smoothing operators are small by the estimates of Proposition 6.1 , and the energy estimate with forcing at order zero (Proposition 8.3 (a) at N = 0 N=0 N = 0 ) bounds the error by the residual, which is controlled by the derivatives of (a). The coefficients of the successor are explicit combinations of those of G G G , each carrying a power of ε m − 1 \varepsilon_{m-1} ε m − 1 .
Proposition 5.4 (Passage between consecutive levels). There exists C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ C(\alpha,s,q,\zeta,\Lambda,n,t_0,T)<\infty C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ with the following property. For every M ≥ 1 M\geq1 M ≥ 1 , κ ∈ K M \kappa\in K_M κ ∈ K M , m 0 < m < M m_0<m<M m 0 < m < M and every G G G and G ′ G' G ′ as in Proposition 5.3 at level m m m , there exist t p r t_{\mathrm{pr}} t pr and Π \Pi Π satisfying (a) and (b) of Proposition 5.3 with this C C C such that, for every datum g g g of G G G , with U U U and Y Y Y as there, and every t ∈ [ t p r , t f ] t\in[t_{\mathrm{pr}},t_{\mathrm{f}}] t ∈ [ t pr , t f ] , the following hold: U ( t ) U(t) U ( t ) and Y ( t ) Y(t) Y ( t ) are of class C N r e p C^{N_{\mathrm{rep}}} C N rep on T 2 \mathbb{T}^2 T 2 , and for every representation R ′ R' R ′ of Y ( t ) Y(t) Y ( t ) at wavelength ε m − 1 \varepsilon_{m-1} ε m − 1 there exists a representation R R R of U ( t ) U(t) U ( t ) at wavelength ε m \varepsilon_m ε m such that
size ε m ( R ) ≤ ( 1 + C ε m − 1 η ) size ε m − 1 ( R ′ ) + C ε m − 1 ∥ g ∥ L G 2 ( T 2 ) . (42) \operatorname{size}_{\varepsilon_m}(R)\leq\left(1+C\varepsilon_{m-1}^{\eta}\right)\operatorname{size}_{\varepsilon_{m-1}}(R')+C\varepsilon_{m-1}\lVert g\rVert_{L^2_G(\mathbb{T}^2)}. \tag*{(42)} size ε m ( R ) ≤ ( 1 + C ε m − 1 η ) size ε m − 1 ( R ′ ) + C ε m − 1 ∥ g ∥ L G 2 ( T 2 ) . (42)
Proposition 5.3 is proved in Sections 12 and 13, and Proposition 5.4 in Section 14. Idea of the proof. Let S S S be a smoothing below the scale ρ ( m ) \rho(m) ρ ( m ) . Then U U U is the sum of Γ m Y \Gamma_mY Γ m Y (old profiles copied by a constant map), the exponential and corrector parts of the lift of S Y SY S Y and of ( I − S ) Y (I-S)Y ( I − S ) Y , and the error U − Y ♯ U-Y^\sharp U − Y ♯ . On S Y SY S Y the corrector part costs (30) and the exponential part a power of ε m − 1 \varepsilon_{m-1} ε m − 1 . On ( I − S ) Y (I-S)Y ( I − S ) Y the moments of S S S (Proposition 6.1 ) make both parts small, and U − Y ♯ U-Y^\sharp U − Y ♯ , which is small in L ω 2 ( T 2 ) L^2_\omega(\mathbb{T}^2) L ω 2 ( T 2 ) by Proposition 5.3(b) and smooth at the molecular radius ρ m o l ( m ) \rho_{\mathrm{mol}}(m) ρ mol ( m ) of (115) , is small at the wavelength ε m \varepsilon_m ε m by interpolation (Step 6 of the proof in Section 14).
Energy estimates and paired smoothing
Fix a smooth function u ≥ 0 u\geq0 u ≥ 0 of integral one supported in the unit disc, and, for an integer ℓ ≥ 1 \ell\geq1 ℓ ≥ 1 , let W ℓ W_\ell W ℓ be the kernel with Fourier transform
W ℓ ^ ( ξ ) : = 1 − ( 1 − ∣ u ^ ( ξ / ( 2 ℓ ) ) ∣ 2 ) ℓ , ξ ∈ R 2 . (43) \widehat{W_\ell}(\xi):=1-\left(1-\lvert\widehat{u}(\xi/(2\ell))\rvert^2\right)^\ell,\qquad\xi\in\mathbb{R}^2. \tag*{(43)} W ℓ ( ξ ) := 1 − ( 1 − ∣ u ( ξ / ( 2 ℓ )) ∣ 2 ) ℓ , ξ ∈ R 2 . (43)
It is smooth, supported in the unit disc, of integral one, with vanishing moments of orders 1 1 1 to 2 ℓ − 1 2\ell-1 2 ℓ − 1 , and W ℓ ^ \widehat{W_\ell} W ℓ takes values in [ 0 , 1 ] [0,1] [ 0 , 1 ] . For a linear map Ω \Omega Ω of R 2 \mathbb{R}^2 R 2 and r > 0 r>0 r > 0 , the operator
S ℓ , r , Ω g ( y ) : = ∫ R 2 W ℓ ( z ) g ( y − r Ω z ) d z (44) S_{\ell,r,\Omega}g(y):=\int_{\mathbb{R}^2}W_\ell(z)\,g(y-r\Omega z)\,\mathrm{d}z \tag*{(44)} S ℓ , r , Ω g ( y ) := ∫ R 2 W ℓ ( z ) g ( y − r Ω z ) d z (44)
takes the mean of g g g over the ellipse { y − r Ω z : ∣ z ∣ < 1 } \{y-r\Omega z:\lvert z\rvert<1\} { y − r Ω z : ∣ z ∣ < 1 } (a segment or a point if Ω \Omega Ω is singular) with the signed weight W ℓ W_\ell W ℓ .
The proofs
Proposition 6.1 (Paired smoothing in eccentric geometry). Let ℓ ≥ 1 \ell\ge1 ℓ ≥ 1 and 1 ≤ N 1 ≤ N 2 1 \le N_1 \le N_2 1 ≤ N 1 ≤ N 2 be integers with N 1 < 2 ℓ N_1 < 2\ell N 1 < 2 ℓ . There exists C ( u , ℓ , N 1 , N 2 ) < ∞ C(u,\ell,N_1,N_2) < \infty C ( u , ℓ , N 1 , N 2 ) < ∞ with the following property. Let C Ω > 0 C_\Omega> 0 C Ω > 0 , and let Ω 1 , Ω 2 \Omega_1,\Omega_2 Ω 1 , Ω 2 be invertible linear maps of R 2 \mathbb{R}^2 R 2 with ∣ Ω i ∣ ∞ ≤ C Ω \lvert\Omega_i\rvert_\infty\le C_\Omega ∣ Ω i ∣ ∞ ≤ C Ω whose vectors Ω 1 e 1 , Ω 2 e 2 \Omega_1 e_1,\Omega_2 e_2 Ω 1 e 1 , Ω 2 e 2 are uniformly transverse, in the sense that the matrix with these columns has an inverse of ∣ ⋅ ∣ ∞ \lvert\cdot\rvert_\infty ∣ ⋅ ∣ ∞ -norm at most 2. Then, for every ρ > 0 \rho> 0 ρ > 0 , 0 < r ≤ ρ 0 < r \le\rho 0 < r ≤ ρ and g ∈ H N 1 ( T 2 ) g \in H^{N_1}(\mathbb{T}^2) g ∈ H N 1 ( T 2 ) , the operator S : = S ℓ , r , Ω 1 S ℓ , r , Ω 2 S := S_{\ell,r,\Omega_1}S_{\ell,r,\Omega_2} S := S ℓ , r , Ω 1 S ℓ , r , Ω 2 satisfies
∥ S g ∥ H ρ N 2 ( T 2 ) ≤ C ( ρ r ) N 2 − N 1 ∥ g ∥ H ρ N 1 ( T 2 ) , ∥ S g ∥ H ρ N 1 ( T 2 ) ≤ ∥ g ∥ H ρ N 1 ( T 2 ) , \lVert Sg\rVert_{H_\rho^{N_2}(\mathbb{T}^2)} \le C\left(\frac{\rho}{r}\right)^{N_2-N_1}\lVert g\rVert_{H_\rho^{N_1}(\mathbb{T}^2)}, \qquad\lVert Sg\rVert_{H_\rho^{N_1}(\mathbb{T}^2)} \le\lVert g\rVert_{H_\rho^{N_1}(\mathbb{T}^2)}, ∥ S g ∥ H ρ N 2 ( T 2 ) ≤ C ( r ρ ) N 2 − N 1 ∥ g ∥ H ρ N 1 ( T 2 ) , ∥ S g ∥ H ρ N 1 ( T 2 ) ≤ ∥ g ∥ H ρ N 1 ( T 2 ) , ∥ g − S g ∥ L 2 ( T 2 ) ≤ C C Ω N 1 ( r ρ ) N 1 ∥ g ∥ H ρ N 1 ( T 2 ) . \lVert g-Sg\rVert_{L^2(\mathbb{T}^2)} \le CC_\Omega^{N_1}\left(\frac{r}{\rho}\right)^{N_1}\lVert g\rVert_{H_\rho^{N_1}(\mathbb{T}^2)}. ∥ g − S g ∥ L 2 ( T 2 ) ≤ C C Ω N 1 ( ρ r ) N 1 ∥ g ∥ H ρ N 1 ( T 2 ) . The third bound holds for all linear maps Ω 1 , Ω 2 \Omega_1,\Omega_2 Ω 1 , Ω 2 of R 2 \mathbb{R}^2 R 2 with ∣ Ω i ∣ ∞ ≤ C Ω \lvert\Omega_i\rvert_\infty\le C_\Omega ∣ Ω i ∣ ∞ ≤ C Ω and all r , ρ > 0 r,\rho> 0 r , ρ > 0 . The constant of the third bound depends only on u u u and N 1 N_1 N 1 .
Two ellipses whose vectors Ω 1 e 1 , Ω 2 e 2 \Omega_1e_1,\Omega_2e_2 Ω 1 e 1 , Ω 2 e 2 are transverse together smooth in every direction, whatever their eccentricity; in the applications these vectors are the long axes. For the sources of a seed the ellipses are round discs. Idea of the proof. Each ∂ i \partial_i ∂ i is a combination of the derivatives along Ω 1 e 1 \Omega_1e_1 Ω 1 e 1 and Ω 2 e 2 \Omega_2e_2 Ω 2 e 2 , and along each of these vectors the corresponding smoothing gains one power of ρ / r \rho/r ρ / r per derivative, while its vanishing moments give the approximation.
Proposition 6.2 (Energy of coupled systems). Let ν > 0 \nu> 0 ν > 0 and t 1 < t 2 t_1 < t_2 t 1 < t 2 , let N s r c ≥ 2 N_{\mathrm{src}} \ge2 N src ≥ 2 be the order of Subsection 5.3, let Φ \Phi Φ be Z 2 \mathbb{Z}^2 Z 2 -periodic and smooth in x x x , with ∇ ⊥ Φ \nabla^\perp\Phi ∇ ⊥ Φ bounded and Borel measurable in ( t , x ) (t,x) ( t , x ) , and let V V V be a finite set with masses ω v > 0 \omega_v > 0 ω v > 0 . Let finitely many sources σ : v ′ → v \sigma:v' \to v σ : v ′ → v , v , v ′ ∈ V v,v' \in V v , v ′ ∈ V , be given, with coefficients A σ , d A_{\sigma,d} A σ , d , 2 ≤ d ≤ N s r c 2 \le d \le N_{\mathrm{src}} 2 ≤ d ≤ N src , that are Borel measurable in t t t and bounded, continuous and periodic in x x x , and linear operators S σ ( t ) S_\sigma(t) S σ ( t ) on L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) that commute with derivatives and such that t ↦ S σ ( t ) g ∈ L 2 ( T 2 ) t \mapsto S_\sigma(t)g \in L^2(\mathbb{T}^2) t ↦ S σ ( t ) g ∈ L 2 ( T 2 ) is Borel measurable for every g ∈ L 2 ( T 2 ) g \in L^2(\mathbb{T}^2) g ∈ L 2 ( T 2 ) . Let constants c j ≥ 0 c_j \ge0 c j ≥ 0 and widths r σ > 0 r_\sigma> 0 r σ > 0 be such that, for every σ \sigma σ , t ∈ [ t 1 , t 2 ] t \in[t_1,t_2] t ∈ [ t 1 , t 2 ] , g ∈ L 2 ( T 2 ) g \in L^2(\mathbb{T}^2) g ∈ L 2 ( T 2 ) and j ≤ N s r c − 2 j \le N_{\mathrm{src}}-2 j ≤ N src − 2 ,
∥ ∇ j S σ ( t ) g ∥ L 2 ( T 2 ) ≤ c j r σ − j ∥ g ∥ L 2 ( T 2 ) . \lVert\nabla^j S_\sigma(t)g\rVert_{L^2(\mathbb{T}^2)} \le c_j r_\sigma^{-j}\lVert g\rVert_{L^2(\mathbb{T}^2)}. ∥ ∇ j S σ ( t ) g ∥ L 2 ( T 2 ) ≤ c j r σ − j ∥ g ∥ L 2 ( T 2 ) . Consider the problem
{ ∂ t U v + ∇ ⊥ Φ ⋅ ∇ U v − ν Δ U v = ∇ ⋅ ∑ v ′ ∈ V ∑ σ : v ′ → v ∑ d = 2 N s r c A σ , d ∇ d − 1 ( S σ U v ′ ) in ( t 1 , t 2 ) × T 2 , v ∈ V , U v ( t 1 , ⋅ ) = g v on T 2 , (45) \begin{cases}
\partial_t U_v+\nabla^\perp\Phi\cdot\nabla U_v-\nu\Delta U_v
=\nabla\cdot\sum_{v'\in V}\sum_{\sigma:v'\to v}\sum_{d=2}^{N_{\mathrm{src}}} A_{\sigma,d}\nabla^{d-1}(S_\sigma U_{v'}) & \text{in }(t_1,t_2)\times\mathbb{T}^2,\quad v\in V,\\
U_v(t_1,\cdot)=g_v & \text{on }\mathbb{T}^2, \tag*{(45)}
\end{cases} { ∂ t U v + ∇ ⊥ Φ ⋅ ∇ U v − ν Δ U v = ∇ ⋅ ∑ v ′ ∈ V ∑ σ : v ′ → v ∑ d = 2 N src A σ , d ∇ d − 1 ( S σ U v ′ ) U v ( t 1 , ⋅ ) = g v in ( t 1 , t 2 ) × T 2 , v ∈ V , on T 2 , (45) If all row and column sums of the coupling matrix
B v v ′ : = ( ω v ′ ω v ) 1 / 2 ∑ σ : v ′ → v ∑ d = 2 N s r c c d − 2 sup [ t 1 , t 2 ] × T 2 ∣ A σ , d ∣ ν r σ d − 2 B_{vv'}:=\left(\frac{\omega_{v'}}{\omega_v}\right)^{1/2}\sum_{\sigma:v'\to v}\sum_{d=2}^{N_{\mathrm{src}}}c_{d-2}\frac{\sup_{[t_1,t_2]\times\mathbb{T}^2}\lvert A_{\sigma,d}\rvert}{\nu r_\sigma^{d-2}} B v v ′ := ( ω v ω v ′ ) 1/2 σ : v ′ → v ∑ d = 2 ∑ N src c d − 2 ν r σ d − 2 sup [ t 1 , t 2 ] × T 2 ∣ A σ , d ∣ are at most 1 8 \frac{1}{8} 8 1 , then, for every datum g = ( g v ) v ∈ V ∈ L 2 ( T 2 ) V g=(g_v)_{v\in V}\in L^2(\mathbb{T}^2)^V g = ( g v ) v ∈ V ∈ L 2 ( T 2 ) V , there exists a unique solution U ∈ C ( [ t 1 , t 2 ] ; L 2 ( T 2 ) V ) ∩ L 2 ( ( t 1 , t 2 ) ; H 1 ( T 2 ) V ) U\in C([t_1,t_2];L^2(\mathbb{T}^2)^V)\cap L^2((t_1,t_2);H^1(\mathbb{T}^2)^V) U ∈ C ([ t 1 , t 2 ] ; L 2 ( T 2 ) V ) ∩ L 2 (( t 1 , t 2 ) ; H 1 ( T 2 ) V ) of (45) (in the integrated sense of (67) , without forcing), and, for every t ′ , t ∈ [ t 1 , t 2 ] t',t\in[t_1,t_2] t ′ , t ∈ [ t 1 , t 2 ] with t ′ ≤ t t'\le t t ′ ≤ t ,
∥ U ( t ) ∥ L ω 2 ( T 2 ) 2 + ν ∥ ∇ U ∥ L 2 ( ( t ′ , t ) ; L ω 2 ( T 2 ) ) 2 ≤ ∥ U ( t ′ ) ∥ L ω 2 ( T 2 ) 2 . \lVert U(t)\rVert_{L^2_\omega(\mathbb{T}^2)}^2+\nu\lVert\nabla U\rVert_{L^2((t',t);L^2_\omega(\mathbb{T}^2))}^2\le\lVert U(t')\rVert_{L^2_\omega(\mathbb{T}^2)}^2. ∥ U ( t ) ∥ L ω 2 ( T 2 ) 2 + ν ∥ ∇ U ∥ L 2 (( t ′ , t ) ; L ω 2 ( T 2 )) 2 ≤ ∥ U ( t ′ ) ∥ L ω 2 ( T 2 ) 2 .
Idea of the proof. We test the equation for U v U_v U v with U v / ω v U_v/\omega_v U v / ω v and bound the coupling by Schur’s test, as after (37) . Its analogue at higher order, Proposition 8.3 , gives the derivatives of Y Y Y used after Proposition 5.3 and the C N r e p C^{N_{\mathrm{rep}}} C N rep regularity in Proposition 5.4 .
Part II. The proofs
Properties of the velocity field: flows and effective diffusivities
This section proves the properties of the construction that § 1.1 states without proof, and Proposition 2.1 . Other sections use the flows of Subsection 7.3: the short flows carry the tubes of Theorem C, and the renewal charts are the coordinates of § 5.2 –§ 5.8 .
Throughout, Λ ≥ 128 \Lambda\ge128 Λ ≥ 128 is an integer, and C C C and c c c denote positive constants depending only on α , q , ζ \alpha,q,\zeta α , q , ζ , unless another dependence is stated; they may change from line to line. Matrices and arrays are normed as in § 1.1 .
Scales and cutoffs
Lemma 7.1 (Scales and cutoffs). For every integer m ≥ 1 m \ge1 m ≥ 1 the following hold.
(a) Wavelengths. ε m − 1 / ε m ≥ Λ \varepsilon_{m-1}/\varepsilon_m \ge\Lambda ε m − 1 / ε m ≥ Λ and ε m ≤ Λ − m \varepsilon_m \le\Lambda^{-m} ε m ≤ Λ − m . If m ≥ 2 m \ge2 m ≥ 2 , then 1 2 ε m − 1 q ≤ ε m ≤ 2 ε m − 1 q \frac{1}{2}\varepsilon_{m-1}^q \le\varepsilon_m \le2\varepsilon_{m-1}^q 2 1 ε m − 1 q ≤ ε m ≤ 2 ε m − 1 q .
(b) Times. 4 ε m − 1 − δ ≤ 4 ⌈ ε m − 1 − δ ⌉ + 1 ≤ 9 ε m − 1 − δ 4\varepsilon_{m-1}^{-\delta} \le4\lceil\varepsilon_{m-1}^{-\delta}\rceil+ 1 \le9\varepsilon_{m-1}^{-\delta} 4 ε m − 1 − δ ≤ 4 ⌈ ε m − 1 − δ ⌉ + 1 ≤ 9 ε m − 1 − δ ; (7) and (15) hold, and ε m ≤ 2 35 τ m \varepsilon_m \le2^{35}\tau_m ε m ≤ 2 35 τ m .
(c) Exponents. q ( β − γ ) = β + γ q(\beta-\gamma)=\beta+\gamma q ( β − γ ) = β + γ , q ( 2 − β − γ ) = 2 − β + 8 δ q(2-\beta-\gamma)=2-\beta+8\delta q ( 2 − β − γ ) = 2 − β + 8 δ , ( q − 1 ) ( β + γ ) = 2 q γ (q-1)(\beta+\gamma)=2q\gamma ( q − 1 ) ( β + γ ) = 2 q γ , 3 δ < γ 3\delta<\gamma 3 δ < γ , and q − ( 2 − β + 4 δ ) = α + ( q + 1 2 ) γ > 0 q-(2-\beta+4\delta)=\alpha+(q+\frac{1}{2})\gamma>0 q − ( 2 − β + 4 δ ) = α + ( q + 2 1 ) γ > 0 .
(d) Cutoffs. Let k , l ∈ Z k,l\in\mathbb{Z} k , l ∈ Z and let j ≥ 0 j\ge0 j ≥ 0 be an integer. There exist C ( ζ , j ) C(\zeta,j) C ( ζ , j ) and C ( j ) C(j) C ( j ) such that
∥ ∂ t j ζ m , k ∥ L ∞ ( R ) ≤ C ( ζ , j ) τ m − j \lVert\partial_t^j\zeta_{m,k}\rVert_{L^\infty(\mathbb{R})}\le C(\zeta,j)\tau_m^{-j} ∥ ∂ t j ζ m , k ∥ L ∞ ( R ) ≤ C ( ζ , j ) τ m − j and
∥ ∂ t j ζ ^ m , l ∥ L ∞ ( R ) ≤ C ( j ) ( τ m ′ ) − j . \lVert\partial_t^j\widehat{\zeta}_{m,l}\rVert_{L^\infty(\mathbb{R})}\le C(j)(\tau'_m)^{-j}. ∥ ∂ t j ζ m , l ∥ L ∞ ( R ) ≤ C ( j ) ( τ m ′ ) − j . The fast cutoff ζ m , k \zeta_{m,k} ζ m , k vanishes outside [ ( k − 2 3 ) τ m , ( k + 2 3 ) τ m ] [(k-\frac{2}{3})\tau_m,(k+\frac{2}{3})\tau_m] [( k − 3 2 ) τ m , ( k + 3 2 ) τ m ] . At each time at most two fast cutoffs are nonzero, and at most one with k k k odd. The slow cutoff satisfies 0 ≤ ζ ^ m , l ≤ 1 0\le\widehat{\zeta}_{m,l}\le1 0 ≤ ζ m , l ≤ 1 and
1 [ ( l − 1 2 ) h m + 2 τ m ′ , ( l + 1 2 ) h m − 2 τ m ′ ] ≤ ζ ^ m , l ≤ 1 ( ( l − 1 2 ) h m + τ m ′ , ( l + 1 2 ) h m − τ m ′ ) , (46) \mathbf{1}_{[(l-\frac{1}{2})h_m+2\tau'_m,(l+\frac{1}{2})h_m-2\tau'_m]}
\le\widehat{\zeta}_{m,l}\le
\mathbf{1}_{((l-\frac{1}{2})h_m+\tau'_m,(l+\frac{1}{2})h_m-\tau'_m)}, \tag*{(46)} 1 [( l − 2 1 ) h m + 2 τ m ′ , ( l + 2 1 ) h m − 2 τ m ′ ] ≤ ζ m , l ≤ 1 (( l − 2 1 ) h m + τ m ′ , ( l + 2 1 ) h m − τ m ′ ) , (46) where 1 E \mathbf{1}_E 1 E is the indicator of a set E E E . The slot [ ( k − 1 2 ) τ m , ( k + 1 2 ) τ m ] [(k-\frac{1}{2})\tau_m,(k+\frac{1}{2})\tau_m] [( k − 2 1 ) τ m , ( k + 2 1 ) τ m ] lies in the block [ ( l m , k − 1 2 ) h m , ( l m , k + 1 2 ) h m ] [(l_{m,k}-\frac{1}{2})h_m,(l_{m,k}+\frac{1}{2})h_m] [( l m , k − 2 1 ) h m , ( l m , k + 2 1 ) h m ] .
Finally, for α , s > 0 \alpha,s>0 α , s > 0 a number q q q satisfying (3) exists if and only if α < 1 3 \alpha<\frac{1}{3} α < 3 1 and s < 1 2 ( 1 − α ) s<\frac{1}{2}(1-\alpha) s < 2 1 ( 1 − α ) ; and for q > 1 q>1 q > 1 , δ > 0 \delta>0 δ > 0 if and only if q < 1 − α 2 α q<\frac{1-\alpha}{2\alpha} q < 2 α 1 − α , while μ > 0 \mu>0 μ > 0 if and only if q < 1 − s α + s q<\frac{1-s}{\alpha+s} q < α + s 1 − s .
Proof. (a) Let m ≥ 1 m\ge1 m ≥ 1 . Since Λ q m / ( q − 1 ) ≥ Λ 1 / ( q − 1 ) ≥ log Λ / ( q − 1 ) \Lambda^{q^m/(q-1)}\ge\Lambda^{1/(q-1)}\ge\log\Lambda/(q-1) Λ q m / ( q − 1 ) ≥ Λ 1/ ( q − 1 ) ≥ log Λ/ ( q − 1 ) and Λ q m − 1 − 1 ≥ ( q − 1 ) log Λ \Lambda^{q^m-1}-1\ge(q-1)\log\Lambda Λ q m − 1 − 1 ≥ ( q − 1 ) log Λ ,
Λ q m + 1 / ( q − 1 ) − Λ ⋅ Λ q m / ( q − 1 ) = Λ ⋅ Λ q m / ( q − 1 ) ( Λ q m − 1 − 1 ) ≥ Λ ( log Λ ) 2 ≥ Λ . \Lambda^{q^{m+1}/(q-1)}-\Lambda\cdot\Lambda^{q^m/(q-1)}
=\Lambda\cdot\Lambda^{q^m/(q-1)}\left(\Lambda^{q^m-1}-1\right)
\ge\Lambda(\log\Lambda)^2\ge\Lambda. Λ q m + 1 / ( q − 1 ) − Λ ⋅ Λ q m / ( q − 1 ) = Λ ⋅ Λ q m / ( q − 1 ) ( Λ q m − 1 − 1 ) ≥ Λ ( log Λ ) 2 ≥ Λ.
So Λ ε m − 1 ≤ Λ ( Λ q m / ( q − 1 ) + 1 ) ≤ ε m + 1 − 1 \Lambda\varepsilon_m^{-1}\le\Lambda(\Lambda^{q^m/(q-1)}+1)\le\varepsilon_{m+1}^{-1} Λ ε m − 1 ≤ Λ ( Λ q m / ( q − 1 ) + 1 ) ≤ ε m + 1 − 1 by (6) ; with ε 0 = 1 \varepsilon_0=1 ε 0 = 1 and ε 1 − 1 ≥ Λ q / ( q − 1 ) ≥ Λ \varepsilon_1^{-1}\ge\Lambda^{q/(q-1)}\ge\Lambda ε 1 − 1 ≥ Λ q / ( q − 1 ) ≥ Λ , the first two bounds follow. For m ≥ 2 m\ge2 m ≥ 2 , Λ q m − 1 / ( q − 1 ) ≥ 128 \Lambda^{q^{m-1}/(q-1)}\ge128 Λ q m − 1 / ( q − 1 ) ≥ 128 , so its ceiling is at most 129 128 \frac{129}{128} 128 129 times it, and ( 129 128 ) q < 2 (\frac{129}{128})^q<2 ( 128 129 ) q < 2 gives ε m − 1 − q ≤ 2 Λ q m / ( q − 1 ) \varepsilon_{m-1}^{-q}\le2\Lambda^{q^m/(q-1)} ε m − 1 − q ≤ 2 Λ q m / ( q − 1 ) . Hence ε m ≤ Λ − q m / ( q − 1 ) ≤ 2 ε m − 1 q \varepsilon_m\le\Lambda^{-q^m/(q-1)}\le2\varepsilon_{m-1}^q ε m ≤ Λ − q m / ( q − 1 ) ≤ 2 ε m − 1 q , and ε m ≥ 1 2 Λ − q m / ( q − 1 ) ≥ 1 2 ε m − 1 q \varepsilon_m\ge\frac{1}{2}\Lambda^{-q^m/(q-1)}\ge\frac{1}{2}\varepsilon_{m-1}^q ε m ≥ 2 1 Λ − q m / ( q − 1 ) ≥ 2 1 ε m − 1 q , since the ceiling of a number at least 1 1 1 is at most twice the number.
(b) By (6) , h m = 2 − 25 / ⌈ ε m − 1 − ( 2 − β + 2 δ ) ⌉ h_m=2^{-25}/\lceil\varepsilon_{m-1}^{-(2-\beta+2\delta)}\rceil h m = 2 − 25 / ⌈ ε m − 1 − ( 2 − β + 2 δ ) ⌉ , and τ m ′ \tau'_m τ m ′ and τ m \tau_m τ m are h m h_m h m divided once and twice by 4 ⌈ ε m − 1 − δ ⌉ + 1 4\lceil\varepsilon_{m-1}^{-\delta}\rceil+1 4 ⌈ ε m − 1 − δ ⌉ + 1 , so the bounds y ≤ ⌈ y ⌉ ≤ 2 y y\le\lceil y\rceil\le2y y ≤ ⌈ y ⌉ ≤ 2 y for y ≥ 1 y\ge1 y ≥ 1 give the first bounds, (7) , h m ≤ 2 − 25 ε m − 1 2 − β + 2 δ h_m\le2^{-25}\varepsilon_{m-1}^{2-\beta+2\delta} h m ≤ 2 − 25 ε m − 1 2 − β + 2 δ , which is (15) , and τ m ≥ 2 − 26 ε m − 1 2 − β + 4 δ / 81 \tau_m\ge2^{-26}\varepsilon_{m-1}^{2-\beta+4\delta}/81 τ m ≥ 2 − 26 ε m − 1 2 − β + 4 δ /81 . For m ≥ 2 m\ge2 m ≥ 2 , (a) and (c) then give ε m ≤ 2 ε m − 1 q ≤ 2 ε m − 1 2 − β + 4 δ ≤ 2 35 τ m \varepsilon_m\le2\varepsilon_{m-1}^q\le2\varepsilon_{m-1}^{2-\beta+4\delta}\le2^{35}\tau_m ε m ≤ 2 ε m − 1 q ≤ 2 ε m − 1 2 − β + 4 δ ≤ 2 35 τ m ; for m = 1 m=1 m = 1 , ε 1 ≤ Λ − 1 ≤ 2 35 τ 1 \varepsilon_1\le\Lambda^{-1}\le2^{35}\tau_1 ε 1 ≤ Λ − 1 ≤ 2 35 τ 1 . (c) and the equivalences are algebra on (3) and (5) . (d) follows from (4) , (9) , (11) and the properties of H; the centres of odd slots are 2 τ m 2\tau_m 2 τ m apart, and ζ m , k \zeta_{m,k} ζ m , k has support of length 4 3 τ m \frac{4}{3}\tau_m 3 4 τ m . Finally, h m / τ m h_m/\tau_m h m / τ m is the odd integer ( 4 ⌈ ε m − 1 − δ ⌉ + 1 ) 2 (4\lceil\varepsilon_{m-1}^{-\delta}\rceil+1)^2 ( 4 ⌈ ε m − 1 − δ ⌉ + 1 ) 2 , so k + 1 2 ( h m / τ m − 1 ) k+\frac{1}{2}(h_m/\tau_m-1) k + 2 1 ( h m / τ m − 1 ) is an integer, and (10) gives
l m , k h m − 1 2 ( h m − τ m ) ≤ k τ m ≤ l m , k h m + 1 2 ( h m − τ m ) . l_{m,k}h_m-\frac{1}{2}(h_m-\tau_m)\le k\tau_m\le l_{m,k}h_m+\frac{1}{2}(h_m-\tau_m). l m , k h m − 2 1 ( h m − τ m ) ≤ k τ m ≤ l m , k h m + 2 1 ( h m − τ m ) .
In particular, the envelope ζ ^ m , k = ζ ^ m , l m , k ζ m , k \widehat{\zeta}_{m,k}=\widehat{\zeta}_{m,l_m,k}\zeta_{m,k} ζ m , k = ζ m , l m , k ζ m , k of slot k k k vanishes outside the block of l m , k l_{m,k} l m , k and within τ m ′ \tau'_m τ m ′ of its ends. Cutoffs as in (4) exist: for 0 < w ≤ 13 80 0<w\le\frac{13}{80} 0 < w ≤ 80 13 , the even function equal to 1 1 1 on [ 0 , 1 2 − w ] [0,\frac{1}{2}-w] [ 0 , 2 1 − w ] , to 0 0 0 on [ 1 2 + w , ∞ ) [\frac{1}{2}+w,\infty) [ 2 1 + w , ∞ ) and affine in between, convolved with an even mollifier of width 1 1000 \frac{1}{1000} 1000 1 , satisfies the first two conditions and lies in C c ∞ ( ( − 2 3 , 2 3 ) ) C^\infty_c((-\frac{2}{3},\frac{2}{3})) C c ∞ (( − 3 2 , 3 2 )) ; its square integral is continuous in w w w , above 9 10 \frac{9}{10} 10 9 at w = 1 10 w=\frac{1}{10} w = 10 1 and at most 1 − 2 3 w < 9 10 1-\frac{2}{3}w<\frac{9}{10} 1 − 3 2 w < 10 9 at w = 13 80 w=\frac{13}{80} w = 80 13 , hence equal to 9 10 \frac{9}{10} 10 9 in between.
Reliance on [5 ] . The proofs of [5 Proposition 2.2 and Corollary 2.4] use the scales and the cutoffs only through the facts of Lemma 7.1 . These facts hold for every q q q satisfying (3) , every integer Λ ≥ 128 \Lambda\ge128 Λ ≥ 128 and every ζ \zeta ζ as in § 1.1 , so those proofs apply to our construction, with ε m ≤ 2 35 τ m \varepsilon_m\le2^{35}\tau_m ε m ≤ 2 35 τ m of Lemma 7.1 (b) in place of the exponent that [5 ] computes in [5 (2.61)] from its choice of q q q . We write out the arguments that differ.
The velocity field
The functions ϕ m \phi_m ϕ m and b m \mathbf{b}_m b m of Lemma 7.2 are well defined and smooth by Proposition 7.3 (i), whose proof does not use Lemma 7.2 .
Lemma 7.2 (Increments of the stream functions). There exists C ( α , q , ζ , Λ ) < ∞ C(\alpha,q,\zeta,\Lambda)<\infty C ( α , q , ζ , Λ ) < ∞ such that, for every m ≥ 1 m\ge1 m ≥ 1 and i , i ′ ∈ { 1 , 2 } i,i'\in\{1,2\} i , i ′ ∈ { 1 , 2 } ,
∥ ϕ m − ϕ m − 1 ∥ L ∞ ( R × T 2 ) ≤ 10 ε m β , ∥ ∂ i ( ϕ m − ϕ m − 1 ) ∥ L ∞ ( R × T 2 ) ≤ 320 ε m α , \|\phi_m-\phi_{m-1}\|_{L^\infty(\mathbb{R}\times\mathbb{T}^2)}\le10\varepsilon_m^\beta,\qquad\|\partial_i(\phi_m-\phi_{m-1})\|_{L^\infty(\mathbb{R}\times\mathbb{T}^2)}\le320\varepsilon_m^\alpha, ∥ ϕ m − ϕ m − 1 ∥ L ∞ ( R × T 2 ) ≤ 10 ε m β , ∥ ∂ i ( ϕ m − ϕ m − 1 ) ∥ L ∞ ( R × T 2 ) ≤ 320 ε m α , ∥ ∂ i ∂ i ′ ( ϕ m − ϕ m − 1 ) ∥ L ∞ ( R × T 2 ) ≤ 36500 ε m β − 2 , (47) \|\partial_i\partial_{i'}(\phi_m-\phi_{m-1})\|_{L^\infty(\mathbb{R}\times\mathbb{T}^2)}\le36500\varepsilon_m^{\beta-2}, \tag*{(47)} ∥ ∂ i ∂ i ′ ( ϕ m − ϕ m − 1 ) ∥ L ∞ ( R × T 2 ) ≤ 36500 ε m β − 2 , (47) and, for every t , t ′ ∈ R t,t'\in\mathbb{R} t , t ′ ∈ R and x , y ∈ R 2 x,y\in\mathbb{R}^2 x , y ∈ R 2 ,
∣ ( b m − b m − 1 ) ( t , x ) − ( b m − b m − 1 ) ( t ′ , y ) ∣ ≤ C ( ∣ t − t ′ ∣ + ∣ x − y ∣ ) ε m β − 2 . (48) \left|(\mathbf{b}_m-\mathbf{b}_{m-1})(t,x)-(\mathbf{b}_m-\mathbf{b}_{m-1})(t',y)\right|\le C\left(|t-t'|+|x-y|\right)\varepsilon_m^{\beta-2}. \tag*{(48)} ∣ ( b m − b m − 1 ) ( t , x ) − ( b m − b m − 1 ) ( t ′ , y ) ∣ ≤ C ( ∣ t − t ′ ∣ + ∣ x − y ∣ ) ε m β − 2 . (48)
Proof. The bounds (47) are [5 Proposition 2.2, (2.39)] : the norm [5 (1.21)] at the radius 2 7 / ε m 2^7/\varepsilon_m 2 7 / ε m bounds each partial derivative of order j j j by j ! ( 2 7 / ε m ) j / ( j + 1 ) 2 j!(2^7/\varepsilon_m)^j/(j+1)^2 j ! ( 2 7 / ε m ) j / ( j + 1 ) 2 times the norm, so the bound 10 ε m β 10\varepsilon_m^\beta 10 ε m β gives 320 ε m α 320\varepsilon_m^\alpha 320 ε m α and 36500 ε m β − 2 36500\varepsilon_m^{\beta-2} 36500 ε m β − 2 . The bound (48) follows from them and from the bound ∣ ∂ t ( b m − b m − 1 ) ∣ ≤ C ε m β − 2 |\partial_t(\mathbf{b}_m-\mathbf{b}_{m-1})|\le C\varepsilon_m^{\beta-2} ∣ ∂ t ( b m − b m − 1 ) ∣ ≤ C ε m β − 2 in the proof of [5 Proposition 2.2] , after [5 (2.61)] ; the first term of that bound, C ε m β − 2 ε m / τ m C\varepsilon_m^{\beta-2}\varepsilon_m/\tau_m C ε m β − 2 ε m / τ m , is at most 2 35 C ε m β − 2 2^{35}C\varepsilon_m^{\beta-2} 2 35 C ε m β − 2 by Lemma 7.1 (b).
Proposition 7.3 (The velocity field). The following hold.
(i) For every m ≥ 0 m\ge0 m ≥ 0 , ϕ m \phi_m ϕ m is well defined, smooth and Z × Z 2 \mathbb{Z}\times\mathbb{Z}^2 Z × Z 2 -periodic, and b m \mathbf{b}_m b m is divergence free.
(ii) For every m ≥ 0 m\ge0 m ≥ 0 ,
∥ ∇ b m ∥ L ∞ ( R × T 2 ) = ∥ ∇ 2 ϕ m ∥ L ∞ ( R × T 2 ) ≤ 78000 ε m β − 2 . (49) \|\nabla\mathbf{b}_m\|_{L^\infty(\mathbb{R}\times\mathbb{T}^2)}=\|\nabla^2\phi_m\|_{L^\infty(\mathbb{R}\times\mathbb{T}^2)}\le78000\varepsilon_m^{\beta-2}. \tag*{(49)} ∥∇ b m ∥ L ∞ ( R × T 2 ) = ∥ ∇ 2 ϕ m ∥ L ∞ ( R × T 2 ) ≤ 78000 ε m β − 2 . (49) (iii) As m → ∞ m\to\infty m → ∞ , ϕ m \phi_m ϕ m and b m \mathbf{b}_m b m converge uniformly on R × R 2 \mathbb{R}\times\mathbb{R}^2 R × R 2 to functions ϕ \phi ϕ and b \mathbf{b} b . The function ϕ \phi ϕ is continuous, Z 2 \mathbb{Z}^2 Z 2 -periodic in x x x and 1 1 1 -periodic in t t t ; for every t ∈ R t\in\mathbb{R} t ∈ R , ϕ ( t , ⋅ ) ∈ C 1 ( R 2 ) \phi(t,\cdot)\in C^1(\mathbb{R}^2) ϕ ( t , ⋅ ) ∈ C 1 ( R 2 ) and ∇ ⊥ ϕ ( t , ⋅ ) = b ( t , ⋅ ) \nabla^\perp\phi(t,\cdot)=\mathbf{b}(t,\cdot) ∇ ⊥ ϕ ( t , ⋅ ) = b ( t , ⋅ ) ; and (16) holds.
(iv) b \mathbf{b} b is bounded, continuous, Z 2 \mathbb{Z}^2 Z 2 -periodic in x x x and 1 1 1 -periodic in t t t ; for every t ∈ R t\in\mathbb{R} t ∈ R , the field b ( t , ⋅ ) \mathbf{b}(t,\cdot) b ( t , ⋅ ) is divergence free in the sense of distributions, and (17) holds.
The constants in (16) and (17) depend only on α , q , ζ , Λ \alpha,q,\zeta,\Lambda α , q , ζ , Λ .
Proof. (i) By induction on m m m : if ϕ m − 1 \phi_{m-1} ϕ m − 1 is smooth and Z × Z 2 \mathbb{Z}\times\mathbb{Z}^2 Z × Z 2 -periodic, its flow X m − 1 X_{m-1} X m − 1 exists, is smooth, and, by uniqueness, commutes with the translations by Z 2 \mathbb{Z}^2 Z 2 and satisfies $X_{m-1}(t+1,x,t'+1)=
X m − 1 ( t , x , t ′ ) X_{m-1}(t,x,t') X m − 1 ( t , x , t ′ ) . By Lemma 7.1(d), finitely many terms of (14) are nonzero near each time, and each ψ m , k \psi_{m,k} ψ m , k is Z 2 \mathbb{Z}^{2} Z 2 -periodic, since ε m − 1 \varepsilon_m^{-1} ε m − 1 is an integer; so ϕ m \phi_m ϕ m is smooth and Z 2 \mathbb{Z}^{2} Z 2 -periodic in x x x . It is 1-periodic in t t t because 1 / h m 1/h_m 1/ h m and 1 / τ m 1/\tau_m 1/ τ m are integers divisible by 4, by (6) : a shift of time by 1 shifts k k k by 1 / τ m 1/\tau_m 1/ τ m and ℓ m , k \ell_{m,k} ℓ m , k by 1 / h m 1/h_m 1/ h m , and keeps the class of k k k modulo 4. Finally, b m = ∇ ⊥ ϕ m b_m=\nabla^\perp\phi_m b m = ∇ ⊥ ϕ m is divergence free.
(ii) As ϕ 0 = 0 \phi_0=0 ϕ 0 = 0 , Lemma 7.2 gives ∣ ∂ i ∂ i ′ ϕ m ∣ ≤ 36500 ∑ m ′ = 1 m ε m ′ β − 2 |\partial_i\partial_{i'}\phi_m|\le36500\sum_{m'=1}^{m}\varepsilon_{m'}^{\beta-2} ∣ ∂ i ∂ i ′ ϕ m ∣ ≤ 36500 ∑ m ′ = 1 m ε m ′ β − 2 , and ε m ′ β − 2 ≤ Λ ( 2 − β ) ( m ′ − m ) 16 m ′ − m ε m β − 2 \varepsilon_{m'}^{\beta-2}\le\Lambda^{(2-\beta)(m'-m)}16^{m'-m}\varepsilon_m^{\beta-2} ε m ′ β − 2 ≤ Λ ( 2 − β ) ( m ′ − m ) 1 6 m ′ − m ε m β − 2 by Lemma 7.1(a), since 2 − β = 1 − α > 2 3 2-\beta=1-\alpha>\frac{2}{3} 2 − β = 1 − α > 3 2 gives Λ 2 − β ≥ 128 2 / 3 > 16 \Lambda^{2-\beta}\ge128^{2/3}>16 Λ 2 − β ≥ 12 8 2/3 > 16 . So every entry of ∇ 2 ϕ m \nabla^2\phi_m ∇ 2 ϕ m is at most 16 15 ⋅ 36500 ε m β − 2 ≤ 39000 ε m β − 2 \frac{16}{15}\cdot36500\varepsilon_m^{\beta-2}\le39000\varepsilon_m^{\beta-2} 15 16 ⋅ 36500 ε m β − 2 ≤ 39000 ε m β − 2 in absolute value. The norm of a 2 × 2 2\times2 2 × 2 matrix is at most twice its largest entry, and ∇ b m \nabla b_m ∇ b m is ∇ 2 ϕ m \nabla^2\phi_m ∇ 2 ϕ m composed with a rotation, which gives (49) .
(iii) By Lemma 7.2 and ε m ′ + 1 ≤ Λ − 1 ε m ′ \varepsilon_{m'+1}\le\Lambda^{-1}\varepsilon_{m'} ε m ′ + 1 ≤ Λ − 1 ε m ′ , the sums ∑ m ′ > m ∣ ϕ m ′ − ϕ m ′ − 1 ∣ \sum_{m'>m}|\phi_{m'}-\phi_{m'-1}| ∑ m ′ > m ∣ ϕ m ′ − ϕ m ′ − 1 ∣ and ∑ m ′ > m ∣ b m ′ − b m ′ − 1 ∣ \sum_{m'>m}|b_{m'}-b_{m'-1}| ∑ m ′ > m ∣ b m ′ − b m ′ − 1 ∣ are at most 10 ε m + 1 β 1 − Λ − β \frac{10\varepsilon_{m+1}^{\beta}}{1-\Lambda^{-\beta}} 1 − Λ − β 10 ε m + 1 β and 2 ⋅ 320 ε m + 1 α 1 − Λ − α \frac{2\cdot320\varepsilon_{m+1}^{\alpha}}{1-\Lambda^{-\alpha}} 1 − Λ − α 2 ⋅ 320 ε m + 1 α , uniformly in ( t , x ) (t,x) ( t , x ) . Hence ϕ m \phi_m ϕ m and b m = ∇ ⊥ ϕ m b_m=\nabla^\perp\phi_m b m = ∇ ⊥ ϕ m converge uniformly, and (16) holds. The limit ϕ \phi ϕ is continuous; since ϕ m ( t , ⋅ ) \phi_m(t,\cdot) ϕ m ( t , ⋅ ) and its gradient converge uniformly, ϕ ( t , ⋅ ) \phi(t,\cdot) ϕ ( t , ⋅ ) is C 1 C^1 C 1 with ∇ ⊥ ϕ = b \nabla^\perp\phi=b ∇ ⊥ ϕ = b ; and the periodicities persist.
(iv) By (iii), b b b is continuous and periodic, and ∣ b ∣ ≤ ∑ m ′ ≥ 1 640 ε m ′ α ≤ C |b|\le\sum_{m'\ge1}640\varepsilon_{m'}^\alpha\le C ∣ b ∣ ≤ ∑ m ′ ≥ 1 640 ε m ′ α ≤ C . Each b m ( t , ⋅ ) b_m(t,\cdot) b m ( t , ⋅ ) is divergence free, and this passes to the uniform limit in the sense of distributions. For (17) , let t , t ′ ∈ R t,t'\in\mathbb{R} t , t ′ ∈ R and x , y ∈ R 2 x,y\in\mathbb{R}^2 x , y ∈ R 2 . If ( t , x ) = ( t ′ , y ) (t,x)=(t',y) ( t , x ) = ( t ′ , y ) there is nothing to prove, and if ∣ t − t ′ ∣ + ∣ x − y ∣ ≥ 1 |t-t'|+|x-y|\ge1 ∣ t − t ′ ∣ + ∣ x − y ∣ ≥ 1 , (17) follows from the bound on b b b . Otherwise, let m ‾ ≥ 1 \overline{m}\ge1 m ≥ 1 be the least level with ε m ‾ ≤ ∣ t − t ′ ∣ + ∣ x − y ∣ \varepsilon_{\overline{m}}\le|t-t'|+|x-y| ε m ≤ ∣ t − t ′ ∣ + ∣ x − y ∣ . We write b = ∑ m ′ ≥ 1 ( b m ′ − b m ′ − 1 ) b=\sum_{m'\ge1}(b_{m'}-b_{m'-1}) b = ∑ m ′ ≥ 1 ( b m ′ − b m ′ − 1 ) , bound the terms with m ′ < m ‾ m'<\overline{m} m ′ < m by (48) and those with m ′ ≥ m ‾ m'\ge\overline{m} m ′ ≥ m by (47) at each of the two points:
∣ b ( t , x ) − b ( t ′ , y ) ∣ ≤ C ∣ t − t ′ ∣ + ∣ x − y ∣ ∑ 1 ≤ m ′ < m ‾ ε m ′ β − 2 + 1280 ∑ m ′ ≥ m ‾ ε m ′ α ≤ C ∣ t − t ′ ∣ + ∣ x − y ∣ ε m ‾ − 1 α − 1 + C ε m ‾ α ≤ C ∣ t − t ′ ∣ + ∣ x − y ∣ α . \begin{aligned}
\lvert b(t,x)-b(t',y)\rvert\le C\lvert t-t'\rvert+\lvert x-y\rvert\sum_{1\le m'<\overline{m}}\varepsilon_{m'}^{\beta-2}+1280\sum_{m'\ge\overline{m}}\varepsilon_{m'}^\alpha\\
&\le C\lvert t-t'\rvert+\lvert x-y\rvert\varepsilon_{\overline{m}-1}^{\alpha-1}+C\varepsilon_{\overline{m}}^\alpha\le C\lvert t-t'\rvert+\lvert x-y\rvert^\alpha.
\end{aligned} ∣ b ( t , x ) − b ( t ′ , y )∣ ≤ C ∣ t − t ′ ∣ + ∣ x − y ∣ 1 ≤ m ′ < m ∑ ε m ′ β − 2 + 1280 m ′ ≥ m ∑ ε m ′ α ≤ C ∣ t − t ′ ∣ + ∣ x − y ∣ ε m − 1 α − 1 + C ε m α ≤ C ∣ t − t ′ ∣ + ∣ x − y ∣ α .
since β − 2 = α − 1 \beta-2=\alpha-1 β − 2 = α − 1 , the two sums are geometric (Lemma 7.1(a)) and ε m ‾ ≤ ∣ t − t ′ ∣ + ∣ x − y ∣ < ε m ‾ − 1 \varepsilon_{\overline{m}}\le|t-t'|+|x-y|<\varepsilon_{\overline{m}-1} ε m ≤ ∣ t − t ′ ∣ + ∣ x − y ∣ < ε m − 1 ; for m ‾ = 1 \overline{m}=1 m = 1 the first sum is empty. This is (17) .
Flows and renewal charts
Lemma 7.4 (Short flows). Let m ≥ 0 m\ge0 m ≥ 0 and let t , t 1 ∈ R t,t_1\in\mathbb{R} t , t 1 ∈ R satisfy ∣ t − t 1 ∣ ε m β − 2 ≤ 2 − 22 |t-t_1|\varepsilon_m^{\beta-2}\le2^{-22} ∣ t − t 1 ∣ ε m β − 2 ≤ 2 − 22 . Then y ↦ X m ( t , y , t 1 ) y\mapsto X_m(t,y,t_1) y ↦ X m ( t , y , t 1 ) is a smooth, volume-preserving diffeomorphism of R 2 \mathbb{R}^2 R 2 commuting with the translations by Z 2 \mathbb{Z}^2 Z 2 , with inverse x ↦ X m ( t 1 , x , t ) x\mapsto X_m(t_1,x,t) x ↦ X m ( t 1 , x , t ) , and, for every x , y ∈ R 2 x,y\in\mathbb{R}^2 x , y ∈ R 2 ,
∣ ∇ y X m ( t , y , t 1 ) − I ∣ ≤ 1 2 , ∣ ∇ x X m ( t 1 , x , t ) − I ∣ ≤ 1 2 ; (50) \lvert\nabla_yX_m(t,y,t_1)-I\rvert\le\frac12,\qquad\lvert\nabla_xX_m(t_1,x,t)-I\rvert\le\frac12; \tag*{(50)} ∣ ∇ y X m ( t , y , t 1 ) − I ∣ ≤ 2 1 , ∣ ∇ x X m ( t 1 , x , t ) − I ∣ ≤ 2 1 ; (50) in particular both maps are 2-Lipschitz. This applies when ∣ t − t 1 ∣ ≤ 8 h m + 1 |t-t_1|\le8h_{m+1} ∣ t − t 1 ∣ ≤ 8 h m + 1 , by (15) .
Proof. By Proposition 7.3(i), b m b_m b m is smooth, bounded, Z 2 \mathbb{Z}^2 Z 2 -periodic and divergence free, so the two maps are smooth, mutually inverse by the group law of the flow, commute with the translations by Z 2 \mathbb{Z}^2 Z 2 by uniqueness, and preserve Lebesgue measure by Liouville’s theorem. The derivative solves ∂ t ∇ y X m ( t , y , t 1 ) = ∇ b m ( t , X m ( t , y , t 1 ) ) ∇ y X m ( t , y , t 1 ) \partial_t\nabla_yX_m(t,y,t_1)=\nabla b_m(t,X_m(t,y,t_1))\nabla_yX_m(t,y,t_1) ∂ t ∇ y X m ( t , y , t 1 ) = ∇ b m ( t , X m ( t , y , t 1 )) ∇ y X m ( t , y , t 1 ) , so Grönwall’s inequality and (49) give
∣ ∇ y X m ( t , y , t 1 ) − I ∣ ≤ exp ( 78000 ε m β − 2 ∣ t − t 1 ∣ ) − 1 ≤ exp ( 1 50 ) − 1 < 1 2 , \lvert\nabla_yX_m(t,y,t_1)-I\rvert\le\exp(78000\varepsilon_m^{\beta-2}|t-t_1|)-1\le\exp\left(\frac{1}{50}\right)-1<\frac12, ∣ ∇ y X m ( t , y , t 1 ) − I ∣ ≤ exp ( 78000 ε m β − 2 ∣ t − t 1 ∣ ) − 1 ≤ exp ( 50 1 ) − 1 < 2 1 ,
and likewise for ∇ x X m ( t 1 , x , t ) \nabla_xX_m(t_1,x,t) ∇ x X m ( t 1 , x , t ) .
Complex extensions. Section 11 bounds the derivatives of its families in their parameters by Cauchy’s estimates, which needs holomorphic extensions of the flows to complex labels; they come from those of the velocity fields. For a vector of C 2 \mathbb{C}^2 C 2 , Im \operatorname{Im} Im is its imaginary part, taken coordinatewise, ∣ ⋅ ∣ 1 |\cdot|_1 ∣ ⋅ ∣ 1 is the sum of the moduli of its coordinates, and ∣ ⋅ ∣ ∞ |\cdot|_\infty ∣ ⋅ ∣ ∞ is as in Section 2, on vectors and on matrices. On the strips { ∣ Im x ∣ 1 < c ε m } \{\lvert\operatorname{Im}x\rvert_1<c\varepsilon_m\} {∣ Im x ∣ 1 < c ε m } , for a small fixed c c c , the shear streams ε m β sin ( 2 π y i / ε m ) \varepsilon_m^\beta\sin(2\pi y_i/\varepsilon_m) ε m β sin ( 2 π y i / ε m ) stay below 2 ε m β 2\varepsilon_m^\beta 2 ε m β , so Cauchy’s estimates return the real Lipschitz scale ε m β − 2 \varepsilon_m^{\beta-2} ε m β − 2 ; the imaginary part of a trajectory grows at most at this rate, so for 2 21 ε m β − 2 ∣ t − t ′ ∣ ≤ 3 40 2^{21}\varepsilon_m^{\beta-2}\lvert t-t'\rvert\le\frac{3}{40} 2 21 ε m β − 2 ∣ t − t ′ ∣ ≤ 40 3 trajectories started in a slightly narrower strip stay in the strip. Complex trajectories X m ( t , y , t ′ ) X_m(t,y,t') X m ( t , y , t ′ ) are defined by (13) .
Proposition 7.5 (Analytic velocity fields). For every m ≥ 0 m\ge0 m ≥ 0 and t ∈ R t\in\mathbb{R} t ∈ R , the field b m ( t , ⋅ ) \mathbf{b}_m(t,\cdot) b m ( t , ⋅ ) extends to a holomorphic map on the strip { x ∈ C 2 : ∣ Im x ∣ 1 < ε m / 50 } \{x\in\mathbb{C}^2:\lvert\operatorname{Im}x\rvert_1<\varepsilon_m/50\} { x ∈ C 2 : ∣ Im x ∣ 1 < ε m /50 } , which is continuous in ( t , x ) (t,x) ( t , x ) , real at real points, and satisfies, for every x , x ′ x,x' x , x ′ in the strip,
∣ b m ( t , x ) − b m ( t , x ′ ) ∣ 1 ≤ 2 21 ε m β − 2 ∣ x − x ′ ∣ 1 . (51) \lvert\mathbf{b}_m(t,x)-\mathbf{b}_m(t,x')\rvert_1\le2^{21}\varepsilon_m^{\beta-2}\lvert x-x'\rvert_1. \tag*{(51)} ∣ b m ( t , x ) − b m ( t , x ′ ) ∣ 1 ≤ 2 21 ε m β − 2 ∣ x − x ′ ∣ 1 . (51)
Lemma 7.6 (Complex flows). Let m ≥ 0 m\ge0 m ≥ 0 , t 1 < t 2 t_1<t_2 t 1 < t 2 with t 2 − t 1 ≤ 6 5 h m + 1 t_2-t_1\le\frac{6}{5}h_{m+1} t 2 − t 1 ≤ 5 6 h m + 1 , and t , t ′ , t ′ ′ ∈ [ t 1 , t 2 ] t,t',t''\in[t_1,t_2] t , t ′ , t ′′ ∈ [ t 1 , t 2 ] . For every y ∈ C 2 y\in\mathbb{C}^2 y ∈ C 2 with ∣ Im y ∣ 1 ≤ ( 1 − 2 21 ε m β − 2 ( t 2 − t 1 ) ) 19 1000 ε m \lvert\operatorname{Im}y\rvert_1\le(1-2^{21}\varepsilon_m^{\beta-2}(t_2-t_1))\frac{19}{1000}\varepsilon_m ∣ Im y ∣ 1 ≤ ( 1 − 2 21 ε m β − 2 ( t 2 − t 1 )) 1000 19 ε m , the trajectory of b m \mathbf{b}_m b m through y y y at time t ′ t' t ′ exists on [ t 1 , t 2 ] [t_1,t_2] [ t 1 , t 2 ] and satisfies the first of the bounds
∣ Im X m ( t , y , t ′ ) ∣ 1 ≤ ∣ Im y ∣ 1 1 − 2 21 ε m β − 2 ( t 2 − t 1 ) , \lvert\operatorname{Im}X_m(t,y,t')\rvert_1\le\frac{\lvert\operatorname{Im}y\rvert_1}{1-2^{21}\varepsilon_m^{\beta-2}(t_2-t_1)}, ∣ Im X m ( t , y , t ′ ) ∣ 1 ≤ 1 − 2 21 ε m β − 2 ( t 2 − t 1 ) ∣ Im y ∣ 1 , ∥ ∇ y X m ( t , y , t ′ ) − I ∥ ∞ ≤ 10 ⋅ 2 21 ε m β − 2 ∣ t − t ′ ∣ ≤ 2 25 ε m β − 2 ∣ t − t ′ ∣ . (52) \lVert\nabla_yX_m(t,y,t')-I\rVert_\infty\le10\cdot2^{21}\varepsilon_m^{\beta-2}\lvert t-t'\rvert\le2^{25}\varepsilon_m^{\beta-2}\lvert t-t'\rvert. \tag*{(52)} ∥ ∇ y X m ( t , y , t ′ ) − I ∥ ∞ ≤ 10 ⋅ 2 21 ε m β − 2 ∣ t − t ′ ∣ ≤ 2 25 ε m β − 2 ∣ t − t ′ ∣ . (52) and, if the inequality on ∣ Im y ∣ 1 \lvert\operatorname{Im}y\rvert_1 ∣ Im y ∣ 1 is strict, also the second. The map ( t , t ′ , y ) ↦ X m ( t , y , t ′ ) (t,t',y)\mapsto X_m(t,y,t') ( t , t ′ , y ) ↦ X m ( t , y , t ′ ) is continuous for t , t ′ ∈ [ t 1 , t 2 ] t,t'\in[t_1,t_2] t , t ′ ∈ [ t 1 , t 2 ] and these y y y , extends the real flow and, where the inequality on ∣ Im y ∣ 1 \lvert\operatorname{Im}y\rvert_1 ∣ Im y ∣ 1 is strict, is holomorphic in y y y . If ∣ Im y ∣ 1 ≤ ( 1 − 2 21 ε m β − 2 ( t 2 − t 1 ) ) 2 19 1000 ε m \lvert\operatorname{Im}y\rvert_1\le(1-2^{21}\varepsilon_m^{\beta-2}(t_2-t_1))^2\frac{19}{1000}\varepsilon_m ∣ Im y ∣ 1 ≤ ( 1 − 2 21 ε m β − 2 ( t 2 − t 1 ) ) 2 1000 19 ε m , then X m ( t , X m ( t ′ ′ , y , t ′ ) , t ′ ′ ) = X m ( t , y , t ′ ) X_m(t,X_m(t'',y,t'),t'')=X_m(t,y,t') X m ( t , X m ( t ′′ , y , t ′ ) , t ′′ ) = X m ( t , y , t ′ ) . Since 2 21 ε m β − 2 ( t 2 − t 1 ) ≤ 3 4 2^{21}\varepsilon_m^{\beta-2}(t_2-t_1)\le\frac{3}{4} 2 21 ε m β − 2 ( t 2 − t 1 ) ≤ 4 3 by (15) , these domains contain { ∣ Im y ∣ 1 < 17 1000 ε m } \{\lvert\operatorname{Im}y\rvert_1<\frac{17}{1000}\varepsilon_m\} {∣ Im y ∣ 1 < 1000 17 ε m } and { ∣ Im y ∣ 1 ≤ 15 1000 ε m } \{\lvert\operatorname{Im}y\rvert_1\le\frac{15}{1000}\varepsilon_m\} {∣ Im y ∣ 1 ≤ 1000 15 ε m } , respectively, and the factor in the first bound is at most 11 10 \frac{11}{10} 10 11 .
Proof of Proposition 7.5 and Lemma 7.6 . Since b 0 = 0 \mathbf{b}_0=0 b 0 = 0 , both follow from two implications: Proposition 7.5 at level m m m implies Lemma 7.6 at level m m m , and Proposition 7.5 and Lemma 7.6 at level m − 1 m-1 m − 1 imply Proposition 7.5 at level m m m .
Lemma 7.6 at level $m. Since b m \mathbf{b}_m b m is real at real points, (51) , comparing b m ( t , x ) \mathbf{b}_m(t,x) b m ( t , x ) with b m ( t , Re x ) \mathbf{b}_m(t,\operatorname{Re}x) b m ( t , Re x ) , gives ∣ Im b m ( t , x ) ∣ 1 ≤ 2 21 ε m β − 2 ∣ Im x ∣ 1 \lvert\operatorname{Im}\mathbf{b}_m(t,x)\rvert_1\le2^{21}\varepsilon_m^{\beta-2}\lvert\operatorname{Im}x\rvert_1 ∣ Im b m ( t , x ) ∣ 1 ≤ 2 21 ε m β − 2 ∣ Im x ∣ 1 . By induction, every Picard iterate of the trajectory equation, started from the constant y y y , therefore has ∣ Im ⋅ ∣ 1 ≤ ∣ Im y ∣ 1 / ( 1 − 2 21 ε m β − 2 ( t 2 − t 1 ) ) ≤ 19 1000 ε m < ε m / 50 \lvert\operatorname{Im}\cdot\rvert_1\le\lvert\operatorname{Im}y\rvert_1/(1-2^{21}\varepsilon_m^{\beta-2}(t_2-t_1))\le\frac{19}{1000}\varepsilon_m<\varepsilon_m/50 ∣ Im ⋅ ∣ 1 ≤ ∣ Im y ∣ 1 / ( 1 − 2 21 ε m β − 2 ( t 2 − t 1 )) ≤ 1000 19 ε m < ε m /50 on [ t 1 , t 2 ] [t_1,t_2] [ t 1 , t 2 ] . Since the first increment is bounded (as b m \mathbf{b}_m b m is bounded at real points, hence on the strip by (51) ), the iterates converge uniformly by (51) , geometrically with ratio $2^{21}\varepsilon_m^{\beta-2}(t_2-t_1)$. Their limit is the trajectory, unique by (51) ; it satisfies the first bound in (52) , and it is continuous in ( t , t ′ , y ) (t,t',y) ( t , t ′ , y ) and holomorphic in y y y , since the iterates are. Their derivatives in y y y solve the linearized iteration and converge; since ∥ ∇ b m ∥ ∞ ≤ 2 ⋅ 2 21 ε m β − 2 \lVert\nabla\mathbf{b}_m\rVert_\infty\le2\cdot2^{21}\varepsilon_m^{\beta-2} ∥ ∇ b m ∥ ∞ ≤ 2 ⋅ 2 21 ε m β − 2 by (51) (each entry is at most 2 21 ε m β − 2 2^{21}\varepsilon_m^{\beta-2} 2 21 ε m β − 2 ), Grönwall’s inequality bounds them by exp ( 2 ⋅ 21 40 ) < 5 \exp(2\cdot\frac{21}{40})<5 exp ( 2 ⋅ 40 21 ) < 5 , and their distance to I I I by 10 ⋅ 2 21 ε m β − 2 ∣ t − t ′ ∣ 10\cdot2^{21}\varepsilon_m^{\beta-2}\lvert t-t'\rvert 10 ⋅ 2 21 ε m β − 2 ∣ t − t ′ ∣ . At real points the trajectory is the real flow, since b m \mathbf{b}_m b m is real there. The group law holds by uniqueness, since, by the first bound, X m ( t ′ ′ , y , t ′ ) X_m(t'',y,t') X m ( t ′′ , y , t ′ ) lies in the domain of existence.
*Proposition 7.5 at level m ≥ 1. ∗ F o r m\ge1.* For m ≥ 1. ∗ F or \lvert t-l_{m,k}h_m\rvert\le\frac12h_m$, Lemma 7.6 at level m − 1 m-1 m − 1 extends x ↦ X m − 1 ( l m , k h m , x , t ) x\mapsto X_{m-1}(l_{m,k}h_m,x,t) x ↦ X m − 1 ( l m , k h m , x , t ) holomorphically, and continuously in ( t , x ) (t,x) ( t , x ) , to { ∣ Im x ∣ 1 < ε m / 25 } \{\lvert\operatorname{Im}x\rvert_1<\varepsilon_m/25\} {∣ Im x ∣ 1 < ε m /25 } , since ε m / 25 < 17 1000 ε m − 1 \varepsilon_m/25<\frac{17}{1000}\varepsilon_{m-1} ε m /25 < 1000 17 ε m − 1 , with values where ∣ Im ⋅ ∣ 1 ≤ 11 1000 ε m / 25 \lvert\operatorname{Im}\cdot\rvert_1\le\frac{11}{1000}\varepsilon_m/25 ∣ Im ⋅ ∣ 1 ≤ 1000 11 ε m /25 . The shear streams (8) are entire, and, as ∣ sin z ∣ ≤ exp ( ∣ Im z ∣ ) \lvert\sin z\rvert\le\exp(\lvert\operatorname{Im}z\rvert) ∣ sin z ∣ ≤ exp (∣ Im z ∣) , ∣ ψ m , k ∣ ≤ ε m β exp ( 2 π ⋅ 11 250 ) ≤ 2 ε m β \lvert\psi_{m,k}\rvert\le\varepsilon_m^\beta\exp(2\pi\cdot\frac{11}{250})\le2\varepsilon_m^\beta ∣ ψ m , k ∣ ≤ ε m β exp ( 2 π ⋅ 250 11 ) ≤ 2 ε m β there. By Lemma 7.1 (d), only the pulses with l m , k h m l_{m,k}h_m l m , k h m within 1 2 h m \frac12h_m 2 1 h m of t t t contribute to (14) , finitely many near each time, so ϕ m − ϕ m − 1 \phi_m-\phi_{m-1} ϕ m − ϕ m − 1 extends to a function on R × { ∣ Im x ∣ 1 < ε m / 25 } \mathbb{R}\times\{\lvert\operatorname{Im}x\rvert_1<\varepsilon_m/25\} R × {∣ Im x ∣ 1 < ε m /25 } , holomorphic in x x x , continuous in ( t , x ) (t,x) ( t , x ) (with its derivatives, by Cauchy’s integral formula) and bounded by $2\varepsilon_m^\beta$, since ∑ k ξ m , k ≤ 1 \sum_k\xi_{m,k}\le1 ∑ k ξ m , k ≤ 1 . Cauchy’s estimate on polydiscs of radius ε m / 300 \varepsilon_m/300 ε m /300 about the points of { ∣ Im x ∣ 1 < ε m / 50 } \{\lvert\operatorname{Im}x\rvert_1<\varepsilon_m/50\} {∣ Im x ∣ 1 < ε m /50 } , which lie in the strip of width ε m / 25 \varepsilon_m/25 ε m /25 , bounds the second derivatives of ϕ m − ϕ m − 1 \phi_m-\phi_{m-1} ϕ m − ϕ m − 1 at these points by 2 ⋅ 2 ε m β ( 300 / ε m ) 2 = 360000 ε m β − 2 2\cdot2\varepsilon_m^\beta(300/\varepsilon_m)^2=360000\varepsilon_m^{\beta-2} 2 ⋅ 2 ε m β ( 300/ ε m ) 2 = 360000 ε m β − 2 , which gives the Lipschitz bound 720000 ε m β − 2 720000 \varepsilon_{m}^{\beta-2} 720000 ε m β − 2 for ∇ ⊥ ( ϕ m − ϕ m − 1 ) \nabla^\perp(\phi_m-\phi_{m-1}) ∇ ⊥ ( ϕ m − ϕ m − 1 ) on { ∣ Im x ∣ 1 < ε m / 50 } \{\lvert\operatorname{Im}x\rvert_1<\varepsilon_m/50\} {∣ Im x ∣ 1 < ε m /50 } . Since ε m − 1 β − 2 ≤ ε m β − 2 / 16 \varepsilon_{m-1}^{\beta-2}\le\varepsilon_m^{\beta-2}/16 ε m − 1 β − 2 ≤ ε m β − 2 /16 , the field b m = b m − 1 + ∇ ⊥ ( ϕ m − ϕ m − 1 ) b_m=b_{m-1}+\nabla^\perp(\phi_m-\phi_{m-1}) b m = b m − 1 + ∇ ⊥ ( ϕ m − ϕ m − 1 ) satisfies (51) with 2 21 / 16 + 720000 ≤ 2 21 2^{21}/16+720000\le2^{21} 2 21 /16 + 720000 ≤ 2 21 . At real points the extension is b m b_m b m , so it is real there.
Renewal charts. Let m ≥ 1 m\ge1 m ≥ 1 and l ∈ Z l\in\mathbb{Z} l ∈ Z . In the block of l l l , the pulses of level m m m are written in the variable y = X m − 1 ( l h m , x , t ) y=X_{m-1}(lh_m,x,t) y = X m − 1 ( l h m , x , t ) of (14) . We call y y y the Lagrangian label of x x x at time t t t , the inverse map y ↦ X m − 1 ( t , y , l h m ) y\mapsto X_{m-1}(t,y,lh_m) y ↦ X m − 1 ( t , y , l h m ) the renewal chart of the block at time t t t , and the chart of the block of l m , k l_{m,k} l m , k the chart of pulse k k k . The conductivity of the chart is
K m , l ( t , y ) : = ( ∇ y X m − 1 ( t , y , l h m ) ) − 1 ( ∇ y X m − 1 ( t , y , l h m ) ) − T , (53) \mathbf{K}_{m,l}(t,y):=\left(\nabla_yX_{m-1}(t,y,lh_m)\right)^{-1}\left(\nabla_yX_{m-1}(t,y,lh_m)\right)^{-\mathsf{T}}, \tag*{(53)} K m , l ( t , y ) := ( ∇ y X m − 1 ( t , y , l h m ) ) − 1 ( ∇ y X m − 1 ( t , y , l h m ) ) − T , (53)
the coefficient of the diffusion term in these coordinates (Lemma 7.9 ). For m = 1 m=1 m = 1 the chart is the identity, since b 0 = 0 b_0=0 b 0 = 0 , and K 1 , l = I K_{1,l}=I K 1 , l = I .
Lemma 7.7 (Renewal charts). Let m ≥ 2 m\ge2 m ≥ 2 and l ∈ Z l\in\mathbb{Z} l ∈ Z .
(a) Let ∣ t − l h m ∣ ≤ h m \lvert t-lh_m\rvert\le h_m ∣ t − l h m ∣ ≤ h m . Lemma 7.4 applies to X m − 1 X_{m-1} X m − 1 with t 1 = l h m t_1=lh_m t 1 = l h m , and, for every integer j ≥ 0 j\ge0 j ≥ 0 , there exists C ( j ) C(j) C ( j ) such that, for every x , y ∈ R 2 x,y\in\mathbb{R}^2 x , y ∈ R 2 ,
ε m − 1 j ∣ ∇ y j ( ∇ y X m − 1 ( t , y , l h m ) − I ) ∣ + ε m − 1 j ∣ ∇ y j ( K m , l ( t , y ) − I ) ∣ ≤ C ( j ) ε m − 1 2 δ , (54) \varepsilon_{m-1}^{j}\left\lvert\nabla_y^j\left(\nabla_yX_{m-1}(t,y,lh_m)-I\right)\right\rvert+\varepsilon_{m-1}^{j}\left\lvert\nabla_y^j\left(\mathbf{K}_{m,l}(t,y)-I\right)\right\rvert\le C(j)\varepsilon_{m-1}^{2\delta}, \tag*{(54)} ε m − 1 j ∇ y j ( ∇ y X m − 1 ( t , y , l h m ) − I ) + ε m − 1 j ∇ y j ( K m , l ( t , y ) − I ) ≤ C ( j ) ε m − 1 2 δ , (54) ε m − 1 j + 1 ∣ ∇ x j + 1 X m − 1 ( l h m , x , t ) ∣ ≤ C ( j ) . \varepsilon_{m-1}^{j+1}\left\lvert\nabla_x^{j+1}X_{m-1}(lh_m,x,t)\right\rvert\le C(j). ε m − 1 j + 1 ∇ x j + 1 X m − 1 ( l h m , x , t ) ≤ C ( j ) . (b) For every t ∈ R t\in\mathbb{R} t ∈ R and y ∈ R 2 y\in\mathbb{R}^2 y ∈ R 2 , with y ′ : = X m − 1 ( l h m , y , ( l + 1 ) h m ) y':=X_{m-1}(lh_m,y,(l+1)h_m) y ′ := X m − 1 ( l h m , y , ( l + 1 ) h m ) ,
X m − 1 ( t , y , ( l + 1 ) h m ) = X m − 1 ( t , y ′ , l h m ) , X_{m-1}(t,y,(l+1)h_m)=X_{m-1}(t,y',lh_m), X m − 1 ( t , y , ( l + 1 ) h m ) = X m − 1 ( t , y ′ , l h m ) , K m , l + 1 ( t , y ) = ( ∇ y y ′ ) − 1 K m , l ( t , y ′ ) ( ∇ y y ′ ) − T , (55) \mathbf{K}_{m,l+1}(t,y)=(\nabla_y y')^{-1}\mathbf{K}_{m,l}(t,y')(\nabla_y y')^{-\mathsf{T}}, \tag*{(55)} K m , l + 1 ( t , y ) = ( ∇ y y ′ ) − 1 K m , l ( t , y ′ ) ( ∇ y y ′ ) − T , (55) and the map y ↦ y ′ y\mapsto y' y ↦ y ′ preserves area.
(c) Lemma 7.6 at level m − 1 m-1 m − 1 applies on the interval [ l h m − 3 5 h m , l h m + 3 5 h m ] [lh_m-\frac{3}{5}h_m,lh_m+\frac{3}{5}h_m] [ l h m − 5 3 h m , l h m + 5 3 h m ] (of length 6 5 h m \frac{6}{5}h_m 5 6 h m ) about the centre of the block, to y ↦ X m − 1 ( t , y , t ′ ) y\mapsto X_{m-1}(t,y,t') y ↦ X m − 1 ( t , y , t ′ ) , in particular to the chart (t ′ = l h m t'=lh_m t ′ = l h m ) and to its inverse (t = l h m t=lh_m t = l h m ).
Proof. Lemmas 7.4 and 7.6 apply in (a) and (c), since h m ε m − 1 β − 2 ≤ 2 − 25 h_m\varepsilon_{m-1}^{\beta-2}\le2^{-25} h m ε m − 1 β − 2 ≤ 2 − 25 by (15) . Next, [5 Proposition 2.2, (2.40)] bounds ∇ x X m − 1 ( t ′ , x , t ′ ′ ) − I \nabla_xX_{m-1}(t',x,t'')-I ∇ x X m − 1 ( t ′ , x , t ′′ ) − I , for every pair of times t ′ , t ′ ′ t',t'' t ′ , t ′′ with ∣ t ′ − t ′ ′ ∣ ≤ 2 − 25 − 2 δ ε m − 1 2 δ \lvert t'-t''\rvert\le2^{-25-2\delta}\varepsilon_{m-1}^{2\delta} ∣ t ′ − t ′′ ∣ ≤ 2 − 25 − 2 δ ε m − 1 2 δ , in the norm [5 (1.21)] at the radius 2 11 / ε m − 1 2^{11}/\varepsilon_{m-1} 2 11 / ε m − 1 by 2 23 ∣ t ′ − t ′ ′ ∣ ε m − 1 β − 2 2^{23}\lvert t'-t''\rvert\varepsilon_{m-1}^{\beta-2} 2 23 ∣ t ′ − t ′′ ∣ ε m − 1 β − 2 ; read as in the proof of Lemma 7.2 , this is ε m − 1 j ∣ ∇ x j ( ∇ x X m − 1 ( t ′ , x , t ′ ′ ) − I ) ∣ ≤ C ( j ) ∣ t ′ − t ′ ′ ∣ ε m − 1 β − 2 \varepsilon_{m-1}^{j}\lvert\nabla_x^j(\nabla_xX_{m-1}(t',x,t'')-I)\rvert\le C(j)\lvert t'-t''\rvert\varepsilon_{m-1}^{\beta-2} ε m − 1 j ∣ ∇ x j ( ∇ x X m − 1 ( t ′ , x , t ′′ ) − I )∣ ≤ C ( j ) ∣ t ′ − t ′′ ∣ ε m − 1 β − 2 . With ( t ′ , t ′ ′ ) = ( t , l h m ) (t',t'')=(t,lh_m) ( t ′ , t ′′ ) = ( t , l h m ) and ( l h m , t ) (lh_m,t) ( l h m , t ) , and h m ε m − 1 β − 2 ≤ 2 − 25 ε m − 1 2 δ h_m\varepsilon_{m-1}^{\beta-2}\le2^{-25}\varepsilon_{m-1}^{2\delta} h m ε m − 1 β − 2 ≤ 2 − 25 ε m − 1 2 δ , this gives the bounds on ∇ y X m − 1 ( t , y , l h m ) − I \nabla_yX_{m-1}(t,y,lh_m)-I ∇ y X m − 1 ( t , y , l h m ) − I and on X m − 1 ( l h m , x , t ) X_{m-1}(lh_m,x,t) X m − 1 ( l h m , x , t ) in (54) . Since det ∇ y X m − 1 = 1 \det\nabla_yX_{m-1}=1 det ∇ y X m − 1 = 1 , the inverse of ∇ y X m − 1 \nabla_yX_{m-1} ∇ y X m − 1 is its adjugate, so K m , l − I \mathbf{K}_{m,l}-I K m , l − I is a quadratic polynomial without constant term in the entries of ∇ y X m − 1 − I \nabla_yX_{m-1}-I ∇ y X m − 1 − I , and the Leibniz rule bounds its jets. Finally, in (b), the first identity of (55) is the group law, the second follows from it by the chain rule, and y ↦ y ′ y\mapsto y' y ↦ y ′ preserves area by Liouville’s theorem.
For m ≥ 1 m\ge1 m ≥ 1 , the stream of the level-m m m pulses [5 (3.3)] is
ψ m ( t , y ) : = ∑ k ∈ 2 Z + 1 ζ ~ m , k ( t ) ψ m , k ( y ) . (56) \psi_m(t,y):=\sum_{k\in2\mathbb{Z}+1}\widetilde{\zeta}_{m,k}(t)\psi_{m,k}(y). \tag*{(56)} ψ m ( t , y ) := k ∈ 2 Z + 1 ∑ ζ m , k ( t ) ψ m , k ( y ) . (56)
Lemma 7.8 (The level-m m m stream in a chart). Let m ≥ 1 m\ge1 m ≥ 1 and l ∈ Z l\in\mathbb{Z} l ∈ Z .
(i) For every t ∈ [ ( l − 1 2 ) h m , ( l + 1 2 ) h m ] t \in[(l-\frac{1}{2})h_m,(l+\frac{1}{2})h_m] t ∈ [( l − 2 1 ) h m , ( l + 2 1 ) h m ] and y ∈ R 2 y \in\mathbb{R}^2 y ∈ R 2 ,
( ϕ m − ϕ m − 1 ) ( t , X m − 1 ( t , y , l h m ) ) = ψ m ( t , y ) , (\phi_m-\phi_{m-1})(t,X_{m-1}(t,y,lh_m))=\psi_m(t,y), ( ϕ m − ϕ m − 1 ) ( t , X m − 1 ( t , y , l h m )) = ψ m ( t , y ) ,
( ∇ y X m − 1 ( t , y , l h m ) ) − 1 ( b m − b m − 1 ) ( t , X m − 1 ( t , y , l h m ) ) = ∇ ⊥ ψ m ( t , y ) . (57) (\nabla_yX_{m-1}(t,y,lh_m))^{-1}(b_m-b_{m-1})(t,X_{m-1}(t,y,lh_m))=\nabla^\perp\psi_m(t,y). \tag*{(57)} ( ∇ y X m − 1 ( t , y , l h m ) ) − 1 ( b m − b m − 1 ) ( t , X m − 1 ( t , y , l h m )) = ∇ ⊥ ψ m ( t , y ) . (57)
(ii) Let k k k be odd with l m , k = l l_{m,k}=l l m , k = l , and let t , t ′ ∈ R t,t'\in\mathbb{R} t , t ′ ∈ R with ∣ t − k τ m ∣ < 4 3 τ m |t-k\tau_m|<\frac{4}{3}\tau_m ∣ t − k τ m ∣ < 3 4 τ m and ∣ t ′ − k τ m ∣ < 4 3 τ m |t'-k\tau_m|<\frac{4}{3}\tau_m ∣ t ′ − k τ m ∣ < 3 4 τ m . Then, for every x ∈ R 2 x\in\mathbb{R}^2 x ∈ R 2 , with y : = X m − 1 ( l h m , x , t ) y:=X_{m-1}(lh_m,x,t) y := X m − 1 ( l h m , x , t ) ,
X m ( t ′ , x , t ) = X m − 1 ( t ′ , y + ( ∫ t t ′ ζ ~ m , k ( t ′ ′ ) d t ′ ′ ) ∇ ⊥ ψ m , k ( y ) , l h m ) . (58) X_m(t',x,t)=X_{m-1}\left(t',y+\left(\int_t^{t'}\widetilde{\zeta}_{m,k}(t'')\,\mathrm{d}t''\right)\nabla^\perp\psi_{m,k}(y),lh_m\right). \tag*{(58)} X m ( t ′ , x , t ) = X m − 1 ( t ′ , y + ( ∫ t t ′ ζ m , k ( t ′′ ) d t ′′ ) ∇ ⊥ ψ m , k ( y ) , l h m ) . (58)
Proof. (i) By Lemma 7.1 (d), at a time t t t of the block only the pulses with l m , k = l l_{m,k}=l l m , k = l contribute to (14) , and for them X m − 1 ( l h m , X m − 1 ( t , y , l h m ) , t ) = y X_{m-1}(lh_m,X_{m-1}(t,y,lh_m),t)=y X m − 1 ( l h m , X m − 1 ( t , y , l h m ) , t ) = y by the group law. This is the first identity. The second follows from it, since det ∇ y X m − 1 = 1 \det\nabla_yX_{m-1}=1 det ∇ y X m − 1 = 1 gives
( 0 − 1 1 0 ) ( ∇ y X m − 1 ) T = ( ∇ y X m − 1 ) − 1 ( 0 − 1 1 0 ) , \begin{pmatrix}0&-1\\1&0\end{pmatrix}(\nabla_yX_{m-1})^{\mathsf{T}}=(\nabla_yX_{m-1})^{-1}\begin{pmatrix}0&-1\\1&0\end{pmatrix}, ( 0 1 − 1 0 ) ( ∇ y X m − 1 ) T = ( ∇ y X m − 1 ) − 1 ( 0 1 − 1 0 ) ,
so that ∇ y ⊥ \nabla^\perp_y ∇ y ⊥ of a function composed with X m − 1 ( t , ⋅ , l h m ) X_{m-1}(t,\cdot,lh_m) X m − 1 ( t , ⋅ , l h m ) is ( ∇ y X m − 1 ) − 1 (\nabla_yX_{m-1})^{-1} ( ∇ y X m − 1 ) − 1 times its ∇ ⊥ \nabla^\perp ∇ ⊥ at X m − 1 ( t , y , l h m ) X_{m-1}(t,y,lh_m) X m − 1 ( t , y , l h m ) .
(ii) Let t ′ ′ t'' t ′′ lie between t t t and t ′ t' t ′ , and x ′ ′ ∈ R 2 x''\in\mathbb{R}^2 x ′′ ∈ R 2 . No pulse other than k k k is active at t ′ ′ t'' t ′′ : for odd k ′ ≠ k k'\ne k k ′ = k , the cutoff ζ m , k ′ \zeta_{m,k'} ζ m , k ′ vanishes outside [ ( k ′ − 2 3 ) τ m , ( k ′ + 2 3 ) τ m ] [(k'-\frac{2}{3})\tau_m,(k'+\frac{2}{3})\tau_m] [( k ′ − 3 2 ) τ m , ( k ′ + 3 2 ) τ m ] , which is at distance at least 4 3 τ m \frac{4}{3}\tau_m 3 4 τ m from k τ m k\tau_m k τ m , while ∣ t ′ ′ − k τ m ∣ < 4 3 τ m |t''-k\tau_m|<\frac{4}{3}\tau_m ∣ t ′′ − k τ m ∣ < 3 4 τ m . If t ′ ′ t'' t ′′ lies in the block of l l l , (i) gives
( b m − b m − 1 ) ( t ′ ′ , x ′ ′ ) = ζ ~ m , k ( t ′ ′ ) ∇ y X m − 1 ( t ′ ′ , y ′ ′ , l h m ) ∇ ⊥ ψ m , k ( y ′ ′ ) , y ′ ′ : = X m − 1 ( l h m , x ′ ′ , t ′ ′ ) . (b_m-b_{m-1})(t'',x'')=\widetilde{\zeta}_{m,k}(t'')\nabla_yX_{m-1}(t'',y'',lh_m)\nabla^\perp\psi_{m,k}(y''),\qquad y'':=X_{m-1}(lh_m,x'',t''). ( b m − b m − 1 ) ( t ′′ , x ′′ ) = ζ m , k ( t ′′ ) ∇ y X m − 1 ( t ′′ , y ′′ , l h m ) ∇ ⊥ ψ m , k ( y ′′ ) , y ′′ := X m − 1 ( l h m , x ′′ , t ′′ ) .
Otherwise, by Lemma 7.1 (d), t ′ ′ t'' t ′′ lies within 5 6 τ m < 5 τ m ≤ τ m ′ \frac{5}{6}\tau_m<5\tau_m\leq\tau'_m 6 5 τ m < 5 τ m ≤ τ m ′ of an end of the block, where all slow cutoffs vanish, and both sides vanish. Now consider the curve t ′ ′ ↦ X m − 1 ( t ′ ′ , y + ( ∫ t t ′ ′ ζ ~ m , k ) ∇ ⊥ ψ m , k ( y ) , l h m ) t''\mapsto X_{m-1}(t'',y+(\int_t^{t''}\widetilde{\zeta}_{m,k})\nabla^\perp\psi_{m,k}(y),lh_m) t ′′ ↦ X m − 1 ( t ′′ , y + ( ∫ t t ′′ ζ m , k ) ∇ ⊥ ψ m , k ( y ) , l h m ) . By (2.6), ∇ ⊥ ψ m , k \nabla^\perp\psi_{m,k} ∇ ⊥ ψ m , k is a multiple of e ( k ) e_{(k)} e ( k ) that depends only on the coordinate of its argument orthogonal to e ( k ) e_{(k)} e ( k ) , so it takes the value ∇ ⊥ ψ m , k ( y ) \nabla^\perp\psi_{m,k}(y) ∇ ⊥ ψ m , k ( y ) at the displaced label. By the chain rule, the derivative of the curve is b m − 1 b_{m-1} b m − 1 at the curve plus ζ ~ m , k ( t ′ ′ ) ∇ y X m − 1 ( t ′ ′ , ⋅ , l h m ) ∇ ⊥ ψ m , k \widetilde{\zeta}_{m,k}(t'')\nabla_yX_{m-1}(t'',\cdot,lh_m)\nabla^\perp\psi_{m,k} ζ m , k ( t ′′ ) ∇ y X m − 1 ( t ′′ , ⋅ , l h m ) ∇ ⊥ ψ m , k at the displaced label, which is the label y ′ ′ y'' y ′′ of the point of the curve by the group law; by the previous display, the second term is b m − b m − 1 b_m-b_{m-1} b m − b m − 1 at the curve. The curve passes through x x x at time t t t , so it is t ′ ′ ↦ X m ( t ′ ′ , x , t ) t''\mapsto X_m(t'',x,t) t ′′ ↦ X m ( t ′′ , x , t ) . □
Shear displacements and phases. We call y ↦ y + ( ∫ t t ′ ζ ~ m , k ( t ′ ′ ) d t ′ ′ ) ∇ ⊥ ψ m , k ( y ) y\mapsto y+(\int_t^{t'}\widetilde{\zeta}_{m,k}(t'')\,\mathrm{d}t'')\nabla^\perp\psi_{m,k}(y) y ↦ y + ( ∫ t t ′ ζ m , k ( t ′′ ) d t ′′ ) ∇ ⊥ ψ m , k ( y ) in (58) the shear displacement of pulse k k k ; it acts at the Lagrangian label y y y and depends on y y y only through y / ε m y/\varepsilon_m y / ε m modulo Z 2 \mathbb{Z}^2 Z 2 , its phase . § 5.4 composes the shear displacements across levels, with labels and phases as parameters.
Lemma 7.9 (The equation in a chart). Let m ≥ 1 m\geq1 m ≥ 1 , l ∈ Z l\in\mathbb{Z} l ∈ Z , ( l − 1 2 ) h m ≤ t 1 ≤ t 2 ≤ ( l + 1 2 ) h m (l-\frac{1}{2})h_m\leq t_1\leq t_2\leq(l+\frac{1}{2})h_m ( l − 2 1 ) h m ≤ t 1 ≤ t 2 ≤ ( l + 2 1 ) h m and ν > 0 \nu>0 ν > 0 . Let F ∈ L 2 ( ( t 1 , t 2 ) × T 2 ; R 2 ) F\in L^2((t_1,t_2)\times\mathbb{T}^2;\mathbb{R}^2) F ∈ L 2 (( t 1 , t 2 ) × T 2 ; R 2 ) , F 0 ∈ L 2 ( ( t 1 , t 2 ) × T 2 ) F^0\in L^2((t_1,t_2)\times\mathbb{T}^2) F 0 ∈ L 2 (( t 1 , t 2 ) × T 2 ) and g ∈ L 2 ( T 2 ) g\in L^2(\mathbb{T}^2) g ∈ L 2 ( T 2 ) . A function θ \theta θ solves
{ ∂ t θ + b m ⋅ ∇ θ − ν Δ θ = ∇ ⋅ F + F 0 in ( t 1 , t 2 ) × T 2 , θ ( t 1 , ⋅ ) = g on T 2 \begin{cases}
\partial_t\theta+b_m\cdot\nabla\theta-\nu\Delta\theta=\nabla\cdot F+F^0 & \text{in }(t_1,t_2)\times\mathbb{T}^2,\\
\theta(t_1,\cdot)=g & \text{on }\mathbb{T}^2
\end{cases} { ∂ t θ + b m ⋅ ∇ θ − ν Δ θ = ∇ ⋅ F + F 0 θ ( t 1 , ⋅ ) = g in ( t 1 , t 2 ) × T 2 , on T 2 if and only if U ( t , y ) : = θ ( t , X m − 1 ( t , y , l h m ) ) U(t,y):=\theta(t,X_{m-1}(t,y,lh_m)) U ( t , y ) := θ ( t , X m − 1 ( t , y , l h m )) solves
{ ∂ t U + ∇ ⊥ ψ m ⋅ ∇ U − ν ∇ ⋅ ( K m , l ∇ U ) = ∇ ⋅ ( ( ∇ y X m − 1 ) − 1 F ( t , X m − 1 ) ) + F 0 ( t , X m − 1 ) in ( t 1 , t 2 ) × T 2 , U ( t 1 , ⋅ ) = g ( X m − 1 ( t 1 , ⋅ , l h m ) ) on T 2 (59) \begin{cases}
\partial_tU+\nabla^\perp\psi_m\cdot\nabla U-\nu\nabla\cdot(K_{m,l}\nabla U)\\
\qquad=\nabla\cdot\left((\nabla_yX_{m-1})^{-1}F(t,X_{m-1})\right)+F^0(t,X_{m-1}) & \text{in }(t_1,t_2)\times\mathbb{T}^2,\\
U(t_1,\cdot)=g(X_{m-1}(t_1,\cdot,lh_m)) & \text{on }\mathbb{T}^2 \tag*{(59)}
\end{cases} ⎩ ⎨ ⎧ ∂ t U + ∇ ⊥ ψ m ⋅ ∇ U − ν ∇ ⋅ ( K m , l ∇ U ) = ∇ ⋅ ( ( ∇ y X m − 1 ) − 1 F ( t , X m − 1 ) ) + F 0 ( t , X m − 1 ) U ( t 1 , ⋅ ) = g ( X m − 1 ( t 1 , ⋅ , l h m )) in ( t 1 , t 2 ) × T 2 , on T 2 (59) where X m − 1 X_{m-1} X m − 1 and ∇ y X m − 1 \nabla_yX_{m-1} ∇ y X m − 1 are evaluated at ( t , y , l h m ) (t,y,lh_m) ( t , y , l h m ) , and ∇ \nabla ∇ acts in y y y . Solutions are energy solutions, as in (67) .
In the chart the drift b m − 1 \mathbf{b}_{m-1} b m − 1 disappears, the identity in the diffusion term becomes the conductivity K m , l \mathbf{K}_{m,l} K m , l , which is close to I I I by Lemma 7.7 , and the pulses become the explicit shears (56) . Since, for U ∈ H 1 ( T 2 ) U \in H^1(\mathbb{T}^2) U ∈ H 1 ( T 2 ) , − ∇ ⋅ ( ψ m ∇ ⊥ U ) = ∇ ⊥ ψ m ⋅ ∇ U -\nabla\cdot(\psi_m\nabla^\perp U)=\nabla^\perp\psi_m\cdot\nabla U − ∇ ⋅ ( ψ m ∇ ⊥ U ) = ∇ ⊥ ψ m ⋅ ∇ U , the equation in (59) can equally be written in divergence form, with the matrix ν K m , l + ψ m ( 0 − 1 1 0 ) \nu\mathbf{K}_{m,l}+\psi_m\begin{pmatrix}0 & -1 \\ 1 & 0\end{pmatrix} ν K m , l + ψ m ( 0 1 − 1 0 ) .
Proof. By Proposition 7.3(i) and as in the proof of Lemma 7.4 , at each time the chart is a smooth, volume-preserving diffeomorphism commuting with the translations by Z 2 \mathbb{Z}^2 Z 2 , so the change of variables preserves the class of solutions and the weak formulation: for φ ∈ C ∞ ( T 2 ) \varphi\in C^\infty(\mathbb{T}^2) φ ∈ C ∞ ( T 2 ) and φ t : = φ ( X m − 1 ( l h m , ⋅ , t ) ) \varphi_t:=\varphi(X_{m-1}(lh_m,\cdot,t)) φ t := φ ( X m − 1 ( l h m , ⋅ , t )) , ⟨ U ( t ) , φ ⟩ L 2 ( T 2 ) = ⟨ θ ( t ) , φ t ⟩ L 2 ( T 2 ) \langle U(t),\varphi\rangle_{L^2(\mathbb{T}^2)}=\langle\theta(t),\varphi_t\rangle_{L^2(\mathbb{T}^2)} ⟨ U ( t ) , φ ⟩ L 2 ( T 2 ) = ⟨ θ ( t ) , φ t ⟩ L 2 ( T 2 ) , so the equation for U U U tested against φ \varphi φ is the equation for θ \theta θ tested against φ t \varphi_t φ t . As φ t \varphi_t φ t is transported by b m − 1 \mathbf{b}_{m-1} b m − 1 , ∂ t φ t + b m − 1 ⋅ ∇ φ t = 0 \partial_t\varphi_t+\mathbf{b}_{m-1}\cdot\nabla\varphi_t=0 ∂ t φ t + b m − 1 ⋅ ∇ φ t = 0 , so the drift b m − 1 \mathbf{b}_{m-1} b m − 1 is absorbed into the time derivative of U U U . Since ∇ y U = ( ∇ y X m − 1 ) T ∇ θ ( t , X m − 1 ) \nabla_yU=(\nabla_yX_{m-1})^{\mathsf{T}}\nabla\theta(t,X_{m-1}) ∇ y U = ( ∇ y X m − 1 ) T ∇ θ ( t , X m − 1 ) , the second identity of (57) turns ( b m − b m − 1 ) ⋅ ∇ θ (\mathbf{b}_m-\mathbf{b}_{m-1})\cdot\nabla\theta ( b m − b m − 1 ) ⋅ ∇ θ into ∇ ⊥ ψ m ⋅ ∇ y U \nabla^\perp\psi_m\cdot\nabla_yU ∇ ⊥ ψ m ⋅ ∇ y U . Finally, since det ∇ y X m − 1 = 1 \det\nabla_yX_{m-1}=1 det ∇ y X m − 1 = 1 , the Piola identity (the rows of the cofactor matrix of a Jacobian matrix are divergence free) gives, for C 1 C^1 C 1 fields and, in the weak formulation, for fields in L 2 L^2 L 2 , ( ∇ ⋅ F ) ( t , X m − 1 ) = ∇ y ⋅ ( ( ∇ y X m − 1 ) − 1 F ( t , X m − 1 ) ) (\nabla\cdot F)(t,X_{m-1})=\nabla_y\cdot((\nabla_yX_{m-1})^{-1}F(t,X_{m-1})) ( ∇ ⋅ F ) ( t , X m − 1 ) = ∇ y ⋅ (( ∇ y X m − 1 ) − 1 F ( t , X m − 1 )) . Applied to F F F and to ν ∇ θ = ν ( ∇ y X m − 1 ) − T ∇ y U \nu\nabla\theta=\nu(\nabla_yX_{m-1})^{-\mathsf{T}}\nabla_yU ν ∇ θ = ν ( ∇ y X m − 1 ) − T ∇ y U , it gives the forcing and the term ν ∇ ⋅ ( K m , l ∇ U ) \nu\nabla\cdot(\mathbf{K}_{m,l}\nabla U) ν ∇ ⋅ ( K m , l ∇ U ) .
The averaged flux
Lemma 7.10 (The averaged flux). Let m ≥ 1 m\ge1 m ≥ 1 and ν > 0 \nu>0 ν > 0 .
(i) For every t ∈ R t\in\mathbb{R} t ∈ R , J m ν ( t ) \mathbf{J}_m^\nu(t) J m ν ( t ) is diagonal, ν I ≤ J m ν ( t ) ≤ ( ν + ε m 2 β / ( 2 ν ) ) I \nu I\le\mathbf{J}_m^\nu(t)\le(\nu+\varepsilon_m^{2\beta}/(2\nu))I ν I ≤ J m ν ( t ) ≤ ( ν + ε m 2 β / ( 2 ν )) I , and, if ζ ~ m , k ( t ) = 0 \widetilde{\zeta}_{m,k}(t)=0 ζ m , k ( t ) = 0 for every odd k k k , then J m ν ( t ) = ν I \mathbf{J}_m^\nu(t)=\nu I J m ν ( t ) = ν I .
(ii) For every t ∈ R t\in\mathbb{R} t ∈ R , J m ν ( t + 2 h m ) \mathbf{J}_m^\nu(t+2h_m) J m ν ( t + 2 h m ) is J m ν ( t ) \mathbf{J}_m^\nu(t) J m ν ( t ) with its two diagonal entries exchanged, and J m ν ( t + 4 h m ) = J m ν ( t ) \mathbf{J}_m^\nu(t+4h_m)=\mathbf{J}_m^\nu(t) J m ν ( t + 4 h m ) = J m ν ( t ) .
(iii) (19) holds, and K ‾ m ( ν ) ≥ ν \overline{K}_m(\nu)\ge\nu K m ( ν ) ≥ ν .
Proof. (i) The matrices e ( k ) ⊗ e ( k ) e_{(k)}\otimes e_{(k)} e ( k ) ⊗ e ( k ) are diagonal, and 0 ≤ ζ ~ m , k ≤ 1 0\le\widetilde{\zeta}_{m,k}\le1 0 ≤ ζ m , k ≤ 1 , so the time integral in (18) lies between 0 0 0 and ∫ − ∞ t exp ( − 4 π 2 ν ( t − t ′ ) / ε m 2 ) d t ′ = ε m 2 / ( 4 π 2 ν ) \int_{-\infty}^t\exp(-4\pi^2\nu(t-t')/\varepsilon_m^2)\,\mathrm{d}t'=\varepsilon_m^2/(4\pi^2\nu) ∫ − ∞ t exp ( − 4 π 2 ν ( t − t ′ ) / ε m 2 ) d t ′ = ε m 2 / ( 4 π 2 ν ) . At most one pulse is active at each time (Lemma 7.1 (d)), so 0 ≤ J m ν ( t ) − ν I ≤ 2 π 2 β m − 2 ⋅ ε m 2 β / ( 4 π 2 ν ) I = ε m 2 β / ( 2 ν ) I 0\le\mathbf{J}_m^\nu(t)-\nu I\le2\pi^2\beta_m^{-2}\cdot\varepsilon_m^{2\beta}/(4\pi^2\nu)I=\varepsilon_m^{2\beta}/(2\nu)I 0 ≤ J m ν ( t ) − ν I ≤ 2 π 2 β m − 2 ⋅ ε m 2 β / ( 4 π 2 ν ) I = ε m 2 β / ( 2 ν ) I .
(ii) Let j ∈ 2 Z j\in2\mathbb{Z} j ∈ 2 Z . Since h m / τ m h_m/\tau_m h m / τ m is an integer, (10) , (9) and (11) give l m , k + j h m / τ m = l m , k + j l_{m,k+jh_m/\tau_m}=l_{m,k}+j l m , k + j h m / τ m = l m , k + j and ζ ~ m , k + j h m / τ m ( t + j h m ) = ζ ~ m , k ( t ) \widetilde{\zeta}_{m,k+jh_m/\tau_m}(t+jh_m)=\widetilde{\zeta}_{m,k}(t) ζ m , k + j h m / τ m ( t + j h m ) = ζ m , k ( t ) , so the terms of J m ν ( t + j h m ) \mathbf{J}_m^\nu(t+jh_m) J m ν ( t + j h m ) are those of J m ν ( t ) \mathbf{J}_m^\nu(t) J m ν ( t ) with k k k replaced by k + j h m / τ m k+jh_m/\tau_m k + j h m / τ m . The integer h m / τ m h_m/\tau_m h m / τ m is odd, so for j = 2 j=2 j = 2 the shift keeps k k k odd and exchanges $4\mathbb{Z}+1a n d and an d 4\mathbb{Z}+3, hence e 1 ⊗ e 1 e_1\otimes e_1 e 1 ⊗ e 1 and e 2 ⊗ e 2 e_2\otimes e_2 e 2 ⊗ e 2 , and for j = 4 j=4 j = 4 it fixes the classes.
(iii) By (ii), J m ν \mathbf{J}_m^\nu J m ν is 4 h m 4h_m 4 h m -periodic and a shift by 2 h m 2h_m 2 h m exchanges its diagonal entries. So its averages over [ t , t + 4 h m ] [t,t+4h_m] [ t , t + 4 h m ] and [ t + 2 h m , t + 6 h m ] [t+2h_m,t+6h_m] [ t + 2 h m , t + 6 h m ] are equal and have exchanged entries; by (i) they are diagonal, hence half their trace times I I I , which is (19) ; and K ‾ m ( ν ) ≥ ν \overline{K}_m(\nu)\ge\nu K m ( ν ) ≥ ν .
Section 10 identifies J m ν \mathbf{J}_m^\nu J m ν with the spatial average of the flux of the correctors of the level-m m m pulses. The bands (20) are disjoint and accumulate at 0 0 0 : by Lemma 7.1 (a), 2 ε M + 1 β − γ ≤ 2 Λ − ( β − γ ) ε M β − γ < 1 2 ε M β − γ 2\varepsilon_{M+1}^{\beta-\gamma}\le2\Lambda^{-(\beta-\gamma)}\varepsilon_M^{\beta-\gamma}<\frac{1}{2}\varepsilon_M^{\beta-\gamma} 2 ε M + 1 β − γ ≤ 2 Λ − ( β − γ ) ε M β − γ < 2 1 ε M β − γ , since β − γ = 2 β / ( q + 1 ) > 32 33 \beta-\gamma=2\beta/(q+1)>\frac{32}{33} β − γ = 2 β / ( q + 1 ) > 33 32 and Λ ≥ 128 \Lambda\ge128 Λ ≥ 128 .
Effective diffusivities
This subsection proves Proposition 2.1; its part (a) holds for every Λ ≥ 128 \Lambda\ge128 Λ ≥ 128 , with C C C depending only on ζ \zeta ζ .
Proof of Proposition 2.1(a). Fix m ≥ 1 m\ge1 m ≥ 1 and ν > 0 \nu>0 ν > 0 . By the recursion (21) , we must show that K ‾ m ( ν ) \overline{K}_m(\nu) K m ( ν ) is within C ( ε m 2 β / ν ) ( ε m 2 / ( ν τ m ) + ε m − 1 δ ) C(\varepsilon_m^{2\beta}/\nu)(\varepsilon_m^2/(\nu\tau_m)+\varepsilon_{m-1}^\delta) C ( ε m 2 β / ν ) ( ε m 2 / ( ν τ m ) + ε m − 1 δ ) of ν + 9 80 ε m 2 β / ν \nu+\frac{9}{80}\varepsilon_m^{2\beta}/\nu ν + 80 9 ε m 2 β / ν .
Step 1: The time integral. Each pulse acts on the averaged flux as if it were held at its current value. Let k k k be odd. By Lemma 7.1 (d) and τ m ≤ τ m ′ \tau_m \le\tau'_m τ m ≤ τ m ′ , ∣ ∂ t ζ ~ m , k ∣ ≤ C / τ m |\partial_t\widetilde{\zeta}_{m,k}| \le C/\tau_m ∣ ∂ t ζ m , k ∣ ≤ C / τ m , so
∣ ∫ − ∞ t exp ( − 4 π 2 ν ( t − t ′ ) ε m 2 ) ζ ~ m , k ( t ′ ) d t ′ − ε m 2 4 π 2 ν ζ ~ m , k ( t ) ∣ ≤ C τ m ∫ 0 ∞ t ′ exp ( − 4 π 2 ν t ′ ε m 2 ) d t ′ ≤ C ε m 4 ν 2 τ m . \left|\int_{-\infty}^{t}\exp\left(-\frac{4\pi^2\nu(t-t')}{\varepsilon_m^2}\right)\widetilde{\zeta}_{m,k}(t')\,\mathrm{d}t'-\frac{\varepsilon_m^2}{4\pi^2\nu}\widetilde{\zeta}_{m,k}(t)\right|\le\frac{C}{\tau_m}\int_0^\infty t'\exp\left(-\frac{4\pi^2\nu t'}{\varepsilon_m^2}\right)\,\mathrm{d}t'\le\frac{C\varepsilon_m^4}{\nu^2\tau_m}. ∫ − ∞ t exp ( − ε m 2 4 π 2 ν ( t − t ′ ) ) ζ m , k ( t ′ ) d t ′ − 4 π 2 ν ε m 2 ζ m , k ( t ) ≤ τ m C ∫ 0 ∞ t ′ exp ( − ε m 2 4 π 2 ν t ′ ) d t ′ ≤ ν 2 τ m C ε m 4 .
As tr ( e ( k ) ⊗ e ( k ) ) = 1 \operatorname{tr}(e_{(k)}\otimes e_{(k)})=1 tr ( e ( k ) ⊗ e ( k ) ) = 1 , 0 ≤ ζ ~ m , k ≤ 1 0\le\widetilde{\zeta}_{m,k}\le1 0 ≤ ζ m , k ≤ 1 and at most one pulse is active at a time, (18) gives, for every t ∈ R t\in\mathbb{R} t ∈ R ,
∣ tr J m ν ( t ) − 2 ν − ε m 2 β 2 ν ∑ k ∈ 2 Z + 1 ζ ~ m , k ( t ) 2 ∣ ≤ C ε m 2 β ν ε m 2 ν τ m . (60) \left|\operatorname{tr}J_m^\nu(t)-2\nu-\frac{\varepsilon_m^{2\beta}}{2\nu}\sum_{k\in2\mathbb{Z}+1}\widetilde{\zeta}_{m,k}(t)^2\right|\le C\frac{\varepsilon_m^{2\beta}}{\nu}\frac{\varepsilon_m^2}{\nu\tau_m}. \tag*{(60)} tr J m ν ( t ) − 2 ν − 2 ν ε m 2 β k ∈ 2 Z + 1 ∑ ζ m , k ( t ) 2 ≤ C ν ε m 2 β ν τ m ε m 2 . (60)
Step 2: The average of the squared envelopes. The function ∑ k ∈ 2 Z + 1 ζ ~ m , k 2 \sum_{k\in2\mathbb{Z}+1}\widetilde{\zeta}_{m,k}^2 ∑ k ∈ 2 Z + 1 ζ m , k 2 is 2 τ m ′ 2\tau'_m 2 τ m ′ -periodic, and its integral over a period is τ m ∫ R ζ m 2 = 9 10 τ m \tau_m\int_{\mathbb{R}}\zeta_m^2=\frac{9}{10}\tau_m τ m ∫ R ζ m 2 = 10 9 τ m by (4) . Since h m / τ m h_m/\tau_m h m / τ m is an integer, its average over [ 0 , 4 h m ] [0,4h_m] [ 0 , 4 h m ] is exactly 9 20 \frac{9}{20} 20 9 : half of the slots are odd, and each carries 9 10 \frac{9}{10} 10 9 . The slow cutoffs change this sum only where ζ ~ m , l m , k ( t ) < 1 \widetilde{\zeta}_{m,l_m,k}(t)<1 ζ m , l m , k ( t ) < 1 for an active pulse k k k . By Lemma 7.1 (d), the slot of pulse k k k lies in its block and ζ m , k \zeta_{m,k} ζ m , k vanishes outside the slot enlarged by 1 τ m τ m ′ < 2 τ m ′ \frac{1}{\tau_m}\tau'_m<2\tau'_m τ m 1 τ m ′ < 2 τ m ′ ; so this happens only within 2 τ m ′ 2\tau'_m 2 τ m ′ of the ends of blocks, a fraction at most 4 τ m ′ / h m 4\tau'_m/h_m 4 τ m ′ / h m of [ 0 , 4 h m ] [0,4h_m] [ 0 , 4 h m ] . As both sums lie in [ 0 , 1 ] [0,1] [ 0 , 1 ] , Lemma 7.1 (b) gives
∣ ∫ 0 4 h m ∑ k ∈ 2 Z + 1 ζ ~ m , k ( t ) 2 d t − 9 20 ∣ ≤ 4 τ m ′ h m = 4 4 ∣ ε m − δ ∣ + 1 ≤ ε m − 1 δ . (61) \left|\int_0^{4h_m}\sum_{k\in2\mathbb{Z}+1}\widetilde{\zeta}_{m,k}(t)^2\,\mathrm{d}t-\frac{9}{20}\right|\le\frac{4\tau'_m}{h_m}=\frac{4}{4\lvert\varepsilon_m^{-\delta}\rvert+1}\le\varepsilon_{m-1}^{\delta}. \tag*{(61)} ∫ 0 4 h m k ∈ 2 Z + 1 ∑ ζ m , k ( t ) 2 d t − 20 9 ≤ h m 4 τ m ′ = 4 ∣ ε m − δ ∣ + 1 4 ≤ ε m − 1 δ . (61)
Step 3: Conclusion. By (19) , K ‾ m ( ν ) \overline{K}_m(\nu) K m ( ν ) is half the average of tr J m ν \operatorname{tr}J_m^\nu tr J m ν over [ 0 , 4 h m ] [0,4h_m] [ 0 , 4 h m ] , so (60) and (61) give
∣ K ‾ m ( ν ) − ν − ε m 2 β 4 ν ⋅ 9 20 ∣ ≤ C ε m 2 β ν ( ε m 2 ν τ m + ε m − 1 δ ) , \left|\overline{K}_m(\nu)-\nu-\frac{\varepsilon_m^{2\beta}}{4\nu}\cdot\frac{9}{20}\right|\le C\frac{\varepsilon_m^{2\beta}}{\nu}\left(\frac{\varepsilon_m^2}{\nu\tau_m}+\varepsilon_{m-1}^{\delta}\right), K m ( ν ) − ν − 4 ν ε m 2 β ⋅ 20 9 ≤ C ν ε m 2 β ( ν τ m ε m 2 + ε m − 1 δ ) ,
and ε m 2 β 4 ν ⋅ 9 20 = 9 80 ε m 2 β ν \frac{\varepsilon_m^{2\beta}}{4\nu}\cdot\frac{9}{20}=\frac{9}{80}\frac{\varepsilon_m^{2\beta}}{\nu} 4 ν ε m 2 β ⋅ 20 9 = 80 9 ν ε m 2 β . This is (a).
Proof of Proposition 2.1(b). We may assume M ≥ 2 M\ge2 M ≥ 2 . Going up from M M M , we bound the logarithm of the ratio s m : = κ m / ( ( 9 / 80 ) 1 / 2 ε m β + γ ) s_m:=\kappa_m/((9/80)^{1/2}\varepsilon_m^{\beta+\gamma}) s m := κ m / (( 9/80 ) 1/2 ε m β + γ ) , uniformly in m m m . This normalization is the one a level reproduces: by (62) and (63) below, s m = 1 s_m=1 s m = 1 gives s m − 1 ≈ ( ε m / ε m − 1 q ) β − γ ( 1 + ε m 2 γ ) s_{m-1}\approx(\varepsilon_m/\varepsilon_{m-1}^q)^{\beta-\gamma}(1+\varepsilon_m^{2\gamma}) s m − 1 ≈ ( ε m / ε m − 1 q ) β − γ ( 1 + ε m 2 γ ) , a number close to 1.
Step 1: One level. Let 2 ≤ m ≤ M 2\le m\le M 2 ≤ m ≤ M . By (a), since κ m + 9 80 ε m 2 β / κ m ≥ 9 80 ε m 2 β / κ m \kappa_m+\frac{9}{80}\varepsilon_m^{2\beta}/\kappa_m\ge\frac{9}{80}\varepsilon_m^{2\beta}/\kappa_m κ m + 80 9 ε m 2 β / κ m ≥ 80 9 ε m 2 β / κ m ,
∣ κ m − 1 κ m + 9 80 ε m 2 β / κ m − 1 ∣ ≤ C ( ε m 2 κ m τ m + ε m − 1 δ ) . (62) \left|\frac{\kappa_{m-1}}{\kappa_m+\frac{9}{80}\varepsilon_m^{2\beta}/\kappa_m}-1\right|\le C\left(\frac{\varepsilon_m^2}{\kappa_m\tau_m}+\varepsilon_{m-1}^{\delta}\right). \tag*{(62)} κ m + 80 9 ε m 2 β / κ m κ m − 1 − 1 ≤ C ( κ m τ m ε m 2 + ε m − 1 δ ) . (62)
Since q ( β − γ ) = β + γ q(\beta-\gamma)=\beta+\gamma q ( β − γ ) = β + γ (Lemma 7.1 (c)), the comparison value satisfies the exact identity
κ m + 9 80 ε m 2 β / κ m ( 9 / 80 ) 1 / 2 ε m − 1 β + γ = ( ε m ε m − 1 q ) β − γ ( ε m 2 γ s m + 1 s m ) . (63) \frac{\kappa_m+\frac{9}{80}\varepsilon_m^{2\beta}/\kappa_m}{(9/80)^{1/2}\varepsilon_{m-1}^{\beta+\gamma}}=\left(\frac{\varepsilon_m}{\varepsilon_{m-1}^q}\right)^{\beta-\gamma}\left(\varepsilon_m^{2\gamma}s_m+\frac{1}{s_m}\right). \tag*{(63)} ( 9/80 ) 1/2 ε m − 1 β + γ κ m + 80 9 ε m 2 β / κ m = ( ε m − 1 q ε m ) β − γ ( ε m 2 γ s m + s m 1 ) . (63)
So, up to the factor ( ε m / ε m − 1 q ) β − γ (\varepsilon_m/\varepsilon_{m-1}^q)^{\beta-\gamma} ( ε m / ε m − 1 q ) β − γ and the error in (62) , a level maps s m s_m s m to ε m 2 γ s m + 1 / s m \varepsilon_m^{2\gamma}s_m+1/s_m ε m 2 γ s m + 1/ s m , essentially 1 / s m 1/s_m 1/ s m , and inversion preserves ∣ log ∣ |\log| ∣ log ∣ : precisely, ∣ log ( ε m 2 γ s m + 1 / s m ) ∣ ≤ ∣ log s m ∣ + ε m 2 γ |\log(\varepsilon_m^{2\gamma}s_m+1/s_m)|\le|\log s_m|+\varepsilon_m^{2\gamma} ∣ log ( ε m 2 γ s m + 1/ s m ) ∣ ≤ ∣ log s m ∣ + ε m 2 γ , since ε m 2 γ s m + 1 / s m \varepsilon_m^{2\gamma}s_m+1/s_m ε m 2 γ s m + 1/ s m lies between 1 / s m 1/s_m 1/ s m and ( 1 + ε m 2 γ ) max { s m , 1 / s m } (1+\varepsilon_m^{2\gamma})\max\{s_m,1/s_m\} ( 1 + ε m 2 γ ) max { s m , 1/ s m } and log ( 1 + ε m 2 γ ) ≤ ε m 2 γ \log(1+\varepsilon_m^{2\gamma})\le\varepsilon_m^{2\gamma} log ( 1 + ε m 2 γ ) ≤ ε m 2 γ . Moreover:
The error of (a). By Lemma 7.1 (a)–(c), ε m 2 − β − γ ≤ 2 ε m − 1 2 − β + 8 δ \varepsilon_m^{2-\beta-\gamma}\le2\varepsilon_{m-1}^{2-\beta+8\delta} ε m 2 − β − γ ≤ 2 ε m − 1 2 − β + 8 δ and τ m ≥ c ε m − 1 2 − β + 4 δ \tau_m\ge c\varepsilon_{m-1}^{2-\beta+4\delta} τ m ≥ c ε m − 1 2 − β + 4 δ , so ε m 2 / ( κ m τ m ) = ( ε m 2 − β − γ / τ m ) ( ε m β + γ / κ m ) ≤ C ε m − 1 4 δ / s m \varepsilon_m^2/(\kappa_m\tau_m)=(\varepsilon_m^{2-\beta-\gamma}/\tau_m)(\varepsilon_m^{\beta+\gamma}/\kappa_m)\le C\varepsilon_{m-1}^{4\delta}/s_m ε m 2 / ( κ m τ m ) = ( ε m 2 − β − γ / τ m ) ( ε m β + γ / κ m ) ≤ C ε m − 1 4 δ / s m .
The factor. ∣ ( ε m / ε m − 1 q ) β − γ − 1 ∣ ≤ C ε m − 1 \left|(\varepsilon_m/\varepsilon_{m-1}^q)^{\beta-\gamma}-1\right|\le C\varepsilon_{m-1} ( ε m / ε m − 1 q ) β − γ − 1 ≤ C ε m − 1 : with λ : = Λ − q m − 1 / ( q − 1 ) \lambda:=\Lambda^{-q^{m-1}/(q-1)} λ := Λ − q m − 1 / ( q − 1 ) , ε m − 1 − 1 \varepsilon_{m-1}^{-1} ε m − 1 − 1 and ε m − 1 \varepsilon_m^{-1} ε m − 1 are the ceilings of λ − 1 \lambda^{-1} λ − 1 and of λ − q \lambda^{-q} λ − q , so ε m / ε m − 1 q = ⌈ λ − 1 ⌉ q / ⌈ λ − q ⌉ \varepsilon_m/\varepsilon_{m-1}^q=\lceil\lambda^{-1}\rceil^q/\lceil\lambda^{-q}\rceil ε m / ε m − 1 q = ⌈ λ − 1 ⌉ q / ⌈ λ − q ⌉ lies between ( 1 + λ q ) − 1 (1+\lambda^q)^{-1} ( 1 + λ q ) − 1 and ( 1 + λ ) q (1+\lambda)^q ( 1 + λ ) q , and λ ≤ 2 ε m − 1 \lambda\le2\varepsilon_{m-1} λ ≤ 2 ε m − 1 .
If the right side of (62) is at most 1 2 \frac{1}{2} 2 1 , then, since the logarithm of a number in [ 1 2 , 2 ] \left[\frac{1}{2},2\right] [ 2 1 , 2 ] is at most twice its distance to 1 in absolute value, (63) gives
∣ log s m − 1 ∣ ≤ ∣ log s m ∣ + C ε m − 1 + ε m 2 γ + C ( 1 + 1 s m ) ε m − 1 δ . (64) \left|\log s_{m-1}\right|\le\left|\log s_m\right|+C\varepsilon_{m-1}+\varepsilon_m^{2\gamma}+C\left(1+\frac{1}{s_m}\right)\varepsilon_{m-1}^{\delta}. \tag*{(64)} ∣ log s m − 1 ∣ ≤ ∣ log s m ∣ + C ε m − 1 + ε m 2 γ + C ( 1 + s m 1 ) ε m − 1 δ . (64)
Step 2: The start. Let m = M m=M m = M . Since 1 2 ε M β − γ ≤ κ M ≤ 2 ε M β − γ \frac{1}{2}\varepsilon_M^{\beta-\gamma}\le\kappa_M\le2\varepsilon_M^{\beta-\gamma} 2 1 ε M β − γ ≤ κ M ≤ 2 ε M β − γ , the two terms of the sum on the right side of (63) lie in [ 1 2 ( 80 / 9 ) 1 / 2 , 2 ( 80 / 9 ) 1 / 2 ] ⊂ [ 1.49 , 5.97 ] \left[\frac{1}{2}(80/9)^{1/2},2(80/9)^{1/2}\right]\subset[1.49,5.97] [ 2 1 ( 80/9 ) 1/2 , 2 ( 80/9 ) 1/2 ] ⊂ [ 1.49 , 5.97 ] and in [ 0 , 2 ( 9 / 80 ) 1 / 2 ε M 2 γ ] ⊂ [ 0 , 0.68 ] \left[0,2(9/80)^{1/2}\varepsilon_M^{2\gamma}\right]\subset[0,0.68] [ 0 , 2 ( 9/80 ) 1/2 ε M 2 γ ] ⊂ [ 0 , 0.68 ] , and the factor ( ε M / ε M − 1 q ) β − γ (\varepsilon_M/\varepsilon_{M-1}^q)^{\beta-\gamma} ( ε M / ε M − 1 q ) β − γ lies in [ 2 − 4 / 3 , 2 4 / 3 ] \left[2^{-4/3},2^{4/3}\right] [ 2 − 4/3 , 2 4/3 ] , by Lemma 7.1 (a) and 0 < β − γ < 4 3 0<\beta-\gamma<\frac{4}{3} 0 < β − γ < 3 4 . By the first bullet, if Λ ≥ C \Lambda\ge C Λ ≥ C , the right side of (62) is at most C ε M − 1 δ ≤ 1 2 C\varepsilon_{M-1}^{\delta}\le\frac{1}{2} C ε M − 1 δ ≤ 2 1 . So s M − 1 s_{M-1} s M − 1 lies in [ 1 2 ⋅ 2 − 4 / 3 ⋅ 1.49 , 3 2 ⋅ 2 4 / 3 ⋅ ( 5.97 + 0.68 ) ] ⊂ [ exp ( − 4 ) , exp ( 4 ) ] \left[\frac{1}{2}\cdot2^{-4/3}\cdot1.49,\frac{3}{2}\cdot2^{4/3}\cdot(5.97+0.68)\right]\subset[\exp(-4),\exp(4)] [ 2 1 ⋅ 2 − 4/3 ⋅ 1.49 , 2 3 ⋅ 2 4/3 ⋅ ( 5.97 + 0.68 ) ] ⊂ [ exp ( − 4 ) , exp ( 4 )] .
Step 3: Up the levels. The constants C C C of Step 1 come from (a) and Lemma 7.1 and do not depend on Λ \Lambda Λ , and the start bound 4 is absolute; so we fix the bound 5 = 4 + 1 5=4+1 5 = 4 + 1 first and Λ 0 \Lambda_0 Λ 0 last. We choose Λ 0 ≥ 128 \Lambda_0\ge128 Λ 0 ≥ 128 so large that, for Λ ≥ Λ 0 \Lambda\ge\Lambda_0 Λ ≥ Λ 0 : the requirement Λ ≥ C \Lambda\ge C Λ ≥ C of Step 2 holds; for every m ≥ 2 m\ge2 m ≥ 2 , C ( 1 + exp ( 5 ) ) ( ε m − 1 4 δ + ε m − 1 δ ) ≤ 1 2 C(1+\exp(5))(\varepsilon_{m-1}^{4\delta}+\varepsilon_{m-1}^{\delta})\le\frac{1}{2} C ( 1 + exp ( 5 )) ( ε m − 1 4 δ + ε m − 1 δ ) ≤ 2 1 ; and
∑ m ≥ 2 ( C ε m − 1 + ε m 2 γ + C ( 1 + exp ( 5 ) ) ε m − 1 δ ) ≤ 1 , \sum_{m\ge2}\left(C\varepsilon_{m-1}+\varepsilon_m^{2\gamma}+C(1+\exp(5))\varepsilon_{m-1}^{\delta}\right)\le1, m ≥ 2 ∑ ( C ε m − 1 + ε m 2 γ + C ( 1 + exp ( 5 )) ε m − 1 δ ) ≤ 1 ,
which is possible since ε m ′ ≤ Λ − m ′ \varepsilon_{m'}\le\Lambda^{-m'} ε m ′ ≤ Λ − m ′ by Lemma 7.1 (a). We prove by induction on m m m , for m = M − 1 , M − 2 , … , 1 m=M-1,M-2,\ldots,1 m = M − 1 , M − 2 , … , 1 in turn, that
∣ log s m ∣ ≤ 4 + ∑ m ′ = m + 1 M − 1 ( C ε m ′ − 1 + ε m ′ 2 γ + C ( 1 + exp ( 5 ) ) ε m ′ − 1 δ ) . \left|\log s_m\right|\le4+\sum_{m'=m+1}^{M-1}\left(C\varepsilon_{m'-1}+\varepsilon_{m'}^{2\gamma}+C(1+\exp(5))\varepsilon_{m'-1}^{\delta}\right). ∣ log s m ∣ ≤ 4 + m ′ = m + 1 ∑ M − 1 ( C ε m ′ − 1 + ε m ′ 2 γ + C ( 1 + exp ( 5 )) ε m ′ − 1 δ ) .
For m = M − 1 m=M-1 m = M − 1 this is Step 2. If it holds at some m ≥ 2 m\ge2 m ≥ 2 , then ∣ log s m ∣ ≤ 5 \left|\log s_m\right|\le5 ∣ log s m ∣ ≤ 5 by the display above, so the right side of (62) is at most 1 2 \frac{1}{2} 2 1 , and (64) with 1 / s m ≤ exp ( 5 ) 1/s_m\le\exp(5) 1/ s m ≤ exp ( 5 ) gives it at m − 1 m-1 m − 1 . Since the sum is at most 1, ∣ log s m ∣ ≤ 5 \left|\log s_m\right|\le5 ∣ log s m ∣ ≤ 5 for 1 ≤ m ≤ M − 1 1\le m\le M-1 1 ≤ m ≤ M − 1 : this is (b).
Proof of Proposition 2.1(c). Let 2 ≤ m ≤ M 2\le m\le M 2 ≤ m ≤ M . First, κ m ≤ K ‾ m ( κ m ) = κ m − 1 \kappa_m\le\overline{K}_m(\kappa_m)=\kappa_{m-1} κ m ≤ K m ( κ m ) = κ m − 1 by Lemma 7.10(iii). Since 1 ≤ m − 1 < M 1\le m-1<M 1 ≤ m − 1 < M , (b) gives κ m − 1 ≍ ε m − 1 β + γ \kappa_{m-1}\asymp\varepsilon_{m-1}^{\beta+\gamma} κ m − 1 ≍ ε m − 1 β + γ , and (7) gives the bound on ( κ m − 1 h m ) 1 / 2 (\kappa_{m-1}h_m)^{1/2} ( κ m − 1 h m ) 1/2 . If m < M m<M m < M , then (b) gives κ m ≍ ε m β + γ \kappa_m\asymp\varepsilon_m^{\beta+\gamma} κ m ≍ ε m β + γ , and Lemma 7.1 (a)–(c) give
ε m 2 κ m τ m ≤ C ε m 2 − β − γ τ m ≤ C ε m − 1 q ( 2 − β − γ ) − ( 2 − β + 4 δ ) = C ε m − 1 4 δ , κ m − 1 κ m ≍ ε m − 1 β + γ ε m − 1 q ( β + γ ) = ε m − 1 − 2 q γ . \frac{\varepsilon_m^2}{\kappa_m\tau_m}\le C\frac{\varepsilon_m^{2-\beta-\gamma}}{\tau_m}\le C\varepsilon_{m-1}^{q(2-\beta-\gamma)-(2-\beta+4\delta)}=C\varepsilon_{m-1}^{4\delta},\qquad\frac{\kappa_{m-1}}{\kappa_m}\asymp\frac{\varepsilon_{m-1}^{\beta+\gamma}}{\varepsilon_{m-1}^{q(\beta+\gamma)}}=\varepsilon_{m-1}^{-2q\gamma}. κ m τ m ε m 2 ≤ C τ m ε m 2 − β − γ ≤ C ε m − 1 q ( 2 − β − γ ) − ( 2 − β + 4 δ ) = C ε m − 1 4 δ , κ m κ m − 1 ≍ ε m − 1 q ( β + γ ) ε m − 1 β + γ = ε m − 1 − 2 q γ .
If m = M m=M m = M , then κ M ≥ 1 2 ε M β − γ = 1 2 ε M β + γ ε M − 2 γ \kappa_M\ge\frac{1}{2}\varepsilon_M^{\beta-\gamma}=\frac{1}{2}\varepsilon_M^{\beta+\gamma}\varepsilon_M^{-2\gamma} κ M ≥ 2 1 ε M β − γ = 2 1 ε M β + γ ε M − 2 γ , so the same computation gives ε M 2 / ( κ M τ M ) ≤ C ε M − 1 4 δ + 2 q γ ≤ C ε M − 1 4 δ \varepsilon_M^2/(\kappa_M\tau_M)\le C\varepsilon_{M-1}^{4\delta+2q\gamma}\le C\varepsilon_{M-1}^{4\delta} ε M 2 / ( κ M τ M ) ≤ C ε M − 1 4 δ + 2 q γ ≤ C ε M − 1 4 δ . Moreover κ M ≍ ε M β − γ ≍ ε M − 1 q ( β − γ ) = ε M − 1 β + γ ≍ κ M − 1 \kappa_M\asymp\varepsilon_M^{\beta-\gamma}\asymp\varepsilon_{M-1}^{q(\beta-\gamma)}=\varepsilon_{M-1}^{\beta+\gamma}\asymp\kappa_{M-1} κ M ≍ ε M β − γ ≍ ε M − 1 q ( β − γ ) = ε M − 1 β + γ ≍ κ M − 1 , so κ M − 1 ≤ C κ M \kappa_{M-1}\le C\kappa_M κ M − 1 ≤ C κ M and ( κ M − 1 / κ M ) 1 / 2 ≤ C ≤ C ε M − 1 − q γ (\kappa_{M-1}/\kappa_M)^{1/2}\le C\le C\varepsilon_{M-1}^{-q\gamma} ( κ M − 1 / κ M ) 1/2 ≤ C ≤ C ε M − 1 − q γ .
Energy estimates for families and paired smoothing
In this section we prove Propositions 6.1 and 6.2 . We prove Proposition 6.2 in the form at higher order announced after it, Proposition 8.3 , together with its analogue for sources without smoothing, Proposition 8.6; Sections 12–15 use these two.
Arrays and coefficients. For a function g g g on T 2 \mathbb{T}^{2} T 2 with values in a Euclidean space, ∇ j g \nabla^{j}g ∇ j g is the array of its 2 j 2^{j} 2 j ordered partial derivatives of order j j j , with the Euclidean norm, as in (36) . A coefficient is a function A A A on T 2 \mathbb{T}^{2} T 2 , or on [ t 1 , t 2 ] × T 2 [t_{1},t_{2}]\times\mathbb{T}^{2} [ t 1 , t 2 ] × T 2 , whose values are linear maps between Euclidean spaces, normed as in Section 2 2 2 . Its derivatives are measured by
∣ ∇ j A ( x ) ∣ : = sup { ∣ ∇ j ( A z ) ( x ) ∣ : ∣ z ∣ ≤ 1 } . \lvert\nabla^{j}A(x)\rvert:= \sup\{\lvert\nabla^{j}(Az)(x)\rvert: \lvert z\rvert\le1\}. ∣ ∇ j A ( x )∣ := sup {∣ ∇ j ( A z ) ( x )∣ : ∣ z ∣ ≤ 1 } .
where A z Az A z is the function x ↦ A ( x ) z x\mapsto A(x)z x ↦ A ( x ) z for a fixed z z z . For the drift, the norm ∣ ∇ j ∇ ⊥ Φ ( x ) ∣ \lvert\nabla^{j}\nabla^{\perp}\Phi(x)\rvert ∣ ∇ j ∇ ⊥ Φ ( x )∣ of the j j j -linear map of Section 2 2 2 equals this norm for the coefficient z ↦ ∇ ⊥ Φ ⋅ z z\mapsto\nabla^{\perp}\Phi\cdot z z ↦ ∇ ⊥ Φ ⋅ z , since the two matrices, acting on arrays, are transposes of each other. For functions g v g_{v} g v on T 2 \mathbb{T}^{2} T 2 , v ∈ V v\in V v ∈ V , masses ω v > 0 \omega_{v}>0 ω v > 0 and g = ( g v ) v ∈ V g=(g_{v})_{v\in V} g = ( g v ) v ∈ V , we write
∥ g ∥ H ρ , ω N ( T 2 ) 2 : = ∑ v ∈ V ∥ g v ∥ H ρ N ( T 2 ) 2 ω v , ∥ g ∥ H ρ , ω ( N v ) ( T 2 ) 2 : = ∑ v ∈ V ∥ g v ∥ H ρ N v ( T 2 ) 2 ω v , (65) \lVert g\rVert_{H_{\rho,\omega}^{N}(\mathbb{T}^{2})}^{2}:=\sum_{v\in V}\frac{\lVert g_{v}\rVert_{H_{\rho}^{N}(\mathbb{T}^{2})}^{2}}{\omega_{v}},\qquad
\lVert g\rVert_{H_{\rho,\omega}^{(N_{v})}(\mathbb{T}^{2})}^{2}:=\sum_{v\in V}\frac{\lVert g_{v}\rVert_{H_{\rho}^{N_{v}}(\mathbb{T}^{2})}^{2}}{\omega_{v}}, \tag*{(65)} ∥ g ∥ H ρ , ω N ( T 2 ) 2 := v ∈ V ∑ ω v ∥ g v ∥ H ρ N ( T 2 ) 2 , ∥ g ∥ H ρ , ω ( N v ) ( T 2 ) 2 := v ∈ V ∑ ω v ∥ g v ∥ H ρ N v ( T 2 ) 2 , (65)
the second for integers N v ≥ 0 N_{v}\ge0 N v ≥ 0 ; for N = 0 N=0 N = 0 the first is the norm ∥ g ∥ L ω 2 ( T 2 ) \lVert g\rVert_{L_{\omega}^{2}(\mathbb{T}^{2})} ∥ g ∥ L ω 2 ( T 2 ) of (35) .
Products and derivatives in H ρ N H_{\rho}^{N} H ρ N
Lemma 8.1 (Products and derivatives at the scale ρ \rho ρ ). Let N ≥ 0 N\ge0 N ≥ 0 be an integer and ρ > 0 \rho>0 ρ > 0 .
(a) For every coefficient A A A of class C N C^{N} C N on T 2 \mathbb{T}^{2} T 2 and every g ∈ H N ( T 2 ) g\in H^{N}(\mathbb{T}^{2}) g ∈ H N ( T 2 ) with values in the domain of A A A ,
∥ A g ∥ H ρ N ( T 2 ) ≤ ( ∑ j = 0 N ρ j j ! sup T 2 ∣ ∇ j A ∣ ) ∥ g ∥ H ρ N ( T 2 ) . \lVert Ag\rVert_{H_{\rho}^{N}(\mathbb{T}^{2})}\le\left(\sum_{j=0}^{N}\frac{\rho^{j}}{j!}\sup_{\mathbb{T}^{2}}\lvert\nabla^{j}A\rvert\right)\lVert g\rVert_{H_{\rho}^{N}(\mathbb{T}^{2})}. ∥ A g ∥ H ρ N ( T 2 ) ≤ ( j = 0 ∑ N j ! ρ j T 2 sup ∣ ∇ j A ∣ ) ∥ g ∥ H ρ N ( T 2 ) . (b) For every integer i ≥ 0 i\ge0 i ≥ 0 and every g ∈ H N + i ( T 2 ) g\in H^{N+i}(\mathbb{T}^{2}) g ∈ H N + i ( T 2 ) , ∥ ∇ i g ∥ H ρ N ( T 2 ) ≤ ρ − i ( ( N + i ) ! / N ! ) ∥ g ∥ H ρ N + i ( T 2 ) \lVert\nabla^{i}g\rVert_{H_{\rho}^{N}(\mathbb{T}^{2})}\le\rho^{-i}((N+i)!/N!)\lVert g\rVert_{H_{\rho}^{N+i}(\mathbb{T}^{2})} ∥ ∇ i g ∥ H ρ N ( T 2 ) ≤ ρ − i (( N + i )! / N !) ∥ g ∥ H ρ N + i ( T 2 ) .
Proof. (a) The weights turn the Leibniz rule into a convolution: there are ( j i ) \binom{j}{i} ( i j ) ways to let i i i of the j j j derivatives of ∇ j ( A g ) \nabla^{j}(Ag) ∇ j ( A g ) fall on A A A , each giving an array of norm at most ∣ ∇ i A ∣ ∣ ∇ j − i g ∣ \lvert\nabla^{i}A\rvert\lvert\nabla^{j-i}g\rvert ∣ ∇ i A ∣ ∣ ∇ j − i g ∣ at every point, and ( j i ) ρ j / j ! = ( ρ i / i ! ) ( ρ j − i / ( j − i ) ! ) \binom{j}{i}\rho^{j}/j!=(\rho^{i}/i!)(\rho^{j-i}/(j-i)!) ( i j ) ρ j / j ! = ( ρ i / i !) ( ρ j − i / ( j − i )!) . So the sequence ( ρ j / j ! ) ∥ ∇ j ( A g ) ∥ L 2 ( T 2 ) (\rho^{j}/j!)\lVert\nabla^{j}(Ag)\rVert_{L^{2}(\mathbb{T}^{2})} ( ρ j / j !) ∥ ∇ j ( A g ) ∥ L 2 ( T 2 ) is bounded by the convolution of ( ρ i / i ! ) sup T 2 ∣ ∇ i A ∣ (\rho^{i}/i!)\sup_{\mathbb{T}^{2}}\lvert\nabla^{i}A\rvert ( ρ i / i !) sup T 2 ∣ ∇ i A ∣ and ( ρ i / i ! ) ∥ ∇ i g ∥ L 2 ( T 2 ) (\rho^{i}/i!)\lVert\nabla^{i}g\rVert_{L^{2}(\mathbb{T}^{2})} ( ρ i / i !) ∥ ∇ i g ∥ L 2 ( T 2 ) , and Young’s inequality for a summable sequence convolved with a square-summable one gives (a). (b) For j ≤ N j\le N j ≤ N , ρ j / j ! = ρ − i ( ( j + i ) ! / j ! ) ρ j + i / ( j + i ) ! \rho^{j}/j!=\rho^{-i}((j+i)!/j!)\rho^{j+i}/(j+i)! ρ j / j ! = ρ − i (( j + i )! / j !) ρ j + i / ( j + i )! and ( j + i ) ! / j ! ≤ ( N + i ) ! / N ! (j+i)!/j!\le(N+i)!/N! ( j + i )! / j ! ≤ ( N + i )! / N ! . □ \square □
Paired smoothing
The kernel. We prove the properties of W ℓ W_{\ell} W ℓ stated before Proposition 6.1 . With u ℓ ( z ) : = ( 2 ℓ ) 2 u ( 2 ℓ z ) u_{\ell}(z):=(2\ell)^{2}u(2\ell z) u ℓ ( z ) := ( 2 ℓ ) 2 u ( 2 ℓ z ) , the autocorrelation u ℓ ∗ u ℓ ( − ⋅ ) u_{\ell}*u_{\ell}(-\mathord\cdot) u ℓ ∗ u ℓ ( − ⋅ ) has Fourier transform ∣ u ^ ( ξ / ( 2 ℓ ) ) ∣ 2 \lvert\widehat{u}(\xi/(2\ell))\rvert^{2} ∣ u ( ξ / ( 2 ℓ )) ∣ 2 , so expanding the power in (43) gives W ℓ = ∑ j = 1 2 ℓ ( − 1 ) j + 1 ( 2 ℓ j ) ( u ℓ ∗ u ℓ ( − ⋅ ) ) ∗ j W_{\ell}=\sum_{j=1}^{2\ell}(-1)^{j+1}\binom{2\ell}{j}(u_{\ell}*u_{\ell}(-\mathord\cdot))^{*j} W ℓ = ∑ j = 1 2 ℓ ( − 1 ) j + 1 ( j 2 ℓ ) ( u ℓ ∗ u ℓ ( − ⋅ ) ) ∗ j , where ( ⋅ ) ∗ j (\mathord\cdot)^{*j} ( ⋅ ) ∗ j is the j j j -fold convolution. The autocorrelation is smooth with support in the disc of radius 1 / ℓ 1/\ell 1/ ℓ , so W ℓ W_{\ell} W ℓ is smooth with support in the unit disc, and ∫ R 2 W ℓ = W ^ ℓ ( 0 ) = 1 \int_{\mathbb{R}^{2}}W_{\ell}=\widehat{W}_{\ell}(0)=1 ∫ R 2 W ℓ = W ℓ ( 0 ) = 1 . Since u ≥ 0 u\ge0 u ≥ 0 has integral one, 0 ≤ ∣ u ^ ∣ 2 ≤ 1 0\le\lvert\widehat{u}\rvert^{2}\le1 0 ≤ ∣ u ∣ 2 ≤ 1 , so W ^ ℓ \widehat{W}_{\ell} W ℓ takes values in [ 0 , 1 ] [0,1] [ 0 , 1 ] . The smooth function 1 − ∣ u ^ ∣ 2 1-\lvert\widehat{u}\rvert^{2} 1 − ∣ u ∣ 2 vanishes at its minimum, the origin, so 1 − W ^ ℓ = ( 1 − ∣ u ^ ( ξ / ( 2 ℓ ) ) ∣ 2 ) 2 ℓ = O ( ∣ ξ ∣ 2 ℓ ) 1-\widehat{W}_{\ell}=(1-\lvert\widehat{u}(\xi/(2\ell))\rvert^{2})^{2\ell}=O(\lvert\xi\rvert^{2\ell}) 1 − W ℓ = ( 1 − ∣ u ( ξ / ( 2 ℓ )) ∣ 2 ) 2 ℓ = O (∣ ξ ∣ 2 ℓ ) , and the derivatives of W ^ ℓ \widehat{W}_{\ell} W ℓ of orders 1 1 1 to 2 ℓ − 1 2\ell-1 2 ℓ − 1 , which are constant multiples of the moments of W ℓ W_{\ell} W ℓ , vanish at the origin. Consequently, each average S ℓ , r , Ω S_{\ell,r,\Omega} S ℓ , r , Ω multiplies the Fourier coefficient of a function on T 2 \mathbb{T}^{2} T 2 at the frequency ξ \xi ξ by W ^ ℓ ( r Ω T ξ ) ∈ [ 0 , 1 ] \widehat{W}_{\ell}(r\Omega^{T}\xi)\in[0,1] W ℓ ( r Ω T ξ ) ∈ [ 0 , 1 ] ; it commutes with derivatives and, by Plancherel’s theorem, contracts every H ρ N ( T 2 ) H_{\rho}^{N}(\mathbb{T}^{2}) H ρ N ( T 2 ) .
Proof of Proposition 6.1. Step 1: Derivatives along Ω 1 e 1 \Omega_{1}e_{1} Ω 1 e 1 and Ω 2 e 2 \Omega_{2}e_{2} Ω 2 e 2 . Each of these derivatives falls on one average and costs r − 1 r^{-1} r − 1 , whatever the short axes of the ellipses (Figure 8.1 ). Let $d_{(1)}:=\Omega_{1}e_{1}\cdot\nabla and ∂ ( 2 ) : = Ω 2 e 2 ⋅ ∇ \partial_{(2)} := \Omega_{2}e_{2} \cdot\nabla ∂ ( 2 ) := Ω 2 e 2 ⋅ ∇ . On y ↦ g ( y − r Ω 1 z ) y \mapsto g(y-r\Omega_{1}z) y ↦ g ( y − r Ω 1 z ) , ∂ ( 1 ) \partial_{(1)} ∂ ( 1 ) is − r − 1 ∂ z 1 -r^{-1}\partial_{z_{1}} − r − 1 ∂ z 1 , and integration by parts in z 1 z_{1} z 1 in (44) gives
Figure 8.1. The two thin ellipses { y − r Ω 1 z : ∣ z ∣ < 1 } \{y-r\Omega_{1}z:|z|<1\} { y − r Ω 1 z : ∣ z ∣ < 1 } and { y − r Ω 2 z : ∣ z ∣ < 1 } \{y-r\Omega_{2}z:|z|<1\} { y − r Ω 2 z : ∣ z ∣ < 1 } , centred at y y y , over which the two averages of S ℓ , r , Ω 1 S ℓ , r , Ω 2 S_{\ell,r,\Omega_{1}}S_{\ell,r,\Omega_{2}} S ℓ , r , Ω 1 S ℓ , r , Ω 2 are taken, and the vectors r Ω 1 e 1 r\Omega_{1}e_{1} r Ω 1 e 1 and r Ω 2 e 2 r\Omega_{2}e_{2} r Ω 2 e 2 .
∂ ( 1 ) S ℓ , r , Ω 1 g ( y ) = r − 1 ∫ R 2 ∂ 1 W ℓ ( z ) g ( y − r Ω 1 z ) d z , \partial_{(1)}S_{\ell,r,\Omega_{1}}g(y)=r^{-1}\int_{\mathbb{R}^{2}}\partial_{1}W_{\ell}(z)g(y-r\Omega_{1}z)\,dz, ∂ ( 1 ) S ℓ , r , Ω 1 g ( y ) = r − 1 ∫ R 2 ∂ 1 W ℓ ( z ) g ( y − r Ω 1 z ) d z ,
and, by Minkowski’s inequality, for every j ≥ 0 j\geq0 j ≥ 0 , ∥ ∂ ( 1 ) j S ℓ , r , Ω 1 g ∥ L 2 ( T 2 ) ≤ r − j ∥ ∂ 1 j W ℓ ∥ L 1 ( R 2 ) ∥ g ∥ L 2 ( T 2 ) \|\partial_{(1)}^{j}S_{\ell,r,\Omega_{1}}g\|_{L^{2}(\mathbb{T}^{2})}\leq r^{-j}\|\partial_{1}^{j}W_{\ell}\|_{L^{1}(\mathbb{R}^{2})}\|g\|_{L^{2}(\mathbb{T}^{2})} ∥ ∂ ( 1 ) j S ℓ , r , Ω 1 g ∥ L 2 ( T 2 ) ≤ r − j ∥ ∂ 1 j W ℓ ∥ L 1 ( R 2 ) ∥ g ∥ L 2 ( T 2 ) ; similarly for ∂ ( 2 ) \partial_{(2)} ∂ ( 2 ) , S ℓ , r , Ω 2 S_{\ell,r,\Omega_{2}} S ℓ , r , Ω 2 and ∂ 2 \partial_{2} ∂ 2 . The pair ( ∂ ( 1 ) , ∂ ( 2 ) ) (\partial_{(1)},\partial_{(2)}) ( ∂ ( 1 ) , ∂ ( 2 ) ) is the image of ( ∂ 1 , ∂ 2 ) (\partial_{1},\partial_{2}) ( ∂ 1 , ∂ 2 ) under the transpose of the matrix with columns Ω 1 e 1 \Omega_{1}e_{1} Ω 1 e 1 , Ω 2 e 2 \Omega_{2}e_{2} Ω 2 e 2 , so each ∂ i \partial_{i} ∂ i is a combination of ∂ ( 1 ) \partial_{(1)} ∂ ( 1 ) and ∂ ( 2 ) \partial_{(2)} ∂ ( 2 ) with coefficients that are entries of the inverse of that matrix, of absolute value at most its ∣ ⋅ ∣ ∞ |\cdot|_{\infty} ∣ ⋅ ∣ ∞ -norm, hence at most 2 2 2 . An ordered derivative of order j j j is thus a sum of 2 j 2^{j} 2 j products of j j j derivatives ∂ ( 1 ) \partial_{(1)} ∂ ( 1 ) or ∂ ( 2 ) \partial_{(2)} ∂ ( 2 ) , with coefficients of absolute value at most 2 j 2^{j} 2 j . The two averages commute with each other and with derivatives; letting ∂ ( 1 ) \partial_{(1)} ∂ ( 1 ) act on S ℓ , r , Ω 1 S_{\ell,r,\Omega_{1}} S ℓ , r , Ω 1 and ∂ ( 2 ) \partial_{(2)} ∂ ( 2 ) on S ℓ , r , Ω 2 S_{\ell,r,\Omega_{2}} S ℓ , r , Ω 2 gives, for every j ≥ 0 j\geq0 j ≥ 0 ,
∥ ∇ j S g ∥ L 2 ( T 2 ) ≤ c j r − j ∥ g ∥ L 2 ( T 2 ) , \|\nabla^{j}Sg\|_{L^{2}(\mathbb{T}^{2})}\leq c_{j}r^{-j}\|g\|_{L^{2}(\mathbb{T}^{2})}, ∥ ∇ j S g ∥ L 2 ( T 2 ) ≤ c j r − j ∥ g ∥ L 2 ( T 2 ) ,
where c j c_{j} c j depends only on u u u , ℓ \ell ℓ and j j j . Proved for smooth g g g , this bound extends to every g ∈ L 2 ( T 2 ) g\in L^{2}(\mathbb{T}^{2}) g ∈ L 2 ( T 2 ) by density, as the averages are bounded on L 2 ( T 2 ) L^{2}(\mathbb{T}^{2}) L 2 ( T 2 ) .
Step 2: The first two bounds. The averages contract H ρ N 1 ( T 2 ) H_{\rho}^{N_{1}}(\mathbb{T}^{2}) H ρ N 1 ( T 2 ) , which gives the second bound and the terms of order j ≤ N 1 j\leq N_{1} j ≤ N 1 of the first. For N 1 < j ≤ N 2 N_{1}<j\leq N_{2} N 1 < j ≤ N 2 , ∇ j S g = ∇ j − N 1 S ∇ N 1 g \nabla^{j}Sg=\nabla^{j-N_{1}}S\nabla^{N_{1}}g ∇ j S g = ∇ j − N 1 S ∇ N 1 g , so Step 1 applied to the entries of ∇ N 1 g \nabla^{N_{1}}g ∇ N 1 g , with r ≤ ρ r\leq\rho r ≤ ρ and N 1 ! ≤ j ! N_{1}!\leq j! N 1 ! ≤ j ! , gives the following bound, whose squares summed over j ≤ N 2 j\leq N_{2} j ≤ N 2 give the first bound:
ρ j j ! ∥ ∇ j S g ∥ L 2 ( T 2 ) ≤ c j − N 1 ( ρ r ) j − N 1 ρ N 1 N 1 ! ∥ ∇ N 1 g ∥ L 2 ( T 2 ) ≤ c j − N 1 ( ρ r ) N 2 − N 1 ∥ g ∥ H ρ N 1 ( T 2 ) . \frac{\rho^{j}}{j!}\|\nabla^{j}Sg\|_{L^{2}(\mathbb{T}^{2})}
\leq c_{j-N_{1}}\left(\frac{\rho}{r}\right)^{j-N_{1}}\frac{\rho^{N_{1}}}{N_{1}!}\|\nabla^{N_{1}}g\|_{L^{2}(\mathbb{T}^{2})}
\leq c_{j-N_{1}}\left(\frac{\rho}{r}\right)^{N_{2}-N_{1}}\|g\|_{H_{\rho}^{N_{1}}(\mathbb{T}^{2})}. j ! ρ j ∥ ∇ j S g ∥ L 2 ( T 2 ) ≤ c j − N 1 ( r ρ ) j − N 1 N 1 ! ρ N 1 ∥ ∇ N 1 g ∥ L 2 ( T 2 ) ≤ c j − N 1 ( r ρ ) N 2 − N 1 ∥ g ∥ H ρ N 1 ( T 2 ) .
Step 3: The approximation. We use the vanishing moments in Fourier form. As 1 − ∣ u ^ ∣ 2 1-|\widehat{u}|^{2} 1 − ∣ u ∣ 2 is smooth, with values in [ 0 , 1 ] [0,1] [ 0 , 1 ] , and vanishes at its minimum, the origin, 1 − ∣ u ^ ∣ 2 ≤ C u ∣ ⋅ ∣ 2 1-|\widehat{u}|^{2}\leq C_{u}|\cdot|^{2} 1 − ∣ u ∣ 2 ≤ C u ∣ ⋅ ∣ 2 with C u C_{u} C u depending only on u u u ; so, by (43) , 2 ℓ > N 1 2\ell>N_{1} 2 ℓ > N 1 and sup ℓ > 0 ( C u / ( 4 ℓ 2 ) ) ℓ = exp ( C u / e ) \sup_{\ell>0}(C_{u}/(4\ell^{2}))^{\ell}=\exp(\sqrt{C_{u}}/e) sup ℓ > 0 ( C u / ( 4 ℓ 2 ) ) ℓ = exp ( C u / e ) , for every ξ ∈ R 2 \xi\in\mathbb{R}^{2} ξ ∈ R 2 ,
0 ≤ 1 − W ^ ℓ ( ξ ) = ( 1 − ∣ u ^ ( ξ 2 ℓ ) ∣ 2 ) ℓ ≤ min { 1 , ( C u ∣ ξ ∣ 2 4 ℓ 2 ) ℓ } ≤ exp ( C u e ) ∣ ξ ∣ N 1 . 0\leq1-\widehat{W}_{\ell}(\xi)=\left(1-\left|\widehat{u}\left(\frac{\xi}{2\ell}\right)\right|^{2}\right)^{\ell}\leq\min\left\{1,\left(\frac{C_{u}|\xi|^{2}}{4\ell^{2}}\right)^{\ell}\right\}\leq\exp\left(\frac{\sqrt{C_{u}}}{e}\right)|\xi|^{N_{1}}. 0 ≤ 1 − W ℓ ( ξ ) = ( 1 − u ( 2 ℓ ξ ) 2 ) ℓ ≤ min { 1 , ( 4 ℓ 2 C u ∣ ξ ∣ 2 ) ℓ } ≤ exp ( e C u ) ∣ ξ ∣ N 1 .
Let i ∈ { 1 , 2 } i\in\{1,2\} i ∈ { 1 , 2 } . As S ℓ , r , Ω i S_{\ell,r,\Omega_{i}} S ℓ , r , Ω i multiplies the Fourier coefficient at ξ \xi ξ by W ^ ℓ ( r Ω i T ξ ) \widehat{W}_{\ell}(r\Omega_{i}^{\mathsf{T}}\xi) W ℓ ( r Ω i T ξ ) , and ∣ Ω i T ξ ∣ ≤ 2 ∣ Ω i ∣ ∞ ∣ ξ ∣ |\Omega_{i}^{\mathsf{T}}\xi|\leq\sqrt{2}|\Omega_{i}|_{\infty}|\xi| ∣ Ω i T ξ ∣ ≤ 2 ∣ Ω i ∣ ∞ ∣ ξ ∣ , Plancherel’s theorem gives
∥ g − S ℓ , r , Ω i g ∥ L 2 ( T 2 ) ≤ exp ( C u e ) ( 2 r C Ω ) N 1 ∥ ∇ N 1 g ∥ L 2 ( T 2 ) ≤ C C Ω N 1 ( r ρ ) N 1 ∥ g ∥ H ρ N 1 ( T 2 ) , \|g-S_{\ell,r,\Omega_{i}}g\|_{L^{2}(\mathbb{T}^{2})}
\leq\exp\left(\frac{\sqrt{C_{u}}}{e}\right)(\sqrt{2}\,rC_{\Omega})^{N_{1}}\|\nabla^{N_{1}}g\|_{L^{2}(\mathbb{T}^{2})}
\leq C C_{\Omega}^{N_{1}}\left(\frac{r}{\rho}\right)^{N_{1}}\|g\|_{H_{\rho}^{N_{1}}(\mathbb{T}^{2})}, ∥ g − S ℓ , r , Ω i g ∥ L 2 ( T 2 ) ≤ exp ( e C u ) ( 2 r C Ω ) N 1 ∥ ∇ N 1 g ∥ L 2 ( T 2 ) ≤ C C Ω N 1 ( ρ r ) N 1 ∥ g ∥ H ρ N 1 ( T 2 ) ,
with C C C depending only on u u u and N 1 N_1 N 1 . Since g − S g = ( g − S ℓ , r , Ω 1 g ) + S ℓ , r , Ω 1 ( g − S ℓ , r , Ω 2 g ) g-Sg=(g-S_{\ell,r,\Omega_1}g)+S_{\ell,r,\Omega_1}(g-S_{\ell,r,\Omega_2}g) g − S g = ( g − S ℓ , r , Ω 1 g ) + S ℓ , r , Ω 1 ( g − S ℓ , r , Ω 2 g ) and 0 ≤ W ^ ℓ ≤ 1 0\leq\widehat{W}_\ell\leq1 0 ≤ W ℓ ≤ 1 , this gives the third bound with the constant 2 C 2C 2 C . This step uses neither the invertibility of Ω i \Omega_i Ω i , nor the transversality of Ω 1 e 1 \Omega_1e_1 Ω 1 e 1 and Ω 2 e 2 \Omega_2e_2 Ω 2 e 2 , nor r ≤ ρ r\leq\rho r ≤ ρ , which proves the last sentence of the proposition. □
Lemma 8.2 (Time-dependent paired smoothing). *Let ℓ ≥ 1 \ell\geq1 ℓ ≥ 1 and r > 0 r>0 r > 0 , and let Ω 1 ( t ) \Omega_1(t) Ω 1 ( t ) , Ω 2 ( t ) \Omega_2(t) Ω 2 ( t ) , t ∈ R t\in\mathbb{R} t ∈ R , be invertible linear maps of R 2 \mathbb{R}^2 R 2 , Borel measurable in t t t , such that, for every t ∈ R t\in\mathbb{R} t ∈ R , the matrix with columns Ω 1 ( t ) e 1 \Omega_1(t)e_1 Ω 1 ( t ) e 1 , Ω 2 ( t ) e 2 \Omega_2(t)e_2 Ω 2 ( t ) e 2 has an inverse of ∣ ⋅ ∣ ∞ \lvert\cdot\rvert_\infty ∣ ⋅ ∣ ∞ -norm at most 2 2 2 . Then the operators S ( t ) : = S ℓ , r , Ω 1 ( t ) S ℓ , r , Ω 2 ( t ) S(t):=S_{\ell,r,\Omega_1(t)}S_{\ell,r,\Omega_2(t)} S ( t ) := S ℓ , r , Ω 1 ( t ) S ℓ , r , Ω 2 ( t ) commute with derivatives; for every g ∈ L 2 ( T 2 ) g\in L^2(\mathbb{T}^2) g ∈ L 2 ( T 2 ) , t ↦ S ( t ) g ∈ L 2 ( T 2 ) t\mapsto S(t)g\in L^2(\mathbb{T}^2) t ↦ S ( t ) g ∈ L 2 ( T 2 ) is Borel measurable; and there exist constants c j ( u , ℓ ) c_j(u,\ell) c j ( u , ℓ ) such that, for every t ∈ R t\in\mathbb{R} t ∈ R , j ≥ 0 j\geq0 j ≥ 0 and g ∈ L 2 ( T 2 ) g\in L^2(\mathbb{T}^2) g ∈ L 2 ( T 2 ) ,
∥ ∇ j S ( t ) g ∥ L 2 ( T 2 ) ≤ c j r − j ∥ g ∥ L 2 ( T 2 ) . \lVert\nabla^jS(t)g\rVert_{L^2(\mathbb{T}^2)}\leq c_jr^{-j}\lVert g\rVert_{L^2(\mathbb{T}^2)}. ∥ ∇ j S ( t ) g ∥ L 2 ( T 2 ) ≤ c j r − j ∥ g ∥ L 2 ( T 2 ) . Thus S ( t ) S(t) S ( t ) is a smoothing operator of width r r r (Subsection 5.4 and the hypotheses of Proposition 6.2 ), with no bound on ∣ Ω i ( t ) ∣ \lvert\Omega_i(t)\rvert ∣ Ω i ( t )∣ .
Proof. Step 1 of the proof of Proposition 6.1 at the time t t t gives the bound, and the multiplier form of S ( t ) S(t) S ( t ) gives the commutation. The multiplier at each frequency depends continuously on the matrices, so t ↦ S ( t ) g t\mapsto S(t)g t ↦ S ( t ) g is Borel measurable, by Plancherel’s theorem and dominated convergence. □
Energy of coupled systems
We keep the setting of Proposition 6.2 , without its condition on B B B , except that U v U_v U v , g v g_v g v and F v 0 F_v^0 F v 0 may take values in a finite-dimensional Euclidean space E v E_v E v (E v = R E_v=\mathbb{R} E v = R for families), and F v F_v F v in E v 2 E_v^2 E v 2 ; A σ , d A_{\sigma,d} A σ , d maps arrays of order d − 1 d-1 d − 1 over E v ′ E_{v'} E v ′ to E v 2 E_v^2 E v 2 , S σ ( t ) S_\sigma(t) S σ ( t ) acts on each coordinate, and, in the component v v v , L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) and H N ( T 2 ) H^N(\mathbb{T}^2) H N ( T 2 ) stand for L 2 ( T 2 ; E v ) L^2(\mathbb{T}^2;E_v) L 2 ( T 2 ; E v ) and H N ( T 2 ; E v ) H^N(\mathbb{T}^2;E_v) H N ( T 2 ; E v ) . We consider, for v ∈ V v\in V v ∈ V ,
{ ∂ t U v + ∇ ⊥ Φ ⋅ ∇ U v − ν Δ U v = ∇ ⋅ ( ∑ v ′ ∈ V ∑ σ : v ′ → v ∑ d = 2 N s r c A σ , d ∇ d − 1 ( S σ U v ′ ) + F v ) + F v 0 in ( t 1 , t 2 ) × T 2 , U v ( t 1 , ⋅ ) = g v on T 2 , (66) \begin{cases}
\partial_tU_v+\nabla^\perp\Phi\cdot\nabla U_v-\nu\Delta U_v
=\nabla\cdot\left(\sum_{v'\in V}\sum_{\sigma:v'\to v}\sum_{d=2}^{N_{\mathrm{src}}}A_{\sigma,d}\nabla^{d-1}(S_\sigma U_{v'})+F_v\right)+F_v^0 & \text{in }(t_1,t_2)\times\mathbb{T}^2,\\
U_v(t_1,\cdot)=g_v & \text{on }\mathbb{T}^2, \tag*{(66)}
\end{cases} { ∂ t U v + ∇ ⊥ Φ ⋅ ∇ U v − ν Δ U v = ∇ ⋅ ( ∑ v ′ ∈ V ∑ σ : v ′ → v ∑ d = 2 N src A σ , d ∇ d − 1 ( S σ U v ′ ) + F v ) + F v 0 U v ( t 1 , ⋅ ) = g v in ( t 1 , t 2 ) × T 2 , on T 2 , (66)
with a flux forcing F v F_v F v and a scalar forcing F v 0 F_v^0 F v 0 . A solution of (66) is a function U ∈ C ( [ t 1 , t 2 ] ; L 2 ( T 2 ) V ) ∩ L 2 ( ( t 1 , t 2 ) ; H 1 ( T 2 ) V ) U\in C([t_1,t_2];L^2(\mathbb{T}^2)^V)\cap L^2((t_1,t_2);H^1(\mathbb{T}^2)^V) U ∈ C ([ t 1 , t 2 ] ; L 2 ( T 2 ) V ) ∩ L 2 (( t 1 , t 2 ) ; H 1 ( T 2 ) V ) such that, for every v ∈ V v\in V v ∈ V and t ∈ [ t 1 , t 2 ] t\in[t_1,t_2] t ∈ [ t 1 , t 2 ] ,
U v ( t ) − g v = ∫ t 1 t ( − ∇ ⊥ Φ ⋅ ∇ U v + ν Δ U v + ∇ ⋅ ( ∑ v ′ , σ , d A σ , d ∇ d − 1 ( S σ U v ′ ) + F v ) + F v 0 ) ( t ′ ′ ) d t ′ ′ . (67) U_v(t)-g_v=\int_{t_1}^{t}\left(-\nabla^\perp\Phi\cdot\nabla U_v+\nu\Delta U_v+\nabla\cdot\left(\sum_{v',\sigma,d}A_{\sigma,d}\nabla^{d-1}(S_\sigma U_{v'})+F_v\right)+F_v^0\right)(t'')\,\mathrm{d}t''. \tag*{(67)} U v ( t ) − g v = ∫ t 1 t − ∇ ⊥ Φ ⋅ ∇ U v + ν Δ U v + ∇ ⋅ v ′ , σ , d ∑ A σ , d ∇ d − 1 ( S σ U v ′ ) + F v + F v 0 ( t ′′ ) d t ′′ . (67)
Here ∇ d − 1 ( S σ U v ′ ) = ∇ d − 2 S σ ∇ U v ′ \nabla^{d-1}(S_\sigma U_{v'})=\nabla^{d-2}S_\sigma\nabla U_{v'} ∇ d − 1 ( S σ U v ′ ) = ∇ d − 2 S σ ∇ U v ′ is measurable in time (by the linearity of S σ ( t ) S_\sigma(t) S σ ( t ) , its bound and the measurability of t ↦ S σ ( t ) h t\mapsto S_\sigma(t)h t ↦ S σ ( t ) h ), with norm at most c d − 2 r σ d − 2 ∥ ∇ U v ′ ∥ L 2 ( ( t 1 , t 2 ) ; L 2 ( T 2 ) ) c_{d-2}r_\sigma^{d-2}\lVert\nabla U_{v'}\rVert_{L^2((t_1,t_2);L^2(\mathbb{T}^2))} c d − 2 r σ d − 2 ∥ ∇ U v ′ ∥ L 2 (( t 1 , t 2 ) ; L 2 ( T 2 )) in L 2 ( ( t 1 , t 2 ) ; L 2 ( T 2 ) ) L^2((t_1,t_2);L^2(\mathbb{T}^2)) L 2 (( t 1 , t 2 ) ; L 2 ( T 2 )) ; (67) is an identity in H − 1 ( T 2 ) H^{-1}(\mathbb{T}^2) H − 1 ( T 2 ) .
For an integer N ≥ 0 N\geq0 N ≥ 0 and ρ > 0 \rho>0 ρ > 0 , the coupling matrix of order N N N at scale ρ \rho ρ is
B v v ′ N , ρ : = ( ω v ′ ω v ) 1 / 2 ∑ σ : v ′ → v ∑ d = 2 N s r c c d − 2 ν r σ d − 2 ∑ j = 0 N ρ j j ! sup [ t 1 , t 2 ] × T 2 ∣ ∇ j A σ , d ∣ , v , v ′ ∈ V . (68) B_{vv'}^{N,\rho}:=\left(\frac{\omega_{v'}}{\omega_v}\right)^{1/2}\sum_{\sigma:v'\to v}\sum_{d=2}^{N_{\mathrm{src}}}\frac{c_{d-2}}{\nu r_\sigma^{d-2}}\sum_{j=0}^{N}\frac{\rho^j}{j!}\sup_{[t_1,t_2]\times\mathbb{T}^2}\lvert\nabla^jA_{\sigma,d}\rvert,\qquad v,v'\in V. \tag*{(68)} B v v ′ N , ρ := ( ω v ω v ′ ) 1/2 σ : v ′ → v ∑ d = 2 ∑ N src ν r σ d − 2 c d − 2 j = 0 ∑ N j ! ρ j [ t 1 , t 2 ] × T 2 sup ∣ ∇ j A σ , d ∣ , v , v ′ ∈ V . (68)
For N = 0 N=0 N = 0 and every ρ \rho ρ , it is the coupling matrix B B B of Proposition 6.2 .
Proposition 8.3 (Energy of coupled systems at higher order). In the setting of (66) , let N ≥ 0 N\geq0 N ≥ 0 be an integer and ρ > 0 \rho>0 ρ > 0 , and assume the following.
(i) For every ( t , x ) ∈ [ t 1 , t 2 ] × T 2 (t,x)\in[t_1,t_2]\times\mathbb{T}^2 ( t , x ) ∈ [ t 1 , t 2 ] × T 2 and every integer 1 ≤ i ≤ N 1\le i\le N 1 ≤ i ≤ N ,
ρ i + 1 ∣ ∇ i ∇ ⊥ Φ ( t , x ) ∣ ≤ 2 − i − 2 i ! ν . \rho^{i+1}\lvert\nabla^i\nabla^\perp\Phi(t,x)\rvert\le2^{-i-2}i!\,\nu. ρ i + 1 ∣ ∇ i ∇ ⊥ Φ ( t , x )∣ ≤ 2 − i − 2 i ! ν .
(ii) For every t ∈ [ t 1 , t 2 ] t\in[t_1,t_2] t ∈ [ t 1 , t 2 ] , the coefficients A σ , d ( t , ⋅ ) A_{\sigma,d}(t,\cdot) A σ , d ( t , ⋅ ) are of class C N C^N C N on T 2 \mathbb{T}^2 T 2 , and, for every j ≤ N j\le N j ≤ N , ∇ j A σ , d \nabla^j A_{\sigma,d} ∇ j A σ , d is bounded on [ t 1 , t 2 ] × T 2 [t_1,t_2]\times\mathbb{T}^2 [ t 1 , t 2 ] × T 2 .
(iii) All row and column sums of B N , ρ B^{N,\rho} B N , ρ are at most 1 8 \frac{1}{8} 8 1 .
Then, for every datum g g g and forcing F F F , F 0 F^0 F 0 such that, for every v ∈ V v\in V v ∈ V , g v ∈ H N ( T 2 ) g_v\in H^N(\mathbb{T}^2) g v ∈ H N ( T 2 ) , F v ∈ L 2 ( ( t 1 , t 2 ) ; H N ( T 2 ) ) F_v\in L^2((t_1,t_2);H^N(\mathbb{T}^2)) F v ∈ L 2 (( t 1 , t 2 ) ; H N ( T 2 )) and F v 0 ∈ L 1 ( ( t 1 , t 2 ) ; H N ( T 2 ) ) F_v^0\in L^1((t_1,t_2);H^N(\mathbb{T}^2)) F v 0 ∈ L 1 (( t 1 , t 2 ) ; H N ( T 2 )) , there exists a unique solution U U U of (66) . It belongs to C ( [ t 1 , t 2 ] ; H N ( T 2 ) V ) ∩ L 2 ( ( t 1 , t 2 ) ; H N + 1 ( T 2 ) V ) C([t_1,t_2];H^N(\mathbb{T}^2)^V)\cap L^2((t_1,t_2);H^{N+1}(\mathbb{T}^2)^V) C ([ t 1 , t 2 ] ; H N ( T 2 ) V ) ∩ L 2 (( t 1 , t 2 ) ; H N + 1 ( T 2 ) V ) , and the following hold.
(a) The solution satisfies
∥ U ∥ L ∞ ( [ t 1 , t 2 ] ; H ρ , ω N ( T 2 ) ) 2 + ν 2 ∥ ∇ U ∥ L 2 ( ( t 1 , t 2 ) ; H ρ , ω N ( T 2 ) ) 2 ≤ 2 ∥ g ∥ H ρ , ω N ( T 2 ) 2 + 8 ν ∥ F ∥ L 2 ( ( t 1 , t 2 ) ; H ρ , ω N ( T 2 ) ) 2 + 10 ∥ F 0 ∥ L 1 ( ( t 1 , t 2 ) ; H ρ , ω N ( T 2 ) ) 2 . \|U\|_{L^\infty([t_1,t_2];H^N_{\rho,\omega}(\mathbb{T}^2))}^2+\frac{\nu}{2}\|\nabla U\|_{L^2((t_1,t_2);H^N_{\rho,\omega}(\mathbb{T}^2))}^2\le2\|g\|_{H^N_{\rho,\omega}(\mathbb{T}^2)}^2+\frac{8}{\nu}\|F\|_{L^2((t_1,t_2);H^N_{\rho,\omega}(\mathbb{T}^2))}^2+10\|F^0\|_{L^1((t_1,t_2);H^N_{\rho,\omega}(\mathbb{T}^2))}^2. ∥ U ∥ L ∞ ([ t 1 , t 2 ] ; H ρ , ω N ( T 2 )) 2 + 2 ν ∥∇ U ∥ L 2 (( t 1 , t 2 ) ; H ρ , ω N ( T 2 )) 2 ≤ 2∥ g ∥ H ρ , ω N ( T 2 ) 2 + ν 8 ∥ F ∥ L 2 (( t 1 , t 2 ) ; H ρ , ω N ( T 2 )) 2 + 10∥ F 0 ∥ L 1 (( t 1 , t 2 ) ; H ρ , ω N ( T 2 )) 2 .
(b) Without forcing, for every t 1 ≤ t ′ ≤ t ≤ t 2 t_1\le t'\le t\le t_2 t 1 ≤ t ′ ≤ t ≤ t 2 ,
∥ U ( t ) ∥ H ρ , ω N ( T 2 ) 2 + ν ∥ ∇ U ∥ L 2 ( ( t ′ , t ) ; H ρ , ω N ( T 2 ) ) 2 ≤ ∥ U ( t ′ ) ∥ H ρ , ω N ( T 2 ) 2 . \|U(t)\|_{H^N_{\rho,\omega}(\mathbb{T}^2)}^2+\nu\|\nabla U\|_{L^2((t',t);H^N_{\rho,\omega}(\mathbb{T}^2))}^2\le\|U(t')\|_{H^N_{\rho,\omega}(\mathbb{T}^2)}^2. ∥ U ( t ) ∥ H ρ , ω N ( T 2 ) 2 + ν ∥∇ U ∥ L 2 (( t ′ , t ) ; H ρ , ω N ( T 2 )) 2 ≤ ∥ U ( t ′ ) ∥ H ρ , ω N ( T 2 ) 2 .
(c) Without forcing, the components satisfy, each with a separate supremum in time,
∑ v ∈ V 1 ω v ( ∥ U v ∥ L ∞ ( [ t 1 , t 2 ] ; H ρ N ( T 2 ) ) 2 + ν ∥ ∇ U v ∥ L 2 ( ( t 1 , t 2 ) ; H ρ N ( T 2 ) ) 2 ) ≤ 2 ∥ g ∥ H ρ , ω N ( T 2 ) 2 . \sum_{v\in V}\frac{1}{\omega_v}\left(\|U_v\|_{L^\infty([t_1,t_2];H^N_\rho(\mathbb{T}^2))}^2+\nu\|\nabla U_v\|_{L^2((t_1,t_2);H^N_\rho(\mathbb{T}^2))}^2\right)\le2\|g\|_{H^N_{\rho,\omega}(\mathbb{T}^2)}^2. v ∈ V ∑ ω v 1 ( ∥ U v ∥ L ∞ ([ t 1 , t 2 ] ; H ρ N ( T 2 )) 2 + ν ∥∇ U v ∥ L 2 (( t 1 , t 2 ) ; H ρ N ( T 2 )) 2 ) ≤ 2∥ g ∥ H ρ , ω N ( T 2 ) 2 .
By (i), the velocity gradient times the diffusion time ρ 2 / ν \rho^2/\nu ρ 2 / ν of the scale ρ \rho ρ is at most 1 8 \frac{1}{8} 8 1 . The sources cost at most an eighth of the dissipation 2 ν ∥ ∇ U ∥ H ρ , ω N ( T 2 ) 2 2\nu\|\nabla U\|_{H^N_{\rho,\omega}(\mathbb{T}^2)}^2 2 ν ∥∇ U ∥ H ρ , ω N ( T 2 ) 2 (Lemma 8.4), and the drift a quarter (Lemma 8.5).
Lemma 8.4 (The source term). Let N ≥ 0 N\ge0 N ≥ 0 , ρ > 0 \rho>0 ρ > 0 and ϑ ≥ 0 \vartheta\ge0 ϑ ≥ 0 , assume (ii) of Proposition 8.3 , and assume that all row and column sums of B N , ρ B^{N,\rho} B N , ρ are at most ϑ \vartheta ϑ . Then, for every t ∈ [ t 1 , t 2 ] t\in[t_1,t_2] t ∈ [ t 1 , t 2 ] and every U = ( U v ) v ∈ V U=(U_v)_{v\in V} U = ( U v ) v ∈ V with U v ∈ H N + 1 ( T 2 ) U_v\in H^{N+1}(\mathbb{T}^2) U v ∈ H N + 1 ( T 2 ) ,
∥ ( ∑ v ′ , σ , d A σ , d ( t ) ∇ d − 1 ( S σ ( t ) U v ′ ) ) v ∈ V ∥ H ρ , ω N ( T 2 ) ≤ ϑ ν ∥ ∇ U ∥ H ρ , ω N ( T 2 ) . \left\|\left(\sum_{v',\sigma,d} A_{\sigma,d}(t)\nabla^{d-1}(S_\sigma(t)U_{v'})\right)_{v\in V}\right\|_{H^N_{\rho,\omega}(\mathbb{T}^2)}\le\vartheta\nu\|\nabla U\|_{H^N_{\rho,\omega}(\mathbb{T}^2)}. v ′ , σ , d ∑ A σ , d ( t ) ∇ d − 1 ( S σ ( t ) U v ′ ) v ∈ V H ρ , ω N ( T 2 ) ≤ ϑ ν ∥∇ U ∥ H ρ , ω N ( T 2 ) . Under (iii) of Proposition 8.3 , this holds with ϑ = 1 8 \vartheta=\frac{1}{8} ϑ = 8 1 .
Proof. Fix σ : v ′ → v \sigma:v'\to v σ : v ′ → v and d d d ; we omit t t t . Since S σ S_\sigma S σ commutes with derivatives, its bound in L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) holds in H ρ N ( T 2 ) H^N_\rho(\mathbb{T}^2) H ρ N ( T 2 ) , and with Lemma 8.1(a),
∥ A σ , d ∇ d − 1 ( S σ U v ′ ) ∥ H ρ N ( T 2 ) ≤ c d − 2 r σ d − 2 ( ∑ j = 0 N ρ j j ! sup [ t 1 , t 2 ] × T 2 ∣ ∇ j A σ , d ∣ ) ∥ ∇ U v ′ ∥ H ρ N ( T 2 ) . \|A_{\sigma,d}\nabla^{d-1}(S_\sigma U_{v'})\|_{H^N_\rho(\mathbb{T}^2)}
\le\frac{c_{d-2}}{r_\sigma^{d-2}}\left(\sum_{j=0}^{N}\frac{\rho^j}{j!}\sup_{[t_1,t_2]\times\mathbb{T}^2}\lvert\nabla^j A_{\sigma,d}\rvert\right)\|\nabla U_{v'}\|_{H^N_\rho(\mathbb{T}^2)}. ∥ A σ , d ∇ d − 1 ( S σ U v ′ ) ∥ H ρ N ( T 2 ) ≤ r σ d − 2 c d − 2 ( j = 0 ∑ N j ! ρ j [ t 1 , t 2 ] × T 2 sup ∣ ∇ j A σ , d ∣ ) ∥∇ U v ′ ∥ H ρ N ( T 2 ) .
So the v v v -th component of the source term, divided by ω v 1 / 2 \omega_v^{1/2} ω v 1/2 , has norm in H ρ N ( T 2 ) H^N_\rho(\mathbb{T}^2) H ρ N ( T 2 ) at most ν ∑ v ′ B v v ′ N , ρ ω v ′ − 1 / 2 ∥ ∇ U v ′ ∥ H ρ N ( T 2 ) \nu\sum_{v'}B^{N,\rho}_{vv'}\omega_{v'}^{-1/2}\|\nabla U_{v'}\|_{H^N_\rho(\mathbb{T}^2)} ν ∑ v ′ B v v ′ N , ρ ω v ′ − 1/2 ∥∇ U v ′ ∥ H ρ N ( T 2 ) , and Schur’s test (Subsection 5.4) gives the lemma.
Lemma 8.5 (The drift commutator). Let N ≥ 1 N \ge1 N ≥ 1 , ρ > 0 \rho> 0 ρ > 0 and t ∈ [ t 1 , t 2 ] t \in[t_1,t_2] t ∈ [ t 1 , t 2 ] , and assume (i) of Proposition 8.3 at t t t . Then, for every smooth g g g on T 2 \mathbb{T}^2 T 2 , with Φ \Phi Φ taken at the time t t t ,
∑ j = 1 N ρ 2 j ( j ! ) 2 ⟨ ∇ j ( ∇ ⊥ Φ ⋅ ∇ g ) − ∇ ⊥ Φ ⋅ ∇ ( ∇ j g ) , ∇ j g ⟩ L 2 ( T 2 ) ≤ ν 4 ∥ ∇ g ∥ H ρ N ( T 2 ) 2 . \sum_{j=1}^{N} \frac{\rho^{2j}}{(j!)^2} \left\langle\nabla^j(\nabla^\perp\Phi\cdot\nabla g)-\nabla^\perp\Phi\cdot\nabla(\nabla^j g),\nabla^j g\right\rangle_{L^2(\mathbb{T}^2)} \le\frac{\nu}{4}\|\nabla g\|_{H^N_\rho(\mathbb{T}^2)}^2. j = 1 ∑ N ( j ! ) 2 ρ 2 j ⟨ ∇ j ( ∇ ⊥ Φ ⋅ ∇ g ) − ∇ ⊥ Φ ⋅ ∇ ( ∇ j g ) , ∇ j g ⟩ L 2 ( T 2 ) ≤ 4 ν ∥∇ g ∥ H ρ N ( T 2 ) 2 .
Proof. By the Leibniz rule, as in the proof of Lemma 8.1(a), the commutator is the sum over 1 ≤ i ≤ j 1 \le i \le j 1 ≤ i ≤ j of ( j i ) \binom{j}{i} ( i j ) arrays in which i i i derivatives fall on ∇ ⊥ Φ \nabla^\perp\Phi ∇ ⊥ Φ and j − i j-i j − i on ∇ g \nabla g ∇ g , each of norm at most ∣ ∇ i ∇ ⊥ Φ ∣ ∣ ∇ j − i + 1 g ∣ |\nabla^i\nabla^\perp\Phi|\,|\nabla^{j-i+1}g| ∣ ∇ i ∇ ⊥ Φ∣ ∣ ∇ j − i + 1 g ∣ at every point. Since ( j i ) ρ 2 j / ( j ! ) 2 \binom{j}{i}\rho^{2j}/(j!)^2 ( i j ) ρ 2 j / ( j ! ) 2 is the product of ρ i + 1 / ( i ! j ) \rho^{i+1}/(i!j) ρ i + 1 / ( i ! j ) , ρ j − i / ( j − i ) ! \rho^{j-i}/(j-i)! ρ j − i / ( j − i )! and ρ j − 1 / ( j − 1 ) ! \rho^{j-1}/(j-1)! ρ j − 1 / ( j − 1 )! , the hypothesis and 1 / j ≤ 1 1/j \le1 1/ j ≤ 1 bound the contribution of i i i to the term of j j j by
2 − i − 2 ν ⋅ ρ j − i ( j − i ) ! ∥ ∇ j − i + 1 g ∥ L 2 ( T 2 ) ⋅ ρ j − 1 ( j − 1 ) ! ∥ ∇ j g ∥ L 2 ( T 2 ) . 2^{-i-2}\nu\cdot\frac{\rho^{j-i}}{(j-i)!}\|\nabla^{j-i+1}g\|_{L^2(\mathbb{T}^2)}\cdot\frac{\rho^{j-1}}{(j-1)!}\|\nabla^j g\|_{L^2(\mathbb{T}^2)}. 2 − i − 2 ν ⋅ ( j − i )! ρ j − i ∥ ∇ j − i + 1 g ∥ L 2 ( T 2 ) ⋅ ( j − 1 )! ρ j − 1 ∥ ∇ j g ∥ L 2 ( T 2 ) .
For each i i i , the sum over j j j of these products is at most 2 − i − 2 ν ∥ ∇ g ∥ H ρ N ( T 2 ) 2 2^{-i-2}\nu\|\nabla g\|_{H^N_\rho(\mathbb{T}^2)}^2 2 − i − 2 ν ∥∇ g ∥ H ρ N ( T 2 ) 2 , by the Cauchy–Schwarz inequality, and ∑ i ≥ 1 2 − i − 2 = 1 4 \sum_{i\ge1}2^{-i-2}=\frac14 ∑ i ≥ 1 2 − i − 2 = 4 1 .
Proof of Proposition 8.3. Step 1: The energy inequalities. Let U U U be a solution without forcing that is smooth in x x x , or one of the Galerkin system of Step 3 (the system projected onto trigonometric polynomials). We apply ∇ j \nabla^j ∇ j , j ≤ N j \le N j ≤ N , to the equation of U v U_v U v and take the inner product in L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) with ( ρ 2 j / ( j ! ) 2 ) ∇ j U v (\rho^{2j}/(j!)^2)\nabla^jU_v ( ρ 2 j / ( j ! ) 2 ) ∇ j U v : the transport term ⟨ ∇ ⊥ Φ ⋅ ∇ ( ∇ j U v ) , ∇ j U v ⟩ L 2 ( T 2 ) \langle\nabla^\perp\Phi\cdot\nabla(\nabla^jU_v),\nabla^jU_v\rangle_{L^2(\mathbb{T}^2)} ⟨ ∇ ⊥ Φ ⋅ ∇ ( ∇ j U v ) , ∇ j U v ⟩ L 2 ( T 2 ) vanishes because ∇ ⊥ Φ \nabla^\perp\Phi ∇ ⊥ Φ is divergence free, the commutator is bounded by Lemma 8.5 applied to U v ( t ) U_v(t) U v ( t ) , and the source term pairs with ∇ U v \nabla U_v ∇ U v after an integration by parts. Summing over j j j , we obtain
d d t ∥ U v ( t ) ∥ H ρ N ( T 2 ) 2 + 3 2 ν ∥ ∇ U v ( t ) ∥ H ρ N ( T 2 ) 2 ≤ 2 ⟨ ∑ v ′ , σ , d A σ , d ∇ d − 1 ( S σ U v ′ ) , ∇ U v ⟩ H ρ N ( T 2 ) , (69) \frac{\mathrm{d}}{\mathrm{d}t}\|U_v(t)\|_{H^N_\rho(\mathbb{T}^2)}^2+\frac{3}{2}\nu\|\nabla U_v(t)\|_{H^N_\rho(\mathbb{T}^2)}^2\le2\left\langle\sum_{v',\sigma,d}A_{\sigma,d}\nabla^{d-1}(S_{\sigma}U_{v'}),\nabla U_v\right\rangle_{H^N_\rho(\mathbb{T}^2)}, \tag*{(69)} d t d ∥ U v ( t ) ∥ H ρ N ( T 2 ) 2 + 2 3 ν ∥∇ U v ( t ) ∥ H ρ N ( T 2 ) 2 ≤ 2 ⟨ v ′ , σ , d ∑ A σ , d ∇ d − 1 ( S σ U v ′ ) , ∇ U v ⟩ H ρ N ( T 2 ) , (69)
with the inner product of the norm (36) . Divided by ω v \omega_v ω v and summed over v v v , the right sides are at most ν 4 ∥ ∇ U ( t ) ∥ H ρ , ω N ( T 2 ) 2 \frac{\nu}{4}\|\nabla U(t)\|_{H^N_{\rho,\omega}(\mathbb{T}^2)}^2 4 ν ∥∇ U ( t ) ∥ H ρ , ω N ( T 2 ) 2 , by the Cauchy–Schwarz inequality and Lemma 8.4, so
d d t ∥ U ( t ) ∥ H ρ , ω N ( T 2 ) 2 + 5 4 ν ∥ ∇ U ( t ) ∥ H ρ , ω N ( T 2 ) 2 ≤ 0. (70) \frac{\mathrm{d}}{\mathrm{d}t}\|U(t)\|_{H^N_{\rho,\omega}(\mathbb{T}^2)}^2+\frac{5}{4}\nu\|\nabla U(t)\|_{H^N_{\rho,\omega}(\mathbb{T}^2)}^2\le0. \tag*{(70)} d t d ∥ U ( t ) ∥ H ρ , ω N ( T 2 ) 2 + 4 5 ν ∥∇ U ( t ) ∥ H ρ , ω N ( T 2 ) 2 ≤ 0. (70)
Integrating (70) over [ t ′ , t ] [t',t] [ t ′ , t ] gives (b). For (c), we integrate (69) over [ t 1 , t ] [t_1,t] [ t 1 , t ] , drop its dissipation, take the supremum in t t t , divide by ω v \omega_v ω v and sum. By the bound of the right sides above, and then by (70) integrated over [ t 1 , t 2 ] [t_1,t_2] [ t 1 , t 2 ] , the left side of (c) is at most
∥ g ∥ H ρ , ω N ( T 2 ) 2 + 5 4 ν ∥ ∇ U ∥ L 2 ( ( t 1 , t 2 ) ; H ρ , ω N ( T 2 ) ) 2 ≤ 2 ∥ g ∥ H ρ , ω N ( T 2 ) 2 . \|g\|_{H^N_{\rho,\omega}(\mathbb{T}^2)}^2+\frac{5}{4}\nu\|\nabla U\|_{L^2((t_1,t_2);H^N_{\rho,\omega}(\mathbb{T}^2))}^2\le2\|g\|_{H^N_{\rho,\omega}(\mathbb{T}^2)}^2. ∥ g ∥ H ρ , ω N ( T 2 ) 2 + 4 5 ν ∥∇ U ∥ L 2 (( t 1 , t 2 ) ; H ρ , ω N ( T 2 )) 2 ≤ 2∥ g ∥ H ρ , ω N ( T 2 ) 2 .
Step 2: The forced estimate. With forcing, the computation of Step 1, summed over v v v , has the further terms − 2 ⟨ F , ∇ U ⟩ H ρ , ω N ( T 2 ) -2\langle F,\nabla U\rangle_{H^N_{\rho,\omega}(\mathbb{T}^2)} − 2 ⟨ F , ∇ U ⟩ H ρ , ω N ( T 2 ) and 2 ⟨ F 0 , U ⟩ H ρ , ω N ( T 2 ) 2\langle F^0,U\rangle_{H^N_{\rho,\omega}(\mathbb{T}^2)} 2 ⟨ F 0 , U ⟩ H ρ , ω N ( T 2 ) , with the inner product of the norm (65) . Bounding the first by ν 4 ∥ ∇ U ∥ H ρ , ω N ( T 2 ) 2 + 4 ν ∥ F ∥ H ρ , ω N ( T 2 ) 2 \frac{\nu}{4}\|\nabla U\|_{H^N_{\rho,\omega}(\mathbb{T}^2)}^2+\frac{4}{\nu}\|F\|_{H^N_{\rho,\omega}(\mathbb{T}^2)}^2 4 ν ∥∇ U ∥ H ρ , ω N ( T 2 ) 2 + ν 4 ∥ F ∥ H ρ , ω N ( T 2 ) 2 and integrating gives, for every t ∈ [ t 1 , t 2 ] t \in[t_1,t_2] t ∈ [ t 1 , t 2 ] ,
∥ U ( t ) ∥ H ρ , ω N ( T 2 ) 2 + ν ∥ ∇ U ∥ L 2 ( ( t 1 , t ) ; H ρ , ω N ( T 2 ) ) 2 ≤ ∥ g ∥ H ρ , ω N ( T 2 ) 2 + 4 ν ∥ F ∥ L 2 ( ( t 1 , t 2 ) ; H ρ , ω N ( T 2 ) ) 2 + 2 ∥ U ∥ L ∞ ( [ t 1 , t 2 ] ; H ρ , ω N ( T 2 ) ) ∥ F 0 ∥ L 1 ( ( t 1 , t 2 ) ; H ρ , ω N ( T 2 ) ) . (71) \|U(t)\|_{H^N_{\rho,\omega}(\mathbb{T}^2)}^2+\nu\|\nabla U\|_{L^2((t_1,t);H^N_{\rho,\omega}(\mathbb{T}^2))}^2\le\|g\|_{H^N_{\rho,\omega}(\mathbb{T}^2)}^2+\frac{4}{\nu}\|F\|_{L^2((t_1,t_2);H^N_{\rho,\omega}(\mathbb{T}^2))}^2+2\|U\|_{L^\infty([t_1,t_2];H^N_{\rho,\omega}(\mathbb{T}^2))}\|F^0\|_{L^1((t_1,t_2);H^N_{\rho,\omega}(\mathbb{T}^2))}. \tag*{(71)} ∥ U ( t ) ∥ H ρ , ω N ( T 2 ) 2 + ν ∥∇ U ∥ L 2 (( t 1 , t ) ; H ρ , ω N ( T 2 )) 2 ≤ ∥ g ∥ H ρ , ω N ( T 2 ) 2 + ν 4 ∥ F ∥ L 2 (( t 1 , t 2 ) ; H ρ , ω N ( T 2 )) 2 + 2∥ U ∥ L ∞ ([ t 1 , t 2 ] ; H ρ , ω N ( T 2 )) ∥ F 0 ∥ L 1 (( t 1 , t 2 ) ; H ρ , ω N ( T 2 )) . (71)
Young’s inequality absorbs the last term, in two stages. On the right side of (71) , let Y Y Y be the sum of the first two terms and X X X and Z Z Z the two norms in the last, and let D : = ν ∥ ∇ U ∥ L 2 ( ( t 1 , t 2 ) ; H ρ , ω N ( T 2 ) ) 2 D := \nu\|\nabla U\|_{L^2((t_1,t_2);H^N_{\rho,\omega}(\mathbb{T}^2))}^2 D := ν ∥∇ U ∥ L 2 (( t 1 , t 2 ) ; H ρ , ω N ( T 2 )) 2 . The supremum in t t t of (71) gives the first line below, and this supremum plus half of (71) at t = t 2 t=t_2 t = t 2 gives the second, which is (a), with the first line in its last step:
X 2 ≤ Y + 2 X Z ≤ Y + 1 2 X 2 + 2 Z 2 , so X 2 ≤ 2 Y + 4 Z 2 , X^2 \le Y+2XZ \le Y+\frac{1}{2}X^2+2Z^2,\qquad\text{so}\qquad X^2\le2Y+4Z^2, X 2 ≤ Y + 2 X Z ≤ Y + 2 1 X 2 + 2 Z 2 , so X 2 ≤ 2 Y + 4 Z 2 ,
X 2 + 1 2 D ≤ 3 2 Y + 3 X Z ≤ 3 2 Y + 1 4 X 2 + 9 Z 2 ≤ 2 Y + 10 Z 2 . X^2+\frac{1}{2}D\le\frac{3}{2}Y+3XZ\le\frac{3}{2}Y+\frac{1}{4}X^2+9Z^2\le2Y+10Z^2. X 2 + 2 1 D ≤ 2 3 Y + 3 X Z ≤ 2 3 Y + 4 1 X 2 + 9 Z 2 ≤ 2 Y + 10 Z 2 .
Step 3: Existence, uniqueness and regularity. By the standard Galerkin argument, with the Fourier truncations P n P_n P n , which are self-adjoint and commute with derivatives, Steps 1 and 2 hold for the Galerkin approximations uniformly in n n n , and a subsequence converges weakly-∗ * ∗ in L ∞ ( [ t 1 , t 2 ] ; H N ( T 2 ) V ) ∩ L 2 ( ( t 1 , t 2 ) ; H N + 1 ( T 2 ) V ) L^\infty([t_1,t_2];H^N(\mathbb{T}^2)^V)\cap L^2((t_1,t_2);H^{N+1}(\mathbb{T}^2)^V) L ∞ ([ t 1 , t 2 ] ; H N ( T 2 ) V ) ∩ L 2 (( t 1 , t 2 ) ; H N + 1 ( T 2 ) V ) to a solution U U U . For j ≤ N j\le N j ≤ N , ∂ t ∇ j U v \partial_t\nabla^jU_v ∂ t ∇ j U v lies in L 2 ( ( t 1 , t 2 ) ; H − 1 ( T 2 ) ) + L 1 ( ( t 1 , t 2 ) ; L 2 ( T 2 ) ) L^2((t_1,t_2);H^{-1}(\mathbb{T}^2))+L^1((t_1,t_2);L^2(\mathbb{T}^2)) L 2 (( t 1 , t 2 ) ; H − 1 ( T 2 )) + L 1 (( t 1 , t 2 ) ; L 2 ( T 2 )) , by (i), Lemma 8.4 and the classes of F F F and F 0 F^0 F 0 ; testing with Fourier truncations then gives the energy identity for ∇ j U v \nabla^jU_v ∇ j U v , hence U ∈ C ( [ t 1 , t 2 ] ; H N ( T 2 ) V ) U\in C([t_1,t_2];H^N(\mathbb{T}^2)^V) U ∈ C ([ t 1 , t 2 ] ; H N ( T 2 ) V ) and Steps 1 and 2 at order N N N (Lemma 8.5 extends to H N + 1 ( T 2 ) H^{N+1}(\mathbb{T}^2) H N + 1 ( T 2 ) by density), and so (a)–(c). At order 0 the same lines apply to every solution, with only the boundedness of ∇ ⊥ Φ \nabla^\perp\Phi ∇ ⊥ Φ , so the difference of two solutions vanishes by Step 1 with N = 0 N=0 N = 0 , since (i) is then void and B 0 , ρ ≤ B N , ρ B^{0,\rho}\le B^{N,\rho} B 0 , ρ ≤ B N , ρ .
By the uniqueness at order 0 (Step 3), the restriction of a solution to [ t ′ , t 2 ] [t',t_2] [ t ′ , t 2 ] is the solution from U ( t ′ ) U(t') U ( t ′ ) with the restricted forcing, as it satisfies (67) there and Proposition 8.3 applies on [ t ′ , t 2 ] [t',t_2] [ t ′ , t 2 ] .
Proof of Proposition 6.2. Take N = 0 N=0 N = 0 in Proposition 8.3, without forcing and with real components: (i) is void, (ii) is the boundedness of the coefficients, B 0 , ρ = B B^{0,\rho}=B B 0 , ρ = B , and (b) is the energy inequality.
Reduction and proofs of the main results
Proposition 8.6 (Systems without smoothing). In the setting of (66) , let ρ > 0 \rho>0 ρ > 0 , let N v ≥ 0 N_v\ge0 N v ≥ 0 , v ∈ V v\in V v ∈ V , be integers, and assume (i) of Proposition 8.3 at the order max v N v \max_vN_v max v N v . Let A v v ′ , d A_{vv',d} A v v ′ , d , for v , v ′ ∈ V v,v'\in V v , v ′ ∈ V and finitely many d ≥ 2 d\ge2 d ≥ 2 , be coefficients as in Proposition 6.2 and in (ii) of Proposition 8.3 at the order N v N_v N v , vanishing unless N v ′ ≥ N v + d − 2 N_{v'}\ge N_v+d-2 N v ′ ≥ N v + d − 2 , such that all row and column sums of the matrix with entries
( ω v ′ ω v ) 1 / 2 ∑ d ( N v + d − 2 ) ! N v ! ν d − 2 ∑ j = 0 N v ρ j j ! sup [ t 1 , t 2 ] × T 2 ∣ ∇ j A v v ′ , d ∣ , v , v ′ ∈ V , \left(\frac{\omega_{v'}}{\omega_v}\right)^{1/2}\sum_d\frac{(N_v+d-2)!}{N_v!\,\nu^{d-2}}\sum_{j=0}^{N_v}\frac{\rho^j}{j!}\sup_{[t_1,t_2]\times\mathbb{T}^2}\left|\nabla^jA_{vv',d}\right|,\qquad v,v'\in V, ( ω v ω v ′ ) 1/2 d ∑ N v ! ν d − 2 ( N v + d − 2 )! j = 0 ∑ N v j ! ρ j [ t 1 , t 2 ] × T 2 sup ∇ j A v v ′ , d , v , v ′ ∈ V ,
are at most 1 2 \frac{1}{2} 2 1 . Then, for every g g g , F F F , F 0 F^0 F 0 such that, for every v ∈ V v\in V v ∈ V , g v ∈ H N v ( T 2 ) g_v\in H^{N_v}(\mathbb{T}^2) g v ∈ H N v ( T 2 ) , F v ∈ L 2 ( ( t 1 , t 2 ) ; H N v ( T 2 ) ) F_v\in L^2((t_1,t_2);H^{N_v}(\mathbb{T}^2)) F v ∈ L 2 (( t 1 , t 2 ) ; H N v ( T 2 )) and F v 0 ∈ L 1 ( ( t 1 , t 2 ) ; H N v ( T 2 ) ) F_v^0\in L^1((t_1,t_2);H^{N_v}(\mathbb{T}^2)) F v 0 ∈ L 1 (( t 1 , t 2 ) ; H N v ( T 2 )) , the problem, for v ∈ V v\in V v ∈ V ,
{ ∂ t U v + ∇ ⊥ Φ ⋅ ∇ U v − ν Δ U v = ∇ ⋅ ( ∑ v ′ ∈ V ∑ d A v v ′ , d ∇ d − 1 U v ′ + F v ) + F v 0 in ( t 1 , t 2 ) × T 2 , U v ( t 1 , ⋅ ) = g v on T 2 , \begin{cases}
\partial_tU_v+\nabla^\perp\Phi\cdot\nabla U_v-\nu\Delta U_v
=\nabla\cdot\left(\sum_{v'\in V}\sum_d A_{vv',d}\nabla^{d-1}U_{v'}+F_v\right)+F_v^0
& \text{in }(t_1,t_2)\times\mathbb{T}^2,\\
U_v(t_1,\cdot)=g_v & \text{on }\mathbb{T}^2,
\end{cases} { ∂ t U v + ∇ ⊥ Φ ⋅ ∇ U v − ν Δ U v = ∇ ⋅ ( ∑ v ′ ∈ V ∑ d A v v ′ , d ∇ d − 1 U v ′ + F v ) + F v 0 U v ( t 1 , ⋅ ) = g v in ( t 1 , t 2 ) × T 2 , on T 2 ,
has a unique solution U U U (as in (67) ) with U v ∈ C ( [ t 1 , t 2 ] ; H N v ( T 2 ) ) ∩ L 2 ( ( t 1 , t 2 ) ; H N v + 1 ( T 2 ) ) U_v\in C([t_1,t_2];H^{N_v}(\mathbb{T}^2))\cap L^2((t_1,t_2);H^{N_v+1}(\mathbb{T}^2)) U v ∈ C ([ t 1 , t 2 ] ; H N v ( T 2 )) ∩ L 2 (( t 1 , t 2 ) ; H N v + 1 ( T 2 )) , and
∥ U ∥ L ∞ ( [ t 1 , t 2 ] ; H ρ ( N v ) ( T 2 ) ) 2 + ν 2 ∥ ∇ U ∥ L 2 ( ( t 1 , t 2 ) ; H ρ ( N v ) ( T 2 ) ) 2 ≤ 2 ∥ g ∥ H ρ ( N v ) ( T 2 ) 2 + 8 ν ∥ F ∥ L 2 ( ( t 1 , t 2 ) ; H ρ ( N v ) ( T 2 ) ) 2 + 10 ∥ F 0 ∥ L 1 ( ( t 1 , t 2 ) ; H ρ ( N v ) ( T 2 ) ) 2 . \|U\|_{L^\infty([t_1,t_2];H_\rho^{(N_v)}(\mathbb{T}^2))}^2+\frac{\nu}{2}\|\nabla U\|_{L^2((t_1,t_2);H_\rho^{(N_v)}(\mathbb{T}^2))}^2
\le2\|g\|_{H_\rho^{(N_v)}(\mathbb{T}^2)}^2+\frac{8}{\nu}\|F\|_{L^2((t_1,t_2);H_\rho^{(N_v)}(\mathbb{T}^2))}^2+10\|F^0\|_{L^1((t_1,t_2);H_\rho^{(N_v)}(\mathbb{T}^2))}^2. ∥ U ∥ L ∞ ([ t 1 , t 2 ] ; H ρ ( N v ) ( T 2 )) 2 + 2 ν ∥∇ U ∥ L 2 (( t 1 , t 2 ) ; H ρ ( N v ) ( T 2 )) 2 ≤ 2∥ g ∥ H ρ ( N v ) ( T 2 ) 2 + ν 8 ∥ F ∥ L 2 (( t 1 , t 2 ) ; H ρ ( N v ) ( T 2 )) 2 + 10∥ F 0 ∥ L 1 (( t 1 , t 2 ) ; H ρ ( N v ) ( T 2 )) 2 .
The system (66) with every S σ S_\sigma S σ replaced by the identity is of this form, with A v v ′ , d = ∑ σ : v ′ → v A σ , d A_{vv',d}=\sum_{\sigma:v'\to v}A_{\sigma,d} A v v ′ , d = ∑ σ : v ′ → v A σ , d .
Proof. The proof is that of Proposition 8.3, for each U v U_v U v at the order N v N_v N v and with uniqueness by (a) for differences, except for the source term: by Lemma 8.1, with i = d − 2 i=d-2 i = d − 2 ,
∥ A v v ′ , d ∇ d − 1 U v ′ ∥ H ρ N v ( T 2 ) ≤ ( ∑ j = 0 N v ρ j j ! sup [ t 1 , t 2 ] × T 2 ∣ ∇ j A v v ′ , d ∣ ) ( N v + d − 2 ) ! N v ! ρ d − 2 ∥ ∇ U v ′ ∥ H ρ N v + d − 2 ( T 2 ) , \|A_{vv',d}\nabla^{d-1}U_{v'}\|_{H_\rho^{N_v}(\mathbb{T}^2)}
\le\left(\sum_{j=0}^{N_v}\frac{\rho^j}{j!}\sup_{[t_1,t_2]\times\mathbb{T}^2}|\nabla^jA_{vv',d}|\right)\frac{(N_v+d-2)!}{N_v!\,\rho^{d-2}}\|\nabla U_{v'}\|_{H_\rho^{N_v+d-2}(\mathbb{T}^2)}, ∥ A v v ′ , d ∇ d − 1 U v ′ ∥ H ρ N v ( T 2 ) ≤ ( j = 0 ∑ N v j ! ρ j [ t 1 , t 2 ] × T 2 sup ∣ ∇ j A v v ′ , d ∣ ) N v ! ρ d − 2 ( N v + d − 2 )! ∥∇ U v ′ ∥ H ρ N v + d − 2 ( T 2 ) ,
and, when A v v ′ , d ≠ 0 A_{vv',d}\ne0 A v v ′ , d = 0 , ∥ ∇ U v ′ ∥ H ρ N v + d − 2 ( T 2 ) ≤ ∥ ∇ U v ′ ∥ H ρ N v ′ ( T 2 ) \|\nabla U_{v'}\|_{H_\rho^{N_v+d-2}(\mathbb{T}^2)}\le\|\nabla U_{v'}\|_{H_\rho^{N_{v'}}(\mathbb{T}^2)} ∥∇ U v ′ ∥ H ρ N v + d − 2 ( T 2 ) ≤ ∥∇ U v ′ ∥ H ρ N v ′ ( T 2 ) ; Schur’s test concludes.
Reduction and proofs of the main results
In this section we prove Lemma 4.3 and Proposition 4.2, deduce Theorem A from them and Proposition 4.1, state and prove Theorem C and Corollary D, and prove Theorem B.
Conventions. Throughout, C C C and c c c denote positive constants depending only on α , s , q , ζ , Λ \alpha,s,q,\zeta,\Lambda α , s , q , ζ , Λ and on the quantities named in each statement. We write D r ( x ) : = { y ∈ R 2 : ∣ y − x ∣ < r } D_r(x):=\{y\in\mathbb{R}^2:|y-x|<r\} D r ( x ) := { y ∈ R 2 : ∣ y − x ∣ < r } and D r : = D r ( 0 ) D_r:=D_r(0) D r := D r ( 0 ) ; for E ⊂ R 2 E\subset\mathbb{R}^2 E ⊂ R 2 , ∥ g ∥ C 0 , s ( E ) \|g\|_{C^{0,s}(E)} ∥ g ∥ C 0 , s ( E ) and [ g ] C 0 , s ( E ) [g]_{C^{0,s}(E)} [ g ] C 0 , s ( E ) are defined by (22) , with E E E in place of T 2 \mathbb{T}^2 T 2 and ∣ x − y ∣ |x-y| ∣ x − y ∣ in place of ∣ x − y ∣ T 2 |x-y|_{\mathbb{T}^2} ∣ x − y ∣ T 2 . We fix C 0 ( α , s , q , ζ , Λ ) ≥ 1 C_0(\alpha,s,q,\zeta,\Lambda)\geq1 C 0 ( α , s , q , ζ , Λ ) ≥ 1 such that, for every m ≥ 0 m\geq0 m ≥ 0 ,
∥ ϕ − ϕ m ∥ L ∞ ( R × T 2 ) ≤ C 0 ε m β + 1 , ∥ b − b m ∥ L ∞ ( R × T 2 ) ≤ C 0 ε m α + 1 , ∥ ∇ 2 ϕ m ∥ L ∞ ( R × T 2 ) ≤ C 0 ε m β − 2 . (72) \|\phi-\phi_m\|_{L^\infty(\mathbb{R}\times\mathbb{T}^2)}\leq C_0\varepsilon_m^{\beta+1},\qquad\|\mathbf{b}-\mathbf{b}_m\|_{L^\infty(\mathbb{R}\times\mathbb{T}^2)}\leq C_0\varepsilon_m^{\alpha+1},\qquad\|\nabla^2\phi_m\|_{L^\infty(\mathbb{R}\times\mathbb{T}^2)}\leq C_0\varepsilon_m^{\beta-2}. \tag*{(72)} ∥ ϕ − ϕ m ∥ L ∞ ( R × T 2 ) ≤ C 0 ε m β + 1 , ∥ b − b m ∥ L ∞ ( R × T 2 ) ≤ C 0 ε m α + 1 , ∥ ∇ 2 ϕ m ∥ L ∞ ( R × T 2 ) ≤ C 0 ε m β − 2 . (72)
by (16) and Proposition 7.3(ii). A disc of radius at most 1 2 \frac{1}{2} 2 1 meets at most four translates of the unit cell, so, for every periodic g g g , y ∈ R 2 y\in\mathbb{R}^2 y ∈ R 2 and r ≤ 1 2 r\leq\frac{1}{2} r ≤ 2 1 ,
∫ D r ( y ) g ( x ) 2 d x ≤ 4 ∥ g ∥ L 2 ( T 2 ) 2 . \int_{D_r(y)}g(x)^2\,dx\leq4\|g\|_{L^2(\mathbb{T}^2)}^2. ∫ D r ( y ) g ( x ) 2 d x ≤ 4∥ g ∥ L 2 ( T 2 ) 2 .
Weak solutions are those of Section 3, and a weak solution on ( t i n , ∞ ) (t_{\mathrm{in}},\infty) ( t in , ∞ ) is one on every ( t i n , t i n + T ) (t_{\mathrm{in}},t_{\mathrm{in}}+T) ( t in , t in + T ) . For a bounded, measurable, Z 2 \mathbb{Z}^2 Z 2 -periodic and divergence-free velocity field and κ > 0 \kappa>0 κ > 0 , the initial-value problem (23) has exactly one weak solution on every ( t i n , t i n + T ) (t_{\mathrm{in}},t_{\mathrm{in}}+T) ( t in , t in + T ) ; its restriction to ( t i n ′ , t i n + T ) (t'_{\mathrm{in}},t_{\mathrm{in}}+T) ( t in ′ , t in + T ) is the weak solution with datum θ ( t i n ′ , ⋅ ) \theta(t'_{\mathrm{in}},\cdot) θ ( t in ′ , ⋅ ) at time t i n ′ t'_{\mathrm{in}} t in ′ , and, for every t ∈ ( t i n , t i n + T ) t\in(t_{\mathrm{in}},t_{\mathrm{in}}+T) t ∈ ( t in , t in + T ) ,
∥ θ ( t , ⋅ ) ∥ L 2 ( T 2 ) 2 + 2 κ ∥ ∇ θ ∥ L 2 ( ( t i n , t ) × T 2 ) 2 = ∥ θ i n ∥ L 2 ( T 2 ) 2 . (73) \|\theta(t,\cdot)\|_{L^2(\mathbb{T}^2)}^2+2\kappa\|\nabla\theta\|_{L^2((t_{\mathrm{in}},t)\times\mathbb{T}^2)}^2=\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)}^2. \tag*{(73)} ∥ θ ( t , ⋅ ) ∥ L 2 ( T 2 ) 2 + 2 κ ∥∇ θ ∥ L 2 (( t in , t ) × T 2 ) 2 = ∥ θ in ∥ L 2 ( T 2 ) 2 . (73)
These facts are standard (Galerkin approximation, Lions–Magenes), and so is the first assertion below; the second follows from it, since the balls of C 0 , s ( T 2 ) C^{0,s}(\mathbb{T}^2) C 0 , s ( T 2 ) are weakly closed in L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) : for every [ t 1 , t 2 ] ⊂ [ t i n , t i n + T ] [t_1,t_2]\subset[t_{\mathrm{in}},t_{\mathrm{in}}+T] [ t 1 , t 2 ] ⊂ [ t in , t in + T ] and t ∈ [ t 1 , t 2 ] t\in[t_1,t_2] t ∈ [ t 1 , t 2 ] ,
θ ∈ C ( [ t i n , t i n + T ] ; L 2 ( T 2 ) ) , ∥ θ ( t , ⋅ ) ∥ C 0 , s ( T 2 ) ≤ ∥ θ ∥ L ∞ ( [ t 1 , t 2 ] ; C 0 , s ( T 2 ) ) . (74) \theta\in C([t_{\mathrm{in}},t_{\mathrm{in}}+T];L^2(\mathbb{T}^2)),\qquad\|\theta(t,\cdot)\|_{C^{0,s}(\mathbb{T}^2)}\leq\|\theta\|_{L^\infty([t_1,t_2];C^{0,s}(\mathbb{T}^2))}. \tag*{(74)} θ ∈ C ([ t in , t in + T ] ; L 2 ( T 2 )) , ∥ θ ( t , ⋅ ) ∥ C 0 , s ( T 2 ) ≤ ∥ θ ∥ L ∞ ([ t 1 , t 2 ] ; C 0 , s ( T 2 )) . (74)
So the limit in L l o c 2 L^2_{\mathrm{loc}} L loc 2 of θ ( t ′ , ⋅ ) \theta(t',\cdot) θ ( t ′ , ⋅ ) as t ′ ↑ t t'\uparrow t t ′ ↑ t , when it exists, is θ ( t , ⋅ ) \theta(t,\cdot) θ ( t , ⋅ ) .
For an open set O ⊂ R × R 2 \mathcal{O}\subset\mathbb{R}\times\mathbb{R}^2 O ⊂ R × R 2 with time sections O t : = { x : ( t , x ) ∈ O } \mathcal{O}_t:=\{x:(t,x)\in\mathcal{O}\} O t := { x : ( t , x ) ∈ O } , a local weak solution of ∂ t θ + b ⋅ ∇ θ − κ Δ θ = 0 \partial_t\theta+\mathbf{b}\cdot\nabla\theta-\kappa\Delta\theta=0 ∂ t θ + b ⋅ ∇ θ − κ Δ θ = 0 in O \mathcal{O} O is a function θ ∈ L l o c 2 ( O ) \theta\in L^2_{\mathrm{loc}}(\mathcal{O}) θ ∈ L loc 2 ( O ) such that, for almost every t t t , θ ( t , ⋅ ) \theta(t,\cdot) θ ( t , ⋅ ) has a weak gradient on O t \mathcal{O}_t O t , with ∇ θ ∈ L l o c 2 ( O ) \nabla\theta\in L^2_{\mathrm{loc}}(\mathcal{O}) ∇ θ ∈ L loc 2 ( O ) , and, for every g ∈ C c ∞ ( O ) g\in C^\infty_c(\mathcal{O}) g ∈ C c ∞ ( O ) ,
∫ O ( − θ ∂ t g + ( b ⋅ ∇ θ ) g + κ ∇ θ ⋅ ∇ g ) d t d x = 0. (75) \int_{\mathcal{O}}\left(-\theta\,\partial_tg+(\mathbf{b}\cdot\nabla\theta)g+\kappa\nabla\theta\cdot\nabla g\right)\,dt\,dx=0. \tag*{(75)} ∫ O ( − θ ∂ t g + ( b ⋅ ∇ θ ) g + κ ∇ θ ⋅ ∇ g ) d t d x = 0. (75)
A weak solution on ( t i n , t i n + T ) (t_{\mathrm{in}},t_{\mathrm{in}}+T) ( t in , t in + T ) is a local weak solution in ( t i n , t i n + T ) × R 2 (t_{\mathrm{in}},t_{\mathrm{in}}+T)\times\mathbb{R}^2 ( t in , t in + T ) × R 2 .
Finally, by Lemma 7.4 and (15) , for every m ≥ 0 m\geq0 m ≥ 0 and ∣ t − t f ∣ ≤ 8 h m + 1 |t-t_f|\leq8h_{m+1} ∣ t − t f ∣ ≤ 8 h m + 1 , the map y ↦ X m ( t , y , t f ) y\mapsto X_m(t,y,t_f) y ↦ X m ( t , y , t f ) of (13) is a smooth, volume-preserving diffeomorphism of R 2 \mathbb{R}^2 R 2 commuting with the translations by Z 2 \mathbb{Z}^2 Z 2 , with inverse x ↦ X m ( t f , x , t ) x\mapsto X_m(t_f,x,t) x ↦ X m ( t f , x , t ) , and both maps are 2-Lipschitz, since
∣ ∇ y X m ( t , y , t f ) − I ∣ ≤ 1 2 , ∣ ∇ x X m ( t f , x , t ) − I ∣ ≤ 1 2 . (76) \left|\nabla_yX_m(t,y,t_f)-I\right|\leq\frac{1}{2},\qquad\left|\nabla_xX_m(t_f,x,t)-I\right|\leq\frac{1}{2}. \tag*{(76)} ∣ ∇ y X m ( t , y , t f ) − I ∣ ≤ 2 1 , ∣ ∇ x X m ( t f , x , t ) − I ∣ ≤ 2 1 . (76)
Perturbed heat equations and moving cylinders
Proof of Lemma 4.3. We adapt Campanato’s method of [31 Chapter III, Section 2] to parabolic equations, with caloric comparison functions and the energy estimates of [42 ] . In this proof, C C C depends only on s s s . We write Q r ( t , x ) : = ( t − r 2 , t ) × D r ( x ) Q_r(t,x):=(t-r^2,t)\times D_r(x) Q r ( t , x ) := ( t − r 2 , t ) × D r ( x ) and Q r : = Q r ( 0 , 0 ) Q_r:=Q_r(0,0) Q r := Q r ( 0 , 0 ) , and take ω 1 ≤ 1 \omega_1\leq1 ω 1 ≤ 1 .
Step 1: Energy and a caloric comparison function. Testing the equation ( U − ( U ) Q 1 ) g 2 (U-(U)_{Q_1})g^2 ( U − ( U ) Q 1 ) g 2 , for a cutoff g g g equal to 1 on Q 3 / 4 Q_{3/4} Q 3/4 and vanishing near the parabolic boundary of Q 1 Q_1 Q 1 , gives
∫ Q 3 / 4 ∣ ∇ U ∣ 2 d t d x ≤ C ∫ Q 1 ∣ U − ( U ) Q 1 ∣ 2 d t d x . (77) \int_{Q_{3/4}}|\nabla U|^2\,dt\,dx\leq C\int_{Q_1}|U-(U)_{Q_1}|^2\,dt\,dx. \tag*{(77)} ∫ Q 3/4 ∣∇ U ∣ 2 d t d x ≤ C ∫ Q 1 ∣ U − ( U ) Q 1 ∣ 2 d t d x . (77)
the skew part has no principal term, since ∇ U ⋅ ( a − I ) ∇ U = 0 \nabla U \cdot(\mathbf{a}-I)\nabla U=0 ∇ U ⋅ ( a − I ) ∇ U = 0 , and the divergence-free drift contributes only 1 2 ∫ ( U − ( U ) Q 1 ) 2 b ∗ ⋅ ∇ g 2 d t d x \frac{1}{2}\int(U-(U)_{Q_1})^2\mathbf{b}_*\cdot\nabla g^2\,\mathrm{d}t\,\mathrm{d}x 2 1 ∫ ( U − ( U ) Q 1 ) 2 b ∗ ⋅ ∇ g 2 d t d x . Since b ∗ ⋅ ∇ U = ∇ ⋅ ( ( U − ( U ) Q 1 ) b ∗ ) \mathbf{b}_*\cdot\nabla U=\nabla\cdot((U-(U)_{Q_1})\mathbf{b}_*) b ∗ ⋅ ∇ U = ∇ ⋅ (( U − ( U ) Q 1 ) b ∗ ) , U U U solves ∂ t U − Δ U = ∇ ⋅ F \partial_tU-\Delta U=\nabla\cdot F ∂ t U − Δ U = ∇ ⋅ F with F : = ( a − I ) ∇ U − ( U − ( U ) Q 1 ) b ∗ F:=(\mathbf{a}-I)\nabla U-(U-(U)_{Q_1})\mathbf{b}_* F := ( a − I ) ∇ U − ( U − ( U ) Q 1 ) b ∗ , and ∣ F ∣ ≤ ϖ 1 ( ∣ ∇ U ∣ + ∣ U − ( U ) Q 1 ∣ ) |F|\leq\varpi_1(|\nabla U|+|U-(U)_{Q_1}|) ∣ F ∣ ≤ ϖ 1 ( ∣∇ U ∣ + ∣ U − ( U ) Q 1 ∣ ) . Let W W W solve this forced heat equation in Q 3 / 4 Q_{3/4} Q 3/4 with zero data, and set U c a l : = U − W U_{\mathrm{cal}}:=U-W U cal := U − W . Then U c a l − ( U ) Q 1 U_{\mathrm{cal}}-(U)_{Q_1} U cal − ( U ) Q 1 is caloric in Q 3 / 4 Q_{3/4} Q 3/4 , and by the energy estimate and (77)
∫ Q 3 / 4 ∣ W ∣ 2 d t d x ≤ C ∫ Q 3 / 4 ∣ F ∣ 2 d t d x ≤ C ϖ 1 2 ∫ Q 1 ∣ U − ( U ) Q 1 ∣ 2 d t d x . \int_{Q_{3/4}} |W|^2\,\mathrm{d}t\,\mathrm{d}x \leq C\int_{Q_{3/4}} |F|^2\,\mathrm{d}t\,\mathrm{d}x \leq C\varpi_1^2\int_{Q_1}|U-(U)_{Q_1}|^2\,\mathrm{d}t\,\mathrm{d}x. ∫ Q 3/4 ∣ W ∣ 2 d t d x ≤ C ∫ Q 3/4 ∣ F ∣ 2 d t d x ≤ C ϖ 1 2 ∫ Q 1 ∣ U − ( U ) Q 1 ∣ 2 d t d x .
Step 2: One step of decay. By the heat kernel representation of a cut-off caloric function, for every ( t , x ) , ( t ′ , x ′ ) ∈ ( − 1 4 , 0 ] × D 1 / 2 (t,x),(t',x')\in(-\frac{1}{4},0]\times D_{1/2} ( t , x ) , ( t ′ , x ′ ) ∈ ( − 4 1 , 0 ] × D 1/2 ,
∣ U c a l ( t , x ) − U c a l ( t ′ , x ′ ) ∣ ≤ C ( ∣ t − t ′ ∣ + ∣ x − x ′ ∣ ) ∫ Q 3 / 4 ∣ U c a l ( τ , z ) − ( U ) Q 1 ∣ d τ d z . |U_{\mathrm{cal}}(t,x)-U_{\mathrm{cal}}(t',x')|\leq C(|t-t'|+|x-x'|)\int_{Q_{3/4}}|U_{\mathrm{cal}}(\tau,z)-(U)_{Q_1}|\,\mathrm{d}\tau\,\mathrm{d}z. ∣ U cal ( t , x ) − U cal ( t ′ , x ′ ) ∣ ≤ C ( ∣ t − t ′ ∣ + ∣ x − x ′ ∣ ) ∫ Q 3/4 ∣ U cal ( τ , z ) − ( U ) Q 1 ∣ d τ d z .
With Step 1 and ∣ Q 1 ∣ / ∣ Q r ∣ = r − 4 |Q_1|/|Q_r|=r^{-4} ∣ Q 1 ∣/∣ Q r ∣ = r − 4 , this gives, for every 0 < r ≤ 1 2 0<r\leq\frac{1}{2} 0 < r ≤ 2 1 ,
∫ Q r ∣ U − ( U ) Q r ∣ 2 d t d x ≤ C ( r 2 + ϖ 1 2 r − 4 ) ∫ Q 1 ∣ U − ( U ) Q 1 ∣ 2 d t d x . (78) \int_{Q_r}|U-(U)_{Q_r}|^2\,\mathrm{d}t\,\mathrm{d}x\leq C(r^2+\varpi_1^2r^{-4})\int_{Q_1}|U-(U)_{Q_1}|^2\,\mathrm{d}t\,\mathrm{d}x. \tag*{(78)} ∫ Q r ∣ U − ( U ) Q r ∣ 2 d t d x ≤ C ( r 2 + ϖ 1 2 r − 4 ) ∫ Q 1 ∣ U − ( U ) Q 1 ∣ 2 d t d x . (78)
Step 3: Iteration. Since 1 + s < 2 1+s<2 1 + s < 2 , we may fix r 0 ∈ ( 0 , 1 2 ] r_0\in(0,\frac{1}{2}] r 0 ∈ ( 0 , 2 1 ] with C r 0 2 ≤ 1 2 r 0 1 + s Cr_0^2\leq\frac{1}{2}r_0^{1+s} C r 0 2 ≤ 2 1 r 0 1 + s and then ϖ 1 \varpi_1 ϖ 1 with C ϖ 1 2 r 0 − 4 ≤ 1 2 r 0 1 + s C\varpi_1^2r_0^{-4}\leq\frac{1}{2}r_0^{1+s} C ϖ 1 2 r 0 − 4 ≤ 2 1 r 0 1 + s ; then, at r = r 0 r=r_0 r = r 0 , the constant on the right side of (78) is at most r 0 1 + s r_0^{1+s} r 0 1 + s . For ( t , x ) ∈ R × R 2 (t,x)\in\mathbb{R}\times\mathbb{R}^2 ( t , x ) ∈ R × R 2 and 0 < r ≤ 1 0<r\leq1 0 < r ≤ 1 , the rescaled function ( τ , z ) ↦ U ( t + r 2 τ , x + r z ) (\tau,z)\mapsto U(t+r^2\tau,x+rz) ( τ , z ) ↦ U ( t + r 2 τ , x + r z ) solves an equation of the same form, with the drift multiplied by r ≤ 1 r\leq1 r ≤ 1 . We iterate with the radii 1 2 r 0 j \frac{1}{2}r_0^j 2 1 r 0 j , j ≥ 1 j\geq1 j ≥ 1 , starting from Q 1 / 2 ( t , x ) ⊂ Q 1 Q_{1/2}(t,x)\subset Q_1 Q 1/2 ( t , x ) ⊂ Q 1 , with a constant depending only on r 0 r_0 r 0 for the first step and for the radii between two consecutive ones, and obtain, for every ( t , x ) ∈ ( − 1 4 , 0 ] × D 1 / 2 (t,x)\in(-\frac{1}{4},0]\times D_{1/2} ( t , x ) ∈ ( − 4 1 , 0 ] × D 1/2 and 0 < r ≤ 1 2 0<r\leq\frac{1}{2} 0 < r ≤ 2 1 ,
∫ Q r ( t , x ) ∣ U ( τ , z ) − ( U ) Q r ( t , x ) ∣ 2 d τ d z ≤ C r 1 + s ∫ Q 1 ∣ U ( τ , z ) − ( U ) Q 1 ∣ 2 d τ d z . (79) \int_{Q_r(t,x)}|U(\tau,z)-(U)_{Q_r(t,x)}|^2\,\mathrm{d}\tau\,\mathrm{d}z\leq Cr^{1+s}\int_{Q_1}|U(\tau,z)-(U)_{Q_1}|^2\,\mathrm{d}\tau\,\mathrm{d}z. \tag*{(79)} ∫ Q r ( t , x ) ∣ U ( τ , z ) − ( U ) Q r ( t , x ) ∣ 2 d τ d z ≤ C r 1 + s ∫ Q 1 ∣ U ( τ , z ) − ( U ) Q 1 ∣ 2 d τ d z . (79)
Step 4: Hölder continuity. By (79) , the averages of U U U over Q r ( t , x ) Q_r(t,x) Q r ( t , x ) and Q r / 2 ( t , x ) Q_{r/2}(t,x) Q r /2 ( t , x ) differ by at most C r ( 1 + s ) / 2 ( ∫ Q 1 ∣ U − ( U ) Q 1 ∣ 2 ) 1 / 2 Cr^{(1+s)/2}\left(\int_{Q_1}|U-(U)_{Q_1}|^2\right)^{1/2} C r ( 1 + s ) /2 ( ∫ Q 1 ∣ U − ( U ) Q 1 ∣ 2 ) 1/2 . So the averages over Q 2 − i ( t , x ) Q_{2^{-i}}(t,x) Q 2 − i ( t , x ) converge uniformly in ( t , x ) (t,x) ( t , x ) to U U U , which is within C C C times this bound of ( U ) Q r ( t , x ) (U)_{Q_r(t,x)} ( U ) Q r ( t , x ) . One comparison of ( U ) Q 1 / 2 ( t , x ) (U)_{Q_{1/2}(t,x)} ( U ) Q 1/2 ( t , x ) with ( U ) Q 1 (U)_{Q_1} ( U ) Q 1 gives the L ∞ L^\infty L ∞ term of Lemma 4.3. Comparing averages over Q 2 r ( t , x ) ⊃ Q r ( t ′ , x ′ ) Q_{2r}(t,x)\supset Q_r(t',x') Q 2 r ( t , x ) ⊃ Q r ( t ′ , x ′ ) , at the radius r = ∣ x − x ′ ∣ + ∣ t − t ′ ∣ 1 / 2 r=|x-x'|+|t-t'|^{1/2} r = ∣ x − x ′ ∣ + ∣ t − t ′ ∣ 1/2 for the points ( t , x ) (t,x) ( t , x ) and ( t ′ , x ′ ) (t',x') ( t ′ , x ′ ) with t ′ ≤ t t'\leq t t ′ ≤ t , gives the seminorm for r ≤ 1 4 r\leq\frac{1}{4} r ≤ 4 1 , where (79) applies at the radius 2 r 2r 2 r . For r > 1 4 r>\frac{1}{4} r > 4 1 the seminorm follows from the L ∞ L^\infty L ∞ term, since ∣ U ( t , x ) − U ( t ′ , x ′ ) ∣ ≤ 2 ∥ U − ( U ) Q 1 ∥ L ∞ ( ( − 1 4 , 0 ] × D 1 / 2 ) |U(t,x)-U(t',x')|\leq2\|U-(U)_{Q_1}\|_{L^\infty((-\frac{1}{4},0]\times D_{1/2})} ∣ U ( t , x ) − U ( t ′ , x ′ ) ∣ ≤ 2∥ U − ( U ) Q 1 ∥ L ∞ (( − 4 1 , 0 ] × D 1/2 ) and 1 < ( 4 r ) ( 1 + s ) / 2 1<(4r)^{(1+s)/2} 1 < ( 4 r ) ( 1 + s ) /2 . The uniform continuity of U U U on ( − 1 4 , 0 ] × D 1 / 2 (-\frac{1}{4},0]\times D_{1/2} ( − 4 1 , 0 ] × D 1/2 gives the convergence of U ( t , ⋅ ) U(t,\cdot) U ( t , ⋅ ) as t ↑ 0 t\uparrow0 t ↑ 0 .
A trajectory of b \mathbf{b} b on a time interval is a C 1 C^1 C 1 curve X X X with ∂ t ′ X ( t ′ ) = b ( t ′ , X ( t ′ ) ) \partial_{t'}X(t')=\mathbf{b}(t',X(t')) ∂ t ′ X ( t ′ ) = b ( t ′ , X ( t ′ )) ; since b \mathbf{b} b is bounded and continuous, a trajectory ends at every point at every time.
Lemma 9.1 (Hölder bounds along moving cylinders). There exist c c y l > 0 c_{\mathrm{cyl}}>0 c cyl > 0 and C < ∞ C<\infty C < ∞ , depending only on α \alpha α , s s s , q q q , ζ \zeta ζ , Λ \Lambda Λ , with the following property. Let κ > 0 \kappa>0 κ > 0 , let O ⊂ R × R 2 \mathcal{O}\subset\mathbb{R}\times\mathbb{R}^2 O ⊂ R × R 2 be open, let θ \theta θ be a local weak solution of ∂ t θ + b ⋅ ∇ θ − κ Δ θ = 0 \partial_t\theta+\mathbf{b}\cdot\nabla\theta-\kappa\Delta\theta=0 ∂ t θ + b ⋅ ∇ θ − κ Δ θ = 0 in O \mathcal{O} O , and let E ⊆ R E\subseteq\mathbb{R} E ⊆ R , E ⊂ R 2 E\subset\mathbb{R}^2 E ⊂ R 2 and C ≥ min { 1 ξ , c c y l 1 / β } C\geq\min\{\frac{1}{\xi},c_{\mathrm{cyl}}^{1/\beta}\} C ≥ min { ξ 1 , c cyl 1/ β } . Assume that, for almost every t ′ ∈ R t'\in\mathbb{R} t ′ ∈ R and every disc D r ( y ) ⊂ O t ′ D_r(y)\subset\mathcal{O}_{t'} D r ( y ) ⊂ O t ′ , θ ( t ′ , ⋅ ) 2 d x ′ ≤ Ξ r 2 \theta(t',\cdot)^2\,\mathrm{d}x'\leq\Xi r^2 θ ( t ′ , ⋅ ) 2 d x ′ ≤ Ξ r 2 , and that, for every x ∈ E x\in E x ∈ E and every trajectory X X X of b \mathbf{b} b on [ t − r 2 / κ , t ] [t-r^2/\kappa,t] [ t − r 2 / κ , t ] with X ( t ) = x X(t)=x X ( t ) = x , the moving cylinder { ( t ′ , y ) : t − r 2 / κ ≤ t ′ ≤ t , ∣ y − X ( t ′ ) ∣ ≤ r } \{(t',y):t-r^2/\kappa\leq t'\leq t,\ |y-X(t')|\leq r\} {( t ′ , y ) : t − r 2 / κ ≤ t ′ ≤ t , ∣ y − X ( t ′ ) ∣ ≤ r } lies in O \mathcal{O} O . Then, for every x ∈ E x\in E x ∈ E , θ ( t ′ , ⋅ ) \theta(t',\cdot) θ ( t ′ , ⋅ ) converges in L 2 ( D r / 4 ( x ) ) L^2(D_{r/4}(x)) L 2 ( D r /4 ( x )) as t ′ ↑ t t'\uparrow t t ′ ↑ t ; the limit θ ( t , ⋅ ) \theta(t,\cdot) θ ( t , ⋅ ) is continuous on ⋃ x ∈ E D r / 4 ( x ) \bigcup_{x\in E}D_{r/4}(x) ⋃ x ∈ E D r /4 ( x ) , and
∥ θ ( t , ⋅ ) ∥ C 0 , s ( E ) ≤ C r − 1 − s Ξ . (80) \|\theta(t,\cdot)\|_{C^{0,s}(E)}\leq Cr^{-1-s}\Xi. \tag*{(80)} ∥ θ ( t , ⋅ ) ∥ C 0 , s ( E ) ≤ C r − 1 − s Ξ. (80)
Proof. Fix x ∈ E x \in E x ∈ E and a trajectory X X X of b b b on [ t − r 2 / κ , t ] [t-r^{2}/\kappa,t] [ t − r 2 / κ , t ] with X ( t ) = x X(t)=x X ( t ) = x , and set
U ( t ′ , y ) : = θ ( t + r 2 κ t ′ , X ( t + r 2 κ t ′ ) + r y ) , ( t ′ , y ) ∈ ( − 1 , 0 ) × D 1 . U(t',y):=\theta\left(t+\frac{r^{2}}{\kappa}t',X\left(t+\frac{r^{2}}{\kappa}t'\right)+ry\right),\qquad(t',y)\in(-1,0)\times D_{1}. U ( t ′ , y ) := θ ( t + κ r 2 t ′ , X ( t + κ r 2 t ′ ) + r y ) , ( t ′ , y ) ∈ ( − 1 , 0 ) × D 1 .
Then U U U solves the equation of Lemma 4.3 with a = I a=I a = I and b ∗ ( t ′ , y ) = r κ [ b ( t ′ ′ , X ( t ′ ′ ) + r y ) − b ( t ′ ′ , X ( t ′ ′ ) ) ] b_{*}(t',y)=\frac{r}{\kappa}[b(t'',X(t'')+ry)-b(t'',X(t''))] b ∗ ( t ′ , y ) = κ r [ b ( t ′′ , X ( t ′′ ) + r y ) − b ( t ′′ , X ( t ′′ ))] , t ′ ′ = t + r 2 t ′ / κ t''=t+r^{2}t'/\kappa t ′′ = t + r 2 t ′ / κ : the frame moves with b b b at its centre. This drift is divergence free, and by (17) ∣ b ∗ ∣ ≤ C r 1 + α / κ ≤ ϖ 1 |b_{*}|\le Cr^{1+\alpha}/\kappa\le\varpi_{1} ∣ b ∗ ∣ ≤ C r 1 + α / κ ≤ ϖ 1 once c c y l c_{\mathrm{cyl}} c cyl is small. Since the discs D r ( X ( t ′ ′ ) ) D_{r}(X(t'')) D r ( X ( t ′′ )) lie in O t ′ O_{t'} O t ′ ,
∫ ( − 1 , 0 ) × D 1 U ( τ , z ) 2 d τ d z ≤ Ξ 2 π r 2 . \int_{(-1,0)\times D_{1}}U(\tau,z)^{2}\,d\tau\,dz\le\frac{\Xi^{2}}{\pi r^{2}}. ∫ ( − 1 , 0 ) × D 1 U ( τ , z ) 2 d τ d z ≤ π r 2 Ξ 2 .
Lemma 4.3 bounds U U U on ( − 1 4 , 0 ] × D 1 / 2 (-\frac14,0]\times D_{1/2} ( − 4 1 , 0 ] × D 1/2 ; since X ( t ′ ′ ) → x X(t'')\to x X ( t ′′ ) → x as t ′ ′ ↑ t t''\uparrow t t ′′ ↑ t , its Hölder bound gives the convergence of θ ( t ′ ′ , ⋅ ) \theta(t'',\cdot) θ ( t ′′ , ⋅ ) to U ( 0 , ( ⋅ − x ) / r ) U(0,(\cdot-x)/r) U ( 0 , ( ⋅ − x ) / r ) , uniformly on D r / 4 ( x ) D_{r/4}(x) D r /4 ( x ) and locally uniformly on D r / 2 ( x ) D_{r/2}(x) D r /2 ( x ) , and, on D r / 2 ( x ) D_{r/2}(x) D r /2 ( x ) , the limit θ ( t , ⋅ ) \theta(t,\cdot) θ ( t , ⋅ ) satisfies the bounds ∣ θ ( t , ⋅ ) ∣ ≤ C Ξ / r |\theta(t,\cdot)|\le C\Xi/r ∣ θ ( t , ⋅ ) ∣ ≤ C Ξ/ r and ∣ θ ( t , z ) − θ ( t , z ′ ) ∣ ≤ C Ξ r − 1 ( ∣ z − z ′ ∣ / r ) ( 1 + s ) / 2 |\theta(t,z)-\theta(t,z')|\le C\Xi r^{-1}(|z-z'|/r)^{(1+s)/2} ∣ θ ( t , z ) − θ ( t , z ′ ) ∣ ≤ C Ξ r − 1 ( ∣ z − z ′ ∣/ r ) ( 1 + s ) /2 ; the limits from different cylinders agree on overlaps. For z , z ′ ∈ E z,z'\in E z , z ′ ∈ E with ∣ z − z ′ ∣ < r / 4 |z-z'|<r/4 ∣ z − z ′ ∣ < r /4 , the cylinder at z z z gives ∣ θ ( t , z ) − θ ( t , z ′ ) ∣ ≤ C Ξ r − 1 − s ∣ z − z ′ ∣ s |\theta(t,z)-\theta(t,z')|\le C\Xi r^{-1-s}|z-z'|^{s} ∣ θ ( t , z ) − θ ( t , z ′ ) ∣ ≤ C Ξ r − 1 − s ∣ z − z ′ ∣ s , since s < 1 2 ( 1 + s ) s<\frac12(1+s) s < 2 1 ( 1 + s ) ; for ∣ z − z ′ ∣ ≥ r / 4 |z-z'|\ge r/4 ∣ z − z ′ ∣ ≥ r /4 the bound on ∣ θ ( t , ⋅ ) ∣ |\theta(t,\cdot)| ∣ θ ( t , ⋅ ) ∣ gives the same. □
Transfer to the full velocity field
Lemma 9.2 (Stream comparison). Let Φ \Phi Φ and Φ ~ \widetilde{\Phi} Φ be as in Proposition 4.2 , and let κ > 0 \kappa>0 κ > 0 and ϖ ≥ 0 \varpi\ge0 ϖ ≥ 0 with ∥ Φ − Φ ~ ∥ L ∞ ( R × T 2 ) ≤ κ ϖ \|\Phi-\widetilde{\Phi}\|_{L^{\infty}(\mathbb{R}\times\mathbb{T}^{2})}\le\kappa\varpi ∥Φ − Φ ∥ L ∞ ( R × T 2 ) ≤ κϖ . Let t i n ∈ R t_{\mathrm{in}}\in\mathbb{R} t in ∈ R , T > 0 T>0 T > 0 , θ i n ∈ L 2 ( T 2 ) \theta_{\mathrm{in}}\in L^{2}(\mathbb{T}^{2}) θ in ∈ L 2 ( T 2 ) , and let θ \theta θ and θ ~ \widetilde{\theta} θ be weak solutions on ( t i n , t i n + T ) (t_{\mathrm{in}},t_{\mathrm{in}}+T) ( t in , t in + T ) of the two equations of Proposition 4.2 , with the datum θ i n \theta_{\mathrm{in}} θ in at time t i n t_{\mathrm{in}} t in . Then
∥ θ − θ ~ ∥ L ∞ ( [ t i n , t i n + T ] ; L 2 ( T 2 ) ) ≤ ϖ ∥ θ i n ∥ L 2 ( T 2 ) . \|\theta-\widetilde{\theta}\|_{L^{\infty}([t_{\mathrm{in}},t_{\mathrm{in}}+T];L^{2}(\mathbb{T}^{2}))}\le\varpi\|\theta_{\mathrm{in}}\|_{L^{2}(\mathbb{T}^{2})}. ∥ θ − θ ∥ L ∞ ([ t in , t in + T ] ; L 2 ( T 2 )) ≤ ϖ ∥ θ in ∥ L 2 ( T 2 ) .
Proof. Since ∇ ⊥ ( Φ − Φ ~ ) ⋅ ∇ θ ~ = − ∇ ⋅ ( ( Φ − Φ ~ ) ∇ ⊥ θ ~ ) \nabla^{\perp}(\Phi-\widetilde{\Phi})\cdot\nabla\widetilde{\theta}=-\nabla\cdot((\Phi-\widetilde{\Phi})\nabla^{\perp}\widetilde{\theta}) ∇ ⊥ ( Φ − Φ ) ⋅ ∇ θ = − ∇ ⋅ (( Φ − Φ ) ∇ ⊥ θ ) , the difference θ − θ ~ \theta-\widetilde{\theta} θ − θ solves
∂ t ( θ − θ ~ ) + ∇ ⊥ Φ ⋅ ∇ ( θ − θ ~ ) − κ Δ ( θ − θ ~ ) = ∇ ⋅ ( ( Φ − Φ ~ ) ∇ ⊥ θ ~ ) \partial_{t}(\theta-\widetilde{\theta})+\nabla^{\perp}\Phi\cdot\nabla(\theta-\widetilde{\theta})-\kappa\Delta(\theta-\widetilde{\theta})=\nabla\cdot((\Phi-\widetilde{\Phi})\nabla^{\perp}\widetilde{\theta}) ∂ t ( θ − θ ) + ∇ ⊥ Φ ⋅ ∇ ( θ − θ ) − κ Δ ( θ − θ ) = ∇ ⋅ (( Φ − Φ ) ∇ ⊥ θ )
with zero datum. Its energy identity, Young’s inequality and (73) for θ ~ \widetilde{\theta} θ give, for t i n < t < t i n + T t_{\mathrm{in}}<t<t_{\mathrm{in}}+T t in < t < t in + T ,
∥ ( θ − θ ~ ) ( t , ⋅ ) ∥ L 2 ( T 2 ) 2 ≤ κ ϖ 2 ∥ ∇ θ ~ ∥ L 2 ( ( t i n , t ) × T 2 ) 2 ≤ ϖ 2 ∥ θ i n ∥ L 2 ( T 2 ) 2 . \|(\theta-\widetilde{\theta})(t,\cdot)\|_{L^{2}(\mathbb{T}^{2})}^{2}\le\kappa\varpi^{2}\|\nabla\widetilde{\theta}\|_{L^{2}((t_{\mathrm{in}},t)\times\mathbb{T}^{2})}^{2}\le\varpi^{2}\|\theta_{\mathrm{in}}\|_{L^{2}(\mathbb{T}^{2})}^{2}. ∥ ( θ − θ ) ( t , ⋅ ) ∥ L 2 ( T 2 ) 2 ≤ κ ϖ 2 ∥∇ θ ∥ L 2 (( t in , t ) × T 2 ) 2 ≤ ϖ 2 ∥ θ in ∥ L 2 ( T 2 ) 2 .
Proof of Proposition 4.2 . Let ϖ 0 : = ϖ 1 \varpi_{0}:=\varpi_{1} ϖ 0 := ϖ 1 , the constant of Lemma 4.3, and fix t ∈ [ t 1 , t 2 ] t\in[t_{1},t_{2}] t ∈ [ t 1 , t 2 ] .
Step 1: The inputs. By Lemma 9.2 with t i n + T = t 2 + r 2 / κ t_{\mathrm{in}}+T=t_{2}+r^{2}/\kappa t in + T = t 2 + r 2 / κ ,
∥ θ − θ ~ ∥ L ∞ ( [ t i n , t 2 + r 2 / κ ] ; L 2 ( T 2 ) ) ≤ ϖ ∥ θ i n ∥ L 2 ( T 2 ) ; \|\theta-\widetilde{\theta}\|_{L^{\infty}([t_{\mathrm{in}},t_{2}+r^{2}/\kappa];L^{2}(\mathbb{T}^{2}))}\le\varpi\|\theta_{\mathrm{in}}\|_{L^{2}(\mathbb{T}^{2})}; ∥ θ − θ ∥ L ∞ ([ t in , t 2 + r 2 / κ ] ; L 2 ( T 2 )) ≤ ϖ ∥ θ in ∥ L 2 ( T 2 ) ;
by (9.2′), the bound Ξ \Xi Ξ on θ ~ \widetilde{\theta} θ holds at every time of [ t 1 − r 2 / κ , t 2 + r 2 / κ ] [t_{1}-r^{2}/\kappa,t_{2}+r^{2}/\kappa] [ t 1 − r 2 / κ , t 2 + r 2 / κ ] .
Step 2: A cylinder moving with ∇ ⊥ Φ ~ \nabla^{\perp}\widetilde{\Phi} ∇ ⊥ Φ . Fix x ∈ R 2 x\in\mathbb{R}^{2} x ∈ R 2 and the trajectory X ~ \widetilde{X} X of ∇ ⊥ Φ ~ \nabla^{\perp}\widetilde{\Phi} ∇ ⊥ Φ with X ~ ( t ) = x \widetilde{X}(t)=x X ( t ) = x , and set, for ( t ′ , y ) ∈ ( − 1 , 0 ) × D 1 (t',y)\in(-1,0)\times D_{1} ( t ′ , y ) ∈ ( − 1 , 0 ) × D 1 and t ′ ′ : = t + r 2 t ′ / ( 4 κ ) t'':=t+r^{2}t'/(4\kappa) t ′′ := t + r 2 t ′ / ( 4 κ ) ,
U ( t ′ , y ) : = θ ( t ′ ′ , X ~ ( t ′ ′ ) + r 2 y ) , U ~ ( t ′ , y ) : = θ ~ ( t ′ ′ , X ~ ( t ′ ′ ) + r 2 y ) . U(t',y):=\theta\left(t'',\widetilde{X}(t'')+\frac{r}{2}y\right),\qquad\widetilde{U}(t',y):=\widetilde{\theta}\left(t'',\widetilde{X}(t'')+\frac{r}{2}y\right). U ( t ′ , y ) := θ ( t ′′ , X ( t ′′ ) + 2 r y ) , U ( t ′ , y ) := θ ( t ′′ , X ( t ′′ ) + 2 r y ) .
By the identity in the proof of Lemma 9.2, applied to θ \theta θ , U U U solves the equation of Lemma 4.3 with
a − I = Φ − Φ ~ κ ( 0 − 1 1 0 ) , b ∗ ( t ′ , y ) = r 2 κ [ ∇ ⊥ Φ ~ ( t ′ ′ , X ~ ( t ′ ′ ) + r 2 y ) − ∇ ⊥ Φ ~ ( t ′ ′ , X ~ ( t ′ ′ ) ) ] , a-I=\frac{\Phi-\widetilde{\Phi}}{\kappa}
\begin{pmatrix}
0 & -1\\
1 & 0
\end{pmatrix},
\qquad
b_{*}(t',y)=\frac{r}{2\kappa}\left[\nabla^{\perp}\widetilde{\Phi}\left(t'',\widetilde{X}(t'')+\frac{r}{2}y\right)-\nabla^{\perp}\widetilde{\Phi}(t'',\widetilde{X}(t''))\right], a − I = κ Φ − Φ ( 0 1 − 1 0 ) , b ∗ ( t ′ , y ) = 2 κ r [ ∇ ⊥ Φ ( t ′′ , X ( t ′′ ) + 2 r y ) − ∇ ⊥ Φ ( t ′′ , X ( t ′′ )) ] ,
the first evaluated at ( t ′ ′ , X ~ ( t ′ ′ ) + r 2 y ) (t'',\widetilde{X}(t'')+\frac{r}{2}y) ( t ′′ , X ( t ′′ ) + 2 r y ) ; a − I a-I a − I is skew with ∣ a − I ∣ ≤ ϖ |a-I|\le\varpi ∣ a − I ∣ ≤ ϖ , and b ∗ b_{*} b ∗ is divergence free with ∣ b ∗ ∣ ≤ r 2 4 κ ∥ ∇ 2 Φ ~ ∥ L ∞ ( R × T 2 ) |b_{*}|\le\frac{r^{2}}{4\kappa}\|\nabla^{2}\widetilde{\Phi}\|_{L^{\infty}(\mathbb{R}\times\mathbb{T}^{2})} ∣ b ∗ ∣ ≤ 4 κ r 2 ∥ ∇ 2 Φ ∥ L ∞ ( R × T 2 ) , so ∣ a − I ∣ + ∣ b ∗ ∣ ≤ ϖ 0 |a-I|+|b_{*}|\le\varpi_{0} ∣ a − I ∣ + ∣ b ∗ ∣ ≤ ϖ 0 ; U ~ \widetilde{U} U solves it with a = I a=I a = I ; and t ′ ′ ∈ ( t 1 − r 2 / κ , t 2 ] t''\in(t_{1}-r^{2}/\kappa,t_{2}] t ′′ ∈ ( t 1 − r 2 / κ , t 2 ] .
Step 3: Closeness to a constant. By Step 1, U ~ ( t ′ , ⋅ ) \widetilde{U}(t',\cdot) U ( t ′ , ⋅ ) oscillates on D 1 D_{1} D 1 by at most Ξ r s \Xi r^{s} Ξ r s . The function u 1 : = 4 u ( 2 ⋅ ) u_{1}:=4u(2\cdot) u 1 := 4 u ( 2 ⋅ ) , with u u u fixed at the start of Section 6, is smooth, nonnegative, of integral one and supported in D 1 / 2 D_{1/2} D 1/2 . Testing the equation of U ~ \widetilde{U} U with it gives, in the sense of distributions in t ′ t' t ′ ,
d d t ′ ⟨ U ~ ( t ′ , ⋅ ) , u 1 ⟩ L 2 ( D 1 ) = ∫ D 1 ( U ~ ( t ′ , y ) − U ~ ( t ′ , 0 ) ) ( Δ + b ∗ ⋅ ∇ ) u 1 ( y ) d y (81) \frac{d}{dt'}\langle\widetilde{U}(t',\cdot),u_{1}\rangle_{L^{2}(D_{1})}
=\int_{D_{1}}\bigl(\widetilde{U}(t',y)-\widetilde{U}(t',0)\bigr)(\Delta+b_{*}\cdot\nabla)u_{1}(y)\,dy \tag*{(81)} d t ′ d ⟨ U ( t ′ , ⋅ ) , u 1 ⟩ L 2 ( D 1 ) = ∫ D 1 ( U ( t ′ , y ) − U ( t ′ , 0 ) ) ( Δ + b ∗ ⋅ ∇ ) u 1 ( y ) d y (81)
because ( Δ + b ∗ ⋅ ∇ ) u 1 (\Delta+b_* \cdot\nabla)u_1 ( Δ + b ∗ ⋅ ∇ ) u 1 has integral zero ( ∇ ⋅ b ∗ = 0 ) (\nabla\cdot b_*=0) ( ∇ ⋅ b ∗ = 0 ) . Hence the pairing varies by at most C Ξ r s C\Xi r^s C Ξ r s over ( − 1 , 0 ) (-1,0) ( − 1 , 0 ) , and U ~ \widetilde U U stays within C Ξ r s C\Xi r^s C Ξ r s of ⟨ U ~ ( 0 , ⋅ ) , u 1 ⟩ L 2 ( D 1 ) \langle\widetilde U(0,\cdot),u_1\rangle_{L^2(D_1)} ⟨ U ( 0 , ⋅ ) , u 1 ⟩ L 2 ( D 1 ) . By Step 1 and the four translates of the unit cell, ∫ D 1 ∣ U − U ~ ∣ 2 ( t ′ , y ) d y ≤ 16 r − 2 ϖ 2 ∥ θ i n ∥ L 2 ( T 2 ) 2 \int_{D_1}|U-\widetilde U|^2(t',y)\,dy\le16r^{-2}\varpi^2\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)}^2 ∫ D 1 ∣ U − U ∣ 2 ( t ′ , y ) d y ≤ 16 r − 2 ϖ 2 ∥ θ in ∥ L 2 ( T 2 ) 2 : normalizing the L 2 L^2 L 2 deviation to the cylinder costs r − 1 r^{-1} r − 1 . Altogether,
\left(\fint_{(-1,0)\times D_1}\left|U(t',y)-\langle\widetilde U(0,\cdot),u_1\rangle_{L^2(D_1)}\right|^2\,dt'\,dy\right)^{1/2}\le C\left(\varpi r^{-1}\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)}+\Xi r^s\right). \tag*{(82)}
Step 4: All distances. Since X ~ ( t ′ ′ ) → x \widetilde X(t'')\to x X ( t ′′ ) → x as t ′ ′ ↑ t t''\uparrow t t ′′ ↑ t , Lemma 4.3 and (82) give that θ ( t ′ ′ , ⋅ ) \theta(t'',\cdot) θ ( t ′′ , ⋅ ) converges to U ( 0 , 2 ( ⋅ − x ) / r ) U(0,2(\cdot-x)/r) U ( 0 , 2 ( ⋅ − x ) / r ) in L l o c 2 ( D r / 4 ( x ) ) L^2_{\mathrm{loc}}(D_{r/4}(x)) L loc 2 ( D r /4 ( x )) , and, by (9.2'), this limit is θ ( t , ⋅ ) \theta(t,\cdot) θ ( t , ⋅ ) . It is within C ( ϖ r − 1 ∥ θ i n ∥ L 2 ( T 2 ) + Ξ r s ) C(\varpi r^{-1}\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)}+\Xi r^s) C ( ϖ r − 1 ∥ θ in ∥ L 2 ( T 2 ) + Ξ r s ) of the constant in (82) , hence of θ ~ ( t , x ) = U ~ ( 0 , 0 ) \widetilde\theta(t,x)=\widetilde U(0,0) θ ( t , x ) = U ( 0 , 0 ) , and its increments over ∣ z − z ′ ∣ ≤ r 4 |z-z'|\le\frac r4 ∣ z − z ′ ∣ ≤ 4 r are at most C ( ϖ r − 1 ∥ θ i n ∥ L 2 ( T 2 ) + Ξ r s ) ( ∣ z − z ′ ∣ / r ) ( 1 + s ) / 2 C(\varpi r^{-1}\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)}+\Xi r^s)(|z-z'|/r)^{(1+s)/2} C ( ϖ r − 1 ∥ θ in ∥ L 2 ( T 2 ) + Ξ r s ) ( ∣ z − z ′ ∣/ r ) ( 1 + s ) /2 . Since x x x is arbitrary,
∥ θ ( t , ⋅ ) − θ ~ ( t , ⋅ ) ∥ L ∞ ( T 2 ) ≤ C ( ϖ r − 1 ∥ θ i n ∥ L 2 ( T 2 ) + Ξ r s ) . \|\theta(t,\cdot)-\widetilde\theta(t,\cdot)\|_{L^\infty(\mathbb{T}^2)}\le C\left(\varpi r^{-1}\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)}+\Xi r^s\right). ∥ θ ( t , ⋅ ) − θ ( t , ⋅ ) ∥ L ∞ ( T 2 ) ≤ C ( ϖ r − 1 ∥ θ in ∥ L 2 ( T 2 ) + Ξ r s ) .
For ∣ z − z ′ ∣ T 2 < r 4 |z-z'|_{\mathbb{T}^2}<\frac r4 ∣ z − z ′ ∣ T 2 < 4 r the increment bound gives ∣ θ ( t , z ) − θ ( t , z ′ ) ∣ ≤ C ( ϖ r − 1 − s ∥ θ i n ∥ L 2 ( T 2 ) + Ξ ) ∣ z − z ′ ∣ T 2 s |\theta(t,z)-\theta(t,z')|\le C(\varpi r^{-1-s}\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)}+\Xi)|z-z'|_{\mathbb{T}^2}^s ∣ θ ( t , z ) − θ ( t , z ′ ) ∣ ≤ C ( ϖ r − 1 − s ∥ θ in ∥ L 2 ( T 2 ) + Ξ ) ∣ z − z ′ ∣ T 2 s ; for ∣ z − z ′ ∣ T 2 ≥ r 4 |z-z'|_{\mathbb{T}^2}\ge\frac r4 ∣ z − z ′ ∣ T 2 ≥ 4 r , the comparison with θ ~ ( t , ⋅ ) \widetilde\theta(t,\cdot) θ ( t , ⋅ ) and the bound on θ ~ \widetilde\theta θ give the same, since 1 ≤ 4 s r − s ∣ z − z ′ ∣ T 2 s 1\le4^s r^{-s}|z-z'|_{\mathbb{T}^2}^s 1 ≤ 4 s r − s ∣ z − z ′ ∣ T 2 s . Hence ∥ θ ( t , ⋅ ) ∥ C 0 , s ( T 2 ) ≤ C ( Ξ + ϖ r − 1 − s ∥ θ i n ∥ L 2 ( T 2 ) ) , \|\theta(t,\cdot)\|_{C^{0,s}(\mathbb{T}^2)}\le C(\Xi+\varpi r^{-1-s}\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)}), ∥ θ ( t , ⋅ ) ∥ C 0 , s ( T 2 ) ≤ C ( Ξ + ϖ r − 1 − s ∥ θ in ∥ L 2 ( T 2 ) ) , since r ≤ 1 r\le1 r ≤ 1 ; and t ∈ [ t 1 , t 2 ] t\in[t_1,t_2] t ∈ [ t 1 , t 2 ] is arbitrary. □
Proof of Theorem A
Proof of Theorem A. We choose the number n n n of extra levels (Step 1); for the bands far out, Proposition 4.1 bounds the solution for the truncated velocity field b M + n b_{M+n} b M + n , and Proposition 4.2 transfers this bound to θ \theta θ (Step 2); the finitely many remaining bands follow from Lemma 9.1 (Step 3). Let Λ 0 \Lambda_0 Λ 0 be the threshold of Proposition 4.1, and fix Λ ≥ Λ 0 \Lambda\ge\Lambda_0 Λ ≥ Λ 0 and 0 < t 0 < T 0<t_0<T 0 < t 0 < T ; the constants below are independent of t i n t_{\mathrm{in}} t in , κ \kappa κ , θ i n \theta_{\mathrm{in}} θ in and θ \theta θ . We continue θ \theta θ to the weak solution on ( t i n , t i n + T + 1 ) (t_{\mathrm{in}},t_{\mathrm{in}}+T+1) ( t in , t in + T + 1 ) with datum θ i n \theta_{\mathrm{in}} θ in , which agrees with it on ( t i n , t i n + T ) (t_{\mathrm{in}},t_{\mathrm{in}}+T) ( t in , t in + T ) .
Step 1: The number of extra levels. The bracket β q − 1 + s 2 ( 2 − β ) \beta q-\frac{1+s}{2}(2-\beta) β q − 2 1 + s ( 2 − β ) is positive, since β q > β \beta q>\beta β q > β and, as β = 1 + α \beta=1+\alpha β = 1 + α and s < 1 2 ( 1 − α ) s<\frac12(1-\alpha) s < 2 1 ( 1 − α ) , ( 1 + s ) ( 2 − β ) < 1 2 ( 3 − α ) ( 1 − α ) < 2 β (1+s)(2-\beta)<\frac12(3-\alpha)(1-\alpha)<2\beta ( 1 + s ) ( 2 − β ) < 2 1 ( 3 − α ) ( 1 − α ) < 2 β . Fix an integer n ≥ 1 n\ge1 n ≥ 1 with
q n ( β q − ( 1 + s ) ( 2 − β ) 2 ) − ( β − γ ) ( 3 + s ) 2 ≥ 0 ; (83) q^n\left(\beta q-\frac{(1+s)(2-\beta)}{2}\right)-\frac{(\beta-\gamma)(3+s)}{2}\ge0; \tag*{(83)} q n ( β q − 2 ( 1 + s ) ( 2 − β ) ) − 2 ( β − γ ) ( 3 + s ) ≥ 0 ; (83)
it depends only on α \alpha α , s s s , q q q .
Step 2: The bands far out. Let κ ∈ K M \kappa\in K_M κ ∈ K M , so that 1 2 ε M β − γ ≤ κ ≤ 2 ε M β − γ \frac12\varepsilon_M^{\beta-\gamma}\le\kappa\le2\varepsilon_M^{\beta-\gamma} 2 1 ε M β − γ ≤ κ ≤ 2 ε M β − γ . We apply Proposition 4.2, whose regularity hypotheses hold by Proposition 7.3(i), (iii) and (iv), with Φ = ϕ \Phi=\phi Φ = ϕ , Φ ~ = ϕ M + n \widetilde\Phi=\phi_{M+n} Φ = ϕ M + n and
r : = ( ϖ κ 2 C 0 ) 1 / 2 ε M + n ( 2 − β ) / 2 , ϖ : = C 0 ε M + n + 1 β κ , r:=\left(\frac{\varpi\kappa}{2C_0}\right)^{1/2}\varepsilon_{M+n}^{(2-\beta)/2},\qquad\varpi:=\frac{C_0\varepsilon_{M+n+1}^{\beta}}{\kappa}, r := ( 2 C 0 ϖ κ ) 1/2 ε M + n ( 2 − β ) /2 , ϖ := κ C 0 ε M + n + 1 β ,
so that, by (72) , ∥ ϕ − ϕ M + n ∥ L ∞ ( R × T 2 ) ≤ κ ϖ \|\phi-\phi_{M+n}\|_{L^\infty(\mathbb{R}\times\mathbb{T}^2)}\le\kappa\varpi ∥ ϕ − ϕ M + n ∥ L ∞ ( R × T 2 ) ≤ κϖ and r 2 κ ∥ ∇ 2 ϕ M + n ∥ L ∞ ( R × T 2 ) ≤ 1 2 ϖ 0 \frac{r^2}{\kappa}\|\nabla^2\phi_{M+n}\|_{L^\infty(\mathbb{R}\times\mathbb{T}^2)}\le\frac12\varpi_0 κ r 2 ∥ ∇ 2 ϕ M + n ∥ L ∞ ( R × T 2 ) ≤ 2 1 ϖ 0 . Since ε m ≍ ε m − 1 q \varepsilon_m\asymp\varepsilon_{m-1}^q ε m ≍ ε m − 1 q , we have ε M + n + 1 ≤ C ε M q n + 1 \varepsilon_{M+n+1}\le C\varepsilon_M^{q^{n+1}} ε M + n + 1 ≤ C ε M q n + 1 and ε M + n ≥ c ε M q n \varepsilon_{M+n}\ge c\varepsilon_M^{q^n} ε M + n ≥ c ε M q n , with constants depending on n n n , and, as κ ≥ 1 2 ε M β − γ \kappa\ge\frac12\varepsilon_M^{\beta-\gamma} κ ≥ 2 1 ε M β − γ ,
ϖ r − 1 − s ≤ C ε M + n + 1 β κ − ( 3 + s ) / 2 ε M + n − ( 2 − β ) ( 1 + s ) / 2 ≤ C ε M q n ( β q − ( 1 + s ) ( 2 − β ) / 2 ) − ( β − γ ) ( 3 + s ) / 2 ≤ C \varpi r^{-1-s}\le C\varepsilon_{M+n+1}^{\beta}\kappa^{-(3+s)/2}\varepsilon_{M+n}^{-(2-\beta)(1+s)/2}\le C\varepsilon_M^{q^n(\beta q-(1+s)(2-\beta)/2)-(\beta-\gamma)(3+s)/2}\le C ϖ r − 1 − s ≤ C ε M + n + 1 β κ − ( 3 + s ) /2 ε M + n − ( 2 − β ) ( 1 + s ) /2 ≤ C ε M q n ( β q − ( 1 + s ) ( 2 − β ) /2 ) − ( β − γ ) ( 3 + s ) /2 ≤ C
by (83) . Moreover r ≤ C ε M ( β − γ ) / 2 r\le C\varepsilon_M^{(\beta-\gamma)/2} r ≤ C ε M ( β − γ ) /2 , r 2 / κ ≤ C ε M 2 − β r^2/\kappa\le C\varepsilon_M^{2-\beta} r 2 / κ ≤ C ε M 2 − β and ϖ ≤ C ε M β q n + 1 − ( β − γ ) \varpi\le C\varepsilon_M^{\beta q^{n+1}-(\beta-\gamma)} ϖ ≤ C ε M β q n + 1 − ( β − γ ) tend to zero as M → ∞ M\to\infty M → ∞ , uniformly in κ ∈ K M \kappa\in K_M κ ∈ K M ; we fix M Λ M_\Lambda M Λ , depending only on α \alpha α , s s s , q q q , ζ \zeta ζ , Λ \Lambda Λ , t 0 t_0 t 0 , T T T , such that, for M ≥ M Λ M\ge M_\Lambda M ≥ M Λ , r ≤ 1 8 r\le\frac18 r ≤ 8 1 , r 2 / κ ≤ min { 1 2 t 0 , 1 } r^2/\kappa\le\min\{\frac12t_0,1\} r 2 / κ ≤ min { 2 1 t 0 , 1 } , ϖ ≤ 1 2 ϖ 0 \varpi\le\frac12\varpi_0 ϖ ≤ 2 1 ϖ 0 , and M M M is at least the threshold M 1 M_1 M 1 of Proposition 4.1 for the times 1 2 t 0 < T + 1 \frac{1}{2}t_0 < T + 1 2 1 t 0 < T + 1 . For such M M M , Proposition 4.1 bounds the solution θ ~ \widetilde{\theta} θ of (28) on ( t i n , t i n + T + 1 ) (t_{\mathrm{in}},t_{\mathrm{in}}+T+1) ( t in , t in + T + 1 ) in L ∞ ( [ t i n + 1 2 t 0 , t i n + T + 1 ] ; C 0 , s ( T 2 ) ) L^\infty([t_{\mathrm{in}}+\frac{1}{2}t_0,t_{\mathrm{in}}+T+1];C^{0,s}(\mathbb{T}^2)) L ∞ ([ t in + 2 1 t 0 , t in + T + 1 ] ; C 0 , s ( T 2 )) by C ∥ θ i n ∥ L 2 ( T 2 ) C\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)} C ∥ θ in ∥ L 2 ( T 2 ) , and, by Proposition 4.2 with t 1 = t i n + t 0 t_1=t_{\mathrm{in}}+t_0 t 1 = t in + t 0 and t 2 = t i n + T t_2=t_{\mathrm{in}}+T t 2 = t in + T , we obtain ∥ θ ∥ L ∞ ( [ t i n + t 0 , t i n + T ] ; C 0 , s ( T 2 ) ) ≤ C ∥ θ i n ∥ L 2 ( T 2 ) \|\theta\|_{L^\infty([t_{\mathrm{in}}+t_0,t_{\mathrm{in}}+T];C^{0,s}(\mathbb{T}^2))}\le C\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)} ∥ θ ∥ L ∞ ([ t in + t 0 , t in + T ] ; C 0 , s ( T 2 )) ≤ C ∥ θ in ∥ L 2 ( T 2 ) .
Step 3: The remaining bands. If κ ∈ K M \kappa\in\mathcal{K}_M κ ∈ K M with M < M A M<M_A M < M A , then κ ≥ 1 2 ε M A β − γ \kappa\ge\frac{1}{2}\varepsilon_{M_A}^{\beta-\gamma} κ ≥ 2 1 ε M A β − γ . Let t ∈ [ t i n + t 0 , t i n + T ] t\in[t_{\mathrm{in}}+t_0,t_{\mathrm{in}}+T] t ∈ [ t in + t 0 , t in + T ] and
r : = min { 1 8 , c c y l ( 1 2 ε M A β − γ ) 1 / β , ( 1 4 ε M A β − γ − 1 t 0 ) 1 / 2 } . r:=\min\left\{\frac{1}{8},c_{\mathrm{cyl}}\left(\frac{1}{2}\varepsilon_{M_A}^{\beta-\gamma}\right)^{1/\beta},\left(\frac{1}{4}\varepsilon_{M_A}^{\beta-\gamma-1}t_0\right)^{1/2}\right\}. r := min { 8 1 , c cyl ( 2 1 ε M A β − γ ) 1/ β , ( 4 1 ε M A β − γ − 1 t 0 ) 1/2 } .
Then r ≤ c c y l κ 1 / β r\le c_{\mathrm{cyl}}\kappa^{1/\beta} r ≤ c cyl κ 1/ β and r 2 / κ ≤ 1 2 t 0 r^2/\kappa\le\frac{1}{2}t_0 r 2 / κ ≤ 2 1 t 0 , so every moving cylinder ending at time t t t lies in O : = ( t i n , t i n + T + 1 ) × R 2 \mathcal{O}:=(t_{\mathrm{in}},t_{\mathrm{in}}+T+1)\times\mathbb{R}^2 O := ( t in , t in + T + 1 ) × R 2 . By (73) and the four translates of the unit cell, the disc bound of Lemma 9.1 holds with Ξ = 2 ∥ θ i n ∥ L 2 ( T 2 ) \Xi=2\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)} Ξ = 2∥ θ in ∥ L 2 ( T 2 ) . By the lemma with E = R 2 E=\mathbb{R}^2 E = R 2 , we obtain ∥ θ ( t , ⋅ ) ∥ C 0 , s ( T 2 ) ≤ C r − 1 − s ∥ θ i n ∥ L 2 ( T 2 ) \|\theta(t,\cdot)\|_{C^{0,s}(\mathbb{T}^2)}\le Cr^{-1-s}\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)} ∥ θ ( t , ⋅ ) ∥ C 0 , s ( T 2 ) ≤ C r − 1 − s ∥ θ in ∥ L 2 ( T 2 ) (for periodic functions the norms on R 2 \mathbb{R}^2 R 2 and T 2 \mathbb{T}^2 T 2 agree), and r r r depends only on α , s , q , ζ , Λ , t 0 , T \alpha,s,q,\zeta,\Lambda,t_0,T α , s , q , ζ , Λ , t 0 , T .
Local regularity in tubes: Theorem C
For an integer m ≥ 0 m\ge0 m ≥ 0 , a radius r t u b e > 0 r_{\mathrm{tube}}>0 r tube > 0 and a final time t f ∈ R t_{\mathrm{f}}\in\mathbb{R} t f ∈ R , the tube is
T : = { ( t , X m ( t , y , t f ) ) : t f − 8 h m + 1 < t < t f , ∣ y ∣ < r t u b e } . (84) \mathcal{T}:=\{(t,X_m(t,y,t_{\mathrm{f}})):t_{\mathrm{f}}-8h_{m+1}<t<t_{\mathrm{f}},\ |y|<r_{\mathrm{tube}}\}. \tag*{(84)} T := {( t , X m ( t , y , t f )) : t f − 8 h m + 1 < t < t f , ∣ y ∣ < r tube } . (84)
the disc D r t u b e D_{r_{\mathrm{tube}}} D r tube carried by the trajectories (13) of b m \mathbf{b}_m b m over the time 8 h m + 1 8h_{m+1} 8 h m + 1 . For κ > 0 \kappa>0 κ > 0 , a local energy solution of ∂ t θ + b ⋅ ∇ θ − κ Δ θ = 0 \partial_t\theta+\mathbf{b}\cdot\nabla\theta-\kappa\Delta\theta=0 ∂ t θ + b ⋅ ∇ θ − κ Δ θ = 0 in T \mathcal{T} T is a function θ ∈ L 2 ( T ) \theta\in L^2(\mathcal{T}) θ ∈ L 2 ( T ) such that, for every g ∈ C c ∞ ( T ) g\in C^\infty_c(\mathcal{T}) g ∈ C c ∞ ( T ) ,
∫ T θ ( ∂ t g + b ⋅ ∇ g + κ Δ g ) d t d x = 0 , (85) \int_{\mathcal{T}}\theta\left(\partial_tg+\mathbf{b}\cdot\nabla g+\kappa\Delta g\right)\,dt\,dx=0, \tag*{(85)} ∫ T θ ( ∂ t g + b ⋅ ∇ g + κ Δ g ) d t d x = 0 , (85)
which is the equation, written as ∂ t θ + ∇ ⋅ ( b θ ) − κ Δ θ = 0 \partial_t\theta+\nabla\cdot(\mathbf{b}\theta)-\kappa\Delta\theta=0 ∂ t θ + ∇ ⋅ ( b θ ) − κ Δ θ = 0 since ∇ ⋅ b = 0 \nabla\cdot\mathbf{b}=0 ∇ ⋅ b = 0 , and such that, in the coordinates y y y , U ( t , y ) : = θ ( t , X m ( t , y , t f ) ) U(t,y):=\theta(t,X_m(t,y,t_{\mathrm{f}})) U ( t , y ) := θ ( t , X m ( t , y , t f )) satisfies U ∈ L ∞ ( ( t f − 8 h m + 1 , t f ) ; L 2 ( D r t u b e ) ) U\in L^\infty((t_{\mathrm{f}}-8h_{m+1},t_{\mathrm{f}});L^2(D_{r_{\mathrm{tube}}})) U ∈ L ∞ (( t f − 8 h m + 1 , t f ) ; L 2 ( D r tube )) and ∇ y U ∈ L 2 ( ( t f − 8 h m + 1 , t f ) × D r t u b e ) \nabla_yU\in L^2((t_{\mathrm{f}}-8h_{m+1},t_{\mathrm{f}})\times D_{r_{\mathrm{tube}}}) ∇ y U ∈ L 2 (( t f − 8 h m + 1 , t f ) × D r tube ) . No lateral data are prescribed.
Theorem C (Local interior regularity). In the setting of Theorem A, with Λ 0 \Lambda_0 Λ 0 as there, let Λ ≥ Λ 0 \Lambda\ge\Lambda_0 Λ ≥ Λ 0 be an integer. There exists C 1 ( α , s , q , ζ , Λ ) ≥ 1 C_1(\alpha,s,q,\zeta,\Lambda)\ge1 C 1 ( α , s , q , ζ , Λ ) ≥ 1 with the following property. Let m ≥ 0 m\ge0 m ≥ 0 be an integer and let r t u b e r_{\mathrm{tube}} r tube satisfy
C 1 max { ε m + 1 α h m + 1 , ( h m + 1 sup K ) 1 / 2 } ≤ r t u b e ≤ 1 8 . (86) C_1\max\left\{\varepsilon_{m+1}^{\alpha}h_{m+1},(h_{m+1}\sup\mathcal{K})^{1/2}\right\}\le r_{\mathrm{tube}}\le\frac{1}{8}. \tag*{(86)} C 1 max { ε m + 1 α h m + 1 , ( h m + 1 sup K ) 1/2 } ≤ r tube ≤ 8 1 . (86) For every 0 < r o b s < 1 8 r t u b e 0<r_{\mathrm{obs}}<\frac{1}{8}r_{\mathrm{tube}} 0 < r obs < 8 1 r tube and 4 h m + 1 < t 0 < 8 h m + 1 4h_{m+1}<t_0<8h_{m+1} 4 h m + 1 < t 0 < 8 h m + 1 there exists C ( α , s , q , ζ , Λ , m , r t u b e , r o b s , t 0 ) < ∞ C(\alpha,s,q,\zeta,\Lambda,m,r_{\mathrm{tube}},r_{\mathrm{obs}},t_0)<\infty C ( α , s , q , ζ , Λ , m , r tube , r obs , t 0 ) < ∞ such that, for every t f ∈ R t_{\mathrm{f}}\in\mathbb{R} t f ∈ R , every κ ∈ K \kappa\in\mathcal{K} κ ∈ K and every local energy solution θ \theta θ of ∂ t θ + b ⋅ ∇ θ − κ Δ θ = 0 \partial_t\theta+\mathbf{b}\cdot\nabla\theta-\kappa\Delta\theta=0 ∂ t θ + b ⋅ ∇ θ − κ Δ θ = 0 in the tube T \mathcal{T} T of (84) , with U ( t , y ) : = θ ( t , X m ( t , y , t f ) ) U(t,y):=\theta(t,X_m(t,y,t_{\mathrm{f}})) U ( t , y ) := θ ( t , X m ( t , y , t f )) , the following hold. For every t ∈ [ t f − 8 h m + 1 + t 0 , t f ] t\in[t_{\mathrm{f}}-8h_{m+1}+t_0,t_{\mathrm{f}}] t ∈ [ t f − 8 h m + 1 + t 0 , t f ] , U ( t ′ , ⋅ ) U(t',\cdot) U ( t ′ , ⋅ ) converges in L 2 ( D r o b s ) L^2(D_{r_{\mathrm{obs}}}) L 2 ( D r obs ) as t ′ ↑ t t'\uparrow t t ′ ↑ t to a continuous function U ( t , ⋅ ) U(t,\cdot) U ( t , ⋅ ) , and
∥ U ∥ L ∞ ( [ t f − 8 h m + 1 + t 0 , t f ] ; C 0 , s ( D r o b s ) ) ≤ C ( ∫ T θ ( t , x ) 2 d t d x ) 1 / 2 . \|U\|_{L^\infty([t_{\mathrm{f}}-8h_{m+1}+t_0,t_{\mathrm{f}}];C^{0,s}(D_{r_{\mathrm{obs}}}))}\le C\left(\int_{\mathcal{T}}\theta(t,x)^2\,dt\,dx\right)^{1/2}. ∥ U ∥ L ∞ ([ t f − 8 h m + 1 + t 0 , t f ] ; C 0 , s ( D r obs )) ≤ C ( ∫ T θ ( t , x ) 2 d t d x ) 1/2 . The bound holds from a time after the middle of the tube up to its end. At the end, where X m ( t f , y , t f ) = y X_m(t_{\mathrm{f}},y,t_{\mathrm{f}})=y X m ( t f , y , t f ) = y , the limit of U ( t ′ , ⋅ ) U(t',\cdot) U ( t ′ , ⋅ ) defines θ ( t f , ⋅ ) : = U ( t f , ⋅ ) \theta(t_{\mathrm{f}},\cdot):=U(t_{\mathrm{f}},\cdot) θ ( t f , ⋅ ) := U ( t f , ⋅ ) on D r t u b e / 8 D_{r_{\mathrm{tube}}/8} D r tube /8 . Since K \mathcal{K} K is bounded and h m + 1 → 0 h_{m+1}\to0 h m + 1 → 0 and ε m + 1 → 0 \varepsilon_{m+1}\to0 ε m + 1 → 0 as m → ∞ m\to\infty m → ∞ , the range (86) is nonempty for all large m m m . Since X m ( t , ⋅ , t f ) X_m(t,\cdot,t_{\mathrm{f}}) X m ( t , ⋅ , t f ) and its inverse are 2-Lipschitz by (76) , Theorem C bounds θ ( t , ⋅ ) \theta(t,\cdot) θ ( t , ⋅ ) in C 0 , s C^{0,s} C 0 , s on the moving disc X m ( t , D r o b s , t f ) X_m(t,D_{r_{\mathrm{obs}}},t_{\mathrm{f}}) X m ( t , D r obs , t f ) , with 2 C 2C 2 C in place of C C C . The proof compares θ \theta θ with the solution on the whole torus from a cut-off datum, to which Theorem A applies; the localization error is exponentially small in r t u b e 2 / ( κ h m + 1 ) r_{\mathrm{tube}}^2/(\kappa h_{m+1}) r tube 2 / ( κ h m + 1 ) , and this offsets the smoothing at the radius r ≤ c c y l κ 1 / β r\le c_{\mathrm{cyl}}\kappa^{1/\beta} r ≤ c cyl κ 1/ β of Lemma 9.1 .
Lemma 9.3 (Localization). In the setting of Theorem A, with Λ 0 \Lambda_0 Λ 0 as there, let Λ ≥ Λ 0 \Lambda\ge\Lambda_0 Λ ≥ Λ 0 be an integer. There exist C 1 ≥ 1 C_1 \ge1 C 1 ≥ 1 , C < ∞ C < \infty C < ∞ and c > 0 c > 0 c > 0 , depending only on α \alpha α , s s s , q q q , ζ \zeta ζ , Λ \Lambda Λ , with the following property. Let m ≥ 0 m \ge0 m ≥ 0 , let r t u b e ≤ 1 8 r_{\mathrm{tube}} \le\frac{1}{8} r tube ≤ 8 1 satisfy C 1 ε m + 1 α h m + 1 ≤ r t u b e C_1\varepsilon_{m+1}^{\alpha}h_{m+1} \le r_{\mathrm{tube}} C 1 ε m + 1 α h m + 1 ≤ r tube , let t f ∈ R t_f \in\mathbb{R} t f ∈ R , let κ > 0 \kappa> 0 κ > 0 satisfy C 1 ( κ h m + 1 ) 1 / 2 ≤ r t u b e C_1(\kappa h_{m+1})^{1/2} \le r_{\mathrm{tube}} C 1 ( κ h m + 1 ) 1/2 ≤ r tube , let θ \theta θ be a local energy solution in T \mathbb{T} T , and let t i n : = t f − 9 2 h m + 1 t_{\mathrm{in}} := t_f-\frac{9}{2}h_{m+1} t in := t f − 2 9 h m + 1 . There exists θ i n ∈ L 2 ( T 2 ) \theta_{\mathrm{in}} \in L^2(\mathbb{T}^2) θ in ∈ L 2 ( T 2 ) with ∥ θ i n ∥ L 2 ( T 2 ) ≤ C r t u b e ( ∫ T θ ( t , x ) 2 d t d x ) 1 / 2 \|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)} \le Cr_{\mathrm{tube}}\left(\int_{\mathcal{T}}\theta(t,x)^2\,\mathrm{d}t\,\mathrm{d}x\right)^{1/2} ∥ θ in ∥ L 2 ( T 2 ) ≤ C r tube ( ∫ T θ ( t , x ) 2 d t d x ) 1/2 such that the weak solution θ p e r \theta^{\mathrm{per}} θ per of (23) on ( t i n , t i n + 5 h m + 1 ) (t_{\mathrm{in}},t_{\mathrm{in}}+5h_{m+1}) ( t in , t in + 5 h m + 1 ) with datum θ i n \theta_{\mathrm{in}} θ in satisfies, with U p e r ( t , y ) : = θ p e r ( t , X m ( t , y , t f ) ) U^{\mathrm{per}}(t,y):=\theta^{\mathrm{per}}(t,X_m(t,y,t_f)) U per ( t , y ) := θ per ( t , X m ( t , y , t f )) ,
∥ U − U p e r ∥ L ∞ ( ( t i n , t f ) ; L 2 ( D r t u b e / 8 ) ) ≤ C r t u b e ( ∫ T θ ( t , x ) 2 d t d x ) 1 / 2 exp ( − c r t u b e 2 κ h m + 1 ) . (87) \|U-U^{\mathrm{per}}\|_{L^\infty((t_{\mathrm{in}},t_f);L^2(D_{r_{\mathrm{tube}}/8}))}\le Cr_{\mathrm{tube}}\left(\int_{\mathcal{T}}\theta(t,x)^2\,\mathrm{d}t\,\mathrm{d}x\right)^{1/2}\exp\left(-\frac{cr_{\mathrm{tube}}^2}{\kappa h_{m+1}}\right). \tag*{(87)} ∥ U − U per ∥ L ∞ (( t in , t f ) ; L 2 ( D r tube /8 )) ≤ C r tube ( ∫ T θ ( t , x ) 2 d t d x ) 1/2 exp ( − κ h m + 1 c r tube 2 ) . (87)
Proof. Step 1: The equation in the coordinates y y y . For t f − 8 h m + 1 ≤ t ≤ t f t_f-8h_{m+1}\le t\le t_f t f − 8 h m + 1 ≤ t ≤ t f , changing variables by the volume-preserving map y ↦ X m ( t , y , t f ) y\mapsto X_m(t,y,t_f) y ↦ X m ( t , y , t f ) , whose time derivative is b m b_m b m , turns (85) into
∂ t U = ∇ y ⋅ ( κ M ∇ y U − U b ∗ ) , M : = ( ∇ y X m ) − 1 ( ∇ y X m ) − T , b ∗ : = ( ∇ y X m ) − 1 ( b − b m ) ( t , X m ( t , y , t f ) ) , (88) \partial_tU=\nabla_y\cdot\left(\kappa M\nabla_yU-Ub_*\right),\qquad
M:=(\nabla_yX_m)^{-1}(\nabla_yX_m)^{-\mathsf{T}},\qquad
b_*:=(\nabla_yX_m)^{-1}(b-b_m)(t,X_m(t,y,t_f)), \tag*{(88)} ∂ t U = ∇ y ⋅ ( κ M ∇ y U − U b ∗ ) , M := ( ∇ y X m ) − 1 ( ∇ y X m ) − T , b ∗ := ( ∇ y X m ) − 1 ( b − b m ) ( t , X m ( t , y , t f )) , (88)
in ( t f − 8 h m + 1 , t f ) × D r t u b e (t_f-8h_{m+1},t_f)\times D_{r_{\mathrm{tube}}} ( t f − 8 h m + 1 , t f ) × D r tube , where ∇ y X m \nabla_yX_m ∇ y X m is evaluated at ( t , y , t f ) (t,y,t_f) ( t , y , t f ) , and U p e r U^{\mathrm{per}} U per solves it in ( t i n , t f ) × R 2 (t_{\mathrm{in}},t_f)\times\mathbb{R}^2 ( t in , t f ) × R 2 . Since det ∇ y X m = 1 \det\nabla_yX_m=1 det ∇ y X m = 1 , the Piola identity (the cofactor matrix of a smooth map is divergence free) gives ∇ y ⋅ b ∗ = ( ∇ ⋅ ( b − b m ) ) ∘ X m = 0 \nabla_y\cdot b_*=(\nabla\cdot(b-b_m))\circ X_m=0 ∇ y ⋅ b ∗ = ( ∇ ⋅ ( b − b m )) ∘ X m = 0 , and, applied to ∇ θ \nabla\theta ∇ θ in place of b − b m b-b_m b − b m , it puts the equation in this divergence form. By (72) and (76) , ∣ b ∗ ∣ ≤ 2 C 0 ε m + 1 α |b_*|\le2C_0\varepsilon_{m+1}^{\alpha} ∣ b ∗ ∣ ≤ 2 C 0 ε m + 1 α , and, as det ∇ y X m = 1 \det\nabla_yX_m=1 det ∇ y X m = 1 , M M M has the eigenvalues of ( ∇ y X m ) T ∇ y X m (\nabla_yX_m)^{\mathsf{T}}\nabla_yX_m ( ∇ y X m ) T ∇ y X m , which lie in [ 1 4 , 9 4 ] \left[\frac{1}{4},\frac{9}{4}\right] [ 4 1 , 4 9 ] .
Step 2: Energy after the start of the tube. Let g ∈ C c ∞ ( D r t u b e ) g\in C^\infty_c(D_{r_{\mathrm{tube}}}) g ∈ C c ∞ ( D r tube ) with 0 ≤ g ≤ 1 0\le g\le1 0 ≤ g ≤ 1 and ∣ ∇ y g ∣ ≤ C / r t u b e |\nabla_y g|\le C/r_{\mathrm{tube}} ∣ ∇ y g ∣ ≤ C / r tube . Testing (88) with g 2 U g^2U g 2 U times a time cutoff that rises on ( t f − 8 h m + 1 , t i n ) (t_f-8h_{m+1},t_{\mathrm{in}}) ( t f − 8 h m + 1 , t in ) shows that g U gU g U is continuous in time with values in L 2 ( D r t u b e ) L^2(D_{r_{\mathrm{tube}}}) L 2 ( D r tube ) on [ t i n , t f ] [t_{\mathrm{in}},t_f] [ t in , t f ] (Lions–Magenes) and, since ∫ ( t f − 8 h m + 1 , t f ) × D r t u b e U 2 = 8 π h m + 1 r t u b e 2 ∫ T θ 2 \int_{(t_f-8h_{m+1},t_f)\times D_{r_{\mathrm{tube}}}}U^2=8\pi h_{m+1}r_{\mathrm{tube}}^2\int_{\mathcal{T}}\theta^2 ∫ ( t f − 8 h m + 1 , t f ) × D r tube U 2 = 8 π h m + 1 r tube 2 ∫ T θ 2 by volume preservation,
∥ g U ∥ L ∞ ( [ t i n , t f ] ; L 2 ( D r t u b e ) ) 2 ≤ C ( 1 h m + 1 + κ r t u b e 2 + ε m + 1 α r t u b e ) ∫ T θ 2 d t d x ≤ C r t u b e 2 ∫ T θ 2 d t d x , (89) \|gU\|_{L^\infty([t_{\mathrm{in}},t_f];L^2(D_{r_{\mathrm{tube}}}))}^2
\le C\left(\frac{1}{h_{m+1}}+\frac{\kappa}{r_{\mathrm{tube}}^2}+\frac{\varepsilon_{m+1}^{\alpha}}{r_{\mathrm{tube}}}\right)\int_{\mathcal{T}}\theta^2\,\mathrm{d}t\,\mathrm{d}x
\le Cr_{\mathrm{tube}}^2\int_{\mathcal{T}}\theta^2\,\mathrm{d}t\,\mathrm{d}x, \tag*{(89)} ∥ g U ∥ L ∞ ([ t in , t f ] ; L 2 ( D r tube )) 2 ≤ C ( h m + 1 1 + r tube 2 κ + r tube ε m + 1 α ) ∫ T θ 2 d t d x ≤ C r tube 2 ∫ T θ 2 d t d x , (89)
by the hypotheses on r t u b e r_{\mathrm{tube}} r tube and κ \kappa κ , with C 1 ≥ 1 C_1\ge1 C 1 ≥ 1 . We apply (89) with g = g 1 g=g_1 g = g 1 and with g = g 2 g=g_2 g = g 2 , where g 1 ( y ) : = H ( 3 − 4 ∣ y ∣ / r t u b e ) g_1(y):=H(3-4|y|/r_{\mathrm{tube}}) g 1 ( y ) := H ( 3 − 4∣ y ∣/ r tube ) equals 1 1 1 on D r t u b e / 2 D_{r_{\mathrm{tube}}/2} D r tube /2 and 0 0 0 outside D 3 r t u b e / 4 D_{3r_{\mathrm{tube}}/4} D 3 r tube /4 , and g 2 ( y ) : = H ( 7 − 8 ∣ y ∣ / r t u b e ) g_2(y):=H(7-8|y|/r_{\mathrm{tube}}) g 2 ( y ) := H ( 7 − 8∣ y ∣/ r tube ) equals 1 1 1 on D 3 r t u b e / 4 D_{3r_{\mathrm{tube}}/4} D 3 r tube /4 and 0 0 0 outside D 7 r t u b e / 8 D_{7r_{\mathrm{tube}}/8} D 7 r tube /8 .
Step 3: The datum. The function g 1 U ( t i n , ⋅ ) g_1U(t_{\mathrm{in}},\cdot) g 1 U ( t in , ⋅ ) is supported in D 3 r t u b e / 4 ⊂ D 1 / 8 D_{3r_{\mathrm{tube}}/4}\subset D_{1/8} D 3 r tube /4 ⊂ D 1/8 , so its translates by Z 2 \mathbb{Z}^2 Z 2 have disjoint supports. Let θ i n \theta_{\mathrm{in}} θ in be their sum, composed with x ↦ X m ( t f , x , t i n ) x\mapsto X_m(t_f,x,t_{\mathrm{in}}) x ↦ X m ( t f , x , t in ) . It is Z 2 \mathbb{Z}^2 Z 2 -periodic, U p e r ( t i n , ⋅ ) U^{\mathrm{per}}(t_{\mathrm{in}},\cdot) U per ( t in , ⋅ ) equals g 1 U ( t i n , ⋅ ) g_1U(t_{\mathrm{in}},\cdot) g 1 U ( t in , ⋅ ) on D r t u b e D_{r_{\mathrm{tube}}} D r tube , and ∥ θ i n ∥ L 2 ( T 2 ) ≤ C r t u b e ( ∫ T θ 2 ) 1 / 2 \|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)}\le Cr_{\mathrm{tube}}\left(\int_{\mathcal{T}}\theta^2\right)^{1/2} ∥ θ in ∥ L 2 ( T 2 ) ≤ C r tube ( ∫ T θ 2 ) 1/2 by (89) with g = g 1 g=g_1 g = g 1 . By (73) and the four translates of the unit cell, since X m ( t , D r t u b e , t f ) X_m(t,D_{r_{\mathrm{tube}}},t_f) X m ( t , D r tube , t f ) lies in a disc of radius 2 r t u b e < 1 2 2r_{\mathrm{tube}}<\frac{1}{2} 2 r tube < 2 1 by (76) , ∥ U p e r ∥ L ∞ ( ( t i n , t f ) ; L 2 ( D r t u b e ) ) 2 ≤ 4 ∥ θ i n ∥ L 2 ( T 2 ) 2 ≤ C r t u b e 2 ∫ T θ 2 \|U^{\mathrm{per}}\|_{L^\infty((t_{\mathrm{in}},t_f);L^2(D_{r_{\mathrm{tube}}}))}^2\le4\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)}^2\le Cr_{\mathrm{tube}}^2\int_{\mathcal{T}}\theta^2 ∥ U per ∥ L ∞ (( t in , t f ) ; L 2 ( D r tube )) 2 ≤ 4∥ θ in ∥ L 2 ( T 2 ) 2 ≤ C r tube 2 ∫ T θ 2 .
Step 4: Exponential separation. We use the energy estimates with an exponential weight of the form used in [22 Section 2] . For a length r ′ > 0 r'>0 r ′ > 0 , the weight
exp ( − r t u b e 8 r ′ H ( 4 ∣ y ∣ r t u b e − 1 2 ) ) \exp\left(-\frac{r_{\mathrm{tube}}}{8r'}H\left(\frac{4|y|}{r_{\mathrm{tube}}}-\frac{1}{2}\right)\right) exp ( − 8 r ′ r tube H ( r tube 4∣ y ∣ − 2 1 ) )
equals 1 1 1 on D r t u b e / 8 D_{r_{\mathrm{tube}}/8} D r tube /8 , equals exp ( − r t u b e / ( 8 r ′ ) ) \exp(-r_{\mathrm{tube}}/(8r')) exp ( − r tube / ( 8 r ′ )) outside D 3 r t u b e / 8 D_{3r_{\mathrm{tube}}/8} D 3 r tube /8 , and the gradient of its logarithm is at most C / r ′ C/r' C / r ′ . Testing the equation of U − U p e r U-U^{\mathrm{per}} U − U per shows that the integral over D r t u b e D_{r_{\mathrm{tube}}} D r tube of the square of this weight times g 2 ( y ) 2 ( U − U p e r ) ( t , y ) 2 g_2(y)^2(U-U^{\mathrm{per}})(t,y)^2 g 2 ( y ) 2 ( U − U per ) ( t , y ) 2 grows at most at the rate C ( κ / r ′ 2 + ε m + 1 α / r ′ ) C(\kappa/r'^2+\varepsilon_{m+1}^{\alpha}/r') C ( κ / r ′2 + ε m + 1 α / r ′ ) : the derivatives of the weight cost 1 / r ′ 1/r' 1/ r ′ against the diffusion and the divergence-free drift, and Young’s inequality absorbs the rest into the dissipation. The other terms contain derivatives of g 2 g_2 g 2 , which vanish on D 3 r t u b e / 4 D_{3r_{\mathrm{tube}}/4} D 3 r tube /4 , so they carry the factor exp ( − r t u b e / ( 4 r ′ ) ) \exp(-r_{\mathrm{tube}}/(4r')) exp ( − r tube / ( 4 r ′ )) , the square of the weight outside
Vanishing diffusivity: Corollary D
D r t u b e / 8 D_{r_{\mathrm{tube}}/8} D r tube /8 . After Young’s inequality, which keeps g 2 g_2 g 2 in front of the gradient of U − U p e r U-U^{\mathrm{per}} U − U per , their integral in time is at most C ( κ / r t u b e 2 + ε m + 1 α / r t u b e ) C(\kappa/r_{\mathrm{tube}}^2+\varepsilon_{m+1}^{\alpha}/r_{\mathrm{tube}}) C ( κ / r tube 2 + ε m + 1 α / r tube ) times the integral of U 2 + ( U p e r ) 2 U^2+(U^{\mathrm{per}})^2 U 2 + ( U per ) 2 over ( t i n , t f ) × D r t u b e (t_{\mathrm{in}},t_f)\times D_{r_{\mathrm{tube}}} ( t in , t f ) × D r tube , which is at most C h m + 1 r t u b e − 2 ∫ T θ 2 Ch_{m+1}r_{\mathrm{tube}}^{-2}\int_T\theta^2 C h m + 1 r tube − 2 ∫ T θ 2 by Steps 2 and 3; and ( κ / r t u b e 2 + ε m + 1 α / r t u b e ) h m + 1 ≤ 2 (\kappa/r_{\mathrm{tube}}^2+\varepsilon_{m+1}^{\alpha}/r_{\mathrm{tube}})h_{m+1}\le2 ( κ / r tube 2 + ε m + 1 α / r tube ) h m + 1 ≤ 2 by the hypotheses. The initial value is bounded by (89) with g = g 2 g=g_2 g = g 2 , since at time t i n t_{\mathrm{in}} t in the difference U − U p e r = ( 1 − g 1 ) U ( t i n , ⋅ ) U-U^{\mathrm{per}}=(1-g_1)U(t_{\mathrm{in}},\cdot) U − U per = ( 1 − g 1 ) U ( t in , ⋅ ) vanishes on D r t u b e / 2 D_{r_{\mathrm{tube}}/2} D r tube /2 . Since t f − t i n = 9 2 h m + 1 t_f-t_{\mathrm{in}}=\frac{9}{2}h_{m+1} t f − t in = 2 9 h m + 1 , Grönwall’s inequality gives, for t i n < t < t f t_{\mathrm{in}}<t<t_f t in < t < t f ,
∫ D r t u b e / 8 ( U − U p e r ) ( t , y ) 2 d y ≤ C r t u b e 2 ∫ T θ ( t ′ , x ) 2 d t ′ d x ⋅ exp ( C ( κ / r ′ 2 + ε m + 1 α / r ′ ) h m + 1 − r t u b e 4 r ′ ) . \int_{D_{r_{\mathrm{tube}}/8}}(U-U^{\mathrm{per}})(t,y)^2\,dy\le Cr_{\mathrm{tube}}^2\int_T\theta(t',x)^2\,dt'\,dx\cdot\exp\left(C(\kappa/r'^2+\varepsilon_{m+1}^{\alpha}/r')h_{m+1}-\frac{r_{\mathrm{tube}}}{4r'}\right). ∫ D r tube /8 ( U − U per ) ( t , y ) 2 d y ≤ C r tube 2 ∫ T θ ( t ′ , x ) 2 d t ′ d x ⋅ exp ( C ( κ / r ′2 + ε m + 1 α / r ′ ) h m + 1 − 4 r ′ r tube ) .
With r ′ : = C ′ κ h m + 1 / r t u b e r':=C'\kappa h_{m+1}/r_{\mathrm{tube}} r ′ := C ′ κ h m + 1 / r tube , and C ′ C' C ′ and then C 1 C_1 C 1 large, the exponent is at most − c r t u b e 2 / ( κ h m + 1 ) -cr_{\mathrm{tube}}^2/(\kappa h_{m+1}) − c r tube 2 / ( κ h m + 1 ) , using C 1 ε m + 1 α h m + 1 ≤ r t u b e C_1\varepsilon_{m+1}^{\alpha}h_{m+1}\le r_{\mathrm{tube}} C 1 ε m + 1 α h m + 1 ≤ r tube for the drift term. This proves (87) .
Proof of Theorem C. Let C 1 C_1 C 1 be that of Lemma 9.3 9.3 9.3 ; for κ ∈ K \kappa\in\mathcal{K} κ ∈ K , (86) gives its hypotheses. Fix r o b s r_{\mathrm{obs}} r obs , t 0 t_0 t 0 , t f t_f t f , κ \kappa κ and θ \theta θ , and let t i n t_{\mathrm{in}} t in , θ i n \theta_{\mathrm{in}} θ in and θ p e r \theta^{\mathrm{per}} θ per be as in Lemma 9.3 9.3 9.3 .
Step 1: The comparison solution. By Theorem A on ( t i n , t i n + 5 h m + 1 ) (t_{\mathrm{in}},t_{\mathrm{in}}+5h_{m+1}) ( t in , t in + 5 h m + 1 ) , with the waiting time t 0 − 4 h m + 1 ∈ ( 0 , 4 h m + 1 ) t_0-4h_{m+1}\in(0,4h_{m+1}) t 0 − 4 h m + 1 ∈ ( 0 , 4 h m + 1 ) , we obtain
∥ θ p e r ∥ L ∞ ( [ t i n + t 0 − 4 h m + 1 , t i n + 5 h m + 1 ] ; C 0 , s ( T 2 ) ) ≤ C ∥ θ i n ∥ L 2 ( T 2 ) ≤ C r t u b e ( ∫ T θ ( t , x ) 2 d t d x ) 1 / 2 , \|\theta^{\mathrm{per}}\|_{L^\infty([t_{\mathrm{in}}+t_0-4h_{m+1},t_{\mathrm{in}}+5h_{m+1}];C^{0,s}(\mathbb{T}^2))}\le C\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)}\le Cr_{\mathrm{tube}}\left(\int_T\theta(t,x)^2\,dt\,dx\right)^{1/2}, ∥ θ per ∥ L ∞ ([ t in + t 0 − 4 h m + 1 , t in + 5 h m + 1 ] ; C 0 , s ( T 2 )) ≤ C ∥ θ in ∥ L 2 ( T 2 ) ≤ C r tube ( ∫ T θ ( t , x ) 2 d t d x ) 1/2 ,
at every time of this interval, by (9.2′), which contains [ t f − 8 h m + 1 + t 0 , t f ] [t_f-8h_{m+1}+t_0,t_f] [ t f − 8 h m + 1 + t 0 , t f ] .
Step 2: The difference. In the open set O : = { ( t , X m ( t , y , t f ) ) : t i n < t < t f , ∣ y ∣ < 1 8 r t u b e } \mathcal{O}:=\{(t,X_m(t,y,t_f)):t_{\mathrm{in}}<t<t_f,\ |y|<\frac{1}{8}r_{\mathrm{tube}}\} O := {( t , X m ( t , y , t f )) : t in < t < t f , ∣ y ∣ < 8 1 r tube } , θ − θ p e r \theta-\theta^{\mathrm{per}} θ − θ per is a local weak solution (the weak gradients are those of U U U and U p e r U^{\mathrm{per}} U per , in the coordinates y y y ). By volume preservation and (87) , ∫ O t ( θ − θ p e r ) ( t ′ , ⋅ ) 2 ≤ Ξ 2 \int_{\mathcal{O}_t}(\theta-\theta^{\mathrm{per}})(t',\cdot)^2\le\Xi^2 ∫ O t ( θ − θ per ) ( t ′ , ⋅ ) 2 ≤ Ξ 2 for almost every t ′ ∈ ( t i n , t f ) t'\in(t_{\mathrm{in}},t_f) t ′ ∈ ( t in , t f ) , with
Ξ : = C r t u b e ( ∫ T θ 2 ) 1 / 2 exp ( − c r t u b e 2 κ h m + 1 ) . \Xi:=Cr_{\mathrm{tube}}\left(\int_T\theta^2\right)^{1/2}\exp\left(-\frac{cr_{\mathrm{tube}}^2}{\kappa h_{m+1}}\right). Ξ := C r tube ( ∫ T θ 2 ) 1/2 exp ( − κ h m + 1 c r tube 2 ) .
Step 3: Moving cylinders stay in O \mathcal{O} O . Let
r : = min { 1 8 , 1 8 r t u b e − r o b s 8 , c c y l κ 1 / β , ( κ min { t 0 − 4 h m + 1 2 , 1 8 r t u b e − r o b s 24 ( 1 + C 0 ε m + 1 α ) } ) 1 / 2 } , r:=\min\left\{\frac{1}{8},\frac{\frac{1}{8}r_{\mathrm{tube}}-r_{\mathrm{obs}}}{8},c_{\mathrm{cyl}}\kappa^{1/\beta},\left(\kappa\min\left\{\frac{t_0-4h_{m+1}}{2},\frac{\frac{1}{8}r_{\mathrm{tube}}-r_{\mathrm{obs}}}{24(1+C_0\varepsilon_{m+1}^{\alpha})}\right\}\right)^{1/2}\right\}, r := min { 8 1 , 8 8 1 r tube − r obs , c cyl κ 1/ β , ( κ min { 2 t 0 − 4 h m + 1 , 24 ( 1 + C 0 ε m + 1 α ) 8 1 r tube − r obs } ) 1/2 } ,
so that
2 r + 6 C 0 ε m + 1 α r 2 κ ≤ 1 2 ( 1 8 r t u b e − r o b s ) , r 2 κ ≤ t 0 − 4 h m + 1 2 . (90) 2r+6C_0\varepsilon_{m+1}^{\alpha}\frac{r^2}{\kappa}\le\frac{1}{2}\left(\frac{1}{8}r_{\mathrm{tube}}-r_{\mathrm{obs}}\right),\qquad\frac{r^2}{\kappa}\le\frac{t_0-4h_{m+1}}{2}. \tag*{(90)} 2 r + 6 C 0 ε m + 1 α κ r 2 ≤ 2 1 ( 8 1 r tube − r obs ) , κ r 2 ≤ 2 t 0 − 4 h m + 1 . (90)
Let t ∈ [ t f − 8 h m + 1 + t 0 , t f ) t\in[t_f-8h_{m+1}+t_0,t_f) t ∈ [ t f − 8 h m + 1 + t 0 , t f ) , E t : = X m ( t , D r o b s , t f ) E_t:=X_m(t,D_{r_{\mathrm{obs}}},t_f) E t := X m ( t , D r obs , t f ) , and let X X X be a trajectory of b b b on [ t − r 2 / κ , t ] [t-r^2/\kappa,t] [ t − r 2 / κ , t ] ending in E t E_t E t . The second bound in (90) gives t − r 2 / κ ≥ t f − 6 h m + 1 + 1 2 t 0 > t i n t-r^2/\kappa\ge t_f-6h_{m+1}+\frac{1}{2}t_0>t_{\mathrm{in}} t − r 2 / κ ≥ t f − 6 h m + 1 + 2 1 t 0 > t in . In the coordinates y y y , the point X ( t ′ ) X(t') X ( t ′ ) moves with the velocity b ∗ b_* b ∗ of (88) , which equals ∇ x X m ( t f , x , t ′ ) ( b − b m ) ( t ′ , x ) \nabla_xX_m(t_f,x,t')(b-b_m)(t',x) ∇ x X m ( t f , x , t ′ ) ( b − b m ) ( t ′ , x ) at x = X ( t ′ ) x=X(t') x = X ( t ′ ) and has size at most 3 2 C 0 ε m + 1 α \frac{3}{2}C_0\varepsilon_{m+1}^{\alpha} 2 3 C 0 ε m + 1 α by (72) and (76) , so it stays within 3 2 C 0 ε m + 1 α r 2 / κ \frac{3}{2}C_0\varepsilon_{m+1}^{\alpha}r^2/\kappa 2 3 C 0 ε m + 1 α r 2 / κ of D r o b s D_{r_{\mathrm{obs}}} D r obs , a quarter of the term 6 C 0 ε m + 1 α r 2 / κ 6C_0\varepsilon_{m+1}^{\alpha}r^2/\kappa 6 C 0 ε m + 1 α r 2 / κ allowed by (90) ; and x ↦ X m ( t f , x , t ′ ) x\mapsto X_m(t_f,x,t') x ↦ X m ( t f , x , t ′ ) is 2-Lipschitz. By the first bound in (90) , the moving cylinder of radius r r r around X X X lies in O \mathcal{O} O .
Step 4: The bound before the end of the tube. Lemma 9.1 9.1 9.1 applies to θ − θ p e r \theta-\theta^{\mathrm{per}} θ − θ per in O \mathcal{O} O at every t ∈ [ t f − 8 h m + 1 + t 0 , t f ) t\in[t_f-8h_{m+1}+t_0,t_f) t ∈ [ t f − 8 h m + 1 + t 0 , t f ) , with E = E t E=E_t E = E t and Ξ \Xi Ξ of Step 2, and gives ∥ ( θ − θ p e r ) ( t , ⋅ ) ∥ C 0 , s ( E t ) ≤ C r − 1 − s Ξ \|(\theta-\theta^{\mathrm{per}})(t,\cdot)\|_{C^{0,s}(E_t)}\le Cr^{-1-s}\Xi ∥ ( θ − θ per ) ( t , ⋅ ) ∥ C 0 , s ( E t ) ≤ C r − 1 − s Ξ . The power of κ \kappa κ in r − 1 − s r^{-1-s} r − 1 − s is absorbed by the exponential in Ξ \Xi Ξ , since κ ↦ κ − ( 1 + s ) / β exp ( − c r t u b e 2 / ( κ h m + 1 ) ) \kappa\mapsto\kappa^{-(1+s)/\beta}\exp(-cr_{\mathrm{tube}}^2/(\kappa h_{m+1})) κ ↦ κ − ( 1 + s ) / β exp ( − c r tube 2 / ( κ h m + 1 )) and κ ↦ κ − ( 1 + s ) / 2 exp ( − c r t u b e 2 / ( κ h m + 1 ) ) \kappa\mapsto\kappa^{-(1+s)/2}\exp(-cr_{\mathrm{tube}}^2/(\kappa h_{m+1})) κ ↦ κ − ( 1 + s ) /2 exp ( − c r tube 2 / ( κ h m + 1 )) are bounded on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) ; r − 1 − s Ξ ≤ C ( ∫ T θ 2 ) 1 / 2 r^{-1-s}\Xi\le C(\int_T\theta^2)^{1/2} r − 1 − s Ξ ≤ C ( ∫ T θ 2 ) 1/2 , with C C C depending on m m m , r t u b e r_{\mathrm{tube}} r tube , r o b s r_{\mathrm{obs}} r obs and t 0 t_0 t 0 . With Step 1, θ ( t , ⋅ ) \theta(t,\cdot) θ ( t , ⋅ ) is continuous on E t E_t E t and ∥ θ ( t , ⋅ ) ∥ C 0 , s ( E t ) ≤ C ( ∫ T θ 2 ) 1 / 2 \|\theta(t,\cdot)\|_{C^{0,s}(E_t)}\le C(\int_T\theta^2)^{1/2} ∥ θ ( t , ⋅ ) ∥ C 0 , s ( E t ) ≤ C ( ∫ T θ 2 ) 1/2 . Since y ↦ X m ( t , y , t f ) y\mapsto X_m(t,y,t_f) y ↦ X m ( t , y , t f ) is 2-Lipschitz by (76) , U ( t , ⋅ ) U(t,\cdot) U ( t , ⋅ ) is continuous and ∥ U ( t , ⋅ ) ∥ C 0 , s ( D r o b s ) ≤ 2 s ∥ θ ( t , ⋅ ) ∥ C 0 , s ( E t ) \|U(t,\cdot)\|_{C^{0,s}(D_{r_{\mathrm{obs}}})}\le2^s\|\theta(t,\cdot)\|_{C^{0,s}(E_t)} ∥ U ( t , ⋅ ) ∥ C 0 , s ( D r obs ) ≤ 2 s ∥ θ ( t , ⋅ ) ∥ C 0 , s ( E t ) ; the convergence of U ( t ′ , ⋅ ) U(t',\cdot) U ( t ′ , ⋅ ) to U ( t , ⋅ ) U(t,\cdot) U ( t , ⋅ ) in L 2 ( D r o b s ) L^2(D_{r_{\mathrm{obs}}}) L 2 ( D r obs ) as t ′ ↑ t t'\uparrow t t ′ ↑ t follows from that of Lemma 9.1 9.1 9.1 for θ − θ p e r \theta-\theta^{\mathrm{per}} θ − θ per , the continuity in time of θ p e r \theta^{\mathrm{per}} θ per and the continuity of X m X_m X m .
Step 5: The end of the tube. By Step 2 of the proof of Lemma 9.3 9.3 9.3 , U ( t ′ , ⋅ ) U(t',\cdot) U ( t ′ , ⋅ ) converges in L 2 ( D r o b s ) L^2(D_{r_{\mathrm{obs}}}) L 2 ( D r obs ) as t ′ ↑ t f t'\uparrow t_f t ′ ↑ t f ; by the uniform bound of Step 4, the limit is continuous and satisfies the same bound. ∎
Vanishing diffusivity: Corollary D
As ∇ ⋅ b = 0 \nabla\cdot\mathbf{b}=0 ∇ ⋅ b = 0 , the transport equation reads ∂ t θ + ∇ ⋅ ( b θ ) = 0 \partial_t\theta+\nabla\cdot(\mathbf{b}\theta)=0 ∂ t θ + ∇ ⋅ ( b θ ) = 0 , meaningful for continuous θ \theta θ .
Corollary D (Vanishing-diffusivity limits). In the setting of Theorem A, with Λ 0 \Lambda_0 Λ 0 as there, let Λ ≥ Λ 0 \Lambda\geq\Lambda_0 Λ ≥ Λ 0 be an integer.
(a) There exist constants C ( α , s , q , ζ , Λ , t 0 , T ) C(\alpha,s,q,\zeta,\Lambda,t_0,T) C ( α , s , q , ζ , Λ , t 0 , T ) , for 0 < t 0 < T 0<t_0<T 0 < t 0 < T , with the following property. Let t i n ∈ R t_{\mathrm{in}}\in\mathbb{R} t in ∈ R , let K ′ ⊂ K K'\subset K K ′ ⊂ K be an infinite subset with 0 as its only accumulation point, and, for κ ∈ K ′ \kappa\in K' κ ∈ K ′ , let θ i n κ ∈ L 2 ( T 2 ) \theta_{\mathrm{in}}^\kappa\in L^2(\mathbb{T}^2) θ in κ ∈ L 2 ( T 2 ) with A 0 : = sup κ ∈ K ′ ∥ θ i n κ ∥ L 2 ( T 2 ) < ∞ A_0:=\sup_{\kappa\in K'}\lVert\theta_{\mathrm{in}}^\kappa\rVert_{L^2(\mathbb{T}^2)}<\infty A 0 := sup κ ∈ K ′ ∥ θ in κ ∥ L 2 ( T 2 ) < ∞ , let θ κ \theta^\kappa θ κ be the weak solution of
{ ∂ t θ κ + b ⋅ ∇ θ κ − κ Δ θ κ = 0 in ( t i n , ∞ ) × T 2 , θ κ ( t i n , ⋅ ) = θ i n κ on T 2 , \begin{cases}
\partial_t\theta^\kappa+\mathbf{b}\cdot\nabla\theta^\kappa-\kappa\Delta\theta^\kappa=0 & \text{in }(t_{\mathrm{in}},\infty)\times\mathbb{T}^2,\\
\theta^\kappa(t_{\mathrm{in}},\cdot)=\theta_{\mathrm{in}}^\kappa& \text{on }\mathbb{T}^2,
\end{cases} { ∂ t θ κ + b ⋅ ∇ θ κ − κ Δ θ κ = 0 θ κ ( t in , ⋅ ) = θ in κ in ( t in , ∞ ) × T 2 , on T 2 , and, with M M M such that κ ∈ K M \kappa\in K_M κ ∈ K M , let θ ~ κ \widetilde{\theta}^\kappa θ κ be the weak solution of the same problem with b M \mathbf{b}_M b M in place of b \mathbf{b} b . Then every θ κ \theta^\kappa θ κ is continuous on ( t i n , ∞ ) × T 2 (t_{\mathrm{in}},\infty)\times\mathbb{T}^2 ( t in , ∞ ) × T 2 , and there exist an infinite subset K ′ ′ ⊂ K ′ K''\subset K' K ′′ ⊂ K ′ , θ i n , ∗ ∈ L 2 ( T 2 ) \theta_{\mathrm{in},*}\in L^2(\mathbb{T}^2) θ in , ∗ ∈ L 2 ( T 2 ) and a continuous function θ ∗ \theta_* θ ∗ on ( t i n , ∞ ) × T 2 (t_{\mathrm{in}},\infty)\times\mathbb{T}^2 ( t in , ∞ ) × T 2 such that:
(i) θ i n κ ⇀ θ i n , ∗ \theta_{\mathrm{in}}^\kappa\rightharpoonup\theta_{\mathrm{in},*} θ in κ ⇀ θ in , ∗ weakly in L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) , and θ κ → θ ∗ \theta^\kappa\to\theta_* θ κ → θ ∗ locally uniformly on ( t i n , ∞ ) × T 2 (t_{\mathrm{in}},\infty)\times\mathbb{T}^2 ( t in , ∞ ) × T 2 , as κ → 0 \kappa\to0 κ → 0 in K ′ ′ K'' K ′′ ;
(ii) ∂ t θ ∗ + ∇ ⋅ ( b θ ∗ ) = 0 \partial_t\theta_*+\nabla\cdot(\mathbf{b}\theta_*)=0 ∂ t θ ∗ + ∇ ⋅ ( b θ ∗ ) = 0 in the sense of distributions on ( t i n , ∞ ) × R 2 (t_{\mathrm{in}},\infty)\times\mathbb{R}^2 ( t in , ∞ ) × R 2 , and θ ∗ ( t , ⋅ ) ⇀ θ i n , ∗ \theta_*(t,\cdot)\rightharpoonup\theta_{\mathrm{in},*} θ ∗ ( t , ⋅ ) ⇀ θ in , ∗ weakly in L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) as t ↓ t i n t\downarrow t_{\mathrm{in}} t ↓ t in ;
(iii) for all such t 0 t_0 t 0 and T T T , ∥ θ ∗ ∥ L ∞ ( [ t i n + t 0 , t i n + T ] ; C 0 , s ( T 2 ) ) ≤ C ( t 0 , T ) A 0 \lVert\theta_*\rVert_{L^\infty([t_{\mathrm{in}}+t_0,t_{\mathrm{in}}+T];C^{0,s}(\mathbb{T}^2))}\leq C(t_0,T)A_0 ∥ θ ∗ ∥ L ∞ ([ t in + t 0 , t in + T ] ; C 0 , s ( T 2 )) ≤ C ( t 0 , T ) A 0 , and θ ∗ \theta_* θ ∗ is Hölder continuous in time on [ t i n + t 0 , t i n + T ] × T 2 [t_{\mathrm{in}}+t_0,t_{\mathrm{in}}+T]\times\mathbb{T}^2 [ t in + t 0 , t in + T ] × T 2 , uniformly in space, with exponent s / ( 1 + s ) s/(1+s) s / ( 1 + s ) ;
(iv) θ κ − θ ~ κ → 0 \theta^\kappa-\widetilde{\theta}^\kappa\to0 θ κ − θ κ → 0 in the sense of distributions on ( t i n , ∞ ) × R 2 (t_{\mathrm{in}},\infty)\times\mathbb{R}^2 ( t in , ∞ ) × R 2 as κ → 0 \kappa\to0 κ → 0 in K ′ K' K ′ .
(b) Let C 1 C_1 C 1 , m m m , r t u b e r_{\mathrm{tube}} r tube , t f t_f t f and the tube T \mathcal{T} T be as in Theorem C, and let T ′ : = { ( t , X m ( t , y , t f ) ) : t f − 4 h m + 1 < t ≤ t f , ∣ y ∣ < 1 8 r t u b e } \mathcal{T}' := \{(t,X_m(t,y,t_f)):t_f-4h_{m+1}<t\leq t_f,\ |y|<\frac{1}{8}r_{\mathrm{tube}}\} T ′ := {( t , X m ( t , y , t f )) : t f − 4 h m + 1 < t ≤ t f , ∣ y ∣ < 8 1 r tube } . There exist constants C ( α , s , q , ζ , Λ , m , r t u b e , r o b s , t 0 ) C(\alpha,s,q,\zeta,\Lambda,m,r_{\mathrm{tube}},r_{\mathrm{obs}},t_0) C ( α , s , q , ζ , Λ , m , r tube , r obs , t 0 ) , for 0 < r o b s < 1 8 r t u b e 0<r_{\mathrm{obs}}<\frac{1}{8}r_{\mathrm{tube}} 0 < r obs < 8 1 r tube and 4 h m + 1 < t 0 < 8 h m + 1 4h_{m+1}<t_0<8h_{m+1} 4 h m + 1 < t 0 < 8 h m + 1 , with the following property. Let K ′ ⊂ K K'\subset K K ′ ⊂ K be an infinite subset with 0 as its only accumulation point, and let θ κ \theta^\kappa θ κ , κ ∈ K ′ \kappa\in K' κ ∈ K ′ , be local energy solutions in T \mathcal{T} T with sup κ ∈ K ′ ∫ T ∣ θ κ ( t , x ) ∣ 2 d t d x < ∞ \sup_{\kappa\in K'}\int_{\mathcal{T}}|\theta^\kappa(t,x)|^2\,dt\,dx<\infty sup κ ∈ K ′ ∫ T ∣ θ κ ( t , x ) ∣ 2 d t d x < ∞ . Then every θ κ \theta^\kappa θ κ is continuous on T ′ \mathcal{T}' T ′ , and there exist an infinite subset K ′ ′ ⊂ K ′ K''\subset K' K ′′ ⊂ K ′ and a continuous function θ ∗ \theta_* θ ∗ on T ′ \mathcal{T}' T ′ such that θ κ → θ ∗ \theta^\kappa\to\theta_* θ κ → θ ∗ uniformly on the compact subsets of T ′ \mathcal{T}' T ′ as κ → 0 \kappa\to0 κ → 0 in K ′ ′ K'' K ′′ ; the function θ ∗ \theta_* θ ∗ satisfies ∂ t θ ∗ + ∇ ⋅ ( b θ ∗ ) = 0 \partial_t\theta_*+\nabla\cdot(\mathbf{b}\theta_*)=0 ∂ t θ ∗ + ∇ ⋅ ( b θ ∗ ) = 0 in the sense of distributions in the interior of T ′ \mathcal{T}' T ′ and, for every such r o b s r_{\mathrm{obs}} r obs and t 0 t_0 t 0 , with U ∗ ( t , y ) : = θ ∗ ( t , X m ( t , y , t f ) ) U_*(t,y):=\theta_*(t,X_m(t,y,t_f)) U ∗ ( t , y ) := θ ∗ ( t , X m ( t , y , t f )) ,
∥ U ∗ ∥ L ∞ ( [ t f − 8 h m + 1 + t 0 , t f ] ; C 0 , s ( D r o b s ) ) ≤ C sup κ ∈ K ′ ( ∫ T ∣ θ κ ( t , x ) ∣ 2 d t d x ) 1 / 2 . \lVert U_*\rVert_{L^\infty([t_f-8h_{m+1}+t_0,t_f];C^{0,s}(D_{r_{\mathrm{obs}}}))}\leq C\sup_{\kappa\in K'}\left(\int_{\mathcal{T}}|\theta^\kappa(t,x)|^2\,dt\,dx\right)^{1/2}. ∥ U ∗ ∥ L ∞ ([ t f − 8 h m + 1 + t 0 , t f ] ; C 0 , s ( D r obs )) ≤ C κ ∈ K ′ sup ( ∫ T ∣ θ κ ( t , x ) ∣ 2 d t d x ) 1/2 .
Proof. (a), compactness. For 0 < t 0 < T 0<t_0<T 0 < t 0 < T and every t ∈ [ t i n + t 0 , t i n + T ] t\in[t_{\mathrm{in}}+t_0,t_{\mathrm{in}}+T] t ∈ [ t in + t 0 , t in + T ] , Theorem A, with (9.2'), bounds θ κ ( t , ⋅ ) \theta^\kappa(t,\cdot) θ κ ( t , ⋅ ) in C 0 , s ( T 2 ) C^{0,s}(\mathbb{T}^2) C 0 , s ( T 2 ) by C A 0 CA_0 C A 0 . Let x ∈ T 2 x\in\mathbb{T}^2 x ∈ T 2 , t , t ′ ∈ [ t i n + t 0 , t i n + T ] t,t'\in[t_{\mathrm{in}}+t_0,t_{\mathrm{in}}+T] t , t ′ ∈ [ t in + t 0 , t in + T ] and 0 < r ≤ 1 0<r\leq1 0 < r ≤ 1 . We average θ κ ( t , ⋅ ) \theta^\kappa(t,\cdot) θ κ ( t , ⋅ ) against r − 2 u ( r − 1 ( x − ⋅ ) ) r^{-2}u(r^{-1}(x-\cdot)) r − 2 u ( r − 1 ( x − ⋅ )) , with u u u the function fixed at the start of § 2.4 . The average differs from θ κ ( t , x ) \theta^\kappa(t,x) θ κ ( t , x ) by at most C r s A 0 Cr^sA_0 C r s A 0 ; testing the equation with this function and using the bound of θ κ \theta^\kappa θ κ in L ∞ L^\infty L ∞ at the times between t ′ t' t ′ and t t t , we obtain
∣ ∫ R 2 ( θ κ ( t , x ′ ) − θ κ ( t ′ , x ′ ) ) r − 2 u ( r − 1 ( x − x ′ ) ) d x ′ ∣ ≤ C ∣ t − t ′ ∣ ( ∥ b ∥ L ∞ ( R × T 2 ) r − 1 + κ r − 2 ) A 0 . \left|\int_{\mathbb{R}^2}\left(\theta^\kappa(t,x')-\theta^\kappa(t',x')\right)r^{-2}u\left(r^{-1}(x-x')\right)\,dx'\right|\leq C|t-t'|\left(\lVert\mathbf{b}\rVert_{L^\infty(\mathbb{R}\times\mathbb{T}^2)}r^{-1}+\kappa r^{-2}\right)A_0. ∫ R 2 ( θ κ ( t , x ′ ) − θ κ ( t ′ , x ′ ) ) r − 2 u ( r − 1 ( x − x ′ ) ) d x ′ ≤ C ∣ t − t ′ ∣ ( ∥ b ∥ L ∞ ( R × T 2 ) r − 1 + κ r − 2 ) A 0 .
Since K ′ K' K ′ is bounded, for ∣ t − t ′ ∣ ≤ 1 |t-t'|\leq1 ∣ t − t ′ ∣ ≤ 1 the choice r = ∣ t − t ′ ∣ 1 / ( s + 2 ) r=|t-t'|^{1/(s+2)} r = ∣ t − t ′ ∣ 1/ ( s + 2 ) gives a modulus C ∣ t − t ′ ∣ s / ( s + 2 ) C|t-t'|^{s/(s+2)} C ∣ t − t ′ ∣ s / ( s + 2 ) in time, uniformly in κ \kappa κ ; for ∣ t − t ′ ∣ > 1 |t-t'|>1 ∣ t − t ′ ∣ > 1 the bound C A 0 CA_0 C A 0 on the slices gives it. So every θ κ \theta^\kappa θ κ is continuous, and the θ κ \theta^\kappa θ κ are bounded and equicontinuous on every [ t i n + t 0 , t i n + T ] × T 2 [t_{\mathrm{in}}+t_0,t_{\mathrm{in}}+T]\times\mathbb{T}^2 [ t in + t 0 , t in + T ] × T 2 ; the theorem of Arzelà and Ascoli with a diagonal argument, and the weak compactness of the balls of L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) , give (i).
( a ) (a) ( a ) , the limit . For g ∈ C c ∞ ( ( t i n , ∞ ) × R 2 ) g\in C^\infty_c((t_{\mathrm{in}},\infty)\times\mathbb{R}^2) g ∈ C c ∞ (( t in , ∞ ) × R 2 ) , ∫ θ κ ( ∂ t g + b ⋅ ∇ g + κ Δ g ) = 0 \int\theta^\kappa(\partial_tg+\mathbf{b}\cdot\nabla g+\kappa\Delta g)=0 ∫ θ κ ( ∂ t g + b ⋅ ∇ g + κ Δ g ) = 0 ; since b κ θ → b θ ∗ \mathbf{b}^\theta_\kappa\to\mathbf{b}^{\theta_*} b κ θ → b θ ∗ uniformly on the support of g g g as κ → 0 \kappa\to0 κ → 0 in K ′ ′ \mathcal{K}'' K ′′ , the limit is (ii). The bound of (iii) passes to the limit pointwise, and the computation above with κ = 0 \kappa=0 κ = 0 , for θ ∗ \theta_* θ ∗ , and r = ∣ t − t ′ ∣ 1 / ( 1 + s ) r=|t-t'|^{1/(1+s)} r = ∣ t − t ′ ∣ 1/ ( 1 + s ) for ∣ t − t ′ ∣ ≤ 1 |t-t'|\le1 ∣ t − t ′ ∣ ≤ 1 give the exponent s / ( 1 + s ) s/(1+s) s / ( 1 + s ) . For the initial trace, let g g g be smooth and Z 2 \mathbb{Z}^2 Z 2 -periodic. By the weak form of the equation, the Cauchy–Schwarz inequality and (73) , for t > t i n t>t_{\mathrm{in}} t > t in ,
∣ ∫ T 2 ( θ κ ( t , x ) − θ i n κ ( x ) ) g ( x ) d x ∣ ≤ ( t − t i n ) A 0 ∥ b ⋅ ∇ g + κ Δ g ∥ L ∞ ( ( t i n , t ) × T 2 ) . \left|\int_{\mathbb{T}^2}(\theta^\kappa(t,x)-\theta^\kappa_{\mathrm{in}}(x))g(x)\,\mathrm{d}x\right|\le(t-t_{\mathrm{in}})A_0\|\mathbf{b}\cdot\nabla g+\kappa\Delta g\|_{L^\infty((t_{\mathrm{in}},t)\times\mathbb{T}^2)}. ∫ T 2 ( θ κ ( t , x ) − θ in κ ( x )) g ( x ) d x ≤ ( t − t in ) A 0 ∥ b ⋅ ∇ g + κ Δ g ∥ L ∞ (( t in , t ) × T 2 ) .
Letting first κ → 0 \kappa\to0 κ → 0 in K ′ ′ \mathcal{K}'' K ′′ and then t ↓ t i n t\downarrow t_{\mathrm{in}} t ↓ t in gives ∫ θ ∗ ( t , ⋅ ) g → ∫ θ i n , ∗ g \int\theta_*(t,\cdot)g\to\int\theta_{\mathrm{in},*}g ∫ θ ∗ ( t , ⋅ ) g → ∫ θ in , ∗ g ; since ∥ θ ∗ ( t , ⋅ ) ∥ L 2 ( T 2 ) ≤ A 0 \|\theta_*(t,\cdot)\|_{L^2(\mathbb{T}^2)}\le A_0 ∥ θ ∗ ( t , ⋅ ) ∥ L 2 ( T 2 ) ≤ A 0 , this extends to every g ∈ L 2 ( T 2 ) g\in L^2(\mathbb{T}^2) g ∈ L 2 ( T 2 ) .
( a ) (a) ( a ) , part (iv) . Each finite union of bands is bounded away from zero, so M → ∞ M\to\infty M → ∞ as κ → 0 \kappa\to0 κ → 0 in K ′ \mathcal{K}' K ′ , where κ ∈ K M \kappa\in\mathcal{K}_M κ ∈ K M . Lemma 9.2, with Φ = ϕ \Phi=\phi Φ = ϕ , Φ ~ = ϕ M \widetilde{\Phi}=\phi_M Φ = ϕ M and, by (72) and κ ≥ 1 2 ε M β − γ \kappa\ge\frac{1}{2}\varepsilon_M^{\beta-\gamma} κ ≥ 2 1 ε M β − γ ,
ϖ : = C 0 ε M β + 1 κ ≤ C ε M q β − β + γ = C ε M ( q + 2 ) γ → 0 , \varpi:=\frac{C_0\varepsilon_M^\beta+1}{\kappa}\le C\varepsilon_M^{q\beta-\beta+\gamma}=C\varepsilon_M^{(q+2)\gamma}\to0, ϖ := κ C 0 ε M β + 1 ≤ C ε M q β − β + γ = C ε M ( q + 2 ) γ → 0 ,
where β ( q − 1 ) = ( q + 1 ) γ \beta(q-1)=(q+1)\gamma β ( q − 1 ) = ( q + 1 ) γ , gives, for every T > 0 T>0 T > 0 , ∥ θ κ − θ ~ κ ∥ L ∞ ( [ t i n , t i n + T ] ; L 2 ( T 2 ) ) ≤ ϖ A 0 \|\theta^\kappa-\widetilde{\theta}^\kappa\|_{L^\infty([t_{\mathrm{in}},t_{\mathrm{in}}+T];L^2(\mathbb{T}^2))}\le\varpi A_0 ∥ θ κ − θ κ ∥ L ∞ ([ t in , t in + T ] ; L 2 ( T 2 )) ≤ ϖ A 0 , hence (iv).
( b ) (b) ( b ) . This is the same argument in the tube, with Theorem C in place of Theorem A. The sets { ( t , X m ( t , y , t f ) ) : t f − 8 h m + 1 + t 0 ≤ t ≤ t f , ∣ y ∣ < r o b s } \{(t,X_m(t,y,t_f)):t_f-8h_{m+1}+t_0\le t\le t_f,\ |y|<r_{\mathrm{obs}}\} {( t , X m ( t , y , t f )) : t f − 8 h m + 1 + t 0 ≤ t ≤ t f , ∣ y ∣ < r obs } , for r o b s ↑ 1 8 r t u b e r_{\mathrm{obs}}\uparrow\frac18r_{\mathrm{tube}} r obs ↑ 8 1 r tube and t 0 ↓ 4 h m + 1 t_0\downarrow4h_{m+1} t 0 ↓ 4 h m + 1 , exhaust T ′ \mathcal{T}' T ′ , and Theorem C bounds the θ κ \theta^\kappa θ κ in C 0 , s C^{0,s} C 0 , s on each of them, uniformly in κ \kappa κ and in time, up to t f t_f t f . The averages of (a), centred at a point x = X m ( t , y , t f ) x=X_m(t,y,t_f) x = X m ( t , y , t f ) of such a set and with r r r and t − t ′ > 0 t-t'>0 t − t ′ > 0 small, depending on m m m , 1 8 r t u b e − r o b s \frac18r_{\mathrm{tube}}-r_{\mathrm{obs}} 8 1 r tube − r obs and t 0 − 4 h m + 1 t_0-4h_{m+1} t 0 − 4 h m + 1 , are supported in a slightly larger such set at every time τ ∈ [ t ′ , t ] \tau\in[t',t] τ ∈ [ t ′ , t ] , since, for z ∈ D r ( x ) z\in D_r(x) z ∈ D r ( x ) , ∣ X m ( t f , z , τ ) − y ∣ ≤ 2 ( r + ∥ b m ∥ L ∞ ( R × T 2 ) ( t − τ ) ) |X_m(t_f,z,\tau)-y|\le2(r+\|\mathbf{b}_m\|_{L^\infty(\mathbb{R}\times\mathbb{T}^2)}(t-\tau)) ∣ X m ( t f , z , τ ) − y ∣ ≤ 2 ( r + ∥ b m ∥ L ∞ ( R × T 2 ) ( t − τ )) by (76) . They give the same modulus in time for t < t f t<t_f t < t f , and, by the limit of Theorem C at the end of the tube, for t = t f t=t_f t = t f .
Uniform anomalous dissipation: Theorem B
Lemma 9.4 (Positive Hölder-smoothing operators contract). Let N ≥ 1 N\ge1 N ≥ 1 be an integer, and let S S S be a linear map from L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) to the continuous functions on T 2 \mathbb{T}^2 T 2 such that S 1 = 1 S1=1 S 1 = 1 and, for every f ∈ L 2 ( T 2 ) f\in L^2(\mathbb{T}^2) f ∈ L 2 ( T 2 ) , ∫ T 2 S f ( x ) d x = ∫ T 2 f ( x ) d x \int_{\mathbb{T}^2}Sf(x)\,\mathrm{d}x=\int_{\mathbb{T}^2}f(x)\,\mathrm{d}x ∫ T 2 S f ( x ) d x = ∫ T 2 f ( x ) d x , ∥ S f ∥ C 0 , s ( T 2 ) ≤ N s ∥ f ∥ L 2 ( T 2 ) \|Sf\|_{C^{0,s}(\mathbb{T}^2)}\le N^s\|f\|_{L^2(\mathbb{T}^2)} ∥ S f ∥ C 0 , s ( T 2 ) ≤ N s ∥ f ∥ L 2 ( T 2 ) and, if f ≥ 0 f\ge0 f ≥ 0 , S f ≥ 0 Sf\ge0 S f ≥ 0 . Then, for every f ∈ L 2 ( T 2 ) f\in L^2(\mathbb{T}^2) f ∈ L 2 ( T 2 ) with ∫ T 2 f ( x ) d x = 0 \int_{\mathbb{T}^2}f(x)\,\mathrm{d}x=0 ∫ T 2 f ( x ) d x = 0 ,
∥ S f ∥ L 2 ( T 2 ) 2 ≤ ( 1 − 1 2 N 4 ) ∥ f ∥ L 2 ( T 2 ) 2 . \|Sf\|_{L^2(\mathbb{T}^2)}^2\le\left(1-\frac{1}{2N^4}\right)\|f\|_{L^2(\mathbb{T}^2)}^2. ∥ S f ∥ L 2 ( T 2 ) 2 ≤ ( 1 − 2 N 4 1 ) ∥ f ∥ L 2 ( T 2 ) 2 .
Proof. Step 1: Rows. For every x ∈ T 2 x\in\mathbb{T}^2 x ∈ T 2 , f ↦ S f ( x ) f\mapsto Sf(x) f ↦ S f ( x ) is bounded on L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) , so, by the Riesz representation theorem, S f ( x ) = ⟨ S x , f ⟩ L 2 ( T 2 ) Sf(x)=\langle S_x,f\rangle_{L^2(\mathbb{T}^2)} S f ( x ) = ⟨ S x , f ⟩ L 2 ( T 2 ) for some S x ∈ L 2 ( T 2 ) S_x\in L^2(\mathbb{T}^2) S x ∈ L 2 ( T 2 ) , the row of S S S at x x x . The Hölder bound gives ∥ S x − S y ∥ L 2 ( T 2 ) ≤ N s ∣ x − y ∣ T 2 s \|S_x-S_y\|_{L^2(\mathbb{T}^2)}\le N^s|x-y|_{\mathbb{T}^2}^s ∥ S x − S y ∥ L 2 ( T 2 ) ≤ N s ∣ x − y ∣ T 2 s . Positivity gives S x ≥ 0 S_x\ge0 S x ≥ 0 , since 0 ≤ S ( S x ) ( x ) = ∥ S x ∥ L 2 ( T 2 ) 2 − ∥ S x − S x ∥ L 2 ( T 2 ) 2 0\le S(S_x)(x)=\|S_x\|_{L^2(\mathbb{T}^2)}^2-\|S_x-S_x\|^2_{L^2(\mathbb{T}^2)} 0 ≤ S ( S x ) ( x ) = ∥ S x ∥ L 2 ( T 2 ) 2 − ∥ S x − S x ∥ L 2 ( T 2 ) 2 for S x S_x S x ; and S 1 = 1 S1=1 S 1 = 1 gives ∫ T 2 S x ( z ) d z = 1 \int_{\mathbb{T}^2}S_x(z)\,\mathrm{d}z=1 ∫ T 2 S x ( z ) d z = 1 , hence ∥ S x ∥ L 2 ( T 2 ) ≥ 1 \|S_x\|_{L^2(\mathbb{T}^2)}\ge1 ∥ S x ∥ L 2 ( T 2 ) ≥ 1 ; and the preservation of integrals gives ∫ T 2 S x d x = 1 \int_{\mathbb{T}^2}S_x\,\mathrm{d}x=1 ∫ T 2 S x d x = 1 .
Step 2: The kernel of S ∗ S S^*S S ∗ S . The function ⟨ S x , S y ⟩ L 2 ( T 2 ) \langle S_x,S_y\rangle_{L^2(\mathbb{T}^2)} ⟨ S x , S y ⟩ L 2 ( T 2 ) is continuous, symmetric and nonnegative, for every x x x its integral in y y y is 1 1 1 , and, by polarization, it equals 1 2 ( ∥ S x ∥ L 2 ( T 2 ) 2 + ∥ S y ∥ L 2 ( T 2 ) 2 − ∥ S x − S y ∥ L 2 ( T 2 ) 2 ) ≥ 1 − 1 2 N 2 s ∣ x − y ∣ T 2 2 s \frac12(\|S_x\|_{L^2(\mathbb{T}^2)}^2+\|S_y\|_{L^2(\mathbb{T}^2)}^2-\|S_x-S_y\|_{L^2(\mathbb{T}^2)}^2)\ge1-\frac12N^{2s}|x-y|_{\mathbb{T}^2}^{2s} 2 1 ( ∥ S x ∥ L 2 ( T 2 ) 2 + ∥ S y ∥ L 2 ( T 2 ) 2 − ∥ S x − S y ∥ L 2 ( T 2 ) 2 ) ≥ 1 − 2 1 N 2 s ∣ x − y ∣ T 2 2 s , which, if ∣ x − y ∣ T 2 ≤ 1 / N |x-y|_{\mathbb{T}^2}\le1/N ∣ x − y ∣ T 2 ≤ 1/ N , is at least 1 2 \frac12 2 1 . The adjoint is S ∗ g = ∫ T 2 g ( x ) S x d x S^*g=\int_{\mathbb{T}^2}g(x)S_x\,\mathrm{d}x S ∗ g = ∫ T 2 g ( x ) S x d x , and expanding ∥ S ∗ g ∥ L 2 ( T 2 ) 2 \|S^*g\|_{L^2(\mathbb{T}^2)}^2 ∥ S ∗ g ∥ L 2 ( T 2 ) 2 gives, for continuous g g g ,
∥ g ∥ L 2 ( T 2 ) 2 − ∥ S ∗ g ∥ L 2 ( T 2 ) 2 = 1 2 ∫ T 2 × T 2 ⟨ S x , S y ⟩ L 2 ( T 2 ) ∣ g ( x ) − g ( y ) ∣ 2 d x d y . (91) \|g\|_{L^2(\mathbb{T}^2)}^2-\|S^*g\|_{L^2(\mathbb{T}^2)}^2=\frac12\int_{\mathbb{T}^2\times\mathbb{T}^2}\langle S_x,S_y\rangle_{L^2(\mathbb{T}^2)}|g(x)-g(y)|^2\,\mathrm{d}x\,\mathrm{d}y. \tag*{(91)} ∥ g ∥ L 2 ( T 2 ) 2 − ∥ S ∗ g ∥ L 2 ( T 2 ) 2 = 2 1 ∫ T 2 × T 2 ⟨ S x , S y ⟩ L 2 ( T 2 ) ∣ g ( x ) − g ( y ) ∣ 2 d x d y . (91)
Step 3: A Poincaré inequality by translations. Let g g g be continuous, periodic and of mean zero. Then the average over z ∈ [ − 1 2 , 1 2 ] 2 z \in\left[-\frac{1}{2},\frac{1}{2}\right]^2 z ∈ [ − 2 1 , 2 1 ] 2 of ∫ T 2 ∣ g ( x + z ) − g ( x ) ∣ 2 d x \int_{\mathbb{T}^2} |g(x+z)-g(x)|^2 \, \mathrm{d}x ∫ T 2 ∣ g ( x + z ) − g ( x ) ∣ 2 d x is 2 ∥ g ∥ L 2 ( T 2 ) 2 2\|g\|_{L^2(\mathbb{T}^2)}^2 2∥ g ∥ L 2 ( T 2 ) 2 . Writing g ( ⋅ + N z ) − g g(\cdot+Nz)-g g ( ⋅ + N z ) − g as a sum of N N N increments by z z z , Cauchy–Schwarz bounds the integral at N z Nz N z by N 2 N^2 N 2 times the integral at z z z , and the change of variables from N z Nz N z to z z z gives
2 ∥ g ∥ L 2 ( T 2 ) 2 ≤ N 4 ∫ [ − 1 2 N , 1 2 N ] 2 × T 2 ∣ g ( x + z ) − g ( x ) ∣ 2 d z d x . 2\|g\|_{L^2(\mathbb{T}^2)}^2 \leq N^4 \int_{\left[-\frac{1}{2N},\frac{1}{2N}\right]^2 \times\mathbb{T}^2} |g(x+z)-g(x)|^2 \, \mathrm{d}z \, \mathrm{d}x . 2∥ g ∥ L 2 ( T 2 ) 2 ≤ N 4 ∫ [ − 2 N 1 , 2 N 1 ] 2 × T 2 ∣ g ( x + z ) − g ( x ) ∣ 2 d z d x .
For z ∈ [ − 1 2 N , 1 2 N ] 2 z \in\left[-\frac{1}{2N},\frac{1}{2N}\right]^2 z ∈ [ − 2 N 1 , 2 N 1 ] 2 , ∣ z ∣ ≤ 1 / N |z| \leq1/N ∣ z ∣ ≤ 1/ N , so, by Step 2 and the change of variables y = x + z y=x+z y = x + z in (91) , the right side of (91) is at least 1 4 ∫ [ − 1 2 N , 1 2 N ] 2 × T 2 ∣ g ( x + z ) − g ( x ) ∣ 2 d z d x ≥ 1 2 N 4 ∥ g ∥ L 2 ( T 2 ) 2 \frac{1}{4}\int_{\left[-\frac{1}{2N},\frac{1}{2N}\right]^2 \times\mathbb{T}^2} |g(x+z)-g(x)|^2 \, \mathrm{d}z \, \mathrm{d}x \geq\frac{1}{2N^4}\|g\|_{L^2(\mathbb{T}^2)}^2 4 1 ∫ [ − 2 N 1 , 2 N 1 ] 2 × T 2 ∣ g ( x + z ) − g ( x ) ∣ 2 d z d x ≥ 2 N 4 1 ∥ g ∥ L 2 ( T 2 ) 2 .
Step 4: Conclusion. Let ∫ T 2 f = 0 \int_{\mathbb{T}^2} f=0 ∫ T 2 f = 0 and g : = S f g:=Sf g := S f , which is continuous and of mean zero. Then
∥ g ∥ L 2 ( T 2 ) 2 = ⟨ f , S ∗ g ⟩ L 2 ( T 2 ) ≤ ∥ f ∥ L 2 ( T 2 ) ∥ S ∗ g ∥ L 2 ( T 2 ) ≤ ( 1 − 1 2 N 4 ) 1 / 2 ∥ f ∥ L 2 ( T 2 ) ∥ g ∥ L 2 ( T 2 ) \|g\|_{L^2(\mathbb{T}^2)}^2=\langle f,S^*g\rangle_{L^2(\mathbb{T}^2)}\leq\|f\|_{L^2(\mathbb{T}^2)}\|S^*g\|_{L^2(\mathbb{T}^2)}\leq\left(1-\frac{1}{2N^4}\right)^{1/2}\|f\|_{L^2(\mathbb{T}^2)}\|g\|_{L^2(\mathbb{T}^2)} ∥ g ∥ L 2 ( T 2 ) 2 = ⟨ f , S ∗ g ⟩ L 2 ( T 2 ) ≤ ∥ f ∥ L 2 ( T 2 ) ∥ S ∗ g ∥ L 2 ( T 2 ) ≤ ( 1 − 2 N 4 1 ) 1/2 ∥ f ∥ L 2 ( T 2 ) ∥ g ∥ L 2 ( T 2 )
by Step 3.
Proof of Theorem B. Fix t i n ∈ R t_{\mathrm{in}}\in\mathbb{R} t in ∈ R and κ ∈ K \kappa\in\mathcal{K} κ ∈ K , and let S θ i n : = θ ( t i n + 1 , ⋅ ) S\theta_{\mathrm{in}}:=\theta(t_{\mathrm{in}}+1,\cdot) S θ in := θ ( t in + 1 , ⋅ ) , with θ \theta θ the weak solution with datum θ i n \theta_{\mathrm{in}} θ in . By uniqueness, S S S is linear; constants are solutions, so S 1 = 1 S1=1 S 1 = 1 ; testing the equation with a function of time only gives ∫ T 2 S θ i n = ∫ T 2 θ i n \int_{\mathbb{T}^2}S\theta_{\mathrm{in}}=\int_{\mathbb{T}^2}\theta_{\mathrm{in}} ∫ T 2 S θ in = ∫ T 2 θ in , since ∇ ⋅ b = 0 \nabla\cdot b=0 ∇ ⋅ b = 0 . The operator S S S is positive by the maximum principle: if θ i n ≥ 0 \theta_{\mathrm{in}}\geq0 θ in ≥ 0 , testing the equation with min { θ , 0 } \min\{\theta,0\} min { θ , 0 } (Stampacchia’s truncation) gives ∫ T 2 min { θ ( t , ⋅ ) , 0 } 2 d x ≤ ∫ T 2 min { θ i n , 0 } 2 d x = 0 \int_{\mathbb{T}^2}\min\{\theta(t,\cdot),0\}^2\,\mathrm{d}x\leq\int_{\mathbb{T}^2}\min\{\theta_{\mathrm{in}},0\}^2\,\mathrm{d}x=0 ∫ T 2 min { θ ( t , ⋅ ) , 0 } 2 d x ≤ ∫ T 2 min { θ in , 0 } 2 d x = 0 for every t > t i n t>t_{\mathrm{in}} t > t in : the diffusion term has a sign, and the transport term vanishes, as ∇ ⋅ b = 0 \nabla\cdot b=0 ∇ ⋅ b = 0 .
By Theorem A on ( t i n , t i n + 1 ) (t_{\mathrm{in}},t_{\mathrm{in}}+1) ( t in , t in + 1 ) with t 0 = 1 2 t_0=\frac{1}{2} t 0 = 2 1 and (9.2′), S S S is bounded from L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) into C 0 , s ( T 2 ) C^{0,s}(\mathbb{T}^2) C 0 , s ( T 2 ) , and, by Lemma 9.4 with N N N the least integer such that N N N is at least the constant of Theorem A, we obtain, with c : = 1 / ( 4 N 4 ) c:=1/(4N^4) c := 1/ ( 4 N 4 ) ,
∥ S g ∥ L 2 ( T 2 ) 2 ≤ ( 1 − 2 c ) ∥ g ∥ L 2 ( T 2 ) 2 \|Sg\|_{L^2(\mathbb{T}^2)}^2\leq(1-2c)\|g\|_{L^2(\mathbb{T}^2)}^2 ∥ S g ∥ L 2 ( T 2 ) 2 ≤ ( 1 − 2 c ) ∥ g ∥ L 2 ( T 2 ) 2
for every g ∈ L 2 ( T 2 ) g\in L^2(\mathbb{T}^2) g ∈ L 2 ( T 2 ) with ∫ T 2 g ( x ) d x = 0 \int_{\mathbb{T}^2}g(x)\,\mathrm{d}x=0 ∫ T 2 g ( x ) d x = 0 .
Since the constant of Theorem A is independent of t i n t_{\mathrm{in}} t in and of κ ∈ K \kappa\in\mathcal{K} κ ∈ K , so are N N N and c c c . Since constants are solutions, θ − ( θ i n ) T 2 \theta-(\theta_{\mathrm{in}})_{\mathbb{T}^2} θ − ( θ in ) T 2 is the weak solution with datum θ i n − ( θ i n ) T 2 \theta_{\mathrm{in}}-(\theta_{\mathrm{in}})_{\mathbb{T}^2} θ in − ( θ in ) T 2 , and θ ( t i n + 1 , ⋅ ) − ( θ i n ) T 2 = S ( θ i n − ( θ i n ) T 2 ) \theta(t_{\mathrm{in}}+1,\cdot)-(\theta_{\mathrm{in}})_{\mathbb{T}^2}=S(\theta_{\mathrm{in}}-(\theta_{\mathrm{in}})_{\mathbb{T}^2}) θ ( t in + 1 , ⋅ ) − ( θ in ) T 2 = S ( θ in − ( θ in ) T 2 ) . The energy identity (73) for θ − ( θ i n ) T 2 \theta-(\theta_{\mathrm{in}})_{\mathbb{T}^2} θ − ( θ in ) T 2 gives
2 κ ∥ ∇ θ ∥ L 2 ( ( t i n , t i n + 1 ) × T 2 ) 2 = ∥ θ i n − ( θ i n ) T 2 ∥ L 2 ( T 2 ) 2 − ∥ S ( θ i n − ( θ i n ) T 2 ) ∥ L 2 ( T 2 ) 2 2\kappa\|\nabla\theta\|_{L^2((t_{\mathrm{in}},t_{\mathrm{in}}+1)\times\mathbb{T}^2)}^2=\|\theta_{\mathrm{in}}-(\theta_{\mathrm{in}})_{\mathbb{T}^2}\|_{L^2(\mathbb{T}^2)}^2-\|S(\theta_{\mathrm{in}}-(\theta_{\mathrm{in}})_{\mathbb{T}^2})\|_{L^2(\mathbb{T}^2)}^2 2 κ ∥∇ θ ∥ L 2 (( t in , t in + 1 ) × T 2 ) 2 = ∥ θ in − ( θ in ) T 2 ∥ L 2 ( T 2 ) 2 − ∥ S ( θ in − ( θ in ) T 2 ) ∥ L 2 ( T 2 ) 2
≥ 2 c ∥ θ i n − ( θ i n ) T 2 ∥ L 2 ( T 2 ) 2 , \geq2c\|\theta_{\mathrm{in}}-(\theta_{\mathrm{in}})_{\mathbb{T}^2}\|_{L^2(\mathbb{T}^2)}^2, ≥ 2 c ∥ θ in − ( θ in ) T 2 ∥ L 2 ( T 2 ) 2 ,
which is (26) . For t i n ′ > t i n t'_{\mathrm{in}}>t_{\mathrm{in}} t in ′ > t in , the restriction of θ \theta θ to ( t i n ′ , ∞ ) (t'_{\mathrm{in}},\infty) ( t in ′ , ∞ ) is the weak solution with datum θ ( t i n ′ , ⋅ ) \theta(t'_{\mathrm{in}},\cdot) θ ( t in ′ , ⋅ ) , whose mean is ( θ i n ) T 2 (\theta_{\mathrm{in}})_{\mathbb{T}^2} ( θ in ) T 2 . So the contraction, with t i n + k t_{\mathrm{in}}+k t in + k in place of t i n t_{\mathrm{in}} t in , gives ∥ θ ( t i n + k + 1 , ⋅ ) − ( θ i n ) T 2 ∥ L 2 ( T 2 ) 2 ≤ ( 1 − 2 c ) ∥ θ ( t i n + k , ⋅ ) − ( θ i n ) T 2 ∥ L 2 ( T 2 ) 2 \|\theta(t_{\mathrm{in}}+k+1,\cdot)-(\theta_{\mathrm{in}})_{\mathbb{T}^2}\|_{L^2(\mathbb{T}^2)}^2\leq(1-2c)\|\theta(t_{\mathrm{in}}+k,\cdot)-(\theta_{\mathrm{in}})_{\mathbb{T}^2}\|_{L^2(\mathbb{T}^2)}^2 ∥ θ ( t in + k + 1 , ⋅ ) − ( θ in ) T 2 ∥ L 2 ( T 2 ) 2 ≤ ( 1 − 2 c ) ∥ θ ( t in + k , ⋅ ) − ( θ in ) T 2 ∥ L 2 ( T 2 ) 2 for every integer k ≥ 0 k\geq0 k ≥ 0 , and, by (73) , ∥ θ ( t , ⋅ ) − ( θ i n ) T 2 ∥ L 2 ( T 2 ) 2 \|\theta(t,\cdot)-(\theta_{\mathrm{in}})_{\mathbb{T}^2}\|_{L^2(\mathbb{T}^2)}^2 ∥ θ ( t , ⋅ ) − ( θ in ) T 2 ∥ L 2 ( T 2 ) 2 does not increase in t t t . Hence, for every t ≥ t i n t\geq t_{\mathrm{in}} t ≥ t in , as 1 − 2 c ≤ exp ( − 2 c ) 1-2c\leq\exp(-2c) 1 − 2 c ≤ exp ( − 2 c ) and c ≤ 1 2 c\leq\frac{1}{2} c ≤ 2 1 , we obtain (27) :
∥ θ ( t , ⋅ ) − ( θ i n ) T 2 ∥ L 2 ( T 2 ) 2 ≤ ( 1 − 2 c ) ⌊ t − t i n ⌋ ∥ θ i n − ( θ i n ) T 2 ∥ L 2 ( T 2 ) 2 ≤ 4 exp ( − 2 c ( t − t i n ) ) ∥ θ i n − ( θ i n ) T 2 ∥ L 2 ( T 2 ) 2 . \|\theta(t,\cdot)-(\theta_{\mathrm{in}})_{\mathbb{T}^2}\|_{L^2(\mathbb{T}^2)}^2\leq(1-2c)^{\lfloor t-t_{\mathrm{in}}\rfloor}\|\theta_{\mathrm{in}}-(\theta_{\mathrm{in}})_{\mathbb{T}^2}\|_{L^2(\mathbb{T}^2)}^2\leq4\exp(-2c(t-t_{\mathrm{in}}))\|\theta_{\mathrm{in}}-(\theta_{\mathrm{in}})_{\mathbb{T}^2}\|_{L^2(\mathbb{T}^2)}^2. ∥ θ ( t , ⋅ ) − ( θ in ) T 2 ∥ L 2 ( T 2 ) 2 ≤ ( 1 − 2 c ) ⌊ t − t in ⌋ ∥ θ in − ( θ in ) T 2 ∥ L 2 ( T 2 ) 2 ≤ 4 exp ( − 2 c ( t − t in )) ∥ θ in − ( θ in ) T 2 ∥ L 2 ( T 2 ) 2 .
Cell problems and the averaged flux
In this section we construct, on each pulse, the finite lift of a system with sources: correctors of every degree up to N c o r N_{\mathrm{cor}} N cor , which oscillate at the wavelength ε m \varepsilon_m ε m , and the means they leave behind, the sources of the successor. In the chart of its block (§ 4.1 ), a pulse of level m m m is a single shear, whose cell problems are one-dimensional and explicitly solvable. The first correctors are those of [AV25, Section 3.2]. Throughout, Λ ≥ 128 \Lambda\ge128 Λ ≥ 128 is an integer, and C C C , c c c denote positive constants depending only on α \alpha α , s s s , q q q , ζ \zeta ζ , unless another dependence is stated. For M ≥ 1 M \ge1 M ≥ 1 , κ ∈ K M \kappa\in\mathcal{K}_M κ ∈ K M and 1 ≤ m ≤ M 1 \le m \le M 1 ≤ m ≤ M , recall that ε ˉ m = ε m ( κ m − 1 / κ m ) 1 / 2 \bar{\varepsilon}_m = \varepsilon_m(\kappa_{m-1}/\kappa_m)^{1/2} ε ˉ m = ε m ( κ m − 1 / κ m ) 1/2 is the corrector amplitude of Subsection 5.1, and ρ ( m ) = ( κ m − 1 h m ) 1 / 2 \rho(m) = (\kappa_{m-1}h_m)^{1/2} ρ ( m ) = ( κ m − 1 h m ) 1/2 is the effective diffusion length (33) .
The window of a pulse. Let m ≥ 1 m \ge1 m ≥ 1 and k k k be odd, and let e ( k ) ⊥ : = e 1 e_{(k)}^\perp:= e_1 e ( k ) ⊥ := e 1 for k ∈ 4 Z + 1 k \in4\mathbb{Z}+1 k ∈ 4 Z + 1 and e ( k ) ⊥ : = e 2 e_{(k)}^\perp:= e_2 e ( k ) ⊥ := e 2 for k ∈ 4 Z + 3 k \in4\mathbb{Z}+3 k ∈ 4 Z + 3 , the coordinate vector other than e ( k ) e_{(k)} e ( k ) . By (8) , ψ m , k ( y ) = ε m β sin ( 2 π y ⋅ e ( k ) ⊥ / ε m ) \psi_{m,k}(y) = \varepsilon_m^\beta\sin(2\pi y \cdot e_{(k)}^\perp/\varepsilon_m) ψ m , k ( y ) = ε m β sin ( 2 π y ⋅ e ( k ) ⊥ / ε m ) and
∇ ⊥ ψ m , k ( y ) = ± 2 π ε m β − 1 cos ( 2 π y ⋅ e ( k ) ⊥ / ε m ) e ( k ) (92) \nabla^\perp\psi_{m,k}(y) = \pm2\pi\varepsilon_m^{\beta-1}\cos(2\pi y \cdot e_{(k)}^\perp/\varepsilon_m)e_{(k)} \tag*{(92)} ∇ ⊥ ψ m , k ( y ) = ± 2 π ε m β − 1 cos ( 2 π y ⋅ e ( k ) ⊥ / ε m ) e ( k ) (92)
with + + + for k ∈ 4 Z + 1 k \in4\mathbb{Z}+1 k ∈ 4 Z + 1 and − - − for k ∈ 4 Z + 3 k \in4\mathbb{Z}+3 k ∈ 4 Z + 3 . The window of pulse k k k and the time cutoff of k k k are
∣ t − k τ m ∣ ≤ 61 48 τ m , H m , k ( t ) : = H ( 5 2 − 2 ∣ t − k τ m ∣ τ m ) (93) |t-k\tau_m| \le\frac{61}{48}\tau_m,\qquad H_{m,k}(t) := H\left(\frac{5}{2}-\frac{2|t-k\tau_m|}{\tau_m}\right) \tag*{(93)} ∣ t − k τ m ∣ ≤ 48 61 τ m , H m , k ( t ) := H ( 2 5 − τ m 2∣ t − k τ m ∣ ) (93)
with H H H of (11) , which is 0 on ( − ∞ , 0 ] (-\infty,0] ( − ∞ , 0 ] and 1 on [ 1 , ∞ ) [1,\infty) [ 1 , ∞ ) . As the envelope ζ ~ m , k \widetilde{\zeta}_{m,k} ζ m , k satisfies 0 ≤ ζ ~ m , k ≤ ζ m , k 0 \le\widetilde{\zeta}_{m,k} \le\zeta_{m,k} 0 ≤ ζ m , k ≤ ζ m , k (by (4) , Lemma 7.1 (d)), it vanishes for ∣ t − k τ m ∣ ≥ 2 3 τ m |t-k\tau_m| \ge\frac{2}{3}\tau_m ∣ t − k τ m ∣ ≥ 3 2 τ m ; as the centres of odd slots are 2 τ m 2\tau_m 2 τ m apart, no other pulse is active on the window of pulse k k k , where therefore ψ m ( t , y ) = ζ ~ m , k ( t ) ψ m , k ( y ) \psi_m(t,y) = \widetilde{\zeta}_{m,k}(t)\psi_{m,k}(y) ψ m ( t , y ) = ζ m , k ( t ) ψ m , k ( y ) (see (56) ). The cutoff H m , k H_{m,k} H m , k is 1 for ∣ t − k τ m ∣ ≤ 3 4 τ m |t-k\tau_m| \le\frac{3}{4}\tau_m ∣ t − k τ m ∣ ≤ 4 3 τ m , hence where ζ ~ m , k ≠ 0 \widetilde{\zeta}_{m,k} \ne0 ζ m , k = 0 ; it vanishes for ∣ t − k τ m ∣ ≥ 5 4 τ m |t-k\tau_m| \ge\frac{5}{4}\tau_m ∣ t − k τ m ∣ ≥ 4 5 τ m ; and ∑ k ∈ 2 Z + 1 H m , k = 1 \sum_{k\in2\mathbb{Z}+1}H_{m,k}=1 ∑ k ∈ 2 Z + 1 H m , k = 1 , since H ( x ) + H ( 1 − x ) = 1 H(x)+H(1-x)=1 H ( x ) + H ( 1 − x ) = 1 . By Lemma 7.8(i), in the chart y ↦ X m − 1 ( t , y , ℓ m , k h m ) y \mapsto X_{m-1}(t,y,\ell_{m,k}h_m) y ↦ X m − 1 ( t , y , ℓ m , k h m ) of k k k the velocity added at level m m m is ∇ ⊥ ψ m ( t , y ) \nabla^\perp\psi_m(t,y) ∇ ⊥ ψ m ( t , y ) in the block of k k k , and it vanishes on the rest of the window; the chart and K m , ℓ m , k \mathcal{K}_{m,\ell_{m,k}} K m , ℓ m , k are data on the whole window.
A two-scale form on the window of pulse k k k is a function A ( t , y , z ) A(t,y,z) A ( t , y , z ) of t t t in the window and y , z ∈ R 2 y,z \in\mathbb{R}^2 y , z ∈ R 2 , 1-periodic in the fast variable z z z , read at z = y / ε m z=y/\varepsilon_m z = y / ε m ; it is transverse if it depends on z z z only through z ⋅ e ( k ) ⊥ z\cdot e_{(k)}^\perp z ⋅ e ( k ) ⊥ . We write ( A ) T 2 ( t , y ) : = ∫ [ 0 , 1 ] 2 A ( t , y , z ) d z (A)_{\mathrm{T2}}(t,y):=\int_{[0,1]^2}A(t,y,z)\,dz ( A ) T2 ( t , y ) := ∫ [ 0 , 1 ] 2 A ( t , y , z ) d z for its average in z z z , and, for a transverse form, A ^ ( t , y ; ξ ) \widehat{A}(t,y;\xi) A ( t , y ; ξ ) for its ξ \xi ξ -th Fourier coefficient in z ⋅ e ( k ) ⊥ z\cdot e_{(k)}^\perp z ⋅ e ( k ) ⊥ , ξ ∈ Z \xi\in\mathbb{Z} ξ ∈ Z , as in § 5.4 .
The corrector of a pulse and the averaged flux
Lemma 10.1 (The corrector of a pulse). Let m ≥ 1 m \ge1 m ≥ 1 , ν > 0 \nu> 0 ν > 0 and k k k odd, and let
χ m , k ν ( t , y ) : = − ( ∫ 0 ∞ exp ( − 4 π 2 ν t ′ ε m 2 ) ζ ~ m , k ( t − t ′ ) d t ′ ) ∇ ⊥ ψ m , k ( y ) . (94) \chi^\nu_{m,k}(t,y):=-\left(\int_0^\infty\exp\left(-\frac{4\pi^2\nu t'}{\varepsilon_m^2}\right)\widetilde{\zeta}_{m,k}(t-t')\,dt'\right)\nabla^\perp\psi_{m,k}(y). \tag*{(94)} χ m , k ν ( t , y ) := − ( ∫ 0 ∞ exp ( − ε m 2 4 π 2 ν t ′ ) ζ m , k ( t − t ′ ) d t ′ ) ∇ ⊥ ψ m , k ( y ) . (94) (a) χ m , k ν \chi^\nu_{m,k} χ m , k ν is Z 2 \mathbb{Z}^2 Z 2 -periodic and smooth in y y y , continuously differentiable in t t t , of mean zero over [ 0 , 1 ] 2 [0,1]^2 [ 0 , 1 ] 2 , parallel to e ( k ) e_{(k)} e ( k ) , a function of y ⋅ e ( k ) ⊥ y\cdot e_{(k)}^\perp y ⋅ e ( k ) ⊥ , and it solves, componentwise,
∂ t χ m , k ν + ζ ~ m , k ∇ ⊥ ψ m , k ⋅ ∇ χ m , k ν − ν Δ χ m , k ν = − ζ ~ m , k ∇ ⊥ ψ m , k in R × R 2 , χ m , k ν = 0 on ( − ∞ , ( k − 2 3 ) τ m ] × R 2 , (95) \begin{aligned}
\partial_t\chi^\nu_{m,k}+\widetilde{\zeta}_{m,k}\nabla^\perp\psi_{m,k}\cdot\nabla\chi^\nu_{m,k}-\nu\Delta\chi^\nu_{m,k}&=-\widetilde{\zeta}_{m,k}\nabla^\perp\psi_{m,k} &&\text{in }\mathbb{R}\times\mathbb{R}^2,\\
\chi^\nu_{m,k}&=0 &&\text{on }(-\infty,(k-\tfrac{2}{3})\tau_m]\times\mathbb{R}^2, \tag*{(95)}
\end{aligned} ∂ t χ m , k ν + ζ m , k ∇ ⊥ ψ m , k ⋅ ∇ χ m , k ν − ν Δ χ m , k ν χ m , k ν = − ζ m , k ∇ ⊥ ψ m , k = 0 in R × R 2 , on ( − ∞ , ( k − 3 2 ) τ m ] × R 2 , (95) whose transport term vanishes identically.
(b) ∥ χ m , k ν ∥ L ∞ ( R × T 2 ) ≤ ε m β + 1 / ( 2 π ν ) \|\chi^\nu_{m,k}\|_{L^\infty(\mathbb{R}\times\mathbb{T}^2)} \le\varepsilon_m^{\beta+1}/(2\pi\nu) ∥ χ m , k ν ∥ L ∞ ( R × T 2 ) ≤ ε m β + 1 / ( 2 π ν ) .
(c) If m ≥ 2 m \ge2 m ≥ 2 , there exist Λ 0 ( α , s , q , ζ ) ≥ 128 \Lambda_0(\alpha,s,q,\zeta) \ge128 Λ 0 ( α , s , q , ζ ) ≥ 128 and C ( α , s , q , ζ ) < ∞ C(\alpha,s,q,\zeta) < \infty C ( α , s , q , ζ ) < ∞ such that, for every Λ ≥ Λ 0 \Lambda\ge\Lambda_0 Λ ≥ Λ 0 , M ≥ m M \ge m M ≥ m and κ ∈ K M \kappa\in\mathcal{K}_M κ ∈ K M ,
ε m 2 β κ m ≤ 20 κ m − 1 , ∥ χ m , k κ m ∥ L ∞ ( R × T 2 ) ≤ C ε ˉ m ≤ C ε m − 1 3 δ ρ ( m ) . (96) \frac{\varepsilon_m^{2\beta}}{\kappa_m}\le20\kappa_{m-1},\qquad\|\chi^{\kappa_m}_{m,k}\|_{L^\infty(\mathbb{R}\times\mathbb{T}^2)}\le C\bar{\varepsilon}_m\le C\varepsilon_{m-1}^{3\delta}\rho(m). \tag*{(96)} κ m ε m 2 β ≤ 20 κ m − 1 , ∥ χ m , k κ m ∥ L ∞ ( R × T 2 ) ≤ C ε ˉ m ≤ C ε m − 1 3 δ ρ ( m ) . (96)
This is the corrector of [5 (3.6)] ; (b) sharpens [5 (3.10)] , and (c) bounds χ m , k R \chi^{\mathrm{R}}_{m,k} χ m , k R by a constant times the corrector amplitude ε ˉ m \bar{\varepsilon}_m ε ˉ m of Part I.
Proof. (a) The integral in (94) is ∫ − ∞ t exp ( − 4 π 2 ν ( t − t ′ ) / ε m 2 ) ζ ~ m , k ( t ′ ) d t ′ \int_{-\infty}^{t} \exp(-4\pi^{2}\nu(t-t')/\varepsilon_m^{2})\widetilde{\zeta}_{m,k}(t')\,\mathrm{d}t' ∫ − ∞ t exp ( − 4 π 2 ν ( t − t ′ ) / ε m 2 ) ζ m , k ( t ′ ) d t ′ ; it vanishes before the pulse, and its derivative is ζ ~ m , k ( t ) \widetilde{\zeta}_{m,k}(t) ζ m , k ( t ) minus 4 π 2 ν / ε m 2 4\pi^{2}\nu/\varepsilon_m^{2} 4 π 2 ν / ε m 2 times itself. By (92) , − ν Δ ∇ ⊥ ψ m , k = ( 4 π 2 ν / ε m 2 ) ∇ ⊥ ψ m , k -\nu\Delta\nabla^{\perp}\psi_{m,k}=(4\pi^{2}\nu/\varepsilon_m^{2})\nabla^{\perp}\psi_{m,k} − ν Δ ∇ ⊥ ψ m , k = ( 4 π 2 ν / ε m 2 ) ∇ ⊥ ψ m , k , which gives the equation; the transport term vanishes because the gradient of a function of y ⋅ e 1 ⊥ y\cdot e_{1}^{\perp} y ⋅ e 1 ⊥ is orthogonal to e ( k ) e_{(k)} e ( k ) . (b) The integral is at most ε m 2 / ( 4 π 2 ν ) \varepsilon_m^{2}/(4\pi^{2}\nu) ε m 2 / ( 4 π 2 ν ) , and ∣ ∇ ⊥ ψ m , k ∣ ≤ 2 π ε m β − 1 |\nabla^{\perp}\psi_{m,k}|\leq2\pi\varepsilon_m^{\beta-1} ∣ ∇ ⊥ ψ m , k ∣ ≤ 2 π ε m β − 1 . (c) By Proposition 2.1(a) and (c), and as ε m − 1 ≤ Λ − 1 \varepsilon_{m-1}\leq\Lambda^{-1} ε m − 1 ≤ Λ − 1 (Lemma 7.1(a)), for Λ 0 \Lambda_0 Λ 0 large,
κ m − 1 ≥ ( 9 80 − C ε m − 1 4 δ − C ε m − 1 δ ) ε m 2 β κ m ≥ 1 20 ε m 2 β κ m , \kappa_{m-1}\geq\left(\frac{9}{80}-C\varepsilon_{m-1}^{4\delta}-C\varepsilon_{m-1}^{\delta}\right)\frac{\varepsilon_m^{2\beta}}{\kappa_m}\geq\frac{1}{20}\frac{\varepsilon_m^{2\beta}}{\kappa_m}, κ m − 1 ≥ ( 80 9 − C ε m − 1 4 δ − C ε m − 1 δ ) κ m ε m 2 β ≥ 20 1 κ m ε m 2 β ,
ε m β + 1 κ m = ε m ( ε m 2 β κ m ) 1 / 2 κ m − 1 / 2 ≤ C ε m ( κ m − 1 κ m ) 1 / 2 = C ε ˉ m . \frac{\varepsilon_m^{\beta+1}}{\kappa_m}=\varepsilon_m\left(\frac{\varepsilon_m^{2\beta}}{\kappa_m}\right)^{1/2}\kappa_m^{-1/2}\leq C\varepsilon_m\left(\frac{\kappa_{m-1}}{\kappa_m}\right)^{1/2}=C\bar{\varepsilon}_m. κ m ε m β + 1 = ε m ( κ m ε m 2 β ) 1/2 κ m − 1/2 ≤ C ε m ( κ m κ m − 1 ) 1/2 = C ε ˉ m .
the first line is the first bound of (96) , and the second, with (b), gives the second. The third follows from ( κ m − 1 / κ m ) 1 / 2 ≤ C ε − q γ (\kappa_{m-1}/\kappa_m)^{1/2}\leq C\varepsilon^{-q\gamma} ( κ m − 1 / κ m ) 1/2 ≤ C ε − q γ , ρ ( m ) ≥ c ε m − 1 1 + γ / 2 + δ \rho(m)\geq c\varepsilon_{m-1}^{1+\gamma/2+\delta} ρ ( m ) ≥ c ε m − 1 1 + γ /2 + δ (Proposition 2.1(c)), ε m ≤ 2 ε m − 1 q \varepsilon_m\leq2\varepsilon_{m-1}^{q} ε m ≤ 2 ε m − 1 q (Lemma 7.1(a)) and q − q γ − 1 − γ 2 − δ = 3 δ q-q\gamma-1-\frac{\gamma}{2}-\delta=3\delta q − q γ − 1 − 2 γ − δ = 3 δ , by (5) .
Lemma 10.2
Lemma 10.2 (The averaged flux is the flux of the correctors). Let m ≥ 1 m\geq1 m ≥ 1 and ν > 0 \nu>0 ν > 0 .
(a) For every odd k k k , t ∈ R t\in\mathbb{R} t ∈ R and e ∈ R 2 e\in\mathbb{R}^{2} e ∈ R 2 , with the term k k k of (18) on the right,
∫ [ 0 , 1 ] 2 ( ν I + ζ ~ m , k ( t ) ψ m , k ( y ) ( 0 − 1 1 0 ) ) ( e + ∇ ( e ⋅ χ m , k ν ( t , y ) ) ) d y = ν e + 2 π 2 ε m 2 β − 2 ζ ~ m , k ( t ) ( ∫ 0 ∞ exp ( − 4 π 2 ν t ′ ε m 2 ) ζ ~ m , k ( t − t ′ ) d t ′ ) ( e ( k ) ⊗ e ( k ) ) e . \int_{[0,1]^{2}}\left(\nu I+\widetilde{\zeta}_{m,k}(t)\psi_{m,k}(y)\begin{pmatrix}0&-1\\1&0\end{pmatrix}\right)\left(e+\nabla(e\cdot\chi_{m,k}^{\nu}(t,y))\right)\,\mathrm{d}y
=\nu e+2\pi^{2}\varepsilon_m^{2\beta-2}\widetilde{\zeta}_{m,k}(t)\left(\int_{0}^{\infty}\exp\left(-\frac{4\pi^{2}\nu t'}{\varepsilon_m^{2}}\right)\widetilde{\zeta}_{m,k}(t-t')\,\mathrm{d}t'\right)(e_{(k)}\otimes e_{(k)})e. ∫ [ 0 , 1 ] 2 ( ν I + ζ m , k ( t ) ψ m , k ( y ) ( 0 1 − 1 0 ) ) ( e + ∇ ( e ⋅ χ m , k ν ( t , y )) ) d y = ν e + 2 π 2 ε m 2 β − 2 ζ m , k ( t ) ( ∫ 0 ∞ exp ( − ε m 2 4 π 2 ν t ′ ) ζ m , k ( t − t ′ ) d t ′ ) ( e ( k ) ⊗ e ( k ) ) e .
(b) The glued corrector χ m ν : = ∑ k ∈ 2 Z + 1 H m , k χ m , k ν \chi_m^{\nu}:=\sum_{k\in2\mathbb{Z}+1}H_{m,k}\chi_{m,k}^{\nu} χ m ν := ∑ k ∈ 2 Z + 1 H m , k χ m , k ν satisfies, for every t ∈ R t\in\mathbb{R} t ∈ R and e ∈ R 2 e\in\mathbb{R}^{2} e ∈ R 2 ,
∫ [ 0 , 1 ] 2 ( ν I + ψ m ( t , y ) ( 0 − 1 1 0 ) ) ( e + ∇ ( e ⋅ χ m ν ( t , y ) ) ) d y = J m ν ( t ) e , (97) \int_{[0,1]^{2}}\left(\nu I+\psi_m(t,y)\begin{pmatrix}0&-1\\1&0\end{pmatrix}\right)\left(e+\nabla(e\cdot\chi_m^{\nu}(t,y))\right)\,\mathrm{d}y=J_m^{\nu}(t)e, \tag*{(97)} ∫ [ 0 , 1 ] 2 ( ν I + ψ m ( t , y ) ( 0 1 − 1 0 ) ) ( e + ∇ ( e ⋅ χ m ν ( t , y )) ) d y = J m ν ( t ) e , (97)
and there exists C ( α , q , ζ ) < ∞ C(\alpha,q,\zeta)<\infty C ( α , q , ζ ) < ∞ such that
∂ t χ m ν + ∇ ⊥ ψ m ⋅ ∇ χ m ν − ν Δ χ m ν + ∇ ⊥ ψ m = ∑ k ∈ 2 Z + 1 H m , k ′ χ m , k ν , \partial_t\chi_m^{\nu}+\nabla^{\perp}\psi_m\cdot\nabla\chi_m^{\nu}-\nu\Delta\chi_m^{\nu}+\nabla^{\perp}\psi_m=\sum_{k\in2\mathbb{Z}+1}H'_{m,k}\chi_{m,k}^{\nu}, ∂ t χ m ν + ∇ ⊥ ψ m ⋅ ∇ χ m ν − ν Δ χ m ν + ∇ ⊥ ψ m = k ∈ 2 Z + 1 ∑ H m , k ′ χ m , k ν ,
∥ ∑ k ∈ 2 Z + 1 H m , k ′ χ m , k ν ∥ L ∞ ( R × T 2 ) ≤ C ε m β + 1 ν τ m exp ( − π 2 ν τ m 3 ε m 2 ) . (98) \left\|\sum_{k\in2\mathbb{Z}+1}H'_{m,k}\chi_{m,k}^{\nu}\right\|_{L^{\infty}(\mathbb{R}\times\mathbb{T}^{2})}\leq\frac{C\varepsilon_m^{\beta+1}}{\nu\tau_m}\exp\left(-\frac{\pi^{2}\nu\tau_m}{3\varepsilon_m^{2}}\right). \tag*{(98)} k ∈ 2 Z + 1 ∑ H m , k ′ χ m , k ν L ∞ ( R × T 2 ) ≤ ν τ m C ε m β + 1 exp ( − 3 ε m 2 π 2 ν τ m ) . (98)
The integrand of (97) is the flux of the corrected coordinate y ⋅ e + e ⋅ χ m ν y\cdot e+e\cdot\chi_m^{\nu} y ⋅ e + e ⋅ χ m ν (the divergence form after Lemma 7.9, with I I I in place of K m , l K_{m,l} K m , l ; [5 (3.15), (3.18)] ), and the glued correctors solve the cell problem up to (98) ([5 (3.11)–(3.13)] ).
Proof. (a) Let k ∈ 4 Z + 1 k\in4\mathbb{Z}+1 k ∈ 4 Z + 1 (the other case is symmetric). Then ψ m , k = ε m β sin ( 2 π y 1 / ε m ) \psi_{m,k}=\varepsilon_m^{\beta}\sin(2\pi y_1/\varepsilon_m) ψ m , k = ε m β sin ( 2 π y 1 / ε m ) , and χ m , k ν \chi_{m,k}^{\nu} χ m , k ν , ∇ ( e ⋅ χ m , k ν ) \nabla(e\cdot\chi_{m,k}^{\nu}) ∇ ( e ⋅ χ m , k ν ) are − 2 π ε m β − 1 cos ( 2 π y 1 / ε m ) e 2 -2\pi\varepsilon_m^{\beta-1}\cos(2\pi y_1/\varepsilon_m)e_2 − 2 π ε m β − 1 cos ( 2 π y 1 / ε m ) e 2 and 4 π 2 ε m 2 β − 2 sin ( 2 π y 1 / ε m ) ( e ⋅ e 2 ) e 1 4\pi^{2}\varepsilon_m^{2\beta-2}\sin(2\pi y_1/\varepsilon_m)(e\cdot e_2)e_1 4 π 2 ε m 2 β − 2 sin ( 2 π y 1 / ε m ) ( e ⋅ e 2 ) e 1 times the integral in (94) . Only the product of ψ m , k \psi_{m,k} ψ m , k and ∇ ( e ⋅ χ m , k ν ) \nabla(e\cdot\chi_{m,k}^{\nu}) ∇ ( e ⋅ χ m , k ν ) has nonzero mean; since ( 0 − 1 1 0 ) e 1 = e 2 \begin{pmatrix}0&-1\\1&0\end{pmatrix}e_1=e_2 ( 0 1 − 1 0 ) e 1 = e 2 and ∫ sin 2 = 1 2 \int\sin^{2}=\frac{1}{2} ∫ sin 2 = 2 1 , its mean is 2 π 2 ε m 2 β − 2 ζ ~ m , k ( t ) 2\pi^{2}\varepsilon_m^{2\beta-2}\widetilde{\zeta}_{m,k}(t) 2 π 2 ε m 2 β − 2 ζ m , k ( t ) times the integral times ( e ⋅ e 2 ) e 2 (e\cdot e_2)e_2 ( e ⋅ e 2 ) e 2 , and e ( k ) = e 2 e_{(k)}=e_2 e ( k ) = e 2 .
(b) If ψ m ( t , ⋅ ) ≠ 0 \psi_m(t,\cdot)\neq0 ψ m ( t , ⋅ ) = 0 , exactly one pulse k k k is active, H m , k ( t ) = 1 H_{m,k}(t)=1 H m , k ( t ) = 1 and the other cutoffs vanish at t t t , as ∣ t − k ′ τ m ∣ ≥ 4 3 τ m |t-k'\tau_m|\geq\frac{4}{3}\tau_m ∣ t − k ′ τ m ∣ ≥ 3 4 τ m for odd k ′ ≠ k k'\neq k k ′ = k ; so χ m ν ( t ) = χ m , k ν ( t ) \chi_m^{\nu}(t)=\chi_{m,k}^{\nu}(t) χ m ν ( t ) = χ m , k ν ( t ) , and (a) gives (97) . Otherwise both sides are ν e \nu e ν e (Lemma 7.10(i)). The product rule and (95) give the equation, since H m , k = 1 H_{m,k}=1 H m , k = 1 where ζ ~ m , k ≠ 0 \widetilde{\zeta}_{m,k}\ne0 ζ m , k = 0 and the transport term vanishes. Finally, H m , k ′ ≠ 0 H'_{m,k}\ne0 H m , k ′ = 0 only for 3 4 τ m ≤ ∣ t − k τ m ∣ ≤ 5 4 τ m \frac{3}{4}\tau_m\le|t-k\tau_m|\le\frac{5}{4}\tau_m 4 3 τ m ≤ ∣ t − k τ m ∣ ≤ 4 5 τ m , where ∣ H m , k ′ ∣ ≤ C / τ m |H'_{m,k}|\le C/\tau_m ∣ H m , k ′ ∣ ≤ C / τ m and either χ m , k ν = 0 \chi^{\nu}_{m,k}=0 χ m , k ν = 0 or the pulse ended at least 1 12 τ m \frac{1}{12}\tau_m 12 1 τ m earlier; after the pulse the integral in (94) decays at the rate 4 π 2 ν / ε m 2 4\pi^2\nu/\varepsilon_m^2 4 π 2 ν / ε m 2 , so ∣ χ m , k ν ∣ ≤ ( ε m β + 1 / 2 π ν ) exp ( − π 2 ν τ m / ( 3 ε m 2 ) ) |\chi^{\nu}_{m,k}|\le(\varepsilon_m^{\beta+1}/2\pi\nu)\exp(-\pi^2\nu\tau_m/(3\varepsilon_m^2)) ∣ χ m , k ν ∣ ≤ ( ε m β + 1 /2 π ν ) exp ( − π 2 ν τ m / ( 3 ε m 2 )) there. □
The covariance
Lemma 10.3 (The covariance). (a) There exists C ( ζ ) < ∞ C(\zeta)<\infty C ( ζ ) < ∞ such that, for every m ≥ 1 m\ge1 m ≥ 1 , ν > 0 \nu>0 ν > 0 and t , t ′ ∈ R t,t'\in\mathbb{R} t , t ′ ∈ R ,
∣ ∫ t ′ t ( J m ν ( t ′ ′ ) − K ‾ m ( ν ) I ) d t ′ ′ ∣ ≤ C ε m 2 β ν ( τ m ′ + ε m 2 h m ν τ m ) . (99) \left|\int_{t'}^{t}\left(\mathbf{J}^{\nu}_{m}(t'')-\overline{K}_{m}(\nu)I\right)\,\mathrm{d}t''\right|\le C\frac{\varepsilon_m^{2\beta}}{\nu}\left(\tau'_m+\frac{\varepsilon_m^2h_m}{\nu\tau_m}\right). \tag*{(99)} ∫ t ′ t ( J m ν ( t ′′ ) − K m ( ν ) I ) d t ′′ ≤ C ν ε m 2 β ( τ m ′ + ν τ m ε m 2 h m ) . (99)
(b) There exist Λ 0 ( α , s , q , ζ ) ≥ 128 \Lambda_0(\alpha,s,q,\zeta)\ge128 Λ 0 ( α , s , q , ζ ) ≥ 128 and C ( α , s , q , ζ ) < ∞ C(\alpha,s,q,\zeta)<\infty C ( α , s , q , ζ ) < ∞ such that, for every Λ ≥ Λ 0 \Lambda\ge\Lambda_0 Λ ≥ Λ 0 , M ≥ 2 M\ge2 M ≥ 2 , κ ∈ K M \kappa\in K_M κ ∈ K M , 2 ≤ m ≤ M 2\le m\le M 2 ≤ m ≤ M and t , t ′ ∈ R t,t'\in\mathbb{R} t , t ′ ∈ R , the covariance (39) satisfies
∣ Γ m ( t ) ∣ + ∣ Γ m ( t ) − Γ m ( t ′ ) ∣ ≤ C κ m − 1 τ m ′ ≤ C ε m δ ρ ( m ) 2 . (100) |\Gamma_m(t)|+|\Gamma_m(t)-\Gamma_m(t')|\le C\kappa_m^{-1}\tau'_m\le C\varepsilon_m^{\delta}\rho(m)^2. \tag*{(100)} ∣ Γ m ( t ) ∣ + ∣ Γ m ( t ) − Γ m ( t ′ ) ∣ ≤ C κ m − 1 τ m ′ ≤ C ε m δ ρ ( m ) 2 . (100)
The matrices are diagonal (Lemma 7.10(i)); § 6 uses the two-time form (b).
Proof. (a) We compare J m ν \mathbf{J}^{\nu}_m J m ν with the flux J m ν , f a s t : = ν I + ( ε m 2 β / 2 ν ) ∑ k ∈ 2 Z + 1 ζ m , k 2 e ( k ) ⊗ e ( k ) \mathbf{J}^{\nu,\mathrm{fast}}_m:=\nu I+(\varepsilon_m^{2\beta}/2\nu)\sum_{k\in2\mathbb{Z}+1}\zeta_{m,k}^2e(k)\otimes e(k) J m ν , fast := ν I + ( ε m 2 β /2 ν ) ∑ k ∈ 2 Z + 1 ζ m , k 2 e ( k ) ⊗ e ( k ) of pulses held at their current value and without slow cutoffs. By Step 1 of the proof of Proposition 2.1(a), holding the pulses changes J m ν \mathbf{J}^{\nu}_m J m ν by at most C ( ε m 2 β / ν ) ε m 2 / ( ν τ m ) C(\varepsilon_m^{2\beta}/\nu)\varepsilon_m^2/(\nu\tau_m) C ( ε m 2 β / ν ) ε m 2 / ( ν τ m ) at each time; removing the slow cutoffs changes it only within 2 τ m ′ 2\tau'_m 2 τ m ′ of the ends of the blocks, where the slow cutoffs differ from 1 1 1 by (46) , as in (61) . So ∫ ∣ J m ν − J m ν , f a s t ∣ ≤ C ( ε m 2 β / ν ) ( τ m ′ + ε m 2 h m / ( ν τ m ) ) \int|\mathbf{J}^{\nu}_m-\mathbf{J}^{\nu,\mathrm{fast}}_m|\le C(\varepsilon_m^{2\beta}/\nu)(\tau'_m+\varepsilon_m^2h_m/(\nu\tau_m)) ∫ ∣ J m ν − J m ν , fast ∣ ≤ C ( ε m 2 β / ν ) ( τ m ′ + ε m 2 h m / ( ν τ m )) over every interval of length 4 h m 4h_m 4 h m . The fast flux is 4 τ m 4\tau_m 4 τ m -periodic with mean ( ν + 9 80 ε m 2 β / ν ) I (\nu+\frac{9}{80}\varepsilon_m^{2\beta}/\nu)I ( ν + 80 9 ε m 2 β / ν ) I , since each period holds two odd slots, one for each direction e ( k ) e(k) e ( k ) , and ∫ R ζ m , k 2 = 9 10 τ m \int_{\mathbb{R}}\zeta_{m,k}^2=\frac{9}{10}\tau_m ∫ R ζ m , k 2 = 10 9 τ m , so its centred primitive is at most C ( ε m 2 β / ν ) τ m C(\varepsilon_m^{2\beta}/\nu)\tau_m C ( ε m 2 β / ν ) τ m . The flux J m ν \mathbf{J}^{\nu}_m J m ν is 4 h m 4h_m 4 h m -periodic with mean K ‾ m ( ν ) I \overline{K}_m(\nu)I K m ( ν ) I (Lemma 7.10), and 4 h m 4h_m 4 h m is a multiple of 4 τ m 4\tau_m 4 τ m , so the two means differ by at most ( 4 h m ) − 1 (4h_m)^{-1} ( 4 h m ) − 1 times the integral above. The left side of (99) is unchanged when t t t or t ′ t' t ′ moves by 4 h m 4h_m 4 h m , since the integral of J m ν − K ‾ m ( ν ) I \mathbf{J}^{\nu}_m-\overline{K}_m(\nu)I J m ν − K m ( ν ) I over a period vanishes; so it suffices to bound the centred primitive over one period. Adding the errors, with τ m ≤ τ m ′ \tau_m\le\tau'_m τ m ≤ τ m ′ , gives (99) .
(b) By (39) , Γ m ( 0 ) = 0 \Gamma_m(0)=0 Γ m ( 0 ) = 0 and Γ m ( t ) − Γ m ( t ′ ) = ∫ t ′ t ( J m κ m ( t ′ ′ ) − κ m − 1 I ) d t ′ ′ \Gamma_m(t)-\Gamma_m(t')=\int_{t'}^t(\mathbf{J}^{\kappa_m}_m(t'')-\kappa_m^{-1}I)\,\mathrm{d}t'' Γ m ( t ) − Γ m ( t ′ ) = ∫ t ′ t ( J m κ m ( t ′′ ) − κ m − 1 I ) d t ′′ . We take ν = κ m \nu=\kappa_m ν = κ m in (a), so that K ‾ m ( κ m ) = κ m − 1 \overline{K}_m(\kappa_m)=\kappa_m^{-1} K m ( κ m ) = κ m − 1 by (21) , and it suffices to bound the right side of (99) by C κ m − 1 τ m ′ C\kappa_m^{-1}\tau'_m C κ m − 1 τ m ′ . By (96) , Proposition 2.1(c) and Lemma 7.1(b), ε m 2 β / κ m ≤ C κ m − 1 \varepsilon_m^{2\beta}/\kappa_m\le C\kappa_m^{-1} ε m 2 β / κ m ≤ C κ m − 1 and ε m 2 h m / ( κ m τ m ) ≤ C ε m 4 δ h m ≤ C τ m ′ \varepsilon_m^2h_m/(\kappa_m\tau_m)\le C\varepsilon_m^{4\delta}h_m\le C\tau'_m ε m 2 h m / ( κ m τ m ) ≤ C ε m 4 δ h m ≤ C τ m ′ , because h m / τ m ′ = 4 ⌈ ε m − δ ⌉ + 1 ≤ 9 ε m − δ h_m/\tau'_m=4\lceil\varepsilon_m^{-\delta}\rceil+1\le9\varepsilon_m^{-\delta} h m / τ m ′ = 4 ⌈ ε m − δ ⌉ + 1 ≤ 9 ε m − δ ; and κ m − 1 τ m ′ ≤ 1 4 ε m δ κ m − 1 − 1 h m = 1 4 ε m δ ρ ( m ) 2 \kappa_m^{-1}\tau'_m\le\frac14\varepsilon_m^{\delta}\kappa_{m-1}^{-1}h_m=\frac14\varepsilon_m^{\delta}\rho(m)^2 κ m − 1 τ m ′ ≤ 4 1 ε m δ κ m − 1 − 1 h m = 4 1 ε m δ ρ ( m ) 2 , as ρ ( m ) = ( κ m − 1 − 1 h m ) 1 / 2 \rho(m)=(\kappa_{m-1}^{-1}h_m)^{1/2} ρ ( m ) = ( κ m − 1 − 1 h m ) 1/2 . □
The cell inverse, the finite lift and the means
In this subsection and the next one, M ≥ 2 M\ge2 M ≥ 2 , κ ∈ K M \kappa\in K_M κ ∈ K M , 2 ≤ m ≤ M 2\le m\le M 2 ≤ m ≤ M , k k k is odd, N c o r ≥ 1 N_{\mathrm{cor}}\ge1 N cor ≥ 1 , and K ( t , y ) K(t,y) K ( t , y ) is a conductivity on the windown of pulse k k k : a symmetric matrix, measurable in t t t , smooth in y y y , with 3 4 I ⪯ K ⪯ 5 4 I \frac{3}{4}I\preceq K\preceq\frac{5}{4}I 4 3 I ⪯ K ⪯ 4 5 I , for instance K m , l , m , k K_{m,l,m,k} K m , l , m , k (Lemma 7.7(a), Λ \Lambda Λ large). The cell operator is the part of ∂ t + ∇ ⊥ ψ m ⋅ ∇ − κ m ∇ ⋅ ( K ∇ ) \partial_t+\nabla^\perp\psi_m\cdot\nabla-\kappa_m\nabla\cdot(K\nabla) ∂ t + ∇ ⊥ ψ m ⋅ ∇ − κ m ∇ ⋅ ( K ∇ ) , applied to a form A \mathcal{A} A read at z = y / ε m z=y/\varepsilon_m z = y / ε m , without derivative in y y y :
L 0 A : = ∂ t A + ε m − 1 ( ∇ ⊥ ψ m ) ( t , ε m z ) ⋅ ∇ z A − κ m ε m − 2 ∇ z ⋅ ( K ∇ z A ) . (101) \mathcal{L}_0\mathcal{A}:=\partial_t\mathcal{A}+\varepsilon_m^{-1}(\nabla^\perp\psi_m)(t,\varepsilon_m z)\cdot\nabla_z\mathcal{A}-\kappa_m\varepsilon_m^{-2}\nabla_z\cdot(K\nabla_z\mathcal{A}). \tag*{(101)} L 0 A := ∂ t A + ε m − 1 ( ∇ ⊥ ψ m ) ( t , ε m z ) ⋅ ∇ z A − κ m ε m − 2 ∇ z ⋅ ( K ∇ z A ) . (101)
On transverse forms its shear term vanishes by (92) , and L 0 = ∂ t − κ m ε m − 2 ( e k ⊥ ⋅ K e k ⊥ ) ( e k ⊥ ⋅ ∇ z ) 2 \mathcal{L}_0=\partial_t-\kappa_m\varepsilon_m^{-2}(e_k^\perp\cdot Ke_k^\perp)(e_k^\perp\cdot\nabla_z)^2 L 0 = ∂ t − κ m ε m − 2 ( e k ⊥ ⋅ K e k ⊥ ) ( e k ⊥ ⋅ ∇ z ) 2 : the cell problems across the pulse are one-dimensional heat equations.
Lemma 10.4 (The cell inverse). Let F 0 F^0 F 0 and F F F be transverse forms of zero average in z z z , measurable in t t t , of class C 2 C^2 C 2 and C 3 C^3 C 3 in z z z respectively, whose derivatives in z z z up to these orders are bounded on [ ( k − 61 / 48 ) τ m , ( k + 61 / 48 ) τ m ] × R 2 × R 2 [(k-61/48)\tau_m,(k+61/48)\tau_m]\times\mathbb{R}^2\times\mathbb{R}^2 [( k − 61/48 ) τ m , ( k + 61/48 ) τ m ] × R 2 × R 2 . There exists exactly one energy solution A \mathcal{A} A , a transverse form of zero average in z z z with, for every y ∈ R 2 y\in\mathbb{R}^2 y ∈ R 2 , ∇ z A ( ⋅ , y , ⋅ ) ∈ L 2 ( [ ( k − 61 / 48 ) τ m , ( k + 61 / 48 ) τ m ] × [ 0 , 1 ] 2 ) \nabla_z\mathcal{A}(\cdot,y,\cdot)\in L^2([(k-61/48)\tau_m,(k+61/48)\tau_m]\times[0,1]^2) ∇ z A ( ⋅ , y , ⋅ ) ∈ L 2 ([( k − 61/48 ) τ m , ( k + 61/48 ) τ m ] × [ 0 , 1 ] 2 ) and A ( ⋅ , y , ⋅ ) ∈ C ( [ ( k − 61 / 48 ) τ m , ( k + 61 / 48 ) τ m ] ; L 2 ( [ 0 , 1 ] 2 ) ) \mathcal{A}(\cdot,y,\cdot)\in C([(k-61/48)\tau_m,(k+61/48)\tau_m];L^2([0,1]^2)) A ( ⋅ , y , ⋅ ) ∈ C ([( k − 61/48 ) τ m , ( k + 61/48 ) τ m ] ; L 2 ([ 0 , 1 ] 2 )) , of the following problem, in the integrated form in t t t tested against smooth periodic functions of z z z , for every y y y :
{ L 0 A = F 0 + e ( k ) ⊥ ⋅ ∇ z F in ( ( k − 61 48 ) τ m , ( k + 61 48 ) τ m ) × R 2 × R 2 , A = 0 on { ( k − 61 48 ) τ m } × R 2 × R 2 . (102) \begin{cases}
\mathcal{L}_0\mathcal{A}=F^0+e_{(k)}^\perp\cdot\nabla_zF & \text{in }\left((k-\frac{61}{48})\tau_m,(k+\frac{61}{48})\tau_m\right)\times\mathbb{R}^2\times\mathbb{R}^2,\\
\mathcal{A}=0 & \text{on }\{(k-\frac{61}{48})\tau_m\}\times\mathbb{R}^2\times\mathbb{R}^2. \tag*{(102)}
\end{cases} { L 0 A = F 0 + e ( k ) ⊥ ⋅ ∇ z F A = 0 in ( ( k − 48 61 ) τ m , ( k + 48 61 ) τ m ) × R 2 × R 2 , on {( k − 48 61 ) τ m } × R 2 × R 2 . (102)
it is the cell inverse of F 0 + e ( k ) ⊥ ⋅ ∇ z F F^0+e_{(k)}^\perp\cdot\nabla_zF F 0 + e ( k ) ⊥ ⋅ ∇ z F . For every ξ ∈ Z ∖ { 0 } \xi\in\mathbb{Z}\setminus\{0\} ξ ∈ Z ∖ { 0 } , y ∈ R 2 y\in\mathbb{R}^2 y ∈ R 2 and t ∈ [ ( k − 61 / 48 ) τ m , ( k + 61 / 48 ) τ m ] t\in[(k-61/48)\tau_m,(k+61/48)\tau_m] t ∈ [( k − 61/48 ) τ m , ( k + 61/48 ) τ m ] ,
A ^ ( t , y ; ξ ) = ∫ ( k − 61 / 48 ) τ m t exp ( − 4 π 2 ξ 2 κ m ε m 2 ∫ t ′ t e ( k ) ⊥ ⋅ K ( t ′ ′ , y ) e ( k ) ⊥ d t ′ ′ ) ( F 0 ^ + 2 π i ξ F ^ ) ( t ′ , y ; ξ ) d t ′ , (103) \widehat{\mathcal{A}}(t,y;\xi)=\int_{(k-61/48)\tau_m}^{t}\exp\left(-\frac{4\pi^2\xi^2\kappa_m}{\varepsilon_m^2}\int_{t'}^{t}e_{(k)}^\perp\cdot\mathbf{K}(t'',y)e_{(k)}^\perp\,dt''\right)\left(\widehat{F^0}+2\pi i\xi\widehat{F}\right)(t',y;\xi)\,dt', \tag*{(103)} A ( t , y ; ξ ) = ∫ ( k − 61/48 ) τ m t exp ( − ε m 2 4 π 2 ξ 2 κ m ∫ t ′ t e ( k ) ⊥ ⋅ K ( t ′′ , y ) e ( k ) ⊥ d t ′′ ) ( F 0 + 2 π i ξ F ) ( t ′ , y ; ξ ) d t ′ , (103)
∥ A ^ ( ⋅ , y ; ξ ) ∥ L ∞ ( [ ( k − 61 / 48 ) τ m , ( k + 61 / 48 ) τ m ] ) ≤ ε m 2 3 π ξ 2 κ m ∥ F 0 ^ ( ⋅ , y ; ξ ) ∥ L ∞ ( [ ( k − 61 / 48 ) τ m , ( k + 61 / 48 ) τ m ] ) + 2 ε m 2 3 π ∣ ξ ∣ κ m ∥ F ^ ( ⋅ , y ; ξ ) ∥ L ∞ ( [ ( k − 61 / 48 ) τ m , ( k + 61 / 48 ) τ m ] ) . \begin{aligned}
\|\widehat{\mathcal{A}}(\cdot,y;\xi)\|_{L^\infty([(k-61/48)\tau_m,(k+61/48)\tau_m])}
&\leq\frac{\varepsilon_m^2}{3\pi\xi^2\kappa_m}\|\widehat{F^0}(\cdot,y;\xi)\|_{L^\infty([(k-61/48)\tau_m,(k+61/48)\tau_m])}\\
&\quad+\frac{2\varepsilon_m^2}{3\pi|\xi|\kappa_m}\|\widehat{F}(\cdot,y;\xi)\|_{L^\infty([(k-61/48)\tau_m,(k+61/48)\tau_m])}.
\end{aligned} ∥ A ( ⋅ , y ; ξ ) ∥ L ∞ ([( k − 61/48 ) τ m , ( k + 61/48 ) τ m ]) ≤ 3 π ξ 2 κ m ε m 2 ∥ F 0 ( ⋅ , y ; ξ ) ∥ L ∞ ([( k − 61/48 ) τ m , ( k + 61/48 ) τ m ]) + 3 π ∣ ξ ∣ κ m 2 ε m 2 ∥ F ( ⋅ , y ; ξ ) ∥ L ∞ ([( k − 61/48 ) τ m , ( k + 61/48 ) τ m ]) .
If K \mathbf{K} K is instead a complex symmetric matrix, measurable in t t t , whose entry e ( k ) ⊥ ⋅ K e ( k ) ⊥ e_{(k)}^\perp\cdot\mathbf{K}e_{(k)}^\perp e ( k ) ⊥ ⋅ K e ( k ) ⊥ is bounded with real part at least 3 4 \frac{3}{4} 4 3 (as in § 5.4 ), the right side of (103) defines a transverse form A \mathcal{A} A of zero average in z z z , the cell inverse at K \mathbf{K} K , with the same bound. The cell inverse of − ( ∇ ⊥ ψ m ) ( t , ε m z ) -(\nabla^\perp\psi_m)(t,\varepsilon_m z) − ( ∇ ⊥ ψ m ) ( t , ε m z ) is the corrector in the conductivity K \mathbf{K} K , written χ K ( t , y , z ) \chi_{\mathbf{K}}(t,y,z) χ K ( t , y , z ) ; χ I ( t , y , z ) = χ m , k m ( t , ε m z ) \chi_{\mathbf{I}}(t,y,z)=\chi^{m}_{m,k}(t,\varepsilon_m z) χ I ( t , y , z ) = χ m , k m ( t , ε m z ) .
Proof. On the mode ξ \xi ξ , L 0 \mathcal{L}_0 L 0 is ∂ t + ( 4 π 2 ξ 2 κ m / ε m 2 ) e ( k ) ⊥ ⋅ K e ( k ) ⊥ \partial_t+(4\pi^2\xi^2\kappa_m/\varepsilon_m^2)e_{(k)}^\perp\cdot\mathbf{K}e_{(k)}^\perp ∂ t + ( 4 π 2 ξ 2 κ m / ε m 2 ) e ( k ) ⊥ ⋅ K e ( k ) ⊥ : Duhamel’s formula gives (103) ; for a conductivity, the energy identity in z z z gives uniqueness. As Re e ( k ) ⊥ ⋅ K e ( k ) ⊥ ≥ 3 4 \operatorname{Re}e_{(k)}^\perp\cdot\mathbf{K}e_{(k)}^\perp\geq\frac{3}{4} Re e ( k ) ⊥ ⋅ K e ( k ) ⊥ ≥ 4 3 , the exponential is at most exp ( − 3 π 2 ξ 2 κ m ( t − t ′ ) / ε m 2 ) \exp(-3\pi^2\xi^2\kappa_m(t-t')/\varepsilon_m^2) exp ( − 3 π 2 ξ 2 κ m ( t − t ′ ) / ε m 2 ) , of integral in t ′ t' t ′ at most ε m 2 / ( 3 π 2 ξ 2 κ m ) \varepsilon_m^2/(3\pi^2\xi^2\kappa_m) ε m 2 / ( 3 π 2 ξ 2 κ m ) , which gives the bound. The source F 0 + e ( k ) ⊥ ⋅ ∇ z F F^0+e_{(k)}^\perp\cdot\nabla_zF F 0 + e ( k ) ⊥ ⋅ ∇ z F is of class C 2 C^2 C 2 in z z z with bounded derivatives, so its modes are O ( ξ − 2 ) O(\xi^{-2}) O ( ξ − 2 ) and, by (103) , those of A \mathcal{A} A are O ( ξ − 4 ) O(\xi^{-4}) O ( ξ − 4 ) ; the series of A \mathcal{A} A converges with two derivatives in z z z , continuously in t t t with values in L 2 ( [ 0 , 1 ] 2 ) L^2([0,1]^2) L 2 ([ 0 , 1 ] 2 ) , and is an energy solution. For K = I \mathbf{K}=\mathbf{I} K = I , (103) on the modes ξ = ± 1 \xi=\pm1 ξ = ± 1 is (94) , since the pulse starts at ( k − 2 3 ) τ m (k-\frac{2}{3})\tau_m ( k − 3 2 ) τ m , inside the window. □
A system on the window of pulse k k k has finitely many components v v v , with masses ω v > 0 \omega_v>0 ω v > 0 , and finitely many sources σ : v ′ → v \sigma:v'\to v σ : v ′ → v . A source has a birth order 2 ≤ d σ ≤ N s r c 2\leq d_\sigma\leq N_{\mathrm{src}} 2 ≤ d σ ≤ N src , its order when created (Subsection 11.1), a factor fac ( σ ) > 0 \operatorname{fac}(\sigma)>0 fac ( σ ) > 0 , a majorant maj ( σ ) ≥ 0 \operatorname{maj}(\sigma)\geq0 maj ( σ ) ≥ 0 , and transverse two-scale forms A σ , d \mathcal{A}_{\sigma,d} A σ , d , 2 ≤ d ≤ d σ 2\leq d\leq d_\sigma 2 ≤ d ≤ d σ , whose values are coefficients from arrays of order d − 1 d-1 d − 1 to R 2 \mathbb{R}^2 R 2 (Section 8); we set A σ , d : = 0 \mathcal{A}_{\sigma,d}:=0 A σ , d := 0 for d > d σ d>d_\sigma d > d σ . The system acts on fields Y = ( Y v ) v Y=(Y_v)_v Y = ( Y v ) v by
∂ t Y v + ∇ ⊥ ψ m ⋅ ∇ Y v − κ m ∇ ⋅ ( K ∇ Y v ) − ∑ σ : v ′ → v ∑ d = 2 d σ ∇ ⋅ ( κ m ε m d − 2 fac ( σ ) A σ , d ( t , y / ε m ) ∇ d − 1 Y v ′ ) ; (104) \partial_tY_v+\nabla^\perp\psi_m\cdot\nabla Y_v-\kappa_m\nabla\cdot(\mathbf{K}\nabla Y_v)
-\sum_{\sigma:v'\to v}\sum_{d=2}^{d_\sigma}\nabla\cdot\left(\kappa_m\varepsilon_m^{d-2}\operatorname{fac}(\sigma)\mathcal{A}_{\sigma,d}(t,y/\varepsilon_m)\nabla^{d-1}Y_{v'}\right); \tag*{(104)} ∂ t Y v + ∇ ⊥ ψ m ⋅ ∇ Y v − κ m ∇ ⋅ ( K ∇ Y v ) − σ : v ′ → v ∑ d = 2 ∑ d σ ∇ ⋅ ( κ m ε m d − 2 fac ( σ ) A σ , d ( t , y / ε m ) ∇ d − 1 Y v ′ ) ; (104)
for the families of Section 11 these are the coefficients (116) in the chart of k k k . A path σ 1 ⋯ σ i \sigma_1\cdots\sigma_i σ 1 ⋯ σ i is a sequence of sources, each starting where the previous one ends; the empty path ∅ \varnothing ∅ has i = 0 i=0 i = 0 . We assume that every path has at most N p a t h N_{\mathrm{path}} N path sources, as for the families of Section 11. On a form A \mathcal{A} A read at z = y / ε m z = y/\varepsilon_m z = y / ε m , the chain rule ∇ = ∇ y + ε m − 1 ∇ z \nabla= \nabla_y + \varepsilon_m^{-1}\nabla_z ∇ = ∇ y + ε m − 1 ∇ z splits (104) by the number of derivatives in y y y into L 0 \mathcal{L}_0 L 0 ,
L 1 A : = ( ∇ ⊥ ψ m ) ( t , ε m z ) ⋅ ∇ y A − κ m ε m − 1 ( ∇ z ⋅ ( K ∇ y A ) + ∇ y ⋅ ( K ∇ z A ) ) , L 2 A : = − κ m ∇ y ⋅ ( K ∇ y A ) , (105) \begin{aligned}
\mathcal{L}_1\mathcal{A} &:= (\nabla^\perp\psi_m)(t,\varepsilon_m z)\cdot\nabla_y\mathcal{A}
-\kappa_m\varepsilon_m^{-1}\left(\nabla_z\cdot(K\nabla_y\mathcal{A})+\nabla_y\cdot(K\nabla_z\mathcal{A})\right),\\
\mathcal{L}_2\mathcal{A} &:= -\kappa_m\nabla_y\cdot(K\nabla_y\mathcal{A}), \tag*{(105)}
\end{aligned} L 1 A L 2 A := ( ∇ ⊥ ψ m ) ( t , ε m z ) ⋅ ∇ y A − κ m ε m − 1 ( ∇ z ⋅ ( K ∇ y A ) + ∇ y ⋅ ( K ∇ z A ) ) , := − κ m ∇ y ⋅ ( K ∇ y A ) , (105)
and the terms − L j s r c [ σ ] -\mathcal{L}^{\mathrm{src}}_j[\sigma] − L j src [ σ ] , where the source parts L j s r c [ σ ] \mathcal{L}^{\mathrm{src}}_j[\sigma] L j src [ σ ] are the parts with exactly j j j derivatives in y y y of the term of σ \sigma σ in (104) without its minus sign, those that fall on A σ , d \mathcal{A}_{\sigma,d} A σ , d included.
The finite lift below collects by degree the terms of the operator (104) applied to the sum of V j [ σ 1 ⋯ σ i ] Y \mathcal{V}_j[\sigma_1\cdots\sigma_i]Y V j [ σ 1 ⋯ σ i ] Y over the degrees and the paths, for a field Y Y Y that solves the averaged equation stated after Proposition 10.6 . The time derivative falls on the coefficients of the correctors, which is the term ∂ t \partial_t ∂ t of L 0 \mathcal{L}_0 L 0 , and on Y Y Y , where the averaged equation replaces it; this gives the last sum of (107) , in which a path splits at the component where the mean ends and the corrector begins. The proof of Lemma 13.2 carries out this computation. In M j , d \mathcal{M}_{j,d} M j , d , as in A σ , d \mathcal{A}_{\sigma,d} A σ , d and in the coefficients of (34) , the order d d d counts the derivatives on the field together with the outer divergence, while in V j , d \mathcal{V}_{j,d} V j , d it counts the derivatives on the field. The means start at d = 2 d=2 d = 2 : in (106) the field Y Y Y appears only differentiated, since V 0 [ ∅ ] Y = Y \mathcal{V}_0[\varnothing]Y=Y V 0 [ ∅ ] Y = Y appears only under a derivative and the correctors of positive degree annihilate constants (Proposition 10.6 (a)).
Definition 10.5 (The finite lift on a pulse). The finite lift of a system on the window of pulse k k k assigns to every path σ 1 ⋯ σ i \sigma_1\cdots\sigma_i σ 1 ⋯ σ i and every degree 0 ≤ j ≤ N c o r 0\le j\le N_{\mathrm{cor}} 0 ≤ j ≤ N cor , the number of derivatives in y y y it carries (on the field or the coefficients), a corrector V j [ σ 1 ⋯ σ i ] \mathcal{V}_j[\sigma_1\cdots\sigma_i] V j [ σ 1 ⋯ σ i ] , a linear map from smooth fields Y ( y ) Y(y) Y ( y ) to transverse two-scale forms, V j [ σ 1 ⋯ σ i ] Y = ∑ d = 0 j V j , d [ σ 1 ⋯ σ i ] ∇ d Y \mathcal{V}_j[\sigma_1\cdots\sigma_i]Y=\sum_{d=0}^j\mathcal{V}_{j,d}[\sigma_1\cdots\sigma_i]\nabla^dY V j [ σ 1 ⋯ σ i ] Y = ∑ d = 0 j V j , d [ σ 1 ⋯ σ i ] ∇ d Y , and to every degree 2 ≤ j ≤ N c o r + N s r c 2\le j\le N_{\mathrm{cor}}+N_{\mathrm{src}} 2 ≤ j ≤ N cor + N src the means M j , d [ σ 1 ⋯ σ i ] ( t , y ) \mathcal{M}_{j,d}[\sigma_1\cdots\sigma_i](t,y) M j , d [ σ 1 ⋯ σ i ] ( t , y ) , 2 ≤ d ≤ j 2\le d\le j 2 ≤ d ≤ j , arrays symmetric in the order of the derivatives. With V 0 [ ∅ ] : = I \mathcal{V}_0[\varnothing]:=I V 0 [ ∅ ] := I , V 0 [ σ 1 ⋯ σ i ] : = 0 \mathcal{V}_0[\sigma_1\cdots\sigma_i]:=0 V 0 [ σ 1 ⋯ σ i ] := 0 for i ≥ 1 i\ge1 i ≥ 1 , and V j [ σ 1 ⋯ σ i ] : = 0 \mathcal{V}_j[\sigma_1\cdots\sigma_i]:=0 V j [ σ 1 ⋯ σ i ] := 0 for j < 0 j<0 j < 0 and j > N c o r j>N_{\mathrm{cor}} j > N cor , they are defined by induction on i i i , then on j j j . The means are the coefficients in
∑ d = 2 j M j , d [ σ 1 ⋯ σ i ] ∇ d − 1 Y : = 1 { j = 2 , i = 0 } κ m K ∇ Y − ( ( ∇ ⊥ ψ m ) ( t , ε m z ) V j − 1 [ σ 1 ⋯ σ i ] Y ) T 2 + ( the part with j − 1 derivatives in y of ∑ d ′ = 2 d σ i κ m ε m d ′ − 2 fac ( σ i ) A σ i , d ′ ( ∇ y + ε m − 1 ∇ z ) d ′ − 1 ∑ j ′ ′ = 0 N c o r V j ′ ′ [ σ 1 ⋯ σ i − 1 ] Y ) T 2 , (106) \begin{aligned}
\sum_{d=2}^{j}\mathcal{M}_{j,d}[\sigma_1\cdots\sigma_i]\nabla^{d-1}Y
&:=\mathbf{1}_{\{j=2,i=0\}}\kappa_mK\nabla Y
-\left((\nabla^\perp\psi_m)(t,\varepsilon_m z)\mathcal{V}_{j-1}[\sigma_1\cdots\sigma_i]Y\right)_{\mathbb{T}^2}\\
&\quad+\left(\text{the part with }j-1\text{ derivatives in }y\text{ of }
\sum_{d'=2}^{d_{\sigma_i}}\kappa_m\varepsilon_m^{d'-2}
\operatorname{fac}(\sigma_i)\mathcal{A}_{\sigma_i,d'}
(\nabla_y+\varepsilon_m^{-1}\nabla_z)^{d'-1}
\sum_{j''=0}^{N_{\mathrm{cor}}}
\mathcal{V}_{j''}[\sigma_1\cdots\sigma_{i-1}]Y\right)_{\mathbb{T}^2}, \tag*{(106)}
\end{aligned} d = 2 ∑ j M j , d [ σ 1 ⋯ σ i ] ∇ d − 1 Y := 1 { j = 2 , i = 0 } κ m K ∇ Y − ( ( ∇ ⊥ ψ m ) ( t , ε m z ) V j − 1 [ σ 1 ⋯ σ i ] Y ) T 2 + the part with j − 1 derivatives in y of d ′ = 2 ∑ d σ i κ m ε m d ′ − 2 fac ( σ i ) A σ i , d ′ ( ∇ y + ε m − 1 ∇ z ) d ′ − 1 j ′′ = 0 ∑ N cor V j ′′ [ σ 1 ⋯ σ i − 1 ] Y T 2 , (106) where, for i = 0 i=0 i = 0 , the last term is absent, and V j ′ ′ \mathcal{V}_{j''} V j ′′ counts as j ′ ′ j'' j ′′ derivatives in y y y . For every 1 ≤ j ≤ N c o r 1\le j\le N_{\mathrm{cor}} 1 ≤ j ≤ N cor , V j [ σ 1 ⋯ σ i ] \mathcal{V}_j[\sigma_1\cdots\sigma_i] V j [ σ 1 ⋯ σ i ] is the cell inverse, coefficient by coefficient, of the part of zero average in z z z of
− ( L 1 V j − 1 [ σ 1 ⋯ σ i ] Y + L 2 V j − 2 [ σ 1 ⋯ σ i ] Y − ∑ j ′ = 0 min { j , N s r c } L j ′ s r c [ σ i ] V j − j ′ [ σ 1 ⋯ σ i − 1 ] Y − ∑ j ′ ′ = 0 j − 2 ∑ i ′ = 0 i V j ′ ′ [ σ i ′ + 1 ⋯ σ i ] ∇ ⋅ ( ∑ d = 2 j − j ′ ′ M j − j ′ ′ , d [ σ 1 ⋯ σ i ′ ] ∇ d − 1 Y ) ) , (107) \begin{aligned}
-\Bigg(&\mathcal{L}_1\mathcal{V}_{j-1}[\sigma_1\cdots\sigma_i]Y
+\mathcal{L}_2\mathcal{V}_{j-2}[\sigma_1\cdots\sigma_i]Y\\
&\quad-\sum_{j'=0}^{\min\{j,N_{\mathrm{src}}\}}
\mathcal{L}^{\mathrm{src}}_{j'}[\sigma_i]\mathcal{V}_{j-j'}[\sigma_1\cdots\sigma_{i-1}]Y\\
&\quad-\sum_{j''=0}^{j-2}\sum_{i'=0}^{i}
\mathcal{V}_{j''}[\sigma_{i'+1}\cdots\sigma_i]\nabla\cdot
\left(\sum_{d=2}^{j-j''}\mathcal{M}_{j-j'',d}[\sigma_1\cdots\sigma_{i'}]\nabla^{d-1}Y\right)\Bigg), \tag*{(107)}
\end{aligned} − ( L 1 V j − 1 [ σ 1 ⋯ σ i ] Y + L 2 V j − 2 [ σ 1 ⋯ σ i ] Y − j ′ = 0 ∑ m i n { j , N src } L j ′ src [ σ i ] V j − j ′ [ σ 1 ⋯ σ i − 1 ] Y − j ′′ = 0 ∑ j − 2 i ′ = 0 ∑ i V j ′′ [ σ i ′ + 1 ⋯ σ i ] ∇ ⋅ d = 2 ∑ j − j ′′ M j − j ′′ , d [ σ 1 ⋯ σ i ′ ] ∇ d − 1 Y ) , (107) with, for i = 0 i=0 i = 0 , no source sum; the parenthesis, for any j ≥ 1 j\ge1 j ≥ 1 , is the *bracket of degree j j j . The finite lift of the system (104) itself is (106) –(107) read for matrices over the components, with the source sum over the sources into each component and matrix products in the last sum.
As L 0 s r c \mathcal{L}^{\mathrm{src}}_0 L 0 src acts along a shorter path and the means of degree j j j use only lower degrees or shorter paths, the induction is well founded. The definition extends to any cell operator with an inverse as in Lemma 10.4 ; the statements below are for transverse forms.
Proposition 10.6 (The finite lift and its means). There exists Λ 0 ( α , s , q , ζ ) ≥ 128 \Lambda_0(\alpha,s,q,\zeta) \ge128 Λ 0 ( α , s , q , ζ ) ≥ 128 with the following property. Let Λ ≥ Λ 0 \Lambda\ge\Lambda_0 Λ ≥ Λ 0 and N ≥ 0 N \ge0 N ≥ 0 , and consider a system on the window of pulse k k k whose majorant matrix, with the entries ( ω v ′ / ω v ) 1 / 2 ∑ σ : v ′ → v fac ( σ ) maj ( σ ) (\omega_{v'}/\omega_v)^{1/2}\sum_{\sigma:v'\to v}\operatorname{fac}(\sigma)\operatorname{maj}(\sigma) ( ω v ′ / ω v ) 1/2 ∑ σ : v ′ → v fac ( σ ) maj ( σ ) as in (117) , has row and column sums at most 1 1 1 , and whose conductivity and forms are measurable in t t t and satisfy, with constants C ( j 1 ) C(j_1) C ( j 1 ) , C ( j 1 , j 2 ) C(j_1,j_2) C ( j 1 , j 2 ) , for every t t t in the window of pulse k k k , y , z ∈ R 2 y,z\in\mathbb{R}^2 y , z ∈ R 2 , 2 ≤ d ≤ N s r c 2\le d\le N_{\mathrm{src}} 2 ≤ d ≤ N src , j 1 ≤ N + 4 ( N p a t h + 1 ) ( N c o r + N s r c ) 2 j_1\le N+4(N_{\mathrm{path}}+1)(N_{\mathrm{cor}}+N_{\mathrm{src}})^2 j 1 ≤ N + 4 ( N path + 1 ) ( N cor + N src ) 2 and j 2 ≤ N + 2 + N p a t h N s r c + 2 N c o r + 2 N s r c j_2\le N+2+N_{\mathrm{path}}N_{\mathrm{src}}+2N_{\mathrm{cor}}+2N_{\mathrm{src}} j 2 ≤ N + 2 + N path N src + 2 N cor + 2 N src ,
ε m − 1 j 1 ∣ ∇ y j 1 K ( t , y ) ∣ ≤ C ( j 1 ) , ε m − 1 j 1 ∣ ∇ y j 1 ∇ z j 2 A σ , d ( t , y , z ) ∣ ≤ C ( j 1 , j 2 ) maj ( σ ) . (108) \varepsilon_{m-1}^{j_1}\lvert\nabla_y^{j_1}K(t,y)\rvert\le C(j_1),\qquad\varepsilon_{m-1}^{j_1}\lvert\nabla_y^{j_1}\nabla_z^{j_2}A_{\sigma,d}(t,y,z)\rvert\le C(j_1,j_2)\operatorname{maj}(\sigma). \tag*{(108)} ε m − 1 j 1 ∣ ∇ y j 1 K ( t , y )∣ ≤ C ( j 1 ) , ε m − 1 j 1 ∣ ∇ y j 1 ∇ z j 2 A σ , d ( t , y , z )∣ ≤ C ( j 1 , j 2 ) maj ( σ ) . (108) Then:
(a) Existence and means. The correctors of positive degree exist, are unique, transverse, of zero average in z z z , zero at the start of the window of pulse k k k , and annihilate constants, and the means are N + 1 N+1 N + 1 times differentiable in y y y . For every path σ 1 ⋯ σ i \sigma_1\cdots\sigma_i σ 1 ⋯ σ i and 1 ≤ j ≤ N c o r + N s r c 1\le j\le N_{\mathrm{cor}}+N_{\mathrm{src}} 1 ≤ j ≤ N cor + N src , the average in z z z of the bracket of degree j j j along it is − ∇ ⋅ ( ∑ d = 2 j M j , d [ σ 1 ⋯ σ i ] ∇ d − 1 Y ) -\nabla\cdot\left(\sum_{d=2}^{j}\mathcal{M}_{j,d}[\sigma_1\cdots\sigma_i]\nabla^{d-1}Y\right) − ∇ ⋅ ( ∑ d = 2 j M j , d [ σ 1 ⋯ σ i ] ∇ d − 1 Y ) , and, for j = 1 j=1 j = 1 , it vanishes.
(b) The cell equations. For every 1 ≤ j ≤ N c o r 1\le j\le N_{\mathrm{cor}} 1 ≤ j ≤ N cor , (107) has zero average in z z z , and V j [ σ 1 ⋯ σ i ] Y \mathcal{V}_j[\sigma_1\cdots\sigma_i]Y V j [ σ 1 ⋯ σ i ] Y solves (10.11) with right side (107) . The means of degree j ≥ N c o r + 2 j\ge N_{\mathrm{cor}}+2 j ≥ N cor + 2 along ∅ \varnothing ∅ vanish.
(c) The averaged flux. If K = I \mathbf{K}=I K = I , then M 2 , 2 [ ∅ ] = J m κ ( t ) \mathcal{M}_{2,2}[\varnothing]=J_m^{\kappa}(t) M 2 , 2 [ ∅ ] = J m κ ( t ) . If, for every t t t in the window of pulse k k k , y ∈ R 2 y\in\mathbb{R}^2 y ∈ R 2 and j 1 j_1 j 1 as in (108) , ε m − 1 j 1 ∣ ∇ y j 1 ( K − I ) ∣ ≤ C ( j 1 ) ε m − 1 2 δ \varepsilon_{m-1}^{j_1}\lvert\nabla_y^{j_1}(\mathbf{K}-I)\rvert\le C(j_1)\varepsilon_{m-1}^{2\delta} ε m − 1 j 1 ∣ ∇ y j 1 ( K − I )∣ ≤ C ( j 1 ) ε m − 1 2 δ , with C ( j 1 ) C(j_1) C ( j 1 ) of (108) , then there exists C C C , depending only on α , s , q , ζ , N c o r , N s r c , N \alpha,s,q,\zeta,N_{\mathrm{cor}},N_{\mathrm{src}},N α , s , q , ζ , N cor , N src , N and the constants of (108) , such that, for every such t t t and y y y and every j 1 ≤ N j_1\le N j 1 ≤ N , ε m − 1 j 1 ∣ ∇ y j 1 ( M 2 , 2 [ ∅ ] − J m κ ( t ) ) ∣ ≤ C κ m − 1 ε m − 1 2 δ \varepsilon_{m-1}^{j_1}\lvert\nabla_y^{j_1}(\mathcal{M}_{2,2}[\varnothing]-J_m^{\kappa}(t))\rvert\le C\kappa_m^{-1}\varepsilon_{m-1}^{2\delta} ε m − 1 j 1 ∣ ∇ y j 1 ( M 2 , 2 [ ∅ ] − J m κ ( t ))∣ ≤ C κ m − 1 ε m − 1 2 δ .
(d) Quiet times. Let N ≥ N s r c − 2 N\ge N_{\mathrm{src}}-2 N ≥ N src − 2 , and let t t t be a time of the window of pulse k k k at which ζ ~ m , k = 0 ∗ a n d t h e f o r m s d o n o t d e p e n d o n ∗ \widetilde{\zeta}_{m,k}=0 *and the forms do not depend on* ζ m , k = 0 ∗ an d t h e f or m s d o n o t d e p e n d o n ∗ z$. Then, at t t t : M 2 , 2 [ ∅ ] = κ m K \mathcal{M}_{2,2}[\varnothing]=\kappa_m\mathbf{K} M 2 , 2 [ ∅ ] = κ m K ; for every path $\sigma of one source, every degree 2 ≤ j ≤ N c o r + N s r c 2\le j\le N_{\mathrm{cor}}+N_{\mathrm{src}} 2 ≤ j ≤ N cor + N src and every smooth field Y Y Y , M j , j [ σ ] ∇ j − 1 Y = κ m ε m j − 2 fac ( σ ) A σ , j ∇ j − 1 Y \mathcal{M}_{j,j}[\sigma]\nabla^{j-1}Y=\kappa_m\varepsilon_m^{j-2}\operatorname{fac}(\sigma)A_{\sigma,j}\nabla^{j-1}Y M j , j [ σ ] ∇ j − 1 Y = κ m ε m j − 2 fac ( σ ) A σ , j ∇ j − 1 Y ; and all other coefficients of the means vanish.
(e) Quiet vanishing. Let t t t be in the window of pulse k k k . If, at all times of the window up to t t t , ζ ~ m , k = 0 ∗ a n d t h e f o r m s d o n o t d e p e n d o n ∗ \widetilde{\zeta}_{m,k}=0 *and the forms do not depend on* ζ m , k = 0 ∗ an d t h e f or m s d o n o t d e p e n d o n ∗ z, ∗ t h e n , f o r e v e r y ∗ , *then, for every* , ∗ t h e n , f or e v er y ∗ 1\le j\le N_{\mathrm{cor}} and every path σ 1 ⋯ σ i \sigma_1\cdots\sigma_i σ 1 ⋯ σ i , V j [ σ 1 ⋯ σ i ] = 0 \mathcal{V}_j[\sigma_1\cdots\sigma_i]=0 V j [ σ 1 ⋯ σ i ] = 0 at t t t .
(f) The system. The finite lift of the system (104) has from v ′ v' v ′ to v v v the correctors and means ∑ V j [ σ 1 ⋯ σ i ] \sum\mathcal{V}_j[\sigma_1\cdots\sigma_i] ∑ V j [ σ 1 ⋯ σ i ] and ∑ M j , d [ σ 1 ⋯ σ i ] \sum\mathcal{M}_{j,d}[\sigma_1\cdots\sigma_i] ∑ M j , d [ σ 1 ⋯ σ i ] , summed over the paths σ 1 ⋯ σ i \sigma_1\cdots\sigma_i σ 1 ⋯ σ i from v ′ v' v ′ to v v v .
So, up to the cell tail (§ 5.8 ), the averaged equation of the finite lift is ∂ t Y v = ∑ ∇ ⋅ ( M j , d [ σ 1 ⋯ σ i ] ∇ d − 1 Y v ′ ) \partial_tY_v=\sum\nabla\cdot\left(\mathcal{M}_{j,d}[\sigma_1\cdots\sigma_i]\nabla^{d-1}Y_{v'}\right) ∂ t Y v = ∑ ∇ ⋅ ( M j , d [ σ 1 ⋯ σ i ] ∇ d − 1 Y v ′ ) , over j j j , d d d and the paths, with the sign of (34) .
Proof. Centring. The average in z z z of a derivative in z z z vanishes, and ( ∇ ⊥ ψ m ) ( t , ε m z ) (\nabla^\perp\psi_m)(t,\varepsilon_m z) ( ∇ ⊥ ψ m ) ( t , ε m z ) does not depend on y y y . So, for a transverse form A \mathcal{A} A of zero average in z z z , ( L 1 A ) T 2 = ∇ ⋅ ( ( ∇ ⊥ ψ m ) ( t , ε m z ) A ) T 2 (\mathcal{L}_1\mathcal{A})_{\mathbb{T}^2}=\nabla\cdot((\nabla^\perp\psi_m)(t,\varepsilon_m z)\mathcal{A})_{\mathbb{T}^2} ( L 1 A ) T 2 = ∇ ⋅ (( ∇ ⊥ ψ m ) ( t , ε m z ) A ) T 2 and ( L 2 A ) T 2 = 0 (\mathcal{L}_2\mathcal{A})_{\mathbb{T}^2}=0 ( L 2 A ) T 2 = 0 , while ( L 1 Y ) T 2 = 0 (\mathcal{L}_1Y)_{\mathbb{T}^2}=0 ( L 1 Y ) T 2 = 0 and ( L 2 Y ) T 2 = − κ m ∇ ⋅ ( K ∇ Y ) (\mathcal{L}_2Y)_{\mathbb{T}^2}=-\kappa_m\nabla\cdot(\mathbf{K}\nabla Y) ( L 2 Y ) T 2 = − κ m ∇ ⋅ ( K ∇ Y ) . A source part is ∇ y + ε m − 1 ∇ z \nabla_y+\varepsilon_m^{-1}\nabla_z ∇ y + ε m − 1 ∇ z applied to a flux, with average ∇ ⋅ \nabla\cdot ∇ ⋅ of the averaged flux.
( a ) (a) ( a ) , ( b ) (b) ( b ) . By induction on i i i and then on j j j . The centring and (106) give the average of the bracket of degree j j j . At j = 1 j=1 j = 1 every term vanishes: along ∅ \varnothing ∅ since the shear has mean zero; along a path of positive length since L 0 s r c \mathcal{L}^{\mathrm{src}}_0 L 0 src is a divergence in z z z ; and, on V 0 \mathcal{V}_0 V 0 , which does not depend on z z z , the part of a source flux with no derivative in y y y vanishes, as d ′ ≥ 2 d'\ge2 d ′ ≥ 2 . So in (107) the term j ′ ′ = 0 j''=0 j ′′ = 0 , to which only i ′ = i i'=i i ′ = i contributes, cancels the average of the bracket, and the terms j ′ ′ ≥ 1 j'' \ge1 j ′′ ≥ 1 have zero average: (107) has zero average in z z z . It is transverse, since K \mathbf{K} K does not depend on z z z , and Lemma 10.4 gives V j [ σ 1 ⋯ σ i ] \mathcal{V}_j[\sigma_1\cdots\sigma_i] V j [ σ 1 ⋯ σ i ] and (b). On constants, L 1 \mathcal{L}_1 L 1 , L 2 \mathcal{L}_2 L 2 and every L j ′ ′ s r c [ σ ] \mathcal{L}^{\mathrm{src}}_{j''}[\sigma] L j ′′ src [ σ ] vanish, hence so do the correctors of positive degree. The regularity of the means is that of Proposition 10.7 , which also gives the derivatives in y y y of the lower correctors and means used to form (107) . Existence and these bounds are proved together, by the induction of its proof.
(c). Along ∅ \varnothing ∅ , at j = 1 j=1 j = 1 the bracket is ( ∇ ⊥ ψ m ) ( t , ε m z ) ⋅ ∇ Y (\nabla^\perp\psi_m)(t,\varepsilon_m z)\cdot\nabla Y ( ∇ ⊥ ψ m ) ( t , ε m z ) ⋅ ∇ Y , so V 1 [ ∅ ] Y = χ K ⋅ ∇ Y \mathcal{V}_1[\varnothing]Y=\chi_{\mathbf{K}}\cdot\nabla Y V 1 [ ∅ ] Y = χ K ⋅ ∇ Y and, by (106) , M 2 , 2 [ ∅ ] = κ m K − ( ( ∇ ⊥ ψ m ) ( t , ε m z ) ⊗ χ K ) T 2 \mathcal{M}_{2,2}[\varnothing]=\kappa_m\mathbf{K}-((\nabla^\perp\psi_m)(t,\varepsilon_m z)\otimes\chi_{\mathbf{K}})^{\mathsf{T}2} M 2 , 2 [ ∅ ] = κ m K − (( ∇ ⊥ ψ m ) ( t , ε m z ) ⊗ χ K ) T 2 . For K = I \mathbf{K}=\mathbf{I} K = I , χ I = χ m κ m ( t , ε m z ) \chi_{\mathbf{I}}=\chi_m^{\kappa_m}(t,\varepsilon_m z) χ I = χ m κ m ( t , ε m z ) , and the mean of cos 2 \cos^2 cos 2 is 1 2 \frac{1}{2} 2 1 , so M 2 , 2 [ ∅ ] \mathcal{M}_{2,2}[\varnothing] M 2 , 2 [ ∅ ] is κ m I \kappa_m\mathbf{I} κ m I plus the term k k k of (18) , that is, J m κ m ( t ) J_m^{\kappa_m}(t) J m κ m ( t ) . For general K \mathbf{K} K , by (103) the modes ξ = ± 1 \xi=\pm1 ξ = ± 1 of χ K \chi_{\mathbf{K}} χ K and χ I \chi_{\mathbf{I}} χ I differ by the integral of ζ ~ m , k \widetilde{\zeta}_{m,k} ζ m , k against the difference of two exponentials, at most ( 4 π 2 κ m / ε m 2 ) ( t − t ′ ) sup ∣ K − I ∣ exp ( − 3 π 2 κ m ( t − t ′ ) / ε m 2 ) (4\pi^2\kappa_m/\varepsilon_m^2)(t-t')\sup|\mathbf{K}-\mathbf{I}|\exp(-3\pi^2\kappa_m(t-t')/\varepsilon_m^2) ( 4 π 2 κ m / ε m 2 ) ( t − t ′ ) sup ∣ K − I ∣ exp ( − 3 π 2 κ m ( t − t ′ ) / ε m 2 ) , of integral in t ′ t' t ′ at most ( 4 ε m 2 / 9 π 2 κ m ) sup ∣ K − I ∣ (4\varepsilon_m^2/9\pi^2\kappa_m)\sup|\mathbf{K}-\mathbf{I}| ( 4 ε m 2 /9 π 2 κ m ) sup ∣ K − I ∣ . With ∣ ∇ ⊥ ψ m , k ∣ ≤ 2 π ε m β − 1 |\nabla^\perp\psi_{m,k}|\le2\pi\varepsilon_m^{\beta-1} ∣ ∇ ⊥ ψ m , k ∣ ≤ 2 π ε m β − 1 and (96) , ∣ M 2 , 2 [ ∅ ] − J m κ m ( t ) ∣ ≤ C ( κ m + ε m 2 β / κ m ) sup ∣ K − I ∣ ≤ C κ m − 1 ε m 2 δ − 1 |\mathcal{M}_{2,2}[\varnothing]-J_m^{\kappa_m}(t)|\le C(\kappa_m+\varepsilon_m^{2\beta}/\kappa_m)\sup|\mathbf{K}-\mathbf{I}|\le C\kappa_{m-1}\varepsilon_m^{2\delta-1} ∣ M 2 , 2 [ ∅ ] − J m κ m ( t ) ∣ ≤ C ( κ m + ε m 2 β / κ m ) sup ∣ K − I ∣ ≤ C κ m − 1 ε m 2 δ − 1 ; derivatives in y y y fall on K \mathbf{K} K and cost C ε m − 1 − 1 C\varepsilon_{m-1}^{-1} C ε m − 1 − 1 each.
(d), (e). At a time as in (d), the shear term of (106) vanishes, and so does the average of the source flux of a corrector of positive degree; what remains is κ m K ∇ Y \kappa_m\mathbf{K}\nabla Y κ m K ∇ Y and the sources applied to V 0 [ ∅ ] Y = Y \mathcal{V}_0[\varnothing]Y=Y V 0 [ ∅ ] Y = Y , of degree d d d for the order d d d . For (e), by induction: if the lower correctors of positive degree vanish up to t t t , then, up to t t t , the bracket is L 2 Y \mathcal{L}_2Y L 2 Y (for i = 0 i=0 i = 0 and j = 2 j=2 j = 2 ), a source part applied to Y Y Y (for i = 1 i=1 i = 1 ), or zero, and it does not depend on z z z . The terms j ′ ′ ≥ 1 j''\ge1 j ′′ ≥ 1 of the last sum of (107) vanish, and its term j ′ ′ = 0 j''=0 j ′′ = 0 does not depend on z z z . So (107) , of zero average by (b), vanishes up to t t t , and (103) gives V j [ σ 1 ⋯ σ i ] = 0 \mathcal{V}_j[\sigma_1\cdots\sigma_i]=0 V j [ σ 1 ⋯ σ i ] = 0 at t t t .
(f). The definition is linear and the cell inverse unique. In the finite lift of the system, a term carrying i i i sources is a product of the forms along one path, in order; its source parts extend a shorter path by its last source, and its products split a path at one point. So the sums over paths satisfy the definition for the system. Its equations are triangular in the length of the longest path into v v v , which is at most N p a t h N_{\mathrm{path}} N path . At degree j j j , the source part L 0 s r c [ σ ] \mathcal{L}^{\mathrm{src}}_0[\sigma] L 0 src [ σ ] of a source σ : w → v \sigma:w\to v σ : w → v couples the corrector from v ′ v' v ′ to v v v to the corrector from v ′ v' v ′ to w w w at the same degree, and the longest path into w w w is shorter than that into v v v . So the definition for the system has exactly one solution, and uniqueness gives (f).
Paths and the threshold. The system along a path (its components, one source per step) satisfies the same hypotheses, and (a) along the path is (a) for it. The threshold Λ 0 \Lambda_0 Λ 0 is the larger of that of Lemma 10.1 (c) and the one beyond which ε ˉ m ≤ ρ ( m ) ≤ ε m − 1 \bar{\varepsilon}_m\le\rho(m)\le\varepsilon_{m-1} ε ˉ m ≤ ρ ( m ) ≤ ε m − 1 , by (96) and Proposition 2.1 (c). □ \square □
Bounds on the correctors and the means
For a transverse form A \mathcal{A} A and integers N , j 1 ≥ 0 N,j_1\ge0 N , j 1 ≥ 0 , the Wiener norm
∥ A ∥ N , j 1 : = ∑ ξ ∈ Z ( 1 + ∣ ξ ∣ ) N ∥ ∇ y j 1 A ( ⋅ , ⋅ ; ξ ) ∥ L ∞ ( [ ( k − 61 / 48 ) τ m , ( k + 61 / 48 ) τ m ] × R 2 ) (109) \|\mathcal{A}\|_{N,j_1}:=\sum_{\xi\in\mathbb{Z}}(1+|\xi|)^N\|\nabla_y^{j_1}\mathcal{A}(\cdot,\cdot;\xi)\|_{L^\infty([(k-61/48)\tau_m,(k+61/48)\tau_m]\times\mathbb{R}^2)} \tag*{(109)} ∥ A ∥ N , j 1 := ξ ∈ Z ∑ ( 1 + ∣ ξ ∣ ) N ∥ ∇ y j 1 A ( ⋅ , ⋅ ; ξ ) ∥ L ∞ ([( k − 61/48 ) τ m , ( k + 61/48 ) τ m ] × R 2 ) (109)
is summed over the entries for arrays. For every j 2 ≤ N j_2\le N j 2 ≤ N , it bounds ( 2 π ) − j 2 (2\pi)^{-j_2} ( 2 π ) − j 2 times the norm of ∇ y j 1 ∇ z j 2 A \nabla_y^{j_1}\nabla_z^{j_2}\mathcal{A} ∇ y j 1 ∇ z j 2 A in L ∞ ( [ ( k − 61 / 48 ) τ m , ( k + 61 / 48 ) τ m ] × R 2 × R 2 ) L^\infty([(k-61/48)\tau_m,(k+61/48)\tau_m]\times\mathbb{R}^2\times\mathbb{R}^2) L ∞ ([( k − 61/48 ) τ m , ( k + 61/48 ) τ m ] × R 2 × R 2 ) , and obeys the Leibniz rule, as 1 + ξ + ξ ′ ≤ ( 1 + ∣ ξ ∣ ) ( 1 + ∣ ξ ′ ∣ ) 1+\xi+\xi'\le(1+|\xi|)(1+|\xi'|) 1 + ξ + ξ ′ ≤ ( 1 + ∣ ξ ∣ ) ( 1 + ∣ ξ ′ ∣ ) . A matrix over the components is bounded in row and column sums if the sum of its entries over each row and over each column is.
Normalized variables. In the time κ m t / ε m 2 \kappa_m t/\varepsilon_m^2 κ m t / ε m 2 and the slow variable y / ρ ( m ) y/\rho(m) y / ρ ( m ) , ( ε m 2 / κ m ) L 0 (\varepsilon_m^2/\kappa_m)\mathcal{L}_0 ( ε m 2 / κ m ) L 0 has the diffusivity e ( k ) ⊥ ⋅ K e ( k ) ⊥ ∈ [ 3 / 4 , 5 / 4 ] e_{(k)}^{\perp}\cdot\mathbf{K}e_{(k)}^{\perp}\in[3/4,5/4] e ( k ) ⊥ ⋅ K e ( k ) ⊥ ∈ [ 3/4 , 5/4 ] in the fast variable. Relative to ( ε m 2 / κ m ) L 0 (\varepsilon_m^2/\kappa_m)\mathcal{L}_0 ( ε m 2 / κ m ) L 0 and to the derivatives in the slow variable, ( ε m 2 / κ m ) L 1 (\varepsilon_m^2/\kappa_m)\mathcal{L}_1 ( ε m 2 / κ m ) L 1 , ( ε m 2 / κ m ) L 2 (\varepsilon_m^2/\kappa_m)\mathcal{L}_2 ( ε m 2 / κ m ) L 2 , ( ε m 2 / κ m ) L j ′ s r c [ σ ] (\varepsilon_m^2/\kappa_m)\mathcal{L}^{\mathrm{src}}_{j'}[\sigma] ( ε m 2 / κ m ) L j ′ src [ σ ] have the sizes of the powers 1 , 2 , j ′ 1,2,j' 1 , 2 , j ′ of ε ˉ m / ρ ( m ) ≤ C ε m − 1 3 δ \bar{\varepsilon}_m/\rho(m)\le C\varepsilon_{m-1}^{3\delta} ε ˉ m / ρ ( m ) ≤ C ε m − 1 3 δ : the drift ε m ( κ m κ m − 1 ) − 1 / 2 ∇ ⊥ ψ m \varepsilon_m(\kappa_m\kappa_{m-1})^{-1/2}\nabla^\perp\psi_m ε m ( κ m κ m − 1 ) − 1/2 ∇ ⊥ ψ m of L 1 \mathcal{L}_1 L 1 is at most C C C by (96) , and the sources carry ( ε m / ρ ( m ) ) j ′ (\varepsilon_m/\rho(m))^{j'} ( ε m / ρ ( m ) ) j ′ , smaller as κ m ≤ κ m − 1 \kappa_m\le\kappa_{m-1} κ m ≤ κ m − 1 .
Proposition 10.7 (Sizes). In the setting of Proposition 10.6 , there exists C C C , depending only on α \alpha α , s s s , q q q , ζ \zeta ζ , N c o r N_{\mathrm{cor}} N cor , N s r c N_{\mathrm{src}} N src , N p a t h N_{\mathrm{path}} N path , N N N and the constants of (108) , such that, with sums over the paths σ 1 ⋯ σ i \sigma_1\cdots\sigma_i σ 1 ⋯ σ i from v ′ v' v ′ to v v v :
(a) for every 1 ≤ d ≤ j ≤ N c o r 1\le d\le j\le N_{\mathrm{cor}} 1 ≤ d ≤ j ≤ N cor and j 1 ≤ N j_1\le N j 1 ≤ N , in row and column sums over the components v v v , v ′ v' v ′ ,
( ω v ′ ω v ) 1 / 2 ε m − 1 j 1 ∥ ∑ V j , d [ σ 1 ⋯ σ i ] ∥ N , j 1 ≤ C ε m j ε m − 1 d − j ; (110) \left(\frac{\omega_{v'}}{\omega_v}\right)^{1/2}\varepsilon_{m-1}^{j_1}\left\|\sum\mathcal{V}_{j,d}[\sigma_1\cdots\sigma_i]\right\|_{N,j_1}\le C\varepsilon_m^j\varepsilon_{m-1}^{d-j}; \tag*{(110)} ( ω v ω v ′ ) 1/2 ε m − 1 j 1 ∑ V j , d [ σ 1 ⋯ σ i ] N , j 1 ≤ C ε m j ε m − 1 d − j ; (110) (b) for every 2 ≤ d ≤ j ≤ N c o r + N s r c 2\le d\le j\le N_{\mathrm{cor}}+N_{\mathrm{src}} 2 ≤ d ≤ j ≤ N cor + N src and j 1 ≤ N j_1\le N j 1 ≤ N , in row and column sums as in (a),
( ω v ′ ω v ) 1 / 2 ε m − 1 j 1 ∥ ∇ y j 1 ∑ M j , d [ σ 1 ⋯ σ i ] ∥ L ∞ ( [ ( k − 61 / 48 ) τ m , ( k + 61 / 48 ) τ m ] × R 2 ) ≤ C κ m − 1 ε m j − 2 ε m − 1 d − j ; (111) \left(\frac{\omega_{v'}}{\omega_v}\right)^{1/2}\varepsilon_{m-1}^{j_1}\left\|\nabla_y^{j_1}\sum\mathcal{M}_{j,d}[\sigma_1\cdots\sigma_i]\right\|_{L^\infty([(k-61/48)\tau_m,(k+61/48)\tau_m]\times\mathbb{R}^2)}
\le C\kappa_{m-1}\varepsilon_m^{j-2}\varepsilon_{m-1}^{d-j}; \tag*{(111)} ( ω v ω v ′ ) 1/2 ε m − 1 j 1 ∇ y j 1 ∑ M j , d [ σ 1 ⋯ σ i ] L ∞ ([( k − 61/48 ) τ m , ( k + 61/48 ) τ m ] × R 2 ) ≤ C κ m − 1 ε m j − 2 ε m − 1 d − j ; (111) (c) for every path σ 1 ⋯ σ i \sigma_1\cdots\sigma_i σ 1 ⋯ σ i , 2 ≤ d ≤ j ≤ N c o r + N s r c 2\le d\le j\le N_{\mathrm{cor}}+N_{\mathrm{src}} 2 ≤ d ≤ j ≤ N cor + N src and j 1 ≤ N + 1 j_1\le N+1 j 1 ≤ N + 1 ,
ε m − 1 j 1 ∥ ∇ y j 1 M j , d [ σ 1 ⋯ σ i ] ∥ L ∞ ( [ ( k − 61 / 48 ) τ m , ( k + 61 / 48 ) τ m ] × R 2 ) ≤ C fac ( σ 1 ) maj ( σ 1 ) ⋯ fac ( σ i ) maj ( σ i ) κ m − 1 ε m j − 2 ε m − 1 d − j . (112) \varepsilon_{m-1}^{j_1}\left\|\nabla_y^{j_1}\mathcal{M}_{j,d}[\sigma_1\cdots\sigma_i]\right\|_{L^\infty([(k-61/48)\tau_m,(k+61/48)\tau_m]\times\mathbb{R}^2)}
\le C\operatorname{fac}(\sigma_1)\operatorname{maj}(\sigma_1)\cdots\operatorname{fac}(\sigma_i)\operatorname{maj}(\sigma_i)\kappa_{m-1}\varepsilon_m^{j-2}\varepsilon_{m-1}^{d-j}. \tag*{(112)} ε m − 1 j 1 ∇ y j 1 M j , d [ σ 1 ⋯ σ i ] L ∞ ([( k − 61/48 ) τ m , ( k + 61/48 ) τ m ] × R 2 ) ≤ C fac ( σ 1 ) maj ( σ 1 ) ⋯ fac ( σ i ) maj ( σ i ) κ m − 1 ε m j − 2 ε m − 1 d − j . (112)
Proof. By induction on the length of the paths and then on the degree; for (c), the majorant matrix is replaced by the entries fac ( σ ) maj ( σ ) \operatorname{fac}(\sigma)\operatorname{maj}(\sigma) fac ( σ ) maj ( σ ) of the sources of the path.
The inverse and the four products. By Lemma 10.4, in the norms (109) with the weights ε m − 1 j 1 \varepsilon_{m-1}^{j_1} ε m − 1 j 1 , the cell inverse costs C ε m 2 / κ m C\varepsilon_m^2/\kappa_m C ε m 2 / κ m and returns two derivatives in z z z on F 0 F^0 F 0 and one on F F F of (102) . The cross terms of L 1 \mathcal{L}_1 L 1 on a transverse form B B B have the form e ( k ) ⊥ ⋅ ∇ z F e_{(k)}^\perp\cdot\nabla_zF e ( k ) ⊥ ⋅ ∇ z F , as ∇ z ⋅ ( K ∇ y B ) = e ( k ) ⊥ ⋅ ∇ z ( e ( k ) ⊥ ⋅ K ∇ y B ) \nabla_z\cdot(K\nabla_yB)=e_{(k)}^\perp\cdot\nabla_z(e_{(k)}^\perp\cdot K\nabla_yB) ∇ z ⋅ ( K ∇ y B ) = e ( k ) ⊥ ⋅ ∇ z ( e ( k ) ⊥ ⋅ K ∇ y B ) and ∇ y ⋅ ( K ∇ z B ) = e ( k ) ⊥ ⋅ ∇ z ∇ y ⋅ ( K e ( k ) ⊥ B ) \nabla_y\cdot(K\nabla_zB)=e_{(k)}^\perp\cdot\nabla_z\nabla_y\cdot(Ke_{(k)}^\perp B) ∇ y ⋅ ( K ∇ z B ) = e ( k ) ⊥ ⋅ ∇ z ∇ y ⋅ ( K e ( k ) ⊥ B ) . A derivative in y y y of the exponential of (103) brings down − ( 4 π 2 ξ 2 κ m / ε m 2 ) ∫ t ′ t e ( k ) ⊥ ⋅ ∇ y K ( t ′ ′ , y ) e ( k ) ⊥ d t ′ ′ -(4\pi^2\xi^2\kappa_m/\varepsilon_m^2)\int_{t'}^t e_{(k)}^\perp\cdot\nabla_yK(t'',y)e_{(k)}^\perp\,\mathrm{d}t'' − ( 4 π 2 ξ 2 κ m / ε m 2 ) ∫ t ′ t e ( k ) ⊥ ⋅ ∇ y K ( t ′′ , y ) e ( k ) ⊥ d t ′′ , of size at most ( 4 π 2 ξ 2 κ m / ε m 2 ) ( t − t ′ ) sup ∣ ∇ y K ∣ (4\pi^2\xi^2\kappa_m/\varepsilon_m^2)(t-t')\sup|\nabla_yK| ( 4 π 2 ξ 2 κ m / ε m 2 ) ( t − t ′ ) sup ∣ ∇ y K ∣ . So j 1 j_1 j 1 derivatives bring down products of at most j 1 j_1 j 1 such terms, with j 1 j_1 j 1 derivatives on K K K in all. After the integration in t ′ t' t ′ each derivative costs C ε m − 1 − 1 C\varepsilon_{m-1}^{-1} C ε m − 1 − 1 , as ∫ 0 ∞ ( a τ ) l e − 3 a τ / 4 d τ = l ! ( 4 / 3 ) l + 1 / a \int_0^\infty(a\tau)^l e^{-3a\tau/4}\,\mathrm{d}\tau=l!(4/3)^{l+1}/a ∫ 0 ∞ ( a τ ) l e − 3 a τ /4 d τ = l ! ( 4/3 ) l + 1 / a for a > 0 a>0 a > 0 . Each term of (107) is a lower corrector or mean times one of four coefficients, which the inverse turns into powers of the corrector amplitude; by (96) , the definition of ε ˉ m \bar\varepsilon_m ε ˉ m and ε m ≤ ε ˉ m \varepsilon_m\le\bar\varepsilon_m ε m ≤ ε ˉ m (Proposition 2.1(c)), the four products satisfy
ε m 2 κ m ε m β − 1 ≤ C ε ˉ m , ε m 2 κ m κ m ε m = ε m , ε m 2 κ m κ m ε m j ′ − 2 = ε m j ′ , ε m 2 κ m κ m − 1 = ε m 2 κ m − 1 κ m = ε m 2 . (113) \frac{\varepsilon_m^2}{\kappa_m}\varepsilon_m^{\beta-1}\le C\bar\varepsilon_m,\qquad
\frac{\varepsilon_m^2}{\kappa_m}\frac{\kappa_m}{\varepsilon_m}=\varepsilon_m,\qquad
\frac{\varepsilon_m^2}{\kappa_m}\kappa_m\varepsilon_m^{j'-2}=\varepsilon_m^{j'},\qquad
\frac{\varepsilon_m^2}{\kappa_m}\kappa_{m-1}=\varepsilon_m^2\frac{\kappa_{m-1}}{\kappa_m}=\varepsilon_m^2. \tag*{(113)} κ m ε m 2 ε m β − 1 ≤ C ε ˉ m , κ m ε m 2 ε m κ m = ε m , κ m ε m 2 κ m ε m j ′ − 2 = ε m j ′ , κ m ε m 2 κ m − 1 = ε m 2 κ m κ m − 1 = ε m 2 . (113)
The step. Assume the bounds along shorter paths, and along the same path at lower degrees. In each term, a derivative in y y y raises the order of the derivatives of Y Y Y or falls on a coefficient, at the cost ε m − 1 − 1 \varepsilon_{m-1}^{-1} ε m − 1 − 1 , which keeps the power ε m − 1 d − j \varepsilon_{m-1}^{d-j} ε m − 1 d − j . The shear term of L 1 V j − 1 \mathcal{L}_1\mathcal{V}_{j-1} L 1 V j − 1 , of size 2 π ε m β − 1 2\pi\varepsilon_m^{\beta-1} 2 π ε m β − 1 with two modes, gets the first product; the cross terms of L 1 V j − 1 \mathcal{L}_1\mathcal{V}_{j-1} L 1 V j − 1 , with one derivative in z z z that the inverse returns, the second; L 2 V j − 2 \mathcal{L}_2\mathcal{V}_{j-2} L 2 V j − 2 the third, with j ′ = 2 j'=2 j ′ = 2 . A term L j ′ s r c [ σ i ] V j − j ′ [ σ 1 ⋯ σ i − 1 ] \mathcal{L}_{j'}^{\mathrm{src}}[\sigma_i]\mathcal{V}_{j-j'}[\sigma_1\cdots\sigma_{i-1}] L j ′ src [ σ i ] V j − j ′ [ σ 1 ⋯ σ i − 1 ] has the coefficient κ m ε m d ′ − 2 fac ( σ i ) A σ i , d ′ \kappa_m\varepsilon_m^{d'-2}\operatorname{fac}(\sigma_i)A_{\sigma_i,d'} κ m ε m d ′ − 2 fac ( σ i ) A σ i , d ′ and d ′ − j ′ d'-j' d ′ − j ′ derivatives in z z z , each costing ε m − 1 − 1 \varepsilon_{m-1}^{-1} ε m − 1 − 1 , so it gets the third product times C fac ( σ i ) maj ( σ i ) C\operatorname{fac}(\sigma_i)\operatorname{maj}(\sigma_i) C fac ( σ i ) maj ( σ i ) . A product term of (107) with j ′ ′ ≥ 1 j''\ge1 j ′′ ≥ 1 carries κ m − 1 \kappa_{m-1} κ m − 1 and j − 2 j-2 j − 2 powers of the corrector amplitude, and the fourth product turns κ m − 1 \kappa_{m-1} κ m − 1 into two more. This gives (a) at j j j .
The means. By (106) , the means are κ m K \kappa_m\mathbf{K} κ m K , at most 5 4 κ m − 1 \frac54\kappa_{m-1} 4 5 κ m − 1 at the degree 2, plus two averages. For the shear term, ε m β − 1 ε ˉ m = ( κ m − 1 ε m 2 / κ m ) 1 / 2 ≤ C κ m − 1 \varepsilon_m^{\beta-1}\bar\varepsilon_m=(\kappa_{m-1}\varepsilon_m^2/\kappa_m)^{1/2}\le C\kappa_{m-1} ε m β − 1 ε ˉ m = ( κ m − 1 ε m 2 / κ m ) 1/2 ≤ C κ m − 1 by (96) , so the bounds of the induction on V j − 1 \mathcal{V}_{j-1} V j − 1 , those of (a) with the N + 1 N+1 N + 1 derivatives in y y y of its reserve and, for (c), along the path, give (b) and (c). For the source term, the part with j ′ j' j ′ derivatives in y y y of a source applied to V j ′ ′ V_{j''} V j ′′ , j ′ ′ + j ′ = j − 1 j'' + j' = j - 1 j ′′ + j ′ = j − 1 , is κ m ε m j ′ − 1 fac ( σ i ) maj ( σ i ) \kappa_m \varepsilon_m^{j'-1}\operatorname{fac}(\sigma_i)\operatorname{maj}(\sigma_i) κ m ε m j ′ − 1 fac ( σ i ) maj ( σ i ) times the bound of the induction for V j ′ ′ V_{j''} V j ′′ , and this is at most the bound (b): by ε m ≤ ε ˉ m \varepsilon_m \le\bar{\varepsilon}_m ε m ≤ ε ˉ m if j ′ ≥ 1 j' \ge1 j ′ ≥ 1 , and by ( κ m / ε m ) ε ˉ m = ( κ m κ m − 1 ) 1 / 2 ≤ κ m − 1 (\kappa_m/\varepsilon_m)\bar{\varepsilon}_m = (\kappa_m\kappa_{m-1})^{1/2} \le\kappa_{m-1} ( κ m / ε m ) ε ˉ m = ( κ m κ m − 1 ) 1/2 ≤ κ m − 1 if j ′ = 0 j'=0 j ′ = 0 .
Components and orders. Over the components, each term is a product of the majorant matrix and of lower correctors and means; row and column sums are submultiplicative, and at most 1 for the majorant matrix. The derivatives are counted by reserves: the induction at path length i i i and degree j j j carries bounds on the correctors with
N + 1 + ( N c o r + N s r c − j ) derivatives in y and N + 2 + ( N p a t h − i ) N s r c + 2 ( N c o r + N s r c − j ) in z , N + 1 + (N_{\mathrm{cor}} + N_{\mathrm{src}} - j)\text{ derivatives in }y \quad\text{and}\quad N + 2 + (N_{\mathrm{path}} - i)N_{\mathrm{src}} + 2(N_{\mathrm{cor}} + N_{\mathrm{src}} - j)\text{ in }z, N + 1 + ( N cor + N src − j ) derivatives in y and N + 2 + ( N path − i ) N src + 2 ( N cor + N src − j ) in z ,
its reserves at ( i , j ) (i,j) ( i , j ) , and on the means, which do not depend on z z z , with N + 2 + ( N c o r + N s r c − j ) N + 2 + (N_{\mathrm{cor}} + N_{\mathrm{src}} - j) N + 2 + ( N cor + N src − j ) derivatives in y y y , one more than a corrector of the same degree. A reserve r r r in z z z is carried in the Wiener norm (109) of order r − 2 r-2 r − 2 , as the cell inverse acts mode by mode. A form with r r r bounded derivatives in z z z has this norm, since ( 1 + ∣ ξ ∣ ) r ∣ A ^ ( ξ ) ∣ ≤ 2 r ( ∥ A ∥ L ∞ + ( 2 π ) − r ∥ ( e k ⊥ ⋅ ∇ z ) r A ∥ L ∞ ) (1+|\xi|)^r|\widehat{A}(\xi)| \le2^r(\|A\|_{L^\infty}+(2\pi)^{-r}\|(e_k^\perp\cdot\nabla_z)^rA\|_{L^\infty}) ( 1 + ∣ ξ ∣ ) r ∣ A ( ξ ) ∣ ≤ 2 r ( ∥ A ∥ L ∞ + ( 2 π ) − r ∥ ( e k ⊥ ⋅ ∇ z ) r A ∥ L ∞ ) and ∑ ξ ∈ Z 2 ( 1 + ∣ ξ ∣ ) − 2 < ∞ \sum_{\xi\in\mathbb{Z}^2}(1+|\xi|)^{-2}<\infty ∑ ξ ∈ Z 2 ( 1 + ∣ ξ ∣ ) − 2 < ∞ . So the bases N + 1 N+1 N + 1 and N + 2 N+2 N + 2 give the N + 1 N+1 N + 1 derivatives in y y y of (c) and the Wiener norms (109) of order N N N . Lowering the degree by j ′ j' j ′ raises the reserve in y y y by j ′ j' j ′ and that in z z z by $2j'$, and shortening the path by one raises the reserve in z z z by N s r c N_{\mathrm{src}} N src . A source part L j ′ s r c [ σ i ] \mathcal{L}_{j'}^{\mathrm{src}}[\sigma_i] L j ′ src [ σ i ] , i ≥ 1 i\ge1 i ≥ 1 , 0 ≤ j ′ ≤ min { j , N s r c } 0\le j'\le\min\{j,N_{\mathrm{src}}\} 0 ≤ j ′ ≤ min { j , N src } , puts at most j ′ j' j ′ derivatives in y y y and N s r c − j ′ N_{\mathrm{src}}-j' N src − j ′ in z z z on its predecessor, of path length i − 1 i-1 i − 1 and degree j − j ′ j-j' j − j ′ , whose reserves exceed those at ( i , j ) (i,j) ( i , j ) by j ′ j' j ′ in y y y and by N s r c + 2 j ′ ≥ N s r c − j ′ N_{\mathrm{src}}+2j'\ge N_{\mathrm{src}}-j' N src + 2 j ′ ≥ N src − j ′ in z z z . The terms with L 1 \mathcal{L}_1 L 1 and L 2 \mathcal{L}_2 L 2 put at most j ′ j' j ′ derivatives in y y y and 2 j ′ 2j' 2 j ′ in z z z on V j − j ′ V_{j-j'} V j − j ′ , j ′ = 1 , 2 j'=1,2 j ′ = 1 , 2 , whose reserves exceed those at ( i , j ) (i,j) ( i , j ) by exactly these numbers. By (106) , a mean of degree j j j asks of V j − 1 V_{j-1} V j − 1 the reserve in y y y of the means at j j j , which is that of V j − 1 V_{j-1} V j − 1 , and of a source part with j ′ j' j ′ derivatives in y y y on V j ′ ′ V_{j''} V j ′′ , j ′ ′ + j ′ = j − 1 j''+j'=j-1 j ′′ + j ′ = j − 1 , that reserve plus j ′ j' j ′ , which is that of V j ′ ′ V_{j''} V j ′′ . The term κ m K \kappa_m\mathbf{K} κ m K of (106) asks at most N + N c o r + N s r c N+N_{\mathrm{cor}}+N_{\mathrm{src}} N + N cor + N src derivatives of K \mathbf{K} K . In a product term with j ′ ′ ≥ 1 j''\ge1 j ′′ ≥ 1 (the term j ′ ′ = 0 j''=0 j ′′ = 0 does not depend on z z z , and the part of zero average removes it), the path splits into two paths, of total length i i i , with the degrees j ′ ′ j'' j ′′ and j − j ′ ′ j-j'' j − j ′′ , both below j j j . By Leibniz’s rule, the mean, of degree j − j ′ ′ j-j'' j − j ′′ , receives at most j ′ ′ + 1 j''+1 j ′′ + 1 derivatives in y y y beyond those of the target, and its reserve in y y y exceeds that of the correctors at ( i , j ) (i,j) ( i , j ) by exactly $j''+1. The corrector, of degree j ′ ′ j'' j ′′ , receives at most those of the target, within its reserve, and both have reserves in z z z at least those at ( i , j ) (i,j) ( i , j ) . So the induction closes. This induction runs together with that of Proposition 10.6(a): at ( i , j ) (i,j) ( i , j ) , the bounds at the lower stages bound (107) in the Wiener norm of order r − 4 r-4 r − 4 , with r ≥ N + 2 + 2 N s r c ≥ 6 r\ge N+2+2N_{\mathrm{src}}\ge6 r ≥ N + 2 + 2 N src ≥ 6 the reserve in z z z at ( i , j ) (i,j) ( i , j ) , so that (107) , as a scalar source, has the two bounded derivatives in z z z that Lemma 10.4 assumes. The largest reserves, at ( 0 , 0 ) (0,0) ( 0 , 0 ) , are N + 2 + N p a t h N s r c + 2 N c o r + 2 N s r c N+2+N_{\mathrm{path}}N_{\mathrm{src}}+2N_{\mathrm{cor}}+2N_{\mathrm{src}} N + 2 + N path N src + 2 N cor + 2 N src in z z z , which is the order in z z z of (108) , and N + 1 + N c o r + N s r c ≤ N + 4 ( N p a t h + 1 ) ( N c o r + N s r c ) 2 N+1+N_{\mathrm{cor}}+N_{\mathrm{src}}\le N+4(N_{\mathrm{path}}+1)(N_{\mathrm{cor}}+N_{\mathrm{src}})^2 N + 1 + N cor + N src ≤ N + 4 ( N path + 1 ) ( N cor + N src ) 2 in y y y , since N s r c ≥ 2 N_{\mathrm{src}}\ge2 N src ≥ 2 , which is within its order in y y y . □
The corrector operator and the mean flux
Let G G G be a family of level m m m , 2 ≤ m ≤ M 2\le m\le M 2 ≤ m ≤ M (Subsection 5.4), whose sources have coefficients (116) satisfying the hypotheses of Proposition 10.6 on the window of every pulse k k k , in its chart, with K = K m , l m , k \mathbf{K}=\mathbf{K}_{m,l_m,k} K = K m , l m , k . Let V j [ σ 1 ⋯ σ i ] V_j[\sigma_1\cdots\sigma_i] V j [ σ 1 ⋯ σ i ] and M j , d [ σ 1 ⋯ σ i ] \mathcal{M}_{j,d}[\sigma_1\cdots\sigma_i] M j , d [ σ 1 ⋯ σ i ] be the correctors and means of its finite lift on the window of pulse k k k , and, for a field Y Y Y on the components of G G G , let Y ~ k ( t , y ) : = Y ( t , X m − 1 ( t , y , l m , k h m ) ) \widetilde{Y}_k(t,y):=Y(t,X_{m-1}(t,y,l_m,k h_m)) Y k ( t , y ) := Y ( t , X m − 1 ( t , y , l m , k h m )) .
Definition 10.8 (The corrector operator and the mean flux). For a field Y Y Y as above, a component v v v , t ∈ R t\in\mathbb{R} t ∈ R and x ∈ R 2 x\in\mathbb{R}^2 x ∈ R 2 , with y : = X m − 1 ( l m , k h m , x , t ) y:=X_{m-1}(l_m,k h_m,x,t) y := X m − 1 ( l m , k h m , x , t ) , ∇ y X m − 1 \nabla_yX_{m-1} ∇ y X m − 1 taken at ( t , y , l m , k h m ) (t,y,l_m,k h_m) ( t , y , l m , k h m ) , and V j [ σ 1 ⋯ σ i ] V_j[\sigma_1\cdots\sigma_i] V j [ σ 1 ⋯ σ i ] , M j , d [ σ 1 ⋯ σ i ] \mathcal{M}_{j,d}[\sigma_1\cdots\sigma_i] M j , d [ σ 1 ⋯ σ i ] those of the window of pulse k k k , in the term of k k k , with the sums over the paths $\sigma_1\cdots\sigma_i
ending at v v v , from the component v ′ v' v ′ ,
( Corr m , k Y ) v ( t , x ) : = ∑ j = 0 N c o r V j [ σ 1 ⋯ σ i ] Y ~ k , v ′ ( t , y , y / ε m ) , Corr m Y : = ∑ k ∈ 2 Z + 1 H m , k Corr m , k Y , (\operatorname{Corr}_{m,k}Y)_v(t,x):=\sum_{j=0}^{N_{\mathrm{cor}}}\mathcal{V}_j[\sigma_1\cdots\sigma_i]\widetilde{Y}_{k,v'}(t,y,y/\varepsilon_m),\qquad\operatorname{Corr}_mY:=\sum_{k\in2\mathbb{Z}+1}H_{m,k}\operatorname{Corr}_{m,k}Y, ( Corr m , k Y ) v ( t , x ) := j = 0 ∑ N cor V j [ σ 1 ⋯ σ i ] Y k , v ′ ( t , y , y / ε m ) , Corr m Y := k ∈ 2 Z + 1 ∑ H m , k Corr m , k Y ,
( M m Y ) v ( t , x ) : = ∑ k ∈ 2 Z + 1 H m , k ( t ) ∇ y X m − 1 ∑ j = 2 N c o r + N s r c ∑ d = 2 j ∑ ( M j , d [ σ 1 ⋯ σ i ] ∇ d − 1 Y ~ k , v ′ ) ( t , y ) . (114) (\mathcal{M}_mY)_v(t,x):=\sum_{k\in2\mathbb{Z}+1}H_{m,k}(t)\nabla_yX_{m-1}\sum_{j=2}^{N_{\mathrm{cor}}+N_{\mathrm{src}}}\sum_{d=2}^{j}\sum(\mathcal{M}_{j,d}[\sigma_1\cdots\sigma_i]\nabla^{d-1}\widetilde{Y}_{k,v'})(t,y). \tag*{(114)} ( M m Y ) v ( t , x ) := k ∈ 2 Z + 1 ∑ H m , k ( t ) ∇ y X m − 1 j = 2 ∑ N cor + N src d = 2 ∑ j ∑ ( M j , d [ σ 1 ⋯ σ i ] ∇ d − 1 Y k , v ′ ) ( t , y ) . (114)
The operator Corr m \operatorname{Corr}_m Corr m is that of (38) . In (114) , the term of k k k vanishes outside the window of k k k , where H m , k = 0 H_{m,k}=0 H m , k = 0 , as 5 4 < 61 48 \frac{5}{4}<\frac{61}{48} 4 5 < 48 61 , and ∇ d − 1 Y ~ k , v ′ \nabla^{d-1}\widetilde{Y}_{k,v'} ∇ d − 1 Y k , v ′ is taken in the label y y y . The mean flux M m \mathcal{M}_m M m glues the means, carried to Eulerian coordinates by the Piola identity of Lemma 7.9 , under which a flux in the chart is multiplied by ∇ y X m − 1 \nabla_yX_{m-1} ∇ y X m − 1 . By Propositions 10.6 (c) and 10.7 (b) and Lemma 7.7 (a), the part along the empty paths is close to J m n m ( t ) ∇ J_m^{\mathrm{nm}}(t)\nabla J m nm ( t ) ∇ ; the difference, of all orders, is the defect of the averaged flux, a source of the successor (§ 5.8 ).
Admissible families
In this section we define the admissible families, those that descend from an admissible seed by cycles (Subsection 5.4), and prove by induction along the cycles that they satisfy the invariant of Definition 11.1 (Proposition 11.5 ), whose clauses (a2)–(a4), (a6), (b)–(d) are used in § 5.6 –§ 5.11 .
Parameters and sources
Parameters. The following parameters depend only on α \alpha α , s s s , q q q , ζ \zeta ζ , Λ \Lambda Λ , n n n , t 0 t_0 t 0 , T T T . With those of Subsection 5.3 and § 5.11 , they are fixed in the order of Table 11.1 , which omits only the absolute constant c a n c_{\mathrm{an}} c an of Subsection 11.3 and computed orders such as N r e c N_{\mathrm{rec}} N rec and N w o r d N_{\mathrm{word}} N word . The orders of the entry, among them N s r c , e N_{\mathrm{src,e}} N src , e and N o u t , e N_{\mathrm{out,e}} N out , e , and the bound B B B on the paths of the seed are defined in Subsection 15.1, in (203) , item (3) there and (207) .
Table 11.1. The order in which the parameters of Parts I and II are chosen. Each row may use the rows above it; the middle column names the earlier choices it uses (in the first three rows, the parameters it uses). The local paragraphs (Subsection 5.3, the items above, Sections 12–15) say why each condition is needed. Everything above the double line depends only on α , s , q , ζ , Λ , n , t 0 , T \alpha,s,q,\zeta,\Lambda,n,t_0,T α , s , q , ζ , Λ , n , t 0 , T and is fixed before the levels M M M , m m m and the diffusivity κ \kappa κ ; below it are thresholds on the levels, chosen after everything above and changing none of it, each so large that the inequalities of the proofs hold beyond it. The constants c b r , c l o , c d i s , c h i c_{\mathrm{br}},c_{\mathrm{lo}},c_{\mathrm{dis}},c_{\mathrm{hi}} c br , c lo , c dis , c hi of Section 12 are absolute.
The orders of the entry, which fix Part I’s N r e p N_{\mathrm{rep}} N rep , and the seed’s bounds (§ 6 ): orders at most N s r c , e N_{\mathrm{src,e}} N src , e , and at most B B B sources on a path. Then, with the other parameters of Subsection 5.3 and the orders of § 5.11 : N g e n N_{\mathrm{gen}} N gen , the number of generations kept; N s r c N_{\mathrm{src}} N src ; the width exponent Θ \Theta Θ of (178) ; N l e n N_{\mathrm{len}} N len , the length of the longest word kept, and N p a t h N_{\mathrm{path}} N path ; N m o m N_{\mathrm{mom}} N mom , the moment order of the smoothings, with 2 ℓ − 1 ≥ N m o m 2\ell-1\ge N_{\mathrm{mom}} 2 ℓ − 1 ≥ N mom and the conditions of (157) ; N c o e f N_{\mathrm{coef}} N coef , the order of the derivative bounds, at least N r e t N_{\mathrm{ret}} N ret and the orders at which the preparation uses these bounds (§ 5.6 ). A generation adds at most 2 N c o r + 4 p + 4 2N_{\mathrm{cor}}+4p+4 2 N cor + 4 p + 4 to the orders and multiplies the number of components on a path by at most N l e n + 2 N_{\mathrm{len}}+2 N len + 2 ; this gives the orders N s r c N_{\mathrm{src}} N src at generation N g e n + 1 N_{\mathrm{gen}}+1 N gen + 1 and the path bound N p a t h N_{\mathrm{path}} N path at generation N g e n N_{\mathrm{gen}} N gen (the paragraph after Definition 11.4 ), and Θ \Theta Θ is small:
N s r c : = N s r c , e + ( N g e n + 1 ) ( 2 N c o r + 4 p + 4 ) , N p a t h : = ( B + 1 ) ( N l e n + 2 ) N g e n − 1 , ( N s r c − 2 ) Θ ≤ δ 8 . N_{\mathrm{src}}:=N_{\mathrm{src,e}}+(N_{\mathrm{gen}}+1)(2N_{\mathrm{cor}}+4p+4),\qquad N_{\mathrm{path}}:=(B+1)(N_{\mathrm{len}}+2)^{N_{\mathrm{gen}}-1},\qquad(N_{\mathrm{src}}-2)\Theta\le\frac{\delta}{8}. N src := N src , e + ( N gen + 1 ) ( 2 N cor + 4 p + 4 ) , N path := ( B + 1 ) ( N len + 2 ) N gen − 1 , ( N src − 2 ) Θ ≤ 8 δ .
The constant c m o l ∈ ( 0 , 1 ] c_{\mathrm{mol}}\in(0,1] c mol ∈ ( 0 , 1 ] of the molecular radius
ρ m o l ( m ) : = c m o l min { ε m , ( κ m ε m 2 − β ) 1 / 2 } , (115) \rho_{\mathrm{mol}}(m):=c_{\mathrm{mol}}\min\{\varepsilon_m,(\kappa_m\varepsilon_m^{2-\beta})^{1/2}\}, \tag*{(115)} ρ mol ( m ) := c mol min { ε m , ( κ m ε m 2 − β ) 1/2 } , (115)
the scale at which a solution of a family of level m m m has derivatives after the time ρ m o l ( m ) 2 / κ m \rho_{\mathrm{mol}}(m)^2/\kappa_m ρ mol ( m ) 2 / κ m ; c m o l c_{\mathrm{mol}} c mol is fixed in § 5.6 , before the orders of item (1). Then the seed’s orders and constants (§ 6 ): its derivative order, at least max { N c o e f , N o u t , e } \max\{N_{\mathrm{coef}},N_{\mathrm{out,e}}\} max { N coef , N out , e } and large enough in terms of the orders of item (1); its weight order (the exponent of 1 + ∣ ξ ∣ 1 + |\xi| 1 + ∣ ξ ∣ in (11.9)), at least its derivative order; the offset N o f f ≥ n N_{\mathrm{off}} \ge n N off ≥ n of the analytic norm (Subsection 11.3; used in Definition 15.3 ); C s e e d ≥ 1 C_{\mathrm{seed}} \ge1 C seed ≥ 1 (its mass, coefficients and majorant sums); and its constants of the kinds in item (3).
(3) The invariant’s constants, weaker than the seed’s (constants larger; orders smaller): C Ω ≥ 3 C_{\Omega} \ge3 C Ω ≥ 3 (∣ ⋅ ∣ ∞ |\cdot|_{\infty} ∣ ⋅ ∣ ∞ -norms of the matrices of the smoothings and of the inverses of their long axes); c j c_j c j (the smoothings); C w i d C_{\mathrm{wid}} C wid (the widths); C d e r C_{\mathrm{der}} C der , C j e t ( j ) C_{\mathrm{jet}}(j) C jet ( j ) (the derivative bounds); the weight order N a n ≥ N c o e f N_{\mathrm{an}} \ge N_{\mathrm{coef}} N an ≥ N coef of the analytic norm. They are so large that Steps 6 and 7 of the proof of Lemma 11.7 close; in particular, the c j c_j c j are at least the c j ( u , ℓ ) c_j(u,\ell) c j ( u , ℓ ) of Lemma 8.2 for the kernel orders in use.
(4) C a d m ≥ C s e e d C_{\mathrm{adm}} \ge C_{\mathrm{seed}} C adm ≥ C seed , the constant of the majorant and coupling sums, and C m a j C_{\mathrm{maj}} C maj , the factor that a cycle applies to majorants, both large enough given item (3) (Lemma 11.6 ; Step 5 of the proof of Lemma 11.7 ); the carrier level m c m_c m c , the coarsest level of the chart parameters ((121) ); the top level m 0 ≥ max { m c , 2 } m_0 \ge\max\{m_c,2\} m 0 ≥ max { m c , 2 } of Subsection 5.3, the largest of the thresholds on the level in Sections 11–14, in particular so large that C a d m ε q d / 4 m − 1 ≤ 1 8 C_{\mathrm{adm}}\varepsilon_q^{d/4}m^{-1} \le\frac{1}{8} C adm ε q d /4 m − 1 ≤ 8 1 for every m ≥ m 0 m \ge m_0 m ≥ m 0 , that ∑ i > m 0 ( 3 τ i + 4 h i ) < 1 2 t 0 \sum_{i>m_0}(3\tau_i+4h_i) < \frac{1}{2}t_0 ∑ i > m 0 ( 3 τ i + 4 h i ) < 2 1 t 0 (Lemma 11.8 (iv)), and that the fresh smoothings (those of the sources a cycle creates, Definition 11.3 (iv)) created at the levels finer than m 0 m_0 m 0 have matrices of long axes with inverses of ∣ ⋅ ∣ ∞ |\cdot|_{\infty} ∣ ⋅ ∣ ∞ -norm ≤ 2 \le2 ≤ 2 . Also, C m a j C_{\mathrm{maj}} C maj is large enough given the constants of Lemma 15.4 (b) and of Steps 4 and 5 of the proof of Lemma 11.7 for the terminal cycle (Proposition 15.10 ).
Definitions and the main proposition
Sources. A source σ : v ′ → v \sigma:v'\to v σ : v ′ → v carries a birth order d σ d_\sigma d σ , its order when created, and a current order , beyond which its coefficients vanish; transports keep both, and only for a seed’s source may the current order be below d σ d_\sigma d σ . It also carries a birth radius ρ σ \rho_\sigma ρ σ ; a factor fac ( σ ) > 0 \operatorname{fac}(\sigma)>0 fac ( σ ) > 0 , the constant in front of its two-scale forms in (116) ; and a majorant maj ( σ ) ≥ 0 \operatorname{maj}(\sigma)\ge0 maj ( σ ) ≥ 0 , which bounds its two-scale forms, so that fac ( σ ) maj ( σ ) \operatorname{fac}(\sigma)\operatorname{maj}(\sigma) fac ( σ ) maj ( σ ) bounds the source. The generation gen ( v ) \operatorname{gen}(v) gen ( v ) counts the cycles that created v v v .
Two-scale forms. On the window ∣ t − k τ m ∣ ≤ 61 48 τ m |t-k\tau_m|\le\frac{61}{48}\tau_m ∣ t − k τ m ∣ ≤ 48 61 τ m of a pulse k k k of level m m m (Section 10), σ \sigma σ acts in the chart of k k k (Section 7), at the label y y y , through the coefficients
k m ε m d − 2 fac ( σ ) A σ , d ( t , y , y ε m ) , 2 ≤ d ≤ d σ , (116) k_m\varepsilon_m^{d-2}\operatorname{fac}(\sigma)\mathcal{A}_{\sigma,d}\left(t,y,\frac{y}{\varepsilon_m}\right),\qquad2\le d\le d_\sigma, \tag*{(116)} k m ε m d − 2 fac ( σ ) A σ , d ( t , y , ε m y ) , 2 ≤ d ≤ d σ , (116)
with two-scale forms A σ , d ( t , y , z ) \mathcal{A}_{\sigma,d}(t,y,z) A σ , d ( t , y , z ) in the sense of Section 10, periodic in the fast variable z z z , with values indexed by the two coordinates of the flux and the entries of ∇ d − 1 \nabla^{d-1} ∇ d − 1 . They also depend on chart parameters , the shears of the coarser levels (precise form (11.7)), and (116) uses those of the chart of k k k . The A σ , d \mathcal{A}_{\sigma,d} A σ , d of (34) are the sums over k k k , with time cutoffs, of the transports of the coefficients (116) from y y y to x x x , order d d d giving orders 2 , … , d 2,\ldots,d 2 , … , d (precise form (11.8)).
A two-scale form A \mathcal{A} A has an analytic norm ∥ A ∥ a n , v \|\mathcal{A}\|_{\mathrm{an},v} ∥ A ∥ an , v , a weighted sum over its Fourier modes in z z z of its suprema over a complex neighbourhood of the chart parameters that shrinks by a fixed step per generation of v v v (precise form (11.9)). Cauchy’s estimate bounds the derivatives of a cycle at the cost of this step, for at most N g e n N_{\mathrm{gen}} N gen generations. The majorant matrix and its priced form are
( ω v ′ ω v ) 1 / 2 ∑ σ : v ′ → v fac ( σ ) maj ( σ ) , \left(\frac{\omega_{v'}}{\omega_v}\right)^{1/2}\sum_{\sigma:v'\to v}\operatorname{fac}(\sigma)\operatorname{maj}(\sigma), ( ω v ω v ′ ) 1/2 σ : v ′ → v ∑ fac ( σ ) maj ( σ ) ,
( ω v ′ ω v ) 1 / 2 ∑ σ : v ′ → v fac ( σ ) max { 1 , ( ρ σ r σ ) d σ − 2 } maj ( σ ) ; (117) \left(\frac{\omega_{v'}}{\omega_v}\right)^{1/2}\sum_{\sigma:v'\to v}\operatorname{fac}(\sigma)\max\left\{1,\left(\frac{\rho_\sigma}{r_\sigma}\right)^{d_\sigma-2}\right\}\operatorname{maj}(\sigma); \tag*{(117)} ( ω v ω v ′ ) 1/2 σ : v ′ → v ∑ fac ( σ ) max { 1 , ( r σ ρ σ ) d σ − 2 } maj ( σ ) ; (117)
the price compensates for the d − 2 d-2 d − 2 derivatives that (37) puts on the kernel of width r σ r_\sigma r σ ; for the masses ω v / Ψ v 2 \omega_v/\Psi_v^2 ω v / Ψ v 2 , with the markers Ψ v \Psi_v Ψ v of Subsection 5.4, the entries are multiplied by Ψ v / Ψ v ′ \Psi_v/\Psi_{v'} Ψ v / Ψ v ′ .
Definitions and the main proposition
Each clause is stated here in words; its precise form, given in Subsection 11.3, is the definition.
Definition 11.1 (The invariant). A family of level m m m satisfies the invariant on [ t A , t f ] [t_A,t_f] [ t A , t f ] if, with the constants of items (1), (3) and (4) of Subsection 11.1 (for a seed: its constants, its derivative order (item (2)) in place of N c o e f N_{\mathrm{coef}} N coef in (b2), (b2′), (c3) and (d), and C s e e d C_{\mathrm{seed}} C seed for C a d m C_{\mathrm{adm}} C adm ):
(a) Cell structure (cycles solve one-dimensional cell problems; charts exact; precise forms ((11.10)–(11.13)}):
(a1) the two-scale forms are built from the construction’s functions and the family’s cell solutions (built from the solutions of its two-dimensional cell problems) by sums, products, derivatives, averages and primitives in z z z , the cell inverse across the pulse and changes of chart;
(a2) on the window of pulse k k k , at all chart parameters, the two-scale forms of k k k are transverse (§ 5.2 ), that is, independent of the coordinate of z z z along e ( k ) e(k) e ( k ) ;
(a3) at every t ∈ [ t A , t f ] t \in[t_A,t_f] t ∈ [ t A , t f ] with ∣ t − k τ m ∣ ≤ 3 4 τ m |t-k\tau_m| \le\frac{3}{4}\tau_m ∣ t − k τ m ∣ ≤ 4 3 τ m , the forms of all pulses whose windows contain t t t are invariant along e ( k ) e(k) e ( k ) ;
(a4) at every t ∈ [ t A , t f ] t \in[t_A,t_f] t ∈ [ t A , t f ] such that the velocity added at level m m m vanishes on [ t − τ m 24 , t + τ m 24 ] \left[t-\frac{\tau_m}{24},t+\frac{\tau_m}{24}\right] [ t − 24 τ m , t + 24 τ m ] , the forms of every pulse whose window contains t t t do not depend on z z z ;
(a5) an integer shift of the base label does not change the forms;
(a6) where the velocity added at level m m m vanishes, consecutive charts describe one coefficient: the change of chart maps (116) of one to the next’s, with z z z fixed;
(a7) near each chart parameter, a pulse’s forms depend only on the flow over its history .
(b) Majorants (the derivative bounds of § 5.6 –§ 5.8 ). Here ∥ ⋅ ∥ L m ∞ ( [ t A , t f ] ) \lVert\cdot\rVert_{L^\infty_m([t_A,t_f])} ∥ ⋅ ∥ L m ∞ ([ t A , t f ]) is the norm of (11.14), and (b2)–(b3) hold for both masses.
(b1) for every source σ : v ′ → v \sigma:v'\to v σ : v ′ → v and 2 ≤ d ≤ d σ 2\le d\le d_\sigma 2 ≤ d ≤ d σ , the entries of A σ , d \mathcal{A}_{\sigma,d} A σ , d extend as in (11.9) and ( ε m / ρ σ ) d − 2 ∥ A σ , d ∥ a n , v ≤ maj ( σ ) (\varepsilon_m/\rho_\sigma)^{d-2}\lVert\mathcal{A}_{\sigma,d}\rVert_{\mathrm{an},v}\le\operatorname{maj}(\sigma) ( ε m / ρ σ ) d − 2 ∥ A σ , d ∥ an , v ≤ maj ( σ ) ;
(b2) for every v v v , v ′ v' v ′ , in each pulse’s chart, the sum
( ω v ′ ω v ) 1 / 2 ∑ σ : v ′ → v fac ( σ ) ∑ d = 2 d σ ∑ j , j ′ ≤ N c o e f ε m − 1 j ∥ ∇ y j ∇ z j ′ A σ , d ∥ L m ∞ ( [ t A , t f ] ) (118) \left(\frac{\omega_{v'}}{\omega_v}\right)^{1/2}\sum_{\sigma:v'\to v}\operatorname{fac}(\sigma)\sum_{d=2}^{d_\sigma}\sum_{j,j'\le N_{\mathrm{coef}}}\varepsilon_{m-1}^{j}\lVert\nabla_y^j\nabla_z^{j'}\mathcal{A}_{\sigma,d}\rVert_{L^\infty_m([t_A,t_f])} \tag*{(118)} ( ω v ω v ′ ) 1/2 σ : v ′ → v ∑ fac ( σ ) d = 2 ∑ d σ j , j ′ ≤ N coef ∑ ε m − 1 j ∥ ∇ y j ∇ z j ′ A σ , d ∥ L m ∞ ([ t A , t f ]) (118)
is at most C d e r C_{\mathrm{der}} C der times the entry ( v , v ′ ) (v,v') ( v , v ′ ) of the majorant matrix;
(b2′) for every N N N , j ≤ N c o e f j\le N_{\mathrm{coef}} j ≤ N coef , the sums of (68) of order N N N at scale ρ m o l ( m ) \rho_{\mathrm{mol}}(m) ρ mol ( m ) for (116) , after j j j more slow derivatives (in y y y ), are at most C j e t ( j ) ε m − 1 − j C_{\mathrm{jet}}(j)\varepsilon_{m-1}^{-j} C jet ( j ) ε m − 1 − j times the majorant matrix (precise form (11.15));
(b3) the priced majorant matrix has row and column sums at most C a d m ε m δ / 4 C_{\mathrm{adm}}\varepsilon_m^{\delta/4} C adm ε m δ /4 .
(c) Smoothing (the c j c_j c j of Subsection 5.4; moments; small widths), for every source σ \sigma σ :
(c1) S σ ( t ) = S ℓ , r σ , Ω 1 ( t ) S ℓ , r σ , Ω 2 ( t ) S_\sigma(t)=S_{\ell,r_{\sigma,\Omega_1}}(t)S_{\ell,r_{\sigma,\Omega_2}}(t) S σ ( t ) = S ℓ , r σ , Ω 1 ( t ) S ℓ , r σ , Ω 2 ( t ) as in (44) (ℓ \ell ℓ , Ω i \Omega_i Ω i depending on σ \sigma σ ), 2 ℓ − 1 ≥ N m o m 2\ell-1\ge N_{\mathrm{mom}} 2 ℓ − 1 ≥ N mom , Ω i \Omega_i Ω i Borel measurable in t t t , ∣ Ω i ( t ) ∣ ≤ C Ω |\Omega_i(t)|\le C_\Omega ∣ Ω i ( t ) ∣ ≤ C Ω , and the matrix of long axes Ω 1 ( t ) e 1 , Ω 2 ( t ) e 2 \Omega_1(t)e_1,\Omega_2(t)e_2 Ω 1 ( t ) e 1 , Ω 2 ( t ) e 2 invertible, with inverse of ∣ ⋅ ∣ ∞ \lvert\cdot\rvert_\infty ∣ ⋅ ∣ ∞ -norm ≤ C Ω \le C_\Omega ≤ C Ω ;
(c2) r σ ≤ C w i d ρ m o l ( m ) ε m δ r_\sigma\le C_{\mathrm{wid}}\rho_{\mathrm{mol}}(m)\varepsilon_m^\delta r σ ≤ C wid ρ mol ( m ) ε m δ and r σ ≤ C w i d ρ ( m ) ε m − 1 4 δ r_\sigma\le C_{\mathrm{wid}}\rho(m)\varepsilon_{m-1}^{4\delta} r σ ≤ C wid ρ ( m ) ε m − 1 4 δ ;
(c3) for every t ∈ [ t A , t f ] t\in[t_A,t_f] t ∈ [ t A , t f ] , ρ > 0 \rho>0 ρ > 0 , N ≤ N c o e f N\le N_{\mathrm{coef}} N ≤ N coef and j ≤ N s r c − 2 j\le N_{\mathrm{src}}-2 j ≤ N src − 2 , ∥ ∇ j S σ ( t ) g ∥ H ρ N ( T 2 ) ≤ c j r σ − j ∥ g ∥ H ρ N ( T 2 ) \lVert\nabla^j S_\sigma(t)g\rVert_{H_\rho^N(\mathbb{T}^2)}\le c_jr_\sigma^{-j}\lVert g\rVert_{H_\rho^N(\mathbb{T}^2)} ∥ ∇ j S σ ( t ) g ∥ H ρ N ( T 2 ) ≤ c j r σ − j ∥ g ∥ H ρ N ( T 2 ) (precise form (11.16)). (d) Coupling (for Proposition 8.3 ): for every N ≤ N c o e f N \le N_{\mathrm{coef}} N ≤ N coef and both masses, the coupling matrices (68) on [ t A , t f ] [t_A,t_f] [ t A , t f ] of order N N N at scale ρ m o l ( m ) \rho_{\mathrm{mol}}(m) ρ mol ( m ) have row and column sums at most C a d m ε m − 1 δ q / 4 C_{\mathrm{adm}}\varepsilon_{m-1}^{\delta q/4} C adm ε m − 1 δ q /4 .
Definition 11.2 (Admissible seed). An admissible seed of level m S < M m_S<M m S < M on [ t S , t f ] [t_S,t_f] [ t S , t f ] , t f − t S ≤ T t_f-t_S\le T t f − t S ≤ T , is a family of level m S m_S m S whose components have generation 0 and marker 1, such that:
(i) its total mass is at most C s e e d C_{\mathrm{seed}} C seed ; its paths of sources and current orders obey the seed’s bounds; its widths are at most ε m S \varepsilon_{m_S} ε m S , and so are the birth radii of its sources of birth order at least 3;
(ii) for every source σ \sigma σ , 0 < ρ ≤ ε m S 0<\rho\le\varepsilon_{m_S} 0 < ρ ≤ ε m S , N N N at most the seed’s derivative order (item (2)) and 2 ≤ d ≤ d σ 2\le d\le d_\sigma 2 ≤ d ≤ d σ , the sum of (68) satisfies
∑ j ≤ N ρ j j ! sup [ t S , t f ] × T 2 ∣ ∇ j A σ , d ∣ ≤ C s e e d maj ( σ ) κ m S ρ σ d − 2 ; \sum_{j\le N}\frac{\rho^j}{j!}\sup_{[t_S,t_f]\times\mathbb{T}^2}\lvert\nabla^j A_{\sigma,d}\rvert\le C_{\mathrm{seed}}\operatorname{maj}(\sigma)\kappa_{m_S}\rho_\sigma^{d-2}; j ≤ N ∑ j ! ρ j [ t S , t f ] × T 2 sup ∣ ∇ j A σ , d ∣ ≤ C seed maj ( σ ) κ m S ρ σ d − 2 ; (iii) its cell solutions are built from finitely many two-dimensional cell problems, each solved once on the window of its pulse, and are unchanged when the phases move along the pulses (precise form (11.17);
(iv) it satisfies the invariant on [ t S , t f ] [t_S,t_f] [ t S , t f ] , also with the supremum in time in (b2′) (precise form (11.17)).
Definition 11.3 (Successor). Let G G G be a family of level m > m 0 m>m_0 m > m 0 with finite lifts along its paths of sources (§ 5.2 ). A successor G ′ G' G ′ of G G G , of level m − 1 m-1 m − 1 , is a family that the comparison of § 5.8 constructs with the lift (38) , solving no new two-dimensional cell problem; its data are:
(i) Letters : the defect letter (the defect of the averaged flux at level m m m ; birth order N c o r + 1 + 4 p N_{\mathrm{cor}}+1+4p N cor + 1 + 4 p ); for every 3 ≤ d ≤ N s r c − N c o r − 4 p − 4 3\le d\le N_{\mathrm{src}}-N_{\mathrm{cor}}-4p-4 3 ≤ d ≤ N src − N cor − 4 p − 4 and every nonzero path length up to the path bound N p a t h N_{\mathrm{path}} N path , a mean letter (birth order d d d ) for the means of degree d d d of the lift along paths of that length (for a path of one source, less the cell average of that source in the order d d d , which its transport carries); for every 2 ≤ d ≤ N s r c − N c o r − 4 p − 4 2\le d\le N_{\mathrm{src}}-N_{\mathrm{cor}}-4p-4 2 ≤ d ≤ N src − N cor − 4 p − 4 , every nonzero path length up to N p a t h N_{\mathrm{path}} N path and every j , j ′ ≤ p j,j'\le p j , j ′ ≤ p with j + j ′ ≥ 1 j+j'\ge1 j + j ′ ≥ 1 , a covariance letter (birth order d + 2 ( j + j ′ ) d+2(j+j') d + 2 ( j + j ′ ) ) for the means of degree d d d along paths of that length, with no subtraction, conjugated by ( − Γ m : ∇ 2 ) j / j ! (-\Gamma_m:\nabla^2)^j/j! ( − Γ m : ∇ 2 ) j / j ! on the left and ( Γ m : ∇ 2 ) j ′ / j ′ ! (\Gamma_m:\nabla^2)^{j'}/j'! ( Γ m : ∇ 2 ) j ′ / j ′ ! on the right (summed over j , j ′ ≤ p j,j'\le p j , j ′ ≤ p , the term j = j ′ = 0 j=j'=0 j = j ′ = 0 included, these give the mean times the truncated exponentials (39) on both sides, as in the flux (172) ).
(ii) Components : those of G G G , with their masses and markers, and their children v ′ v' v ′ , labelled by nonempty words of letters, with gen ( v ′ ) = gen ( v ) + 1 ≤ N g e n \operatorname{gen}(v')=\operatorname{gen}(v)+1\le N_{\mathrm{gen}} gen ( v ′ ) = gen ( v ) + 1 ≤ N gen , # v ′ ≤ N l e n \#v'\le N_{\mathrm{len}} # v ′ ≤ N len and
ω v ′ : = 1 − 128 − η / 2 2 N w o r d ε m − 1 η / 2 ω v , Ψ v ′ : = ε m − 1 − η / 8 Ψ v , (119) \omega_{v'}:=\frac{1-128^{-\eta/2}}{2N_{\mathrm{word}}}\varepsilon_{m-1}^{\eta/2}\omega_v,\qquad\Psi_{v'}:=\varepsilon_{m-1}^{-\eta/8}\Psi_v, \tag*{(119)} ω v ′ := 2 N word 1 − 12 8 − η /2 ε m − 1 η /2 ω v , Ψ v ′ := ε m − 1 − η /8 Ψ v , (119) where N w o r d N_{\mathrm{word}} N word counts the nonempty words of length ≤ N l e n + 1 \le N_{\mathrm{len}}+1 ≤ N len + 1 .
(iii) Old sources , in each word: the transport of the cell average (the average in z z z ) of every source, in its orders d ≥ 3 d\ge3 d ≥ 3 (the same factor, width, smoothing, birth radius, birth order and current order; majorant times C m a j C_{\mathrm{maj}} C maj ); the mean of degree 2 along each nonempty path of sources (birth order 2; width, smoothing and birth radius of its widest source; product of factors; C m a j × C_{\mathrm{maj}}\times C maj × product of majorants).
(iv) New sources : for each letter and path it labels (each component, for the defect letter), a source from the first component’s word to the last one’s word plus the letter: coefficients multiplied by ε m − 1 − η \varepsilon_{m-1}^{-\eta} ε m − 1 − η , the fresh smoothing (186) with the ellipses of (11.18), birth radius ρ ( m ) \rho(m) ρ ( m ) , the letter’s birth order, factor the path’s product of factors (1 for the defect letter), majorant C m a j ε m − 1 δ − η C_{\mathrm{maj}}\varepsilon_{m-1}^{\delta-\eta} C maj ε m − 1 δ − η times the product of the majorants along the path.
The components of G ′ G' G ′ are thus the pairs ( v , w ) (v,w) ( v , w ) of a component v v v of G G G and a word w w w of length at most N l e n N_{\mathrm{len}} N len , with gen ( v ) < N g e n \operatorname{gen}(v)<N_{\mathrm{gen}} gen ( v ) < N gen if w w w is nonempty (the empty word gives v v v , a nonempty one a child
of v v v ). In each word w w w , a source of (iii) along a path from v ′ v' v ′ to v v v (one source, for a transport) goes from ( v ′ , w ) (v',w) ( v ′ , w ) to ( v , w ) (v,w) ( v , w ) , and a source of (iv) of a letter ℓ \ell ℓ along such a path (with v ′ = v v'=v v ′ = v for the defect letter) from ( v ′ , w ) (v',w) ( v ′ , w ) to ( v , w ℓ ) (v,w\ell) ( v , w ℓ ) . The children of word length N_{\len}+1 or generation N_{\gen}+1 are created with the masses and markers of (119) , then dropped with the sources at them (Subsection 13.3).
Definition 11.4 (Admissible family). A family G G G of level m ≥ m 0 m\ge m_0 m ≥ m 0 is admissible on [ t A , t f ] [t_A,t_f] [ t A , t f ] if there exist an admissible seed G m S G_{m_S} G m S of level m S ≥ m m_S\ge m m S ≥ m , m S > m 0 m_S>m_0 m S > m 0 , on [ t S , t f ] [t_S,t_f] [ t S , t f ] with t S + 1 2 t 0 < t f t_S+\frac{1}{2}t_0<t_f t S + 2 1 t 0 < t f , and families G m S − 1 , … , G m = G G_{m_S-1},\ldots,G_m=G G m S − 1 , … , G m = G , each G i − 1 G_{i-1} G i − 1 a successor of G i G_i G i , such that t A = t A ( m ) t_A=t_A^{(m)} t A = t A ( m ) , where t A ( m S ) : = t S t_A^{(m_S)}:=t_S t A ( m S ) := t S and, for every m < i ≤ m S m<i\le m_S m < i ≤ m S ,
t A ( i − 1 ) : = t A ( i ) + 3 τ i ≤ t f − 4 h i . (120) t_A^{(i-1)}:=t_A^{(i)}+3\tau_i\le t_f-4h_i. \tag*{(120)} t A ( i − 1 ) := t A ( i ) + 3 τ i ≤ t f − 4 h i . (120)
The path bound. For a family G G G and g ≥ 0 g\ge0 g ≥ 0 , let P g ( G ) P_g(G) P g ( G ) be the largest number of sources on a path of G G G whose components have generation at most g g g . We claim that in the families G i G_i G i of Definition 11.4 no source goes to a smaller generation and P_g(G_i)\le B+gN_{\len} for every g g g ; so every path of G i G_i G i has at most B+N_{\gen}N_{\len}\le N_{\path} sources. This holds for the seed (generation 0) and passes to a successor G ′ G' G ′ of G G G . A component ( u , w ) (u,w) ( u , w ) of G ′ G' G ′ , with u u u a component of G G G and w w w a word, has generation gen ( u ) + 1 \operatorname{gen}(u)+1 gen ( u ) + 1 if w w w is nonempty and gen ( u ) \operatorname{gen}(u) gen ( u ) otherwise; an old source goes from ( u , w ) (u,w) ( u , w ) to ( u ′ ′ , w ) (u'',w) ( u ′′ , w ) , and a new one from ( u , w ) (u,w) ( u , w ) to ( u ′ ′ , w ℓ ) (u'',w\ell) ( u ′′ , w ℓ ) , along a nonempty path of G G G from u u u to u ′ ′ u'' u ′′ , or with u ′ ′ = u u''=u u ′′ = u for the defect letter (Definition 11.3 ). So an old source adds the same 0 or 1 to both generations, and a new one ends at generation gen ( u ′ ′ ) + 1 ≥ gen ( u ) + 1 \operatorname{gen}(u'')+1\ge\operatorname{gen}(u)+1 gen ( u ′′ ) + 1 ≥ gen ( u ) + 1 , by the claim for G G G . Along a path of G ′ G' G ′ the word only grows, by a letter at each new source, so the path has at most N_{\len} new sources; it projects onto a path of G G G , each old source onto at least one source. If it ends at generation at most g g g with an empty word, its sources are all old, at most P g ( G ) P_g(G) P g ( G ) ; with a nonempty word, the projected path ends at generation at most g − 1 g-1 g − 1 , so lies in the generations at most g − 1 g-1 g − 1 by the monotonicity in G G G , and the path has at most P_{g-1}(G)+N_{\len} sources. Thus P_g(G')\le\max\{P_g(G),P_{g-1}(G)+N_{\len}\} .
Proposition 11.5 (Admissible families). An admissible family of level m m m on [ t A , t f ] [t_A,t_f] [ t A , t f ] satisfies the invariant with the constants of items (1), (3) and (4) of Subsection 11.1, which depend only on α \alpha α , s s s , q q q , ζ \zeta ζ , Λ \Lambda Λ , n n n , t 0 t_0 t 0 , T T T . Moreover, for its sources with a fresh smoothing the inverse in (c1) has ∣ ⋅ ∣ ∞ \lvert\cdot\rvert_\infty ∣ ⋅ ∣ ∞ -norm ≤ 2 \le2 ≤ 2 , so that the long axes are uniformly transverse in the sense of Proposition 6.1 , and (c3) holds with the constants c j ( u , ℓ ) c_j(u,\ell) c j ( u , ℓ ) of Lemma 8.2 ; and the sums in (d) are at most C a d m ε m − 1 δ 0 / 4 ≤ 1 8 C_{\mathrm{adm}}\varepsilon_{m-1}^{\delta_0/4}\le\frac18 C adm ε m − 1 δ 0 /4 ≤ 8 1 .
In the main estimate of the proof, the recursion for (b3), a new source gains ε m − 1 δ − η \varepsilon_{m-1}^{\delta-\eta} ε m − 1 δ − η , loses \varepsilon_{m-1}^{-3\eta/8-(N_{\src}-2)\Theta} to its masses, markers and price, and keeps ε m − 1 4 δ / 64 ≤ ε m − 1 δ / 4 \varepsilon_{m-1}^{4\delta/64}\le\varepsilon_{m-1}^{\delta/4} ε m − 1 4 δ /64 ≤ ε m − 1 δ /4 , as \eta,(N_{\src}-2)\Theta\le\delta/8 .
Precise form of the admissibility conditions
Pulses and chains. The chart of a pulse k k k of level m m m is used where its time cutoff H m , k H_{m,k} H m , k (93) is positive; quiet times are those at which the velocity added at level m m m vanishes; c_{\an}:=1/20000 is the strip constant.
(11.7) Chart parameters. A chain C C C of k k k consists of a level m ′ ≤ m m'\le m m ′ ≤ m , pulses k i k_i k i of the levels m ′ ≤ i ≤ m m'\le i\le m m ′ ≤ i ≤ m , k ˉ m = k \bar{k}_m=k k ˉ m = k , each k i + 1 k_{i+1} k i + 1 with its chart used in the window of k i k_i k i , a base label y ( m ′ − 1 ) ∈ R 2 y^{(m'-1)}\in\mathbb{R}^2 y ( m ′ − 1 ) ∈ R 2 , and, for each m ′ ≤ i ≤ m m'\le i\le m m ′ ≤ i ≤ m , a centre y ( i ) ∈ R 2 y^{(i)}\in\mathbb{R}^2 y ( i ) ∈ R 2 and a phase z ( i ) ∈ R 2 z^{(i)}\in\mathbb{R}^2 z ( i ) ∈ R 2 . With t ˉ i , k : = ℓ i k i \bar{t}_{i,k}:=\ell_i k_i t ˉ i , k := ℓ i k i , the centre of the block of a pulse k k k of level i i i , and t ˉ i : = t ˉ i , k i \bar{t}_i:=\bar{t}_{i,k_i} t ˉ i := t ˉ i , k i , its flows X i \mathcal{X}_i X i and centre points x ˉ i \bar{x}_i x ˉ i are X m ′ − 1 : = X m ′ − 1 \mathcal{X}_{m'-1}:=\mathcal{X}_{m'-1} X m ′ − 1 := X m ′ − 1 , x ˉ m ′ − 1 : = y ( m ′ − 1 ) \bar{x}_{m'-1}:=y^{(m'-1)} x ˉ m ′ − 1 := y ( m ′ − 1 ) and, for m ′ ≤ i < m m'\le i<m m ′ ≤ i < m , x ˉ i : = X i − 1 ( t ˉ i + 1 , x ˉ i + 1 + y ( i ) , t ˉ i ) \bar{x}_i:=\mathcal{X}_{i-1}(\bar{t}_{i+1},\bar{x}_{i+1}+y^{(i)},\bar{t}_i) x ˉ i := X i − 1 ( t ˉ i + 1 , x ˉ i + 1 + y ( i ) , t ˉ i ) and
X i ( t ′ , ⋅ , t ) : = X i − 1 ( t ′ , ⋅ , t ˉ i ) ∘ [ y ↦ y + ( ∫ t t ′ ξ ~ i , k i ( t ′ ′ ) d t ′ ′ ) ∇ ψ i , k i ( y + ε i z ( i ) − x ˉ i − 1 − y ( i ) ) ] ∘ X i − 1 ( t ˉ i , ⋅ , t ) ; \mathcal{X}_i(t',\cdot,t):=\mathcal{X}_{i-1}(t',\cdot,\bar{t}_i)\circ\left[y\mapsto y+\left(\int_t^{t'}\widetilde{\xi}_{i,k_i}(t'')\,\mathrm{d}t''\right)\nabla\psi_{i,k_i}(y+\varepsilon_i z^{(i)}-\bar{x}_{i-1}-y^{(i)})\right]\circ\mathcal{X}_{i-1}(\bar{t}_i,\cdot,t); X i ( t ′ , ⋅ , t ) := X i − 1 ( t ′ , ⋅ , t ˉ i ) ∘ [ y ↦ y + ( ∫ t t ′ ξ i , k i ( t ′′ ) d t ′′ ) ∇ ψ i , k i ( y + ε i z ( i ) − x ˉ i − 1 − y ( i ) ) ] ∘ X i − 1 ( t ˉ i , ⋅ , t ) ;
at the matched phase ε i z ( i ) = x ˉ i − 1 + y ( i ) \varepsilon_i z^{(i)}=\bar{x}_{i-1}+y^{(i)} ε i z ( i ) = x ˉ i − 1 + y ( i ) the bracket is the shear displacement of § 4.1 . The chain is allowed if m ′ ≥ m c m' \ge m_c m ′ ≥ m c , the carrier level, and, for every m ′ ≤ i < m m' \le i < m m ′ ≤ i < m , ∣ y ( i ) ∣ 1 < 3 c a n ε i / 4 \lvert y^{(i)}\rvert_1 < 3c_{\mathrm{an}}\varepsilon_i/4 ∣ y ( i ) ∣ 1 < 3 c an ε i /4 . The chain C [ y ] C^{[y]} C [ y ] recentred by y ∈ R 2 y \in\mathbb{R}^2 y ∈ R 2 has the centre point x ˉ m − 1 + y \bar{x}_{m-1}+y x ˉ m − 1 + y : it moves the base label if m ′ = m m'=m m ′ = m , and otherwise the last centre, keeping ε m − 1 z ( m − 1 ) − y ( m − 1 ) \varepsilon_{m-1}z^{(m-1)}-y^{(m-1)} ε m − 1 z ( m − 1 ) − y ( m − 1 ) fixed. For m ′ = m m'=m m ′ = m the chain is the chart of k k k with base label y ( m − 1 ) y^{(m-1)} y ( m − 1 ) . The two-scale forms of k k k depend also on the chain: each is a function A ( t , C , z ) \mathcal{A}(t,C,z) A ( t , C , z ) of t t t in the window of k k k , an allowed chain C C C of k k k and z z z , whose value at the chain with m ′ = m m'=m m ′ = m and base label y y y (the chain without shear ) is the two-scale form A ( t , y , z ) \mathcal{A}(t,y,z) A ( t , y , z ) of § 5.2 in (116) . The history of k k k is the smallest interval containing its window, t ˉ m \bar{t}_m t ˉ m and the histories of the pulses of level m + 1 m+1 m + 1 whose charts are used in the window of k k k (at the finest level M + n M+n M + n of b M + n \mathbf{b}_{M+n} b M + n : its window and t ˉ m \bar{t}_m t ˉ m ). For t t t in the window of k k k , a chain C C C of k k k , a pulse k ′ k' k ′ of level m + 1 m+1 m + 1 with H m + 1 , k ′ ( t ) > 0 H_{m+1,k'}(t)>0 H m + 1 , k ′ ( t ) > 0 and z ∈ C 2 z\in\mathbb{C}^2 z ∈ C 2 in the strip of the phases of (11.9), the chain extended at ( t , z ) (t,z) ( t , z ) has the data of C C C , k m + 1 = k ′ k_{m+1}=k' k m + 1 = k ′ , the phase z ( m ) = z z^{(m)}=z z ( m ) = z and the centre y ( m ) = ( ∫ t ˉ m + 1 t ζ ~ m , k ( t ′ ′ ) d t ′ ′ ) ∇ ⊥ ψ m , k ( ε m z ) y^{(m)}=\left(\int_{\bar{t}_{m+1}}^t\tilde{\zeta}_{m,k}(t'')\,\mathrm{d}t''\right)\nabla^\perp\psi_{m,k}(\varepsilon_m z) y ( m ) = ( ∫ t ˉ m + 1 t ζ ~ m , k ( t ′′ ) d t ′′ ) ∇ ⊥ ψ m , k ( ε m z ) . Its centre point at the time t t t is that of C C C , as the shear over [ t ˉ m + 1 , t ] [\bar{t}_{m+1},t] [ t ˉ m + 1 , t ] moves x ˉ m − 1 + y ( m ) \bar{x}_{m-1}+y^{(m)} x ˉ m − 1 + y ( m ) to x ˉ m − 1 \bar{x}_{m-1} x ˉ m − 1 ; the centre is along e ( k ) e_{(k)} e ( k ) , depends on z z z only through z ⋅ e ( k ) ⊥ z\cdot e^\perp_{(k)} z ⋅ e ( k ) ⊥ , and has ∣ y ( m ) ∣ 1 ≤ 4 ⋅ 10 − 6 c a n ε m \lvert y^{(m)}\rvert_1\le4\cdot10^{-6}c_{\mathrm{an}}\varepsilon_m ∣ y ( m ) ∣ 1 ≤ 4 ⋅ 1 0 − 6 c an ε m by (121) , as ∣ t − t ˉ m + 1 ∣ ≤ 1 2 h m + 1 + 3 4 τ m + 1 \lvert t-\bar{t}_{m+1}\rvert\le\frac{1}{2}h_{m+1}+\frac{3}{4}\tau_{m+1} ∣ t − t ˉ m + 1 ∣ ≤ 2 1 h m + 1 + 4 3 τ m + 1 . At ε m z = x ˉ m − 1 \varepsilon_m z=\bar{x}_{m-1} ε m z = x ˉ m − 1 its map on the history of k ′ k' k ′ is the shear displacement of (58) .
(11.8) Transport. Let x ↦ y x\mapsto y x ↦ y be an area-preserving change of variables. The Piola identity writes the divergence in y y y of a vector field F F F as ∇ x ⋅ ( ( ∇ x y ) − 1 F ) \nabla_x\cdot\left((\nabla_x y)^{-1}F\right) ∇ x ⋅ ( ( ∇ x y ) − 1 F ) , and the chain rule, applied one derivative at a time in the order of an entry of ∇ y d − 1 \nabla_y^{d-1} ∇ y d − 1 , writes that entry as a combination of the entries of ∇ x , … , ∇ x d − 1 \nabla_x,\ldots,\nabla_x^{d-1} ∇ x , … , ∇ x d − 1 . Substituting both into ∇ y ⋅ ( coefficient ⋅ ∇ y d − 1 g ) \nabla_y\cdot(\text{coefficient}\cdot\nabla_y^{d-1}g) ∇ y ⋅ ( coefficient ⋅ ∇ y d − 1 g ) for the orders 2 ≤ d ≤ d σ 2\le d\le d_\sigma 2 ≤ d ≤ d σ and collecting the terms in each entry of each ∇ x d ′ g \nabla_x^{d'}g ∇ x d ′ g gives the transported coefficients, of orders 2 ≤ d ′ ≤ d σ 2\le d'\le d_\sigma 2 ≤ d ′ ≤ d σ , indexed by the entries of ∇ x d ′ − 1 \nabla_x^{d'-1} ∇ x d ′ − 1 (not symmetrized).
(11.9) Analytic norm. For a component v v v , the reserve res ( v ) : = ( N g e n + 2 − gen ( v ) ) / ( N g e n + 2 + N o f f ) \operatorname{res}(v):=(N_{\mathrm{gen}}+2^{-\operatorname{gen}(v)})/(N_{\mathrm{gen}}+2+N_{\mathrm{off}}) res ( v ) := ( N gen + 2 − gen ( v ) ) / ( N gen + 2 + N off ) , with N o f f N_{\mathrm{off}} N off the offset of item (2). Let D v D_v D v be the set of complex chart parameters, with the discrete data of an allowed chain, such that ∣ Im y ( m ′ − 1 ) ∣ 1 < ( 1 + res ( v ) / 4 ) ε m ′ − 1 / 100 \lvert\operatorname{Im}y^{(m'-1)}\rvert_1<(1+\operatorname{res}(v)/4)\varepsilon_{m'-1}/100 ∣ Im y ( m ′ − 1 ) ∣ 1 < ( 1 + res ( v ) /4 ) ε m ′ − 1 /100 and, for every m ′ ≤ i < m m'\le i<m m ′ ≤ i < m ,
∣ y ( i ) ∣ 1 < 1 + res ( v ) 4 c a n ε i , ∣ Im z ( i ) ⋅ e ( k i ) ⊥ ∣ < ( 3 + res ( v ) 4 + 1 64 ) c a n , \lvert y^{(i)}\rvert_1<\frac{1+\operatorname{res}(v)}{4}c_{\mathrm{an}}\varepsilon_i,\qquad
\lvert\operatorname{Im}z^{(i)}\cdot e^\perp_{(k_i)}\rvert<\left(3+\frac{\operatorname{res}(v)}{4}+\frac{1}{64}\right)c_{\mathrm{an}}, ∣ y ( i ) ∣ 1 < 4 1 + res ( v ) c an ε i , ∣ Im z ( i ) ⋅ e ( k i ) ⊥ ∣ < ( 3 + 4 res ( v ) + 64 1 ) c an ,
with e ( k ) ⊥ e^\perp_{(k)} e ( k ) ⊥ of § 5.2 ; the outer domain is defined with 2 2 2 in place of res ( v ) \operatorname{res}(v) res ( v ) , and its real points are the allowed chains. If the entries of a two-scale form A \mathcal{A} A of k k k extend holomorphically to the outer domain (for each t t t ; Borel measurable in t t t , locally bounded uniformly in t t t ; periodic in the phases and independent of their coordinates along the pulses), then
∥ A ∥ a n , v : = ∑ ξ ∈ Z ∑ entries ξ ∈ Z exp ( 3 π c a n ∣ ξ ∣ ) ( 1 + ∣ ξ ∣ ) N a n sup ∣ A ^ ( t , C ; ξ ) ∣ , \lVert\mathcal{A}\rVert_{\mathrm{an},v}:=\sum_{\xi\in\mathbb{Z}}\sum_{\text{entries }\xi\in\mathbb{Z}}\exp(3\pi c_{\mathrm{an}}\lvert\xi\rvert)(1+\lvert\xi\rvert)^{N_{\mathrm{an}}}\sup\lvert\widehat{\mathcal{A}}(t,C;\xi)\rvert, ∥ A ∥ an , v := ξ ∈ Z ∑ entries ξ ∈ Z ∑ exp ( 3 π c an ∣ ξ ∣) ( 1 + ∣ ξ ∣ ) N an sup ∣ A ( t , C ; ξ )∣ ,
with A ^ ( t , C ; ξ ) \widehat{\mathcal{A}}(t,C;\xi) A ( t , C ; ξ ) the ξ \xi ξ -th Fourier coefficient in z ⋅ e ( k ) ⊥ z\cdot e^\perp_{(k)} z ⋅ e ( k ) ⊥ , N a n N_{\mathrm{an}} N an the weight order of item (3), and the supremum over the pulses k k k of level m m m , t t t in the window of k k k and C ∈ D v C\in D_v C ∈ D v .
(11.10) Clause (a1). Every entry of a two-scale form of k k k is built from: the Jacobian of X m − 1 ( t , ⋅ , t ˉ m ) \mathcal{X}_{m-1}(t,\cdot,\bar{t}_m) X m − 1 ( t , ⋅ , t ˉ m ) at x ˉ m − 1 \bar{x}_{m-1} x ˉ m − 1 , its inverse, the conductivity (as for the renewal charts of § 4.1 ) and the mass strain (the time derivative of the Jacobian times its inverse); the same for the chain extended at ( t , z ) (t,z) ( t , z ) (end of (11.7)) toward a pulse k ′ k' k ′ of level $m+1 whose chart is used in the window of k k k , where the Jacobian, its inverse and the conductivity take their identity values outside the support of the time cutoff of k ′ k' k ′ ; the drift of the cell problem of k k k ; the time cutoffs; the matrices J i κ i \mathbf{J}_i^{\kappa_i} J i κ i of (18) and Γ i \Gamma_i Γ i of (39) ; constants; and the family’s cell solutions. The operations are sums, products, derivatives in z z z across the pulse, in the phases and along recentring, averages and transverse primitives in z z z , the inverse of the one-dimensional cell operator across the pulse (zero at the start of the window), and the evaluation of an entry built in this way for a pulse k ′ k' k ′ as above, at the level m + 1 m+1 m + 1 and independent of the fast variable of k ′ k' k ′ , at the chain extended at ( t , z ) (t,z) ( t , z ) toward k ′ k' k ′ , where the time cutoff of k ′ k' k ′ is positive, times that cutoff.
(11.11) Clauses (a2)–(a5). They hold at all allowed chains; in (a5), the base label moves in Z 2 \mathbb{Z}^2 Z 2 .
(11.12) Clause (a6). If k ′ ≠ k k' \ne k k ′ = k is in the block of k k k or the next one and t ∈ [ t A , t f ] t \in[t_A,t_f] t ∈ [ t A , t f ] is quiet in both windows, then at t t t the forms of k k k , k ′ k' k ′ at chains without shear do not depend on z z z , and, as coefficients of symmetric arrays ∇ d − 1 \nabla^{d-1} ∇ d − 1 , (116) of k ′ k' k ′ at y ′ y' y ′ is the transport of that of k k k by y = X m − 1 ( t ˉ m , k , y ′ , t ˉ m , k ′ ) y=X_{m-1}(\bar{t}_{m,k},y',\bar{t}_{m,k'}) y = X m − 1 ( t ˉ m , k , y ′ , t ˉ m , k ′ ) .
(11.13) Clause (a7). If allowed chains C , C ′ C,C' C , C ′ of k k k have the same map y ↦ X m − 1 ( t , x ˉ m − 1 + y , t ˉ m ) y \mapsto X_{m-1}(t,\bar{x}_{m-1}+y,\bar{t}_m) y ↦ X m − 1 ( t , x ˉ m − 1 + y , t ˉ m ) on the history of k k k , then, for y y y near 0 0 0 and every t , z t,z t , z , their two-scale forms at C [ y ] C[y] C [ y ] and C ′ [ y ] C'[y] C ′ [ y ] agree.
(11.14) Clause (b2). In (118) , the norm is ∥ ⋅ ∥ L m ∞ ( [ t A , t f ] ) \lVert\cdot\rVert_{L^\infty_m([t_A,t_f])} ∥ ⋅ ∥ L m ∞ ([ t A , t f ]) , the essential supremum over t ∈ [ t A , t f ] t\in[t_A,t_f] t ∈ [ t A , t f ] of the supremum over the pulses k k k whose window contains t t t , the allowed chains C C C of k k k with all ∣ y ( i ) ∣ < c a n ε i / 4 |y^{(i)}|<c_{\mathrm{an}}\varepsilon_i/4 ∣ y ( i ) ∣ < c an ε i /4 and z ∈ R 2 z\in\mathbb{R}^2 z ∈ R 2 , summed over the entries; and ∇ y \nabla_y ∇ y differentiates A σ , d ( t , C [ y ] , z ) A_{\sigma,d}(t,C[y],z) A σ , d ( t , C [ y ] , z ) at y = 0 y=0 y = 0 (for a chain without shear, this is the ∇ y \nabla_y ∇ y of (118) ).
(11.15) Clause (b2′). For both masses, every v v v , v ′ v' v ′ and every N N N , j ≤ N c o e f j\le N_{\mathrm{coef}} j ≤ N coef , the sum (118) , with ∥ ⋅ ∥ L m ∞ ( [ t A , t f ] ) \lVert\cdot\rVert_{L^\infty_m([t_A,t_f])} ∥ ⋅ ∥ L m ∞ ([ t A , t f ]) and ∇ y \nabla_y ∇ y as in (11.14) and with the inner double sum of (118) replaced by
∑ j 1 + j 2 ≤ N ( N + d − 2 ) ! ρ m o l ( m ) j 1 + j 2 + 2 − d N ! j 1 ! j 2 ! ε m j 2 + 2 − d ∥ ∇ y j 1 ∇ z j 2 A σ , d ∥ L m ∞ ( [ t A , t f ] ) , \sum_{j_1+j_2\le N}\frac{(N+d-2)!\,\rho_{\mathrm{mol}}(m)^{j_1+j_2+2-d}}{N!\,j_1!\,j_2!\,\varepsilon_m^{j_2+2-d}}\left\lVert\nabla_y^{j_1}\nabla_z^{j_2}A_{\sigma,d}\right\rVert_{L^\infty_m([t_A,t_f])}, j 1 + j 2 ≤ N ∑ N ! j 1 ! j 2 ! ε m j 2 + 2 − d ( N + d − 2 )! ρ mol ( m ) j 1 + j 2 + 2 − d ∇ y j 1 ∇ z j 2 A σ , d L m ∞ ([ t A , t f ]) ,
is at most C j e t ( j ) ε m − 1 − j C_{\mathrm{jet}}(j)\varepsilon_{m-1}^{-j} C jet ( j ) ε m − 1 − j times the entry ( v , v ′ ) (v,v') ( v , v ′ ) of the majorant matrix.
(11.16) Clause (c3). The operators S σ ( t ) S_\sigma(t) S σ ( t ) commute with derivatives, t ↦ S σ ( t ) g t\mapsto S_\sigma(t)g t ↦ S σ ( t ) g is Borel measurable for every g ∈ L 2 ( T 2 ) g\in L^2(\mathbb{T}^2) g ∈ L 2 ( T 2 ) , and, for every g ∈ H ρ N ( T 2 ) g\in H^N_\rho(\mathbb{T}^2) g ∈ H ρ N ( T 2 ) , the bound holds.
(11.17) Seeds, (iii)–(iv). (iii) Each cell solution of the seed is given by an expression built as in (11.10) from finitely many two-dimensional cell problems, each solved once on the window of its pulse, in the cell time t κ min { i , M } / ε i 2 t\kappa_{\min\{i,M\}}/\varepsilon_i^2 t κ m i n { i , M } / ε i 2 of its level i i i , from its initial value, and depending, among the pulses of level i + 1 i+1 i + 1 , only on those whose charts are used in that window. It is unchanged when the phases of an allowed chain move along their pulses, and any two functions given by the same expression agree on the window and the allowed chains. (iv) In (11.15) at m = m S m=m_S m = m S , the essential supremum in time is replaced by the supremum.
(11.18) Fresh smoothings (Definition 11.3(iv)). The averaged flux J m κ m \mathbf{J}^{\kappa_m}_m J m κ m is diagonal and 4 h m 4h_m 4 h m -periodic, with its entries exchanged by a shift of 2 h m 2h_m 2 h m (Lemma 7.10). For ℓ ∈ Z \ell\in\mathbb{Z} ℓ ∈ Z , let t h o r ∈ [ ( ℓ + 1 2 ) h m , ( ℓ + 9 2 ) h m ) t_{\mathrm{hor}}\in[(\ell+\frac{1}{2})h_m,(\ell+\frac{9}{2})h_m) t hor ∈ [( ℓ + 2 1 ) h m , ( ℓ + 2 9 ) h m ) be a time at which the first entry of J m κ m \mathbf{J}^{\kappa_m}_m J m κ m is largest, t v e r t_{\mathrm{ver}} t ver the time of this interval at distance 2 h m 2h_m 2 h m from it, where the second entry is largest, and ℓ h o r \ell_{\mathrm{hor}} ℓ hor , ℓ v e r \ell_{\mathrm{ver}} ℓ ver the blocks containing them. The fresh smoothing of the successor of a family of level m m m is (186) with, for ( ℓ + 1 2 ) h m ≤ t < ( ℓ + 3 2 ) h m (\ell+\frac{1}{2})h_m\le t<(\ell+\frac{3}{2})h_m ( ℓ + 2 1 ) h m ≤ t < ( ℓ + 2 3 ) h m ,
Ω 1 ( t ) : = ∇ y X m − 1 ( t , y h o r , ℓ h o r h m ) ( κ m − 1 − 1 J m κ m ( t h o r ) ) 1 / 2 , y h o r : = X m − 1 ( ℓ h o r h m , 0 , ( ℓ + 1 2 ) h m ) , \Omega_1(t):=\nabla_yX_{m-1}(t,y_{\mathrm{hor}},\ell_{\mathrm{hor}}h_m)\left(\kappa_{m-1}^{-1}\mathbf{J}^{\kappa_m}_m(t_{\mathrm{hor}})\right)^{1/2},\qquad y_{\mathrm{hor}}:=X_{m-1}(\ell_{\mathrm{hor}}h_m,0,(\ell+\tfrac{1}{2})h_m), Ω 1 ( t ) := ∇ y X m − 1 ( t , y hor , ℓ hor h m ) ( κ m − 1 − 1 J m κ m ( t hor ) ) 1/2 , y hor := X m − 1 ( ℓ hor h m , 0 , ( ℓ + 2 1 ) h m ) ,
Ω 2 ( t ) : = ∇ y X m − 1 ( t , y v e r , ℓ v e r h m ) ( κ m − 1 − 1 J m κ m ( t v e r ) ) 1 / 2 , y v e r : = X m − 1 ( ℓ v e r h m , 0 , ( ℓ + 1 2 ) h m ) , \Omega_2(t):=\nabla_yX_{m-1}(t,y_{\mathrm{ver}},\ell_{\mathrm{ver}}h_m)\left(\kappa_{m-1}^{-1}\mathbf{J}^{\kappa_m}_m(t_{\mathrm{ver}})\right)^{1/2},\qquad y_{\mathrm{ver}}:=X_{m-1}(\ell_{\mathrm{ver}}h_m,0,(\ell+\tfrac{1}{2})h_m), Ω 2 ( t ) := ∇ y X m − 1 ( t , y ver , ℓ ver h m ) ( κ m − 1 − 1 J m κ m ( t ver ) ) 1/2 , y ver := X m − 1 ( ℓ ver h m , 0 , ( ℓ + 2 1 ) h m ) ,
with entrywise square roots. These are the ellipses of the averaged flux at its horizontal and vertical maxima, in the label of their block, carried to the time t t t by the coarse flow.
The induction
Proposition 11.5 is proved by induction along the cycles of Definition 11.4: Lemma 11.6 starts it at the seed, and Lemma 11.7, proved in Subsection 11.6 with the complex chains of Subsection 11.5, passes the invariant to the successor, at the cost of the time 3 τ m 3\tau_m 3 τ m . Lemma 11.8 records what § 5.8 and § 5.11 use about successors. In Subsections 11.4–11.6, C C C and c c c denote positive constants depending only on α , s , q , ζ , Λ , n , t 0 , T \alpha,s,q,\zeta,\Lambda,n,t_0,T α , s , q , ζ , Λ , n , t 0 , T , fixed before m 0 ≥ m c m_0\ge m_c m 0 ≥ m c , beyond which the inequalities below hold.
Lemma 11.6 (Seeds). *An admissible seed of level m S > m 0 m_{\mathrm S}>m_0 m S > m 0 on [ t S , t f ] [t_{\mathrm S},t_{\mathrm f}] [ t S , t f ] with t S + 1 2 t 0 < t f t_{\mathrm S}+\frac{1}{2}t_0<t_{\mathrm f} t S + 2 1 t 0 < t f satisfies the invariant on [ t S , t f ] [t_{\mathrm S},t_{\mathrm f}] [ t S , t f ] with the constants of items (1), (3) and (4) of Subsection 11.1.
Proof. By Definition 11.2(iv), the seed satisfies the invariant with its constants. Each clause of Definition 11.1 other than (d) weakens when a constant grows or an order decreases (in (b1), a smaller weight order decreases the analytic norm, and the reserves are the same, since the offset is), and (d) is increasing in the c j c_j c j . As the constants of families are weaker than the seed’s (items (1)–(3)), this gives the invariant except for the sums of (d); for the sums of (b3), C s e e d ε m S δ / 4 ≤ C a d m ε m S δ / 4 C_{\mathrm{seed}}\varepsilon_{\mathrm{m_S}}^{\delta/4}\le C_{\mathrm{adm}}\varepsilon_{\mathrm{m_S}}^{\delta/4} C seed ε m S δ /4 ≤ C adm ε m S δ /4 . By (b1) for the seed, the coefficient (11.2) of order d d d of a source σ \sigma σ is at most κ m S ρ d − 2 fac ( σ ) maj ( σ ) \kappa_{m_{\mathrm S}}\rho^{d-2}\operatorname{fac}(\sigma)\operatorname{maj}(\sigma) κ m S ρ d − 2 fac ( σ ) maj ( σ ) . Its transport to x x x gives the orders d ′ ≤ d d'\le d d ′ ≤ d at the cost ε m S − 1 d ′ − d ≤ ρ d ′ − d \varepsilon_{\mathrm{m_S}-1}^{d'-d}\le\rho^{d'-d} ε m S − 1 d ′ − d ≤ ρ d ′ − d : for d ′ < d d'<d d ′ < d the birth order of σ \sigma σ is at least 3, so ρ σ ≤ ε m S \rho_\sigma\le\varepsilon_{\mathrm{m_S}} ρ σ ≤ ε m S by Definition 11.2(i). Hence the argument of Step 8 of Subsection 11.6, at the level m S m_{\mathrm S} m S in place of m − 1 m-1 m − 1 and with the Cauchy estimates of Step 6, shows that the coupling matrices of (d), formed with the c j c_j c j of families, are at most C C C times the priced majorant matrix, entry by entry. By Lemma 7.1(a) and as C a d m C_{\mathrm{adm}} C adm is large given item (3), their sums are at most
C C s e e d ε m S δ / 4 ≤ 2 δ / 4 C C s e e d ε m S − 1 δ / 4 ≤ C a d m ε m S − 1 δ / 4 . CC_{\mathrm{seed}}\varepsilon_{\mathrm{m_S}}^{\delta/4}\le2^{\delta/4}CC_{\mathrm{seed}}\varepsilon_{\mathrm{m_S}-1}^{\delta/4}\le C_{\mathrm{adm}}\varepsilon_{\mathrm{m_S}-1}^{\delta/4}. C C seed ε m S δ /4 ≤ 2 δ /4 C C seed ε m S − 1 δ /4 ≤ C adm ε m S − 1 δ /4 .
Lemma 11.7 (One cycle). Let m > m 0 m>m_0 m > m 0 and t A + 3 τ m + 4 h m ≤ t f t_A+3\tau_m+4h_m\le t_{\mathrm f} t A + 3 τ m + 4 h m ≤ t f . Let G G G be a family of level m m m obtained by successors from an admissible seed on [ t S , t f ] [t_{\mathrm S},t_{\mathrm f}] [ t S , t f ] , with t S ≤ t A t_{\mathrm S}\le t_A t S ≤ t A , that satisfies the invariant on [ t A , t f ] [t_A,t_{\mathrm f}] [ t A , t f ] with the constants of items (1), (3) and (4) of Subsection 11.1, and let G ′ G' G ′ be a successor of G G G . Then G ′ G' G ′ satisfies the invariant on [ t A + 3 τ m , t f ] [t_A+3\tau_m,t_{\mathrm f}] [ t A + 3 τ m , t f ] with the same constants.
Proof of Proposition 11.5. Let G G G be admissible of level m m m on [ t A , t f ] [t_A,t_{\mathrm f}] [ t A , t f ] , with the seed G m S G_{m_{\mathrm S}} G m S and the families G i G_i G i of Definition 11.4. By Lemma 11.6, G m S G_{m_{\mathrm S}} G m S satisfies the invariant on [ t A ( m S ) , t f ] = [ t S , t f ] [t_A^{(m_{\mathrm S})},t_{\mathrm f}]=[t_{\mathrm S},t_{\mathrm f}] [ t A ( m S ) , t f ] = [ t S , t f ] , and for i = m S , … , m + 1 i=m_{\mathrm S},\ldots,m+1 i = m S , … , m + 1 in turn Lemma 11.7 passes it from G i G_i G i on [ t A ( i ) , t f ] [t_A^{(i)},t_{\mathrm f}] [ t A ( i ) , t f ] to G i − 1 G_{i-1} G i − 1 on [ t A ( i − 1 ) , t f ] [t_A^{(i-1)},t_{\mathrm f}] [ t A ( i − 1 ) , t f ] , since i > m 0 i>m_0 i > m 0 , t S ≤ t A ( i ) t_{\mathrm S}\le t_A^{(i)} t S ≤ t A ( i ) and t A ( i ) + 3 τ i + 4 h i ≤ t f t_A^{(i)}+3\tau_i+4h_i\le t_{\mathrm f} t A ( i ) + 3 τ i + 4 h i ≤ t f by (11.6). The fresh smoothings of the sources of G G G were created above m ≥ m 0 m\ge m_0 m ≥ m 0 , so Step 7 of Subsection 11.6 gives the inverse bound 2 and (c3) with the constants of Lemma 8.2; the bound 1 8 \frac{1}{8} 8 1 holds by item (4).
Lemma 11.8 (Successors). Let G G G be a family of level m > m 0 m>m_0 m > m 0 obtained from an admissible seed by successors, and let G ′ G' G ′ be a successor of G G G .
Each component of G G G is one of G ′ G' G ′ , with the same masses and markers.
The total mass of G ′ G' G ′ is at most twice that of the seed.
The readout (5.10) satisfies ∣ γ m ∣ ≤ 1 + N w o r d 1 / 2 ε m − 1 5 η / 4 |\gamma_m|\le1+N_{\mathrm{word}}^{1/2}\varepsilon_{m-1}^{5\eta/4} ∣ γ m ∣ ≤ 1 + N word 1/2 ε m − 1 5 η /4 .
For every admissible seed G m S G_{m_{\mathrm S}} G m S of level m S > m 0 m_{\mathrm S}>m_0 m S > m 0 on [ t S , t f ] [t_{\mathrm S},t_{\mathrm f}] [ t S , t f ] with t S + 1 2 t 0 < t f t_{\mathrm S}+\frac{1}{2}t_0<t_{\mathrm f} t S + 2 1 t 0 < t f , there exist families G m S − 1 , … , G m 0 G_{m_{\mathrm S}-1},\ldots,G_{m_0} G m S − 1 , … , G m 0 , each G i − 1 G_{i-1} G i − 1 a successor of G i G_i G i , for which (11.6) holds with m = m 0 m=m_0 m = m 0 . In particular, each G i G_i G i is an admissible family of level i i i on [ t A ( i ) , t f ] [t_A^{(i)},t_{\mathrm f}] [ t A ( i ) , t f ] .
Proof. (i) A component of G G G is labelled in G ′ G' G ′ by the empty word and has generation at most N g e n N_{\mathrm{gen}} N gen , so Definition 11.3(ii) keeps it with its mass and marker. (ii) By (11.5), a cycle at level i i i gives each component v v v at most N w o r d N_{\mathrm{word}} N word children, of total mass at most 1 2 ( 1 − 128 − η / 2 ) ε i − 1 η / 2 ω v \frac{1}{2}(1-128^{-\eta/2})\varepsilon_{i-1}^{\eta/2}\omega_v 2 1 ( 1 − 12 8 − η /2 ) ε i − 1 η /2 ω v , and ε i − 1 ≤ 128 − ( i − 1 ) \varepsilon_{i-1}\le128^{-(i-1)} ε i − 1 ≤ 12 8 − ( i − 1 ) by Lemma 7.1(a). So the cycles multiply the total mass by at most exp ( 1 2 ( 1 − 128 − η / 2 ) ∑ i ≥ 2 128 − η ( i − 1 ) / 2 ) ≤ exp ( 1 / 2 ) < 2 \exp(\frac{1}{2}(1-128^{-\eta/2})\sum_{i\ge2}128^{-\eta(i-1)/2})\le\exp(1/2)<2 exp ( 2 1 ( 1 − 12 8 − η /2 ) ∑ i ≥ 2 12 8 − η ( i − 1 ) /2 ) ≤ exp ( 1/2 ) < 2 . (iii) In the coordinates scaled by the masses to the power − 1 2 -\frac{1}{2} − 2 1 , the readout (5.10) is the identity on the components of G G G plus a matrix with one entry γ v ≠ v ′ ( ω v ′ / ω v ) 1 / 2 ≤ ε m − 1 5 η / 4 \gamma_{v\ne v'}(\omega_{v'}/\omega_v)^{1/2}\le\varepsilon_{m-1}^{5\eta/4} γ v = v ′ ( ω v ′ / ω v ) 1/2 ≤ ε m − 1 5 η /4 , by (11.5), for each component v v v and child v ′ v' v ′ of v v v . A row has at most N w o r d N_{\mathrm{word}} N word entries and a column one, so Schur’s test (Subsection 5.4) bounds this matrix by N w o r d 1 / 2 ε m − 1 5 η / 4 N_{\mathrm{word}}^{1/2}\varepsilon_{m-1}^{5\eta/4} N word 1/2 ε m − 1 5 η /4 . (iv)
By Proposition 11.5 and the setting of the lift (§ 5.8 ), where (b1), (b3), Cauchy’s estimate and (54) give the hypotheses of Proposition 10.6 , an admissible family of a level finer than m 0 m_0 m 0 has finite lifts up to t f t_f t f on the windows that start at or after t A t_A t A , as the cell inverse is solved forward from the start of the window ((11.10)); these include every window that meets [ t A + 3 τ m , t f ] [t_A+3\tau_m,t_f] [ t A + 3 τ m , t f ] , of length 61 24 τ m < 3 τ m \frac{61}{24}\tau_m<3\tau_m 24 61 τ m < 3 τ m by (93) . So § 5.8 constructs its successor. As Proposition 11.5 for G i G_i G i uses only G m S , … , G i G_{m_S},\ldots,G_i G m S , … , G i , the families are constructed one level at a time, up to G m 0 G_{m_0} G m 0 as long as (120) holds, and it holds: as t A ( i ) = t S + ∑ i ′ = i + 1 m S 3 τ i ′ t_A^{(i)}=t_S+\sum_{i'=i+1}^{m_S}3\tau_{i'} t A ( i ) = t S + ∑ i ′ = i + 1 m S 3 τ i ′ , item (4) gives, for every m 0 < i ≤ m S m_0<i\le m_S m 0 < i ≤ m S ,
t A ( i ) + 3 τ i + 4 h i ≤ t S + ∑ i ′ > m 0 ( 3 τ i ′ + 4 h i ′ ) < t S + 1 2 t 0 < t f . t_A^{(i)}+3\tau_i+4h_i\le t_S+\sum_{i'>m_0}(3\tau_{i'}+4h_{i'})<t_S+\frac{1}{2}t_0<t_f. t A ( i ) + 3 τ i + 4 h i ≤ t S + i ′ > m 0 ∑ ( 3 τ i ′ + 4 h i ′ ) < t S + 2 1 t 0 < t f .
Holomorphic extension of the flows
The analytic norm takes suprema over complex chart parameters, so the flows X i \mathcal{X}_i X i of a chain (11.7) must extend holomorphically to the outer domain. They are compositions of a coarse flow, extended by Lemma 7.6 , with entire shear displacements. The domains close because the renewal times of a chain are short, which the carrier level provides: m c m_c m c satisfies, for every level i ≥ m c − 1 i\ge m_c-1 i ≥ m c − 1 ,
ε i β − 2 h i + 1 ≤ 10 − 6 c a n , 48 h i + 2 < τ i + 1 , (121) \varepsilon_i^{\beta-2}h_{i+1}\le10^{-6}c_{\mathrm{an}},\qquad48h_{i+2}<\tau_{i+1}, \tag*{(121)} ε i β − 2 h i + 1 ≤ 1 0 − 6 c an , 48 h i + 2 < τ i + 1 , (121)
and such a level exists, by (15) and, for the second bound, by (7) and ε i + 1 ≤ 2 ε i q \varepsilon_{i+1}\le2\varepsilon_i^q ε i + 1 ≤ 2 ε i q , as q ( 2 − β + 2 δ ) − ( 2 − β + 4 δ ) = ( q − 1 ) ( 1 − β 2 q + 2 ) + 2 q δ > 0 q(2-\beta+2\delta)-(2-\beta+4\delta)=(q-1)(1-\frac{\beta}{2q+2})+2q\delta>0 q ( 2 − β + 2 δ ) − ( 2 − β + 4 δ ) = ( q − 1 ) ( 1 − 2 q + 2 β ) + 2 q δ > 0 by (5) . For a chain of a pulse k k k of level m m m and m ′ − 1 ≤ i ≤ m − 1 m'-1\le i\le m-1 m ′ − 1 ≤ i ≤ m − 1 , the flow X i \mathcal{X}_i X i is anchored at the centre t ˉ i + 1 \bar t_{i+1} t ˉ i + 1 of the block of k i + 1 k_{i+1} k i + 1 (k m = k k_m=k k m = k ); we call ∣ t − t ˉ i + 1 ∣ ≤ 3 5 h i + 1 |t-\bar t_{i+1}|\le\frac{3}{5}h_{i+1} ∣ t − t ˉ i + 1 ∣ ≤ 5 3 h i + 1 the interval of X i \mathcal{X}_i X i , and write x ˉ i ( t ) : = X i ( t , x ˉ i , t ˉ i + 1 ) \bar x_i(t):=\mathcal{X}_i(t,\bar x_i,\bar t_{i+1}) x ˉ i ( t ) := X i ( t , x ˉ i , t ˉ i + 1 ) for the centre point at time t t t .
The intervals of the flows and the histories of the pulses are nested. Let i ≥ m c i\ge m_c i ≥ m c , and let a pulse k i + 1 k_{i+1} k i + 1 of level i + 1 i+1 i + 1 have its chart used at a time s s s of the window of a pulse k i k_i k i of level i i i , so that ∣ s − k i + 1 τ i + 1 ∣ < 5 4 τ i + 1 |s-k_{i+1}\tau_{i+1}|<\frac{5}{4}\tau_{i+1} ∣ s − k i + 1 τ i + 1 ∣ < 4 5 τ i + 1 and ∣ s − k i τ i ∣ ≤ 61 48 τ i |s-k_i\tau_i|\le\frac{61}{48}\tau_i ∣ s − k i τ i ∣ ≤ 48 61 τ i by (93) . The time of a pulse of level j j j is within 1 2 h j \frac{1}{2}h_j 2 1 h j of the centre of its block (Lemma 7.1 (d)), and τ j ≤ h j / 25 \tau_j\le h_j/25 τ j ≤ h j /25 by (6) . Hence, by the second bound of (121) at the level i − 1 ≥ m c − 1 i-1\ge m_c-1 i − 1 ≥ m c − 1 and induction on the level, starting from the finest level (where a history is the window and t ˉ M + n \bar t_{M+n} t ˉ M + n ), the interval ∣ t − t ˉ i + 1 , k i + 1 ∣ ≤ 3 5 h i + 1 |t-\bar t_{i+1,k_{i+1}}|\le\frac{3}{5}h_{i+1} ∣ t − t ˉ i + 1 , k i + 1 ∣ ≤ 5 3 h i + 1 and the history of k i + 1 k_{i+1} k i + 1 lie within
( 3 5 + 1 2 ) h i + 1 + 5 4 τ i + 1 ≤ 6 5 h i + 1 < τ i 40 \left(\frac{3}{5}+\frac{1}{2}\right)h_{i+1}+\frac{5}{4}\tau_{i+1}\le\frac{6}{5}h_{i+1}<\frac{\tau_i}{40} ( 5 3 + 2 1 ) h i + 1 + 4 5 τ i + 1 ≤ 5 6 h i + 1 < 40 τ i
of s s s , hence within 1 2 h i + 61 48 τ i + τ i 40 < 3 5 h i \frac{1}{2}h_i+\frac{61}{48}\tau_i+\frac{\tau_i}{40}<\frac{3}{5}h_i 2 1 h i + 48 61 τ i + 40 τ i < 5 3 h i of t ˉ i , k i \bar t_{i,k_i} t ˉ i , k i , as 1 2 + 1 25 ( 61 48 + 1 40 ) < 3 5 \frac{1}{2}+\frac{1}{25}\left(\frac{61}{48}+\frac{1}{40}\right)<\frac{3}{5} 2 1 + 25 1 ( 48 61 + 40 1 ) < 5 3 ; so does the history of k i k_i k i .
Lemma 11.9 (Complex chains). Fix the levels and pulses of a chain of a pulse k k k of level m m m , with m ′ ≥ m c m'\ge m_c m ′ ≥ m c . For every m ′ − 1 ≤ i ≤ m − 1 m'-1\le i\le m-1 m ′ − 1 ≤ i ≤ m − 1 , the maps X i ( t ′ , y , t ) \mathcal{X}_i(t',y,t) X i ( t ′ , y , t ) , for t , t ′ t,t' t , t ′ in the interval of X i \mathcal{X}_i X i , and the centre points x ˉ i ( t ) \bar x_i(t) x ˉ i ( t ) defined above extend from the allowed chains to the parameters of the outer domain of (11.9) and to the points y y y with ∣ y − x ˉ i ( t ) ∣ ∞ < 8 c a n ε i |y-\bar x_i(t)|_\infty<8c_{\mathrm{an}}\varepsilon_i ∣ y − x ˉ i ( t ) ∣ ∞ < 8 c an ε i , holomorphically in the parameters and in y y y , such that, for every t , t ′ , t ′ ′ t,t',t'' t , t ′ , t ′′ in the interval of X i \mathcal{X}_i X i and every y y y with ∣ y − x ˉ i ( t ) ∣ ∞ < 8 c a n ε i |y-\bar x_i(t)|_\infty<8c_{\mathrm{an}}\varepsilon_i ∣ y − x ˉ i ( t ) ∣ ∞ < 8 c an ε i :
(i) X i ( t , y , t ) = y \mathcal{X}_i(t,y,t)=y X i ( t , y , t ) = y and ∣ ∇ y X i ( t ′ , y , t ) − I ∣ ∞ ≤ 2 25 ε i β − 2 ∣ t ′ − t ∣ |\nabla_y\mathcal{X}_i(t',y,t)-I|_\infty\le2^{25}\varepsilon_i^{\beta-2}|t'-t| ∣ ∇ y X i ( t ′ , y , t ) − I ∣ ∞ ≤ 2 25 ε i β − 2 ∣ t ′ − t ∣ ;
(ii) X i ( t ′ , x ˉ i ( t ) , t ) = x ˉ i ( t ′ ) \mathcal{X}_i(t',\bar x_i(t),t)=\bar x_i(t') X i ( t ′ , x ˉ i ( t ) , t ) = x ˉ i ( t ′ ) , and, if ∣ y − x ˉ i ( t ) ∣ ∞ < 2 c a n ε i |y-\bar x_i(t)|_\infty<2c_{\mathrm{an}}\varepsilon_i ∣ y − x ˉ i ( t ) ∣ ∞ < 2 c an ε i , then X i ( t ′ ′ , X i ( t ′ , y , t ) , t ′ ) = X i ( t ′ ′ , y , t ) \mathcal{X}_i(t'',\mathcal{X}_i(t',y,t),t')=\mathcal{X}_i(t'',y,t) X i ( t ′′ , X i ( t ′ , y , t ) , t ′ ) = X i ( t ′′ , y , t ) .
By (i), (ii) and 2 25 ε i β − 2 6 5 h i + 1 ≤ 1 100 2^{25}\varepsilon_i^{\beta-2}\frac{6}{5}h_{i+1}\le\frac{1}{100} 2 25 ε i β − 2 5 6 h i + 1 ≤ 100 1 , from (121) , the map X i ( t ′ , ⋅ , t ) \mathcal{X}_i(t',\cdot,t) X i ( t ′ , ⋅ , t ) sends every ball about x ˉ i ( t ) \bar x_i(t) x ˉ i ( t ) of radius at most 8 c a n ε i 8c_{\mathrm{an}}\varepsilon_i 8 c an ε i into the ball about x ˉ i ( t ′ ) \bar x_i(t') x ˉ i ( t ′ ) of 101 100 \frac{101}{100} 100 101 times that radius, and, by Cauchy’s estimate, its derivative is Lipschitz on the ball of radius 4 c a n ε i 4c_{\mathrm{an}}\varepsilon_i 4 c an ε i , with constant at most 1 / ( 4 c a n ε i ) 1/(4c_{\mathrm{an}}\varepsilon_i) 1/ ( 4 c an ε i ) . Proof. By induction on i i i . The first level. We apply Lemma 7.6 at the level m ′ − 1 m'-1 m ′ − 1 on the interval of X m ′ − 1 = X m ′ − 1 \mathcal{X}_{m'-1} = X_{m'-1} X m ′ − 1 = X m ′ − 1 , of length t 2 − t 1 = 6 5 h m ′ t_2-t_1=\frac{6}{5}h_{m'} t 2 − t 1 = 5 6 h m ′ , where x ˉ m ′ − 1 ( t ) = X m ′ − 1 ( t , y ( m ′ − 1 ) , t ˉ m ′ ) \bar{x}_{m'-1}(t)=X_{m'-1}(t,y^{(m'-1)},\bar{t}_{m'}) x ˉ m ′ − 1 ( t ) = X m ′ − 1 ( t , y ( m ′ − 1 ) , t ˉ m ′ ) . By (121) , the number 1 − 2 21 ε m ′ − 1 β − 2 ( t 2 − t 1 ) 1-2^{21}\varepsilon_{m'-1}^{\beta-2}(t_2-t_1) 1 − 2 21 ε m ′ − 1 β − 2 ( t 2 − t 1 ) of that lemma satisfies
1 − 2 21 ε m ′ − 1 β − 2 ⋅ 6 5 h m ′ ≥ 1 − 6 5 ⋅ 2 21 ⋅ 10 − 6 c a n ≥ 0.9998. 1-2^{21}\varepsilon_{m'-1}^{\beta-2}\cdot\frac{6}{5}h_{m'}\geq1-\frac{6}{5}\cdot2^{21}\cdot10^{-6}c_{\mathrm{an}}\geq0.9998. 1 − 2 21 ε m ′ − 1 β − 2 ⋅ 5 6 h m ′ ≥ 1 − 5 6 ⋅ 2 21 ⋅ 1 0 − 6 c an ≥ 0.9998.
In units of ε m ′ − 1 / 1000 \varepsilon_{m'-1}/1000 ε m ′ − 1 /1000 , in which c a n ε m ′ − 1 = 1 20 c_{\mathrm{an}}\varepsilon_{m'-1}=\frac{1}{20} c an ε m ′ − 1 = 20 1 , Lemma 7.6 with this bound gives holomorphy and (52) where ∣ Im y ∣ 1 < 0.9998 ⋅ 19 |\operatorname{Im}y|_1<0.9998\cdot19 ∣ Im y ∣ 1 < 0.9998 ⋅ 19 , and the group law where ∣ Im y ∣ 1 ≤ 0.9998 2 ⋅ 19 |\operatorname{Im}y|_1\leq0.9998^2\cdot19 ∣ Im y ∣ 1 ≤ 0.999 8 2 ⋅ 19 . In the outer domain ∣ Im y ( m ′ − 1 ) ∣ 1 < 15 |\operatorname{Im}y^{(m'-1)}|_1<15 ∣ Im y ( m ′ − 1 ) ∣ 1 < 15 , so ∣ Im x ˉ m ′ − 1 ( t ) ∣ 1 < 15 / 0.9998 < 15.2 |\operatorname{Im}\bar{x}_{m'-1}(t)|_1<15/0.9998<15.2 ∣ Im x ˉ m ′ − 1 ( t ) ∣ 1 < 15/0.9998 < 15.2 by (52) . As ∣ ⋅ ∣ 1 ≤ 2 ∣ ⋅ ∣ ∞ |\cdot|_1\leq2|\cdot|_\infty ∣ ⋅ ∣ 1 ≤ 2∣ ⋅ ∣ ∞ , the balls about x ˉ m ′ − 1 ( t ) \bar{x}_{m'-1}(t) x ˉ m ′ − 1 ( t ) of radii 8 c a n ε m ′ − 1 8c_{\mathrm{an}}\varepsilon_{m'-1} 8 c an ε m ′ − 1 and 2 c a n ε m ′ − 1 2c_{\mathrm{an}}\varepsilon_{m'-1} 2 c an ε m ′ − 1 , that is, 2 5 \frac{2}{5} 5 2 and 1 10 \frac{1}{10} 10 1 , lie in ∣ Im y ∣ 1 < 16 |\operatorname{Im}y|_1<16 ∣ Im y ∣ 1 < 16 and ∣ Im y ∣ 1 < 15.4 |\operatorname{Im}y|_1<15.4 ∣ Im y ∣ 1 < 15.4 . This gives (i), as 10 ⋅ 2 21 ≤ 2 25 10\cdot2^{21}\leq2^{25} 10 ⋅ 2 21 ≤ 2 25 , and, with the group law at the base label, (ii).
From i − 1 i-1 i − 1 to i i i . As the chart of k i + 1 k_{i+1} k i + 1 is used in the window of k i k_i k i , the interval of X i \mathcal{X}_i X i lies within 3 5 h i \frac{3}{5}h_i 5 3 h i of t ˉ i \bar{t}_i t ˉ i , inside that of X i − 1 \mathcal{X}_{i-1} X i − 1 , by the nesting above (i ≥ m ′ ≥ m c i\geq m'\geq m_c i ≥ m ′ ≥ m c ). For t , t ′ t,t' t , t ′ in it, (11.7) composes X i − 1 ( t ˉ i , ⋅ , t ) \mathcal{X}_{i-1}(\bar{t}_i,\cdot,t) X i − 1 ( t ˉ i , ⋅ , t ) , the shear map (the bracket in (11.7)) and X i − 1 ( t ′ , ⋅ , t ˉ i ) \mathcal{X}_{i-1}(t',\cdot,\bar{t}_i) X i − 1 ( t ′ , ⋅ , t ˉ i ) . The shear ∇ ⊥ ψ i , k i \nabla^\perp\psi_{i,k_i} ∇ ⊥ ψ i , k i is entire and depends on its argument only through the coordinate along e ( k i ) ⊥ e_{(k_i)}^\perp e ( k i ) ⊥ , which the shear map therefore does not change. Where the imaginary part of this coordinate is less than 15 c a n ε i 15c_{\mathrm{an}}\varepsilon_i 15 c an ε i , the shear is at most 2 π ε i β − 1 exp ( 30 π c a n ) 2\pi\varepsilon_i^{\beta-1}\exp(30\pi c_{\mathrm{an}}) 2 π ε i β − 1 exp ( 30 π c an ) and its derivative at most 4 π ε i β − 2 exp ( 30 π c a n ) 4\pi\varepsilon_i^{\beta-2}\exp(30\pi c_{\mathrm{an}}) 4 π ε i β − 2 exp ( 30 π c an ) , by (8) .
As 0 ≤ ζ ~ i , k i ≤ 1 0\leq\widetilde{\zeta}_{i,k_i}\leq1 0 ≤ ζ i , k i ≤ 1 by (4) and (12) , (121) shows that the shear map over an interval of length at most 6 5 h i + 1 \frac{6}{5}h_{i+1} 5 6 h i + 1 moves each point of the strip by at most 6 5 h i + 1 ⋅ 2 π ε i β − 1 exp ( 30 π c a n ) ≤ 10 − 5 c a n ε i \frac{6}{5}h_{i+1}\cdot2\pi\varepsilon_i^{\beta-1}\exp(30\pi c_{\mathrm{an}})\leq10^{-5}c_{\mathrm{an}}\varepsilon_i 5 6 h i + 1 ⋅ 2 π ε i β − 1 exp ( 30 π c an ) ≤ 1 0 − 5 c an ε i . The point x ˉ i − 1 + y ( i ) \bar{x}_{i-1}+y^{(i)} x ˉ i − 1 + y ( i ) , and its image under the shear map over [ t ˉ i + 1 , t ] [\bar{t}_{i+1},t] [ t ˉ i + 1 , t ] , which X i − 1 ( t , ⋅ , t ˉ i ) \mathcal{X}_{i-1}(t,\cdot,\bar{t}_i) X i − 1 ( t , ⋅ , t ˉ i ) maps to x ˉ i ( t ) \bar{x}_i(t) x ˉ i ( t ) by (11.7) and (ii) for X i − 1 \mathcal{X}_{i-1} X i − 1 , lie within ( 3 4 + 10 − 5 ) c a n ε i < 2 c a n ε i − 1 (\frac{3}{4}+10^{-5})c_{\mathrm{an}}\varepsilon_i<2c_{\mathrm{an}}\varepsilon_{i-1} ( 4 3 + 1 0 − 5 ) c an ε i < 2 c an ε i − 1 of x ˉ i − 1 = x ˉ i − 1 ( t ˉ i ) \bar{x}_{i-1}=\bar{x}_{i-1}(\bar{t}_i) x ˉ i − 1 = x ˉ i − 1 ( t ˉ i ) , as ∣ y ( i ) ∣ 1 < 3 4 c a n ε i |y^{(i)}|_1<\frac{3}{4}c_{\mathrm{an}}\varepsilon_i ∣ y ( i ) ∣ 1 < 4 3 c an ε i . So ∣ x ˉ i ( t ) − x ˉ i − 1 ( t ) ∣ ∞ ≤ 101 100 ( 3 4 + 10 − 5 ) c a n ε i ≤ c a n ε i |\bar{x}_i(t)-\bar{x}_{i-1}(t)|_\infty\leq\frac{101}{100}(\frac{3}{4}+10^{-5})c_{\mathrm{an}}\varepsilon_i\leq c_{\mathrm{an}}\varepsilon_i ∣ x ˉ i ( t ) − x ˉ i − 1 ( t ) ∣ ∞ ≤ 100 101 ( 4 3 + 1 0 − 5 ) c an ε i ≤ c an ε i , by (i) for X i − 1 \mathcal{X}_{i-1} X i − 1 , and all points below lie within 10 c a n ε i ≤ 10 128 c a n ε i − 1 10c_{\mathrm{an}}\varepsilon_i\leq\frac{10}{128}c_{\mathrm{an}}\varepsilon_{i-1} 10 c an ε i ≤ 128 10 c an ε i − 1 of x ˉ i − 1 ( t ) \bar{x}_{i-1}(t) x ˉ i − 1 ( t ) or x ˉ i − 1 \bar{x}_{i-1} x ˉ i − 1 , in the balls of (i), (ii) and of the Lipschitz bound for X i − 1 \mathcal{X}_{i-1} X i − 1 . Hence X i − 1 ( t ˉ i , ⋅ , t ) \mathcal{X}_{i-1}(\bar{t}_i,\cdot,t) X i − 1 ( t ˉ i , ⋅ , t ) sends the ball ∣ y − x ˉ i ( t ) ∣ ∞ < 8 c a n ε i |y-\bar{x}_i(t)|_\infty<8c_{\mathrm{an}}\varepsilon_i ∣ y − x ˉ i ( t ) ∣ ∞ < 8 c an ε i into the ball of radius 101 100 ⋅ 8 c a n ε i \frac{101}{100}\cdot8c_{\mathrm{an}}\varepsilon_i 100 101 ⋅ 8 c an ε i about the image of x ˉ i ( t ) \bar{x}_i(t) x ˉ i ( t ) , which is x ˉ i − 1 + y ( i ) \bar{x}_{i-1}+y^{(i)} x ˉ i − 1 + y ( i ) moved along e ( k i ) e_{(k_i)} e ( k i ) by a shear map. For y ′ y' y ′ in the image ball, the bound on Im z ⋅ e ( k i ) ⊥ \operatorname{Im}z\cdot e_{(k_i)}^\perp Im z ⋅ e ( k i ) ⊥ of the outer domain gives
∣ Im ( ( y ′ + ε i z ( i ) − x ˉ i − 1 − y ( i ) ) ⋅ e ( k i ) ⊥ ) ∣ < 101 100 ⋅ 8 c a n ε i + ( 3 + 1 2 + 1 64 ) c a n ε i < 12 c a n ε i . \left|\operatorname{Im}\left((y'+\varepsilon_i z^{(i)}-\bar{x}_{i-1}-y^{(i)})\cdot e_{(k_i)}^\perp\right)\right|<\frac{101}{100}\cdot8c_{\mathrm{an}}\varepsilon_i+\left(3+\frac{1}{2}+\frac{1}{64}\right)c_{\mathrm{an}}\varepsilon_i<12c_{\mathrm{an}}\varepsilon_i. Im ( ( y ′ + ε i z ( i ) − x ˉ i − 1 − y ( i ) ) ⋅ e ( k i ) ⊥ ) < 100 101 ⋅ 8 c an ε i + ( 3 + 2 1 + 64 1 ) c an ε i < 12 c an ε i .
the argument of the shear stays in the strip, and X i \mathcal{X}_i X i and x ˉ i ( t ) \bar{x}_i(t) x ˉ i ( t ) are holomorphic. At t ′ = t t'=t t ′ = t the shear map is the identity, and the shear maps over [ t , t ′ ] [t,t'] [ t , t ′ ] and [ t ′ , t ′ ′ ] [t',t''] [ t ′ , t ′′ ] compose to the one over [ t , t ′ ′ ] [t,t''] [ t , t ′′ ] , since the integrals of ζ ~ i , k i \widetilde{\zeta}_{i,k_i} ζ i , k i add and the shear map does not change the coordinate along e ( k i ) ⊥ e_{(k_i)}^\perp e ( k i ) ⊥ ; with (i), (ii) for X i − 1 \mathcal{X}_{i-1} X i − 1 , this gives the identities of (i), (ii). By the chain rule and the group law (ii) for X i − 1 \mathcal{X}_{i-1} X i − 1 , ∇ y X i ( t , y , t ) − I \nabla_y\mathcal{X}_i(t,y,t)-I ∇ y X i ( t , y , t ) − I is the sum of three terms: ∇ y X i − 1 ( t , y , t ) − I \nabla_y\mathcal{X}_{i-1}(t,y,t)-I ∇ y X i − 1 ( t , y , t ) − I , bounded by (i) for X i − 1 \mathcal{X}_{i-1} X i − 1 ; the change of ∇ y X i − 1 ( t ′ , ⋅ , t ˉ i ) \nabla_y\mathcal{X}_{i-1}(t',\cdot,\bar{t}_i) ∇ y X i − 1 ( t ′ , ⋅ , t ˉ i ) over the shear displacement (at most ∣ t ′ − t ∣ 2 π ε i β − 1 exp ( 30 π c a n ) |t'-t|2\pi\varepsilon_i^{\beta-1}\exp(30\pi c_{\mathrm{an}}) ∣ t ′ − t ∣2 π ε i β − 1 exp ( 30 π c an ) ), bounded by the Lipschitz bound above, times ∇ y X i − 1 ( t ˉ i , y , t ) \nabla_y\mathcal{X}_{i-1}(\bar{t}_i,y,t) ∇ y X i − 1 ( t ˉ i , y , t ) , of norm at most 101 100 \frac{101}{100} 100 101 ; and ∫ t t ′ ζ i , k i ( t ′ ′ ) d t ′ ′ \int_t^{t'}\zeta_{i,k_i}(t'')\,dt'' ∫ t t ′ ζ i , k i ( t ′′ ) d t ′′ , at most ∣ t ′ − t ∣ |t'-t| ∣ t ′ − t ∣ , times the derivative of the shear between two matrices of norm at most 101 100 \frac{101}{100} 100 101 . As ε i − 1 β − 2 ≤ 128 β − 2 ε i β − 2 ≤ 1 2 ε i β − 2 \varepsilon_{i-1}^{\beta-2}\leq128^{\beta-2}\varepsilon_i^{\beta-2}\leq\frac{1}{2}\varepsilon_i^{\beta-2} ε i − 1 β − 2 ≤ 12 8 β − 2 ε i β − 2 ≤ 2 1 ε i β − 2 , since 2 − β = 1 − α > 2 3 2-\beta=1-\alpha>\frac{2}{3} 2 − β = 1 − α > 3 2 ,
∣ ∇ y X i ( t ′ , y , t ) − I ∣ ∞ ∣ t ′ − t ∣ ≤ 2 25 ε i − 1 β − 2 + 101 100 2 π ε i β − 1 exp ( 30 π c a n ) 4 c a n ε i − 1 + ( 101 100 ) 2 4 π 2 ε i β − 2 exp ( 30 π c a n ) ≤ ( 2 24 + 10 3 ) ε i β − 2 , \begin{aligned}
\frac{|\nabla_y\mathcal{X}_i(t',y,t)-I|_\infty}{|t'-t|}
\leq2^{25}\varepsilon_{i-1}^{\beta-2}+\frac{101}{100}\frac{2\pi\varepsilon_i^{\beta-1}\exp(30\pi c_{\mathrm{an}})}{4c_{\mathrm{an}}\varepsilon_{i-1}}+\left(\frac{101}{100}\right)^2 4\pi^2\varepsilon_i^{\beta-2}\exp(30\pi c_{\mathrm{an}}) \\
&\leq(2^{24}+10^3)\varepsilon_i^{\beta-2},
\end{aligned} ∣ t ′ − t ∣ ∣ ∇ y X i ( t ′ , y , t ) − I ∣ ∞ ≤ 2 25 ε i − 1 β − 2 + 100 101 4 c an ε i − 1 2 π ε i β − 1 exp ( 30 π c an ) + ( 100 101 ) 2 4 π 2 ε i β − 2 exp ( 30 π c an ) ≤ ( 2 24 + 1 0 3 ) ε i β − 2 ,
which is at most 2 25 ε i β − 2 2^{25}\varepsilon_i^{\beta-2} 2 25 ε i β − 2 . At real parameters the extensions are given by the formulas (11.7). □ \square □
Proof of Lemma 11.7
Proof of Lemma 11.7. Each clause for G ′ G' G ′ , of level m − 1 m-1 m − 1 , is read off its explicit data (Definition 11.3), those of G G G and the invariant of G G G , on [ t A + 3 τ m , t f ] [t_A+3\tau_m,t_f] [ t A + 3 τ m , t f ] , the time window of Proposition 5.3(b). At a time t t t , the forms of G ′ G' G ′ read those of G G G at the pulses k k k of level m m m with H m , k ( t ) > 0 H_{m,k}(t)>0 H m , k ( t ) > 0 (Step 1) at times of the window of k k k up to t t t , the lift being solved forward on it (11.10), and only the flows of the chains, functions of the construction, reach back along the histories; for t ≥ t A + 3 τ m t\ge t_A+3\tau_m t ≥ t A + 3 τ m these times lie in [ t A , t f ] [t_A,t_f] [ t A , t f ] , where G G G satisfies the invariant, as in the proof of Lemma 11.8 (iv). Step 3, the recursion for (b3), is the main estimate. It comes before (b1) because the majorants of G ′ G' G ′ are numbers fixed by Definition 11.3 , so (b3) is a recursion on numbers, while (b1) says that the forms obey them, which needs the analytic calculus of Step 4. Step 5 proves (b1), and (b2), (b2′ ' ′ ), (d) follow. Components of G G G are written v v v , v ′ v' v ′ ; a component v ~ \widetilde{v} v of G ′ G' G ′ over v v v is v v v or a child of v v v .
Step 1: (a1)–(a4). By Definition 11.3 and § 5.8 , the forms of a source of G ′ G' G ′ at a pulse k ′ k' k ′ of level m − 1 m-1 m − 1 are of four kinds: transports (11.8) of cell averages of forms of G G G ; means of the lift along paths of sources of G G G ; these means conjugated by powers of Γ m : ∇ 2 \Gamma_m:\nabla^2 Γ m : ∇ 2 ; and the defect of the averaged flux, with the commutator of the covariance with b m − 1 b_{m-1} b m − 1 , of flux (169) . At a time t t t of the window of k ′ k' k ′ , they read the construction’s functions and the forms of G G G at the pulses k k k of level m m m with H m , k ( t ) > 0 H_{m,k}(t)>0 H m , k ( t ) > 0 , at the chains of k k k extended at ( t , z ) (t,z) ( t , z ) from those of k ′ k' k ′ (end of (11.7), one level lower), whose phase z z z is the fast variable of k ′ k' k ′ ; as the forms of G G G may contain evaluations toward the level m + 1 m+1 m + 1 , these evaluations are nested, which (11.10) allows. The lift of § 5.2 is built from these forms by the cell inverse across the pulse, derivatives and averages in the fast variable of level m m m , products and slow derivatives. So the forms of G ′ G' G ′ are built as in (11.10) from the construction’s functions and the cell solutions of the seed, and no two-dimensional cell problem is solved: this is (a1). They depend on z z z only through the coordinate along e ( k ′ ) ⊥ e^\perp_{(k')} e ( k ′ ) ⊥ of the phase of the shear of k ′ k' k ′ : so do the shear map and the centre of the extended chain, and so do the forms of G G G , since forms of finite analytic norm do not depend on the coordinates of the phases along the pulses (11.9). So the forms of G ′ G' G ′ are transverse: this is (a2). The same argument, with (b1) for G G G , shows that they are periodic in the phases and independent of their coordinates along the pulses, as (11.9) requires.
For (a4), let t t t , in the window of k ′ k' k ′ , be as in (a4) at the level m − 1 m-1 m − 1 . The forms of k ′ k' k ′ at t t t read those of G G G at the pulses k k k of level m m m with H m , k ( t ) > 0 H_{m,k}(t)>0 H m , k ( t ) > 0 on their histories, within 6 5 h m < τ m − 1 40 \frac{6}{5}h_m<\frac{\tau_{m-1}}{40} 5 6 h m < 40 τ m − 1 of t t t by the nesting after (121) at the level m − 1 ≥ m c m-1\ge m_c m − 1 ≥ m c . There the shear of k ′ k' k ′ vanishes, so the extended chains have centre 0 0 0 and their maps on the history of k k k and centre points do not depend on z z z ; by (a7) for G G G , neither do the forms of G G G and the correctors at these chains, nor the forms of k ′ k' k ′ at t t t .
For (a3), let ∣ t − k ′ τ m − 1 ∣ ≤ 3 4 τ m − 1 |t-k'\tau_{m-1}|\le\frac{3}{4}\tau_{m-1} ∣ t − k ′ τ m − 1 ∣ ≤ 4 3 τ m − 1 : the forms of k ′ k' k ′ are transverse by (a2), and the window of another pulse meets this interval only where ∣ t − k ′ ′ τ m − 1 ∣ ≥ 35 48 τ m − 1 |t-k''\tau_{m-1}|\ge\frac{35}{48}\tau_{m-1} ∣ t − k ′′ τ m − 1 ∣ ≥ 48 35 τ m − 1 , and there no pulse is active within τ m − 1 24 \frac{\tau_{m-1}}{24} 24 τ m − 1 of t t t , as ζ m − 1 , k ′ ′ \zeta_{m-1,k''} ζ m − 1 , k ′′ vanishes outside ∣ t − k ′ ′ τ m − 1 ∣ ≤ 2 3 τ m − 1 |t-k''\tau_{m-1}|\le\frac{2}{3}\tau_{m-1} ∣ t − k ′′ τ m − 1 ∣ ≤ 3 2 τ m − 1 (Lemma 7.1 (d)), so that (a4) applies. The commutator term of the defect letter reads no form of G G G : in its flux (169) , ∇ b m − 1 \nabla b_{m-1} ∇ b m − 1 read in the chart of k ′ k' k ′ is the strain of X m − 2 \mathcal{X}_{m-2} X m − 2 plus ∇ y ( ∇ y X m − 2 w ) ( ∇ y X m − 2 ) − 1 \nabla_y(\nabla_y\mathcal{X}_{m-2}w)(\nabla_y\mathcal{X}_{m-2})^{-1} ∇ y ( ∇ y X m − 2 w ) ( ∇ y X m − 2 ) − 1 , with w w w the drift of k ′ k' k ′ , which depends only on z ⋅ e ( k ′ ) ⊥ z\cdot e^\perp_{(k')} z ⋅ e ( k ′ ) ⊥ and vanishes at the times of (a4) and (a6). So this term is built as in (11.10) and is transverse; at those times it is the strain, the transport of the one Eulerian coefficient ∇ b m − 1 \nabla b_{m-1} ∇ b m − 1 , which gives (a4) and (a6); and, a function of the flow, it satisfies (a5) and (a7).
Step 2: (a5)–(a7). Moving the base label by an integer vector commutes with every operation of (11.10): the flows commute with the translations by Z 2 \mathbb{Z}^2 Z 2 (Lemma 7.4 ), the arguments of the shear streams and the centres of the extended chains do not change, the cutoffs do not depend on the label, and the forms of G G G are invariant by (a5) for G G G : this gives (a5). Clauses (a6) and (a7) follow from (a7) for G G G , as a form of G ′ G' G ′ at k ′ k' k ′ is built from forms of G G G over the histories of the pulses of level m m m whose charts are used in the window of k ′ k' k ′ , which lie in the history of k ′ k' k ′ . For (a7), two allowed chains of k ′ k' k ′ with the same map over this history, extended at the same ( t , z ) (t,z) ( t , z ) , have the same centre, which depends only on k ′ k' k ′ , k k k , t t t , z z z , and the same maps over these histories; so the forms of G G G agree there near the chain, by (a7) for G G G , and so do the correctors, solved from zero, and the forms of G ′ G' G ′ . For (a6), the windows of two pulses k ′ k' k ′ , k ′ ′ k'' k ′′ of level m − 1 m-1 m − 1 meet only if their indices differ by 2 2 2 , and their overlap lies at distance at least ( 35 48 − 2 3 ) τ m − 1 = τ m − 1 16 \left(\frac{35}{48}-\frac{2}{3}\right)\tau_{m-1}=\frac{\tau_{m-1}}{16} ( 48 35 − 3 2 ) τ m − 1 = 16 τ m − 1 from the supports of both fast cutoffs, so both shears vanish within τ m − 1 16 \frac{\tau_{m-1}}{16} 16 τ m − 1 of a time t t t of the overlap, and, as in Step 1 and by (a7) for G G G , the extended chains reduce to the charts of k ′ k' k ′ and k ′ ′ k'' k ′′ , related by Lemma 7.7(b); the forms of G ′ G' G ′ at k ′ k' k ′ , k ′ ′ k'' k ′′ are then transports of the same forms of G G G through the two charts, and transport composes exactly, as the chain rule taken one derivative at a time is associative: this is (11.12).
Step 3: the recursion for (b3). Fix one of the two masses. By (b3) for G G G , every row and column sum of the priced majorant matrix (117) of G G G is at most C a d m ε m δ / 4 C_{\mathrm{adm}}\varepsilon_m^{\delta/4} C adm ε m δ /4 , at most 1 2 \frac{1}{2} 2 1 for m > m 0 m>m_0 m > m 0 by item (4) and Lemma 7.1(a). For a path of sources of G G G from v ′ v' v ′ to v v v , the product of the corresponding entries of this matrix is at least ( ω v ′ / ω v ) 1 / 2 ∏ σ ∈ p a t h fac ( σ ) maj ( σ ) (\omega_{v'}/\omega_v)^{1/2}\prod_{\sigma\in\mathrm{path}}\operatorname{fac}(\sigma)\operatorname{maj}(\sigma) ( ω v ′ / ω v ) 1/2 ∏ σ ∈ path fac ( σ ) maj ( σ ) : the mass ratios telescope, and every price is at least 1. Their sums over the paths of a given length into or out of a component are at most the row or column sums of a power of the matrix, so over all nonzero lengths at most 2 C a d m ε m δ / 4 2C_{\mathrm{adm}}\varepsilon_m^{\delta/4} 2 C adm ε m δ /4 .
Old sources. Each word contains a copy of G G G (Definition 11.3(iii)), its sources joining the components of the word over v ′ v' v ′ and v v v , whose masses and markers are those of v ′ v' v ′ and v v v times numbers depending only on the word (Lemma 11.8(i), (119) ); so the masses in the entries are those of G G G . A transported source keeps its factor, birth radius, birth order and width, and its majorant is multiplied by C m a j C_{\mathrm{maj}} C maj ; a second-order mean has birth order 2, hence price 1, factor the product of the factors and majorant C m a j C_{\mathrm{maj}} C maj times the product of the majorants. Counting transports as paths of one source, the copies into or out of a component of a word contribute at most 2 C m a j C a d m ε m δ / 4 2C_{\mathrm{maj}}C_{\mathrm{adm}}\varepsilon_m^{\delta/4} 2 C maj C adm ε m δ /4 per kind.
New sources. A letter labelling a path from v ′ v' v ′ to v v v (v ′ = v v'=v v ′ = v for the defect letter) gives a source from the component of a word over v ′ v' v ′ to the child over v v v of the extended word. By (119) , the square root of its mass ratio is at most C ε m − 1 − η / 4 ( ω v ′ / ω v ) 1 / 2 C\varepsilon_{m-1}^{-\eta/4}(\omega_{v'}/\omega_v)^{1/2} C ε m − 1 − η /4 ( ω v ′ / ω v ) 1/2 , and its marker ratio at most ε m − 1 − η / 8 Ψ v / Ψ v ′ \varepsilon_{m-1}^{-\eta/8}\Psi_v/\Psi_{v'} ε m − 1 − η /8 Ψ v / Ψ v ′ . Its birth radius is ρ ( m ) \rho(m) ρ ( m ) , its width r m = ρ ( m ) ε m − 1 Θ r_m=\rho(m)\varepsilon_{m-1}^{\Theta} r m = ρ ( m ) ε m − 1 Θ by (178) , and its birth order at most N s r c N_{\mathrm{src}} N src , so its price is at most ε m − 1 − ( N s r c − 2 ) Θ \varepsilon_{m-1}^{-(N_{\mathrm{src}}-2)\Theta} ε m − 1 − ( N src − 2 ) Θ . Its entry is thus at most
C C m a j ε m − 1 δ − η − η 4 − η 8 − ( N s r c − 2 ) Θ CC_{\mathrm{maj}}\varepsilon_{m-1}^{\delta-\eta-\frac{\eta}{4}-\frac{\eta}{8}-(N_{\mathrm{src}}-2)\Theta} C C maj ε m − 1 δ − η − 4 η − 8 η − ( N src − 2 ) Θ
times the product along its path. Into or out of a component, the entries of one letter add up to at most twice this, by the bound on path sums.
The recursion. Adding over the at most N w o r d + 1 N_{\mathrm{word}}+1 N word + 1 words and over the letters, every row and column sum of the priced majorant matrix of G ′ G' G ′ is at most
C ε m − 1 δ − 11 8 η − ( N s r c − 2 ) Θ + 4 ( N w o r d + 1 ) ( C m a j + 1 ) C a d m ε m δ / 4 (122) C\varepsilon_{m-1}^{\delta-\frac{11}{8}\eta-(N_{\mathrm{src}}-2)\Theta}
+4(N_{\mathrm{word}}+1)(C_{\mathrm{maj}}+1)C_{\mathrm{adm}}\varepsilon_m^{\delta/4} \tag*{(122)} C ε m − 1 δ − 8 11 η − ( N src − 2 ) Θ + 4 ( N word + 1 ) ( C maj + 1 ) C adm ε m δ /4 (122)
with C C C depending on η \eta η , N w o r d N_{\mathrm{word}} N word , the number of letters and C m a j C_{\mathrm{maj}} C maj , and
δ − 11 8 η − ( N s r c − 2 ) Θ ≥ 45 64 δ > δ 4 , \delta-\frac{11}{8}\eta-(N_{\mathrm{src}}-2)\Theta
\ge\frac{45}{64}\delta>\frac{\delta}{4}, δ − 8 11 η − ( N src − 2 ) Θ ≥ 64 45 δ > 4 δ ,
as η \eta η , ( N s r c − 2 ) Θ ≤ δ / 8 (N_{\mathrm{src}}-2)\Theta\le\delta/8 ( N src − 2 ) Θ ≤ δ /8 by (5) and item (1). In the second term, ε m δ / 4 ≤ 2 ε m − 1 ( q − 1 ) δ / 4 ε m − 1 δ / 4 \varepsilon_m^{\delta/4}\le2\varepsilon_{m-1}^{(q-1)\delta/4}\varepsilon_{m-1}^{\delta/4} ε m δ /4 ≤ 2 ε m − 1 ( q − 1 ) δ /4 ε m − 1 δ /4 by Lemma 7.1(a): the old sums gain one step of the scales. So, for m > m 0 m>m_0 m > m 0 , both terms of (122) are at most 1 2 C a d m ε m − 1 δ / 4 \frac{1}{2}C_{\mathrm{adm}}\varepsilon_{m-1}^{\delta/4} 2 1 C adm ε m − 1 δ /4 , m 0 m_0 m 0 being chosen after C a d m C_{\mathrm{adm}} C adm , C m a j C_{\mathrm{maj}} C maj and the exponents: this is (b3). For the masses ω v / Ψ v 2 \omega_v/\Psi_v^2 ω v / Ψ v 2 , the markers appear only through the ratio ε m − 1 − η / 8 \varepsilon_{m-1}^{-\eta/8} ε m − 1 − η /8 of a first letter, counted above.
Step 4: analyticity. The construction’s functions extend to the outer domain: the flows of the chains and their centre points by Lemma 11.9; the shear streams are entire; the cutoffs, K m τ K_m^{\tau} K m τ and Γ m \Gamma_m Γ m depend on time only; and the Jacobians, their inverses, the conductivity and the strain are built from the flows. The strain is defined on the complex domain, as ∂ t ∇ y X i = ∇ B i ( t , X i ) ∇ y X i \partial_t\nabla_y X_i=\nabla B_i(t,X_i)\nabla_y X_i ∂ t ∇ y X i = ∇ B i ( t , X i ) ∇ y X i is continuous in t t t and holomorphic in y y y (Proposition 7.5, Lemma 7.6), and so are the compositions of (11.7), with C 1 C^1 C 1 time weights ∫ ζ i , k i \int\zeta_{i,k_i} ∫ ζ i , k i ; near x ˉ i ( t ) \bar{x}_i(t) x ˉ i ( t ) , the strain of X i X_i X i is the limit of h − 1 ( ∇ y X i ( t + h , ⋅ ) − I ) h^{-1}(\nabla_y X_i(t+h,\cdot)-I) h − 1 ( ∇ y X i ( t + h , ⋅ ) − I ) as h → 0 h\to0 h → 0 , so its ∣ ⋅ ∣ ∞ |\cdot|_\infty ∣ ⋅ ∣ ∞ -norm is at most 2 25 ε i β − 2 2^{25}\varepsilon_i^{\beta-2} 2 25 ε i β − 2 , by Lemma 11.9. As 2 25 ε m − 1 β − 2 6 5 h m ≤ 1 100 2^{25}\varepsilon_{m-1}^{\beta-2}\frac{6}{5}h_m\le\frac{1}{100} 2 25 ε m − 1 β − 2 5 6 h m ≤ 100 1 by (121) , on the outer domain and in the window of k k k the Jacobian of X m − 1 ( t , ⋅ , t ˉ m ) \mathcal{X}_{m-1}(t,\cdot,\bar{t}_m) X m − 1 ( t , ⋅ , t ˉ m ) is within 1 100 \frac{1}{100} 100 1 of I I I , by Lemma 11.9(i), and its inverse within 1 99 \frac{1}{99} 99 1 . The conductivity of the chain is ( ∇ y X m − 1 ) − 1 ( ∇ y X m − 1 ) − T (\nabla_y\mathcal{X}_{m-1})^{-1}(\nabla_y\mathcal{X}_{m-1})^{-T} ( ∇ y X m − 1 ) − 1 ( ∇ y X m − 1 ) − T , as in (53) , with the transpose and no complex conjugation, so that it is holomorphic; its entry across the pulse, the only one the cell inverse uses, is the sum of the squares of the entries of the row of the inverse along e ( k ) e_{(k)} e ( k ) , within 1 99 \frac{1}{99} 99 1 of e ( k ) e_{(k)} e ( k ) in ∣ ⋅ ∣ 1 \lvert\cdot\rvert_1 ∣ ⋅ ∣ 1 , so it is within 2 99 + 1 99 2 ≤ 1 10 \frac{2}{99}+\frac{1}{99^2}\leq\frac{1}{10} 99 2 + 9 9 2 1 ≤ 10 1 of 1 and has real part at least 9 10 \frac{9}{10} 10 9 .
Let v v v be a component of G G G and v ~ \tilde{v} v ~ one of G ′ G' G ′ over v v v , so that res ( v ~ ) ≥ res ( v ) − 1 / ( N g e n + 2 + N o f f ) \operatorname{res}(\tilde{v})\geq\operatorname{res}(v)-1/(N_{\mathrm{gen}}+2+N_{\mathrm{off}}) res ( v ~ ) ≥ res ( v ) − 1/ ( N gen + 2 + N off ) . As sources never go to a smaller generation (Subsection 11.1), the forms of the sources on a path into v v v are bounded on D v D_v D v by their analytic norms. We split the step 1 / ( N g e n + 2 + N o f f ) 1/(N_{\mathrm{gen}}+2+N_{\mathrm{off}}) 1/ ( N gen + 2 + N off ) of the reserve into N s r c N_{\mathrm{src}} N src equal parts, one for each slow derivative a cycle takes; D \mathcal{D} D and ∥ ⋅ ∥ a n \|\cdot\|_{\mathrm{an}} ∥ ⋅ ∥ an are those of (11.9) at the level of the form, with a number r r r in place of the reserve where written. The extension at ( t , z ) (t,z) ( t , z ) of a chain of D v ~ D_{\tilde{v}} D v ~ at the level m − 1 m-1 m − 1 , with ∣ Im z ⋅ e ( k ) ⊥ ∣ < ( 3 + 1 64 ) c a n \lvert\operatorname{Im}z\cdot e_{(k)}^{\perp}\rvert<(3+\frac{1}{64})c_{\mathrm{an}} ∣ Im z ⋅ e ( k ) ⊥ ∣ < ( 3 + 64 1 ) c an , keeps the lower parameters and has a centre of ∣ ⋅ ∣ 1 \lvert\cdot\rvert_1 ∣ ⋅ ∣ 1 -norm at most 4 ⋅ 10 − 6 c a n ε m − 1 4\cdot10^{-6}c_{\mathrm{an}}\varepsilon_{m-1} 4 ⋅ 1 0 − 6 c an ε m − 1 , below all radii of centres, so it lies in D v ~ D_{\tilde{v}} D v ~ at the level m m m , hence in D v D_v D v .
A slow derivative costs one part. Let A \mathcal{A} A have finite analytic norm at the reserve res ( v ) \operatorname{res}(v) res ( v ) , and let res ( v ~ ) ≤ res ( v ) − 1 / ( N s r c ( N g e n + 2 + N o f f ) ) \operatorname{res}(\tilde{v})\leq\operatorname{res}(v)-1/(N_{\mathrm{src}}(N_{\mathrm{gen}}+2+N_{\mathrm{off}})) res ( v ~ ) ≤ res ( v ) − 1/ ( N src ( N gen + 2 + N off )) . For a chain C C C in D v ~ D_{\tilde{v}} D v ~ , the recentred chains C [ y ] C^{[y]} C [ y ] with ∣ y ∣ ∞ < c a n ε m − 1 / ( 9 N s r c ( N g e n + 2 + N o f f ) ) \lvert y\rvert_\infty<c_{\mathrm{an}}\varepsilon_{m-1}/(9N_{\mathrm{src}}(N_{\mathrm{gen}}+2+N_{\mathrm{off}})) ∣ y ∣ ∞ < c an ε m − 1 / ( 9 N src ( N gen + 2 + N off )) lie in D v D_v D v . Indeed, recentring moves the base label by y y y if m ′ = m m'=m m ′ = m ; otherwise, by the group law of Lemma 11.9(ii), it moves the last centre by w : = X m − 2 ( t ˉ m − 1 , x ˉ m − 1 + y , t ˉ m ) − X m − 2 ( t ˉ m − 1 , x ˉ m − 1 , t ˉ m ) w:=\mathcal{X}_{m-2}(\bar{t}_{m-1},\bar{x}_{m-1}+y,\bar{t}_m)-\mathcal{X}_{m-2}(\bar{t}_{m-1},\bar{x}_{m-1},\bar{t}_m) w := X m − 2 ( t ˉ m − 1 , x ˉ m − 1 + y , t ˉ m ) − X m − 2 ( t ˉ m − 1 , x ˉ m − 1 , t ˉ m ) and the last phase by w / ε m − 1 w/\varepsilon_{m-1} w / ε m − 1 , with ∣ w ∣ ∞ ≤ ( 1 + a ) ∣ y ∣ ∞ \lvert w\rvert_\infty\leq(1+a)\lvert y\rvert_\infty ∣ w ∣ ∞ ≤ ( 1 + a ) ∣ y ∣ ∞ , a : = 2 25 ⋅ 3 5 ⋅ 10 − 6 c a n < 2 ⋅ 10 − 3 a:=2^{25}\cdot\frac{3}{5}\cdot10^{-6}c_{\mathrm{an}}<2\cdot10^{-3} a := 2 25 ⋅ 5 3 ⋅ 1 0 − 6 c an < 2 ⋅ 1 0 − 3 , by Lemma 11.9(i) and (11.19) at the level m − 2 m-2 m − 2 , as ∣ t ˉ m − t ˉ m − 1 ∣ ≤ 3 5 h m − 1 \lvert\bar{t}_m-\bar{t}_{m-1}\rvert\leq\frac{3}{5}h_{m-1} ∣ t ˉ m − t ˉ m − 1 ∣ ≤ 5 3 h m − 1 by the nesting after (11.19) and the segment from x ˉ m − 1 \bar{x}_{m-1} x ˉ m − 1 to x ˉ m − 1 + y \bar{x}_{m-1}+y x ˉ m − 1 + y lies within c a n ε m − 1 c_{\mathrm{an}}\varepsilon_{m-1} c an ε m − 1 of x ˉ m − 2 ( t ˉ m ) \bar{x}_{m-2}(\bar{t}_m) x ˉ m − 2 ( t ˉ m ) (proof of Lemma 11.9). By (11.9), the radii of D v D_v D v exceed those of D v ~ D_{\tilde{v}} D v ~ by c a n ε m − 1 c_{\mathrm{an}}\varepsilon_{m-1} c an ε m − 1 , c a n c_{\mathrm{an}} c an and ε m − 1 / 100 \varepsilon_{m-1}/100 ε m − 1 /100 times 1 / ( 4 N s r c ( N g e n + 2 + N o f f ) ) 1/(4N_{\mathrm{src}}(N_{\mathrm{gen}}+2+N_{\mathrm{off}})) 1/ ( 4 N src ( N gen + 2 + N off )) , and 2 ( 1 + a ) / 9 < 1 4 2(1+a)/9<\frac{1}{4} 2 ( 1 + a ) /9 < 4 1 . Cauchy’s estimate in y 1 y_1 y 1 , mode by mode, gives
ε m − 1 ∂ y 1 A ^ ( t , C [ y ] ; ξ ) ∣ y = 0 ≤ 9 N s r c ( N g e n + 2 + N o f f ) c a n sup D v ∣ A ^ ( t , ⋅ ; ξ ) ∣ , \varepsilon_{m-1}\left.\partial_{y_1}\widehat{\mathcal{A}}(t,C^{[y]};\xi)\right|_{y=0}\leq\frac{9N_{\mathrm{src}}(N_{\mathrm{gen}}+2+N_{\mathrm{off}})}{c_{\mathrm{an}}}\sup_{D_v}\left|\widehat{\mathcal{A}}(t,\cdot;\xi)\right|, ε m − 1 ∂ y 1 A ( t , C [ y ] ; ξ ) y = 0 ≤ c an 9 N src ( N gen + 2 + N off ) D v sup A ( t , ⋅ ; ξ ) ,
and likewise in y 2 y_2 y 2 , so that ε m − 1 ∥ ∇ y A ∥ a n , v ~ ≤ 18 N s r c ( N g e n + 2 + N o f f ) c a n − 1 ∥ A ∥ a n , v \varepsilon_{m-1}\|\nabla_y\mathcal{A}\|_{\mathrm{an},\tilde{v}}\leq18N_{\mathrm{src}}(N_{\mathrm{gen}}+2+N_{\mathrm{off}})c_{\mathrm{an}}^{-1}\|\mathcal{A}\|_{\mathrm{an},v} ε m − 1 ∥ ∇ y A ∥ an , v ~ ≤ 18 N src ( N gen + 2 + N off ) c an − 1 ∥ A ∥ an , v .
A transverse primitive costs nothing. On a transverse form with zero average in z z z , it multiplies the mode ξ ≠ 0 \xi\neq0 ξ = 0 of z ⋅ e ( k ) ⊥ z\cdot e_{(k)}^{\perp} z ⋅ e ( k ) ⊥ by ( 2 π i ξ ) − 1 (2\pi i\xi)^{-1} ( 2 π i ξ ) − 1 , of modulus at most 1 2 π \frac{1}{2\pi} 2 π 1 , and keeps the mode 0 at zero. The other operations of (1.10) cost nothing , a constant or one part, by the same two mechanisms. Products cost a constant, the Fourier weights exp ( 3 π c a n ∣ ξ ∣ ) ( 1 + ∣ ξ ∣ ) N a n \exp(3\pi c_{\mathrm{an}}\lvert\xi\rvert)(1+\lvert\xi\rvert)^{N_{\mathrm{an}}} exp ( 3 π c an ∣ ξ ∣) ( 1 + ∣ ξ ∣ ) N an of the analytic norm being submultiplicative in ξ \xi ξ ; the average in z z z keeps the mode 0; the cell inverse of § 5.2 maps the mode ξ \xi ξ of a flux written ( κ m / ε m 2 ) e ( k ) ⊥ ⋅ ∇ z F (\kappa_m/\varepsilon_m^2)e_{(k)}^{\perp}\cdot\nabla_zF ( κ m / ε m 2 ) e ( k ) ⊥ ⋅ ∇ z F to that of the gradient of the solution by a time integral of $4\pi^2\xi^2\kappa_m/\varepsilon_m^2$ times exponentials whose exponents have real parts at most − 3 π 2 ξ 2 κ m / ε m 2 -3\pi^2\xi^2\kappa_m/\varepsilon_m^2 − 3 π 2 ξ 2 κ m / ε m 2 times the elapsed time, as the real part of the entry across the pulse is at least 9 10 ≥ 3 4 \frac{9}{10}\geq\frac{3}{4} 10 9 ≥ 4 3 , so with a constant at most 4 3 \frac{4}{3} 3 4 ; on a scalar source it costs the power ε m 2 / ( 3 π 2 ξ 2 κ m ) \varepsilon_m^2/(3\pi^2\xi^2\kappa_m) ε m 2 / ( 3 π 2 ξ 2 κ m ) of Lemma 10.4 , which the proof of Proposition 10.7 pairs with the coefficients of (10.16). A derivative in the fast variable of level m m m multiplies its mode by 2 π i ξ 2\pi i\xi 2 π i ξ , which the polynomial weight absorbs: a cycle takes at most 2 + N p a t h N s r c + 2 N c o r + 2 N s r c 2+N_{\mathrm{path}}N_{\mathrm{src}}+2N_{\mathrm{cor}}+2N_{\mathrm{src}} 2 + N path N src + 2 N cor + 2 N src of them, the order in z z z of (10.17) at N = 0 N=0 N = 0 , and N a n ≥ N c o e f ≥ N r e t N_{\mathrm{an}}\geq N_{\mathrm{coef}}\geq N_{\mathrm{ret}} N an ≥ N coef ≥ N ret (items (1) and (3)) exceeds this number by (13.6). A derivative in the analytic costs one part. After the averages in the fast variable of level m m m , the Fourier weights of the analytic norm of G ′ G' G ′ are restored: its fast variable is the phase of level m − 1 m-1 m − 1 of the chains of G G G , in which the forms of G G G are holomorphic on ∣ Im z ( m − 1 ) ⋅ e ( k ′ ) ⊥ ∣ < ( 3 + 1 64 ) c a n \lvert\operatorname{Im}z^{(m-1)}\cdot e_{(k')}^{\perp}\rvert<(3+\frac{1}{64})c_{\mathrm{an}} ∣ Im z ( m − 1 ) ⋅ e ( k ′ ) ⊥ ∣ < ( 3 + 64 1 ) c an , so their modes in it decay faster than exp ( − 6 π c a n ∣ ξ ∣ ) \exp(-6\pi c_{\mathrm{an}}\lvert\xi\rvert) exp ( − 6 π c an ∣ ξ ∣) . Transports and second-order means use only operations that cost nothing or a constant. A letter takes at most N s r c − 2 N_{\mathrm{src}}-2 N src − 2 slow derivatives and goes to a child, so its forms have analytic norms at res ( v ~ ) \operatorname{res}(\tilde{v}) res ( v ~ ) bounded by constants times those of forms of G G G at res ( v ) \operatorname{res}(v) res ( v ) . With P : = N s r c ( N g e n + 2 + N o f f ) P:=N_{\mathrm{src}}(N_{\mathrm{gen}}+2+N_{\mathrm{off}}) P := N src ( N gen + 2 + N off ) and D ( r ) \mathcal{D}(r) D ( r ) the set D \mathcal{D} D of the level m m m with r r r in place of the reserve, these costs give, for a path σ 1 ⋯ σ i \sigma_1\cdots\sigma_i σ 1 ⋯ σ i of sources of $G into v v v , r = res ( v ) r=\operatorname{res}(v) r = res ( v ) and 2 ≤ d ≤ j ≤ N c o r + N s r c 2\le d\le j\le N_{\mathrm{cor}}+N_{\mathrm{src}} 2 ≤ d ≤ j ≤ N cor + N src with j − d ≤ N s r c − 2 j-d\le N_{\mathrm{src}}-2 j − d ≤ N src − 2 ,
sup D ( r − ( j − d ) / P ) ∣ M j , d [ σ 1 ⋯ σ i ] ∣ ≤ C fac ( σ 1 ) maj ( σ 1 ) ⋯ fac ( σ i ) maj ( σ i ) κ m − 1 ε ˉ m j − 2 ε m − 1 d − j . \sup_{\mathcal{D}(r-(j-d)/P)}\left|\mathcal{M}_{j,d}[\sigma_1\cdots\sigma_i]\right|\le C\operatorname{fac}(\sigma_1)\operatorname{maj}(\sigma_1)\cdots\operatorname{fac}(\sigma_i)\operatorname{maj}(\sigma_i)\kappa_{m-1}\bar{\varepsilon}_m^{j-2}\varepsilon_{m-1}^{d-j}. D ( r − ( j − d ) / P ) sup ∣ M j , d [ σ 1 ⋯ σ i ] ∣ ≤ C fac ( σ 1 ) maj ( σ 1 ) ⋯ fac ( σ i ) maj ( σ i ) κ m − 1 ε ˉ m j − 2 ε m − 1 d − j .
This is the proof of Proposition 10.7(c), with Lemma 10.4 in its complex form (it uses only Re e ( k ) ⊥ ⋅ K e ( k ) ⊥ ≥ 3 4 \operatorname{Re}e_{(k)}^{\perp}\cdot K e_{(k)}^{\perp}\ge\frac{3}{4} Re e ( k ) ⊥ ⋅ K e ( k ) ⊥ ≥ 4 3 ), the costs above for derivatives, and the monotone generations along the path; Proposition 10.6(c) passes in the same way to complex conductivities, as ∣ e − a − e − b ∣ ≤ ∣ a − b ∣ max { ∣ e − a ∣ , ∣ e − b ∣ } \lvert e^{-a}-e^{-b}\rvert\le\lvert a-b\rvert\max\{\lvert e^{-a}\rvert,\lvert e^{-b}\rvert\} ∣ e − a − e − b ∣ ≤ ∣ a − b ∣ max {∣ e − a ∣ , ∣ e − b ∣} for complex a a a , b b b . Their hypotheses (108) hold on every window by (b1) for G G G , as in Step 6.
Step 5: (b1). For a source σ ′ \sigma' σ ′ of G ′ G' G ′ into v ~ \widetilde v v we check, kind by kind, that the forms satisfy ( ε m − 1 / ρ σ ′ ) d − 2 ∥ A σ ′ , d ∥ a n , v ~ ≤ maj ( σ ′ ) (\varepsilon_{m-1}/\rho_{\sigma'})^{d-2}\lVert A_{\sigma',d}\rVert_{\mathrm{an},\widetilde v}\le\operatorname{maj}(\sigma') ( ε m − 1 / ρ σ ′ ) d − 2 ∥ A σ ′ , d ∥ an , v ≤ maj ( σ ′ ) ; the constants produced by Step 4 are at most C m a j C_{\mathrm{maj}} C maj , which is large given item (3). Each form of G ′ G' G ′ at k ′ k' k ′ is, at a time t t t , a sum over the at most two pulses k k k of level m m m with H m , k ( t ) > 0 H_{m,k}(t)>0 H m , k ( t ) > 0 of the transport of a form at k k k by the maps of Lemma 11.9, with the Piola factor of (114) ; this costs a constant (Lemma 11.9, Cauchy’s estimate on its balls), the coefficient of order d ′ d' d ′ from the order d d d carrying ε m − 1 d ′ − d \varepsilon_{m-1}^{d'-d} ε m − 1 d ′ − d , so we bound the forms at k k k . The birth radius of a source of G G G of birth order at least 3 is at most ε m \varepsilon_m ε m : for one of the seed by Definition 11.2(i), and for one created at a level i > m i>m i > m since ρ ( i ) ≤ C ε i − 1 1 + γ / 2 + δ ≤ ε i − 1 \rho(i)\le C\varepsilon_{i-1}^{1+\gamma/2+\delta}\le\varepsilon_{i-1} ρ ( i ) ≤ C ε i − 1 1 + γ /2 + δ ≤ ε i − 1 by (184) , as i > m 0 i>m_0 i > m 0 .
Transports. Let σ \sigma σ be a source of G G G . Its coefficients (116) of orders d ≥ 3 d\ge3 d ≥ 3 , averaged in z z z , are transported to the chart of k ′ k' k ′ ; the order d d d gives the orders d ′ ≤ d d'\le d d ′ ≤ d , with the d − d ′ d-d' d − d ′ derivatives of the change of variables at the scale ε m − 1 \varepsilon_{m-1} ε m − 1 (Lemma 11.9). Writing the result of order d ′ d' d ′ as κ m − 1 ε m − 1 d ′ − 2 fac ( σ ) A σ ′ , d ′ \kappa_{m-1}\varepsilon_{m-1}^{d'-2}\operatorname{fac}(\sigma)A_{\sigma',d'} κ m − 1 ε m − 1 d ′ − 2 fac ( σ ) A σ ′ , d ′ , we obtain, by (b1) for G G G , κ m ≤ κ m − 1 \kappa_m\le\kappa_{m-1} κ m ≤ κ m − 1 (Proposition 2.1(c)) and ρ σ ≤ ε m \rho_\sigma\le\varepsilon_m ρ σ ≤ ε m ,
( ε m − 1 ρ σ ) d ′ − 2 ∥ A σ ′ , d ′ ∥ a n , v ~ ≤ C κ m κ m − 1 ∑ d = max { 3 , d ′ } d σ ( ρ σ ε m − 1 ) d − d ′ ( ε m ρ σ ) d − 2 ∥ A σ , d ∥ a n , v ≤ C maj ( σ ) . \left(\frac{\varepsilon_{m-1}}{\rho_\sigma}\right)^{d'-2}\lVert A_{\sigma',d'}\rVert_{\mathrm{an},\widetilde v}\le C\frac{\kappa_m}{\kappa_{m-1}}\sum_{d=\max\{3,d'\}}^{d_\sigma}\left(\frac{\rho_\sigma}{\varepsilon_{m-1}}\right)^{d-d'}\left(\frac{\varepsilon_m}{\rho_\sigma}\right)^{d-2}\lVert A_{\sigma,d}\rVert_{\mathrm{an},v}\le C\operatorname{maj}(\sigma). ( ρ σ ε m − 1 ) d ′ − 2 ∥ A σ ′ , d ′ ∥ an , v ≤ C κ m − 1 κ m d = m a x { 3 , d ′ } ∑ d σ ( ε m − 1 ρ σ ) d − d ′ ( ρ σ ε m ) d − 2 ∥ A σ , d ∥ an , v ≤ C maj ( σ ) .
Second-order means are at most C C C times the product of the fac ( σ ) maj ( σ ) \operatorname{fac}(\sigma)\operatorname{maj}(\sigma) fac ( σ ) maj ( σ ) along their path (Section 10), in the analytic norm with no loss of reserve, as averages of products of forms and correctors (Step 4).
Mean letters. Let σ ′ \sigma' σ ′ be the source of a mean letter of birth order d σ ′ ≥ 3 d_{\sigma'}\ge3 d σ ′ ≥ 3 along a path of G G G into v v v . By Proposition 10.7(c), the coefficient of order d ≤ d σ ′ d\le d_{\sigma'} d ≤ d σ ′ of the mean of degree d σ ′ d_{\sigma'} d σ ′ along the path is κ m − 1 ε ˉ m d σ ′ − 2 ε m − 1 d − d σ ′ \kappa_{m-1}\bar{\varepsilon}_m^{d_{\sigma'}-2}\varepsilon_{m-1}^{d-d_{\sigma'}} κ m − 1 ε ˉ m d σ ′ − 2 ε m − 1 d − d σ ′ times a form bounded by C ∏ σ ∈ path fac ( σ ) maj ( σ ) C\prod_{\sigma\in\operatorname{path}}\operatorname{fac}(\sigma)\operatorname{maj}(\sigma) C ∏ σ ∈ path fac ( σ ) maj ( σ ) . Its recursion takes d σ ′ − d d_{\sigma'}-d d σ ′ − d slow derivatives and otherwise the operations of Step 4, so the bound holds in the analytic norm at the reserve of v ~ \widetilde v v . By Definition 11.3(iv), A σ ′ , d A_{\sigma',d} A σ ′ , d is ε m − 1 d − 2 \varepsilon_{m-1}^{d-2} ε m − 1 d − 2 times this coefficient divided by κ m − 1 ε m − 1 d σ ′ − 2 ∏ σ ∈ path fac ( σ ) \kappa_{m-1}\varepsilon_{m-1}^{d_{\sigma'}-2}\prod_{\sigma\in\operatorname{path}}\operatorname{fac}(\sigma) κ m − 1 ε m − 1 d σ ′ − 2 ∏ σ ∈ path fac ( σ ) , and ρ σ ′ = ρ ( m ) ≤ ε m − 1 \rho_{\sigma'}=\rho(m)\le\varepsilon_{m-1} ρ σ ′ = ρ ( m ) ≤ ε m − 1 , by (184) as m > m 0 m>m_0 m > m 0 , so
( ε m − 1 ρ σ ′ ) d − 2 ∥ A σ ′ , d ∥ a n , v ~ ≤ C ε m − 1 − η ( ε ˉ m ρ ( m ) ) d σ ′ − 2 ∏ σ ∈ path maj ( σ ) ≤ C ε m − 1 δ − η ∏ σ ∈ path maj ( σ ) . \left(\frac{\varepsilon_{m-1}}{\rho_{\sigma'}}\right)^{d-2}\lVert A_{\sigma',d}\rVert_{\mathrm{an},\widetilde v}\le C\varepsilon_{m-1}^{-\eta}\left(\frac{\bar{\varepsilon}_m}{\rho(m)}\right)^{d_{\sigma'}-2}\prod_{\sigma\in\operatorname{path}}\operatorname{maj}(\sigma)\le C\varepsilon_{m-1}^{\delta-\eta}\prod_{\sigma\in\operatorname{path}}\operatorname{maj}(\sigma). ( ρ σ ′ ε m − 1 ) d − 2 ∥ A σ ′ , d ∥ an , v ≤ C ε m − 1 − η ( ρ ( m ) ε ˉ m ) d σ ′ − 2 σ ∈ path ∏ maj ( σ ) ≤ C ε m − 1 δ − η σ ∈ path ∏ maj ( σ ) .
The last step is the smallness of the correctors: by (96) , ε ˉ m ≤ C ε m 3 δ ρ ( m ) \bar{\varepsilon}_m\le C\varepsilon_m^{3\delta}\rho(m) ε ˉ m ≤ C ε m 3 δ ρ ( m ) , so the ratio in parentheses is at most C ε m − 1 3 δ C\varepsilon_{m-1}^{3\delta} C ε m − 1 3 δ , and 3 δ ( d σ ′ − 2 ) ≥ δ 3\delta(d_{\sigma'}-2)\ge\delta 3 δ ( d σ ′ − 2 ) ≥ δ . For a path of one source σ \sigma σ , the letter subtracts its cell average, at most C ε m − 1 − η ( κ m / κ m − 1 ) maj ( σ ) C\varepsilon_{m-1}^{-\eta}(\kappa_m/\kappa_{m-1})\operatorname{maj}(\sigma) C ε m − 1 − η ( κ m / κ m − 1 ) maj ( σ ) by the transport bound and ρ σ ≤ ε m ≤ ρ σ ′ \rho_\sigma\le\varepsilon_m\le\rho_{\sigma'} ρ σ ≤ ε m ≤ ρ σ ′ ; and κ m / κ m − 1 ≤ C ε m − 1 2 γ ≤ C ε m − 1 δ \kappa_m/\kappa_{m-1}\le C\varepsilon_{m-1}^{2\gamma}\le C\varepsilon_{m-1}^{\delta} κ m / κ m − 1 ≤ C ε m − 1 2 γ ≤ C ε m − 1 δ , by Proposition 2.1(c) (m ≤ m s < M m\le m_s<M m ≤ m s < M ) and 3 δ < γ 3\delta<\gamma 3 δ < γ .
Covariance letters (j + j ′ ≥ 1 j+j'\ge1 j + j ′ ≥ 1 ) are identical, with the mean conjugated: each operator Γ m : ∇ 2 \Gamma_m:\nabla^2 Γ m : ∇ 2 adds two orders and a coefficient ∣ Γ m ( t ) ∣ ≤ C ε m − 1 δ ρ ( m ) 2 \lvert\Gamma_m(t)\rvert\le C\varepsilon_{m-1}^{\delta}\rho(m)^2 ∣ Γ m ( t )∣ ≤ C ε m − 1 δ ρ ( m ) 2 (Lemma 10.3(b)), so the gain ( ∣ Γ m ∣ / ρ σ ′ ) j + j ′ ≤ C ε m − 1 δ (\lvert\Gamma_m\rvert/\rho_{\sigma'})^{j+j'}\le C\varepsilon_{m-1}^{\delta} (∣ Γ m ∣ / ρ σ ′ ) j + j ′ ≤ C ε m − 1 δ replaces that of the correctors, and derivatives on a coefficient cost ρ σ ′ / ε m − 1 ≤ 1 \rho_{\sigma'}/\varepsilon_{m-1}\le1 ρ σ ′ / ε m − 1 ≤ 1 each.
The defect letter has no path. Its forms are ε ˉ m − 1 − η \bar{\varepsilon}_{m-1}^{-\eta} ε ˉ m − 1 − η times the defect of the averaged flux at level m m m (Section 13), normalized by κ m − 1 ε m − 1 d − 2 \kappa_{m-1}\varepsilon_{m-1}^{d-2} κ m − 1 ε m − 1 d − 2 : the diagonal means of the lift less J m κ m ∇ J_m^{\kappa_m}\nabla J m κ m ∇ , and the commutator of the covariance with the coarse velocity, conjugated by the truncated exponentials. Each part is small by a separation of scales. With the conductivity I I I , the second-order diagonal mean is exactly J m κ m J^{\kappa_m}_{m} J m κ m (Proposition 10.6 (c)); in a chain the conductivity differs from I I I by at most C ε m − 1 β − 2 h m ≤ C ε m − 1 2 δ C\varepsilon_{m-1}^{\beta-2}h_m \le C\varepsilon_{m-1}^{2\delta} C ε m − 1 β − 2 h m ≤ C ε m − 1 2 δ by Step 4 and (15) , which changes the mean by at most C κ m − 1 ε m − 1 2 δ C\kappa_{m-1}\varepsilon_{m-1}^{2\delta} C κ m − 1 ε m − 1 2 δ (Proposition 10.6 (c)). The Piola factor ∇ y X m − 1 \nabla_yX_{m-1} ∇ y X m − 1 of (114) , on the left and, through the gradient of Y ~ k \widetilde{Y}_k Y k , transposed on the right, changes it by at most C ∣ ∇ y X m − 1 − I ∣ ∣ J m κ m ∣ ≤ C κ m − 1 ε m − 1 2 δ C\lvert\nabla_yX_{m-1}-I\rvert\lvert J^{\kappa_m}_{m}\rvert\le C\kappa_{m-1}\varepsilon_{m-1}^{2\delta} C ∣ ∇ y X m − 1 − I ∣ ∣ J m κ m ∣ ≤ C κ m − 1 ε m − 1 2 δ , by Lemma 7.10(i) and (96) . The diagonal means of degrees 3 3 3 to N c o r + 1 N_{\mathrm{cor}}+1 N cor + 1 gain C ε m − 1 3 δ C\varepsilon_{m-1}^{3\delta} C ε m − 1 3 δ , as above. The commutator is at most ∣ Γ m ( t ) ∣ \lvert\Gamma_m(t)\rvert ∣ Γ m ( t )∣ times the strain of the chain, which is ∇ b m − 1 \nabla b_{m-1} ∇ b m − 1 at a real chain with matched phases and has ∣ ⋅ ∣ ∞ \lvert\cdot\rvert_{\infty} ∣ ⋅ ∣ ∞ -norm at most 2 25 ε m − 1 β − 2 2^{25}\varepsilon_{m-1}^{\beta-2} 2 25 ε m − 1 β − 2 on the outer domain (Step 4), its derivatives costing C ε m − 1 − 1 C\varepsilon_{m-1}^{-1} C ε m − 1 − 1 each (Cauchy’s estimate). So the commutator is at most C ε m − 1 β − 2 κ m − 1 τ m ′ ≤ C κ m − 1 ε m − 1 2 δ C\varepsilon_{m-1}^{\beta-2}\kappa_{m-1}\tau'_m \le C\kappa_{m-1}\varepsilon_{m-1}^{2\delta} C ε m − 1 β − 2 κ m − 1 τ m ′ ≤ C κ m − 1 ε m − 1 2 δ , by Lemma 10.3(b) and (15) , and the truncated exponentials cost at most ( 1 + C ∣ Γ m ∣ / ρ ( m ) 2 ) 2 p ≤ C (1+C\lvert\Gamma_m\rvert/\rho(m)^2)^{2p}\le C ( 1 + C ∣ Γ m ∣ / ρ ( m ) 2 ) 2 p ≤ C , derivatives on a coefficient costing at most 1 each. So the coefficient of order d d d of the defect is at most C ε m − 1 δ κ m − 1 ρ ( m ) d − 2 C\varepsilon_{m-1}^{\delta}\kappa_{m-1}\rho(m)^{d-2} C ε m − 1 δ κ m − 1 ρ ( m ) d − 2 in the analytic norm, and with the birth radius ρ ( m ) \rho(m) ρ ( m ) this is (b1) with the majorant C m a j ε m − 1 δ − η C_{\mathrm{maj}}\varepsilon_{m-1}^{\delta-\eta} C maj ε m − 1 δ − η .
*Step 6: (b2) and (b2′), in their precise forms (11.14) and (11.15) at the level m − 1 m-1 m − 1 . Let C C C be an allowed chain with all ∣ y ( i ) ∣ 1 < 1 4 C c a n ε i \lvert y^{(i)}\rvert_1 < \frac{1}{4}C_{\mathrm{can}}\varepsilon_i ∣ y ( i ) ∣ 1 < 4 1 C can ε i . Every component has reserve at least 2 / ( N g e n + 2 + N o f f ) 2/(N_{\mathrm{gen}}+2+N_{\mathrm{off}}) 2/ ( N gen + 2 + N off ) , so D v ~ \mathcal{D}_{\widetilde{v}} D v contains a complex neighbourhood of C C C and of its recentrings by ∣ y ∣ ∞ \lvert y\rvert_\infty ∣ y ∣ ∞ at most a fixed multiple of ε m − 2 \varepsilon_{m-2} ε m − 2 . Cauchy’s estimate in y y y , and the weight exp ( 3 π C c a n ∣ ξ ∣ ) \exp(3\pi C_{\mathrm{can}}\lvert\xi\rvert) exp ( 3 π C can ∣ ξ ∣) in z z z , give, for all j j j and j ′ j' j ′ , with C C C depending also on j j j and j ′ j' j ′ , by (b1) and ρ σ ′ ≤ ε m − 1 \rho_{\sigma'}\le\varepsilon_{m-1} ρ σ ′ ≤ ε m − 1 when d ≥ 3 d\ge3 d ≥ 3 ,
ε m − 2 j + j ′ ∥ ∇ y j ∇ y ′ j ′ A σ ′ , d ∥ L m − 1 ∞ ( [ t A + 3 τ m , t f ] ) ≤ C ∥ A σ ′ , d ∥ a n , v ~ ≤ C ( ρ σ ′ ε m − 1 ) d − 2 C m a j ( σ ′ ) ≤ C C m a j ( σ ′ ) . \varepsilon_{m-2}^{j+j'}\left\lVert\nabla_y^j\nabla_{y'}^{j'}A_{\sigma',d}\right\rVert_{L^\infty_{m-1}([t_A+3\tau_m,t_f])}\le C\left\lVert A_{\sigma',d}\right\rVert_{\mathrm{an},\widetilde{v}}\le C\left(\frac{\rho_{\sigma'}}{\varepsilon_{m-1}}\right)^{d-2}C_{\mathrm{maj}}(\sigma')\le C C_{\mathrm{maj}}(\sigma'). ε m − 2 j + j ′ ∇ y j ∇ y ′ j ′ A σ ′ , d L m − 1 ∞ ([ t A + 3 τ m , t f ]) ≤ C ∥ A σ ′ , d ∥ an , v ≤ C ( ε m − 1 ρ σ ′ ) d − 2 C maj ( σ ′ ) ≤ C C maj ( σ ′ ) .
Summing with the weights of (118) gives (b2), as C d e r ≥ C C_{\mathrm{der}}\ge C C der ≥ C . In (11.15) at level m − 1 m-1 m − 1 , slow and fast derivatives cost ( ρ m o l ( m − 1 ) / ε m − 2 ) j 1 ≤ 1 (\rho_{\mathrm{mol}}(m-1)/\varepsilon_{m-2})^{j_1}\le1 ( ρ mol ( m − 1 ) / ε m − 2 ) j 1 ≤ 1 and ( ρ m o l ( m − 1 ) / ε m − 1 ) j 2 ≤ 1 (\rho_{\mathrm{mol}}(m-1)/\varepsilon_{m-1})^{j_2}\le1 ( ρ mol ( m − 1 ) / ε m − 1 ) j 2 ≤ 1 , and the power ( ε m − 1 / ρ m o l ( m − 1 ) ) d − 2 (\varepsilon_{m-1}/\rho_{\mathrm{mol}}(m-1))^{d-2} ( ε m − 1 / ρ mol ( m − 1 ) ) d − 2 is absorbed by (b1) if every source of G ′ G' G ′ of current order at least 3 has birth radius at most ρ m o l ( m − 1 ) \rho_{\mathrm{mol}}(m-1) ρ mol ( m − 1 ) . A transported source keeps its birth radius ≤ ε m \le\varepsilon_m ≤ ε m ; a new one has ρ ( m ) \rho(m) ρ ( m ) . By Lemma 7.1 (a), (c), Proposition 2.1 (b) (m − 1 < M m-1<M m − 1 < M ) and (184) ,
ε m ≤ 2 ε m − 1 q , ρ ( m ) ≤ C ε m − 1 1 + γ / 2 + δ , ρ m o l ( m − 1 ) ≥ c ε m − 1 1 + γ / 2 , \varepsilon_m\le2\varepsilon_{m-1}^{q},\qquad\rho(m)\le C\varepsilon_{m-1}^{1+\gamma/2+\delta},\qquad\rho_{\mathrm{mol}}(m-1)\ge c\varepsilon_{m-1}^{1+\gamma/2}, ε m ≤ 2 ε m − 1 q , ρ ( m ) ≤ C ε m − 1 1 + γ /2 + δ , ρ mol ( m − 1 ) ≥ c ε m − 1 1 + γ /2 ,
and q − 1 − γ 2 = 4 δ + q γ q-1-\frac{\gamma}{2}=4\delta+q\gamma q − 1 − 2 γ = 4 δ + q γ (Lemma 7.1 (c)), so, for m > m 0 m>m_0 m > m 0 , both birth radii are at most ρ m o l ( m − 1 ) \rho_{\mathrm{mol}}(m-1) ρ mol ( m − 1 ) . Hence (b2′); for the masses ω v / Ψ v 2 \omega_v/\Psi_v^2 ω v / Ψ v 2 , both sides gain the factor Ψ v / Ψ v ′ \Psi_v/\Psi_{v'} Ψ v / Ψ v ′ .
Step 7: (c). A transported source keeps its smoothing, and a second-order mean takes that of its widest source, so (c1) and (c3) are inherited, and (c2) follows from (c2) for G G G , since ρ m o l ( m ) ε m δ ≤ ρ m o l ( m − 1 ) ε m − 1 4 δ \rho_{\mathrm{mol}}(m)\varepsilon_m^\delta\le\rho_{\mathrm{mol}}(m-1)\varepsilon_{m-1}^{4\delta} ρ mol ( m ) ε m δ ≤ ρ mol ( m − 1 ) ε m − 1 4 δ and ρ ( m ) ε m δ ≤ ρ ( m − 1 ) ε m − 2 4 δ \rho(m)\varepsilon_m^\delta\le\rho(m-1)\varepsilon_{m-2}^{4\delta} ρ ( m ) ε m δ ≤ ρ ( m − 1 ) ε m − 2 4 δ (ε m ≤ ε m − 1 \varepsilon_m\le\varepsilon_{m-1} ε m ≤ ε m − 1 , κ m ≤ κ m − 1 ≤ κ m − 2 \kappa_m\le\kappa_{m-1}\le\kappa_{m-2} κ m ≤ κ m − 1 ≤ κ m − 2 , h m ≤ h m − 1 h_m\le h_{m-1} h m ≤ h m − 1 ).
The new sources have the fresh smoothing (186) , of width r m r_m r m , with the ellipses of (11.18). Then Ω 1 , Ω 2 \Omega_1,\Omega_2 Ω 1 , Ω 2 are piecewise continuous in t t t , and ∣ Ω i ( t ) ∣ ∞ < C ≤ C Ω \lvert\Omega_i(t)\rvert_\infty<C\le C_\Omega ∣ Ω i ( t ) ∣ ∞ < C ≤ C Ω , by Lemma 7.4 (∣ t − t h o r h m ∣ ≤ 9 2 h m \lvert t-t_{\mathrm{hor}}h_m\rvert\le\frac{9}{2}h_m ∣ t − t hor h m ∣ ≤ 2 9 h m ), Lemma 7.10(i) and Proposition 2.1 (a), (c). Both are invertible, as det ∇ y X m − 1 = 1 \det\nabla_yX_{m-1}=1 det ∇ y X m − 1 = 1 (Lemma 7.4 ) and J m κ m ≥ κ m I J^{\kappa_m}_{m}\ge\kappa_m I J m κ m ≥ κ m I (Lemma 7.10(ii)). The long axes Ω 1 ( t ) e 1 , Ω 2 ( t ) e 2 \Omega_1(t)e_1,\Omega_2(t)e_2 Ω 1 ( t ) e 1 , Ω 2 ( t ) e 2 are the columns e 1 , e 2 e_1,e_2 e 1 , e 2 of Jacobians within C ε m − 1 2 δ C\varepsilon_{m-1}^{2\delta} C ε m − 1 2 δ of I I I (proof of Lemma 7.4 , (15) ), times square roots of maxima of the diagonal entries of κ m − 1 − 1 J m κ m \kappa_{m-1}^{-1}J^{\kappa_m}_{m} κ m − 1 − 1 J m κ m , at least 1 as these entries have average 1 over a period by (19) and (21) . So the matrix of long axes is within C ε m − 1 2 δ C\varepsilon_{m-1}^{2\delta} C ε m − 1 2 δ of I I I times a diagonal matrix with entries ≥ 1 \ge1 ≥ 1 , and, for m > m 0 m>m_0 m > m 0 , its inverse has ∣ ⋅ ∣ ∞ \lvert\cdot\rvert_\infty ∣ ⋅ ∣ ∞ -norm at most ( 1 − C ε m − 1 2 δ ) − 1 ≤ 2 ≤ C Ω (1-C\varepsilon_{m-1}^{2\delta})^{-1}\le2\le C_\Omega ( 1 − C ε m − 1 2 δ ) − 1 ≤ 2 ≤ C Ω .
This gives (c1), with 2 ℓ − 1 ≥ N m o m 2\ell-1\ge N_{\mathrm{mom}} 2 ℓ − 1 ≥ N mom by item (1). Lemma 8.2 gives the bound of (c3) with its constants, which the c j c_j c j of item (3) dominate; the bound passes to H ρ N ( T 2 ) H^N_\rho(\mathbb{T}^2) H ρ N ( T 2 ) since S ( t ) S(t) S ( t ) commutes with derivatives. The widths follow from (178) , (184) , ε m − 1 ≤ 2 ε m − 2 q \varepsilon_{m-1}\le2\varepsilon_{m-2}^{q} ε m − 1 ≤ 2 ε m − 2 q and 4 δ < q − 1 4\delta<q-1 4 δ < q − 1 (Step 6):
r m ≤ C ε m − 1 1 + γ / 2 + δ + Θ ≤ C ρ m o l ( m − 1 ) ε m − 1 δ + Θ ≤ C w i d ρ m o l ( m − 1 ) ε m − 1 δ − η , r_m\le C\varepsilon_{m-1}^{1+\gamma/2+\delta+\Theta}\le C\rho_{\mathrm{mol}}(m-1)\varepsilon_{m-1}^{\delta+\Theta}\le C_{\mathrm{wid}}\rho_{\mathrm{mol}}(m-1)\varepsilon_{m-1}^{\delta-\eta}, r m ≤ C ε m − 1 1 + γ /2 + δ + Θ ≤ C ρ mol ( m − 1 ) ε m − 1 δ + Θ ≤ C wid ρ mol ( m − 1 ) ε m − 1 δ − η ,
r m ≤ C ρ ( m − 1 ) ε m − 2 ( q − 1 ) ( 1 + γ / 2 + δ ) + q Θ ≤ C w i d ρ ( m − 1 ) ε m − 2 4 δ . r_m\le C\rho(m-1)\varepsilon_{m-2}^{(q-1)(1+\gamma/2+\delta)+q\Theta}\le C_{\mathrm{wid}}\rho(m-1)\varepsilon_{m-2}^{4\delta}. r m ≤ C ρ ( m − 1 ) ε m − 2 ( q − 1 ) ( 1 + γ /2 + δ ) + q Θ ≤ C wid ρ ( m − 1 ) ε m − 2 4 δ .
Step 8: (d). The coupling matrix (68) of G ′ G' G ′ of order N N N at scale ρ m o l ( m − 1 ) \rho_{\mathrm{mol}}(m-1) ρ mol ( m − 1 ) is at most C C C times the priced majorant matrix, entry by entry, for both masses. Indeed, by (b1), the coefficient (116) of order d d d of a source σ ′ \sigma' σ ′ is at most κ m − 1 ρ σ ′ d − 2 fac ( σ ′ ) maj ( σ ′ ) \kappa_{m-1}\rho_{\sigma'}^{d-2}\operatorname{fac}(\sigma')\operatorname{maj}(\sigma') κ m − 1 ρ σ ′ d − 2 fac ( σ ′ ) maj ( σ ′ ) . Its transport to x x x gives the orders d ′ ≤ d d'\le d d ′ ≤ d at the cost ε m − 2 d ′ − d ≤ ρ σ ′ d ′ − d \varepsilon_{m-2}^{d'-d}\le\rho_{\sigma'}^{d'-d} ε m − 2 d ′ − d ≤ ρ σ ′ d ′ − d ; if d ′ < d d'<d d ′ < d , then d ≥ 3 d\ge3 d ≥ 3 , so σ ′ \sigma' σ ′ has current order at least 3 and ρ σ ′ ≤ ε m − 2 \rho_{\sigma'}\le\varepsilon_{m-2} ρ σ ′ ≤ ε m − 2 , by Step 6. Its derivatives at the scale ρ m o l ( m − 1 ) ≤ ε m − 1 \rho_{\mathrm{mol}}(m-1)\le\varepsilon_{m-1} ρ mol ( m − 1 ) ≤ ε m − 1 cost constants, as in Step 6: by the chain rule, ∇ x j \nabla_x^j ∇ x j of the transported coefficient, at z = y / ε m − 1 z=y/\varepsilon_{m-1} z = y / ε m − 1 , combines ∇ z j 1 ∇ z j 2 \nabla_z^{j_1}\nabla_z^{j_2} ∇ z j 1 ∇ z j 2 of the form, j 1 + j 2 ≤ j j_1+j_2\le j j 1 + j 2 ≤ j , at the cost ε m − 2 − j 1 ε m − 1 − j 2 \varepsilon_{m-2}^{-j_1}\varepsilon_{m-1}^{-j_2} ε m − 2 − j 1 ε m − 1 − j 2 , as the derivatives of order j + 1 j+1 j + 1 of X m − 2 ( t , ⋅ , t ˉ m − 1 ) X_{m-2}(t,\cdot,\bar t_{m-1}) X m − 2 ( t , ⋅ , t ˉ m − 1 ) and of its inverse are at most C ε m − 2 − j C\varepsilon_{m-2}^{-j} C ε m − 2 − j (Lemma 11.9(i), Cauchy’s estimate), and at most two time cutoffs are positive at once. So
c d ′ − 2 κ m − 1 r σ ′ d ′ − 2 ∑ j ≤ N ρ m o l ( m − 1 ) j j ! sup [ t A + 3 τ m , t f ] × T 2 ∣ ∇ j A σ ′ , d ′ ∣ ≤ C fac ( σ ′ ) maj ( σ ′ ) ( ρ σ ′ r σ ′ ) d ′ − 2 . \frac{c_{d'-2}}{\kappa_{m-1}r_{\sigma'}^{d'-2}}\sum_{j\le N}\frac{\rho_{\mathrm{mol}}(m-1)^j}{j!}\sup_{[t_A+3\tau_m,t_f]\times\mathbb{T}^2}\lvert\nabla^j A_{\sigma',d'}\rvert\le C\operatorname{fac}(\sigma')\operatorname{maj}(\sigma')\left(\frac{\rho_{\sigma'}}{r_{\sigma'}}\right)^{d'-2}. κ m − 1 r σ ′ d ′ − 2 c d ′ − 2 j ≤ N ∑ j ! ρ mol ( m − 1 ) j [ t A + 3 τ m , t f ] × T 2 sup ∣ ∇ j A σ ′ , d ′ ∣ ≤ C fac ( σ ′ ) maj ( σ ′ ) ( r σ ′ ρ σ ′ ) d ′ − 2 .
and the last power is at most the price max { 1 , ( ρ σ ′ / r σ ′ ) d σ ′ − 2 } \max\{1,(\rho_{\sigma'}/r_{\sigma'})^{d_{\sigma'}-2}\} max { 1 , ( ρ σ ′ / r σ ′ ) d σ ′ − 2 } , which compensates for the coefficient c d ′ − 2 r σ ′ 2 − d ′ c_{d'-2}r_{\sigma'}^{2-d'} c d ′ − 2 r σ ′ 2 − d ′ of the kernel. So the row and column sums of the coupling matrix are at most C C C times (122) . As in Step 3, with one more step of the scales ( ε m − 1 δ / 4 ≤ 2 ε m − 2 q δ / 4 ) (\varepsilon_{m-1}^{\delta/4}\le2\varepsilon_{m-2}^{q\delta/4}) ( ε m − 1 δ /4 ≤ 2 ε m − 2 q δ /4 ) , for m > m 0 m>m_0 m > m 0 , m 0 m_0 m 0 being chosen after C C C , C a d m C_{\mathrm{adm}} C adm and C m a j C_{\mathrm{maj}} C maj , both terms are at most 1 2 C a d m ε m − 2 q δ / 4 \frac12 C_{\mathrm{adm}}\varepsilon_{m-2}^{q\delta/4} 2 1 C adm ε m − 2 q δ /4 : this is (d).
Smoothing within one period: Bloch wave analysis
This section proves part (a) of Proposition 5.3 5.3 5.3 , the preparation, by Bloch wave analysis; part (b) is proved in Section 13 13 13 , for the time t p r t_{\mathrm{pr}} t pr and the map Π \Pi Π constructed here. Part (a) says that within one period 4 h m 4h_m 4 h m of the pulses the solution of an admissible family of level m m m becomes, up to a small error, smooth at the scale ρ ( m ) \rho(m) ρ ( m ) , with bounds by the L 2 L^2 L 2 norm of the datum. This scale is the distance over which the coarse diffusivity κ m − 1 \kappa_{m-1} κ m − 1 spreads in one renewal time; below the terminal level it is far larger than the distance that molecular diffusion reaches in that time. So the smoothing must come from the homogenization of the pulses of level m m m , and it must hold to the accuracy ε m − 1 q \varepsilon_{m-1}^q ε m − 1 q of the cycle, uniformly in the contrast, and without any derivative of the datum.
We obtain it from an exact decomposition of the evolution. In the chart of a renewal block, with the conductivity frozen at a point, Bloch waves split the evolution into evolutions on the period cell, one for each frequency (Subsection 12.2 12.2 12.2 ). At small frequencies, the cell mean of the principal branch of these evolutions evolves at a scalar rate, the symbol (Subsection 12.3 12.3 12.3 ). To leading order the symbol is the homogenized operator of the pulses; as a block carries pulses of both orientations, over one block the principal branch is damped by a Gaussian in the frequency at the scale 1 / ρ ( m ) 1/\rho(m) 1/ ρ ( m ) , which is the smoothing we need (Subsection 12.4 12.4 12.4 ). Everything else decays at the cell rate, the inverse of the time that molecular diffusion takes to cross a period cell, and over a block its norm is multiplied by a number smaller than any power of ε m − 1 \varepsilon_{m-1} ε m − 1 (Subsection 12.5 12.5 12.5 ). Subsections 12.6 12.6 12.6 –12.9 12.9 12.9 add the sources of the family, which take derivatives of the solution. We expand the solution in its sources by Duhamel’s formula, freeze the coefficients on patches of size ρ ( m ) \rho(m) ρ ( m ) , and estimate each term by duality, against solutions of the adjoint cell evolutions, which carry the derivatives that the sources take, except near the ends of a block (Lemma 12.12 (iii)). On the principal branch a source then costs, over a block, at most a power of 1 + ρ ( m ) ∣ ξ ∣ 1+\rho(m)|\xi| 1 + ρ ( m ) ∣ ξ ∣ , which the Gaussian absorbs; everything else is damped. Subsection 12.10 12.10 12.10 assembles the proof.
Conventions. In Subsections 12.2 12.2 12.2 –12.5 12.5 12.5 every constant of a statement is named in it, and C C C is absolute in the proofs of Subsections 12.2 12.2 12.2 and 12.3 12.3 12.3 ; Subsection 12.6 12.6 12.6 states the conventions of Subsections 12.6 12.6 12.6 –12.9 12.9 12.9 . Elsewhere, C C C and c c c denote positive constants depending only on α \alpha α , s s s , q q q , ζ \zeta ζ , Λ \Lambda Λ , n n n , t 0 t_0 t 0 , T T T and on the orders and constants fixed in Subsection 11.1 11.1 11.1 and below, which may change from line to line; C ( j ) C(j) C ( j ) depends also on j j j , and C ( J ) C(J) C ( J ) and c ( J ) c(J) c ( J ) on J J J . The corrector amplitude ε ˉ m = ε m ( κ m − 1 / κ m ) 1 / 2 \bar{\varepsilon}_m = \varepsilon_m(\kappa_{m-1}/\kappa_m)^{1/2} ε ˉ m = ε m ( κ m − 1 / κ m ) 1/2 of (30) , with the contrast κ m − 1 / κ m \kappa_{m-1}/\kappa_m κ m − 1 / κ m , is at most C ε m 3 δ ρ ( m ) C\varepsilon_m^{3\delta}\rho(m) C ε m 3 δ ρ ( m ) , by (96) . The cell rate is κ m / ε m 2 \kappa_m/\varepsilon_m^2 κ m / ε m 2 . Two identities, from the definition of ε ˉ m \bar{\varepsilon}_m ε ˉ m and (33) , translate between cell units and the coarse scale:
κ m ε m 2 ε ˉ m 2 = κ m − 1 , ρ ( m ) 2 ε ˉ m 2 = κ m h m ε m 2 . \frac{\kappa_m}{\varepsilon_m^2}\bar{\varepsilon}_m^2=\kappa_{m-1},\qquad\frac{\rho(m)^2}{\bar{\varepsilon}_m^2}=\frac{\kappa_m h_m}{\varepsilon_m^2}. ε m 2 κ m ε ˉ m 2 = κ m − 1 , ε ˉ m 2 ρ ( m ) 2 = ε m 2 κ m h m .
in cell units the homogenized equation damps the frequency ξ \xi ξ at the rate ( ε ˉ m ∣ ξ ∣ ) 2 (\bar{\varepsilon}_m|\xi|)^2 ( ε ˉ m ∣ ξ ∣ ) 2 , and a renewal block lasts ( ρ ( m ) / ε ˉ m ) 2 (\rho(m)/\bar{\varepsilon}_m)^2 ( ρ ( m ) / ε ˉ m ) 2 cell diffusion times.
The constant c m o l c_{\mathrm{mol}} c mol . The constant c m o l ∈ ( 0 , 1 ] c_{\mathrm{mol}}\in(0,1] c mol ∈ ( 0 , 1 ] of the molecular radius (115) is fixed here, after Λ \Lambda Λ : it is at most a threshold depending only on α , s , q , ζ , Λ \alpha,s,q,\zeta,\Lambda α , s , q , ζ , Λ , below which, for every M M M , κ ∈ K M \kappa\in\mathcal{K}_M κ ∈ K M and 2 ≤ m ≤ M 2\le m\le M 2 ≤ m ≤ M ,
ρ m o l ( m ) j + 1 ∥ ∇ j b m ∥ L ∞ ( R × T 2 ) ≤ 2 − j − 2 j ! κ m ( j ≥ 1 ) , \rho_{\mathrm{mol}}(m)^{j+1}\|\nabla^j\mathbf{b}_m\|_{L^\infty(\mathbb{R}\times\mathbb{T}^2)}\le2^{-j-2}j!\,\kappa_m\qquad(j\ge1), ρ mol ( m ) j + 1 ∥ ∇ j b m ∥ L ∞ ( R × T 2 ) ≤ 2 − j − 2 j ! κ m ( j ≥ 1 ) ,
{#eq:12.1}
\frac{\rho_{\mathrm{mol}}(m)}{\varepsilon_m}\le 2^{-11},\qquad \varepsilon_m^{\beta-2}\frac{\rho_{\mathrm{mol}}(m)^2}{\kappa_m}\le 2^{-20}.
The first condition is (i) of Proposition 8.3 for the velocity b m \mathbf{b}_m b m at the scale ρ m o l ( m ) \rho_{\mathrm{mol}}(m) ρ mol ( m ) and every order. It holds for small c m o l c_{\mathrm{mol}} c mol : Proposition 7.5 and Cauchy’s estimate on polydiscs of radius ε m / 100 \varepsilon_m/100 ε m /100 give ∥ ∇ j b m ∥ L ∞ ( R × T 2 ) ≤ C j ( j − 1 ) ! ε m β − 1 − j \|\nabla^j\mathbf{b}_m\|_{L^\infty(\mathbb{R}\times\mathbb{T}^2)}\le C^j(j-1)!\varepsilon_m^{\beta-1-j} ∥ ∇ j b m ∥ L ∞ ( R × T 2 ) ≤ C j ( j − 1 )! ε m β − 1 − j with an absolute C C C , and (115) bounds the left side by c m o l j + 1 C j ( j − 1 ) ! κ m c_{\mathrm{mol}}^{j+1}C^j(j-1)!\kappa_m c mol j + 1 C j ( j − 1 )! κ m . The last condition makes the molecular time ρ m o l ( m ) 2 / κ m \rho_{\mathrm{mol}}(m)^2/\kappa_m ρ mol ( m ) 2 / κ m short compared with the strain time ε m 2 − β \varepsilon_m^{2-\beta} ε m 2 − β of b m \mathbf{b}_m b m (49) .
The order of choices. The depth N d e p N_{\mathrm{dep}} N dep of the Duhamel expansion (Subsection 12.6) is the least integer with the second condition below, N b u f N_{\mathrm{buf}} N buf is the order of the molecular smoothing, and the moment order N m o m N_{\mathrm{mom}} N mom of the smoothings satisfies the first condition (item (1) of Subsection 11.1, Table 11.1):
q δ N m o m > Q + 3 , q δ 4 ( N d e p + 1 ) > Q + 3 , N b u f : = ( N d e p + 1 ) ( N s r c − 2 ) + N m o m . (123) q\delta N_{\mathrm{mom}}>Q+3,\qquad\frac{q\delta}{4}(N_{\mathrm{dep}}+1)>Q+3,\qquad N_{\mathrm{buf}}:=(N_{\mathrm{dep}}+1)(N_{\mathrm{src}}-2)+N_{\mathrm{mom}}. \tag*{(123)} q δ N mom > Q + 3 , 4 q δ ( N dep + 1 ) > Q + 3 , N buf := ( N dep + 1 ) ( N src − 2 ) + N mom . (123)
The integers introduced in Subsections 12.6–12.9 depend only on these orders, on N s r c N_{\mathrm{src}} N src , Q Q Q and N o u t N_{\mathrm{out}} N out , and N c o e f N_{\mathrm{coef}} N coef is at least each of them and N b u f N_{\mathrm{buf}} N buf . The top level m 0 m_0 m 0 is fixed after c m o l c_{\mathrm{mol}} c mol .
Statements
The preparation runs over the interval of Figure 12.1 . From t c t_c t c the solution runs for the molecular time, which gives it derivatives at the scale ρ m o l ( m ) \rho_{\mathrm{mol}}(m) ρ mol ( m ) (Lemma 12.17 ), and from the end t 1 t_1 t 1 of the molecular interval it is expanded in its sources. The rest of the block of t 1 t_1 t 1 may carry pulses of one orientation only; the next, complete, block smooths, and t p r t_{\mathrm{pr}} t pr is its end, inside a quiet margin. Proposition 12.1 gives four further properties of t p r t_{\mathrm{pr}} t pr and Π \Pi Π . By (i) the comparison of § 5.8 starts at t p r t_{\mathrm{pr}} t pr , where the lift adds no corrector (Lemma 13.1 ); (iii) gives the derivatives that the passage of § 5.11 spends; and (iv) bounds the datum of the successor, the retained trace of Π g \Pi g Π g .
Figure 12.1. The preparation of a cycle, in time (not to scale): the molecular interval from t c t_c t c to t 1 t_1 t 1 , the rest of the block containing t 1 t_1 t 1 (the old partial block), and one complete renewal block, from the block boundary t 2 t_2 t 2 to t p r t_{\mathrm{pr}} t pr . Vertical strokes: pulses, solid and dotted for the two orientations. Shaded: the quiet margins around t 2 t_2 t 2 and t p r t_{\mathrm{pr}} t pr , of half-width τ m ′ \tau'_m τ m ′ .
Proposition 12.1 (Further properties of the preparation). There exists C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ C(\alpha,s,q,\zeta,\Lambda,n,t_0,T)<\infty C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ such that, for every M M M , κ \kappa κ , m m m , t A t_A t A , t c t_c t c , t f t_f t f , G G G and G ′ G' G ′ as in Proposition 5.3, the time t p r t_{\mathrm{pr}} t pr and the map Π \Pi Π of Proposition 5.3 can be chosen, satisfying (a) and (b) with this C C C , such that also the following hold.
(i) The time t p r t_{\mathrm{pr}} t pr comes after the molecular time, and the velocity added at level m m m vanishes near it:
t c + ρ m o l ( m ) 2 κ m ≤ t p r , b m = b m − 1 on [ t p r − τ m ′ , t p r + τ m ′ ] × T 2 . t_c+\frac{\rho_{\mathrm{mol}}(m)^2}{\kappa_m}\le t_{\mathrm{pr}},\qquad\mathbf{b}_m=\mathbf{b}_{m-1}\quad\text{on }[t_{\mathrm{pr}}-\tau'_m,t_{\mathrm{pr}}+\tau'_m]\times\mathbb{T}^2. t c + κ m ρ mol ( m ) 2 ≤ t pr , b m = b m − 1 on [ t pr − τ m ′ , t pr + τ m ′ ] × T 2 .
(ii) Π \Pi Π maps every field of G G G whose components are constant functions to itself.
(iii) For every datum g g g , the solution U U U of G G G from g g g at time t c t_c t c satisfies
∥ U ∥ L ∞ ( [ t p r , t f ] ; H ρ m o l ( m ) , ω 2 N r e p + 4 ( T 2 ) ) ≤ C ∥ g ∥ L G 2 ( T 2 ) . \lVert U\rVert_{L^\infty([t_{\mathrm{pr}},t_f];H^{2N_{\mathrm{rep}}+4}_{\rho_{\mathrm{mol}}(m),\omega}(\mathbb{T}^2))} \le C\lVert g\rVert_{L^2_G(\mathbb{T}^2)}. ∥ U ∥ L ∞ ([ t pr , t f ] ; H ρ mol ( m ) , ω 2 N rep + 4 ( T 2 )) ≤ C ∥ g ∥ L G 2 ( T 2 ) .
(iv) For every datum g g g ,
∥ exp p ( − Γ m ( t p r ) : ∇ 2 ) Π g ∥ H ρ ( m ) , G N r e t ( T 2 ) ≤ C ∥ g ∥ L G 2 ( T 2 ) . \left\lVert\exp_p(-\Gamma_m(t_{\mathrm{pr}}):\nabla^2)\Pi g\right\rVert_{H^{N_{\mathrm{ret}}}_{\rho(m),G}(\mathbb{T}^2)} \le C\lVert g\rVert_{L^2_G(\mathbb{T}^2)}. exp p ( − Γ m ( t pr ) : ∇ 2 ) Π g H ρ ( m ) , G N ret ( T 2 ) ≤ C ∥ g ∥ L G 2 ( T 2 ) .
The frozen operator and its Bloch waves
In the chart of a renewal block (Lemma 7.9 ) the coarse velocity is gone: what remains is the periodic shear of the pulses and a conductivity close to the identity that varies only on the scale ε m − 1 \varepsilon_{m-1} ε m − 1 , much larger than ρ ( m ) \rho(m) ρ ( m ) . As a point of the datum influences, over one block, a neighbourhood of size about ρ ( m ) \rho(m) ρ ( m ) , we freeze the conductivity at a point y 0 y_0 y 0 ; Subsection 12.9 bounds the error. The frozen operator has periodic coefficients, and Bloch waves reduce it to one evolution on the period cell for each frequency, the fibre equation (126) .
Standing hypotheses. Throughout Subsections 12.2–12.5, let Λ ≥ Λ 0 \Lambda\ge\Lambda_0 Λ ≥ Λ 0 , where Λ 0 ( α , s , q , ζ ) ≥ 128 \Lambda_0(\alpha,s,q,\zeta) \ge128 Λ 0 ( α , s , q , ζ ) ≥ 128 is at least the thresholds of Proposition 2.1 and Lemma 10.1 and is so large that
4 π 2 κ m τ m ε m 2 ≥ 6 , ε m − 1 δ ≤ 1 16 for every M ≥ 2 , κ ∈ K M , 2 ≤ m ≤ M (124) \frac{4\pi^2\kappa_m\tau_m}{\varepsilon_m^2} \ge6,\qquad\varepsilon_{m-1}^{\delta} \le\frac{1}{16}\qquad\text{for every }M\ge2,\ \kappa\in\mathcal{K}_M,\ 2\le m\le M \tag*{(124)} ε m 2 4 π 2 κ m τ m ≥ 6 , ε m − 1 δ ≤ 16 1 for every M ≥ 2 , κ ∈ K M , 2 ≤ m ≤ M (124)
(by Proposition 2.1 (c)). Fix M ≥ 2 M\ge2 M ≥ 2 , κ ∈ K M \kappa\in\mathcal{K}_M κ ∈ K M , 2 ≤ m ≤ M 2\le m\le M 2 ≤ m ≤ M , l ∈ Z l\in\mathbb{Z} l ∈ Z , a point y 0 ∈ R 2 y_0\in\mathbb{R}^2 y 0 ∈ R 2 and a measurable map K \mathbf{K} K from R \mathbb{R} R to the symmetric 2 × 2 2\times2 2 × 2 matrices with ∥ K − I ∥ L ∞ ( R ) ≤ ε m − 1 δ \lVert\mathbf{K}-I\rVert_{L^\infty(\mathbb{R})}\le\varepsilon_{m-1}^{\delta} ∥ K − I ∥ L ∞ ( R ) ≤ ε m − 1 δ . In the applications K ( t ) = K m , l ( t , y 0 ) \mathbf{K}(t)=\mathbf{K}_{m,l}(t,y_0) K ( t ) = K m , l ( t , y 0 ) for ∣ t − l h m ∣ ≤ h m \lvert t-lh_m\rvert\le h_m ∣ t − l h m ∣ ≤ h m , the conductivity of the chart of the block of l l l frozen at the label y 0 y_0 y 0 , and K ( t ) = I \mathbf{K}(t)=I K ( t ) = I otherwise; by (54) , ∥ K − I ∥ L ∞ ( R ) ≤ C ε m − 1 2 δ ≤ ε m − 1 δ \lVert\mathbf{K}-I\rVert_{L^\infty(\mathbb{R})}\le C\varepsilon_{m-1}^{2\delta}\le\varepsilon_{m-1}^{\delta} ∥ K − I ∥ L ∞ ( R ) ≤ C ε m − 1 2 δ ≤ ε m − 1 δ for Λ 0 \Lambda_0 Λ 0 large. By Lemma 7.9 , the level-m m m operator in the chart of the block is ∂ t + ∇ ⊥ ψ m ⋅ ∇ − κ m ∇ ⋅ ( K m , l ∇ ⋅ ) \partial_t+\nabla^\perp\psi_m\cdot\nabla-\kappa_m\nabla\cdot(\mathbf{K}_{m,l}\nabla\cdot) ∂ t + ∇ ⊥ ψ m ⋅ ∇ − κ m ∇ ⋅ ( K m , l ∇ ⋅ ) , with ψ m \psi_m ψ m of (56) . With its conductivity frozen at y 0 y_0 y 0 , it becomes the frozen operator
∂ t + ∇ ⊥ ψ m ( t , y ) ⋅ ∇ − κ m ∇ ⋅ K ( t ) ∇ ⋅ on R × R 2 , (125) \partial_t+\nabla^\perp\psi_m(t,y)\cdot\nabla-\kappa_m\nabla\cdot\mathbf{K}(t)\nabla\cdot\quad\text{on }\mathbb{R}\times\mathbb{R}^2, \tag*{(125)} ∂ t + ∇ ⊥ ψ m ( t , y ) ⋅ ∇ − κ m ∇ ⋅ K ( t ) ∇ ⋅ on R × R 2 , (125)
whose coefficients are ε m \varepsilon_m ε m -periodic in y y y (§ 1.1 ).
Fibres. We write z : = y / ε m z := y/\varepsilon_m z := y / ε m for the fast variable, as in § 5.2 . The velocity ( ∇ ⊥ ψ m ) ( t , ε m z ) (\nabla^\perp\psi_m)(t,\varepsilon_m z) ( ∇ ⊥ ψ m ) ( t , ε m z ) is 1-periodic in z z z , divergence free and of mean zero over the cell. On C 2 \mathbb{C}^2 C 2 the dot is bilinear, ξ ⋅ η : = ξ 1 η 1 + ξ 2 η 2 \xi\cdot\eta:= \xi_1\eta_1+\xi_2\eta_2 ξ ⋅ η := ξ 1 η 1 + ξ 2 η 2 , so that ∣ ξ ∣ 2 = ξ ⋅ ξ ˉ |\xi|^2=\xi\cdot\bar{\xi} ∣ ξ ∣ 2 = ξ ⋅ ξ ˉ . For ξ ∈ C 2 \xi\in\mathbb{C}^2 ξ ∈ C 2 , the function exp ( i ξ ⋅ ( y − y 0 ) ) U ( t , y / ε m ) \exp(i\xi\cdot(y-y_0))U(t,y/\varepsilon_m) exp ( i ξ ⋅ ( y − y 0 )) U ( t , y / ε m ) , with U U U periodic in z z z , is annihilated by (125) if and only if U U U solves the fibre equation at ξ \xi ξ ,
∂ t U = κ m ε m 2 ( ∇ + i ε m ξ ) ⋅ K ( t ) ( ∇ + i ε m ξ ) U − 1 ε m ( ∇ ⊥ ψ m ) ( t , ε m z ) ⋅ ( ∇ + i ε m ξ ) U (126) \partial_t U = \frac{\kappa_m}{\varepsilon_m^2}(\nabla+i\varepsilon_m\xi)\cdot K(t)(\nabla+i\varepsilon_m\xi)U-\frac{1}{\varepsilon_m}(\nabla^\perp\psi_m)(t,\varepsilon_m z)\cdot(\nabla+i\varepsilon_m\xi)U \tag*{(126)} ∂ t U = ε m 2 κ m ( ∇ + i ε m ξ ) ⋅ K ( t ) ( ∇ + i ε m ξ ) U − ε m 1 ( ∇ ⊥ ψ m ) ( t , ε m z ) ⋅ ( ∇ + i ε m ξ ) U (126)
on T 2 \mathbb{T}^2 T 2 , with ∇ \nabla ∇ in z z z and ∇ ⊥ ψ m \nabla^\perp\psi_m ∇ ⊥ ψ m the velocity of (125) , in y y y . Solutions are energy solutions, one for each datum. The real frequencies with ε m ∣ ξ ∣ ∞ ≤ π \varepsilon_m|\xi|_\infty\leq\pi ε m ∣ ξ ∣ ∞ ≤ π form the (Brillouin) zone . For real ξ \xi ξ the transport term of (126) is skew-adjoint, as the velocity is divergence free, so
d d t ∥ U ( t ) ∥ L 2 ( T 2 ) 2 = − 2 κ m ε m 2 ⟨ K ( t ) ( ∇ + i ε m ξ ) U , ( ∇ + i ε m ξ ) U ⟩ L 2 ( T 2 ) ≤ 0 , (127) \frac{\mathrm{d}}{\mathrm{d}t}\|U(t)\|_{L^2(\mathbb{T}^2)}^2=-\frac{2\kappa_m}{\varepsilon_m^2}\left\langle K(t)(\nabla+i\varepsilon_m\xi)U,(\nabla+i\varepsilon_m\xi)U\right\rangle_{L^2(\mathbb{T}^2)}\leq0, \tag*{(127)} d t d ∥ U ( t ) ∥ L 2 ( T 2 ) 2 = − ε m 2 2 κ m ⟨ K ( t ) ( ∇ + i ε m ξ ) U , ( ∇ + i ε m ξ ) U ⟩ L 2 ( T 2 ) ≤ 0 , (127)
and, by Fourier series, for real ξ \xi ξ in the zone and every U ∈ H 1 ( T 2 ) U\in H^1(\mathbb{T}^2) U ∈ H 1 ( T 2 ) ,
∥ ( ∇ + i ε m ξ ) U ∥ L 2 ( T 2 ) ≥ max { ε m ∣ ξ ∣ ∥ U ∥ L 2 ( T 2 ) , π ∥ U − ( U ) T 2 ∥ L 2 ( T 2 ) } , (128) \|(\nabla+i\varepsilon_m\xi)U\|_{L^2(\mathbb{T}^2)}\geq\max\left\{\varepsilon_m|\xi|\|U\|_{L^2(\mathbb{T}^2)},\ \pi\|U-(U)_{\mathbb{T}^2}\|_{L^2(\mathbb{T}^2)}\right\}, \tag*{(128)} ∥ ( ∇ + i ε m ξ ) U ∥ L 2 ( T 2 ) ≥ max { ε m ∣ ξ ∣∥ U ∥ L 2 ( T 2 ) , π ∥ U − ( U ) T 2 ∥ L 2 ( T 2 ) } , (128)
since ∣ 2 π j + ε m ξ ∣ ≥ ε m ∣ ξ ∣ |2\pi j+\varepsilon_m\xi|\geq\varepsilon_m|\xi| ∣2 π j + ε m ξ ∣ ≥ ε m ∣ ξ ∣ for every j ∈ Z 2 j\in\mathbb{Z}^2 j ∈ Z 2 , and ≥ π \geq\pi ≥ π for j ≠ 0 j\neq0 j = 0 .
Bloch sums. Fields g = ( g v ) v ∈ V g=(g_v)_{v\in V} g = ( g v ) v ∈ V on R 2 \mathbb{R}^2 R 2 carry the masses ω v \omega_v ω v of (35) , and their moments about y 0 y_0 y 0 at the scale ρ ( m ) \rho(m) ρ ( m ) are measured by
∥ g ∥ L ω , N ′ 2 ( R 2 ) : = ∥ ( 1 + ∣ y − y 0 ∣ 2 ρ ( m ) 2 ) N ′ / 2 g ∥ L ω 2 ( R 2 ) , \|g\|_{L^2_{\omega,N'}(\mathbb{R}^2)}:=\left\|\left(1+\frac{|y-y_0|^2}{\rho(m)^2}\right)^{N'/2}g\right\|_{L^2_\omega(\mathbb{R}^2)}, ∥ g ∥ L ω , N ′ 2 ( R 2 ) := ( 1 + ρ ( m ) 2 ∣ y − y 0 ∣ 2 ) N ′ /2 g L ω 2 ( R 2 ) ,
∥ g ∥ H ρ , ω , N ′ N ( R 2 ) 2 : = ∑ j ≤ N ρ 2 j ( j ! ) 2 ∥ ∇ j g ∥ L ω , N ′ 2 ( R 2 ) 2 ; (129) \|g\|_{H^N_{\rho,\omega,N'}(\mathbb{R}^2)}^2:=\sum_{j\leq N}\frac{\rho^{2j}}{(j!)^2}\|\nabla^jg\|_{L^2_{\omega,N'}(\mathbb{R}^2)}^2; \tag*{(129)} ∥ g ∥ H ρ , ω , N ′ N ( R 2 ) 2 := j ≤ N ∑ ( j ! ) 2 ρ 2 j ∥ ∇ j g ∥ L ω , N ′ 2 ( R 2 ) 2 ; (129)
the centre y 0 y_0 y 0 is always the point of the frozen operator. For a measurable field g g g on R 2 \mathbb{R}^2 R 2 with ∥ g ∥ L ω , 2 2 ( R 2 ) < ∞ \|g\|_{L^2_{\omega,2}(\mathbb{R}^2)}<\infty ∥ g ∥ L ω , 2 2 ( R 2 ) < ∞ , its Bloch sums about y 0 y_0 y 0 are the sums
B g ( ξ , z ) : = ∑ j ∈ Z 2 exp ( − i ξ ⋅ ( ε m ( z + j ) − y 0 ) ) g ( ε m ( z + j ) ) , (130) \mathcal{B}g(\xi,z):=\sum_{j\in\mathbb{Z}^2}\exp\left(-i\xi\cdot\left(\varepsilon_m(z+j)-y_0\right)\right)g\left(\varepsilon_m(z+j)\right), \tag*{(130)} B g ( ξ , z ) := j ∈ Z 2 ∑ exp ( − i ξ ⋅ ( ε m ( z + j ) − y 0 ) ) g ( ε m ( z + j ) ) , (130)
periodic in z z z . At every real ξ \xi ξ they converge absolutely for almost every z z z , and in L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) , by the Cauchy–Schwarz inequality against the weight ( 1 + ∣ y − y 0 ∣ 2 / ρ ( m ) 2 ) − 2 \left(1+|y-y_0|^2/\rho(m)^2\right)^{-2} ( 1 + ∣ y − y 0 ∣ 2 / ρ ( m ) 2 ) − 2 , which is summable over the points y = ε m ( z + j ) y=\varepsilon_m(z+j) y = ε m ( z + j ) , j ∈ Z 2 j\in\mathbb{Z}^2 j ∈ Z 2 . Their mean over T 2 \mathbb{T}^2 T 2 is ε m − 2 g ^ ( ξ ) \varepsilon_m^{-2}\widehat{g}(\xi) ε m − 2 g ( ξ ) , where g ^ ( ξ ) : = ∫ R 2 exp ( − i ξ ⋅ ( y − y 0 ) ) g ( y ) d y \widehat{g}(\xi):=\int_{\mathbb{R}^2}\exp(-i\xi\cdot(y-y_0))g(y)\,\mathrm{d}y g ( ξ ) := ∫ R 2 exp ( − i ξ ⋅ ( y − y 0 )) g ( y ) d y is the Fourier transform about y 0 y_0 y 0 . They satisfy B ( ∇ g ) = ε m − 1 ( ∇ + i ε m ξ ) B g \mathcal{B}(\nabla g)=\varepsilon_m^{-1}(\nabla+i\varepsilon_m\xi)\mathcal{B}g B ( ∇ g ) = ε m − 1 ( ∇ + i ε m ξ ) B g , with ∇ \nabla ∇ in y y y on the left and in z z z on the right, and B ( a ( ⋅ / ε m ) g ) = a B g \mathcal{B}(a(\cdot/\varepsilon_m)g)=a\mathcal{B}g B ( a ( ⋅ / ε m ) g ) = a B g for every 1-periodic function a a a . The following lemma, Plancherel’s identity for Bloch sums, returns bounds proved at each frequency to the plane.
Lemma 12.2 (Bloch sums). Let g g g be a measurable field on R 2 \mathbb{R}^2 R 2 with ∥ g ∥ L ω , 2 2 ( R 2 ) < ∞ \|g\|_{L^2_{\omega,2}(\mathbb{R}^2)}<\infty ∥ g ∥ L ω , 2 2 ( R 2 ) < ∞ . Then the Bloch sums of g g g satisfy
∫ z o n e ∥ B g ( ξ , ⋅ ) ∥ L ω 2 ( T 2 ) 2 d ξ = ( 2 π ) 2 ε m − 4 ∥ g ∥ L ω 2 ( R 2 ) 2 . \int_{\mathrm{zone}}\|\mathcal{B}g(\xi,\cdot)\|_{L^2_\omega(\mathbb{T}^2)}^2\,\mathrm{d}\xi=(2\pi)^2\varepsilon_m^{-4}\|g\|_{L^2_\omega(\mathbb{R}^2)}^2. ∫ zone ∥ B g ( ξ , ⋅ ) ∥ L ω 2 ( T 2 ) 2 d ξ = ( 2 π ) 2 ε m − 4 ∥ g ∥ L ω 2 ( R 2 ) 2 . Proof. For almost every z z z , B g ( ξ , z ) \mathcal{B}g(\xi,z) B g ( ξ , z ) is a complex number of modulus one times the trigonometric series ∑ j ∈ Z 2 exp ( − i ε m ξ ⋅ j ) g ( ε m ( z + j ) ) \sum_{j\in\mathbb{Z}^2}\exp(-i\varepsilon_m\xi\cdot j)g(\varepsilon_m(z+j)) ∑ j ∈ Z 2 exp ( − i ε m ξ ⋅ j ) g ( ε m ( z + j )) of the cell frequency ε m ξ ∈ [ − π , π ] 2 \varepsilon_m\xi\in[-\pi,\pi]^2 ε m ξ ∈ [ − π , π ] 2 , and Parseval’s identity gives ∫ z o n e ∣ B g ( ξ , z ) ∣ 2 d ξ = ( 2 π ) 2 ε m − 2 ∑ j ∣ g ( ε m ( z + j ) ) ∣ 2 \int_{\mathrm{zone}}|\mathcal{B}g(\xi,z)|^2\,\mathrm{d}\xi=(2\pi)^2\varepsilon_m^{-2}\sum_j|g(\varepsilon_m(z+j))|^2 ∫ zone ∣ B g ( ξ , z ) ∣ 2 d ξ = ( 2 π ) 2 ε m − 2 ∑ j ∣ g ( ε m ( z + j )) ∣ 2 ; integrating in z z z gives the identity. □ \square □
We write L 0 \mathcal{L}_0 L 0 for the cell operator (101) with the conductivity K ( t ) \mathbf{K}(t) K ( t ) of the standing hypotheses, at every t ∈ R t \in\mathbb{R} t ∈ R . The next subsection constructs the principal branch as a fixed point of the cell problem forced at all times, which Lemma 12.3 solves with a constant independent of the velocity: the transport is skew and drops out of the energy, and the diffusion damps functions of mean zero at the cell rate. For a Banach space X X X , forcings and solutions with values in X X X are measured over intervals of one cell diffusion time ε m 2 / κ m \varepsilon_m^2/\kappa_m ε m 2 / κ m ,
∥ F ∥ L u l 2 ( R ; X ) : = sup t ∈ R ( κ m ε m 2 ∫ t t + ε m 2 / κ m ∥ F ( t ′ ) ∥ X 2 d t ′ ) 1 / 2 , \|F\|_{L^2_{\mathrm{ul}}(\mathbb{R};X)} := \sup_{t\in\mathbb{R}}\left(\frac{\kappa_m}{\varepsilon_m^2}\int_t^{t+\varepsilon_m^2/\kappa_m}\|F(t')\|_X^2\,\mathrm{d}t'\right)^{1/2}, ∥ F ∥ L ul 2 ( R ; X ) := t ∈ R sup ( ε m 2 κ m ∫ t t + ε m 2 / κ m ∥ F ( t ′ ) ∥ X 2 d t ′ ) 1/2 ,
and the space L ∞ ( R ; L 2 ( T 2 ) ) ∩ L u l 2 ( R ; H 1 ( T 2 ) ) L^\infty(\mathbb{R};L^2(\mathbb{T}^2))\cap L^2_{\mathrm{ul}}(\mathbb{R};H^1(\mathbb{T}^2)) L ∞ ( R ; L 2 ( T 2 )) ∩ L ul 2 ( R ; H 1 ( T 2 )) carries the norm ∥ w ∥ L ∞ ( R ; L 2 ( T 2 ) ) + ∥ ∇ w ∥ L u l 2 ( R ; L 2 ( T 2 ) ) \|w\|_{L^\infty(\mathbb{R};L^2(\mathbb{T}^2))}+\|\nabla w\|_{L^2_{\mathrm{ul}}(\mathbb{R};L^2(\mathbb{T}^2))} ∥ w ∥ L ∞ ( R ; L 2 ( T 2 )) + ∥∇ w ∥ L ul 2 ( R ; L 2 ( T 2 )) .
Lemma 12.3 (Bounded solutions at every time). Let F 0 F^0 F 0 and F F F be measurable on R × T 2 \mathbb{R}\times\mathbb{T}^2 R × T 2 , with values in C \mathbb{C} C and C 2 \mathbb{C}^2 C 2 , with finite L u l 2 ( R ; L 2 ( T 2 ) ) L^2_{\mathrm{ul}}(\mathbb{R};L^2(\mathbb{T}^2)) L ul 2 ( R ; L 2 ( T 2 )) norms. Then the problem
L 0 U = κ m ε m 2 ( F 0 − ( F 0 ) T 2 + ∇ ⋅ F ) in R × T 2 \mathcal{L}_0 U=\frac{\kappa_m}{\varepsilon_m^2}\left(F^0-(F^0)_{\mathbb{T}^2}+\nabla\cdot F\right)\quad\text{in }\mathbb{R}\times\mathbb{T}^2 L 0 U = ε m 2 κ m ( F 0 − ( F 0 ) T 2 + ∇ ⋅ F ) in R × T 2 has exactly one energy solution U ∈ L ∞ ( R ; L 2 ( T 2 ) ) U\in L^\infty(\mathbb{R};L^2(\mathbb{T}^2)) U ∈ L ∞ ( R ; L 2 ( T 2 )) with ( U ( t ) ) T 2 = 0 (U(t))_{\mathbb{T}^2}=0 ( U ( t ) ) T 2 = 0 for every t t t , and
∥ U ∥ L ∞ ( R ; L 2 ( T 2 ) ) + ∥ ∇ U ∥ L u l 2 ( R ; L 2 ( T 2 ) ) ≤ 100 ( ∥ F 0 ∥ L u l 2 ( R ; L 2 ( T 2 ) ) + ∥ F ∥ L u l 2 ( R ; L 2 ( T 2 ) ) ) . \|U\|_{L^\infty(\mathbb{R};L^2(\mathbb{T}^2))}+\|\nabla U\|_{L^2_{\mathrm{ul}}(\mathbb{R};L^2(\mathbb{T}^2))}\le100\left(\|F^0\|_{L^2_{\mathrm{ul}}(\mathbb{R};L^2(\mathbb{T}^2))}+\|F\|_{L^2_{\mathrm{ul}}(\mathbb{R};L^2(\mathbb{T}^2))}\right). ∥ U ∥ L ∞ ( R ; L 2 ( T 2 )) + ∥∇ U ∥ L ul 2 ( R ; L 2 ( T 2 )) ≤ 100 ( ∥ F 0 ∥ L ul 2 ( R ; L 2 ( T 2 )) + ∥ F ∥ L ul 2 ( R ; L 2 ( T 2 )) ) .
Proof. We measure time in units of ε m 2 / κ m \varepsilon_m^2/\kappa_m ε m 2 / κ m , so that the coefficients κ m / ε m 2 \kappa_m/\varepsilon_m^2 κ m / ε m 2 disappear. Testing with U ‾ \overline{U} U and using K ≥ 1 2 I \mathbf{K}\ge\frac{1}{2}\mathbf{I} K ≥ 2 1 I and Poincaré’s inequality ∥ ∇ U ∥ L 2 ( T 2 ) ≥ 2 π ∥ U ∥ L 2 ( T 2 ) \|\nabla U\|_{L^2(\mathbb{T}^2)}\ge2\pi\|U\|_{L^2(\mathbb{T}^2)} ∥∇ U ∥ L 2 ( T 2 ) ≥ 2 π ∥ U ∥ L 2 ( T 2 ) for mean-zero U U U , we obtain
d d t ∥ U ∥ L 2 ( T 2 ) 2 + π 2 ∥ U ∥ L 2 ( T 2 ) 2 + 1 4 ∥ ∇ U ∥ L 2 ( T 2 ) 2 ≤ 4 ( ∥ F 0 ∥ L 2 ( T 2 ) 2 + ∥ F ∥ L 2 ( T 2 ) 2 ) . \frac{\mathrm{d}}{\mathrm{d}t}\|U\|_{L^2(\mathbb{T}^2)}^2+\pi^2\|U\|_{L^2(\mathbb{T}^2)}^2+\frac{1}{4}\|\nabla U\|_{L^2(\mathbb{T}^2)}^2\le4\left(\|F^0\|_{L^2(\mathbb{T}^2)}^2+\|F\|_{L^2(\mathbb{T}^2)}^2\right). d t d ∥ U ∥ L 2 ( T 2 ) 2 + π 2 ∥ U ∥ L 2 ( T 2 ) 2 + 4 1 ∥∇ U ∥ L 2 ( T 2 ) 2 ≤ 4 ( ∥ F 0 ∥ L 2 ( T 2 ) 2 + ∥ F ∥ L 2 ( T 2 ) 2 ) .
Summing over intervals of unit length, we find that every energy solution on [ t ′ , ∞ ) [t',\infty) [ t ′ , ∞ ) satisfies
∥ U ( t ) ∥ L 2 ( T 2 ) ≤ e − π 2 ( t − t ′ ) / 2 ∥ U ( t ′ ) ∥ L 2 ( T 2 ) + 2.01 ( ∥ F 0 ∥ L u l 2 ( R ; L 2 ( T 2 ) ) 2 + ∥ F ∥ L u l 2 ( R ; L 2 ( T 2 ) ) 2 ) 1 / 2 , \|U(t)\|_{L^2(\mathbb{T}^2)}\le e^{-\pi^2(t-t')/2}\|U(t')\|_{L^2(\mathbb{T}^2)}+2.01\left(\|F^0\|_{L^2_{\mathrm{ul}}(\mathbb{R};L^2(\mathbb{T}^2))}^2+\|F\|_{L^2_{\mathrm{ul}}(\mathbb{R};L^2(\mathbb{T}^2))}^2\right)^{1/2}, ∥ U ( t ) ∥ L 2 ( T 2 ) ≤ e − π 2 ( t − t ′ ) /2 ∥ U ( t ′ ) ∥ L 2 ( T 2 ) + 2.01 ( ∥ F 0 ∥ L ul 2 ( R ; L 2 ( T 2 )) 2 + ∥ F ∥ L ul 2 ( R ; L 2 ( T 2 )) 2 ) 1/2 ,
and the first display, integrated over an interval of unit length, then bounds ∇ U \nabla U ∇ U . The solutions from zero at the times t ′ = − 1 , − 2 , … t'=-1,-2,\ldots t ′ = − 1 , − 2 , … converge to one with the bound, and two bounded solutions coincide (let t ′ → − ∞ t'\to-\infty t ′ → − ∞ ).
The principal branch at small frequencies
At ξ = 0 \xi=0 ξ = 0 the constants solve the fibre equation (126) , and every other solution approaches its cell mean at the cell rate. We show that this splitting persists at small frequencies. As the coefficients depend on time, there is no eigenvalue to continue. We seek instead a cell function R ξ ( t ) \mathcal{R}_\xi(t) R ξ ( t ) of mean one and a scalar rate Z ξ ( t ) \mathcal{Z}_\xi(t) Z ξ ( t ) such that R ξ \mathcal{R}_\xi R ξ , multiplied by a scalar function of time that grows at the rate Z ξ \mathcal{Z}_\xi Z ξ , solves (126) ; in the plane this is a plane wave corrected at the cell scale. Averaging the fibre equation over T 2 \mathbb{T}^2 T 2 , with the velocity of mean zero, gives the symbol
Z ξ = − κ m ξ ⋅ K ξ − i ξ ⋅ ( ( ∇ ⊥ ψ m ) ( t , ε m ⋅ ) ( R ξ − 1 ) ) T 2 , (131) \mathcal{Z}_\xi=-\kappa_m\,\xi\cdot\mathbf{K}\xi-\mathrm{i}\,\xi\cdot\left((\nabla^\perp\psi_m)(t,\varepsilon_m\cdot)(\mathcal{R}_\xi-1)\right)_{\mathbb{T}^2}, \tag*{(131)} Z ξ = − κ m ξ ⋅ K ξ − i ξ ⋅ ( ( ∇ ⊥ ψ m ) ( t , ε m ⋅ ) ( R ξ − 1 ) ) T 2 , (131)
and its part of mean zero is the graph equation
L 0 ( R ξ − 1 ) = 2 i κ m ε m ∇ ⋅ ( K ξ ( R ξ − 1 ) ) − i ( ξ ⋅ ( ∇ ⊥ ψ m ) ( t , ε m ⋅ ) R ξ − ( ξ ⋅ ( ∇ ⊥ ψ m ) ( t , ε m ⋅ ) ( R ξ − 1 ) ) T 2 ) + i ( R ξ − 1 ) ( ξ ⋅ ( ∇ ⊥ ψ m ) ( t , ε m ⋅ ) ( R ξ − 1 ) ) T 2 . (132) \mathcal{L}_0(\mathcal{R}_\xi-1)=\frac{2\mathrm{i}\kappa_m}{\varepsilon_m}\nabla\cdot\left(\mathbf{K}\xi(\mathcal{R}_\xi-1)\right)
-\mathrm{i}\left(\xi\cdot(\nabla^\perp\psi_m)(t,\varepsilon_m\cdot)\mathcal{R}_\xi-\left(\xi\cdot(\nabla^\perp\psi_m)(t,\varepsilon_m\cdot)(\mathcal{R}_\xi-1)\right)_{\mathbb{T}^2}\right)
+\mathrm{i}(\mathcal{R}_\xi-1)\left(\xi\cdot(\nabla^\perp\psi_m)(t,\varepsilon_m\cdot)(\mathcal{R}_\xi-1)\right)_{\mathbb{T}^2}. \tag*{(132)} L 0 ( R ξ − 1 ) = ε m 2 i κ m ∇ ⋅ ( K ξ ( R ξ − 1 ) ) − i ( ξ ⋅ ( ∇ ⊥ ψ m ) ( t , ε m ⋅ ) R ξ − ( ξ ⋅ ( ∇ ⊥ ψ m ) ( t , ε m ⋅ ) ( R ξ − 1 ) ) T 2 ) + i ( R ξ − 1 ) ( ξ ⋅ ( ∇ ⊥ ψ m ) ( t , ε m ⋅ ) ( R ξ − 1 ) ) T 2 . (132)
This is the cell problem of Lemma 12.3 for R ξ − 1 R_{\xi}-1 R ξ − 1 , with a forcing that is small in cell units when ε ˉ m ∣ ξ ∣ \bar{\varepsilon}_{m}|\xi| ε ˉ m ∣ ξ ∣ is small: whatever the Péclet number (the velocity in cell units), the transport term on a mode of frequency ξ \xi ξ has relative size
ε m 2 κ m ∥ ∇ ⊥ ψ m ∥ L ∞ ( R × R 2 ) ∣ ξ ∣ ≤ 2 π ε m β + 1 ∣ ξ ∣ κ m ≤ C ε ˉ m ∣ ξ ∣ (133) \frac{\varepsilon_{m}^{2}}{\kappa_{m}}\|\nabla^{\perp}\psi_{m}\|_{L^{\infty}(\mathbb{R}\times\mathbb{R}^{2})}|\xi| \le\frac{2\pi\varepsilon_{m}^{\beta+1}|\xi|}{\kappa_{m}} \le C\bar{\varepsilon}_{m}|\xi| \tag*{(133)} κ m ε m 2 ∥ ∇ ⊥ ψ m ∥ L ∞ ( R × R 2 ) ∣ ξ ∣ ≤ κ m 2 π ε m β + 1 ∣ ξ ∣ ≤ C ε ˉ m ∣ ξ ∣ (133)
with an absolute C C C (Step 1 of the proof of Proposition 12.4). So the graph equation has exactly one small solution of mean zero, and R ξ R_{\xi} R ξ is the principal branch , the branch that continues the constants. By (a) below it differs from 1 1 1 by the relative size of the first corrector of a plane wave, and by (b) the symbol is at most of the size of the rate of the homogenized equation at the frequency ξ \xi ξ .
Proposition 12.4 (The principal branch). There exist absolute constants c b r ∈ ( 0 , 1 ] c_{\mathrm{br}}\in(0,1] c br ∈ ( 0 , 1 ] and C b r C_{\mathrm{br}} C br such that, under the standing hypotheses, there exist R ξ ( t ) ∈ L 2 ( T 2 ) R_{\xi}(t)\in L^{2}(\mathbb{T}^{2}) R ξ ( t ) ∈ L 2 ( T 2 ) and Z ξ ( t ) ∈ C Z_{\xi}(t)\in\mathbb{C} Z ξ ( t ) ∈ C , for t ∈ R t\in\mathbb{R} t ∈ R and ξ ∈ C 2 \xi\in\mathbb{C}^{2} ξ ∈ C 2 with ε ˉ m ∣ ξ ∣ < c b r \bar{\varepsilon}_{m}|\xi|<c_{\mathrm{br}} ε ˉ m ∣ ξ ∣ < c br , power series in ξ \xi ξ , such that for every such ξ \xi ξ the following hold.
(a) For every t ∈ R t\in\mathbb{R} t ∈ R , ( R ξ ( t ) ) T 2 = 1 (R_{\xi}(t))_{\mathbb{T}^{2}}=1 ( R ξ ( t ) ) T 2 = 1 , and
∥ R ξ − 1 ∥ L ∞ ( R ; L 2 ( T 2 ) ) ≤ C b r ε ˉ m ∣ ξ ∣ . \|R_{\xi}-1\|_{L^{\infty}(\mathbb{R};L^{2}(\mathbb{T}^{2}))}\le C_{\mathrm{br}}\bar{\varepsilon}_{m}|\xi|. ∥ R ξ − 1 ∥ L ∞ ( R ; L 2 ( T 2 )) ≤ C br ε ˉ m ∣ ξ ∣. (b) Z ξ = ∑ j ≥ 2 Z ξ , j Z_{\xi}=\sum_{j\ge2}Z_{\xi,j} Z ξ = ∑ j ≥ 2 Z ξ , j , where Z ξ , j Z_{\xi,j} Z ξ , j is homogeneous of degree j j j in ξ \xi ξ , and for every j ≥ 2 j\ge2 j ≥ 2
∥ Z ξ , j ∥ L ∞ ( R ) ≤ C b r κ m ε m 2 ( 2 ε ˉ m ∣ ξ ∣ c b r ) j . \|Z_{\xi,j}\|_{L^{\infty}(\mathbb{R})}\le C_{\mathrm{br}}\frac{\kappa_{m}}{\varepsilon_{m}^{2}}\left(\frac{2\bar{\varepsilon}_{m}|\xi|}{c_{\mathrm{br}}}\right)^{j}. ∥ Z ξ , j ∥ L ∞ ( R ) ≤ C br ε m 2 κ m ( c br 2 ε ˉ m ∣ ξ ∣ ) j . (c) For every t ′ ∈ R t'\in\mathbb{R} t ′ ∈ R , the function t ↦ R ξ ( t ) exp ( ∫ t ′ t Z ξ ( t ′ ′ ) d t ′ ′ ) t\mapsto R_{\xi}(t)\exp\left(\int_{t'}^{t}Z_{\xi}(t'')\,\mathrm{d}t''\right) t ↦ R ξ ( t ) exp ( ∫ t ′ t Z ξ ( t ′′ ) d t ′′ ) solves the fibre equation (12.5) on [ t ′ , ∞ ) [t',\infty) [ t ′ , ∞ ) .
(d) For every ξ ∈ R 2 \xi\in\mathbb{R}^{2} ξ ∈ R 2 with ε ˉ m ∣ ξ ∣ < c b r \bar{\varepsilon}_{m}|\xi|<c_{\mathrm{br}} ε ˉ m ∣ ξ ∣ < c br and every t ′ ≤ t t'\le t t ′ ≤ t ,
∣ exp ( ∫ t ′ t Z ξ ( t ′ ′ ) d t ′ ′ ) ∣ ≤ C b r . \left|\exp\left(\int_{t'}^{t}Z_{\xi}(t'')\,\mathrm{d}t''\right)\right|\le C_{\mathrm{br}}. exp ( ∫ t ′ t Z ξ ( t ′′ ) d t ′′ ) ≤ C br . (e) R ξ − 1 R_{\xi}-1 R ξ − 1 is the only solution of the graph equation (12.11) that has mean zero at every time and norm at most one in L ∞ ( R ; L 2 ( T 2 ) ) ∩ L u l 2 ( R ; H 1 ( T 2 ) ) L^{\infty}(\mathbb{R};L^{2}(\mathbb{T}^{2}))\cap L^{2}_{\mathrm{ul}}(\mathbb{R};H^{1}(\mathbb{T}^{2})) L ∞ ( R ; L 2 ( T 2 )) ∩ L ul 2 ( R ; H 1 ( T 2 )) , and Z ξ Z_{\xi} Z ξ is given by (131) .
Proof. Step 1: The branch and the symbol. As ∥ ∇ ⊥ ψ m ∥ L ∞ ( R × R 2 ) ≤ 2 π ε m β − 1 \|\nabla^{\perp}\psi_{m}\|_{L^{\infty}(\mathbb{R}\times\mathbb{R}^{2})}\le2\pi\varepsilon_{m}^{\beta-1} ∥ ∇ ⊥ ψ m ∥ L ∞ ( R × R 2 ) ≤ 2 π ε m β − 1 and ε m 2 β / κ m ≤ 20 κ m − 1 \varepsilon_{m}^{2\beta}/\kappa_{m}\le20\kappa_{m-1} ε m 2 β / κ m ≤ 20 κ m − 1 (Lemma 10.1 (c)), the constant of (133) is absolute. The right side of the graph equation (132) is the cell rate times F 0 − ( F 0 ) T 2 + ∇ ⋅ F F^{0}-(F^{0})_{\mathbb{T}^{2}}+\nabla\cdot F F 0 − ( F 0 ) T 2 + ∇ ⋅ F , where F = 2 i ε m K ξ ( R ξ − 1 ) F=2i\varepsilon_{m}K\xi(R_{\xi}-1) F = 2 i ε m K ξ ( R ξ − 1 ) and F 0 F^{0} F 0 is ε m 2 / κ m \varepsilon_{m}^{2}/\kappa_{m} ε m 2 / κ m times the transport terms, with ∥ F ∥ L 2 ( T 2 ) ≤ 4 ε ˉ m ∣ ξ ∣ ∥ R ξ − 1 ∥ L 2 ( T 2 ) \|F\|_{L^{2}(\mathbb{T}^{2})}\le4\bar{\varepsilon}_{m}|\xi|\|R_{\xi}-1\|_{L^{2}(\mathbb{T}^{2})} ∥ F ∥ L 2 ( T 2 ) ≤ 4 ε ˉ m ∣ ξ ∣∥ R ξ − 1 ∥ L 2 ( T 2 ) and ∥ F 0 ∥ L 2 ( T 2 ) ≤ C ε ˉ m ∣ ξ ∣ ( 1 + ∥ R ξ − 1 ∥ L 2 ( T 2 ) 2 ) \|F^{0}\|_{L^{2}(\mathbb{T}^{2})}\le C\bar{\varepsilon}_{m}|\xi|(1+\|R_{\xi}-1\|_{L^{2}(\mathbb{T}^{2})}^{2}) ∥ F 0 ∥ L 2 ( T 2 ) ≤ C ε ˉ m ∣ ξ ∣ ( 1 + ∥ R ξ − 1 ∥ L 2 ( T 2 ) 2 ) , by (133) and ε m ∣ ξ ∣ ≤ ε ˉ m ∣ ξ ∣ \varepsilon_{m}|\xi|\le\bar{\varepsilon}_{m}|\xi| ε m ∣ ξ ∣ ≤ ε ˉ m ∣ ξ ∣ . So Lemma 12.3 makes the map that sends a mean-zero function, put in place of R ξ − 1 R_{\xi}-1 R ξ − 1 on the right side, to the bounded mean-zero solution of (132) a contraction of the unit ball of L ∞ ( R ; L 2 ( T 2 ) ) ∩ L u l 2 ( R ; H 1 ( T 2 ) ) L^{\infty}(\mathbb{R};L^{2}(\mathbb{T}^{2}))\cap L^{2}_{\mathrm{ul}}(\mathbb{R};H^{1}(\mathbb{T}^{2})) L ∞ ( R ; L 2 ( T 2 )) ∩ L ul 2 ( R ; H 1 ( T 2 )) when ε ˉ m ∣ ξ ∣ < c b r \bar{\varepsilon}_{m}|\xi|<c_{\mathrm{br}} ε ˉ m ∣ ξ ∣ < c br , with an absolute c b r ≤ 1 2 c_{\mathrm{br}}\le\frac{1}{2} c br ≤ 2 1 ; its fixed point has norm at most C ε ˉ m ∣ ξ ∣ C\bar{\varepsilon}_{m}|\xi| C ε ˉ m ∣ ξ ∣ . The map is a polynomial in ξ \xi ξ and in the function, so its Picard iterates are polynomials in ξ \xi ξ and converge uniformly on the complex ball ε ˉ m ∣ ξ ∣ < c b r \bar{\varepsilon}_{m}|\xi|<c_{\mathrm{br}} ε ˉ m ∣ ξ ∣ < c br : R ξ R_{\xi} R ξ and Z ξ Z_{\xi} Z ξ are power series in ξ \xi ξ . For ε ˉ m ∣ ξ ∣ ≤ c b r / 2 \bar{\varepsilon}_{m}|\xi|\le c_{\mathrm{br}}/2 ε ˉ m ∣ ξ ∣ ≤ c br /2 , ∣ Z ξ ∣ |Z_{\xi}| ∣ Z ξ ∣ is at most an absolute constant times κ m / ε m 2 \kappa_{m}/\varepsilon_{m}^{2} κ m / ε m 2 , by (131) and (133) ; we take C b r ≥ 2 C_{\mathrm{br}}\ge2 C br ≥ 2 , at least this constant and the constant of the bound on the fixed point, which gives the bound of (a). Cauchy’s estimate on the ball of radius c b r / 2 c_{\mathrm{br}}/2 c br /2 bounds the term of degree j j j by ( 2 ε ˉ m ∣ ξ ∣ / c b r ) j (2\bar{\varepsilon}_{m}|\xi|/c_{\mathrm{br}})^{j} ( 2 ε ˉ m ∣ ξ ∣/ c br ) j times the supremum, which gives (b); the terms of degree 0 and 1 vanish, as R ξ − 1 R_{\xi}-1 R ξ − 1 is of degree at least one in (131) . A mean-zero solution of (132) in that unit @references 重庆时时彩彩 ball is a fixed point of the map, by the uniqueness in Lemma 12.3 ; this is (e), and (c) and the mean in (a) hold by construction.
Step 2: Real frequencies. By (127) , and by (a) with c b r ≤ 1 2 c_{\mathrm{br}} \le\frac{1}{2} c br ≤ 2 1 and C b r ≥ 2 C_{\mathrm{br}} \ge2 C br ≥ 2 , respectively,
∥ R ξ ( t ) ∥ L 2 ( T 2 ) ∣ exp ( ∫ t ′ t Z ξ ( t ′ ′ ) d t ′ ′ ) ∣ ≤ ∥ R ξ ( t ′ ) ∥ L 2 ( T 2 ) , 1 ≤ ∥ R ξ ∥ L 2 ( T 2 ) ≤ 1 + C b r ε ˉ m ∣ ξ ∣ ≤ C b r , \left\|\mathcal{R}_{\xi}(t)\right\|_{L^{2}(\mathbb{T}^{2})}\left|\exp\left(\int_{t'}^{t}\mathcal{Z}_{\xi}(t'')\,\mathrm{d}t''\right)\right| \le\left\|\mathcal{R}_{\xi}(t')\right\|_{L^{2}(\mathbb{T}^{2})}, \qquad1 \le\left\|\mathcal{R}_{\xi}\right\|_{L^{2}(\mathbb{T}^{2})} \le1+C_{\mathrm{br}}\bar{\varepsilon}_{m}|\xi| \le C_{\mathrm{br}}, ∥ R ξ ( t ) ∥ L 2 ( T 2 ) exp ( ∫ t ′ t Z ξ ( t ′′ ) d t ′′ ) ≤ ∥ R ξ ( t ′ ) ∥ L 2 ( T 2 ) , 1 ≤ ∥ R ξ ∥ L 2 ( T 2 ) ≤ 1 + C br ε ˉ m ∣ ξ ∣ ≤ C br ,
which gives (d).
Dissipation of the branch: the averaged flux
The quadratic part of the symbol is given by the averaged flux of Part I, up to an error small relative to κ m − 1 ∣ ξ ∣ 2 \kappa_{m-1}|\xi|^{2} κ m − 1 ∣ ξ ∣ 2 , from the conductivity and from switching the pulses on and off, because the first-order part of the branch is the corrector of § 5.2 . At each time the averaged flux enhances diffusion only along the flow of the active pulse; a block, however, carries pulses of both orientations, and over a block the averaged flux integrates to at least a constant times ρ ( m ) 2 \rho(m)^{2} ρ ( m ) 2 in every direction. So over one block the branch is damped by the Gaussian exp ( − c d i s ρ ( m ) 2 ∣ ξ ∣ 2 ) \exp(-c_{\mathrm{dis}}\rho(m)^{2}|\xi|^{2}) exp ( − c dis ρ ( m ) 2 ∣ ξ ∣ 2 ) , by (b) below: this is the homogenized equation of level m − 1 m-1 m − 1 seen over the time h m h_m h m . Item (c) also covers intervals that are not complete blocks: the symbol is at most π 2 / 16 \pi^{2}/16 π 2 /16 times the cell rate, so the branch changes more slowly than diffusion damps functions of mean zero on the cell, and frequency derivatives of its integral, at the scale 1 / ρ ( m ) 1/\rho(m) 1/ ρ ( m ) , cost only powers of 1 + ρ ( m ) ∣ ξ ∣ 1+\rho(m)|\xi| 1 + ρ ( m ) ∣ ξ ∣ . Below, the core of a pulse k k k is [ ( k − 1 3 ) τ m , ( k + 1 3 ) τ m ] [(k-\frac{1}{3})\tau_m,(k+\frac{1}{3})\tau_m] [( k − 3 1 ) τ m , ( k + 3 1 ) τ m ] ; it is complete if the slow cutoff of its block is 1 on it.
Proposition 12.5 (Dissipation of the branch). There exist Λ 0 ( α , s , q , ζ ) ≥ 128 \Lambda_{0}(\alpha,s,q,\zeta) \ge128 Λ 0 ( α , s , q , ζ ) ≥ 128 , C ( α , s , q , ζ ) C(\alpha,s,q,\zeta) C ( α , s , q , ζ ) and absolute constants c l o ∈ ( 0 , c b r / 8 ] c_{\mathrm{lo}} \in(0,c_{\mathrm{br}}/8] c lo ∈ ( 0 , c br /8 ] and c d i s > 0 c_{\mathrm{dis}}>0 c dis > 0 such that, under the standing hypotheses with this Λ 0 \Lambda_{0} Λ 0 , the following hold.
(a) For every ξ ∈ C 2 \xi\in\mathbb{C}^{2} ξ ∈ C 2 ,
∥ Z ξ , 2 + ξ ⋅ J m κ m ξ ∥ L ∞ ( R ) ≤ C ε ˉ m − 1 δ κ m − 1 ∣ ξ ∣ 2 . \left\|\mathcal{Z}_{\xi,2}+\xi\cdot J^{\kappa_m}_{m}\xi\right\|_{L^{\infty}(\mathbb{R})} \le C\bar{\varepsilon}_{m-1}^{\delta}\kappa_{m-1}|\xi|^{2}. ∥ Z ξ , 2 + ξ ⋅ J m κ m ξ ∥ L ∞ ( R ) ≤ C ε ˉ m − 1 δ κ m − 1 ∣ ξ ∣ 2 . (b) For every l ′ ∈ Z l' \in\mathbb{Z} l ′ ∈ Z , every ξ ∈ R 2 \xi\in\mathbb{R}^{2} ξ ∈ R 2 with ε ˉ m ∣ ξ ∣ ≤ 2 c l o \bar{\varepsilon}_{m}|\xi| \le2c_{\mathrm{lo}} ε ˉ m ∣ ξ ∣ ≤ 2 c lo and every integer N ≥ 2 N \ge2 N ≥ 2 , on the renewal block [ t − , t + ] : = [ ( l ′ − 1 2 ) h m , ( l ′ + 1 2 ) h m ] [t_{-},t_{+}]:=[(l'-\frac{1}{2})h_m,(l'+\frac{1}{2})h_m] [ t − , t + ] := [( l ′ − 2 1 ) h m , ( l ′ + 2 1 ) h m ] ,
Re ∫ t − t + Z ξ ( t ) d t ≤ − c d i s ρ ( m ) 2 ∣ ξ ∣ 2 , Re ∫ t − t + ∑ j = 2 N Z ξ , j ( t ) d t ≤ − c d i s ρ ( m ) 2 ∣ ξ ∣ 2 . \operatorname{Re}\int_{t_{-}}^{t_{+}}\mathcal{Z}_{\xi}(t)\,\mathrm{d}t \le-c_{\mathrm{dis}}\rho(m)^{2}|\xi|^{2}, \qquad\operatorname{Re}\int_{t_{-}}^{t_{+}}\sum_{j=2}^{N}\mathcal{Z}_{\xi,j}(t)\,\mathrm{d}t \le-c_{\mathrm{dis}}\rho(m)^{2}|\xi|^{2}. Re ∫ t − t + Z ξ ( t ) d t ≤ − c dis ρ ( m ) 2 ∣ ξ ∣ 2 , Re ∫ t − t + j = 2 ∑ N Z ξ , j ( t ) d t ≤ − c dis ρ ( m ) 2 ∣ ξ ∣ 2 . (c) For every ξ ∈ C 2 \xi\in\mathbb{C}^{2} ξ ∈ C 2 with ε ˉ m ∣ ξ ∣ ≤ 2 c l o \bar{\varepsilon}_{m}|\xi| \le2c_{\mathrm{lo}} ε ˉ m ∣ ξ ∣ ≤ 2 c lo ,
C b r ε ˉ m ∣ ξ ∣ ≤ 1 3 , ∥ Z ξ ∥ L ∞ ( R ) ≤ π 2 κ m 16 ε ˉ m 2 . C_{\mathrm{br}}\bar{\varepsilon}_{m}|\xi| \le\frac{1}{3}, \qquad\left\|\mathcal{Z}_{\xi}\right\|_{L^{\infty}(\mathbb{R})} \le\frac{\pi^{2}\kappa_m}{16\bar{\varepsilon}_{m}^{2}}. C br ε ˉ m ∣ ξ ∣ ≤ 3 1 , ∥ Z ξ ∥ L ∞ ( R ) ≤ 16 ε ˉ m 2 π 2 κ m . For every interval [ t ′ , t ′ ′ ] [t',t''] [ t ′ , t ′′ ] of length at most 2 h m 2h_m 2 h m and every ξ ∈ R 2 \xi\in\mathbb{R}^{2} ξ ∈ R 2 with ε ˉ m ∣ ξ ∣ ≤ 2 c l o \bar{\varepsilon}_{m}|\xi| \le2c_{\mathrm{lo}} ε ˉ m ∣ ξ ∣ ≤ 2 c lo , with ∇ ξ j \nabla_{\xi}^{j} ∇ ξ j the array of the derivatives of order j j j and C ( j ) C(j) C ( j ) depending only on j j j , the following hold for every integer j ≥ 1 j \ge1 j ≥ 1 :
∣ ρ ( m ) − j ∇ ξ j ∫ t ′ t ′ ′ Z ξ ( t ) d t ∣ ≤ { C ( 1 ) ρ ( m ) ∣ ξ ∣ , j = 1 , C ( j ) , j ≥ 2 , \left|\rho(m)^{-j}\nabla_{\xi}^{j}\int_{t'}^{t''}\mathcal{Z}_{\xi}(t)\,\mathrm{d}t\right| \le
\begin{cases}
C(1)\rho(m)|\xi|, & j=1,\\
C(j), & j\ge2,
\end{cases} ρ ( m ) − j ∇ ξ j ∫ t ′ t ′′ Z ξ ( t ) d t ≤ { C ( 1 ) ρ ( m ) ∣ ξ ∣ , C ( j ) , j = 1 , j ≥ 2 , and
∣ ρ ( m ) − j ∇ ξ j exp ( ± ∫ t ′ t ′ ′ Z ξ ( t ) d t ) ∣ ≤ C ( j ) ( 1 + ρ ( m ) ∣ ξ ∣ ) j ∣ exp ( ± ∫ t ′ t ′ ′ Z ξ ( t ) d t ) ∣ . \left|\rho(m)^{-j}\nabla_{\xi}^{j}\exp\left(\pm\int_{t'}^{t''}\mathcal{Z}_{\xi}(t)\,\mathrm{d}t\right)\right| \le C(j)(1+\rho(m)|\xi|)^{j}\left|\exp\left(\pm\int_{t'}^{t''}\mathcal{Z}_{\xi}(t)\,\mathrm{d}t\right)\right|. ρ ( m ) − j ∇ ξ j exp ( ± ∫ t ′ t ′′ Z ξ ( t ) d t ) ≤ C ( j ) ( 1 + ρ ( m ) ∣ ξ ∣ ) j exp ( ± ∫ t ′ t ′′ Z ξ ( t ) d t ) .
Proof. Step 1: The quadratic part. The solution operator of Lemma 12.3 is linear and independent of ξ \xi ξ . So the term R ξ , 1 \mathcal{R}_{\xi,1} R ξ , 1 of degree one in ξ \xi ξ of the branch is the bounded mean-zero solution of (132) with R ξ , 1 \mathcal{R}_{\xi,1} R ξ , 1 in place of R ξ − 1 \mathcal{R}_{\xi}-1 R ξ − 1 on the left side and with the part of degree one of the right side, − i ξ ⋅ ( ∇ ⊥ ψ m ) ( t , ε m z ) -\mathrm{i}\xi\cdot(\nabla^{\perp}\psi_m)(t,\varepsilon_m z) − i ξ ⋅ ( ∇ ⊥ ψ m ) ( t , ε m z ) , on the right. By (131) ,
Z ξ , 2 = − κ m ξ ⋅ K ( t ) ξ − i ( ξ ⋅ ( ∇ ⊥ ψ m ) ( t , ε m ⋅ ) R ξ , 1 ) T 2 . (134) Z_{\xi,2}=-\kappa_m\xi\cdot\mathbf{K}(t)\xi-\mathrm{i}\left(\xi\cdot(\nabla^{\perp}\psi_m)(t,\varepsilon_m\cdot)\mathcal{R}_{\xi,1}\right)_{\mathbb{T}^2}. \tag*{(134)} Z ξ , 2 = − κ m ξ ⋅ K ( t ) ξ − i ( ξ ⋅ ( ∇ ⊥ ψ m ) ( t , ε m ⋅ ) R ξ , 1 ) T 2 . (134)
Step 2: Comparison with the averaged flux. In this step C C C depends only on α \alpha α , s s s , q q q , ζ \zeta ζ , as in (a). We use the glued corrector χ m κ m \chi_m^{\kappa_m} χ m κ m of Lemma 10.2(b). For e ∈ R 2 e\in\mathbb{R}^2 e ∈ R 2 , since ( ψ m ) T 2 = 0 (\psi_m)_{\mathbb{T}^2}=0 ( ψ m ) T 2 = 0 , ( ∇ ( e ⋅ χ m κ m ) ) T 2 = 0 (\nabla(e\cdot\chi_m^{\kappa_m}))_{\mathbb{T}^2}=0 ( ∇ ( e ⋅ χ m κ m ) ) T 2 = 0 and ∇ ⊥ ψ m = ( 0 − 1 1 0 ) ∇ ψ m \nabla^{\perp}\psi_m=\begin{pmatrix}0&-1\\1&0\end{pmatrix}\nabla\psi_m ∇ ⊥ ψ m = ( 0 1 − 1 0 ) ∇ ψ m , an integration by parts in (97) gives the exact identity
J m κ m ( t ) e = κ m e − ∫ T 2 ∇ ⊥ ψ m ( t , y ) ( e ⋅ χ m κ m ( t , y ) ) d y , (135) \mathbf{J}_m^{\kappa_m}(t)e=\kappa_m e-\int_{\mathbb{T}^2}\nabla^{\perp}\psi_m(t,y)\left(e\cdot\chi_m^{\kappa_m}(t,y)\right)\,dy, \tag*{(135)} J m κ m ( t ) e = κ m e − ∫ T 2 ∇ ⊥ ψ m ( t , y ) ( e ⋅ χ m κ m ( t , y ) ) d y , (135)
so that − ξ ⋅ J m κ m ξ = − κ m ξ ⋅ ξ + ∫ T 2 ( ξ ⋅ ∇ ⊥ ψ m ) ( ξ ⋅ χ m κ m ) d y -\xi\cdot\mathbf{J}_m^{\kappa_m}\xi=-\kappa_m\xi\cdot\xi+\int_{\mathbb{T}^2}(\xi\cdot\nabla^{\perp}\psi_m)(\xi\cdot\chi_m^{\kappa_m})\,dy − ξ ⋅ J m κ m ξ = − κ m ξ ⋅ ξ + ∫ T 2 ( ξ ⋅ ∇ ⊥ ψ m ) ( ξ ⋅ χ m κ m ) d y for every ξ ∈ C 2 \xi\in\mathbb{C}^2 ξ ∈ C 2 , by bilinearity. By (98) , read in the fast variable, i ξ ⋅ χ m κ m ( t , ε m z ) \mathrm{i}\xi\cdot\chi_m^{\kappa_m}(t,\varepsilon_m z) i ξ ⋅ χ m κ m ( t , ε m z ) solves the equation of R ξ , 1 \mathcal{R}_{\xi,1} R ξ , 1 at K = I \mathbf{K}=I K = I up to the switching residual. Hence R ξ , 1 − i ξ ⋅ χ m κ m ( t , ε m z ) \mathcal{R}_{\xi,1}-\mathrm{i}\xi\cdot\chi_m^{\kappa_m}(t,\varepsilon_m z) R ξ , 1 − i ξ ⋅ χ m κ m ( t , ε m z ) is the bounded mean-zero solution of Lemma 12.3 with F 0 = − ( ε m 2 / κ m ) i ξ ⋅ ∑ k H m , k ′ χ m , k κ m ( t , ε m z ) F^0=-(\varepsilon_m^2/\kappa_m)\mathrm{i}\xi\cdot\sum_k H'_{m,k}\chi_{m,k}^{\kappa_m}(t,\varepsilon_m z) F 0 = − ( ε m 2 / κ m ) i ξ ⋅ ∑ k H m , k ′ χ m , k κ m ( t , ε m z ) , with H m , k ′ H'_{m,k} H m , k ′ the derivative of H m , k H_{m,k} H m , k in t t t , and F = ( K − I ) ∇ z ( i ξ ⋅ χ m κ m ( t , ε m z ) ) F=(\mathbf{K}-I)\nabla_z(\mathrm{i}\xi\cdot\chi_m^{\kappa_m}(t,\varepsilon_m z)) F = ( K − I ) ∇ z ( i ξ ⋅ χ m κ m ( t , ε m z )) . The gradient in z z z of χ m κ m ( t , ε m z ) \chi_m^{\kappa_m}(t,\varepsilon_m z) χ m κ m ( t , ε m z ) is at most ε m β + 1 / κ m \varepsilon_m^{\beta+1}/\kappa_m ε m β + 1 / κ m , by (94) , (92) and ∑ k H m , k = 1 \sum_k H_{m,k}=1 ∑ k H m , k = 1 ; with (98) and Lemma 12.3 ,
∥ R ξ , 1 − i ξ ⋅ χ m κ m ( ⋅ , ε m ⋅ ) ∥ L ∞ ( R ; L 2 ( T 2 ) ) ≤ C ∣ ξ ∣ ε m β + 1 κ m ( ε m 2 κ m τ m + ε m − 1 δ ) . \left\|\mathcal{R}_{\xi,1}-\mathrm{i}\xi\cdot\chi_m^{\kappa_m}(\cdot,\varepsilon_m\cdot)\right\|_{L^\infty(\mathbb{R};L^2(\mathbb{T}^2))}\le C|\xi|\frac{\varepsilon_m^{\beta+1}}{\kappa_m}\left(\frac{\varepsilon_m^2}{\kappa_m\tau_m}+\varepsilon_{m-1}^{\delta}\right). ∥ R ξ , 1 − i ξ ⋅ χ m κ m ( ⋅ , ε m ⋅ ) ∥ L ∞ ( R ; L 2 ( T 2 )) ≤ C ∣ ξ ∣ κ m ε m β + 1 ( κ m τ m ε m 2 + ε m − 1 δ ) .
The averages over T 2 \mathbb{T}^2 T 2 in z z z are those over T 2 \mathbb{T}^2 T 2 in y y y , as the functions are ε m \varepsilon_m ε m -periodic; so (134) and (135) give
Z ξ , 2 + ξ ⋅ J m κ m ξ = − κ m ξ ⋅ ( K ( t ) − I ) ξ − i ( ξ ⋅ ( ∇ ⊥ ψ m ) ( t , ε m ⋅ ) ( R ξ , 1 − i ξ ⋅ χ m κ m ( t , ε m ⋅ ) ) ) T 2 , Z_{\xi,2}+\xi\cdot\mathbf{J}_m^{\kappa_m}\xi=-\kappa_m\xi\cdot(\mathbf{K}(t)-I)\xi-\mathrm{i}\left(\xi\cdot(\nabla^{\perp}\psi_m)(t,\varepsilon_m\cdot)\left(\mathcal{R}_{\xi,1}-\mathrm{i}\xi\cdot\chi_m^{\kappa_m}(t,\varepsilon_m\cdot)\right)\right)_{\mathbb{T}^2}, Z ξ , 2 + ξ ⋅ J m κ m ξ = − κ m ξ ⋅ ( K ( t ) − I ) ξ − i ( ξ ⋅ ( ∇ ⊥ ψ m ) ( t , ε m ⋅ ) ( R ξ , 1 − i ξ ⋅ χ m κ m ( t , ε m ⋅ ) ) ) T 2 ,
∣ Z ξ , 2 + ξ ⋅ J m κ m ξ ∣ ≤ κ m ε m − 1 δ ∣ ξ ∣ 2 + C ε m 2 β κ m ( ε m 2 κ m τ m + ε m − 1 δ ) ∣ ξ ∣ 2 ≤ C ε m − 1 δ κ m − 1 ∣ ξ ∣ 2 , \left|Z_{\xi,2}+\xi\cdot\mathbf{J}_m^{\kappa_m}\xi\right|\le\kappa_m\varepsilon_{m-1}^{\delta}|\xi|^2+C\frac{\varepsilon_m^{2\beta}}{\kappa_m}\left(\frac{\varepsilon_m^2}{\kappa_m\tau_m}+\varepsilon_{m-1}^{\delta}\right)|\xi|^2\le C\varepsilon_{m-1}^{\delta}\kappa_{m-1}|\xi|^2, ∣ Z ξ , 2 + ξ ⋅ J m κ m ξ ∣ ≤ κ m ε m − 1 δ ∣ ξ ∣ 2 + C κ m ε m 2 β ( κ m τ m ε m 2 + ε m − 1 δ ) ∣ ξ ∣ 2 ≤ C ε m − 1 δ κ m − 1 ∣ ξ ∣ 2 ,
the last by κ m ≤ κ m − 1 \kappa_m\le\kappa_{m-1} κ m ≤ κ m − 1 , ε m 2 β / κ m ≤ 20 κ m − 1 \varepsilon_m^{2\beta}/\kappa_m\le20\kappa_{m-1} ε m 2 β / κ m ≤ 20 κ m − 1 and ε m 2 / ( κ m τ m ) ≤ C ε m − 1 4 δ \varepsilon_m^2/(\kappa_m\tau_m)\le C\varepsilon_{m-1}^{4\delta} ε m 2 / ( κ m τ m ) ≤ C ε m − 1 4 δ (Proposition 2.1 (c) and (96) ). This is (a).
Step 3: Coercivity over a block. We show, with c d i s : = ( 1 − exp ( − 2 ) ) / 192 c_{\mathrm{dis}}:=(1-\exp(-2))/192 c dis := ( 1 − exp ( − 2 )) /192 ,
∫ t − t + ξ ⋅ J m κ m ( t ) ξ d t ≥ 2 c d i s κ m − 1 h m ∣ ξ ∣ 2 ( ξ ∈ R 2 ) . (136) \int_{t_-}^{t_+}\xi\cdot\mathbf{J}_m^{\kappa_m}(t)\xi\,dt\ge2c_{\mathrm{dis}}\kappa_{m-1}h_m|\xi|^2\qquad(\xi\in\mathbb{R}^2). \tag*{(136)} ∫ t − t + ξ ⋅ J m κ m ( t ) ξ d t ≥ 2 c dis κ m − 1 h m ∣ ξ ∣ 2 ( ξ ∈ R 2 ) . (136)
On a complete core ζ ~ m , k = 1 \widetilde{\zeta}_{m,k}=1 ζ m , k = 1 , since ζ ( ⋅ − k ) \zeta(\cdot-k) ζ ( ⋅ − k ) is the only term of the partition (4) that is nonzero there. So for t ∈ [ k τ m , ( k + 1 3 ) τ m ] t\in[k\tau_m,(k+\frac{1}{3})\tau_m] t ∈ [ k τ m , ( k + 3 1 ) τ m ] the time integral of (18) is at least ∫ 0 τ m / 3 exp ( − 4 π 2 κ m t ~ / ε m 2 ) d t ~ ≥ ( 1 − exp ( − 2 ) ) ε m 2 / ( 4 π 2 κ m ) \int_0^{\tau_m/3}\exp(-4\pi^2\kappa_m\widetilde{t}/\varepsilon_m^2)\,d\widetilde{t}\ge(1-\exp(-2))\varepsilon_m^2/(4\pi^2\kappa_m) ∫ 0 τ m /3 exp ( − 4 π 2 κ m t / ε m 2 ) d t ≥ ( 1 − exp ( − 2 )) ε m 2 / ( 4 π 2 κ m ) , by (124) , and (18) gives
e ( k ) ⋅ J m κ m ( t ) e ( k ) ≥ κ m + ( 1 − exp ( − 2 ) ) ε m 2 β 2 κ m . e_{(k)}\cdot\mathbf{J}_m^{\kappa_m}(t)e_{(k)}\ge\kappa_m+(1-\exp(-2))\frac{\varepsilon_m^{2\beta}}{2\kappa_m}. e ( k ) ⋅ J m κ m ( t ) e ( k ) ≥ κ m + ( 1 − exp ( − 2 )) 2 κ m ε m 2 β .
A block has h m / τ m = ( 4 ⌈ ε m − 1 − δ ⌉ + 1 ) 2 h_m/\tau_m=(4\lceil\varepsilon_{m-1}^{-\delta}\rceil+1)^2 h m / τ m = ( 4 ⌈ ε m − 1 − δ ⌉ + 1 ) 2 slots, and only those within 2 τ m ′ + τ m 2\tau'_m+\tau_m 2 τ m ′ + τ m of its ends fail to have complete cores (Lemma 7.1 (d)); so the complete cores are those of h m / τ m − 4 τ m ′ / τ m h_m/\tau_m-4\tau'_m/\tau_m h m / τ m − 4 τ m ′ / τ m consecutive integers, with τ m ′ / τ m = 4 ⌈ ε m − 1 − δ ⌉ + 1 ≥ 65 \tau'_m/\tau_m=4\lceil\varepsilon_{m-1}^{-\delta}\rceil+1\ge65 τ m ′ / τ m = 4 ⌈ ε m − 1 − δ ⌉ + 1 ≥ 65 by (124) , and it has at least h m / ( 32 τ m ) h_m/(32\tau_m) h m / ( 32 τ m ) complete cores of each orientation. Since J m κ m \mathbf{J}_m^{\kappa_m} J m κ m is diagonal and at least κ m I \kappa_m I κ m I (Lemma 7.10), for e ∈ { e 1 , e 2 } e\in\{e_1,e_2\} e ∈ { e 1 , e 2 }
∫ t − t + e ⋅ J m κ m e d t ≥ κ m h m + h m 32 τ m ⋅ τ m 3 ⋅ ( 1 − exp ( − 2 ) ) ε m 2 β 2 κ m ≥ 2 c d i s h m ( κ m + ε m 2 β 2 κ m ) ≥ 2 c d i s h m κ m − 1 , \int_{t_-}^{t_+}e\cdot\mathbf{J}_m^{\kappa_m}e\,dt\ge\kappa_m h_m+\frac{h_m}{32\tau_m}\cdot\frac{\tau_m}{3}\cdot(1-\exp(-2))\frac{\varepsilon_m^{2\beta}}{2\kappa_m}\ge2c_{\mathrm{dis}}h_m\left(\kappa_m+\frac{\varepsilon_m^{2\beta}}{2\kappa_m}\right)\ge2c_{\mathrm{dis}}h_m\kappa_{m-1}, ∫ t − t + e ⋅ J m κ m e d t ≥ κ m h m + 32 τ m h m ⋅ 3 τ m ⋅ ( 1 − exp ( − 2 )) 2 κ m ε m 2 β ≥ 2 c dis h m ( κ m + 2 κ m ε m 2 β ) ≥ 2 c dis h m κ m − 1 ,
because κ m − 1 = K ‾ m ( κ m ) ≤ κ m + ε m 2 β / ( 2 κ m ) \kappa_{m-1}=\overline{K}_m(\kappa_m)\leq\kappa_m+\varepsilon_m^{2\beta}/(2\kappa_m) κ m − 1 = K m ( κ m ) ≤ κ m + ε m 2 β / ( 2 κ m ) (Lemma 7.10(i) and (19) ). This is (136) .
Step 4: The block integral. For real ξ \xi ξ with ε ˉ m ∣ ξ ∣ ≤ c b r / 4 \bar{\varepsilon}_m|\xi|\leq c_{\mathrm{br}}/4 ε ˉ m ∣ ξ ∣ ≤ c br /4 , by (a), (136) , Proposition 12.4(b) and ε ˉ m 2 ε m 2 ( ε ˉ m ∣ ξ ∣ ) 2 = κ m − 1 ∣ ξ ∣ 2 \frac{\bar{\varepsilon}_m^2}{\varepsilon_m^2}(\bar{\varepsilon}_m|\xi|)^2=\kappa_{m-1}|\xi|^2 ε m 2 ε ˉ m 2 ( ε ˉ m ∣ ξ ∣ ) 2 = κ m − 1 ∣ ξ ∣ 2 ,
Re ∫ t − t + Z ξ ( t ) d t ≤ − ( 2 c d i s − C ε m − 1 δ ) κ m − 1 h m ∣ ξ ∣ 2 + C b r h m κ m ε m 2 ∑ j ≥ 3 ( 2 ε ˉ m ∣ ξ ∣ c b r ) j ≤ − ( 2 c d i s − C ε m − 1 δ − 16 C b r c b r 3 ε ˉ m ∣ ξ ∣ ) ρ ( m ) 2 ∣ ξ ∣ 2 . \begin{aligned}
\operatorname{Re}\int_{t_-}^{t_+}Z_\xi(t)\,\mathrm{d}t\leq-\left(2c_{\mathrm{dis}}-C\varepsilon_{m-1}^{\delta}\right)\kappa_{m-1}h_m|\xi|^2+C_{\mathrm{br}}h_m\frac{\kappa_m}{\varepsilon_m^2}\sum_{j\geq3}\left(\frac{2\bar{\varepsilon}_m|\xi|}{c_{\mathrm{br}}}\right)^j \\
&\leq-\left(2c_{\mathrm{dis}}-C\varepsilon_{m-1}^{\delta}-\frac{16C_{\mathrm{br}}}{c_{\mathrm{br}}^3}\bar{\varepsilon}_m|\xi|\right)\rho(m)^2|\xi|^2.
\end{aligned} Re ∫ t − t + Z ξ ( t ) d t ≤ − ( 2 c dis − C ε m − 1 δ ) κ m − 1 h m ∣ ξ ∣ 2 + C br h m ε m 2 κ m j ≥ 3 ∑ ( c br 2 ε ˉ m ∣ ξ ∣ ) j ≤ − ( 2 c dis − C ε m − 1 δ − c br 3 16 C br ε ˉ m ∣ ξ ∣ ) ρ ( m ) 2 ∣ ξ ∣ 2 .
We choose c l o : = min { c b r / 8 , c b r 3 c d i s / ( 64 C b r ) , π c b r / ( 32 C b r 1 / 2 ) , 1 / ( 6 C b r ) } c_{\mathrm{lo}}:=\min\{c_{\mathrm{br}}/8,c_{\mathrm{br}}^3c_{\mathrm{dis}}/(64C_{\mathrm{br}}),\pi c_{\mathrm{br}}/(32C_{\mathrm{br}}^{1/2}),1/(6C_{\mathrm{br}})\} c lo := min { c br /8 , c br 3 c dis / ( 64 C br ) , π c br / ( 32 C br 1/2 ) , 1/ ( 6 C br )} , and Λ 0 \Lambda_0 Λ 0 so large that C ε m − 1 δ ≤ c d i s / 2 C\varepsilon_{m-1}^{\delta}\leq c_{\mathrm{dis}}/2 C ε m − 1 δ ≤ c dis /2 . Then the right side, and that of every truncation, is at most − c d i s ρ ( m ) 2 ∣ ξ ∣ 2 -c_{\mathrm{dis}}\rho(m)^2|\xi|^2 − c dis ρ ( m ) 2 ∣ ξ ∣ 2 for ε ˉ m ∣ ξ ∣ ≤ 2 c l o \bar{\varepsilon}_m|\xi|\leq2c_{\mathrm{lo}} ε ˉ m ∣ ξ ∣ ≤ 2 c lo .
Step 5: Frequency derivatives. By Proposition 12.4(b), summing a geometric series of ratio at most 3 / 4 3/4 3/4 , ∥ Z ξ ′ ∥ L ∞ ( R ) ≤ 16 C b r c b r − 2 κ m ε m 2 ( ε ˉ m ∣ ξ ′ ∣ ) 2 \|Z_{\xi'}\|_{L^\infty(\mathbb{R})}\leq16C_{\mathrm{br}}c_{\mathrm{br}}^{-2}\frac{\kappa_m}{\varepsilon_m^2}(\bar{\varepsilon}_m|\xi'|)^2 ∥ Z ξ ′ ∥ L ∞ ( R ) ≤ 16 C br c br − 2 ε m 2 κ m ( ε ˉ m ∣ ξ ′ ∣ ) 2 for complex ξ ′ \xi' ξ ′ with ε ˉ m ∣ ξ ′ ∣ ≤ 3 c b r / 8 \bar{\varepsilon}_m|\xi'|\leq3c_{\mathrm{br}}/8 ε ˉ m ∣ ξ ′ ∣ ≤ 3 c br /8 ; as 2 c l o ≤ c b r / 4 2c_{\mathrm{lo}}\leq c_{\mathrm{br}}/4 2 c lo ≤ c br /4 , the last two entries of the minimum defining c l o c_{\mathrm{lo}} c lo give the first sentence of (c). There the integral over [ t ′ , t ′ ′ ] [t',t''] [ t ′ , t ′′ ] is holomorphic and at most 32 C b r c b r − 2 ρ ( m ) 2 ∣ ξ ∣ 2 32C_{\mathrm{br}}c_{\mathrm{br}}^{-2}\rho(m)^2|\xi|^2 32 C br c br − 2 ρ ( m ) 2 ∣ ξ ∣ 2 , as t ′ ′ − t ′ ≤ 2 h m t''-t'\leq2h_m t ′′ − t ′ ≤ 2 h m and ε ˉ m 2 κ m h m / ε m 2 = ρ ( m ) 2 \bar{\varepsilon}_m^2\kappa_mh_m/\varepsilon_m^2=\rho(m)^2 ε ˉ m 2 κ m h m / ε m 2 = ρ ( m ) 2 . Cauchy’s estimates on the polydiscs about a real ξ \xi ξ of radii c b r / ( 16 ε ˉ m ) c_{\mathrm{br}}/(16\bar{\varepsilon}_m) c br / ( 16 ε ˉ m ) and ∣ ξ ∣ / 4 |\xi|/4 ∣ ξ ∣/4 give the bounds of (c) on this integral for j ≥ 2 j\geq2 j ≥ 2 (as ρ ( m ) 2 / ε ˉ m 2 = κ m h m / ε m 2 ≥ 1 \rho(m)^2/\bar{\varepsilon}_m^2=\kappa_mh_m/\varepsilon_m^2\geq1 ρ ( m ) 2 / ε ˉ m 2 = κ m h m / ε m 2 ≥ 1 ) and j = 1 j=1 j = 1 (at ξ = 0 \xi=0 ξ = 0 the gradient vanishes, as Z ξ Z_\xi Z ξ has no term of degree one). Faà di Bruno’s formula gives the last bound.
The other frequencies: two orientations
At the frequencies with ε ˉ m ∣ ξ ∣ ≥ c l o \bar{\varepsilon}_m|\xi|\geq c_{\mathrm{lo}} ε ˉ m ∣ ξ ∣ ≥ c lo the whole fibre must decay at the cell rate. If the contrast is bounded, as only at the terminal level, the inequality (128) suffices. If it is large, then at the frequencies with ε m ∣ ξ ∣ ≪ 1 \varepsilon_m|\xi|\ll1 ε m ∣ ξ ∣ ≪ 1 molecular diffusion alone damps the cell mean, the plane wave of frequency ξ \xi ξ , far more slowly than at the cell rate. The damping comes instead from the shear: a pulse flowing along e 2 e_2 e 2 moves the cell mean into the first cell harmonic, at a rate at least comparable to the cell rate once ε ˉ m ∣ ξ 2 ∣ ≥ c l o / 2 \bar{\varepsilon}_m|\xi_2|\geq c_{\mathrm{lo}}/2 ε ˉ m ∣ ξ 2 ∣ ≥ c lo /2 , and diffusion damps the harmonic. Such a pulse does not see the frequencies along e 1 e_1 e 1 , as it moves functions of y 1 y_1 y 1 along their level lines; so both orientations are needed, and a block has many pulses of each.
Proposition 12.6 (Damping of the other frequencies). There exist absolute constants χ i , X i > 0 \chi_{\mathrm{i}},\Chi_{\mathrm{i}}>0 χ i , X i > 0 such that the following holds under the standing hypotheses, with m m m above a level depending only on α , s , q , ζ , Λ \alpha,s,q,\zeta,\Lambda α , s , q , ζ , Λ . For every l ′ ∈ Z l'\in\mathbb{Z} l ′ ∈ Z , every ξ ∈ R 2 \xi\in\mathbb{R}^2 ξ ∈ R 2 in the zone with ε ˉ m ∣ ξ ∣ ≥ c l o \bar{\varepsilon}_m|\xi|\geq c_{\mathrm{lo}} ε ˉ m ∣ ξ ∣ ≥ c lo and every solution U U U of the fibre equation (126) on the renewal block [ t − , t + ] : = [ ( l ′ − 1 2 ) h m , ( l ′ + 1 2 ) h m ] [t_-,t_+]:=[(l'-\frac12)h_m,(l'+\frac12)h_m] [ t − , t + ] := [( l ′ − 2 1 ) h m , ( l ′ + 2 1 ) h m ] ,
∥ U ( t + ) ∥ L 2 ( T 2 ) ≤ exp ( − χ i κ m h m ε m 2 ) ∥ U ( t − ) ∥ L 2 ( T 2 ) . \|U(t_+)\|_{L^2(\mathbb{T}^2)}\leq\exp\left(-\chi_{\mathrm{i}}\frac{\kappa_mh_m}{\varepsilon_m^2}\right)\|U(t_-)\|_{L^2(\mathbb{T}^2)}. ∥ U ( t + ) ∥ L 2 ( T 2 ) ≤ exp ( − χ i ε m 2 κ m h m ) ∥ U ( t − ) ∥ L 2 ( T 2 ) . If moreover κ m ≤ 1 2 κ m − 1 \kappa_m\leq\frac12\kappa_{m-1} κ m ≤ 2 1 κ m − 1 , then, for every interval [ t ′ , t ′ ′ ] ⊂ [ t − , t + ] [t',t'']\subset[t_-,t_+] [ t ′ , t ′′ ] ⊂ [ t − , t + ] of length at least 4 X i ε m 2 / κ m 4\Chi_{\mathrm{i}}\varepsilon_m^2/\kappa_m 4 X i ε m 2 / κ m that lies in a complete core of a pulse k k k with ε ˉ m ∣ ξ ⋅ e ( k ) ∣ ≥ c l o / 2 \bar{\varepsilon}_m|\xi\cdot e(k)|\geq c_{\mathrm{lo}}/2 ε ˉ m ∣ ξ ⋅ e ( k ) ∣ ≥ c lo /2 ,
∥ U ( t ′ ′ ) ∥ L 2 ( T 2 ) ≤ 1 2 ∥ U ( t ′ ) ∥ L 2 ( T 2 ) . \|U(t'')\|_{L^2(\mathbb{T}^2)}\leq\frac12\|U(t')\|_{L^2(\mathbb{T}^2)}. ∥ U ( t ′′ ) ∥ L 2 ( T 2 ) ≤ 2 1 ∥ U ( t ′ ) ∥ L 2 ( T 2 ) .
Proof. By (127) the norm never grows. We distinguish two cases.
Bounded contrast, κ m ≥ 1 2 κ m − 1 \kappa_m\geq\frac12\kappa_{m-1} κ m ≥ 2 1 κ m − 1 . * Then ε m ∣ ξ ∣ ≥ 2 − 1 / 2 ε ˉ m ∣ ξ ∣ ≥ c l o / 2 \varepsilon_m|\xi|\geq2^{-1/2}\bar{\varepsilon}_m|\xi|\geq c_{\mathrm{lo}}/2 ε m ∣ ξ ∣ ≥ 2 − 1/2 ε ˉ m ∣ ξ ∣ ≥ c lo /2 , and (127) , (128) and K ≥ 1 2 I K\geq\frac12 I K ≥ 2 1 I give d d t ∥ U ∥ L 2 ( T 2 ) 2 ≤ − κ m ε m 2 c l o 2 4 ∥ U ∥ L 2 ( T 2 ) 2 \frac{\mathrm{d}}{\mathrm{d}t}\|U\|_{L^2(\mathbb{T}^2)}^2\leq-\frac{\kappa_m}{\varepsilon_m^2}\frac{c_{\mathrm{lo}}^2}{4}\|U\|_{L^2(\mathbb{T}^2)}^2 d t d ∥ U ∥ L 2 ( T 2 ) 2 ≤ − ε m 2 κ m 4 c lo 2 ∥ U ∥ L 2 ( T 2 ) 2 , whatever the velocity. For m m m large, this case is m = M m=M m = M (Proposition 2.1).
Large contrast, κ m ≤ 1 2 κ m − 1 \kappa_m\leq\frac12\kappa_{m-1} κ m ≤ 2 1 κ m − 1 . * Then κ m − 1 = K ‾ m ( κ m ) ≤ κ m + ε m 2 β / ( 2 κ m ) \kappa_{m-1}=\overline{K}_m(\kappa_m)\leq\kappa_m+\varepsilon_m^{2\beta}/(2\kappa_m) κ m − 1 = K m ( κ m ) ≤ κ m + ε m 2 β / ( 2 κ m ) (Lemma 7.10(i) and (19) ) forces ε m 2 β / κ m ≥ κ m − 1 \varepsilon_m^{2\beta}/\kappa_m\geq\kappa_{m-1} ε m 2 β / κ m ≥ κ m − 1 . One of the coordinates of ξ \xi ξ , say ξ 2 \xi_2 ξ 2 , has $\bar{\varepsilon}_m|\xi_2|\geq c_{\mathrm{lo}}/2. Let k ∈ 4 Z + 1 k \in4\mathbb{Z}+1 k ∈ 4 Z + 1 be a pulse, whose flow direction is e 2 e_2 e 2 , with a complete core. On the core only pulse k k k is active and ζ m , k = 1 \zeta_{m,k}=1 ζ m , k = 1 , so by (92) the velocity of (126) is the stationary shear 2 π ε m β − 1 cos ( 2 π z 1 ) e 2 2\pi\varepsilon_m^{\beta-1}\cos(2\pi z_1)e_2 2 π ε m β − 1 cos ( 2 π z 1 ) e 2 . The shear couples the cell mean ( U ) T 2 (U)_{\mathbb{T}^2} ( U ) T 2 to the first harmonic ( 2 cos ( 2 π z 1 ) U ) T 2 (\sqrt{2}\cos(2\pi z_1)U)_{\mathbb{T}^2} ( 2 cos ( 2 π z 1 ) U ) T 2 : testing (126) with 1 1 1 and with 2 cos ( 2 π z 1 ) \sqrt{2}\cos(2\pi z_1) 2 cos ( 2 π z 1 ) , and integrating by parts in z 2 z_2 z 2 and, where derivatives fall on the test function, in z 1 z_1 z 1 , we obtain
d d t ( U ) T 2 = − κ m ξ ⋅ K ( t ) ξ ( U ) T 2 − i 2 π ε m β − 1 ξ 2 ( 2 cos ( 2 π z 1 ) U ) T 2 , d d t ( 2 cos ( 2 π z 1 ) U ) T 2 = − i 2 π ε m β − 1 ξ 2 ( U ) T 2 + a remainder , (137) \begin{aligned}
\frac{\mathrm{d}}{\mathrm{d}t}(U)_{\mathbb{T}^2}&=-\kappa_m\xi\cdot\mathbf{K}(t)\xi(U)_{\mathbb{T}^2}-\mathrm{i}\sqrt{2}\pi\varepsilon_m^{\beta-1}\xi_2(\sqrt{2}\cos(2\pi z_1)U)_{\mathbb{T}^2},\\
\frac{\mathrm{d}}{\mathrm{d}t}(\sqrt{2}\cos(2\pi z_1)U)_{\mathbb{T}^2}&=-\mathrm{i}\sqrt{2}\pi\varepsilon_m^{\beta-1}\xi_2(U)_{\mathbb{T}^2}+\text{a remainder}, \tag*{(137)}
\end{aligned} d t d ( U ) T 2 d t d ( 2 cos ( 2 π z 1 ) U ) T 2 = − κ m ξ ⋅ K ( t ) ξ ( U ) T 2 − i 2 π ε m β − 1 ξ 2 ( 2 cos ( 2 π z 1 ) U ) T 2 , = − i 2 π ε m β − 1 ξ 2 ( U ) T 2 + a remainder , (137)
with the remainder at most C ( ε m β − 1 ∣ ξ 2 ∣ + κ m / ε m 2 ) ∥ U − ( U ) T 2 ∥ L 2 ( T 2 ) C(\varepsilon_m^{\beta-1}|\xi_2|+\kappa_m/\varepsilon_m^2)\lVert U-(U)_{\mathbb{T}^2}\rVert_{L^2(\mathbb{T}^2)} C ( ε m β − 1 ∣ ξ 2 ∣ + κ m / ε m 2 ) ∥ U − ( U ) T 2 ∥ L 2 ( T 2 ) for an absolute C C C , as K \mathbf{K} K does not depend on z z z and ε m ∣ ξ ∣ ≤ 2 π \varepsilon_m|\xi|\le\sqrt{2}\pi ε m ∣ ξ ∣ ≤ 2 π ; here ∣ ( 2 cos ( 2 π z 1 ) U ) T 2 ∣ ≤ ∥ U − ( U ) T 2 ∥ L 2 ( T 2 ) \lvert(\sqrt{2}\cos(2\pi z_1)U)_{\mathbb{T}^2}\rvert\le\lVert U-(U)_{\mathbb{T}^2}\rVert_{L^2(\mathbb{T}^2)} ∣( 2 cos ( 2 π z 1 ) U ) T 2 ∣ ≤ ∥ U − ( U ) T 2 ∥ L 2 ( T 2 ) , since the harmonic has mean zero. Hence the imaginary part of ( U ) T 2 ( 2 cos ( 2 π z 1 ) U ) T 2 (U)_{\mathbb{T}^2}(\sqrt{2}\cos(2\pi z_1)U)_{\mathbb{T}^2} ( U ) T 2 ( 2 cos ( 2 π z 1 ) U ) T 2 has the derivative 2 π ε m β − 1 ξ 2 ( ∣ ( 2 cos ( 2 π z 1 ) U ) T 2 ∣ 2 − ∣ ( U ) T 2 ∣ 2 ) \sqrt{2}\pi\varepsilon_m^{\beta-1}\xi_2(\lvert(\sqrt{2}\cos(2\pi z_1)U)_{\mathbb{T}^2}\rvert^2-\lvert(U)_{\mathbb{T}^2}\rvert^2) 2 π ε m β − 1 ξ 2 (∣( 2 cos ( 2 π z 1 ) U ) T 2 ∣ 2 − ∣( U ) T 2 ∣ 2 ) up to C ( ε m β − 1 ∣ ξ 2 ∣ + κ m / ε m ) ( U ) T 2 ∥ U − ( U ) T 2 ∥ L 2 ( T 2 ) C(\varepsilon_m^{\beta-1}|\xi_2|+\kappa_m/\varepsilon_m)(U)_{\mathbb{T}^2}\lVert U-(U)_{\mathbb{T}^2}\rVert_{L^2(\mathbb{T}^2)} C ( ε m β − 1 ∣ ξ 2 ∣ + κ m / ε m ) ( U ) T 2 ∥ U − ( U ) T 2 ∥ L 2 ( T 2 ) , the damping of the first line included, as ε m ∣ ξ ∣ ≤ 2 π \varepsilon_m|\xi|\le\sqrt{2}\pi ε m ∣ ξ ∣ ≤ 2 π . We integrate over an interval [ t ′ , t ′ ′ ] [t',t''] [ t ′ , t ′′ ] of the core; at its ends the imaginary part is at most 1 2 ∥ U ( t ′ ) ∥ L 2 ( T 2 ) 2 \frac{1}{2}\lVert U(t')\rVert_{L^2(\mathbb{T}^2)}^2 2 1 ∥ U ( t ′ ) ∥ L 2 ( T 2 ) 2 , as the norm does not increase, and Young’s inequality on the error gives
ε m β − 1 ∣ ξ 2 ∣ ∫ t ′ t ′ ′ ∣ ( U ) T 2 ∣ 2 d t ≤ C ∥ U ( t ′ ) ∥ L 2 ( T 2 ) 2 + C ε m β − 1 ∣ ξ 2 ∣ ( 1 + κ m ε m β + 1 ∣ ξ 2 ∣ ) 2 ∫ t ′ t ′ ′ ∥ U − ( U ) T 2 ∥ L 2 ( T 2 ) 2 d t . \varepsilon_m^{\beta-1}|\xi_2|\int_{t'}^{t''}|(U)_{\mathbb{T}^2}|^2\,\mathrm{d}t\le C\lVert U(t')\rVert_{L^2(\mathbb{T}^2)}^2+C\varepsilon_m^{\beta-1}|\xi_2|\left(1+\frac{\kappa_m}{\varepsilon_m^{\beta+1}|\xi_2|}\right)^2\int_{t'}^{t''}\lVert U-(U)_{\mathbb{T}^2}\rVert_{L^2(\mathbb{T}^2)}^2\,\mathrm{d}t. ε m β − 1 ∣ ξ 2 ∣ ∫ t ′ t ′′ ∣ ( U ) T 2 ∣ 2 d t ≤ C ∥ U ( t ′ ) ∥ L 2 ( T 2 ) 2 + C ε m β − 1 ∣ ξ 2 ∣ ( 1 + ε m β + 1 ∣ ξ 2 ∣ κ m ) 2 ∫ t ′ t ′′ ∥ U − ( U ) T 2 ∥ L 2 ( T 2 ) 2 d t .
By (127) and (128) , the mean-free part satisfies π 2 ( κ m / ε m 2 ) ∫ t ′ t ′ ′ ∥ U − ( U ) T 2 ∥ L 2 ( T 2 ) 2 d t ≤ ∥ U ( t ′ ) ∥ L 2 ( T 2 ) 2 \pi^2(\kappa_m/\varepsilon_m^2)\int_{t'}^{t''}\lVert U-(U)_{\mathbb{T}^2}\rVert_{L^2(\mathbb{T}^2)}^2\,\mathrm{d}t\le\lVert U(t')\rVert_{L^2(\mathbb{T}^2)}^2 π 2 ( κ m / ε m 2 ) ∫ t ′ t ′′ ∥ U − ( U ) T 2 ∥ L 2 ( T 2 ) 2 d t ≤ ∥ U ( t ′ ) ∥ L 2 ( T 2 ) 2 . Now ε m β + 1 ∣ ξ 2 ∣ / κ m ≥ ε ˉ m ∣ ξ 2 ∣ ≥ c l o / 2 \varepsilon_m^{\beta+1}|\xi_2|/\kappa_m\ge\bar{\varepsilon}_m|\xi_2|\ge c_{\mathrm{lo}}/2 ε m β + 1 ∣ ξ 2 ∣/ κ m ≥ ε ˉ m ∣ ξ 2 ∣ ≥ c lo /2 , by ε m 2 / κ m ≥ κ m − 1 \varepsilon_m^2/\kappa_m\ge\kappa_{m-1} ε m 2 / κ m ≥ κ m − 1 , so that
( ε m β − 1 ∣ ξ 2 ∣ ) − 1 = ( κ m / ( ε m β + 1 ∣ ξ 2 ∣ ) ) ε m 2 / κ m ≤ ( 2 / c l o ) ε m 2 / κ m . (\varepsilon_m^{\beta-1}|\xi_2|)^{-1}=(\kappa_m/(\varepsilon_m^{\beta+1}|\xi_2|))\varepsilon_m^2/\kappa_m\le(2/c_{\mathrm{lo}})\varepsilon_m^2/\kappa_m. ( ε m β − 1 ∣ ξ 2 ∣ ) − 1 = ( κ m / ( ε m β + 1 ∣ ξ 2 ∣ )) ε m 2 / κ m ≤ ( 2/ c lo ) ε m 2 / κ m .
Dividing the first bound by ε m β − 1 ∣ ξ 2 ∣ \varepsilon_m^{\beta-1}|\xi_2| ε m β − 1 ∣ ξ 2 ∣ and adding the two bounds, we obtain ∫ t ′ t ′ ′ ∥ U ∥ L 2 ( T 2 ) 2 d t ≤ C h i ( ε m 2 / κ m ) ∥ U ( t ′ ) ∥ L 2 ( T 2 ) 2 \int_{t'}^{t''}\lVert U\rVert_{L^2(\mathbb{T}^2)}^2\,\mathrm{d}t\le C_{\mathrm{hi}}(\varepsilon_m^2/\kappa_m)\lVert U(t')\rVert_{L^2(\mathbb{T}^2)}^2 ∫ t ′ t ′′ ∥ U ∥ L 2 ( T 2 ) 2 d t ≤ C hi ( ε m 2 / κ m ) ∥ U ( t ′ ) ∥ L 2 ( T 2 ) 2 with an absolute C h i C_{\mathrm{hi}} C hi , as c l o c_{\mathrm{lo}} c lo is absolute; and since the norm does not increase, ∥ U ( t ′ ′ ) ∥ L 2 ( T 2 ) ≤ 1 2 ∥ U ( t ′ ) ∥ L 2 ( T 2 ) \lVert U(t'')\rVert_{L^2(\mathbb{T}^2)}\le\frac{1}{2}\lVert U(t')\rVert_{L^2(\mathbb{T}^2)} ∥ U ( t ′′ ) ∥ L 2 ( T 2 ) ≤ 2 1 ∥ U ( t ′ ) ∥ L 2 ( T 2 ) as soon as t ′ ′ − t ′ ≥ 4 C h i ε m 2 / κ m t''-t'\ge4C_{\mathrm{hi}}\varepsilon_m^2/\kappa_m t ′′ − t ′ ≥ 4 C hi ε m 2 / κ m . This is the last assertion of the proposition for k ∈ 4 Z + 1 k\in4\mathbb{Z}+1 k ∈ 4 Z + 1 ; for k ∈ 4 Z + 3 k\in4\mathbb{Z}+3 k ∈ 4 Z + 3 we exchange the coordinates.
A core has length 2 3 τ m \frac{2}{3}\tau_m 3 2 τ m , and κ m τ m / ε m 2 → ∞ \kappa_m\tau_m/\varepsilon_m^2\to\infty κ m τ m / ε m 2 → ∞ with m m m (Proposition 2.1 (c)); for m m m large, each complete core of the right orientation halves the norm at least κ m τ m / ( 12 C h i ε m 2 ) \kappa_m\tau_m/(12C_{\mathrm{hi}}\varepsilon_m^2) κ m τ m / ( 12 C hi ε m 2 ) times. The block has at least h m / ( 32 τ m ) h_m/(32\tau_m) h m / ( 32 τ m ) such cores (Step 3 of the proof of Proposition 12.5), so the norm decreases by the factor exp ( − c h i κ m h m / ε m 2 ) \exp(-c_{\mathrm{hi}}\kappa_mh_m/\varepsilon_m^2) exp ( − c hi κ m h m / ε m 2 ) , with c h i : = min { log 2 / ( 384 C h i ) , c l o 2 / 8 } c_{\mathrm{hi}}:=\min\{\log2/(384C_{\mathrm{hi}}),c_{\mathrm{lo}}^2/8\} c hi := min { log 2/ ( 384 C hi ) , c lo 2 /8 } .
The Duhamel expansion and the adjoint equation
In a family, sources couple the components, and a source of order d d d takes d d d derivatives. The preparation expands the solution in its sources by Duhamel’s formula, and as the sources are small, finitely many terms suffice (Subsection 12.10). Proposition 12.7 , the main result of Subsections 12.6–12.9, prepares their sum. A direct estimate of each term would lose derivatives of the datum and, through the transport, powers of the contrast; the map Π ~ \widetilde{\Pi} Π of that proposition loses neither, and only its error, of size ε m Q + 3 \varepsilon_m^{Q+3} ε m Q + 3 , uses derivatives of the datum, at the molecular scale. The Duhamel expansion of a family G G G of level m m m from a datum g g g at a time t 1 t_1 t 1 is the sequence U { 0 } , U { 1 } , … U^{\{0\}},U^{\{1\}},\ldots U { 0 } , U { 1 } , … of solutions of
( ∂ t + b m ⋅ ∇ − κ m Δ ) U ν { j } = { 0 , j = 0 , ∇ ⋅ ∑ ν ′ → ν ∑ σ : ν ′ → ν ∑ d = 2 N s r c A σ , d ∇ d − 1 U ν ′ { j − 1 } , j ≥ 1. (\partial_t+\mathbf{b}_m\cdot\nabla-\kappa_m\Delta)U_{\nu}^{\{j\}}=
\begin{cases}
0, & j=0,\\
\nabla\cdot\sum_{\nu'\to\nu}\sum_{\sigma:\,\nu'\to\nu}\sum_{d=2}^{N^{\mathrm{src}}}A_{\sigma,d}\nabla^{d-1}U_{\nu'}^{\{j-1\}}, & j\ge1.
\end{cases} ( ∂ t + b m ⋅ ∇ − κ m Δ ) U ν { j } = { 0 , ∇ ⋅ ∑ ν ′ → ν ∑ σ : ν ′ → ν ∑ d = 2 N src A σ , d ∇ d − 1 U ν ′ { j − 1 } , j = 0 , j ≥ 1.
with U { 0 } ( t 1 ) = g U^{\{0\}}(t_1)=g U { 0 } ( t 1 ) = g and U { j } ( t 1 ) = 0 U^{\{j\}}(t_1)=0 U { j } ( t 1 ) = 0 for j ≥ 1 j\geq1 j ≥ 1 : the system (34) with its sources applied one at a time and without the smoothing operators S σ S_\sigma S σ . The terms exist in C ( [ t 1 , t f ] ; H N b u f − j ( N s r c − 2 ) ( T 2 ) V ) C([t_1,t_f];H^{N_{\mathrm{buf}}-j(N_{\mathrm{src}}-2)}(\mathbb{T}^2)^V) C ([ t 1 , t f ] ; H N buf − j ( N src − 2 ) ( T 2 ) V ) for j ≤ N d e p + 1 j\leq N_{\mathrm{dep}}+1 j ≤ N dep + 1 when g ∈ H N b u f ( T 2 ) V g\in H^{N_{\mathrm{buf}}}(\mathbb{T}^2)^V g ∈ H N buf ( T 2 ) V , by Proposition 8.6, whose condition at the scale ρ m o l ( m ) \rho_{\mathrm{mol}}(m) ρ mol ( m ) is (b2′) of Definition 11.1 with majorant sums at most C ∗ 2 ε m δ / 4 C_*^2\varepsilon_m^{\delta/4} C ∗ 2 ε m δ /4 .
Proposition 12.7 (Preparation of the Duhamel expansion). Let C ∗ ≥ 1 C_*\geq1 C ∗ ≥ 1 . There exist C C C and a level, both depending only on α , s , q , ζ , Λ , n , t 0 , T \alpha,s,q,\zeta,\Lambda,n,t_0,T α , s , q , ζ , Λ , n , t 0 , T and C ∗ C_* C ∗ , such that the following holds for every M M M , κ ∈ K M \kappa\in\mathcal{K}_M κ ∈ K M and every m ≤ M m\leq M m ≤ M above this level, the terminal level included. Let G G G be a family of level m m m that satisfies the invariant (Definition 11.1) on [ t A , t f ] [t_A,t_f] [ t A , t f ] , with C ∗ C_* C ∗ in place of the constants C Ω C_\Omega C Ω , c j c_j c j (j ≤ N s r c − 2 j\leq N_{\mathrm{src}}-2 j ≤ N src − 2 ), C w i d C_{\mathrm{wid}} C wid , C d e r C_{\mathrm{der}} C der , C j e t ( j ) C_{\mathrm{jet}}(j) C jet ( j ) (j ≤ N c o e f j\leq N_{\mathrm{coef}} j ≤ N coef ) and C a d m C_{\mathrm{adm}} C adm of items (3)–(4) of Subsection 11.1, and whose majorant matrix has nonzero entries that, as arrows between components, form no path of more than N p a t h N_{\mathrm{path}} N path arrows, let t 1 ∈ [ t A , t f − 2 h m ] t_1\in[t_A,t_f-2h_m] t 1 ∈ [ t A , t f − 2 h m ] , and let t p r t_{\mathrm{pr}} t pr be the end of the block that follows the block containing t 1 t_1 t 1 : with ( l − 1 2 ) h m ≤ t 1 < ( l + 1 2 ) h m (l-\frac12)h_m\leq t_1<(l+\frac12)h_m ( l − 2 1 ) h m ≤ t 1 < ( l + 2 1 ) h m and t 2 : = ( l + 1 2 ) h m t_2:=(l+\frac12)h_m t 2 := ( l + 2 1 ) h m ,
t p r : = t 2 + h m , h m < t p r − t 1 ≤ 2 h m . (138) t_{\mathrm{pr}}:=t_2+h_m,\qquad h_m<t_{\mathrm{pr}}-t_1\leq2h_m. \tag*{(138)} t pr := t 2 + h m , h m < t pr − t 1 ≤ 2 h m . (138) Then the velocity added at level m m m vanishes near t p r t_{\mathrm{pr}} t pr ,
b m = b m − 1 on [ t p r − τ m ′ , t p r + τ m ′ ] × T 2 , \mathbf{b}_m=\mathbf{b}_{m-1}\quad\text{on }[t_{\mathrm{pr}}-\tau'_m,t_{\mathrm{pr}}+\tau'_m]\times\mathbb{T}^2, b m = b m − 1 on [ t pr − τ m ′ , t pr + τ m ′ ] × T 2 , and there exists a linear map Π ~ : L 2 ( T 2 ) V → H N o u t ( T 2 ) V \widetilde{\Pi}:L^2(\mathbb{T}^2)^V\to H^{N_{\mathrm{out}}}(\mathbb{T}^2)^V Π : L 2 ( T 2 ) V → H N out ( T 2 ) V that maps fields with constant components to themselves, such that, for both choices ω v \omega_v ω v and ω v / Ψ v 2 \omega_v/\Psi_v^2 ω v / Ψ v 2 of the masses and for every g ∈ H N b u f ( T 2 ) V g\in H^{N_{\mathrm{buf}}}(\mathbb{T}^2)^V g ∈ H N buf ( T 2 ) V , with U { j } U^{\{j\}} U { j } the Duhamel expansion of G G G from g g g at t 1 t_1 t 1 ,
∥ Π ~ g ∥ H ρ ( m ) , ω N o u t ( T 2 ) ≤ C ∥ g ∥ L ω 2 ( T 2 ) , \|\widetilde{\Pi}g\|_{H^{N_{\mathrm{out}}}_{\rho(m),\omega}(\mathbb{T}^2)}
\leq C\|g\|_{L^2_\omega(\mathbb{T}^2)}, ∥ Π g ∥ H ρ ( m ) , ω N out ( T 2 ) ≤ C ∥ g ∥ L ω 2 ( T 2 ) , ∥ ∑ j = 0 N d e p U { j } ( t p r ) − Π ~ g ∥ L ω 2 ( T 2 ) ≤ C ε m Q + 3 ∥ g ∥ H ρ m o l ( m ) , ω N b u f ( T 2 ) . \left\|\sum_{j=0}^{N_{\mathrm{dep}}}U^{\{j\}}(t_{\mathrm{pr}})-\widetilde{\Pi}g\right\|_{L^2_\omega(\mathbb{T}^2)}
\leq C\varepsilon_m^{Q+3}\|g\|_{H^{N_{\mathrm{buf}}}_{\rho_{\mathrm{mol}}(m),\omega}(\mathbb{T}^2)}. j = 0 ∑ N dep U { j } ( t pr ) − Π g L ω 2 ( T 2 ) ≤ C ε m Q + 3 ∥ g ∥ H ρ mol ( m ) , ω N buf ( T 2 ) . With the masses ω v / Ψ v 2 \omega_v/\Psi_v^2 ω v / Ψ v 2 , these norms are those of L G 2 L^2_G L G 2 and H ρ , G N H^N_{\rho,G} H ρ , G N in (35) .
The plan. The proof (end of Subsection 12.9) cuts the datum into pieces within ρ ( m ) / 4 \rho(m)/4 ρ ( m ) /4 of their centres y 0 y_0 y 0 and freezes the conductivity and the forms of the sources at y 0 y_0 y 0 as Taylor polynomials, which become insertions , operators inserted between free evolutions. Each term of the expansion becomes a finite sum of Duhamel terms of the frozen operator (125) , across the old partial block, the change of chart at t 2 t_2 t 2 , and one complete block (Figure 12.1 ); Lemma 12.15 gives the model of each and its error.
Sources moved onto tests. In the bounds that determine the model, the derivatives that the sources take are never taken of a term of the expansion: we pair the term with a test , a solution of the adjoint fibre equation run backward from the final time, and move each source onto the test, which carries all of them. Against the principal branch R ξ R_\xi R ξ , a source in divergence form acts at most at a constant times the sum of the rate κ m − 1 ∣ ξ ∣ 2 \kappa_{m-1}|\xi|^2 κ m − 1 ∣ ξ ∣ 2 of the homogenized equation and one over the length h m h_m h m of a block (Lemma 12.10 ), and over a block it costs at most a power of 1 + ρ ( m ) ∣ ξ ∣ 1+\rho(m)|\xi| 1 + ρ ( m ) ∣ ξ ∣ , which the Gaussian of Proposition 12.5 (b) absorbs. These bounds lose no power of the contrast. Everything else decays over a block by a factor smaller than any power of ε m − 1 \varepsilon_{m-1} ε m − 1 ; it is estimated with losses polynomial in the contrast factor and in the number of cell times ε m 2 / κ m \varepsilon_m^2/\kappa_m ε m 2 / κ m in a block, and it goes into the error.
Conventions for Subsections 12.6–12.9. The standing hypotheses hold with m ≤ M m\leq M m ≤ M , the terminal level included, and m m m above a level that is at least that of Proposition 12.6 and that each lemma may enlarge; l ∈ Z l \in\mathbb{Z} l ∈ Z is fixed, and the block is [ ( l − 1 2 ) h m , ( l + 1 2 ) h m ] [(l-\frac{1}{2})h_m,(l+\frac{1}{2})h_m] [( l − 2 1 ) h m , ( l + 2 1 ) h m ] . Unless a statement says otherwise, t ′ < t ′ ′ t' < t'' t ′ < t ′′ lie in the block, ξ \xi ξ is real with ε ˉ m ∣ ξ ∣ ≤ 2 c l o \bar{\varepsilon}_m|\xi| \le2c_{\mathrm{lo}} ε ˉ m ∣ ξ ∣ ≤ 2 c lo , and insertions have strength at most one; for insertions of any strengths, every bound on a Duhamel term or its moment terms holds with its constant multiplied by the product of the strengths of its insertions, as these terms are linear in each insertion. The constants C C C , C ( N ) C(N) C ( N ) , c > 0 c > 0 c > 0 , the integers N 0 N_0 N 0 , N 2 N_2 N 2 , N 3 N_3 N 3 and the level of a statement depend only on α \alpha α , s s s , q q q , ζ \zeta ζ , Λ \Lambda Λ , n n n , t 0 t_0 t 0 , T T T , on N p a t h N_{\mathrm{path}} N path and on the integers that the statement fixes: the lengths of its Duhamel terms, the orders and Taylor degrees of their insertions, the moment order N 1 N_1 N 1 , and its Sobolev orders and weight exponents. Proofs measure time in cell units κ m t / ε m 2 \kappa_m t/\varepsilon_m^2 κ m t / ε m 2 . Quiet times are those of § 5.4 , at which no pulse of level m m m is active; no pulse is active within τ m ′ ≥ 5 τ m \tau'_m \ge5\tau_m τ m ′ ≥ 5 τ m of the ends of the block, its quiet margins. Windows of pulses are those of (93) , and complete cores those of Subsection 12.4. We call ( 1 + ε ˉ m / ε m ) = 1 + ( κ m − 1 / κ m ) 1 / 2 (1+\bar{\varepsilon}_m/\varepsilon_m)=1+(\kappa_{m-1}/\kappa_m)^{1/2} ( 1 + ε ˉ m / ε m ) = 1 + ( κ m − 1 / κ m ) 1/2 the contrast factor , and we write ∥ f ∥ H J ( T 2 ) : = ∥ f ∥ H 1 J ( T 2 ) \|f\|_{H_J(\mathbb{T}^2)}:=\|f\|_{H^J_1(\mathbb{T}^2)} ∥ f ∥ H J ( T 2 ) := ∥ f ∥ H 1 J ( T 2 ) for the norm (36) at ρ = 1 \rho=1 ρ = 1 .
Insertions. After freezing, the sources become insertions; by (139) , in cell units an insertion of order d d d is the cell rate times an operator of order d d d in z z z , with coefficients bounded in z z z and polynomial in y − y 0 y-y_0 y − y 0 . The frozen operator (125) acts componentwise on fields of a finite set V V V of components with masses ω v \omega_v ω v ; we write ⟨ f , f ′ ⟩ L 2 ( T 2 ) : = ( f f ′ ‾ ) T 2 \langle f,f'\rangle_{L^2(\mathbb{T}^2)}:=(f\overline{f'})_{\mathbb{T}^2} ⟨ f , f ′ ⟩ L 2 ( T 2 ) := ( f f ′ ) T 2 and ⟨ U , U ′ ⟩ L ω 2 ( T 2 ) : = ∑ v ω v − 1 ⟨ U v , U v ′ ⟩ L 2 ( T 2 ) \langle U,U'\rangle_{L^2_\omega(\mathbb{T}^2)}:=\sum_v\omega_v^{-1}\langle U_v,U'_v\rangle_{L^2(\mathbb{T}^2)} ⟨ U , U ′ ⟩ L ω 2 ( T 2 ) := ∑ v ω v − 1 ⟨ U v , U v ′ ⟩ L 2 ( T 2 ) . An insertion is an operator I U : = ∇ ⋅ ∑ d = 2 N s r c A d ∇ d − 1 U IU:=\nabla\cdot\sum_{d=2}^{N_{\mathrm{src}}}A_d\nabla^{d-1}U I U := ∇ ⋅ ∑ d = 2 N src A d ∇ d − 1 U between such fields with coefficients
A d ( t , y ) = κ m ε m d − 2 A d ( t , y , y ε m ) , (139) A_d(t,y)=\kappa_m\varepsilon_m^{d-2}A_d\left(t,y,\frac{y}{\varepsilon_m}\right), \tag*{(139)} A d ( t , y ) = κ m ε m d − 2 A d ( t , y , ε m y ) , (139)
where A d ( t , y , z ) = ∑ a ( ( y − y 0 ) / ρ ( m ) ) a A d , a ( t , z ) A_d(t,y,z)=\sum_{a}((y-y_0)/\rho(m))^aA_{d,a}(t,z) A d ( t , y , z ) = ∑ a (( y − y 0 ) / ρ ( m ) ) a A d , a ( t , z ) , the sum over the multi-indices a ∈ N 2 a\in\mathbb{N}^2 a ∈ N 2 with ∣ a ∣ |a| ∣ a ∣ at most the Taylor degree , has coefficients A d , a A_{d,a} A d , a that are two-scale forms (§ 5.2 ) independent of y y y , measurable in t t t . On the block they satisfy:
(i) the structure clauses (a2)–(a4) of Definition 11.1: transverse (independent of z ⋅ e ( k ) z\cdot e_{(k)} z ⋅ e ( k ) ) on the window of each pulse k k k , invariant along e ( k ) e_{(k)} e ( k ) where ∣ t − k τ m ∣ ≤ 3 4 τ m |t-k\tau_m|\le\frac{3}{4}\tau_m ∣ t − k τ m ∣ ≤ 4 3 τ m , and independent of z z z at times with no pulse active within τ m / 24 \tau_m/24 τ m /24 ;
(ii) the derivative bound: with M v v ′ : = ∑ d ∑ a ∑ j ′ ∥ ∇ z j ′ ( A d , a ) v v ′ ∥ L ∞ M_{vv'}:=\sum_d\sum_a\sum_{j'}\|\nabla_z^{j'}(A_{d,a})_{vv'}\|_{L^\infty} M v v ′ := ∑ d ∑ a ∑ j ′ ∥ ∇ z j ′ ( A d , a ) v v ′ ∥ L ∞ , the sum over j ′ j' j ′ up to the derivative counts below, the matrix of entries ( ω v ′ / ω v ) 1 / 2 M v v ′ (\omega_{v'}/\omega_v)^{1/2}M_{vv'} ( ω v ′ / ω v ) 1/2 M v v ′ has row and column sums at most the strength ;
(iii) either this matrix is diagonal and A d A_d A d has no term of degree 0 0 0 in y − y 0 y-y_0 y − y 0 , or its nonzero entries, as arrows between components, form no path of more than N p a t h N_{\mathrm{path}} N path arrows. The sources of § 5.4 satisfy the second alternative, and the conductivity insertions the first.
The order is the largest d d d with A d ≠ 0 A_d\ne0 A d = 0 . We sum the forms of a source over the pulses, with their time cutoffs, each replaced by its Taylor polynomial at y 0 y_0 y 0 and written in the chart of the block by clause (a6) of Definition 11.1. On the block, extended by zero outside the interval of the Duhamel term, this sum is an insertion of strength C C C times its majorant sums. The Taylor polynomial at y 0 y_0 y 0 of the conductivity minus its value gives diagonal insertions of order 2 2 2 and strength C ε m − 1 2 δ ρ ( m ) / ε m − 1 C\varepsilon_{m-1}^{2\delta}\rho(m)/\varepsilon_{m-1} C ε m − 1 2 δ ρ ( m ) / ε m − 1 on fields supported near y 0 y_0 y 0 , each power of ( y − y 0 ) / ε m − 1 (y-y_0)/\varepsilon_{m-1} ( y − y 0 ) / ε m − 1 costing one moment at the scale ρ ( m ) \rho(m) ρ ( m ) .
Duhamel terms. A Duhamel term of length N N N on an interval [ t 1 , t 2 ] [t_1,t_2] [ t 1 , t 2 ] of the block from a datum g g g is the sequence U [ 0 ] , … , U [ N ] U^{[0]},\ldots,U^{[N]} U [ 0 ] , … , U [ N ] of energy solutions of (125) on [ t 1 , t 2 ] [t_1,t_2] [ t 1 , t 2 ] with U [ 0 ] ( t 1 ) = g U^{[0]}(t_1)=g U [ 0 ] ( t 1 ) = g and, for 1 ≤ j ≤ N 1\le j\le N 1 ≤ j ≤ N , the forcing I j U [ j − 1 ] \mathcal{I}_jU^{[j-1]} I j U [ j − 1 ] and U [ j ] ( t 1 ) = 0 U^{[j]}(t_1)=0 U [ j ] ( t 1 ) = 0 , where I 1 , … , I N \mathcal{I}_1,\ldots,\mathcal{I}_N I 1 , … , I N are insertions of orders d 1 , … , d N d_1,\ldots,d_N d 1 , … , d N . It is centred at y 0 y_0 y 0 when the conductivity of (125) is frozen at y 0 y_0 y 0 and the insertions are polynomials in y − y 0 y-y_0 y − y 0 . For an integer K ≥ 0 K\ge0 K ≥ 0 , let D K \mathcal{D}_K D K be the class of fields g g g on R 2 \mathbb{R}^2 R 2 with ∇ g g ∈ L ω , K 2 ( R 2 ) V \nabla^{g}g \in L^{2}_{\omega,K}(\mathbb{R}^{2})^{V} ∇ g g ∈ L ω , K 2 ( R 2 ) V for every b ≤ d 1 + ⋯ + d N − 2 N b \le d_{1}+\cdots+d_{N}-2N b ≤ d 1 + ⋯ + d N − 2 N , with the moments of (129) . For data in D K D_{K} D K , with K K K the sum of the Taylor degrees of the insertions, the Duhamel term exists and is unique, by the energy of Proposition 8.6 with the moment weights (the estimate of Lemma 8.5 is pointwise). Along the sequence the orders decrease by d j − 2 d_{j}-2 d j − 2 and the weights by the Taylor degree of I j \mathcal{I}_{j} I j at step j j j .
Tests. For real ξ \xi ξ , a test on [ t ′ , t ′ ′ ] [t',t''] [ t ′ , t ′′ ] is an energy solution U † U^{\dagger} U † of the adjoint fibre equation
− ∂ t U † = κ m ε m 2 ( ∇ + i ε m ξ ) ⋅ K ( t ) ( ∇ + i ε m ξ ) U † + 1 ε m ∇ ⊥ ψ m ( t , ε m z ) ⋅ ( ∇ + i ε m ξ ) U † (140) -\partial_{t}U^{\dagger}=\frac{\kappa_{m}}{\varepsilon_{m}^{2}}(\nabla+i\varepsilon_{m}\xi)\cdot\mathbf{K}(t)(\nabla+i\varepsilon_{m}\xi)U^{\dagger}+\frac{1}{\varepsilon_{m}}\nabla^{\perp}\psi_{m}(t,\varepsilon_{m}z)\cdot(\nabla+i\varepsilon_{m}\xi)U^{\dagger} \tag*{(140)} − ∂ t U † = ε m 2 κ m ( ∇ + i ε m ξ ) ⋅ K ( t ) ( ∇ + i ε m ξ ) U † + ε m 1 ∇ ⊥ ψ m ( t , ε m z ) ⋅ ( ∇ + i ε m ξ ) U † (140)
on [ t ′ , t ′ ′ ] × T 2 [t',t'']\times\mathbb{T}^{2} [ t ′ , t ′′ ] × T 2 with a datum at t ′ ′ t'' t ′′ (componentwise for fields): the fibre equation (126) run backward in time with the velocity reversed. The transport term is skew-adjoint and the diffusion self-adjoint, so for every solution U U U of (126) with a forcing F F F ,
d d t ⟨ U ( t ) , U † ( t ) ⟩ L 2 ( T 2 ) = ⟨ F ( t ) , U † ( t ) ⟩ L 2 ( T 2 ) . (141) \frac{\mathrm{d}}{\mathrm{d}t}\langle U(t),U^{\dagger}(t)\rangle_{L^{2}(\mathbb{T}^{2})}=\langle F(t),U^{\dagger}(t)\rangle_{L^{2}(\mathbb{T}^{2})}. \tag*{(141)} d t d ⟨ U ( t ) , U † ( t ) ⟩ L 2 ( T 2 ) = ⟨ F ( t ) , U † ( t ) ⟩ L 2 ( T 2 ) . (141)
As R ξ ( t ) exp ( ∫ t ′ t Z ξ ( t ~ ) d t ~ ) \mathcal{R}_{\xi}(t)\exp\left(\int_{t'}^{t}\mathcal{Z}_{\xi}(\widetilde{t})\,\mathrm{d}\widetilde{t}\right) R ξ ( t ) exp ( ∫ t ′ t Z ξ ( t ) d t ) solves (126) (Proposition 12.4(c)), exp ( ∫ t ′ t Z ξ ( t ~ ) d t ~ ) ⟨ R ξ ( t ) , U v † ( t ) ⟩ L 2 ( T 2 ) \exp\left(\int_{t'}^{t}\mathcal{Z}_{\xi}(\widetilde{t})\,\mathrm{d}\widetilde{t}\right)\langle\mathcal{R}_{\xi}(t),U^{\dagger}_{v}(t)\rangle_{L^{2}(\mathbb{T}^{2})} exp ( ∫ t ′ t Z ξ ( t ) d t ) ⟨ R ξ ( t ) , U v † ( t ) ⟩ L 2 ( T 2 ) is constant. A test is off-branch at t t t if ⟨ R ξ ( t ) , U v † ( t ) ⟩ L 2 ( T 2 ) = 0 \langle\mathcal{R}_{\xi}(t),U^{\dagger}_{v}(t)\rangle_{L^{2}(\mathbb{T}^{2})}=0 ⟨ R ξ ( t ) , U v † ( t ) ⟩ L 2 ( T 2 ) = 0 for every v v v ; it is then off-branch at every earlier time, and it reads only the part of a solution off the branch. For a time t ′ ′ t'' t ′′ of the block with no pulse of level m m m active within τ m / 24 \tau_{m}/24 τ m /24 of it, the branch test U b r † U^{\dagger}_{\mathrm{br}} U br † is the test with U b r † ( t ′ ′ ) = 1 U^{\dagger}_{\mathrm{br}}(t'')=1 U br † ( t ′′ ) = 1 ; paired with a Bloch sum at t ′ ′ t'' t ′′ , it reads the cell mean, that is, by (143) below, the Fourier transform.
Moment terms. The coefficients of insertions are polynomials in y − y 0 y-y_{0} y − y 0 , which do not act fibre by fibre; so we carry the moments of the Duhamel terms as further unknowns. A power of y − y 0 y-y_{0} y − y 0 then maps moment terms to higher ones, every coefficient becomes periodic in the fast variable, and Bloch sums decouple the fibres. The moments are also the frequency derivatives of the Fourier transform (143) , which localize the model near y 0 y_{0} y 0 . Let U [ 0 ] , … , U [ N ] U^{[0]},\ldots,U^{[N]} U [ 0 ] , … , U [ N ] be a Duhamel term on [ t 1 , t 2 ] [t_{1},t_{2}] [ t 1 , t 2 ] , let N 1 ≥ 0 N_{1}\ge0 N 1 ≥ 0 , let N ′ N' N ′ be N 1 N_{1} N 1 plus the sum of the Taylor degrees of its insertions, and let its datum lie in D N ′ + 2 D_{N'+2} D N ′ + 2 . For a multi-index i ∈ N 2 i\in\mathbb{N}^{2} i ∈ N 2 with ∣ i ∣ |i| ∣ i ∣ at most N 1 N_{1} N 1 plus the sum of the Taylor degrees of I j + 1 , … , I N \mathcal{I}_{j+1},\ldots,\mathcal{I}_{N} I j + 1 , … , I N , the moment term ( j , i ) (j,i) ( j , i ) is U [ j , i ] : = ( ( y − y 0 ) / ρ ( m ) ) i U [ j ] U^{[j,i]}:=((y-y_{0})/\rho(m))^{i}U^{[j]} U [ j , i ] := (( y − y 0 ) / ρ ( m ) ) i U [ j ] . Its degree is ∣ i ∣ |i| ∣ i ∣ plus the sum of the Taylor degrees of I 1 , … , I j \mathcal{I}_{1},\ldots,\mathcal{I}_{j} I 1 , … , I j , at most N ′ N' N ′ ; a moment term is lower than ( j , i ) (j,i) ( j , i ) if its degree is smaller, or equal with a smaller j j j , and the moment terms with given j j j and ∣ i ∣ |i| ∣ i ∣ form a group . Commuting the weight with the frozen operator, we obtain
( ∂ t + ∇ ⊥ ψ m ⋅ ∇ − κ m ∇ ⋅ ( K ( t ) ∇ ⋅ ) ) U [ j , i ] = ( y − y 0 ρ ( m ) ) i I j U [ j − 1 ] + ∑ e i ⋅ e ρ ( m ) ( ∇ ⊥ ψ m ⋅ e − 2 κ m K ( t ) e ⋅ ∇ ) U [ j , i − e ] + κ m ∑ e , e ′ ( i ⋅ e ) ( i ⋅ e ′ − e ⋅ e ′ ) ρ ( m ) 2 e ⋅ K ( t ) e ′ U [ j , i − e − e ′ ] (142) \begin{aligned}
\left(\partial_{t}+\nabla^{\perp}\psi_{m}\cdot\nabla-\kappa_{m}\nabla\cdot(\mathbf{K}(t)\nabla\cdot)\right)U^{[j,i]}
={}&\left(\frac{y-y_{0}}{\rho(m)}\right)^{i}\mathcal{I}_{j}U^{[j-1]}+\sum_{e}\frac{i\cdot e}{\rho(m)}\left(\nabla^{\perp}\psi_{m}\cdot e-2\kappa_{m}\mathbf{K}(t)e\cdot\nabla\right)U^{[j,i-e]}\\
&+\kappa_{m}\sum_{e,e'}\frac{(i\cdot e)(i\cdot e'-e\cdot e')}{\rho(m)^{2}}e\cdot\mathbf{K}(t)e'U^{[j,i-e-e']} \tag*{(142)}
\end{aligned} ( ∂ t + ∇ ⊥ ψ m ⋅ ∇ − κ m ∇ ⋅ ( K ( t ) ∇ ⋅ ) ) U [ j , i ] = ( ρ ( m ) y − y 0 ) i I j U [ j − 1 ] + e ∑ ρ ( m ) i ⋅ e ( ∇ ⊥ ψ m ⋅ e − 2 κ m K ( t ) e ⋅ ∇ ) U [ j , i − e ] + κ m e , e ′ ∑ ρ ( m ) 2 ( i ⋅ e ) ( i ⋅ e ′ − e ⋅ e ′ ) e ⋅ K ( t ) e ′ U [ j , i − e − e ′ ] (142)
(the sums over e , e ′ ∈ { e 1 , e 2 } e,e'\in\{e_{1},e_{2}\} e , e ′ ∈ { e 1 , e 2 } ; without the first term for j = 0 j=0 j = 0 ). Moving the weight through I j \mathcal{I}_{j} I j , whose coefficients are polynomials in ( y − y 0 ) / ε m − 1 = ( ρ ( m ) / ε m − 1 ) ( y − y 0 ) / ρ ( m ) (y-y_{0})/\varepsilon_{m-1}=(\rho(m)/\varepsilon_{m-1})(y-y_{0})/\rho(m) ( y − y 0 ) / ε m − 1 = ( ρ ( m ) / ε m − 1 ) ( y − y 0 ) / ρ ( m ) , makes the first term a sum of operators of order at most d j d_{j} d j on the moment terms ( j − 1 , i ′ ) (j-1,i') ( j − 1 , i ′ ) , ∣ i ′ ∣ |i'| ∣ i ′ ∣ at most ∣ i ∣ |i| ∣ i ∣ plus the Taylor degree of I j \mathcal{I}_{j} I j , with coefficients bounded as those of I j \mathcal{I}_{j} I j and ρ ( m ) − 1 \rho(m)^{-1} ρ ( m ) − 1 per derivative on the weight. All moment terms on the right are lower than ( j , i ) (j,i) ( j , i ) .
By the rules for Bloch sums after (130) , B U [ j , i ] ( t ) \mathcal{B}U^{[j,i]}(t) B U [ j , i ] ( t ) , short for B ( U [ j , i ] ( t ) ) ( ξ , ⋅ ) \mathcal{B}(U^{[j,i]}(t))(\xi,\cdot) B ( U [ j , i ] ( t )) ( ξ , ⋅ ) , solves (126) forced by the sum over the lower ( j ′ , i ′ ) (j',i') ( j ′ , i ′ ) of I ( j , i ) , ( j ′ , i ′ ) ( t ) B U [ j ′ , i ′ ] ( t ) \mathcal{I}_{(j,i),(j',i')}(t)\mathcal{B}U^{[j',i']}(t) I ( j , i ) , ( j ′ , i ′ ) ( t ) B U [ j ′ , i ′ ] ( t ) , the fibre forms of the terms of (142) ; T ( j , i ) , ( j ′ , i ′ ) ∗ ( t ) \mathcal{T}^{*}_{(j,i),(j',i')}(t) T ( j , i ) , ( j ′ , i ′ ) ∗ ( t ) is the adjoint for ⟨ ⋅ , ⋅ ⟩ L 2 ( T 2 ) \langle\cdot,\cdot\rangle_{L^{2}(\mathbb{T}^{2})} ⟨ ⋅ , ⋅ ⟩ L 2 ( T 2 ) . As the mean of a Bloch sum is ε m − 2 \varepsilon_m^{-2} ε m − 2 times the Fourier transform (Subsection 12.2), with ∇ ξ i : = ∂ ξ 1 i 1 ∂ ξ 2 i 2 \nabla_{\xi}^{i}:=\partial_{\xi_1}^{i_1}\partial_{\xi_2}^{i_2} ∇ ξ i := ∂ ξ 1 i 1 ∂ ξ 2 i 2 ,
ρ ( m ) − ∣ i ∣ ∇ ξ i U [ j ] ( t ) ^ ( ξ ) = ( − i ) ∣ i ∣ U [ j , i ] ^ ( t ) ( ξ ) = ( − i ) ∣ i ∣ ε m 2 B U [ j , i ] ( t ) ^ T 2 . (143) \rho^{(m)-|i|}\widehat{\nabla_{\xi}^{i}U^{[j]}(t)}(\xi)=(-i)^{|i|}\widehat{U^{[j,i]}}(t)(\xi)=(-i)^{|i|}\varepsilon_m^2\widehat{B U^{[j,i]}(t)}_{\mathbb{T}^2}. \tag*{(143)} ρ ( m ) − ∣ i ∣ ∇ ξ i U [ j ] ( t ) ( ξ ) = ( − i ) ∣ i ∣ U [ j , i ] ( t ) ( ξ ) = ( − i ) ∣ i ∣ ε m 2 B U [ j , i ] ( t ) T 2 . (143)
The moment terms have two further moments: at step j j j the weight available, N ′ + 2 N' + 2 N ′ + 2 minus the Taylor degrees of I 1 , … , I j I_1,\ldots,I_j I 1 , … , I j , exceeds the largest ∣ i ∣ |i| ∣ i ∣ by two. So, by the moment-weighted energy of the Duhamel term, each U [ j , i ] U^{[j,i]} U [ j , i ] , its gradient (in L 2 L^2 L 2 in time) and the flux of its forcing lie in L ω , 2 2 ( R 2 ) L^2_{\omega,2}(\mathbb{R}^2) L ω , 2 2 ( R 2 ) , their Bloch sums converge (paragraph of (130) ), and Lemma 12.2 applies. For an array of tests U † = ( U ( j , i ) † ) U^\dagger=(U^{\dagger}_{(j,i)}) U † = ( U ( j , i ) † ) indexed by a group, and for a lower group, T ∗ U † \mathcal{T}^{*}U^\dagger T ∗ U † is the array indexed by the lower group with entries ∑ ( j , i ) T ( j , i ) , ( j ′ , i ′ ) ∗ U ( j , i ) † \sum_{(j,i)}\mathcal{T}^{*}_{(j,i),(j',i')}U^\dagger_{(j,i)} ∑ ( j , i ) T ( j , i ) , ( j ′ , i ′ ) ∗ U ( j , i ) † . Arrays are normed by the square sums of the norms of their entries; for a function f f f on T 2 \mathbb{T}^2 T 2 and a field or array U U U , ∣ ⟨ U , f ⟩ ∣ = ∣ ⟨ f , U ⟩ ∣ |\langle U,f\rangle|=|\langle f,U\rangle| ∣ ⟨ U , f ⟩ ∣ = ∣ ⟨ f , U ⟩ ∣ is the square root of the sum over the entries and components of ω v − 1 ∣ ⟨ U v , f ⟩ L 2 ( T 2 ) ∣ 2 \omega_v^{-1}|\langle U_v,f\rangle_{L^2(\mathbb{T}^2)}|^2 ω v − 1 ∣ ⟨ U v , f ⟩ L 2 ( T 2 ) ∣ 2 .
Derivative counts. The proofs below differentiate the coefficients of the insertions of a Duhamel term at most 2 ( N 1 + N ′ + d 1 + ⋯ + d N + 2 ) + max { d 1 , … , d N } + 2 2(N_1+N'+d_1+\cdots+d_N+2)+\max\{d_1,\ldots,d_N\}+2 2 ( N 1 + N ′ + d 1 + ⋯ + d N + 2 ) + max { d 1 , … , d N } + 2 times in the fast variable, and, for a Duhamel term across the renewal (Lemma 12.15), at most 2 ( 3 N 1 + d 1 + ⋯ + d N − + N + + 4 ) + max { d 1 , … , d N − + N + } + 2 2(3N_1+d_1+\cdots+d_{N_-+N_+}+4)+\max\{d_1,\ldots,d_{N_-+N_+}\}+2 2 ( 3 N 1 + d 1 + ⋯ + d N − + N + + 4 ) + max { d 1 , … , d N − + N + } + 2 plus twice the sum of the Taylor degrees; N c o e f N_{\mathrm{coef}} N coef is at least these counts.
Estimates for the adjoint equation
Tests carry the derivatives that the sources take, and in cell units the velocity of the fibre equation has the size of the Péclet number, at most a constant times the contrast factor, by (133) with ∣ ξ ∣ = 1 / ε m |\xi|=1/\varepsilon_m ∣ ξ ∣ = 1/ ε m ; so a derivative of a test may cost a power of the contrast factor. Lemma 12.8 gives the bounds that accept this loss, for the parts that decay at the cell rate. The bounds without loss, Lemma 12.9, rest on a feature of shears: on the window of a pulse, a function independent of the coordinate along its flow is not transported, and on such functions the adjoint fibre equation is a heat equation with a potential of the size of ε ˉ m ∣ ξ ∣ \bar{\varepsilon}_m|\xi| ε ˉ m ∣ ξ ∣ , whatever the contrast. Lemma 12.10 says how insertions act on tests.
Halving intervals. For real ξ \xi ξ in the zone with ε ˉ m ∣ ξ ∣ ≥ c l o \bar{\varepsilon}_m|\xi|\ge c_{\mathrm{lo}} ε ˉ m ∣ ξ ∣ ≥ c lo , a halving interval is an interval of the block of length max { 4 C h i , 16 c l o − 2 } ε m 2 / κ m \max\{4C_{\mathrm{hi}},16c_{\mathrm{lo}}^{-2}\}\varepsilon_m^2/\kappa_m max { 4 C hi , 16 c lo − 2 } ε m 2 / κ m , contained, if κ m ≤ 1 2 κ m − 1 \kappa_m\le\frac12\kappa_{m-1} κ m ≤ 2 1 κ m − 1 , in a complete core of a pulse k k k with ε ˉ m ∣ ξ ⋅ e ( k ) ∣ ≥ c l o / 2 \bar{\varepsilon}_m|\xi\cdot e^{(k)}|\ge c_{\mathrm{lo}}/2 ε ˉ m ∣ ξ ⋅ e ( k ) ∣ ≥ c lo /2 . Solutions of (126) lose at least half of their L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) norm across each (Proposition 12.6, and for κ m ≥ 1 2 κ m − 1 \kappa_m\ge\frac12\kappa_{m-1} κ m ≥ 2 1 κ m − 1 the first case of its proof), and so do tests, as by (141) their map from the right end to the left end is the adjoint of the solution map. The block contains c κ m h m / ε m 2 c\kappa_m h_m/\varepsilon_m^2 c κ m h m / ε m 2 disjoint ones (the count at the end of that proof). If # [ t ′ , t ′ ′ ] \#[t',t''] # [ t ′ , t ′′ ] is the largest number of disjoint halving intervals in [ t ′ , t ′ ′ ] [t',t''] [ t ′ , t ′′ ] (and 0 0 0 if ε ˉ m ∣ ξ ∣ < c l o \bar{\varepsilon}_m|\xi|<c_{\mathrm{lo}} ε ˉ m ∣ ξ ∣ < c lo ), then # [ t ′ , t ′ ] + # [ t , t ′ ′ ] ≥ # [ t ′ , t ′ ′ ] − 1 \#[t',t']+\#[t,t'']\ge\#[t',t'']-1 # [ t ′ , t ′ ] + # [ t , t ′′ ] ≥ # [ t ′ , t ′′ ] − 1 , and shortening [ t ′ , t ′ ′ ] [t',t''] [ t ′ , t ′′ ] by a cell time ε m 2 / κ m \varepsilon_m^2/\kappa_m ε m 2 / κ m at either end lowers it by at most 2 2 2 .
In Lemma 12.8, (a) and (b) say that the fibre evolution propagates and creates derivatives at the cell scale ((b) over one cell time, from a datum of negative order and with a forcing), at the cost of a power of the contrast factor, which the decay of Proposition 12.6 over a block absorbs; by (c), without pulses each cell mode evolves separately, and the part of mean zero decays with constant one. The negative norm of (b) is ∥ f ∥ H − J ′ ( T 2 ) : = sup { ∣ ⟨ f , v ⟩ L 2 ( T 2 ) ∣ : ∥ v ∥ H J ′ ( T 2 ) ≤ 1 } \|f\|_{H^{-J'}(\mathbb{T}^2)}:=\sup\{|\langle f,v\rangle_{L^2(\mathbb{T}^2)}|:\|v\|_{H^{J'}(\mathbb{T}^2)}\le1\} ∥ f ∥ H − J ′ ( T 2 ) := sup { ∣ ⟨ f , v ⟩ L 2 ( T 2 ) ∣ : ∥ v ∥ H J ′ ( T 2 ) ≤ 1 } .
Lemma 12.8 (The fibre evolution smooths). For every choice of integers J , J ′ > 0 J,J'>0 J , J ′ > 0 there exists C C C such that, for every real ξ \xi ξ in the zone, every solution U U U of (126) on [ t ′ , t ′ ′ ] [t',t''] [ t ′ , t ′′ ] satisfies the following. Tests satisfy the same bounds with U † ( t ′ ) U^\dagger(t') U † ( t ′ ) on the left side and U † ( t ′ ′ ) U^\dagger(t'') U † ( t ′′ ) on the right side. (a) Long times.
∥ U ( t ′ ′ ) ∥ H J ( T 2 ) ≤ C ( 1 + ε ˉ m ε m ) J 2 − # [ t ′ , t ′ ′ ] × min { ∥ U ( t ′ ) ∥ H J ( T 2 ) , ( min { 1 , κ m ( t ′ ′ − t ′ ) ε m 2 } ) − J / 2 ∥ U ( t ′ ) ∥ L 2 ( T 2 ) } . \lVert U(t^{\prime\prime})\rVert_{H^J(\mathbb{T}^2)} \leq C\left(1+\frac{\bar{\varepsilon}_m}{\varepsilon_m}\right)^J 2^{-\#[t^{\prime},t^{\prime\prime}]} \times\min\left\{\lVert U(t^{\prime})\rVert_{H^J(\mathbb{T}^2)},\left(\min\left\{1,\frac{\kappa_m(t^{\prime\prime}-t^{\prime})}{\varepsilon_m^2}\right\}\right)^{-J/2}\lVert U(t^{\prime})\rVert_{L^2(\mathbb{T}^2)}\right\}. ∥ U ( t ′′ ) ∥ H J ( T 2 ) ≤ C ( 1 + ε m ε ˉ m ) J 2 − # [ t ′ , t ′′ ] × min { ∥ U ( t ′ ) ∥ H J ( T 2 ) , ( min { 1 , ε m 2 κ m ( t ′′ − t ′ ) } ) − J /2 ∥ U ( t ′ ) ∥ L 2 ( T 2 ) } .
(b) One cell time. Let κ m ( t ′ ′ − t ′ ) ≤ ε m 2 \kappa_m(t^{\prime\prime}-t^{\prime})\leq\varepsilon_m^2 κ m ( t ′′ − t ′ ) ≤ ε m 2 , and let U U U have a forcing F 0 F^0 F 0 , possibly zero. Then
∥ U ( t ′ ′ ) ∥ H J ( T 2 ) ≤ C ( 1 + ε ˉ m ε m ) J ( min { ∥ U ( t ′ ) ∥ H J ( T 2 ) , ( κ m ( t ′ ′ − t ′ ) ε m 2 ) − J + J ′ 2 ∥ U ( t ′ ) ∥ H − J ′ ( T 2 ) } + ∫ t ′ t ′ ′ ∥ F 0 ( t ) ∥ H J ( T 2 ) d t ) . \lVert U(t^{\prime\prime})\rVert_{H^J(\mathbb{T}^2)} \leq C\left(1+\frac{\bar{\varepsilon}_m}{\varepsilon_m}\right)^J\left(\min\left\{\lVert U(t^{\prime})\rVert_{H^J(\mathbb{T}^2)},\left(\frac{\kappa_m(t^{\prime\prime}-t^{\prime})}{\varepsilon_m^2}\right)^{-\frac{J+J^{\prime}}{2}}\lVert U(t^{\prime})\rVert_{H^{-J^{\prime}}(\mathbb{T}^2)}\right\}+\int_{t^{\prime}}^{t^{\prime\prime}}\lVert F^0(t)\rVert_{H^J(\mathbb{T}^2)}\,dt\right). ∥ U ( t ′′ ) ∥ H J ( T 2 ) ≤ C ( 1 + ε m ε ˉ m ) J min ⎩ ⎨ ⎧ ∥ U ( t ′ ) ∥ H J ( T 2 ) , ( ε m 2 κ m ( t ′′ − t ′ ) ) − 2 J + J ′ ∥ U ( t ′ ) ∥ H − J ′ ( T 2 ) ⎭ ⎬ ⎫ + ∫ t ′ t ′′ ∥ F 0 ( t ) ∥ H J ( T 2 ) d t .
(c) Quiet stretches. If no pulse of level m m m is active on [ t ′ , t ′ ′ ] [t^{\prime},t^{\prime\prime}] [ t ′ , t ′′ ] , then, with constant 1,
( U ( t ′ ′ ) ) T 2 = exp ( − κ m ∫ t ′ t ′ ′ ξ ⋅ K ( t ) ξ d t ) ( U ( t ′ ) ) T 2 , (U(t^{\prime\prime}))_{\mathbb{T}^2}=\exp\left(-\kappa_m\int_{t^{\prime}}^{t^{\prime\prime}}\xi\cdot\mathbf{K}(t)\xi\,dt\right)(U(t^{\prime}))_{\mathbb{T}^2}, ( U ( t ′′ ) ) T 2 = exp ( − κ m ∫ t ′ t ′′ ξ ⋅ K ( t ) ξ d t ) ( U ( t ′ ) ) T 2 ,
∥ U ( t ′ ′ ) − ( U ( t ′ ′ ) ) T 2 ∥ H J ( T 2 ) ≤ e − π 2 κ m ( t ′ ′ − t ′ ) 2 ε m 2 ∥ U ( t ′ ) − ( U ( t ′ ) ) T 2 ∥ H J ( T 2 ) . \lVert U(t^{\prime\prime})-(U(t^{\prime\prime}))_{\mathbb{T}^2}\rVert_{H^J(\mathbb{T}^2)}
\leq e^{-\frac{\pi^2\kappa_m(t^{\prime\prime}-t^{\prime})}{2\varepsilon_m^2}}\lVert U(t^{\prime})-(U(t^{\prime}))_{\mathbb{T}^2}\rVert_{H^J(\mathbb{T}^2)}. ∥ U ( t ′′ ) − ( U ( t ′′ ) ) T 2 ∥ H J ( T 2 ) ≤ e − 2 ε m 2 π 2 κ m ( t ′′ − t ′ ) ∥ U ( t ′ ) − ( U ( t ′ ) ) T 2 ∥ H J ( T 2 ) .
Proof. (a) The L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) norm does not increase, by (127) , and halves across halving intervals. When we differentiate (126) , only the terms in which derivatives fall on the velocity, of size C ( J ) ( 1 + ε ˉ m / ε m ) C(J)(1+\bar{\varepsilon}_m/\varepsilon_m) C ( J ) ( 1 + ε ˉ m / ε m ) , are neither skew nor dissipative. In cell time and in the zone, as ∣ 2 π n + ε m ξ ∣ ≥ 1 2 ∣ 2 π n ∣ \lvert2\pi\mathbf{n}+\varepsilon_m\xi\rvert\geq\frac{1}{2}\lvert2\pi\mathbf{n}\rvert ∣ 2 π n + ε m ξ ∣ ≥ 2 1 ∣ 2 π n ∣ for n ∈ Z 2 \mathbf{n}\in\mathbb{Z}^2 n ∈ Z 2 and K ≥ 15 16 I \mathbf{K}\geq\frac{15}{16}I K ≥ 16 15 I ,
1 4 ∥ ∇ j + 1 U ∥ L 2 ( T 2 ) 2 ≤ ∥ ( ∇ + i ε m ξ ) ∇ j U ∥ L 2 ( T 2 ) 2 ≤ 2 ∥ ∇ j + 1 U ∥ L 2 ( T 2 ) 2 + 4 π 2 ∥ ∇ j U ∥ L 2 ( T 2 ) 2 , \frac{1}{4}\lVert\nabla^{j+1}U\rVert_{L^2(\mathbb{T}^2)}^2\leq\lVert(\nabla+i\varepsilon_m\xi)\nabla^jU\rVert_{L^2(\mathbb{T}^2)}^2\leq2\lVert\nabla^{j+1}U\rVert_{L^2(\mathbb{T}^2)}^2+4\pi^2\lVert\nabla^jU\rVert_{L^2(\mathbb{T}^2)}^2, 4 1 ∥ ∇ j + 1 U ∥ L 2 ( T 2 ) 2 ≤ ∥( ∇ + i ε m ξ ) ∇ j U ∥ L 2 ( T 2 ) 2 ≤ 2 ∥ ∇ j + 1 U ∥ L 2 ( T 2 ) 2 + 4 π 2 ∥ ∇ j U ∥ L 2 ( T 2 ) 2 ,
d d t ∥ ∇ j U ∥ L 2 ( T 2 ) 2 ≤ − 15 8 ∥ ( ∇ + i ε m ξ ) ∇ j U ∥ L 2 ( T 2 ) 2 + C ( j ) ( 1 + ε ˉ m ε m ) ∑ i < j ∥ ( ∇ + i ε m ξ ) ∇ i U ∥ L 2 ( T 2 ) ∥ ∇ j U ∥ L 2 ( T 2 ) . \frac{\mathrm{d}}{\mathrm{d}t}\lVert\nabla^jU\rVert_{L^2(\mathbb{T}^2)}^2\leq-\frac{15}{8}\lVert(\nabla+i\varepsilon_m\xi)\nabla^jU\rVert_{L^2(\mathbb{T}^2)}^2+C(j)\left(1+\frac{\bar{\varepsilon}_m}{\varepsilon_m}\right)\sum_{i<j}\lVert(\nabla+i\varepsilon_m\xi)\nabla^iU\rVert_{L^2(\mathbb{T}^2)}\lVert\nabla^jU\rVert_{L^2(\mathbb{T}^2)}. d t d ∥ ∇ j U ∥ L 2 ( T 2 ) 2 ≤ − 8 15 ∥( ∇ + i ε m ξ ) ∇ j U ∥ L 2 ( T 2 ) 2 + C ( j ) ( 1 + ε m ε ˉ m ) i < j ∑ ∥( ∇ + i ε m ξ ) ∇ i U ∥ L 2 ( T 2 ) ∥ ∇ j U ∥ L 2 ( T 2 ) .
So, for t − t ′ ≤ c ( J ) / ( 1 + ε ˉ m / ε m ) 2 t-t^{\prime}\leq c(J)/(1+\bar{\varepsilon}_m/\varepsilon_m)^2 t − t ′ ≤ c ( J ) / ( 1 + ε ˉ m / ε m ) 2 , the weighted energy ∑ j ≤ J 8 − j ( j ! ) − 1 ( t − t ′ ) j ∥ ∇ j U ( t ) ∥ L 2 ( T 2 ) 2 \sum_{j\leq J}8^{-j}(j!)^{-1}(t-t^{\prime})^j\lVert\nabla^jU(t)\rVert_{L^2(\mathbb{T}^2)}^2 ∑ j ≤ J 8 − j ( j ! ) − 1 ( t − t ′ ) j ∥ ∇ j U ( t ) ∥ L 2 ( T 2 ) 2 does not increase, and the unweighted one at most doubles, by Young’s inequality on the cross terms:
( t − t ′ ) J ∥ U ( t ) ∥ H J ( T 2 ) 2 ≤ C ( J ) ∥ U ( t ′ ) ∥ L 2 ( T 2 ) 2 , ∥ U ( t ) ∥ H J ( T 2 ) ≤ C ( J ) ∥ U ( t ′ ) ∥ H J ( T 2 ) . (t-t^{\prime})^J\lVert U(t)\rVert_{H^J(\mathbb{T}^2)}^2\leq C(J)\lVert U(t^{\prime})\rVert_{L^2(\mathbb{T}^2)}^2,\qquad\lVert U(t)\rVert_{H^J(\mathbb{T}^2)}\leq C(J)\lVert U(t^{\prime})\rVert_{H^J(\mathbb{T}^2)}. ( t − t ′ ) J ∥ U ( t ) ∥ H J ( T 2 ) 2 ≤ C ( J ) ∥ U ( t ′ ) ∥ L 2 ( T 2 ) 2 , ∥ U ( t ) ∥ H J ( T 2 ) ≤ C ( J ) ∥ U ( t ′ ) ∥ H J ( T 2 ) .
Longer intervals use the L 2 L^2 L 2 bound up to t ′ ′ − c ( J ) / ( 1 + ε ˉ m / ε m ) 2 t^{\prime\prime}-c(J)/(1+\bar{\varepsilon}_m/\varepsilon_m)^2 t ′′ − c ( J ) / ( 1 + ε ˉ m / ε m ) 2 , which loses at most two halving intervals and so gives a constant multiple of 2 − # [ t ′ , t ′ ′ ] 2^{-\#[t^{\prime},t^{\prime\prime}]} 2 − # [ t ′ , t ′′ ] , and then the first of these bounds, which costs ( 1 + ε ˉ m / ε m ) J (1+\bar{\varepsilon}_m/\varepsilon_m)^J ( 1 + ε ˉ m / ε m ) J .
(b) Forcings are added by Duhamel’s formula. For a datum of negative order, we pair U U U at the midpoint with the tests from L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) , by (141) , and smooth after it.
(c) The coefficients do not depend on z z z : the mode exp ( 2 π i n ⋅ z ) \exp(2\pi i\mathbf{n}\cdot z) exp ( 2 π i n ⋅ z ) , n ∈ Z 2 \mathbf{n}\in\mathbb{Z}^2 n ∈ Z 2 , is multiplied by exp ( − κ m ε m 2 ∫ t ′ t ′ ′ ( 2 π n + ε m ξ ) ⋅ K ( t ) ( 2 π n + ε m ξ ) d t ) \exp(-\frac{\kappa_m}{\varepsilon_m^2}\int_{t^{\prime}}^{t^{\prime\prime}}(2\pi\mathbf{n}+\varepsilon_m\xi)\cdot\mathbf{K}(t)(2\pi\mathbf{n}+\varepsilon_m\xi)\,dt) exp ( − ε m 2 κ m ∫ t ′ t ′′ ( 2 π n + ε m ξ ) ⋅ K ( t ) ( 2 π n + ε m ξ ) d t ) , and ∣ 2 π n + ε m ξ ∣ ≥ π \lvert2\pi\mathbf{n}+\varepsilon_m\xi\rvert\geq\pi ∣ 2 π n + ε m ξ ∣ ≥ π for n ≠ 0 \mathbf{n}\neq0 n = 0 , with K ≥ 1 2 I \mathbf{K}\geq\frac{1}{2}I K ≥ 2 1 I . □ \square □
Activity. A test transverse to a pulse, that is, independent of the coordinate z ⋅ e ( k ) z\cdot e_{(k)} z ⋅ e ( k ) along its flow, has derivatives bounded without loss on the window of the pulse. The rest of a test does not: the shear turns its dependence along the flow into gradients across it, larger by at most a power of the Péclet number, which diffusion damps after about C log ( 1 + ε ˉ m / ε m ) C\log(1+\bar{\varepsilon}_m/\varepsilon_m) C log ( 1 + ε ˉ m / ε m ) cell times. The activity weight carries a power of the contrast factor on this part over that time. For t t t in the block, t a c t ( t ) t_{\mathrm{act}}(t) t act ( t ) is the supremum of the times of the block up to t t t at which a pulse of level m m m is active, e ( k ) e_{(k)} e ( k ) is the flow direction of the pulse k k k with ∣ t a c t ( t ) − k τ m ∣ ≤ 2 3 τ m \lvert t_{\mathrm{act}}(t)-k\tau_m\rvert\le\frac{2}{3}\tau_m ∣ t act ( t ) − k τ m ∣ ≤ 3 2 τ m (any direction if there is none), and ( U ) e ( k ) (U)_{e_{(k)}} ( U ) e ( k ) is the average of U U U along the lines parallel to e ( k ) e_{(k)} e ( k ) . For integers J , N a c t ≥ 0 J,N_{\mathrm{act}}\ge0 J , N act ≥ 0 , the activity weight and the weighted norm are
w N a c t ( t ) : = max { 1 , ( 1 + ε ˉ m ε m ) N a c t exp ( − 3 π 2 κ m ( t − t a c t ( t ) ) 8 ε m 2 ) } , w_{N_{\mathrm{act}}}(t):=\max\left\{1,\left(1+\frac{\bar{\varepsilon}_m}{\varepsilon_m}\right)^{N_{\mathrm{act}}}\exp\left(-\frac{3\pi^2\kappa_m(t-t_{\mathrm{act}}(t))}{8\varepsilon_m^2}\right)\right\}, w N act ( t ) := max { 1 , ( 1 + ε m ε ˉ m ) N act exp ( − 8 ε m 2 3 π 2 κ m ( t − t act ( t )) ) } ,
∥ U † ∥ H J ( T 2 ) ; N a c t , t : = ∥ ( U † ) e ( k ) ∥ H J ( T 2 ) + w N a c t ( t ) ∥ U † − ( U † ) e ( k ) ∥ H J ( T 2 ) (144) \left\|U^\dagger\right\|_{H^J(\mathbb{T}^2);N_{\mathrm{act}},t}:=\left\|(U^\dagger)_{e_{(k)}}\right\|_{H^J(\mathbb{T}^2)}+w_{N_{\mathrm{act}}}(t)\left\|U^\dagger-(U^\dagger)_{e_{(k)}}\right\|_{H^J(\mathbb{T}^2)} \tag*{(144)} U † H J ( T 2 ) ; N act , t := ( U † ) e ( k ) H J ( T 2 ) + w N act ( t ) U † − ( U † ) e ( k ) H J ( T 2 ) (144)
(masses for fields, square sums for arrays), with w N a c t ( t ) : = 1 w_{N_{\mathrm{act}}}(t):=1 w N act ( t ) := 1 if no pulse is active in the block before t t t . The norm is at least ∥ U † ∥ H J ( T 2 ) \left\|U^\dagger\right\|_{H^J(\mathbb{T}^2)} U † H J ( T 2 ) , as ( ⋅ ) e ( k ) (\mathord{\cdot})_{e_{(k)}} ( ⋅ ) e ( k ) is an orthogonal projection in H J ( T 2 ) H^J(\mathbb{T}^2) H J ( T 2 ) .
In Lemma 12.9 no constant depends on the contrast. By (a), a transverse test has the bounds of a heat equation over two cell times, and from 1 1 1 it stays within a constant times ε ˉ m ∣ ξ ∣ \bar{\varepsilon}_m|\xi| ε ˉ m ∣ ξ ∣ of 1 1 1 . By (b), an off-branch test has a small cell mean and decays backward at a rate comparable to the cell rate, and in the norm (144) it does so with N 0 N_0 N 0 fewer powers of the contrast factor in the weight at the earlier time, which compensates for crossing a pulse. By (c), the branch test is, up to a relative error of the size of ε ˉ m ∣ ξ ∣ \bar{\varepsilon}_m|\xi| ε ˉ m ∣ ξ ∣ , the complex conjugate of the scalar evolution of the branch, run backward.
Lemma 12.9 (Three kinds of tests). For every integer J ≥ 0 J\ge0 J ≥ 0 there exists an integer N 0 N_0 N 0 and, for every integer N a c t ≥ 0 N_{\mathrm{act}}\ge0 N act ≥ 0 , a constant C C C such that the following hold.
(a) Transverse tests. Let k k k be a pulse such that ∇ ⊥ ψ m ( t , ⋅ ) \nabla^\perp\psi_m(t,\mathord{\cdot}) ∇ ⊥ ψ m ( t , ⋅ ) is a multiple of ∇ ⊥ ψ m , k \nabla^\perp\psi_{m,k} ∇ ⊥ ψ m , k for every t ∈ [ t ′ , t ′ ′ ] t\in[t',t''] t ∈ [ t ′ , t ′′ ] , as it is on the window of k k k , and let the test U † U^\dagger U † be transverse at t ′ ′ t'' t ′′ . Then it is transverse on [ t ′ , t ′ ′ ] [t',t''] [ t ′ , t ′′ ] , and
∥ U † ( t ′ ) ∥ H J ( T 2 ) ≤ { C ∥ U † ( t ′ ′ ) ∥ H J ( T 2 ) if κ m ( t ′ ′ − t ′ ) ≤ ε m 2 , C ∥ U † ( t ′ ′ ) ∥ L 2 ( T 2 ) if ε m 2 ≤ κ m ( t ′ ′ − t ′ ) ≤ 2 ε m 2 . \left\|U^\dagger(t')\right\|_{H^J(\mathbb{T}^2)}
\le
\begin{cases}
C\left\|U^\dagger(t'')\right\|_{H^J(\mathbb{T}^2)} & \text{if }\kappa_m(t''-t')\le\varepsilon_m^2,\\
C\left\|U^\dagger(t'')\right\|_{L^2(\mathbb{T}^2)} & \text{if }\varepsilon_m^2\le\kappa_m(t''-t')\le2\varepsilon_m^2.
\end{cases} U † ( t ′ ) H J ( T 2 ) ≤ { C U † ( t ′′ ) H J ( T 2 ) C U † ( t ′′ ) L 2 ( T 2 ) if κ m ( t ′′ − t ′ ) ≤ ε m 2 , if ε m 2 ≤ κ m ( t ′′ − t ′ ) ≤ 2 ε m 2 . If moreover U † ( t ′ ′ ) = 1 U^\dagger(t'')=1 U † ( t ′′ ) = 1 and κ m ( t ′ ′ − t ′ ) ≤ 2 ε m 2 \kappa_m(t''-t')\le2\varepsilon_m^2 κ m ( t ′′ − t ′ ) ≤ 2 ε m 2 , then
∥ U † ( t ′ ) − 1 ∥ H J ( T 2 ) ≤ C ε ˉ m ∣ ξ ∣ . \left\|U^\dagger(t')-1\right\|_{H^J(\mathbb{T}^2)}\le C\bar{\varepsilon}_m|\xi|. U † ( t ′ ) − 1 H J ( T 2 ) ≤ C ε ˉ m ∣ ξ ∣. With a transverse forcing F 0 F^0 F 0 , the first two bounds hold with C ∫ t ′ t ′ ′ ∥ F 0 ( t ) ∥ H J ( T 2 ) d t C\int_{t'}^{t''}\left\|F^0(t)\right\|_{H^J(\mathbb{T}^2)}\,\mathrm{d}t C ∫ t ′ t ′′ F 0 ( t ) H J ( T 2 ) d t added to their right sides.
(b) Off-branch tests. Let U † U^\dagger U † be off-branch at t ′ ′ t'' t ′′ , and N a c t ≥ N a c t ′ ′ ≥ N a c t ′ + N 0 N_{\mathrm{act}}\ge N''_{\mathrm{act}}\ge N'_{\mathrm{act}}+N_0 N act ≥ N act ′′ ≥ N act ′ + N 0 , N a c t ′ ≥ 0 N'_{\mathrm{act}}\ge0 N act ′ ≥ 0 . Then
∥ U † ( t ′ ) ∥ L 2 ( T 2 ) ≤ e − π 2 κ m ( t ′ ′ − t ′ ) / ( 8 ε m 2 ) ∥ U † ( t ′ ′ ) ∥ L 2 ( T 2 ) , \left\|U^\dagger(t')\right\|_{L^2(\mathbb{T}^2)}
\le e^{-\pi^2\kappa_m(t''-t')/(8\varepsilon_m^2)}
\left\|U^\dagger(t'')\right\|_{L^2(\mathbb{T}^2)}, U † ( t ′ ) L 2 ( T 2 ) ≤ e − π 2 κ m ( t ′′ − t ′ ) / ( 8 ε m 2 ) U † ( t ′′ ) L 2 ( T 2 ) , ∥ U † ( t ′ ) ∥ H J ( T 2 ) ; N a c t ′ , t ′ ≤ C e − π 2 κ m ( t ′ ′ − t ′ ) / ( 8 ε m 2 ) ∥ U † ( t ′ ′ ) ∥ H J ( T 2 ) ; N a c t ′ ′ , t ′ ′ . \left\|U^\dagger(t')\right\|_{H^J(\mathbb{T}^2);N'_{\mathrm{act}},t'}
\le Ce^{-\pi^2\kappa_m(t''-t')/(8\varepsilon_m^2)}
\left\|U^\dagger(t'')\right\|_{H^J(\mathbb{T}^2);N''_{\mathrm{act}},t''}. U † ( t ′ ) H J ( T 2 ) ; N act ′ , t ′ ≤ C e − π 2 κ m ( t ′′ − t ′ ) / ( 8 ε m 2 ) U † ( t ′′ ) H J ( T 2 ) ; N act ′′ , t ′′ . (c) The branch test. For every time t ′ ′ t'' t ′′ of the block with no pulse of level m m m active within τ m / 24 \tau_m/24 τ m /24 of it, every 0 ≤ N a c t ′ ≤ N a c t 0\le N'_{\mathrm{act}}\le N_{\mathrm{act}} 0 ≤ N act ′ ≤ N act and every t ≤ t ′ ′ t\le t'' t ≤ t ′′ of the block,
⟨ R ξ ( t ) , U b r † ( t ) ⟩ L 2 ( T 2 ) = exp ( ∫ t t ′ ′ Z ξ ( t ~ ) d t ~ ) , \left\langle\mathcal{R}_{\xi}(t),U^\dagger_{\mathrm{br}}(t)\right\rangle_{L^2(\mathbb{T}^2)}
=\exp\left(\int_t^{t''}\mathcal{Z}_{\xi}(\widetilde{t})\,\mathrm{d}\widetilde{t}\right), ⟨ R ξ ( t ) , U br † ( t ) ⟩ L 2 ( T 2 ) = exp ( ∫ t t ′′ Z ξ ( t ) d t ) , ∥ U b r † ( t ) − exp ( ∫ t t ′ ′ Z ξ ( t ~ ) d t ~ ) ∥ H J ( T 2 ) ; N a c t ′ , t ≤ C ε ˉ m ∣ ξ ∣ ∣ exp ( ∫ t t ′ ′ Z ξ ( t ~ ) d t ~ ) ∣ . \left\|U^\dagger_{\mathrm{br}}(t)-\exp\left(\int_t^{t''}\mathcal{Z}_{\xi}(\widetilde{t})\,\mathrm{d}\widetilde{t}\right)\right\|_{H^J(\mathbb{T}^2);N'_{\mathrm{act}},t}
\le C\bar{\varepsilon}_m|\xi|\left|\exp\left(\int_t^{t''}\mathcal{Z}_{\xi}(\widetilde{t})\,\mathrm{d}\widetilde{t}\right)\right|. U br † ( t ) − exp ( ∫ t t ′′ Z ξ ( t ) d t ) H J ( T 2 ) ; N act ′ , t ≤ C ε ˉ m ∣ ξ ∣ exp ( ∫ t t ′′ Z ξ ( t ) d t ) .
Proof. Time is in cell units, so that τ m \tau_m τ m stands for κ m τ m / ε m 2 \kappa_m\tau_m/\varepsilon_m^2 κ m τ m / ε m 2 and Z ξ Z_\xi Z ξ for ε m 2 Z ξ / κ m \varepsilon_m^2Z_\xi/\kappa_m ε m 2 Z ξ / κ m ; tests run backward, and “before” and “after” refer to t t t . On the window of k k k the velocity is a shear parallel to e ( k ) e_{(k)} e ( k ) depending on z ⋅ e ( k ) ⊥ z\cdot e_{(k)}^\perp z ⋅ e ( k ) ⊥ (paragraph of (93) ); it vanishes outside the active intervals ∣ t − k τ m ∣ ≤ 2 3 τ m \lvert t-k\tau_m\rvert\le\frac{2}{3}\tau_m ∣ t − k τ m ∣ ≤ 3 2 τ m .
(a) If U † U^\dagger U † is transverse, e ( k ) ⋅ ∇ U † = 0 e_{(k)}\cdot\nabla U^\dagger=0 e ( k ) ⋅ ∇ U † = 0 , so on transverse functions (140) is the heat equation with a potential
− ∂ t U † = ( ∇ + i ε m ξ ) ⋅ K ( ∇ + i ε m ξ ) U † + i ε m 2 κ m ξ ⋅ ( ∇ ⊥ ψ m ) ( t , ε m z ) U † , -\partial_tU^\dagger=(\nabla+i\varepsilon_m\xi)\cdot K(\nabla+i\varepsilon_m\xi)U^\dagger+i\frac{\varepsilon_m^2}{\kappa_m}\xi\cdot(\nabla^\perp\psi_m)(t,\varepsilon_m z)U^\dagger, − ∂ t U † = ( ∇ + i ε m ξ ) ⋅ K ( ∇ + i ε m ξ ) U † + i κ m ε m 2 ξ ⋅ ( ∇ ⊥ ψ m ) ( t , ε m z ) U † ,
∥ ∇ z j ( ε m 2 κ m ξ ⋅ ∇ ⊥ ψ m ) ∥ L ∞ ( T 2 ) ≤ ( 2 π ) j C ε ˉ m ∣ ξ ∣ ( j ≥ 0 ) , \left\|\nabla_z^j\left(\frac{\varepsilon_m^2}{\kappa_m}\xi\cdot\nabla^\perp\psi_m\right)\right\|_{L^\infty(\mathbb{T}^2)}\le(2\pi)^j C\bar\varepsilon_m\lvert\xi\rvert\quad(j\ge0), ∇ z j ( κ m ε m 2 ξ ⋅ ∇ ⊥ ψ m ) L ∞ ( T 2 ) ≤ ( 2 π ) j C ε ˉ m ∣ ξ ∣ ( j ≥ 0 ) ,
by (92) and (133) ; it maps transverse functions to transverse ones, and its energy estimates over two cell times (with the standard weight ( t ′ ′ − t ) J (t''-t)^J ( t ′′ − t ) J for the second bound) do not see the contrast; from 1 1 1 , U † − 1 U^\dagger-1 U † − 1 is forced by − ε m 2 ξ ⋅ K ξ -\varepsilon_m^2\xi\cdot K\xi − ε m 2 ξ ⋅ K ξ plus the potential, of size C ε ˉ m ∣ ξ ∣ C\bar\varepsilon_m\lvert\xi\rvert C ε ˉ m ∣ ξ ∣ . With a transverse forcing F 0 F^0 F 0 , Duhamel’s formula gives the forced bounds: for t ′ ≤ t ~ ≤ t ′ ′ t'\le\tilde t\le t'' t ′ ≤ t ~ ≤ t ′′ the test from the datum F 0 ( t ~ ) F^0(\tilde t) F 0 ( t ~ ) at t ~ \tilde t t ~ is at most C ∥ F 0 ( t ~ ) ∥ H J ( T 2 ) C\lVert F^0(\tilde t)\rVert_{H^J(\mathbb{T}^2)} C ∥ F 0 ( t ~ ) ∥ H J ( T 2 ) in H J ( T 2 ) H^J(\mathbb{T}^2) H J ( T 2 ) at t ′ t' t ′ , by the first bound if t ~ − t ′ ≤ 1 \tilde t-t'\le1 t ~ − t ′ ≤ 1 and by the second if 1 ≤ t ~ − t ′ ≤ 2 1\le\tilde t-t'\le2 1 ≤ t ~ − t ′ ≤ 2 .
(b), Step 1: the L 2 L^2 L 2 rate. By Propositions 12.4(a) and 12.5(c), then (127) , (128) and K ≥ 15 16 I K\ge\frac{15}{16}I K ≥ 16 15 I , for each component,
∣ ( U † ) T 2 ∣ = ∣ ⟨ U † , 1 − R ξ ⟩ L 2 ( T 2 ) ∣ ≤ 1 3 ∥ U † ∥ L 2 ( T 2 ) , d d t ∥ U † ∥ L 2 ( T 2 ) 2 ≥ 15 π 2 8 ⋅ 8 9 ∥ U † ∥ L 2 ( T 2 ) 2 . \left|(U^\dagger)_{\mathbb{T}^2}\right|=\left|\langle U^\dagger,1-R_\xi\rangle_{L^2(\mathbb{T}^2)}\right|\le\frac{1}{3}\lVert U^\dagger\rVert_{L^2(\mathbb{T}^2)},\qquad\frac{\mathrm d}{\mathrm dt}\lVert U^\dagger\rVert_{L^2(\mathbb{T}^2)}^2\ge\frac{15\pi^2}{8}\cdot\frac{8}{9}\lVert U^\dagger\rVert_{L^2(\mathbb{T}^2)}^2. ( U † ) T 2 = ⟨ U † , 1 − R ξ ⟩ L 2 ( T 2 ) ≤ 3 1 ∥ U † ∥ L 2 ( T 2 ) , d t d ∥ U † ∥ L 2 ( T 2 ) 2 ≥ 8 15 π 2 ⋅ 9 8 ∥ U † ∥ L 2 ( T 2 ) 2 .
Step 2: two bounds. Let N 0 N_0 N 0 be the least integer at least the exponent C C C of ( 1 + ε ˉ m / ε m ) (1+\bar\varepsilon_m/\varepsilon_m) ( 1 + ε ˉ m / ε m ) in Lemma 12.8(a) for J J J . For U † U^\dagger U † off-branch at t ′ ′ t'' t ′′ ,
∥ U † ( t ′ ) ∥ H J ( T 2 ) ≤ { C ( 1 + ε ˉ m / ε m ) N 0 e − π 2 ( t ′ ′ − t ′ ) / 8 ∥ U † ( t ′ ′ ) ∥ H J ( T 2 ) (crude) , 3 2 e − π 2 ( t ′ ′ − t ′ ) / 2 ∥ U † ( t ′ ′ ) ∥ H J ( T 2 ) (quiet) , (145) \lVert U^\dagger(t')\rVert_{H^J(\mathbb{T}^2)}
\le
\begin{cases}
C(1+\bar\varepsilon_m/\varepsilon_m)^{N_0}e^{-\pi^2(t''-t')/8}\lVert U^\dagger(t'')\rVert_{H^J(\mathbb{T}^2)} & \text{(crude)},\\
\frac{3}{2}e^{-\pi^2(t''-t')/2}\lVert U^\dagger(t'')\rVert_{H^J(\mathbb{T}^2)} & \text{(quiet)},
\end{cases} \tag*{(145)} ∥ U † ( t ′ ) ∥ H J ( T 2 ) ≤ { C ( 1 + ε ˉ m / ε m ) N 0 e − π 2 ( t ′′ − t ′ ) /8 ∥ U † ( t ′′ ) ∥ H J ( T 2 ) 2 3 e − π 2 ( t ′′ − t ′ ) /2 ∥ U † ( t ′′ ) ∥ H J ( T 2 ) (crude) , (quiet) , (145)
the second if no pulse is active on [ t ′ , t ′ ′ ] [t',t''] [ t ′ , t ′′ ] : by Lemma 12.8(a) on [ t ′ , min { t ′ + 1 , t ′ ′ } ] [t',\min\{t'+1,t''\}] [ t ′ , min { t ′ + 1 , t ′′ }] and Step 1, and by Lemma 12.8(c), the mean being at most half the mean-free part (Step 1). An off-branch datum U U U at a time t ∗ t_* t ∗ splits along e ( k ) e_{(k)} e ( k ) into off-branch parts:
U t r : = ( U ) e ( k ) − ⟨ ( U ) e ( k ) , R ξ ( t ∗ ) ⟩ L 2 ( T 2 ) , ∥ U − U t r ∥ H J ( T 2 ) ≤ 7 3 ∥ U − ( U ) e ( k ) ∥ H J ( T 2 ) , (146) U^{\mathrm{tr}}:=(U)_{e_{(k)}}-\langle(U)_{e_{(k)}},R_\xi(t_*)\rangle_{L^2(\mathbb{T}^2)},\qquad\lVert U-U^{\mathrm{tr}}\rVert_{H^J(\mathbb{T}^2)}\le\frac{7}{3}\lVert U-(U)_{e_{(k)}}\rVert_{H^J(\mathbb{T}^2)}, \tag*{(146)} U tr := ( U ) e ( k ) − ⟨( U ) e ( k ) , R ξ ( t ∗ ) ⟩ L 2 ( T 2 ) , ∥ U − U tr ∥ H J ( T 2 ) ≤ 3 7 ∥ U − ( U ) e ( k ) ∥ H J ( T 2 ) , (146)
componentwise, since ⟨ ( U ) e ( k ) , R ξ ⟩ L 2 ( T 2 ) = − ⟨ U − ( U ) e ( k ) , R ξ ⟩ L 2 ( T 2 ) \langle(U)_{e_{(k)}},R_\xi\rangle_{L^2(\mathbb{T}^2)}=-\langle U-(U)_{e_{(k)}},R_\xi\rangle_{L^2(\mathbb{T}^2)} ⟨( U ) e ( k ) , R ξ ⟩ L 2 ( T 2 ) = − ⟨ U − ( U ) e ( k ) , R ξ ⟩ L 2 ( T 2 ) and ∥ R ξ ∥ L 2 ( T 2 ) ≤ 4 3 \lVert R_\xi\rVert_{L^2(\mathbb{T}^2)}\le\frac{4}{3} ∥ R ξ ∥ L 2 ( T 2 ) ≤ 3 4 .
Step 3: the windows. Let k ^ 0 \hat k_0 k ^ 0 be the pulse of t a c t ( t ′ ′ ) t_{\mathrm{act}}(t'') t act ( t ′′ ) (if there is none, (145) suffices), and for i ≥ 1 i\ge1 i ≥ 1 let k ^ i : = k ^ 0 − 2 i \hat k_i:=\hat k_0-2i k ^ i := k ^ 0 − 2 i and t ^ i : = ( k ^ i + 11 12 ) τ m \hat t_i:=(\hat k_i+\frac{11}{12})\tau_m t ^ i := ( k ^ i + 12 11 ) τ m , a quarter of τ m \tau_m τ m after the activity of k ^ i \hat k_i k ^ i ; t ^ 0 : = t ′ ′ \hat t_0:=t'' t ^ 0 := t ′′ . On [ t ^ i + 1 , t ^ i ] [\hat t_{i+1},\hat t_i] [ t ^ i + 1 , t ^ i ] the velocity is a multiple of ∇ ⊥ ψ m , k ^ i \nabla^\perp\psi_{m,\hat k_i} ∇ ⊥ ψ m , k ^ i , for i ≥ 1 i\ge1 i ≥ 1 no pulse is active on [ t ^ i , t ^ i + 5 12 τ m ] [\hat t_i,\hat t_i+\frac{5}{12}\tau_m] [ t ^ i , t ^ i + 12 5 τ m ] , and, as τ m ≥ c ε m − 1 − 4 δ \tau_m\ge c\varepsilon_{m-1}^{-4\delta} τ m ≥ c ε m − 1 − 4 δ and ( 1 + ε ˉ m / ε m ) ≤ C ε m − 1 − q γ (1+\bar\varepsilon_m/\varepsilon_m)\le C\varepsilon_{m-1}^{-q\gamma} ( 1 + ε ˉ m / ε m ) ≤ C ε m − 1 − q γ (Proposition 2.1(c)), weights of exponent at most N a c t + N 0 N_{\mathrm{act}}+N_0 N act + N 0 are 1 1 1 for m m m large where the direction in (144) is not e ( k ^ i ) e_{(\hat k_i)} e ( k ^ i ) . We normalize the right side of (b) to 1 1 1 . With U − 1 † : = U † U^\dagger_{-1}:=U^\dagger U − 1 † := U † , let U i † : = ( U i − 1 † ( t ^ i ) ) t r U^\dagger_i:=(U^\dagger_{i-1}(\hat t_i))^{\mathrm{tr}} U i † := ( U i − 1 † ( t ^ i ) ) tr along e ( k ^ i ) e_{(\hat k_i)} e ( k ^ i ) , by (146) , i ≥ 0 i\ge0 i ≥ 0 . With the same letters for the tests from these data,
U † ( t ) = U i † ( t ) + ∑ i ′ = 0 i ( U i ′ − 1 † − U i ′ † ) ( t ) ( t ≤ t ^ i ) , U i † transverse to e ( k ^ i ) on [ t ^ i + 1 , t ^ i ] . U^\dagger(t)=U^\dagger_i(t)+\sum_{i'=0}^{i}(U^\dagger_{i'-1}-U^\dagger_{i'})(t)\quad(t\le\hat t_i),\qquad U^\dagger_i\text{ transverse to }e_{(\hat k_i)}\text{ on }[\hat t_{i+1},\hat t_i]. U † ( t ) = U i † ( t ) + i ′ = 0 ∑ i ( U i ′ − 1 † − U i ′ † ) ( t ) ( t ≤ t ^ i ) , U i † transverse to e ( k ^ i ) on [ t ^ i + 1 , t ^ i ] .
Three mechanisms bound the parts at the rate π 2 / 8 \pi^2/8 π 2 /8 , without compounding constants.
The part U † − U 0 † U^{\dagger}-U_{0}^{\dagger} U † − U 0 † not transverse at t ′ ′ t^{\prime\prime} t ′′ . After t a c t ( t ′ ′ ) t_{\mathrm{act}}(t^{\prime\prime}) t act ( t ′′ ) the weight grows backward at the rate 3 π 2 / 8 3\pi^{2}/8 3 π 2 /8 and the quiet bound gains π 2 / 2 = 3 π 2 / 8 + π 2 / 8 \pi^{2}/2=3\pi^{2}/8+\pi^{2}/8 π 2 /2 = 3 π 2 /8 + π 2 /8 ; before it, the crude loss ( 1 + ε ˉ m / ε m ) N 0 (1+\bar{\varepsilon}_{m}/\varepsilon_{m})^{N_{0}} ( 1 + ε ˉ m / ε m ) N 0 is absorbed by N a c t ′ ′ ≥ N a c t ′ + N 0 N_{\mathrm{act}}^{\prime\prime}\geq N_{\mathrm{act}}^{\prime}+N_{0} N act ′′ ≥ N act ′ + N 0 :
w N a c t ′ ( t ) e − 3 π 2 ( t ′ ′ − t ) / 8 ≤ w N a c t ′ ′ ( t ′ ′ ) ( t ≥ t a c t ( t ′ ′ ) ) , w_{N_{\mathrm{act}}^{\prime}}(t)e^{-3\pi^{2}(t^{\prime\prime}-t)/8}\leq w_{N_{\mathrm{act}}^{\prime\prime}}(t^{\prime\prime})\qquad(t\geq t_{\mathrm{act}}(t^{\prime\prime})), w N act ′ ( t ) e − 3 π 2 ( t ′′ − t ) /8 ≤ w N act ′′ ( t ′′ ) ( t ≥ t act ( t ′′ )) ,
( 1 + ε ˉ m / ε m ) N a c t ′ + N 0 e − 3 π 2 ( t ′ ′ − t a c t ( t ′ ′ ) ) / 8 ≤ w N a c t ′ ′ ( t ′ ′ ) , (1+\bar{\varepsilon}_{m}/\varepsilon_{m})^{N_{\mathrm{act}}^{\prime}+N_{0}}e^{-3\pi^{2}(t^{\prime\prime}-t_{\mathrm{act}}(t^{\prime\prime}))/8}\leq w_{N_{\mathrm{act}}^{\prime\prime}}(t^{\prime\prime}), ( 1 + ε ˉ m / ε m ) N act ′ + N 0 e − 3 π 2 ( t ′′ − t act ( t ′′ )) /8 ≤ w N act ′′ ( t ′′ ) ,
so w N a c t ′ ( t ) ∥ ( U † − U 0 † ) ( t ) ∥ H J ( T 2 ) ≤ C e − π 2 ( t ′ ′ − t ) / 8 w_{N_{\mathrm{act}}^{\prime}}(t)\lVert(U^{\dagger}-U_{0}^{\dagger})(t)\rVert_{H^{J}(\mathbb{T}^{2})}\leq Ce^{-\pi^{2}(t^{\prime\prime}-t)/8} w N act ′ ( t ) ∥( U † − U 0 † ) ( t ) ∥ H J ( T 2 ) ≤ C e − π 2 ( t ′′ − t ) /8 .
The parts U i − 1 † − U i † U_{i-1}^{\dagger}-U_{i}^{\dagger} U i − 1 † − U i † , i ≥ 1 i\geq1 i ≥ 1 , by induction on i i i (the third item for U i − 1 † U_{i-1}^{\dagger} U i − 1 † bounds it at the end of the stretch; the C C C below is fixed before the level). At t ^ i \hat{t}_{i} t ^ i , after the quiet stretch, the part of U i − 1 † U_{i-1}^{\dagger} U i − 1 † not transverse to e ( k ^ i ) e_{(\hat{k}_{i})} e ( k ^ i ) is mean-free (at t ^ i \hat{t}_{i} t ^ i , U i − 1 † U_{i-1}^{\dagger} U i − 1 † is transverse to e ( k ^ i − 1 ) ⊥ e ( k ^ i ) e_{(\hat{k}_{i-1})}^{\perp}e_{(\hat{k}_{i})} e ( k ^ i − 1 ) ⊥ e ( k ^ i ) , so its average along e ( k ^ i ) e_{(\hat{k}_{i})} e ( k ^ i ) is its mean) and damped by e − 5 π 2 τ m / 24 e^{-5\pi^{2}\tau_{m}/24} e − 5 π 2 τ m /24 (Lemma 12.8(c)); with the crude bound,
∥ ( U i − 1 † − U i † ) ( t ) ∥ H J ( T 2 ) ≤ C ( 1 + ε ˉ m / ε m ) N 0 e − 5 π 2 τ m / 32 e − π 2 ( t ′ ′ − t ) / 8 . \lVert(U_{i-1}^{\dagger}-U_{i}^{\dagger})(t)\rVert_{H^{J}(\mathbb{T}^{2})}\leq C(1+\bar{\varepsilon}_{m}/\varepsilon_{m})^{N_{0}}e^{-5\pi^{2}\tau_{m}/32}e^{-\pi^{2}(t^{\prime\prime}-t)/8}. ∥( U i − 1 † − U i † ) ( t ) ∥ H J ( T 2 ) ≤ C ( 1 + ε ˉ m / ε m ) N 0 e − 5 π 2 τ m /32 e − π 2 ( t ′′ − t ) /8 .
There are at most C ε m − 2 δ C\varepsilon_{m}^{-2\delta} C ε m − 2 δ of them, so their sum times ( 1 + ε ˉ m / ε m ) N a c t (1+\bar{\varepsilon}_{m}/\varepsilon_{m})^{N_{\mathrm{act}}} ( 1 + ε ˉ m / ε m ) N act is at most e − π 2 ( t ′ ′ − t ) / 8 e^{-\pi^{2}(t^{\prime\prime}-t)/8} e − π 2 ( t ′′ − t ) /8 for m m m large.
The transverse parts. By (a), ∥ U i † ( t ) ∥ H J ( T 2 ) \lVert U_{i}^{\dagger}(t)\rVert_{H^{J}(\mathbb{T}^{2})} ∥ U i † ( t ) ∥ H J ( T 2 ) is at most C C C times ∥ U i † ( t + 1 ) ∥ L 2 ( T 2 ) \lVert U_{i}^{\dagger}(t+1)\rVert_{L^{2}(\mathbb{T}^{2})} ∥ U i † ( t + 1 ) ∥ L 2 ( T 2 ) (or times ∥ U i † ( t ^ i ) ∥ H J ( T 2 ) \lVert U_{i}^{\dagger}(\hat{t}_{i})\rVert_{H^{J}(\mathbb{T}^{2})} ∥ U i † ( t ^ i ) ∥ H J ( T 2 ) near t ^ i \hat{t}_{i} t ^ i ; by (12.25) and ∥ R ξ ∥ L 2 ( T 2 ) ≤ 4 3 \lVert R_{\xi}\rVert_{L^{2}(\mathbb{T}^{2})}\leq\frac{4}{3} ∥ R ξ ∥ L 2 ( T 2 ) ≤ 3 4 this is at most 7 3 \frac{7}{3} 3 7 for i = 0 i=0 i = 0 , and at most 7 3 ∥ U i − 1 † ( t ^ i ) ∥ H J ( T 2 ) ≤ C ∥ U i − 1 † ( t ^ i + 1 ) ∥ L 2 ( T 2 ) \frac{7}{3}\lVert U_{i-1}^{\dagger}(\hat{t}_{i})\rVert_{H^{J}(\mathbb{T}^{2})}\leq C\lVert U_{i-1}^{\dagger}(\hat{t}_{i}+1)\rVert_{L^{2}(\mathbb{T}^{2})} 3 7 ∥ U i − 1 † ( t ^ i ) ∥ H J ( T 2 ) ≤ C ∥ U i − 1 † ( t ^ i + 1 ) ∥ L 2 ( T 2 ) for i ≥ 1 i\geq1 i ≥ 1 , by (a), as t ^ i − 1 − t ^ i ≥ 1 \hat{t}_{i-1}-\hat{t}_{i}\geq1 t ^ i − 1 − t ^ i ≥ 1 for m m m large), and ∥ U i † ∥ L 2 ( T 2 ) ≤ ∥ U † ∥ L 2 ( T 2 ) + ∑ i ′ ≤ i ∥ U i ′ − 1 † − U i ′ † ∥ H J ( T 2 ) ≤ C e − π 2 ( t ′ ′ − t ) / 8 \lVert U_{i}^{\dagger}\rVert_{L^{2}(\mathbb{T}^{2})}\leq\lVert U^{\dagger}\rVert_{L^{2}(\mathbb{T}^{2})}+\sum_{i^{\prime}\leq i}\lVert U_{i^{\prime}-1}^{\dagger}-U_{i^{\prime}}^{\dagger}\rVert_{H^{J}(\mathbb{T}^{2})}\leq Ce^{-\pi^{2}(t^{\prime\prime}-t)/8} ∥ U i † ∥ L 2 ( T 2 ) ≤ ∥ U † ∥ L 2 ( T 2 ) + ∑ i ′ ≤ i ∥ U i ′ − 1 † − U i ′ † ∥ H J ( T 2 ) ≤ C e − π 2 ( t ′′ − t ) /8 by Step 1 and the two items above: every step returns to the L 2 L^{2} L 2 bound. With (12.23) this gives (b).
(c), Step 1: the identity and the L 2 L^{2} L 2 bound. The first line below, for t ′ ≤ t ≤ t ′ ′ t^{\prime}\leq t\leq t^{\prime\prime} t ′ ≤ t ≤ t ′′ , follows from Proposition 12.4 (c) and (12.20); we call it the identity. Let U b r † ˚ : = U b r † − ( U b r † ) T 2 U_{\mathrm{br}}^{\mathring{\dagger}}:=U_{\mathrm{br}}^{\dagger}-(U_{\mathrm{br}}^{\dagger})_{\mathbb{T}^{2}} U br † ˚ := U br † − ( U br † ) T 2 ; it vanishes at t ′ ′ t^{\prime\prime} t ′′ and solves the mean-free part of (12.19), forced by the mean and damped at the rate 15 16 π 2 \frac{15}{16}\pi^{2} 16 15 π 2 . The identity and Proposition 12.4 (a) give the other two lines:
exp ( ∫ t ′ t Z ξ ) ⟨ R ξ ( t ) , U b r † ( t ) ⟩ L 2 ( T 2 ) = exp ( ∫ t ′ t ′ ′ Z ξ ) ⟨ R ξ ( t ′ ′ ) , 1 ⟩ L 2 ( T 2 ) = exp ( ∫ t ′ t ′ ′ Z ξ ) , \exp\left(\int_{t^{\prime}}^{t}Z_{\xi}\right)\langle R_{\xi}(t),U_{\mathrm{br}}^{\dagger}(t)\rangle_{L^{2}(\mathbb{T}^{2})}=\exp\left(\int_{t^{\prime}}^{t^{\prime\prime}}Z_{\xi}\right)\langle R_{\xi}(t^{\prime\prime}),1\rangle_{L^{2}(\mathbb{T}^{2})}=\exp\left(\int_{t^{\prime}}^{t^{\prime\prime}}Z_{\xi}\right), exp ( ∫ t ′ t Z ξ ) ⟨ R ξ ( t ) , U br † ( t ) ⟩ L 2 ( T 2 ) = exp ( ∫ t ′ t ′′ Z ξ ) ⟨ R ξ ( t ′′ ) , 1 ⟩ L 2 ( T 2 ) = exp ( ∫ t ′ t ′′ Z ξ ) ,
∣ ( U b r † ( t ) ) T 2 − exp ( ∫ t t ′ ′ Z ξ ) ∣ ≤ C b r ε ˉ m ∣ ξ ∣ ∥ U b r † ˚ ( t ) ∥ L 2 ( T 2 ) , \left|(U_{\mathrm{br}}^{\dagger}(t))_{\mathbb{T}^{2}}-\exp\left(\int_{t}^{t^{\prime\prime}}Z_{\xi}\right)\right|\leq C_{\mathrm{br}}\bar{\varepsilon}_{m}|\xi|\lVert U_{\mathrm{br}}^{\mathring{\dagger}}(t)\rVert_{L^{2}(\mathbb{T}^{2})}, ( U br † ( t ) ) T 2 − exp ( ∫ t t ′′ Z ξ ) ≤ C br ε ˉ m ∣ ξ ∣ ∥ U br † ˚ ( t ) ∥ L 2 ( T 2 ) ,
∥ U b r † ˚ ( t ) ∥ L 2 ( T 2 ) ≤ C ε ˉ m ∣ ξ ∣ ∫ t t ′ ′ e − 15 π 2 ( t ~ − t ) / 16 ∣ ( U b r † ( t ~ ) ) T 2 ∣ d t ~ . \lVert U_{\mathrm{br}}^{\mathring{\dagger}}(t)\rVert_{L^{2}(\mathbb{T}^{2})}\leq C\bar{\varepsilon}_{m}|\xi|\int_{t}^{t^{\prime\prime}}e^{-15\pi^{2}(\tilde{t}-t)/16}|(U_{\mathrm{br}}^{\dagger}(\tilde{t}))_{\mathbb{T}^{2}}|\,d\tilde{t}. ∥ U br † ˚ ( t ) ∥ L 2 ( T 2 ) ≤ C ε ˉ m ∣ ξ ∣ ∫ t t ′′ e − 15 π 2 ( t ~ − t ) /16 ∣ ( U br † ( t ~ ) ) T 2 ∣ d t ~ .
As the scalar factor changes at most at the rate π 2 / 16 \pi^{2}/16 π 2 /16 (Proposition 12.5 (c)), a bootstrap in the ratio of ∥ U b r † ˚ ∥ L 2 ( T 2 ) \lVert U_{\mathrm{br}}^{\mathring{\dagger}}\rVert_{L^{2}(\mathbb{T}^{2})} ∥ U br † ˚ ∥ L 2 ( T 2 ) to it (closing since ε ˉ m ∣ ξ ∣ ≤ 1 4 \bar{\varepsilon}_{m}|\xi|\leq\frac{1}{4} ε ˉ m ∣ ξ ∣ ≤ 4 1 , C b r ε ˉ m ∣ ξ ∣ ≤ 1 3 C_{\mathrm{br}}\bar{\varepsilon}_{m}|\xi|\leq\frac{1}{3} C br ε ˉ m ∣ ξ ∣ ≤ 3 1 and the constant of (12.12) is at most 2 π 20 2\pi\sqrt{20} 2 π 20 ) gives (c) in L 2 ( T 2 ) L^{2}(\mathbb{T}^{2}) L 2 ( T 2 ) , for every such t ′ ′ t^{\prime\prime} t ′′ .
Step 2: the pulses. If no pulse is active in the block before t ′ ′ t^{\prime\prime} t ′′ , then the velocity vanishes on the block up to t ′ ′ t^{\prime\prime} t ′′ , U b r † ( t ) U_{\mathrm{br}}^{\dagger}(t) U br † ( t ) is the constant exp ∫ t t ′ ′ Z ξ \exp\int_{t}^{t^{\prime\prime}}Z_{\xi} exp ∫ t t ′′ Z ξ , by the identity, and the left side of the second line of (c) vanishes. Otherwise, with k ^ i , t ^ i \hat{k}_{i},\hat{t}_{i} k ^ i , t ^ i as in (b), Step 3, let U b r , i † U_{\mathrm{br},i}^{\dagger} U br , i † be the branch test from the quiet time t ^ i \hat{t}_{i} t ^ i ; it is transverse on [ t ^ i + 1 , t ^ i ] [\hat{t}_{i+1},\hat{t}_{i}] [ t ^ i + 1 , t ^ i ] . Its difference from the conjugate scalar factor, zero at t ^ i \hat{t}_{i} t ^ i , is forced transversally with size C ε ˉ m ∣ ξ ∣ C\bar{\varepsilon}_{m}|\xi| C ε ˉ m ∣ ξ ∣ times the scalar factor, as ∣ Z ξ ∣ ≤ C ( ε ˉ m ∣ ξ ∣ ) 2 |Z_{\xi}|\leq C(\bar{\varepsilon}_{m}|\xi|)^{2} ∣ Z ξ ∣ ≤ C ( ε ˉ m ∣ ξ ∣ ) 2 by Proposition 12.4 (b). By Step 1 at t + 1 t+1 t + 1 and the forced bounds of (a) on [ t , t + 1 ] [t,t+1] [ t , t + 1 ] (from the datum 0 0 0 at t ^ i \hat{t}_{i} t ^ i when t + 1 > t ^ i t+1>\hat{t}_{i} t + 1 > t ^ i ), over which the scalar factor changes in modulus by at most e π 2 / 16 e^{\pi^2/16} e π 2 /16 (Proposition 12.5 (c)), with C t r C_{\mathrm{tr}} C tr independent of the contrast,
∥ U b r , i † ( t ) − exp ( ∫ t t ^ i Z ξ ) ∥ H J ( T 2 ) ≤ C t r ε ˉ m ∣ ξ ∣ ∣ exp ( ∫ t t ^ i Z ξ ) ∣ ( t ^ i + 1 ≤ t ≤ t ^ i ) . \left\|U_{\mathrm{br},i}^{\dagger}(t)-\exp\left(\int_{t}^{\hat{t}_i}\mathcal{Z}_{\xi}\right)\right\|_{H^J(\mathbb{T}^2)}\leq C_{\mathrm{tr}}\bar{\varepsilon}_m|\xi|\left|\exp\left(\int_{t}^{\hat{t}_i}\mathcal{Z}_{\xi}\right)\right| \qquad(\hat{t}_{i+1}\leq t\leq\hat{t}_i). U br , i † ( t ) − exp ( ∫ t t ^ i Z ξ ) H J ( T 2 ) ≤ C tr ε ˉ m ∣ ξ ∣ exp ( ∫ t t ^ i Z ξ ) ( t ^ i + 1 ≤ t ≤ t ^ i ) .
The remainder U o f f , i † U_{\mathrm{off},i}^{\dagger} U off , i † below is off-branch at t ^ i \hat{t}_i t ^ i , by the identity, and U o f f , 0 † = 0 U_{\mathrm{off},0}^{\dagger}=0 U off , 0 † = 0 :
U o f f , i † : = U b r † − exp ( ∫ t ^ i t ′ ′ Z ξ ) U b r , i † , U_{\mathrm{off},i}^{\dagger}:=U_{\mathrm{br}}^{\dagger}-\exp\left(\int_{\hat{t}_i}^{t''}\mathcal{Z}_{\xi}\right)U_{\mathrm{br},i}^{\dagger}, U off , i † := U br † − exp ( ∫ t ^ i t ′′ Z ξ ) U br , i † ,
U b r † ( t ) − exp ( ∫ t t ′ ′ Z ξ ) = U o f f , i † ( t ) + exp ( ∫ t ^ i t ′ ′ Z ξ ) ( U b r , i † ( t ) − exp ( ∫ t t ^ i Z ξ ) ) . U_{\mathrm{br}}^{\dagger}(t)-\exp\left(\int_{t}^{t''}\mathcal{Z}_{\xi}\right)=U_{\mathrm{off},i}^{\dagger}(t)+\exp\left(\int_{\hat{t}_i}^{t''}\mathcal{Z}_{\xi}\right)\left(U_{\mathrm{br},i}^{\dagger}(t)-\exp\left(\int_{t}^{\hat{t}_i}\mathcal{Z}_{\xi}\right)\right). U br † ( t ) − exp ( ∫ t t ′′ Z ξ ) = U off , i † ( t ) + exp ( ∫ t ^ i t ′′ Z ξ ) ( U br , i † ( t ) − exp ( ∫ t t ^ i Z ξ ) ) .
By (b), with its constant C o f f C_{\mathrm{off}} C off (exponents N a c t ′ + N 0 N'_{\mathrm{act}}+N_0 N act ′ + N 0 at t ^ i \hat{t}_i t ^ i , where the weight is 1, and N a c t ′ N'_{\mathrm{act}} N act ′ at t t t ; at most N a c t + N 0 N_{\mathrm{act}}+N_0 N act + N 0 , for which the level is taken), and Proposition 12.5 (c),
∥ U o f f , i † ( t ) ∥ H J ( T 2 ) ; N a c t ′ , t ≤ 2 C o f f e − π 2 ( t ^ i − t ) / 16 ∥ U o f f , i † ( t ^ i ) ∥ H J ( T 2 ) ∣ exp ( ∫ t t ′ ′ Z ξ ) ∣ ∣ exp ( ∫ t ^ i t ′ ′ Z ξ ) ∣ . \left\|U_{\mathrm{off},i}^{\dagger}(t)\right\|_{H^J(\mathbb{T}^2);N'_{\mathrm{act}},t}\leq2C_{\mathrm{off}}e^{-\pi^2(\hat{t}_i-t)/16}\left\|U_{\mathrm{off},i}^{\dagger}(\hat{t}_i)\right\|_{H^J(\mathbb{T}^2)}\frac{\left|\exp\left(\int_{t}^{t''}\mathcal{Z}_{\xi}\right)\right|}{\left|\exp\left(\int_{\hat{t}_i}^{t''}\mathcal{Z}_{\xi}\right)\right|}. U off , i † ( t ) H J ( T 2 ) ; N act ′ , t ≤ 2 C off e − π 2 ( t ^ i − t ) /16 U off , i † ( t ^ i ) H J ( T 2 ) exp ( ∫ t ^ i t ′′ Z ξ ) exp ( ∫ t t ′′ Z ξ ) .
Step 3: the induction. As t ^ i − t ^ i + 1 ≥ 5 12 τ m \hat{t}_i-\hat{t}_{i+1}\geq\frac{5}{12}\tau_m t ^ i − t ^ i + 1 ≥ 12 5 τ m , for m m m large:
∥ U o f f , i † ( t ^ i ) ∥ H J ( T 2 ) ≤ 4 C t r ε ˉ m ∣ ξ ∣ ∣ exp ( ∫ t ^ i t ′ ′ Z ξ ) ∣ (true for i = 0 ) \left\|U_{\mathrm{off},i}^{\dagger}(\hat{t}_i)\right\|_{H^J(\mathbb{T}^2)}\leq4C_{\mathrm{tr}}\bar{\varepsilon}_m|\xi|\left|\exp\left(\int_{\hat{t}_i}^{t''}\mathcal{Z}_{\xi}\right)\right| \qquad\text{(true for }i=0\text{)} U off , i † ( t ^ i ) H J ( T 2 ) ≤ 4 C tr ε ˉ m ∣ ξ ∣ exp ( ∫ t ^ i t ′′ Z ξ ) (true for i = 0 )
⟹ (c) at t holds with ( 8 C o f f C t r e − π 2 ( t ^ i − t ) / 16 + 2 C t r ) for C on [ t ^ i + 1 , t ^ i ] , by Step 2 \Longrightarrow\text{(c) at }t\text{ holds with }\left(8C_{\mathrm{off}}C_{\mathrm{tr}}e^{-\pi^2(\hat{t}_i-t)/16}+2C_{\mathrm{tr}}\right)\text{ for }C\text{ on }[\hat{t}_{i+1},\hat{t}_i],\text{ by Step 2} ⟹ (c) at t holds with ( 8 C off C tr e − π 2 ( t ^ i − t ) /16 + 2 C tr ) for C on [ t ^ i + 1 , t ^ i ] , by Step 2
⟹ ∥ U o f f , i + 1 † ( t ^ i + 1 ) ∥ H J ( T 2 ) ≤ 4 C t r ε ˉ m ∣ ξ ∣ ∣ exp ( ∫ t ^ i + 1 t ′ ′ Z ξ ) ∣ at t = t ^ i + 1 , as 4 C o f f e − 5 π 2 τ m / 192 ≤ 1. \Longrightarrow\left\|U_{\mathrm{off},i+1}^{\dagger}(\hat{t}_{i+1})\right\|_{H^J(\mathbb{T}^2)}\leq4C_{\mathrm{tr}}\bar{\varepsilon}_m|\xi|\left|\exp\left(\int_{\hat{t}_{i+1}}^{t''}\mathcal{Z}_{\xi}\right)\right| \qquad\text{at }t=\hat{t}_{i+1},\text{ as }4C_{\mathrm{off}}e^{-5\pi^2\tau_m/192}\leq1. ⟹ U off , i + 1 † ( t ^ i + 1 ) H J ( T 2 ) ≤ 4 C tr ε ˉ m ∣ ξ ∣ exp ( ∫ t ^ i + 1 t ′′ Z ξ ) at t = t ^ i + 1 , as 4 C off e − 5 π 2 τ m /192 ≤ 1.
Insertions against tests. In a Duhamel term an insertion acts at each time on the previous term, and the result is evolved by the frozen operator; by (141) , the pairing of the next term with a test is therefore an integral in time of the pairing of the previous term with the adjoint insertion I ∗ I^* I ∗ applied to the test. So what matters is what I ∗ I^* I ∗ does to a test at one time: how many derivatives it takes, and how large a multiple of the branch test its output contains. The pairing with R ξ R_{\xi} R ξ measures that multiple; the rest of the output is orthogonal to R ξ R_{\xi} R ξ , and the test it starts decays at the cell rate (Lemma 12.9 (b)). The insertion acts at the cell rate, and its pairing with R ξ R_{\xi} R ξ is at most C κ m / ε ˉ m 2 C\kappa_m/\bar{\varepsilon}_m^2 C κ m / ε ˉ m 2 times the factor ε ˉ m ( ∣ ξ ∣ + ρ ( m ) − 1 ) \bar{\varepsilon}_m(|\xi|+\rho(m)^{-1}) ε ˉ m ( ∣ ξ ∣ + ρ ( m ) − 1 ) , the corrector amplitude divided by the length on which the field varies. Indeed, paired with 1 1 1 , the outer divergence of the insertion leaves only ε m ξ \varepsilon_m\xi ε m ξ times a cell mean, R ξ R_{\xi} R ξ differs from 1 1 1 by the size of the corrector, and the commutators of (142) carry the corrector amplitude, or ε m \varepsilon_m ε m , divided by ρ ( m ) \rho(m) ρ ( m ) . If the test is itself nearly constant, as the branch test is, the innermost derivative of the insertion gives a second factor, at most the same one. The factor, and the rate at which an insertion acts when it carries the factor twice, satisfy
ε ˉ m ( ∣ ξ ∣ + 1 ρ ( m ) ) ≤ 1 , κ m − 1 ∣ ξ ∣ 2 + 1 h m ≤ κ m − 2 ε ˉ m 2 ε ˉ m 2 ( ∣ ξ ∣ + 1 ρ ( m ) ) 2 ≤ 2 ( κ m − 1 ∣ ξ ∣ 2 + 1 h m ) (147) \bar{\varepsilon}_m\left(|\xi|+\frac{1}{\rho(m)}\right)\leq1,\qquad\kappa_{m-1}|\xi|^2+\frac{1}{h_m}\leq\frac{\kappa_{m-2}}{\bar{\varepsilon}_m^2}\bar{\varepsilon}_m^2\left(|\xi|+\frac{1}{\rho(m)}\right)^2\leq2\left(\kappa_{m-1}|\xi|^2+\frac{1}{h_m}\right) \tag*{(147)} ε ˉ m ( ∣ ξ ∣ + ρ ( m ) 1 ) ≤ 1 , κ m − 1 ∣ ξ ∣ 2 + h m 1 ≤ ε ˉ m 2 κ m − 2 ε ˉ m 2 ( ∣ ξ ∣ + ρ ( m ) 1 ) 2 ≤ 2 ( κ m − 1 ∣ ξ ∣ 2 + h m 1 ) (147)
by (96) for m m m large, which gives ε ˉ m / ρ ( m ) ≤ 3 4 \bar{\varepsilon}_m/\rho(m)\leq\frac{3}{4} ε ˉ m / ρ ( m ) ≤ 4 3 , by ε ˉ m ∣ ξ ∣ ≤ 2 c l o ≤ 1 4 \bar{\varepsilon}_m|\xi|\leq2c_{\mathrm{lo}}\leq\frac{1}{4} ε ˉ m ∣ ξ ∣ ≤ 2 c lo ≤ 4 1 , and, for the bounds on the rate, by the identities of the Conventions. So over a block the rate (147) integrates to at most 2 ( 1 + ρ ( m ) ∣ ξ ∣ ) 2 2(1+\rho(m)|\xi|)^2 2 ( 1 + ρ ( m ) ∣ ξ ∣ ) 2 . A single factor is the cost of passing between the branch test and the tests orthogonal to R ξ R_\xi R ξ . These decay faster than the scalar factor of the branch changes (Proposition 12.5(c)), so leaving the branch and returning cost the factor twice over about one cell time, at most a constant times the cost of one cell time on the branch.
Lemma 12.10(a),(b) makes these bounds precise, and (c) is the forward bound used in Lemma 12.12(iii). The map I ∗ I^* I ∗ keeps the norm (144) , as, near a pulse, insertions do not depend on the coordinate along its flow.
Lemma 12.10 (Insertions against tests). Consider the moment terms of a Duhamel term on [ t 1 , t 2 ] [t_1,t_2] [ t 1 , t 2 ] , with moment order N 1 N_1 N 1 . Let a group and a lower group be given, let d d d be the order of the terms of (12.21) that connect them (the order of the insertion that connects them, or 1 for the commutator terms), and let J , N a c t ≥ 0 J,N_{\mathrm{act}}\geq0 J , N act ≥ 0 be integers, with J + d − 1 J+d-1 J + d − 1 at most the derivative counts of Subsection 12.6. There exists C C C such that, for almost every t ∈ [ t 1 , t 2 ] t\in[t_1,t_2] t ∈ [ t 1 , t 2 ] , every real ξ \xi ξ in the zone, every array U † U^\dagger U † of fields in H J + d ( T 2 ) V H^{J+d}(T^2)^V H J + d ( T 2 ) V indexed by the group and every 0 ≤ N a c t ′ ≤ N a c t 0\leq N'_{\mathrm{act}}\leq N_{\mathrm{act}} 0 ≤ N act ′ ≤ N act , the following hold.
(a) Mapping, and the pairing with R ξ R_\xi R ξ . The first bound below holds; if moreover ε ˉ m ∣ ξ ∣ ≤ 2 c l o \bar{\varepsilon}_m|\xi|\leq2c_{\mathrm{lo}} ε ˉ m ∣ ξ ∣ ≤ 2 c lo and J + d ≥ max { d 1 , … , d N } + 3 J+d\geq\max\{d_1,\ldots,d_N\}+3 J + d ≥ max { d 1 , … , d N } + 3 , so does the second:
∥ I ∗ U † ∥ H J ( T 2 ) ; N a c t ′ , t ≤ C κ k m ε m 2 ∥ U † ∥ H J + d ( T 2 ) ; N a c t ′ , t , \|I^*U^\dagger\|_{H^J(T^2);N'_{\mathrm{act}},t}\leq C\frac{\kappa^{{k_m}}}{\varepsilon_m^2}\|U^\dagger\|_{H^{J+d}(T^2);N'_{\mathrm{act}},t}, ∥ I ∗ U † ∥ H J ( T 2 ) ; N act ′ , t ≤ C ε m 2 κ k m ∥ U † ∥ H J + d ( T 2 ) ; N act ′ , t , ∣ ⟨ R ξ ( t ) , I ∗ U † ⟩ ∣ ≤ C κ k m ε m 2 ε ˉ m ( ∣ ξ ∣ + 1 ρ ( m ) ) ∥ U † ∥ H J + d ( T 2 ) . |\langle R_\xi(t),I^*U^\dagger\rangle|\leq C\frac{\kappa^{k_m}}{\varepsilon_m^2}\bar{\varepsilon}_m\left(|\xi|+\frac{1}{\rho(m)}\right)\|U^\dagger\|_{H^{J+d}(T^2)}. ∣ ⟨ R ξ ( t ) , I ∗ U † ⟩ ∣ ≤ C ε m 2 κ k m ε ˉ m ( ∣ ξ ∣ + ρ ( m ) 1 ) ∥ U † ∥ H J + d ( T 2 ) . (b) The branch test. Let ε m ∣ ξ ∣ ≤ 2 c l o \varepsilon_m|\xi|\leq2c_{\mathrm{lo}} ε m ∣ ξ ∣ ≤ 2 c lo , let t ′ ′ ≥ t t''\geq t t ′′ ≥ t be a time as in Lemma 12.9(c), and let every entry of U † U^\dagger U † be the branch test of that lemma at t t t times a constant vector of C V \mathbb{C}^V C V , these vectors having square sum one in ∣ ⋅ ∣ ω |\cdot|_\omega ∣ ⋅ ∣ ω . Then
∣ ⟨ R ξ ( t ) , I ∗ U † ⟩ ∣ ≤ C ( κ m − 1 ∣ ξ ∣ 2 + 1 h m ) ∣ exp ( ∫ t t ′ ′ Z ξ ( t ~ ) d t ~ ) ∣ , |\langle R_\xi(t),I^*U^\dagger\rangle|\leq C\left(\kappa_{m-1}|\xi|^2+\frac{1}{h_m}\right)\left|\exp\left(\int_t^{t''}Z_\xi(\tilde t)\,\mathrm{d}\tilde t\right)\right|, ∣ ⟨ R ξ ( t ) , I ∗ U † ⟩ ∣ ≤ C ( κ m − 1 ∣ ξ ∣ 2 + h m 1 ) exp ( ∫ t t ′′ Z ξ ( t ~ ) d t ~ ) , ∥ I ∗ U † ∥ H J ( T 2 ) ; N a c t ′ , t ≤ C κ k m ε m 2 ε ˉ m ( ∣ ξ ∣ + 1 ρ ( m ) ) ∣ exp ( ∫ t t ′ ′ Z ξ ( t ~ ) d t ~ ) ∣ . (148) \|I^*U^\dagger\|_{H^J(T^2);N'_{\mathrm{act}},t}\leq C\frac{\kappa^{k_m}}{\varepsilon_m^2}\bar{\varepsilon}_m\left(|\xi|+\frac{1}{\rho(m)}\right)\left|\exp\left(\int_t^{t''}Z_\xi(\tilde t)\,\mathrm{d}\tilde t\right)\right|. \tag*{(148)} ∥ I ∗ U † ∥ H J ( T 2 ) ; N act ′ , t ≤ C ε m 2 κ k m ε ˉ m ( ∣ ξ ∣ + ρ ( m ) 1 ) exp ( ∫ t t ′′ Z ξ ( t ~ ) d t ~ ) . (148) (c) Forward. For every array U ′ U' U ′ in H J + d ( T 2 ) V H^{J+d}(T^2)^V H J + d ( T 2 ) V indexed by the lower group and every ( j , i ) (j,i) ( j , i ) in the group,
∥ ∑ ( j ′ , i ′ ) I ( j , i ) , ( j ′ , i ′ ) ( t ) U ( j ′ , i ′ ) ′ ∥ H J ( T 2 ) ≤ C κ k m ε m 2 ∥ U ′ ∥ H J + d ( T 2 ) . \left\|\sum_{(j',i')} \mathcal{I}_{(j,i),(j',i')}(t)U'_{(j',i')}\right\|_{H^J(T^2)}
\leq C\frac{\kappa^{k_m}}{\varepsilon_m^2}\|U'\|_{H^{J+d}(T^2)}. ( j ′ , i ′ ) ∑ I ( j , i ) , ( j ′ , i ′ ) ( t ) U ( j ′ , i ′ ) ′ H J ( T 2 ) ≤ C ε m 2 κ k m ∥ U ′ ∥ H J + d ( T 2 ) .
Proof. We bound one entry and sum by Schur’s test in the components with the weighted row and column sums of the matrices of the strengths, so that C C C does not depend on the number of components. Recall ε m ≤ ε ˉ m \varepsilon_m\leq\bar{\varepsilon}_m ε m ≤ ε ˉ m , by Proposition 2.1(c), and ε ˉ m / ρ ( m ) ≤ 3 4 \bar{\varepsilon}_m/\rho(m)\leq\frac{3}{4} ε ˉ m / ρ ( m ) ≤ 4 3 (paragraph of (147) ).
(a), the mapping, and (c). In cell time the insertion part of an entry of I ∗ I^* I ∗ is, up to sign and terms with ε m / ρ ( m ) \varepsilon_m/\rho(m) ε m / ρ ( m ) in place of a derivative,
∑ d ′ ≤ d ( ∇ + i ε m ξ ) d ′ − 1 ⋅ ( A d ′ ∗ ( ∇ + i ε m ξ ) ⋅ ) , \sum_{d'\leq d}(\nabla+i\varepsilon_m\xi)^{d'-1}\cdot\left(\mathcal{A}_{d'}^*(\nabla+i\varepsilon_m\xi)\cdot\right), d ′ ≤ d ∑ ( ∇ + i ε m ξ ) d ′ − 1 ⋅ ( A d ′ ∗ ( ∇ + i ε m ξ ) ⋅ ) ,
with coefficients bounded with their z z z -derivatives by clause (ii) of the insertions and ρ ( m ) ≤ C ε m − 1 \rho(m)\leq C\varepsilon_m^{-1} ρ ( m ) ≤ C ε m − 1 ; the commutator part has coefficients ε m / ρ ( m ) \varepsilon_m/\rho(m) ε m / ρ ( m ) , ( ε m / ρ ( m ) ) 2 (\varepsilon_m/\rho(m))^2 ( ε m / ρ ( m ) ) 2 and at most C ε ˉ m / ρ ( m ) C\bar{\varepsilon}_m/\rho(m) C ε ˉ m / ρ ( m ) , with their z z z -derivatives, by (133) and the display in the proof of Lemma 12.9(a). The forward operator of (c) is the adjoint of this one, with the same conjugated matrix and at most J + 1 ≤ J + d − 1 J+1\leq J+d-1 J + 1 ≤ J + d − 1 derivatives on the coefficients of an insertion. For the weights, let k k k be the pulse of (144) at t t t , with ∣ t a c t ( t ) − k τ m ∣ ≤ 2 3 τ m \lvert t_{\mathrm{act}}(t)-k\tau_m\rvert\le\frac{2}{3}\tau_m ∣ t act ( t ) − k τ m ∣ ≤ 3 2 τ m . Where the activity weight exceeds 1, t t t lies within C log ( 1 + ε ˉ m / ε m ) C\log(1+\bar{\varepsilon}_m/\varepsilon_m) C log ( 1 + ε ˉ m / ε m ) cell times after the activity of k k k , so ∣ t − k τ m ∣ ≤ 3 4 τ m \lvert t-k\tau_m\rvert\le\frac{3}{4}\tau_m ∣ t − k τ m ∣ ≤ 4 3 τ m for m m m large, and there the coefficients do not depend on z ⋅ e ( k ) z\cdot e_{(k)} z ⋅ e ( k ) , by item (i) of the insertions, the velocity being parallel to e ( k ) e_{(k)} e ( k ) ; hence
∥ I ⋆ U † − ( I ⋆ U † ) e ( k ) ∥ H J ( T 2 ) ≤ C κ m ε m 2 ∥ U † − ( U † ) e ( k ) ∥ H J + d ( T 2 ) . \left\lVert\mathcal{I}^{\star}U^{\dagger}-(\mathcal{I}^{\star}U^{\dagger})_{e_{(k)}}\right\rVert_{H^J(\mathbb{T}^2)}\le C\frac{\kappa_m}{\varepsilon_m^2}\left\lVert U^{\dagger}-(U^{\dagger})_{e_{(k)}}\right\rVert_{H^{J+d}(\mathbb{T}^2)}. I ⋆ U † − ( I ⋆ U † ) e ( k ) H J ( T 2 ) ≤ C ε m 2 κ m U † − ( U † ) e ( k ) H J + d ( T 2 ) .
Where the weight is 1, the bound is the plain one, by the last sentence after (144) ; elsewhere I ⋆ \mathcal{I}^{\star} I ⋆ commutes with the average along e ( k ) e_{(k)} e ( k ) .
(a), the pairing with R ξ \mathcal{R}_{\xi} R ξ . For one entry, the mean of a divergence is that of its zeroth-order part:
⟨ R ξ , I ⋆ U † ⟩ = ( I ⋆ U † ) ‾ T 2 + ⟨ R ξ − 1 , I ⋆ U † ⟩ , ( ( ∇ + i ε m ξ ) ⋅ F ) T 2 = i ε m ξ ⋅ ( F ) T 2 , \langle\mathcal{R}_{\xi},\mathcal{I}^{\star}U^{\dagger}\rangle=\overline{(\mathcal{I}^{\star}U^{\dagger})}_{\mathbb{T}^2}+\langle\mathcal{R}_{\xi}-1,\mathcal{I}^{\star}U^{\dagger}\rangle,\qquad\bigl((\nabla+i\varepsilon_m\xi)\cdot F\bigr)_{\mathbb{T}^2}=i\varepsilon_m\xi\cdot(F)_{\mathbb{T}^2}, ⟨ R ξ , I ⋆ U † ⟩ = ( I ⋆ U † ) T 2 + ⟨ R ξ − 1 , I ⋆ U † ⟩ , ( ( ∇ + i ε m ξ ) ⋅ F ) T 2 = i ε m ξ ⋅ ( F ) T 2 ,
for a vector field F F F on T 2 \mathbb{T}^2 T 2 , and the other terms carry ε m / ρ ( m ) \varepsilon_m/\rho(m) ε m / ρ ( m ) , ε ˉ m / ρ ( m ) \bar{\varepsilon}_m/\rho(m) ε ˉ m / ρ ( m ) or C b r ε ˉ m ∣ ξ ∣ C_{\mathrm{br}}\bar{\varepsilon}_m\lvert\xi\rvert C br ε ˉ m ∣ ξ ∣ . The first bound with J = 0 J=0 J = 0 and N a c t ′ = 0 N'_{\mathrm{act}}=0 N act ′ = 0 gives ∥ I ⋆ U † ∥ L 2 ( T 2 ) ≤ C κ m ε m 2 ∥ U † ∥ H d ( T 2 ) \lVert\mathcal{I}^{\star}U^{\dagger}\rVert_{L^2(\mathbb{T}^2)}\le C\frac{\kappa_m}{\varepsilon_m^2}\lVert U^{\dagger}\rVert_{H^d(\mathbb{T}^2)} ∥ I ⋆ U † ∥ L 2 ( T 2 ) ≤ C ε m 2 κ m ∥ U † ∥ H d ( T 2 ) , so only ∥ U † ∥ H d ( T 2 ) \lVert U^{\dagger}\rVert_{H^d(\mathbb{T}^2)} ∥ U † ∥ H d ( T 2 ) appears.
(b) Let a a a be the array of the constant vectors, so that
U † = a ( U b r † ( t ) − exp ( ∫ t t ′ ′ Z ξ ) ) + a exp ( ∫ t t ′ ′ Z ξ ) . U^{\dagger}=a\left(U_{\mathrm{br}}^{\dagger}(t)-\exp\left(\int_t^{t''}\mathcal{Z}_{\xi}\right)\right)+a\exp\left(\int_t^{t''}\mathcal{Z}_{\xi}\right). U † = a ( U br † ( t ) − exp ( ∫ t t ′′ Z ξ ) ) + a exp ( ∫ t t ′′ Z ξ ) .
By the first bound of (a) and Lemma 12.9(c) at the order J + d J+d J + d with the exponent N a c t ′ N'_{\mathrm{act}} N act ′ , the image under I ⋆ \mathcal{I}^{\star} I ⋆ of the first part is at most C κ m ε m 2 ε ˉ m ∣ ξ ∣ ∥ exp ∫ t t ′ ′ Z ξ ∥ C\frac{\kappa_m}{\varepsilon_m^2}\bar{\varepsilon}_m\lvert\xi\rvert\left\lVert\exp\int_t^{t''}\mathcal{Z}_{\xi}\right\rVert C ε m 2 κ m ε ˉ m ∣ ξ ∣ exp ∫ t t ′′ Z ξ in the norm ∥ ⋅ ∥ H J ( T 2 ) ; N a c t ′ , t \lVert\cdot\rVert_{H^J(\mathbb{T}^2);N'_{\mathrm{act}},t} ∥ ⋅ ∥ H J ( T 2 ) ; N act ′ , t . On the second part, the innermost derivative of an insertion term gives i ε m ξ i\varepsilon_m\xi i ε m ξ , or ε m / ρ ( m ) \varepsilon_m/\rho(m) ε m / ρ ( m ) stands in its place; the commutator terms carry ( ε m / ρ ( m ) ) ε m ∣ ξ ∣ (\varepsilon_m/\rho(m))\varepsilon_m\lvert\xi\rvert ( ε m / ρ ( m )) ε m ∣ ξ ∣ , C ε m / ρ ( m ) C\varepsilon_m/\rho(m) C ε m / ρ ( m ) and ( ε m / ρ ( m ) ) 2 (\varepsilon_m/\rho(m))^2 ( ε m / ρ ( m ) ) 2 . So the image of the second part is at most C κ m ε m 2 ε ˉ m ( ∣ ξ ∣ + 1 ρ ( m ) ) ∥ exp ∫ t t ′ ′ Z ξ ∥ C\frac{\kappa_m}{\varepsilon_m^2}\bar{\varepsilon}_m\left(\lvert\xi\rvert+\frac{1}{\rho(m)}\right)\left\lVert\exp\int_t^{t''}\mathcal{Z}_{\xi}\right\rVert C ε m 2 κ m ε ˉ m ( ∣ ξ ∣ + ρ ( m ) 1 ) exp ∫ t t ′′ Z ξ in H J ( T 2 ) H^J(\mathbb{T}^2) H J ( T 2 ) ; where the weight exceeds 1 this field is invariant along e ( k ) e_{(k)} e ( k ) , so its weighted norm is at most 2 \sqrt{2} 2 times this in every case. This is the second line of (148) . For the first, in the mean a term of order d ′ d' d ′ carries d ′ ≥ 2 d'\ge2 d ′ ≥ 2 quantities among ε m ∣ ξ ∣ \varepsilon_m\lvert\xi\rvert ε m ∣ ξ ∣ , ε m / ρ ( m ) \varepsilon_m/\rho(m) ε m / ρ ( m ) and the bound C ε ˉ m ∣ ξ ∣ ∥ exp ∫ t t ′ ′ Z ξ ∥ C\bar{\varepsilon}_m\lvert\xi\rvert\left\lVert\exp\int_t^{t''}\mathcal{Z}_{\xi}\right\rVert C ε ˉ m ∣ ξ ∣ exp ∫ t t ′′ Z ξ of ( ∇ + i ε m ξ ) U b r † ( t ) (\nabla+i\varepsilon_m\xi)U_{\mathrm{br}}^{\dagger}(t) ( ∇ + i ε m ξ ) U br † ( t ) given by Lemma 12.9(c). The commutators carry ( ε m / ρ ( m ) ) ε m ∣ ξ ∣ (\varepsilon_m/\rho(m))\varepsilon_m\lvert\xi\rvert ( ε m / ρ ( m )) ε m ∣ ξ ∣ and ( ε m / ρ ( m ) ) 2 (\varepsilon_m/\rho(m))^2 ( ε m / ρ ( m ) ) 2 , and the drift, of mean zero against a constant, carries ( ε ˉ m / ρ ( m ) ) ε ˉ m ∣ ξ ∣ (\bar{\varepsilon}_m/\rho(m))\bar{\varepsilon}_m\lvert\xi\rvert ( ε ˉ m / ρ ( m )) ε ˉ m ∣ ξ ∣ ; against R ξ − 1 \mathcal{R}_{\xi}-1 R ξ − 1 the second line gains C b r ε ˉ m ∣ ξ ∣ C_{\mathrm{br}}\bar{\varepsilon}_m\lvert\xi\rvert C br ε ˉ m ∣ ξ ∣ . We conclude with (147) .
The derivative count. In H J ( T 2 ) H^J(\mathbb{T}^2) H J ( T 2 ) , the product rule puts at most J + d − 1 J+d-1 J + d − 1 derivatives on the coefficients of an insertion of order d d d , as d − 1 d-1 d − 1 of the derivatives stand outside them. □
Duhamel terms over a renewal block
We follow one Duhamel term of the frozen operator, centred at y 0 y_0 y 0 , over a segment that ends at a block boundary, and look at the Fourier transform of its last term at the end of the segment, at the low frequencies. There, by (143) , it is ε ˉ m 2 \bar{\varepsilon}_m^2 ε ˉ m 2 times the pairing with the branch test, which is 1 at the end, and Lemma 12.11 shows this pairing across the segment.
Over a complete block, Lemma 12.12 splits the term into a model Π b l k , g \Pi_{\mathrm{blk},g} Π blk , g , smooth at the scale ρ ( m ) \rho(m) ρ ( m ) , and an error smaller than any power of ε m − 1 \varepsilon_{m-1} ε m − 1 ; the model is the last term at the end of the block, cut off smoothly to low frequencies, which is its part in the principal branch. The low-frequency cutoff is H m o ( ξ ) : = H ( 2 − ε ˉ m ∣ ξ ∣ / c o ) H_m^{\mathrm{o}}(\xi):=H(2-\bar{\varepsilon}_m\lvert\xi\rvert/c_{\mathrm{o}}) H m o ( ξ ) := H ( 2 − ε ˉ m ∣ ξ ∣ / c o ) , where H H H is the smooth step of (11) and c o c_{\mathrm{o}} c o the absolute constant of Proposition 12.5: the cutoff is 1 where ε ˉ m ∣ ξ ∣ ≤ c o \bar{\varepsilon}_m\lvert\xi\rvert\le c_{\mathrm{o}} ε ˉ m ∣ ξ ∣ ≤ c o and 0 where ε ˉ m ∣ ξ ∣ ≥ 2 c o \bar{\varepsilon}_m\lvert\xi\rvert\ge2c_{\mathrm{o}} ε ˉ m ∣ ξ ∣ ≥ 2 c o . For a field F F F on R 2 \mathbb{R}^2 R 2 (or any function of ξ \xi ξ with values in C V \mathbb{C}^V C V in place of F ^ \widehat{F} F ) and a datum g g g , we write
∣ F ^ ∣ ρ ( m ) , ω , N 1 ; ξ : = ∑ j ≤ N 1 ∣ ∇ ξ j F ^ ( ξ ) ∣ ω ρ ( m ) j j ! , ∥ g ∥ ρ ( m ) , ω , N ′ ; ξ : = ∑ ∣ i ∣ ≤ N ′ ∥ B ( ( y − y 0 ρ ( m ) ) i g ) ( ξ , ⋅ ) ∥ L ω 2 ( T 2 ) . \lvert\widehat{F}\rvert_{\rho(m),\omega,N_1;\xi}:=\sum_{j\le N_1}\frac{\lvert\nabla_\xi^j\widehat{F}(\xi)\rvert_\omega}{\rho(m)^j j!},\qquad\lVert g\rVert_{\rho(m),\omega,N';\xi}:=\sum_{\lvert i\rvert\le N'}\left\lVert\mathcal{B}\left(\left(\frac{y-y_0}{\rho(m)}\right)^i g\right)(\xi,\cdot)\right\rVert_{L^2_\omega(\mathbb{T}^2)}. ∣ F ∣ ρ ( m ) , ω , N 1 ; ξ := j ≤ N 1 ∑ ρ ( m ) j j ! ∣ ∇ ξ j F ( ξ ) ∣ ω , ∥ g ∥ ρ ( m ) , ω , N ′ ; ξ := ∣ i ∣ ≤ N ′ ∑ B ( ( ρ ( m ) y − y 0 ) i g ) ( ξ , ⋅ ) L ω 2 ( T 2 ) .
with ∇ ξ j \nabla_\xi^j ∇ ξ j the array of the derivatives of order j j j in ξ \xi ξ , and ∣ c ∣ ω 2 : = ∑ ν ω ν − 1 ∣ c ν ∣ 2 |c|_\omega^2 := \sum_\nu\omega_\nu^{-1}|c_\nu|^2 ∣ c ∣ ω 2 := ∑ ν ω ν − 1 ∣ c ν ∣ 2 for c ∈ C V c\in\mathbb{C}^V c ∈ C V ; the second is the moment data of g g g at ξ \xi ξ .
Without insertions, the left side of (149) is, up to the corrector, the Fourier transform of the datum, whose frequency derivatives are its moments. By Lemma 12.11, insertions of any orders change this only by a polynomial factor in ρ ( m ) ∣ ξ ∣ \rho(m)|\xi| ρ ( m ) ∣ ξ ∣ , with no power of the contrast, on a segment that may be a whole block or arbitrarily short. In the proof (Figure 12.2 ) we pair the moment terms with the branch test and with off-branch tests: insertions act on the branch with the square of the factor ε ˉ m ( ∣ ξ ∣ + ρ ( m ) − 1 ) \bar{\varepsilon}_m(|\xi|+\rho(m)^{-1}) ε ˉ m ( ∣ ξ ∣ + ρ ( m ) − 1 ) of (147) , a passage between the branch and the off-branch part costs the factor once, and off-branch tests decay at the cell rate, so a triangular Grönwall inequality closes with a polynomial loss in ρ ( m ) ∣ ξ ∣ \rho(m)|\xi| ρ ( m ) ∣ ξ ∣ . The part of the datum off the branch is read by tests of norm at most one, which is why no derivative of the datum enters.
Figure 12.2. The proof of Lemma 12.11 in cell time. Tracks: pairings with the branch test (1 at t 2 t_2 t 2 ) and with off-branch tests; arrows from g g g : the split of the datum; ticks: insertion times; dots: insertions within a track; vertical arrows: passages between the tracks; labels: the power of ε ˉ m ( ∣ ξ ∣ + ρ ( m ) − 1 ) \bar{\varepsilon}_m(|\xi|+\rho(m)^{-1}) ε ˉ m ( ∣ ξ ∣ + ρ ( m ) − 1 ) each costs; thick lower segments: decay at the cell rate.
Lemma 12.11 (The Fourier transform on a segment). Let t 2 : = ( l + 1 2 ) h m t_2 := (l+\frac{1}{2})h_m t 2 := ( l + 2 1 ) h m , t 1 ∈ [ t 2 − h m , t 2 ] t_1\in[t_2-h_m,t_2] t 1 ∈ [ t 2 − h m , t 2 ] and N 1 ≥ 0 N_1\geq0 N 1 ≥ 0 an integer, and let U [ 0 ] , … , U [ N ] U^{[0]},\ldots,U^{[N]} U [ 0 ] , … , U [ N ] be a Duhamel term on [ t 1 , t 2 ] [t_1,t_2] [ t 1 , t 2 ] , centred at y 0 y_0 y 0 , from a datum g ∈ H d 1 + ⋯ + d N − 2 N ( R V ) g\in H^{d_1+\cdots+d_N-2N}(\mathbb{R}^V) g ∈ H d 1 + ⋯ + d N − 2 N ( R V ) with compact support, with N ′ N' N ′ as for the moment terms. There exist C C C and an integer N 2 N_2 N 2 such that the function of ξ \xi ξ inside the norm on the left of (12.28) is N 1 N_1 N 1 times differentiable and
∣ exp ( − ∫ t 1 t 2 Z ξ ( t ) d t ) H m l o U [ N ] ( t 2 ) ∣ ρ ( m ) , ω , N 1 ; ξ ≤ C ( 1 + ρ ( m ) ∣ ξ ∣ ) N 2 ε m 2 ∥ g ∥ ρ ( m ) , ω , N ′ (149) \left|\exp\left(-\int_{t_1}^{t_2}\mathcal{Z}_\xi(t)\,\mathrm{d}t\right)H_m^{\mathrm{lo}}U^{[N]}(t_2)\right|_{\rho(m),\omega,N_1;\xi}
\leq C(1+\rho(m)|\xi|)^{N_2}\varepsilon_m^2\|g\|_{\rho(m),\omega,N'} \tag*{(149)} exp ( − ∫ t 1 t 2 Z ξ ( t ) d t ) H m lo U [ N ] ( t 2 ) ρ ( m ) , ω , N 1 ; ξ ≤ C ( 1 + ρ ( m ) ∣ ξ ∣ ) N 2 ε m 2 ∥ g ∥ ρ ( m ) , ω , N ′ (149)
Proof. Fix ξ \xi ξ , and let B U [ j , i ] ( t ) BU^{[j,i]}(t) B U [ j , i ] ( t ) be the Bloch sums at ξ \xi ξ of the moment terms with N 1 N_1 N 1 .
Step 1: Coordinates. Let U b r † U_{\mathrm{br}}^\dagger U br † be the branch test of Lemma 12.9(c) with t ′ ′ = t 2 t''=t_2 t ′′ = t 2 , which lies in the quiet margin at the end of the block. Each moment term ( j , i ) (j,i) ( j , i ) gets a test order N j , i N_{j,i} N j , i and an exponent of the weight, fixed in Step 7; the norm (12.23) of this order and exponent is the weighted norm of ( j , i ) (j,i) ( j , i ) . The branch coordinate and the off-branch seminorm of ( j , i ) (j,i) ( j , i ) at t t t are, with the supremum over the U † U^\dagger U † off-branch at t t t of weighted norm at most one,
∣ B U [ j , i ] ( t ) ∣ b r , t : = ∣ ⟨ B U [ j , i ] ( t ) , U b r † ( t ) ⟩ ∣ ∣ exp ∫ t 1 t 2 Z ξ ∣ , ∥ B U [ j , i ] ( t ) ∥ o f f , t : = sup U † ∣ ⟨ B U [ j , i ] ( t ) , U † ⟩ L ω 2 ( T 2 ) ∣ ∣ exp ∫ t 1 t Z ξ ∣ ; |BU^{[j,i]}(t)|_{\mathrm{br},t}:=\frac{|\langle BU^{[j,i]}(t),U_{\mathrm{br}}^\dagger(t)\rangle|}{\left|\exp\int_{t_1}^{t_2}\mathcal{Z}_\xi\right|},\qquad
\|BU^{[j,i]}(t)\|_{\mathrm{off},t}:=\frac{\sup_{U^\dagger}|\langle BU^{[j,i]}(t),U^\dagger\rangle_{L_\omega^2(T^2)}|}{\left|\exp\int_{t_1}^{t}\mathcal{Z}_\xi\right|}; ∣ B U [ j , i ] ( t ) ∣ br , t := exp ∫ t 1 t 2 Z ξ ∣ ⟨ B U [ j , i ] ( t ) , U br † ( t )⟩ ∣ , ∥ B U [ j , i ] ( t ) ∥ off , t := exp ∫ t 1 t Z ξ sup U † ∣ ⟨ B U [ j , i ] ( t ) , U † ⟩ L ω 2 ( T 2 ) ∣ ;
the scalar factors normalize the branch: on the solution R ξ ( t ) exp ( ∫ t 1 t Z ξ ) R_\xi(t)\exp\left(\int_{t_1}^{t}\mathcal{Z}_\xi\right) R ξ ( t ) exp ( ∫ t 1 t Z ξ ) the branch coordinate is 1 and the off-branch seminorm 0.
Step 2: The branch inequality. Fix a constant vector of C V \mathbb{C}^{V} C V of norm one in ∣ ⋅ ∣ ω |\cdot|_{\omega} ∣ ⋅ ∣ ω . By (141) , the derivative of the pairing of B U [ j , i ] BU^{[j,i]} B U [ j , i ] with U b r † U_{\mathrm{br}}^{\dagger} U br † times this vector is the sum over the lower ( j ′ , i ′ ) (j',i') ( j ′ , i ′ ) of the pairings of B U [ j ′ , i ′ ] BU^{[j',i']} B U [ j ′ , i ′ ] with I ( j , i ) , ( j ′ , i ′ ) ∗ I^{*}_{(j,i),(j',i')} I ( j , i ) , ( j ′ , i ′ ) ∗ applied to it. We split the latter, component by component, into a multiple of U b r † ( t ) U_{\mathrm{br}}^{\dagger}(t) U br † ( t ) and a rest off-branch at t t t : by Lemmas 12.9(c) and 12.10(b) and the chain in (147) , the multiples are at most C C C times the square of the factor ε ˉ m ( ∣ ξ ∣ + ρ ( m ) − 1 ) \bar{\varepsilon}_{m}(|\xi|+\rho(m)^{-1}) ε ˉ m ( ∣ ξ ∣ + ρ ( m ) − 1 ) , and the rest at most C C C times the factor times ∣ exp ∫ t ′ t 2 Z ξ ∣ \left|\exp\int_{t'}^{t_{2}}Z_{\xi}\right| exp ∫ t ′ t 2 Z ξ in the weighted norm of ( j ′ , i ′ ) (j',i') ( j ′ , i ′ ) . As the scalar factors multiply, integrating gives, with the sum over the terms on the right of (142) ,
∣ B U [ j , i ] ( t ) ∣ b r , t ≤ ∣ B U [ j , i ] ( t 1 ) ∣ b r , t 1 + C ∫ t 1 t ∑ ( j ′ , i ′ ) ( ε ˉ m 2 ( ∣ ξ ∣ + 1 ρ ( m ) ) ∣ B U [ j ′ , i ′ ] ( t ′ ) ∣ b r , t ′ + ε ˉ m ( ∣ ξ ∣ + 1 ρ ( m ) ) ∥ B U [ j ′ , i ′ ] ( t ′ ) ∥ o f f , t ′ ) d t ′ . \left|BU^{[j,i]}(t)\right|_{\mathrm{br},t}\leq\left|BU^{[j,i]}(t_{1})\right|_{\mathrm{br},t_{1}}+C\int_{t_{1}}^{t}\sum_{(j',i')}\left(\bar{\varepsilon}_{m}^{2}\left(|\xi|+\frac{1}{\rho(m)}\right)\left|BU^{[j',i']}(t')\right|_{\mathrm{br},t'}+\bar{\varepsilon}_{m}\left(|\xi|+\frac{1}{\rho(m)}\right)\left\|BU^{[j',i']}(t')\right\|_{\mathrm{off},t'}\right)\,dt'. B U [ j , i ] ( t ) br , t ≤ B U [ j , i ] ( t 1 ) br , t 1 + C ∫ t 1 t ( j ′ , i ′ ) ∑ ( ε ˉ m 2 ( ∣ ξ ∣ + ρ ( m ) 1 ) B U [ j ′ , i ′ ] ( t ′ ) br , t ′ + ε ˉ m ( ∣ ξ ∣ + ρ ( m ) 1 ) B U [ j ′ , i ′ ] ( t ′ ) off , t ′ ) d t ′ .
Step 3: The off-branch inequality. Let U † U^{\dagger} U † be the test on [ t 1 , t ] [t_{1},t] [ t 1 , t ] from a field off-branch at t t t of weighted norm at most one for ( j , i ) (j,i) ( j , i ) . By Lemma 12.9(b) it stays off-branch, and, for every term ( j ′ , i ′ ) (j',i') ( j ′ , i ′ ) on the right of (142) , its norm of order N j , i N_{j,i} N j , i and exponent N j ′ , i ′ N 0 N_{j',i'}N_{0} N j ′ , i ′ N 0 at t ′ t' t ′ is at most C e − π 2 ( t − t ′ ) / 8 Ce^{-\pi^{2}(t-t')/8} C e − π 2 ( t − t ′ ) /8 . By (141) its pairing with B U [ j , i ] ( t ) BU^{[j,i]}(t) B U [ j , i ] ( t ) is that at t 1 t_{1} t 1 plus the integral of the pairings of the lower terms with I ∗ U † I^{*}U^{\dagger} I ∗ U † . We split I ∗ U † ( t ′ ) I^{*}U^{\dagger}(t') I ∗ U † ( t ′ ) as in Step 2: by Lemma 12.10(a), the multiples of U b r † ( t ′ ) U_{\mathrm{br}}^{\dagger}(t') U br † ( t ′ ) are at most C C C times the factor ε ˉ m ( ∣ ξ ∣ + ρ ( m ) − 1 ) \bar{\varepsilon}_{m}(|\xi|+\rho(m)^{-1}) ε ˉ m ( ∣ ξ ∣ + ρ ( m ) − 1 ) times e − π 2 ( t − t ′ ) / 8 / ∣ exp ∫ t ′ t 2 Z ξ ∣ e^{-\pi^{2}(t-t')/8}/\left|\exp\int_{t'}^{t_{2}}Z_{\xi}\right| e − π 2 ( t − t ′ ) /8 / exp ∫ t ′ t 2 Z ξ , and the rest is at most C e − π 2 ( t − t ′ ) / 8 Ce^{-\pi^{2}(t-t')/8} C e − π 2 ( t − t ′ ) /8 in the weighted norm. With ∣ exp ( − ∫ t ′ t Z ξ ) ∣ ≤ e π 2 ( t − t ′ ) / 16 \left|\exp\left(-\int_{t'}^{t}Z_{\xi}\right)\right|\leq e^{\pi^{2}(t-t')/16} exp ( − ∫ t ′ t Z ξ ) ≤ e π 2 ( t − t ′ ) /16 (Proposition 12.5(c)),
∥ B U [ j , i ] ( t ) ∥ o f f , t ≤ C e − π 2 ( t − t 1 ) / 16 ∥ B U [ j , i ] ( t 1 ) ∥ L ω 2 ( T 2 ) + C ∫ t 1 t e − π 2 ( t − t ′ ) / 16 ∑ ( j ′ , i ′ ) ( ε ˉ m ( ∣ ξ ∣ + 1 ρ ( m ) ) ∣ B U [ j ′ , i ′ ] ( t ′ ) ∣ b r , t ′ + ∥ B U [ j ′ , i ′ ] ( t ′ ) ∥ o f f , t ′ ) d t ′ . \left\|BU^{[j,i]}(t)\right\|_{\mathrm{off},t}\leq Ce^{-\pi^{2}(t-t_{1})/16}\left\|BU^{[j,i]}(t_{1})\right\|_{L^{2}_{\omega}(\mathbb{T}^{2})}+C\int_{t_{1}}^{t}e^{-\pi^{2}(t-t')/16}\sum_{(j',i')}\left(\bar{\varepsilon}_{m}\left(|\xi|+\frac{1}{\rho(m)}\right)\left|BU^{[j',i']}(t')\right|_{\mathrm{br},t'}+\left\|BU^{[j',i']}(t')\right\|_{\mathrm{off},t'}\right)\,dt'. B U [ j , i ] ( t ) off , t ≤ C e − π 2 ( t − t 1 ) /16 B U [ j , i ] ( t 1 ) L ω 2 ( T 2 ) + C ∫ t 1 t e − π 2 ( t − t ′ ) /16 ( j ′ , i ′ ) ∑ ( ε ˉ m ( ∣ ξ ∣ + ρ ( m ) 1 ) B U [ j ′ , i ′ ] ( t ′ ) br , t ′ + B U [ j ′ , i ′ ] ( t ′ ) off , t ′ ) d t ′ .
the coupling between off-branch parts has no small coefficient.
Step 4: Data. For j ≥ 1 j\geq1 j ≥ 1 the moment terms vanish at t 1 t_{1} t 1 . For j = 0 j=0 j = 0 , ∥ U b r † ( t 1 ) ∥ L 2 ( T 2 ) ≤ C ∣ exp ∫ t 1 t 2 Z ξ ∣ \left\|U_{\mathrm{br}}^{\dagger}(t_{1})\right\|_{L^{2}(\mathbb{T}^{2})}\leq C\left|\exp\int_{t_{1}}^{t_{2}}Z_{\xi}\right| U br † ( t 1 ) L 2 ( T 2 ) ≤ C exp ∫ t 1 t 2 Z ξ by Lemma 12.9(c), and the norm (144) is at least the L 2 L^{2} L 2 norm; so
∣ B U [ 0 , i ] ( t 1 ) ∣ b r , t 1 + ∥ B U [ 0 , i ] ( t 1 ) ∥ o f f , t 1 ≤ C ∥ g ∥ ρ ( m ) , ω , N ′ ; ξ . \left|BU^{[0,i]}(t_{1})\right|_{\mathrm{br},t_{1}}+\left\|BU^{[0,i]}(t_{1})\right\|_{\mathrm{off},t_{1}}\leq C\left\|g\right\|_{\rho(m),\omega,N';\xi}. B U [ 0 , i ] ( t 1 ) br , t 1 + B U [ 0 , i ] ( t 1 ) off , t 1 ≤ C ∥ g ∥ ρ ( m ) , ω , N ′ ; ξ .
Step 5: The triangular Grönwall inequality. We claim that, at t t t , the branch coordinate and the off-branch seminorm of ( j , i ) (j,i) ( j , i ) are at most ∥ g ∥ ρ ( m ) , ω , N ′ ; ξ \left\|g\right\|_{\rho(m),\omega,N';\xi} ∥ g ∥ ρ ( m ) , ω , N ′ ; ξ times, respectively,
C j , i ( 1 + ε ˉ m 2 ( ∣ ξ ∣ + 1 ρ ( m ) ) 2 ( t − t 1 ) ) N j , i , the same times ( ε ˉ m ( ∣ ξ ∣ + 1 ρ ( m ) ) + e − π 2 ( t − t 1 ) / 32 ) , C_{j,i}\left(1+\bar{\varepsilon}_{m}^{2}\left(|\xi|+\frac{1}{\rho(m)}\right)^{2}(t-t_{1})\right)^{N_{j,i}},
\qquad
\text{the same times }\left(\bar{\varepsilon}_{m}\left(|\xi|+\frac{1}{\rho(m)}\right)+e^{-\pi^{2}(t-t_{1})/32}\right), C j , i ( 1 + ε ˉ m 2 ( ∣ ξ ∣ + ρ ( m ) 1 ) 2 ( t − t 1 ) ) N j , i , the same times ( ε ˉ m ( ∣ ξ ∣ + ρ ( m ) 1 ) + e − π 2 ( t − t 1 ) /32 ) ,
C j , i C_{j,i} C j , i depending only on N j , i N_{j,i} N j , i and the fixed integers. The off-branch seminorm, a supremum over tests that change with t t t , is not integrated in t t t : Steps 2–3 hold with any continuous functions above the lower quantities. We prove the claim by induction on N j , i N_{j,i} N j , i , inserting the majorants of the lower terms, which are the terms on the right of (142) , of test orders lower by at least one (Step 7). In Step 2 the square of the factor times the first majorant and the factor times the second integrate to at most the first majorant, as ∫ 0 ∞ e − π 2 t ′ / 32 d t ′ ≤ 4 \int_{0}^{\infty}e^{-\pi^{2}t'/32}\,dt'\leq4 ∫ 0 ∞ e − π 2 t ′ /32 d t ′ ≤ 4 and the factor is at most 1. In Step 3 the integrand is the same power times twice the factor plus e − π 2 ( t ′ − t 1 ) / 32 e^{-\pi^{2}(t'-t_{1})/32} e − π 2 ( t ′ − t 1 ) /32 , which against e − π 2 ( t − t ′ ) / 16 e^{-\pi^{2}(t-t')/16} e − π 2 ( t − t ′ ) /16 gives at most the second majorant. Over the segment t − t 1 ≤ κ m h m / ε m 2 t-t_{1}\leq\kappa_{m}h_{m}/\varepsilon_{m}^{2} t − t 1 ≤ κ m h m / ε m 2 , and t − t 1 t-t_{1} t − t 1 times the square of the factor is at most 2 ( 1 + ρ ( m ) ∣ ξ ∣ ) 2 2(1+\rho(m)|\xi|)^{2} 2 ( 1 + ρ ( m ) ∣ ξ ∣ ) 2 by (147) .
Step 6: Output. The function of ξ \xi ξ in (149) is N 1 N_{1} N 1 times differentiable, as U [ N ] ( t 2 ) U^{[N]}(t_{2}) U [ N ] ( t 2 ) has N 1 + 2 N_{1}+2 N 1 + 2 moments and ∫ t 1 t 2 Z ξ \int_{t_{1}}^{t_{2}}Z_{\xi} ∫ t 1 t 2 Z ξ is analytic on the support of H m ′ H_{m}^{\prime} H m ′ (Proposition 12.4). At t 2 t_{2} t 2 the branch test is 1, so, by (143) and Step 5, for ∣ i ∣ ≤ N 1 |i| \le N_1 ∣ i ∣ ≤ N 1 ,
ρ ( m ) − ∣ i ∣ ∣ ∇ ξ i U [ N ] ( t 2 ) ( ξ ) ∣ ω = ε m 2 ∣ B U [ N , i ] ( t 2 , 1 ) ^ ∣ ≤ C ε m 2 ∣ exp ( ∫ t 1 t 2 Z ξ ) ∣ ( 1 + ρ ( m ) ∣ ξ ∣ ) C ∥ g ∥ ρ ( m ) , ω , N ′ ; ξ . \rho(m)^{-|i|}\left|\nabla_{\xi}^{i}U^{[N]}(t_2)(\xi)\right|_{\omega}=\varepsilon_m^2\left|\widehat{\mathcal{B}U^{[N,i]}(t_2,1)}\right|\le C\varepsilon_m^2\left|\exp\left(\int_{t_1}^{t_2}Z_{\xi}\right)\right|(1+\rho(m)|\xi|)^{C}\|g\|_{\rho(m),\omega,N';\xi}. ρ ( m ) − ∣ i ∣ ∇ ξ i U [ N ] ( t 2 ) ( ξ ) ω = ε m 2 B U [ N , i ] ( t 2 , 1 ) ≤ C ε m 2 exp ( ∫ t 1 t 2 Z ξ ) ( 1 + ρ ( m ) ∣ ξ ∣ ) C ∥ g ∥ ρ ( m ) , ω , N ′ ; ξ .
By Leibniz’s rule, the derivatives in (149) are sums of products of these, of derivatives of H m 0 H_m^0 H m 0 , bounded at the scale ρ ( m ) \rho(m) ρ ( m ) since ε ˉ m ≤ ρ ( m ) \bar{\varepsilon}_m\le\rho(m) ε ˉ m ≤ ρ ( m ) , and of derivatives of exp ( − ∫ t 1 t 2 Z ξ ) \exp\left(-\int_{t_1}^{t_2}Z_{\xi}\right) exp ( − ∫ t 1 t 2 Z ξ ) , normalized as in (149) , at most C ( 1 + ρ ( m ) ∣ ξ ∣ ) N 1 ∣ exp ( − ∫ t 1 t 2 Z ξ ) ∣ C(1+\rho(m)|\xi|)^{N_1}\left|\exp\left(-\int_{t_1}^{t_2}Z_{\xi}\right)\right| C ( 1 + ρ ( m ) ∣ ξ ∣ ) N 1 exp ( − ∫ t 1 t 2 Z ξ ) by Proposition 12.5(c); the factors ∣ exp ( ± ∫ t 1 t 2 Z ξ ) ∣ \left|\exp\left(\pm\int_{t_1}^{t_2}Z_{\xi}\right)\right| exp ( ± ∫ t 1 t 2 Z ξ ) cancel, and N 2 N_2 N 2 is the power above plus N 1 N_1 N 1 .
Step 7: Test orders. The moment term ( j , i ) (j,i) ( j , i ) gets the test order N j , i : = max { d 1 , … , d N } + 3 + d 1 + ⋯ + d j N_{j,i}:=\max\{d_1,\ldots,d_N\}+3+d_1+\cdots+d_j N j , i := max { d 1 , … , d N } + 3 + d 1 + ⋯ + d j plus its degree, and the exponent of the weight N j , i N 0 N_{j,i}N_0 N j , i N 0 , with N 0 N_0 N 0 the largest of the integers of Lemma 12.9(b) for these orders (the statement of (b) holds with any larger integer); Lemmas 12.9 and 12.10 are applied with N a c t N_{\mathrm{act}} N act the largest of these exponents. Along (142) the test order drops by at least the order d d d of the connecting terms: by at least d j d_j d j from ( j , i ) (j,i) ( j , i ) to the terms ( j − 1 , i ′ ) (j-1,i') ( j − 1 , i ′ ) , as ∣ i ′ ∣ |i'| ∣ i ′ ∣ is at most ∣ i ∣ |i| ∣ i ∣ plus the Taylor degree of I j \mathcal{I}_j I j , and by at least 1 1 1 to the commutator terms. So Lemma 12.10 applies to tests of order N j , i N_{j,i} N j , i with J = N j , i ′ J=N_{j,i'} J = N j , i ′ , the test order of the lower term, and the exponents drop by at least N 0 N_0 N 0 , as Step 3 requires. Lemma 12.10 differentiates the coefficients at most J + d − 1 ≤ N j , i − 1 ≤ max { d 1 , … , d N } + 2 + d 1 + ⋯ + d N + N ′ J+d-1\le N_{j,i}-1\le\max\{d_1,\ldots,d_N\}+2+d_1+\cdots+d_N+N' J + d − 1 ≤ N j , i − 1 ≤ max { d 1 , … , d N } + 2 + d 1 + ⋯ + d N + N ′ times, within the derivative counts of Subsection 12.6; the constants C j , i C_{j,i} C j , i are fixed in the order of the test orders.
Over a complete block the scalar factor ∣ exp ( ∫ t 1 t 2 Z ξ ) ∣ \left|\exp\left(\int_{t_1}^{t_2}Z_{\xi}\right)\right| exp ( ∫ t 1 t 2 Z ξ ) is at most the Gaussian of Proposition 12.5(b), and Lemma 12.11 gives (ii): the model Π b l k g \Pi_{\mathrm{blk}}g Π blk g is smooth at the scale ρ ( m ) \rho(m) ρ ( m ) and localized near y 0 y_0 y 0 by its moments. Item (iii) bounds the rest, which the block damps at the cell rate, with losses in powers of ( 1 + ε ˉ m / ε m ) (1+\bar{\varepsilon}_m/\varepsilon_m) ( 1 + ε ˉ m / ε m ) and of the number of cell times, and with derivatives taken forward over the last cell time; the damping, smaller than any power of ε m − 1 \varepsilon_m^{-1} ε m − 1 , absorbs these losses.
Lemma 12.12 (Duhamel terms over a renewal block). Let N 1 ≥ 0 N_1\ge0 N 1 ≥ 0 be an integer, and let U [ 0 ] , … , U [ N ] U^{[0]},\ldots,U^{[N]} U [ 0 ] , … , U [ N ] be a Duhamel term over the block [ t 1 , t 2 ] = [ ( l − 1 2 ) h m , ( l + 1 2 ) h m ] [t_1,t_2]=[(l-\frac{1}{2})h_m,(l+\frac{1}{2})h_m] [ t 1 , t 2 ] = [( l − 2 1 ) h m , ( l + 2 1 ) h m ] , centred at y 0 y_0 y 0 , with N ′ N' N ′ and the class D N ′ + 2 \mathcal{D}^{N'+2} D N ′ + 2 of its data as for the moment terms. There exist C , c > 0 C,c>0 C , c > 0 , an integer N 2 N_2 N 2 and a linear map Π b l k \Pi_{\mathrm{blk}} Π blk , independent of the datum g g g , such that:
(i) for every g ∈ D N ′ + 2 g\in\mathcal{D}^{N'+2} g ∈ D N ′ + 2 , Π b l k g \Pi_{\mathrm{blk}}g Π blk g is smooth, and Π b l k g ^ ( ξ ) = 0 \widehat{\Pi_{\mathrm{blk}}g}(\xi)=0 Π blk g ( ξ ) = 0 for every ξ \xi ξ with ε ˉ m ∣ ξ ∣ > 2 c l o \bar{\varepsilon}_m|\xi|>2c_{\mathrm{lo}} ε ˉ m ∣ ξ ∣ > 2 c lo ;
(ii) for every g ∈ D N ′ + 2 g\in\mathcal{D}^{N'+2} g ∈ D N ′ + 2 and every real ξ \xi ξ ,
∣ Π b l k g ^ ∣ ρ ( m ) , ω , N 1 ; ξ ≤ C ( 1 + ρ ( m ) ∣ ξ ∣ ) N 2 e − c ρ ( m ) 2 ∣ ξ ∣ 2 ε m 2 ∥ g ∥ ρ ( m ) , ω , N ′ ; ξ ; \left|\widehat{\Pi_{\mathrm{blk}}g}\right|_{\rho(m),\omega,N_1;\xi}\le C(1+\rho(m)|\xi|)^{N_2}e^{-c\rho(m)^2|\xi|^2}\varepsilon_m^2\|g\|_{\rho(m),\omega,N';\xi}; Π blk g ρ ( m ) , ω , N 1 ; ξ ≤ C ( 1 + ρ ( m ) ∣ ξ ∣ ) N 2 e − c ρ ( m ) 2 ∣ ξ ∣ 2 ε m 2 ∥ g ∥ ρ ( m ) , ω , N ′ ; ξ ; (iii) for every g ∈ D N ′ + 2 ∩ H N 1 + d 1 + ⋯ + d N − 2 N ( R 2 ; V ) g\in\mathcal{D}^{N'+2}\cap H^{N_1+d_1+\cdots+d_N-2N}(\mathbb{R}^2;V) g ∈ D N ′ + 2 ∩ H N 1 + d 1 + ⋯ + d N − 2 N ( R 2 ; V ) ,
∥ U [ N ] ( t 2 ) − Π b l k g ∥ H ρ ( m ) , ω , N 1 N 1 ( R 2 ) ≤ C ( 1 + κ m h m ε m 2 ) N 2 ( 1 + ε ˉ m ε m ) N 2 exp ( − c κ m τ m ε m 2 ) ∥ g ∥ L ω , N ′ 2 ( R 2 ) ≤ C ε m − 1 Q + 3 ∥ g ∥ L ω , N ′ 2 ( R 2 ) . \left\|U^{[N]}(t_2)-\Pi_{\mathrm{blk}}g\right\|_{H^{N_1}_{\rho(m),\omega,N_1}(\mathbb{R}^2)}
\le C\left(1+\frac{\kappa_m h_m}{\varepsilon_m^2}\right)^{N_2}\left(1+\frac{\bar{\varepsilon}_m}{\varepsilon_m}\right)^{N_2}\exp\left(-\frac{c\kappa_m\tau_m}{\varepsilon_m^2}\right)\|g\|_{L^2_{\omega,N'}(\mathbb{R}^2)}
\le C\varepsilon_{m-1}^{Q+3}\|g\|_{L^2_{\omega,N'}(\mathbb{R}^2)}. U [ N ] ( t 2 ) − Π blk g H ρ ( m ) , ω , N 1 N 1 ( R 2 ) ≤ C ( 1 + ε m 2 κ m h m ) N 2 ( 1 + ε m ε ˉ m ) N 2 exp ( − ε m 2 c κ m τ m ) ∥ g ∥ L ω , N ′ 2 ( R 2 ) ≤ C ε m − 1 Q + 3 ∥ g ∥ L ω , N ′ 2 ( R 2 ) .
Proof. The map. For g ∈ D N ′ + 2 g\in\mathcal{D}^{N'+2} g ∈ D N ′ + 2 let Π b l k g ^ ( ξ ) \widehat{\Pi_{\mathrm{blk}}g}(\xi) Π blk g ( ξ ) be H m 0 ( ξ ) N ′ + 1 H_m^0(\xi)^{N'+1} H m 0 ( ξ ) N ′ + 1 times U [ N ] ( t 2 ) ^ ( ξ ) \widehat{U^{[N]}(t_2)}(\xi) U [ N ] ( t 2 ) ( ξ ) . It is linear, with the Fourier support of (i), and smooth since its Fourier transform has compact support.
(ii). Lemma 12.11 over the block, times H m 0 ( ξ ) N ′ H_m^0(\xi)^{N'} H m 0 ( ξ ) N ′ , bounds the derivatives of exp ( − ∫ t 1 t 2 Z ξ ) Π b l k g ^ ( ξ ) \exp\left(-\int_{t_1}^{t_2}Z_{\xi}\right)\widehat{\Pi_{\mathrm{blk}}g}(\xi) exp ( − ∫ t 1 t 2 Z ξ ) Π blk g ( ξ ) for compactly supported g ∈ H d 1 + ⋯ + d N − 2 N ( R 2 ; V ) g\in H^{d_1+\cdots+d_N-2N}(\mathbb{R}^2;V) g ∈ H d 1 + ⋯ + d N − 2 N ( R 2 ; V ) ; those of exp ( ∫ t 1 t 2 Z ξ ) \exp\left(\int_{t_1}^{t_2}Z_{\xi}\right) exp ( ∫ t 1 t 2 Z ξ ) of order j j j , times ρ ( m ) − j \rho(m)^{-j} ρ ( m ) − j , are at most C ( j ) ( 1 + ρ ( m ) ∣ ξ ∣ ) j e − c d i s ρ ( m ) 2 ∣ ξ ∣ 2 C(j)(1+\rho(m)|\xi|)^j e^{-c_{\mathrm{dis}}\rho(m)^2|\xi|^2} C ( j ) ( 1 + ρ ( m ) ∣ ξ ∣ ) j e − c dis ρ ( m ) 2 ∣ ξ ∣ 2 by Proposition 12.5(b),(c), and Leibniz’s rule gives the bound with c = c d i s c=c_{\mathrm{dis}} c = c dis , the power of 1 + ρ ( m ) ∣ ξ ∣ 1+\rho(m)|\xi| 1 + ρ ( m ) ∣ ξ ∣ rising by N 1 N_1 N 1 . It extends to g ∈ D N ′ + 2 g\in\mathcal{D}_{N^{\prime}+2} g ∈ D N ′ + 2 by the limit g ( r ) : = χ ( ( y − y 0 ) / r ) g → g g^{(r)}:=\chi((y-y_0)/r)g\to g g ( r ) := χ (( y − y 0 ) / r ) g → g in D N ′ + 2 \mathcal{D}_{N^{\prime}+2} D N ′ + 2 as r → ∞ r\to\infty r → ∞ , χ \chi χ a smooth cutoff equal to 1 near 0: by the moment-weighted energy, U g ( r ) [ N ] ( t 2 ) → U g [ N ] ( t 2 ) U_{g^{(r)}}^{[N]}(t_2)\to U_g^{[N]}(t_2) U g ( r ) [ N ] ( t 2 ) → U g [ N ] ( t 2 ) in L ω , N 1 + 2 2 ( R 2 ) L^2_{\omega,N_1+2}(\mathbb{R}^2) L ω , N 1 + 2 2 ( R 2 ) , so the Fourier transforms converge in C N 1 ( R 2 ) C^{N_1}(\mathbb{R}^2) C N 1 ( R 2 ) ; and the moment data of g ( r ) g^{(r)} g ( r ) converge to those of g g g at every ξ \xi ξ (paragraph of (130) , with two extra moments).
( i i i ) (iii) ( iii ) , Step 1: Against tests. Let ξ \xi ξ be in the zone and N j , i N_{j,i} N j , i the test orders of the proof of Lemma 12.11. For every moment term, t ∈ [ t 1 , t 2 ] t\in[t_1,t_2] t ∈ [ t 1 , t 2 ] and U † ∈ H N j , i ( T 2 ) U^\dagger\in H^{N_{j,i}}(\mathbb{T}^2) U † ∈ H N j , i ( T 2 ) , no orthogonality required,
∣ ⟨ B U [ j , i ] ( t ) , U † ⟩ L ω 2 ( T 2 ) ∣ ≤ C ( 1 + κ m ( t − t 1 ) ε m 2 ) C ( 1 + ε ˉ m ε m ) C 2 − # [ t 1 , t ] ∥ U † ∥ H N j , i ( T 2 ) ∥ g ∥ ρ ( m ) , ω , N ′ ; ξ . (150) \left|\left\langle\mathcal{B}U^{[j,i]}(t),U^\dagger\right\rangle_{L^2_\omega(\mathbb{T}^2)}\right|\leq C\left(1+\frac{\kappa_m(t-t_1)}{\varepsilon_m^2}\right)^C\left(1+\frac{\bar{\varepsilon}_m}{\varepsilon_m}\right)^C2^{-\#[t_1,t]}\|U^\dagger\|_{H^{N_{j,i}}(\mathbb{T}^2)}\|g\|_{\rho(m),\omega,N^{\prime};\xi}. \tag*{(150)} ⟨ B U [ j , i ] ( t ) , U † ⟩ L ω 2 ( T 2 ) ≤ C ( 1 + ε m 2 κ m ( t − t 1 ) ) C ( 1 + ε m ε ˉ m ) C 2 − # [ t 1 , t ] ∥ U † ∥ H N j , i ( T 2 ) ∥ g ∥ ρ ( m ) , ω , N ′ ; ξ . (150)
Indeed, by (141) the pairing is that at t 1 t_1 t 1 with the test from U † U^\dagger U † , plus the integral of the pairings of the lower moment terms with I ∗ \mathcal{I}^* I ∗ of the test, which Lemmas 12.8(a) and 12.10(a) bound by C ( 1 + ε ˉ m / ε m ) C 2 − # [ t ′ , t ] C(1+\bar{\varepsilon}_m/\varepsilon_m)^C2^{-\#[t^{\prime},t]} C ( 1 + ε ˉ m / ε m ) C 2 − # [ t ′ , t ] in H N j ′ , i ′ H^{N_{j^{\prime},i^{\prime}}} H N j ′ , i ′ . By induction, with 2 − # [ t 1 , t ′ ] 2 − # [ t ′ , t ] ≤ 2 ⋅ 2 − # [ t 1 , t ] 2^{-\#[t_1,t^{\prime}]}2^{-\#[t^{\prime},t]}\leq2\cdot2^{-\#[t_1,t]} 2 − # [ t 1 , t ′ ] 2 − # [ t ′ , t ] ≤ 2 ⋅ 2 − # [ t 1 , t ] , each pairing is a nested integral over a simplex of insertion times, of volume polynomial in the cell duration, with every derivative on a smooth test: there is no short-time singularity.
Step 2: The full field. On N + N ′ + 1 N+N^{\prime}+1 N + N ′ + 1 equal parts of the last cell time of the block, Lemma 12.8(b) smooths the moment terms one after the other, from the negative order of (150) for the part without forcing and with the forcing for the Duhamel part, to the forward order of ( j , i ) (j,i) ( j , i ) : N 1 + d j + 1 + ⋯ + d N N_1+d_{j+1}+\cdots+d_N N 1 + d j + 1 + ⋯ + d N plus N 1 N_1 N 1 plus the Taylor degrees of I j + 1 , … , I N \mathcal{I}_{j+1},\ldots,\mathcal{I}_N I j + 1 , … , I N minus ∣ i ∣ |i| ∣ i ∣ . The rank of ( j , i ) (j,i) ( j , i ) , j j j plus its degree, lies in { 0 , … , N + N ′ } \{0,\ldots,N+N^{\prime}\} { 0 , … , N + N ′ } and drops by at least one along every arrow of (142) , as the degree does not increase from ( j , i ) (j,i) ( j , i ) to the terms ( j − 1 , i ′ ) (j-1,i^{\prime}) ( j − 1 , i ′ ) and drops along the commutator terms; so the terms of rank r r r are smoothed on the part r + 1 r+1 r + 1 , their forcings being smooth from its start. The forward order drops along (142) by at least the order of the connecting terms, so the forcing lies in the next forward order by Lemma 12.10(c), which differentiates the coefficients at most N 1 + N ′ + d 1 + ⋯ + d N + max { d 1 , … , d N } − 1 N_1+N^{\prime}+d_1+\cdots+d_N+\max\{d_1,\ldots,d_N\}-1 N 1 + N ′ + d 1 + ⋯ + d N + max { d 1 , … , d N } − 1 times here, within the derivative counts. So every ( N , i ) (N,i) ( N , i ) is at most C ( 1 + κ m h m / ε m 2 ) C ( 1 + ε ˉ m / ε m ) C ∥ g ∥ ρ ( m ) , ω , N ′ ; ξ C(1+\kappa_mh_m/\varepsilon_m^2)^C(1+\bar{\varepsilon}_m/\varepsilon_m)^C\|g\|_{\rho(m),\omega,N^{\prime};\xi} C ( 1 + κ m h m / ε m 2 ) C ( 1 + ε ˉ m / ε m ) C ∥ g ∥ ρ ( m ) , ω , N ′ ; ξ in H N 1 ( T 2 ) H^{N_1}(\mathbb{T}^2) H N 1 ( T 2 ) at t 2 t_2 t 2 .
Step 3: The discarded frequencies. The Fourier transform of U [ N ] ( t 2 ) − Π b l k g U^{[N]}(t_2)-\Pi_{\mathrm{blk}}g U [ N ] ( t 2 ) − Π blk g is ( 1 − H m ∘ ( ξ ) N ′ + 1 ) U [ N ] ( t 2 ) ( ξ ) (1-\mathrm{H}_m^{\circ}(\xi)^{N^{\prime}+1})U^{[N]}(t_2)(\xi) ( 1 − H m ∘ ( ξ ) N ′ + 1 ) U [ N ] ( t 2 ) ( ξ ) , supported where ε ˉ m ∣ ξ ∣ ≥ c l o \bar{\varepsilon}_m|\xi|\geq c_{\mathrm{lo}} ε ˉ m ∣ ξ ∣ ≥ c lo ; we split it by the smooth step H ( 2 − 4 ε m ∣ ξ ∣ / π ) \mathrm{H}(2-4\varepsilon_m|\xi|/\pi) H ( 2 − 4 ε m ∣ ξ ∣/ π ) , which is 1 for ε m ∣ ξ ∣ ≤ π / 4 \varepsilon_m|\xi|\leq\pi/4 ε m ∣ ξ ∣ ≤ π /4 and 0 for ε m ∣ ξ ∣ ≥ π / 2 \varepsilon_m|\xi|\geq\pi/2 ε m ∣ ξ ∣ ≥ π /2 . At moderate frequencies, ε m ∣ ξ ∣ ≤ π / 2 \varepsilon_m|\xi|\leq\pi/2 ε m ∣ ξ ∣ ≤ π /2 , (143) makes ρ ( m ) − i ∣ ∇ ξ i U [ N ] ( t 2 ) ( ξ ) ∣ ω \rho(m)^{-i|\nabla_\xi^iU^{[N]}(t_2)(\xi)|_\omega} ρ ( m ) − i ∣ ∇ ξ i U [ N ] ( t 2 ) ( ξ ) ∣ ω equal to ε m 2 ∣ ⟨ B U [ N , i ] ( t 2 ) , 1 ⟩ ∣ \varepsilon_m^2|\langle\mathcal{B}U^{[N,i]}(t_2),1\rangle| ε m 2 ∣ ⟨ B U [ N , i ] ( t 2 ) , 1 ⟩ ∣ , which (150) , with the test 1 and the κ m h m / ε m 2 \kappa_mh_m/\varepsilon_m^2 κ m h m / ε m 2 halving intervals of the block, bounds by
C ε m 2 ( 1 + κ m h m ε m 2 ) C ( 1 + ε ˉ m ε m ) C e − c κ m h m / ε m 2 ∥ g ∥ ρ ( m ) , ω , N ′ ; ξ . C\varepsilon_m^2\left(1+\frac{\kappa_mh_m}{\varepsilon_m^2}\right)^C\left(1+\frac{\bar{\varepsilon}_m}{\varepsilon_m}\right)^Ce^{-c\kappa_mh_m/\varepsilon_m^2}\|g\|_{\rho(m),\omega,N^{\prime};\xi}. C ε m 2 ( 1 + ε m 2 κ m h m ) C ( 1 + ε m ε ˉ m ) C e − c κ m h m / ε m 2 ∥ g ∥ ρ ( m ) , ω , N ′ ; ξ .
As ρ m o l ( m ) ∣ ξ ∣ ≤ π / 2 \rho_{\mathrm{mol}}(m)|\xi|\leq\pi/2 ρ mol ( m ) ∣ ξ ∣ ≤ π /2 on this band, Plancherel’s identity, Leibniz’s rule on the two cutoffs and Lemma 12.2 give the bound of (iii) for this part. At high frequencies, ε m ∣ ξ ∣ ≥ π / 4 \varepsilon_m|\xi|\geq\pi/4 ε m ∣ ξ ∣ ≥ π /4 , we use the modes of Bloch sums: for h h h with ∥ h ∥ L ω , 2 2 ( R 2 ) < ∞ \|h\|_{L^2_{\omega,2}(\mathbb{R}^2)}<\infty ∥ h ∥ L ω , 2 2 ( R 2 ) < ∞ , ξ ′ \xi^{\prime} ξ ′ in the zone and n ∈ Z 2 \mathbf{n}\in\mathbb{Z}^2 n ∈ Z 2 , by Poisson’s summation formula,
∫ T 2 B h ( ξ ′ , z ) e − 2 π i n ⋅ z d z = e − 2 π i n ⋅ y 0 / ε m ε m 2 h ^ ( ξ ′ + 2 π n ε m ) , (151) \int_{\mathbb{T}^2}\mathcal{B}h(\xi^{\prime},z)e^{-2\pi i\mathbf{n}\cdot z}\,dz=\frac{e^{-2\pi i\mathbf{n}\cdot y_0/\varepsilon_m}}{\varepsilon_m^2}\widehat{h}\left(\xi^{\prime}+\frac{2\pi\mathbf{n}}{\varepsilon_m}\right), \tag*{(151)} ∫ T 2 B h ( ξ ′ , z ) e − 2 π i n ⋅ z d z = ε m 2 e − 2 π i n ⋅ y 0 / ε m h ( ξ ′ + ε m 2 π n ) , (151)
and the translates of the zone by 2 π Z 2 / ε m 2\pi\mathbb{Z}^2/\varepsilon_m 2 π Z 2 / ε m tile R 2 \mathbb{R}^2 R 2 . So, with ξ ′ \xi^{\prime} ξ ′ the frequency of the zone with ε m ( ξ − ξ ′ ) ∈ 2 π Z 2 \varepsilon_m(\xi-\xi^{\prime})\in2\pi\mathbb{Z}^2 ε m ( ξ − ξ ′ ) ∈ 2 π Z 2 , U [ N , i ] ( t 2 ) ( ξ ) U^{[N,i]}(t_2)(\xi) U [ N , i ] ( t 2 ) ( ξ ) is ε m 2 \varepsilon_m^2 ε m 2 times a unimodular number times the mode e i ε m ( ξ − ξ ′ ) ⋅ z e^{i\varepsilon_m(\xi-\xi^{\prime})\cdot z} e i ε m ( ξ − ξ ′ ) ⋅ z of the Bloch sum at ξ ′ \xi^{\prime} ξ ′ . Let t t t be the start of the quiet stretch, of length at least τ m ′ − τ m / 24 \tau_m^{\prime}-\tau_m/24 τ m ′ − τ m /24 , that ends the block. By (150) with this test,
∣ ⟨ B U [ j , i ] ( t ) ( ξ ′ , ⋅ ) , e i ε m ( ξ − ξ ′ ) ⋅ z ⟩ ∣ ≤ C ( 1 + κ m h m ε m 2 ) C ( 1 + ε ˉ m ε m ) C ( 1 + ε m ∣ ξ ∣ ) N j , i ∥ g ∥ ρ ( m ) , ω , N ′ ; ξ . \left|\left\langle\mathcal{B}U^{[j,i]}(t)(\xi^{\prime},\cdot),e^{i\varepsilon_m(\xi-\xi^{\prime})\cdot z}\right\rangle\right|\leq C\left(1+\frac{\kappa_mh_m}{\varepsilon_m^2}\right)^C\left(1+\frac{\bar{\varepsilon}_m}{\varepsilon_m}\right)^C(1+\varepsilon_m|\xi|)^{N_{j,i}}\|g\|_{\rho(m),\omega,N^{\prime};\xi}. ⟨ B U [ j , i ] ( t ) ( ξ ′ , ⋅ ) , e i ε m ( ξ − ξ ′ ) ⋅ z ⟩ ≤ C ( 1 + ε m 2 κ m h m ) C ( 1 + ε m ε ˉ m ) C ( 1 + ε m ∣ ξ ∣ ) N j , i ∥ g ∥ ρ ( m ) , ω , N ′ ; ξ .
Across the stretch the velocity vanishes and the insertions do not depend on z z z , by clause (i) of the definition of insertions, so each mode evolves separately and is damped at the rate 1 2 ε m 2 ∣ ξ ∣ 2 ≥ π 2 32 \frac{1}{2}\varepsilon_m^2|\xi|^2 \ge\frac{\pi^2}{32} 2 1 ε m 2 ∣ ξ ∣ 2 ≥ 32 π 2 (the proof of Lemma 12.8(c)). The triangular forcing costs a polynomial in the length of the stretch, and, summed over the modes, the Gaussian damping in ε m ∣ ξ ∣ \varepsilon_m|\xi| ε m ∣ ξ ∣ absorbs the growth ( 1 + ε m ∣ ξ ∣ ) N j , i (1+\varepsilon_m|\xi|)^{N_{j,i}} ( 1 + ε m ∣ ξ ∣ ) N j , i and leaves the factor e − c κ m τ m ′ / ε m 2 e^{-c\kappa_m\tau'_m/\varepsilon_m^2} e − c κ m τ m ′ / ε m 2 in H 1 N ( T 2 ) H^N_1(\mathbb{T}^2) H 1 N ( T 2 ) at t 2 t_2 t 2 . We return to the plane: by (151) , ∇ + i ε m ξ ′ \nabla+i\varepsilon_m\xi' ∇ + i ε m ξ ′ acts on the mode n \mathbf{n} n of B h ( ξ ′ , ⋅ ) Bh(\xi',\cdot) B h ( ξ ′ , ⋅ ) as i ε m ξ i\varepsilon_m\xi i ε m ξ , with ξ = ξ ′ + 2 π n / ε m \xi=\xi'+2\pi\mathbf{n}/\varepsilon_m ξ = ξ ′ + 2 π n / ε m , so, by Parseval’s identity,
∫ z o n e ∥ ( ∇ + i ε m ξ ′ ) j B h ( ξ ′ , ⋅ ) ∥ L ω 2 ( T 2 ) 2 d ξ ′ = ( 2 π ) 2 ε m 2 j − 4 ∥ ∇ j h ∥ L ω 2 ( R 2 ) 2 \int_{\mathrm{zone}}\left\|(\nabla+i\varepsilon_m\xi')^jBh(\xi',\cdot)\right\|_{L^2_\omega(\mathbb{T}^2)}^2\,\mathrm{d}\xi'=(2\pi)^2\varepsilon_m^{2j-4}\|\nabla^j h\|_{L^2_\omega(\mathbb{R}^2)}^2 ∫ zone ( ∇ + i ε m ξ ′ ) j B h ( ξ ′ , ⋅ ) L ω 2 ( T 2 ) 2 d ξ ′ = ( 2 π ) 2 ε m 2 j − 4 ∥ ∇ j h ∥ L ω 2 ( R 2 ) 2
whenever the left side is finite, and ∇ j h \nabla^j h ∇ j h then lies in L ω 2 ( R 2 ) L^2_\omega(\mathbb{R}^2) L ω 2 ( R 2 ) . We apply this with h = ( ( y − y 0 ) / ρ ( m ) ) a U [ N ] ( t 2 ) h=((y-y_0)/\rho(m))^aU^{[N]}(t_2) h = (( y − y 0 ) / ρ ( m ) ) a U [ N ] ( t 2 ) , ∣ a ∣ ≤ N 1 |a|\le N_1 ∣ a ∣ ≤ N 1 , and commute the weight with ρ m o l ( m ) ∇ \rho_{\mathrm{mol}}(m)\nabla ρ mol ( m ) ∇ at the cost ρ m o l ( m ) / ρ ( m ) ≤ 1 \rho_{\mathrm{mol}}(m)/\rho(m)\le1 ρ mol ( m ) / ρ ( m ) ≤ 1 , the derivatives at the scale ρ m o l ( m ) ≤ ε m \rho_{\mathrm{mol}}(m)\le\varepsilon_m ρ mol ( m ) ≤ ε m costing a constant. As h m ≥ τ ‾ m ≥ τ m h_m\ge\overline{\tau}_m\ge\tau_m h m ≥ τ m ≥ τ m , this is the first bound of (iii) for g ∈ D N ′ + 2 g\in D^{N'+2} g ∈ D N ′ + 2 .
Step 4: The power of ε m − 1 \varepsilon_{m-1} ε m − 1 . By Proposition 2.1(c) and (7) , κ m τ m / ε m 2 ≥ c ε m − 1 − 4 δ \kappa_m\tau_m/\varepsilon_m^2\ge c\varepsilon_{m-1}^{-4\delta} κ m τ m / ε m 2 ≥ c ε m − 1 − 4 δ , ε ‾ m / ε m ≤ C ε m − 1 − q γ \overline{\varepsilon}_m/\varepsilon_m\le C\varepsilon_{m-1}^{-q\gamma} ε m / ε m ≤ C ε m − 1 − q γ , and κ m h m / ε m 2 ≤ ρ ( m ) 2 / ε m 2 ≤ C ε m − 1 − 2 q \kappa_mh_m/\varepsilon_m^2\le\rho(m)^2/\varepsilon_m^2\le C\varepsilon_{m-1}^{-2q} κ m h m / ε m 2 ≤ ρ ( m ) 2 / ε m 2 ≤ C ε m − 1 − 2 q as κ m ≤ κ m − 1 \kappa_m\le\kappa_{m-1} κ m ≤ κ m − 1 , ρ ( m ) ≤ 1 \rho(m)\le1 ρ ( m ) ≤ 1 and ε m ≥ 1 2 ε m − 1 q \varepsilon_m\ge\frac{1}{2}\varepsilon_{m-1}^q ε m ≥ 2 1 ε m − 1 q (Lemma 7.1(a)). So the first bound of (iii) is at most C ε m − 1 − C exp ( − c ε m − 1 − 4 δ ) ≤ C ε m − 1 Q + 3 C\varepsilon_{m-1}^{-C}\exp(-c\varepsilon_{m-1}^{-4\delta})\le C\varepsilon_{m-1}^{Q+3} C ε m − 1 − C exp ( − c ε m − 1 − 4 δ ) ≤ C ε m − 1 Q + 3 for m m m large.
The change of Lagrangian coordinates, and the proof of Proposition 12.7
The old partial block, from t 1 t_1 t 1 to the block boundary t 2 t_2 t 2 , may be short and carry pulses of one orientation only (Figure 12.1 ): it need not damp the frequencies away from the branch, and its scalar factor has no Gaussian. At t 2 t_2 t 2 the chart changes, by an area-preserving map that does not keep the support of the Fourier transform. So the old term is carried across t 2 t_2 t 2 at low frequency, where it is bounded only in negative norms, and the next block, the complete future block , smooths it (Figure 12.3 ): its model sees its datum only through the zone part below, and its Gaussian absorbs any power of 1 + ρ ( m ) ∣ ξ ∣ 1+\rho(m)|\xi| 1 + ρ ( m ) ∣ ξ ∣ .
Figure 12.3. The first renewal (Lemmas 12.13–12.15 ). At t 2 t_2 t 2 , in the quiet margin (shaded), U [ N − ] ( t 2 ) U^{[N_-]}(t_2) U [ N − ] ( t 2 ) splits into its low-frequency part, transported to the chart of the future block and cut to its zone part, and the rest. Solid: the model Π r e n g \Pi_{\mathrm{reng}} Π reng ; dashed: its two errors; (a)–(c): the items of Lemma 12.13; Π + \Pi_+ Π + : the model of the future block.
Let t 2 : = ( l + 1 2 ) h m t_2:=(l+\frac{1}{2})h_m t 2 := ( l + 2 1 ) h m and t p r : = t 2 + h m t_{\mathrm{pr}}:=t_2+h_m t pr := t 2 + h m , as in (138) , let y 0 ′ y'_0 y 0 ′ be a label of the chart of the block of l + 1 l+1 l + 1 , and let y 0 : = X m − 1 ( l h m , y 0 ′ , ( l + 1 ) h m ) y_0:=X_{m-1}(lh_m,y'_0,(l+1)h_m) y 0 := X m − 1 ( l h m , y 0 ′ , ( l + 1 ) h m ) be the label of the same point in the chart of the block of l l l (Lemma 7.7(b)). A Duhamel term across the renewal of lengths N − , N + N_-,N_+ N − , N + , from a datum g g g at a time t 1 ∈ [ t 2 − h m , t 2 ] t_1\in[t_2-h_m,t_2] t 1 ∈ [ t 2 − h m , t 2 ] , is the sequence U [ 0 ] , … , U [ N − + N + ] U^{[0]},\ldots,U^{[N_-+N_+]} U [ 0 ] , … , U [ N − + N + ] in which U [ 0 ] , … , U [ N − ] U^{[0]},\ldots,U^{[N_-]} U [ 0 ] , … , U [ N − ] is a Duhamel term on [ t 1 , t 2 ] [t_1,t_2] [ t 1 , t 2 ] for the block of l l l , centred at y 0 y_0 y 0 , from g g g , with the old insertions I 1 , … , I N − \mathcal{I}_1,\ldots,\mathcal{I}_{N_-} I 1 , … , I N − , of orders d 1 , … , d N − d_1,\ldots,d_{N_-} d 1 , … , d N − ; U [ N − ] U^{[N_-]} U [ N − ] continues on [ t 2 , t p r ] [t_2,t_{\mathrm{pr}}] [ t 2 , t pr ] by the frozen operator of the block of l + 1 l+1 l + 1 , centred at y 0 ′ y'_0 y 0 ′ , from U [ N − ] ( t 2 ) ∘ X m − 1 ( l h m , ⋅ , ( l + 1 ) h m ) U^{[N_-]}(t_2)\circ X_{m-1}(lh_m,\cdot,(l+1)h_m) U [ N − ] ( t 2 ) ∘ X m − 1 ( l h m , ⋅ , ( l + 1 ) h m ) ; and U [ N − + 1 ] , … , U [ N − + N + ] U^{[N_-+1]},\ldots,U^{[N_-+N_+]} U [ N − + 1 ] , … , U [ N − + N + ] start from zero at t 2 t_2 t 2 , forced by N + N_+ N + future insertions I N − + 1 , … , I N − + N + \mathcal{I}_{N_-+1},\ldots,\mathcal{I}_{N_-+N_+} I N − + 1 , … , I N − + N + of the block of l + 1 l+1 l + 1 , of orders d N − + 1 , … , d N − + N + d_{N_-+1},\ldots,d_{N_-+N_+} d N − + 1 , … , d N − + N + . Here and below, U [ N − ] ( t 2 ) U^{[N_-]}(t_2) U [ N − ] ( t 2 ) always denotes the value at t 2 t_2 t 2 in the chart of the block of l l l .
The low-frequency part and the zone part of a field F F F on R 2 \mathbb{R}^2 R 2 are given, with F F F a function of the labels of a chart and the Fourier transform about its centre, by
F l o ^ ( ξ ) : = H m l o ( ξ ) F ^ ( ξ ) , F z o n e ^ ( ξ ) : = H ( 4 − 8 ε m ∣ ξ ∣ π ) F ^ ( ξ ) ; \widehat{F^{\mathrm{lo}}}(\xi):=H_m^{\mathrm{lo}}(\xi)\widehat{F}(\xi),\qquad\widehat{F^{\mathrm{zone}}}(\xi):=H\left(4-\frac{8\varepsilon_m|\xi|}{\pi}\right)\widehat{F}(\xi); F lo ( ξ ) := H m lo ( ξ ) F ( ξ ) , F zone ( ξ ) := H ( 4 − π 8 ε m ∣ ξ ∣ ) F ( ξ ) ;
the second multiplier, the zone multiplier , is 1 for ε m ∣ ξ ∣ ≤ 3 π / 8 \varepsilon_m|\xi|\le3\pi/8 ε m ∣ ξ ∣ ≤ 3 π /8 and 0 for ε m ∣ ξ ∣ ≥ π / 2 \varepsilon_m|\xi|\ge\pi/2 ε m ∣ ξ ∣ ≥ π /2 , so the Fourier transform of F z o n e F^{\mathrm{zone}} F zone is supported in the zone ε m ∣ ξ ∣ ∞ ≤ π \varepsilon_m|\xi|_\infty\le\pi ε m ∣ ξ ∣ ∞ ≤ π . The zone part is taken after the change of chart.
Negative norms. Fix an integer N 1 ≥ 0 N_1\ge0 N 1 ≥ 0 , the moment order of the output, and let N ′ N' N ′ be that of the future block: N 1 N_1 N 1 plus the sum of the Taylor degrees of the future insertions. For an integer N ≥ 0 N\ge0 N ≥ 0 and ρ ∈ { ε m , ρ ( m ) } \rho\in\{\varepsilon_m,\rho(m)\} ρ ∈ { ε m , ρ ( m )} , over Schwartz fields ϑ \vartheta ϑ ,
∥ F ∥ H ρ , ω , N ′ + 2 − N ( R 2 ) : = sup { ∣ ⟨ F , ϑ ⟩ L ω 2 ( R 2 ) ∣ : ∥ ϑ ∥ H ρ , ω , − ( N ′ + 2 ) N ( R 2 ) ≤ 1 } \|F\|_{H^{-N}_{\rho,\omega,N'+2}(\mathbb{R}^2)}:=\sup\left\{\left|\langle F,\vartheta\rangle_{L^2_\omega(\mathbb{R}^2)}\right|:\|\vartheta\|_{H^{N}_{\rho,\omega,-(N'+2)}(\mathbb{R}^2)}\le1\right\} ∥ F ∥ H ρ , ω , N ′ + 2 − N ( R 2 ) := sup { ⟨ F , ϑ ⟩ L ω 2 ( R 2 ) : ∥ ϑ ∥ H ρ , ω , − ( N ′ + 2 ) N ( R 2 ) ≤ 1 }
with ⟨ F , ϑ ⟩ L ω 2 ( R 2 ) : = ∑ v ω v − 1 ∫ R 2 F v ϑ v ‾ d y \langle F,\vartheta\rangle_{L^2_\omega(\mathbb{R}^2)} := \sum_v \omega_v^{-1}\int_{\mathbb{R}^2}F_v\overline{\vartheta_v}\,dy ⟨ F , ϑ ⟩ L ω 2 ( R 2 ) := ∑ v ω v − 1 ∫ R 2 F v ϑ v d y , and with the weight ( 1 + ∣ y − y ∗ ∣ 2 / ρ ( m ) 2 ) − ( N ′ + 2 ) / 2 (1+|y-y_*|^2/\rho(m)^2)^{-(N'+2)/2} ( 1 + ∣ y − y ∗ ∣ 2 / ρ ( m ) 2 ) − ( N ′ + 2 ) /2 in (129) , y ∗ y_* y ∗ the centre of the chart of F F F , about which the moments are taken. A larger N N N or the scale ρ ( m ) \rho(m) ρ ( m ) in place of ε m \varepsilon_m ε m gives a smaller norm. The negative order N n e g : = N 1 + d 1 + ⋯ + d N − + 2 N_{\mathrm{neg}} := N_1+d_1+\cdots+d_{N_-}+2 N neg := N 1 + d 1 + ⋯ + d N − + 2 exceeds the number of derivatives that the old insertions put on a test field.
In Lemma 12.13, (a) bounds the old term with losses, and (b) its low-frequency part without loss but, with no Gaussian to absorb the polynomial factor of Lemma 12.11 , only in a negative norm at the scale ρ ( m ) \rho(m) ρ ( m ) . By (c) the change of chart preserves these norms, and the high frequencies, far above 1 / ρ ( m ) 1/\rho(m) 1/ ρ ( m ) , gain a power of ε ˉ m / ρ ( m ) \bar{\varepsilon}_m/\rho(m) ε ˉ m / ρ ( m ) at that scale, which absorbs the losses of (a).
Lemma 12.13 (The old block, carried across). In the setting above (a Duhamel term across the renewal from g g g at t 1 t_1 t 1 ), there exist C C C , an integer N 3 N_3 N 3 and, for every integer N ≥ 0 N \ge0 N ≥ 0 , a constant C ( N ) C(N) C ( N ) such that:
(a) In a negative norm. For every g ∈ H d 1 + ⋯ + d N − − 2 N − ( R 2 ) V g \in H^{d_1+\cdots+d_{N_-}-2N_-}(\mathbb{R}^2)^V g ∈ H d 1 + ⋯ + d N − − 2 N − ( R 2 ) V with compact support,
∥ U [ N − ] ( t 2 ) ∥ H ε m , ω , N ′ + 2 − N n e g ( R 2 ) ≤ C ( 1 + κ m h m ε m 2 ) N 3 ( 1 + ε ˉ m ε m ) N 3 ∥ g ∥ L ω , N 3 2 ( R 2 ) . \left\|U^{[N_-]}(t_2)\right\|_{H^{-N_{\mathrm{neg}}}_{\varepsilon_m,\omega,N'+2}(\mathbb{R}^2)}
\le C\left(1+\frac{\kappa_m h_m}{\varepsilon_m^2}\right)^{N_3}\left(1+\frac{\bar{\varepsilon}_m}{\varepsilon_m}\right)^{N_3}\|g\|_{L^2_{\omega,N_3}(\mathbb{R}^2)}. U [ N − ] ( t 2 ) H ε m , ω , N ′ + 2 − N neg ( R 2 ) ≤ C ( 1 + ε m 2 κ m h m ) N 3 ( 1 + ε m ε ˉ m ) N 3 ∥ g ∥ L ω , N 3 2 ( R 2 ) . (b) At low frequency, for the same g g g , with no loss in the contrast or the number of cell times:
∥ ( U [ N − ] ( t 2 ) ) l o ∥ H ρ ( m ) , ω , N ′ + 2 − N 3 ( R 2 ) ≤ C ∥ g ∥ L ω , N 3 2 ( R 2 ) . \left\|\left(U^{[N_-]}(t_2)\right)^{\mathrm{lo}}\right\|_{H^{-N_3}_{\rho(m),\omega,N'+2}(\mathbb{R}^2)}
\le C\|g\|_{L^2_{\omega,N_3}(\mathbb{R}^2)}. ( U [ N − ] ( t 2 ) ) lo H ρ ( m ) , ω , N ′ + 2 − N 3 ( R 2 ) ≤ C ∥ g ∥ L ω , N 3 2 ( R 2 ) . (c) Transport and the cutoff. For every F ∈ L 2 ( R 2 ) V F \in L^2(\mathbb{R}^2)^V F ∈ L 2 ( R 2 ) V , every N N N and ρ ∈ { ε m , ρ ( m ) } \rho\in\{\varepsilon_m,\rho(m)\} ρ ∈ { ε m , ρ ( m )} , with F F F in the chart of the block of l l l ,
∥ F ∘ X m − 1 ( l h m , ⋅ , ( l + 1 ) h m ) ∥ H ρ , ω , N ′ + 2 − N ( R 2 ) ≤ C ( N ) ∥ F ∥ H ρ , ω , N ′ + 2 − N ( R 2 ) , \left\|F\circ X_{m-1}(lh_m,\cdot,(l+1)h_m)\right\|_{H^{-N}_{\rho,\omega,N'+2}(\mathbb{R}^2)}
\le C(N)\|F\|_{H^{-N}_{\rho,\omega,N'+2}(\mathbb{R}^2)}, ∥ F ∘ X m − 1 ( l h m , ⋅ , ( l + 1 ) h m ) ∥ H ρ , ω , N ′ + 2 − N ( R 2 ) ≤ C ( N ) ∥ F ∥ H ρ , ω , N ′ + 2 − N ( R 2 ) , ∥ F − F l o ∥ H ρ ( m ) , ω , N ′ + 2 − ( N + N n e g ) ( R 2 ) ≤ C ( N ) ( ε ˉ m ρ ( m ) ) N ∥ F ∥ H ε m , ω , N ′ + 2 − N n e g ( R 2 ) . \left\|F-F^{\mathrm{lo}}\right\|_{H^{-(N+N_{\mathrm{neg}})}_{\rho(m),\omega,N'+2}(\mathbb{R}^2)}
\le C(N)\left(\frac{\bar{\varepsilon}_m}{\rho(m)}\right)^N
\|F\|_{H^{-N_{\mathrm{neg}}}_{\varepsilon_m,\omega,N'+2}(\mathbb{R}^2)}. F − F lo H ρ ( m ) , ω , N ′ + 2 − ( N + N neg ) ( R 2 ) ≤ C ( N ) ( ρ ( m ) ε ˉ m ) N ∥ F ∥ H ε m , ω , N ′ + 2 − N neg ( R 2 ) .
Proof. (a) By (141) , the pairing of U [ N − ] ( t 2 ) U^{[N_-]}(t_2) U [ N − ] ( t 2 ) with a Schwartz field of norm one in H^{N_{\mathrm{neg}}_{\varepsilon_m,\omega,-(N'+2)}(\mathbb{R}^2)} is a sum of nested integrals, over a simplex of insertion times, of the pairing of g g g with this field run backward by the adjoint frozen operator on R 2 \mathbb{R}^{2} R 2 and acted on by the adjoint old insertions, each costing its order in derivatives and its Taylor degree in the weight. The simplex volume, the cost of the adjoint evolution in these weighted norms, and the rate from the transport of the moment weight w w w , by ∇ ⊥ ψ m ⋅ ∇ w = − ∇ ⋅ ( ψ m ∇ ⊥ w ) \nabla^\perp\psi_m \cdot\nabla w=-\nabla\cdot(\psi_m\nabla^\perp w) ∇ ⊥ ψ m ⋅ ∇ w = − ∇ ⋅ ( ψ m ∇ ⊥ w ) and the Cauchy–Schwarz inequality against the dissipation (a constant over t 2 − t 1 ≤ h m t_2-t_1\le h_m t 2 − t 1 ≤ h m , by Grönwall’s inequality), are
1 N − ! ( κ m ( t 2 − t 1 ) ε m 2 ) N − , C ( 1 + ε ˉ m ε m ) C ( 1 + κ m ( t 2 − t 1 ) ε m 2 ) C , C ∣ ψ m ∣ 2 κ m ρ ( m ) 2 ≤ C h m . \frac{1}{N_-!}\left(\frac{\kappa_m(t_2-t_1)}{\varepsilon_m^2}\right)^{N_-},\qquad C\left(1+\frac{\bar{\varepsilon}_m}{\varepsilon_m}\right)^C\left(1+\frac{\kappa_m(t_2-t_1)}{\varepsilon_m^2}\right)^C,\qquad\frac{C|\psi_m|^2}{\kappa_m\rho(m)^2}\le\frac{C}{h_m}. N − ! 1 ( ε m 2 κ m ( t 2 − t 1 ) ) N − , C ( 1 + ε m ε ˉ m ) C ( 1 + ε m 2 κ m ( t 2 − t 1 ) ) C , κ m ρ ( m ) 2 C ∣ ψ m ∣ 2 ≤ h m C .
the last as ∣ ψ m ∣ 2 ≤ C ε m 2 β |\psi_m|^2\le C\varepsilon_m^{2\beta} ∣ ψ m ∣ 2 ≤ C ε m 2 β and ε m 2 β / κ m ≤ 20 κ m − 1 − 1 \varepsilon_m^{2\beta}/\kappa_m\le20\kappa_{m-1}^{-1} ε m 2 β / κ m ≤ 20 κ m − 1 − 1 ; there is no short-time singularity, as at the scale ε m / ( C ( 1 + ε ˉ m / ε m ) ) \varepsilon_m/(C(1+\bar{\varepsilon}_m/\varepsilon_m)) ε m / ( C ( 1 + ε ˉ m / ε m )) the field ∇ ⊥ ψ m \nabla^\perp\psi_m ∇ ⊥ ψ m satisfies (i) of Proposition 8.3, whose energy holds with the weight on R 2 \mathbb{R}^2 R 2 , its commutator estimate being pointwise.
( b ) (b) ( b ) Let N 3 N_3 N 3 be even, at least N 2 + N ′ + 2 N_2+N'+2 N 2 + N ′ + 2 , at least the exponents of 1 + κ m h m / ε m 2 1+\kappa_mh_m/\varepsilon_m^2 1 + κ m h m / ε m 2 and of ( 1 + ε ˉ m / ε m ) (1+\bar{\varepsilon}_m/\varepsilon_m) ( 1 + ε ˉ m / ε m ) in the proof of (a), and at least N ′ + 2 N'+2 N ′ + 2 plus the sum of the Taylor degrees of the old insertions, the weight used in (a), with N 2 N_2 N 2 the integer of Lemma 12.11 with N ′ + 2 N'+2 N ′ + 2 in place of N 1 N_1 N 1 . That lemma on [ t 1 , t 2 ] [t_1,t_2] [ t 1 , t 2 ] , and Propositions 12.5(c) and 12.4(d) for the scalar factor (no Gaussian, but no growth either), give
∣ ( U [ N − ] ^ ( t 2 ) ) ρ ( m ) , ω , N ′ + 2 ; ξ l o ∣ ≤ C ( 1 + ρ ( m ) ∣ ξ ∣ ) N 2 + N ′ + 2 ε m 2 ∥ g ∥ ρ ( m ) , ω , N 3 ; ξ . \left|\left(\widehat{U^{[N_-]}}(t_2)\right)^{\mathrm{lo}}_{\rho(m),\omega,N'+2;\xi}\right|\le C(1+\rho(m)|\xi|)^{N_2+N'+2}\varepsilon_m^2\|g\|_{\rho(m),\omega,N_3;\xi}. ( U [ N − ] ( t 2 ) ) ρ ( m ) , ω , N ′ + 2 ; ξ lo ≤ C ( 1 + ρ ( m ) ∣ ξ ∣ ) N 2 + N ′ + 2 ε m 2 ∥ g ∥ ρ ( m ) , ω , N 3 ; ξ .
As N 3 N_3 N 3 is even, ( 1 − ρ ( m ) 2 Δ ) N 3 / 2 (1-\rho(m)^2\Delta)^{N_3/2} ( 1 − ρ ( m ) 2 Δ ) N 3 /2 moves onto the test fields of the negative norm at the cost C C C , and the factor ( 1 + ρ ( m ) 2 ∣ ξ ∣ 2 ) − N 3 / 2 (1+\rho(m)^2|\xi|^2)^{-N_3/2} ( 1 + ρ ( m ) 2 ∣ ξ ∣ 2 ) − N 3 /2 it leaves absorbs the power of 1 + ρ ( m ) ∣ ξ ∣ 1+\rho(m)|\xi| 1 + ρ ( m ) ∣ ξ ∣ ; by Plancherel, moments becoming ξ \xi ξ -derivatives, and Lemma 12.2 summing over ξ \xi ξ ,
∥ ( U [ N − ] ^ ( t 2 ) ) l o ∥ H ρ ( m ) , ω , − ( N ′ + 2 ) − N 3 ( R 2 ) ≤ C ε m 2 ( ∫ z o n e ∥ g ∥ ρ ( m ) , ω , N 3 ; ξ 2 d ξ ) 1 / 2 ≤ C ∥ g ∥ L ω , N 3 2 ( R 2 ) . \left\|\left(\widehat{U^{[N_-]}}(t_2)\right)^{\mathrm{lo}}\right\|_{H^{-N_3}_{\rho(m),\omega,-(N'+2)}(\mathbb{R}^2)}
\le C\varepsilon_m^2\left(\int_{\mathrm{zone}}\|g\|_{\rho(m),\omega,N_3;\xi}^2\,d\xi\right)^{1/2}
\le C\|g\|_{L^2_{\omega,N_3}(\mathbb{R}^2)}. ( U [ N − ] ( t 2 ) ) lo H ρ ( m ) , ω , − ( N ′ + 2 ) − N 3 ( R 2 ) ≤ C ε m 2 ( ∫ zone ∥ g ∥ ρ ( m ) , ω , N 3 ; ξ 2 d ξ ) 1/2 ≤ C ∥ g ∥ L ω , N 3 2 ( R 2 ) .
( c ) (c) ( c ) The map X m − 1 ( l h m , ⋅ , ( l + 1 ) h m ) X_{m-1}(lh_m,\cdot,(l+1)h_m) X m − 1 ( l h m , ⋅ , ( l + 1 ) h m ) preserves area, maps y 0 ′ y'_0 y 0 ′ to y 0 y_0 y 0 , and has a gradient within C ε m − 1 2 δ C\varepsilon_{m-1}^{2\delta} C ε m − 1 2 δ of the identity, with derivatives at the scale ε m − 1 ≥ ρ ( m ) \varepsilon_{m-1}\ge\rho(m) ε m − 1 ≥ ρ ( m ) ((7.9, Lemma 7.7(b)}); by duality, the first bound follows from the chain rule for the inverse map, as the moment weights 1 + ∣ X m − 1 ( l h m , y , ( l + 1 ) h m ) − y 0 ∣ 2 / ρ ( m ) 2 1+|X_{m-1}(lh_m,y,(l+1)h_m)-y_0|^2/\rho(m)^2 1 + ∣ X m − 1 ( l h m , y , ( l + 1 ) h m ) − y 0 ∣ 2 / ρ ( m ) 2 and 1 + ∣ y − y 0 ′ ∣ 2 / ρ ( m ) 2 1+|y-y'_0|^2/\rho(m)^2 1 + ∣ y − y 0 ′ ∣ 2 / ρ ( m ) 2 are comparable. The Fourier transform of F − F l o F-F^{\mathrm{lo}} F − F lo is supported in ε ˉ m ∣ ξ ∣ ≥ c l o \bar{\varepsilon}_m|\xi|\ge c_{\mathrm{lo}} ε ˉ m ∣ ξ ∣ ≥ c lo , as ρ ( m ) 2 = κ m − 1 h m \rho(m)^2=\kappa_{m-1}h_m ρ ( m ) 2 = κ m − 1 h m ,
ρ ( m ) ε ˉ m = ( κ m h m ε m 2 ) 1 / 2 . (152) \frac{\rho(m)}{\bar{\varepsilon}_m}=\left(\frac{\kappa_mh_m}{\varepsilon_m^2}\right)^{1/2}. \tag*{(152)} ε ˉ m ρ ( m ) = ( ε m 2 κ m h m ) 1/2 . (152)
By duality it suffices that 1 − H m l o ( − i ∇ ) 1-H_m^{\mathrm{lo}}(-i\nabla) 1 − H m lo ( − i ∇ ) maps H ρ ( m ) , ω , − ( N ′ + 2 ) N + N n e g ( R 2 ) H^{N+N_{\mathrm{neg}}}_{\rho(m),\omega,-(N'+2)}(\mathbb{R}^2) H ρ ( m ) , ω , − ( N ′ + 2 ) N + N neg ( R 2 ) to H ρ ( m ) , ω , − ( N ′ + 2 ) N n e g ( R 2 ) H^{N_{\mathrm{neg}}}_{\rho(m),\omega,-(N'+2)}(\mathbb{R}^2) H ρ ( m ) , ω , − ( N ′ + 2 ) N neg ( R 2 ) , whose norm exceeds that at the scale ε m \varepsilon_m ε m , with norm C ( N ) ( ε ˉ m / ρ ( m ) ) N C(N)(\bar{\varepsilon}_m/\rho(m))^N C ( N ) ( ε ˉ m / ρ ( m ) ) N . The kernel of H m l o ( − i ∇ ) H_m^{\mathrm{lo}}(-i\nabla) H m lo ( − i ∇ ) is a Schwartz function at the scale ε ˉ m / c l o \bar{\varepsilon}_m/c_{\mathrm{lo}} ε ˉ m / c lo with integral 1 1 1 and vanishing moments, as H m l o ≡ 1 H_m^{\mathrm{lo}}\equiv1 H m lo ≡ 1 near 0 0 0 ; so ( 1 − H m l o ( − i ∇ ) ) ϑ (1-H_m^{\mathrm{lo}}(-i\nabla))\vartheta ( 1 − H m lo ( − i ∇ )) ϑ is its integral against the Taylor remainder of order N N N of ϑ \vartheta ϑ , of size ∣ y ~ ∣ N ∣ ∇ N ϑ ∣ |\widetilde y|^N|\nabla^N\vartheta| ∣ y ∣ N ∣ ∇ N ϑ ∣ . Peetre’s inequality moves the moment weight through the shift by y ~ \widetilde y y at the cost ( 1 + ∣ y ~ ∣ / ρ ( m ) ) N ′ + 2 (1+|\widetilde y|/\rho(m))^{N'+2} ( 1 + ∣ y ∣/ ρ ( m ) ) N ′ + 2 , and Minkowski’s inequality gives the coefficient C ( N ) ( ε ˉ m / ρ ( m ) ) N C(N)(\bar{\varepsilon}_m/\rho(m))^N C ( N ) ( ε ˉ m / ρ ( m ) ) N .
Lemma 12.14 (The future block). In the setting above, let Π + \Pi_+ Π + be the map Π b l k \Pi_{\mathrm{blk}} Π blk of Lemma 12.12 for the future block, centred at y 0 ′ y'_0 y 0 ′ , with the future insertions and N 1 N_1 N 1 . There exist C C C , c > 0 c>0 c > 0 , an integer N 3 N_3 N 3 and, for every integer N ≥ 0 N\ge0 N ≥ 0 , a constant C ( N ) C(N) C ( N ) such that:
(a) The future model on negative norms. For every F ∈ L 2 ( R 2 ) V F \in L^{2}(\mathbb{R}^{2})^{V} F ∈ L 2 ( R 2 ) V with finite right side,
∥ Π + ( F z o n e ) ∥ H ρ ( m ) , ω , N 1 N ( R 2 ) ≤ C ( N ) ∥ F ∥ H ρ ( m ) , ω , N ′ + 2 − N ( R 2 ) . \left\|\Pi_{+}(F^{\mathrm{zone}})\right\|_{H^{N}_{\rho(m),\omega,N_{1}}(\mathbb{R}^{2})} \le C(N)\left\|F\right\|_{H^{-N}_{\rho(m),\omega,N^{\prime}+2}(\mathbb{R}^{2})}. ∥ Π + ( F zone ) ∥ H ρ ( m ) , ω , N 1 N ( R 2 ) ≤ C ( N ) ∥ F ∥ H ρ ( m ) , ω , N ′ + 2 − N ( R 2 ) .
(b) The rest of the future block. Let F F F satisfy ∇ b F ∈ L ω , N ′ 2 ( R 2 ) V \nabla^{b}F \in L^{2}_{\omega,N^{\prime}}(\mathbb{R}^{2})^{V} ∇ b F ∈ L ω , N ′ 2 ( R 2 ) V for b ≤ N 1 + d N − + 1 + ⋯ + d N − + N + − 2 N + b \le N_{1}+d_{N_{-}+1}+\cdots+d_{N_{-}+N_{+}}-2N_{+} b ≤ N 1 + d N − + 1 + ⋯ + d N − + N + − 2 N + and ∥ F ∥ L ω , N ′ + 2 2 ( R 2 ) < ∞ \|F\|_{L^{2}_{\omega,N^{\prime}+2}(\mathbb{R}^{2})}<\infty ∥ F ∥ L ω , N ′ + 2 2 ( R 2 ) < ∞ , and let U F [ N + ] U_{F}^{[N_{+}]} U F [ N + ] be the last term of its Duhamel term over the block of l + 1 l+1 l + 1 from t 2 t_{2} t 2 , with the future insertions. Then
∥ U F [ N + ] ( t p r ) − Π + ( F z o n e ) ∥ H ρ m o l ( m ) , ω , N 1 N 1 ( R 2 ) ≤ C ( 1 + κ m h m ε m 2 ) N 3 ( 1 + ε ˉ m ε m ) N 3 e − c κ m τ m / ε m 2 ∥ F ∥ H ε m , ω , N ′ + 2 − N n e g ( R 2 ) . \begin{aligned}
\left\|U_{F}^{[N_{+}]}(t_{\mathrm{pr}})-\Pi_{+}(F^{\mathrm{zone}})\right\|_{H^{N_{1}}_{\rho_{\mathrm{mol}}(m),\omega,N_{1}}(\mathbb{R}^{2})} \\
&\le C\left(1+\frac{\kappa_{m}h_{m}}{\varepsilon_{m}^{2}}\right)^{N_{3}}\left(1+\frac{\bar{\varepsilon}_{m}}{\varepsilon_{m}}\right)^{N_{3}}e^{-c\kappa_{m}\tau_{m}/\varepsilon_{m}^{2}}\|F\|_{H^{-N_{\mathrm{neg}}}_{\varepsilon_{m},\omega,N^{\prime}+2}(\mathbb{R}^{2})}.
\end{aligned} U F [ N + ] ( t pr ) − Π + ( F zone ) H ρ mol ( m ) , ω , N 1 N 1 ( R 2 ) ≤ C ( 1 + ε m 2 κ m h m ) N 3 ( 1 + ε m ε ˉ m ) N 3 e − c κ m τ m / ε m 2 ∥ F ∥ H ε m , ω , N ′ + 2 − N neg ( R 2 ) .
Proof. (a) As the zone multiplier has a Schwartz kernel at the scale ε m \varepsilon_{m} ε m and the right side is finite, F z o n e F^{\mathrm{zone}} F zone and its derivatives have finite moments of order N ′ + 2 N^{\prime}+2 N ′ + 2 , by Peetre’s inequality; so F z o n e F^{\mathrm{zone}} F zone lies in the class D N ′ + 2 D_{N^{\prime}+2} D N ′ + 2 of the future block, and Lemma 12.12(ii) applies. At ξ \xi ξ in the zone its Bloch sums are constants, by (151) , as ∣ ε m ξ + 2 π n ∣ ≥ π |\varepsilon_{m}\xi+2\pi\mathbf{n}|\ge\pi ∣ ε m ξ + 2 π n ∣ ≥ π for n ∈ Z 2 ∖ { 0 } \mathbf{n}\in\mathbb{Z}^{2}\setminus\{0\} n ∈ Z 2 ∖ { 0 } :
B ( ( y − y 0 ′ ρ ( m ) ) i F z o n e ) ( ξ , ⋅ ) = i ∣ i ∣ ε m 2 ρ ( m ) ∣ i ∣ ∇ ξ i F z o n e ^ ( ξ ) . \mathcal{B}\left(\left(\frac{y-y_{0}^{\prime}}{\rho(m)}\right)^{i}F^{\mathrm{zone}}\right)(\xi,\cdot)=\frac{i^{|i|}}{\varepsilon_{m}^{2}\rho(m)|i|}\nabla_{\xi}^{i}\widehat{F^{\mathrm{zone}}}(\xi). B ( ( ρ ( m ) y − y 0 ′ ) i F zone ) ( ξ , ⋅ ) = ε m 2 ρ ( m ) ∣ i ∣ i ∣ i ∣ ∇ ξ i F zone ( ξ ) .
So Plancherel’s identity and Lemma 12.12(ii), with its integer N 2 N_{2} N 2 , the Gaussian absorbing ( 1 + ρ ( m ) ∣ ξ ∣ ) 4 N + 2 N 2 (1+\rho(m)|\xi|)^{4N+2N_{2}} ( 1 + ρ ( m ) ∣ ξ ∣ ) 4 N + 2 N 2 , give
∥ Π + ( F z o n e ) ∥ H ρ ( m ) , ω , N 1 2 N ( R 2 ) 2 ≤ C ( N ) ∑ ∣ i ∣ ≤ N ′ ∫ R 2 ( 1 + ρ ( m ) 2 ∣ ξ ∣ 2 ) − N ρ ( m ) − 2 ∣ i ∣ ∣ ∇ ξ i F z o n e ^ ( ξ ) ∣ ω 2 d ξ . \left\|\Pi_{+}(F^{\mathrm{zone}})\right\|_{H^{2N}_{\rho(m),\omega,N_{1}}(\mathbb{R}^{2})}^{2}
\le C(N)\sum_{|i|\le N^{\prime}}\int_{\mathbb{R}^{2}}\left(1+\rho(m)^{2}|\xi|^{2}\right)^{-N}\rho(m)^{-2|i|}\left|\nabla_{\xi}^{i}\widehat{F^{\mathrm{zone}}}(\xi)\right|_{\omega}^{2}\,d\xi. ∥ Π + ( F zone ) ∥ H ρ ( m ) , ω , N 1 2 N ( R 2 ) 2 ≤ C ( N ) ∣ i ∣ ≤ N ′ ∑ ∫ R 2 ( 1 + ρ ( m ) 2 ∣ ξ ∣ 2 ) − N ρ ( m ) − 2∣ i ∣ ∇ ξ i F zone ( ξ ) ω 2 d ξ .
The zone multiplier has derivatives bounded at the scale ε m ≤ ρ ( m ) \varepsilon_{m}\le\rho(m) ε m ≤ ρ ( m ) , so F F F may replace F z o n e F^{\mathrm{zone}} F zone on the right; by Plancherel this is at most C ( N ) C(N) C ( N ) times the sum over ∣ i ∣ ≤ N ′ |i|\le N^{\prime} ∣ i ∣ ≤ N ′ of the squared H ρ ( m ) , ω , 0 − N ( R 2 ) H^{-N}_{\rho(m),\omega,0}(\mathbb{R}^{2}) H ρ ( m ) , ω , 0 − N ( R 2 ) norms of ( ( y − y 0 ′ ) / ρ ( m ) ) i F ((y-y_{0}^{\prime})/\rho(m))^{i}F (( y − y 0 ′ ) / ρ ( m ) ) i F , and, as multiplication by ( ( y − y 0 ′ ) / ρ ( m ) ) i ((y-y_{0}^{\prime})/\rho(m))^{i} (( y − y 0 ′ ) / ρ ( m ) ) i maps H ρ ( m ) , ω , 0 N ( R 2 ) H^{N}_{\rho(m),\omega,0}(\mathbb{R}^{2}) H ρ ( m ) , ω , 0 N ( R 2 ) into H ρ ( m ) , ω , − ( N ′ + 2 ) N ( R 2 ) H^{N}_{\rho(m),\omega,-(N^{\prime}+2)}(\mathbb{R}^{2}) H ρ ( m ) , ω , − ( N ′ + 2 ) N ( R 2 ) , duality gives (a).
(b) The Duhamel terms from F z o n e F^{\mathrm{zone}} F zone and F h i : = F − F z o n e F^{\mathrm{hi}}:=F-F^{\mathrm{zone}} F hi := F − F zone exist, as both lie in the class of (b) (F z o n e F^{\mathrm{zone}} F zone by Peetre’s inequality, as in (a)), and by linearity U F [ N + ] − Π + ( F z o n e ) U_{F}^{[N_{+}]}-\Pi_{+}(F^{\mathrm{zone}}) U F [ N + ] − Π + ( F zone ) is the sum of U F z o n e [ N + ] − Π + ( F z o n e ) U_{F^{\mathrm{zone}}}^{[N_{+}]}-\Pi_{+}(F^{\mathrm{zone}}) U F zone [ N + ] − Π + ( F zone ) and U F h i [ N + ] U_{F^{\mathrm{hi}}}^{[N_{+}]} U F hi [ N + ] . As in (a), at the scale ε m \varepsilon_{m} ε m , ∥ F z o n e ∥ L ω , N ′ 2 ( R 2 ) ≤ C ∥ F ∥ H ε m , ω , N ′ + 2 − N n e g ( R 2 ) \|F^{\mathrm{zone}}\|_{L^{2}_{\omega,N^{\prime}}(\mathbb{R}^{2})}\le C\|F\|_{H^{-N_{\mathrm{neg}}}_{\varepsilon_{m},\omega,N^{\prime}+2}(\mathbb{R}^{2})} ∥ F zone ∥ L ω , N ′ 2 ( R 2 ) ≤ C ∥ F ∥ H ε m , ω , N ′ + 2 − N neg ( R 2 ) , so Lemma 12.12(iii) bounds the first term at t p r t_{\mathrm{pr}} t pr . For the second we use the quiet stretch [ t 2 , t q ] [t_{2},t_{q}] [ t 2 , t q ] , t q : = t 2 + τ m ′ − τ m / 24 t_{q}:=t_{2}+\tau_{m}^{\prime}-\tau_{m}/24 t q := t 2 + τ m ′ − τ m /24 , the first part of the quiet margin of the block of l + 1 l+1 l + 1 . As τ m ′ ≥ 5 τ m \tau_{m}^{\prime}\ge5\tau_{m} τ m ′ ≥ 5 τ m , its length in cell units lies between κ m τ m ′ / ( 2 ε m ) \kappa_{m}\tau_{m}^{\prime}/(2\varepsilon_{m}) κ m τ m ′ / ( 2 ε m ) , which is at least 1 1 1 for m m m large, and κ m h m / ε m 2 \kappa_{m}h_{m}/\varepsilon_{m}^{2} κ m h m / ε m 2 . Let U F h i [ j , i ] U_{F^{\mathrm{hi}}}^{[j,i]} U F hi [ j , i ] be the moment terms of the Duhamel term of F h i F^{\mathrm{hi}} F hi over the block of l + 1 l+1 l + 1 , centred at y 0 ′ y_{0}^{\prime} y 0 ′ .
The quiet stretch. On [ t 2 , t q ] [t_{2},t_{q}] [ t 2 , t q ] no pulse is active within τ m / 24 \tau_{m}/24 τ m /24 , so ψ m = 0 \psi_{m}=0 ψ m = 0 and the coefficients of the insertions do not depend on z z z , by clause (i) of the definition of insertions; once the weight is moved through as in the sentence after (142) , they do not depend on y y y either. So at each frequency η \eta η the system (142) for the Fourier transforms of the U F h i [ j , i ] U_{F^{\mathrm{hi}}}^{[j,i]} U F hi [ j , i ] is triangular, with the common diagonal − κ m η ⋅ K ( t ) η -\kappa_{m}\eta\cdot\mathbf{K}(t)\eta − κ m η ⋅ K ( t ) η and off-diagonal entries κ m / ε m 2 \kappa_{m}/\varepsilon_{m}^{2} κ m / ε m 2 times polynomials in ε m η \varepsilon_{m}\eta ε m η with bounded coefficients (the commutator terms carry ε m / ρ ( m ) \varepsilon_{m}/\rho(m) ε m / ρ ( m ) and ( ε m / ρ ( m ) ) 2 (\varepsilon_{m}/\rho(m))^{2} ( ε m / ρ ( m ) ) 2 , and ε m ≤ ρ ( m ) ≤ ε m − 1 \varepsilon_{m}\le\rho(m)\le\varepsilon_{m-1} ε m ≤ ρ ( m ) ≤ ε m − 1 ). As the terms with j ≥ 1 j\ge1 j ≥ 1 vanish at t 2 t_{2} t 2 ,
U F h i [ j , i ] ^ ( t q , η ) = exp ( − κ m ∫ t 2 t q η ⋅ K ( t ) η d t ) ∑ ∣ i ′ ∣ ≤ N ′ P j , i , i ′ ( ε m η ) U F h i [ 0 , i ′ ] ^ ( t 2 , η ) . \widehat{U_{F^{\mathrm{hi}}}^{[j,i]}}(t_{q},\eta)=\exp\left(-\kappa_{m}\int_{t_{2}}^{t_{q}}\eta\cdot\mathbf{K}(t)\eta\,dt\right)\sum_{|i'|\le N^{\prime}}P_{j,i,i'}(\varepsilon_{m}\eta)\widehat{U_{F^{\mathrm{hi}}}^{[0,i']}}(t_{2},\eta). U F hi [ j , i ] ( t q , η ) = exp ( − κ m ∫ t 2 t q η ⋅ K ( t ) η d t ) ∣ i ′ ∣ ≤ N ′ ∑ P j , i , i ′ ( ε m η ) U F hi [ 0 , i ′ ] ( t 2 , η ) .
with P j , i , i ′ ( x ) P_{j,i,i'}(x) P j , i , i ′ ( x ) a linear map on C V \mathbb{C}^{V} C V , polynomial in the length of the stretch, whose norm on ( C V , ∣ ⋅ ∣ ω ) (\mathbb{C}^{V},|\cdot|_{\omega}) ( C V , ∣ ⋅ ∣ ω ) is at most C ( 1 + κ m h m / ε m 2 ) C ( 1 + ∣ x ∣ ) C C(1+\kappa_m h_m/\varepsilon_m^2)^C(1+|x|)^C C ( 1 + κ m h m / ε m 2 ) C ( 1 + ∣ x ∣ ) C . By (143) , U ^ F h i [ 0 , i ′ ] ( t 2 , η ) = ∣ i ′ ∣ ∣ ρ ( m ) − ∣ i ∣ ∣ ∇ η i ′ F h i ^ ( η ) \widehat{U}^{[0,i']}_{F h_i}(t_2,\eta)=\lvert i'\rvert^{\lvert\rho(m)-\lvert i\rvert\rvert}\nabla_{\eta}^{i'}\widehat{F h_i}(\eta) U F h i [ 0 , i ′ ] ( t 2 , η ) = ∣ i ′ ∣ ∣ ρ ( m ) − ∣ i ∣∣ ∇ η i ′ F h i ( η ) , which vanishes for ε m ∣ η ∣ ≤ 3 π / 8 \varepsilon_m|\eta|\leq3\pi/8 ε m ∣ η ∣ ≤ 3 π /8 . As K ≥ 15 16 I K\geq\frac{15}{16}I K ≥ 16 15 I , the exponential is at most exp ( − 15 32 κ m τ m ′ ∣ η ∣ 2 ) \exp(-\frac{15}{32}\kappa_m\tau'_m|\eta|^2) exp ( − 32 15 κ m τ m ′ ∣ η ∣ 2 ) : half of the exponent absorbs ( 1 + ε m ∣ η ∣ ) C (1+\varepsilon_m|\eta|)^C ( 1 + ε m ∣ η ∣ ) C , as κ m τ m ′ / ε m 2 ≥ 2 \kappa_m\tau'_m/\varepsilon_m^2\geq2 κ m τ m ′ / ε m 2 ≥ 2 , and the other half is at most exp ( − κ m τ m ′ / ( 4 ε m 2 ) ) \exp(-\kappa_m\tau'_m/(4\varepsilon_m^2)) exp ( − κ m τ m ′ / ( 4 ε m 2 )) where ε m ∣ η ∣ ≥ 3 π / 8 \varepsilon_m|\eta|\geq3\pi/8 ε m ∣ η ∣ ≥ 3 π /8 . Each ∇ η \nabla_\eta ∇ η that falls on the zone multiplier costs ε m ≤ ρ ( m ) \varepsilon_m\leq\rho(m) ε m ≤ ρ ( m ) , and multiplication by ( ( y − y 0 ) / ρ ( m ) ) i ′ ((y-y^0)/\rho(m))^{i'} (( y − y 0 ) / ρ ( m ) ) i ′ , ∣ i ′ ∣ ≤ N ′ |i'|\leq N' ∣ i ′ ∣ ≤ N ′ , maps H ε m , ω , 0 N n e g ( R 2 ) H^{N_{\mathrm{neg}}}_{\varepsilon_m,\omega,0}(\mathbb{R}^2) H ε m , ω , 0 N neg ( R 2 ) into H ε m , ω , − ( N ′ + 2 ) N n e g ( R 2 ) H^{N_{\mathrm{neg}}}_{\varepsilon_m,\omega,-(N'+2)}(\mathbb{R}^2) H ε m , ω , − ( N ′ + 2 ) N neg ( R 2 ) , as in (a) at the scale ε m ≤ ρ ( m ) \varepsilon_m\leq\rho(m) ε m ≤ ρ ( m ) . So
∑ ( j , i ) ∫ R 2 ( 1 + ε m 2 ∣ η ∣ 2 ) − N n e g ∣ U ^ F h i [ j , i ] ( t q , η ) ∣ ω 2 d η ≤ C ( 1 + κ m h m ε m 2 ) C exp ( − κ m τ m ′ 2 ε m 2 ) ∥ F ∥ H ε m , ω , N ′ + 2 − N n e g ( R 2 ) 2 . \sum_{(j,i)}\int_{\mathbb{R}^2}(1+\varepsilon_m^2|\eta|^2)^{-N_{\mathrm{neg}}}\left|\widehat{U}^{[j,i]}_{F h_i}(t_q,\eta)\right|_\omega^2\,d\eta
\leq C\left(1+\frac{\kappa_m h_m}{\varepsilon_m^2}\right)^C\exp\left(-\frac{\kappa_m\tau'_m}{2\varepsilon_m^2}\right)\|F\|_{H^{-N_{\mathrm{neg}}}_{\varepsilon_m,\omega,N'+2}(\mathbb{R}^2)}^2. ( j , i ) ∑ ∫ R 2 ( 1 + ε m 2 ∣ η ∣ 2 ) − N neg U F h i [ j , i ] ( t q , η ) ω 2 d η ≤ C ( 1 + ε m 2 κ m h m ) C exp ( − 2 ε m 2 κ m τ m ′ ) ∥ F ∥ H ε m , ω , N ′ + 2 − N neg ( R 2 ) 2 .
The same formula, with up to two further η \eta η -derivatives, shows that the moment terms have at t q t_q t q every derivative with two moments, so their Bloch sums are defined; the moments persist for t ≥ t q t\geq t_q t ≥ t q by the moment-weighted energy (paragraph Duhamel terms ), and the Bloch sums solve the fibre system, by the paragraph of (130) .
From the plane to the fibres. Let h h h be a field on R 2 \mathbb{R}^2 R 2 with two moments and ξ \xi ξ in the zone. The modes of B h ( ξ , ⋅ ) B h(\xi,\cdot) B h ( ξ , ⋅ ) are given by (151) , and, as ∣ ε m ξ ∣ 2 ≤ 2 π 2 |\varepsilon_m\xi|^2\leq2\pi^2 ∣ ε m ξ ∣ 2 ≤ 2 π 2 , the weights 1 + ∣ 2 π n ∣ 2 1+|2\pi n|^2 1 + ∣2 π n ∣ 2 and 1 + ∣ 2 π n + ε m ξ ∣ 2 1+|2\pi n+\varepsilon_m\xi|^2 1 + ∣2 π n + ε m ξ ∣ 2 are comparable with the constant 2 + 4 π 2 2+4\pi^2 2 + 4 π 2 . So, componentwise and then with the masses, for every integer K ≥ 0 K\geq0 K ≥ 0 , with ∥ ⋅ ∥ H − K ( T 2 ) \|\cdot\|_{H^{-K}(\mathbb{T}^2)} ∥ ⋅ ∥ H − K ( T 2 ) the norm dual to H K ( T 2 ) H^K(\mathbb{T}^2) H K ( T 2 ) for ⟨ ⋅ , ⋅ ⟩ L ω 2 ( T 2 ) \langle\cdot,\cdot\rangle_{L^2_\omega(\mathbb{T}^2)} ⟨ ⋅ , ⋅ ⟩ L ω 2 ( T 2 ) ,
∫ z o n e ∥ B h ( ξ , ⋅ ) ∥ H − K ( T 2 ) 2 d ξ ≤ C ( K ) ε m − 4 ∫ R 2 ( 1 + ε m 2 ∣ η ∣ 2 ) − K ∣ h ^ ( η ) ∣ ω 2 d η . \int_{\mathrm{zone}}\|B h(\xi,\cdot)\|_{H^{-K}(\mathbb{T}^2)}^2\,d\xi
\leq C(K)\varepsilon_m^{-4}\int_{\mathbb{R}^2}(1+\varepsilon_m^2|\eta|^2)^{-K}|\widehat{h}(\eta)|_\omega^2\,d\eta. ∫ zone ∥ B h ( ξ , ⋅ ) ∥ H − K ( T 2 ) 2 d ξ ≤ C ( K ) ε m − 4 ∫ R 2 ( 1 + ε m 2 ∣ η ∣ 2 ) − K ∣ h ( η ) ∣ ω 2 d η .
The rest of the block. Let t ∈ [ t q , t p r ] t\in[t_q,t_{\mathrm{pr}}] t ∈ [ t q , t pr ] , let ξ \xi ξ be in the zone, and let U † ∈ H N j , i + N n e g ( T 2 ) V U^\dagger\in H^{N_{j,i}+N_{\mathrm{neg}}}(\mathbb{T}^2)^V U † ∈ H N j , i + N neg ( T 2 ) V . The induction of Step 1 of the proof of Lemma 12.12(iii), started at t q t_q t q with a data term for each moment term and at the test orders raised by N n e g N_{\mathrm{neg}} N neg , gives
∣ ⟨ B U F h i [ j , i ] ( t ) , U † ⟩ L ω 2 ( T 2 ) ∣ ≤ C ( 1 + κ m ( t − t q ) ε m 2 ) C ( 1 + ε ˉ m ε m ) C ∥ U † ∥ H N j , i + N n e g ( T 2 ) × ∑ ( j ′ , i ′ ) ∣ B U F h i [ j ′ , i ′ ] ( t q ) ∣ H − N n e g ( T 2 ) , \left|\left\langle B U^{[j,i]}_{F h_i}(t),U^\dagger\right\rangle_{L^2_\omega(\mathbb{T}^2)}\right|
\leq C\left(1+\frac{\kappa_m(t-t_q)}{\varepsilon_m^2}\right)^C\left(1+\frac{\bar{\varepsilon}_m}{\varepsilon_m}\right)^C\|U^\dagger\|_{H^{N_{j,i}+N_{\mathrm{neg}}}(\mathbb{T}^2)}
\times\sum_{(j',i')}\left|B U^{[j',i']}_{F h_i}(t_q)\right|_{H^{-N_{\mathrm{neg}}}(\mathbb{T}^2)}, ⟨ B U F h i [ j , i ] ( t ) , U † ⟩ L ω 2 ( T 2 ) ≤ C ( 1 + ε m 2 κ m ( t − t q ) ) C ( 1 + ε m ε ˉ m ) C ∥ U † ∥ H N j , i + N neg ( T 2 ) × ( j ′ , i ′ ) ∑ B U F h i [ j ′ , i ′ ] ( t q ) H − N neg ( T 2 ) ,
the sum over the moment terms ( j ′ , i ′ ) (j',i') ( j ′ , i ′ ) equal to or lower than ( j , i ) (j,i) ( j , i ) , with the coefficients differentiated within the derivative counts for a Duhamel term across the renewal. Step 2 of that proof, on N + + N ′ + 1 N_++N'+1 N + + N ′ + 1 equal parts of the last cell time of the block, which lies in [ t q , t p r ] [t_q,t_{\mathrm{pr}}] [ t q , t pr ] , from the negative order raised by N n e g N_{\mathrm{neg}} N neg , bounds every ( N + , i ) (N_+,i) ( N + , i ) , ∣ i ∣ ≤ N 1 |i|\leq N_1 ∣ i ∣ ≤ N 1 , in H N 1 ( T 2 ) H^{N_1}(\mathbb{T}^2) H N 1 ( T 2 ) at t p r t_{\mathrm{pr}} t pr by C ( 1 + κ m h m / ε m 2 ) C ( 1 + ε ˉ m / ε m ) C C(1+\kappa_m h_m/\varepsilon_m^2)^C(1+\bar{\varepsilon}_m/\varepsilon_m)^C C ( 1 + κ m h m / ε m 2 ) C ( 1 + ε ˉ m / ε m ) C times the same sum. We square and integrate over the zone: the Parseval identity of Step 3 of that proof on the left side, and the bound from the plane to the fibres with K = N n e g K=N_{\mathrm{neg}} K = N neg and the bound at t q t_q t q on the right side, give, the powers ε m 4 \varepsilon_m^4 ε m 4 and ε m − 4 \varepsilon_m^{-4} ε m − 4 cancelling and as τ m ′ ≥ τ m \tau'_m\geq\tau_m τ m ′ ≥ τ m ,
∥ U F h i [ N + ] ( t p r ) ∥ H ρ m o l ( m ) , ω , N 1 N 1 ( R 2 ) ≤ C ( 1 + κ m h m ε m 2 ) C ( 1 + ε ˉ m ε m ) C exp ( − κ m τ m 4 ε m 2 ) ∥ F ∥ H ε m , ω , N ′ + 2 − N n e g ( R 2 ) . \|U^{[N_+]}_{F h_i}(t_{\mathrm{pr}})\|_{H^{N_1}_{\rho_{\mathrm{mol}}(m),\omega,N_1}(\mathbb{R}^2)}
\leq C\left(1+\frac{\kappa_m h_m}{\varepsilon_m^2}\right)^C\left(1+\frac{\bar{\varepsilon}_m}{\varepsilon_m}\right)^C\exp\left(-\frac{\kappa_m\tau_m}{4\varepsilon_m^2}\right)\|F\|_{H^{-N_{\mathrm{neg}}}_{\varepsilon_m,\omega,N'+2}(\mathbb{R}^2)}. ∥ U F h i [ N + ] ( t pr ) ∥ H ρ mol ( m ) , ω , N 1 N 1 ( R 2 ) ≤ C ( 1 + ε m 2 κ m h m ) C ( 1 + ε m ε ˉ m ) C exp ( − 4 ε m 2 κ m τ m ) ∥ F ∥ H ε m , ω , N ′ + 2 − N neg ( R 2 ) .
This is (b) for the second term, and Lemma 12.12(iii) for the future block gives (b) for the first, for c c c at most 1 4 \frac14 4 1 and the c c c of that lemma, and N 3 N_3 N 3 at least the exponents above and its N 2 N_2 N 2 . ∎
Lemma 12.15 assembles the two lemmas along Figure 12.3 : its model is bounded without loss, and the losses in its two errors, the future block less its model and the high frequencies of the old term, are offset by the damping of the future block and the gain of Lemma 12.13(c).
Lemma 12.15 (The first renewal). Let J ≥ 0 J \ge0 J ≥ 0 be an integer, and consider a Duhamel term across the renewal from a datum g ∈ H N 1 + d 1 + ⋯ + d N − + N + − 2 ( N − + N + ) ( R 2 ) g \in H^{N_1+d_1+\cdots+d_{N_-}+N_+-2(N_-+N_+)}(\mathbb{R}^2) g ∈ H N 1 + d 1 + ⋯ + d N − + N + − 2 ( N − + N + ) ( R 2 ) supported within distance ρ ( m ) \rho(m) ρ ( m ) of y 0 y_0 y 0 . There exist C C C and a linear map Π r e n \Pi_{\mathrm{ren}} Π ren , independent of g g g , such that, with the moments about y 0 y_0 y 0 on the left sides,
∥ Π r e n g ∥ H ρ ( m ) , ω , N 1 J ( R 2 ) ≤ C ∥ g ∥ L ω 2 ( R 2 ) , \|\Pi_{\mathrm{ren}}g\|_{H^J_{\rho(m),\omega,N_1}(\mathbb{R}^2)} \le C\|g\|_{L^2_\omega(\mathbb{R}^2)}, ∥ Π ren g ∥ H ρ ( m ) , ω , N 1 J ( R 2 ) ≤ C ∥ g ∥ L ω 2 ( R 2 ) , ∥ U [ N − + N + ] ( t p r ) − Π r e n g ∥ H ρ m o l ( m ) , ω , N 1 N 1 ( R 2 ) ≤ C ε m − 1 Q + 3 ∥ g ∥ L ω 2 ( R 2 ) . \|U^{[N_-+N_+]}(t_{\mathrm{pr}})-\Pi_{\mathrm{ren}}g\|_{H^{N_1}_{\rho_{\mathrm{mol}}(m),\omega,N_1}(\mathbb{R}^2)} \le C\varepsilon_{m-1}^{Q+3}\|g\|_{L^2_\omega(\mathbb{R}^2)}. ∥ U [ N − + N + ] ( t pr ) − Π ren g ∥ H ρ mol ( m ) , ω , N 1 N 1 ( R 2 ) ≤ C ε m − 1 Q + 3 ∥ g ∥ L ω 2 ( R 2 ) .
Proof. Let N 3 N_3 N 3 be the larger of the integers of Lemmas 12.13 and 12.14; both hold with it. The model is the solid path of Figure 12.3 ,
Π r e n g : = Π + ( ( ( U [ N − ] ( t 2 ) ) o ∘ X m − 1 ( ℓ h m , ⋅ , ( ℓ + 1 ) h m ) ) z o n e ) . \Pi_{\mathrm{ren}}g:=\Pi_+\left(\left((U^{[N_-]}(t_2))^\mathrm{o}\circ X_{m-1}(\ell h_m,\cdot,(\ell+1)h_m)\right)^\mathrm{zone}\right). Π ren g := Π + ( ( ( U [ N − ] ( t 2 ) ) o ∘ X m − 1 ( ℓ h m , ⋅ , ( ℓ + 1 ) h m ) ) zone ) .
It is linear, and ∥ g ∥ L ω , N 3 2 ( R 2 ) ≤ 2 N 3 / 2 ∥ g ∥ L ω 2 ( R 2 ) \|g\|_{L^2_{\omega,N_3}(\mathbb{R}^2)}\le2^{N_3/2}\|g\|_{L^2_\omega(\mathbb{R}^2)} ∥ g ∥ L ω , N 3 2 ( R 2 ) ≤ 2 N 3 /2 ∥ g ∥ L ω 2 ( R 2 ) by the support of $g. By Lemma 12.14(a) at the order max { J , N 3 } \max\{J,N_3\} max { J , N 3 } , and Lemma 12.13(c) at the scale ρ ( m ) \rho(m) ρ ( m ) and Lemma 12.13(b), ∥ Π r e n g ∥ H ρ ( m ) , ω , N 1 J ( R 2 ) ≤ C ∥ g ∥ L ω 2 ( R 2 ) \|\Pi_{\mathrm{ren}}g\|_{H^J_{\rho(m),\omega,N_1}(\mathbb{R}^2)}\le C\|g\|_{L^2_\omega(\mathbb{R}^2)} ∥ Π ren g ∥ H ρ ( m ) , ω , N 1 J ( R 2 ) ≤ C ∥ g ∥ L ω 2 ( R 2 ) .
The error. U [ N − + N + ] ( t p r ) U^{[N_-+N_+]}(t_{\mathrm{pr}}) U [ N − + N + ] ( t pr ) is the last term at t p r t_{\mathrm{pr}} t pr of the Duhamel term over the block of ℓ + 1 \ell+1 ℓ + 1 from F : = U [ N − ] ( t 2 ) ∘ X m − 1 ( ℓ h m , ⋅ , ( ℓ + 1 ) h m ) F:=U^{[N_-]}(t_2)\circ X_{m-1}(\ell h_m,\cdot,(\ell+1)h_m) F := U [ N − ] ( t 2 ) ∘ X m − 1 ( ℓ h m , ⋅ , ( ℓ + 1 ) h m ) , which lies in the class of Lemma 12.14(b): as g g g has compact support, the weighted energy of Proposition 8.6 with the moment weights gives finite moments to U [ N − ] ( t 2 ) U^{[N_-]}(t_2) U [ N − ] ( t 2 ) and to its derivatives of order at most N 1 + d N − + 1 + ⋯ + d N − + N + − 2 N + N_1+d_{N_-+1}+\cdots+d_{N_-+N_+}-2N_+ N 1 + d N − + 1 + ⋯ + d N − + N + − 2 N + , the j j j -th old insertion lowering the order by d j − 2 d_j-2 d j − 2 , and the change of variables, area preserving with derivatives bounded at the scale ε m − 1 ≥ ρ ( m ) \varepsilon_{m-1}\ge\rho(m) ε m − 1 ≥ ρ ( m ) , keeps the class. So U [ N − + N + ] ( t p r ) − Π r e n g U^{[N_-+N_+]}(t_{\mathrm{pr}})-\Pi_{\mathrm{ren}}g U [ N − + N + ] ( t pr ) − Π ren g is the sum of U [ N − + N + ] ( t p r ) − Π + ( F z o n e ) U^{[N_-+N_+]}(t_{\mathrm{pr}})-\Pi_+(F^\mathrm{zone}) U [ N − + N + ] ( t pr ) − Π + ( F zone ) and of Π + \Pi_+ Π + applied to the zone part of ( U [ N − ] ( t 2 ) − ( U [ N − ] ( t 2 ) ) o ) ∘ X m − 1 ( ℓ h m , ⋅ , ( ℓ + 1 ) h m ) (U^{[N_-]}(t_2)-(U^{[N_-]}(t_2))^\mathrm{o})\circ X_{m-1}(\ell h_m,\cdot,(\ell+1)h_m) ( U [ N − ] ( t 2 ) − ( U [ N − ] ( t 2 ) ) o ) ∘ X m − 1 ( ℓ h m , ⋅ , ( ℓ + 1 ) h m ) . By Lemmas 12.14(b), 12.13(c) at the scale ε m \varepsilon_m ε m and 12.13(a) for the first, and by Lemmas 12.14(a) with N e r r + N n e g N_{\mathrm{err}}+N_{\mathrm{neg}} N err + N neg in place of N N N (N e r r ≥ N 1 N_{\mathrm{err}}\ge N_1 N err ≥ N 1 is chosen below; ρ m o l ( m ) ≤ ρ ( m ) \rho_{\mathrm{mol}}(m)\le\rho(m) ρ mol ( m ) ≤ ρ ( m ) ), 12.13(c) with (152) , and 12.13(a) for the second, they are at most, times ∥ g ∥ L ω 2 ( R 2 ) \|g\|_{L^2_\omega(\mathbb{R}^2)} ∥ g ∥ L ω 2 ( R 2 ) ,
C ( 1 + κ m h m ε m 2 ) 2 N 3 ( 1 + ε ˉ m / ε m ) 2 N 3 e − c κ m τ m ε m 2 , C ( N e r r ) ( κ m h m ε m 2 ) − N e r r / 2 ( 1 + κ m h m ε m 2 ) N 3 ( 1 + ε ˉ m / ε m ) N 3 . C\left(1+\frac{\kappa_mh_m}{\varepsilon_m^2}\right)^{2N_3}(1+\bar\varepsilon_m/\varepsilon_m)^{2N_3}e^{-\frac{c\kappa_m\tau_m}{\varepsilon_m^2}},\qquad C(N_{\mathrm{err}})\left(\frac{\kappa_mh_m}{\varepsilon_m^2}\right)^{-N_{\mathrm{err}}/2}\left(1+\frac{\kappa_mh_m}{\varepsilon_m^2}\right)^{N_3}(1+\bar\varepsilon_m/\varepsilon_m)^{N_3}. C ( 1 + ε m 2 κ m h m ) 2 N 3 ( 1 + ε ˉ m / ε m ) 2 N 3 e − ε m 2 c κ m τ m , C ( N err ) ( ε m 2 κ m h m ) − N err /2 ( 1 + ε m 2 κ m h m ) N 3 ( 1 + ε ˉ m / ε m ) N 3 .
The powers of ε m − 1 \varepsilon_{m-1} ε m − 1 . With the bounds of Step 4 of the proof of Lemma 12.12, which with h m / τ m ≥ ε m − 1 − 2 δ h_m/\tau_m\ge\varepsilon_{m-1}^{-2\delta} h m / τ m ≥ ε m − 1 − 2 δ (by (6) ) give κ m h m / ε m 2 ≥ c ε m − 1 − 6 δ \kappa_mh_m/\varepsilon_m^2\ge c\varepsilon_{m-1}^{-6\delta} κ m h m / ε m 2 ≥ c ε m − 1 − 6 δ , the first term is at most C ε m − 1 Q + 3 ∥ g ∥ L ω 2 ( R 2 ) C\varepsilon_{m-1}^{Q+3}\|g\|_{L^2_\omega(\mathbb{R}^2)} C ε m − 1 Q + 3 ∥ g ∥ L ω 2 ( R 2 ) for m m m large, and the second, as N e r r ≥ 2 N 3 N_{\mathrm{err}}\ge2N_3 N err ≥ 2 N 3 by the choice below, at most C ( N e r r ) ε m − 1 6 δ ( N e r r / 2 − N 3 ) − q γ N 3 ∥ g ∥ L ω 2 ( R 2 ) C(N_{\mathrm{err}})\varepsilon_{m-1}^{6\delta(N_{\mathrm{err}}/2-N_3)-q\gamma N_3}\|g\|_{L^2_\omega(\mathbb{R}^2)} C ( N err ) ε m − 1 6 δ ( N err /2 − N 3 ) − q γ N 3 ∥ g ∥ L ω 2 ( R 2 ) ; we choose N e r r N_{\mathrm{err}} N err , after N 3 N_3 N 3 , with 6 δ ( N e r r / 2 − N 3 ) − q γ N 3 ≥ Q + 3 6\delta(N_{\mathrm{err}}/2-N_3)-q\gamma N_3\ge Q+3 6 δ ( N err /2 − N 3 ) − q γ N 3 ≥ Q + 3 . □ \square □
In the proof of Proposition 12.7, moments control the freezing: the coefficients vary on the scale ε m − 1 \varepsilon_{m-1} ε m − 1 , much larger than ρ ( m ) \rho(m) ρ ( m ) , and the moments of a piece are multiplied by at most a constant over each block (Step 3); the models of the pieces, each localized near its centre, are almost orthogonal (Step 5).
Proof of Proposition 12.7. Step 1: The time interval. With t 2 t_2 t 2 and t p r t_{\mathrm{pr}} t pr as in (138) , t 1 < t 2 ≤ t 1 + h m t_1<t_2\le t_1+h_m t 1 < t 2 ≤ t 1 + h m gives the bounds there, and the slow cutoff (11) vanishes within τ m ′ \tau'_m τ m ′ of every block boundary, which gives the margin.
Step 2: Patches. We cover the labels at time t 1 t_1 t 1 , in the chart of the block of t 1 t_1 t 1 , by a smooth partition of unity at the scale ρ ( m ) / 8 \rho(m)/8 ρ ( m ) /8 , with bounded overlap, whose functions are supported within ρ ( m ) / 4 \rho(m)/4 ρ ( m ) /4 of their centres. Their products with g g g , the pieces, have squared L ω 2 ( T 2 ) L^2_\omega(\mathbb{T}^2) L ω 2 ( T 2 ) norms of sum at most ∥ g ∥ L ω 2 ( T 2 ) 2 \|g\|^2_{L^2_\omega(\mathbb{T}^2)} ∥ g ∥ L ω 2 ( T 2 ) 2 , and the same in H ρ m o l ( m ) , ω N b u f ( T 2 ) H^{N_{\mathrm{buf}}}_{\rho_{\mathrm{mol}}(m),\omega}(\mathbb{T}^2) H ρ mol ( m ) , ω N buf ( T 2 ) up to a constant, as the partition varies on the scale ρ ( m ) ≥ ρ m o l ( m ) \rho(m) \ge\rho_{\mathrm{mol}}(m) ρ ( m ) ≥ ρ mol ( m ) and the chart of the block of t 1 t_1 t 1 preserves area and has derivatives bounded at the scale ε m − 1 ≥ ρ ( m ) \varepsilon_{m-1} \ge\rho(m) ε m − 1 ≥ ρ ( m ) , by (54) . Each piece is a datum on R 2 \mathbb{R}^2 R 2 . Fix a piece, and let y 0 y_0 y 0 be its centre.
Step 3: Freezing (the decisive estimate). On the patch of y 0 y_0 y 0 , we replace the conductivity and the two-scale forms of the sources by their Taylor polynomials of degree N T a y N_{\mathrm{Tay}} N Tay at y 0 y_0 y 0 , on the block of l + 1 l+1 l + 1 at y 0 ′ y'_0 y 0 ′ , the label of the same point in its chart ( X m − 1 ( l h m , y 0 ′ , ( l + 1 ) h m ) = y 0 ) (X_{m-1}(lh_m,y'_0,(l+1)h_m)=y_0) ( X m − 1 ( l h m , y 0 ′ , ( l + 1 ) h m ) = y 0 ) . We expand the conductivity polynomial as insertions (Subsection 12.6), up to N T a y N_{\mathrm{Tay}} N Tay of them on the two blocks together. Near the ends of a block, the forms given in the chart of the neighbouring block are insertions of the block, of C C C times the same strength (paragraph Insertions ; (a4), (a6), (54) ; the items (a1)–(d) cited in this proof are those of Definition 11.1). Moments bound the remainders: by (54) , the conductivity differs from its Taylor polynomial by at most
C ε m − 1 2 δ ( ∣ y − y 0 ∣ ε m − 1 ) N T a y + 1 = C ε m − 1 2 δ ( ρ ( m ) ε m − 1 ) N T a y + 1 ( ∣ y − y 0 ∣ ρ ( m ) ) N T a y + 1 , C\varepsilon_{m-1}^{2\delta}\left(\frac{|y-y_0|}{\varepsilon_{m-1}}\right)^{N_{\mathrm{Tay}}+1}
= C\varepsilon_{m-1}^{2\delta}\left(\frac{\rho(m)}{\varepsilon_{m-1}}\right)^{N_{\mathrm{Tay}}+1}\left(\frac{|y-y_0|}{\rho(m)}\right)^{N_{\mathrm{Tay}}+1}, C ε m − 1 2 δ ( ε m − 1 ∣ y − y 0 ∣ ) N Tay + 1 = C ε m − 1 2 δ ( ε m − 1 ρ ( m ) ) N Tay + 1 ( ρ ( m ) ∣ y − y 0 ∣ ) N Tay + 1 ,
and the same holds for the forms in the sums of (b2′) with j = N T a y + 1 j=N_{\mathrm{Tay}}+1 j = N Tay + 1 , with C ∗ C_* C ∗ times the entry of the majorant matrix in place of ε m − 1 2 δ \varepsilon_{m-1}^{2\delta} ε m − 1 2 δ ; here N T a y + 1 N_{\mathrm{Tay}}+1 N Tay + 1 is an internal order of Proposition 12.7, so at most N c o e f N_{\mathrm{coef}} N coef (the table of Subsection 11.1). By Proposition 2.1(c), and as each power of ( y − y 0 ) / ε m − 1 (y-y_0)/\varepsilon_{m-1} ( y − y 0 ) / ε m − 1 costs one moment and the strength below,
ρ ( m ) ε m − 1 ≤ C ε γ / 2 + δ , C ε m − 1 2 δ ρ ( m ) ε m − 1 ≤ C ε γ / 2 + 3 δ \frac{\rho(m)}{\varepsilon_{m-1}} \le C\varepsilon^{\gamma/2+\delta}, \qquad C\varepsilon_{m-1}^{2\delta}\frac{\rho(m)}{\varepsilon_{m-1}} \le C\varepsilon^{\gamma/2+3\delta} ε m − 1 ρ ( m ) ≤ C ε γ /2 + δ , C ε m − 1 2 δ ε m − 1 ρ ( m ) ≤ C ε γ /2 + 3 δ
(the strength of a conductivity insertion).
Telescoping the frozen evolution with its insertions against the actual one leaves, as forcing of the difference, the Taylor remainders and the term with N T a y + 1 N_{\mathrm{Tay}}+1 N Tay + 1 conductivity insertions. These forcings take at most N ♯ : = N T a y ( N T a y + N d e p + 1 ) + 1 N_\sharp:= N_{\mathrm{Tay}}(N_{\mathrm{Tay}}+N_{\mathrm{dep}}+1)+1 N ♯ := N Tay ( N Tay + N dep + 1 ) + 1 moments from the datum, as at most N T a y + N d e p N_{\mathrm{Tay}}+N_{\mathrm{dep}} N Tay + N dep insertions, of Taylor degrees at most N T a y N_{\mathrm{Tay}} N Tay , precede a forcing of degree at most N T a y + 1 N_{\mathrm{Tay}}+1 N Tay + 1 . We use an energy with the moment weights w : = ( 1 + ∣ y − y 0 ∣ 2 / ρ ( m ) 2 ) N w w := (1+|y-y_0|^2/\rho(m)^2)^{N_w} w := ( 1 + ∣ y − y 0 ∣ 2 / ρ ( m ) 2 ) N w , about y 0 ′ y'_0 y 0 ′ in place of y 0 y_0 y 0 on the block of l + 1 l+1 l + 1 , as in Proposition 8.6, at moment orders N w N_w N w that decrease along the chain from N ♯ + 3 N_\sharp+3 N ♯ + 3 for the datum to 3 3 3 for the difference, as the Sobolev orders do. Each term is bounded through the previous ones, and the moments of the previous term absorb the polynomial growth of the frozen coefficients. The difference at t p r t_{\mathrm{pr}} t pr is therefore at most, in L ω , 3 2 ( R 2 ) L^2_{\omega,3}(\mathbb{R}^2) L ω , 3 2 ( R 2 ) about y 0 ′ y'_0 y 0 ′ , with g y 0 g_{y_0} g y 0 the piece and the norm (129) ,
C [ ( C ε m − 1 γ / 2 + δ ) N T a y + 1 + ( C ε m − 1 γ / 2 + 3 δ ) N T a y + 1 ] ∥ g y 0 ∥ H ρ m o l ( m ) , ω , N ♯ + 3 N b u f ( R 2 ) . C\left[\left(C\varepsilon_{m-1}^{\gamma/2+\delta}\right)^{N_{\mathrm{Tay}}+1}+\left(C\varepsilon_{m-1}^{\gamma/2+3\delta}\right)^{N_{\mathrm{Tay}}+1}\right]\lVert g_{y_0}\rVert_{H^{N_{\mathrm{buf}}}_{\rho_{\mathrm{mol}}(m),\omega,N_\sharp+3}(\mathbb{R}^2).} C [ ( C ε m − 1 γ /2 + δ ) N Tay + 1 + ( C ε m − 1 γ /2 + 3 δ ) N Tay + 1 ] ∥ g y 0 ∥ H ρ mol ( m ) , ω , N ♯ + 3 N buf ( R 2 ) .
The shears of the chart move labels by far more than ρ ( m ) \rho(m) ρ ( m ) over a block, but they enter the weighted energy only through their stream function ψ m \psi_m ψ m , which is small. For a solution D D D of the equation of the block with a forcing f f f ,
d d t ∫ R 2 w D 2 d y + 2 κ m ∫ R 2 w ∇ D ⋅ ∇ D d y = ∫ R 2 D 2 ∇ ⊥ ψ m ⋅ ∇ w d y + κ m ∫ R 2 D 2 Δ w d y − 2 κ m ∫ R 2 D ∇ w ⋅ ( K − I ) ∇ D d y + 2 ∫ R 2 w D f d y . \begin{aligned}
\frac{\mathrm{d}}{\mathrm{d}t}\int_{\mathbb{R}^2}wD^2\,\mathrm{d}y+2\kappa_m\int_{\mathbb{R}^2}w\nabla D\cdot\nabla D\,\mathrm{d}y
={}&\int_{\mathbb{R}^2}D^2\nabla^\perp\psi_m\cdot\nabla w\,\mathrm{d}y+\kappa_m\int_{\mathbb{R}^2}D^2\Delta w\,\mathrm{d}y\\
&-2\kappa_m\int_{\mathbb{R}^2}D\nabla w\cdot(\mathbf{K}-\mathbf{I})\nabla D\,\mathrm{d}y+2\int_{\mathbb{R}^2}wDf\,\mathrm{d}y.
\end{aligned} d t d ∫ R 2 w D 2 d y + 2 κ m ∫ R 2 w ∇ D ⋅ ∇ D d y = ∫ R 2 D 2 ∇ ⊥ ψ m ⋅ ∇ w d y + κ m ∫ R 2 D 2 Δ w d y − 2 κ m ∫ R 2 D ∇ w ⋅ ( K − I ) ∇ D d y + 2 ∫ R 2 w D f d y .
As in the proof of Lemma 12.13(a), the first term on the right is at most a quarter of the dissipation plus C ∣ ψ m ∣ 2 / ( κ m ρ ( m ) 2 ) ≤ C / h m C|\psi_m|^2/(\kappa_m\rho(m)^2)\le C/h_m C ∣ ψ m ∣ 2 / ( κ m ρ ( m ) 2 ) ≤ C / h m times the weighted energy, and the second at most C κ m / ρ ( m ) 2 ≤ C / h m C\kappa_m/\rho(m)^2\le C/h_m C κ m / ρ ( m ) 2 ≤ C / h m times it; as κ m ∣ ∇ w ∣ 2 / w ≤ C w / h m \kappa_m|\nabla w|^2/w\le Cw/h_m κ m ∣∇ w ∣ 2 / w ≤ C w / h m , the Cauchy–Schwarz inequality bounds the third by half of the dissipation plus C / h m C/h_m C / h m times the weighted energy. At the orders i i i of derivatives at the scale ρ m o l ( m ) \rho_{\mathrm{mol}}(m) ρ mol ( m ) , the sources add 2 ∫ ∇ w ⋅ ∇ i D ∇ i F d y 2\int\nabla w\cdot\nabla^iD\nabla^iF\,\mathrm{d}y 2 ∫ ∇ w ⋅ ∇ i D ∇ i F d y , with F F F the flux of the forcing of D D D , which Young’s inequality bounds by C κ m − 1 ∫ w ∣ ∇ i D ∣ 2 d y + κ m − 1 ∫ w ∣ ∇ i F ∣ 2 d y C\kappa_m^{-1}\int w|\nabla^iD|^2\,\mathrm{d}y+\kappa_m^{-1}\int w|\nabla^iF|^2\,\mathrm{d}y C κ m − 1 ∫ w ∣ ∇ i D ∣ 2 d y + κ m − 1 ∫ w ∣ ∇ i F ∣ 2 d y , the size of the forcing in Proposition 8.6. With condition (i) of Proposition 8.3 for ∇ ⊥ ψ m \nabla^\perp\psi_m ∇ ⊥ ψ m at the scale ρ m o l ( m ) \rho_{\mathrm{mol}}(m) ρ mol ( m ) , by the second line of (12.1), the energy runs on R 2 \mathbb{R}^2 R 2 with the weight w w w , and Grönwall’s inequality at the rate C / h m C/h_m C / h m multiplies it by at most a constant on each of the two blocks, the weights about y 0 y_0 y 0 and y 0 ′ y'_0 y 0 ′ being matched at t 2 t_2 t 2 as in the proof of Lemma 12.13(c). The degree N T a y N_{\mathrm{Tay}} N Tay is chosen so that, for m m m large, the difference is at most ε ε m Q + 4 ∥ g y 0 ∥ H ρ m o l ( m ) , ω N b u f ( R 2 ) \varepsilon\varepsilon_m^{Q+4}\lVert g_{y_0}\rVert_{H^{N_{\mathrm{buf}}}_{\rho_{\mathrm{mol}}(m),\omega}(\mathbb{R}^2)} ε ε m Q + 4 ∥ g y 0 ∥ H ρ mol ( m ) , ω N buf ( R 2 ) ; the norm of the display is at most a constant times this norm, as g y 0 g_{y_0} g y 0 is supported within ρ ( m ) / 4 \rho(m)/4 ρ ( m ) /4 of y 0 y_0 y 0 .
Step 4: The frozen terms. Each frozen patch evolution is a finite sum of Duhamel terms across the renewal at t 2 t_2 t 2 , of lengths at most N d e p + N T a y N_{\mathrm{dep}}+N_{\mathrm{Tay}} N dep + N Tay . Lemma 12.15 , with J : = N o u t J:=N_{\mathrm{out}} J := N out and N 1 : = 3 N_1:=3 N 1 := 3 , gives their models and errors C ε m Q + 3 C\varepsilon_m^{Q+3} C ε m Q + 3 , multiplied by the product of the strengths of the insertions, which for n s r c n_{\mathrm{src}} n src source insertions, by (b2) and (b3), and n c o n n_{\mathrm{con}} n con conductivity insertions is at most
( C C ∗ ε m δ / 4 ) n s r c ( C ε m 2 δ − 1 ρ ( m ) ε m − 1 ) n c o n . \left(C C_* \varepsilon_m^{\delta/4}\right)^{n_{\mathrm{src}}}\left(C\varepsilon_m^{2\delta-1}\frac{\rho(m)}{\varepsilon_{m-1}}\right)^{n_{\mathrm{con}}}. ( C C ∗ ε m δ /4 ) n src ( C ε m 2 δ − 1 ε m − 1 ρ ( m ) ) n con .
A piece of g ∈ H N b u f ( T 2 ) ∨ g\in H^{N_{\mathrm{buf}}}(\mathbb{T}^2)^{\vee} g ∈ H N buf ( T 2 ) ∨ has the regularity that the lemma asks, as the conductivity insertions have order 2: N 1 + d 1 + ⋯ − 2 ( N − + N + ) ≤ 3 + N d e p ( N s r c − 2 ) ≤ N b u f N_1+d_1+\cdots-2(N_-+N_+)\le3+N_{\mathrm{dep}}(N_{\mathrm{src}}-2)\le N_{\mathrm{buf}} N 1 + d 1 + ⋯ − 2 ( N − + N + ) ≤ 3 + N dep ( N src − 2 ) ≤ N buf .
Step 5: Sums. Let Π s u m g \Pi^{\mathrm{sum}}g Π sum g be the sum, over the pieces and over the lattice images, of the models of all these Duhamel terms, returned to T 2 \mathbb{T}^2 T 2 by the chart at t p r t_{\mathrm{pr}} t pr , and for the piece g y 0 g_{y_0} g y 0 centred at y 0 y_0 y 0 let Π y 0 s u m g \Pi^{\mathrm{sum}}_{y_0}g Π y 0 sum g be the sum of the models of its Duhamel terms. By (a5), the periodicity of K \mathbf{K} K and X m − 1 ( t , y + n , t ′ ) = X m − 1 ( t , y , t ′ ) + n X_{m-1}(t,y+n,t')=X_{m-1}(t,y,t')+n X m − 1 ( t , y + n , t ′ ) = X m − 1 ( t , y , t ′ ) + n for n ∈ Z 2 n\in\mathbb{Z}^2 n ∈ Z 2 , the model of a shifted piece is the shifted model; the chart at t p r t_{\mathrm{pr}} t pr changes the norm of H ρ ( m ) , ω N o u t H^{N_{\mathrm{out}}}_{\rho(m),\omega} H ρ ( m ) , ω N out by at most a constant, as in Step 2. For two centres y 0 y_0 y 0 and y ~ 0 \widetilde{y}_0 y 0 , let y 0 ′ y'_0 y 0 ′ and y ~ 0 ′ \widetilde{y}'_0 y 0 ′ be their labels in the chart of the block of l + 1 l+1 l + 1 , and r : = ∣ y 0 ′ − y ~ 0 ′ ∣ r:=|y'_0-\widetilde{y}'_0| r := ∣ y 0 ′ − y 0 ′ ∣ . By the moments of order N 1 = 3 N_1=3 N 1 = 3 about y 0 ′ y'_0 y 0 ′ in Lemma 12.15 and Peetre’s inequality, for j ≤ N o u t j\le N_{\mathrm{out}} j ≤ N out ,
∣ ⟨ ∇ j Π y 0 s u m g , ∇ j Π y ~ 0 s u m g ⟩ L ω 2 ( R 2 ) ∣ ≤ C ρ ( m ) − 2 j ( 1 + r ρ ( m ) ) − 3 ∥ g y 0 ∥ L ω 2 ( R 2 ) ∥ g y ~ 0 ∥ L ω 2 ( R 2 ) , \left|\left\langle\nabla^j\Pi^{\mathrm{sum}}_{y_0}g,\nabla^j\Pi^{\mathrm{sum}}_{\widetilde{y}_0}g\right\rangle_{L^2_\omega(\mathbb{R}^2)}\right|\le C\rho(m)^{-2j}\left(1+\frac{r}{\rho(m)}\right)^{-3}\lVert g_{y_0}\rVert_{L^2_\omega(\mathbb{R}^2)}\lVert g_{\widetilde{y}_0}\rVert_{L^2_\omega(\mathbb{R}^2)}, ⟨ ∇ j Π y 0 sum g , ∇ j Π y 0 sum g ⟩ L ω 2 ( R 2 ) ≤ C ρ ( m ) − 2 j ( 1 + ρ ( m ) r ) − 3 ∥ g y 0 ∥ L ω 2 ( R 2 ) ∥ g y 0 ∥ L ω 2 ( R 2 ) ,
and the matrix ( 1 + r / ρ ( m ) ) − 3 (1+r/\rho(m))^{-3} ( 1 + r / ρ ( m ) ) − 3 , indexed by the centres, has bounded row sums: the labels of the centres in the chart of the block of l + 1 l+1 l + 1 form a net of mesh comparable to ρ ( m ) / 8 \rho(m)/8 ρ ( m ) /8 in R 2 \mathbb{R}^2 R 2 , as the transition X m − 1 ( l h m , ⋅ , ( l + 1 ) h m ) X_{m-1}(lh_m,\cdot,(l+1)h_m) X m − 1 ( l h m , ⋅ , ( l + 1 ) h m ) has gradient within C ε m − 1 2 δ C\varepsilon_{m-1}^{2\delta} C ε m − 1 2 δ of the identity, by (54) . So Schur’s test in the patch index gives, by (36) ,
∥ Π s u m g ∥ H ρ ( m ) , ω N o u t ( T 2 ) ≤ C ∥ g ∥ L ω 2 ( T 2 ) . \left\|\Pi^{\mathrm{sum}}g\right\|_{H^{N_{\mathrm{out}}}_{\rho(m),\omega}(\mathbb{T}^2)}\le C\lVert g\rVert_{L^2_\omega(\mathbb{T}^2)}. ∥ Π sum g ∥ H ρ ( m ) , ω N out ( T 2 ) ≤ C ∥ g ∥ L ω 2 ( T 2 ) .
By density, Π s u m \Pi^{\mathrm{sum}} Π sum extends to a bounded linear map on L 2 ( T 2 ) ∨ L^2(\mathbb{T}^2)^{\vee} L 2 ( T 2 ) ∨ .
The errors add in the same way. For a piece g y 0 g_{y_0} g y 0 , let E y 0 E_{y_0} E y 0 be its error at t p r t_{\mathrm{pr}} t pr in the chart of the block of l + 1 l+1 l + 1 , the sum of the difference of Step 3 and of the errors of Step 4. By Step 3, and by Lemma 12.15 with N 1 = 3 N_1=3 N 1 = 3 and the strengths of Step 4, at most 1 for m m m large,
∥ E y 0 ∥ L ω , 3 2 ( R 2 ) ≤ C a y 0 , a y 0 : = ε m Q + 3 ∥ g y 0 ∥ H ρ m o l ( m ) , ω N b u f ( R 2 ) , \lVert E_{y_0}\rVert_{L^2_{\omega,3}(\mathbb{R}^2)}\le Ca_{y_0},\qquad a_{y_0}:=\varepsilon_m^{Q+3}\lVert g_{y_0}\rVert_{H^{N_{\mathrm{buf}}}_{\rho_{\mathrm{mol}}(m),\omega}(\mathbb{R}^2),} ∥ E y 0 ∥ L ω , 3 2 ( R 2 ) ≤ C a y 0 , a y 0 := ε m Q + 3 ∥ g y 0 ∥ H ρ mol ( m ) , ω N buf ( R 2 ) ,
with the moments about y 0 ′ y'_0 y 0 ′ . By the moments, (a5), the periodicity of the coefficients and the uniqueness of the energy solutions on T 2 \mathbb{T}^2 T 2 , the lattice sum over the pieces of the terms on R 2 \mathbb{R}^2 R 2 is ⋃ j = 0 N d e p U { j } \bigcup_{j=0}^{N_{\mathrm{dep}}}U^{\{j\}} ⋃ j = 0 N dep U { j } , for each j j j , and the charts preserve area. So ∑ j = 0 N d e p U { j } ( t p r ) − Π s u m g \sum_{j=0}^{N_{\mathrm{dep}}}U^{\{j\}}(t_{\mathrm{pr}})-\Pi^{\mathrm{sum}}g ∑ j = 0 N dep U { j } ( t pr ) − Π sum g is the sum of the errors E y 0 E_{y_0} E y 0 over the pieces and their lattice images. At each y y y , one of ∣ y − y 0 ′ ∣ |y-y'_0| ∣ y − y 0 ′ ∣ and ∣ y − y ~ 0 ′ ∣ |y-\widetilde{y}'_0| ∣ y − y 0 ′ ∣ is at least r / 2 r/2 r /2 , so ∣ ⟨ E y 0 , E y ~ 0 ⟩ L ω 2 ( R 2 ) ∣ ≤ C ( 1 + r / ρ ( m ) ) − 3 a y 0 a y ~ 0 \left|\langle E_{y_0},E_{\widetilde{y}_0}\rangle_{L^2_\omega(\mathbb{R}^2)}\right|\le C(1+r/\rho(m))^{-3}a_{y_0}a_{\widetilde{y}_0} ⟨ E y 0 , E y 0 ⟩ L ω 2 ( R 2 ) ≤ C ( 1 + r / ρ ( m ) ) − 3 a y 0 a y 0 , and Schur’s test over the centres and their lattice images, with Step 2, gives
∥ ∑ j = 0 N d e p U { j } ( t p r ) − Π s u m g ∥ L ω 2 ( T 2 ) 2 ≤ C ∑ y 0 a y 0 2 ≤ C ε m − 1 2 ( Q + 3 ) ∥ g ∥ H ρ m o l ( m ) , ω N b u f ( T 2 ) 2 . \left\|\sum_{j=0}^{N_{\mathrm{dep}}}U^{\{j\}}(t_{\mathrm{pr}})-\Pi^{\mathrm{sum}}g\right\|_{L^2_\omega(\mathbb{T}^2)}^2\le C\sum_{y_0}a_{y_0}^2\le C\varepsilon_{m-1}^{2(Q+3)}\lVert g\rVert_{H^{N_{\mathrm{buf}}}_{\rho_{\mathrm{mol}}(m),\omega}(\mathbb{T}^2)}^2. j = 0 ∑ N dep U { j } ( t pr ) − Π sum g L ω 2 ( T 2 ) 2 ≤ C y 0 ∑ a y 0 2 ≤ C ε m − 1 2 ( Q + 3 ) ∥ g ∥ H ρ mol ( m ) , ω N buf ( T 2 ) 2 .
Step 6: Constants. A field with constant components solves the system without smoothing, whose sources carry at least one derivative, so its Duhamel expansion is the field itself in U { 0 } U^{\{0\}} U { 0 } and zero in U { j } U^{\{j\}} U { j } , j ≥ 1 j \ge1 j ≥ 1 . With ( g ) T 2 (g)_{\mathbb{T}^2} ( g ) T 2 the field of the means of the components, Steps 3–5 therefore give the first bound below, and we define Π ~ \widetilde{\Pi} Π by the formula on the right:
∥ Π s u m [ ( g ) T 2 ] − ( g ) T 2 ∥ L ω 2 ( T 2 ) ≤ C ε m Q + 3 ∥ g ∥ L ω 2 ( T 2 ) , Π ~ g : = Π s u m g − Π s u m [ ( g ) T 2 ] + ( g ) T 2 . \left\|\Pi^{\mathrm{sum}}[(g)_{\mathbb{T}^2}]-(g)_{\mathbb{T}^2}\right\|_{L^2_\omega(\mathbb{T}^2)} \le C\varepsilon_m^{Q+3}\|g\|_{L^2_\omega(\mathbb{T}^2)}, \qquad\widetilde{\Pi}g := \Pi^{\mathrm{sum}}g-\Pi^{\mathrm{sum}}[(g)_{\mathbb{T}^2}]+(g)_{\mathbb{T}^2}. ∥ Π sum [( g ) T 2 ] − ( g ) T 2 ∥ L ω 2 ( T 2 ) ≤ C ε m Q + 3 ∥ g ∥ L ω 2 ( T 2 ) , Π g := Π sum g − Π sum [( g ) T 2 ] + ( g ) T 2 .
The map Π ~ \widetilde{\Pi} Π fixes the fields with constant components and satisfies both bounds of the proposition, since ∥ ( g ) T 2 ∥ L ω 2 ( T 2 ) ≤ ∥ g ∥ L ω 2 ( T 2 ) \|(g)_{\mathbb{T}^2}\|_{L^2_\omega(\mathbb{T}^2)} \le\|g\|_{L^2_\omega(\mathbb{T}^2)} ∥ ( g ) T 2 ∥ L ω 2 ( T 2 ) ≤ ∥ g ∥ L ω 2 ( T 2 ) .
Both choices of the masses. The construction of t p r t_{\mathrm{pr}} t pr and Π ~ \widetilde{\Pi} Π does not use the masses, and the hypotheses hold for both choices (Definition 11.1), so one Π ~ \widetilde{\Pi} Π serves both.
Remark 12.16 (The terminal level). Propositions 12.4–12.7 and Lemmas 12.8–12.15 hold at the terminal level m = M m=M m = M ; § 6 uses them, with Lemma 15.8 in place of Lemma 12.17, which is stated for m < M m<M m < M .
Proof of Proposition 5.3(a) and of Proposition 12.1
The Duhamel expansion applies the sources without their smoothing operators, and each source it applies takes derivatives of the datum. Lemma 12.17 supplies them, at the molecular scale; the truncated expansion is within C ε m − 1 Q + 3 C\varepsilon_{m-1}^{Q+3} C ε m − 1 Q + 3 of the solution (Step 4 below), and Proposition 12.7 prepares it.
Lemma 12.17 (Molecular smoothing). For every integer 0 ≤ N ≤ N c o e f 0 \le N \le N_{\mathrm{coef}} 0 ≤ N ≤ N coef there exists a constant C ( N ) C(N) C ( N ) , depending only on N N N , such that the following holds. Let κ ∈ K M \kappa\in\mathcal{K}_M κ ∈ K M and m 0 < m < M m_0<m<M m 0 < m < M , let G G G be an admissible family of level m m m on [ t A , t f ] [t_A,t_f] [ t A , t f ] , and let t c ∈ [ t A , t f − ρ m o l ( m ) 2 / κ m ] t_c \in[t_A,t_f-\rho_{\mathrm{mol}}(m)^2/\kappa_m] t c ∈ [ t A , t f − ρ mol ( m ) 2 / κ m ] . For both choices ω v \omega_v ω v and ω v / Ψ v 2 \omega_v/\Psi_v^2 ω v / Ψ v 2 of the masses and for every datum g g g , the solution U U U of G G G from g g g at t c t_c t c satisfies
∥ U ∥ L ∞ ( [ t c + ρ m o l ( m ) 2 / κ m , t f ] ; H ρ m o l ( m ) , ω N ( T 2 ) ) ≤ C ( N ) ∥ g ∥ L ω 2 ( T 2 ) . \|U\|_{L^\infty([t_c+\rho_{\mathrm{mol}}(m)^2/\kappa_m,t_f];H^N_{\rho_{\mathrm{mol}}(m),\omega}(\mathbb{T}^2))} \le C(N)\|g\|_{L^2_\omega(\mathbb{T}^2)}. ∥ U ∥ L ∞ ([ t c + ρ mol ( m ) 2 / κ m , t f ] ; H ρ mol ( m ) , ω N ( T 2 )) ≤ C ( N ) ∥ g ∥ L ω 2 ( T 2 ) .
Proof. Proposition 8.3 applies at every order j ≤ N j \le N j ≤ N at the scale ρ m o l ( m ) \rho_{\mathrm{mol}}(m) ρ mol ( m ) , with its diffusivity equal to κ m \kappa_m κ m : its hypothesis (i) is (12.1); (ii) follows from (b1), (b2) of Definition 11.1 and (7.9); and (iii) is its (d), with sums at most $\frac{1}{8}$ for $m>m_0$ (Proposition 11.5`); by its part (b), ∥ U ( t ) ∥ H ρ m o l ( m ) , ω j ( T 2 ) \|U(t)\|_{H^j_{\rho_{\mathrm{mol}}(m),\omega}(\mathbb{T}^2)} ∥ U ( t ) ∥ H ρ mol ( m ) , ω j ( T 2 ) does not increase in t t t . The ( t − t c ) j (t-t_c)^j ( t − t c ) j -weighted energy estimate, by induction on j j j with
∥ U ∥ H ρ m o l ( m ) , ω j ( T 2 ) 2 ≤ ∥ U ∥ L ω 2 ( T 2 ) 2 + ρ m o l ( m ) 2 ∥ ∇ U ∥ H ρ m o l ( m ) , ω j − 1 ( T 2 ) 2 \|U\|_{H^j_{\rho_{\mathrm{mol}}(m),\omega}(\mathbb{T}^2)}^2 \le\|U\|_{L^2_\omega(\mathbb{T}^2)}^2+\rho_{\mathrm{mol}}(m)^2\|\nabla U\|_{H^{j-1}_{\rho_{\mathrm{mol}}(m),\omega}(\mathbb{T}^2)}^2 ∥ U ∥ H ρ mol ( m ) , ω j ( T 2 ) 2 ≤ ∥ U ∥ L ω 2 ( T 2 ) 2 + ρ mol ( m ) 2 ∥∇ U ∥ H ρ mol ( m ) , ω j − 1 ( T 2 ) 2
(by (36) ), gives for 0 < t − t c ≤ ρ m o l ( m ) 2 / κ m 0<t-t_c\le\rho_{\mathrm{mol}}(m)^2/\kappa_m 0 < t − t c ≤ ρ mol ( m ) 2 / κ m
( t − t c ) j ∥ U ( t ) ∥ H ρ m o l ( m ) , ω j ( T 2 ) 2 + κ m ∫ t c t ( t ′ − t c ) j ∥ ∇ U ( t ′ ) ∥ H ρ m o l ( m ) , ω j ( T 2 ) 2 d t ′ ≤ ( j + 1 ) ! ( ρ m o l ( m ) 2 κ m ) j ∥ g ∥ L ω 2 ( T 2 ) 2 , (t-t_c)^j\|U(t)\|_{H^j_{\rho_{\mathrm{mol}}(m),\omega}(\mathbb{T}^2)}^2+\kappa_m\int_{t_c}^{t}(t'-t_c)^j\|\nabla U(t')\|_{H^j_{\rho_{\mathrm{mol}}(m),\omega}(\mathbb{T}^2)}^2\,\mathrm{d}t'
\le(j+1)!\left(\frac{\rho_{\mathrm{mol}}(m)^2}{\kappa_m}\right)^j\|g\|_{L^2_\omega(\mathbb{T}^2)}^2, ( t − t c ) j ∥ U ( t ) ∥ H ρ mol ( m ) , ω j ( T 2 ) 2 + κ m ∫ t c t ( t ′ − t c ) j ∥∇ U ( t ′ ) ∥ H ρ mol ( m ) , ω j ( T 2 ) 2 d t ′ ≤ ( j + 1 )! ( κ m ρ mol ( m ) 2 ) j ∥ g ∥ L ω 2 ( T 2 ) 2 ,
as 1 + j ⋅ j ! ≤ ( j + 1 ) ! 1+j\cdot j! \le(j+1)! 1 + j ⋅ j ! ≤ ( j + 1 )! : the weight vanishes at t c t_c t c , whatever the derivatives of the datum. At t = t c + ρ m o l ( m ) 2 / κ m t=t_c+\rho_{\mathrm{mol}}(m)^2/\kappa_m t = t c + ρ mol ( m ) 2 / κ m this is the bound, with C ( N ) : = ( ( N + 1 ) ! ) 1 / 2 C(N):=((N+1)!)^{1/2} C ( N ) := (( N + 1 )! ) 1/2 , which persists. □ \square □
Proof of Proposition 5.3(a). The top level m 0 m_0 m 0 is at least the levels of Propositions 12.5–12.7 and Lemmas 12.8–12.15, and so large that ρ m o l ( m ) 2 / κ m ≤ C ε m q ( 2 − β ) ≤ 2 h m \rho_{\mathrm{mol}}(m)^2/\kappa_m \le C\varepsilon_m^{q(2-\beta)} \le2h_m ρ mol ( m ) 2 / κ m ≤ C ε m q ( 2 − β ) ≤ 2 h m , by (12.1), Lemma 7.1(a), (5) and (7) . Fix g g g , let U U U be the solution of G G G from g g g at t c t_c t c , and put t 1 : = t c + ρ m o l ( m ) 2 / κ m t_1:=t_c+\rho_{\mathrm{mol}}(m)^2/\kappa_m t 1 := t c + ρ mol ( m ) 2 / κ m . Then t 1 ≤ t c + 2 h m ≤ t f − 2 h m t_1\le t_c+2h_m\le t_f-2h_m t 1 ≤ t c + 2 h m ≤ t f − 2 h m , as t c + 4 h m ≤ t f t_c+4h_m\le t_f t c + 4 h m ≤ t f (Proposition 5.3).
Step 1: Smoothing. By Lemma 12.17 with N = N b u f ≤ N c o e f N=N_{\mathrm{buf}}\le N_{\mathrm{coef}} N = N buf ≤ N coef ,
∥ U ( t 1 ) ∥ H ρ m o l ( m ) , ω N b u f ( T 2 ) ≤ C ∥ g ∥ L ω 2 ( T 2 ) . \lVert U(t_1)\rVert_{H^{N_{\mathrm{buf}}}_{\rho_{\mathrm{mol}}(m),\omega}(\mathbb{T}^2)}\le C\lVert g\rVert_{L^2_\omega(\mathbb{T}^2)}. ∥ U ( t 1 ) ∥ H ρ mol ( m ) , ω N buf ( T 2 ) ≤ C ∥ g ∥ L ω 2 ( T 2 ) .
Step 2: The remainder of the expansion. Let U { j } U^{\{j\}} U { j } be the Duhamel expansion of G G G from U ( t 1 ) U(t_1) U ( t 1 ) at t 1 t_1 t 1 . Applying the operator of level m m m to each term, the remainder U − ∑ j = 0 N d e p U { j } U-\sum_{j=0}^{N_{\mathrm{dep}}}U^{\{j\}} U − ∑ j = 0 N dep U { j } solves on [ t 1 , t f ] [t_1,t_f] [ t 1 , t f ] the system (66) of G G G , with its smoothings, from the datum 0 at t 1 t_1 t 1 , with the flux forcing F : = F ( 1 ) + F ( 2 ) F:=F^{(1)}+F^{(2)} F := F ( 1 ) + F ( 2 ) ,
F ν ( 1 ) : = ∑ ν ′ , σ , d A σ , d ∇ d − 1 ( S σ U ν ′ { N d e p } ) , F ν ( 2 ) : = ∑ ν ′ , σ , d A σ , d ∇ d − 1 ( ( S σ − I ) ∑ j < N d e p U ν ′ { j } ) : F^{(1)}_\nu:=\sum_{\nu',\sigma,d}A_{\sigma,d}\nabla^{d-1}\left(S_\sigma U^{\{N_{\mathrm{dep}}\}}_{\nu'}\right),\qquad F^{(2)}_\nu:=\sum_{\nu',\sigma,d}A_{\sigma,d}\nabla^{d-1}\left((S_\sigma-I)\sum_{j<N_{\mathrm{dep}}}U^{\{j\}}_{\nu'}\right): F ν ( 1 ) := ν ′ , σ , d ∑ A σ , d ∇ d − 1 ( S σ U ν ′ { N dep } ) , F ν ( 2 ) := ν ′ , σ , d ∑ A σ , d ∇ d − 1 ( S σ − I ) j < N dep ∑ U ν ′ { j } :
the last term of the expansion fed back through the sources, and the error of applying the sources without their smoothing. By Proposition 8.3(a) at the order 0, whose (iii) is clause (d) of Definition 11.1,
∥ U − ∑ j = 0 N d e p U { j } ∥ L ∞ ( [ t 1 , t f ] ; L ω 2 ( T 2 ) ) ≤ ( 8 κ m ) 1 / 2 ∥ F ∥ L 2 ( ( t 1 , t f ) ; L ω 2 ( T 2 ) ) . \left\lVert U-\sum_{j=0}^{N_{\mathrm{dep}}}U^{\{j\}}\right\rVert_{L^\infty([t_1,t_f];L^2_\omega(\mathbb{T}^2))}\le\left(\frac{8}{\kappa_m}\right)^{1/2}\lVert F\rVert_{L^2((t_1,t_f);L^2_\omega(\mathbb{T}^2))}. U − j = 0 ∑ N dep U { j } L ∞ ([ t 1 , t f ] ; L ω 2 ( T 2 )) ≤ ( κ m 8 ) 1/2 ∥ F ∥ L 2 (( t 1 , t f ) ; L ω 2 ( T 2 )) .
Step 3: The terms of the expansion. The coefficients of the sources are summed by Schur’s test: by (b2′) and the transport to x x x (as in Step 8 of the proof of Lemma 11.7, at the cost of a constant), for every N ≤ N b u f N\le N_{\mathrm{buf}} N ≤ N buf the matrix with entries
( ω ν ω ν ′ ) 1 / 2 ∑ σ : ν ′ → ν ∑ d = 2 N s r c ( N + d − 2 ) ! N ! κ m ρ m o l ( m ) d − 2 ∑ i ≤ N ρ m o l ( m ) i i ! ∥ ∇ i A σ , d ∥ L ∞ ( ( t 1 , t f ) × T 2 ) \left(\frac{\omega_\nu}{\omega_\nu'}\right)^{1/2}\sum_{\sigma:\,\nu'\to\nu}\sum_{d=2}^{N_{\mathrm{src}}}\frac{(N+d-2)!}{N!\,\kappa_m\rho_{\mathrm{mol}}(m)^{d-2}}\sum_{i\le N}\frac{\rho_{\mathrm{mol}}(m)^i}{i!}\lVert\nabla^iA_{\sigma,d}\rVert_{L^\infty((t_1,t_f)\times\mathbb{T}^2)} ( ω ν ′ ω ν ) 1/2 σ : ν ′ → ν ∑ d = 2 ∑ N src N ! κ m ρ mol ( m ) d − 2 ( N + d − 2 )! i ≤ N ∑ i ! ρ mol ( m ) i ∥ ∇ i A σ , d ∥ L ∞ (( t 1 , t f ) × T 2 )
is at most C C C times the majorant matrix, so its row and column sums are at most C ε m δ / 4 C\varepsilon_m^{\delta/4} C ε m δ /4 , by (b3). Proposition 8.3(a) at the scale ρ m o l ( m ) \rho_{\mathrm{mol}}(m) ρ mol ( m ) and the order N b u f − j ( N s r c − 2 ) N_{\mathrm{buf}}-j(N_{\mathrm{src}}-2) N buf − j ( N src − 2 ) , with (12.1) for its (i) and with Lemma 8.1 and these sums for the forcing as in the proof of Proposition 8.6, gives by induction on j j j , with a constant C C C ,
κ m 1 / 2 ∥ ∇ U { j } ∥ L 2 ( ( t 1 , t f ) ; H ρ m o l ( m ) , ω N b u f − j ( N s r c − 2 ) ( T 2 ) ) ≤ C j + 1 ε m j δ / 4 ∥ U ( t 1 ) ∥ H ρ m o l ( m ) , ω N b u f ( T 2 ) . \kappa_m^{1/2}\lVert\nabla U^{\{j\}}\rVert_{L^2((t_1,t_f);H^{N_{\mathrm{buf}}-j(N_{\mathrm{src}}-2)}_{\rho_{\mathrm{mol}}(m),\omega}(\mathbb{T}^2))}\le C^{j+1}\varepsilon_m^{j\delta/4}\lVert U(t_1)\rVert_{H^{N_{\mathrm{buf}}}_{\rho_{\mathrm{mol}}(m),\omega}(\mathbb{T}^2)}. κ m 1/2 ∥ ∇ U { j } ∥ L 2 (( t 1 , t f ) ; H ρ mol ( m ) , ω N buf − j ( N src − 2 ) ( T 2 )) ≤ C j + 1 ε m j δ /4 ∥ U ( t 1 ) ∥ H ρ mol ( m ) , ω N buf ( T 2 ) .
Step 4: The forcing. We use the order and width conditions, with ℓ \ell ℓ the order of the kernels W ℓ W_\ell W ℓ of S σ S_\sigma S σ in (c1), which depends on σ \sigma σ ,
N m o m ≤ 2 ℓ − 1 , N b u f − j ( N s r c − 2 ) ≥ N m o m + N s r c − 2 ( j < N d e p ) , ( r σ ρ m o l ( m ) ) N m o m ≤ C ε m − 1 q δ N m o m , N_{\mathrm{mom}}\le2\ell-1,\qquad N_{\mathrm{buf}}-j(N_{\mathrm{src}}-2)\ge N_{\mathrm{mom}}+N_{\mathrm{src}}-2\ (j<N_{\mathrm{dep}}),\qquad\left(\frac{r_\sigma}{\rho_{\mathrm{mol}}(m)}\right)^{N_{\mathrm{mom}}}\le C\varepsilon_{m-1}^{q\delta N_{\mathrm{mom}}}, N mom ≤ 2 ℓ − 1 , N buf − j ( N src − 2 ) ≥ N mom + N src − 2 ( j < N dep ) , ( ρ mol ( m ) r σ ) N mom ≤ C ε m − 1 q δ N mom ,
the first by (c1), the second by the definition (123) of N b u f N_{\mathrm{buf}} N buf , the third by (c2) and ε m ≤ 2 ε m − 1 q \varepsilon_m\le2\varepsilon_{m-1}^q ε m ≤ 2 ε m − 1 q . In L 2 ( ( t 1 , t f ) ; L ω 2 ( T 2 ) ) L^2((t_1,t_f);L^2_\omega(\mathbb{T}^2)) L 2 (( t 1 , t f ) ; L ω 2 ( T 2 )) , the two parts of the forcing satisfy
∥ F ( 1 ) ∥ ≤ C a d m ε m − 1 q δ / 4 κ m ∥ ∇ U { N d e p } ∥ L 2 ( ( t 1 , t f ) ; L ω 2 ( T 2 ) ) \lVert F^{(1)}\rVert\le C_{\mathrm{adm}}\varepsilon_{m-1}^{q\delta/4}\kappa_m\lVert\nabla U^{\{N_{\mathrm{dep}}\}}\rVert_{L^2((t_1,t_f);L^2_\omega(\mathbb{T}^2))} ∥ F ( 1 ) ∥ ≤ C adm ε m − 1 q δ /4 κ m ∥ ∇ U { N dep } ∥ L 2 (( t 1 , t f ) ; L ω 2 ( T 2 ))
(Lemma 8.4 and (d)),
∥ F ( 2 ) ∥ ≤ C κ m 1 / 2 ε m − 1 q δ N m o m ∥ U ( t 1 ) ∥ H ρ m o l ( m ) , ω N b u f ( T 2 ) \lVert F^{(2)}\rVert\le C\kappa_m^{1/2}\varepsilon_{m-1}^{q\delta N_{\mathrm{mom}}}\lVert U(t_1)\rVert_{H^{N_{\mathrm{buf}}}_{\rho_{\mathrm{mol}}(m),\omega}(\mathbb{T}^2)} ∥ F ( 2 ) ∥ ≤ C κ m 1/2 ε m − 1 q δ N mom ∥ U ( t 1 ) ∥ H ρ mol ( m ) , ω N buf ( T 2 )
(Proposition 6.1, (b2′), Step 3); in the first line, the sums C a d m ε m − 1 q δ / 4 C_{\mathrm{adm}}\varepsilon_{m-1}^{q\delta/4} C adm ε m − 1 q δ /4 of (d) replace the 1 8 \frac{1}{8} 8 1 of Lemma 8.4; in the second line, the third bound of Proposition 6.1, which holds for every r , ρ > 0 r,\rho>0 r , ρ > 0 and ∣ Ω i ∣ ∞ ≤ C Ω \lvert\Omega_i\rvert_\infty\le C_\Omega ∣ Ω i ∣ ∞ ≤ C Ω , with a constant independent of ℓ \ell ℓ as N m o m ≤ 2 ℓ − 1 N_{\mathrm{mom}}\le2\ell-1 N mom ≤ 2 ℓ − 1 , at the scale ρ m o l ( m ) \rho_{\mathrm{mol}}(m) ρ mol ( m ) with N 1 = N m o m N_1=N_{\mathrm{mom}} N 1 = N mom , bounds ( I − S σ ) ∇ d − 1 U { j } (I-S_\sigma)\nabla^{d-1}U^{\{j\}} ( I − S σ ) ∇ d − 1 U { j } by C ( r σ / ρ m o l ( m ) ) N m o m ∥ ∇ d − 1 U { j } ∥ H ρ ( m ) N m o m ( T 2 ) C(r_{\sigma}/\rho_{\mathrm{mol}}(m))^{N_{\mathrm{mom}}}\lVert\nabla^{d-1}U^{\{j\}}\rVert_{H_{\rho(m)}^{N_{\mathrm{mom}}}(\mathbb{T}^{2})} C ( r σ / ρ mol ( m ) ) N mom ∥ ∇ d − 1 U { j } ∥ H ρ ( m ) N mom ( T 2 ) , and the coefficients A σ , d A_{\sigma,d} A σ , d are summed by the matrix of Step 3 at N = 0 N=0 N = 0 . By Steps 2 and 3, ε m δ / 4 ≤ 2 ε m − 1 q δ / 4 \varepsilon_m^{\delta/4}\le2\varepsilon_{m-1}^{q\delta/4} ε m δ /4 ≤ 2 ε m − 1 q δ /4 , the choice of N d e p N_{\mathrm{dep}} N dep , (123) and Step 1,
∥ U − ∑ j = 0 N d e p U { j } ∥ L ∞ ( [ t 1 , t f ] ; L ω 2 ( T 2 ) ) ≤ C ( ε m − 1 q δ ( N d e p + 1 ) / 4 + ε m − 1 q δ N m o m ) ∥ g ∥ L ω 2 ( T 2 ) ≤ C ε m − 1 Q + 3 ∥ g ∥ L ω 2 ( T 2 ) . \left\lVert U-\sum_{j=0}^{N_{\mathrm{dep}}}U^{\{j\}}\right\rVert_{L^{\infty}([t_{1},t_{\mathrm{f}}];L_{\omega}^{2}(\mathbb{T}^{2}))}\le C\left(\varepsilon_{m-1}^{q\delta(N_{\mathrm{dep}}+1)/4}+\varepsilon_{m-1}^{q\delta N_{\mathrm{mom}}}\right)\lVert g\rVert_{L_{\omega}^{2}(\mathbb{T}^{2})}\le C\varepsilon_{m-1}^{Q+3}\lVert g\rVert_{L_{\omega}^{2}(\mathbb{T}^{2})}. U − j = 0 ∑ N dep U { j } L ∞ ([ t 1 , t f ] ; L ω 2 ( T 2 )) ≤ C ( ε m − 1 q δ ( N dep + 1 ) /4 + ε m − 1 q δ N mom ) ∥ g ∥ L ω 2 ( T 2 ) ≤ C ε m − 1 Q + 3 ∥ g ∥ L ω 2 ( T 2 ) .
and the same holds with the masses ω v / Ψ v 2 \omega_{v}/\Psi_{v}^{2} ω v / Ψ v 2 , for which (b2′), (b3) and (d) also hold.
Step 5: The preparation. Proposition 12.7, with the constants of Definition 11.1 (Proposition 11.5) and with its path hypothesis (an admissible family has paths of at most N p a t h N_{\mathrm{path}} N path sources, item (1) of Subsection 11.1), gives t p r ∈ ( t 1 , t 1 + 2 h m ] ⊂ [ t c , t c + 4 h m ] t_{\mathrm{pr}}\in(t_{1},t_{1}+2h_{m}]\subset[t_{\mathrm{c}},t_{\mathrm{c}}+4h_{m}] t pr ∈ ( t 1 , t 1 + 2 h m ] ⊂ [ t c , t c + 4 h m ] and Π ~ \widetilde{\Pi} Π ; we put Π g : = Π ~ U ( t 1 ) \Pi g:=\widetilde{\Pi}U(t_{1}) Π g := Π U ( t 1 ) , which is linear in g g g . Step 4, Proposition 12.7 and Step 1, for the masses ω v / Ψ v 2 \omega_{v}/\Psi_{v}^{2} ω v / Ψ v 2 , give the first bound of (a),
∥ Π g − U ( t p r ) ∥ L G 2 ( T 2 ) ≤ C ε m − 1 Q + 3 ∥ g ∥ L G 2 ( T 2 ) . \lVert\Pi g-U(t_{\mathrm{pr}})\rVert_{L_{G}^{2}(\mathbb{T}^{2})}\le C\varepsilon_{m-1}^{Q+3}\lVert g\rVert_{L_{G}^{2}(\mathbb{T}^{2})}. ∥ Π g − U ( t pr ) ∥ L G 2 ( T 2 ) ≤ C ε m − 1 Q + 3 ∥ g ∥ L G 2 ( T 2 ) .
Step 6: The other two bounds. The solution U U U does not grow in L G 2 ( T 2 ) L_{G}^{2}(\mathbb{T}^{2}) L G 2 ( T 2 ) , by Proposition 6.2 for the masses ω v / Ψ v 2 \omega_{v}/\Psi_{v}^{2} ω v / Ψ v 2 , whose coupling sums are at most 1 8 \frac{1}{8} 8 1 by (d); the near-unit norm of Π \Pi Π comes from this energy, not from the model. With Step 5 and the first bound of Proposition 12.7 for these weights,
∥ Π g ∥ L G 2 ( T 2 ) ≤ ∥ U ( t p r ) ∥ L G 2 ( T 2 ) + C ε m − 1 Q ∥ g ∥ L G 2 ( T 2 ) ≤ ( 1 + C ε m − 1 Q ) ∥ g ∥ L G 2 ( T 2 ) , \lVert\Pi g\rVert_{L_{G}^{2}(\mathbb{T}^{2})}\le\lVert U(t_{\mathrm{pr}})\rVert_{L_{G}^{2}(\mathbb{T}^{2})}+C\varepsilon_{m-1}^{Q}\lVert g\rVert_{L_{G}^{2}(\mathbb{T}^{2})}\le(1+C\varepsilon_{m-1}^{Q})\lVert g\rVert_{L_{G}^{2}(\mathbb{T}^{2})}, ∥ Π g ∥ L G 2 ( T 2 ) ≤ ∥ U ( t pr ) ∥ L G 2 ( T 2 ) + C ε m − 1 Q ∥ g ∥ L G 2 ( T 2 ) ≤ ( 1 + C ε m − 1 Q ) ∥ g ∥ L G 2 ( T 2 ) ,
∥ Π g ∥ H ρ ( m ) , G N o u t ( T 2 ) ≤ C ∥ U ( t 1 ) ∥ L G 2 ( T 2 ) ≤ C ∥ g ∥ L G 2 ( T 2 ) . \lVert\Pi g\rVert_{H_{\rho(m),G}^{N_{\mathrm{out}}}(\mathbb{T}^{2})}\le C\lVert U(t_{1})\rVert_{L_{G}^{2}(\mathbb{T}^{2})}\le C\lVert g\rVert_{L_{G}^{2}(\mathbb{T}^{2})}. ∥ Π g ∥ H ρ ( m ) , G N out ( T 2 ) ≤ C ∥ U ( t 1 ) ∥ L G 2 ( T 2 ) ≤ C ∥ g ∥ L G 2 ( T 2 ) .
Proof of Proposition 12.1. We take t p r t_{\mathrm{pr}} t pr and Π \Pi Π as just constructed. (i) t p r > t 1 = t c + ρ m o l ( m ) 2 / κ m t_{\mathrm{pr}}>t_{1}=t_{\mathrm{c}}+\rho_{\mathrm{mol}}(m)^{2}/\kappa_{m} t pr > t 1 = t c + ρ mol ( m ) 2 / κ m , and the margin is that of Proposition 12.7. (ii) A field with constant components is a solution of G G G at every time, so U ( t 1 ) = g U(t_{1})=g U ( t 1 ) = g , and Π ~ \widetilde{\Pi} Π fixes it. (iii) Lemma 12.17 with N = 2 N r e p + 4 ≤ N r e t ≤ N c o e f N=2N_{\mathrm{rep}}+4\le N_{\mathrm{ret}}\le N_{\mathrm{coef}} N = 2 N rep + 4 ≤ N ret ≤ N coef , as t p r ≥ t c + ρ m o l ( m ) 2 / κ m t_{\mathrm{pr}}\ge t_{\mathrm{c}}+\rho_{\mathrm{mol}}(m)^{2}/\kappa_{m} t pr ≥ t c + ρ mol ( m ) 2 / κ m , and ∥ g ∥ L ω 2 ( T 2 ) ≤ ∥ g ∥ L G 2 ( T 2 ) \lVert g\rVert_{L_{\omega}^{2}(\mathbb{T}^{2})}\le\lVert g\rVert_{L_{G}^{2}(\mathbb{T}^{2})} ∥ g ∥ L ω 2 ( T 2 ) ≤ ∥ g ∥ L G 2 ( T 2 ) , as Ψ v ≥ 1 \Psi_{v}\ge1 Ψ v ≥ 1 . (iv) We expand exp p ( − Γ m ( t p r ) : ∇ 2 ) \exp_{p}(-\Gamma_{m}(t_{\mathrm{pr}}):\nabla^{2}) exp p ( − Γ m ( t pr ) : ∇ 2 ) : by Lemma 8.1(b) each application of Γ m ( t p r ) : ∇ 2 \Gamma_{m}(t_{\mathrm{pr}}):\nabla^{2} Γ m ( t pr ) : ∇ 2 costs two orders and at most C ( N r e t , p ) ∣ Γ m ( t p r ) ∣ / ρ ( m ) 2 ≤ C ε m − 1 δ C(N_{\mathrm{ret}},p)|\Gamma_{m}(t_{\mathrm{pr}})|/\rho(m)^{2}\le C\varepsilon_{m-1}^{\delta} C ( N ret , p ) ∣ Γ m ( t pr ) ∣/ ρ ( m ) 2 ≤ C ε m − 1 δ in H ρ ( m ) , G N r e t ( T 2 ) H_{\rho(m),G}^{N_{\mathrm{ret}}}(\mathbb{T}^{2}) H ρ ( m ) , G N ret ( T 2 ) (Lemma 10.3(b)), and N r e t + 2 p ≤ N o u t N_{\mathrm{ret}}+2p\le N_{\mathrm{out}} N ret + 2 p ≤ N out ; the third bound of Proposition 5.3(a) concludes.
Two-scale expansion and the homogenization error
In this section we prove Proposition 5.3(b), the comparison of the solution U U U of G G G with its two-scale expansion, the lift Y ♯ Y^{\sharp} Y ♯ of (38) : the successor G ′ G' G ′ of Definition 11.3 is admissible, the solution Y Y Y of G ′ G' G ′ keeps the amplitude of the datum, and the homogenization error U − Y ♯ U-Y^{\sharp} U − Y ♯ is small. We also give the formulas for the sources of the successor, whose data Definition 11.3 lists. Section 12 proves part (a) and constructs the preparation time t p r t_{\mathrm{pr}} t pr and the map Π \Pi Π used below.
The route. Each of the three steps of the lift (38) turns one equation into another, exactly, up to explicit error terms (Figure 13.1).
The correctors (Subsection 13.1). The correctors of Section 10 cancel every term of degree at most N c o r N_{\mathrm{cor}} N cor of the operator of G G G applied to a lifted field. There remain the averaged operator ∂ t + b m − 1 ⋅ ∇ − ∇ ⋅ M m \partial_{t}+b_{m-1}\cdot\nabla-\nabla\cdot\mathcal{M}_{m} ∂ t + b m − 1 ⋅ ∇ − ∇ ⋅ M m , with the mean flux M m \mathcal{M}_{m} M m of the lift, whose part of order 2 without sources is J m κ m ∇ J_{m}^{\kappa_{m}}\nabla J m κ m ∇ up to a defect, and a tail of mean zero in the fast variable. Each degree of the tail costs the square root ( ε m 2 / ( κ m h m ) ) 1 / 2 (\varepsilon_{m}^{2}/(\kappa_{m}h_{m}))^{1/2} ( ε m 2 / ( κ m h m ) ) 1/2 of the ratio of the cell time to the renewal time. The correctors are switched from pulse to pulse in the quiet intervals, where they decay exponentially.
The truncated exponential (Subsection 13.2). Conjugation by exp p ( Γ m : ∇ 2 ) \exp_p(\Gamma_m : \nabla^2) exp p ( Γ m : ∇ 2 ) turns κ m − 1 Δ \kappa_{m-1}\Delta κ m − 1 Δ into J m κ m ( t ) : ∇ 2 J_m^{\kappa_m}(t) : \nabla^2 J m κ m ( t ) : ∇ 2 , up to a term of degree p p p in Γ m \Gamma_m Γ m times ∂ t Γ m = J m κ m − κ m − 1 I \partial_t\Gamma_m = J_m^{\kappa_m} - \kappa_{m-1}I ∂ t Γ m = J m κ m − κ m − 1 I , and commutes with b m − 1 ⋅ ∇ \mathbf{b}_{m-1}\cdot\nabla b m − 1 ⋅ ∇ up to the divergence of an explicit flux.
The readout (Subsection 13.3). The part of M m \mathcal{M}_m M m beyond J m κ m ∇ J_m^{\kappa_m}\nabla J m κ m ∇ and the commutator, conjugated by the truncated exponential, are not small. They are kept as the sources of G ′ G' G ′ : this is the construction of the successor. The readout weight ε m − 1 η \varepsilon_{m-1}^{\eta} ε m − 1 η that each letter (Definition 11.3 (i)) adds cancels the power ε m − 1 − η \varepsilon_{m-1}^{-\eta} ε m − 1 − η of the coefficients of the new source that carries it (Lemma 13.8 ), so the letters telescope.
The residual of Y ♯ Y^{\sharp} Y ♯ in the system of G G G is then a forcing of size C ε m − 1 Q ∥ g ∥ L G 2 ( T 2 ) C\varepsilon_{m-1}^{Q}\lVert g\rVert_{L^2_G(\mathbb{T}^2)} C ε m − 1 Q ∥ g ∥ L G 2 ( T 2 ) (Subsection 13.4), and Proposition 8.3 (a) bounds U − Y ♯ U-Y^{\sharp} U − Y ♯ by it (Subsection 13.5).
Figure 13.1. The three steps of the lift (38) , Y ♯ = Corr m [ exp p ( Γ m : ∇ 2 ) Υ m Y ] Y^{\sharp}=\operatorname{Corr}_m[\exp_p(\Gamma_m:\nabla^2)\Upsilon_mY] Y ♯ = Corr m [ exp p ( Γ m : ∇ 2 ) Υ m Y ] , from top to bottom. Each box names a field and the equation it solves, up to error terms; J m κ m J_m^{\kappa_m} J m κ m is the averaged flux (18) .
Setting. Throughout, C C C and c c c denote positive constants depending only on α , s , q , ζ , Λ , n , t 0 , T \alpha,s,q,\zeta,\Lambda,n,t_0,T α , s , q , ζ , Λ , n , t 0 , T . In Subsection 13.1, the setting of the lift is: M ≥ 1 M\ge1 M ≥ 1 , κ ∈ K M \kappa\in\mathcal{K}_M κ ∈ K M , 2 ≤ m ≤ M 2\le m\le M 2 ≤ m ≤ M , and a family G G G of level m m m that satisfies the invariant (Definition 11.1) on an interval [ t A , t f ] [t_A,t_f] [ t A , t f ] , whose sources have current order at most N s r c N_{\mathrm{src}} N src (the sums of (104) stop at the current order), whose paths of sources have at most N p a t h N_{\mathrm{path}} N path sources, and whose sources of birth order at least 3 have birth radius at most ε m \varepsilon_m ε m . An admissible family has these properties (item (1) of Subsection 11.1, Definition 11.2 (i), and Step 5 of the proof of Lemma 11.7 ). Its constants are those of § 5.4 , or other constants fixed depending only on α , s , q , ζ , Λ , n , t 0 , T \alpha,s,q,\zeta,\Lambda,n,t_0,T α , s , q , ζ , Λ , n , t 0 , T (for the terminal family of § 6 , with the terminal orders and path bound). Above a level that depends only on α , s , q , ζ , Λ , n , t 0 , T \alpha,s,q,\zeta,\Lambda,n,t_0,T α , s , q , ζ , Λ , n , t 0 , T and is at least m c m_c m c , such a family has the finite lifts of § 5.2 at every order N N N , with constants depending only on N N N and α , s , q , ζ , Λ , n , t 0 , T \alpha,s,q,\zeta,\Lambda,n,t_0,T α , s , q , ζ , Λ , n , t 0 , T . Indeed, the majorant matrix is at most the priced one, so its row and column sums are at most 1 by (b3). By (b1) and the bound on the birth radii, ∥ A σ , d ∥ a n , v \lVert A_{\sigma,d}\rVert_{\mathrm{an},v} ∥ A σ , d ∥ an , v is at most a constant times maj ( σ ) \operatorname{maj}(\sigma) maj ( σ ) . Cauchy’s estimate in the base label on D v D_v D v , whose supremum runs over the whole window of k k k , and the exponential weight in z z z give (108) for the forms, at every j 1 , j 2 j_1,j_2 j 1 , j 2 ; and (54) gives it for the conductivity, as the window of k k k lies in ∣ t − l m , k h m ∣ ≤ h m |t-l_{m,k}h_m|\le h_m ∣ t − l m , k h m ∣ ≤ h m . Below, Proposition 10.7 is used at the order N = N r e t N=N_{\mathrm{ret}} N = N ret . Clauses (a1)–(d) are those of Definition 11.1.
For a field F F F of a family of level m m m we write
( A m F ) v : = ∑ v ′ ∈ V ∑ σ : v ′ → v ∑ d = 2 N s r c A σ , d ∇ d − 1 F v ′ , ( A m S F ) v : = ∑ v ′ ∈ V ∑ σ : v ′ → v ∑ d = 2 N s r c A σ , d ∇ d − 1 ( S σ F v ′ ) (153) (A_mF)_v := \sum_{v'\in V}\sum_{\sigma:v'\to v}\sum_{d=2}^{N_{\mathrm{src}}} A_{\sigma,d}\nabla^{d-1}F_{v'}, \qquad(A_m^SF)_v := \sum_{v'\in V}\sum_{\sigma:v'\to v}\sum_{d=2}^{N_{\mathrm{src}}} A_{\sigma,d}\nabla^{d-1}(S_\sigma F_{v'}) \tag*{(153)} ( A m F ) v := v ′ ∈ V ∑ σ : v ′ → v ∑ d = 2 ∑ N src A σ , d ∇ d − 1 F v ′ , ( A m S F ) v := v ′ ∈ V ∑ σ : v ′ → v ∑ d = 2 ∑ N src A σ , d ∇ d − 1 ( S σ F v ′ ) (153)
for the flux of its sources without and with the smoothing, so that the system (34) of the family reads ∂ t U + b m ⋅ ∇ U − κ m Δ U = ∇ ⋅ A m S U \partial_tU+\mathbf{b}_m\cdot\nabla U-\kappa_m\Delta U=\nabla\cdot A_m^SU ∂ t U + b m ⋅ ∇ U − κ m Δ U = ∇ ⋅ A m S U .
The two-scale expansion
Subsections 13.3–13.5 work in the setting of Proposition 5.3: m 0 < m < M m_0<m<M m 0 < m < M , an admissible family G G G of level m m m on [ t A , t f ] [t_A,t_f] [ t A , t f ] and its successor G ′ G' G ′ , with set of components V ′ V' V ′ , a time t c t_c t c with t A + 3 τ m ≤ t c ≤ t f − 4 h m t_A+3\tau_m\le t_c\le t_f-4h_m t A + 3 τ m ≤ t c ≤ t f − 4 h m , the time t p r t_{\mathrm{pr}} t pr and the map Π \Pi Π of Section 12, a datum g g g of G G G , the solution U U U of G G G from g g g at time t c t_c t c , and the solution Y Y Y of G ′ G' G ′ from the retained trace of Π g \Pi g Π g at time t p r t_{\mathrm{pr}} t pr . Recall from Subsection 5.5 that the readout is ( Υ m Y ) v = Y v + ∑ v ′ ε m # v ′ Y v ′ (\Upsilon_mY)_v=Y_v+\sum_{v'}\varepsilon_m^{\#v'}Y_{v'} ( Υ m Y ) v = Y v + ∑ v ′ ε m # v ′ Y v ′ , over the children v ′ v' v ′ of v v v in G ′ G' G ′ , where # v ′ \#v' # v ′ is the number of letters of the word of v ′ v' v ′ . The retained trace of Π g \Pi g Π g at time t p r t_{\mathrm{pr}} t pr is exp p ( − Γ m ( t p r ) : ∇ 2 ) ( Π g ) v \exp_p(-\Gamma_m(t_{\mathrm{pr}}):\nabla^2)(\Pi g)_v exp p ( − Γ m ( t pr ) : ∇ 2 ) ( Π g ) v on each component v v v of G G G and 0 0 0 on the new components. Here t f − t p r ≤ T t_f-t_{\mathrm{pr}}\le T t f − t pr ≤ T (Definitions 11.2 and 11.4). The family G ′ G' G ′ is admissible on [ t A + 3 τ m , t f ] [t_A+3\tau_m,t_f] [ t A + 3 τ m , t f ] (Subsection 13.5, which uses only Definition 11.4), so, by Proposition 11.5, G G G and G ′ G' G ′ satisfy the invariant, with sums at most 1 8 \frac18 8 1 in clause (d) for both masses, and below 1 8 \frac18 8 1 at the order 0 when m m m is above a level depending only on α , s , q , ζ , Λ , n , t 0 , T \alpha,s,q,\zeta,\Lambda,n,t_0,T α , s , q , ζ , Λ , n , t 0 , T . Being admissible, G G G is in the setting of the lift. We write
Y b ( t ) : = exp p ( Γ m ( t ) : ∇ 2 ) Υ m Y ( t ) , so that Y ♯ ( t ) = Corr m Y b ( t ) . (154) Y^{\mathrm{b}}(t):=\exp_p\bigl(\Gamma_m(t):\nabla^2\bigr)\Upsilon_mY(t),\qquad\text{so that}\qquad Y^{\sharp}(t)=\operatorname{Corr}_mY^{\mathrm{b}}(t). \tag*{(154)} Y b ( t ) := exp p ( Γ m ( t ) : ∇ 2 ) Υ m Y ( t ) , so that Y ♯ ( t ) = Corr m Y b ( t ) . (154)
Two norms. For a field F F F of a family with masses ω v \omega_v ω v on [ t 1 , t 2 ] [t_1,t_2] [ t 1 , t 2 ] , an integer N ≥ 0 N\ge0 N ≥ 0 , ρ > 0 \rho>0 ρ > 0 and ν > 0 \nu>0 ν > 0 , the component energy of F F F of order N N N at the scale ρ \rho ρ , with the diffusivity ν \nu ν , on [ t 1 , t 2 ] [t_1,t_2] [ t 1 , t 2 ] is
( ∑ v ∈ V 1 ω v ( ∥ F v ∥ L ∞ ( ( t 1 , t 2 ) ; H ρ N ( T 2 ) 2 + ν ∥ ∇ F v ∥ L 2 ( ( t 1 , t 2 ) ; H ρ N ( T 2 ) 2 ) ) 1 / 2 , (155) \left(\sum_{v\in V}\frac{1}{\omega_v}\left(\|F_v\|_{L^\infty((t_1,t_2);H_\rho^N(\mathbb{T}^2)}^2+\nu\|\nabla F_v\|_{L^2((t_1,t_2);H_\rho^N(\mathbb{T}^2)}^2\right)\right)^{1/2}, \tag*{(155)} ( v ∈ V ∑ ω v 1 ( ∥ F v ∥ L ∞ (( t 1 , t 2 ) ; H ρ N ( T 2 ) 2 + ν ∥∇ F v ∥ L 2 (( t 1 , t 2 ) ; H ρ N ( T 2 ) 2 ) ) 1/2 , (155)
the square root of the left side of Proposition 8.3(c), with the supremum in time taken in each component. In this section, the energy of F F F of order N N N is its component energy (155) of order N N N at the scale ρ ( m ) \rho(m) ρ ( m ) , with the diffusivity κ m − 1 \kappa_{m-1} κ m − 1 , on the interval at hand. For a forcing ( F f , F f , 0 ) (F^f,F^{f,0}) ( F f , F f , 0 ) of the system (66) of a family of level m m m on [ t 1 , t 2 ] [t_1,t_2] [ t 1 , t 2 ] , its norm is
κ m − 1 / 2 ∥ F f ∥ L 2 ( ( t 1 , t 2 ) ; L ω 2 ( T 2 ) ) + ∥ F f , 0 ∥ L 1 ( ( t 1 , t 2 ) ; L ω 2 ( T 2 ) ) . (156) \kappa_m^{-1/2}\|F^f\|_{L^2((t_1,t_2);L^2_\omega(\mathbb{T}^2))}+\|F^{f,0}\|_{L^1((t_1,t_2);L^2_\omega(\mathbb{T}^2))}. \tag*{(156)} κ m − 1/2 ∥ F f ∥ L 2 (( t 1 , t 2 ) ; L ω 2 ( T 2 )) + ∥ F f , 0 ∥ L 1 (( t 1 , t 2 ) ; L ω 2 ( T 2 )) . (156)
Orders. Besides the parameters of Subsection 5.3 and Sections 11 and 14, the comparison uses the number N c l N_{\mathrm{cl}} N cl of antidivergence steps in the reconstruction of the cell tail (Lemma 13.4; “cl” for cleanup), and fixes the moment order N m o m N_{\mathrm{mom}} N mom for the smoothings of the sources of G ′ G' G ′ at the scale ρ ( m ) \rho(m) ρ ( m ) and of G G G at the scale ρ m o l ( m ) \rho_{\mathrm{mol}}(m) ρ mol ( m ) (Step 3 of the proof of Proposition 13.11). They are fixed in the order of item (1) of Subsection 11.1 (Table 11.1), as follows. Given Q Q Q (Section 14) and the seed’s bounds on its orders and path lengths (Section 15), we choose N g e n N_{\mathrm{gen}} N gen , N c o r N_{\mathrm{cor}} N cor , p p p and N c l N_{\mathrm{cl}} N cl ; then N s r c N_{\mathrm{src}} N src and Θ \Theta Θ , with ( N s r c − 2 ) Θ ≤ δ / 8 (N_{\mathrm{src}}-2)\Theta\le\delta/8 ( N src − 2 ) Θ ≤ δ /8 ; then N l e n N_{\mathrm{len}} N len ; then N m o m N_{\mathrm{mom}} N mom , such that
η ( N g e n + 1 ) 8 , δ ( N l e n + 1 ) 2 , 3 δ N c o r , δ p , Θ N m o m > Q + q γ + 4 , \frac{\eta(N_{\mathrm{gen}}+1)}{8},\qquad\frac{\delta(N_{\mathrm{len}}+1)}{2},\qquad3\delta N_{\mathrm{cor}},\qquad\delta p,\qquad\Theta N_{\mathrm{mom}}>Q+q\gamma+4, 8 η ( N gen + 1 ) , 2 δ ( N len + 1 ) , 3 δ N cor , δ p , Θ N mom > Q + q γ + 4 ,
( q γ + 3 δ ) N c l ≥ 2 − β + 8 δ . (157) (q\gamma+3\delta)N_{\mathrm{cl}}\ge2-\beta+8\delta. \tag*{(157)} ( q γ + 3 δ ) N cl ≥ 2 − β + 8 δ . (157)
these give q δ N m o m > 12 Θ N m o m > Q + 4 q\delta N_{\mathrm{mom}}>12\Theta N_{\mathrm{mom}}>Q+4 q δ N mom > 12Θ N mom > Q + 4 , since q δ > δ > 12 Θ q\delta>\delta>12\Theta q δ > δ > 12Θ by (181) ; the kernels then have vanishing moments up to N m o m N_{\mathrm{mom}} N mom (clause (c1)), and the third bound of Proposition 6.1 applies with N 1 = N m o m N_1=N_{\mathrm{mom}} N 1 = N mom . Last, we choose N r e t N_{\mathrm{ret}} N ret and set N o u t : = N r e t + 4 p N_{\mathrm{out}}:=N_{\mathrm{ret}}+4p N out := N ret + 4 p . With N r e c : = 4 ( N c o r + N s r c ) + 2 N c l + 16 N_{\mathrm{rec}}:=4(N_{\mathrm{cor}}+N_{\mathrm{src}})+2N_{\mathrm{cl}}+16 N rec := 4 ( N cor + N src ) + 2 N cl + 16 , the order of the reconstruction, and the path bound N p a t h N_{\mathrm{path}} N path of Section 11, N r e t N_{\mathrm{ret}} N ret is at least the passage floor of (180) and
N r e t ≥ max { N r e c + N s r c + N m o m + 6 p , N s r c − 2 + N m o m + N c o r + 2 p + 1 } , N_{\mathrm{ret}}\ge\max\{N_{\mathrm{rec}}+N_{\mathrm{src}}+N_{\mathrm{mom}}+6p,\ N_{\mathrm{src}}-2+N_{\mathrm{mom}}+N_{\mathrm{cor}}+2p+1\}, N ret ≥ max { N rec + N src + N mom + 6 p , N src − 2 + N mom + N cor + 2 p + 1 } ,
N r e t ≥ N r e c + 2 N s r c N p a t h + max { N s r c + N m o m + N c o r + 6 p + 4 , ( N p a t h + 3 ) ( N c o r + N s r c + 12 ) + 2 p } . (158) N_{\mathrm{ret}}\ge N_{\mathrm{rec}}+2N_{\mathrm{src}}N_{\mathrm{path}}
+\max\{N_{\mathrm{src}}+N_{\mathrm{mom}}+N_{\mathrm{cor}}+6p+4,\ (N_{\mathrm{path}}+3)(N_{\mathrm{cor}}+N_{\mathrm{src}}+12)+2p\}. \tag*{(158)} N ret ≥ N rec + 2 N src N path + max { N src + N mom + N cor + 6 p + 4 , ( N path + 3 ) ( N cor + N src + 12 ) + 2 p } . (158)
The second floor bounds the number of derivatives in the fast variable that a cycle takes along the paths of sources, which have at most N p a t h N_{\mathrm{path}} N path sources (proof of Lemma 11.7); it depends on N l e n N_{\mathrm{len}} N len through N p a t h N_{\mathrm{path}} N path .
The two-scale expansion
For a pulse k k k of level m m m , a field F F F of G G G is read in the chart of its block as F ~ k ( t , y ) \widetilde{F}_k(t,y) F k ( t , y ) , with the label y = X m − 1 ( l m , k h m , x , t ) y=X_{m-1}(l_m,k h_m,x,t) y = X m − 1 ( l m , k h m , x , t ) and the fast variable z = y / ε m z=y/\varepsilon_m z = y / ε m . The parts L 0 \mathcal{L}_0 L 0 , L 1 \mathcal{L}_1 L 1 , L 2 \mathcal{L}_2 L 2 and L j s r c [ σ ] \mathcal{L}^{\mathrm{src}}_j[\sigma] L j src [ σ ] of the operator of the system are those of (101) and (105) . Written without a path, V j \mathcal{V}_j V j and M j , d \mathcal{M}_{j,d} M j , d are the correctors and means of the system, the sums over the paths of Proposition 10.6(f), with V 0 = I \mathcal{V}_0=I V 0 = I . The operators Corr m , k \operatorname{Corr}_{m,k} Corr m , k , Corr m \operatorname{Corr}_m Corr m and the mean flux M m \mathcal{M}_m M m are those of Definition 10.8, the part of M m \mathcal{M}_m M m of degree d d d along the paths of i i i sources being the sum of the means of degree d d d along them.
The cost of a degree. Since ρ ( m ) 2 = κ m − 1 h m \rho(m)^2=\kappa_{m-1}h_m ρ ( m ) 2 = κ m − 1 h m , the corrector amplitude ε ˉ m \bar{\varepsilon}_m ε ˉ m of Subsection 5.1, divided by ρ ( m ) \rho(m) ρ ( m ) , is the square root of the ratio of the cell time ε m 2 / κ m \varepsilon_m^2/\kappa_m ε m 2 / κ m to the renewal time:
ε ˉ m ρ ( m ) = ( ε m 2 κ m h m ) 1 / 2 ≤ C ε m − 1 3 δ , ε m ρ ( m ) = ( κ m κ m − 1 ) 1 / 2 ( ε m 2 κ m h m ) 1 / 2 ≤ C ε m − 1 q γ + 3 δ . (159) \frac{\bar{\varepsilon}_m}{\rho(m)}=\left(\frac{\varepsilon_m^2}{\kappa_m h_m}\right)^{1/2}\le C\varepsilon_{m-1}^{3\delta},\qquad
\frac{\varepsilon_m}{\rho(m)}=\left(\frac{\kappa_m}{\kappa_{m-1}}\right)^{1/2}\left(\frac{\varepsilon_m^2}{\kappa_m h_m}\right)^{1/2}\le C\varepsilon_{m-1}^{q\gamma+3\delta}. \tag*{(159)} ρ ( m ) ε ˉ m = ( κ m h m ε m 2 ) 1/2 ≤ C ε m − 1 3 δ , ρ ( m ) ε m = ( κ m − 1 κ m ) 1/2 ( κ m h m ε m 2 ) 1/2 ≤ C ε m − 1 q γ + 3 δ . (159)
The first bound is Lemma 10.1(c); the second holds at every 2 ≤ m ≤ M 2\le m\le M 2 ≤ m ≤ M , since ε m ≤ 2 ε m − 1 q \varepsilon_m\le2\varepsilon_{m-1}^{q} ε m ≤ 2 ε m − 1 q (Lemma 7.1(a)), ρ ( m ) = ( κ m − 1 h m ) 1 / 2 ≥ c ε m − 1 1 + γ / 2 + δ \rho(m)=(\kappa_{m-1}h_m)^{1/2}\ge c\varepsilon_{m-1}^{1+\gamma/2+\delta} ρ ( m ) = ( κ m − 1 h m ) 1/2 ≥ c ε m − 1 1 + γ /2 + δ (Proposition 2.1(c)), and q − 1 − γ 2 − δ = q γ + 3 δ q-1-\frac{\gamma}{2}-\delta=q\gamma+3\delta q − 1 − 2 γ − δ = q γ + 3 δ by (5) . The degree is the number of derivatives in y y y , and every derivative that falls on the field is one of them, so the order never exceeds the degree. By Proposition 10.7, the coefficients of order d d d of a corrector of degree j j j are bounded, with their derivatives (each slow derivative costing ε m − 1 − 1 \varepsilon_{m-1}^{-1} ε m − 1 − 1 , each fast one a constant), by
C ε ˉ m j ε m − 1 d − j = C ρ ( m ) d ( ε m 2 κ m h m ) j / 2 ( ρ ( m ) ε m − 1 ) j − d , (160) C\bar{\varepsilon}_m^j\varepsilon_{m-1}^{d-j}
=C\rho(m)^d\left(\frac{\varepsilon_m^2}{\kappa_m h_m}\right)^{j/2}
\left(\frac{\rho(m)}{\varepsilon_{m-1}}\right)^{j-d}, \tag*{(160)} C ε ˉ m j ε m − 1 d − j = C ρ ( m ) d ( κ m h m ε m 2 ) j /2 ( ε m − 1 ρ ( m ) ) j − d , (160)
in row and column sums of the matrices with the entries multiplied by ( ω v ′ / ω v ) 1 / 2 (\omega_{v'}/\omega_v)^{1/2} ( ω v ′ / ω v ) 1/2 , as in (117) . The excess of the degree over the order costs ρ ( m ) / ε m − 1 ≤ C ε m − 1 γ / 2 + δ \rho(m)/\varepsilon_{m-1}\le C\varepsilon_{m-1}^{\gamma/2+\delta} ρ ( m ) / ε m − 1 ≤ C ε m − 1 γ /2 + δ . The coefficients of the means of degree j j j satisfy the same bound times κ m / ε m 2 \kappa_m/\varepsilon_m^2 κ m / ε m 2 . We also use c ε m ≤ ρ ( m ) ≤ ε m − 1 c\varepsilon_m\le\rho(m)\le\varepsilon_{m-1} c ε m ≤ ρ ( m ) ≤ ε m − 1 , so that ( ρ ( m ) / ε m − 1 ) j − d ≤ 1 (\rho(m)/\varepsilon_{m-1})^{j-d}\le1 ( ρ ( m ) / ε m − 1 ) j − d ≤ 1 .
The correctors of a pulse start from zero when its window opens, and stay zero while the shear is off and the forms do not depend on the fast variable (Proposition 10.6(e)). So at a time of [ t A + 3 τ m , t f ] [t_A+3\tau_m,t_f] [ t A + 3 τ m , t f ] with no pulse of level m m m active within τ m ′ \tau'_m τ m ′ , such as t p r t_{\rm pr} t pr (Proposition 12.1(i)), the lift adds no corrector.
Lemma 13.1 (Quiet entrance). The following holds in the setting of the lift, with m m m above a level depending only on α \alpha α , s s s , q q q , ζ \zeta ζ , Λ \Lambda Λ , n n n , t 0 t_0 t 0 , T T T . Let t ∈ [ t A + 3 τ m , t f ] t\in[t_A+3\tau_m,t_f] t ∈ [ t A + 3 τ m , t f ] be such that the velocity added at level m m m vanishes on [ t − τ m ′ , t + τ m ′ ] [t-\tau'_m,t+\tau'_m] [ t − τ m ′ , t + τ m ′ ] . Then, for every pulse k k k with H m , k ( t ) ≠ 0 H_{m,k}(t)\ne0 H m , k ( t ) = 0 , every corrector V j \mathcal{V}_j V j of k k k with j ≥ 1 j\ge1 j ≥ 1 vanishes at t t t . In particular, for every field F F F of G G G , Corr m F = F \operatorname{Corr}_m F=F Corr m F = F at the time t t t .
Proof. Let H m , k ( t ) ≠ 0 H_{m,k}(t)\ne0 H m , k ( t ) = 0 , so ∣ t − k τ m ∣ ≤ 5 4 τ m |t-k\tau_m|\le\frac{5}{4}\tau_m ∣ t − k τ m ∣ ≤ 4 5 τ m . The window of k k k starts at ( k − 61 48 ) τ m > t − 121 48 τ m (k-\frac{61}{48})\tau_m>t-\frac{121}{48}\tau_m ( k − 48 61 ) τ m > t − 48 121 τ m . Since τ m ′ / τ m = 4 ⌈ ε m − 1 − δ ⌉ + 1 ≥ 5 \tau'_m/\tau_m=4\lceil\varepsilon_{m-1}^{-\delta}\rceil+1\ge5 τ m ′ / τ m = 4 ⌈ ε m − 1 − δ ⌉ + 1 ≥ 5 , every t ′ t' t ′ between the start of the window and t t t has [ t ′ − τ m 24 , t ′ + τ m 24 ] ⊂ [ t − τ m ′ , t + τ m ′ ] [t'-\frac{\tau_m}{24},t'+\frac{\tau_m}{24}]\subset[t-\tau'_m,t+\tau'_m] [ t ′ − 24 τ m , t ′ + 24 τ m ] ⊂ [ t − τ m ′ , t + τ m ′ ] , and t ′ ≥ t − 3 τ m ≥ t A t'\ge t-3\tau_m\ge t_A t ′ ≥ t − 3 τ m ≥ t A . So the shear of k k k vanishes there, and, by (a4), the two-scale forms of G G G do not depend on z z z . For m m m large, the conductivity of the chart lies between 3 4 I \frac34 I 4 3 I and 5 4 I \frac54 I 4 5 I , by (54) . So, by Proposition 10.6(e), the correctors of positive degree vanish at t t t . □ \square □
When we apply the operator of the system to the field with its correctors and sort the terms by degree, each corrector of degree j ≤ N c o r j\le N_{\mathrm{cor}} j ≤ N cor removes the part of mean zero in z z z of the terms of degree j j j , and their means move into the averaged equation of F F F (Definition 10.5). The cell tail is what is left, of degrees above N c o r N_{\mathrm{cor}} N cor , each degree carrying the small factor ε ˉ m / ρ ( m ) \bar{\varepsilon}_m/\rho(m) ε ˉ m / ρ ( m ) of (159) ; in Lemma 13.4 one of them pays for the cell rate κ m / ε m 2 \kappa_m/\varepsilon_m^2 κ m / ε m 2 that multiplies its coefficients.
Lemma 13.2 (The correctors cancel the leading terms). There exists C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ C(\alpha,s,q,\zeta,\Lambda,n,t_0,T)<\infty C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ such that the following holds in the setting of the lift, with m m m above a level depending only on α , s , q , ζ , Λ , n , t 0 , T \alpha,s,q,\zeta,\Lambda,n,t_0,T α , s , q , ζ , Λ , n , t 0 , T . Let k k k be a pulse, let I I I be an interval in the window of k k k , and let F ~ ( t , y ) \widetilde{F}(t,y) F ( t , y ) be a field on I I I , in the chart of k k k , with F ~ ( t , ⋅ ) ∈ C 2 N c o r + N s r c + 1 ( R 2 ) \widetilde{F}(t,\cdot)\in C^{2N_{\mathrm{cor}}+N_{\mathrm{src}}+1}(\mathbb{R}^2) F ( t , ⋅ ) ∈ C 2 N cor + N src + 1 ( R 2 ) with bounded derivatives, and with ∇ y a F ~ \nabla_y^a\widetilde{F} ∇ y a F , ∣ a ∣ ≤ N c o r + 1 |a|\le N_{\mathrm{cor}}+1 ∣ a ∣ ≤ N cor + 1 , Lipschitz in t t t , uniformly in y y y . Then there exists a differential operator T k T_k T k in y y y , the cell tail on the pulse k k k , depending on no field, of orders 1 1 1 to 2 N c o r + N s r c 2N_{\mathrm{cor}}+N_{\mathrm{src}} 2 N cor + N src , whose coefficients have mean zero in z z z and whose terms have degrees N c o r + 1 N_{\mathrm{cor}}+1 N cor + 1 to 2 N c o r + N s r c 2N_{\mathrm{cor}}+N_{\mathrm{src}} 2 N cor + N src , such that, off a null set of t t t independent of ( y , z ) (y,z) ( y , z ) ,
( L 0 + L 1 + L 2 − ∑ σ ∑ j ′ = 0 N s r c L σ , j ′ s r c [ σ ] ∑ j ′ ′ = 0 N c o r V j ′ ′ ) F ~ − ∑ j ′ ′ = 0 N c o r V j ′ ′ ( ∂ t F ~ − ∑ j = 2 N c o r + N s r c ∑ d = 2 j ∇ y ⋅ ( M j , d ∇ y d − 1 F ~ ) ) = T k F ~ . (161) \left(\mathcal{L}_0+\mathcal{L}_1+\mathcal{L}_2-\sum_{\sigma}\sum_{j'=0}^{N_{\mathrm{src}}}\mathcal{L}^{\mathrm{src}}_{\sigma,j'}[\sigma]\sum_{j''=0}^{N_{\mathrm{cor}}}\mathcal{V}_{j''}\right)\widetilde{F}
-\sum_{j''=0}^{N_{\mathrm{cor}}}\mathcal{V}_{j''}\left(\partial_t\widetilde{F}-\sum_{j=2}^{N_{\mathrm{cor}}+N_{\mathrm{src}}}\sum_{d=2}^{j}\nabla_y\cdot\left(\mathcal{M}_{j,d}\nabla_y^{d-1}\widetilde{F}\right)\right)=T_k\widetilde{F}. \tag*{(161)} L 0 + L 1 + L 2 − σ ∑ j ′ = 0 ∑ N src L σ , j ′ src [ σ ] j ′′ = 0 ∑ N cor V j ′′ F − j ′′ = 0 ∑ N cor V j ′′ ( ∂ t F − j = 2 ∑ N cor + N src d = 2 ∑ j ∇ y ⋅ ( M j , d ∇ y d − 1 F ) ) = T k F . (161) Its coefficients of order d d d and degree j j j satisfy (160) times κ m / ε m 2 \kappa_m/\varepsilon_m^2 κ m / ε m 2 , in row and column sums as there, each slow derivative up to the order N c l N_{\mathrm{cl}} N cl costing ε m − 1 − 1 \varepsilon_{m-1}^{-1} ε m − 1 − 1 .
Proof. Off a null set of t t t , the derivatives in t t t of the ∇ y a F ~ \nabla_y^a\widetilde{F} ∇ y a F , ∣ a ∣ ≤ N c o r |a|\le N_{\mathrm{cor}} ∣ a ∣ ≤ N cor , and of the coefficients V j , d \mathcal{V}_{j,d} V j , d exist at every ( y , z ) (y,z) ( y , z ) : their difference quotients are Lipschitz in ( y , z ) (y,z) ( y , z ) , uniformly in the increment (by the Lipschitz bound one derivative higher, and by (103) with the Wiener norm of Proposition 10.7), and they converge on a countable dense set, at the Lebesgue points of the data of the cell equations. So the limits for the ∇ y a F ~ \nabla_y^a\widetilde{F} ∇ y a F commute with ∇ y \nabla_y ∇ y , and the limit for V j , d \mathcal{V}_{j,d} V j , d is the right side of its cell equation at every ( y , z ) (y,z) ( y , z ) , by continuity. Thus, off this null set, the product rule holds for V j , d ∇ y d F ~ \mathcal{V}_{j,d}\nabla_y^d\widetilde{F} V j , d ∇ y d F , Proposition 10.6(b) holds at every ( y , z ) (y,z) ( y , z ) , and the left side of (161) is computed at the time t t t as an identity between differential operators applied to F ~ \widetilde{F} F . We sort its terms by degree. The cell operator L 0 \mathcal{L}_0 L 0 contains ∂ t \partial_t ∂ t , so on V j F ~ \mathcal{V}_j\widetilde{F} V j F it acts on the coefficients and gives V j ∂ t F ~ \mathcal{V}_j\partial_t\widetilde{F} V j ∂ t F , which cancels. The remaining terms act on F ~ \widetilde{F} F through its derivatives in y y y , with coefficients built from the correctors, the means and the parts of the operator, which depend on no field. At the degree 0 0 0 only ∑ σ L 0 , σ s r c [ σ ] F ~ \sum_{\sigma}\mathcal{L}^{\mathrm{src}}_{0,\sigma}[\sigma]\widetilde{F} ∑ σ L 0 , σ src [ σ ] F remains, and it vanishes, as F ~ \widetilde{F} F does not depend on z z z . The terms of degree j ≥ 1 j\ge1 j ≥ 1 are then L 0 \mathcal{L}_0 L 0 acting on the coefficients of V j F ~ \mathcal{V}_j\widetilde{F} V j F , plus the bracket of degree j j j of Definition 10.5 (for the system, by Proposition 10.6(f)), plus ∑ j ′ ′ = 0 j − 2 V j ′ ′ ∑ d ∇ y ⋅ ( M j − j ′ ′ , d ∇ y d − 1 F ~ ) \sum_{j''=0}^{j-2}\mathcal{V}_{j''}\sum_d\nabla_y\cdot(\mathcal{M}_{j-j'',d}\nabla_y^{d-1}\widetilde{F}) ∑ j ′′ = 0 j − 2 V j ′′ ∑ d ∇ y ⋅ ( M j − j ′′ , d ∇ y d − 1 F ) , with V j ′ ′ = 0 \mathcal{V}_{j''}=0 V j ′′ = 0 for j ′ ′ > N c o r j''>N_{\mathrm{cor}} j ′′ > N cor and no term for j − j ′ ′ > N c o r + N s r c j-j''>N_{\mathrm{cor}}+N_{\mathrm{src}} j − j ′′ > N cor + N src . For 1 ≤ j ≤ N c o r 1\le j\le N_{\mathrm{cor}} 1 ≤ j ≤ N cor this is L 0 V j F ~ \mathcal{L}_0\mathcal{V}_j\widetilde{F} L 0 V j F minus (107) , which vanishes by Proposition 10.6(b).
For j > N c o r j>N_{\mathrm{cor}} j > N cor no corrector is added, and the terms of degree j j j are the bracket plus the sum. The term j ′ ′ = 0 j''=0 j ′′ = 0 of the sum cancels the mean of the bracket (Proposition 10.6(a)); the bracket and the means of Definition 10.5 are both formed with V j = 0 \mathcal{V}_j=0 V j = 0 for j > N c o r j>N_{\mathrm{cor}} j > N cor , and the terms j ′ ′ ≥ 1 j''\ge1 j ′′ ≥ 1 have mean zero, as the correctors do. For j > N c o r + N s r c j>N_{\mathrm{cor}}+N_{\mathrm{src}} j > N cor + N src the bracket and the term j ′ ′ = 0 j''=0 j ′′ = 0 vanish, since a source has at most N s r c N_{\mathrm{src}} N src order derivatives. So the degrees are N c o r + 1 N_{\mathrm{cor}}+1 N cor + 1 to 2 N c o r + N s r c 2N_{\mathrm{cor}}+N_{\mathrm{src}} 2 N cor + N src , and every term has order at least 1 1 1 , since the correctors of positive degree annihilate constants and L 1 , L 2 \mathcal{L}_1,\mathcal{L}_2 L 1 , L 2 , the source parts and the means differentiate.
Each term of degree j j j is a lower corrector times one of the four products (113) divided by ε m 2 / κ m \varepsilon_m^2/\kappa_m ε m 2 / κ m , or a corrector times a mean. By Proposition 10.7, its coefficients obey (160) times κ m / ε m 2 \kappa_m/\varepsilon_m^2 κ m / ε m 2 , in row and column sums, which are submultiplicative, those of the majorant matrix being at most 1 1 1 by (b3). For the slow derivatives up to the order N c l N_{\mathrm{cl}} N cl , we use Proposition 10.7 at an order N ≥ N c l + N c o r + N s r c N\ge N_{\mathrm{cl}}+N_{\mathrm{cor}}+N_{\mathrm{src}} N ≥ N cl + N cor + N src (such as N r e t N_{\mathrm{ret}} N ret , by (158) ). In a coefficient of the tail, a corrector carries at most N s r c N_{\mathrm{src}} N src slow derivatives and a mean at most N c o r + 1 N_{\mathrm{cor}}+1 N cor + 1 (one from the divergence and at most N c o r N_{\mathrm{cor}} N cor from the corrector applied
to it), so that after N c l N_{\mathrm{cl}} N cl further slow derivatives each carries at most N N N . Each slow derivative of a corrector, a mean, the conductivity or a form costs ε ˉ m − 1 − 1 \bar{\varepsilon}_{m-1}^{-1} ε ˉ m − 1 − 1 (Proposition 10.7 , (108) ).
At one time, the correctors change a field by at most a constant times ε ˉ m / ρ ( m ) \bar{\varepsilon}_m/\rho(m) ε ˉ m / ρ ( m ) times its derivatives at the scale ρ ( m ) \rho(m) ρ ( m ) , and, as they oscillate at the wavelength ε m \varepsilon_m ε m , each derivative of the change costs ε m − 1 \varepsilon_m^{-1} ε m − 1 (Lemma 13.3 ). On a field smooth at the wavelength ε m − 1 \varepsilon_{m-1} ε m − 1 , this gives the cost (30) of the correctors in Step 4 of the proof of Proposition 5.4 , where the condition s < 1 2 ( 1 − α ) s < \frac{1}{2}(1-\alpha) s < 2 1 ( 1 − α ) is used.
Lemma 13.3 (The correctors on a field). There exists C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ C(\alpha,s,q,\zeta,\Lambda,n,t_0,T) < \infty C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ such that, in the setting of the lift with m 0 < m < M m_0 < m < M m 0 < m < M , for every t ∈ [ t A + 3 τ m , t f ] t \in[t_{\mathrm A}+3\tau_m,t_{\mathrm f}] t ∈ [ t A + 3 τ m , t f ] , every field F F F of G G G of class C N r e p + N c o r + 2 C^{N_{\mathrm{rep}}+N_{\mathrm{cor}}+2} C N rep + N cor + 2 on T 2 \mathbb{T}^2 T 2 , every multi-index a a a with ∣ a ∣ ≤ N r e p |a| \le N_{\mathrm{rep}} ∣ a ∣ ≤ N rep and every x ∈ T 2 x \in\mathbb{T}^2 x ∈ T 2 ,
ε ˉ m ∣ a ∣ ∣ ∂ a ( Corr m F − F ) ( x ) ∣ ω ≤ C ε ˉ m ρ ( m ) ∑ 1 ≤ ∣ a ′ ∣ ≤ N r e p + N c o r ρ ( m ) ∣ a ′ ∣ ∣ ∂ a ′ F ( x ) ∣ ω , (162) \bar{\varepsilon}_m^{|a|} \left|\partial^a(\operatorname{Corr}_m F-F)(x)\right|_\omega\le C\frac{\bar{\varepsilon}_m}{\rho(m)} \sum_{1\le|a'|\le N_{\mathrm{rep}}+N_{\mathrm{cor}}} \rho(m)^{|a'|}|\partial^{a'}F(x)|_\omega, \tag*{(162)} ε ˉ m ∣ a ∣ ∣ ∂ a ( Corr m F − F ) ( x ) ∣ ω ≤ C ρ ( m ) ε ˉ m 1 ≤ ∣ a ′ ∣ ≤ N rep + N cor ∑ ρ ( m ) ∣ a ′ ∣ ∣ ∂ a ′ F ( x ) ∣ ω , (162) where Corr m \operatorname{Corr}_m Corr m acts at the time t t t , as in (38) .
Proof. At each time at most two time cutoffs are positive, so it suffices to bound Corr m , k F − F = ∑ j = 1 N c o r V j F ~ k \operatorname{Corr}_{m,k}F-F=\sum_{j=1}^{N_{\mathrm{cor}}}\mathcal{V}_j\widetilde{F}_k Corr m , k F − F = ∑ j = 1 N cor V j F k , read at y = X m − 1 ( l m , k h m , x , t ) y=X_{m-1}(l_m,k h_m,x,t) y = X m − 1 ( l m , k h m , x , t ) , z = y / ε m z=y/\varepsilon_m z = y / ε m . The coefficients of the correctors of positive degree act on derivatives of order d ≥ 1 d\ge1 d ≥ 1 , since these correctors annihilate constants. A derivative in x x x falls on the fast variable (cost ε m − 1 \varepsilon_m^{-1} ε m − 1 ), on a slow coefficient or on the chart (cost C ε m − 1 − 1 C\varepsilon_{m-1}^{-1} C ε m − 1 − 1 , by (54) ), or on F F F . So, by (160) and the chain rule, ε ˉ m ∣ a ∣ ∣ ∂ a V j F ~ k ∣ \bar{\varepsilon}_m^{|a|}|\partial^a\mathcal{V}_j\widetilde{F}_k| ε ˉ m ∣ a ∣ ∣ ∂ a V j F k ∣ is at most a sum, over 1 ≤ d ′ ≤ d ≤ j 1\le d'\le d\le j 1 ≤ d ′ ≤ d ≤ j and d ′ ≤ ∣ a ′ ∣ ≤ d ′ + ∣ a ∣ d'\le|a'|\le d'+|a| d ′ ≤ ∣ a ′ ∣ ≤ d ′ + ∣ a ∣ , of C ρ ( m ) d ′ ( ε m 2 / ( κ m h m ) ) j / 2 ( ρ ( m ) / ε m − 1 ) j − d ′ ε m ∣ a ′ ∣ − d ′ ∣ ∂ a ′ F ∣ C\rho(m)^{d'}(\varepsilon_m^2/(\kappa_mh_m))^{j/2}(\rho(m)/\varepsilon_{m-1})^{j-d'}\varepsilon_m^{|a'|-d'}|\partial^{a'}F| C ρ ( m ) d ′ ( ε m 2 / ( κ m h m ) ) j /2 ( ρ ( m ) / ε m − 1 ) j − d ′ ε m ∣ a ′ ∣ − d ′ ∣ ∂ a ′ F ∣ . Indeed, ∇ y d F ~ k \nabla_y^d\widetilde{F}_k ∇ y d F k contains the derivatives of F F F of every order 1 ≤ d ′ ≤ d 1\le d'\le d 1 ≤ d ′ ≤ d , the other d − d ′ d-d' d − d ′ derivatives falling on the chart at the cost C ε m − 1 − d C\varepsilon_{m-1}^{-d} C ε m − 1 − d , which turns (160) at the order d d d into (160) at the order d ′ d' d ′ . Moreover
ρ ( m ) d ′ ( ε m 2 κ m h m ) d ′ / 2 ( ρ ( m ) ε m − 1 ) j − d ′ ε m ∣ a ′ ∣ − d ′ = ε ˉ m ρ ( m ) ρ ( m ) ∣ a ′ ∣ ( ε m 2 κ m h m ) j − 1 2 ( ρ ( m ) ε m − 1 ) j − d ′ ( ε m ρ ( m ) ) ∣ a ′ ∣ − d ′ , \rho(m)^{d'}\left(\frac{\varepsilon_m^2}{\kappa_m h_m}\right)^{d'/2}\left(\frac{\rho(m)}{\varepsilon_{m-1}}\right)^{j-d'}\varepsilon_m^{|a'|-d'}
=
\frac{\bar{\varepsilon}_m}{\rho(m)}\rho(m)^{|a'|}\left(\frac{\varepsilon_m^2}{\kappa_m h_m}\right)^{\frac{j-1}{2}}\left(\frac{\rho(m)}{\varepsilon_{m-1}}\right)^{j-d'}\left(\frac{\varepsilon_m}{\rho(m)}\right)^{|a'|-d'}, ρ ( m ) d ′ ( κ m h m ε m 2 ) d ′ /2 ( ε m − 1 ρ ( m ) ) j − d ′ ε m ∣ a ′ ∣ − d ′ = ρ ( m ) ε ˉ m ρ ( m ) ∣ a ′ ∣ ( κ m h m ε m 2 ) 2 j − 1 ( ε m − 1 ρ ( m ) ) j − d ′ ( ρ ( m ) ε m ) ∣ a ′ ∣ − d ′ ,
and the last three powers are bounded: the first and the third by (159) , as j ≥ 1 j\ge1 j ≥ 1 and ∣ a ′ ∣ ≥ d ′ |a'|\ge d' ∣ a ′ ∣ ≥ d ′ , and the second by ρ ( m ) ≤ ε m − 1 \rho(m)\le\varepsilon_{m-1} ρ ( m ) ≤ ε m − 1 , as j ≥ d ′ j\ge d' j ≥ d ′ . Here 1 ≤ ∣ a ′ ∣ ≤ N c o r + N r e p 1\le|a'|\le N_{\mathrm{cor}}+N_{\mathrm{rep}} 1 ≤ ∣ a ′ ∣ ≤ N cor + N rep . By Proposition 10.7 (a), whose hypothesis holds by (b3), and Schur’s test, the coefficients ∑ V j , d \sum\mathcal{V}_{j,d} ∑ V j , d , summed over the paths, act on ( R ν , ∣ ⋅ ∣ ω ) (\mathbb{R}^{\nu},|\cdot|_\omega) ( R ν , ∣ ⋅ ∣ ω ) with norm at most C ε ˉ m j ε m − 1 d − j C\bar{\varepsilon}_m^{j}\varepsilon_{m-1}^{d-j} C ε ˉ m j ε m − 1 d − j , so these bounds hold in the norm ∣ ⋅ ∣ ω |\cdot|_\omega ∣ ⋅ ∣ ω .
The tail in physical space. The cell tail of (161) has mean zero in the fast variable, and is read at z = y / ε m z=y/\varepsilon_m z = y / ε m . A function of mean zero in z z z is a derivative in z z z , and a derivative in z z z read at z = y / ε m z=y/\varepsilon_m z = y / ε m is ε m \varepsilon_m ε m times a derivative in y y y , up to a new slow derivative. So each antidivergence step gains the factor ε m / ρ ( m ) \varepsilon_m/\rho(m) ε m / ρ ( m ) , and after N c l N_{\mathrm{cl}} N cl steps only a scalar remainder of that size is left. In (163) the factor before the parenthesis is the power N c o r N_{\mathrm{cor}} N cor of ( ε m 2 / ( κ m h m ) ) 1 / 2 = ε ˉ m / ρ ( m ) (\varepsilon_m^2/(\kappa_mh_m))^{1/2}=\bar{\varepsilon}_m/\rho(m) ( ε m 2 / ( κ m h m ) ) 1/2 = ε ˉ m / ρ ( m ) (by (159) ), one degree paying for the cell rate against the dissipation. In the second term, the remainder, the power N c l N_{\mathrm{cl}} N cl of ε m / ρ ( m ) \varepsilon_m/\rho(m) ε m / ρ ( m ) beats, by the choice of N c l N_{\mathrm{cl}} N cl in (157) , the square root of the number of cell diffusion times in [ t 1 , t 2 ] [t_1,t_2] [ t 1 , t 2 ] when t 2 − t 1 ≤ T t_2-t_1\le T t 2 − t 1 ≤ T .
Lemma 13.4 (The cell tail). *There exists C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ C(\alpha,s,q,\zeta,\Lambda,n,t_0,T) < \infty C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ such that the following holds in the setting of the lift, with m m m above a level depending only on α , s , q , ζ , Λ , n , t 0 , T \alpha,s,q,\zeta,\Lambda,n,t_0,T α , s , q , ζ , Λ , n , t 0 , T . Let [ t 1 , t 2 ] ⊂ [ t A + 3 τ m , t f ] [t_1,t_2]\subset[t_{\mathrm A}+3\tau_m,t_{\mathrm f}] [ t 1 , t 2 ] ⊂ [ t A + 3 τ m , t f ] , and let F F F be a field of G G G on [ t 1 , t 2 ] [t_1,t_2] [ t 1 , t 2 ] , of class C N r e p + 1 C^{N_{\mathrm{rep}}+1} C N rep + 1 in x x x at each time, with derivatives measurable in ( t , x ) (t,x) ( t , x ) . Then there exists a forcing ( F t , 0 , F t , 1 ) (F^{\mathrm{t},0},F^{\mathrm{t},1}) ( F t , 0 , F t , 1 ) such that the fields T k F ~ k \mathcal{T}_k\widetilde{F}_k T k F k , with T k \mathcal{T}_k T k the cell tail of Lemma 13.2 , read at z = y / ε m z=y/\varepsilon_m z = y / ε m , carried to x x x by the charts and glued by the
cutoffs H m , k ( t ) \mathrm{H}_{m,k}(t) H m , k ( t ) , are ∇ ⋅ F t l + F t l , 0 \nabla\cdot F^{\mathrm{tl}} + F^{\mathrm{tl},0} ∇ ⋅ F tl + F tl , 0 , and such that its norm (156) is at most the energy of F F F of order N r e c N_{\mathrm{rec}} N rec on [ t 1 , t 2 ] [t_1,t_2] [ t 1 , t 2 ] times
C ( ε m 2 κ m h m ) N c o r / 2 ( 1 + ( t 2 − t 1 ) 1 / 2 κ m 1 / 2 ε m ( ε m ρ ( m ) ) N c l ) . (163) C\left(\frac{\varepsilon_m^2}{\kappa_m h_m}\right)^{N_{\mathrm{cor}}/2}\left(1+(t_2-t_1)^{1/2}\frac{\kappa_m^{1/2}}{\varepsilon_m}\left(\frac{\varepsilon_m}{\rho(m)}\right)^{N_{\mathrm{cl}}}\right). \tag*{(163)} C ( κ m h m ε m 2 ) N cor /2 ( 1 + ( t 2 − t 1 ) 1/2 ε m κ m 1/2 ( ρ ( m ) ε m ) N cl ) . (163)
Proof. Step 1: Antidivergence. The forms are transverse, by clause (a2), so each coefficient of the cell tail depends on z z z only through z ⋅ e ( k ) ⊥ z\cdot e^{\perp}_{(k)} z ⋅ e ( k ) ⊥ . A function of mean zero in z ⋅ e ( k ) ⊥ z\cdot e^{\perp}_{(k)} z ⋅ e ( k ) ⊥ has a primitive ( e ( k ) ⊥ ⋅ ∇ z ) − 1 (e^{\perp}_{(k)}\cdot\nabla_z)^{-1} ( e ( k ) ⊥ ⋅ ∇ z ) − 1 of mean zero, bounded by its supremum, with the same slow derivatives. Let F 0 c l F^{\mathrm{cl}}_0 F 0 cl be one term of the cell tail of F ~ k \widetilde{F}_k F k , a function of ( t , y , z ) (t,y,z) ( t , y , z ) of mean zero in z z z , and let the cleanup iterates be F i + 1 c l : = − e ( k ) ⊥ ⋅ ∇ y ( e ( k ) ⊥ ⋅ ∇ z ) − 1 F i c l F^{\mathrm{cl}}_{i+1}:=-e^{\perp}_{(k)}\cdot\nabla_y(e^{\perp}_{(k)}\cdot\nabla_z)^{-1}F^{\mathrm{cl}}_i F i + 1 cl := − e ( k ) ⊥ ⋅ ∇ y ( e ( k ) ⊥ ⋅ ∇ z ) − 1 F i cl , again of mean zero. The chain rule gives
F 0 c l ( t , y , y ε m ) = ∇ y ⋅ ( ∑ i = 0 N c l − 1 ε m i + 1 e ( k ) ⊥ ( ( e ( k ) ⊥ ⋅ ∇ z ) − 1 F i c l ) ( t , y , y ε m ) ) + ε m N c l F N c l c l ( t , y , y ε m ) . (164) F^{\mathrm{cl}}_0\left(t,y,\frac{y}{\varepsilon_m}\right)=\nabla_y\cdot\left(\sum_{i=0}^{N_{\mathrm{cl}}-1}\varepsilon_m^{i+1}e^{\perp}_{(k)}\left((e^{\perp}_{(k)}\cdot\nabla_z)^{-1}F^{\mathrm{cl}}_i\right)\left(t,y,\frac{y}{\varepsilon_m}\right)\right)+\varepsilon_m^{N_{\mathrm{cl}}}F^{\mathrm{cl}}_{N_{\mathrm{cl}}}\left(t,y,\frac{y}{\varepsilon_m}\right). \tag*{(164)} F 0 cl ( t , y , ε m y ) = ∇ y ⋅ ( i = 0 ∑ N cl − 1 ε m i + 1 e ( k ) ⊥ ( ( e ( k ) ⊥ ⋅ ∇ z ) − 1 F i cl ) ( t , y , ε m y ) ) + ε m N cl F N cl cl ( t , y , ε m y ) . (164)
Each slow derivative in F i + 1 c l F^{\mathrm{cl}}_{i+1} F i + 1 cl falls on a coefficient (cost ε m − 1 − 1 ≤ ρ ( m ) − 1 \varepsilon_{m-1}^{-1}\leq\rho(m)^{-1} ε m − 1 − 1 ≤ ρ ( m ) − 1 ) or on F ~ k \widetilde{F}_k F k .
Step 2: Size of the flux. Let the term F 0 c l F^{\mathrm{cl}}_0 F 0 cl have degree j ≥ N c o r + 1 j\geq N_{\mathrm{cor}}+1 j ≥ N cor + 1 and order d d d . Lemma 13.2 bounds the flux of (164) with i = 0 i=0 i = 0 by C ε m κ m / ε m 2 C\varepsilon_m\kappa_m/\varepsilon_m^2 C ε m κ m / ε m 2 times the bound (160) times ∣ ∇ y d F ~ k ∣ \lvert\nabla_y^d\widetilde{F}_k\rvert ∣ ∇ y d F k ∣ . By the chain rule, the first line of (54) , area preservation and ρ ( m ) ≤ ε m − 1 \rho(m)\leq\varepsilon_{m-1} ρ ( m ) ≤ ε m − 1 , and by (36) and (155) , for d ≥ 1 d\geq1 d ≥ 1 , ∥ ∇ y d F ~ k ∥ L 2 ( ( t 1 , t 2 ) ; L ω 2 ( T 2 ) ) \|\nabla_y^d\widetilde{F}_k\|_{L^2((t_1,t_2);L^2_\omega(\mathbb{T}^2))} ∥ ∇ y d F k ∥ L 2 (( t 1 , t 2 ) ; L ω 2 ( T 2 )) is at most C ρ ( m ) 1 − d κ m − 1 − 1 / 2 C\rho(m)^{1-d}\kappa_{m-1}^{-1/2} C ρ ( m ) 1 − d κ m − 1 − 1/2 times the energy of F F F of order d d d . Also κ m − 1 / 2 κ m ε m − 1 / 2 κ m − 1 − 1 ρ ( m ) = ( κ m h m / ε m 2 ) 1 / 2 \kappa_m^{-1/2}\kappa_m\varepsilon_m^{-1/2}\kappa_{m-1}^{-1}\rho(m)=(\kappa_m h_m/\varepsilon_m^2)^{1/2} κ m − 1/2 κ m ε m − 1/2 κ m − 1 − 1 ρ ( m ) = ( κ m h m / ε m 2 ) 1/2 . So the flux contributes to (156) at most
C ( ε m 2 κ m h m ) j − 1 2 ( ρ ( m ) ε m − 1 ) j − d ≤ C ( ε m 2 κ m h m ) N c o r / 2 C\left(\frac{\varepsilon_m^2}{\kappa_m h_m}\right)^{\frac{j-1}{2}}\left(\frac{\rho(m)}{\varepsilon_{m-1}}\right)^{j-d}\leq C\left(\frac{\varepsilon_m^2}{\kappa_m h_m}\right)^{N_{\mathrm{cor}}/2} C ( κ m h m ε m 2 ) 2 j − 1 ( ε m − 1 ρ ( m ) ) j − d ≤ C ( κ m h m ε m 2 ) N cor /2
times the energy, by (159) . The steps i ≥ 1 i\geq1 i ≥ 1 add the factors ( ε m / ρ ( m ) ) i ≤ C (\varepsilon_m/\rho(m))^i\leq C ( ε m / ρ ( m ) ) i ≤ C .
Step 3: The remainder. The remainder ε m N c l F N c l c l \varepsilon_m^{N_{\mathrm{cl}}}F^{\mathrm{cl}}_{N_{\mathrm{cl}}} ε m N cl F N cl cl is at most ( ε m / ρ ( m ) ) N c l (\varepsilon_m/\rho(m))^{N_{\mathrm{cl}}} ( ε m / ρ ( m ) ) N cl times the bound of Step 2 divided by ε m \varepsilon_m ε m , on derivatives ∇ y d ′ F ~ k \nabla_y^{d'}\widetilde{F}_k ∇ y d ′ F k with d ≤ d ′ ≤ d + N c l d\leq d'\leq d+N_{\mathrm{cl}} d ≤ d ′ ≤ d + N cl . Its norm in L 1 ( ( t 1 , t 2 ) ; L 2 ( T 2 ) ) L^1((t_1,t_2);L^2(\mathbb{T}^2)) L 1 (( t 1 , t 2 ) ; L 2 ( T 2 )) is at most ( t 2 − t 1 ) 1 / 2 (t_2-t_1)^{1/2} ( t 2 − t 1 ) 1/2 times its norm in L 2 L^2 L 2 . As in Step 2, with the relation κ m ε m − 2 κ m − 1 − 1 ρ ( m ) = ( κ m 1 / 2 / ε m ) ( κ m h m / ε m 2 ) 1 / 2 \kappa_m\varepsilon_m^{-2}\kappa_{m-1}^{-1}\rho(m)=(\kappa_m^{1/2}/\varepsilon_m)(\kappa_mh_m/\varepsilon_m^2)^{1/2} κ m ε m − 2 κ m − 1 − 1 ρ ( m ) = ( κ m 1/2 / ε m ) ( κ m h m / ε m 2 ) 1/2 , this gives the second term of (163) .
Step 4: Back to x x x . The charts preserve area and turn the divergence in y y y into that in x x x (Piola, Lemma 7.9, with ∣ ∇ y X m − 1 ∣ ≤ 2 \lvert\nabla_yX_{m-1}\rvert\leq2 ∣ ∇ y X m − 1 ∣ ≤ 2 ), and derivatives in y y y of order d ′ d' d ′ into derivatives in x x x of orders d ′ ′ ≤ d ′ d''\leq d' d ′′ ≤ d ′ at the cost C ε m − 1 d ′ ′ − d ′ C\varepsilon_{m-1}^{d''-d'} C ε m − 1 d ′′ − d ′ (Lemma 7.7); at most two cutoffs are positive at each time, and the orders used are at most 2 N c o r + N s r c + N c l ≤ N r e c 2N_{\mathrm{cor}}+N_{\mathrm{src}}+N_{\mathrm{cl}}\leq N_{\mathrm{rec}} 2 N cor + N src + N cl ≤ N rec , by Step 3 and Lemma 13.2. On the paths with sources, the sums over the paths are bounded by (b3). □
Switching. Lemma 13.5 glues the identities of Lemma 13.2 in the coordinates x x x , using the lift of each pulse where its time cutoff is positive. The gluing produces the derivatives H m , k ′ = ∂ t H m , k \mathrm{H}'_{m,k}=\partial_t\mathrm{H}_{m,k} H m , k ′ = ∂ t H m , k of the cutoffs, supported in the quiet intervals [ ( k + 2 3 ) τ m , ( k + 4 3 ) τ m ] [(k+\frac{2}{3})\tau_m,(k+\frac{4}{3})\tau_m] [( k + 3 2 ) τ m , ( k + 3 4 ) τ m ] , k ∈ 2 Z + 1 k\in2\mathbb{Z}+1 k ∈ 2 Z + 1 , between consecutive pulses, where the correctors of the pulse that has just switched off decay at the rate c κ m / ε m 2 c\kappa_m/\varepsilon_m^2 c κ m / ε m 2 , and those of the next pulse have not started.
Lemma 13.5 (Switching between pulses). There exist C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ C(\alpha,s,q,\zeta,\Lambda,n,t_0,T)<\infty C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ and c ( α , s , q , ζ , Λ , n , t 0 , T ) > 0 c(\alpha,s,q,\zeta,\Lambda,n,t_0,T)>0 c ( α , s , q , ζ , Λ , n , t 0 , T ) > 0 such that the following holds in the setting of the lift, with α \alpha α a level depending only on α , s , q , ζ , Λ , n , t 0 , T \alpha,s,q,\zeta,\Lambda,n,t_0,T α , s , q , ζ , Λ , n , t 0 , T . Let [ t 1 , t 2 ] ⊂ [ t A + 3 τ m , t f ] [t_1,t_2]\subset[t_A+3\tau_m,t_f] [ t 1 , t 2 ] ⊂ [ t A + 3 τ m , t f ] with t 2 − t 1 ≤ T t_2-t_1\leq T t 2 − t 1 ≤ T , and let F F F be a field of G G G on [ t 1 , t 2 ] [t_1,t_2] [ t 1 , t 2 ] , of class C N r e c + 1 C^{N_{\mathrm{rec}}+1} C N rec + 1 in x x x at each time, with derivatives measurable in ( t , x ) (t,x) ( t , x ) , with ∇ x a F \nabla_x^aF ∇ x a F ,
∣ a ∣ ≤ N c o r + 2 |a| \le N_{\mathrm{cor}}+2 ∣ a ∣ ≤ N cor + 2 , bounded on [ t 1 , t 2 ] × T 2 [t_1,t_2]\times\mathbb{T}^2 [ t 1 , t 2 ] × T 2 , and with ∇ x a F \nabla_x^aF ∇ x a F , ∣ a ∣ ≤ N c o r + 1 |a|\le N_{\mathrm{cor}}+1 ∣ a ∣ ≤ N cor + 1 , Lipschitz in t t t , uniformly in x x x . Then, with a forcing ( F t l , F t l , 0 ) (F^{\mathrm{tl}},F^{\mathrm{tl},0}) ( F tl , F tl , 0 ) as in Lemma 13.4, at almost every t ∈ ( t 1 , t 2 ) t\in(t_1,t_2) t ∈ ( t 1 , t 2 ) ,
( ∂ t + b m ⋅ ∇ − κ m Δ ) Corr m F − ∇ ⋅ ( A m Corr m F ) − Corr m ( ∂ t F + b m − 1 ⋅ ∇ F − ∇ ⋅ ( M m F ) ) = ∇ ⋅ F t l + F t l , 0 + ∑ k ∈ 2 Z + 1 H m , k ′ ( t ) ( Corr m , k F − F ) , (165) \begin{aligned}
&(\partial_t+\mathbf{b}_m\cdot\nabla-\kappa_m\Delta)\operatorname{Corr}_m F-\nabla\cdot(\mathbf{A}_m\operatorname{Corr}_m F)-\operatorname{Corr}_m(\partial_tF+\mathbf{b}_{m-1}\cdot\nabla F-\nabla\cdot(\mathcal{M}_mF))\\
&\qquad=\nabla\cdot F^{\mathrm{tl}}+F^{\mathrm{tl},0}+\sum_{k\in2\mathbb{Z}+1}\mathrm{H}'_{m,k}(t)(\operatorname{Corr}_{m,k}F-F), \tag*{(165)}
\end{aligned} ( ∂ t + b m ⋅ ∇ − κ m Δ ) Corr m F − ∇ ⋅ ( A m Corr m F ) − Corr m ( ∂ t F + b m − 1 ⋅ ∇ F − ∇ ⋅ ( M m F )) = ∇ ⋅ F tl + F tl , 0 + k ∈ 2 Z + 1 ∑ H m , k ′ ( t ) ( Corr m , k F − F ) , (165)
and the norm of the last sum in L 1 ( ( t 1 , t 2 ) ; L ω 2 ( T 2 ) ) L^1((t_1,t_2);L^2_\omega(\mathbb{T}^2)) L 1 (( t 1 , t 2 ) ; L ω 2 ( T 2 )) is at most the energy of F F F of order N r e c N_{\mathrm{rec}} N rec on [ t 1 , t 2 ] [t_1,t_2] [ t 1 , t 2 ] times
C ( ε m 2 κ m ) 1 / 2 ( 1 + ( t 2 − t 1 ) / τ m τ m ) 1 / 2 exp ( − c κ m τ m ε m 2 ) . (166) C\left(\frac{\varepsilon_m^2}{\kappa_m}\right)^{1/2}\left(1+\frac{(t_2-t_1)/\tau_m}{\tau_m}\right)^{1/2}\exp\left(-c\frac{\kappa_m\tau_m}{\varepsilon_m^2}\right). \tag*{(166)} C ( κ m ε m 2 ) 1/2 ( 1 + τ m ( t 2 − t 1 ) / τ m ) 1/2 exp ( − c ε m 2 κ m τ m ) . (166)
Moreover κ m τ m / ε m 2 ≥ c ε m − 4 δ \kappa_m\tau_m/\varepsilon_m^2\ge c\varepsilon_m^{-4\delta} κ m τ m / ε m 2 ≥ c ε m − 4 δ , so the exponential is smaller than any power of ε m − 1 \varepsilon_m^{-1} ε m − 1 . The field Corr m F \operatorname{Corr}_m F Corr m F is Lipschitz in time in L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) , so (165) holds in the form (67) .
Proof. Step 1: One pulse. Where H m , k > 0 \mathrm{H}_{m,k}>0 H m , k > 0 , the velocity added at level m m m is the shear of k k k , and A m \mathbf{A}_m A m is the transport of the coefficients of k k k . Where H m , k = 1 \mathrm{H}_{m,k}=1 H m , k = 1 , this holds as the other cutoffs vanish; where a neighbouring cutoff is also positive, the velocity added at level m m m vanishes, and by (a6) the transports of the coefficients of the two pulses agree. By Lemma 7.9, the chart turns ∂ t + b m ⋅ ∇ − κ m Δ − ∇ ⋅ A m \partial_t+\mathbf{b}_m\cdot\nabla-\kappa_m\Delta-\nabla\cdot\mathbf{A}_m ∂ t + b m ⋅ ∇ − κ m Δ − ∇ ⋅ A m into the operator on the left of (161) in the labels, and ∂ t \partial_t ∂ t at fixed label into ∂ t + b m − 1 ⋅ ∇ \partial_t+\mathbf{b}_{m-1}\cdot\nabla ∂ t + b m − 1 ⋅ ∇ at fixed x x x . On the part of the window outside the block of l m , k l_{m,k} l m , k , the velocity added at level m m m vanishes, and the proof of Lemma 7.9 applies verbatim. Lemma 13.2 applies to F ~ k \widetilde{F}_k F k on the interval I I I where [ t 1 , t 2 ] [t_1,t_2] [ t 1 , t 2 ] meets the window of k k k : its class follows from that of F F F , by ∂ t X m − 1 = b m − 1 ( t , X m − 1 ) \partial_tX_{m-1}=\mathbf{b}_{m-1}(t,X_{m-1}) ∂ t X m − 1 = b m − 1 ( t , X m − 1 ) and (54) . So Lemma 13.2, carried to x x x , is (165) with Corr m , k \operatorname{Corr}_{m,k} Corr m , k in place of Corr m \operatorname{Corr}_m Corr m , the means of the pulse k k k carried to x x x in place of M m \mathcal{M}_m M m , the cell tail of k k k , and without the last sum.
Step 2: Gluing. The cutoffs depend only on t t t , ∑ k H m , k = 1 \sum_k\mathrm{H}_{m,k}=1 ∑ k H m , k = 1 and ∑ k H m , k ′ = 0 \sum_k\mathrm{H}'_{m,k}=0 ∑ k H m , k ′ = 0 . So the operator of G G G without smoothing, applied to Corr m F = ∑ k H m , k Corr m , k F \operatorname{Corr}_m F=\sum_k\mathrm{H}_{m,k}\operatorname{Corr}_{m,k}F Corr m F = ∑ k H m , k Corr m , k F , is ∑ k H m , k \sum_k\mathrm{H}_{m,k} ∑ k H m , k times its application to Corr m , k F \operatorname{Corr}_{m,k}F Corr m , k F , plus ∑ k H m , k ′ ( Corr m , k F − F ) \sum_k\mathrm{H}'_{m,k}(\operatorname{Corr}_{m,k}F-F) ∑ k H m , k ′ ( Corr m , k F − F ) . By Definition 10.8, M m \mathcal{M}_m M m is the sum over k k k of H m , k \mathrm{H}_{m,k} H m , k times the means of the pulse k k k carried to x x x . By Step 1, the first sum gives (165) when, in the term of k k k , the means of k k k are replaced by M m \mathcal{M}_m M m . Their difference is nonzero only where the cutoffs of k k k and of a neighbour k ′ k' k ′ overlap, where 3 4 τ m ≤ ∣ t − k τ m ∣ ≤ 5 4 τ m \frac{3}{4}\tau_m\le|t-k\tau_m|\le\frac{5}{4}\tau_m 4 3 τ m ≤ ∣ t − k τ m ∣ ≤ 4 5 τ m , inside the quiet intervals. There the means of each pulse are the operator of G G G itself, read in its chart (Proposition 10.6(d)). Their part κ m K \kappa_m\mathbf{K} κ m K carries to the flux κ m ∇ F \kappa_m\nabla F κ m ∇ F in every chart, as ∇ y X m − 1 K m , l ( ∇ y X m − 1 ) T = I \nabla_yX_{m-1}\mathbf{K}_{m,l}(\nabla_yX_{m-1})^{\mathsf{T}}=I ∇ y X m − 1 K m , l ( ∇ y X m − 1 ) T = I by (53) , and consecutive charts describe the same coefficients of the sources, by (a6); so the replacement costs nothing. The field Corr m F \operatorname{Corr}_mF Corr m F is a finite sum of products of H m , k ( t ) \mathrm{H}_{m,k}(t) H m , k ( t ) , the coefficients V j , d \mathcal{V}_{j,d} V j , d , derivatives of the chart and ∇ y d F ~ k \nabla_y^d\widetilde{F}_k ∇ y d F k , each Lipschitz in t t t uniformly in x x x , which proves the last assertion of the lemma.
Step 3: The switching term. The derivative H m , k ′ \mathrm{H}'_{m,k} H m , k ′ is supported where 3 4 τ m ≤ ∣ t − k τ m ∣ ≤ 5 4 τ m \frac{3}{4}\tau_m\le|t-k\tau_m|\le\frac{5}{4}\tau_m 4 3 τ m ≤ ∣ t − k τ m ∣ ≤ 4 5 τ m , and ∣ H m , k ′ ∣ ≤ C / τ m |\mathrm{H}'_{m,k}|\le C/\tau_m ∣ H m , k ′ ∣ ≤ C / τ m . There the shear of k k k vanishes, since its envelope is supported where ∣ t − k τ m ∣ ≤ 2 3 τ m |t-k\tau_m|\le\frac{2}{3}\tau_m ∣ t − k τ m ∣ ≤ 3 2 τ m . On the side before k k k , the positive correctors of k k k vanish, as in Lemma 13.1, since the stretch from the start ( k − 61 48 ) τ m (k-\frac{61}{48})\tau_m ( k − 48 61 ) τ m of the window to ( k − 3 4 ) τ m (k-\frac{3}{4})\tau_m ( k − 4 3 ) τ m , enlarged by τ m 24 \frac{\tau_m}{24} 24 τ m , lies in the quiet interval between k − 2 k-2 k − 2 and k k k . On the side after k k k , for t ′ ≥ ( k + 17 24 ) τ m t'\ge(k+\frac{17}{24})\tau_m t ′ ≥ ( k + 24 17 ) τ m , [ t ′ − τ m 24 , t ′ + τ m 24 ] [t'-\frac{\tau_m}{24},t'+\frac{\tau_m}{24}] [ t ′ − 24 τ m , t ′ + 24 τ m ] lies in the quiet interval [ ( k + 2 3 ) τ m , ( k + 4 3 ) τ m ] [(k+\frac{2}{3})\tau_m,(k+\frac{4}{3})\tau_m] [( k + 3 2 ) τ m , ( k + 3 4 ) τ m ] , on which the velocity added at level m m m vanishes, as ζ \zeta ζ has compact support in ( − 2 3 , 2 3 ) (-\frac{2}{3},\frac{2}{3}) ( − 3 2 , 3 2 ) . So, at these times, the shear of k k k vanishes, the forms do not depend on z z z , by (a4), and neither do the conductivity and the means. As in the proof of Proposition 10.6(e), the terms of the brackets (107) with V 0 [ 0 ] = I \mathcal{V}_0[0]=I V 0 [ 0 ] = I then do not depend on z z z and give nothing, and the positive correctors solve a triangular system of homogeneous cell equations, with no datum except the lower correctors, the mode ξ ⋅ e ( k ) ⊥ \xi\cdot e^{\perp}_{(k)} ξ ⋅ e ( k ) ⊥ of each forced only by the modes ξ \xi ξ of the lower ones. On every transverse Fourier mode the cell inverse decays at a rate at least c κ m / ε m 2 c\kappa_m/\varepsilon_m^2 c κ m / ε m 2 , by the ellipticity of the conductivity (Lemma 10.4 ). After the time τ m 24 \frac{\tau_m}{24} 24 τ m the positive correctors are therefore bounded by C exp ( − c κ m τ m / ε m 2 ) C\exp(-c\kappa_m\tau_m/\varepsilon_m^2) C exp ( − c κ m τ m / ε m 2 ) times their sizes ((160) ); the polynomial factors of the triangular system are absorbed by lowering c c c . The supports of the H m , k ′ H'_{m,k} H m , k ′ in ( t 1 , t 2 ) (t_1,t_2) ( t 1 , t 2 ) are at most C ( 1 + ( t 2 − t 1 ) / τ m ) C(1+(t_2-t_1)/\tau_m) C ( 1 + ( t 2 − t 1 ) / τ m ) intervals of length at most τ m \tau_m τ m . So, by the Cauchy–Schwarz inequality in time, the norm of the last sum of ((165) ) in L 1 ( ( t 1 , t 2 ) ; L ω 2 ( T 2 ) ) L^1((t_1,t_2);L^2_{\omega}(\mathbb{T}^2)) L 1 (( t 1 , t 2 ) ; L ω 2 ( T 2 )) is at most C ( ( 1 + ( t 2 − t 1 ) / τ m ) / τ m ) 1 / 2 C((1+(t_2-t_1)/\tau_m)/\tau_m)^{1/2} C (( 1 + ( t 2 − t 1 ) / τ m ) / τ m ) 1/2 times the norm of the fields Corr m , k F − F \operatorname{Corr}_{m,k}F-F Corr m , k F − F in L 2 ( ( t 1 , t 2 ) ; L ω 2 ( T 2 ) ) L^2((t_1,t_2);L^2_{\omega}(\mathbb{T}^2)) L 2 (( t 1 , t 2 ) ; L ω 2 ( T 2 )) over these supports. As in Lemma 13.4 , for j ≥ 1 j\geq1 j ≥ 1 , ((160) ) times ρ ( m ) 1 − d κ m − 1 − 1 / 2 \rho(m)^{1-d}\kappa_{m-1}^{-1/2} ρ ( m ) 1 − d κ m − 1 − 1/2 is at most ρ ( m ) m − 1 − 1 2 ( ε m 2 / ( κ m h m ) ) 1 / 2 = ( ε m 2 / κ m ) 1 / 2 \rho(m)^{-\frac12}_{m-1}(\varepsilon_m^2/(\kappa_mh_m))^{1/2}=(\varepsilon_m^2/\kappa_m)^{1/2} ρ ( m ) m − 1 − 2 1 ( ε m 2 / ( κ m h m ) ) 1/2 = ( ε m 2 / κ m ) 1/2 , which gives ((166) ). Last, κ m τ m / ε m 2 ≥ c ε m − 1 − 4 δ \kappa_m\tau_m/\varepsilon_m^2\geq c\varepsilon_{m-1}^{-4\delta} κ m τ m / ε m 2 ≥ c ε m − 1 − 4 δ by Proposition 2.1 (c).
In Proposition 13.11 , the field Y ♭ Y^\flat Y ♭ of ((154) ) solves the averaged equation up to the divergence of a small flux F a v F^{\mathrm{av}} F av , which the corrector operator carries into the system of G G G . The correctors of positive degree give the second, small term of ((167) ), as in Lemma 13.4 . The first term is the cost of the identity part: the flux is bounded against the dissipation of level m − 1 m-1 m − 1 , but the norm ((156) ) uses κ m \kappa_m κ m . For m < M m<M m < M this cost is at most C ε m − 1 − q γ C\varepsilon_{m-1}^{-q\gamma} C ε m − 1 − q γ , which the term q γ q\gamma q γ in ((157) ) absorbs, and at m = M m=M m = M at most C C C , by Proposition 2.1 (c).
Lemma 13.6 (A coarse flux under the lift). There exists C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ C(\alpha,s,q,\zeta,\Lambda,n,t_0,T)<\infty C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ such that the following holds in the setting of the lift, with m m m above a level depending only on α , s , q , ζ , Λ , n , t 0 , T \alpha,s,q,\zeta,\Lambda,n,t_0,T α , s , q , ζ , Λ , n , t 0 , T . Let [ t 1 , t 2 ] ⊂ [ t A + 3 τ m , t f ] [t_1,t_2]\subset[t_A+3\tau_m,t_f] [ t 1 , t 2 ] ⊂ [ t A + 3 τ m , t f ] , and let F a v F^{\mathrm{av}} F av be a flux, a vector field for each component of G G G , on [ t 1 , t 2 ] [t_1,t_2] [ t 1 , t 2 ] , of class C N r e c + 1 C^{N_{\mathrm{rec}}+1} C N rec + 1 in x x x at each time, with derivatives measurable in ( t , x ) (t,x) ( t , x ) . Then there exists a forcing ( F c o , F c o , 0 ) (F^{\mathrm{co}},F^{\mathrm{co},0}) ( F co , F co , 0 ) such that Corr m ( ∇ ⋅ F a v ) = ∇ ⋅ F c o + F c o , 0 \operatorname{Corr}_m(\nabla\cdot F^{\mathrm{av}})=\nabla\cdot F^{\mathrm{co}}+F^{\mathrm{co},0} Corr m ( ∇ ⋅ F av ) = ∇ ⋅ F co + F co , 0 and such that its norm ((156) ) is at most
C [ ( κ m − 1 κ m ) 1 / 2 + ε m 2 κ m h m ( 1 + ( t 2 − t 1 ) 1 / 2 κ m 1 / 2 ε m ( ε m ρ ( m ) ) N c l ) ] κ m − 1 − 1 / 2 ∥ F a v ∥ L 2 ( ( t 1 , t 2 ) ; H ρ ( m ) , ω N r e c ( T 2 ) ) . (167) C\left[\left(\frac{\kappa_{m-1}}{\kappa_m}\right)^{1/2}+\frac{\varepsilon_m^2}{\kappa_mh_m}\left(1+(t_2-t_1)^{1/2}\frac{\kappa_m^{1/2}}{\varepsilon_m}\left(\frac{\varepsilon_m}{\rho(m)}\right)^{N_{\mathrm{cl}}}\right)\right]\kappa_{m-1}^{-1/2}\|F^{\mathrm{av}}\|_{L^2((t_1,t_2);H_{\rho(m),\omega}^{N_{\mathrm{rec}}}(\mathbb{T}^2))}. \tag*{(167)} C [ ( κ m κ m − 1 ) 1/2 + κ m h m ε m 2 ( 1 + ( t 2 − t 1 ) 1/2 ε m κ m 1/2 ( ρ ( m ) ε m ) N cl ) ] κ m − 1 − 1/2 ∥ F av ∥ L 2 (( t 1 , t 2 ) ; H ρ ( m ) , ω N rec ( T 2 )) . (167)
Proof. The identity part of Corr m \operatorname{Corr}_m Corr m gives ∇ ⋅ F a v \nabla\cdot F^{\mathrm{av}} ∇ ⋅ F av , whose flux contributes κ m − 1 / 2 ∥ F a v ∥ L 2 ( ( t 1 , t 2 ) ; L ω 2 ( T 2 ) ) = ( κ m − 1 / κ m ) 1 / 2 κ m − 1 − 1 / 2 ∥ F a v ∥ L 2 ( ( t 1 , t 2 ) ; L ω 2 ( T 2 ) ) \kappa_m^{-1/2}\|F^{\mathrm{av}}\|_{L^2((t_1,t_2);L^2_\omega(\mathbb{T}^2))}=(\kappa_{m-1}/\kappa_m)^{1/2}\kappa_{m-1}^{-1/2}\|F^{\mathrm{av}}\|_{L^2((t_1,t_2);L^2_\omega(\mathbb{T}^2))} κ m − 1/2 ∥ F av ∥ L 2 (( t 1 , t 2 ) ; L ω 2 ( T 2 )) = ( κ m − 1 / κ m ) 1/2 κ m − 1 − 1/2 ∥ F av ∥ L 2 (( t 1 , t 2 ) ; L ω 2 ( T 2 )) . A corrector of positive degree j j j has mean zero in z z z , so its action on ∇ ⋅ F a v \nabla\cdot F^{\mathrm{av}} ∇ ⋅ F av is reconstructed by ((164) ), as in Lemma 13.4 . The first flux is at most ε m \varepsilon_m ε m times ((160) ) times ∣ ∇ y d ( ( ∇ ⋅ F a v ) ∘ X m − 1 ) ∣ \lvert\nabla_y^d((\nabla\cdot F^{\mathrm{av}})\circ X_{m-1})\rvert ∣ ∇ y d (( ∇ ⋅ F av ) ∘ X m − 1 )∣ , whose norm in L 2 ( ( t 1 , t 2 ) ; L ω 2 ( T 2 ) ) L^2((t_1,t_2);L^2_\omega(\mathbb{T}^2)) L 2 (( t 1 , t 2 ) ; L ω 2 ( T 2 )) is at most C ρ ( m ) − d − 1 ∥ F a v ∥ L 2 ( ( t 1 , t 2 ) ; H ρ ( m ) , ω d + 1 ( T 2 ) ) C\rho(m)^{-d-1}\|F^{\mathrm{av}}\|_{L^2((t_1,t_2);H_{\rho(m),\omega}^{d+1}(\mathbb{T}^2))} C ρ ( m ) − d − 1 ∥ F av ∥ L 2 (( t 1 , t 2 ) ; H ρ ( m ) , ω d + 1 ( T 2 )) , by the chain rule and ((54) ). As κ m − 1 / 2 ε m ρ ( m ) − 1 = ( ε m 2 / ( κ m h m ) ) 1 / 2 κ m − 1 − 1 / 2 \kappa_m^{-1/2}\varepsilon_m\rho(m)^{-1}=(\varepsilon_m^2/(\kappa_mh_m))^{1/2}\kappa_{m-1}^{-1/2} κ m − 1/2 ε m ρ ( m ) − 1 = ( ε m 2 / ( κ m h m ) ) 1/2 κ m − 1 − 1/2 , it contributes at most
C ( ε m 2 κ m h m ) j + 1 2 ( ρ ( m ) ε m − 1 ) j − d ≤ C ε m 2 κ m h m C\left(\frac{\varepsilon_m^2}{\kappa_mh_m}\right)^{\frac{j+1}{2}}\left(\frac{\rho(m)}{\varepsilon_{m-1}}\right)^{j-d}\leq C\frac{\varepsilon_m^2}{\kappa_mh_m} C ( κ m h m ε m 2 ) 2 j + 1 ( ε m − 1 ρ ( m ) ) j − d ≤ C κ m h m ε m 2
times κ m − 1 − 1 / 2 ∥ F a v ∥ L 2 ( ( t 1 , t 2 ) ; H ρ ( m ) , ω d + 1 ( T 2 ) ) \kappa_{m-1}^{-1/2}\|F^{\mathrm{av}}\|_{L^2((t_1,t_2);H_{\rho(m),\omega}^{d+1}(\mathbb{T}^2))} κ m − 1 − 1/2 ∥ F av ∥ L 2 (( t 1 , t 2 ) ; H ρ ( m ) , ω d + 1 ( T 2 )) , since j ≥ 1 j\geq1 j ≥ 1 . The remainder after N c l N_{\mathrm{cl}} N cl steps is bounded as in Step 3 of Lemma 13.4 . as the covariance is a small multiple of ρ ( m ) 2 \rho(m)^2 ρ ( m ) 2 (Lemma 10.3(b)), each power of Γ m : ∇ 2 \Gamma_m : \nabla^2 Γ m : ∇ 2 gains ε δ m − 1 δ \varepsilon\delta_{m-1}^{\delta} ε δ m − 1 δ on fields smooth at the scale ρ ( m ) \rho(m) ρ ( m ) , at the cost of two derivatives.
Lemma 13.7 (Temporal conjugation). Let M ≥ 1 M \ge1 M ≥ 1 , κ ∈ K M \kappa\in\mathcal{K}_M κ ∈ K M and 2 ≤ m ≤ M 2 \le m \le M 2 ≤ m ≤ M .
(i) For every t ∈ R t \in\mathbb{R} t ∈ R , Γ m = Γ m ( t ) \Gamma_m = \Gamma_m(t) Γ m = Γ m ( t ) is a symmetric matrix independent of x x x , and, for every function F F F of class C 2 p + 2 C^{2p+2} C 2 p + 2 on T 2 \mathbb{T}^2 T 2 ,
[ b m − 1 ⋅ ∇ , Γ m : ∇ 2 ] F = ∇ ⋅ ( − ( ∇ b m − 1 Γ m + Γ m ( ∇ b m − 1 ) T ) ∇ F ) , (168) [\mathbf{b}_{m-1} \cdot\nabla, \Gamma_m : \nabla^2]F = \nabla\cdot\left(-(\nabla\mathbf{b}_{m-1}\Gamma_m+\Gamma_m(\nabla\mathbf{b}_{m-1})^{\mathsf{T}})\nabla F\right), \tag*{(168)} [ b m − 1 ⋅ ∇ , Γ m : ∇ 2 ] F = ∇ ⋅ ( − ( ∇ b m − 1 Γ m + Γ m ( ∇ b m − 1 ) T ) ∇ F ) , (168) where ( ∇ b m − 1 ) i i ′ : = ∂ i ′ ( b m − 1 ) i (\nabla\mathbf{b}_{m-1})_{ii'} := \partial_{i'}(\mathbf{b}_{m-1})_i ( ∇ b m − 1 ) i i ′ := ∂ i ′ ( b m − 1 ) i . Consequently the commutator with the truncated exponential is a divergence,
[ b m − 1 ⋅ ∇ , exp p ( Γ m : ∇ 2 ) ] F = ∇ ⋅ ( ∑ j = 1 p 1 j ! ∑ j ′ + j ′ ′ = j − 1 ( Γ m : ∇ 2 ) j ′ ( − ( ∇ b m − 1 Γ m + Γ m ( ∇ b m − 1 ) T ) ∇ ( Γ m : ∇ 2 ) j ′ ′ F ) ) , (169) [\mathbf{b}_{m-1}\cdot\nabla,\exp_p(\Gamma_m:\nabla^2)]F
= \nabla\cdot\left(\sum_{j=1}^{p}\frac{1}{j!}\sum_{j'+j''=j-1}(\Gamma_m:\nabla^2)^{j'}\left(-(\nabla\mathbf{b}_{m-1}\Gamma_m+\Gamma_m(\nabla\mathbf{b}_{m-1})^{\mathsf{T}})\nabla(\Gamma_m:\nabla^2)^{j''}F\right)\right), \tag*{(169)} [ b m − 1 ⋅ ∇ , exp p ( Γ m : ∇ 2 )] F = ∇ ⋅ j = 1 ∑ p j ! 1 j ′ + j ′′ = j − 1 ∑ ( Γ m : ∇ 2 ) j ′ ( − ( ∇ b m − 1 Γ m + Γ m ( ∇ b m − 1 ) T ) ∇ ( Γ m : ∇ 2 ) j ′′ F ) , (169) and we call the field in the divergence the flux of the commutator at F F F .
(ii) For every t 1 < t 2 t_1 < t_2 t 1 < t 2 and every field F F F on [ t 1 , t 2 ] [t_1,t_2] [ t 1 , t 2 ] of class C 2 p + 2 C^{2p+2} C 2 p + 2 in x x x at each time, with ∇ x a F \nabla_x^a F ∇ x a F , ∣ a ∣ ≤ 2 p + 1 |a| \le2p+1 ∣ a ∣ ≤ 2 p + 1 , Lipschitz in t t t , uniformly in x x x , there exists a null set of t t t , independent of x x x , off which
( ∂ t + b m − 1 ⋅ ∇ − J m κ m ( t ) : ∇ 2 ) exp p ( Γ m : ∇ 2 ) F = exp p ( Γ m : ∇ 2 ) ( ∂ t + b m − 1 ⋅ ∇ − κ m − 1 Δ ) F + [ b m − 1 ⋅ ∇ , exp p ( Γ m : ∇ 2 ) ] F − ( J m κ m ( t ) − κ m − 1 I ) : ∇ 2 ( Γ m : ∇ 2 ) p p ! F . (170) \left(\partial_t+\mathbf{b}_{m-1}\cdot\nabla-\mathbf{J}^{\kappa_m}_m(t):\nabla^2\right)\exp_p(\Gamma_m:\nabla^2)F
= \exp_p(\Gamma_m:\nabla^2)(\partial_t+\mathbf{b}_{m-1}\cdot\nabla-\kappa_{m-1}\Delta)F
+[\mathbf{b}_{m-1}\cdot\nabla,\exp_p(\Gamma_m:\nabla^2)]F
-\left(\mathbf{J}^{\kappa_m}_m(t)-\kappa_{m-1}I\right):\nabla^2\frac{(\Gamma_m:\nabla^2)^p}{p!}F. \tag*{(170)} ( ∂ t + b m − 1 ⋅ ∇ − J m κ m ( t ) : ∇ 2 ) exp p ( Γ m : ∇ 2 ) F = exp p ( Γ m : ∇ 2 ) ( ∂ t + b m − 1 ⋅ ∇ − κ m − 1 Δ ) F + [ b m − 1 ⋅ ∇ , exp p ( Γ m : ∇ 2 )] F − ( J m κ m ( t ) − κ m − 1 I ) : ∇ 2 p ! ( Γ m : ∇ 2 ) p F . (170) (iii) There exist Λ 0 ( α , s , q , ζ ) ≥ 128 \Lambda_0(\alpha,s,q,\zeta)\ge128 Λ 0 ( α , s , q , ζ ) ≥ 128 and, for every integer N ≥ 0 N\ge0 N ≥ 0 , a constant C ( α , s , q , ζ , p , N ) < ∞ C(\alpha,s,q,\zeta,p,N)<\infty C ( α , s , q , ζ , p , N ) < ∞ such that, for every Λ ≥ Λ 0 \Lambda\ge\Lambda_0 Λ ≥ Λ 0 , t ∈ R t\in\mathbb{R} t ∈ R and F ∈ H N + 4 p ( T 2 ) F\in H^{N+4p}(\mathbb{T}^2) F ∈ H N + 4 p ( T 2 ) (in the second bound, F ∈ H N + 2 p ( T 2 ) F\in H^{N+2p}(\mathbb{T}^2) F ∈ H N + 2 p ( T 2 ) ), with Γ m = Γ m ( t ) \Gamma_m=\Gamma_m(t) Γ m = Γ m ( t ) ,
∥ ( exp p ( Γ m : ∇ 2 ) exp p ( − Γ m : ∇ 2 ) − I ) F ∥ H ρ ( m ) N ( T 2 ) ≤ C ε δ m − 1 δ ( p + 1 ) ∥ F ∥ H ρ ( m ) N + 4 p ( T 2 ) , (171) \left\|\left(\exp_p(\Gamma_m:\nabla^2)\exp_p(-\Gamma_m:\nabla^2)-I\right)F\right\|_{H^N_{\rho(m)}(\mathbb{T}^2)}
\le C\varepsilon\delta_{m-1}^{\delta(p+1)}\|F\|_{H^{N+4p}_{\rho(m)}(\mathbb{T}^2)}, \tag*{(171)} ( exp p ( Γ m : ∇ 2 ) exp p ( − Γ m : ∇ 2 ) − I ) F H ρ ( m ) N ( T 2 ) ≤ C ε δ m − 1 δ ( p + 1 ) ∥ F ∥ H ρ ( m ) N + 4 p ( T 2 ) , (171) ∥ ( exp p ( ± Γ m : ∇ 2 ) − I ) F ∥ H ρ ( m ) N ( T 2 ) ≤ C ε δ m − 1 ∥ F ∥ H ρ ( m ) N + 2 p ( T 2 ) . \left\|\left(\exp_p(\pm\Gamma_m:\nabla^2)-I\right)F\right\|_{H^N_{\rho(m)}(\mathbb{T}^2)}
\le C\varepsilon\delta_{m-1}\|F\|_{H^{N+2p}_{\rho(m)}(\mathbb{T}^2)}. ( exp p ( ± Γ m : ∇ 2 ) − I ) F H ρ ( m ) N ( T 2 ) ≤ C ε δ m − 1 ∥ F ∥ H ρ ( m ) N + 2 p ( T 2 ) .
Proof. (i) The matrix Γ m ( t ) \Gamma_m(t) Γ m ( t ) is symmetric by ((39) ). With summation over the repeated coordinate indices i , i ′ , i ′ ′ i,i',i'' i , i ′ , i ′′ ,
[ b m − 1 ⋅ ∇ , Γ m : ∇ 2 ] F = − ( Γ m ) i i ′ ∂ i ∂ i ′ ( b m − 1 ) i ′ ′ ∂ i ′ ′ F − 2 ( Γ m ) i i ′ ′ ∂ i ( b m − 1 ) i ′ ∂ i ′ ∂ i ′ ′ F , [\mathbf{b}_{m-1}\cdot\nabla,\Gamma_m:\nabla^2]F
=-(\Gamma_m)_{ii'}\partial_i\partial_{i'}(\mathbf{b}_{m-1})_{i''}\partial_{i''}F
-2(\Gamma_m)_{ii''}\partial_i(\mathbf{b}_{m-1})_{i'}\partial_{i'}\partial_{i''}F, [ b m − 1 ⋅ ∇ , Γ m : ∇ 2 ] F = − ( Γ m ) i i ′ ∂ i ∂ i ′ ( b m − 1 ) i ′′ ∂ i ′′ F − 2 ( Γ m ) i i ′′ ∂ i ( b m − 1 ) i ′ ∂ i ′ ∂ i ′′ F ,
and expanding the divergence on the right side of ((168) ) gives the same two terms, as ∇ ⋅ b m − 1 = 0 \nabla\cdot\mathbf{b}_{m-1}=0 ∇ ⋅ b m − 1 = 0 and Γ m \Gamma_m Γ m is symmetric. The operators Γ m : ∇ 2 \Gamma_m:\nabla^2 Γ m : ∇ 2 and ∇ ⋅ \nabla\cdot ∇ ⋅ commute, and
[ b m − 1 ⋅ ∇ , ( Γ m : ∇ 2 ) j ] = ∑ j ′ + j ′ ′ = j − 1 ( Γ m : ∇ 2 ) j ′ [ b m − 1 ⋅ ∇ , Γ m : ∇ 2 ] ( Γ m : ∇ 2 ) j ′ ′ , [\mathbf{b}_{m-1}\cdot\nabla,(\Gamma_m:\nabla^2)^j]
=\sum_{j'+j''=j-1}(\Gamma_m:\nabla^2)^{j'}[\mathbf{b}_{m-1}\cdot\nabla,\Gamma_m:\nabla^2](\Gamma_m:\nabla^2)^{j''}, [ b m − 1 ⋅ ∇ , ( Γ m : ∇ 2 ) j ] = j ′ + j ′′ = j − 1 ∑ ( Γ m : ∇ 2 ) j ′ [ b m − 1 ⋅ ∇ , Γ m : ∇ 2 ] ( Γ m : ∇ 2 ) j ′′ ,
which gives ((169) ).
(ii) By ((39) ), Γ m \Gamma_m Γ m is Lipschitz in time with ∂ t Γ m = J m κ m − κ m − 1 I \partial_t\Gamma_m=\mathbf{J}^{\kappa_m}_m-\kappa_{m-1}I ∂ t Γ m = J m κ m − κ m − 1 I , and constant-coefficient operators commute (off a null set of t t t independent of x x x , the derivatives ∂ t ∇ a F \partial_t\nabla^aF ∂ t ∇ a F , ∣ a ∣ ≤ 2 p |a|\le2p ∣ a ∣ ≤ 2 p , exist at every x x x and equal ∇ a ∂ t F \nabla^a\partial_tF ∇ a ∂ t F , as in the proof of Lemma 13.2 ). So
∂ t exp p ( Γ m : ∇ 2 ) F = exp p ( Γ m : ∇ 2 ) ∂ t F + ( J m κ m − κ m − 1 I ) : ∇ 2 ( exp p ( Γ m : ∇ 2 ) − ( Γ m : ∇ 2 ) p p ! ) F . \partial_t\exp_p(\Gamma_m:\nabla^2)F
=\exp_p(\Gamma_m:\nabla^2)\partial_tF
+(\mathbf{J}^{\kappa_m}_m-\kappa_{m-1}I):\nabla^2\left(\exp_p(\Gamma_m:\nabla^2)-\frac{(\Gamma_m:\nabla^2)^p}{p!}\right)F. ∂ t exp p ( Γ m : ∇ 2 ) F = exp p ( Γ m : ∇ 2 ) ∂ t F + ( J m κ m − κ m − 1 I ) : ∇ 2 ( exp p ( Γ m : ∇ 2 ) − p ! ( Γ m : ∇ 2 ) p ) F .
Adding b m − 1 ⋅ ∇ \mathbf{b}_{m-1}\cdot\nabla b m − 1 ⋅ ∇ , which commutes with exp p ( Γ m : ∇ 2 ) \exp_p(\Gamma_m:\nabla^2) exp p ( Γ m : ∇ 2 ) up to (i), and subtracting J m κ m : ∇ 2 exp p ( Γ m : ∇ 2 ) F = exp p ( Γ m : ∇ 2 ) ( κ m − 1 Δ F ) + ( J m κ m − κ m − 1 I ) : ∇ 2 exp p ( Γ m : ∇ 2 ) F J_m^{\kappa_m}:\nabla^2\exp_p(\Gamma_m:\nabla^2)F=\exp_p(\Gamma_m:\nabla^2)(\kappa_{m-1}\Delta F)+(J_m^{\kappa_m}-\kappa_{m-1}I):\nabla^2\exp_p(\Gamma_m:\nabla^2)F J m κ m : ∇ 2 exp p ( Γ m : ∇ 2 ) F = exp p ( Γ m : ∇ 2 ) ( κ m − 1 Δ F ) + ( J m κ m − κ m − 1 I ) : ∇ 2 exp p ( Γ m : ∇ 2 ) F gives (170) .
(iii) The operator ( Γ m : ∇ 2 ) j (\Gamma_m:\nabla^2)^j ( Γ m : ∇ 2 ) j is a sum of 4 j 4^j 4 j derivatives of order 2 j 2j 2 j with coefficients of absolute value at most ∣ Γ m ∣ j |\Gamma_m|^j ∣ Γ m ∣ j , so Lemma 8.1(b) gives, for ρ > 0 \rho>0 ρ > 0 ,
∥ ( Γ m : ∇ 2 ) j F ∥ H ρ N ( T 2 ) ≤ ( 4 ∣ Γ m ∣ ρ 2 ) j ( N + 2 j ) ! N ! ∥ F ∥ H ρ N + 2 j ( T 2 ) \left\|(\Gamma_m:\nabla^2)^jF\right\|_{H_\rho^N(\mathbb{T}^2)}\leq\left(\frac{4|\Gamma_m|}{\rho^2}\right)^j\frac{(N+2j)!}{N!}\left\|F\right\|_{H_\rho^{N+2j}(\mathbb{T}^2)} ( Γ m : ∇ 2 ) j F H ρ N ( T 2 ) ≤ ( ρ 2 4∣ Γ m ∣ ) j N ! ( N + 2 j )! ∥ F ∥ H ρ N + 2 j ( T 2 )
In exp p ( Γ m : ∇ 2 ) exp p ( − Γ m : ∇ 2 ) − I \exp_p(\Gamma_m:\nabla^2)\exp_p(-\Gamma_m:\nabla^2)-I exp p ( Γ m : ∇ 2 ) exp p ( − Γ m : ∇ 2 ) − I the terms of powers 1 1 1 to p p p of Γ m : ∇ 2 \Gamma_m:\nabla^2 Γ m : ∇ 2 cancel, since they are those of exp ( Γ m : ∇ 2 ) exp ( − Γ m : ∇ 2 ) \exp(\Gamma_m:\nabla^2)\exp(-\Gamma_m:\nabla^2) exp ( Γ m : ∇ 2 ) exp ( − Γ m : ∇ 2 ) , which is the identity, and the powers p + 1 p+1 p + 1 to 2 p 2p 2 p remain. Lemma 10.3(b) gives ∣ Γ m ( t ) ∣ ≤ C ε m δ m − 1 ρ ( m ) 2 |\Gamma_m(t)|\leq C\varepsilon_m\delta_{m-1}\rho(m)^2 ∣ Γ m ( t ) ∣ ≤ C ε m δ m − 1 ρ ( m ) 2 .
The coupled system at the coarser level
The sources of the successor. Let G G G be as in Definition 11.3. Its successor G ′ G' G ′ has the components and weights of Definition 11.3(ii). Its sources carry the pieces of the operator
F ↦ ∇ ⋅ ( exp p ( − Γ m : ∇ 2 ) ( ( M m − J m κ m ∇ ) exp p ( Γ m : ∇ 2 ) F ) ) − exp p ( − Γ m : ∇ 2 ) [ b m − 1 ⋅ ∇ , exp p ( Γ m : ∇ 2 ) ] F (172) F\mapsto\nabla\cdot\left(\exp_p(-\Gamma_m:\nabla^2)\left((\mathcal{M}_m-J_m^{\kappa_m}\nabla)\exp_p(\Gamma_m:\nabla^2)F\right)\right)-\exp_p(-\Gamma_m:\nabla^2)\left[\mathbf{b}_{m-1}\cdot\nabla,\exp_p(\Gamma_m:\nabla^2)\right]F \tag*{(172)} F ↦ ∇ ⋅ ( exp p ( − Γ m : ∇ 2 ) ( ( M m − J m κ m ∇ ) exp p ( Γ m : ∇ 2 ) F ) ) − exp p ( − Γ m : ∇ 2 ) [ b m − 1 ⋅ ∇ , exp p ( Γ m : ∇ 2 ) ] F (172)
on fields F F F of G G G , which is in divergence form by (169) . The pieces are the following.
The defect letter: the part of M m \mathcal{M}_m M m on the paths without sources, of all orders, less J m κ m ∇ J_m^{\kappa_m}\nabla J m κ m ∇ (the defect of the averaged flux), conjugated as in (172) , together with the commutator term of (172) , whose flux is − exp p ( − Γ m : ∇ 2 ) -\exp_p(-\Gamma_m:\nabla^2) − exp p ( − Γ m : ∇ 2 ) applied to the flux of the commutator at F F F .
Old sources: the part of M m \mathcal{M}_m M m of degree 2 2 2 along the nonempty paths of sources (Definition 11.3(iii)); and the cell averages of the sources of G G G in their orders d ≥ 3 d\geq3 d ≥ 3 , carried to the charts of level m − 1 m-1 m − 1 , which are the part of M m \mathcal{M}_m M m along the paths of one source that involves no corrector.
Mean letters: for d ≥ 3 d\geq3 d ≥ 3 and i ≥ 1 i\geq1 i ≥ 1 , the part of M m \mathcal{M}_m M m of degree d d d along the paths of i i i sources; for i = 1 i=1 i = 1 , less the cell average of the source in the order d d d , which the old source carries.
Covariance letters: for d ≥ 2 d\geq2 d ≥ 2 , i ≥ 1 i\geq1 i ≥ 1 , and j , j ′ ≤ p j,j'\leq p j , j ′ ≤ p with j + j ′ ≥ 1 j+j'\geq1 j + j ′ ≥ 1 , the part of M m \mathcal{M}_m M m of degree d d d along the paths of i i i sources, times ( − Γ m : ∇ 2 ) j / j ! (-\Gamma_m:\nabla^2)^j/j! ( − Γ m : ∇ 2 ) j / j ! on the left and ( Γ m : ∇ 2 ) j ′ / j ′ ! (\Gamma_m:\nabla^2)^{j'}/j'! ( Γ m : ∇ 2 ) j ′ / j ′ ! on the right.
Their sum is (172) , since the sum over j , j ′ ≤ p j,j'\leq p j , j ′ ≤ p of a part multiplied in this way is that part multiplied by exp p ( − Γ m : ∇ 2 ) \exp_p(-\Gamma_m:\nabla^2) exp p ( − Γ m : ∇ 2 ) on the left and by exp p ( Γ m : ∇ 2 ) \exp_p(\Gamma_m:\nabla^2) exp p ( Γ m : ∇ 2 ) on the right. By item (1) of Subsection 11.1, the sources of G G G have current orders at most N s r c − 2 N c o r − 4 p − 4 N_{\mathrm{src}}-2N_{\mathrm{cor}}-4p-4 N src − 2 N cor − 4 p − 4 , so by (106) the means along their paths vanish above the degree N s r c − N c o r − 4 p − 4 N_{\mathrm{src}}-N_{\mathrm{cor}}-4p-4 N src − N cor − 4 p − 4 ; as these paths have at most N p a t h N_{\mathrm{path}} N path sources, every piece above has its letter in Definition 11.3(i). They are carried by the sources of Definition 11.3(iii)–(iv), with its words, smoothings and the power ε m − 1 − η \varepsilon_{m-1}^{-\eta} ε m − 1 − η of the coefficients of the new sources; the components of generation above N g e n N_{\mathrm{gen}} N gen or word length above N l e n N_{\mathrm{len}} N len are omitted, with the sources into them. In the charts of level m − 1 m-1 m − 1 these fluxes are the transports (11.8) of their two-scale forms. So the successor exists for every such G G G : its sources are explicit in the correctors and the means of Section 10, and no further cell problem is solved.
The successor at the scale ρ ( m ) \rho(m) ρ ( m ) . In the setting of Proposition 5.3, for m m m above a level that depends only on α , s , q , ζ , Λ , n , t 0 , T \alpha,s,q,\zeta,\Lambda,n,t_0,T α , s , q , ζ , Λ , n , t 0 , T , ρ ( m ) ≤ ρ m o l ( m − 1 ) \rho(m)\leq\rho_{\mathrm{mol}}(m-1) ρ ( m ) ≤ ρ mol ( m − 1 ) , as ρ ( m ) \rho(m) ρ ( m ) is below c m o l c_{\mathrm{mol}} c mol times each entry of the minimum (115) :
ρ ( m ) ≤ C ε m − 1 1 + γ / 2 + δ ≤ c m o l ε m − 1 , ρ ( m ) 2 = κ m − 1 h m ≤ 2 − 25 ε m − 1 2 δ κ m − 1 ε m − 1 2 ≤ c m o l 2 κ m − 1 ε m − 1 2 , \rho(m)\leq C\varepsilon_{m-1}^{1+\gamma/2+\delta}\leq c_{\mathrm{mol}}\varepsilon_{m-1},\qquad\rho(m)^2=\kappa_{m-1}h_m\leq2^{-25}\varepsilon_{m-1}^{2\delta}\kappa_{m-1}\varepsilon_{m-1}^2\leq c_{\mathrm{mol}}^2\kappa_{m-1}\varepsilon_{m-1}^2, ρ ( m ) ≤ C ε m − 1 1 + γ /2 + δ ≤ c mol ε m − 1 , ρ ( m ) 2 = κ m − 1 h m ≤ 2 − 25 ε m − 1 2 δ κ m − 1 ε m − 1 2 ≤ c mol 2 κ m − 1 ε m − 1 2 ,
by Proposition 2.1(c) and (15) , as c m o l c_{\mathrm{mol}} c mol is fixed before this level (Section 12). So Proposition 8.3 applies to G ′ G' G ′ on [ t p r , t f ] [t_{\mathrm{pr}},t_f] [ t pr , t f ] , at every order N ≤ N r e t N \le N_{\mathrm{ret}} N ≤ N ret and the scale ρ ( m ) \rho(m) ρ ( m ) , with the masses ω v / Ψ v 2 \omega_v/\Psi_v^2 ω v / Ψ v 2 (the marked norm) and with the masses ω v \omega_v ω v : its hypothesis (iii) holds by clause (d) of G ′ G' G ′ (N r e t ≤ N c o e f N_{\mathrm{ret}} \le N_{\mathrm{coef}} N ret ≤ N coef ), since the coupling matrix (68) increases with the scale, and (i) by (12.1) at the level m − 1 m-1 m − 1 ; hypothesis (ii) holds by clause (b1), whose analytic norm runs over every time of the window, through Cauchy’s estimate on the polydiscs of Step 6 of Subsection 11.6, the chains being smooth (Lemma 11.9). The datum of Y Y Y is the retained trace, with the weights of G G G on the components of G G G (Lemma 11.8(i)), so Proposition 12.1(iv) gives ∥ Y ( t p r ) ∥ H ρ ( m ) , G ′ N r e t ( T 2 ) ≤ C ∥ g ∥ L G 2 ( T 2 ) \|Y(t_{\mathrm{pr}})\|_{H_{\rho(m),G'}^{N_{\mathrm{ret}}}(\mathbb{T}^2)} \le C\|g\|_{L_G^2(\mathbb{T}^2)} ∥ Y ( t pr ) ∥ H ρ ( m ) , G ′ N ret ( T 2 ) ≤ C ∥ g ∥ L G 2 ( T 2 ) . Lemma 13.9 and Proposition 13.10 use these facts, as m 0 m_0 m 0 is at least this level.
Lemma 13.8 checks the construction: summed with the readout weights, the equations of the components of G ′ G' G ′ give the equation of level m − 1 m-1 m − 1 for Υ m Y \Upsilon_mY Υ m Y , forced by (172) , up to the fluxes of the smoothings and of the omitted components.
Lemma 13.8 (The successor read out). In the setting of Proposition 5.3, let Y Y Y be a solution of the system of G ′ G' G ′ on [ t p r , t f ] [t_{\mathrm{pr}},t_f] [ t pr , t f ] in C ( [ t p r , t f ] ; H N r e t + 3 ( T 2 ) V ′ ) C([t_{\mathrm{pr}},t_f];H^{N_{\mathrm{ret}}+3}(\mathbb{T}^2)^{V'}) C ([ t pr , t f ] ; H N ret + 3 ( T 2 ) V ′ ) . Then
( ∂ t + b m − 1 ⋅ ∇ − κ m − 1 Δ ) Υ m Y = ∇ ⋅ ( exp p ( − Γ m : ∇ 2 ) ( M m − J m k ∇ ) exp p ( Γ m : ∇ 2 ) Υ m Y ) − exp p ( − Γ m : ∇ 2 ) [ b m − 1 ⋅ ∇ , exp p ( Γ m : ∇ 2 ) ] Υ m Y − ∇ ⋅ ( F s m + F f r ) (173) (\partial_t+\mathbf b_{m-1}\cdot\nabla-\kappa_{m-1}\Delta)\Upsilon_mY
=
\nabla\cdot\left(\exp_{\mathrm p}(-\Gamma_m:\nabla^2)\left(\mathcal M_m-\mathcal J_m^{\mathrm k}\nabla\right)\exp_{\mathrm p}(\Gamma_m:\nabla^2)\Upsilon_mY\right)
-\exp_{\mathrm p}(-\Gamma_m:\nabla^2)\left[\mathbf b_{m-1}\cdot\nabla,\exp_{\mathrm p}(\Gamma_m:\nabla^2)\right]\Upsilon_mY
-\nabla\cdot(F^{\mathrm{sm}}+F^{\mathrm{fr}}) \tag*{(173)} ( ∂ t + b m − 1 ⋅ ∇ − κ m − 1 Δ ) Υ m Y = ∇ ⋅ ( exp p ( − Γ m : ∇ 2 ) ( M m − J m k ∇ ) exp p ( Γ m : ∇ 2 ) Υ m Y ) − exp p ( − Γ m : ∇ 2 ) [ b m − 1 ⋅ ∇ , exp p ( Γ m : ∇ 2 ) ] Υ m Y − ∇ ⋅ ( F sm + F fr ) (173) in the weak sense of (67) . Here F s m F^{\mathrm{sm}} F sm is the smoothing defect: the sum, over the pieces of (172) , of the flux of each piece acting on ( I − S σ ) Υ m Y (I-S_\sigma)\Upsilon_mY ( I − S σ ) Υ m Y , with S σ S_\sigma S σ the smoothing of the source of the construction that carries it, before the omission. And F f r F^{\mathrm{fr}} F fr is the frontier flux: the readout of the flux of the construction into the omitted components, from the retained ones.
Proof. By the readout formula, the readout weight of a component v ′ v' v ′ over a component v v v of G G G is ε m − 1 η # v ′ \varepsilon_{m-1}^{\eta\#v'} ε m − 1 η # v ′ (# v ′ = 0 \#v'=0 # v ′ = 0 for v ′ = v v'=v v ′ = v ). We multiply the equation of each component over v v v by its readout weight and sum. Old sources act between components with the same word, with the same coefficient and smoothing, so they sum to the old pieces acting on S σ Υ m Y S_\sigma\Upsilon_mY S σ Υ m Y . A letter source (a new source of Definition 11.3(iv)), with the same coefficients and smoothing for every word of its origin, runs from a component over a component v ′ ′ v'' v ′′ of G G G to the component v ′ v' v ′ over v v v whose word is that of the first followed by the letter, so the word of its origin has length # v ′ − 1 \#v'-1 # v ′ − 1 . Its coefficients carry ε m − 1 − η \varepsilon_{m-1}^{-\eta} ε m − 1 − η , so
ε m − 1 η # v ′ ε m − 1 − η = ε m − 1 η ( # v ′ − 1 ) , \varepsilon_{m-1}^{\eta\#v'}\varepsilon_{m-1}^{-\eta}
=
\varepsilon_{m-1}^{\eta(\#v'-1)}, ε m − 1 η # v ′ ε m − 1 − η = ε m − 1 η ( # v ′ − 1 ) ,
the readout weight of its origin, and the letter sources sum to the letter pieces acting on S σ Υ m Y S_\sigma\Upsilon_mY S σ Υ m Y , as long as their targets are retained. The sources into omitted components are missing from the sum, and their readout is F f r F^{\mathrm{fr}} F fr . Writing S σ Υ m Y = Υ m Y − ( I − S σ ) Υ m Y S_\sigma\Upsilon_mY=\Upsilon_mY-(I-S_\sigma)\Upsilon_mY S σ Υ m Y = Υ m Y − ( I − S σ ) Υ m Y and summing the pieces into (172) gives (173) . As Y ∈ C ( [ t p r , t f ] ; H N r e t + 3 ( T 2 ) V ′ ) Y\in C([t_{\mathrm{pr}},t_f];H^{N_{\mathrm{ret}}+3}(\mathbb{T}^2)^{V'}) Y ∈ C ([ t pr , t f ] ; H N ret + 3 ( T 2 ) V ′ ) , the identities (67) of the components, combined with the readout weights, give (173) in that sense. □ \square □
Lemma 13.9 bounds the flux into the omitted components, in the norm in which Lemma 13.6 reads a coarse flux. Those of generation N g e n + 1 N_{\mathrm{gen}}+1 N gen + 1 have markers at least ε m − 1 − η ( N g e n + 1 ) / 8 \varepsilon_{m-1}^{-\eta(N_{\mathrm{gen}}+1)/8} ε m − 1 − η ( N gen + 1 ) /8 , and the solution is bounded in the marked norm: the first term of (174) . Those with words of length N l e n + 1 N_{\mathrm{len}}+1 N len + 1 are reached through N l e n + 1 N_{\mathrm{len}}+1 N len + 1 letters, each gaining ε m − 1 δ / 2 \varepsilon_{m-1}^{\delta/2} ε m − 1 δ /2 in the coupling: the second.
Lemma 13.9 (The omitted components). There exists C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ C(\alpha,s,q,\zeta,\Lambda,n,t_0,T)<\infty C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ such that, in the setting of Proposition 5.3, for every datum g g g and every integer 0 ≤ N ≤ N r e t 0\le N\le N_{\mathrm{ret}} 0 ≤ N ≤ N ret ,
κ m − 1 − 1 / 2 ∥ F f r ∥ L 2 ( ( t p r , t f ) ; H ρ ( m ) , ω N ( T 2 ) ) ≤ C ( ε m − 1 η ( N g e n + 1 ) / 8 + ε m − 1 δ ( N l e n + 1 ) / 2 ) ∥ g ∥ L G 2 ( T 2 ) . (174) \kappa_{m-1}^{-1/2}\|F^{\mathrm{fr}}\|_{L^2((t_{\mathrm{pr}},t_f);H_{\rho(m),\omega}^{N}(\mathbb{T}^2))}
\le
C\left(\varepsilon_{m-1}^{\eta(N_{\mathrm{gen}}+1)/8}
+\varepsilon_{m-1}^{\delta(N_{\mathrm{len}}+1)/2}\right)\|g\|_{L_G^2(\mathbb{T}^2)}. \tag*{(174)} κ m − 1 − 1/2 ∥ F fr ∥ L 2 (( t pr , t f ) ; H ρ ( m ) , ω N ( T 2 )) ≤ C ( ε m − 1 η ( N gen + 1 ) /8 + ε m − 1 δ ( N len + 1 ) /2 ) ∥ g ∥ L G 2 ( T 2 ) . (174)
Proof. The datum of Y Y Y satisfies ∥ Y ( t p r ) ∥ H ρ ( m ) , G ′ N ( T 2 ) ≤ C ∥ g ∥ L G 2 ( T 2 ) \|Y(t_{\mathrm{pr}})\|_{H_{\rho(m),G'}^{N}(\mathbb{T}^{2})} \le C\|g\|_{L_{G}^{2}(\mathbb{T}^{2})} ∥ Y ( t pr ) ∥ H ρ ( m ) , G ′ N ( T 2 ) ≤ C ∥ g ∥ L G 2 ( T 2 ) for 0 ≤ N ≤ N r e t 0 \le N \le N_{\mathrm{ret}} 0 ≤ N ≤ N ret , by the paragraph on the successor at the scale ρ ( m ) \rho(m) ρ ( m ) , as the norm increases with N N N .
Omitted generation. A child v ′ v' v ′ of generation N g e n + 1 N_{\mathrm{gen}}+1 N gen + 1 , an omitted component, was created by N g e n + 1 N_{\mathrm{gen}}+1 N gen + 1 cycles, at levels m ′ ≥ m m' \ge m m ′ ≥ m . By (119) , applied before the omission, its marker is that of its root, at least 1 1 1 , times ε m ′ − 1 − η / 8 ≥ ε m − 1 − η / 8 \varepsilon_{m'-1}^{-\eta/8} \ge\varepsilon_{m-1}^{-\eta/8} ε m ′ − 1 − η /8 ≥ ε m − 1 − η /8 for each of these cycles, so at least ε m − 1 − η ( N g e n + 1 ) / 8 \varepsilon_{m-1}^{-\eta(N_{\mathrm{gen}}+1)/8} ε m − 1 − η ( N gen + 1 ) /8 . The sources of the construction into v ′ v' v ′ , before the omission, have row and column sums at most C C C , at the order N N N and the scale ρ ( m ) \rho(m) ρ ( m ) , in the coupling matrix for the masses ω v / Ψ v 2 \omega_v/\Psi_v^2 ω v / Ψ v 2 . Indeed, Steps 3 and 8 of the proof of Lemma 11.7 use Definition 11.3 only through the masses and markers of the copies and the bound of a new source by a fixed number times the product along its path, which hold before the omission. So do Steps 4 and 5, as a child of generation N g e n + 1 N_{\mathrm{gen}}+1 N gen + 1 has reserve 1 / ( N g e n + 2 + N o f f ) > 0 1/(N_{\mathrm{gen}}+2+N_{\mathrm{off}})>0 1/ ( N gen + 2 + N off ) > 0 ; so (b1) holds for the sources into it, and Step 8 reads their derivatives by Cauchy’s estimate on the margin of this reserve. Lemma 8.4, with ϑ = C \vartheta=C ϑ = C , bounds their flux, measured with the mass and the marker of v ′ v' v ′ , by C κ m − 1 C\kappa_{m-1} C κ m − 1 times the norm of ∇ Y \nabla Y ∇ Y in L 2 ( ( t p r , t f ) ; H ρ ( m ) , G ′ N ( T 2 ) ) L^{2}((t_{\mathrm{pr}},t_{\mathrm{f}});H_{\rho(m),G'}^{N}(\mathbb{T}^{2})) L 2 (( t pr , t f ) ; H ρ ( m ) , G ′ N ( T 2 )) . Dividing by the marker of v ′ v' v ′ gives the first term of (174) ; the readout weight and the ratio of the masses of v ′ v' v ′ and of the component of G G G whose child it is cost at most C C C . Proposition 6.2 for G ′ G' G ′ with the masses ω v / Ψ v 2 \omega_v/\Psi_v^2 ω v / Ψ v 2 bounds the norm of ∇ Y \nabla Y ∇ Y in the marked norm by κ m − 1 − 1 / 2 ∥ Y ( t p r ) ∥ L G 2 ( T 2 ) \kappa_{m-1}^{-1/2}\|Y(t_{\mathrm{pr}})\|_{L_G^2(\mathbb{T}^2)} κ m − 1 − 1/2 ∥ Y ( t pr ) ∥ L G 2 ( T 2 ) , and at order N N N Proposition 8.3(b) does so, by the paragraph on the successor at the scale ρ ( m ) \rho(m) ρ ( m ) .
Omitted length. The components of G ′ G' G ′ whose words have length j j j form a family, the layer of length j j j , whose coupling matrix is a principal submatrix of that of G ′ G' G ′ , with sums at most 1 / 8 1/8 1/8 ; the sources within a word are old sources, and those from the layer of length j − 1 j-1 j − 1 letter sources. For j ≥ 1 j \ge1 j ≥ 1 its datum is zero, and its forcing is the flux of the letter sources from the layer of length j − 1 j-1 j − 1 . By Steps 3 and 8 of the proof of Lemma 11.7, for each letter the entries of the coupling matrix of these sources into a component, or out of one, add up to at most C ε m − 1 δ − 11 η / 8 − ( N s r c − 2 ) Θ ≤ C ε m − 1 45 δ / 64 ≤ C ε m − 1 δ / 2 C\varepsilon_{m-1}^{\delta-11\eta/8-(N_{\mathrm{src}}-2)\Theta} \le C\varepsilon_{m-1}^{45\delta/64} \le C\varepsilon_{m-1}^{\delta/2} C ε m − 1 δ − 11 η /8 − ( N src − 2 ) Θ ≤ C ε m − 1 45 δ /64 ≤ C ε m − 1 δ /2 (as η ≤ δ / 8 \eta\le\delta/8 η ≤ δ /8 by (5) , and ( N s r c − 2 ) Θ ≤ δ / 8 (N_{\mathrm{src}}-2)\Theta\le\delta/8 ( N src − 2 ) Θ ≤ δ /8 ), the factor ε m − 1 − η \varepsilon_{m-1}^{-\eta} ε m − 1 − η and the mass and marker ratios included. So Proposition 8.3(a), with Lemma 8.4 at ϑ = C ε m − 1 δ / 2 \vartheta=C\varepsilon_{m-1}^{\delta/2} ϑ = C ε m − 1 δ /2 , bounds the dissipation κ m − 1 1 / 2 ∥ ∇ Y ∥ L 2 ( ( t p r , t f ) ; H ρ ( m ) , G ′ N ( T 2 ) ) \kappa_{m-1}^{1/2}\|\nabla Y\|_{L^{2}((t_{\mathrm{pr}},t_{\mathrm{f}});H_{\rho(m),G'}^{N}(\mathbb{T}^{2}))} κ m − 1 1/2 ∥∇ Y ∥ L 2 (( t pr , t f ) ; H ρ ( m ) , G ′ N ( T 2 )) of the components of Y Y Y in the layer of length j ≥ 1 j \ge1 j ≥ 1 by C ε m − 1 δ / 2 C\varepsilon_{m-1}^{\delta/2} C ε m − 1 δ /2 times that of the layer of length j − 1 j-1 j − 1 , with the marked norms (35) . One more letter, from the layer of length N l e n N_{\mathrm{len}} N len , bounds the flux into the omitted layer of length N l e n + 1 N_{\mathrm{len}}+1 N len + 1 : in all, N l e n + 1 N_{\mathrm{len}}+1 N len + 1 letters. The readout weight costs at most C C C . □
Bounds on the residual of the two-scale expansion
Lemmas 13.4–13.6 bound the residual of the lift by energies of the fields they act on; Proposition 13.10, with Lemma 13.7(iii), bounds these and the energy of the lift. By item (i) of Proposition 13.10, the solution of the successor keeps the N r e t N_{\mathrm{ret}} N ret derivatives at the scale ρ ( m ) \rho(m) ρ ( m ) of its datum. By item (ii), the lift has derivatives at the scale ρ m o l ( m ) \rho_{\mathrm{mol}}(m) ρ mol ( m ) , with the dissipation κ m \kappa_m κ m : a fast derivative of a corrector costs ε m − 1 \varepsilon_m^{-1} ε m − 1 , against κ m 1 / 2 \kappa_m^{1/2} κ m 1/2 in the dissipation, and one degree offsets it.
Proposition 13.10 (Energy of the successor and of the lift). There exists C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ C(\alpha,s,q,\zeta,\Lambda,n,t_0,T)<\infty C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ such that, in the setting of Proposition 5.3, for every datum g g g of G G G the following hold.
(i) The solution Y Y Y lies in C ( [ t p r , t f ] ; H ρ ( m ) , G ′ N r e t + 3 ( T 2 ) V ) C([t_{\mathrm{pr}},t_{\mathrm{f}}];H_{\rho(m),G'}^{N_{\mathrm{ret}+3}}(\mathbb{T}^{2})^{V}) C ([ t pr , t f ] ; H ρ ( m ) , G ′ N ret + 3 ( T 2 ) V ) , and
∥ Y ∥ L ∞ ( [ t p r , t f ] ; H ρ ( m ) , G ′ N r e t ( T 2 ) ) ≤ C ∥ g ∥ L G 2 ( T 2 ) ; \|Y\|_{L^\infty([t_{\mathrm{pr}},t_{\mathrm{f}}];H_{\rho(m),G'}^{N_{\mathrm{ret}}}(\mathbb{T}^{2}))} \le C\|g\|_{L_G^2(\mathbb{T}^{2})}; ∥ Y ∥ L ∞ ([ t pr , t f ] ; H ρ ( m ) , G ′ N ret ( T 2 )) ≤ C ∥ g ∥ L G 2 ( T 2 ) ; and the energy of T m Y \mathcal{T}_mY T m Y of order N r e t N_{\mathrm{ret}} N ret on [ t p r , t f ] [t_{\mathrm{pr}},t_{\mathrm{f}}] [ t pr , t f ] is at most C ∥ g ∥ L G 2 ( T 2 ) C\|g\|_{L_G^2(\mathbb{T}^{2})} C ∥ g ∥ L G 2 ( T 2 ) .
(ii) For every integer 0 ≤ N ≤ N r e t − N c o r − 2 ρ 0 \le N \le N_{\mathrm{ret}}-N_{\mathrm{cor}}-2\rho 0 ≤ N ≤ N ret − N cor − 2 ρ , the component energy of Y ♯ Y^\sharp Y ♯ of order N N N at the
scale ρ m o l ( m ) \rho_{\mathrm{mol}}(m) ρ mol ( m ) , with the diffusivity κ m \kappa_m κ m , on [ t p r , t f ] [t_{\mathrm{pr}},t_{\mathrm{f}}] [ t pr , t f ] , is at most C ∥ g ∥ L G 2 ( T 2 ) C\lVert g\rVert_{L^2_G(\mathbb{T}^2)} C ∥ g ∥ L G 2 ( T 2 ) . In particular, for every t ∈ [ t p r , t f ] t\in[t_{\mathrm{pr}},t_{\mathrm{f}}] t ∈ [ t pr , t f ] , ∥ Y ♯ ( t ) ∥ H ρ m o l ( m ) , ω 2 N r e p + 4 ( T 2 ) ≤ C ∥ g ∥ L G 2 ( T 2 ) \lVert Y^\sharp(t)\rVert_{H^{2N_{\mathrm{rep}}+4}_{\rho_{\mathrm{mol}}(m),\omega}(\mathbb{T}^2)}\le C\lVert g\rVert_{L^2_G(\mathbb{T}^2)} ∥ Y ♯ ( t ) ∥ H ρ mol ( m ) , ω 2 N rep + 4 ( T 2 ) ≤ C ∥ g ∥ L G 2 ( T 2 ) .
Proof. (i) By the paragraph on the successor at the scale ρ ( m ) \rho(m) ρ ( m ) , Proposition 8.3 (b) with the marked masses gives the bound on Y Y Y and the continuity in H N r e t ( T 2 ) V ′ H^{N_{\mathrm{ret}}}(\mathbb{T}^2)^{V'} H N ret ( T 2 ) V ′ , and (c) with the masses ω v \omega_v ω v gives the component energy of Y Y Y , since the markers are at least 1. By Lemma 11.8 (iii), the readout has norm at most 1 + N w o r d 1 / 2 ε m − 1 5 η / 4 ≤ 2 1+N_{\mathrm{word}}^{1/2}\varepsilon_{m-1}^{5\eta/4}\le2 1 + N word 1/2 ε m − 1 5 η /4 ≤ 2 in the coordinates scaled by ω v − 1 / 2 \omega_v^{-1/2} ω v − 1/2 , with nonnegative entries; applied to the component norms, it bounds the component energy of Υ m Y \Upsilon_mY Υ m Y by twice that of Y Y Y .
The solution Y Y Y lies in C ( [ t p r , t f ] ; H N r e t + 3 ( T 2 ) V ′ ) C([t_{\mathrm{pr}},t_{\mathrm{f}}];H^{N_{\mathrm{ret}}+3}(\mathbb{T}^2)^{V'}) C ([ t pr , t f ] ; H N ret + 3 ( T 2 ) V ′ ) , by Proposition 8.3 at the order N r e t + 3 N_{\mathrm{ret}}+3 N ret + 3 and a scale ρ ′ ≤ ρ ( m ) \rho'\le\rho(m) ρ ′ ≤ ρ ( m ) small enough, and the uniqueness at the order 0. Indeed, the datum lies in H N o u t − 2 p ( T 2 ) V ′ H^{N_{\mathrm{out}}-2p}(\mathbb{T}^2)^{V'} H N out − 2 p ( T 2 ) V ′ , with N o u t − 2 p = N r e t + 2 p ≥ N r e t + 3 N_{\mathrm{out}}-2p=N_{\mathrm{ret}}+2p\ge N_{\mathrm{ret}}+3 N out − 2 p = N ret + 2 p ≥ N ret + 3 (p ≥ 2 p\ge2 p ≥ 2 by (157) ); hypothesis (i) holds at ρ ′ \rho' ρ ′ as at ρ ( m ) \rho(m) ρ ( m ) ; (ii) holds as the coefficients of G ′ G' G ′ have bounded derivatives of every order, by Cauchy’s estimate on its forms at positive reserve (Steps 4 and 6 of Subsection 11.6) and Lemma 11.9; and (iii) holds as, when ρ ′ → 0 \rho'\to0 ρ ′ → 0 , the coupling sums of order N r e t + 3 N_{\mathrm{ret}}+3 N ret + 3 tend to those at the order 0, below 1 8 \frac{1}{8} 8 1 .
(ii) By Lemma 13.7 (iii), the energy of Y ♭ Y^\flat Y ♭ of order N + N c o r N+N_{\mathrm{cor}} N + N cor is at most C C C times that of Υ m Y \Upsilon_mY Υ m Y of order N + N c o r + 2 p N+N_{\mathrm{cor}}+2p N + N cor + 2 p . In Y ♯ = Corr m Y ♭ Y^\sharp=\operatorname{Corr}_mY^\flat Y ♯ = Corr m Y ♭ , we consider the part of a corrector of degree j ≥ 1 j\ge1 j ≥ 1 on the paths without sources, acting on derivatives of order d d d of Y ♭ Y^\flat Y ♭ , with 1 ≤ d ≤ j 1\le d\le j 1 ≤ d ≤ j (the order 0 vanishes, as these correctors annihilate constants). In the dissipation, the costly derivative is the fast one, ε m − 1 − 1 \varepsilon_{m-1}^{-1} ε m − 1 − 1 , against κ m 1 / 2 \kappa_m^{1/2} κ m 1/2 . Its contribution to κ m 1 / 2 ∥ ∇ Y ♯ ∥ L 2 ( ( t p r , t f ) ; H ρ m o l ( m ) , ω N ( T 2 ) ) \kappa_m^{1/2}\lVert\nabla Y^\sharp\rVert_{L^2((t_{\mathrm{pr}},t_{\mathrm{f}});H^N_{\rho_{\mathrm{mol}}(m),\omega}(\mathbb{T}^2))} κ m 1/2 ∥ ∇ Y ♯ ∥ L 2 (( t pr , t f ) ; H ρ mol ( m ) , ω N ( T 2 )) is at most
C κ m 1 / 2 ε m − 1 − 1 ρ ( m ) κ m − 1 − 1 / 2 ( ε m 2 κ m h m ) j / 2 ( ρ ( m ) ε m − 1 ) j − d = C ( ε m 2 κ m h m ) ( j − 1 ) / 2 ( ρ ( m ) ε m − 1 ) j − d ≤ C C\kappa_m^{1/2}\varepsilon_{m-1}^{-1}\rho(m)\kappa_{m-1}^{-1/2}\left(\frac{\varepsilon_m^2}{\kappa_mh_m}\right)^{j/2}\left(\frac{\rho(m)}{\varepsilon_{m-1}}\right)^{j-d}
=
C\left(\frac{\varepsilon_m^2}{\kappa_mh_m}\right)^{(j-1)/2}\left(\frac{\rho(m)}{\varepsilon_{m-1}}\right)^{j-d}
\le C C κ m 1/2 ε m − 1 − 1 ρ ( m ) κ m − 1 − 1/2 ( κ m h m ε m 2 ) j /2 ( ε m − 1 ρ ( m ) ) j − d = C ( κ m h m ε m 2 ) ( j − 1 ) /2 ( ε m − 1 ρ ( m ) ) j − d ≤ C
times the energy of Y ♭ Y^\flat Y ♭ , by (160) and the exact relation κ m 1 / 2 ε m − 1 − 1 ρ ( m ) κ m − 1 − 1 / 2 = ρ ( m ) / ε ˉ m = ( κ m h m / ε m 2 ) 1 / 2 \kappa_m^{1/2}\varepsilon_{m-1}^{-1}\rho(m)\kappa_{m-1}^{-1/2}=\rho(m)/\bar{\varepsilon}_m=(\kappa_mh_m/\varepsilon_m^2)^{1/2} κ m 1/2 ε m − 1 − 1 ρ ( m ) κ m − 1 − 1/2 = ρ ( m ) / ε ˉ m = ( κ m h m / ε m 2 ) 1/2 of (159) . A derivative that falls on ∇ d Y ♭ \nabla^dY^\flat ∇ d Y ♭ in place of the fast variable costs instead ( κ m / κ m − 1 ) 1 / 2 ( ε m 2 / ( κ m h m ) ) j / 2 ( ρ ( m ) / ε m − 1 ) j − d ≤ C (\kappa_m/\kappa_{m-1})^{1/2}(\varepsilon_m^2/(\kappa_mh_m))^{j/2}(\rho(m)/\varepsilon_{m-1})^{j-d}\le C ( κ m / κ m − 1 ) 1/2 ( ε m 2 / ( κ m h m ) ) j /2 ( ρ ( m ) / ε m − 1 ) j − d ≤ C times the energy; the transport of the coefficients to x x x by (54) lowers the order d d d at the cost ε m − 1 − 1 \varepsilon_{m-1}^{-1} ε m − 1 − 1 per order, which keeps the form of (160) ; and ρ m o l ( m ) ≤ c m o l ε m ≤ C ρ ( m ) \rho_{\mathrm{mol}}(m)\le c_{\mathrm{mol}}\varepsilon_m\le C\rho(m) ρ mol ( m ) ≤ c mol ε m ≤ C ρ ( m ) . The supremum in time is at most C ( ε m 2 / ( κ m h m ) ) j / 2 C(\varepsilon_m^2/(\kappa_mh_m))^{j/2} C ( ε m 2 / ( κ m h m ) ) j /2 times the energy. Derivatives at the scale ρ m o l ( m ) \rho_{\mathrm{mol}}(m) ρ mol ( m ) cost at most C C C , and the identity part of Corr m \operatorname{Corr}_m Corr m costs ( κ m / κ m − 1 ) 1 / 2 ≤ 1 (\kappa_m/\kappa_{m-1})^{1/2}\le1 ( κ m / κ m − 1 ) 1/2 ≤ 1 in the dissipation. The paths with sources are bounded in the same way, the sums over the paths by (b3). With 2 N r e p + 4 ≤ N r e t − N c o r − 2 p 2N_{\mathrm{rep}}+4\le N_{\mathrm{ret}}-N_{\mathrm{cor}}-2p 2 N rep + 4 ≤ N ret − N cor − 2 p by (180) , this gives the last bound.
The square marker.
In Proposition 13.11, the lift starts from the preparation, up to the two truncated exponentials (175) , since it adds no corrector at the preparation time (Lemma 13.1 ). Its residual has four terms: the coarse flux (Lemma 13.6), the cell tail, the switching term, and the fine smoothing defect (the error of the smoothings of G G G ); Table 13.1 lists their bounds and the origin of their smallness.
Table 13.1. The residual of the lift (Proposition 13.11) in the norm (156) , the coarse flux split as in Step 3, each part with its cost ε m − 1 − q γ \varepsilon_{m-1}^{-q\gamma} ε m − 1 − q γ (Lemma 13.6). Each bound is at most C ε m − 1 Q C\varepsilon_{m-1}^{Q} C ε m − 1 Q by (157) .
Proposition 13.11 (The residual of the lift). *There exists C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ C(\alpha,s,q,\zeta,\Lambda,n,t_0,T)<\infty C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ such that, in the setting of Proposition 5.3, for every datum g g g of G G G , the lift Y ♯ Y^\sharp Y ♯ solves the system (66) of G G G on [ t p r , t f ] [t_{\mathrm{pr}},t_{\mathrm{f}}] [ t pr , t f ] from
Y ♯ ( t p r ) = exp ρ ( Γ m ( t p r ) : ∇ 2 ) exp ρ ( − Γ m ( t p r ) : ∇ 2 ) Π g , (175) Y^\sharp(t_{\mathrm{pr}})=\exp_{\rho}(\Gamma_m(t_{\mathrm{pr}}):\nabla^2)\exp_{\rho}(-\Gamma_m(t_{\mathrm{pr}}):\nabla^2)\Pi g, \tag*{(175)} Y ♯ ( t pr ) = exp ρ ( Γ m ( t pr ) : ∇ 2 ) exp ρ ( − Γ m ( t pr ) : ∇ 2 ) Π g , (175) with a forcing ( F , F 0 ) (F,F^0) ( F , F 0 ) such that
κ m − 1 / 2 ∥ F ∥ L 2 ( ( t p r , t f ) ; L ω 2 ( T 2 ) ) + ∥ F 0 ∥ L 1 ( ( t p r , t f ) ; L ω 2 ( T 2 ) ) ≤ C ε m Q − 1 ∥ g ∥ L G 2 ( T 2 ) . (176) \kappa_m^{-1/2}\lVert F\rVert_{L^2((t_{\mathrm{pr}},t_{\mathrm{f}});L^2_{\omega}(\mathbb{T}^2))}+\lVert F^0\rVert_{L^1((t_{\mathrm{pr}},t_{\mathrm{f}});L^2_{\omega}(\mathbb{T}^2))}\le C\varepsilon_m^{Q-1}\lVert g\rVert_{L^2_G(\mathbb{T}^2)}. \tag*{(176)} κ m − 1/2 ∥ F ∥ L 2 (( t pr , t f ) ; L ω 2 ( T 2 )) + ∥ F 0 ∥ L 1 (( t pr , t f ) ; L ω 2 ( T 2 )) ≤ C ε m Q − 1 ∥ g ∥ L G 2 ( T 2 ) . (176)
Proof. The top level m 0 m_0 m 0 is at least the levels of the lemmas above, and so large that the finitely many inequalities below between powers of ε m − 1 \varepsilon_{m-1} ε m − 1 hold for m > m 0 m>m_0 m > m 0 .
Step 1: The initial value. The velocity added at level m m m vanishes on [ t p r − τ m , t p r + τ m ] [t_{\mathrm{pr}}-\tau_m,t_{\mathrm{pr}}+\tau_m] [ t pr − τ m , t pr + τ m ] (Proposition 12.1(i)), and t p r ≥ t c ≥ t A + 3 τ m t_{\mathrm{pr}}\ge t_{\mathrm{c}}\ge t_{\mathrm{A}}+3\tau_m t pr ≥ t c ≥ t A + 3 τ m , so Lemma 13.1 gives Y ♯ ( t p r ) = Y ♭ ( t p r ) Y^{\sharp}(t_{\mathrm{pr}})=Y^{\flat}(t_{\mathrm{pr}}) Y ♯ ( t pr ) = Y ♭ ( t pr ) . As Y ♯ ( t p r ) Y^{\sharp}(t_{\mathrm{pr}}) Y ♯ ( t pr ) vanishes on the new components, Υ m Y ♯ ( t p r ) = exp p ( − Γ m ( t p r ) : ∇ 2 ) Π g \Upsilon_mY^{\sharp}(t_{\mathrm{pr}})=\exp_p(-\Gamma_m(t_{\mathrm{pr}}):\nabla^2)\Pi g Υ m Y ♯ ( t pr ) = exp p ( − Γ m ( t pr ) : ∇ 2 ) Π g , and (175) follows.
Step 2: The exact identity. By Proposition 13.10(i) and (67) , Υ m Y \Upsilon_mY Υ m Y is Lipschitz in t t t with values in H N r e t + 3 − N s r c ( T 2 ) H^{N_{\mathrm{ret}}+3-N_{\mathrm{src}}}(\mathbb{T}^2) H N ret + 3 − N src ( T 2 ) : the drift costs one order, κ m − 1 Δ \kappa_{m-1}\Delta κ m − 1 Δ two, a source at most N s r c N_{\mathrm{src}} N src , and the coefficients of G ′ G' G ′ have bounded derivatives of every order. As Γ m \Gamma_m Γ m is Lipschitz in t t t , Y ♭ Y^{\flat} Y ♭ of (154) is Lipschitz in t t t with values in H N r e t + 3 − N s r c − 2 p ( T 2 ) H^{N_{\mathrm{ret}}+3-N_{\mathrm{src}}-2p}(\mathbb{T}^2) H N ret + 3 − N src − 2 p ( T 2 ) and bounded in H N r e t + 3 − 2 p ( T 2 ) H^{N_{\mathrm{ret}}+3-2p}(\mathbb{T}^2) H N ret + 3 − 2 p ( T 2 ) , and by (158) and Sobolev’s embedding these fields have the classes of Lemmas 13.7(ii) and 13.5. So (173) holds at almost every t t t , and, with Lemma 13.7(ii), it gives the averaged equation of Y ♭ Y^{\flat} Y ♭ , with the flux of the averaged equation F a v F^{\mathrm{av}} F av :
∂ t Y ♭ + b m − 1 ⋅ ∇ Y ♭ − ∇ ⋅ ( M m Y ♭ ) = ∇ ⋅ F a v , \partial_tY^{\flat}+\mathbf{b}_{m-1}\cdot\nabla Y^{\flat}-\nabla\cdot(\mathcal{M}_mY^{\flat})=\nabla\cdot F^{\mathrm{av}}, ∂ t Y ♭ + b m − 1 ⋅ ∇ Y ♭ − ∇ ⋅ ( M m Y ♭ ) = ∇ ⋅ F av ,
F a v : = ( exp p ( Γ m : ∇ 2 ) exp p ( − Γ m : ∇ 2 ) − I ) ( ( M m − J m κ m ∇ ) Y ♭ − F c o m ) − exp p ( Γ m : ∇ 2 ) ( F s m + F f r ) − ( J m κ m − κ m − 1 I ) ∇ ( Γ m : ∇ 2 ) p p ! Υ m Y . (177) F^{\mathrm{av}}:=\left(\exp_p(\Gamma_m:\nabla^2)\exp_p(-\Gamma_m:\nabla^2)-I\right)\left((\mathcal{M}_m-\mathbf{J}^{\kappa_m}_m\nabla)Y^{\flat}-F^{\mathrm{com}}\right)
-\exp_p(\Gamma_m:\nabla^2)(F^{\mathrm{sm}}+F^{\mathrm{fr}})-(\mathbf{J}^{\kappa_m}_m-\kappa_{m-1}I)\nabla\frac{(\Gamma_m:\nabla^2)^p}{p!}\Upsilon_mY. \tag*{(177)} F av := ( exp p ( Γ m : ∇ 2 ) exp p ( − Γ m : ∇ 2 ) − I ) ( ( M m − J m κ m ∇ ) Y ♭ − F com ) − exp p ( Γ m : ∇ 2 ) ( F sm + F fr ) − ( J m κ m − κ m − 1 I ) ∇ p ! ( Γ m : ∇ 2 ) p Υ m Y . (177)
Here F c o m F^{\mathrm{com}} F com is the flux of the commutator at Υ m Y \Upsilon_mY Υ m Y ; the first line of F a v F^{\mathrm{av}} F av is the truncation part , and its last term the Taylor part . Indeed, [ b m − 1 ⋅ ∇ , exp p ( Γ m : ∇ 2 ) ] Υ m Y = ∇ ⋅ F c o m [\mathbf{b}_{m-1}\cdot\nabla,\exp_p(\Gamma_m:\nabla^2)]\Upsilon_mY=\nabla\cdot F^{\mathrm{com}} [ b m − 1 ⋅ ∇ , exp p ( Γ m : ∇ 2 )] Υ m Y = ∇ ⋅ F com by (169) , J m κ m : ∇ 2 = ∇ ⋅ ( J m κ m ∇ ) \mathbf{J}^{\kappa_m}_m:\nabla^2=\nabla\cdot(\mathbf{J}^{\kappa_m}_m\nabla) J m κ m : ∇ 2 = ∇ ⋅ ( J m κ m ∇ ) , and exp p ( ± Γ m : ∇ 2 ) \exp_p(\pm\Gamma_m:\nabla^2) exp p ( ± Γ m : ∇ 2 ) commute with ∇ ⋅ \nabla\cdot ∇ ⋅ . Lemma 13.5, applied to the field Y ♭ Y^{\flat} Y ♭ , then gives, with A m = A m S + ( A m − A m S ) \mathbf{A}_m=\mathbf{A}^{S}_m+(\mathbf{A}_m-\mathbf{A}^{S}_m) A m = A m S + ( A m − A m S ) ,
( ∂ t + b m ⋅ ∇ − κ m Δ ) Y ♯ − ∇ ⋅ ( A m S Y ♯ ) = Corr m ( ∇ ⋅ F a v ) + ∇ ⋅ F t l + F t l , 0 + ∑ k H m , k ′ ( Corr m , k Y ♭ − Y ♭ ) + ∇ ⋅ F f i n e . (\partial_t+\mathbf{b}_m\cdot\nabla-\kappa_m\Delta)Y^{\sharp}-\nabla\cdot(\mathbf{A}^{S}_mY^{\sharp})
=\operatorname{Corr}_m(\nabla\cdot F^{\mathrm{av}})+\nabla\cdot F^{\mathrm{tl}}+F^{\mathrm{tl},0}+\sum_k H'_{m,k}(\operatorname{Corr}_{m,k}Y^{\flat}-Y^{\flat})+\nabla\cdot F^{\mathrm{fine}}. ( ∂ t + b m ⋅ ∇ − κ m Δ ) Y ♯ − ∇ ⋅ ( A m S Y ♯ ) = Corr m ( ∇ ⋅ F av ) + ∇ ⋅ F tl + F tl , 0 + k ∑ H m , k ′ ( Corr m , k Y ♭ − Y ♭ ) + ∇ ⋅ F fine .
with the fine smoothing defect F f i n e : = ( A m − A m S ) Y ♯ F^{\mathrm{fine}}:=(\mathbf{A}_m-\mathbf{A}^{S}_m)Y^{\sharp} F fine := ( A m − A m S ) Y ♯ , the sources of G G G acting on ( I − S σ ) Y ♯ (I-S_\sigma)Y^{\sharp} ( I − S σ ) Y ♯ . The forcing ( F , F 0 ) (F,F^0) ( F , F 0 ) of (176) collects the four terms on the right side, the coarse flux, the cell tail, the switching term and the fine smoothing defect: with the forcing ( F c o , F c o , 0 ) (F^{\mathrm{co}},F^{\mathrm{co},0}) ( F co , F co , 0 ) of Lemma 13.6 for F a v F^{\mathrm{av}} F av , it is F = F c o + F t l + F f i n e F=F^{\mathrm{co}}+F^{\mathrm{tl}}+F^{\mathrm{fine}} F = F co + F tl + F fine and F 0 = F c o , 0 + F t l , 0 + ∑ k H m , k ′ ( Corr m , k Y ♭ − Y ♭ ) F^0=F^{\mathrm{co},0}+F^{\mathrm{tl},0}+\sum_k H'_{m,k}(\operatorname{Corr}_{m,k}Y^{\flat}-Y^{\flat}) F 0 = F co , 0 + F tl , 0 + ∑ k H m , k ′ ( Corr m , k Y ♭ − Y ♭ ) . And Y ♯ Y^{\sharp} Y ♯ lies in the class of solutions: by Lemma 13.5 it is Lipschitz with values in L 2 ( T 2 ) V L^2(\mathbb{T}^2)^V L 2 ( T 2 ) V , the display holds in the form (67) , and Proposition 13.10(ii) gives the rest.
Step 3: The four terms. By Proposition 13.10(i), the energy of Υ m Y \Upsilon_mY Υ m Y of order N r e t N_{\mathrm{ret}} N ret is at most C ∥ g ∥ L G 2 ( T 2 ) C\lVert g\rVert_{L^2_G(\mathbb{T}^2)} C ∥ g ∥ L G 2 ( T 2 ) , and so, by Lemma 13.7(iii), is that of Y ♭ Y^{\flat} Y ♭ of order N r e t − 2 p N_{\mathrm{ret}}-2p N ret − 2 p . The truncation part reads the energy of Υ m Y \Upsilon_mY Υ m Y of order at most N r e c + 6 p + N s r c − 2 N_{\mathrm{rec}}+6p+N_{\mathrm{src}}-2 N rec + 6 p + N src − 2 , and the smoothing defect that of order at most N r e c + 2 p + N s r c + N m o m − 2 N_{\mathrm{rec}}+2p+N_{\mathrm{src}}+N_{\mathrm{mom}}-2 N rec + 2 p + N src + N mom − 2 , both at most N r e t N_{\mathrm{ret}} N ret by the first line of (158) . The fine smoothing defect reads the component energy of Y ♯ Y^{\sharp} Y ♯ of Proposition 13.10(ii) at the order N m o m + N s r c − 2 ≤ N r e t − N c o r − 2 p N_{\mathrm{mom}}+N_{\mathrm{src}}-2\le N_{\mathrm{ret}}-N_{\mathrm{cor}}-2p N mom + N src − 2 ≤ N ret − N cor − 2 p , by the same line. Moreover, at each time Y ♭ Y^{\flat} Y ♭ and F a v F^{\mathrm{av}} F av have N r e c + 3 N_{\mathrm{rec}}+3 N rec + 3 derivatives in L 2 L^2 L 2 , by Proposition 13.10(i) and (158) , hence N r e c + 1 N_{\mathrm{rec}}+1 N rec + 1 continuous ones (measurable in ( t , x ) (t,x) ( t , x ) , being continuous in x x x and measurable in t t t , the smoothings by clause (c3) and Lemma 8.2), as Lemmas 13.4–13.6 require. Table 13.1 lists the terms.
The coarse flux , by Lemma 13.6: its cost is C ( κ m − 1 / κ m ) 1 / 2 ≤ C ε m − 1 − q γ C(\kappa_{m-1}/\kappa_m)^{1/2}\le C\varepsilon_{m-1}^{-q\gamma} C ( κ m − 1 / κ m ) 1/2 ≤ C ε m − 1 − q γ , since its second term is at most C C C , by (159) and the condition on N c l N_{\mathrm{cl}} N cl in (157) (see the cell tail below). Then κ m − 1 − 1 ∥ F a v ∥ L 2 ( ( t p r , t f ) ; H c o e f f , ω N r e c ( T 2 ) ) \kappa_{m-1}^{-1}\lVert F^{\mathrm{av}}\rVert_{L^2((t_{\mathrm{pr}},t_{\mathrm{f}});H^{N_{\mathrm{rec}}}_{\mathrm{coeff},\omega}(\mathbb{T}^2))} κ m − 1 − 1 ∥ F av ∥ L 2 (( t pr , t f ) ; H coeff , ω N rec ( T 2 )) has four parts. The truncation part is at most C ε m − 1 δ ( p + 1 ) ∥ g ∥ L G 2 ( T 2 ) C\varepsilon_{m-1}^{\delta(p+1)}\lVert g\rVert_{L^2_G(\mathbb{T}^2)} C ε m − 1 δ ( p + 1 ) ∥ g ∥ L G 2 ( T 2 ) , by (171) : the coefficients of M m − J m κ m ∇ \mathcal{M}_m-\mathbf{J}^{\kappa_m}_m\nabla M m − J m κ m ∇ and of the flux of the commutator are at most C κ m − 1 C\kappa_{m-1} C κ m − 1 at the scale ρ ( m ) \rho(m) ρ ( m ) , by Proposition 10.7 and, for the flux of the commutator, by
ρ ( m ) l ∥ ∇ l + 1 b m − 1 ∥ L ∞ ( R × T 2 ) ∣ Γ m ( t ) ∣ ≤ C ( l ) κ m − 1 ε m − 1 δ , l ≥ 0. \rho(m)^l\lVert\nabla^{l+1}\mathbf{b}_{m-1}\rVert_{L^\infty(\mathbb{R}\times\mathbb{T}^2)}\lvert\Gamma_m(t)\rvert\le C(l)\kappa_{m-1}\varepsilon_{m-1}^{\delta},\qquad l\ge0. ρ ( m ) l ∥ ∇ l + 1 b m − 1 ∥ L ∞ ( R × T 2 ) ∣ Γ m ( t )∣ ≤ C ( l ) κ m − 1 ε m − 1 δ , l ≥ 0.
Proof of Proposition 5.3(b)
which holds by (100) , with (49) and (15) at l = 0 l=0 l = 0 , and with the first line of (12.1) at the level m − 1 m-1 m − 1 and ρ ( m ) ≤ ρ m o l ( m − 1 ) \rho(m)\leq\rho_{\mathrm{mol}}(m-1) ρ ( m ) ≤ ρ mol ( m − 1 ) at l ≥ 1 l\geq1 l ≥ 1 . The Taylor part is at most C ε m δ p ∥ g ∥ L G 2 ( T 2 ) C\varepsilon_m^{\delta p}\|g\|_{L^2_G(\mathbb{T}^2)} C ε m δ p ∥ g ∥ L G 2 ( T 2 ) , since ∣ J m κ m ( t ) ∣ ≤ κ m + ε m 2 / ( 2 κ m ) ≤ C κ m − 1 \lvert\mathbf{J}^{\kappa_m}_m(t)\rvert\leq\kappa_m+\varepsilon_m^2/(2\kappa_m)\leq C\kappa_{m-1} ∣ J m κ m ( t )∣ ≤ κ m + ε m 2 / ( 2 κ m ) ≤ C κ m − 1 . The smoothing defect F s m F^{\mathrm{sm}} F sm appears under the truncated exponential, so by the second bound of (171) we bound it in H ρ ( m ) , ω N r e c + 2 p ( T 2 ) H^{N_{\mathrm{rec}}+2p}_{\rho(m),\omega}(\mathbb{T}^2) H ρ ( m ) , ω N rec + 2 p ( T 2 ) . It is at most C\varepsilon_m^{\Theta N_{\mathrm{mom}}}_{m-1}\|g\|_{L^2_G(\mathbb{T}^2)} : by the third bound of Proposition 6.1 at each time (which needs no transversality and has a constant independent of ℓ \ell ℓ ), with N 1 = N m o m N_1=N_{\mathrm{mom}} N 1 = N mom and the bound C Ω C_\Omega C Ω of (c1), each piece acting on ( I − S σ ) Υ m Y (I-S_\sigma)\Upsilon_mY ( I − S σ ) Υ m Y costs ( r σ / ρ ( m ) ) N m o m (r_\sigma/\rho(m))^{N_{\mathrm{mom}}} ( r σ / ρ ( m ) ) N mom , with r σ ≤ C ρ ( m ) ε m − 1 ϑ r_\sigma\leq C\rho(m)\varepsilon_{m-1}^{\vartheta} r σ ≤ C ρ ( m ) ε m − 1 ϑ by (c2) for G G G , whose old sources G ′ G' G ′ copies, by (178) for the letters, and 4 δ > Θ 4\delta>\Theta 4 δ > Θ (by (181) ). The coefficients are bounded by clauses (b1)–(b3) of the admissible family G ′ G' G ′ ; a letter piece into a component of G G G of generation N g e n N_{\mathrm{gen}} N gen has no source in G ′ G' G ′ , and its coefficients obey the same bounds before the omission, as in the proof of Lemma 13.9. The frontier part is Lemma 13.9 at the order N r e c + 2 p N_{\mathrm{rec}}+2p N rec + 2 p , under the truncated exponential as for F s m F^{\mathrm{sm}} F sm .
( t f − t p r ) 1 / 2 κ m 1 / 2 ε m ( ε m ρ ( m ) ) N c l ≤ C T 1 / 2 ε m − 1 − ( 2 − β + 8 δ ) / 2 + ( q γ + 3 δ ) N c l ≤ C . (t_{\mathrm{f}}-t_{\mathrm{pr}})^{1/2}\frac{\kappa_m^{1/2}}{\varepsilon_m}\left(\frac{\varepsilon_m}{\rho(m)}\right)^{N_{\mathrm{cl}}}\leq CT^{1/2}\varepsilon_{m-1}^{-(2-\beta+8\delta)/2+(q\gamma+3\delta)N_{\mathrm{cl}}}\leq C. ( t f − t pr ) 1/2 ε m κ m 1/2 ( ρ ( m ) ε m ) N cl ≤ C T 1/2 ε m − 1 − ( 2 − β + 8 δ ) /2 + ( q γ + 3 δ ) N cl ≤ C .
by (159) , the condition on N c l N_{\mathrm{cl}} N cl in (157) , and κ m / ε m 2 ≤ C ε m β + γ − 2 ≤ C ε m − 1 − ( 2 − β + 8 δ ) \kappa_m/\varepsilon_m^2\leq C\varepsilon_m^{\beta+\gamma-2}\leq C\varepsilon_{m-1}^{-(2-\beta+8\delta)} κ m / ε m 2 ≤ C ε m β + γ − 2 ≤ C ε m − 1 − ( 2 − β + 8 δ ) (Proposition 2.1(b) and Lemma 7.1(c)).
Switching , by Lemma 13.5: at most C ε m − 1 − C exp ( − c ε m − 1 − 4 δ ) ≤ C ε m − 1 Q C\varepsilon_{m-1}^{-C}\exp(-c\varepsilon_{m-1}^{-4\delta})\leq C\varepsilon_{m-1}^{Q} C ε m − 1 − C exp ( − c ε m − 1 − 4 δ ) ≤ C ε m − 1 Q , since t f − t p r ≤ T t_{\mathrm{f}}-t_{\mathrm{pr}}\leq T t f − t pr ≤ T , τ m ≍ ε m − 1 2 − β + 4 δ \tau_m\asymp\varepsilon_{m-1}^{2-\beta+4\delta} τ m ≍ ε m − 1 2 − β + 4 δ by (7) , and ε m 2 / κ m ≤ C τ m \varepsilon_m^2/\kappa_m\leq C\tau_m ε m 2 / κ m ≤ C τ m by Proposition 2.1(c).
The fine smoothing defect. The sources of G G G at the scale ρ m o l ( m ) \rho_{\mathrm{mol}}(m) ρ mol ( m ) are bounded by clause (b2′). By the third bound of Proposition 6.1 at this scale, at each time, as for F s m F^{\mathrm{sm}} F sm , each costs ( r σ / ρ m o l ( m ) ) N m o m ≤ C ε m δ N m o m ≤ C ε m − 1 q δ N m o m (r_\sigma/\rho_{\mathrm{mol}}(m))^{N_{\mathrm{mom}}}\leq C\varepsilon_m^{\delta N_{\mathrm{mom}}}\leq C\varepsilon_{m-1}^{q\delta N_{\mathrm{mom}}} ( r σ / ρ mol ( m ) ) N mom ≤ C ε m δ N mom ≤ C ε m − 1 q δ N mom , by (c2). The remaining norm of Y ♯ Y^\sharp Y ♯ , of order N m o m + N s r c − 2 N_{\mathrm{mom}}+N_{\mathrm{src}}-2 N mom + N src − 2 at the scale ρ m o l ( m ) \rho_{\mathrm{mol}}(m) ρ mol ( m ) , is bounded by Proposition 13.10(ii).
Step 4: The orders. By (157) , whose margin q γ q\gamma q γ absorbs the cost ε m − 1 − q γ \varepsilon_{m-1}^{-q\gamma} ε m − 1 − q γ of the coarse flux, the norm (156) of each term is at most C ε m − 1 Q + 4 ∥ g ∥ L G 2 ( T 2 ) C\varepsilon_{m-1}^{Q+4}\|g\|_{L^2_G(\mathbb{T}^2)} C ε m − 1 Q + 4 ∥ g ∥ L G 2 ( T 2 ) , and (176) follows.
Proof of Proposition 5.3(b)
Here m 0 m_0 m 0 is the top level of item (4) of Subsection 11.1, at least every threshold below (Table 11.1 ).
Admissibility. By Definition 11.4, G G G descends from an admissible seed by successors, with ((120) ) and t A = t A ( m ) t_A=t_A^{(m)} t A = t A ( m ) ; we append G ′ G' G ′ to it. The new condition ((120) ) is t A ( m − 1 ) = t A + 3 τ m ≤ t f − 4 h m t_A^{(m-1)}=t_A+3\tau_m\leq t_f-4h_m t A ( m − 1 ) = t A + 3 τ m ≤ t f − 4 h m , which holds since t A + 3 τ m ≤ t c ≤ t f − 4 h m t_A+3\tau_m\leq t_c\leq t_f-4h_m t A + 3 τ m ≤ t c ≤ t f − 4 h m , and m − 1 ≥ m 0 m-1\geq m_0 m − 1 ≥ m 0 . So G ′ G' G ′ is admissible on [ t A + 3 τ m , t f ] [t_A+3\tau_m,t_f] [ t A + 3 τ m , t f ] .
The marked amplitude. So G ′ G' G ′ satisfies clause (d) on [ t p r , t f ] [t_{\rm pr},t_f] [ t pr , t f ] (Proposition 11.5), whose matrix at the order 0 is the coupling matrix of Proposition 6.2; with the masses ω v / Ψ v 2 \omega_v/\Psi_v^2 ω v / Ψ v 2 , this proposition gives ∥ Y ∥ L ∞ ( [ t p r , t f ] ; L G ′ 2 ( T 2 ) ) ≤ ∥ Y ( t p r ) ∥ L G ′ 2 ( T 2 ) \|Y\|_{L^\infty([t_{\rm pr},t_f];L^2_{G'}(\mathbb{T}^2))}\leq\|Y(t_{\rm pr})\|_{L^2_{G'}(\mathbb{T}^2)} ∥ Y ∥ L ∞ ([ t pr , t f ] ; L G ′ 2 ( T 2 )) ≤ ∥ Y ( t pr ) ∥ L G ′ 2 ( T 2 ) . The datum, the retained trace, has the masses and markers of G G G on the components of G G G (Lemma 11.8(i)), so Lemma 13.7(iii) and Proposition 5.3(a), with N o u t ≥ 2 p N_{\rm out}\geq2p N out ≥ 2 p , give the first bound of Proposition 5.3(b):
∥ Y ( t p r ) ∥ L G ′ 2 ( T 2 ) ≤ ∥ Π g ∥ L G 2 ( T 2 ) + C ε m δ m − 1 ∥ Π g ∥ H ρ ( m ) , G N o u t ( T 2 ) ≤ ( 1 + C ε m Q ) ∥ g ∥ L G 2 ( T 2 ) + C ε m δ m − 1 ∥ g ∥ L G 2 ( T 2 ) . \begin{aligned}
\|Y(t_{\rm pr})\|_{L^2_{G'}(\mathbb{T}^2)}
&\leq\|\Pi g\|_{L^2_G(\mathbb{T}^2)}+C\varepsilon_m^{\delta_{m-1}}\|\Pi g\|_{H_{\rho(m),G}^{N_{\rm out}}(\mathbb{T}^2)} \\
&\leq(1+C\varepsilon_m^Q)\|g\|_{L^2_G(\mathbb{T}^2)}+C\varepsilon_m^{\delta_{m-1}}\|g\|_{L^2_G(\mathbb{T}^2)}.
\end{aligned} ∥ Y ( t pr ) ∥ L G ′ 2 ( T 2 ) ≤ ∥Π g ∥ L G 2 ( T 2 ) + C ε m δ m − 1 ∥Π g ∥ H ρ ( m ) , G N out ( T 2 ) ≤ ( 1 + C ε m Q ) ∥ g ∥ L G 2 ( T 2 ) + C ε m δ m − 1 ∥ g ∥ L G 2 ( T 2 ) .
The comparison. The restriction of U U U to [ t p r , t f ] [t_{\rm pr},t_f] [ t pr , t f ] solves the system of G G G from U ( t p r ) U(t_{\rm pr}) U ( t pr ) (Section 8). By Proposition 13.11, U − Y ♯ U-Y^\sharp U − Y ♯ solves ((66) ) with the forcing − ( F , F 0 ) -(F,F^0) − ( F , F 0 ) from
U ( t p r ) − Y ♯ ( t p r ) = ( U ( t p r ) − Π g ) − ( exp p ( Γ m ( t p r ) : ∇ 2 ) exp p ( − Γ m ( t p r ) : ∇ 2 ) − I ) Π g . U(t_{\rm pr})-Y^\sharp(t_{\rm pr})=(U(t_{\rm pr})-\Pi g)-\left(\exp_p(\Gamma_m(t_{\rm pr}):\nabla^2)\exp_p(-\Gamma_m(t_{\rm pr}):\nabla^2)-I\right)\Pi g. U ( t pr ) − Y ♯ ( t pr ) = ( U ( t pr ) − Π g ) − ( exp p ( Γ m ( t pr ) : ∇ 2 ) exp p ( − Γ m ( t pr ) : ∇ 2 ) − I ) Π g .
We bound both terms in L ω 2 ( T 2 ) L^2_\omega(\mathbb{T}^2) L ω 2 ( T 2 ) by their norms with the markers, which are at least 1. The first term is at most C ε m Q ∥ g ∥ L G 2 ( T 2 ) C\varepsilon_m^Q\|g\|_{L^2_G(\mathbb{T}^2)} C ε m Q ∥ g ∥ L G 2 ( T 2 ) by the first bound of Proposition 5.3(a). By ((171) ) at N = 0 N=0 N = 0 , the third bound of Proposition 5.3(a) with N o u t ≥ 4 p N_{\rm out}\geq4p N out ≥ 4 p , and δ p > Q \delta p>Q δ p > Q , the second term is at most
C ε m δ ( p + 1 ) ∥ Π g ∥ H ρ ( m ) , G 4 p ( T 2 ) ≤ C ε m Q ∥ g ∥ L G 2 ( T 2 ) . C\varepsilon_m^{\delta(p+1)}\|\Pi g\|_{H_{\rho(m),G}^{4p}(\mathbb{T}^2)}
\leq C\varepsilon_m^Q\|g\|_{L^2_G(\mathbb{T}^2)}. C ε m δ ( p + 1 ) ∥Π g ∥ H ρ ( m ) , G 4 p ( T 2 ) ≤ C ε m Q ∥ g ∥ L G 2 ( T 2 ) .
As G G G satisfies (d) with sums at most 1 8 \frac18 8 1 for the masses ω v \omega_v ω v (Proposition 11.5), by linearity and uniqueness (Proposition 8.3, N = 0 N=0 N = 0 ), U − Y ♯ U-Y^\sharp U − Y ♯ is the solution given there for this datum and the forcing − ( F , F 0 ) -(F,F^0) − ( F , F 0 ) , and Proposition 8.3(a) bounds ∥ U − Y ♯ ∥ L ∞ ( [ t p r , t f ] ; L ω 2 ( T 2 ) ) + κ m 1 / 2 ∥ ∇ ( U − Y ♯ ) ∥ L 2 ( ( t p r , t f ) ; L ω 2 ( T 2 ) ) \|U-Y^\sharp\|_{L^\infty([t_{\rm pr},t_f];L^2_\omega(\mathbb{T}^2))}+\kappa_m^{1/2}\|\nabla(U-Y^\sharp)\|_{L^2((t_{\rm pr},t_f);L^2_\omega(\mathbb{T}^2))} ∥ U − Y ♯ ∥ L ∞ ([ t pr , t f ] ; L ω 2 ( T 2 )) + κ m 1/2 ∥∇ ( U − Y ♯ ) ∥ L 2 (( t pr , t f ) ; L ω 2 ( T 2 )) by C C C times the norms of this datum in L ω 2 ( T 2 ) L^2_\omega(\mathbb{T}^2) L ω 2 ( T 2 ) and of the forcing in ((156) ), hence by C ε m Q ∥ g ∥ L G 2 ( T 2 ) C\varepsilon_m^Q\|g\|_{L^2_G(\mathbb{T}^2)} C ε m Q ∥ g ∥ L G 2 ( T 2 ) by ((176) ): the second bound of Proposition 5.3(b). The third is Proposition 13.10(i).
The terminal cycle. Section 15 compares the terminal cycle, at level M M M , in the same way: Lemmas 13.1, 13.2 and 13.4–13.6 hold at m = M m=M m = M , where the cost ( κ M − 1 / κ M ) 1 / 2 (\kappa_{M-1}/\kappa_M)^{1/2} ( κ M − 1 / κ M ) 1/2 in Lemma 13.6 is bounded; Lemma 13.3 is for m < M m<M m < M . Lemma 13.7(i)–(ii) hold at every level, and ((171) ) holds at m = M m=M m = M , with its proof, for the anchored covariance Γ M ( t ) − Γ M ( t p r ( M ) ) \Gamma_M(t)-\Gamma_M(t_{\rm pr}^{(M)}) Γ M ( t ) − Γ M ( t pr ( M ) ) , which Lemma 10.3(b) bounds.
Passage between consecutive levels and induction on the levels
In this section we prove Proposition 5.4, one passage: the cycle from level m m m to the coarser level m − 1 m-1 m − 1 , read on representations; and Proposition 5.2, the induction on the levels, which runs the two dual cascades.
The five profiles. In the notation of Proposition 5.3, a passage writes U ( t ) U(t) U ( t ) as Υ m Y ( t ) \Upsilon_mY(t) Υ m Y ( t ) plus the five profiles of Table 14.1 : the exponential and corrector parts ((187) ) of the lift, applied to S Y ( t ) SY(t) S Y ( t ) , where S = S ( t ) S=S(t) S = S ( t ) is the fresh smoothing ((186) ), and to ( I − S ) Y ( t ) (I-S)Y(t) ( I − S ) Y ( t ) , and the comparison error U ( t ) − Y ♯ ( t ) U(t)-Y^{\sharp}(t) U ( t ) − Y ♯ ( t ) . The first two profiles are bounded by a small multiple of the size of a representation R ′ R' R ′ of Y ( t ) Y(t) Y ( t ) at wavelength ε m − 1 \varepsilon_{m-1} ε m − 1 (Lemma 14.2(a)), the last three by C ε m − 1 2 ∥ g ∥ L G ( T 2 ) 2 C\varepsilon_{m-1}^{2}\lVert g\rVert_{L^{2}_{G(\mathbb{T}^{2})}} C ε m − 1 2 ∥ g ∥ L G ( T 2 ) 2 . This absolute smallness absorbs the power ε m − 1 − Z \varepsilon_{m-1}^{-Z} ε m − 1 − Z of (179) lost when a Sobolev bound is read as a profile.
Parameters. Throughout, C C C and c c c denote positive constants depending only on α \alpha α , s s s , q q q , ζ \zeta ζ , Λ \Lambda Λ , n n n , t 0 t_0 t 0 , T T T , which may change from line to line. Besides the parameters of Subsection 5.3, among them the order N r e t N_{\mathrm{ret}} N ret , fixed in Section 13 with the floors (158) and (180) , the proofs use the following ones, depending only on α \alpha α , s s s , q q q , ζ \zeta ζ , Λ \Lambda Λ , n n n , t 0 t_0 t 0 , T T T (Table 11.1 ):
r m : = ρ ( m ) ε m − 1 Θ (178) r_m := \rho(m)\varepsilon_{m-1}^{\Theta} \tag*{(178)} r m := ρ ( m ) ε m − 1 Θ (178)
of the fresh smoothing (186) of the new sources of a successor created at level m m m , and the order N m o m N_{\mathrm{mom}} N mom of its vanishing moments (Section 11);
the constant c m o l ∈ ( 0 , 1 ] c_{\mathrm{mol}}\in(0,1] c mol ∈ ( 0 , 1 ] of the molecular radius ρ m o l ( m ) \rho_{\mathrm{mol}}(m) ρ mol ( m ) of (115) , the scale at which a solution of a family of level m m m has derivatives after the time ρ m o l ( m ) 2 / κ m \rho_{\mathrm{mol}}(m)^2/\kappa_m ρ mol ( m ) 2 / κ m (Section 12);
Z : = 1 + s + ( γ 2 + δ ) ( N r e p + 1 ) , (179) Z:=1+s+\left(\frac{\gamma}{2}+\delta\right)(N_{\mathrm{rep}}+1), \tag*{(179)} Z := 1 + s + ( 2 γ + δ ) ( N rep + 1 ) , (179)
lost when a Sobolev bound is read as a profile.
These parameters and N r e t N_{\mathrm{ret}} N ret satisfy
N r e t ≥ 2 N r e p + 4 + N c o r + 2 p + N m o m , Q ≥ 4 + 2 q Z , (180) N_{\mathrm{ret}}\ge2N_{\mathrm{rep}}+4+N_{\mathrm{cor}}+2p+N_{\mathrm{mom}},\qquad Q\ge4+2qZ, \tag*{(180)} N ret ≥ 2 N rep + 4 + N cor + 2 p + N mom , Q ≥ 4 + 2 q Z , (180)
and the width exponent is so small that
12 Θ < δ , Θ N c o r ≤ ( γ 2 + δ ) ( N r e p − 1 ) . (181) 12\Theta<\delta,\qquad\Theta N_{\mathrm{cor}}\le\left(\frac{\gamma}{2}+\delta\right)(N_{\mathrm{rep}}-1). \tag*{(181)} 12Θ < δ , Θ N cor ≤ ( 2 γ + δ ) ( N rep − 1 ) . (181)
Estimates from the homogenization cycle
For a field F F F of a family G G G and ρ > 0 \rho>0 ρ > 0 , the norm (65) of H ρ , ω N ( T 2 ) H_{\rho,\omega}^{N}(\mathbb{T}^{2}) H ρ , ω N ( T 2 ) is the norm (35) of H ρ , G N ( T 2 ) H_{\rho,G}^{N}(\mathbb{T}^{2}) H ρ , G N ( T 2 ) with all markers replaced by 1. Since the markers are at least 1,
∥ F ∥ H ρ , ω N ( T 2 ) ≤ ∥ F ∥ H ρ , G N ( T 2 ) . (182) \lVert F\rVert_{H_{\rho,\omega}^{N}(\mathbb{T}^{2})}\le\lVert F\rVert_{H_{\rho,G}^{N}(\mathbb{T}^{2})}. \tag*{(182)} ∥ F ∥ H ρ , ω N ( T 2 ) ≤ ∥ F ∥ H ρ , G N ( T 2 ) . (182)
A passage works at one time. Proposition 14.1 bounds the factor by which the readout multiplies the size of a representation (Lemma 14.2(b)), states the comparison at every time, and gives U ( t ) U(t) U ( t ) the 2 N r e p + 4 2N_{\mathrm{rep}}+4 2 N rep + 4 derivatives at the scale ρ m o l ( m ) \rho_{\mathrm{mol}}(m) ρ mol ( m ) that the lift has (Proposition 13.10(ii)); Step 6 of the proof of Proposition 5.4 interpolates between the comparison and these derivatives.
Proposition 14.1 (Further properties of a cycle). There exists C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ C(\alpha,s,q,\zeta,\Lambda,n,t_0,T)<\infty C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ such that, for every M M M , κ \kappa κ , m m m , t A t_A t A , t c t_c t c , t f t_f t f , G G G and G ′ G' G ′ as in Proposition 5.3, there exist t p r t_{\mathrm{pr}} t pr and Π \Pi Π satisfying (a) and (b) of Proposition 5.3 with this C C C , such that the following holds for every datum g g g of G G G , with U U U and Y ♯ Y^\sharp Y ♯ as there.
(i) The readout satisfies ∣ Υ m ∣ ≤ 1 + C ε m − 1 5 η / 4 |\Upsilon_m|\leq1+C\varepsilon_{m-1}^{5\eta/4} ∣ Υ m ∣ ≤ 1 + C ε m − 1 5 η /4 .
(ii) For every t ∈ [ t p r , t f ] t\in[t_{\mathrm{pr}},t_f] t ∈ [ t pr , t f ] ,
∥ U ( t ) − Y ♯ ( t ) ∥ L ω 2 ( T 2 ) ≤ C ε m − 1 Q ∥ g ∥ L G 2 ( T 2 ) , \|U(t)-Y^\sharp(t)\|_{L^2_\omega(\mathbb{T}^2)}\leq C\varepsilon_{m-1}^{Q}\|g\|_{L^2_G(\mathbb{T}^2)}, ∥ U ( t ) − Y ♯ ( t ) ∥ L ω 2 ( T 2 ) ≤ C ε m − 1 Q ∥ g ∥ L G 2 ( T 2 ) , ∥ U ( t ) ∥ H ρ m o l ( m ) , ω 2 N r e p + 4 ( T 2 ) ≤ C ∥ g ∥ L G 2 ( T 2 ) . \|U(t)\|_{H^{2N_{\mathrm{rep}}+4}_{\rho_{\mathrm{mol}}(m),\omega}(\mathbb{T}^2)}\leq C\|g\|_{L^2_G(\mathbb{T}^2)}. ∥ U ( t ) ∥ H ρ mol ( m ) , ω 2 N rep + 4 ( T 2 ) ≤ C ∥ g ∥ L G 2 ( T 2 ) .
Proposition 14.1 holds for the time t p r t_{\mathrm{pr}} t pr and the map Π \Pi Π constructed in § 5.6 : part (i) is Lemma 11.8(iii). The first bound of (ii) is Proposition 5.3(b), and the second is Proposition 12.1(iii), each at every time, as U U U (Proposition 6.2) and Y ♯ Y^\sharp Y ♯ (Proposition 13.11) are continuous in time with values in L 2 ( T 2 ) V L^2(\mathbb{T}^2)^{V} L 2 ( T 2 ) V . For the same t p r t_{\mathrm{pr}} t pr and Π \Pi Π , Proposition 13.10 bounds Y Y Y and Y ♯ ( t ) Y^\sharp(t) Y ♯ ( t ) . Proposition 5.4 is proved below with these t p r t_{\mathrm{pr}} t pr and Π \Pi Π . Lemma 13.3 applies to the family G G G of Proposition 5.3 at every time t ∈ [ t c , t f ] t\in[t_c,t_f] t ∈ [ t c , t f ] , since G G G is admissible on [ t A , t f ] [t_A,t_f] [ t A , t f ] , hence in the setting of the lift of § 5.8 (Proposition 11.5 gives the invariant), and t c ≥ t A + 3 τ m t_c\geq t_A+3\tau_m t c ≥ t A + 3 τ m .
By Lemma 10.3(b) and (178) , the covariance is small against the square of the width:
∣ Γ m ( t ) ∣ ≤ C ε m − 1 δ − 2 Θ r m 2 . (183) |\Gamma_m(t)|\leq C\varepsilon_{m-1}^{\delta-2\Theta}r_m^2 . \tag*{(183)} ∣ Γ m ( t ) ∣ ≤ C ε m − 1 δ − 2Θ r m 2 . (183)
By Proposition 2.1(c), for m ≥ 2 m\geq2 m ≥ 2 , ρ ( m ) ≍ ε m − 1 1 + γ / 2 + δ \rho(m)\asymp\varepsilon_{m-1}^{1+\gamma/2+\delta} ρ ( m ) ≍ ε m − 1 1 + γ /2 + δ , so that
C ε m − 1 1 + γ / 2 + δ ≤ ρ ( m ) ≤ C ε m − 1 1 + γ / 2 + δ ≤ C ε m − 1 , r m ≤ ρ ( m ) . (184) C\varepsilon_{m-1}^{1+\gamma/2+\delta}\leq\rho(m)\leq C\varepsilon_{m-1}^{1+\gamma/2+\delta}\leq C\varepsilon_{m-1},\qquad r_m\leq\rho(m). \tag*{(184)} C ε m − 1 1 + γ /2 + δ ≤ ρ ( m ) ≤ C ε m − 1 1 + γ /2 + δ ≤ C ε m − 1 , r m ≤ ρ ( m ) . (184)
Representations and the smoothing
By Lemma 14.2(a), the size of a representation bounds the field, its derivatives at the wavelength and its Hölder norm. By (b), the readout carries a representation to a finer wavelength at the cost of the norm of the readout, as each profile keeps its wavelength and the base, read at a finer wavelength, does not grow; added profiles add their norms.
Lemma 14.2 (Bounds from a representation, and copies). Let 0 < λ ≤ λ ′ ≤ 1 0<\lambda\leq\lambda'\leq1 0 < λ ≤ λ ′ ≤ 1 .
(a) For every field U U U of a family and every representation R R R of U U U at wavelength λ \lambda λ ,
∥ U ∥ L ∞ ( T 2 ) + λ − s [ U ] C λ N r e p ( T 2 ) ≤ size λ ( R ) , ∥ U ∥ C 0 , s ( T 2 ) ≤ 3 size λ ( R ) . \|U\|_{L^\infty(\mathbb{T}^2)}+\lambda^{-s}[U]_{C^{N_{\mathrm{rep}}}_\lambda(\mathbb{T}^2)}\leq\operatorname{size}_\lambda(R),\qquad\|U\|_{C^{0,s}(\mathbb{T}^2)}\leq3\operatorname{size}_\lambda(R). ∥ U ∥ L ∞ ( T 2 ) + λ − s [ U ] C λ N rep ( T 2 ) ≤ size λ ( R ) , ∥ U ∥ C 0 , s ( T 2 ) ≤ 3 size λ ( R ) . (b) Let R ′ R' R ′ be a representation, with base E E E and profiles P \mathcal{P} P , of a field Y Y Y of a successor G ′ G' G ′ of a family G G G of level m m m at wavelength λ ′ \lambda' λ ′ . Let finitely many fields P i ∈ C N r e p ( T 2 ) V P_i\in C^{N_{\mathrm{rep}}}(\mathbb{T}^2)^V P i ∈ C N rep ( T 2 ) V of G G G be given, with wavelengths λ i ∈ [ λ , 1 ] \lambda_i\in[\lambda,1] λ i ∈ [ λ , 1 ] . Then the base Υ m E \Upsilon_mE Υ m E , the profiles Υ m P \Upsilon_mP Υ m P , P ∈ P P\in\mathcal{P} P ∈ P , with the wavelengths λ P \lambda_P λ P , and the profiles P i P_i P i with the wavelengths λ i \lambda_i λ i form a representation R R R of Υ m Y + ∑ i P i \Upsilon_mY+\sum_iP_i Υ m Y + ∑ i P i at wavelength λ \lambda λ , and
size λ ( R ) ≤ ∣ Υ m ∣ size λ ′ ( R ′ ) + ∑ i λ i − s ∥ P i ∥ C λ i N r e p ( T 2 ) . \operatorname{size}_\lambda(R)\leq|\Upsilon_m|\operatorname{size}_{\lambda'}(R')+\sum_i\lambda_i^{-s}\|P_i\|_{C^{N_{\mathrm{rep}}}_{\lambda_i}(\mathbb{T}^2)}. size λ ( R ) ≤ ∣ Υ m ∣ size λ ′ ( R ′ ) + i ∑ λ i − s ∥ P i ∥ C λ i N rep ( T 2 ) .
Proof. (a) For a profile P P P and ∣ j ∣ ≥ 1 > s |j| \ge1 > s ∣ j ∣ ≥ 1 > s , λ ≤ λ P ≤ 1 \lambda\le\lambda_P \le1 λ ≤ λ P ≤ 1 gives λ ∣ j ∣ − s ≤ λ P ∣ j ∣ − s \lambda^{|j|-s} \le\lambda_P^{|j|-s} λ ∣ j ∣ − s ≤ λ P ∣ j ∣ − s and 1 ≤ λ − s 1 \le\lambda^{-s} 1 ≤ λ − s ; adding the definitions (40) over the base and the profiles gives the first bound. For the second, for x , y ∈ T 2 x,y \in\mathbb{T}^2 x , y ∈ T 2 ,
∣ P ( x ) − P ( y ) ∣ ω ≤ min { 2 ∥ P ∥ L ∞ ( T 2 ) , ( ∥ ∂ 1 P ∥ L ∞ ( T 2 ) + ∥ ∂ 2 P ∥ L ∞ ( T 2 ) ) ∣ x − y ∣ T 2 } |P(x)-P(y)|_\omega\le\min\left\{2\|P\|_{L^\infty(\mathbb{T}^2)},\left(\|\partial_1P\|_{L^\infty(\mathbb{T}^2)}+\|\partial_2P\|_{L^\infty(\mathbb{T}^2)}\right)|x-y|_{\mathbb{T}^2}\right\} ∣ P ( x ) − P ( y ) ∣ ω ≤ min { 2∥ P ∥ L ∞ ( T 2 ) , ( ∥ ∂ 1 P ∥ L ∞ ( T 2 ) + ∥ ∂ 2 P ∥ L ∞ ( T 2 ) ) ∣ x − y ∣ T 2 }
≤ 2 λ P − s ∥ P ∥ C λ P N r e p ( T 2 ) ∣ x − y ∣ T 2 , \le2\lambda_P^{-s}\|P\|_{C^{N_{\mathrm{rep}}}_{\lambda_P}(\mathbb{T}^2)}|x-y|_{\mathbb{T}^2}, ≤ 2 λ P − s ∥ P ∥ C λ P N rep ( T 2 ) ∣ x − y ∣ T 2 ,
since λ P ( ∥ ∂ 1 P ∥ L ∞ ( T 2 ) + ∥ ∂ 2 P ∥ L ∞ ( T 2 ) ) ≤ [ P ] C λ P N r e p ( T 2 ) \lambda_P(\|\partial_1P\|_{L^\infty(\mathbb{T}^2)}+\|\partial_2P\|_{L^\infty(\mathbb{T}^2)}) \le[P]_{C^{N_{\mathrm{rep}}}_{\lambda_P}(\mathbb{T}^2)} λ P ( ∥ ∂ 1 P ∥ L ∞ ( T 2 ) + ∥ ∂ 2 P ∥ L ∞ ( T 2 ) ) ≤ [ P ] C λ P N rep ( T 2 ) and s < 1 s<1 s < 1 , and adding [ E ] C 0 , s ( T 2 ) [E]_{C^{0,s}(\mathbb{T}^2)} [ E ] C 0 , s ( T 2 ) and the first bound gives ∥ U ∥ C 0 , s ( T 2 ) ≤ 3 size λ ( R ) \|U\|_{C^{0,s}(\mathbb{T}^2)} \le3\operatorname{size}_\lambda(R) ∥ U ∥ C 0 , s ( T 2 ) ≤ 3 size λ ( R ) .
(b) The readout, a constant linear map, commutes with derivatives and at every point costs at most its operator norm ∣ Υ m ∣ |\Upsilon_m| ∣ Υ m ∣ for ∣ ⋅ ∣ ω |\cdot|_\omega ∣ ⋅ ∣ ω , hence at most ∣ Υ m ∣ |\Upsilon_m| ∣ Υ m ∣ in each term of (41) . The base, measured at λ ≤ λ ′ \lambda\le\lambda' λ ≤ λ ′ , only decreases by the monotonicity of (a), and the profiles P i P_i P i add their terms. □
Lemma 14.3 turns a Sobolev bound into a profile: N r e p + 2 N_{\mathrm{rep}}+2 N rep + 2 derivatives in L 2 L^2 L 2 at a scale ρ \rho ρ , read at a wavelength λ \lambda λ , cost the fixed power λ − Z \lambda^{-Z} λ − Z where the maximum in (185) is bounded.
Lemma 14.3 (Sobolev bounds read at a wavelength). There exists C ( N r e p ) < ∞ C(N_{\mathrm{rep}})<\infty C ( N rep ) < ∞ such that, for every 0 < λ ≤ 1 0<\lambda\le1 0 < λ ≤ 1 , 0 < ρ ≤ 1 0<\rho\le1 0 < ρ ≤ 1 and every field F ∈ H N r e p + 2 ( T 2 ) V F\in H^{N_{\mathrm{rep}}+2}(\mathbb{T}^2)^V F ∈ H N rep + 2 ( T 2 ) V of a family,
λ − s ∥ F ∥ C λ N r e p ( T 2 ) ≤ C max { 1 , λ 1 + γ / 2 + δ ρ } N r e p + 1 λ − Z ∥ F ∥ H ρ , ω N r e p + 2 ( T 2 ) . (185) \lambda^{-s}\|F\|_{C^{N_{\mathrm{rep}}}_\lambda(\mathbb{T}^2)}
\le C\max\left\{1,\frac{\lambda^{1+\gamma/2+\delta}}{\rho}\right\}^{N_{\mathrm{rep}}+1}\lambda^{-Z}\|F\|_{H^{N_{\mathrm{rep}}+2}_{\rho,\omega}(\mathbb{T}^2)}. \tag*{(185)} λ − s ∥ F ∥ C λ N rep ( T 2 ) ≤ C max { 1 , ρ λ 1 + γ /2 + δ } N rep + 1 λ − Z ∥ F ∥ H ρ , ω N rep + 2 ( T 2 ) . (185) The pair ( F , ∅ ) (F,\varnothing) ( F , ∅ ) , with base F F F and no profile, is a representation of F F F at wavelength λ \lambda λ with size λ ( ( F , ∅ ) ) ≤ 4 λ − s ∥ F ∥ C λ N r e p ( T 2 ) \operatorname{size}_\lambda((F,\varnothing))\le4\lambda^{-s}\|F\|_{C^{N_{\mathrm{rep}}}_\lambda(\mathbb{T}^2)} size λ (( F , ∅ )) ≤ 4 λ − s ∥ F ∥ C λ N rep ( T 2 ) .
Proof. For a scalar function g g g , a multi-index j j j and ρ ≤ 1 \rho\le1 ρ ≤ 1 , the Cauchy–Schwarz inequality in Fourier gives ∥ ∂ j g ∥ L ∞ ( T 2 ) ≤ C ρ − ∣ j ∣ − 1 ∥ g ∥ H ρ ∣ j ∣ + 2 ( T 2 ) \|\partial^j g\|_{L^\infty(\mathbb{T}^2)}\le C\rho^{-|j|-1}\|g\|_{H^{|j|+2}_\rho(\mathbb{T}^2)} ∥ ∂ j g ∥ L ∞ ( T 2 ) ≤ C ρ − ∣ j ∣ − 1 ∥ g ∥ H ρ ∣ j ∣ + 2 ( T 2 ) . Applied to each component, it bounds the left side of (185) by C ∑ i = 0 N r e p λ i − s ρ − i − 1 ∥ F ∥ H ρ , ω N r e p + 2 ( T 2 ) C\sum_{i=0}^{N_{\mathrm{rep}}}\lambda^{i-s}\rho^{-i-1}\|F\|_{H^{N_{\mathrm{rep}}+2}_{\rho,\omega}(\mathbb{T}^2)} C ∑ i = 0 N rep λ i − s ρ − i − 1 ∥ F ∥ H ρ , ω N rep + 2 ( T 2 ) , and for 0 ≤ i ≤ N r e p 0\le i\le N_{\mathrm{rep}} 0 ≤ i ≤ N rep , since λ ≤ 1 \lambda\le1 λ ≤ 1 ,
λ i − s ρ − i − 1 = ( λ 1 + γ / 2 + δ ρ ) i + 1 λ − ( 1 + s + ( γ / 2 + δ ) ( i + 1 ) ) ≤ max { 1 , λ 1 + γ / 2 + δ ρ } N r e p + 1 λ − Z . \lambda^{i-s}\rho^{-i-1}
=\left(\frac{\lambda^{1+\gamma/2+\delta}}{\rho}\right)^{i+1}\lambda^{-(1+s+(\gamma/2+\delta)(i+1))}
\le\max\left\{1,\frac{\lambda^{1+\gamma/2+\delta}}{\rho}\right\}^{N_{\mathrm{rep}}+1}\lambda^{-Z}. λ i − s ρ − i − 1 = ( ρ λ 1 + γ /2 + δ ) i + 1 λ − ( 1 + s + ( γ /2 + δ ) ( i + 1 )) ≤ max { 1 , ρ λ 1 + γ /2 + δ } N rep + 1 λ − Z . Finally, F ∈ C N r e p ( T 2 ) V F\in C^{N_{\mathrm{rep}}}(\mathbb{T}^2)^V F ∈ C N rep ( T 2 ) V by the Sobolev embedding, and the size of ( F , ∅ ) (F,\varnothing) ( F , ∅ ) is bounded as in the proof of Lemma 14.2 (a). □
At the time t t t , the fresh smoothing of the new sources of a successor created at level m m m is the product of two averages (44) along ellipses of width r m r_m r m ,
S ( t ) : = S ℓ , r m , Ω 1 ( t ) S ℓ , r m , Ω 2 ( t ) , (186) S(t):=S_{\ell,r_m,\Omega_1(t)}S_{\ell,r_m,\Omega_2(t)}, \tag*{(186)} S ( t ) := S ℓ , r m , Ω 1 ( t ) S ℓ , r m , Ω 2 ( t ) , (186) where ℓ \ell ℓ is a fixed integer with 2 ℓ > N m o m 2\ell>N_{\mathrm{mom}} 2 ℓ > N mom , depending only on α \alpha α , s s s , q q q , ζ \zeta ζ , Λ \Lambda Λ , n n n , t 0 t_0 t 0 , T T T , and Ω 1 ( t ) \Omega_1(t) Ω 1 ( t ) , Ω 2 ( t ) \Omega_2(t) Ω 2 ( t ) are the maps of the ellipses (11.18). By Step 7 of the proof of Lemma 11.7 , they are invertible with ∣ Ω i ( t ) ∣ ∞ ≤ C Ω |\Omega_i(t)|_\infty\le C_\Omega ∣ Ω i ( t ) ∣ ∞ ≤ C Ω , and, at the levels finer than the top level of Subsection 11.1, their long axes Ω 1 ( t ) e 1 \Omega_1(t)e_1 Ω 1 ( t ) e 1 , Ω 2 ( t ) e 2 \Omega_2(t)e_2 Ω 2 ( t ) e 2 are uniformly transverse in the sense of Proposition 6.1 .
A passage splits Y ( t ) Y(t) Y ( t ) with this smoothing. By Lemma 14.4(a), S ( t ) S(t) S ( t ) moves derivatives onto its kernel at the cost r m − 1 r_m^{-1} r m − 1 each, uniformly in the eccentricity, as every derivative combines derivatives along the two long axes; by (b), the vanishing moments of the kernel make ( I − S ) Y ( t ) (I-S)Y(t) ( I − S ) Y ( t ) small.
Lemma 14.4 (The smoothing). There exists C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ C(\alpha,s,q,\zeta,\Lambda,n,t_0,T)<\infty C ( α , s , q , ζ , Λ , n , t 0 , T ) < ∞ such that the following holds for every M ≥ 1 M\ge1 M ≥ 1 , κ ∈ K M \kappa\in\mathcal{K}_M κ ∈ K M , m 0 < m ≤ M m_0<m\le M m 0 < m ≤ M , t ∈ R t\in\mathbb{R} t ∈ R and every successor G ′ G' G ′ of a family of level m m m .
(a) For every pair of integers 1 ≤ N 1 ≤ N 2 ≤ N r e p + N c o r + 2 p 1 \le N_1 \le N_2 \le N_{\mathrm{rep}}+N_{\mathrm{cor}}+2p 1 ≤ N 1 ≤ N 2 ≤ N rep + N cor + 2 p and every field F F F of G ′ G' G ′ of class C N 1 C^{N_1} C N 1 on T 2 \mathbb{T}^2 T 2 ,
∑ ∣ i ∣ = N 2 ∥ ∂ i S ( t ) F ∥ L ∞ ( T 2 ) ≤ C r m N 1 − N 2 ∑ ∣ j ∣ = N 1 ∥ ∂ j F ∥ L ∞ ( T 2 ) . \sum_{|i|=N_2}\|\partial^iS(t)F\|_{L^\infty(\mathbb{T}^2)}\le C r_m^{N_1-N_2}\sum_{|j|=N_1}\|\partial^jF\|_{L^\infty(\mathbb{T}^2)}. ∣ i ∣ = N 2 ∑ ∥ ∂ i S ( t ) F ∥ L ∞ ( T 2 ) ≤ C r m N 1 − N 2 ∣ j ∣ = N 1 ∑ ∥ ∂ j F ∥ L ∞ ( T 2 ) .
(b) For every integer 0 ≤ N ≤ N r e t − N m o m − 2 0\le N\le N_{\mathrm{ret}}-N_{\mathrm{mom}}-2 0 ≤ N ≤ N ret − N mom − 2 , every ρ ≥ r m \rho\ge r_m ρ ≥ r m and every field F ∈ H N + N m o m + 2 ( T 2 ) V ′ F\in H^{N+N_{\mathrm{mom}}+2}(\mathbb{T}^2)^{V'} F ∈ H N + N mom + 2 ( T 2 ) V ′ of G ′ G' G ′ , where V ′ V' V ′ is the set of components of G ′ G' G ′ ,
∥ F − S ( t ) F ∥ H ρ , G ′ N ( T 2 ) ≤ C ( r m ρ ) N m o m ∥ F ∥ H ρ , G ′ N + N m o m + 2 ( T 2 ) . \|F-S(t)F\|_{H^N_{\rho,G'}(\mathbb{T}^2)}\le C\left(\frac{r_m}{\rho}\right)^{N_{\mathrm{mom}}}\|F\|_{H^{N+N_{\mathrm{mom}}+2}_{\rho,G'}(\mathbb{T}^2)}. ∥ F − S ( t ) F ∥ H ρ , G ′ N ( T 2 ) ≤ C ( ρ r m ) N mom ∥ F ∥ H ρ , G ′ N + N mom + 2 ( T 2 ) .
Proof. (a) This is Step 1 of the proof of Proposition 6.1 , with r m r_m r m in place of r r r , Ω i ( t ) \Omega_i(t) Ω i ( t ) in place of Ω i \Omega_i Ω i , and L ∞ ( T 2 ) L^\infty(\mathbb{T}^2) L ∞ ( T 2 ) in place of L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) , applied to the derivatives ∂ j F \partial^jF ∂ j F , ∣ j ∣ = N 1 |j|=N_1 ∣ j ∣ = N 1 : every ∂ i \partial^i ∂ i with ∣ i ∣ = N 2 |i|=N_2 ∣ i ∣ = N 2 is a combination of products of N 2 − N 1 N_2-N_1 N 2 − N 1 derivatives ∂ ( 1 ) \partial_{(1)} ∂ ( 1 ) or ∂ ( 2 ) \partial_{(2)} ∂ ( 2 ) with one ∂ j \partial^j ∂ j , ∣ j ∣ = N 1 |j|=N_1 ∣ j ∣ = N 1 , with coefficients that are products of entries of the inverse of the matrix of the long axes Ω 1 ( t ) e 1 \Omega_1(t)e_1 Ω 1 ( t ) e 1 , Ω 2 ( t ) e 2 \Omega_2(t)e_2 Ω 2 ( t ) e 2 , which are at most 2 2 2 in absolute value. Only this inverse enters, not the inverse of the short axes. Each ∂ ( 1 ) \partial_{(1)} ∂ ( 1 ) or ∂ ( 2 ) \partial_{(2)} ∂ ( 2 ) costs r m − 1 r_m^{-1} r m − 1 , and C C C depends only on N 2 N_2 N 2 and on norms in L 1 ( R 2 ) L^1(\mathbb{R}^2) L 1 ( R 2 ) of derivatives of W ℓ W_\ell W ℓ of orders at most N 2 − N 1 N_2-N_1 N 2 − N 1 . (b) is the third bound of Proposition 6.1 at the order N m o m < 2 ℓ N_{\mathrm{mom}}<2\ell N mom < 2 ℓ , applied to the derivatives ∂ j F v \partial^jF_v ∂ j F v , ∣ j ∣ ≤ N |j|\le N ∣ j ∣ ≤ N , which commute with S ( t ) S(t) S ( t ) , and summed with the weights of H ρ , G ′ N ( T 2 ) H^N_{\rho,G'}(\mathbb{T}^2) H ρ , G ′ N ( T 2 ) ; the factorial weights cost a constant depending on N + N m o m ≤ N r e t N+N_{\mathrm{mom}}\le N_{\mathrm{ret}} N + N mom ≤ N ret , and the constant of (b) contains C Ω N m o m C_{\Omega}^{N_{\mathrm{mom}}} C Ω N mom . Both parts use the bounds on Ω i ( t ) \Omega_i(t) Ω i ( t ) recalled after (186) , which hold for every t ∈ R t\in\mathbb{R} t ∈ R , as m > m 0 m>m_0 m > m 0 . □
Proof of Proposition 5.4
Proof of Proposition 5.4. The top level m 0 ≥ 2 m_0\ge2 m 0 ≥ 2 (so that m ≥ 2 m\ge2 m ≥ 2 , as Lemma 7.1 (a) and (184) require) is so large that, for m > m 0 m>m_0 m > m 0 , ρ ( m ) ≤ 1 \rho(m)\le1 ρ ( m ) ≤ 1 and C ε m − 1 5 η / 4 ≤ 1 C\varepsilon_{m-1}^{5\eta/4}\le1 C ε m − 1 5 η /4 ≤ 1 , with the constant C C C of Proposition 14.1 (i); then ∣ Υ m ∣ ≤ 2 |\Upsilon_m|\le2 ∣ Υ m ∣ ≤ 2 . Let t p r t_{\mathrm{pr}} t pr and Π \Pi Π be those of the paragraph after Proposition 14.1 . Fix a datum g g g , a time t ∈ [ t p r , t f ] t\in[t_{\mathrm{pr}},t_f] t ∈ [ t pr , t f ] and a representation R ′ R' R ′ of Y ( t ) Y(t) Y ( t ) at wavelength ε m − 1 \varepsilon_{m-1} ε m − 1 , and write S S S for S ( t ) S(t) S ( t ) . The proof writes U ( t ) U(t) U ( t ) as Υ m Y ( t ) \Upsilon_mY(t) Υ m Y ( t ) plus five profiles (Step 2), bounds their profile norms (Steps 3–6), and adds them (Step 7).
Step 1: Regularity. By Proposition 13.10(i), Proposition 14.1 (ii) and the Sobolev embedding, Y ( t ) Y(t) Y ( t ) is of class C N r e t − 2 C^{N_{\mathrm{ret}}-2} C N ret − 2 on T 2 \mathbb{T}^2 T 2 , with N r e t − 2 ≥ N r e p + N c o r + 2 p + 2 N_{\mathrm{ret}}-2\ge N_{\mathrm{rep}}+N_{\mathrm{cor}}+2p+2 N ret − 2 ≥ N rep + N cor + 2 p + 2 by (180) , and U ( t ) U(t) U ( t ) is of class C 2 N r e p + 2 C^{2N_{\mathrm{rep}}+2} C 2 N rep + 2 on T 2 \mathbb{T}^2 T 2 ; this is the first claim of Proposition 5.4 . Since S S S keeps regularity, exp p ( Γ m ( t ) : ∇ 2 ) \exp_p(\Gamma_m(t):\nabla^2) exp p ( Γ m ( t ) : ∇ 2 ) costs 2 p 2p 2 p derivatives and Corr m \operatorname{Corr}_m Corr m maps fields of class C N r e p + N c o r C^{N_{\mathrm{rep}}+N_{\mathrm{cor}}} C N rep + N cor to fields of class C N r e p C^{N_{\mathrm{rep}}} C N rep , the five profiles defined in Step 2 are of class C N r e p C^{N_{\mathrm{rep}}} C N rep on T 2 \mathbb{T}^2 T 2 .
Step 2: The exact split. The operator exp p ( Γ m ( t ) : ∇ 2 ) \exp_p(\Gamma_m(t):\nabla^2) exp p ( Γ m ( t ) : ∇ 2 ) acts on every component by the same scalar differential operator with constant coefficients, so it commutes with the constant matrix Υ m \Upsilon_m Υ m . For a field F F F of G ′ G' G ′ of class C N r e p + N c o r + 2 p + 2 C^{N_{\mathrm{rep}}+N_{\mathrm{cor}}+2p+2} C N rep + N cor + 2 p + 2 on T 2 \mathbb{T}^2 T 2 , define, as in (154) , F ♭ : = exp p ( Γ m ( t ) : ∇ 2 ) Υ m F F^\flat:=\exp_p(\Gamma_m(t):\nabla^2)\Upsilon_mF F ♭ := exp p ( Γ m ( t ) : ∇ 2 ) Υ m F , and its exponential and corrector parts
F exp : = Υ m ( exp p ( Γ m ( t ) : ∇ 2 ) − I ) F , F c o r : = ( Corr m − I ) F ♭ . (187) F^{\exp}:=\Upsilon_m\bigl(\exp_p(\Gamma_m(t):\nabla^2)-I\bigr)F,\qquad F^{\mathrm{cor}}:=(\operatorname{Corr}_m-I)F^\flat. \tag*{(187)} F e x p := Υ m ( exp p ( Γ m ( t ) : ∇ 2 ) − I ) F , F cor := ( Corr m − I ) F ♭ . (187)
Then the lift (38) of F F F is F ♯ = Corr m F ♭ = Υ m F + F exp + F c o r F^\sharp=\operatorname{Corr}_mF^\flat=\Upsilon_mF+F^{\exp}+F^{\mathrm{cor}} F ♯ = Corr m F ♭ = Υ m F + F e x p + F cor , since F ♭ = Υ m F + F exp F^\flat=\Upsilon_mF+F^{\exp} F ♭ = Υ m F + F e x p . Both parts are linear in F F F , so the splitting Y ( t ) = S Y ( t ) + ( I − S ) Y ( t ) Y(t)=SY(t)+(I-S)Y(t) Y ( t ) = S Y ( t ) + ( I − S ) Y ( t ) gives the exact identity
U ( t ) = Υ m Y ( t ) + ( S Y ( t ) ) exp + ( S Y ( t ) ) c o r + ( ( I − S ) Y ( t ) ) exp + ( ( I − S ) Y ( t ) ) c o r + ( U ( t ) − Y ♯ ( t ) ) . (188) \begin{aligned}
U(t)={}&\Upsilon_mY(t)+(SY(t))^{\exp}+(SY(t))^{\mathrm{cor}}\\
&+((I-S)Y(t))^{\exp}+((I-S)Y(t))^{\mathrm{cor}}+(U(t)-Y^\sharp(t)). \tag*{(188)}
\end{aligned} U ( t ) = Υ m Y ( t ) + ( S Y ( t ) ) e x p + ( S Y ( t ) ) cor + (( I − S ) Y ( t ) ) e x p + (( I − S ) Y ( t ) ) cor + ( U ( t ) − Y ♯ ( t )) . (188)
We give the two exponential parts the wavelength ε m − 1 \varepsilon_{m-1} ε m − 1 , and the two corrector parts and U ( t ) − Y ♯ ( t ) U(t)-Y^\sharp(t) U ( t ) − Y ♯ ( t ) the wavelength ε m \varepsilon_m ε m . Lemma 14.2 (b), with λ ′ = ε m − 1 \lambda'=\varepsilon_{m-1} λ ′ = ε m − 1 and λ = ε m \lambda=\varepsilon_m λ = ε m , turns R ′ R' R ′ and these five profiles into a representation R R R of U ( t ) U(t) U ( t ) at wavelength ε m \varepsilon_m ε m , whose size is at most ∣ Γ m ∣ size ε m − 1 ( R ′ ) |\Gamma_m| \operatorname{size}_{\varepsilon_{m-1}}(R') ∣ Γ m ∣ size ε m − 1 ( R ′ ) plus the five profile norms bounded in Steps 3–6.
Step 3: The exponential part of S Y ( t ) SY(t) S Y ( t ) . We show
ε m − 1 − s ∥ ( S Y ( t ) ) exp ∥ C ε m − 1 N r e p ( T 2 ) ≤ C ε m − 1 δ − 2 Θ ε m − 1 − s [ Y ( t ) ] C ε m − 1 N r e p ( T 2 ) . (189) \varepsilon_{m-1}^{-s}\left\|(SY(t))^{\exp}\right\|_{C^{N_{\mathrm{rep}}}_{\varepsilon_{m-1}}(\mathbb{T}^2)} \le C\varepsilon_{m-1}^{\delta-2\Theta}\varepsilon_{m-1}^{-s}[Y(t)]_{C^{N_{\mathrm{rep}}}_{\varepsilon_{m-1}}(\mathbb{T}^2)} . \tag*{(189)} ε m − 1 − s ∥ ( S Y ( t ) ) e x p ∥ C ε m − 1 N rep ( T 2 ) ≤ C ε m − 1 δ − 2Θ ε m − 1 − s [ Y ( t ) ] C ε m − 1 N rep ( T 2 ) . (189)
Every term carries Γ m ( t ) \Gamma_m(t) Γ m ( t ) at least once, which is small compared with r m 2 r_m^2 r m 2 by (183) , and r m 2 r_m^2 r m 2 absorbs the two derivatives that come with it: on S S S each costs r m − 1 r_m^{-1} r m − 1 , and on Y ( t ) Y(t) Y ( t ) each costs ε m − 1 − 1 ≤ C r m − 1 \varepsilon_{m-1}^{-1} \le Cr_m^{-1} ε m − 1 − 1 ≤ C r m − 1 . In detail, every term of exp p ( Γ m ( t ) : ∇ 2 ) − I \exp_p(\Gamma_m(t):\nabla^2)-I exp p ( Γ m ( t ) : ∇ 2 ) − I is ( Γ m ( t ) : ∇ 2 ) l / l ! (\Gamma_m(t):\nabla^2)^l/l! ( Γ m ( t ) : ∇ 2 ) l / l ! with 1 ≤ l ≤ p 1\le l\le p 1 ≤ l ≤ p , and ( Γ m ( t ) : ∇ 2 ) l (\Gamma_m(t):\nabla^2)^l ( Γ m ( t ) : ∇ 2 ) l is a sum of 4 l 4^l 4 l derivatives of order 2 l 2l 2 l with coefficients of absolute value at most ∣ Γ m ( t ) ∣ l |\Gamma_m(t)|^l ∣ Γ m ( t ) ∣ l . Let j j j be a multi-index with ∣ j ∣ ≤ N r e p |j|\le N_{\mathrm{rep}} ∣ j ∣ ≤ N rep , and put N 2 : = ∣ j ∣ + 2 l N_2:=|j|+2l N 2 := ∣ j ∣ + 2 l and N 1 : = min { N r e p , N 2 } ≥ 1 N_1:=\min\{N_{\mathrm{rep}},N_2\}\ge1 N 1 := min { N rep , N 2 } ≥ 1 . Lemma 14.4(a) keeps N 1 N_1 N 1 derivatives on Y ( t ) Y(t) Y ( t ) and puts the other N 2 − N 1 N_2-N_1 N 2 − N 1 on S S S :
∥ ∂ j ( Γ m ( t ) : ∇ 2 ) l S Y ( t ) ∥ L ∞ ( T 2 ) ≤ 4 l ∣ Γ m ( t ) ∣ l ∑ ∣ i ∣ = N 2 ∥ ∂ i S Y ( t ) ∥ L ∞ ( T 2 ) ≤ C ∣ Γ m ( t ) ∣ l r m N 1 − N 2 ∑ ∣ i ∣ = N 1 ∥ ∂ i Y ( t ) ∥ L ∞ ( T 2 ) . \begin{aligned}
\left\|\partial^j(\Gamma_m(t):\nabla^2)^lSY(t)\right\|_{L^\infty(\mathbb{T}^2)}
\le4^l|\Gamma_m(t)|^l\sum_{|i|=N_2}\left\|\partial^iSY(t)\right\|_{L^\infty(\mathbb{T}^2)} \\
&\le C|\Gamma_m(t)|^l r_m^{N_1-N_2}\sum_{|i|=N_1}\left\|\partial^iY(t)\right\|_{L^\infty(\mathbb{T}^2)} .
\end{aligned} ∂ j ( Γ m ( t ) : ∇ 2 ) l S Y ( t ) L ∞ ( T 2 ) ≤ 4 l ∣ Γ m ( t ) ∣ l ∣ i ∣ = N 2 ∑ ∂ i S Y ( t ) L ∞ ( T 2 ) ≤ C ∣ Γ m ( t ) ∣ l r m N 1 − N 2 ∣ i ∣ = N 1 ∑ ∂ i Y ( t ) L ∞ ( T 2 ) .
We multiply by ε m − 1 ∣ j ∣ − s \varepsilon_{m-1}^{|j|-s} ε m − 1 ∣ j ∣ − s and write ∣ Γ m ( t ) ∣ l r m N 1 − N 2 ε m − 1 ∣ j ∣ = ( ∣ Γ m ( t ) ∣ / r m 2 ) l ( r m / ε m − 1 ) N 1 − ∣ j ∣ ε m − 1 N 1 |\Gamma_m(t)|^l r_m^{N_1-N_2}\varepsilon_{m-1}^{|j|}=(|\Gamma_m(t)|/r_m^2)^l(r_m/\varepsilon_{m-1})^{N_1-|j|}\varepsilon_{m-1}^{N_1} ∣ Γ m ( t ) ∣ l r m N 1 − N 2 ε m − 1 ∣ j ∣ = ( ∣ Γ m ( t ) ∣/ r m 2 ) l ( r m / ε m − 1 ) N 1 − ∣ j ∣ ε m − 1 N 1 . Here ( ∣ Γ m ( t ) ∣ / r m 2 ) l ≤ C ε m − 1 δ − 2 Θ (|\Gamma_m(t)|/r_m^2)^l\le C\varepsilon_{m-1}^{\delta-2\Theta} ( ∣ Γ m ( t ) ∣/ r m 2 ) l ≤ C ε m − 1 δ − 2Θ by (183) and 2 Θ < δ 2\Theta<\delta 2Θ < δ , and ( r m / ε m − 1 ) N 1 − ∣ j ∣ ≤ C (r_m/\varepsilon_{m-1})^{N_1-|j|}\le C ( r m / ε m − 1 ) N 1 − ∣ j ∣ ≤ C by (184) , since N 1 ≥ ∣ j ∣ N_1\ge|j| N 1 ≥ ∣ j ∣ . As 1 ≤ N 1 ≤ N r e p 1\le N_1\le N_{\mathrm{rep}} 1 ≤ N 1 ≤ N rep , for ∣ i ∣ = N 1 |i|=N_1 ∣ i ∣ = N 1 we have ε m − 1 N 1 ∥ ∂ i Y ( t ) ∥ L ∞ ( T 2 ) ≤ i ! [ Y ( t ) ] C ε m − 1 N r e p ( T 2 ) \varepsilon_{m-1}^{N_1}\|\partial^iY(t)\|_{L^\infty(\mathbb{T}^2)}\le i![Y(t)]_{C^{N_{\mathrm{rep}}}_{\varepsilon_{m-1}}(\mathbb{T}^2)} ε m − 1 N 1 ∥ ∂ i Y ( t ) ∥ L ∞ ( T 2 ) ≤ i ! [ Y ( t ) ] C ε m − 1 N rep ( T 2 ) , so
ε m − 1 ∣ j ∣ − s ∥ ∂ j ( Γ m ( t ) : ∇ 2 ) l S Y ( t ) ∥ L ∞ ( T 2 ) ≤ C ε m − 1 δ − 2 Θ ε m − 1 − s [ Y ( t ) ] C ε m − 1 N r e p ( T 2 ) . \varepsilon_{m-1}^{|j|-s}\left\|\partial^j(\Gamma_m(t):\nabla^2)^lSY(t)\right\|_{L^\infty(\mathbb{T}^2)}
\le C\varepsilon_{m-1}^{\delta-2\Theta}\varepsilon_{m-1}^{-s}[Y(t)]_{C^{N_{\mathrm{rep}}}_{\varepsilon_{m-1}}(\mathbb{T}^2)} . ε m − 1 ∣ j ∣ − s ∂ j ( Γ m ( t ) : ∇ 2 ) l S Y ( t ) L ∞ ( T 2 ) ≤ C ε m − 1 δ − 2Θ ε m − 1 − s [ Y ( t ) ] C ε m − 1 N rep ( T 2 ) .
The readout commutes with derivatives and ∣ Γ m ∣ ≤ 2 |\Gamma_m|\le2 ∣ Γ m ∣ ≤ 2 ; dividing by j ! j! j ! and summing over ∣ j ∣ ≤ N r e p |j|\le N_{\mathrm{rep}} ∣ j ∣ ≤ N rep and 1 ≤ l ≤ p 1\le l\le p 1 ≤ l ≤ p gives (189) .
Step 4: The corrector part of S Y ( t ) SY(t) S Y ( t ) . We show
ε m s ∥ ( S Y ( t ) ) c o r ∥ C ε m N r e p ( T 2 ) ≤ C ε m − 1 μ ε m − 1 − s [ Y ( t ) ] C ε m − 1 N r e p ( T 2 ) . (190) \varepsilon_m^s\left\|(SY(t))^{\mathrm{cor}}\right\|_{C^{N_{\mathrm{rep}}}_{\varepsilon_m}(\mathbb{T}^2)}
\le C\varepsilon_{m-1}^{\mu}\varepsilon_{m-1}^{-s}[Y(t)]_{C^{N_{\mathrm{rep}}}_{\varepsilon_{m-1}}(\mathbb{T}^2)} . \tag*{(190)} ε m s ∥ ( S Y ( t ) ) cor ∥ C ε m N rep ( T 2 ) ≤ C ε m − 1 μ ε m − 1 − s [ Y ( t ) ] C ε m − 1 N rep ( T 2 ) . (190)
The correction costs the corrector amplitude times one derivative of Y ( t ) Y(t) Y ( t ) at the wavelength ε m − 1 \varepsilon_{m-1} ε m − 1 ; the length ρ ( m ) \rho(m) ρ ( m ) cancels, and (30) says that this is small when measured at the wavelength ε m \varepsilon_m ε m . This is where s < 1 2 ( 1 − α ) s<\frac{1}{2}(1-\alpha) s < 2 1 ( 1 − α ) is used. We apply Lemma 13.3 to the field ( S Y ( t ) ) ♭ (SY(t))^\flat ( S Y ( t ) ) ♭ of G G G . Its derivatives at the scale ρ ( m ) \rho(m) ρ ( m ) are controlled by the seminorm of Y ( t ) Y(t) Y ( t ) at the wavelength ε m − 1 \varepsilon_{m-1} ε m − 1 :
∑ 1 ≤ ∣ i ∣ ≤ N r e p + N c o r ρ ( m ) ∣ i ∣ ∥ ∂ i ( S Y ( t ) ) ♭ ∥ L ∞ ( T 2 ) ≤ C ρ ( m ) ε m − 1 [ Y ( t ) ] C ε m − 1 N r e p ( T 2 ) . (191) \sum_{1\le|i|\le N_{\mathrm{rep}}+N_{\mathrm{cor}}}\rho(m)^{|i|}\left\|\partial^i(SY(t))^\flat\right\|_{L^\infty(\mathbb{T}^2)}
\le C\frac{\rho(m)}{\varepsilon_{m-1}}[Y(t)]_{C^{N_{\mathrm{rep}}}_{\varepsilon_{m-1}}(\mathbb{T}^2)} . \tag*{(191)} 1 ≤ ∣ i ∣ ≤ N rep + N cor ∑ ρ ( m ) ∣ i ∣ ∂ i ( S Y ( t ) ) ♭ L ∞ ( T 2 ) ≤ C ε m − 1 ρ ( m ) [ Y ( t ) ] C ε m − 1 N rep ( T 2 ) . (191)
Indeed, we expand exp p ( Γ m ( t ) : ∇ 2 ) \exp_p(\Gamma_m(t):\nabla^2) exp p ( Γ m ( t ) : ∇ 2 ) as in Step 3, now with 0 ≤ l ≤ p 0\le l\le p 0 ≤ l ≤ p ; the readout commutes with derivatives, ∣ Γ m ∣ ≤ 2 |\Gamma_m|\le2 ∣ Γ m ∣ ≤ 2 , and ∣ Γ m ( t ) ∣ ≤ C r m 2 |\Gamma_m(t)|\le Cr_m^2 ∣ Γ m ( t ) ∣ ≤ C r m 2 by (183) , as 2 Θ < δ 2\Theta<\delta 2Θ < δ . For a multi-index i i i with 1 ≤ ∣ i ∣ ≤ N r e p + N c o r 1\le|i|\le N_{\mathrm{rep}}+N_{\mathrm{cor}} 1 ≤ ∣ i ∣ ≤ N rep + N cor , Lemma 14.4(a) with N 1 : = min { N r e p , ∣ i ∣ + 2 l } ≥ 1 N_1:=\min\{N_{\mathrm{rep}},|i|+2l\}\ge1 N 1 := min { N rep , ∣ i ∣ + 2 l } ≥ 1 bounds the term of degree l l l by C ρ ( m ) ∣ i ∣ r m N 1 − ∣ i ∣ ∑ ∣ j ∣ = N 1 ∥ ∂ j Y ( t ) ∥ L ∞ ( T 2 ) C\rho(m)^{|i|}r_m^{N_1-|i|}\sum_{|j|=N_1}\|\partial^jY(t)\|_{L^\infty(\mathbb{T}^2)} C ρ ( m ) ∣ i ∣ r m N 1 − ∣ i ∣ ∑ ∣ j ∣ = N 1 ∥ ∂ j Y ( t ) ∥ L ∞ ( T 2 ) . If ∣ i ∣ ≤ N r e p |i|\le N_{\mathrm{rep}} ∣ i ∣ ≤ N rep , then N 1 ≥ ∣ i ∣ N_1\ge|i| N 1 ≥ ∣ i ∣ and ρ ( m ) ∣ i ∣ r m N 1 − ∣ i ∣ ≤ ρ ( m ) N 1 ≤ C ρ ( m ) ε m − 1 N 1 − 1 \rho(m)^{|i|}r_m^{N_1-|i|}\le\rho(m)^{N_1}\le C\rho(m)\varepsilon_{m-1}^{N_1-1} ρ ( m ) ∣ i ∣ r m N 1 − ∣ i ∣ ≤ ρ ( m ) N 1 ≤ C ρ ( m ) ε m − 1 N 1 − 1 by (184) . If ∣ i ∣ > N r e p |i|>N_{\mathrm{rep}} ∣ i ∣ > N rep , then N 1 = N r e p N_1=N_{\mathrm{rep}} N 1 = N rep , and the ∣ i ∣ − N r e p ≤ N c o r |i|-N_{\mathrm{rep}}\le N_{\mathrm{cor}} ∣ i ∣ − N rep ≤ N cor derivatives beyond N r e p N_{\mathrm{rep}} N rep fall on S S S , each at the cost ρ ( m ) / r m = ε m − 1 − Θ \rho(m)/r_m=\varepsilon_{m-1}^{-\Theta} ρ ( m ) / r m = ε m − 1 − Θ . The power ρ ( m ) N r e p − 1 ≤ C ε m − 1 N r e p − 1 ε m − 1 ( γ / 2 + δ ) ( N r e p − 1 ) \rho(m)^{N_{\mathrm{rep}}-1}\le C\varepsilon_{m-1}^{N_{\mathrm{rep}}-1}\varepsilon_{m-1}^{(\gamma/2+\delta)(N_{\mathrm{rep}}-1)} ρ ( m ) N rep − 1 ≤ C ε m − 1 N rep − 1 ε m − 1 ( γ /2 + δ ) ( N rep − 1 ) (by (184) ) absorbs them:
ρ ( m ) ∣ i ∣ r m N r e p − ∣ i ∣ = ρ ( m ) N r e p ( ρ ( m ) r m ) ∣ i ∣ − N r e p ≤ C ρ ( m ) ε m − 1 N r e p − 1 ε m − 1 ( γ / 2 + δ ) ( N r e p − 1 ) − Θ N c o r , \rho(m)^{|i|}r_m^{N_{\mathrm{rep}}-|i|}
=\rho(m)^{N_{\mathrm{rep}}}\left(\frac{\rho(m)}{r_m}\right)^{|i|-N_{\mathrm{rep}}}
\le C\rho(m)\varepsilon_{m-1}^{N_{\mathrm{rep}}-1}\varepsilon_{m-1}^{(\gamma/2+\delta)(N_{\mathrm{rep}}-1)-\Theta N_{\mathrm{cor}}} , ρ ( m ) ∣ i ∣ r m N rep − ∣ i ∣ = ρ ( m ) N rep ( r m ρ ( m ) ) ∣ i ∣ − N rep ≤ C ρ ( m ) ε m − 1 N rep − 1 ε m − 1 ( γ /2 + δ ) ( N rep − 1 ) − Θ N cor ,
and the last exponent is nonnegative by (181) . In both cases the term is at most
C ρ ( m ) ε m − 1 ε m − 1 N 1 ∑ ∣ j ∣ = N 1 ∥ ∂ j Y ( t ) ∥ L ∞ ( T 2 ) ≤ C ρ ( m ) ε m − 1 [ Y ( t ) ] C ε m − 1 N r e p ( T 2 ) . C\frac{\rho(m)}{\varepsilon_{m-1}}\varepsilon_{m-1}^{N_1}\sum_{|j|=N_1}\|\partial^jY(t)\|_{L^\infty(\mathbb{T}^2)}\le C\frac{\rho(m)}{\varepsilon_{m-1}}[Y(t)]_{C_{\varepsilon_{m-1}}^{N_{\mathrm{rep}}(\mathbb{T}^2)}}. C ε m − 1 ρ ( m ) ε m − 1 N 1 ∣ j ∣ = N 1 ∑ ∥ ∂ j Y ( t ) ∥ L ∞ ( T 2 ) ≤ C ε m − 1 ρ ( m ) [ Y ( t ) ] C ε m − 1 N rep ( T 2 ) .
which proves (191) . Now Lemma 13.3 and (191) give, for ∣ j ∣ ≤ N r e p |j|\le N_{\mathrm{rep}} ∣ j ∣ ≤ N rep ,
ε m ∣ j ∣ − s ∥ ∂ j ( S Y ( t ) ) c o r ∥ L ∞ ( T 2 ) ≤ C ε m − s ε ˉ m ρ ( m ) ρ ( m ) ε m − 1 [ Y ( t ) ] C ε m − 1 N r e p ( T 2 ) = C ( ε m − s ε ˉ m ε m − 1 s − 1 ) ε m − 1 − s [ Y ( t ) ] C ε m − 1 N r e p ( T 2 ) . \begin{aligned}
\varepsilon_m^{|j|-s}\|\partial^j(SY(t))^{\mathrm{cor}}\|_{L^\infty(\mathbb{T}^2)}
\le C\varepsilon_m^{-s}\frac{\bar{\varepsilon}_m}{\rho(m)}\frac{\rho(m)}{\varepsilon_{m-1}}[Y(t)]_{C_{\varepsilon_{m-1}}^{N_{\mathrm{rep}}(\mathbb{T}^2)}} \\
&= C\left(\varepsilon_m^{-s}\bar{\varepsilon}_m\varepsilon_{m-1}^{s-1}\right)\varepsilon_{m-1}^{-s}[Y(t)]_{C_{\varepsilon_{m-1}}^{N_{\mathrm{rep}}(\mathbb{T}^2)}}.
\end{aligned} ε m ∣ j ∣ − s ∥ ∂ j ( S Y ( t ) ) cor ∥ L ∞ ( T 2 ) ≤ C ε m − s ρ ( m ) ε ˉ m ε m − 1 ρ ( m ) [ Y ( t ) ] C ε m − 1 N rep ( T 2 ) = C ( ε m − s ε ˉ m ε m − 1 s − 1 ) ε m − 1 − s [ Y ( t ) ] C ε m − 1 N rep ( T 2 ) .
and the first factor is at most C ε m − 1 μ C\varepsilon_{m-1}^{\mu} C ε m − 1 μ by (30) . Summing over j j j proves (190) .
Step 5: The parts of ( I − S ) Y ( t ) (I-S)Y(t) ( I − S ) Y ( t ) . We show
ε m − 1 − s ∥ ( I − S ) Y ( t ) e x p ∥ C ε m − 1 N r e p ( T 2 ) + ε m − s ∥ ( I − S ) Y ( t ) c o r ∥ C ε m N r e p ( T 2 ) ≤ C ε m − 1 2 ∥ g ∥ L G 2 ( T 2 ) . (192) \varepsilon_{m-1}^{-s}\|(I-S)Y(t)^{\mathrm{exp}}\|_{C_{\varepsilon_{m-1}}^{N_{\mathrm{rep}}(\mathbb{T}^2)}}+\varepsilon_m^{-s}\|(I-S)Y(t)^{\mathrm{cor}}\|_{C_{\varepsilon_m}^{N_{\mathrm{rep}}(\mathbb{T}^2)}}\le C\varepsilon_{m-1}^{2}\|g\|_{L_G^2(\mathbb{T}^2)}. \tag*{(192)} ε m − 1 − s ∥ ( I − S ) Y ( t ) exp ∥ C ε m − 1 N rep ( T 2 ) + ε m − s ∥ ( I − S ) Y ( t ) cor ∥ C ε m N rep ( T 2 ) ≤ C ε m − 1 2 ∥ g ∥ L G 2 ( T 2 ) . (192)
By Lemma 14.4(b) with ρ = ρ ( m ) \rho=\rho(m) ρ = ρ ( m ) , the bound r m / ρ ( m ) ≤ ε m − 1 Θ r_m/\rho(m)\le\varepsilon_{m-1}^{\Theta} r m / ρ ( m ) ≤ ε m − 1 Θ , (180) and Proposition 13.10(i),
∥ ( I − S ) Y ( t ) ∥ H ρ ( m ) , G ′ N r e p + N c o r + 2 p + 2 ( T 2 ) ≤ C ε m − 1 Θ N m o m ∥ Y ( t ) ∥ H ρ ( m ) , G ′ N r e t ( T 2 ) ≤ C ε m − 1 Θ N m o m ∥ g ∥ L G 2 ( T 2 ) . (193) \|(I-S)Y(t)\|_{H_{\rho(m),G'}^{N_{\mathrm{rep}}+N_{\mathrm{cor}}+2p+2}(\mathbb{T}^2)}
\le C\varepsilon_{m-1}^{\Theta N_{\mathrm{mom}}}\|Y(t)\|_{H_{\rho(m),G'}^{N_{\mathrm{ret}}(\mathbb{T}^2)}
\le C\varepsilon_{m-1}^{\Theta N_{\mathrm{mom}}}\|g\|_{L_G^2(\mathbb{T}^2)}.} \tag*{(193)} ∥ ( I − S ) Y ( t ) ∥ H ρ ( m ) , G ′ N rep + N cor + 2 p + 2 ( T 2 ) ≤ C ε m − 1 Θ N mom ∥ Y ( t ) ∥ H ρ ( m ) , G ′ N ret ( T 2 ) ≤ C ε m − 1 Θ N mom ∥ g ∥ L G 2 ( T 2 ) . (193)
For the exponential part, in the norms H ρ ( m ) , ω N ( T 2 ) H_{\rho(m),\omega}^{N}(\mathbb{T}^2) H ρ ( m ) , ω N ( T 2 ) each operator Γ m ( t ) : ∇ 2 \Gamma_m(t):\nabla^2 Γ m ( t ) : ∇ 2 costs two orders (Lemma 8.1(b)) and the factor C ∣ Γ m ( t ) ∣ / ρ ( m ) 2 ≤ C ε m − 1 δ C|\Gamma_m(t)|/\rho(m)^2\le C\varepsilon_{m-1}^{\delta} C ∣ Γ m ( t ) ∣/ ρ ( m ) 2 ≤ C ε m − 1 δ (Lemma 10.3(b)), and the readout costs ∣ Υ m ∣ ≤ 2 |\Upsilon_m|\le2 ∣ Υ m ∣ ≤ 2 . Lemma 14.3 with λ = ε m − 1 \lambda=\varepsilon_{m-1} λ = ε m − 1 and ρ = ρ ( m ) \rho=\rho(m) ρ = ρ ( m ) , where ε m − 1 1 + γ / 2 + δ / ρ ( m ) ≤ C \varepsilon_{m-1}^{1+\gamma/2+\delta}/\rho(m)\le C ε m − 1 1 + γ /2 + δ / ρ ( m ) ≤ C by (184) , the bound (182) and (193) give
ε m − 1 − s ∥ ( I − S ) Y ( t ) e x p ∥ C ε m − 1 N r e p ( T 2 ) ≤ C ε m − 1 − Z ∥ ( I − S ) Y ( t ) e x p ∥ H ρ ( m ) , ω N r e p + 2 ( T 2 ) ≤ C ε m − 1 δ − Z ∥ ( I − S ) Y ( t ) ∥ H ρ ( m ) , G ′ N r e p + 2 p + 2 ( T 2 ) ≤ C ε m − 1 Θ N m o m − Z ∥ g ∥ L G 2 ( T 2 ) . \begin{aligned}
\varepsilon_{m-1}^{-s}\|(I-S)Y(t)^{\mathrm{exp}}\|_{C_{\varepsilon_{m-1}}^{N_{\mathrm{rep}}(\mathbb{T}^2)}}
\le C\varepsilon_{m-1}^{-Z}\|(I-S)Y(t)^{\mathrm{exp}}\|_{H_{\rho(m),\omega}^{N_{\mathrm{rep}}+2}(\mathbb{T}^2)} \\
&\le C\varepsilon_{m-1}^{\delta-Z}\|(I-S)Y(t)\|_{H_{\rho(m),G'}^{N_{\mathrm{rep}}+2p+2}(\mathbb{T}^2)}
\le C\varepsilon_{m-1}^{\Theta N_{\mathrm{mom}}-Z}\|g\|_{L_G^2(\mathbb{T}^2)}.
\end{aligned} ε m − 1 − s ∥ ( I − S ) Y ( t ) exp ∥ C ε m − 1 N rep ( T 2 ) ≤ C ε m − 1 − Z ∥ ( I − S ) Y ( t ) exp ∥ H ρ ( m ) , ω N rep + 2 ( T 2 ) ≤ C ε m − 1 δ − Z ∥ ( I − S ) Y ( t ) ∥ H ρ ( m ) , G ′ N rep + 2 p + 2 ( T 2 ) ≤ C ε m − 1 Θ N mom − Z ∥ g ∥ L G 2 ( T 2 ) .
For the corrector part, the inequality of the proof of Lemma 14.3, with the costs of exp p ( Γ m ( t ) : ∇ 2 ) \exp_p(\Gamma_m(t):\nabla^2) exp p ( Γ m ( t ) : ∇ 2 ) and Υ m \Upsilon_m Υ m just used and (182) , bounds ρ ( m ) ∣ i ∣ ∥ ∂ i ( ( I − S ) Y ( t ) ) ♭ ∥ L ∞ ( T 2 ) \rho(m)^{|i|}\|\partial^i((I-S)Y(t))^\flat\|_{L^\infty(\mathbb{T}^2)} ρ ( m ) ∣ i ∣ ∥ ∂ i (( I − S ) Y ( t ) ) ♭ ∥ L ∞ ( T 2 ) , for ∣ i ∣ ≤ N r e p + N c o r |i|\le N_{\mathrm{rep}}+N_{\mathrm{cor}} ∣ i ∣ ≤ N rep + N cor , by C ρ ( m ) C\rho(m) C ρ ( m ) times the left side of (193) . Moreover, ε ˉ m ≤ C ρ ( m ) \bar{\varepsilon}_m\le C\rho(m) ε ˉ m ≤ C ρ ( m ) and ρ ( m ) ≥ c ε m \rho(m)\ge c\varepsilon_m ρ ( m ) ≥ c ε m by (159) , and ε m ≥ c ε m − 1 q \varepsilon_m\ge c\varepsilon_{m-1}^{q} ε m ≥ c ε m − 1 q by Lemma 7.1(a). Lemma 13.3 and (193) then give
ε m − s ∥ ( I − S ) Y ( t ) c o r ∥ C ε m N r e p ( T 2 ) ≤ C ε m − s ρ ( m ) − 1 ε m − 1 Θ N m o m ∥ g ∥ L G 2 ( T 2 ) ≤ C ε m − 1 Θ N m o m − q ( 1 + s ) ∥ g ∥ L G 2 ( T 2 ) . \varepsilon_m^{-s}\|(I-S)Y(t)^{\mathrm{cor}}\|_{C_{\varepsilon_m}^{N_{\mathrm{rep}}(\mathbb{T}^2)}}
\le C\varepsilon_m^{-s}\rho(m)^{-1}\varepsilon_{m-1}^{\Theta N_{\mathrm{mom}}}\|g\|_{L_G^2(\mathbb{T}^2)}
\le C\varepsilon_{m-1}^{\Theta N_{\mathrm{mom}}-q(1+s)}\|g\|_{L_G^2(\mathbb{T}^2)}. ε m − s ∥ ( I − S ) Y ( t ) cor ∥ C ε m N rep ( T 2 ) ≤ C ε m − s ρ ( m ) − 1 ε m − 1 Θ N mom ∥ g ∥ L G 2 ( T 2 ) ≤ C ε m − 1 Θ N mom − q ( 1 + s ) ∥ g ∥ L G 2 ( T 2 ) .
The moments of S S S beat both losses: by (157) , and as Q ≥ 4 + 2 q Z Q\ge4+2qZ Q ≥ 4 + 2 q Z by (180) , Z ≥ 1 + s Z\ge1+s Z ≥ 1 + s and q > 1 q>1 q > 1 ,
Θ N m o m > Q + q γ + 4 ≥ 2 + max { Z , q ( 1 + s ) } , \Theta N_{\mathrm{mom}}>Q+q\gamma+4\ge2+\max\{Z,q(1+s)\}, Θ N mom > Q + q γ + 4 ≥ 2 + max { Z , q ( 1 + s )} ,
which proves (192) .
Step 6: The comparison error. We show
ε m − s ∥ U ( t ) − Y ♯ ( t ) ∥ C ε m N r e p ( T 2 ) ≤ C ε m − 1 2 ∥ g ∥ L G 2 ( T 2 ) . (194) \varepsilon_m^{-s}\|U(t)-Y^\sharp(t)\|_{C_{\varepsilon_m}^{N_{\mathrm{rep}}(\mathbb{T}^2)}}\le C\varepsilon_{m-1}^{2}\|g\|_{L_G^2(\mathbb{T}^2)}. \tag*{(194)} ε m − s ∥ U ( t ) − Y ♯ ( t ) ∥ C ε m N rep ( T 2 ) ≤ C ε m − 1 2 ∥ g ∥ L G 2 ( T 2 ) . (194)
Proposition 14.1(ii) bounds U ( t ) − Y ♯ ( t ) U(t)-Y^\sharp(t) U ( t ) − Y ♯ ( t ) in L ω 2 ( T 2 ) L_\omega^2(\mathbb{T}^2) L ω 2 ( T 2 ) by C ε m − 1 Q ∥ g ∥ L G 2 ( T 2 ) C\varepsilon_{m-1}^{Q}\|g\|_{L_G^2(\mathbb{T}^2)} C ε m − 1 Q ∥ g ∥ L G 2 ( T 2 ) , and, with Proposition 13.10(ii), in H ρ m o l ( m ) , ω 2 N r e p + 4 ( T 2 ) H_{\rho_{\mathrm{mol}}(m),\omega}^{2N_{\mathrm{rep}}+4}(\mathbb{T}^2) H ρ mol ( m ) , ω 2 N rep + 4 ( T 2 ) by C ∥ g ∥ L G 2 ( T 2 ) C\|g\|_{L_G^2(\mathbb{T}^2)} C ∥ g ∥ L G 2 ( T 2 ) . Interpolation between these two bounds (in Fourier, by the Cauchy–Schwarz inequality) gives N r e p + 2 N_{\mathrm{rep}}+2 N rep + 2 derivatives with half the smallness:
∥ U ( t ) − Y ♯ ( t ) ∥ H ρ m o l ( m ) , ω N r e p + 2 ( T 2 ) ≤ C ε m − 1 Q / 2 ∥ g ∥ L G 2 ( T 2 ) . (195) \|U(t)-Y^\sharp(t)\|_{H_{\rho_{\mathrm{mol}}(m),\omega}^{N_{\mathrm{rep}}+2}(\mathbb{T}^2)}
\le C\varepsilon_{m-1}^{Q/2}\|g\|_{L_G^2(\mathbb{T}^2)}. \tag*{(195)} ∥ U ( t ) − Y ♯ ( t ) ∥ H ρ mol ( m ) , ω N rep + 2 ( T 2 ) ≤ C ε m − 1 Q /2 ∥ g ∥ L G 2 ( T 2 ) . (195)
Since m < M m < M m < M , Proposition 2.1 (b) gives κ m ≥ c ε m γ \kappa_m \ge c\varepsilon_m^{\gamma} κ m ≥ c ε m γ , hence, by (115) , c ε m 1 + γ / 2 + δ ≤ ρ m o l ( m ) ≤ 1 c\varepsilon_m^{1+\gamma/2+\delta} \le\rho_{\mathrm{mol}}(m) \le1 c ε m 1 + γ /2 + δ ≤ ρ mol ( m ) ≤ 1 . Lemma 14.3 with λ = ε m \lambda= \varepsilon_m λ = ε m and ρ = ρ m o l ( m ) \rho= \rho_{\mathrm{mol}}(m) ρ = ρ mol ( m ) , (195) and ε m ≥ c ε m − 1 μ \varepsilon_m \ge c\varepsilon_{m-1}^{\mu} ε m ≥ c ε m − 1 μ (Lemma 7.1 (a)) give
ε m − s ∥ U ( t ) − Y ♯ ( t ) ∥ C ε m N r e p ( T 2 ) ≤ C ε m − 1 − Z ε m Q / 2 ∥ g ∥ L G 2 ( T 2 ) ≤ C ε m − 1 Q / 2 − q Z ∥ g ∥ L G 2 ( T 2 ) , \varepsilon_m^{-s}\lVert U(t)-Y^{\sharp}(t)\rVert_{C_{\varepsilon_m}^{N_{\mathrm{rep}}}(\mathsf{T}_2)} \le C\varepsilon_{m-1}^{-Z}\varepsilon_m^{Q/2}\lVert g\rVert_{L_G^2(\mathsf{T}_2)} \le C\varepsilon_{m-1}^{Q/2-qZ}\lVert g\rVert_{L_G^2(\mathsf{T}_2)}, ε m − s ∥ U ( t ) − Y ♯ ( t ) ∥ C ε m N rep ( T 2 ) ≤ C ε m − 1 − Z ε m Q /2 ∥ g ∥ L G 2 ( T 2 ) ≤ C ε m − 1 Q /2 − q Z ∥ g ∥ L G 2 ( T 2 ) ,
and Q / 2 − q Z ≥ 2 Q/2-qZ \ge2 Q /2 − q Z ≥ 2 by (180) .
Step 7: Conclusion. By Lemma 14.2 (a) applied to R ′ R' R ′ , ε m − 1 − s [ Y ( t ) ] C ε m − 1 N r e p ( T 2 ) ≤ size ε m − 1 ( R ′ ) \varepsilon_{m-1}^{-s}[Y(t)]_{C_{\varepsilon_{m-1}}^{N_{\mathrm{rep}}}(\mathsf{T}_2)} \le\operatorname{size}_{\varepsilon_{m-1}}(R') ε m − 1 − s [ Y ( t ) ] C ε m − 1 N rep ( T 2 ) ≤ size ε m − 1 ( R ′ ) .
Inserting this, (189) –(194) and Proposition 14.1 (i) into the bound of Step 2 gives
size ε m ( R ) ≤ ( 1 + C ( ε m − 1 5 η / 4 + ε m − 1 δ − 2 Θ + ε m − 1 μ ) ) size ε m − 1 ( R ′ ) + C ε m − 1 2 ∥ g ∥ L G 2 ( T 2 ) . \operatorname{size}_{\varepsilon_m}(R) \le\left(1+C\left(\varepsilon_{m-1}^{5\eta/4}+\varepsilon_{m-1}^{\delta-2\Theta}+\varepsilon_{m-1}^{\mu}\right)\right)\operatorname{size}_{\varepsilon_{m-1}}(R')+C\varepsilon_{m-1}^{2}\lVert g\rVert_{L_G^2(\mathsf{T}_2)}. size ε m ( R ) ≤ ( 1 + C ( ε m − 1 5 η /4 + ε m − 1 δ − 2Θ + ε m − 1 μ ) ) size ε m − 1 ( R ′ ) + C ε m − 1 2 ∥ g ∥ L G 2 ( T 2 ) .
By (5) , η = min { δ / 8 , μ / 4 } \eta=\min\{\delta/8,\mu/4\} η = min { δ /8 , μ /4 } , and by (181) , δ − 2 Θ > 5 6 δ \delta-2\Theta>\frac{5}{6}\delta δ − 2Θ > 6 5 δ , so each of 5 η / 4 5\eta/4 5 η /4 , δ − 2 Θ \delta-2\Theta δ − 2Θ and μ \mu μ is at least η \eta η . Since ε m − 1 ≤ 1 \varepsilon_{m-1}\le1 ε m − 1 ≤ 1 , this proves Proposition 5.4 .
Proof of Proposition 5.2
Proof of Proposition 5.2. Each cycle starts once the previous preparation is complete and the admissibility window has moved forward by 3 τ j 3\tau_j 3 τ j .
Step 1: Levels finer than m 0 m_0 m 0 . By the choice of m 0 m_0 m 0 (item (4) of Subsection 11.1), ρ ( j ) ≤ 1 \rho(j)\le1 ρ ( j ) ≤ 1 for every level j > m 0 j>m_0 j > m 0 . An admissible seed of level j > m 0 j>m_0 j > m 0 on a window [ t S , t f ] [t_S,t_f] [ t S , t f ] with t S + 1 2 t 0 < t f t_S+\frac{1}{2}t_0<t_f t S + 2 1 t 0 < t f is an admissible family of level j j j on [ t S , t f ] [t_S,t_f] [ t S , t f ] (Definition 11.4 , with no successor). For every j ≥ m 0 j\ge m_0 j ≥ m 0 , the row and column sums of the coupling matrices of an admissible family of level j j j , for the masses ω v \omega_v ω v and ω v / Ψ 2 \omega_v/\Psi^2 ω v / Ψ 2 , are at most 1 8 \frac{1}{8} 8 1 (Proposition 11.5 ). By the bounds (7) on τ j \tau_j τ j and h j h_j h j (Lemma 7.1 (b)), and since ε j ≤ Λ − j \varepsilon_j\le\Lambda^{-j} ε j ≤ Λ − j (Lemma 7.1 (a)), the series of the 3 τ j + 4 h j 3\tau_j+4h_j 3 τ j + 4 h j converges, and item (4) takes m 0 m_0 m 0 so large that, for every m > m 0 m>m_0 m > m 0 ,
∑ j = m 0 + 1 m ( 3 τ j + 4 h j ) < 1 2 t 0 . (196) \sum_{j=m_0+1}^{m}(3\tau_j+4h_j)<\frac{1}{2}t_0. \tag*{(196)} j = m 0 + 1 ∑ m ( 3 τ j + 4 h j ) < 2 1 t 0 . (196)
Step 2: The homogenization cascade. Let m m m , κ \kappa κ , G G G , t S t_S t S , t f t_f t f , g g g and U U U be as in Proposition 5.2 . Set G m : = G G_m:=G G m := G , U m : = U U_m:=U U m := U , t ( m ) : = t S t_{(m)}:=t_S t ( m ) := t S and t A ( m ) : = t S t_A^{(m)}:=t_S t A ( m ) := t S ; by Step 1, G m G_m G m is an admissible family of level m m m on [ t A ( m ) , t f ] [t_A^{(m)},t_f] [ t A ( m ) , t f ] . For j = m , m − 1 , … , m 0 + 1 j=m,m-1,\ldots,m_0+1 j = m , m − 1 , … , m 0 + 1 in turn, given an admissible family G j G_j G j of level j j j on [ t A ( j ) , t f ] [t_A^{(j)},t_f] [ t A ( j ) , t f ] and a solution U j U_j U j of G j G_j G j on [ t ( j ) , t f ] [t_{(j)},t_f] [ t ( j ) , t f ] , put
t c ( j ) : = max { t ( j ) , t A ( j ) + 3 τ j } . (197) t_c^{(j)}:=\max\{t_{(j)},t_A^{(j)}+3\tau_j\}. \tag*{(197)} t c ( j ) := max { t ( j ) , t A ( j ) + 3 τ j } . (197)
Let G j ′ G'_j G j ′ be the successor of G j G_j G j in the descent of the seed G G G given by Lemma 11.8 (iv), and let t p r ( j ) ∈ [ t c ( j ) , t c ( j ) + 4 h j ] t_{\mathrm{pr}}^{(j)}\in[t_c^{(j)},t_c^{(j)}+4h_j] t pr ( j ) ∈ [ t c ( j ) , t c ( j ) + 4 h j ] and Π j \Pi_j Π j be the time and the map constructed in § 5.6 for G j G_j G j and G j ′ G'_j G j ′ , with t A = t A ( j ) t_A=t_A^{(j)} t A = t A ( j ) and t c = t c ( j ) t_c=t_c^{(j)} t c = t c ( j ) . By the paragraph after Proposition 14.1 , Propositions 5.3, 14.1, 13.10 and 5.4 hold for this pair. The hypothesis t c ( j ) + 4 h j ≤ t f t_c^{(j)}+4h_j\le t_f t c ( j ) + 4 h j ≤ t f of Proposition 5.3 follows from (198) at level j j j , whose proof below uses only the times t p r ( i ) t_{\mathrm{pr}}^{(i)} t pr ( i ) of the levels i > j i>j i > j , constructed before. By Proposition 5.3(b), G j − 1 : = G j ′ G_{j-1}:=G'_j G j − 1 := G j ′ is an admissible family of level j − 1 j-1 j − 1 on [ t A ( j − 1 ) , t f ] [t_A^{(j-1)},t_f] [ t A ( j − 1 ) , t f ] , where t A ( j − 1 ) : = t A ( j ) + 3 τ j t_A^{(j-1)}:=t_A^{(j)}+3\tau_j t A ( j − 1 ) := t A ( j ) + 3 τ j . Let U j − 1 U_{j-1} U j − 1 be the solution of G j − 1 G_{j-1} G j − 1 from the retained trace of Π j U j ( t c ( j ) ) \Pi_jU_j(t_c^{(j)}) Π j U j ( t c ( j ) ) at time t ( j − 1 ) : = t p r ( j ) t_{(j-1)}:=t_{\mathrm{pr}}^{(j)} t ( j − 1 ) := t pr ( j ) . By the uniqueness in Proposition 6.2 and (67) , subtracted at the times t c ( j ) t_c^{(j)} t c ( j ) and t ≥ t c ( j ) t\ge t_c^{(j)} t ≥ t c ( j ) , the restriction of U j U_j U j to [ t c ( j ) , t f ] [t_c^{(j)},t_f] [ t c ( j ) , t f ] is the solution of G j G_j G j from U j ( t c ( j ) ) U_j(t_c^{(j)}) U j ( t c ( j ) ) at time t c ( j ) t_c^{(j)} t c ( j ) . Hence these four propositions apply at level j j j to the datum U j ( t c ( j ) ) U_j(t_c^{(j)}) U j ( t c ( j ) ) , with U = U j U=U_j U = U j and Y = U j − 1 Y=U_{j-1} Y = U j − 1 .
Initialization at the finest levels
The times satisfy, for m 0 < j ≤ m m_0<j\le m m 0 < j ≤ m ,
t c ( j ) + 4 h j ≤ t S + ∑ i = j m ( 3 τ i + 4 h i ) < t S + 1 2 t 0 < t f . (198) t_c^{(j)}+4h_j\le t_S+\sum_{i=j}^{m}(3\tau_i+4h_i)<t_S+\frac{1}{2}t_0<t_f. \tag*{(198)} t c ( j ) + 4 h j ≤ t S + i = j ∑ m ( 3 τ i + 4 h i ) < t S + 2 1 t 0 < t f . (198)
For j = m j=m j = m , t c ( m ) = t S + 3 τ m t_c^{(m)}=t_S+3\tau_m t c ( m ) = t S + 3 τ m . For j < m j<m j < m , t ( j ) = t p r ( j + 1 ) ≤ t c ( j + 1 ) + 4 h j + 1 t_{(j)}=t_{\mathrm{pr}}^{(j+1)}\le t_c^{(j+1)}+4h_{j+1} t ( j ) = t pr ( j + 1 ) ≤ t c ( j + 1 ) + 4 h j + 1 and t A ( j ) + 3 τ j = t S + 3 ∑ i = j m τ i t_A^{(j)}+3\tau_j=t_S+3\sum_{i=j}^{m}\tau_i t A ( j ) + 3 τ j = t S + 3 ∑ i = j m τ i , which gives the first inequality by induction on j j j , for j = m , m − 1 , … j=m,m-1,\ldots j = m , m − 1 , … in turn; the second is (196) .
In particular, for m 0 < j ≤ m m_0<j\le m m 0 < j ≤ m , t p r ( j ) ≤ t c ( j ) + 4 h j < t S + 1 2 t 0 t_{\mathrm{pr}}^{(j)}\le t_c^{(j)}+4h_j<t_S+\frac{1}{2}t_0 t pr ( j ) ≤ t c ( j ) + 4 h j < t S + 2 1 t 0 . Moreover, t ( m ) = t A ( m ) t_{(m)}=t_A^{(m)} t ( m ) = t A ( m ) and, for m 0 ≤ j < m m_0\le j<m m 0 ≤ j < m ,
t ( j ) = t p r ( j + 1 ) ≥ t c ( j + 1 ) ≥ t A ( j + 1 ) + 3 τ j + 1 = t A ( j ) , t_{(j)}=t_{\mathrm{pr}}^{(j+1)}\ge t_c^{(j+1)}\ge t_A^{(j+1)}+3\tau_{j+1}=t_A^{(j)}, t ( j ) = t pr ( j + 1 ) ≥ t c ( j + 1 ) ≥ t A ( j + 1 ) + 3 τ j + 1 = t A ( j ) ,
so U j U_j U j lives in the window [ t A ( j ) , t f ] [t_A^{(j)},t_f] [ t A ( j ) , t f ] of G j G_j G j , on which Step 1 bounds its coupling matrices.
Step 3: The marked amplitudes. We show that, for m 0 ≤ j ≤ m m_0\le j\le m m 0 ≤ j ≤ m ,
∥ U j ∥ L ∞ ( [ t ( j ) , t f ] ; L G j 2 ( T 2 ) ) ≤ ∏ i = j + 1 m ( 1 + C ( ε i − 1 q + ε i − 1 δ ) ) ∥ g ∥ L ω 2 ( T 2 ) ≤ C ∥ g ∥ L ω 2 ( T 2 ) . (199) \lVert U_j\rVert_{L^\infty([t_{(j)},t_f];L^2_{G_j}(\mathbb{T}^2))}\le\prod_{i=j+1}^{m}(1+C(\varepsilon_{i-1}^{q}+\varepsilon_{i-1}^{\delta}))\lVert g\rVert_{L^2_\omega(\mathbb{T}^2)}\le C\lVert g\rVert_{L^2_\omega(\mathbb{T}^2)}. \tag*{(199)} ∥ U j ∥ L ∞ ([ t ( j ) , t f ] ; L G j 2 ( T 2 )) ≤ i = j + 1 ∏ m ( 1 + C ( ε i − 1 q + ε i − 1 δ )) ∥ g ∥ L ω 2 ( T 2 ) ≤ C ∥ g ∥ L ω 2 ( T 2 ) . (199)
which is (31) iterated; the product is bounded as in Subsection 5.2. For j = m j=m j = m , this is Proposition 6.2 for G m G_m G m , since the markers of a seed are 1; the proposition applies by Step 1 and Definition 11.2(iv).
Let m 0 < j ≤ m m_0<j\le m m 0 < j ≤ m . Since U j U_j U j is continuous in time (Proposition 6.2), the norm of the datum U j ( t c ( j ) ) U_j(t_c^{(j)}) U j ( t c ( j ) ) in L G j 2 ( T 2 ) L^2_{G_j}(\mathbb{T}^2) L G j 2 ( T 2 ) is at most the left side of (199) at level j j j . Proposition 5.3(b) at level j j j , with Y = U j − 1 Y=U_{j-1} Y = U j − 1 , bounds ∥ U j − 1 ∥ L ∞ ( [ t ( j − 1 ) , t f ] ; L G j − 1 2 ( T 2 ) ) \lVert U_{j-1}\rVert_{L^\infty([t_{(j-1)},t_f];L^2_{G_{j-1}}(\mathbb{T}^2))} ∥ U j − 1 ∥ L ∞ ([ t ( j − 1 ) , t f ] ; L G j − 1 2 ( T 2 )) by 1 + C ( ε j − 1 q + ε j − 1 δ ) 1+C(\varepsilon_{j-1}^{q}+\varepsilon_{j-1}^{\delta}) 1 + C ( ε j − 1 q + ε j − 1 δ ) times this norm, which gives (199) at level j − 1 j-1 j − 1 .
Step 4: The top level. Let t ∈ [ t S + 1 2 t 0 , t f ] t\in[t_S+\frac{1}{2}t_0,t_f] t ∈ [ t S + 2 1 t 0 , t f ] ; by (198) , t ≥ t p r ( m 0 + 1 ) = t ( m 0 ) t\ge t_{\mathrm{pr}}^{(m_0+1)}=t_{(m_0)} t ≥ t pr ( m 0 + 1 ) = t ( m 0 ) . At level m 0 + 1 m_0+1 m 0 + 1 the solution Y Y Y is U m 0 U_{m_0} U m 0 , so Proposition 13.10(i) and (199) give
∥ U m 0 ( t ) ∥ H ρ ( m 0 + 1 ) , G m 0 N r e t ( T 2 ) ≤ C ∥ U m 0 + 1 ( t c ( m 0 + 1 ) ) ∥ L G m 0 + 1 2 ( T 2 ) ≤ C ∥ g ∥ L ω 2 ( T 2 ) . \lVert U_{m_0}(t)\rVert_{H^{N_{\mathrm{ret}}}_{\rho(m_0+1),G_{m_0}}(\mathbb{T}^2)}
\le C\lVert U_{m_0+1}(t_c^{(m_0+1)})\rVert_{L^2_{G_{m_0+1}}(\mathbb{T}^2)}
\le C\lVert g\rVert_{L^2_\omega(\mathbb{T}^2)}. ∥ U m 0 ( t ) ∥ H ρ ( m 0 + 1 ) , G m 0 N ret ( T 2 ) ≤ C ∥ U m 0 + 1 ( t c ( m 0 + 1 ) ) ∥ L G m 0 + 1 2 ( T 2 ) ≤ C ∥ g ∥ L ω 2 ( T 2 ) .
Since ρ ( m 0 + 1 ) ≤ 1 \rho(m_0+1)\le1 ρ ( m 0 + 1 ) ≤ 1 and ε m 0 1 + γ / 2 + δ / ρ ( m 0 + 1 ) ≤ C \varepsilon_{m_0}^{1+\gamma/2+\delta}/\rho(m_0+1)\le C ε m 0 1 + γ /2 + δ / ρ ( m 0 + 1 ) ≤ C by (184) , Lemma 14.3 with λ = ε m 0 \lambda=\varepsilon_{m_0} λ = ε m 0 , together with (182) and N r e t ≥ N r e p + 2 N_{\mathrm{ret}}\ge N_{\mathrm{rep}}+2 N ret ≥ N rep + 2 , shows that R m 0 ( t ) : = ( U m 0 ( t ) , ∅ ) R_{m_0}(t):=(U_{m_0}(t),\varnothing) R m 0 ( t ) := ( U m 0 ( t ) , ∅ ) is a representation of U m 0 ( t ) U_{m_0}(t) U m 0 ( t ) at wavelength ε m 0 \varepsilon_{m_0} ε m 0 with
size ε m 0 ( R m 0 ( t ) ) ≤ C ε m 0 − Z ∥ U m 0 ( t ) ∥ H ρ ( m 0 + 1 ) , ω N r e p + 2 ( T 2 ) ≤ C ε m 0 − Z ∥ g ∥ L ω 2 ( T 2 ) . (200) \operatorname{size}_{\varepsilon_{m_0}}(R_{m_0}(t))\le C\varepsilon_{m_0}^{-Z}\lVert U_{m_0}(t)\rVert_{H^{N_{\mathrm{rep}}+2}_{\rho(m_0+1),\omega}(\mathbb{T}^2)}
\le C\varepsilon_{m_0}^{-Z}\lVert g\rVert_{L^2_\omega(\mathbb{T}^2)}. \tag*{(200)} size ε m 0 ( R m 0 ( t )) ≤ C ε m 0 − Z ∥ U m 0 ( t ) ∥ H ρ ( m 0 + 1 ) , ω N rep + 2 ( T 2 ) ≤ C ε m 0 − Z ∥ g ∥ L ω 2 ( T 2 ) . (200)
Step 5: The regularity cascade. For j = m 0 + 1 , … , m j=m_0+1,\ldots,m j = m 0 + 1 , … , m in turn, t ∈ [ t p r ( j ) , t f ] t\in[t_{\mathrm{pr}}^{(j)},t_f] t ∈ [ t pr ( j ) , t f ] by (198) , and Proposition 5.4 at level j j j , applied to the representation R j − 1 ( t ) R_{j-1}(t) R j − 1 ( t ) of U j − 1 ( t ) U_{j-1}(t) U j − 1 ( t ) at wavelength ε j − 1 \varepsilon_{j-1} ε j − 1 , gives a representation R j ( t ) R_j(t) R j ( t ) of U j ( t ) U_j(t) U j ( t ) at wavelength ε j \varepsilon_j ε j . The datum of this passage is U j ( t c ( j ) ) U_j(t_c^{(j)}) U j ( t c ( j ) ) , whose norm is at most C ∥ g ∥ L ω 2 ( T 2 ) C\lVert g\rVert_{L^2_\omega(\mathbb{T}^2)} C ∥ g ∥ L ω 2 ( T 2 ) by (199) , so
size ε j ( R j ( t ) ) ≤ ( 1 + C ε j − 1 η ) size ε j − 1 ( R j − 1 ( t ) ) + C ε j − 1 ∥ g ∥ L ω 2 ( T 2 ) , \operatorname{size}_{\varepsilon_j}(R_j(t))\le(1+C\varepsilon_{j-1}^{\eta})\operatorname{size}_{\varepsilon_{j-1}}(R_{j-1}(t))+C\varepsilon_{j-1}\lVert g\rVert_{L^2_\omega(\mathbb{T}^2)}, size ε j ( R j ( t )) ≤ ( 1 + C ε j − 1 η ) size ε j − 1 ( R j − 1 ( t )) + C ε j − 1 ∥ g ∥ L ω 2 ( T 2 ) ,
which is (5.4) together with (199) . Iterating, with ∏ j ( 1 + C ε j − 1 η ) ≤ exp ( C ∑ j ε j − 1 η ) \prod_j(1+C\varepsilon_{j-1}^{\eta})\le\exp(C\sum_j\varepsilon_{j-1}^{\eta}) ∏ j ( 1 + C ε j − 1 η ) ≤ exp ( C ∑ j ε j − 1 η ) , the bounded sums of Subsection 5.2 and (200) ,
size ε m ( R m ( t ) ) ≤ exp ( C ∑ j = m 0 + 1 m ε j − 1 η ) ( size ε m 0 ( R m 0 ( t ) ) + C ∑ j = m 0 + 1 m ε j − 1 ∥ g ∥ L ω 2 ( T 2 ) ) ≤ C ( ε m 0 − Z + 1 ) ∥ g ∥ L ω 2 ( T 2 ) . \begin{aligned}
\operatorname{size}_{\varepsilon_m}(R_m(t))
&\le\exp\left(C\sum_{j=m_0+1}^{m}\varepsilon_{j-1}^{\eta}\right)\left(\operatorname{size}_{\varepsilon_{m_0}}(R_{m_0}(t))+C\sum_{j=m_0+1}^{m}\varepsilon_{j-1}\lVert g\rVert_{L^2_\omega(\mathbb{T}^2)}\right)\\
&\le C(\varepsilon_{m_0}^{-Z}+1)\lVert g\rVert_{L^2_\omega(\mathbb{T}^2)}.
\end{aligned} size ε m ( R m ( t )) ≤ exp ( C j = m 0 + 1 ∑ m ε j − 1 η ) ( size ε m 0 ( R m 0 ( t )) + C j = m 0 + 1 ∑ m ε j − 1 ∥ g ∥ L ω 2 ( T 2 ) ) ≤ C ( ε m 0 − Z + 1 ) ∥ g ∥ L ω 2 ( T 2 ) .
Since m 0 m_0 m 0 depends only on α , s , q , ζ , Λ , n , t 0 , T \alpha,s,q,\zeta,\Lambda,n,t_0,T α , s , q , ζ , Λ , n , t 0 , T , the representation R ( t ) : = R m ( t ) R(t):=R_m(t) R ( t ) := R m ( t ) proves Proposition 5.2, for every t ∈ [ t S + 1 2 t 0 , t f ] t\in[t_S+\frac{1}{2}t_0,t_f] t ∈ [ t S + 2 1 t 0 , t f ] .
Initialization at the finest levels
In this section we prove Proposition 5.1 , the entry, which initializes the induction at the finest levels: it produces the admissible seed of level M − 1 M-1 M − 1 , depending only on M M M , κ \kappa κ and t i n t_{\mathrm{in}} t in , from which the homogenization cascade (Proposition 5.2 ) starts. A weak solution θ \theta θ of (28) , whose velocity field b M + n \mathbf{b}_{M+n} b M + n has n n n levels finer than M M M , only provides the datum of the seed. Throughout, κ ∈ K M \kappa\in\mathcal{K}_M κ ∈ K M , so that κ M = κ \kappa_M=\kappa κ M = κ , t i n ∈ R t_{\mathrm{in}}\in\mathbb{R} t in ∈ R and t f : = t i n + T t_{\mathrm{f}}:=t_{\mathrm{in}}+T t f := t in + T ; C C C and c c c denote positive constants depending only on α , s , q , ζ , Λ , n , t 0 , T \alpha,s,q,\zeta,\Lambda,n,t_0,T α , s , q , ζ , Λ , n , t 0 , T , which may change from line to line; and M ≥ M ∗ ( α , s , q , ζ , Λ , n , t 0 , T ) M\geq M_*(\alpha,s,q,\zeta,\Lambda,n,t_0,T) M ≥ M ∗ ( α , s , q , ζ , Λ , n , t 0 , T ) , the threshold of Proposition 5.1 , fixed last, above the finitely many levels beyond which the inequalities below hold.
The route. The entry walks up from level M + n M+n M + n to level M − 1 M-1 M − 1 in three stages, and then back down.
Weak passages , from level i i i to level i − 1 i-1 i − 1 for i = M + n , … , M + 1 i=M+n,\ldots,M+1 i = M + n , … , M + 1 . At the levels finer than M M M , the Péclet number ε i β / κ \varepsilon_i^\beta/\kappa ε i β / κ is small (Lemma 15.1), so a passage keeps the diffusivity κ \kappa κ : the enhancement of level i i i has the relative size ( ε i β / κ ) 2 (\varepsilon_i^\beta/\kappa)^2 ( ε i β / κ ) 2 , and it is kept as a source of the coarser family instead of being added to the diffusivity. On the quiet margins of level i i i the velocity field is b i − 1 \mathbf{b}_{i-1} b i − 1 , and molecular diffusion over a margin reaches the wavelength ε i − 1 \varepsilon_{i-1} ε i − 1 ; the solution regains its derivatives there, and the next passage starts at the block boundary between two margins.
One smoothing , at level M M M : the sources are averaged over discs of radius r w r_{\mathrm{w}} r w , slightly below ε M \varepsilon_M ε M , so that the terminal cycle can use § 5.6 and § 5.8 .
The terminal cycle at level M M M , by the method of Proposition 5.3, at the bounded contrast κ M − 1 / κ ≤ 2 \kappa_{M-1}/\kappa\leq2 κ M − 1 / κ ≤ 2 . It is prepared over a time of order h M h_M h M and creates one generation of components. Its successor, restarted h M h_M h M later with its components declared of generation 0, is the seed.
Back down, the terminal lift and the n n n weak lifts carry a representation at ε M − 1 \varepsilon_{M-1} ε M − 1 to one at ε M + n \varepsilon_{M+n} ε M + n at a constant cost, and θ \theta θ differs from the lifted solution by little in C 0 , s ( T 2 ) C^{0,s}(\mathbb{T}^2) C 0 , s ( T 2 ) .
One ladder of families. We write G i G_i G i and U i U_i U i , M ≤ i ≤ M + n M\leq i\leq M+n M ≤ i ≤ M + n , for the families of the weak levels and their solutions, and t ( i ) t_{(i)} t ( i ) for the time at which U i U_i U i starts; U M + n = θ U_{M+n}=\theta U M + n = θ . This continues upward the notation G j G_j G j , U j U_j U j , t ( j ) t_{(j)} t ( j ) of the proof of Proposition 5.2 : the seed is G M − 1 = G G_{M-1}=G G M − 1 = G , and its solution is U M − 1 = U U_{M-1}=U U M − 1 = U , from t ( M − 1 ) = t e t_{(M-1)}=t_{\mathrm{e}} t ( M − 1 ) = t e . Between them stand G ~ M \widetilde{G}_M G M , the family G M G_M G M with smoothed sources, its solution U ~ M \widetilde{U}_M U M , and the terminal successor G ′ G' G ′ of G ~ M \widetilde{G}_M G M , of level M − 1 M-1 M − 1 , with its solution Y Y Y (Subsection 15.3). Figure 15.1 shows the times.
Figure 15.1. The times of the entry, not to scale. Shaded: the quiet margins, the two of level i i i around t ( i − 1 ) t_{(i-1)} t ( i − 1 ) , forming [ t ( i − 1 ) − τ i ′ , t ( i − 1 ) + τ i ′ ] [t_{(i-1)}-\tau'_i,t_{(i-1)}+\tau'_i] [ t ( i − 1 ) − τ i ′ , t ( i − 1 ) + τ i ′ ] , M < i ≤ M + n M<i\leq M+n M < i ≤ M + n , and the two of level M M M around t p r ( M ) t_{\mathrm{pr}}^{(M)} t pr ( M ) . Bars: the solutions of the ladder from their starts, with the preparation time t p r ( M ) t_{\mathrm{pr}}^{(M)} t pr ( M ) and t e = t p r ( M ) + h M t_{\mathrm{e}}=t_{\mathrm{pr}}^{(M)}+h_M t e = t pr ( M ) + h M ; a bar ends where the next solution takes over, or at t f t_{\mathrm{f}} t f . Dots: the levels between M + n − 1 M+n-1 M + n − 1 and M M M .
Parameters and scales
Parameters. The entry uses the readout weight ε i − 1 δ / 16 \varepsilon_{i-1}^{\delta/16} ε i − 1 δ /16 of a letter at level i i i , the mass factor ( 4 n N c o r , w ) − 1 ε i − 1 δ / 32 (4nN_{\mathrm{cor,w}})^{-1}\varepsilon_{i-1}^{\delta/32} ( 4 n N cor , w ) − 1 ε i − 1 δ /32 of a letter of a weak passage at level i i i , with N c o r , w N_{\mathrm{cor,w}} N cor , w of item (1), and the width
r w : = ε M 1 + γ / 4 + δ / 2 r_{\mathrm{w}}:=\varepsilon_{M}^{1+\gamma/4+\delta/2} r w := ε M 1 + γ /4 + δ /2
of the smoothing at level M M M . Since δ ≤ 1 4 ( q − 1 ) \delta\le\frac{1}{4}(q-1) δ ≤ 4 1 ( q − 1 ) and γ < 2 3 ( q − 1 ) \gamma< \frac{2}{3}(q-1) γ < 3 2 ( q − 1 ) by (5) , γ 4 + δ 2 < 7 24 ( q − 1 ) \frac{\gamma}{4}+\frac{\delta}{2}<\frac{7}{24}(q-1) 4 γ + 2 δ < 24 7 ( q − 1 ) . With L : = Λ q M / ( q − 1 ) L:=\Lambda q^{M}/(q-1) L := Λ q M / ( q − 1 ) , (6) gives ε M + 1 ≤ L − q \varepsilon_{M+1}\le L^{-q} ε M + 1 ≤ L − q and ε M ≥ ( 2 L ) − 1 \varepsilon_M\ge(2L)^{-1} ε M ≥ ( 2 L ) − 1 , and L q − 1 − γ / 4 − δ / 2 ≥ Λ 17 / 24 ≥ 4 L^{q-1-\gamma/4-\delta/2}\ge\Lambda^{17/24}\ge4 L q − 1 − γ /4 − δ /2 ≥ Λ 17/24 ≥ 4 as Λ ≥ 128 \Lambda\ge128 Λ ≥ 128 , so that ε M + 1 ≤ r w ≤ ε M \varepsilon_{M+1}\le r_{\mathrm{w}}\le\varepsilon_M ε M + 1 ≤ r w ≤ ε M . The entry then fixes the following orders and exponents, in this order, depending only on α , s , q , ζ , Λ , n , t 0 , T \alpha,s,q,\zeta,\Lambda,n,t_0,T α , s , q , ζ , Λ , n , t 0 , T (Table 11.1 ).
(1) The weak orders: the corrector order N c o r , w N_{\mathrm{cor,w}} N cor , w and word length N l e n , w N_{\mathrm{len,w}} N len , w of a weak passage, with
( q − 1 ) N c o r , w > 5 + 2 q n , 2 ( q + 2 ) γ ( N l e n , w + 1 ) > 4 + 3 q n . (201) (q-1)N_{\mathrm{cor,w}}>5+2q^n,\qquad2(q+2)\gamma(N_{\mathrm{len,w}}+1)>4+3q^n. \tag*{(201)} ( q − 1 ) N cor , w > 5 + 2 q n , 2 ( q + 2 ) γ ( N len , w + 1 ) > 4 + 3 q n . (201)
They absorb the cell tail and the frontier of a weak passage (Lemma 15.7 ). With them, θ \theta θ is recovered from a field of G M G_M G M at the order n N c o r , w + 5 nN_{\mathrm{cor,w}}+5 n N cor , w + 5 (Lemma 15.13 (b)).
(2) The terminal orders: the accuracy Q e : = 2 ( 5 + q n + 1 ( 1 + s ) ) Q_{\mathrm{e}}:=2(5+q^{n+1}(1+s)) Q e := 2 ( 5 + q n + 1 ( 1 + s )) of the terminal cycle; then its corrector order N c o r , e N_{\mathrm{cor,e}} N cor , e , truncation order p e p_{\mathrm{e}} p e , word length N l e n , e N_{\mathrm{len,e}} N len , e and number N c l , e N_{\mathrm{cl,e}} N cl , e of antidivergence steps, with
( q γ + 3 δ ) N c o r , e > Q e + 2 , δ p e > Q e + 2 , 1 2 δ ( N l e n , e + 1 ) > Q e + 2 , ( q γ + 3 δ ) N c l , e > 3 ; (202) \begin{aligned}
(q\gamma+3\delta)N_{\mathrm{cor,e}}&>Q_{\mathrm{e}}+2,\qquad\delta p_{\mathrm{e}}>Q_{\mathrm{e}}+2,\\
\frac{1}{2}\delta(N_{\mathrm{len,e}}+1)&>Q_{\mathrm{e}}+2,\qquad(q\gamma+3\delta)N_{\mathrm{cl,e}}>3; \tag*{(202)}
\end{aligned} ( q γ + 3 δ ) N cor , e 2 1 δ ( N len , e + 1 ) > Q e + 2 , δ p e > Q e + 2 , > Q e + 2 , ( q γ + 3 δ ) N cl , e > 3 ; (202)
then the largest birth order of the sources the terminal cycle creates and the width exponent of their smoothing,
N s r c , e : = N c o r , e + n N c o r , w + 4 p e + 1 , Θ e : = δ 16 ( N s r c , e − 1 ) . (203) N_{\mathrm{src,e}}:=N_{\mathrm{cor,e}}+nN_{\mathrm{cor,w}}+4p_{\mathrm{e}}+1,\qquad\Theta_{\mathrm{e}}:=\frac{\delta}{16(N_{\mathrm{src,e}}-1)}. \tag*{(203)} N src , e := N cor , e + n N cor , w + 4 p e + 1 , Θ e := 16 ( N src , e − 1 ) δ . (203)
The accuracy absorbs the loss ε M + n − 1 − s ≤ C ε M − 1 − q n + 1 ( 1 + s ) \varepsilon_{M+n}^{-1-s}\le C\varepsilon_{M-1}^{-q^{n+1}(1+s)} ε M + n − 1 − s ≤ C ε M − 1 − q n + 1 ( 1 + s ) of the recovery of θ \theta θ (Lemma 15.14 ), and (202) the residual of the terminal cycle (Table 15.1 ).
Table 15.1. The residual of the terminal lift Y ♯ Y^\sharp Y ♯ in the system of G ~ M \widetilde{G}_M G M on [ t p r ( M ) , t f ] [t_{\mathrm{pr}}^{(M)},t_f] [ t pr ( M ) , t f ] , measured as in (156) at the order N r d , e N_{\mathrm{rd},e} N rd , e and the scale ε M \varepsilon_M ε M . The cell tail has the degrees above N c o r , e N_{\mathrm{cor},e} N cor , e ; the frontier the words of length N l e n , e + 1 N_{\mathrm{len},e}+1 N len , e + 1 , each letter gaining ε M − 1 δ / 2 \varepsilon_{M-1}^{\delta/2} ε M − 1 δ /2 (proof of Proposition 15.11 ); the new smoothings are bounded by the priced majorant entry of that proof times the gain ε M − 1 Θ e N m o m , e \varepsilon_{M-1}^{\Theta_eN_{\mathrm{mom},e}} ε M − 1 Θ e N mom , e of the moments. The cleanup factor 1 + ( κ T ) 1 / 2 ε M − 1 ( ε M / ρ ( M ) ) N c l , e 1+(\kappa T)^{1/2}\varepsilon_M^{-1}(\varepsilon_M/\rho(M))^{N_{\mathrm{cl},e}} 1 + ( κ T ) 1/2 ε M − 1 ( ε M / ρ ( M ) ) N cl , e is at most C C C , as ( q γ + 3 δ ) N c l , e > 3 (q\gamma+3\delta)N_{\mathrm{cl},e}>3 ( q γ + 3 δ ) N cl , e > 3 , and so is the cost of a coarse flux under the lift (Lemma 13.6; fourth column), as κ M − 1 / κ ≤ 2 \kappa_{M-1}/\kappa\leq2 κ M − 1 / κ ≤ 2 . The truncation row includes the Taylor part, as in the proof of Proposition 13.11 .
(3) The representation order
N r e p : = n N c o r , w + N c o r , e + 2 p e + 4. (204) N_{\mathrm{rep}}:=nN_{\mathrm{cor,w}}+N_{\mathrm{cor,e}}+2p_{\mathrm{e}}+4. \tag*{(204)} N rep := n N cor , w + N cor , e + 2 p e + 4. (204)
This is Part I’s N r e p N_{\mathrm{rep}} N rep : the terminal lift spends N c o r , e + 2 p e + 1 N_{\mathrm{cor,e}}+2p_{\mathrm{e}}+1 N cor , e + 2 p e + 1 derivatives and the weak lifts n N c o r , w nN_{\mathrm{cor,w}} n N cor , w , which leaves three. The orders of items (1) and (2) fix the seed’s bounds: birth orders at most N s r c , e N_{\mathrm{src,e}} N src , e , at most N p a t h , e : = ( N l e n , e + 2 ) ( n N l e n , w + 1 ) N_{\mathrm{path,e}}:=(N_{\mathrm{len,e}}+2)(nN_{\mathrm{len,w}}+1) N path , e := ( N len , e + 2 ) ( n N len , w + 1 ) components along a path of sources, that is, at most B : = N p a t h , e − 1 B:=N_{\mathrm{path,e}}-1 B := N path , e − 1 sources on a path, the bound B B B of item (1) of Subsection 11.1. Section 11 then fixes the remaining parameters of its item (1).
(4) After these: the moment orders N m o m , w N_{\mathrm{mom,w}} N mom , w of the smoothing at level M M M and N m o m , e N_{\mathrm{mom,e}} N mom , e of the smoothing of the sources the terminal cycle creates, with
δ − δ 16 − δ 64 − Θ e ( N s r c , e − 2 ) + Θ e N m o m , e > Q e + 2 , q ( 2 ( q + 2 ) γ − 5 δ 64 + ( q 4 + δ 2 ) N m o m , w ) > 2 Q e + 4 ; (205) \begin{aligned}
\delta-\frac{\delta}{16}-\frac{\delta}{64}-\Theta_{\mathrm{e}}(N_{\mathrm{src,e}}-2)+\Theta_{\mathrm{e}}N_{\mathrm{mom,e}}&>Q_{\mathrm{e}}+2,\\
q\left(2(q+2)\gamma-\frac{5\delta}{64}+\left(\frac{q}{4}+\frac{\delta}{2}\right)N_{\mathrm{mom,w}}\right)&>2Q_{\mathrm{e}}+4; \tag*{(205)}
\end{aligned} δ − 16 δ − 64 δ − Θ e ( N src , e − 2 ) + Θ e N mom , e q ( 2 ( q + 2 ) γ − 64 5 δ + ( 4 q + 2 δ ) N mom , w ) > Q e + 2 , > 2 Q e + 4 ; (205)
they absorb the smoothings in Table 15.1 . Since 2 ( q + 2 ) γ < 1 2(q+2)\gamma<1 2 ( q + 2 ) γ < 1 and q < 17 16 q<\frac{17}{16} q < 16 17 by (5) and (3) , the second condition gives ( q 4 + δ 2 ) N m o m , w > ( 2 Q e + 4 ) / q − 1 > 2 + 2 q n \left(\frac{q}{4}+\frac{\delta}{2}\right)N_{\mathrm{mom,w}}>(2Q_{\mathrm{e}}+4)/q-1>2+2q^n ( 4 q + 2 δ ) N mom , w > ( 2 Q e + 4 ) / q − 1 > 2 + 2 q n , which Lemma 15.14 uses. Then the recovery order N r d , e ≥ n N c o r , w + 5 N_{\mathrm{rd,e}}\ge nN_{\mathrm{cor,w}}+5 N rd , e ≥ n N cor , w + 5 of the terminal comparison (Proposition 15.10 (c)), which, with N r e c , e : = 4 ( N c o r , e + N s r c , e ) + 2 N c l , e + 16 N_{\mathrm{rec,e}} := 4(N_{\mathrm{cor,e}} + N_{\mathrm{src,e}}) + 2N_{\mathrm{cl,e}} + 16 N rec , e := 4 ( N cor , e + N src , e ) + 2 N cl , e + 16 and N ˉ m o m : = max { N m o m , w , N m o m , e , N m o m } \bar{N}_{\mathrm{mom}} := \max\{N_{\mathrm{mom,w}}, N_{\mathrm{mom,e}}, N_{\mathrm{mom}}\} N ˉ mom := max { N mom , w , N mom , e , N mom } , is also at least the floor (158) of the comparison at m = M m = M m = M , with the terminal orders,
N r d , e ≥ max { N r e c , e + N s r c , e + N ˉ m o m + 6 p e , N s r c , e − 2 + N ˉ m o m + N c o r , e + 2 p e + 1 } , N r d , e ≥ N r e c , e + 2 N s r c , e N p a t h , e + max { N s r c , e + N ˉ m o m + N c o r , e + 6 p e + 4 , ( N p a t h , e + 3 ) ( N c o r , e + N s r c , e + 12 ) + 2 p e } ; (206) \begin{aligned}
N_{\mathrm{rd,e}} &\ge\max\{N_{\mathrm{rec,e}} + N_{\mathrm{src,e}} + \bar{N}_{\mathrm{mom}} + 6p_{\mathrm{e}}, N_{\mathrm{src,e}} - 2 + \bar{N}_{\mathrm{mom}} + N_{\mathrm{cor,e}} + 2p_{\mathrm{e}} + 1\},\\
N_{\mathrm{rd,e}} &\ge N_{\mathrm{rec,e}} + 2N_{\mathrm{src,e}}N_{\mathrm{path,e}}\\
&\quad+ \max\{N_{\mathrm{src,e}} + \bar{N}_{\mathrm{mom}} + N_{\mathrm{cor,e}} + 6p_{\mathrm{e}} + 4, (N_{\mathrm{path,e}} + 3)(N_{\mathrm{cor,e}} + N_{\mathrm{src,e}} + 12) + 2p_{\mathrm{e}}\}; \tag*{(206)}
\end{aligned} N rd , e N rd , e ≥ max { N rec , e + N src , e + N ˉ mom + 6 p e , N src , e − 2 + N ˉ mom + N cor , e + 2 p e + 1 } , ≥ N rec , e + 2 N src , e N path , e + max { N src , e + N ˉ mom + N cor , e + 6 p e + 4 , ( N path , e + 3 ) ( N cor , e + N src , e + 12 ) + 2 p e } ; (206)
then the output order N o u t , e N_{\mathrm{out,e}} N out , e of the terminal preparation and the trace order N t r , w N_{\mathrm{tr,w}} N tr , w of a weak passage,
N o u t , e : = max { N r e p , 2 N r d , e + 2 , N c o r , e + 2 p e + N r d , e + 2 } , N t r , w ≥ max { N m o m , w + n N c o r , w + 1 + N r d , e , 2 N r d , e + 2 } . (207) \begin{aligned}
N_{\mathrm{out,e}} &:= \max\{N_{\mathrm{rep}}, 2N_{\mathrm{rd,e}} + 2, N_{\mathrm{cor,e}} + 2p_{\mathrm{e}} + N_{\mathrm{rd,e}} + 2\},\\
N_{\mathrm{tr,w}} &\ge\max\{N_{\mathrm{mom,w}} + nN_{\mathrm{cor,w}} + 1 + N_{\mathrm{rd,e}}, 2N_{\mathrm{rd,e}} + 2\}. \tag*{(207)}
\end{aligned} N out , e N tr , w := max { N rep , 2 N rd , e + 2 , N cor , e + 2 p e + N rd , e + 2 } , ≥ max { N mom , w + n N cor , w + 1 + N rd , e , 2 N rd , e + 2 } . (207)
As N o u t , e ≥ 2 N r d , e + 2 N_{\mathrm{out,e}} \ge2N_{\mathrm{rd,e}} + 2 N out , e ≥ 2 N rd , e + 2 , the proof of Proposition 15.10 runs at the order N o u t , e N_{\mathrm{out,e}} N out , e , above this floor. The condition on N t r , w N_{\mathrm{tr,w}} N tr , w in (207) leaves room for the smoothing and the sources of G M G_M G M in Lemma 15.8, above N r d , e N_{\mathrm{rd,e}} N rd , e . The order N t r , w N_{\mathrm{tr,w}} N tr , w is also at least the order N b u f N_{\mathrm{buf}} N buf of (123) that the preparation reads at the terminal level to produce Π \Pi Π at N o u t , e N_{\mathrm{out,e}} N out , e , above the order 2 N r d , e 2N_{\mathrm{rd,e}} 2 N rd , e at which the proof of Proposition 15.10 (a) interpolates. Then the seed’s constants (§ 5.4 , item (2)).
(5) The threshold M ∗ M_\ast M ∗ , last.
The levels finer than M, at small Péclet number
Lemma 15.1 separates the weak levels from the terminal one. At the levels finer than M M M , by (a), the Péclet number is small, the strain of b i − 1 b_{i-1} b i − 1 is small over the time diffusion needs to reach ϵ i − 1 \epsilon_{i-1} ϵ i − 1 , the quiet margins are long enough for item 1 of the route (as γ > 3 δ \gamma> 3\delta γ > 3 δ ), and a pulse lasts many cell diffusion times. At level M M M , by (b), the margins are too short, and the contrast is at most 2.
Lemma 15.1 (Scales of the entry). There exist Λ 0 ( α , s , q , ζ ) ≥ 128 \Lambda_0(\alpha, s, q, \zeta) \ge128 Λ 0 ( α , s , q , ζ ) ≥ 128 and C ( α , s , q , ζ ) C(\alpha, s, q, \zeta) C ( α , s , q , ζ ) with the following property. For every Λ ≥ Λ 0 \Lambda\ge\Lambda_0 Λ ≥ Λ 0 there exists an integer M 2 ( α , s , q , ζ , Λ ) ≥ 2 M_2(\alpha, s, q, \zeta, \Lambda) \ge2 M 2 ( α , s , q , ζ , Λ ) ≥ 2 such that, for every M ≥ 2 M \ge2 M ≥ 2 and κ ∈ K M \kappa\in K_M κ ∈ K M , the following hold.
(a) For every M < i ≤ M + n M < i \le M + n M < i ≤ M + n ,
ϵ i β κ ≤ C ϵ i − 1 ( q + 2 ) γ , ϵ i − 1 β κ ≤ C ϵ i − 1 γ , ϵ i − 1 2 κ τ i ′ ≤ C ϵ i − 1 γ − 3 δ , κ τ i ϵ i 2 ≥ 1 C ϵ i − 1 − ( 2 ( q − 1 ) + γ − 4 δ ) . \frac{\epsilon_i^\beta}{\kappa} \le C\epsilon_{i-1}^{(q+2)\gamma}, \qquad
\frac{\epsilon_{i-1}^\beta}{\kappa} \le C\epsilon_{i-1}^\gamma, \qquad
\frac{\epsilon_{i-1}^2}{\kappa\tau_i'} \le C\epsilon_{i-1}^{\gamma-3\delta}, \qquad
\frac{\kappa\tau_i}{\epsilon_i^2} \ge\frac{1}{C}\epsilon_{i-1}^{-(2(q-1)+\gamma-4\delta)}. κ ϵ i β ≤ C ϵ i − 1 ( q + 2 ) γ , κ ϵ i − 1 β ≤ C ϵ i − 1 γ , κ τ i ′ ϵ i − 1 2 ≤ C ϵ i − 1 γ − 3 δ , ϵ i 2 κ τ i ≥ C 1 ϵ i − 1 − ( 2 ( q − 1 ) + γ − 4 δ ) .
(b) At level M M M , κ τ M ′ ≲ ϵ M − 1 2 + γ + 3 δ \kappa\tau_M' \lesssim\epsilon_{M-1}^{2+\gamma+3\delta} κ τ M ′ ≲ ϵ M − 1 2 + γ + 3 δ and
ϵ M β κ ≤ C ϵ M γ , ϵ ˉ M ρ ( M ) ≳ ϵ M − 1 q γ + 3 δ , κ h M ϵ M 2 ≲ ϵ M − 1 − 2 ( q γ + 3 δ ) , κ τ M ϵ M 2 ≲ ϵ M − 1 − ( 2 q γ + 4 δ ) . \frac{\epsilon_M^\beta}{\kappa} \le C\epsilon_M^\gamma, \qquad
\frac{\bar{\epsilon}_M}{\rho(M)} \gtrsim\epsilon_{M-1}^{q\gamma+3\delta}, \qquad
\frac{\kappa h_M}{\epsilon_M^2} \lesssim\epsilon_{M-1}^{-2(q\gamma+3\delta)}, \qquad
\frac{\kappa\tau_M}{\epsilon_M^2} \lesssim\epsilon_{M-1}^{-(2q\gamma+4\delta)}. κ ϵ M β ≤ C ϵ M γ , ρ ( M ) ϵ ˉ M ≳ ϵ M − 1 q γ + 3 δ , ϵ M 2 κ h M ≲ ϵ M − 1 − 2 ( q γ + 3 δ ) , ϵ M 2 κ τ M ≲ ϵ M − 1 − ( 2 q γ + 4 δ ) .
Moreover κ M − 1 / κ ≥ 1 \kappa_{M-1}/\kappa\ge1 κ M − 1 / κ ≥ 1 , and, if M ≥ M 2 M \ge M_2 M ≥ M 2 , then κ M − 1 / κ ≤ 2 \kappa_{M-1}/\kappa\le2 κ M − 1 / κ ≤ 2 .
Proof. All the bounds are arithmetic. By (20) and Lemma 7.1 (a), for i − 1 ≥ M i - 1 \ge M i − 1 ≥ M ,
κ ≥ 1 2 ϵ i − 1 β − γ , ϵ i ≤ 2 ϵ i − 1 , κ ≍ ϵ M β − γ ≍ ϵ M − 1 β + γ . \kappa\ge\frac{1}{2}\epsilon_{i-1}^{\beta-\gamma}, \qquad
\epsilon_i \le2\epsilon_{i-1}, \qquad
\kappa\asymp\epsilon_M^{\beta-\gamma} \asymp\epsilon_{M-1}^{\beta+\gamma}. κ ≥ 2 1 ϵ i − 1 β − γ , ϵ i ≤ 2 ϵ i − 1 , κ ≍ ϵ M β − γ ≍ ϵ M − 1 β + γ .
With (7) , they give the exponents of (a) and (b) by two identities: q ( β − γ ) = β + γ q(\beta-\gamma) = \beta+\gamma q ( β − γ ) = β + γ , which turns q β − β + γ q\beta-\beta+\gamma q β − β + γ into ( q + 2 ) γ (q+2)\gamma ( q + 2 ) γ , and q − 1 − γ 2 − δ = q γ + 3 δ q-1-\frac{\gamma}{2}-\delta= q\gamma+3\delta q − 1 − 2 γ − δ = q γ + 3 δ (Lemma 7.1 (c), as β = 1 + α \beta= 1+\alpha β = 1 + α ), which gives ϵ ˉ M / ρ ( M ) = ϵ M ( κ h M ) − 1 / 2 ≍ ϵ M − 1 q − 1 − γ / 2 − δ \bar{\epsilon}_M/\rho(M) = \epsilon_M(\kappa h_M)^{-1/2} \asymp\epsilon_{M-1}^{q-1-\gamma/2-\delta} ϵ ˉ M / ρ ( M ) = ϵ M ( κ h M ) − 1/2 ≍ ϵ M − 1 q − 1 − γ /2 − δ and κ τ M / ϵ M 2 ≍ ϵ M − 1 2 δ − 2 ( q − 1 − γ / 2 − δ ) \kappa\tau_M/\epsilon_M^2 \asymp\epsilon_{M-1}^{2\delta-2(q-1-\gamma/2-\delta)} κ τ M / ϵ M 2 ≍ ϵ M − 1 2 δ − 2 ( q − 1 − γ /2 − δ ) . By Proposition 2.1 (c), and by Proposition 2.1 (a) and Lemma 7.1 (a),
1 ≤ κ M − 1 κ ≤ 1 + C ϵ M 2 β κ 2 ≤ 1 + C ϵ M 2 γ ≤ 1 + C Λ − 2 γ M , 1 \le\frac{\kappa_{M-1}}{\kappa} \le1 + C\frac{\epsilon_M^{2\beta}}{\kappa^2} \le1 + C\epsilon_M^{2\gamma} \le1 + C\Lambda^{-2\gamma M}, 1 ≤ κ κ M − 1 ≤ 1 + C κ 2 ϵ M 2 β ≤ 1 + C ϵ M 2 γ ≤ 1 + C Λ − 2 γ M ,
which is at most 2 for M ≥ M 2 M \ge M_2 M ≥ M 2 , the least integer M 2 ≥ 2 M_2 \ge2 M 2 ≥ 2 with C Λ − 2 γ M 2 ≤ 1 C\Lambda^{-2\gamma M_2} \le1 C Λ − 2 γ M 2 ≤ 1 .
The levels finer than M M M , at small Péclet number
The two-dimensional cell inverse. Let M < i ≤ M + n M < i \le M + n M < i ≤ M + n and let k k k be a pulse of level i i i . In the chart of k k k (§ 4.1 ), with z = y / ε i z = y/\varepsilon_i z = y / ε i , the shear added at level i i i is ∇ ⊥ ψ i ( t , y ) \nabla^\perp\psi_i(t,y) ∇ ⊥ ψ i ( t , y ) of ((56) ), of size 2 π ε i β − 1 2\pi\varepsilon_i^{\beta-1} 2 π ε i β − 1 ; in the cell time κ t / ε i 2 \kappa t/\varepsilon_i^2 κ t / ε i 2 it is the Péclet number ε i β / κ \varepsilon_i^\beta/\kappa ε i β / κ times a bounded field. A weak passage solves its cell problems with the two-dimensional cell inverse below, as Definition 11.2 (iii) allows: once, from zero, on the window of each pulse. As the divergence-free shear drops out of the energy, solutions without forcing decay at the cell rate of level i i i , which absorbs the switching in Lemma 15.7 .
Lemma 15.2 (The two-dimensional cell inverse). Let N ≥ 1 N \ge1 N ≥ 1 be one of the orders used below; they depend only on α , s , q , ζ , Λ , n , t 0 , T \alpha,s,q,\zeta,\Lambda,n,t_0,T α , s , q , ζ , Λ , n , t 0 , T , and M ∗ M_* M ∗ is fixed after them. There exists C ( α , s , q , ζ , Λ , n , t 0 , T , N ) C(\alpha,s,q,\zeta,\Lambda,n,t_0,T,N) C ( α , s , q , ζ , Λ , n , t 0 , T , N ) such that the following hold for every M ≥ M ∗ M \ge M_* M ≥ M ∗ , M < i ≤ M + n M < i \le M+n M < i ≤ M + n , every pulse k k k of level i i i , every conductivity K ( t , y ) \mathbf{K}(t,y) K ( t , y ) on the window of k k k as in § 5.2 , and every pair F 0 , F F^0,F F 0 , F of two-scale forms on the window, of zero average in z z z , measurable in t t t , of class C N + 3 C^{N+3} C N + 3 in z z z with bounded derivatives. A solution on an interval I I I of the equation below is a two-scale form A \mathcal{A} A with A ( ⋅ , y , ⋅ ) ∈ C ( I ; L 2 ( [ 0 , 1 ] 2 ) ) \mathcal{A}(\cdot,y,\cdot) \in C(I;L^2([0,1]^2)) A ( ⋅ , y , ⋅ ) ∈ C ( I ; L 2 ([ 0 , 1 ] 2 )) for every y ∈ R 2 y \in\mathbb{R}^2 y ∈ R 2 that solves it in the integrated form in t t t , tested against smooth periodic functions of z z z , as in Lemma 10.4 .
(a) There exists exactly one solution A \mathcal{A} A on the window, of zero average in z z z and with ∇ z A ( ⋅ , y , ⋅ ) \nabla_z\mathcal{A}(\cdot,y,\cdot) ∇ z A ( ⋅ , y , ⋅ ) square integrable over the window times [ 0 , 1 ] 2 [0,1]^2 [ 0 , 1 ] 2 for every y ∈ R 2 y \in\mathbb{R}^2 y ∈ R 2 , of
∂ t A + ε i − 1 ( ∇ ⊥ ψ i ) ( t , ε i z ) ⋅ ∇ z A − κ ε i − 2 ∇ z ⋅ ( K ∇ z A ) = κ ε i − 2 ( F 0 + ∇ z ⋅ F ) in the window of k , A = 0 at its start, t = ( k − 61 48 ) τ i ; \begin{aligned}
\partial_t\mathcal{A}+\varepsilon_i^{-1}(\nabla^\perp\psi_i)(t,\varepsilon_i z)\cdot\nabla_z\mathcal{A}-\kappa\varepsilon_i^{-2}\nabla_z\cdot(\mathbf{K}\nabla_z\mathcal{A})&=\kappa\varepsilon_i^{-2}(F^0+\nabla_z\cdot F) &&\text{in the window of }k,\\
\mathcal{A}&=0 &&\text{at its start, }t=\left(k-\frac{61}{48}\right)\tau_i;
\end{aligned} ∂ t A + ε i − 1 ( ∇ ⊥ ψ i ) ( t , ε i z ) ⋅ ∇ z A − κ ε i − 2 ∇ z ⋅ ( K ∇ z A ) A = κ ε i − 2 ( F 0 + ∇ z ⋅ F ) = 0 in the window of k , at its start, t = ( k − 48 61 ) τ i ; it is of class C N C^N C N in z z z .
(b) Let F 0 = F = 0 F^0=F=0 F 0 = F = 0 , and let A \mathcal{A} A be a solution on an interval [ t 1 , t 2 ] [t_1,t_2] [ t 1 , t 2 ] within the window, of class C N C^N C N in z z z , with A ( t 1 ) \mathcal{A}(t_1) A ( t 1 ) of zero average and of class C N + 2 C^{N+2} C N + 2 in z z z , and with ∇ z A ( ⋅ , y , ⋅ ) \nabla_z\mathcal{A}(\cdot,y,\cdot) ∇ z A ( ⋅ , y , ⋅ ) square integrable over [ t 1 , t 2 ] × [ 0 , 1 ] 2 [t_1,t_2]\times[0,1]^2 [ t 1 , t 2 ] × [ 0 , 1 ] 2 for every y ∈ R 2 y \in\mathbb{R}^2 y ∈ R 2 . Then, for every t ∈ [ t 1 , t 2 ] t \in[t_1,t_2] t ∈ [ t 1 , t 2 ] and y ∈ R 2 y \in\mathbb{R}^2 y ∈ R 2 ,
∥ A ( t , y , ⋅ ) ∥ L 2 ( [ 0 , 1 ] 2 ) ≤ C exp ( − π 2 κ ( t − t 1 ) ε i 2 ) ∥ A ( t 1 , y , ⋅ ) ∥ L 2 ( [ 0 , 1 ] 2 ) . \lVert\mathcal{A}(t,y,\cdot)\rVert_{L^2([0,1]^2)} \le C\exp\left(-\frac{\pi^2\kappa(t-t_1)}{\varepsilon_i^2}\right)\lVert\mathcal{A}(t_1,y,\cdot)\rVert_{L^2([0,1]^2)}. ∥ A ( t , y , ⋅ ) ∥ L 2 ([ 0 , 1 ] 2 ) ≤ C exp ( − ε i 2 π 2 κ ( t − t 1 ) ) ∥ A ( t 1 , y , ⋅ ) ∥ L 2 ([ 0 , 1 ] 2 ) .
Proof. The solution of (a) is constructed by Galerkin approximation. The shear is divergence free, so it drops out of the energy identity; with K ≥ 3 4 I \mathbf{K} \ge\frac{3}{4}I K ≥ 4 3 I and ∥ ∇ z A ∥ L 2 ( [ 0 , 1 ] 2 ) 2 ≥ 4 π 2 ∥ A ∥ L 2 ( [ 0 , 1 ] 2 ) 2 \lVert\nabla_z\mathcal{A}\rVert_{L^2([0,1]^2)}^2 \ge4\pi^2\lVert\mathcal{A}\rVert_{L^2([0,1]^2)}^2 ∥ ∇ z A ∥ L 2 ([ 0 , 1 ] 2 ) 2 ≥ 4 π 2 ∥ A ∥ L 2 ([ 0 , 1 ] 2 ) 2 for forms of zero average, the identity gives (b), with C = 1 C=1 C = 1 , and, for the difference of two solutions, the uniqueness in (a). The derivatives of the shear in z z z are at most C ε i β / κ ≤ C C\varepsilon_i^\beta/\kappa\le C C ε i β / κ ≤ C , by Lemma 15.1(a); so, differentiating in z z z , the energy identity and the Sobolev embedding give the regularity in (a). □ \square □
The weak corrector operator. The weak finite lift of level i i i is the finite lift of Definition 10.5 for the pulses of level i i i , with κ \kappa κ for both diffusivities, the corrector order N c o r , w N_{\mathrm{cor,w}} N cor , w , the slow length ε i − 1 \varepsilon_{i-1} ε i − 1 and the two-dimensional cell inverse: the inverse of a right side H 0 + ∇ z ⋅ H H^0+\nabla_z\cdot H H 0 + ∇ z ⋅ H is the solution of Lemma 15.2 (a) with F 0 = ε i 2 κ − 1 H 0 F^0=\varepsilon_i^2\kappa^{-1}H^0 F 0 = ε i 2 κ − 1 H 0 and F = ε i 2 κ − 1 H F=\varepsilon_i^2\kappa^{-1}H F = ε i 2 κ − 1 H . The definition, existence and uniqueness of the finite lift use only the cell inverse (§ 5.2 ): the correctors exist at every order, each solved once on the window of its pulse from zero at its start, and the lift is determined by its values on the window. The weak corrector operator Corr i w ( t ) \operatorname{Corr}^{\mathrm{w}}_i(t) Corr i w ( t ) is the corrector operator of Definition 10.8 built from the weak finite lift. For a field U U U of ( t , x ) (t,x) ( t , x ) , Corr i w U \operatorname{Corr}^{\mathrm{w}}_i U Corr i w U is the field t ↦ Corr i w ( t ) U ( t ) t\mapsto\operatorname{Corr}^{\mathrm{w}}_i(t)U(t) t ↦ Corr i w ( t ) U ( t ) .
In Definition 15.3 , the means of the weak finite lift along the paths of sources become the inherited sources (ii), and those along the empty path, less the molecular flux, become the letters (iii).
Definition 15.3 (The weak families). The family G M + n G_{M+n} G M + n has one component, of mass 1 2 \frac{1}{2} 2 1 , and no source. For M < i ≤ M + n M < i \le M+n M < i ≤ M + n , the weak family G i − 1 G_{i-1} G i − 1 , of level i − 1 i-1 i − 1 , with the diffusivity κ \kappa κ , the velocity field b i − 1 \mathbf{b}_{i-1} b i − 1 and no smoothing, is defined from G i G_i G i as follows.
(i) Components: the components v v v of G i G_i G i , with their masses, and their children v ′ v' v ′ , labelled by the nonempty words of length # v ′ ≤ N l e n , w \#v' \le N_{\mathrm{len,w}} # v ′ ≤ N len , w in the letters 2 , … , N c o r , w + 1 2,\ldots,N_{\mathrm{cor,w}}+1 2 , … , N cor , w + 1 , with the masses ω v ′ : = ( 4 n N c o r , w ) − δ / 32 ε i − 1 δ / 32 ω v \omega_{v'} := \left(4nN_{\mathrm{cor,w}}\right)^{-\delta/32}\varepsilon_{i-1}^{\delta/32}\omega_v ω v ′ := ( 4 n N cor , w ) − δ /32 ε i − 1 δ /32 ω v . All markers are 1. The components over v v v are v v v , with the empty word, and its children.
(ii) Inherited sources: for every word of length at most N l e n , w N_{\mathrm{len,w}} N len , w , every nonempty path of sources of G i G_i G i from v 1 v_1 v 1 to v 2 v_2 v 2 and every 2 ≤ d ≤ N c o r , w + 1 + ( M + n − i ) N c o r , w 2 \le d \le N_{\mathrm{cor,w}}+1+(M+n-i)N_{\mathrm{cor,w}} 2 ≤ d ≤ N cor , w + 1 + ( M + n − i ) N cor , w , a source of birth order d d d from the component over v 1 v_1 v 1 with this word to the component over v 2 v_2 v 2 with this word. Its coefficients are the means of degree d d d of the weak finite lift along the path (Definition 10.5 ), transported to x x x ((11.8)); its factor is the product of the factors along the path.
(iii) Letters: for every component with a word of length less than N l e n , w N_{\mathrm{len,w}} N len , w and every letter 2 ≤ d ≤ N c o r , w + 1 2 \le d \le N_{\mathrm{cor,w}}+1 2 ≤ d ≤ N cor , w + 1 , a source of birth order d d d to the component over the same v v v whose word is extended by d d d . Its coefficients are ε i − 1 − δ / 16 \varepsilon_{i-1}^{-\delta/16} ε i − 1 − δ /16 times the mean of degree d d d of the weak finite lift along the empty path, less 1 { d = 2 } κ K \mathbf{1}_{\{d=2\}}\kappa K 1 { d = 2 } κ K (the eddy diffusivity at degree d d d ), transported to x x x ; its factor is ε i − 1 − δ / 16 ( ε i δ / κ ) 2 \varepsilon_{i-1}^{-\delta/16}(\varepsilon_i^\delta/\kappa)^2 ε i − 1 − δ /16 ( ε i δ / κ ) 2 .
Every source of G i − 1 G_{i-1} G i − 1 has the majorant 1 and the birth radius ε i \varepsilon_i ε i . For M ≤ i ≤ M + n M \le i \le M+n M ≤ i ≤ M + n , every component of G i G_i G i has the reserve res i : = ( N g e n + 3 + i − M ) / ( N g e n + 2 + N o f f ) \operatorname{res}_i := (N_{\mathrm{gen}}+3+i-M)/(N_{\mathrm{gen}}+2+N_{\mathrm{off}}) res i := ( N gen + 3 + i − M ) / ( N gen + 2 + N off ) , at most 1 for i < M + n i < M+n i < M + n as N o f f ≥ n N_{\mathrm{off}} \ge n N off ≥ n (item (2) of Subsection 11.1), which replaces res ( v ) \operatorname{res}(v) res ( v ) in (11.9) for this family; the smoothed family of Lemma 15.4(b) has the components of G M G_M G M , at the reserve res M \operatorname{res}_M res M .
The power ε i − 1 − δ / 16 \varepsilon_{i-1}^{-\delta/16} ε i − 1 − δ /16 in the coefficients of a letter is the inverse of the readout weight it adds, as for a successor ((38) ). The readout of the weak passage is accordingly ( Υ i w F ) v : = F v + ∑ v ′ ε i − 1 δ # v ′ / 16 F v ′ (\Upsilon_i^{\mathrm{w}}F)_v := F_v+\sum_{v'}\varepsilon_{i-1}^{\delta\#v'/16}F_{v'} ( Υ i w F ) v := F v + ∑ v ′ ε i − 1 δ # v ′ /16 F v ′ , over the children v ′ v' v ′ of v v v , and the weak lift of level i i i is the map F ↦ Corr i w ( t ) Υ i w F F\mapsto\operatorname{Corr}_{i}^{\mathrm{w}}(t)\Upsilon_i^{\mathrm{w}}F F ↦ Corr i w ( t ) Υ i w F . The birth orders of G i G_i G i are at most 1 + ( M + n − i ) N c o r , w 1+(M+n-i)N_{\mathrm{cor,w}} 1 + ( M + n − i ) N cor , w , and its paths have at most ( M + n − i ) N l e n , w (M+n-i)N_{\mathrm{len,w}} ( M + n − i ) N len , w sources.
By Lemma 15.4(a), a source of birth order d σ d_\sigma d σ has coefficients of that order at most C κ C_\kappa C κ times its factor and ε i d σ − 2 ε i − 1 − 1 \varepsilon_i^{d_\sigma-2}\varepsilon_{i-1}^{-1} ε i d σ − 2 ε i − 1 − 1 , each derivative and each lower order costing ε i − 1 − 1 \varepsilon_{i-1}^{-1} ε i − 1 − 1 . The factor of a letter is small because the shear is weak: the first corrector, forced by the shear alone, has the size of the Péclet number, so the eddy diffusivity at degree 2, the average of the shear against it, is κ \kappa κ times the square of the Péclet number times a bounded form. The masses and readout weights turn this gain into the small coupling of (b), with which the smoothed family of level M M M satisfies the invariant.
Lemma 15.4 (The weak coefficients). There exists C ( α , s , q , ζ , Λ , n , t 0 , T ) C(\alpha,s,q,\zeta,\Lambda,n,t_0,T) C ( α , s , q , ζ , Λ , n , t 0 , T ) such that, for every M ≥ M ∗ M \ge M_* M ≥ M ∗ , the following hold.
(a) For every M < i ≤ M + n M < i \le M+n M < i ≤ M + n , every source σ \sigma σ of G i − 1 G_{i-1} G i − 1 , of birth order d σ d_\sigma d σ , every 2 ≤ d ≤ d σ 2 \le d \le d_\sigma 2 ≤ d ≤ d σ and every order N N N used below,
∑ j 1 ≤ N ε i − 1 δ j 1 j 1 ! sup R × T 2 ∣ ∇ j 1 A σ , d ∣ ≤ C fac ( σ ) κ ε i d σ − 2 ε i − 1 d − d σ . \sum_{j_1\le N}\frac{\varepsilon_{i-1}^{\delta j_1}}{j_1!}\sup_{\mathbb{R}\times\mathbb{T}^2}\left|\nabla^{j_1}A_{\sigma,d}\right|\le C\operatorname{fac}(\sigma)\kappa\varepsilon_i^{d_\sigma-2}\varepsilon_{i-1}^{d-d_\sigma}. j 1 ≤ N ∑ j 1 ! ε i − 1 δ j 1 R × T 2 sup ∇ j 1 A σ , d ≤ C fac ( σ ) κ ε i d σ − 2 ε i − 1 d − d σ . (b) The family G M G_M G M has total mass at most 2 3 \frac{2}{3} 3 2 . The family G ~ M \widetilde{G}_M G M , which is G M G_M G M with every source smoothed by the paired smoothing S w S_{\mathrm{w}} S w of width r w r_{\mathrm{w}} r w over round discs (Subsection 15.3), satisfies the clauses of Definition 11.1 on [ t ( M ) , t f ] [t_{(M)},t_{\mathrm{f}}] [ t ( M ) , t f ] with constants depending only on α , s , q , ζ , Λ , n , t 0 , T \alpha,s,q,\zeta,\Lambda,n,t_0,T α , s , q , ζ , Λ , n , t 0 , T (in clause (b1) the majorant of a source is multiplied by such a constant), the time t ( M ) t_{(M)} t ( M ) being that of Proposition 15.5; the row and column sums of its priced majorant matrix and
of its coupling matrices (68) of every order used below, at every scale ρ ≤ ε M \rho\le\varepsilon_M ρ ≤ ε M , are at most
C ε M 2 ( q + 2 ) γ − 5 δ / 64 ≤ C ε M q − 1 . C\varepsilon_M^{2(q+2)\gamma-5\delta/64} \le C\varepsilon_M^{q-1}. C ε M 2 ( q + 2 ) γ − 5 δ /64 ≤ C ε M q − 1 .
Proof. (a) By induction on M + n − i M+n-i M + n − i , from G M + n G_{M+n} G M + n , which has no source. In the chart of a pulse k k k of level i i i , with z = y / ε i z=y/\varepsilon_i z = y / ε i , the system of G i G_i G i is a system on the window of pulse k k k in the sense of Section 10, with κ \kappa κ for κ m \kappa_m κ m and κ m − 1 \kappa_{m-1} κ m − 1 , ε i \varepsilon_i ε i for ε m \varepsilon_m ε m and ε i − 1 \varepsilon_{i-1} ε i − 1 for ε m − 1 \varepsilon_{m-1} ε m − 1 : by (a) at the level i + 1 i+1 i + 1 and the clauses of Definition 11.1 for G i G_i G i , proved below at every level together with (a), its forms are transverse and satisfy (108) . On transverse forms the shear term of the cell operator vanishes, so the inverse of Lemma 15.2 is that of Lemma 10.4, by the uniqueness in Lemma 15.2(a). The proofs of Propositions 10.6 and 10.7 then apply with ε i \varepsilon_i ε i in place of ε ˉ m \bar{\varepsilon}_m ε ˉ m : the four products (113) become ( ε i β / κ ) ε i (\varepsilon_i^\beta/\kappa)\varepsilon_i ( ε i β / κ ) ε i , ε i \varepsilon_i ε i , ε i β \varepsilon_i^\beta ε i β and ε i 2 \varepsilon_i^2 ε i 2 , with ε i β / κ ≤ C \varepsilon_i^\beta/\kappa\le C ε i β / κ ≤ C by Lemma 15.1(a). Along the empty path the bracket of degree 1 is the shear term alone and L 2 Y \mathcal{L}_2Y L 2 Y has no part of zero average, as K \mathbf{K} K does not depend on z z z ; so every corrector of positive degree along ∅ \varnothing ∅ carries one power of the Péclet number ε i β / κ \varepsilon_i^\beta/\kappa ε i β / κ , and the means along ∅ \varnothing ∅ differ from 1 { j = 2 } κ K \mathbf{1}_{\{j=2\}}\kappa\mathbf{K} 1 { j = 2 } κ K by the average of the shear, of size ( κ / ε i ) ( ε i β / κ ) (\kappa/\varepsilon_i)(\varepsilon_i^\beta/\kappa) ( κ / ε i ) ( ε i β / κ ) , against such a corrector. Hence, for every path σ 1 ⋯ σ J \sigma_1\cdots\sigma_J σ 1 ⋯ σ J of G i G_i G i , 2 ≤ d ≤ j 2\le d\le j 2 ≤ d ≤ j and j 1 ≤ N j_1\le N j 1 ≤ N ,
ε i − 1 j 1 ∥ ∇ y j 1 M j , d [ σ 1 ⋯ σ J ] ∥ L ∞ ( [ ( k − 61 48 ) τ i , ( k + 61 48 ) τ i ] × R 2 ) ≤ C fac ( σ 1 ) ⋯ fac ( σ J ) κ ε i j − 2 ε i − 1 d − j , \varepsilon_{i-1}^{j_1}\left\|\nabla_y^{j_1}\mathcal{M}_{j,d}[\sigma_1\cdots\sigma_J]\right\|_{L^\infty\left([(k-\frac{61}{48})\tau_i,(k+\frac{61}{48})\tau_i]\times\mathbb{R}^2\right)}
\le C\operatorname{fac}(\sigma_1)\cdots\operatorname{fac}(\sigma_J)\kappa\varepsilon_i^{j-2}\varepsilon_{i-1}^{d-j}, ε i − 1 j 1 ∇ y j 1 M j , d [ σ 1 ⋯ σ J ] L ∞ ( [( k − 48 61 ) τ i , ( k + 48 61 ) τ i ] × R 2 ) ≤ C fac ( σ 1 ) ⋯ fac ( σ J ) κ ε i j − 2 ε i − 1 d − j ,
ε i − 1 j 1 ∥ ∇ y j 1 ( M j , d [ ∅ ] − 1 { j = 2 } κ K ) ∥ L ∞ ( [ ( k − 61 48 ) τ i , ( k + 61 48 ) τ i ] × R 2 ) ≤ C κ ( ε i β κ ) 2 ε i j − 2 ε i − 1 d − j . \varepsilon_{i-1}^{j_1}\left\|\nabla_y^{j_1}\left(\mathcal{M}_{j,d}[\varnothing]-\mathbf{1}_{\{j=2\}}\kappa\mathbf{K}\right)\right\|_{L^\infty\left([(k-\frac{61}{48})\tau_i,(k+\frac{61}{48})\tau_i]\times\mathbb{R}^2\right)}
\le C\kappa\left(\frac{\varepsilon_i^\beta}{\kappa}\right)^2\varepsilon_i^{j-2}\varepsilon_{i-1}^{d-j}. ε i − 1 j 1 ∇ y j 1 ( M j , d [ ∅ ] − 1 { j = 2 } κ K ) L ∞ ( [( k − 48 61 ) τ i , ( k + 48 61 ) τ i ] × R 2 ) ≤ C κ ( κ ε i β ) 2 ε i j − 2 ε i − 1 d − j .
These means of degree d σ d_\sigma d σ , with the power ε i − 1 − δ / 16 \varepsilon_{i-1}^{-\delta/16} ε i − 1 − δ /16 for the letters, are the coefficients of Definition 15.3, transported to x x x with the derivatives of the charts at the scale ε i − 1 \varepsilon_{i-1} ε i − 1 (Lemma 7.7); this gives (a).
(b) Row and column sums. Every source of G M G_M G M was created at the weak passage from level M + 1 M+1 M + 1 , with birth radius ε M + 1 ≤ r w \varepsilon_{M+1}\le r_w ε M + 1 ≤ r w , so its price in (117) is 1. By (a) with i = M + 1 i=M+1 i = M + 1 , its coefficients of order d d d are at most C fac ( σ ) κ r w d − 2 C\operatorname{fac}(\sigma)\kappa r_w^{d-2} C fac ( σ ) κ r w d − 2 , with derivatives at the scale ε M \varepsilon_M ε M ; so the entries of the coupling matrices (68) of G ~ M \widetilde{G}_M G M of every order used below, at the scale ε M \varepsilon_M ε M , are at most C C C times those of the majorant matrix (117) . These entries increase with ρ \rho ρ , so the same holds at every scale ρ ≤ ε M \rho\le\varepsilon_M ρ ≤ ε M , among them ρ m o l ( M ) = c m o l ε M \rho_{\mathrm{mol}}(M)=c_{\mathrm{mol}}\varepsilon_M ρ mol ( M ) = c mol ε M (by (115) ), as κ ε M 2 − β ≥ 1 2 ε M 2 − γ ≥ ε M 2 \kappa\varepsilon_M^{2-\beta}\ge\frac12\varepsilon_M^{2-\gamma}\ge\varepsilon_M^2 κ ε M 2 − β ≥ 2 1 ε M 2 − γ ≥ ε M 2 by the proof of Lemma 15.1 and ε M γ ≤ 1 2 \varepsilon_M^\gamma\le\frac12 ε M γ ≤ 2 1 for M ≥ M ∗ M\ge M_* M ≥ M ∗ . Along a letter from a component to its child, created at the weak passage from a level i > M i>M i > M , the square root of the mass ratio and the factor give
( 4 n N c o r , w ) 1 / 2 ε i − 1 − δ / 64 ⋅ ε i − 1 − δ / 16 ( ε i β κ ) 2 ≤ C ε i 2 ( q + 2 ) γ − 5 δ / 64 ≤ C ε M 2 ( q + 2 ) γ − 5 δ / 64 , (4nN_{\mathrm{cor},w})^{1/2}\varepsilon_{i-1}^{-\delta/64}\cdot\varepsilon_{i-1}^{-\delta/16}\left(\frac{\varepsilon_i^\beta}{\kappa}\right)^2
\le C\varepsilon_i^{2(q+2)\gamma-5\delta/64}\le C\varepsilon_M^{2(q+2)\gamma-5\delta/64}, ( 4 n N cor , w ) 1/2 ε i − 1 − δ /64 ⋅ ε i − 1 − δ /16 ( κ ε i β ) 2 ≤ C ε i 2 ( q + 2 ) γ − 5 δ /64 ≤ C ε M 2 ( q + 2 ) γ − 5 δ /64 ,
by Lemma 15.1(a). For an inherited source, the factor is the product of the factors along its path, the square roots of the mass ratios telescope, and the word multiplies the masses at both ends by the same number; so its entry is the product of the entries of the letters in its history. Since G M + n G_{M+n} G M + n has no source, every source of G M G_M G M has at least one letter, and at most N l e n , w N_{\mathrm{len},w} N len , w , in its history, and every entry is at most C ε M 2 ( q + 2 ) γ − 5 δ / 64 C\varepsilon_M^{2(q+2)\gamma-5\delta/64} C ε M 2 ( q + 2 ) γ − 5 δ /64 . A row or a column has a number of entries bounded by the numbers of words, paths and degrees; this gives the sums, and 2 ( q + 2 ) γ − 5 δ 64 > q − 1 2(q+2)\gamma-\frac{5\delta}{64}>q-1 2 ( q + 2 ) γ − 64 5 δ > q − 1 since ( q + 2 ) γ ≥ q − 1 (q+2)\gamma\ge q-1 ( q + 2 ) γ ≥ q − 1 (as β > 1 \beta>1 β > 1 ) and 5 4 δ < q − 1 \frac54\delta<q-1 4 5 δ < q − 1 . The masses: a weak passage multiplies the total mass by at most 1 + ∑ L ≥ 1 ( 4 n ) − L = 1 + 1 / ( 4 n − 1 ) 1+\sum_{L\ge1}(4n)^{-L}=1+1/(4n-1) 1 + ∑ L ≥ 1 ( 4 n ) − L = 1 + 1/ ( 4 n − 1 ) , and ( 1 + 1 / ( 4 n − 1 ) ) n ≤ 4 3 (1+1/(4n-1))^n\le\frac43 ( 1 + 1/ ( 4 n − 1 ) ) n ≤ 3 4 .
The other clauses of Definition 11.1 are proved at every level, for G i − 1 G_{i-1} G i − 1 , M < i ≤ M + n M<i\le M+n M < i ≤ M + n , and G ~ M \widetilde{G}_M G M , together with (a), which uses them at the level i i i . They are checked as for the successors in Subsection 11.6, the weak finite lift replacing the finite lift of Section 10, with the inverse of Lemma 10.4 on transverse forms, as in (a): clauses (a1)–(a7) by Steps 1 and 2 there, the weak cell solutions being solved once on their windows and unique there (Lemma 15.2(a)). Clause (b1) holds by Steps 4 and 5, with res ( v ) = res i \operatorname{res}(v)=\operatorname{res}_i res ( v ) = res i and res ( v ~ ) = res i − 1 = res i − 1 / ( N g e n + 2 + N o f f ) \operatorname{res}(\widetilde{v})=\operatorname{res}_{i-1}=\operatorname{res}_i-1/(N_{\mathrm{gen}}+2+N_{\mathrm{off}}) res ( v ) = res i − 1 = res i − 1/ ( N gen + 2 + N off ) for every component of G i − 1 G_{i-1} G i − 1 (within G i G_i G i all reserves are equal, which replaces the monotone generations of Step 4), a weak passage taking at most n N c o r , w − 1 ≤ N s r c − 2 nN_{\mathrm{cor,w}}-1\le N_{\mathrm{src}}-2 n N cor , w − 1 ≤ N src − 2 slow derivatives (N s r c ≥ N s r c , e ≥ n N c o r , w + 6 N_{\mathrm{src}}\ge N_{\mathrm{src,e}}\ge nN_{\mathrm{cor,w}}+6 N src ≥ N src , e ≥ n N cor , w + 6 , as N s r c N_{\mathrm{src}} N src counts from the seed’s birth orders, item (1) of Subsection 11.1). The entry across the pulse of the conductivity has real part at least 9 10 \frac{9}{10} 10 9 on the outer domain by (15) at the level i i i , so the bounds of (a) hold in the analytic norms; with the birth radius ε i \varepsilon_i ε i , the left side of (b1) is at most C ( ε i / ε i − 1 ) d σ − d C(\varepsilon_i/\varepsilon_{i-1})^{d_\sigma-d} C ( ε i / ε i − 1 ) d σ − d . Clauses (b2) and (b2′) hold by Cauchy’s estimate, as in Step 6, the birth radii being ε i ≤ ε i − 1 \varepsilon_i\le\varepsilon_{i-1} ε i ≤ ε i − 1 , and, for G ~ M \widetilde{G}_M G M , ε M + 1 ≤ ρ m o l ( M ) = c m o l ε M \varepsilon_{M+1}\le\rho_{\mathrm{mol}}(M)=c_{\mathrm{mol}}\varepsilon_M ε M + 1 ≤ ρ mol ( M ) = c mol ε M ; (b3) and (d) by the sums above; and, for G ~ M \widetilde{G}_M G M , (c1) as the long axes of S w S_w S w have the inverse 2 1 / 2 I 2^{1/2}I 2 1/2 I , (c3) by Lemma 8.2, and (c2) as r w / ρ m o l ( M ) = ε M γ / 4 + δ / 2 / c m o l ≤ C ε M δ r_w/\rho_{\mathrm{mol}}(M)=\varepsilon_M^{\gamma/4+\delta/2}/c_{\mathrm{mol}}\le C\varepsilon_M^\delta r w / ρ mol ( M ) = ε M γ /4 + δ /2 / c mol ≤ C ε M δ and r w / ρ ( M ) ≤ C ε M − 1 3 δ + 5 2 q γ + 1 ≤ C ε M − 1 4 δ r_w/\rho(M)\le C\varepsilon_{M-1}^{3\delta+\frac{5}{2}q\gamma+1}\le C\varepsilon_{M-1}^{4\delta} r w / ρ ( M ) ≤ C ε M − 1 3 δ + 2 5 q γ + 1 ≤ C ε M − 1 4 δ , by 3 δ < γ 3\delta<\gamma 3 δ < γ , Proposition 2.1(c) and q − 1 − γ 2 − δ = q γ + 3 δ q-1-\frac{\gamma}{2}-\delta=q\gamma+3\delta q − 1 − 2 γ − δ = q γ + 3 δ .
Proposition 15.5 carries out item 1 of the route: items (i)–(iii) restart each solution in the next family, with the new components starting from zero, and (iv) follows from Lemma 15.6.
Proposition 15.5 (The weak reduction). There exists C ( α , s , q , ζ , Λ , n , t 0 , T ) C(\alpha,s,q,\zeta,\Lambda,n,t_0,T) C ( α , s , q , ζ , Λ , n , t 0 , T ) such that, for every M ≥ M ∗ M\ge M_* M ≥ M ∗ , the families G i G_i G i do not depend on t i n t_{\mathrm{in}} t in , θ i n \theta_{\mathrm{in}} θ in or θ \theta θ , and there exists t ( M ) ∈ [ t i n , t i n + 6 n h M + 1 ] t_{(M)}\in[t_{\mathrm{in}},t_{\mathrm{in}}+6nh_{M+1}] t ( M ) ∈ [ t in , t in + 6 n h M + 1 ] , depending only on M M M , κ \kappa κ and t i n t_{\mathrm{in}} t in , with the following property. For every θ i n ∈ L 2 ( T 2 ) \theta_{\mathrm{in}}\in L^2(\mathbb{T}^2) θ in ∈ L 2 ( T 2 ) and every weak solution θ \theta θ of (28) there exist times t i n < t ( M + n ) < t ( M + n − 1 ) < ⋯ < t ( M ) t_{\mathrm{in}}<t_{(M+n)}<t_{(M+n-1)}<\cdots<t_{(M)} t in < t ( M + n ) < t ( M + n − 1 ) < ⋯ < t ( M ) and solutions U i U_i U i of the systems (34) of G i G_i G i , with κ \kappa κ in place of their diffusivity κ m \kappa_m κ m and without smoothing, on [ t ( i ) , t f ] [t_{(i)},t_f] [ t ( i ) , t f ] , in the class of Proposition 8.6, such that:
(i) U M + n = θ U_{M+n}=\theta U M + n = θ on [ t ( M + n ) , t f ] [t_{(M+n)},t_f] [ t ( M + n ) , t f ] ;
(ii) for every M < i ≤ M + n M<i\le M+n M < i ≤ M + n , t ( i − 1 ) t_{(i-1)} t ( i − 1 ) is the block boundary at the centre of the first pair of quiet margins [ ( l + 1 2 ) h i − τ i , ( l + 1 2 ) h i + τ i ] [(l+\frac12)h_i-\tau_i,(l+\frac12)h_i+\tau_i] [( l + 2 1 ) h i − τ i , ( l + 2 1 ) h i + τ i ] , l ∈ Z l\in\mathbb{Z} l ∈ Z , whose left end is at least t ( i ) t_{(i)} t ( i ) ; so t ( i ) + τ i ≤ t ( i − 1 ) ≤ t ( i ) + 2 h i t_{(i)}+\tau_i\le t_{(i-1)}\le t_{(i)}+2h_i t ( i ) + τ i ≤ t ( i − 1 ) ≤ t ( i ) + 2 h i , and b i = b i − 1 \mathbf{b}_i=\mathbf{b}_{i-1} b i = b i − 1 on these margins;
(iii) for every M < i ≤ M + n M<i\le M+n M < i ≤ M + n , U i − 1 ( t ( i − 1 ) ) U_{i-1}(t_{(i-1)}) U i − 1 ( t ( i − 1 ) ) is U i ( t ( i − 1 ) ) U_i(t_{(i-1)}) U i ( t ( i − 1 ) ) on the components of G i G_i G i and 0 0 0 on their children;
(iv) for every M ≤ i ≤ M + n M\le i\le M+n M ≤ i ≤ M + n , ∥ U i ∥ L ∞ ( [ t ( i ) , t f ] ; L ω 2 ( T 2 ) ) ≤ C ∥ θ i n ∥ L 2 ( T 2 ) \|U_i\|_{L^\infty([t_{(i)},t_f];L^2_\omega(\mathbb{T}^2))}\le C\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)} ∥ U i ∥ L ∞ ([ t ( i ) , t f ] ; L ω 2 ( T 2 )) ≤ C ∥ θ in ∥ L 2 ( T 2 ) , and
∥ U M ∥ L ∞ ( [ t ( M ) , t f ] ; H ε M , ω N t r , w ( T 2 ) + κ 1 / 2 ∥ ∇ U M ∥ L 2 ( ( t ( M ) , t f ) ; H ε M , ω N t r , w ( T 2 ) ) ≤ C ∥ θ i n ∥ L 2 ( T 2 ) . \|U_M\|_{L^\infty([t_{(M)},t_f];H^{N_{\mathrm{tr,w}}}_{\varepsilon_M,\omega}(\mathbb{T}^2)}
+\kappa^{1/2}\|\nabla U_M\|_{L^2((t_{(M)},t_f);H^{N_{\mathrm{tr,w}}}_{\varepsilon_M,\omega}(\mathbb{T}^2))}
\le C\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)}. ∥ U M ∥ L ∞ ([ t ( M ) , t f ] ; H ε M , ω N tr , w ( T 2 ) + κ 1/2 ∥∇ U M ∥ L 2 (( t ( M ) , t f ) ; H ε M , ω N tr , w ( T 2 )) ≤ C ∥ θ in ∥ L 2 ( T 2 ) .
Lemma 15.6 is the regain of derivatives on the quiet margins of item 1 of the route.
Lemma 15.6 (A complete trace on the quiet margins). There exists C ( α , s , q , ζ , Λ , n , t 0 , T ) C(\alpha,s,q,\zeta,\Lambda,n,t_0,T) C ( α , s , q , ζ , Λ , n , t 0 , T ) such that the following holds for every M ≥ M ∗ M\ge M_* M ≥ M ∗ , M < i ≤ M + n M<i\le M+n M < i ≤ M + n , θ i n \theta_{\mathrm{in}} θ in and θ \theta θ . For a component v v v of G i − 1 G_{i-1} G i − 1 , let N v N_v N v be N t r , w N_{\mathrm{tr,w}} N tr , w plus n N c o r , w − 1 nN_{\mathrm{cor,w}}-1 n N cor , w − 1 times the largest number of sources on a path of G i − 1 G_{i-1} G i − 1 that starts at v v v (the components of G i G_i G i are among those of G i − 1 G_{i-1} G i − 1 ). Then the solutions of Proposition 15.5 satisfy
∥ U i ( t ( i − 1 ) ) ∥ H ε i − 1 , ω N v ( T 2 ) ≤ C ∥ U i ∥ L ∞ ( [ t ( i ) , t ( i − 1 ) ] ; L ω 2 ( T 2 ) ) . \|U_i(t_{(i-1)})\|_{H^{N_v}_{\varepsilon_{i-1},\omega}(\mathbb{T}^2)}
\le C\|U_i\|_{L^\infty([t_{(i)},t_{(i-1)}];L^2_\omega(\mathbb{T}^2))}. ∥ U i ( t ( i − 1 ) ) ∥ H ε i − 1 , ω N v ( T 2 ) ≤ C ∥ U i ∥ L ∞ ([ t ( i ) , t ( i − 1 ) ] ; L ω 2 ( T 2 )) .
Proof. The heat interval [ t ( i − 1 ) − ε i − 1 2 / κ , t ( i − 1 ) ] [t_{(i-1)}-\varepsilon_{i-1}^2/\kappa,t_{(i-1)}] [ t ( i − 1 ) − ε i − 1 2 / κ , t ( i − 1 ) ] lies in the second half of the quiet margin that ends at t ( i − 1 ) t_{(i-1)} t ( i − 1 ) , since ε i − 1 2 / κ ≤ C ε i γ − 3 δ τ i 2 ≤ 1 2 τ i 2 \varepsilon_{i-1}^2/\kappa\le C\varepsilon_i^{\gamma-3\delta}\tau_i^2\le\frac12\tau_i^2 ε i − 1 2 / κ ≤ C ε i γ − 3 δ τ i 2 ≤ 2 1 τ i 2 by Lemma 15.1(a); and every time within 5 6 h i + 1 \frac{5}{6}h_{i+1} 6 5 h i + 1 of it lies in the two quiet margins around t ( i − 1 ) t_{(i-1)} t ( i − 1 ) , since 5 6 h i + 1 ≤ 1 2 τ i \frac{5}{6}h_{i+1}\le\frac12\tau_i 6 5 h i + 1 ≤ 2 1 τ i by (7) and Lemma 7.1(a), (c).
Step 1: the coefficients on the heat interval. Let N ∗ : = max v N v + n N l e n , w ( 1 + n N c o r , w ) N_*:=\max_vN_v+nN_{\mathrm{len,w}}(1+nN_{\mathrm{cor,w}}) N ∗ := max v N v + n N len , w ( 1 + n N cor , w ) , an auxiliary order depending only on α , s , q , ζ , Λ , n , t 0 , T \alpha,s,q,\zeta,\Lambda,n,t_0,T α , s , q , ζ , Λ , n , t 0 , T (M ∗ M_* M ∗ is taken after it). On the heat interval the coefficients of G i G_i G i have their slow derivatives at the wavelength ε i − 1 \varepsilon_{i-1} ε i − 1 , up to the order N ∗ N_* N ∗ : for a source σ : v ′ → v \sigma:v'\to v σ : v ′ → v of G i G_i G i , of birth order d σ d_\sigma d σ , let
B ~ σ : = ( ω v ′ ω v ) 1 / 2 ∑ d = 2 d σ 1 κ ε i − 1 d − 2 ∑ j = 0 N ∗ ε i − 1 j j ! sup ∣ ∇ j A σ , d ∣ , \widetilde B_\sigma:=\left(\frac{\omega_{v'}}{\omega_v}\right)^{1/2}\sum_{d=2}^{d_\sigma}\frac{1}{\kappa\varepsilon_{i-1}^{d-2}}\sum_{j=0}^{N_*}\frac{\varepsilon_{i-1}^{j}}{j!}\sup\left|\nabla^j A_{\sigma,d}\right|, B σ := ( ω v ω v ′ ) 1/2 d = 2 ∑ d σ κ ε i − 1 d − 2 1 j = 0 ∑ N ∗ j ! ε i − 1 j sup ∇ j A σ , d ,
the supremum over the heat interval and T 2 \mathbb{T}^2 T 2 ; then the matrix B ~ v v ′ : = ∑ σ : v ′ → v B ~ σ \widetilde B_{vv'}:=\sum_{\sigma:v'\to v}\widetilde B_\sigma B v v ′ := ∑ σ : v ′ → v B σ has row and column sums at most C C C . Indeed, a coefficient of G i G_i G i at a time t t t reads the flow of b i \mathbf b_i b i only over the histories of the pulses of level i + 1 i+1 i + 1 whose cutoff is positive at t t t , which lie within 5 6 h i + 1 \frac{5}{6}h_{i+1} 6 5 h i + 1 of t t t (the nesting after (121) ), hence, for t t t in the heat interval, in the margins, where b i = b i − 1 \mathbf b_i=\mathbf b_{i-1} b i = b i − 1 . So there the coefficients do not depend on the fast variable of level i i i , by (a4), and their slow derivatives cost ε i − 1 − 1 \varepsilon_{i-1}^{-1} ε i − 1 − 1 at every order, by Cauchy’s estimate in the chart parameters from (b1) and by the transport (11.8); (a4) and (b1) hold for G i G_i G i by the proof of Lemma 15.4 . With the sizes of Lemma 15.4 (a) at the level i + 1 i+1 i + 1 and ε i + 1 d σ − 2 ε i d − d σ ≤ ε i − 1 d − 2 \varepsilon_{i+1}^{d_\sigma-2}\varepsilon_i^{d-d_\sigma}\leq\varepsilon_{i-1}^{d-2} ε i + 1 d σ − 2 ε i d − d σ ≤ ε i − 1 d − 2 , this gives B ~ σ ≤ C ( ω v ′ / ω v ) 1 / 2 fac ( σ ) \widetilde B_\sigma\leq C(\omega_{v'}/\omega_v)^{1/2}\operatorname{fac}(\sigma) B σ ≤ C ( ω v ′ / ω v ) 1/2 fac ( σ ) , whose row and column sums are at most C C C by the telescoping in the proof of Lemma 15.4 (b).
Step 2: one path. We rescale time so that the heat interval is [ 0 , 1 ] [0,1] [ 0 , 1 ] , t = t ( i − 1 ) − ( 1 − t ) ε i − 1 2 / κ t=t_{(i-1)}-(1-t)\varepsilon_{i-1}^2/\kappa t = t ( i − 1 ) − ( 1 − t ) ε i − 1 2 / κ , write U v ( t ) : = ω v − 1 / 2 U i , v ( t ) U_v(t):=\omega_v^{-1/2}U_{i,v}(t) U v ( t ) := ω v − 1/2 U i , v ( t ) , and measure lengths at the scale ε i − 1 \varepsilon_{i-1} ε i − 1 , in the norms H ε i − 1 N ( T 2 ) H^N_{\varepsilon_{i-1}}(\mathbb{T}^2) H ε i − 1 N ( T 2 ) , negative orders N N N being the dual norms. The drift b i − 1 \mathbf b_{i-1} b i − 1 then has derivatives at most C ε i − 1 β / κ ≤ C ε i − 1 γ C\varepsilon_{i-1}^{\beta}/\kappa\leq C\varepsilon_{i-1}^{\gamma} C ε i − 1 β / κ ≤ C ε i − 1 γ at the unit scale, by Lemma 15.1(a) and the bound ∥ ∇ j b i − 1 ∥ L ∞ ( R × T 2 ) ≤ C j ( j − 1 ) ! ε i − 1 β − 1 − j \|\nabla^j\mathbf b_{i-1}\|_{L^\infty(\mathbb{R}\times\mathbb{T}^2)}\leq C^j(j-1)!\varepsilon_{i-1}^{\beta-1-j} ∥ ∇ j b i − 1 ∥ L ∞ ( R × T 2 ) ≤ C j ( j − 1 )! ε i − 1 β − 1 − j of § 5.6 . Let E ( t , t ′ ) \mathcal E(t,t') E ( t , t ′ ) be the evolution of the scalar part of the system of G i G_i G i , drift and diffusion, and F σ ( t ) g : = ( ε i − 1 2 / κ ) ( ω v ′ / ω v ) 1 / 2 ∇ ⋅ ∑ d A σ , d ∇ d − 1 g F_\sigma(t)g:=(\varepsilon_{i-1}^2/\kappa)(\omega_{v'}/\omega_v)^{1/2}\nabla\cdot\sum_d A_{\sigma,d}\nabla^{d-1}g F σ ( t ) g := ( ε i − 1 2 / κ ) ( ω v ′ / ω v ) 1/2 ∇ ⋅ ∑ d A σ , d ∇ d − 1 g the source term of σ \sigma σ . The rate κ / ε i − 1 2 \kappa/\varepsilon_{i-1}^2 κ / ε i − 1 2 of a source cancels against the length of the heat interval, and the product rule and its dual give, for ∣ N ∣ , ∣ N − d σ ∣ ≤ N ∗ |N|,|N-d_\sigma|\leq N_* ∣ N ∣ , ∣ N − d σ ∣ ≤ N ∗ ,
∥ F σ ( t ) g ∥ H ε i − 1 N − d σ ( T 2 ) ≤ C B ~ σ ∥ g ∥ H ε i − 1 N ( T 2 ) : \|F_\sigma(t)g\|_{H^{N-d_\sigma}_{\varepsilon_{i-1}}(\mathbb{T}^2)}\leq C\widetilde B_\sigma\|g\|_{H^N_{\varepsilon_{i-1}}(\mathbb{T}^2)}: ∥ F σ ( t ) g ∥ H ε i − 1 N − d σ ( T 2 ) ≤ C B σ ∥ g ∥ H ε i − 1 N ( T 2 ) :
at a fixed time a source loses its whole birth order d σ d_\sigma d σ , not the d σ − 2 d_\sigma-2 d σ − 2 of the energy estimates. As the sources of G i G_i G i form no cycle and its paths have at most ( M + n − i ) N l e n , w (M+n-i)N_{\mathrm{len,w}} ( M + n − i ) N len , w sources, Duhamel’s formula, iterated in the spaces H ε i − 1 N ( T 2 ) H^N_{\varepsilon_{i-1}}(\mathbb{T}^2) H ε i − 1 N ( T 2 ) , ∣ N ∣ ≤ N ∗ |N|\leq N_* ∣ N ∣ ≤ N ∗ , writes U v ( 1 ) U_v(1) U v ( 1 ) as the sum, over the paths σ 1 ⋯ σ J \sigma_1\cdots\sigma_J σ 1 ⋯ σ J of G i G_i G i from a component v ′ v' v ′ to v v v , of
∫ 0 < t 1 < ⋯ < t J < 1 E ( 1 , t J ) F σ J ( t J ) E ( t J , t J − 1 ) ⋯ F σ 1 ( t 1 ) E ( t 1 , 0 ) U v ′ ( 0 ) d t 1 ⋯ d t J , \int_{0<t_1<\cdots<t_J<1}\mathcal E(1,t_J)F_{\sigma_J}(t_J)\mathcal E(t_J,t_{J-1})\cdots F_{\sigma_1}(t_1)\mathcal E(t_1,0)U_{v'}(0)\,\mathrm{d}t_1\cdots\mathrm{d}t_J, ∫ 0 < t 1 < ⋯ < t J < 1 E ( 1 , t J ) F σ J ( t J ) E ( t J , t J − 1 ) ⋯ F σ 1 ( t 1 ) E ( t 1 , 0 ) U v ′ ( 0 ) d t 1 ⋯ d t J ,
the empty path, J = 0 J=0 J = 0 , giving E ( 1 , 0 ) U v ( 0 ) \mathcal E(1,0)U_v(0) E ( 1 , 0 ) U v ( 0 ) . Fix the path and the times. One of the J + 1 J+1 J + 1 intervals between 0 , t 1 , … , t J , 1 0,t_1,\ldots,t_J,1 0 , t 1 , … , t J , 1 has length at least 1 / ( J + 1 ) 1/(J+1) 1/ ( J + 1 ) ; let d − d_- d − and d + d_+ d + be the sums of the birth orders of the sources before and after it, so that N v + d − + d + ≤ N ∗ N_v+d_-+d_+\leq N_* N v + d − + d + ≤ N ∗ . Before this long interval we follow the integrand in H ε i − 1 − N ( T 2 ) H^{-N}_{\varepsilon_{i-1}}(\mathbb{T}^2) H ε i − 1 − N ( T 2 ) , down to N = d − N=d_- N = d − , and after it in H ε i − 1 N v + N ( T 2 ) H^{N_v+N}_{\varepsilon_{i-1}}(\mathbb{T}^2) H ε i − 1 N v + N ( T 2 ) , from N = d + N=d_+ N = d + down to 0 0 0 , spaces on which the scalar evolutions are bounded, uniformly in the lengths of their intervals, by the energy estimate and its dual. Across the long interval, split at its midpoint, the adjoint evolution maps L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) to H ε i − 1 d − ( T 2 ) H^{d_-}_{\varepsilon_{i-1}}(\mathbb{T}^2) H ε i − 1 d − ( T 2 ) over the first half and E \mathcal E E maps L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) to H ε i − 1 N v + d + ( T 2 ) H^{N_v+d_+}_{\varepsilon_{i-1}}(\mathbb{T}^2) H ε i − 1 N v + d + ( T 2 ) over the second, by energy estimates with time weights, with constants C ( J ) C(J) C ( J ) , as both halves have length at least 1 / ( 2 ( J + 1 ) ) 1/(2(J+1)) 1/ ( 2 ( J + 1 )) and the orders are at most N ∗ N_* N ∗ . So the integrand is at most C ( J ) B ~ σ 1 ⋯ B ~ σ J ∥ U v ′ ( 0 ) ∥ L 2 ( T 2 ) C(J)\widetilde B_{\sigma_1}\cdots\widetilde B_{\sigma_J}\|U_{v'}(0)\|_{L^2(\mathbb{T}^2)} C ( J ) B σ 1 ⋯ B σ J ∥ U v ′ ( 0 ) ∥ L 2 ( T 2 ) in H ε i − 1 N v ( T 2 ) H^{N_v}_{\varepsilon_{i-1}}(\mathbb{T}^2) H ε i − 1 N v ( T 2 ) , and the ordered simplex has volume 1 / J ! ≤ 1 1/J!\leq1 1/ J ! ≤ 1 .
Step 3: the sum over paths. Summing over the paths,
∥ U v ( 1 ) ∥ H ε i − 1 N v ( T 2 ) ≤ ∑ v ′ ( ∑ J = 0 ( M + n − i ) N l e n , w C ( J ) ( B ~ ) J ) ∥ U v ′ ( 0 ) ∥ L 2 ( T 2 ) . \|U_v(1)\|_{H^{N_v}_{\varepsilon_{i-1}}(\mathbb{T}^2)}
\leq\sum_{v'}\left(\sum_{J=0}^{(M+n-i)N_{\mathrm{len,w}}}C(J)(\widetilde B)^J\right)\|U_{v'}(0)\|_{L^2(\mathbb{T}^2)}. ∥ U v ( 1 ) ∥ H ε i − 1 N v ( T 2 ) ≤ v ′ ∑ J = 0 ∑ ( M + n − i ) N len , w C ( J ) ( B ) J ∥ U v ′ ( 0 ) ∥ L 2 ( T 2 ) .
The row and column sums of this matrix are at most C C C , as its degree, the C ( J ) C(J) C ( J ) and the row and column sums of B ~ \widetilde{B} B (Step 1) depend only on α , s , q , ζ , Λ , n , t 0 , T \alpha, s, q, \zeta, \Lambda, n, t_0, T α , s , q , ζ , Λ , n , t 0 , T . So, by Schur’s test, ∥ U i ( t ( i − 1 ) ) ∥ H ε i − 1 , ω N v ( T 2 ) ≤ C ∥ U i ( t ( i − 1 ) − ε i − 1 2 / κ ) ∥ L ω 2 ( T 2 ) \|U_i(t_{(i-1)})\|_{H^{N_v}_{\varepsilon_{i-1},\omega}(T^2)} \le C\|U_i(t_{(i-1)}-\varepsilon_{i-1}^2/\kappa)\|_{L^2_\omega(T^2)} ∥ U i ( t ( i − 1 ) ) ∥ H ε i − 1 , ω N v ( T 2 ) ≤ C ∥ U i ( t ( i − 1 ) − ε i − 1 2 / κ ) ∥ L ω 2 ( T 2 ) , and t ( i − 1 ) − ε i − 1 2 / κ ≥ t ( i ) t_{(i-1)}-\varepsilon_{i-1}^2/\kappa\ge t_{(i)} t ( i − 1 ) − ε i − 1 2 / κ ≥ t ( i ) . □ \square □
Proof of Proposition 15.5. Let t ( M + n ) t_{(M+n)} t ( M + n ) be the left end of the first pair of quiet margins of level M + n M+n M + n whose left end is greater than t i n t_{\mathrm{in}} t in , so that t i n < t ( M + n ) ≤ t i n + h M + n t_{\mathrm{in}} < t_{(M+n)} \le t_{\mathrm{in}}+h_{M+n} t in < t ( M + n ) ≤ t in + h M + n , and let U M + n : = θ U_{M+n}:=\theta U M + n := θ on [ t ( M + n ) , t f ] [t_{(M+n)},t_f] [ t ( M + n ) , t f ] , the solution of the single equation of G M + n G_{M+n} G M + n , with ∥ θ ∥ L ∞ ( [ t i n , t f ] ; L 2 ( T 2 ) ) ≤ ∥ θ i n ∥ L 2 ( T 2 ) \|\theta\|_{L^\infty([t_{\mathrm{in}},t_f];L^2(T^2))}\le\|\theta_{\mathrm{in}}\|_{L^2(T^2)} ∥ θ ∥ L ∞ ([ t in , t f ] ; L 2 ( T 2 )) ≤ ∥ θ in ∥ L 2 ( T 2 ) . Since b M + n b_{M+n} b M + n is smooth, θ \theta θ is smooth for t > t i n t>t_{\mathrm{in}} t > t in , by parabolic regularity, so it lies in the class of Proposition 8.6 at every order on [ t ( M + n ) , t f ] [t_{(M+n)},t_f] [ t ( M + n ) , t f ] . Let M < i ≤ M + n M<i\le M+n M < i ≤ M + n , let U i U_i U i be constructed, with ∥ U i ∥ L ∞ ( [ t ( i ) , t f ] ; L ω 2 ( T 2 ) ) ≤ C ∥ θ i n ∥ L 2 ( T 2 ) \|U_i\|_{L^\infty([t_{(i)},t_f];L^2_\omega(T^2))}\le C\|\theta_{\mathrm{in}}\|_{L^2(T^2)} ∥ U i ∥ L ∞ ([ t ( i ) , t f ] ; L ω 2 ( T 2 )) ≤ C ∥ θ in ∥ L 2 ( T 2 ) , and define t ( i − 1 ) t_{(i-1)} t ( i − 1 ) by (ii); on the quiet margins of level i i i the pulses of level i i i vanish, so b i = b i − 1 b_i=b_{i-1} b i = b i − 1 there. By Lemma 15.6, the datum (iii) has the orders N ν N_\nu N ν of that lemma at the scale ε i − 1 \varepsilon_{i-1} ε i − 1 , with norm at most C ∥ θ i n ∥ L 2 ( T 2 ) C\|\theta_{\mathrm{in}}\|_{L^2(T^2)} C ∥ θ in ∥ L 2 ( T 2 ) . Proposition 8.6 applies to G i − 1 G_{i-1} G i − 1 at these orders, which decrease along every source σ \sigma σ by at least n N c o r , w − 1 ≥ d σ − 2 nN_{\mathrm{cor,w}}-1\ge d_\sigma-2 n N cor , w − 1 ≥ d σ − 2 , as it asks, and at the scale ε i − 1 \varepsilon_{i-1} ε i − 1 : its drift condition holds by Step 2 of the proof of Lemma 15.6, and its matrix is at most C C C times the majorant matrix, by Lemma 15.4(a), with row and column sums at most C ε M q − 1 ≤ 1 8 C\varepsilon_{M}^{q-1}\le\frac18 C ε M q − 1 ≤ 8 1 as in the proof of Lemma 15.4(b). It gives U i − 1 U_{i-1} U i − 1 on [ t ( i − 1 ) , t f ] [t_{(i-1)},t_f] [ t ( i − 1 ) , t f ] , with the energy of that proposition at most C C C times the norm of the datum. After n n n restarts this gives (iv), the constant growing by a fixed factor per restart, as the orders of the components of G M G_M G M are at least N t r , w N_{\mathrm{tr,w}} N tr , w . By (ii), the times t ( i ) t_{(i)} t ( i ) are determined by t i n t_{\mathrm{in}} t in and the margins; as h i ≤ h M + 1 h_i\le h_{M+1} h i ≤ h M + 1 , t ( M ) ≤ t ( M + n ) + 2 n h M + 1 ≤ t i n + 6 n h M + 1 < t f t_{(M)}\le t_{(M+n)}+2nh_{M+1}\le t_{\mathrm{in}}+6nh_{M+1}<t_f t ( M ) ≤ t ( M + n ) + 2 n h M + 1 ≤ t in + 6 n h M + 1 < t f for M ≥ M ∗ M\ge M_* M ≥ M ∗ . □ \square □
For M ≤ i ≤ M + n M\le i\le M+n M ≤ i ≤ M + n we write N i : = ( M + n − i ) N c o r , w + 5 N_i:=(M+n-i)N_{\mathrm{cor,w}}+5 N i := ( M + n − i ) N cor , w + 5 : the composed weak lifts from level i i i to level M + n M+n M + n spend ( M + n − i ) N c o r , w (M+n-i)N_{\mathrm{cor,w}} ( M + n − i ) N cor , w derivatives.
A weak passage is the comparison of § 5.8 without the truncated exponential. Of its residual, the cell tail is small by the separation of scales, the flux into the omitted words by the square of the Péclet number for each letter, and the switching as a pulse lasts many cell diffusion times; by (201) and Lemma 15.1(a), their size absorbs the loss of Lemma 15.13(b) in Lemma 15.14.
Lemma 15.7 (One weak passage). There exists C ( α , s , q , ζ , Λ , n , t 0 , T ) C(\alpha,s,q,\zeta,\Lambda,n,t_0,T) C ( α , s , q , ζ , Λ , n , t 0 , T ) such that, for every M ≥ M ∗ M\ge M_* M ≥ M ∗ , M < i ≤ M + n M<i\le M+n M < i ≤ M + n , θ i n \theta_{\mathrm{in}} θ in and θ \theta θ , the solutions of Proposition 15.5 satisfy
∥ U i − Corr i w Y i w U i − 1 ∥ L ∞ ( [ t ( i − 1 ) , t f ] ; H ε i , ω N i ( T 2 ) ) ≤ C ε M 3 + 2 q n ∥ U i ∥ L ∞ ( [ t ( i ) , t f ] ; L ω 2 ( T 2 ) ) . \|U_i-\operatorname{Corr}^{\mathrm{w}}_i\mathcal{Y}^{\mathrm{w}}_iU_{i-1}\|_{L^\infty([t_{(i-1)},t_f];H^{N_i}_{\varepsilon_i,\omega}(T^2))}\le C\varepsilon_M^{3+2q^n}\|U_i\|_{L^\infty([t_{(i)},t_f];L^2_\omega(T^2))}. ∥ U i − Corr i w Y i w U i − 1 ∥ L ∞ ([ t ( i − 1 ) , t f ] ; H ε i , ω N i ( T 2 )) ≤ C ε M 3 + 2 q n ∥ U i ∥ L ∞ ([ t ( i ) , t f ] ; L ω 2 ( T 2 )) .
Proof. Let k : = M + n − i k:=M+n-i k := M + n − i . For a component ν \nu ν of G i G_i G i , let P i ( ν ) P_i(\nu) P i ( ν ) be the largest number of sources on a path of G i G_i G i that starts at ν \nu ν , and N ~ ν : = N i + ( N c o r , w − 1 ) P i ( ν ) \widetilde{N}_\nu:=N_i+(N_{\mathrm{cor,w}}-1)P_i(\nu) N ν := N i + ( N cor , w − 1 ) P i ( ν ) , the orders of Lemma 15.6 for G i G_i G i , with N i N_i N i in place of N t r , w N_{\mathrm{tr,w}} N tr , w . Let R : = 4 ( N c o r , w + 1 + k N c o r , w ) + 18 + N c o r , w − 1 R:=4(N_{\mathrm{cor,w}}+1+kN_{\mathrm{cor,w}})+18+N_{\mathrm{cor,w}}-1 R := 4 ( N cor , w + 1 + k N cor , w ) + 18 + N cor , w − 1 , the order N r e c = 4 ( N c o r + N s r c ) + 2 N c l + 16 N_{\mathrm{rec}}=4(N_{\mathrm{cor}}+N_{\mathrm{src}})+2N_{\mathrm{cl}}+16 N rec = 4 ( N cor + N src ) + 2 N cl + 16 of the reconstruction at N c o r = N c o r , w N_{\mathrm{cor}}=N_{\mathrm{cor,w}} N cor = N cor , w , N s r c = 1 + k N c o r , w N_{\mathrm{src}}=1+kN_{\mathrm{cor,w}} N src = 1 + k N cor , w and N c l = 1 N_{\mathrm{cl}}=1 N cl = 1 , plus the d − 2 ≤ N c o r , w − 1 d-2\le N_{\mathrm{cor,w}}-1 d − 2 ≤ N cor , w − 1 of a frontier letter. As k ≤ n − 1 k\le n-1 k ≤ n − 1 , N s r c , e ≥ n N c o r , w + 6 N_{\mathrm{src},e}\ge nN_{\mathrm{cor,w}}+6 N src , e ≥ n N cor , w + 6 by (203) , and N r d , e ≥ N r e c , e + 2 N s r c , e N p a t h , e ≥ 8 N s r c , e N_{\mathrm{rd},e}\ge N_{\mathrm{rec},e}+2N_{\mathrm{src},e}N_{\mathrm{path},e}\ge8N_{\mathrm{src},e} N rd , e ≥ N rec , e + 2 N src , e N path , e ≥ 8 N src , e by (206) , and N t r , w ≥ 2 N r d , e + 2 N_{\mathrm{tr,w}}\ge2N_{\mathrm{rd},e}+2 N tr , w ≥ 2 N rd , e + 2 by (207) ,
N i + R = ( 5 k + 5 ) N c o r , w + 26 ≤ 5 n N c o r , w + 26 < 16 N s r c , e + 2 ≤ 2 N r d , e + 2 ≤ N t r , w . N_i+R=(5k+5)N_{\mathrm{cor,w}}+26\le5nN_{\mathrm{cor,w}}+26<16N_{\mathrm{src},e}+2\le2N_{\mathrm{rd},e}+2\le N_{\mathrm{tr,w}}. N i + R = ( 5 k + 5 ) N cor , w + 26 ≤ 5 n N cor , w + 26 < 16 N src , e + 2 ≤ 2 N rd , e + 2 ≤ N tr , w .
A component of U i − 1 U_{i-1} U i − 1 of order K ≥ N t r , w K\ge N_{\mathrm{tr,w}} K ≥ N tr , w in Lemma 15.6 is bounded in H K ( T 2 ) H^K(T^2) H K ( T 2 ) and, by its equation and the order condition of Proposition 8.6, Lipschitz in t t t with values in H K − 2 ( T 2 ) H^{K-2}(T^2) H K − 2 ( T 2 ) . So, by the Sobolev embedding, U i − 1 ( t ) U_{i-1}(t) U i − 1 ( t ) is of class C max { n N c o r , w , R + 1 } C^{\max\{nN_{\mathrm{cor,w}},R+1\}} C m a x { n N cor , w , R + 1 } , these derivatives being Lipschitz in t t t , uniformly in x x x : the classes that the weak lift and Lemmas 13.2 and 13.4–13.6 ask for; the corrector coefficients are absolutely continuous in t t t , with a bounded derivative at each mode.
Step 1: zero initial difference. At the block boundary t ( i − 1 ) t_{(i-1)} t ( i − 1 ) , between two quiet margins, the weak corrector operator is the identity: the correctors of the pulses whose cutoff is positive at t ( i − 1 ) t_{(i-1)} t ( i − 1 ) are solved from zero on windows within 121 48 τ i \frac{121}{48}\tau_i 48 121 τ i of t ( i − 1 ) t_{(i-1)} t ( i − 1 ) , which lie in these margins as 121 48 τ i + 1 24 τ i ≤ 5 τ i ≤ τ i ′ \frac{121}{48}\tau_i+\frac{1}{24}\tau_i\le5\tau_i\le\tau'_i 48 121 τ i + 24 1 τ i ≤ 5 τ i ≤ τ i ′ ; there the shear of level i i i vanishes and, by (a4), the forms of G i G_i G i do not depend on the fast variable, so the correctors of positive degree vanish (Proposition 10.6 (e), as in Lemma 13.1 ). Since Υ i w U i − 1 ( t ( i − 1 ) ) = U i ( t ( i − 1 ) ) \Upsilon_i^{\mathrm{w}}U_{i-1}(t_{(i-1)})=U_i(t_{(i-1)}) Υ i w U i − 1 ( t ( i − 1 ) ) = U i ( t ( i − 1 ) ) by Proposition 15.5 (iii), the difference vanishes at t ( i − 1 ) t_{(i-1)} t ( i − 1 ) .
Step 2: the residual. The computations of Lemmas 13.2 and 13.5 and of the readout identity of Lemma 13.8 use only the recursion of Definition 10.5 , the equations of the cell problems, and the masses and factors of the new components. As these lemmas are stated in the setting of the lift, we rerun their proofs, and those of Lemmas 13.4 –13.6 for the sizes, with ϵ m = ϵ ˉ m = ϵ i \epsilon_m=\bar{\epsilon}_m=\epsilon_i ϵ m = ϵ ˉ m = ϵ i , ρ ( m ) = ϵ i − 1 \rho(m)=\epsilon_{i-1} ρ ( m ) = ϵ i − 1 , κ m = κ m − 1 = κ \kappa_m=\kappa_{m-1}=\kappa κ m = κ m − 1 = κ , κ m h m = ϵ i − 1 2 \kappa_m h_m=\epsilon_{i-1}^2 κ m h m = ϵ i − 1 2 , N c o r = N c o r , w N_{\mathrm{cor}}=N_{\mathrm{cor,w}} N cor = N cor , w , N s r c = 1 + k N c o r , w N_{\mathrm{src}}=1+kN_{\mathrm{cor,w}} N src = 1 + k N cor , w , N c l = 1 N_{\mathrm{cl}}=1 N cl = 1 , with τ i \tau_i τ i in (166) and the bounds of the proof of Lemma 15.4 (a) for (160) . With the weak finite lift and Definition 15.3 they show that U i − Corr i w Υ i w U i − 1 U_i-\operatorname{Corr}^{\mathrm{w}}_i\Upsilon^{\mathrm{w}}_iU_{i-1} U i − Corr i w Υ i w U i − 1 solves the system of G i G_i G i on [ t ( i − 1 ) , t f ] [t_{(i-1)},t_f] [ t ( i − 1 ) , t f ] with a forcing in three parts:
the cell tail , the degrees above N c o r , w N_{\mathrm{cor,w}} N cor , w of the weak finite lift, written in divergence form as in Lemma 13.4 , with one antidivergence step at the wavelength ϵ i − 1 \epsilon_{i-1} ϵ i − 1 , of size ( ϵ i / ϵ i − 1 ) N c o r , w (\epsilon_i/\epsilon_{i-1})^{N_{\mathrm{cor,w}}} ( ϵ i / ϵ i − 1 ) N cor , w times the cleanup factor 1 + ( κ T ) 1 / 2 / ϵ i − 1 1+(\kappa T)^{1/2}/\epsilon_{i-1} 1 + ( κ T ) 1/2 / ϵ i − 1 ;
the frontier , the letters out of the words of length N l e n , w N_{\mathrm{len,w}} N len , w , whose targets are not components of G i − 1 G_{i-1} G i − 1 , of size ( 1 + ( ϵ i / ϵ i − 1 ) 2 ( 1 + ( κ T ) 1 / 2 / ϵ i − 1 ) ) ϵ M 2 ( q + 2 ) γ ( N l e n , w + 1 ) (1+(\epsilon_i/\epsilon_{i-1})^2(1+(\kappa T)^{1/2}/\epsilon_{i-1}))\epsilon_M^{2(q+2)\gamma(N_{\mathrm{len,w}}+1)} ( 1 + ( ϵ i / ϵ i − 1 ) 2 ( 1 + ( κ T ) 1/2 / ϵ i − 1 )) ϵ M 2 ( q + 2 ) γ ( N len , w + 1 ) , the first factor being the cost of a coarse flux under the weak lift (Lemma 13.6 with κ \kappa κ for both diffusivities); read out, each of its N l e n , w + 1 N_{\mathrm{len,w}}+1 N len , w + 1 letters contributes the relative size ( ϵ i p / κ ) 2 (\epsilon_i^p/\kappa)^2 ( ϵ i p / κ ) 2 of its eddy diffusivity;
the switching , the time derivatives of the cutoffs between consecutive pulses, where the correctors of the earlier pulse have decayed for a time at least 1 24 τ i \frac{1}{24}\tau_i 24 1 τ i , of size ϵ i − C exp ( − c κ τ i / ϵ i 2 ) \epsilon_i^{-C}\exp(-c\kappa\tau_i/\epsilon_i^2) ϵ i − C exp ( − c κ τ i / ϵ i 2 ) (Step 3 of the proof of Lemma 13.5 at the level i i i , with (a4) for G i G_i G i , and with Lemma 15.2 (b) on each equation of the triangular system, Duhamel’s formula and induction on the path length and the degree in place of the decay of the cell inverse on transverse modes).
The forcing is measured in the norm of the right side of Proposition 8.6 at the orders N ~ v \widetilde{N}_v N v and the scale ϵ i \epsilon_i ϵ i ; in it each part is at most its size times the energy of U i − 1 U_{i-1} U i − 1 at the orders of Lemma 15.6 and the scale ϵ i − 1 \epsilon_{i-1} ϵ i − 1 (for the frontier, layer by layer), which Proposition 8.6 and Lemma 15.6 bound by C ∥ U i ∥ L ∞ ( [ t ( i ) , t ( i − 1 ) ] ; L ω 2 ( T 2 ) ) C\|U_i\|_{L^\infty([t_{(i)},t_{(i-1)}];L^2_\omega(\mathbb{T}^2))} C ∥ U i ∥ L ∞ ([ t ( i ) , t ( i − 1 ) ] ; L ω 2 ( T 2 )) . Indeed, each term of the three parts is a two-scale form read at ( y , y / ϵ i ) (y,y/\epsilon_i) ( y , y / ϵ i ) and carried to x x x by the chart, times a derivative of the field or of a flux, and a derivative at the scale ϵ i \epsilon_i ϵ i costs a constant on the two-scale form (Proposition 10.7 , with (108) from (b1) by Cauchy’s estimate), at most C ϵ i / ϵ i − 1 C\epsilon_i/\epsilon_{i-1} C ϵ i / ϵ i − 1 on a slow coefficient or a chart (Lemmas 15.4 (a) and 7.7 ), and, on the field, ϵ i / ϵ i − 1 \epsilon_i/\epsilon_{i-1} ϵ i / ϵ i − 1 times one derivative at the scale ϵ i − 1 \epsilon_{i-1} ϵ i − 1 . So only the order at which U i − 1 U_{i-1} U i − 1 is read rises: the forcing in v v v at the order N ~ v \widetilde{N}_v N v reads it at the order N ~ v + R \widetilde{N}_v+R N v + R at the components ( v ′ , w ) (v',w) ( v ′ , w ) , for every word w w w and every v ′ v' v ′ with a path of J ≥ 0 J\ge0 J ≥ 0 sources of G i G_i G i into v v v ; as this path, followed by a longest path from v v v , lifts with the word w w w to inherited sources (Definition 15.3 (ii)), N ( v ′ , w ) ≥ N t r , w + ( n N c o r , w − 1 ) P i ( v ) ≥ N ~ v + R N_{(v',w)}\ge N_{\mathrm{tr},w}+(nN_{\mathrm{cor,w}}-1)P_i(v)\ge\widetilde{N}_v+R N ( v ′ , w ) ≥ N tr , w + ( n N cor , w − 1 ) P i ( v ) ≥ N v + R .
For the frontier we argue as for the omitted length in the proof of Lemma 13.9 , layer by layer in the length l l l of the words. The layer of length l l l receives sources only from itself and, by letters, from the layer of length l − 1 l-1 l − 1 ; its matrix in Proposition 8.6 at the orders N ( v , w ) N_{(v,w)} N ( v , w ) and the scale ϵ i − 1 \epsilon_{i-1} ϵ i − 1 has sums at most 1 8 \frac{1}{8} 8 1 (proof of Proposition 15.5 ), and its datum vanishes for l ≥ 1 l\ge1 l ≥ 1 . Let D l D_l D l be κ 1 / 2 \kappa^{1/2} κ 1/2 times the norm of ∇ U i − 1 \nabla U_{i-1} ∇ U i − 1 on it in L 2 ( ( t ( i − 1 ) , t f ) ; H ϵ i − 1 , ω 1 ( T 2 ) ) L^2((t_{(i-1)},t_f);H^1_{\epsilon_{i-1},\omega}(\mathbb{T}^2)) L 2 (( t ( i − 1 ) , t f ) ; H ϵ i − 1 , ω 1 ( T 2 )) . A letter ( v , w ) → ( v , w d ) (v,w)\to(v,wd) ( v , w ) → ( v , w d ) has coefficients at most C ϵ i − 6 / 16 ( ϵ i p / κ ) 2 κ C\epsilon_i^{-6/16}(\epsilon_i^p/\kappa)^2\kappa C ϵ i − 6/16 ( ϵ i p / κ ) 2 κ times the powers of Lemma 15.4 (a), and N ( v , w ) ≥ N ( v , w d ) + n N c o r , w − 1 ≥ N ( v , w d ) + d − 2 N_{(v,w)}\ge N_{(v,wd)}+nN_{\mathrm{cor,w}}-1\ge N_{(v,wd)}+d-2 N ( v , w ) ≥ N ( v , w d ) + n N cor , w − 1 ≥ N ( v , w d ) + d − 2 , as a path from ( v , w d ) (v,wd) ( v , w d ) extends by this letter. With Lemma 8.1, the square root ( 4 n N c o r , w ) 1 / 2 ϵ i − 1 − 6 / 64 (4nN_{\mathrm{cor,w}})^{1/2}\epsilon_{i-1}^{-6/64} ( 4 n N cor , w ) 1/2 ϵ i − 1 − 6/64 of the mass ratio and Schur’s test (a child has one parent, a parent N c o r , w N_{\mathrm{cor,w}} N cor , w children), Proposition 8.6 gives D l ≤ C ϵ i − 1 − 56 / 64 ( ϵ i p / κ ) 2 D l − 1 D_l\le C\epsilon_{i-1}^{-56/64}(\epsilon_i^p/\kappa)^2D_{l-1} D l ≤ C ϵ i − 1 − 56/64 ( ϵ i p / κ ) 2 D l − 1 for l ≥ 1 l\ge1 l ≥ 1 , and bounds D 0 D_0 D 0 by the datum, which Lemma 15.6 bounds. The frontier flux F f r F^{\mathrm{fr}} F fr carries the readout weight ϵ i − 1 δ N l e n , w / 16 \epsilon_{i-1}^{\delta N_{\mathrm{len,w}}/16} ϵ i − 1 δ N len , w /16 , the square root ( 4 n N c o r , w ) − N l e n , w / 2 ε i − 1 δ N l e n , w / 64 (4nN_{\mathrm{cor},w})^{-N_{\mathrm{len},w}/2}\varepsilon_{i-1}^{\delta N_{\mathrm{len},w}/64} ( 4 n N cor , w ) − N len , w /2 ε i − 1 δ N len , w /64 of the mass ratio and the eddy coefficient κ ( ε i β / κ ) 2 \kappa(\varepsilon_i^\beta/\kappa)^2 κ ( ε i β / κ ) 2 ; so, in the norm of the forcing and before the coarse-flux cost, the powers of ε i − 1 \varepsilon_{i-1} ε i − 1 cancel and, by Lemma 15.1(a),
κ − 1 / 2 ∥ F f r ∥ ≤ C ε i − 1 5 δ N l e n , w / 64 ( ε i β κ ) 2 D N l e n , w ≤ C ( ε i β κ ) 2 ( N l e n , w + 1 ) D 0 ≤ C ε M 2 ( q + 2 ) γ ( N l e n , w + 1 ) D 0 . \kappa^{-1/2}\lVert F^{\mathrm{fr}}\rVert\le C\varepsilon_{i-1}^{5\delta N_{\mathrm{len},w}/64}\left(\frac{\varepsilon_i^\beta}{\kappa}\right)^2D_{N_{\mathrm{len},w}}\le C\left(\frac{\varepsilon_i^\beta}{\kappa}\right)^{2(N_{\mathrm{len},w}+1)}D_0\le C\varepsilon_M^{2(q+2)\gamma(N_{\mathrm{len},w}+1)}D_0. κ − 1/2 ∥ F fr ∥ ≤ C ε i − 1 5 δ N len , w /64 ( κ ε i β ) 2 D N len , w ≤ C ( κ ε i β ) 2 ( N len , w + 1 ) D 0 ≤ C ε M 2 ( q + 2 ) γ ( N len , w + 1 ) D 0 .
Step 3: the sizes. We use ε i ≤ 2 ε i − 1 a \varepsilon_i\le2\varepsilon_{i-1}^a ε i ≤ 2 ε i − 1 a and c ε M q n − 1 ≤ ε i − 1 ≤ ε M c\varepsilon_M^{q^n-1}\le\varepsilon_{i-1}\le\varepsilon_M c ε M q n − 1 ≤ ε i − 1 ≤ ε M (Lemma 7.1 (a)), and ( κ T ) 1 / 2 / ε i − 1 ≤ C ε i − 1 − 1 (\kappa T)^{1/2}/\varepsilon_{i-1}\le C\varepsilon_{i-1}^{-1} ( κ T ) 1/2 / ε i − 1 ≤ C ε i − 1 − 1 . The cell tail is at most C ε i − 1 ( q − 1 ) N c o r , w − 1 ≤ C ε M 4 + 2 q n C\varepsilon_{i-1}^{(q-1)N_{\mathrm{cor},w}-1}\le C\varepsilon_M^{4+2q^n} C ε i − 1 ( q − 1 ) N cor , w − 1 ≤ C ε M 4 + 2 q n by the first condition of (201) . The frontier is at most C ε i − 1 − 1 ε M 2 ( q + 2 ) γ ( N l e n , w + 1 ) ≤ C ε M 4 + 3 q n − q n − 1 ≤ C ε M 4 + 2 q n C\varepsilon_{i-1}^{-1}\varepsilon_M^{2(q+2)\gamma(N_{\mathrm{len},w}+1)}\le C\varepsilon_M^{4+3q^n-q^{n-1}}\le C\varepsilon_M^{4+2q^n} C ε i − 1 − 1 ε M 2 ( q + 2 ) γ ( N len , w + 1 ) ≤ C ε M 4 + 3 q n − q n − 1 ≤ C ε M 4 + 2 q n by the second. The switching is at most C ε i − 1 − C exp ( − c ε i − 1 − ( 2 ( q − 1 ) + γ − 4 δ ) ) ≤ C ε M 4 + 2 q n C\varepsilon_{i-1}^{-C}\exp(-c\varepsilon_{i-1}^{-(2(q-1)+\gamma-4\delta)})\le C\varepsilon_M^{4+2q^n} C ε i − 1 − C exp ( − c ε i − 1 − ( 2 ( q − 1 ) + γ − 4 δ ) ) ≤ C ε M 4 + 2 q n by Lemma 15.1(a), since 2 ( q − 1 ) − 4 δ > 0 2(q-1)-4\delta>0 2 ( q − 1 ) − 4 δ > 0 .
Step 4: energy. Proposition 8.6 for G i G_i G i at the scale ε i \varepsilon_i ε i and the orders N ~ v \widetilde N_v N v , which decrease along the sources of G i G_i G i as it asks ( d σ − 2 ≤ k N c o r , w − 1 ≤ n N c o r , w − 1 ) (d_\sigma-2\le kN_{\mathrm{cor},w}-1\le nN_{\mathrm{cor},w}-1) ( d σ − 2 ≤ k N cor , w − 1 ≤ n N cor , w − 1 ) , bounds the energy of the difference by C C C times its forcing; its hypotheses hold by Lemmas 15.1(a) and 15.4 , and its uniqueness identifies the difference with the solution from zero, as the difference lies in its class: the lift by the count of Step 2, and U i U_i U i by the proof of Proposition 15.5 , as N t r , w ≥ N i N_{\mathrm{tr},w}\ge N_i N tr , w ≥ N i (for i = M + n i=M+n i = M + n , θ \theta θ is smooth). With Steps 1 and 3 this gives the lemma, as N ~ v ≥ N i \widetilde N_v\ge N_i N v ≥ N i , with one power of ε M \varepsilon_M ε M to spare. □
Level M M M : smoothing and a homogenization cycle
The smoothing. The weak families have unsmoothed sources: Proposition 8.6 solves their systems at orders that decrease along the paths of sources, and the derivatives are regained on each quiet margin. The terminal cycle uses § 5.6 and § 5.8 , which need the clauses of the invariant, and clause (c) asks for paired smoothings with vanishing moments and widths below ρ m o l ( M ) \rho_{\mathrm{mol}}(M) ρ mol ( M ) and ρ ( M ) \rho(M) ρ ( M ) ; so the sources are smoothed once, at level M M M .
Let S w : = S ℓ w , r w , 2 − 1 / 2 I S ℓ w , r w , 2 − 1 / 2 I S_w:=S_{\ell_w,r_w,2^{-1/2}I}S_{\ell_w,r_w,2^{-1/2}I} S w := S ℓ w , r w , 2 − 1/2 I S ℓ w , r w , 2 − 1/2 I , the product of two averages ((44) ) over round discs of radius 2 − 1 / 2 r w 2^{-1/2}r_w 2 − 1/2 r w , with 2 ℓ w − 1 ≥ max { N m o m , w , N m o m } 2\ell_w-1\ge\max\{N_{\mathrm{mom},w},N_{\mathrm{mom}}\} 2 ℓ w − 1 ≥ max { N mom , w , N mom } . Its matrix of long axes, 2 − 1 / 2 I 2^{-1/2}I 2 − 1/2 I , has the inverse 2 1 / 2 I 2^{1/2}I 2 1/2 I , of ∣ ⋅ ∣ ∞ \lvert\cdot\rvert_\infty ∣ ⋅ ∣ ∞ -norm 2 1 / 2 ≤ 2 2^{1/2}\le2 2 1/2 ≤ 2 , so Proposition 6.1 applies. Let G ~ M \widetilde G_M G M be the family G M G_M G M with every source smoothed by S w S_w S w , and let U ~ M \widetilde U_M U M be the solution of G ~ M \widetilde G_M G M from U M ( t ( M ) ) U_M(t_{(M)}) U M ( t ( M ) ) at time t ( M ) t_{(M)} t ( M ) . By Lemma 15.8, U ~ M \widetilde U_M U M keeps the derivatives of U M U_M U M at the scale ε M \varepsilon_M ε M , and, at a lower order, differs from it by at most a constant times the gain of the moments of S w S_w S w times the small coupling of Lemma 15.4 (b).
Lemma 15.8 (The smoothing at level M M M ). *There exists C ( α , s , q , ζ , Λ , n , t 0 , T ) C(\alpha,s,q,\zeta,\Lambda,n,t_0,T) C ( α , s , q , ζ , Λ , n , t 0 , T ) such that, for every M ≥ M ∗ M\ge M_* M ≥ M ∗ , θ i n \theta_{\mathrm{in}} θ in and θ \theta θ , the component energy ((155) ) of U ~ M \widetilde U_M U M of order N t r , w N_{\mathrm{tr},w} N tr , w at the scale ε M \varepsilon_M ε M , with the diffusivity κ \kappa κ , on [ t ( M ) , t f ] [t_{(M)},t_f] [ t ( M ) , t f ] is at most C ∥ θ i n ∥ L 2 ( T 2 ) C\lVert\theta_{\mathrm{in}}\rVert_{L^2(\mathbb T^2)} C ∥ θ in ∥ L 2 ( T 2 ) , and, with N : = N t r , w − N m o m , w − n N c o r , w − 1 N:=N_{\mathrm{tr},w}-N_{\mathrm{mom},w}-nN_{\mathrm{cor},w}-1 N := N tr , w − N mom , w − n N cor , w − 1 ,
∥ U M − U ~ M ∥ L ∞ ( [ t ( M ) , t f ] ; H ε M , ω N ( T 2 ) ) ≤ C ε M q − 1 + ( γ / 4 + δ / 2 ) N m o m , w ∥ θ i n ∥ L 2 ( T 2 ) . \lVert U_M-\widetilde U_M\rVert_{L^\infty([t_{(M)},t_f];H_{\varepsilon_M,\omega}^{N}(\mathbb T^2))}\le C\varepsilon_M^{q-1+(\gamma/4+\delta/2)N_{\mathrm{mom},w}}\lVert\theta_{\mathrm{in}}\rVert_{L^2(\mathbb T^2)}. ∥ U M − U M ∥ L ∞ ([ t ( M ) , t f ] ; H ε M , ω N ( T 2 )) ≤ C ε M q − 1 + ( γ /4 + δ /2 ) N mom , w ∥ θ in ∥ L 2 ( T 2 ) .
Proof. Proposition 8.3 applies to G ~ M \widetilde G_M G M at every order up to N t r , w N_{\mathrm{tr},w} N tr , w and the scale ε M \varepsilon_M ε M : its condition (i) holds by the bound on ∇ b M \nabla b_M ∇ b M stated after (12.1) and ε M β / κ ≤ C ε M γ \varepsilon_M^\beta/\kappa\le C\varepsilon_M^\gamma ε M β / κ ≤ C ε M γ (Lemma 15.1(b)), and (ii) and (iii) by Lemma 15.4 (a) at i = M + 1 i=M+1 i = M + 1 and the order N t r , w N_{\mathrm{tr},w} N tr , w , with the sums of the proof of Lemma 15.4 (b). The solution U ~ M \widetilde U_M U M has no forcing, and its datum U M ( t ( M ) ) U_M(t_{(M)}) U M ( t ( M ) ) has norm at most C ∥ θ i n ∥ L 2 ( T 2 ) C\lVert\theta_{\mathrm{in}}\rVert_{L^2(\mathbb T^2)} C ∥ θ in ∥ L 2 ( T 2 ) in H ε M N t r , w ( T 2 ) H_{\varepsilon_M}^{N_{\mathrm{tr},w}}(\mathbb T^2) H ε M N tr , w ( T 2 ) by Proposition 15.5 (iv); so part (c) gives the first bound.
The difference U M − U ~ M U_M-\widetilde U_M U M − U M vanishes at t ( M ) t_{(M)} t ( M ) and solves the system of G ~ M \widetilde G_M G M with the forcing ∇ ⋅ ∑ σ ∑ d A σ , d ∇ d − 2 ( I − S w ) ∇ U M , v ′ \nabla\cdot\sum_\sigma\sum_d A_{\sigma,d}\nabla^{d-2}(I-S_w)\nabla U_{M,v'} ∇ ⋅ ∑ σ ∑ d A σ , d ∇ d − 2 ( I − S w ) ∇ U M , v ′ , the sums over the sources σ : v ′ → v \sigma:v'\to v σ : v ′ → v . By the third bound of Proposition 6.1 at the scale ε M \varepsilon_M ε M and the order N m o m , w < 2 ℓ w N_{\mathrm{mom},w}<2\ell_w N mom , w < 2 ℓ w , I − S w I-S_w I − S w costs C ( r w / ε M ) N m o m , w = C ε M ( γ / 4 + δ / 2 ) N m o m , w C(r_w/\varepsilon_M)^{N_{\mathrm{mom},w}}=C\varepsilon_M^{(\gamma/4+\delta/2)N_{\mathrm{mom},w}} C ( r w / ε M ) N mom , w = C ε M ( γ /4 + δ /2 ) N mom , w and N m o m , w N_{\mathrm{mom},w} N mom , w more derivatives at the scale ε M \varepsilon_M ε M , and d − 2 ≤ n N c o r , w − 1 d-2\le nN_{\mathrm{cor},w}-1 d − 2 ≤ n N cor , w − 1 , as the birth orders of G M G_M G M are at most 1 + n N c o r , w 1+nN_{\mathrm{cor},w} 1 + n N cor , w . By Lemma 15.4 , κ − 1 ε M − 2 − d \kappa^{-1}\varepsilon_M^{-2-d} κ − 1 ε M − 2 − d times the coefficients, with the mass ratios, has row and column sums at most C ε M q − 1 C\varepsilon_M^{q-1} C ε M q − 1 . So the forcing, measured as in (156) at the order N N N , is at most C ε M q − 1 + ( γ / 4 + δ / 2 ) N m o m , w C\varepsilon_M^{q-1+(\gamma/4+\delta/2)N_{\mathrm{mom,w}}} C ε M q − 1 + ( γ /4 + δ /2 ) N mom , w times κ 1 / 2 ∥ ∇ U M ∥ L 2 ( ( t ( M ) , t f ) ; H ε M , ω N t r , w ( T 2 ) ) \kappa^{1/2}\lVert\nabla U_M\rVert_{L^2((t_{(M)},t_f);H^{N_{\mathrm{tr,w}}}_{\varepsilon_M,\omega}(\mathbb{T}^2))} κ 1/2 ∥ ∇ U M ∥ L 2 (( t ( M ) , t f ) ; H ε M , ω N tr , w ( T 2 )) , which Proposition 15.5(iv) bounds by C ∥ θ i n ∥ L 2 ( T 2 ) C\lVert\theta_{\mathrm{in}}\rVert_{L^2(\mathbb{T}^2)} C ∥ θ in ∥ L 2 ( T 2 ) . Proposition 8.3(a) at the order N N N then gives the second bound.
The terminal successor of Definition 15.9 is the successor of Section 13 at the level M M M , with four changes: the orders; one generation of children, with the marker 1, so that it can be declared a seed without change (Proposition 15.11); the coefficients, majorant and smoothing of the new sources; and the covariance, anchored at the preparation time so that the retained trace there is the identity on the components of G ~ M \widetilde G_M G M . The preparation time depends on t ( M ) t_{(M)} t ( M ) alone: with l ∈ Z l\in\mathbb{Z} l ∈ Z such that ( l − 1 2 ) h M ≤ t ( M ) < ( l + 1 2 ) h M (l-\frac{1}{2})h_M\leq t_{(M)}<(l+\frac{1}{2})h_M ( l − 2 1 ) h M ≤ t ( M ) < ( l + 2 1 ) h M , it is t p r ( M ) : = ( l + 3 2 ) h M t_{\mathrm{pr}}^{(M)}:=(l+\frac{3}{2})h_M t pr ( M ) := ( l + 2 3 ) h M , the time (138) at m = M m=M m = M with t 1 = t ( M ) t_1=t_{(M)} t 1 = t ( M ) .
Definition 15.9 (The terminal successor). The terminal successor G ′ G' G ′ of G ~ M \widetilde G_M G M is the family of level M − 1 M-1 M − 1 that the comparison of Section 13 constructs at m = M m=M m = M (Definition 11.3), with these changes.
(i) The orders are N c o r , e N_{\mathrm{cor,e}} N cor , e and p e p_e p e ; the mean and covariance letters run over the degrees up to N c o r , e + 1 + n N c o r , w N_{\mathrm{cor,e}}+1+nN_{\mathrm{cor,w}} N cor , e + 1 + n N cor , w and the paths of sources of G ~ M \widetilde G_M G M , so that the birth orders are at most N s r c , e N_{\mathrm{src,e}} N src , e .
(ii) One generation: the components of G ~ M \widetilde G_M G M , with their masses, and their children v ′ v' v ′ , with nonempty words of length # v ′ ≤ N l e n , e \#v'\leq N_{\mathrm{len,e}} # v ′ ≤ N len , e , each with the mass c e ε M − 1 δ / 32 ω v c_e\varepsilon_{M-1}^{\delta/32}\omega_v c e ε M − 1 δ /32 ω v divided by the number of these words, for a constant c e ( α , s , q , ζ , Λ , n , t 0 , T ) > 0 c_e(\alpha,s,q,\zeta,\Lambda,n,t_0,T)>0 c e ( α , s , q , ζ , Λ , n , t 0 , T ) > 0 , the readout weight ε M − 1 δ # v ′ / 16 \varepsilon_{M-1}^{\delta\#v'/16} ε M − 1 δ # v ′ /16 in (38) , and the marker 1.
(iii) The coefficients of the new sources are multiplied by ε M − 1 − δ / 16 \varepsilon_{M-1}^{-\delta/16} ε M − 1 − δ /16 . The new sources have the birth radius ρ ( M ) \rho(M) ρ ( M ) , the majorant C m a j ε M − 1 δ − δ / 16 C_{\mathrm{maj}}\varepsilon_{M-1}^{\delta-\delta/16} C maj ε M − 1 δ − δ /16 times the product of the majorants along the path, and the round smoothing of S w S_w S w , with the width ρ ( M ) ε M − 1 θ e \rho(M)\varepsilon_{M-1}^{\theta_e} ρ ( M ) ε M − 1 θ e in place of r w r_w r w and with the kernel order ℓ e \ell_e ℓ e in place of ℓ w \ell_w ℓ w , where 2 ℓ e − 1 ≥ max { N m o m , e , N m o m } 2\ell_e-1\geq\max\{N_{\mathrm{mom,e}},N_{\mathrm{mom}}\} 2 ℓ e − 1 ≥ max { N mom , e , N mom } .
(iv) The covariance is anchored at the preparation time t p r ( M ) t_{\mathrm{pr}}^{(M)} t pr ( M ) : Γ M ( t ) − Γ M ( t p r ( M ) ) \Gamma_M(t)-\Gamma_M(t_{\mathrm{pr}}^{(M)}) Γ M ( t ) − Γ M ( t pr ( M ) ) replaces Γ M ( t ) \Gamma_M(t) Γ M ( t ) in Definition 11.3, in (172) (the defect and covariance letters) and in the lift.
The terminal lift of a field Y Y Y of G ′ G' G ′ is
Y e ( t ) : = Corr e ( t ) [ exp p e ( ( Γ M ( t ) − Γ M ( t p r ( M ) ) ) : ∇ 2 ) e Y ( t ) ] , (208) Y^{\mathrm{e}}(t):=\operatorname{Corr}^{\mathrm{e}}(t)\left[\exp_{p_e}\left(\left(\Gamma_M(t)-\Gamma_M(t_{\mathrm{pr}}^{(M)})\right):\nabla^2\right)^{\mathrm{e}}Y(t)\right], \tag*{(208)} Y e ( t ) := Corr e ( t ) [ exp p e ( ( Γ M ( t ) − Γ M ( t pr ( M ) ) ) : ∇ 2 ) e Y ( t ) ] , (208) where Υ e \Upsilon^{\mathrm{e}} Υ e is the readout (38) of G ′ G' G ′ , exp p e \exp_{p_e} exp p e is the truncated exponential (39) with p e p_e p e in place of p p p , and Corr e \operatorname{Corr}^{\mathrm{e}} Corr e is the corrector operator Corr M \operatorname{Corr}_M Corr M of Definition 10.8 for G ~ M \widetilde G_M G M with the corrector order N c o r , e N_{\mathrm{cor,e}} N cor , e .
The anchored covariance is small: by Lemmas 10.3(b) and 15.1(b),
∣ Γ M ( t ) − Γ M ( t p r ( M ) ) ∣ ≤ C κ M − 1 τ M ′ ≤ C ε M − 1 δ ρ ( M ) 2 ≤ C ε M − 1 γ + 3 δ ε M − 1 2 . \lvert\Gamma_M(t)-\Gamma_M(t_{\mathrm{pr}}^{(M)})\rvert\leq C\kappa_{M-1}\tau'_M\leq C\varepsilon_{M-1}^{\delta}\rho(M)^2\leq C\varepsilon_{M-1}^{\gamma+3\delta}\varepsilon_{M-1}^2. ∣ Γ M ( t ) − Γ M ( t pr ( M ) )∣ ≤ C κ M − 1 τ M ′ ≤ C ε M − 1 δ ρ ( M ) 2 ≤ C ε M − 1 γ + 3 δ ε M − 1 2 . Proposition 15.10 is Proposition 5.3 at the terminal level, with a preparation for G ~ M \widetilde G_M G M , which is not admissible but satisfies the clauses of the invariant (Lemma 15.4(b)); the contrast κ M − 1 / κ \kappa_{M-1}/\kappa κ M − 1 / κ is at most 2, and the datum already has derivatives at the scale ρ m o l ( M ) = c m o l ε M \rho_{\mathrm{mol}}(M)=c_{\mathrm{mol}}\varepsilon_M ρ mol ( M ) = c mol ε M . The last bound of (a), by interpolation, and the comparison (c) carry N r d , e N_{\mathrm{rd,e}} N rd , e derivatives at the scale ε M \varepsilon_M ε M at half the accuracy of the L 2 L^2 L 2 bound; the weak lifts read (c) in Lemma 15.14. Parts (b) and (c) use sums of G ′ G' G ′ bounded in the proof of Proposition 15.11, which does not use (b) or (c).
Proposition 15.10 (The terminal cycle). There exist C ( α , s , q , ζ , Λ , n , t 0 , T ) C(\alpha,s,q,\zeta,\Lambda,n,t_0,T) C ( α , s , q , ζ , Λ , n , t 0 , T ) and c ( α , s , q , ζ , Λ , n , t 0 , T ) > 0 c(\alpha,s,q,\zeta,\Lambda,n,t_0,T)>0 c ( α , s , q , ζ , Λ , n , t 0 , T ) > 0 such that, for every M ≥ M ∗ M\geq M_* M ≥ M ∗ , there exist t p r ( M ) t_{\mathrm{pr}}^{(M)} t pr ( M ) and a linear map Π : H N t r , w ( T 2 ) V → H N o u t , e ( T 2 ) V \Pi:H^{N_{\mathrm{tr,w}}}(\mathbb{T}^2)^V\to H^{N_{\mathrm{out,e}}}(\mathbb{T}^2)^V Π : H N tr , w ( T 2 ) V → H N out , e ( T 2 ) V , V V V the set of components of G ~ M \widetilde G_M G M , both depending only on M M M , κ \kappa κ and t i n t_{\mathrm{in}} t in , with the
following properties. For every θ i n \theta_{\mathrm{in}} θ in and θ \theta θ , let Y Y Y be the solution of G ′ G^{\prime} G ′ on [ t p r ( M ) , t f ] [t_{\mathrm{pr}}^{(M)},t_f] [ t pr ( M ) , t f ] from the datum equal to Π U ~ M ( t ( M ) ) \Pi\widetilde{U}_M(t_{(M)}) Π U M ( t ( M ) ) on the components of G ~ M \widetilde{G}_M G M and to 0 0 0 on their children.
(a) t ( M ) + c h M ≤ t p r ( M ) ≤ t ( M ) + C h M t_{(M)}+ch_M\leq t_{\mathrm{pr}}^{(M)}\leq t_{(M)}+Ch_M t ( M ) + c h M ≤ t pr ( M ) ≤ t ( M ) + C h M , b M = b M − 1 \mathbf{b}_M=\mathbf{b}_{M-1} b M = b M − 1 on [ t p r ( M ) − τ M ′ , t p r ( M ) + τ M ′ ] [t_{\mathrm{pr}}^{(M)}-\tau'_M,t_{\mathrm{pr}}^{(M)}+\tau'_M] [ t pr ( M ) − τ M ′ , t pr ( M ) + τ M ′ ] , and
∥ Π U ~ M ( t ( M ) ) ∥ H ρ ( M ) , ω N o u t , e ( T 2 ) ≤ C ∥ θ i n ∥ L 2 ( T 2 ) , \left\|\Pi\widetilde{U}_M(t_{(M)})\right\|_{H_{\rho(M),\omega}^{N_{\mathrm{out},e}}(\mathbb{T}^2)}\leq C\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)}, Π U M ( t ( M ) ) H ρ ( M ) , ω N out , e ( T 2 ) ≤ C ∥ θ in ∥ L 2 ( T 2 ) ,
∥ U ~ M ( t p r ( M ) ) − Π U ~ M ( t ( M ) ) ∥ L ω 2 ( T 2 ) ≤ C ε M − 1 Q e ∥ θ i n ∥ L 2 ( T 2 ) , \left\|\widetilde{U}_M(t_{\mathrm{pr}}^{(M)})-\Pi\widetilde{U}_M(t_{(M)})\right\|_{L_\omega^2(\mathbb{T}^2)}\leq C\varepsilon_{M-1}^{Q_e}\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)}, U M ( t pr ( M ) ) − Π U M ( t ( M ) ) L ω 2 ( T 2 ) ≤ C ε M − 1 Q e ∥ θ in ∥ L 2 ( T 2 ) ,
∥ U ~ M ( t p r ( M ) ) − Π U ~ M ( t ( M ) ) ∥ H ε M , ω N r d , e ( T 2 ) ≤ C ε M − 1 Q e / 2 ∥ θ i n ∥ L 2 ( T 2 ) . \left\|\widetilde{U}_M(t_{\mathrm{pr}}^{(M)})-\Pi\widetilde{U}_M(t_{(M)})\right\|_{H_{\varepsilon_M,\omega}^{N_{\mathrm{rd},e}}(\mathbb{T}^2)}\leq C\varepsilon_{M-1}^{Q_e/2}\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)}. U M ( t pr ( M ) ) − Π U M ( t ( M ) ) H ε M , ω N rd , e ( T 2 ) ≤ C ε M − 1 Q e /2 ∥ θ in ∥ L 2 ( T 2 ) .
(b) The component energy (155) of Y Y Y of order N o u t , e N_{\mathrm{out},e} N out , e at the scale ρ ( M ) \rho(M) ρ ( M ) , with the diffusivity κ M − 1 \kappa_{M-1} κ M − 1 , on [ t p r ( M ) , t f ] [t_{\mathrm{pr}}^{(M)},t_f] [ t pr ( M ) , t f ] is at most C ∥ θ i n ∥ L 2 ( T 2 ) C\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)} C ∥ θ in ∥ L 2 ( T 2 ) .
(c) For every t ∈ [ t p r ( M ) , t f ] t\in[t_{\mathrm{pr}}^{(M)},t_f] t ∈ [ t pr ( M ) , t f ] , Y ♯ ( t ) ∈ H N r d , e ( T 2 ) V Y^\sharp(t)\in H^{N_{\mathrm{rd},e}}(\mathbb{T}^2)^V Y ♯ ( t ) ∈ H N rd , e ( T 2 ) V , and
∥ U ~ M − Y ♯ ∥ L ∞ ( [ t p r ( M ) , t f ] ; H ε M , ω N r d , e ( T 2 ) ) ≤ C ε M − 1 Q e / 2 ∥ θ i n ∥ L 2 ( T 2 ) . \left\|\widetilde{U}_M-Y^\sharp\right\|_{L^\infty([t_{\mathrm{pr}}^{(M)},t_f];H_{\varepsilon_M,\omega}^{N_{\mathrm{rd},e}}(\mathbb{T}^2))}\leq C\varepsilon_{M-1}^{Q_e/2}\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)}. U M − Y ♯ L ∞ ([ t pr ( M ) , t f ] ; H ε M , ω N rd , e ( T 2 )) ≤ C ε M − 1 Q e /2 ∥ θ in ∥ L 2 ( T 2 ) .
Proof. (a) The preparation of § 5.6 runs at the terminal level for G ~ M \widetilde{G}_M G M from t 1 = t ( M ) t_1=t_{(M)} t 1 = t ( M ) , the start of the interval [ t ( M ) , t f ] [t_{(M)},t_f] [ t ( M ) , t f ] on which G ~ M \widetilde{G}_M G M satisfies the invariant (Lemma 15.4 (b)), so that t A = t ( M ) t_A=t_{(M)} t A = t ( M ) and, for M ≥ M ∗ M\geq M_* M ≥ M ∗ , t 1 + 2 h M ≤ t i n + 6 n h M + 1 + 2 h M ≤ t f t_1+2h_M\leq t_{\mathrm{in}}+6nh_{M+1}+2h_M\leq t_f t 1 + 2 h M ≤ t in + 6 n h M + 1 + 2 h M ≤ t f , as Proposition 12.7 asks. It uses that proposition at m = M m=M m = M for G ~ M \widetilde{G}_M G M , with N o u t , e N_{\mathrm{out},e} N out , e in place of N o u t N_{\mathrm{out}} N out and the constants of Lemma 15.4 (b), and Propositions 12.4 –12.6 and Lemmas 12.8–12.15 , which hold at m = M m=M m = M (Remark 12.16), where the contrast κ M − 1 / κ \kappa_{M-1}/\kappa κ M − 1 / κ is at most 2 2 2 (Lemma 15.1(b)), so that their powers of the contrast factor are constants. Assembled as in Subsection 12.10, with the bound ∥ U ~ M ( t ( M ) ) ∥ H ρ m o l ( M ) , ω N b u f ( T 2 ) ≤ C ∥ θ i n ∥ L 2 ( T 2 ) \|\widetilde{U}_M(t_{(M)})\|_{H_{\rho_{\mathrm{mol}}(M),\omega}^{N_{\mathrm{buf}}}(\mathbb{T}^2)}\leq C\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)} ∥ U M ( t ( M ) ) ∥ H ρ mol ( M ) , ω N buf ( T 2 ) ≤ C ∥ θ in ∥ L 2 ( T 2 ) of Proposition 15.5 (iv) (U ~ M ( t ( M ) ) = U M ( t ( M ) ) \widetilde{U}_M(t_{(M)})=U_M(t_{(M)}) U M ( t ( M ) ) = U M ( t ( M ) ) , ρ m o l ( M ) = c m o l ε M \rho_{\mathrm{mol}}(M)=c_{\mathrm{mol}}\varepsilon_M ρ mol ( M ) = c mol ε M , N b u f ≤ N t r , w N_{\mathrm{buf}}\leq N_{\mathrm{tr},w} N buf ≤ N tr , w ) in place of the molecular smoothing, they give Π \Pi Π , depending only on G ~ M \widetilde{G}_M G M and t ( M ) t_{(M)} t ( M ) , the first two bounds, the second as Q ≥ 4 + 2 q Z > Q e Q\geq4+2qZ>Q_e Q ≥ 4 + 2 q Z > Q e by (179) , (180) , (204) and (202) , and the time t p r ( M ) t_{\mathrm{pr}}^{(M)} t pr ( M ) , a block boundary between two quiet margins, with t ( M ) + h M < t p r ( M ) ≤ t ( M ) + 2 h M t_{(M)}+h_M<t_{\mathrm{pr}}^{(M)}\le t_{(M)}+2h_M t ( M ) + h M < t pr ( M ) ≤ t ( M ) + 2 h M . Here G ~ M \widetilde{G}_M G M has paths of at most n N l e n , w ≤ N p a t h nN_{\mathrm{len},w}\le N_{\mathrm{path}} n N len , w ≤ N path sources (the remark after Definition 15.3, and item (1) of Subsection 11.1), as Proposition 12.7 asks; and that proposition holds with N o u t , e N_{\mathrm{out,e}} N out , e in place of N o u t N_{\mathrm{out}} N out , as its proof uses N o u t N_{\mathrm{out}} N out only as the order J J J of Lemma 12.15, which holds for every J J J , in Step 4, and as the order of the norms of Step 5, which the chart at t p r t_{\mathrm{pr}} t pr changes by at most a constant. The third bound follows by interpolation in Fourier between the second bound and the bounds of U ~ M ( t p r ( M ) ) \widetilde{U}_M(t_{\mathrm{pr}}^{(M)}) U M ( t pr ( M ) ) in H ε M , ω N t r , w ( T 2 ) H_{\varepsilon_M,\omega}^{N_{\mathrm{tr},w}}(\mathbb{T}^2) H ε M , ω N tr , w ( T 2 ) (Lemma 15.8) and of Π U ~ M ( t ( M ) ) \Pi\widetilde{U}_M(t_{(M)}) Π U M ( t ( M ) ) in H ρ ( M ) , ω N o u t , e ( T 2 ) H_{\rho(M),\omega}^{N_{\mathrm{out,e}}}(\mathbb{T}^2) H ρ ( M ) , ω N out , e ( T 2 ) (the first bound), as 2 N r d , e ≤ min { N t r , w , N o u t , e } 2N_{\mathrm{rd,e}}\le\min\{N_{\mathrm{tr},w},N_{\mathrm{out,e}}\} 2 N rd , e ≤ min { N tr , w , N out , e } and ε M ≤ ρ ( M ) \varepsilon_M\le\rho(M) ε M ≤ ρ ( M ) .
(b) Proposition 8.3(c) applies to G ′ G' G ′ at the order N o u t , e N_{\mathrm{out,e}} N out , e , the scale ρ ( M ) \rho(M) ρ ( M ) and the diffusivity κ M − 1 \kappa_{M-1} κ M − 1 . Its condition (i) holds since ρ ( M ) ≤ ε M − 1 \rho(M)\le\varepsilon_{M-1} ρ ( M ) ≤ ε M − 1 and ρ ( M ) 2 ε M − 1 β − 2 = κ M − 1 h M ε M − 1 β − 2 ≤ C ε M − 1 2 δ κ M − 1 \rho(M)^2\varepsilon_{M-1}^{\beta-2}=\kappa_{M-1}h_M\varepsilon_{M-1}^{\beta-2}\le C\varepsilon_{M-1}^{2\delta}\kappa_{M-1} ρ ( M ) 2 ε M − 1 β − 2 = κ M − 1 h M ε M − 1 β − 2 ≤ C ε M − 1 2 δ κ M − 1 by (2.5); (ii) as in the proof of Proposition 13.10(i); and the coupling sums of G ′ G' G ′ are at most C ε M − 1 δ / 2 C\varepsilon_{M-1}^{\delta/2} C ε M − 1 δ /2 by clauses (b3) and (d) in the proof of Proposition 15.11, which bound the matrix (8.4) at the order N o u t , e N_{\mathrm{out,e}} N out , e and the scale ρ ( M ) \rho(M) ρ ( M ) , as it increases with both, ρ ( M ) ≤ ρ m o l ( M − 1 ) \rho(M)\le\rho_{\mathrm{mol}}(M-1) ρ ( M ) ≤ ρ mol ( M − 1 ) and the seed’s derivative order is at least N o u t , e N_{\mathrm{out,e}} N out , e (item (2) of Subsection 11.1). That proof uses Lemma 15.4(b), on [ t ( M ) , t f ] [t_{(M)},t_f] [ t ( M ) , t f ] , and bounds that hold at every time, so these clauses hold on [ t ( M ) + 3 τ M , t f ] ⊃ [ t p r ( M ) , t f ] [t_{(M)}+3\tau_M,t_f]\supset[t_{\mathrm{pr}}^{(M)},t_f] [ t ( M ) + 3 τ M , t f ] ⊃ [ t pr ( M ) , t f ] , as t p r ( M ) > t ( M ) + h M ≥ t ( M ) + 3 τ M t_{\mathrm{pr}}^{(M)}>t_{(M)}+h_M\ge t_{(M)}+3\tau_M t pr ( M ) > t ( M ) + h M ≥ t ( M ) + 3 τ M (proof of (a), (2.4)). The datum is bounded by (a).
(c) The identities. Lemmas 13.1, 13.2 and 13.4–13.6 hold at m = M m=M m = M for G ~ M \widetilde{G}_M G M , with N c o r , e N_{\mathrm{cor,e}} N cor , e , N s r c , e N_{\mathrm{src,e}} N src , e , N c l , e N_{\mathrm{cl,e}} N cl , e , N r e c , e N_{\mathrm{rec,e}} N rec , e , n N l e n , w nN_{\mathrm{len},w} n N len , w in place of N c o r N_{\mathrm{cor}} N cor , N s r c N_{\mathrm{src}} N src , N c l N_{\mathrm{cl}} N cl , N r e c N_{\mathrm{rec}} N rec , N p a t h N_{\mathrm{path}} N path : G ~ M \widetilde{G}_M G M is in the setting of the lift (Section 13) with these orders, as it satisfies the clauses of the invariant (Lemma 15.4(b)), its birth orders are at most 1 + n N c o r , w ≤ N s r c , e 1+nN_{\mathrm{cor},w}\le N_{\mathrm{src,e}} 1 + n N cor , w ≤ N src , e , its paths have at most n N l e n , w nN_{\mathrm{len},w} n N len , w sources (proof of (a)), and its birth radii, those of G M G_M G M , are ε M + 1 ≤ 2 ε M q − 1 ε M ≤ ε M \varepsilon_{M+1}\le2\varepsilon_M^{q-1}\varepsilon_M\le\varepsilon_M ε M + 1 ≤ 2 ε M q − 1 ε M ≤ ε M for M ≥ M ∗ M\ge M_* M ≥ M ∗ (Definition 15.3, Lemma 7.1(a)). The identity of Lemma 13.7 is exact, and the anchor does not change it, as only the derivative J M k − κ M − 1 I J_M^k-\kappa_{M-1}I J M k − κ M − 1 I of the covariance appears in it; the readout identity of G ′ G' G ′ is that of Lemma 13.8, with the same proof. By (b), Υ e Y \Upsilon^eY Υ e Y and its anchored exponential have the classes of Lemmas 13.7(ii) and 13.5, as in Step 2 of the proof of Proposition 13.11 with N o u t , e N_{\mathrm{out,e}} N out , e , N s r c , e N_{\mathrm{src,e}} N src , e , p e p_e p e in place of N r e t + 3 N_{\mathrm{ret}}+3 N ret + 3 , N s r c N_{\mathrm{src}} N src , p p p : by (15.7) and the first line of (15.6), N o u t , e − N s r c , e − 2 p e − 2 ≥ N r e c , e + 1 N_{\mathrm{out,e}}-N_{\mathrm{src,e}}-2p_e-2\ge N_{\mathrm{rec,e}}+1 N out , e − N src , e − 2 p e − 2 ≥ N rec , e + 1 , which gives the orders 2 p e + 1 2p_e+1 2 p e + 1 and N c o r , e + 1 N_{\mathrm{cor,e}}+1 N cor , e + 1 of these lemmas. As in Step 2 of the proof of Lemma 15.7, with the cost C ε M / ρ ( M ) C\varepsilon_M/\rho(M) C ε M / ρ ( M ) of a derivative at the scale ε M \varepsilon_M ε M on slow factors, the proofs of Lemmas 13.4–13.6 give, for N ≤ N r d , e N\le N_{\mathrm{rd,e}} N ≤ N rd , e , the same bounds with H ε M , ω N ( T 2 ) H_{\varepsilon_M,\omega}^N(\mathbb{T}^2) H ε M , ω N ( T 2 ) in place of L 2 ( T 2 ) L^2(\mathbb{T}^2) L 2 ( T 2 ) in the norm (13.4), reading N N N more orders. At N = N r d , e N=N_{\mathrm{rd,e}} N = N rd , e the residual reads Y Y Y at most at the order 2 N r d , e ≤ N o u t , e − 2 2N_{\mathrm{rd,e}}\le N_{\mathrm{out,e}}-2 2 N rd , e ≤ N out , e − 2 , and the row of the old smoothings at c m o l ε M c_{\mathrm{mol}}\varepsilon_M c mol ε M in Table 15.1 reads Y ♯ Y^\sharp Y ♯ at the order N r d , e + N m o m , w + n N c o r , w − 1 ≤ N o u t , e − N c o r , e − 2 p e N_{\mathrm{rd,e}}+N_{\mathrm{mom},w}+nN_{\mathrm{cor},w}-1\le N_{\mathrm{out,e}}-N_{\mathrm{cor,e}}-2p_e N rd , e + N mom , w + n N cor , w − 1 ≤ N out , e − N cor , e − 2 p e (first line of (15.6)), up to which the component energy of Y ♯ Y^\sharp Y ♯ at the scale ρ m o l ( M ) \rho_{\mathrm{mol}}(M) ρ mol ( M ) , with the diffusivity κ \kappa κ , is at most C ∥ θ i n ∥ L 2 ( T 2 ) C\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)} C ∥ θ in ∥ L 2 ( T 2 ) , by the proof of Proposition 13.10(ii) with the sizes of Lemma 15.1(b). So Y ♯ Y^\sharp Y ♯ solves the system of G ~ M \widetilde{G}_M G M on [ t p r ( M ) , t f ] [t_{\mathrm{pr}}^{(M)},t_f] [ t pr ( M ) , t f ] with a forcing, the residual, listed in Table 15.1 ; by (15.2), (15.5) and Lemma 15.1(b), each is at most C ε M − 1 Q e + 2 ∥ θ i n ∥ L 2 ( T 2 ) C\varepsilon_{M-1}^{Q_e+2}\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)} C ε M − 1 Q e + 2 ∥ θ in ∥ L 2 ( T 2 ) , the energy of Y Y Y being bounded by (b).
The initial difference. At t p r ( M ) t_{\mathrm{pr}}^{(M)} t pr ( M ) the anchored exponential is the identity, Y Y Y vanishes on the children, and the correctors of the pulses whose cutoff is positive vanish, by Lemma 13.1, as the velocity added at level M M M vanishes on [ t p r ( M ) − τ M ′ , t p r ( M ) + τ M ′ ] [t_{\mathrm{pr}}^{(M)}-\tau_M',t_{\mathrm{pr}}^{(M)}+\tau_M'] [ t pr ( M ) − τ M ′ , t pr ( M ) + τ M ′ ] and t p r ( M ) > t ( M ) + 3 τ M t_{\mathrm{pr}}^{(M)}>t_{(M)}+3\tau_M t pr ( M ) > t ( M ) + 3 τ M . So Y ♯ ( t p r ( M ) ) = Π U ~ M ( t ( M ) ) Y^\sharp(t_{\mathrm{pr}}^{(M)})=\Pi\widetilde{U}_M(t_{(M)}) Y ♯ ( t pr ( M ) ) = Π U M ( t ( M ) ) , and by (a) the difference U ~ M − Y ♯ \widetilde{U}_M-Y^\sharp U M − Y ♯ at t p r ( M ) t_{\mathrm{pr}}^{(M)} t pr ( M ) is at most C ε M Q e / 2 ∥ θ i n ∥ L 2 ( T 2 ) C\varepsilon_M^{Q_e/2}\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)} C ε M Q e /2 ∥ θ in ∥ L 2 ( T 2 ) in H ε M , ω N r d , e ( T 2 ) H_{\varepsilon_M,\omega}^{N_{\mathrm{rd,e}}}(\mathbb{T}^2) H ε M , ω N rd , e ( T 2 ) .
Energy. Proposition 8.3(a) for G ~ M \widetilde{G}_M G M at the order N r d , e N_{\mathrm{rd,e}} N rd , e and the scale ε M \varepsilon_M ε M , whose hypotheses hold as in the proof of Lemma 15.8, bounds the left side of (c) by C ′ C' C ′ times the sum of the initial difference and the residual, and ε M − 1 Q e + 2 ≤ ε M − 1 Q e / 2 \varepsilon_{M-1}^{Q_e+2}\le\varepsilon_{M-1}^{Q_e/2} ε M − 1 Q e + 2 ≤ ε M − 1 Q e /2 . By (b), Y ♯ ( t ) ∈ H ε M , ω N r d , e ( T 2 ) V Y^\sharp(t)\in H_{\varepsilon_M,\omega}^{N_{\mathrm{rd,e}}}(\mathbb{T}^2)^V Y ♯ ( t ) ∈ H ε M , ω N rd , e ( T 2 ) V , as N o u t , e ≥ N c o r , e + 2 p e + N r d , e + 2 N_{\mathrm{out,e}}\ge N_{\mathrm{cor,e}}+2p_e+N_{\mathrm{rd,e}}+2 N out , e ≥ N cor , e + 2 p e + N rd , e + 2 and the lift spends N c o r , e + 2 p e N_{\mathrm{cor,e}}+2p_e N cor , e + 2 p e derivatives, and two more in the Sobolev embedding.
The first family of the induction
Proposition 15.11 makes the terminal successor, restarted at t e t_{\mathrm e} t e , an admissible seed G = G M − 1 G=G_{M-1} G = G M − 1 . The reserve of generation 0 is the one the terminal cycle reaches from res M \operatorname{res}_{M} res M , and the markers of G ′ G' G ′ are 1, so declaring its components of generation 0 changes no coefficient, smoothing or mass. As G ′ G' G ′ does not descend from a seed, Lemma 11.7 does not apply; the main estimate is the recursion for (b3): a new source gains ε M − 1 δ \varepsilon^{\delta}_{M-1} ε M − 1 δ from its majorant and loses less than ε M − 1 − δ / 2 \varepsilon^{-\delta/2}_{M-1} ε M − 1 − δ /2 .
Proposition 15.11 (The seed). Let M ≥ M ∗ M\ge M_* M ≥ M ∗ and t e : = t p r ( M ) + h M t_{\mathrm e}:=t_{\mathrm{pr}}^{(M)}+h_M t e := t pr ( M ) + h M . Then t e ≤ t i n + 1 4 t 0 t_{\mathrm e}\le t_{\mathrm{in}}+\frac{1}{4}t_0 t e ≤ t in + 4 1 t 0 , and the family G G G of level M − 1 M-1 M − 1 obtained from G ′ G' G ′ by declaring all its components of generation 0 is an admissible seed of level M − 1 M-1 M − 1 on [ t e , t f ] [t_{\mathrm e},t_{\mathrm f}] [ t e , t f ] (Definition 11.2), with the bounds of item (3) of Subsection 15.1 and constants depending only on α , s , q , ζ , Λ , n , t 0 , T \alpha,s,q,\zeta,\Lambda,n,t_0,T α , s , q , ζ , Λ , n , t 0 , T .
Proof. By Propositions 15.5 and 15.10(a), and as M ≥ M ∗ M\ge M_* M ≥ M ∗ , t e ≤ t i n + 6 n h M + 1 + ( C + 1 ) h M ≤ t i n + 1 4 t 0 t_{\mathrm e}\le t_{\mathrm{in}}+6nh_{M+1}+(C+1)h_M\le t_{\mathrm{in}}+\frac{1}{4}t_0 t e ≤ t in + 6 n h M + 1 + ( C + 1 ) h M ≤ t in + 4 1 t 0 , and t e ≥ t p r ( M ) ≥ t ( M ) ≥ t i n t_{\mathrm e}\ge t_{\mathrm{pr}}^{(M)}\ge t_{(M)}\ge t_{\mathrm{in}} t e ≥ t pr ( M ) ≥ t ( M ) ≥ t in , so t f − t e ≤ T t_{\mathrm f}-t_{\mathrm e}\le T t f − t e ≤ T ; we check Definition 11.2 with t S = t e t_S=t_{\mathrm e} t S = t e .
(i). The components of G ~ M \widetilde{G}_M G M have total mass at most 2 3 \frac{2}{3} 3 2 (Lemma 15.4(b)), their children at most c ε ε M − 1 δ / 32 c_{\varepsilon}\varepsilon_{M-1}^{\delta/32} c ε ε M − 1 δ /32 times that. The sources of G ~ M \widetilde{G}_M G M have birth orders at most 1 + n N c o r , w 1+nN_{\mathrm{cor,w}} 1 + n N cor , w , and the new ones at most N s r c , e N_{\mathrm{src,e}} N src , e . Along a path, at most N l e n , e + 1 N_{\mathrm{len,e}}+1 N len , e + 1 new sources separate at most N l e n , e + 2 N_{\mathrm{len,e}}+2 N len , e + 2 runs of sources of G ~ M \widetilde{G}_M G M , each of at most n N l e n , w nN_{\mathrm{len,w}} n N len , w sources, so a path has at most ( N l e n , e + 2 ) ( n N l e n , w + 1 ) (N_{\mathrm{len,e}}+2)(nN_{\mathrm{len,w}}+1) ( N len , e + 2 ) ( n N len , w + 1 ) components. The widths are r w ≤ ε M r_{\mathrm w}\le\varepsilon_M r w ≤ ε M and ρ ( M ) ε M − 1 Θ e ≤ ε M − 1 \rho(M)\varepsilon_{M-1}^{\Theta_{\mathrm e}}\le\varepsilon_{M-1} ρ ( M ) ε M − 1 Θ e ≤ ε M − 1 , and the birth radii of the sources from G ~ M \widetilde{G}_M G M are at most ε M + 1 \varepsilon_{M+1} ε M + 1 , those of the new ones ρ ( M ) ≤ ε M − 1 \rho(M)\le\varepsilon_{M-1} ρ ( M ) ≤ ε M − 1 .
(ii). The coefficients of G G G are those of G ′ G' G ′ . By (b1) of (iv) and the transport, as in Steps 6 and 8 of Subsection 11.6, each derivative of a coefficient costs at most ε M − 1 − 1 \varepsilon^{-1}_{M-1} ε M − 1 − 1 (the fast variable of level M − 1 M-1 M − 1 ; the slow ones cost ε M − 2 − 1 \varepsilon^{-1}_{M-2} ε M − 2 − 1 ) and the transport costs ε M − 2 d − d ′ ≤ ρ σ d − d ′ \varepsilon^{d-d'}_{M-2}\le\rho^{d-d'}_{\sigma} ε M − 2 d − d ′ ≤ ρ σ d − d ′ ; so the sum in (ii) at ρ = ε M − 1 \rho=\varepsilon_{M-1} ρ = ε M − 1 , which increases with ρ \rho ρ , is at most C fac ( σ ) maj ( σ ) κ M − 1 ρ d − 2 ≤ C maj ( σ ) κ M − 1 ρ d − 2 C\operatorname{fac}(\sigma)\operatorname{maj}(\sigma)\kappa_{M-1}\rho^{d-2}\le C\operatorname{maj}(\sigma)\kappa_{M-1}\rho^{d-2} C fac ( σ ) maj ( σ ) κ M − 1 ρ d − 2 ≤ C maj ( σ ) κ M − 1 ρ d − 2 , as fac ( σ ) ≤ 1 \operatorname{fac}(\sigma)\le1 fac ( σ ) ≤ 1 : it is 1 or a product of factors of letters of the G i − 1 G_{i-1} G i − 1 (Definitions 11.3, 15.9 and the proof of Lemma 15.4(b)), each at most C ε i − 1 2 ( q + 2 ) γ − δ / 16 ≤ 1 C\varepsilon_{i-1}^{2(q+2)\gamma-\delta/16}\le1 C ε i − 1 2 ( q + 2 ) γ − δ /16 ≤ 1 for M ≥ M ∗ M\ge M_* M ≥ M ∗ (Lemma 15.1(a)).
(iii). The cell solutions are those of the weak finite lifts, finitely many two-dimensional cell problems per pulse (Lemma 15.2), each solved once, from zero, on the window of its pulse, in the cell time κ t / ε i 2 \kappa t/\varepsilon_i^2 κ t / ε i 2 of its level i > M i>M i > M , depending, among the pulses of level i + 1 i+1 i + 1 , only on those whose charts are used there; the terminal cycle solves none. Their invariance when the phases move along the pulses and their determination on the window are proved as in Steps 1 and 2 of Subsection 11.6.
(iv): clauses (a), (b1), (b2), (b2′). As in Steps 1, 2 and 4–6 of Subsection 11.6, for G ~ M \widetilde{G}_M G M and G ′ G' G ′ at m = M m=M m = M , using Lemma 15.4(b) in place of the invariant 11.2, for all components of G ~ M \widetilde{G}_M G M , one step 1 / ( N g e n + 2 + N o f f ) 1/(N_{\mathrm{gen}}+2+N_{\mathrm{off}}) 1/ ( N gen + 2 + N off ) above generation 0, and with the supremum in time, as every bound used holds at every time. Three scale facts change: the correctors and the covariance gain ε ˉ M / ρ ( M ) ≤ C ε M − 1 γ + 3 δ \bar{\varepsilon}_M/\rho(M)\le C\varepsilon_{M-1}^{\gamma+3\delta} ε ˉ M / ρ ( M ) ≤ C ε M − 1 γ + 3 δ (Lemma 15.1(b)) and ∣ Γ M ( t ) − Γ M ( t p r ( M ) ) ∣ / ρ ( M ) 2 ≤ C ε M − 1 δ |\Gamma_M(t)-\Gamma_M(t_{\mathrm{pr}}^{(M)})|/\rho(M)^2\le C\varepsilon_{M-1}^{\delta} ∣ Γ M ( t ) − Γ M ( t pr ( M ) ) ∣/ ρ ( M ) 2 ≤ C ε M − 1 δ (after (208) ); in Step 5, the cell average of a source σ \sigma σ of G ~ M \widetilde{G}_M G M subtracted by a mean letter along σ \sigma σ gains ( ρ σ / ρ ( M ) ) d ′ − 2 (\rho_\sigma/\rho(M))^{d'-2} ( ρ σ / ρ ( M ) ) d ′ − 2 or ( ρ σ / ε M − 1 ) d − d ′ (\rho_\sigma/\varepsilon_{M-1})^{d-d'} ( ρ σ / ε M − 1 ) d − d ′ , at most C ε M − 1 γ + 3 δ C\varepsilon_{M-1}^{\gamma+3\delta} C ε M − 1 γ + 3 δ as ρ σ ≤ ε M + 1 \rho_\sigma\le\varepsilon_{M+1} ρ σ ≤ ε M + 1 , in place of κ M / κ M − 1 \kappa_M/\kappa_{M-1} κ M / κ M − 1 , which is not small here; and in Step 6 the birth radii ε M + 1 \varepsilon_{M+1} ε M + 1 and ρ ( M ) ≤ C ε M − 1 1 + γ / 2 + δ \rho(M)\le C\varepsilon_{M-1}^{1+\gamma/2+\delta} ρ ( M ) ≤ C ε M − 1 1 + γ /2 + δ are below ρ m o l ( M − 1 ) ≥ c ε M − 1 1 + γ / 2 \rho_{\mathrm{mol}}(M-1)\ge c\varepsilon_{M-1}^{1+\gamma/2} ρ mol ( M − 1 ) ≥ c ε M − 1 1 + γ /2 , and the widths r w , ρ ( M ) ε M − 1 Θ e r_{\mathrm w},\rho(M)\varepsilon_{M-1}^{\Theta_{\mathrm e}} r w , ρ ( M ) ε M − 1 Θ e below ε M − 1 \varepsilon_{M-1} ε M − 1 .
(iv): the recursion for (b3). The markers are 1, so the two choices ω w \omega_{\mathrm w} ω w and ω v / Ψ v 2 \omega_{\mathrm v}/\Psi_{\mathrm v}^2 ω v / Ψ v 2 of the masses agree. Step 3 of Subsection 11.6 uses Definition 11.3 only through two facts, which hold for G ~ M \widetilde{G}_M G M and G ′ G' G ′ by Definition 15.9: the copies keep the masses of the family, and the entry of a new source is at most a fixed number times the product along its path. It uses the invariant only through the old priced sums, which must be at most 1 2 \frac{1}{2} 2 1 and are at most C ε M 2 ( q + 2 ) γ − 5 δ / 64 C\varepsilon_M^{2(q+2)\gamma-5\delta/64} C ε M 2 ( q + 2 ) γ − 5 δ /64 here, with prices 1 (Lemma 15.4(b)). So the computation of (122) applies, with C C C depending only on the numbers of words and letters and on C m a j C_{\mathrm{maj}} C maj . A new source gains ε M − 1 δ \varepsilon_{M-1}^{\delta} ε M − 1 δ from its majorant C m a j ε M − 1 δ − δ / 16 C_{\mathrm{maj}}\varepsilon_{M-1}^{\delta-\delta/16} C maj ε M − 1 δ − δ /16 , which clause (b1) justifies as in Step 5 of Subsection 11.6: the mean letters gain powers of ε ˉ M / ρ ( M ) \bar{\varepsilon}_M/\rho(M) ε ˉ M / ρ ( M ) , and the covariance and defect letters ε M − 1 δ \varepsilon_{M-1}^{\delta} ε M − 1 δ , by the scale facts above (for the defect letter, the bound after (208) replaces Lemma 10.3(b)). It costs the power ε M − 1 − δ / 16 \varepsilon_{M-1}^{-\delta/16} ε M − 1 − δ /16 of its coefficients, the square root C ε M − 1 − δ / 64 C\varepsilon_{M-1}^{-\delta/64} C ε M − 1 − δ /64 of the mass ratio, and the price ε M − 1 − Θ e ( N s r c , e − 2 ) \varepsilon_{M-1}^{-\Theta_{\mathrm e}(N_{\mathrm{src,e}}-2)} ε M − 1 − Θ e ( N src , e − 2 ) of its width ρ ( M ) ε M − 1 Θ e \rho(M)\varepsilon_{M-1}^{\Theta_{\mathrm e}} ρ ( M ) ε M − 1 Θ e . Since Θ e ( N s r c , e − 2 ) < 1 16 δ \Theta_{\mathrm e}(N_{\mathrm{src,e}}-2)<\frac{1}{16}\delta Θ e ( N src , e − 2 ) < 16 1 δ by (203) , the exponent δ − δ 16 − δ 64 − Θ e ( N s r c , e − 2 ) \delta-\frac{\delta}{16}-\frac{\delta}{64}-\Theta_{\mathrm e}(N_{\mathrm{src,e}}-2) δ − 16 δ − 64 δ − Θ e ( N src , e − 2 ) exceeds 55 64 δ > 1 2 δ \frac{55}{64}\delta>\frac{1}{2}\delta 64 55 δ > 2 1 δ , and, as ε M q − 1 ≤ C ε M q ( q − 1 ) \varepsilon_M^{q-1}\le C\varepsilon_M^{q(q-1)} ε M q − 1 ≤ C ε M q ( q − 1 ) and q ( q − 1 ) ≥ 4 δ q(q-1)\ge4\delta q ( q − 1 ) ≥ 4 δ , every row and column sum of the priced majorant matrix of G ′ G' G ′ is at most
C ε M 2 ( q + 2 ) γ − 5 δ / 64 + C ε M − 1 δ − δ / 16 − δ / 64 − Θ e ( N s r c , e − 2 ) ≤ C ε M − 1 δ / 2 ≤ C s e e d ε M − 1 δ / 4 . C\varepsilon_M^{2(q+2)\gamma-5\delta/64}+C\varepsilon_{M-1}^{\delta-\delta/16-\delta/64-\Theta_{\mathrm e}(N_{\mathrm{src,e}}-2)}
\le C\varepsilon_{M-1}^{\delta/2}\le C_{\mathrm{seed}}\varepsilon_{M-1}^{\delta/4}. C ε M 2 ( q + 2 ) γ − 5 δ /64 + C ε M − 1 δ − δ /16 − δ /64 − Θ e ( N src , e − 2 ) ≤ C ε M − 1 δ /2 ≤ C seed ε M − 1 δ /4 .
(iv): clause (c). (c1): the sources from G ~ M \widetilde G_M G M have the smoothing S w S_w S w , and the new ones the round smoothings of Definition 15.9(iii); all have the inverse 2 1 / 2 I 2^{1/2}I 2 1/2 I of the matrix of long axes, of ∣ ⋅ ∣ ∞ \lvert\cdot\rvert_\infty ∣ ⋅ ∣ ∞ -norm 2 1 / 2 ≤ 2 2^{1/2}\le2 2 1/2 ≤ 2 , and the order ℓ \ell ℓ of the kernel W ℓ W_\ell W ℓ of their smoothings satisfies 2 ℓ − 1 ≥ N m o m 2\ell-1\ge N_{\mathrm{mom}} 2 ℓ − 1 ≥ N mom . (c2): by Lemma 7.1(a), (c) and Proposition 2.1(c), ρ m o l ( M − 1 ) ≥ c ε M − 1 1 + γ / 2 \rho_{\mathrm{mol}}(M-1)\ge c\varepsilon_{M-1}^{1+\gamma/2} ρ mol ( M − 1 ) ≥ c ε M − 1 1 + γ /2 and ρ ( M − 1 ) ≥ c ε M − 1 1 + γ / 2 + δ \rho(M-1)\ge c\varepsilon_{M-1}^{1+\gamma/2+\delta} ρ ( M − 1 ) ≥ c ε M − 1 1 + γ /2 + δ , and, with q − 1 − γ 2 = 4 δ + q γ q-1-\frac{\gamma}{2}=4\delta+q\gamma q − 1 − 2 γ = 4 δ + q γ ,
r w ρ m o l ( M − 1 ) ≤ C ε M − 1 q γ + 4 δ + q ( γ / 4 + δ / 2 ) , ρ ( M ) ε M − 1 Θ e ρ m o l ( M − 1 ) ≤ C ε M − 1 δ + Θ e , \frac{r_w}{\rho_{\mathrm{mol}}(M-1)}\le C\varepsilon_{M-1}^{q\gamma+4\delta+q(\gamma/4+\delta/2)},\qquad
\frac{\rho(M)\varepsilon_{M-1}^{\Theta_{\mathrm e}}}{\rho_{\mathrm{mol}}(M-1)}\le C\varepsilon_{M-1}^{\delta+\Theta_{\mathrm e}}, ρ mol ( M − 1 ) r w ≤ C ε M − 1 q γ + 4 δ + q ( γ /4 + δ /2 ) , ρ mol ( M − 1 ) ρ ( M ) ε M − 1 Θ e ≤ C ε M − 1 δ + Θ e ,
r w ρ ( M − 1 ) ≤ C ε M − 2 q 2 ( 1 + γ / 4 + δ / 2 ) − 1 − γ / 2 − δ , ρ ( M ) ε M − 1 Θ e ρ ( M − 1 ) ≤ C ε M − 2 ( q − 1 ) ( 1 + γ / 2 + δ ) + q Θ e . \frac{r_w}{\rho(M-1)}\le C\varepsilon_{M-2}^{q^2(1+\gamma/4+\delta/2)-1-\gamma/2-\delta},\qquad
\frac{\rho(M)\varepsilon_{M-1}^{\Theta_{\mathrm e}}}{\rho(M-1)}\le C\varepsilon_{M-2}^{(q-1)(1+\gamma/2+\delta)+q\Theta_{\mathrm e}}. ρ ( M − 1 ) r w ≤ C ε M − 2 q 2 ( 1 + γ /4 + δ /2 ) − 1 − γ /2 − δ , ρ ( M − 1 ) ρ ( M ) ε M − 1 Θ e ≤ C ε M − 2 ( q − 1 ) ( 1 + γ /2 + δ ) + q Θ e .
The exponents of the first line are at least δ \delta δ , those of the second at least q − 1 ≥ 4 δ q-1\ge4\delta q − 1 ≥ 4 δ , as γ 2 + δ < 7 12 ( q − 1 ) \frac{\gamma}{2}+\delta<\frac{7}{12}(q-1) 2 γ + δ < 12 7 ( q − 1 ) . (c3): Lemma 8.2, with the constant matrices 2 − 1 / 2 I 2^{-1/2}I 2 − 1/2 I .
(iv): clause (d). As in Step 8 of Subsection 11.6, the coupling matrices of G G G at the scale ρ m o l ( M − 1 ) \rho_{\mathrm{mol}}(M-1) ρ mol ( M − 1 ) are at most C C C times the priced majorant matrix, entry by entry, so their row and column sums are at most C ε M − 1 δ / 2 ≤ C ε M − 2 q δ / 2 ≤ C s e e d ε M − 2 q δ / 4 C\varepsilon_{M-1}^{\delta/2}\le C\varepsilon_{M-2}^{q\delta/2}\le C_{\mathrm{seed}}\varepsilon_{M-2}^{q\delta/4} C ε M − 1 δ /2 ≤ C ε M − 2 q δ /2 ≤ C seed ε M − 2 q δ /4 , by Lemma 7.1(a). □
From the first family to the solution θ \theta θ
Representations of any order. Each lift spends derivatives, so the representations on the way back down have the orders n N c o r , w + 3 nN_{\mathrm{cor,w}}+3 n N cor , w + 3 (at ε M \varepsilon_M ε M ) down to 3 3 3 (at ε M + n \varepsilon_{M+n} ε M + n ): for N ≥ 1 N\ge1 N ≥ 1 , a representation of order N N N at wavelength λ \lambda λ and its size size λ , N ( R ) \operatorname{size}_{\lambda,N}(R) size λ , N ( R ) are defined by Subsection 5.6 and (41) with N N N in place of N r e p N_{\mathrm{rep}} N rep (size λ = size λ , N r e p \operatorname{size}_{\lambda}=\operatorname{size}_{\lambda,N_{\mathrm{rep}}} size λ = size λ , N rep ), and Lemma 14.2 holds for them, with the same proof.
The composed weak lift. For t ∈ R t\in\mathbb{R} t ∈ R , let W M + n ( t ) U : = U W_{M+n}(t)U:=U W M + n ( t ) U := U for a field U U U of G M + n G_{M+n} G M + n , a function, and, for M < i ≤ M + n M<i\le M+n M < i ≤ M + n , W i − 1 ( t ) U : = W i ( t ) Corr i w ( t ) Υ i w U W_{i-1}(t)U:=W_i(t)\operatorname{Corr}^{w}_{i}(t)\Upsilon^{w}_{i}U W i − 1 ( t ) U := W i ( t ) Corr i w ( t ) Υ i w U for a field U U U of G i − 1 G_{i-1} G i − 1 of class C ( M + n − i + 1 ) N c o r , w + 1 C^{(M+n-i+1)N_{\mathrm{cor,w}}+1} C ( M + n − i + 1 ) N cor , w + 1 ; so W M ( t ) U W_M(t)U W M ( t ) U is a function for every field U U U of G M G_M G M of class C n N c o r , w + 1 C^{nN_{\mathrm{cor,w}}+1} C n N cor , w + 1 .
Lemma 15.12 carries a representation through the terminal lift at a constant cost, harmless as only n + 1 n+1 n + 1 lifts remain: the anchored covariance is small against ε M − 1 δ \varepsilon_{M-1}^{\delta} ε M − 1 δ and the contrast bounded.
Lemma 15.12 (The terminal lift of representations). There exists C ( α , s , q , ζ , Λ , n , t 0 , T ) C(\alpha,s,q,\zeta,\Lambda,n,t_0,T) C ( α , s , q , ζ , Λ , n , t 0 , T ) such that, for every M ≥ M ∗ M\ge M_* M ≥ M ∗ , every field Y Y Y of G ′ G' G ′ on [ t p r ( M ) , t f ] [t_{\mathrm{pr}}^{(M)},t_{\mathrm f}] [ t pr ( M ) , t f ] , every t ∈ [ t p r ( M ) , t f ] t\in[t_{\mathrm{pr}}^{(M)},t_{\mathrm f}] t ∈ [ t pr ( M ) , t f ] and every representation R R R of Y ( t ) Y(t) Y ( t ) at wavelength ε M − 1 \varepsilon_{M-1} ε M − 1 (of order N r e p N_{\mathrm{rep}} N rep ), there exists a representation R ♯ R^\sharp R ♯ of its terminal lift Y ♯ ( t ) Y^\sharp(t) Y ♯ ( t ) (208) of order n N c o r , w + 3 nN_{\mathrm{cor,w}}+3 n N cor , w + 3 at wavelength ε M \varepsilon_M ε M with size ε M , n N c o r , w + 3 ( R ♯ ) ≤ C size ε M − 1 ( R ) \operatorname{size}_{\varepsilon_M,nN_{\mathrm{cor,w}}+3}(R^\sharp)\le C\operatorname{size}_{\varepsilon_{M-1}}(R) size ε M , n N cor , w + 3 ( R ♯ ) ≤ C size ε M − 1 ( R ) .
Proof. Let E E E be the base of R R R and N : = n N c o r , w + 3 N:=nN_{\mathrm{cor,w}}+3 N := n N cor , w + 3 ; we follow the three steps of (208) . Step 1: the readout. Scaled by ω v − 1 / 2 \omega_v^{-1/2} ω v − 1/2 , R e \mathbb{R}^{e} R e is the identity plus, for each child v ′ v' v ′ of v v v , one entry ε M − 1 δ # v ′ / 16 ( ω v ′ / ω v ) 1 / 2 ≤ C ε M − 1 5 δ / 64 \varepsilon_{M-1}^{\delta^{\#v'/16}}(\omega_{v'}/\omega_v)^{1/2} \le C\varepsilon_{M-1}^{5\delta/64} ε M − 1 δ # v ′ /16 ( ω v ′ / ω v ) 1/2 ≤ C ε M − 1 5 δ /64 , so ∣ R e ∣ ≤ 2 |\mathbb{R}^{e}| \le2 ∣ R e ∣ ≤ 2 for the norms ∣ ⋅ ∣ ω |\cdot|_{\omega} ∣ ⋅ ∣ ω , by Schur’s test, and it maps R R R to a representation of size at most 2 size ε M − 1 ( R ) 2\operatorname{size}_{\varepsilon_{M-1}}(R) 2 size ε M − 1 ( R ) (Subsection 5.6).
Step 2: the exponential. For 1 ≤ j ≤ p e 1 \le j \le p_e 1 ≤ j ≤ p e , on a profile of wavelength λ P ≥ ε M − 1 \lambda_P \ge\varepsilon_{M-1} λ P ≥ ε M − 1 , ( ( Γ M ( t ) − Γ M ( t p r ( M ) ) ) : ∇ 2 ) j ((\Gamma_M(t)-\Gamma_M(t_{\mathrm{pr}}^{(M)})):\nabla^2)^j (( Γ M ( t ) − Γ M ( t pr ( M ) )) : ∇ 2 ) j costs ( ∣ Γ M ( t ) − Γ M ( t p r ( M ) ) ∣ / λ P ) j ≤ C (|\Gamma_M(t)-\Gamma_M(t_{\mathrm{pr}}^{(M)})|/\lambda_P)^j \le C ( ∣ Γ M ( t ) − Γ M ( t pr ( M ) ) ∣/ λ P ) j ≤ C (after (208) ) and 2 j 2j 2 j derivatives; so it maps the profiles, and the base read as a profile of wavelength ε M − 1 \varepsilon_{M-1} ε M − 1 , to profiles of the same wavelengths, of order N + N c o r , e + 1 N+N_{\mathrm{cor},e}+1 N + N cor , e + 1 and size at most C size ε M − 1 ( R ) C\operatorname{size}_{\varepsilon_{M-1}}(R) C size ε M − 1 ( R ) .
Step 3: the correctors. Corr e ( t ) P − P \operatorname{Corr}^{e}(t)P-P Corr e ( t ) P − P is the sum, over the pulses k k k with H M , k ( t ) ≠ 0 H_{M,k}(t)\ne0 H M , k ( t ) = 0 , of the correctors of degrees 1 , … , N c o r , e 1,\ldots,N_{\mathrm{cor},e} 1 , … , N cor , e of P P P , read in the chart of k k k at z = y / ε M z=y/\varepsilon_M z = y / ε M . By the computation of Proposition 10.7(a), with the bounds of Lemma 15.4(b) in place of its hypotheses and ε ˉ M ≤ 2 1 / 2 ε M \bar{\varepsilon}_M\le2^{1/2}\varepsilon_M ε ˉ M ≤ 2 1/2 ε M (as κ M − 1 / κ ≤ 2 \kappa_{M-1}/\kappa\le2 κ M − 1 / κ ≤ 2 ), the coefficient of ∇ d P \nabla^dP ∇ d P in a corrector of degree j ≥ d ≥ 1 j\ge d\ge1 j ≥ d ≥ 1 is at most C ( 2 1 / 2 ε M ) j ε M − 1 d − j ≤ C ε M d C(2^{1/2}\varepsilon_M)^j\varepsilon_{M-1}^{d-j}\le C\varepsilon_M^d C ( 2 1/2 ε M ) j ε M − 1 d − j ≤ C ε M d , and each of its derivatives costs C ε M − 1 C\varepsilon_M^{-1} C ε M − 1 ; the chart has bounded derivatives at the scale ε M − 1 \varepsilon_{M-1} ε M − 1 .
So, for a profile P P P of wavelength λ P ≥ ε M − 1 \lambda_P\ge\varepsilon_{M-1} λ P ≥ ε M − 1 , Corr e ( t ) P − P \operatorname{Corr}^{e}(t)P-P Corr e ( t ) P − P is a profile of wavelength ε M \varepsilon_M ε M with ε M − s ∥ Corr e ( t ) P − P ∥ C ε M N ( T 2 ) ≤ C ∑ d ≥ 1 ε M − s ( ε M / λ P ) d [ P ] C λ P N + N c o r , e ( T 2 ) ≤ C λ P − s [ P ] C λ P N + N c o r , e ( T 2 ) \varepsilon_M^{-s}\|\operatorname{Corr}^{e}(t)P-P\|_{C^N_{\varepsilon_M}(\mathbb{T}^2)}\le C\sum_{d\ge1}\varepsilon_M^{-s}(\varepsilon_M/\lambda_P)^d[P]_{C^{N+N_{\mathrm{cor},e}}_{\lambda_P}(\mathbb{T}^2)}\le C\lambda_P^{-s}[P]_{C^{N+N_{\mathrm{cor},e}}_{\lambda_P}(\mathbb{T}^2)} ε M − s ∥ Corr e ( t ) P − P ∥ C ε M N ( T 2 ) ≤ C ∑ d ≥ 1 ε M − s ( ε M / λ P ) d [ P ] C λ P N + N cor , e ( T 2 ) ≤ C λ P − s [ P ] C λ P N + N cor , e ( T 2 ) , since d ≥ 1 > s d\ge1>s d ≥ 1 > s . The profiles of Step 2 keep their wavelengths, and the base, read at the wavelength ε M \varepsilon_M ε M , has ε M − s [ E ] C ε M N ( T 2 ) ≤ ε M − 1 − s [ E ] C ε M − 1 N ( T 2 ) \varepsilon_M^{-s}[E]_{C^N_{\varepsilon_M}(\mathbb{T}^2)}\le\varepsilon_{M-1}^{-s}[E]_{C^N_{\varepsilon_{M-1}}(\mathbb{T}^2)} ε M − s [ E ] C ε M N ( T 2 ) ≤ ε M − 1 − s [ E ] C ε M − 1 N ( T 2 ) , as ( ε M / ε M − 1 ) 1 − s ≤ 1 (\varepsilon_M/\varepsilon_{M-1})^{1-s}\le1 ( ε M / ε M − 1 ) 1 − s ≤ 1 . The three steps use at most N + N c o r , e + 2 p e + 1 = N r e p N+N_{\mathrm{cor},e}+2p_e+1=N_{\mathrm{rep}} N + N cor , e + 2 p e + 1 = N rep derivatives, which gives the lemma.
By Lemma 15.13(a), the weak lifts do the same, each spending N c o r , w N_{\mathrm{cor},w} N cor , w derivatives, as their correctors take derivatives of the field with a factor ε i \varepsilon_i ε i each. By (b), a Sobolev bound at the wavelength ε i \varepsilon_i ε i becomes, after the lifts, a Hölder bound with the loss ε M + n − 1 − s \varepsilon_{M+n}^{-1-s} ε M + n − 1 − s , offset by the accuracies of the weak passages, of the smoothing and of the terminal cycle.
Lemma 15.13 (The weak lifts of representations). There exists C ( α , s , q , ζ , Λ , n , t 0 , T ) C(\alpha,s,q,\zeta,\Lambda,n,t_0,T) C ( α , s , q , ζ , Λ , n , t 0 , T ) such that, for every M ≥ M ∗ M\ge M_* M ≥ M ∗ , M ≤ i ≤ M + n M\le i\le M+n M ≤ i ≤ M + n and t ∈ R t\in\mathbb{R} t ∈ R , the following hold.
(a) For every representation R R R of a field U U U of G i G_i G i of order ( M + n − i ) N c o r , w + 3 (M+n-i)N_{\mathrm{cor},w}+3 ( M + n − i ) N cor , w + 3 at wavelength ε i \varepsilon_i ε i , there exists a representation R ♯ R^\sharp R ♯ of W i ( t ) U \mathcal{W}_i(t)U W i ( t ) U of order 3 3 3 at wavelength ε M + n \varepsilon_{M+n} ε M + n with size ε M + n , 3 ( R ♯ ) ≤ C size ε i , ( M + n − i ) N c o r , w + 3 ( R ) \operatorname{size}_{\varepsilon_{M+n},3}(R^\sharp)\le C\operatorname{size}_{\varepsilon_i,(M+n-i)N_{\mathrm{cor},w}+3}(R) size ε M + n , 3 ( R ♯ ) ≤ C size ε i , ( M + n − i ) N cor , w + 3 ( R ) .
(b) For every field U U U of G i G_i G i with components in H N i ( T 2 ) H^{N_i}(\mathbb{T}^2) H N i ( T 2 ) ,
∥ W i ( t ) U ∥ C 0 , s ( T 2 ) ≤ C ε M + n − 1 − s ∥ U ∥ H ε i , ω N i ( T 2 ) . \|\mathcal{W}_i(t)U\|_{C^{0,s}(\mathbb{T}^2)}\le C\varepsilon_{M+n}^{-1-s}\|U\|_{H^{N_i}_{\varepsilon_i,\omega}(\mathbb{T}^2)}. ∥ W i ( t ) U ∥ C 0 , s ( T 2 ) ≤ C ε M + n − 1 − s ∥ U ∥ H ε i , ω N i ( T 2 ) .
:::
Proof. (a) By induction on M + n − i M+n-i M + n − i : ∣ R i w ∣ ≤ 2 |\mathbb{R}_i^w|\le2 ∣ R i w ∣ ≤ 2 as in Step 1 of the proof of Lemma 15.12, and the coefficient of ∇ d P \nabla^dP ∇ d P in a corrector of degree j ≥ d ≥ 1 j\ge d\ge1 j ≥ d ≥ 1 of Corr i w \operatorname{Corr}_i^w Corr i w is at most C ε i j ε i − 1 d − j ≤ C ε i d C\varepsilon_i^{j}\varepsilon_{i-1}^{d-j}\le C\varepsilon_i^d C ε i j ε i − 1 d − j ≤ C ε i d , each derivative costing C ε i − 1 C\varepsilon_i^{-1} C ε i − 1 (proof of Lemma 15.4(a)); so Step 3 there, with ε i \varepsilon_i ε i for ε M \varepsilon_M ε M and exponential, gives the passage, which spends N c o r , w N_{\mathrm{cor},w} N cor , w derivatives. (b) As in the proof of Lemma 14.3, U U U , with no profile, is a representation of order N i − 2 = ( M + n − i ) N c o r , w + 3 N_i-2=(M+n-i)N_{\mathrm{cor},w}+3 N i − 2 = ( M + n − i ) N cor , w + 3 at wavelength ε i \varepsilon_i ε i of size at most C ε i − 1 − s ∥ U ∥ H ε i , ω N i ( T 2 ) C\varepsilon_i^{-1-s}\|U\|_{H^{N_i}_{\varepsilon_i,\omega}(\mathbb{T}^2)} C ε i − 1 − s ∥ U ∥ H ε i , ω N i ( T 2 ) ; (a) and Lemma 14.2(a) with the order 3 give (b), as ε i ≥ ε M + n \varepsilon_i\ge\varepsilon_{M+n} ε i ≥ ε M + n .
In Lemma 15.14, the difference between θ \theta θ and the solution of the terminal successor, lifted through the terminal cycle and the n n n weak passages, telescopes into the errors of Lemmas 15.7 and 15.8 and Proposition 15.10(c), each read in C 0 , s ( T 2 ) C^{0,s}(\mathbb{T}^2) C 0 , s ( T 2 ) by Lemma 15.13(b).
Lemma 15.14 (Recovering θ \theta θ ). There exists C ( α , s , q , ζ , Λ , n , t 0 , T ) C(\alpha,s,q,\zeta,\Lambda,n,t_0,T) C ( α , s , q , ζ , Λ , n , t 0 , T ) such that, for every M ≥ M ∗ M\ge M_* M ≥ M ∗ , θ i n \theta_{\mathrm{in}} θ in and θ \theta θ , with ( W M Y ♯ ) ( t ) : = W M ( t ) Y ♯ ( t ) (\mathcal{W}_{M}\mathsf{Y}^{\sharp})(t):=\mathcal{W}_{M}(t)\mathsf{Y}^{\sharp}(t) ( W M Y ♯ ) ( t ) := W M ( t ) Y ♯ ( t ) ,
∥ θ − W M Y ♯ ∥ L ∞ ( [ t i n + 1 4 t 0 , t f ] ; C 0 , s ( T 2 ) ) ≤ C ε M − 1 2 ∥ θ i n ∥ L 2 ( T 2 ) . \|\theta-\mathcal{W}_{M}\mathsf{Y}^{\sharp}\|_{L^\infty([t_{\mathrm{in}}+\frac14t_0,t_f];C^{0,s}(\mathbb{T}^2))}\le C\varepsilon_{M-1}^{2}\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)}. ∥ θ − W M Y ♯ ∥ L ∞ ([ t in + 4 1 t 0 , t f ] ; C 0 , s ( T 2 )) ≤ C ε M − 1 2 ∥ θ in ∥ L 2 ( T 2 ) .
:::
Proof. Let t ∈ [ t i n + 1 4 t 0 , t f ] t \in[t_{\mathrm{in}}+\frac{1}{4}t_0,t_f] t ∈ [ t in + 4 1 t 0 , t f ] , so t ≥ t e ≥ t ( M ) t \ge t_e \ge t_{(M)} t ≥ t e ≥ t ( M ) . As N t r , w ≥ N r d , e ≥ n N c o r , w + 5 N_{\mathrm{tr,w}} \ge N_{\mathrm{rd,e}} \ge nN_{\mathrm{cor,w}}+5 N tr , w ≥ N rd , e ≥ n N cor , w + 5 (item (4) of Subsection 15.1), by the Sobolev embedding W i − 1 ( t ) \mathcal{W}_{i-1}(t) W i − 1 ( t ) applies to U i − 1 ( t ) U_{i-1}(t) U i − 1 ( t ) and W M ( t ) \mathcal{W}_M(t) W M ( t ) to Y ♯ ( t ) Y^{\sharp}(t) Y ♯ ( t ) , whose components lie in H N t r , w ( T 2 ) H^{N_{\mathrm{tr,w}}}(\mathbb{T}^2) H N tr , w ( T 2 ) and H N r d , e ( T 2 ) H^{N_{\mathrm{rd,e}}}(\mathbb{T}^2) H N rd , e ( T 2 ) (proof of Lemma 15.7, Proposition 15.10 (c)). By Proposition 15.5 (i) and the definition of the W i \mathcal{W}_i W i ,
θ ( t ) − W M ( t ) Y ♯ ( t ) = ∑ i = M + 1 M + n W i ( t ) [ U i ( t ) − Corr w ( t ) Υ i w U i − 1 ( t ) ] + W M ( t ) [ U M ( t ) − Y ♯ ( t ) ] . \theta(t)-\mathcal{W}_M(t)Y^{\sharp}(t)=\sum_{i=M+1}^{M+n}\mathcal{W}_i(t)\left[U_i(t)-\operatorname{Corr}^{\mathrm{w}}(t)\Upsilon_i^{\mathrm{w}}U_{i-1}(t)\right]+\mathcal{W}_M(t)\left[U_M(t)-Y^{\sharp}(t)\right]. θ ( t ) − W M ( t ) Y ♯ ( t ) = i = M + 1 ∑ M + n W i ( t ) [ U i ( t ) − Corr w ( t ) Υ i w U i − 1 ( t ) ] + W M ( t ) [ U M ( t ) − Y ♯ ( t ) ] .
and U M − Y ♯ = ( U M − U ~ M ) + ( U ~ M − Y ♯ ) U_M-Y^{\sharp}=(U_M-\widetilde{U}_M)+(\widetilde{U}_M-Y^{\sharp}) U M − Y ♯ = ( U M − U M ) + ( U M − Y ♯ ) . We bound each term by Lemma 15.13 (b), with ε M + n − 1 − s ≤ C ε M − 2 q \varepsilon_{M+n}^{-1-s}\le C\varepsilon_M^{-2q} ε M + n − 1 − s ≤ C ε M − 2 q and ε M + n − 1 − s ≤ C ε M − 1 − q n + 1 ( 1 + s ) \varepsilon_{M+n}^{-1-s}\le C\varepsilon_{M-1}^{-q^{n+1}(1+s)} ε M + n − 1 − s ≤ C ε M − 1 − q n + 1 ( 1 + s ) (Lemma 7.1 (a), s < 1 s<1 s < 1 ). By Lemma 15.7 and Proposition 15.5 (iv), each term of the sum is at most C ε M − 2 q ε M 3 + 2 q ∥ θ i n ∥ L 2 ( T 2 ) C\varepsilon_M^{-2q}\varepsilon_M^{3+2q}\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)} C ε M − 2 q ε M 3 + 2 q ∥ θ in ∥ L 2 ( T 2 ) ; by Lemma 15.8, whose order N t r , w − N m o m , w − n N c o r , w − 1 N_{\mathrm{tr,w}}-N_{\mathrm{mom,w}}-nN_{\mathrm{cor,w}}-1 N tr , w − N mom , w − n N cor , w − 1 is at least N r d , e ≥ N M N_{\mathrm{rd,e}}\ge N_M N rd , e ≥ N M , and as ( 7 4 + δ 2 ) N m o m , w > 2 + 2 q \left(\frac{7}{4}+\frac{\delta}{2}\right)N_{\mathrm{mom,w}}>2+2q ( 4 7 + 2 δ ) N mom , w > 2 + 2 q (item (4)), the term with U M − U ~ M U_M-\widetilde{U}_M U M − U M is at most C ε M − 2 q ε M q + 1 + 2 q ∥ θ i n ∥ L 2 ( T 2 ) C\varepsilon_M^{-2q}\varepsilon_M^{q+1+2q}\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)} C ε M − 2 q ε M q + 1 + 2 q ∥ θ in ∥ L 2 ( T 2 ) ; by Proposition 15.10 (c), the term with U ~ M − Y ♯ \widetilde{U}_M-Y^{\sharp} U M − Y ♯ is at most C ε M − 1 Q e / 2 − q n + 1 ( 1 + s ) ∥ θ i n ∥ L 2 ( T 2 ) = C ε M − 1 5 ∥ θ i n ∥ L 2 ( T 2 ) C\varepsilon_{M-1}^{Q_e/2-q^{n+1}(1+s)}\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)}=C\varepsilon_{M-1}^{5}\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)} C ε M − 1 Q e /2 − q n + 1 ( 1 + s ) ∥ θ in ∥ L 2 ( T 2 ) = C ε M − 1 5 ∥ θ in ∥ L 2 ( T 2 ) . As ε M ≤ ε M − 1 \varepsilon_M\le\varepsilon_{M-1} ε M ≤ ε M − 1 , the sum is at most C ε M − 1 2 ∥ θ i n ∥ L 2 ( T 2 ) C\varepsilon_{M-1}^{2}\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)} C ε M − 1 2 ∥ θ in ∥ L 2 ( T 2 ) .
Proof of Proposition 5.1
Proof of Proposition 5.1. Let M ≥ M ∗ M\ge M_* M ≥ M ∗ , κ ∈ K M \kappa\in K_M κ ∈ K M and t i n ∈ R t_{\mathrm{in}}\in\mathbb{R} t in ∈ R . The times t p r ( M ) , t e ∈ [ t i n , t i n + 1 4 t 0 ] t_{\mathrm{pr}}^{(M)},t_e\in[t_{\mathrm{in}},t_{\mathrm{in}}+\frac{1}{4}t_0] t pr ( M ) , t e ∈ [ t in , t in + 4 1 t 0 ] , the map Π \Pi Π and the admissible seed G G G of level M − 1 M-1 M − 1 on [ t e , t f ] [t_e,t_f] [ t e , t f ] depend only on M M M , κ \kappa κ and t i n t_{\mathrm{in}} t in (Propositions 15.5 , 15.10 and 15.11 ). Given θ i n \theta_{\mathrm{in}} θ in and θ \theta θ , let Y Y Y be the solution of G ′ G' G ′ of Proposition 15.10 and g : = Y ( t e ) g:=Y(t_e) g := Y ( t e ) , a datum of G G G with ∥ g ∥ L ω 2 ( T 2 ) ≤ C ∥ θ i n ∥ L 2 ( T 2 ) \|g\|_{L^2_\omega(\mathbb{T}^2)}\le C\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)} ∥ g ∥ L ω 2 ( T 2 ) ≤ C ∥ θ in ∥ L 2 ( T 2 ) by (b) there. As the systems of G G G and G ′ G' G ′ coincide, the solution U U U of G G G from g g g at time t e t_e t e is Y Y Y on [ t e , t f ] [t_e,t_f] [ t e , t f ] (Proposition 6.2 ). For [ t 1 , t 2 ] [t_1,t_2] [ t 1 , t 2 ] and Ξ \Xi Ξ as in the statement and t ∈ [ t 1 , t 2 ] t\in[t_1,t_2] t ∈ [ t 1 , t 2 ] at which R ( t ) R(t) R ( t ) exists (so t ≥ t e ≥ t p r ( M ) t\ge t_e\ge t_{\mathrm{pr}}^{(M)} t ≥ t e ≥ t pr ( M ) ), Lemma 15.12 , applied to the representation R ( t ) R(t) R ( t ) of U ( t ) = Y ( t ) U(t)=Y(t) U ( t ) = Y ( t ) , and then Lemma 15.13 (a) at i = M i=M i = M give a representation of W M ( t ) Y ♯ ( t ) \mathcal{W}_M(t)Y^{\sharp}(t) W M ( t ) Y ♯ ( t ) of order 3 at wavelength ε M + n \varepsilon_{M+n} ε M + n of size at most C Ξ C\Xi C Ξ , so ∥ W M ( t ) Y ♯ ( t ) ∥ C 0 , s ( T 2 ) ≤ C Ξ \|\mathcal{W}_M(t)Y^{\sharp}(t)\|_{C^{0,s}(\mathbb{T}^2)}\le C\Xi ∥ W M ( t ) Y ♯ ( t ) ∥ C 0 , s ( T 2 ) ≤ C Ξ by Lemma 14.2 (a). With Lemma 15.14 , ∥ θ ( t ) ∥ C 0 , s ( T 2 ) ≤ C Ξ + C ∥ θ i n ∥ L 2 ( T 2 ) \|\theta(t)\|_{C^{0,s}(\mathbb{T}^2)}\le C\Xi+C\|\theta_{\mathrm{in}}\|_{L^2(\mathbb{T}^2)} ∥ θ ( t ) ∥ C 0 , s ( T 2 ) ≤ C Ξ + C ∥ θ in ∥ L 2 ( T 2 ) for almost every t ∈ [ t 1 , t 2 ] t\in[t_1,t_2] t ∈ [ t 1 , t 2 ] .
Index of notation and terms
The main symbols, by topic, then a glossary of terms, each with the place of its definition; local symbols are left out. Throughout, k k k indexes pulses, l l l renewal blocks and m m m levels (a second level is i i i : in a chain, in Definition 11.4 and in the entry; Subsection 14.4 writes j j j and i i i ); M M M is the terminal level (κ ∈ K M \kappa\in K_M κ ∈ K M ).
General notation
Table 2.
Parameters and exponents
Table 3.
@page 158/165 starts_mid=1 ends_mid=1
The reduction and the main results
Scales and windows
The velocity field and its flows
Diffusivities
Families and their norms
A σ , d A_{\sigma,d} A σ , d ; g g g , U N U_N U N coefficient of order d d d of σ \sigma σ , 2 ≤ d ≤ N s r c 2 \le d \le N_{\mathrm{src}} 2 ≤ d ≤ N src ; a datum and the solution (34)
L ω 2 L^2_\omega L ω 2 , L G 2 L^2_G L G 2 , H ρ , ω N H^N_{\rho,\omega} H ρ , ω N weighted norm, marked norm and Sobolev norm of a family (35)
H ρ N H^N_\rho H ρ N ; H ρ , ω N H^N_{\rho,\omega} H ρ , ω N , H ρ , ω ( N v ) H^{(N_v)}_{\rho,\omega} H ρ , ω ( N v ) N N N derivatives at scale ρ \rho ρ ; weighted, at one order or per component (36) , (65)
B v v ′ B_{vv'} B v v ′ , B N , ρ N B^{N}_{N,\rho} B N , ρ N the coupling matrix; the coupling matrix of order N N N at the scale ρ \rho ρ (37) , (68)
F F F , F 0 F^0 F 0 flux forcing and scalar forcing of a system (66)
G ′ G' G ′ ; # v ′ \#v' # v ′ the successor; the number of letters of the word of v ′ v' v ′ § 2.2.2 , (38)
G m G_m G m , U m U_m U m the families of the homogenization cascade and of the entry, and their solutions § 2.2 , § 6
G ~ M \widetilde{G}_M G M , U ~ M \widetilde{U}_M U M G M G_M G M with its sources smoothed by S w S_w S w ; its solution from U M ( t ( M ) ) U_M(t_{(M)}) U M ( t ( M ) ) § 6.2.1
d σ d_\sigma d σ , ρ σ \rho_\sigma ρ σ , fac, maj birth order, birth radius, factor fac ( σ ) \operatorname{fac}(\sigma) fac ( σ ) and majorant maj ( σ ) \operatorname{maj}(\sigma) maj ( σ ) § 5.4.1
A ( t , y , z ) A(t,y,z) A ( t , y , z ) two-scale form, read at z = y / ε m z=y/\varepsilon_m z = y / ε m ; transverse if a function of z ⋅ e ( k ) ⊥ z \cdot e^{\perp}_{(k)} z ⋅ e ( k ) ⊥ § 5.2
∥ A ∥ N , j 1 \lVert A\rVert_{N,j_1} ∥ A ∥ N , j 1 , ∥ A ∥ a n , v \lVert A\rVert_{\mathrm{an},v} ∥ A ∥ an , v Wiener norm on the window of a pulse (109) ; analytic norm (11.9)
The lift, the correctors and the representations
Υ m \Upsilon_m Υ m , Y ♯ Y^\sharp Y ♯ , Corr m \operatorname{Corr}_m Corr m readout; lift of a field Y Y Y of the successor; corrector operator (38)
exp p \operatorname{exp}_p exp p the exponential series truncated after the degree p p p (39)
Y ♭ Y^\flat Y ♭ exp p ( Γ m : ∇ 2 ) Υ m Y \operatorname{exp}_p(\Gamma_m:\nabla^2)\Upsilon_mY exp p ( Γ m : ∇ 2 ) Υ m Y , so that Y ♯ = Corr m Y ♭ Y^\sharp=\operatorname{Corr}_m Y^\flat Y ♯ = Corr m Y ♭ (154)
Π \Pi Π the preparation map Prop. 5.3
χ m , k m \chi^m_{m,k} χ m , k m , χ m \chi^m χ m the corrector of pulse k k k ; the glued corrector (Lemma 10.2) (94)
K ( t , y ) K(t,y) K ( t , y ) ; L 0 L_0 L 0 a conductivity on the window of a pulse; the cell operator § 5.3.2 , (101)
V j V_j V j , M j , d M_{j,d} M j , d correctors of degree j j j and means of the finite lift, per path or summed Def. 10.5
Corr m , k \operatorname{Corr}_{m,k} Corr m , k , M m \mathcal{M}_m M m the corrector operator on pulse k k k ; the mean flux (114)
∣ U ( x ) ∣ ω \lvert U(x)\rvert_\omega ∣ U ( x ) ∣ ω , C λ N C^N_\lambda C λ N pointwise norm of a field of a family; norm, seminorm at scale λ \lambda λ (40)
R R R ; E E E , P ∈ P P \in\mathcal{P} P ∈ P , λ P \lambda_P λ P representation at wavelength λ ∗ \lambda_* λ ∗ ; base, profiles, their wavelengths § 2.3
R ♯ R^\sharp R ♯ the representation of a lifted field, built from R R R Lemmas 15.12 , 15.13 , §15.5
Ξ \Xi Ξ a bound on the sizes of the representations of U ( t ) U(t) U ( t ) Prop. 5.1
The preparation (Section 12)
y 0 y_0 y 0 , K ( t ) K(t) K ( t ) the point at which the conductivity is frozen; the frozen conductivity § 5.6.2
L ω , N ′ 2 L^2_{\omega,N'} L ω , N ′ 2 , H ρ , ω , N ′ N H^N_{\rho,\omega,N'} H ρ , ω , N ′ N the moment norms about y 0 y_0 y 0 (129)
R ξ R_\xi R ξ , Z ξ Z_\xi Z ξ the principal branch and its symbol Prop. 12.4
H m 0 ( ξ ) H_m^0(\xi) H m 0 ( ξ ) , c l o c_{\mathrm{lo}} c lo low-frequency cutoff H ( 2 − ε ˉ m ∣ ξ ∣ / c l o ) H(2-\bar{\varepsilon}_m\lvert\xi\rvert/c_{\mathrm{lo}}) H ( 2 − ε ˉ m ∣ ξ ∣ / c lo ) , c l o c_{\mathrm{lo}} c lo of Prop. 12.5 § 5.7.5
I I I an insertion: a source term with polynomial coefficients (139)
U { j } U^{\{j\}} U { j } the terms of the Duhamel expansion of a family § 5.7.3
U [ 0 ] , … , U [ N ] U^{[0]},\ldots,U^{[N]} U [ 0 ] , … , U [ N ] a Duhamel term of length N N N § 5.7.3
U † U^\dagger U † , U b r † U^\dagger_{\mathrm{br}} U br † a test, a solution of the adjoint fibre equation; the branch test (140)
∣ F ^ b ∣ … ; ξ \lvert\widehat{F}_b\rvert_{\ldots;\xi} ∣ F b ∣ … ; ξ , ∥ g ∥ … ; ξ \lVert g\rVert_{\ldots;\xi} ∥ g ∥ … ; ξ the normalized ξ \xi ξ -jet and the moment data at ξ \xi ξ , of orders N 1 N_1 N 1 , N ′ N' N ′ § 5.7.5
Π b l k \Pi_{\mathrm{blk}} Π blk , Π + \Pi_+ Π + , Π r e n \Pi_{\mathrm{ren}} Π ren models of a Duhamel term over a block, the future block, the renewal §§§ 5.7.5 –§ 5.7.6
H ρ , ω , N ′ + 2 − N H^{-\!N}_{\rho,\omega,N'+2} H ρ , ω , N ′ + 2 − N , N n e g N_{\mathrm{neg}} N neg negative norms, dual to the moment norms; the negative order § 5.7.6
Orders (chosen in the order of Table 11.1 , which gives their conditions)
N c o r N_{\mathrm{cor}} N cor , p p p corrector order (largest corrector degree); truncation order of exp p \operatorname{exp}_p exp p § 2.2.1
N r e t N_{\mathrm{ret}} N ret , N o u t N_{\mathrm{out}} N out order controlled by a cycle; order of the preparations § 2.2.1
N r e p N_{\mathrm{rep}} N rep , N s r c N_{\mathrm{src}} N src , Q Q Q representation order; source order; accuracy exponent § 2.2.1
Z Z Z ; N m o m N_{\mathrm{mom}} N mom power-loss exponent; moment order of the smoothings (179) , § 5.4.1
N g e n N_{\mathrm{gen}} N gen , N l e n N_{\mathrm{len}} N len generations kept; length of the longest word § 5.4.1
N w o r d N_{\mathrm{word}} N word , N p a t h N_{\mathrm{path}} N path number of words; path bound (119) , § 5.4.1
N c o e f N_{\mathrm{coef}} N coef , N a n N_{\mathrm{an}} N an , N o f f N_{\mathrm{off}} N off derivative-bound order; weight order and offset of the analytic norm § 5.4.1
N d e p N_{\mathrm{dep}} N dep , N b u f N_{\mathrm{buf}} N buf depth of the Duhamel expansion; order of the molecular smoothing (123)
N c l N_{\mathrm{cl}} N cl , N r e c N_{\mathrm{rec}} N rec cleanup steps for the cell tail; order of the reconstruction (157)
N c o r , w N_{\mathrm{cor},w} N cor , w , N l e n , w N_{\mathrm{len},w} N len , w corrector order and word length of the weak passages (201)
Q e Q_e Q e , N s r c , e N_{\mathrm{src},e} N src , e , Θ e \Theta_e Θ e accuracy, largest new birth order, width exponent of the terminal cycle (203)
N c o r , e N_{\mathrm{cor},e} N cor , e , p e p_e p e corrector order and truncation order of the terminal cycle (202) N l e n , e , N c l , e N_{\mathrm{len,e}}, N_{\mathrm{cl,e}} N len , e , N cl , e — word length and antidivergence steps of the terminal cycle (202)
N m o m , w , N m o m , e N_{\mathrm{mom,w}}, N_{\mathrm{mom,e}} N mom , w , N mom , e — moment orders of the smoothings at level M M M and of the terminal cycle (205)
N r d , e , N t r , w , N o u t , e N_{\mathrm{rd,e}}, N_{\mathrm{tr,w}}, N_{\mathrm{out,e}} N rd , e , N tr , w , N out , e — terminal recovery order; weak trace order; terminal output order (207)
N i N_i N i — ( M + n − i ) N c o r , w + 5 (M+n-i)N_{\mathrm{cor,w}}+5 ( M + n − i ) N cor , w + 5 , the order at which a field of G i G_i G i is read back down § 6.2
Constants and thresholds
C C C , c c c — generic constants; dependence declared in each section and statement each section
m 0 m_0 m 0 — the top level, the largest threshold on the level in Sections 11–14 § 2.2.1 , § 5.4.1
M ∗ M_* M ∗ , M 1 M_1 M 1 — thresholds for the terminal level (M 1 M_1 M 1 : Proposition 4.1) Prop. 5.1
C s e e d , C a d m , C m a j C_{\mathrm{seed}}, C_{\mathrm{adm}}, C_{\mathrm{maj}} C seed , C adm , C maj — seed constant; bound on majorant and coupling sums; cycle factor § 5.4.1
C Ω , C w i d , C d e r , C j e t C_{\Omega}, C_{\mathrm{wid}}, C_{\mathrm{der}}, C_{\mathrm{jet}} C Ω , C wid , C der , C jet — constants of the invariant (with the c j c_j c j ): smoothings, widths, derivatives § 5.4.1
c a n c_{\mathrm{an}} c an — the strip constant 1 / 20000 1/20000 1/20000 of the chart parameters § 5.5.1
Terms
activity weight — the weight w N a c t ( t ) w_{N_{\mathrm{act}}}(t) w N act ( t ) of the norm ∥ ⋅ ∥ H J ( T 2 ) ; N a c t , t \lVert\cdot\rVert_{H_J(T^2);N_{\mathrm{act}},t} ∥ ⋅ ∥ H J ( T 2 ) ; N act , t (144)
admissible family — a family descended by cycles from an admissible seed Def. 11.4
admissible seed — a seed that satisfies Definition 11.2 Def. 11.2
allowed chain — a chain of levels at least m c m_c m c whose centres satisfy ∣ y ( i ) ∣ 1 < 3 c a n ε i / 4 \lvert y^{(i)}\rvert_1<3c_{\mathrm{an}}\varepsilon_i/4 ∣ y ( i ) ∣ 1 < 3 c an ε i /4 (11.7)
averaged flux — J m ν \mathbf{J}^{\nu}_m J m ν , the spatial average of the flux of the correctors of the pulses (18)
base, profile — the smooth part and the oscillating parts of a representation § 2.3
Bloch sum — B g ( ξ , z ) \mathcal{B}g(\xi,z) B g ( ξ , z ) , the sum of a field g g g about y 0 y_0 y 0 (130)
branch test — the test from 1 1 1 at a quiet time (off-branch: orthogonal to the principal branch) § 5.7.3
cell average — the average in z z z of a two-scale form of a source, carried by its transport Def. 11.3
cell inverse — the solution of a cell problem across a pulse; χ K \chi_K χ K , the corrector in K \mathbf{K} K Lem. 10.4
cell solution — a function built from two-dimensional cell problems of a family Def. 11.1
cell tail — T k T_k T k , the operator left by the correctors of pulse k k k , of degrees above N c o r N_{\mathrm{cor}} N cor Lem. 13.2
cell time — ε m 2 / κ m \varepsilon_m^2/\kappa_m ε m 2 / κ m , the time diffusion takes to cross a period cell § 1.3.2
child, parent — a component created by a cycle; the component it comes from § 2.2.2
comparison — the bound on U − Y ♯ U-Y^{\sharp} U − Y ♯ in a cycle Prop. 5.3
component energy — the quantity (155) of a field, of an order, at a scale, with a diffusivity (155)
contrast factor — 1 + ε ˉ m / ε m = 1 + ( κ m − 1 / κ m ) 1 / 2 1+\bar{\varepsilon}_m/\varepsilon_m=1+(\kappa_{m-1}/\kappa_m)^{1/2} 1 + ε ˉ m / ε m = 1 + ( κ m − 1 / κ m ) 1/2 § 5.7.3
corrector amplitude — ε ˉ m \bar{\varepsilon}_m ε ˉ m , the size of the correctors of level m m m per derivative §5.1
covariance — Γ m ( t ) \Gamma_m(t) Γ m ( t ) , the accumulated surplus of the averaged flux over κ m − 1 I \kappa_{m-1}I κ m − 1 I (39)
covariance letter — a letter for the means along a path conjugated by powers of ∓ Γ m : ∇ 2 \mp\Gamma_m:\nabla^2 ∓ Γ m : ∇ 2 Def. 11.3
cycle — one homogenization step, from level m m m to level m − 1 m-1 m − 1 Prop. 5.3
defect, defect letter — the mean flux along the empty paths minus J m κ ( t ) ∇ \mathbf{J}^{\kappa}_{m}(t)\nabla J m κ ( t ) ∇ ; its letter § 5.3.4
entry — from θ \theta θ at level M + n M+n M + n up to the seed, and the lift back down Prop. 5.1
family — a coupled system of equations of one level § 2.2.2
fibre equation — the equation of the periodic part of a Bloch wave at the frequency ξ \xi ξ (126)
finite lift — the correctors and means of a system on the window of one pulse Def. 10.5
frontier flux — F f r F^{\mathrm{fr}} F fr , the readout of the flux into the omitted components Lem. 13.8
generation — gen ( v ) \operatorname{gen}(v) gen ( v ) , the number of cycles that created the component v v v § 5.4.1
halving interval — a few cell times in which solutions and tests halve; # { t ′ , t ′ ′ } \#\{t',t''\} # { t ′ , t ′′ } counts them § 5.7.4
homogenization cascade — the entry and the cycles, up the levels, forward in time § 2.2
invariant — the bounds satisfied by the admissible families Def. 11.1
lift — Y ♯ Y^{\sharp} Y ♯ , from a field of the successor to the components of G G G (38)
marked amplitude, A m A_m A m — the norm ∥ U m ∥ L ( t m , t ] ∞ ; L G m 2 ( T 2 ) \lVert U_m\rVert_{L^\infty_{(t_{m},t]};L^2_{G_m}(\mathbb{T}^2)} ∥ U m ∥ L ( t m , t ] ∞ ; L G m 2 ( T 2 ) , carried up the levels § 2.2
marker — the number Ψ v \Psi_v Ψ v , which grows with the generation; marked norm: ∥ ⋅ ∥ L G 2 \lVert\cdot\rVert_{L^2_{G}} ∥ ⋅ ∥ L G 2 § 2.2.2
mass — the number ω v \omega_v ω v of a component § 2.2.2
mean letter — the letter for the means of the lift along a path Def. 11.3
moment term — U [ j , i ] U^{[j,i]} U [ j , i ] , a term U [ j ] U^{[j]} U [ j ] of a Duhamel term times ( ( y − y 0 ) / ρ ( m ) ) i ((y-y_0)/\rho(m))^i (( y − y 0 ) / ρ ( m ) ) i § 5.7.3 outer domain D v D_v D v with 2 in place of res ( v ) \operatorname{res}(v) res ( v ) ; its real points are the allowed chains (11.9)
passage a cycle read back down on representations, at a fixed time Prop. 5.4
path sources σ 1 ⋯ σ i \sigma_1\cdots\sigma_i σ 1 ⋯ σ i , each starting where the previous one ends § 5.3.2
preparation Π g \Pi_g Π g , close to the solution at t p r t_{\mathrm{pr}} t pr and smooth at the scale ρ ( m ) \rho(m) ρ ( m ) Prop. 5.3
price max { 1 , ( ρ σ / r σ ) d σ − 2 } \max\{1,(\rho_\sigma/r_\sigma)^{d_\sigma-2}\} max { 1 , ( ρ σ / r σ ) d σ − 2 } , which enters the priced majorant matrix (117)
pulse an odd slot k k k of level m m m , active for a time of about τ m \tau_m τ m around k τ m k\tau_m k τ m (8)
quiet margin the pulse-free end of a renewal block, of length τ m \tau_m τ m (12)
readout Υ m \Upsilon_m Υ m , which adds each new component to its parent (38)
regularity cascade the passages and the lifts, back down the levels, at a fixed time Prop. 5.2
renewal block [ ( l − 1 2 ) h m , ( l + 1 2 ) h m ] [(l-\frac{1}{2})h_m,(l+\frac{1}{2})h_m] [( l − 2 1 ) h m , ( l + 2 1 ) h m ] , with the Lagrangian coordinates of b m − 1 b_{m-1} b m − 1 (10)
renewal chart y ↦ X m − 1 ( t , y , l h m ) y\mapsto X_{m-1}(t,y,lh_m) y ↦ X m − 1 ( t , y , l h m ) , inverse to the Lagrangian label y y y § 4.1.3
representation a base plus profiles, each measured at its wavelength § 2.3
reserve res ( v ) \operatorname{res}(v) res ( v ) , which fixes the radii of the domain D v D_v D v of the analytic norm (11.9)
retained trace the datum of G ′ G' G ′ obtained by inverting the truncated exponential § 2.2.3
seed a family whose components have generation 0 and marker 1 § 2.2.2
shear displacement the map y ↦ y + ( ∫ t t ′ ζ m , k ) ∇ ⊥ ψ m , k ( y ) y\mapsto y+(\int_t^{t'}\zeta_{m,k})\nabla^\perp\psi_{m,k}(y) y ↦ y + ( ∫ t t ′ ζ m , k ) ∇ ⊥ ψ m , k ( y ) of (58) § 4.1.3
size size λ ∗ ( R ) \operatorname{size}_{\lambda^*}(R) size λ ∗ ( R ) , which bounds the C 0 , s C^{0,s} C 0 , s norm (41)
source a coupling term between two components, with a smoothing § 2.2.2
successor the family of level m − 1 m-1 m − 1 that a cycle produces Def. 11.3
terminal cycle the cycle of the entry at level M M M , which passes to κ M − 1 \kappa_{M-1} κ M − 1 § 2.3.1
terminal level the level M M M of the band K M \mathcal{K}_M K M that contains κ \kappa κ (20)
terminal lift the lift of a field of G ′ G' G ′ with the anchored covariance Γ M ( t ) − Γ M ( t p r M ) \Gamma_M(t)-\Gamma_M(t_{\mathrm{pr}}^{M}) Γ M ( t ) − Γ M ( t pr M ) (208)
terminal successor the successor G ′ G' G ′ of G ~ M \widetilde{G}_M G M , of level M − 1 M-1 M − 1 Def. 15.9
test a solution of the adjoint fibre equation, paired with Duhamel terms (140)
transport coefficients carried from y y y to x x x (Piola identity, chain rule) (11.8)
weak family the family G i − 1 G_{i-1} G i − 1 , M < i ≤ M + n M<i\le M+n M < i ≤ M + n , made from G i G_i G i at the diffusivity κ \kappa κ Def. 15.3
weak lift Corr i w ( t ) Υ i w \operatorname{Corr}^{w}_{i}(t)\Upsilon^{w}_{i} Corr i w ( t ) Υ i w , weak readout, then correctors; W i ( t ) W_i(t) W i ( t ) , the composed lifts § 6.2
weak passage a step of the entry at a level finer than M M M , at the diffusivity κ \kappa κ § 2.3.1
word, letter the label of a created component; a letter names a source term § 2.2.2
zone the frequencies with ε m ∣ ξ ∣ ∞ ≤ π \varepsilon_m|\xi|_\infty\le\pi ε m ∣ ξ ∣ ∞ ≤ π , the cell frequencies § 5.6.2
zone part, F l o F^{\mathrm{lo}} F lo F z o n e F^{\mathrm{zone}} F zone , the part of a field at the frequencies ε m ∣ ξ ∣ < π / 2 \varepsilon_m|\xi|<\pi/2 ε m ∣ ξ ∣ < π /2 ; F ~ l o : = H m l o F ~ \widetilde{F}^{\mathrm{lo}}:=H_m^{\mathrm{lo}}\widetilde{F} F lo := H m lo F § 5.7.6
Acknowledgments. S.A. was supported by NSF grant DMS-2350340 and by the European Research Council (ERC) under the European Union’s Horizon Europe research and innovation programme, grant agreement number 101200828. V.V. was supported by the Collaborative NSF grant DMS-2307681 and by a Simons Investigator Award.
References