Introduction
We consider a nonlocal curve evolution approximating the motion of an inextensible elastic filament immersed in a Stokes fluid in R 3 \mathbb{R}^3 R 3 . Our model belongs to a family of curve evolution equations arising in the following way. The thin filament is treated as a rod with rigid cross sections of constant radius 0 < ϵ ≪ 1 0 < \epsilon\ll1 0 < ϵ ≪ 1 whose centerline X ( s , t ) : T × [ 0 , T ] → R 3 \mathbf{X}(s,t): \mathbb{T} \times[0,T] \to\mathbb{R}^3 X ( s , t ) : T × [ 0 , T ] → R 3 for T : = R / Z \mathbb{T} := \mathbb{R}/\mathbb{Z} T := R / Z deforms subject to a 1D elasticity law (Figure 1). Here the elastic force density
f ( s ) = ( X s s s − τ X s ) s , ( ⋅ ) s : = ∂ ∂ s , (1) \mathbf{f}(s) = \left(\mathbf{X}_{sss} - \tau\mathbf{X}_s\right)_s,\qquad(\cdot)_s := \frac{\partial}{\partial s}, \tag*{(1)} f ( s ) = ( X sss − τ X s ) s , ( ⋅ ) s := ∂ s ∂ , (1)
along the filament centerline comes from Euler-Bernoulli beam theory [3 , 4 , 14 , 40 , 42 , 43 ] , with the tension τ ( s , t ) \tau(s,t) τ ( s , t ) serving as a Lagrange multiplier to enforce the local inextensibility constraint ∣ X s ∣ 2 = 1 \lvert\mathbf{X}_s\rvert^2 = 1 ∣ X s ∣ 2 = 1 . The elastic forcing is coupled to the filament motion via a force-to-velocity map
L ϵ ( X ) : f ( s ) ↦ v ( s ) (2) \mathcal{L}_{\epsilon}(\mathbf{X}):\ \mathbf{f}(s) \mapsto\mathbf{v}(s) \tag*{(2)} L ϵ ( X ) : f ( s ) ↦ v ( s ) (2)
which incorporates the effects of the surrounding Stokes fluid. Models in this family are essentially distinguished by the level of detail with which the fluid effects are included in the map L ϵ ( X ) \mathcal{L}_{\epsilon}(\mathbf{X}) L ϵ ( X ) . The position of the curve then evolves according to an equation of the form
∂ X ∂ t = − L ϵ ( X ) [ ( X s s s − τ X s ) s ] , ∣ X s ∣ 2 = 1. (3) \frac{\partial\mathbf{X}}{\partial t} = -\mathcal{L}_{\epsilon}(\mathbf{X})\left[(\mathbf{X}_{sss} - \tau\mathbf{X}_s)_s\right],\qquad\lvert\mathbf{X}_s\rvert^2 = 1. \tag*{(3)} ∂ t ∂ X = − L ϵ ( X ) [ ( X sss − τ X s ) s ] , ∣ X s ∣ 2 = 1. (3)
Figure 1. We consider X ( s , t ) \mathbf{X}(s,t) X ( s , t ) as the centerline of an immersed 3D filament with constant cross sectional radius 0 < ϵ ≪ 1 0 < \epsilon\ll1 0 < ϵ ≪ 1 .
The simplest choice of force-to-velocity map which still incorporates meaningful hydrodynamic effects is the resistive force theory (local slender body theory) [12 , 16 , 18 , 37 ] approximation
L ϵ , R F T ( X ) = ∣ log ϵ ∣ 4 π ( I + X s ⊗ X s ) . (4) \mathcal{L}_{\epsilon,\mathrm{RFT}}(\mathbf{X}) = \frac{|\log\epsilon|}{4\pi}\left(\mathbf{I}+\mathbf{X}_s\otimes\mathbf{X}_s\right). \tag*{(4)} L ϵ , RFT ( X ) = 4 π ∣ log ϵ ∣ ( I + X s ⊗ X s ) . (4)
The map L ϵ , R F T \mathcal{L}_{\epsilon,\mathrm{RFT}} L ϵ , RFT includes only the leading order O ( ∣ log ϵ ∣ ) O(|\log\epsilon|) O ( ∣ log ϵ ∣ ) effect of the surrounding viscous fluid on the filament, resulting in a local force-to-velocity operator. The main feature of L ϵ , R F T \mathcal{L}_{\epsilon,\mathrm{RFT}} L ϵ , RFT is a drag anisotropy weighting the viscous drag in the tangential (X s \mathbf{X}_s X s ) direction along the filament twice as much as in the normal directions. This drag anisotropy is a simple but important nonlinear effect that, among other consequences, allows for net displacement (swimming) of the filament. Resistive force theory models of elastohydrodynamics are therefore useful for understanding undulatory swimming at low Reynolds number [7 , 5 , 8 , 9 , 15 , 25 , 23 , 22 , 39 , 24 , 29 , 34 ] . These local dynamics also serve as a useful testing ground for efficient numerical implementations of inextensibility [27 , 13 , 41 ] .
Resistive force theory is an important member of the immersed filament modeling hierarchy, but for certain applications, its treatment of hydrodynamic effects is overly simplistic. A natural, more detailed candidate for a force-to-velocity map L ϵ \mathcal{L}_{\epsilon} L ϵ is nonlocal slender body theory1 [19 , 17 , 40 ] :
L ϵ , n l o c [ f ] ( s ) = 1 8 π [ ( I − 3 X s ⊗ X s ) − 2 ( I + X s ⊗ X s ) log ( π ϵ 4 ) ] f ( s ) + 1 8 π ∫ T [ ( I ∣ X ^ ∣ + X ^ ⊗ X ^ ∣ X ^ ∣ 3 ) f ( s ′ ) − I + X s ⊗ X s ∣ sin ( π ( s − s ′ ) / π ) ∣ f ( s ) ] d s ′ . (5) \begin{aligned}
\mathcal{L}_{\epsilon,\mathrm{nloc}}[f](s)={}&\frac{1}{8\pi}\left[\left(\mathbf{I}-3\mathbf{X}_s\otimes\mathbf{X}_s\right)-2\left(\mathbf{I}+\mathbf{X}_s\otimes\mathbf{X}_s\right)\log\left(\frac{\pi\epsilon}{4}\right)\right]f(s)\\
&+\frac{1}{8\pi}\int_{\mathbb{T}}\left[\left(\frac{\mathbf{I}}{|\widehat{\mathbf{X}}|}+\frac{\widehat{\mathbf{X}}\otimes\widehat{\mathbf{X}}}{|\widehat{\mathbf{X}}|^3}\right)f(s')-\frac{\mathbf{I}+\mathbf{X}_s\otimes\mathbf{X}_s}{\left|\sin\left(\pi(s-s')/\pi\right)\right|}f(s)\right]\,ds'. \tag*{(5)}
\end{aligned} L ϵ , nloc [ f ] ( s ) = 8 π 1 [ ( I − 3 X s ⊗ X s ) − 2 ( I + X s ⊗ X s ) log ( 4 π ϵ ) ] f ( s ) + 8 π 1 ∫ T [ ( ∣ X ∣ I + ∣ X ∣ 3 X ⊗ X ) f ( s ′ ) − ∣ sin ( π ( s − s ′ ) / π ) ∣ I + X s ⊗ X s f ( s ) ] d s ′ . (5)
where X ^ = X ( s ) − X ( s ′ ) \widehat{\mathbf{X}}=\mathbf{X}(s)-\mathbf{X}(s') X = X ( s ) − X ( s ′ ) . We mention this operator due to its numerical utility; however, due to well-known issues [11 , 40 , 38 , 28 ] at high wavenumbers ∼ 1 ϵ \sim\frac{1}{\epsilon} ∼ ϵ 1 , an evolution of the form (3) using L ϵ , n l o c \mathcal{L}_{\epsilon,\mathrm{nloc}} L ϵ , nloc is ill-posed, even at analytic regularity.
We thus turn to arguably the most detailed choice of force-to-velocity map suitable for (3) : the slender body Neumann-to-Dirichlet (NtD) map , introduced by the author together with Mori and Spirn in [30 , 31 ] . The slender body NtD map comes from solving the slender body boundary value problem for 3D Stokes flow about a filament of radius ϵ > 0 \epsilon>0 ϵ > 0 using force data defined on a curve. The full definition of the slender body NtD map, which we will denote by L ϵ , N t D \mathcal{L}_{\epsilon,\mathrm{NtD}} L ϵ , NtD , appears in section 1.3. The evolution (1.3) using L ϵ , N t D \mathcal{L}_{\epsilon,\mathrm{NtD}} L ϵ , NtD gives rise to the slender body free boundary problem, a curve evolution in which the fluid surrounding the 3D filament exactly satisfies a Stokes boundary value problem at each moment in time. In [35 , 33 ] , we showed that the principal behavior of L ϵ , N t D \mathcal{L}_{\epsilon,\mathrm{NtD}} L ϵ , NtD about a curved filament is captured by the slender body NtD map about a straight cylinder, for which we derived an explicit symbol in [28 ] . From this decomposition, in [36 ] , we showed that the slender body free boundary problem is locally well-posed in C 4 , α ( T ) C^{4,\alpha}(\mathbb{T}) C 4 , α ( T ) . Questions of global well-posedness and long-time convergence to equilibrium remain completely open.
This brings us to the curve evolution studied in this paper. Here we consider a natural and novel intermediate model of the form (1.3) between the full slender body free boundary problem (1.26) and the classical resistive force theory model (1.18). Our choice of the force-to-velocity map, which we will denote by L ϵ \mathcal{L}_{\epsilon} L ϵ , is a pseudodifferential operator whose symbol in the tangential and normal directions to the filament centerline is the explicit Fourier multiplier corresponding to the exact slender body NtD map about a straight cylinder. In particular, we simply apply the straight map to a curved filament (see Definition 1.1). At low wavenumbers, the map behaves like multiplication by ∣ log ϵ ∣ |\log\epsilon| ∣ log ϵ ∣ , with the tangential direction along the filament weighted twice as much as the normal directions. At high wavenumbers ∣ k ∣ ≳ 1 ϵ |k| \gtrsim\frac{1}{\epsilon} ∣ k ∣ ≳ ϵ 1 , the map behaves like multiplication by ( ϵ ∣ k ∣ ) − 1 (\epsilon|k|)^{-1} ( ϵ ∣ k ∣ ) − 1 on the Fourier side, i.e., a (scaled) inverse derivative. In a sense, the force-to-velocity map L ϵ \mathcal{L}_{\epsilon} L ϵ interpolates between resistive force theory at low wavenumbers and the full slender body NtD map at high wavenumbers.
Our aim is twofold: (I.) We seek to probe the global well-posedness question for the full slender body free boundary problem (1.26) by replacing L ϵ , N t D \mathcal{L}_{\epsilon,\mathrm{NtD}} L ϵ , NtD with its principal part. Here we prove global well-posedness for (1.3) using the map L ‾ ϵ \overline{\mathcal{L}}_{\epsilon} L ϵ in the natural energy space. One caveat is that we treat (1.3) with L ‾ ϵ \overline{\mathcal{L}}_{\epsilon} L ϵ as a pure curve evolution and do not worry about self-intersection of the filament. While self-intersection would lead to breakdown of the full free boundary problem, we can continue to make sense of the model with L ‾ ϵ \overline{\mathcal{L}}_{\epsilon} L ϵ even if the filament passes through itself. Self-intersection is inherently a low-wavenumber phenomenon, and at low wavenumbers, L ‾ ϵ \overline{\mathcal{L}}_{\epsilon} L ϵ looks like resistive force theory rather than L ϵ , N t D \mathcal{L}_{\epsilon,\mathrm{NtD}} L ϵ , NtD . However, our global well-posedness result for (1.3) using L ‾ ϵ \overline{\mathcal{L}}_{\epsilon} L ϵ loosely suggests that if the large-scale geometry of the full slender body free boundary evolution remains under control, then the full free boundary problem should be globally well-posed.
(II.) In addition, we seek to illustrate how resistive force theory dynamics using L ϵ , R F T \mathcal{L}_{\epsilon,\mathrm{RFT}} L ϵ , RFT emerge from more detailed models in the singular limit ϵ → 0 + \epsilon\to0^{+} ϵ → 0 + . We prove that the evolution (1.3) with L ‾ ϵ \overline{\mathcal{L}}_{\epsilon} L ϵ converges to the resistive force theory evolution with L ϵ , R F T \mathcal{L}_{\epsilon,\mathrm{RFT}} L ϵ , RFT as the radius parameter ϵ → 0 \epsilon\to0 ϵ → 0 , thereby establishing the first rigorous dynamical connection between different levels of the curve evolution hierarchy. The convergence is at an expected but very slow logarithmic-in-ϵ \epsilon ϵ rate, since (1.4) only captures the very leading order local effects of the Stokes fluid on the filament. The slow convergence does make the proof delicate, as there is no room for waste.
In [1 ] , we recognize (1.3) with L ϵ , R F T \mathcal{L}_{\epsilon,\mathrm{RFT}} L ϵ , RFT as a gradient flow of the filament bending energy 1 2 ∫ T ∣ X s s ∣ 2 d s \frac{1}{2}\int_{\mathbb{T}}|\mathbf{X}_{ss}|^{2}\,ds 2 1 ∫ T ∣ X ss ∣ 2 d s , which allows us to show long-time convergence to 3D Euler elasticae, the critical points of the bending energy. The evolution with L ‾ ϵ \overline{\mathcal{L}}_{\epsilon} L ϵ is also a gradient flow of the bending energy with respect to the metric
⟨ V , U ⟩ : = ⟨ L ‾ ϵ − 1 V , U ⟩ L 2 ( T ) (6) \langle V,U\rangle:= \langle\overline{\mathcal{L}}_{\epsilon}^{-1}V,U\rangle_{L^{2}(\mathbb{T})} \tag*{(6)} ⟨ V , U ⟩ := ⟨ L ϵ − 1 V , U ⟩ L 2 ( T ) (6)
defined on perturbations V , U V,U V , U of an inextensible curve X \mathbf{X} X . We may thus frame our convergence result as that of two different gradient flows for the same energy whose metrics converge as ϵ → 0 \epsilon\to0 ϵ → 0 .
The force-to-velocity map
1.1. The force-to-velocity map. We now define the force-to-velocity map L ‾ ϵ \overline{\mathcal{L}}_{\epsilon} L ϵ used in (1.3). Given h ( s ) : T → R 3 h(s):\mathbb{T}\to\mathbb{R}^{3} h ( s ) : T → R 3 defined along a filament centerline X \mathbf{X} X , we denote the projection of h h h onto the tangential and normal directions about X X X as
P X s h = ( X s ⋅ h ) X s , P X s ⊥ h = h − P X s h . (7) \mathbb{P}_{X_s}h=(X_s\cdot h)X_s,\qquad\mathbb{P}_{X_s}^{\perp}h=h-\mathbb{P}_{X_s}h. \tag*{(7)} P X s h = ( X s ⋅ h ) X s , P X s ⊥ h = h − P X s h . (7)
In addition, we denote the Fourier transform of h h h by
F [ h ( s ) ] ( k ) = ∫ T h ( s ) e i 2 π k s d s . (8) \mathcal{F}[h(s)](k)=\int_{\mathbb{T}}h(s)e^{i2\pi ks}\,ds. \tag*{(8)} F [ h ( s )] ( k ) = ∫ T h ( s ) e i 2 π k s d s . (8)
Given a scalar-valued Fourier multiplier m ( k ) m(k) m ( k ) , we will use the notation T m T_m T m to denote the operator
T m h = F − 1 [ m F [ h ] ] . (9) T_mh=\mathcal{F}^{-1}[m\mathcal{F}[h]]. \tag*{(9)} T m h = F − 1 [ m F [ h ]] . (9)
Definition 1.1 (Force to velocity map). Given X ∈ H 2 ( T ) X\in H^2(\mathbb{T}) X ∈ H 2 ( T ) and a line force density f ∈ L 2 ( T ) f\in L^2(\mathbb{T}) f ∈ L 2 ( T ) , the force-to-velocity map $\overline{\mathcal{L}}_\epsilon is given by
L ‾ ϵ [ f ] = P X s T m ϵ t ( P X s f ) + P X s ⊥ T m ϵ n ( P X s ⊥ f ) , (10) \overline{\mathcal{L}}_\epsilon[f]=\mathbb{P}_{X_s}T_{m_\epsilon^t}(\mathbb{P}_{X_s}f)+\mathbb{P}_{X_s}^{\perp}T_{m_\epsilon^n}(\mathbb{P}_{X_s}^{\perp}f), \tag*{(10)} L ϵ [ f ] = P X s T m ϵ t ( P X s f ) + P X s ⊥ T m ϵ n ( P X s ⊥ f ) , (10) where, for k ≠ 0 k\ne0 k = 0 , the multipliers m ϵ t ( k ) m_\epsilon^t(k) m ϵ t ( k ) , m ϵ n ( k ) m_\epsilon^n(k) m ϵ n ( k ) are given by
m ϵ t ( k ) = 2 K 0 K 1 + 2 π ϵ ∣ k ∣ ( K 0 2 − K 1 2 ) 8 π 2 ϵ ∣ k ∣ K 1 2 , (11) m_\epsilon^t(k)=\frac{2K_0K_1+2\pi\epsilon|k|(K_0^2-K_1^2)}{8\pi^2\epsilon|k|K_1^2}, \tag*{(11)} m ϵ t ( k ) = 8 π 2 ϵ ∣ k ∣ K 1 2 2 K 0 K 1 + 2 π ϵ ∣ k ∣ ( K 0 2 − K 1 2 ) , (11) m ϵ n ( k ) = 2 K 0 K 1 K 2 + 2 π ϵ ∣ k ∣ ( K 1 2 ( K 0 + K 2 ) − 2 K 0 2 K 2 ) 4 π 2 ϵ ∣ k ∣ ( 4 K 1 2 K 2 + 2 π ϵ ∣ k ∣ K 1 ( K 1 2 − K 0 K 2 ) ) , (12) m_\epsilon^n(k)=\frac{2K_0K_1K_2+2\pi\epsilon|k|(K_1^2(K_0+K_2)-2K_0^2K_2)}{4\pi^2\epsilon|k|(4K_1^2K_2+2\pi\epsilon|k|K_1(K_1^2-K_0K_2))}, \tag*{(12)} m ϵ n ( k ) = 4 π 2 ϵ ∣ k ∣ ( 4 K 1 2 K 2 + 2 π ϵ ∣ k ∣ K 1 ( K 1 2 − K 0 K 2 )) 2 K 0 K 1 K 2 + 2 π ϵ ∣ k ∣ ( K 1 2 ( K 0 + K 2 ) − 2 K 0 2 K 2 ) , (12) where K j = K j ( 2 π ϵ ∣ k ∣ ) K_j=K_j(2\pi\epsilon|k|) K j = K j ( 2 π ϵ ∣ k ∣ ) , j = 0 , 1 , 2 j=0,1,2 j = 0 , 1 , 2 , are j th j^{\text{th}} j th order modified Bessel functions of the second kind.
For k = 0 k=0 k = 0 , we define
m ϵ t ( 0 ) = ∣ log ϵ ∣ 2 π , m ϵ n ( 0 ) = ∣ log ϵ ∣ 4 π . (13) m_\epsilon^t(0)=\frac{|\log\epsilon|}{2\pi},\qquad m_\epsilon^n(0)=\frac{|\log\epsilon|}{4\pi}. \tag*{(13)} m ϵ t ( 0 ) = 2 π ∣ log ϵ ∣ , m ϵ n ( 0 ) = 4 π ∣ log ϵ ∣ . (13) We provide plots of the inverse multipliers m ϵ t ( k ) − 1 m_\epsilon^t(k)^{-1} m ϵ t ( k ) − 1 and m ϵ n ( k ) − 1 m_\epsilon^n(k)^{-1} m ϵ n ( k ) − 1 in Figure 2 to more clearly display the transition from logarithmic behavior ( m ϵ t ( k ) − 1 , m ϵ n ( k ) − 1 ∼ ∣ log ( ϵ ∣ k ∣ ) ∣ − 1 ) (m_\epsilon^t(k)^{-1},m_\epsilon^n(k)^{-1}\sim|\log(\epsilon|k|)|^{-1}) ( m ϵ t ( k ) − 1 , m ϵ n ( k ) − 1 ∼ ∣ log ( ϵ ∣ k ∣ ) ∣ − 1 ) at low wavenumbers to linear growth ( m ϵ t ( k ) − 1 , m ϵ n ( k ) − 1 ∼ ϵ ∣ k ∣ ) (m_\epsilon^t(k)^{-1},m_\epsilon^n(k)^{-1}\sim\epsilon|k|) ( m ϵ t ( k ) − 1 , m ϵ n ( k ) − 1 ∼ ϵ ∣ k ∣ ) at high wavenumbers.
Figure 2. Plots of the inverse tangential (left) and normal (right) direction multipliers m ϵ t ( k ) − 1 m_\epsilon^t(k)^{-1} m ϵ t ( k ) − 1 and m ϵ n ( k ) − 1 m_\epsilon^n(k)^{-1} m ϵ n ( k ) − 1 versus 2 π ϵ ∣ k ∣ 2\pi\epsilon|k| 2 π ϵ ∣ k ∣ . Here the inverses are used to more clearly display the linear (∼ ϵ ∣ k ∣ \sim\epsilon|k| ∼ ϵ ∣ k ∣ ) behavior at high wavenumbers ∣ k ∣ ≳ 1 ϵ |k|\gtrsim\frac{1}{\epsilon} ∣ k ∣ ≳ ϵ 1 .
For k ≠ 0 k\ne0 k = 0 , the multipliers m ϵ t ( k ) m_\epsilon^t(k) m ϵ t ( k ) and m ϵ n ( k ) m_\epsilon^n(k) m ϵ n ( k ) are precisely the tangential and normal eigenvalues, respectively, of the slender body Neumann-to-Dirichlet map in the special filament geometry of a straight cylinder.
The regularization (1.13) for the k = 0 k=0 k = 0 mode is used since the expressions (1.11), (1.12) blow up like ∣ log ( ϵ ∣ k ∣ ) ∣ |\log(\epsilon|k|)| ∣ log ( ϵ ∣ k ∣ ) ∣ as k → 0 k\to0 k → 0 . This stems from Stokes’ paradox for the straight (periodic/infinite) cylinder, for which these multipliers are calculated, and is not expected to be a physically realistic or meaningful approximation of the zero-mode behavior for a curved, closed filament. In particular, the sense in which L ‾ ϵ \overline{\mathcal{L}}_\epsilon L ϵ serves as an approximation for the ‘full’ slender body NtD map (see section 1.3) is primarily through high wavenumber behavior. Here the choice of regularization for the k = 0 k=0 k = 0 mode is chosen to match with resistive force theory (1.4) and thus facilitate the convergence result of Theorem 1.3. We emphasize that L ‾ ϵ \overline{\mathcal{L}}_\epsilon L ϵ in a sense interpolates between resistive force theory at low wavenumbers and the full slender body NtD map at high wavenumbers.
Main theorem statements
Through multiplier and tension bounds outlined in the following sections, we show that the evolution
∂ X ∂ t = − L ‾ ϵ ( X ) [ ( X s s s − τ X s ) s ] , ∣ X s ∣ 2 = 1. (14) \frac{\partial X}{\partial t}=-\overline{\mathcal{L}}_\epsilon(X)\left[(X_{sss}-\tau X_s)_s\right],\qquad|X_s|^2=1. \tag*{(14)} ∂ t ∂ X = − L ϵ ( X ) [ ( X sss − τ X s ) s ] , ∣ X s ∣ 2 = 1. (14)
is a third-order, semilinear, parabolic equation for X X X , coupled with an elliptic equation for the filament tension τ \tau τ . It therefore shares features of both resistive force theory and the full slender body free boundary problem.
In section 3, we show:
Theorem 1.2 (Global well-posedness). Given ϵ > 0 \epsilon>0 ϵ > 0 and an initial curve X i n ∈ H 2 ( T ) X^{\mathrm{in}}\in H^2(\mathbb{T}) X in ∈ H 2 ( T ) , the evolution (1.14) with $\overline{\mathcal{L}}_\epsilon as in Definition 1.1 admits a unique global solution
X ∈ C ( [ 0 , + ∞ ) ; H 2 ) ∩ L t , l o c 2 H s 7 / 2 ( [ 0 , + ∞ ) × T ) (15) X\in C([0,+\infty);H^2)\cap L^2_{t,\mathrm{loc}}H_s^{7/2}([0,+\infty)\times\mathbb{T}) \tag*{(15)} X ∈ C ([ 0 , + ∞ ) ; H 2 ) ∩ L t , loc 2 H s 7/2 ([ 0 , + ∞ ) × T ) (15) with initial data X i n X^{\mathrm{in}} X in .
The starting point for Theorem 1.2 is the energy identity
1 2 ∂ t ∫ T ∣ X s ∣ 2 d s = − ∫ T ( X s s s − τ X s ) s ⋅ L ‾ ϵ [ ( X s s s − τ X s ) s ] d s , (16) \frac{1}{2}\partial_t\int_{\mathbb{T}}|X_s|^2\,ds=-\int_{\mathbb{T}}(X_{sss}-\tau X_s)_s\cdot\overline{\mathcal{L}}_\epsilon[(X_{sss}-\tau X_s)_s]\,ds, \tag*{(16)} 2 1 ∂ t ∫ T ∣ X s ∣ 2 d s = − ∫ T ( X sss − τ X s ) s ⋅ L ϵ [( X sss − τ X s ) s ] d s , (16)
which is explored in greater detail in sections 2.2 and 3.5. The main difficulty is that it is not immediately clear that the right hand side dissipation controls the curve X s s s s X_{ssss} X ssss alone, due to the presence of the tension. In addition, necessary ingredients in the proof of local well-posedness in the energy space are a priori estimates for the tension, which in section 3.2 is shown to satisfy an elliptic equation at each time. Interestingly, and somewhat surprisingly, the tension is better behaved than its counterpart in resistive force theory, which is critical in a certain sense.2 This may be due to the more physically realistic behavior of L ‾ ϵ \overline{\mathcal{L}}_\epsilon L ϵ at high wavenumbers for capturing fluid effects on immersed filament dynamics. A final key element is our decomposition of the main operator L ‾ ϵ ∂ s 4 \overline{\mathcal{L}}_\epsilon\partial_s^4 L ϵ ∂ s 4 in section 3.3, which exploits the inextensibility in a crucial way.
Our next contribution is to compare the dynamics of the curve evolution (1.14) directly with the resistive force theory evolution
∂ Y ∂ t = − L ϵ , R F T ( Y ) [ ( Y s s s − τ Y Y s ) s ] , ∣ Y s ∣ 2 = 1 , (17) \frac{\partial Y}{\partial t}=-\mathcal{L}_{\epsilon,\mathrm{RFT}}(Y)\left[(Y_{sss}-\tau_Y Y_s)_s\right],\qquad|Y_s|^2=1, \tag*{(17)} ∂ t ∂ Y = − L ϵ , RFT ( Y ) [ ( Y sss − τ Y Y s ) s ] , ∣ Y s ∣ 2 = 1 , (17)
where L ϵ , R F T \mathcal{L}_{\epsilon,\mathrm{RFT}} L ϵ , RFT is given by the local operator (1.4). Here we use Y Y Y to distinguish from the evolution (1.14). Since L ϵ , R F T \mathcal{L}_{\epsilon,\mathrm{RFT}} L ϵ , RFT is local, (1.18) is a fourth-order semilinear parabolic equation for Y ( s , t ) Y(s,t) Y ( s , t ) , so the natural energy space L t ∞ H s 2 ∩ L t 2 H s 4 L_t^\infty H_s^2\cap L_t^2 H_s^4 L t ∞ H s 2 ∩ L t 2 H s 4 is more regular than that of the evolution (1.14). The
∥ τ Y ∥ H 1 ( T ) ≤ c ∥ Y s s ∥ H 1 ( T ) 2 ≤ c ∥ Y s s ∥ L 2 ( T ) ∥ Y s s ∥ H 2 ( T ) , (18) \|\tau_Y\|_{H^1(\mathbb{T})}\le c\|Y_{ss}\|_{H^1(\mathbb{T})}^2\le c\|Y_{ss}\|_{L^2(\mathbb{T})}\|Y_{ss}\|_{H^2(\mathbb{T})}, \tag*{(18)} ∥ τ Y ∥ H 1 ( T ) ≤ c ∥ Y ss ∥ H 1 ( T ) 2 ≤ c ∥ Y ss ∥ L 2 ( T ) ∥ Y ss ∥ H 2 ( T ) , (18)
where we emphasise that the right hand side involves a full copy of Y ∈ H s 4 Y\in H_s^4 Y ∈ H s 4 and thus requires the full smoothing power of the semigroup e − t ∂ s 4 e^{-t\partial_s^4} e − t ∂ s 4 to estimate in the energy space. See [29 Remark 2.4] for additional discussion.
PDE theory of (1.18) has been developed by the author and others in [1 , 29 , 26 ] . By [1 Theorem 1.1] , starting from initial data X i n ∈ H 2 ( T ) \mathbf{X}^{\mathrm{in}} \in H^2(\mathbb{T}) X in ∈ H 2 ( T ) , equation (17) admits a unique global solution Y ∈ C t H s 2 ∩ L t , l o c 2 H s 4 \mathbf{Y} \in C_t H_s^2 \cap L_{t,\mathrm{loc}}^2 H_s^4 Y ∈ C t H s 2 ∩ L t , loc 2 H s 4 , and starting from X i n ∈ H 4 ( T ) \mathbf{X}^{\mathrm{in}} \in H^4(\mathbb{T}) X in ∈ H 4 ( T ) , the unique global solution may be upgraded to Y ∈ C t H s 4 ∩ L t , l o c 2 H s 6 \mathbf{Y} \in C_t H_s^4 \cap L_{t,\mathrm{loc}}^2 H_s^6 Y ∈ C t H s 4 ∩ L t , loc 2 H s 6 .
Noting that the behavior of both L ‾ ϵ \overline{\mathcal{L}}_\epsilon L ϵ (see Lemma 2.1 ) and L ϵ , R F T \mathcal{L}_{\epsilon,\mathrm{RFT}} L ϵ , RFT blows up logarithmically in ϵ \epsilon ϵ as ϵ → 0 \epsilon\to0 ϵ → 0 , we consider convergence under the time rescaling
t ‾ = ∣ log ϵ ∣ t , ∣ log ϵ ∣ ∂ ∂ t = ∂ ∂ t ‾ . (19) \underline{t} = \lvert\log\epsilon\rvert t,\qquad\lvert\log\epsilon\rvert\frac{\partial}{\partial t} = \frac{\partial}{\partial\underline{t}}. \tag*{(19)} t = ∣ log ϵ ∣ t , ∣ log ϵ ∣ ∂ t ∂ = ∂ t ∂ . (19)
In section 4, we then show the following convergence result:
Theorem 1.3 (Convergence to RFT dynamics). Given any 0 < T < ∞ 0 < T < \infty 0 < T < ∞ and an initial curve X i n ∈ H 4 ( T ) \mathbf{X}^{\mathrm{in}} \in H^4(\mathbb{T}) X in ∈ H 4 ( T ) , let X ∈ C t H s 2 ∩ L t 2 H ˙ s 7 / 2 ( [ 0 , T ] × T ) \mathbf{X} \in C_t H_s^2 \cap L_t^2 \dot{H}_s^{7/2}([0,T] \times\mathbb{T}) X ∈ C t H s 2 ∩ L t 2 H ˙ s 7/2 ([ 0 , T ] × T ) and Y ∈ C t H s 4 ∩ L t 2 H s 6 ( [ 0 , T ] × T ) \mathbf{Y} \in C_t H_s^4 \cap L_t^2 H_s^6([0,T] \times\mathbb{T}) Y ∈ C t H s 4 ∩ L t 2 H s 6 ([ 0 , T ] × T ) be the unique solutions to (1.14) and the resistive force theory evolution (1.18), respectively, under the time rescaling (1.19) and with initial data X i n \mathbf{X}^{\mathrm{in}} X in . For ϵ > 0 \epsilon> 0 ϵ > 0 sufficiently small, there exists an ϵ \epsilon ϵ -independent constant C R F T = C R F T ( T , ∥ X ∥ L t ∞ H s 2 , ∥ Y ∥ L t ∞ H s 4 ∩ L t 2 H s 6 ) C_{\mathrm{RFT}} = C_{\mathrm{RFT}}(T,\|\mathbf{X}\|_{L_t^\infty H_s^2},\|\mathbf{Y}\|_{L_t^\infty H_s^4 \cap L_t^2 H_s^6}) C RFT = C RFT ( T , ∥ X ∥ L t ∞ H s 2 , ∥ Y ∥ L t ∞ H s 4 ∩ L t 2 H s 6 ) such that
∥ X − Y ∥ L t ∞ H s 2 + ∣ log ϵ ∣ 1 / 2 ∥ X − Y ∥ L t 2 H ˙ s 7 / 2 ≤ ∣ log ϵ ∣ − 1 / 2 C R F T . (20) \|\mathbf{X}-\mathbf{Y}\|_{L_t^\infty H_s^2}+\lvert\log\epsilon\rvert^{1/2}\|\mathbf{X}-\mathbf{Y}\|_{L_t^2\dot{H}_s^{7/2}}\leq\lvert\log\epsilon\rvert^{-1/2}C_{\mathrm{RFT}}. \tag*{(20)} ∥ X − Y ∥ L t ∞ H s 2 + ∣ log ϵ ∣ 1/2 ∥ X − Y ∥ L t 2 H ˙ s 7/2 ≤ ∣ log ϵ ∣ − 1/2 C RFT . (20) In particular, on any fixed time interval, the curve X ( s , t ) \mathbf{X}(s,t) X ( s , t ) converges to Y ( s , t ) \mathbf{Y}(s,t) Y ( s , t ) as ϵ → 0 \epsilon\to0 ϵ → 0 .
Note that the very slow convergence rate is expected, and, for L t 2 H ˙ s 7 / 2 L_t^2\dot{H}_s^{7/2} L t 2 H ˙ s 7/2 , is likely optimal.
Connection with the slender body Neumann-to-Dirichlet map
As mentioned, arguably the most detailed method for incorporating 3D fluid effects into the force-to-velocity map L ϵ \mathcal{L}_\epsilon L ϵ is by solving the following slender body boundary value problem , introduced by the author together with Mori and Spirn in [30 , 31 ] . Given 0 < ϵ ≪ 1 0 < \epsilon\ll1 0 < ϵ ≪ 1 sufficiently small, we may define a ‘fattened’ version of the curve X \mathbf{X} X as
Σ ϵ ( t ) = { x ∈ X s ( s , t ) ⊥ : dist ( x , X ( s , t ) ) < ϵ , s ∈ T } . (21) \Sigma_\epsilon(t)=\left\{x\in\mathbf{X}_s(s,t)^\perp:\operatorname{dist}(x,\mathbf{X}(s,t))<\epsilon,\ s\in\mathbb{T}\right\}. \tag*{(21)} Σ ϵ ( t ) = { x ∈ X s ( s , t ) ⊥ : dist ( x , X ( s , t )) < ϵ , s ∈ T } . (21)
The 3D body Σ ϵ \Sigma_\epsilon Σ ϵ may be considered as a regularization of the filament centerline which allows us to seek ( u , p ) : R 3 \ Σ ‾ ϵ → R 3 × R (u,p):\mathbb{R}^3\backslash\overline{\Sigma}_\epsilon\to\mathbb{R}^3\times\mathbb{R} ( u , p ) : R 3 \ Σ ϵ → R 3 × R satisfying the boundary value problem
− Δ u + ∇ p = 0 , div u = 0 in R 3 \ Σ ‾ ϵ -\Delta u+\nabla p=0,\qquad\operatorname{div}u=0\qquad\text{in }\mathbb{R}^3\backslash\overline{\Sigma}_\epsilon − Δ u + ∇ p = 0 , div u = 0 in R 3 \ Σ ϵ
∫ 0 2 π ( σ [ u ] n ) J ϵ ( s , θ ) d θ = f ( s ) on ∂ Σ ϵ (22) \int_0^{2\pi}(\boldsymbol{\sigma}[u]n)\mathcal{J}_\epsilon(s,\theta)\,\mathrm{d}\theta=f(s)\qquad\text{on }\partial\Sigma_\epsilon \tag*{(22)} ∫ 0 2 π ( σ [ u ] n ) J ϵ ( s , θ ) d θ = f ( s ) on ∂ Σ ϵ (22)
u ∣ ∂ Σ ϵ = v ( s ) , unknown but independent of θ , u\big|_{\partial\Sigma_\epsilon}=v(s),\qquad\text{unknown but independent of }\theta, u ∂ Σ ϵ = v ( s ) , unknown but independent of θ ,
and ∣ u ∣ → 0 |u|\to0 ∣ u ∣ → 0 as ∣ x ∣ → ∞ |x|\to\infty ∣ x ∣ → ∞ . In this boundary value problem, we make sense of the force data f ( s ) f(s) f ( s ) defined along a 1D curve as the total surface stress σ [ u ] n = ( ∇ u + ∇ u T − p I ) n \boldsymbol{\sigma}[u]n=(\nabla u+\nabla u^{\mathsf{T}}-pI)n σ [ u ] n = ( ∇ u + ∇ u T − p I ) n per cross section of the 3D body Σ ϵ \Sigma_\epsilon Σ ϵ , weighted by the body’s surface area through a Jacobian factor J ϵ \mathcal{J}_\epsilon J ϵ . Since this is only a partial, ‘angle-averaged’ Neumann boundary condition, it must be supplemented with an additional partial boundary condition in the form of a geometric constraint on the Dirichlet boundary value of u u u known as a fiber integrity condition. In particular, u ∣ ∂ Σ ϵ u|_{\partial\Sigma_\epsilon} u ∣ ∂ Σ ϵ is unknown but constrained to be a function of arclength s s s only.
The slender body Neumann-to-Dirichlet (NtD) map L ϵ , N t D \mathcal{L}_{\epsilon,\mathrm{NtD}} L ϵ , NtD is then defined as the map
L ϵ , N t D ( X ) : f ( s ) ↦ v ( s ) (23) \mathcal{L}_{\epsilon,\mathrm{NtD}}(\mathbf{X}):f(s)\mapsto v(s) \tag*{(23)} L ϵ , NtD ( X ) : f ( s ) ↦ v ( s ) (23)
which comes from solving the (Neumann) boundary value problem (22) for u u u and evaluating the Dirichlet boundary value on ∂ Σ ϵ \partial\Sigma_\epsilon ∂ Σ ϵ . The solution to this boundary value problem depends on both the filament radius ϵ \epsilon ϵ and the centerline geometry X ( s ) \mathbf{X}(s) X ( s ) in a complicated way. To isolate these complexities, we can consider the slender body NtD map in the simplest possible geometry, which here is the straight, constant-radius filament with periodic boundary conditions at the ends.
In particular, let C ϵ C_\epsilon C ϵ denote the body (21) with straight centerline X ( s ) = s e z \mathbf{X}(s)=s\mathbf{e}_z X ( s ) = s e z , s ∈ T s\in\mathbb{T} s ∈ T . The behavior of the slender body NtD map (23) about C ϵ C_\epsilon C ϵ conveniently diagonalizes, with differing behavior in directions tangent and normal to the filament centerline. In particular, for k ≠ 0 k\ne0 k = 0 , it may be shown (see [28 ] ) that
L ϵ , N t D ( s e z ) [ e 2 π i k s e z ] = m ϵ t ( k ) e 2 π i k s e z , L ϵ , N t D ( s e z ) [ e 2 π i k s e j ] = m ϵ n ( k ) e 2 π i k s e j , j = x , y , (24) \begin{aligned}
\mathcal{L}_{\epsilon,\mathrm{NtD}}(s\mathbf{e}_z)[e^{2\pi i k s}\mathbf{e}_z] &= m_\epsilon^{\mathrm{t}}(k)e^{2\pi i k s}\mathbf{e}_z,\\
\mathcal{L}_{\epsilon,\mathrm{NtD}}(s\mathbf{e}_z)[e^{2\pi i k s}\mathbf{e}_j] &= m_\epsilon^{\mathrm{n}}(k)e^{2\pi i k s}\mathbf{e}_j,\qquad j=x,y, \tag*{(24)}
\end{aligned} L ϵ , NtD ( s e z ) [ e 2 π ik s e z ] L ϵ , NtD ( s e z ) [ e 2 π ik s e j ] = m ϵ t ( k ) e 2 π ik s e z , = m ϵ n ( k ) e 2 π ik s e j , j = x , y , (24)
where the eigenvalues m ϵ t ( k ) m_\epsilon^{\mathrm{t}}(k) m ϵ t ( k ) and m ϵ n ( k ) m_\epsilon^{\mathrm{n}}(k) m ϵ n ( k ) are as in Definition 1.1. More specifically, for the straight filament, we have that the slender body NtD map L ϵ , N t D ( s e z ) [ ⋅ ] \mathcal{L}_{\epsilon,\mathrm{NtD}}(s\mathbf{e}_z)[\cdot] L ϵ , NtD ( s e z ) [ ⋅ ] is exactly the operator L ϵ ( s e z ) [ ⋅ ] \mathcal{L}_\epsilon(s\mathbf{e}_z)[\cdot] L ϵ ( s e z ) [ ⋅ ] of Definition 1.1 (after removing the k = 0 k=0 k = 0 mode).
For a general curved filament, the map L ϵ , N t D ( X ) \mathcal{L}_{\epsilon,\mathrm{NtD}}(\mathbf{X}) L ϵ , NtD ( X ) no longer admits an explicit Fourier multiplier representation. However, in [33 Theorem 1.5 & Corollary 1.6] , it is shown that the principal part of the general operator L ϵ , N t D ( X ) \mathcal{L}_{\epsilon,\mathrm{NtD}}(\mathbf{X}) L ϵ , NtD ( X ) is given by the straight NtD map L ϵ , N t D ( s e z ) \mathcal{L}_{\epsilon,\mathrm{NtD}}(s\mathbf{e}_z) L ϵ , NtD ( s e z ) in the sense that
L ϵ , N t D ( X ) [ ∂ s 4 X ] = L ϵ , N t D ( s e z ) [ ∂ s 4 X ] + R e r r , (25) \mathcal{L}_{\epsilon,\mathrm{NtD}}(\mathbf{X})[\partial_s^4\mathbf{X}]
=
\mathcal{L}_{\epsilon,\mathrm{NtD}}(s\mathbf{e}_z)[\partial_s^4\mathbf{X}]
+\mathcal{R}_{\mathrm{err}}, \tag*{(25)} L ϵ , NtD ( X ) [ ∂ s 4 X ] = L ϵ , NtD ( s e z ) [ ∂ s 4 X ] + R err , (25)
where the remainder terms R e r r \mathcal{R}_{\mathrm{err}} R err are lower order with respect to regularity or size in ϵ \epsilon ϵ . In [36 Theorem 1.1] , this decomposition is leveraged to show qualitative local well-posedness for the full slender body free boundary problem
∂ X ∂ t = − L ϵ , N t D ( X ) [ ( X s s s − τ X s ) s ] , ∣ X s ∣ 2 = 1 , (26) \frac{\partial\mathbf{X}}{\partial t}
=
-\mathcal{L}_{\epsilon,\mathrm{NtD}}(\mathbf{X})[(\mathbf{X}_{sss}-\tau\mathbf{X}_s)_s],
\qquad
|\mathbf{X}_s|^2=1, \tag*{(26)} ∂ t ∂ X = − L ϵ , NtD ( X ) [( X sss − τ X s ) s ] , ∣ X s ∣ 2 = 1 , (26)
where L ϵ , N t D ( X ) \mathcal{L}_{\epsilon,\mathrm{NtD}}(\mathbf{X}) L ϵ , NtD ( X ) is given by (23) for the instantaneous curve shape X \mathbf{X} X . More precisely, local well-posedness is shown in C ( [ 0 , T ] ; h 4 , α ( T ) ) C([0,T];h^{4,\alpha}(\mathbb{T})) C ([ 0 , T ] ; h 4 , α ( T )) for an initial curve belonging to the little Hölder space h 4 , α ( T ) h^{4,\alpha}(\mathbb{T}) h 4 , α ( T ) . Note that, by (25) , the evolution (26) is a third order, quasilinear parabolic equation for X \mathbf{X} X . In particular, this regularity is shared by the evolution (14) considered here.
The setting of Hölder spaces is more convenient for the operator decomposition (25) , but less convenient for obtaining insight into long-time dynamics. Here our choice of operator L ϵ \mathcal{L}_\epsilon L ϵ as in Definition 1.1 is inspired by the principal straight part of the slender body NtD map, adapted slightly for application directly to a curved filament. This allows us to develop the solution theory of the principal dynamics in a natural energy space while avoiding the additional difficulties associated with (a.) decomposing the full operator as in (25) in lower regularity spaces, and (b.) accounting for filament self-intersection.
Multiplier bounds and energetics
Before proceeding to the proof of Theorem 1.2, we make note of some useful bounds for the multipliers m ϵ t m_\epsilon^{\mathrm{t}} m ϵ t and m ϵ n m_\epsilon^{\mathrm{n}} m ϵ n comprising the operator L ϵ \mathcal{L}_\epsilon L ϵ . These may be used to show that the energy identity satisfied by solutions to (14) is in fact signed, which will be crucial for the global-in-time behavior of the filament.
Multiplier bounds
We begin with a collection of lemmas bounding the behavior of the multipliers m ϵ t ( k ) m_\epsilon^{\mathrm{t}}(k) m ϵ t ( k ) and m ϵ n ( k ) m_\epsilon^{\mathrm{n}}(k) m ϵ n ( k ) in both k k k and ϵ \epsilon ϵ . First, from [33 Lemmas 3.4 & 3.5] , we note the following general bounds for the behavior of m ϵ t m_\epsilon^{\mathrm{t}} m ϵ t and m ϵ n m_\epsilon^{\mathrm{n}} m ϵ n , which differs at high versus low wavenumbers.
Lemma 2.1 (Multiplier behavior [33 ] ). Given ϵ > 0 \epsilon> 0 ϵ > 0 and m ϵ t ( k ) m_\epsilon^t(k) m ϵ t ( k ) and m ϵ n ( k ) m_\epsilon^n(k) m ϵ n ( k ) as in (11) , (12) , there exist constants c > 0 c > 0 c > 0 such that
m ϵ t ( k ) , m ϵ n ( k ) ≤ { c ∣ log ϵ ∣ , ∣ k ∣ < 1 2 π ϵ , c ϵ − 1 ∣ k ∣ − 1 , ∣ k ∣ ≥ 1 2 π ϵ , m ϵ t ( k ) − 1 , m ϵ n ( k ) − 1 ≤ { c ∣ log ϵ ∣ − 1 , ∣ k ∣ < 1 2 π ϵ , c ϵ ∣ k ∣ , ∣ k ∣ ≥ 1 2 π ϵ . (27) \begin{aligned}
m_\epsilon^t(k),\ m_\epsilon^n(k) &\le\begin{cases}
c|\log\epsilon|, & |k| < \frac{1}{2\pi\epsilon},\\
c\epsilon^{-1}|k|^{-1}, & |k| \ge\frac{1}{2\pi\epsilon},
\end{cases}\\
m_\epsilon^t(k)^{-1},\ m_\epsilon^n(k)^{-1} &\le\begin{cases}
c|\log\epsilon|^{-1}, & |k| < \frac{1}{2\pi\epsilon},\\
c\epsilon|k|, & |k| \ge\frac{1}{2\pi\epsilon}.
\end{cases} \tag*{(27)}
\end{aligned} m ϵ t ( k ) , m ϵ n ( k ) m ϵ t ( k ) − 1 , m ϵ n ( k ) − 1 ≤ { c ∣ log ϵ ∣ , c ϵ − 1 ∣ k ∣ − 1 , ∣ k ∣ < 2 π ϵ 1 , ∣ k ∣ ≥ 2 π ϵ 1 , ≤ { c ∣ log ϵ ∣ − 1 , cϵ ∣ k ∣ , ∣ k ∣ < 2 π ϵ 1 , ∣ k ∣ ≥ 2 π ϵ 1 . (27)
This behavior is displayed in Figure 2 . In addition, we will require some refinements to Lemma 2.1 at both high and low wavenumbers. For high wavenumbers, by [28 Proposition 1.4] , we may obtain a more precise linear growth rate for ( m ϵ t ) − 1 (m_\epsilon^t)^{-1} ( m ϵ t ) − 1 and ( m ϵ n ) − 1 (m_\epsilon^n)^{-1} ( m ϵ n ) − 1 as follows.
Lemma 2.2 (High wavenumber refinement [28 ] ). There exist constants c t , c n > 0 c_t, c_n > 0 c t , c n > 0 such that for all k > 0 k > 0 k > 0 , the multipliers m ϵ t m_\epsilon^t m ϵ t and m ϵ n m_\epsilon^n m ϵ n in (11) , (12) satisfy
c t − 1 ϵ ∣ k ∣ ≤ m ϵ t ( k ) − 1 ≤ c t − 1 ( ϵ ∣ k ∣ + 1 ) c n − 1 ϵ ∣ k ∣ ≤ m ϵ n ( k ) − 1 ≤ c n − 1 ( ϵ ∣ k ∣ + 1 ) . (28) c_t^{-1}\epsilon|k| \le m_\epsilon^t(k)^{-1} \le c_t^{-1}(\epsilon|k|+1) \qquad c_n^{-1}\epsilon|k| \le m_\epsilon^n(k)^{-1} \le c_n^{-1}(\epsilon|k|+1). \tag*{(28)} c t − 1 ϵ ∣ k ∣ ≤ m ϵ t ( k ) − 1 ≤ c t − 1 ( ϵ ∣ k ∣ + 1 ) c n − 1 ϵ ∣ k ∣ ≤ m ϵ n ( k ) − 1 ≤ c n − 1 ( ϵ ∣ k ∣ + 1 ) . (28)
For low wavenumbers, we note the following refined asymptotics, which will allow us to prove the convergence result of Theorem 1.3 .
Lemma 2.3 (Low wavenumber refinement). For 0 < ϵ < 1 0 < \epsilon< 1 0 < ϵ < 1 and for 0 < ∣ k ∣ < 1 2 π ϵ 0 < |k| < \frac{1}{2\pi\epsilon} 0 < ∣ k ∣ < 2 π ϵ 1 , we have the refined low wavenumber bounds
m ϵ t ( k ) − ∣ log ϵ ∣ 2 π ≤ c ( 1 + ∣ log ∣ k ∣ ∣ ) , m ϵ n ( k ) − ∣ log ϵ ∣ 4 π ≤ c ( 1 + ∣ log ∣ k ∣ ∣ ) , (29) m_\epsilon^t(k)-\frac{|\log\epsilon|}{2\pi} \le c(1+|\log|k||), \qquad m_\epsilon^n(k)-\frac{|\log\epsilon|}{4\pi} \le c(1+|\log|k||), \tag*{(29)} m ϵ t ( k ) − 2 π ∣ log ϵ ∣ ≤ c ( 1 + ∣ log ∣ k ∣∣ ) , m ϵ n ( k ) − 4 π ∣ log ϵ ∣ ≤ c ( 1 + ∣ log ∣ k ∣∣ ) , (29) along with m ϵ t ( 0 ) − ∣ log ϵ ∣ 2 π = m ϵ n ( 0 ) − ∣ log ϵ ∣ 4 π = 0 m_\epsilon^t(0)-\frac{|\log\epsilon|}{2\pi}=m_\epsilon^n(0)-\frac{|\log\epsilon|}{4\pi}=0 m ϵ t ( 0 ) − 2 π ∣ l o g ϵ ∣ = m ϵ n ( 0 ) − 4 π ∣ l o g ϵ ∣ = 0 .
In particular, this difference is bounded independent of ϵ \epsilon ϵ at low wavenumbers.
Proof of Lemma 2.3. Expanding m ϵ t ( k ) m_\epsilon^t(k) m ϵ t ( k ) and m ϵ n ( k ) m_\epsilon^n(k) m ϵ n ( k ) in a Taylor series about ∣ k ∣ = 0 |k|=0 ∣ k ∣ = 0 yields
m ϵ t ( k ) = − 1 − 2 γ + 2 log π − 2 log ( ϵ ∣ k ∣ ) 4 π + O ( ϵ 2 k 2 log 2 ( ϵ ∣ k ∣ ) ) m_\epsilon^t(k)=\frac{-1-2\gamma+2\log\pi-2\log(\epsilon|k|)}{4\pi}+O(\epsilon^2 k^2\log^2(\epsilon|k|)) m ϵ t ( k ) = 4 π − 1 − 2 γ + 2 log π − 2 log ( ϵ ∣ k ∣ ) + O ( ϵ 2 k 2 log 2 ( ϵ ∣ k ∣ ))
m ϵ n ( k ) = 1 − 2 γ − 2 log π − 2 log ( ϵ ∣ k ∣ ) 8 π + O ( ϵ 2 k 2 log 2 ( ϵ ∣ k ∣ ) ) . (30) m_\epsilon^n(k)=\frac{1-2\gamma-2\log\pi-2\log(\epsilon|k|)}{8\pi}+O(\epsilon^2 k^2\log^2(\epsilon|k|)). \tag*{(30)} m ϵ n ( k ) = 8 π 1 − 2 γ − 2 log π − 2 log ( ϵ ∣ k ∣ ) + O ( ϵ 2 k 2 log 2 ( ϵ ∣ k ∣ )) . (30)
Here γ ≈ 0.5772 \gamma\approx0.5772 γ ≈ 0.5772 is the Euler gamma. Subtracting ∣ log ϵ ∣ 2 π \frac{|\log\epsilon|}{2\pi} 2 π ∣ l o g ϵ ∣ from m ϵ t ( k ) m_\epsilon^t(k) m ϵ t ( k ) and ∣ log ϵ ∣ 4 π \frac{|\log\epsilon|}{4\pi} 4 π ∣ l o g ϵ ∣ from m ϵ n ( k ) m_\epsilon^n(k) m ϵ n ( k ) yields remainder terms satisfying (29) . □ \square □
Energetics
The filament evolution (14) admits the following energy identity:
1 2 ∂ t ∫ T ∣ X s s ∣ 2 d s = − ∫ T ( X s s s − τ X s ) s ⋅ L ‾ ϵ [ ( X s s s − τ X s ) s ] d s (31) \frac{1}{2}\partial_t\int_{\mathbb{T}}|\mathbf{X}_{ss}|^2\,ds=-\int_{\mathbb{T}}(\mathbf{X}_{sss}-\tau\mathbf{X}_s)_s\cdot\overline{\mathcal{L}}_\epsilon[(\mathbf{X}_{sss}-\tau\mathbf{X}_s)_s]\,ds \tag*{(31)} 2 1 ∂ t ∫ T ∣ X ss ∣ 2 d s = − ∫ T ( X sss − τ X s ) s ⋅ L ϵ [( X sss − τ X s ) s ] d s (31)
which may be obtained by dotting equation (14) with ( X s s s − τ X s ) s (\mathbf{X}_{sss}-\tau\mathbf{X}_s)_s ( X sss − τ X s ) s and integrating over T \mathbb{T} T . Noting that the inextensibility constraint ∣ X s ∣ 2 = 1 |\mathbf{X}_s|^2=1 ∣ X s ∣ 2 = 1 implies ∂ t X s ⋅ X s = 0 \partial_t\mathbf{X}_s\cdot\mathbf{X}_s=0 ∂ t X s ⋅ X s = 0 , we have
∫ T ∂ X ∂ t ⋅ ( X s s s − τ X s ) s d s = − ∫ T ∂ X s ∂ t ⋅ ( X s s s − τ X s ) d s = − ∫ T ∂ X s ∂ t ⋅ X s s s d s = 1 2 ∂ t ∫ T ∣ X s s ∣ 2 d s (32) \begin{aligned}
\int_{\mathbb{T}}\frac{\partial\mathbf{X}}{\partial t}\cdot(\mathbf{X}_{sss}-\tau\mathbf{X}_s)_s\,ds=-\int_{\mathbb{T}}\frac{\partial\mathbf{X}_s}{\partial t}\cdot(\mathbf{X}_{sss}-\tau\mathbf{X}_s)\,ds \\
&=-\int_{\mathbb{T}}\frac{\partial\mathbf{X}_s}{\partial t}\cdot\mathbf{X}_{sss}\,ds=\frac{1}{2}\partial_t\int_{\mathbb{T}}|\mathbf{X}_{ss}|^2\,ds \tag*{(32)}
\end{aligned} ∫ T ∂ t ∂ X ⋅ ( X sss − τ X s ) s d s = − ∫ T ∂ t ∂ X s ⋅ ( X sss − τ X s ) d s = − ∫ T ∂ t ∂ X s ⋅ X sss d s = 2 1 ∂ t ∫ T ∣ X ss ∣ 2 d s (32)
yielding the left hand side of (31) .
To see that the right hand side of (31) is signed, we make note of the following proposition.
Proposition 2.4. Given f ∈ H − 1 / 2 ( T ) f \in H^{-1/2}(\mathbb{T}) f ∈ H − 1/2 ( T ) , we have
∫ T f ⋅ L ‾ ϵ [ f ] d s ≥ c ∣ log ϵ ∣ ( ∥ P X s f ∥ H − 1 / 2 ( T ) 2 + ∥ P X s ⊥ f ∥ H − 1 / 2 ( T ) 2 ) ≥ c ∣ log ϵ ∣ ∥ f ∥ H − 1 / 2 ( T ) 2 . (33) \int_{\mathbb{T}} f \cdot\overline{\mathcal{L}}_{\epsilon}[f]\,ds \ge c|\log\epsilon|\left(\|\mathbb{P}_{X_s}f\|_{H^{-1/2}(\mathbb{T})}^{2}+\|\mathbb{P}_{X_s}^{\perp}f\|_{H^{-1/2}(\mathbb{T})}^{2}\right)
\ge c|\log\epsilon|\|f\|_{H^{-1/2}(\mathbb{T})}^{2}. \tag*{(33)} ∫ T f ⋅ L ϵ [ f ] d s ≥ c ∣ log ϵ ∣ ( ∥ P X s f ∥ H − 1/2 ( T ) 2 + ∥ P X s ⊥ f ∥ H − 1/2 ( T ) 2 ) ≥ c ∣ log ϵ ∣∥ f ∥ H − 1/2 ( T ) 2 . (33)
We note that this lower bound is a bit pessimistic (in ϵ \epsilon ϵ ) due to the differing ϵ \epsilon ϵ -scaling of the multipliers at low versus high wavenumbers (Lemma 2.1 ); however, it will be sufficient for establishing local and global well-posedness.
Proof. Using Definition 1.1 for L ‾ ϵ \overline{\mathcal{L}}_{\epsilon} L ϵ along with Parseval’s theorem, we may write
∫ T f ⋅ L ‾ ϵ [ f ] d s = ∫ T ( ( P X s f ) ⋅ T ϵ m t ( P X s f ) + ( P X s ⊥ f ) ⋅ T ϵ m n ( P X s ⊥ f ) ) d s = ∑ ∣ k ∣ = 0 ∞ ∣ F [ P X s f ] ( k ) ∣ 2 m ϵ t ( k ) + ∣ F [ P X s ⊥ f ] ( k ) ∣ 2 m ϵ n ( k ) ≥ c ∣ log ϵ ∣ ∑ ∣ k ∣ = 0 ⌊ 1 2 π ϵ ⌋ ∣ F [ P X s f ] ( k ) ∣ 2 + ∣ F [ P X s ⊥ f ] ( k ) ∣ 2 + c ∣ log ϵ ∣ ∑ ∣ k ∣ = ⌊ 1 2 π ϵ ⌋ + 1 ∞ ∣ k ∣ − 1 ( ∣ F [ P X s f ] ( k ) ∣ 2 + ∣ F [ P X s ⊥ f ] ( k ) ∣ 2 ) ≥ c ∣ log ϵ ∣ ( ∥ P X s f ∥ H − 1 / 2 ( T ) 2 + ∥ P X s ⊥ f ∥ H − 1 / 2 ( T ) 2 ) ≥ c ∣ log ϵ ∣ ∥ f ∥ H − 1 / 2 ( T ) 2 . (34) \begin{aligned}
\int_{\mathbb{T}} f\cdot\overline{\mathcal{L}}_{\epsilon}[f]\,ds
=
\int_{\mathbb{T}}\left((\mathbb{P}_{X_s}f)\cdot T_{\epsilon}^{m_t}(\mathbb{P}_{X_s}f)+(\mathbb{P}_{X_s}^{\perp}f)\cdot T_{\epsilon}^{m_n}(\mathbb{P}_{X_s}^{\perp}f)\right)\,ds \\
&=\sum_{|k|=0}^{\infty}\left|\mathcal{F}[\mathbb{P}_{X_s}f](k)\right|^{2}m_{\epsilon}^{t}(k)+\left|\mathcal{F}[\mathbb{P}_{X_s}^{\perp}f](k)\right|^{2}m_{\epsilon}^{n}(k) \\
&\ge c|\log\epsilon|\sum_{|k|=0}^{\left\lfloor\frac{1}{2\pi\epsilon}\right\rfloor}\left|\mathcal{F}[\mathbb{P}_{X_s}f](k)\right|^{2}+\left|\mathcal{F}[\mathbb{P}_{X_s}^{\perp}f](k)\right|^{2} \\
&+c|\log\epsilon|\sum_{|k|=\left\lfloor\frac{1}{2\pi\epsilon}\right\rfloor+1}^{\infty}|k|^{-1}\left(\left|\mathcal{F}[\mathbb{P}_{X_s}f](k)\right|^{2}+\left|\mathcal{F}[\mathbb{P}_{X_s}^{\perp}f](k)\right|^{2}\right) \\
&\ge c|\log\epsilon|\left(\|\mathbb{P}_{X_s}f\|_{H^{-1/2}(\mathbb{T})}^{2}+\|\mathbb{P}_{X_s}^{\perp}f\|_{H^{-1/2}(\mathbb{T})}^{2}\right)
\ge c|\log\epsilon|\|f\|_{H^{-1/2}(\mathbb{T})}^{2}. \tag*{(34)}
\end{aligned} ∫ T f ⋅ L ϵ [ f ] d s = ∫ T ( ( P X s f ) ⋅ T ϵ m t ( P X s f ) + ( P X s ⊥ f ) ⋅ T ϵ m n ( P X s ⊥ f ) ) d s = ∣ k ∣ = 0 ∑ ∞ ∣ F [ P X s f ] ( k ) ∣ 2 m ϵ t ( k ) + F [ P X s ⊥ f ] ( k ) 2 m ϵ n ( k ) ≥ c ∣ log ϵ ∣ ∣ k ∣ = 0 ∑ ⌊ 2 π ϵ 1 ⌋ ∣ F [ P X s f ] ( k ) ∣ 2 + F [ P X s ⊥ f ] ( k ) 2 + c ∣ log ϵ ∣ ∣ k ∣ = ⌊ 2 π ϵ 1 ⌋ + 1 ∑ ∞ ∣ k ∣ − 1 ( ∣ F [ P X s f ] ( k ) ∣ 2 + F [ P X s ⊥ f ] ( k ) 2 ) ≥ c ∣ log ϵ ∣ ( ∥ P X s f ∥ H − 1/2 ( T ) 2 + ∥ P X s ⊥ f ∥ H − 1/2 ( T ) 2 ) ≥ c ∣ log ϵ ∣∥ f ∥ H − 1/2 ( T ) 2 . (34)
by the triangle inequality. Here in the third line we have used the behavior of the multipliers m ϵ t m_{\epsilon}^{t} m ϵ t and m ϵ n m_{\epsilon}^{n} m ϵ n from Lemma 2.1 . □
Using Proposition 2.4 , the filament bending energy is non-increasing in time; in particular,
1 2 ∂ t ∫ T ∣ X s s ∣ 2 d s ≤ − c ∣ log ϵ ∣ ∥ ( X s s s − τ X s ) s ∥ H − 1 / 2 ( T ) 2 ≤ − c ∣ log ϵ ∣ ∥ X s s s − τ X s ∥ H ˙ 1 / 2 ( T ) 2 . (35) \begin{aligned}
\frac{1}{2}\partial_t\int_{\mathbb{T}}|X_{ss}|^{2}\,ds\le-c|\log\epsilon|\| (X_{sss}-\tau X_s)_s\|_{H^{-1/2}(\mathbb{T})}^{2} \\
&\le-c|\log\epsilon|\|X_{sss}-\tau X_s\|_{\dot{H}^{1/2}(\mathbb{T})}^{2}. \tag*{(35)}
\end{aligned} 2 1 ∂ t ∫ T ∣ X ss ∣ 2 d s ≤ − c ∣ log ϵ ∣∥ ( X sss − τ X s ) s ∥ H − 1/2 ( T ) 2 ≤ − c ∣ log ϵ ∣∥ X sss − τ X s ∥ H ˙ 1/2 ( T ) 2 . (35)
The energy inequality (35) will form the basis for our proof of global well-posedness.
Global well-posedness
This section is devoted to the proof of Theorem 1.2 . We begin with bounds for the tension τ ( s , t ) \tau(s,t) τ ( s , t ) appearing as a Lagrange multiplier in (1.14). These bounds are leveraged to show local well-posedness in the energy space, which is then upgraded to global well-posedness using the energy inequality (35) .
Preliminaries
3.1. Preliminaries. Here we collect some identities and bounds that will be used throughout the following sections. We begin with the observation that, by differentiating the inextensibility constraint ∣ X s ∣ 2 = 1 |X_s|^2=1 ∣ X s ∣ 2 = 1 , we obtain the following identities:
X s ⋅ X s s = 0 , ∣ X s s ∣ 2 = − X s ⋅ X s s s , 3 X s s s ⋅ X s s = − X s ⋅ X s s s s . (36) X_s\cdot X_{ss}=0,\qquad|X_{ss}|^2=-X_s\cdot X_{sss},\qquad3X_{sss}\cdot X_{ss}=-X_s\cdot X_{ssss}. \tag*{(36)} X s ⋅ X ss = 0 , ∣ X ss ∣ 2 = − X s ⋅ X sss , 3 X sss ⋅ X ss = − X s ⋅ X ssss . (36)
Next, we make note of some useful interpolation inequalities on T \mathbb{T} T . Given h : T → R h:\mathbb{T}\to\mathbb{R} h : T → R with ∫ T h d s = 0 \int_{\mathbb{T}}h\,ds=0 ∫ T h d s = 0 , we may bound
∥ h ∥ H ˙ 1 / 2 ≤ c ∥ ∂ s h ∥ L 2 1 / 2 ∥ h ∥ L 2 1 / 2 , ∥ h s ∥ L 2 ≤ c ∥ h s ∥ H ˙ 1 / 2 2 / 3 ∥ h ∥ L 2 1 / 3 , ∥ h ∥ L ∞ ≤ c ∥ h s ∥ H ˙ 1 / 2 1 / 3 ∥ h ∥ L 2 2 / 3 . (37) \begin{aligned}
\|h\|_{\dot{H}^{1/2}} &\le c\|\partial_s h\|_{L^2}^{1/2}\|h\|_{L^2}^{1/2},\\
\|h_s\|_{L^2} &\le c\|h_s\|_{\dot{H}^{1/2}}^{2/3}\|h\|_{L^2}^{1/3},\\
\|h\|_{L^\infty} &\le c\|h_s\|_{\dot{H}^{1/2}}^{1/3}\|h\|_{L^2}^{2/3}. \tag*{(37)}
\end{aligned} ∥ h ∥ H ˙ 1/2 ∥ h s ∥ L 2 ∥ h ∥ L ∞ ≤ c ∥ ∂ s h ∥ L 2 1/2 ∥ h ∥ L 2 1/2 , ≤ c ∥ h s ∥ H ˙ 1/2 2/3 ∥ h ∥ L 2 1/3 , ≤ c ∥ h s ∥ H ˙ 1/2 1/3 ∥ h ∥ L 2 2/3 . (37)
The proof of these may be found, for example, in [2 ] .
Finally, we will require the following proposition.
Proposition 3.1. Given scalar-valued h ∈ H − 1 / 2 ( T ) h\in H^{-1/2}(\mathbb{T}) h ∈ H − 1/2 ( T ) and vector-valued v ∈ H 1 ( T ) v\in H^1(\mathbb{T}) v ∈ H 1 ( T ) with ∣ v ∣ = 1 |v|=1 ∣ v ∣ = 1 , we may bound
∥ h ∥ H − 1 / 2 ( T ) ≤ c ∥ v ∥ H 1 ( T ) ∥ h v ∥ H − 1 / 2 ( T ) . (38) \|h\|_{H^{-1/2}(\mathbb{T})}\le c\|v\|_{H^1(\mathbb{T})}\|hv\|_{H^{-1/2}(\mathbb{T})}. \tag*{(38)} ∥ h ∥ H − 1/2 ( T ) ≤ c ∥ v ∥ H 1 ( T ) ∥ h v ∥ H − 1/2 ( T ) . (38) In addition, we may bound
∥ h v ∥ H − 1 / 2 ( T ) ≤ c ∥ v ∥ H 1 ( T ) ∥ h ∥ H − 1 / 2 ( T ) . (39) \|hv\|_{H^{-1/2}(\mathbb{T})}\le c\|v\|_{H^1(\mathbb{T})}\|h\|_{H^{-1/2}(\mathbb{T})}. \tag*{(39)} ∥ h v ∥ H − 1/2 ( T ) ≤ c ∥ v ∥ H 1 ( T ) ∥ h ∥ H − 1/2 ( T ) . (39)
Proof. For any scalar-valued φ ∈ H 1 / 2 ( T ) \varphi\in H^{1/2}(\mathbb{T}) φ ∈ H 1/2 ( T ) with ∥ φ ∥ H 1 / 2 = 1 \|\varphi\|_{H^{1/2}}=1 ∥ φ ∥ H 1/2 = 1 , we may calculate
∣ ∫ T h φ d s ∣ = ∣ ∫ T ( h v ) ⋅ ( v φ ) d s ∣ ≤ c ∥ h v ∥ H − 1 / 2 ∥ v φ ∥ H 1 / 2 ≤ c ∥ h v ∥ H − 1 / 2 ∥ v ∥ H 1 ∥ φ ∥ H 1 / 2 = c ∥ h v ∥ H − 1 / 2 ∥ v ∥ H 1 . (40) \begin{aligned}
\left|\int_{\mathbb{T}}h\varphi\,ds\right|=\left|\int_{\mathbb{T}}(hv)\cdot(v\varphi)\,ds\right|\le c\|hv\|_{H^{-1/2}}\|v\varphi\|_{H^{1/2}} \\
&\le c\|hv\|_{H^{-1/2}}\|v\|_{H^1}\|\varphi\|_{H^{1/2}}=c\|hv\|_{H^{-1/2}}\|v\|_{H^1}. \tag*{(40)}
\end{aligned} ∫ T h φ d s = ∫ T ( h v ) ⋅ ( v φ ) d s ≤ c ∥ h v ∥ H − 1/2 ∥ v φ ∥ H 1/2 ≤ c ∥ h v ∥ H − 1/2 ∥ v ∥ H 1 ∥ φ ∥ H 1/2 = c ∥ h v ∥ H − 1/2 ∥ v ∥ H 1 . (40)
Taking the supremum over all such φ \varphi φ yields (3.3).
Similarly, taking any vector-valued φ ∈ H 1 / 2 ( T ) \varphi\in H^{1/2}(\mathbb{T}) φ ∈ H 1/2 ( T ) with ∥ φ ∥ H 1 / 2 = 1 \|\varphi\|_{H^{1/2}}=1 ∥ φ ∥ H 1/2 = 1 , we may calculate
∣ ∫ T h v ⋅ φ d s ∣ ≤ ∥ h ∥ H − 1 / 2 ∥ v ⋅ φ ∥ H 1 / 2 ≤ c ∥ h ∥ H − 1 / 2 ∥ v ∥ H 1 ∥ φ ∥ H 1 / 2 = c ∥ h ∥ H − 1 / 2 ∥ v ∥ H 1 . (41) \begin{aligned}
\left|\int_{\mathbb{T}}hv\cdot\varphi\,ds\right|\le\|h\|_{H^{-1/2}}\|v\cdot\varphi\|_{H^{1/2}} \\
&\le c\|h\|_{H^{-1/2}}\|v\|_{H^1}\|\varphi\|_{H^{1/2}}=c\|h\|_{H^{-1/2}}\|v\|_{H^1}. \tag*{(41)}
\end{aligned} ∫ T h v ⋅ φ d s ≤ ∥ h ∥ H − 1/2 ∥ v ⋅ φ ∥ H 1/2 ≤ c ∥ h ∥ H − 1/2 ∥ v ∥ H 1 ∥ φ ∥ H 1/2 = c ∥ h ∥ H − 1/2 ∥ v ∥ H 1 . (41)
Again taking the supremum over all such φ \varphi φ yields (3.4). □
Tension estimates
As in the analogous tension determination problems of [29 , 21 , 10 , 1 , 36 ] , we will use that ∂ t X s ⋅ X s = 0 \partial_tX_s\cdot X_s=0 ∂ t X s ⋅ X s = 0 to obtain an equation and subsequent estimates for the tension τ \tau τ . Differentiating (1.14) in s s s and dotting with X s X_s X s , we have that at each time t t t (which we will omit from our notation throughout this section), τ \tau τ satisfies the equation
( L ‾ ϵ [ ( τ X s ) s ] ) s ⋅ X s = ( L ‾ ϵ [ X s s s s ] ) s ⋅ X s . (42) \left(\overline{\mathcal{L}}_\epsilon[(\tau X_s)_s]\right)_s\cdot X_s=\left(\overline{\mathcal{L}}_\epsilon[X_{ssss}]\right)_s\cdot X_s. \tag*{(42)} ( L ϵ [( τ X s ) s ] ) s ⋅ X s = ( L ϵ [ X ssss ] ) s ⋅ X s . (42)
We show the following.
Lemma 3.2 (Tension bounds). Given ϵ > 0 \epsilon>0 ϵ > 0 and X ∈ H 7 / 2 ( T ) X\in H^{7/2}(\mathbb{T}) X ∈ H 7/2 ( T ) , the tension equation (42) admits a unique solution τ ∈ H 1 / 2 ( T ) \tau\in H^{1/2}(\mathbb{T}) τ ∈ H 1/2 ( T ) . Furthermore, τ \tau τ satisfies the bound
∥ τ ∥ H 1 / 2 ≤ c ϵ − 1 ( 1 + ∥ X s ∥ H 1 4 ) ( ∥ X s s s ∥ H ˙ 1 / 2 ∥ X s ∥ H 1 3 / 4 + ∥ X s s s ∥ H ˙ 1 / 2 ∥ X s ∥ H 1 10 / 3 ) . (43) \|\tau\|_{H^{1/2}}\le c\epsilon^{-1}(1+\|X_s\|_{H^1}^{4})\left(\|X_{sss}\|_{\dot{H}^{1/2}}\|X_s\|_{H^1}^{3/4}+\|X_{sss}\|_{\dot{H}^{1/2}}\|X_s\|_{H^1}^{10/3}\right). \tag*{(43)} ∥ τ ∥ H 1/2 ≤ c ϵ − 1 ( 1 + ∥ X s ∥ H 1 4 ) ( ∥ X sss ∥ H ˙ 1/2 ∥ X s ∥ H 1 3/4 + ∥ X sss ∥ H ˙ 1/2 ∥ X s ∥ H 1 10/3 ) . (43)
Given two nearby curves X 1 , X 2 ∈ H 7 / 2 ( T ) \mathbf{X}_1,\mathbf{X}_2 \in H^{7/2}(\mathbb{T}) X 1 , X 2 ∈ H 7/2 ( T ) , the solutions τ 1 , τ 2 \tau_1,\tau_2 τ 1 , τ 2 to the corresponding tension determination problems (42) satisfy
∥ τ 1 − τ 2 ∥ H 1 / 2 ≤ c ϵ − 1 ( 1 + ∥ ( X 1 ) s ∥ H 1 9 + ∥ ( X 2 ) s ∥ H 1 9 ) × [ ∥ ( X 1 − X 2 ) s ∥ H 1 ( ∥ ( X 1 ) s s s ∥ H ˙ 1 / 2 2 / 3 + ∥ ( X 2 ) s s s ∥ H ˙ 1 / 2 2 / 3 + ∥ ( X 1 ) s s s ∥ H ˙ 1 / 2 3 / 4 ) + ∥ ( X 1 − X 2 ) s s s ∥ H ˙ 1 / 2 1 / 3 ∥ ( X 1 − X 2 ) s ∥ H 1 2 / 3 ( ∥ ( X 1 ) s s s ∥ H ˙ 1 / 2 1 / 3 + ∥ ( X 2 ) s s s ∥ H ˙ 1 / 2 1 / 3 ) + ∥ ( X 1 − X 2 ) s s s ∥ H ˙ 1 / 2 3 / 4 ∥ ( X 1 − X 2 ) s ∥ H 1 1 / 4 ] . (44) \begin{aligned}
\|\tau_1-\tau_2\|_{H^{1/2}} \leq{}& c\epsilon^{-1}(1+\|(\mathbf{X}_1)_s\|_{H^1}^{9}+\|(\mathbf{X}_2)_s\|_{H^1}^{9})\times\\
&\Bigl[\|(\mathbf{X}_1-\mathbf{X}_2)_s\|_{H^1}\bigl(\|(\mathbf{X}_1)_{sss}\|_{\dot{H}^{1/2}}^{2/3}+\|(\mathbf{X}_2)_{sss}\|_{\dot{H}^{1/2}}^{2/3}+\|(\mathbf{X}_1)_{sss}\|_{\dot{H}^{1/2}}^{3/4}\bigr) \\
&\quad+\|(\mathbf{X}_1-\mathbf{X}_2)_{sss}\|_{\dot{H}^{1/2}}^{1/3}\|(\mathbf{X}_1-\mathbf{X}_2)_s\|_{H^1}^{2/3}\bigl(\|(\mathbf{X}_1)_{sss}\|_{\dot{H}^{1/2}}^{1/3}+\|(\mathbf{X}_2)_{sss}\|_{\dot{H}^{1/2}}^{1/3}\bigr) \\
&\quad+\|(\mathbf{X}_1-\mathbf{X}_2)_{sss}\|_{\dot{H}^{1/2}}^{3/4}\|(\mathbf{X}_1-\mathbf{X}_2)_s\|_{H^1}^{1/4}\Bigr]. \tag*{(44)}
\end{aligned} ∥ τ 1 − τ 2 ∥ H 1/2 ≤ c ϵ − 1 ( 1 + ∥ ( X 1 ) s ∥ H 1 9 + ∥ ( X 2 ) s ∥ H 1 9 ) × [ ∥ ( X 1 − X 2 ) s ∥ H 1 ( ∥ ( X 1 ) sss ∥ H ˙ 1/2 2/3 + ∥ ( X 2 ) sss ∥ H ˙ 1/2 2/3 + ∥ ( X 1 ) sss ∥ H ˙ 1/2 3/4 ) + ∥ ( X 1 − X 2 ) sss ∥ H ˙ 1/2 1/3 ∥ ( X 1 − X 2 ) s ∥ H 1 2/3 ( ∥ ( X 1 ) sss ∥ H ˙ 1/2 1/3 + ∥ ( X 2 ) sss ∥ H ˙ 1/2 1/3 ) + ∥ ( X 1 − X 2 ) sss ∥ H ˙ 1/2 3/4 ∥ ( X 1 − X 2 ) s ∥ H 1 1/4 ] . (44)
Note that the right hand side of (43) involves less than one copy of the highest derivative of X \mathbf{X} X (here we use ∥ X s s s ∥ H ˙ 1 / 2 3 / 4 \|X_{sss}\|_{\dot{H}^{1/2}}^{3/4} ∥ X sss ∥ H ˙ 1/2 3/4 ). In contrast, the analogous bound (1.17) in the resistive force theory setting involves a full copy of the highest derivative of the curve in the corresponding energy space, which makes the tension behavior there critical.
Proof. For ( τ , φ ) ∈ H 1 / 2 ( T ) × H 1 / 2 ( T ) (\tau,\varphi)\in H^{1/2}(\mathbb{T})\times H^{1/2}(\mathbb{T}) ( τ , φ ) ∈ H 1/2 ( T ) × H 1/2 ( T ) , we may define the bilinear form
B ϵ ( τ , φ ) : = ∫ T L ‾ ϵ [ ( τ X s ) s ] ⋅ ( φ X s ) s d s , (45) B_\epsilon(\tau,\varphi):=\int_{\mathbb{T}}\overline{\mathcal{L}}_\epsilon[(\tau\mathbf{X}_s)_s]\cdot(\varphi\mathbf{X}_s)_s\,ds, \tag*{(45)} B ϵ ( τ , φ ) := ∫ T L ϵ [( τ X s ) s ] ⋅ ( φ X s ) s d s , (45)
which, using Parseval’s theorem, may be seen to satisfy
B ϵ ( τ , φ ) = ∫ T ( ( φ s X s ) ⋅ T ϵ m t ( τ s X s ) + ( φ X s s ) ⋅ T ϵ m n ( τ X s s ) ) d s = ∑ ∣ k ∣ = 0 ∞ ( m ϵ t ( k ) F [ φ s X s ] ⋅ F [ τ s X s ] ‾ + m ϵ n ( k ) F [ φ X s s ] ⋅ F [ τ X s s ] ‾ ) = ∫ T ( T ϵ m t ( φ s X s ) ⋅ ( τ s X s ) + T ϵ m n ( φ X s s ) ⋅ ( τ X s s ) ) d s = B ϵ ( φ , τ ) . (46) \begin{aligned}
B_\epsilon(\tau,\varphi)
={}&\int_{\mathbb{T}}\left((\varphi_s\mathbf{X}_s)\cdot T_\epsilon^{\mathrm{mt}}(\tau_s\mathbf{X}_s)+(\varphi\mathbf{X}_{ss})\cdot T_\epsilon^{\mathrm{mn}}(\tau\mathbf{X}_{ss})\right)\,ds \\
={}&\sum_{|k|=0}^{\infty}\left(m_\epsilon^{\mathrm{t}}(k)\mathcal{F}[\varphi_s\mathbf{X}_s]\cdot\overline{\mathcal{F}[\tau_s\mathbf{X}_s]}+m_\epsilon^{\mathrm{n}}(k)\mathcal{F}[\varphi\mathbf{X}_{ss}]\cdot\overline{\mathcal{F}[\tau\mathbf{X}_{ss}]}\right) \\
={}&\int_{\mathbb{T}}\left(T_\epsilon^{\mathrm{mt}}(\varphi_s\mathbf{X}_s)\cdot(\tau_s\mathbf{X}_s)+T_\epsilon^{\mathrm{mn}}(\varphi\mathbf{X}_{ss})\cdot(\tau\mathbf{X}_{ss})\right)\,ds \\
={}&B_\epsilon(\varphi,\tau). \tag*{(46)}
\end{aligned} B ϵ ( τ , φ ) = = = = ∫ T ( ( φ s X s ) ⋅ T ϵ mt ( τ s X s ) + ( φ X ss ) ⋅ T ϵ mn ( τ X ss ) ) d s ∣ k ∣ = 0 ∑ ∞ ( m ϵ t ( k ) F [ φ s X s ] ⋅ F [ τ s X s ] + m ϵ n ( k ) F [ φ X ss ] ⋅ F [ τ X ss ] ) ∫ T ( T ϵ mt ( φ s X s ) ⋅ ( τ s X s ) + T ϵ mn ( φ X ss ) ⋅ ( τ X ss ) ) d s B ϵ ( φ , τ ) . (46)
In addition, using the multiplier bounds of Lemma 2.1 and Parseval’s theorem, we may bound
∣ B ϵ ( τ , φ ) ∣ ≤ c ∣ log ϵ ∣ ∑ ∣ k ∣ = 0 ⌊ 1 2 π ϵ ⌋ ( ∣ F [ φ s X s ] ∣ ∣ F [ τ s X s ] ∣ + ∣ F [ φ X s s ] ∣ ∣ F [ τ X s s ] ∣ ) + c ϵ − 1 ∑ ∣ k ∣ = ⌊ 1 2 π ϵ ⌋ + 1 ∞ ∣ k ∣ − 1 ( ∣ F [ φ s X s ] ∣ ∣ F [ τ s X s ] ∣ + ∣ F [ φ X s s ] ∣ ∣ F [ τ X s s ] ∣ ) ≤ c ϵ − 1 ∣ log ϵ ∣ ( ∥ φ s X s ∥ H − 1 / 2 ∥ τ s X s ∥ H − 1 / 2 + ∥ φ X s s ∥ H − 1 / 2 ∥ τ X s s ∥ H − 1 / 2 ) ≤ c ϵ − 1 ∣ log ϵ ∣ ∥ X s ∥ H 1 2 ∥ τ ∥ H 1 / 2 ∥ φ ∥ H 1 / 2 . (47) \begin{aligned}
|B_\epsilon(\tau,\varphi)|\leq{}&c|\log\epsilon|\sum_{|k|=0}^{\lfloor\frac{1}{2\pi\epsilon}\rfloor}\left(|\mathcal{F}[\varphi_s\mathbf{X}_s]||\mathcal{F}[\tau_s\mathbf{X}_s]|+|\mathcal{F}[\varphi\mathbf{X}_{ss}]||\mathcal{F}[\tau\mathbf{X}_{ss}]|\right)\\
&+c\epsilon^{-1}\sum_{|k|=\lfloor\frac{1}{2\pi\epsilon}\rfloor+1}^{\infty}|k|^{-1}\left(|\mathcal{F}[\varphi_s\mathbf{X}_s]||\mathcal{F}[\tau_s\mathbf{X}_s]|+|\mathcal{F}[\varphi\mathbf{X}_{ss}]||\mathcal{F}[\tau\mathbf{X}_{ss}]|\right)\\
\leq{}&c\epsilon^{-1}|\log\epsilon|\left(\|\varphi_s\mathbf{X}_s\|_{H^{-1/2}}\|\tau_s\mathbf{X}_s\|_{H^{-1/2}}+\|\varphi\mathbf{X}_{ss}\|_{H^{-1/2}}\|\tau\mathbf{X}_{ss}\|_{H^{-1/2}}\right)\\
\leq{}&c\epsilon^{-1}|\log\epsilon|\|\mathbf{X}_s\|_{H^1}^{2}\|\tau\|_{H^{1/2}}\|\varphi\|_{H^{1/2}}. \tag*{(47)}
\end{aligned} ∣ B ϵ ( τ , φ ) ∣ ≤ ≤ ≤ c ∣ log ϵ ∣ ∣ k ∣ = 0 ∑ ⌊ 2 π ϵ 1 ⌋ ( ∣ F [ φ s X s ] ∣∣ F [ τ s X s ] ∣ + ∣ F [ φ X ss ] ∣∣ F [ τ X ss ] ∣ ) + c ϵ − 1 ∣ k ∣ = ⌊ 2 π ϵ 1 ⌋ + 1 ∑ ∞ ∣ k ∣ − 1 ( ∣ F [ φ s X s ] ∣∣ F [ τ s X s ] ∣ + ∣ F [ φ X ss ] ∣∣ F [ τ X ss ] ∣ ) c ϵ − 1 ∣ log ϵ ∣ ( ∥ φ s X s ∥ H − 1/2 ∥ τ s X s ∥ H − 1/2 + ∥ φ X ss ∥ H − 1/2 ∥ τ X ss ∥ H − 1/2 ) c ϵ − 1 ∣ log ϵ ∣∥ X s ∥ H 1 2 ∥ τ ∥ H 1/2 ∥ φ ∥ H 1/2 . (47)
Finally, we note that, by Proposition 2.4 , we may bound B ϵ ( τ , τ ) B_\epsilon(\tau,\tau) B ϵ ( τ , τ ) below as
B ϵ ( τ , τ ) ≥ c ∣ log ϵ ∣ ( ∥ τ s X s ∥ H − 1 / 2 ( T ) 2 + ∥ τ X s s ∥ H − 1 / 2 ( T ) 2 ) . (48) B_\epsilon(\tau,\tau)\geq c|\log\epsilon|\left(\|\tau_s\mathbf{X}_s\|_{H^{-1/2}(\mathbb{T})}^{2}+\|\tau\mathbf{X}_{ss}\|_{H^{-1/2}(\mathbb{T})}^{2}\right). \tag*{(48)} B ϵ ( τ , τ ) ≥ c ∣ log ϵ ∣ ( ∥ τ s X s ∥ H − 1/2 ( T ) 2 + ∥ τ X ss ∥ H − 1/2 ( T ) 2 ) . (48)
Using Proposition 3.1 , bound (38) , we may estimate
∥ τ s ∥ H − 1 / 2 ≤ c ∥ X s ∥ H 1 ∥ τ s X s ∥ H − 1 / 2 . (49) \|\tau_s\|_{H^{-1/2}}\leq c\|\mathbf{X}_s\|_{H^1}\|\tau_s\mathbf{X}_s\|_{H^{-1/2}}. \tag*{(49)} ∥ τ s ∥ H − 1/2 ≤ c ∥ X s ∥ H 1 ∥ τ s X s ∥ H − 1/2 . (49)
To obtain a bound for the L 2 L^2 L 2 part of τ \tau τ , as in [1 ] , we will need Fenchel’s Theorem [6 ] : A closed C 2 C^2 C 2 curve X \mathbf{X} X must satisfy
∫ T ∣ X s s ∣ 2 d s ≥ ∫ T ∣ X s s ∣ d s ≥ 2 π . (50) \int_{\mathbb{T}} |\mathbf{X}_{ss}|^2\,ds \ge\int_{\mathbb{T}} |\mathbf{X}_{ss}|\,ds \ge2\pi. \tag*{(50)} ∫ T ∣ X ss ∣ 2 d s ≥ ∫ T ∣ X ss ∣ d s ≥ 2 π . (50)
Letting τ 0 \tau_0 τ 0 denote the mean of τ \tau τ ,
τ 0 = ∫ T τ d s , (51) \tau_0=\int_{\mathbb{T}}\tau\,ds , \tag*{(51)} τ 0 = ∫ T τ d s , (51)
on T \mathbb{T} T , we have
∥ τ ∥ L 2 2 ≤ ∥ τ − τ 0 ∥ L 2 2 + ∥ τ 0 ∥ L 2 2 ≤ c ( ∥ τ ∥ H ˙ 1 / 2 2 + ∣ τ 0 ∣ 2 ) ≤ c ( ∥ X s s ∥ H 1 2 ∥ τ s X s ∥ H − 1 / 2 2 + ∣ τ 0 ∣ 2 ) . (52) \begin{aligned}
\|\tau\|_{L^2}^2 &\le\|\tau-\tau_0\|_{L^2}^2+\|\tau_0\|_{L^2}^2 \le c(\|\tau\|_{\dot{H}^{1/2}}^2+|\tau_0|^2)\\
&\le c(\|\mathbf{X}_{ss}\|_{H^1}^2\|\tau_s\mathbf{X}_s\|_{H^{-1/2}}^2+|\tau_0|^2). \tag*{(52)}
\end{aligned} ∥ τ ∥ L 2 2 ≤ ∥ τ − τ 0 ∥ L 2 2 + ∥ τ 0 ∥ L 2 2 ≤ c ( ∥ τ ∥ H ˙ 1/2 2 + ∣ τ 0 ∣ 2 ) ≤ c ( ∥ X ss ∥ H 1 2 ∥ τ s X s ∥ H − 1/2 2 + ∣ τ 0 ∣ 2 ) . (52)
We may estimate τ 0 \tau_0 τ 0 as
∣ τ 0 ∣ = ∣ τ 0 ∫ T ∣ X s s ∣ 2 d s ∥ X s s ∥ L 2 2 ∣ = 1 ∥ X s s ∥ L 2 2 ∣ ∫ T τ 0 X s s ⋅ X s s d s ∣ ≤ 1 2 π ∣ ∫ T τ 0 X s s ⋅ X s s ∥ X s s ∥ H 1 / 2 d s ∣ ≤ c ∥ τ 0 X s s ∥ H − 1 / 2 . (53) \begin{aligned}
|\tau_0|&=\left|\tau_0\frac{\int_{\mathbb{T}}|\mathbf{X}_{ss}|^2\,ds}{\|\mathbf{X}_{ss}\|_{L^2}^2}\right|=\frac{1}{\|\mathbf{X}_{ss}\|_{L^2}^2}\left|\int_{\mathbb{T}}\tau_0\mathbf{X}_{ss}\cdot\mathbf{X}_{ss}\,ds\right|\\
&\le\frac{1}{2\pi}\left|\int_{\mathbb{T}}\tau_0\mathbf{X}_{ss}\cdot\frac{\mathbf{X}_{ss}}{\|\mathbf{X}_{ss}\|_{H^{1/2}}}\,ds\right|\le c\|\tau_0\mathbf{X}_{ss}\|_{H^{-1/2}} . \tag*{(53)}
\end{aligned} ∣ τ 0 ∣ = τ 0 ∥ X ss ∥ L 2 2 ∫ T ∣ X ss ∣ 2 d s = ∥ X ss ∥ L 2 2 1 ∫ T τ 0 X ss ⋅ X ss d s ≤ 2 π 1 ∫ T τ 0 X ss ⋅ ∥ X ss ∥ H 1/2 X ss d s ≤ c ∥ τ 0 X ss ∥ H − 1/2 . (53)
Noting that for any v ∈ H 1 / 2 ( T ) \boldsymbol{v}\in H^{1/2}(\mathbb{T}) v ∈ H 1/2 ( T ) with ∥ v ∥ H 1 / 2 = 1 \|\boldsymbol{v}\|_{H^{1/2}}=1 ∥ v ∥ H 1/2 = 1 we may calculate
∣ ∫ T ( τ − τ 0 ) X s s ⋅ v d s ∣ ≤ ∥ τ − τ 0 ∥ L 2 ∥ X s s ∥ L 2 ∥ v ∥ L ∞ ≤ c ∥ τ ∥ H ˙ 1 / 2 ∥ X s s ∥ L 2 ≤ c ∥ X s s ∥ H 1 2 ∥ τ s X s ∥ H − 1 / 2 , (54) \begin{aligned}
\left|\int_{\mathbb{T}}(\tau-\tau_0)\mathbf{X}_{ss}\cdot\boldsymbol{v}\,ds\right|&\le\|\tau-\tau_0\|_{L^2}\|\mathbf{X}_{ss}\|_{L^2}\|\boldsymbol{v}\|_{L^\infty}\le c\|\tau\|_{\dot{H}^{1/2}}\|\mathbf{X}_{ss}\|_{L^2}\\
&\le c\|\mathbf{X}_{ss}\|_{H^1}^2\|\tau_s\mathbf{X}_s\|_{H^{-1/2}}, \tag*{(54)}
\end{aligned} ∫ T ( τ − τ 0 ) X ss ⋅ v d s ≤ ∥ τ − τ 0 ∥ L 2 ∥ X ss ∥ L 2 ∥ v ∥ L ∞ ≤ c ∥ τ ∥ H ˙ 1/2 ∥ X ss ∥ L 2 ≤ c ∥ X ss ∥ H 1 2 ∥ τ s X s ∥ H − 1/2 , (54)
we may bound
∥ τ 0 X s s ∥ H − 1 / 2 ≤ ∥ ( τ − τ 0 ) X s s ∥ H − 1 / 2 + ∥ τ X s s ∥ H − 1 / 2 ≤ c ∥ X s s ∥ H 1 2 ∥ τ s X s ∥ H − 1 / 2 + ∥ τ X s s ∥ H − 1 / 2 . (55) \begin{aligned}
\|\tau_0\mathbf{X}_{ss}\|_{H^{-1/2}}&\le\|(\tau-\tau_0)\mathbf{X}_{ss}\|_{H^{-1/2}}+\|\tau\mathbf{X}_{ss}\|_{H^{-1/2}}\\
&\le c\|\mathbf{X}_{ss}\|_{H^1}^2\|\tau_s\mathbf{X}_s\|_{H^{-1/2}}+\|\tau\mathbf{X}_{ss}\|_{H^{-1/2}} . \tag*{(55)}
\end{aligned} ∥ τ 0 X ss ∥ H − 1/2 ≤ ∥ ( τ − τ 0 ) X ss ∥ H − 1/2 + ∥ τ X ss ∥ H − 1/2 ≤ c ∥ X ss ∥ H 1 2 ∥ τ s X s ∥ H − 1/2 + ∥ τ X ss ∥ H − 1/2 . (55)
Thus the mean of τ \tau τ satisfies
∣ τ 0 ∣ ≤ c ( ∥ X s ∥ H 1 2 ∥ τ s X s ∥ H − 1 / 2 + ∥ τ X s s ∥ H − 1 / 2 ) , (56) |\tau_0|\le c(\|\mathbf{X}_s\|_{H^1}^2\|\tau_s\mathbf{X}_s\|_{H^{-1/2}}+\|\tau\mathbf{X}_{ss}\|_{H^{-1/2}}) , \tag*{(56)} ∣ τ 0 ∣ ≤ c ( ∥ X s ∥ H 1 2 ∥ τ s X s ∥ H − 1/2 + ∥ τ X ss ∥ H − 1/2 ) , (56)
and altogether we may bound
∥ τ ∥ H 1 / 2 ≤ c ( ( ∥ X s ∥ H 1 + ∥ X s ∥ H 1 2 ) ∥ τ s X s ∥ H − 1 / 2 + ∥ τ X s s ∥ H − 1 / 2 ) ≤ c ∣ log ϵ ∣ − 1 / 2 ( 1 + ∥ X s ∥ H 1 2 ) B ϵ ( τ , τ ) 1 / 2 . (57) \begin{aligned}
\|\tau\|_{H^{1/2}}&\le c((\|\mathbf{X}_s\|_{H^1}+\|\mathbf{X}_s\|_{H^1}^2)\|\tau_s\mathbf{X}_s\|_{H^{-1/2}}+\|\tau\mathbf{X}_{ss}\|_{H^{-1/2}})\\
&\le c|\log\epsilon|^{-1/2}(1+\|\mathbf{X}_s\|_{H^1}^2)\mathcal{B}_{\epsilon}(\tau,\tau)^{1/2}. \tag*{(57)}
\end{aligned} ∥ τ ∥ H 1/2 ≤ c (( ∥ X s ∥ H 1 + ∥ X s ∥ H 1 2 ) ∥ τ s X s ∥ H − 1/2 + ∥ τ X ss ∥ H − 1/2 ) ≤ c ∣ log ϵ ∣ − 1/2 ( 1 + ∥ X s ∥ H 1 2 ) B ϵ ( τ , τ ) 1/2 . (57)
By the Lax-Milgram lemma, given g ∈ H − 1 / 2 ( T ) g\in H^{-1/2}(\mathbb{T}) g ∈ H − 1/2 ( T ) , there is then a unique τ ∈ H 1 / 2 ( T ) \tau\in H^{1/2}(\mathbb{T}) τ ∈ H 1/2 ( T ) satisfying
B ϵ ( τ , φ ) = ∫ T g φ d s ∀ φ ∈ H 1 / 2 ( T ) . (58) \mathcal{B}_{\epsilon}(\tau,\varphi)=\int_{\mathbb{T}}g\varphi\,ds \qquad\forall\ \varphi\in H^{1/2}(\mathbb{T}) . \tag*{(58)} B ϵ ( τ , φ ) = ∫ T g φ d s ∀ φ ∈ H 1/2 ( T ) . (58)
We consider g g g given by the right hand side of (3.7). In particular, upon multiplying both sides of (3.7) by τ \tau τ and integrating by parts over T \mathbb{T} T , we have
B ϵ ( τ , τ ) = ∫ T L ‾ ϵ [ X s s s s ] ⋅ ( τ X s ) s d s . (59) \mathcal{B}_{\epsilon}(\tau,\tau)=\int_{\mathbb{T}}\overline{\mathcal{L}}_{\epsilon}[\mathbf{X}_{ssss}]\cdot(\tau\mathbf{X}_s)_s\,ds . \tag*{(59)} B ϵ ( τ , τ ) = ∫ T L ϵ [ X ssss ] ⋅ ( τ X s ) s d s . (59)
Using Parseval’s theorem and the multiplier behavior of Lemma 2.1, the right hand side of (3.24) may be estimated as
∣ ∫ T L ϵ ‾ [ X s s s s ] ⋅ ( τ X s ) s d s ∣ ≤ ∣ ∫ T τ s X s ⋅ T m ϵ P X s X s s s s d s ∣ + ∣ ∫ T τ X s s ⋅ T m n P X s ⊥ X s s s s d s ∣ = ∣ ∑ ∣ k ∣ = 0 ∞ m k t ( E ) F [ τ s X s ] ⋅ F [ P X s X s s s s ] ‾ ∣ + ∣ ∑ ∣ k ∣ = 0 ∞ m k n ( E ) F [ τ X s s ] ⋅ F [ P X s ⊥ X s s s s ] ‾ ∣ ≤ c ∣ log ϵ ∣ ∣ ∑ ∣ k ∣ = 0 ⌊ 1 2 π ϵ ⌋ F [ τ s X s ] ⋅ F [ P X s X s s s s ] ‾ + F [ τ X s s ] ⋅ F [ P X s ⊥ X s s s s ] ‾ ∣ + c ϵ − 1 ∑ ∣ k ∣ = ⌊ 1 2 π ϵ ⌋ + 1 ∞ ∣ k ∣ − 1 ( ∣ F [ τ s X s ] ⋅ F [ P X s X s s s s ] ‾ ∣ + ∣ F [ τ X s s ] ⋅ F [ P X s ⊥ X s s s s ] ‾ ∣ ) ≤ c ϵ − 1 ∣ log ϵ ∣ ( ∥ τ s X s ∥ H − 1 / 2 ∥ P X s X s s s s ∥ H − 1 / 2 + ∥ τ X s s ∥ H − 1 / 8 ∥ P X s ⊥ X s s s s ∥ H − 7 / 8 ) . (60) \begin{aligned}
\left|\int_{\mathbb T}\overline{\mathcal L_\epsilon}[X_{ssss}]\cdot(\tau X_s)_s\,ds\right|
&\leq\left|\int_{\mathbb T}\tau_sX_s\cdot T_m^\epsilon\mathbb P_{X_s}X_{ssss}\,ds\right|
+\left|\int_{\mathbb T}\tau X_{ss}\cdot T_m^n\mathbb P^\perp_{X_s}X_{ssss}\,ds\right| \\
&=\left|\sum_{|k|=0}^{\infty}m_k^t(\mathcal E)\mathcal F[\tau_sX_s]\cdot\overline{\mathcal F[\mathbb P_{X_s}X_{ssss}]}\right|
+\left|\sum_{|k|=0}^{\infty}m_k^n(\mathcal E)\mathcal F[\tau X_{ss}]\cdot\overline{\mathcal F[\mathbb P^\perp_{X_s}X_{ssss}]}\right| \\
&\leq c|\log\epsilon|\left|\sum_{|k|=0}^{\left\lfloor\frac{1}{2\pi\epsilon}\right\rfloor}\mathcal F[\tau_sX_s]\cdot\overline{\mathcal F[\mathbb P_{X_s}X_{ssss}]}+\mathcal F[\tau X_{ss}]\cdot\overline{\mathcal F[\mathbb P^\perp_{X_s}X_{ssss}]}\right| \\
&\quad+c\epsilon^{-1}\sum_{|k|=\left\lfloor\frac{1}{2\pi\epsilon}\right\rfloor+1}^{\infty}|k|^{-1}\left(\left|\mathcal F[\tau_sX_s]\cdot\overline{\mathcal F[\mathbb P_{X_s}X_{ssss}]}\right|
+\left|\mathcal F[\tau X_{ss}]\cdot\overline{\mathcal F[\mathbb P^\perp_{X_s}X_{ssss}]}\right|\right) \\
&\leq c\epsilon^{-1}|\log\epsilon|\left(\|\tau_sX_s\|_{H^{-1/2}}\|\mathbb P_{X_s}X_{ssss}\|_{H^{-1/2}}
+\|\tau X_{ss}\|_{H^{-1/8}}\|\mathbb P^\perp_{X_s}X_{ssss}\|_{H^{-7/8}}\right). \tag*{(60)}
\end{aligned} ∫ T L ϵ [ X ssss ] ⋅ ( τ X s ) s d s ≤ ∫ T τ s X s ⋅ T m ϵ P X s X ssss d s + ∫ T τ X ss ⋅ T m n P X s ⊥ X ssss d s = ∣ k ∣ = 0 ∑ ∞ m k t ( E ) F [ τ s X s ] ⋅ F [ P X s X ssss ] + ∣ k ∣ = 0 ∑ ∞ m k n ( E ) F [ τ X ss ] ⋅ F [ P X s ⊥ X ssss ] ≤ c ∣ log ϵ ∣ ∣ k ∣ = 0 ∑ ⌊ 2 π ϵ 1 ⌋ F [ τ s X s ] ⋅ F [ P X s X ssss ] + F [ τ X ss ] ⋅ F [ P X s ⊥ X ssss ] + c ϵ − 1 ∣ k ∣ = ⌊ 2 π ϵ 1 ⌋ + 1 ∑ ∞ ∣ k ∣ − 1 ( F [ τ s X s ] ⋅ F [ P X s X ssss ] + F [ τ X ss ] ⋅ F [ P X s ⊥ X ssss ] ) ≤ c ϵ − 1 ∣ log ϵ ∣ ( ∥ τ s X s ∥ H − 1/2 ∥ P X s X ssss ∥ H − 1/2 + ∥ τ X ss ∥ H − 1/8 ∥ P X s ⊥ X ssss ∥ H − 7/8 ) . (60)
Note that in the final inequality, for the term corresponding to the normal directions along the filament, we crucially make use of the fact that we can share the ∣ k ∣ − 1 |k|^{-1} ∣ k ∣ − 1 equally between X s s X_{ss} X ss and P X s ⊥ X s s s s \mathbb P^\perp_{X_s}X_{ssss} P X s ⊥ X ssss . In particular, we can leverage the additional regularity of τ X s s \tau X_{ss} τ X ss (compared to τ s X s \tau_sX_s τ s X s ) and require less from the normal direction term P X s ⊥ X s s s s \mathbb P^\perp_{X_s}X_{ssss} P X s ⊥ X ssss . This will be useful for closing a fixed point argument later on.
We may estimate
∥ τ X s s ∥ H − 1 / 8 ≤ c ∥ X s s ∥ L 2 ∥ τ ∥ L p (61) \|\tau X_{ss}\|_{H^{-1/8}}\leq c\|X_{ss}\|_{L^2}\|\tau\|_{L^p} \tag*{(61)} ∥ τ X ss ∥ H − 1/8 ≤ c ∥ X ss ∥ L 2 ∥ τ ∥ L p (61)
for any p > 2 p>2 p > 2 ; in particular, we have
∥ τ X s s ∥ H − 1 / 8 ≤ c ∥ X s s ∥ L 2 ∥ τ ∥ H 1 / 2 . (62) \|\tau X_{ss}\|_{H^{-1/8}}\leq c\|X_{ss}\|_{L^2}\|\tau\|_{H^{1/2}}. \tag*{(62)} ∥ τ X ss ∥ H − 1/8 ≤ c ∥ X ss ∥ L 2 ∥ τ ∥ H 1/2 . (62)
Furthermore, by the triangle inequality and interpolation, we may bound
∥ P X s ⊥ X s s s s ∥ H − 7 / 8 ≤ ∥ X s s ∥ H ˙ 9 / 8 + ∥ P X s X s s s s ∥ H − 7 / 8 ≤ c ∥ X s s ∥ H ˙ 3 / 2 3 / 4 ∥ X s s ∥ L 2 1 / 4 + c ∥ P X s X s s s s ∥ H − 1 / 2 . (63) \begin{aligned}
\|\mathbb P^\perp_{X_s}X_{ssss}\|_{H^{-7/8}}
&\leq\|X_{ss}\|_{\dot H^{9/8}}+\|\mathbb P_{X_s}X_{ssss}\|_{H^{-7/8}} \\
&\leq c\|X_{ss}\|_{\dot H^{3/2}}^{3/4}\|X_{ss}\|_{L^2}^{1/4}
+c\|\mathbb P_{X_s}X_{ssss}\|_{H^{-1/2}}. \tag*{(63)}
\end{aligned} ∥ P X s ⊥ X ssss ∥ H − 7/8 ≤ ∥ X ss ∥ H ˙ 9/8 + ∥ P X s X ssss ∥ H − 7/8 ≤ c ∥ X ss ∥ H ˙ 3/2 3/4 ∥ X ss ∥ L 2 1/4 + c ∥ P X s X ssss ∥ H − 1/2 . (63)
So far, we may thus bound B ϵ ( τ , τ ) \mathcal B_\epsilon(\tau,\tau) B ϵ ( τ , τ ) as
B ϵ ( τ , τ ) ≤ c ϵ − 1 ∣ log ϵ ∣ ∥ τ ∥ H 1 / 2 ( ∥ X s s ∥ L 2 ∥ P X s X s s s s ∥ H − 1 / 2 + ∥ X s s ∥ H ˙ 1 / 2 3 / 4 ∥ X s s ∥ L 2 5 / 4 ) . (64) \mathcal B_\epsilon(\tau,\tau)\leq c\epsilon^{-1}|\log\epsilon|\|\tau\|_{H^{1/2}}\left(\|X_{ss}\|_{L^2}\|\mathbb P_{X_s}X_{ssss}\|_{H^{-1/2}}+\|X_{ss}\|_{\dot H^{1/2}}^{3/4}\|X_{ss}\|_{L^2}^{5/4}\right). \tag*{(64)} B ϵ ( τ , τ ) ≤ c ϵ − 1 ∣ log ϵ ∣∥ τ ∥ H 1/2 ( ∥ X ss ∥ L 2 ∥ P X s X ssss ∥ H − 1/2 + ∥ X ss ∥ H ˙ 1/2 3/4 ∥ X ss ∥ L 2 5/4 ) . (64)
Using the identities (3.1) coming from the inextensibility constraint, we may write the tangential part of X s s s s X_{ssss} X ssss as
P X s X s s s s = − 3 2 ∂ s ∣ X s s ∣ 2 X s . (65) \mathbb P_{X_s}X_{ssss}=-\frac{3}{2}\partial_s|X_{ss}|^2X_s. \tag*{(65)} P X s X ssss = − 2 3 ∂ s ∣ X ss ∣ 2 X s . (65)
By Proposition 3.1, bound (3.4), along with the interpolation inequalities (3.2), we may then estimate P X s X s s s s \mathbb P_{X_s}X_{ssss} P X s X ssss as
∥ P X s X s s s s ∥ H − 1 / 2 ( T ) ≤ c ∥ ∣ X s s ∣ 2 ∥ H ˙ 1 / 2 ∥ X s s ∥ H 1 ≤ c ∥ ∂ s ∣ X s s ∣ 2 ∥ L 2 1 / 2 ∥ ∣ X s s ∣ 2 ∥ L 2 1 / 2 ∥ X s s ∥ H 1 ≤ c ∥ X s s s s ∥ H 1 1 / 2 ∥ X s s ∥ L ∞ ∥ X s s ∥ H 1 3 / 2 ≤ c ∥ X s s s s ∥ H 1 2 / 3 ∥ X s s ∥ H 1 7 / 3 . (66) \begin{aligned}
\|\mathbb P_{X_s}X_{ssss}\|_{H^{-1/2}(\mathbb T)}
&\leq c\||X_{ss}|^2\|_{\dot H^{1/2}}\|X_{ss}\|_{H^1}
\leq c\|\partial_s|X_{ss}|^2\|_{L^2}^{1/2}\||X_{ss}|^2\|_{L^2}^{1/2}\|X_{ss}\|_{H^1} \\
&\leq c\|X_{ssss}\|_{H^1}^{1/2}\|X_{ss}\|_{L^\infty}\|X_{ss}\|_{H^1}^{3/2}
\leq c\|X_{ssss}\|_{H^1}^{2/3}\|X_{ss}\|_{H^1}^{7/3}. \tag*{(66)}
\end{aligned} ∥ P X s X ssss ∥ H − 1/2 ( T ) ≤ c ∥∣ X ss ∣ 2 ∥ H ˙ 1/2 ∥ X ss ∥ H 1 ≤ c ∥ ∂ s ∣ X ss ∣ 2 ∥ L 2 1/2 ∥∣ X ss ∣ 2 ∥ L 2 1/2 ∥ X ss ∥ H 1 ≤ c ∥ X ssss ∥ H 1 1/2 ∥ X ss ∥ L ∞ ∥ X ss ∥ H 1 3/2 ≤ c ∥ X ssss ∥ H 1 2/3 ∥ X ss ∥ H 1 7/3 . (66)
We thus obtain
B ϵ ( τ , τ ) ≤ c ϵ − 1 ∣ log ϵ ∣ ∥ τ ∥ H 1 / 2 ( ∥ X s s s s ∥ H ˙ 1 / 2 3 / 4 ∥ X s s ∥ H 1 5 / 4 + ∥ X s s s s ∥ H ˙ 1 / 2 2 / 3 ∥ X s s ∥ H 1 10 / 3 ) . (67) \mathcal B_\epsilon(\tau,\tau)\leq c\epsilon^{-1}|\log\epsilon|\|\tau\|_{H^{1/2}}\left(\|X_{ssss}\|_{\dot H^{1/2}}^{3/4}\|X_{ss}\|_{H^1}^{5/4}+\|X_{ssss}\|_{\dot H^{1/2}}^{2/3}\|X_{ss}\|_{H^1}^{10/3}\right). \tag*{(67)} B ϵ ( τ , τ ) ≤ c ϵ − 1 ∣ log ϵ ∣∥ τ ∥ H 1/2 ( ∥ X ssss ∥ H ˙ 1/2 3/4 ∥ X ss ∥ H 1 5/4 + ∥ X ssss ∥ H ˙ 1/2 2/3 ∥ X ss ∥ H 1 10/3 ) . (67)
The lower bound (57) then yields
∥ τ ∥ H 1 / 2 2 ≤ c ϵ − 1 ∥ τ ∥ H 1 / 2 ( 1 + ∥ X s ∥ H 1 4 ) ( ∥ X s s s ∥ H ˙ 1 / 2 3 / 4 ∥ X s ∥ H 1 5 / 4 + ∥ X s s s ∥ H ˙ 1 / 2 2 / 3 ∥ X s ∥ H 1 10 / 3 ) , (68) \|\tau\|_{H^{1/2}}^2 \le c\epsilon^{-1}\|\tau\|_{H^{1/2}}(1+\|X_s\|_{H^1}^4)\left(\|X_{sss}\|_{\dot H^{1/2}}^{3/4}\|X_s\|_{H^1}^{5/4}+\|X_{sss}\|_{\dot H^{1/2}}^{2/3}\|X_s\|_{H^1}^{10/3}\right), \tag*{(68)} ∥ τ ∥ H 1/2 2 ≤ c ϵ − 1 ∥ τ ∥ H 1/2 ( 1 + ∥ X s ∥ H 1 4 ) ( ∥ X sss ∥ H ˙ 1/2 3/4 ∥ X s ∥ H 1 5/4 + ∥ X sss ∥ H ˙ 1/2 2/3 ∥ X s ∥ H 1 10/3 ) , (68)
and an application of Young’s inequality yields (43) .
We next turn to the Lipschitz bound (44) . Let X 1 X_1 X 1 , X 2 X_2 X 2 denote two nearby curves3 , and let τ 1 \tau_1 τ 1 , τ 2 \tau_2 τ 2 satisfy the corresponding tension determination problems (42) . Letting B ϵ , 1 B_{\epsilon,1} B ϵ , 1 , B ϵ , 2 B_{\epsilon,2} B ϵ , 2 denote the associated bilinear forms (45) , we consider the weak form of the tension equations satisfied by τ 1 \tau_1 τ 1 and τ 2 \tau_2 τ 2 . In particular, for any φ ∈ H 1 / 2 ( T ) \varphi\in H^{1/2}(\mathbb{T}) φ ∈ H 1/2 ( T ) , we have that τ 1 \tau_1 τ 1 and τ 2 \tau_2 τ 2 satisfy
B ϵ , 1 ( τ 1 , φ ) − B ϵ , 2 ( τ 2 , φ ) = ∫ T L ‾ ϵ , 1 [ ( X 1 ) s s s s ] ⋅ ( φ ( X 1 ) s ) s d s − ∫ T L ‾ ϵ , 2 [ ( X 2 ) s s s s ] ⋅ ( φ ( X 2 ) s ) s d s ⏟ R H S . (69) B_{\epsilon,1}(\tau_1,\varphi)-B_{\epsilon,2}(\tau_2,\varphi)
=
\underbrace{\int_{\mathbb{T}}\overline{\mathcal{L}}_{\epsilon,1}[(X_1)_{ssss}]\cdot(\varphi(X_1)_s)_s\,ds-\int_{\mathbb{T}}\overline{\mathcal{L}}_{\epsilon,2}[(X_2)_{ssss}]\cdot(\varphi(X_2)_s)_s\,ds}_{\mathrm{RHS}}. \tag*{(69)} B ϵ , 1 ( τ 1 , φ ) − B ϵ , 2 ( τ 2 , φ ) = RHS ∫ T L ϵ , 1 [( X 1 ) ssss ] ⋅ ( φ ( X 1 ) s ) s d s − ∫ T L ϵ , 2 [( X 2 ) ssss ] ⋅ ( φ ( X 2 ) s ) s d s . (69)
Note that, using the form (45) of B ϵ B_\epsilon B ϵ , we may write
B ϵ , 1 ( τ 1 , φ ) − B ϵ , 2 ( τ 2 , φ ) = B ϵ , 2 ( τ 1 − τ 2 , φ ) + L H S , B_{\epsilon,1}(\tau_1,\varphi)-B_{\epsilon,2}(\tau_2,\varphi)=B_{\epsilon,2}(\tau_1-\tau_2,\varphi)+\mathrm{LHS}, B ϵ , 1 ( τ 1 , φ ) − B ϵ , 2 ( τ 2 , φ ) = B ϵ , 2 ( τ 1 − τ 2 , φ ) + LHS ,
L H S = ∫ T ( φ s ( X 1 − X 2 ) s ⋅ T ϵ m t ( ( τ 1 ) s ( X 1 ) s ) + φ ( X 1 − X 2 ) s s ⋅ T ϵ m n ( τ 1 ( X 1 ) s s ) ) d s + ∫ T ( φ s ( X 2 ) s ⋅ T ϵ m t ( ( τ 1 ) s ( X 1 − X 2 ) s ) + φ ( X 2 ) s s ⋅ T ϵ m n ( τ 1 ( X 1 − X 2 ) s s ) ) d s . (70) \begin{aligned}
\mathrm{LHS}={}&\int_{\mathbb{T}}\left(\varphi_s(X_1-X_2)_s\cdot T^{\mathrm{mt}}_{\epsilon}((\tau_1)_s(X_1)_s)+\varphi(X_1-X_2)_{ss}\cdot T^{\mathrm{mn}}_{\epsilon}(\tau_1(X_1)_{ss})\right)\,ds\\
&+\int_{\mathbb{T}}\left(\varphi_s(X_2)_s\cdot T^{\mathrm{mt}}_{\epsilon}((\tau_1)_s(X_1-X_2)_s)+\varphi(X_2)_{ss}\cdot T^{\mathrm{mn}}_{\epsilon}(\tau_1(X_1-X_2)_{ss})\right)\,ds. \tag*{(70)}
\end{aligned} LHS = ∫ T ( φ s ( X 1 − X 2 ) s ⋅ T ϵ mt (( τ 1 ) s ( X 1 ) s ) + φ ( X 1 − X 2 ) ss ⋅ T ϵ mn ( τ 1 ( X 1 ) ss ) ) d s + ∫ T ( φ s ( X 2 ) s ⋅ T ϵ mt (( τ 1 ) s ( X 1 − X 2 ) s ) + φ ( X 2 ) ss ⋅ T ϵ mn ( τ 1 ( X 1 − X 2 ) ss ) ) d s . (70)
Following the estimates above for a single curve, we may bound the terms in LHS as
∣ L H S ∣ ≤ c ϵ − 1 ∣ log ϵ ∣ ( ∥ φ s ( X 1 − X 2 ) s ∥ H − 1 / 2 ∥ ( τ 1 ) s ( X 1 ) s ∥ H − 1 / 2 + ∥ φ ( X 1 − X 2 ) s s ∥ H − 1 / 2 ∥ τ 1 ( X 1 ) s s ∥ H − 1 / 2 + ∥ φ s ( X 2 ) s ∥ H − 1 / 2 ∥ ( τ 1 ) s ( X 1 − X 2 ) s ∥ H − 1 / 2 + ∥ φ ( X 2 ) s s ∥ H − 1 / 2 ∥ τ 1 ( X 1 − X 2 ) s s ∥ H − 1 / 2 ) ≤ c ϵ − 1 ∣ log ϵ ∣ ∥ φ ∥ H 1 / 2 ∥ ( X 1 − X 2 ) s ∥ H 1 ∥ τ 1 ∥ H 1 / 2 ( ∥ ( X 1 ) s ∥ H 1 + ∥ ( X 2 ) s ∥ H 1 ) ≤ c ϵ − 1 ∣ log ϵ ∣ ∥ φ ∥ H 1 / 2 ∥ ( X 1 − X 2 ) s ∥ H 1 ( ∥ ( X 1 ) s ∥ H 1 + ∥ ( X 2 ) s ∥ H 1 ) × ( 1 + ∥ ( X 1 ) s ∥ H 1 22 / 3 ) ( ∥ ( X 1 ) s s s ∥ H ˙ 1 / 2 3 / 4 + ∥ ( X 1 ) s s s ∥ H ˙ 1 / 2 2 / 3 ) . (71) \begin{aligned}
|\mathrm{LHS}| \le{}& c\epsilon^{-1}|\log\epsilon|\Bigl(\|\varphi_s(X_1-X_2)_s\|_{H^{-1/2}}\|(\tau_1)_s(X_1)_s\|_{H^{-1/2}}\\
&+\|\varphi(X_1-X_2)_{ss}\|_{H^{-1/2}}\|\tau_1(X_1)_{ss}\|_{H^{-1/2}}+\|\varphi_s(X_2)_s\|_{H^{-1/2}}\|(\tau_1)_s(X_1-X_2)_s\|_{H^{-1/2}}\\
&+\|\varphi(X_2)_{ss}\|_{H^{-1/2}}\|\tau_1(X_1-X_2)_{ss}\|_{H^{-1/2}}\Bigr)\\
\le{}&c\epsilon^{-1}|\log\epsilon|\|\varphi\|_{H^{1/2}}\|(X_1-X_2)_s\|_{H^1}\|\tau_1\|_{H^{1/2}}\left(\|(X_1)_s\|_{H^1}+\|(X_2)_s\|_{H^1}\right)\\
\le{}&c\epsilon^{-1}|\log\epsilon|\|\varphi\|_{H^{1/2}}\|(X_1-X_2)_s\|_{H^1}\left(\|(X_1)_s\|_{H^1}+\|(X_2)_s\|_{H^1}\right)\\
&\qquad\times\left(1+\|(X_1)_s\|_{H^1}^{22/3}\right)\left(\|(X_1)_{sss}\|_{\dot H^{1/2}}^{3/4}+\|(X_1)_{sss}\|_{\dot H^{1/2}}^{2/3}\right). \tag*{(71)}
\end{aligned} ∣ LHS ∣ ≤ ≤ ≤ c ϵ − 1 ∣ log ϵ ∣ ( ∥ φ s ( X 1 − X 2 ) s ∥ H − 1/2 ∥ ( τ 1 ) s ( X 1 ) s ∥ H − 1/2 + ∥ φ ( X 1 − X 2 ) ss ∥ H − 1/2 ∥ τ 1 ( X 1 ) ss ∥ H − 1/2 + ∥ φ s ( X 2 ) s ∥ H − 1/2 ∥ ( τ 1 ) s ( X 1 − X 2 ) s ∥ H − 1/2 + ∥ φ ( X 2 ) ss ∥ H − 1/2 ∥ τ 1 ( X 1 − X 2 ) ss ∥ H − 1/2 ) c ϵ − 1 ∣ log ϵ ∣∥ φ ∥ H 1/2 ∥ ( X 1 − X 2 ) s ∥ H 1 ∥ τ 1 ∥ H 1/2 ( ∥ ( X 1 ) s ∥ H 1 + ∥ ( X 2 ) s ∥ H 1 ) c ϵ − 1 ∣ log ϵ ∣∥ φ ∥ H 1/2 ∥ ( X 1 − X 2 ) s ∥ H 1 ( ∥ ( X 1 ) s ∥ H 1 + ∥ ( X 2 ) s ∥ H 1 ) × ( 1 + ∥ ( X 1 ) s ∥ H 1 22/3 ) ( ∥ ( X 1 ) sss ∥ H ˙ 1/2 3/4 + ∥ ( X 1 ) sss ∥ H ˙ 1/2 2/3 ) . (71)
Here in the final line, we have used the tension bound (43) (with some simplification).
In addition, we may bound the terms RHS in (69) as follows. Using the identity (79) for the tangential direction along X 1 X_1 X 1 and X 2 X_2 X 2 , we may write out RHS as
R H S = − 3 2 ∫ T ( φ s ( X 1 − X 2 ) s ⋅ T ϵ m t ( ∂ s ∣ ( X 1 ) s s ∣ 2 ( X 1 ) s ) + φ s ( X 2 ) s ⋅ T ϵ m t [ ∂ s ( ∣ ( X 1 ) s s ∣ 2 − ∣ ( X 2 ) s s ∣ 2 ) ( X 1 ) s + ∂ s ∣ ( X 2 ) s s ∣ 2 ( X 1 − X 2 ) s ] ) d s + ∫ T ( φ ( X 1 − X 2 ) s s ⋅ T ϵ m n ( ( P ( X 1 ) s ⊥ ) ( X 1 ) s s s s ) + φ ( X 2 ) s s ⋅ T ϵ m n ( ( ( P ( X 1 ) s ⊥ − P ( X 2 ) s ⊥ ) ( X 1 ) s s s s + P ( X 2 ) s ⊥ ( X 1 − X 2 ) s s s s ) ) d s . (72) \begin{aligned}
\mathrm{RHS}={}&-\frac{3}{2}\int_{\mathbb{T}}\left(\varphi_s(X_1-X_2)_s\cdot T^{\mathrm{mt}}_{\epsilon}\left(\partial_s\left|(X_1)_{ss}\right|^2(X_1)_s\right)\right.\\
&\qquad+\varphi_s(X_2)_s\cdot T^{\mathrm{mt}}_{\epsilon}\left[\partial_s\left(\left|(X_1)_{ss}\right|^2-\left|(X_2)_{ss}\right|^2\right)(X_1)_s+\partial_s\left|(X_2)_{ss}\right|^2(X_1-X_2)_s\right]\Bigr)\,ds\\
&+\int_{\mathbb{T}}\left(\varphi(X_1-X_2)_{ss}\cdot T^{\mathrm{mn}}_{\epsilon}\left((P^\perp_{(X_1)_s})(X_1)_{ssss}\right)\right.\\
&\qquad+\varphi(X_2)_{ss}\cdot T^{\mathrm{mn}}_{\epsilon}\left(\left((P^\perp_{(X_1)_s}-P^\perp_{(X_2)_s})(X_1)_{ssss}+P^\perp_{(X_2)_s}(X_1-X_2)_{ssss}\right)\right)\,ds. \tag*{(72)}
\end{aligned} RHS = − 2 3 ∫ T ( φ s ( X 1 − X 2 ) s ⋅ T ϵ mt ( ∂ s ∣ ( X 1 ) ss ∣ 2 ( X 1 ) s ) + φ s ( X 2 ) s ⋅ T ϵ mt [ ∂ s ( ∣ ( X 1 ) ss ∣ 2 − ∣ ( X 2 ) ss ∣ 2 ) ( X 1 ) s + ∂ s ∣ ( X 2 ) ss ∣ 2 ( X 1 − X 2 ) s ] ) d s + ∫ T ( φ ( X 1 − X 2 ) ss ⋅ T ϵ mn ( ( P ( X 1 ) s ⊥ ) ( X 1 ) ssss ) + φ ( X 2 ) ss ⋅ T ϵ mn ( ( ( P ( X 1 ) s ⊥ − P ( X 2 ) s ⊥ ) ( X 1 ) ssss + P ( X 2 ) s ⊥ ( X 1 − X 2 ) ssss ) ) d s . (72)
Then, again following the estimates above for a single curve, we may bound
∣ R H S ∣ ≤ c ϵ − 1 ∣ log ϵ ∣ ( ∥ φ s ( X 1 − X 2 ) s ∥ H − 1 / 2 ∥ ∂ s ∣ ( X 1 ) s s ∣ 2 ( X 1 ) s ∥ H − 1 / 2 + ∥ φ s ( X 2 ) s ∥ H − 1 / 2 ∥ ∂ s ( ( X 1 − X 2 ) s s ⋅ ( X 1 + X 2 ) s s ) ( X 1 ) s ∥ H − 1 / 2 + ∥ φ s ( X 2 ) s ∥ H − 1 / 2 ∥ ∂ s ∣ ( X 2 ) s s ∣ 2 ( X 1 − X 2 ) s ∥ H − 1 / 2 + ∥ φ ( X 1 − X 2 ) s s ∥ H − 1 / 8 ∥ P ( X 1 ) s ⊥ ( X 1 ) s s s s ∥ H − 7 / 8 + ∥ φ ( X 2 ) s s ∥ H − 1 / 8 ∥ ( P ( X 1 ) s ⊥ − P ( X 2 ) s ⊥ ) ( X 1 ) s s s s ∥ H − 7 / 8 + ∥ φ ( X 2 ) s s ∥ H − 1 / 8 ∥ P ( X 2 ) s ⊥ ( X 1 − X 2 ) s s s s ∥ H − 7 / 8 } ≤ c ϵ − 1 ∣ log ϵ ∣ ∥ φ ∥ H 1 / 2 ( 1 + ∥ ( X 1 ) s ∥ H 1 10 / 3 + ∥ ( X 2 ) s ∥ H 1 10 / 3 ) [ ∥ ( X 1 − X 2 ) s ∥ H 1 × ( ∥ ( X 1 ) s s s ∥ H ˙ 1 / 2 2 / 3 + ∥ ( X 2 ) s s s ∥ H ˙ 1 / 2 2 / 3 + ∥ ( X 1 ) s s s ∥ H ˙ 1 / 2 3 / 4 ) + ∥ ( X 1 − X 2 ) s s s ∥ H ˙ 1 / 2 1 / 3 ∥ ( X 1 − X 2 ) s ∥ H 1 2 / 3 ( ∥ ( X 1 ) s s s ∥ H ˙ 1 / 2 1 / 3 + ∥ ( X 2 ) s s s ∥ H ˙ 1 / 2 1 / 3 ) + ∥ ( X 1 − X 2 ) s s s ∥ H ˙ 1 / 2 3 / 4 ∥ ( X 1 − X 2 ) s ∥ H 1 1 / 4 ] . (73) \begin{aligned}
|\mathrm{RHS}| \leq{}& c\epsilon^{-1}|\log\epsilon|\left(\|\varphi_s(X_1-X_2)_s\|_{H^{-1/2}}\|\partial_s|(X_1)_{ss}|^2(X_1)_s\|_{H^{-1/2}}\right.\\
&+\|\varphi_s(X_2)_s\|_{H^{-1/2}}\|\partial_s((X_1-X_2)_{ss}\cdot(X_1+X_2)_{ss})(X_1)_s\|_{H^{-1/2}}\\
&+\|\varphi_s(X_2)_s\|_{H^{-1/2}}\|\partial_s|(X_2)_{ss}|^2(X_1-X_2)_s\|_{H^{-1/2}}\\
&+\|\varphi(X_1-X_2)_{ss}\|_{H^{-1/8}}\|\mathbb{P}^{\perp}_{(X_1)_s}(X_1)_{ssss}\|_{H^{-7/8}}\\
&+\|\varphi(X_2)_{ss}\|_{H^{-1/8}}\|(\mathbb{P}^{\perp}_{(X_1)_s}-\mathbb{P}^{\perp}_{(X_2)_s})(X_1)_{ssss}\|_{H^{-7/8}}\\
&\left.+\|\varphi(X_2)_{ss}\|_{H^{-1/8}}\|\mathbb{P}^{\perp}_{(X_2)_s}(X_1-X_2)_{ssss}\|_{H^{-7/8}}\right\}\\
\leq{}& c\epsilon^{-1}|\log\epsilon|\|\varphi\|_{H^{1/2}}\left(1+\|(X_1)_s\|_{H^1}^{10/3}+\|(X_2)_s\|_{H^1}^{10/3}\right)\left[\|(X_1-X_2)_s\|_{H^1}\right.\\
&\quad\times\left(\|(X_1)_{sss}\|_{\dot H^{1/2}}^{2/3}+\|(X_2)_{sss}\|_{\dot H^{1/2}}^{2/3}+\|(X_1)_{sss}\|_{\dot H^{1/2}}^{3/4}\right)\\
&\quad+\|(X_1-X_2)_{sss}\|_{\dot H^{1/2}}^{1/3}\|(X_1-X_2)_s\|_{H^1}^{2/3}\left(\|(X_1)_{sss}\|_{\dot H^{1/2}}^{1/3}+\|(X_2)_{sss}\|_{\dot H^{1/2}}^{1/3}\right)\\
&\quad\left.+\|(X_1-X_2)_{sss}\|_{\dot H^{1/2}}^{3/4}\|(X_1-X_2)_s\|_{H^1}^{1/4}\right]. \tag*{(73)}
\end{aligned} ∣ RHS ∣ ≤ ≤ c ϵ − 1 ∣ log ϵ ∣ ( ∥ φ s ( X 1 − X 2 ) s ∥ H − 1/2 ∥ ∂ s ∣ ( X 1 ) ss ∣ 2 ( X 1 ) s ∥ H − 1/2 + ∥ φ s ( X 2 ) s ∥ H − 1/2 ∥ ∂ s (( X 1 − X 2 ) ss ⋅ ( X 1 + X 2 ) ss ) ( X 1 ) s ∥ H − 1/2 + ∥ φ s ( X 2 ) s ∥ H − 1/2 ∥ ∂ s ∣ ( X 2 ) ss ∣ 2 ( X 1 − X 2 ) s ∥ H − 1/2 + ∥ φ ( X 1 − X 2 ) ss ∥ H − 1/8 ∥ P ( X 1 ) s ⊥ ( X 1 ) ssss ∥ H − 7/8 + ∥ φ ( X 2 ) ss ∥ H − 1/8 ∥ ( P ( X 1 ) s ⊥ − P ( X 2 ) s ⊥ ) ( X 1 ) ssss ∥ H − 7/8 + ∥ φ ( X 2 ) ss ∥ H − 1/8 ∥ P ( X 2 ) s ⊥ ( X 1 − X 2 ) ssss ∥ H − 7/8 } c ϵ − 1 ∣ log ϵ ∣∥ φ ∥ H 1/2 ( 1 + ∥ ( X 1 ) s ∥ H 1 10/3 + ∥ ( X 2 ) s ∥ H 1 10/3 ) [ ∥ ( X 1 − X 2 ) s ∥ H 1 × ( ∥ ( X 1 ) sss ∥ H ˙ 1/2 2/3 + ∥ ( X 2 ) sss ∥ H ˙ 1/2 2/3 + ∥ ( X 1 ) sss ∥ H ˙ 1/2 3/4 ) + ∥ ( X 1 − X 2 ) sss ∥ H ˙ 1/2 1/3 ∥ ( X 1 − X 2 ) s ∥ H 1 2/3 ( ∥ ( X 1 ) sss ∥ H ˙ 1/2 1/3 + ∥ ( X 2 ) sss ∥ H ˙ 1/2 1/3 ) + ∥ ( X 1 − X 2 ) sss ∥ H ˙ 1/2 3/4 ∥ ( X 1 − X 2 ) s ∥ H 1 1/4 ] . (73)
Altogether, taking φ = τ 1 − τ 2 \varphi=\tau_1-\tau_2 φ = τ 1 − τ 2 in (69) and the subsequent bounds, we may estimate
B ϵ , 2 ( τ 1 − τ 2 , τ 1 − τ 2 ) ≤ ∣ L H S ∣ + ∣ R H S ∣ ≤ c ϵ − 1 ∣ log ϵ ∣ ∥ τ 1 − τ 2 ∥ H 1 / 2 ( 1 + ∥ ( X 1 ) s ∥ H 1 10 / 3 + ∥ ( X 2 ) s ∥ H 1 10 / 3 ) [ ∥ ( X 1 − X 2 ) s ∥ H 1 × ( ∥ ( X 1 ) s s s ∥ H ˙ 1 / 2 2 / 3 + ∥ ( X 2 ) s s s ∥ H ˙ 1 / 2 2 / 3 + ∥ ( X 1 ) s s s ∥ H ˙ 1 / 2 3 / 4 ) ( 1 + ∥ ( X 1 ) s ∥ H 1 5 / 1 ) + ∥ ( X 1 − X 2 ) s s s ∥ H ˙ 1 / 2 1 / 3 ∥ ( X 1 − X 2 ) s ∥ H 1 2 / 3 ( ∥ ( X 1 ) s s s ∥ H ˙ 1 / 2 1 / 3 + ∥ ( X 2 ) s s s ∥ H ˙ 1 / 2 1 / 3 ) + ∥ ( X 1 − X 2 ) s s s ∥ H ˙ 1 / 2 3 / 4 ∥ ( X 1 − X 2 ) s ∥ H 1 1 / 4 ] . (74) \begin{aligned}
\mathcal{B}_{\epsilon,2}(\tau_1-\tau_2,\tau_1-\tau_2)&\leq|\mathrm{LHS}|+|\mathrm{RHS}|\\
&\leq c\epsilon^{-1}|\log\epsilon|\|\tau_1-\tau_2\|_{H^{1/2}}\left(1+\|(X_1)_s\|_{H^1}^{10/3}+\|(X_2)_s\|_{H^1}^{10/3}\right)\left[\|(X_1-X_2)_s\|_{H^1}\right.\\
&\quad\times\left(\|(X_1)_{sss}\|_{\dot H^{1/2}}^{2/3}+\|(X_2)_{sss}\|_{\dot H^{1/2}}^{2/3}+\|(X_1)_{sss}\|_{\dot H^{1/2}}^{3/4}\right)\left(1+\|(X_1)_s\|_{H^1}^{5/1}\right)\\
&\quad+\|(X_1-X_2)_{sss}\|_{\dot H^{1/2}}^{1/3}\|(X_1-X_2)_s\|_{H^1}^{2/3}\left(\|(X_1)_{sss}\|_{\dot H^{1/2}}^{1/3}+\|(X_2)_{sss}\|_{\dot H^{1/2}}^{1/3}\right)\\
&\quad\left.+\|(X_1-X_2)_{sss}\|_{\dot H^{1/2}}^{3/4}\|(X_1-X_2)_s\|_{H^1}^{1/4}\right]. \tag*{(74)}
\end{aligned} B ϵ , 2 ( τ 1 − τ 2 , τ 1 − τ 2 ) ≤ ∣ LHS ∣ + ∣ RHS ∣ ≤ c ϵ − 1 ∣ log ϵ ∣∥ τ 1 − τ 2 ∥ H 1/2 ( 1 + ∥ ( X 1 ) s ∥ H 1 10/3 + ∥ ( X 2 ) s ∥ H 1 10/3 ) [ ∥ ( X 1 − X 2 ) s ∥ H 1 × ( ∥ ( X 1 ) sss ∥ H ˙ 1/2 2/3 + ∥ ( X 2 ) sss ∥ H ˙ 1/2 2/3 + ∥ ( X 1 ) sss ∥ H ˙ 1/2 3/4 ) ( 1 + ∥ ( X 1 ) s ∥ H 1 5/1 ) + ∥ ( X 1 − X 2 ) sss ∥ H ˙ 1/2 1/3 ∥ ( X 1 − X 2 ) s ∥ H 1 2/3 ( ∥ ( X 1 ) sss ∥ H ˙ 1/2 1/3 + ∥ ( X 2 ) sss ∥ H ˙ 1/2 1/3 ) + ∥ ( X 1 − X 2 ) sss ∥ H ˙ 1/2 3/4 ∥ ( X 1 − X 2 ) s ∥ H 1 1/4 ] . (74)
Applying the lower bound (57) and Young’s inequality yields the bound (44) . □
Main operator decomposition and linear bounds
Since the operator L ‾ ϵ ∂ s 4 \overline{\mathcal{L}}_\epsilon\partial_s^4 L ϵ ∂ s 4 used for the evolution (14) involves a nonlinear dependence on the curve X X X , we begin by extracting the principal linear part for use in the local existence theory. Using the inextensibility constraint for X X X , we show the following.
Lemma 3.3 (Main operator decomposition). Given X ∈ H 7 / 2 ( T ) X\in H^{7/2}(\mathbb{T}) X ∈ H 7/2 ( T ) satisfying ∣ X s ∣ = 1 |X_s|=1 ∣ X s ∣ = 1 and L ‾ ϵ \overline{\mathcal{L}}_\epsilon L ϵ as in Definition 1.1, the operator L ‾ ϵ [ ∂ s 4 ⋅ ] \overline{\mathcal{L}}_\epsilon[\partial_s^4\cdot] L ϵ [ ∂ s 4 ⋅ ] applied to X X X may be decomposed as
L ‾ ϵ [ ∂ s 4 X ] = T m ϵ n ( ∂ s 4 X ) + R [ X ] , (75) \overline{\mathcal{L}}_\epsilon[\partial_s^4X]=T_{m_\epsilon^n}(\partial_s^4X)+\mathcal{R}[X], \tag*{(75)} L ϵ [ ∂ s 4 X ] = T m ϵ n ( ∂ s 4 X ) + R [ X ] , (75) where m ϵ n m_\epsilon^n m ϵ n is the normal direction multiplier (1.12)–(1.13), and the remainder term R [ X ] \mathcal{R}[X] R [ X ] satisfies
∥ R [ X ] ∥ H 1 / 2 ( T ) ≤ c ϵ − 3 / 2 ( ∥ ( X ) s s s ∥ H ˙ 1 / 2 ∥ X s ∥ H 1 2 / 3 + ∥ X s ∥ H 1 4 / 3 + ∥ X s ∥ H 1 2 ) ( 1 + ∥ X s ∥ H 1 2 ) . (76) \|\mathcal{R}[X]\|_{H^{1/2}(\mathbb{T})}\leq c\epsilon^{-3/2}\left(\|(X)_{sss}\|_{\dot H^{1/2}}\|X_s\|_{H^1}^{2/3}+\|X_s\|_{H^1}^{4/3}+\|X_s\|_{H^1}^{2}\right)(1+\|X_s\|_{H^1}^{2}). \tag*{(76)} ∥ R [ X ] ∥ H 1/2 ( T ) ≤ c ϵ − 3/2 ( ∥ ( X ) sss ∥ H ˙ 1/2 ∥ X s ∥ H 1 2/3 + ∥ X s ∥ H 1 4/3 + ∥ X s ∥ H 1 2 ) ( 1 + ∥ X s ∥ H 1 2 ) . (76)
In addition, given two nearby curves X 1 , X 2 ∈ H 7 / 2 ( T ) X_1,X_2 \in H^{7/2}(\mathbb{T}) X 1 , X 2 ∈ H 7/2 ( T ) both satisfying ∣ ( X j ) s ∣ = 1 \lvert(X_j)_s\rvert= 1 ∣( X j ) s ∣ = 1 , we may bound the difference between the corresponding remainder terms as
∥ R [ X 1 ] − R [ X 2 ] ∥ H 1 / 2 ( T ) ≤ c ε − 3 / 2 ( 1 + ∥ ( X 1 ) s ∥ H 1 10 / 3 + ∥ ( X 2 ) s ∥ H 1 10 / 3 ) × [ ∥ ( X 1 − X 2 ) s ∥ H 1 ( ∥ ( X 1 ) s s s ∥ H ˙ 1 / 2 2 / 3 + ∥ ( X 2 ) s s s ∥ H ˙ 1 / 2 2 / 3 + ∥ ( X 1 ) s ∥ H 1 2 / 3 + 1 ) + ∥ ( X 1 − X 2 ) s ∥ H 1 2 / 3 ∥ ( X 1 − X 2 ) s s s ∥ H ˙ 1 / 2 1 / 3 ( ∥ ( X 1 ) s s s ∥ H ˙ 1 / 2 1 / 3 + ∥ ( X 2 ) s s s ∥ H ˙ 1 / 2 1 / 3 ) + ∥ ( X 1 − X 2 ) s s s ∥ H ˙ 1 / 2 2 / 3 ∥ ( X 1 − X 2 ) s ∥ H 1 1 / 3 ] . (77) \begin{aligned}
\|\mathcal{R}[X_1]-\mathcal{R}[X_2]\|_{H^{1/2}(\mathbb{T})}
&\le c\varepsilon^{-3/2}\left(1+\|(X_1)_s\|_{H^1}^{10/3}+\|(X_2)_s\|_{H^1}^{10/3}\right)\\
&\quad\times\left[\|(X_1-X_2)_s\|_{H^1}\left(\|(X_1)_{sss}\|_{\dot H^{1/2}}^{2/3}+\|(X_2)_{sss}\|_{\dot H^{1/2}}^{2/3}+\|(X_1)_s\|_{H^1}^{2/3}+1\right)\right.\\
&\qquad\left.+\|(X_1-X_2)_s\|_{H^1}^{2/3}\|(X_1-X_2)_{sss}\|_{\dot H^{1/2}}^{1/3}\left(\|(X_1)_{sss}\|_{\dot H^{1/2}}^{1/3}+\|(X_2)_{sss}\|_{\dot H^{1/2}}^{1/3}\right)\right.\\
&\qquad\left.+\|(X_1-X_2)_{sss}\|_{\dot H^{1/2}}^{2/3}\|(X_1-X_2)_s\|_{H^1}^{1/3}\right]. \tag*{(77)}
\end{aligned} ∥ R [ X 1 ] − R [ X 2 ] ∥ H 1/2 ( T ) ≤ c ε − 3/2 ( 1 + ∥ ( X 1 ) s ∥ H 1 10/3 + ∥ ( X 2 ) s ∥ H 1 10/3 ) × [ ∥ ( X 1 − X 2 ) s ∥ H 1 ( ∥ ( X 1 ) sss ∥ H ˙ 1/2 2/3 + ∥ ( X 2 ) sss ∥ H ˙ 1/2 2/3 + ∥ ( X 1 ) s ∥ H 1 2/3 + 1 ) + ∥ ( X 1 − X 2 ) s ∥ H 1 2/3 ∥ ( X 1 − X 2 ) sss ∥ H ˙ 1/2 1/3 ( ∥ ( X 1 ) sss ∥ H ˙ 1/2 1/3 + ∥ ( X 2 ) sss ∥ H ˙ 1/2 1/3 ) + ∥ ( X 1 − X 2 ) sss ∥ H ˙ 1/2 2/3 ∥ ( X 1 − X 2 ) s ∥ H 1 1/3 ] . (77)
Proof. We begin by writing
L ‾ ε [ ∂ s 4 X ] = T m ε ( ∂ s 4 X ) + R [ X ] , \overline{\mathcal{L}}_\varepsilon[\partial_s^4X]=T_{m_\varepsilon}(\partial_s^4X)+\mathcal{R}[X], L ε [ ∂ s 4 X ] = T m ε ( ∂ s 4 X ) + R [ X ] ,
R [ X ] = − ( X s ⋅ T m ε ( ∂ s 4 X ) ) X s + P X s T m ε ( P X s ∂ s 4 X ) − P X s ⊥ T m ε ( P X s ∂ s 4 X ) . (78) \mathcal{R}[X]=-(X_s\cdot T_{m_\varepsilon}(\partial_s^4X))X_s+\mathbb{P}_{X_s}T_{m_\varepsilon}(\mathbb{P}_{X_s}\partial_s^4X)-\mathbb{P}_{X_s}^{\perp}T_{m_\varepsilon}(\mathbb{P}_{X_s}\partial_s^4X). \tag*{(78)} R [ X ] = − ( X s ⋅ T m ε ( ∂ s 4 X )) X s + P X s T m ε ( P X s ∂ s 4 X ) − P X s ⊥ T m ε ( P X s ∂ s 4 X ) . (78)
The terms R b \mathcal{R}_b R b and R c \mathcal{R}_c R c are the simplest to estimate. In particular, using the identities (36) coming from the inextensibility constraint, we may write
P X s ∂ s 4 X = − 3 ( X s s ⋅ X s s s ) X s (79) \mathbb{P}_{X_s}\partial_s^4X=-3(X_{ss}\cdot X_{sss})X_s \tag*{(79)} P X s ∂ s 4 X = − 3 ( X ss ⋅ X sss ) X s (79)
so that, using Parseval’s theorem, we may bound
∥ T m ε ( P X s ∂ s 4 X ) ∥ H 1 / 2 2 = ∑ ∣ k ∣ = 0 ∞ m ε 2 ( k ) ( 1 + ∣ k ∣ 2 ) 1 / 2 ∣ F [ P X s ∂ s 4 X ] ∣ 2 ≤ c ∣ log ε ∣ 2 ∑ ∣ k ∣ = 0 ⌊ 1 2 π ε ⌋ ( 1 + ∣ k ∣ 2 ) 1 / 2 ∣ F [ P X s ∂ s 4 X ] ∣ 2 + c ε − 2 ∑ ∣ k ∣ = ⌊ 1 2 π ε ⌋ + 1 ∞ ∣ k ∣ − 2 ( 1 + ∣ k ∣ 2 ) 1 / 2 ∣ F [ P X s ∂ s 4 X ] ∣ 2 ≤ c ε − 2 ∣ log ε ∣ 2 ∥ ( X s s ⋅ X s s s ) X s ∥ H − 1 / 2 2 ≤ c ε − 2 ∣ log ε ∣ 2 ∥ X s ∥ H 1 2 ∥ ∂ s ∣ X s s ∣ 2 ∥ H − 1 / 2 2 . (80) \begin{aligned}
\|T_{m_\varepsilon}(\mathbb{P}_{X_s}\partial_s^4X)\|_{H^{1/2}}^2
&=\sum_{|k|=0}^{\infty}m_\varepsilon^2(k)(1+|k|^2)^{1/2}\left|\mathcal{F}[\mathbb{P}_{X_s}\partial_s^4X]\right|^2\\
&\le c|\log\varepsilon|^2\sum_{|k|=0}^{\left\lfloor\frac{1}{2\pi\varepsilon}\right\rfloor}(1+|k|^2)^{1/2}\left|\mathcal{F}[\mathbb{P}_{X_s}\partial_s^4X]\right|^2\\
&\quad+c\varepsilon^{-2}\sum_{|k|=\left\lfloor\frac{1}{2\pi\varepsilon}\right\rfloor+1}^{\infty}|k|^{-2}(1+|k|^2)^{1/2}\left|\mathcal{F}[\mathbb{P}_{X_s}\partial_s^4X]\right|^2\\
&\le c\varepsilon^{-2}|\log\varepsilon|^2\|(X_{ss}\cdot X_{sss})X_s\|_{H^{-1/2}}^2\\
&\le c\varepsilon^{-2}|\log\varepsilon|^2\|X_s\|_{H^1}^2\|\partial_s|X_{ss}|^2\|_{H^{-1/2}}^2. \tag*{(80)}
\end{aligned} ∥ T m ε ( P X s ∂ s 4 X ) ∥ H 1/2 2 = ∣ k ∣ = 0 ∑ ∞ m ε 2 ( k ) ( 1 + ∣ k ∣ 2 ) 1/2 F [ P X s ∂ s 4 X ] 2 ≤ c ∣ log ε ∣ 2 ∣ k ∣ = 0 ∑ ⌊ 2 π ε 1 ⌋ ( 1 + ∣ k ∣ 2 ) 1/2 F [ P X s ∂ s 4 X ] 2 + c ε − 2 ∣ k ∣ = ⌊ 2 π ε 1 ⌋ + 1 ∑ ∞ ∣ k ∣ − 2 ( 1 + ∣ k ∣ 2 ) 1/2 F [ P X s ∂ s 4 X ] 2 ≤ c ε − 2 ∣ log ε ∣ 2 ∥ ( X ss ⋅ X sss ) X s ∥ H − 1/2 2 ≤ c ε − 2 ∣ log ε ∣ 2 ∥ X s ∥ H 1 2 ∥ ∂ s ∣ X ss ∣ 2 ∥ H − 1/2 2 . (80)
Here we have used the multiplier bounds of Lemma 2.1 in the first inequality, and the second bound of Proposition 3.1 in the final inequality. The interpolation inequalities (37) then yield
∥ T m ε ( P X s ∂ s 4 X ) ∥ H 1 / 2 ≤ c ε − 1 ∣ log ε ∣ ∥ X s ∥ H 1 ∥ ∣ X s s ∣ ∥ H ˙ 1 / 2 ≤ c ε − 1 ∣ log ε ∣ ∥ X s ∥ H 1 ∥ X s s s ∥ L 2 1 / 2 ∥ X s s ∥ L ∞ ≤ c ε − 1 ∣ log ε ∣ ∥ X s ∥ H 1 ∥ X s s s ∥ H ˙ 1 / 2 2 / 3 . (81) \begin{aligned}
\|T_{m_\varepsilon}(\mathbb{P}_{X_s}\partial_s^4X)\|_{H^{1/2}}
&\le c\varepsilon^{-1}|\log\varepsilon|\|X_s\|_{H^1}\||X_{ss}|\|_{\dot H^{1/2}}\\
&\le c\varepsilon^{-1}|\log\varepsilon|\|X_s\|_{H^1}\|X_{sss}\|_{L^2}^{1/2}\|X_{ss}\|_{L^\infty}\\
&\le c\varepsilon^{-1}|\log\varepsilon|\|X_s\|_{H^1}\|X_{sss}\|_{\dot H^{1/2}}^{2/3}. \tag*{(81)}
\end{aligned} ∥ T m ε ( P X s ∂ s 4 X ) ∥ H 1/2 ≤ c ε − 1 ∣ log ε ∣∥ X s ∥ H 1 ∥∣ X ss ∣ ∥ H ˙ 1/2 ≤ c ε − 1 ∣ log ε ∣∥ X s ∥ H 1 ∥ X sss ∥ L 2 1/2 ∥ X ss ∥ L ∞ ≤ c ε − 1 ∣ log ε ∣∥ X s ∥ H 1 ∥ X sss ∥ H ˙ 1/2 2/3 . (81)
We may thus estimate R b \mathcal{R}_b R b as
∥ R b ∥ H 1 / 2 = ∥ P X s T m ε ( P X s ∂ s 4 X ) ∥ H 1 / 2 ≤ c ∥ X s ∥ H 1 ∥ T m ε ( P X s ∂ s 4 X ) ∥ H 1 / 2 ≤ c ε − 1 ∣ log ε ∣ ∥ X s ∥ H 1 10 / 3 ∥ X s s s ∥ H ˙ 1 / 2 2 / 3 . (82) \begin{aligned}
\|\mathcal{R}_b\|_{H^{1/2}}
&=\|\mathbb{P}_{X_s}T_{m_\varepsilon}(\mathbb{P}_{X_s}\partial_s^4X)\|_{H^{1/2}}\\
&\le c\|X_s\|_{H^1}\|T_{m_\varepsilon}(\mathbb{P}_{X_s}\partial_s^4X)\|_{H^{1/2}}\\
&\le c\varepsilon^{-1}|\log\varepsilon|\|X_s\|_{H^1}^{10/3}\|X_{sss}\|_{\dot H^{1/2}}^{2/3}. \tag*{(82)}
\end{aligned} ∥ R b ∥ H 1/2 = ∥ P X s T m ε ( P X s ∂ s 4 X ) ∥ H 1/2 ≤ c ∥ X s ∥ H 1 ∥ T m ε ( P X s ∂ s 4 X ) ∥ H 1/2 ≤ c ε − 1 ∣ log ε ∣∥ X s ∥ H 1 10/3 ∥ X sss ∥ H ˙ 1/2 2/3 . (82)
A similar calculation for the normal direction term R c \mathcal{R}_c R c yields
∥ R c ∥ H 1 / 2 = ∥ P X s T ϵ m n ( P X s ∂ s 4 X ) ∥ H 1 / 2 ≤ c ϵ − 1 ∣ log ϵ ∣ ∥ X s ∥ H 1 10 / 3 ∥ X s s s ∥ H ˙ 1 / 2 2 / 3 . (83) \left\|\mathcal{R}_c\right\|_{H^{1/2}}=\left\|\mathbb{P}_{X_s}T_{\epsilon}^{mn}\left(\mathbb{P}_{X_s}\partial_s^4\mathbf{X}\right)\right\|_{H^{1/2}}\le c\epsilon^{-1}\lvert\log\epsilon\rvert\left\|\mathbf{X}_s\right\|_{H^1}^{10/3}\left\|\mathbf{X}_{sss}\right\|_{\dot{H}^{1/2}}^{2/3}. \tag*{(83)} ∥ R c ∥ H 1/2 = P X s T ϵ mn ( P X s ∂ s 4 X ) H 1/2 ≤ c ϵ − 1 ∣ log ϵ ∣ ∥ X s ∥ H 1 10/3 ∥ X sss ∥ H ˙ 1/2 2/3 . (83)
To estimate R a \mathcal{R}_a R a , we first use that ∂ s \partial_s ∂ s commutes with T ϵ m n T_{\epsilon}^{mn} T ϵ mn to write
X s ⋅ T ϵ m n ( ∂ s 4 X ) = X s ⋅ ∂ s 3 ( T ϵ m n X s ) . (84) \mathbf{X}_s\cdot T_{\epsilon}^{mn}\left(\partial_s^4\mathbf{X}\right)=\mathbf{X}_s\cdot\partial_s^3\left(T_{\epsilon}^{mn}\mathbf{X}_s\right). \tag*{(84)} X s ⋅ T ϵ mn ( ∂ s 4 X ) = X s ⋅ ∂ s 3 ( T ϵ mn X s ) . (84)
We then consider the difference
R a , d i f f = ∂ s 3 ( T ϵ m n X s ) − c n ϵ − 1 ∂ s 2 X s , (85) \mathcal{R}_{a,\mathrm{diff}}=\partial_s^3\left(T_{\epsilon}^{mn}\mathbf{X}_s\right)-c_n\epsilon^{-1}\partial_s^2\mathbf{X}_s, \tag*{(85)} R a , diff = ∂ s 3 ( T ϵ mn X s ) − c n ϵ − 1 ∂ s 2 X s , (85)
where c n c_n c n is the constant from Lemma 2.2. Then, using the bounds of Lemma 2.2, we may estimate
∥ R a , d i f f ∥ H 1 / 2 2 = ∑ ∣ k ∣ = 0 ∞ ( 1 + ∣ k ∣ 2 ) 1 / 2 ( ∣ k ∣ m ϵ n ( k ) − c n ϵ − 1 ) 2 ∣ k ∣ 4 ∣ F [ X s ] ∣ 2 ≤ ∑ ∣ k ∣ = 0 ∞ ( 1 + ∣ k ∣ 2 ) 1 / 2 ( c n ϵ − 1 ) 2 ( ϵ ∣ k ∣ + 1 ) − 2 ∣ k ∣ 4 ∣ F [ X s ] ∣ 2 ≤ c ϵ − 3 ∑ ∣ k ∣ = 0 ∞ ( 1 + ∣ k ∣ 2 ) 1 / 2 ∣ k ∣ 3 ∣ F [ X s ] ∣ 2 ≤ c ϵ − 3 ∥ X s s ∥ H 1 2 . (86) \begin{aligned}
\left\|\mathcal{R}_{a,\mathrm{diff}}\right\|_{H^{1/2}}^2
&=\sum_{|k|=0}^{\infty}(1+|k|^2)^{1/2}\left(|k|m_{\epsilon}^{n}(k)-c_n\epsilon^{-1}\right)^2|k|^4\left|\mathcal{F}[\mathbf{X}_s]\right|^2\\
&\le\sum_{|k|=0}^{\infty}(1+|k|^2)^{1/2}(c_n\epsilon^{-1})^2(\epsilon|k|+1)^{-2}|k|^4\left|\mathcal{F}[\mathbf{X}_s]\right|^2\\
&\le c\epsilon^{-3}\sum_{|k|=0}^{\infty}(1+|k|^2)^{1/2}|k|^3\left|\mathcal{F}[\mathbf{X}_s]\right|^2\\
&\le c\epsilon^{-3}\left\|\mathbf{X}_{ss}\right\|_{H^1}^2. \tag*{(86)}
\end{aligned} ∥ R a , diff ∥ H 1/2 2 = ∣ k ∣ = 0 ∑ ∞ ( 1 + ∣ k ∣ 2 ) 1/2 ( ∣ k ∣ m ϵ n ( k ) − c n ϵ − 1 ) 2 ∣ k ∣ 4 ∣ F [ X s ] ∣ 2 ≤ ∣ k ∣ = 0 ∑ ∞ ( 1 + ∣ k ∣ 2 ) 1/2 ( c n ϵ − 1 ) 2 ( ϵ ∣ k ∣ + 1 ) − 2 ∣ k ∣ 4 ∣ F [ X s ] ∣ 2 ≤ c ϵ − 3 ∣ k ∣ = 0 ∑ ∞ ( 1 + ∣ k ∣ 2 ) 1/2 ∣ k ∣ 3 ∣ F [ X s ] ∣ 2 ≤ c ϵ − 3 ∥ X ss ∥ H 1 2 . (86)
Using (84) , we may write
X s ⋅ T ϵ m n ( ∂ s 4 X ) = c n ϵ − 1 X s ⋅ X s s s + X s ⋅ R a , d i f f = − c n ϵ − 1 ∣ X s s ∣ 2 + X s ⋅ R a , d i f f , (87) \mathbf{X}_s\cdot T_{\epsilon}^{mn}\left(\partial_s^4\mathbf{X}\right)=c_n\epsilon^{-1}\mathbf{X}_s\cdot\mathbf{X}_{sss}+\mathbf{X}_s\cdot\mathcal{R}_{a,\mathrm{diff}}=-c_n\epsilon^{-1}|\mathbf{X}_{ss}|^2+\mathbf{X}_s\cdot\mathcal{R}_{a,\mathrm{diff}}, \tag*{(87)} X s ⋅ T ϵ mn ( ∂ s 4 X ) = c n ϵ − 1 X s ⋅ X sss + X s ⋅ R a , diff = − c n ϵ − 1 ∣ X ss ∣ 2 + X s ⋅ R a , diff , (87)
so that, by (86) and the interpolation inequalities (37) , we may estimate the final remainder term as
∥ R a ∥ H 1 / 2 ≤ c n ϵ − 1 ∥ X s ∥ H 1 ∥ ∣ X s s ∣ 2 ∥ H 1 / 2 + ∥ X s ∥ H 1 ∥ R a , d i f f ∥ H 1 / 2 ≤ c ϵ − 1 ( ∥ X s s s ∥ H ˙ 1 / 2 2 / 3 ∥ X s ∥ H 1 7 / 3 + ∥ X s s s ∥ H ˙ 1 / 2 1 / 3 ∥ X s ∥ H 1 8 / 3 ) + c ϵ − 3 / 2 ( ∥ X s s s ∥ H ˙ 1 / 2 2 / 3 ∥ X s ∥ H 1 4 / 3 + ∥ X s ∥ H 1 2 ) ≤ c ϵ − 3 / 2 ( ∥ X s s s ∥ H ˙ 1 / 2 2 / 3 ∥ X s ∥ H 1 4 / 3 + ∥ X s ∥ H 1 2 ) ( 1 + ∥ X s ∥ H 1 ) . (88) \begin{aligned}
\left\|\mathcal{R}_a\right\|_{H^{1/2}}
&\le c_n\epsilon^{-1}\left\|\mathbf{X}_s\right\|_{H^1}\left\||\mathbf{X}_{ss}|^2\right\|_{H^{1/2}}+\left\|\mathbf{X}_s\right\|_{H^1}\left\|\mathcal{R}_{a,\mathrm{diff}}\right\|_{H^{1/2}}\\
&\le c\epsilon^{-1}\left(\left\|\mathbf{X}_{sss}\right\|_{\dot{H}^{1/2}}^{2/3}\left\|\mathbf{X}_s\right\|_{H^1}^{7/3}+\left\|\mathbf{X}_{sss}\right\|_{\dot{H}^{1/2}}^{1/3}\left\|\mathbf{X}_s\right\|_{H^1}^{8/3}\right)\\
&\qquad+c\epsilon^{-3/2}\left(\left\|\mathbf{X}_{sss}\right\|_{\dot{H}^{1/2}}^{2/3}\left\|\mathbf{X}_s\right\|_{H^1}^{4/3}+\left\|\mathbf{X}_s\right\|_{H^1}^2\right)\\
&\le c\epsilon^{-3/2}\left(\left\|\mathbf{X}_{sss}\right\|_{\dot{H}^{1/2}}^{2/3}\left\|\mathbf{X}_s\right\|_{H^1}^{4/3}+\left\|\mathbf{X}_s\right\|_{H^1}^2\right)\left(1+\left\|\mathbf{X}_s\right\|_{H^1}\right). \tag*{(88)}
\end{aligned} ∥ R a ∥ H 1/2 ≤ c n ϵ − 1 ∥ X s ∥ H 1 ∣ X ss ∣ 2 H 1/2 + ∥ X s ∥ H 1 ∥ R a , diff ∥ H 1/2 ≤ c ϵ − 1 ( ∥ X sss ∥ H ˙ 1/2 2/3 ∥ X s ∥ H 1 7/3 + ∥ X sss ∥ H ˙ 1/2 1/3 ∥ X s ∥ H 1 8/3 ) + c ϵ − 3/2 ( ∥ X sss ∥ H ˙ 1/2 2/3 ∥ X s ∥ H 1 4/3 + ∥ X s ∥ H 1 2 ) ≤ c ϵ − 3/2 ( ∥ X sss ∥ H ˙ 1/2 2/3 ∥ X s ∥ H 1 4/3 + ∥ X s ∥ H 1 2 ) ( 1 + ∥ X s ∥ H 1 ) . (88)
Combining the bounds for R a \mathcal{R}_a R a , R b \mathcal{R}_b R b , and R c \mathcal{R}_c R c , we obtain the estimate (76) .
To show the Lipschitz bound (77) , we again begin with R b \mathcal{R}_b R b and R c \mathcal{R}_c R c . Using (79) , we have
∥ R b [ X 1 ] − R b [ X 2 ] ∥ H 1 / 2 ≤ c [ ∥ ( X 1 − X 2 ) s ∥ H 1 ( ∥ ( X 1 ) s ∥ H 1 + ∥ ( X 2 ) s ∥ H 1 ) ∥ T ϵ m ( ∂ s ∣ ( X 1 ) s s ∣ 2 ( X 1 ) s ) ∥ H 1 / 2 + ∥ ( X 2 ) s s ∥ H 1 ( ∥ T ϵ m ( ∂ s ∣ ( X 2 ) s s ∣ 2 ( X 1 − X 2 ) s ) ∥ H 1 / 2 + ∥ T ϵ m ( ∂ s ( ∣ ( X 1 ) s s ∣ 2 − ∣ ( X 2 ) s s ∣ 2 ) ( X 1 ) s ) ∥ H 1 / 2 ) ] ≤ c ϵ − 1 ∣ log ϵ ∣ [ ∥ ( X 1 − X 2 ) s ∥ H 1 ( 1 + ∥ ( X 1 ) s ∥ H 1 2 + ∥ ( X 2 ) s ∥ H 1 2 ) × ( ∥ ∣ ( X 1 ) s s ∣ 2 ∥ H ˙ 1 / 2 + ∥ ∣ ( X 2 ) s s ∣ 2 ∥ H ˙ 1 / 2 ) + ∥ ( X 1 ) s ∥ H 1 ∥ ( X 1 ) s s − ( X 2 ) s s ⋅ ( ( X 1 ) s s + ( X 2 ) s s ) ∥ H 1 / 2 ] ≤ c ϵ − 1 ∣ log ϵ ∣ ( 1 + ∥ ( X 1 ) s ∥ H 1 10 / 3 + ∥ ( X 2 ) s ∥ H 1 10 / 3 ) × [ ∥ ( X 1 − X 2 ) s ∥ H 1 ( ∥ ( X 1 ) s s s ∥ H ˙ 1 / 2 2 / 3 + ∥ ( X 2 ) s s s ∥ H ˙ 1 / 2 2 / 3 ) + ∥ ( X 1 − X 2 ) s ∥ H 1 2 / 3 ∥ ( X 1 − X 2 ) s s s ∥ H ˙ 1 / 2 1 / 3 ( ∥ ( X 1 ) s s s ∥ H ˙ 1 / 2 1 / 3 + ∥ ( X 2 ) s s s ∥ H ˙ 1 / 2 1 / 3 ) ] . (89) \begin{aligned}
\|\mathcal{R}_b[X_1]-\mathcal{R}_b[X_2]\|_{H^{1/2}}
&\le c\Big[ \|(X_1-X_2)_s\|_{H^1}\big(\|(X_1)_s\|_{H^1}+\|(X_2)_s\|_{H^1}\big)
\big\|\mathcal{T}_{\epsilon}^{m}\big(\partial_s|(X_1)_{ss}|^2(X_1)_s\big)\big\|_{H^{1/2}} \\
&\quad+\|(X_2)_{ss}\|_{H^1}\Big(\big\|\mathcal{T}_{\epsilon}^{m}\big(\partial_s|(X_2)_{ss}|^2(X_1-X_2)_s\big)\big\|_{H^{1/2}} \\
&\qquad+\big\|\mathcal{T}_{\epsilon}^{m}\big(\partial_s\big(|(X_1)_{ss}|^2-|(X_2)_{ss}|^2\big)(X_1)_s\big)\big\|_{H^{1/2}}\Big)\Big] \\
&\le c\epsilon^{-1}|\log\epsilon|\Big[\|(X_1-X_2)_s\|_{H^1}\big(1+\|(X_1)_s\|_{H^1}^2+\|(X_2)_s\|_{H^1}^2\big) \\
&\quad\times\Big(\big\|| (X_1)_{ss}|^2\big\|_{\dot H^{1/2}}+\big\|| (X_2)_{ss}|^2\big\|_{\dot H^{1/2}}\Big) \\
&\qquad+\|(X_1)_s\|_{H^1}\big\|(X_1)_{ss}-(X_2)_{ss}\cdot\big((X_1)_{ss}+(X_2)_{ss}\big)\big\|_{H^{1/2}}\Big] \\
&\le c\epsilon^{-1}|\log\epsilon|\big(1+\|(X_1)_s\|_{H^1}^{10/3}+\|(X_2)_s\|_{H^1}^{10/3}\big) \\
&\quad\times\Big[\|(X_1-X_2)_s\|_{H^1}\Big(\|(X_1)_{sss}\|_{\dot H^{1/2}}^{2/3}+\|(X_2)_{sss}\|_{\dot H^{1/2}}^{2/3}\Big) \\
&\qquad+\|(X_1-X_2)_s\|_{H^1}^{2/3}\|(X_1-X_2)_{sss}\|_{\dot H^{1/2}}^{1/3}
\Big(\|(X_1)_{sss}\|_{\dot H^{1/2}}^{1/3}+\|(X_2)_{sss}\|_{\dot H^{1/2}}^{1/3}\Big)\Big]. \tag*{(89)}
\end{aligned} ∥ R b [ X 1 ] − R b [ X 2 ] ∥ H 1/2 ≤ c [ ∥ ( X 1 − X 2 ) s ∥ H 1 ( ∥ ( X 1 ) s ∥ H 1 + ∥ ( X 2 ) s ∥ H 1 ) T ϵ m ( ∂ s ∣ ( X 1 ) ss ∣ 2 ( X 1 ) s ) H 1/2 + ∥ ( X 2 ) ss ∥ H 1 ( T ϵ m ( ∂ s ∣ ( X 2 ) ss ∣ 2 ( X 1 − X 2 ) s ) H 1/2 + T ϵ m ( ∂ s ( ∣ ( X 1 ) ss ∣ 2 − ∣ ( X 2 ) ss ∣ 2 ) ( X 1 ) s ) H 1/2 ) ] ≤ c ϵ − 1 ∣ log ϵ ∣ [ ∥ ( X 1 − X 2 ) s ∥ H 1 ( 1 + ∥ ( X 1 ) s ∥ H 1 2 + ∥ ( X 2 ) s ∥ H 1 2 ) × ( ∣ ( X 1 ) ss ∣ 2 H ˙ 1/2 + ∣ ( X 2 ) ss ∣ 2 H ˙ 1/2 ) + ∥ ( X 1 ) s ∥ H 1 ( X 1 ) ss − ( X 2 ) ss ⋅ ( ( X 1 ) ss + ( X 2 ) ss ) H 1/2 ] ≤ c ϵ − 1 ∣ log ϵ ∣ ( 1 + ∥ ( X 1 ) s ∥ H 1 10/3 + ∥ ( X 2 ) s ∥ H 1 10/3 ) × [ ∥ ( X 1 − X 2 ) s ∥ H 1 ( ∥ ( X 1 ) sss ∥ H ˙ 1/2 2/3 + ∥ ( X 2 ) sss ∥ H ˙ 1/2 2/3 ) + ∥ ( X 1 − X 2 ) s ∥ H 1 2/3 ∥ ( X 1 − X 2 ) sss ∥ H ˙ 1/2 1/3 ( ∥ ( X 1 ) sss ∥ H ˙ 1/2 1/3 + ∥ ( X 2 ) sss ∥ H ˙ 1/2 1/3 ) ] . (89)
An identical bound holds for ∥ R c [ X 1 ] − R c [ X 2 ] ∥ H 1 / 2 \|\mathcal{R}_c[X_1]-\mathcal{R}_c[X_2]\|_{H^{1/2}} ∥ R c [ X 1 ] − R c [ X 2 ] ∥ H 1/2 by an analogous calculation.
For R a \mathcal{R}_a R a , as above, we first note the linear estimate
∥ R a , d i f f [ X 1 − X 2 ] ∥ H 1 / 2 ≤ c ϵ − 3 / 2 ∥ ( X 1 − X 2 ) s s ∥ H 1 . (90) \|\mathcal{R}_{a,\mathrm{diff}}[X_1-X_2]\|_{H^{1/2}}\le c\epsilon^{-3/2}\|(X_1-X_2)_{ss}\|_{H^1}. \tag*{(90)} ∥ R a , diff [ X 1 − X 2 ] ∥ H 1/2 ≤ c ϵ − 3/2 ∥ ( X 1 − X 2 ) ss ∥ H 1 . (90)
We then have
∥ R a [ X 1 ] − R a [ X 2 ] ∥ H 1 / 2 ≤ c [ ∥ ( X 1 − X 2 ) s ∥ H 1 ( ϵ − 1 ∥ ∣ ( X 1 ) s s ∣ 2 ∥ H 1 / 2 + ( ∥ ( X 1 ) s ∥ H 1 + ∥ ( X 2 ) s ∥ H 1 ) ∥ R a , d i f f [ X 1 ] ∥ H 1 / 2 ) + ( 1 + ∥ ( X 2 ) s ∥ H 1 2 ) ( ϵ − 1 ∥ ( X 1 − X 2 ) s s ⋅ ( X 1 + X 2 ) s s ∥ H 1 / 2 + ∥ R a , d i f f [ X 1 − X 2 ] ∥ H 1 / 2 ) ] ≤ c ϵ − 3 / 2 ( 1 + ∥ ( X 1 ) s ∥ H 1 3 + ∥ ( X 2 ) s ∥ H 1 3 ) [ ∥ ( X 1 − X 2 ) s s s ∥ H ˙ 1 / 2 2 / 3 ∥ ( X 1 − X 2 ) s ∥ H 1 1 / 3 + ∥ ( X 1 − X 2 ) s ∥ H 1 ( ∥ ( X 1 ) s s s ∥ H ˙ 1 / 2 2 / 3 + ∥ ( X 2 ) s s s ∥ H ˙ 1 / 2 2 / 3 + ∥ ( X 1 ) s ∥ H 1 2 / 3 + 1 ) ] . (91) \begin{aligned}
\|\mathcal{R}_a[X_1]-\mathcal{R}_a[X_2]\|_{H^{1/2}}
&\le c\Big[\|(X_1-X_2)_s\|_{H^1}\Big(\epsilon^{-1}\big\|| (X_1)_{ss}|^2\big\|_{H^{1/2}} \\
&\quad+\big(\|(X_1)_s\|_{H^1}+\|(X_2)_s\|_{H^1}\big)\|\mathcal{R}_{a,\mathrm{diff}}[X_1]\|_{H^{1/2}}\Big) \\
&\quad+\big(1+\|(X_2)_s\|_{H^1}^2\big)\Big(\epsilon^{-1}\|(X_1-X_2)_{ss}\cdot(X_1+X_2)_{ss}\|_{H^{1/2}}+\|\mathcal{R}_{a,\mathrm{diff}}[X_1-X_2]\|_{H^{1/2}}\Big)\Big] \\
&\le c\epsilon^{-3/2}\big(1+\|(X_1)_s\|_{H^1}^3+\|(X_2)_s\|_{H^1}^3\big)\Big[\|(X_1-X_2)_{sss}\|_{\dot H^{1/2}}^{2/3}\|(X_1-X_2)_s\|_{H^1}^{1/3} \\
&\quad+\|(X_1-X_2)_s\|_{H^1}\Big(\|(X_1)_{sss}\|_{\dot H^{1/2}}^{2/3}+\|(X_2)_{sss}\|_{\dot H^{1/2}}^{2/3}+\|(X_1)_s\|_{H^1}^{2/3}+1\Big)\Big]. \tag*{(91)}
\end{aligned} ∥ R a [ X 1 ] − R a [ X 2 ] ∥ H 1/2 ≤ c [ ∥ ( X 1 − X 2 ) s ∥ H 1 ( ϵ − 1 ∣ ( X 1 ) ss ∣ 2 H 1/2 + ( ∥ ( X 1 ) s ∥ H 1 + ∥ ( X 2 ) s ∥ H 1 ) ∥ R a , diff [ X 1 ] ∥ H 1/2 ) + ( 1 + ∥ ( X 2 ) s ∥ H 1 2 ) ( ϵ − 1 ∥ ( X 1 − X 2 ) ss ⋅ ( X 1 + X 2 ) ss ∥ H 1/2 + ∥ R a , diff [ X 1 − X 2 ] ∥ H 1/2 ) ] ≤ c ϵ − 3/2 ( 1 + ∥ ( X 1 ) s ∥ H 1 3 + ∥ ( X 2 ) s ∥ H 1 3 ) [ ∥ ( X 1 − X 2 ) sss ∥ H ˙ 1/2 2/3 ∥ ( X 1 − X 2 ) s ∥ H 1 1/3 + ∥ ( X 1 − X 2 ) s ∥ H 1 ( ∥ ( X 1 ) sss ∥ H ˙ 1/2 2/3 + ∥ ( X 2 ) sss ∥ H ˙ 1/2 2/3 + ∥ ( X 1 ) s ∥ H 1 2/3 + 1 ) ] . (91)
Again combining the bounds for R a \mathcal{R}_a R a , R b \mathcal{R}_b R b , and R c \mathcal{R}_c R c , we obtain the Lipschitz estimate (77) . □
Given the decomposition of Lemma 3.3 , we consider the curve evolution (14) as
∂ X ∂ t = − T ϵ m ∂ s 4 X − R [ X ] + L ϵ [ ( τ X s ) s ] , ∣ X s ∣ 2 = 1 , (92) \frac{\partial X}{\partial t}=-\mathcal{T}_{\epsilon}^{m}\partial_s^4X-\mathcal{R}[X]+\mathcal{L}_{\epsilon}[(\tau X_s)_s],\qquad|X_s|^2=1, \tag*{(92)} ∂ t ∂ X = − T ϵ m ∂ s 4 X − R [ X ] + L ϵ [( τ X s ) s ] , ∣ X s ∣ 2 = 1 , (92)
where the principal linear part of the evolution is given by T ϵ m ∂ s 4 ∼ ϵ − 1 ∂ s 3 \mathcal{T}_{\epsilon}^{m}\partial_s^4\sim\epsilon^{-1}\partial_s^3 T ϵ m ∂ s 4 ∼ ϵ − 1 ∂ s 3 , and the tension term and R \mathcal{R} R will be treated as remainders. We will make use of some general estimates for the linear part of (92) . Given a forcing function f ∈ L t 2 H s 1 / 2 f\in L_t^2H_s^{1/2} f ∈ L t 2 H s 1/2 , we consider the following linear equation on T:
∂ t u = − T m ϵ n ∂ s 4 u + f , u ∣ t = 0 = u i n . (93) \partial_t u=-T_{m_{\epsilon}^{n}}\partial_s^4u+f,\qquad u\big|_{t=0}=u^{\mathrm{in}}. \tag*{(93)} ∂ t u = − T m ϵ n ∂ s 4 u + f , u t = 0 = u in . (93)
Lemma 3.4 (Linear estimates). For T > 0 T>0 T > 0 and any u i n ∈ H s 2 ( T ) u^{\mathrm{in}}\in H_s^2(\mathbb{T}) u in ∈ H s 2 ( T ) , there exists a unique weak solution u ∈ L t ∞ H s 2 ∩ L t 2 H s 7 / 2 ( ( 0 , T ) × T ) u\in L_t^\infty H_s^2\cap L_t^2H_s^{7/2}((0,T)\times\mathbb{T}) u ∈ L t ∞ H s 2 ∩ L t 2 H s 7/2 (( 0 , T ) × T ) to the linear equation (93) satisfying
∥ u − u i n ∥ H s 2 → 0 as t → 0 + . (94) \lVert u-u^{\mathrm{in}}\rVert_{H_s^2}\to0\qquad\text{as }t\to0^+. \tag*{(94)} ∥ u − u in ∥ H s 2 → 0 as t → 0 + . (94) In addition, u u u is continuous on [ 0 , T ] [0,T] [ 0 , T ] with values in H s 2 H_s^2 H s 2 and satisfies the estimate
∥ u ∥ L t ∞ H s 2 + ∣ log ϵ ∣ 1 / 2 ∥ ∂ s 4 u ∥ L t 2 H s − 1 / 2 ≤ c ( ∥ u i n ∥ H s 2 + ( T 1 / 2 + ∣ log ϵ ∣ − 1 / 2 ) ∥ f ∥ L t 2 H s 1 / 2 ) (95) \lVert u\rVert_{L_t^\infty H_s^2}+|\log\epsilon|^{1/2}\lVert\partial_s^4u\rVert_{L_t^2H_s^{-1/2}}\leq c\left(\lVert u^{\mathrm{in}}\rVert_{H_s^2}+\left(T^{1/2}+|\log\epsilon|^{-1/2}\right)\lVert f\rVert_{L_t^2H_s^{1/2}}\right) \tag*{(95)} ∥ u ∥ L t ∞ H s 2 + ∣ log ϵ ∣ 1/2 ∥ ∂ s 4 u ∥ L t 2 H s − 1/2 ≤ c ( ∥ u in ∥ H s 2 + ( T 1/2 + ∣ log ϵ ∣ − 1/2 ) ∥ f ∥ L t 2 H s 1/2 ) (95) on ( 0 , T ) × T (0,T)\times\mathbb{T} ( 0 , T ) × T .
Proof. The mapping properties of the solution operator for u u u are immediate from the explicit solution formula using the Fourier multiplier behavior of Lemma 2.1, so it remains to show the bound (95) . We begin with a bound for ∂ s 2 u \partial_s^2u ∂ s 2 u . Upon differentiating (93) in s s s , we have that ∂ s 2 u \partial_s^2u ∂ s 2 u satisfies
∂ t ( ∂ s 2 u ) = − T m ϵ n ∂ s 4 ( ∂ s 2 u ) + ∂ s 2 f \partial_t(\partial_s^2u)=-T_{m_{\epsilon}^{n}}\partial_s^4(\partial_s^2u)+\partial_s^2f ∂ t ( ∂ s 2 u ) = − T m ϵ n ∂ s 4 ( ∂ s 2 u ) + ∂ s 2 f
∂ s 2 u ∣ t = 0 = ∂ s 2 u i n . (96) \partial_s^2u\big|_{t=0}=\partial_s^2u^{\mathrm{in}}. \tag*{(96)} ∂ s 2 u t = 0 = ∂ s 2 u in . (96)
Multiplying by ∂ s 2 u \partial_s^2u ∂ s 2 u and integrating by parts, we have that ∂ s 2 u \partial_s^2u ∂ s 2 u satisfies the energy identity
1 2 ∂ t ∫ T ( ∂ s 2 u ) 2 d s + ∫ T ∂ s 4 u T m ϵ n ∂ s 4 u d s = ∫ T ∂ s 4 u f d s . (97) \frac{1}{2}\partial_t\int_{\mathbb{T}}(\partial_s^2u)^2\,ds+\int_{\mathbb{T}}\partial_s^4u\,T_{m_{\epsilon}^{n}}\partial_s^4u\,ds=\int_{\mathbb{T}}\partial_s^4u f\,ds. \tag*{(97)} 2 1 ∂ t ∫ T ( ∂ s 2 u ) 2 d s + ∫ T ∂ s 4 u T m ϵ n ∂ s 4 u d s = ∫ T ∂ s 4 u f d s . (97)
where we have used that ∂ s \partial_s ∂ s commutes with T m ϵ n T_{m_{\epsilon}^{n}} T m ϵ n . We may bound the term involving T m ϵ n T_{m_{\epsilon}^{n}} T m ϵ n as
∫ T ∂ s 4 u T m ϵ n ∂ s 4 u d s = ∑ ∣ k ∣ = 0 ∞ ∣ F [ u ] ( k ) ∣ 2 ∣ k ∣ 8 m ϵ n ( k ) ≥ c ∣ log ϵ ∣ ∑ ∣ k ∣ = 0 ⌊ 1 2 π ϵ ⌋ ∣ F [ u ] ( k ) ∣ 2 ∣ k ∣ 8 + c ϵ − 1 ∑ ∣ k ∣ = ⌊ 1 2 π ϵ ⌋ + 1 ∞ ∣ k ∣ 7 ∣ F [ u ] ( k ) ∣ 2 ≥ c ∣ log ϵ ∣ ∥ ∂ s 4 u ∥ H − 1 / 2 ( T ) 2 . (98) \begin{aligned}
\int_{\mathbb{T}}\partial_s^4u\,T_{m_{\epsilon}^{n}}\partial_s^4u\,ds=\sum_{|k|=0}^{\infty}|\mathcal{F}[u](k)|^2|k|^8m_{\epsilon}^{n}(k) \\
&\geq c|\log\epsilon|\sum_{|k|=0}^{\left\lfloor\frac{1}{2\pi\epsilon}\right\rfloor}|\mathcal{F}[u](k)|^2|k|^8+c\epsilon^{-1}\sum_{|k|=\left\lfloor\frac{1}{2\pi\epsilon}\right\rfloor+1}^{\infty}|k|^7|\mathcal{F}[u](k)|^2 \\
&\geq c|\log\epsilon|\lVert\partial_s^4u\rVert_{H^{-1/2}(\mathbb{T})}^2. \tag*{(98)}
\end{aligned} ∫ T ∂ s 4 u T m ϵ n ∂ s 4 u d s = ∣ k ∣ = 0 ∑ ∞ ∣ F [ u ] ( k ) ∣ 2 ∣ k ∣ 8 m ϵ n ( k ) ≥ c ∣ log ϵ ∣ ∣ k ∣ = 0 ∑ ⌊ 2 π ϵ 1 ⌋ ∣ F [ u ] ( k ) ∣ 2 ∣ k ∣ 8 + c ϵ − 1 ∣ k ∣ = ⌊ 2 π ϵ 1 ⌋ + 1 ∑ ∞ ∣ k ∣ 7 ∣ F [ u ] ( k ) ∣ 2 ≥ c ∣ log ϵ ∣ ∥ ∂ s 4 u ∥ H − 1/2 ( T ) 2 . (98)
Integrating (97) in t t t , we then obtain the bound
∥ ∂ s 2 u ∥ L t ∞ L s 2 2 + ∣ log ϵ ∣ ∥ ∂ s 4 u ∥ L t 2 H s − 1 / 2 2 ≤ c ( ∥ ∂ s 2 u i n ∥ L t ∞ L s 2 2 + ∥ ∂ s 4 u ∥ L t 2 H s − 1 / 2 ∥ f ∥ L t 2 H s 1 / 2 ) . (99) \lVert\partial_s^2u\rVert_{L_t^\infty L_s^2}^2+|\log\epsilon|\lVert\partial_s^4u\rVert_{L_t^2H_s^{-1/2}}^2\leq c\left(\lVert\partial_s^2u^{\mathrm{in}}\rVert_{L_t^\infty L_s^2}^2+\lVert\partial_s^4u\rVert_{L_t^2H_s^{-1/2}}\lVert f\rVert_{L_t^2H_s^{1/2}}\right). \tag*{(99)} ∥ ∂ s 2 u ∥ L t ∞ L s 2 2 + ∣ log ϵ ∣ ∥ ∂ s 4 u ∥ L t 2 H s − 1/2 2 ≤ c ( ∥ ∂ s 2 u in ∥ L t ∞ L s 2 2 + ∥ ∂ s 4 u ∥ L t 2 H s − 1/2 ∥ f ∥ L t 2 H s 1/2 ) . (99)
An application of Young’s inequality yields (95) with L t ∞ H ˙ s 2 L_t^\infty\dot H_s^2 L t ∞ H ˙ s 2 in place of L t ∞ H s 2 L_t^\infty H_s^2 L t ∞ H s 2 .
It remains to bound the mean of u u u , which we denote by u 0 u_0 u 0 . We have that u 0 u_0 u 0 satisfies
∂ t u 0 = ∫ T f d s , u 0 ∣ t = 0 = u 0 i n , (100) \partial_tu_0=\int_{\mathbb{T}}f\,ds,\qquad u_0\big|_{t=0}=u_0^{\mathrm{in}}, \tag*{(100)} ∂ t u 0 = ∫ T f d s , u 0 t = 0 = u 0 in , (100)
so that
∥ u 0 ∥ L t ∞ ≤ ∣ u 0 i n ∣ + T 1 / 2 ∥ f ∥ L t 2 L s 2 . (101) \lVert u_0\rVert_{L_t^\infty}\leq|u_0^{\mathrm{in}}|+T^{1/2}\lVert f\rVert_{L_t^2L_s^2}. \tag*{(101)} ∥ u 0 ∥ L t ∞ ≤ ∣ u 0 in ∣ + T 1/2 ∥ f ∥ L t 2 L s 2 . (101)
Local well-posedness
Here we establish local-in-time well-posedness for the evolution (1.14), rewritten as (3.57).
Lemma 3.5 (Local well-posedness). Given ϵ > 0 \epsilon> 0 ϵ > 0 and X i n ∈ H s 2 ( T ) \mathbf{X}^{\mathrm{in}} \in H_s^2(\mathbb{T}) X in ∈ H s 2 ( T ) , there exists T > 0 T > 0 T > 0 and a unique solution X ∈ C ( [ 0 , T ] ; H s 2 ( T ) ) ∩ L t 2 H ˙ s 7 / 2 \mathbf{X} \in C([0,T]; H_s^2(\mathbb{T})) \cap L_t^2 \dot{H}_s^{7/2} X ∈ C ([ 0 , T ] ; H s 2 ( T )) ∩ L t 2 H ˙ s 7/2 to (3.57) with initial condition X i n \mathbf{X}^{\mathrm{in}} X in .
Given T f > 0 T_f > 0 T f > 0 , if X ∈ C ( [ 0 , T ] ; H s 2 ( T ) ) ∩ L t 2 H ˙ s 7 / 2 \mathbf{X} \in C([0,T]; H_s^2(\mathbb{T})) \cap L_t^2 \dot{H}_s^{7/2} X ∈ C ([ 0 , T ] ; H s 2 ( T )) ∩ L t 2 H ˙ s 7/2 is a solution for all T ∈ ( 0 , T f ) T \in(0,T_f) T ∈ ( 0 , T f ) and
∥ X ∥ L t ∞ H s 2 ∩ L t 2 H ˙ s 7 / 2 ( [ 0 , T ] × T ) < + ∞ , (102) \|\mathbf{X}\|_{L_t^\infty H_s^2\cap L_t^2\dot{H}_s^{7/2}([0,T]\times\mathbb{T})}<+\infty, \tag*{(102)} ∥ X ∥ L t ∞ H s 2 ∩ L t 2 H ˙ s 7/2 ([ 0 , T ] × T ) < + ∞ , (102) then the solution may be extended continuously in time past T f T_f T f , i.e. there exists δ > 0 \delta> 0 δ > 0 such that X ∈ C ( [ 0 , T f + δ ] ; H s 2 ( T ) ) ∩ L t 2 H ˙ s 7 / 2 \mathbf{X} \in C([0,T_f+\delta]; H_s^2(\mathbb{T})) \cap L_t^2\dot{H}_s^{7/2} X ∈ C ([ 0 , T f + δ ] ; H s 2 ( T )) ∩ L t 2 H ˙ s 7/2 .
Proof. For T > 0 T > 0 T > 0 , let Φ T ( X ) \Phi^T(\mathbf{X}) Φ T ( X ) denote the time-T T T solution map for the evolution (3.57). We proceed by a fixed point argument in the space Y 0 ∩ Y 1 \mathcal{Y}_0 \cap\mathcal{Y}_1 Y 0 ∩ Y 1 , where
Y 0 ( T ) = C ( [ 0 , T ] ; H s 2 ) , ∥ ⋅ ∥ Y 0 = sup t ∈ [ 0 , T ] ∥ ⋅ ∥ H s 2 ( T ) , Y 1 ( T ) = L t 2 H ˙ s 7 / 2 , ∥ ⋅ ∥ Y 1 = ( ∫ 0 T ∥ ⋅ ∥ H ˙ s 7 / 2 ( T ) 2 d t ′ ) 1 / 2 . (103) \begin{aligned}
\mathcal{Y}_0(T) &= C([0,T]; H_s^2), \qquad\|\cdot\|_{\mathcal{Y}_0} = \sup_{t\in[0,T]}\|\cdot\|_{H_s^2(T)},\\
\mathcal{Y}_1(T) &= L_t^2\dot{H}_s^{7/2}, \qquad\|\cdot\|_{\mathcal{Y}_1} = \left(\int_0^T \|\cdot\|_{\dot{H}_s^{7/2}(T)}^2\,dt'\right)^{1/2}. \tag*{(103)}
\end{aligned} Y 0 ( T ) Y 1 ( T ) = C ([ 0 , T ] ; H s 2 ) , ∥ ⋅ ∥ Y 0 = t ∈ [ 0 , T ] sup ∥ ⋅ ∥ H s 2 ( T ) , = L t 2 H ˙ s 7/2 , ∥ ⋅ ∥ Y 1 = ( ∫ 0 T ∥ ⋅ ∥ H ˙ s 7/2 ( T ) 2 d t ′ ) 1/2 . (103)
For some M 0 , M 1 > 0 M_0,M_1>0 M 0 , M 1 > 0 , we consider X \mathbf{X} X belonging to the closed ball
B 0 , 1 = { X ∈ Y 0 ( T ) ∩ Y 1 ( T ) : ∥ X ∥ Y 0 ≤ M 0 , ∥ X ∥ Y 1 ≤ M 1 } , (104) B_{0,1}=\{\mathbf{X}\in\mathcal{Y}_0(T)\cap\mathcal{Y}_1(T):\|\mathbf{X}\|_{\mathcal{Y}_0}\le M_0,\ \|\mathbf{X}\|_{\mathcal{Y}_1}\le M_1\}, \tag*{(104)} B 0 , 1 = { X ∈ Y 0 ( T ) ∩ Y 1 ( T ) : ∥ X ∥ Y 0 ≤ M 0 , ∥ X ∥ Y 1 ≤ M 1 } , (104)
and show that for some choice of T T T , the solution map Φ T ( ⋅ ) \Phi^T(\cdot) Φ T ( ⋅ ) maps B 0 , 1 B_{0,1} B 0 , 1 to itself.
We begin with the tension terms. First, we may note the following general bounds. Using Parseval’s theorem, the form (1.9) of T m ϵ t T_{m_\epsilon}^t T m ϵ t , and the multiplier bounds of Lemma 2.1, we have that for any f ∈ H − 1 / 2 ( T ) f\in H^{-1/2}(\mathbb{T}) f ∈ H − 1/2 ( T ) ,
∥ T m ϵ t ( P X s f ) ∥ H 1 / 2 2 = ∑ ∣ k ∣ = 0 ∞ m ϵ t ( k ) 2 ( 1 + ∣ k ∣ 2 ) 1 / 2 ∣ F [ P X s f ] ∣ 2 ≤ c ∣ log ϵ ∣ 2 ∑ ∣ k ∣ = 0 ⌊ 1 2 π ϵ ⌋ ( 1 + ∣ k ∣ 2 ) 1 / 2 ∣ F [ P X s f ] ∣ 2 + c ϵ − 2 ∑ ∣ k ∣ = ⌊ 1 2 π ϵ ⌋ + 1 ∞ ∣ k ∣ − 2 ( 1 + ∣ k ∣ 2 ) 1 / 2 ∣ F [ P X s f ] ∣ 2 ≤ c ϵ − 2 ∣ log ϵ ∣ 2 ∥ P X s f ∥ H − 1 / 2 2 . (105) \begin{aligned}
\|T_{m_\epsilon}^t(\mathbb{P}_{X_s}f)\|_{H^{1/2}}^2
&=\sum_{|k|=0}^{\infty}m_\epsilon^t(k)^2(1+|k|^2)^{1/2}\left|\mathcal{F}[\mathbb{P}_{X_s}f]\right|^2\\
&\le c|\log\epsilon|^2\sum_{|k|=0}^{\left\lfloor\frac{1}{2\pi\epsilon}\right\rfloor}(1+|k|^2)^{1/2}\left|\mathcal{F}[\mathbb{P}_{X_s}f]\right|^2\\
&\quad+c\epsilon^{-2}\sum_{|k|=\left\lfloor\frac{1}{2\pi\epsilon}\right\rfloor+1}^{\infty}|k|^{-2}(1+|k|^2)^{1/2}\left|\mathcal{F}[\mathbb{P}_{X_s}f]\right|^2\\
&\le c\epsilon^{-2}|\log\epsilon|^2\|\mathbb{P}_{X_s}f\|_{H^{-1/2}}^2. \tag*{(105)}
\end{aligned} ∥ T m ϵ t ( P X s f ) ∥ H 1/2 2 = ∣ k ∣ = 0 ∑ ∞ m ϵ t ( k ) 2 ( 1 + ∣ k ∣ 2 ) 1/2 ∣ F [ P X s f ] ∣ 2 ≤ c ∣ log ϵ ∣ 2 ∣ k ∣ = 0 ∑ ⌊ 2 π ϵ 1 ⌋ ( 1 + ∣ k ∣ 2 ) 1/2 ∣ F [ P X s f ] ∣ 2 + c ϵ − 2 ∣ k ∣ = ⌊ 2 π ϵ 1 ⌋ + 1 ∑ ∞ ∣ k ∣ − 2 ( 1 + ∣ k ∣ 2 ) 1/2 ∣ F [ P X s f ] ∣ 2 ≤ c ϵ − 2 ∣ log ϵ ∣ 2 ∥ P X s f ∥ H − 1/2 2 . (105)
Similarly,
∥ T m ϵ n ( P X s ⊥ f ) ∥ H 1 / 2 2 ≤ c ϵ − 2 ∣ log ϵ ∣ 2 ∥ P X s ⊥ f ∥ H − 1 / 2 2 . (106) \|T_{m_\epsilon}^n(\mathbb{P}_{X_s}^{\perp}f)\|_{H^{1/2}}^2\le c\epsilon^{-2}|\log\epsilon|^2\|\mathbb{P}_{X_s}^{\perp}f\|_{H^{-1/2}}^2. \tag*{(106)} ∥ T m ϵ n ( P X s ⊥ f ) ∥ H 1/2 2 ≤ c ϵ − 2 ∣ log ϵ ∣ 2 ∥ P X s ⊥ f ∥ H − 1/2 2 . (106)
We may thus estimate
∥ L ~ ϵ [ f ] ∥ H 1 / 2 ≤ c ( 1 + ∥ X s ∥ H 1 ) ( ∥ T m ϵ t ( P X s f ) ∥ H 1 / 2 + ∥ T m ϵ n ( P X s ⊥ f ) ∥ H 1 / 2 ) ≤ c ϵ − 1 ∣ log ϵ ∣ ( 1 + ∥ X s ∥ H 1 ) ( ∥ P X s f ∥ H − 1 / 2 + ∥ P X s ⊥ f ∥ H − 1 / 2 ) ≤ c ϵ − 1 ∣ log ϵ ∣ ( 1 + ∥ X s ∥ H 1 2 ) ∥ f ∥ H − 1 / 2 . (107) \begin{aligned}
\|\widetilde{\mathcal{L}}_\epsilon[f]\|_{H^{1/2}}
&\le c(1+\|X_s\|_{H^1})\left(\|T_{m_\epsilon}^t(\mathbb{P}_{X_s}f)\|_{H^{1/2}}+\|T_{m_\epsilon}^n(\mathbb{P}_{X_s}^{\perp}f)\|_{H^{1/2}}\right)\\
&\le c\epsilon^{-1}|\log\epsilon|(1+\|X_s\|_{H^1})\left(\|\mathbb{P}_{X_s}f\|_{H^{-1/2}}+\|\mathbb{P}_{X_s}^{\perp}f\|_{H^{-1/2}}\right)\\
&\le c\epsilon^{-1}|\log\epsilon|(1+\|X_s\|_{H^1}^2)\|f\|_{H^{-1/2}}. \tag*{(107)}
\end{aligned} ∥ L ϵ [ f ] ∥ H 1/2 ≤ c ( 1 + ∥ X s ∥ H 1 ) ( ∥ T m ϵ t ( P X s f ) ∥ H 1/2 + ∥ T m ϵ n ( P X s ⊥ f ) ∥ H 1/2 ) ≤ c ϵ − 1 ∣ log ϵ ∣ ( 1 + ∥ X s ∥ H 1 ) ( ∥ P X s f ∥ H − 1/2 + ∥ P X s ⊥ f ∥ H − 1/2 ) ≤ c ϵ − 1 ∣ log ϵ ∣ ( 1 + ∥ X s ∥ H 1 2 ) ∥ f ∥ H − 1/2 . (107)
Applying this bound to the tension term ( τ X s ) s (\tau X_s)_s ( τ X s ) s , we have
∥ L ~ ϵ [ ( τ X s ) s ] ∥ H 1 / 2 ≤ c ϵ − 1 ∣ log ϵ ∣ ( 1 + ∥ X s ∥ H s 1 2 ) ∥ τ X s ∥ H s 1 / 2 ≤ c ϵ − 2 ∣ log ϵ ∣ ( 1 + ∥ X s ∥ H s 1 6 ) ( ∥ X s s s ∥ H s 1 / 2 3 / 4 ∥ X s ∥ H s 1 9 / 4 + ∥ X s s s ∥ H s 1 / 2 2 / 3 ∥ X s ∥ H s 1 13 / 3 ) . (108) \begin{aligned}
\|\widetilde{\mathcal{L}}_\epsilon[(\tau X_s)_s]\|_{H^{1/2}}
&\le c\epsilon^{-1}|\log\epsilon|(1+\|X_s\|_{H_s^1}^2)\|\tau X_s\|_{H_s^{1/2}}\\
&\le c\epsilon^{-2}|\log\epsilon|(1+\|X_s\|_{H_s^1}^6)
\left(\|X_{sss}\|_{H_s^{1/2}}^{3/4}\|X_s\|_{H_s^1}^{9/4}
+\|X_{sss}\|_{H_s^{1/2}}^{2/3}\|X_s\|_{H_s^1}^{13/3}\right). \tag*{(108)}
\end{aligned} ∥ L ϵ [( τ X s ) s ] ∥ H 1/2 ≤ c ϵ − 1 ∣ log ϵ ∣ ( 1 + ∥ X s ∥ H s 1 2 ) ∥ τ X s ∥ H s 1/2 ≤ c ϵ − 2 ∣ log ϵ ∣ ( 1 + ∥ X s ∥ H s 1 6 ) ( ∥ X sss ∥ H s 1/2 3/4 ∥ X s ∥ H s 1 9/4 + ∥ X sss ∥ H s 1/2 2/3 ∥ X s ∥ H s 1 13/3 ) . (108)
by Lemma 3.2. Integrating in time from 0 to T T T , for X ∈ B 0 , 1 X \in B_{0,1} X ∈ B 0 , 1 , we have
∥ L ‾ ϵ [ ( τ X s ) s ] ∥ L t 2 H ˙ s 1 / 2 2 ≤ c ϵ − 2 ∣ log ϵ ∣ ( ( 1 + ∥ X s ∥ L t ∞ H s 1 6 ) ( T 1 / 8 ∥ X s s s ∥ L t 2 H ˙ s 1 / 2 3 / 4 ∥ X s ∥ L t ∞ H s 1 9 / 4 + T 1 / 6 ∥ X s s s ∥ L t 2 H ˙ s 1 / 2 2 / 3 ∥ X s ∥ L t ∞ H s 1 13 / 3 ) ) ≤ c ϵ − 2 ∣ log ϵ ∣ ( ( 1 + M 0 6 ) ( T 1 / 8 M 1 3 / 4 M 0 9 / 4 + T 1 / 6 M 1 2 / 3 M 0 13 / 3 ) ) . (109) \begin{aligned}
\left\|\overline{\mathcal{L}}_\epsilon[(\tau X_s)_s]\right\|_{L_t^2\dot{H}_s^{1/2}}^2 \le c\epsilon^{-2}|\log\epsilon|\left((1+\|X_s\|_{L_t^\infty H_s^1}^6)\left(T^{1/8}\|X_{sss}\|_{L_t^2\dot{H}_s^{1/2}}^{3/4}\|X_s\|_{L_t^\infty H_s^1}^{9/4}+T^{1/6}\|X_{sss}\|_{L_t^2\dot{H}_s^{1/2}}^{2/3}\|X_s\|_{L_t^\infty H_s^1}^{13/3}\right)\right) \\
&\le c\epsilon^{-2}|\log\epsilon|\left((1+M_0^6)\left(T^{1/8}M_1^{3/4}M_0^{9/4}+T^{1/6}M_1^{2/3}M_0^{13/3}\right)\right). \tag*{(109)}
\end{aligned} L ϵ [( τ X s ) s ] L t 2 H ˙ s 1/2 2 ≤ c ϵ − 2 ∣ log ϵ ∣ ( ( 1 + ∥ X s ∥ L t ∞ H s 1 6 ) ( T 1/8 ∥ X sss ∥ L t 2 H ˙ s 1/2 3/4 ∥ X s ∥ L t ∞ H s 1 9/4 + T 1/6 ∥ X sss ∥ L t 2 H ˙ s 1/2 2/3 ∥ X s ∥ L t ∞ H s 1 13/3 ) ) ≤ c ϵ − 2 ∣ log ϵ ∣ ( ( 1 + M 0 6 ) ( T 1/8 M 1 3/4 M 0 9/4 + T 1/6 M 1 2/3 M 0 13/3 ) ) . (109)
For the remainder terms R [ X ] R[X] R [ X ] , by Lemma 3.3, we additionally have
∥ R [ X ] ∥ L t 2 H ˙ s 1 / 2 2 ≤ c ϵ − 3 / 2 ( T 1 / 6 ∥ X s s s ∥ L t 2 H ˙ s 1 / 2 2 / 3 ∥ X s ∥ L t ∞ H s 1 4 / 3 + T 1 / 2 ∥ X s ∥ L t ∞ H s 1 2 ) ( 1 + ∥ X s ∥ L t ∞ H s 1 2 ) ≤ c ϵ − 3 / 2 ( T 1 / 6 M 1 2 / 3 M 0 4 / 3 + T 1 / 2 M 0 2 ) ( 1 + M 0 2 ) . (110) \begin{aligned}
\|R[X]\|_{L_t^2\dot{H}_s^{1/2}}^2 \le c\epsilon^{-3/2}\left(T^{1/6}\|X_{sss}\|_{L_t^2\dot{H}_s^{1/2}}^{2/3}\|X_s\|_{L_t^\infty H_s^1}^{4/3}+T^{1/2}\|X_s\|_{L_t^\infty H_s^1}^2\right)(1+\|X_s\|_{L_t^\infty H_s^1}^2) \\
&\le c\epsilon^{-3/2}\left(T^{1/6}M_1^{2/3}M_0^{4/3}+T^{1/2}M_0^2\right)(1+M_0^2). \tag*{(110)}
\end{aligned} ∥ R [ X ] ∥ L t 2 H ˙ s 1/2 2 ≤ c ϵ − 3/2 ( T 1/6 ∥ X sss ∥ L t 2 H ˙ s 1/2 2/3 ∥ X s ∥ L t ∞ H s 1 4/3 + T 1/2 ∥ X s ∥ L t ∞ H s 1 2 ) ( 1 + ∥ X s ∥ L t ∞ H s 1 2 ) ≤ c ϵ − 3/2 ( T 1/6 M 1 2/3 M 0 4/3 + T 1/2 M 0 2 ) ( 1 + M 0 2 ) . (110)
Altogether, using Lemma 3.4, we may bound
∥ X ∥ Y 0 ≤ ∥ X i n ∥ H s 2 + c ϵ − 3 / 2 ( T 1 / 2 + ∣ log ϵ ∣ − 1 / 2 ) ( T 1 / 6 M 1 2 / 3 M 0 4 / 3 + T 1 / 2 M 0 2 ) ( 1 + M 0 2 ) + c ϵ − 2 ∣ log ϵ ∣ ( T 1 / 2 + ∣ log ϵ ∣ − 1 / 2 ) ( 1 + M 0 6 ) ( T 1 / 8 M 1 3 / 4 M 0 9 / 4 + T 1 / 6 M 1 2 / 3 M 0 13 / 3 ) ≤ ∥ X i n ∥ H s 2 + c 0 ( M 0 , M 1 ) ϵ − 2 ∣ log ϵ ∣ T 1 / 8 , \begin{aligned}
\|X\|_{\mathcal{Y}_0}\le\|X^{\mathrm{in}}\|_{H_s^2}+c\epsilon^{-3/2}(T^{1/2}+|\log\epsilon|^{-1/2})(T^{1/6}M_1^{2/3}M_0^{4/3}+T^{1/2}M_0^2)(1+M_0^2) \\
&+c\epsilon^{-2}|\log\epsilon|(T^{1/2}+|\log\epsilon|^{-1/2})(1+M_0^6)(T^{1/8}M_1^{3/4}M_0^{9/4}+T^{1/6}M_1^{2/3}M_0^{13/3}) \\
&\le\|X^{\mathrm{in}}\|_{H_s^2}+c_0(M_0,M_1)\epsilon^{-2}|\log\epsilon|T^{1/8},
\end{aligned} ∥ X ∥ Y 0 ≤ ∥ X in ∥ H s 2 + c ϵ − 3/2 ( T 1/2 + ∣ log ϵ ∣ − 1/2 ) ( T 1/6 M 1 2/3 M 0 4/3 + T 1/2 M 0 2 ) ( 1 + M 0 2 ) + c ϵ − 2 ∣ log ϵ ∣ ( T 1/2 + ∣ log ϵ ∣ − 1/2 ) ( 1 + M 0 6 ) ( T 1/8 M 1 3/4 M 0 9/4 + T 1/6 M 1 2/3 M 0 13/3 ) ≤ ∥ X in ∥ H s 2 + c 0 ( M 0 , M 1 ) ϵ − 2 ∣ log ϵ ∣ T 1/8 ,
∥ X ∥ Y 1 ≤ ∣ log ϵ ∣ − 1 / 2 ∥ e − t T m n ϵ ∂ s 4 X i n ∥ Y 1 + c ϵ − 3 / 2 ∣ log ϵ ∣ − 1 / 2 ( T 1 / 2 + ∣ log ϵ ∣ − 1 / 2 ) ( T 1 / 6 M 1 2 / 3 M 0 4 / 3 + T 1 / 2 M 0 2 ) ( 1 + M 0 2 ) + c ϵ − 2 ∣ log ϵ ∣ 1 / 2 ( T 1 / 2 + ∣ log ϵ ∣ − 1 / 2 ) ( 1 + M 0 6 ) ( T 1 / 8 M 1 3 / 4 M 0 9 / 4 + T 1 / 6 M 1 2 / 3 M 0 13 / 3 ) ≤ ∣ log ϵ ∣ − 1 / 2 M 1 i n ( T ) + c 1 ( M 0 , M 1 ) ϵ − 2 ∣ log ϵ ∣ 1 / 2 T 1 / 8 , (111) \begin{aligned}
\|X\|_{\mathcal{Y}_1}\le|\log\epsilon|^{-1/2}\left\|e^{-tT_m^{n_\epsilon}\partial_s^4}X^{\mathrm{in}}\right\|_{\mathcal{Y}_1} \\
&+c\epsilon^{-3/2}|\log\epsilon|^{-1/2}(T^{1/2}+|\log\epsilon|^{-1/2})(T^{1/6}M_1^{2/3}M_0^{4/3}+T^{1/2}M_0^2)(1+M_0^2) \\
&+c\epsilon^{-2}|\log\epsilon|^{1/2}(T^{1/2}+|\log\epsilon|^{-1/2})(1+M_0^6)(T^{1/8}M_1^{3/4}M_0^{9/4}+T^{1/6}M_1^{2/3}M_0^{13/3}) \\
&\le|\log\epsilon|^{-1/2}M_1^{\mathrm{in}}(T)+c_1(M_0,M_1)\epsilon^{-2}|\log\epsilon|^{1/2}T^{1/8}, \tag*{(111)}
\end{aligned} ∥ X ∥ Y 1 ≤ ∣ log ϵ ∣ − 1/2 e − t T m n ϵ ∂ s 4 X in Y 1 + c ϵ − 3/2 ∣ log ϵ ∣ − 1/2 ( T 1/2 + ∣ log ϵ ∣ − 1/2 ) ( T 1/6 M 1 2/3 M 0 4/3 + T 1/2 M 0 2 ) ( 1 + M 0 2 ) + c ϵ − 2 ∣ log ϵ ∣ 1/2 ( T 1/2 + ∣ log ϵ ∣ − 1/2 ) ( 1 + M 0 6 ) ( T 1/8 M 1 3/4 M 0 9/4 + T 1/6 M 1 2/3 M 0 13/3 ) ≤ ∣ log ϵ ∣ − 1/2 M 1 in ( T ) + c 1 ( M 0 , M 1 ) ϵ − 2 ∣ log ϵ ∣ 1/2 T 1/8 , (111)
where M 1 i n ( T ) : = ∥ e − t T m n ϵ ∂ s 4 X i n ∥ Y 1 → 0 M_1^{\mathrm{in}}(T):=\left\|e^{-tT_m^{n_\epsilon}\partial_s^4}X^{\mathrm{in}}\right\|_{\mathcal{Y}_1}\to0 M 1 in ( T ) := e − t T m n ϵ ∂ s 4 X in Y 1 → 0 as T → 0 + T\to0^+ T → 0 + since ∥ ⋅ ∥ Y 1 = ∥ ⋅ ∥ L t 2 H ˙ s 7 / 2 \|\cdot\|_{\mathcal{Y}_1}=\|\cdot\|_{L_t^2\dot{H}_s^{7/2}} ∥ ⋅ ∥ Y 1 = ∥ ⋅ ∥ L t 2 H ˙ s 7/2 is an integral quantity in t t t . Taking M 0 = 4 ∥ X i n ∥ H s 2 M_0=4\|X^{\mathrm{in}}\|_{H_s^2} M 0 = 4∥ X in ∥ H s 2 and then choosing T T T small enough that
c 0 ( M 0 , M 1 ) ϵ − 2 ∣ log ϵ ∣ T 1 / 8 ≤ M 0 2 c_0(M_0,M_1)\epsilon^{-2}|\log\epsilon|T^{1/8}\le\frac{M_0}{2} c 0 ( M 0 , M 1 ) ϵ − 2 ∣ log ϵ ∣ T 1/8 ≤ 2 M 0
and
∣ log ϵ ∣ − 1 / 2 M 1 i n ( T ) + c 1 ( M 0 , M 1 ) ϵ − 2 ∣ log ϵ ∣ 1 / 2 T 1 / 8 ≤ 3 M 1 4 , (112) |\log\epsilon|^{-1/2}M_1^{\mathrm{in}}(T)+c_1(M_0,M_1)\epsilon^{-2}|\log\epsilon|^{1/2}T^{1/8}\le\frac{3M_1}{4}, \tag*{(112)} ∣ log ϵ ∣ − 1/2 M 1 in ( T ) + c 1 ( M 0 , M 1 ) ϵ − 2 ∣ log ϵ ∣ 1/2 T 1/8 ≤ 4 3 M 1 , (112)
we obtain that the time-T T T solution map Φ T ( X ) \Phi^T(X) Φ T ( X ) maps B 0 , 1 B_{0,1} B 0 , 1 to itself.
We next show that the solution map Φ T \Phi^T Φ T is a contraction on B 0 , 1 B_{0,1} B 0 , 1 . Again, we begin with the tension term. First, for any f 1 , f 2 : T → R 3 f_1,f_2:\mathbb{T}\to\mathbb{R}^3 f 1 , f 2 : T → R 3 along corresponding nearby curves X 1 , X 2 X_1,X_2 X 1 , X 2 , we may write
L ‾ ϵ [ f 1 ] − L ‾ ϵ [ f 2 ] = P ( X 1 ) s T m ϵ m ( P ( X 1 ) s f 1 ) − P ( X 2 ) s T m ϵ m ( P ( X 2 ) s f 2 ) + P ( X 1 ) s ⊥ T m ϵ m ( P ( X 1 ) s ⊥ f 1 ) − P ( X 2 ) s ⊥ T m ϵ m ( P ( X 2 ) s ⊥ f 2 ) . (113) \begin{aligned}
\overline{\mathcal{L}}_\epsilon[f_1]-\overline{\mathcal{L}}_\epsilon[f_2]
=\mathbb{P}_{(X_1)_s}T_{m_\epsilon}^{m}\bigl(\mathbb{P}_{(X_1)_s}f_1\bigr)-\mathbb{P}_{(X_2)_s}T_{m_\epsilon}^{m}\bigl(\mathbb{P}_{(X_2)_s}f_2\bigr) \\
&+\mathbb{P}_{(X_1)_s}^{\perp}T_{m_\epsilon}^{m}\bigl(\mathbb{P}_{(X_1)_s}^{\perp}f_1\bigr)-\mathbb{P}_{(X_2)_s}^{\perp}T_{m_\epsilon}^{m}\bigl(\mathbb{P}_{(X_2)_s}^{\perp}f_2\bigr). \tag*{(113)}
\end{aligned} L ϵ [ f 1 ] − L ϵ [ f 2 ] = P ( X 1 ) s T m ϵ m ( P ( X 1 ) s f 1 ) − P ( X 2 ) s T m ϵ m ( P ( X 2 ) s f 2 ) + P ( X 1 ) s ⊥ T m ϵ m ( P ( X 1 ) s ⊥ f 1 ) − P ( X 2 ) s ⊥ T m ϵ m ( P ( X 2 ) s ⊥ f 2 ) . (113)
We may then estimate
∥ L ‾ ϵ [ f 1 ] − L ‾ ϵ [ f 2 ] ∥ H 1 / 2 ≤ ∥ P ( X 1 ) s − P ( X 2 ) s ∥ H 1 ∥ T ϵ m , t ( P ( X 1 ) s f 1 ) ∥ H 1 / 2 + ∥ P ( X 2 ) s ∥ H 1 ( ∥ T ϵ m , t ( ( P ( X 1 ) s − P ( X 2 ) s ) f 1 ) ∥ H 1 / 2 + ∥ T ϵ m , t ( P ( X 2 ) s ( f 1 − f 2 ) ) ∥ H 1 / 2 ) + ∥ P ⊥ ( X 1 ) s − P ⊥ ( X 2 ) s ∥ H 1 ∥ T ϵ m , n ( P ⊥ ( X 1 ) s f 1 ) ∥ H 1 / 2 + ∥ P ⊥ ( X 2 ) s ∥ H 1 ( ∥ T ϵ m , n ( ( P ⊥ ( X 1 ) s − P ⊥ ( X 2 ) s ) f 1 ) ∥ H 1 / 2 + ∥ T ϵ m , n ( P ⊥ ( X 2 ) s ( f 1 − f 2 ) ) ∥ H 1 / 2 ) ≤ c ϵ − 1 ∣ log ϵ ∣ ∥ ( X 1 − X 2 ) s ∥ H 1 ( ∥ P ( X 1 ) s f 1 ∥ H − 1 / 2 + ∥ P ⊥ ( X 1 ) s f 1 ∥ H − 1 / 2 ) + c ϵ − 1 ∣ log ϵ ∣ ( 1 + ∥ ( X 2 ) s ∥ H 1 ) ∥ ( P ( X 1 ) s − P ( X 2 ) s ) f 1 ∥ H − 1 / 2 + c ϵ − 1 ∣ log ϵ ∣ ( 1 + ∥ ( X 2 ) s ∥ H 1 ) ( ∥ P ( X 2 ) s ( f 1 − f 2 ) ∥ H − 1 / 2 + ∥ P ⊥ ( X 2 ) s ( f 1 − f 2 ) ∥ H − 1 / 2 ) ≤ c ϵ − 1 ∣ log ϵ ∣ ∥ ( X 1 − X 2 ) s ∥ H 1 ( 1 + ∥ ( X 1 ) s ∥ H 1 + ∥ ( X 2 ) s ∥ H 1 ) ∥ f 1 ∥ H − 1 / 2 + c ϵ − 1 ∣ log ϵ ∣ ( 1 + ∥ ( X 2 ) s ∥ H 1 ) 2 ∥ f 1 − f 2 ∥ H − 1 / 2 . (114) \begin{aligned}
\left\|\overline{\mathcal{L}}_{\epsilon}[f_1]-\overline{\mathcal{L}}_{\epsilon}[f_2]\right\|_{H^{1/2}}
&\leq\left\|\mathbb{P}(X_1)_s-\mathbb{P}(X_2)_s\right\|_{H^1}\left\|T_{\epsilon}^{m,t}\left(\mathbb{P}(X_1)_s f_1\right)\right\|_{H^{1/2}}\\
&\quad+\left\|\mathbb{P}(X_2)_s\right\|_{H^1}\left(\left\|T_{\epsilon}^{m,t}\left((\mathbb{P}(X_1)_s-\mathbb{P}(X_2)_s)f_1\right)\right\|_{H^{1/2}}+\left\|T_{\epsilon}^{m,t}\left(\mathbb{P}(X_2)_s(f_1-f_2)\right)\right\|_{H^{1/2}}\right)\\
&\quad+\left\|\mathbb{P}^{\perp}(X_1)_s-\mathbb{P}^{\perp}(X_2)_s\right\|_{H^1}\left\|T_{\epsilon}^{m,n}\left(\mathbb{P}^{\perp}(X_1)_s f_1\right)\right\|_{H^{1/2}}\\
&\quad+\left\|\mathbb{P}^{\perp}(X_2)_s\right\|_{H^1}\left(\left\|T_{\epsilon}^{m,n}\left((\mathbb{P}^{\perp}(X_1)_s-\mathbb{P}^{\perp}(X_2)_s)f_1\right)\right\|_{H^{1/2}}+\left\|T_{\epsilon}^{m,n}\left(\mathbb{P}^{\perp}(X_2)_s(f_1-f_2)\right)\right\|_{H^{1/2}}\right)\\
&\leq c\epsilon^{-1}|\log\epsilon|\left\|(X_1-X_2)_s\right\|_{H^1}\left(\left\|\mathbb{P}(X_1)_s f_1\right\|_{H^{-1/2}}+\left\|\mathbb{P}^{\perp}(X_1)_s f_1\right\|_{H^{-1/2}}\right)\\
&\quad+c\epsilon^{-1}|\log\epsilon|\left(1+\left\|(X_2)_s\right\|_{H^1}\right)\left\|(\mathbb{P}(X_1)_s-\mathbb{P}(X_2)_s)f_1\right\|_{H^{-1/2}}\\
&\quad+c\epsilon^{-1}|\log\epsilon|\left(1+\left\|(X_2)_s\right\|_{H^1}\right)\left(\left\|\mathbb{P}(X_2)_s(f_1-f_2)\right\|_{H^{-1/2}}+\left\|\mathbb{P}^{\perp}(X_2)_s(f_1-f_2)\right\|_{H^{-1/2}}\right)\\
&\leq c\epsilon^{-1}|\log\epsilon|\left\|(X_1-X_2)_s\right\|_{H^1}\left(1+\left\|(X_1)_s\right\|_{H^1}+\left\|(X_2)_s\right\|_{H^1}\right)\left\|f_1\right\|_{H^{-1/2}}\\
&\quad+c\epsilon^{-1}|\log\epsilon|\left(1+\left\|(X_2)_s\right\|_{H^1}\right)^2\left\|f_1-f_2\right\|_{H^{-1/2}} . \tag*{(114)}
\end{aligned} L ϵ [ f 1 ] − L ϵ [ f 2 ] H 1/2 ≤ ∥ P ( X 1 ) s − P ( X 2 ) s ∥ H 1 T ϵ m , t ( P ( X 1 ) s f 1 ) H 1/2 + ∥ P ( X 2 ) s ∥ H 1 ( T ϵ m , t ( ( P ( X 1 ) s − P ( X 2 ) s ) f 1 ) H 1/2 + T ϵ m , t ( P ( X 2 ) s ( f 1 − f 2 ) ) H 1/2 ) + P ⊥ ( X 1 ) s − P ⊥ ( X 2 ) s H 1 T ϵ m , n ( P ⊥ ( X 1 ) s f 1 ) H 1/2 + P ⊥ ( X 2 ) s H 1 ( T ϵ m , n ( ( P ⊥ ( X 1 ) s − P ⊥ ( X 2 ) s ) f 1 ) H 1/2 + T ϵ m , n ( P ⊥ ( X 2 ) s ( f 1 − f 2 ) ) H 1/2 ) ≤ c ϵ − 1 ∣ log ϵ ∣ ∥ ( X 1 − X 2 ) s ∥ H 1 ( ∥ P ( X 1 ) s f 1 ∥ H − 1/2 + P ⊥ ( X 1 ) s f 1 H − 1/2 ) + c ϵ − 1 ∣ log ϵ ∣ ( 1 + ∥ ( X 2 ) s ∥ H 1 ) ∥ ( P ( X 1 ) s − P ( X 2 ) s ) f 1 ∥ H − 1/2 + c ϵ − 1 ∣ log ϵ ∣ ( 1 + ∥ ( X 2 ) s ∥ H 1 ) ( ∥ P ( X 2 ) s ( f 1 − f 2 ) ∥ H − 1/2 + P ⊥ ( X 2 ) s ( f 1 − f 2 ) H − 1/2 ) ≤ c ϵ − 1 ∣ log ϵ ∣ ∥ ( X 1 − X 2 ) s ∥ H 1 ( 1 + ∥ ( X 1 ) s ∥ H 1 + ∥ ( X 2 ) s ∥ H 1 ) ∥ f 1 ∥ H − 1/2 + c ϵ − 1 ∣ log ϵ ∣ ( 1 + ∥ ( X 2 ) s ∥ H 1 ) 2 ∥ f 1 − f 2 ∥ H − 1/2 . (114)
Plugging in f 1 = ( τ 1 ( X 1 ) s ) s f_1=(\tau_1(X_1)_s)_s f 1 = ( τ 1 ( X 1 ) s ) s and f 2 = ( τ 2 ( X 2 ) s ) s f_2=(\tau_2(X_2)_s)_s f 2 = ( τ 2 ( X 2 ) s ) s for τ 1 \tau_1 τ 1 , τ 2 \tau_2 τ 2 satisfying the tension equations (42) corresponding to X 1 X_1 X 1 and X 2 X_2 X 2 , respectively, we have
∥ L ‾ ϵ [ ( τ 1 ( X 1 ) s ) s ] − L ‾ ϵ [ ( τ 2 ( X 2 ) s ) s ] ∥ H s 1 / 2 ≤ c ϵ − 1 ∣ log ϵ ∣ ∥ ( X 1 − X 2 ) s ∥ H s 1 ( 1 + ∥ ( X 1 ) s ∥ H s 1 + ∥ ( X 2 ) s ∥ H s 1 ) ∥ τ 1 ( X 1 ) s ∥ H s 1 / 2 + c ϵ − 1 ∣ log ϵ ∣ ( 1 + ∥ ( X 2 ) s ∥ H s 1 ) 2 ∥ τ 1 ( X 1 ) s − τ 2 ( X 2 ) s ∥ H s 1 / 2 ≤ c ϵ − 1 ∣ log ϵ ∣ ∥ ( X 1 − X 2 ) s ∥ H s 1 ( 1 + ∥ ( X 1 ) s ∥ H s 1 + ∥ ( X 2 ) s ∥ H s 1 ) 2 ( ∥ τ 1 ∥ H s 1 / 2 + ∥ τ 2 ∥ H s 1 / 2 ) + c ϵ − 1 ∣ log ϵ ∣ ( 1 + ∥ ( X 2 ) s ∥ H s 1 ) 2 ∥ ( X 1 ) s ∥ H s 1 ∥ τ 1 − τ 2 ∥ H s 1 / 2 ≤ c ϵ − 2 ∣ log ϵ ∣ ( 1 + ∥ ( X 1 ) s ∥ H s 1 2 + ∥ ( X 2 ) s ∥ H s 1 2 ) × [ ∥ ( X 1 − X 2 ) s s s ∥ H s 1 / 2 1 / 3 ∥ ( X 1 − X 2 ) s ∥ H s 1 2 / 3 ( ∥ ( X 1 ) s s s ∥ H s 1 / 2 1 / 3 + ∥ ( X 2 ) s s s ∥ H s 1 / 2 1 / 3 ) + ∥ ( X 1 − X 2 ) s ∥ H s 1 ( ∥ ( X 1 ) s s s ∥ H s 1 / 2 2 / 3 + ∥ ( X 2 ) s s s ∥ H s 1 / 2 2 / 3 + ∥ ( X 1 ) s s s ∥ H s 1 / 2 3 / 4 + ∥ ( X 2 ) s s s ∥ H s 1 / 2 3 / 4 ) + ∥ ( X 1 − X 2 ) s s s ∥ H s 1 / 2 3 / 4 ∥ ( X 1 − X 2 ) s ∥ H s 1 1 / 4 ] . (115) \begin{aligned}
\left\|\overline{\mathcal{L}}_{\epsilon}[(\tau_1(X_1)_s)_s]-\overline{\mathcal{L}}_{\epsilon}[(\tau_2(X_2)_s)_s]\right\|_{H_s^{1/2}}
&\leq c\epsilon^{-1}|\log\epsilon|\left\|(X_1-X_2)_s\right\|_{H_s^1}\left(1+\left\|(X_1)_s\right\|_{H_s^1}+\left\|(X_2)_s\right\|_{H_s^1}\right)\left\|\tau_1(X_1)_s\right\|_{H_s^{1/2}}\\
&\quad+c\epsilon^{-1}|\log\epsilon|\left(1+\left\|(X_2)_s\right\|_{H_s^1}\right)^2\left\|\tau_1(X_1)_s-\tau_2(X_2)_s\right\|_{H_s^{1/2}}\\
&\leq c\epsilon^{-1}|\log\epsilon|\left\|(X_1-X_2)_s\right\|_{H_s^1}\left(1+\left\|(X_1)_s\right\|_{H_s^1}+\left\|(X_2)_s\right\|_{H_s^1}\right)^2\left(\left\|\tau_1\right\|_{H_s^{1/2}}+\left\|\tau_2\right\|_{H_s^{1/2}}\right)\\
&\quad+c\epsilon^{-1}|\log\epsilon|\left(1+\left\|(X_2)_s\right\|_{H_s^1}\right)^2\left\|(X_1)_s\right\|_{H_s^1}\left\|\tau_1-\tau_2\right\|_{H_s^{1/2}}\\
&\leq c\epsilon^{-2}|\log\epsilon|\left(1+\left\|(X_1)_s\right\|_{H_s^1}^2+\left\|(X_2)_s\right\|_{H_s^1}^2\right)\\
&\quad\times\left[\left\|(X_1-X_2)_{sss}\right\|_{H_s^{1/2}}^{1/3}\left\|(X_1-X_2)_s\right\|_{H_s^1}^{2/3}\left(\left\|(X_1)_{sss}\right\|_{H_s^{1/2}}^{1/3}+\left\|(X_2)_{sss}\right\|_{H_s^{1/2}}^{1/3}\right)\right.\\
&\qquad+\left\|(X_1-X_2)_s\right\|_{H_s^1}\left(\left\|(X_1)_{sss}\right\|_{H_s^{1/2}}^{2/3}+\left\|(X_2)_{sss}\right\|_{H_s^{1/2}}^{2/3}+\left\|(X_1)_{sss}\right\|_{H_s^{1/2}}^{3/4}+\left\|(X_2)_{sss}\right\|_{H_s^{1/2}}^{3/4}\right)\\
&\qquad\left.+\left\|(X_1-X_2)_{sss}\right\|_{H_s^{1/2}}^{3/4}\left\|(X_1-X_2)_s\right\|_{H_s^1}^{1/4}\right]. \tag*{(115)}
\end{aligned} L ϵ [( τ 1 ( X 1 ) s ) s ] − L ϵ [( τ 2 ( X 2 ) s ) s ] H s 1/2 ≤ c ϵ − 1 ∣ log ϵ ∣ ∥ ( X 1 − X 2 ) s ∥ H s 1 ( 1 + ∥ ( X 1 ) s ∥ H s 1 + ∥ ( X 2 ) s ∥ H s 1 ) ∥ τ 1 ( X 1 ) s ∥ H s 1/2 + c ϵ − 1 ∣ log ϵ ∣ ( 1 + ∥ ( X 2 ) s ∥ H s 1 ) 2 ∥ τ 1 ( X 1 ) s − τ 2 ( X 2 ) s ∥ H s 1/2 ≤ c ϵ − 1 ∣ log ϵ ∣ ∥ ( X 1 − X 2 ) s ∥ H s 1 ( 1 + ∥ ( X 1 ) s ∥ H s 1 + ∥ ( X 2 ) s ∥ H s 1 ) 2 ( ∥ τ 1 ∥ H s 1/2 + ∥ τ 2 ∥ H s 1/2 ) + c ϵ − 1 ∣ log ϵ ∣ ( 1 + ∥ ( X 2 ) s ∥ H s 1 ) 2 ∥ ( X 1 ) s ∥ H s 1 ∥ τ 1 − τ 2 ∥ H s 1/2 ≤ c ϵ − 2 ∣ log ϵ ∣ ( 1 + ∥ ( X 1 ) s ∥ H s 1 2 + ∥ ( X 2 ) s ∥ H s 1 2 ) × [ ∥ ( X 1 − X 2 ) sss ∥ H s 1/2 1/3 ∥ ( X 1 − X 2 ) s ∥ H s 1 2/3 ( ∥ ( X 1 ) sss ∥ H s 1/2 1/3 + ∥ ( X 2 ) sss ∥ H s 1/2 1/3 ) + ∥ ( X 1 − X 2 ) s ∥ H s 1 ( ∥ ( X 1 ) sss ∥ H s 1/2 2/3 + ∥ ( X 2 ) sss ∥ H s 1/2 2/3 + ∥ ( X 1 ) sss ∥ H s 1/2 3/4 + ∥ ( X 2 ) sss ∥ H s 1/2 3/4 ) + ∥ ( X 1 − X 2 ) sss ∥ H s 1/2 3/4 ∥ ( X 1 − X 2 ) s ∥ H s 1 1/4 ] . (115)
Here we have used the tension bounds of Lemma 3.2. Integrating in time from 0 to T T T , we obtain the L t 2 L_t^2 L t 2 bound
∥ L ‾ ϵ [ ( τ 1 ( X 1 ) s ) s ] − L ‾ ϵ [ ( τ 2 ( X 2 ) s ) s ] ∥ L T 2 H s 1 / 2 ≤ c ϵ − 2 ∣ log ϵ ∣ ( 1 + M 0 12 ) [ T 1 / 6 ∥ X 1 − X 2 ∥ Y 1 1 / 3 ∥ X 1 − X 2 ∥ Y 0 2 / 3 M 1 1 / 3 + T 1 / 8 ∥ X 1 − X 2 ∥ Y 1 3 / 4 ∥ X 1 − X 2 ∥ Y 0 1 / 4 + ∥ X 1 − X 2 ∥ Y 0 ( T 1 / 6 M 1 2 / 3 + T 1 / 8 M 1 3 / 4 ) ] . (116) \begin{aligned}
\left\|\overline{\mathcal{L}}_{\epsilon}[(\tau_1(X_1)_s)_s]-\overline{\mathcal{L}}_{\epsilon}[(\tau_2(X_2)_s)_s]\right\|_{L_T^2H_s^{1/2}}
&\leq c\epsilon^{-2}|\log\epsilon|\left(1+M_0^{12}\right)\left[T^{1/6}\left\|X_1-X_2\right\|_{Y_1}^{1/3}\left\|X_1-X_2\right\|_{Y_0}^{2/3}M_1^{1/3}\right.\\
&\qquad\left.+T^{1/8}\left\|X_1-X_2\right\|_{Y_1}^{3/4}\left\|X_1-X_2\right\|_{Y_0}^{1/4}+\left\|X_1-X_2\right\|_{Y_0}\left(T^{1/6}M_1^{2/3}+T^{1/8}M_1^{3/4}\right)\right]. \tag*{(116)}
\end{aligned} L ϵ [( τ 1 ( X 1 ) s ) s ] − L ϵ [( τ 2 ( X 2 ) s ) s ] L T 2 H s 1/2 ≤ c ϵ − 2 ∣ log ϵ ∣ ( 1 + M 0 12 ) [ T 1/6 ∥ X 1 − X 2 ∥ Y 1 1/3 ∥ X 1 − X 2 ∥ Y 0 2/3 M 1 1/3 + T 1/8 ∥ X 1 − X 2 ∥ Y 1 3/4 ∥ X 1 − X 2 ∥ Y 0 1/4 + ∥ X 1 − X 2 ∥ Y 0 ( T 1/6 M 1 2/3 + T 1/8 M 1 3/4 ) ] . (116)
For the remainder term R [ X ] \mathcal{R}[X] R [ X ] , integrating the Lipschitz bound of Lemma 3.3 in time from 0 0 0 to T T T , we have
∥ R [ X 1 ] − R [ X 2 ] ∥ L t 2 H s 1 / 2 ≤ c ε − 3 / 2 ( 1 + M 0 10 / 3 ) T 1 / 6 [ ∥ X 1 − X 2 ∥ Y 0 ( M 1 + T 1 / 3 ( M 0 2 / 3 + 1 ) ) + ∥ X 1 − X 2 ∥ Y 0 2 / 3 ∥ X 1 − X 2 ∥ Y 1 1 / 3 M 1 1 / 3 + ∥ X 1 − X 2 ∥ Y 1 2 / 3 ∥ X 1 − X 2 ∥ Y 0 1 / 3 ] . (117) \left\|\mathcal{R}[X_1]-\mathcal{R}[X_2]\right\|_{L_t^2H_s^{1/2}}
\le c\varepsilon^{-3/2}(1+M_0^{10/3})T^{1/6}\left[\left\|X_1-X_2\right\|_{Y_0}\left(M_1+T^{1/3}(M_0^{2/3}+1)\right)
+\left\|X_1-X_2\right\|_{Y_0}^{2/3}\left\|X_1-X_2\right\|_{Y_1}^{1/3}M_1^{1/3}+\left\|X_1-X_2\right\|_{Y_1}^{2/3}\left\|X_1-X_2\right\|_{Y_0}^{1/3}\right]. \tag*{(117)} ∥ R [ X 1 ] − R [ X 2 ] ∥ L t 2 H s 1/2 ≤ c ε − 3/2 ( 1 + M 0 10/3 ) T 1/6 [ ∥ X 1 − X 2 ∥ Y 0 ( M 1 + T 1/3 ( M 0 2/3 + 1 ) ) + ∥ X 1 − X 2 ∥ Y 0 2/3 ∥ X 1 − X 2 ∥ Y 1 1/3 M 1 1/3 + ∥ X 1 − X 2 ∥ Y 1 2/3 ∥ X 1 − X 2 ∥ Y 0 1/3 ] . (117)
Altogether, using Lemma 3.4, we may estimate
∥ Φ T ( X 1 ) − Φ T ( X 2 ) ∥ Y 0 ∩ Y 1 ≤ ∥ L ε [ ( τ 1 ( X 1 ) s ) s ] − L ε [ ( τ 2 ( X 2 ) s ) s ] ∥ L t 2 H s 1 / 2 + ∥ R [ X 1 ] − R [ X 2 ] ∥ L t 2 H s 1 / 2 ≤ c ( M 0 , M 1 ) ε − 2 ∣ log ε ∣ T 1 / 8 ∥ X 1 − X 2 ∥ Y 0 ∩ Y 1 . (118) \begin{aligned}
\left\|\Phi^T(X_1)-\Phi^T(X_2)\right\|_{Y_0\cap Y_1}
\le\left\|\mathcal{L}_{\varepsilon}\left[(\tau_1(X_1)_s)_s\right]-\mathcal{L}_{\varepsilon}\left[(\tau_2(X_2)_s)_s\right]\right\|_{L_t^2H_s^{1/2}}+\left\|\mathcal{R}[X_1]-\mathcal{R}[X_2]\right\|_{L_t^2H_s^{1/2}} \\
&\le c(M_0,M_1)\varepsilon^{-2}|\log\varepsilon|T^{1/8}\left\|X_1-X_2\right\|_{Y_0\cap Y_1}. \tag*{(118)}
\end{aligned} Φ T ( X 1 ) − Φ T ( X 2 ) Y 0 ∩ Y 1 ≤ ∥ L ε [ ( τ 1 ( X 1 ) s ) s ] − L ε [ ( τ 2 ( X 2 ) s ) s ] ∥ L t 2 H s 1/2 + ∥ R [ X 1 ] − R [ X 2 ] ∥ L t 2 H s 1/2 ≤ c ( M 0 , M 1 ) ε − 2 ∣ log ε ∣ T 1/8 ∥ X 1 − X 2 ∥ Y 0 ∩ Y 1 . (118)
Taking T T T small enough that c ( M 0 , M 1 ) ε − 2 ∣ log ε ∣ T 1 / 8 ≤ 1 2 c(M_0,M_1)\varepsilon^{-2}|\log\varepsilon|T^{1/8}\le\frac{1}{2} c ( M 0 , M 1 ) ε − 2 ∣ log ε ∣ T 1/8 ≤ 2 1 in addition to the above bounds (3.77), we obtain that Φ T \Phi^T Φ T is a contraction on B 0 , 1 B_{0,1} B 0 , 1 and hence admits a unique fixed point. □
Global well-posedness
Given the local well-posedness result of Lemma 3.5, we may proceed to the proof of Theorem (1.2) regarding global well-posedness for the evolution (1.14). Given the form (2.9) of the energy inequality satisfied by the filament and following [20 , 32 , 1 ] , it will be convenient to define the quantity
Z : = X s s s − τ X s . (119) Z:=X_{sss}-\tau X_s. \tag*{(119)} Z := X sss − τ X s . (119)
We define the energy and dissipation quantities E E E and D D D as
E ( t ) : = 1 2 ∥ X s s ∥ L s 2 ( T ) 2 , D ( t ) : = ∥ Z ∥ H s 1 / 2 ( T ) 2 , (120) E(t):=\frac{1}{2}\left\|X_{ss}\right\|_{L^2_s(\mathbb{T})}^2,\qquad D(t):=\left\|Z\right\|_{H_s^{1/2}(\mathbb{T})}^2, \tag*{(120)} E ( t ) := 2 1 ∥ X ss ∥ L s 2 ( T ) 2 , D ( t ) := ∥ Z ∥ H s 1/2 ( T ) 2 , (120)
so that, by (2.9),
E ( t ) ≤ − c ∣ log ε ∣ D ( t ) . (121) E(t)\le-c|\log\varepsilon|D(t). \tag*{(121)} E ( t ) ≤ − c ∣ log ε ∣ D ( t ) . (121)
We aim to extract control on ∥ X s s s ∥ H s 1 / 2 \left\|X_{sss}\right\|_{H_s^{1/2}} ∥ X sss ∥ H s 1/2 , i.e. on the curve alone, in terms of E E E and D D D . We show the following.
Lemma 3.6 (Control on X s s s X_{sss} X sss ). A curve X X X satisfying (1.14) may be controlled by the energy and dissipation quantities as follows:
∥ X s s s ∥ H s 1 / 2 ( T ) ≤ c ( D ( t ) 1 / 2 ( 1 + E ( t ) 1 / 2 ) + E ( t ) 1 / 2 + E ( t ) 4 ) . (122) \left\|X_{sss}\right\|_{H_s^{1/2}(\mathbb{T})}\le c\left(D(t)^{1/2}(1+E(t)^{1/2})+E(t)^{1/2}+E(t)^4\right). \tag*{(122)} ∥ X sss ∥ H s 1/2 ( T ) ≤ c ( D ( t ) 1/2 ( 1 + E ( t ) 1/2 ) + E ( t ) 1/2 + E ( t ) 4 ) . (122)
Proof. We begin by noting that the quantities D D D and E E E control the full H s 1 / 2 H_s^{1/2} H s 1/2 norm of Z Z Z in the following way. Using (3.1), we may calculate that the mean of Z Z Z satisfies
Z 0 : = ∫ T Z d s = ∫ T ( X s s s − τ X s ) d s = ∫ T τ X s d s = − ∫ T ( ( Z − Z 0 ) ⋅ X s + Z 0 ⋅ X s + ∣ X s s ∣ 2 ) X s d s . (123) \begin{aligned}
Z_0:=\int_{\mathbb{T}}Z\,ds=\int_{\mathbb{T}}(X_{sss}-\tau X_s)\,ds=\int_{\mathbb{T}}\tau X_s\,ds \\
&=-\int_{\mathbb{T}}\left((Z-Z_0)\cdot X_s+Z_0\cdot X_s+\left|X_{ss}\right|^2\right)X_s\,ds. \tag*{(123)}
\end{aligned} Z 0 := ∫ T Z d s = ∫ T ( X sss − τ X s ) d s = ∫ T τ X s d s = − ∫ T ( ( Z − Z 0 ) ⋅ X s + Z 0 ⋅ X s + ∣ X ss ∣ 2 ) X s d s . (123)
From this we obtain
( ∫ T ( I + X s ⊗ X s ) d s ) Z 0 = − ∫ T ( ( Z − Z 0 ) ⋅ X s + ∣ X s s ∣ 2 ) X s d s . (124) \left(\int_{\mathbb{T}}(I+X_s\otimes X_s)\,ds\right)Z_0=-\int_{\mathbb{T}}\left((Z-Z_0)\cdot X_s+\left|X_{ss}\right|^2\right)X_s\,ds. \tag*{(124)} ( ∫ T ( I + X s ⊗ X s ) d s ) Z 0 = − ∫ T ( ( Z − Z 0 ) ⋅ X s + ∣ X ss ∣ 2 ) X s d s . (124)
Using that the left hand side integrand is bounded below pointwise and that, on T \mathbb{T} T , we have
∥ Z − Z 0 ∥ L s 2 ( T ) 2 = ∑ ∣ k ∣ = 1 ∞ F [ Z ] F [ Z ] ‾ ≤ ∑ ∣ k ∣ = 1 ∞ ∣ k ∣ F [ Z ] F [ Z ] ‾ ≤ ∥ Z ∥ H s 1 / 2 ( T ) 2 , \left\|Z-Z_0\right\|_{L_s^2(\mathbb{T})}^2=\sum_{|k|=1}^{\infty}\mathcal{F}[Z]\overline{\mathcal{F}[Z]}\le\sum_{|k|=1}^{\infty}|k|\mathcal{F}[Z]\overline{\mathcal{F}[Z]}\le\left\|Z\right\|_{H_s^{1/2}(\mathbb{T})}^2, ∥ Z − Z 0 ∥ L s 2 ( T ) 2 = ∣ k ∣ = 1 ∑ ∞ F [ Z ] F [ Z ] ≤ ∣ k ∣ = 1 ∑ ∞ ∣ k ∣ F [ Z ] F [ Z ] ≤ ∥ Z ∥ H s 1/2 ( T ) 2 ,
we may then estimate
∣ Z 0 ∣ ≤ c ( ∥ Z ∥ H ˙ s 1 / 2 + ∥ X s s ∥ L s 2 2 ) ≤ c ( D ( t ) 1 / 2 + E ( t ) ) . (125) |Z_0| \le c\left(\|Z\|_{\dot H_s^{1/2}}+\|X_{ss}\|_{L_s^2}^2\right)\le c(D(t)^{1/2}+E(t)). \tag*{(125)} ∣ Z 0 ∣ ≤ c ( ∥ Z ∥ H ˙ s 1/2 + ∥ X ss ∥ L s 2 2 ) ≤ c ( D ( t ) 1/2 + E ( t )) . (125)
We thus have
∥ Z ∥ H s 1 / 2 ( T ) ≤ c ( D ( t ) 1 / 2 + E ( t ) ) . (126) \|Z\|_{H_s^{1/2}(\mathbb{T})}\le c(D(t)^{1/2}+E(t)). \tag*{(126)} ∥ Z ∥ H s 1/2 ( T ) ≤ c ( D ( t ) 1/2 + E ( t )) . (126)
We now proceed to estimate ∥ X s s s ∥ H s 1 / 2 \|X_{sss}\|_{H_s^{1/2}} ∥ X sss ∥ H s 1/2 . First note that we may write the normal components of X s s s X_{sss} X sss as
P X s ⊥ X s s s = P X s ⊥ Z = Z − P X s Z . (127) \mathbb{P}_{X_s}^{\perp}X_{sss}=\mathbb{P}_{X_s}^{\perp}Z=Z-\mathbb{P}_{X_s}Z. \tag*{(127)} P X s ⊥ X sss = P X s ⊥ Z = Z − P X s Z . (127)
Using (126) , we may estimate
∥ P X s Z ∥ H ˙ s 1 / 2 ≤ ∥ Z ∥ H s 1 / 2 ∥ X s ⊗ X s ∥ H s 1 ≤ c ( D ( t ) 1 / 2 + E ( t ) ) ( 1 + E ( t ) 1 / 2 ) , (128) \|\mathbb{P}_{X_s}Z\|_{\dot H_s^{1/2}}\le\|Z\|_{H_s^{1/2}}\|X_s\otimes X_s\|_{H_s^1}\le c(D(t)^{1/2}+E(t))(1+E(t)^{1/2}), \tag*{(128)} ∥ P X s Z ∥ H ˙ s 1/2 ≤ ∥ Z ∥ H s 1/2 ∥ X s ⊗ X s ∥ H s 1 ≤ c ( D ( t ) 1/2 + E ( t )) ( 1 + E ( t ) 1/2 ) , (128)
so that
∥ P X s ⊥ X s s s ∥ H s 1 / 2 ≤ c ( D ( t ) 1 / 2 + E ( t ) ) ( 1 + E ( t ) 1 / 2 ) . (129) \|\mathbb{P}_{X_s}^{\perp}X_{sss}\|_{H_s^{1/2}}\le c(D(t)^{1/2}+E(t))(1+E(t)^{1/2}). \tag*{(129)} ∥ P X s ⊥ X sss ∥ H s 1/2 ≤ c ( D ( t ) 1/2 + E ( t )) ( 1 + E ( t ) 1/2 ) . (129)
Next, using (36) , we may write the tangential component of X s s s X_{sss} X sss as
P X s X s s s = ( X s ⋅ X s s s ) X s = − ∣ X s s ∣ 2 X s . (130) \mathbb{P}_{X_s}X_{sss}=(X_s\cdot X_{sss})X_s=-|X_{ss}|^2X_s. \tag*{(130)} P X s X sss = ( X s ⋅ X sss ) X s = − ∣ X ss ∣ 2 X s . (130)
To estimate this in H ˙ s 1 / 2 \dot H_s^{1/2} H ˙ s 1/2 , we will make use of the interpolation inequalities (37) on T \mathbb{T} T to obtain
∥ ∣ X s s ∣ 2 X s ∥ H ˙ s 1 / 2 ≤ c ∥ ( ∣ X s s ∣ 2 X s ) s ∥ L s 2 1 / 2 ∥ ∣ X s s ∣ 2 X s ∥ L s 2 1 / 2 ≤ c ∥ ( ∣ X s s ∣ 2 X s ) s ∥ L s 2 1 / 2 ∥ X s s ∥ L s 5 / 6 ∥ X s s s ∥ H ˙ s 1 / 2 1 / 6 . (131) \big\||X_{ss}|^2X_s\big\|_{\dot H_s^{1/2}}\le c\big\|(|X_{ss}|^2X_s)_s\big\|_{L_s^2}^{1/2}\big\||X_{ss}|^2X_s\big\|_{L_s^2}^{1/2}
\le c\big\|(|X_{ss}|^2X_s)_s\big\|_{L_s^2}^{1/2}\|X_{ss}\|_{L_s^{5/6}}\|X_{sss}\|_{\dot H_s^{1/2}}^{1/6}. \tag*{(131)} ∣ X ss ∣ 2 X s H ˙ s 1/2 ≤ c ( ∣ X ss ∣ 2 X s ) s L s 2 1/2 ∣ X ss ∣ 2 X s L s 2 1/2 ≤ c ( ∣ X ss ∣ 2 X s ) s L s 2 1/2 ∥ X ss ∥ L s 5/6 ∥ X sss ∥ H ˙ s 1/2 1/6 . (131)
Expanding
( ∣ X s s ∣ 2 X s ) s = 2 ( X s s ⋅ X s s s ) X s + ∣ X s s ∣ 2 X s s , (132) (|X_{ss}|^2X_s)_s=2(X_{ss}\cdot X_{sss})X_s+|X_{ss}|^2X_{ss}, \tag*{(132)} ( ∣ X ss ∣ 2 X s ) s = 2 ( X ss ⋅ X sss ) X s + ∣ X ss ∣ 2 X ss , (132)
we may further estimate
∥ ( ∣ X s s ∣ 2 X s ) s ∥ L s 2 ≤ 2 ∥ X s s ∥ L s ∞ ∥ X s s s ∥ L s 2 + ∥ X s s ∥ L s ∞ 2 ∥ X s s ∥ L s 2 ≤ ∥ X s s s ∥ H s 1 / 2 ∥ X s s ∥ L s 2 + ∥ X s s s ∥ H s 1 / 2 2 / 3 ∥ X s s ∥ L s 2 7 / 3 . (133) \big\|(|X_{ss}|^2X_s)_s\big\|_{L_s^2}\le2\|X_{ss}\|_{L_s^\infty}\|X_{sss}\|_{L_s^2}+\|X_{ss}\|_{L_s^\infty}^2\|X_{ss}\|_{L_s^2}
\le\|X_{sss}\|_{H_s^{1/2}}\|X_{ss}\|_{L_s^2}+\|X_{sss}\|_{H_s^{1/2}}^{2/3}\|X_{ss}\|_{L_s^2}^{7/3}. \tag*{(133)} ( ∣ X ss ∣ 2 X s ) s L s 2 ≤ 2∥ X ss ∥ L s ∞ ∥ X sss ∥ L s 2 + ∥ X ss ∥ L s ∞ 2 ∥ X ss ∥ L s 2 ≤ ∥ X sss ∥ H s 1/2 ∥ X ss ∥ L s 2 + ∥ X sss ∥ H s 1/2 2/3 ∥ X ss ∥ L s 2 7/3 . (133)
In total, the right hand side of (130) satisfies
∥ ∣ X s s ∣ 2 X s ∥ H ˙ s 1 / 2 ≤ c ( ∥ X s s s ∥ H s 1 / 2 2 / 3 ∥ X s s ∥ L s 2 4 / 3 + ∥ X s s s ∥ H s 1 / 2 1 / 2 ∥ X s s ∥ L s 2 2 ) . (134) \big\||X_{ss}|^2X_s\big\|_{\dot H_s^{1/2}}\le c\left(\|X_{sss}\|_{H_s^{1/2}}^{2/3}\|X_{ss}\|_{L_s^2}^{4/3}+\|X_{sss}\|_{H_s^{1/2}}^{1/2}\|X_{ss}\|_{L_s^2}^2\right). \tag*{(134)} ∣ X ss ∣ 2 X s H ˙ s 1/2 ≤ c ( ∥ X sss ∥ H s 1/2 2/3 ∥ X ss ∥ L s 2 4/3 + ∥ X sss ∥ H s 1/2 1/2 ∥ X ss ∥ L s 2 2 ) . (134)
Using (134) and (129) , we may thus bound
∥ X s s s ∥ H s 1 / 2 ≤ ∥ P X s ⊥ X s s s ∥ H s 1 / 2 + ∥ P X s X s s s ∥ H s 1 / 2 ≤ c ( D ( t ) 1 / 2 + E ( t ) ) ( 1 + E ( t ) 1 / 2 ) + c ( ∥ X s s s ∥ H s 1 / 2 2 / 3 E ( t ) 4 / 3 + ∥ X s s s ∥ H s 1 / 2 1 / 2 E ( t ) 2 ) ≤ c ( D ( t ) 1 / 2 ( 1 + E ( t ) 1 / 2 ) + E ( t ) + E ( t ) 4 ) , (135) \begin{aligned}
\|X_{sss}\|_{H_s^{1/2}}&\le\|\mathbb{P}_{X_s}^{\perp}X_{sss}\|_{H_s^{1/2}}+\|\mathbb{P}_{X_s}X_{sss}\|_{H_s^{1/2}}\\
&\le c(D(t)^{1/2}+E(t))(1+E(t)^{1/2})+c\left(\|X_{sss}\|_{H_s^{1/2}}^{2/3}E(t)^{4/3}+\|X_{sss}\|_{H_s^{1/2}}^{1/2}E(t)^2\right)\\
&\le c\left(D(t)^{1/2}(1+E(t)^{1/2})+E(t)+E(t)^4\right), \tag*{(135)}
\end{aligned} ∥ X sss ∥ H s 1/2 ≤ ∥ P X s ⊥ X sss ∥ H s 1/2 + ∥ P X s X sss ∥ H s 1/2 ≤ c ( D ( t ) 1/2 + E ( t )) ( 1 + E ( t ) 1/2 ) + c ( ∥ X sss ∥ H s 1/2 2/3 E ( t ) 4/3 + ∥ X sss ∥ H s 1/2 1/2 E ( t ) 2 ) ≤ c ( D ( t ) 1/2 ( 1 + E ( t ) 1/2 ) + E ( t ) + E ( t ) 4 ) , (135)
where we have used Young’s inequality to split the quantities in the second line. Using (37) , we may further estimate
∥ X s s s ∥ L s 2 ≤ c ( ∥ X s s s ∥ H ˙ s 1 / 2 + ∥ X s s ∥ L s 2 ) ≤ c ( D ( t ) 1 / 2 ( 1 + E ( t ) 1 / 2 ) + E ( t ) 1 / 2 + E ( t ) 4 ) , (136) \|X_{sss}\|_{L_s^2}\le c(\|X_{sss}\|_{\dot H_s^{1/2}}+\|X_{ss}\|_{L_s^2})\le c(D(t)^{1/2}(1+E(t)^{1/2})+E(t)^{1/2}+E(t)^4), \tag*{(136)} ∥ X sss ∥ L s 2 ≤ c ( ∥ X sss ∥ H ˙ s 1/2 + ∥ X ss ∥ L s 2 ) ≤ c ( D ( t ) 1/2 ( 1 + E ( t ) 1/2 ) + E ( t ) 1/2 + E ( t ) 4 ) , (136)
from which we obtain (122) . □
Given Lemma 3.6 and the continuation criterion (102) , the local solution of Lemma 3.5 may be extended globally in time to yield Theorem 1.2.
Convergence to resistive force theory dynamics
We may next turn to the limiting dynamics of (1.14) as ε → 0 \varepsilon\to0 ε → 0 .
Setup
Given an initial curve4 X i n ∈ H 4 ( T ) \mathbf{X}^{\mathrm{in}} \in H^4(\mathbb{T}) X in ∈ H 4 ( T ) , let Y ∈ C ( [ 0 , T ] ; H 4 ( T ) ) ∩ L t , l o c 2 H s 6 \mathbf{Y} \in C([0,T];H^4(\mathbb{T})) \cap L^2_{t,\mathrm{loc}}H^6_s Y ∈ C ([ 0 , T ] ; H 4 ( T )) ∩ L t , loc 2 H s 6 be the resulting unique global solution to the resistive force theory evolution (1.18), guaranteed by [1 Theorem 1.1] . In particular, Y \mathbf{Y} Y satisfies
∂ Y ∂ t = − ∣ log ε ∣ 4 π ( I + Y s ⊗ Y s ) ( Y s s s − τ Y Y s ) s = − ∣ log ε ∣ 2 π P Y s ( Y s s s − τ Y Y s ) s − ∣ log ε ∣ 4 π P Y s ⊥ ( Y s s s − τ Y Y s ) s , ∣ Y s ∣ = 1 , Y ∣ t = 0 = X i n . (137) \begin{aligned}
\frac{\partial\mathbf{Y}}{\partial t}
= -\frac{|\log\varepsilon|}{4\pi}(\mathbf{I}+\mathbf{Y}_s\otimes\mathbf{Y}_s)(\mathbf{Y}_{sss}-\tau_{\mathbf{Y}}\mathbf{Y}_s)_s \\
&= -\frac{|\log\varepsilon|}{2\pi}\mathbb{P}_{\mathbf{Y}_s}(\mathbf{Y}_{sss}-\tau_{\mathbf{Y}}\mathbf{Y}_s)_s
-\frac{|\log\varepsilon|}{4\pi}\mathbb{P}_{\mathbf{Y}_s}^{\perp}(\mathbf{Y}_{sss}-\tau_{\mathbf{Y}}\mathbf{Y}_s)_s,
\qquad|\mathbf{Y}_s|=1,\qquad\left.\mathbf{Y}\right|_{t=0}=\mathbf{X}^{\mathrm{in}}. \tag*{(137)}
\end{aligned} ∂ t ∂ Y = − 4 π ∣ log ε ∣ ( I + Y s ⊗ Y s ) ( Y sss − τ Y Y s ) s = − 2 π ∣ log ε ∣ P Y s ( Y sss − τ Y Y s ) s − 4 π ∣ log ε ∣ P Y s ⊥ ( Y sss − τ Y Y s ) s , ∣ Y s ∣ = 1 , Y ∣ t = 0 = X in . (137)
Letting X ( s , t ) ∈ C ( [ 0 , T ] ; H s 2 ( T ) ) ∩ L t , l o c 2 H s 7 / 2 \mathbf{X}(s,t) \in C([0,T];H^2_s(\mathbb{T}))\cap L^2_{t,\mathrm{loc}}H^{7/2}_s X ( s , t ) ∈ C ([ 0 , T ] ; H s 2 ( T )) ∩ L t , loc 2 H s 7/2 denote the global solution to (1.14) starting from X i n \mathbf{X}^{\mathrm{in}} X in , we consider the difference W = X − Y \mathbf{W}=\mathbf{X}-\mathbf{Y} W = X − Y , whose evolution then satisfies
∂ W ∂ t = − L ε ( X ) [ ( X s s s − τ X X s ) s ] + ∣ log ε ∣ 4 π ( I + Y s ⊗ Y s ) ( Y s s s − τ Y Y s ) s , W ∣ t = 0 = 0. (138) \frac{\partial\mathbf{W}}{\partial t}
=-\mathcal{L}_{\varepsilon}(\mathbf{X})[(\mathbf{X}_{sss}-\tau_{\mathbf{X}}\mathbf{X}_s)_s]
+\frac{|\log\varepsilon|}{4\pi}(\mathbf{I}+\mathbf{Y}_s\otimes\mathbf{Y}_s)(\mathbf{Y}_{sss}-\tau_{\mathbf{Y}}\mathbf{Y}_s)_s,
\qquad\left.\mathbf{W}\right|_{t=0}=0. \tag*{(138)} ∂ t ∂ W = − L ε ( X ) [( X sss − τ X X s ) s ] + 4 π ∣ log ε ∣ ( I + Y s ⊗ Y s ) ( Y sss − τ Y Y s ) s , W ∣ t = 0 = 0. (138)
Now, noting that both the map L ε ( X ) \mathcal{L}_{\varepsilon}(\mathbf{X}) L ε ( X ) and the resistive force theory operator L ε , R F T ( Y ) = ∣ log ε ∣ 4 π ( I + Y s ⊗ Y s ) \mathcal{L}_{\varepsilon,\mathrm{RFT}}(\mathbf{Y})=\frac{|\log\varepsilon|}{4\pi}(\mathbf{I}+\mathbf{Y}_s\otimes\mathbf{Y}_s) L ε , RFT ( Y ) = 4 π ∣ l o g ε ∣ ( I + Y s ⊗ Y s ) blow up logarithmically as ε → 0 \varepsilon\to0 ε → 0 , to obtain a convergence result for their dynamics, we consider (4.2) under the following rescaling of time:
t = ∣ log ε ∣ t ‾ , ∂ t = ∣ log ε ∣ − 1 ∂ t ‾ . (139) t=|\log\varepsilon|\underline{t},\qquad\partial_t=|\log\varepsilon|^{-1}\partial_{\underline{t}}. \tag*{(139)} t = ∣ log ε ∣ t , ∂ t = ∣ log ε ∣ − 1 ∂ t . (139)
The left hand side of the evolution (4.2) may then be written
∂ W ∂ t = ∣ log ε ∣ ∂ W ∂ t ‾ . (140) \frac{\partial\mathbf{W}}{\partial t}=|\log\varepsilon|\frac{\partial\mathbf{W}}{\partial\underline{t}}. \tag*{(140)} ∂ t ∂ W = ∣ log ε ∣ ∂ t ∂ W . (140)
We may split the right hand side of (4.2) as
− P X s T ε m t P X s ( X s s s − τ X X s ) s + P Y s ∣ log ε ∣ 2 π P Y s ( Y s s s − τ Y Y s ) s -\mathbb{P}_{\mathbf{X}_s}T_{\varepsilon}^{m_t}\mathbb{P}_{\mathbf{X}_s}(\mathbf{X}_{sss}-\tau_{\mathbf{X}}\mathbf{X}_s)_s
+\mathbb{P}_{\mathbf{Y}_s}\frac{|\log\varepsilon|}{2\pi}\mathbb{P}_{\mathbf{Y}_s}(\mathbf{Y}_{sss}-\tau_{\mathbf{Y}}\mathbf{Y}_s)_s − P X s T ε m t P X s ( X sss − τ X X s ) s + P Y s 2 π ∣ log ε ∣ P Y s ( Y sss − τ Y Y s ) s
− P X s ⊥ T ε m n P X s ⊥ ( X s s s − τ X X s ) s + P Y s ⊥ ∣ log ε ∣ 4 π P Y s ⊥ ( Y s s s − τ Y Y s ) s = − P X s T ε m t P X s ( W s s s − ( τ X X s − τ Y Y s ) ) s \begin{aligned}
-\mathbb{P}_{\mathbf{X}_s}^{\perp}T_{\varepsilon}^{m_n}\mathbb{P}_{\mathbf{X}_s}^{\perp}(\mathbf{X}_{sss}-\tau_{\mathbf{X}}\mathbf{X}_s)_s
+\mathbb{P}_{\mathbf{Y}_s}^{\perp}\frac{|\log\varepsilon|}{4\pi}\mathbb{P}_{\mathbf{Y}_s}^{\perp}(\mathbf{Y}_{sss}-\tau_{\mathbf{Y}}\mathbf{Y}_s)_s \\
&=-\mathbb{P}_{\mathbf{X}_s}T_{\varepsilon}^{m_t}\mathbb{P}_{\mathbf{X}_s}\bigl(\mathbf{W}_{sss}-(\tau_{\mathbf{X}}\mathbf{X}_s-\tau_{\mathbf{Y}}\mathbf{Y}_s)\bigr)_s
\end{aligned} − P X s ⊥ T ε m n P X s ⊥ ( X sss − τ X X s ) s + P Y s ⊥ 4 π ∣ log ε ∣ P Y s ⊥ ( Y sss − τ Y Y s ) s = − P X s T ε m t P X s ( W sss − ( τ X X s − τ Y Y s ) ) s
− ( P X s T ε m t P X s − P Y s ∣ log ε ∣ 2 π P Y s ) ( Y s s s − τ Y Y s ) s -\left(\mathbb{P}_{\mathbf{X}_s}T_{\varepsilon}^{m_t}\mathbb{P}_{\mathbf{X}_s}
-\mathbb{P}_{\mathbf{Y}_s}\frac{|\log\varepsilon|}{2\pi}\mathbb{P}_{\mathbf{Y}_s}\right)(\mathbf{Y}_{sss}-\tau_{\mathbf{Y}}\mathbf{Y}_s)_s − ( P X s T ε m t P X s − P Y s 2 π ∣ log ε ∣ P Y s ) ( Y sss − τ Y Y s ) s
− P X s ⊥ T ε m n P X s ⊥ ( W s s s − ( τ X X s − τ Y Y s ) ) s -\mathbb{P}_{\mathbf{X}_s}^{\perp}T_{\varepsilon}^{m_n}\mathbb{P}_{\mathbf{X}_s}^{\perp}\bigl(\mathbf{W}_{sss}-(\tau_{\mathbf{X}}\mathbf{X}_s-\tau_{\mathbf{Y}}\mathbf{Y}_s)\bigr)_s − P X s ⊥ T ε m n P X s ⊥ ( W sss − ( τ X X s − τ Y Y s ) ) s
− ( P X s ⊥ T ε m n P X s ⊥ − P Y s ⊥ ∣ log ε ∣ 4 π P Y s ⊥ ) ( Y s s s − τ Y Y s ) s , (141) -\left(\mathbb{P}_{\mathbf{X}_s}^{\perp}T_{\varepsilon}^{m_n}\mathbb{P}_{\mathbf{X}_s}^{\perp}
-\mathbb{P}_{\mathbf{Y}_s}^{\perp}\frac{|\log\varepsilon|}{4\pi}\mathbb{P}_{\mathbf{Y}_s}^{\perp}\right)(\mathbf{Y}_{sss}-\tau_{\mathbf{Y}}\mathbf{Y}_s)_s, \tag*{(141)} − ( P X s ⊥ T ε m n P X s ⊥ − P Y s ⊥ 4 π ∣ log ε ∣ P Y s ⊥ ) ( Y sss − τ Y Y s ) s , (141)
so that, upon multiplying (4.2) by W s s s s W_{ssss} W ssss and integrating over T \mathbb{T} T , we obtain the energy identity
∣ log ϵ ∣ ∂ ∂ t ∫ T ∣ W s s ∣ 2 d s + ∫ T ( P X s W s s s s ) ⋅ T m ϵ t ( P X s W s s s s ) d s + ∫ T ( P X s ⊥ W s s s s ) ⋅ T m ϵ n ( P X s ⊥ W s s s s ) d s = A ∥ + A ⊥ , |\log\epsilon| \frac{\partial}{\partial t}\int_{\mathbb{T}} |W_{ss}|^2\,ds+\int_{\mathbb{T}}(\mathbb{P}_{X_s}W_{ssss})\cdot T_{m_\epsilon}^{\mathrm{t}}(\mathbb{P}_{X_s}W_{ssss})\,ds
+\int_{\mathbb{T}}(\mathbb{P}_{X_s}^{\perp}W_{ssss})\cdot T_{m_\epsilon}^{\mathrm{n}}(\mathbb{P}_{X_s}^{\perp}W_{ssss})\,ds=A^{\parallel}+A^{\perp}, ∣ log ϵ ∣ ∂ t ∂ ∫ T ∣ W ss ∣ 2 d s + ∫ T ( P X s W ssss ) ⋅ T m ϵ t ( P X s W ssss ) d s + ∫ T ( P X s ⊥ W ssss ) ⋅ T m ϵ n ( P X s ⊥ W ssss ) d s = A ∥ + A ⊥ ,
A ∥ = ∫ T ( P X s W s s s s ) ⋅ T m ϵ t P X s ( τ W X s + τ Y W s ) d s − ∫ T W s s s s ⋅ ( P X s T m ϵ t P X s − P Y s ∣ log ϵ ∣ 2 π P Y s ) ( Y s s s − τ Y Y s ) d s A^{\parallel}=\int_{\mathbb{T}}(\mathbb{P}_{X_s}W_{ssss})\cdot T_{m_\epsilon}^{\mathrm{t}}\mathbb{P}_{X_s}(\tau_WX_s+\tau_YW_s)\,ds
-\int_{\mathbb{T}}W_{ssss}\cdot\left(\mathbb{P}_{X_s}T_{m_\epsilon}^{\mathrm{t}}\mathbb{P}_{X_s}-\mathbb{P}_{Y_s}\frac{|\log\epsilon|}{2\pi}\mathbb{P}_{Y_s}\right)(Y_{sss}-\tau_YY_s)\,ds A ∥ = ∫ T ( P X s W ssss ) ⋅ T m ϵ t P X s ( τ W X s + τ Y W s ) d s − ∫ T W ssss ⋅ ( P X s T m ϵ t P X s − P Y s 2 π ∣ log ϵ ∣ P Y s ) ( Y sss − τ Y Y s ) d s
A ⊥ = ∫ T ( P X s ⊥ W s s s s ) ⋅ T m ϵ n P X s ⊥ ( τ W X s + τ Y W s ) d s − ∫ T W s s s s ⋅ ( P X s ⊥ T m ϵ n P X s ⊥ − P Y s ⊥ ∣ log ϵ ∣ 4 π P Y s ⊥ ) ( Y s s s − τ Y Y s ) d s . (142) A^{\perp}=\int_{\mathbb{T}}(\mathbb{P}_{X_s}^{\perp}W_{ssss})\cdot T_{m_\epsilon}^{\mathrm{n}}\mathbb{P}_{X_s}^{\perp}(\tau_WX_s+\tau_YW_s)\,ds
-\int_{\mathbb{T}}W_{ssss}\cdot\left(\mathbb{P}_{X_s}^{\perp}T_{m_\epsilon}^{\mathrm{n}}\mathbb{P}_{X_s}^{\perp}-\mathbb{P}_{Y_s}^{\perp}\frac{|\log\epsilon|}{4\pi}\mathbb{P}_{Y_s}^{\perp}\right)(Y_{sss}-\tau_YY_s)\,ds. \tag*{(142)} A ⊥ = ∫ T ( P X s ⊥ W ssss ) ⋅ T m ϵ n P X s ⊥ ( τ W X s + τ Y W s ) d s − ∫ T W ssss ⋅ ( P X s ⊥ T m ϵ n P X s ⊥ − P Y s ⊥ 4 π ∣ log ϵ ∣ P Y s ⊥ ) ( Y sss − τ Y Y s ) d s . (142)
Our goal will be to bound A ∥ A^{\parallel} A ∥ and A ⊥ A^{\perp} A ⊥ in terms of the energy and dissipation quantities appearing on the left hand side of (142) . Since convergence is expected only at a very slow logarithmic rate as ϵ → 0 \epsilon\to0 ϵ → 0 (see Theorem 1.3 and introductory discussion), these estimates are a bit delicate. In particular, we see from the local well-posedness theory that because of the difference in ϵ \epsilon ϵ -scaling for L ϵ \mathcal{L}_{\epsilon} L ϵ at high versus low wavenumbers, the ‘coarse’ upper and lower bounds differ by a factor of ϵ − 1 \epsilon^{-1} ϵ − 1 . This will be far too lossy for convergence, so whenever possible, we will aim to avoid actually evaluating the operators T m ϵ t T_{m_\epsilon}^{\mathrm{t}} T m ϵ t and T m ϵ n T_{m_\epsilon}^{\mathrm{n}} T m ϵ n . Instead, we will rely on the fact that both T m ϵ t T_{m_\epsilon}^{\mathrm{t}} T m ϵ t and T m ϵ n T_{m_\epsilon}^{\mathrm{n}} T m ϵ n are positive operators, and, in particular, we may apply ( T m ϵ t ) 1 / 2 (T_{m_\epsilon}^{\mathrm{t}})^{1/2} ( T m ϵ t ) 1/2 and ( T m ϵ n ) 1 / 2 (T_{m_\epsilon}^{\mathrm{n}})^{1/2} ( T m ϵ n ) 1/2 . Furthermore, since m ϵ n ( k ) ≲ m ϵ t ( k ) ≲ m ϵ n ( k ) m_\epsilon^{\mathrm{n}}(k)\lesssim m_\epsilon^{\mathrm{t}}(k)\lesssim m_\epsilon^{\mathrm{n}}(k) m ϵ n ( k ) ≲ m ϵ t ( k ) ≲ m ϵ n ( k ) independent of ϵ \epsilon ϵ , by Lemma 2.1 , we may interchange T m ϵ t T_{m_\epsilon}^{\mathrm{t}} T m ϵ t and T m ϵ n T_{m_\epsilon}^{\mathrm{n}} T m ϵ n in upper and lower bounds as needed.
To proceed, we will need a few additional ingredients. The first is a bound for powers of the inverse operators T m ϵ t , − 1 = T 1 / m ϵ t T_{m_\epsilon}^{\mathrm{t},-1}=T_{1/m_\epsilon^{\mathrm{t}}} T m ϵ t , − 1 = T 1/ m ϵ t , T m ϵ n , − 1 = T 1 / m ϵ n T_{m_\epsilon}^{\mathrm{n},-1}=T_{1/m_\epsilon^{\mathrm{n}}} T m ϵ n , − 1 = T 1/ m ϵ n .
Lemma 4.1 (Inverse multiplier operators). For θ ≥ 0 \theta\geq0 θ ≥ 0 , the inverse operators
T m ϵ t , − θ ( ⋅ ) = F − 1 [ m ϵ t ( k ) − θ F [ ⋅ ] ] , T m ϵ n , − θ ( ⋅ ) = F − 1 [ m ϵ n ( k ) − θ F [ ⋅ ] ] (143) T_{m_\epsilon}^{\mathrm{t},-\theta}(\cdot)=\mathcal{F}^{-1}\left[m_\epsilon^{\mathrm{t}}(k)^{-\theta}\mathcal{F}[\cdot]\right],\qquad T_{m_\epsilon}^{\mathrm{n},-\theta}(\cdot)=\mathcal{F}^{-1}\left[m_\epsilon^{\mathrm{n}}(k)^{-\theta}\mathcal{F}[\cdot]\right] \tag*{(143)} T m ϵ t , − θ ( ⋅ ) = F − 1 [ m ϵ t ( k ) − θ F [ ⋅ ] ] , T m ϵ n , − θ ( ⋅ ) = F − 1 [ m ϵ n ( k ) − θ F [ ⋅ ] ] (143) satisfy, for j = t , n j=\mathrm{t},\mathrm{n} j = t , n ,
∥ T m ϵ j , − θ f ∥ H ℓ ≤ c ∣ log ϵ ∣ − θ ∥ f ∥ H ℓ + θ . (144) \left\|T_{m_\epsilon}^{j,-\theta}f\right\|_{H^\ell}\leq c|\log\epsilon|^{-\theta}\|f\|_{H^{\ell+\theta}}. \tag*{(144)} T m ϵ j , − θ f H ℓ ≤ c ∣ log ϵ ∣ − θ ∥ f ∥ H ℓ + θ . (144)
The proof follows directly from the multiplier bounds of Lemma 2.1 . We will use Lemma 4.1 along the way in estimating A ∥ A^{\parallel} A ∥ and A ⊥ A^{\perp} A ⊥ .
Operator difference
We will require bounds for the terms in A ∥ A^{\parallel} A ∥ and A ⊥ A^{\perp} A ⊥ involving the difference between the multiplier operators T m ϵ t T_{m_\epsilon}^{\mathrm{t}} T m ϵ t , T m ϵ n T_{m_\epsilon}^{\mathrm{n}} T m ϵ n and their resistive force theory counterparts, the constants ∣ log ϵ ∣ 2 π \frac{|\log\epsilon|}{2\pi} 2 π ∣ l o g ϵ ∣ , ∣ log ϵ ∣ 4 π \frac{|\log\epsilon|}{4\pi} 4 π ∣ l o g ϵ ∣ . Here we exploit Lemma 2.3 : the difference between the multipliers and the resistive force theory constants is bounded independent of ϵ \epsilon ϵ at low wavenumber.
Lemma 4.2 (Multiplier operator and RFT difference). Given φ ∈ H − 1 / 2 ( T ) \varphi\in H^{-1/2}(\mathbb{T}) φ ∈ H − 1/2 ( T ) and f ∈ L 2 ( T ) f\in L^2(\mathbb{T}) f ∈ L 2 ( T ) , we may estimate
∣ ∫ T φ ⋅ ( T m ϵ t − ∣ log ϵ ∣ 2 π ) [ f ] d s ∣ ≤ c ∥ ( T m ϵ t ) 1 / 2 φ ∥ L 2 ∥ f ∥ L 2 \left|\int_{\mathbb{T}}\varphi\cdot\left(T_{m_\epsilon}^{\mathrm{t}}-\frac{|\log\epsilon|}{2\pi}\right)[f]\,ds\right|\leq c\left\|(T_{m_\epsilon}^{\mathrm{t}})^{1/2}\varphi\right\|_{L^2}\|f\|_{L^2} ∫ T φ ⋅ ( T m ϵ t − 2 π ∣ log ϵ ∣ ) [ f ] d s ≤ c ( T m ϵ t ) 1/2 φ L 2 ∥ f ∥ L 2 ∣ ∫ T φ ⋅ ( T m ϵ n − ∣ log ϵ ∣ 4 π ) [ f ] d s ∣ ≤ c ∥ ( T m ϵ n ) 1 / 2 φ ∥ L 2 ∥ f ∥ L 2 . (145) \left|\int_{\mathbb{T}}\varphi\cdot\left(T_{m_\epsilon}^{\mathrm{n}}-\frac{|\log\epsilon|}{4\pi}\right)[f]\,ds\right|\leq c\left\|(T_{m_\epsilon}^{\mathrm{n}})^{1/2}\varphi\right\|_{L^2}\|f\|_{L^2}. \tag*{(145)} ∫ T φ ⋅ ( T m ϵ n − 4 π ∣ log ϵ ∣ ) [ f ] d s ≤ c ( T m ϵ n ) 1/2 φ L 2 ∥ f ∥ L 2 . (145)
As a corollary, given a bit more regularity on f f f , we may obtain the following bound:
Corollary 4.3. Given φ ∈ H − 1 / 2 ( T ) \varphi\in H^{-1/2}(\mathbb{T}) φ ∈ H − 1/2 ( T ) and f ∈ H 1 / 2 ( T ) f \in H^{1/2}(\mathbb{T}) f ∈ H 1/2 ( T ) , we have
∣ ∫ T φ ⋅ ( P X s T m ϵ n P X s − P Y s ∣ log ϵ ∣ 2 π P Y s ) [ f ] d s ∣ ≤ c ∥ T m ϵ 1 / 2 ( P X s φ ) ∥ L 2 ∥ P X s f ∥ L 2 \left|\int_{\mathbb{T}}\varphi\cdot\left(\mathbb{P}_{X_s}T_{m_\epsilon}^{n}\mathbb{P}_{X_s}-\mathbb{P}_{Y_s}\frac{|\log\epsilon|}{2\pi}\mathbb{P}_{Y_s}\right)[f]\,ds\right|\le c\left\|T_{m_\epsilon}^{1/2}(\mathbb{P}_{X_s}\varphi)\right\|_{L^2}\left\|\mathbb{P}_{X_s}f\right\|_{L^2} ∫ T φ ⋅ ( P X s T m ϵ n P X s − P Y s 2 π ∣ log ϵ ∣ P Y s ) [ f ] d s ≤ c T m ϵ 1/2 ( P X s φ ) L 2 ∥ P X s f ∥ L 2 + c ∣ log ϵ ∣ 1 / 2 ∥ W s ∥ H 1 ( ∥ X s ∥ H 1 2 + ∥ Y s ∥ H 1 2 ) ∥ T m ϵ 1 / 2 φ ∥ L 2 ∥ f ∥ H 1 / 2 {}+c|\log\epsilon|^{1/2}\left\|\mathbf{W}_s\right\|_{H^1}\left(\left\|\mathbf{X}_s\right\|_{H^1}^2+\left\|\mathbf{Y}_s\right\|_{H^1}^2\right)\left\|T_{m_\epsilon}^{1/2}\varphi\right\|_{L^2}\left\|f\right\|_{H^{1/2}} + c ∣ log ϵ ∣ 1/2 ∥ W s ∥ H 1 ( ∥ X s ∥ H 1 2 + ∥ Y s ∥ H 1 2 ) T m ϵ 1/2 φ L 2 ∥ f ∥ H 1/2 ∣ ∫ T φ ⋅ ( P X s T m ϵ n P X s − P Y s ∣ log ϵ ∣ 4 π P Y s ) [ f ] d s ∣ ≤ c ∥ T m ϵ 1 / 2 ( P X s φ ) ∥ L 2 ∥ P X s ⊥ f ∥ L 2 \left|\int_{\mathbb{T}}\varphi\cdot\left(\mathbb{P}_{X_s}T_{m_\epsilon}^{n}\mathbb{P}_{X_s}-\mathbb{P}_{Y_s}\frac{|\log\epsilon|}{4\pi}\mathbb{P}_{Y_s}\right)[f]\,ds\right|\le c\left\|T_{m_\epsilon}^{1/2}(\mathbb{P}_{X_s}\varphi)\right\|_{L^2}\left\|\mathbb{P}_{X_s}^{\perp}f\right\|_{L^2} ∫ T φ ⋅ ( P X s T m ϵ n P X s − P Y s 4 π ∣ log ϵ ∣ P Y s ) [ f ] d s ≤ c T m ϵ 1/2 ( P X s φ ) L 2 P X s ⊥ f L 2 + c ∣ log ϵ ∣ 1 / 2 ∥ W s ∥ H 1 ( 1 + ∥ X s ∥ H 1 2 + ∥ Y s ∥ H 1 2 ) ∥ T m ϵ 1 / 2 φ ∥ L 2 ∥ f ∥ H 1 / 2 . (146) {}+c|\log\epsilon|^{1/2}\left\|\mathbf{W}_s\right\|_{H^1}\left(1+\left\|\mathbf{X}_s\right\|_{H^1}^2+\left\|\mathbf{Y}_s\right\|_{H^1}^2\right)\left\|T_{m_\epsilon}^{1/2}\varphi\right\|_{L^2}\left\|f\right\|_{H^{1/2}}. \tag*{(146)} + c ∣ log ϵ ∣ 1/2 ∥ W s ∥ H 1 ( 1 + ∥ X s ∥ H 1 2 + ∥ Y s ∥ H 1 2 ) T m ϵ 1/2 φ L 2 ∥ f ∥ H 1/2 . (146)
Note that the use of this corollary will be the only place requiring additional regularity on the resistive force theory curve Y \mathbf{Y} Y beyond the natural energy space. We proceed to prove Lemma 4.2 and Corollary 4.3 together.
Proof. We begin with the tangential direction of Lemma 4.2. Using Parseval’s theorem and Lemma 2.1 with the low wavenumber correction of Lemma 2.3, we have
∣ ∫ T φ ⋅ ( T m ϵ n − ∣ log ϵ ∣ 2 π ) f d s ∣ = ∣ ∑ ∣ k ∣ = 0 ∞ ( m ϵ t ( k ) − ∣ log ϵ ∣ 2 π ) F [ φ ] ⋅ F [ f ] ‾ ∣ ≤ c ∑ ∣ k ∣ = 1 ⌊ 1 2 π ϵ ⌋ ( 1 + log ∣ k ∣ ) ∣ F [ φ ] ⋅ F [ f ] ‾ ∣ + c ∑ ∣ k ∣ = ⌈ 1 2 π ϵ ⌉ ∞ ( ϵ ∣ k ∣ ) − 1 / 2 ∣ m ϵ t ( k ) 1 / 2 F [ φ ] ⋅ F [ f ] ‾ ∣ ≤ c ∥ T m ϵ 1 / 2 φ ∥ L 2 ∥ f ∥ L 2 . (147) \begin{aligned}
\left|\int_{\mathbb{T}}\varphi\cdot\left(T_{m_\epsilon}^{n}-\frac{|\log\epsilon|}{2\pi}\right)f\,ds\right|=\left|\sum_{|k|=0}^{\infty}\left(m_\epsilon^{t}(k)-\frac{|\log\epsilon|}{2\pi}\right)\mathcal{F}[\varphi]\cdot\overline{\mathcal{F}[f]}\right| \\
&\le c\sum_{|k|=1}^{\left\lfloor\frac{1}{2\pi\epsilon}\right\rfloor}(1+\log|k|)\left|\mathcal{F}[\varphi]\cdot\overline{\mathcal{F}[f]}\right|+c\sum_{|k|=\left\lceil\frac{1}{2\pi\epsilon}\right\rceil}^{\infty}(\epsilon|k|)^{-1/2}\left|m_\epsilon^{t}(k)^{1/2}\mathcal{F}[\varphi]\cdot\overline{\mathcal{F}[f]}\right| \\
&\le c\left\|T_{m_\epsilon}^{1/2}\varphi\right\|_{L^2}\left\|f\right\|_{L^2}. \tag*{(147)}
\end{aligned} ∫ T φ ⋅ ( T m ϵ n − 2 π ∣ log ϵ ∣ ) f d s = ∣ k ∣ = 0 ∑ ∞ ( m ϵ t ( k ) − 2 π ∣ log ϵ ∣ ) F [ φ ] ⋅ F [ f ] ≤ c ∣ k ∣ = 1 ∑ ⌊ 2 π ϵ 1 ⌋ ( 1 + log ∣ k ∣ ) F [ φ ] ⋅ F [ f ] + c ∣ k ∣ = ⌈ 2 π ϵ 1 ⌉ ∑ ∞ ( ϵ ∣ k ∣ ) − 1/2 m ϵ t ( k ) 1/2 F [ φ ] ⋅ F [ f ] ≤ c T m ϵ 1/2 φ L 2 ∥ f ∥ L 2 . (147)
Here in the sum over high wavenumbers, we have used that ∣ k ∣ − 1 ≲ ϵ |k|^{-1}\lesssim\epsilon ∣ k ∣ − 1 ≲ ϵ , in particular, we may trade the extra regularity of f f f to avoid losing factors of ϵ \epsilon ϵ . The normal direction bound follows by a nearly identical calculation using the low wavenumber correction of Lemma 2.3 for m ϵ n ( k ) m_\epsilon^{n}(k) m ϵ n ( k ) .
To show Corollary 4.3, we again treat the tangential direction explicitly and note that the normal direction follows similarly. We may write
∫ T φ ⋅ ( P X s T m ϵ n P X s − P Y s ∣ log ϵ ∣ 2 π P Y s ) [ f ] d s = I 1 + I 2 + I 3 , \int_{\mathbb{T}}\varphi\cdot\left(\mathbb{P}_{X_s}T_{m_\epsilon}^{n}\mathbb{P}_{X_s}-\mathbb{P}_{Y_s}\frac{|\log\epsilon|}{2\pi}\mathbb{P}_{Y_s}\right)[f]\,ds=I_1+I_2+I_3, ∫ T φ ⋅ ( P X s T m ϵ n P X s − P Y s 2 π ∣ log ϵ ∣ P Y s ) [ f ] d s = I 1 + I 2 + I 3 ,
I 1 = ∫ T P X s φ ⋅ ( T m ϵ n − ∣ log ϵ ∣ 2 π ) [ P X s f ] d s I_1=\int_{\mathbb{T}}\mathbb{P}_{X_s}\varphi\cdot\left(T_{m_\epsilon}^{n}-\frac{|\log\epsilon|}{2\pi}\right)[\mathbb{P}_{X_s}f]\,ds I 1 = ∫ T P X s φ ⋅ ( T m ϵ n − 2 π ∣ log ϵ ∣ ) [ P X s f ] d s
I 2 = ∣ log ϵ ∣ 2 π ∫ T P X s φ ⋅ ( P X s − P Y s ) f d s I_2=\frac{|\log\epsilon|}{2\pi}\int_{\mathbb{T}}\mathbb{P}_{X_s}\varphi\cdot(\mathbb{P}_{X_s}-\mathbb{P}_{Y_s})f\,ds I 2 = 2 π ∣ log ϵ ∣ ∫ T P X s φ ⋅ ( P X s − P Y s ) f d s
I 3 = ∣ log ϵ ∣ 2 π ∫ T φ ⋅ ( P X s − P Y s ) P X s f d s . (148) I_3=\frac{|\log\epsilon|}{2\pi}\int_{\mathbb{T}}\varphi\cdot(\mathbb{P}_{X_s}-\mathbb{P}_{Y_s})\mathbb{P}_{X_s}f\,ds. \tag*{(148)} I 3 = 2 π ∣ log ϵ ∣ ∫ T φ ⋅ ( P X s − P Y s ) P X s f d s . (148)
By Lemma 4.2, we immediately have
∣ I 1 ∣ ≤ c ∥ T m ϵ 1 / 2 ( P X s φ ) ∥ L 2 ∥ P X s f ∥ L 2 . (149) |I_1|\le c\left\|T_{m_\epsilon}^{1/2}(\mathbb{P}_{X_s}\varphi)\right\|_{L^2}\left\|\mathbb{P}_{X_s}f\right\|_{L^2}. \tag*{(149)} ∣ I 1 ∣ ≤ c T m ϵ 1/2 ( P X s φ ) L 2 ∥ P X s f ∥ L 2 . (149)
For I 2 I_2 I 2 and I 3 I_3 I 3 , writing φ = T m ϵ − 1 / 2 T m ϵ 1 / 2 φ \varphi= T_{m_\epsilon}^{-1/2}T_{m_\epsilon}^{1/2}\varphi φ = T m ϵ − 1/2 T m ϵ 1/2 φ , we may use Lemma 4.1 and the additional regularity of f f f to estimate
∣ I 2 ∣ ≤ c ∣ log ϵ ∣ ∣ ∫ T ( T m ϵ − 1 / 2 T m ϵ 1 / 2 φ ) ⋅ X s ) X s ⋅ ( ( W s ⊗ X s + Y s ⊗ W s ) f ) d s ∣ ≤ c ∣ log ϵ ∣ ∥ T m ϵ 1 / 2 φ ∥ L 2 ∥ T m ϵ − 1 / 2 ( X s ⊗ X s ( W s ⊗ X s + Y s ⊗ W s ) f ) ∥ L 2 ≤ c ∣ log ϵ ∣ 1 / 2 ∥ T m ϵ 1 / 2 φ ∥ L 2 ∥ X s ⊗ X s ( W s ⊗ X s + Y s ⊗ W s ) ∥ H 1 ∥ f ∥ H 1 / 2 ≤ c ∣ log ϵ ∣ 1 / 2 ∥ W s ∥ H 1 ( ∥ X s ∥ H 1 2 + ∥ Y s ∥ H 1 2 ) ∥ T m ϵ 1 / 2 φ ∥ L 2 ∥ f ∥ H 1 / 2 , (150) \begin{aligned}
|I_2| &\le c|\log\epsilon| \left|\int_{\mathbb{T}}(T_{m_\epsilon}^{-1/2}T_{m_\epsilon}^{1/2}\varphi)\cdot X_s)X_s\cdot((W_s\otimes X_s+Y_s\otimes W_s)f)\,ds\right| \\
&\le c|\log\epsilon| \left\|T_{m_\epsilon}^{1/2}\varphi\right\|_{L^2}\left\|T_{m_\epsilon}^{-1/2}(X_s\otimes X_s(W_s\otimes X_s+Y_s\otimes W_s)f)\right\|_{L^2} \\
&\le c|\log\epsilon|^{1/2}\left\|T_{m_\epsilon}^{1/2}\varphi\right\|_{L^2}\left\|X_s\otimes X_s(W_s\otimes X_s+Y_s\otimes W_s)\right\|_{H^1}\|f\|_{H^{1/2}} \\
&\le c|\log\epsilon|^{1/2}\|W_s\|_{H^1}(\|X_s\|_{H^1}^2+\|Y_s\|_{H^1}^2)\left\|T_{m_\epsilon}^{1/2}\varphi\right\|_{L^2}\|f\|_{H^{1/2}}, \tag*{(150)}
\end{aligned} ∣ I 2 ∣ ≤ c ∣ log ϵ ∣ ∫ T ( T m ϵ − 1/2 T m ϵ 1/2 φ ) ⋅ X s ) X s ⋅ (( W s ⊗ X s + Y s ⊗ W s ) f ) d s ≤ c ∣ log ϵ ∣ T m ϵ 1/2 φ L 2 T m ϵ − 1/2 ( X s ⊗ X s ( W s ⊗ X s + Y s ⊗ W s ) f ) L 2 ≤ c ∣ log ϵ ∣ 1/2 T m ϵ 1/2 φ L 2 ∥ X s ⊗ X s ( W s ⊗ X s + Y s ⊗ W s ) ∥ H 1 ∥ f ∥ H 1/2 ≤ c ∣ log ϵ ∣ 1/2 ∥ W s ∥ H 1 ( ∥ X s ∥ H 1 2 + ∥ Y s ∥ H 1 2 ) T m ϵ 1/2 φ L 2 ∥ f ∥ H 1/2 , (150)
as well as
∣ I 3 ∣ ≤ c ∣ log ϵ ∣ ∣ ∫ T ( T m ϵ − 1 / 2 T m ϵ 1 / 2 φ ) ⋅ ( W s ⊗ X s + Y s ⊗ W s ) X s ( X s ⋅ f ) d s ∣ ≤ c ∣ log ϵ ∣ ∥ T m ϵ 1 / 2 φ ∥ L 2 ∥ T m ϵ − 1 / 2 ( ( W s ⊗ X s + Y s ⊗ W s ) X s ( X s ⋅ f ) ) ∥ L 2 ≤ c ∣ log ϵ ∣ 1 / 2 ∥ T m ϵ 1 / 2 φ ∥ L 2 ∥ ( W s ⊗ X s + Y s ⊗ W s ) X s ⊗ X s ∥ H 1 ∥ f ∥ H 1 / 2 ≤ c ∣ log ϵ ∣ 1 / 2 ∥ W s ∥ H 1 ( ∥ X s ∥ H 1 2 + ∥ Y s ∥ H 1 2 ) ∥ T m ϵ 1 / 2 φ ∥ L 2 ∥ f ∥ H 1 / 2 . (151) \begin{aligned}
|I_3| &\le c|\log\epsilon| \left|\int_{\mathbb{T}}(T_{m_\epsilon}^{-1/2}T_{m_\epsilon}^{1/2}\varphi)\cdot(W_s\otimes X_s+Y_s\otimes W_s)X_s(X_s\cdot f)\,ds\right| \\
&\le c|\log\epsilon|\left\|T_{m_\epsilon}^{1/2}\varphi\right\|_{L^2}\left\|T_{m_\epsilon}^{-1/2}((W_s\otimes X_s+Y_s\otimes W_s)X_s(X_s\cdot f))\right\|_{L^2} \\
&\le c|\log\epsilon|^{1/2}\left\|T_{m_\epsilon}^{1/2}\varphi\right\|_{L^2}\left\|(W_s\otimes X_s+Y_s\otimes W_s)X_s\otimes X_s\right\|_{H^1}\|f\|_{H^{1/2}} \\
&\le c|\log\epsilon|^{1/2}\|W_s\|_{H^1}(\|X_s\|_{H^1}^2+\|Y_s\|_{H^1}^2)\left\|T_{m_\epsilon}^{1/2}\varphi\right\|_{L^2}\|f\|_{H^{1/2}}. \tag*{(151)}
\end{aligned} ∣ I 3 ∣ ≤ c ∣ log ϵ ∣ ∫ T ( T m ϵ − 1/2 T m ϵ 1/2 φ ) ⋅ ( W s ⊗ X s + Y s ⊗ W s ) X s ( X s ⋅ f ) d s ≤ c ∣ log ϵ ∣ T m ϵ 1/2 φ L 2 T m ϵ − 1/2 (( W s ⊗ X s + Y s ⊗ W s ) X s ( X s ⋅ f )) L 2 ≤ c ∣ log ϵ ∣ 1/2 T m ϵ 1/2 φ L 2 ∥ ( W s ⊗ X s + Y s ⊗ W s ) X s ⊗ X s ∥ H 1 ∥ f ∥ H 1/2 ≤ c ∣ log ϵ ∣ 1/2 ∥ W s ∥ H 1 ( ∥ X s ∥ H 1 2 + ∥ Y s ∥ H 1 2 ) T m ϵ 1/2 φ L 2 ∥ f ∥ H 1/2 . (151)
In total, we may bound
∣ I 1 ∣ + ∣ I 2 ∣ + ∣ I 3 ∣ ≤ c ∥ T m ϵ 1 / 2 ( P X s φ ) ∥ L 2 ∥ P X s f ∥ L 2 + c ∣ log ϵ ∣ 1 / 2 ∥ W s ∥ H 1 ( ∥ X s ∥ H 1 2 + ∥ Y s ∥ H 1 2 ) ∥ T m ϵ 1 / 2 φ ∥ L 2 ∥ f ∥ H 1 / 2 (152) \begin{aligned}
|I_1|+|I_2|+|I_3| &\le c\left\|T_{m_\epsilon}^{1/2}(\mathbb{P}_{X_s}\varphi)\right\|_{L^2}\|\mathbb{P}_{X_s}f\|_{L^2} \\
&\quad+c|\log\epsilon|^{1/2}\|W_s\|_{H^1}(\|X_s\|_{H^1}^2+\|Y_s\|_{H^1}^2)\left\|T_{m_\epsilon}^{1/2}\varphi\right\|_{L^2}\|f\|_{H^{1/2}} \tag*{(152)}
\end{aligned} ∣ I 1 ∣ + ∣ I 2 ∣ + ∣ I 3 ∣ ≤ c T m ϵ 1/2 ( P X s φ ) L 2 ∥ P X s f ∥ L 2 + c ∣ log ϵ ∣ 1/2 ∥ W s ∥ H 1 ( ∥ X s ∥ H 1 2 + ∥ Y s ∥ H 1 2 ) T m ϵ 1/2 φ L 2 ∥ f ∥ H 1/2 (152)
to obtain Corollary 4.3. □
Tension bounds and difference
In addition to the operator difference estimates from section 4.2, we will require bounds for terms resulting from the difference in tension behavior between the L ϵ ‾ ( X ) \overline{L_\epsilon}(X) L ϵ ( X ) curve X X X and the resistive force theory curve Y Y Y .
We first note that the resistive force theory equation for the tension τ Y \tau_Y τ Y may be obtained in an analogous manner to (42) . Differentiating (137) in s s s and dotting with Y s Y_s Y s , the inextensibility constraint yields the equation
( ( I + Y s ⊗ Y s ) ( τ Y Y s ) s ) ⋅ Y s = ( ( I + Y s ⊗ Y s ) Y s s s s ) s ⋅ Y s . (153) ((\mathbf{I}+Y_s\otimes Y_s)(\tau_Y Y_s)_s)\cdot Y_s=((\mathbf{I}+Y_s\otimes Y_s)Y_{ssss})_s\cdot Y_s. \tag*{(153)} (( I + Y s ⊗ Y s ) ( τ Y Y s ) s ) ⋅ Y s = (( I + Y s ⊗ Y s ) Y ssss ) s ⋅ Y s . (153)
We further note that, by [1 Lemma 3.3] , τ Y \tau_Y τ Y satisfying (153) obeys the bounds
∥ τ Y ∥ H 1 ≤ c ∥ Y s s ∥ H 1 2 , ∥ τ Y ∥ H 2 ≤ c ( ∥ Y s ∥ H 2 4 + ∥ Y s ∥ H 2 2 ∥ Y s ∥ H 3 ) . (154) \|\tau_Y\|_{H^1}\le c\|Y_{ss}\|_{H^1}^2,
\qquad
\|\tau_Y\|_{H^2}\le c(\|Y_s\|_{H^2}^4+\|Y_s\|_{H^2}^2\|Y_s\|_{H^3}). \tag*{(154)} ∥ τ Y ∥ H 1 ≤ c ∥ Y ss ∥ H 1 2 , ∥ τ Y ∥ H 2 ≤ c ( ∥ Y s ∥ H 2 4 + ∥ Y s ∥ H 2 2 ∥ Y s ∥ H 3 ) . (154)
We may immediately estimate the terms involving τ Y \tau_Y τ Y in A ∥ A^\parallel A ∥ and A ⊥ A^\perp A ⊥ of (142) as
∥ T m ϵ 1 / 2 ( P X s ( τ Y W s ) s ) ∥ L 2 ≤ c ∣ log ϵ ∣ 1 / 2 ∥ P X s ( τ Y W s ) s ∥ L 2 ≤ c ∣ log ϵ ∣ 1 / 2 ∥ ( τ Y ) s ∥ L ∞ ∥ W s ∥ H 1 ≤ c ∣ log ϵ ∣ 1 / 2 ( ∥ Y s ∥ H 2 4 + ∥ Y s ∥ H 2 2 ∥ Y s ∥ H 3 ) ∥ W s ∥ H 1 , (155) \begin{aligned}
\left\|T_{m_\epsilon}^{1/2}(\mathbb{P}_{X_s}(\tau_Y W_s)_s)\right\|_{L^2}
&\le c|\log\epsilon|^{1/2}\|\mathbb{P}_{X_s}(\tau_Y W_s)_s\|_{L^2}
\le c|\log\epsilon|^{1/2}\|(\tau_Y)_s\|_{L^\infty}\|W_s\|_{H^1} \\
&\le c|\log\epsilon|^{1/2}(\|Y_s\|_{H^2}^4+\|Y_s\|_{H^2}^2\|Y_s\|_{H^3})\|W_s\|_{H^1}, \tag*{(155)}
\end{aligned} T m ϵ 1/2 ( P X s ( τ Y W s ) s ) L 2 ≤ c ∣ log ϵ ∣ 1/2 ∥ P X s ( τ Y W s ) s ∥ L 2 ≤ c ∣ log ϵ ∣ 1/2 ∥ ( τ Y ) s ∥ L ∞ ∥ W s ∥ H 1 ≤ c ∣ log ϵ ∣ 1/2 ( ∥ Y s ∥ H 2 4 + ∥ Y s ∥ H 2 2 ∥ Y s ∥ H 3 ) ∥ W s ∥ H 1 , (155)
with an identical bound for the normal direction term ∥ T m ϵ 1 / 2 ( P X s ⊥ ( τ Y W s ) s ) ∥ L 2 \left\|T_{m_\epsilon}^{1/2}(\mathbb{P}_{X_s}^{\perp}(\tau_Y W_s)_s)\right\|_{L^2} T m ϵ 1/2 ( P X s ⊥ ( τ Y W s ) s ) L 2 .
We will additionally require a bound for the terms in A ∥ A^\parallel A ∥ and A ⊥ A^\perp A ⊥ involving the difference τ W = τ X − τ Y \tau_W=\tau_X-\tau_Y τ W = τ X − τ Y . We show the following.
Lemma 4.4 (Tension difference). The difference τ W = τ X − τ Y \tau_{\mathrm W}=\tau_X-\tau_Y τ W = τ X − τ Y between the tensions corresponding to curves X \mathbf X X and Y \mathbf Y Y , respectively, satisfies the bound
∥ T m ϵ 1 / 2 ( τ W ) s X s ∥ L 2 + ∥ T m ϵ 1 / 2 ( τ W X s s ) ∥ L 2 ≤ c ( ∣ log ϵ ∣ 1 / 2 ∥ W s ∥ H 1 ( 1 + ∥ Y s ∥ H 3 2 ) + ( ∥ T m ϵ 1 / 2 W s s s s ∥ L 2 + 1 + ∥ Y s ∥ H 3 2 ) ( 1 + ∥ X s ∥ H 1 + ∥ Y s ∥ H 3 2 ) ) , (156) \left\|T_{m_\epsilon}^{1/2}(\tau_{\mathrm W})_s\mathbf X_s\right\|_{L^2}+\left\|T_{m_\epsilon}^{1/2}(\tau_{\mathrm W}\mathbf X_{ss})\right\|_{L^2}\leq c\left(\lvert\log\epsilon\rvert^{1/2}\|\mathbf W_s\|_{H^1}\left(1+\|\mathbf Y_s\|_{H^3}^2\right)+\left(\left\|T_{m_\epsilon}^{1/2}\mathbf W_{ssss}\right\|_{L^2}+1+\|\mathbf Y_s\|_{H^3}^2\right)\left(1+\|\mathbf X_s\|_{H^1}+\|\mathbf Y_s\|_{H^3}^2\right)\right), \tag*{(156)} T m ϵ 1/2 ( τ W ) s X s L 2 + T m ϵ 1/2 ( τ W X ss ) L 2 ≤ c ( ∣ log ϵ ∣ 1/2 ∥ W s ∥ H 1 ( 1 + ∥ Y s ∥ H 3 2 ) + ( T m ϵ 1/2 W ssss L 2 + 1 + ∥ Y s ∥ H 3 2 ) ( 1 + ∥ X s ∥ H 1 + ∥ Y s ∥ H 3 2 ) ) , (156) where W = X − Y \mathbf W=\mathbf X-\mathbf Y W = X − Y .
Proof. For any φ ∈ H 1 / 2 ( T ) \varphi\in H^{1/2}(\mathbb T) φ ∈ H 1/2 ( T ) , defining the bilinear forms
B ϵ X ( τ X , φ ) = ∫ T L ‾ ϵ ( X ) [ ( τ X X s ) s ] ⋅ ( φ X s ) s d s , B_\epsilon^X(\tau_X,\varphi)=\int_{\mathbb T}\overline{\mathcal L}_\epsilon(\mathbf X)[(\tau_X\mathbf X_s)_s]\cdot(\varphi\mathbf X_s)_s\,ds, B ϵ X ( τ X , φ ) = ∫ T L ϵ ( X ) [( τ X X s ) s ] ⋅ ( φ X s ) s d s ,
B ϵ Y ( τ Y , φ ) = ∣ log ϵ ∣ 4 π ∫ T ( ( I + Y s ⊗ Y s ) ( τ Y Y s ) s ) ⋅ ( φ Y s ) s d s , (157) B_\epsilon^Y(\tau_Y,\varphi)=\frac{\lvert\log\epsilon\rvert}{4\pi}\int_{\mathbb T}\left((\mathbf I+\mathbf Y_s\otimes\mathbf Y_s)(\tau_Y\mathbf Y_s)_s\right)\cdot(\varphi\mathbf Y_s)_s\,ds, \tag*{(157)} B ϵ Y ( τ Y , φ ) = 4 π ∣ log ϵ ∣ ∫ T ( ( I + Y s ⊗ Y s ) ( τ Y Y s ) s ) ⋅ ( φ Y s ) s d s , (157)
we may write the difference between the weak form of the equations satisfied by τ X \tau_X τ X and τ Y \tau_Y τ Y as
B ϵ X ( τ X , φ ) − B ϵ Y ( τ Y , φ ) = ∫ T ( L ‾ ϵ ( X ) [ X s s s s ] ⋅ ( φ X s ) s − ∣ log ϵ ∣ 4 π ( ( I + Y s ⊗ Y s ) Y s s s s ) ⋅ ( φ Y s ) s ) d s . (158) B_\epsilon^X(\tau_X,\varphi)-B_\epsilon^Y(\tau_Y,\varphi)
=\int_{\mathbb T}\left(\overline{\mathcal L}_\epsilon(\mathbf X)[\mathbf X_{ssss}]\cdot(\varphi\mathbf X_s)_s-\frac{\lvert\log\epsilon\rvert}{4\pi}\left((\mathbf I+\mathbf Y_s\otimes\mathbf Y_s)\mathbf Y_{ssss}\right)\cdot(\varphi\mathbf Y_s)_s\right)\,ds. \tag*{(158)} B ϵ X ( τ X , φ ) − B ϵ Y ( τ Y , φ ) = ∫ T ( L ϵ ( X ) [ X ssss ] ⋅ ( φ X s ) s − 4 π ∣ log ϵ ∣ ( ( I + Y s ⊗ Y s ) Y ssss ) ⋅ ( φ Y s ) s ) d s . (158)
The left hand side of (158) may be written as
B ϵ X ( τ X , φ ) − B ϵ Y ( τ Y , φ ) = B ∥ + B ⊥ , B_\epsilon^X(\tau_X,\varphi)-B_\epsilon^Y(\tau_Y,\varphi)=B^\parallel+B^\perp, B ϵ X ( τ X , φ ) − B ϵ Y ( τ Y , φ ) = B ∥ + B ⊥ ,
B ∥ = ∫ T ( ( P X s T m ϵ ∥ ( ( τ X ) s X s ) ) ⋅ ( φ X s ) s − ∣ log ϵ ∣ 2 π ( τ Y ) s Y s ⋅ ( φ Y s ) s ) d s , B^\parallel=\int_{\mathbb T}\left((\mathbb P_{\mathbf X_s}T_{m_\epsilon}^{\parallel}((\tau_X)_s\mathbf X_s))\cdot(\varphi\mathbf X_s)_s-\frac{\lvert\log\epsilon\rvert}{2\pi}(\tau_Y)_s\mathbf Y_s\cdot(\varphi\mathbf Y_s)_s\right)\,ds, B ∥ = ∫ T ( ( P X s T m ϵ ∥ (( τ X ) s X s )) ⋅ ( φ X s ) s − 2 π ∣ log ϵ ∣ ( τ Y ) s Y s ⋅ ( φ Y s ) s ) d s ,
B ⊥ = ∫ T ( ( P X s ⊥ T m ϵ ⊥ ( τ X X s s ) ) ⋅ ( φ X s ) s − ∣ log ϵ ∣ 4 π τ Y Y s s ⋅ ( φ Y s ) s ) d s . (159) B^\perp=\int_{\mathbb T}\left((\mathbb P_{\mathbf X_s}^{\perp}T_{m_\epsilon}^{\perp}(\tau_X\mathbf X_{ss}))\cdot(\varphi\mathbf X_s)_s-\frac{\lvert\log\epsilon\rvert}{4\pi}\tau_Y\mathbf Y_{ss}\cdot(\varphi\mathbf Y_s)_s\right)\,ds. \tag*{(159)} B ⊥ = ∫ T ( ( P X s ⊥ T m ϵ ⊥ ( τ X X ss )) ⋅ ( φ X s ) s − 4 π ∣ log ϵ ∣ τ Y Y ss ⋅ ( φ Y s ) s ) d s . (159)
We may further write
B ∥ = ∫ T ( ( X s ⋅ T m ϵ ∥ ( ( τ X ) s X s ) ) − ∣ log ϵ ∣ 2 π ( τ Y ) s ) φ s d s = ∫ T ( ( X s ⋅ T m ϵ ∥ ( ( τ W ) s X s ) ) + X s ⋅ ( T m ϵ ∥ − ∣ log ϵ ∣ 2 π ) [ ( τ Y ) s X s ] ) φ s d s . (160) \begin{aligned}
B^\parallel=\int_{\mathbb T}\left(\left(\mathbf X_s\cdot T_{m_\epsilon}^{\parallel}((\tau_X)_s\mathbf X_s)\right)-\frac{\lvert\log\epsilon\rvert}{2\pi}(\tau_Y)_s\right)\varphi_s\,ds \\
&=\int_{\mathbb T}\left(\left(\mathbf X_s\cdot T_{m_\epsilon}^{\parallel}((\tau_{\mathrm W})_s\mathbf X_s)\right)+\mathbf X_s\cdot\left(T_{m_\epsilon}^{\parallel}-\frac{\lvert\log\epsilon\rvert}{2\pi}\right)[(\tau_Y)_s\mathbf X_s]\right)\varphi_s\,ds. \tag*{(160)}
\end{aligned} B ∥ = ∫ T ( ( X s ⋅ T m ϵ ∥ (( τ X ) s X s ) ) − 2 π ∣ log ϵ ∣ ( τ Y ) s ) φ s d s = ∫ T ( ( X s ⋅ T m ϵ ∥ (( τ W ) s X s ) ) + X s ⋅ ( T m ϵ ∥ − 2 π ∣ log ϵ ∣ ) [( τ Y ) s X s ] ) φ s d s . (160)
as well as
B ⊥ = ∫ T ( X s s ⋅ T m ϵ ⊥ ( τ X X s s ) − ∣ log ϵ ∣ 4 π τ Y ∣ Y s s ∣ 2 ) φ d s = ∫ T ( X s s ⋅ T m ϵ ⊥ ( τ W X s s ) + X s s ⋅ ( T m ϵ ⊥ − ∣ log ϵ ∣ 4 π ) [ τ Y X s s ] + ∣ log ϵ ∣ 4 π τ Y ( ∣ X s s ∣ 2 − ∣ Y s s ∣ 2 ) ) φ d s . (161) \begin{aligned}
B^\perp=\int_{\mathbb T}\left(\mathbf X_{ss}\cdot T_{m_\epsilon}^{\perp}(\tau_X\mathbf X_{ss})-\frac{\lvert\log\epsilon\rvert}{4\pi}\tau_Y\lvert\mathbf Y_{ss}\rvert^2\right)\varphi\,ds \\
&=\int_{\mathbb T}\left(\mathbf X_{ss}\cdot T_{m_\epsilon}^{\perp}(\tau_{\mathrm W}\mathbf X_{ss})+\mathbf X_{ss}\cdot\left(T_{m_\epsilon}^{\perp}-\frac{\lvert\log\epsilon\rvert}{4\pi}\right)[\tau_Y\mathbf X_{ss}]+\frac{\lvert\log\epsilon\rvert}{4\pi}\tau_Y\left(\lvert\mathbf X_{ss}\rvert^2-\lvert\mathbf Y_{ss}\rvert^2\right)\right)\varphi\,ds. \tag*{(161)}
\end{aligned} B ⊥ = ∫ T ( X ss ⋅ T m ϵ ⊥ ( τ X X ss ) − 4 π ∣ log ϵ ∣ τ Y ∣ Y ss ∣ 2 ) φ d s = ∫ T ( X ss ⋅ T m ϵ ⊥ ( τ W X ss ) + X ss ⋅ ( T m ϵ ⊥ − 4 π ∣ log ϵ ∣ ) [ τ Y X ss ] + 4 π ∣ log ϵ ∣ τ Y ( ∣ X ss ∣ 2 − ∣ Y ss ∣ 2 ) ) φ d s . (161)
We may thus rewrite the left hand side of (158) as
B ϵ X ( τ X , φ ) − B ϵ Y ( τ Y , φ ) = B ϵ X ( τ W , φ ) + L H S , B_\epsilon^X(\tau_X,\varphi)-B_\epsilon^Y(\tau_Y,\varphi)=B_\epsilon^X(\tau_{\mathrm W},\varphi)+\mathrm{LHS}, B ϵ X ( τ X , φ ) − B ϵ Y ( τ Y , φ ) = B ϵ X ( τ W , φ ) + LHS ,
L H S = ∫ T ( φ s X s ⋅ ( T m ϵ ∥ − ∣ log ϵ ∣ 2 π ) [ ( τ Y ) s X s ] + φ X s s ⋅ ( T m ϵ ⊥ − ∣ log ϵ ∣ 4 π ) [ τ Y X s s ] + ∣ log ϵ ∣ 4 π τ Y ( ∣ X s s ∣ 2 − ∣ Y s s ∣ 2 ) φ ) d s . (162) \mathrm{LHS}=\int_{\mathbb T}\left(\varphi_s\mathbf X_s\cdot\left(T_{m_\epsilon}^{\parallel}-\frac{\lvert\log\epsilon\rvert}{2\pi}\right)[(\tau_Y)_s\mathbf X_s]+\varphi\mathbf X_{ss}\cdot\left(T_{m_\epsilon}^{\perp}-\frac{\lvert\log\epsilon\rvert}{4\pi}\right)[\tau_Y\mathbf X_{ss}]+\frac{\lvert\log\epsilon\rvert}{4\pi}\tau_Y\left(\lvert\mathbf X_{ss}\rvert^2-\lvert\mathbf Y_{ss}\rvert^2\right)\varphi\right)\,ds. \tag*{(162)} LHS = ∫ T ( φ s X s ⋅ ( T m ϵ ∥ − 2 π ∣ log ϵ ∣ ) [( τ Y ) s X s ] + φ X ss ⋅ ( T m ϵ ⊥ − 4 π ∣ log ϵ ∣ ) [ τ Y X ss ] + 4 π ∣ log ϵ ∣ τ Y ( ∣ X ss ∣ 2 − ∣ Y ss ∣ 2 ) φ ) d s . (162)
Using Lemma 4.2 and the τ Y \tau_Y τ Y bound (154) , we may bound the terms appearing in LHS as
∣ L H S ∣ ≤ c ( ∥ T m ϵ t 1 / 2 ( φ s X s ) ∥ L 2 ∥ ( τ Y ) s X s ∥ L 2 + ∥ T m ϵ n 1 / 2 ( φ X s s ) ∥ L 2 ∥ τ Y X s s ∥ L 2 + ∣ log ϵ ∣ ∥ W s ∥ H 1 ∥ τ Y ∥ L ∞ ( ∥ W s s ∥ L ∞ + ∥ Y s s ∥ L ∞ ) ∥ φ ∥ L 2 ) ≤ c ( ∥ T m ϵ t 1 / 2 ( φ s X s ) ∥ L 2 ∥ Y s s ∥ H 1 2 + ∥ T m ϵ n 1 / 2 ( φ X s s ) ∥ L 2 ∥ Y s s ∥ H 1 2 ∥ X s ∥ H 1 + ∣ log ϵ ∣ ∥ W s ∥ H 1 ∥ Y s s ∥ H 1 2 ( ∥ W s s s ∥ H ˙ 1 / 2 2 / 3 ∥ W s ∥ H 1 1 / 3 + ∥ Y s s ∥ H 1 ) × ∥ T m ϵ n − 1 T m ϵ n n ( φ X s ) ⋅ X s ∥ L 2 ) ≤ c ∥ Y s s ∥ H 1 2 ( ∥ T m ϵ t 1 / 2 ( φ s X s ) ∥ L 2 + ∥ T m ϵ n 1 / 2 ( φ X s s ) ∥ L 2 ) ∥ X s ∥ H 1 + c ∥ W s ∥ H 1 ( ∥ W s s s ∥ H ˙ 1 / 2 2 / 3 ∥ W s ∥ H 1 1 / 3 + ∥ Y s s ∥ H 1 ) ∥ T m ϵ n 1 / 2 ( φ X s ) ∥ L 2 ≤ c ( ∥ T m ϵ t 1 / 2 ( φ s X s ) ∥ L 2 + ∥ T m ϵ n 1 / 2 ( φ X s s ) ∥ L 2 ) ∥ Y s s ∥ H 1 2 ( 1 + ∥ X s ∥ H 1 2 + ∥ Y s s ∥ H 1 2 ) + ∣ log ϵ ∣ 1 / 2 ∥ W s ∥ H 1 4 / 3 ∥ W s s s ∥ H ˙ 1 / 2 2 / 3 . (163) \begin{aligned}
|\mathrm{LHS}| &\le c\left(\left\|T_{m_\epsilon^t}^{1/2}(\varphi_sX_s)\right\|_{L^2}\left\|(\tau_Y)_sX_s\right\|_{L^2}+\left\|T_{m_\epsilon^n}^{1/2}(\varphi X_{ss})\right\|_{L^2}\left\|\tau_YX_{ss}\right\|_{L^2}\right.\\
&\qquad\left.+|\log\epsilon|\left\|W_s\right\|_{H^1}\left\|\tau_Y\right\|_{L^\infty}\left(\left\|W_{ss}\right\|_{L^\infty}+\left\|Y_{ss}\right\|_{L^\infty}\right)\|\varphi\|_{L^2}\right)\\
&\le c\left(\left\|T_{m_\epsilon^t}^{1/2}(\varphi_sX_s)\right\|_{L^2}\left\|Y_{ss}\right\|_{H^1}^2+\left\|T_{m_\epsilon^n}^{1/2}(\varphi X_{ss})\right\|_{L^2}\left\|Y_{ss}\right\|_{H^1}^2\left\|X_s\right\|_{H^1}\right.\\
&\qquad\left.+|\log\epsilon|\left\|W_s\right\|_{H^1}\left\|Y_{ss}\right\|_{H^1}^2\left(\left\|W_{sss}\right\|_{\dot H^{1/2}}^{2/3}\left\|W_s\right\|_{H^1}^{1/3}+\left\|Y_{ss}\right\|_{H^1}\right)\right.\\
&\qquad\left.\qquad\times\left\|T_{m_\epsilon^n}^{-1}T_{m_\epsilon^n}^{n}(\varphi X_s)\cdot X_s\right\|_{L^2}\right)\\
&\le c\left\|Y_{ss}\right\|_{H^1}^2\left(\left\|T_{m_\epsilon^t}^{1/2}(\varphi_sX_s)\right\|_{L^2}+\left\|T_{m_\epsilon^n}^{1/2}(\varphi X_{ss})\right\|_{L^2}\right)\left\|X_s\right\|_{H^1}\\
&\qquad+c\left\|W_s\right\|_{H^1}\left(\left\|W_{sss}\right\|_{\dot H^{1/2}}^{2/3}\left\|W_s\right\|_{H^1}^{1/3}+\left\|Y_{ss}\right\|_{H^1}\right)\left\|T_{m_\epsilon^n}^{1/2}(\varphi X_s)\right\|_{L^2}\\
&\le c\left(\left\|T_{m_\epsilon^t}^{1/2}(\varphi_sX_s)\right\|_{L^2}+\left\|T_{m_\epsilon^n}^{1/2}(\varphi X_{ss})\right\|_{L^2}\right)\left\|Y_{ss}\right\|_{H^1}^2\left(1+\left\|X_s\right\|_{H^1}^2+\left\|Y_{ss}\right\|_{H^1}^2\right)\\
&\qquad+|\log\epsilon|^{1/2}\left\|W_s\right\|_{H^1}^{4/3}\left\|W_{sss}\right\|_{\dot H^{1/2}}^{2/3}. \tag*{(163)}
\end{aligned} ∣ LHS ∣ ≤ c ( T m ϵ t 1/2 ( φ s X s ) L 2 ∥ ( τ Y ) s X s ∥ L 2 + T m ϵ n 1/2 ( φ X ss ) L 2 ∥ τ Y X ss ∥ L 2 + ∣ log ϵ ∣ ∥ W s ∥ H 1 ∥ τ Y ∥ L ∞ ( ∥ W ss ∥ L ∞ + ∥ Y ss ∥ L ∞ ) ∥ φ ∥ L 2 ) ≤ c ( T m ϵ t 1/2 ( φ s X s ) L 2 ∥ Y ss ∥ H 1 2 + T m ϵ n 1/2 ( φ X ss ) L 2 ∥ Y ss ∥ H 1 2 ∥ X s ∥ H 1 + ∣ log ϵ ∣ ∥ W s ∥ H 1 ∥ Y ss ∥ H 1 2 ( ∥ W sss ∥ H ˙ 1/2 2/3 ∥ W s ∥ H 1 1/3 + ∥ Y ss ∥ H 1 ) × T m ϵ n − 1 T m ϵ n n ( φ X s ) ⋅ X s L 2 ) ≤ c ∥ Y ss ∥ H 1 2 ( T m ϵ t 1/2 ( φ s X s ) L 2 + T m ϵ n 1/2 ( φ X ss ) L 2 ) ∥ X s ∥ H 1 + c ∥ W s ∥ H 1 ( ∥ W sss ∥ H ˙ 1/2 2/3 ∥ W s ∥ H 1 1/3 + ∥ Y ss ∥ H 1 ) T m ϵ n 1/2 ( φ X s ) L 2 ≤ c ( T m ϵ t 1/2 ( φ s X s ) L 2 + T m ϵ n 1/2 ( φ X ss ) L 2 ) ∥ Y ss ∥ H 1 2 ( 1 + ∥ X s ∥ H 1 2 + ∥ Y ss ∥ H 1 2 ) + ∣ log ϵ ∣ 1/2 ∥ W s ∥ H 1 4/3 ∥ W sss ∥ H ˙ 1/2 2/3 . (163)
Here we have also exploited Lemma 4.1 to bound T m ϵ n − 1 T_{m_\epsilon^n}^{-1} T m ϵ n − 1 , and have used that we may opt not to use the regularity gain from T m ϵ n 1 / 2 T_{m_\epsilon^n}^{1/2} T m ϵ n 1/2 to avoid an additional factor of ϵ − 1 / 2 \epsilon^{-1/2} ϵ − 1/2 from the high wavenumber scaling of m ϵ n ( k ) m_\epsilon^n(k) m ϵ n ( k ) . We have further used that m ϵ n ( k ) ≲ m ϵ t ( k ) ≲ m ϵ n ( k ) m_\epsilon^n(k)\lesssim m_\epsilon^t(k)\lesssim m_\epsilon^n(k) m ϵ n ( k ) ≲ m ϵ t ( k ) ≲ m ϵ n ( k ) , so we may interchange T m ϵ t T_{m_\epsilon^t} T m ϵ t and T m ϵ n T_{m_\epsilon^n} T m ϵ n in the above bounds.
We next turn to the right hand side terms RHS of (158) , which we may write as
R H S = R ∥ + R ⊥ , \mathrm{RHS}=R^\parallel+R^\perp, RHS = R ∥ + R ⊥ ,
R ∥ = ∫ T ( X s ⋅ T m ϵ t ( X s ⋅ X s s s s X s ) − ∣ log ϵ ∣ 2 π Y s ⋅ Y s s s s ) φ s d s R^\parallel=\int_{\mathbb{T}}\left(X_s\cdot T_{m_\epsilon^t}(X_s\cdot X_{ssss}X_s)-\frac{|\log\epsilon|}{2\pi}Y_s\cdot Y_{ssss}\right)\varphi_s\,ds R ∥ = ∫ T ( X s ⋅ T m ϵ t ( X s ⋅ X ssss X s ) − 2 π ∣ log ϵ ∣ Y s ⋅ Y ssss ) φ s d s
R ⊥ = ∫ T ( X s s ⋅ T m ϵ n ( P X s ⊥ X s s s s ) − ∣ log ϵ ∣ 4 π Y s s s s ⋅ Y s s ) φ d s . (164) R^\perp=\int_{\mathbb{T}}\left(X_{ss}\cdot T_{m_\epsilon^n}(\mathbb{P}^{\perp}_{X_s}X_{ssss})-\frac{|\log\epsilon|}{4\pi}Y_{ssss}\cdot Y_{ss}\right)\varphi\,ds. \tag*{(164)} R ⊥ = ∫ T ( X ss ⋅ T m ϵ n ( P X s ⊥ X ssss ) − 4 π ∣ log ϵ ∣ Y ssss ⋅ Y ss ) φ d s . (164)
Using the inextensibility identities (36) , we may write
R ∥ = − 3 ∫ T φ s X s ⋅ T m ϵ t ( ( X s s ⋅ W s s s + W s s ⋅ Y s s s ) X s ) d s − 3 ∫ T φ s X s ⋅ ( T m ϵ t − ∣ log ϵ ∣ 2 π ) [ X s Y s s ⋅ Y s s s ] d s = − 3 ∫ T φ s X s ⋅ T m ϵ t ( 1 2 ∂ s ∣ W s s ∣ 2 X s + ∂ s ( Y s s ⋅ W s s ) X s ) d s − 3 2 ∫ T φ s X s ⋅ ( T m ϵ t − ∣ log ϵ ∣ 2 π ) [ ∂ s ∣ Y s s ∣ 2 X s ] d s . (165) \begin{aligned}
R^\parallel={}&-3\int_{\mathbb{T}}\varphi_sX_s\cdot T_{m_\epsilon^t}\left((X_{ss}\cdot W_{sss}+W_{ss}\cdot Y_{sss})X_s\right)\,ds\\
&-3\int_{\mathbb{T}}\varphi_sX_s\cdot\left(T_{m_\epsilon^t}-\frac{|\log\epsilon|}{2\pi}\right)\left[X_sY_{ss}\cdot Y_{sss}\right]\,ds\\
={}&-3\int_{\mathbb{T}}\varphi_sX_s\cdot T_{m_\epsilon^t}\left(\frac{1}{2}\partial_s|W_{ss}|^2X_s+\partial_s(Y_{ss}\cdot W_{ss})X_s\right)\,ds\\
&-\frac{3}{2}\int_{\mathbb{T}}\varphi_sX_s\cdot\left(T_{m_\epsilon^t}-\frac{|\log\epsilon|}{2\pi}\right)\left[\partial_s|Y_{ss}|^2X_s\right]\,ds. \tag*{(165)}
\end{aligned} R ∥ = = − 3 ∫ T φ s X s ⋅ T m ϵ t ( ( X ss ⋅ W sss + W ss ⋅ Y sss ) X s ) d s − 3 ∫ T φ s X s ⋅ ( T m ϵ t − 2 π ∣ log ϵ ∣ ) [ X s Y ss ⋅ Y sss ] d s − 3 ∫ T φ s X s ⋅ T m ϵ t ( 2 1 ∂ s ∣ W ss ∣ 2 X s + ∂ s ( Y ss ⋅ W ss ) X s ) d s − 2 3 ∫ T φ s X s ⋅ ( T m ϵ t − 2 π ∣ log ϵ ∣ ) [ ∂ s ∣ Y ss ∣ 2 X s ] d s . (165)
Using Lemma 4.2 and the interpolation inequalities (37) , we may estimate R ∥ R^{\parallel} R ∥ as
∣ R ∥ ∣ ≤ c ∥ T m ϵ t 1 / 2 ( φ X s ) ∥ L 2 ( ∥ T m ϵ t 1 / 2 ( ∂ s ∣ W s s ∣ 2 X s ) ∥ L 2 + ∥ T m ϵ t 1 / 2 ( ∂ s ( W s s ⋅ Y s s ) X s ) ∥ L 2 + ∥ X s Y s s ⋅ Y s s s ∥ L 2 ) ≤ c ∥ T m ϵ t 1 / 2 ( φ X s ) ∥ L 2 ( ∣ log ϵ ∣ 1 / 2 ∥ W s s 2 ∥ H ˙ 1 + ∣ log ϵ ∣ 1 / 2 ∥ W s s ⋅ Y s s ∥ H ˙ 1 + ∥ Y s s 2 ∥ H ˙ 1 ) ≤ c ∥ T m ϵ t 1 / 2 ( φ X s ) ∥ L 2 ( ∣ log ϵ ∣ 1 / 2 ∥ W s s s ∥ H ˙ 1 / 2 ( ∥ W s ∥ H 1 + ∥ Y s ∥ H 3 ) + ∣ log ϵ ∣ 1 / 2 ∥ W s ∥ H 1 ∥ Y s ∥ H 3 + ∥ Y s s ∥ H 1 2 ) ≤ c ∥ T m ϵ t 1 / 2 ( φ X s ) ∥ L 2 ( ∥ T m ϵ t 1 / 2 W s s s s ∥ L 2 ( ∥ W s ∥ H 1 + ∥ Y s ∥ H 3 ) + ∥ Y s s ∥ H 1 2 + ∣ log ϵ ∣ 1 / 2 ∥ W s ∥ H 1 ∥ Y s ∥ H 3 ) . (166) \begin{aligned}
\left|R^{\parallel}\right| &\leq c\left\|T_{m_\epsilon^t}^{1/2}(\varphi X_s)\right\|_{L^2}\left(\left\|T_{m_\epsilon^t}^{1/2}(\partial_s\left|W_{ss}\right|^2X_s)\right\|_{L^2}+\left\|T_{m_\epsilon^t}^{1/2}(\partial_s(W_{ss}\cdot Y_{ss})X_s)\right\|_{L^2}\right.\\
&\qquad\left.+\left\|X_sY_{ss}\cdot Y_{sss}\right\|_{L^2}\right)\\
&\leq c\left\|T_{m_\epsilon^t}^{1/2}(\varphi X_s)\right\|_{L^2}\left(\left|\log\epsilon\right|^{1/2}\left\|W_{ss}^2\right\|_{\dot H^1}+\left|\log\epsilon\right|^{1/2}\left\|W_{ss}\cdot Y_{ss}\right\|_{\dot H^1}+\left\|Y_{ss}^2\right\|_{\dot H^1}\right)\\
&\leq c\left\|T_{m_\epsilon^t}^{1/2}(\varphi X_s)\right\|_{L^2}\left(\left|\log\epsilon\right|^{1/2}\left\|W_{sss}\right\|_{\dot H^{1/2}}\left(\left\|W_s\right\|_{H^1}+\left\|Y_s\right\|_{H^3}\right)\right.\\
&\qquad\left.+\left|\log\epsilon\right|^{1/2}\left\|W_s\right\|_{H^1}\left\|Y_s\right\|_{H^3}+\left\|Y_{ss}\right\|_{H^1}^2\right)\\
&\leq c\left\|T_{m_\epsilon^t}^{1/2}(\varphi X_s)\right\|_{L^2}\left(\left\|T_{m_\epsilon^t}^{1/2}W_{ssss}\right\|_{L^2}\left(\left\|W_s\right\|_{H^1}+\left\|Y_s\right\|_{H^3}\right)+\left\|Y_{ss}\right\|_{H^1}^2\right.\\
&\qquad\left.+\left|\log\epsilon\right|^{1/2}\left\|W_s\right\|_{H^1}\left\|Y_s\right\|_{H^3}\right). \tag*{(166)}
\end{aligned} R ∥ ≤ c T m ϵ t 1/2 ( φ X s ) L 2 ( T m ϵ t 1/2 ( ∂ s ∣ W ss ∣ 2 X s ) L 2 + T m ϵ t 1/2 ( ∂ s ( W ss ⋅ Y ss ) X s ) L 2 + ∥ X s Y ss ⋅ Y sss ∥ L 2 ) ≤ c T m ϵ t 1/2 ( φ X s ) L 2 ( ∣ log ϵ ∣ 1/2 W ss 2 H ˙ 1 + ∣ log ϵ ∣ 1/2 ∥ W ss ⋅ Y ss ∥ H ˙ 1 + Y ss 2 H ˙ 1 ) ≤ c T m ϵ t 1/2 ( φ X s ) L 2 ( ∣ log ϵ ∣ 1/2 ∥ W sss ∥ H ˙ 1/2 ( ∥ W s ∥ H 1 + ∥ Y s ∥ H 3 ) + ∣ log ϵ ∣ 1/2 ∥ W s ∥ H 1 ∥ Y s ∥ H 3 + ∥ Y ss ∥ H 1 2 ) ≤ c T m ϵ t 1/2 ( φ X s ) L 2 ( T m ϵ t 1/2 W ssss L 2 ( ∥ W s ∥ H 1 + ∥ Y s ∥ H 3 ) + ∥ Y ss ∥ H 1 2 + ∣ log ϵ ∣ 1/2 ∥ W s ∥ H 1 ∥ Y s ∥ H 3 ) . (166)
Here, as in the estimate (163) for LHS, we have opted to forego the regularity gain from T m ϵ t 1 / 2 T_{m_\epsilon^t}^{1/2} T m ϵ t 1/2 in order to avoid an additional factor of ϵ − 1 / 2 \epsilon^{-1/2} ϵ − 1/2 from the high wavenumber behavior of m ϵ t ( k ) m_\epsilon^t(k) m ϵ t ( k ) . Furthermore, in the last line, we have used that
∥ W s s s ∥ H ˙ 1 / 2 = ∥ T m ϵ t − 1 / 2 T m ϵ t 1 / 2 W s s s ∥ H ˙ 1 / 2 ≤ c ∣ log ϵ ∣ − 1 / 2 ∥ T m ϵ t 1 / 2 W s s s ∥ H ˙ 1 , (167) \left\|W_{sss}\right\|_{\dot H^{1/2}}=\left\|T_{m_\epsilon^t}^{-1/2}T_{m_\epsilon^t}^{1/2}W_{sss}\right\|_{\dot H^{1/2}}\leq c\left|\log\epsilon\right|^{-1/2}\left\|T_{m_\epsilon^t}^{1/2}W_{sss}\right\|_{\dot H^1}, \tag*{(167)} ∥ W sss ∥ H ˙ 1/2 = T m ϵ t − 1/2 T m ϵ t 1/2 W sss H ˙ 1/2 ≤ c ∣ log ϵ ∣ − 1/2 T m ϵ t 1/2 W sss H ˙ 1 , (167)
by Lemma 4.1 .
We may similarly write R ⊥ R^{\perp} R ⊥ as
R ⊥ = ∫ T ( X s s ⋅ T ϵ m n ( X s s s s + 3 X s s ⋅ X s s s X s ) − ∣ log ϵ ∣ 4 π ( Y s s s s + 3 Y s s ⋅ Y s s s Y s ) ⋅ Y s s ) φ d s = ∫ T φ X s s ⋅ T ϵ m n ( W s s s s + 3 X s s ⋅ W s s s X s + 3 W s s ⋅ Y s s s X s + 3 Y s s ⋅ Y s s s W s ) d s + ∫ T φ ( X s s ⋅ ( T ϵ m n − ∣ log ϵ ∣ 4 π ) + ∣ log ϵ ∣ 4 π W s s ⋅ ) [ Y s s s s + 3 Y s s ⋅ Y s s s Y s ] d s . (168) \begin{aligned}
R^{\perp}={}&\int_{\mathbb{T}}\left(X_{ss}\cdot T_\epsilon^{mn}(X_{ssss}+3X_{ss}\cdot X_{sss}X_s)-\frac{\left|\log\epsilon\right|}{4\pi}(Y_{ssss}+3Y_{ss}\cdot Y_{sss}Y_s)\cdot Y_{ss}\right)\varphi\,ds\\
={}&\int_{\mathbb{T}}\varphi X_{ss}\cdot T_\epsilon^{mn}(W_{ssss}+3X_{ss}\cdot W_{sss}X_s+3W_{ss}\cdot Y_{sss}X_s+3Y_{ss}\cdot Y_{sss}W_s)\,ds\\
&+\int_{\mathbb{T}}\varphi\left(X_{ss}\cdot\left(T_\epsilon^{mn}-\frac{\left|\log\epsilon\right|}{4\pi}\right)+\frac{\left|\log\epsilon\right|}{4\pi}W_{ss}\cdot\right)[Y_{ssss}+3Y_{ss}\cdot Y_{sss}Y_s]\,ds. \tag*{(168)}
\end{aligned} R ⊥ = = ∫ T ( X ss ⋅ T ϵ mn ( X ssss + 3 X ss ⋅ X sss X s ) − 4 π ∣ log ϵ ∣ ( Y ssss + 3 Y ss ⋅ Y sss Y s ) ⋅ Y ss ) φ d s ∫ T φ X ss ⋅ T ϵ mn ( W ssss + 3 X ss ⋅ W sss X s + 3 W ss ⋅ Y sss X s + 3 Y ss ⋅ Y sss W s ) d s + ∫ T φ ( X ss ⋅ ( T ϵ mn − 4 π ∣ log ϵ ∣ ) + 4 π ∣ log ϵ ∣ W ss ⋅ ) [ Y ssss + 3 Y ss ⋅ Y sss Y s ] d s . (168)
Again using Lemma 4.2 and the inequalities (3.2), we may estimate
∣ R ∥ ∣ ≤ c ∥ T m ϵ n 1 / 2 ( φ X s s ) ∥ L 2 ( ∥ T m ϵ n 1 / 2 W s s s s ∥ L 2 + ∥ T m ϵ n 1 / 2 ( ∂ s ∣ W s s ∣ 2 X s ) ∥ L 2 + ∥ T m ϵ n 1 / 2 ( ∂ s ( W s s ⋅ Y s s ) X s ) ∥ L 2 + ∥ T m ϵ n 1 / 2 ( ∂ s ∣ Y s s ∣ 2 W s ) ∥ L 2 ) + c ( ∥ T m ϵ n 1 / 2 ( φ X s s ) ∥ L 2 + ∣ log ϵ ∣ ∥ φ ∥ L 2 ∥ W s s ∥ L ∞ ) ( ∥ Y s ∥ H 3 + ∥ Y s s ∥ H 1 2 ) ≤ c ∥ T m ϵ n 1 / 2 ( φ X s s ) ∥ L 2 ( ∥ T m ϵ n 1 / 2 W s s s s ∥ L 2 ( 1 + ∥ W s ∥ H 1 + ∥ Y s ∥ H 3 ) + ( 1 + ∣ log ϵ ∣ 1 / 2 ∥ W s ∥ H 1 ) ( ∥ Y s ∥ H 3 + ∥ Y s s ∥ H 1 2 ) ) + c ∣ log ϵ ∣ ∥ T m ϵ n − 1 T m ϵ n ( φ X s ) ⋅ X s ∥ L 2 ∥ W s s s ∥ H ˙ 1 / 2 2 / 3 ∥ W s ∥ H 1 1 / 3 ( ∥ Y s ∥ H 3 + ∥ Y s s ∥ H 1 2 ) ≤ c ∥ T m ϵ n 1 / 2 ( φ X s s ) ∥ L 2 ( ∥ T m ϵ n 1 / 2 W s s s s ∥ L 2 ( 1 + ∥ W s ∥ H 1 + ∥ Y s ∥ H 3 ) + ( 1 + ∣ log ϵ ∣ 1 / 2 ∥ W s ∥ H 1 ) ( ∥ Y s ∥ H 3 + ∥ Y s s ∥ H 1 2 ) ) + c ∣ log ϵ ∣ 1 / 2 ∥ T m ϵ n 1 / 2 ( φ X s ) s ∥ L 2 ∥ W s s s ∥ H ˙ 1 / 2 2 / 3 ∥ W s ∥ H 1 1 / 3 ( ∥ Y s ∥ H 3 + ∥ Y s s ∥ H 1 2 ) . (169) \begin{aligned}
\lvert R^{\parallel}\rvert&\le c\left\|T_{m_{\epsilon}^{n}}^{1/2}(\varphi X_{ss})\right\|_{L^{2}}\left(\left\|T_{m_{\epsilon}^{n}}^{1/2}W_{ssss}\right\|_{L^{2}}+\left\|T_{m_{\epsilon}^{n}}^{1/2}(\partial_{s}\lvert W_{ss}\rvert^{2}X_{s})\right\|_{L^{2}}\right.\\
&\qquad\left.+\left\|T_{m_{\epsilon}^{n}}^{1/2}(\partial_{s}(W_{ss}\cdot Y_{ss})X_{s})\right\|_{L^{2}}+\left\|T_{m_{\epsilon}^{n}}^{1/2}(\partial_{s}\lvert Y_{ss}\rvert^{2}W_{s})\right\|_{L^{2}}\right)\\
&\qquad+c\left(\left\|T_{m_{\epsilon}^{n}}^{1/2}(\varphi X_{ss})\right\|_{L^{2}}+\lvert\log\epsilon\rvert\|\varphi\|_{L^{2}}\|W_{ss}\|_{L^{\infty}}\right)\left(\|Y_{s}\|_{H^{3}}+\|Y_{ss}\|_{H^{1}}^{2}\right)\\
&\le c\left\|T_{m_{\epsilon}^{n}}^{1/2}(\varphi X_{ss})\right\|_{L^{2}}\left(\left\|T_{m_{\epsilon}^{n}}^{1/2}W_{ssss}\right\|_{L^{2}}\left(1+\|W_{s}\|_{H^{1}}+\|Y_{s}\|_{H^{3}}\right)\right.\\
&\qquad\left.+\left(1+\lvert\log\epsilon\rvert^{1/2}\|W_{s}\|_{H^{1}}\right)\left(\|Y_{s}\|_{H^{3}}+\|Y_{ss}\|_{H^{1}}^{2}\right)\right)\\
&\qquad+c\lvert\log\epsilon\rvert\left\|T_{m_{\epsilon}^{n}}^{-1}T_{m_{\epsilon}^{n}}(\varphi X_{s})\cdot X_{s}\right\|_{L^{2}}\|W_{sss}\|_{\dot H^{1/2}}^{2/3}\|W_{s}\|_{H^{1}}^{1/3}\left(\|Y_{s}\|_{H^{3}}+\|Y_{ss}\|_{H^{1}}^{2}\right)\\
&\le c\left\|T_{m_{\epsilon}^{n}}^{1/2}(\varphi X_{ss})\right\|_{L^{2}}\left(\left\|T_{m_{\epsilon}^{n}}^{1/2}W_{ssss}\right\|_{L^{2}}\left(1+\|W_{s}\|_{H^{1}}+\|Y_{s}\|_{H^{3}}\right)\right.\\
&\qquad\left.+\left(1+\lvert\log\epsilon\rvert^{1/2}\|W_{s}\|_{H^{1}}\right)\left(\|Y_{s}\|_{H^{3}}+\|Y_{ss}\|_{H^{1}}^{2}\right)\right)\\
&\qquad+c\lvert\log\epsilon\rvert^{1/2}\left\|T_{m_{\epsilon}^{n}}^{1/2}(\varphi X_{s})_{s}\right\|_{L^{2}}\|W_{sss}\|_{\dot H^{1/2}}^{2/3}\|W_{s}\|_{H^{1}}^{1/3}\left(\|Y_{s}\|_{H^{3}}+\|Y_{ss}\|_{H^{1}}^{2}\right). \tag*{(169)}
\end{aligned} ∣ R ∥ ∣ ≤ c T m ϵ n 1/2 ( φ X ss ) L 2 ( T m ϵ n 1/2 W ssss L 2 + T m ϵ n 1/2 ( ∂ s ∣ W ss ∣ 2 X s ) L 2 + T m ϵ n 1/2 ( ∂ s ( W ss ⋅ Y ss ) X s ) L 2 + T m ϵ n 1/2 ( ∂ s ∣ Y ss ∣ 2 W s ) L 2 ) + c ( T m ϵ n 1/2 ( φ X ss ) L 2 + ∣ log ϵ ∣ ∥ φ ∥ L 2 ∥ W ss ∥ L ∞ ) ( ∥ Y s ∥ H 3 + ∥ Y ss ∥ H 1 2 ) ≤ c T m ϵ n 1/2 ( φ X ss ) L 2 ( T m ϵ n 1/2 W ssss L 2 ( 1 + ∥ W s ∥ H 1 + ∥ Y s ∥ H 3 ) + ( 1 + ∣ log ϵ ∣ 1/2 ∥ W s ∥ H 1 ) ( ∥ Y s ∥ H 3 + ∥ Y ss ∥ H 1 2 ) ) + c ∣ log ϵ ∣ T m ϵ n − 1 T m ϵ n ( φ X s ) ⋅ X s L 2 ∥ W sss ∥ H ˙ 1/2 2/3 ∥ W s ∥ H 1 1/3 ( ∥ Y s ∥ H 3 + ∥ Y ss ∥ H 1 2 ) ≤ c T m ϵ n 1/2 ( φ X ss ) L 2 ( T m ϵ n 1/2 W ssss L 2 ( 1 + ∥ W s ∥ H 1 + ∥ Y s ∥ H 3 ) + ( 1 + ∣ log ϵ ∣ 1/2 ∥ W s ∥ H 1 ) ( ∥ Y s ∥ H 3 + ∥ Y ss ∥ H 1 2 ) ) + c ∣ log ϵ ∣ 1/2 T m ϵ n 1/2 ( φ X s ) s L 2 ∥ W sss ∥ H ˙ 1/2 2/3 ∥ W s ∥ H 1 1/3 ( ∥ Y s ∥ H 3 + ∥ Y ss ∥ H 1 2 ) . (169)
Here we have reused some of the estimates from the LHS bound (4.27) and the R ∥ R^{\parallel} R ∥ bound (4.30).
Altogether, taking φ = τ W \varphi=\tau W φ = τ W in (4.22) and combining the bounds for LHS and RHS, we may estimate
B ϵ X ( τ W , τ W ) ≤ ∣ L H S ∣ + ∣ R H S ∣ ≤ c ( ∥ T m ϵ t 1 / 2 ( ( τ W ) s X s ) ∥ L 2 + ∥ T m ϵ n 1 / 2 ( τ W X s s ) ∥ L 2 ) × ( ∣ log ϵ ∣ 1 / 2 ∥ W s ∥ H 1 ( ∥ Y s ∥ H 3 + ∥ Y s ∥ H 2 2 ) + ∣ log ϵ ∣ 1 / 2 ∥ W s ∥ H 1 ∥ W s s s ∥ H ˙ 1 / 2 2 / 3 ( ∥ Y s ∥ H 3 + ∥ Y s s ∥ H 1 2 ) + ∥ T m ϵ t 1 / 2 W s s s s ∥ L 2 ( 1 + ∥ W s ∥ H 1 + ∥ Y s ∥ H 3 ) + ( ∥ Y s ∥ H 3 + ∥ Y s s ∥ H 1 2 ) ( 1 + ∥ X s ∥ H 1 2 + ∥ Y s s ∥ H 1 2 ) ) ≤ c ( ∥ T m ϵ t 1 / 2 ( ( τ W ) s X s ) ∥ L 2 + ∥ T m ϵ n 1 / 2 ( τ W X s s ) ∥ L 2 ) × ( ∣ log ϵ ∣ 1 / 2 ∥ W s ∥ H 1 ( ∥ Y s ∥ H 3 + ∥ Y s ∥ H 2 2 ) + ∥ T m ϵ t 1 / 2 W s s s s ∥ L 2 ( 1 + ∥ X s ∥ H 1 + ∥ Y s ∥ H 3 + ∥ Y s s ∥ H 1 2 ) + ( ∥ Y s ∥ H 3 + ∥ Y s s ∥ H 1 2 ) ( 1 + ∥ X s ∥ H 1 2 + ∥ Y s s ∥ H 1 2 ) ) . (170) \begin{aligned}
B_{\epsilon}^{X}(\tau W,\tau W)&\le\lvert\mathrm{LHS}\rvert+\lvert\mathrm{RHS}\rvert\\
&\le c\left(\left\|T_{m_{\epsilon}^{t}}^{1/2}((\tau W)_{s}X_{s})\right\|_{L^{2}}+\left\|T_{m_{\epsilon}^{n}}^{1/2}(\tau W X_{ss})\right\|_{L^{2}}\right)\\
&\quad\times\left(\lvert\log\epsilon\rvert^{1/2}\|W_{s}\|_{H^{1}}\left(\|Y_{s}\|_{H^{3}}+\|Y_{s}\|_{H^{2}}^{2}\right)\right.\\
&\qquad+\lvert\log\epsilon\rvert^{1/2}\|W_{s}\|_{H^{1}}\|W_{sss}\|_{\dot H^{1/2}}^{2/3}\left(\|Y_{s}\|_{H^{3}}+\|Y_{ss}\|_{H^{1}}^{2}\right)\\
&\qquad+\left\|T_{m_{\epsilon}^{t}}^{1/2}W_{ssss}\right\|_{L^{2}}\left(1+\|W_{s}\|_{H^{1}}+\|Y_{s}\|_{H^{3}}\right)\\
&\qquad\left.+\left(\|Y_{s}\|_{H^{3}}+\|Y_{ss}\|_{H^{1}}^{2}\right)\left(1+\|X_{s}\|_{H^{1}}^{2}+\|Y_{ss}\|_{H^{1}}^{2}\right)\right)\\
&\le c\left(\left\|T_{m_{\epsilon}^{t}}^{1/2}((\tau W)_{s}X_{s})\right\|_{L^{2}}+\left\|T_{m_{\epsilon}^{n}}^{1/2}(\tau W X_{ss})\right\|_{L^{2}}\right)\\
&\quad\times\left(\lvert\log\epsilon\rvert^{1/2}\|W_{s}\|_{H^{1}}\left(\|Y_{s}\|_{H^{3}}+\|Y_{s}\|_{H^{2}}^{2}\right)\right.\\
&\qquad+\left\|T_{m_{\epsilon}^{t}}^{1/2}W_{ssss}\right\|_{L^{2}}\left(1+\|X_{s}\|_{H^{1}}+\|Y_{s}\|_{H^{3}}+\|Y_{ss}\|_{H^{1}}^{2}\right)\\
&\qquad\left.+\left(\|Y_{s}\|_{H^{3}}+\|Y_{ss}\|_{H^{1}}^{2}\right)\left(1+\|X_{s}\|_{H^{1}}^{2}+\|Y_{ss}\|_{H^{1}}^{2}\right)\right). \tag*{(170)}
\end{aligned} B ϵ X ( τ W , τ W ) ≤ ∣ LHS ∣ + ∣ RHS ∣ ≤ c ( T m ϵ t 1/2 (( τ W ) s X s ) L 2 + T m ϵ n 1/2 ( τ W X ss ) L 2 ) × ( ∣ log ϵ ∣ 1/2 ∥ W s ∥ H 1 ( ∥ Y s ∥ H 3 + ∥ Y s ∥ H 2 2 ) + ∣ log ϵ ∣ 1/2 ∥ W s ∥ H 1 ∥ W sss ∥ H ˙ 1/2 2/3 ( ∥ Y s ∥ H 3 + ∥ Y ss ∥ H 1 2 ) + T m ϵ t 1/2 W ssss L 2 ( 1 + ∥ W s ∥ H 1 + ∥ Y s ∥ H 3 ) + ( ∥ Y s ∥ H 3 + ∥ Y ss ∥ H 1 2 ) ( 1 + ∥ X s ∥ H 1 2 + ∥ Y ss ∥ H 1 2 ) ) ≤ c ( T m ϵ t 1/2 (( τ W ) s X s ) L 2 + T m ϵ n 1/2 ( τ W X ss ) L 2 ) × ( ∣ log ϵ ∣ 1/2 ∥ W s ∥ H 1 ( ∥ Y s ∥ H 3 + ∥ Y s ∥ H 2 2 ) + T m ϵ t 1/2 W ssss L 2 ( 1 + ∥ X s ∥ H 1 + ∥ Y s ∥ H 3 + ∥ Y ss ∥ H 1 2 ) + ( ∥ Y s ∥ H 3 + ∥ Y ss ∥ H 1 2 ) ( 1 + ∥ X s ∥ H 1 2 + ∥ Y ss ∥ H 1 2 ) ) . (170)
Here, to simplify, we have used that m ϵ t ( k ) ≲ m ϵ n ( k ) ≲ m ϵ t ( k ) m_{\epsilon}^{t}(k)\lesssim m_{\epsilon}^{n}(k)\lesssim m_{\epsilon}^{t}(k) m ϵ t ( k ) ≲ m ϵ n ( k ) ≲ m ϵ t ( k ) , so we may interchange T m ϵ t T_{m_{\epsilon}^{t}} T m ϵ t and T m ϵ n T_{m_{\epsilon}^{n}} T m ϵ n . In addition, we have replaced one instance of ∥ W s ∥ H 1 ≤ ∥ X s ∥ H 1 + ∥ Y s ∥ H 1 \|W_{s}\|_{H^{1}}\le\|X_{s}\|_{H^{1}}+\|Y_{s}\|_{H^{1}} ∥ W s ∥ H 1 ≤ ∥ X s ∥ H 1 + ∥ Y s ∥ H 1 , and used that
∥ W s s s ∥ H ˙ 1 / 2 2 / 3 ∥ W s ∥ H 1 1 / 3 ≤ c ∣ log ϵ ∣ − 1 / 3 ∥ T m ϵ t W s s s ∥ L 2 2 / 3 ∥ W s ∥ H 1 1 / 3 ≤ c ( ∣ log ϵ ∣ − 1 / 2 ∥ T m ϵ t W s s s s ∥ L 2 + ∥ W s ∥ H 1 ) . (171) \begin{aligned}
\|W_{sss}\|_{\dot H^{1/2}}^{2/3}\|W_{s}\|_{H^{1}}^{1/3}
&\le c\lvert\log\epsilon\rvert^{-1/3}\|T_{m_{\epsilon}^{t}}W_{sss}\|_{L^{2}}^{2/3}\|W_{s}\|_{H^{1}}^{1/3}\\
&\le c\left(\lvert\log\epsilon\rvert^{-1/2}\|T_{m_{\epsilon}^{t}}W_{ssss}\|_{L^{2}}+\|W_{s}\|_{H^{1}}\right). \tag*{(171)}
\end{aligned} ∥ W sss ∥ H ˙ 1/2 2/3 ∥ W s ∥ H 1 1/3 ≤ c ∣ log ϵ ∣ − 1/3 ∥ T m ϵ t W sss ∥ L 2 2/3 ∥ W s ∥ H 1 1/3 ≤ c ( ∣ log ϵ ∣ − 1/2 ∥ T m ϵ t W ssss ∥ L 2 + ∥ W s ∥ H 1 ) . (171)
Finally, using that
B ϵ X ( τ W , τ W ) = ∥ T m ϵ t 1 / 2 ( ( τ W ) s X s ) ∥ L 2 2 + ∥ T m ϵ n 1 / 2 ( τ W X s s ) ∥ L 2 2 , (172) B_{\epsilon}^{X}(\tau W,\tau W)=\left\|T_{m_{\epsilon}^{t}}^{1/2}((\tau W)_{s}X_{s})\right\|_{L^{2}}^{2}+\left\|T_{m_{\epsilon}^{n}}^{1/2}(\tau W X_{ss})\right\|_{L^{2}}^{2}, \tag*{(172)} B ϵ X ( τ W , τ W ) = T m ϵ t 1/2 (( τ W ) s X s ) L 2 2 + T m ϵ n 1/2 ( τ W X ss ) L 2 2 , (172)
applying Young’s inequality to (4.34) then yields (4.20). □
Energy bound for curve difference
Using the results of sections 4.2 and 4.3, we may proceed to estimate the terms A ∥ A^{\parallel} A ∥ and A ⊥ A^{\perp} A ⊥ appearing on the right hand side of the energy identity (4.6).
We may first apply Corollary 4.3 to estimate A ∥ A^{\parallel} A ∥ as
∣ A ∥ ∣ ≤ ∥ T m ϵ 1 / 2 ( P X s W s s s s ) ∥ L 2 ( ∥ T m ϵ 1 / 2 ( ( τ W ) s X s ) ∥ L 2 + ∥ T m ϵ 1 / 2 ( P X s ( τ Y W s ) s ) ∥ L 2 ) + c ∥ T m ϵ 1 / 2 ( P X s W s s s s ) ∥ L 2 ∥ P X s ( Y s s s − τ Y Y s ) ∥ L 2 + c ∣ log ϵ ∣ 1 / 2 ∥ W s ∥ H 1 ( ∥ X s ∥ H 1 2 + ∥ Y s ∥ H 1 2 ) ∥ T m ϵ 1 / 2 W s s s s ∥ L 2 ∥ Y s s s − τ Y Y s ∥ H 3 / 2 . (173) \begin{aligned}
\left|A^{\parallel}\right| \leq{}& \left\|T_{m_\epsilon}^{1/2}(\mathbb{P}_{X_s}W_{ssss})\right\|_{L^2}
\left(\left\|T_{m_\epsilon}^{1/2}((\tau W)_sX_s)\right\|_{L^2}
+\left\|T_{m_\epsilon}^{1/2}(\mathbb{P}_{X_s}(\tau_YW_s)_s)\right\|_{L^2}\right) \\
&+c\left\|T_{m_\epsilon}^{1/2}(\mathbb{P}_{X_s}W_{ssss})\right\|_{L^2}
\left\|\mathbb{P}_{X_s}(Y_{sss}-\tau_YY_s)\right\|_{L^2} \\
&+c|\log\epsilon|^{1/2}\|W_s\|_{H^1}
\left(\|X_s\|_{H^1}^2+\|Y_s\|_{H^1}^2\right)
\left\|T_{m_\epsilon}^{1/2}W_{ssss}\right\|_{L^2}
\|Y_{sss}-\tau_YY_s\|_{H^{3/2}} . \tag*{(173)}
\end{aligned} A ∥ ≤ T m ϵ 1/2 ( P X s W ssss ) L 2 ( T m ϵ 1/2 (( τ W ) s X s ) L 2 + T m ϵ 1/2 ( P X s ( τ Y W s ) s ) L 2 ) + c T m ϵ 1/2 ( P X s W ssss ) L 2 ∥ P X s ( Y sss − τ Y Y s ) ∥ L 2 + c ∣ log ϵ ∣ 1/2 ∥ W s ∥ H 1 ( ∥ X s ∥ H 1 2 + ∥ Y s ∥ H 1 2 ) T m ϵ 1/2 W ssss L 2 ∥ Y sss − τ Y Y s ∥ H 3/2 . (173)
and A ⊥ A^{\perp} A ⊥ as
∣ A ⊥ ∣ ≤ ∥ T m ϵ 1 / 2 ( P X s ⊥ W s s s s ) ∥ L 2 ( ∥ T m ϵ 1 / 2 ( τ W s s ) ∥ L 2 + ∥ T m ϵ 1 / 2 ( P X s ⊥ ( τ Y W s ) s ) ∥ L 2 ) + c ∥ T m ϵ 1 / 2 ( P X s ⊥ W s s s s ) ∥ L 2 ∥ P X s ⊥ ( Y s s s − τ Y Y s ) ∥ L 2 + c ∣ log ϵ ∣ 1 / 2 ∥ W s ∥ H 1 ( 1 + ∥ X s ∥ H 1 2 + ∥ Y s ∥ H 1 2 ) ∥ T m ϵ 1 / 2 W s s s s ∥ L 2 ∥ Y s s s − τ Y Y s ∥ H 3 / 2 . (174) \begin{aligned}
\left|A^{\perp}\right| \leq{}& \left\|T_{m_\epsilon}^{1/2}(\mathbb{P}^{\perp}_{X_s}W_{ssss})\right\|_{L^2}
\left(\left\|T_{m_\epsilon}^{1/2}(\tau W_{ss})\right\|_{L^2}
+\left\|T_{m_\epsilon}^{1/2}(\mathbb{P}^{\perp}_{X_s}(\tau_YW_s)_s)\right\|_{L^2}\right) \\
&+c\left\|T_{m_\epsilon}^{1/2}(\mathbb{P}^{\perp}_{X_s}W_{ssss})\right\|_{L^2}
\left\|\mathbb{P}^{\perp}_{X_s}(Y_{sss}-\tau_YY_s)\right\|_{L^2} \\
&+c|\log\epsilon|^{1/2}\|W_s\|_{H^1}
\left(1+\|X_s\|_{H^1}^2+\|Y_s\|_{H^1}^2\right)
\left\|T_{m_\epsilon}^{1/2}W_{ssss}\right\|_{L^2}
\|Y_{sss}-\tau_YY_s\|_{H^{3/2}} . \tag*{(174)}
\end{aligned} A ⊥ ≤ T m ϵ 1/2 ( P X s ⊥ W ssss ) L 2 ( T m ϵ 1/2 ( τ W ss ) L 2 + T m ϵ 1/2 ( P X s ⊥ ( τ Y W s ) s ) L 2 ) + c T m ϵ 1/2 ( P X s ⊥ W ssss ) L 2 P X s ⊥ ( Y sss − τ Y Y s ) L 2 + c ∣ log ϵ ∣ 1/2 ∥ W s ∥ H 1 ( 1 + ∥ X s ∥ H 1 2 + ∥ Y s ∥ H 1 2 ) T m ϵ 1/2 W ssss L 2 ∥ Y sss − τ Y Y s ∥ H 3/2 . (174)
Together, using Lemma 4.4 and the bound (4.19) for the tension terms involving τ Y \tau_Y τ Y , we obtain the bound
∣ A ∥ ∣ + ∣ A ⊥ ∣ ≤ c ( ∥ T m ϵ 1 / 2 ( P X s W s s s s ) ∥ L 2 + ∥ T m ϵ 1 / 2 ( P X s ⊥ W s s s s ) ∥ L 2 ) × ( ∣ log ϵ ∣ 1 / 2 ∥ W s ∥ H 1 ( 1 + ∥ X s ∥ H 1 2 + ∥ Y s ∥ H 1 2 ) ( 1 + ∥ Y s ∥ H 3 5 + ∥ Y s ∥ H 7 / 2 ) + ( ∥ T m ϵ 1 / 2 W s s s s ∥ L 2 + 1 + ∥ Y s ∥ H 3 2 ) ( 1 + ∥ X s ∥ H 1 + ∥ Y s ∥ H 3 2 ) ) . (175) \begin{aligned}
|A^{\parallel}|+|A^{\perp}| \leq{}&
c\left(\left\|T_{m_\epsilon}^{1/2}(\mathbb{P}_{X_s}W_{ssss})\right\|_{L^2}
+\left\|T_{m_\epsilon}^{1/2}(\mathbb{P}^{\perp}_{X_s}W_{ssss})\right\|_{L^2}\right) \\
&\quad\times\left(|\log\epsilon|^{1/2}\|W_s\|_{H^1}
(1+\|X_s\|_{H^1}^2+\|Y_s\|_{H^1}^2)
(1+\|Y_s\|_{H^3}^5+\|Y_s\|_{H^{7/2}})\right. \\
&\qquad\left.+
\left(\left\|T_{m_\epsilon}^{1/2}W_{ssss}\right\|_{L^2}
+1+\|Y_s\|_{H^3}^2\right)
(1+\|X_s\|_{H^1}+\|Y_s\|_{H^3}^2)\right). \tag*{(175)}
\end{aligned} ∣ A ∥ ∣ + ∣ A ⊥ ∣ ≤ c ( T m ϵ 1/2 ( P X s W ssss ) L 2 + T m ϵ 1/2 ( P X s ⊥ W ssss ) L 2 ) × ( ∣ log ϵ ∣ 1/2 ∥ W s ∥ H 1 ( 1 + ∥ X s ∥ H 1 2 + ∥ Y s ∥ H 1 2 ) ( 1 + ∥ Y s ∥ H 3 5 + ∥ Y s ∥ H 7/2 ) + ( T m ϵ 1/2 W ssss L 2 + 1 + ∥ Y s ∥ H 3 2 ) ( 1 + ∥ X s ∥ H 1 + ∥ Y s ∥ H 3 2 ) ) . (175)
Now, recalling the form of the energy identity (4.6) under the time rescaling (4.3), we have
∣ log ϵ ∣ ∂ ∂ t ∫ T ∣ W s ∣ 2 d s + ∥ T m ϵ 1 / 2 ( P X s W s s s s ) ∥ L s 2 2 + ∥ T m ϵ 1 / 2 ( P X s ⊥ W s s s s ) ∥ L s 2 2 ≤ ∣ A ∥ ∣ + ∣ A ⊥ ∣ . (176) |\log\epsilon|\frac{\partial}{\partial t}\int_{\mathbb{T}}|W_s|^2\,ds
+\left\|T_{m_\epsilon}^{1/2}(\mathbb{P}_{X_s}W_{ssss})\right\|_{L_s^2}^2
+\left\|T_{m_\epsilon}^{1/2}(\mathbb{P}^{\perp}_{X_s}W_{ssss})\right\|_{L_s^2}^2
\leq|A^{\parallel}|+|A^{\perp}|. \tag*{(176)} ∣ log ϵ ∣ ∂ t ∂ ∫ T ∣ W s ∣ 2 d s + T m ϵ 1/2 ( P X s W ssss ) L s 2 2 + T m ϵ 1/2 ( P X s ⊥ W ssss ) L s 2 2 ≤ ∣ A ∥ ∣ + ∣ A ⊥ ∣. (176)
Applying Young’s inequality to (4.39) thus yields the energy bound
∂ ∂ t ∫ T ∣ W s ∣ 2 d s + ∥ T m ϵ 1 / 2 ( P X s W s s s s ) ∥ L s 2 2 + ∥ T m ϵ 1 / 2 ( P X s ⊥ W s s s s ) ∥ L s 2 2 ≤ c ( ∥ W s ∥ H s 1 2 ( 1 + ∥ X s ∥ H s 1 2 + ∥ Y s ∥ H s 1 2 ) 2 ( 1 + ∥ Y s ∥ H s 3 5 + ∥ Y s ∥ H s 7 / 2 ) 2 + ∣ log ϵ ∣ − 1 ( ∥ T m ϵ 1 / 2 W s s s s ∥ L s 2 2 + ( 1 + ∥ Y s ∥ H s 3 2 ) 2 ) ( 1 + ∥ X s ∥ H s 1 + ∥ Y s ∥ H s 3 2 ) 2 ) . (177) \begin{aligned}
\frac{\partial}{\partial t}\int_{\mathbb{T}}|W_s|^2\,ds
&+\left\|T_{m_\epsilon}^{1/2}(\mathbb{P}_{X_s}W_{ssss})\right\|_{L_s^2}^2
+\left\|T_{m_\epsilon}^{1/2}(\mathbb{P}^{\perp}_{X_s}W_{ssss})\right\|_{L_s^2}^2 \\
&\leq c\left(\|W_s\|_{H_s^1}^2
(1+\|X_s\|_{H_s^1}^2+\|Y_s\|_{H_s^1}^2)^2
(1+\|Y_s\|_{H_s^3}^5+\|Y_s\|_{H_s^{7/2}})^2\right. \\
&\qquad\left.+|\log\epsilon|^{-1}
\left(\left\|T_{m_\epsilon}^{1/2}W_{ssss}\right\|_{L_s^2}^2
+(1+\|Y_s\|_{H_s^3}^2)^2\right)
(1+\|X_s\|_{H_s^1}+\|Y_s\|_{H_s^3}^2)^2\right). \tag*{(177)}
\end{aligned} ∂ t ∂ ∫ T ∣ W s ∣ 2 d s + T m ϵ 1/2 ( P X s W ssss ) L s 2 2 + T m ϵ 1/2 ( P X s ⊥ W ssss ) L s 2 2 ≤ c ( ∥ W s ∥ H s 1 2 ( 1 + ∥ X s ∥ H s 1 2 + ∥ Y s ∥ H s 1 2 ) 2 ( 1 + ∥ Y s ∥ H s 3 5 + ∥ Y s ∥ H s 7/2 ) 2 + ∣ log ϵ ∣ − 1 ( T m ϵ 1/2 W ssss L s 2 2 + ( 1 + ∥ Y s ∥ H s 3 2 ) 2 ) ( 1 + ∥ X s ∥ H s 1 + ∥ Y s ∥ H s 3 2 ) 2 ) . (177)
We may define the energy and dissipation quantities
E W ( t ) = ∥ W s s ∥ L s 2 2 , D W ( t ) = ∥ T m ϵ 1 / 2 ( P X s W s s s s ) ∥ L s 2 2 + ∥ T m ϵ 1 / 2 ( P X s ⊥ W s s s s ) ∥ L s 2 2 , (178) E_W(t)=\|W_{ss}\|_{L_s^2}^2,\qquad
D_W(t)=\left\|T_{m_\epsilon}^{1/2}(\mathbb{P}_{X_s}W_{ssss})\right\|_{L_s^2}^2
+\left\|T_{m_\epsilon}^{1/2}(\mathbb{P}^{\perp}_{X_s}W_{ssss})\right\|_{L_s^2}^2, \tag*{(178)} E W ( t ) = ∥ W ss ∥ L s 2 2 , D W ( t ) = T m ϵ 1/2 ( P X s W ssss ) L s 2 2 + T m ϵ 1/2 ( P X s ⊥ W ssss ) L s 2 2 , (178)
and note that ∥ W s ∥ H s 1 2 ≲ E W ( t ) \|W_s\|_{H_s^1}^2\lesssim E_W(t) ∥ W s ∥ H s 1 2 ≲ E W ( t ) since W s W_s W s has zero mean on T \mathbb{T} T . We further define the quantities
C X Y 1 ( t ) = c ( 1 + ∥ X s ∥ H s 1 2 + ∥ Y s ∥ H s 1 2 ) 2 ( 1 + ∥ Y s ∥ H s 3 5 + ∥ Y s ∥ H s 7 / 2 ) 2 , C X Y 2 ( t ) = c ( 1 + ∥ X s ∥ H s 1 + ∥ Y s ∥ H s 3 2 ) 2 , F X Y ( t ) = c ( 1 + ∥ Y s ∥ H s 3 2 ) 2 ( 1 + ∥ X s ∥ H s 1 + ∥ Y s ∥ H s 3 2 ) 2 . (179) \begin{aligned}
C_{XY1}(t)&=c(1+\|X_s\|_{H_s^1}^2+\|Y_s\|_{H_s^1}^2)^2(1+\|Y_s\|_{H_s^3}^5+\|Y_s\|_{H_s^{7/2}})^2,\\
C_{XY2}(t)&=c(1+\|X_s\|_{H_s^1}+\|Y_s\|_{H_s^3}^2)^2,\\
F_{XY}(t)&=c(1+\|Y_s\|_{H_s^3}^2)^2(1+\|X_s\|_{H_s^1}+\|Y_s\|_{H_s^3}^2)^2. \tag*{(179)}
\end{aligned} C X Y 1 ( t ) C X Y 2 ( t ) F X Y ( t ) = c ( 1 + ∥ X s ∥ H s 1 2 + ∥ Y s ∥ H s 1 2 ) 2 ( 1 + ∥ Y s ∥ H s 3 5 + ∥ Y s ∥ H s 7/2 ) 2 , = c ( 1 + ∥ X s ∥ H s 1 + ∥ Y s ∥ H s 3 2 ) 2 , = c ( 1 + ∥ Y s ∥ H s 3 2 ) 2 ( 1 + ∥ X s ∥ H s 1 + ∥ Y s ∥ H s 3 2 ) 2 . (179)
By (2.9), we have that ∥ X s ∥ H s 1 \left\|X_s\right\|_{H_s^1} ∥ X s ∥ H s 1 is uniformly bounded in time, independent of ϵ \epsilon ϵ . Furthermore, since Y ∈ C t H s 4 ∩ L t , l o c 2 H s 6 Y \in C_tH_s^4 \cap L_{t,\mathrm{loc}}^2H_s^6 Y ∈ C t H s 4 ∩ L t , loc 2 H s 6 , we have C X Y 1 ∈ L t 1 C_{XY1} \in L_t^1 C X Y 1 ∈ L t 1 while C X Y 2 , F X Y ∈ L t ∞ C_{XY2}, F_{XY} \in L_t^\infty C X Y 2 , F X Y ∈ L t ∞ with ϵ \epsilon ϵ -independent bounds, as the time-rescaled evolution for Y Y Y no longer depends on ϵ \epsilon ϵ .
We may thus rewrite (177) as
∂ t E W ( t ) + D W ( t ) ≤ C X Y 1 ( t ) E W ( t ) + ∣ log ϵ ∣ − 1 C X Y 2 ( t ) D W ( t ) + ∣ log ϵ ∣ − 1 F X Y ( t ) . (180) \partial_t E_W(t) + D_W(t) \le C_{XY1}(t)E_W(t) + |\log\epsilon|^{-1}C_{XY2}(t)D_W(t) + |\log\epsilon|^{-1}F_{XY}(t). \tag*{(180)} ∂ t E W ( t ) + D W ( t ) ≤ C X Y 1 ( t ) E W ( t ) + ∣ log ϵ ∣ − 1 C X Y 2 ( t ) D W ( t ) + ∣ log ϵ ∣ − 1 F X Y ( t ) . (180)
Then, given a fixed time interval [ 0 , T ] [0,T] [ 0 , T ] , for ϵ \epsilon ϵ sufficiently small, we may absorb ∣ log ϵ ∣ − 1 C X Y 2 D W ( t ) |\log\epsilon|^{-1}C_{XY2}D_W(t) ∣ log ϵ ∣ − 1 C X Y 2 D W ( t ) into the left hand side terms to obtain
∂ t E W ( t ) + 1 2 D W ( t ) ≤ C X Y 1 ( t ) E W ( t ) + ∣ log ϵ ∣ − 1 F X Y ( t ) , E W ( 0 ) = 0. (181) \partial_t E_W(t) + \frac{1}{2}D_W(t) \le C_{XY1}(t)E_W(t) + |\log\epsilon|^{-1}F_{XY}(t), \qquad E_W(0)=0. \tag*{(181)} ∂ t E W ( t ) + 2 1 D W ( t ) ≤ C X Y 1 ( t ) E W ( t ) + ∣ log ϵ ∣ − 1 F X Y ( t ) , E W ( 0 ) = 0. (181)
For every T < ∞ T < \infty T < ∞ , a Grönwall inequality applied to (181) then yields that there exists C W = C W ( T , ∥ X ∥ L t ∞ H s 2 , ∥ Y ∥ L t ∞ H s 4 ∩ L t 2 H s 6 ) C_W = C_W(T,\left\|X\right\|_{L_t^\infty H_s^2},\left\|Y\right\|_{L_t^\infty H_s^4\cap L_t^2H_s^6}) C W = C W ( T , ∥ X ∥ L t ∞ H s 2 , ∥ Y ∥ L t ∞ H s 4 ∩ L t 2 H s 6 ) independent of ϵ \epsilon ϵ such that
E W ( t ) + ∫ 0 t D W ( t ′ ) d t ′ ≤ ∣ log ϵ ∣ − 1 C W ( T , ∥ X ∥ L t ∞ H s 2 , ∥ Y ∥ L t ∞ H s 4 ∩ L t 2 H s 6 ) . (182) E_W(t) + \int_0^t D_W(t')\,dt' \le|\log\epsilon|^{-1}C_W(T,\left\|X\right\|_{L_t^\infty H_s^2},\left\|Y\right\|_{L_t^\infty H_s^4\cap L_t^2H_s^6}). \tag*{(182)} E W ( t ) + ∫ 0 t D W ( t ′ ) d t ′ ≤ ∣ log ϵ ∣ − 1 C W ( T , ∥ X ∥ L t ∞ H s 2 , ∥ Y ∥ L t ∞ H s 4 ∩ L t 2 H s 6 ) . (182)
In particular, on any fixed time interval, the left hand side goes to zero as ϵ → 0 \epsilon\to0 ϵ → 0 .
It remains to control the mean of W \mathbf{W} W , which we denote by W 0 \mathbf{W}_0 W 0 . Using equation (4.2) under the time rescaling (4.3), we have
∂ W 0 ∂ t = − ∣ log ϵ ∣ − 1 ∫ T L ‾ ϵ ( X ) [ ( X s s s − τ X X s ) s ] d s + 1 4 π ∫ T ( I + Y s ⊗ Y s ) ( Y s s s − τ Y Y s ) s d s = − ∣ log ϵ ∣ − 1 ∫ T L ‾ ϵ ( X ) [ ( W s s s − τ W X s − τ Y W s ) s ] d s ⏟ J 1 − ∣ log ϵ ∣ − 1 ∫ T ( L ‾ ϵ ( X ) − 1 4 π ( I + Y s ⊗ Y s ) ) [ ( Y s s s − τ Y Y s ) s ] d s ⏟ J 2 . (183) \begin{aligned}
\frac{\partial\mathbf{W}_0}{\partial t}
&= -|\log\epsilon|^{-1}\int_{\mathbb{T}}\overline{\mathcal{L}}_\epsilon(X)[(X_{sss}-\tau_XX_s)_s]\,ds
+\frac{1}{4\pi}\int_{\mathbb{T}}(I+Y_s\otimes Y_s)(Y_{sss}-\tau_YY_s)_s\,ds \\
&= -|\log\epsilon|^{-1}\underbrace{\int_{\mathbb{T}}\overline{\mathcal{L}}_\epsilon(X)[(W_{sss}-\tau_WX_s-\tau_YW_s)_s]\,ds}_{J_1} \\
&\quad-|\log\epsilon|^{-1}\underbrace{\int_{\mathbb{T}}\left(\overline{\mathcal{L}}_\epsilon(X)-\frac{1}{4\pi}(I+Y_s\otimes Y_s)\right)[(Y_{sss}-\tau_YY_s)_s]\,ds}_{J_2}. \tag*{(183)}
\end{aligned} ∂ t ∂ W 0 = − ∣ log ϵ ∣ − 1 ∫ T L ϵ ( X ) [( X sss − τ X X s ) s ] d s + 4 π 1 ∫ T ( I + Y s ⊗ Y s ) ( Y sss − τ Y Y s ) s d s = − ∣ log ϵ ∣ − 1 J 1 ∫ T L ϵ ( X ) [( W sss − τ W X s − τ Y W s ) s ] d s − ∣ log ϵ ∣ − 1 J 2 ∫ T ( L ϵ ( X ) − 4 π 1 ( I + Y s ⊗ Y s ) ) [( Y sss − τ Y Y s ) s ] d s . (183)
with W 0 ∣ t = 0 = 0 \left.\mathbf{W}_0\right|_{t=0}=0 W 0 ∣ t = 0 = 0 . Using equation (4.19) and Lemma 4.4, we may estimate J 1 J_1 J 1 as
∣ J 1 ∣ ≤ c ( ∥ T ϵ m ( P X s W s s s s ) ∥ L s 2 + ∥ T ϵ n ( P X s ⊥ W s s s s ) ∥ L s 2 ) + c ( ∥ T ϵ m ( ( τ W ) s X s ) ∥ L s 2 + ∥ T ϵ n ( τ W X s s ) ∥ L s 2 ) + c ( ∥ T ϵ m ( P X s ( τ Y W s ) s ) ∥ L s 2 + ∥ T ϵ n ( P X s ⊥ ( τ Y W s ) s ) ∥ L s 2 ) ≤ c ∣ log ϵ ∣ 1 / 2 D W ( t ) 1 / 2 + c ∣ log ϵ ∣ 1 / 2 ( ∣ log ϵ ∣ 1 / 2 ∥ W s ∥ H 1 ( 1 + ∥ Y s ∥ H 3 ) + ( D W ( t ) 1 / 2 + 1 + ∥ Y s ∥ H 3 ) ( 1 + ∥ X s ∥ H 1 + ∥ Y s ∥ H 3 ) ) + c ∣ log ϵ ∣ ∥ W s ∥ H 1 ( ∥ Y s ∥ H 2 4 + ∥ Y s ∥ H 2 ∥ Y s ∥ H 3 ) ≤ c ∣ log ϵ ∣ 1 / 2 D W ( t ) 1 / 2 + c ∣ log ϵ ∣ E W ( t ) 1 / 2 ( 1 + ∥ Y s ∥ H 3 4 ) + c ∣ log ϵ ∣ 1 / 2 ( 1 + ∥ Y s ∥ H 3 2 ) ( 1 + ∥ X s ∥ H 1 + ∥ Y s ∥ H 3 ) ≤ ∣ log ϵ ∣ 1 / 2 ( c D W ( t ) 1 / 2 + C ( T , ∥ X ∥ L t ∞ H s 2 , ∥ Y ∥ L t ∞ H s 4 ∩ L t 2 H s 6 ) ) , (184) \begin{aligned}
|J_1|
&\le c\left(\left\|T_\epsilon^{m}(\mathbb{P}_{X_s}W_{ssss})\right\|_{L_s^2}
+\left\|T_\epsilon^{n}(\mathbb{P}_{X_s}^{\perp}W_{ssss})\right\|_{L_s^2}\right) \\
&\quad+c\left(\left\|T_\epsilon^{m}((\tau_W)_sX_s)\right\|_{L_s^2}
+\left\|T_\epsilon^{n}(\tau_WX_{ss})\right\|_{L_s^2}\right) \\
&\quad+c\left(\left\|T_\epsilon^{m}(\mathbb{P}_{X_s}(\tau_YW_s)_s)\right\|_{L_s^2}
+\left\|T_\epsilon^{n}(\mathbb{P}_{X_s}^{\perp}(\tau_YW_s)_s)\right\|_{L_s^2}\right) \\
&\le c|\log\epsilon|^{1/2}D_W(t)^{1/2}
+c|\log\epsilon|^{1/2}\left(|\log\epsilon|^{1/2}\left\|W_s\right\|_{H^1}(1+\left\|Y_s\right\|_{H^3})\right. \\
&\qquad\left.+(D_W(t)^{1/2}+1+\left\|Y_s\right\|_{H^3})(1+\left\|X_s\right\|_{H^1}+\left\|Y_s\right\|_{H^3})\right) \\
&\quad+c|\log\epsilon|\left\|W_s\right\|_{H^1}\left(\left\|Y_s\right\|_{H^2}^4+\left\|Y_s\right\|_{H^2}\left\|Y_s\right\|_{H^3}\right) \\
&\le c|\log\epsilon|^{1/2}D_W(t)^{1/2}
+c|\log\epsilon|E_W(t)^{1/2}(1+\left\|Y_s\right\|_{H^3}^4) \\
&\quad+c|\log\epsilon|^{1/2}(1+\left\|Y_s\right\|_{H^3}^2)(1+\left\|X_s\right\|_{H^1}+\left\|Y_s\right\|_{H^3}) \\
&\le|\log\epsilon|^{1/2}\left(cD_W(t)^{1/2}
+C(T,\left\|X\right\|_{L_t^\infty H_s^2},\left\|Y\right\|_{L_t^\infty H_s^4\cap L_t^2H_s^6})\right), \tag*{(184)}
\end{aligned} ∣ J 1 ∣ ≤ c ( ∥ T ϵ m ( P X s W ssss ) ∥ L s 2 + T ϵ n ( P X s ⊥ W ssss ) L s 2 ) + c ( ∥ T ϵ m (( τ W ) s X s ) ∥ L s 2 + ∥ T ϵ n ( τ W X ss ) ∥ L s 2 ) + c ( ∥ T ϵ m ( P X s ( τ Y W s ) s ) ∥ L s 2 + T ϵ n ( P X s ⊥ ( τ Y W s ) s ) L s 2 ) ≤ c ∣ log ϵ ∣ 1/2 D W ( t ) 1/2 + c ∣ log ϵ ∣ 1/2 ( ∣ log ϵ ∣ 1/2 ∥ W s ∥ H 1 ( 1 + ∥ Y s ∥ H 3 ) + ( D W ( t ) 1/2 + 1 + ∥ Y s ∥ H 3 ) ( 1 + ∥ X s ∥ H 1 + ∥ Y s ∥ H 3 ) ) + c ∣ log ϵ ∣ ∥ W s ∥ H 1 ( ∥ Y s ∥ H 2 4 + ∥ Y s ∥ H 2 ∥ Y s ∥ H 3 ) ≤ c ∣ log ϵ ∣ 1/2 D W ( t ) 1/2 + c ∣ log ϵ ∣ E W ( t ) 1/2 ( 1 + ∥ Y s ∥ H 3 4 ) + c ∣ log ϵ ∣ 1/2 ( 1 + ∥ Y s ∥ H 3 2 ) ( 1 + ∥ X s ∥ H 1 + ∥ Y s ∥ H 3 ) ≤ ∣ log ϵ ∣ 1/2 ( c D W ( t ) 1/2 + C ( T , ∥ X ∥ L t ∞ H s 2 , ∥ Y ∥ L t ∞ H s 4 ∩ L t 2 H s 6 ) ) , (184)
using (4.46). Furthermore, using Corollary 4.3 , we may bound J 2 J_2 J 2 as
∣ J 2 ∣ ≤ c ( ∥ T m ϵ 1 / 2 ( X s ⊗ X s ) ∥ L s 2 + ∥ T m ϵ 1 / 2 ( I − X s ⊗ X s ) ∥ L s 2 ) × ( ∥ P X s ( Y s s s − τ Y Y s ) s ∥ L s 2 + ∥ P X s ⊥ ( Y s s s − τ Y Y s ) s ∥ L s 2 ) + ∣ log ϵ ∣ 1 / 2 ∥ W s ∥ H s 1 ( 1 + ∥ X s ∥ H s 1 2 + ∥ Y s ∥ H s 1 2 ) ∥ ( Y s s s − τ Y Y s ) s ∥ H s 1 / 2 ≤ c ∣ log ϵ ∣ 1 / 2 ( 1 + ∥ Y s ∥ H s 3 3 ) + c ∣ log ϵ ∣ E W 1 / 2 ( t ) ( 1 + ∥ X s ∥ H s 1 2 + ∥ Y s ∥ H s 1 2 ) ( 1 + ∥ Y s ∥ H s 3 5 + ∥ Y s ∥ H s 7 / 2 ) ≤ ∣ log ϵ ∣ 1 / 2 C ( T , ∥ X ∥ L t ∞ H s 2 , ∥ Y ∥ L t ∞ H s 4 ∩ L t 2 H s 6 ) . (185) \begin{aligned}
|J_2| &\le c\left(\left\|T_{m_\epsilon}^{1/2}(X_s\otimes X_s)\right\|_{L_s^2}+\left\|T_{m_\epsilon}^{1/2}(I-X_s\otimes X_s)\right\|_{L_s^2}\right) \\
&\quad\times\left(\left\|P_{X_s}(Y_{sss}-\tau_Y Y_s)_s\right\|_{L_s^2}+\left\|P_{X_s}^{\perp}(Y_{sss}-\tau_Y Y_s)_s\right\|_{L_s^2}\right)\\
&\quad+|\log\epsilon|^{1/2}\|W_s\|_{H_s^1}\left(1+\|X_s\|_{H_s^1}^2+\|Y_s\|_{H_s^1}^2\right)\|(Y_{sss}-\tau_Y Y_s)_s\|_{H_s^{1/2}}\\
&\le c|\log\epsilon|^{1/2}\left(1+\|Y_s\|_{H_s^3}^3\right)\\
&\quad+c|\log\epsilon|E_W^{1/2}(t)\left(1+\|X_s\|_{H_s^1}^2+\|Y_s\|_{H_s^1}^2\right)\left(1+\|Y_s\|_{H_s^3}^5+\|Y_s\|_{H_s^{7/2}}\right)\\
&\le|\log\epsilon|^{1/2}C\left(T,\|X\|_{L_t^\infty H_s^2},\|Y\|_{L_t^\infty H_s^4\cap L_t^2H_s^6}\right). \tag*{(185)}
\end{aligned} ∣ J 2 ∣ ≤ c ( T m ϵ 1/2 ( X s ⊗ X s ) L s 2 + T m ϵ 1/2 ( I − X s ⊗ X s ) L s 2 ) × ( ∥ P X s ( Y sss − τ Y Y s ) s ∥ L s 2 + P X s ⊥ ( Y sss − τ Y Y s ) s L s 2 ) + ∣ log ϵ ∣ 1/2 ∥ W s ∥ H s 1 ( 1 + ∥ X s ∥ H s 1 2 + ∥ Y s ∥ H s 1 2 ) ∥ ( Y sss − τ Y Y s ) s ∥ H s 1/2 ≤ c ∣ log ϵ ∣ 1/2 ( 1 + ∥ Y s ∥ H s 3 3 ) + c ∣ log ϵ ∣ E W 1/2 ( t ) ( 1 + ∥ X s ∥ H s 1 2 + ∥ Y s ∥ H s 1 2 ) ( 1 + ∥ Y s ∥ H s 3 5 + ∥ Y s ∥ H s 7/2 ) ≤ ∣ log ϵ ∣ 1/2 C ( T , ∥ X ∥ L t ∞ H s 2 , ∥ Y ∥ L t ∞ H s 4 ∩ L t 2 H s 6 ) . (185)
again by (4.46). In total, we have that W 0 W_0 W 0 satisfies
∂ ∂ t ∣ W 0 ∣ 2 ≤ ∣ log ϵ ∣ − 1 ( c D W ( t ) + C ( T , ∥ X ∥ L t ∞ H s 2 , ∥ Y ∥ L t ∞ H s 4 ∩ L t 2 H s 6 ) ) , (186) \frac{\partial}{\partial t}|W_0|^2\le|\log\epsilon|^{-1}\left(cD_W(t)+C\left(T,\|X\|_{L_t^\infty H_s^2},\|Y\|_{L_t^\infty H_s^4\cap L_t^2H_s^6}\right)\right), \tag*{(186)} ∂ t ∂ ∣ W 0 ∣ 2 ≤ ∣ log ϵ ∣ − 1 ( c D W ( t ) + C ( T , ∥ X ∥ L t ∞ H s 2 , ∥ Y ∥ L t ∞ H s 4 ∩ L t 2 H s 6 ) ) , (186)
with ∣ W 0 ∣ 2 ∣ t = 0 = 0 |W_0|^2\big|_{t=0}=0 ∣ W 0 ∣ 2 t = 0 = 0 . Integrating in time from 0 to T T T , we obtain
∣ W 0 ∣ 2 ≤ ∣ log ϵ ∣ − 1 C ( T , ∥ X ∥ L t ∞ H s 2 , ∥ Y ∥ L t ∞ H s 4 ∩ L t 2 H s 6 ) . (187) |W_0|^2\le|\log\epsilon|^{-1}C\left(T,\|X\|_{L_t^\infty H_s^2},\|Y\|_{L_t^\infty H_s^4\cap L_t^2H_s^6}\right). \tag*{(187)} ∣ W 0 ∣ 2 ≤ ∣ log ϵ ∣ − 1 C ( T , ∥ X ∥ L t ∞ H s 2 , ∥ Y ∥ L t ∞ H s 4 ∩ L t 2 H s 6 ) . (187)
Combining the bounds for E W E_W E W , D W D_W D W , and ∣ W 0 ∣ 2 |W_0|^2 ∣ W 0 ∣ 2 , we obtain
∥ W ∥ L t ∞ H s 2 + ∥ T m ϵ 1 / 2 ( P X s W s s s s ) ∥ L t 2 L s 2 + ∥ T m ϵ 1 / 2 ( P X s W s s s s ) ∥ L t 2 L s 2 ≤ ∣ log ϵ ∣ − 1 / 2 C ( T , ∥ X ∥ L t ∞ H s 2 , ∥ Y ∥ L t ∞ H s 4 ∩ L t 2 H s 6 ) . (188) \begin{aligned}
\|W\|_{L_t^\infty H_s^2}
&+\left\|T_{m_\epsilon}^{1/2}(P_{X_s}W_{ssss})\right\|_{L_t^2L_s^2}
+\left\|T_{m_\epsilon}^{1/2}(P_{X_s}W_{ssss})\right\|_{L_t^2L_s^2}\\
&\le|\log\epsilon|^{-1/2}C\left(T,\|X\|_{L_t^\infty H_s^2},\|Y\|_{L_t^\infty H_s^4\cap L_t^2H_s^6}\right). \tag*{(188)}
\end{aligned} ∥ W ∥ L t ∞ H s 2 + T m ϵ 1/2 ( P X s W ssss ) L t 2 L s 2 + T m ϵ 1/2 ( P X s W ssss ) L t 2 L s 2 ≤ ∣ log ϵ ∣ − 1/2 C ( T , ∥ X ∥ L t ∞ H s 2 , ∥ Y ∥ L t ∞ H s 4 ∩ L t 2 H s 6 ) . (188)
Using the lower bound of Lemma 2.1 for T m ϵ T_{m_\epsilon} T m ϵ and T m ϵ n T_{m_\epsilon}^n T m ϵ n yields Theorem 1.3 . □
Acknowledgments. LO acknowledges support from NSF grant DMS-2406003 and thanks Dallas Albritton for helpful comments on a draft of this paper.
References
Department of Mathematics, University of Wisconsin - Madison, Madison, WI 53706
Email address: lohm2@wisc.edu