1+\lVert\nabla u\rVert_{B^{0}_{\infty,1}}+\lVert\Log C\rVert_{H^{1+\varepsilon}}.
The abstract coefficient class is only a compact-window quotient template. The model consequences are: strong two-dimensional Oldroyd–B solutions continue under ∇ u ∈ L t 1 B ∞ , 1 0 \nabla u\in L^{1}_{t}B^{0}_{\infty,1} ∇ u ∈ L t 1 B ∞ , 1 0 , while strong two-dimensional FENE-P solutions continue under ∇ u ∈ L t 2 B ∞ , 1 0 \nabla u\in L^{2}_{t}B^{0}_{\infty,1} ∇ u ∈ L t 2 B ∞ , 1 0 . In both cases the needed compact window and logarithmic bound are derived from the velocity clock and the model barriers, not assumed as independent hypotheses. All criteria are integer-Sobolev strong-solution criteria; no Leray-type weak-solution or critical-space local theory is asserted. The sharpness examples are static operator obstructions for the pressure-free stress map, not dynamic blow-up constructions. In three dimensions the quotient has the residual split T ( Z ) ∘ = ( q 1 + 2 a q 2 ) Y + q 2 ( Y 2 ) ∘ T(Z)^\circ=(q_1+2aq_2)Y+q_2(Y^2)^\circ T ( Z ) ∘ = ( q 1 + 2 a q 2 ) Y + q 2 ( Y 2 ) ∘ . Since ( Y 2 ) ∘ (Y^2)^\circ ( Y 2 ) ∘ is generally independent of Y Y Y , the exact scalar quotient closure is intrinsically two-dimensional. On prescribed compact windows this residual is absorbed by viscosity; for Oldroyd–B and FENE-P it vanishes because q 2 ≡ 0 q_2\equiv0 q 2 ≡ 0 . No claim is made for general anisotropic or non-spectral stresses.
Introduction
Stress-diffusion-free viscoelastic systems couple parabolic fluid smoothing to purely transported tensor dynamics. The velocity equation is viscous, but the conformation tensor has no spatial diffusion and enters the momentum equation through the divergence of an elastic stress. A continuation criterion must therefore identify which part of the transported tensor can feed derivatives back into the velocity equation. This paper shows that, in two dimensions and for spectral isotropic stresses, the relevant object is not the full elastic stress but its pressure quotient.
An incompressible viscoelastic equation does not use the full elastic stress as an absolute tensor. It uses the stress only modulo pressure: adding an isotropic tensor r ( C ) I r(C)I r ( C ) I changes the scalar pressure but not the projected velocity dynamics. The basic object is therefore the pressure-free active part of the stress, namely the component that survives in vorticity and can exchange derivatives with the non-diffusive conformation tensor. This quotient viewpoint is especially rigid in two space dimensions.
Indeed, let C ∈ S + + 2 C \in\mathbb{S}_{++}^{2} C ∈ S ++ 2 be a positive conformation tensor and let T ( C ) T(C) T ( C ) be a smooth spectral isotropic elastic stress. By the smooth isotropic representation theorem, followed by the two-dimensional Cayley–Hamilton reduction,
T ( C ) = q 0 ( tr C , det C ) I + q 1 ( tr C , det C ) C . T(C) = q_0(\operatorname{tr} C, \det C)I + q_1(\operatorname{tr} C, \det C)C. T ( C ) = q 0 ( tr C , det C ) I + q 1 ( tr C , det C ) C .
Writing C = a I + Y C = aI + Y C = a I + Y , with a = 1 2 tr C a = \frac{1}{2}\operatorname{tr} C a = 2 1 tr C and Y = C ∘ Y = C^\circ Y = C ∘ , gives
T ( C ) = ( q 0 + a q 1 ) I + q 1 Y . T(C) = (q_0 + aq_1)I + q_1Y. T ( C ) = ( q 0 + a q 1 ) I + q 1 Y .
Since det C = a 2 − 1 2 ∣ Y ∣ 2 \det C = a^2 - \frac{1}{2}\lvert Y\rvert^2 det C = a 2 − 2 1 ∣ Y ∣ 2 , the scalar q 1 q_1 q 1 may be regarded as a function of ( a , ∣ Y ∣ 2 ) (a,\lvert Y\rvert^2) ( a , ∣ Y ∣ 2 ) . Thus, after pressure projection, every two-dimensional spectral isotropic stress has the same active direction:
[ T ( C ) ] a c t i v e = q 1 ( a , ∣ Y ∣ 2 ) Y . [T(C)]_{\mathrm{active}} = q_1(a,\lvert Y\rvert^2)Y. [ T ( C ) ] active = q 1 ( a , ∣ Y ∣ 2 ) Y .
The pressure-free anisotropic channel is not a feature of Oldroyd–B or FENE-P; it is the natural quotient-level form of a two-dimensional isotropic tensorial stress. The scalar factor carries the constitutive law, while the deviatoric tensor Y Y Y carries the only stress direction visible to the incompressible velocity.
This observation changes the role of model classes in the continuation problem. The central regularity difficulty in stress-diffusion-free viscoelastic flow is a derivative imbalance between the viscous velocity and a non-diffusive conformation tensor. The elastic stress enters the momentum equation through a divergence, so one derivative of a transported quantity is fed back into the parabolic velocity equation. A useful criterion should therefore be formulated at the level of the active pressure class of the stress, not at the level of a chosen spring law. In two dimensions this active pressure class is exactly a scalar spectral coefficient times Y Y Y .
The first result of the paper is the corresponding active-deviatoric projection principle. For upper-convected stretching, the trace-free equation has the universal principal term 2 a S ( u ) 2aS(u) 2 a S ( u ) . Testing this equation with the weight q 1 / a q_1/a q 1 / a cancels the top-order velocity–stress coupling produced by div ( q 1 Y ) \operatorname{div}(q_1Y) div ( q 1 Y ) . Thus the leading continuation mechanism is dictated by the pressure quotient and by two-dimensional isotropic spectral algebra, rather than by a special cancellation in a particular constitutive model.
This projection is exactly two-dimensional, but the three-dimensional obstruction can be isolated rather than hidden. In three dimensions, the Cayley–Hamilton representation of a spectral stress generally contains a quadratic term,
T ( C ) = q 0 I + q 1 C + q 2 C 2 . T(C) = q_0I + q_1C + q_2C^2. T ( C ) = q 0 I + q 1 C + q 2 C 2 .
Writing C = a I + Y C = aI + Y C = a I + Y , a = 1 3 tr C a = \frac{1}{3}\operatorname{tr} C a = 3 1 tr C , tr Y = 0 \operatorname{tr}Y = 0 tr Y = 0 , gives the pressure-free identity
T ( C ) ∘ = ( q 1 + 2 a q 2 ) Y + q 2 ( Y 2 ) ∘ . T(C)^\circ= (q_1 + 2aq_2)Y + q_2(Y^2)^\circ. T ( C ) ∘ = ( q 1 + 2 a q 2 ) Y + q 2 ( Y 2 ) ∘ .
Thus the single active direction is replaced by a cancellative Y Y Y -channel and a Cayley–Hamilton residual ( Y 2 ) ∘ (Y^2)^\circ ( Y 2 ) ∘ . This is an algebraic boundary, not just a technical loss: in S 0 3 \mathbb{S}_0^3 S 0 3 , ( Y 2 ) ∘ (Y^2)^\circ ( Y 2 ) ∘ is not generally a scalar multiple of Y Y Y . The paper treats this residual explicitly in § 9 : on compact conformation windows it is a viscosity-absorbed high-order remainder, while for Oldroyd–B and FENE-P the residual coefficient q 2 q_2 q 2 is identically zero. This is the precise sense in which the two-dimensional quotient is exact and its three-dimensional remnant remains controlled for the principal spectral models. The statement is not a classification of arbitrary three-dimensional stresses. A non-spectral or anisotropic stress can carry pressure-free tensor components not expressible through I I I , C C C , C 2 C^2 C 2 ; such components require their own coercive structure or a direct stress clock, and are outside the quotient closure proved here. The second result packages this quotient structure into a compact-window analytic template.
For
C = a I + Y , η = ∣ Y ∣ 2 , C = aI + Y, \qquad\eta= |Y|^2, C = a I + Y , η = ∣ Y ∣ 2 ,
the compact-window normal form is
∂ t u + u ⋅ ∇ u − ν Δ u + ∇ p = α div ( τ ( a , η ) Y ) , \partial_t u + u \cdot\nabla u - \nu\Delta u + \nabla p = \alpha\operatorname{div}(\tau(a,\eta)Y), ∂ t u + u ⋅ ∇ u − ν Δ u + ∇ p = α div ( τ ( a , η ) Y ) ,
D t a + λ − 1 G ( a , η ) = S ( u ) : Y , D t Y + λ − 1 μ ( a , η ) Y = 2 a S ( u ) + [ ∇ u Y + Y ( ∇ u ) T ] . D_t a + \lambda^{-1}G(a,\eta) = S(u) : Y, \qquad D_tY + \lambda^{-1}\mu(a,\eta)Y = 2aS(u) + [\nabla u Y + Y(\nabla u)^{T}]. D t a + λ − 1 G ( a , η ) = S ( u ) : Y , D t Y + λ − 1 μ ( a , η ) Y = 2 a S ( u ) + [ ∇ u Y + Y ( ∇ u ) T ] .
Here τ ( a , η ) Y \tau(a,\eta)Y τ ( a , η ) Y is the pressure-free active stress allowed by two-dimensional spectral isotropy. The relaxation terms G ( a , η ) G(a,\eta) G ( a , η ) and μ ( a , η ) Y \mu(a,\eta)Y μ ( a , η ) Y record the scalar and deviatoric relaxation channels. The conditions imposed on these coefficients are compact-window continuation conditions: positive active and deviatoric relaxation coefficients, together with monotone scalar relaxation. They are not proposed as necessary thermodynamic axioms for all polymer laws. When a physical interpretation is wanted, one should additionally impose the free-energy compatibility and relaxation-dissipation condition in Remark 6.1. Under these hypotheses the principal cancellation uses the weight τ ( a , η ) / a \tau(a,\eta)/a τ ( a , η ) / a , and the high-order energy closes with coefficient
1 + ∥ ∇ u ∥ B ∞ , 1 0 + ∥ Log C ∥ H 1 + ε . 1 + \|\nabla u\|_{B^0_{\infty,1}} + \|\operatorname{Log} C\|_{H^{1+\varepsilon}}. 1 + ∥∇ u ∥ B ∞ , 1 0 + ∥ Log C ∥ H 1 + ε .
The compact-window theorem should therefore be read as a quotient estimate, not as a classification of polymer models. The concrete constitutive laws in the paper are Oldroyd–B and FENE-P. Remark 6.1 identifies the additional free-energy compatibility needed for thermodynamic models, and Example 6.1 records a polynomial free-energy perturbation with a genuine nonzero three-dimensional quadratic Cayley–Hamilton coefficient. This example calibrates the abstraction; it is not used to promote arbitrary coefficient choices to physical models.
The concrete model criteria are obtained after the relevant compact window is propagated. For Oldroyd–B, the endpoint clock
∫ 0 T ∥ ∇ u ( t ) ∥ B ∞ , 1 0 d t < ∞ \int_0^T \|\nabla u(t)\|_{B^0_{\infty,1}}\,dt < \infty ∫ 0 T ∥∇ u ( t ) ∥ B ∞ , 1 0 d t < ∞
propagates the positive-cone window. On that window a low-order estimate gives
Log A ∈ L 2 ( 0 , T ; H 1 + ε ) . \operatorname{Log} A \in L^2(0,T;H^{1+\varepsilon}). Log A ∈ L 2 ( 0 , T ; H 1 + ε ) .
For FENE-P, the positive-cone boundary is supplemented by the finite-extensibility boundary tr C = b \operatorname{tr} C = b tr C = b ; the trace-gap barrier is propagated by the squared clock
∇ u ∈ L t 2 B ∞ , 1 0 . \nabla u \in L^2_t B^0_{\infty,1}. ∇ u ∈ L t 2 B ∞ , 1 0 .
The same pressure-free anisotropic energy then closes the high-order estimate on this propagated FENE window.
The Besov endpoint is the zero-order velocity-gradient control that supplies the Lagrangian Lipschitz modulus. The criteria below are nevertheless high-order strong-solution restart criteria. They do not assert existence, uniqueness, or regularization for energy-level or Leray-type weak solutions; such statements would require a separate low-regularity framework. Nor are the criteria intended as critical-space local well-posedness theorems. Their ambient class is the standard integer Sobolev class used in the local continuation principles below. Thus the hypotheses are not empty: for every admissible datum in that class with a compact initial conformation window there is a positive local existence time, and the endpoint clock rules out breakdown of that strong solution.
Conformation and dimension conventions. In model-specific statements we use A A A only for the Oldroyd–B conformation tensor and C C C for the FENE-P conformation tensor. In the abstract spectral quotient statements, C C C is a dummy positive conformation tensor; after specialization it is replaced by A A A for Oldroyd–B and by C C C for FENE-P. The FENE-P spring factor is dimension dependent. In dimension d d d ,
f b , d ( C ) = b − d b − tr C , T b , d ( C ) = f b , d ( C ) C − I , D b , d = { C ∈ S + + d : tr C < b } , f_{b,d}(C)=\frac{b-d}{b-\operatorname{tr}C},\qquad T_{b,d}(C)=f_{b,d}(C)C-I,\qquad\mathcal{D}_{b,d}=\{C\in\mathbb{S}_{++}^{d}:\operatorname{tr}C<b\}, f b , d ( C ) = b − tr C b − d , T b , d ( C ) = f b , d ( C ) C − I , D b , d = { C ∈ S ++ d : tr C < b } ,
and the physical parameter range is b > d b>d b > d . Thus the two-dimensional sections use f b , 2 = ( b − 2 ) / ( b − tr C ) f_{b,2}=(b-2)/(b-\operatorname{tr}C) f b , 2 = ( b − 2 ) / ( b − tr C ) and b > 2 b>2 b > 2 , while § 9 uses f b , 3 = ( b − 3 ) / ( b − tr C ) f_{b,3}=(b-3)/(b-\operatorname{tr}C) f b , 3 = ( b − 3 ) / ( b − tr C ) and b > 3 b>3 b > 3 . These are the same condition b > d b>d b > d in different spatial dimensions, not two competing assumptions on one model.
The incompressible Oldroyd–B system without artificial stress diffusion is
∂ t u + u ⋅ ∇ u − ν Δ u + ∇ p = α div ( A − I ) , (1) \partial_tu+u\cdot\nabla u-\nu\Delta u+\nabla p=\alpha\operatorname{div}(A-I), \tag*{(1)} ∂ t u + u ⋅ ∇ u − ν Δ u + ∇ p = α div ( A − I ) , (1)
∂ t A + u ⋅ ∇ A = ∇ u A + A ( ∇ u ) T − λ − 1 ( A − I ) , (2) \partial_tA+u\cdot\nabla A=\nabla_uA+A(\nabla u)^{T}-\lambda^{-1}(A-I), \tag*{(2)} ∂ t A + u ⋅ ∇ A = ∇ u A + A ( ∇ u ) T − λ − 1 ( A − I ) , (2)
div u = 0. (3) \operatorname{div}u=0. \tag*{(3)} div u = 0. (3)
Here u : T 2 → R 2 u:\mathbb{T}^{2}\to\mathbb{R}^{2} u : T 2 → R 2 , A : T 2 → S + + 2 A:\mathbb{T}^{2}\to\mathbb{S}_{++}^{2} A : T 2 → S ++ 2 , and ν , α , λ > 0 \nu,\alpha,\lambda>0 ν , α , λ > 0 . The conformation tensor has no spatial diffusion. The velocity is parabolic, but the force in the vorticity equation contains curl div A \operatorname{curl}\operatorname{div}A curl div A , so derivatives of the stress enter the velocity equation at the same level at which parabolic smoothing is being used.
The corresponding stress-diffusion-free FENE-P system is
∂ t u + u ⋅ ∇ u − ν Δ u + ∇ p = α div T b ( C ) , (4) \partial_tu+u\cdot\nabla u-\nu\Delta u+\nabla p=\alpha\operatorname{div}T_b(C), \tag*{(4)} ∂ t u + u ⋅ ∇ u − ν Δ u + ∇ p = α div T b ( C ) , (4)
∂ t C + u ⋅ ∇ C = ∇ u C + C ( ∇ u ) T − λ − 1 ( f b ( C ) C − I ) , (5) \partial_tC+u\cdot\nabla C=\nabla_uC+C(\nabla u)^{T}-\lambda^{-1}\bigl(f_b(C)C-I\bigr), \tag*{(5)} ∂ t C + u ⋅ ∇ C = ∇ u C + C ( ∇ u ) T − λ − 1 ( f b ( C ) C − I ) , (5)
div u = 0 , (6) \operatorname{div}u=0, \tag*{(6)} div u = 0 , (6)
where, in the two-dimensional normalization used below,
f b ( C ) = f b , 2 ( C ) = b − 2 b − tr C , T b ( C ) = f b ( C ) C − I . f_b(C)=f_{b,2}(C)=\frac{b-2}{b-\operatorname{tr}C},\qquad T_b(C)=f_b(C)C-I. f b ( C ) = f b , 2 ( C ) = b − tr C b − 2 , T b ( C ) = f b ( C ) C − I .
The Oldroyd–B system is recovered formally on bounded trace windows as the Hookean limit b → ∞ b\to\infty b → ∞ , where f b ( C ) → 1 f_b(C)\to1 f b ( C ) → 1 and T b ( C ) → C − I T_b(C)\to C-I T b ( C ) → C − I . At fixed b b b , however, derivatives of the spring factor contain powers of ( b − tr C ) − 1 (b-\operatorname{tr}C)^{-1} ( b − tr C ) − 1 , and these coefficients cannot be read from the Oldroyd–B logarithmic variable alone.
The continuation result is stated in terms of these geometric variables, but for Oldroyd–B the spectral window is not an additional hypothesis. The endpoint velocity clock
∫ 0 T ∥ ∇ u ( t ) ∥ B ∞ , 1 0 d t < ∞ \int_{0}^{T}\|\nabla u(t)\|_{B^{0}_{\infty,1}}\,dt<\infty ∫ 0 T ∥∇ u ( t ) ∥ B ∞ , 1 0 d t < ∞
propagates both upper and lower eigenvalue bounds for A A A by a Lagrangian comparison argument. Once this automatic cone control is separated out, the low-order pressure-free estimate gives \Log A\in L^{2}(0,T;H^{1+\epsilon}) , and the high-order estimate then restarts the solution. The improvement from the purely logarithmic L t 4 L^{4}_{t} L t 4 closure to an L t 2 L^{2}_{t} L t 2 logarithmic estimate is obtained by the pressure-free physical decomposition A = a I + Y A=aI+Y A = a I + Y . For FENE-P the same positive-cone geometry remains, and the finite-extensibility boundary is controlled by a scalar barrier. The restoring force near tr C = b \operatorname{tr}C=b tr C = b prevents trace-gap collapse under ∇ u ∈ L t 2 B ∞ , 1 0 \nabla u\in L^{2}_{t}B^{0}_{\infty,1} ∇ u ∈ L t 2 B ∞ , 1 0 . On the resulting compact FENE window, the low-order estimate again gives \Log C\in L^{2}_{t}H^{1+\epsilon}_{x} , and the pressure-free anisotropic energy controls the remaining differentiability channel.
The main theorem may be summarized as follows.
Main theorem (Informal form). In two dimensions, every spectral isotropic elastic stress has an active deviatoric part q 1 Y q_1Y q 1 Y after pressure projection, and this active part cancels at top order against upper-convected stretching when the trace-free equation is tested with weight q 1 / a q_1/a q 1 / a . More generally, for compact-window spectrally admissible active channels
div ( τ ( a , ∣ Y ∣ 2 ) Y ) , D t Y + λ − 1 μ ( a , ∣ Y ∣ 2 ) Y = ⋯ , \operatorname{div}\bigl(\tau(a,\lvert Y\rvert^2)Y\bigr), \qquad D_tY+\lambda^{-1}\mu(a,\lvert Y\rvert^2)Y=\cdots, div ( τ ( a , ∣ Y ∣ 2 ) Y ) , D t Y + λ − 1 μ ( a , ∣ Y ∣ 2 ) Y = ⋯ , with positive τ , μ \tau,\mu τ , μ and stable scalar relaxation, this gives a compact-window endpoint estimate with coefficient
1 + ∥ ∇ u ∥ B ∞ , 1 0 + ∥ log C ∥ H 1 + ε 2 . 1+\lVert\nabla u\rVert_{B^0_{\infty,1}}+\lVert\log C\rVert_{H^{1+\varepsilon}}^2. 1 + ∥ ∇ u ∥ B ∞ , 1 0 + ∥ log C ∥ H 1 + ε 2 . Consequently, a strong two-dimensional Oldroyd–B solution continues beyond T T T if
∫ 0 T ∥ ∇ u ( t ) ∥ B ∞ , 1 0 d t < ∞ . \int_0^T\lVert\nabla u(t)\rVert_{B^0_{\infty,1}}\,dt<\infty. ∫ 0 T ∥ ∇ u ( t ) ∥ B ∞ , 1 0 d t < ∞. and a strong FENE-P solution continues beyond T T T if
∫ 0 T ∥ ∇ u ( t ) ∥ B ∞ , 1 0 2 d t < ∞ . \int_0^T\lVert\nabla u(t)\rVert_{B^0_{\infty,1}}^2\,dt<\infty. ∫ 0 T ∥ ∇ u ( t ) ∥ B ∞ , 1 0 2 d t < ∞. In both cases the L t 2 H x 1 + ε L^2_tH^{1+\varepsilon}_x L t 2 H x 1 + ε logarithmic conformation bound is derived on the propagated compact window rather than assumed as an external continuation hypothesis. The different velocity clocks come from window propagation: the Oldroyd–B positive cone is propagated by the endpoint flow-map clock, while the FENE-P trace gap has a finite-extensibility restoring force with boundary singularity exponent one. In the general trace-gap comparison, a restoring singularity ( b − tr C ) − θ (b-\operatorname{tr}C)^{-\theta} ( b − tr C ) − θ gives the velocity-clock exponent ( 1 + θ ) / θ (1+\theta)/\theta ( 1 + θ ) / θ . In three dimensions the full two-dimensional spectral reduction is no longer available. For a dummy tensor Z = a I + Y Z=aI+Y Z = a I + Y and spectral isotropic laws with a smooth local Cayley–Hamilton representation on the compact window, the pressure quotient instead splits as
T ( Z ) ∘ = τ 1 ( Z ) Y + τ 2 ( Z ) ( Y 2 ) ∘ , T(Z)^\circ=\tau_1(Z)Y+\tau_2(Z)(Y^2)^\circ, T ( Z ) ∘ = τ 1 ( Z ) Y + τ 2 ( Z ) ( Y 2 ) ∘ , where the second term is the Cayley–Hamilton residual. The Y Y Y -channel is the only channel that participates in the principal cancellation; the residual channel is estimated separately and absorbed by viscosity on compact windows. This compact-window statement does not cover arbitrary anisotropic or non-spectral stresses, and the quotient algebra by itself does not propagate the window. For Oldroyd–B and FENE-P, however, τ 2 ≡ 0 \tau_2\equiv0 τ 2 ≡ 0 and the needed windows are propagated by the model-specific flow-map and trace-gap barriers. Thus the principal three-dimensional model consequences are a pure velocity-clock criterion for Oldroyd–B, a pure-clock blow-up alternative for FENE-P, and a squared-clock window-propagated criterion for FENE-P. Separately, if one tries to replace this compact-window route by a fully logarithmic entropy route for three-dimensional FENE-P, the entropy variables have a finite-dimensional mobility threshold: they are monotone on every compact FENE window when b ≥ 15 / 4 b\ge15/4 b ≥ 15/4 , while for 3 < b < 15 / 4 3<b<15/4 3 < b < 15/4 an additional anisotropy-controlled spectral window is required. This side calculation explains the parameter q ∗ q_\ast q ∗ ; it is not an extra hypothesis in the velocity-clock criteria.
A model-facing reading guide. For readers interested primarily in Oldroyd–B and FENE-P, the proof can be read as a three-step mechanism. First, the pressure projection removes the isotropic stress and leaves only the deviatoric active channel. Second, the trace-free conformation equation contains exactly the stretching term needed to cancel the highest-order coupling between this channel and the velocity. Third, the endpoint velocity clock propagates the conformation window for the model at hand, after which the low-order logarithmic estimate and the high-order restart estimate close the continuation argument. Definition 6.1 only records the compact-window quotient estimate behind this mechanism; it may be skipped on a first reading of the two concrete model criteria.
Main results
We keep the model notation fixed: A A A denotes the Oldroyd–B conformation tensor and C C C denotes the FENE-P conformation tensor. Model-free spectral lemmas use a dummy tensor, denoted by C C C in two dimensions and by Z Z Z in § 9 . For FENE-P the Peterlin factor always uses the ambient dimension: f b , 2 = ( b − 2 ) / ( b − tr C ) f_{b,2}=(b-2)/(b-\operatorname{tr} C) f b , 2 = ( b − 2 ) / ( b − tr C ) in two dimensions and f b , 3 = ( b − 3 ) / ( b − tr C ) f_{b,3}=(b-3)/(b-\operatorname{tr} C) f b , 3 = ( b − 3 ) / ( b − tr C ) in three dimensions. The first result is the pressure-free active-deviatoric projection principle proved in § 6 . For every two-dimensional spectral isotropic stress law, the active stress after pressure projection has the form q 1 Y q_1Y q 1 Y . This identifies the unique stress component that enters vorticity and the top-order velocity–conformation coupling.
The second result is a compact-window quotient endpoint estimate. For active channels τ ( a , ∣ Y ∣ 2 ) Y \tau(a,|Y|^2)Y τ ( a , ∣ Y ∣ 2 ) Y , coercive deviatoric relaxation μ ( a , ∣ Y ∣ 2 ) Y \mu(a,|Y|^2)Y μ ( a , ∣ Y ∣ 2 ) Y , and monotone scalar relaxation G ( a , ∣ Y ∣ 2 ) G(a,|Y|^2) G ( a , ∣ Y ∣ 2 ) , the weighted active-deviatoric energy closes under the coefficient
1 + ∥ ∇ u ∥ B ∞ , 1 0 + ∥ Log C ∥ H 1 + ε 2 . 1+\|\nabla u\|_{B^0_{\infty,1}}+\|\operatorname{Log} C\|_{H^{1+\varepsilon}}^2. 1 + ∥∇ u ∥ B ∞ , 1 0 + ∥ Log C ∥ H 1 + ε 2 .
This is a high-order analytic estimate on a prescribed compact window, not a constitutive classification. Thermodynamic polymer closures form the subclass for which the stress and relaxation are derived from a common free-energy density in the sense of Remark 6.1. Example 6.1 gives a nontrivial member of this subclass whose three-dimensional stress contains an actual C 2 C^2 C 2 component. The scalar-Peterlin theorem is the trace-only specialization
τ = ψ ( 2 a ) , μ = ϕ ( 2 a ) , G = a ϕ ( 2 a ) − χ ( 2 a ) , \tau=\psi(2a),\qquad\mu=\phi(2a),\qquad G=a\phi(2a)-\chi(2a), τ = ψ ( 2 a ) , μ = ϕ ( 2 a ) , G = a ϕ ( 2 a ) − χ ( 2 a ) ,
and the Oldroyd–B and FENE-P estimates are recovered from this specialization.
The third result is the Oldroyd–B continuation theorem. A strong solution can be continued as long as
∫ 0 T ∥ ∇ u ( t ) ∥ B ∞ , 1 0 d t < ∞ . \int_0^T \|\nabla u(t)\|_{B^0_{\infty,1}}\,dt<\infty. ∫ 0 T ∥∇ u ( t ) ∥ B ∞ , 1 0 d t < ∞.
The compact positive-cone window needed by the high-order energy estimate is not assumed; it is propagated by the velocity clock. On this window, the low-order pressure-free estimate gives Log A ∈ L 2 ( 0 , T ; H 1 + ε ) \operatorname{Log} A\in L^2(0,T;H^{1+\varepsilon}) Log A ∈ L 2 ( 0 , T ; H 1 + ε ) , which makes the high-order coefficient integrable. Equivalently, finite-time Oldroyd–B breakdown forces divergence of the velocity Besov clock.
The fourth result is the FENE-P continuation theorem proved in § 8 . The FENE-P compact window has two boundaries:
C > 0 , tr C < b . C>0,\qquad\operatorname{tr} C<b. C > 0 , tr C < b .
The lower spectral boundary is propagated by the velocity clock, as in Oldroyd–B. The finite-extensibility boundary is propagated by the scalar barrier equation once ∇ u ∈ L 2 ( 0 , T ; B ∞ , 1 0 ) \nabla u\in L^2(0,T;B^0_{\infty,1}) ∇ u ∈ L 2 ( 0 , T ; B ∞ , 1 0 ) . On the resulting compact FENE window, the pressure-free splitting T b ( C ) = ( f b a − 1 ) I + f b Y T_b(C)=(f_ba-1)I+f_bY T b ( C ) = ( f b a − 1 ) I + f b Y gives the same principal anisotropic cancellation with weight f b / a f_b/a f b / a . Thus the FENE-P continuation criterion is
∫ 0 T ∥ ∇ u ( t ) ∥ B ∞ , 1 0 2 d t < ∞ . \int_0^T \|\nabla u(t)\|_{B^0_{\infty,1}}^2\,dt<\infty. ∫ 0 T ∥∇ u ( t ) ∥ B ∞ , 1 0 2 d t < ∞.
The logarithmic conformation bound needed by the high-order estimate is derived on the propagated compact FENE window.
The fifth result is a three-dimensional compact-window extension with an explicit Cayley–Hamilton residual split. This result has a narrower scope than the two-dimensional theorem. In the model-free three-dimensional statements we write Z = a I + Y Z=aI+Y Z = a I + Y , a = 1 3 tr Z a=\frac{1}{3}\operatorname{tr} Z a = 3 1 tr Z , and remove scalar stress through the pressure. For spectral isotropic stress laws admitting a smooth local Cayley–Hamilton representation on a compact spectral window,
T ( Z ) = q 0 ( Z ) I + q 1 ( Z ) Z + q 2 ( Z ) Z 2 , (7) T(Z)=q_0(Z)I+q_1(Z)Z+q_2(Z)Z^2, \tag*{(7)} T ( Z ) = q 0 ( Z ) I + q 1 ( Z ) Z + q 2 ( Z ) Z 2 , (7)
the pressure-free part is
T ( Z ) ∘ = ( q 1 + 2 a q 2 ) Y + q 2 ( Y 2 ) ∘ . T(Z)^\circ= (q_1 + 2a q_2)Y + q_2(Y^2)^\circ. T ( Z ) ∘ = ( q 1 + 2 a q 2 ) Y + q 2 ( Y 2 ) ∘ .
The first term is the cancellative active channel. The second term is a quadratic residual. Proposition 9.1 shows that this residual is generically an independent pressure-free direction, so a single scalar quotient closure is a genuinely two-dimensional feature. After one integration by parts the residual pairs with the viscous velocity dissipation and is controlled by compact-window Moser estimates. This is not a theorem for arbitrary anisotropic or non-spectral stresses, for which additional pressure-free tensor directions may be present. Nor does the algebraic split itself propagate the compact window: the compact-window criterion is conditional until a model-specific barrier supplies that window. For Oldroyd–B and FENE-P the residual is absent, because their constitutive representations are explicit with q 2 = 0 q_2 = 0 q 2 = 0 , and the required windows are obtained from the flow-map and trace-gap estimates. Hence a smooth three-dimensional Oldroyd–B solution can break down only if
\int_0^{T_*} \|\nabla u(t)\|_{B^\mathord{\infty,1}_\infty}\,\mathrm{d}t = \infty.
For three-dimensional FENE-P, finite-time breakdown forces at least one of the following: divergence of the same pure velocity clock, loss of the lower positive-cone bound, or collapse of the trace gap b − tr C b - \operatorname{tr} C b − tr C . Under the squared clock
\int_0^{T_*} \|\nabla u(t)\|_{B^\mathord{\infty,1}_\infty}^{2}\,\mathrm{d}t < \infty,
the compact FENE window is propagated and the same continuation mechanism applies. Finally, a separate finite-dimensional entropy-mobility calculation is recorded for a possible three-dimensional FENE-P logarithmic formulation. It is not used in the velocity-clock proof. Its role is to identify when the entropy-variable mobility is monotone: this is automatic on compact FENE windows when b ≥ 15 / 4 b \ge15/4 b ≥ 15/4 , while for 3 < b < 15 / 4 3 < b < 15/4 3 < b < 15/4 the spectral variance must remain below the explicit threshold q ∗ ( b , r ) q_*(b,r) q ∗ ( b , r ) .
The final result records a static operator obstruction associated with the logarithmic regularity used in the proof. For active forcing measured in H − 1 + γ H^{-1+\gamma} H − 1 + γ , the estimate based on the pressure-free stress map cannot be closed below Log C ∈ H 1 + γ \operatorname{Log} C \in H^{1+\gamma} Log C ∈ H 1 + γ without additional dynamical input. Indeed, below this threshold there are fixed-spectrum sequences for which every zeroth-order spectral density is pointwise constant, while
∥ curl div T ( C N ) ∥ H − 1 + γ → ∞ . \|\operatorname{curl}\operatorname{div} T(C_N)\|_{H^{-1+\gamma}} \to\infty. ∥ curl div T ( C N ) ∥ H − 1 + γ → ∞.
The obstruction is high-frequency rotation of the active deviatoric channel. It explains why the logarithmic term appears in the proof, but it is not a finite-time blow-up or dynamic optimality theorem.
The paper is organized as follows. § 3 records the analytic estimates used in the proof. § 4 records the positive cone and entropy structure inherited by both models. § 5 gives the logarithmic equation and the good unknown. § 6 proves the active-deviatoric projection principle, the full compact-window quotient estimate, its scalar-Peterlin specialization, and the Oldroyd–B continuation criterion, including the propagation of the compact spectral window from the endpoint velocity clock. § 7 proves the entropy-blind obstruction and the static H 1 + γ H^{1+\gamma} H 1 + γ active-forcing obstruction. § 8 derives the FENE-P endpoint continuation criterion and the propagation of the finite-extensibility trace gap. § 9 gives the three-dimensional Cayley–Hamilton residual split, states its exact compact-window scope, derives model-specific consequences for Oldroyd–B and FENE-P, and records the FENE-P entropy-mobility boundary. § 10 compares the FENE-P criterion with Oldroyd–B in the Hookean limit.
Main mechanism and proof structure
The pressure quotient leaves a single active direction in two dimensions. For C = a I + Y C=aI+Y C = a I + Y , Y = C ∘ Y=C^{\circ} Y = C ∘ , a smooth spectral isotropic stress has the pressure-free form
[ T ( C ) ] a c t i v e = q 1 ( a , ∣ Y ∣ 2 ) Y . [T(C)]_{\mathrm{active}}=q_1(a,|Y|^2)Y. [ T ( C ) ] active = q 1 ( a , ∣ Y ∣ 2 ) Y .
Thus the same deviatoric variable that appears in the trace-free conformation equation is the only stress component visible to the incompressible velocity. For such compact-window quotient channels the differentiated velocity equation contains the leading term
− α ∫ τ ( a , η ) ∂ β Y : S ( ∂ β u ) d x , η = ∣ Y ∣ 2 , -\alpha\int\tau(a,\eta)\,\partial^\beta Y:S(\partial^\beta u)\,dx,\qquad\eta=|Y|^2, − α ∫ τ ( a , η ) ∂ β Y : S ( ∂ β u ) d x , η = ∣ Y ∣ 2 ,
while the trace-free equation, tested with the weight
W ( a , η ) = τ ( a , η ) a , W(a,\eta)=\frac{\tau(a,\eta)}{a}, W ( a , η ) = a τ ( a , η ) ,
contains the opposite contribution
α 2 ∫ W ( a , η ) ∂ β Y : 2 a S ( ∂ β u ) d x = α ∫ τ ( a , η ) ∂ β Y : S ( ∂ β u ) d x . \frac{\alpha}{2}\int W(a,\eta)\partial^\beta Y:2aS(\partial^\beta u)\,dx
=\alpha\int\tau(a,\eta)\,\partial^\beta Y:S(\partial^\beta u)\,dx. 2 α ∫ W ( a , η ) ∂ β Y : 2 a S ( ∂ β u ) d x = α ∫ τ ( a , η ) ∂ β Y : S ( ∂ β u ) d x .
The top-order velocity–stress coupling therefore cancels at the level of the pressure-free variables. After the endpoint transport commutators and derivatives of the weight are estimated, the compact-window high-order energy has coefficient
1+\|\nabla u\|_{B^0_{\infty,1}}+\|\Log C\|_{H^{1+\varepsilon}}^2.
A low-order pressure-free estimate gives the logarithmic term on the propagated compact window. The model-specific part of the argument is the window propagation: Oldroyd–B uses the Lagrangian positive-cone comparison, whereas FENE-P also uses the scalar trace-gap barrier. This yields the L t 1 B ∞ , 1 0 L_t^1B^0_{\infty,1} L t 1 B ∞ , 1 0 velocity clock for Oldroyd–B and the squared L t 2 B ∞ , 1 0 L_t^2B^0_{\infty,1} L t 2 B ∞ , 1 0 clock for FENE-P.
The proof is organized accordingly. First, the stress is reduced modulo pressure to q 1 Y q_1Y q 1 Y . Second, the weighted trace-free energy cancels the principal force div ( q 1 Y ) \operatorname{div}(q_1Y) div ( q 1 Y ) . Third, the relevant compact spectral window is propagated by the scalar comparison available in the model under consideration. On this propagated window, the low-order estimate yields the logarithmic regularity. Fourth, this logarithmic bound closes the high-order estimate and allows the strong solution to be restarted.
Relation with Euler and stress-diffusive models
The regularity problem for the two-dimensional Oldroyd–B system without stress diffusion lies between two regimes that are much better understood. If stress diffusion is imposed, the missing derivative in the conformation equation is restored by parabolic smoothing. If the data are small, perturbative mechanisms can keep the non-diffusive stress from feeding a large cascade back into the velocity. The regime considered here has neither source of control: no artificial stress dissipation and no smallness.
For the two-dimensional Euler equation, vorticity is transported and the classical obstruction is growth of the Lipschitz norm of the velocity. For Oldroyd–B without stress diffusion, the vorticity equation contains the extra forcing curl div A \operatorname{curl}\operatorname{div} A curl div A . The stress is transported and stretched, but it has no parabolic regularization. Thus the Euler-type vorticity criterion becomes a coupled velocity–conformation criterion; see Corollary 6.2. The endpoint velocity clock prevents Oldroyd–B escape from the positive cone, while the logarithmic channel controls the remaining high-frequency stress mechanism.
Related work
The Oldroyd model goes back to Oldroyd’s constitutive theory [25 ] ; standard continuum and kinetic accounts include [7 , 8 , 11 , 18 , 6 , 27 , 26 ] . Strong and weak solution theories for differential viscoelastic models were developed in several directions, including the early work of Guillopé and Saut [15 ] , the global weak-solution framework of Lions and Masmoudi [22 ] , and the small-data smooth theory of Lin, Liu, and Zhang [20 ] , and the boundary-value theory of Lin and Zhang [21 ] . Lifespan and breakdown criteria for related viscoelastic systems were studied by Chemin and Masmoudi [9 ] and by Lei, Masmoudi, and Zhou [19 ] . When additional stress diffusion is present, two-dimensional regularity becomes substantially more tractable; see, for example, Constantin and Kliegl [10 ] , the global regularity results of Elgindi and Rousset [12 ] , and the regularized and numerical frameworks of Barrett–Boyaval and Barrett–Süli [2 , 3 ] . The pressure handling in these diffusive works is different from the quotient used here. There, Leray projection or auxiliary vorticity–stress combinations enter parabolic or transformed energy estimates, where stress diffusion, damping, or a model-specific transformed variable supplies the missing derivative. Here the isotropic stress is removed algebraically before the high-order estimate, and the remaining active deviatoric channel is paired directly with upper-convected stretching; no stress Laplacian is available. In the stress-diffusion-free system, global control is also known in small-data regimes; a recent example is the work of Tu, Wang, and Wen [28 ] . The results below address the regime without stress diffusion and without smallness, using variables adapted to the conformation geometry.
The logarithmic conformation representation was introduced in numerical rheology by Fattal and Kupferman [13 , 14 ] and further developed in high-Weissenberg computations by Hulsen, Fattal, and Kupferman [16 ] to preserve positivity and improve stability. The present use is analytic: the logarithmic chart separates spectral control from high-frequency concentration. The velocity part of the criterion follows the spirit of the Beale–Kato–Majda continuation principle [5 ] , but the non-diffusive stress equation forces the additional logarithmic stress diagnostic.
The finitely extensible models considered below originate from the standard FENE closures for dilute polymeric fluids; see, for example, [8 , 11 ] . Mathematical work on FENE dumbbell models has also emphasized the role of the finite-extension boundary; representative global well-posedness and weak-solution results are due to Masmoudi [23 , 24 ] , while finite-element and weak-solution approaches for regularized kinetic closures are developed in [3 , 4 ] . The FENE-P criterion below isolates the finite-extensibility boundary in an endpoint continuation framework: the trace-gap barrier is propagated by the squared endpoint velocity clock, and the same pressure-free active-deviatoric cancellation closes the high-order continuation estimate.
Relation with direct continuation criteria
Classical non-blowup criteria for Oldroyd-type systems monitor the stress itself. Chemin and Masmoudi obtained criteria involving direct stress norms, and Lei, Masmoudi, and Zhou gave a BKM-type formulation with τ ∈ L 1 ( 0 , T ; B M O ) \tau\in L^{1}(0,T;\mathrm{BMO}) τ ∈ L 1 ( 0 , T ; BMO ) , together with low-integrability stress control. The criterion here is not a norm inclusion between stress and velocity clocks. It is a pressure-quotient certification: the isotropic stress is removed before the high-order coupling is estimated. This is also distinct from the pressure manipulations used in stress-diffusive global regularity arguments, where the projection is coupled to parabolic or transformed energy control.
This distinction is visible already for pure pressure modes. If p N ( x ) = N sin ( N x 1 ) p_N(x)=N\sin(Nx_1) p N ( x ) = N sin ( N x 1 ) and σ N = p N I \sigma_N=p_N I σ N = p N I , then
P div σ N = 0 , ∥ σ N ∥ B M O ≳ N , \mathbb{P}\operatorname{div}\sigma_N=0,\qquad\|\sigma_N\|_{\mathrm{BMO}}\gtrsim N, P div σ N = 0 , ∥ σ N ∥ BMO ≳ N ,
where P \mathbb{P} P is the Leray projector. A direct full-stress clock can therefore be large because of a component that exerts no force on the incompressible velocity.
Conversely, once the velocity clock has propagated a compact positive-cone window for Oldroyd–B, the usual direct stress clocks are recovered. If c I ≤ A ( t , x ) ≤ C I cI\le A(t,x)\le CI c I ≤ A ( t , x ) ≤ C I , then τ = A − I \tau=A-I τ = A − I satisfies
∥ τ ( t ) ∥ B M O ≤ 2 ∥ τ ( t ) ∥ L ∞ ≤ C c , C , ∥ τ ( t ) ∥ L 1 + ∥ τ ( t ) ∥ L 2 ≤ C c , C . \|\tau(t)\|_{\mathrm{BMO}} \le2\|\tau(t)\|_{L^\infty} \le C_{c,C}, \qquad\|\tau(t)\|_{L^1}+\|\tau(t)\|_{L^2} \le C_{c,C}. ∥ τ ( t ) ∥ BMO ≤ 2∥ τ ( t ) ∥ L ∞ ≤ C c , C , ∥ τ ( t ) ∥ L 1 + ∥ τ ( t ) ∥ L 2 ≤ C c , C .
Lemma 6.1 shows that this window is propagated by the endpoint velocity clock from positive initial data. Thus, along the solutions covered by Theorem 6.1, the direct stress hypotheses are consequences rather than independent assumptions.
The reverse implication does not follow from direct stress information alone. For A ε = ε I A_\varepsilon=\varepsilon I A ε = ε I ,
∥ A ε − I ∥ B M O = 0 , ∥ A ε − I ∥ L 1 ≤ C , ∥ Log A ε ∥ L ∞ = ∣ log ε ∣ → ∞ . \|A_\varepsilon-I\|_{\mathrm{BMO}}=0,\qquad\|A_\varepsilon-I\|_{L^1}\le C,\qquad\|\operatorname{Log} A_\varepsilon\|_{L^\infty}=|\log\varepsilon|\to\infty. ∥ A ε − I ∥ BMO = 0 , ∥ A ε − I ∥ L 1 ≤ C , ∥ Log A ε ∥ L ∞ = ∣ log ε ∣ → ∞.
Direct stress BMO and low-integrability bounds do not encode the lower positive-cone boundary or the logarithmic chart. The pressure-quotient criterion should therefore be read as a one-way geometric mechanism: a finite endpoint velocity clock propagates the conformation window, the low-order estimate gives the logarithmic regularity used in the high-order estimate, and the argument then implies the standard direct stress clocks on that window.
Functional Setting and Continuation Framework
We work on the periodic torus T 2 \mathbb{T}^2 T 2 to avoid boundary effects. The same local estimates apply on the whole plane with the standard modifications for decay and low frequencies. The continuation theorems are formulated at integer Sobolev levels s = m ∈ N s=m\in\mathbb{N} s = m ∈ N , m ≥ 3 m\ge3 m ≥ 3 . Thus the top-order cancellations are proved by classical differentiations and the endpoint clock is used to prevent breakdown of an already existing strong solution. The only fractional Sobolev input used in the argument is the low-order H 1 + ε H^{1+\varepsilon} H 1 + ε logarithmic bound, and its endpoint transport estimate is proved below in dyadic Littlewood–Paley form.
Notation. Constants denoted by C C C may change from line to line. Constants denoted by C K C_K C K may also depend on a fixed spectral window ∥ B ∥ L ∞ ≤ K \|B\|_{L^\infty}\le K ∥ B ∥ L ∞ ≤ K . In the two-dimensional FENE-P part, K K K also denotes a compact subdomain of
D b = D b , 2 = { C ∈ S + + 2 : tr C < b } , b > 2 ; \mathcal{D}_b=\mathcal{D}_{b,2}=\{C\in S_{++}^2:\operatorname{tr}C<b\},\qquad b>2; D b = D b , 2 = { C ∈ S ++ 2 : tr C < b } , b > 2 ;
the constants may then depend on b b b , the lower spectral gap of C C C , and the upper trace gap b − tr C b-\operatorname{tr}C b − tr C . Throughout the high-order index s s s is an integer m ≥ 3 m\ge3 m ≥ 3 . This keeps the continuation statement aligned with the estimates actually proved. The fractional exponent 1 + ε 1+\varepsilon 1 + ε appearing later is handled explicitly by the endpoint dyadic commutator Lemma 3.2 ; no separate fractional high-order continuation theorem is invoked.
Definition 2.1 (Strong positive-cone solution). Let T > 0 T>0 T > 0 . A strong positive-cone solution on [ 0 , T ] [0,T] [ 0 , T ] is a pair ( u , A ) (u,A) ( u , A ) such that
u ∈ C ( [ 0 , T ] ; H s ) , A ∈ C ( [ 0 , T ] ; H s ( T 2 ; S + + 2 ) ) , u\in C([0,T];H^s),\qquad A\in C([0,T];H^s(\mathbb{T}^2;S_{++}^2)), u ∈ C ([ 0 , T ] ; H s ) , A ∈ C ([ 0 , T ] ; H s ( T 2 ; S ++ 2 )) , u ∈ L 2 ( 0 , T ; H s + 1 ) , div u = 0 , u\in L^2(0,T;H^{s+1}),\qquad\operatorname{div}u=0, u ∈ L 2 ( 0 , T ; H s + 1 ) , div u = 0 , and (1) –(2) hold in the classical Sobolev sense.
Definition 2.2 (Strong FENE-P solution). Let b > 2 b>2 b > 2 and T > 0 T>0 T > 0 . A strong FENE-P solution on [ 0 , T ] [0,T] [ 0 , T ] is a pair ( u , C ) (u,C) ( u , C ) such that
u ∈ C ( [ 0 , T ] ; H s ) , C ∈ C ( [ 0 , T ] ; H s ( T 2 ; S + + 2 ) ) , u\in C([0,T];H^s),\qquad C\in C([0,T];H^s(\mathbb{T}^2;S_{++}^2)), u ∈ C ([ 0 , T ] ; H s ) , C ∈ C ([ 0 , T ] ; H s ( T 2 ; S ++ 2 )) , u ∈ L 2 ( 0 , T ; H s + 1 ) , div u = 0 , C ( t , x ) ∈ D b , u\in L^2(0,T;H^{s+1}),\qquad\operatorname{div}u=0,\qquad C(t,x)\in\mathcal{D}_b, u ∈ L 2 ( 0 , T ; H s + 1 ) , div u = 0 , C ( t , x ) ∈ D b , and (4) –(5) hold in the classical Sobolev sense.
Proposition 2.1 (Local continuation principle). Let s ∈ N s \in\mathbb{N} s ∈ N , s ≥ 3 s \ge3 s ≥ 3 , u 0 ∈ H s u_0 \in H^s u 0 ∈ H s , div u 0 = 0 \operatorname{div}u_0 = 0 div u 0 = 0 , and A 0 ∈ H s ( T 2 ; S + + 2 ) A_0 \in H^s(\mathbb{T}^2;\mathbb{S}_{++}^2) A 0 ∈ H s ( T 2 ; S ++ 2 ) with spectral range contained in [ m 0 , M 0 ] [m_0,M_0] [ m 0 , M 0 ] , 0 < m 0 < M 0 < ∞ 0 < m_0 < M_0 < \infty 0 < m 0 < M 0 < ∞ . Then there exists a unique strong positive-cone solution on a time interval [ 0 , T l o c ] [0,T_{\mathrm{loc}}] [ 0 , T loc ] . Moreover, if a solution on [ 0 , T ∗ ) [0,T_\ast) [ 0 , T ∗ ) satisfies
sup t < T ∗ ( ∥ u ( t ) ∥ H s + ∥ A ( t ) ∥ H s + ∥ A ( t ) ∥ L ∞ + ∥ A ( t ) − 1 ∥ L ∞ ) < ∞ , \sup_{t<T_\ast}\left(\|u(t)\|_{H^s}+\|A(t)\|_{H^s}+\|A(t)\|_{L^\infty}+\|A(t)^{-1}\|_{L^\infty}\right)<\infty, t < T ∗ sup ( ∥ u ( t ) ∥ H s + ∥ A ( t ) ∥ H s + ∥ A ( t ) ∥ L ∞ + ∥ A ( t ) − 1 ∥ L ∞ ) < ∞ , then it extends as a strong positive-cone solution beyond T ∗ T_\ast T ∗ .
Proof. Regularize the system, solve by Picard iteration, and obtain estimates in H s H^s H s for A A A and in H s ∩ L t 2 H s + 1 H^s \cap L^2_tH^{s+1} H s ∩ L t 2 H s + 1 for u u u . Positivity is propagated along characteristics by Lemma 4.1. The lower spectral bound prevents degeneration of the cone, while the upper spectral and Sobolev bounds control all nonlinear coefficients.
More explicitly, the regularized estimates give a local existence time
T l o c ≥ τ ( ∥ u 0 ∥ H s + ∥ A 0 ∥ H s , ∥ A 0 ∥ L ∞ , ∥ A 0 − 1 ∥ L ∞ ) > 0 , T_{\mathrm{loc}} \ge\tau\left(\|u_0\|_{H^s}+\|A_0\|_{H^s},\|A_0\|_{L^\infty},\|A_0^{-1}\|_{L^\infty}\right)>0, T loc ≥ τ ( ∥ u 0 ∥ H s + ∥ A 0 ∥ H s , ∥ A 0 ∥ L ∞ , ∥ A 0 − 1 ∥ L ∞ ) > 0 , where τ \tau τ is nondecreasing in its arguments. In particular, the integer-Sobolev positive-cone class contains a full local solution class for every datum satisfying the displayed bounds. The same estimate applies when the construction is restarted from any time t 0 < T ∗ t_0<T_\ast t 0 < T ∗ . Under (2.1), these arguments remain bounded uniformly for t 0 < T ∗ t_0<T_\ast t 0 < T ∗ . Hence there is a uniform τ ∗ > 0 \tau_\ast>0 τ ∗ > 0 such that the solution restarts on [ t 0 , t 0 + τ ∗ ] [t_0,t_0+\tau_\ast] [ t 0 , t 0 + τ ∗ ] for every t 0 < T ∗ t_0<T_\ast t 0 < T ∗ sufficiently close to T ∗ T_\ast T ∗ . Choosing t 0 > T ∗ − τ ∗ / 2 t_0>T_\ast-\tau_\ast/2 t 0 > T ∗ − τ ∗ /2 gives an extension beyond T ∗ T_\ast T ∗ .
Proposition 2.2 (Local continuation in the FENE-P domain). Let b > 2 b>2 b > 2 , s ∈ N s\in\mathbb{N} s ∈ N , s ≥ 3 s\ge3 s ≥ 3 , u 0 ∈ H s u_0\in H^s u 0 ∈ H s , div u 0 = 0 \operatorname{div}u_0=0 div u 0 = 0 , and C 0 ∈ H s ( T 2 ; D b ) C_0\in H^s(\mathbb{T}^2;D_b) C 0 ∈ H s ( T 2 ; D b ) . Assume that the range of C 0 C_0 C 0 is contained in a
compact set K 0 ⋐ D b K_0 \Subset D_b K 0 ⋐ D b . Then there exists a unique strong FENE-P solution on a time interval [ 0 , T l o c ] [0,T_{\mathrm{loc}}] [ 0 , T loc ] . Moreover, if a solution on [ 0 , T ∗ ) [0,T_*) [ 0 , T ∗ ) satisfies
sup t < T ∗ ( ∥ u ( t ) ∥ H s + ∥ C ( t ) ∥ H s ) < ∞ \sup_{t<T_*}\left(\|u(t)\|_{H^s}+\|C(t)\|_{H^s}\right)<\infty t < T ∗ sup ( ∥ u ( t ) ∥ H s + ∥ C ( t ) ∥ H s ) < ∞
and the range of C ( t , ⋅ ) C(t,\cdot) C ( t , ⋅ ) remains in a fixed compact set K ⋐ D b K\Subset D_b K ⋐ D b for all t < T ∗ t<T_* t < T ∗ , then the solution extends beyond T ∗ T_* T ∗ as a strong FENE-P solution.
Proof. On K ⋐ D b K\Subset D_b K ⋐ D b , the maps
C ↦ f b ( C ) , C ↦ T b ( C ) , C ↦ f b ( C ) C − I C\mapsto f_b(C),\qquad C\mapsto T_b(C),\qquad C\mapsto f_b(C)C-I C ↦ f b ( C ) , C ↦ T b ( C ) , C ↦ f b ( C ) C − I
are smooth with bounded derivatives of every finite order. The FENE-P system therefore has the same quasilinear parabolic-transport structure as the Oldroyd–B system, with composition constants depending only on K K K , b b b , and the H s H^s H s size of C C C . Regularization and Picard iteration give a local existence time bounded below by a nonincreasing function of
∥ u 0 ∥ H s + ∥ C 0 ∥ H s \|u_0\|_{H^s}+\|C_0\|_{H^s} ∥ u 0 ∥ H s + ∥ C 0 ∥ H s
and the compact-set constants of K 0 K_0 K 0 . Thus the FENE-P strong class is a genuine local solution class for compact finite-extensibility data, not an additional a priori assumption. Restarting the construction at times t 0 < T ∗ t_0<T_* t 0 < T ∗ gives the same lower bound with K K K in place of K 0 K_0 K 0 . The stated uniform bounds therefore give a uniform restart time and extend the solution beyond T ∗ T_* T ∗ . □
Analytic Preliminaries
We collect the analytic estimates used below. We use the standard Littlewood–Paley and paraproduct framework of [1 ] and the commutator estimates of [17 ] . The endpoint transport estimate is written in dyadic form because it is the only place where the critical clock B ∞ , 1 0 B^0_{\infty,1} B ∞ , 1 0 interacts with a fractional Sobolev index. The remaining high-order coefficient estimates are used at integer levels.
Lemma 3.1 (Standard Moser and tame calculus). Let r > 1 r>1 r > 1 in two dimensions. Then H r ( T 2 ) H^r(\mathbb{T}^2) H r ( T 2 ) is an algebra and
∥ f g ∥ H r ≤ C r ( ∥ f ∥ L ∞ ∥ g ∥ H r + ∥ g ∥ L ∞ ∥ f ∥ H r ) . \|fg\|_{H^r}\leq C_r\left(\|f\|_{L^\infty}\|g\|_{H^r}+\|g\|_{L^\infty}\|f\|_{H^r}\right). ∥ f g ∥ H r ≤ C r ( ∥ f ∥ L ∞ ∥ g ∥ H r + ∥ g ∥ L ∞ ∥ f ∥ H r ) . If F ∈ C ∞ ( R N ) F\in C^\infty(\mathbb{R}^N) F ∈ C ∞ ( R N ) , F ( 0 ) = 0 F(0)=0 F ( 0 ) = 0 , and ∥ v ∥ L ∞ ≤ K \|v\|_{L^\infty}\leq K ∥ v ∥ L ∞ ≤ K , then
∥ F ( v ) ∥ H r ≤ C F , K , r ∥ v ∥ H r . \|F(v)\|_{H^r}\leq C_{F,K,r}\|v\|_{H^r}. ∥ F ( v ) ∥ H r ≤ C F , K , r ∥ v ∥ H r . The same estimates hold componentwise for matrix-valued functions. Moreover, if m ≥ 2 m\geq2 m ≥ 2 , 0 < ε < 1 0<\varepsilon<1 0 < ε < 1 , and f 1 , … , f N f_1,\ldots,f_N f 1 , … , f N be smooth scalar, vector, or matrix fields on T 2 \mathbb{T}^2 T 2 . If at most one factor is measured at order m m m , then
∥ ∏ j = 1 N f j ∥ H m ≤ C ∏ j = 1 N ( 1 + ∥ f j ∥ H 1 + ε ) ∑ j = 1 N ∥ f j ∥ H m . \left\|\prod_{j=1}^{N}f_j\right\|_{H^m}\leq C\prod_{j=1}^{N}\left(1+\|f_j\|_{H^{1+\varepsilon}}\right)\sum_{j=1}^{N}\|f_j\|_{H^m}. j = 1 ∏ N f j H m ≤ C j = 1 ∏ N ( 1 + ∥ f j ∥ H 1 + ε ) j = 1 ∑ N ∥ f j ∥ H m . The constant depends on m m m , ε \varepsilon ε , N N N and on fixed L ∞ L^\infty L ∞ bounds for smooth coefficient functions of the f j f_j f j 's.
Proof. These are standard consequences of the Bony paraproduct decomposition f g = T f g + T g f + R ( f , g ) fg=T_f g+T_g f+R(f,g) f g = T f g + T g f + R ( f , g ) . Since r > 1 = d / 2 r>1=d/2 r > 1 = d /2 , H r ( T 2 ) H^r(\mathbb{T}^2) H r ( T 2 ) is an algebra. The composition bound follows by writing
F ( v ) = ∫ 0 1 D F ( θ v ) v d θ . F(v)=\int_0^1 DF(\theta v)v\,d\theta. F ( v ) = ∫ 0 1 D F ( θ v ) v d θ .
For (3.1), put the highest derivative on one factor and place all remaining factors in H 1 + ε ↪ L ∞ H^{1+\varepsilon}\hookrightarrow L^\infty H 1 + ε ↪ L ∞ . □
Lemma 3.2 (Endpoint Besov transport commutator). Let ( Δ j ) j ≥ − 1 (\Delta_j)_{j\geq-1} ( Δ j ) j ≥ − 1 be a nonhomogeneous Littlewood–Paley partition on T 2 \mathbb{T}^2 T 2 . Let σ > 0 \sigma> 0 σ > 0 , div u = 0 \operatorname{div} u = 0 div u = 0 , and let f f f be scalar, vector, or matrix valued. Then
∣ ∑ j ≥ − 1 2 2 σ j ∫ T 2 Δ j f : [ Δ j , u ⋅ ∇ ] f d x ∣ ≤ C σ ∥ ∇ u ∥ B ∞ , 1 0 ∥ f ∥ H σ 2 . \left|\sum_{j\geq-1}2^{2\sigma j}\int_{\mathbb{T}^2}\Delta_j f:[\Delta_j,u\cdot\nabla]f\,\mathrm{d}x\right|\leq C_\sigma\|\nabla u\|_{B^0_{\infty,1}}\|f\|_{H^\sigma}^2. j ≥ − 1 ∑ 2 2 σ j ∫ T 2 Δ j f : [ Δ j , u ⋅ ∇ ] f d x ≤ C σ ∥∇ u ∥ B ∞ , 1 0 ∥ f ∥ H σ 2 . Consequently, for every integer $m\geq 1,
∣ ∑ ∣ γ ∣ ≤ m ∫ T 2 ∂ γ f : [ ∂ γ , u ⋅ ∇ ] f d x ∣ ≤ C m ∥ ∇ u ∥ B ∞ , 1 0 ∥ f ∥ H m 2 . \left|\sum_{|\gamma|\leq m}\int_{\mathbb{T}^2}\partial^\gamma f:[\partial^\gamma,u\cdot\nabla]f\,\mathrm{d}x\right|\leq C_m\|\nabla u\|_{B^0_{\infty,1}}\|f\|_{H^m}^2. ∣ γ ∣ ≤ m ∑ ∫ T 2 ∂ γ f : [ ∂ γ , u ⋅ ∇ ] f d x ≤ C m ∥∇ u ∥ B ∞ , 1 0 ∥ f ∥ H m 2 .
Proof. The proof is included to make the endpoint use of B ∞ , 1 0 B^0_{\infty,1} B ∞ , 1 0 explicit. Set b k = ∥ Δ k ∇ u ∥ L ∞ b_k=\|\Delta_k\nabla u\|_{L^\infty} b k = ∥ Δ k ∇ u ∥ L ∞ , so that ∑ k ≥ − 1 b k ≲ ∥ ∇ u ∥ B ∞ , 1 0 \sum_{k\geq-1}b_k\lesssim\|\nabla u\|_{B^0_{\infty,1}} ∑ k ≥ − 1 b k ≲ ∥∇ u ∥ B ∞ , 1 0 . Bony’s splitting of u ⋅ ∇ f u\cdot\nabla f u ⋅ ∇ f gives
[ Δ j , u ⋅ ∇ ] f = R j 1 + R j 2 + R j 3 , [\Delta_j,u\cdot\nabla]f=R_j^1+R_j^2+R_j^3, [ Δ j , u ⋅ ∇ ] f = R j 1 + R j 2 + R j 3 ,
where, up to harmless finite changes of the index ranges,
R j 1 = ∑ ∣ k − j ∣ ≤ 4 [ Δ j , S k − 1 u ⋅ ∇ ] Δ k f , R_j^1=\sum_{|k-j|\leq4}[\Delta_j,S_{k-1}u\cdot\nabla]\Delta_k f, R j 1 = ∣ k − j ∣ ≤ 4 ∑ [ Δ j , S k − 1 u ⋅ ∇ ] Δ k f ,
R j 2 R_j^2 R j 2 contains the high-frequency velocity acting on the low-frequency part of f f f , and R j 3 R_j^3 R j 3 contains the high-high remainder. The kernel representation of Δ j \Delta_j Δ j gives
∥ R j 1 ∥ L 2 ≤ C ∑ ∣ k − j ∣ ≤ 4 ∥ ∇ S k − 1 u ∥ L ∞ ∥ Δ k f ∥ L 2 ≤ C ∥ ∇ u ∥ B ∞ , 1 0 ∑ ∣ k − j ∣ ≤ 4 ∥ Δ k f ∥ L 2 . \|R_j^1\|_{L^2}\leq C\sum_{|k-j|\leq4}\|\nabla S_{k-1}u\|_{L^\infty}\|\Delta_k f\|_{L^2}\leq C\|\nabla u\|_{B^0_{\infty,1}}\sum_{|k-j|\leq4}\|\Delta_k f\|_{L^2}. ∥ R j 1 ∥ L 2 ≤ C ∣ k − j ∣ ≤ 4 ∑ ∥∇ S k − 1 u ∥ L ∞ ∥ Δ k f ∥ L 2 ≤ C ∥∇ u ∥ B ∞ , 1 0 ∣ k − j ∣ ≤ 4 ∑ ∥ Δ k f ∥ L 2 .
For the two remaining terms, the derivative on the high-frequency velocity is recorded in the summable sequence b k b_k b k , and the dyadic interactions are localized by ℓ 1 \ell^1 ℓ 1 kernels. More precisely,
2 σ j ∥ R j 2 + R j 3 ∥ L 2 ≤ C σ ∑ k ≥ − 1 2 − c σ ∣ j − k ∣ b k 2 σ k ∥ Δ k f ∥ L 2 , 2^{\sigma j}\|R_j^2+R_j^3\|_{L^2}\leq C_\sigma\sum_{k\geq-1}2^{-c_\sigma|j-k|}b_k2^{\sigma k}\|\Delta_k f\|_{L^2}, 2 σ j ∥ R j 2 + R j 3 ∥ L 2 ≤ C σ k ≥ − 1 ∑ 2 − c σ ∣ j − k ∣ b k 2 σ k ∥ Δ k f ∥ L 2 ,
for some c σ > 0 c_\sigma>0 c σ > 0 . The estimate follows by Young’s inequality for sequences, using b ∈ ℓ 1 b\in\ell^1 b ∈ ℓ 1 . Equivalently, there is a sequence c j ∈ ℓ 2 c_j\in\ell^2 c j ∈ ℓ 2 , ∥ c ∥ ℓ 2 ≤ 1 \|c\|_{\ell^2}\leq1 ∥ c ∥ ℓ 2 ≤ 1 , such that
2 σ j ∥ [ Δ j , u ⋅ ∇ ] f ∥ L 2 ≤ C σ c j ∥ ∇ u ∥ B ∞ , 1 0 ∥ f ∥ H σ . 2^{\sigma j}\|[\Delta_j,u\cdot\nabla]f\|_{L^2}\leq C_\sigma c_j\|\nabla u\|_{B^0_{\infty,1}}\|f\|_{H^\sigma}. 2 σ j ∥ [ Δ j , u ⋅ ∇ ] f ∥ L 2 ≤ C σ c j ∥∇ u ∥ B ∞ , 1 0 ∥ f ∥ H σ .
Multiplying by 2 σ j ∥ Δ j f ∥ L 2 2^{\sigma j}\|\Delta_jf\|_{L^2} 2 σ j ∥ Δ j f ∥ L 2 and summing in j j j gives (3.2). The divergence-free condition has removed the principal transport contribution ∫ u ⋅ ∇ ∣ Δ j f ∣ 2 / 2 = 0 \int u\cdot\nabla|\Delta_jf|^2/2=0 ∫ u ⋅ ∇∣ Δ j f ∣ 2 /2 = 0 . The integer derivative form (3.2) follows from the equivalence between the nonhomogeneous dyadic H m H^m H m norm and the classical derivative norm. □ \square □
Remark 3.1 (Scope of fractional estimates). Lemma 3.2 is the fractional Littlewood–Paley input used in the paper. The high-order continuation estimates themselves are proved for integer s = m ≥ 3 s=m\geq3 s = m ≥ 3 , where all coefficient commutators are expanded by ordinary derivatives. Thus the proof does not rely on an unstated fractional high-order commutator calculus for the full nonlinear system, nor on a low-regularity passage below the strong-solution class of Remark 2.1.
Entropy and Positive-Cone Geometry
Definition 4.1 (Free energy). For a smooth positive-cone solution define
E ( t ) = 1 2 ∫ T 2 ∣ u ∣ 2 d x + α 2 ∫ T 2 ( tr A − log det A − 2 ) d x . \mathcal{E}(t)=\frac{1}{2}\int_{\mathbb{T}^2}|u|^2\,dx+\frac{\alpha}{2}\int_{\mathbb{T}^2}\left(\operatorname{tr}A-\log\det A-2\right)\,dx. E ( t ) = 2 1 ∫ T 2 ∣ u ∣ 2 d x + 2 α ∫ T 2 ( tr A − log det A − 2 ) d x .
Proposition 4.1 (Entropy identity). Every smooth solution of (1)–(2) satisfies
d d t E ( t ) + ν ∫ T 2 ∣ ∇ u ∣ 2 d x + α 2 λ ∫ T 2 tr ( A + A − 1 − 2 I ) d x = 0. \frac{d}{dt}\mathcal{E}(t)+\nu\int_{\mathbb{T}^2}|\nabla u|^2\,dx+\frac{\alpha}{2\lambda}\int_{\mathbb{T}^2}\operatorname{tr}\left(A+A^{-1}-2I\right)\,dx=0. d t d E ( t ) + ν ∫ T 2 ∣∇ u ∣ 2 d x + 2 λ α ∫ T 2 tr ( A + A − 1 − 2 I ) d x = 0. Proof. Testing the momentum equation by u u u gives
1 2 d d t ∫ ∣ u ∣ 2 d x + ν ∫ ∣ ∇ u ∣ 2 d x = − α ∫ ( A − I ) : ∇ u d x . \frac{1}{2}\frac{d}{dt}\int|u|^2\,dx+\nu\int|\nabla u|^2\,dx=-\alpha\int(A-I):\nabla u\,dx. 2 1 d t d ∫ ∣ u ∣ 2 d x + ν ∫ ∣∇ u ∣ 2 d x = − α ∫ ( A − I ) : ∇ u d x . Testing the conformation equation by I − A − 1 I-A^{-1} I − A − 1 gives
d d t ∫ ( tr A − log det A − 2 ) d x = 2 ∫ ( A − I ) : ∇ u d x − λ − 1 ∫ tr ( A + A − 1 − 2 I ) d x . \frac{d}{dt}\int\left(\operatorname{tr}A-\log\det A-2\right)\,dx=2\int(A-I):\nabla u\,dx-\lambda^{-1}\int\operatorname{tr}\left(A+A^{-1}-2I\right)\,dx. d t d ∫ ( tr A − log det A − 2 ) d x = 2 ∫ ( A − I ) : ∇ u d x − λ − 1 ∫ tr ( A + A − 1 − 2 I ) d x . The stretching terms cancel after multiplication by α / 2 \alpha/2 α /2 .
Lemma 4.1 (Cone propagation). If A 0 ( x ) ∈ S + + 2 A_0(x)\in\mathbb{S}_{++}^2 A 0 ( x ) ∈ S ++ 2 and ∫ 0 T ∥ ∇ u ( t ) ∥ L ∞ d t < ∞ \int_0^T\|\nabla u(t)\|_{L^\infty}\,dt<\infty ∫ 0 T ∥∇ u ( t ) ∥ L ∞ d t < ∞ , then A ( t , x ) ∈ S + + 2 A(t,x)\in\mathbb{S}_{++}^2 A ( t , x ) ∈ S ++ 2 for all t ≤ T t\le T t ≤ T . Moreover, along characteristics, the extreme eigenvalues obey
λ max A ( t ) ≤ C ( λ max A 0 + 1 ) exp ( 2 ∫ 0 t ∥ ∇ u ( s ) ∥ L ∞ d s ) , \lambda_{\max}A(t)\le C\left(\lambda_{\max}A_0+1\right)\exp\left(2\int_0^t\|\nabla u(s)\|_{L^\infty}\,ds\right), λ m a x A ( t ) ≤ C ( λ m a x A 0 + 1 ) exp ( 2 ∫ 0 t ∥∇ u ( s ) ∥ L ∞ d s ) , and an analogous lower bound holds for λ min A ( t ) \lambda_{\min}A(t) λ m i n A ( t ) .
Proof. Along a trajectory X ( t ) X(t) X ( t ) , the matrix satisfies
d d t A = ∇ u A + A ( ∇ u ) T − λ − 1 ( A − I ) . \frac{d}{dt}A=\nabla u\,A+A(\nabla u)^{T}-\lambda^{-1}(A-I). d t d A = ∇ u A + A ( ∇ u ) T − λ − 1 ( A − I ) . Testing against unit vectors and applying Gronwall gives the upper spectral bound. Applying the same argument to A − 1 A^{-1} A − 1 , whose equation is obtained by differentiating A − 1 A = I A^{-1}A=I A − 1 A = I , gives the lower bound.
Logarithmic Coordinates and Good Unknowns
Let B = Log A B=\operatorname{Log}A B = Log A . The derivative of the matrix exponential is
d exp B ( C ) = ∫ 0 1 e ( 1 − θ ) B C e θ B d θ . d\exp_B(C)=\int_0^1 e^{(1-\theta)B}Ce^{\theta B}\,d\theta. d exp B ( C ) = ∫ 0 1 e ( 1 − θ ) B C e θ B d θ .
If the spectrum of B B B is contained in [ − K , K ] [-K,K] [ − K , K ] , then
c K ∣ C ∣ ≤ ∣ d exp B ( C ) ∣ ≤ C K ∣ C ∣ ( C = C T ) , c_K|C|\le|d\exp_B(C)|\le C_K|C|\qquad(C=C^T), c K ∣ C ∣ ≤ ∣ d exp B ( C ) ∣ ≤ C K ∣ C ∣ ( C = C T ) ,
and the constants depend only on K K K . Thus the logarithmic coordinate is not merely a change of notation: it gives an elliptic chart on every compact spectral subset of the positive cone.
Lemma 5.1 (Sobolev calculus for the exponential chart). Let s > 1 s>1 s > 1 in two dimensions and suppose ∥ B ∥ L ∞ ≤ K \lVert B\rVert_{L^\infty}\le K ∥ B ∥ L ∞ ≤ K . Then
∥ e B − I ∥ H s ≤ C K ( 1 + ∥ B ∥ H s ) . \lVert e^B-I\rVert_{H^s}\le C_K\left(1+\lVert B\rVert_{H^s}\right). ∥ e B − I ∥ H s ≤ C K ( 1 + ∥ B ∥ H s ) . and, for two symmetric fields B 1 , B 2 B_1,B_2 B 1 , B 2 with the same L ∞ L^\infty L ∞ bound,
∥ e B 1 − e B 2 ∥ H s − 1 ≤ C K ( 1 + ∥ B 1 ∥ H s + ∥ B 2 ∥ H s ) ∥ B 1 − B 2 ∥ H s − 1 . \lVert e^{B_1}-e^{B_2}\rVert_{H^{s-1}}\le C_K\left(1+\lVert B_1\rVert_{H^s}+\lVert B_2\rVert_{H^s}\right)\lVert B_1-B_2\rVert_{H^{s-1}}. ∥ e B 1 − e B 2 ∥ H s − 1 ≤ C K ( 1 + ∥ B 1 ∥ H s + ∥ B 2 ∥ H s ) ∥ B 1 − B 2 ∥ H s − 1 . The same estimates hold for B ↦ ( d exp B ) − 1 B\mapsto(\mathrm{d}\exp_B)^{-1} B ↦ ( d exp B ) − 1 on symmetric matrices.
Proof. Use the integral representation
e B − I = ∫ 0 1 e θ B B d θ . e^B-I=\int_0^1 e^{\theta B}B\,\mathrm{d}\theta. e B − I = ∫ 0 1 e θ B B d θ . and Lemma 3.1 . The Lipschitz bound follows from
e B 1 − e B 2 = ∫ 0 1 d exp B 2 + θ ( B 1 − B 2 ) ( B 1 − B 2 ) d θ . e^{B_1}-e^{B_2}=\int_0^1\mathrm{d}\exp_{B_2+\theta(B_1-B_2)}(B_1-B_2)\,\mathrm{d}\theta. e B 1 − e B 2 = ∫ 0 1 d exp B 2 + θ ( B 1 − B 2 ) ( B 1 − B 2 ) d θ . The inverse derivative is controlled by (5) and the smooth dependence of the inverse operator on B B B on compact spectral sets.
In logarithmic variables, the conformation equation can be written as
d exp B ( ∂ t B + u ⋅ ∇ B ) = ∇ u e B + e B ( ∇ u ) T − λ − 1 ( e B − I ) . \mathrm{d}\exp_B\left(\partial_tB+u\cdot\nabla B\right)=\nabla u e^B+e^B(\nabla u)^T-\lambda^{-1}(e^B-I). d exp B ( ∂ t B + u ⋅ ∇ B ) = ∇ u e B + e B ( ∇ u ) T − λ − 1 ( e B − I ) .
When the spectrum of B B B is bounded, d exp B \mathrm{d}\exp_B d exp B and its inverse are bounded on symmetric matrices. Thus the equation is transport-dominated, but the source is nonlinear in B B B and ∇ u \nabla u ∇ u .
The naive differentiated equation contains two bad-looking terms. The first is the standard transport commutator [ ∂ γ , u ⋅ ∇ ] B [\partial^\gamma,u\cdot\nabla]B [ ∂ γ , u ⋅ ∇ ] B , which is controlled by a critical Besov norm of ∇ u \nabla u ∇ u . The second is more specific to the positive cone: when one differentiates e B e^B e B , the top derivative does not simply equal e B ∂ γ B e^B\partial^\gamma B e B ∂ γ B . It contains non-commuting matrix factors and lower-order products. The good unknown below subtracts the paralinear part of this matrix-exponential commutator. Geometrically, it applies the inverse tangent map of the exponential chart and therefore pulls the differentiated conformation tensor ∂ γ A \partial^\gamma A ∂ γ A back to the logarithmic tangent space at B B B . This is the point at which the positive-cone geometry enters the continuation argument.
Definition 5.1 (Logarithmic good unknown). For a multi-index γ \gamma γ , define
G γ = ( d exp B ) − 1 ( ∂ γ ( e B ) ) . G_\gamma=(\mathrm{d}\exp_B)^{-1}\left(\partial^\gamma(e^B)\right). G γ = ( d exp B ) − 1 ( ∂ γ ( e B ) ) . By Lemma 5.2,
G γ = ∂ γ B + ( d exp B ) − 1 C γ ( B ) . G_\gamma=\partial^\gamma B+(\mathrm{d}\exp_B)^{-1}C_\gamma(B). G γ = ∂ γ B + ( d exp B ) − 1 C γ ( B ) . Thus G γ G_\gamma G γ is an Alinhac-type correction of ∂ γ B \partial^\gamma B ∂ γ B : it is defined intrinsically from the differentiated conformation tensor and differs from ∂ γ B \partial^\gamma B ∂ γ B only by lower-order matrix-exponential commutators.
Lemma 5.2 (Commutator structure). Let m ≥ 2 m\ge2 m ≥ 2 , 0 < ε < 1 0<\varepsilon<1 0 < ε < 1 , and assume ∥ B ∥ L ∞ ≤ K \lVert B\rVert_{L^\infty}\le K ∥ B ∥ L ∞ ≤ K . Then for ∣ γ ∣ = m |\gamma|=m ∣ γ ∣ = m ,
∂ γ ( e B ) = d exp B ( ∂ γ B ) + C γ ( B ) , \partial^\gamma(e^B)=\mathrm{d}\exp_B(\partial^\gamma B)+C_\gamma(B), ∂ γ ( e B ) = d exp B ( ∂ γ B ) + C γ ( B ) , where
∥ C γ ( B ) ∥ L 2 ≤ C K ( 1 + ∥ B ∥ H 1 + ε 2 ) ( 1 + ∥ B ∥ H m − 1 ) . \lVert C_\gamma(B)\rVert_{L^2}\le C_K\left(1+\lVert B\rVert_{H^{1+\varepsilon}}^2\right)\left(1+\lVert B\rVert_{H^{m-1}}\right). ∥ C γ ( B ) ∥ L 2 ≤ C K ( 1 + ∥ B ∥ H 1 + ε 2 ) ( 1 + ∥ B ∥ H m − 1 ) .
Proof. Expanding ∂ γ ( e B ) \partial^\gamma(e^B) ∂ γ ( e B ) by repeated differentiation of the integral formula for d exp B d\exp_B d exp B , the unique term containing m m m derivatives of B B B is d exp B ( ∂ γ B ) d\exp_B(\partial^\gamma B) d exp B ( ∂ γ B ) . Every other term is a finite product with at least two positive-order derivatives of B B B and total order at most m m m . In two dimensions the H 1 + ε H^{1+\varepsilon} H 1 + ε factor controls the low-order products, while the remaining derivative is placed in H m − 1 H^{m-1} H m − 1 . The spectral bound absorbs the exponential coefficients.
Proposition 5.1 (Good-unknown estimate). Let s ∈ N s \in\mathbb{N} s ∈ N , s ≥ 3 s \ge3 s ≥ 3 , 0 < ε < 1 0 < \varepsilon< 1 0 < ε < 1 , and let ( u , B ) (u,B) ( u , B ) be smooth. If B B B has bounded spectrum on [ 0 , T ] [0,T] [ 0 , T ] , then for every 2 ≤ m ≤ s 2 \le m \le s 2 ≤ m ≤ s ,
d d t ∥ B ∥ H m ≤ C ( 1 + ∥ ∇ u ∥ B ∞ , 1 ∞ + ∥ B ∥ H 1 + ε 2 ) ( 1 + ∥ B ∥ H m + ∥ u ∥ H m + 1 ) . \frac{d}{dt}\|B\|_{H^m} \le C\left(1+\|\nabla u\|_{B^\infty_{\infty,1}}+\|B\|_{H^{1+\varepsilon}}^2\right)\left(1+\|B\|_{H^m}+\|u\|_{H^{m+1}}\right). d t d ∥ B ∥ H m ≤ C ( 1 + ∥∇ u ∥ B ∞ , 1 ∞ + ∥ B ∥ H 1 + ε 2 ) ( 1 + ∥ B ∥ H m + ∥ u ∥ H m + 1 ) .
Proof. Write the logarithmic equation as
∂ t B + u ⋅ ∇ B = F ( B , ∇ u ) , F ( B , M ) = ( d exp B ) − 1 ( M e B + e B M T − λ − 1 ( e B − I ) ) . \partial_t B+u\cdot\nabla B=\mathfrak{F}(B,\nabla u),\qquad\mathfrak{F}(B,M)=(d\exp_B)^{-1}\left(Me^B+e^BM^T-\lambda^{-1}(e^B-I)\right). ∂ t B + u ⋅ ∇ B = F ( B , ∇ u ) , F ( B , M ) = ( d exp B ) − 1 ( M e B + e B M T − λ − 1 ( e B − I ) ) .
On a fixed spectral window, Lemma 5.1 and Lemma 3.1 give the tame bound
∥ F ( B , ∇ u ) ∥ H m ≤ C K ( 1 + ∥ B ∥ H 1 + ε 2 ) ( 1 + ∥ B ∥ H m + ∥ u ∥ H m + 1 ) . \|\mathfrak{F}(B,\nabla u)\|_{H^m}\le C_K\left(1+\|B\|_{H^{1+\varepsilon}}^2\right)\left(1+\|B\|_{H^m}+\|u\|_{H^{m+1}}\right). ∥ F ( B , ∇ u ) ∥ H m ≤ C K ( 1 + ∥ B ∥ H 1 + ε 2 ) ( 1 + ∥ B ∥ H m + ∥ u ∥ H m + 1 ) .
Indeed the top derivative falls either on ∇ u \nabla u ∇ u , giving ∥ u ∥ H m + 1 \|u\|_{H^{m+1}} ∥ u ∥ H m + 1 , or on one copy of B B B in the exponential coefficients, giving a term linear in ∥ B ∥ H m \|B\|_{H^m} ∥ B ∥ H m ; all remaining factors are placed in H 1 + ε ↪ L ∞ H^{1+\varepsilon}\hookrightarrow L^\infty H 1 + ε ↪ L ∞ . Differentiating the transport equation and pairing with ∂ γ B \partial^\gamma B ∂ γ B yields
1 2 d d t ∥ ∂ γ B ∥ L 2 2 = − ∫ T 2 ∂ γ B [ ∂ γ , u ⋅ ∇ ] B d x + ∫ T 2 ∂ γ B ∂ γ F d x . \frac{1}{2}\frac{d}{dt}\|\partial^\gamma B\|_{L^2}^2=-\int_{\mathbb{T}^2}\partial^\gamma B[\partial^\gamma,u\cdot\nabla]B\,dx+\int_{\mathbb{T}^2}\partial^\gamma B\partial^\gamma\mathfrak{F}\,dx. 2 1 d t d ∥ ∂ γ B ∥ L 2 2 = − ∫ T 2 ∂ γ B [ ∂ γ , u ⋅ ∇ ] B d x + ∫ T 2 ∂ γ B ∂ γ F d x .
The endpoint transport commutator is bounded by Lemma 3.2:
∣ ∫ T 2 ∂ γ B [ ∂ γ , u ⋅ ∇ ] B d x ∣ ≤ C ∥ ∇ u ∥ B ∞ , 1 ∞ ∥ B ∥ H m 2 . \left|\int_{\mathbb{T}^2}\partial^\gamma B[\partial^\gamma,u\cdot\nabla]B\,dx\right|\le C\|\nabla u\|_{B^\infty_{\infty,1}}\|B\|_{H^m}^2. ∫ T 2 ∂ γ B [ ∂ γ , u ⋅ ∇ ] B d x ≤ C ∥∇ u ∥ B ∞ , 1 ∞ ∥ B ∥ H m 2 .
After summing over ∣ γ ∣ ≤ m |\gamma|\le m ∣ γ ∣ ≤ m , using (5), and taking the square root of the resulting differential inequality, we obtain (5.1). The good unknown G γ G_\gamma G γ gives an equivalent way to phrase the same estimate in the differentiated conformation variable ∂ γ A \partial^\gamma A ∂ γ A : by Lemma 5.2, G γ − ∂ γ B G_\gamma-\partial^\gamma B G γ − ∂ γ B is lower order and is absorbed by the ∥ B ∥ H 1 + ε 2 ( 1 + ∥ B ∥ H m ) \|B\|_{H^{1+\varepsilon}}^2(1+\|B\|_{H^m}) ∥ B ∥ H 1 + ε 2 ( 1 + ∥ B ∥ H m ) term.
Active-Deviatoric Projection and Endpoint Continuation
Let ω = curl u \omega=\operatorname{curl}u ω = curl u . The endpoint velocity modulus is controlled by the periodic Biot–Savart law,
∥ ∇ u ∥ B ∞ , 1 ∞ ≤ C ( ∥ ω ∥ B ∞ , 1 ∞ + ∥ u ∥ L 2 ) . \|\nabla u\|_{B^\infty_{\infty,1}}\le C\left(\|\omega\|_{B^\infty_{\infty,1}}+\|u\|_{L^2}\right). ∥∇ u ∥ B ∞ , 1 ∞ ≤ C ( ∥ ω ∥ B ∞ , 1 ∞ + ∥ u ∥ L 2 ) .
The logical dependence of the argument is as follows.
Table 1.
We begin with the structural projection which is independent of the particular Oldroyd–B or FENE-P closure.
Proposition 6.1 (Pressure-free active-deviatoric projection). Let T : S + + 2 → S 2 T:\mathbb{S}_{++}^2\to\mathbb{S}^2 T : S ++ 2 → S 2 be a smooth spectral isotropic stress law in two space dimensions:
T ( Q C Q T ) = Q T ( C ) Q T , Q ∈ O ( 2 ) . T(QCQ^{\mathsf T})=QT(C)Q^{\mathsf T},\qquad Q\in O(2). T ( QC Q T ) = QT ( C ) Q T , Q ∈ O ( 2 ) . For every compact K ⋐ S + + 2 K\Subset\mathbb{S}_{++}^2 K ⋐ S ++ 2 , there are smooth scalar functions q 0 , q 1 q_0,q_1 q 0 , q 1 , depending on s = tr C s=\operatorname{tr}C s = tr C and d = det C d=\det C d = det C , such that
T ( C ) = q 0 ( tr C , det C ) I + q 1 ( tr C , det C ) C . T(C)=q_0(\operatorname{tr}C,\det C)I+q_1(\operatorname{tr}C,\det C)C. T ( C ) = q 0 ( tr C , det C ) I + q 1 ( tr C , det C ) C . Writing C = a I + Y C=aI+Y C = a I + Y , a = 1 2 tr C a=\frac{1}{2}\operatorname{tr}C a = 2 1 tr C , Y = C ∘ Y=C^\circ Y = C ∘ , one has
T ( C ) = ( q 0 + a q 1 ) I + q 1 Y , T ( C ) ∘ = q 1 Y . T(C)=(q_0+aq_1)I+q_1Y,\qquad T(C)^\circ=q_1Y. T ( C ) = ( q 0 + a q 1 ) I + q 1 Y , T ( C ) ∘ = q 1 Y . Consequently, in the incompressible momentum equation, the isotropic part is a pressure mode and only div ( q 1 Y ) \operatorname{div}(q_1Y) div ( q 1 Y ) enters the vorticity and the top-order velocity–conformation coupling. If the conformation equation has upper-convected stretching
D t C = ∇ u C + C ( ∇ u ) T + R ( C ) , D_tC=\nabla u\,C+C(\nabla u)^{\mathsf T}+R(C), D t C = ∇ u C + C ( ∇ u ) T + R ( C ) , then the principal active coupling cancels when the trace-free equation is tested with weight q 1 / a q_1/a q 1 / a .
Proof. Spectral isotropy implies that T ( C ) T(C) T ( C ) and C C C are coaxial. If C C C has distinct eigenvalues λ 1 , λ 2 \lambda_1,\lambda_2 λ 1 , λ 2 and T ( C ) T(C) T ( C ) has corresponding eigenvalues t 1 , t 2 t_1,t_2 t 1 , t 2 , define
q 1 = t 1 − t 2 λ 1 − λ 2 , q 0 = λ 1 t 2 − λ 2 t 1 λ 1 − λ 2 . q_1=\frac{t_1-t_2}{\lambda_1-\lambda_2},\qquad q_0=\frac{\lambda_1t_2-\lambda_2t_1}{\lambda_1-\lambda_2}. q 1 = λ 1 − λ 2 t 1 − t 2 , q 0 = λ 1 − λ 2 λ 1 t 2 − λ 2 t 1 .
The smooth spectral representation theorem, equivalently the smooth divided-difference lemma for symmetric spectral functions, extends q 0 , q 1 q_0,q_1 q 0 , q 1 smoothly across the repeated-eigenvalue set. Since symmetric smooth functions of λ 1 , λ 2 \lambda_1,\lambda_2 λ 1 , λ 2 are smooth functions of tr C \operatorname{tr}C tr C and det C \det C det C on compact positive-cone windows, the displayed representation follows.
The pressure projection is then algebraic:
div T ( C ) = ∇ ( q 0 + a q 1 ) + div ( q 1 Y ) . \operatorname{div} T(C) = \nabla(q_0 + aq_1) + \operatorname{div}(q_1Y). div T ( C ) = ∇ ( q 0 + a q 1 ) + div ( q 1 Y ) .
The first term is absorbed into the pressure and has zero curl. Finally, the trace-free part of the upper-convected stretching is
{ ∇ u C + C ( ∇ u ) T } ∘ = 2 a S ( u ) + [ ∇ u Y + Y ( ∇ u ) T ] ∘ . \left\{\nabla u\,C + C(\nabla u)^{\mathsf T}\right\}^{\circ} = 2aS(u) + \left[\nabla u\,Y + Y(\nabla u)^{\mathsf T}\right]^{\circ}. { ∇ u C + C ( ∇ u ) T } ∘ = 2 a S ( u ) + [ ∇ u Y + Y ( ∇ u ) T ] ∘ .
At derivative order β \beta β , the velocity equation contributes
− α ∫ q 1 ∂ β Y : S ( ∂ β u ) d x , -\alpha\int q_1\,\partial^\beta Y:S(\partial^\beta u)\,\mathrm{d}x, − α ∫ q 1 ∂ β Y : S ( ∂ β u ) d x ,
while testing the trace-free equation by ( α / 2 ) ( q 1 / a ) ∂ β Y (\alpha/2)(q_1/a)\partial^\beta Y ( α /2 ) ( q 1 / a ) ∂ β Y gives
α 2 ∫ q 1 a ∂ β Y : 2 a S ( ∂ β u ) d x = α ∫ q 1 ∂ β Y : S ( ∂ β u ) d x . \frac{\alpha}{2}\int\frac{q_1}{a}\partial^\beta Y:2aS(\partial^\beta u)\,\mathrm{d}x
=
\alpha\int q_1\partial^\beta Y:S(\partial^\beta u)\,\mathrm{d}x. 2 α ∫ a q 1 ∂ β Y : 2 a S ( ∂ β u ) d x = α ∫ q 1 ∂ β Y : S ( ∂ β u ) d x .
The principal terms cancel exactly.
We next record the compact-window quotient estimate abstracted from the active-deviatoric projection above. This is an analytic template for the high-order proof, not a proposal for a new class of physical constitutive laws.
Definition 6.1 (Compact-window quotient closure). Let K ⋐ S + + 2 K \Subset S_{++}^{2} K ⋐ S ++ 2 be a compact conformation window and write
C = a I + Y , a = 1 2 tr C , Y = C ∘ , η = ∣ Y ∣ 2 . C=aI+Y,\qquad a=\frac{1}{2}\operatorname{tr}C,\qquad Y=C^\circ,\qquad\eta=\lvert Y\rvert^2. C = a I + Y , a = 2 1 tr C , Y = C ∘ , η = ∣ Y ∣ 2 . A compact-window closure is spectrally admissible on K K K if, in the variables ( a , Y ) (a,Y) ( a , Y ) , it has the form
∂ t u + u ⋅ ∇ u − ν Δ u + ∇ p = α div ( τ ( a , η ) Y ) , (8) \partial_tu+u\cdot\nabla u-\nu\Delta u+\nabla p=\alpha\operatorname{div}(\tau(a,\eta)Y), \tag*{(8)} ∂ t u + u ⋅ ∇ u − ν Δ u + ∇ p = α div ( τ ( a , η ) Y ) , (8) D t a + λ − 1 G ( a , η ) = S ( u ) : Y , (9) D_ta+\lambda^{-1}G(a,\eta)=S(u):Y, \tag*{(9)} D t a + λ − 1 G ( a , η ) = S ( u ) : Y , (9) D t Y + λ − 1 μ ( a , η ) Y = 2 a S ( u ) + [ ∇ u Y + Y ( ∇ u ) T ] ∘ . (10) D_tY+\lambda^{-1}\mu(a,\eta)Y=2aS(u)+\left[\nabla u\,Y+Y(\nabla u)^{\mathsf T}\right]^\circ. \tag*{(10)} D t Y + λ − 1 μ ( a , η ) Y = 2 a S ( u ) + [ ∇ u Y + Y ( ∇ u ) T ] ∘ . (10) The coefficients are smooth on the range of ( a , η ) (a,\eta) ( a , η ) determined by K K K . Moreover, there are constants a 0 , τ 0 , μ 0 , κ > 0 a_0,\tau_0,\mu_0,\kappa>0 a 0 , τ 0 , μ 0 , κ > 0 and a ∗ a_\ast a ∗ such that
a ≥ a 0 , τ ( a , η ) ≥ τ 0 , μ ( a , η ) ≥ μ 0 , G ( a ∗ , 0 ) = 0 , ∂ a G ( a , η ) ≥ κ a\ge a_0,\qquad\tau(a,\eta)\ge\tau_0,\qquad\mu(a,\eta)\ge\mu_0,\qquad G(a_\ast,0)=0,\qquad\partial_aG(a,\eta)\ge\kappa a ≥ a 0 , τ ( a , η ) ≥ τ 0 , μ ( a , η ) ≥ μ 0 , G ( a ∗ , 0 ) = 0 , ∂ a G ( a , η ) ≥ κ on this range. These are compact-window continuation assumptions. They do not by themselves assert a thermodynamic free-energy law for arbitrary τ , μ , G \tau,\mu,G τ , μ , G .
which gives (11) . The relaxation part of (9) –(10) gives D t a = − λ − 1 G D_t a=-\lambda^{-1}G D t a = − λ − 1 G and D t η = − 2 λ − 1 μ η D_t\eta=-2\lambda^{-1}\mu\eta D t η = − 2 λ − 1 μ η ; therefore (12) is precisely the local entropy production inequality for F ~ \widetilde{\mathscr F} F .
The endpoint theorem below does not require (11) –(12) ; it requires only the continuation assumptions in Definition 6.1. Thus the abstract theorem is not presented as a classification of all thermodynamic isotropic polymer models. It is a pressure-quotient continuation mechanism, and thermodynamic closures are recovered by adding the compatibility conditions above. The Oldroyd–B and FENE-P specializations satisfy these conditions with the standard positive-cone free energies.
Example 6.1 (A free-energy-compatible channel with a genuine quadratic component). Let 0 < θ < 1 0<\theta<1 0 < θ < 1 and define, for C ∈ S + + d C\in\mathbb{S}^{d}_{++} C ∈ S ++ d
F ~ θ ( C ) = tr ( C − Log C − I ) + θ 2 tr ( C − I ) 2 . \widetilde{\mathscr F}_{\theta}(C)=\operatorname{tr}(C-\operatorname{Log}C-I)+\frac{\theta}{2}\operatorname{tr}(C-I)^2. F θ ( C ) = tr ( C − Log C − I ) + 2 θ tr ( C − I ) 2 . Take the elastic stress and relaxation to be
T θ ( C ) = C ∂ C F ~ θ ( C ) = C − I + θ ( C 2 − C ) , R ( C ) = − ( C − I ) . T_{\theta}(C)=C\,\partial_C\widetilde{\mathscr F}_{\theta}(C)=C-I+\theta(C^2-C),\qquad R(C)=-(C-I). T θ ( C ) = C ∂ C F θ ( C ) = C − I + θ ( C 2 − C ) , R ( C ) = − ( C − I ) . The relaxation dissipates the same density, since
∂ C F ~ θ ( C ) : ( C − I ) = tr ( C + C − 1 − 2 I ) + θ ∣ C − I ∣ 2 ≥ 0. \partial_C\widetilde{\mathscr F}_{\theta}(C):(C-I)=\operatorname{tr}(C+C^{-1}-2I)+\theta|C-I|^2\ge0. ∂ C F θ ( C ) : ( C − I ) = tr ( C + C − 1 − 2 I ) + θ ∣ C − I ∣ 2 ≥ 0. In two dimensions, writing C = a I + Y C=aI+Y C = a I + Y , η = ∣ Y ∣ 2 \eta=|Y|^2 η = ∣ Y ∣ 2 , this example gives
G ( a , η ) = a − 1 , μ ( a , η ) = 1 , τ θ ( a , η ) = 1 + θ ( 2 a − 1 ) . G(a,\eta)=a-1,\qquad\mu(a,\eta)=1,\qquad\tau_{\theta}(a,\eta)=1+\theta(2a-1). G ( a , η ) = a − 1 , μ ( a , η ) = 1 , τ θ ( a , η ) = 1 + θ ( 2 a − 1 ) . Hence ∂ a G = 1 \partial_aG=1 ∂ a G = 1 , μ = 1 \mu=1 μ = 1 , and τ θ ≥ 1 − θ > 0 \tau_{\theta}\ge1-\theta>0 τ θ ≥ 1 − θ > 0 on the positive cone. It is therefore a member of Definition 6.1 and satisfies the compatibility conditions in Remark 6.1. In three dimensions its local Cayley–Hamilton representation is explicit,
T θ ( C ) = − I + ( 1 − θ ) C + θ C 2 , T_{\theta}(C)=-I+(1-\theta)C+\theta C^2, T θ ( C ) = − I + ( 1 − θ ) C + θ C 2 , so q 2 ≡ θ ≠ 0 q_2\equiv\theta\ne0 q 2 ≡ θ = 0 , and the pressure-free split contains the genuine residual channel
T θ ( C ) ∘ = ( 1 − θ + 2 a θ ) Y + θ ( Y 2 ) ∘ . T_{\theta}(C)^\circ=(1-\theta+2a\theta)Y+\theta(Y^2)^\circ. T θ ( C ) ∘ = ( 1 − θ + 2 a θ ) Y + θ ( Y 2 ) ∘ . This example is included to calibrate the abstract compact-window mechanism; it is not used as an additional physical model theorem.
Proposition 6.2 (Compact-window quotient endpoint estimate). Let m ≥ 3 m\ge3 m ≥ 3 , 0 < ε < 1 0<\varepsilon<1 0 < ε < 1 , and let ( u , a , Y ) (u,a,Y) ( u , a , Y ) be a smooth solution of (8) –(10) on [ 0 , T ] [0,T] [ 0 , T ] , with C = a I + Y C=aI+Y C = a I + Y remaining in K ⋐ S + + 2 K\Subset\mathbb{S}^{2}_{++} K ⋐ S ++ 2 . Put
W ( a , η ) = τ ( a , η ) a , h = a − a ∗ . W(a,\eta)=\frac{\tau(a,\eta)}{a},\qquad h=a-a_*. W ( a , η ) = a τ ( a , η ) , h = a − a ∗ . Then there exists σ 0 = σ 0 ( K ) > 0 \sigma_0=\sigma_0(K)>0 σ 0 = σ 0 ( K ) > 0 such that, for every 0 < σ ≤ σ 0 0<\sigma\le\sigma_0 0 < σ ≤ σ 0 , the energy
E m s p e c = ∥ u ∥ H m 2 + σ ∥ h ∥ H m 2 + α 4 ∑ ∣ β ∣ ≤ m ∫ T 2 W ( a , η ) ∣ ∂ β Y ∣ 2 d x \mathcal{E}^{\mathrm{spec}}_m=\|u\|_{H^m}^2+\sigma\|h\|_{H^m}^2+\frac{\alpha}{4}\sum_{|\beta|\le m}\int_{\mathbb{T}^2}W(a,\eta)|\partial^\beta Y|^2\,dx E m spec = ∥ u ∥ H m 2 + σ ∥ h ∥ H m 2 + 4 α ∣ β ∣ ≤ m ∑ ∫ T 2 W ( a , η ) ∣ ∂ β Y ∣ 2 d x satisfies
d d t E m s p e c + c ν ∥ u ∥ H m + 1 2 + c λ − 1 ∥ Y ∥ H m 2 + c σ λ − 1 ∥ h ∥ H m 2 ≤ C K ( 1 + ∥ ∇ u ∥ B ∞ , 1 ∞ + ∥ Log C ∥ H 1 + ε 2 ) E m s p e c . \frac{d}{dt}\mathcal{E}^{\mathrm{spec}}_m+c\nu\|u\|_{H^{m+1}}^2+c\lambda^{-1}\|Y\|_{H^m}^2+c\sigma\lambda^{-1}\|h\|_{H^m}^2
\le C_K\left(1+\|\nabla u\|_{B^\infty_{\infty,1}}+\|\operatorname{Log}C\|_{H^{1+\varepsilon}}^2\right)\mathcal{E}^{\mathrm{spec}}_m. d t d E m spec + c ν ∥ u ∥ H m + 1 2 + c λ − 1 ∥ Y ∥ H m 2 + c σ λ − 1 ∥ h ∥ H m 2 ≤ C K ( 1 + ∥∇ u ∥ B ∞ , 1 ∞ + ∥ Log C ∥ H 1 + ε 2 ) E m spec .
Proof. The compact window makes E m s p e c \mathcal{E}^{\mathrm{spec}}_m E m spec equivalent to ∥ u ∥ H m + σ 2 + ∥ h ∥ H m 2 + ∥ Y ∥ H m 2 \|u\|_{H^m+\sigma}^2+\|h\|_{H^m}^2+\|Y\|_{H^m}^2 ∥ u ∥ H m + σ 2 + ∥ h ∥ H m 2 + ∥ Y ∥ H m 2 , with constants depending on K K K and σ \sigma σ . It also gives the composition bound
∥ a ∥ H 1 + ε + ∥ Y ∥ H 1 + ε ≤ C K ( 1 + ∥ Log C ∥ H 1 + ε ) . \|a\|_{H^{1+\varepsilon}}+\|Y\|_{H^{1+\varepsilon}}\le C_K(1+\|\operatorname{Log} C\|_{H^{1+\varepsilon}}). ∥ a ∥ H 1 + ε + ∥ Y ∥ H 1 + ε ≤ C K ( 1 + ∥ Log C ∥ H 1 + ε ) .
We first isolate the new coefficient commutators. For
M β τ = ∂ β ( τ ( a , η ) Y ) − τ ( a , η ) ∂ β Y , M^\tau_\beta=\partial^\beta\bigl(\tau(a,\eta)Y\bigr)-\tau(a,\eta)\partial^\beta Y, M β τ = ∂ β ( τ ( a , η ) Y ) − τ ( a , η ) ∂ β Y ,
the tame composition lemma gives, for ∣ β ∣ ≤ m |\beta|\le m ∣ β ∣ ≤ m ,
∥ M β τ ∥ L 2 ≤ C K ( 1 + ∥ Log C ∥ H 1 + ε ) ( ∥ h ∥ H m + ∥ Y ∥ H m ) . \|M^\tau_\beta\|_{L^2}\le C_K(1+\|\operatorname{Log} C\|_{H^{1+\varepsilon}})(\|h\|_{H^m}+\|Y\|_{H^m}). ∥ M β τ ∥ L 2 ≤ C K ( 1 + ∥ Log C ∥ H 1 + ε ) ( ∥ h ∥ H m + ∥ Y ∥ H m ) .
Indeed, for ∣ β ∣ ≥ 1 |\beta|\ge1 ∣ β ∣ ≥ 1 ,
∂ β τ = τ a ∂ β a + 2 τ η ( Y : ∂ β Y ) + R β τ , \partial^\beta\tau=\tau_a\partial^\beta a+2\tau_\eta(Y:\partial^\beta Y)+R^\tau_\beta, ∂ β τ = τ a ∂ β a + 2 τ η ( Y : ∂ β Y ) + R β τ ,
where
∥ R β τ ∥ L 2 ≤ C K ∥ ( a , Y ) ∥ H 1 + ε ∥ ( h , Y ) ∥ H m . \|R^\tau_\beta\|_{L^2}\le C_K\|(a,Y)\|_{H^{1+\varepsilon}}\|(h,Y)\|_{H^m}. ∥ R β τ ∥ L 2 ≤ C K ∥ ( a , Y ) ∥ H 1 + ε ∥ ( h , Y ) ∥ H m .
All remaining terms in M β τ M^\tau_\beta M β τ contain at least two positive derivatives and obey the same Moser bound. Therefore, for every δ > 0 \delta>0 δ > 0 ,
∣ ∫ M β τ : S ( ∂ β u ) d x ∣ ≤ δ ν ∥ u ∥ H m + 1 2 + C δ , K ( 1 + ∥ Log C ∥ H 1 + ε 2 ) E m s p e c . \left|\int M^\tau_\beta:S(\partial^\beta u)\,dx\right|\le\delta\nu\|u\|_{H^{m+1}}^2+C_{\delta,K}\left(1+\|\operatorname{Log} C\|_{H^{1+\varepsilon}}^2\right)\mathcal{E}^{\mathrm{spec}}_m. ∫ M β τ : S ( ∂ β u ) d x ≤ δ ν ∥ u ∥ H m + 1 2 + C δ , K ( 1 + ∥ Log C ∥ H 1 + ε 2 ) E m spec .
The identical estimate holds with τ \tau τ replaced by μ \mu μ in ∂ β ( μ Y ) − μ ∂ β Y \partial^\beta(\mu Y)-\mu\partial^\beta Y ∂ β ( μ Y ) − μ ∂ β Y , without the viscous absorption.
Apply ∂ β \partial^\beta ∂ β to the velocity equation and test by ∂ β u \partial^\beta u ∂ β u . The principal active term is
− α ∫ τ ( a , η ) ∂ β Y : S ( ∂ β u ) d x , -\alpha\int\tau(a,\eta)\partial^\beta Y:S(\partial^\beta u)\,dx, − α ∫ τ ( a , η ) ∂ β Y : S ( ∂ β u ) d x ,
and the difference between ∂ β ( τ Y ) \partial^\beta(\tau Y) ∂ β ( τ Y ) and τ ∂ β Y \tau\partial^\beta Y τ ∂ β Y is controlled by (6) . The transport and viscous terms are controlled as in Lemma 6.2.
Next apply ∂ β \partial^\beta ∂ β to the Y Y Y -equation and test by ( α / 2 ) W ∂ β Y (\alpha/2)W\partial^\beta Y ( α /2 ) W ∂ β Y . The stretching term gives
α 2 ∫ W ∂ β Y : 2 a S ( ∂ β u ) d x = α ∫ τ ( a , η ) ∂ β Y : S ( ∂ β u ) d x , \frac{\alpha}{2}\int W\partial^\beta Y:2aS(\partial^\beta u)\,dx=\alpha\int\tau(a,\eta)\partial^\beta Y:S(\partial^\beta u)\,dx, 2 α ∫ W ∂ β Y : 2 a S ( ∂ β u ) d x = α ∫ τ ( a , η ) ∂ β Y : S ( ∂ β u ) d x ,
which cancels (6) . The main relaxation term gives
α 2 λ ∫ W μ ∣ ∂ β Y ∣ 2 d x ≥ c K λ − 1 ∥ ∂ β Y ∥ L 2 2 . \frac{\alpha}{2\lambda}\int W\mu|\partial^\beta Y|^2\,dx\ge c_K\lambda^{-1}\|\partial^\beta Y\|_{L^2}^2. 2 λ α ∫ W μ ∣ ∂ β Y ∣ 2 d x ≥ c K λ − 1 ∥ ∂ β Y ∥ L 2 2 .
The coefficient commutator ∂ β ( μ Y ) − μ ∂ β Y \partial^\beta(\mu Y)-\mu\partial^\beta Y ∂ β ( μ Y ) − μ ∂ β Y is controlled by the tame bound above. Since
D t W = W a D t a + W η D t η D_tW=W_aD_ta+W_\eta D_t\eta D t W = W a D t a + W η D t η
and
D t η = 4 a Y : S ( u ) + 2 Y : [ ∇ u Y + Y ( ∇ u ) T ] − 2 λ − 1 μ ∣ Y ∣ 2 , D_t\eta=4aY:S(u)+2Y:[\nabla u\,Y+Y(\nabla u)^\mathsf{T}]-2\lambda^{-1}\mu|Y|^2, D t η = 4 aY : S ( u ) + 2 Y : [ ∇ u Y + Y ( ∇ u ) T ] − 2 λ − 1 μ ∣ Y ∣ 2 ,
the compact window gives
∥ D t W ∥ L ∞ ≤ C K ( 1 + ∥ ∇ u ∥ L ∞ ) ≤ C K ( 1 + ∥ ∇ u ∥ B ∞ , 1 0 ) . \|D_tW\|_{L^\infty}\le C_K(1+\|\nabla u\|_{L^\infty})\le C_K(1+\|\nabla u\|_{B^0_{\infty,1}}). ∥ D t W ∥ L ∞ ≤ C K ( 1 + ∥∇ u ∥ L ∞ ) ≤ C K ( 1 + ∥∇ u ∥ B ∞ , 1 0 ) .
The weighted transport commutator is then estimated by Lemma 6.2, and the remaining stretching commutators have the same form as in the Oldroyd–B physical estimate.
It remains to treat the scalar equation. For ∣ β ∣ ≥ 1 |\beta| \ge1 ∣ β ∣ ≥ 1 ,
∂ β G ( a , η ) = G a ∂ β h + 2 G η ( Y : ∂ β Y ) + R β G , \partial^\beta G(a,\eta) = G_a\partial^\beta h + 2G_\eta(Y:\partial^\beta Y) + R^G_\beta, ∂ β G ( a , η ) = G a ∂ β h + 2 G η ( Y : ∂ β Y ) + R β G ,
with
∥ R β G ∥ L 2 ≤ C K ∥ ( a , Y ) ∥ H 1 + ε ∥ ( h , Y ) ∥ H m . \|R^G_\beta\|_{L^2} \le C_K\|(a,Y)\|_{H^{1+\varepsilon}}\|(h,Y)\|_{H^m}. ∥ R β G ∥ L 2 ≤ C K ∥ ( a , Y ) ∥ H 1 + ε ∥ ( h , Y ) ∥ H m .
Testing the differentiated scalar equation by σ ∂ β h \sigma\partial^\beta h σ ∂ β h , the top relaxation quadratic form is
σ G a ∣ ∂ β h ∣ 2 + 2 σ G η ∂ β h ( Y : ∂ β Y ) + α 2 W μ ∣ ∂ β Y ∣ 2 . \sigma G_a|\partial^\beta h|^2 + 2\sigma G_\eta\partial^\beta h\,(Y:\partial^\beta Y) + \frac{\alpha}{2}W\mu|\partial^\beta Y|^2. σ G a ∣ ∂ β h ∣ 2 + 2 σ G η ∂ β h ( Y : ∂ β Y ) + 2 α W μ ∣ ∂ β Y ∣ 2 .
Since G a ≥ κ G_a \ge\kappa G a ≥ κ , W μ ≥ c K > 0 W\mu\ge c_K > 0 W μ ≥ c K > 0 , and Y Y Y is bounded on K K K , Young’s inequality shows that this form is bounded below by
c σ ∣ ∂ β h ∣ 2 + c ∣ ∂ β Y ∣ 2 c\sigma|\partial^\beta h|^2 + c|\partial^\beta Y|^2 c σ ∣ ∂ β h ∣ 2 + c ∣ ∂ β Y ∣ 2
provided 0 < σ ≤ σ 0 ( K ) 0 < \sigma\le\sigma_0(K) 0 < σ ≤ σ 0 ( K ) . At order zero,
G ( a , η ) h = G ( a , 0 ) h + { G ( a , η ) − G ( a , 0 ) } h . G(a,\eta)h = G(a,0)h + \{G(a,\eta)-G(a,0)\}h. G ( a , η ) h = G ( a , 0 ) h + { G ( a , η ) − G ( a , 0 )} h .
The first term controls h 2 h^2 h 2 by the same monotonicity in a a a , while the second is bounded by C K ∣ Y ∣ 2 ∣ h ∣ C_K|Y|^2|h| C K ∣ Y ∣ 2 ∣ h ∣ and is absorbed by the Y Y Y -relaxation and the small scalar weight. The source S ( u ) : Y S(u):Y S ( u ) : Y is estimated by
C ∥ Y ∥ H 1 + ε ∥ u ∥ H m + 1 + C ∥ ∇ u ∥ B 0 , 1 ∞ ∥ Y ∥ H m , C\|Y\|_{H^{1+\varepsilon}}\|u\|_{H^{m+1}} + C\|\nabla u\|_{B^\infty_{0,1}}\|Y\|_{H^m}, C ∥ Y ∥ H 1 + ε ∥ u ∥ H m + 1 + C ∥∇ u ∥ B 0 , 1 ∞ ∥ Y ∥ H m ,
followed by Young’s inequality.
Summing the velocity, trace-free, and scalar estimates over ∣ β ∣ ≤ m |\beta| \le m ∣ β ∣ ≤ m , choosing δ > 0 \delta> 0 δ > 0 small and then σ ≤ σ 0 ( K ) \sigma\le\sigma_0(K) σ ≤ σ 0 ( K ) , gives (6.2).
The scalar-Peterlin theorem is the trace-only specialization of Proposition 6.2. We keep the explicit statement because it is the form used for Oldroyd–B and FENE-P.
Definition 6.2 (Continuation-admissible scalar-Peterlin closure). Let I ⋐ ( 0 , ∞ ) I \Subset(0,\infty) I ⋐ ( 0 , ∞ ) be a compact trace interval. A triple ( ψ , ϕ , χ ) (\psi,\phi,\chi) ( ψ , ϕ , χ ) is continuation-admissible on I I I if the coefficients are smooth on I I I and, with
Ψ ( a ) = ψ ( 2 a ) , Φ ( a ) = ϕ ( 2 a ) , G ( a ) = a ϕ ( 2 a ) − χ ( 2 a ) , \Psi(a)=\psi(2a), \qquad\Phi(a)=\phi(2a), \qquad G(a)=a\phi(2a)-\chi(2a), Ψ ( a ) = ψ ( 2 a ) , Φ ( a ) = ϕ ( 2 a ) , G ( a ) = a ϕ ( 2 a ) − χ ( 2 a ) , there are constants a 0 , Ψ 0 , Φ 0 , κ > 0 a_0,\Psi_0,\Phi_0,\kappa>0 a 0 , Ψ 0 , Φ 0 , κ > 0 and a ∗ ∈ { a : 2 a ∈ I } a_* \in\{a:2a\in I\} a ∗ ∈ { a : 2 a ∈ I } such that
a ≥ a 0 , Ψ ( a ) ≥ Ψ 0 , Φ ( a ) ≥ Φ 0 , G ( a ∗ ) = 0 , G ′ ( a ) ≥ κ a\ge a_0,\qquad\Psi(a)\ge\Psi_0,\qquad\Phi(a)\ge\Phi_0,\qquad G(a_*)=0,\qquad G'(a)\ge\kappa a ≥ a 0 , Ψ ( a ) ≥ Ψ 0 , Φ ( a ) ≥ Φ 0 , G ( a ∗ ) = 0 , G ′ ( a ) ≥ κ for 2 a ∈ I 2a\in I 2 a ∈ I . These are the hypotheses used in the endpoint continuation estimate.
This scalar-Peterlin definition is again a continuation condition. The thermodynamic compatibility in Remark 6.1 reduces here to a familiar trace potential condition: it is sufficient that there exist F F F with F ′ = ψ F'=\psi F ′ = ψ such that
( ψ ( tr C ) I − C − 1 ) : ( χ ( tr C ) I − ϕ ( tr C ) C ) ≤ 0 (\psi(\operatorname{tr} C)I-C^{-1}):(\chi(\operatorname{tr} C)I-\phi(\operatorname{tr} C)C)\le0 ( ψ ( tr C ) I − C − 1 ) : ( χ ( tr C ) I − ϕ ( tr C ) C ) ≤ 0
for C C C in the compact conformation window. Then ∫ T 2 ( F ( tr C ) − log det C ) d x \int_{\mathbb{T}^2}(F(\operatorname{tr} C)-\log\det C)\,dx ∫ T 2 ( F ( tr C ) − log det C ) d x is dissipated by the relaxation part. Oldroyd–B and FENE-P satisfy this compatibility, but Proposition 6.3 only uses the continuation admissibility stated above.
Proposition 6.3 (Scalar-Peterlin compact-window endpoint estimate). Let m ≥ 3 m\ge3 m ≥ 3 , 0 < ε < 1 0<\varepsilon<1 0 < ε < 1 , and consider a smooth solution of
∂ t u + u ⋅ ∇ u − ν Δ u + ∇ p = α div ( ψ ( r ) C − I ) , r = tr C , \partial_tu+u\cdot\nabla u-\nu\Delta u+\nabla p=\alpha\operatorname{div}(\psi(r)C-I), \qquad r=\operatorname{tr}C, ∂ t u + u ⋅ ∇ u − ν Δ u + ∇ p = α div ( ψ ( r ) C − I ) , r = tr C , D t C = ∇ u C + C ( ∇ u ) T + λ − 1 χ ( r ) I − λ − 1 ϕ ( r ) C , div u = 0. D_tC=\nabla u\,C+C(\nabla u)^{T}+\lambda^{-1}\chi(r)I-\lambda^{-1}\phi(r)C,\qquad\operatorname{div}u=0. D t C = ∇ u C + C ( ∇ u ) T + λ − 1 χ ( r ) I − λ − 1 ϕ ( r ) C , div u = 0.
Write C = a I + Y C=aI+Y C = a I + Y , a = 1 2 tr C a=\frac{1}{2}\operatorname{tr} C a = 2 1 tr C , Y = C ∘ Y=C^\circ Y = C ∘ , and set
Ψ ( a ) = ψ ( 2 a ) , Φ ( a ) = ϕ ( 2 a ) , G ( a ) = a ϕ ( 2 a ) − χ ( 2 a ) , W ( a ) = Ψ ( a ) a . \Psi(a)=\psi(2a),\qquad\Phi(a)=\phi(2a),\qquad G(a)=a\phi(2a)-\chi(2a),\qquad W(a)=\frac{\Psi(a)}{a}. Ψ ( a ) = ψ ( 2 a ) , Φ ( a ) = ϕ ( 2 a ) , G ( a ) = a ϕ ( 2 a ) − χ ( 2 a ) , W ( a ) = a Ψ ( a ) .
Assume that C ( t , x ) C(t,x) C ( t , x ) remains in a compact conformation window K K K , that
a ≥ a 0 > 0 , Ψ ( a ) ≥ Ψ 0 > 0 , Φ ( a ) ≥ Φ 0 > 0 a\ge a_0>0,\qquad\Psi(a)\ge\Psi_0>0,\qquad\Phi(a)\ge\Phi_0>0 a ≥ a 0 > 0 , Ψ ( a ) ≥ Ψ 0 > 0 , Φ ( a ) ≥ Φ 0 > 0
on this window, and that for some a ∗ a_* a ∗ and κ > 0 \kappa>0 κ > 0 ,
G ( a ∗ ) = 0 , G ′ ( a ) ≥ κ G(a_*)=0,\qquad G'(a)\ge\kappa G ( a ∗ ) = 0 , G ′ ( a ) ≥ κ
on the scalar range of K K K . Define
E m ψ = ∥ u ∥ H m 2 + ∥ a − a ∗ ∥ H m 2 + α 4 ∑ ∣ β ∣ ≤ m ∫ T 2 W ( a ) ∣ ∂ β Y ∣ 2 d x . \mathcal{E}^{\psi}_m=\|u\|_{H^m}^2+\|a-a_*\|_{H^m}^2+\frac{\alpha}{4}\sum_{|\beta|\le m}\int_{\mathbb{T}^2}W(a)|\partial^\beta Y|^2\,dx. E m ψ = ∥ u ∥ H m 2 + ∥ a − a ∗ ∥ H m 2 + 4 α ∣ β ∣ ≤ m ∑ ∫ T 2 W ( a ) ∣ ∂ β Y ∣ 2 d x .
Then
d d t E m ψ + c ν ∥ u ∥ H m + 1 2 + c λ − 1 ∥ Y ∥ H m 2 + c λ − 1 ∥ a − a ∗ ∥ H m 2 ≤ C K ( 1 + ∥ ∇ u ∥ B 0 , 1 ∞ + ∥ Log C ∥ H 1 + ε 2 ) E m ψ . (13) \frac{d}{dt}\mathcal{E}^{\psi}_m+c\nu\|u\|_{H^{m+1}}^2+c\lambda^{-1}\|Y\|_{H^m}^2+c\lambda^{-1}\|a-a_*\|_{H^m}^2
\le C_K\left(1+\|\nabla u\|_{B^\infty_{0,1}}+\|\operatorname{Log} C\|_{H^{1+\varepsilon}}^2\right)\mathcal{E}^{\psi}_m. \tag*{(13)} d t d E m ψ + c ν ∥ u ∥ H m + 1 2 + c λ − 1 ∥ Y ∥ H m 2 + c λ − 1 ∥ a − a ∗ ∥ H m 2 ≤ C K ( 1 + ∥∇ u ∥ B 0 , 1 ∞ + ∥ Log C ∥ H 1 + ε 2 ) E m ψ . (13)
Proof. The pressure-free splitting is
ψ ( r ) C − I = ( Ψ a − 1 ) I + Ψ Y . \psi(r)C-I=(\Psi a-1)I+\Psi Y. ψ ( r ) C − I = ( Ψ a − 1 ) I + Ψ Y .
Thus the momentum equation becomes
∂ t u + u ⋅ ∇ u − ν Δ u + ∇ π = α div ( Ψ Y ) . \partial_tu+u\cdot\nabla u-\nu\Delta u+\nabla\pi=\alpha\operatorname{div}(\Psi Y). ∂ t u + u ⋅ ∇ u − ν Δ u + ∇ π = α div ( Ψ Y ) .
The scalar and trace-free equations are
D t ( a − a ∗ ) + λ − 1 G ( a ) = S ( u ) : Y , D_t(a-a_*)+\lambda^{-1}G(a)=S(u):Y, D t ( a − a ∗ ) + λ − 1 G ( a ) = S ( u ) : Y ,
D t Y + λ − 1 Φ Y = 2 a S ( u ) + [ ∇ u Y + Y ( ∇ u ) T ] ∘ . D_tY+\lambda^{-1}\Phi Y=2aS(u)+[\nabla u\,Y+Y(\nabla u)^{\mathsf T}]^\circ. D t Y + λ − 1 Φ Y = 2 a S ( u ) + [ ∇ u Y + Y ( ∇ u ) T ] ∘ .
The compact window makes E m ψ \mathcal{E}^{\psi}_m E m ψ equivalent to ∥ u ∥ H m 2 + ∥ a − a ∗ ∥ H m 2 + ∥ Y ∥ H m 2 \|u\|_{H^m}^2+\|a-a_*\|_{H^m}^2+\|Y\|_{H^m}^2 ∥ u ∥ H m 2 + ∥ a − a ∗ ∥ H m 2 + ∥ Y ∥ H m 2 .
Apply ∂ β \partial^\beta ∂ β , ∣ β ∣ ≤ m |\beta|\le m ∣ β ∣ ≤ m , to the velocity equation and test by ∂ β u \partial^\beta u ∂ β u . The principal elastic term is
− α ∫ Ψ ∂ β Y : S ( ∂ β u ) d x . -\alpha\int\Psi\,\partial^\beta Y:S(\partial^\beta u)\,dx. − α ∫ Ψ ∂ β Y : S ( ∂ β u ) d x .
The difference between ∂ β ( Ψ Y ) \partial^\beta(\Psi Y) ∂ β ( Ψ Y ) and Ψ ∂ β Y \Psi\partial^\beta Y Ψ ∂ β Y is estimated by Moser calculus:
∥ ∂ β ( Ψ Y ) − Ψ ∂ β Y ∥ L 2 ≤ C K ( 1 + ∥ Log C ∥ H 1 + ε ) ( ∥ a − a ∗ ∥ H m + ∥ Y ∥ H m ) . \left\|\partial^\beta(\Psi Y)-\Psi\partial^\beta Y\right\|_{L^2}
\le C_K\left(1+\|\operatorname{Log} C\|_{H^{1+\varepsilon}}\right)\left(\|a-a_*\|_{H^m}+\|Y\|_{H^m}\right). ∂ β ( Ψ Y ) − Ψ ∂ β Y L 2 ≤ C K ( 1 + ∥ Log C ∥ H 1 + ε ) ( ∥ a − a ∗ ∥ H m + ∥ Y ∥ H m ) .
There is no additive constant: the commutator vanishes for β = 0 \beta=0 β = 0 , and for ∣ β ∣ > 0 |\beta|>0 ∣ β ∣ > 0 every term contains a positive derivative of a − a ∗ a-a_* a − a ∗ or Y Y Y . Young’s inequality bounds the commutator contribution by
δ ν ∥ u ∥ H m + 1 2 + C δ , K ( 1 + ∥ Log C ∥ H 1 + ε 2 ) E m ψ . \delta\nu\|u\|_{H^{m+1}}^2+C_{\delta,K}\left(1+\|\operatorname{Log} C\|_{H^{1+\varepsilon}}^2\right)\mathcal{E}^{\psi}_m. δ ν ∥ u ∥ H m + 1 2 + C δ , K ( 1 + ∥ Log C ∥ H 1 + ε 2 ) E m ψ .
Next differentiate the Y Y Y -equation and test by ( α / 2 ) W ∂ β Y (\alpha/2)W\partial^\beta Y ( α /2 ) W ∂ β Y . The principal stretching term gives
α 2 ∫ W ∂ β Y : 2 a S ( ∂ β u ) d x = α ∫ Ψ ∂ β Y : S ( ∂ β u ) d x . \frac{\alpha}{2}\int W\partial^\beta Y:2aS(\partial^\beta u)\,dx
=\alpha\int\Psi\,\partial^\beta Y:S(\partial^\beta u)\,dx. 2 α ∫ W ∂ β Y : 2 a S ( ∂ β u ) d x = α ∫ Ψ ∂ β Y : S ( ∂ β u ) d x .
which cancels the principal velocity contribution. The relaxation term gives
α 2 λ ∫ W Φ ∣ ∂ β Y ∣ 2 d x ≥ c λ − 1 ∥ ∂ β Y ∥ L 2 2 . \frac{\alpha}{2\lambda}\int W\Phi|\partial^\beta Y|^2\,dx \ge c\lambda^{-1}\|\partial^\beta Y\|_{L^2}^2. 2 λ α ∫ W Φ∣ ∂ β Y ∣ 2 d x ≥ c λ − 1 ∥ ∂ β Y ∥ L 2 2 .
The material derivative of the weight satisfies
D t W = W ′ ( a ) { S ( u ) : Y − λ − 1 G ( a ) } , D_tW=W'(a)\{S(u):Y-\lambda^{-1}G(a)\}, D t W = W ′ ( a ) { S ( u ) : Y − λ − 1 G ( a )} ,
and hence
∥ D t W ∥ L ∞ ≤ C K ( 1 + ∥ ∇ u ∥ L ∞ ) ≤ C K ( 1 + ∥ ∇ u ∥ B ∞ , 1 0 ) . \|D_tW\|_{L^\infty}\le C_K(1+\|\nabla u\|_{L^\infty})\le C_K(1+\|\nabla u\|_{B^0_{\infty,1}}). ∥ D t W ∥ L ∞ ≤ C K ( 1 + ∥∇ u ∥ L ∞ ) ≤ C K ( 1 + ∥∇ u ∥ B ∞ , 1 0 ) .
The weighted transport commutator is controlled by the endpoint Coifman–Meyer estimate
∣ ∑ ∣ β ∣ ≤ m ∫ W ∂ β Y : [ ∂ β , u ⋅ ∇ ] Y d x ∣ ≤ C K ∥ ∇ u ∥ B ∞ , 1 0 ∥ Y ∥ H m 2 + C K ∥ Y ∥ H 1 + ε ∥ u ∥ H m + 1 ∥ Y ∥ H m . \left|\sum_{|\beta|\le m}\int W\partial^\beta Y:[\partial^\beta,u\cdot\nabla]Y\,dx\right|
\le C_K\|\nabla u\|_{B^0_{\infty,1}}\|Y\|_{H^m}^2+C_K\|Y\|_{H^{1+\varepsilon}}\|u\|_{H^{m+1}}\|Y\|_{H^m}. ∣ β ∣ ≤ m ∑ ∫ W ∂ β Y : [ ∂ β , u ⋅ ∇ ] Y d x ≤ C K ∥∇ u ∥ B ∞ , 1 0 ∥ Y ∥ H m 2 + C K ∥ Y ∥ H 1 + ε ∥ u ∥ H m + 1 ∥ Y ∥ H m .
This is precisely the weighted form of Lemma 6.2: the weight is placed in L ∞ L^\infty L ∞ , while the evolution of the weight is handled by D t W D_tW D t W . The remaining stretching and relaxation commutators have the same tame form. Using the compact-window composition bound
∥ a − a ∗ ∥ H 1 + ε + ∥ Y ∥ H 1 + ε ≤ C K ( 1 + ∥ Log C ∥ H 1 + ε ) , \|a-a_*\|_{H^{1+\varepsilon}}+\|Y\|_{H^{1+\varepsilon}}\le C_K(1+\|\operatorname{Log}C\|_{H^{1+\varepsilon}}), ∥ a − a ∗ ∥ H 1 + ε + ∥ Y ∥ H 1 + ε ≤ C K ( 1 + ∥ Log C ∥ H 1 + ε ) ,
their total contribution is bounded by
δ ν ∥ u ∥ H m + 1 2 + C δ , K ( 1 + ∥ ∇ u ∥ B ∞ , 1 0 + ∥ Log C ∥ H 1 + ε ) E m ψ . \delta\nu\|u\|_{H^{m+1}}^2+C_{\delta,K}(1+\|\nabla u\|_{B^0_{\infty,1}}+\|\operatorname{Log}C\|_{H^{1+\varepsilon}})\mathcal{E}_m^\psi. δ ν ∥ u ∥ H m + 1 2 + C δ , K ( 1 + ∥∇ u ∥ B ∞ , 1 0 + ∥ Log C ∥ H 1 + ε ) E m ψ .
It remains to estimate the scalar equation. Put h = a − a ∗ h=a-a_* h = a − a ∗ and F ( h ) = G ( a ∗ + h ) F(h)=G(a_*+h) F ( h ) = G ( a ∗ + h ) . Then F ( 0 ) = 0 F(0)=0 F ( 0 ) = 0 and F ′ ( h ) ≥ κ F'(h)\ge\kappa F ′ ( h ) ≥ κ . For ∣ β ∣ ≥ 1 |\beta|\ge1 ∣ β ∣ ≥ 1 ,
∂ β F ( h ) = F ′ ( h ) ∂ β h + R β F , \partial^\beta F(h)=F'(h)\partial^\beta h+R^F_\beta, ∂ β F ( h ) = F ′ ( h ) ∂ β h + R β F ,
where R β F = 0 R^F_\beta=0 R β F = 0 if ∣ β ∣ = 1 |\beta|=1 ∣ β ∣ = 1 , and if ∣ β ∣ ≥ 2 |\beta|\ge2 ∣ β ∣ ≥ 2 ,
∥ R β F ∥ L 2 ≤ C K ∥ h ∥ H 1 + ε ∥ h ∥ H m . \|R^F_\beta\|_{L^2}\le C_K\|h\|_{H^{1+\varepsilon}}\|h\|_{H^m}. ∥ R β F ∥ L 2 ≤ C K ∥ h ∥ H 1 + ε ∥ h ∥ H m .
This is the standard scalar Moser remainder: every monomial in R β F R^F_\beta R β F contains at least two positive derivatives of h h h . Therefore the principal part yields
∫ F ′ ( h ) ∣ ∂ β h ∣ 2 d x ≥ κ ∥ ∂ β h ∥ L 2 2 , \int F'(h)|\partial^\beta h|^2\,dx\ge\kappa\|\partial^\beta h\|_{L^2}^2, ∫ F ′ ( h ) ∣ ∂ β h ∣ 2 d x ≥ κ ∥ ∂ β h ∥ L 2 2 ,
while the remainder is bounded by
C K ( 1 + ∥ Log C ∥ H 1 + ε ) E m ψ . C_K(1+\|\operatorname{Log}C\|_{H^{1+\varepsilon}})\mathcal{E}_m^\psi. C K ( 1 + ∥ Log C ∥ H 1 + ε ) E m ψ .
The zero-order contribution satisfies
∫ F ( h ) h d x = ∫ h 2 ∫ 0 1 F ′ ( θ h ) d θ d x ≥ κ ∥ h ∥ L 2 2 . \int F(h)h\,dx=\int h^2\int_0^1F'(\theta h)\,d\theta\,dx\ge\kappa\|h\|_{L^2}^2. ∫ F ( h ) h d x = ∫ h 2 ∫ 0 1 F ′ ( θ h ) d θ d x ≥ κ ∥ h ∥ L 2 2 .
Finally, the source ∂ β ( S ( u ) : Y ) \partial^\beta(S(u):Y) ∂ β ( S ( u ) : Y ) is bounded by
C ∥ Y ∥ H 1 + ε ∥ u ∥ H m + 1 + C ∥ ∇ u ∥ B ∞ , 1 0 ∥ Y ∥ H m , C\|Y\|_{H^{1+\varepsilon}}\|u\|_{H^{m+1}}+C\|\nabla u\|_{B^0_{\infty,1}}\|Y\|_{H^m}, C ∥ Y ∥ H 1 + ε ∥ u ∥ H m + 1 + C ∥∇ u ∥ B ∞ , 1 0 ∥ Y ∥ H m ,
and is treated by Young’s inequality.
Summing the velocity, trace-free, and scalar estimates, choosing δ > 0 \delta>0 δ > 0 small, and absorbing the viscous terms gives (13) .
We now specialize this principle to Oldroyd–B. In that case ψ = ϕ = χ = 1 \psi=\phi=\chi=1 ψ = ϕ = χ = 1 , so a ∗ = 1 a_\ast=1 a ∗ = 1 , G ( a ) = a − 1 G(a)=a-1 G ( a ) = a − 1 , and W = a − 1 W=a^{-1} W = a − 1 . The key refinement of the Oldroyd–B estimate is therefore that the stress force should not be estimated directly in logarithmic coordinates. Instead write
A = a I + Y , a = 1 2 tr A , Y = A ∘ = A − 1 2 ( tr A ) I . A=aI+Y,\qquad a=\frac{1}{2}\operatorname{tr}A,\qquad Y=A^\circ=A-\frac{1}{2}(\operatorname{tr}A)I. A = a I + Y , a = 2 1 tr A , Y = A ∘ = A − 2 1 ( tr A ) I .
Then div ( a I ) = ∇ a \operatorname{div}(aI)=\nabla a div ( a I ) = ∇ a is absorbed into the pressure and (1) –(2) become
∂ t u + u ⋅ ∇ u − ν Δ u + ∇ π = α div Y , (14) \partial_tu+u\cdot\nabla u-\nu\Delta u+\nabla\pi=\alpha\operatorname{div}Y, \tag*{(14)} ∂ t u + u ⋅ ∇ u − ν Δ u + ∇ π = α div Y , (14)
( ∂ t + u ⋅ ∇ ) a + λ − 1 ( a − 1 ) = S ( u ) : Y , (15) (\partial_t+u\cdot\nabla)a+\lambda^{-1}(a-1)=S(u):Y, \tag*{(15)} ( ∂ t + u ⋅ ∇ ) a + λ − 1 ( a − 1 ) = S ( u ) : Y , (15)
( ∂ t + u ⋅ ∇ ) Y + λ − 1 Y = 2 a S ( u ) + [ ∇ u Y + Y ( ∇ u ) T ] . (16) (\partial_t+u\cdot\nabla)Y+\lambda^{-1}Y=2aS(u)+[\nabla uY+Y(\nabla u)^{T}]. \tag*{(16)} ( ∂ t + u ⋅ ∇ ) Y + λ − 1 Y = 2 a S ( u ) + [ ∇ u Y + Y ( ∇ u ) T ] . (16)
Here S ( u ) = ( ∇ u + ∇ u T ) / 2 S(u)=(\nabla u+\nabla u^{T})/2 S ( u ) = ( ∇ u + ∇ u T ) /2 . The principal coupling in (14) and (16) cancels in the energy below. At derivative order β \beta β , the top-order velocity contribution is
− α ∫ ∂ β Y : S ( ∂ β u ) d x . -\alpha\int\partial^\beta Y:S(\partial^\beta u)\,dx. − α ∫ ∂ β Y : S ( ∂ β u ) d x .
while the top-order anisotropic contribution, after testing the Y Y Y -equation by ( α / 2 ) a − 1 ∂ β Y (\alpha/2)a^{-1}\partial^\beta Y ( α /2 ) a − 1 ∂ β Y , is
α ∫ ∂ β Y : S ( ∂ β u ) d x . \alpha\int\partial^\beta Y:S(\partial^\beta u)\,dx. α ∫ ∂ β Y : S ( ∂ β u ) d x .
Thus the leading stress–velocity interaction is removed by an exact identity, not by a smallness assumption. This pressure-free cancellation is the mechanism which gives the L t 2 H x 1 + ε L_t^2H_x^{1+\varepsilon} L t 2 H x 1 + ε criterion.
Lemma 6.1 (Propagation of Oldroyd–B spectral windows). Let ( u , A ) (u,A) ( u , A ) be a smooth positive-cone solution of (1) –(2) on [ 0 , T ) [0,T) [ 0 , T ) , and assume that
∫ 0 T ∥ ∇ u ( t ) ∥ B 0 ∞ , 1 d t < ∞ . \int_0^T\|\nabla u(t)\|_{B^\ 0_{\infty,1}}\,dt<\infty. ∫ 0 T ∥∇ u ( t ) ∥ B 0 ∞ , 1 d t < ∞. If the initial spectrum of A 0 A_0 A 0 is contained in [ m 0 , M 0 ] [m_0,M_0] [ m 0 , M 0 ] , with 0 < m 0 ≤ M 0 < ∞ 0<m_0\le M_0<\infty 0 < m 0 ≤ M 0 < ∞ , then there are constants m T , M T m_T,M_T m T , M T , depending only on m 0 , M 0 , T , λ m_0,M_0,T,\lambda m 0 , M 0 , T , λ and ∫ 0 T ∥ ∇ u ( t ) ∥ B 0 ∞ , 1 d t \int_0^T\|\nabla u(t)\|_{B^\ 0_{\infty,1}}\,dt ∫ 0 T ∥∇ u ( t ) ∥ B 0 ∞ , 1 d t , such that
0 < m T I ≤ A ( t , x ) ≤ M T I for all ( t , x ) ∈ [ 0 , T ) × T 2 . 0<m_T I\le A(t,x)\le M_T I\qquad\text{for all }(t,x)\in[0,T)\times\mathbb{T}^2. 0 < m T I ≤ A ( t , x ) ≤ M T I for all ( t , x ) ∈ [ 0 , T ) × T 2 . In particular, the compact spectral window required by the high-order physical energy estimate is propagated by the endpoint velocity clock.
Proof. Let X ( t ; x 0 ) X(t;x_0) X ( t ; x 0 ) be the Lagrangian flow. Along this trajectory write A ( t ) = A ( t , X ( t ; x 0 ) ) \mathcal{A}(t)=A(t,X(t;x_0)) A ( t ) = A ( t , X ( t ; x 0 )) . Then
d d t A = ∇ u A + A ( ∇ u ) T − λ − 1 ( A − I ) . \frac{d}{dt}\mathcal{A}=\nabla u\,\mathcal{A}+\mathcal{A}(\nabla u)^T-\lambda^{-1}(\mathcal{A}-I). d t d A = ∇ u A + A ( ∇ u ) T − λ − 1 ( A − I ) .
Let M ( t ) M(t) M ( t ) and m ( t ) m(t) m ( t ) denote the largest and smallest eigenvalues of A ( t ) \mathcal{A}(t) A ( t ) . At times of differentiability of these Lipschitz functions, using a unit eigenvector for the corresponding eigenvalue gives
M ˙ ( t ) ≤ 2 ∥ ∇ u ( t ) ∥ L ∞ M ( t ) + λ − 1 , \dot M(t)\le2\|\nabla u(t)\|_{L^\infty}M(t)+\lambda^{-1}, M ˙ ( t ) ≤ 2∥∇ u ( t ) ∥ L ∞ M ( t ) + λ − 1 ,
and
m ˙ ( t ) ≥ − ( 2 ∥ ∇ u ( t ) ∥ L ∞ + λ − 1 ) m ( t ) . \dot m(t)\ge-\left(2\|\nabla u(t)\|_{L^\infty}+\lambda^{-1}\right)m(t). m ˙ ( t ) ≥ − ( 2∥∇ u ( t ) ∥ L ∞ + λ − 1 ) m ( t ) .
The same inequalities hold for the upper and lower Dini derivatives. Since the endpoint embedding gives
∥ ∇ u ∥ L ∞ ≤ C ∥ ∇ u ∥ B 0 ∞ , 1 , \|\nabla u\|_{L^\infty}\le C\|\nabla u\|_{B^\ 0_{\infty,1}}, ∥∇ u ∥ L ∞ ≤ C ∥∇ u ∥ B 0 ∞ , 1 ,
Gronwall’s inequality gives
M ( t ) ≤ exp ( 2 C ∫ 0 t ∥ ∇ u ( τ ) ∥ B ∞ , 1 0 d τ ) ( M 0 + λ − 1 t ) , M(t) \le\exp\left(2C\int_0^t \|\nabla u(\tau)\|_{B^0_{\infty,1}}\,d\tau\right)\left(M_0+\lambda^{-1}t\right), M ( t ) ≤ exp ( 2 C ∫ 0 t ∥∇ u ( τ ) ∥ B ∞ , 1 0 d τ ) ( M 0 + λ − 1 t ) ,
and
m ( t ) ≥ m 0 exp ( − λ − 1 t − 2 C ∫ 0 t ∥ ∇ u ( τ ) ∥ B ∞ , 1 0 d τ ) . m(t) \ge m_0\exp\left(-\lambda^{-1}t-2C\int_0^t \|\nabla u(\tau)\|_{B^0_{\infty,1}}\,d\tau\right). m ( t ) ≥ m 0 exp ( − λ − 1 t − 2 C ∫ 0 t ∥∇ u ( τ ) ∥ B ∞ , 1 0 d τ ) .
Taking the infimum over trajectories for the lower bound and the supremum for the upper bound proves the claim.
Lemma 6.2 (Endpoint transport and physical commutator bounds). Let m ≥ 3 m \ge3 m ≥ 3 and 0 < ε < 1 0 < \varepsilon< 1 0 < ε < 1 . For smooth fields on T 2 \mathbb{T}^2 T 2 and div u = 0 , \operatorname{div} u = 0, div u = 0 ,
∣ ∑ ∣ β ∣ ≤ m ∫ ∂ β f : [ ∂ β , u ⋅ ∇ ] f d x ∣ ≤ C ∥ ∇ u ∥ B ∞ , 1 0 ∥ f ∥ H m 2 . \left|\sum_{|\beta|\le m}\int\partial^\beta f : [\partial^\beta,u\cdot\nabla]f\,dx\right|
\le C\|\nabla u\|_{B^0_{\infty,1}}\|f\|_{H^m}^2. ∣ β ∣ ≤ m ∑ ∫ ∂ β f : [ ∂ β , u ⋅ ∇ ] f d x ≤ C ∥∇ u ∥ B ∞ , 1 0 ∥ f ∥ H m 2 . Moreover, if w ∈ L ∞ w\in L^\infty w ∈ L ∞ , then
∣ ∑ ∣ β ∣ ≤ m ∫ w ∂ β f : [ ∂ β , u ⋅ ∇ ] f d x ∣ ≤ C ∥ w ∥ L ∞ ∥ ∇ u ∥ B ∞ , 1 0 ∥ f ∥ H m 2 + C ∥ w ∥ L ∞ ∥ f ∥ H 1 + ε ∥ u ∥ H m + 1 ∥ f ∥ H m . \left|\sum_{|\beta|\le m}\int w\partial^\beta f : [\partial^\beta,u\cdot\nabla]f\,dx\right|
\le C\|w\|_{L^\infty}\|\nabla u\|_{B^0_{\infty,1}}\|f\|_{H^m}^2
+C\|w\|_{L^\infty}\|f\|_{H^{1+\varepsilon}}\|u\|_{H^{m+1}}\|f\|_{H^m}. ∣ β ∣ ≤ m ∑ ∫ w ∂ β f : [ ∂ β , u ⋅ ∇ ] f d x ≤ C ∥ w ∥ L ∞ ∥∇ u ∥ B ∞ , 1 0 ∥ f ∥ H m 2 + C ∥ w ∥ L ∞ ∥ f ∥ H 1 + ε ∥ u ∥ H m + 1 ∥ f ∥ H m . Moreover, for ∣ β ∣ ≤ m , |\beta|\le m, ∣ β ∣ ≤ m ,
∥ [ ∂ β , a ] ∇ u ∥ L 2 ≤ C ∥ a ∥ H 1 + ε ∥ u ∥ H m + 1 + C ∥ ∇ u ∥ B ∞ , 1 0 ∥ a ∥ H m , \|[\partial^\beta,a]\nabla u\|_{L^2}
\le C\|a\|_{H^{1+\varepsilon}}\|u\|_{H^{m+1}}
+C\|\nabla u\|_{B^0_{\infty,1}}\|a\|_{H^m}, ∥ [ ∂ β , a ] ∇ u ∥ L 2 ≤ C ∥ a ∥ H 1 + ε ∥ u ∥ H m + 1 + C ∥∇ u ∥ B ∞ , 1 0 ∥ a ∥ H m , and the same estimate holds with a a a replaced by any component of Y Y Y . In particular,
∥ [ ∂ β , Y ] ∇ u ∥ L 2 + ∥ [ ∂ β , ∇ u ] Y ∥ L 2 ≤ C ∥ Y ∥ H 1 + ε ∥ u ∥ H m + 1 + C ∥ ∇ u ∥ B ∞ , 1 0 ∥ Y ∥ H m . \|[\partial^\beta,Y]\nabla u\|_{L^2}
+\|[\partial^\beta,\nabla u]Y\|_{L^2}
\le C\|Y\|_{H^{1+\varepsilon}}\|u\|_{H^{m+1}}
+C\|\nabla u\|_{B^0_{\infty,1}}\|Y\|_{H^m}. ∥ [ ∂ β , Y ] ∇ u ∥ L 2 + ∥ [ ∂ β , ∇ u ] Y ∥ L 2 ≤ C ∥ Y ∥ H 1 + ε ∥ u ∥ H m + 1 + C ∥∇ u ∥ B ∞ , 1 0 ∥ Y ∥ H m . The same right-hand side controls
∥ ∂ β ( ( ∇ u ) Y + Y ( ∇ u ) T ) ∥ L 2 \left\|\partial^\beta\left((\nabla u)Y+Y(\nabla u)^{T}\right)\right\|_{L^2} ∂ β ( ( ∇ u ) Y + Y ( ∇ u ) T ) L 2 after subtracting any explicitly displayed top-order term.
Proof. The first bound is the endpoint transport energy estimate of Lemma Lemma 3.2 . For the weighted bound, decompose the commutator by Bony’s paraproduct. The low-frequency velocity interactions are estimated by the endpoint modulus ∥ ∇ u ∥ B ∞ , 1 0 \|\nabla u\|_{B^0_{\infty,1}} ∥∇ u ∥ B ∞ , 1 0 , with the coefficient w w w placed in L ∞ L^\infty L ∞ . In the remaining high-frequency velocity interactions, the top derivative falls on u u u ; these terms are bounded by putting the corresponding derivative of f f f in H ε ↪ L p H^\varepsilon\hookrightarrow L^p H ε ↪ L p and the derivative of u u u in H 1 ↪ L q H^1\hookrightarrow L^q H 1 ↪ L q , giving
C ∥ w ∥ L ∞ ∥ f ∥ H 1 + ε ∥ u ∥ H m + 1 ∥ f ∥ H m . C\|w\|_{L^\infty}\|f\|_{H^{1+\varepsilon}}\|u\|_{H^{m+1}}\|f\|_{H^m}. C ∥ w ∥ L ∞ ∥ f ∥ H 1 + ε ∥ u ∥ H m + 1 ∥ f ∥ H m .
This is the form used below: after Young’s inequality, the last term is absorbed by the viscous dissipation at the cost of a coefficient proportional to ∥ f ∥ H 1 + ε 2 \|f\|_{H^{1+\varepsilon}}^2 ∥ f ∥ H 1 + ε 2 . For (6.2), decompose [ ∂ β , a ] ∇ u [\partial^\beta,a]\nabla u [ ∂ β , a ] ∇ u by Bony’s paraproduct. In the high-low and balanced pieces the derivative falling on a a a is placed in H ε ↪ L p H^\varepsilon\hookrightarrow L^p H ε ↪ L p , while the remaining derivative of u u u is placed in H 1 ↪ L q H^1\hookrightarrow L^q H 1 ↪ L q , with 1 / p = ( 1 − ε ) / 2 1/p=(1-\varepsilon)/2 1/ p = ( 1 − ε ) /2 , 1 / q = ε / 2 1/q=\varepsilon/2 1/ q = ε /2 . This gives
∥ ∇ a ∥ L p ∥ u ∥ W m , q ≤ C ∥ a ∥ H 1 + ε ∥ u ∥ H m + 1 . \|\nabla a\|_{L^p}\|u\|_{W^{m,q}}
\le C\|a\|_{H^{1+\varepsilon}}\|u\|_{H^{m+1}}. ∥∇ a ∥ L p ∥ u ∥ W m , q ≤ C ∥ a ∥ H 1 + ε ∥ u ∥ H m + 1 .
The low-high piece is the endpoint paraproduct controlled by ∥ ∇ u ∥ B ∞ , 1 0 ∥ a ∥ H m \|\nabla u\|_{B^0_{\infty,1}}\|a\|_{H^m} ∥∇ u ∥ B ∞ , 1 0 ∥ a ∥ H m . The estimates with Y Y Y are componentwise identical. Expanding ∂ β ( ( ∇ u ) Y ) \partial^\beta((\nabla u)Y) ∂ β (( ∇ u ) Y ) or ∂ β ( Y ( ∇ u ) T ) \partial^\beta(Y(\nabla u)^T) ∂ β ( Y ( ∇ u ) T ) and removing the chosen top-order term leaves exactly the same commutator structure.
Proposition 6.4 (Oldroyd–B active-deviatoric endpoint estimate). Let m ≥ 3 m \ge3 m ≥ 3 , 0 < ε < 1 0 < \varepsilon< 1 0 < ε < 1 , and suppose that
0 < c 0 I ≤ A ( t , x ) ≤ C 0 I 0 < c_0 I \le A(t,x) \le C_0 I 0 < c 0 I ≤ A ( t , x ) ≤ C 0 I on [ 0 , T ] [0,T] [ 0 , T ] . Define
E m ( t ) = ∥ u ∥ H m 2 + ∥ a − 1 ∥ H m 2 + α 4 ∑ ∣ β ∣ ≤ m ∫ a − 1 ∣ ∂ β Y ∣ 2 d x . \mathcal{E}_m(t)=\|u\|_{H^m}^2+\|a-1\|_{H^m}^2+\frac{\alpha}{4}\sum_{|\beta|\le m}\int a^{-1}|\partial^\beta Y|^2\,dx. E m ( t ) = ∥ u ∥ H m 2 + ∥ a − 1 ∥ H m 2 + 4 α ∣ β ∣ ≤ m ∑ ∫ a − 1 ∣ ∂ β Y ∣ 2 d x . Then
d d t E m + c ν ∥ u ∥ H m + 1 2 + c λ − 1 ( ∥ a − 1 ∥ H m 2 + α ∥ Y ∥ H m 2 ) ≤ C c 0 , C 0 ( 1 + ∥ ∇ u ∥ B 0 , 1 ∞ + ∥ Log A ∥ H 1 + ε 2 ) E m . \frac{d}{dt}\mathcal{E}_m+c\nu\|u\|_{H^{m+1}}^2+c\lambda^{-1}\bigl(\|a-1\|_{H^m}^2+\alpha\|Y\|_{H^m}^2\bigr)
\le C_{c_0,C_0}\left(1+\|\nabla u\|_{B^\infty_{0,1}}+\|\operatorname{Log}A\|_{H^{1+\varepsilon}}^2\right)\mathcal{E}_m. d t d E m + c ν ∥ u ∥ H m + 1 2 + c λ − 1 ( ∥ a − 1 ∥ H m 2 + α ∥ Y ∥ H m 2 ) ≤ C c 0 , C 0 ( 1 + ∥∇ u ∥ B 0 , 1 ∞ + ∥ Log A ∥ H 1 + ε 2 ) E m . Proof. Write D t = ∂ t + u ⋅ ∇ D_t=\partial_t+u\cdot\nabla D t = ∂ t + u ⋅ ∇ . The compact spectral window gives uniform upper and lower bounds for a a a and a − 1 a^{-1} a − 1 , so E m \mathcal{E}_m E m is equivalent to ∥ u ∥ H m 2 + ∥ a − 1 ∥ H m 2 + α ∥ Y ∥ H m 2 \|u\|_{H^m}^2+\|a-1\|_{H^m}^2+\alpha\|Y\|_{H^m}^2 ∥ u ∥ H m 2 + ∥ a − 1 ∥ H m 2 + α ∥ Y ∥ H m 2 .
Apply ∂ β \partial^\beta ∂ β , ∣ β ∣ ≤ m |\beta|\le m ∣ β ∣ ≤ m , to (14) and test by ∂ β u \partial^\beta u ∂ β u . Since div u = 0 \operatorname{div}u=0 div u = 0 , the transport term is a commutator and the pressure term vanishes. The elastic term is
α ∫ ∂ β u ⋅ ∂ β div Y d x = − α ∫ ∂ β Y : S ( ∂ β u ) d x . \alpha\int\partial^\beta u\cdot\partial^\beta\operatorname{div}Y\,dx=-\alpha\int\partial^\beta Y:S(\partial^\beta u)\,dx. α ∫ ∂ β u ⋅ ∂ β div Y d x = − α ∫ ∂ β Y : S ( ∂ β u ) d x . The remaining velocity terms are bounded by
C ∥ ∇ u ∥ B 0 , 1 ∞ ∥ u ∥ H m 2 − ν ∥ ∇ ∂ β u ∥ L 2 2 . C\|\nabla u\|_{B^\infty_{0,1}}\|u\|_{H^m}^2-\nu\|\nabla\partial^\beta u\|_{L^2}^2. C ∥∇ u ∥ B 0 , 1 ∞ ∥ u ∥ H m 2 − ν ∥∇ ∂ β u ∥ L 2 2 . Next differentiate the anisotropic equation (16) . After commuting ∂ β \partial^\beta ∂ β with the material derivative,
D t ∂ β Y + λ − 1 ∂ β Y = 2 a S ( ∂ β u ) + 2 [ ∂ β , a ] S ( u ) + ∂ β ( [ ∇ u Y + Y ( ∇ u ) T ] ) − [ ∂ β , u ⋅ ∇ ] Y . D_t\partial^\beta Y+\lambda^{-1}\partial^\beta Y=2aS(\partial^\beta u)+2[\partial^\beta,a]S(u)+\partial^\beta\bigl([\nabla u\,Y+Y(\nabla u)^{\mathsf T}]\bigr)-[\partial^\beta,u\cdot\nabla]Y. D t ∂ β Y + λ − 1 ∂ β Y = 2 a S ( ∂ β u ) + 2 [ ∂ β , a ] S ( u ) + ∂ β ( [ ∇ u Y + Y ( ∇ u ) T ] ) − [ ∂ β , u ⋅ ∇ ] Y . Test this identity by ( α / 2 ) a − 1 ∂ β Y (\alpha/2)a^{-1}\partial^\beta Y ( α /2 ) a − 1 ∂ β Y . Because div u = 0 \operatorname{div}u=0 div u = 0 ,
α 4 d d t ∫ a − 1 ∣ ∂ β Y ∣ 2 d x + α 2 λ ∫ a − 1 ∣ ∂ β Y ∣ 2 d x = α ∫ ∂ β Y : S ( ∂ β u ) d x + R β a + R β Y + R β t r + R β w , \frac{\alpha}{4}\frac{d}{dt}\int a^{-1}|\partial^\beta Y|^2\,dx+\frac{\alpha}{2\lambda}\int a^{-1}|\partial^\beta Y|^2\,dx
=\alpha\int\partial^\beta Y:S(\partial^\beta u)\,dx+\mathcal{R}^a_\beta+\mathcal{R}^Y_\beta+\mathcal{R}^{\mathrm{tr}}_\beta+\mathcal{R}^w_\beta, 4 α d t d ∫ a − 1 ∣ ∂ β Y ∣ 2 d x + 2 λ α ∫ a − 1 ∣ ∂ β Y ∣ 2 d x = α ∫ ∂ β Y : S ( ∂ β u ) d x + R β a + R β Y + R β tr + R β w , where
R β a = α ∫ a − 1 ∂ β Y : [ ∂ β , a ] S ( u ) d x , \mathcal{R}^a_\beta=\alpha\int a^{-1}\partial^\beta Y:[\partial^\beta,a]S(u)\,dx, R β a = α ∫ a − 1 ∂ β Y : [ ∂ β , a ] S ( u ) d x , R β Y \mathcal{R}^Y_\beta R β Y contains the quadratic stretching term ∂ β ( [ ∇ u Y + Y ( ∇ u ) T ] ) \partial^\beta([\nabla u\,Y+Y(\nabla u)^{\mathsf T}]) ∂ β ([ ∇ u Y + Y ( ∇ u ) T ]) , R β t r \mathcal{R}^{\mathrm{tr}}_\beta R β tr contains the transport commutator, and
R β w = α 4 ∫ D t ( a − 1 ) ∣ ∂ β Y ∣ 2 d x \mathcal{R}^w_\beta=\frac{\alpha}{4}\int D_t(a^{-1})|\partial^\beta Y|^2\,dx R β w = 4 α ∫ D t ( a − 1 ) ∣ ∂ β Y ∣ 2 d x is the contribution of the time-dependent weight. The first term on the right-hand side of the weighted identity cancels exactly with the elastic term from the velocity equation:
− α ∫ ∂ β Y : S ( ∂ β u ) d x + α ∫ ∂ β Y : S ( ∂ β u ) d x = 0. -\alpha\int\partial^\beta Y:S(\partial^\beta u)\,dx+\alpha\int\partial^\beta Y:S(\partial^\beta u)\,dx=0. − α ∫ ∂ β Y : S ( ∂ β u ) d x + α ∫ ∂ β Y : S ( ∂ β u ) d x = 0. This is the top-order cancellation identity. It occurs after the isotropic stress has been removed into the pressure and before the use of Young’s inequality. Consequently the remaining estimates have to control only commutators, the time derivative of the weight a − 1 a^{-1} a − 1 , the scalar equation for a a a , and lower-order stretching terms. This is the source of the L t 2 L^2_t L t 2 logarithmic coefficient.
We now bound the remainders. For the coefficient commutator, Lemma 6.2 gives, for every δ > 0 \delta>0 δ > 0 ,
∣ R β a ∣ ≤ C ∥ ∂ β Y ∥ L 2 ∥ [ ∂ β , a ] S ( u ) ∥ L 2 ≤ δ ν ∥ u ∥ H m + 1 2 + C δ ∥ a ∥ H 1 + ε 2 ∥ Y ∥ H m 2 + C ∥ ∇ u ∥ B ∞ , 1 0 E m . |R_{\beta}^{a}| \le C\|\partial^\beta Y\|_{L^2}\|[\partial^\beta,a]S(u)\|_{L^2}
\le\delta\nu\|u\|_{H^{m+1}}^2+C_\delta\|a\|_{H^{1+\varepsilon}}^2\|Y\|_{H^m}^2+C\|\nabla u\|_{B^0_{\infty,1}}\mathcal{E}_m. ∣ R β a ∣ ≤ C ∥ ∂ β Y ∥ L 2 ∥ [ ∂ β , a ] S ( u ) ∥ L 2 ≤ δ ν ∥ u ∥ H m + 1 2 + C δ ∥ a ∥ H 1 + ε 2 ∥ Y ∥ H m 2 + C ∥∇ u ∥ B ∞ , 1 0 E m .
The same lemma gives
∣ R β Y ∣ ≤ δ ν ∥ u ∥ H m + 1 2 + C δ ∥ Y ∥ H 1 + ε 2 ∥ Y ∥ H m 2 + C ∥ ∇ u ∥ B ∞ , 1 0 E m . |R_{\beta}^{Y}| \le\delta\nu\|u\|_{H^{m+1}}^2+C_\delta\|Y\|_{H^{1+\varepsilon}}^2\|Y\|_{H^m}^2+C\|\nabla u\|_{B^0_{\infty,1}}\mathcal{E}_m. ∣ R β Y ∣ ≤ δ ν ∥ u ∥ H m + 1 2 + C δ ∥ Y ∥ H 1 + ε 2 ∥ Y ∥ H m 2 + C ∥∇ u ∥ B ∞ , 1 0 E m .
The transport commutator is controlled by the weighted transport estimate in Lemma 6.2. The part with a top derivative on u u u is absorbed by viscosity:
∣ R β t r ∣ ≤ δ ν ∥ u ∥ H m + 1 2 + C δ ∥ Y ∥ H 1 + ε 2 ∥ Y ∥ H m 2 + C c 0 , C 0 ∥ ∇ u ∥ B ∞ , 1 0 ∥ Y ∥ H m 2 . |R_{\beta}^{tr}| \le\delta\nu\|u\|_{H^{m+1}}^2+C_\delta\|Y\|_{H^{1+\varepsilon}}^2\|Y\|_{H^m}^2+C_{c_0,C_0}\|\nabla u\|_{B^0_{\infty,1}}\|Y\|_{H^m}^2. ∣ R β t r ∣ ≤ δ ν ∥ u ∥ H m + 1 2 + C δ ∥ Y ∥ H 1 + ε 2 ∥ Y ∥ H m 2 + C c 0 , C 0 ∥∇ u ∥ B ∞ , 1 0 ∥ Y ∥ H m 2 .
Finally, from (15) ,
D t ( a − 1 ) = λ − 1 a − 2 ( a − 1 ) − a − 2 S ( u ) : Y . D_t(a^{-1})=\lambda^{-1}a^{-2}(a-1)-a^{-2}S(u):Y. D t ( a − 1 ) = λ − 1 a − 2 ( a − 1 ) − a − 2 S ( u ) : Y .
Hence, using the compact spectral window and ∥ ∇ u ∥ L ∞ ≤ C ∥ ∇ u ∥ B ∞ , 1 0 \|\nabla u\|_{L^\infty}\le C\|\nabla u\|_{B^0_{\infty,1}} ∥∇ u ∥ L ∞ ≤ C ∥∇ u ∥ B ∞ , 1 0 ,
∣ R β w ∣ ≤ C c 0 , C 0 ( 1 + ∥ ∇ u ∥ B ∞ , 1 0 ) ∥ Y ∥ H m 2 . |R_{\beta}^{w}|\le C_{c_0,C_0}\left(1+\|\nabla u\|_{B^0_{\infty,1}}\right)\|Y\|_{H^m}^2. ∣ R β w ∣ ≤ C c 0 , C 0 ( 1 + ∥∇ u ∥ B ∞ , 1 0 ) ∥ Y ∥ H m 2 .
It remains to estimate the scalar equation. Applying ∂ β \partial^\beta ∂ β to (15) , testing by ∂ β ( a − 1 ) \partial^\beta(a-1) ∂ β ( a − 1 ) , and using Lemma 6.2 gives
1 2 d d t ∥ a − 1 ∥ H m 2 + λ − 1 ∥ a − 1 ∥ H m 2 ≤ δ ν ∥ u ∥ H m + 1 2 + C δ ∥ Y ∥ H 1 + ε 2 ∥ a − 1 ∥ H m 2 + C ∥ ∇ u ∥ B ∞ , 1 0 E m . \frac{1}{2}\frac{d}{dt}\|a-1\|_{H^m}^2+\lambda^{-1}\|a-1\|_{H^m}^2\le\delta\nu\|u\|_{H^{m+1}}^2+C_\delta\|Y\|_{H^{1+\varepsilon}}^2\|a-1\|_{H^m}^2+C\|\nabla u\|_{B^0_{\infty,1}}\mathcal{E}_m. 2 1 d t d ∥ a − 1 ∥ H m 2 + λ − 1 ∥ a − 1 ∥ H m 2 ≤ δ ν ∥ u ∥ H m + 1 2 + C δ ∥ Y ∥ H 1 + ε 2 ∥ a − 1 ∥ H m 2 + C ∥∇ u ∥ B ∞ , 1 0 E m .
On the compact spectral window, the smooth maps B = Log A ↦ a B=\operatorname{Log}A\mapsto a B = Log A ↦ a and B ↦ Y B\mapsto Y B ↦ Y satisfy
∥ a ∥ H 1 + ε + ∥ Y ∥ H 1 + ε ≤ C c 0 , C 0 ( 1 + ∥ Log A ∥ H 1 + ε ) . \|a\|_{H^{1+\varepsilon}}+\|Y\|_{H^{1+\varepsilon}}\le C_{c_0,C_0}\left(1+\|\operatorname{Log}A\|_{H^{1+\varepsilon}}\right). ∥ a ∥ H 1 + ε + ∥ Y ∥ H 1 + ε ≤ C c 0 , C 0 ( 1 + ∥ Log A ∥ H 1 + ε ) .
Choose δ > 0 \delta>0 δ > 0 small enough that the velocity terms in the displayed commutator, stretching, and scalar estimates are absorbed by the viscous dissipation. Summing over ∣ β ∣ ≤ m |\beta|\le m ∣ β ∣ ≤ m , using the compact-window composition estimate, and using the weighted equivalence of E m \mathcal{E}_m E m gives the claimed differential inequality.
Proposition 6.5 (Compact-window logarithmic estimate). Let 0 < ε < 1 0<\varepsilon<1 0 < ε < 1 , s = 1 + ε s=1+\varepsilon s = 1 + ε , and let ( u , a , Y ) (u,a,Y) ( u , a , Y ) be a smooth solution of (7)–(9) on [ 0 , T ] [0,T] [ 0 , T ] , with C = a I + Y C=aI+Y C = a I + Y remaining in a compact conformation window K ⋐ S + + 2 K\Subset S_{++}^2 K ⋐ S ++ 2 . Put h = a − a ∗ h=a-a_* h = a − a ∗ and
B ( t ) = ∥ ∇ u ( t ) ∥ B ∞ , 1 0 . B(t)=\|\nabla u(t)\|_{B^0_{\infty,1}}. B ( t ) = ∥∇ u ( t ) ∥ B ∞ , 1 0 . Then, for a sufficiently small fixed scalar weight σ > 0 \sigma>0 σ > 0 , the low-order energy
E s l o w = ∥ u ∥ H s 2 + σ ∥ h ∥ H s 2 + ∥ Y ∥ H s 2 \mathcal{E}^{\mathrm{low}}_s=\|u\|_{H^s}^2+\sigma\|h\|_{H^s}^2+\|Y\|_{H^s}^2 E s low = ∥ u ∥ H s 2 + σ ∥ h ∥ H s 2 + ∥ Y ∥ H s 2 satisfies
d d t E s l o w + c ν ∥ u ∥ H s + 1 2 + c λ − 1 ( ∥ h ∥ H s 2 + ∥ Y ∥ H s 2 ) ≤ C K ( 1 + B ( t ) ) E s l o w . \frac{d}{dt}\mathcal{E}^{\mathrm{low}}_s+c\nu\|u\|_{H^{s+1}}^2+c\lambda^{-1}\left(\|h\|_{H^s}^2+\|Y\|_{H^s}^2\right)\le C_K(1+B(t))\mathcal{E}^{\mathrm{low}}_s. d t d E s low + c ν ∥ u ∥ H s + 1 2 + c λ − 1 ( ∥ h ∥ H s 2 + ∥ Y ∥ H s 2 ) ≤ C K ( 1 + B ( t )) E s low . Consequently, if T < ∞ T<\infty T < ∞ and ∫ 0 T B ( t ) d t < ∞ \int_0^T B(t)\,dt<\infty ∫ 0 T B ( t ) d t < ∞ , then
Log C ∈ L 2 ( 0 , T ; H 1 + ε ) . \operatorname{Log}C\in L^2(0,T;H^{1+\varepsilon}). Log C ∈ L 2 ( 0 , T ; H 1 + ε ) .
Proof. At the fractional level s = 1 + ε s=1+\varepsilon s = 1 + ε , the transport terms are controlled by the endpoint commutator estimate in Lemma 3.2 , giving the velocity clock B ( t ) B(t) B ( t ) . On the compact window, all coefficient functions and their derivatives are bounded in L ∞ L^\infty L ∞ . The source S ( u ) : Y S(u):Y S ( u ) : Y in the scalar equation and the stretching source 2 a S ( u ) 2aS(u) 2 a S ( u ) in the trace-free equation are controlled by Young’s inequality and the viscous dissipation, using the compact L ∞ L^\infty L ∞ bounds for a a a and Y Y Y . The positive deviatoric relaxation μ ≥ μ 0 \mu\ge\mu_0 μ ≥ μ 0 gives Y Y Y -damping. The scalar relaxation is split as
G ( a , η ) = ∂ a G ( a ∗ , 0 ) h + G ~ ( h , Y ) , G(a,\eta)=\partial_aG(a_\ast,0)h+\widetilde{G}(h,Y), G ( a , η ) = ∂ a G ( a ∗ , 0 ) h + G ( h , Y ) ,
and the monotonicity ∂ a G ≥ κ \partial_aG\ge\kappa ∂ a G ≥ κ , together with a small scalar energy weight, gives h h h -damping. The part of the smooth remainder coming from the η \eta η -dependence is at least quadratic in Y Y Y ; after testing the scalar equation by σ h \sigma h σ h , it is absorbed by the Y Y Y -relaxation and the scalar damping, up to the right-hand side of (6.5). Summing the velocity, scalar, and trace-free estimates gives (6.5).
Gronwall’s inequality gives a uniform bound for E s l o w \mathcal{E}^{\mathrm{low}}_s E s low and an L t 2 H x s L^2_tH^s_x L t 2 H x s bound for h h h and Y Y Y . On K K K , the map ( h , Y ) ↦ Log ( a I + Y ) (h,Y)\mapsto\operatorname{Log}(aI+Y) ( h , Y ) ↦ Log ( a I + Y ) is smooth on Sobolev spaces with s > 1 s>1 s > 1 . Hence
∥ Log C ∥ H s ≤ C K ( 1 + ∥ h ∥ H s + ∥ Y ∥ H s ) , \|\operatorname{Log}C\|_{H^s}\le C_K(1+\|h\|_{H^s}+\|Y\|_{H^s}), ∥ Log C ∥ H s ≤ C K ( 1 + ∥ h ∥ H s + ∥ Y ∥ H s ) ,
which proves the stated logarithmic estimate.
This also records the time exponent. After the pressure renormalization and weighted anisotropic cancellation, all remaining high-order remainders are bounded by the square of ∥ Log A ∥ H 1 + ε \|\operatorname{Log}A\|_{H^{1+\varepsilon}} ∥ Log A ∥ H 1 + ε . The continuation argument therefore needs the L t 2 H x 1 + ε L^2_tH^{1+\varepsilon}_x L t 2 H x 1 + ε logarithmic estimate, rather than an external L t 4 L^4_t L t 4 logarithmic hypothesis.
Theorem 6.1 (Endpoint Oldroyd–B continuation criterion). Let m ≥ 3 m\ge3 m ≥ 3 , 0 < ε < 1 0<\varepsilon<1 0 < ε < 1 , and let ( u , A ) (u,A) ( u , A ) be a strong positive-cone solution with Sobolev index s = m s=m s = m on [ 0 , T ) [0,T) [ 0 , T ) . If
∫ 0 T ∥ ∇ u ( t ) ∥ B ∞ , 1 0 d t < ∞ , \int_0^T\|\nabla u(t)\|_{B^0_{\infty,1}}\,dt<\infty, ∫ 0 T ∥∇ u ( t ) ∥ B ∞ , 1 0 d t < ∞ , then A A A remains in a compact spectral window of the positive cone on [ 0 , T ) [0,T) [ 0 , T ) , and the strong solution continues beyond T T T .
Proof. The criterion propagates a compact spectral window by Lemma 6.1. On this propagated window, Proposition 6.5 gives
Log A ∈ L 2 ( 0 , T ; H 1 + ε ) . \operatorname{Log}A\in L^2(0,T;H^{1+\varepsilon}). Log A ∈ L 2 ( 0 , T ; H 1 + ε ) .
Proposition 6.4 gives
d d t E m ( t ) ≤ C ( 1 + ∥ ∇ u ( t ) ∥ B ∞ , 1 0 + ∥ Log A ( t ) ∥ H 1 + ε 2 ) E m ( t ) , \frac{d}{dt}\mathcal{E}_m(t)\le C\left(1+\|\nabla u(t)\|_{B^0_{\infty,1}}+\|\operatorname{Log}A(t)\|_{H^{1+\varepsilon}}^2\right)\mathcal{E}_m(t), d t d E m ( t ) ≤ C ( 1 + ∥∇ u ( t ) ∥ B ∞ , 1 0 + ∥ Log A ( t ) ∥ H 1 + ε 2 ) E m ( t ) ,
where the constant depends only on the propagated window and the fixed parameters. The coefficient is integrable by the velocity clock and by the logarithmic bound. Gronwall’s inequality therefore gives
sup t < T E m ( t ) < ∞ . \sup_{t<T}\mathcal{E}_m(t)<\infty. t < T sup E m ( t ) < ∞.
Since the spectral window is compact, the bound on ( a , Y ) (a,Y) ( a , Y ) is equivalent to an H m H^m H m bound on A A A . The local continuation principle, Proposition 2.1, then extends the solution beyond T T T .
Corollary 6.1 (Oldroyd–B breakdown alternative). Let T ∗ T_\ast T ∗ be the first breakdown time of a strong Oldroyd–B solution. Then the endpoint velocity clock diverges:
∫ 0 T ∗ ∥ ∇ u ( t ) ∥ B ∞ , 1 0 d t = ∞ . \int_0^{T_\ast}\|\nabla u(t)\|_{B^0_{\infty,1}}\,dt=\infty. ∫ 0 T ∗ ∥∇ u ( t ) ∥ B ∞ , 1 0 d t = ∞. Loss of a compact positive-cone spectral window is not a separate Oldroyd–B alternative under finite endpoint velocity clock; it is ruled out by Lemma 6.1.
Proof. If the clock were finite, Theorem 6.1 would continue the solution beyond T ∗ T_* T ∗ , a contradiction.
Corollary 6.2 (Coupled vorticity–stress form). At a finite Oldroyd–B breakdown time,
∫ 0 T ∗ ∥ ω ( t ) ∥ B ∞ , 1 0 d t = ∞ . \int_0^{T_*} \|\omega(t)\|_{B^0_{\infty,1}}\,dt = \infty. ∫ 0 T ∗ ∥ ω ( t ) ∥ B ∞ , 1 0 d t = ∞.
Proof. Use the periodic Biot–Savart estimate for the endpoint Besov norm and Corollary 6.1; the entropy identity gives u ∈ L t ∞ L x 2 u \in L^\infty_t L^2_x u ∈ L t ∞ L x 2 on every smooth finite time interval, so the L 2 L^2 L 2 term in the Biot–Savart bound is harmless.
A Functional Obstruction at the Logarithmic Threshold
The endpoint theorem removes the compact spectral window as an independent Oldroyd–B hypothesis and, through Proposition 6.5, provides the logarithmic differentiability needed in the high-order estimate. The threshold is a real derivative threshold in the pressure-free stress map, rather than a disguised entropy bound. Entropy and relaxation control zeroth-order functions of the eigenvalues of A A A , and the endpoint velocity clock propagates pointwise spectral bounds, but zero-order spectral information alone does not control rapid oscillation of the active deviatoric stress. The next results record this obstruction first through bounded logarithmic amplitudes and then through a fixed-spectrum eigenframe oscillation, where every spectral density is exactly constant while the active deviatoric stress oscillates at high frequency.
Theorem 7.1 (High-frequency entropy-blind sequence). Let H ∈ S 2 H \in\mathbb{S}^2 H ∈ S 2 be nonzero and trace free, and let
B N ( x ) = δ sin ( N x 1 ) H , A N = e B N . B_N(x) = \delta\sin(Nx_1)H,\qquad A_N = e^{B_N}. B N ( x ) = δ sin ( N x 1 ) H , A N = e B N . For each fixed δ > 0 \delta> 0 δ > 0 , the fields A N A_N A N remain in a compact spectral window independent of N N N , and the entropy density
tr A N − log det A N − 2 \operatorname{tr} A_N - \log\det A_N - 2 tr A N − log det A N − 2 is bounded uniformly in N N N in every L p L^p L p , 1 ≤ p < ∞ 1 \le p < \infty 1 ≤ p < ∞ , while
∥ B N ∥ H 1 + ε → ∞ as N → ∞ . \|B_N\|_{H^{1+\varepsilon}} \to\infty\qquad\text{as } N \to\infty. ∥ B N ∥ H 1 + ε → ∞ as N → ∞.
Proof. The spectrum of B N B_N B N is contained in a fixed compact interval depending only on δ \delta δ and H H H . Therefore the eigenvalues of A N = e B N A_N = e^{B_N} A N = e B N remain in a fixed compact subinterval of ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) , and the entropy density is uniformly bounded. The H 1 + ε H^{1+\varepsilon} H 1 + ε norm grows like N 1 + ε N^{1+\varepsilon} N 1 + ε .
Remark 7.1 (Interpretation). Theorem 7.1 does not construct a solution. It shows that any argument using only entropy, relaxation, and positivity cannot control the high-frequency logarithmic term. The continuation proof therefore needs the flow-map clock, which propagates the compact window and yields the logarithmic differentiability used by the high-order estimate.
Proposition 7.1 (Spectral energies are blind to the active high-frequency channel). Let T : S + + 2 → S 2 T : \mathbb{S}^2_{++} \to\mathbb{S}^2 T : S ++ 2 → S 2 be a smooth spectral isotropic stress law. Fix distinct eigenvalues λ 1 ≠ λ 2 \lambda_1 \ne\lambda_2 λ 1 = λ 2 in a compact positive-cone window and assume that the active coefficient in
T ( C ) = q 0 ( tr C , det C ) I + q 1 ( tr C , det C ) C T(C) = q_0(\operatorname{tr} C,\det C)I + q_1(\operatorname{tr} C,\det C)C T ( C ) = q 0 ( tr C , det C ) I + q 1 ( tr C , det C ) C satisfies q 1 ( λ 1 + λ 2 , λ 1 λ 2 ) ≠ 0 q_1(\lambda_1+\lambda_2,\lambda_1\lambda_2) \ne0 q 1 ( λ 1 + λ 2 , λ 1 λ 2 ) = 0 . Then there are smooth fields C N : T 2 → S + + 2 C_N : \mathbb{T}^2 \to\mathbb{S}^2_{++} C N : T 2 → S ++ 2 such that every zero-order spectral density E ( tr C N , det C N ) E(\operatorname{tr} C_N,\det C_N) E ( tr C N , det C N ) is independent of x x x and N N N , while
∥ log C N ∥ H 1 + ε + ∥ T ( C N ) ∥ H 1 + ε → ∞ . \|\log C_N\|_{H^{1+\varepsilon}}+\|T(C_N)\|_{H^{1+\varepsilon}} \to\infty. ∥ log C N ∥ H 1 + ε + ∥ T ( C N ) ∥ H 1 + ε → ∞.
Proof. Let
Λ = ( λ 1 0 0 λ 2 ) , R ϑ = ( cos ϑ − sin ϑ sin ϑ cos ϑ ) , \Lambda= \begin{pmatrix} \lambda_1 & 0 \\ 0 & \lambda_2 \end{pmatrix}, \qquad R_\vartheta= \begin{pmatrix} \cos\vartheta& -\sin\vartheta\\ \sin\vartheta& \cos\vartheta\end{pmatrix}, Λ = ( λ 1 0 0 λ 2 ) , R ϑ = ( cos ϑ sin ϑ − sin ϑ cos ϑ ) ,
and set C N ( x ) = R N x 1 Λ R N x 1 T C_N(x)=R_{Nx_1}\Lambda R_{Nx_1}^{T} C N ( x ) = R N x 1 Λ R N x 1 T . The eigenvalues of C N C_N C N are constant. Hence tr C N = λ 1 + λ 2 \operatorname{tr} C_N=\lambda_1+\lambda_2 tr C N = λ 1 + λ 2 , det C N = λ 1 λ 2 \det C_N=\lambda_1\lambda_2 det C N = λ 1 λ 2 , and every zero-order spectral density is constant.
On the other hand,
log C N = R N x 1 ( log Λ ) R N x 1 T \log C_N=R_{Nx_1}(\log\Lambda)R_{Nx_1}^{T} log C N = R N x 1 ( log Λ ) R N x 1 T
has a nonzero oscillatory trace-free part at frequency 2 N 2N 2 N , because λ 1 ≠ λ 2 \lambda_1\ne\lambda_2 λ 1 = λ 2 . Therefore
∥ log C N ∥ H 1 + ε ≳ N 1 + ε . \lVert\log C_N\rVert_{H^{1+\varepsilon}}\gtrsim N^{1+\varepsilon}. ∥ log C N ∥ H 1 + ε ≳ N 1 + ε .
Moreover
T ( C N ) ∘ = q 1 ( λ 1 + λ 2 , λ 1 λ 2 ) C N ∘ , T(C_N)^\circ=q_1(\lambda_1+\lambda_2,\lambda_1\lambda_2)C_N^\circ, T ( C N ) ∘ = q 1 ( λ 1 + λ 2 , λ 1 λ 2 ) C N ∘ ,
and C N ∘ C_N^\circ C N ∘ has the same frequency 2 N 2N 2 N . Since the displayed active coefficient is nonzero,
∥ T ( C N ) ∘ ∥ H 1 + ε ≳ N 1 + ε . \lVert T(C_N)^\circ\rVert_{H^{1+\varepsilon}}\gtrsim N^{1+\varepsilon}. ∥ T ( C N ) ∘ ∥ H 1 + ε ≳ N 1 + ε .
Theorem 7.2 (Static active-forcing obstruction). Let T : S + + 2 → S 2 T:\mathbb{S}^2_{++}\to\mathbb{S}^2 T : S ++ 2 → S 2 be a smooth spectral isotropic stress law and let K ⋐ S + + 2 K\Subset\mathbb{S}^2_{++} K ⋐ S ++ 2 . Suppose K K K contains the fixed spectral orbit of an anisotropic state
C ∗ = a I + r E 1 ∈ K , E 1 = ( 1 0 0 − 1 ) , r > 0 , C_\ast=aI+rE_1\in K,\qquad E_1=\begin{pmatrix}1&0\\0&-1\end{pmatrix},\qquad r>0, C ∗ = a I + r E 1 ∈ K , E 1 = ( 1 0 0 − 1 ) , r > 0 , such that the active coefficient in
T ( C ) = q 0 ( tr C , det C ) I + q 1 ( tr C , det C ) C T(C)=q_0(\operatorname{tr}C,\det C)I+q_1(\operatorname{tr}C,\det C)C T ( C ) = q 0 ( tr C , det C ) I + q 1 ( tr C , det C ) C satisfies q 1 ( 2 a , a 2 − r 2 ) r ≠ 0 q_1(2a,a^2-r^2)r\ne0 q 1 ( 2 a , a 2 − r 2 ) r = 0 . Let 0 ≤ γ < 1 0\le\gamma<1 0 ≤ γ < 1 . For every 0 < σ < 1 + γ 0<\sigma<1+\gamma 0 < σ < 1 + γ , there exist smooth fields C N : T 2 → K C_N:\mathbb{T}^2\to K C N : T 2 → K with fixed eigenvalues a + r a+r a + r and a − r a-r a − r such that every zeroth-order spectral density is pointwise independent of N N N ,
sup N ∥ log C N ∥ H σ < ∞ , \sup_N\lVert\log C_N\rVert_{H^\sigma}<\infty, N sup ∥ log C N ∥ H σ < ∞ , but
∥ curl div T ( C N ) ∥ H − 1 + γ → ∞ . \lVert\operatorname{curl}\operatorname{div}T(C_N)\rVert_{H^{-1+\gamma}}\to\infty. ∥ curl div T ( C N ) ∥ H − 1 + γ → ∞. Conversely, for every smooth C : T 2 → K C:\mathbb{T}^2\to K C : T 2 → K ,
∥ curl div T ( C ) ∥ H − 1 + γ ≤ C K ( 1 + ∥ log C ∥ H 1 + γ ) . \lVert\operatorname{curl}\operatorname{div}T(C)\rVert_{H^{-1+\gamma}}\le C_K\left(1+\lVert\log C\rVert_{H^{1+\gamma}}\right). ∥ curl div T ( C ) ∥ H − 1 + γ ≤ C K ( 1 + ∥ log C ∥ H 1 + γ ) . Thus this static pressure-free stress estimate cannot be formulated below the H 1 + γ H^{1+\gamma} H 1 + γ logarithmic threshold without adding further dynamical information or a different structure.
Proof. Let
R θ = ( cos θ − sin θ sin θ cos θ ) , θ N = N − σ sin ( N x 1 ) , R_\theta=\begin{pmatrix}\cos\theta&-\sin\theta\\\sin\theta&\cos\theta\end{pmatrix},\qquad\theta_N=N^{-\sigma}\sin(Nx_1), R θ = ( cos θ sin θ − sin θ cos θ ) , θ N = N − σ sin ( N x 1 ) ,
and set
C N = a I + r R θ N E 1 R θ N T . C_N=aI+rR_{\theta_N}E_1R_{\theta_N}^{T}. C N = a I + r R θ N E 1 R θ N T .
By the orbit assumption, C N ∈ K C_N\in K C N ∈ K for all N N N . The eigenvalues of C N C_N C N are exactly a + r a+r a + r and a − r a-r a − r , so every zeroth-order spectral density is pointwise constant. Since θ ↦ log ( a I + r R θ E 1 R θ T ) \theta\mapsto\log(aI+rR_\theta E_1R_\theta^T) θ ↦ log ( a I + r R θ E 1 R θ T ) is smooth and θ N \theta_N θ N has amplitude N − σ N^{-\sigma} N − σ and frequency N N N , the Sobolev bound sup N ∥ log C N ∥ H σ < ∞ \sup_N\lVert\log C_N\rVert_{H^\sigma}<\infty sup N ∥ log C N ∥ H σ < ∞ follows. On this fixed spectral orbit,
T ( C N ) ∘ = q 1 ( 2 a , a 2 − r 2 ) r R θ N E 1 R θ N T T(C_N)^\circ= q_1(2a,a^2-r^2)\,rR_{\theta_N}E_1R_{\theta_N}^{\mathsf T} T ( C N ) ∘ = q 1 ( 2 a , a 2 − r 2 ) r R θ N E 1 R θ N T
The 12-entry is
q 1 ( 2 a , a 2 − r 2 ) r sin ( 2 θ N ) = 2 q 1 ( 2 a , a 2 − r 2 ) r N − σ sin ( N x 1 ) + O ( N − 2 σ ) . q_1(2a,a^2-r^2)\,r\sin(2\theta_N)=2q_1(2a,a^2-r^2)rN^{-\sigma}\sin(Nx_1)+O(N^{-2\sigma}). q 1 ( 2 a , a 2 − r 2 ) r sin ( 2 θ N ) = 2 q 1 ( 2 a , a 2 − r 2 ) r N − σ sin ( N x 1 ) + O ( N − 2 σ ) .
Since C N C_N C N depends only on x 1 x_1 x 1 ,
curl div T ( C N ) = ∂ 11 ( T ( C N ) 12 ) , \operatorname{curl}\operatorname{div}T(C_N)=\partial_{11}(T(C_N)_{12}), curl div T ( C N ) = ∂ 11 ( T ( C N ) 12 ) ,
and therefore
∥ curl div T ( C N ) ∥ H − 1 + γ ≳ N 1 + γ − σ → ∞ . \|\operatorname{curl}\operatorname{div}T(C_N)\|_{H^{-1+\gamma}}\gtrsim N^{1+\gamma-\sigma}\to\infty. ∥ curl div T ( C N ) ∥ H − 1 + γ ≳ N 1 + γ − σ → ∞.
For the converse estimate, the pressure part of T ( C ) T(C) T ( C ) has zero curl-divergence. On K K K , the map Log C ↦ T ( C ) ∘ \operatorname{Log}C\mapsto T(C)^\circ Log C ↦ T ( C ) ∘ is smooth on Sobolev spaces, so
∥ T ( C ) ∘ ∥ H 1 + γ ≤ C K ( 1 + ∥ Log C ∥ H 1 + γ ) . \|T(C)^\circ\|_{H^{1+\gamma}}\leq C_K(1+\|\operatorname{Log}C\|_{H^{1+\gamma}}). ∥ T ( C ) ∘ ∥ H 1 + γ ≤ C K ( 1 + ∥ Log C ∥ H 1 + γ ) .
Since curl div is a second-order operator, it maps H 1 + γ H^{1+\gamma} H 1 + γ to H − 1 + γ H^{-1+\gamma} H − 1 + γ , which proves the endpoint bound.
Proposition 7.2 (Relaxation does not give a derivative). Let B N , A N B_N,A_N B N , A N be as in Theorem 7.1. Then the relaxation density
tr ( A N + A N − 1 − 2 I ) \operatorname{tr}(A_N+A_N^{-1}-2I) tr ( A N + A N − 1 − 2 I ) is uniformly bounded in every L p L^p L p , 1 ≤ p < ∞ 1\leq p<\infty 1 ≤ p < ∞ , while ∥ B N ∥ H 1 + ε → ∞ \|B_N\|_{H^{1+\varepsilon}}\to\infty ∥ B N ∥ H 1 + ε → ∞ . In particular, the entropy dissipation term in (4.1) cannot control the logarithmic concentration appearing in Theorem 6.1.
Proof. The eigenvalues of A N A_N A N and A N − 1 A_N^{-1} A N − 1 remain in a compact interval depending only on δ \delta δ and H H H . Hence the relaxation density is uniformly bounded pointwise. The Sobolev growth of B N B_N B N is unchanged from Theorem 7.1.
The finite-extensibility barrier for FENE-P
This section is two-dimensional: b > 2 b>2 b > 2 , f b = f b , 2 f_b=f_{b,2} f b = f b , 2 , and D b = D b , 2 D_b=D_{b,2} D b = D b , 2 . The three-dimensional FENE-P normalization is recorded separately in § 9 .
We now record the part of the argument which is not present in Oldroyd–B. In the FENE-P system the conformation tensor must remain in
D b = D b , 2 = { C ∈ S + + 2 : tr C < b } . \mathcal{D}_b = \mathcal{D}_{b,2} = \{C \in\mathbb{S}_{++}^{2} : \operatorname{tr} C < b\}. D b = D b , 2 = { C ∈ S ++ 2 : tr C < b } .
The logarithmic conformation B = Log C B = \operatorname{Log} C B = Log C still controls the lower spectral boundary of the positive cone, but it does not control the upper trace boundary. The missing coordinate is
ϕ b ( C ) = − log ( b − tr C ) . \phi_b(C) = -\log(b-\operatorname{tr} C). ϕ b ( C ) = − log ( b − tr C ) .
The role of this section is to show that this boundary is propagated by the squared endpoint velocity clock and that the high-order estimate closes in the same pressure-free active-deviatoric variables.
Definition 8.1 (Compact FENE window). For fixed b > 2 b > 2 b > 2 , a FENE-P solution lies in a compact FENE window on [ 0 , T ] [0,T] [ 0 , T ] if there exist constants 0 < c 0 < c 1 < ∞ 0 < c_0 < c_1 < \infty 0 < c 0 < c 1 < ∞ and δ > 0 \delta> 0 δ > 0 such that
c 0 I ≤ C ( t , x ) ≤ c 1 I , tr C ( t , x ) ≤ b − δ c_0 I \le C(t,x) \le c_1 I,\qquad\operatorname{tr} C(t,x) \le b-\delta c 0 I ≤ C ( t , x ) ≤ c 1 I , tr C ( t , x ) ≤ b − δ for all ( t , x ) ∈ [ 0 , T ] × T 2 (t,x) \in[0,T] \times\mathbb{T}^{2} ( t , x ) ∈ [ 0 , T ] × T 2 .
Lemma 8.1 (FENE window recovery). Let C ( x ) ∈ D b C(x) \in\mathcal{D}_b C ( x ) ∈ D b . If
∥ Log C ∥ L ∞ ≤ M , ∥ ϕ b ( C ) ∥ L ∞ ≤ M b , \|\operatorname{Log} C\|_{L^\infty} \le M,\qquad\|\phi_b(C)\|_{L^\infty} \le M_b, ∥ Log C ∥ L ∞ ≤ M , ∥ ϕ b ( C ) ∥ L ∞ ≤ M b , then
e − M I ≤ C ≤ e M I , tr C ≤ b − e − M b . e^{-M}I \le C \le e^{M}I,\qquad\operatorname{tr} C \le b-e^{-M_b}. e − M I ≤ C ≤ e M I , tr C ≤ b − e − M b . In particular the logarithmic field and the barrier field recover a compact FENE window.
Proof. The eigenvalues of C C C are exponentials of the eigenvalues of Log C \operatorname{Log} C Log C , which gives the spectral bounds. The barrier bound gives − log ( b − tr C ) ≤ M b -\log(b-\operatorname{tr} C) \le M_b − log ( b − tr C ) ≤ M b , hence b − tr C ≥ e − M b b-\operatorname{tr} C \ge e^{-M_b} b − tr C ≥ e − M b .
Lemma 8.2 (Smooth coordinates on compact FENE windows). Let s > 1 s > 1 s > 1 . On every compact FENE window K ⋐ D b K \Subset\mathcal{D}_b K ⋐ D b , the quantities
∥ C ∥ H s , ∥ Log C ∥ H s , ∥ T b ( C ) ∥ H s \|C\|_{H^s},\qquad\|\operatorname{Log} C\|_{H^s},\qquad\|T_b(C)\|_{H^s} ∥ C ∥ H s , ∥ Log C ∥ H s , ∥ T b ( C ) ∥ H s are mutually controlled, with constants depending on K , s , b K,s,b K , s , b . Moreover
∥ f b ( C ) ∥ H s ≤ C K , s , b ( 1 + ∥ ϕ b ( C ) ∥ H s ) . \|f_b(C)\|_{H^s} \le C_{K,s,b}\left(1+\|\phi_b(C)\|_{H^s}\right). ∥ f b ( C ) ∥ H s ≤ C K , s , b ( 1 + ∥ ϕ b ( C ) ∥ H s ) .
Proof. The first statement follows from the Sobolev composition theorem on compact subsets of D b \mathcal{D}_b D b . For the spring factor,
f b ( C ) = b − 2 b − tr C = ( b − 2 ) e ϕ b ( C ) . f_b(C)=\frac{b-2}{b-\operatorname{tr} C}=(b-2)e^{\phi_b(C)}. f b ( C ) = b − tr C b − 2 = ( b − 2 ) e ϕ b ( C ) .
The L ∞ L^\infty L ∞ bound supplied by the compact window and the Moser composition estimate give the displayed inequality.
The FENE-P free energy is
F b ( C ) = − ( b − 2 ) log ( 1 − tr C b ) − log det C . \mathcal{F}_b(C)=-(b-2)\log\left(1-\frac{\operatorname{tr} C}{b}\right)-\log\det C. F b ( C ) = − ( b − 2 ) log ( 1 − b tr C ) − log det C .
It contains both boundaries of D b \mathcal{D}_b D b : the logarithmic determinant sees loss of positive definiteness, while the first term sees the trace boundary. This zeroth-order information is important, but it is not the same as the pointwise barrier propagation and derivative control needed for continuation.
Proposition 8.1 (FENE-P free-energy identity). For smooth FENE-P solutions with C ∈ D b C \in D_b C ∈ D b ,
d d t [ 1 2 ∥ u ∥ L 2 2 + α 2 ∫ T 2 F b ( C ) d x ] + ν ∥ ∇ u ∥ L 2 2 + α 2 λ ∫ T 2 D b r e l ( C ) d x = 0 , (17) \frac{d}{dt}\left[\frac{1}{2}\lVert u\rVert_{L^2}^2+\frac{\alpha}{2}\int_{\mathbb{T}^2}F_b(C)\,dx\right]+\nu\lVert\nabla u\rVert_{L^2}^2+\frac{\alpha}{2\lambda}\int_{\mathbb{T}^2}D_b^{\mathrm{rel}}(C)\,dx=0, \tag*{(17)} d t d [ 2 1 ∥ u ∥ L 2 2 + 2 α ∫ T 2 F b ( C ) d x ] + ν ∥ ∇ u ∥ L 2 2 + 2 λ α ∫ T 2 D b rel ( C ) d x = 0 , (17) where D b r e l ( C ) ≥ 0 D_b^{\mathrm{rel}}(C)\geq0 D b rel ( C ) ≥ 0 is the FENE-P relaxation dissipation.
Proof. Multiplying the velocity equation by u u u gives the coupling term − α ∫ T b ( C ) : ∇ u d x -\alpha\int T_b(C):\nabla u\,dx − α ∫ T b ( C ) : ∇ u d x . The variational derivative of F b F_b F b is f b ( C ) I − C − 1 f_b(C)I-C^{-1} f b ( C ) I − C − 1 . Pairing this derivative with the conformation equation gives the stretching contribution 2 ∫ T b ( C ) : ∇ u d x 2\int T_b(C):\nabla u\,dx 2 ∫ T b ( C ) : ∇ u d x . After multiplication by α / 2 \alpha/2 α /2 , this cancels the velocity coupling. The remaining relaxation contribution is nonnegative by convexity of F b F_b F b on D b D_b D b .
The new differential structure is the trace-gap equation. Taking the trace of (5) gives
D t tr C = 2 C : ∇ u − λ − 1 ( f b ( C ) tr C − 2 ) , D t = ∂ t + u ⋅ ∇ . D_t\operatorname{tr}C=2C:\nabla u-\lambda^{-1}\left(f_b(C)\operatorname{tr}C-2\right),\qquad D_t=\partial_t+u\cdot\nabla. D t tr C = 2 C : ∇ u − λ − 1 ( f b ( C ) tr C − 2 ) , D t = ∂ t + u ⋅ ∇.
Writing g = b − tr C g=b-\operatorname{tr}C g = b − tr C , we obtain
D t g = − 2 C : ∇ u + λ − 1 ( b − 2 g tr C − 2 ) . (18) D_tg=-2C:\nabla u+\lambda^{-1}\left(\frac{b-2}{g}\operatorname{tr}C-2\right). \tag*{(18)} D t g = − 2 C : ∇ u + λ − 1 ( g b − 2 tr C − 2 ) . (18)
Equivalently,
D t ϕ b = 2 C : ∇ u g − λ − 1 ( b − 2 ) tr C g 2 + 2 λ − 1 1 g . (19) D_t\phi_b=\frac{2C:\nabla u}{g}-\lambda^{-1}\frac{(b-2)\operatorname{tr}C}{g^2}+2\lambda^{-1}\frac{1}{g}. \tag*{(19)} D t ϕ b = g 2 C : ∇ u − λ − 1 g 2 ( b − 2 ) tr C + 2 λ − 1 g 1 . (19)
The negative g − 2 g^{-2} g − 2 term is the finite-extensibility restoring force. It propagates the trace gap under a squared endpoint velocity clock.
Lemma 8.3 (Finite-extensibility clock-exponent law). Let r = tr C < b r=\operatorname{tr}C<b r = tr C < b , C ∈ S + + 2 C\in\mathbb{S}_{++}^2 C ∈ S ++ 2 , and suppose that along a smooth Lagrangian trajectory
D t r = 2 C : ∇ u − λ − 1 H ( r ) . D_tr=2C:\nabla u-\lambda^{-1}H(r). D t r = 2 C : ∇ u − λ − 1 H ( r ) . Assume that for some θ > 0 \theta>0 θ > 0 , c 0 > 0 c_0>0 c 0 > 0 , and c 1 ≥ 0 c_1\geq0 c 1 ≥ 0 ,
H ( r ) ≥ c 0 ( b − r ) − θ − c 1 H(r)\geq c_0(b-r)^{-\theta}-c_1 H ( r ) ≥ c 0 ( b − r ) − θ − c 1 on the trace range reached by the solution. Put
p θ = 1 + θ θ . p_\theta=\frac{1+\theta}{\theta}. p θ = θ 1 + θ . If b − r ( 0 , x ) ≥ η 0 > 0 b-r(0,x)\geq\eta_0>0 b − r ( 0 , x ) ≥ η 0 > 0 and
∫ 0 T ∥ ∇ u ( t ) ∥ B ∞ , 1 0 p θ d t < ∞ , \int_0^T\lVert\nabla u(t)\rVert_{B_{\infty,1}^{0}}^{p_\theta}\,dt<\infty, ∫ 0 T ∥ ∇ u ( t ) ∥ B ∞ , 1 0 p θ d t < ∞ , then there exists η > 0 \eta>0 η > 0 , depending only on the displayed quantities and the initial trace gap, such that
b − r ( t , x ) ≥ η on [ 0 , T ) × T 2 . b-r(t,x)\geq\eta\qquad\text{on }[0,T)\times\mathbb{T}^2. b − r ( t , x ) ≥ η on [ 0 , T ) × T 2 .
Proof. Let Φ g = − log ( b − r ) \Phi_g=-\log(b-r) Φ g = − log ( b − r ) , z = e Φ g z=e^{\Phi_g} z = e Φ g , and L ( t ) = ∥ ∇ u ( t ) ∥ L ∞ L(t)=\lVert\nabla u(t)\rVert_{L^\infty} L ( t ) = ∥ ∇ u ( t ) ∥ L ∞ . Since C > 0 C>0 C > 0 and r < b r<b r < b , ∣ C : ∇ u ∣ = ∣ C : S ( u ) ∣ ≤ b L ( t ) |C:\nabla u|=|C:S(u)|\leq bL(t) ∣ C : ∇ u ∣ = ∣ C : S ( u ) ∣ ≤ b L ( t ) . Therefore
D t Φ g = 2 C : ∇ u − λ − 1 H ( r ) b − r ≤ C ( 1 + L ( t ) ) z − c z 1 + θ . D_t\Phi_g=\frac{2C:\nabla u-\lambda^{-1}H(r)}{b-r}\leq C(1+L(t))z-cz^{1+\theta}. D t Φ g = b − r 2 C : ∇ u − λ − 1 H ( r ) ≤ C ( 1 + L ( t )) z − c z 1 + θ .
The elementary inequality
A z − c z 1 + θ ≤ C θ , c ( 1 + A p θ ) , A , z ≥ 0 , Az-cz^{1+\theta}\leq C_{\theta,c}(1+A^{p_\theta}),\qquad A,z\geq0, A z − c z 1 + θ ≤ C θ , c ( 1 + A p θ ) , A , z ≥ 0 ,
follows by maximizing the left-hand side in z z z , with the maximum occurring at scale z ∼ A 1 / θ z \sim A^{1/\theta} z ∼ A 1/ θ . Hence
D t Φ g ≤ C ( 1 + L ( t ) p θ ) ≤ C ( 1 + ∥ ∇ u ( t ) ∥ B ∞ , 1 0 p θ ) . D_t\Phi_g \le C(1+L(t)^{p_\theta}) \le C\left(1+\|\nabla u(t)\|_{B^0_{\infty,1}}^{p_\theta}\right). D t Φ g ≤ C ( 1 + L ( t ) p θ ) ≤ C ( 1 + ∥∇ u ( t ) ∥ B ∞ , 1 0 p θ ) .
Integrating along Lagrangian trajectories gives a uniform upper bound for Φ g \Phi_g Φ g , which is equivalent to a positive lower bound for b − r b-r b − r .
For FENE-P,
H b ( r ) = f b ( r ) r − 2 = ( b − 2 ) r b − r − 2 = b ( b − 2 ) b − r − b . H_b(r)=f_b(r)r-2=\frac{(b-2)r}{b-r}-2=\frac{b(b-2)}{b-r}-b. H b ( r ) = f b ( r ) r − 2 = b − r ( b − 2 ) r − 2 = b − r b ( b − 2 ) − b .
Thus θ = 1 \theta=1 θ = 1 in Lemma 8.3 , and the algebraic clock exponent is p θ = 2 p_\theta=2 p θ = 2 .
Lemma 8.4 (Propagation of compact FENE windows). Let ( u , C ) (u,C) ( u , C ) be a smooth FENE-P solution on [ 0 , T ) [0,T) [ 0 , T ) , and assume
∫ 0 T ∥ ∇ u ( t ) ∥ B ∞ , 1 0 2 d t < ∞ . \int_0^T \|\nabla u(t)\|_{B^0_{\infty,1}}^2\,dt<\infty. ∫ 0 T ∥∇ u ( t ) ∥ B ∞ , 1 0 2 d t < ∞. If the initial range of C 0 C_0 C 0 is contained in a compact subset of D b \mathcal{D}_b D b , then C ( t , ⋅ ) C(t,\cdot) C ( t , ⋅ ) remains in a compact FENE window on [ 0 , T ) [0,T) [ 0 , T ) .
Proof. Let L ( t ) = ∥ ∇ u ( t ) ∥ L ∞ L(t)=\|\nabla u(t)\|_{L^\infty} L ( t ) = ∥∇ u ( t ) ∥ L ∞ . Since C > 0 C>0 C > 0 and tr C < b \operatorname{tr}C<b tr C < b , ∣ C : S ( u ) ∣ ≤ b L ( t ) |C:S(u)|\le bL(t) ∣ C : S ( u ) ∣ ≤ b L ( t ) . From (19) , with g = e − ϕ b g=e^{-\phi_b} g = e − ϕ b and tr C = b − g \operatorname{tr}C=b-g tr C = b − g , we get
D t ϕ b ≤ ( 2 b L ( t ) + b λ − 1 ) e ϕ b − b ( b − 2 ) λ − 1 e 2 ϕ b . D_t\phi_b\le(2bL(t)+b\lambda^{-1})e^{\phi_b}-b(b-2)\lambda^{-1}e^{2\phi_b}. D t ϕ b ≤ ( 2 b L ( t ) + b λ − 1 ) e ϕ b − b ( b − 2 ) λ − 1 e 2 ϕ b .
Completing the square in e ϕ b e^{\phi_b} e ϕ b gives
D t ϕ b ≤ C b , λ ( 1 + L ( t ) 2 ) . D_t\phi_b\le C_{b,\lambda}(1+L(t)^2). D t ϕ b ≤ C b , λ ( 1 + L ( t ) 2 ) .
Taking the supremum along Lagrangian trajectories and using L ( t ) ≤ C ∥ ∇ u ( t ) ∥ B ∞ , 1 0 L(t)\le C\|\nabla u(t)\|_{B^0_{\infty,1}} L ( t ) ≤ C ∥∇ u ( t ) ∥ B ∞ , 1 0 propagates a positive lower bound for b − tr C b-\operatorname{tr}C b − tr C .
The trace bound gives C ≤ b I C\le bI C ≤ b I . The smallest eigenvalue m ( t ) m(t) m ( t ) satisfies the Dini inequality
m ˙ ( t ) ≥ − ( 2 L ( t ) + λ − 1 ∥ f b ( C ( t ) ) ∥ L ∞ ) m ( t ) . \dot m(t)\ge-\left(2L(t)+\lambda^{-1}\|f_b(C(t))\|_{L^\infty}\right)m(t). m ˙ ( t ) ≥ − ( 2 L ( t ) + λ − 1 ∥ f b ( C ( t )) ∥ L ∞ ) m ( t ) .
The propagated trace gap bounds f b ( C ) f_b(C) f b ( C ) in L ∞ L^\infty L ∞ , and ∫ 0 T L ( t ) d t < ∞ \int_0^T L(t)\,dt<\infty ∫ 0 T L ( t ) d t < ∞ follows from the squared clock. Gronwall’s inequality gives a positive lower spectral bound. Hence C C C stays in a compact subset of D b \mathcal{D}_b D b . □
We now use physical FENE-P variables
C = a I + Y , a = 1 2 tr C , Y = C ∘ . C=aI+Y,\qquad a=\frac{1}{2}\operatorname{tr}C,\qquad Y=C^\circ. C = a I + Y , a = 2 1 tr C , Y = C ∘ .
Since
T b ( C ) = f b ( C ) C − I = ( f b a − 1 ) I + f b Y , T_b(C)=f_b(C)C-I=(f_ba-1)I+f_bY, T b ( C ) = f b ( C ) C − I = ( f b a − 1 ) I + f b Y ,
the isotropic part is absorbed into the pressure. With f = f b ( C ) = ( b − 2 ) / ( b − 2 a ) f=f_b(C)=(b-2)/(b-2a) f = f b ( C ) = ( b − 2 ) / ( b − 2 a ) , the equations become
∂ t u + u ⋅ ∇ u − ν Δ u + ∇ π = α div ( f Y ) (20) \partial_t u+u\cdot\nabla u-\nu\Delta u+\nabla\pi=\alpha\operatorname{div}(fY) \tag*{(20)} ∂ t u + u ⋅ ∇ u − ν Δ u + ∇ π = α div ( f Y ) (20)
D t a + λ − 1 ( a f − 1 ) = S ( u ) : Y (21) D_ta+\lambda^{-1}(af-1)=S(u):Y \tag*{(21)} D t a + λ − 1 ( a f − 1 ) = S ( u ) : Y (21)
D t Y + λ − 1 f Y = 2 a S ( u ) + [ ∇ u Y + Y ( ∇ u ) T ] ∘ (22) D_tY+\lambda^{-1}fY=2aS(u)+[\nabla u\,Y+Y(\nabla u)^{T}]^\circ \tag*{(22)} D t Y + λ − 1 f Y = 2 a S ( u ) + [ ∇ u Y + Y ( ∇ u ) T ] ∘ (22)
Lemma 8.5 (FENE-P coefficient commutators). Let K ⋐ D b K\Subset D_b K ⋐ D b , m ≥ 3 m\geq3 m ≥ 3 , and 0 < ε < 1 0<\varepsilon<1 0 < ε < 1 . On K K K , for ∣ β ∣ ≤ m |\beta|\leq m ∣ β ∣ ≤ m ,
∥ ∂ β ( f Y ) − f ∂ β Y ∥ L 2 + ∥ [ ∂ β , f ] Y ∥ L 2 ≤ C K ( 1 + ∥ Log C ∥ H 1 + ε ) ( ∥ a − 1 ∥ H m + ∥ Y ∥ H m ) . \left\|\partial^\beta(fY)-f\partial^\beta Y\right\|_{L^2}+\left\|[\partial^\beta,f]Y\right\|_{L^2}\leq C_K\left(1+\|\operatorname{Log}C\|_{H^{1+\varepsilon}}\right)\left(\|a-1\|_{H^m}+\|Y\|_{H^m}\right). ∂ β ( f Y ) − f ∂ β Y L 2 + [ ∂ β , f ] Y L 2 ≤ C K ( 1 + ∥ Log C ∥ H 1 + ε ) ( ∥ a − 1 ∥ H m + ∥ Y ∥ H m ) .
Proof. On K K K , f = ( b − 2 ) / ( b − 2 a ) f=(b-2)/(b-2a) f = ( b − 2 ) / ( b − 2 a ) , a a a , and Y Y Y are smooth functions of Log C \operatorname{Log}C Log C . The Sobolev composition theorem gives
∥ a ∥ H 1 + ε + ∥ Y ∥ H 1 + ε + ∥ f ∥ H 1 + ε ≤ C K ( 1 + ∥ Log C ∥ H 1 + ε ) . \|a\|_{H^{1+\varepsilon}}+\|Y\|_{H^{1+\varepsilon}}+\|f\|_{H^{1+\varepsilon}}\leq C_K\left(1+\|\operatorname{Log}C\|_{H^{1+\varepsilon}}\right). ∥ a ∥ H 1 + ε + ∥ Y ∥ H 1 + ε + ∥ f ∥ H 1 + ε ≤ C K ( 1 + ∥ Log C ∥ H 1 + ε ) .
For positive derivatives, the tame composition estimate also gives
∥ ∇ f ∥ H m − 1 ≤ C K ( 1 + ∥ Log C ∥ H 1 + ε ) ∥ a − 1 ∥ H m , \|\nabla f\|_{H^{m-1}}\leq C_K\left(1+\|\operatorname{Log}C\|_{H^{1+\varepsilon}}\right)\|a-1\|_{H^m}, ∥∇ f ∥ H m − 1 ≤ C K ( 1 + ∥ Log C ∥ H 1 + ε ) ∥ a − 1 ∥ H m ,
because every positive derivative of f ( a ) f(a) f ( a ) contains a positive derivative of a a a , and ∂ γ a = ∂ γ ( a − 1 ) \partial^\gamma a=\partial^\gamma(a-1) ∂ γ a = ∂ γ ( a − 1 ) for ∣ γ ∣ > 0 |\gamma|>0 ∣ γ ∣ > 0 . Expanding
∂ β ( f Y ) − f ∂ β Y = ∑ 0 < γ ≤ β c β , γ ∂ γ f ∂ β − γ Y \partial^\beta(fY)-f\partial^\beta Y=\sum_{0<\gamma\leq\beta}c_{\beta,\gamma}\partial^\gamma f\,\partial^{\beta-\gamma}Y ∂ β ( f Y ) − f ∂ β Y = 0 < γ ≤ β ∑ c β , γ ∂ γ f ∂ β − γ Y
and applying the tame product estimate gives the first bound: one factor carries the high norm ∥ a − 1 ∥ H m \|a-1\|_{H^m} ∥ a − 1 ∥ H m or ∥ Y ∥ H m \|Y\|_{H^m} ∥ Y ∥ H m , while all remaining factors are placed in H 1 + ε H^{1+\varepsilon} H 1 + ε or L ∞ L^\infty L ∞ . No additive constant is produced, since for β = 0 \beta=0 β = 0 the left-hand side vanishes and for ∣ β ∣ > 0 |\beta|>0 ∣ β ∣ > 0 each term contains a positive derivative. The commutator [ ∂ β , f ] Y [\partial^\beta,f]Y [ ∂ β , f ] Y has the same expansion. □
Proposition 8.2 (Pressure-free FENE-P endpoint estimate). Let m ≥ 3 m\geq3 m ≥ 3 , 0 < ε < 1 0<\varepsilon<1 0 < ε < 1 , and let ( u , C ) (u,C) ( u , C ) be a smooth FENE-P solution on [ 0 , T ] [0,T] [ 0 , T ] whose conformation tensor remains in a compact FENE window K K K . Set C = a I + Y C=aI+Y C = a I + Y , a = 1 2 tr C a=\frac{1}{2}\operatorname{tr}C a = 2 1 tr C , Y = C ∘ Y=C^\circ Y = C ∘ , and f = f b ( C ) f=f_b(C) f = f b ( C ) . Define
E m F ( t ) = ∥ u ∥ H m 2 + ∥ a − 1 ∥ H m 2 + α 4 ∑ ∣ β ∣ ≤ m ∫ T 2 f a ∣ ∂ β Y ∣ 2 d x . \mathcal{E}^{F}_m(t)=\|u\|_{H^m}^2+\|a-1\|_{H^m}^2+\frac{\alpha}{4}\sum_{|\beta|\leq m}\int_{\mathbb{T}^2}\frac{f}{a}|\partial^\beta Y|^2\,dx. E m F ( t ) = ∥ u ∥ H m 2 + ∥ a − 1 ∥ H m 2 + 4 α ∣ β ∣ ≤ m ∑ ∫ T 2 a f ∣ ∂ β Y ∣ 2 d x . Then
d d t E m F + c ν ∥ u ∥ H m + 1 2 + c λ − 1 ∥ Y ∥ H m 2 ≤ C K ( 1 + ∥ ∇ u ∥ B 0 , 1 ∞ + ∥ Log C ∥ H 1 + ε ) E m F . \frac{d}{dt}\mathcal{E}^{F}_m+c\nu\|u\|_{H^{m+1}}^2+c\lambda^{-1}\|Y\|_{H^m}^2
\leq C_K\left(1+\|\nabla u\|_{B^\infty_{0,1}}+\|\operatorname{Log}C\|_{H^{1+\varepsilon}}\right)\mathcal{E}^{F}_m. d t d E m F + c ν ∥ u ∥ H m + 1 2 + c λ − 1 ∥ Y ∥ H m 2 ≤ C K ( 1 + ∥∇ u ∥ B 0 , 1 ∞ + ∥ Log C ∥ H 1 + ε ) E m F . where C K C_K C K depends on the compact FENE window and the fixed parameters.
Proof. The compact FENE window makes E m F \mathcal{E}^{F}_m E m F equivalent to ∥ u ∥ H m 2 + ∥ a − 1 ∥ H m 2 + ∥ Y ∥ H m 2 \|u\|_{H^m}^2+\|a-1\|_{H^m}^2+\|Y\|_{H^m}^2 ∥ u ∥ H m 2 + ∥ a − 1 ∥ H m 2 + ∥ Y ∥ H m 2 . Apply ∂ β \partial^\beta ∂ β , ∣ β ∣ ≤ m |\beta|\leq m ∣ β ∣ ≤ m , to (20) and test by ∂ β u \partial^\beta u ∂ β u . The principal elastic contribution is
− α ∫ f ∂ β Y : S ( ∂ β u ) d x . -\alpha\int f\partial^\beta Y:S(\partial^\beta u)\,dx. − α ∫ f ∂ β Y : S ( ∂ β u ) d x . The difference between ∂ β ( f Y ) \partial^\beta(fY) ∂ β ( f Y ) and f ∂ β Y f\partial^\beta Y f ∂ β Y is bounded by Lemma 8.5 and Young’s inequality:
δ ν ∥ u ∥ H m + 1 2 + C δ , K ( 1 + ∥ Log C ∥ H 1 + ε 2 ) E m F . \delta\nu\|u\|_{H^{m+1}}^2+C_{\delta,K}\left(1+\|\operatorname{Log}C\|_{H^{1+\varepsilon}}^2\right)\mathcal{E}^{F}_m. δ ν ∥ u ∥ H m + 1 2 + C δ , K ( 1 + ∥ Log C ∥ H 1 + ε 2 ) E m F .
Next differentiate (22) and test by ( α / 2 ) ( f / a ) ∂ β Y (\alpha/2)(f/a)\partial^\beta Y ( α /2 ) ( f / a ) ∂ β Y . The term 2 α S ( ∂ β u ) 2\alpha S(\partial^\beta u) 2 α S ( ∂ β u ) produces
α ∫ f ∂ β Y : S ( ∂ β u ) d x , \alpha\int f\,\partial^\beta Y:S(\partial^\beta u)\,dx, α ∫ f ∂ β Y : S ( ∂ β u ) d x ,
which cancels the principal elastic term from the velocity equation. The relaxation term contributes
α 2 λ ∫ f 2 a ∣ ∂ β Y ∣ 2 d x , \frac{\alpha}{2\lambda}\int\frac{f^2}{a}\lvert\partial^\beta Y\rvert^2\,dx, 2 λ α ∫ a f 2 ∣ ∂ β Y ∣ 2 d x ,
which controls ∥ Y ∥ H m 2 \lVert Y\rVert_{H^m}^2 ∥ Y ∥ H m 2 on K K K . The time-dependent weight satisfies
D t ( f a ) = ( f a ) ′ ( a ) ( S ( u ) : Y − λ − 1 ( a f − 1 ) ) , D_t\left(\frac{f}{a}\right)=\left(\frac{f}{a}\right)'(a)\left(S(u):Y-\lambda^{-1}(af-1)\right), D t ( a f ) = ( a f ) ′ ( a ) ( S ( u ) : Y − λ − 1 ( a f − 1 ) ) ,
and hence
∣ D t ( f a ) ∣ ≤ C K ( 1 + ∥ ∇ u ∥ L ∞ ) . \left\lvert D_t\left(\frac{f}{a}\right)\right\rvert\leq C_K(1+\lVert\nabla u\rVert_{L^\infty}). D t ( a f ) ≤ C K ( 1 + ∥ ∇ u ∥ L ∞ ) .
The weighted transport commutator is controlled by Lemma 6.2, with w = f / a w=f/a w = f / a . The coefficient, stretching, and relaxation commutators are controlled by Lemmas 8.5 and 6.2. Their total contribution is bounded by
δ ν ∥ u ∥ H m + 1 2 + C δ , K ( 1 + ∥ ∇ u ∥ B ∞ , 1 0 + ∥ Log C ∥ H 1 + ε 2 ) E m F . \delta\nu\lVert u\rVert_{H^{m+1}}^2+C_{\delta,K}\left(1+\lVert\nabla u\rVert_{B^0_{\infty,1}}+\lVert\operatorname{Log}C\rVert_{H^{1+\varepsilon}}^2\right)\mathcal{E}_m^F. δ ν ∥ u ∥ H m + 1 2 + C δ , K ( 1 + ∥ ∇ u ∥ B ∞ , 1 0 + ∥ Log C ∥ H 1 + ε 2 ) E m F .
Finally, apply ∂ β \partial^\beta ∂ β to (21) . Since
a f ( a ) − 1 = 0 at a = 1 , d d a ( a f ( a ) − 1 ) = b ( b − 2 ) ( b − 2 a ) 2 > 0 af(a)-1=0\quad\text{at }a=1,\qquad
\frac{d}{da}(af(a)-1)=\frac{b(b-2)}{(b-2a)^2}>0 a f ( a ) − 1 = 0 at a = 1 , d a d ( a f ( a ) − 1 ) = ( b − 2 a ) 2 b ( b − 2 ) > 0
on K K K , the relaxation part is coercive for a − 1 a-1 a − 1 , up to lower-order commutators. The source ∂ β ( S ( u ) : Y ) \partial^\beta(S(u):Y) ∂ β ( S ( u ) : Y ) is estimated by placing the top derivative of u u u in the viscous dissipation and all remaining factors in H 1 + ε H^{1+\varepsilon} H 1 + ε or L ∞ L^\infty L ∞ . Thus
d d t ∥ a − 1 ∥ H m 2 ≤ δ ν ∥ u ∥ H m + 1 2 + C δ , K ( 1 + ∥ ∇ u ∥ B ∞ , 1 0 + ∥ Log C ∥ H 1 + ε 2 ) E m F . \frac{d}{dt}\lVert a-1\rVert_{H^m}^2\leq\delta\nu\lVert u\rVert_{H^{m+1}}^2+C_{\delta,K}\left(1+\lVert\nabla u\rVert_{B^0_{\infty,1}}+\lVert\operatorname{Log}C\rVert_{H^{1+\varepsilon}}^2\right)\mathcal{E}_m^F. d t d ∥ a − 1 ∥ H m 2 ≤ δ ν ∥ u ∥ H m + 1 2 + C δ , K ( 1 + ∥ ∇ u ∥ B ∞ , 1 0 + ∥ Log C ∥ H 1 + ε 2 ) E m F .
Choosing δ \delta δ sufficiently small and summing over ∣ β ∣ ≤ m \lvert\beta\rvert\leq m ∣ β ∣ ≤ m proves (8.2).
Theorem 8.1 (Unified geometric continuation criterion). Let s ∈ N s\in\mathbb{N} s ∈ N , s ≥ 3 s\geq3 s ≥ 3 , and 0 < ε < 1 0<\varepsilon<1 0 < ε < 1 .
For Oldroyd–B, let ( u , A ) (u,A) ( u , A ) be a strong positive-cone solution on [ 0 , T ) [0,T) [ 0 , T ) . If
∫ 0 T ∥ ∇ u ( t ) ∥ B ∞ , 1 0 0 d t < ∞ , \int_0^T\lVert\nabla u(t)\rVert_{B^0_{\infty,1}}^0\,dt<\infty, ∫ 0 T ∥ ∇ u ( t ) ∥ B ∞ , 1 0 0 d t < ∞ , then A A A automatically remains in a compact spectral window and the strong solution continues beyond T T T .
For FENE-P, fix b > 2 b>2 b > 2 and let ( u , C ) (u,C) ( u , C ) be a strong FENE-P solution on [ 0 , T ) [0,T) [ 0 , T ) , so that C ( t , x ) ∈ D b C(t,x)\in\mathcal{D}_b C ( t , x ) ∈ D b . If
∫ 0 T ∥ ∇ u ( t ) ∥ B ∞ , 1 0 2 d t < ∞ , \int_0^T\lVert\nabla u(t)\rVert_{B^0_{\infty,1}}^2\,dt<\infty, ∫ 0 T ∥ ∇ u ( t ) ∥ B ∞ , 1 0 2 d t < ∞ , then the FENE-P strong solution continues beyond T T T .
Proof. The Oldroyd–B statement is Theorem 6.1. For FENE-P, the squared endpoint velocity clock propagates a compact FENE window by Lemma 8.4. Since T < ∞ T<\infty T < ∞ , this squared clock also gives ∇ u ∈ L 1 ( 0 , T ; B ∞ , 1 0 ) \nabla u\in L^1(0,T;B^0_{\infty,1}) ∇ u ∈ L 1 ( 0 , T ; B ∞ , 1 0 ) . On the compact FENE window, Proposition 6.5 gives Log C ∈ L 2 ( 0 , T ; H 1 + ε ) \operatorname{Log}C\in L^2(0,T;H^{1+\varepsilon}) Log C ∈ L 2 ( 0 , T ; H 1 + ε ) .
Proposition 8.2 gives
d d t ξ s F ≤ C K ( 1 + ∥ ∇ u ∥ B ∞ , 1 0 + ∥ Log C ∥ H 1 + ε 2 ) ξ s F . \frac{d}{dt}\xi_s^F \le C_K\left(1+\|\nabla u\|_{B^0_{\infty,1}}+\|\operatorname{Log} C\|_{H^{1+\varepsilon}}^2\right)\xi_s^F. d t d ξ s F ≤ C K ( 1 + ∥∇ u ∥ B ∞ , 1 0 + ∥ Log C ∥ H 1 + ε 2 ) ξ s F .
The coefficient is integrable by the squared velocity clock, the finite time interval, and the logarithmic bound. Gronwall’s inequality gives a uniform H s H^s H s bound for ( u , a − 1 , Y ) (u,a-1,Y) ( u , a − 1 , Y ) . On the compact FENE window, this is equivalent to an H s H^s H s bound for C C C and T b ( C ) T_b(C) T b ( C ) . Proposition 2.2 then restarts the solution and extends it beyond T T T .
Corollary 8.1 (Two-dimensional endpoint breakdown alternatives). At the first breakdown time of a strong two-dimensional stress-diffusion-free solution, the Oldroyd–B model can fail only through the endpoint flow-map channel:
∫ 0 T ∗ ∥ ∇ u ( t ) ∥ B ∞ , 1 0 d t = ∞ . \int_0^{T_*}\|\nabla u(t)\|_{B^0_{\infty,1}}\,dt=\infty. ∫ 0 T ∗ ∥∇ u ( t ) ∥ B ∞ , 1 0 d t = ∞. For FENE-P, finite-time breakdown forces loss of the squared endpoint velocity clock:
∫ 0 T ∗ ∥ ∇ u ( t ) ∥ B ∞ , 1 0 2 d t = ∞ . \int_0^{T_*}\|\nabla u(t)\|_{B^0_{\infty,1}}^2\,dt=\infty. ∫ 0 T ∗ ∥∇ u ( t ) ∥ B ∞ , 1 0 2 d t = ∞. In particular, positive-cone loss, trace-gap collapse, and logarithmic high-frequency concentration cannot occur while the relevant velocity clock stays finite.
Proof. If none of the listed quantities diverges, then the relevant continuation criterion in Theorem 8.1 applies and extends the solution beyond T ∗ T_* T ∗ , a contradiction.
Three-dimensional residual criteria and structural boundary
The two-dimensional spectral quotient used above is sharp because the smooth isotropic representation, together with the two-dimensional Cayley–Hamilton identity, leaves only one deviatoric active channel after quotienting by pressure. In three dimensions a smooth local Cayley–Hamilton representation has one additional pressure-free direction. This is the basic structural boundary between the two-dimensional theorem and the three-dimensional statements below. The aim of this section is therefore not to force a false three-dimensional analogue of the scalar quotient closure, but to separate the additional direction explicitly, estimate it as a residual on compact conformation windows, and then apply the resulting pressure-free mechanism to Oldroyd–B and FENE-P, where the residual coefficient is in fact zero. The section is deliberately not a general three-dimensional theory of anisotropic or non-spectral tensor stresses.
Throughout this section the domain is T 3 \mathbb{T}^3 T 3 , m ≥ 4 m \ge4 m ≥ 4 , and
B ( t ) = ∥ ∇ u ( t ) ∥ B ∞ , 1 0 . B(t)=\lVert\nabla u(t)\rVert_{B^{0}_{\infty,1}}. B ( t ) = ∥ ∇ u ( t ) ∥ B ∞ , 1 0 .
The index m ≥ 4 m \ge4 m ≥ 4 is part of the strong-solution framework in three dimensions: it gives the algebra and compact-window Moser bounds needed for the residual channel and for classical spectral barriers. No weak or Leray-type three-dimensional theory is inferred from these estimates. The high-order argument below uses a two-tier closure: the endpoint clock first controls an H 3 H^3 H 3 compact-window energy, and this lower tier supplies the W 1 , ∞ W^{1,\infty} W 1 , ∞ coefficient bounds needed in the H m H^m H m , m ≥ 4 m \ge4 m ≥ 4 , restart estimate. For a generic three-dimensional tensor we write
a = 1 3 tr Z , h = a − 1 , Y = Z − a I . a=\frac{1}{3}\operatorname{tr}Z,\qquad h=a-1,\qquad Y=Z-aI. a = 3 1 tr Z , h = a − 1 , Y = Z − a I .
The letter Z Z Z is reserved here for model-free spectral statements. In the Oldroyd–B specialization Z = A Z=A Z = A , while in the FENE-P specialization Z = C Z=C Z = C . For three-dimensional FENE-P, throughout this section
f b ( C ) = f b , 3 ( C ) = b − 3 b − tr C , b > 3. f_b(C)=f_{b,3}(C)=\frac{b-3}{b-\operatorname{tr}C},\qquad b>3. f b ( C ) = f b , 3 ( C ) = b − tr C b − 3 , b > 3.
We also decompose
∇ u = S + Ω , S = 1 2 ( ∇ u + ∇ u T ) , Ω = 1 2 ( ∇ u − ∇ u T ) . \nabla u=S+\Omega,\qquad S=\frac{1}{2}\left(\nabla u+\nabla u^T\right),\qquad\Omega=\frac{1}{2}\left(\nabla u-\nabla u^T\right). ∇ u = S + Ω , S = 2 1 ( ∇ u + ∇ u T ) , Ω = 2 1 ( ∇ u − ∇ u T ) .
We also set
R 2 ( Y ) = ( Y 2 ) ∘ = Y 2 − 1 3 tr ( Y 2 ) I . R_2(Y)=(Y^2)^\circ=Y^2-\frac{1}{3}\operatorname{tr}(Y^2)I. R 2 ( Y ) = ( Y 2 ) ∘ = Y 2 − 3 1 tr ( Y 2 ) I .
Proposition 9.1 (Algebraic boundary of the three-dimensional quotient). The map Y ↦ R 2 ( Y ) = ( Y 2 ) ∘ Y\mapsto R_2(Y)=(Y^2)^\circ Y ↦ R 2 ( Y ) = ( Y 2 ) ∘ is not reducible, in S 0 3 \mathbb{S}^3_0 S 0 3 , to a scalar multiple of Y Y Y . More precisely, if
Y ∗ = diag ( 1 , 0 , − 1 ) , Y_*=\operatorname{diag}(1,0,-1), Y ∗ = diag ( 1 , 0 , − 1 ) , then
R 2 ( Y ∗ ) = diag ( 1 3 , − 2 3 , 1 3 ) ∉ span { Y ∗ } . R_2(Y_*)=\operatorname{diag}\left(\frac{1}{3},-\frac{2}{3},\frac{1}{3}\right)\notin\operatorname{span}\{Y_*\}. R 2 ( Y ∗ ) = diag ( 3 1 , − 3 2 , 3 1 ) ∈ / span { Y ∗ } . Consequently, there is no neighbourhood of Y ∗ Y_* Y ∗ on which R 2 ( Y ) = λ ( Y ) Y R_2(Y)=\lambda(Y)Y R 2 ( Y ) = λ ( Y ) Y for a scalar function λ \lambda λ . Thus a three-dimensional spectral stress with a nonzero q 2 q_2 q 2 -channel cannot, in general, be reduced after quotienting by pressure to a single active scalar multiple of Y Y Y . The remaining possibilities are either a residual estimate on a compact conformation window or an additional model-specific mechanism that removes or controls this channel.
Proof. For Y ∗ Y_* Y ∗ as above, tr Y ∗ = 0 \operatorname{tr}Y_*=0 tr Y ∗ = 0 and
Y ∗ 2 = diag ( 1 , 0 , 1 ) , tr ( Y ∗ 2 ) = 2. Y_*^2=\operatorname{diag}(1,0,1),\qquad\operatorname{tr}(Y_*^2)=2. Y ∗ 2 = diag ( 1 , 0 , 1 ) , tr ( Y ∗ 2 ) = 2.
Hence
( Y ∗ 2 ) ∘ = Y ∗ 2 − 2 3 I = diag ( 1 3 , − 2 3 , 1 3 ) . (Y_*^2)^\circ=Y_*^2-\frac{2}{3}I=\operatorname{diag}\left(\frac{1}{3},-\frac{2}{3},\frac{1}{3}\right). ( Y ∗ 2 ) ∘ = Y ∗ 2 − 3 2 I = diag ( 3 1 , − 3 2 , 3 1 ) .
This matrix cannot be λ Y ∗ \lambda Y_* λ Y ∗ , since the middle diagonal entry of λ Y ∗ \lambda Y_* λ Y ∗ is zero for every λ \lambda λ , whereas the middle diagonal entry of R 2 ( Y ∗ ) R_2(Y_*) R 2 ( Y ∗ ) is − 2 / 3 -2/3 − 2/3 . This proves the non-reducibility at Y ∗ Y_* Y ∗ , and a neighbourhood representation would in particular hold at Y ∗ Y_* Y ∗ , giving the same contradiction. The final assertion follows from the split (24) .
Lemma 9.1 (Three-dimensional pressure quotient with Cayley–Hamilton residual). Let K ⋐ S + + 3 K \Subset\mathbb{S}_{++}^{3} K ⋐ S ++ 3 be a compact spectral window and let T T T be a spectral isotropic stress law defined on a neighbourhood of K K K . Assume that, on K K K , T T T admits a smooth local Cayley–Hamilton representation
T ( Z ) = q 0 ( Z ) I + q 1 ( Z ) Z + q 2 ( Z ) Z 2 , (23) T(Z)=q_0(Z)I+q_1(Z)Z+q_2(Z)Z^2, \tag*{(23)} T ( Z ) = q 0 ( Z ) I + q 1 ( Z ) Z + q 2 ( Z ) Z 2 , (23) where q 0 , q 1 , q 2 q_0,q_1,q_2 q 0 , q 1 , q 2 are smooth spectral scalar functions. If Z = a I + Y Z=aI+Y Z = a I + Y , a = 1 3 tr Z a=\frac{1}{3}\operatorname{tr}Z a = 3 1 tr Z , tr Y = 0 \operatorname{tr}Y=0 tr Y = 0 , then its pressure-free part is
T ( Z ) ∘ = τ 1 ( Z ) Y + τ 2 ( Z ) R 2 ( Y ) , τ 1 = q 1 + 2 a q 2 , τ 2 = q 2 . (24) T(Z)^\circ=\tau_1(Z)Y+\tau_2(Z)R_2(Y),\qquad\tau_1=q_1+2aq_2,\qquad\tau_2=q_2. \tag*{(24)} T ( Z ) ∘ = τ 1 ( Z ) Y + τ 2 ( Z ) R 2 ( Y ) , τ 1 = q 1 + 2 a q 2 , τ 2 = q 2 . (24) Thus, within the class (23) , the only three-dimensional pressure-free channel beyond the cancellative Y Y Y -direction is the Cayley–Hamilton residual R 2 ( Y ) R_2(Y) R 2 ( Y ) . For Oldroyd–B and FENE-P the representation is explicit and has q 2 ≡ 0 q_2\equiv0 q 2 ≡ 0 , so this residual vanishes.
Proof. The representation (23) is the hypothesis under which the general spectral statement is used. The only point requiring care is the smoothness of the scalar coefficients at multiple spectra. We do not use an eigenvalue labelling: for smooth isotropic tensor laws this is the standard local representation on compact spectral windows, while the lemma may also be read as a purely algebraic statement conditional on (23) . The two model stresses used below satisfy it directly: A − I = ( a − 1 ) I + A A-I=(a-1)I+A A − I = ( a − 1 ) I + A for Oldroyd–B and f b ( C ) C − I = ( f b a − 1 ) I + f b Y f_b(C)C-I=(f_ba-1)I+f_bY f b ( C ) C − I = ( f b a − 1 ) I + f b Y for FENE-P.
The rest is algebraic and does not use spectral diagonalization. Since
Z 2 = a 2 I + 2 a Y + Y 2 , Z^2=a^2I+2aY+Y^2, Z 2 = a 2 I + 2 aY + Y 2 , we have
( Z 2 ) ∘ = 2 a Y + ( Y 2 ) ∘ . (Z^2)^\circ=2aY+(Y^2)^\circ. ( Z 2 ) ∘ = 2 aY + ( Y 2 ) ∘ . Taking the trace-free part of q 0 I + q 1 Z + q 2 Z 2 q_0I+q_1Z+q_2Z^2 q 0 I + q 1 Z + q 2 Z 2 gives (24) . The last assertion follows from A − I = ( a − 1 ) I + Y A-I=(a-1)I+Y A − I = ( a − 1 ) I + Y for Oldroyd–B and f b ( C ) C − I = ( f b a − 1 ) I + f b Y f_b(C)C-I=(f_ba-1)I+f_bY f b ( C ) C − I = ( f b a − 1 ) I + f b Y for FENE-P. □ \square □
Lemma 9.2 (Compact-window estimate for the residual channel). Let K ⋐ S + + 3 K\Subset\mathbb{S}_{++}^{3} K ⋐ S ++ 3 be a compact conformation window and let τ 2 \tau_2 τ 2 be a smooth spectral scalar on a neighbourhood of K K K . Suppose Z ( t , x ) ∈ K Z(t,x)\in K Z ( t , x ) ∈ K on T 3 \mathbb{T}^3 T 3 , set Y = Z − 1 3 ( tr Z ) I Y=Z-\frac{1}{3}(\operatorname{tr}Z)I Y = Z − 3 1 ( tr Z ) I , and let m ≥ 4 m\ge4 m ≥ 4 . Then for every multi-index ∣ β ∣ ≤ m |\beta|\le m ∣ β ∣ ≤ m and every δ > 0 \delta>0 δ > 0 ,
∣ ⟨ ∂ β div ( τ 2 ( Z ) R 2 ( Y ) ) , ∂ β u ⟩ L 2 ( T 3 ) ∣ ≤ δ ∥ ∇ ∂ β u ∥ L 2 2 + C δ , K , m , τ 2 ( 1 + ∥ Z − I ∥ H m 2 ) . \left|\left\langle\partial^\beta\operatorname{div}\bigl(\tau_2(Z)R_2(Y)\bigr),\partial^\beta u\right\rangle_{L^2(\mathbb{T}^3)}\right|
\le\delta\left\|\nabla\partial^\beta u\right\|_{L^2}^{2}+C_{\delta,K,m,\tau_2}\left(1+\left\|Z-I\right\|_{H^m}^{2}\right). ⟨ ∂ β div ( τ 2 ( Z ) R 2 ( Y ) ) , ∂ β u ⟩ L 2 ( T 3 ) ≤ δ ∇ ∂ β u L 2 2 + C δ , K , m , τ 2 ( 1 + ∥ Z − I ∥ H m 2 ) .
Consequently, if the velocity equation has viscosity ν > 0 \nu> 0 ν > 0 , choosing δ \delta δ so that N m δ < ν / 4 N_m\delta< \nu/4 N m δ < ν /4 , where N m = # { β : ∣ β ∣ ≤ m } N_m = \#\{\beta: |\beta| \le m\} N m = # { β : ∣ β ∣ ≤ m } , allows the summed residual to be absorbed into the viscous velocity dissipation. It contributes only compact-window Gronwall terms and is not a second cancellation mechanism.
Proof. On the torus, integration by parts gives
⟨ ∂ β div ( τ 2 ( Z ) R 2 ( Y ) ) , ∂ β u ⟩ = − ⟨ ∂ β ( τ 2 ( Z ) R 2 ( Y ) ) , ∇ ∂ β u ⟩ . \left\langle\partial^\beta\operatorname{div}\left(\tau_2(Z)R_2(Y)\right),\partial^\beta u\right\rangle= -\left\langle\partial^\beta\left(\tau_2(Z)R_2(Y)\right),\nabla\partial^\beta u\right\rangle. ⟨ ∂ β div ( τ 2 ( Z ) R 2 ( Y ) ) , ∂ β u ⟩ = − ⟨ ∂ β ( τ 2 ( Z ) R 2 ( Y ) ) , ∇ ∂ β u ⟩ .
Since Z ∈ K Z \in K Z ∈ K pointwise, Y Y Y , R 2 ( Y ) R_2(Y) R 2 ( Y ) , τ 2 ( Z ) \tau_2(Z) τ 2 ( Z ) , and all derivatives of τ 2 \tau_2 τ 2 needed in the Moser calculus are uniformly bounded in L ∞ L^\infty L ∞ by constants depending only on K K K and τ 2 \tau_2 τ 2 . Because m ≥ 4 m \ge4 m ≥ 4 , H m ( T 3 ) H^m(\mathbb{T}^3) H m ( T 3 ) is an algebra and the compact-window Moser estimates imply
∥ ∂ β ( τ 2 ( Z ) R 2 ( Y ) ) ∥ L 2 ≤ C K , m , τ 2 ( 1 + ∥ Z − I ∥ H m ) . \left\|\partial^\beta\left(\tau_2(Z)R_2(Y)\right)\right\|_{L^2} \le C_{K,m,\tau_2}\left(1+\|Z-I\|_{H^m}\right). ∂ β ( τ 2 ( Z ) R 2 ( Y ) ) L 2 ≤ C K , m , τ 2 ( 1 + ∥ Z − I ∥ H m ) .
Cauchy’s inequality followed by Young’s inequality gives the displayed bound. The smallness condition on δ \delta δ relative to ν \nu ν and N m N_m N m is not part of the algebraic estimate; it is only the choice used when this term is inserted into the summed viscous high-order energy inequality. □
The compact-window velocity-clock results below do not use entropy variables in three dimensions. After the Oldroyd–B and FENE-P criteria are proved, we record one finite-dimensional calculation for a different purpose: it identifies when the three-dimensional FENE-P entropy variables are coercive enough to support a possible logarithmic route. The parameter q ∗ q_* q ∗ introduced there is therefore a spectral-variance boundary for the entropy-variable Jacobian, not a hidden assumption in the pressure-free velocity-clock argument.
The trace-free stretching identity is
∇ u Y + Y ∇ u T + 2 a S − 2 3 ( Y : S ) I = 2 a S + ( S Y + Y S − 2 3 ( Y : S ) I ) + ( Ω Y − Y Ω ) . \nabla u\,Y+Y\nabla u^{T}+2aS-\frac{2}{3}(Y:S)I=2aS+\left(SY+YS-\frac{2}{3}(Y:S)I\right)+(\Omega Y-Y\Omega). ∇ u Y + Y ∇ u T + 2 a S − 3 2 ( Y : S ) I = 2 a S + ( S Y + Y S − 3 2 ( Y : S ) I ) + ( Ω Y − Y Ω ) .
The first term is the active stretching block. The second is passive and is absorbed by viscosity. The third is objective rotation and is skew in the principal dyadic energy. Lemma 9.1 shows that the stress side has the same structure: a cancellative active channel and, in the general spectral case, the residual channel estimated in Lemma 9.2.
Proposition 9.2 (Three-dimensional compact-window pressure-free criterion). Let ( u , C ) (u,C) ( u , C ) be a strong three-dimensional Oldroyd–B or FENE-P solution on [ 0 , T ) [0,T) [ 0 , T ) , with ( u 0 , C 0 − I ) ∈ H m (u_0,C_0-I) \in H^m ( u 0 , C 0 − I ) ∈ H m , m ≥ 4 m \ge4 m ≥ 4 . In this proposition, C C C is the common conformation variable; in the Oldroyd–B specialization it should be read as the tensor denoted A A A in the two-dimensional model sections. Assume that C C C remains in a fixed compact conformation window:
κ I ≤ C ( t , x ) ≤ K I for Oldroyd–B, \kappa I \le C(t,x) \le KI \qquad\text{for Oldroyd--B,} κ I ≤ C ( t , x ) ≤ K I for Oldroyd–B, or
κ I ≤ C ( t , x ) , tr C ( t , x ) ≤ b − κ for FENE-P. \kappa I \le C(t,x),\qquad\operatorname{tr} C(t,x) \le b-\kappa\qquad\text{for FENE-P.} κ I ≤ C ( t , x ) , tr C ( t , x ) ≤ b − κ for FENE-P. If
∫ 0 T B ( t ) d t < ∞ , \int_0^T B(t)\,\mathrm{d}t < \infty, ∫ 0 T B ( t ) d t < ∞ , then
sup 0 ≤ t < T ( ∥ u ( t ) ∥ H m + ∥ C ( t ) − I ∥ H m ) < ∞ . \sup_{0\le t<T}\left(\|u(t)\|_{H^m}+\|C(t)-I\|_{H^m}\right)<\infty. 0 ≤ t < T sup ( ∥ u ( t ) ∥ H m + ∥ C ( t ) − I ∥ H m ) < ∞. Consequently, if T T T is the maximal existence time and the same compact window persists up to T T T , the solution continues beyond T T T .
Proof. The proof has two steps. We first close a fixed subcritical tier, and then use that tier as the coefficient control in the integer high-order estimate. This spells out where the three-dimensional assumptions enter and avoids using the unknown H m H^m H m norm as a coefficient in its own Gronwall inequality.
Let
E s ( t ) ≃ ∥ u ( t ) ∥ H s 2 + ∥ h ( t ) ∥ H s 2 + ∥ Y ( t ) ∥ H s 2 , D s ( t ) = ∥ ∇ u ( t ) ∥ H s 2 , h = a − 1 , E_s(t) \simeq\|u(t)\|_{H^s}^2+\|h(t)\|_{H^s}^2+\|Y(t)\|_{H^s}^2,\qquad\mathcal{D}_s(t)=\|\nabla u(t)\|_{H^s}^2,\qquad h=a-1, E s ( t ) ≃ ∥ u ( t ) ∥ H s 2 + ∥ h ( t ) ∥ H s 2 + ∥ Y ( t ) ∥ H s 2 , D s ( t ) = ∥∇ u ( t ) ∥ H s 2 , h = a − 1 ,
where the equivalence constants depend only on the compact window. In FENE-P we use the positive metric
3 α 2 f b ′ ( a ) ∣ h ∣ 2 + α f b ( a ) 2 a ∣ Y ∣ 2 , f b = b − 3 b − 3 a , \frac{3\alpha}{2}f_b'(a)|h|^2+\frac{\alpha f_b(a)}{2a}|Y|^2,\qquad f_b=\frac{b-3}{b-3a}, 2 3 α f b ′ ( a ) ∣ h ∣ 2 + 2 a α f b ( a ) ∣ Y ∣ 2 , f b = b − 3 a b − 3 ,
and in Oldroyd–B the corresponding metric is a scalar multiple of a − 1 ∣ Y ∣ 2 a^{-1}|Y|^2 a − 1 ∣ Y ∣ 2 on the trace-free component. The window makes these weights and a finite number of their derivatives bounded and uniformly positive.
The pressure-free principal terms are the same at every Sobolev level. The trace-free equation contains
2 a S + ( S Y + Y S − 2 3 ( Y : S ) I ) + ( Ω Y − Y Ω ) . 2aS+\left(SY+YS-\frac{2}{3}(Y:S)I\right)+(\Omega Y-Y\Omega). 2 a S + ( S Y + Y S − 3 2 ( Y : S ) I ) + ( Ω Y − Y Ω ) .
The first term is the only active stretching block. The passive symmetric block is estimated by η D s + C η , K E s \eta D_s+C_{\eta,K}E_s η D s + C η , K E s , while the rotation block is skew in the principal dyadic energy and contributes only commutators. On the stress side, Lemma 9.1 gives a cancellative Y Y Y -channel plus the residual R 2 ( Y ) R_2(Y) R 2 ( Y ) ; for Oldroyd–B and FENE-P the residual coefficient is zero. The active pairings are therefore
⟨ α 2 a Z , 2 a S ⟩ = α Z : S \left\langle\frac{\alpha}{2a}Z,2aS\right\rangle=\alpha Z:S ⟨ 2 a α Z , 2 a S ⟩ = α Z : S
for Oldroyd–B, and
⟨ ( 3 α 2 f b ′ ( a ) ξ , α f b ( a ) 2 a Z ) , ( 2 3 Y : S , 2 a S ) ⟩ = α ( f b ′ ( a ) ξ Y + f b ( a ) Z ) : S \left\langle\left(\frac{3\alpha}{2}f_b'(a)\xi,\frac{\alpha f_b(a)}{2a}Z\right),\left(\frac{2}{3}Y:S,2aS\right)\right\rangle=\alpha\left(f_b'(a)\xi Y+f_b(a)Z\right):S ⟨ ( 2 3 α f b ′ ( a ) ξ , 2 a α f b ( a ) Z ) , ( 3 2 Y : S , 2 a S ) ⟩ = α ( f b ′ ( a ) ξ Y + f b ( a ) Z ) : S
for FENE-P. These cancel exactly the top-order pairings with div Y \operatorname{div}Y div Y and div ( f b Y ) \operatorname{div}(f_bY) div ( f b Y ) , respectively, after projection onto divergence-free velocities.
We now record the estimates used to close the lower and upper tiers. For ∣ β ∣ ≤ s |\beta|\le s ∣ β ∣ ≤ s , the transport terms satisfy the endpoint commutator bound
∣ ⟨ ∂ β ( u ⋅ ∇ W ) , ∂ β W ⟩ ∣ ≤ C s B ( t ) ∥ W ∥ H s 2 , W ∈ { u , h , Y } , \left|\left\langle\partial^\beta(u\cdot\nabla W),\partial^\beta W\right\rangle\right|\le C_sB(t)\|W\|_{H^s}^2,\qquad W\in\{u,h,Y\}, ⟨ ∂ β ( u ⋅ ∇ W ) , ∂ β W ⟩ ≤ C s B ( t ) ∥ W ∥ H s 2 , W ∈ { u , h , Y } ,
where B ( t ) = ∥ ∇ u ( t ) ∥ B ∞ , 1 ∞ B(t)=\|\nabla u(t)\|_{B^\infty_{\infty,1}} B ( t ) = ∥∇ u ( t ) ∥ B ∞ , 1 ∞ . The coefficient and composition remainders are treated in tame form: for any smooth coefficient Φ ( C ) \Phi(C) Φ ( C ) on the compact window and s ≥ 3 s\ge3 s ≥ 3 ,
∥ [ ∂ β , Φ ( C ) ] F ∥ L 2 ≤ C s , K ( 1 + ∥ C − I ∥ W 1 , ∞ s ) ∥ C − I ∥ H s ∥ F ∥ H s − 1 . (25) \|[\partial^\beta,\Phi(C)]F\|_{L^2}\le C_{s,K}\left(1+\|C-I\|_{W^{1,\infty}}^{s}\right)\|C-I\|_{H^s}\|F\|_{H^{s-1}}. \tag*{(25)} ∥ [ ∂ β , Φ ( C )] F ∥ L 2 ≤ C s , K ( 1 + ∥ C − I ∥ W 1 , ∞ s ) ∥ C − I ∥ H s ∥ F ∥ H s − 1 . (25)
with constants depending only on finitely many derivatives of Φ \Phi Φ on the window. This is where the three-dimensional argument uses one low tier beyond the energy level.
Take first s 0 = 3 s_0=3 s 0 = 3 . Since H 3 ( T 3 ) ↪ W 1 , ∞ H^3(\mathbb{T}^3)\hookrightarrow W^{1,\infty} H 3 ( T 3 ) ↪ W 1 , ∞ , the same pressure-free cancellation, (25) , and Young’s inequality give
d d t E 3 ( t ) + c D 3 ( t ) ≤ C K ( 1 + B ( t ) ) ( 1 + E 3 ( t ) ) . (26) \frac{d}{dt}E_3(t)+cD_3(t)\le C_K(1+B(t))(1+E_3(t)). \tag*{(26)} d t d E 3 ( t ) + c D 3 ( t ) ≤ C K ( 1 + B ( t )) ( 1 + E 3 ( t )) . (26)
Thus
M 3 : = sup 0 ≤ t < T E 3 ( t ) < ∞ M_3:=\sup_{0\le t<T}E_3(t)<\infty M 3 := 0 ≤ t < T sup E 3 ( t ) < ∞
whenever ∫ 0 T B ( t ) d t < ∞ \int_0^T B(t)\,\mathrm{d}t < \infty ∫ 0 T B ( t ) d t < ∞ . In particular ∥ C − I ∥ W 1 , ∞ \|C-I\|_{W^{1,\infty}} ∥ C − I ∥ W 1 , ∞ is bounded by a constant depending on M 3 M_3 M 3 .
Now let m ≥ 4 m \ge4 m ≥ 4 . Repeating the differentiated estimate at order m m m , all coefficient derivatives in (25) are controlled by M 3 M_3 M 3 , while the unique top-order velocity–conformation coupling has already been cancelled by the weighted active pairing above. The nonlinear convection term is controlled by B ( t ) E m B(t)E_m B ( t ) E m , and the remaining passive stretching, coefficient, and pressure-free stress remainders satisfy
∣ R m ( t ) ∣ ≤ η D m ( t ) + C m , K , M 3 ( 1 + B ( t ) ) ( 1 + E m ( t ) ) . |\mathcal{R}_m(t)| \le\eta\mathcal{D}_m(t) + C_{m,K,M_3}(1+B(t))(1+E_m(t)). ∣ R m ( t ) ∣ ≤ η D m ( t ) + C m , K , M 3 ( 1 + B ( t )) ( 1 + E m ( t )) .
Choosing η \eta η small gives the closed high-order inequality
d d t E m ( t ) + c D m ( t ) ≤ C m , K , M 3 ( 1 + B ( t ) ) ( 1 + E m ( t ) ) . (27) \frac{\mathrm{d}}{\mathrm{d}t}E_m(t) + c\mathcal{D}_m(t) \le C_{m,K,M_3}(1+B(t))(1+E_m(t)). \tag*{(27)} d t d E m ( t ) + c D m ( t ) ≤ C m , K , M 3 ( 1 + B ( t )) ( 1 + E m ( t )) . (27)
The coefficient in (27) is integrable on [ 0 , T ) [0,T) [ 0 , T ) , since M 3 < ∞ M_3 < \infty M 3 < ∞ and B ∈ L 1 ( 0 , T ) B \in L^1(0,T) B ∈ L 1 ( 0 , T ) . Gronwall therefore yields sup t < T E m ( t ) < ∞ \sup_{t<T} E_m(t) < \infty sup t < T E m ( t ) < ∞ . This proves the stated bound for ∥ u ∥ H m + ∥ C − I ∥ H m \|u\|_{H^m}+\|C-I\|_{H^m} ∥ u ∥ H m + ∥ C − I ∥ H m , because the compact window makes the ( h , Y ) (h,Y) ( h , Y ) and C − I C-I C − I norms equivalent.
The continuation statement follows by restarting the standard local strong solution theory from times approaching T T T . The restart time depends only on the compact window and on the uniform H m H^m H m bound above, not on the particular approaching time.
Theorem 9.1 (Three-dimensional Oldroyd–B pure velocity clock). Let ( u , C ) (u,C) ( u , C ) be a strong three-dimensional Oldroyd–B solution on its maximal interval [ 0 , T ∗ ) [0,T_*) [ 0 , T ∗ ) , with C 0 > 0 C_0>0 C 0 > 0 , ( u 0 , C 0 − I ) ∈ H m (u_0,C_0-I)\in H^m ( u 0 , C 0 − I ) ∈ H m , and m ≥ 4 m\ge4 m ≥ 4 . If T ∗ < ∞ T_*<\infty T ∗ < ∞ , then
∫ 0 T ∗ ∥ ∇ u ( t ) ∥ B ∞ , 1 0 d t = ∞ . \int_0^{T_*}\|\nabla u(t)\|_{B^0_{\infty,1}}\,\mathrm{d}t=\infty. ∫ 0 T ∗ ∥∇ u ( t ) ∥ B ∞ , 1 0 d t = ∞.
Proof. Assume the clock is finite. Along particle trajectories the maximal and minimal eigenvalues satisfy
λ ˙ max ≤ 2 ∥ ∇ u ∥ L ∞ λ max − λ − 1 λ max + λ − 1 , \dot{\lambda}_{\max}\le2\|\nabla u\|_{L^\infty}\lambda_{\max}-\lambda^{-1}\lambda_{\max}+\lambda^{-1}, λ ˙ m a x ≤ 2∥∇ u ∥ L ∞ λ m a x − λ − 1 λ m a x + λ − 1 ,
and
λ ˙ min ≥ − 2 ∥ ∇ u ∥ L ∞ λ min − λ − 1 λ min + λ − 1 . \dot{\lambda}_{\min}\ge-2\|\nabla u\|_{L^\infty}\lambda_{\min}-\lambda^{-1}\lambda_{\min}+\lambda^{-1}. λ ˙ m i n ≥ − 2∥∇ u ∥ L ∞ λ m i n − λ − 1 λ m i n + λ − 1 .
Since ∥ ∇ u ∥ L ∞ ≲ B ( t ) \|\nabla u\|_{L^\infty}\lesssim B(t) ∥∇ u ∥ L ∞ ≲ B ( t ) , Gronwall propagates a compact positive-definite window for C C C . Proposition 9.2 then gives a uniform H m H^m H m bound and restarts the solution beyond T ∗ T_* T ∗ , a contradiction.
Theorem 9.2 (Three-dimensional FENE-P blow-up alternative). Let b > 3 b>3 b > 3 , and let ( u , C ) (u,C) ( u , C ) be a strong three-dimensional FENE-P solution on its maximal interval [ 0 , T ∗ ) [0,T_*) [ 0 , T ∗ ) , with C 0 > 0 C_0>0 C 0 > 0 , tr C 0 < b \operatorname{tr} C_0<b tr C 0 < b , ( u 0 , C 0 − I ) ∈ H m (u_0,C_0-I)\in H^m ( u 0 , C 0 − I ) ∈ H m , and m ≥ 4 m\ge4 m ≥ 4 . If T ∗ < ∞ T_*<\infty T ∗ < ∞ , then at least one of the following alternatives occurs:
∫ 0 T ∗ ∥ ∇ u ( t ) ∥ B ∞ , 1 0 d t = ∞ , \int_0^{T_*}\|\nabla u(t)\|_{B^0_{\infty,1}}\,\mathrm{d}t=\infty, ∫ 0 T ∗ ∥∇ u ( t ) ∥ B ∞ , 1 0 d t = ∞ ,
lim inf t ↑ T ∗ inf x λ min C ( t , x ) = 0 , \liminf_{t\uparrow T_*}\inf_x\lambda_{\min}C(t,x)=0, t ↑ T ∗ lim inf x inf λ m i n C ( t , x ) = 0 ,
or
lim inf t ↑ T ∗ inf x ( b − tr C ( t , x ) ) = 0. \liminf_{t\uparrow T_*}\inf_x\bigl(b-\operatorname{tr}C(t,x)\bigr)=0. t ↑ T ∗ lim inf x inf ( b − tr C ( t , x ) ) = 0.
Proof. If none of the three alternatives occurs, then C C C remains in a compact FENE window and the pure clock is finite. Proposition 9.2 therefore gives the uniform H m H^m H m bound and a restart beyond T ∗ T_* T ∗ , a contradiction.
Theorem 9.3 (Three-dimensional FENE-P squared-clock criterion). Under the hypotheses of Theorem 9.2, finite-time breakdown implies
∫ 0 T ∗ ∥ ∇ u ( t ) ∥ B ∞ , 1 0 2 d t = ∞ . \int_0^{T_*}\|\nabla u(t)\|_{B^0_{\infty,1}}^2\,\mathrm{d}t=\infty. ∫ 0 T ∗ ∥∇ u ( t ) ∥ B ∞ , 1 0 2 d t = ∞.
Proof. It remains to show that the squared clock propagates the compact FENE window. Set r = tr C r=\operatorname{tr} C r = tr C and g = b − r g=b-r g = b − r . Taking the trace of the FENE-P equation gives
D t g = − 2 S : C + λ − 1 b ( b − 3 − g ) g . D_t g=-2S:C+\lambda^{-1}\frac{b(b-3-g)}{g}. D t g = − 2 S : C + λ − 1 g b ( b − 3 − g ) .
Thus D t ( − log g ) D_t(-\log g) D t ( − log g ) is bounded above by
C ( 1 + ∥ ∇ u ∥ L ∞ 2 ) , C\left(1+\|\nabla u\|_{L^\infty}^{2}\right), C ( 1 + ∥∇ u ∥ L ∞ 2 ) ,
after completing the square against the positive singular term g − 2 g^{-2} g − 2 . The squared Besov clock therefore propagates a positive trace gap on finite time intervals. The lower eigenvalue satisfies the usual Dini inequality
λ ˙ min ≥ − C ∥ ∇ u ∥ L ∞ λ min − C T λ min + c T , \dot{\lambda}_{\min}\ge-C\|\nabla u\|_{L^\infty}\lambda_{\min}-C_T\lambda_{\min}+c_T, λ ˙ m i n ≥ − C ∥∇ u ∥ L ∞ λ m i n − C T λ m i n + c T ,
once the trace gap bounds f b f_b f b . Since the squared clock implies the L t 1 L x ∞ L_t^1L_x^\infty L t 1 L x ∞ velocity-gradient clock on finite intervals, the positive lower eigenvalue bound is also propagated. Hence C C C stays in a compact FENE window, and Theorem 9.2 rules out finite-time breakdown under the squared clock. □
Entropy-mobility side condition. The preceding three-dimensional theorems are velocity-clock results. They use spectral barriers only to keep the solution in a compact physical window; after that, the pressure-free high-order estimate closes without a three-dimensional logarithmic conformation energy. The following finite-dimensional calculation is included to locate the obstruction faced by a stronger program: deriving the compact-window logarithmic estimate directly in three-dimensional FENE-P entropy variables. In that program the mobility map from the Peterlin stress variables to the entropy variables must be monotone. The number q ∗ ( b , r ) q_*(b,r) q ∗ ( b , r ) below is precisely the trace-dependent variance threshold for that monotonicity. Thus Proposition 9.3 supports the scope statement of this section; it is not an additional step in the preceding velocity-clock proofs.
Definition 9.1 (Entropy-mobility admissible windows for three-dimensional FENE-P). Let b > 3 b>3 b > 3 . For a positive spectrum ℓ = ( ℓ 1 , ℓ 2 , ℓ 3 ) \ell=(\ell_1,\ell_2,\ell_3) ℓ = ( ℓ 1 , ℓ 2 , ℓ 3 ) with r = ℓ 1 + ℓ 2 + ℓ 3 < b r=\ell_1+\ell_2+\ell_3<b r = ℓ 1 + ℓ 2 + ℓ 3 < b , set
q ( ℓ ) = ∑ i = 1 3 ( ℓ i − r 3 ) 2 , θ = b − r b − 3 . q(\ell)=\sum_{i=1}^{3}\left(\ell_i-\frac{r}{3}\right)^2,\qquad\theta=\frac{b-r}{b-3}. q ( ℓ ) = i = 1 ∑ 3 ( ℓ i − 3 r ) 2 , θ = b − 3 b − r . When θ > 4 b / 3 \theta>4b/3 θ > 4 b /3 , define
A ∗ ( b , r ) = θ ( b − r ) 3 b + r 2 9 b , q ∗ ( b , r ) = A ∗ ( b , r ) b 2 θ / 4 − b / 3 . A_*(b,r)=\frac{\theta(b-r)}{3b}+\frac{r^2}{9b},\qquad q_*(b,r)=\frac{A_*(b,r)b^2}{\theta/4-b/3}. A ∗ ( b , r ) = 3 b θ ( b − r ) + 9 b r 2 , q ∗ ( b , r ) = θ /4 − b /3 A ∗ ( b , r ) b 2 . The quantity q ( ℓ ) q(\ell) q ( ℓ ) is the spectral variance at fixed trace, and q ∗ ( b , r ) q_*(b,r) q ∗ ( b , r ) is the largest variance allowed by positivity of the entropy-variable mobility Jacobian when the unconditional estimate is unavailable. A compact FENE window K ⋐ { C > 0 , tr C < b } K\Subset\{C>0,\operatorname{tr}C<b\} K ⋐ { C > 0 , tr C < b } is entropy-mobility admissible if either b ≥ 15 / 4 b\ge15/4 b ≥ 15/4 , or there exists γ > 0 \gamma>0 γ > 0 such that, for every spectrum of every C ∈ K C\in K C ∈ K with θ > 4 b / 3 \theta>4b/3 θ > 4 b /3 ,
q ( ℓ ) ≤ q ∗ ( b , r ) − γ . q(\ell)\le q_*(b,r)-\gamma. q ( ℓ ) ≤ q ∗ ( b , r ) − γ .
Proposition 9.3 (Entropy-variable mobility condition for three-dimensional FENE-P). Let b > 3 b>3 b > 3 , c = b − 3 c=b-3 c = b − 3 , r = ℓ 1 + ℓ 2 + ℓ 3 < b r=\ell_1+\ell_2+\ell_3<b r = ℓ 1 + ℓ 2 + ℓ 3 < b , and
ψ ( r ) = c b − r , p i = ψ ( r ) ℓ i − 1 , F i = ψ ( r ) − ℓ i − 1 . \psi(r)=\frac{c}{b-r},\qquad p_i=\psi(r)\ell_i-1,\qquad F_i=\psi(r)-\ell_i^{-1}. ψ ( r ) = b − r c , p i = ψ ( r ) ℓ i − 1 , F i = ψ ( r ) − ℓ i − 1 .
Then the symmetric part of the Jacobian D p F D_pF D p F is positive definite for every positive spectrum with r < b r < b r < b if b ≥ 15 / 4 b \ge15/4 b ≥ 15/4 . More generally, on every entropy-mobility admissible compact FENE window K K K there is c K > 0 c_K > 0 c K > 0 such that
1 2 ( D p F + ( D p F ) T ) ≥ c K I . \frac{1}{2}\left(D_pF + (D_pF)^{\mathsf{T}}\right) \ge c_K I. 2 1 ( D p F + ( D p F ) T ) ≥ c K I .
For 3 < b < 15 / 4 3 < b < 15/4 3 < b < 15/4 , the condition in Definition 9.1 is sharp at the level of this finite-dimensional Jacobian: without an anisotropy restriction, the symmetric part can be indefinite.
Proof. Let e = ( 1 , 1 , 1 ) T \boldsymbol{e} = (1,1,1)^{\mathsf{T}} e = ( 1 , 1 , 1 ) T , D = diag ( ℓ i − 1 ) D = \operatorname{diag}(\ell_i^{-1}) D = diag ( ℓ i − 1 ) , and
θ = b − r b − 3 . \theta= \frac{b-r}{b-3}. θ = b − 3 b − r .
The Jacobians with respect to ℓ \ell ℓ are
J p = ψ I + ψ ′ ℓ e T , J F = diag ( ℓ i − 2 ) + ψ ′ e e T , ψ ′ = b − 3 ( b − r ) 2 . J_p = \psi I + \psi' \ell\boldsymbol{e}^{\mathsf{T}}, \qquad J_F = \operatorname{diag}(\ell_i^{-2}) + \psi' \boldsymbol{e}\boldsymbol{e}^{\mathsf{T}}, \qquad\psi' = \frac{b-3}{(b-r)^2}. J p = ψ I + ψ ′ ℓ e T , J F = diag ( ℓ i − 2 ) + ψ ′ e e T , ψ ′ = ( b − r ) 2 b − 3 .
Since
J p − 1 = θ ( I − 1 b ℓ e T ) , J_p^{-1} = \theta\left(I-\frac{1}{b}\ell\boldsymbol{e}^{\mathsf{T}}\right), J p − 1 = θ ( I − b 1 ℓ e T ) ,
we obtain
D p F = θ diag ( ℓ i − 2 ) − θ b diag ( ℓ i − 1 ) e e T + 1 b e e T . D_pF = \theta\operatorname{diag}(\ell_i^{-2})-\frac{\theta}{b}\operatorname{diag}(\ell_i^{-1})\boldsymbol{e}\boldsymbol{e}^{\mathsf{T}}+\frac{1}{b}\boldsymbol{e}\boldsymbol{e}^{\mathsf{T}}. D p F = θ diag ( ℓ i − 2 ) − b θ diag ( ℓ i − 1 ) e e T + b 1 e e T .
Hence, for z ∈ R 3 z \in\mathbb{R}^3 z ∈ R 3 ,
z T D p F + ( D p F ) T 2 z = θ ∑ i z i 2 ℓ i 2 − θ b ( ∑ i z i ℓ i ) ( ∑ i z i ) + 1 b ( ∑ i z i ) 2 . z^{\mathsf{T}}\frac{D_pF+(D_pF)^{\mathsf{T}}}{2}z
= \theta\sum_i\frac{z_i^2}{\ell_i^2}
-\frac{\theta}{b}\left(\sum_i\frac{z_i}{\ell_i}\right)\left(\sum_i z_i\right)
+\frac{1}{b}\left(\sum_i z_i\right)^2. z T 2 D p F + ( D p F ) T z = θ i ∑ ℓ i 2 z i 2 − b θ ( i ∑ ℓ i z i ) ( i ∑ z i ) + b 1 ( i ∑ z i ) 2 .
Put w i = z i / ℓ i w_i = z_i/\ell_i w i = z i / ℓ i . Then the quadratic form becomes
Q ( w ) = θ ∣ w ∣ 2 − θ b ( e ⋅ w ) ( ℓ ⋅ w ) + 1 b ( ℓ ⋅ w ) 2 . Q(w) = \theta|w|^2 - \frac{\theta}{b}(\boldsymbol{e}\cdot w)(\ell\cdot w) + \frac{1}{b}(\ell\cdot w)^2. Q ( w ) = θ ∣ w ∣ 2 − b θ ( e ⋅ w ) ( ℓ ⋅ w ) + b 1 ( ℓ ⋅ w ) 2 .
Write
ℓ = r 3 e + m , e ⋅ m = 0 , q = ∣ m ∣ 2 . \ell= \frac{r}{3}\boldsymbol{e}+m, \qquad\boldsymbol{e}\cdot m = 0, \qquad q = |m|^2. ℓ = 3 r e + m , e ⋅ m = 0 , q = ∣ m ∣ 2 .
If q = 0 q = 0 q = 0 , the displayed form is plainly positive. If q > 0 q > 0 q > 0 , set
A = e ⋅ w , M = m ⋅ w . A = \boldsymbol{e}\cdot w, \qquad M = m\cdot w. A = e ⋅ w , M = m ⋅ w .
For fixed A A A and M M M , the smallest possible ∣ w ∣ 2 |w|^2 ∣ w ∣ 2 is
A 2 3 + M 2 q . \frac{A^2}{3}+\frac{M^2}{q}. 3 A 2 + q M 2 .
Thus positivity is equivalent to positivity of
a ∗ A 2 + g ∗ M 2 + h ∗ A M , a_*A^2+g_*M^2+h_*AM, a ∗ A 2 + g ∗ M 2 + h ∗ A M ,
where
a ∗ = θ ( b − r ) 3 b + r 2 9 b , g ∗ = θ q + 1 b , h ∗ = 2 r / 3 − θ b . a_* = \frac{\theta(b-r)}{3b}+\frac{r^2}{9b}, \qquad g_* = \frac{\theta}{q}+\frac{1}{b}, \qquad h_* = \frac{2r/3-\theta}{b}. a ∗ = 3 b θ ( b − r ) + 9 b r 2 , g ∗ = q θ + b 1 , h ∗ = b 2 r /3 − θ .
A direct simplification gives
a ∗ g ∗ − h ∗ 2 4 = θ { a ∗ q − θ / 4 − b / 3 b 2 } . a_*g_*-\frac{h_*^2}{4}
= \theta\left\{\frac{a_*}{q}-\frac{\theta/4-b/3}{b^2}\right\}. a ∗ g ∗ − 4 h ∗ 2 = θ { q a ∗ − b 2 θ /4 − b /3 } .
If θ ≤ 4 b / 3 \theta\le4b/3 θ ≤ 4 b /3 , the determinant is positive. Since
θ ≤ b b − 3 ≤ 4 b 3 \theta\le\frac{b}{b-3} \le\frac{4b}{3} θ ≤ b − 3 b ≤ 3 4 b
whenever b ≥ 15 / 4 b \ge15/4 b ≥ 15/4 , this proves unconditional positivity in that range. If θ > 4 b / 3 \theta> 4b/3 θ > 4 b /3 , the determinant condition is exactly
q < a ∗ b 2 θ / 4 − b / 3 = q ∗ ( b , r ) . q < \frac{a_*b^2}{\theta/4-b/3}=q_*(b,r). q < θ /4 − b /3 a ∗ b 2 = q ∗ ( b , r ) .
On a compact entropy-mobility admissible window, the strict gap gives a uniform lower bound c K c_K c K . Conversely, if 3 < b < 15 / 4 3<b<15/4 3 < b < 15/4 , spectra with sufficiently large variance and θ > 4 b / 3 \theta>4b/3 θ > 4 b /3 violate the determinant condition, so the symmetric Jacobian is indefinite.
Hookean limit and relation between the two criteria
The FENE-P criterion reduces to the Oldroyd–B criterion only in a controlled limit. Suppose 0 < C ≤ M I 0<C\le MI 0 < C ≤ M I and tr C ≤ M \operatorname{tr}C\le M tr C ≤ M , with M M M independent of b b b . Then
f b ( C ) = b − 2 b − tr C = 1 + tr C − 2 b + O M ( b − 2 ) . f_b(C)=\frac{b-2}{b-\operatorname{tr}C}=1+\frac{\operatorname{tr}C-2}{b}+O_M(b^{-2}). f b ( C ) = b − tr C b − 2 = 1 + b tr C − 2 + O M ( b − 2 ) .
and hence
T b ( C ) = C − I + tr C − 2 b C + O M ( b − 2 ) . T_b(C)=C-I+\frac{\operatorname{tr}C-2}{b}C+O_M(b^{-2}). T b ( C ) = C − I + b tr C − 2 C + O M ( b − 2 ) .
Moreover
ϕ b ( C ) = − log b + O M ( b − 1 ) , ∇ ϕ b ( C ) = ∇ tr C b − tr C = O M ( b − 1 ) ∇ C . \phi_b(C)=-\log b+O_M(b^{-1}),\qquad\nabla\phi_b(C)=\frac{\nabla\operatorname{tr}C}{b-\operatorname{tr}C}=O_M(b^{-1})\nabla C. ϕ b ( C ) = − log b + O M ( b − 1 ) , ∇ ϕ b ( C ) = b − tr C ∇ tr C = O M ( b − 1 ) ∇ C .
Thus, on bounded trace windows independent of b b b , the FENE-P physical criterion formally reduces to the Oldroyd–B positive-cone criterion. At fixed b b b , or along sequences for which tr C \operatorname{tr}C tr C approaches b b b , this reduction is invalid at the level of pointwise barrier control. The endpoint theorem therefore propagates the trace gap before applying the pressure-free high-order estimate.
Proposition 10.1 (Entropy does not control trace-gap oscillation). Fix b > 2 b>2 b > 2 . There are smooth diagonal fields C N : T 2 → D b C_N:\mathbb{T}^2\to D_b C N : T 2 → D b whose eigenvalues stay in a fixed positive interval and whose positive-cone entropy
∫ T 2 ( tr C N − log det C N ) d x \int_{\mathbb{T}^2}(\operatorname{tr} C_N-\log\det C_N)\,dx ∫ T 2 ( tr C N − log det C N ) d x is uniformly bounded, while
∥ ϕ b ( C N ) ∥ H 1 + ε → ∞ . \|\phi_b(C_N)\|_{H^{1+\varepsilon}}\to\infty. ∥ ϕ b ( C N ) ∥ H 1 + ε → ∞. Proof. Choose
θ N ( x ) = θ 0 + a N sin ( N x 1 ) , \theta_N(x)=\theta_0+a_N\sin(Nx_1), θ N ( x ) = θ 0 + a N sin ( N x 1 ) , with 0 < θ 0 < b 0<\theta_0<b 0 < θ 0 < b , a N → 0 a_N\to0 a N → 0 , and a N N 1 + ε → ∞ a_NN^{1+\varepsilon}\to\infty a N N 1 + ε → ∞ . Let
C N ( x ) = θ N ( x ) 2 I . C_N(x)=\frac{\theta_N(x)}{2}I. C N ( x ) = 2 θ N ( x ) I . For N N N large, the eigenvalues remain in a fixed positive interval and the positive-cone entropy is uniformly bounded. However
ϕ b ( C N ) = − log ( b − θ N ) , \phi_b(C_N)=-\log(b-\theta_N), ϕ b ( C N ) = − log ( b − θ N ) , whose H 1 + ε H^{1+\varepsilon} H 1 + ε norm grows like a N N 1 + ε a_NN^{1+\varepsilon} a N N 1 + ε . Thus positive-cone entropy does not control high-frequency oscillation of the trace-gap barrier. The FENE-P endpoint criterion above avoids assuming this barrier norm directly; it uses the pointwise barrier equation to recover the compact window and the pressure-free physical variables to close the derivative estimate.
Conclusion
The pressure quotient isolates the part of an elastic stress that can force an incompressible velocity. In two dimensions, every smooth spectral isotropic stress has active part q 1 Y q_1Y q 1 Y after pressure projection. The trace-free conformation equation then supplies the matching principal stretching term 2 a S ( u ) 2aS(u) 2 a S ( u ) , and the weight q 1 / a q_1/a q 1 / a cancels the top-order velocity–stress coupling. On compact conformation windows this gives the endpoint high-order coefficient
1 + ∥ ∇ u ∥ B ∞ , 1 0 + ∥ log C ∥ H 1 + ε 2 . 1+\|\nabla u\|_{B^0_{\infty,1}}+\|\log C\|_{H^{1+\varepsilon}}^2. 1 + ∥∇ u ∥ B ∞ , 1 0 + ∥ log C ∥ H 1 + ε 2 .
For Oldroyd–B the compact positive-cone window is propagated by the L t 1 B ∞ , 1 0 L^1_tB^0_{\infty,1} L t 1 B ∞ , 1 0 velocity clock, and the pressure-free low-order energy provides the logarithmic channel required by the high-order estimate. For FENE-P the same active-deviatoric cancellation applies after the splitting T b ( C ) = ( f b a − 1 ) I + f b Y T_b(C)=(f_ba^{-1})I+f_bY T b ( C ) = ( f b a − 1 ) I + f b Y ; the finite-extensibility trace gap is propagated by the squared L t 2 B ∞ , 1 0 L^2_tB^0_{\infty,1} L t 2 B ∞ , 1 0 clock.
The three-dimensional statements have a narrower scope, and the restriction is structural. For spectral isotropic laws with a smooth local Cayley–Hamilton representation on a compact spectral window, the pressure-free stress splits into a cancellative Y Y Y -channel and a quadratic residual ( Y 2 ) ∘ (Y^2)^\circ ( Y 2 ) ∘ . The residual is generically independent of Y Y Y ; hence the exact scalar quotient closure is a two-dimensional phenomenon. In three dimensions the residual is estimated separately and absorbed by viscosity on compact windows. General anisotropic or non-spectral closures may contain additional pressure-free directions and are not covered by this quotient closure. For Oldroyd–B and FENE-P the quadratic coefficient is zero, and the compact window used in the continuation criteria are supplied by the corresponding model-specific barriers.
All criteria are strong-solution continuation criteria. The proof uses integer-order Sobolev differentiations, Moser estimates for spectral functions, and pointwise conformation barriers. It does not construct Leray-type weak solutions or prove weak-solution regularization. This does not leave the criterion without objects: the local principles in § 2 give positive-time strong solutions for integer-Sobolev data with compact initial conformation windows. The endpoint clocks are therefore blow-up exclusion mechanisms for maximal strong solutions in that established local class, not substitutes for a critical H 1 H^{1} H 1 or energy-level well-posedness theory. The spectrally admissible class is used only as a compact-window quotient template. Thermodynamic polymer laws require the additional free-energy compatibility and relaxation-dissipation conditions stated in Remark 6.1.
The fixed-spectrum examples show why the logarithmic channel enters the estimates: entropy, trace, determinant, and finite-extensibility bounds do not control high-frequency rotation of the active deviatoric stress. These examples are static obstructions for the pressure-free stress map. They do not prove that loss of logarithmic regularity forces finite-time blow-up of the evolution. For the two model systems, the continuation loop closes instead through the propagated conformation window, the logarithmic regularity estimate, and the endpoint velocity clock.
Acknowledgements
The author acknowledges financial support from the National Natural Science Foundation of China (NSFC, Grant No. 12501602), the Education Department of Hunan Province (Grant No. 24C0055), the Science and Technology Department of Hunan Province (Grant No. 2025JJ60052), and the Scientific Research Start-up Fund of Xiangtan University (Grant No. KZ0810769).