Introduction
The Ericksen–Leslie theory couples incompressible flow to a unit-vector director describing the local orientation of a nematic liquid crystal. We refer to Ericksen [14 ] and Leslie [36 ] , the monographs [3 , 11 ] , and the surveys [2 , 46 ] for a comprehensive overview of liquid crystal models. The pointwise constraint ∣ d ∣ = 1 |\mathbf{d}| = 1 ∣ d ∣ = 1 present in Ericksen–Leslie models is analytically delicate in three dimensions, because the natural energy bounds give only weak H 1 H^1 H 1 -compactness, which does not identify the quadratic Ericksen stress and allows concentration of the director energy.
The interaction problem of an Ericksen–Leslie model with a rigid body combines the Ericksen–Leslie description of a nematic flow with Newton’s laws, giving rise to a moving boundary problem in which the fluid velocity, the molecular orientation, expressed by the director field, and the rigid motion are coupled. Mechanically, a significant difficulty arises due to the so-called anchoring at the surface of the rigid body. It produces viscous forces and additionally distorts the director field, generating elastic forces. Mathematically, this is expressed by a non-homogeneous Dirichlet boundary condition of the director field. Another important aspect in this context of the possibility of contact of the rigid body with the outer boundary of the fluid container.
In this paper, we study this coupling for the simplified Ericksen–Leslie system introduced by Lin [37 ] . The fluid is subject to no-slip boundary conditions, while the director d = ( d 1 , d 2 , d 3 ) \mathbf{d} = (d_1,d_2,d_3) d = ( d 1 , d 2 , d 3 ) satisfies the condition ∣ d ∣ = 1 |\mathbf{d}| = 1 ∣ d ∣ = 1 as well as d 3 ≥ 0 d_3 \ge0 d 3 ≥ 0 , and its boundary values on the body are fixed in material coordinate. Our main result, Theorem 2.3 , constructs a finite-energy weak solution up to the first wall contact and continues it afterwards under permanent sticking, i.e., the rigid body is assumed to remain attached to the outer fluid boundary after contact. Moreover, in Corollary 2.10 , we provide an explicit lower bound for the contact time.
The solution to the interaction problem of the Ericksen–Leslie model including the aforementioned director condition is obtained by passing to the limit in a Ginzburg–Landau regularization. On fixed domains, the corresponding regularized model was analyzed by Lin and Liu [38 ] . In their Markdown 2020 Mathematics Subject Classification. 35Q35, 76A15, 74F10, 76D03, 35K59.
Key words and phrases. Simplified Ericksen–Leslie model, fluid – rigid body interaction, global weak solutions, moving-domain singular limits, Ginzburg–Landau approximation, wall contact. seminal work [39 ] , Lin and Wang proved global existence of a weak solution to the simplified Ericksen–Leslie model with director condition. The essential input from [39 ] for the present work is a strong compactness result for approximate harmonic maps in the interior of the fluid domain, see Lemma 4.3. An additional difficulty in this article is that we need to extend the director through the body. This extension satisfies a transport equation on the whole fluid–rigid body container, which paves the way for global strong L 2 L^2 L 2 -compactness.
Since the domain resulting from a contact of the rigid body with the outer boundary may have cusps or several components, we prove in Theorem 2.6 the existence of a weak solution to the simplified Ericksen–Leslie model with director condition on arbitrary open sets. The proof uses the variational Dirichlet Laplacian, compactness of the zero-extension, and a compactness argument à la Lin–Wang in the interior, without imposing regularity on the boundary.
For the analysis of Ericksen–Leslie models on fixed domains, we refer to [20 ] for smooth relaxation limits, [35 ] for weak solutions in the axisymmetric setting, [9 , 30 , 31 , 48 ] for strong solutions, or [34 , 49 ] for singularity formation and regularity. A different description of liquid crystals is due to Beris and Edwards [3 ] , where the molecular orientation is described by tensors that are symmetric and have trace zero. The mathematical analysis of such Q Q Q -tensor models on fixed domains has been carried out in [1 , 18 , 29 , 42 , 43 , 44 , 47 ] .
In the context of interaction problems of viscous incompressible fluids with a rigid body, we refer, e.g., to [10 , 12 , 13 , 16 , 17 , 27 , 45 ] for existence results of weak solutions, while strong well-posedness and the long-time behavior of such fluid–rigid interaction problems has, for example, been discussed in [15 , 22 ] . Recent work concerns regularity of weak solutions [41 ] , small-body and vanishing-body limits [8 , 19 , 28 ] , and multi-particle or long-time regimes [7 ] . The behavior near contact is especially sensitive to the boundary condition. Under no slip, non-collision and blow-up mechanisms were studied in [32 , 33 ] , while the case with slip was analyzed in [25 , 26 ] .
For liquid crystals coupled to a rigid body, Binz et al. [4 ] proved local strong well-posedness and global strong existence near equilibria for the simplified Ericksen–Leslie system with Neumann director conditions. Brandt et al. [6 ] proved similar results in the context of the general Q Q Q -tensor model. Geng et al. [24 ] constructed global weak solutions for a rigid body in a Q Q Q -tensor fluid, including the discussion of wall contact. Mensah [40 ] considered a two-dimensional simplified Ericksen–Leslie fluid coupled to a flexible shell.
It is worthwhile comparing our result to the previous works [4 , 24 ] . In fact, we allow for general data in the natural finite-energy class, so in particular, compared to [4 ] , we consider lower regularity for the initial data, and we do not require a smallness assumption to infer the global-in-time existence of a solution. In addition, we discuss the case of contact of the rigid body with the fluid boundary, which has been omitted in the work [4 ] on strong solutions. Let us emphasize that we address the simplified Ericksen–Leslie model including the director condition ∣ d ∣ = 1 |d| = 1 ∣ d ∣ = 1 . This condition as well as the aforementioned anchoring of the rigid body, which is expressed via non-homogeneous Dirichlet boundary conditions, and which requires a delicate extension procedure, pose significant additional challenges compared to the work [24 ] in the context of the Q Q Q -tensor model. To the best of our knowledge, this work is the first result on the global-in-time existence of solutions to the fluid–rigid body interaction problem of an Ericksen–Leslie liquid crystal model in three dimensions.
The rest of the paper is organized as follows. § 2 introduces the model, the weak formulations, and the main results, including energy estimates for the rigid displacement. In § 3 , we construct solutions to the moving domain-problem of the Ginzburg–Landau regularization by removing a Galerkin approximation, a penalization of rigidity, and an artificial viscosity. This section also develops the fixed-domain Ginzburg–Landau theory on arbitrary bounded open sets. § 4 then carries out the singular limit ε → 0 \varepsilon\to0 ε → 0 on the moving domain. In § 5 , the solution obtained from the singular limit passage is continued up to the first contact time and then also continued after impact, completing the proof of Theorem 2.3 . Auxiliary results are proved in Appendix A.
Statement of the problem and main results
First of all, let us introduce some notations. We use C C C for a general positive constant, C ( a , b ) C(a,b) C ( a , b ) for a positive constant that may depend on a a a and b b b and may change from line to line. For a measurable set Q Q Q , let
Char ( Q ) : = { χ ∈ L ∞ ( Q ) : χ ∈ { 0 , 1 } a.e. in Q } \operatorname{Char}(Q):=\{\chi\in L^\infty(Q):\chi\in\{0,1\}\ \text{a.e. in }Q\} Char ( Q ) := { χ ∈ L ∞ ( Q ) : χ ∈ { 0 , 1 } a.e. in Q }
denote the class of characteristic functions on Q Q Q . For a fixed spatial domain G G G , we abbreviate L t p X x : = L p ( 0 , T ; X ( G ) ) L^p_tX_x:=L^p(0,T;X(G)) L t p X x := L p ( 0 , T ; X ( G )) and L t , x p : = L p ( ( 0 , T ) × G ) L^p_{t,x}:=L^p((0,T)\times G) L t , x p := L p (( 0 , T ) × G ) , and we use the same abbreviations for norms. Here G = Ω G=\Omega G = Ω unless another domain is specified, and moving spatial domains are always displayed explicitly. We set H − 1 ( G ) : = ( H 0 1 ( G ) ) ′ H^{-1}(G):=(H^1_0(G))' H − 1 ( G ) := ( H 0 1 ( G ) ) ′ . Time slices of Bochner functions are evaluated at their precise representatives on a common set of full measure, unless a continuous representative is specified.
The simplified Ericksen–Leslie model coupled with a rigid body
The simplified Ericksen–Leslie model coupled with a rigid body. We consider a rigid body interacting with a simplified Ericksen–Leslie fluid. The fluid density, viscosity, and solid density are normalized to one, and vector-valued quantities are written in boldface.
2.1.1. Geometric setting. Let Ω ⊂ R 3 \Omega\subset\mathbb{R}^3 Ω ⊂ R 3 be an open, bounded, and connected domain with boundary ∂ Ω \partial\Omega ∂ Ω of class C 2 C^2 C 2 , representing the container in which the system evolves. A rigid body occupies, at initial time, an open, bounded, and connected domain S 0 ⊂ Ω S_0\subset\Omega S 0 ⊂ Ω with C 2 C^2 C 2 -boundary, whose center of mass is located at the origin, i.e.,
1 ∣ S 0 ∣ ∫ S 0 x d x = 0. \frac{1}{|S_0|}\int_{S_0}x\,\mathrm{d}x=0. ∣ S 0 ∣ 1 ∫ S 0 x d x = 0.
Let T > 0 T>0 T > 0 be a given final time. At each time t ∈ ( 0 , T ) t\in(0,T) t ∈ ( 0 , T ) , the position of the rigid body is characterized by its center of mass h ( t ) ∈ R 3 h(t)\in\mathbb{R}^3 h ( t ) ∈ R 3 and a rotation matrix O ( t ) ∈ S O ( 3 ) O(t)\in SO(3) O ( t ) ∈ S O ( 3 ) , so that the solid domain is given by
S ( t ) = { h ( t ) + O ( t ) y : y ∈ S 0 } . S(t)=\{h(t)+O(t)y:y\in S_0\}. S ( t ) = { h ( t ) + O ( t ) y : y ∈ S 0 } .
Before contact, S ( t ) ‾ ⊂ Ω \overline{S(t)}\subset\Omega S ( t ) ⊂ Ω , and the fluid domain at time t t t is
F ( t ) : = Ω ∖ S ( t ) ‾ , \mathcal{F}(t):=\Omega\setminus\overline{S(t)}, F ( t ) := Ω ∖ S ( t ) ,
with initial fluid domain F 0 : = Ω ∖ S 0 ‾ \mathcal{F}_0:=\Omega\setminus\overline{S_0} F 0 := Ω ∖ S 0 . The rigid velocity is
u S ( t , x ) = ℓ ( t ) + ω ( t ) × ( x − h ( t ) ) , ( t , x ) ∈ ( 0 , T ) × S ( t ) , u^S(t,x)=\ell(t)+\omega(t)\times(x-h(t)),\qquad(t,x)\in(0,T)\times S(t), u S ( t , x ) = ℓ ( t ) + ω ( t ) × ( x − h ( t )) , ( t , x ) ∈ ( 0 , T ) × S ( t ) ,
where ℓ ( t ) = h ′ ( t ) \ell(t)=h'(t) ℓ ( t ) = h ′ ( t ) denotes the translational velocity and ω ( t ) \omega(t) ω ( t ) the angular velocity. To prescribe the director anchoring on the moving interface, let Z ( t , y ) Z(t,y) Z ( t , y ) be the rigid flow map. Denote by m ( t ) m(t) m ( t ) the skew-symmetric matrix satisfying m ( t ) x = ω ( t ) × x m(t)x=\omega(t)\times x m ( t ) x = ω ( t ) × x . Then
∂ t Z ( t , y ) = m ( t ) ( Z ( t , y ) − h ( t ) ) + ℓ ( t ) , in ( 0 , T ) × R 3 , Z ( 0 , y ) = y , for y ∈ R 3 . (1) \begin{aligned}
\partial_tZ(t,y)&=m(t)(Z(t,y)-h(t))+\ell(t), &&\text{in }(0,T)\times\mathbb{R}^3,\\
Z(0,y)&=y, &&\text{for }y\in\mathbb{R}^3. \tag*{(1)}
\end{aligned} ∂ t Z ( t , y ) Z ( 0 , y ) = m ( t ) ( Z ( t , y ) − h ( t )) + ℓ ( t ) , = y , in ( 0 , T ) × R 3 , for y ∈ R 3 . (1)
Let O ( t ) ∈ S O ( 3 ) O(t)\in SO(3) O ( t ) ∈ S O ( 3 ) solve O ′ ( t ) = m ( t ) O ( t ) O'(t)=m(t)O(t) O ′ ( t ) = m ( t ) O ( t ) with O ( 0 ) = I 3 O(0)=\mathbb{I}_3 O ( 0 ) = I 3 . Since h ( 0 ) = 0 h(0)=0 h ( 0 ) = 0 , the solution of (1) is
Z ( t , y ) = h ( t ) + O ( t ) y . Z(t,y)=h(t)+O(t)y. Z ( t , y ) = h ( t ) + O ( t ) y .
Write Z t ( y ) : = Z ( t , y ) Z_t(y):=Z(t,y) Z t ( y ) := Z ( t , y ) . Then S ( t ) = Z t ( S 0 ) S(t)=Z_t(S_0) S ( t ) = Z t ( S 0 ) . The inverse ( Z t ) − 1 (Z_t)^{-1} ( Z t ) − 1 solves
∂ t ( Z t ) − 1 ( x ) = − O ⊤ ( t ) m ( t ) O ( t ) ( Z t ) − 1 ( x ) − O ⊤ ( t ) ℓ ( t ) , in ( 0 , T ) × R 3 , ( Z 0 ) − 1 ( x ) = x , for x ∈ R 3 . (2) \begin{aligned}
\partial_t(Z_t)^{-1}(x)&=-O^\top(t)m(t)O(t)(Z_t)^{-1}(x)-O^\top(t)\ell(t), &&\text{in }(0,T)\times\mathbb{R}^3,\\
(Z_0)^{-1}(x)&=x, &&\text{for }x\in\mathbb{R}^3. \tag*{(2)}
\end{aligned} ∂ t ( Z t ) − 1 ( x ) ( Z 0 ) − 1 ( x ) = − O ⊤ ( t ) m ( t ) O ( t ) ( Z t ) − 1 ( x ) − O ⊤ ( t ) ℓ ( t ) , = x , in ( 0 , T ) × R 3 , for x ∈ R 3 . (2)
Simplified Ericksen–Leslie equations
The fluid phase is governed by the simplified Ericksen–Leslie system introduced by Lin [37 ] and Lin–Liu [38 ] . The unknowns in the fluid region are the velocity u F \mathbf{u}^{\mathcal{F}} u F , the pressure π \pi π , and the director field d \mathbf{d} d . They are defined on ( 0 , T ) × F ( t ) (0,T) \times\mathcal{F}(t) ( 0 , T ) × F ( t ) . Their values lie in R 3 \mathbb{R}^{3} R 3 , R \mathbb{R} R , and S 2 \mathbb{S}^{2} S 2 , respectively, where
S 2 = { f ∈ R 3 : ∣ f ∣ = 1 } \mathbb{S}^{2} = \{f \in\mathbb{R}^{3} : \lvert f\rvert= 1\} S 2 = { f ∈ R 3 : ∣ f ∣ = 1 }
denotes the unit sphere in R 3 \mathbb{R}^{3} R 3 . The condition ∣ d ∣ = 1 \lvert\mathbf{d}\rvert= 1 ∣ d ∣ = 1 is the director constraint.
The equations in the fluid domain are
{ ∂ t u F + ( u F ⋅ ∇ ) u F − Δ u F + ∇ π = − div ( ( ∇ d ) T ∇ d ) , in ( 0 , T ) × F ( t ) , div u F = 0 , in ( 0 , T ) × F ( t ) , ∂ t d + ( u F ⋅ ∇ ) d − Δ d = ∣ ∇ d ∣ 2 d , in ( 0 , T ) × F ( t ) , ∣ d ∣ = 1 , in ( 0 , T ) × F ( t ) . (3) \begin{cases}
\partial_{t}\mathbf{u}^{\mathcal{F}} + (\mathbf{u}^{\mathcal{F}} \cdot\nabla)\mathbf{u}^{\mathcal{F}} - \Delta\mathbf{u}^{\mathcal{F}} + \nabla\pi= -\operatorname{div}\left((\nabla\mathbf{d})^{\mathsf{T}}\nabla\mathbf{d}\right), & \text{in } (0,T) \times\mathcal{F}(t),\\
\operatorname{div}\mathbf{u}^{\mathcal{F}} = 0, & \text{in } (0,T) \times\mathcal{F}(t),\\
\partial_{t}\mathbf{d} + (\mathbf{u}^{\mathcal{F}} \cdot\nabla)\mathbf{d} - \Delta\mathbf{d} = \lvert\nabla\mathbf{d}\rvert^{2}\mathbf{d}, & \text{in } (0,T) \times\mathcal{F}(t),\\
\lvert\mathbf{d}\rvert= 1, & \text{in } (0,T) \times\mathcal{F}(t). \tag*{(3)}
\end{cases} ⎩ ⎨ ⎧ ∂ t u F + ( u F ⋅ ∇ ) u F − Δ u F + ∇ π = − div ( ( ∇ d ) T ∇ d ) , div u F = 0 , ∂ t d + ( u F ⋅ ∇ ) d − Δ d = ∣ ∇ d ∣ 2 d , ∣ d ∣ = 1 , in ( 0 , T ) × F ( t ) , in ( 0 , T ) × F ( t ) , in ( 0 , T ) × F ( t ) , in ( 0 , T ) × F ( t ) . (3)
For ∣ d ∣ = 1 \lvert\mathbf{d}\rvert= 1 ∣ d ∣ = 1 , the scalar ∣ ∇ d ∣ 2 \lvert\nabla\mathbf{d}\rvert^{2} ∣ ∇ d ∣ 2 is the Lagrange multiplier associated with the unit-length constraint in (3) 3 _3 3 , and − div ( ( ∇ d ) T ∇ d ) -\operatorname{div}\left((\nabla\mathbf{d})^{\mathsf{T}}\nabla\mathbf{d}\right) − div ( ( ∇ d ) T ∇ d ) in (3) 1 _1 1 is the Ericksen elastic force. The isotropic part of the elastic stress has been absorbed into the pressure.
Stress tensor
To formulate the fluid–structure coupling through Newton’s laws, we introduce the relevant stress tensors. Let D ( u F ) = 1 2 ( ∇ u F + ( ∇ u F ) T ) \mathbb{D}(\mathbf{u}^{\mathcal{F}}) = \frac{1}{2}\left(\nabla\mathbf{u}^{\mathcal{F}} + (\nabla\mathbf{u}^{\mathcal{F}})^{\mathsf{T}}\right) D ( u F ) = 2 1 ( ∇ u F + ( ∇ u F ) T ) denote the symmetric part of the gradient. The Newtonian stress tensor is given by
T N ( u F , π ) = 2 D ( u F ) − π I 3 , \mathbb{T}_{\mathrm{N}}(\mathbf{u}^{\mathcal{F}},\pi) = 2\mathbb{D}(\mathbf{u}^{\mathcal{F}}) - \pi\mathbb{I}_{3}, T N ( u F , π ) = 2 D ( u F ) − π I 3 ,
where I 3 \mathbb{I}_{3} I 3 denotes the 3 × 3 3 \times3 3 × 3 identity matrix. The Ericksen elastic stress is
T E ( d ) = − ( ∇ d ) T ∇ d . \mathbb{T}_{\mathrm{E}}(\mathbf{d}) = -(\nabla\mathbf{d})^{\mathsf{T}}\nabla\mathbf{d}. T E ( d ) = − ( ∇ d ) T ∇ d .
The total Cauchy stress is
T ( u F , π , d ) : = T N ( u F , π ) + T E ( d ) = 2 D ( u F ) − π I 3 − ( ∇ d ) T ∇ d . (4) \mathbb{T}(\mathbf{u}^{\mathcal{F}},\pi,\mathbf{d}) := \mathbb{T}_{\mathrm{N}}(\mathbf{u}^{\mathcal{F}},\pi) + \mathbb{T}_{\mathrm{E}}(\mathbf{d}) = 2\mathbb{D}(\mathbf{u}^{\mathcal{F}}) - \pi\mathbb{I}_{3} - (\nabla\mathbf{d})^{\mathsf{T}}\nabla\mathbf{d}. \tag*{(4)} T ( u F , π , d ) := T N ( u F , π ) + T E ( d ) = 2 D ( u F ) − π I 3 − ( ∇ d ) T ∇ d . (4)
Rigid body dynamics
With unit solid density, the mass and center of mass are
m = ∣ S ( t ) ∣ and h ( t ) = 1 m ∫ S ( t ) x d x , m = \lvert S(t)\rvert\quad\text{and} \quad\mathbf{h}(t) = \frac{1}{m}\int_{S(t)} \mathbf{x}\,\mathrm{d}x, m = ∣ S ( t )∣ and h ( t ) = m 1 ∫ S ( t ) x d x ,
and the moment of inertia is defined by
[ J ( t ) a ] ⋅ b = ∫ S ( t ) ( a × ( x − h ( t ) ) ) ⋅ ( b × ( x − h ( t ) ) ) d x , a , b ∈ R 3 . [\mathbb{J}(t)\mathbf{a}] \cdot\mathbf{b} = \int_{S(t)} \left(\mathbf{a} \times(\mathbf{x} - \mathbf{h}(t))\right) \cdot\left(\mathbf{b} \times(\mathbf{x} - \mathbf{h}(t))\right)\,\mathrm{d}x,\qquad\mathbf{a},\mathbf{b} \in\mathbb{R}^{3}. [ J ( t ) a ] ⋅ b = ∫ S ( t ) ( a × ( x − h ( t )) ) ⋅ ( b × ( x − h ( t )) ) d x , a , b ∈ R 3 .
By Newton’s laws of motion, the translational and rotational dynamics of the rigid body satisfy
m ℓ ′ ( t ) = − ∫ ∂ S ( t ) T ( u F , π , d ) ⋅ n d σ , m\boldsymbol{\ell}'(t) = -\int_{\partial S(t)} \mathbb{T}(\mathbf{u}^{\mathcal{F}},\pi,\mathbf{d}) \cdot\mathbf{n}\,\mathrm{d}\sigma, m ℓ ′ ( t ) = − ∫ ∂ S ( t ) T ( u F , π , d ) ⋅ n d σ ,
J ( t ) ω ′ ( t ) = J ( t ) ω ( t ) × ω ( t ) − ∫ ∂ S ( t ) ( x − h ( t ) ) × T ( u F , π , d ) ⋅ n d σ . (5) \mathbb{J}(t)\boldsymbol{\omega}'(t) = \mathbb{J}(t)\boldsymbol{\omega}(t) \times\boldsymbol{\omega}(t) - \int_{\partial S(t)} (\mathbf{x} - \mathbf{h}(t)) \times\mathbb{T}(\mathbf{u}^{\mathcal{F}},\pi,\mathbf{d}) \cdot\mathbf{n}\,\mathrm{d}\sigma. \tag*{(5)} J ( t ) ω ′ ( t ) = J ( t ) ω ( t ) × ω ( t ) − ∫ ∂ S ( t ) ( x − h ( t )) × T ( u F , π , d ) ⋅ n d σ . (5)
where n \mathbf{n} n denotes the unit outward normal to the fluid domain F ( t ) \mathcal{F}(t) F ( t ) on the interface ∂ S ( t ) \partial S(t) ∂ S ( t ) (equivalently, the unit inward normal to S ( t ) S(t) S ( t ) ), and d σ \mathrm{d}\sigma d σ is the surface measure.
2.1.5. Coupling, boundary, and initial conditions. At the fluid–solid interface and at the container boundary, we impose the no-slip conditions
u F = u S on ( 0 , T ) × ∂ S ( t ) and u F = 0 on ( 0 , T ) × ∂ Ω . u^{F}=u^{S}\quad\text{on }(0,T)\times\partial S(t)\quad\text{and}\quad u^{F}=0\quad\text{on }(0,T)\times\partial\Omega. u F = u S on ( 0 , T ) × ∂ S ( t ) and u F = 0 on ( 0 , T ) × ∂ Ω.
We denote the upper hemisphere and its closed convex hull respectively by
S + 2 : = { f = ( f 1 , f 2 , f 3 ) ∈ R 3 : ∣ f ∣ = 1 , f 3 ≥ 0 } , S^{2}_{+}:=\{f=(f_{1},f_{2},f_{3})\in\mathbb{R}^{3}:|f|=1,\ f_{3}\geq0\}, S + 2 := { f = ( f 1 , f 2 , f 3 ) ∈ R 3 : ∣ f ∣ = 1 , f 3 ≥ 0 } ,
K + : = { f ∈ R 3 : ∣ f ∣ ≤ 1 , f 3 ≥ 0 } . (6) K_{+}:=\{f\in\mathbb{R}^{3}:|f|\leq1,\ f_{3}\geq0\}. \tag*{(6)} K + := { f ∈ R 3 : ∣ f ∣ ≤ 1 , f 3 ≥ 0 } . (6)
The anchoring is described by a single finite-energy map d 0 ∈ H 1 ( Ω ; S + 2 ) d_{0}\in H^{1}(\Omega;S^{2}_{+}) d 0 ∈ H 1 ( Ω ; S + 2 ) which, as in [39 ] , serves both as initial director and as Dirichlet datum. Its traces on ∂ Ω \partial\Omega ∂ Ω and ∂ S 0 \partial S_{0} ∂ S 0 are understood in H 1 / 2 H^{1/2} H 1/2 . Only d 0 ∣ F 0 d_{0}|_{F_{0}} d 0 ∣ F 0 is the physical initial director. The values of d 0 d_{0} d 0 on S 0 S_{0} S 0 prescribe the inner boundary datum in material coordinates and define the extension of the director through the solid. The Dirichlet energy of this extension in the solid is invariant under the rigid motion and drops out of the energy balance. Anchoring maps in H 2 ( Ω ; S + 2 ) H^{2}(\Omega;S^{2}_{+}) H 2 ( Ω ; S + 2 ) are used only within the approximation scheme. This additional regularity is removed by the hemisphere-preserving approximation of Lemma A.1 . With the flow map Z t \boldsymbol{Z}_{t} Z t and its inverse ( Z t ) − 1 (\boldsymbol{Z}_{t})^{-1} ( Z t ) − 1 as introduced in (2) , the interface and boundary conditions for the director field are respectively given by
d ( t , x ) = d 0 ( ( Z t ) − 1 ( x ) ) , for ( t , x ) ∈ ( 0 , T ) × ∂ S ( t ) , d(t,x)=d_{0}\left((\boldsymbol{Z}_{t})^{-1}(x)\right),\quad\text{for }(t,x)\in(0,T)\times\partial S(t), d ( t , x ) = d 0 ( ( Z t ) − 1 ( x ) ) , for ( t , x ) ∈ ( 0 , T ) × ∂ S ( t ) ,
d ( t , x ) = d 0 ( x ) , for ( t , x ) ∈ ( 0 , T ) × ∂ Ω . d(t,x)=d_{0}(x),\quad\text{for }(t,x)\in(0,T)\times\partial\Omega. d ( t , x ) = d 0 ( x ) , for ( t , x ) ∈ ( 0 , T ) × ∂ Ω.
The anchoring values on ∂ S ( t ) \partial S(t) ∂ S ( t ) are transported with the body but not rotated. In particular, they remain in S + 2 S^{2}_{+} S + 2 , and the transported lifting used in § 3 is constant along particle paths. Finally, we prescribe the initial director in the fluid and the initial position of the body by
d ( 0 , ⋅ ) = d 0 in F 0 , h ( 0 ) = 0 , d(0,\cdot)=d_{0}\quad\text{in }F_{0},\qquad h(0)=0, d ( 0 , ⋅ ) = d 0 in F 0 , h ( 0 ) = 0 ,
and we use a unified initial velocity on the whole container,
u 0 ∈ H ( Ω ) , u 0 ( x ) = ℓ 0 + ω 0 × x for a.e. x ∈ S 0 . (7) u_{0}\in\mathcal{H}(\Omega),\qquad u_{0}(x)=\ell_{0}+\omega_{0}\times x\quad\text{for a.e. }x\in S_{0}. \tag*{(7)} u 0 ∈ H ( Ω ) , u 0 ( x ) = ℓ 0 + ω 0 × x for a.e. x ∈ S 0 . (7)
where the space H ( Ω ) \mathcal{H}(\Omega) H ( Ω ) will be made precise in (8) below. Thus the physical initial fluid velocity is u 0 ∣ F 0 u_{0}|_{F_{0}} u 0 ∣ F 0 , while the restriction to the solid determines the initial translational and angular velocities ℓ 0 , ω 0 \ell_{0},\omega_{0} ℓ 0 , ω 0 uniquely. In particular, no boundary trace of the finite-energy datum u 0 u_{0} u 0 is assumed. If one additionally assumes H 1 H^{1} H 1 -regularity of the initial velocity together with the classical boundary traces, then this unified formulation is consistent with the usual no-slip conditions on ∂ S 0 \partial S_{0} ∂ S 0 and ∂ Ω \partial\Omega ∂ Ω .
Weak solution and statements of main results
We first introduce the function spaces and the notion of weak solution for the constrained system on an arbitrary bounded open set (Definition 2.1 ), which governs the evolution after contact and whose spaces are reused later, and then the weak formulation of the coupled fluid–rigid-body problem before contact (Definition 2.2 ). Our main results are Theorem 2.3 on global existence through wall contact, Proposition 2.4 on the impact law, and Theorem 2.6 on existence on arbitrary bounded open sets.
We use a whole-container formulation for the velocity. Let
H ( Ω ) : = { v ∈ L 2 ( Ω ; R 3 ) : div v = 0 in D ′ ( Ω ) , v ⋅ n = 0 in H − 1 / 2 ( ∂ Ω ) } . (8) \mathcal{H}(\Omega):=\left\{v\in L^{2}(\Omega;\mathbb{R}^{3}):\operatorname{div}v=0\text{ in }\mathcal{D}'(\Omega),\ v\cdot n=0\text{ in }H^{-1/2}(\partial\Omega)\right\}. \tag*{(8)} H ( Ω ) := { v ∈ L 2 ( Ω ; R 3 ) : div v = 0 in D ′ ( Ω ) , v ⋅ n = 0 in H − 1/2 ( ∂ Ω ) } . (8)
and set
V ( Ω ) : = { v ∈ H 0 1 ( Ω ; R 3 ) : div v = 0 } . \mathcal{V}(\Omega):=\left\{v\in H^{1}_{0}(\Omega;\mathbb{R}^{3}):\operatorname{div}v=0\right\}. V ( Ω ) := { v ∈ H 0 1 ( Ω ; R 3 ) : div v = 0 } .
For any connected open set S ⊂ Ω \mathcal{S}\subset\Omega S ⊂ Ω representing a rigid configuration, set
H S ( Ω ) : = { v ∈ H ( Ω ) : D ( v ) = 0 in D ′ ( S ) } . (9) \mathcal{H}_{\mathcal{S}}(\Omega):=\left\{v\in\mathcal{H}(\Omega):\mathcal{D}(v)=0\text{ in }\mathcal{D}'(\mathcal{S})\right\}. \tag*{(9)} H S ( Ω ) := { v ∈ H ( Ω ) : D ( v ) = 0 in D ′ ( S ) } . (9)
The last condition is meaningful for v ∈ L 2 ( Ω ; R 3 ) v\in L^{2}(\Omega;\mathbb{R}^{3}) v ∈ L 2 ( Ω ; R 3 ) in the distributional sense and, because S \mathcal{S} S is connected, implies that the restriction of v v v to S \mathcal{S} S is a rigid velocity. In particular, for u 0 ∈ H S 0 ( Ω ) u_{0}\in\mathcal{H}_{S_{0}}(\Omega) u 0 ∈ H S 0 ( Ω ) , the initial translational and angular velocities ℓ 0 \ell_0 ℓ 0 and ω 0 \omega_0 ω 0 appearing in (7) are the unique coefficients of this rigid restriction.
For any bounded open set G ⊂ R 3 G \subset\mathbb{R}^3 G ⊂ R 3 , denote
D G : = { v ∈ C c ∞ ( G ; R 3 ) : div v = 0 } , H G : = D G ‾ L 2 ( G ) , V G : = D G ‾ H 0 1 ( G ) . (10) \mathcal{D}_G := \{v \in C^\infty_c(G;\mathbb{R}^3) : \operatorname{div}v = 0\}, \qquad\mathcal{H}_G := \overline{\mathcal{D}_G}^{L^2(G)}, \qquad\mathcal{V}_G := \overline{\mathcal{D}_G}^{H^1_0(G)}. \tag*{(10)} D G := { v ∈ C c ∞ ( G ; R 3 ) : div v = 0 } , H G := D G L 2 ( G ) , V G := D G H 0 1 ( G ) . (10)
These definitions require no regularity of ∂ G \partial G ∂ G and will be used both for the residual contact domain and for the arbitrary-open-set existence theorem below. For given 0 < T < ∞ 0<T<\infty 0 < T < ∞ , define
X G : = { ψ ∈ H 1 ( 0 , T ; L 2 ( G ; R 3 ) ) ∩ L 2 ( 0 , T ; H 0 1 ( G ; R 3 ) ) : ψ ( T ) = 0 } . \mathcal{X}_G := \{\psi\in H^1(0,T;L^2(G;\mathbb{R}^3)) \cap L^2(0,T;H^1_0(G;\mathbb{R}^3)) : \psi(T)=0\}. X G := { ψ ∈ H 1 ( 0 , T ; L 2 ( G ; R 3 )) ∩ L 2 ( 0 , T ; H 0 1 ( G ; R 3 )) : ψ ( T ) = 0 } .
We consider the system restricted to the fixed domain G G G given by
{ ∂ t u + ( u ⋅ ∇ ) u − Δ u + ∇ π = − div ( ( ∇ d ) T ∇ d ) , in ( 0 , T ) × G , div u = 0 , in ( 0 , T ) × G , ∂ t d + ( u ⋅ ∇ ) d − Δ d = ∣ ∇ d ∣ 2 d , in ( 0 , T ) × G , ∣ d ∣ = 1 , in ( 0 , T ) × G . (11) \left\{
\begin{aligned}
\partial_t u+(u\cdot\nabla)u-\Delta u+\nabla\pi&= -\operatorname{div}\left((\nabla d)^{\mathsf{T}}\nabla d\right), &&\text{in }(0,T)\times G,\\
\operatorname{div}u&=0, &&\text{in }(0,T)\times G,\\
\partial_t d+(u\cdot\nabla)d-\Delta d&=|\nabla d|^2d, &&\text{in }(0,T)\times G,\\
|d|&=1, &&\text{in }(0,T)\times G.
\end{aligned}
\right. \tag*{(11)} ⎩ ⎨ ⎧ ∂ t u + ( u ⋅ ∇ ) u − Δ u + ∇ π div u ∂ t d + ( u ⋅ ∇ ) d − Δ d ∣ d ∣ = − div ( ( ∇ d ) T ∇ d ) , = 0 , = ∣∇ d ∣ 2 d , = 1 , in ( 0 , T ) × G , in ( 0 , T ) × G , in ( 0 , T ) × G , in ( 0 , T ) × G . (11)
Definition 2.1. Let G ⊂ R 3 G\subset\mathbb{R}^3 G ⊂ R 3 be bounded and open, u 0 ∈ H G u_0\in\mathcal{H}_G u 0 ∈ H G , and d 0 ∈ H 1 ( G ; S + 2 ) d_0\in H^1(G;\mathbb{S}^2_+) d 0 ∈ H 1 ( G ; S + 2 ) . A weak solution to (11) on [ 0 , T ] [0,T] [ 0 , T ] with initial data ( u 0 , d 0 ) (u_0,d_0) ( u 0 , d 0 ) is a pair ( u , d ) (u,d) ( u , d ) with
u ∈ L ∞ ( 0 , T ; H G ) ∩ L 2 ( 0 , T ; V G ) , u\in L^\infty(0,T;\mathcal{H}_G)\cap L^2(0,T;\mathcal{V}_G), u ∈ L ∞ ( 0 , T ; H G ) ∩ L 2 ( 0 , T ; V G ) , d ∈ L ∞ ( 0 , T ; H 1 ( G ; S + 2 ) ) , d − d 0 ∈ L ∞ ( 0 , T ; H 0 1 ( G ; R 3 ) ) , d\in L^\infty(0,T;H^1(G;\mathbb{S}^2_+)), \qquad d-d_0\in L^\infty(0,T;H^1_0(G;\mathbb{R}^3)), d ∈ L ∞ ( 0 , T ; H 1 ( G ; S + 2 )) , d − d 0 ∈ L ∞ ( 0 , T ; H 0 1 ( G ; R 3 )) , whose tension τ : = Δ d + ∣ ∇ d ∣ 2 d \tau:=\Delta d+|\nabla d|^2d τ := Δ d + ∣∇ d ∣ 2 d , defined in the sense of distributions, belongs to L 2 ( ( 0 , T ) × G ; R 3 ) L^2((0,T)\times G;\mathbb{R}^3) L 2 (( 0 , T ) × G ; R 3 ) , such that the following hold.
(i) For every ζ ∈ C c 1 ( [ 0 , T ) ; D G ) \zeta\in C^1_c([0,T);\mathcal{D}_G) ζ ∈ C c 1 ([ 0 , T ) ; D G ) and every ψ ∈ X G \psi\in\mathcal{X}_G ψ ∈ X G ,
∫ 0 T ∫ G [ − u ⋅ ∂ t ζ − ( u ⊗ u ) : ∇ ζ + 2 D ( u ) : D ( ζ ) ] d x d t = ∫ G u 0 ⋅ ζ ( 0 ) d x + ∫ 0 T ∫ G ( ∇ d ) T ∇ d : ∇ ζ d x d t , \int_0^T\int_G[-u\cdot\partial_t\zeta-(u\otimes u):\nabla\zeta+2D(u):D(\zeta)]\,\mathrm{d}x\,\mathrm{d}t
=
\int_G u_0\cdot\zeta(0)\,\mathrm{d}x+\int_0^T\int_G(\nabla d)^{\mathsf{T}}\nabla d:\nabla\zeta\,\mathrm{d}x\,\mathrm{d}t, ∫ 0 T ∫ G [ − u ⋅ ∂ t ζ − ( u ⊗ u ) : ∇ ζ + 2 D ( u ) : D ( ζ )] d x d t = ∫ G u 0 ⋅ ζ ( 0 ) d x + ∫ 0 T ∫ G ( ∇ d ) T ∇ d : ∇ ζ d x d t , ∫ 0 T ∫ G [ − d ⋅ ∂ t ψ + ( u ⋅ ∇ ) d ⋅ ψ − τ ⋅ ψ ] d x d t = ∫ G d 0 ⋅ ψ ( 0 ) d x . \int_0^T\int_G[-d\cdot\partial_t\psi+(u\cdot\nabla)d\cdot\psi-\tau\cdot\psi]\,\mathrm{d}x\,\mathrm{d}t
=
\int_G d_0\cdot\psi(0)\,\mathrm{d}x. ∫ 0 T ∫ G [ − d ⋅ ∂ t ψ + ( u ⋅ ∇ ) d ⋅ ψ − τ ⋅ ψ ] d x d t = ∫ G d 0 ⋅ ψ ( 0 ) d x . (ii) For a.e. t ∈ ( 0 , T ) t\in(0,T) t ∈ ( 0 , T ) ,
1 2 ∥ u ( t ) ∥ L 2 ( G ) 2 + 1 2 ∥ ∇ d ( t ) ∥ L 2 ( G ) 2 + 2 ∫ 0 t ∫ G ∣ D ( u ) ∣ 2 d x d s + ∫ 0 t ∫ G ∣ τ ∣ 2 d x d s ≤ 1 2 ∥ u 0 ∥ L 2 ( G ) 2 + 1 2 ∥ ∇ d 0 ∥ L 2 ( G ) 2 . (12) \frac{1}{2}\|u(t)\|_{L^2(G)}^2+\frac{1}{2}\|\nabla d(t)\|_{L^2(G)}^2+2\int_0^t\int_G|D(u)|^2\,\mathrm{d}x\,\mathrm{d}s
+\int_0^t\int_G|\tau|^2\,\mathrm{d}x\,\mathrm{d}s
\leq
\frac{1}{2}\|u_0\|_{L^2(G)}^2+\frac{1}{2}\|\nabla d_0\|_{L^2(G)}^2. \tag*{(12)} 2 1 ∥ u ( t ) ∥ L 2 ( G ) 2 + 2 1 ∥∇ d ( t ) ∥ L 2 ( G ) 2 + 2 ∫ 0 t ∫ G ∣ D ( u ) ∣ 2 d x d s + ∫ 0 t ∫ G ∣ τ ∣ 2 d x d s ≤ 2 1 ∥ u 0 ∥ L 2 ( G ) 2 + 2 1 ∥∇ d 0 ∥ L 2 ( G ) 2 . (12)
A global weak solution is a pair defined on [ 0 , ∞ ) [0,\infty) [ 0 , ∞ ) which is a weak solution on [ 0 , T ] [0,T] [ 0 , T ] for every T > 0 T>0 T > 0 .
For a set A ⊂ Ω A\subset\Omega A ⊂ Ω and r > 0 r>0 r > 0 , write
A r : = { x ∈ Ω : dist ( x , A ) < r } . A_r:=\{x\in\Omega:\operatorname{dist}(x,A)<r\}. A r := { x ∈ Ω : dist ( x , A ) < r } .
Whenever a characteristic function χ \chi χ is used to encode a solid, an open set S S S is part of the data and we use the pointwise representative χ = 1 S \chi=1_S χ = 1 S . We then set S ( χ ) : = S S(\chi):=S S ( χ ) := S . Thus S ( χ ) r S(\chi)_r S ( χ ) r denotes the open r r r -collar of the represented solid. For an open set A ⊂ Ω A\subset\Omega A ⊂ Ω , define
K ( A ) : = { v ∈ V ( Ω ) : D ( v ) = 0 a.e. in A } . \mathcal{K}(A):=\{v\in\mathcal{V}(\Omega):D(v)=0\ \text{a.e. in }A\}. K ( A ) := { v ∈ V ( Ω ) : D ( v ) = 0 a.e. in A } .
For a represented characteristic function χ = 1 S ( χ ) \chi=1_{S(\chi)} χ = 1 S ( χ ) , set
K ( χ ) : = K ( S ( χ ) ) , K r ( χ ) : = K ( S ( χ ) r ) . \mathcal{K}(\chi):=\mathcal{K}(S(\chi)), \qquad\mathcal{K}_r(\chi):=\mathcal{K}(S(\chi)_r). K ( χ ) := K ( S ( χ )) , K r ( χ ) := K ( S ( χ ) r ) .
For a spacetime characteristic function φ \varphi φ , we write
L 2 ( 0 , T ; K ( φ ) ) : = { v ∈ L 2 ( 0 , T ; V ( Ω ) ) : φ D ( v ) = 0 a.e. in ( 0 , T ) × Ω } . L^2(0,T;\mathcal{K}(\varphi)) := \{v \in L^2(0,T;\mathcal{V}(\Omega)) : \varphi\mathbb{D}(v) = 0\ \text{a.e. in }(0,T) \times\Omega\}. L 2 ( 0 , T ; K ( φ )) := { v ∈ L 2 ( 0 , T ; V ( Ω )) : φ D ( v ) = 0 a.e. in ( 0 , T ) × Ω } .
The unified velocity is
u ( t , x ) = { u F ( t , x ) , x ∈ F ( t ) , u S ( t , x ) , x ∈ S ( t ) . u(t,x) =
\begin{cases}
u^{F}(t,x), & x \in\mathcal{F}(t),\\
u^{S}(t,x), & x \in S(t).
\end{cases} u ( t , x ) = { u F ( t , x ) , u S ( t , x ) , x ∈ F ( t ) , x ∈ S ( t ) .
For a rigid trajectory S ( t ) = h ( t ) + O ( t ) S 0 S(t) = h(t) + O(t)S_0 S ( t ) = h ( t ) + O ( t ) S 0 , recall that F ( t ) = Ω ∖ S ( t ) ‾ \mathcal{F}(t) = \Omega\setminus\overline{S(t)} F ( t ) = Ω ∖ S ( t ) is the corresponding fluid region, and introduce the fluid spacetime
Q F : = { ( t , x ) ∈ ( 0 , T ) × Ω : x ∈ F ( t ) } . Q_{\mathcal{F}} := \{(t,x) \in(0,T) \times\Omega: x \in\mathcal{F}(t)\}. Q F := {( t , x ) ∈ ( 0 , T ) × Ω : x ∈ F ( t )} .
The standard momentum test class is
T u : = { ζ ∈ C c 1 ( [ 0 , T ) ; C ∞ ( Ω ; R 3 ) ) : div ζ = 0 , D ( ζ ( t ) ) = 0 on S ( t ) for every t ∈ [ 0 , T ] } . (13) \mathcal{T}_u := \{\zeta\in C^1_c([0,T);C^\infty(\Omega;\mathbb{R}^3)) : \operatorname{div}\zeta= 0,\ \mathbb{D}(\zeta(t)) = 0\ \text{on }S(t)\ \text{for every }t \in[0,T]\}. \tag*{(13)} T u := { ζ ∈ C c 1 ([ 0 , T ) ; C ∞ ( Ω ; R 3 )) : div ζ = 0 , D ( ζ ( t )) = 0 on S ( t ) for every t ∈ [ 0 , T ]} . (13)
For the director equation, we use the molecular field and the moving energy test space
X d ( φ ) : = { ψ ∈ H 1 ( 0 , T ; L 2 ( Ω ; R 3 ) ) ∩ L 2 ( 0 , T ; H 0 1 ( Ω ; R 3 ) ) : ψ ( T ) = 0 , \mathcal{X}_d(\varphi) := \{\psi\in H^1(0,T;L^2(\Omega;\mathbb{R}^3)) \cap L^2(0,T;H^1_0(\Omega;\mathbb{R}^3)) : \psi(T) = 0, X d ( φ ) := { ψ ∈ H 1 ( 0 , T ; L 2 ( Ω ; R 3 )) ∩ L 2 ( 0 , T ; H 0 1 ( Ω ; R 3 )) : ψ ( T ) = 0 ,
ψ ( t ) = 0 a.e. in S ( t ) , ψ ( t ) ∣ F ( t ) ∈ H 0 1 ( F ( t ) ; R 3 ) for a.e. t } . (14) \psi(t) = 0\ \text{a.e. in }S(t),\quad\psi(t)|_{\mathcal{F}(t)} \in H^1_0(\mathcal{F}(t);\mathbb{R}^3)\ \text{for a.e. }t\}. \tag*{(14)} ψ ( t ) = 0 a.e. in S ( t ) , ψ ( t ) ∣ F ( t ) ∈ H 0 1 ( F ( t ) ; R 3 ) for a.e. t } . (14)
The smooth localized class
T d : = { ψ ∈ C c 1 ( [ 0 , T ) × Ω ; R 3 ) : supp ψ ∩ ( ( 0 , T ) × ∂ Ω ) = ∅ , supp ψ ⋐ Q F ∪ ( { 0 } × F 0 ) } (15) \mathcal{T}_d := \{\psi\in C^1_c([0,T) \times\Omega;\mathbb{R}^3) : \operatorname{supp}\psi\cap((0,T) \times\partial\Omega) = \varnothing,\ \operatorname{supp}\psi\Subset Q_{\mathcal{F}} \cup(\{0\} \times\mathcal{F}_0)\} \tag*{(15)} T d := { ψ ∈ C c 1 ([ 0 , T ) × Ω ; R 3 ) : supp ψ ∩ (( 0 , T ) × ∂ Ω ) = ∅ , supp ψ ⋐ Q F ∪ ({ 0 } × F 0 )} (15)
is dense in (14) (see Lemma 3.13 ). The tension formulation below is well defined on this larger energy test space, whereas the formulation containing ∣ ∇ d ∣ 2 d |\nabla d|^2d ∣∇ d ∣ 2 d is used only for localized smooth tests.
The pre-contact weak solution is defined as follows.
Definition 2.2. Let u 0 ∈ H S 0 ( Ω ) u_0 \in\mathcal{H}_{S_0}(\Omega) u 0 ∈ H S 0 ( Ω ) , d 0 ∈ H 1 ( Ω ; S + 2 ) d_0 \in H^1(\Omega;\mathbb{S}^2_+) d 0 ∈ H 1 ( Ω ; S + 2 ) , and φ 0 = 1 S 0 \varphi_0 = \mathbb{1}_{S_0} φ 0 = 1 S 0 . A pre-contact weak solution of (3) coupled to (5) , with the stated boundary and initial conditions, on [ 0 , T ] [0,T] [ 0 , T ] is a triple ( u , d , φ ) (u,d,\varphi) ( u , d , φ ) together with a rigid trajectory satisfying
h ∈ W 1 , ∞ ( 0 , T ; R 3 ) , O ∈ W 1 , ∞ ( 0 , T ; R 3 × 3 ) , O ( t ) ∈ SO ( 3 ) for every t , h \in W^{1,\infty}(0,T;\mathbb{R}^3),\qquad O \in W^{1,\infty}(0,T;\mathbb{R}^{3 \times3}),\qquad O(t) \in\operatorname{SO}(3)\ \text{for every }t, h ∈ W 1 , ∞ ( 0 , T ; R 3 ) , O ∈ W 1 , ∞ ( 0 , T ; R 3 × 3 ) , O ( t ) ∈ SO ( 3 ) for every t , such that
h ( 0 ) = 0 , O ( 0 ) = I 3 , φ ( t ) = 1 S ( t ) , S ( t ) = h ( t ) + O ( t ) S 0 , S ( t ) ‾ ⊂ Ω , h(0) = 0,\qquad O(0) = \mathbb{I}_3,\qquad\varphi(t) = \mathbb{1}_{S(t)},\qquad S(t) = h(t) + O(t)S_0,\qquad\overline{S(t)} \subset\Omega, h ( 0 ) = 0 , O ( 0 ) = I 3 , φ ( t ) = 1 S ( t ) , S ( t ) = h ( t ) + O ( t ) S 0 , S ( t ) ⊂ Ω , u ∈ L ∞ ( 0 , T ; H ( Ω ) ) ∩ L 2 ( 0 , T ; K ( φ ) ) , u \in L^\infty(0,T;\mathcal{H}(\Omega)) \cap L^2(0,T;\mathcal{K}(\varphi)), u ∈ L ∞ ( 0 , T ; H ( Ω )) ∩ L 2 ( 0 , T ; K ( φ )) , d ∈ L ∞ ( 0 , T ; H 1 ( F ( t ) ; S + 2 ) ) , d \in L^\infty(0,T;H^1(\mathcal{F}(t);\mathbb{S}^2_+)), d ∈ L ∞ ( 0 , T ; H 1 ( F ( t ) ; S + 2 )) , τ : = Δ d + ∣ ∇ d ∣ 2 d ∈ L 2 ( Q F ; R 3 ) in the sense of distributions , \tau:= \Delta d + |\nabla d|^2d \in L^2(Q_{\mathcal{F}};\mathbb{R}^3)\quad\text{in the sense of distributions}, τ := Δ d + ∣∇ d ∣ 2 d ∈ L 2 ( Q F ; R 3 ) in the sense of distributions , and the following hold.
(i) The restriction of u u u to S ( t ) S(t) S ( t ) is the rigid velocity
u ( t , x ) = h ′ ( t ) + ω ( t ) × ( x − h ( t ) ) for a.e. x ∈ S ( t ) , u(t,x) = h'(t) + \omega(t) \times(x-h(t))\quad\text{for a.e. }x \in S(t), u ( t , x ) = h ′ ( t ) + ω ( t ) × ( x − h ( t )) for a.e. x ∈ S ( t ) , where O ′ = m O O' = mO O ′ = m O and m x = ω × x mx = \omega\times x m x = ω × x .
(ii) The director satisfies the strong anchoring condition in the trace sense,
Tr ∂ Ω d ( t ) = d 0 , Tr ∂ S ( t ) d ( t , x ) = d 0 ( O ( t ) T ( x − h ( t ) ) ) for a.e. t . (16) \operatorname{Tr}_{\partial\Omega}d(t) = d_0,\qquad\operatorname{Tr}_{\partial S(t)}d(t,x) = d_0\left(O(t)^\mathsf{T}(x-h(t))\right)\quad\text{for a.e. }t. \tag*{(16)} Tr ∂ Ω d ( t ) = d 0 , Tr ∂ S ( t ) d ( t , x ) = d 0 ( O ( t ) T ( x − h ( t )) ) for a.e. t . (16)
(iii) For every ζ ∈ T u \zeta\in\mathcal{T}_u ζ ∈ T u , ψ ∈ X d ( φ ) \psi\in\mathcal{X}_d(\varphi) ψ ∈ X d ( φ ) , and η ∈ C c 1 ( [ 0 , T ) × Ω ) \eta\in C^1_c([0,T) \times\Omega) η ∈ C c 1 ([ 0 , T ) × Ω ) ,
∫ 0 T ∫ Ω [ − u ⋅ ∂ t ζ − ( u ⊗ u ) : ∇ ζ + 2 D ( u ) : D ( ζ ) ] d x d t = ∫ Ω u 0 ⋅ ζ ( 0 ) d x + ∫ 0 T ∫ F ( t ) ( ∇ d ) T ∇ d : ∇ ζ d x d t , (17) \int_0^T \int_\Omega\left[-u \cdot\partial_t\zeta- (u \otimes u) : \nabla\zeta+ 2D(u) : D(\zeta)\right]\,\mathrm{d}x\,\mathrm{d}t
= \int_\Omega u_0 \cdot\zeta(0)\,\mathrm{d}x + \int_0^T \int_{\mathcal{F}(t)} (\nabla d)^{\mathsf{T}}\nabla d : \nabla\zeta\,\mathrm{d}x\,\mathrm{d}t, \tag*{(17)} ∫ 0 T ∫ Ω [ − u ⋅ ∂ t ζ − ( u ⊗ u ) : ∇ ζ + 2 D ( u ) : D ( ζ ) ] d x d t = ∫ Ω u 0 ⋅ ζ ( 0 ) d x + ∫ 0 T ∫ F ( t ) ( ∇ d ) T ∇ d : ∇ ζ d x d t , (17)
∫ 0 T ∫ F ( t ) [ − d ⋅ ∂ t ψ + ( u ⋅ ∇ ) d ⋅ ψ − τ ⋅ ψ ] d x d t = ∫ F 0 d 0 ⋅ ψ ( 0 ) d x , (18) \int_0^T \int_{\mathcal{F}(t)} \left[-d \cdot\partial_t\psi+ (u \cdot\nabla)d \cdot\psi- \tau\cdot\psi\right]\,\mathrm{d}x\,\mathrm{d}t = \int_{\mathcal{F}_0} d_0 \cdot\psi(0)\,\mathrm{d}x, \tag*{(18)} ∫ 0 T ∫ F ( t ) [ − d ⋅ ∂ t ψ + ( u ⋅ ∇ ) d ⋅ ψ − τ ⋅ ψ ] d x d t = ∫ F 0 d 0 ⋅ ψ ( 0 ) d x , (18)
∫ 0 T ∫ Ω φ [ ∂ t η + u ⋅ ∇ η ] d x d t = − ∫ Ω φ 0 η ( 0 ) d x . (19) \int_0^T \int_\Omega\varphi\left[\partial_t\eta+ u \cdot\nabla\eta\right]\,\mathrm{d}x\,\mathrm{d}t = -\int_\Omega\varphi_0\eta(0)\,\mathrm{d}x. \tag*{(19)} ∫ 0 T ∫ Ω φ [ ∂ t η + u ⋅ ∇ η ] d x d t = − ∫ Ω φ 0 η ( 0 ) d x . (19)
(iv) Finally, for a.e. t ∈ ( 0 , T ) t \in(0,T) t ∈ ( 0 , T ) , we have the energy inequality
E ( t ) + 2 ∫ 0 t ∫ Ω ∣ D ( u ) ∣ 2 d x d s + ∫ 0 t ∫ F ( s ) ∣ τ ∣ 2 d x d s ≤ E ( 0 ) , (20) E(t) + 2\int_0^t \int_\Omega|D(u)|^2\,\mathrm{d}x\,\mathrm{d}s + \int_0^t \int_{\mathcal{F}(s)} |\tau|^2\,\mathrm{d}x\,\mathrm{d}s \le E(0), \tag*{(20)} E ( t ) + 2 ∫ 0 t ∫ Ω ∣ D ( u ) ∣ 2 d x d s + ∫ 0 t ∫ F ( s ) ∣ τ ∣ 2 d x d s ≤ E ( 0 ) , (20)
where
E ( t ) : = 1 2 ∫ Ω ∣ u ( t ) ∣ 2 d x + 1 2 ∫ F ( t ) ∣ ∇ d ( t ) ∣ 2 d x for a.e. t ∈ ( 0 , T ) , E(t) := \frac{1}{2}\int_\Omega|u(t)|^2\,\mathrm{d}x + \frac{1}{2}\int_{\mathcal{F}(t)} |\nabla d(t)|^2\,\mathrm{d}x \quad\text{for a.e. } t \in(0,T), E ( t ) := 2 1 ∫ Ω ∣ u ( t ) ∣ 2 d x + 2 1 ∫ F ( t ) ∣∇ d ( t ) ∣ 2 d x for a.e. t ∈ ( 0 , T ) ,
E ( 0 ) : = 1 2 ∫ Ω ∣ u 0 ∣ 2 d x + 1 2 ∫ F 0 ∣ ∇ d 0 ∣ 2 d x . E(0) := \frac{1}{2}\int_\Omega|u_0|^2\,\mathrm{d}x + \frac{1}{2}\int_{\mathcal{F}_0} |\nabla d_0|^2\,\mathrm{d}x. E ( 0 ) := 2 1 ∫ Ω ∣ u 0 ∣ 2 d x + 2 1 ∫ F 0 ∣∇ d 0 ∣ 2 d x .
Note that h ( 0 ) = 0 h(0)=0 h ( 0 ) = 0 in the above definition means precisely that the center of mass at time zero is the origin. For ψ ∈ T d \psi\in\mathcal{T}_d ψ ∈ T d , (18) is equivalently
∫ 0 T ∫ F ( t ) [ − d ⋅ ∂ t ψ + ( u ⋅ ∇ ) d ⋅ ψ + ∇ d : ∇ ψ − ∣ ∇ d ∣ 2 d ⋅ ψ ] d x d t = ∫ F 0 d 0 ⋅ ψ ( 0 ) d x . \int_0^T \int_{\mathcal{F}(t)} \left[-d \cdot\partial_t\psi+ (u \cdot\nabla)d \cdot\psi+ \nabla d : \nabla\psi- |\nabla d|^2 d \cdot\psi\right]\,\mathrm{d}x\,\mathrm{d}t = \int_{\mathcal{F}_0} d_0 \cdot\psi(0)\,\mathrm{d}x. ∫ 0 T ∫ F ( t ) [ − d ⋅ ∂ t ψ + ( u ⋅ ∇ ) d ⋅ ψ + ∇ d : ∇ ψ − ∣∇ d ∣ 2 d ⋅ ψ ] d x d t = ∫ F 0 d 0 ⋅ ψ ( 0 ) d x .
For data satisfying the hypotheses below, the initial distance to the wall and the initial energy E ( 0 ) E(0) E ( 0 ) in (20) are denoted by
δ 0 : = dist ( S 0 , ∂ Ω ) , E 0 : = 1 2 ∥ u 0 ∥ L 2 ( Ω ) 2 + 1 2 ∥ ∇ d 0 ∥ L 2 ( F 0 ) 2 . (21) \delta_0 := \operatorname{dist}(\mathcal{S}_0,\partial\Omega), \qquad E_0 := \frac{1}{2}\|u_0\|_{L^2(\Omega)}^2 + \frac{1}{2}\|\nabla d_0\|_{L^2(\mathcal{F}_0)}^2. \tag*{(21)} δ 0 := dist ( S 0 , ∂ Ω ) , E 0 := 2 1 ∥ u 0 ∥ L 2 ( Ω ) 2 + 2 1 ∥∇ d 0 ∥ L 2 ( F 0 ) 2 . (21)
For a bounded open set G G G and ( v , q ) ∈ L 2 ( G ; R 3 ) × H 1 ( G ; R 3 ) (v,q) \in L^2(G;\mathbb{R}^3) \times H^1(G;\mathbb{R}^3) ( v , q ) ∈ L 2 ( G ; R 3 ) × H 1 ( G ; R 3 ) , write
E G ( v , q ) : = 1 2 ∥ v ∥ L 2 ( G ) 2 + 1 2 ∥ ∇ q ∥ L 2 ( G ) 2 . \mathcal{E}_G(v,q) := \frac{1}{2}\|v\|_{L^2(G)}^2 + \frac{1}{2}\|\nabla q\|_{L^2(G)}^2. E G ( v , q ) := 2 1 ∥ v ∥ L 2 ( G ) 2 + 2 1 ∥∇ q ∥ L 2 ( G ) 2 .
For a collision-free rigid trajectory and a director satisfying the material-coordinate anchoring, we denote by d ˉ \bar d d ˉ the material extension
d ˉ ( t , x ) : = { d ( t , x ) , x ∈ F ( t ) , d 0 ( O ( t ) T ( x − h ( t ) ) ) , x ∈ S ( t ) . (22) \bar d(t,x) :=
\begin{cases}
d(t,x), & x \in\mathcal{F}(t),\\
d_0(O(t)^{\mathsf{T}}(x-h(t))), & x \in\mathcal{S}(t).
\end{cases} \tag*{(22)} d ˉ ( t , x ) := { d ( t , x ) , d 0 ( O ( t ) T ( x − h ( t ))) , x ∈ F ( t ) , x ∈ S ( t ) . (22)
The main result provides global weak solutions under the perfectly inelastic sticking law.
Theorem 2.3. Assume ∂ Ω , ∂ S 0 ∈ C 2 \partial\Omega,\partial\mathcal{S}_0 \in C^2 ∂ Ω , ∂ S 0 ∈ C 2 , S 0 \mathcal{S}_0 S 0 is connected, δ 0 > 0 \delta_0 > 0 δ 0 > 0 , and
u 0 ∈ H S 0 ( Ω ) , d 0 ∈ H 1 ( Ω ; S + 2 ) . u_0 \in\mathcal{H}_{\mathcal{S}_0}(\Omega), \qquad d_0 \in H^1(\Omega;\mathbb{S}^2_+). u 0 ∈ H S 0 ( Ω ) , d 0 ∈ H 1 ( Ω ; S + 2 ) . Then there exist T c o l ∈ ( 0 , ∞ ] T_{\mathrm{col}} \in(0,\infty] T col ∈ ( 0 , ∞ ] , a rigid trajectory S ( t ) = h ( t ) + O ( t ) S 0 \mathcal{S}(t)=h(t)+O(t)\mathcal{S}_0 S ( t ) = h ( t ) + O ( t ) S 0 and a triple ( u , d , φ ) (u,d,\varphi) ( u , d , φ ) on [ 0 , T c o l ) [0,T_{\mathrm{col}}) [ 0 , T col ) , and, if T c o l < ∞ T_{\mathrm{col}}<\infty T col < ∞ , a pair ( u , d ) (u,d) ( u , d ) on [ T c o l , ∞ ) [T_{\mathrm{col}},\infty) [ T col , ∞ ) , denoted by the same symbols, with the following properties.
(i) For every T < T c o l T<T_{\mathrm{col}} T < T col , the triple ( u , d , φ ) (u,d,\varphi) ( u , d , φ ) is a weak solution on [ 0 , T ] [0,T] [ 0 , T ] in the sense of Definition 2.2 , and dist ( S ( t ) , ∂ Ω ) > 0 \operatorname{dist}(\mathcal{S}(t),\partial\Omega)>0 dist ( S ( t ) , ∂ Ω ) > 0 for 0 ≤ t < T c o l 0 \le t < T_{\mathrm{col}} 0 ≤ t < T col .
(ii) If T c o l < ∞ T_{\mathrm{col}} < \infty T col < ∞ , the rigid motion extends continuously to T c o l T_{\mathrm{col}} T col and, with S c : = S ( T c o l ) S_c := S(T_{\mathrm{col}}) S c := S ( T col ) and Ω c : = Ω ∖ S c ‾ \Omega_c := \Omega\setminus\overline{S_c} Ω c := Ω ∖ S c ,
dist ( S c , ∂ Ω ) = 0. (23) \operatorname{dist}(S_c,\partial\Omega)=0. \tag*{(23)} dist ( S c , ∂ Ω ) = 0. (23)
As t → T c o l t \to T_{\mathrm{col}} t → T col , the material extension (22) converges to some d c ∈ H 1 ( Ω ; S + 2 ) d_c \in\mathrm{H}^1(\Omega;\mathbb{S}^{2}_{+}) d c ∈ H 1 ( Ω ; S + 2 ) , strongly in L 2 ( Ω ) \mathrm{L}^2(\Omega) L 2 ( Ω ) and weakly in H 1 ( Ω ) \mathrm{H}^1(\Omega) H 1 ( Ω ) , and there is a unique u c + ∈ H Ω c \mathbf{u}_c^+ \in\mathcal{H}_{\Omega_c} u c + ∈ H Ω c such that ( u ( t ) , ζ ) L 2 ( Ω ) → ( u c + , ζ ) L 2 ( Ω c ) (u(t),\zeta)_{\mathrm{L}^2(\Omega)} \to(\mathbf{u}_c^+,\zeta)_{\mathrm{L}^2(\Omega_c)} ( u ( t ) , ζ ) L 2 ( Ω ) → ( u c + , ζ ) L 2 ( Ω c ) for every ζ ∈ D Ω c \zeta\in\mathcal{D}_{\Omega_c} ζ ∈ D Ω c .
(iii) If T c o l < ∞ T_{\mathrm{col}} < \infty T col < ∞ , the body stays at S c S_c S c for t ≥ T c o l t \ge T_{\mathrm{col}} t ≥ T col , the velocity vanishes in S c S_c S c , and ( u , d ) ( T c o l + ⋅ ) ∣ Ω c (u,d)(T_{\mathrm{col}}+\cdot)|_{\Omega_c} ( u , d ) ( T col + ⋅ ) ∣ Ω c is a global weak solution on Ω c \Omega_c Ω c in the sense of Definition 2.1 with initial data ( u c + , d c ∣ Ω c ) (\mathbf{u}_c^+,d_c|_{\Omega_c}) ( u c + , d c ∣ Ω c ) . Moreover, with F ( s ) : = Ω c \mathcal{F}(s) := \Omega_c F ( s ) := Ω c for s ≥ T c o l s \ge T_{\mathrm{col}} s ≥ T col and the dissipation
D ( 0 , t ) : = 2 ∫ 0 t ∫ Ω ∣ D ( u ) ∣ 2 d x d s + ∫ 0 t ∫ F ( s ) ∣ τ ∣ 2 d x d s , \mathcal{D}(0,t):=2\int_0^t\int_{\Omega}|\mathbb{D}(u)|^2\,dx\,ds+\int_0^t\int_{\mathcal{F}(s)}|\tau|^2\,dx\,ds, D ( 0 , t ) := 2 ∫ 0 t ∫ Ω ∣ D ( u ) ∣ 2 d x d s + ∫ 0 t ∫ F ( s ) ∣ τ ∣ 2 d x d s ,
we have
E Ω c ( u ( t ) , d ( t ) ) + D ( 0 , t ) ≤ E 0 for a.e. t > T c o l . (24) \mathcal{E}_{\Omega_c}(u(t),d(t))+\mathcal{D}(0,t)\le E_0\qquad\text{for a.e. }t>T_{\mathrm{col}}. \tag*{(24)} E Ω c ( u ( t ) , d ( t )) + D ( 0 , t ) ≤ E 0 for a.e. t > T col . (24)
The first contact time is bounded from below explicitly. By Corollary 2.10 in § 2.3 , T c o l ≥ T e d T_{\mathrm{col}}\ge T_{\mathrm{ed}} T col ≥ T ed , where T e d > 0 T_{\mathrm{ed}}>0 T ed > 0 , defined in (31) , depends only on δ 0 \delta_0 δ 0 , E 0 E_0 E 0 , Ω \Omega Ω , and S 0 S_0 S 0 .
If contact occurs, the impact law takes the following form. For a connected rigid configuration S ⊂ Ω S\subset\Omega S ⊂ Ω , let G S : = Ω ∖ S ‾ G_S:=\Omega\setminus\overline{S} G S := Ω ∖ S , let P H G S P_{\mathcal{H}_{G_S}} P H G S be the L 2 ( G S ) \mathrm{L}^2(G_S) L 2 ( G S ) -orthogonal projection onto H G S \mathcal{H}_{G_S} H G S , and define the impact dissipation of v ∈ H S ( Ω ) v\in\mathcal{H}_S(\Omega) v ∈ H S ( Ω ) by
J i m p ( S , v ) : = 1 2 ∥ v ∥ L 2 ( S ) 2 + 1 2 ∥ v ∣ G S − P H G S ( v ∣ G S ) ∥ L 2 ( G S ) 2 . (25) \mathcal{J}_{\mathrm{imp}}(S,v):=\frac{1}{2}\|v\|_{\mathrm{L}^2(S)}^2+\frac{1}{2}\|v|_{G_S}-P_{\mathcal{H}_{G_S}}(v|_{G_S})\|_{\mathrm{L}^2(G_S)}^2. \tag*{(25)} J imp ( S , v ) := 2 1 ∥ v ∥ L 2 ( S ) 2 + 2 1 ∥ v ∣ G S − P H G S ( v ∣ G S ) ∥ L 2 ( G S ) 2 . (25)
It is the kinetic energy of the body plus the kinetic energy of the part of the fluid velocity that is not admissible in the residual fluid domain G S G_S G S .
Proposition 2.4. Let T c o l < ∞ T_{\mathrm{col}}<\infty T col < ∞ in Theorem 2.3 , and let u ( t j ) ⇀ u c − u(t_j)\rightharpoonup\mathbf{u}_c^- u ( t j ) ⇀ u c − in L 2 ( Ω ) \mathrm{L}^2(\Omega) L 2 ( Ω ) along a sequence t j → T c o l t_j\to T_{\mathrm{col}} t j → T col in the full-measure set of times at which u ( t j ) u(t_j) u ( t j ) is rigid on S ( t j ) S(t_j) S ( t j ) and the energy inequality (20) holds. Then
u c − ∈ H S c ( Ω ) and u c + = P H Ω c ( u c − ∣ Ω c ) , (26) \mathbf{u}_c^-\in\mathcal{H}_{S_c}(\Omega)\quad\text{and}\quad\mathbf{u}_c^+=P_{\mathcal{H}_{\Omega_c}}(\mathbf{u}_c^-|_{\Omega_c}), \tag*{(26)} u c − ∈ H S c ( Ω ) and u c + = P H Ω c ( u c − ∣ Ω c ) , (26) and consequently
1 2 ∥ u c − ∥ L 2 ( Ω ) 2 = 1 2 ∥ u c + ∥ L 2 ( Ω c ) 2 + J i m p ( S c , u c − ) . (27) \frac{1}{2}\|\mathbf{u}_c^-\|_{\mathrm{L}^2(\Omega)}^2=\frac{1}{2}\|\mathbf{u}_c^+\|_{\mathrm{L}^2(\Omega_c)}^2+\mathcal{J}_{\mathrm{imp}}(S_c,\mathbf{u}_c^-). \tag*{(27)} 2 1 ∥ u c − ∥ L 2 ( Ω ) 2 = 2 1 ∥ u c + ∥ L 2 ( Ω c ) 2 + J imp ( S c , u c − ) . (27) Moreover, the sequence ( t j ) (t_j) ( t j ) can be chosen such that, in addition,
E Ω c ( u c + , d c ) + J i m p ( S c , u c − ) ≤ ess lim inf t → T c o l E ( t ) , (28) \mathcal{E}_{\Omega_c}(\mathbf{u}_c^+,d_c)+\mathcal{J}_{\mathrm{imp}}(S_c,\mathbf{u}_c^-)\le\underset{t\to T_{\mathrm{col}}}{\operatorname{ess\,lim\,inf}}\,E(t), \tag*{(28)} E Ω c ( u c + , d c ) + J imp ( S c , u c − ) ≤ t → T col ess lim inf E ( t ) , (28) and the energy inequality (24) holds with J i m p ( S c , u c − ) \mathcal{J}_{\mathrm{imp}}(S_c,\mathbf{u}_c^-) J imp ( S c , u c − ) added to its left-hand side.
Remark 2.5. In Proposition 2.4, the rigid velocity need not converge as t → T c o l t\to T_{\mathrm{col}} t → T col , so the weak pre-impact trace u c − \mathbf{u}_c^- u c − and the impact dissipation J i m p ( S c , u c − ) \mathcal{J}_{\mathrm{imp}}(S_c,\mathbf{u}_c^-) J imp ( S c , u c − ) may depend on the chosen sequence, whereas the post-impact state ( u c + , d c ) (\mathbf{u}_c^+,d_c) ( u c + , d c ) does not. We observe that the inequality (28) may be strict, and the difference indicates a possible loss of compactness of the kinetic or elastic energy at contact.
The post-contact construction uses the following fixed-domain existence theorem, which is independent of the rigid-body geometry and requires no boundary regularity.
Theorem 2.6. Consider a bounded and open set G ⊂ R 3 G\subset\mathbb{R}^3 G ⊂ R 3 with no regularity assumption on ∂ G \partial G ∂ G , and recall the spaces D G , H G , V G \mathcal{D}_G,\mathcal{H}_G,\mathcal{V}_G D G , H G , V G from (10) . If u 0 ∈ H G u_0\in\mathcal{H}_G u 0 ∈ H G and d 0 ∈ H 1 ( G ; S + 2 ) d_0\in\mathrm{H}^1(G;\mathbb{S}^{2}_{+}) d 0 ∈ H 1 ( G ; S + 2 ) , then there exists a global weak solution in the sense of Definition 2.1 .
Energy control of the rigid displacement
In this subsection, we discuss a lower bound for the collision time T c o l T_{\mathrm{col}} T col by the time T e d T_{\mathrm{ed}} T ed associated with the energy dissipation, see (31) below. We observe that the energy inequality controls the motion of the body in two ways. The kinetic energy bounds the rigid velocity, and the viscous dissipation bounds its time integral upon using Poincaré’s and Korn’s inequalities, see (29) below. The estimates of this subsection are used at every level of the construction. They yield collision-free intervals that are independent of the approximation parameters. Throughout the paper, we use the following two constants. First, by C K = C K ( Ω ) > 0 C_K=C_K(\Omega)>0 C K = C K ( Ω ) > 0 we denote a Korn–Poincaré constant, i.e.,
∥ v ∥ L 2 ( Ω ) ≤ C K ∥ D ( v ) ∥ L 2 ( Ω ) for v ∈ V ( Ω ) . (29) \lVert v\rVert_{L^2(\Omega)}\le C_K\lVert\mathbb{D}(v)\rVert_{L^2(\Omega)}\quad\text{for }v\in\mathcal{V}(\Omega). \tag*{(29)} ∥ v ∥ L 2 ( Ω ) ≤ C K ∥ D ( v ) ∥ L 2 ( Ω ) for v ∈ V ( Ω ) . (29)
Next, we define
C S 0 : = m − 1 / 2 + diam ( S 0 ) λ min ( J ( 0 ) ) − 1 / 2 , (30) C_{\mathcal{S}_0}:=m^{-1/2}+\operatorname{diam}(\mathcal{S}_0)\lambda_{\min}(\mathbb{J}(0))^{-1/2}, \tag*{(30)} C S 0 := m − 1/2 + diam ( S 0 ) λ m i n ( J ( 0 ) ) − 1/2 , (30)
where λ min ( J ( 0 ) ) > 0 \lambda_{\min}(\mathbb{J}(0))>0 λ m i n ( J ( 0 )) > 0 denotes the smallest eigenvalue of the inertia tensor of S 0 \mathcal{S}_0 S 0 . For E 0 > 0 E_0>0 E 0 > 0 , let T e d > 0 T_{\mathrm{ed}}>0 T ed > 0 be the unique solution of
T e d ( 1 − e − 4 T e d / C K 2 ) = 2 δ 0 2 C S 0 2 C K 2 E 0 . (31) T_{\mathrm{ed}}\left(1-e^{-4T_{\mathrm{ed}}/C_K^2}\right)=\frac{2\delta_0^2}{C_{\mathcal{S}_0}^2C_K^2E_0}. \tag*{(31)} T ed ( 1 − e − 4 T ed / C K 2 ) = C S 0 2 C K 2 E 0 2 δ 0 2 . (31)
The existence of T e d > 0 T_{\mathrm{ed}}>0 T ed > 0 is due to the fact that the left-hand side of (31) increases strictly from zero to infinity. If E 0 = 0 E_0=0 E 0 = 0 , we set T e d : = + ∞ T_{\mathrm{ed}}:=+\infty T ed := + ∞ .
Lemma 2.7. Let S ( t ) = h ( t ) + O ( t ) S 0 \mathcal{S}(t)=h(t)+O(t)\mathcal{S}_0 S ( t ) = h ( t ) + O ( t ) S 0 denote the position of the rigid body, where we assume that h h h and O O O are Lipschitz continuous, and where O ′ = m O O'=mO O ′ = m O and m x = ω × x mx=\omega\times x m x = ω × x , capturing the motion of the rigid body. Moreover, suppose that the fluid and rigid body velocities coincide, i.e.,
u ( t , x ) = h ′ ( t ) + ω ( t ) × ( x − h ( t ) ) for a.e. x ∈ S ( t ) . u(t,x)=h'(t)+\omega(t)\times(x-h(t))\quad\text{for a.e. }x\in\mathcal{S}(t). u ( t , x ) = h ′ ( t ) + ω ( t ) × ( x − h ( t )) for a.e. x ∈ S ( t ) . Then, with C S 0 C_{\mathcal{S}_0} C S 0 from (2.30), for a.e. t t t , we obtain the bound
sup y ∈ S 0 ∣ d d t ( h ( t ) + O ( t ) y ) ∣ ≤ C S 0 ∥ u ( t ) ∥ L 2 ( S ( t ) ) . \sup_{y\in\mathcal{S}_0}\left|\frac{\mathrm{d}}{\mathrm{d}t}\bigl(h(t)+O(t)y\bigr)\right|\le C_{\mathcal{S}_0}\lVert u(t)\rVert_{L^2(\mathcal{S}(t))}. y ∈ S 0 sup d t d ( h ( t ) + O ( t ) y ) ≤ C S 0 ∥ u ( t ) ∥ L 2 ( S ( t )) .
Proof. First, since by assumption the center of mass of S 0 \mathcal{S}_0 S 0 is the origin and the density equals one, we get
∫ S ( t ) ∣ u ∣ 2 d x = m ∣ h ′ ∣ 2 + ω ⋅ J ( t ) ω . \int_{\mathcal{S}(t)}|u|^2\,\mathrm{d}x=m|h'|^2+\omega\cdot\mathbb{J}(t)\omega. ∫ S ( t ) ∣ u ∣ 2 d x = m ∣ h ′ ∣ 2 + ω ⋅ J ( t ) ω .
Moreover, J ( t ) = O ( t ) J ( 0 ) O ( t ) T \mathbb{J}(t)=O(t)\mathbb{J}(0)O(t)^{\mathsf{T}} J ( t ) = O ( t ) J ( 0 ) O ( t ) T has the same eigenvalues as J ( 0 ) \mathbb{J}(0) J ( 0 ) , so that ∣ h ′ ∣ ≤ m − 1 / 2 ∥ u ∥ L 2 ( S ( t ) ) |h'|\le m^{-1/2}\lVert u\rVert_{L^2(\mathcal{S}(t))} ∣ h ′ ∣ ≤ m − 1/2 ∥ u ∥ L 2 ( S ( t )) and ∣ ω ∣ ≤ λ min ( J ( 0 ) ) − 1 / 2 ∥ u ∥ L 2 ( S ( t ) ) |\omega|\le\lambda_{\min}(\mathbb{J}(0))^{-1/2}\lVert u\rVert_{L^2(\mathcal{S}(t))} ∣ ω ∣ ≤ λ m i n ( J ( 0 ) ) − 1/2 ∥ u ∥ L 2 ( S ( t )) . As the origin lies in the convex hull of S 0 \mathcal{S}_0 S 0 , we have ∣ y ∣ ≤ diam ( S 0 ) |y|\le\operatorname{diam}(\mathcal{S}_0) ∣ y ∣ ≤ diam ( S 0 ) for y ∈ S 0 y\in\mathcal{S}_0 y ∈ S 0 , and therefore
∣ d d t ( h ( t ) + O ( t ) y ) ∣ = ∣ h ′ + ω × O ( t ) y ∣ ≤ ∣ h ′ ∣ + diam ( S 0 ) ∣ ω ∣ ≤ C S 0 ∥ u ( t ) ∥ L 2 ( S ( t ) ) . \left|\frac{\mathrm{d}}{\mathrm{d}t}\bigl(h(t)+O(t)y\bigr)\right|=|h'+\omega\times O(t)y|\le|h'|+\operatorname{diam}(\mathcal{S}_0)|\omega|\le C_{\mathcal{S}_0}\lVert u(t)\rVert_{L^2(\mathcal{S}(t))}. d t d ( h ( t ) + O ( t ) y ) = ∣ h ′ + ω × O ( t ) y ∣ ≤ ∣ h ′ ∣ + diam ( S 0 ) ∣ ω ∣ ≤ C S 0 ∥ u ( t ) ∥ L 2 ( S ( t )) .
□ \square □
Lemma 2.8. Let T > 0 T>0 T > 0 , and let S ( t ) \mathcal{S}(t) S ( t ) and u u u be as in Lemma 2.7 on [ 0 , T ] [0,T] [ 0 , T ] , with h ( 0 ) = 0 h(0)=0 h ( 0 ) = 0 and O ( 0 ) = I 3 O(0)=I_3 O ( 0 ) = I 3 , such that S ( t ) ⊂ Ω \mathcal{S}(t)\subset\Omega S ( t ) ⊂ Ω and u ∈ L ∞ ( 0 , T ; H ( Ω ) ) ∩ L 2 ( 0 , T ; V ( Ω ) ) u\in L^\infty(0,T;\mathcal{H}(\Omega))\cap L^2(0,T;\mathcal{V}(\Omega)) u ∈ L ∞ ( 0 , T ; H ( Ω )) ∩ L 2 ( 0 , T ; V ( Ω )) . Assume that, for a.e. t ∈ ( 0 , T ) t\in(0,T) t ∈ ( 0 , T ) ,
1 2 ∥ u ( t ) ∥ L 2 ( Ω ) 2 + 2 ∫ 0 t ∥ D ( u ( s ) ) ∥ L 2 ( Ω ) 2 d s ≤ E 0 . (32) \frac{1}{2}\lVert u(t)\rVert_{L^2(\Omega)}^2+2\int_0^t\lVert\mathbb{D}(u(s))\rVert_{L^2(\Omega)}^2\,\mathrm{d}s\le E_0. \tag*{(32)} 2 1 ∥ u ( t ) ∥ L 2 ( Ω ) 2 + 2 ∫ 0 t ∥ D ( u ( s )) ∥ L 2 ( Ω ) 2 d s ≤ E 0 . (32) Then, for every 0 ≤ t ≤ T 0\le t\le T 0 ≤ t ≤ T ,
d H ( S ( t ) , S 0 ) ≤ C S 0 C K E 0 t 2 ( 1 − e − 4 t / C K 2 ) ≤ C S 0 min { 2 E 0 t , C K E 0 t 2 } , (33) d_{\mathrm{H}}(\mathcal{S}(t),\mathcal{S}_0)\le C_{\mathcal{S}_0}C_K\sqrt{\frac{E_0t}{2}\left(1-e^{-4t/C_K^2}\right)}
\le C_{\mathcal{S}_0}\min\left\{\sqrt{2E_0t},C_K\sqrt{\frac{E_0t}{2}}\right\}, \tag*{(33)} d H ( S ( t ) , S 0 ) ≤ C S 0 C K 2 E 0 t ( 1 − e − 4 t / C K 2 ) ≤ C S 0 min { 2 E 0 t , C K 2 E 0 t } , (33) and consequently
dist ( S ( t ) , ∂ Ω ) ≥ δ 0 − C S 0 C K E 0 t 2 ( 1 − e − 4 t / C K 2 ) . (34) \operatorname{dist}(\mathcal{S}(t),\partial\Omega)\ge\delta_0-C_{\mathcal{S}_0}C_K\sqrt{\frac{E_0t}{2}\left(1-e^{-4t/C_K^2}\right)}. \tag*{(34)} dist ( S ( t ) , ∂ Ω ) ≥ δ 0 − C S 0 C K 2 E 0 t ( 1 − e − 4 t / C K 2 ) . (34)
Proof. Set Y ( t ) : = ∫ 0 t ∥ u ( s ) ∥ L 2 ( Ω ) 2 d s Y(t) := \int_0^t \lVert u(s)\rVert_{L^2(\Omega)}^2\,\mathrm{d}s Y ( t ) := ∫ 0 t ∥ u ( s ) ∥ L 2 ( Ω ) 2 d s . It readily follows that Y ′ ( t ) = ∥ u ( t ) ∥ L 2 ( Ω ) 2 Y'(t) = \lVert u(t)\rVert_{L^2(\Omega)}^2 Y ′ ( t ) = ∥ u ( t ) ∥ L 2 ( Ω ) 2 for a.e. t t t , and the assumption (32) together with the Korn–Poincaré inequality (29) yields, for a.e. t ∈ ( 0 , T ) t \in(0,T) t ∈ ( 0 , T ) , the estimate 1 2 Y ′ ( t ) + 2 C K 2 Y ( t ) ≤ E 0 \frac{1}{2}Y'(t) + \frac{2}{C_K^2}Y(t) \le E_0 2 1 Y ′ ( t ) + C K 2 2 Y ( t ) ≤ E 0 . Multiplying by e 4 t / C K 2 e^{4t/C_K^2} e 4 t / C K 2 , integrating, and using Y ( 0 ) = 0 Y(0)=0 Y ( 0 ) = 0 , we get
Y ( t ) ≤ C K 2 E 0 2 ( 1 − e − 4 t / C K 2 ) . Y(t) \le\frac{C_K^2 E_0}{2}\left(1-e^{-4t/C_K^2}\right). Y ( t ) ≤ 2 C K 2 E 0 ( 1 − e − 4 t / C K 2 ) .
By Lemma 2.7 and the Cauchy–Schwarz inequality in time, every y ∈ S 0 y \in S_0 y ∈ S 0 satisfies the estimate
∣ h ( t ) + O ( t ) y − y ∣ ≤ C S 0 ∫ 0 t ∥ u ( s ) ∥ L 2 ( Ω ) d s ≤ C S 0 t Y ( t ) , \lvert h(t)+O(t)y-y\rvert\le C_{S_0}\int_0^t \lVert u(s)\rVert_{L^2(\Omega)}\,\mathrm{d}s \le C_{S_0}\sqrt{tY(t)}, ∣ h ( t ) + O ( t ) y − y ∣ ≤ C S 0 ∫ 0 t ∥ u ( s ) ∥ L 2 ( Ω ) d s ≤ C S 0 t Y ( t ) ,
proving the first inequality in (33) . The second inequality in (33) then follows from the elementary estimate 1 − e − x ≤ min { x , 1 } 1-e^{-x}\le\min\{x,1\} 1 − e − x ≤ min { x , 1 } for x ≥ 0 x\ge0 x ≥ 0 . Finally, dist ( S ( t ) , ∂ Ω ) ≥ δ 0 − d H ( S ( t ) , S 0 ) \operatorname{dist}(S(t),\partial\Omega)\ge\delta_0-d_H(S(t),S_0) dist ( S ( t ) , ∂ Ω ) ≥ δ 0 − d H ( S ( t ) , S 0 ) . □
Every energy inequality in this paper contains the kinetic energy and the viscous dissipation on its left-hand side, so Lemma 2.8 applies at all approximation levels. The construction mostly uses the interval of length
T ∗ : = { δ 0 2 C S 0 2 E 0 , E 0 > 0 , + ∞ , E 0 = 0. (35) T_* := \begin{cases}
\frac{\delta_0}{2C_{S_0}\sqrt{2E_0}}, & E_0>0,\\
+\infty, & E_0=0.
\end{cases} \tag*{(35)} T ∗ := { 2 C S 0 2 E 0 δ 0 , + ∞ , E 0 > 0 , E 0 = 0. (35)
Corollary 2.9. Under the assumptions of Lemma 2.8,
dist ( S ( t ) , ∂ Ω ) ≥ δ 0 2 ( 0 ≤ t ≤ min { T , T ∗ } ) , dist ( S ( t ) , ∂ Ω ) > 0 ( 0 ≤ t < min { T , T e d } ) . \operatorname{dist}(S(t),\partial\Omega)\ge\frac{\delta_0}{2}\quad(0\le t\le\min\{T,T_*\}),\qquad\operatorname{dist}(S(t),\partial\Omega)>0\quad(0\le t<\min\{T,T_{\mathrm{ed}}\}). dist ( S ( t ) , ∂ Ω ) ≥ 2 δ 0 ( 0 ≤ t ≤ min { T , T ∗ }) , dist ( S ( t ) , ∂ Ω ) > 0 ( 0 ≤ t < min { T , T ed }) . Proof. The first assertion follows from the bound C S 0 2 E 0 t C_{S_0}\sqrt{2E_0}\,t C S 0 2 E 0 t in (33) . The second follows from (34) , the definition (31) , and the strict monotonicity of t ↦ t ( 1 − e − 4 t / C K 2 ) t\mapsto t(1-e^{-4t/C_K^2}) t ↦ t ( 1 − e − 4 t / C K 2 ) . □
Applied to the solution constructed in this paper, Lemma 2.8 gives the following explicit separation estimate. Its proof is completed in § 5.1 .
Corollary 2.10. The solution of Theorem 2.3 satisfies, for every 0 ≤ t < T c o l 0\le t<T_{\mathrm{col}} 0 ≤ t < T col ,
dist ( S ( t ) , ∂ Ω ) ≥ δ 0 − C S 0 C K E 0 t 2 ( 1 − e − 4 t / C K 2 ) . (36) \operatorname{dist}(S(t),\partial\Omega)\ge\delta_0-C_{S_0}C_K\sqrt{\frac{E_0t}{2}\left(1-e^{-4t/C_K^2}\right)}. \tag*{(36)} dist ( S ( t ) , ∂ Ω ) ≥ δ 0 − C S 0 C K 2 E 0 t ( 1 − e − 4 t / C K 2 ) . (36) In particular,
T c o l ≥ T e d , (37) T_{\mathrm{col}}\ge T_{\mathrm{ed}}, \tag*{(37)} T col ≥ T ed , (37) and if E 0 = 0 E_0=0 E 0 = 0 , then the body is at rest and T c o l = ∞ T_{\mathrm{col}}=\infty T col = ∞ .
Remark 2.11. Since 1 − e − x < min { x , 1 } 1-e^{-x}<\min\{x,1\} 1 − e − x < min { x , 1 } for x > 0 x>0 x > 0 , the definition (31) implies, for E 0 > 0 E_0>0 E 0 > 0 ,
T e d > max { δ 0 C S 0 2 E 0 , 2 δ 0 2 C S 0 2 C K 2 E 0 } . T_{\mathrm{ed}}>\max\left\{\frac{\delta_0}{C_{S_0}\sqrt{2E_0}},\frac{2\delta_0^2}{C_{S_0}^2C_K^2E_0}\right\}. T ed > max { C S 0 2 E 0 δ 0 , C S 0 2 C K 2 E 0 2 δ 0 2 } .
The Ginzburg–Landau approximation
In this section, ε > 0 \varepsilon>0 ε > 0 is fixed, and we construct weak solutions of the Ginzburg–Landau relaxation on the moving domain, see Theorem 3.2. The construction uses three further approximation levels, namely a Galerkin approximation of the velocity (index k k k ), a penalization of the rigidity constraint and of the anchoring condition in the solid (index n n n ), and an H 3 H^3 H 3 artificial viscosity regularization (parameter δ \delta δ ). These are removed in the order k → ∞ k\to\infty k → ∞ , n → ∞ n\to\infty n → ∞ , and δ → 0 \delta\to0 δ → 0 . For fixed δ \delta δ , the approximate velocities are Lipschitz and vanish on ∂ Ω \partial\Omega ∂ Ω , so the body cannot reach the wall in finite time, see (48) . Separation uniformly in δ \delta δ then follows from § 2.3 .
Define
F ε ( d ) : = 1 4 ε 2 ( 1 − ∣ d ∣ 2 ) 2 and f ε ( d ) : = ∇ d F ε ( d ) = − 1 ε 2 ( 1 − ∣ d ∣ 2 ) d . F_\varepsilon(d):=\frac{1}{4\varepsilon^2}(1-\lvert d\rvert^2)^2
\qquad\text{and}\qquad
f_\varepsilon(d):=\nabla_dF_\varepsilon(d)=-\frac{1}{\varepsilon^2}(1-\lvert d\rvert^2)d. F ε ( d ) := 4 ε 2 1 ( 1 − ∣ d ∣ 2 ) 2 and f ε ( d ) := ∇ d F ε ( d ) = − ε 2 1 ( 1 − ∣ d ∣ 2 ) d .
The Ginzburg–Landau approximation then reads
∂ t u F + ( u F ⋅ ∇ ) u F + ∇ π = Δ u F − div ( ( ∇ d ) T ∇ d ) , in ( 0 , T ) × F ( t ) , div u F = 0 , in ( 0 , T ) × F ( t ) , ∂ t d + ( u F ⋅ ∇ ) d = Δ d + 1 ε 2 ( 1 − ∣ d ∣ 2 ) d , in ( 0 , T ) × F ( t ) . (38) \begin{aligned}
\partial_t u^{\mathcal{F}} + (u^{\mathcal{F}} \cdot\nabla)u^{\mathcal{F}} + \nabla\pi&= \Delta u^{\mathcal{F}} - \operatorname{div}\left((\nabla d)^{\mathsf{T}}\nabla d\right), && \text{in } (0,T) \times\mathcal{F}(t),\\
\operatorname{div}u^{\mathcal{F}} &= 0, && \text{in } (0,T) \times\mathcal{F}(t),\\
\partial_t d + (u^{\mathcal{F}} \cdot\nabla)d &= \Delta d + \frac{1}{\varepsilon^2}(1-|d|^2)d, && \text{in } (0,T) \times\mathcal{F}(t). \tag*{(38)}
\end{aligned} ∂ t u F + ( u F ⋅ ∇ ) u F + ∇ π div u F ∂ t d + ( u F ⋅ ∇ ) d = Δ u F − div ( ( ∇ d ) T ∇ d ) , = 0 , = Δ d + ε 2 1 ( 1 − ∣ d ∣ 2 ) d , in ( 0 , T ) × F ( t ) , in ( 0 , T ) × F ( t ) , in ( 0 , T ) × F ( t ) . (38)
The stress tensor T ( u F , π , d ) \mathbb{T}(u^{\mathcal{F}},\pi,d) T ( u F , π , d ) remains as in (4) , and the rigid body dynamics are governed by
m ℓ ′ ( t ) = − ∫ ∂ S ( t ) T ( u F , π , d ) ⋅ n d σ , m\ell'(t)=-\int_{\partial S(t)}\mathbb{T}(u^{\mathcal{F}},\pi,d)\cdot n\,\mathrm{d}\sigma, m ℓ ′ ( t ) = − ∫ ∂ S ( t ) T ( u F , π , d ) ⋅ n d σ ,
J ( t ) ω ′ ( t ) = J ( t ) ω ( t ) × ω ( t ) − ∫ ∂ S ( t ) ( x − h ( t ) ) × T ( u F , π , d ) ⋅ n d σ . \mathbb{J}(t)\omega'(t)=\mathbb{J}(t)\omega(t)\times\omega(t)-\int_{\partial S(t)}(x-h(t))\times\mathbb{T}(u^{\mathcal{F}},\pi,d)\cdot n\,\mathrm{d}\sigma. J ( t ) ω ′ ( t ) = J ( t ) ω ( t ) × ω ( t ) − ∫ ∂ S ( t ) ( x − h ( t )) × T ( u F , π , d ) ⋅ n d σ .
The coupling and boundary conditions are
u F = u S on ( 0 , T ) × ∂ S ( t ) , u F = 0 on ( 0 , T ) × ∂ Ω , u^{\mathcal{F}}=u^{S}\ \text{on }(0,T)\times\partial S(t),\qquad u^{\mathcal{F}}=0\ \text{on }(0,T)\times\partial\Omega, u F = u S on ( 0 , T ) × ∂ S ( t ) , u F = 0 on ( 0 , T ) × ∂ Ω ,
d ( t , x ) = d 0 ( ( Z t ) − 1 ( x ) ) for t ∈ ( 0 , T ) , x ∈ ∂ S ( t ) , d = d 0 on ( 0 , T ) × ∂ Ω , d(t,x)=d_0\left((Z_t)^{-1}(x)\right)\ \text{for }t\in(0,T),\ x\in\partial S(t),\qquad d=d_0\ \text{on }(0,T)\times\partial\Omega, d ( t , x ) = d 0 ( ( Z t ) − 1 ( x ) ) for t ∈ ( 0 , T ) , x ∈ ∂ S ( t ) , d = d 0 on ( 0 , T ) × ∂ Ω ,
where we invoked the flow map Z t Z_t Z t as well as its inverse ( Z t ) − 1 (Z_t)^{-1} ( Z t ) − 1 from (1) and (2) , respectively. The initial conditions are
u F ( 0 , x ) = u 0 ( x ) in F 0 , d ( 0 , x ) = d 0 ( x ) in F 0 , h ( 0 ) = 0 , h ′ ( 0 ) = ℓ 0 , ω ( 0 ) = ω 0 . (39) u^{\mathcal{F}}(0,x)=u_0(x)\ \text{in }\mathcal{F}_0,\qquad d(0,x)=d_0(x)\ \text{in }\mathcal{F}_0,\qquad h(0)=0,\quad h'(0)=\ell_0,\quad\omega(0)=\omega_0. \tag*{(39)} u F ( 0 , x ) = u 0 ( x ) in F 0 , d ( 0 , x ) = d 0 ( x ) in F 0 , h ( 0 ) = 0 , h ′ ( 0 ) = ℓ 0 , ω ( 0 ) = ω 0 . (39)
For the function spaces defined in § 2.2 , the weak solution of the approximate problem (38) –(39) is defined as follows.
Definition 3.1. Let u 0 ∈ H S 0 ( Ω ) u_0 \in\mathcal{H}_{S_0}(\Omega) u 0 ∈ H S 0 ( Ω ) , d 0 ∈ H 2 ( Ω ; S + 2 ) d_0 \in H^2(\Omega;S^2_+) d 0 ∈ H 2 ( Ω ; S + 2 ) , and φ 0 = 1 S 0 \varphi_0=\mathbf{1}_{S_0} φ 0 = 1 S 0 . A weak Ginzburg–Landau solution consists of a rigid trajectory and a triple ( u , d , φ ) (u,d,\varphi) ( u , d , φ ) satisfying all geometric, kinematic, velocity, anchoring, and solid-transport requirements of Definition 2.2 . Moreover, d ∈ L ∞ ( 0 , T ; H 1 ( F ( t ) ; R 3 ) ) d\in L^\infty(0,T;H^1(\mathcal{F}(t);\mathbb{R}^3)) d ∈ L ∞ ( 0 , T ; H 1 ( F ( t ) ; R 3 )) satisfies ∣ d ∣ ≤ 1 |d|\leq1 ∣ d ∣ ≤ 1 and d 3 ≥ 0 d_3\geq0 d 3 ≥ 0 as well as
∫ 0 T ∫ F ( t ) [ − d ⋅ ∂ t ψ + ( u ⋅ ∇ ) d ⋅ ψ + ∇ d : ∇ ψ + f ε ( d ) ⋅ ψ ] d x d t = ∫ F 0 d 0 ⋅ ψ ( 0 ) d x \int_0^T\int_{\mathcal{F}(t)}\left[-d\cdot\partial_t\psi+(u\cdot\nabla)d\cdot\psi+\nabla d:\nabla\psi+f_\varepsilon(d)\cdot\psi\right]\mathrm{d}x\,\mathrm{d}t
=\int_{\mathcal{F}_0}d_0\cdot\psi(0)\,\mathrm{d}x ∫ 0 T ∫ F ( t ) [ − d ⋅ ∂ t ψ + ( u ⋅ ∇ ) d ⋅ ψ + ∇ d : ∇ ψ + f ε ( d ) ⋅ ψ ] d x d t = ∫ F 0 d 0 ⋅ ψ ( 0 ) d x for every ψ ∈ X d ( φ ) \psi\in\mathcal{X}_d(\varphi) ψ ∈ X d ( φ ) . Furthermore, for every ζ ∈ T u \zeta\in\mathcal{T}_u ζ ∈ T u and η ∈ C c 1 ( [ 0 , T ) × Ω ) \eta\in C^1_c([0,T)\times\Omega) η ∈ C c 1 ([ 0 , T ) × Ω ) , the momentum and transport identities read
∫ 0 T ∫ Ω [ − u ⋅ ∂ t ζ − ( u ⊗ u ) : ∇ ζ + 2 D ( u ) : D ( ζ ) ] d x d t = ∫ Ω u 0 ⋅ ζ ( 0 ) d x + ∫ 0 T ∫ F ( t ) ( ∇ d ) T ∇ d : ∇ ζ d x d t , \int_0^T\int_{\Omega}\left[-u\cdot\partial_t\zeta-(u\otimes u):\nabla\zeta+2\mathbb{D}(u):\mathbb{D}(\zeta)\right]\mathrm{d}x\,\mathrm{d}t
=\int_{\Omega}u_0\cdot\zeta(0)\,\mathrm{d}x+\int_0^T\int_{\mathcal{F}(t)}(\nabla d)^{\mathsf{T}}\nabla d:\nabla\zeta\,\mathrm{d}x\,\mathrm{d}t, ∫ 0 T ∫ Ω [ − u ⋅ ∂ t ζ − ( u ⊗ u ) : ∇ ζ + 2 D ( u ) : D ( ζ ) ] d x d t = ∫ Ω u 0 ⋅ ζ ( 0 ) d x + ∫ 0 T ∫ F ( t ) ( ∇ d ) T ∇ d : ∇ ζ d x d t , ∫ 0 T ∫ Ω φ [ ∂ t η + u ⋅ ∇ η ] d x d t = − ∫ Ω φ 0 η ( 0 ) d x . \int_0^T\int_{\Omega}\varphi[\partial_t\eta+u\cdot\nabla\eta]\,\mathrm{d}x\,\mathrm{d}t=-\int_{\Omega}\varphi_0\eta(0)\,\mathrm{d}x. ∫ 0 T ∫ Ω φ [ ∂ t η + u ⋅ ∇ η ] d x d t = − ∫ Ω φ 0 η ( 0 ) d x . The strong anchoring condition (16) is imposed in the trace sense.
We are in a position to state the result on the existence of a weak solution to the Ginzburg–Landau system for fixed ε > 0 \varepsilon>0 ε > 0 .
Theorem 3.2. Assume u 0 ∈ H S 0 ( Ω ) u_0\in\mathcal{H}_{S_0}(\Omega) u 0 ∈ H S 0 ( Ω ) and d 0 ∈ H 2 ( Ω ; S + 2 ) d_0\in H^2(\Omega;S^2_+) d 0 ∈ H 2 ( Ω ; S + 2 ) , let E 0 E_0 E 0 be as in (21) , and let T ∗ T_* T ∗ be given by (35) . If E 0 = 0 E_0=0 E 0 = 0 , we replace T ∗ T_* T ∗ by an arbitrary finite time. Then, for every ε > 0 \varepsilon>0 ε > 0 , the Ginzburg–Landau system (38) –(39) admits a weak solution on [ 0 , T ∗ ] [0,T_*] [ 0 , T ∗ ] in the sense of Definition 3.1
such that dist ( S ( t ) , ∂ Ω ) ≥ δ 0 / 2 \operatorname{dist}(S(t),\partial\Omega)\ge\delta_0/2 dist ( S ( t ) , ∂ Ω ) ≥ δ 0 /2 for 0 ≤ t ≤ T ∗ 0\le t\le T_* 0 ≤ t ≤ T ∗ and which satisfies, for a.e. t ∈ ( 0 , T ∗ ) t\in(0,T_*) t ∈ ( 0 , T ∗ ) , the energy inequality
E ε G L ( t ) + 2 ∫ 0 t ∫ Ω ∣ D ( u ) ∣ 2 d x d s + ∫ 0 t ∫ F ( s ) ∣ Δ d − f ε ( d ) ∣ 2 d x d s ≤ E ε G L ( 0 ) , (40) E_\varepsilon^{\mathrm{GL}}(t)+2\int_0^t\int_\Omega|D(\boldsymbol{u})|^2\,\mathrm{d}x\,\mathrm{d}s+\int_0^t\int_{F(s)}|\Delta\boldsymbol{d}-\boldsymbol{f}_\varepsilon(\boldsymbol{d})|^2\,\mathrm{d}x\,\mathrm{d}s\le E_\varepsilon^{\mathrm{GL}}(0), \tag*{(40)} E ε GL ( t ) + 2 ∫ 0 t ∫ Ω ∣ D ( u ) ∣ 2 d x d s + ∫ 0 t ∫ F ( s ) ∣Δ d − f ε ( d ) ∣ 2 d x d s ≤ E ε GL ( 0 ) , (40)
where
E ε G L ( t ) : = 1 2 ∫ Ω ∣ u ( t ) ∣ 2 d x + 1 2 ∫ F ( t ) ∣ ∇ d ( t ) ∣ 2 d x + ∫ F ( t ) F ε ( d ( t ) ) d x . E_\varepsilon^{\mathrm{GL}}(t):=\frac{1}{2}\int_\Omega|\boldsymbol{u}(t)|^2\,\mathrm{d}x+\frac{1}{2}\int_{F(t)}|\nabla\boldsymbol{d}(t)|^2\,\mathrm{d}x+\int_{F(t)}F_\varepsilon(\boldsymbol{d}(t))\,\mathrm{d}x. E ε GL ( t ) := 2 1 ∫ Ω ∣ u ( t ) ∣ 2 d x + 2 1 ∫ F ( t ) ∣∇ d ( t ) ∣ 2 d x + ∫ F ( t ) F ε ( d ( t )) d x .
We note here that the initial value E ε G L ( 0 ) E_\varepsilon^{\mathrm{GL}}(0) E ε GL ( 0 ) in the above theorem is independent of ε \varepsilon ε since ∣ d 0 ∣ = 1 |\boldsymbol{d}_0|=1 ∣ d 0 ∣ = 1 .
The material anchoring condition is imposed by the following transported lifting, which also preserves the fixed outer boundary datum. Let X s , t \boldsymbol{X}_{s,t} X s , t be the volume-preserving flow generated by u ∈ L ∞ ( 0 , T ; W 1 , ∞ ( Ω ; R 3 ) ) ∩ C ( [ 0 , T ] ; V ( Ω ) ) \boldsymbol{u}\in L^\infty(0,T;W^{1,\infty}(\Omega;\mathbb{R}^3))\cap C([0,T];V(\Omega)) u ∈ L ∞ ( 0 , T ; W 1 , ∞ ( Ω ; R 3 )) ∩ C ([ 0 , T ] ; V ( Ω )) , and put Y ( t ) : = X t , 0 \boldsymbol{Y}(t):=\boldsymbol{X}_{t,0} Y ( t ) := X t , 0 . Since the trace of u \boldsymbol{u} u vanishes on ∂ Ω \partial\Omega ∂ Ω , the flow fixes the outer boundary. Define
φ ( t ) : = φ 0 ∘ Y ( t ) , b ( t ) : = d 0 ∘ Y ( t ) . (41) \varphi(t):=\varphi_0\circ\boldsymbol{Y}(t),\qquad\boldsymbol{b}(t):=\boldsymbol{d}_0\circ\boldsymbol{Y}(t). \tag*{(41)} φ ( t ) := φ 0 ∘ Y ( t ) , b ( t ) := d 0 ∘ Y ( t ) . (41)
Then we have
∂ t φ + u ⋅ ∇ φ = ∂ t b + u ⋅ ∇ b = 0 , φ ( t ) = 1 X 0 , t ( S 0 ) , \partial_t\varphi+\boldsymbol{u}\cdot\nabla\varphi=\partial_t\boldsymbol{b}+\boldsymbol{u}\cdot\nabla\boldsymbol{b}=0,\qquad\varphi(t)=\mathbb{1}_{X_{0,t}(S_0)}, ∂ t φ + u ⋅ ∇ φ = ∂ t b + u ⋅ ∇ b = 0 , φ ( t ) = 1 X 0 , t ( S 0 ) ,
and b ( t , X 0 , t ( y ) ) = d 0 ( y ) \boldsymbol{b}(t,\boldsymbol{X}_{0,t}(y))=\boldsymbol{d}_0(y) b ( t , X 0 , t ( y )) = d 0 ( y ) for y ∈ ∂ S 0 y\in\partial S_0 y ∈ ∂ S 0 , and b = d 0 \boldsymbol{b}=\boldsymbol{d}_0 b = d 0 on ∂ Ω \partial\Omega ∂ Ω .
Let X u : = V ( Ω ) ∩ H 3 ( Ω ; R 3 ) X_u:=V(\Omega)\cap H^3(\Omega;\mathbb{R}^3) X u := V ( Ω ) ∩ H 3 ( Ω ; R 3 ) , then X u X_u X u is a closed subspace of H 3 ( Ω ; R 3 ) H^3(\Omega;\mathbb{R}^3) H 3 ( Ω ; R 3 ) . We can fix a Hilbert inner product a u ( ⋅ , ⋅ ) a_u(\cdot,\cdot) a u ( ⋅ , ⋅ ) on X u X_u X u , which is equivalent to the H 3 H^3 H 3 -inner product. We write
∥ v ∥ X u 2 : = a u ( v , v ) . \|\boldsymbol{v}\|_{X_u}^2:=a_u(\boldsymbol{v},\boldsymbol{v}). ∥ v ∥ X u 2 := a u ( v , v ) .
The regularizing term is δ a u ( u , v ) \delta a_u(\boldsymbol{u},\boldsymbol{v}) δ a u ( u , v ) . Its energy contribution controls the full X u X_u X u -norm and hence, by the embedding X u ↪ W 1 , ∞ ( Ω ; R 3 ) X_u\hookrightarrow W^{1,\infty}(\Omega;\mathbb{R}^3) X u ↪ W 1 , ∞ ( Ω ; R 3 ) , whose norm we denote by C X C_X C X , the Lipschitz norm used to construct the approximate flow maps.
Penalization and artificial viscosity
The rigidity of the body and the anchoring of the director inside the solid are enforced by penalization. The dissipation term n 2 φ ∣ D ( u ) ∣ 2 \frac{n}{2}\varphi|D(\boldsymbol{u})|^2 2 n φ ∣ D ( u ) ∣ 2 penalizes deformations of the solid, and the energy term n 2 φ ∣ d − b ∣ 2 \frac{n}{2}\varphi|\boldsymbol{d}-\boldsymbol{b}|^2 2 n φ ∣ d − b ∣ 2 penalizes deviations of the director from the transported lifting. The artificial viscosity δ a u \delta a_u δ a u provides the Lipschitz regularity of the velocity that is needed to transport φ \varphi φ and b \boldsymbol{b} b .
Definition 3.3. Let u 0 ∈ H S 0 ( Ω ) \boldsymbol{u}_0\in H_{S_0}(\Omega) u 0 ∈ H S 0 ( Ω ) , d 0 ∈ H 2 ( Ω ; S + 2 ) \boldsymbol{d}_0\in H^2(\Omega;\mathbb{S}^2_+) d 0 ∈ H 2 ( Ω ; S + 2 ) , and let φ 0 \varphi_0 φ 0 denote the characteristic function of S 0 S_0 S 0 . A quadruple ( u n , d n , φ n , b n ) (\boldsymbol{u}^n,\boldsymbol{d}^n,\varphi^n,\boldsymbol{b}^n) ( u n , d n , φ n , b n ) with
u n ∈ L ∞ ( 0 , T ; H ( Ω ) ) ∩ L 2 ( 0 , T ; V ( Ω ) ∩ H 3 ( Ω ; R 3 ) ) , \boldsymbol{u}^n\in L^\infty(0,T;\mathcal{H}(\Omega))\cap L^2(0,T;V(\Omega)\cap H^3(\Omega;\mathbb{R}^3)), u n ∈ L ∞ ( 0 , T ; H ( Ω )) ∩ L 2 ( 0 , T ; V ( Ω ) ∩ H 3 ( Ω ; R 3 )) ,
d n ∈ L ∞ ( 0 , T ; H 1 ( Ω ; R 3 ) ) ∩ L 2 ( 0 , T ; H 2 ( Ω ; R 3 ) ) , \boldsymbol{d}^n\in L^\infty(0,T;H^1(\Omega;\mathbb{R}^3))\cap L^2(0,T;H^2(\Omega;\mathbb{R}^3)), d n ∈ L ∞ ( 0 , T ; H 1 ( Ω ; R 3 )) ∩ L 2 ( 0 , T ; H 2 ( Ω ; R 3 )) ,
d n − d 0 ∈ L ∞ ( 0 , T ; H 0 1 ( Ω ; R 3 ) ) , d n ( t , x ) ∈ K + for a.e. ( t , x ) , \boldsymbol{d}^n-\boldsymbol{d}_0\in L^\infty(0,T;H^1_0(\Omega;\mathbb{R}^3)),\qquad\boldsymbol{d}^n(t,x)\in K_+\quad\text{for a.e. }(t,x), d n − d 0 ∈ L ∞ ( 0 , T ; H 0 1 ( Ω ; R 3 )) , d n ( t , x ) ∈ K + for a.e. ( t , x ) ,
φ n ∈ Char ( [ 0 , T ] × Ω ) ∩ C ( [ 0 , T ] ; L p ( Ω ) ) for every p ∈ [ 1 , ∞ ) , \varphi^n\in\operatorname{Char}([0,T]\times\Omega)\cap C([0,T];L^p(\Omega))\quad\text{for every }p\in[1,\infty), φ n ∈ Char ([ 0 , T ] × Ω ) ∩ C ([ 0 , T ] ; L p ( Ω )) for every p ∈ [ 1 , ∞ ) ,
b n ∈ C ( [ 0 , T ] ; L p ( Ω ; R 3 ) ) for every p ∈ [ 1 , ∞ ) \boldsymbol{b}^n\in C([0,T];L^p(\Omega;\mathbb{R}^3))\quad\text{for every }p\in[1,\infty) b n ∈ C ([ 0 , T ] ; L p ( Ω ; R 3 )) for every p ∈ [ 1 , ∞ )
is called a weak solution to the penalized problem if, for all test functions
ζ ∈ H 1 ( 0 , T ; L 2 ( Ω ; R 3 ) ) ∩ L 2 ( 0 , T ; V ( Ω ) ∩ H 3 ( Ω ; R 3 ) ) , ζ ( T ) = 0 , \zeta\in H^1(0,T;L^2(\Omega;\mathbb{R}^3))\cap L^2(0,T;V(\Omega)\cap H^3(\Omega;\mathbb{R}^3)),\qquad\zeta(T)=0, ζ ∈ H 1 ( 0 , T ; L 2 ( Ω ; R 3 )) ∩ L 2 ( 0 , T ; V ( Ω ) ∩ H 3 ( Ω ; R 3 )) , ζ ( T ) = 0 ,
ψ ∈ H 1 ( 0 , T ; L 2 ( Ω ; R 3 ) ) ∩ L 2 ( 0 , T ; H 0 1 ( Ω ; R 3 ) ) , ψ ( T ) = 0 , \psi\in H^1(0,T;L^2(\Omega;\mathbb{R}^3))\cap L^2(0,T;H^1_0(\Omega;\mathbb{R}^3)),\qquad\psi(T)=0, ψ ∈ H 1 ( 0 , T ; L 2 ( Ω ; R 3 )) ∩ L 2 ( 0 , T ; H 0 1 ( Ω ; R 3 )) , ψ ( T ) = 0 ,
η ∈ C 1 ( [ 0 , T ] × Ω ‾ ) , η ( T ) = 0 , \eta\in C^1([0,T]\times\overline{\Omega}),\qquad\eta(T)=0, η ∈ C 1 ([ 0 , T ] × Ω ) , η ( T ) = 0 ,
ξ ∈ C 1 ( [ 0 , T ] × Ω ‾ ; R 3 ) , ξ ( T ) = 0 , \boldsymbol{\xi}\in C^1([0,T]\times\overline{\Omega};\mathbb{R}^3),\qquad\boldsymbol{\xi}(T)=0, ξ ∈ C 1 ([ 0 , T ] × Ω ; R 3 ) , ξ ( T ) = 0 ,
one has
∫ 0 T ∫ Ω [ − u n ⋅ ∂ t ζ − ( u n ⊗ u n ) : ∇ ζ + ( 2 + n φ n ) D ( u n ) : D ( ζ ) ] d x d t + δ ∫ 0 T a u ( u n , ζ ) d t = ∫ Ω u 0 ⋅ ζ ( 0 ) d x − ∫ 0 T ∫ Ω ( ζ ⋅ ∇ ) d n ⋅ Δ d n d x d t , (42) \int_0^T\int_\Omega\left[-u^n\cdot\partial_t\zeta-(u^n\otimes u^n):\nabla\zeta+(2+n\varphi^n)\mathop{\mathrm{D}}(u^n):\mathop{\mathrm{D}}(\zeta)\right]\,\mathrm{d}x\,\mathrm{d}t+\delta\int_0^T a_u(u^n,\zeta)\,\mathrm{d}t
=\int_\Omega u_0\cdot\zeta(0)\,\mathrm{d}x-\int_0^T\int_\Omega(\zeta\cdot\nabla)d^n\cdot\Delta d^n\,\mathrm{d}x\,\mathrm{d}t, \tag*{(42)} ∫ 0 T ∫ Ω [ − u n ⋅ ∂ t ζ − ( u n ⊗ u n ) : ∇ ζ + ( 2 + n φ n ) D ( u n ) : D ( ζ ) ] d x d t + δ ∫ 0 T a u ( u n , ζ ) d t = ∫ Ω u 0 ⋅ ζ ( 0 ) d x − ∫ 0 T ∫ Ω ( ζ ⋅ ∇ ) d n ⋅ Δ d n d x d t , (42)
∫ 0 T ∫ Ω [ − d n ⋅ ∂ t ψ + ( u n ⋅ ∇ ) d n ⋅ ψ − Δ d n ⋅ ψ + f ε ( d n ) ⋅ ψ ] d x d t = ∫ Ω d 0 ⋅ ψ ( 0 ) d x − ∫ 0 T ∫ Ω n φ n ( d n − b n ) ⋅ ψ d x d t , (43) \int_0^T\int_\Omega\left[-d^n\cdot\partial_t\psi+(u^n\cdot\nabla)d^n\cdot\psi-\Delta d^n\cdot\psi+f_\varepsilon(d^n)\cdot\psi\right]\,\mathrm{d}x\,\mathrm{d}t
=\int_\Omega d_0\cdot\psi(0)\,\mathrm{d}x-\int_0^T\int_\Omega n\varphi^n(d^n-b^n)\cdot\psi\,\mathrm{d}x\,\mathrm{d}t, \tag*{(43)} ∫ 0 T ∫ Ω [ − d n ⋅ ∂ t ψ + ( u n ⋅ ∇ ) d n ⋅ ψ − Δ d n ⋅ ψ + f ε ( d n ) ⋅ ψ ] d x d t = ∫ Ω d 0 ⋅ ψ ( 0 ) d x − ∫ 0 T ∫ Ω n φ n ( d n − b n ) ⋅ ψ d x d t , (43)
∫ 0 T ∫ Ω φ n [ ∂ t η + ( u n ⋅ ∇ ) η ] d x d t = − ∫ Ω φ 0 η ( 0 ) d x , and (44) \int_0^T\int_\Omega\varphi^n\left[\partial_t\eta+(u^n\cdot\nabla)\eta\right]\,\mathrm{d}x\,\mathrm{d}t=-\int_\Omega\varphi_0\eta(0)\,\mathrm{d}x,\quad\text{and} \tag*{(44)} ∫ 0 T ∫ Ω φ n [ ∂ t η + ( u n ⋅ ∇ ) η ] d x d t = − ∫ Ω φ 0 η ( 0 ) d x , and (44)
∫ 0 T ∫ Ω b n [ ∂ t ξ + ( u n ⋅ ∇ ) ξ ] d x d t = − ∫ Ω d 0 ⋅ ξ ( 0 ) d x . (45) \int_0^T\int_\Omega b^n\left[\partial_t\xi+(u^n\cdot\nabla)\xi\right]\,\mathrm{d}x\,\mathrm{d}t=-\int_\Omega d_0\cdot\xi(0)\,\mathrm{d}x. \tag*{(45)} ∫ 0 T ∫ Ω b n [ ∂ t ξ + ( u n ⋅ ∇ ) ξ ] d x d t = − ∫ Ω d 0 ⋅ ξ ( 0 ) d x . (45)
We define M n M^n M n by
M n : = Δ d n + 1 ε 2 ( 1 − ∣ d n ∣ 2 ) d n − n φ n ( d n − b n ) . (46) M^n:=\Delta d^n+\frac{1}{\varepsilon^2}(1-\lvert d^n\rvert^2)d^n-n\varphi^n(d^n-b^n). \tag*{(46)} M n := Δ d n + ε 2 1 ( 1 − ∣ d n ∣ 2 ) d n − n φ n ( d n − b n ) . (46)
Theorem 3.4. Let u 0 ∈ H S 0 ( Ω ) u_0\in\mathcal{H}_{\mathrm{S}0}(\Omega) u 0 ∈ H S 0 ( Ω ) , d 0 ∈ H 2 ( Ω ; S + 2 ) d_0\in H^2(\Omega;S^2_+) d 0 ∈ H 2 ( Ω ; S + 2 ) , and let φ 0 \varphi_0 φ 0 be the characteristic function of S 0 S_0 S 0 . Then, for every T > 0 T>0 T > 0 , the penalized problem (42) –(45) admits a solution ( u n , d n , φ n , b n ) (u^n,d^n,\varphi^n,b^n) ( u n , d n , φ n , b n ) on [ 0 , T ] [0,T] [ 0 , T ] in the sense of Definition 3.3 which satisfies, for a.e. t ∈ ( 0 , T ) t\in(0,T) t ∈ ( 0 , T ) , the energy inequality
E n , δ , ε ( t ) + ∫ 0 t ∫ Ω ( 2 + n φ n ) ∣ D ( u n ) ∣ 2 d x d s + δ ∫ 0 t ∥ u n ( s ) ∥ X u 2 d s + ∫ 0 t ∫ Ω ∣ M n ∣ 2 d x d s ≤ E n , δ , ε ( 0 ) = E p e n , 0 , (47) E^{n,\delta,\varepsilon}(t)+\int_0^t\int_\Omega(2+n\varphi^n)\lvert\mathop{\mathrm{D}}(u^n)\rvert^2\,\mathrm{d}x\,\mathrm{d}s+\delta\int_0^t\lVert u^n(s)\rVert_{X_u}^2\,\mathrm{d}s
+\int_0^t\int_\Omega\lvert M^n\rvert^2\,\mathrm{d}x\,\mathrm{d}s\leq E^{n,\delta,\varepsilon}(0)=E_{\mathrm{pen},0}, \tag*{(47)} E n , δ , ε ( t ) + ∫ 0 t ∫ Ω ( 2 + n φ n ) ∣ D ( u n ) ∣ 2 d x d s + δ ∫ 0 t ∥ u n ( s ) ∥ X u 2 d s + ∫ 0 t ∫ Ω ∣ M n ∣ 2 d x d s ≤ E n , δ , ε ( 0 ) = E pen , 0 , (47) where the approximate energy E n , δ , ε E^{n,\delta,\varepsilon} E n , δ , ε is defined by
E n , δ , ε ( t ) : = 1 2 ∫ Ω ∣ u n ∣ 2 d x + ∫ Ω ( 1 2 ∣ ∇ d n ∣ 2 + F ε ( d n ) ) d x + n 2 ∫ Ω φ n ∣ d n − b n ∣ 2 d x . E^{n,\delta,\varepsilon}(t):=\frac{1}{2}\int_\Omega\lvert u^n\rvert^2\,\mathrm{d}x+\int_\Omega\left(\frac{1}{2}\lvert\nabla d^n\rvert^2+F_\varepsilon(d^n)\right)\,\mathrm{d}x+\frac{n}{2}\int_\Omega\varphi^n\lvert d^n-b^n\rvert^2\,\mathrm{d}x. E n , δ , ε ( t ) := 2 1 ∫ Ω ∣ u n ∣ 2 d x + ∫ Ω ( 2 1 ∣ ∇ d n ∣ 2 + F ε ( d n ) ) d x + 2 n ∫ Ω φ n ∣ d n − b n ∣ 2 d x . Here
E p e n , 0 : = 1 2 ∫ Ω ∣ u 0 ∣ 2 d x + 1 2 ∫ Ω ∣ ∇ d 0 ∣ 2 d x , E_{\mathrm{pen},0}:=\frac{1}{2}\int_\Omega\lvert u_0\rvert^2\,\mathrm{d}x+\frac{1}{2}\int_\Omega\lvert\nabla d_0\rvert^2\,\mathrm{d}x, E pen , 0 := 2 1 ∫ Ω ∣ u 0 ∣ 2 d x + 2 1 ∫ Ω ∣ ∇ d 0 ∣ 2 d x , which is independent of n n n , δ \delta δ , and ε \varepsilon ε because ∣ d 0 ∣ = 1 \lvert d_0\rvert=1 ∣ d 0 ∣ = 1 and the initial director penalty vanishes. Moreover, φ n ( t ) = 1 S n ( t ) \varphi^n(t)=\mathbf{1}_{S^n(t)} φ n ( t ) = 1 S n ( t ) with S n ( t ) : = X 0 , t n ( S 0 ) S^n(t):=X^n_{0,t}(S_0) S n ( t ) := X 0 , t n ( S 0 ) , where X n X^n X n denotes the flow of u n u^n u n , and
dist ( S n ( t ) , ∂ Ω ) ≥ ρ δ ( t ) : = δ 0 exp ( − C X E p e n , 0 / δ ) ( 0 ≤ t ≤ T ) . (48) \operatorname{dist}(S^n(t),\partial\Omega)\geq\rho_\delta(t):=\delta_0\exp\left(-C_X\sqrt{E_{\mathrm{pen},0}/\delta}\right)\qquad(0\leq t\leq T). \tag*{(48)} dist ( S n ( t ) , ∂ Ω ) ≥ ρ δ ( t ) := δ 0 exp ( − C X E pen , 0 / δ ) ( 0 ≤ t ≤ T ) . (48)
Semi-Galerkin approximation
At the first approximation level, the parameters n ∈ N n\in\mathbb{N} n ∈ N , δ > 0 \delta>0 δ > 0 , and ε > 0 \varepsilon>0 ε > 0 are fixed, and only the Galerkin dimension k ∈ N k\in\mathbb{N} k ∈ N varies. As in the semi-Galerkin scheme of [38 Sec. 2] , only the velocity is discretized, while the director solves a parabolic problem. The moving body requires two modifications. The solid indicator and the inhomogeneous director datum are both transported by the flow of the Galerkin velocity, the latter through the lifting b k b^k b k in (54) .
Let us recall that F 0 = Ω ∖ S 0 ‾ F_0=\Omega\setminus\overline{S_0} F 0 = Ω ∖ S 0 and φ 0 : = 1 S 0 \varphi_0:=\mathbf{1}_{S_0} φ 0 := 1 S 0 . At this approximation level we work with the whole-container anchoring map
d 0 ∈ H 2 ( Ω ; R 3 ) , ∣ d 0 ∣ = 1 a.e. in Ω . d_0\in H^2(\Omega;\mathbb{R}^3),\qquad\lvert d_0\rvert=1\quad\text{a.e. in }\Omega. d 0 ∈ H 2 ( Ω ; R 3 ) , ∣ d 0 ∣ = 1 a.e. in Ω.
We also recall the Hilbert space X u = V ( Ω ) ∩ H 3 ( Ω ; R 3 ) X_u=\mathcal{V}(\Omega)\cap H^3(\Omega;\mathbb{R}^3) X u = V ( Ω ) ∩ H 3 ( Ω ; R 3 ) and the inner product a u a_u a u fixed above. The usual solenoidal density theorem on the C 2 C^2 C 2 -container implies that the embedding X u ↪ H ( Ω ) X_u\hookrightarrow\mathcal{H}(\Omega) X u ↪ H ( Ω ) is dense, continuous, and compact. The spectral theorem for this compact Hilbert embedding follows that there exist 0 < λ 1 ≤ λ 2 ≤ ⋯ 0 < \lambda_1 \le\lambda_2 \le\cdots 0 < λ 1 ≤ λ 2 ≤ ⋯ , λ j → ∞ \lambda_j \to\infty λ j → ∞ , and an L 2 ( Ω ) L^2(\Omega) L 2 ( Ω ) -orthonormal basis { v j } j ∈ N ⊂ X u \{v_j\}_{j\in\mathbb{N}} \subset X_u { v j } j ∈ N ⊂ X u of H ( Ω ) \mathcal{H}(\Omega) H ( Ω ) satisfying
a u ( v j , w ) = λ j ( v j , w ) L 2 ( Ω ) for every w ∈ X u . a_u(v_j,w)=\lambda_j(v_j,w)_{L^2(\Omega)} \qquad\text{for every } w\in X_u. a u ( v j , w ) = λ j ( v j , w ) L 2 ( Ω ) for every w ∈ X u .
Consequently, if P k P_k P k denotes the L 2 L^2 L 2 -orthogonal projection onto V k : = Span { v 1 , … , v k } V_k:=\operatorname{Span}\{v_1,\ldots,v_k\} V k := Span { v 1 , … , v k } , then
∥ P k w ∥ X u ≤ ∥ w ∥ X u , P k w → w strongly in X u ( w ∈ X u ) . (49) \|P_k w\|_{X_u}\le\|w\|_{X_u},\qquad P_k w\to w\quad\text{strongly in }X_u\quad(w\in X_u). \tag*{(49)} ∥ P k w ∥ X u ≤ ∥ w ∥ X u , P k w → w strongly in X u ( w ∈ X u ) . (49)
Indeed, writing w = ∑ j c j v j w=\sum_j c_jv_j w = ∑ j c j v j in L 2 ( Ω ) L^2(\Omega) L 2 ( Ω ) we have a u ( w , w ) = ∑ j λ j ∣ c j ∣ 2 a_u(w,w)=\sum_j\lambda_j|c_j|^2 a u ( w , w ) = ∑ j λ j ∣ c j ∣ 2 . In particular, we imply that v j ∈ X u ↪ W 1 , ∞ ( Ω ; R 3 ) v_j\in X_u\hookrightarrow W^{1,\infty}(\Omega;\mathbb{R}^3) v j ∈ X u ↪ W 1 , ∞ ( Ω ; R 3 ) , while the finite spans V k V_k V k are dense in H ( Ω ) \mathcal{H}(\Omega) H ( Ω ) with respect to the L 2 L^2 L 2 -norm and in X u X_u X u with respect to the H 3 H^3 H 3 -norm. The approximate velocity has the form
u k ( t , x ) = ∑ j = 1 k a j k ( t ) v j ( x ) . (50) u^k(t,x)=\sum_{j=1}^{k}a_j^k(t)v_j(x). \tag*{(50)} u k ( t , x ) = j = 1 ∑ k a j k ( t ) v j ( x ) . (50)
In particular, we have div u k = 0 \operatorname{div}u^k=0 div u k = 0 in Ω \Omega Ω and u k = 0 u^k=0 u k = 0 on ∂ Ω \partial\Omega ∂ Ω .
For a given velocity u k ∈ C ( [ 0 , T ] ; V k ) u^k\in C([0,T];V_k) u k ∈ C ([ 0 , T ] ; V k ) , let X s , t k : Ω → Ω X^k_{s,t}:\Omega\to\Omega X s , t k : Ω → Ω be the corresponding flow map, i.e.,
d d t X s , t k ( y ) = u k ( t , X s , t k ( y ) ) , X s , s k ( y ) = y . (51) \frac{\mathrm{d}}{\mathrm{d}t}X^k_{s,t}(y)=u^k(t,X^k_{s,t}(y)),\qquad X^k_{s,s}(y)=y. \tag*{(51)} d t d X s , t k ( y ) = u k ( t , X s , t k ( y )) , X s , s k ( y ) = y . (51)
We denote the inverse map by
Y k ( t , x ) : = X t , 0 k ( x ) = ( X 0 , t k ) − 1 ( x ) . (52) Y^k(t,x):=X^k_{t,0}(x)=(X^k_{0,t})^{-1}(x). \tag*{(52)} Y k ( t , x ) := X t , 0 k ( x ) = ( X 0 , t k ) − 1 ( x ) . (52)
Note that the flow map X s , t k X^k_{s,t} X s , t k is volume preserving and fixes ∂ Ω \partial\Omega ∂ Ω . The transported solid indicator, the approximate solid, and the transported lifting are respectively defined by
φ k ( t , x ) : = φ 0 ( Y k ( t , x ) ) , S k ( t ) : = X 0 , t k ( S 0 ) , ϕ k ( t ) = 1 S k ( t ) , (53) \varphi^k(t,x):=\varphi_0(Y^k(t,x)),\qquad S^k(t):=X^k_{0,t}(S_0),\qquad\phi^k(t)=1_{S^k(t)}, \tag*{(53)} φ k ( t , x ) := φ 0 ( Y k ( t , x )) , S k ( t ) := X 0 , t k ( S 0 ) , ϕ k ( t ) = 1 S k ( t ) , (53)
b k ( t , x ) : = d 0 ( Y k ( t , x ) ) . (54) b^k(t,x):=d_0(Y^k(t,x)). \tag*{(54)} b k ( t , x ) := d 0 ( Y k ( t , x )) . (54)
Denote D t k : = ∂ t + u k ⋅ ∇ D_t^k:=\partial_t+u^k\cdot\nabla D t k := ∂ t + u k ⋅ ∇ , we have, in the sense of distributions,
D t k φ k = 0 , D t k b k = 0 in ( 0 , T ) × Ω , φ k ( 0 ) = φ 0 , b k ( 0 ) = d 0 . (55) D_t^k\varphi^k=0,\qquad D_t^k b^k=0\quad\text{in }(0,T)\times\Omega,\qquad\varphi^k(0)=\varphi_0,\qquad b^k(0)=d_0. \tag*{(55)} D t k φ k = 0 , D t k b k = 0 in ( 0 , T ) × Ω , φ k ( 0 ) = φ 0 , b k ( 0 ) = d 0 . (55)
Since the flow fixes ∂ Ω \partial\Omega ∂ Ω , we have b k = d 0 b^k=d_0 b k = d 0 on ( 0 , T ) × ∂ Ω (0,T)\times\partial\Omega ( 0 , T ) × ∂ Ω , and b k ( t , X 0 , t k ( y ) ) = d 0 ( y ) b^k(t,X^k_{0,t}(y))=d_0(y) b k ( t , X 0 , t k ( y )) = d 0 ( y ) for y ∈ ∂ S 0 y\in\partial S_0 y ∈ ∂ S 0 .
For fixed u k u^k u k , φ k \varphi^k φ k , and b k b^k b k , the director field d k d^k d k is defined as the solution of the semilinear parabolic problem
{ ∂ t d k + ( u k ⋅ ∇ ) d k = Δ d k + 1 ε 2 ( 1 − ∣ d k ∣ 2 ) d k − n φ k ( d k − b k ) in ( 0 , T ) × Ω , d k = d 0 on ( ( 0 , T ) × ∂ Ω ) ∪ ( { 0 } × Ω ) . (56) \begin{cases}
\partial_t d^k+(u^k\cdot\nabla)d^k=\Delta d^k+\dfrac{1}{\varepsilon^2}(1-|d^k|^2)d^k-n\varphi^k(d^k-b^k) & \text{in }(0,T)\times\Omega,\\
d^k=d_0 & \text{on }((0,T)\times\partial\Omega)\cup(\{0\}\times\Omega). \tag*{(56)}
\end{cases} ⎩ ⎨ ⎧ ∂ t d k + ( u k ⋅ ∇ ) d k = Δ d k + ε 2 1 ( 1 − ∣ d k ∣ 2 ) d k − n φ k ( d k − b k ) d k = d 0 in ( 0 , T ) × Ω , on (( 0 , T ) × ∂ Ω ) ∪ ({ 0 } × Ω ) . (56)
We define the approximate molecular field M n , k M^{n,k} M n , k by
M n , k : = Δ d k + 1 ε 2 ( 1 − ∣ d k ∣ 2 ) d k − n φ k ( d k − b k ) . (57) M^{n,k}:=\Delta d^k+\frac{1}{\varepsilon^2}(1-|d^k|^2)d^k-n\varphi^k(d^k-b^k). \tag*{(57)} M n , k := Δ d k + ε 2 1 ( 1 − ∣ d k ∣ 2 ) d k − n φ k ( d k − b k ) . (57)
Then (56) is equivalently
D t k d k = M n , k . (58) D_t^k d^k=M^{n,k}. \tag*{(58)} D t k d k = M n , k . (58)
Since D t k b k = 0 D_t^k b^k=0 D t k b k = 0 , the lifted field w k : = d k − b k w^k:=d^k-b^k w k := d k − b k , which vanishes on ∂ Ω \partial\Omega ∂ Ω , satisfies D t k w k = M n , k D_t^k w^k=M^{n,k} D t k w k = M n , k .
The velocity coefficients are determined by the following finite-dimensional weak formulation. For all v ∈ V k v\in V_k v ∈ V k , it holds that
( ∂ t u k , v ) L 2 ( Ω ) + ∫ Ω ( ( u k ⋅ ∇ ) u k ) ⋅ v d x + ∫ Ω ( 2 + n φ k ) D ( u k ) : D ( v ) d x + δ a u ( u k , v ) = − ∫ Ω ( v ⋅ ∇ ) d k ⋅ ( M n , k + n φ k ( d k − b k ) ) d x . (59) (\partial_tu^k,v)_{L^2(\Omega)}+\int_\Omega\bigl((u^k\cdot\nabla)u^k\bigr)\cdot v\,\mathrm{d}x+\int_\Omega(2+n\varphi^k)\mathbb{D}(u^k):\mathbb{D}(v)\,\mathrm{d}x+\delta a_u(u^k,v)
=-\int_\Omega(v\cdot\nabla)d^k\cdot\bigl(M^{n,k}+n\varphi^k(d^k-b^k)\bigr)\,\mathrm{d}x. \tag*{(59)} ( ∂ t u k , v ) L 2 ( Ω ) + ∫ Ω ( ( u k ⋅ ∇ ) u k ) ⋅ v d x + ∫ Ω ( 2 + n φ k ) D ( u k ) : D ( v ) d x + δ a u ( u k , v ) = − ∫ Ω ( v ⋅ ∇ ) d k ⋅ ( M n , k + n φ k ( d k − b k ) ) d x . (59)
The initial condition is u k ( 0 ) = P k u 0 u^k(0)=P_k u_0 u k ( 0 ) = P k u 0 , where P k : H ( Ω ) → V k P_k:\mathcal{H}(\Omega)\to V_k P k : H ( Ω ) → V k denotes the L 2 L^2 L 2 orthogonal projection.
Definition 3.5. Let n ∈ N n\in\mathbb{N} n ∈ N , δ > 0 \delta>0 δ > 0 , ε > 0 \varepsilon>0 ε > 0 , and k ∈ N k\in\mathbb{N} k ∈ N be fixed. A quadruple ( u k , d k , φ k , b k ) (u^k,d^k,\varphi^k,b^k) ( u k , d k , φ k , b k ) is called a solution to the semi-Galerkin approximation if
u k ∈ W 1 , 2 ( 0 , T ; V k ) ∩ C ( [ 0 , T ] ; V k ) , u^k\in W^{1,2}(0,T;V_k)\cap C([0,T];V_k), u k ∈ W 1 , 2 ( 0 , T ; V k ) ∩ C ([ 0 , T ] ; V k ) , φ k \varphi^k φ k and b k b^k b k are given by (53) and (54) , d k d^k d k solves (56) , and (59) holds for all v ∈ V k v\in V_k v ∈ V k .
Define the approximate energy E k n , δ , ε E_k^{n,\delta,\varepsilon} E k n , δ , ε by
E k n , δ , ε ( t ) : = 1 2 ∫ Ω ∣ u k ∣ 2 d x + ∫ Ω ( 1 2 ∣ ∇ d k ∣ 2 + 1 4 ε 2 ( 1 − ∣ d k ∣ 2 ) 2 ) d x + n 2 ∫ Ω φ k ∣ d k − b k ∣ 2 d x . E_k^{n,\delta,\varepsilon}(t):=\frac{1}{2}\int_{\Omega}|u^k|^2\,\mathrm{d}x+\int_{\Omega}\left(\frac{1}{2}|\nabla d^k|^2+\frac{1}{4\varepsilon^2}(1-|d^k|^2)^2\right)\,\mathrm{d}x+\frac{n}{2}\int_{\Omega}\varphi^k|d^k-b^k|^2\,\mathrm{d}x. E k n , δ , ε ( t ) := 2 1 ∫ Ω ∣ u k ∣ 2 d x + ∫ Ω ( 2 1 ∣∇ d k ∣ 2 + 4 ε 2 1 ( 1 − ∣ d k ∣ 2 ) 2 ) d x + 2 n ∫ Ω φ k ∣ d k − b k ∣ 2 d x .
The semi-Galerkin system satisfies the following exact energy identity.
Proposition 3.6. Every sufficiently smooth semi-Galerkin approximation satisfies, for all t ∈ [ 0 , T ] t\in[0,T] t ∈ [ 0 , T ] ,
E k n , δ , ε ( t ) + ∫ 0 t ∫ Ω ( 2 + n φ k ) ∣ D ( u k ) ∣ 2 d x d s + δ ∫ 0 t ∥ u k ( s ) ∥ X u 2 d s + ∫ 0 t ∫ Ω ∣ M n , k ∣ 2 d x d s = E k n , δ , ε ( 0 ) . (60) \begin{aligned}
E_k^{n,\delta,\varepsilon}(t)&+\int_0^t\int_{\Omega}(2+n\varphi^k)|\mathbb{D}(u^k)|^2\,\mathrm{d}x\,\mathrm{d}s+\delta\int_0^t\|u^k(s)\|_{\mathcal{X}_u}^2\,\mathrm{d}s\\
&+\int_0^t\int_{\Omega}|M^{n,k}|^2\,\mathrm{d}x\,\mathrm{d}s=E_k^{n,\delta,\varepsilon}(0). \tag*{(60)}
\end{aligned} E k n , δ , ε ( t ) + ∫ 0 t ∫ Ω ( 2 + n φ k ) ∣ D ( u k ) ∣ 2 d x d s + δ ∫ 0 t ∥ u k ( s ) ∥ X u 2 d s + ∫ 0 t ∫ Ω ∣ M n , k ∣ 2 d x d s = E k n , δ , ε ( 0 ) . (60)
In particular, the estimate is uniform with respect to n n n , provided E k n , δ , ε ( 0 ) E_k^{n,\delta,\varepsilon}(0) E k n , δ , ε ( 0 ) is uniformly bounded with respect to n n n .
Proof. Testing (59) with v = u k v=u^k v = u k we have
1 2 d d t ∫ Ω ∣ u k ∣ 2 d x + ∫ Ω ( 2 + n φ k ) ∣ D ( u k ) ∣ 2 d x + δ ∥ u k ∥ X u 2 = − ∫ Ω ( u k ⋅ ∇ ) d k ⋅ ( M n , k + n φ k ( d k − b k ) ) d x . (61) \frac{1}{2}\frac{\mathrm{d}}{\mathrm{d}t}\int_{\Omega}|u^k|^2\,\mathrm{d}x+\int_{\Omega}(2+n\varphi^k)|\mathbb{D}(u^k)|^2\,\mathrm{d}x+\delta\|u^k\|_{\mathcal{X}_u}^2
=-\int_{\Omega}(u^k\cdot\nabla)d^k\cdot(M^{n,k}+n\varphi^k(d^k-b^k))\,\mathrm{d}x. \tag*{(61)} 2 1 d t d ∫ Ω ∣ u k ∣ 2 d x + ∫ Ω ( 2 + n φ k ) ∣ D ( u k ) ∣ 2 d x + δ ∥ u k ∥ X u 2 = − ∫ Ω ( u k ⋅ ∇ ) d k ⋅ ( M n , k + n φ k ( d k − b k )) d x . (61)
Using the boundary condition d k = d 0 d^k=d_0 d k = d 0 on ∂ Ω \partial\Omega ∂ Ω , and hence ∂ t d k = 0 \partial_t d^k=0 ∂ t d k = 0 on ∂ Ω \partial\Omega ∂ Ω , we obtain
d d t ∫ Ω ( 1 2 ∣ ∇ d k ∣ 2 + F ε ( d k ) ) d x = − ∫ Ω ( M n , k + n φ k ( d k − b k ) ) ⋅ ∂ t d k d x . \frac{\mathrm{d}}{\mathrm{d}t}\int_{\Omega}\left(\frac{1}{2}|\nabla d^k|^2+F_\varepsilon(d^k)\right)\,\mathrm{d}x=-\int_{\Omega}(M^{n,k}+n\varphi^k(d^k-b^k))\cdot\partial_t d^k\,\mathrm{d}x. d t d ∫ Ω ( 2 1 ∣∇ d k ∣ 2 + F ε ( d k ) ) d x = − ∫ Ω ( M n , k + n φ k ( d k − b k )) ⋅ ∂ t d k d x .
Since ∂ t d k = M n , k − ( u k ⋅ ∇ ) d k \partial_t d^k=M^{n,k}-(u^k\cdot\nabla)d^k ∂ t d k = M n , k − ( u k ⋅ ∇ ) d k , this yields
d d t ∫ Ω ( 1 2 ∣ ∇ d k ∣ 2 + F ε ( d k ) ) d x = − ∫ Ω ∣ M n , k ∣ 2 d x − n ∫ Ω φ k ( d k − b k ) ⋅ M n , k d x + ∫ Ω ( u k ⋅ ∇ ) d k ⋅ ( M n , k + n φ k ( d k − b k ) ) d x . (62) \begin{aligned}
\frac{\mathrm{d}}{\mathrm{d}t}\int_{\Omega}\left(\frac{1}{2}|\nabla d^k|^2+F_\varepsilon(d^k)\right)\,\mathrm{d}x
&=-\int_{\Omega}|M^{n,k}|^2\,\mathrm{d}x-n\int_{\Omega}\varphi^k(d^k-b^k)\cdot M^{n,k}\,\mathrm{d}x\\
&\quad+\int_{\Omega}(u^k\cdot\nabla)d^k\cdot(M^{n,k}+n\varphi^k(d^k-b^k))\,\mathrm{d}x. \tag*{(62)}
\end{aligned} d t d ∫ Ω ( 2 1 ∣∇ d k ∣ 2 + F ε ( d k ) ) d x = − ∫ Ω ∣ M n , k ∣ 2 d x − n ∫ Ω φ k ( d k − b k ) ⋅ M n , k d x + ∫ Ω ( u k ⋅ ∇ ) d k ⋅ ( M n , k + n φ k ( d k − b k )) d x . (62)
Since D t k φ k = 0 D_t^k\varphi^k=0 D t k φ k = 0 , div u k = 0 \operatorname{div}u^k=0 div u k = 0 , and u k = 0 u^k=0 u k = 0 on ∂ Ω \partial\Omega ∂ Ω , we have
d d t n 2 ∫ Ω φ k ∣ d k − b k ∣ 2 d x = n ∫ Ω φ k ( d k − b k ) ⋅ D t k ( d k − b k ) d x . \frac{\mathrm{d}}{\mathrm{d}t}\frac{n}{2}\int_{\Omega}\varphi^k|d^k-b^k|^2\,\mathrm{d}x=n\int_{\Omega}\varphi^k(d^k-b^k)\cdot D_t^k(d^k-b^k)\,\mathrm{d}x. d t d 2 n ∫ Ω φ k ∣ d k − b k ∣ 2 d x = n ∫ Ω φ k ( d k − b k ) ⋅ D t k ( d k − b k ) d x .
Then, combining with (55) and (58) , we get
d d t n 2 ∫ Ω φ k ∣ d k − b k ∣ 2 d x = n ∫ Ω φ k ( d k − b k ) ⋅ M n , k d x . (63) \frac{\mathrm{d}}{\mathrm{d}t}\frac{n}{2}\int_{\Omega}\varphi^k|d^k-b^k|^2\,\mathrm{d}x=n\int_{\Omega}\varphi^k(d^k-b^k)\cdot M^{n,k}\,\mathrm{d}x. \tag*{(63)} d t d 2 n ∫ Ω φ k ∣ d k − b k ∣ 2 d x = n ∫ Ω φ k ( d k − b k ) ⋅ M n , k d x . (63)
Adding (61) –(63) , the elastic convection terms cancel, and the two n n n -weighted terms cancel, so
d d t E k n , δ , ε ( t ) + ∫ Ω ( 2 + n φ k ) ∣ D ( u k ) ∣ 2 d x + δ ∥ u k ∥ X u 2 + ∫ Ω ∣ M n , k ∣ 2 d x = 0. \frac{\mathrm{d}}{\mathrm{d}t}E_k^{n,\delta,\varepsilon}(t)+\int_{\Omega}(2+n\varphi^k)|\mathbb{D}(u^k)|^2\,\mathrm{d}x+\delta\|u^k\|_{\mathcal{X}_u}^2+\int_{\Omega}|M^{n,k}|^2\,\mathrm{d}x=0. d t d E k n , δ , ε ( t ) + ∫ Ω ( 2 + n φ k ) ∣ D ( u k ) ∣ 2 d x + δ ∥ u k ∥ X u 2 + ∫ Ω ∣ M n , k ∣ 2 d x = 0.
Integrating this equation over ( 0 , t ) (0,t) ( 0 , t ) we obtain (60) , which follows the last statement. □
Next we state the result on the solution existence for the semi-Galerkin system, its proof is given in Section A.3.
Proposition 3.7. Let u 0 ∈ H S 0 ( Ω ) u_0 \in H_{\mathcal{S}_0}(\Omega) u 0 ∈ H S 0 ( Ω ) , d 0 ∈ H 2 ( Ω ; S + 2 ) d_0 \in H^2(\Omega;S_+^2) d 0 ∈ H 2 ( Ω ; S + 2 ) , denote by φ 0 \varphi_0 φ 0 the characteristic function of S 0 S_0 S 0 , and assume that the boundaries ∂ Ω \partial\Omega ∂ Ω and ∂ S 0 \partial S_0 ∂ S 0 are of class C 2 C^2 C 2 . For every k , n ∈ N k,n \in\mathbb{N} k , n ∈ N , δ , ε > 0 \delta,\varepsilon> 0 δ , ε > 0 , and T > 0 T > 0 T > 0 , the semi-Galerkin system comprising (55) , (56) , and (59) has a solution on [ 0 , T ] [0,T] [ 0 , T ] which satisfies (60) and dist ( S k ( t ) , ∂ Ω ) ≥ ρ δ ( t ) \operatorname{dist}(S^k(t),\partial\Omega) \ge\rho_\delta(t) dist ( S k ( t ) , ∂ Ω ) ≥ ρ δ ( t ) for 0 ≤ t ≤ T 0 \le t \le T 0 ≤ t ≤ T , with ρ δ \rho_\delta ρ δ as in (48) .
We are now ready to prove the existence of a weak solution to the penalized problem (42) –(45) as asserted in Theorem 3.4 .
Proof of Theorem 3.4 . The parameters n n n , δ \delta δ , and ε \varepsilon ε are fixed throughout the proof. We will pass to the limit k → ∞ k \to\infty k → ∞ in the semi-Galerkin system. Fix T > 0 T > 0 T > 0 , and let ( u k , d k , φ k , b k ) (u^k,d^k,\varphi^k,b^k) ( u k , d k , φ k , b k ) be the solutions on [ 0 , T ] [0,T] [ 0 , T ] provided by Proposition 3.7, with corresponding flows X s , t k X^k_{s,t} X s , t k , inverse flows Y k Y^k Y k , and molecular fields M n , k M^{n,k} M n , k as in (57) . We recall the transport identities (55) and the director equation (58) .
By the Galerkin energy identity (60) and the director equation (56) , there exists a constant C > 0 C > 0 C > 0 which is independent of k k k such that
∥ u k ∥ L ∞ ( 0 , T ; H ( Ω ) ) + ∥ u k ∥ L 2 ( 0 , T ; V ( Ω ) ∩ H 3 ( Ω ) ) + ∥ d k ∥ L ∞ ( 0 , T ; H 1 ( Ω ) ) + ∥ d k ∥ L 2 ( 0 , T ; H 2 ( Ω ) ) + ∥ M n , k ∥ L 2 ( ( 0 , T ) × Ω ) + n ∥ ( φ k − b k ) ∥ L ∞ ( 0 , T ; L 2 ( Ω ) ) ≤ C . (64) \lVert u^k\rVert_{L^\infty(0,T;\mathcal{H}(\Omega))}+\lVert u^k\rVert_{L^2(0,T;V(\Omega)\cap H^3(\Omega))}+\lVert d^k\rVert_{L^\infty(0,T;H^1(\Omega))}
+\lVert d^k\rVert_{L^2(0,T;H^2(\Omega))}+\lVert M^{n,k}\rVert_{L^2((0,T)\times\Omega)}
+\sqrt{n}\lVert(\varphi^k-b^k)\rVert_{L^\infty(0,T;L^2(\Omega))}\le C. \tag*{(64)} ∥ u k ∥ L ∞ ( 0 , T ; H ( Ω )) + ∥ u k ∥ L 2 ( 0 , T ; V ( Ω ) ∩ H 3 ( Ω )) + ∥ d k ∥ L ∞ ( 0 , T ; H 1 ( Ω )) + ∥ d k ∥ L 2 ( 0 , T ; H 2 ( Ω )) + ∥ M n , k ∥ L 2 (( 0 , T ) × Ω ) + n ∥( φ k − b k ) ∥ L ∞ ( 0 , T ; L 2 ( Ω )) ≤ C . (64)
The required time-derivative bounds are stated next and proved in Section A.3.
Claim 3.8. There exists a constant C > 0 C > 0 C > 0 , independent of k k k , such that
∥ ∂ t u k ∥ L 2 ( 0 , T ; X u ′ ) ≤ C and ∥ ∂ t d k ∥ L 2 ( 0 , T ; L 2 ( Ω ; R 3 ) ) ≤ C . (65) \lVert\partial_tu^k\rVert_{L^2(0,T;X'_u)}\le C\quad\text{and}\quad\lVert\partial_td^k\rVert_{L^2(0,T;L^2(\Omega;\mathbb{R}^3))}\le C. \tag*{(65)} ∥ ∂ t u k ∥ L 2 ( 0 , T ; X u ′ ) ≤ C and ∥ ∂ t d k ∥ L 2 ( 0 , T ; L 2 ( Ω ; R 3 )) ≤ C . (65) In particular, ∂ t d k \partial_td^k ∂ t d k is bounded in L 2 ( 0 , T ; ( H 1 ( Ω ; R 3 ) ) ′ ) L^2(0,T;(H^1(\Omega;\mathbb{R}^3))') L 2 ( 0 , T ; ( H 1 ( Ω ; R 3 ) ) ′ ) .
The uniform bounds (64) , the fact that φ k \varphi^k φ k is uniformly bounded by one as a characteristic function, and the Banach–Alaoglu theorem provide subsequences (not relabeled) such that
u k ⇀ u weakly in L 2 ( 0 , T ; H 3 ( Ω ; R 3 ) ) and weak-* in L ∞ ( 0 , T ; H ( Ω ) ) , (66) u^k \rightharpoonup u\ \text{weakly in }L^2(0,T;H^3(\Omega;\mathbb{R}^3))\ \text{and weak-* in }L^\infty(0,T;\mathcal{H}(\Omega)), \tag*{(66)} u k ⇀ u weakly in L 2 ( 0 , T ; H 3 ( Ω ; R 3 )) and weak-* in L ∞ ( 0 , T ; H ( Ω )) , (66)
d k ⇀ d weakly in L 2 ( 0 , T ; H 2 ( Ω ; R 3 ) ) and weak-* in L ∞ ( 0 , T ; H 1 ( Ω ; R 3 ) ) , (67) d^k \rightharpoonup d\ \text{weakly in }L^2(0,T;H^2(\Omega;\mathbb{R}^3))\ \text{and weak-* in }L^\infty(0,T;H^1(\Omega;\mathbb{R}^3)), \tag*{(67)} d k ⇀ d weakly in L 2 ( 0 , T ; H 2 ( Ω ; R 3 )) and weak-* in L ∞ ( 0 , T ; H 1 ( Ω ; R 3 )) , (67)
M n , k ⇀ M weakly in L 2 ( 0 , T ; L 2 ( Ω ; R 3 ) ) , φ k ⇀ ∗ φ weak-* in L ∞ ( 0 , T ; L ∞ ( Ω ) ) , (68) M^{n,k} \rightharpoonup M\ \text{weakly in }L^2(0,T;L^2(\Omega;\mathbb{R}^3)),\qquad\varphi^k \rightharpoonup^\ast\varphi\ \text{weak-* in }L^\infty(0,T;L^\infty(\Omega)), \tag*{(68)} M n , k ⇀ M weakly in L 2 ( 0 , T ; L 2 ( Ω ; R 3 )) , φ k ⇀ ∗ φ weak-* in L ∞ ( 0 , T ; L ∞ ( Ω )) , (68)
for some limit functions
u ∈ L 2 ( 0 , T ; H 3 ( Ω ; R 3 ) ) ∩ L ∞ ( 0 , T ; H ( Ω ) ) , u \in L^2(0,T;H^3(\Omega;\mathbb{R}^3))\cap L^\infty(0,T;\mathcal{H}(\Omega)), u ∈ L 2 ( 0 , T ; H 3 ( Ω ; R 3 )) ∩ L ∞ ( 0 , T ; H ( Ω )) ,
d ∈ L 2 ( 0 , T ; H 2 ( Ω ; R 3 ) ) ∩ L ∞ ( 0 , T ; H 1 ( Ω ; R 3 ) ) , d \in L^2(0,T;H^2(\Omega;\mathbb{R}^3))\cap L^\infty(0,T;H^1(\Omega;\mathbb{R}^3)), d ∈ L 2 ( 0 , T ; H 2 ( Ω ; R 3 )) ∩ L ∞ ( 0 , T ; H 1 ( Ω ; R 3 )) ,
M ∈ L 2 ( ( 0 , T ) × Ω ; R 3 ) , and φ ∈ L ∞ ( ( 0 , T ) × Ω ) . M \in L^2((0,T)\times\Omega;\mathbb{R}^3),\qquad\text{and}\qquad\varphi\in L^\infty((0,T)\times\Omega). M ∈ L 2 (( 0 , T ) × Ω ; R 3 ) , and φ ∈ L ∞ (( 0 , T ) × Ω ) .
By (64) and (65) , the Banach–Alaoglu theorem and the Aubin–Lions–Simon compactness lemma imply, up to a non-relabeled subsequence,
u k → u strongly in L 2 ( 0 , T ; H s ( Ω ; R 3 ) ) for every s < 3 , (69) u^k \to u\ \text{strongly in }L^2(0,T;H^s(\Omega;\mathbb{R}^3))\ \text{for every }s<3, \tag*{(69)} u k → u strongly in L 2 ( 0 , T ; H s ( Ω ; R 3 )) for every s < 3 , (69)
d k → d strongly in L 2 ( 0 , T ; H 2 − η ( Ω ; R 3 ) ) for every η > 0. (70) d^k \to d\ \text{strongly in }L^2(0,T;H^{2-\eta}(\Omega;\mathbb{R}^3))\ \text{for every }\eta>0. \tag*{(70)} d k → d strongly in L 2 ( 0 , T ; H 2 − η ( Ω ; R 3 )) for every η > 0. (70)
We now identify the limits of the transported solid indicator φ k \varphi^k φ k and of the transported lifting b k b^k b k . By (69) , choosing s ∈ ( 5 2 , 3 ) s\in\left(\frac{5}{2},3\right) s ∈ ( 2 5 , 3 ) , and using H s ( Ω ; R 3 ) ↪ W 1 , ∞ ( Ω ; R 3 ) H^s(\Omega;\mathbb{R}^3)\hookrightarrow W^{1,\infty}(\Omega;\mathbb{R}^3) H s ( Ω ; R 3 ) ↪ W 1 , ∞ ( Ω ; R 3 ) , we obtain
u k → u strongly in L 1 ( 0 , T ; W 1 , ∞ ( Ω ; R 3 ) ) . (71) u^k \to u\ \text{strongly in }L^1(0,T;W^{1,\infty}(\Omega;\mathbb{R}^3)). \tag*{(71)} u k → u strongly in L 1 ( 0 , T ; W 1 , ∞ ( Ω ; R 3 )) . (71)
Indeed, the strong convergence first holds in L 2 ( 0 , T ; W 1 , ∞ ( Ω ; R 3 ) ) L^2(0,T;W^{1,\infty}(\Omega;\mathbb{R}^3)) L 2 ( 0 , T ; W 1 , ∞ ( Ω ; R 3 )) , and thus in L 1 ( 0 , T ; W 1 , ∞ ( Ω ; R 3 ) ) L^1(0,T;W^{1,\infty}(\Omega;\mathbb{R}^3)) L 1 ( 0 , T ; W 1 , ∞ ( Ω ; R 3 )) . Let X s , t X_{s,t} X s , t denote the flow generated by the limiting velocity u u u , that is,
d d t X s , t ( y ) = u ( t , X s , t ( y ) ) , X s , s ( y ) = y . \frac{\mathrm{d}}{\mathrm{d}t}X_{s,t}(y)=u(t,X_{s,t}(y)),\qquad X_{s,s}(y)=y. d t d X s , t ( y ) = u ( t , X s , t ( y )) , X s , s ( y ) = y .
We recall that Y ( t , x ) = X t , 0 ( x ) Y(t,x)=X_{t,0}(x) Y ( t , x ) = X t , 0 ( x ) . Since u ∈ L 1 ( 0 , T ; W 1 , ∞ ( Ω ; R 3 ) ) u\in L^{1}(0,T;W^{1,\infty}(\Omega;\mathbb{R}^{3})) u ∈ L 1 ( 0 , T ; W 1 , ∞ ( Ω ; R 3 )) , this flow is well-defined and unique. Moreover, the stability estimate for flows associated with Sobolev–Lipschitz vector fields, together with (71) , gives
X s , t k → X s , t and Y k ( t , ⋅ ) → Y ( t , ⋅ ) (72) X^{k}_{s,t}\to X_{s,t}\quad\text{and}\quad Y^{k}(t,\cdot)\to Y(t,\cdot) \tag*{(72)} X s , t k → X s , t and Y k ( t , ⋅ ) → Y ( t , ⋅ ) (72)
uniformly on [ 0 , T ] 2 × Ω ‾ [0,T]^{2}\times\overline{\Omega} [ 0 , T ] 2 × Ω . More precisely, for 0 ≤ s , t ≤ T 0\leq s,t\leq T 0 ≤ s , t ≤ T ,
sup y ∈ Ω ∣ X s , t k ( y ) − X s , t ( y ) ∣ ≤ C ∫ 0 T ∥ u k ( τ ) − u ( τ ) ∥ L ∞ ( Ω ) d τ , \sup_{y\in\Omega}\lvert X^{k}_{s,t}(y)-X_{s,t}(y)\rvert\leq C\int_{0}^{T}\lVert u^{k}(\tau)-u(\tau)\rVert_{L^{\infty}(\Omega)}\,\mathrm{d}\tau, y ∈ Ω sup ∣ X s , t k ( y ) − X s , t ( y )∣ ≤ C ∫ 0 T ∥ u k ( τ ) − u ( τ ) ∥ L ∞ ( Ω ) d τ ,
where C C C depends on sup k ∫ 0 T ∥ ∇ u k ( τ ) ∥ L ∞ ( Ω ) d τ \sup_{k}\int_{0}^{T}\lVert\nabla u^{k}(\tau)\rVert_{L^{\infty}(\Omega)}\,\mathrm{d}\tau sup k ∫ 0 T ∥ ∇ u k ( τ ) ∥ L ∞ ( Ω ) d τ , but is independent of k k k .
For the above limiting velocity field u u u , we define φ ( t , x ) : = φ 0 ( Y ( t , x ) ) \varphi(t,x):=\varphi_{0}(Y(t,x)) φ ( t , x ) := φ 0 ( Y ( t , x )) and b ( t , x ) : = d 0 ( Y ( t , x ) ) b(t,x):=d_{0}(Y(t,x)) b ( t , x ) := d 0 ( Y ( t , x )) as in (41) . This identifies the weak-* limit.
First, since φ k ( t , x ) = φ 0 ( Y k ( t , x ) ) = 1 S 0 ( Y k ( t , x ) ) \varphi^{k}(t,x)=\varphi_{0}(Y^{k}(t,x))=\mathbf{1}_{S_{0}}(Y^{k}(t,x)) φ k ( t , x ) = φ 0 ( Y k ( t , x )) = 1 S 0 ( Y k ( t , x )) , we claim that
φ k → φ strongly in C ( [ 0 , T ] ; L p ( Ω ) ) for every 1 ≤ p < ∞ . (73) \varphi^{k}\to\varphi\quad\text{strongly in }C([0,T];L^{p}(\Omega))\quad\text{for every }1\leq p<\infty. \tag*{(73)} φ k → φ strongly in C ([ 0 , T ] ; L p ( Ω )) for every 1 ≤ p < ∞. (73)
Indeed, let
ρ k : = sup t ∈ [ 0 , T ] ∥ Y k ( t , ⋅ ) − Y ( t , ⋅ ) ∥ L ∞ ( Ω ) . \rho_{k}:=\sup_{t\in[0,T]}\lVert Y^{k}(t,\cdot)-Y(t,\cdot)\rVert_{L^{\infty}(\Omega)}. ρ k := t ∈ [ 0 , T ] sup ∥ Y k ( t , ⋅ ) − Y ( t , ⋅ ) ∥ L ∞ ( Ω ) .
Then ρ k → 0 \rho_{k}\to0 ρ k → 0 . For each t ∈ [ 0 , T ] t\in[0,T] t ∈ [ 0 , T ] , the set on which φ k ( t ) ≠ φ ( t ) \varphi^{k}(t)\neq\varphi(t) φ k ( t ) = φ ( t ) is contained in
{ x ∈ Ω : dist ( Y ( t , x ) , ∂ S 0 ) ≤ ρ k } . \{x\in\Omega:\operatorname{dist}(Y(t,x),\partial S_{0})\leq\rho_{k}\}. { x ∈ Ω : dist ( Y ( t , x ) , ∂ S 0 ) ≤ ρ k } .
Since u u u is divergence-free and satisfies the no-slip condition on ∂ Ω \partial\Omega ∂ Ω , the flow Y ( t , ⋅ ) Y(t,\cdot) Y ( t , ⋅ ) is measure-preserving. Hence
∣ { x ∈ Ω : φ k ( t , x ) ≠ φ ( t , x ) } ∣ ≤ ∣ { y ∈ Ω : dist ( y , ∂ S 0 ) ≤ ρ k } ∣ . \lvert\{x\in\Omega:\varphi^{k}(t,x)\neq\varphi(t,x)\}\rvert\leq\lvert\{y\in\Omega:\operatorname{dist}(y,\partial S_{0})\leq\rho_{k}\}\rvert. ∣{ x ∈ Ω : φ k ( t , x ) = φ ( t , x )}∣ ≤ ∣{ y ∈ Ω : dist ( y , ∂ S 0 ) ≤ ρ k }∣ .
The right-hand side tends to zero because ∂ S 0 \partial S_{0} ∂ S 0 has zero Lebesgue measure. Therefore
sup t ∈ [ 0 , T ] ∥ φ k ( t ) − φ ( t ) ∥ L p ( Ω ) p ≤ ∣ { y ∈ Ω : dist ( y , ∂ S 0 ) ≤ ρ k } ∣ → 0 , \sup_{t\in[0,T]}\lVert\varphi^{k}(t)-\varphi(t)\rVert_{L^{p}(\Omega)}^{p}\leq\lvert\{y\in\Omega:\operatorname{dist}(y,\partial S_{0})\leq\rho_{k}\}\rvert\to0, t ∈ [ 0 , T ] sup ∥ φ k ( t ) − φ ( t ) ∥ L p ( Ω ) p ≤ ∣{ y ∈ Ω : dist ( y , ∂ S 0 ) ≤ ρ k }∣ → 0 ,
which proves (73) . In particular,
φ ( t , ⋅ ) = 1 S ( t ) for all t ∈ [ 0 , T ] , S ( t ) : = X 0 , t ( S 0 ) . \varphi(t,\cdot)=\mathbf{1}_{S(t)}\quad\text{for all }t\in[0,T],\qquad S(t):=X_{0,t}(S_{0}). φ ( t , ⋅ ) = 1 S ( t ) for all t ∈ [ 0 , T ] , S ( t ) := X 0 , t ( S 0 ) .
Thus, the weak-* limit of φ k \varphi^{k} φ k is again a characteristic function.
Next, we treat the transported lifting. Since d 0 ∈ H 2 ( Ω ; R 3 ) d_{0}\in H^{2}(\Omega;\mathbb{R}^{3}) d 0 ∈ H 2 ( Ω ; R 3 ) , and since in the dimensions under consideration H 2 ( Ω ; R 3 ) ↪ C 0 , α ( Ω ‾ ; R 3 ) H^{2}(\Omega;\mathbb{R}^{3})\hookrightarrow C^{0,\alpha}(\overline{\Omega};\mathbb{R}^{3}) H 2 ( Ω ; R 3 ) ↪ C 0 , α ( Ω ; R 3 ) for some α ∈ ( 0 , 1 2 ) \alpha\in(0,\frac{1}{2}) α ∈ ( 0 , 2 1 ) , we may choose the Hölder continuous representative of d 0 d_{0} d 0 . Using (72) , we obtain
∥ b k ( t ) − b ( t ) ∥ L ∞ ( Ω ) = ∥ d 0 ( Y k ( t , ⋅ ) ) − d 0 ( Y ( t , ⋅ ) ) ∥ L ∞ ( Ω ) ≤ [ d 0 ] C 0 , α ( Ω ‾ ) ∥ Y k ( t , ⋅ ) − Y ( t , ⋅ ) ∥ L ∞ ( Ω ) . \lVert b^{k}(t)-b(t)\rVert_{L^{\infty}(\Omega)}=\lVert d_{0}(Y^{k}(t,\cdot))-d_{0}(Y(t,\cdot))\rVert_{L^{\infty}(\Omega)}\leq[d_{0}]_{C^{0,\alpha}(\overline{\Omega})}\lVert Y^{k}(t,\cdot)-Y(t,\cdot)\rVert_{L^{\infty}(\Omega)}. ∥ b k ( t ) − b ( t ) ∥ L ∞ ( Ω ) = ∥ d 0 ( Y k ( t , ⋅ )) − d 0 ( Y ( t , ⋅ )) ∥ L ∞ ( Ω ) ≤ [ d 0 ] C 0 , α ( Ω ) ∥ Y k ( t , ⋅ ) − Y ( t , ⋅ ) ∥ L ∞ ( Ω ) .
Consequently,
b k → b strongly in C ( [ 0 , T ] ; L ∞ ( Ω ; R 3 ) ) , (74) b^{k}\to b\quad\text{strongly in }C([0,T];L^{\infty}(\Omega;\mathbb{R}^{3})), \tag*{(74)} b k → b strongly in C ([ 0 , T ] ; L ∞ ( Ω ; R 3 )) , (74)
and hence also strongly in C ( [ 0 , T ] ; L p ( Ω ; R 3 ) ) C([0,T];L^{p}(\Omega;\mathbb{R}^{3})) C ([ 0 , T ] ; L p ( Ω ; R 3 )) for every 1 ≤ p < ∞ 1\leq p<\infty 1 ≤ p < ∞ . Moreover, since ∣ d 0 ∣ = 1 \lvert d_{0}\rvert=1 ∣ d 0 ∣ = 1 a.e. in Ω \Omega Ω , the transported lift satisfies
∣ b k ∣ = 1 and ∣ b ∣ = 1 a.e. in ( 0 , T ) × Ω . \lvert b^{k}\rvert=1\quad\text{and}\quad\lvert b\rvert=1\quad\text{a.e. in }(0,T)\times\Omega. ∣ b k ∣ = 1 and ∣ b ∣ = 1 a.e. in ( 0 , T ) × Ω.
Finally, passing to the limit in the transport equations (55) , using the strong convergence of u k u^{k} u k , φ k \varphi^{k} φ k , and b k b^{k} b k , yields
∂ t φ + div ( φ u ) = 0 , ∂ t b + u ⋅ ∇ b = 0 \partial_{t}\varphi+\operatorname{div}(\varphi u)=0,\qquad\partial_{t}b+u\cdot\nabla b=0 ∂ t φ + div ( φ u ) = 0 , ∂ t b + u ⋅ ∇ b = 0
in D ′ ( ( 0 , T ) × Ω ) \mathcal{D}'((0,T)\times\Omega) D ′ (( 0 , T ) × Ω ) , with initial data φ ( 0 ) = φ 0 \varphi(0)=\varphi_{0} φ ( 0 ) = φ 0 , b ( 0 ) = d 0 b(0)=d_{0} b ( 0 ) = d 0 . Since div u = 0 \operatorname{div}u=0 div u = 0 , the first equation is equivalently ∂ t φ + u ⋅ ∇ φ = 0 \partial_{t}\varphi+u\cdot\nabla\varphi=0 ∂ t φ + u ⋅ ∇ φ = 0 in the distributional sense.
Now we pass to the limit in the momentum equation. Fix m ∈ N m \in\mathbb{N} m ∈ N and let ζ ∈ C 1 ( [ 0 , T ] ; V ( Ω ) ∩ H 3 ( Ω ; R 3 ) ) \zeta\in C^{1}([0,T];\mathcal{V}(\Omega)\cap H^{3}(\Omega;\mathbb{R}^{3})) ζ ∈ C 1 ([ 0 , T ] ; V ( Ω ) ∩ H 3 ( Ω ; R 3 )) be of the form ζ ( t , x ) = ∑ i = 1 m a i ( t ) v i ( x ) \zeta(t,x)=\sum_{i=1}^{m}a^{i}(t)v_{i}(x) ζ ( t , x ) = ∑ i = 1 m a i ( t ) v i ( x ) with a i ( T ) = 0 a^{i}(T)=0 a i ( T ) = 0 . For every k ≥ m k\ge m k ≥ m we may substitute ζ \zeta ζ in place of v v v in (59) , integrate over ( 0 , T ) (0,T) ( 0 , T ) , and integrate by parts in time to obtain
∫ 0 T ∫ Ω [ − u k ⋅ ∂ t ζ − ( u k ⊗ u k ) : ∇ ζ + ( 2 + n φ k ) D ( u k ) : D ( ζ ) + ( ζ ⋅ ∇ ) d k ⋅ ( M n , k + n φ k ( d k − b k ) ) ] d x d t + δ ∫ 0 T a u ( u k , ζ ) d t = ∫ Ω u k ( 0 ) ⋅ ζ ( 0 ) d x . \int_{0}^{T}\int_{\Omega}\left[-u^{k}\cdot\partial_{t}\zeta-(u^{k}\otimes u^{k}):\nabla\zeta+(2+n\varphi^{k})D(u^{k}):D(\zeta)+(\zeta\cdot\nabla)d^{k}\cdot\left(M^{n,k}+n\varphi^{k}(d^{k}-b^{k})\right)\right]\,\mathrm{d}x\,\mathrm{d}t+\delta\int_{0}^{T}a_{u}(u^{k},\zeta)\,\mathrm{d}t=\int_{\Omega}u^{k}(0)\cdot\zeta(0)\,\mathrm{d}x. ∫ 0 T ∫ Ω [ − u k ⋅ ∂ t ζ − ( u k ⊗ u k ) : ∇ ζ + ( 2 + n φ k ) D ( u k ) : D ( ζ ) + ( ζ ⋅ ∇ ) d k ⋅ ( M n , k + n φ k ( d k − b k ) ) ] d x d t + δ ∫ 0 T a u ( u k , ζ ) d t = ∫ Ω u k ( 0 ) ⋅ ζ ( 0 ) d x .
Since u k ( 0 ) = P k u 0 u^{k}(0)=P_{k}u_{0} u k ( 0 ) = P k u 0 and P k u 0 → u 0 P_{k}u_{0}\to u_{0} P k u 0 → u 0 strongly in L 2 ( Ω ) L^{2}(\Omega) L 2 ( Ω ) , the initial term converges to ∫ Ω u 0 ⋅ ζ ( 0 ) d x \int_{\Omega}u_{0}\cdot\zeta(0)\,\mathrm{d}x ∫ Ω u 0 ⋅ ζ ( 0 ) d x . Passing to the remaining terms using the weak and weak-* convergences (3.29)–(3.31) of u k u^{k} u k , d k d^{k} d k , M n , k M^{n,k} M n , k , and φ k \varphi^{k} φ k , and the strong convergences (3.32)–(3.33) of u k u^{k} u k and d k d^{k} d k as well as (3.36)–(3.37) of φ k \varphi^{k} φ k and b k b^{k} b k , we arrive at the identity
∫ 0 T ∫ Ω [ − u ⋅ ∂ t ζ − ( u ⊗ u ) : ∇ ζ + ( 2 + n φ ) D ( u ) : D ( ζ ) + ( ζ ⋅ ∇ ) d ⋅ ( M + n φ ( d − b ) ) ] d x d t + δ ∫ 0 T a u ( u , ζ ) d t = ∫ Ω u 0 ⋅ ζ ( 0 ) d x . \int_{0}^{T}\int_{\Omega}\left[-u\cdot\partial_{t}\zeta-(u\otimes u):\nabla\zeta+(2+n\varphi)D(u):D(\zeta)+(\zeta\cdot\nabla)d\cdot\left(M+n\varphi(d-b)\right)\right]\,\mathrm{d}x\,\mathrm{d}t+\delta\int_{0}^{T}a_{u}(u,\zeta)\,\mathrm{d}t=\int_{\Omega}u_{0}\cdot\zeta(0)\,\mathrm{d}x. ∫ 0 T ∫ Ω [ − u ⋅ ∂ t ζ − ( u ⊗ u ) : ∇ ζ + ( 2 + n φ ) D ( u ) : D ( ζ ) + ( ζ ⋅ ∇ ) d ⋅ ( M + n φ ( d − b ) ) ] d x d t + δ ∫ 0 T a u ( u , ζ ) d t = ∫ Ω u 0 ⋅ ζ ( 0 ) d x .
Because f ε = ∇ d F ε f_{\varepsilon}=\nabla_{d}F_{\varepsilon} f ε = ∇ d F ε and div ζ = 0 \operatorname{div}\zeta=0 div ζ = 0 with zero boundary trace,
∫ Ω ( ζ ⋅ ∇ ) d ⋅ f ε ( d ) d x = ∫ Ω ζ ⋅ ∇ F ε ( d ) d x = 0. \int_{\Omega}(\zeta\cdot\nabla)d\cdot f_{\varepsilon}(d)\,\mathrm{d}x=\int_{\Omega}\zeta\cdot\nabla F_{\varepsilon}(d)\,\mathrm{d}x=0. ∫ Ω ( ζ ⋅ ∇ ) d ⋅ f ε ( d ) d x = ∫ Ω ζ ⋅ ∇ F ε ( d ) d x = 0.
Since M + n φ ( d − b ) = Δ d − f ε ( d ) M+n\varphi(d-b)=\Delta d-f_{\varepsilon}(d) M + n φ ( d − b ) = Δ d − f ε ( d ) , the last term on the left of the preceding identity therefore reduces to ∫ 0 T ∫ Ω ( ζ ⋅ ∇ ) d ⋅ Δ d d x d t \int_{0}^{T}\int_{\Omega}(\zeta\cdot\nabla)d\cdot\Delta d\,\mathrm{d}x\,\mathrm{d}t ∫ 0 T ∫ Ω ( ζ ⋅ ∇ ) d ⋅ Δ d d x d t . Moving it to the right-hand side gives the elastic term with the minus sign in (42) . Because ⋃ m ≥ 1 V m \bigcup_{m\ge1}V_{m} ⋃ m ≥ 1 V m is dense in X u = V ( Ω ) ∩ H 3 ( Ω ) X_{u}=\mathcal{V}(\Omega)\cap H^{3}(\Omega) X u = V ( Ω ) ∩ H 3 ( Ω ) and the spectral projections satisfy (49) , the preceding identity is continuous with respect to the norm ∥ ζ ∥ H 1 ( 0 , T ; L 2 ( Ω ) ) + ∥ ζ ∥ L 2 ( 0 , T ; X u ) \|\zeta\|_{H^{1}(0,T;L^{2}(\Omega))}+\|\zeta\|_{L^{2}(0,T;X_{u})} ∥ ζ ∥ H 1 ( 0 , T ; L 2 ( Ω )) + ∥ ζ ∥ L 2 ( 0 , T ; X u ) . Time mollification followed by the L 2 L^{2} L 2 -orthogonal projection P k P_{k} P k therefore converges in this norm (the projection is contractive both in L 2 ( Ω ) L^{2}(\Omega) L 2 ( Ω ) and in X u X_{u} X u ) and extends the identity to every test function ζ \zeta ζ admitted in Definition 3.3. This proves (42) .
For the director equation, let ψ ∈ H 1 ( 0 , T ; L 2 ( Ω ; R 3 ) ) ∩ L 2 ( 0 , T ; H 0 1 ( Ω ; R 3 ) ) \psi\in H^{1}(0,T;L^{2}(\Omega;\mathbb{R}^{3}))\cap L^{2}(0,T;H_{0}^{1}(\Omega;\mathbb{R}^{3})) ψ ∈ H 1 ( 0 , T ; L 2 ( Ω ; R 3 )) ∩ L 2 ( 0 , T ; H 0 1 ( Ω ; R 3 )) with ψ ( T ) = 0 \psi(T)=0 ψ ( T ) = 0 . The Galerkin director equation reads
∫ 0 T ∫ Ω [ − d k ⋅ ∂ t ψ + ( u k ⋅ ∇ ) d k ⋅ ψ + ∇ d k : ∇ ψ + f ε ( d k ) ⋅ ψ + n φ k ( d k − b k ) ⋅ ψ ] d x d t = ∫ Ω d 0 ⋅ ψ ( 0 ) d x . (75) \int_{0}^{T}\int_{\Omega}\left[-d^{k}\cdot\partial_{t}\psi+(u^{k}\cdot\nabla)d^{k}\cdot\psi+\nabla d^{k}:\nabla\psi+f_{\varepsilon}(d^{k})\cdot\psi+n\varphi^{k}(d^{k}-b^{k})\cdot\psi\right]\,\mathrm{d}x\,\mathrm{d}t=\int_{\Omega}d_{0}\cdot\psi(0)\,\mathrm{d}x. \tag*{(75)} ∫ 0 T ∫ Ω [ − d k ⋅ ∂ t ψ + ( u k ⋅ ∇ ) d k ⋅ ψ + ∇ d k : ∇ ψ + f ε ( d k ) ⋅ ψ + n φ k ( d k − b k ) ⋅ ψ ] d x d t = ∫ Ω d 0 ⋅ ψ ( 0 ) d x . (75)
By (3.30), we can pass the limit for the first and diffusion terms. Since ε > 0 \varepsilon>0 ε > 0 is fixed and ∣ d k ∣ ≤ 1 |d^{k}|\le1 ∣ d k ∣ ≤ 1 , the map f ε f_{\varepsilon} f ε is Lipschitz on the range of the sequence with a constant bounded by C ε − 2 C\varepsilon^{-2} C ε − 2 , uniformly with respect to k k k . Hence (3.33) implies that f ε ( d k ) → f ε ( d ) f_{\varepsilon}(d^{k})\to f_{\varepsilon}(d) f ε ( d k ) → f ε ( d ) strongly in L 2 ( ( 0 , T ) × Ω ) L^{2}((0,T)\times\Omega) L 2 (( 0 , T ) × Ω ) . Similarly, by (3.36), (3.37), and (3.33) we obtain φ k ( d k − b k ) → φ ( d − b ) \varphi^{k}(d^{k}-b^{k})\to\varphi(d-b) φ k ( d k − b k ) → φ ( d − b ) strongly in L 2 ( ( 0 , T ) × Ω ) L^{2}((0,T)\times\Omega) L 2 (( 0 , T ) × Ω ) . For the transport term, choose any s > 3 / 2 s>3/2 s > 3/2 in (3.32), then we have u k → u u^{k}\to u u k → u strongly in L t 2 L x ∞ L^{2}_{t}L^{\infty}_{x} L t 2 L x ∞ and ∇ d k \nabla d^{k} ∇ d k is bounded in L t ∞ L x 2 L^{\infty}_{t}L^{2}_{x} L t ∞ L x 2 . Hence, ( u k ⋅ ∇ ) d k (u^{k}\cdot\nabla)d^{k} ( u k ⋅ ∇ ) d k is bounded in L 2 ( 0 , T ; L 2 ( Ω ) ) L^{2}(0,T;L^{2}(\Omega)) L 2 ( 0 , T ; L 2 ( Ω )) . For an arbitrary ψ ∈ L 2 ( 0 , T ; H 0 1 ( Ω ) ) \psi\in L^{2}(0,T;H_{0}^{1}(\Omega)) ψ ∈ L 2 ( 0 , T ; H 0 1 ( Ω )) , write the difference of the transport pairings as
∫ 0 T ∫ Ω ( ( u k − u ) ⋅ ∇ ) d k ⋅ ψ d x d t + ∫ 0 T ∫ Ω ( u ⋅ ∇ ) ( d k − d ) ⋅ ψ d x d t . \int_{0}^{T}\int_{\Omega}((u^{k}-u)\cdot\nabla)d^{k}\cdot\psi\,\mathrm{d}x\,\mathrm{d}t+\int_{0}^{T}\int_{\Omega}(u\cdot\nabla)(d^{k}-d)\cdot\psi\,\mathrm{d}x\,\mathrm{d}t. ∫ 0 T ∫ Ω (( u k − u ) ⋅ ∇ ) d k ⋅ ψ d x d t + ∫ 0 T ∫ Ω ( u ⋅ ∇ ) ( d k − d ) ⋅ ψ d x d t .
The first term tends to zero by the strong L t 2 L x ∞ L^{2}_{t}L^{\infty}_{x} L t 2 L x ∞ -convergence of u k u^{k} u k , the uniform L t ∞ L x 2 L^{\infty}_{t}L^{2}_{x} L t ∞ L x 2 -bound for ∇ d k \nabla d^{k} ∇ d k , and ψ ∈ L t 2 L x 2 \psi\in L^{2}_{t}L^{2}_{x} ψ ∈ L t 2 L x 2 . For the second term, u ψ ∈ L t 1 L x 2 u\psi\in L^{1}_{t}L^{2}_{x} u ψ ∈ L t 1 L x 2 , so the weak-* convergence of ∇ d k \nabla d^{k} ∇ d k in L t ∞ L x 2 L^{\infty}_{t}L^{2}_{x} L t ∞ L x 2 implies the desired limit. Thus the transport term passes directly for the full stated energy test space. Passing to the limit in (75) we yield (43) .
Finally, the constitutive identity
M n , k = Δ d k − f ε ( d k ) − n φ k ( d k − b k ) M^{n,k}=\Delta d^k-f_\varepsilon(d^k)-n\varphi^k(d^k-b^k) M n , k = Δ d k − f ε ( d k ) − n φ k ( d k − b k )
passes in distributions, using the weak L t 2 H x 2 L^2_tH^2_x L t 2 H x 2 -convergence of d k d^k d k and the two strong convergences just proved. Hence
M = Δ d − f ε ( d ) − n φ ( d − b ) a.e. in ( 0 , T ) × Ω , M=\Delta d-f_\varepsilon(d)-n\varphi(d-b) \quad\text{a.e. in } (0,T)\times\Omega, M = Δ d − f ε ( d ) − n φ ( d − b ) a.e. in ( 0 , T ) × Ω ,
which also identifies the molecular field appearing in the limiting momentum equation. It remains to pass to the energy inequality and to preserve separation from the wall. From (69) and (70) , after one subsequence
u k ( t ) → u ( t ) strongly in L 2 ( Ω ) , d k ( t ) → d ( t ) strongly in H 1 ( Ω ) u^k(t)\to u(t) \quad\text{strongly in } L^2(\Omega), \qquad d^k(t)\to d(t) \quad\text{strongly in } H^1(\Omega) u k ( t ) → u ( t ) strongly in L 2 ( Ω ) , d k ( t ) → d ( t ) strongly in H 1 ( Ω )
for a.e. t t t . At those times the kinetic, director-gradient, Ginzburg–Landau, and director-penalty parts of the energy pass by strong convergence (using also (73) –(74) ). The viscous, artificial-viscosity, and molecular-field dissipations pass by weak lower semicontinuity. Taking the liminf in (60) therefore gives (47) for a.e. t t t . Finally, the separation estimate (48) holds for every k k k by Proposition 3.7 and passes to the limit by the uniform convergence (72) of the flows. This completes the proof. □
Rigidity limit n n n to ∞ \infty ∞
We pass to the rigidity limit n → ∞ n\to\infty n → ∞ with δ \delta δ and ε \varepsilon ε fixed. The director is compact on every spacetime region that remains a positive distance from the limiting solid. For the momentum equation this local compactness is sufficient. Admissible test fields are rigid in a neighborhood of the body, and the symmetric Ericksen tensor contracts only with the symmetric gradient of the test field, hence contributes nothing in that neighborhood. The velocity penalization gives φ D ( u ) = 0 \varphi_D(u)=0 φ D ( u ) = 0 in the limit, from which the limiting transported solid is a genuine rigid motion.
Fix T > 0 T>0 T > 0 . By Theorem 3.4, the penalized problem has a solution ( u n , d n , φ n , b n ) (u^n,d^n,\varphi^n,b^n) ( u n , d n , φ n , b n ) on [ 0 , T ] [0,T] [ 0 , T ] for every n n n , and, by (48) , all approximate solids satisfy dist ( S ( φ n ( t ) ) , ∂ Ω ) ≥ ρ δ ( T ) > 0 \operatorname{dist}(\mathcal{S}(\varphi^n(t)),\partial\Omega)\ge\rho_\delta(T)>0 dist ( S ( φ n ( t )) , ∂ Ω ) ≥ ρ δ ( T ) > 0 for 0 ≤ t ≤ T 0\le t\le T 0 ≤ t ≤ T . The energy inequality (47) provides bounds on u n u^n u n and d n d^n d n that are uniform in n n n . Applying the Banach–Alaoglu theorem, we can extract a subsequence (not relabeled) such that
u n ⇀ u weakly in L 2 ( 0 , T ; H 3 ( Ω ; R 3 ) ) and weak-* in L ∞ ( 0 , T ; H ( Ω ) ) , (76) u^n\rightharpoonup u \quad\text{weakly in } L^2(0,T;H^3(\Omega;\mathbb{R}^3)) \text{ and weak-* in } L^\infty(0,T;\mathcal{H}(\Omega)), \tag*{(76)} u n ⇀ u weakly in L 2 ( 0 , T ; H 3 ( Ω ; R 3 )) and weak-* in L ∞ ( 0 , T ; H ( Ω )) , (76)
d n ⇀ d weak-* in L ∞ ( 0 , T ; H 1 ( Ω ; R 3 ) ) , (77) d^n\rightharpoonup d \quad\text{weak-* in } L^\infty(0,T;H^1(\Omega;\mathbb{R}^3)), \tag*{(77)} d n ⇀ d weak-* in L ∞ ( 0 , T ; H 1 ( Ω ; R 3 )) , (77)
M n ⇀ M weakly in L 2 ( 0 , T ; L 2 ( Ω ; R 3 ) ) , (78) M^n\rightharpoonup M \quad\text{weakly in } L^2(0,T;L^2(\Omega;\mathbb{R}^3)), \tag*{(78)} M n ⇀ M weakly in L 2 ( 0 , T ; L 2 ( Ω ; R 3 )) , (78)
φ n ⇀ φ weak-* in L ∞ ( 0 , T ; L ∞ ( Ω ) ) , (79) \varphi^n\rightharpoonup\varphi\quad\text{weak-* in } L^\infty(0,T;L^\infty(\Omega)), \tag*{(79)} φ n ⇀ φ weak-* in L ∞ ( 0 , T ; L ∞ ( Ω )) , (79)
for some limit functions
u ∈ L 2 ( 0 , T ; H 3 ( Ω ; R 3 ) ) ∩ L ∞ ( 0 , T ; H ( Ω ) ) , u\in L^2(0,T;H^3(\Omega;\mathbb{R}^3))\cap L^\infty(0,T;\mathcal{H}(\Omega)), u ∈ L 2 ( 0 , T ; H 3 ( Ω ; R 3 )) ∩ L ∞ ( 0 , T ; H ( Ω )) ,
d ∈ L ∞ ( 0 , T ; H 1 ( Ω ; R 3 ) ) , M ∈ L 2 ( ( 0 , T ) × Ω ; R 3 ) , 0 ≤ φ ≤ 1. d\in L^\infty(0,T;H^1(\Omega;\mathbb{R}^3)), \qquad M\in L^2((0,T)\times\Omega;\mathbb{R}^3), \qquad0\le\varphi\le1. d ∈ L ∞ ( 0 , T ; H 1 ( Ω ; R 3 )) , M ∈ L 2 (( 0 , T ) × Ω ; R 3 ) , 0 ≤ φ ≤ 1.
We first show that the flows of the approximate velocities converge, identify the limits of the transported indicator and lifting, and deduce that the limiting solid moves rigidly. For fixed δ \delta δ , by the energy estimate and the embedding H 3 ( Ω ) ↪ W 1 , ∞ ( Ω ) ∩ W 2 , 6 ( Ω ) H^3(\Omega)\hookrightarrow W^{1,\infty}(\Omega)\cap W^{2,6}(\Omega) H 3 ( Ω ) ↪ W 1 , ∞ ( Ω ) ∩ W 2 , 6 ( Ω ) , one has uniform bi-Lipschitz flow bounds. Differentiating the flow equation twice in space, we have
d d t ∇ 2 X 0 , t n = ( ∇ 2 u n ) ∘ X 0 , t n [ ∇ X 0 , t n , ∇ X 0 , t n ] + ( ∇ u n ) ∘ X 0 , t n ∇ 2 X 0 , t n , \frac{\mathrm{d}}{\mathrm{d}t}\nabla^2X^n_{0,t}=(\nabla^2u^n)\circ X^n_{0,t}[\nabla X^n_{0,t},\nabla X^n_{0,t}]+(\nabla u^n)\circ X^n_{0,t}\nabla^2X^n_{0,t}, d t d ∇ 2 X 0 , t n = ( ∇ 2 u n ) ∘ X 0 , t n [ ∇ X 0 , t n , ∇ X 0 , t n ] + ( ∇ u n ) ∘ X 0 , t n ∇ 2 X 0 , t n ,
and using volume preservation and Grönwall’s inequality one obtains
sup n sup 0 ≤ t ≤ T ( ∥ X 0 , t n ∥ W 2 , 6 ( Ω ) + ∥ Y n ( t ) ∥ W 2 , 6 ( Ω ) ) ≤ C δ . \sup_n\sup_{0\le t\le T}\left(\|X^n_{0,t}\|_{W^{2,6}(\Omega)}+\|Y^n(t)\|_{W^{2,6}(\Omega)}\right)\le C_\delta. n sup 0 ≤ t ≤ T sup ( ∥ X 0 , t n ∥ W 2 , 6 ( Ω ) + ∥ Y n ( t ) ∥ W 2 , 6 ( Ω ) ) ≤ C δ .
The inverse estimate follows from the derivative formulas for inverse maps. The maps and their inverses are also uniformly 1 / 2 1/2 1/2 -Hölder in time with values in C ( Ω ‾ ) C(\overline{\Omega}) C ( Ω ) . By the Arzelà–Ascoli theorem, we get, up to a subsequence,
X 0 , t n → X 0 , t , Y n ( t , ⋅ ) → Y ( t , ⋅ ) uniformly on [ 0 , T ] × Ω ‾ . (80) X^n_{0,t} \to X_{0,t}, \qquad Y^n(t,\cdot) \to Y(t,\cdot) \quad\text{uniformly on } [0,T]\times\overline{\Omega}. \tag*{(80)} X 0 , t n → X 0 , t , Y n ( t , ⋅ ) → Y ( t , ⋅ ) uniformly on [ 0 , T ] × Ω . (80)
These are the flows of the weak limit u \mathbf{u} u . Indeed, for fixed t t t and y \mathbf{y} y , the error in the integral flow equation splits into
∫ 0 t [ u n ( s , X 0 , s n ( y ) ) − u n ( s , X 0 , s ( y ) ) ] d s + ∫ 0 t ( u n − u ) ( s , X 0 , s ( y ) ) d s . \int_0^t \left[\mathbf{u}^n(s,X^n_{0,s}(\mathbf{y}))-\mathbf{u}^n(s,X_{0,s}(\mathbf{y}))\right]\,\mathrm{d}s+\int_0^t(\mathbf{u}^n-\mathbf{u})(s,X_{0,s}(\mathbf{y}))\,\mathrm{d}s. ∫ 0 t [ u n ( s , X 0 , s n ( y )) − u n ( s , X 0 , s ( y )) ] d s + ∫ 0 t ( u n − u ) ( s , X 0 , s ( y )) d s .
The first term tends to 0 0 0 by the uniform convergence of the maps and the L t 1 W x 1 , ∞ L^1_tW^{1,\infty}_x L t 1 W x 1 , ∞ -bound. The second term tends to 0 0 0 by weak L t 2 H x 3 L^2_tH^3_x L t 2 H x 3 -convergence, since evaluation along the fixed continuous path is a bounded element of L 2 ( 0 , T ; ( H 3 ( Ω ) ) ′ ) L^2(0,T;(H^3(\Omega))') L 2 ( 0 , T ; ( H 3 ( Ω ) ) ′ ) . Uniqueness of the Lipschitz flow identifies the limits. They preserve volume and fix ∂ Ω \partial\Omega ∂ Ω .
The indicator and lifting convergence now follows exactly as in (73) and (74) , namely
φ n → φ : = φ 0 ∘ Y in C ( [ 0 , T ] ; L p ( Ω ) ) , p < ∞ , \varphi^n \to\varphi:= \varphi_0\circ\mathbf{Y}\quad\text{in }C([0,T];L^p(\Omega)),\quad p<\infty, φ n → φ := φ 0 ∘ Y in C ([ 0 , T ] ; L p ( Ω )) , p < ∞ ,
b n → b : = d 0 ∘ Y in C ( [ 0 , T ] ; L ∞ ( Ω ) ) . (81) b^n \to b := d_0\circ\mathbf{Y}\quad\text{in }C([0,T];L^\infty(\Omega)). \tag*{(81)} b n → b := d 0 ∘ Y in C ([ 0 , T ] ; L ∞ ( Ω )) . (81)
Here the indicator estimate uses ∣ ∂ S 0 ∣ = 0 |\partial S_0|=0 ∣ ∂ S 0 ∣ = 0 , and the lifting estimate uses the Hölder continuity of d 0 ∈ H 2 ( Ω ) d_0\in H^2(\Omega) d 0 ∈ H 2 ( Ω ) . In particular, with S ( t ) : = X 0 , t ( S 0 ) S(t):=X_{0,t}(S_0) S ( t ) := X 0 , t ( S 0 ) ,
sup t ∈ [ 0 , T ] d H ( S ( φ n ( t ) ) , S ( t ) ) → 0. (82) \sup_{t\in[0,T]}d_H(S(\varphi^n(t)),S(t))\to0. \tag*{(82)} t ∈ [ 0 , T ] sup d H ( S ( φ n ( t )) , S ( t )) → 0. (82)
Thus both transport identities pass to the limit, and the following inclusions hold.
Lemma 3.9. For every σ > 0 \sigma>0 σ > 0 and all sufficiently large n n n ,
S ( φ n ( t ) ) ⊂ S ( t ) σ , S ( t ) ⊂ S ( φ n ( t ) ) σ ( 0 ≤ t ≤ T ) . S(\varphi^n(t))\subset S(t)_\sigma,\qquad S(t)\subset S(\varphi^n(t))_\sigma\quad(0\le t\le T). S ( φ n ( t )) ⊂ S ( t ) σ , S ( t ) ⊂ S ( φ n ( t ) ) σ ( 0 ≤ t ≤ T ) .
The two penalty bounds give
∥ φ n D ( u n ) ∥ L t , x 2 ≤ C n − 1 / 2 , ∥ φ n ( d n − b n ) ∥ L t ∞ L x 2 ≤ C n − 1 / 2 . \|\varphi^n\mathbb{D}(\mathbf{u}^n)\|_{L^2_{t,x}}\le Cn^{-1/2},\qquad\|\varphi^n(d^n-b^n)\|_{L^\infty_tL^2_x}\le Cn^{-1/2}. ∥ φ n D ( u n ) ∥ L t , x 2 ≤ C n − 1/2 , ∥ φ n ( d n − b n ) ∥ L t ∞ L x 2 ≤ C n − 1/2 .
Combining them with (81) and the weak convergences yields φ D ( u ) = 0 \varphi\mathbb{D}(\mathbf{u})=0 φ D ( u ) = 0 and φ ( d − b ) = 0 \varphi(d-b)=0 φ ( d − b ) = 0 . Set h ( t ) = ∣ S 0 ∣ − 1 ∫ S ( t ) x d x \mathbf{h}(t)=|S_0|^{-1}\int_{S(t)}\mathbf{x}\,\mathrm{d}x h ( t ) = ∣ S 0 ∣ − 1 ∫ S ( t ) x d x . The connected transported set therefore has the rigid velocity ℓ ( t ) + ω ( t ) × ( x − h ( t ) ) \ell(t)+\omega(t)\times(\mathbf{x}-\mathbf{h}(t)) ℓ ( t ) + ω ( t ) × ( x − h ( t )) . Differentiating its center of mass along the flow gives h ′ = ℓ \mathbf{h}'=\ell h ′ = ℓ . The relative trajectories solve ∂ t ( X 0 , t ( y ) − h ) = ω × ( X 0 , t ( y ) − h ) \partial_t(X_{0,t}(\mathbf{y})-\mathbf{h})=\omega\times(X_{0,t}(\mathbf{y})-\mathbf{h}) ∂ t ( X 0 , t ( y ) − h ) = ω × ( X 0 , t ( y ) − h ) , whence
X 0 , t ( y ) = h ( t ) + O ( t ) y ( y ∈ S 0 ) , O ′ = m O , O ( 0 ) = I 3 . X_{0,t}(\mathbf{y})=\mathbf{h}(t)+O(t)\mathbf{y}\quad(\mathbf{y}\in S_0),\qquad O'=mO,\quad O(0)=\mathbb{I}_3. X 0 , t ( y ) = h ( t ) + O ( t ) y ( y ∈ S 0 ) , O ′ = m O , O ( 0 ) = I 3 .
Consequently S ( t ) = h ( t ) + O ( t ) S 0 S(t)=\mathbf{h}(t)+O(t)S_0 S ( t ) = h ( t ) + O ( t ) S 0 and
d = b = d 0 ( O ( t ) ⊤ ( x − h ( t ) ) ) a.e. in S ( t ) . (83) \mathbf{d}=\mathbf{b}=d_0(O(t)^\top(\mathbf{x}-\mathbf{h}(t)))\quad\text{a.e. in }S(t). \tag*{(83)} d = b = d 0 ( O ( t ) ⊤ ( x − h ( t ))) a.e. in S ( t ) . (83)
Next, we prove that the director converges strongly away from the solid and that the velocity converges strongly in the whole container. By (83) , the lifted director field w : = d − b \mathbf{w}:=\mathbf{d}-\mathbf{b} w := d − b vanishes almost everywhere in S ( t ) S(t) S ( t ) .
The director sequence is locally strongly compact as follows. The proof is given in Section A.4.
Lemma 3.10. For every σ > 0 \sigma>0 σ > 0 , the sequence { d n } \{d^n\} { d n } converges strongly to d d d in L 2 ( 0 , T ; H 1 ( F ~ σ ( t ) ; R 3 ) ) L^2(0,T;H^1(\widetilde{F}_\sigma(t);\mathbb{R}^3)) L 2 ( 0 , T ; H 1 ( F σ ( t ) ; R 3 )) , where
F ~ σ ( t ) : = Ω ∖ S ( φ ( t ) ) σ = { x ∈ Ω : dist ( x , S ( φ ( t ) ) ) > σ } . \widetilde{F}_\sigma(t):=\Omega\setminus S(\varphi(t))_\sigma=\{x\in\Omega:\operatorname{dist}(x,S(\varphi(t)))>\sigma\}. F σ ( t ) := Ω ∖ S ( φ ( t ) ) σ = { x ∈ Ω : dist ( x , S ( φ ( t ))) > σ } . Thus F ~ σ ( t ) \widetilde{F}_\sigma(t) F σ ( t ) is an open set compactly separated from the solid whenever σ \sigma σ is below the tubular radius.
The velocity compactness is based on the following criterion. Its proof can be found in Section A.4.
Lemma 3.11. Let S ( t ) = h ( t ) + O ( t ) S 0 S(t)=h(t)+O(t)S_0 S ( t ) = h ( t ) + O ( t ) S 0 be a rigid trajectory of a connected C 2 C^2 C 2 -set S 0 S_0 S 0 . Let χ n = 1 S n ( t ) \chi_n=\mathbf{1}_{S_n(t)} χ n = 1 S n ( t ) and χ = 1 S ( t ) \chi=\mathbf{1}_{S(t)} χ = 1 S ( t ) , where S n ( t ) = Φ n ( t , S 0 ) S_n(t)=\Phi_n(t,S_0) S n ( t ) = Φ n ( t , S 0 ) . Assume that every Φ n ( t ) \Phi_n(t) Φ n ( t ) is volume preserving on a fixed neighborhood of S 0 S_0 S 0 and that
sup n , t ( ∥ Φ n ( t ) ∥ W 1 , ∞ + ∥ Φ n ( t ) − 1 ∥ W 1 , ∞ + ∥ Φ n ( t ) ∥ W 2 , 6 + ∥ Φ n ( t ) − 1 ∥ W 2 , 6 ) < ∞ . (84) \sup_{n,t}\left(\|\Phi_n(t)\|_{W^{1,\infty}}+\|\Phi_n(t)^{-1}\|_{W^{1,\infty}}+\|\Phi_n(t)\|_{W^{2,6}}+\|\Phi_n(t)^{-1}\|_{W^{2,6}}\right)<\infty. \tag*{(84)} n , t sup ( ∥ Φ n ( t ) ∥ W 1 , ∞ + ∥ Φ n ( t ) − 1 ∥ W 1 , ∞ + ∥ Φ n ( t ) ∥ W 2 , 6 + ∥ Φ n ( t ) − 1 ∥ W 2 , 6 ) < ∞. (84) Assume also that S n ( t ) → S ( t ) S_n(t)\to S(t) S n ( t ) → S ( t ) uniformly in Hausdorff distance, the sets remain uniformly separated from ∂ Ω \partial\Omega ∂ Ω , and χ n → χ \chi_n\to\chi χ n → χ strongly in C ( [ 0 , T ] ; L p ( Ω ) ) C([0,T];L^p(\Omega)) C ([ 0 , T ] ; L p ( Ω )) for every finite p p p . Suppose
v n ⇀ ∗ v in L ∞ ( 0 , T ; H ( Ω ) ) , v n ⇀ v in L 2 ( 0 , T ; V ( Ω ) ) , v_n \overset{*}{\rightharpoonup} v\ \text{in }L^\infty(0,T;\mathcal{H}(\Omega)),\qquad v_n\rightharpoonup v\ \text{in }L^2(0,T;\mathcal{V}(\Omega)), v n ⇀ ∗ v in L ∞ ( 0 , T ; H ( Ω )) , v n ⇀ v in L 2 ( 0 , T ; V ( Ω )) , as well as
sup n ∥ v n ∥ L 2 ( 0 , T ; H 3 ( Ω ) ) < ∞ , and (85) \sup_n\|v_n\|_{L^2(0,T;H^3(\Omega))}<\infty,\qquad\text{and} \tag*{(85)} n sup ∥ v n ∥ L 2 ( 0 , T ; H 3 ( Ω )) < ∞ , and (85) n ∫ 0 T ∫ Ω χ n ∣ D ( v n ) ∣ 2 d x d t ≤ C . (86) n\int_0^T\int_\Omega\chi_n|\mathbb{D}(v_n)|^2\,\mathrm{d}x\,\mathrm{d}t\leq C. \tag*{(86)} n ∫ 0 T ∫ Ω χ n ∣ D ( v n ) ∣ 2 d x d t ≤ C . (86) Assume in addition that, for every sufficiently small σ > 0 \sigma>0 σ > 0 , [ 0 , T ] [0,T] [ 0 , T ] admits a finite covering by intervals I j I_j I j and fixed connected C 2 C^2 C 2 -sets S j ⋐ Ω S_j\Subset\Omega S j ⋐ Ω satisfying
S n ( t ) ⊂ S j ⊂ S ( t ) σ ( t ∈ I j , n ≥ n σ ) , S_n(t)\subset S_j\subset S(t)_\sigma\qquad(t\in I_j,\ n\geq n_\sigma), S n ( t ) ⊂ S j ⊂ S ( t ) σ ( t ∈ I j , n ≥ n σ ) , and there is r ∗ > 0 r_*>0 r ∗ > 0 , independent of small σ \sigma σ and of j j j , such that
D j : = ( S j ) r ∗ ∖ S j ‾ ⋐ Ω D_j:=(S_j)_{r_*}\setminus\overline{S_j}\Subset\Omega D j := ( S j ) r ∗ ∖ S j ⋐ Ω are C 2 C^2 C 2 -collars with uniformly bounded C 2 C^2 C 2 -character. Assume finally that on each I j I_j I j there is a constant C σ , j C_{\sigma,j} C σ , j , independent of n n n , such that
∣ ∫ I j ∫ Ω v n ⋅ ∂ t ζ d x d t ∣ ≤ C σ , j ( ∥ ζ ∥ L ∞ ( I j ; H 3 ( Ω ) ) + ∥ ζ ∥ L 2 ( I j ; H 3 ( Ω ) ) ) \left|\int_{I_j}\int_\Omega v_n\cdot\partial_t\zeta\,\mathrm{d}x\,\mathrm{d}t\right|
\leq C_{\sigma,j}\left(\|\zeta\|_{L^\infty(I_j;H^3(\Omega))}+\|\zeta\|_{L^2(I_j;H^3(\Omega))}\right) ∫ I j ∫ Ω v n ⋅ ∂ t ζ d x d t ≤ C σ , j ( ∥ ζ ∥ L ∞ ( I j ; H 3 ( Ω )) + ∥ ζ ∥ L 2 ( I j ; H 3 ( Ω )) ) for every smooth, compactly time-supported, divergence-free ζ \zeta ζ with D ( ζ ) = 0 \mathbb{D}(\zeta)=0 D ( ζ ) = 0 on S j S_j S j . Then
v n → v strongly in L 2 ( 0 , T ; L 2 ( Ω ) ) . v_n\to v\qquad\text{strongly in }L^2(0,T;L^2(\Omega)). v n → v strongly in L 2 ( 0 , T ; L 2 ( Ω )) .
The application of the preceding criterion to the present sequence is stated next. Its proof is given in Section A.4.
Lemma 3.12. The sequence { u n } \{u^n\} { u n } in (76) converges strongly to u u u in L 2 ( 0 , T ; L 2 ( Ω ; R 3 ) ) L^2(0,T;L^2(\Omega;\mathbb{R}^3)) L 2 ( 0 , T ; L 2 ( Ω ; R 3 )) .
Observe that the strong convergence of u n → u u^n\to u u n → u in L 2 ( 0 , T ; L 2 ( Ω ; R 3 ) ) L^2(0,T;L^2(\Omega;\mathbb{R}^3)) L 2 ( 0 , T ; L 2 ( Ω ; R 3 )) together with the uniform boundedness of u n u^n u n in L 2 ( 0 , T ; H 3 ( Ω ; R 3 ) ) L^2(0,T;H^3(\Omega;\mathbb{R}^3)) L 2 ( 0 , T ; H 3 ( Ω ; R 3 )) and interpolation implies that u n → u u^n\to u u n → u strongly in L 2 ( 0 , T ; H s ( Ω ; R 3 ) ) L^2(0,T;H^s(\Omega;\mathbb{R}^3)) L 2 ( 0 , T ; H s ( Ω ; R 3 )) for all s < 3 s<3 s < 3 . In particular, choosing s ∈ ( 5 2 , 3 ) s\in\left(\frac{5}{2},3\right) s ∈ ( 2 5 , 3 ) , we find that
u n → u strongly in L 1 ( 0 , T ; W 1 , ∞ ( Ω ; R 3 ) ) . u^n\to u\qquad\text{strongly in }L^1(0,T;W^{1,\infty}(\Omega;\mathbb{R}^3)). u n → u strongly in L 1 ( 0 , T ; W 1 , ∞ ( Ω ; R 3 )) .
Consequently, the limits X ^ \widehat{X} X and Y ^ \widehat{Y} Y obtained in (80) indeed coincide with the flow map X X X and inverse flow map Y Y Y associated with the limit fluid velocity field u u u , i.e., X ^ = X \widehat{X}=X X = X and Y ^ = Y \widehat{Y}=Y Y = Y .
We now pass to the limit in the director equation, first for test functions supported in the fluid region and then, by density, for the full class of energy tests. Recall the fluid spacetime Q F Q_F Q F from § 2.2 , now associated with the limiting rigid trajectory. Let ψ ∈ C c ∞ ( [ 0 , T ) × Ω ; R 3 ) \psi\in C^\infty_c([0,T)\times\Omega;\mathbb{R}^3) ψ ∈ C c ∞ ([ 0 , T ) × Ω ; R 3 ) with supp ψ ⊂ Q F ∪ ( { 0 } × F 0 ) \operatorname{supp}\psi\subset Q_F\cup(\{0\}\times F_0) supp ψ ⊂ Q F ∪ ({ 0 } × F 0 ) . By compactness of the support and continuity of the rigid body motion, there exists σ > 0 \sigma>0 σ > 0 such that
ψ ( t , ⋅ ) = 0 on S ( t ) σ for all t ∈ [ 0 , T ] . \psi(t,\cdot)=0\quad\text{on }S(t)_\sigma\qquad\text{for all }t\in[0,T]. ψ ( t , ⋅ ) = 0 on S ( t ) σ for all t ∈ [ 0 , T ] .
From Lemma 3.9 , for n n n sufficiently large, we deduce that S ( φ n ( t ) ) ⊂ S ( φ ( t ) ) σ / 2 \mathcal{S}(\varphi^{n}(t)) \subset\mathcal{S}(\varphi(t))_{\sigma/2} S ( φ n ( t )) ⊂ S ( φ ( t ) ) σ /2 for all t ∈ [ 0 , T ] t \in[0,T] t ∈ [ 0 , T ] and therefore φ n ψ = 0 \varphi^{n}\psi=0 φ n ψ = 0 a.e. in ( 0 , T ) × Ω (0,T)\times\Omega ( 0 , T ) × Ω . Using ψ \psi ψ in (43) , the director penalization term n φ n ( d n − b n ) ⋅ ψ n\varphi^{n}(d^{n}-b^{n})\cdot\psi n φ n ( d n − b n ) ⋅ ψ vanishes identically for all sufficiently large n n n , and we obtain
∫ 0 T ∫ Ω [ − d n ⋅ ∂ t ψ + ( u n ⋅ ∇ ) d n ⋅ ψ + ∇ d n : ∇ ψ + f ε ( d n ) ⋅ ψ ] d x d t = ∫ F 0 d 0 ⋅ ψ ( 0 ) d x . (87) \int_{0}^{T}\int_{\Omega}\left[-d^{n}\cdot\partial_{t}\psi+(u^{n}\cdot\nabla)d^{n}\cdot\psi+\nabla d^{n}:\nabla\psi+f_{\varepsilon}(d^{n})\cdot\psi\right]\,\mathrm{d}x\,\mathrm{d}t=\int_{\mathcal{F}_{0}}d_{0}\cdot\psi(0)\,\mathrm{d}x. \tag*{(87)} ∫ 0 T ∫ Ω [ − d n ⋅ ∂ t ψ + ( u n ⋅ ∇ ) d n ⋅ ψ + ∇ d n : ∇ ψ + f ε ( d n ) ⋅ ψ ] d x d t = ∫ F 0 d 0 ⋅ ψ ( 0 ) d x . (87)
Since ψ \psi ψ vanishes on S ( φ ( t ) ) σ \mathcal{S}(\varphi(t))_{\sigma} S ( φ ( t ) ) σ , its support is contained in { x : dist ( x , S ( φ ( t ) ) ) ≥ σ } \{x:\operatorname{dist}(x,\mathcal{S}(\varphi(t)))\geq\sigma\} { x : dist ( x , S ( φ ( t ))) ≥ σ } and hence is compactly contained in the open set F ~ σ / 2 ( t ) \widetilde{\mathcal{F}}_{\sigma/2}(t) F σ /2 ( t ) . The local compactness result, Lemma 3.10 , therefore gives d n → d d^{n}\to d d n → d strongly in L 2 ( 0 , T ; H 1 ( F ~ σ / 2 ( t ) ; R 3 ) ) L^{2}(0,T;H^{1}(\widetilde{\mathcal{F}}_{\sigma/2}(t);\mathbb{R}^{3})) L 2 ( 0 , T ; H 1 ( F σ /2 ( t ) ; R 3 )) . Together with the strong convergence of u n u^{n} u n from Lemma 3.12 , this allows us to pass to the limit in (87) . By the properties of the support of ψ \psi ψ , we find
∫ 0 T ∫ F ( t ) [ − d ⋅ ∂ t ψ + ( u ⋅ ∇ ) d ⋅ ψ + ∇ d : ∇ ψ + f ε ( d ) ⋅ ψ ] d x d t = ∫ F 0 d 0 ⋅ ψ ( 0 ) d x . (88) \int_{0}^{T}\int_{\mathcal{F}(t)}\left[-d\cdot\partial_{t}\psi+(u\cdot\nabla)d\cdot\psi+\nabla d:\nabla\psi+f_{\varepsilon}(d)\cdot\psi\right]\,\mathrm{d}x\,\mathrm{d}t=\int_{\mathcal{F}_{0}}d_{0}\cdot\psi(0)\,\mathrm{d}x. \tag*{(88)} ∫ 0 T ∫ F ( t ) [ − d ⋅ ∂ t ψ + ( u ⋅ ∇ ) d ⋅ ψ + ∇ d : ∇ ψ + f ε ( d ) ⋅ ψ ] d x d t = ∫ F 0 d 0 ⋅ ψ ( 0 ) d x . (88)
Consequently,
∂ t d + ( u ⋅ ∇ ) d = Δ d − f ε ( d ) in D ′ ( Q F ) . \partial_{t}d+(u\cdot\nabla)d=\Delta d-f_{\varepsilon}(d)\quad\text{in } \mathcal{D}'(Q_{\mathcal{F}}). ∂ t d + ( u ⋅ ∇ ) d = Δ d − f ε ( d ) in D ′ ( Q F ) .
To identify the weak limit M M M on the fluid region, let ξ ∈ C c ∞ ( Q F ; R 3 ) \xi\in C_{c}^{\infty}(Q_{\mathcal{F}};\mathbb{R}^{3}) ξ ∈ C c ∞ ( Q F ; R 3 ) . For all sufficiently large n n n , one has φ n ξ = 0 \varphi^{n}\xi=0 φ n ξ = 0 . Hence the constitutive relation
M n = Δ d n − f ε ( d n ) − n φ n ( d n − b n ) M^{n}=\Delta d^{n}-f_{\varepsilon}(d^{n})-n\varphi^{n}(d^{n}-b^{n}) M n = Δ d n − f ε ( d n ) − n φ n ( d n − b n )
implies
∫ 0 T ∫ Ω [ M n ⋅ ξ + ∇ d n : ∇ ξ + f ε ( d n ) ⋅ ξ ] d x d t = 0. \int_{0}^{T}\int_{\Omega}\left[M^{n}\cdot\xi+\nabla d^{n}:\nabla\xi+f_{\varepsilon}(d^{n})\cdot\xi\right]\,\mathrm{d}x\,\mathrm{d}t=0. ∫ 0 T ∫ Ω [ M n ⋅ ξ + ∇ d n : ∇ ξ + f ε ( d n ) ⋅ ξ ] d x d t = 0.
Using the weak convergence of M n M^{n} M n from (78) , the weak convergence of ∇ d n \nabla d^{n} ∇ d n from (77) , and the strong local convergence of d n d^{n} d n as established in Lemma 3.10 , we pass to the limit and obtain
∫ 0 T ∫ Ω [ M ⋅ ξ + ∇ d : ∇ ξ + f ε ( d ) ⋅ ξ ] d x d t = 0. \int_{0}^{T}\int_{\Omega}\left[M\cdot\xi+\nabla d:\nabla\xi+f_{\varepsilon}(d)\cdot\xi\right]\,\mathrm{d}x\,\mathrm{d}t=0. ∫ 0 T ∫ Ω [ M ⋅ ξ + ∇ d : ∇ ξ + f ε ( d ) ⋅ ξ ] d x d t = 0.
Therefore,
M = Δ d − f ε ( d ) in D ′ ( Q F ) . M=\Delta d-f_{\varepsilon}(d)\quad\text{in } \mathcal{D}'(Q_{\mathcal{F}}). M = Δ d − f ε ( d ) in D ′ ( Q F ) .
Since both sides belong locally to L 2 ( Q F ) L^{2}(Q_{\mathcal{F}}) L 2 ( Q F ) , the equality holds almost everywhere in Q F Q_{\mathcal{F}} Q F . We extend (88) to the moving energy test class X d ( φ ) \mathcal{X}_{d}(\varphi) X d ( φ ) from (14) . Here and below, a function defined on F ( t ) \mathcal{F}(t) F ( t ) is identified with its extension by zero to S ( t ) \mathcal{S}(t) S ( t ) . We use the following density result, proved in Section A.4.
Lemma 3.13. Let D F : = { ψ ∈ C ∞ ( [ 0 , T ) × Ω ; R 3 ) : supp ψ ⊂ Q F ∪ ( { 0 } × F 0 ) } \mathcal{D}_{\mathcal{F}}:=\{\psi\in C^{\infty}([0,T)\times\Omega;\mathbb{R}^{3}):\operatorname{supp}\psi\subset Q_{\mathcal{F}}\cup(\{0\}\times\mathcal{F}_{0})\} D F := { ψ ∈ C ∞ ([ 0 , T ) × Ω ; R 3 ) : supp ψ ⊂ Q F ∪ ({ 0 } × F 0 )} . Then D F \mathcal{D}_{\mathcal{F}} D F is dense in the space X d ( φ ) \mathcal{X}_{d}(\varphi) X d ( φ ) defined in (14) with respect to the norm
∥ ψ ∥ H 1 ( 0 , T ; L 2 ( Ω ) ) + ∥ ψ ∥ L 2 ( 0 , T ; H 1 ( Ω ) ) . \|\psi\|_{H^{1}(0,T;L^{2}(\Omega))}+\|\psi\|_{L^{2}(0,T;H^{1}(\Omega))}. ∥ ψ ∥ H 1 ( 0 , T ; L 2 ( Ω )) + ∥ ψ ∥ L 2 ( 0 , T ; H 1 ( Ω )) . Moreover, the approximating sequence may be chosen such that ψ m ( 0 ) → ψ ( 0 ) \psi_{m}(0)\to\psi(0) ψ m ( 0 ) → ψ ( 0 ) in L 2 ( Ω ) L^{2}(\Omega) L 2 ( Ω ) .
By Lemma 3.13 , every ψ ∈ X d ( φ ) \psi\in\mathcal{X}_{d}(\varphi) ψ ∈ X d ( φ ) is the limit, in the norm of that lemma, of a sequence ( ψ m ) ⊂ D F (\psi_{m})\subset\mathcal{D}_{\mathcal{F}} ( ψ m ) ⊂ D F with ψ m ( 0 ) → ψ ( 0 ) \psi_{m}(0)\to\psi(0) ψ m ( 0 ) → ψ ( 0 ) in L 2 ( Ω ) L^{2}(\Omega) L 2 ( Ω ) , and (88) holds for each ψ m \psi_{m} ψ m . Every term in (88) is continuous with respect to this convergence, because
d ∈ L ∞ ( 0 , T ; H 1 ( Ω ; R 3 ) ) , u ∈ L 2 ( 0 , T ; L ∞ ( Ω ; R 3 ) ) , d\in L^{\infty}(0,T;H^{1}(\Omega;\mathbb{R}^{3})),\qquad u\in L^{2}(0,T;L^{\infty}(\Omega;\mathbb{R}^{3})), d ∈ L ∞ ( 0 , T ; H 1 ( Ω ; R 3 )) , u ∈ L 2 ( 0 , T ; L ∞ ( Ω ; R 3 )) ,
f ε ( d ) ∈ L ∞ ( ( 0 , T ) × Ω ; R 3 ) , f_{\varepsilon}(d)\in L^{\infty}((0,T)\times\Omega;\mathbb{R}^{3}), f ε ( d ) ∈ L ∞ (( 0 , T ) × Ω ; R 3 ) ,
where the last property follows from Lemma A.5 and the fixed value of ε \varepsilon ε . Consequently, for every ψ ∈ X d ( φ ) \psi\in\mathcal{X}_{d}(\varphi) ψ ∈ X d ( φ ) ,
∫ 0 T ∫ F ( t ) [ − d ⋅ ∂ t ψ + ( u ⋅ ∇ ) d ⋅ ψ + ∇ d : ∇ ψ + f ε ( d ) ⋅ ψ ] d x d t = ∫ F 0 d 0 ⋅ ψ ( 0 ) d x . \int_{0}^{T}\int_{\mathcal{F}(t)}\left[-d\cdot\partial_{t}\psi+(u\cdot\nabla)d\cdot\psi+\nabla d:\nabla\psi+f_{\varepsilon}(d)\cdot\psi\right]\,\mathrm{d}x\,\mathrm{d}t=\int_{\mathcal{F}_{0}}d_{0}\cdot\psi(0)\,\mathrm{d}x. ∫ 0 T ∫ F ( t ) [ − d ⋅ ∂ t ψ + ( u ⋅ ∇ ) d ⋅ ψ + ∇ d : ∇ ψ + f ε ( d ) ⋅ ψ ] d x d t = ∫ F 0 d 0 ⋅ ψ ( 0 ) d x .
We turn to the momentum equation, in which the local compactness of the director suffices because the test fields are rigid near the solid. Let ζ ∈ C c 1 ( [ 0 , T ) ; V ( Ω ) ∩ H 3 ( Ω ; R 3 ) ) \zeta\in C^{1}_{c}([0,T);\mathcal{V}(\Omega)\cap H^{3}(\Omega;\mathbb{R}^{3})) ζ ∈ C c 1 ([ 0 , T ) ; V ( Ω ) ∩ H 3 ( Ω ; R 3 )) be rigid in a collar of the limiting solid, i.e., for some σ > 0 \sigma>0 σ > 0 ,
D ( ζ ) = 0 a.e. on S ( t ) σ for all t ∈ [ 0 , T ] . (89) \mathbb{D}(\zeta)=0\quad\text{a.e. on }\mathcal{S}(t)_{\sigma}\qquad\text{for all }t\in[0,T]. \tag*{(89)} D ( ζ ) = 0 a.e. on S ( t ) σ for all t ∈ [ 0 , T ] . (89)
By the convergence of the Hausdorff distance as established in (82) , for all sufficiently large n n n , we get
S ( φ n ( t ) ) ⊂ S ( t ) σ / 2 for all t ∈ [ 0 , T ] . \mathcal{S}(\varphi^{n}(t))\subset\mathcal{S}(t)_{\sigma/2}\qquad\text{for all }t\in[0,T]. S ( φ n ( t )) ⊂ S ( t ) σ /2 for all t ∈ [ 0 , T ] .
In particular, φ n D ( ζ ) = 0 \varphi^{n}\mathbb{D}(\zeta)=0 φ n D ( ζ ) = 0 , and hence
n ∫ 0 T ∫ Ω φ n D ( u n ) : D ( ζ ) d x d t = 0. n\int_{0}^{T}\int_{\Omega}\varphi^{n}\mathbb{D}(u^{n}):\mathbb{D}(\zeta)\,\mathrm{d}x\,\mathrm{d}t=0. n ∫ 0 T ∫ Ω φ n D ( u n ) : D ( ζ ) d x d t = 0.
For the director contribution, the identity M n + n φ n ( d n − b n ) = Δ d n − f ε ( d n ) \mathbf{M}^{n}+n\varphi^{n}(d^{n}-b^{n})=\Delta d^{n}-f_{\varepsilon}(d^{n}) M n + n φ n ( d n − b n ) = Δ d n − f ε ( d n ) gives
− ∫ Ω ( ζ ⋅ ∇ ) d n ⋅ ( M n + n φ n ( d n − b n ) ) d x = − ∫ Ω ( ζ ⋅ ∇ ) d n ⋅ ( Δ d n − f ε ( d n ) ) d x . -\int_{\Omega}(\zeta\cdot\nabla)d^{n}\cdot\left(\mathbf{M}^{n}+n\varphi^{n}(d^{n}-b^{n})\right)\,\mathrm{d}x
=-\int_{\Omega}(\zeta\cdot\nabla)d^{n}\cdot\left(\Delta d^{n}-f_{\varepsilon}(d^{n})\right)\,\mathrm{d}x. − ∫ Ω ( ζ ⋅ ∇ ) d n ⋅ ( M n + n φ n ( d n − b n ) ) d x = − ∫ Ω ( ζ ⋅ ∇ ) d n ⋅ ( Δ d n − f ε ( d n ) ) d x .
Since f ε = ∇ d F ε f_{\varepsilon}=\nabla_{d}F_{\varepsilon} f ε = ∇ d F ε , div ζ = 0 \operatorname{div}\zeta=0 div ζ = 0 , and ζ = 0 \zeta=0 ζ = 0 on ∂ Ω \partial\Omega ∂ Ω ,
∫ Ω ( ζ ⋅ ∇ ) d n ⋅ f ε ( d n ) d x = ∫ Ω ζ ⋅ ∇ F ε ( d n ) d x = 0. \int_{\Omega}(\zeta\cdot\nabla)d^{n}\cdot f_{\varepsilon}(d^{n})\,\mathrm{d}x
=\int_{\Omega}\zeta\cdot\nabla F_{\varepsilon}(d^{n})\,\mathrm{d}x=0. ∫ Ω ( ζ ⋅ ∇ ) d n ⋅ f ε ( d n ) d x = ∫ Ω ζ ⋅ ∇ F ε ( d n ) d x = 0.
Integration by parts therefore yields
− ∫ Ω ( ζ ⋅ ∇ ) d n ⋅ ( M n + n φ n ( d n − b n ) ) d x = ∫ Ω ( ∇ d n ) T ∇ d n : ∇ ζ d x = ∫ Ω ( ∇ d n ) T ∇ d n : D ( ζ ) d x . -\int_{\Omega}(\zeta\cdot\nabla)d^{n}\cdot\left(\mathbf{M}^{n}+n\varphi^{n}(d^{n}-b^{n})\right)\,\mathrm{d}x
=\int_{\Omega}(\nabla d^{n})^{\mathsf{T}}\nabla d^{n}:\nabla\zeta\,\mathrm{d}x
=\int_{\Omega}(\nabla d^{n})^{\mathsf{T}}\nabla d^{n}:\mathbb{D}(\zeta)\,\mathrm{d}x. − ∫ Ω ( ζ ⋅ ∇ ) d n ⋅ ( M n + n φ n ( d n − b n ) ) d x = ∫ Ω ( ∇ d n ) T ∇ d n : ∇ ζ d x = ∫ Ω ( ∇ d n ) T ∇ d n : D ( ζ ) d x .
By (89) , the last integrand vanishes on S ( t ) σ \mathcal{S}(t)_{\sigma} S ( t ) σ . Hence,
∫ 0 T ∫ Ω ( ∇ d n ) T ∇ d n : ∇ ζ d x d t = ∫ 0 T ∫ F ~ σ ( t ) ( ∇ d n ) T ∇ d n : ∇ ζ d x d t . \int_{0}^{T}\int_{\Omega}(\nabla d^{n})^{\mathsf{T}}\nabla d^{n}:\nabla\zeta\,\mathrm{d}x\,\mathrm{d}t
=\int_{0}^{T}\int_{\widetilde{\mathcal{F}}_{\sigma}(t)}(\nabla d^{n})^{\mathsf{T}}\nabla d^{n}:\nabla\zeta\,\mathrm{d}x\,\mathrm{d}t. ∫ 0 T ∫ Ω ( ∇ d n ) T ∇ d n : ∇ ζ d x d t = ∫ 0 T ∫ F σ ( t ) ( ∇ d n ) T ∇ d n : ∇ ζ d x d t .
On the open set F ~ σ ( t ) \widetilde{\mathcal{F}}_{\sigma}(t) F σ ( t ) , the local strong compactness established in Lemma 3.10 gives
∇ d n → ∇ d strongly in L 2 ( 0 , T ; L 2 ( F ~ σ ( t ) ; R 3 × 3 ) ) . \nabla d^{n}\to\nabla d\quad\text{strongly in }L^{2}(0,T;L^{2}(\widetilde{\mathcal{F}}_{\sigma}(t);\mathbb{R}^{3\times3})). ∇ d n → ∇ d strongly in L 2 ( 0 , T ; L 2 ( F σ ( t ) ; R 3 × 3 )) .
Consequently, ( ∇ d n ) T ∇ d n → ( ∇ d ) T ∇ d (\nabla d^{n})^{\mathsf{T}}\nabla d^{n}\to(\nabla d)^{\mathsf{T}}\nabla d ( ∇ d n ) T ∇ d n → ( ∇ d ) T ∇ d strongly in L 1 ( 0 , T ; L 1 ( F ~ σ ( t ) ; R 3 × 3 ) ) L^{1}(0,T;L^{1}(\widetilde{\mathcal{F}}_{\sigma}(t);\mathbb{R}^{3\times3})) L 1 ( 0 , T ; L 1 ( F σ ( t ) ; R 3 × 3 )) on this region, and therefore
− ∫ 0 T ∫ Ω ( ζ ⋅ ∇ ) d n ⋅ ( M n + n φ n ( d n − b n ) ) d x d t → ∫ 0 T ∫ F ~ σ ( t ) ( ∇ d ) T ∇ d : ∇ ζ d x d t = ∫ 0 T ∫ F ( t ) ( ∇ d ) T ∇ d : ∇ ζ d x d t , -\int_{0}^{T}\int_{\Omega}(\zeta\cdot\nabla)d^{n}\cdot\left(\mathbf{M}^{n}+n\varphi^{n}(d^{n}-b^{n})\right)\,\mathrm{d}x\,\mathrm{d}t
\to\int_{0}^{T}\int_{\widetilde{\mathcal{F}}_{\sigma}(t)}(\nabla d)^{\mathsf{T}}\nabla d:\nabla\zeta\,\mathrm{d}x\,\mathrm{d}t
=\int_{0}^{T}\int_{\mathcal{F}(t)}(\nabla d)^{\mathsf{T}}\nabla d:\nabla\zeta\,\mathrm{d}x\,\mathrm{d}t, − ∫ 0 T ∫ Ω ( ζ ⋅ ∇ ) d n ⋅ ( M n + n φ n ( d n − b n ) ) d x d t → ∫ 0 T ∫ F σ ( t ) ( ∇ d ) T ∇ d : ∇ ζ d x d t = ∫ 0 T ∫ F ( t ) ( ∇ d ) T ∇ d : ∇ ζ d x d t ,
where we additionally used that D ( ζ ) = 0 \mathbb{D}(\zeta)=0 D ( ζ ) = 0 on S ( t ) σ ⊃ F ( t ) ∩ S ( t ) σ \mathcal{S}(t)_{\sigma}\supset\mathcal{F}(t)\cap\mathcal{S}(t)_{\sigma} S ( t ) σ ⊃ F ( t ) ∩ S ( t ) σ .
The remaining terms in (42) pass to the limit by Lemma 3.12 and (76) . Hence
∫ 0 T ∫ Ω [ − u ⋅ ∂ t ζ − ( u ⊗ u ) : ∇ ζ + 2 D ( u ) : D ( ζ ) ] d x d t + δ ∫ 0 T a u ( u , ζ ) d t = ∫ Ω u 0 ⋅ ζ ( 0 ) d x + ∫ 0 T ∫ F ( t ) ( ∇ d ) T ∇ d : ∇ ζ d x d t \int_{0}^{T}\int_{\Omega}\left[-u\cdot\partial_{t}\zeta-(u\otimes u):\nabla\zeta+2\mathbb{D}(u):\mathbb{D}(\zeta)\right]\,\mathrm{d}x\,\mathrm{d}t+\delta\int_{0}^{T}a_{u}(u,\zeta)\,\mathrm{d}t
=\int_{\Omega}u_{0}\cdot\zeta(0)\,\mathrm{d}x+\int_{0}^{T}\int_{\mathcal{F}(t)}(\nabla d)^{\mathsf{T}}\nabla d:\nabla\zeta\,\mathrm{d}x\,\mathrm{d}t ∫ 0 T ∫ Ω [ − u ⋅ ∂ t ζ − ( u ⊗ u ) : ∇ ζ + 2 D ( u ) : D ( ζ ) ] d x d t + δ ∫ 0 T a u ( u , ζ ) d t = ∫ Ω u 0 ⋅ ζ ( 0 ) d x + ∫ 0 T ∫ F ( t ) ( ∇ d ) T ∇ d : ∇ ζ d x d t
for every ζ ∈ C c 1 ( [ 0 , T ) ; V ( Ω ) ∩ H 3 ( Ω ; R 3 ) ) \zeta\in C^{1}_{c}([0,T);\mathcal{V}(\Omega)\cap H^{3}(\Omega;\mathbb{R}^{3})) ζ ∈ C c 1 ([ 0 , T ) ; V ( Ω ) ∩ H 3 ( Ω ; R 3 )) that is rigid in a collar of the limiting solid.
It remains to derive the energy inequality for the limit and to show that the limiting solid stays away from the container wall. To this end, we pass to the limit in the approximate energy inequality (47) . Besides the weak convergences established above, we use the global compactness
d m → d strongly in C ( [ 0 , T ] ; L 2 ( Ω ; R 3 ) ) . (90) d^m \to d \quad\text{strongly in } C([0,T];L^2(\Omega;\mathbb{R}^3)). \tag*{(90)} d m → d strongly in C ([ 0 , T ] ; L 2 ( Ω ; R 3 )) . (90)
Indeed, by the director equation,
∂ t d m = M m − ( u m ⋅ ∇ ) d m \partial_t d^m = M^m - (u^m \cdot\nabla)d^m ∂ t d m = M m − ( u m ⋅ ∇ ) d m
and the right-hand side is uniformly bounded in L 2 ( 0 , T ; L 2 ( Ω ; R 3 ) ) L^2(0,T;L^2(\Omega;\mathbb{R}^3)) L 2 ( 0 , T ; L 2 ( Ω ; R 3 )) . Together with the uniform bound in L ∞ ( 0 , T ; H 1 ( Ω ; R 3 ) ) L^\infty(0,T;H^1(\Omega;\mathbb{R}^3)) L ∞ ( 0 , T ; H 1 ( Ω ; R 3 )) , this yields the asserted compactness. By interpolation, d m → d d^m \to d d m → d strongly in C ( [ 0 , T ] ; L q ( Ω ; R 3 ) ) C([0,T];L^q(\Omega;\mathbb{R}^3)) C ([ 0 , T ] ; L q ( Ω ; R 3 )) for all q ∈ [ 2 , 6 ) q \in[2,6) q ∈ [ 2 , 6 ) . In particular,
F ε ( d m ) → F ε ( d ) strongly in C ( [ 0 , T ] ; L 1 ( Ω ) ) . F_\varepsilon(d^m) \to F_\varepsilon(d) \quad\text{strongly in } C([0,T];L^1(\Omega)). F ε ( d m ) → F ε ( d ) strongly in C ([ 0 , T ] ; L 1 ( Ω )) .
The strong velocity compactness from Lemma 3.12 and the strong C ( [ 0 , T ] ; L 2 ( Ω ; R 3 ) ) C([0,T];L^2(\Omega;\mathbb{R}^3)) C ([ 0 , T ] ; L 2 ( Ω ; R 3 )) -convergence of the directors from (90) allow one common subsequence such that u m ( t ) → u ( t ) u^m(t) \to u(t) u m ( t ) → u ( t ) and d m ( t ) → d ( t ) d^m(t) \to d(t) d m ( t ) → d ( t ) strongly in L 2 ( Ω ) L^2(\Omega) L 2 ( Ω ) for a.e. t t t . At any such time, the uniform H 1 H^1 H 1 -bound on d m ( t ) d^m(t) d m ( t ) identifies every weak H 1 H^1 H 1 -cluster point with d ( t ) d(t) d ( t ) , so the director gradient is lower semicontinuous. The kinetic energy converges strongly, and the potential converges by the preceding C t L x 1 C_tL_x^1 C t L x 1 -convergence. The viscous and artificial-viscosity dissipations and M m ⇀ M M^m \rightharpoonup M M m ⇀ M pass by weak lower semicontinuity. We obtain, initially for almost every t ∈ ( 0 , T ) t \in(0,T) t ∈ ( 0 , T ) ,
1 2 ∫ Ω ∣ u ( t ) ∣ 2 d x + ∫ Ω ( 1 2 ∣ ∇ d ( t ) ∣ 2 + F ε ( d ( t ) ) ) d x + 2 ∫ 0 t ∫ Ω ∣ D ( u ) ∣ 2 d x d s + δ ∫ 0 t ∥ u ( s ) ∥ X u 2 d s + ∫ 0 t ∫ F ( s ) ∣ M ∣ 2 d x d s ≤ 1 2 ∫ Ω ∣ u 0 ∣ 2 d x + ∫ Ω ( 1 2 ∣ ∇ d 0 ∣ 2 + F ε ( d 0 ) ) d x , \begin{aligned}
\frac{1}{2}\int_\Omega|u(t)|^2\,\mathrm{d}x+\int_\Omega\left(\frac{1}{2}|\nabla d(t)|^2+F_\varepsilon(d(t))\right)\,\mathrm{d}x+2\int_0^t\int_\Omega|\mathbb{D}(u)|^2\,\mathrm{d}x\,\mathrm{d}s+\delta\int_0^t \|u(s)\|_{X_u}^2\,\mathrm{d}s \\
&+\int_0^t\int_{\mathcal{F}(s)}|M|^2\,\mathrm{d}x\,\mathrm{d}s\leq\frac{1}{2}\int_\Omega|u_0|^2\,\mathrm{d}x+\int_\Omega\left(\frac{1}{2}|\nabla d_0|^2+F_\varepsilon(d_0)\right)\,\mathrm{d}x,
\end{aligned} 2 1 ∫ Ω ∣ u ( t ) ∣ 2 d x + ∫ Ω ( 2 1 ∣∇ d ( t ) ∣ 2 + F ε ( d ( t )) ) d x + 2 ∫ 0 t ∫ Ω ∣ D ( u ) ∣ 2 d x d s + δ ∫ 0 t ∥ u ( s ) ∥ X u 2 d s + ∫ 0 t ∫ F ( s ) ∣ M ∣ 2 d x d s ≤ 2 1 ∫ Ω ∣ u 0 ∣ 2 d x + ∫ Ω ( 2 1 ∣∇ d 0 ∣ 2 + F ε ( d 0 ) ) d x ,
where we used the identity M = Δ d + 1 ε 2 ( 1 − ∣ d ∣ 2 ) d M=\Delta d+\frac{1}{\varepsilon^2}(1-|d|^2)d M = Δ d + ε 2 1 ( 1 − ∣ d ∣ 2 ) d a.e. in Q F Q_{\mathcal{F}} Q F .
To remove the director energy contained in the solid, set
C S : = ∫ S 0 ( 1 2 ∣ ∇ d 0 ∣ 2 + F ε ( d 0 ) ) d x = 1 2 ∫ S 0 ∣ ∇ d 0 ∣ 2 d x . C_S:=\int_{S_0}\left(\frac{1}{2}|\nabla d_0|^2+F_\varepsilon(d_0)\right)\,\mathrm{d}x=\frac{1}{2}\int_{S_0}|\nabla d_0|^2\,\mathrm{d}x. C S := ∫ S 0 ( 2 1 ∣∇ d 0 ∣ 2 + F ε ( d 0 ) ) d x = 2 1 ∫ S 0 ∣∇ d 0 ∣ 2 d x .
Since ∣ d 0 ∣ = 1 |d_0|=1 ∣ d 0 ∣ = 1 a.e. in Ω \Omega Ω , one has F ε ( d 0 ) = 0 F_\varepsilon(d_0)=0 F ε ( d 0 ) = 0 . The initial director energy therefore splits as
∫ Ω ( 1 2 ∣ ∇ d 0 ∣ 2 + F ε ( d 0 ) ) d x = ∫ F 0 ( 1 2 ∣ ∇ d 0 ∣ 2 + F ε ( d 0 ) ) d x + C S . \int_\Omega\left(\frac{1}{2}|\nabla d_0|^2+F_\varepsilon(d_0)\right)\,\mathrm{d}x=\int_{\mathcal{F}_0}\left(\frac{1}{2}|\nabla d_0|^2+F_\varepsilon(d_0)\right)\,\mathrm{d}x+C_S. ∫ Ω ( 2 1 ∣∇ d 0 ∣ 2 + F ε ( d 0 ) ) d x = ∫ F 0 ( 2 1 ∣∇ d 0 ∣ 2 + F ε ( d 0 ) ) d x + C S .
Furthermore, d = b d=b d = b in S ( t ) S(t) S ( t ) , and the restriction of the limiting flow to the rigid body is a rigid transformation. Since b ( t , x ) = d 0 ( Y ( t , x ) ) b(t,x)=d_0(Y(t,x)) b ( t , x ) = d 0 ( Y ( t , x )) and ∣ d 0 ∣ = 1 |d_0|=1 ∣ d 0 ∣ = 1 , we obtain
∫ S ( t ) ( 1 2 ∣ ∇ d ( t ) ∣ 2 + F ε ( d ( t ) ) ) d x = C S . \int_{S(t)}\left(\frac{1}{2}|\nabla d(t)|^2+F_\varepsilon(d(t))\right)\,\mathrm{d}x=C_S. ∫ S ( t ) ( 2 1 ∣∇ d ( t ) ∣ 2 + F ε ( d ( t )) ) d x = C S .
Subtracting C S C_S C S from both sides gives
E δ , ε ( t ) + 2 ∫ 0 t ∫ Ω ∣ D ( u ) ∣ 2 d x d s + δ ∫ 0 t ∥ u ( s ) ∥ X u 2 d s + ∫ 0 t ∫ F ( s ) ∣ Δ d + 1 ε 2 ( 1 − ∣ d ∣ 2 ) d ∣ 2 d x d s ≤ E δ , ε ( 0 ) = E 0 , (91) \begin{aligned}
E^{\delta,\varepsilon}(t)+2\int_0^t\int_\Omega|\mathbb{D}(u)|^2\,\mathrm{d}x\,\mathrm{d}s+\delta\int_0^t \|u(s)\|_{X_u}^2\,\mathrm{d}s \\
&+\int_0^t\int_{\mathcal{F}(s)}\left|\Delta d+\frac{1}{\varepsilon^2}(1-|d|^2)d\right|^2\,\mathrm{d}x\,\mathrm{d}s\leq E^{\delta,\varepsilon}(0)=E_0, \tag*{(91)}
\end{aligned} E δ , ε ( t ) + 2 ∫ 0 t ∫ Ω ∣ D ( u ) ∣ 2 d x d s + δ ∫ 0 t ∥ u ( s ) ∥ X u 2 d s + ∫ 0 t ∫ F ( s ) Δ d + ε 2 1 ( 1 − ∣ d ∣ 2 ) d 2 d x d s ≤ E δ , ε ( 0 ) = E 0 , (91)
where
E δ , ε ( t ) : = 1 2 ∫ Ω ∣ u ( t ) ∣ 2 d x + ∫ F ( t ) ( 1 2 ∣ ∇ d ( t ) ∣ 2 + F ε ( d ( t ) ) ) d x , E^{\delta,\varepsilon}(t):=\frac{1}{2}\int_\Omega|u(t)|^2\,\mathrm{d}x+\int_{\mathcal{F}(t)}\left(\frac{1}{2}|\nabla d(t)|^2+F_\varepsilon(d(t))\right)\,\mathrm{d}x, E δ , ε ( t ) := 2 1 ∫ Ω ∣ u ( t ) ∣ 2 d x + ∫ F ( t ) ( 2 1 ∣∇ d ( t ) ∣ 2 + F ε ( d ( t )) ) d x ,
and, because ∣ d 0 ∣ = 1 |d_0|=1 ∣ d 0 ∣ = 1 , the initial value E 0 E_0 E 0 is independent of both δ \delta δ and ε \varepsilon ε .
Finally, (48) and the uniform convergence of the forward flow maps give
dist ( S ( φ ( t ) ) , ∂ Ω ) ≥ ρ δ ( T ) > 0 ( 0 ≤ t ≤ T ) . \operatorname{dist}(S(\varphi(t)),\partial\Omega) \ge\rho_\delta(T) > 0 \qquad(0 \le t \le T). dist ( S ( φ ( t )) , ∂ Ω ) ≥ ρ δ ( T ) > 0 ( 0 ≤ t ≤ T ) .
Since T > 0 T>0 T > 0 was arbitrary, the rigidity limit can be carried out on every bounded time interval. At fixed δ \delta δ , the body never reaches the container wall.
Vanishing artificial-viscosity limit δ \delta δ to zero
Fix ε > 0 \varepsilon>0 ε > 0 and assume E 0 > 0 E_0>0 E 0 > 0 . If E 0 = 0 E_0=0 E 0 = 0 , the time T ∗ T_\ast T ∗ below may be replaced by any finite time. For each δ > 0 \delta>0 δ > 0 , let ( u δ , d δ , M δ , φ δ ) (u_\delta,d_\delta,M_\delta,\varphi_\delta) ( u δ , d δ , M δ , φ δ ) be the post-rigidity solution on [ 0 , T ∗ ] [0,T_\ast] [ 0 , T ∗ ] provided by § 3.2 . We denote
S δ ( t ) = S ( φ δ ( t ) ) , F δ ( t ) = Ω ∖ S δ ( t ) ‾ . S_\delta(t)=S(\varphi_\delta(t)), \qquad\mathcal{F}_\delta(t)=\Omega\setminus\overline{S_\delta(t)}. S δ ( t ) = S ( φ δ ( t )) , F δ ( t ) = Ω ∖ S δ ( t ) .
In the fluid region,
M δ ≔ Δ d δ − f ε ( d δ ) = Δ d δ + 1 ε 2 ( 1 − ∣ d δ ∣ 2 ) d δ . M_\delta\coloneqq\Delta d_\delta-f_\varepsilon(d_\delta)=\Delta d_\delta+\frac{1}{\varepsilon^2}(1-\lvert d_\delta\rvert^2)d_\delta. M δ : = Δ d δ − f ε ( d δ ) = Δ d δ + ε 2 1 ( 1 − ∣ d δ ∣ 2 ) d δ .
The post-rigidity solution satisfies
u δ ∈ L ∞ ( 0 , T ∗ ; H ( Ω ) ) ∩ L 2 ( 0 , T ∗ ; K ( φ δ ) ∩ H 3 ( Ω ; R 3 ) ) , u_\delta\in L^\infty(0,T_\ast;\mathcal{H}(\Omega))\cap L^2(0,T_\ast;\mathcal{K}(\varphi_\delta)\cap H^3(\Omega;\mathbb{R}^3)), u δ ∈ L ∞ ( 0 , T ∗ ; H ( Ω )) ∩ L 2 ( 0 , T ∗ ; K ( φ δ ) ∩ H 3 ( Ω ; R 3 )) ,
d δ ∣ F δ ( t ) ∈ L ∞ ( 0 , T ∗ ; H 1 ( F δ ( t ) ; R 3 ) ) , M δ ∈ L 2 ( Q F δ ; R 3 ) , d_\delta|_{\mathcal{F}_\delta(t)}\in L^\infty(0,T_\ast;H^1(\mathcal{F}_\delta(t);\mathbb{R}^3)), \qquad M_\delta\in L^2(Q_{\mathcal{F}_\delta};\mathbb{R}^3), d δ ∣ F δ ( t ) ∈ L ∞ ( 0 , T ∗ ; H 1 ( F δ ( t ) ; R 3 )) , M δ ∈ L 2 ( Q F δ ; R 3 ) ,
φ δ ∈ Char ( [ 0 , T ∗ ] × Ω ) ∩ C 0 , 1 / p ( [ 0 , T ∗ ] ; L p ( Ω ) ) ( 1 ≤ p < ∞ ) . \varphi_\delta\in\operatorname{Char}([0,T_\ast]\times\Omega)\cap C^{0,1/p}([0,T_\ast];L^p(\Omega)) \qquad(1\le p<\infty). φ δ ∈ Char ([ 0 , T ∗ ] × Ω ) ∩ C 0 , 1/ p ([ 0 , T ∗ ] ; L p ( Ω )) ( 1 ≤ p < ∞ ) .
Its momentum equation reads
∫ 0 T ∗ ∫ Ω [ − u δ ⋅ ∂ t ζ − ( u δ ⊗ u δ ) : ∇ ζ + 2 D ( u δ ) : D ( ζ ) ] d x d t + δ ∫ 0 T ∗ a u ( u δ , ζ ) d t = ∫ Ω u 0 ⋅ ζ ( 0 ) d x + ∫ 0 T ∗ ∫ F δ ( t ) ( ∇ d δ ) T ∇ d δ : ∇ ζ d x d t . (92) \int_0^{T_\ast}\int_\Omega\left[-u_\delta\cdot\partial_t\zeta-(u_\delta\otimes u_\delta):\nabla\zeta+2D(u_\delta):D(\zeta)\right]\mathrm{d}x\,\mathrm{d}t
+\delta\int_0^{T_\ast}a_u(u_\delta,\zeta)\,\mathrm{d}t
=\int_\Omega u_0\cdot\zeta(0)\,\mathrm{d}x+\int_0^{T_\ast}\int_{\mathcal{F}_\delta(t)}(\nabla d_\delta)^{\mathsf{T}}\nabla d_\delta:\nabla\zeta\,\mathrm{d}x\,\mathrm{d}t. \tag*{(92)} ∫ 0 T ∗ ∫ Ω [ − u δ ⋅ ∂ t ζ − ( u δ ⊗ u δ ) : ∇ ζ + 2 D ( u δ ) : D ( ζ ) ] d x d t + δ ∫ 0 T ∗ a u ( u δ , ζ ) d t = ∫ Ω u 0 ⋅ ζ ( 0 ) d x + ∫ 0 T ∗ ∫ F δ ( t ) ( ∇ d δ ) T ∇ d δ : ∇ ζ d x d t . (92)
for every
ζ ∈ H 1 ( 0 , T ∗ ; H 1 ( Ω ; R 3 ) ) ∩ L 2 ( 0 , T ∗ ; V ( Ω ) ∩ H 3 ( Ω ; R 3 ) ) , ζ ( T ∗ ) = 0 , \zeta\in H^1(0,T_\ast;H^1(\Omega;\mathbb{R}^3))\cap L^2(0,T_\ast;\mathcal{V}(\Omega)\cap H^3(\Omega;\mathbb{R}^3)), \qquad\zeta(T_\ast)=0, ζ ∈ H 1 ( 0 , T ∗ ; H 1 ( Ω ; R 3 )) ∩ L 2 ( 0 , T ∗ ; V ( Ω ) ∩ H 3 ( Ω ; R 3 )) , ζ ( T ∗ ) = 0 ,
for which there exists r > 0 r>0 r > 0 such that D ( ζ ) = 0 D(\zeta)=0 D ( ζ ) = 0 on ( S δ ( t ) ) r (S_\delta(t))_r ( S δ ( t ) ) r for a.e. t ∈ ( 0 , T ∗ ) t\in(0,T_\ast) t ∈ ( 0 , T ∗ ) . Its director equation reads
∫ 0 T ∗ ∫ F δ ( t ) [ − d δ ⋅ ∂ t ψ + ( u δ ⋅ ∇ ) d δ ⋅ ψ + ∇ d δ : ∇ ψ + f ε ( d δ ) ⋅ ψ ] d x d t = ∫ F 0 d 0 ⋅ ψ ( 0 ) d x , \int_0^{T_\ast}\int_{\mathcal{F}_\delta(t)}\left[-d_\delta\cdot\partial_t\psi+(u_\delta\cdot\nabla)d_\delta\cdot\psi+\nabla d_\delta:\nabla\psi+f_\varepsilon(d_\delta)\cdot\psi\right]\mathrm{d}x\,\mathrm{d}t
=\int_{\mathcal{F}_0}d_0\cdot\psi(0)\,\mathrm{d}x, ∫ 0 T ∗ ∫ F δ ( t ) [ − d δ ⋅ ∂ t ψ + ( u δ ⋅ ∇ ) d δ ⋅ ψ + ∇ d δ : ∇ ψ + f ε ( d δ ) ⋅ ψ ] d x d t = ∫ F 0 d 0 ⋅ ψ ( 0 ) d x ,
for every ψ ∈ X d ( φ δ ) \psi\in\mathcal{X}_d(\varphi_\delta) ψ ∈ X d ( φ δ ) , see (14) , and its transport equation reads
∫ 0 T ∗ ∫ Ω φ δ [ ∂ t η + u δ ⋅ ∇ η ] d x d t = − ∫ Ω φ 0 η ( 0 ) d x \int_0^{T_\ast}\int_\Omega\varphi_\delta[\partial_t\eta+u_\delta\cdot\nabla\eta]\,\mathrm{d}x\,\mathrm{d}t
=-\int_\Omega\varphi_0\eta(0)\,\mathrm{d}x ∫ 0 T ∗ ∫ Ω φ δ [ ∂ t η + u δ ⋅ ∇ η ] d x d t = − ∫ Ω φ 0 η ( 0 ) d x
for every η ∈ C 1 ( [ 0 , T ∗ ] × Ω ‾ ) \eta\in C^1([0,T_\ast]\times\overline{\Omega}) η ∈ C 1 ([ 0 , T ∗ ] × Ω ) with η ( T ∗ ) = 0 \eta(T_\ast)=0 η ( T ∗ ) = 0 . The energy inequality (91) gives, with C C C independent of δ \delta δ ,
∥ u δ ∥ L t ∞ L x 2 + ∥ u δ ∥ L t 2 H 0 , x 1 + ∥ ∇ d δ ∥ L t ∞ L 2 ( F δ ( t ) ) + ∥ M δ ∥ L 2 ( Q F δ ) + δ ∥ u δ ∥ L 2 ( 0 , T ∗ ; X u ) ≤ C . \lVert u_\delta\rVert_{L^\infty_tL^2_x}+\lVert u_\delta\rVert_{L^2_tH^1_{0,x}}+\lVert\nabla d_\delta\rVert_{L^\infty_tL^2(\mathcal{F}_\delta(t))}+\lVert M_\delta\rVert_{L^2(Q_{\mathcal{F}_\delta})}+\sqrt{\delta}\,\lVert u_\delta\rVert_{L^2(0,T_\ast;\mathcal{X}_u)}\le C. ∥ u δ ∥ L t ∞ L x 2 + ∥ u δ ∥ L t 2 H 0 , x 1 + ∥ ∇ d δ ∥ L t ∞ L 2 ( F δ ( t )) + ∥ M δ ∥ L 2 ( Q F δ ) + δ ∥ u δ ∥ L 2 ( 0 , T ∗ ; X u ) ≤ C .
The maximum principle of Lemma A.5 , applied at the approximation level, also gives
∣ d δ ∣ ≤ 1 , ( d δ ) 3 ≥ 0 a.e. in Q F δ . (93) \lvert d_\delta\rvert\le1,\qquad(d_\delta)_3\ge0\quad\text{a.e. in }Q_{\mathcal{F}_\delta}. \tag*{(93)} ∣ d δ ∣ ≤ 1 , ( d δ ) 3 ≥ 0 a.e. in Q F δ . (93)
We first show that the solids stay uniformly away from the container wall and that their rigid motions are compact. Since u δ u_\delta u δ is rigid in S δ ( t ) S_\delta(t) S δ ( t ) , there are h δ h_\delta h δ , ℓ δ = h δ ′ \ell_\delta=h_\delta' ℓ δ = h δ ′ , ω δ \omega_\delta ω δ , and O δ ∈ SO ( 3 ) O_\delta\in\operatorname{SO}(3) O δ ∈ SO ( 3 ) such that
S δ ( t ) = h δ ( t ) + O δ ( t ) S 0 , u δ = ℓ δ + ω δ × ( x − h δ ) in S δ ( t ) . S_\delta(t)=h_\delta(t)+O_\delta(t)S_0,\qquad u_\delta=\ell_\delta+\omega_\delta\times(x-h_\delta)\quad\text{in }S_\delta(t). S δ ( t ) = h δ ( t ) + O δ ( t ) S 0 , u δ = ℓ δ + ω δ × ( x − h δ ) in S δ ( t ) .
Combining the identity
∫ S δ ( t ) ∣ u δ ∣ 2 d x = m ∣ ℓ δ ∣ 2 + J δ ( t ) ω δ ⋅ ω δ \int_{S_\delta(t)}\lvert u_\delta\rvert^2\,\mathrm{d}x=m\lvert\ell_\delta\rvert^2+J_\delta(t)\omega_\delta\cdot\omega_\delta ∫ S δ ( t ) ∣ u δ ∣ 2 d x = m ∣ ℓ δ ∣ 2 + J δ ( t ) ω δ ⋅ ω δ
and the energy inequality (91) we have
∥ ℓ δ ∥ L ∞ ( 0 , T ∗ ) + ∥ ω δ ∥ L ∞ ( 0 , T ∗ ) ≤ C . \lVert\ell_\delta\rVert_{L^\infty(0,T_\ast)}+\lVert\omega_\delta\rVert_{L^\infty(0,T_\ast)}\le C. ∥ ℓ δ ∥ L ∞ ( 0 , T ∗ ) + ∥ ω δ ∥ L ∞ ( 0 , T ∗ ) ≤ C .
Moreover, (3.54) contains the kinetic energy and the viscous dissipation with initial energy E 0 E_0 E 0 , so that Corollary 2.9 yields, for every δ > 0 \delta>0 δ > 0 ,
dist ( S δ ( t ) , ∂ Ω ) ≥ δ 0 2 ( 0 ≤ t ≤ T ∗ ) . (94) \operatorname{dist}(\mathcal{S}_{\delta}(t),\partial\Omega)\geq\frac{\delta_0}{2}\qquad(0\leq t\leq T_*). \tag*{(94)} dist ( S δ ( t ) , ∂ Ω ) ≥ 2 δ 0 ( 0 ≤ t ≤ T ∗ ) . (94)
After extraction, h δ → h h_{\delta}\to h h δ → h and O δ → O O_{\delta}\to O O δ → O uniformly on [ 0 , T ∗ ] [0,T_*] [ 0 , T ∗ ] . With S ( t ) = h ( t ) + O ( t ) S 0 \mathcal{S}(t)=h(t)+O(t)\mathcal{S}_0 S ( t ) = h ( t ) + O ( t ) S 0 and φ = 1 S ( t ) \varphi=\mathbf{1}_{\mathcal{S}(t)} φ = 1 S ( t ) , this implies
sup t ∈ [ 0 , T ∗ ] d H ( S δ ( t ) , S ( t ) ) → 0 , φ δ → φ strongly in C ( [ 0 , T ∗ ] ; L p ( Ω ) ) , 1 ≤ p < ∞ , (95) \sup_{t\in[0,T_*]}d_{\mathrm H}(\mathcal{S}_{\delta}(t),\mathcal{S}(t))\to0,\qquad\varphi_{\delta}\to\varphi\ \text{strongly in }C([0,T_*];L^p(\Omega)),\qquad1\leq p<\infty, \tag*{(95)} t ∈ [ 0 , T ∗ ] sup d H ( S δ ( t ) , S ( t )) → 0 , φ δ → φ strongly in C ([ 0 , T ∗ ] ; L p ( Ω )) , 1 ≤ p < ∞ , (95)
and, for every σ > 0 \sigma>0 σ > 0 and all small δ \delta δ ,
S δ ( t ) ⊂ S ( t ) σ , S ( t ) ⊂ S δ ( t ) σ ( 0 ≤ t ≤ T ∗ ) . (96) \mathcal{S}_{\delta}(t)\subset\mathcal{S}(t)_{\sigma},\qquad\mathcal{S}(t)\subset\mathcal{S}_{\delta}(t)_{\sigma}\qquad(0\leq t\leq T_*). \tag*{(96)} S δ ( t ) ⊂ S ( t ) σ , S ( t ) ⊂ S δ ( t ) σ ( 0 ≤ t ≤ T ∗ ) . (96)
After a further extraction, as δ → 0 \delta\to0 δ → 0 , we get
ℓ δ ∗ ⇀ ℓ , ω δ ∗ ⇀ ω in L ∞ ( 0 , T ∗ ) , \ell_{\delta}^{*}\rightharpoonup\ell,\qquad\omega_{\delta}^{*}\rightharpoonup\omega\quad\text{in }L^{\infty}(0,T_*), ℓ δ ∗ ⇀ ℓ , ω δ ∗ ⇀ ω in L ∞ ( 0 , T ∗ ) ,
whereas h δ → h h_{\delta}\to h h δ → h and O δ → O O_{\delta}\to O O δ → O uniformly as recorded above. Passing to the integral rigid-motion equations gives
h ′ = ℓ , O ′ = m O , m x = ω × x . h'=\ell,\qquad O'=mO,\qquad mx=\omega\times x. h ′ = ℓ , O ′ = m O , m x = ω × x .
In addition,
φ δ u δ = φ δ [ ℓ δ + ω δ × ( x − h δ ) ] . \varphi_{\delta}u_{\delta}=\varphi_{\delta}\left[\ell_{\delta}+\omega_{\delta}\times(x-h_{\delta})\right]. φ δ u δ = φ δ [ ℓ δ + ω δ × ( x − h δ ) ] .
The strong convergence of φ δ \varphi_{\delta} φ δ and h δ h_{\delta} h δ , together with the weak velocity convergence and the weak-* convergence of the rigid coefficients, passes this identity in distributions. Consequently
u ( t , x ) = ℓ ( t ) + ω ( t ) × ( x − h ( t ) ) for a.e. x ∈ S ( t ) . (97) u(t,x)=\ell(t)+\omega(t)\times(x-h(t))\qquad\text{for a.e. }x\in\mathcal{S}(t). \tag*{(97)} u ( t , x ) = ℓ ( t ) + ω ( t ) × ( x − h ( t )) for a.e. x ∈ S ( t ) . (97)
Thus the limiting rigid trajectory is generated by the rigid restriction of the limiting velocity, including its orientation.
Next, we extend the director through the solid in order to obtain compactness in the whole container, and then strengthen the convergence locally in the fluid region. To this end, define the whole-container material extension
d ‾ δ ( t , x ) : = { d δ ( t , x ) , x ∈ F δ ( t ) , O δ ( t ) T ( x − h δ ( t ) ) , x ∈ S δ ( t ) . \overline{d}_{\delta}(t,x):=
\begin{cases}
d_{\delta}(t,x), & x\in\mathcal{F}_{\delta}(t),\\
O_{\delta}(t)^{\mathsf T}(x-h_{\delta}(t)), & x\in\mathcal{S}_{\delta}(t).
\end{cases} d δ ( t , x ) := { d δ ( t , x ) , O δ ( t ) T ( x − h δ ( t )) , x ∈ F δ ( t ) , x ∈ S δ ( t ) .
The two traces agree on ∂ S δ ( t ) \partial\mathcal{S}_{\delta}(t) ∂ S δ ( t ) , so d ‾ δ ( t ) ∈ H 1 ( Ω ; R 3 ) \overline{d}_{\delta}(t)\in H^1(\Omega;\mathbb{R}^3) d δ ( t ) ∈ H 1 ( Ω ; R 3 ) for a.e. t t t . The solid contribution to its Dirichlet energy is invariant under the rigid motion. Hence
sup δ > 0 ∥ d ‾ δ ∥ L ∞ ( 0 , T ∗ ; H 1 ( Ω ) ) ≤ C . (98) \sup_{\delta>0}\|\overline{d}_{\delta}\|_{L^{\infty}(0,T_*;H^1(\Omega))}\leq C. \tag*{(98)} δ > 0 sup ∥ d δ ∥ L ∞ ( 0 , T ∗ ; H 1 ( Ω )) ≤ C . (98)
Extend M δ M_{\delta} M δ by zero to the solid and denote the extension by M ‾ δ \overline{M}_{\delta} M δ . Applying Lemma A.4 to the fluid equation and the rigidly transported solid branch gives
∂ t d ‾ δ + ( u δ ⋅ ∇ ) d ‾ δ = M ‾ δ in D ′ ( ( 0 , T ∗ ) × Ω ) . (99) \partial_t\overline{d}_{\delta}+(u_{\delta}\cdot\nabla)\overline{d}_{\delta}=\overline{M}_{\delta}\quad\text{in }\mathcal{D}'((0,T_*)\times\Omega). \tag*{(99)} ∂ t d δ + ( u δ ⋅ ∇ ) d δ = M δ in D ′ (( 0 , T ∗ ) × Ω ) . (99)
Because u δ ∈ L t 2 L x 6 u_{\delta}\in L^2_tL^6_x u δ ∈ L t 2 L x 6 and ∇ d ‾ δ ∈ L t ∞ L x 2 \nabla\overline{d}_{\delta}\in L^{\infty}_tL^2_x ∇ d δ ∈ L t ∞ L x 2 , the transport term is bounded in L t 2 L x 3 / 2 ↪ L t 2 H x − 1 L^2_tL^{3/2}_x\hookrightarrow L^2_tH^{-1}_x L t 2 L x 3/2 ↪ L t 2 H x − 1 . Thus, ∂ t d ‾ δ \partial_t\overline{d}_{\delta} ∂ t d δ is bounded in L 2 ( 0 , T ∗ ; H − 1 ( Ω ; R 3 ) ) L^2(0,T_*;H^{-1}(\Omega;\mathbb{R}^3)) L 2 ( 0 , T ∗ ; H − 1 ( Ω ; R 3 )) . By the Aubin–Lions lemma, after extracting a subsequence we have
d ‾ δ → d ‾ strongly in L 2 ( ( 0 , T ∗ ) × Ω ) (100) \overline{d}_{\delta}\to\overline{d}\quad\text{strongly in }L^2((0,T_*)\times\Omega) \tag*{(100)} d δ → d strongly in L 2 (( 0 , T ∗ ) × Ω ) (100)
and weak-* in L t ∞ H x 1 L^{\infty}_tH^1_x L t ∞ H x 1 . In view of (3.56) and the unit-length solid branch, the limit satisfies ∣ d ‾ ∣ ≤ 1 |\overline{d}|\leq1 ∣ d ∣ ≤ 1 and ( d ‾ ) 3 ≥ 0 (\overline{d})_3\geq0 ( d ) 3 ≥ 0 .
We strengthen the convergence on compact subsets of the fluid region. Let K ⋐ Q F K \Subset Q_F K ⋐ Q F . By (95) , a finite cylinder covering of K K K provides V j ⋐ U j ⋐ Ω V_j \Subset U_j \Subset\Omega V j ⋐ U j ⋐ Ω and time intervals I j I_j I j such that U j ⊂ F δ ( t ) ∩ F ( t ) U_j \subset F_\delta(t) \cap F(t) U j ⊂ F δ ( t ) ∩ F ( t ) for t ∈ I j t \in I_j t ∈ I j and all sufficiently small δ \delta δ . On I j × U j I_j \times U_j I j × U j ,
Δ d δ = M δ + f ε ( d δ ) . \Delta d_\delta= M_\delta+ f_\varepsilon(d_\delta). Δ d δ = M δ + f ε ( d δ ) .
For fixed ε \varepsilon ε , (93) makes f ε ( d δ ) f_\varepsilon(d_\delta) f ε ( d δ ) uniformly bounded. Interior elliptic regularity therefore gives a uniform L 2 ( I j ; H 2 ( V j ) ) L^2(I_j;H^2(V_j)) L 2 ( I j ; H 2 ( V j )) -bound. Equation (99) gives the corresponding L 2 ( I j ; H − 1 ( V j ) ) L^2(I_j;H^{-1}(V_j)) L 2 ( I j ; H − 1 ( V j )) -bound on the time derivative. Applying the Aubin–Lions lemma, we yield
d δ → d strongly in L 2 ( I j ; H 1 ( V j ) ) , (101) d_\delta\to d \quad\text{strongly in } L^2(I_j;H^1(V_j)), \tag*{(101)} d δ → d strongly in L 2 ( I j ; H 1 ( V j )) , (101)
where d = d ‾ ∣ Q F d=\overline{d}|_{Q_F} d = d ∣ Q F . Consequently, for every K ⋐ Q F K \Subset Q_F K ⋐ Q F ,
( ∇ d δ ) T ∇ d δ → ( ∇ d ) T ∇ d strongly in L 1 ( K ) . (102) (\nabla d_\delta)^{\mathsf T}\nabla d_\delta\to(\nabla d)^{\mathsf T}\nabla d \quad\text{strongly in } L^1(K). \tag*{(102)} ( ∇ d δ ) T ∇ d δ → ( ∇ d ) T ∇ d strongly in L 1 ( K ) . (102)
Moreover, since f ε f_\varepsilon f ε is Lipschitz on the closed unit ball, (100) gives
f ε ( d δ ‾ ) → f ε ( d ‾ ) strongly in L 2 ( ( 0 , T ∗ ) × Ω ) . (103) f_\varepsilon(\overline{d_\delta}) \to f_\varepsilon(\overline{d}) \quad\text{strongly in } L^2((0,T_*)\times\Omega). \tag*{(103)} f ε ( d δ ) → f ε ( d ) strongly in L 2 (( 0 , T ∗ ) × Ω ) . (103)
After extracting a subsequence,
M ~ δ ⇀ M ~ weakly in L 2 ( ( 0 , T ∗ ) × Ω ) . (104) \widetilde{M}_\delta\rightharpoonup\widetilde{M} \quad\text{weakly in } L^2((0,T_*)\times\Omega). \tag*{(104)} M δ ⇀ M weakly in L 2 (( 0 , T ∗ ) × Ω ) . (104)
Testing on compact subsets of Q F Q_F Q F and using (101) we get
M ~ = Δ d − f ε ( d ) in D ′ ( Q F ) . \widetilde{M}=\Delta d-f_\varepsilon(d) \quad\text{in } \mathcal{D}'(Q_F). M = Δ d − f ε ( d ) in D ′ ( Q F ) .
By construction of the weak extension, M ~ = 0 \widetilde{M}=0 M = 0 a.e. in the spacetime interior of the solid.
The anchoring condition can be passed using only local collars, so no global moving-domain diffeomorphism is needed. Choose r > 0 r>0 r > 0 smaller than one quarter of the uniform separation in (94) and smaller than the tubular radii of ∂ Ω \partial\Omega ∂ Ω and ∂ S 0 \partial S_0 ∂ S 0 . The fixed outer collar
U o u t : = { x ∈ Ω : dist ( x , ∂ Ω ) < r } U_{\mathrm{out}}:=\{x\in\Omega:\operatorname{dist}(x,\partial\Omega)<r\} U out := { x ∈ Ω : dist ( x , ∂ Ω ) < r }
is contained in F δ ( t ) ∩ F ( t ) F_\delta(t)\cap F(t) F δ ( t ) ∩ F ( t ) for every t t t and every small δ \delta δ . By (98) and (100) , the restrictions d δ ‾ ∣ U o u t \overline{d_\delta}|_{U_{\mathrm{out}}} d δ ∣ U out converge weakly in L t 2 H x 1 L^2_tH^1_x L t 2 H x 1 and strongly in L t , x 2 L^2_{t,x} L t , x 2 to d ‾ ∣ U o u t \overline{d}|_{U_{\mathrm{out}}} d ∣ U out . Their traces on ∂ Ω \partial\Omega ∂ Ω are all d 0 d_0 d 0 , so continuity of the trace operator gives Tr ∂ Ω d = d 0 \operatorname{Tr}_{\partial\Omega}d=d_0 Tr ∂ Ω d = d 0 for a.e. time.
For the inner boundary, let
U i n 0 : = { y ∈ F 0 : dist ( y , ∂ S 0 ) < r } . U_{\mathrm{in}}^0:=\{y\in F_0:\operatorname{dist}(y,\partial S_0)<r\}. U in 0 := { y ∈ F 0 : dist ( y , ∂ S 0 ) < r } .
and use only the rigid maps Z δ ( t , y ) = h δ ( t ) + O δ ( t ) y Z_\delta(t,y)=h_\delta(t)+O_\delta(t)y Z δ ( t , y ) = h δ ( t ) + O δ ( t ) y and Z ( t , y ) = h ( t ) + O ( t ) y Z(t,y)=h(t)+O(t)y Z ( t , y ) = h ( t ) + O ( t ) y . The pullbacks
d δ ‾ ( t , y ) : = d δ ( t , Z δ ( t , y ) ) , y ∈ U i n 0 , \overline{d_\delta}(t,y):=d_\delta(t,Z_\delta(t,y)),\qquad y\in U_{\mathrm{in}}^0, d δ ( t , y ) := d δ ( t , Z δ ( t , y )) , y ∈ U in 0 ,
are uniformly bounded in L t ∞ H 1 ( U i n 0 ) L^\infty_tH^1(U_{\mathrm{in}}^0) L t ∞ H 1 ( U in 0 ) because the maps are rigid isometries. The strong convergence (100) together with the uniform convergence of h δ h_\delta h δ and O δ O_\delta O δ implies d δ ‾ → d ‾ ∘ Z \overline{d_\delta}\to\overline{d}\circ Z d δ → d ∘ Z strongly in L 2 ( ( 0 , T ∗ ) × U i n 0 ) L^2((0,T_*)\times U_{\mathrm{in}}^0) L 2 (( 0 , T ∗ ) × U in 0 ) . For the varying composition, we first establish this for smooth approximation of d ‾ \overline{d} d and then by L 2 L^2 L 2 -approximation using preservation of volume. Thus the weak L t 2 H y 1 L^2_tH^1_y L t 2 H y 1 -limit is uniquely d ‾ ∘ Z \overline{d}\circ Z d ∘ Z . On ∂ S 0 \partial S_0 ∂ S 0 every pulled-back trace equals d 0 d_0 d 0 . The fixed-domain trace operator is continuous, so we have
Tr ∂ S ( t ) d ( t , x ) = d 0 ( O ( t ) T ( x − h ( t ) ) ) , for a.e. t . \operatorname{Tr}_{\partial S(t)}d(t,x)=d_0\bigl(O(t)^{\mathsf T}(x-h(t))\bigr),\qquad\text{for a.e. }t. Tr ∂ S ( t ) d ( t , x ) = d 0 ( O ( t ) T ( x − h ( t )) ) , for a.e. t .
The proof of the lemma below is deferred to Section A.4.
Lemma 3.14. Let S j ( t ) = h j ( t ) + O j ( t ) S 0 ⋐ Ω S_j(t)=h_j(t)+O_j(t)S_0\Subset\Omega S j ( t ) = h j ( t ) + O j ( t ) S 0 ⋐ Ω be rigid trajectories with h j , O j ∈ C ( [ 0 , T ] ) h_j,O_j\in C([0,T]) h j , O j ∈ C ([ 0 , T ]) satisfying
inf j , t dist ( S j ( t ) , ∂ Ω ) > 0 , h j → h , O j → O uniformly on [ 0 , T ] . \inf_{j,t}\operatorname{dist}(S_j(t),\partial\Omega)>0,\qquad h_j\to h,\quad O_j\to O\quad\text{uniformly on }[0,T]. j , t inf dist ( S j ( t ) , ∂ Ω ) > 0 , h j → h , O j → O uniformly on [ 0 , T ] . Let v j v_j v j satisfy
v j ∈ L ∞ ( 0 , T ; H ( Ω ) ) ∩ L 2 ( 0 , T ; V ( Ω ) ) , D ( v j ) = 0 a.e. on S j ( t ) , v_j\in L^\infty(0,T;\mathcal{H}(\Omega))\cap L^2(0,T;\mathcal{V}(\Omega)),\qquad\mathbb{D}(v_j)=0\quad\text{a.e. on }S_j(t), v j ∈ L ∞ ( 0 , T ; H ( Ω )) ∩ L 2 ( 0 , T ; V ( Ω )) , D ( v j ) = 0 a.e. on S j ( t ) ,
with a uniform bound in these spaces. Assume moreover that for every σ > 0 \sigma> 0 σ > 0 and every interval I ⋐ ( 0 , T ) I \Subset(0,T) I ⋐ ( 0 , T ) there are j σ j_\sigma j σ and C σ , I C_{\sigma,I} C σ , I such that, for all j ≥ j σ j \ge j_\sigma j ≥ j σ ,
∣ ∫ I ∫ Ω v j ⋅ ∂ t ζ d x d t ∣ ≤ C σ , I ( ∥ ζ ∥ L ∞ ( I ; H 3 ( Ω ) ) + ∥ ζ ∥ L 2 ( I ; H 3 ( Ω ) ) ) (105) \left|\int_I\int_\Omega\mathbf{v}_j \cdot\partial_t \zeta\,\mathrm{d}x\,\mathrm{d}t\right| \le C_{\sigma,I}\left(\|\zeta\|_{L^\infty(I;H^3(\Omega))}+\|\zeta\|_{L^2(I;H^3(\Omega))}\right) \tag*{(105)} ∫ I ∫ Ω v j ⋅ ∂ t ζ d x d t ≤ C σ , I ( ∥ ζ ∥ L ∞ ( I ; H 3 ( Ω )) + ∥ ζ ∥ L 2 ( I ; H 3 ( Ω )) ) (105)
for every ζ ∈ C ∞ ( I × Ω ; R 3 ) \zeta\in C^\infty(I \times\Omega;\mathbb{R}^3) ζ ∈ C ∞ ( I × Ω ; R 3 ) satisfying div ζ = 0 \operatorname{div}\zeta= 0 div ζ = 0 and D ( ζ ( t ) ) = 0 \mathbb{D}(\zeta(t)) = 0 D ( ζ ( t )) = 0 on S ( t ) σ S(t)_\sigma S ( t ) σ for every t ∈ I t \in I t ∈ I . Then, after extraction,
v j → v strongly in L 2 ( 0 , T ; L 2 ( Ω ; R 3 ) ) . \mathbf{v}_j \to\mathbf{v}\qquad\text{strongly in }L^2(0,T;L^2(\Omega;\mathbb{R}^3)). v j → v strongly in L 2 ( 0 , T ; L 2 ( Ω ; R 3 )) .
By the energy inequality (91) , we imply that as δ → 0 \delta\to0 δ → 0 ,
u δ ⇀ ∗ u in L ∞ ( 0 , T ∗ ; H ( Ω ) ) , u δ ⇀ u in L 2 ( 0 , T ∗ ; V ( Ω ) ) . (106) \mathbf{u}_\delta\overset{*}{\rightharpoonup} \mathbf{u}\ \text{in }L^\infty(0,T_*;\mathcal{H}(\Omega)),\qquad\mathbf{u}_\delta\rightharpoonup\mathbf{u}\ \text{in }L^2(0,T_*;\mathcal{V}(\Omega)). \tag*{(106)} u δ ⇀ ∗ u in L ∞ ( 0 , T ∗ ; H ( Ω )) , u δ ⇀ u in L 2 ( 0 , T ∗ ; V ( Ω )) . (106)
The exact rigidity and (95) imply φ D ( u ) = 0 \varphi_{\mathbb{D}}(\mathbf{u})=0 φ D ( u ) = 0 , so u ∈ L 2 ( 0 , T ∗ ; K ( φ ) ) \mathbf{u}\in L^2(0,T_*;\mathcal{K}(\varphi)) u ∈ L 2 ( 0 , T ∗ ; K ( φ )) . For a smooth test rigid on S ( t ) σ S(t)_\sigma S ( t ) σ , (96) makes it admissible in (92) for all small δ \delta δ . The convection and viscous terms have their standard energy bounds, and by (98) ,
∣ ∫ Ω ( ∇ d δ ) T ∇ d δ : ∇ ζ d x ∣ ≤ C ∥ ζ ∥ H 3 ( Ω ) . \left|\int_\Omega(\nabla\mathbf{d}_\delta)^{\mathsf{T}}\nabla\mathbf{d}_\delta:\nabla\zeta\,\mathrm{d}x\right| \le C\|\zeta\|_{H^3(\Omega)}. ∫ Ω ( ∇ d δ ) T ∇ d δ : ∇ ζ d x ≤ C ∥ ζ ∥ H 3 ( Ω ) .
Finally, we have
∣ δ ∫ 0 T ∗ a u ( u δ , ζ ) d t ∣ ≤ δ ∥ δ u δ ∥ L 2 ( 0 , T ∗ ; X u ) ∥ ζ ∥ L 2 ( 0 , T ∗ ; X u ) ≤ C δ ∥ ζ ∥ L 2 ( 0 , T ∗ ; X u ) . (107) \left|\delta\int_0^{T_*}a_{\mathbf{u}}(\mathbf{u}_\delta,\zeta)\,\mathrm{d}t\right|
\le\sqrt{\delta}\|\sqrt{\delta}\mathbf{u}_\delta\|_{L^2(0,T_*;\mathcal{X}_{\mathbf{u}})}\|\zeta\|_{L^2(0,T_*;\mathcal{X}_{\mathbf{u}})}
\le C\sqrt{\delta}\|\zeta\|_{L^2(0,T_*;\mathcal{X}_{\mathbf{u}})}. \tag*{(107)} δ ∫ 0 T ∗ a u ( u δ , ζ ) d t ≤ δ ∥ δ u δ ∥ L 2 ( 0 , T ∗ ; X u ) ∥ ζ ∥ L 2 ( 0 , T ∗ ; X u ) ≤ C δ ∥ ζ ∥ L 2 ( 0 , T ∗ ; X u ) . (107)
Thus (105) holds uniformly in δ \delta δ , and by Lemma 3.14 we have
u δ → u strongly in L 2 ( 0 , T ∗ ; L 2 ( Ω ; R 3 ) ) as δ → 0. (108) \mathbf{u}_\delta\to\mathbf{u}\qquad\text{strongly in }L^2(0,T_*;L^2(\Omega;\mathbb{R}^3))\text{ as }\delta\to0. \tag*{(108)} u δ → u strongly in L 2 ( 0 , T ∗ ; L 2 ( Ω ; R 3 )) as δ → 0. (108)
In particular, u δ ⊗ u δ → u ⊗ u \mathbf{u}_\delta\otimes\mathbf{u}_\delta\to\mathbf{u}\otimes\mathbf{u} u δ ⊗ u δ → u ⊗ u strongly in L 1 ( ( 0 , T ∗ ) × Ω ) L^1((0,T_*)\times\Omega) L 1 (( 0 , T ∗ ) × Ω ) as δ → 0 \delta\to0 δ → 0 .
We can now pass to the limit for the equations. From (95) and (108) , we obtain the transport identity
∫ 0 T ∗ ∫ Ω φ [ ∂ t η + u ⋅ ∇ η ] d x d t = − ∫ Ω φ 0 η ( 0 ) d x . \int_0^{T_*}\int_\Omega\varphi[\partial_t\eta+\mathbf{u}\cdot\nabla\eta]\,\mathrm{d}x\,\mathrm{d}t
=-\int_\Omega\varphi_0\eta(0)\,\mathrm{d}x. ∫ 0 T ∗ ∫ Ω φ [ ∂ t η + u ⋅ ∇ η ] d x d t = − ∫ Ω φ 0 η ( 0 ) d x .
For a director test function compactly supported in Q F ∪ ( { 0 } × F 0 ) Q_F\cup(\{0\}\times F_0) Q F ∪ ({ 0 } × F 0 ) , the support lies in F δ ( t ) F_\delta(t) F δ ( t ) for all small δ \delta δ . Combining (101) , (103) , and (108) we have
∫ 0 T ∗ ∫ F ( t ) [ − d ⋅ ∂ t ψ + ( u ⋅ ∇ ) d ⋅ ψ + ∇ d : ∇ ψ + f ε ( d ) ⋅ ψ ] d x d t = ∫ F 0 d 0 ⋅ ψ ( 0 ) d x . \int_0^{T_*}\int_{F(t)}\left[-\mathbf{d}\cdot\partial_t\psi+(\mathbf{u}\cdot\nabla)\mathbf{d}\cdot\psi+\nabla\mathbf{d}:\nabla\psi+f_\varepsilon(\mathbf{d})\cdot\psi\right]\,\mathrm{d}x\,\mathrm{d}t
=\int_{F_0}\mathbf{d}_0\cdot\psi(0)\,\mathrm{d}x. ∫ 0 T ∗ ∫ F ( t ) [ − d ⋅ ∂ t ψ + ( u ⋅ ∇ ) d ⋅ ψ + ∇ d : ∇ ψ + f ε ( d ) ⋅ ψ ] d x d t = ∫ F 0 d 0 ⋅ ψ ( 0 ) d x .
By Lemma 3.13, this identity can be extended to the natural moving energy test class for fixed ε \varepsilon ε .
For the momentum equation, let ζ \zeta ζ be a smooth test function that is rigid in S ( t ) σ S(t)_\sigma S ( t ) σ . The artificial-viscosity term vanishes by (107) . We can pass to the limit for all velocity terms by (106) and (108) . The director term needs only (102) . Indeed, since the Ericksen tensor is symmetric,
( ∇ d δ ) T ∇ d δ : ∇ ζ = ( ∇ d δ ) T ∇ d δ : D ( ζ ) = 0 on S ( t ) σ , (\nabla\mathbf{d}_\delta)^{\mathsf{T}}\nabla\mathbf{d}_\delta:\nabla\zeta
=(\nabla\mathbf{d}_\delta)^{\mathsf{T}}\nabla\mathbf{d}_\delta:\mathbb{D}(\zeta)=0
\quad\text{on }S(t)_\sigma, ( ∇ d δ ) T ∇ d δ : ∇ ζ = ( ∇ d δ ) T ∇ d δ : D ( ζ ) = 0 on S ( t ) σ ,
and the test function is compactly supported outside the collar. The initial and terminal time layers are controlled directly by the uniform director energy. Hence
∫ 0 T ∗ ∫ Ω [ − u ⋅ ∂ t ζ − ( u ⊗ u ) : ∇ ζ + 2 D ( u ) : D ( ζ ) ] d x d t = ∫ Ω u 0 ⋅ ζ ( 0 ) d x + ∫ 0 T ∗ ∫ F ( t ) ( ∇ d ) T ∇ d : ∇ ζ d x d t . \int_0^{T_*}\int_\Omega\left[-\mathbf{u}\cdot\partial_t\zeta-(\mathbf{u}\otimes\mathbf{u}):\nabla\zeta+2\mathbb{D}(\mathbf{u}):\mathbb{D}(\zeta)\right]\,\mathrm{d}x\,\mathrm{d}t
=\int_\Omega\mathbf{u}_0\cdot\zeta(0)\,\mathrm{d}x+\int_0^{T_*}\int_{F(t)}(\nabla\mathbf{d})^{\mathsf{T}}\nabla\mathbf{d}:\nabla\zeta\,\mathrm{d}x\,\mathrm{d}t. ∫ 0 T ∗ ∫ Ω [ − u ⋅ ∂ t ζ − ( u ⊗ u ) : ∇ ζ + 2 D ( u ) : D ( ζ ) ] d x d t = ∫ Ω u 0 ⋅ ζ ( 0 ) d x + ∫ 0 T ∗ ∫ F ( t ) ( ∇ d ) T ∇ d : ∇ ζ d x d t .
The global H 1 H^1 H 1 material extension makes the limiting momentum functional continuous under (A.1). Hence, by Proposition A.3 this identity can be extended to the standard test class T u \mathcal{T}_{\mathbf{u}} T u . In the subsequent singular limit we use the collar subclass only as an approximation device.
To pass to the limit for the energy inequality, we extract one common subsequence. By strong convergences (100) and (108) , we can take, up to a subsequence,
d ‾ δ ( t ) → d ‾ ( t ) and u δ ( t ) → u ( t ) strongly in L 2 ( Ω ) as δ → 0 , \overline{d}_{\delta}(t) \to\overline{d}(t) \qquad\text{and}\qquad u_{\delta}(t) \to u(t) \qquad\text{strongly in }L^{2}(\Omega)\text{ as }\delta\to0, d δ ( t ) → d ( t ) and u δ ( t ) → u ( t ) strongly in L 2 ( Ω ) as δ → 0 ,
for a.e. t ∈ ( 0 , T ∗ ) t \in(0,T_{\ast}) t ∈ ( 0 , T ∗ ) . Fix such a time that is also a Lebesgue time for all energy terms. The uniform H 1 H^{1} H 1 -bound (98) implies that every weak H 1 H^{1} H 1 -cluster point of d ‾ δ ( t ) \overline{d}_{\delta}(t) d δ ( t ) equals d ‾ ( t ) \overline{d}(t) d ( t ) , so
∥ ∇ d ‾ ( t ) ∥ L 2 ( Ω ) 2 ≤ lim inf δ → 0 ∥ ∇ d ‾ δ ( t ) ∥ L 2 ( Ω ) 2 . \|\nabla\overline{d}(t)\|_{L^{2}(\Omega)}^{2} \leq\liminf_{\delta\to0}\|\nabla\overline{d}_{\delta}(t)\|_{L^{2}(\Omega)}^{2}. ∥∇ d ( t ) ∥ L 2 ( Ω ) 2 ≤ δ → 0 lim inf ∥∇ d δ ( t ) ∥ L 2 ( Ω ) 2 .
For fixed ε \varepsilon ε , by the maximum principle we have ∣ d ‾ δ ∣ ≤ 1 |\overline{d}_{\delta}| \leq1 ∣ d δ ∣ ≤ 1 , and F ε F_{\varepsilon} F ε is Lipschitz on the closed unit ball. Hence, the strong convergence d ‾ δ ( t ) → d ‾ ( t ) \overline{d}_{\delta}(t) \to\overline{d}(t) d δ ( t ) → d ( t ) in L 2 ( Ω ) L^{2}(\Omega) L 2 ( Ω ) at the fixed time t t t implies the strong L 1 L^{1} L 1 -convergence of the Ginzburg–Landau potential. The kinetic energy converges along this common subsequence. For the fluid Dirichlet energy we subtract the rigid-motion invariant solid contribution
1 2 ∫ S δ ( t ) ∣ ∇ d ‾ δ ∣ 2 d x = 1 2 ∫ S 0 ∣ ∇ d 0 ∣ 2 d x = 1 2 ∫ S ( t ) ∣ ∇ d ‾ ∣ 2 d x . \frac{1}{2}\int_{S_{\delta}(t)}|\nabla\overline{d}_{\delta}|^{2}\,dx
=
\frac{1}{2}\int_{S_{0}}|\nabla d_{0}|^{2}\,dx
=
\frac{1}{2}\int_{S(t)}|\nabla\overline{d}|^{2}\,dx. 2 1 ∫ S δ ( t ) ∣∇ d δ ∣ 2 d x = 2 1 ∫ S 0 ∣∇ d 0 ∣ 2 d x = 2 1 ∫ S ( t ) ∣∇ d ∣ 2 d x .
By the weak lower semicontinuity of the viscous and molecular-field dissipations, and by (104) , for a.e. t ∈ ( 0 , T ∗ ) t \in(0,T_{\ast}) t ∈ ( 0 , T ∗ ) we have
E ε ( t ) + 2 ∫ 0 t ∫ Ω ∣ D ( u ) ∣ 2 d x d s + ∫ 0 t ∫ F ( s ) ∣ Δ d − f ε ( d ) ∣ 2 d x d s ≤ E ε ( 0 ) , E_{\varepsilon}(t)+2\int_{0}^{t}\int_{\Omega}|D(u)|^{2}\,dx\,ds+\int_{0}^{t}\int_{\mathcal{F}(s)}|\Delta d-f_{\varepsilon}(d)|^{2}\,dx\,ds\leq E_{\varepsilon}(0), E ε ( t ) + 2 ∫ 0 t ∫ Ω ∣ D ( u ) ∣ 2 d x d s + ∫ 0 t ∫ F ( s ) ∣Δ d − f ε ( d ) ∣ 2 d x d s ≤ E ε ( 0 ) ,
where
E ε ( t ) = 1 2 ∥ u ( t ) ∥ L 2 ( Ω ) 2 + ∫ F ( t ) ( 1 2 ∣ ∇ d ( t ) ∣ 2 + F ε ( d ( t ) ) ) d x . E_{\varepsilon}(t)=\frac{1}{2}\|u(t)\|_{L^{2}(\Omega)}^{2}+\int_{\mathcal{F}(t)}\left(\frac{1}{2}|\nabla d(t)|^{2}+F_{\varepsilon}(d(t))\right)\,dx. E ε ( t ) = 2 1 ∥ u ( t ) ∥ L 2 ( Ω ) 2 + ∫ F ( t ) ( 2 1 ∣∇ d ( t ) ∣ 2 + F ε ( d ( t )) ) d x .
We can pass to the limit in the separation estimate (94) . This completes the limit δ → 0 \delta\to0 δ → 0 and the proof of Theorem 3.2 .
Fixed-domain Ginzburg–Landau theory on arbitrary open sets
Since the post-impact fluid region need not be Lipschitz, we establish fixed-ε \varepsilon ε existence and compactness for the Ginzburg–Landau system on an arbitrary bounded open set. This construction is performed after the limiting contact configuration has been identified.
For a bounded open set G ⊂ R 3 G \subset\mathbb{R}^{3} G ⊂ R 3 , let D G \mathcal{D}_{G} D G , H G \mathcal{H}_{G} H G , and V G \mathcal{V}_{G} V G be defined in (10) . For v ∈ D G v \in\mathcal{D}_{G} v ∈ D G , integrating by parts we have
∥ ∇ v ∥ L 2 ( G ) 2 = 2 ∥ D ( v ) ∥ L 2 ( G ) 2 , \|\nabla v\|_{L^{2}(G)}^{2}=2\|D(v)\|_{L^{2}(G)}^{2}, ∥∇ v ∥ L 2 ( G ) 2 = 2∥ D ( v ) ∥ L 2 ( G ) 2 ,
and by the density argument, we can extend this identity to every v ∈ V G v \in\mathcal{V}_{G} v ∈ V G . Thus the viscous dissipation controls the full H 0 1 H_{0}^{1} H 0 1 -norm on V G \mathcal{V}_{G} V G without any assumption on ∂ G \partial G ∂ G .
Lemma 3.15. Let G ⊂ R 3 G \subset\mathbb{R}^{3} G ⊂ R 3 be bounded and open. Let ( v m ) m ≥ 1 ⊂ D G (v_m)_{m\geq1} \subset\mathcal{D}_{G} ( v m ) m ≥ 1 ⊂ D G be L 2 ( G ) L^{2}(G) L 2 ( G ) -orthonormal, with nested spans X N : = span { v 1 , … , v N } X_N := \operatorname{span}\{v_1,\ldots,v_N\} X N := span { v 1 , … , v N } whose union is dense in both H G \mathcal{H}_{G} H G and V G \mathcal{V}_{G} V G . Suppose that ( w j ) (w_j) ( w j ) is bounded in L ∞ ( 0 , T ; H G ) ∩ L 2 ( 0 , T ; V G ) L^{\infty}(0,T;\mathcal{H}_{G}) \cap L^{2}(0,T;\mathcal{V}_{G}) L ∞ ( 0 , T ; H G ) ∩ L 2 ( 0 , T ; V G ) and that, for every fixed m m m , the scalar functions a j , m ( t ) : = ( w j ( t ) , v m ) L 2 ( G ) a_{j,m}(t):=(w_j(t),v_m)_{L^{2}(G)} a j , m ( t ) := ( w j ( t ) , v m ) L 2 ( G ) have uniformly bounded variation on ( 0 , T ) (0,T) ( 0 , T ) for all sufficiently large j j j . Then ( w j ) (w_j) ( w j ) is relatively compact in L 2 ( ( 0 , T ) × G ) L^{2}((0,T)\times G) L 2 (( 0 , T ) × G ) .
Proof. Let P N P_N P N be the L 2 ( G ) L^{2}(G) L 2 ( G ) -orthogonal projection onto X N X_N X N . Since H 0 1 ( G ) ⋐ L 2 ( G ) H_{0}^{1}(G)\Subset L^{2}(G) H 0 1 ( G ) ⋐ L 2 ( G ) for every bounded open set by the Rellich–Kondrachov theorem with zero extension to a containing ball, the embedding V G ⊂ H G \mathcal{V}_{G}\subset\mathcal{H}_{G} V G ⊂ H G is compact. Hence
α N : = sup z ∈ V G ∥ z ∥ V G ≤ 1 ∥ ( I − P N ) z ∥ L 2 ( G ) → 0. (109) \alpha_N:=\sup_{\substack{z\in\mathcal{V}_{G}\\ \|z\|_{\mathcal{V}_{G}}\leq1}}\|(I-P_N)z\|_{L^{2}(G)}\to0. \tag*{(109)} α N := z ∈ V G ∥ z ∥ V G ≤ 1 sup ∥ ( I − P N ) z ∥ L 2 ( G ) → 0. (109)
Indeed, the image of the unit ball of V G \mathcal{V}_{G} V G is relatively compact in H G \mathcal{H}_{G} H G , and P N z → z P_Nz\to z P N z → z in H G \mathcal{H}_{G} H G for every z ∈ H G z\in\mathcal{H}_{G} z ∈ H G . The convergence is therefore uniform on that compact set.
For fixed N N N , each coefficient of P N w j P_N w_j P N w j is bounded in L ∞ ( 0 , T ) L^\infty(0,T) L ∞ ( 0 , T ) and in BV ( 0 , T ) \operatorname{BV}(0,T) BV ( 0 , T ) . Applying Helly’s compactness theorem, these coefficients are relatively compact in L 1 ( 0 , T ) L^1(0,T) L 1 ( 0 , T ) . The uniform L ∞ ( 0 , T ) L^\infty(0,T) L ∞ ( 0 , T ) -bound implies that they are relatively compact in L 2 ( 0 , T ) L^2(0,T) L 2 ( 0 , T ) . Thus, ( P N w j ) j (P_N w_j)_j ( P N w j ) j is relatively compact in L 2 ( 0 , T ; H G ) L^2(0,T;H_G) L 2 ( 0 , T ; H G ) . On the other hand, by (109) ,
∥ ( I − P N ) w j ∥ L 2 ( 0 , T ; H G ) ≤ α N ∥ w j ∥ L 2 ( 0 , T ; V G ) , \|(I-P_N)w_j\|_{L^2(0,T;H_G)}\leq\alpha_N\|w_j\|_{L^2(0,T;V_G)}, ∥ ( I − P N ) w j ∥ L 2 ( 0 , T ; H G ) ≤ α N ∥ w j ∥ L 2 ( 0 , T ; V G ) ,
uniformly in j j j . Letting N → ∞ N\to\infty N → ∞ , the claim is derived. □
Lemma 3.16. Let G ⊂ R 3 G\subset\mathbb{R}^3 G ⊂ R 3 be bounded and open, let u ∈ L 2 ( 0 , T ; H 0 1 ( G ; R 3 ) ) \mathbf{u}\in L^2(0,T;H_0^1(G;\mathbb{R}^3)) u ∈ L 2 ( 0 , T ; H 0 1 ( G ; R 3 )) be solenoidal, and let b ∈ H 1 ( G ; K + ) \mathbf{b}\in H^1(G;K_+) b ∈ H 1 ( G ; K + ) . Suppose
d ∈ L ∞ ( 0 , T ; H 1 ( G ; R 3 ) ) , d − b ∈ L 2 ( 0 , T ; H 0 1 ( G ; R 3 ) ) , ∂ t d ∈ L 2 ( 0 , T ; H − 1 ( G ; R 3 ) ) , \mathbf{d}\in L^\infty(0,T;H^1(G;\mathbb{R}^3)),\qquad\mathbf{d}-\mathbf{b}\in L^2(0,T;H_0^1(G;\mathbb{R}^3)),\qquad\partial_t\mathbf{d}\in L^2(0,T;H^{-1}(G;\mathbb{R}^3)), d ∈ L ∞ ( 0 , T ; H 1 ( G ; R 3 )) , d − b ∈ L 2 ( 0 , T ; H 0 1 ( G ; R 3 )) , ∂ t d ∈ L 2 ( 0 , T ; H − 1 ( G ; R 3 )) , d ( 0 ) ∈ K + \mathbf{d}(0)\in K_+ d ( 0 ) ∈ K + a.e., and
∂ t d + ( u ⋅ ∇ ) d = Δ d + 1 ε 2 ( 1 − ∣ d ∣ 2 ) d in L 2 ( 0 , T ; H − 1 ( G ) ) . (110) \partial_t\mathbf{d}+(\mathbf{u}\cdot\nabla)\mathbf{d}=\Delta\mathbf{d}+\frac{1}{\varepsilon^2}(1-|\mathbf{d}|^2)\mathbf{d}\quad\text{in }L^2(0,T;H^{-1}(G)). \tag*{(110)} ∂ t d + ( u ⋅ ∇ ) d = Δ d + ε 2 1 ( 1 − ∣ d ∣ 2 ) d in L 2 ( 0 , T ; H − 1 ( G )) . (110) Then ∣ d ∣ ≤ 1 |\mathbf{d}|\leq1 ∣ d ∣ ≤ 1 and d 3 ≥ 0 d_3\geq0 d 3 ≥ 0 a.e. on ( 0 , T ) × G (0,T)\times G ( 0 , T ) × G .
Proof. Let us write d = b + w \mathbf{d}=\mathbf{b}+\mathbf{w} d = b + w with w ∈ L 2 ( 0 , T ; H 0 1 ( G ; R 3 ) ) \mathbf{w}\in L^2(0,T;H_0^1(G;\mathbb{R}^3)) w ∈ L 2 ( 0 , T ; H 0 1 ( G ; R 3 )) . Since ∂ t w = ∂ t d ∈ L 2 ( 0 , T ; H − 1 ( G ; R 3 ) ) \partial_t\mathbf{w}=\partial_t\mathbf{d}\in L^2(0,T;H^{-1}(G;\mathbb{R}^3)) ∂ t w = ∂ t d ∈ L 2 ( 0 , T ; H − 1 ( G ; R 3 )) , we have d ∈ C ( [ 0 , T ] ; L 2 ( G ; R 3 ) ) \mathbf{d}\in C([0,T];L^2(G;\mathbb{R}^3)) d ∈ C ([ 0 , T ] ; L 2 ( G ; R 3 )) .
We first show that if Ψ : R 3 → R k \Psi:\mathbb{R}^3\to\mathbb{R}^k Ψ : R 3 → R k is Lipschitz with Ψ = 0 \Psi=0 Ψ = 0 on K + K_+ K + , then Ψ ( d ( t ) ) ∈ H 0 1 ( G ; R k ) \Psi(\mathbf{d}(t))\in H_0^1(G;\mathbb{R}^k) Ψ ( d ( t )) ∈ H 0 1 ( G ; R k ) for a.e. t t t . Indeed, fix t t t and choose w j ∈ C 0 ∞ ( G ; R 3 ) \mathbf{w}_j\in C_0^\infty(G;\mathbb{R}^3) w j ∈ C 0 ∞ ( G ; R 3 ) with w j → w ( t ) \mathbf{w}_j\to\mathbf{w}(t) w j → w ( t ) in H 1 ( G ) H^1(G) H 1 ( G ) . Since b \mathbf{b} b takes values in K + K_+ K + , the function Ψ ( b + w j ) \Psi(\mathbf{b}+\mathbf{w}_j) Ψ ( b + w j ) vanishes outside supp w j \operatorname{supp}\mathbf{w}_j supp w j . Hence, it has compact support in G G G and belongs to H 0 1 ( G ; R k ) H_0^1(G;\mathbb{R}^k) H 0 1 ( G ; R k ) . By the chain rule for Lipschitz maps, these functions Ψ ( b + w j ) \Psi(\mathbf{b}+\mathbf{w}_j) Ψ ( b + w j ) are bounded in H 1 ( G ) H^1(G) H 1 ( G ) , and they converge to Ψ ( d ( t ) ) \Psi(\mathbf{d}(t)) Ψ ( d ( t )) in L 2 ( G ) L^2(G) L 2 ( G ) . Since H 0 1 ( G ) H_0^1(G) H 0 1 ( G ) is weakly closed in H 1 ( G ) H^1(G) H 1 ( G ) , we conclude that Ψ ( d ( t ) ) ∈ H 0 1 ( G ; R k ) \Psi(\mathbf{d}(t))\in H_0^1(G;\mathbb{R}^k) Ψ ( d ( t )) ∈ H 0 1 ( G ; R k ) .
Next we prove that ∣ d ∣ ≤ 1 |\mathbf{d}|\leq1 ∣ d ∣ ≤ 1 . Let Π \Pi Π be the nearest-point projection onto the closed unit ball B ⊂ R 3 B\subset\mathbb{R}^3 B ⊂ R 3 , and define
F ( z ) : = z − Π ( z ) = ( 1 − 1 ∣ z ∣ ) + z , Φ ( z ) : = 1 2 ∣ F ( z ) ∣ 2 = 1 2 ( ∣ z ∣ − 1 ) + 2 . \mathbf{F}(\mathbf{z}):=\mathbf{z}-\Pi(\mathbf{z})=\left(1-\frac{1}{|\mathbf{z}|}\right)_+\mathbf{z},\qquad\Phi(\mathbf{z}):=\frac{1}{2}|\mathbf{F}(\mathbf{z})|^2=\frac{1}{2}(|\mathbf{z}|-1)_+^2. F ( z ) := z − Π ( z ) = ( 1 − ∣ z ∣ 1 ) + z , Φ ( z ) := 2 1 ∣ F ( z ) ∣ 2 = 2 1 ( ∣ z ∣ − 1 ) + 2 . The function Φ \Phi Φ is convex and of class C 1 , 1 C^{1,1} C 1 , 1 with ∇ Φ = F \nabla\Phi=\mathbf{F} ∇Φ = F , and F \mathbf{F} F is Lipschitz and vanishes on B ⊃ K + B\supset K_+ B ⊃ K + . Hence F ( d ) ∈ L 2 ( 0 , T ; H 0 1 ( G ; R 3 ) ) \mathbf{F}(\mathbf{d})\in L^2(0,T;H_0^1(G;\mathbb{R}^3)) F ( d ) ∈ L 2 ( 0 , T ; H 0 1 ( G ; R 3 )) by the above arguments.
Now test equation (110) with F ( d ) \mathbf{F}(\mathbf{d}) F ( d ) . Due to the convexity of Φ \Phi Φ , for all t , s ∈ ( 0 , T ) t,s\in(0,T) t , s ∈ ( 0 , T ) we have
F ( d ( s ) ) ⋅ ( d ( t ) − d ( s ) ) ≤ Φ ( d ( t ) ) − Φ ( d ( s ) ) ≤ F ( d ( t ) ) ⋅ ( d ( t ) − d ( s ) ) . \mathbf{F}(\mathbf{d}(s))\cdot(\mathbf{d}(t)-\mathbf{d}(s))\leq\Phi(\mathbf{d}(t))-\Phi(\mathbf{d}(s))\leq\mathbf{F}(\mathbf{d}(t))\cdot(\mathbf{d}(t)-\mathbf{d}(s)). F ( d ( s )) ⋅ ( d ( t ) − d ( s )) ≤ Φ ( d ( t )) − Φ ( d ( s )) ≤ F ( d ( t )) ⋅ ( d ( t ) − d ( s )) . Integrating these inequalities over G G G and using the Steklov averages in time, we imply that the integral of Φ ( d ( t ) ) \Phi(\mathbf{d}(t)) Φ ( d ( t )) over G G G is absolutely continuous as a function of t t t and
d d t ∫ G Φ ( d ) d x = ⟨ ∂ t d , F ( d ) ⟩ . (111) \frac{\mathrm{d}}{\mathrm{d}t}\int_G\Phi(\mathbf{d})\,\mathrm{d}x=\langle\partial_t\mathbf{d},\mathbf{F}(\mathbf{d})\rangle. \tag*{(111)} d t d ∫ G Φ ( d ) d x = ⟨ ∂ t d , F ( d )⟩ . (111) By definition of F \mathbf{F} F , we have on { ∣ d ∣ > 1 } \{|\mathbf{d}|>1\} { ∣ d ∣ > 1 } , ∂ i F ( d ) = ( 1 − 1 ∣ d ∣ ) ∂ i d + d ⋅ ∂ i d ∣ d ∣ 3 d \partial_i\mathbf{F}(\mathbf{d})=\left(1-\frac{1}{|\mathbf{d}|}\right)\partial_i\mathbf{d}+\frac{\mathbf{d}\cdot\partial_i\mathbf{d}}{|\mathbf{d}|^3}\mathbf{d} ∂ i F ( d ) = ( 1 − ∣ d ∣ 1 ) ∂ i d + ∣ d ∣ 3 d ⋅ ∂ i d d , and ∇ F ( d ) = 0 \nabla\mathbf{F}(\mathbf{d})=0 ∇ F ( d ) = 0 a.e. on { ∣ d ∣ ≤ 1 } \{|\mathbf{d}|\leq1\} { ∣ d ∣ ≤ 1 } . Consequently,
∇ d : ∇ F ( d ) = ( 1 − 1 ∣ d ∣ ) + ∣ ∇ d ∣ 2 + 1 { ∣ d ∣ > 1 } 1 ∣ d ∣ 3 ∑ i = 1 3 ( d ⋅ ∂ i d ) 2 ≥ 0. (112) \nabla\mathbf{d}:\nabla\mathbf{F}(\mathbf{d})=\left(1-\frac{1}{|\mathbf{d}|}\right)_+|\nabla\mathbf{d}|^2+\mathbf{1}_{\{|\mathbf{d}|>1\}}\frac{1}{|\mathbf{d}|^3}\sum_{i=1}^{3}(\mathbf{d}\cdot\partial_i\mathbf{d})^2\geq0. \tag*{(112)} ∇ d : ∇ F ( d ) = ( 1 − ∣ d ∣ 1 ) + ∣∇ d ∣ 2 + 1 { ∣ d ∣ > 1 } ∣ d ∣ 3 1 i = 1 ∑ 3 ( d ⋅ ∂ i d ) 2 ≥ 0. (112) Similarly, we have F ( d ) ⋅ ∂ i d = F ( d ) ⋅ ∂ i F ( d ) = ∂ i Φ ( d ) \mathbf{F}(\mathbf{d})\cdot\partial_i\mathbf{d}=\mathbf{F}(\mathbf{d})\cdot\partial_i\mathbf{F}(\mathbf{d})=\partial_i\Phi(\mathbf{d}) F ( d ) ⋅ ∂ i d = F ( d ) ⋅ ∂ i F ( d ) = ∂ i Φ ( d ) , hence, ( u ⋅ ∇ ) d ⋅ F ( d ) = u ⋅ ∇ Φ ( d ) (\mathbf{u}\cdot\nabla)\mathbf{d}\cdot\mathbf{F}(\mathbf{d})=\mathbf{u}\cdot\nabla\Phi(\mathbf{d}) ( u ⋅ ∇ ) d ⋅ F ( d ) = u ⋅ ∇Φ ( d ) . Choose φ j ∈ C 0 ∞ ( G ; R 3 ) \boldsymbol{\varphi}_j\in C_0^\infty(G;\mathbb{R}^3) φ j ∈ C 0 ∞ ( G ; R 3 ) with φ j → F ( d ( t ) ) \boldsymbol{\varphi}_j\to\mathbf{F}(\mathbf{d}(t)) φ j → F ( d ( t )) in H 1 ( G ) H^1(G) H 1 ( G ) . Since div u = 0 \operatorname{div}\mathbf{u}=0 div u = 0 , we have ∫ G u ⋅ ∇ 1 2 ∣ φ j ∣ 2 d x = 0 \int_G\mathbf{u}\cdot\nabla\frac{1}{2}|\boldsymbol{\varphi}_j|^2\,\mathrm{d}x=0 ∫ G u ⋅ ∇ 2 1 ∣ φ j ∣ 2 d x = 0 , and since φ j ⋅ ∇ φ j → F ( d ) ⋅ ∇ F ( d ) \boldsymbol{\varphi}_j\cdot\nabla\boldsymbol{\varphi}_j\to\mathbf{F}(\mathbf{d})\cdot\nabla\mathbf{F}(\mathbf{d}) φ j ⋅ ∇ φ j → F ( d ) ⋅ ∇ F ( d ) in L 3 / 2 ( G ) L^{3/2}(G) L 3/2 ( G ) and u ( t ) ∈ L 6 ( G ) \mathbf{u}(t)\in L^6(G) u ( t ) ∈ L 6 ( G ) , we have
∫ G ( u ⋅ ∇ ) d ⋅ F ( d ) d x = 0 for a.e. t . (113) \int_G(\mathbf{u}\cdot\nabla)\mathbf{d}\cdot\mathbf{F}(\mathbf{d})\,\mathrm{d}x=0\quad\text{for a.e. }t. \tag*{(113)} ∫ G ( u ⋅ ∇ ) d ⋅ F ( d ) d x = 0 for a.e. t . (113)
Note that d ⋅ F ( d ) = ∣ d ∣ ( ∣ d ∣ − 1 ) + d \cdot F(d)=|d|(|d|-1)_{+} d ⋅ F ( d ) = ∣ d ∣ ( ∣ d ∣ − 1 ) + , so ( 1 − ∣ d ∣ 2 ) d ⋅ F ( d ) ≤ 0 (1-|d|^{2})d\cdot F(d)\le0 ( 1 − ∣ d ∣ 2 ) d ⋅ F ( d ) ≤ 0 . Combining this inequality with (112) , (113) , we obtain
d d t ∫ G Φ ( d ) d x = − ∫ G ( u ⋅ ∇ ) d ⋅ F ( d ) d x + ∫ G ∇ d : ∇ F ( d ) d x + 1 ε 2 ∫ G ( 1 − ∣ d ∣ 2 ) d ⋅ F ( d ) d x ≤ 0. \frac{\mathrm{d}}{\mathrm{d}t}\int_G\Phi(d)\,\mathrm{d}x=-\int_G(u\cdot\nabla)d\cdot F(d)\,\mathrm{d}x+\int_G\nabla d:\nabla F(d)\,\mathrm{d}x+\frac{1}{\varepsilon^2}\int_G(1-|d|^2)d\cdot F(d)\,\mathrm{d}x\le0. d t d ∫ G Φ ( d ) d x = − ∫ G ( u ⋅ ∇ ) d ⋅ F ( d ) d x + ∫ G ∇ d : ∇ F ( d ) d x + ε 2 1 ∫ G ( 1 − ∣ d ∣ 2 ) d ⋅ F ( d ) d x ≤ 0.
As Φ ≥ 0 \Phi\ge0 Φ ≥ 0 and Φ ( d ( 0 ) ) = 0 \Phi(d(0))=0 Φ ( d ( 0 )) = 0 , it implies Φ ( d ( t ) ) = 0 \Phi(d(t))=0 Φ ( d ( t )) = 0 , that is, ∣ d ∣ ≤ 1 |d|\le1 ∣ d ∣ ≤ 1 a.e. in ( 0 , T ) × G (0,T)\times G ( 0 , T ) × G .
Lastly, we show that d 3 ≥ 0 d_3\ge0 d 3 ≥ 0 . Let us denote q : = d 3 q:=d_3 q := d 3 . By our assumption (110) , it satisfies
∂ t q + u ⋅ ∇ q − Δ q = 1 ε 2 ( 1 − ∣ d ∣ 2 ) q in L 2 ( 0 , T ; H − 1 ( G ) ) , (114) \partial_tq+u\cdot\nabla q-\Delta q=\frac{1}{\varepsilon^2}(1-|d|^2)q\quad\text{in }L^2(0,T;H^{-1}(G)), \tag*{(114)} ∂ t q + u ⋅ ∇ q − Δ q = ε 2 1 ( 1 − ∣ d ∣ 2 ) q in L 2 ( 0 , T ; H − 1 ( G )) , (114)
and 0 ≤ 1 ε 2 ( 1 − ∣ d ∣ 2 ) ≤ ε − 2 0\le\frac{1}{\varepsilon^2}(1-|d|^2)\le\varepsilon^{-2} 0 ≤ ε 2 1 ( 1 − ∣ d ∣ 2 ) ≤ ε − 2 by the above verification that ∣ d ∣ ≤ 1 |d|\le1 ∣ d ∣ ≤ 1 . The negative part q − = max { − q , 0 } q_-=\max\{-q,0\} q − = max { − q , 0 } equals Ψ ( d ) \Psi(d) Ψ ( d ) for the Lipschitz function Ψ ( z ) : = ( z 3 ) − \Psi(z):=(z_3)_- Ψ ( z ) := ( z 3 ) − , which vanishes on K + K_+ K + . Therefore, we have q − ∈ L 2 ( 0 , T ; H 0 1 ( G ) ) q_-\in L^2(0,T;H^1_0(G)) q − ∈ L 2 ( 0 , T ; H 0 1 ( G )) . Moreover, we have the identities
⟨ ∂ t q , − q − ⟩ = 1 2 d d t ∥ q − ∥ L 2 ( G ) 2 , ∇ q ⋅ ∇ ( − q − ) = ∣ ∇ q − ∣ 2 , q ( − q − ) = q − 2 , \langle\partial_tq,-q_-\rangle=\frac{1}{2}\frac{\mathrm{d}}{\mathrm{d}t}\|q_-\|_{L^2(G)}^2,\qquad\nabla q\cdot\nabla(-q_-)=|\nabla q_-|^2,\qquad q(-q_-)=q_-^2, ⟨ ∂ t q , − q − ⟩ = 2 1 d t d ∥ q − ∥ L 2 ( G ) 2 , ∇ q ⋅ ∇ ( − q − ) = ∣∇ q − ∣ 2 , q ( − q − ) = q − 2 ,
where the first identity can be derived as in (111) . Testing (114) with − q − -q_- − q − , using these identities, and noting that the transport term vanishes as in the above computation of ⟨ ∂ t d , F ( d ) ⟩ \langle\partial_td,F(d)\rangle ⟨ ∂ t d , F ( d )⟩ , we get
1 2 d d t ∥ q − ∥ L 2 ( G ) 2 + ∥ ∇ q − ∥ L 2 ( G ) 2 = ∫ G 1 ε 2 ( 1 − ∣ d ∣ 2 ) q − 2 d x ≤ 1 ε 2 ∥ q − ∥ L 2 ( G ) 2 . \frac{1}{2}\frac{\mathrm{d}}{\mathrm{d}t}\|q_-\|_{L^2(G)}^2+\|\nabla q_-\|_{L^2(G)}^2=\int_G\frac{1}{\varepsilon^2}(1-|d|^2)q_-^2\,\mathrm{d}x\le\frac{1}{\varepsilon^2}\|q_-\|_{L^2(G)}^2. 2 1 d t d ∥ q − ∥ L 2 ( G ) 2 + ∥∇ q − ∥ L 2 ( G ) 2 = ∫ G ε 2 1 ( 1 − ∣ d ∣ 2 ) q − 2 d x ≤ ε 2 1 ∥ q − ∥ L 2 ( G ) 2 .
Since q − ( 0 ) = 0 q_-(0)=0 q − ( 0 ) = 0 , by Grönwall’s inequality we have q − = 0 q_-=0 q − = 0 meaning that d 3 ≥ 0 d_3\ge0 d 3 ≥ 0 a.e. in ( 0 , T ) × G (0,T)\times G ( 0 , T ) × G . □
We first solve the Ginzburg–Landau relaxation of (11) on an arbitrary bounded open set G G G ,
{ ∂ t u + ( u ⋅ ∇ ) u − Δ u + ∇ π = − div ( ( ∇ d ) T ∇ d ) , div u = 0 , in ( 0 , T ) × G , ∂ t d + ( u ⋅ ∇ ) d − Δ d = − f ε ( d ) , in ( 0 , T ) × G , (115) \left\{
\begin{aligned}
\partial_tu+(u\cdot\nabla)u-\Delta u+\nabla\pi&=-\operatorname{div}\left((\nabla d)^{\mathsf T}\nabla d\right), &&\operatorname{div}u=0,\quad\text{in }(0,T)\times G,\\
\partial_td+(u\cdot\nabla)d-\Delta d&=-f_\varepsilon(d), &&\text{in }(0,T)\times G,
\end{aligned}
\right. \tag*{(115)} { ∂ t u + ( u ⋅ ∇ ) u − Δ u + ∇ π ∂ t d + ( u ⋅ ∇ ) d − Δ d = − div ( ( ∇ d ) T ∇ d ) , = − f ε ( d ) , div u = 0 , in ( 0 , T ) × G , in ( 0 , T ) × G , (115)
with u = 0 u=0 u = 0 and d = d ∗ d=d_* d = d ∗ on ∂ G \partial G ∂ G in the variational sense, and with initial data ( u ∗ , d ∗ ) (u_*,d_*) ( u ∗ , d ∗ ) .
Proposition 3.17. Let G ⊂ R 3 G\subset\mathbb{R}^3 G ⊂ R 3 be a bounded open set, let u ∗ ∈ H G u_*\in H_G u ∗ ∈ H G , and let d ∗ ∈ H 1 ( G ; K + ) d_*\in H^1(G;K_+) d ∗ ∈ H 1 ( G ; K + ) . Let ε > 0 \varepsilon>0 ε > 0 be fixed. Then there is a global trajectory ( u , d , M ) (u,d,M) ( u , d , M ) on [ 0 , ∞ ) [0,\infty) [ 0 , ∞ ) such that, for every T > 0 T>0 T > 0 ,
u ∈ L ∞ ( 0 , T ; H G ) ∩ L 2 ( 0 , T ; V G ) , u\in L^\infty(0,T;H_G)\cap L^2(0,T;V_G), u ∈ L ∞ ( 0 , T ; H G ) ∩ L 2 ( 0 , T ; V G ) , d ∈ L ∞ ( 0 , T ; H 1 ( G ; K + ) ) , d − d ∗ ∈ L ∞ ( 0 , T ; H 0 1 ( G ; R 3 ) ) , d\in L^\infty(0,T;H^1(G;K_+)),\qquad d-d_*\in L^\infty(0,T;H^1_0(G;\mathbb{R}^3)), d ∈ L ∞ ( 0 , T ; H 1 ( G ; K + )) , d − d ∗ ∈ L ∞ ( 0 , T ; H 0 1 ( G ; R 3 )) , M : = Δ d − f ε ( d ) ∈ L 2 ( ( 0 , T ) × G ; R 3 ) , M:=\Delta d-f_\varepsilon(d)\in L^2((0,T)\times G;\mathbb{R}^3), M := Δ d − f ε ( d ) ∈ L 2 (( 0 , T ) × G ; R 3 ) , where the molecular-field identity is understood variationally, that is, for a.e. t t t
∫ G M ( t ) ⋅ ξ d x = − ∫ G ∇ d ( t ) : ∇ ξ d x − ∫ G f ε ( d ( t ) ) ⋅ ξ d x ∀ ξ ∈ H 0 1 ( G ; R 3 ) . \int_G M(t)\cdot\xi\,\mathrm{d}x=-\int_G\nabla d(t):\nabla\xi\,\mathrm{d}x-\int_Gf_\varepsilon(d(t))\cdot\xi\,\mathrm{d}x\qquad\forall\xi\in H^1_0(G;\mathbb{R}^3). ∫ G M ( t ) ⋅ ξ d x = − ∫ G ∇ d ( t ) : ∇ ξ d x − ∫ G f ε ( d ( t )) ⋅ ξ d x ∀ ξ ∈ H 0 1 ( G ; R 3 ) . The pair ( u , d ) (u,d) ( u , d ) solves (115) in the following weak form: for every ζ ∈ C 1 ( [ 0 , T ) ; D G ) \zeta\in C^1([0,T);D_G) ζ ∈ C 1 ([ 0 , T ) ; D G ) and every ψ ∈ C 1 ( [ 0 , T ) ; C c ∞ ( G ; R 3 ) ) \psi\in C^1([0,T);C^\infty_c(G;\mathbb{R}^3)) ψ ∈ C 1 ([ 0 , T ) ; C c ∞ ( G ; R 3 )) ,
∫ 0 T ∫ G [ − u ⋅ ∂ t ζ − ( u ⊗ u ) : ∇ ζ + 2 D ( u ) : D ( ζ ) ] d x d t = ∫ G u ∗ ⋅ ζ ( 0 ) d x + ∫ 0 T ∫ G ( ∇ d ) T ∇ d : ∇ ζ d x d t , \int_0^T\int_G\left[-u\cdot\partial_t\zeta-(u\otimes u):\nabla\zeta+2D(u):D(\zeta)\right]\mathrm{d}x\,\mathrm{d}t
=\int_Gu_*\cdot\zeta(0)\,\mathrm{d}x+\int_0^T\int_G(\nabla d)^{\mathsf T}\nabla d:\nabla\zeta\,\mathrm{d}x\,\mathrm{d}t, ∫ 0 T ∫ G [ − u ⋅ ∂ t ζ − ( u ⊗ u ) : ∇ ζ + 2 D ( u ) : D ( ζ ) ] d x d t = ∫ G u ∗ ⋅ ζ ( 0 ) d x + ∫ 0 T ∫ G ( ∇ d ) T ∇ d : ∇ ζ d x d t , ∫ 0 T ∫ G [ − d ⋅ ∂ t ψ + ( u ⋅ ∇ ) d ⋅ ψ + ∇ d : ∇ ψ + f ε ( d ) ⋅ ψ ] d x d t = ∫ G d ∗ ⋅ ψ ( 0 ) d x . \int_0^T\int_G\left[-d\cdot\partial_t\psi+(u\cdot\nabla)d\cdot\psi+\nabla d:\nabla\psi+f_\varepsilon(d)\cdot\psi\right]\mathrm{d}x\,\mathrm{d}t=\int_Gd_*\cdot\psi(0)\,\mathrm{d}x. ∫ 0 T ∫ G [ − d ⋅ ∂ t ψ + ( u ⋅ ∇ ) d ⋅ ψ + ∇ d : ∇ ψ + f ε ( d ) ⋅ ψ ] d x d t = ∫ G d ∗ ⋅ ψ ( 0 ) d x .
For a.e. t ∈ ( 0 , T ) t \in(0,T) t ∈ ( 0 , T ) we have
1 2 ∥ u ( t ) ∥ L 2 ( G ) 2 + 1 2 ∥ ∇ d ( t ) ∥ L 2 ( G ) 2 + ∫ G F ε ( d ( t ) ) d x + 2 ∫ 0 t ∫ G ∣ D ( u ) ∣ 2 d x d s + ∫ 0 t ∫ G ∣ M ∣ 2 d x d s ≤ 1 2 ∥ u ∗ ∥ L 2 ( G ) 2 + 1 2 ∥ ∇ d ∗ ∥ L 2 ( G ) 2 + ∫ G F ε ( d ∗ ) d x . (116) \frac{1}{2}\lVert u(t)\rVert_{L^2(G)}^2+\frac{1}{2}\lVert\nabla d(t)\rVert_{L^2(G)}^2+\int_G F_\varepsilon(d(t))\,\mathrm{d}x+2\int_0^t\int_G |\mathsf{D}(u)|^2\,\mathrm{d}x\,\mathrm{d}s
+\int_0^t\int_G |\mathsf{M}|^2\,\mathrm{d}x\,\mathrm{d}s\leq\frac{1}{2}\lVert u_*\rVert_{L^2(G)}^2+\frac{1}{2}\lVert\nabla d_*\rVert_{L^2(G)}^2+\int_G F_\varepsilon(d_*)\,\mathrm{d}x. \tag*{(116)} 2 1 ∥ u ( t ) ∥ L 2 ( G ) 2 + 2 1 ∥ ∇ d ( t ) ∥ L 2 ( G ) 2 + ∫ G F ε ( d ( t )) d x + 2 ∫ 0 t ∫ G ∣ D ( u ) ∣ 2 d x d s + ∫ 0 t ∫ G ∣ M ∣ 2 d x d s ≤ 2 1 ∥ u ∗ ∥ L 2 ( G ) 2 + 2 1 ∥ ∇ d ∗ ∥ L 2 ( G ) 2 + ∫ G F ε ( d ∗ ) d x . (116)
Proof. Note that G G G may fail to be Lipschitz, we will utilize compactness properties that do not require boundary regularity. Let b ∗ b_* b ∗ be the componentwise harmonic representative of the affine class d ∗ + H 0 1 ( G ) d_*+H^1_0(G) d ∗ + H 0 1 ( G ) , i.e.,
b ∗ − d ∗ ∈ H 0 1 ( G ) , ∫ G ∇ b ∗ : ∇ ξ d x = 0 ∀ ξ ∈ H 0 1 ( G ; R 3 ) . b_*-d_*\in H^1_0(G),\qquad\int_G\nabla b_*:\nabla\xi\,\mathrm{d}x=0\quad\forall\xi\in H^1_0(G;\mathbb{R}^3). b ∗ − d ∗ ∈ H 0 1 ( G ) , ∫ G ∇ b ∗ : ∇ ξ d x = 0 ∀ ξ ∈ H 0 1 ( G ; R 3 ) .
We claim that b ∗ b_* b ∗ takes values in K + K_+ K + . By definition (6) , we can write
K + = { z : z 3 ≥ 0 } ∩ { z : e ⋅ z ≤ 1 } . K_+=\{z:z_3\geq0\}\cap\{z:e\cdot z\leq1\}. K + = { z : z 3 ≥ 0 } ∩ { z : e ⋅ z ≤ 1 } .
Let ℓ ( z ) = e ⋅ z − 1 \ell(z)=e\cdot z-1 ℓ ( z ) = e ⋅ z − 1 or ℓ ( z ) = − z 3 \ell(z)=-z_3 ℓ ( z ) = − z 3 , so that ℓ ( d ∗ ) ≤ 0 \ell(d_*)\leq0 ℓ ( d ∗ ) ≤ 0 . We claim that the scalar function h : = ℓ ( b ∗ ) h:=\ell(b_*) h := ℓ ( b ∗ ) is weakly harmonic, and h + = max { h , 0 } ∈ H 0 1 ( G ) h_+=\max\{h,0\}\in H^1_0(G) h + = max { h , 0 } ∈ H 0 1 ( G ) . Indeed, choose q m ∈ C c ∞ ( G ; R 3 ) q_m\in C^\infty_c(G;\mathbb{R}^3) q m ∈ C c ∞ ( G ; R 3 ) with q m → b ∗ − d ∗ q_m\to b_*-d_* q m → b ∗ − d ∗ in H 0 1 ( G ) H^1_0(G) H 0 1 ( G ) . Then ℓ ( d ∗ + q m ) + \ell(d_*+q_m)_+ ℓ ( d ∗ + q m ) + has compact support in G G G because ℓ ( d ∗ ) ≤ 0 \ell(d_*)\leq0 ℓ ( d ∗ ) ≤ 0 , is bounded in H 1 ( G ) H^1(G) H 1 ( G ) , and converges to h + h_+ h + in L 2 ( G ) L^2(G) L 2 ( G ) . By the weak closedness of H 0 1 ( G ) H^1_0(G) H 0 1 ( G ) , h + ∈ H 0 1 ( G ) h_+\in H^1_0(G) h + ∈ H 0 1 ( G ) . The harmonicity of h h h also implies that ∥ ∇ h + ∥ L 2 ( G ) 2 = 0 \lVert\nabla h_+\rVert_{L^2(G)}^2=0 ∥ ∇ h + ∥ L 2 ( G ) 2 = 0 , hence h + = 0 h_+=0 h + = 0 by Poincaré’s inequality. Applying this argument to − z 3 -z_3 − z 3 and to a countable dense set of e ∈ S 2 e\in S^2 e ∈ S 2 , we imply that b ∗ ∈ K + b_*\in K_+ b ∗ ∈ K + almost everywhere.
Choose a sequence ( v j ) j ≥ 1 ⊂ D G (v_j)_{j\geq1}\subset D_G ( v j ) j ≥ 1 ⊂ D G whose span is dense both in H G \mathcal{H}_G H G with respect to L 2 ( G ) L^2(G) L 2 ( G ) and in D G ‾ H 0 1 ( G ) \overline{D_G}^{H^1_0(G)} D G H 0 1 ( G ) with respect to H 0 1 ( G ) H^1_0(G) H 0 1 ( G ) , and orthonormalize it in L 2 ( G ) L^2(G) L 2 ( G ) . For fixed N N N , we consider velocities of the form u N ( t ) = ∑ j = 1 N a j ( t ) v j u^N(t)=\sum_{j=1}^N a_j(t)v_j u N ( t ) = ∑ j = 1 N a j ( t ) v j . If a ∈ C ( [ 0 , T ] ; R N ) a\in C([0,T];\mathbb{R}^N) a ∈ C ([ 0 , T ] ; R N ) is prescribed, the vector field u N u^N u N is smooth in space, divergence free, and has zero trace. We solve the equation
∂ t d + ( u N ⋅ ∇ ) d = Δ d − f ε ( d ) , d ( 0 ) = d ∗ , d − b ∗ ∈ L 2 ( 0 , T ; H 0 1 ( G ) ) , \partial_t d+(u^N\cdot\nabla)d=\Delta d-f_\varepsilon(d),\qquad d(0)=d_*,\qquad d-b_*\in L^2(0,T;H^1_0(G)), ∂ t d + ( u N ⋅ ∇ ) d = Δ d − f ε ( d ) , d ( 0 ) = d ∗ , d − b ∗ ∈ L 2 ( 0 , T ; H 0 1 ( G )) ,
by the eigenfunction Galerkin scheme for the Dirichlet Laplacian associated with H 0 1 ( G ) H^1_0(G) H 0 1 ( G ) . The energy estimate and Banach–Alaoglu give weak and weak-* compactness at the energy level. After the auxiliary Galerkin index is removed, Lemma 3.16 applies with boundary representative b ∗ b_* b ∗ and shows that the resulting parabolic solution satisfies d ∈ K + d\in K_+ d ∈ K + almost everywhere. To show the dependence on the given velocity, let d 1 , d 2 d_1,d_2 d 1 , d 2 correspond to coefficient paths a 1 , a 2 a_1,a_2 a 1 , a 2 in a bounded subset of C ( [ 0 , T ] ; R N ) C([0,T];\mathbb{R}^N) C ([ 0 , T ] ; R N ) and set r = d 1 − d 2 r=d_1-d_2 r = d 1 − d 2 . Testing the difference equation by r r r , using div u [ a 1 ] = 0 \operatorname{div}u[a_1]=0 div u [ a 1 ] = 0 , the Lipschitz continuity of f ε f_\varepsilon f ε on K + K_+ K + , and the finite-dimensional bound ∥ u [ a 1 ] − u [ a 2 ] ∥ L ∞ ( G ) ≤ C N ∣ a 1 − a 2 ∣ \lVert u[a_1]-u[a_2]\rVert_{L^\infty(G)}\leq C_N|a_1-a_2| ∥ u [ a 1 ] − u [ a 2 ] ∥ L ∞ ( G ) ≤ C N ∣ a 1 − a 2 ∣ , we have
1 2 d d t ∥ r ∥ L 2 ( G ) 2 + ∥ ∇ r ∥ L 2 ( G ) 2 ≤ C N , ε , R ∥ r ∥ L 2 ( G ) 2 + C N , R ∣ a 1 − a 2 ∣ 2 ∥ ∇ d 2 ∥ L 2 ( G ) 2 . \frac{1}{2}\frac{\mathrm{d}}{\mathrm{d}t}\lVert r\rVert_{L^2(G)}^2+\lVert\nabla r\rVert_{L^2(G)}^2\leq C_{N,\varepsilon,R}\lVert r\rVert_{L^2(G)}^2+C_{N,R}|a_1-a_2|^2\lVert\nabla d_2\rVert_{L^2(G)}^2. 2 1 d t d ∥ r ∥ L 2 ( G ) 2 + ∥ ∇ r ∥ L 2 ( G ) 2 ≤ C N , ε , R ∥ r ∥ L 2 ( G ) 2 + C N , R ∣ a 1 − a 2 ∣ 2 ∥ ∇ d 2 ∥ L 2 ( G ) 2 .
Applying Grönwall’s lemma and using the uniform director-energy bound we imply that d [ a ] d[a] d [ a ] depends continuously on a a a , as a map from bounded subsets of C ( [ 0 , T ] ; R N ) C([0,T];\mathbb{R}^N) C ([ 0 , T ] ; R N ) into C ( [ 0 , T ] ; L 2 ( G ) ) ∩ L 2 ( 0 , T ; H 1 ( G ) ) C([0,T];L^2(G))\cap L^2(0,T;H^1(G)) C ([ 0 , T ] ; L 2 ( G )) ∩ L 2 ( 0 , T ; H 1 ( G )) .
The finite-dimensional coupling can be closed without any regularity of ∂ G \partial G ∂ G . Given a coefficient path a a a , let d [ a ] d[a] d [ a ] be the director constructed as above and let T N ( a ) = b \mathcal{T}_N(a)=b T N ( a ) = b where b b b solves the following equation
b i ′ ( t ) + 2 ∑ j = 1 N ( D ( v j ) , D ( v i ) ) L 2 ( G ) b j ′ ( t ) = ∫ G [ u [ a ] ⊗ u [ a ] ] : ∇ v i d x + ∫ G ( ∇ d [ a ] ) T ∇ d [ a ] : ∇ v i d x , b_i'(t)+2\sum_{j=1}^N(\mathsf{D}(v_j),\mathsf{D}(v_i))_{L^2(G)}b_j'(t)=\int_G[u[a]\otimes u[a]]:\nabla v_i\,\mathrm{d}x+\int_G(\nabla d[a])^{\mathsf T}\nabla d[a]:\nabla v_i\,\mathrm{d}x, b i ′ ( t ) + 2 j = 1 ∑ N ( D ( v j ) , D ( v i ) ) L 2 ( G ) b j ′ ( t ) = ∫ G [ u [ a ] ⊗ u [ a ]] : ∇ v i d x + ∫ G ( ∇ d [ a ] ) T ∇ d [ a ] : ∇ v i d x ,
with initial datum to be the L 2 ( G ) L^2(G) L 2 ( G ) -orthogonal projection of u ∗ u_* u ∗ . Here u [ a ] = ∑ j = 1 N a j v j u[a]=\sum_{j=1}^N a_jv_j u [ a ] = ∑ j = 1 N a j v j . Let us fix R > 0 R>0 R > 0 and take a a a in the closed ball of radius R R R in C ( [ 0 , T 1 ] ; R N ) C([0,T_1];\mathbb{R}^N) C ([ 0 , T 1 ] ; R N ) for some T 1 T_1 T 1 to be determined. Then, on this ball the parabolic energy estimate for the given smooth drift u [ a ] u[a] u [ a ] implies a bound for d [ a ] d[a] d [ a ] in L ∞ ( 0 , T 1 ; H 1 ( G ) ) L^\infty(0,T_1;H^1(G)) L ∞ ( 0 , T 1 ; H 1 ( G )) depending only on R R R , N N N , ε \varepsilon ε , and the initial energy. Then, using the integral form of the above ODE we have
∥ T N ( a ) ( t ) − T N ( a ) ( s ) ∥ R N ≤ C ∣ t − s ∣ 1 / 2 , \lVert T_N(a)(t)-T_N(a)(s)\rVert_{\mathbb{R}^N}\le C|t-s|^{1/2}, ∥ T N ( a ) ( t ) − T N ( a ) ( s ) ∥ R N ≤ C ∣ t − s ∣ 1/2 ,
where C > 0 C>0 C > 0 depending on N N N , R R R , ε \varepsilon ε and initial energy. Choose R R R larger than the norm of the projected initial coefficients and then choose T 1 T_1 T 1 sufficiently small, T N T_N T N maps the above ball into itself. The preceding difference estimate for the director implies that T N T_N T N is continuous, and the above modulus implies the relative compactness by the Arzelà–Ascoli theorem. Thus, we can apply Schauder’s theorem to conclude that there exists a fixed point on [ 0 , T 1 ] [0,T_1] [ 0 , T 1 ] .
After removing the auxiliary director Galerkin index, we derive the a priori estimate at the fixed point. For the above velocity u N u^N u N , we first remove the auxiliary Dirichlet Galerkin index in the director equation. The resulting parabolic solution has the full molecular field
M N = Δ d N − f ε ( d N ) ∈ L 2 ( ( 0 , T 1 ) × G ) M^N=\Delta d^N-f_\varepsilon(d^N)\in L^2((0,T_1)\times G) M N = Δ d N − f ε ( d N ) ∈ L 2 (( 0 , T 1 ) × G )
and satisfies the director energy identity. Since Δ d N = M N + f ε ( d N ) ∈ L 2 ( G ) \Delta d^N=M^N+f_\varepsilon(d^N)\in L^2(G) Δ d N = M N + f ε ( d N ) ∈ L 2 ( G ) , and since u N u^N u N has compact support in G G G , the interior elliptic regularity implies d N ∈ H 2 d^N\in H^2 d N ∈ H 2 on a neighborhood of supp u N \operatorname{supp}u^N supp u N . Then, since div u N = 0 \operatorname{div}u^N=0 div u N = 0 , by integrating by parts we have
∫ G ( u N ⋅ ∇ ) d N ⋅ M N d x = − ∫ G ( ∇ d N ) ⊤ ∇ d N : ∇ u N d x . \int_G (u^N\cdot\nabla)d^N\cdot M^N\,\mathrm{d}x=-\int_G(\nabla d^N)^\top\nabla d^N:\nabla u^N\,\mathrm{d}x. ∫ G ( u N ⋅ ∇ ) d N ⋅ M N d x = − ∫ G ( ∇ d N ) ⊤ ∇ d N : ∇ u N d x .
Indeed, the contribution of − f ε ( d N ) = − ∇ F ε ( d N ) -f_\varepsilon(d^N)=-\nabla F_\varepsilon(d^N) − f ε ( d N ) = − ∇ F ε ( d N ) is the integral of − u N ⋅ ∇ F ε ( d N ) -u^N\cdot\nabla F_\varepsilon(d^N) − u N ⋅ ∇ F ε ( d N ) and vanishes, while the Laplacian term yields the Ericksen stress. Testing the finite-dimensional momentum equation by u N u^N u N and adding the director energy identity we get (116) . It implies that the semi-Galerkin energy is non-increasing. In particular, the coefficient vector and the director energy remain bounded on every finite interval, so the local Schauder construction can be continued with a lifespan depending only on the current bounded quantities. There is no finite-time accumulation of restart times. Hence ( u N , d N ) (u^N,d^N) ( u N , d N ) exists on every finite interval.
The energy inequality (116) for ( u N , d N , M N ) (u^N,d^N,M^N) ( u N , d N , M N ) implies the uniform bounds
u N bounded in L t ∞ L x 2 ∩ L t 2 H 0 , x 1 , d N bounded in L t ∞ H x 1 , M N bounded in L t , x 2 . (117) u^N\ \text{bounded in }L_t^\infty L_x^2\cap L_t^2H_{0,x}^1,\qquad d^N\ \text{bounded in }L_t^\infty H_x^1,\qquad M^N\ \text{bounded in }L_{t,x}^2. \tag*{(117)} u N bounded in L t ∞ L x 2 ∩ L t 2 H 0 , x 1 , d N bounded in L t ∞ H x 1 , M N bounded in L t , x 2 . (117)
Since d N ∈ K + d^N\in K_+ d N ∈ K + and ε \varepsilon ε is fixed, f ε ( d N ) f_\varepsilon(d^N) f ε ( d N ) is bounded in L t , x ∞ L_{t,x}^\infty L t , x ∞ . Set w N : = d N − b ∗ ∈ H 0 1 ( G ; R 3 ) w^N:=d^N-b_*\in H_0^1(G;\mathbb{R}^3) w N := d N − b ∗ ∈ H 0 1 ( G ; R 3 ) . Since b ∗ b_* b ∗ is weakly harmonic, by the variational molecular-field equation we have
− Δ w N = − M N − f ε ( d N ) -\Delta w^N=-M^N-f_\varepsilon(d^N) − Δ w N = − M N − f ε ( d N )
in H − 1 ( G ) H^{-1}(G) H − 1 ( G ) . The right-hand side belongs to L 2 ( G ) L^2(G) L 2 ( G ) , hence w N w^N w N is bounded in L 2 ( 0 , T ; D ( A D ) ) L^2(0,T;D(A_D)) L 2 ( 0 , T ; D ( A D )) , where A D A_D A D is the Dirichlet Laplacian on the arbitrary open set G G G . Because G G G is bounded, by the Rellich–Kondrachov theorem with zero extension to a containing ball we have H 0 1 ( G ) ⊆ L 2 ( G ) H_0^1(G)\subseteq L^2(G) H 0 1 ( G ) ⊆ L 2 ( G ) , hence A D A_D A D has compact resolvent. If A D e j = λ j e j A_De_j=\lambda_je_j A D e j = λ j e j is its spectral resolution, then, with c j = ( w , e j ) L 2 ( G ) c_j=(w,e_j)_{L^2(G)} c j = ( w , e j ) L 2 ( G ) ,
∥ ∇ w ∥ L 2 ( G ) 2 = ∑ j λ j ∣ c j ∣ 2 , ∥ A D w ∥ L 2 ( G ) 2 = ∑ j λ j 2 ∣ c j ∣ 2 . \lVert\nabla w\rVert_{L^2(G)}^2=\sum_j\lambda_j|c_j|^2,\qquad\lVert A_Dw\rVert_{L^2(G)}^2=\sum_j\lambda_j^2|c_j|^2. ∥ ∇ w ∥ L 2 ( G ) 2 = j ∑ λ j ∣ c j ∣ 2 , ∥ A D w ∥ L 2 ( G ) 2 = j ∑ λ j 2 ∣ c j ∣ 2 .
The first identity is the quadratic-form identity for the variational Dirichlet Laplacian. On the bounded set G G G , its left-hand side is an equivalent norm on H 0 1 ( G ) H_0^1(G) H 0 1 ( G ) by Poincaré’s inequality. Since λ j → ∞ \lambda_j\to\infty λ j → ∞ as j → ∞ j\to\infty j → ∞ , the tails of the first series are uniformly small on bounded subsets of D ( A D ) D(A_D) D ( A D ) . Thus, without any boundary regularity assumption,
D ( A D ) ⊆ H 0 1 ( G ) ⊆ L 2 ( G ) . D(A_D)\subseteq H_0^1(G)\subseteq L^2(G). D ( A D ) ⊆ H 0 1 ( G ) ⊆ L 2 ( G ) .
Moreover
∂ t w N = M N − u N ⋅ ∇ d N \partial_t w^N=M^N-u^N\cdot\nabla d^N ∂ t w N = M N − u N ⋅ ∇ d N
is bounded in L 2 ( 0 , T ; H − 1 ( G ) ) L^2(0,T;H^{-1}(G)) L 2 ( 0 , T ; H − 1 ( G )) . Therefore, by the Aubin–Lions lemma, after extracting a subsequence we have
d N → d strongly in L 2 ( 0 , T ; H 1 ( G ) ) . (118) d^N \to d \qquad\text{strongly in } L^2(0,T;H^1(G)). \tag*{(118)} d N → d strongly in L 2 ( 0 , T ; H 1 ( G )) . (118)
For the velocity we use Lemma 3.15 . The basis ( v m ) (v_m) ( v m ) was chosen above so that its spans are dense in H G \mathcal{H}_G H G and V G \mathcal{V}_G V G as introduced in § 2.2 . For every fixed m m m and every N ≥ m N\geq m N ≥ m , by the Galerkin equation we have
d d t ( u N , v m ) L 2 ( G ) = ∫ G ( u N ⊗ u N ) : ∇ v m d x − 2 ∫ G D ( u N ) : D ( v m ) d x + ∫ G ( ∇ d N ) ⊤ ∇ d N : ∇ v m d x . \frac{\mathrm{d}}{\mathrm{d}t}(u^N,v_m)_{L^2(G)}=\int_G(u^N\otimes u^N):\nabla v_m\,\mathrm{d}x-2\int_G\mathbb{D}(u^N):\mathbb{D}(v_m)\,\mathrm{d}x+\int_G(\nabla d^N)^\top\nabla d^N:\nabla v_m\,\mathrm{d}x. d t d ( u N , v m ) L 2 ( G ) = ∫ G ( u N ⊗ u N ) : ∇ v m d x − 2 ∫ G D ( u N ) : D ( v m ) d x + ∫ G ( ∇ d N ) ⊤ ∇ d N : ∇ v m d x .
The uniform bounds (117) therefore yield
∫ 0 T ∣ d d t ( u N , v m ) L 2 ( G ) ∣ d t ≤ C m ( T ∥ u N ∥ L t ∞ L x 2 2 + T 1 / 2 ∥ D ( u N ) ∥ L t , x 2 2 + T ∥ ∇ d N ∥ L t ∞ L x 2 2 ) ≤ C m , T . \int_0^T\left|\frac{\mathrm{d}}{\mathrm{d}t}(u^N,v_m)_{L^2(G)}\right|\,\mathrm{d}t\leq C_m\left(T\lVert u^N\rVert_{L^\infty_tL^2_x}^2+T^{1/2}\lVert\mathbb{D}(u^N)\rVert_{L^2_{t,x}}^2+T\lVert\nabla d^N\rVert_{L^\infty_tL^2_x}^2\right)\leq C_{m,T}. ∫ 0 T d t d ( u N , v m ) L 2 ( G ) d t ≤ C m ( T ∥ u N ∥ L t ∞ L x 2 2 + T 1/2 ∥ D ( u N ) ∥ L t , x 2 2 + T ∥ ∇ d N ∥ L t ∞ L x 2 2 ) ≤ C m , T .
Thus the functions t ↦ ( u N ( t ) , v m ) L 2 ( G ) t\mapsto(u^N(t),v_m)_{L^2(G)} t ↦ ( u N ( t ) , v m ) L 2 ( G ) , N ≥ m N\geq m N ≥ m , have uniformly bounded variation on ( 0 , T ) (0,T) ( 0 , T ) , and by Lemma 3.15 we have, after extracting a subsequence, u N → u u^N\to u u N → u strongly in L 2 ( ( 0 , T ) × G ) L^2((0,T)\times G) L 2 (( 0 , T ) × G ) . By (117) , we may also assume that M N ⇀ M M^N\rightharpoonup M M N ⇀ M weakly in L 2 ( ( 0 , T ) × G ) L^2((0,T)\times G) L 2 (( 0 , T ) × G ) . Together with (118) , these convergences allow us to pass to the limit in the Galerkin momentum equation, in the director equation for d N d^N d N , and in the relation M N = Δ d N − f ε ( d N ) M^N=\Delta d^N-f_\varepsilon(d^N) M N = Δ d N − f ε ( d N ) . This implies the two weak equations and the molecular-field identity of the statement. Note that ( ∇ d N ) ⊤ ∇ d N → ( ∇ d ) ⊤ ∇ d (\nabla d^N)^\top\nabla d^N\to(\nabla d)^\top\nabla d ( ∇ d N ) ⊤ ∇ d N → ( ∇ d ) ⊤ ∇ d in L 1 ( ( 0 , T ) × G ) L^1((0,T)\times G) L 1 (( 0 , T ) × G ) , which suffices for test fields in D G \mathcal{D}_G D G .
Since d N ∈ K + d^N\in K_+ d N ∈ K + by Lemma 3.16 , the strong convergence implies d ∈ K + d\in K_+ d ∈ K + a.e., and since d N − b ∗ d^N-b_* d N − b ∗ and b ∗ − d ∗ b_*-d_* b ∗ − d ∗ belong to H 1 ( G ; R 3 ) H^1(G;\mathbb{R}^3) H 1 ( G ; R 3 ) , the weak closedness of this space implies d − d ∗ ∈ H 1 ( G ; R 3 ) d-d_*\in H^1(G;\mathbb{R}^3) d − d ∗ ∈ H 1 ( G ; R 3 ) .
Finally, we recall that ( u N , d N , M N ) (u^N,d^N,M^N) ( u N , d N , M N ) satisfies the energy inequality (116) . After a further extraction, u N ( t ) → u ( t ) u^N(t)\to u(t) u N ( t ) → u ( t ) in L 2 ( G ) L^2(G) L 2 ( G ) and d N ( t ) → d ( t ) d^N(t)\to d(t) d N ( t ) → d ( t ) in H 1 ( G ) H^1(G) H 1 ( G ) for almost everywhere t t t , so the first three terms on its left-hand side converge, and the two dissipation integrals are weakly lower semicontinuous. Hence (116) holds for the limit. Since the approximations exist on every bounded time interval with the same bounds, a diagonal argument over T = 1 , 2 , … T=1,2,\ldots T = 1 , 2 , … yields a single trajectory on [ 0 , ∞ ) [0,\infty) [ 0 , ∞ ) . □
The singular limit 0 0 0 on collision-free intervals
In this section, we carry out the limit passage ε → 0 \varepsilon\to0 ε → 0 in the Ginzburg–Landau regularization, see Theorem 3.2 for the existence result. We refer to the resulting solutions in which the constraint ∣ d ∣ = 1 |d|=1 ∣ d ∣ = 1 holds as constrained weak solutions as opposed to the solutions with regularization parameter ε > 0 \varepsilon>0 ε > 0 . The following proposition records the existence of such constrained weak solutions.
Proposition 4.1. Under the assumptions of Theorem 2.3, there exists a constrained weak solution ( u , d , φ ) (u,d,\varphi) ( u , d , φ ) on [ 0 , T ∗ ) [0,T_*) [ 0 , T ∗ ) , that is, a weak solution on [ 0 , T ] [0,T] [ 0 , T ] in the sense of Definition 2.2 for every T < T ∗ T<T_* T < T ∗ , where T ∗ T_* T ∗ is given by (35) . It satisfies dist ( S ( t ) , ∂ Ω ) ≥ δ 0 / 2 \operatorname{dist}(\mathcal{S}(t),\partial\Omega)\geq\delta_0/2 dist ( S ( t ) , ∂ Ω ) ≥ δ 0 /2 for 0 ≤ t < T ∗ 0\leq t<T_* 0 ≤ t < T ∗ .
The rest of this section is devoted to the proof of Proposition 4.1.
In this section and in Section A.2, we also use the collar-rigid subclass of the standard momentum test class (13) ,
T u c o l : = { ζ ∈ T u : ∃ r > 0 such that D ( ζ ( t ) ) = 0 on S ( t ) r for every t ∈ [ 0 , T ] } , (119) \mathcal{T}^{\mathrm{col}}_u:=\{\zeta\in\mathcal{T}_u:\exists r>0\text{ such that }\mathbb{D}(\zeta(t))=0\text{ on }\mathcal{S}(t)_r\text{ for every }t\in[0,T]\}, \tag*{(119)} T u col := { ζ ∈ T u : ∃ r > 0 such that D ( ζ ( t )) = 0 on S ( t ) r for every t ∈ [ 0 , T ]} , (119)
which serves as an approximation device. The final weak formulation uses (13) .
We return to the datum d 0 ∈ H 1 ( Ω ; S + 2 ) d_0\in H^1(\Omega;\mathbb{S}^2_+) d 0 ∈ H 1 ( Ω ; S + 2 ) from Theorem 2.3. By Lemma A.1 , choose a family
d 0 ε ∈ H 2 ( Ω ; S + 2 ) , d 0 ε → d 0 strongly in H 1 ( Ω ) ( ε → 0 ) . (120) d_0^\varepsilon\in H^2(\Omega;\mathbb{S}^2_+),\qquad d_0^\varepsilon\to d_0\quad\text{strongly in }H^1(\Omega)\quad(\varepsilon\to0). \tag*{(120)} d 0 ε ∈ H 2 ( Ω ; S + 2 ) , d 0 ε → d 0 strongly in H 1 ( Ω ) ( ε → 0 ) . (120)
Set
E 0 ε : = 1 2 ∥ u 0 ∥ L 2 ( Ω ) 2 + 1 2 ∥ ∇ d 0 ε ∥ L 2 ( F 0 ) 2 , E 0 ε → E 0 . E_0^\varepsilon:=\frac{1}{2}\lVert u_0\rVert_{L^2(\Omega)}^2+\frac{1}{2}\lVert\nabla d_0^\varepsilon\rVert_{L^2(\mathcal{F}_0)}^2,\qquad E_0^\varepsilon\to E_0. E 0 ε := 2 1 ∥ u 0 ∥ L 2 ( Ω ) 2 + 2 1 ∥ ∇ d 0 ε ∥ L 2 ( F 0 ) 2 , E 0 ε → E 0 .
Let T ∗ T_{\ast} T ∗ be the time in (35) . In the case E 0 = 0 E_{0}=0 E 0 = 0 , we get u 0 = 0 \boldsymbol{u}_{0}=0 u 0 = 0 and ∇ d 0 = 0 \nabla d_{0}=0 ∇ d 0 = 0 a.e. in F 0 \mathcal{F}_{0} F 0 , so the stationary body together with d ( t ) = d 0 ∣ F 0 d(t)=d_{0}|_{\mathcal{F}_{0}} d ( t ) = d 0 ∣ F 0 is a constrained solution. We thus assume E 0 > 0 E_{0}>0 E 0 > 0 in the following.
Fix T < T ∗ T<T_{\ast} T < T ∗ . For every sufficiently small ε \varepsilon ε , Theorem 3.2 applied with anchor d 0 ε d_{0}^{\varepsilon} d 0 ε gives a solution ( u ε , d ε , φ ε ) (\boldsymbol{u}_{\varepsilon},d_{\varepsilon},\varphi_{\varepsilon}) ( u ε , d ε , φ ε ) on [ 0 , T ] [0,T] [ 0 , T ] . Due to the convergence of the energy E 0 ε → E 0 E_{0}^{\varepsilon}\to E_{0} E 0 ε → E 0 , Lemma 2.8 gives a number ρ T > 0 \rho_{T}>0 ρ T > 0 , independent of small ε \varepsilon ε , such that
dist ( S ε ( t ) , ∂ Ω ) ≥ ρ T ( 0 ≤ t ≤ T ) . (121) \operatorname{dist}(\mathcal{S}_{\varepsilon}(t),\partial\Omega)\geq\rho_{T}\qquad(0\leq t\leq T). \tag*{(121)} dist ( S ε ( t ) , ∂ Ω ) ≥ ρ T ( 0 ≤ t ≤ T ) . (121)
Define τ ε : = Δ d ε + 1 ε 2 ( 1 − ∣ d ε ∣ 2 ) d ε \tau_{\varepsilon}:=\Delta d_{\varepsilon}+\frac{1}{\varepsilon^{2}}(1-|d_{\varepsilon}|^{2})d_{\varepsilon} τ ε := Δ d ε + ε 2 1 ( 1 − ∣ d ε ∣ 2 ) d ε . The energy inequality (40) for ( u ε , d ε , φ ε ) (\boldsymbol{u}_{\varepsilon},d_{\varepsilon},\varphi_{\varepsilon}) ( u ε , d ε , φ ε ) , whose right-hand side equals E 0 ε E_{0}^{\varepsilon} E 0 ε because ∣ d 0 ε ∣ = 1 |d_{0}^{\varepsilon}|=1 ∣ d 0 ε ∣ = 1 , together with (120) and Korn’s inequality, implies
sup 0 < ε < ε T [ ∥ u ε ∥ L ∞ ( 0 , T ; L 2 ( Ω ) ) 2 + ∥ u ε ∥ L 2 ( 0 , T ; H 0 1 ( Ω ) ) 2 + ∥ ∇ d ε ∥ L ∞ ( 0 , T ; L 2 ( F ε ( t ) ) ) 2 + ∫ 0 T ∫ F ε ( t ) ( 1 − ∣ d ε ∣ 2 ) 2 ε 2 d x d t + ∥ τ ε ∥ L 2 ( Q F ε ) 2 ] ≤ C T . (122) \sup_{0<\varepsilon<\varepsilon_{T}}\left[\|\boldsymbol{u}_{\varepsilon}\|_{L^{\infty}(0,T;L^{2}(\Omega))}^{2}+\|\boldsymbol{u}_{\varepsilon}\|_{L^{2}(0,T;H_{0}^{1}(\Omega))}^{2}+\|\nabla d_{\varepsilon}\|_{L^{\infty}(0,T;L^{2}(\mathcal{F}_{\varepsilon}(t)))}^{2}
+\int_{0}^{T}\int_{\mathcal{F}_{\varepsilon}(t)}\frac{(1-|d_{\varepsilon}|^{2})^{2}}{\varepsilon^{2}}\,\mathrm{d}x\,\mathrm{d}t+\|\tau_{\varepsilon}\|_{L^{2}(Q_{\mathcal{F}_{\varepsilon}})}^{2}\right]\leq C_{T}. \tag*{(122)} 0 < ε < ε T sup [ ∥ u ε ∥ L ∞ ( 0 , T ; L 2 ( Ω )) 2 + ∥ u ε ∥ L 2 ( 0 , T ; H 0 1 ( Ω )) 2 + ∥∇ d ε ∥ L ∞ ( 0 , T ; L 2 ( F ε ( t ))) 2 + ∫ 0 T ∫ F ε ( t ) ε 2 ( 1 − ∣ d ε ∣ 2 ) 2 d x d t + ∥ τ ε ∥ L 2 ( Q F ε ) 2 ] ≤ C T . (122)
Rigid motion and velocity compactness
Lemma 4.2. For every sufficiently small ε \varepsilon ε ,
∣ d ε ∣ ≤ 1 , ( d ε ) 3 ≥ 0 a.e. in Q F ε . |d_{\varepsilon}|\leq1,\qquad(d_{\varepsilon})_{3}\geq0\qquad\text{a.e. in }Q_{\mathcal{F}_{\varepsilon}}. ∣ d ε ∣ ≤ 1 , ( d ε ) 3 ≥ 0 a.e. in Q F ε .
Proof. By construction, see Theorem 3.2, the pair ( u ε , d ε ) (\boldsymbol{u}_{\varepsilon},d_{\varepsilon}) ( u ε , d ε ) is obtained from solutions of the semi-Galerkin system of § 3.1.1 with anchor d 0 ε d_{0}^{\varepsilon} d 0 ε through the limits k → ∞ k\to\infty k → ∞ , n → ∞ n\to\infty n → ∞ , and δ → 0 \delta\to0 δ → 0 . The Galerkin director equation (56) is of the form (A.7) with
u = u k , a = n φ k ≥ 0 , b = b k = d 0 ε ∘ Y k , \boldsymbol{u}=\boldsymbol{u}^{k},\qquad a=n\varphi^{k}\geq0,\qquad b=b^{k}=d_{0}^{\varepsilon}\circ Y^{k}, u = u k , a = n φ k ≥ 0 , b = b k = d 0 ε ∘ Y k ,
and its initial and boundary datum is d 0 ε d_{0}^{\varepsilon} d 0 ε . Since d 0 ε d_{0}^{\varepsilon} d 0 ε takes values in S + 2 \mathbb{S}_{+}^{2} S + 2 , Lemma A.5 gives ∣ d k ∣ ≤ 1 |d^{k}|\leq1 ∣ d k ∣ ≤ 1 and ( d k ) 3 ≥ 0 (d^{k})_{3}\geq0 ( d k ) 3 ≥ 0 a.e. in ( 0 , T ) × Ω (0,T)\times\Omega ( 0 , T ) × Ω . These closed convex constraints are preserved by the strong L 2 L^{2} L 2 -convergence of the directors in each of the three limits, see (93) for the post-rigidity level. □
Let h ε h_{\varepsilon} h ε and O ε O_{\varepsilon} O ε be the rigid motions. From Lemma 2.7 and (122) , we infer uniform W 1 , ∞ W^{1,\infty} W 1 , ∞ -bounds. After extraction, as ε → 0 \varepsilon\to0 ε → 0 , we have
h ε → h , O ε → O uniformly on [ 0 , T ] , (123) h_{\varepsilon}\to h,\qquad O_{\varepsilon}\to O\qquad\text{uniformly on }[0,T], \tag*{(123)} h ε → h , O ε → O uniformly on [ 0 , T ] , (123)
and, with φ = 1 h ( t ) + O ( t ) S 0 \varphi=\mathbb{1}_{h(t)+O(t)S_{0}} φ = 1 h ( t ) + O ( t ) S 0 capturing the position of the rigid body, we get the convergence
φ ε → φ in C ( [ 0 , T ] ; L p ( Ω ) ) , 1 ≤ p < ∞ . (124) \varphi_{\varepsilon}\to\varphi\quad\text{in }C([0,T];L^{p}(\Omega)),\qquad1\leq p<\infty. \tag*{(124)} φ ε → φ in C ([ 0 , T ] ; L p ( Ω )) , 1 ≤ p < ∞. (124)
This convergence is due to the fact that corresponding material points of S ε ( t ) \mathcal{S}_{\varepsilon}(t) S ε ( t ) and S ( t ) \mathcal{S}(t) S ( t ) have distance at most
η ε : = ∥ h ε − h ∥ L ∞ ( 0 , T ) + diam ( S 0 ) ∥ O ε − O ∥ L ∞ ( 0 , T ) → 0. \eta_{\varepsilon}:=\|h_{\varepsilon}-h\|_{L^{\infty}(0,T)}+\operatorname{diam}(S_{0})\|O_{\varepsilon}-O\|_{L^{\infty}(0,T)}\to0. η ε := ∥ h ε − h ∥ L ∞ ( 0 , T ) + diam ( S 0 ) ∥ O ε − O ∥ L ∞ ( 0 , T ) → 0.
Hence the solids converge uniformly in Hausdorff distance, and their symmetric difference is contained in the layer { x : dist ( x , ∂ S ( t ) ) ≤ 2 η ε } \{x:\operatorname{dist}(x,\partial S(t))\leq2\eta_{\varepsilon}\} { x : dist ( x , ∂ S ( t )) ≤ 2 η ε } around ∂ S ( t ) \partial S(t) ∂ S ( t ) , a rigid image of the C 2 C^{2} C 2 -surface ∂ S 0 \partial S_{0} ∂ S 0 , whose measure is at most C η ε C\eta_{\varepsilon} C η ε . Consequently,
sup t ∈ [ 0 , T ] ∥ φ ε ( t ) − φ ( t ) ∥ L p ( Ω ) p = sup t ∈ [ 0 , T ] ∣ S ε ( t ) △ S ( t ) ∣ ≤ C η ε . \sup_{t\in[0,T]}\|\varphi_{\varepsilon}(t)-\varphi(t)\|_{L^{p}(\Omega)}^{p}=\sup_{t\in[0,T]}|\mathcal{S}_{\varepsilon}(t)\mathbin{\triangle}\mathcal{S}(t)|\leq C\eta_{\varepsilon}. t ∈ [ 0 , T ] sup ∥ φ ε ( t ) − φ ( t ) ∥ L p ( Ω ) p = t ∈ [ 0 , T ] sup ∣ S ε ( t ) △ S ( t ) ∣ ≤ C η ε .
We apply Lemma 3.14 to u ε \boldsymbol{u}_{\varepsilon} u ε and the trajectories S ε ( t ) \mathcal{S}_{\varepsilon}(t) S ε ( t ) . The separation, convergence, and boundedness hypotheses hold by (121) , (123) , and (122) . To verify (105) , let ζ \zeta ζ be a test field as in the lemma, rigid on S ( t ) σ \mathcal{S}(t)_{\sigma} S ( t ) σ . By the uniform Hausdorff convergence of the solids, S ε ( t ) ⊂ S ( t ) σ \mathcal{S}_{\varepsilon}(t)\subset\mathcal{S}(t)_{\sigma} S ε ( t ) ⊂ S ( t ) σ for all small ε \varepsilon ε , so ζ \zeta ζ is admissible in the momentum equation of Definition 3.1. By the Sobolev embedding of H 3 ( Ω ) H^{3}(\Omega) H 3 ( Ω ) into W 1 , ∞ ( Ω ) W^{1,\infty}(\Omega) W 1 , ∞ ( Ω ) , the convective term and the Ericksen stress are bounded by C T ∥ ζ ∥ L ∞ ( I ; H 3 ( Ω ) ) C_{T}\|\zeta\|_{L^{\infty}(I;H^{3}(\Omega))} C T ∥ ζ ∥ L ∞ ( I ; H 3 ( Ω )) and the viscous term by C T ∥ ζ ∥ L 2 ( I ; H 3 ( Ω ) ) C_{T}\|\zeta\|_{L^{2}(I;H^{3}(\Omega))} C T ∥ ζ ∥ L 2 ( I ; H 3 ( Ω )) , uniformly in ε \varepsilon ε by (122) .
Consequently, after extraction,
u ε → u strongly in L 2 ( ( 0 , T ) × Ω ) , u ε ⇀ u weakly in L 2 ( 0 , T ; H 0 1 ( Ω ) ) , u ε ⇀ ∗ u weak-* in L ∞ ( 0 , T ; L 2 ( Ω ) ) . (125) \begin{aligned}
u_{\varepsilon} &\to u &&\text{strongly in } L^{2}((0,T)\times\Omega),\\
u_{\varepsilon} &\rightharpoonup u &&\text{weakly in } L^{2}(0,T;H_{0}^{1}(\Omega)),\\
u_{\varepsilon} &\overset{*}{\rightharpoonup} u &&\text{weak-* in } L^{\infty}(0,T;L^{2}(\Omega)). \tag*{(125)}
\end{aligned} u ε u ε u ε → u ⇀ u ⇀ ∗ u strongly in L 2 (( 0 , T ) × Ω ) , weakly in L 2 ( 0 , T ; H 0 1 ( Ω )) , weak-* in L ∞ ( 0 , T ; L 2 ( Ω )) . (125)
In the same way as in the limit passage δ → 0 \delta\to0 δ → 0 in (97) , for the limit ε → 0 \varepsilon\to0 ε → 0 , we can show that
u ( t , x ) = ℓ ( t ) + ω ( t ) × ( x − h ( t ) ) for a.e. x ∈ S ( t ) , u(t,x)=\ell(t)+\omega(t)\times(x-h(t)) \quad\text{for a.e. }x\in S(t), u ( t , x ) = ℓ ( t ) + ω ( t ) × ( x − h ( t )) for a.e. x ∈ S ( t ) ,
so the limiting rigid trajectory is generated by the rigid restriction of the limiting velocity, including its orientation.
The material extension through the rigid phase also allows us to deduce global compactness of the director. Define
d ‾ ε ( t , x ) : = { d ε ( t , x ) , for x ∈ F ε ( t ) , d 0 ε ( O ε ( t ) ⊤ ( x − h ε ( t ) ) ) , for x ∈ S ε ( t ) . \overline{d}_{\varepsilon}(t,x):=
\begin{cases}
d_{\varepsilon}(t,x), & \text{for }x\in F_{\varepsilon}(t),\\
d_{0}^{\varepsilon}\left(O_{\varepsilon}(t)^{\top}(x-h_{\varepsilon}(t))\right), & \text{for }x\in S_{\varepsilon}(t).
\end{cases} d ε ( t , x ) := { d ε ( t , x ) , d 0 ε ( O ε ( t ) ⊤ ( x − h ε ( t )) ) , for x ∈ F ε ( t ) , for x ∈ S ε ( t ) .
The anchoring condition in Definition 3.1 ensures that the two traces agree on the moving interface, so Lemma A.4 implies that d ‾ ε ( t ) ∈ H 1 ( Ω ; R 3 ) \overline{d}_{\varepsilon}(t)\in H^{1}(\Omega;\mathbb{R}^{3}) d ε ( t ) ∈ H 1 ( Ω ; R 3 ) . The fact that the change of variables x = h ε ( t ) + O ε ( t ) y x=h_{\varepsilon}(t)+O_{\varepsilon}(t)y x = h ε ( t ) + O ε ( t ) y is an isometry of S 0 S_{0} S 0 onto S ε ( t ) S_{\varepsilon}(t) S ε ( t ) leads to
∥ ∇ d ‾ ε ( t ) ∥ L 2 ( Ω ) 2 = ∥ ∇ d ε ( t ) ∥ L 2 ( F ε ( t ) ) 2 + ∥ ∇ d 0 ε ∥ L 2 ( S 0 ) 2 . \|\nabla\overline{d}_{\varepsilon}(t)\|_{L^{2}(\Omega)}^{2}
=
\|\nabla d_{\varepsilon}(t)\|_{L^{2}(F_{\varepsilon}(t))}^{2}
+
\|\nabla d_{0}^{\varepsilon}\|_{L^{2}(S_{0})}^{2}. ∥∇ d ε ( t ) ∥ L 2 ( Ω ) 2 = ∥∇ d ε ( t ) ∥ L 2 ( F ε ( t )) 2 + ∥∇ d 0 ε ∥ L 2 ( S 0 ) 2 .
The first term is bounded by (122) and the second by (120) . Joint with ∣ d ‾ ε ∣ ≤ 1 |\overline{d}_{\varepsilon}|\le1 ∣ d ε ∣ ≤ 1 , this yields
sup ε < ε T ∥ d ‾ ε ∥ L ∞ ( 0 , T ; H 1 ( Ω ) ) ≤ C T . \sup_{\varepsilon<\varepsilon_{T}}\|\overline{d}_{\varepsilon}\|_{L^{\infty}(0,T;H^{1}(\Omega))}\le C_{T}. ε < ε T sup ∥ d ε ∥ L ∞ ( 0 , T ; H 1 ( Ω )) ≤ C T .
Let τ ~ ε \widetilde{\tau}_{\varepsilon} τ ε be the zero extension of τ ε \tau_{\varepsilon} τ ε to the solid. An application of Lemma A.4 with anchor d 0 ε d_{0}^{\varepsilon} d 0 ε gives
∂ t d ‾ ε + ( u ε ⋅ ∇ ) d ‾ ε = τ ~ ε in D ′ ( ( 0 , T ) × Ω ) . \partial_{t}\overline{d}_{\varepsilon}+(u_{\varepsilon}\cdot\nabla)\overline{d}_{\varepsilon}
=\widetilde{\tau}_{\varepsilon}
\quad\text{in }\mathcal{D}'((0,T)\times\Omega). ∂ t d ε + ( u ε ⋅ ∇ ) d ε = τ ε in D ′ (( 0 , T ) × Ω ) .
By u ε ∈ L t 2 L x 6 u_{\varepsilon}\in L_{t}^{2}L_{x}^{6} u ε ∈ L t 2 L x 6 and ∇ d ‾ ε ∈ L t ∞ L x 2 \nabla\overline{d}_{\varepsilon}\in L_{t}^{\infty}L_{x}^{2} ∇ d ε ∈ L t ∞ L x 2 , ( u ε ⋅ ∇ ) d ‾ ε (u_{\varepsilon}\cdot\nabla)\overline{d}_{\varepsilon} ( u ε ⋅ ∇ ) d ε is bounded in L 2 ( 0 , T ; L 3 / 2 ( Ω ) ) ↪ L 2 ( 0 , T ; H − 1 ( Ω ) ) L^{2}(0,T;L^{3/2}(\Omega))\hookrightarrow L^{2}(0,T;H^{-1}(\Omega)) L 2 ( 0 , T ; L 3/2 ( Ω )) ↪ L 2 ( 0 , T ; H − 1 ( Ω )) . On the other hand, τ ~ ε \widetilde{\tau}_{\varepsilon} τ ε is bounded in L t , x 2 L_{t,x}^{2} L t , x 2 . Hence, by the Aubin–Lions lemma, after extracting a subsequence we have
d ‾ ε → d ‾ strongly in L 2 ( ( 0 , T ) × Ω ) , d ‾ ε ⇀ ∗ d ‾ in L ∞ ( 0 , T ; H 1 ( Ω ) ) . (126) \overline{d}_{\varepsilon}\to\overline{d}
\quad\text{strongly in }L^{2}((0,T)\times\Omega),
\qquad
\overline{d}_{\varepsilon}\overset{*}{\rightharpoonup}\overline{d}
\quad\text{in }L^{\infty}(0,T;H^{1}(\Omega)). \tag*{(126)} d ε → d strongly in L 2 (( 0 , T ) × Ω ) , d ε ⇀ ∗ d in L ∞ ( 0 , T ; H 1 ( Ω )) . (126)
On every compact subset of the limiting fluid spacetime, the restriction of d ‾ \overline{d} d is the director identified below by the local Lin–Wang argument. On the solid, it is the rigidly transported limit of d 0 ε d_{0}^{\varepsilon} d 0 ε , namely d 0 ( O ( t ) ⊤ ( x − h ( t ) ) ) d_{0}(O(t)^{\top}(x-h(t))) d 0 ( O ( t ) ⊤ ( x − h ( t ))) .
A quantitative Lin–Wang compactness principle
The singular director limit is localized to compact subsets of the moving fluid region. The required static compactness statement is the following.
Lemma 4.3. Let U ⋐ R 3 U\Subset\mathbb{R}^{3} U ⋐ R 3 be smooth and bounded and let V ⋐ U V\Subset U V ⋐ U . Suppose e j → 0 e_{j}\to0 e j → 0 and v j ∈ H 1 ( U ; R 3 ) v_{j}\in H^{1}(U;\mathbb{R}^{3}) v j ∈ H 1 ( U ; R 3 ) satisfies
Δ v j + 1 e j 2 ( 1 − ∣ v j ∣ 2 ) v j = τ j in U , (127) \Delta v_{j}+\frac{1}{e_{j}^{2}}(1-|v_{j}|^{2})v_{j}=\tau_{j}
\quad\text{in }U, \tag*{(127)} Δ v j + e j 2 1 ( 1 − ∣ v j ∣ 2 ) v j = τ j in U , (127) with
∣ v j ∣ ≤ 1 , ( v j ) 3 ≥ 0 , |v_{j}|\le1,\qquad(v_{j})_{3}\ge0, ∣ v j ∣ ≤ 1 , ( v j ) 3 ≥ 0 , sup j ∫ U ( 1 2 ∣ ∇ v j ∣ 2 + ( 1 − ∣ v j ∣ 2 ) 2 4 e j 2 ) d x < ∞ , sup j ∥ τ j ∥ L 2 ( U ) < ∞ . (128) \sup_{j}\int_{U}\left(\frac{1}{2}|\nabla v_{j}|^{2}
+\frac{(1-|v_{j}|^{2})^{2}}{4e_{j}^{2}}\right)\,dx<\infty,
\qquad
\sup_{j}\|\tau_{j}\|_{L^{2}(U)}<\infty. \tag*{(128)} j sup ∫ U ( 2 1 ∣∇ v j ∣ 2 + 4 e j 2 ( 1 − ∣ v j ∣ 2 ) 2 ) d x < ∞ , j sup ∥ τ j ∥ L 2 ( U ) < ∞. (128) Then ( v j ) (v_{j}) ( v j ) is precompact in H 1 ( V ) H^{1}(V) H 1 ( V ) . Every limit belongs to H 1 ( V ; S + 2 ) H^{1}(V;S_{+}^{2}) H 1 ( V ; S + 2 ) .
Proof. This is a direct local application of [39 Theorem 6.1] . In the notation of that theorem, (4.9)–(4.10) place v j v_j v j in the class X ( L 1 , L 2 , a ; U ) \mathrm{X}(L_1,L_2,a;U) X ( L 1 , L 2 , a ; U ) with a = 1 a=1 a = 1 . Indeed, ∣ v j ∣ ≤ 1 |v_j|\le1 ∣ v j ∣ ≤ 1 , and ( v j ) 3 ≥ 0 (v_j)_3\ge0 ( v j ) 3 ≥ 0 is the condition ( v j ) 3 ≥ − 1 + a (v_j)_3\ge-1+a ( v j ) 3 ≥ − 1 + a for a = 1 a=1 a = 1 , while the two remaining defining bounds are the uniform bound for the Ginzburg–Landau energy of v j v_j v j and the uniform L 2 ( U ) L^2(U) L 2 ( U ) -bound for its tension τ j \tau_j τ j , both of which are assumed in (128) . The cited theorem states that X ( L 1 , L 2 , a ; U ) \mathrm{X}(L_1,L_2,a;U) X ( L 1 , L 2 , a ; U ) is precompact in H l o c 1 ( U ; R 3 ) H^1_{\mathrm{loc}}(U;\mathbb{R}^3) H loc 1 ( U ; R 3 ) . Hence ( v j ) (v_j) ( v j ) is precompact in H 1 ( V ) H^1(V) H 1 ( V ) for every V ⋐ U V\Subset U V ⋐ U .
If v j → v v_j\to v v j → v strongly in H 1 ( V ) H^1(V) H 1 ( V ) along a convergent subsequence, then
∥ 1 − ∣ v j ∣ ∥ L 2 ( V ) 2 ≤ 4 ε j 2 sup i ∫ U ( 1 2 ∣ ∇ v i ∣ 2 + ( 1 − ∣ v i ∣ 2 ) 2 4 ε i 2 ) d x → 0. \lVert1-|v_j|\rVert_{L^2(V)}^2\le4\varepsilon_j^2\sup_i\int_U\left(\frac{1}{2}|\nabla v_i|^2+\frac{(1-|v_i|^2)^2}{4\varepsilon_i^2}\right)\,dx\to0. ∥ 1 − ∣ v j ∣ ∥ L 2 ( V ) 2 ≤ 4 ε j 2 i sup ∫ U ( 2 1 ∣∇ v i ∣ 2 + 4 ε i 2 ( 1 − ∣ v i ∣ 2 ) 2 ) d x → 0.
Thus ∣ v ∣ = 1 |v|=1 ∣ v ∣ = 1 a.e. on V V V . The closed half-space condition ( v j ) 3 ≥ 0 (v_j)_3\ge0 ( v j ) 3 ≥ 0 passes under the strong L 2 L^2 L 2 -convergence, so v ∈ H 1 ( V ; S + 2 ) v\in H^1(V;\mathbb{S}^2_+) v ∈ H 1 ( V ; S + 2 ) .
Remark 4.4. Note that Lemma 4.3 only provides compactness, and it does not identify the weak limit of the tensions τ j \tau_j τ j . In fact, the limiting tension is identified separately, see the cross-product identity and the tangency relation (134) in the proof of Proposition 4.7.
The following lemma establishes a uniform local bound of director fields in H 1 H^1 H 1 .
Lemma 4.5. Consider smooth bounded V ⋐ U ⊂ R 3 V\Subset U\subset\mathbb{R}^3 V ⋐ U ⊂ R 3 as well as numbers L , Λ > 0 L,\Lambda>0 L , Λ > 0 . For every η > 0 \eta>0 η > 0 , there exists r = r ( U , V , L , Λ , η ) > 0 r=r(U,V,L,\Lambda,\eta)>0 r = r ( U , V , L , Λ , η ) > 0 such that if 0 < ε 1 , ε 2 ≤ 1 0<\varepsilon_1,\varepsilon_2\le1 0 < ε 1 , ε 2 ≤ 1 and v i ∈ H 1 ( U ; R 3 ) v_i\in H^1(U;\mathbb{R}^3) v i ∈ H 1 ( U ; R 3 ) satisfy
Δ v i + 1 ε i 2 ( 1 − ∣ v i ∣ 2 ) v i = τ i , ∣ v i ∣ ≤ 1 , ( v i ) 3 ≥ 0 , \Delta v_i+\frac{1}{\varepsilon_i^2}(1-|v_i|^2)v_i=\tau_i,\qquad|v_i|\le1,\qquad(v_i)_3\ge0, Δ v i + ε i 2 1 ( 1 − ∣ v i ∣ 2 ) v i = τ i , ∣ v i ∣ ≤ 1 , ( v i ) 3 ≥ 0 , ∫ U ( 1 2 ∣ ∇ v i ∣ 2 + ( 1 − ∣ v i ∣ 2 ) 2 4 ε i 2 ) d x ≤ L , ∥ τ i ∥ L 2 ( U ) ≤ Λ , \int_U\left(\frac{1}{2}|\nabla v_i|^2+\frac{(1-|v_i|^2)^2}{4\varepsilon_i^2}\right)\,dx\le L,\qquad\|\tau_i\|_{L^2(U)}\le\Lambda, ∫ U ( 2 1 ∣∇ v i ∣ 2 + 4 ε i 2 ( 1 − ∣ v i ∣ 2 ) 2 ) d x ≤ L , ∥ τ i ∥ L 2 ( U ) ≤ Λ , and ∥ v 1 − v 2 ∥ L 2 ( U ) ≤ r \|v_1-v_2\|_{L^2(U)}\le r ∥ v 1 − v 2 ∥ L 2 ( U ) ≤ r , then ∥ v 1 − v 2 ∥ H 1 ( V ) ≤ η \|v_1-v_2\|_{H^1(V)}\le\eta ∥ v 1 − v 2 ∥ H 1 ( V ) ≤ η .
Proof. Suppose by contradiction that the assertion fails. Then there are pairs ( v 1 , j , v 2 , j ) (v_{1,j},v_{2,j}) ( v 1 , j , v 2 , j ) with parameters ε 1 , j , ε 2 , j ∈ ( 0 , 1 ] \varepsilon_{1,j},\varepsilon_{2,j}\in(0,1] ε 1 , j , ε 2 , j ∈ ( 0 , 1 ] such that both maps satisfy the above bounds, ∥ v 1 , j − v 2 , j ∥ L 2 ( U ) → 0 \|v_{1,j}-v_{2,j}\|_{L^2(U)}\to0 ∥ v 1 , j − v 2 , j ∥ L 2 ( U ) → 0 , but ∥ v 1 , j − v 2 , j ∥ H 1 ( V ) > η \|v_{1,j}-v_{2,j}\|_{H^1(V)}>\eta ∥ v 1 , j − v 2 , j ∥ H 1 ( V ) > η for every j j j . Upon passing to a subsequence, for i = 1 , 2 i=1,2 i = 1 , 2 , we may assume ε i , j → ε ˉ i ∈ [ 0 , 1 ] \varepsilon_{i,j}\to\bar\varepsilon_i\in[0,1] ε i , j → ε ˉ i ∈ [ 0 , 1 ] . Next, we choose an intermediate smooth domain W W W such that V ⋐ W ⋐ U V\Subset W\Subset U V ⋐ W ⋐ U . We claim that, for each i i i , the sequence { v i , j } j \{v_{i,j}\}_j { v i , j } j is precompact in H 1 ( V ) H^1(V) H 1 ( V ) .
If ε ˉ i = 0 \bar\varepsilon_i=0 ε ˉ i = 0 , then ( v i , j ) j (v_{i,j})_j ( v i , j ) j satisfies the hypotheses of Lemma 4.3 with ε j = ε i , j → 0 \varepsilon_j=\varepsilon_{i,j}\to0 ε j = ε i , j → 0 and τ j = τ i , j \tau_j=\tau_{i,j} τ j = τ i , j . In fact, (127) and the constraints ∣ v i , j ∣ ≤ 1 |v_{i,j}|\le1 ∣ v i , j ∣ ≤ 1 , ( v i , j ) 3 ≥ 0 (v_{i,j})_3\ge0 ( v i , j ) 3 ≥ 0 hold by assumption, and the bounds (128) hold with the constants L L L and Λ \Lambda Λ , showing precompactness of the sequence in H 1 ( V ) H^1(V) H 1 ( V ) .
On the other hand, if ε ˉ i > 0 \bar\varepsilon_i>0 ε ˉ i > 0 , then ε i , j ≥ ε ˉ i / 2 \varepsilon_{i,j}\ge\bar\varepsilon_i/2 ε i , j ≥ ε ˉ i /2 for all sufficiently large j j j . In
Δ v i , j = τ i , j − 1 ε i , j 2 ( 1 − ∣ v i , j ∣ 2 ) v i , j \Delta v_{i,j}=\tau_{i,j}-\frac{1}{\varepsilon_{i,j}^2}(1-|v_{i,j}|^2)v_{i,j} Δ v i , j = τ i , j − ε i , j 2 1 ( 1 − ∣ v i , j ∣ 2 ) v i , j
the bound ∣ v i , j ∣ ≤ 1 |v_{i,j}|\le1 ∣ v i , j ∣ ≤ 1 yields that the right-hand side is uniformly bounded in L 2 ( U ) L^2(U) L 2 ( U ) . By the standard interior elliptic estimates we get a uniform H 2 ( W ) H^2(W) H 2 ( W ) -bound. From H 2 ( W ) ⋐ H 1 ( V ) H^2(W)\Subset H^1(V) H 2 ( W ) ⋐ H 1 ( V ) , it follows that the sequence is precompact in H 1 ( V ) H^1(V) H 1 ( V ) .
In summary, after passing to a common subsequence, v i , j → v i v_{i,j}\to v_i v i , j → v i strongly in H 1 ( V ) H^1(V) H 1 ( V ) for i = 1 , 2 i=1,2 i = 1 , 2 . The L 2 ( U ) L^2(U) L 2 ( U ) -distance tending to zero forces v 1 = v 2 v_1=v_2 v 1 = v 2 a.e. on V V V , and hence the H 1 ( V ) H^1(V) H 1 ( V ) -distance also tends to zero, a contradiction.
Lemma 4.6. Let I I I be a bounded interval and V ⋐ U V\Subset U V ⋐ U as in Lemma 4.5 . Let ε j ∈ ( 0 , 1 ] \varepsilon_j\in(0,1] ε j ∈ ( 0 , 1 ] , and let d j ∈ L 2 ( I ; H 1 ( U ; R 3 ) ) d_j\in L^2(I;H^1(U;\mathbb{R}^3)) d j ∈ L 2 ( I ; H 1 ( U ; R 3 )) and τ j ∈ L 2 ( I × U ; R 3 ) \tau_j\in L^2(I\times U;\mathbb{R}^3) τ j ∈ L 2 ( I × U ; R 3 ) satisfy, for a.e. t ∈ I t\in I t ∈ I ,
Δ d j ( t ) + 1 ε j 2 ( 1 − ∣ d j ( t ) ∣ 2 ) d j ( t ) = τ j ( t ) in U , ∣ d j ( t ) ∣ ≤ 1 , ( d j ( t ) ) 3 ≥ 0 , (129) \Delta d_j(t)+\frac{1}{\varepsilon_j^2}(1-|d_j(t)|^2)d_j(t)=\tau_j(t)\quad\text{in }U,\qquad|d_j(t)|\le1,\qquad(d_j(t))_3\ge0, \tag*{(129)} Δ d j ( t ) + ε j 2 1 ( 1 − ∣ d j ( t ) ∣ 2 ) d j ( t ) = τ j ( t ) in U , ∣ d j ( t ) ∣ ≤ 1 , ( d j ( t ) ) 3 ≥ 0 , (129)
∫ U ( 1 2 ∣ ∇ d j ( t ) ∣ 2 + ( 1 − ∣ d j ( t ) ∣ 2 ) 2 4 ϵ j 2 ) d x ≤ L , ∫ I ∥ τ j ( t ) ∥ L 2 ( U ) 2 d t ≤ C . (130) \int_U \left(\frac{1}{2}|\nabla d_j(t)|^2+\frac{(1-|d_j(t)|^2)^2}{4\epsilon_j^2}\right)\,dx\le L,\qquad\int_I\|\tau_j(t)\|_{L^2(U)}^2\,dt\le C. \tag*{(130)} ∫ U ( 2 1 ∣∇ d j ( t ) ∣ 2 + 4 ϵ j 2 ( 1 − ∣ d j ( t ) ∣ 2 ) 2 ) d x ≤ L , ∫ I ∥ τ j ( t ) ∥ L 2 ( U ) 2 d t ≤ C . (130)
If ( d j ) (d_j) ( d j ) is a Cauchy sequence in L 2 ( I × U ) L^2(I\times U) L 2 ( I × U ) , then it is a Cauchy sequence in L 2 ( I ; H 1 ( V ) ) L^2(I;H^1(V)) L 2 ( I ; H 1 ( V )) .
Proof. Fix η > 0 \eta>0 η > 0 and Λ > 0 \Lambda>0 Λ > 0 , and let r r r be given by Lemma 4.5 . Define
B j , Λ : = { t ∈ I : ∥ τ j ( t ) ∥ L 2 ( U ) > Λ } , A i j , r : = { t ∈ I : ∥ d i ( t ) − d j ( t ) ∥ L 2 ( U ) > r } . B_{j,\Lambda}:=\{t\in I:\|\tau_j(t)\|_{L^2(U)}>\Lambda\},\qquad A_{ij,r}:=\{t\in I:\|d_i(t)-d_j(t)\|_{L^2(U)}>r\}. B j , Λ := { t ∈ I : ∥ τ j ( t ) ∥ L 2 ( U ) > Λ } , A ij , r := { t ∈ I : ∥ d i ( t ) − d j ( t ) ∥ L 2 ( U ) > r } .
By Chebyshev’s inequality we have
∣ B j , Λ ∣ ≤ C Λ − 2 , ∣ A i j , r ∣ ≤ r − 2 ∥ d i − d j ∥ L 2 ( I × U ) 2 . |B_{j,\Lambda}|\le C\Lambda^{-2},\qquad|A_{ij,r}|\le r^{-2}\|d_i-d_j\|_{L^2(I\times U)}^2. ∣ B j , Λ ∣ ≤ C Λ − 2 , ∣ A ij , r ∣ ≤ r − 2 ∥ d i − d j ∥ L 2 ( I × U ) 2 .
We now address the set I ∖ ( B i , Λ ∪ B j , Λ ∪ A i j , r ) I\setminus(B_{i,\Lambda}\cup B_{j,\Lambda}\cup A_{ij,r}) I ∖ ( B i , Λ ∪ B j , Λ ∪ A ij , r ) . In fact, for a.e. t ∈ I ∖ ( B i , Λ ∪ B j , Λ ∪ A i j , r ) t\in I\setminus(B_{i,\Lambda}\cup B_{j,\Lambda}\cup A_{ij,r}) t ∈ I ∖ ( B i , Λ ∪ B j , Λ ∪ A ij , r ) , (129) and (130) imply that d i ( t ) d_i(t) d i ( t ) and d j ( t ) d_j(t) d j ( t ) satisfy the hypotheses of Lemma 4.5 . Thus, ∥ d i ( t ) − d j ( t ) ∥ H 1 ( V ) ≤ η \|d_i(t)-d_j(t)\|_{H^1(V)}\le\eta ∥ d i ( t ) − d j ( t ) ∥ H 1 ( V ) ≤ η . The energy bound in (130) gives ∫ V ∣ ∇ d i − ∇ d j ∣ 2 d x ≤ 8 L \int_V|\nabla d_i-\nabla d_j|^2\,dx\le8L ∫ V ∣∇ d i − ∇ d j ∣ 2 d x ≤ 8 L and ∣ d i − d j ∣ ≤ 2 |d_i-d_j|\le2 ∣ d i − d j ∣ ≤ 2 . In total, we get
lim sup i , j → ∞ ∥ d i − d j ∥ L 2 ( I ; H 1 ( V ) ) 2 ≤ ∣ I ∣ η 2 + C U , L Λ − 2 . \limsup_{i,j\to\infty}\|d_i-d_j\|_{L^2(I;H^1(V))}^2\le|I|\eta^2+C_{U,L}\Lambda^{-2}. i , j → ∞ lim sup ∥ d i − d j ∥ L 2 ( I ; H 1 ( V )) 2 ≤ ∣ I ∣ η 2 + C U , L Λ − 2 .
The proof is completed upon letting i , j → ∞ i,j\to\infty i , j → ∞ , then Λ → ∞ \Lambda\to\infty Λ → ∞ , and finally η → 0 \eta\to0 η → 0 .
Proposition 4.7. There exist a director field d d d and a tension field τ ∈ L 2 ( Q F ; R 3 ) \tau\in L^2(Q_{\mathcal F};\mathbb{R}^3) τ ∈ L 2 ( Q F ; R 3 ) such that, after extraction, for every compact K ⋐ Q F K\Subset Q_{\mathcal F} K ⋐ Q F , we get
d ϵ → d strongly in L 2 ( K ) , ∇ d ϵ → ∇ d strongly in L 2 ( K ) , (131) d_\epsilon\to d\quad\text{strongly in }L^2(K),\qquad\nabla d_\epsilon\to\nabla d\quad\text{strongly in }L^2(K), \tag*{(131)} d ϵ → d strongly in L 2 ( K ) , ∇ d ϵ → ∇ d strongly in L 2 ( K ) , (131)
and τ ϵ ⇀ τ \tau_\epsilon\rightharpoonup\tau τ ϵ ⇀ τ weakly in L 2 ( K ) L^2(K) L 2 ( K ) . Moreover, we have ∣ d ∣ = 1 |d|=1 ∣ d ∣ = 1 , d 3 ≥ 0 d_3\ge0 d 3 ≥ 0 , and τ = Δ d + ∣ ∇ d ∣ 2 d \tau=\Delta d+|\nabla d|^2d τ = Δ d + ∣∇ d ∣ 2 d in D ′ ( Q F ) D'(Q_{\mathcal F}) D ′ ( Q F ) .
Proof. Let K ⋐ Q F K\Subset Q_{\mathcal F} K ⋐ Q F . A finite covering gives cylinders I j × V j I_j\times V_j I j × V j and smooth domains V j ⋐ U j V_j\Subset U_j V j ⋐ U j such that U j ⊂ F ϵ ( t ) ∩ F ( t ) U_j\subset\mathcal F_\epsilon(t)\cap\mathcal F(t) U j ⊂ F ϵ ( t ) ∩ F ( t ) for t ∈ I j t\in I_j t ∈ I j and all small ϵ \epsilon ϵ . On I j × U j I_j\times U_j I j × U j ,
∂ t d ϵ = τ ϵ − ( u ϵ ⋅ ∇ ) d ϵ . \partial_t d_\epsilon=\tau_\epsilon-(u_\epsilon\cdot\nabla)d_\epsilon. ∂ t d ϵ = τ ϵ − ( u ϵ ⋅ ∇ ) d ϵ .
By (122) , τ ϵ \tau_\epsilon τ ϵ is bounded in L 2 ( I j × U j ) L^2(I_j\times U_j) L 2 ( I j × U j ) , u ϵ u_\epsilon u ϵ is bounded in L 2 ( I j ; L 6 ( U j ) ) L^2(I_j;L^6(U_j)) L 2 ( I j ; L 6 ( U j )) by the Sobolev embedding H 0 1 ( Ω ) ↪ L 6 ( Ω ) H_0^1(\Omega)\hookrightarrow L^6(\Omega) H 0 1 ( Ω ) ↪ L 6 ( Ω ) , and ∇ d ϵ \nabla d_\epsilon ∇ d ϵ is bounded in L ∞ ( I j ; L 2 ( U j ) ) L^\infty(I_j;L^2(U_j)) L ∞ ( I j ; L 2 ( U j )) . Hölder’s inequality then yields that ( u ϵ ⋅ ∇ ) d ϵ (u_\epsilon\cdot\nabla)d_\epsilon ( u ϵ ⋅ ∇ ) d ϵ is bounded in L 2 ( I j ; L 3 / 2 ( U j ) ) L^2(I_j;L^{3/2}(U_j)) L 2 ( I j ; L 3/2 ( U j )) , so in total, the right-hand side is bounded in L 2 ( I j ; W − 1 , 3 / 2 ( U j ) ) L^2(I_j;W^{-1,3/2}(U_j)) L 2 ( I j ; W − 1 , 3/2 ( U j )) . Moreover, due to ∣ d ϵ ∣ ≤ 1 |d_\epsilon|\le1 ∣ d ϵ ∣ ≤ 1 , d ϵ d_\epsilon d ϵ is bounded in L 2 ( I j ; H 1 ( U j ) ) L^2(I_j;H^1(U_j)) L 2 ( I j ; H 1 ( U j )) . By H 1 ( U j ) ⋐ L 2 ( U j ) ↪ W − 1 , 3 / 2 ( U j ) H^1(U_j)\Subset L^2(U_j)\hookrightarrow W^{-1,3/2}(U_j) H 1 ( U j ) ⋐ L 2 ( U j ) ↪ W − 1 , 3/2 ( U j ) , applying the Aubin–Lions lemma, after extracting a subsequence we have
d ϵ → d strongly in L 2 ( I j × U j ) . d_\epsilon\to d\quad\text{strongly in }L^2(I_j\times U_j). d ϵ → d strongly in L 2 ( I j × U j ) .
For a.e. t ∈ I j t\in I_j t ∈ I j , d ϵ ( t ) d_\epsilon(t) d ϵ ( t ) satisfies (129) on U j U_j U j with ϵ j = ϵ \epsilon_j=\epsilon ϵ j = ϵ and tension τ ϵ ( t ) \tau_\epsilon(t) τ ϵ ( t ) , by the definition of τ ϵ \tau_\epsilon τ ϵ and the bounds ∣ d ϵ ∣ ≤ 1 |d_\epsilon|\le1 ∣ d ϵ ∣ ≤ 1 , ( d ϵ ) 3 ≥ 0 (d_\epsilon)_3\ge0 ( d ϵ ) 3 ≥ 0 . On the other hand, the bounds (130) follow from (122) . Lemma 4.6 then upgrades the previous convergence to
d ϵ → d strongly in L 2 ( I j ; H 1 ( V j ) ) . d_\epsilon\to d\quad\text{strongly in }L^2(I_j;H^1(V_j)). d ϵ → d strongly in L 2 ( I j ; H 1 ( V j )) .
Invoking a finite covering, we conclude (131) . The estimate ∫ I j ∫ U j ( 1 − ∣ d ϵ ∣ 2 ) 2 d x d t ≤ C T ϵ 2 \int_{I_j}\int_{U_j}(1-|d_\epsilon|^2)^2\,dx\,dt\le CT\epsilon^2 ∫ I j ∫ U j ( 1 − ∣ d ϵ ∣ 2 ) 2 d x d t ≤ C T ϵ 2 , which follows from (122) , together with the strong convergence gives ∣ d ∣ = 1 |d|=1 ∣ d ∣ = 1 a.e., while ( d ϵ ) 3 ≥ 0 (d_\epsilon)_3\ge0 ( d ϵ ) 3 ≥ 0 yields d 3 ≥ 0 d_3\ge0 d 3 ≥ 0 .
To identify the tension, we take the cross product of the elliptic equation with d ϵ d_\epsilon d ϵ and obtain
d ϵ × τ ϵ = d ϵ × Δ d ϵ = div ( d ϵ × ∇ d ϵ ) . d_\epsilon\times\tau_\epsilon=d_\epsilon\times\Delta d_\epsilon=\operatorname{div}(d_\epsilon\times\nabla d_\epsilon). d ϵ × τ ϵ = d ϵ × Δ d ϵ = div ( d ϵ × ∇ d ϵ ) .
By (122) , the tensions τ ϵ \tau_\epsilon τ ϵ are bounded in L 2 ( K ) L^2(K) L 2 ( K ) for every compact K ⋐ Q F K\Subset Q_{\mathcal F} K ⋐ Q F , uniformly in K K K . Hence, after extraction, we infer the existence of τ ∈ L 2 ( Q F ; R 3 ) \tau\in L^2(Q_{\mathcal F};\mathbb{R}^3) τ ∈ L 2 ( Q F ; R 3 ) such that τ ϵ ⇀ τ \tau_\epsilon\rightharpoonup\tau τ ϵ ⇀ τ weakly in L 2 ( K ) L^2(K) L 2 ( K ) for every such K K K . Testing the above identity against smooth compactly supported fields and using the strong convergence (131) together with this weak convergence gives, in the limit,
d × τ = div ( d × ∇ d ) in D ′ ( Q F ) . (132) d\times\tau=\operatorname{div}(d\times\nabla d)\quad\text{in }D'(Q_{\mathcal F}). \tag*{(132)} d × τ = div ( d × ∇ d ) in D ′ ( Q F ) . (132)
We now fix a cylinder I × U I \times U I × U with I ‾ × U ‾ ⋐ Q F \overline{I} \times\overline{U} \Subset Q_F I × U ⋐ Q F . Concatenating ∣ d ∣ = 1 |d| = 1 ∣ d ∣ = 1 , τ ∈ L 2 ( Q F ) \tau\in L^2(Q_F) τ ∈ L 2 ( Q F ) , and ∇ d ∈ L 2 ( I × U ) \nabla d \in L^2(I \times U) ∇ d ∈ L 2 ( I × U ) by (131) , we find that d × τ d \times\tau d × τ and d × ∇ d d \times\nabla d d × ∇ d belong to L 2 ( I × U ) L^2(I \times U) L 2 ( I × U ) . As (132) contains no time derivative, it extends by density to
∫ I ∫ U ( d × τ ) ⋅ χ d x d t = − ∑ i = 1 3 ∫ I ∫ U ( d × ∂ i d ) ⋅ ∂ i χ d x d t for χ ∈ L 2 ( I ; H 0 1 ( U ; R 3 ) ) . (133) \int_I \int_U (d \times\tau) \cdot\chi\,\mathrm{d}x \,\mathrm{d}t = -\sum_{i=1}^{3} \int_I \int_U (d \times\partial_i d) \cdot\partial_i \chi\,\mathrm{d}x \,\mathrm{d}t \quad\text{for } \chi\in L^2(I;H_0^1(U;\mathbb{R}^3)). \tag*{(133)} ∫ I ∫ U ( d × τ ) ⋅ χ d x d t = − i = 1 ∑ 3 ∫ I ∫ U ( d × ∂ i d ) ⋅ ∂ i χ d x d t for χ ∈ L 2 ( I ; H 0 1 ( U ; R 3 )) . (133)
Next, we show that the tension is tangential. The equation ∂ t d ε + ( u ε ⋅ ∇ ) d ε = τ ε \partial_t d_\varepsilon+ (u_\varepsilon\cdot\nabla)d_\varepsilon= \tau_\varepsilon ∂ t d ε + ( u ε ⋅ ∇ ) d ε = τ ε gives
∂ t ∣ d ε ∣ 2 2 + u ε ⋅ ∇ ∣ d ε ∣ 2 2 = d ε ⋅ τ ε . \partial_t \frac{|d_\varepsilon|^2}{2} + u_\varepsilon\cdot\nabla\frac{|d_\varepsilon|^2}{2} = d_\varepsilon\cdot\tau_\varepsilon. ∂ t 2 ∣ d ε ∣ 2 + u ε ⋅ ∇ 2 ∣ d ε ∣ 2 = d ε ⋅ τ ε .
Let ϕ ∈ C c ∞ ( Q F ) \phi\in C^\infty_c(Q_F) ϕ ∈ C c ∞ ( Q F ) . Since supp ϕ \operatorname{supp}\phi supp ϕ is a compact subset of Q F Q_F Q F and the solids S ε ( t ) S_\varepsilon(t) S ε ( t ) converge to S ( t ) S(t) S ( t ) uniformly in Hausdorff distance, see (123) , we have supp ϕ ⊂ Q F ε \operatorname{supp}\phi\subset Q_{F_\varepsilon} supp ϕ ⊂ Q F ε for all sufficiently small ε \varepsilon ε , and integration by parts yields
∫ 0 T ∫ Ω ϕ d ε ⋅ τ ε d x d t = − 1 2 ∫ 0 T ∫ Ω ∣ d ε ∣ 2 ( ∂ t ϕ + u ε ⋅ ∇ ϕ ) d x d t , \int_0^T \int_\Omega\phi\,d_\varepsilon\cdot\tau_\varepsilon\,\mathrm{d}x \,\mathrm{d}t = -\frac{1}{2}\int_0^T \int_\Omega|d_\varepsilon|^2(\partial_t\phi+ u_\varepsilon\cdot\nabla\phi) \,\mathrm{d}x \,\mathrm{d}t, ∫ 0 T ∫ Ω ϕ d ε ⋅ τ ε d x d t = − 2 1 ∫ 0 T ∫ Ω ∣ d ε ∣ 2 ( ∂ t ϕ + u ε ⋅ ∇ ϕ ) d x d t ,
where we used div u ε = 0 \operatorname{div}u_\varepsilon= 0 div u ε = 0 . The local strong L 2 L^2 L 2 -convergence of d ε d_\varepsilon d ε , the bound ∣ d ε ∣ ≤ 1 |d_\varepsilon| \leq1 ∣ d ε ∣ ≤ 1 , and ∣ d ∣ = 1 |d| = 1 ∣ d ∣ = 1 imply ∣ d ε ∣ 2 → 1 |d_\varepsilon|^2 \to1 ∣ d ε ∣ 2 → 1 strongly in L p ( supp ϕ ) L^p(\operatorname{supp}\phi) L p ( supp ϕ ) for every p < ∞ p < \infty p < ∞ . Together with the strong local L 2 L^2 L 2 -convergence of u ε u_\varepsilon u ε and the weak L 2 L^2 L 2 -convergence of τ ε \tau_\varepsilon τ ε , this yields
∫ 0 T ∫ Ω ϕ d ⋅ τ d x d t = − 1 2 ∫ 0 T ∫ Ω ( ∂ t ϕ + u ⋅ ∇ ϕ ) d x d t = 0 , \int_0^T \int_\Omega\phi\,d \cdot\tau\,\mathrm{d}x \,\mathrm{d}t = -\frac{1}{2}\int_0^T \int_\Omega(\partial_t\phi+ u \cdot\nabla\phi) \,\mathrm{d}x \,\mathrm{d}t = 0, ∫ 0 T ∫ Ω ϕ d ⋅ τ d x d t = − 2 1 ∫ 0 T ∫ Ω ( ∂ t ϕ + u ⋅ ∇ ϕ ) d x d t = 0 ,
because ϕ \phi ϕ has compact support in spacetime and div u = 0 \operatorname{div}u = 0 div u = 0 . This implies
d ⋅ τ = 0 a.e. in Q F . (134) d \cdot\tau= 0 \quad\text{a.e. in } Q_F. \tag*{(134)} d ⋅ τ = 0 a.e. in Q F . (134)
Finally, let ϕ ∈ C c ∞ ( I × U ; R 3 ) \phi\in C^\infty_c(I \times U;\mathbb{R}^3) ϕ ∈ C c ∞ ( I × U ; R 3 ) , and set ϕ T : = ϕ − ( d ⋅ ϕ ) d \phi_T := \phi- (d \cdot\phi)d ϕ T := ϕ − ( d ⋅ ϕ ) d and χ : = d × ϕ T = d × ϕ \chi:= d \times\phi_T = d \times\phi χ := d × ϕ T = d × ϕ . With regard to d ∈ L 2 ( I ; H 1 ( U ) ) d \in L^2(I;H^1(U)) d ∈ L 2 ( I ; H 1 ( U )) and ∣ d ∣ = 1 |d| = 1 ∣ d ∣ = 1 , both fields belong to L 2 ( I ; H 0 1 ( U ; R 3 ) ) L^2(I;H_0^1(U;\mathbb{R}^3)) L 2 ( I ; H 0 1 ( U ; R 3 )) , and no time regularity is needed. As ∣ d ∣ = 1 |d| = 1 ∣ d ∣ = 1 , d ⋅ ∂ i d = 0 d \cdot\partial_i d = 0 d ⋅ ∂ i d = 0 , d ⋅ ϕ T = 0 d \cdot\phi_T = 0 d ⋅ ϕ T = 0 , and (134) holds, we infer that a.e.
τ ⋅ ϕ = τ ⋅ ϕ T = ( d × τ ) ⋅ ( d × ϕ T ) = ( d × τ ) ⋅ χ , \tau\cdot\phi= \tau\cdot\phi_T = (d \times\tau) \cdot(d \times\phi_T) = (d \times\tau) \cdot\chi, τ ⋅ ϕ = τ ⋅ ϕ T = ( d × τ ) ⋅ ( d × ϕ T ) = ( d × τ ) ⋅ χ ,
and
( d × ∂ i d ) ⋅ ∂ i χ = ( d × ∂ i d ) ⋅ ( ∂ i d × ϕ T ) + ( d × ∂ i d ) ⋅ ( d × ∂ i ϕ T ) = ∂ i d ⋅ ∂ i ϕ T . (d \times\partial_i d) \cdot\partial_i \chi= (d \times\partial_i d) \cdot(\partial_i d \times\phi_T) + (d \times\partial_i d) \cdot(d \times\partial_i \phi_T) = \partial_i d \cdot\partial_i \phi_T. ( d × ∂ i d ) ⋅ ∂ i χ = ( d × ∂ i d ) ⋅ ( ∂ i d × ϕ T ) + ( d × ∂ i d ) ⋅ ( d × ∂ i ϕ T ) = ∂ i d ⋅ ∂ i ϕ T .
Inserting χ \chi χ into (133) therefore gives
∫ I ∫ U τ ⋅ ϕ d x d t = − ∫ I ∫ U ∇ d : ∇ ϕ T d x d t . \int_I \int_U \tau\cdot\phi\,\mathrm{d}x \,\mathrm{d}t = -\int_I \int_U \nabla d : \nabla\phi_T \,\mathrm{d}x \,\mathrm{d}t. ∫ I ∫ U τ ⋅ ϕ d x d t = − ∫ I ∫ U ∇ d : ∇ ϕ T d x d t .
Since d ⋅ ∂ i d = 0 d \cdot\partial_i d = 0 d ⋅ ∂ i d = 0 , we have ∇ d : ∇ ( ( d ⋅ ϕ ) d ) = ∣ ∇ d ∣ 2 ( d ⋅ ϕ ) \nabla d : \nabla((d \cdot\phi)d) = |\nabla d|^2(d \cdot\phi) ∇ d : ∇ (( d ⋅ ϕ ) d ) = ∣∇ d ∣ 2 ( d ⋅ ϕ ) , and consequently
∫ I ∫ U τ ⋅ ϕ d x d t = − ∫ I ∫ U ∇ d : ∇ ϕ d x d t + ∫ I ∫ U ∣ ∇ d ∣ 2 d ⋅ ϕ d x d t . (135) \int_I \int_U \tau\cdot\phi\,\mathrm{d}x \,\mathrm{d}t = -\int_I \int_U \nabla d : \nabla\phi\,\mathrm{d}x \,\mathrm{d}t + \int_I \int_U |\nabla d|^2d \cdot\phi\,\mathrm{d}x \,\mathrm{d}t. \tag*{(135)} ∫ I ∫ U τ ⋅ ϕ d x d t = − ∫ I ∫ U ∇ d : ∇ ϕ d x d t + ∫ I ∫ U ∣∇ d ∣ 2 d ⋅ ϕ d x d t . (135)
As the cylinder I × U I \times U I × U was arbitrary, τ = Δ d + ∣ ∇ d ∣ 2 d \tau= \Delta d + |\nabla d|^2d τ = Δ d + ∣∇ d ∣ 2 d in D ′ ( Q F ) \mathcal{D}'(Q_F) D ′ ( Q F ) . □ \square □
Let τ ~ ε \widetilde{\tau}_\varepsilon τ ε denote the extension of τ ε \tau_\varepsilon τ ε by zero to the solid. By (122) , a subsequence converges weakly in L 2 ( ( 0 , T ) × Ω ) L^2((0,T) \times\Omega) L 2 (( 0 , T ) × Ω ) to some τ ~ \widetilde{\tau} τ . By the uniform convergence (123) of the rigid motions, every compact subset of Q F Q_F Q F lies in Q F ε Q_{F_\varepsilon} Q F ε , and every compact subset of the open set { ( t , x ) : x ∈ S ( t ) } \{(t,x):x \in S(t)\} {( t , x ) : x ∈ S ( t )} lies in { ( t , x ) : x ∈ S ε ( t ) } \{(t,x):x \in S_\varepsilon(t)\} {( t , x ) : x ∈ S ε ( t )} , for all sufficiently small ε \varepsilon ε . Hence τ ~ = τ \widetilde{\tau} = \tau τ = τ in Q F Q_F Q F by Proposition 4.7, and τ ~ = 0 \widetilde{\tau} = 0 τ = 0 a.e. in the solid region, since τ ~ ε \widetilde{\tau}_\varepsilon τ ε vanishes there. Due to the measure zero of the interface, τ ~ \widetilde{\tau} τ is the zero extension of τ \tau τ , and the weak lower semicontinuity of the norm gives the estimate
∫ 0 T ∫ F ( t ) ∣ τ ∣ 2 d x d t ≤ lim inf ε → 0 ∫ 0 T ∫ F ε ( t ) ∣ τ ε ∣ 2 d x d t . (136) \int_0^T \int_{F(t)} |\tau|^2 \,\mathrm{d}x \,\mathrm{d}t \leq\liminf_{\varepsilon\to0} \int_0^T \int_{F_\varepsilon(t)} |\tau_\varepsilon|^2 \,\mathrm{d}x \,\mathrm{d}t. \tag*{(136)} ∫ 0 T ∫ F ( t ) ∣ τ ∣ 2 d x d t ≤ ε → 0 lim inf ∫ 0 T ∫ F ε ( t ) ∣ τ ε ∣ 2 d x d t . (136)
Anchoring and passage to the weak equations
In this subsection, we show that the limiting d d d satisfies the anchoring condition (16) with d 0 d_0 d 0 , and we pass to the limit in the director and fluid equations.
Lemma 4.8. The limiting director field d d d satisfies the anchoring condition (16) with the original H 1 H^1 H 1 -datum d 0 d_0 d 0 .
Proof. Let r 0 > 0 r_0>0 r 0 > 0 be such that the nearest-point projections onto ∂ Ω \partial\Omega ∂ Ω and ∂ S 0 \partial S_0 ∂ S 0 are well defined and of class C 1 C^1 C 1 on the r 0 r_0 r 0 -neighborhoods of these surfaces. The existence of such an r 0 r_0 r 0 is guaranteed by the compactness and C 2 C^2 C 2 -regularity of both surfaces. We now choose 0 < r < r 0 0<r<r_0 0 < r < r 0 smaller than one quarter of the uniform separation ρ T \rho_T ρ T in (121) , establishing that U o u t U_{\mathrm{out}} U out and U i n 0 U_{\mathrm{in}}^0 U in 0 defined below are Lipschitz domains, and thereby showing that the trace operators onto ∂ Ω \partial\Omega ∂ Ω and ∂ S 0 \partial S_0 ∂ S 0 are continuous on H 1 H^1 H 1 of U o u t U_{\mathrm{out}} U out and U i n 0 U_{\mathrm{in}}^0 U in 0 . Another consequence is that U o u t U_{\mathrm{out}} U out does not meet any of the solids S ε ( t ) S_\varepsilon(t) S ε ( t ) .
On U o u t = { x ∈ Ω : dist ( x , ∂ Ω ) < r } U_{\mathrm{out}}=\{x\in\Omega:\operatorname{dist}(x,\partial\Omega)<r\} U out = { x ∈ Ω : dist ( x , ∂ Ω ) < r } , the material extensions d ˉ ε \bar d_\varepsilon d ˉ ε coincide with the fluid directors. The convergences from (126) and the uniform L t ∞ H x 1 L^\infty_tH^1_x L t ∞ H x 1 -bound imply that the restrictions of d ˉ ε \bar d_\varepsilon d ˉ ε converge strongly in L t , x 2 L^2_{t,x} L t , x 2 and weakly in L t 2 H x 1 L^2_tH^1_x L t 2 H x 1 . The corresponding outer traces are d 0 ε d_0^\varepsilon d 0 ε , and they converge to d 0 d_0 d 0 in H 1 / 2 ( ∂ Ω ) H^{1/2}(\partial\Omega) H 1/2 ( ∂ Ω ) by (120) . The outer anchoring condition then follows from the continuity of the trace operator.
For the inner boundary, we introduce U i n 0 = { y ∈ F 0 : dist ( y , ∂ S 0 ) < r } U_{\mathrm{in}}^0=\{y\in\mathcal F_0:\operatorname{dist}(y,\partial S_0)<r\} U in 0 = { y ∈ F 0 : dist ( y , ∂ S 0 ) < r } along with the rigid maps Z ε ( t , y ) = h ε ( t ) + O ε ( t ) y Z_\varepsilon(t,y)=h_\varepsilon(t)+O_\varepsilon(t)y Z ε ( t , y ) = h ε ( t ) + O ε ( t ) y . The pullbacks d ^ ε ( t , y ) : = d ε ( t , Z ε ( t , y ) ) \widehat d_\varepsilon(t,y):=d_\varepsilon(t,Z_\varepsilon(t,y)) d ε ( t , y ) := d ε ( t , Z ε ( t , y )) are uniformly bounded in L ∞ ( 0 , T ; H 1 ( U i n 0 ) ) L^\infty(0,T;H^1(U_{\mathrm{in}}^0)) L ∞ ( 0 , T ; H 1 ( U in 0 )) . From (126) and (123) , we deduce that d ^ ε → d ( t , h ( t ) + O ( t ) y ) \widehat d_\varepsilon\to d(t,h(t)+O(t)y) d ε → d ( t , h ( t ) + O ( t ) y ) strongly in L 2 ( ( 0 , T ) × U i n 0 ) L^2((0,T)\times U_{\mathrm{in}}^0) L 2 (( 0 , T ) × U in 0 ) . The weak L t 2 H y 1 L^2_tH^1_y L t 2 H y 1 -limit is thus uniquely the indicated pullback of d d d . On ∂ S 0 \partial S_0 ∂ S 0 the pulled-back traces equal d 0 ε d_0^\varepsilon d 0 ε , so their weak trace limit is d 0 d_0 d 0 . Invoking the push-forward of this identity by the limiting rigid map, we obtain the inner anchoring in (16) . □
Let now ζ ∈ T u c o l \zeta\in T_u^{\mathrm{col}} ζ ∈ T u col , where we recall the test space T u c o l T_u^{\mathrm{col}} T u col from (119) . Its symmetric gradient vanishes in a fixed collar of S ( t ) S(t) S ( t ) , which contains S ε ( t ) S_\varepsilon(t) S ε ( t ) for all sufficiently small ε \varepsilon ε . By symmetry of the Ericksen tensor, its contraction with ∇ ζ \nabla\zeta ∇ ζ vanishes in that collar. On the complement, local strong H 1 H^1 H 1 -convergence from Proposition 4.7 allows for the limit passage in the stress. Note that this convergence does not cover a neighborhood of t = 0 t=0 t = 0 . However, the bound for ∇ d ε \nabla d_\varepsilon ∇ d ε in L ∞ ( 0 , T ; L 2 ( F ε ( t ) ) ) L^\infty(0,T;L^2(\mathcal F_\varepsilon(t))) L ∞ ( 0 , T ; L 2 ( F ε ( t ))) from (122) yields that the contribution of the time layer ( 0 , t 0 ) (0,t_0) ( 0 , t 0 ) to the stress term is at most C t 0 ∥ ∇ ζ ∥ L ∞ ( ( 0 , T ) × Ω ) Ct_0\|\nabla\zeta\|_{L^\infty((0,T)\times\Omega)} C t 0 ∥∇ ζ ∥ L ∞ (( 0 , T ) × Ω ) , uniformly in ε \varepsilon ε . The same bound holds for the limit. Thus, letting first ε → 0 \varepsilon\to0 ε → 0 and then t 0 → 0 t_0\to0 t 0 → 0 gives the momentum identity for every collar-rigid test. Proposition A.3 then extends the identity to every ζ ∈ T u \zeta\in T_u ζ ∈ T u , proving (17) .
For the director, we first consider ψ ∈ T d \psi\in T_d ψ ∈ T d , with T d T_d T d as introduced in (15) . For all sufficiently small ε \varepsilon ε , it is supported in the approximate fluid region, and the molecular-field equation gives
∫ 0 T ∫ F ε ( t ) [ − d ε ⋅ ∂ t ψ + ( u ε ⋅ ∇ ) d ε ⋅ ψ − τ ε ⋅ ψ ] d x d t = ∫ F 0 d 0 ε ⋅ ψ ( 0 ) d x . \int_0^T\int_{\mathcal F_\varepsilon(t)}\left[-d_\varepsilon\cdot\partial_t\psi+(u_\varepsilon\cdot\nabla)d_\varepsilon\cdot\psi-\tau_\varepsilon\cdot\psi\right]\,dx\,dt=\int_{\mathcal F_0}d_0^\varepsilon\cdot\psi(0)\,dx. ∫ 0 T ∫ F ε ( t ) [ − d ε ⋅ ∂ t ψ + ( u ε ⋅ ∇ ) d ε ⋅ ψ − τ ε ⋅ ψ ] d x d t = ∫ F 0 d 0 ε ⋅ ψ ( 0 ) d x .
On every interval [ t 0 , T ] [t_0,T] [ t 0 , T ] , the limit follows from (125) and Proposition 4.7, while on ( 0 , t 0 ) (0,t_0) ( 0 , t 0 ) , we integrate the convection term by parts in space and obtain
∫ 0 t 0 ∫ F ε ( t ) ( u ε ⋅ ∇ ) d ε ⋅ ψ d x d t = − ∫ 0 t 0 ∫ F ε ( t ) ( u ε ⋅ ∇ ) ψ ⋅ d ε d x d t . \int_0^{t_0}\int_{\mathcal F_\varepsilon(t)}(u_\varepsilon\cdot\nabla)d_\varepsilon\cdot\psi\,dx\,dt=-\int_0^{t_0}\int_{\mathcal F_\varepsilon(t)}(u_\varepsilon\cdot\nabla)\psi\cdot d_\varepsilon\,dx\,dt. ∫ 0 t 0 ∫ F ε ( t ) ( u ε ⋅ ∇ ) d ε ⋅ ψ d x d t = − ∫ 0 t 0 ∫ F ε ( t ) ( u ε ⋅ ∇ ) ψ ⋅ d ε d x d t .
Thus, the bound ∣ d ε ∣ ≤ 1 |d_\varepsilon|\le1 ∣ d ε ∣ ≤ 1 from Lemma 4.2 and the bound for u ε u_\varepsilon u ε in L ∞ ( 0 , T ; L 2 ( Ω ) ) L^\infty(0,T;L^2(\Omega)) L ∞ ( 0 , T ; L 2 ( Ω )) from (122) make it O ( t 0 ) O(t_0) O ( t 0 ) . By the Cauchy–Schwarz inequality in time and the bound for τ ε \tau_\varepsilon τ ε in (122) , the molecular-field term is O ( t 0 1 / 2 ) O(t_0^{1/2}) O ( t 0 1/2 ) , and the term containing ∂ t ψ \partial_t\psi ∂ t ψ is O ( t 0 ) O(t_0) O ( t 0 ) since ∣ d ε ∣ ≤ 1 |d_\varepsilon|\le1 ∣ d ε ∣ ≤ 1 . Letting first ε → 0 \varepsilon\to0 ε → 0 and then t 0 → 0 t_0\to0 t 0 → 0 , and using (120) , yields
∫ 0 T ∫ F ( t ) [ − d ⋅ ∂ t ψ + ( u ⋅ ∇ ) d ⋅ ψ − τ ⋅ ψ ] d x d t = ∫ F 0 d 0 ⋅ ψ ( 0 ) d x . (137) \int_0^T\int_{\mathcal F(t)}\left[-d\cdot\partial_t\psi+(u\cdot\nabla)d\cdot\psi-\tau\cdot\psi\right]\,dx\,dt=\int_{\mathcal F_0}d_0\cdot\psi(0)\,dx. \tag*{(137)} ∫ 0 T ∫ F ( t ) [ − d ⋅ ∂ t ψ + ( u ⋅ ∇ ) d ⋅ ψ − τ ⋅ ψ ] d x d t = ∫ F 0 d 0 ⋅ ψ ( 0 ) d x . (137)
The functional in (137) is continuous on X d ( φ ) \mathcal{X}_d(\varphi) X d ( φ ) , since the only nontrivial term ( u ⋅ ∇ ) d ⋅ ψ (u\cdot\nabla)d\cdot\psi ( u ⋅ ∇ ) d ⋅ ψ is bounded by ∥ u ∥ L t 2 L x 6 ∥ ∇ d ∥ L t ∞ L x 2 ∥ ψ ∥ L t 2 L x 3 \|u\|_{L^2_tL^6_x}\|\nabla d\|_{L^\infty_tL^2_x}\|\psi\|_{L^2_tL^3_x} ∥ u ∥ L t 2 L x 6 ∥∇ d ∥ L t ∞ L x 2 ∥ ψ ∥ L t 2 L x 3 . Lemma 3.13 then extends (137) to every ψ ∈ X d ( φ ) \psi\in\mathcal{X}_d(\varphi) ψ ∈ X d ( φ ) , proving (18) in its stronger tension form. Finally, (124) and (125) give (19) . Exact rigidity passes to the limit as before.
Energy inequality and the quantitative interval
We pass to the energy inequality along the same subsequence. By (125) and (126) , after a further subsequence,
u ‾ ε ( t ) → u ‾ ( t ) strongly in L 2 ( Ω ) , d ‾ ε ( t ) → d ‾ ( t ) strongly in L 2 ( Ω ) \overline{u}_{\varepsilon}(t)\to\overline{u}(t)\quad\text{strongly in }L^2(\Omega),\qquad\overline{d}_{\varepsilon}(t)\to\overline{d}(t)\quad\text{strongly in }L^2(\Omega) u ε ( t ) → u ( t ) strongly in L 2 ( Ω ) , d ε ( t ) → d ( t ) strongly in L 2 ( Ω )
for a.e. t ∈ ( 0 , T ) t\in(0,T) t ∈ ( 0 , T ) . Fix such a time t t t at which, in addition, the energy inequality (40) holds for every ε \varepsilon ε in the sequence and ∥ d ‾ ε ( t ) ∥ H 1 ( Ω ) ≤ C T \|\overline{d}_{\varepsilon}(t)\|_{H^1(\Omega)}\le C_T ∥ d ε ( t ) ∥ H 1 ( Ω ) ≤ C T . Since the sequence is countable, a.e. t ∈ ( 0 , T ) t\in(0,T) t ∈ ( 0 , T ) has these properties. Choose a subsequence ε k → 0 \varepsilon_k\to0 ε k → 0 , depending on t t t , such that
lim k → ∞ ∫ Ω ∣ ∇ d ‾ ε k ( t ) ∣ 2 d x = lim inf ε → 0 ∫ Ω ∣ ∇ d ‾ ε ( t ) ∣ 2 d x . \lim_{k\to\infty}\int_{\Omega}|\nabla\overline{d}_{\varepsilon_k}(t)|^2\,dx=\liminf_{\varepsilon\to0}\int_{\Omega}|\nabla\overline{d}_{\varepsilon}(t)|^2\,dx. k → ∞ lim ∫ Ω ∣∇ d ε k ( t ) ∣ 2 d x = ε → 0 lim inf ∫ Ω ∣∇ d ε ( t ) ∣ 2 d x .
Due to the boundedness of ( d ‾ ε k ( t ) ) k (\overline{d}_{\varepsilon_k}(t))_k ( d ε k ( t ) ) k in H 1 ( Ω ) H^1(\Omega) H 1 ( Ω ) , a further subsequence converges weakly in H 1 ( Ω ) H^1(\Omega) H 1 ( Ω ) , and its limit is d ‾ ( t ) \overline{d}(t) d ( t ) thanks to the strong L 2 L^2 L 2 -convergence. By the weak lower semicontinuity of the norm, we thus have
∫ Ω ∣ ∇ d ‾ ( t ) ∣ 2 d x ≤ lim inf ε → 0 ∫ Ω ∣ ∇ d ‾ ε ( t ) ∣ 2 d x . \int_{\Omega}|\nabla\overline{d}(t)|^2\,dx\le\liminf_{\varepsilon\to0}\int_{\Omega}|\nabla\overline{d}_{\varepsilon}(t)|^2\,dx. ∫ Ω ∣∇ d ( t ) ∣ 2 d x ≤ ε → 0 lim inf ∫ Ω ∣∇ d ε ( t ) ∣ 2 d x .
Invoking the change of variables x = h ε ( t ) + O ε ( t ) y x=h_{\varepsilon}(t)+O_{\varepsilon}(t)y x = h ε ( t ) + O ε ( t ) y and (120) , we find that
∫ S ε ( t ) ∣ ∇ d ‾ ε ( t ) ∣ 2 d x = ∫ S 0 ∣ ∇ d ‾ 0 ∣ 2 d x → ∫ S 0 ∣ ∇ d ‾ 0 ∣ 2 d x = ∫ S ( t ) ∣ ∇ d ‾ ( t ) ∣ 2 d x . \int_{S_{\varepsilon}(t)}|\nabla\overline{d}_{\varepsilon}(t)|^2\,dx=\int_{S_0}|\nabla\overline{d}_0|^2\,dx\to\int_{S_0}|\nabla\overline{d}_0|^2\,dx=\int_{S(t)}|\nabla\overline{d}(t)|^2\,dx. ∫ S ε ( t ) ∣∇ d ε ( t ) ∣ 2 d x = ∫ S 0 ∣∇ d 0 ∣ 2 d x → ∫ S 0 ∣∇ d 0 ∣ 2 d x = ∫ S ( t ) ∣∇ d ( t ) ∣ 2 d x .
Subtracting this gives, for a.e. t t t ,
∫ F ( t ) ∣ ∇ d ‾ ( t ) ∣ 2 d x ≤ lim inf ε → 0 ∫ F ε ( t ) ∣ ∇ d ‾ ε ( t ) ∣ 2 d x . \int_{\mathcal{F}(t)}|\nabla\overline{d}(t)|^2\,dx\le\liminf_{\varepsilon\to0}\int_{\mathcal{F}_{\varepsilon}(t)}|\nabla\overline{d}_{\varepsilon}(t)|^2\,dx. ∫ F ( t ) ∣∇ d ( t ) ∣ 2 d x ≤ ε → 0 lim inf ∫ F ε ( t ) ∣∇ d ε ( t ) ∣ 2 d x .
The kinetic energy converges at the same a.e. time slices. Weak lower semicontinuity of the viscous dissipation and (136) , together with the nonnegative Ginzburg–Landau potential and the convergence of the initial energy E 0 ε → E 0 E_0^\varepsilon\to E_0 E 0 ε → E 0 , now yields (20) for a.e. t ∈ ( 0 , T ) t\in(0,T) t ∈ ( 0 , T ) .
The preceding arguments yield a constrained weak solution on [ 0 , T ] [0,T] [ 0 , T ] for every T < T ∗ , T<T_*, T < T ∗ , obtained along a subsequence that may depend on T T T . In order to obtain a single solution on [ 0 , T ∗ ) [0,T_*) [ 0 , T ∗ ) , we use a diagonal argument. To this end, for each ε \varepsilon ε , let ( u ε , d ε , φ ε ) (u_\varepsilon,d_\varepsilon,\varphi_\varepsilon) ( u ε , d ε , φ ε ) be the Ginzburg–Landau solution from Theorem 3.2 with anchor d 0 ε d_0^\varepsilon d 0 ε . This solution exists on [ 0 , T ∗ ε ] [0,T_*^\varepsilon] [ 0 , T ∗ ε ] , where T ∗ ε T_*^\varepsilon T ∗ ε is given by (35) with E 0 ε E_0^\varepsilon E 0 ε in place of E 0 E_0 E 0 , and T ∗ ε → T ∗ T_*^\varepsilon\to T_* T ∗ ε → T ∗ because E 0 ε → E 0 E_0^\varepsilon\to E_0 E 0 ε → E 0 . Choose T m ↑ T ∗ T_m\uparrow T_* T m ↑ T ∗ and a sequence ε j → 0 \varepsilon_j\to0 ε j → 0 . The arguments of this section, applied on [ 0 , T 1 ] [0,T_1] [ 0 , T 1 ] , give a subsequence along which the Ginzburg–Landau solutions converge to a constrained weak solution on [ 0 , T 1 ] [0,T_1] [ 0 , T 1 ] . Applied to this subsequence on [ 0 , T 2 ] [0,T_2] [ 0 , T 2 ] , they give a further subsequence, and so on. Along the diagonal subsequence, the Ginzburg–Landau solutions converge on [ 0 , T m ] [0,T_m] [ 0 , T m ] for every m m m , and the limits on different intervals coincide on the smaller one, since they are limits of the same sequence. Thus, they give rise to a single triple ( u , d , φ ) (u,d,\varphi) ( u , d , φ ) on [ 0 , T ∗ ) [0,T_*) [ 0 , T ∗ ) that is a constrained weak solution on [ 0 , T ] [0,T] [ 0 , T ] for every T < T ∗ T<T_* T < T ∗ . By (20) , ( u , d , φ ) (u,d,\varphi) ( u , d , φ ) satisfies (32) on each such interval, so dist ( S ( t ) , ∂ Ω ) ≥ δ 0 / 2 \operatorname{dist}(S(t),\partial\Omega)\ge\delta_0/2 dist ( S ( t ) , ∂ Ω ) ≥ δ 0 /2 for 0 ≤ t < T ∗ 0\le t<T_* 0 ≤ t < T ∗ by Corollary 2.9 , completing the proof of Proposition 4.1 .
The following section is concerned with the continuation of this solution up to the first contact time in § 5.1 and then after contact in § 5.2 .
Continuation of the solution
In this section, we discuss the continuation of the solution after the limit passage in the Ginzburg–Landau regularization, first before possible contact with the outer fluid boundary, see § 5.1 , and then after contact, see § 5.2 .
Pre-contact continuation of the constrained solution
We iterate the local constrained construction. The lower bound for the next collision-free interval depends only on the current separation and energy, not on an H 2 H^2 H 2 -norm of the current director. Thus the loss of regularity in the limit ε → 0 \varepsilon\to0 ε → 0 does not prevent restarting.
Restart of the constrained solution requires collision-free endpoint traces and a gluing property.
Lemma 5.1. Let ( u , d , φ ) (\boldsymbol{u},\boldsymbol{d},\varphi) ( u , d , φ ) be a constrained weak solution on [ 0 , T 0 ) [0,T_0) [ 0 , T 0 ) , that is, a weak solution on [ 0 , T ] [0,T] [ 0 , T ] in the sense of Definition 2.2 for every T < T 0 T<T_0 T < T 0 , and assume
inf t 1 < t < T 0 dist ( S ( t ) , ∂ Ω ) > 0 \inf_{t_1<t<T_0} \operatorname{dist}(\mathcal{S}(t),\partial\Omega)>0 t 1 < t < T 0 inf dist ( S ( t ) , ∂ Ω ) > 0 for some t 1 < T 0 t_1<T_0 t 1 < T 0 . Then the rigid motion has a limit S ∗ : = h ∗ + O ∗ S 0 ⋐ Ω \mathcal{S}_*:=\boldsymbol{h}_*+O_*\mathcal{S}_0\Subset\Omega S ∗ := h ∗ + O ∗ S 0 ⋐ Ω as t → T 0 t\to T_0 t → T 0 . There exist
u ∗ ∈ H S ∗ ( Ω ) , d ∗ ∈ H 1 ( Ω ∖ S ‾ ∗ ; S + 2 ) , \boldsymbol{u}_*\in\mathcal{H}_{\mathcal{S}_*}(\Omega),\qquad\boldsymbol{d}_*\in H^1(\Omega\setminus\overline{\mathcal{S}}_*;\mathbb{S}^2_+), u ∗ ∈ H S ∗ ( Ω ) , d ∗ ∈ H 1 ( Ω ∖ S ∗ ; S + 2 ) , satisfying the anchoring conditions
Tr ∂ Ω d ∗ = d 0 , Tr ∂ S ∗ d ∗ ( x ) = d 0 ( O ∗ T ( x − h ∗ ) ) , (138) \operatorname{Tr}_{\partial\Omega}\boldsymbol{d}_*=\boldsymbol{d}_0,\qquad\operatorname{Tr}_{\partial\mathcal{S}_*}\boldsymbol{d}_*(x)=\boldsymbol{d}_0\left(O_*^{\mathsf T}(x-\boldsymbol{h}_*)\right), \tag*{(138)} Tr ∂ Ω d ∗ = d 0 , Tr ∂ S ∗ d ∗ ( x ) = d 0 ( O ∗ T ( x − h ∗ ) ) , (138) that is, (16) for the configuration ( h ∗ , O ∗ ) (\boldsymbol{h}_*,O_*) ( h ∗ , O ∗ ) , and a sequence chosen from the common full-measure set of admissible times t j → T 0 t_j\to T_0 t j → T 0 such that
u ( t j ) ⇀ u ∗ in L 2 ( Ω ) , d ‾ ( t ) → d ‾ ∗ strongly in L 2 ( Ω ) as t → T 0 , \boldsymbol{u}(t_j)\rightharpoonup\boldsymbol{u}_*\quad\text{in }L^2(\Omega),\qquad\overline{\boldsymbol{d}}(t)\to\overline{\boldsymbol{d}}_*\quad\text{strongly in }L^2(\Omega)\text{ as }t\to T_0, u ( t j ) ⇀ u ∗ in L 2 ( Ω ) , d ( t ) → d ∗ strongly in L 2 ( Ω ) as t → T 0 , where d ‾ \overline{\boldsymbol{d}} d denotes the material extension through the solid and d ‾ ∗ ∣ Ω ∖ S ‾ ∗ = d ∗ \overline{\boldsymbol{d}}_*|_{\Omega\setminus\overline{\mathcal{S}}_*}=\boldsymbol{d}_* d ∗ ∣ Ω ∖ S ∗ = d ∗ . Moreover
1 2 ∥ u ∗ ∥ L 2 ( Ω ) 2 + 1 2 ∥ ∇ d ∗ ∥ L 2 ( Ω ∖ S ∗ ) 2 ≤ lim inf j → ∞ E ( t j ) . (139) \frac{1}{2}\|\boldsymbol{u}_*\|_{L^2(\Omega)}^2+\frac{1}{2}\|\nabla\boldsymbol{d}_*\|_{L^2(\Omega\setminus\mathcal{S}_*)}^2\leq\liminf_{j\to\infty}E(t_j). \tag*{(139)} 2 1 ∥ u ∗ ∥ L 2 ( Ω ) 2 + 2 1 ∥∇ d ∗ ∥ L 2 ( Ω ∖ S ∗ ) 2 ≤ j → ∞ lim inf E ( t j ) . (139) For every ζ ∈ C ∞ ( Ω ; R 3 ) \boldsymbol{\zeta}\in C^\infty(\Omega;\mathbb{R}^3) ζ ∈ C ∞ ( Ω ; R 3 ) which is divergence free and rigid in a fixed positive neighborhood of S ∗ \mathcal{S}_* S ∗ ,
lim t → T 0 ∫ Ω u ( t ) ⋅ ζ d x = ∫ Ω u ∗ ⋅ ζ d x . (140) \lim_{t\to T_0}\int_\Omega\boldsymbol{u}(t)\cdot\boldsymbol{\zeta}\,\mathrm{d}x=\int_\Omega\boldsymbol{u}_*\cdot\boldsymbol{\zeta}\,\mathrm{d}x. \tag*{(140)} t → T 0 lim ∫ Ω u ( t ) ⋅ ζ d x = ∫ Ω u ∗ ⋅ ζ d x . (140) Finally, suppose that ( u , d , φ ) (\boldsymbol{u},\boldsymbol{d},\varphi) ( u , d , φ ) is continued beyond T 0 T_0 T 0 by a constrained weak solution on [ T 0 , T 1 ] [T_0,T_1] [ T 0 , T 1 ] with initial state ( u ∗ , d ∗ , S ∗ ) (\boldsymbol{u}_*,\boldsymbol{d}_*,\mathcal{S}_*) ( u ∗ , d ∗ , S ∗ ) and initial energy
E ∗ : = 1 2 ∥ u ∗ ∥ L 2 ( Ω ) 2 + 1 2 ∥ ∇ d ∗ ∥ L 2 ( Ω ∖ S ‾ ∗ ) 2 . E_*:=\frac{1}{2}\|\boldsymbol{u}_*\|_{L^2(\Omega)}^2+\frac{1}{2}\|\nabla\boldsymbol{d}_*\|_{L^2(\Omega\setminus\overline{\mathcal{S}}_*)}^2. E ∗ := 2 1 ∥ u ∗ ∥ L 2 ( Ω ) 2 + 2 1 ∥∇ d ∗ ∥ L 2 ( Ω ∖ S ∗ ) 2 . Then the concatenation is a constrained weak solution on [ 0 , T 1 ] [0,T_1] [ 0 , T 1 ] that satisfies the energy inequality (20) with the original initial energy E 0 E_0 E 0 .
Proof. The energy identity for the rigid part and Lemma 2.7 give uniform bounds for h ′ \boldsymbol{h}' h ′ and ω \boldsymbol{\omega} ω , so h ( t ) → h ∗ \boldsymbol{h}(t)\to\boldsymbol{h}_* h ( t ) → h ∗ and O ( t ) → O ∗ O(t)\to O_* O ( t ) → O ∗ as t → T 0 t\to T_0 t → T 0 . Next, we define the extension of d \boldsymbol{d} d on the whole container by
d ‾ ( t , x ) : = { d ( t , x ) , for x ∈ F ( t ) , d 0 ( O ( t ) T ( x − h ( t ) ) ) , for x ∈ S ( t ) . \overline{\boldsymbol{d}}(t,x):=
\begin{cases}
\boldsymbol{d}(t,x), & \text{for }x\in\mathcal{F}(t),\\
\boldsymbol{d}_0\left(O(t)^{\mathsf T}(x-\boldsymbol{h}(t))\right), & \text{for }x\in\mathcal{S}(t).
\end{cases} d ( t , x ) := { d ( t , x ) , d 0 ( O ( t ) T ( x − h ( t )) ) , for x ∈ F ( t ) , for x ∈ S ( t ) .
Moreover, we denote by τ ~ \widetilde{\boldsymbol{\tau}} τ the zero extension of τ ∈ L 2 ( Q F ) \boldsymbol{\tau}\in L^2(Q_{\mathcal{F}}) τ ∈ L 2 ( Q F ) . From Lemma A.4 , it follows that
∂ t d ‾ + u ⋅ ∇ d ‾ = τ ~ in D ′ ( ( t 1 , T 0 ) × Ω ) . \partial_t\overline{\boldsymbol{d}}+\boldsymbol{u}\cdot\nabla\overline{\boldsymbol{d}}=\widetilde{\boldsymbol{\tau}}\quad\text{in }\mathcal{D}'((t_1,T_0)\times\Omega). ∂ t d + u ⋅ ∇ d = τ in D ′ (( t 1 , T 0 ) × Ω ) .
Now we set q : = d ‾ − d 0 \boldsymbol{q}:=\overline{\boldsymbol{d}}-\boldsymbol{d}_0 q := d − d 0 . The energy inequality (20) and the fact that the solid part of d ‾ \overline{\boldsymbol{d}} d is a rigid image of d 0 ∣ S 0 \boldsymbol{d}_0|_{\mathcal{S}_0} d 0 ∣ S 0 yield that the extension d ‾ \overline{\boldsymbol{d}} d is bounded in L ∞ ( t 1 , T 0 ; H 1 ( Ω ) ) L^\infty(t_1,T_0;H^1(\Omega)) L ∞ ( t 1 , T 0 ; H 1 ( Ω )) . Its trace on ∂ Ω \partial\Omega ∂ Ω equals that of d 0 \boldsymbol{d}_0 d 0 by the outer anchoring condition in (16) . Furthermore, ∂ t d ‾ ∈ L 2 ( t 1 , T 0 ; H − 1 ( Ω ) ) \partial_t\overline{\boldsymbol{d}}\in L^2(t_1,T_0;H^{-1}(\Omega)) ∂ t d ∈ L 2 ( t 1 , T 0 ; H − 1 ( Ω )) by the last assertion of Lemma A.4 . Since d 0 \boldsymbol{d}_0 d 0 does not depend on time, we obtain
q ∈ L 2 ( t 1 , T 0 ; H 0 1 ( Ω ) ) , ∂ t q ∈ L 2 ( t 1 , T 0 ; H − 1 ( Ω ) ) . \boldsymbol{q}\in L^2(t_1,T_0;H^1_0(\Omega)),\qquad\partial_t\boldsymbol{q}\in L^2(t_1,T_0;H^{-1}(\Omega)). q ∈ L 2 ( t 1 , T 0 ; H 0 1 ( Ω )) , ∂ t q ∈ L 2 ( t 1 , T 0 ; H − 1 ( Ω )) .
By the Lions–Magenes lemma for the triple H 0 1 ( Ω ) ↪ L 2 ( Ω ) ↪ H − 1 ( Ω ) H^1_0(\Omega)\hookrightarrow L^2(\Omega)\hookrightarrow H^{-1}(\Omega) H 0 1 ( Ω ) ↪ L 2 ( Ω ) ↪ H − 1 ( Ω ) , the function q \boldsymbol{q} q , and hence d ‾ \overline{\boldsymbol{d}} d , belongs to C ( [ t 1 , T 0 ] ; L 2 ( Ω ) ) C([t_1,T_0];L^2(\Omega)) C ([ t 1 , T 0 ] ; L 2 ( Ω )) . In particular, d ‾ ( t ) → d ‾ ∗ \overline{\boldsymbol{d}}(t)\to\overline{\boldsymbol{d}}_* d ( t ) → d ∗ strongly in L 2 ( Ω ) L^2(\Omega) L 2 ( Ω ) as t → T 0 t\to T_0 t → T 0 . By the boundedness of d ‾ ( t ) \overline{\boldsymbol{d}}(t) d ( t ) in H 1 ( Ω ) H^1(\Omega) H 1 ( Ω ) , this convergence carries over to weak convergence in H 1 ( Ω ) H^1(\Omega) H 1 ( Ω ) , and d ‾ ∗ ∈ H 1 ( Ω ) \overline{\boldsymbol{d}}_*\in H^1(\Omega) d ∗ ∈ H 1 ( Ω ) . Thanks to h ( t ) → h ∗ \boldsymbol{h}(t)\to\boldsymbol{h}_* h ( t ) → h ∗ and O ( t ) → O ∗ O(t)\to O_* O ( t ) → O ∗ , the part of d ‾ ( t ) \overline{\boldsymbol{d}}(t) d ( t ) defined on S ( t ) \mathcal{S}(t) S ( t ) converges as well, and a.e. in S ∗ \mathcal{S}_* S ∗ , we get d ‾ ∗ ( x ) = d 0 ( O ∗ T ( x − h ∗ ) ) \overline{\boldsymbol{d}}_*(x)=\boldsymbol{d}_0\left(O_*^{\mathsf T}(x-\boldsymbol{h}_*)\right) d ∗ ( x ) = d 0 ( O ∗ T ( x − h ∗ ) ) . This reveals that d ∗ : = d ‾ ∗ ∣ Ω ∖ S ‾ ∗ \boldsymbol{d}_*:=\overline{\boldsymbol{d}}_*|_{\Omega\setminus\overline{\mathcal{S}}_*} d ∗ := d ∗ ∣ Ω ∖ S ∗ satisfies (138) . The corresponding anchoring condition on the rigid body interface holds because d ˉ ∗ ∈ H 1 ( Ω ) \bar{d}_* \in H^1(\Omega) d ˉ ∗ ∈ H 1 ( Ω ) has the same trace on ∂ S ∗ \partial S_* ∂ S ∗ from both sides, and the anchoring condition on the outer boundary holds because d ˉ ∗ − d 0 ∈ H 0 1 ( Ω ) \bar{d}_* - d_0 \in H^1_0(\Omega) d ˉ ∗ − d 0 ∈ H 0 1 ( Ω ) . The strong L 2 L^2 L 2 -convergence allows us to infer that ∣ d ∗ ∣ = 1 |d_*| = 1 ∣ d ∗ ∣ = 1 and ( d ∗ ) 3 ≥ 0 (d_*)_3 \ge0 ( d ∗ ) 3 ≥ 0 . Finally, the gradient norm of the part of d ˉ \bar{d} d ˉ defined on the solid domain S ( t ) S(t) S ( t ) does not depend on time, i.e.,
∫ S ( t ) ∣ ∇ d ˉ ( t ) ∣ 2 d x = ∫ S 0 ∣ ∇ d 0 ∣ 2 d x = ∫ S ∗ ∣ ∇ d ˉ ∗ ∣ 2 d x . \int_{S(t)} |\nabla\bar{d}(t)|^2 \, \mathrm{d}x = \int_{S_0} |\nabla d_0|^2 \, \mathrm{d}x = \int_{S_*} |\nabla\bar{d}_*|^2 \, \mathrm{d}x. ∫ S ( t ) ∣∇ d ˉ ( t ) ∣ 2 d x = ∫ S 0 ∣∇ d 0 ∣ 2 d x = ∫ S ∗ ∣∇ d ˉ ∗ ∣ 2 d x .
Weak lower semicontinuity of the H 1 ( Ω ) H^1(\Omega) H 1 ( Ω ) -norm thus gives
∥ ∇ d ∗ ∥ L 2 ( Ω ∖ S ∗ ) 2 ≤ lim inf t → T 0 ∥ ∇ d ( t ) ∥ L 2 ( F ( t ) ) 2 , \|\nabla d_*\|^2_{L^2(\Omega\setminus S_*)} \le\liminf_{t\to T_0} \|\nabla d(t)\|^2_{L^2(\mathcal{F}(t))}, ∥∇ d ∗ ∥ L 2 ( Ω ∖ S ∗ ) 2 ≤ t → T 0 lim inf ∥∇ d ( t ) ∥ L 2 ( F ( t )) 2 ,
which is the director part of the energy inequality (139) .
Let E ( t ) E(t) E ( t ) be the energy on the left-hand side of (20) . We then choose times t j → T 0 t_j \to T_0 t j → T 0 , in the set with full measure on which the energy inequality holds and u ( t j ) u(t_j) u ( t j ) is rigid on S ( t j ) S(t_j) S ( t j ) , such that E ( t j ) → ess lim inf t → T 0 E ( t ) E(t_j) \to\operatorname*{ess\,lim\,inf}_{t\to T_0} E(t) E ( t j ) → ess lim inf t → T 0 E ( t ) . Upon extracting, the boundedness of the sequence u ( t j ) u(t_j) u ( t j ) in H ( Ω ) \mathcal{H}(\Omega) H ( Ω ) allows us to derive its weak convergence in L 2 ( Ω ) L^2(\Omega) L 2 ( Ω ) to some u ∗ ∈ H ( Ω ) u_* \in\mathcal{H}(\Omega) u ∗ ∈ H ( Ω ) . On the set S ( t j ) S(t_j) S ( t j ) , we write
u ( t j , x ) = ℓ j + ω j × ( x − h ( t j ) ) . u(t_j,x) = \ell_j + \omega_j \times(x-h(t_j)). u ( t j , x ) = ℓ j + ω j × ( x − h ( t j )) .
The rigid kinetic-energy identity bounds the rigid velocities ( ℓ j , ω j ) (\ell_j,\omega_j) ( ℓ j , ω j ) . A further extraction yields their convergence to ( ℓ ∗ , ω ∗ ) (\ell_*,\omega_*) ( ℓ ∗ , ω ∗ ) . Every compact subset of S ∗ S_* S ∗ lies in S ( t j ) S(t_j) S ( t j ) for all large j j j , so weak convergence against tests supported in that compact set gives
u ∗ ( x ) = ℓ ∗ + ω ∗ × ( x − h ∗ ) for a.e. x ∈ S ∗ . u_*(x) = \ell_* + \omega_* \times(x-h_*) \quad\text{for a.e. } x \in S_*. u ∗ ( x ) = ℓ ∗ + ω ∗ × ( x − h ∗ ) for a.e. x ∈ S ∗ .
This establishes u ∗ ∈ H S ∗ ( Ω ) u_* \in\mathcal{H}_{S_*}(\Omega) u ∗ ∈ H S ∗ ( Ω ) , and weak lower semicontinuity gives the kinetic part of (139) .
Finally, fix a divergence-free ζ ∈ C c ∞ ( Ω ; R 3 ) \zeta\in C^\infty_c(\Omega;\mathbb{R}^3) ζ ∈ C c ∞ ( Ω ; R 3 ) that is rigid on ( S ∗ ) r (S_*)_r ( S ∗ ) r for some r > 0 r>0 r > 0 . By the convergence S ( t ) → S ∗ S(t) \to S_* S ( t ) → S ∗ , there is t 2 < T 0 t_2<T_0 t 2 < T 0 such that ζ \zeta ζ is rigid on S ( t ) r / 2 S(t)_{r/2} S ( t ) r /2 for all t ∈ ( t 2 , T 0 ) t \in(t_2,T_0) t ∈ ( t 2 , T 0 ) . Multiplied by a cut-off function in time whose support is in ( t 2 , T 0 ) (t_2,T_0) ( t 2 , T 0 ) , it is thus an admissible test field in the class (119) . The momentum equation shows that t ↦ ( u ( t ) , ζ ) L 2 ( Ω ) t \mapsto(u(t),\zeta)_{L^2(\Omega)} t ↦ ( u ( t ) , ζ ) L 2 ( Ω ) has a distributional derivative in L 1 ( t 2 , T 0 ) L^1(t_2,T_0) L 1 ( t 2 , T 0 ) , because the convective and Ericksen stresses are bounded in L x ∞ L x 1 L^\infty_xL^1_x L x ∞ L x 1 , while the viscous contribution is in L 2 ( t 2 , T 0 ) L^2(t_2,T_0) L 2 ( t 2 , T 0 ) . Hence the pairing has a unique one-sided limit. Evaluating it along t j t_j t j proves (140) .
It remains to prove the gluing statement. For this purpose, let ζ \zeta ζ be a test field on [ 0 , T 1 ) [0,T_1) [ 0 , T 1 ) in the class T u \mathcal{T}_u T u from (119) . Testing the first piece with ζ \zeta ζ multiplied by a cut-off function in time gives, for a.e. s < T 0 s<T_0 s < T 0 , its momentum identity on [ 0 , s ] [0,s] [ 0 , s ] with the additional term ∫ Ω u ( s ) ⋅ ζ ( s ) d x \int_{\Omega} u(s)\cdot\zeta(s)\,\mathrm{d}x ∫ Ω u ( s ) ⋅ ζ ( s ) d x on the left-hand side. By the uniform convergence ζ ( s ) → ζ ( T 0 ) \zeta(s)\to\zeta(T_0) ζ ( s ) → ζ ( T 0 ) uniformly, this term also converges to ∫ Ω u ∗ ⋅ ζ ( T 0 ) d x \int_{\Omega}u_*\cdot\zeta(T_0)\,\mathrm{d}x ∫ Ω u ∗ ⋅ ζ ( T 0 ) d x as s → T 0 s\to T_0 s → T 0 by (140) , and this limit is the initial term in the momentum identity of the second piece. Adding the two identities gives the momentum identity on [ 0 , T 1 ] [0,T_1] [ 0 , T 1 ] for tests ζ ∈ T u c o l \zeta\in\mathcal{T}^{\mathrm{col}}_u ζ ∈ T u col , and Proposition A.3 extends it to all ζ ∈ T u \zeta\in\mathcal{T}_u ζ ∈ T u . In the same way, the endpoint terms in the director equation cancel because d ˉ ( s ) → d ˉ ∗ \bar{d}(s)\to\bar{d}_* d ˉ ( s ) → d ˉ ∗ strongly in L 2 ( Ω ) L^2(\Omega) L 2 ( Ω ) , and the transport equation for φ \varphi φ holds across T 0 T_0 T 0 because the rigid motion is continuous.
For the energy inequality, let D ( s , t ) \mathcal{D}(s,t) D ( s , t ) denote the two dissipation integrals in (20) taken over the time interval ( s , t ) (s,t) ( s , t ) . For a.e. t ∈ ( T 0 , T 1 ) t\in(T_0,T_1) t ∈ ( T 0 , T 1 ) , the energy inequality of the second piece, (139) , and the energy inequality of the first piece at the times t j t_j t j give
E ( t ) + D ( 0 , t ) ≤ E ∗ + D ( 0 , T 0 ) ≤ lim inf j → ∞ ( E ( t j ) + D ( 0 , t j ) ) ≤ E 0 . E(t)+\mathcal{D}(0,t) \le E_*+\mathcal{D}(0,T_0) \le\liminf_{j\to\infty}\bigl(E(t_j)+\mathcal{D}(0,t_j)\bigr) \le E_0. E ( t ) + D ( 0 , t ) ≤ E ∗ + D ( 0 , T 0 ) ≤ j → ∞ lim inf ( E ( t j ) + D ( 0 , t j ) ) ≤ E 0 .
The constrained system can then be restarted from data in the energy class.
Lemma 5.2. Let S ∗ = h ∗ + O ∗ S 0 S_* = h_* + O_*S_0 S ∗ = h ∗ + O ∗ S 0 satisfy ρ ∗ : = dist ( S ∗ , ∂ Ω ) > 0 \rho_* := \operatorname{dist}(S_*,\partial\Omega)>0 ρ ∗ := dist ( S ∗ , ∂ Ω ) > 0 , consider u ∗ ∈ H S ∗ ( Ω ) u_* \in\mathcal{H}_{S_*}(\Omega) u ∗ ∈ H S ∗ ( Ω ) , and assume that d ∗ ∈ H 1 ( Ω ∖ S ∗ ‾ ; S + 2 ) d_* \in H^1(\Omega\setminus\overline{S_*};S^2_+) d ∗ ∈ H 1 ( Ω ∖ S ∗ ; S + 2 ) satisfies (138) . Define E ∗ : = 1 2 ∥ u ∗ ∥ L 2 ( Ω ) 2 + 1 2 ∥ ∇ d ∗ ∥ L 2 ( Ω ∖ S ∗ ) 2 E_* := \frac{1}{2}\|u_*\|^2_{L^2(\Omega)}+\frac{1}{2}\|\nabla d_*\|^2_{L^2(\Omega\setminus S_*)} E ∗ := 2 1 ∥ u ∗ ∥ L 2 ( Ω ) 2 + 2 1 ∥∇ d ∗ ∥ L 2 ( Ω ∖ S ∗ ) 2 .
Then there is a constrained weak solution starting from ( u ∗ , d ∗ , S ∗ ) (\mathbf{u}_*,\mathbf{d}_*,S_*) ( u ∗ , d ∗ , S ∗ ) on a time interval of length
θ ∗ : = { ρ ∗ 2 C S 0 2 E ∗ , for E ∗ > 0 , + ∞ , for E ∗ = 0. \theta_* :=
\begin{cases}
\frac{\rho_*}{2C_{S_0}\sqrt{2E_*}}, & \text{for } E_* > 0,\\
+\infty, & \text{for } E_* = 0.
\end{cases} θ ∗ := { 2 C S 0 2 E ∗ ρ ∗ , + ∞ , for E ∗ > 0 , for E ∗ = 0.
Moreover, the body remains at distance at least ρ ∗ / 2 \rho_*/2 ρ ∗ /2 from ∂ Ω \partial\Omega ∂ Ω on that interval.
Proof. If E ∗ = 0 E_* = 0 E ∗ = 0 , then u ∗ = 0 \mathbf{u}_* = 0 u ∗ = 0 and ∇ d ∗ = 0 \nabla\mathbf{d}_* = 0 ∇ d ∗ = 0 . This means that the body at rest at S ∗ S_* S ∗ , together with u = 0 \mathbf{u} = 0 u = 0 and d ( t ) = d ∗ \mathbf{d}(t) = \mathbf{d}_* d ( t ) = d ∗ , is a constrained weak solution for all times.
For a global definition, we extend d ∗ \mathbf{d}_* d ∗ through the solid by the associated material datum, i.e.,
d ‾ ∗ ( x ) : = { d ‾ ∗ ( x ) , for x ∈ Ω ∖ S ‾ ∗ , d 0 ( O ∗ T ( x − h ∗ ) ) , for x ∈ S ∗ . \overline{\mathbf{d}}_*(x) :=
\begin{cases}
\overline{\mathbf{d}}_*(x), & \text{for } x \in\Omega\setminus\overline{S}_*,\\
\mathbf{d}_0\bigl(O_*^{\mathsf{T}}(x-\mathbf{h}_*)\bigr), & \text{for } x \in S_*.
\end{cases} d ∗ ( x ) := { d ∗ ( x ) , d 0 ( O ∗ T ( x − h ∗ ) ) , for x ∈ Ω ∖ S ∗ , for x ∈ S ∗ .
From (138) , it follows that d ‾ ∗ ∈ H 1 ( Ω ; S + 2 ) \overline{\mathbf{d}}_* \in H^1(\Omega;\mathbb{S}^2_+) d ∗ ∈ H 1 ( Ω ; S + 2 ) . This allows us to apply Lemma A.1 , where the approximation index tied to the Ginzburg–Landau parameter ε \varepsilon ε , and we choose
d ‾ ∗ ε ∈ H 2 ( Ω ; S + 2 ) , ∣ d ‾ ∗ ε ∣ = 1 , d ‾ ∗ ε → d ‾ ∗ strongly in H 1 ( Ω ) (141) \overline{\mathbf{d}}_*^\varepsilon\in H^2(\Omega;\mathbb{S}^2_+), \qquad\lvert\overline{\mathbf{d}}_*^\varepsilon\rvert= 1, \qquad\overline{\mathbf{d}}_*^\varepsilon\to\overline{\mathbf{d}}_* \quad\text{strongly in } H^1(\Omega) \tag*{(141)} d ∗ ε ∈ H 2 ( Ω ; S + 2 ) , ∣ d ∗ ε ∣ = 1 , d ∗ ε → d ∗ strongly in H 1 ( Ω ) (141)
as ε → 0 \varepsilon\to0 ε → 0 . Next, we apply the construction of § 3 and § 4 with the restart time as new initial time, with S ∗ S_* S ∗ , u ∗ \mathbf{u}_* u ∗ , and d ‾ ∗ ε \overline{\mathbf{d}}_*^\varepsilon d ∗ ε replacing S 0 S_0 S 0 , u 0 \mathbf{u}_0 u 0 , and d 0 ε \mathbf{d}_0^\varepsilon d 0 ε . In particular, φ 0 = 1 S ∗ \varphi_0 = 1_{S_*} φ 0 = 1 S ∗ , and no change of coordinates is needed. We keep the description of the rigid motion through the original reference body, S ( t ) = h ( t ) + O ( t ) S 0 S(t) = \mathbf{h}(t) + O(t)S_0 S ( t ) = h ( t ) + O ( t ) S 0 with h ( 0 ) = h ∗ \mathbf{h}(0) = \mathbf{h}_* h ( 0 ) = h ∗ and O ( 0 ) = O ∗ O(0) = O_* O ( 0 ) = O ∗ , so that the points of S ∗ S_* S ∗ move according to
Z ∗ ( t , x ) = h ( t ) + O ( t ) O ∗ T ( x − h ∗ ) , x ∈ S ∗ . Z_*(t,x) = \mathbf{h}(t) + O(t)O_*^{\mathsf{T}}(x-\mathbf{h}_*), \qquad x \in S_*. Z ∗ ( t , x ) = h ( t ) + O ( t ) O ∗ T ( x − h ∗ ) , x ∈ S ∗ .
By definition, d ‾ ∗ ( x ) = d 0 ( O ∗ T ( x − h ∗ ) ) \overline{\mathbf{d}}_*(x) = \mathbf{d}_0\bigl(O_*^{\mathsf{T}}(x-\mathbf{h}_*)\bigr) d ∗ ( x ) = d 0 ( O ∗ T ( x − h ∗ ) ) in S ∗ S_* S ∗ , so the anchoring datum transported by Z ∗ Z_* Z ∗ is d 0 ( O ( t ) T ( x − h ( t ) ) ) \mathbf{d}_0\bigl(O(t)^{\mathsf{T}}(x-\mathbf{h}(t))\bigr) d 0 ( O ( t ) T ( x − h ( t )) ) on S ( t ) S(t) S ( t ) , as in (16) . Let us observe that the solid mass and the eigenvalues of the inertia tensor are invariant under rigid placement, the Korn–Poincaré constant depends only on Ω \Omega Ω , and the lifespan estimate depends on the initial gap to ∂ Ω \partial\Omega ∂ Ω . As a result, the construction and the uniform separation estimates of § 2.3 and § 3 apply with S ∗ S_* S ∗ as initial solid.
For each sufficiently small ε \varepsilon ε , Theorem 3.2 thus gives a solution to the moving-domain problem with Ginzburg–Landau regularization starting from ( u ∗ , d ‾ ∗ ε , S ∗ ) (\mathbf{u}_*,\overline{\mathbf{d}}_*^\varepsilon,S_*) ( u ∗ , d ∗ ε , S ∗ ) , and with initial energy
E ∗ ε = 1 2 ∥ u ∗ ∥ L 2 ( Ω ) 2 + 1 2 ∥ ∇ d ‾ ∗ ε ∥ L 2 ( Ω ∖ S ∗ ) 2 → E ∗ . E_*^\varepsilon= \frac{1}{2}\lVert\mathbf{u}_*\rVert_{L^2(\Omega)}^2 + \frac{1}{2}\lVert\nabla\overline{\mathbf{d}}_*^\varepsilon\rVert_{L^2(\Omega\setminus S_*)}^2 \to E_*. E ∗ ε = 2 1 ∥ u ∗ ∥ L 2 ( Ω ) 2 + 2 1 ∥ ∇ d ∗ ε ∥ L 2 ( Ω ∖ S ∗ ) 2 → E ∗ .
Here the initial potential energy vanishes due to the sphere-valuedness. By Theorem 3.2 , this solution then exists at least on [ 0 , θ ε ] [0,\theta_\varepsilon] [ 0 , θ ε ] , where, in accordance with (35) with ρ ∗ \rho_* ρ ∗ and E ∗ ε E_*^\varepsilon E ∗ ε in place of δ 0 \delta_0 δ 0 and E 0 E_0 E 0 ,
θ ε : = ρ ∗ 2 C S 0 2 E ∗ ε , θ ε → θ ∗ . \theta_\varepsilon:= \frac{\rho_*}{2C_{S_0}\sqrt{2E_*^\varepsilon}}, \qquad\theta_\varepsilon\to\theta_*. θ ε := 2 C S 0 2 E ∗ ε ρ ∗ , θ ε → θ ∗ .
In particular, by Corollary 2.9 , for every θ < θ ∗ \theta< \theta_* θ < θ ∗ and all sufficiently small ε \varepsilon ε ,
dist ( S ε ( t ) , ∂ Ω ) ≥ ρ ∗ 2 ( 0 ≤ t ≤ θ ) . (142) \operatorname{dist}(S_\varepsilon(t),\partial\Omega) \ge\frac{\rho_*}{2} \qquad(0 \le t \le\theta). \tag*{(142)} dist ( S ε ( t ) , ∂ Ω ) ≥ 2 ρ ∗ ( 0 ≤ t ≤ θ ) . (142)
It remains to pass to the limit with only H 1 H^1 H 1 -regularity of d ∗ \mathbf{d}_* d ∗ . The rigid motions are compact in C ( [ 0 , θ ] ) C([0,\theta]) C ([ 0 , θ ]) due to the uniform energy inequality and (142) , the velocities are strongly compact in L 2 ( ( 0 , θ ) × Ω ) L^2((0,\theta)\times\Omega) L 2 (( 0 , θ ) × Ω ) by Lemma 3.14 , and the directors are strongly compact in L l o c 2 ( 0 , θ ; H l o c 1 ( F ( t ) ) ) L^2_{\mathrm{loc}}(0,\theta;H^1_{\mathrm{loc}}(\mathcal{F}(t))) L loc 2 ( 0 , θ ; H loc 1 ( F ( t ))) by Lemma 4.6 . We recall that (135) from Proposition 4.7 identifies the limiting molecular field as Δ d + ∣ ∇ d ∣ 2 d \Delta\mathbf{d}+\lvert\nabla\mathbf{d}\rvert^2\mathbf{d} Δ d + ∣ ∇ d ∣ 2 d . The argument of Lemma 4.8 , applied with d ‾ ∗ ε \overline{\mathbf{d}}_*^\varepsilon d ∗ ε from (141) instead of (120) , yields that the limiting director satisfies the anchoring condition (16) with the original datum d 0 \mathbf{d}_0 d 0 . The elastic stress first passes against tests ζ ∈ T u c o l \boldsymbol{\zeta}\in\mathcal{T}^{\mathrm{col}}_{\mathbf{u}} ζ ∈ T u col and then against every standard rigid test ζ ∈ T u \boldsymbol{\zeta}\in\mathcal{T}_{\mathbf{u}} ζ ∈ T u thanks to Proposition A.3 .
With regard to the energy inequality, we argue as at the end of § 4 , with the material extensions of the restart director fields. The only change is that the solid part of the gradient norm now equals ∥ ∇ d ˉ ∗ ε ∥ L 2 ( S ∗ ) 2 \lVert\nabla\bar{d}_*^\varepsilon\rVert_{L^2(S_*)}^2 ∥ ∇ d ˉ ∗ ε ∥ L 2 ( S ∗ ) 2 , which converges to ∥ ∇ d ˉ ∗ ∥ L 2 ( S ∗ ) 2 \lVert\nabla\bar{d}_*\rVert_{L^2(S_*)}^2 ∥ ∇ d ˉ ∗ ∥ L 2 ( S ∗ ) 2 by (141) . The convergence E ∗ ε → E ∗ E_*^\varepsilon\to E_* E ∗ ε → E ∗ shown above implies that the limiting energy inequality has initial energy E ∗ E_* E ∗ , so in total, the limit is a constrained weak solution with initial state ( u ∗ , d ∗ , S ∗ ) (u_*,d_*,S_*) ( u ∗ , d ∗ , S ∗ ) .
The preceding arguments give a constrained weak solution on [ 0 , θ ] [0,\theta] [ 0 , θ ] for every θ < θ ∗ \theta<\theta_* θ < θ ∗ . Repeating the diagonal argument at the end of § 4 , with a sequence θ m ↑ θ ∗ \theta_m \uparrow\theta_* θ m ↑ θ ∗ instead of T m ↑ T ∗ T_m \uparrow T_* T m ↑ T ∗ , we obtain a single constrained weak solution on [ 0 , θ ∗ ) [0,\theta_*) [ 0 , θ ∗ ) . Moreover, Corollary 2.9 gives dist ( S ( t ) , ∂ Ω ) ≥ ρ ∗ / 2 \operatorname{dist}(S(t),\partial\Omega)\geq\rho_*/2 dist ( S ( t ) , ∂ Ω ) ≥ ρ ∗ /2 for 0 ≤ t < θ ∗ 0\leq t<\theta_* 0 ≤ t < θ ∗ . If θ ∗ < ∞ \theta_*<\infty θ ∗ < ∞ , Lemma 5.1 , applied with T 0 = θ ∗ T_0=\theta_* T 0 = θ ∗ , provides the endpoint state at θ ∗ \theta_* θ ∗ . □
Proof of Theorem 2.3 (i) and Corollary 2.10 . If E 0 = 0 E_0=0 E 0 = 0 , then u 0 = 0 u_0=0 u 0 = 0 and ∇ d 0 = 0 \nabla d_0=0 ∇ d 0 = 0 a.e. in F 0 F_0 F 0 . Thus, the body at rest together with u = 0 u=0 u = 0 and d ( t ) = d 0 ∣ F 0 d(t)=d_0|_{F_0} d ( t ) = d 0 ∣ F 0 is a global constrained solution, and T c o l = ∞ T_{\mathrm{col}}=\infty T col = ∞ . Therefore, we assume E 0 > 0 E_0>0 E 0 > 0 and write C R : = C S 0 C_R:=C_{S_0} C R := C S 0 in the remainder of the proof.
We start by constructing one branch of the solution. Proposition 4.1 yields that there is a constrained weak solution on [ 0 , T ∗ ) [0,T_*) [ 0 , T ∗ ) . We then fix t 1 ∈ ( 0 , T ∗ ) t_1\in(0,T_*) t 1 ∈ ( 0 , T ∗ ) and restrict the solution to [ 0 , t 1 ] [0,t_1] [ 0 , t 1 ] . By Lemma 5.1 , applied with T 0 = t 1 T_0=t_1 T 0 = t 1 , it has an endpoint state ( u ∗ , d ∗ , S ∗ ) (u_*,d_*,S_*) ( u ∗ , d ∗ , S ∗ ) at t 1 t_1 t 1 that satisfies the hypotheses of Lemma 5.2 . We then extend this branch inductively without modifying pieces that have been constructed previously. Suppose that it has been constructed on [ 0 , t j ] [0,t_j] [ 0 , t j ] , with gap ρ j : = dist ( S ( t j ) , ∂ Ω ) > 0 \rho_j:=\operatorname{dist}(S(t_j),\partial\Omega)>0 ρ j := dist ( S ( t j ) , ∂ Ω ) > 0 and endpoint energy at most E 0 E_0 E 0 . By Lemma 5.2 , the endpoint data at t j t_j t j can be continued to a constrained weak solution for at least ρ j / ( 2 C R 2 E 0 ) \rho_j/(2C_R\sqrt{2E_0}) ρ j / ( 2 C R 2 E 0 ) units of time, during which the gap stays at least ρ j / 2 \rho_j/2 ρ j /2 . We retain the piece of length
t j + 1 − t j : = ρ j 4 C R 2 E 0 , t_{j+1}-t_j:=\frac{\rho_j}{4C_R\sqrt{2E_0}}, t j + 1 − t j := 4 C R 2 E 0 ρ j ,
whose terminal gap is positive, apply Lemma 5.1 at t j + 1 t_{j+1} t j + 1 , and continue. If the energy of an endpoint state vanishes, we continue by the solution at rest, as in the case E 0 = 0 E_0=0 E 0 = 0 . The last assertion of Lemma 5.1 implies that every such concatenation is a constrained weak solution on the combined interval and satisfies the energy inequality (20) with the original initial energy E 0 E_0 E 0 .
We define T c o l : = lim j → ∞ t j ∈ ( 0 , ∞ ] T_{\mathrm{col}}:=\lim_{j\to\infty}t_j\in(0,\infty] T col := lim j → ∞ t j ∈ ( 0 , ∞ ] . The energy inequality (20) yields ∥ u ( t ) ∥ L 2 ( Ω ) ≤ 2 E 0 \lVert u(t)\rVert_{L^2(\Omega)}\leq\sqrt{2E_0} ∥ u ( t ) ∥ L 2 ( Ω ) ≤ 2 E 0 for a.e. t < T c o l t<T_{\mathrm{col}} t < T col . By Lemma 2.7 , the velocity of every point of the body is bounded by C S 0 2 E 0 C_{S_0}\sqrt{2E_0} C S 0 2 E 0 , showing that the rigid trajectory and the gap t ↦ dist ( S ( t ) , ∂ Ω ) t\mapsto\operatorname{dist}(S(t),\partial\Omega) t ↦ dist ( S ( t ) , ∂ Ω ) are Lipschitz continuous on [ 0 , T c o l ) [0,T_{\mathrm{col}}) [ 0 , T col ) . If T c o l < ∞ T_{\mathrm{col}}<\infty T col < ∞ , the gap thus has a limit ρ ≥ 0 \rho\geq0 ρ ≥ 0 as t → T c o l t\to T_{\mathrm{col}} t → T col . The case ρ > 0 \rho>0 ρ > 0 implies ρ j ≥ ρ / 2 \rho_j\geq\rho/2 ρ j ≥ ρ /2 for all large j j j , hence t j + 1 − t j ≥ ρ / ( 8 C R 2 E 0 ) t_{j+1}-t_j\geq\rho/(8C_R\sqrt{2E_0}) t j + 1 − t j ≥ ρ / ( 8 C R 2 E 0 ) , which contradicts the convergence of ( t j ) (t_j) ( t j ) . Therefore, ρ = 0 \rho=0 ρ = 0 , the rigid motion extends continuously to T c o l T_{\mathrm{col}} T col , and (23) holds. By positivity of every ρ j \rho_j ρ j , the time T c o l T_{\mathrm{col}} T col is the first contact time of the constructed branch, proving Theorem 2.3 (i) and the first assertion of Theorem 2.3 (ii).
Finally, we establish a lower bound for the collision time T c o l T_{\mathrm{col}} T col . In fact, the branch satisfies (20) on every compact subinterval of [ 0 , T c o l ) [0,T_{\mathrm{col}}) [ 0 , T col ) , so Lemma 2.8 applies, and (34) is (36) . Recall from (31) that T e d T_{\mathrm{ed}} T ed is the unique time at which the right-hand side of (36) vanishes, and that this right-hand side is positive for t < T e d t<T_{\mathrm{ed}} t < T ed . Hence, if T c o l < T e d T_{\mathrm{col}}<T_{\mathrm{ed}} T col < T ed , the gap would stay bounded away from zero as t → T c o l t\to T_{\mathrm{col}} t → T col , which contradicts ρ = 0 \rho=0 ρ = 0 . Therefore T c o l ≥ T e d T_{\mathrm{col}}\geq T_{\mathrm{ed}} T col ≥ T ed . This completes the proof of Corollary 2.10 . □
Post-impact constrained continuation
We identify the one-sided data at the first wall contact and then solve the constrained Ericksen–Leslie system on the resulting fixed fluid region. The post-impact solution is obtained by a Ginzburg–Landau limit on this fixed domain.
Let G ⊂ R 3 G\subset\mathbb{R}^3 G ⊂ R 3 be bounded and open, and let D G D_G D G , H G \mathcal{H}_G H G , and V G \mathcal{V}_G V G be defined by (10) . We recall that the fixed-domain solution concept was stated in Definition 2.1 .
Proposition 5.3. For every bounded open G ⊂ R 3 G\subset\mathbb{R}^3 G ⊂ R 3 , every u ∗ ∈ H G u_*\in\mathcal{H}_G u ∗ ∈ H G , and every d ∗ ∈ H 1 ( G ; S 2 ) d_*\in H^1(G;\mathbb{S}^2) d ∗ ∈ H 1 ( G ; S 2 ) , there exists a post-impact constrained weak solution in the sense of Definition 2.1 on [ 0 , ∞ ) [0,\infty) [ 0 , ∞ ) . Moreover, there are ε j → 0 \varepsilon_j\to0 ε j → 0 and corresponding global fixed-domain Ginzburg–Landau solutions ( u j , d j , M j ) (u_j,d_j,M_j) ( u j , d j , M j ) with
the same initial data such that, for every T < ∞ T < \infty T < ∞ ,
u j → u strongly in L 2 ( ( 0 , T ) × G ) , d j → d strongly in L 2 ( ( 0 , T ) × G ) , d j ⇀ ∗ d in L ∞ ( 0 , T ; H 1 ( G ) ) , d j → d strongly in L 2 ( 0 , T ; H 1 ( V ) ) for every V ⋐ G , M j ⇀ τ weakly in L 2 ( ( 0 , T ) × G ) . \begin{aligned}
\boldsymbol{u}_j &\to\boldsymbol{u} &&\text{strongly in } L^2((0,T)\times G),\\
\boldsymbol{d}_j &\to\boldsymbol{d} &&\text{strongly in } L^2((0,T)\times G), \qquad\boldsymbol{d}_j \overset{*}{\rightharpoonup} \boldsymbol{d}\text{ in } L^\infty(0,T;H^1(G)),\\
\boldsymbol{d}_j &\to\boldsymbol{d} &&\text{strongly in } L^2(0,T;H^1(V))\quad\text{for every }V\Subset G,\\
M_j &\rightharpoonup\tau&&\text{weakly in } L^2((0,T)\times G).
\end{aligned} u j d j d j M j → u → d → d ⇀ τ strongly in L 2 (( 0 , T ) × G ) , strongly in L 2 (( 0 , T ) × G ) , d j ⇀ ∗ d in L ∞ ( 0 , T ; H 1 ( G )) , strongly in L 2 ( 0 , T ; H 1 ( V )) for every V ⋐ G , weakly in L 2 (( 0 , T ) × G ) .
Proof. We fix T > 0 T>0 T > 0 , let ε j → 0 \varepsilon_j\to0 ε j → 0 , and apply Proposition 3.17 on G G G with the same data ( u ∗ , d ∗ ) (\boldsymbol{u}_*,\boldsymbol{d}_*) ( u ∗ , d ∗ ) for every j j j . The director condition ∣ d ∗ ∣ = 1 |\boldsymbol{d}_*|=1 ∣ d ∗ ∣ = 1 implies that the initial Ginzburg–Landau potential vanishes. From (116) , it then follows, uniformly in j j j ,
∥ u j ∥ L t ∞ L x 2 2 + ∥ u j ∥ L t 2 H x 1 2 + ∥ d j ∥ L t ∞ H x 1 2 + ∥ M j ∥ L t , x 2 2 ≤ C T , \|\boldsymbol{u}_j\|_{L^\infty_tL^2_x}^2+\|\boldsymbol{u}_j\|_{L^2_tH^1_x}^2+\|\boldsymbol{d}_j\|_{L^\infty_tH^1_x}^2+\|M_j\|_{L^2_{t,x}}^2\le C_T, ∥ u j ∥ L t ∞ L x 2 2 + ∥ u j ∥ L t 2 H x 1 2 + ∥ d j ∥ L t ∞ H x 1 2 + ∥ M j ∥ L t , x 2 2 ≤ C T ,
sup 0 < t < T 1 ε j 2 ∫ G ( 1 − ∣ d j ( t ) ∣ 2 ) 2 d x ≤ C T , ∣ d j ∣ ≤ 1 , ( d j ) 3 ≥ 0 , (143) \sup_{0<t<T}\frac{1}{\varepsilon_j^2}\int_G(1-|\boldsymbol{d}_j(t)|^2)^2\,\mathrm{d}x\le C_T,\qquad|\boldsymbol{d}_j|\le1,\qquad(\boldsymbol{d}_j)_3\ge0, \tag*{(143)} 0 < t < T sup ε j 2 1 ∫ G ( 1 − ∣ d j ( t ) ∣ 2 ) 2 d x ≤ C T , ∣ d j ∣ ≤ 1 , ( d j ) 3 ≥ 0 , (143)
where we recall M j = Δ d j + ε j − 2 ( 1 − ∣ d j ∣ 2 ) d j M_j=\Delta\boldsymbol{d}_j+\varepsilon_j^{-2}(1-|\boldsymbol{d}_j|^2)\boldsymbol{d}_j M j = Δ d j + ε j − 2 ( 1 − ∣ d j ∣ 2 ) d j . By u j ∈ L t 2 L x 6 \boldsymbol{u}_j\in L^2_tL^6_x u j ∈ L t 2 L x 6 and ∇ d j ∈ L t ∞ L x 2 \nabla\boldsymbol{d}_j\in L^\infty_tL^2_x ∇ d j ∈ L t ∞ L x 2 , the director equation implies that ∂ t d j = M j − u j ⋅ ∇ d j \partial_t\boldsymbol{d}_j=M_j-\boldsymbol{u}_j\cdot\nabla\boldsymbol{d}_j ∂ t d j = M j − u j ⋅ ∇ d j is bounded in L 2 ( 0 , T ; H − 1 ( G ) ) L^2(0,T;H^{-1}(G)) L 2 ( 0 , T ; H − 1 ( G )) . For q j : = d j − d ∗ \boldsymbol{q}_j:=\boldsymbol{d}_j-\boldsymbol{d}_* q j := d j − d ∗ , we find that q j \boldsymbol{q}_j q j is bounded in L ∞ ( 0 , T ; H 0 1 ( G ) ) L^\infty(0,T;H^1_0(G)) L ∞ ( 0 , T ; H 0 1 ( G )) , while ∂ t q j = ∂ t d j \partial_t\boldsymbol{q}_j=\partial_t\boldsymbol{d}_j ∂ t q j = ∂ t d j is bounded in L 2 ( 0 , T ; H − 1 ( G ) ) L^2(0,T;H^{-1}(G)) L 2 ( 0 , T ; H − 1 ( G )) . The zero-extension map is continuous from H 0 1 ( G ) H^1_0(G) H 0 1 ( G ) into H 0 1 ( B ) H^1_0(B) H 0 1 ( B ) for any fixed ball B ⊇ G ‾ B\supseteq\overline{G} B ⊇ G , so the Rellich–Kondrachov theorem on B B B shows that H 0 1 ( G ) ⋐ L 2 ( G ) H^1_0(G)\Subset L^2(G) H 0 1 ( G ) ⋐ L 2 ( G ) for every bounded open G G G . Applying the Aubin–Lions lemma to q j \boldsymbol{q}_j q j imply that d j → d \boldsymbol{d}_j\to\boldsymbol{d} d j → d strongly in L 2 ( ( 0 , T ) × G ) L^2((0,T)\times G) L 2 (( 0 , T ) × G ) and, after extraction, weak-* in L t ∞ H x 1 L^\infty_tH^1_x L t ∞ H x 1 . The potential bound in (143) therefore implies ∣ d ∣ = 1 |\boldsymbol{d}|=1 ∣ d ∣ = 1 a.e. Since ( d j ) 3 ≥ 0 (\boldsymbol{d}_j)_3\ge0 ( d j ) 3 ≥ 0 by (143) , the strong convergence gives d 3 ≥ 0 d_3\ge0 d 3 ≥ 0 a.e. The fact that d j − d ∗ ∈ H 0 1 ( G ; R 3 ) \boldsymbol{d}_j-\boldsymbol{d}_*\in H^1_0(G;\mathbb{R}^3) d j − d ∗ ∈ H 0 1 ( G ; R 3 ) and the weak closedness of this space imply d − d ∗ ∈ L ∞ ( 0 , T ; H 0 1 ( G ; R 3 ) ) \boldsymbol{d}-\boldsymbol{d}_*\in L^\infty(0,T;H^1_0(G;\mathbb{R}^3)) d − d ∗ ∈ L ∞ ( 0 , T ; H 0 1 ( G ; R 3 )) .
Let now V ⋐ U ⋐ G V\Subset U\Subset G V ⋐ U ⋐ G . The sequence ( d j ) (\boldsymbol{d}_j) ( d j ) is Cauchy in L 2 ( ( 0 , T ) × U ) L^2((0,T)\times U) L 2 (( 0 , T ) × U ) by the preceding Aubin–Lions argument. By (143) , the maps d j ( t ) \boldsymbol{d}_j(t) d j ( t ) satisfy (129) on U U U with e j = ε j e_j=\varepsilon_j e j = ε j and τ j = M j ( t ) \tau_j=M_j(t) τ j = M j ( t ) , as well as (130) . Thus, for every V ⋐ G V\Subset G V ⋐ G , Lemma 4.6 yields
d j → d strongly in L 2 ( 0 , T ; H 1 ( V ) ) . (144) \boldsymbol{d}_j\to\boldsymbol{d}\qquad\text{strongly in }L^2(0,T;H^1(V)). \tag*{(144)} d j → d strongly in L 2 ( 0 , T ; H 1 ( V )) . (144)
After passing to the limit M j ⇀ τ M_j\rightharpoonup\tau M j ⇀ τ in L 2 ( ( 0 , T ) × G ) L^2((0,T)\times G) L 2 (( 0 , T ) × G ) , arguing as in (135) in the proof of Proposition 4.7 gives τ = Δ d + ∣ ∇ d ∣ 2 d \tau=\Delta\boldsymbol{d}+|\nabla\boldsymbol{d}|^2\boldsymbol{d} τ = Δ d + ∣∇ d ∣ 2 d in D ′ ( ( 0 , T ) × G ) \mathcal{D}'((0,T)\times G) D ′ (( 0 , T ) × G ) .
For the velocity, we fix the basis ( v m ) (\boldsymbol{v}_m) ( v m ) from Lemma 3.15 . Testing the weak momentum equation with χ ( t ) v m \chi(t)\boldsymbol{v}_m χ ( t ) v m , χ ∈ C c 1 ( 0 , T ) \chi\in C^1_c(0,T) χ ∈ C c 1 ( 0 , T ) , shows that t ↦ ( u j ( t ) , v m ) L 2 ( G ) t\mapsto(\boldsymbol{u}_j(t),\boldsymbol{v}_m)_{L^2(G)} t ↦ ( u j ( t ) , v m ) L 2 ( G ) belongs to W 1 , 1 ( 0 , T ) W^{1,1}(0,T) W 1 , 1 ( 0 , T ) , with distributional derivative given by the same expression as in the proof of Proposition 3.17. By (143) , we get
∫ 0 T ∣ d d t ( u j , v m ) L 2 ( G ) ∣ d t ≤ C m ( T ∥ u j ∥ L t ∞ L x 2 ∥ D ( u j ) ∥ L t , x 2 + T 1 / 2 ∥ ∇ d j ∥ L t ∞ L x 2 2 ) ≤ C m , T . \int_0^T\left|\frac{\mathrm{d}}{\mathrm{d}t}(\boldsymbol{u}_j,\boldsymbol{v}_m)_{L^2(G)}\right|\,\mathrm{d}t
\le C_m\left(T\|\boldsymbol{u}_j\|_{L^\infty_tL^2_x}\|\mathbb{D}(\boldsymbol{u}_j)\|_{L^2_{t,x}}+T^{1/2}\|\nabla\boldsymbol{d}_j\|_{L^\infty_tL^2_x}^2\right)
\le C_{m,T}. ∫ 0 T d t d ( u j , v m ) L 2 ( G ) d t ≤ C m ( T ∥ u j ∥ L t ∞ L x 2 ∥ D ( u j ) ∥ L t , x 2 + T 1/2 ∥∇ d j ∥ L t ∞ L x 2 2 ) ≤ C m , T .
These functions thus have uniformly bounded variation on ( 0 , T ) (0,T) ( 0 , T ) , so Lemma 3.15 implies
u j → u strongly in L 2 ( ( 0 , T ) × G ) . (145) \boldsymbol{u}_j\to\boldsymbol{u}\qquad\text{strongly in }L^2((0,T)\times G). \tag*{(145)} u j → u strongly in L 2 (( 0 , T ) × G ) . (145)
For test functions with compact support, (144) and (145) allow for the limit passage in the two nonlinear terms in the momentum and director equations. The limiting director equation, written with the tension τ \tau τ , depends continuously on ψ ∈ X G \psi\in\mathcal{X}_G ψ ∈ X G . The transport term is bounded by ∥ u ∥ L t 2 L x 6 ∥ ∇ d ∥ L t ∞ L x 2 ∥ ψ ∥ L t 2 L x 3 \|\boldsymbol{u}\|_{L^2_tL^6_x}\|\nabla\boldsymbol{d}\|_{L^\infty_tL^2_x}\|\psi\|_{L^2_tL^3_x} ∥ u ∥ L t 2 L x 6 ∥∇ d ∥ L t ∞ L x 2 ∥ ψ ∥ L t 2 L x 3 , and the tension term by ∥ τ ∥ L t , x 2 ∥ ψ ∥ L t , x 2 \|\tau\|_{L^2_{t,x}}\|\psi\|_{L^2_{t,x}} ∥ τ ∥ L t , x 2 ∥ ψ ∥ L t , x 2 , where τ ∈ L 2 ( ( 0 , T ) × G ) \tau\in L^2((0,T)\times G) τ ∈ L 2 (( 0 , T ) × G ) as the weak limit of the fields M j M_j M j , which are bounded by (143) . The density of C c ∞ ( ( 0 , T ) × G ) C^\infty_c((0,T)\times G) C c ∞ (( 0 , T ) × G ) in X G \mathcal{X}_G X G allows us to infer that the director equation holds for every ψ ∈ X G \psi\in\mathcal{X}_G ψ ∈ X G .
For the energy inequality, we pass to a further subsequence such that
u j ( t ) → u ( t ) , d j ( t ) → d ( t ) strongly in L 2 ( G ) \boldsymbol{u}_j(t)\to\boldsymbol{u}(t),\qquad\boldsymbol{d}_j(t)\to\boldsymbol{d}(t)\qquad\text{strongly in }L^2(G) u j ( t ) → u ( t ) , d j ( t ) → d ( t ) strongly in L 2 ( G )
for a.e. t ∈ ( 0 , T ) t \in(0,T) t ∈ ( 0 , T ) . Fix such a time t t t at which (116) holds for every j j j . Since ( d j ( t ) ) j (d_j(t))_j ( d j ( t ) ) j is bounded in H 1 ( G ) H^1(G) H 1 ( G ) by (143) , every weak H 1 H^1 H 1 -limit of a subsequence equals d ( t ) d(t) d ( t ) . Hence
∥ ∇ d ( t ) ∥ L 2 ( G ) 2 ≤ lim inf j → ∞ ∥ ∇ d j ( t ) ∥ L 2 ( G ) 2 , ∥ u j ( t ) ∥ L 2 ( G ) 2 → ∥ u ( t ) ∥ L 2 ( G ) 2 . \|\nabla d(t)\|_{L^2(G)}^2 \le\liminf_{j\to\infty}\|\nabla d_j(t)\|_{L^2(G)}^2,\qquad\|u_j(t)\|_{L^2(G)}^2 \to\|u(t)\|_{L^2(G)}^2. ∥∇ d ( t ) ∥ L 2 ( G ) 2 ≤ j → ∞ lim inf ∥∇ d j ( t ) ∥ L 2 ( G ) 2 , ∥ u j ( t ) ∥ L 2 ( G ) 2 → ∥ u ( t ) ∥ L 2 ( G ) 2 .
The Ginzburg–Landau potential F ε j F_{\varepsilon_j} F ε j is nonnegative, and the dissipation terms 2 ∫ 0 t ∫ G ∣ D ( u ) ∣ 2 d x d s 2\int_0^t\int_G|\mathbb{D}(u)|^2\,dx\,ds 2 ∫ 0 t ∫ G ∣ D ( u ) ∣ 2 d x d s and ∫ 0 t ∫ G ∣ M ∣ 2 d x d s \int_0^t\int_G|M|^2\,dx\,ds ∫ 0 t ∫ G ∣ M ∣ 2 d x d s are weakly lower semicontinuous, since D ( u j ) ⇀ D ( u ) \mathbb{D}(u_j)\rightharpoonup\mathbb{D}(u) D ( u j ) ⇀ D ( u ) and M j ⇀ τ M_j\rightharpoonup\tau M j ⇀ τ weakly in L 2 ( ( 0 , T ) × G ) L^2((0,T)\times G) L 2 (( 0 , T ) × G ) . Passing to the liminf in (116) , whose right-hand side is independent of j j j thanks to ∣ d ∗ ∣ = 1 |d_*|=1 ∣ d ∗ ∣ = 1 , thus yields (12) for a.e. t t t .
Finally, note that the trajectories provided by Proposition 3.17 are defined on [ 0 , ∞ ) [0,\infty) [ 0 , ∞ ) . A diagonal argument over T = 1 , 2 , … T=1,2,\ldots T = 1 , 2 , … , as at the end of § 4 , then yields a single solution on [ 0 , ∞ ) [0,\infty) [ 0 , ∞ ) . This completes the proof of this proposition and especially establishes Theorem 2.6.
Remark 5.4. The proof shows that the solution of Theorem 2.6 is obtained as ε \varepsilon ε -limit with u ε → u u_\varepsilon\to u u ε → u strongly in L l o c 2 ( [ 0 , ∞ ) ; L 2 ( G ) ) L^2_{\mathrm{loc}}([0,\infty);L^2(G)) L loc 2 ([ 0 , ∞ ) ; L 2 ( G )) , d ε → d d_\varepsilon\to d d ε → d strongly in L l o c 2 ( [ 0 , ∞ ) ; L 2 ( G ) ) L^2_{\mathrm{loc}}([0,\infty);L^2(G)) L loc 2 ([ 0 , ∞ ) ; L 2 ( G )) and in L 2 ( 0 , T ; H 1 ( V ) ) L^2(0,T;H^1(V)) L 2 ( 0 , T ; H 1 ( V )) for every T < ∞ T<\infty T < ∞ and every V ⋐ G V\Subset G V ⋐ G , while, for every T < ∞ T<\infty T < ∞ , the molecular fields M ε : = Δ d ε + ε − 2 ( 1 − ∣ d ε ∣ 2 ) d ε M_\varepsilon:=\Delta d_\varepsilon+\varepsilon^{-2}(1-|d_\varepsilon|^2)d_\varepsilon M ε := Δ d ε + ε − 2 ( 1 − ∣ d ε ∣ 2 ) d ε satisfy M ε ⇀ τ M_\varepsilon\rightharpoonup\tau M ε ⇀ τ in L 2 ( ( 0 , T ) × G ) L^2((0,T)\times G) L 2 (( 0 , T ) × G ) .
Next, we identify the state of the solution at the first contact time T c T_c T c . We write u c − u_c^- u c − for a pre-impact velocity trace, that is, a weak L 2 ( Ω ) L^2(\Omega) L 2 ( Ω ) -limit of u ( t j ) u(t_j) u ( t j ) along a sequence t j → T c t_j\to T_c t j → T c , which need not be unique, and u c + u_c^+ u c + for the post-impact velocity on the residual fluid domain, which is uniquely determined.
Lemma 5.5. Let ( u , d , ϕ ) (u,d,\phi) ( u , d , ϕ ) be a constrained weak solution on [ 0 , T c ) [0,T_c) [ 0 , T c ) in the sense of Definition 2.2, with energy E ( t ) E(t) E ( t ) as in (20) , such that
dist ( S ( t ) , ∂ Ω ) > 0 ( t < T c ) , dist ( S ( t ) , ∂ Ω ) → 0 ( t → T c ) . \operatorname{dist}(S(t),\partial\Omega)>0\quad(t<T_c),\qquad\operatorname{dist}(S(t),\partial\Omega)\to0\quad(t\to T_c). dist ( S ( t ) , ∂ Ω ) > 0 ( t < T c ) , dist ( S ( t ) , ∂ Ω ) → 0 ( t → T c ) . Then the following assertions hold.
(i) The rigid motion has a unique limit S c = h c + O c S 0 S_c=h_c+O_cS_0 S c = h c + O c S 0 as t → T c t\to T_c t → T c , and there exists d c ∈ H 1 ( Ω ; S + 2 ) d_c\in H^1(\Omega;S^2_+) d c ∈ H 1 ( Ω ; S + 2 ) such that the material extension d ‾ \overline d d of d d d through the solid satisfies
d ‾ ( t ) → d c strongly in L 2 ( Ω ) and weakly in H 1 ( Ω ) ( t → T c ) . (146) \overline d(t)\to d_c\quad\text{strongly in }L^2(\Omega)\text{ and weakly in }H^1(\Omega)\quad(t\to T_c). \tag*{(146)} d ( t ) → d c strongly in L 2 ( Ω ) and weakly in H 1 ( Ω ) ( t → T c ) . (146) (ii) With Ω c : = Ω ∖ S c ‾ \Omega_c:=\Omega\setminus\overline{S_c} Ω c := Ω ∖ S c , there exists a unique u c + ∈ H Ω c u_c^+\in\mathcal{H}_{\Omega_c} u c + ∈ H Ω c such that, for every ζ ∈ D Ω c \zeta\in\mathcal{D}_{\Omega_c} ζ ∈ D Ω c with zero extension ζ ~ \widetilde\zeta ζ to Ω \Omega Ω ,
( u c + , ζ ) L 2 ( Ω c ) = lim t → T c ( u ( t ) , ζ ~ ) L 2 ( Ω ) . (u_c^+,\zeta)_{L^2(\Omega_c)}=\lim_{t\to T_c}(u(t),\widetilde\zeta)_{L^2(\Omega)}. ( u c + , ζ ) L 2 ( Ω c ) = t → T c lim ( u ( t ) , ζ ) L 2 ( Ω ) . (iii) If u ( t j ) ⇀ u c − u(t_j)\rightharpoonup u_c^- u ( t j ) ⇀ u c − in L 2 ( Ω ) L^2(\Omega) L 2 ( Ω ) along a sequence t j → T c t_j\to T_c t j → T c of times at which u ( t j ) u(t_j) u ( t j ) is rigid on S ( t j ) S(t_j) S ( t j ) , then u c − ∈ H S c ( Ω ) u_c^-\in\mathcal{H}_{S_c}(\Omega) u c − ∈ H S c ( Ω ) and u c + = P H Ω c ( u c − ∣ Ω c ) u_c^+=P_{\mathcal{H}_{\Omega_c}}(u_c^-|_{\Omega_c}) u c + = P H Ω c ( u c − ∣ Ω c ) .
(iv) We have
1 2 ∥ u c + ∥ L 2 ( Ω c ) 2 + 1 2 ∥ ∇ d c ∥ L 2 ( Ω c ) 2 ≤ e s s l i m i n f t → T c E ( t ) . (147) \frac{1}{2}\|u_c^+\|_{L^2(\Omega_c)}^2+\frac{1}{2}\|\nabla d_c\|_{L^2(\Omega_c)}^2\le\mathop{\mathrm{ess\,liminf}}_{t\to T_c}E(t). \tag*{(147)} 2 1 ∥ u c + ∥ L 2 ( Ω c ) 2 + 2 1 ∥∇ d c ∥ L 2 ( Ω c ) 2 ≤ ess liminf t → T c E ( t ) . (147) Moreover, there is a sequence t j → T c t_j\to T_c t j → T c as in (iii), chosen from the full-measure set on which the energy inequality holds, such that
E ( t j ) → e s s l i m i n f t → T c E ( t ) , u ( t j ) ⇀ u c − in L 2 ( Ω ) , E(t_j)\to\mathop{\mathrm{ess\,liminf}}_{t\to T_c}E(t),\qquad u(t_j)\rightharpoonup u_c^-\text{ in }L^2(\Omega), E ( t j ) → ess liminf t → T c E ( t ) , u ( t j ) ⇀ u c − in L 2 ( Ω ) , and every weak limit u c − u_c^- u c − obtained along such a sequence satisfies
1 2 ∥ u c − ∥ L 2 ( Ω c ) 2 + 1 2 ∥ ∇ d c ∥ L 2 ( Ω c ) 2 + 1 2 ∥ u c − ∥ L 2 ( S c ) 2 + 1 2 ∥ u c − ∣ Ω c − u c + ∥ L 2 ( Ω c ) 2 ≤ e s s l i m i n f t → T c E ( t ) . (148) \frac{1}{2}\|u_c^-\|_{L^2(\Omega_c)}^2+\frac{1}{2}\|\nabla d_c\|_{L^2(\Omega_c)}^2
+\frac{1}{2}\|u_c^-\|_{L^2(S_c)}^2+\frac{1}{2}\|u_c^-\vert_{\Omega_c}-u_c^+\|_{L^2(\Omega_c)}^2
\le\mathop{\mathrm{ess\,liminf}}_{t\to T_c}E(t). \tag*{(148)} 2 1 ∥ u c − ∥ L 2 ( Ω c ) 2 + 2 1 ∥∇ d c ∥ L 2 ( Ω c ) 2 + 2 1 ∥ u c − ∥ L 2 ( S c ) 2 + 2 1 ∥ u c − ∣ Ω c − u c + ∥ L 2 ( Ω c ) 2 ≤ ess liminf t → T c E ( t ) . (148)
Proof. By the energy inequality (20) and Lemma 2.7 , the functions h h h and O O O are Lipschitz on [ 0 , T c ) [0,T_c) [ 0 , T c ) . Hence h ( t ) → h c h(t)\to h_c h ( t ) → h c and O ( t ) → O c O(t)\to O_c O ( t ) → O c as t → T c t\to T_c t → T c , and we set S c : = h c + O c S 0 S_c:=h_c+O_cS_0 S c := h c + O c S 0 . Let d ˉ \bar{d} d ˉ be the material extension (22) , and let τ ~ \tilde{\tau} τ ~ be the extension of the tension τ = Δ d + ∣ ∇ d ∣ 2 d \tau=\Delta d+\lvert\nabla d\rvert^2d τ = Δ d + ∣ ∇ d ∣ 2 d by zero to the solid. By Lemma A.4 , we get ∂ t d ˉ + u ⋅ ∇ d ˉ = τ ~ \partial_t\bar{d}+u\cdot\nabla\bar{d}=\tilde{\tau} ∂ t d ˉ + u ⋅ ∇ d ˉ = τ ~ in D ′ ( ( 0 , T c ) × Ω ) \mathcal{D}'((0,T_c)\times\Omega) D ′ (( 0 , T c ) × Ω ) . The same argument as in the proof of Lemma 5.1 shows that q : = d ˉ − d 0 q:=\bar{d}-d_0 q := d ˉ − d 0 satisfies q ∈ L 2 ( 0 , T c ; H 1 ( Ω ) ) q\in L^2(0,T_c;H^1(\Omega)) q ∈ L 2 ( 0 , T c ; H 1 ( Ω )) and ∂ t q ∈ L 2 ( 0 , T c ; H − 1 ( Ω ) ) \partial_tq\in L^2(0,T_c;H^{-1}(\Omega)) ∂ t q ∈ L 2 ( 0 , T c ; H − 1 ( Ω )) . The Lions–Magenes lemma then yields d ˉ ∈ C ( [ 0 , T c ] ; L 2 ( Ω ) ) \bar{d}\in C([0,T_c];L^2(\Omega)) d ˉ ∈ C ([ 0 , T c ] ; L 2 ( Ω )) , so d ˉ ( t ) \bar{d}(t) d ˉ ( t ) converges strongly in L 2 ( Ω ) L^2(\Omega) L 2 ( Ω ) as t → T c t\to T_c t → T c and, being bounded in H 1 ( Ω ) H^1(\Omega) H 1 ( Ω ) , also weakly in H 1 ( Ω ) H^1(\Omega) H 1 ( Ω ) . This proves (146) . The limit d c d_c d c takes values in S + 2 S^2_+ S + 2 by the strong convergence, and d c ( x ) = d 0 ( O c T ( x − h c ) ) d_c(x)=d_0(O_c^{\mathsf{T}}(x-h_c)) d c ( x ) = d 0 ( O c T ( x − h c )) for a.e. x ∈ S c x\in S_c x ∈ S c , completing the proof of (i).
For (ii), consider ζ ∈ D Ω c \zeta\in\mathcal{D}_{\Omega_c} ζ ∈ D Ω c . By the convergence S ( t ) → S c S(t)\to S_c S ( t ) → S c , the support of ζ ~ \tilde{\zeta} ζ ~ has positive distance from S ( t ) S(t) S ( t ) for all t t t close to T c T_c T c . Multiplied by a cut-off function in time, ζ ~ \tilde{\zeta} ζ ~ is therefore an admissible test field in the momentum equation. Thanks to (20) , the convective term and the Ericksen stress are bounded in L t ∞ L x 1 L^\infty_tL^1_x L t ∞ L x 1 , and the viscous term is bounded in L t , x 2 L^2_{t,x} L t , x 2 , so the distributional derivative of t ↦ ( u ( t ) , ζ ~ ) L 2 ( Ω ) t\mapsto(u(t),\tilde{\zeta})_{L^2(\Omega)} t ↦ ( u ( t ) , ζ ~ ) L 2 ( Ω ) belongs to L 1 L^1 L 1 near T c T_c T c . Thus, the limit as t → T c t\to T_c t → T c exists, and we get the bound
∣ lim t → T c ( u ( t ) , ζ ~ ) L 2 ( Ω ) ∣ ≤ sup t < T c ∥ u ( t ) ∥ L 2 ( Ω ) ∥ ζ ∥ L 2 ( Ω c ) . \left|\lim_{t\to T_c}(u(t),\tilde{\zeta})_{L^2(\Omega)}\right|\leq\sup_{t<T_c}\|u(t)\|_{L^2(\Omega)}\|\zeta\|_{L^2(\Omega_c)}. t → T c lim ( u ( t ) , ζ ~ ) L 2 ( Ω ) ≤ t < T c sup ∥ u ( t ) ∥ L 2 ( Ω ) ∥ ζ ∥ L 2 ( Ω c ) .
Since D Ω c \mathcal{D}_{\Omega_c} D Ω c is dense in H Ω c \mathcal{H}_{\Omega_c} H Ω c , the limit defines a bounded linear functional on H Ω c \mathcal{H}_{\Omega_c} H Ω c , and the Riesz representation theorem gives a unique u c + ∈ H Ω c u_c^+\in\mathcal{H}_{\Omega_c} u c + ∈ H Ω c and thus proves (ii).
With regard to (iii), let u ( t j ) ⇀ u c − u(t_j)\rightharpoonup u_c^- u ( t j ) ⇀ u c − be as in the statement. Then u c − ∈ H ( Ω ) u_c^-\in\mathcal{H}(\Omega) u c − ∈ H ( Ω ) by weak closedness. On S ( t j ) S(t_j) S ( t j ) , we write
u ( t j , x ) = ℓ j + ω j × ( x − h ( t j ) ) . u(t_j,x)=\ell_j+\omega_j\times(x-h(t_j)). u ( t j , x ) = ℓ j + ω j × ( x − h ( t j )) .
By (20) and the proof of Lemma 2.7 , the sequence ( ℓ j , ω j ) (\ell_j,\omega_j) ( ℓ j , ω j ) is bounded, and after extraction it converges. Since for sufficiently large j j j , every compact subset of S c S_c S c lies in S ( t j ) S(t_j) S ( t j ) , testing the weak convergence against fields with compact support in the set S c S_c S c reveals that u c − ∈ H S c ( Ω ) u_c^-\in\mathcal{H}_{S_c}(\Omega) u c − ∈ H S c ( Ω ) in the sense of (9) . Moreover, (ii) yields ( u c + , ζ ) L 2 ( Ω c ) = ( u c − ∣ Ω c , ζ ) L 2 ( Ω c ) (u_c^+,\zeta)_{L^2(\Omega_c)}=(u_c^-|_{\Omega_c},\zeta)_{L^2(\Omega_c)} ( u c + , ζ ) L 2 ( Ω c ) = ( u c − ∣ Ω c , ζ ) L 2 ( Ω c ) for every ζ ∈ D Ω c \zeta\in\mathcal{D}_{\Omega_c} ζ ∈ D Ω c . By density of D Ω c \mathcal{D}_{\Omega_c} D Ω c in H Ω c \mathcal{H}_{\Omega_c} H Ω c and u c + ∈ H Ω c u_c^+\in\mathcal{H}_{\Omega_c} u c + ∈ H Ω c , this shows u c + = P H Ω c ( u c − ∣ Ω c ) u_c^+=P_{\mathcal{H}_{\Omega_c}}(u_c^-|_{\Omega_c}) u c + = P H Ω c ( u c − ∣ Ω c ) .
Finally, choose times t j → T c t_j\to T_c t j → T c , in the set with full measure on which the energy inequality holds and u ( t j ) u(t_j) u ( t j ) is rigid on S ( t j ) S(t_j) S ( t j ) , such that E ( t j ) → ess liminf t → T c E ( t ) E(t_j)\to\operatorname*{ess\,liminf}_{t\to T_c}E(t) E ( t j ) → ess liminf t → T c E ( t ) . After extraction, we infer that u ( t j ) ⇀ u ˉ u(t_j)\rightharpoonup\bar{u} u ( t j ) ⇀ u ˉ in L 2 ( Ω ) L^2(\Omega) L 2 ( Ω ) . In view of (iii), we have the orthogonal decomposition
∥ u ˉ ∥ L 2 ( Ω ) 2 = ∥ u c + ∥ L 2 ( Ω c ) 2 + ∥ u c − ∥ L 2 ( S c ) 2 + ∥ u c − ∣ Ω c − u c + ∥ L 2 ( Ω c ) 2 . \|\bar{u}\|_{L^2(\Omega)}^2=\|u_c^+\|_{L^2(\Omega_c)}^2+\|u_c^-\|_{L^2(S_c)}^2+\|u_c^-|_{\Omega_c}-u_c^+\|_{L^2(\Omega_c)}^2. ∥ u ˉ ∥ L 2 ( Ω ) 2 = ∥ u c + ∥ L 2 ( Ω c ) 2 + ∥ u c − ∥ L 2 ( S c ) 2 + ∥ u c − ∣ Ω c − u c + ∥ L 2 ( Ω c ) 2 .
Thus, the weak lower semicontinuity of the norm yields the estimate
∥ u c + ∥ L 2 ( Ω c ) 2 + ∥ u c − ∥ L 2 ( S c ) 2 + ∥ u c − ∣ Ω c − u c + ∥ L 2 ( Ω c ) 2 ≤ lim inf j → ∞ ∥ u ( t j ) ∥ L 2 ( Ω ) 2 . \|u_c^+\|_{L^2(\Omega_c)}^2+\|u_c^-\|_{L^2(S_c)}^2+\|u_c^-|_{\Omega_c}-u_c^+\|_{L^2(\Omega_c)}^2\leq\liminf_{j\to\infty}\|u(t_j)\|_{L^2(\Omega)}^2. ∥ u c + ∥ L 2 ( Ω c ) 2 + ∥ u c − ∥ L 2 ( S c ) 2 + ∥ u c − ∣ Ω c − u c + ∥ L 2 ( Ω c ) 2 ≤ j → ∞ lim inf ∥ u ( t j ) ∥ L 2 ( Ω ) 2 .
Moreover, the norm of the gradient of the part of d ˉ ( t ) \bar{d}(t) d ˉ ( t ) defined on the solid equals ∥ ∇ d 0 ∥ L 2 ( S 0 ) \|\nabla d_0\|_{L^2(S_0)} ∥∇ d 0 ∥ L 2 ( S 0 ) for every t < T c t<T_c t < T c , and the same holds for d c d_c d c on S c S_c S c . The weak convergence from (146) and the weak lower semicontinuity of the H 1 ( Ω ) H^1(\Omega) H 1 ( Ω ) -norm then give
∥ ∇ d c ∥ L 2 ( Ω c ) 2 ≤ lim inf j → ∞ ∥ ∇ d ( t j ) ∥ L 2 ( F ( t j ) ) 2 . \|\nabla d_c\|_{L^2(\Omega_c)}^2\leq\liminf_{j\to\infty}\|\nabla d(t_j)\|_{L^2(F(t_j))}^2. ∥∇ d c ∥ L 2 ( Ω c ) 2 ≤ j → ∞ lim inf ∥∇ d ( t j ) ∥ L 2 ( F ( t j )) 2 .
Combining the last two estimates, we find (148) . Dropping the two nonnegative impact terms yields (147) , finishing the proof.
Completion of the proof of Theorem 2.3 . From § 5.1 , we obtain the solution to the moving-body problem on [ 0 , T c o l ) [0,T_{\mathrm{col}}) [ 0 , T col ) , proving Theorem 2.3 (i). If T c o l = ∞ T_{\mathrm{col}}=\infty T col = ∞ , nothing remains to be shown, so assume T c o l < ∞ T_{\mathrm{col}}<\infty T col < ∞ . Parts (i) and (ii) of Lemma 5.5 provide the rigid body position at contact S c S_c S c , the contact value d c ∈ H 1 ( Ω ; S + 2 ) d_c\in H^1(\Omega;S^2_+) d c ∈ H 1 ( Ω ; S + 2 ) of the material extension, and the post-impact velocity u c + ∈ H Ω c u_c^+\in\mathcal{H}_{\Omega_c} u c + ∈ H Ω c . This proves Theorem 2.3 (ii). With regard to Theorem 2.3 (iii), we apply Proposition 5.3 on Ω c \Omega_c Ω c with initial data ( u c + , d c ∣ Ω c ) (u_c^+,d_c|_{\Omega_c}) ( u c + , d c ∣ Ω c ) , keep the body at S c S_c S c for t ≥ T c o l t\geq T_{\mathrm{col}} t ≥ T col , and extend the post-impact velocity by zero to S c \mathcal{S}_{c} S c . The energy inequality (24) follows from the sharpened inequality in Proposition 2.4, proved below, since J i m p ≥ 0 \mathcal{J}_{\mathrm{imp}} \ge0 J imp ≥ 0 .
Proof of Proposition 2.4. Lemma 5.5 (iii) gives (26) for every u c − \mathbf{u}_{c}^{-} u c − , and the orthogonal decomposition
∥ u c − ∥ L 2 ( Ω ) 2 = ∥ u c + ∥ L 2 ( Ω c ) 2 + ∥ u c − ∥ L 2 ( S c ) 2 + ∥ u c − ∣ Ω c − u c + ∥ L 2 ( Ω c ) 2 \lVert\mathbf{u}_{c}^{-}\rVert_{L^{2}(\Omega)}^{2}=\lVert\mathbf{u}_{c}^{+}\rVert_{L^{2}(\Omega_{c})}^{2}+\lVert\mathbf{u}_{c}^{-}\rVert_{L^{2}(\mathcal{S}_{c})}^{2}+\lVert\mathbf{u}_{c}^{-}|_{\Omega_{c}}-\mathbf{u}_{c}^{+}\rVert_{L^{2}(\Omega_{c})}^{2} ∥ u c − ∥ L 2 ( Ω ) 2 = ∥ u c + ∥ L 2 ( Ω c ) 2 + ∥ u c − ∥ L 2 ( S c ) 2 + ∥ u c − ∣ Ω c − u c + ∥ L 2 ( Ω c ) 2
corresponds precisely to (27) . For the energy inequalities, let D \mathcal{D} D be the dissipation from Theorem 2.3 (iii), and choose t j → T c o l t_{j}\to T_{\mathrm{col}} t j → T col and u c − \mathbf{u}_{c}^{-} u c − as in Lemma 5.5 (iv), so that E ( t j ) → ess liminf t → T c o l E ( t ) E(t_{j})\to\operatorname*{ess\,liminf}_{t\to T_{\mathrm{col}}}E(t) E ( t j ) → ess liminf t → T col E ( t ) and u ( t j ) ⇀ u c − \mathbf{u}(t_{j})\rightharpoonup\mathbf{u}_{c}^{-} u ( t j ) ⇀ u c − . By the projection formula (26) , the two impact terms in (148) add up to J i m p ( S c , u c − ) \mathcal{J}_{\mathrm{imp}}(\mathcal{S}_{c},\mathbf{u}_{c}^{-}) J imp ( S c , u c − ) , and therefore
E Ω c ( u c + , d c ) + J i m p ( S c , u c − ) ≤ lim j → ∞ E ( t j ) = ess liminf t → T c o l E ( t ) , \mathcal{E}_{\Omega_{c}}(\mathbf{u}_{c}^{+},d_{c})+\mathcal{J}_{\mathrm{imp}}(\mathcal{S}_{c},\mathbf{u}_{c}^{-})\leq\lim_{j\to\infty}E(t_{j})=\operatorname*{ess\,liminf}_{t\to T_{\mathrm{col}}}E(t), E Ω c ( u c + , d c ) + J imp ( S c , u c − ) ≤ j → ∞ lim E ( t j ) = t → T col ess liminf E ( t ) ,
which is (28) .
It remains to prove the sharpened global energy inequality. Before contact, the energy inequality (20) at the times t j t_{j} t j reads E ( t j ) + D ( 0 , t j ) ≤ E 0 E(t_{j})+\mathcal{D}(0,t_{j})\leq E_{0} E ( t j ) + D ( 0 , t j ) ≤ E 0 . As j → ∞ j\to\infty j → ∞ , we have E ( t j ) → ess liminf t → T c o l E ( t ) E(t_{j})\to\operatorname*{ess\,liminf}_{t\to T_{\mathrm{col}}}E(t) E ( t j ) → ess liminf t → T col E ( t ) and, by monotone convergence, D ( 0 , t j ) → D ( 0 , T c o l ) \mathcal{D}(0,t_{j})\to\mathcal{D}(0,T_{\mathrm{col}}) D ( 0 , t j ) → D ( 0 , T col ) . Together with (28) , this gives
E Ω c ( u c + , d c ) + J i m p ( S c , u c − ) + D ( 0 , T c o l ) ≤ E 0 . \mathcal{E}_{\Omega_{c}}(\mathbf{u}_{c}^{+},d_{c})+\mathcal{J}_{\mathrm{imp}}(\mathcal{S}_{c},\mathbf{u}_{c}^{-})+\mathcal{D}(0,T_{\mathrm{col}})\leq E_{0}. E Ω c ( u c + , d c ) + J imp ( S c , u c − ) + D ( 0 , T col ) ≤ E 0 .
After contact, the solution satisfies (12) on Ω c \Omega_{c} Ω c with initial energy E Ω c ( u c + , d c ) \mathcal{E}_{\Omega_{c}}(\mathbf{u}_{c}^{+},d_{c}) E Ω c ( u c + , d c ) , that is,
E Ω c ( u ( t ) , d ( t ) ) + D ( 0 , t ) − D ( 0 , T c o l ) ≤ E Ω c ( u c + , d c ) for a.e. t > T c o l , \mathcal{E}_{\Omega_{c}}(\mathbf{u}(t),d(t))+\mathcal{D}(0,t)-\mathcal{D}(0,T_{\mathrm{col}})\leq\mathcal{E}_{\Omega_{c}}(\mathbf{u}_{c}^{+},d_{c}) \qquad\text{for a.e. }t>T_{\mathrm{col}}, E Ω c ( u ( t ) , d ( t )) + D ( 0 , t ) − D ( 0 , T col ) ≤ E Ω c ( u c + , d c ) for a.e. t > T col ,
since D ( 0 , t ) − D ( 0 , T c o l ) \mathcal{D}(0,t)-\mathcal{D}(0,T_{\mathrm{col}}) D ( 0 , t ) − D ( 0 , T col ) is the dissipation after contact. Adding the two inequalities yields (24) with J i m p ( S c , u c − ) \mathcal{J}_{\mathrm{imp}}(\mathcal{S}_{c},\mathbf{u}_{c}^{-}) J imp ( S c , u c − ) added to its left-hand side.
Acknowledgements. F.B. acknowledges the support by the German National Academy of Sciences Leopoldina through the Leopoldina Fellowship Program with grant number LPDS 2024-07. V.N. acknowledges partial support from an AMS–Simons travel grant.
Appendix A. Proofs of auxiliary results
Approximation of the director fields
The lemma below asserts that an H 1 H^{1} H 1 -director field with values in S + 2 \mathbb{S}_{+}^{2} S + 2 can be approximated by smooth director fields with values in S + 2 \mathbb{S}_{+}^{2} S + 2 .
Lemma A.1. Let Ω ⊂ R 3 \Omega\subset\mathbb{R}^{3} Ω ⊂ R 3 be a bounded C 2 C^{2} C 2 -domain, and let d ∈ H 1 ( Ω ; S + 2 ) \mathbf{d}\in H^{1}(\Omega;\mathbb{S}_{+}^{2}) d ∈ H 1 ( Ω ; S + 2 ) . Then there are maps d j ∈ C ∞ ( Ω ‾ ; S + 2 ) ⊂ H 2 ( Ω ; S + 2 ) \mathbf{d}_{j}\in C^{\infty}(\overline{\Omega};\mathbb{S}_{+}^{2})\subset H^{2}(\Omega;\mathbb{S}_{+}^{2}) d j ∈ C ∞ ( Ω ; S + 2 ) ⊂ H 2 ( Ω ; S + 2 ) such that d j → d \mathbf{d}_{j}\to\mathbf{d} d j → d strongly in H 1 ( Ω ; R 3 ) H^{1}(\Omega;\mathbb{R}^{3}) H 1 ( Ω ; R 3 ) . The assertion remains valid under fixed rotations of the target hemisphere.
Proof. It is enough to consider S + 2 \mathbb{S}_{+}^{2} S + 2 . To this end, we write d = ( p , q ) \mathbf{d}=(\mathbf{p},q) d = ( p , q ) with p = ( d 1 , d 2 ) \mathbf{p}=(d_{1},d_{2}) p = ( d 1 , d 2 ) and q = d 3 ≥ 0 q=d_{3}\geq0 q = d 3 ≥ 0 . Then ∣ p ∣ 2 + q 2 = 1 \lvert\mathbf{p}\rvert^{2}+q^{2}=1 ∣ p ∣ 2 + q 2 = 1 and, a.e., we get p ⋅ ∇ p + q ∇ q = 0 \mathbf{p}\cdot\nabla\mathbf{p}+q\nabla q=0 p ⋅ ∇ p + q ∇ q = 0 . For 0 < κ < 1 0<\kappa<1 0 < κ < 1 , we set p κ = ( 1 − κ ) p \mathbf{p}^{\kappa}=(1-\kappa)\mathbf{p} p κ = ( 1 − κ ) p , q κ = 1 − ∣ p κ ∣ 2 q^{\kappa}=\sqrt{1-\lvert\mathbf{p}^{\kappa}\rvert^{2}} q κ = 1 − ∣ p κ ∣ 2 , and d κ = ( p κ , q κ ) \mathbf{d}^{\kappa}=(\mathbf{p}^{\kappa},q^{\kappa}) d κ = ( p κ , q κ ) , so we obtain q κ ≥ 1 − ( 1 − κ ) 2 > 0 q^{\kappa}\geq\sqrt{1-(1-\kappa)^{2}}>0 q κ ≥ 1 − ( 1 − κ ) 2 > 0 as well as
q κ = 1 − ( 1 − κ ) 2 ∣ p ∣ 2 . q^{\kappa}=\sqrt{1-(1-\kappa)^{2}\lvert\mathbf{p}\rvert^{2}}. q κ = 1 − ( 1 − κ ) 2 ∣ p ∣ 2 .
∇ q κ = ( 1 − κ ) 2 q q 2 + ( 2 κ − κ 2 ) ∣ p ∣ 2 ∇ q . \nabla q^{\kappa}=\frac{(1-\kappa)^{2}q}{\sqrt{q^{2}+(2\kappa-\kappa^{2})\lvert\mathbf{p}\rvert^{2}}}\nabla q. ∇ q κ = q 2 + ( 2 κ − κ 2 ) ∣ p ∣ 2 ( 1 − κ ) 2 q ∇ q .
Note that the coefficient of ∇ q \nabla q ∇ q is bounded by one and converges to one wherever q > 0 q>0 q > 0 . On the other hand, on the set where the nonnegative Sobolev function q q q vanishes, one has ∇ q = 0 \nabla q=0 ∇ q = 0 a.e. From dominated convergence, it then follows that d κ → d \mathbf{d}^{\kappa}\to\mathbf{d} d κ → d strongly in H 1 ( Ω ) H^{1}(\Omega) H 1 ( Ω ) as κ → 0 \kappa\to0 κ → 0 .
Now we fix κ \kappa κ , extend p κ \mathbf{p}^{\kappa} p κ to an H 1 ( R 3 ; R 2 ) H^{1}(\mathbb{R}^{3};\mathbb{R}^{2}) H 1 ( R 3 ; R 2 ) -map, and compose the extension with the metric projection onto the closed disk of radius 1 − κ / 2 1-\kappa/2 1 − κ /2 . Note that this does not change the map on Ω \Omega Ω . By mollification, we get smooth maps p κ , m \mathbf{p}_{\kappa,m} p κ , m taking values in that same convex disk and converging to p κ \mathbf{p}^{\kappa} p κ in H 1 ( Ω ) H^{1}(\Omega) H 1 ( Ω ) . From the smoothness of the map z ↦ ( z , 1 − ∣ z ∣ 2 ) z\mapsto(z,\sqrt{1-\lvert z\rvert^{2}}) z ↦ ( z , 1 − ∣ z ∣ 2 ) with bounded derivative on { ∣ z ∣ ≤ 1 − κ / 2 } \{\lvert z\rvert\leq1-\kappa/2\} {∣ z ∣ ≤ 1 − κ /2 } , we infer the strong convergence of d κ , m \mathbf{d}_{\kappa,m} d κ , m to d κ \mathbf{d}^{\kappa} d κ in H 1 ( Ω ) H^{1}(\Omega) H 1 ( Ω ) . Choosing a suitable diagonal sequence with κ → 0 \kappa\to0 κ → 0 , m → ∞ m\to\infty m → ∞ proves the claim.
Test-space approximation and material gluing
We first prove density of T u c o l \mathcal{T}^{\mathrm{col}}_u T u col as introduced in (119) in C T u C\mathcal{T}_u C T u
Lemma A.2. Assume that the rigid body S ( t ) = h ( t ) + O ( t ) S 0 S(t)=h(t)+O(t)S_0 S ( t ) = h ( t ) + O ( t ) S 0 maintains a positive distance from ∂ Ω \partial\Omega ∂ Ω , where h h h and O O O belong to W 1 , ∞ ( 0 , T ) W^{1,\infty}(0,T) W 1 , ∞ ( 0 , T ) and ∂ S 0 \partial S_0 ∂ S 0 is of class C 2 C^2 C 2 . Then for every ζ ∈ T u \zeta\in\mathcal{T}_u ζ ∈ T u there are r k → 0 r_k\to0 r k → 0 and ζ k ∈ T u c o l \zeta_k\in\mathcal{T}^{\mathrm{col}}_u ζ k ∈ T u col , with D ( ζ k ) = 0 \mathbb{D}(\zeta_k)=0 D ( ζ k ) = 0 on S ( t ) r k S(t)_{r_k} S ( t ) r k , such that
∥ ζ k − ζ ∥ L ∞ ( 0 , T ; W 1 , ∞ ( Ω ) ) + ∥ ∂ t ( ζ k − ζ ) ∥ L 1 ( 0 , T ; L 2 ( Ω ) ) → 0. (149) \|\zeta_k-\zeta\|_{L^\infty(0,T;W^{1,\infty}(\Omega))}+\|\partial_t(\zeta_k-\zeta)\|_{L^1(0,T;L^2(\Omega))}\to0. \tag*{(149)} ∥ ζ k − ζ ∥ L ∞ ( 0 , T ; W 1 , ∞ ( Ω )) + ∥ ∂ t ( ζ k − ζ ) ∥ L 1 ( 0 , T ; L 2 ( Ω )) → 0. (149)
Proof of Lemma A.2. We start by fixing ζ ∈ T u \zeta\in\mathcal{T}_u ζ ∈ T u . Since S ( t ) S(t) S ( t ) is connected and D ( ζ ( t ) ) = 0 \mathbb{D}(\zeta(t))=0 D ( ζ ( t )) = 0 there, it follows that there exists a unique affine rigid field R ( t , x ) = a ( t ) + B ( t ) x R(t,x)=a(t)+B(t)x R ( t , x ) = a ( t ) + B ( t ) x , with B ( t ) T = − B ( t ) B(t)^\mathsf{T}=-B(t) B ( t ) T = − B ( t ) , and which agrees with ζ ( t ) \zeta(t) ζ ( t ) on S ( t ) S(t) S ( t ) . The coefficients a ( t ) a(t) a ( t ) and B ( t ) B(t) B ( t ) can be written by the mean and angular-momentum formulas on S ( t ) S(t) S ( t ) , showing that they belong to W 1 , ∞ ( 0 , T ) W^{1,\infty}(0,T) W 1 , ∞ ( 0 , T ) . We now set q = ζ − R q=\zeta-R q = ζ − R . Then q = 0 q=0 q = 0 on the open set S ( t ) S(t) S ( t ) . By the smoothness of ζ \zeta ζ in space, every first spatial derivative of q q q has zero trace on ∂ S ( t ) \partial S(t) ∂ S ( t ) . For d t ( x ) : = dist ( x , S ( t ) ) d_t(x):=\operatorname{dist}(x,S(t)) d t ( x ) := dist ( x , S ( t )) and d t ( x ) < r 0 d_t(x)<r_0 d t ( x ) < r 0 , it then follows that
∣ q ( t , x ) ∣ ≤ C d t ( x ) 2 , ∣ ∇ q ( t , x ) ∣ ≤ C d t ( x ) . (150) |q(t,x)|\leq C d_t(x)^2,\qquad|\nabla q(t,x)|\leq C d_t(x). \tag*{(150)} ∣ q ( t , x ) ∣ ≤ C d t ( x ) 2 , ∣∇ q ( t , x ) ∣ ≤ C d t ( x ) . (150)
Differentiating the identity q ( t , h ( t ) + O ( t ) y ) = 0 q(t,h(t)+O(t)y)=0 q ( t , h ( t ) + O ( t ) y ) = 0 for y ∈ ∂ S 0 y\in\partial S_0 y ∈ ∂ S 0 at a.e. time and using ∇ q = 0 \nabla q=0 ∇ q = 0 on the interface, for a.e. t t t , we find
∣ ∂ t q ( t , x ) ∣ ≤ C d t ( x ) (151) |\partial_tq(t,x)|\leq C d_t(x) \tag*{(151)} ∣ ∂ t q ( t , x ) ∣ ≤ C d t ( x ) (151)
in the same tubular neighborhood. Note that the rigidity of the motion and its Lipschitz regularity yield that the constants are uniform in time.
Next, we choose a smooth cutoff ϑ ∈ C ∞ ( R ) \vartheta\in C^\infty(\mathbb{R}) ϑ ∈ C ∞ ( R ) with ϑ = 0 \vartheta=0 ϑ = 0 on ( − ∞ , 1 ] (-\infty,1] ( − ∞ , 1 ] and ϑ = 1 \vartheta=1 ϑ = 1 on [ 2 , ∞ ) [2,\infty) [ 2 , ∞ ) , and we set ϑ r ( t , x ) = ϑ ( d t ( x ) / r ) \vartheta_r(t,x)=\vartheta(d_t(x)/r) ϑ r ( t , x ) = ϑ ( d t ( x ) / r ) . The field ζ r 0 = R + ϑ r q \zeta^0_r=R+\vartheta_rq ζ r 0 = R + ϑ r q equals R R R on S ( t ) r S(t)_r S ( t ) r and equals ζ \zeta ζ outside S ( t ) 2 r S(t)_{2r} S ( t ) 2 r , but div ζ r 0 = f r : = ∇ ϑ r ⋅ q \operatorname{div}\zeta^0_r=f_r:=\nabla\vartheta_r\cdot q div ζ r 0 = f r := ∇ ϑ r ⋅ q . The defect is supported in the tubular annulus A r ( t ) = S ( t ) 2 r ∖ S ( t ) r A_r(t)=S(t)_{2r}\setminus S(t)_r A r ( t ) = S ( t ) 2 r ∖ S ( t ) r . From (A.2), we deduce that ∥ f r ∥ L ∞ ( ( 0 , T ) × Ω ) ≤ C r \|f_r\|_{L^\infty((0,T)\times\Omega)}\leq Cr ∥ f r ∥ L ∞ (( 0 , T ) × Ω ) ≤ C r and ∫ A r ( t ) f r ( t ) d x = 0 \int_{A_r(t)}f_r(t)\,dx=0 ∫ A r ( t ) f r ( t ) d x = 0 , where we note that the second identity follows from the divergence theorem. The cutoff vanishes on the inner parallel boundary, while the flux of q q q through the outer parallel boundary equals its flux through ∂ S ( t ) \partial S(t) ∂ S ( t ) and is therefore zero because div q = 0 \operatorname{div}q=0 div q = 0 and q = 0 q=0 q = 0 on ∂ S ( t ) \partial S(t) ∂ S ( t ) .
Fix r 0 > 0 r_0>0 r 0 > 0 smaller than the uniform tubular radius and the distance from the moving body to ∂ Ω \partial\Omega ∂ Ω . Instead of solving the divergence equation on the shrinking annulus, solve it on the collar with fixed-width collar A r ∗ ( t ) : = S ( t ) r 0 ∖ S ( t ) r A_r^*(t):=S(t)_{r_0}\setminus S(t)_r A r ∗ ( t ) := S ( t ) r 0 ∖ S ( t ) r . Then the defect f r f_r f r is supported in A r ( t ) ⊂ A r ∗ ( t ) A_r(t)\subset A_r^*(t) A r ( t ) ⊂ A r ∗ ( t ) , and its mean over A r ∗ ( t ) A_r^*(t) A r ∗ ( t ) is zero by the divergence theorem. The estimates (A.2) imply, for every fixed p ∈ ( 3 , ∞ ) p\in(3,\infty) p ∈ ( 3 , ∞ ) , the bounds ∥ f r ( t ) ∥ L ∞ ( A r ∗ ( t ) ) ≤ C r \|f_r(t)\|_{L^\infty(A_r^*(t))}\leq Cr ∥ f r ( t ) ∥ L ∞ ( A r ∗ ( t )) ≤ C r and ∥ f r ( t ) ∥ W 1 , p ( A r ∗ ( t ) ) ≤ C r 1 / p \|f_r(t)\|_{W^{1,p}(A_r^*(t))}\leq Cr^{1/p} ∥ f r ( t ) ∥ W 1 , p ( A r ∗ ( t )) ≤ C r 1/ p , since first derivatives of f r = ∇ ϑ r ⋅ q f_r=\nabla\vartheta_r\cdot q f r = ∇ ϑ r ⋅ q are O ( 1 ) O(1) O ( 1 ) and are supported in a layer of volume O ( r ) O(r) O ( r ) .
We observe that the collars A r ∗ ( t ) A_r^*(t) A r ∗ ( t ) , 0 < r < r 0 / 4 0<r<r_0/4 0 < r < r 0 /4 , form a uniformly C 2 C^2 C 2 -family. After undoing the rigid motion, their inner and outer boundaries are parallel surfaces of ∂ S 0 \partial S_0 ∂ S 0 at offsets in [ 0 , r 0 ] [0,r_0] [ 0 , r 0 ] , with r 0 r_0 r 0 fixed. The Bogovskiĭ construction on a finite C 2 C^2 C 2 -atlas then gives operators B r ( t ) \mathcal{B}_r(t) B r ( t ) with constants independent of r r r and t t t such that div B r ( t ) f r = f r \operatorname{div}\mathcal{B}_r(t)f_r=f_r div B r ( t ) f r = f r , B r ( t ) f r ∈ W 2 , p ( A r ∗ ( t ) ; R 3 ) ∩ W 0 1 , p ( A r ∗ ( t ) ; R 3 ) \mathcal{B}_r(t)f_r\in W^{2,p}(A_r^*(t);\mathbb{R}^3)\cap W^{1,p}_0(A_r^*(t);\mathbb{R}^3) B r ( t ) f r ∈ W 2 , p ( A r ∗ ( t ) ; R 3 ) ∩ W 0 1 , p ( A r ∗ ( t ) ; R 3 ) , and
∥ B r f r ∥ W 2 , p ( A r ∗ ( t ) ) ≤ C ∥ f r ∥ W 1 , p ( A r ∗ ( t ) ) ≤ C r 1 / p , (152) \|\mathcal{B}_rf_r\|_{W^{2,p}(A_r^*(t))}\leq C\|f_r\|_{W^{1,p}(A_r^*(t))}\leq Cr^{1/p}, \tag*{(152)} ∥ B r f r ∥ W 2 , p ( A r ∗ ( t )) ≤ C ∥ f r ∥ W 1 , p ( A r ∗ ( t )) ≤ C r 1/ p , (152)
where we refer to [5 , 21 ] for the Bogovskiĭ operator and its higher-order estimates. Due to p > 3 p>3 p > 3 , Sobolev embedding yields ∥ B r f r ∥ W 1 , ∞ ( A r ∗ ( t ) ) ≤ C r 1 / p → 0 \|\mathcal{B}_rf_r\|_{W^{1,\infty}(A_r^*(t))}\leq Cr^{1/p}\to0 ∥ B r f r ∥ W 1 , ∞ ( A r ∗ ( t )) ≤ C r 1/ p → 0 . We then extend the correction by zero outside A r ∗ ( t ) A_r^*(t) A r ∗ ( t ) . As its trace vanishes on the inner boundary, this extension does not alter the rigid field on S ( t ) r S(t)_r S ( t ) r .
The time derivative is estimated separately. A differentiation of f r f_r f r produces only O ( 1 ) O(1) O ( 1 ) terms on the O ( r ) O(r) O ( r ) transition layer, by (A.2) and (A.3). Moreover, we have ∇ f r = O ( 1 ) \nabla f_r=O(1) ∇ f r = O ( 1 ) on the same layer, so for a.e. t t t , we get ∥ ∂ t f r ( t ) ∥ L 2 ( Ω ) + ∥ ∇ f r ( t ) ∥ L 2 ( Ω ) ≤ C r 1 / 2 \|\partial_tf_r(t)\|_{L^2(\Omega)}+\|\nabla f_r(t)\|_{L^2(\Omega)}\leq Cr^{1/2} ∥ ∂ t f r ( t ) ∥ L 2 ( Ω ) + ∥∇ f r ( t ) ∥ L 2 ( Ω ) ≤ C r 1/2 .
For a uniform control of the time derivative, we write the Bogovskiĭ correction on a fixed reference collar. To this end, let Z t ( y ) : = h ( t ) + O ( t ) y Z_t(y):=h(t)+O(t)y Z t ( y ) := h ( t ) + O ( t ) y , A r ∗ , 0 : = S 0 , r 0 ∖ S 0 , r A_{r}^{*,0}:=S_{0,r_0}\setminus S_{0,r} A r ∗ , 0 := S 0 , r 0 ∖ S 0 , r , and invoke the Bogovskiĭ operator B r 0 \mathcal{B}_r^0 B r 0 on A r ∗ , 0 A_r^{*,0} A r ∗ , 0 with the uniform bounds above. We set f ~ r ( t , y ) : = f r ( t , Z t ( y ) ) \widetilde f_r(t,y):=f_r(t,Z_t(y)) f r ( t , y ) := f r ( t , Z t ( y )) as well as w r ( t , Z t ( y ) ) : = O ( t ) B r 0 f ~ r ( t , y ) w_r(t,Z_t(y)):=O(t)B_r^0\widetilde f_r(t,y) w r ( t , Z t ( y )) := O ( t ) B r 0 f r ( t , y ) . Since the divergence is preserved under rotation, we get div x w r = f r \operatorname{div}_x w_r=f_r div x w r = f r . With V R ( t , x ) : = h ′ ( t ) + O ′ ( t ) O ( t ) ⊤ ( x − h ( t ) ) V_R(t,x):=h'(t)+O'(t)O(t)^\top(x-h(t)) V R ( t , x ) := h ′ ( t ) + O ′ ( t ) O ( t ) ⊤ ( x − h ( t )) , it then follows that
∂ t f ~ r = ( ∂ t f r + V R ⋅ ∇ f r ) ∘ Z t . \partial_t\widetilde f_r=\left(\partial_t f_r+V_R\cdot\nabla f_r\right)\circ Z_t. ∂ t f r = ( ∂ t f r + V R ⋅ ∇ f r ) ∘ Z t .
The rigid velocity V R V_R V R is uniformly bounded on the fixed collar, and Z t Z_t Z t maps A r ∗ , 0 A_r^{*,0} A r ∗ , 0 isometrically onto A r ∗ ( t ) A_r^*(t) A r ∗ ( t ) . Therefore
∥ ∂ t f ~ r ( t ) ∥ L 2 ( A r ∗ , 0 ) = ∥ ( ∂ t f r + V R ⋅ ∇ f r ) ( t ) ∥ L 2 ( A r ∗ ( t ) ) ≤ C r 1 / 2 . \|\partial_t\widetilde f_r(t)\|_{L^2(A_r^{*,0})}=\|(\partial_t f_r+V_R\cdot\nabla f_r)(t)\|_{L^2(A_r^*(t))}\le Cr^{1/2}. ∥ ∂ t f r ( t ) ∥ L 2 ( A r ∗ , 0 ) = ∥ ( ∂ t f r + V R ⋅ ∇ f r ) ( t ) ∥ L 2 ( A r ∗ ( t )) ≤ C r 1/2 .
Differentiating the formula for w r w_r w r at fixed x x x gives a rotation term, a term containing B r 0 ∂ t f ~ r B_r^0\partial_t\widetilde f_r B r 0 ∂ t f r , and the Eulerian transport term containing ∇ B r 0 f ~ r \nabla B_r^0\widetilde f_r ∇ B r 0 f r . The uniform L 2 → H 0 1 L^2\to H_0^1 L 2 → H 0 1 -estimate for the Bogovskiĭ operators B r 0 B_r^0 B r 0 on the reference collars, see [5 ] and [21 Chap. III.3] , thus gives ∥ ∂ t w r ( t ) ∥ L 2 ( Ω ) ≤ C r 1 / 2 \|\partial_t w_r(t)\|_{L^2(\Omega)}\le Cr^{1/2} ∥ ∂ t w r ( t ) ∥ L 2 ( Ω ) ≤ C r 1/2 , and we may take B r ( t ) f r = w r B_r(t)f_r=w_r B r ( t ) f r = w r . In particular, the time estimate does not rely on differentiating an abstract operator on the moving collar.
Define ζ ^ r : = ζ r 0 − B r ( t ) f r \widehat\zeta_r:=\zeta_r^0-B_r(t)f_r ζ r := ζ r 0 − B r ( t ) f r . Then div ζ ^ r = 0 \operatorname{div}\widehat\zeta_r=0 div ζ r = 0 and it is rigid on S ( t ) r S(t)_r S ( t ) r . The Bogovskiĭ correction is supported in the fixed collar S ( t ) r 0 S(t)_{r_0} S ( t ) r 0 and tends to zero there in W 1 , ∞ W^{1,\infty} W 1 , ∞ . It need not vanish on all of S ( t ) r 0 ∖ S ( t ) 2 r S(t)_{r_0}\setminus S(t)_{2r} S ( t ) r 0 ∖ S ( t ) 2 r . From the estimates (A.2)–(A.4), it follows that
∥ ζ ^ r − ζ ∥ L ∞ ( 0 , T ; W 1 , ∞ ( Ω ) ) → 0 , ∥ ∂ t ( ζ ^ r − ζ ) ∥ L 1 ( 0 , T ; L 2 ( Ω ) ) → 0 , (153) \|\widehat\zeta_r-\zeta\|_{L^\infty(0,T;W^{1,\infty}(\Omega))}\to0,\qquad\|\partial_t(\widehat\zeta_r-\zeta)\|_{L^1(0,T;L^2(\Omega))}\to0, \tag*{(153)} ∥ ζ r − ζ ∥ L ∞ ( 0 , T ; W 1 , ∞ ( Ω )) → 0 , ∥ ∂ t ( ζ r − ζ ) ∥ L 1 ( 0 , T ; L 2 ( Ω )) → 0 , (153)
where all fields are defined on the whole container Ω \Omega Ω . The field ζ ^ r \widehat\zeta_r ζ r is not necessarily smooth yet because the tubular coordinates inherit only the C 2 C^2 C 2 -regularity of ∂ S 0 \partial S_0 ∂ S 0 . We note that ζ ^ r \widehat\zeta_r ζ r coincides with the affine rigid field R ( t ) R(t) R ( t ) on S ( t ) r S(t)_r S ( t ) r and has compact spatial support that has a fixed distance from ∂ Ω \partial\Omega ∂ Ω . We then extend it by zero to R 3 \mathbb{R}^3 R 3 and convolve in space with a symmetric mollifier of radius ρ r = o ( r ) \rho_r=o(r) ρ r = o ( r ) , where we invoke that convolution preserves the divergence-free condition. Moreover, on S ( t ) r / 2 S(t)_{r/2} S ( t ) r /2 the convolution samples only the affine field R ( t ) R(t) R ( t ) . Symmetry of the mollifier preserves affine functions, so the spatially mollified field is still rigid there. The estimates in (A.5) are preserved. Indeed, writing η ρ r \eta_{\rho_r} η ρ r for the spatial mollifier,
∥ η ρ r ∗ ζ ^ r − ζ ∥ W 1 , ∞ ( Ω ) ≤ ∥ ζ ^ r − ζ ∥ W 1 , ∞ ( Ω ) + ∥ η ρ r ∗ ζ − ζ ∥ W 1 , ∞ ( Ω ) , \|\eta_{\rho_r}*\widehat\zeta_r-\zeta\|_{W^{1,\infty}(\Omega)}\le\|\widehat\zeta_r-\zeta\|_{W^{1,\infty}(\Omega)}+\|\eta_{\rho_r}*\zeta-\zeta\|_{W^{1,\infty}(\Omega)}, ∥ η ρ r ∗ ζ r − ζ ∥ W 1 , ∞ ( Ω ) ≤ ∥ ζ r − ζ ∥ W 1 , ∞ ( Ω ) + ∥ η ρ r ∗ ζ − ζ ∥ W 1 , ∞ ( Ω ) ,
and the second term tends to zero because ζ \zeta ζ is spatially smooth. The same contraction argument in L t 1 L x 2 L_t^1L_x^2 L t 1 L x 2 , applied after differentiating in time, gives the second convergence in (A.5).
At this stage the time regularity is only W 1 , ∞ W^{1,\infty} W 1 , ∞ because the rigid trajectory has that regularity. We then extend the fields slightly beyond the temporal endpoints and use mollification in time on a second scale θ r = o ( r ) \theta_r=o(r) θ r = o ( r ) . From the Lipschitz continuity of S ( t ) S(t) S ( t ) , we deduce that S ( t ) r / 4 ⊂ S ( s ) r / 2 S(t)_{r/4}\subset S(s)_{r/2} S ( t ) r /4 ⊂ S ( s ) r /2 whenever ∣ t − s ∣ ≤ c r |t-s|\le cr ∣ t − s ∣ ≤ cr . Note that the time-mollified field remains rigid on S ( t ) r / 4 S(t)_{r/4} S ( t ) r /4 . A final time cutoff, chosen in a region where the original test already vanishes, produces ζ r ∈ T u c o l \zeta_r\in\mathcal{T}_u^{\mathrm{col}} ζ r ∈ T u col , and (A.5) is preserved. A choice of r = r k → 0 r=r_k\to0 r = r k → 0 proves (A.1) and thus completes the proof. □ \square □
The density result has a formulation independent of the liquid-crystal equations.
Proposition A.3. Let S ( t ) S(t) S ( t ) satisfy the hypotheses of Lemma A.2 . Suppose u u u is solenoidal and satisfies u ∈ L ∞ ( 0 , T ; L 2 ( Ω ; R 3 ) ) ∩ L 2 ( 0 , T ; H 0 1 ( Ω ; R 3 ) ) u\in L^\infty(0,T;L^2(\Omega;\mathbb{R}^3))\cap L^2(0,T;H_0^1(\Omega;\mathbb{R}^3)) u ∈ L ∞ ( 0 , T ; L 2 ( Ω ; R 3 )) ∩ L 2 ( 0 , T ; H 0 1 ( Ω ; R 3 )) as well as Λ ∈ L 1 ( ( 0 , T ) × Ω ; R 3 × 3 ) \Lambda\in L^1((0,T)\times\Omega;\mathbb{R}^{3\times3}) Λ ∈ L 1 (( 0 , T ) × Ω ; R 3 × 3 ) . Assume that the momentum identity
∫ 0 T ∫ Ω [ − u ⋅ ∂ t ζ − ( u ⊗ u ) : ∇ ζ + 2 D ( u ) : D ( ζ ) ] d x d t = ∫ Ω u 0 ⋅ ζ ( 0 ) d x + ∫ 0 T ∫ Ω Λ : ∇ ζ d x d t \int_0^T\int_\Omega\left[-u\cdot\partial_t\zeta-(u\otimes u):\nabla\zeta+2\mathbb{D}(u):\mathbb{D}(\zeta)\right]\mathrm{d}x\,\mathrm{d}t=\int_\Omega u_0\cdot\zeta(0)\,\mathrm{d}x+\int_0^T\int_\Omega\Lambda:\nabla\zeta\,\mathrm{d}x\,\mathrm{d}t ∫ 0 T ∫ Ω [ − u ⋅ ∂ t ζ − ( u ⊗ u ) : ∇ ζ + 2 D ( u ) : D ( ζ ) ] d x d t = ∫ Ω u 0 ⋅ ζ ( 0 ) d x + ∫ 0 T ∫ Ω Λ : ∇ ζ d x d t holds for every ζ ∈ T u c o l \zeta\in\mathcal{T}_u^{\mathrm{col}} ζ ∈ T u col . Then it holds for every ζ ∈ T u \zeta\in\mathcal{T}_u ζ ∈ T u .
Proof of Proposition A.3 . Let ζ k \zeta_k ζ k be the sequence from Lemma A.2 . The convergence (A.1) gives
∣ ∫ 0 T ∫ Ω u ⋅ ∂ t ( ζ k − ζ ) d x d t ∣ ≤ ∥ u ∥ L t ∞ L x 2 ∥ ∂ t ( ζ k − ζ ) ∥ L t 1 L x 2 → 0. \left|\int_0^T\int_\Omega u\cdot\partial_t(\zeta_k-\zeta)\,\mathrm{d}x\,\mathrm{d}t\right|\le\|u\|_{L^\infty_tL_x^2}\|\partial_t(\zeta_k-\zeta)\|_{L_t^1L_x^2}\to0. ∫ 0 T ∫ Ω u ⋅ ∂ t ( ζ k − ζ ) d x d t ≤ ∥ u ∥ L t ∞ L x 2 ∥ ∂ t ( ζ k − ζ ) ∥ L t 1 L x 2 → 0.
Similarly,
∣ ∫ 0 T ∫ Ω ( u ⊗ u ) : ∇ ( ζ k − ζ ) d x d t ∣ ≤ T ∥ u ∥ L t ∞ L x 2 2 ∥ ∇ ( ζ k − ζ ) ∥ L t , x ∞ → 0 , \left|\int_0^T\int_\Omega(\mathbf{u}\otimes\mathbf{u}):\nabla(\zeta_k-\zeta)\,\mathrm{d}x\,\mathrm{d}t\right|
\leq T\|\mathbf{u}\|_{L^\infty_t L^2_x}^2\|\nabla(\zeta_k-\zeta)\|_{L^\infty_{t,x}}\to0, ∫ 0 T ∫ Ω ( u ⊗ u ) : ∇ ( ζ k − ζ ) d x d t ≤ T ∥ u ∥ L t ∞ L x 2 2 ∥∇ ( ζ k − ζ ) ∥ L t , x ∞ → 0 ,
and the viscous term tends to zero by Cauchy–Schwarz and the uniform L t ∞ W x 1 , ∞ L^\infty_tW^{1,\infty}_x L t ∞ W x 1 , ∞ -convergence. The stress term satisfies
∣ ∫ 0 T ∫ Ω A : ∇ ( ζ k − ζ ) d x d t ∣ ≤ ∥ A ∥ L 1 ( ( 0 , T ) × Ω ) ∥ ∇ ( ζ k − ζ ) ∥ L ∞ ( ( 0 , T ) × Ω ) → 0. \left|\int_0^T\int_\Omega\mathbb{A}:\nabla(\zeta_k-\zeta)\,\mathrm{d}x\,\mathrm{d}t\right|
\leq\|\mathbb{A}\|_{L^1((0,T)\times\Omega)}\|\nabla(\zeta_k-\zeta)\|_{L^\infty((0,T)\times\Omega)}\to0. ∫ 0 T ∫ Ω A : ∇ ( ζ k − ζ ) d x d t ≤ ∥ A ∥ L 1 (( 0 , T ) × Ω ) ∥∇ ( ζ k − ζ ) ∥ L ∞ (( 0 , T ) × Ω ) → 0.
The initial term converges uniformly in space. Passing k → ∞ k\to\infty k → ∞ proves the claim. □
The next lemma gives the transport identity for the material extension and shows that no distribution supported on the moving interface is produced.
Lemma A.4. Let S ( t ) = h ( t ) + O ( t ) S 0 ⋐ Ω S(t)=h(t)+O(t)S_0\Subset\Omega S ( t ) = h ( t ) + O ( t ) S 0 ⋐ Ω be a Lipschitz rigid trajectory and F ( t ) = Ω ∖ S ( t ) ‾ F(t)=\Omega\setminus\overline{S(t)} F ( t ) = Ω ∖ S ( t ) . Let d 0 ∈ H 1 ( Ω ; R 3 ) d_0\in H^1(\Omega;\mathbb{R}^3) d 0 ∈ H 1 ( Ω ; R 3 ) , and let u ∈ L 2 ( 0 , T ; H 1 ( Ω ; R 3 ) ) \mathbf{u}\in L^2(0,T;H^1(\Omega;\mathbb{R}^3)) u ∈ L 2 ( 0 , T ; H 1 ( Ω ; R 3 )) be solenoidal and equal in S ( t ) S(t) S ( t ) to the rigid velocity generating the trajectory. Suppose d F ∈ L ∞ ( 0 , T ; H 1 ( F ( t ) ; R 3 ) ) d_F\in L^\infty(0,T;H^1(F(t);\mathbb{R}^3)) d F ∈ L ∞ ( 0 , T ; H 1 ( F ( t ) ; R 3 )) has trace
Tr ∂ S ( t ) d F ( t , x ) = d 0 ( O ( t ) T ( x − h ( t ) ) ) , \operatorname{Tr}_{\partial S(t)}d_F(t,x)=d_0(O(t)^\mathsf{T}(x-h(t))), Tr ∂ S ( t ) d F ( t , x ) = d 0 ( O ( t ) T ( x − h ( t ))) , and satisfies ∂ t d F + ( u ⋅ ∇ ) d F = M \partial_td_F+(\mathbf{u}\cdot\nabla)d_F=M ∂ t d F + ( u ⋅ ∇ ) d F = M in the moving fluid region, with M ∈ L 2 ( Q F ) M\in L^2(Q_F) M ∈ L 2 ( Q F ) , in the energy weak formulation. Define
d S ( t , x ) : = d 0 ( O ( t ) T ( x − h ( t ) ) ) ( x ∈ S ( t ) ) , d ‾ : = d F 1 F ( t ) + d S 1 S ( t ) , d_S(t,x):=d_0(O(t)^\mathsf{T}(x-h(t)))\quad(x\in S(t)),\qquad\overline{d}:=d_F\mathbf{1}_{F(t)}+d_S\mathbf{1}_{S(t)}, d S ( t , x ) := d 0 ( O ( t ) T ( x − h ( t ))) ( x ∈ S ( t )) , d := d F 1 F ( t ) + d S 1 S ( t ) , and let M ~ \widetilde{M} M be the zero extension of M M M through the solid. Then d ‾ ( t ) ∈ H 1 ( Ω ; R 3 ) \overline{d}(t)\in H^1(\Omega;\mathbb{R}^3) d ( t ) ∈ H 1 ( Ω ; R 3 ) for a.e. t t t and
∂ t d ‾ + ( u ⋅ ∇ ) d ‾ = M ~ in D ′ ( ( 0 , T ) × Ω ) . (154) \partial_t\overline{d}+(\mathbf{u}\cdot\nabla)\overline{d}=\widetilde{M}\quad\text{in }\mathcal{D}'((0,T)\times\Omega). \tag*{(154)} ∂ t d + ( u ⋅ ∇ ) d = M in D ′ (( 0 , T ) × Ω ) . (154) In particular, if d ‾ ∈ L ∞ ( 0 , T ; H 1 ( Ω ) ) \overline{d}\in L^\infty(0,T;H^1(\Omega)) d ∈ L ∞ ( 0 , T ; H 1 ( Ω )) and M ~ ∈ L 2 ( ( 0 , T ) × Ω ) \widetilde{M}\in L^2((0,T)\times\Omega) M ∈ L 2 (( 0 , T ) × Ω ) , then ∂ t d ‾ ∈ L 2 ( 0 , T ; H − 1 ( Ω ) ) \partial_t\overline{d}\in L^2(0,T;H^{-1}(\Omega)) ∂ t d ∈ L 2 ( 0 , T ; H − 1 ( Ω )) .
Proof of Lemma A.4. For a.e. time, integration by parts in the two phases shows that the distributional spatial gradient of the glued field is its piecewise L 2 L^2 L 2 -gradient, because the possible interface term is the jump of its trace multiplied by the normal, and this jump is zero. This shows that d ‾ ( t ) ∈ H 1 ( Ω ; R 3 ) \overline{d}(t)\in H^1(\Omega;\mathbb{R}^3) d ( t ) ∈ H 1 ( Ω ; R 3 ) .
We justify the material equation on a compact time interval on which the gap is positive. To this end, let U U U be the affine rigid velocity generating S ( t ) S(t) S ( t ) , and let ρ ( t , x ) = dist ( x , S ( t ) ) \rho(t,x)=\operatorname{dist}(x,S(t)) ρ ( t , x ) = dist ( x , S ( t )) in a fixed fluid-side tubular neighborhood. From rigidity, it follows that ( ∂ t + U ⋅ ∇ ) ρ = 0 (\partial_t+U\cdot\nabla)\rho=0 ( ∂ t + U ⋅ ∇ ) ρ = 0 . We then choose ϑ ∈ C ∞ ( R ) \vartheta\in C^\infty(\mathbb{R}) ϑ ∈ C ∞ ( R ) , equal to zero on ( − ∞ , 1 ] (-\infty,1] ( − ∞ , 1 ] and to one on [ 2 , ∞ ) [2,\infty) [ 2 , ∞ ) , and set η r = ϑ ( ρ / r ) \eta_r=\vartheta(\rho/r) η r = ϑ ( ρ / r ) in the collar, extended by one in the remaining fluid. This shows ( ∂ t + u ⋅ ∇ ) η r = ( u − U ) ⋅ ∇ η r (\partial_t+\mathbf{u}\cdot\nabla)\eta_r=(\mathbf{u}-U)\cdot\nabla\eta_r ( ∂ t + u ⋅ ∇ ) η r = ( u − U ) ⋅ ∇ η r and ∣ ∇ η r ∣ ≤ C / r |\nabla\eta_r|\leq C/r ∣∇ η r ∣ ≤ C / r , and the derivative is supported in A r ( t ) = { x ∈ F ( t ) : r < ρ ( t , x ) < 2 r } A_r(t)=\{x\in F(t):r<\rho(t,x)<2r\} A r ( t ) = { x ∈ F ( t ) : r < ρ ( t , x ) < 2 r } . The trace of u − U \mathbf{u}-U u − U vanishes at the inner interface, so that the one-dimensional Hardy estimate in tubular coordinates gives
∥ u − U ρ ∥ L 2 ( { ( t , x ) : 0 < ρ ( t , x ) < r 0 } ) < ∞ . \left\|\frac{\mathbf{u}-U}{\rho}\right\|_{L^2(\{(t,x):0<\rho(t,x)<r_0\})}<\infty. ρ u − U L 2 ({( t , x ) : 0 < ρ ( t , x ) < r 0 }) < ∞.
Let ξ ∈ C ∞ ( ( 0 , T ) × Ω ; R 3 ) \xi\in C^\infty((0,T)\times\Omega;\mathbb{R}^3) ξ ∈ C ∞ (( 0 , T ) × Ω ; R 3 ) . We then use η r ξ \eta_r\xi η r ξ in the fluid molecular-field equation and observe that it is an admissible test function. Integration by parts in the convective term, we get
∫ 0 T ∫ F ( t ) η r [ − d F ⋅ ( ∂ t ξ + u ⋅ ∇ ξ ) − M ⋅ ξ ] d x d t = ∫ 0 T ∫ A r ( t ) d F ⋅ ξ ( ∂ t + u ⋅ ∇ ) η r d x d t . \int_0^T\int_{F(t)}\eta_r\left[-d_F\cdot(\partial_t\xi+\mathbf{u}\cdot\nabla\xi)-M\cdot\xi\right]\,\mathrm{d}x\,\mathrm{d}t
=
\int_0^T\int_{A_r(t)}d_F\cdot\xi\,(\partial_t+\mathbf{u}\cdot\nabla)\eta_r\,\mathrm{d}x\,\mathrm{d}t. ∫ 0 T ∫ F ( t ) η r [ − d F ⋅ ( ∂ t ξ + u ⋅ ∇ ξ ) − M ⋅ ξ ] d x d t = ∫ 0 T ∫ A r ( t ) d F ⋅ ξ ( ∂ t + u ⋅ ∇ ) η r d x d t .
In this identity, the absolute value of the right-hand side is bounded by
C ∥ ξ ∥ L ∞ ( ( 0 , T ) × Ω ) ∥ d F ∥ L 2 ( { ( t , x ) : x ∈ A r ( t ) } ) ∥ u − U ρ ∥ L 2 ( { ( t , x ) : 0 < ρ < r 0 } ) , C\|\xi\|_{L^\infty((0,T)\times\Omega)}\|d_F\|_{L^2(\{(t,x):x\in A_r(t)\})}
\left\|\frac{\mathbf{u}-U}{\rho}\right\|_{L^2(\{(t,x):0<\rho<r_0\})}, C ∥ ξ ∥ L ∞ (( 0 , T ) × Ω ) ∥ d F ∥ L 2 ({( t , x ) : x ∈ A r ( t )}) ρ u − U L 2 ({( t , x ) : 0 < ρ < r 0 }) ,
which tends to zero by absolute continuity of the first L 2 L^2 L 2 integral. For the other terms, we may use dominated convergence. This proves the fluid material identity for arbitrary ξ \xi ξ without an interface term. The solid branch satisfies the analogous identity with zero source by the rigid change of variables. Adding them proves (A.6). The argument is local in time. The resulting distributional identity holds on the whole open pre-contact interval.
Finally, we use u ∈ L t 2 L x 6 u \in L^2_tL^6_x u ∈ L t 2 L x 6 and ∇ d ‾ ∈ L t ∞ L x 2 \nabla\overline{d} \in L^\infty_tL^2_x ∇ d ∈ L t ∞ L x 2 to get u ⋅ ∇ d ‾ ∈ L 2 ( 0 , T ; L 3 / 2 ( Ω ) ) ↪ L 2 ( 0 , T ; H − 1 ( Ω ) ) u \cdot\nabla\overline{d} \in L^2(0,T;L^{3/2}(\Omega)) \hookrightarrow L^2(0,T;H^{-1}(\Omega)) u ⋅ ∇ d ∈ L 2 ( 0 , T ; L 3/2 ( Ω )) ↪ L 2 ( 0 , T ; H − 1 ( Ω )) . The last assertion of this lemma then follows from (A.6).
Auxiliary estimates for the Galerkin approximation
The Galerkin construction uses the following maximum principle for the relaxed director equation.
Lemma A.5. Let T > 0 T>0 T > 0 , let u ∈ L 2 ( 0 , T ; V ( Ω ) ) u \in L^2(0,T;\mathcal{V}(\Omega)) u ∈ L 2 ( 0 , T ; V ( Ω )) , let a ∈ L ∞ ( ( 0 , T ) × Ω ) a \in L^\infty((0,T)\times\Omega) a ∈ L ∞ (( 0 , T ) × Ω ) satisfy a ≥ 0 a \ge0 a ≥ 0 , and let b ∈ L ∞ ( ( 0 , T ) × Ω ; R 3 ) b \in L^\infty((0,T)\times\Omega;\mathbb{R}^3) b ∈ L ∞ (( 0 , T ) × Ω ; R 3 ) satisfy ∣ b ∣ ≤ 1 |b| \le1 ∣ b ∣ ≤ 1 and ( b ) 3 ≥ 0 (b)_3 \ge0 ( b ) 3 ≥ 0 a.e. Suppose that d d d is a sufficiently regular weak solution of
∂ t d + u ⋅ ∇ d = Δ d + 1 ε 2 ( 1 − ∣ d ∣ 2 ) d − a ( d − b ) (155) \partial_t d + u \cdot\nabla d = \Delta d + \frac{1}{\varepsilon^2}(1-|d|^2)d-a(d-b) \tag*{(155)} ∂ t d + u ⋅ ∇ d = Δ d + ε 2 1 ( 1 − ∣ d ∣ 2 ) d − a ( d − b ) (155) with initial and Dirichlet boundary data satisfying ∣ d ∣ ≤ 1 |d| \le1 ∣ d ∣ ≤ 1 and d 3 ≥ 0 d_3 \ge0 d 3 ≥ 0 . Then for a.e. ( t , x ) ∈ ( 0 , T ) × Ω (t,x) \in(0,T)\times\Omega ( t , x ) ∈ ( 0 , T ) × Ω , it holds that ∣ d ( t , x ) ∣ ≤ 1 |d(t,x)| \le1 ∣ d ( t , x ) ∣ ≤ 1 and d 3 ( t , x ) ≥ 0 d_3(t,x) \ge0 d 3 ( t , x ) ≥ 0 . In particular, the conclusion applies to the Galerkin director equation (56) , with a = η φ k a=\eta\varphi^k a = η φ k and b = b k b=b^k b = b k .
Proof. We set w = ∣ d ∣ 2 w=|d|^2 w = ∣ d ∣ 2 . From (A.7), it follows that
( ∂ t + u ⋅ ∇ − Δ ) w = − 2 ∣ ∇ d ∣ 2 + 2 ε 2 ( 1 − w ) w − 2 a ( w − d ⋅ b ) . (\partial_t+u\cdot\nabla-\Delta)w=-2|\nabla d|^2+\frac{2}{\varepsilon^2}(1-w)w-2a(w-d\cdot b). ( ∂ t + u ⋅ ∇ − Δ ) w = − 2∣∇ d ∣ 2 + ε 2 2 ( 1 − w ) w − 2 a ( w − d ⋅ b ) .
On the set { w > 1 } \{w>1\} { w > 1 } , the bound ∣ b ∣ ≤ 1 |b|\le1 ∣ b ∣ ≤ 1 gives w − d ⋅ b ≥ w − w ≥ 0 w-d\cdot b\ge w-\sqrt{w}\ge0 w − d ⋅ b ≥ w − w ≥ 0 , while ( 1 − w ) w ≤ 0 (1-w)w\le0 ( 1 − w ) w ≤ 0 . Testing the scalar inequality by ( w − 1 ) + (w-1)_+ ( w − 1 ) + , using div u = 0 \operatorname{div}u=0 div u = 0 and the fact that ( w − 1 ) + (w-1)_+ ( w − 1 ) + has zero trace and zero initial value, we conclude that ( w − 1 ) + = 0 (w-1)_+=0 ( w − 1 ) + = 0 .
For q = d 3 q=d_3 q = d 3 , after the first step, we have ( ∂ t + u ⋅ ∇ − Δ ) q = 1 ε 2 ( 1 − ∣ d ∣ 2 ) q − a q + a ( b ) 3 (\partial_t+u\cdot\nabla-\Delta)q=\frac{1}{\varepsilon^2}(1-|d|^2)q-aq+a(b)_3 ( ∂ t + u ⋅ ∇ − Δ ) q = ε 2 1 ( 1 − ∣ d ∣ 2 ) q − a q + a ( b ) 3 . Testing this identity by − q − -q_- − q − and using ( b ) 3 ≥ 0 (b)_3\ge0 ( b ) 3 ≥ 0 , a ≥ 0 a\ge0 a ≥ 0 , and 0 ≤ 1 − ∣ d ∣ 2 ≤ 1 0\le1-|d|^2\le1 0 ≤ 1 − ∣ d ∣ 2 ≤ 1 , we find that
1 2 d d t ∥ q − ∥ L 2 ( Ω ) 2 + ∥ ∇ q − ∥ L 2 ( Ω ) 2 ≤ 1 ε 2 ∥ q − ∥ L 2 ( Ω ) 2 . \frac{1}{2}\frac{\mathrm{d}}{\mathrm{d}t}\|q_-\|_{L^2(\Omega)}^2+\|\nabla q_-\|_{L^2(\Omega)}^2\le\frac{1}{\varepsilon^2}\|q_-\|_{L^2(\Omega)}^2. 2 1 d t d ∥ q − ∥ L 2 ( Ω ) 2 + ∥∇ q − ∥ L 2 ( Ω ) 2 ≤ ε 2 1 ∥ q − ∥ L 2 ( Ω ) 2 .
Since q − ( 0 ) = 0 q_-(0)=0 q − ( 0 ) = 0 and the negative part has zero boundary trace, an application of Grönwall’s lemma yields q − = 0 q_-=0 q − = 0 . The argument is first justified at the Galerkin level and then passes to the weak limits by the strong L 2 L^2 L 2 -convergence available in the construction. We also refer here to [39 Lemmas 2.1–2.2] .
Proof of Proposition 3.7 . We construct solutions to the semi-Galerkin problem introduced in § 3.1.1 , verify the energy identity (60) , and prove the separation estimate (48) . Note that the construction uses the semi-Galerkin scheme of [38 Sec. 2] , coupled to the transported solid and lifting introduced in § 3.1.1 .
We start with a given velocity field u ‾ k ∈ C ( [ 0 , T ] ; V k ) \overline{u}^k\in C([0,T];\mathcal{V}_k) u k ∈ C ([ 0 , T ] ; V k ) . Next, we construct the flow map X ‾ k \overline{X}^k X k as a solution of (51) , the inverse flow map Y ‾ k \overline{Y}^k Y k by (52) , the transported solid indicator φ ‾ k \overline{\varphi}^k φ k by (53) , and the lifting vector field b ‾ k \overline{b}^k b k by (54) . With these quantities fixed, the semilinear parabolic problem (56) has a unique energy solution on a short time interval. We denote it by d ‾ k \overline{d}^k d k and define M ‾ n , k \overline{M}^{n,k} M n , k by (57) .
The velocity field u k u^k u k is constructed by the following fixed-point argument. We use the L 2 L^2 L 2 -orthonormal basis { v j } j ∈ N ⊂ X u = V ( Ω ) ∩ H 3 ( Ω ; R 3 ) \{v_j\}_{j\in\mathbb{N}}\subset X_u=\mathcal{V}(\Omega)\cap H^3(\Omega;\mathbb{R}^3) { v j } j ∈ N ⊂ X u = V ( Ω ) ∩ H 3 ( Ω ; R 3 ) of H ( Ω ) \mathcal{H}(\Omega) H ( Ω ) made precise in § 3.1.1 . Its finite spans are dense in X u X_u X u in the H 3 H^3 H 3 -norm, and every fixed basis vector belongs to W 1 , ∞ ( Ω ; R 3 ) W^{1,\infty}(\Omega;\mathbb{R}^3) W 1 , ∞ ( Ω ; R 3 ) . We recall that V k : = Span { v 1 , … , v k } \mathcal{V}_k:=\operatorname{Span}\{v_1,\ldots,v_k\} V k := Span { v 1 , … , v k } and consider the associated orthogonal projection P k : H ( Ω ) → V k \mathcal{P}_k:\mathcal{H}(\Omega)\to\mathcal{V}_k P k : H ( Ω ) → V k . We then look for a solution of the form as in (50) , i.e., u ˉ k ( t , x ) = ∑ j = 1 k a j k ( t ) v j ( x ) \bar{u}^{k}(t,x)=\sum_{j=1}^{k}a_{j}^{k}(t)v_{j}(x) u ˉ k ( t , x ) = ∑ j = 1 k a j k ( t ) v j ( x ) , satisfying for all l = 1 , … , k l=1,\ldots,k l = 1 , … , k , the identity
( ∂ t u k , v l ) L 2 ( Ω ) + ∫ Ω ( ( u k ⋅ ∇ ) u k ) ⋅ v l d x + ∫ Ω ( 2 + η φ ˉ k ) D ( u k ) : D ( v l ) d x + δ a u ( u k , v l ) = − ∫ Ω ( v l ⋅ ∇ ) d ˉ k ⋅ ( M ‾ n , k + η φ ˉ k ( d ˉ k − b ˉ k ) ) d x . (156) \begin{aligned}
(\partial_{t}u^{k},v_{l})_{L^{2}(\Omega)}+\int_{\Omega}((u^{k}\cdot\nabla)u^{k})\cdot v_{l}\,dx+\int_{\Omega}(2+\eta\bar{\varphi}^{k})\mathbb{D}(u^{k}):\mathbb{D}(v_{l})\,dx+\delta a_{u}(u^{k},v_{l})\\
=-\int_{\Omega}(v_{l}\cdot\nabla)\bar{d}^{k}\cdot\left(\overline{M}^{n,k}+\eta\bar{\varphi}^{k}(\bar{d}^{k}-\bar{b}^{k})\right)\,dx. \tag*{(156)}
\end{aligned} ( ∂ t u k , v l ) L 2 ( Ω ) + ∫ Ω (( u k ⋅ ∇ ) u k ) ⋅ v l d x + ∫ Ω ( 2 + η φ ˉ k ) D ( u k ) : D ( v l ) d x + δ a u ( u k , v l ) = − ∫ Ω ( v l ⋅ ∇ ) d ˉ k ⋅ ( M n , k + η φ ˉ k ( d ˉ k − b ˉ k ) ) d x . (156)
We formulate the fixed-point step with the transported characteristic function and the director penalization included. Fix R > 2 ∥ P k u 0 ∥ L 2 ( Ω ) R>2\lVert P_{k}u_{0}\rVert_{L^{2}(\Omega)} R > 2 ∥ P k u 0 ∥ L 2 ( Ω ) and let
B R : = { u ˉ k ∈ C ( [ 0 , T ] ; V k ) : u ˉ k ( 0 ) = P k u 0 , ∥ u ˉ k ∥ C t L x 2 ≤ R } . B_{R}:=\left\{\bar{u}^{k}\in C([0,T];V_{k}):\bar{u}^{k}(0)=P_{k}u_{0},\ \lVert\bar{u}^{k}\rVert_{C_{t}L^{2}_{x}}\le R\right\}. B R := { u ˉ k ∈ C ([ 0 , T ] ; V k ) : u ˉ k ( 0 ) = P k u 0 , ∥ u ˉ k ∥ C t L x 2 ≤ R } .
Note that all norms on the finite-dimensional space V k V_{k} V k are equivalent. In particular, for u ˉ k ∈ B R \bar{u}^{k}\in B_{R} u ˉ k ∈ B R , the associated flow and inverse flow exist on a time interval depending only on k k k and R R R , and are uniformly bounded in C 1 C^{1} C 1 there. For the transported indicator φ ˉ k \bar{\varphi}^{k} φ ˉ k , it follows that 0 ≤ φ ˉ k ≤ 1 0\le\bar{\varphi}^{k}\le1 0 ≤ φ ˉ k ≤ 1 , while b ˉ k = d 0 ∘ Y ˉ k \bar{b}^{k}=d_{0}\circ\bar{Y}^{k} b ˉ k = d 0 ∘ Y ˉ k is bounded in all Sobolev norms needed in the parabolic problem, amd the constants depend on k k k , R R R , T T T as well as the data but not on the particular element of B R B_{R} B R .
For these prescribed coefficients, the director equation gives rise to a semilinear parabolic Dirichlet problem with bounded zeroth-order coefficient η φ ˉ k \eta\bar{\varphi}^{k} η φ ˉ k . Employing standard Galerkin or monotonicity theory, we find a unique energy solution on the chosen interval. Moreover, Lemma A.5 yields ∣ d ˉ k ∣ ≤ 1 \lvert\bar{d}^{k}\rvert\le1 ∣ d ˉ k ∣ ≤ 1 and ( d ˉ k ) 3 ≥ 0 (\bar{d}^{k})_{3}\ge0 ( d ˉ k ) 3 ≥ 0 . Testing the director equation by its molecular field M ‾ n , k \overline{M}^{n,k} M n , k and using the bounds for the prescribed drift and lift yields, on sufficiently short intervals, the estimate
∥ d ˉ k ∥ L t ∞ H x 1 + ∥ d ˉ k ∥ L t 2 H x 2 + ∥ M ‾ n , k ∥ L t , x 2 ≤ C k , n , δ , ε , R , u 0 , d 0 . \lVert\bar{d}^{k}\rVert_{L^{\infty}_{t}H^{1}_{x}}+\lVert\bar{d}^{k}\rVert_{L^{2}_{t}H^{2}_{x}}+\lVert\overline{M}^{n,k}\rVert_{L^{2}_{t,x}}\le C_{k,n,\delta,\varepsilon,R,u_{0},d_{0}}. ∥ d ˉ k ∥ L t ∞ H x 1 + ∥ d ˉ k ∥ L t 2 H x 2 + ∥ M n , k ∥ L t , x 2 ≤ C k , n , δ , ε , R , u 0 , d 0 .
We observe that the right-hand side of (A.8) belongs to L 2 ( 0 , T ) L^{2}(0,T) L 2 ( 0 , T ) , while the remaining coefficients are continuous or bounded in time. Hence (A.8) is a Carathéodory system of finite-dimensional ODEs for the coefficient vector ( a 1 k , … , a k k ) (a_{1}^{k},\ldots,a_{k}^{k}) ( a 1 k , … , a k k ) , and it has a unique absolutely continuous local solution. The estimates below give u k ∈ W 1 , 2 ( 0 , T ; V k ) ⊂ C ( [ 0 , T ] ; V k ) u^{k}\in W^{1,2}(0,T;V_{k})\subset C([0,T];V_{k}) u k ∈ W 1 , 2 ( 0 , T ; V k ) ⊂ C ([ 0 , T ] ; V k ) . Norm equivalence and the preceding bounds give
sup 0 ≤ t ≤ T ∥ u k ( t ) − P k u 0 ∥ L 2 ( Ω ) ≤ C k , n , δ , ε , R , u 0 , d 0 T θ \sup_{0\le t\le T}\lVert u^{k}(t)-P_{k}u_{0}\rVert_{L^{2}(\Omega)}\le C_{k,n,\delta,\varepsilon,R,u_{0},d_{0}}T^{\theta} 0 ≤ t ≤ T sup ∥ u k ( t ) − P k u 0 ∥ L 2 ( Ω ) ≤ C k , n , δ , ε , R , u 0 , d 0 T θ
for some θ > 0 \theta>0 θ > 0 . Thus, after reducing T T T if necessary, the map T : B R → B R \mathcal{T}:B_{R}\to B_{R} T : B R → B R , T ( u ˉ k ) : = u k \mathcal{T}(\bar{u}^{k}):=u^{k} T ( u ˉ k ) := u k , is well-defined.
We verify continuity of the fixed-point map without assuming additional smoothness of the spectral basis. Fix k , n , δ , ε k,n,\delta,\varepsilon k , n , δ , ε and suppose u ˉ j k → u ˉ k \bar{u}_{j}^{k}\to\bar{u}^{k} u ˉ j k → u ˉ k in C ( [ 0 , T ] ; V k ) C([0,T];V_{k}) C ([ 0 , T ] ; V k ) . Since V k ⊂ H 3 ( Ω ; R 3 ) V_{k}\subset H^{3}(\Omega;\mathbb{R}^{3}) V k ⊂ H 3 ( Ω ; R 3 ) is finite dimensional, this convergence also holds in C ( [ 0 , T ] ; H 3 ( Ω ; R 3 ) ) C([0,T];H^{3}(\Omega;\mathbb{R}^{3})) C ([ 0 , T ] ; H 3 ( Ω ; R 3 )) and hence in C ( [ 0 , T ] ; W 1 , ∞ ( Ω ; R 3 ) ) C([0,T];W^{1,\infty}(\Omega;\mathbb{R}^{3})) C ([ 0 , T ] ; W 1 , ∞ ( Ω ; R 3 )) . Stability of the flow and its inverse gives φ ˉ j k → φ ˉ k \bar{\varphi}_{j}^{k}\to\bar{\varphi}^{k} φ ˉ j k → φ ˉ k in C ( [ 0 , T ] ; L p ( Ω ) ) C([0,T];L^{p}(\Omega)) C ([ 0 , T ] ; L p ( Ω )) whenever p < ∞ p<\infty p < ∞ and b ˉ j k → b ˉ k \bar{b}_{j}^{k}\to\bar{b}^{k} b ˉ j k → b ˉ k in C ( [ 0 , T ] ; L ∞ ( Ω ; R 3 ) ) C([0,T];L^{\infty}(\Omega;\mathbb{R}^{3})) C ([ 0 , T ] ; L ∞ ( Ω ; R 3 )) . Only these convergences of the transported coefficients are needed in the physical director equation.
For brevity in this paragraph, denote the prescribed velocities by v j , v v_{j},v v j , v , their transported coefficients by ϕ j , ϕ \phi_{j},\phi ϕ j , ϕ and b j , b b_{j},b b j , b , and the corresponding director solutions by d j , d d_{j},d d j , d . Set r j = d j − d r_{j}=d_{j}-d r j = d j − d . Then r j r_{j} r j has zero initial value and zero outer trace, and satisfies
∂ t r j + ( v j ⋅ ∇ ) r j − Δ r j + n ϕ j r j = − ( ( v j − v ) ⋅ ∇ ) d − ( f ε ( d j ) − f ε ( d ) ) − n ( ϕ j − ϕ ) ( d − b ) + n ϕ j ( b j − b ) . \begin{aligned}
\partial_{t}r_{j}+(v_{j}\cdot\nabla)r_{j}-\Delta r_{j}+n\phi_{j}r_{j}
&=-((v_{j}-v)\cdot\nabla)d-(f_{\varepsilon}(d_{j})-f_{\varepsilon}(d))\\
&\quad-n(\phi_{j}-\phi)(d-b)+n\phi_{j}(b_{j}-b).
\end{aligned} ∂ t r j + ( v j ⋅ ∇ ) r j − Δ r j + n ϕ j r j = − (( v j − v ) ⋅ ∇ ) d − ( f ε ( d j ) − f ε ( d )) − n ( ϕ j − ϕ ) ( d − b ) + n ϕ j ( b j − b ) .
Testing by r j r_j r j , using div v j = 0 \operatorname{div} v_j=0 div v j = 0 , the maximum principle, and Lipschitz continuity of f ε f_\varepsilon f ε on the closed unit ball, gives
1 2 d d t ∥ r j ∥ L 2 ( Ω ) 2 + ∥ ∇ r j ∥ L 2 ( Ω ) 2 ≤ C n , ε ∥ r j ∥ L 2 ( Ω ) 2 + C ∥ v j − v ∥ L ∞ ( Ω ) 2 ∥ ∇ d ∥ L 2 ( Ω ) 2 + C n 2 ( ∥ ϕ j − ϕ ∥ L 2 ( Ω ) 2 + ∥ b j − b ∥ L 2 ( Ω ) 2 ) . \frac{1}{2}\frac{\mathrm{d}}{\mathrm{d}t}\|r_j\|_{L^2(\Omega)}^2+\|\nabla r_j\|_{L^2(\Omega)}^2\le C_{n,\varepsilon}\|r_j\|_{L^2(\Omega)}^2+C\|v_j-v\|_{L^\infty(\Omega)}^2\|\nabla d\|_{L^2(\Omega)}^2+Cn^2\left(\|\phi_j-\phi\|_{L^2(\Omega)}^2+\|b_j-b\|_{L^2(\Omega)}^2\right). 2 1 d t d ∥ r j ∥ L 2 ( Ω ) 2 + ∥∇ r j ∥ L 2 ( Ω ) 2 ≤ C n , ε ∥ r j ∥ L 2 ( Ω ) 2 + C ∥ v j − v ∥ L ∞ ( Ω ) 2 ∥∇ d ∥ L 2 ( Ω ) 2 + C n 2 ( ∥ ϕ j − ϕ ∥ L 2 ( Ω ) 2 + ∥ b j − b ∥ L 2 ( Ω ) 2 ) .
Grönwall’s inequality then yields d j → d d_j\to d d j → d in C ( [ 0 , T ] ; L 2 ( Ω ; R 3 ) ) ∩ L 2 ( 0 , T ; H 1 ( Ω ; R 3 ) ) C([0,T];L^2(\Omega;\mathbb{R}^3))\cap L^2(0,T;H^1(\Omega;\mathbb{R}^3)) C ([ 0 , T ] ; L 2 ( Ω ; R 3 )) ∩ L 2 ( 0 , T ; H 1 ( Ω ; R 3 )) . For each fixed velocity basis vector v l v_l v l , write the director force in stress form. With F j , l ( t ) : = ∫ Ω ( ∇ d j ) ⊤ ∇ d j : ∇ v l d x F_{j,l}(t):=\int_\Omega(\nabla d_j)^\top\nabla d_j:\nabla v_l\,\mathrm{d}x F j , l ( t ) := ∫ Ω ( ∇ d j ) ⊤ ∇ d j : ∇ v l d x , we have
∥ F j , l − F l ∥ L 1 ( 0 , T ) ≤ ∥ ∇ v l ∥ L ∞ ( Ω ) ( ∥ ∇ d j ∥ L 2 ( ( 0 , T ) × Ω ) + ∥ ∇ d ∥ L 2 ( ( 0 , T ) × Ω ) ) ⋅ ∥ ∇ r j ∥ L 2 ( ( 0 , T ) × Ω ) → 0. \|F_{j,l}-F_l\|_{L^1(0,T)}\le\|\nabla v_l\|_{L^\infty(\Omega)}\left(\|\nabla d_j\|_{L^2((0,T)\times\Omega)}+\|\nabla d\|_{L^2((0,T)\times\Omega)}\right)\cdot\|\nabla r_j\|_{L^2((0,T)\times\Omega)}\to0. ∥ F j , l − F l ∥ L 1 ( 0 , T ) ≤ ∥∇ v l ∥ L ∞ ( Ω ) ( ∥∇ d j ∥ L 2 (( 0 , T ) × Ω ) + ∥∇ d ∥ L 2 (( 0 , T ) × Ω ) ) ⋅ ∥∇ r j ∥ L 2 (( 0 , T ) × Ω ) → 0.
Thanks to the convergence of ϕ j \phi_j ϕ j and the boundedness of the gradients of the fixed basis functions, the viscosity matrices also converge. Continuous dependence in the integral form of the finite-dimensional momentum ODE now gives T ( u ‾ j k ) → T ( u ‾ k ) \mathcal{T}(\overline{u}^{\,k}_j)\to\mathcal{T}(\overline{u}^{\,\,k}) T ( u j k ) → T ( u k ) in C ( [ 0 , T ] ; V k ) C([0,T];V_k) C ([ 0 , T ] ; V k ) . The coefficient derivatives are uniformly bounded in L 2 ( 0 , T ) L^2(0,T) L 2 ( 0 , T ) on the fixed-point ball, which implies equicontinuity. We may thus apply the Arzelà–Ascoli theorem, which yields that T ( u ‾ k ) \mathcal{T}(\overline{u}^{\,k}) T ( u k ) is compact, so Schauder’s theorem implies in turn that there is a fixed point.
At the fixed point, the flow X k X^k X k , indicator φ k \varphi^k φ k , lift b k b^k b k , director d k d^k d k , and velocity u k u^k u k satisfy the equations (53) –(59) . The exact cancellation shown in (61) –(63) in the proof of Proposition 3.6 gives (60) . We observe that these bounds are independent of the local fixed-point radius. Since the director problem is posed on the fixed container and the velocity takes values in the finite-dimensional space V k V_k V k , the construction can be restarted, and the Galerkin solution exists on every bounded time interval.
Finally, we prove the separation estimate (48) for the solids S k ( t ) = X 0 , t k ( S 0 ) S^k(t)=X^k_{0,t}(S_0) S k ( t ) = X 0 , t k ( S 0 ) resulting from the Galerkin approximation. Let x ∈ Ω x\in\Omega x ∈ Ω , and let z ∈ ∂ Ω z\in\partial\Omega z ∈ ∂ Ω be a nearest boundary point. The half-open segment from x x x to z z z lies in Ω \Omega Ω , and u k ( t ) u^k(t) u k ( t ) vanishes on ∂ Ω \partial\Omega ∂ Ω , so ∣ u k ( t , x ) ∣ ≤ ∥ ∇ u k ( t ) ∥ L ∞ ( Ω ) dist ( x , ∂ Ω ) |u^k(t,x)|\le\|\nabla u^k(t)\|_{L^\infty(\Omega)}\operatorname{dist}(x,\partial\Omega) ∣ u k ( t , x ) ∣ ≤ ∥∇ u k ( t ) ∥ L ∞ ( Ω ) dist ( x , ∂ Ω ) . For y ∈ S 0 y\in S_0 y ∈ S 0 , the function g ( t ) : = dist ( X 0 , t k ( y ) , ∂ Ω ) g(t):=\operatorname{dist}(X^k_{0,t}(y),\partial\Omega) g ( t ) := dist ( X 0 , t k ( y ) , ∂ Ω ) is Lipschitz and satisfies, for a.e. t t t , the estimate ∣ g ′ ( t ) ∣ ≤ ∣ u k ( t , X 0 , t k ( y ) ) ∣ ≤ ∥ ∇ u k ( t ) ∥ L ∞ ( Ω ) g ( t ) |g'(t)|\le|u^k(t,X^k_{0,t}(y))|\le\|\nabla u^k(t)\|_{L^\infty(\Omega)}g(t) ∣ g ′ ( t ) ∣ ≤ ∣ u k ( t , X 0 , t k ( y )) ∣ ≤ ∥∇ u k ( t ) ∥ L ∞ ( Ω ) g ( t ) . Since g ( 0 ) ≥ δ 0 g(0)\ge\delta_0 g ( 0 ) ≥ δ 0 and δ ∥ u k ∥ L 2 ( 0 , t ; X u ) ≤ E k n , δ , ε ( 0 ) ≤ E p e n , 0 \delta\|u^k\|_{L^2(0,t;X_u)}\le E_k^{n,\delta,\varepsilon}(0)\le E_{\mathrm{pen},0} δ ∥ u k ∥ L 2 ( 0 , t ; X u ) ≤ E k n , δ , ε ( 0 ) ≤ E pen , 0 by (60) , by Grönwall’s lemma and the Cauchy–Schwarz inequality we have
g ( t ) ≥ δ 0 exp ( − ∫ 0 t ∥ ∇ u k ( s ) ∥ L ∞ ( Ω ) d s ) ≥ δ 0 exp ( − C X t ∥ u k ∥ L 2 ( 0 , t ; X u ) ) ≥ ρ δ ( t ) . g(t)\ge\delta_0\exp\left(-\int_0^t\|\nabla u^k(s)\|_{L^\infty(\Omega)}\,\mathrm{d}s\right)\ge\delta_0\exp\left(-C_X\sqrt{t}\,\|u^k\|_{L^2(0,t;X_u)}\right)\ge\rho_\delta(t). g ( t ) ≥ δ 0 exp ( − ∫ 0 t ∥∇ u k ( s ) ∥ L ∞ ( Ω ) d s ) ≥ δ 0 exp ( − C X t ∥ u k ∥ L 2 ( 0 , t ; X u ) ) ≥ ρ δ ( t ) .
Taking the infimum over y ∈ S 0 y\in S_0 y ∈ S 0 proves the separation estimate for every k k k and n n n . In particular, the approximate solid never reaches the container wall. □
The time-derivative estimates in Claim 3.8 are proved next.
Proof of Claim 3.8 . From (A.8), for every v l ∈ V k v_l\in V_k v l ∈ V k we have
∣ ∫ Ω ∂ t u k ⋅ v l d x ∣ ≤ ∫ Ω ( ∣ ( u k ⊗ u k ) : ∇ v l ∣ + ∣ ( 2 + n φ k ) ( D ( u k ) : D ( v l ) ) ∣ + ∣ ( v l ⋅ ∇ ) d k ⋅ ( M n , k + n φ k ( d k − b k ) ) ∣ ) d x + δ ∣ a u ( u k , v l ) ∣ . (157) \left|\int_\Omega\partial_tu^k\cdot v_l\,\mathrm{d}x\right|\le\int_\Omega\left(\left|(u^k\otimes u^k):\nabla v_l\right|+\left|(2+n\varphi^k)(\mathbb{D}(u^k):\mathbb{D}(v_l))\right|
+\left|(v_l\cdot\nabla)d^k\cdot\left(M^{n,k}+n\varphi^k(d^k-b^k)\right)\right|\right)\,\mathrm{d}x+\delta|a_u(u^k,v_l)|. \tag*{(157)} ∫ Ω ∂ t u k ⋅ v l d x ≤ ∫ Ω ( ( u k ⊗ u k ) : ∇ v l + ( 2 + n φ k ) ( D ( u k ) : D ( v l )) + ( v l ⋅ ∇ ) d k ⋅ ( M n , k + n φ k ( d k − b k ) ) ) d x + δ ∣ a u ( u k , v l ) ∣. (157)
Applying Hölder’s inequality together with the Sobolev embedding H 2 ( Ω ) ↪ L ∞ ( Ω ) H^2(\Omega)\hookrightarrow L^\infty(\Omega) H 2 ( Ω ) ↪ L ∞ ( Ω ) , we estimate the first three terms on the right-hand side of (A.9) by
∫ Ω ∣ ( u k ⊗ u k ) ( t ) : ∇ v l ∣ d x ≤ ∥ u k ( t ) ∥ L 2 ( Ω ) 2 ∥ v l ∥ H 3 ( Ω ) , (158) \int_\Omega\left|(u^k\otimes u^k)(t):\nabla v_l\right|\,\mathrm{d}x\le\|u^k(t)\|_{L^2(\Omega)}^2\|v_l\|_{H^3(\Omega)}, \tag*{(158)} ∫ Ω ( u k ⊗ u k ) ( t ) : ∇ v l d x ≤ ∥ u k ( t ) ∥ L 2 ( Ω ) 2 ∥ v l ∥ H 3 ( Ω ) , (158)
∫ Ω ∣ ( 2 + n φ k ) ( D ( u k ( t ) ) : D ( v l ) ) ∣ d x ≤ C ∥ ∇ u k ( t ) ∥ L 2 ( Ω ) ∥ v l ∥ H 1 ( Ω ) , \int_\Omega\left|(2+n\varphi^k)(\mathbb{D}(u^k(t)):\mathbb{D}(v_l))\right|\,\mathrm{d}x\le C\|\nabla u^k(t)\|_{L^2(\Omega)}\|v_l\|_{H^1(\Omega)}, ∫ Ω ( 2 + n φ k ) ( D ( u k ( t )) : D ( v l )) d x ≤ C ∥∇ u k ( t ) ∥ L 2 ( Ω ) ∥ v l ∥ H 1 ( Ω ) ,
and
δ ∣ a u ( u k , v l ) ∣ ≤ δ ∥ u k ∥ X u ∥ v l ∥ X u . \delta|a_u(u^k,v_l)|\le\delta\|u^k\|_{X_u}\|v_l\|_{X_u}. δ ∣ a u ( u k , v l ) ∣ ≤ δ ∥ u k ∥ X u ∥ v l ∥ X u .
For the last term, we recall from (57) that M n , k + n φ k ( d k − b k ) = Δ d k + 1 ε 2 ( 1 − ∣ d k ∣ 2 ) d k \mathbf{M}^{n,k}+n\varphi^{k}(d^{k}-b^{k})=\Delta d^{k}+\frac{1}{\varepsilon^{2}}(1-|d^{k}|^{2})d^{k} M n , k + n φ k ( d k − b k ) = Δ d k + ε 2 1 ( 1 − ∣ d k ∣ 2 ) d k . Thus, additionally using the embedding H 1 ( Ω ) ↪ L 6 ( Ω ) H^{1}(\Omega)\hookrightarrow L^{6}(\Omega) H 1 ( Ω ) ↪ L 6 ( Ω ) , we find that
∫ Ω ∣ ( v l ⋅ ∇ ) d k ⋅ ( M n , k + n φ k ( d k − b k ) ) ∣ d x = ∫ Ω ∣ ( v l ⋅ ∇ ) d k ⋅ ( Δ d k + 1 ε 2 ( 1 − ∣ d k ∣ 2 ) d k ) ∣ d x ≤ C ∥ d k ∥ H 1 ( Ω ) ( ∥ d k ∥ H 2 ( Ω ) + ∥ d k ∥ L 6 ( Ω ) 3 ) ∥ v l ∥ H 2 ( Ω ) ≤ C ∥ d k ∥ H 1 ( Ω ) ( ∥ d k ∥ H 2 ( Ω ) + ∥ d k ∥ H 1 ( Ω ) 3 ) ∥ v l ∥ H 2 ( Ω ) . (159) \begin{aligned}
\int_{\Omega}\left|(v^{l}\cdot\nabla)d^{k}\cdot\left(\mathbf{M}^{n,k}+n\varphi^{k}(d^{k}-b^{k})\right)\right|\,\mathrm{d}x
&=\int_{\Omega}\left|(v^{l}\cdot\nabla)d^{k}\cdot\left(\Delta d^{k}+\frac{1}{\varepsilon^{2}}(1-|d^{k}|^{2})d^{k}\right)\right|\,\mathrm{d}x\\
&\le C\|d^{k}\|_{H^{1}(\Omega)}\left(\|d^{k}\|_{H^{2}(\Omega)}+\|d^{k}\|_{L^{6}(\Omega)}^{3}\right)\|v^{l}\|_{H^{2}(\Omega)}\\
&\le C\|d^{k}\|_{H^{1}(\Omega)}\left(\|d^{k}\|_{H^{2}(\Omega)}+\|d^{k}\|_{H^{1}(\Omega)}^{3}\right)\|v^{l}\|_{H^{2}(\Omega)}. \tag*{(159)}
\end{aligned} ∫ Ω ( v l ⋅ ∇ ) d k ⋅ ( M n , k + n φ k ( d k − b k ) ) d x = ∫ Ω ( v l ⋅ ∇ ) d k ⋅ ( Δ d k + ε 2 1 ( 1 − ∣ d k ∣ 2 ) d k ) d x ≤ C ∥ d k ∥ H 1 ( Ω ) ( ∥ d k ∥ H 2 ( Ω ) + ∥ d k ∥ L 6 ( Ω ) 3 ) ∥ v l ∥ H 2 ( Ω ) ≤ C ∥ d k ∥ H 1 ( Ω ) ( ∥ d k ∥ H 2 ( Ω ) + ∥ d k ∥ H 1 ( Ω ) 3 ) ∥ v l ∥ H 2 ( Ω ) . (159)
The estimates (A.9)–(A.11) are linear in the test field and therefore give, for every v ∈ V k v\in\mathcal{V}_{k} v ∈ V k ,
∣ ( ∂ t u k ( t ) , v ) L 2 ( Ω ) ∣ ≤ C a k ( t ) ∥ v ∥ X u for a.e. t ∈ ( 0 , T ) , \left|(\partial_{t}u^{k}(t),v)_{L^{2}(\Omega)}\right|\le Ca_{k}(t)\|v\|_{X_{u}}\qquad\text{for a.e. }t\in(0,T), ( ∂ t u k ( t ) , v ) L 2 ( Ω ) ≤ C a k ( t ) ∥ v ∥ X u for a.e. t ∈ ( 0 , T ) ,
where C C C is independent of k k k (with n n n , δ \delta δ , ε \varepsilon ε fixed) and
a k ( t ) : = ∥ u k ( t ) ∥ L 2 ( Ω ) 2 + ∥ ∇ u k ( t ) ∥ L 2 ( Ω ) 2 + δ ∥ u k ( t ) ∥ X u + ∥ d k ∥ H 1 ( Ω ) ( ∥ d k ∥ H 2 ( Ω ) + ∥ d k ∥ H 1 ( Ω ) 3 ) . a_{k}(t):=\|u^{k}(t)\|_{L^{2}(\Omega)}^{2}+\|\nabla u^{k}(t)\|_{L^{2}(\Omega)}^{2}+\delta\|u^{k}(t)\|_{X_{u}}+\|d^{k}\|_{H^{1}(\Omega)}\left(\|d^{k}\|_{H^{2}(\Omega)}+\|d^{k}\|_{H^{1}(\Omega)}^{3}\right). a k ( t ) := ∥ u k ( t ) ∥ L 2 ( Ω ) 2 + ∥∇ u k ( t ) ∥ L 2 ( Ω ) 2 + δ ∥ u k ( t ) ∥ X u + ∥ d k ∥ H 1 ( Ω ) ( ∥ d k ∥ H 2 ( Ω ) + ∥ d k ∥ H 1 ( Ω ) 3 ) .
For arbitrary ζ ∈ X u \zeta\in X_{u} ζ ∈ X u , use the spectral projection P k ζ ∈ V k P_{k}\zeta\in\mathcal{V}_{k} P k ζ ∈ V k . Since ∂ t u k ( t ) ∈ V k \partial_{t}u^{k}(t)\in\mathcal{V}_{k} ∂ t u k ( t ) ∈ V k and P k P_{k} P k is the L 2 L^{2} L 2 -orthogonal projection, we find that ⟨ ∂ t u k , ζ ⟩ = ( ∂ t u k , P k ζ ) L 2 ( Ω ) \langle\partial_{t}u^{k},\zeta\rangle=(\partial_{t}u^{k},P_{k}\zeta)_{L^{2}(\Omega)} ⟨ ∂ t u k , ζ ⟩ = ( ∂ t u k , P k ζ ) L 2 ( Ω ) , and (49) yields the estimate
∣ ⟨ ∂ t u k ( t ) , ζ ⟩ ∣ ≤ C a k ( t ) ∥ P k ζ ∥ X u ≤ C a k ( t ) ∥ ζ ∥ X u . \left|\langle\partial_{t}u^{k}(t),\zeta\rangle\right|\le Ca_{k}(t)\|P_{k}\zeta\|_{X_{u}}\le Ca_{k}(t)\|\zeta\|_{X_{u}}. ⟨ ∂ t u k ( t ) , ζ ⟩ ≤ C a k ( t ) ∥ P k ζ ∥ X u ≤ C a k ( t ) ∥ ζ ∥ X u .
The energy bound in (64) yields a k ∈ L 2 ( 0 , T ) a_{k}\in L^{2}(0,T) a k ∈ L 2 ( 0 , T ) , and therefore (65) follows with a constant independent of k k k .
Writing the director equation (56) 1 _{1} 1 in terms of M n , k \mathbf{M}^{n,k} M n , k as in (58) , for e ∈ L 2 ( Ω ; R 3 ) e\in L^{2}(\Omega;\mathbb{R}^{3}) e ∈ L 2 ( Ω ; R 3 ) , we get
− ∫ Ω ∂ t d k ⋅ e d x = ∫ Ω [ ( u k ⋅ ∇ ) d k ⋅ e − M n , k ⋅ e ] d x . -\int_{\Omega}\partial_{t}d^{k}\cdot e\,\mathrm{d}x=\int_{\Omega}\left[(u^{k}\cdot\nabla)d^{k}\cdot e-\mathbf{M}^{n,k}\cdot e\right]\,\mathrm{d}x. − ∫ Ω ∂ t d k ⋅ e d x = ∫ Ω [ ( u k ⋅ ∇ ) d k ⋅ e − M n , k ⋅ e ] d x .
Estimating each term via Hölder’s inequality and Sobolev embeddings (cf. (A.10)–(A.11) for the analogous velocity estimates), we obtain
∣ ∫ Ω ∂ t d k ⋅ e d x ∣ ≤ C b k ( t ) ∥ e ∥ L 2 ( Ω ) for all e ∈ L 2 ( Ω ; R 3 ) , \left|\int_{\Omega}\partial_{t}d^{k}\cdot e\,\mathrm{d}x\right|\le Cb_{k}(t)\|e\|_{L^{2}(\Omega)}\qquad\text{for all }e\in L^{2}(\Omega;\mathbb{R}^{3}), ∫ Ω ∂ t d k ⋅ e d x ≤ C b k ( t ) ∥ e ∥ L 2 ( Ω ) for all e ∈ L 2 ( Ω ; R 3 ) ,
where b k ( t ) : = ∥ u k ∥ H 3 ( Ω ) ∥ d k ∥ H 1 ( Ω ) + ∥ M n , k ∥ L 2 ( Ω ) b_{k}(t):=\|u^{k}\|_{H^{3}(\Omega)}\|d^{k}\|_{H^{1}(\Omega)}+\|\mathbf{M}^{n,k}\|_{L^{2}(\Omega)} b k ( t ) := ∥ u k ∥ H 3 ( Ω ) ∥ d k ∥ H 1 ( Ω ) + ∥ M n , k ∥ L 2 ( Ω ) . By the bounds for d k d^{k} d k and u k u^{k} u k in (64) , we conclude that b k ∈ L 2 ( 0 , T ) b_{k}\in L^{2}(0,T) b k ∈ L 2 ( 0 , T ) and thus (65) , finishing the proof. □
Auxiliary results for the rigidity and artificial-viscosity limits.
Proof of Lemma 3.10. It is sufficient to consider σ > 0 \sigma>0 σ > 0 smaller than a fixed tubular radius of the rigid body and than one quarter of its distance from ∂ Ω \partial\Omega ∂ Ω . The assertion for larger σ \sigma σ then follows by restriction. By the uniform Hausdorff convergence (82) and the uniform continuity of the limiting rigid trajectory, we can cover [ 0 , T ] [0,T] [ 0 , T ] by finitely many intervals I j I_{j} I j and choose smooth fixed sets A j ⋐ B j ⋐ Ω A_{j}\Subset B_{j}\Subset\Omega A j ⋐ B j ⋐ Ω such that, with G j : = Ω ∖ B j ‾ G_{j}:=\Omega\setminus\overline{B_{j}} G j := Ω ∖ B j , for all sufficiently large n n n and all t ∈ I j t\in I_{j} t ∈ I j ,
S ( φ n ( t ) ) ⋐ A j ⊂ S ( φ ( t ) ) σ / 2 , F ~ σ ( t ) ⊂ G j , dist ( A j , Ω ∖ B j ) ≥ c σ . (160) S(\varphi^{n}(t))\Subset A_{j}\subset S(\varphi(t))_{\sigma/2},\qquad\widetilde{F}_{\sigma}(t)\subset G_{j},\qquad\operatorname{dist}(A_{j},\Omega\setminus B_{j})\ge c\sigma. \tag*{(160)} S ( φ n ( t )) ⋐ A j ⊂ S ( φ ( t ) ) σ /2 , F σ ( t ) ⊂ G j , dist ( A j , Ω ∖ B j ) ≥ c σ . (160)
For example, after choosing a center time t j t_{j} t j so that the oscillation of S ( t ) S(t) S ( t ) on I j I_{j} I j is less than σ / 16 \sigma/16 σ /16 , one may take smooth tubular neighborhoods of S ( t j ) S(t_{j}) S ( t j ) of radii σ / 4 \sigma/4 σ /4 and 3 σ / 4 3\sigma/4 3 σ /4 , with an arbitrarily small adjustment of the radii if needed.
Now we choose χ j ∈ C ∞ ( Ω ) \chi_{j}\in C^{\infty}(\Omega) χ j ∈ C ∞ ( Ω ) with 0 ≤ χ j ≤ 1 0\le\chi_{j}\le1 0 ≤ χ j ≤ 1 , χ j = 0 \chi_{j}=0 χ j = 0 on a neighborhood of A j A_{j} A j , and χ j = 1 \chi_{j}=1 χ j = 1 on a neighborhood of G j G_{j} G j , and with derivatives bounded by constants depending only on σ \sigma σ and the fixed geometry. We then set d ~ n = d n − d 0 \widetilde{d}^{n}=d^{n}-d_{0} d n = d n − d 0 . Due to Tr ∂ Ω d n = 0 \operatorname{Tr}_{\partial\Omega}d^{n}=0 Tr ∂ Ω d n = 0 , χ j d ~ n \chi_{j}\widetilde{d}^{n} χ j d n belongs to H 0 1 ( Ω ) H_{0}^{1}(\Omega) H 0 1 ( Ω ) . Moreover, on supp χ j \operatorname{supp}\chi_{j} supp χ j the penalization vanishes for all large n n n , and the molecular-field identity (46) gives
χ j Δ d n = χ j ( M n + f ε ( d n ) ) , \chi_{j}\Delta d^{n}=\chi_{j}\left(\mathbf{M}^{n}+f_{\varepsilon}(d^{n})\right), χ j Δ d n = χ j ( M n + f ε ( d n ) ) ,
where we recall f ε ( d ) = − ε − 2 ( 1 − ∣ d ∣ 2 ) d f_\varepsilon(d)=-\varepsilon^{-2}(1-|d|^2)d f ε ( d ) = − ε − 2 ( 1 − ∣ d ∣ 2 ) d . The energy inequality and H 1 ( Ω ) ↪ L 6 ( Ω ) H^1(\Omega)\hookrightarrow L^6(\Omega) H 1 ( Ω ) ↪ L 6 ( Ω ) imply, for fixed ε \varepsilon ε , the bound sup n ∥ f ε ( d n ) ∥ L 2 ( 0 , T ; L 2 ( Ω ) ) ≤ C ε \sup_n\|f_\varepsilon(d^n)\|_{L^2(0,T;L^2(\Omega))}\le C_\varepsilon sup n ∥ f ε ( d n ) ∥ L 2 ( 0 , T ; L 2 ( Ω )) ≤ C ε . This shows the uniform boundedness of
Δ ( χ j d ~ n ) = χ j ( M n + f ε ( d n ) − Δ d 0 ) + 2 ∇ χ j ⋅ ∇ d ~ n + ( Δ χ j ) d ~ n \Delta(\chi_j\widetilde d^n)=\chi_j(M^n+f_\varepsilon(d^n)-\Delta d_0)+2\nabla\chi_j\cdot\nabla\widetilde d^n+(\Delta\chi_j)\widetilde d^n Δ ( χ j d n ) = χ j ( M n + f ε ( d n ) − Δ d 0 ) + 2∇ χ j ⋅ ∇ d n + ( Δ χ j ) d n
in L 2 ( I j ; L 2 ( Ω ) ) L^2(I_j;L^2(\Omega)) L 2 ( I j ; L 2 ( Ω )) . An H 2 H^2 H 2 -estimate on Ω \Omega Ω and restricting to G j G_j G j then yield sup n ∥ d n ∥ L 2 ( I j ; H 2 ( G j ) ) ≤ C σ , ε \sup_n\|d^n\|_{L^2(I_j;H^2(G_j))}\le C_{\sigma,\varepsilon} sup n ∥ d n ∥ L 2 ( I j ; H 2 ( G j )) ≤ C σ , ε . By the choice of B j B_j B j as smooth, G j G_j G j is a fixed C 2 C^2 C 2 -domain, so the compact embedding H 2 ( G j ) ⋐ H 1 ( G j ) H^2(G_j)\Subset H^1(G_j) H 2 ( G j ) ⋐ H 1 ( G j ) used below is standard.
For ∂ t d n \partial_t d^n ∂ t d n , we invoke the distributional form of the director equation ∂ t d n + ( u n ⋅ ∇ ) d n = M n \partial_t d^n+(u^n\cdot\nabla)d^n=M^n ∂ t d n + ( u n ⋅ ∇ ) d n = M n . Combining the boundedness of M n M^n M n in L t , x 2 L^2_{t,x} L t , x 2 , u n u^n u n in L t 2 H x 3 ↪ L t 2 L x ∞ L^2_tH^3_x\hookrightarrow L^2_tL^\infty_x L t 2 H x 3 ↪ L t 2 L x ∞ , and ∇ d n \nabla d^n ∇ d n in L x ∞ L x 2 L^\infty_xL^2_x L x ∞ L x 2 , we get sup n ∥ ∂ t d n ∥ L 2 ( 0 , T ; L 2 ( Ω ) ) ≤ C δ , ε \sup_n\|\partial_t d^n\|_{L^2(0,T;L^2(\Omega))}\le C_{\delta,\varepsilon} sup n ∥ ∂ t d n ∥ L 2 ( 0 , T ; L 2 ( Ω )) ≤ C δ , ε . Applying the Aubin–Lions lemma on each fixed domain G j G_j G j , where we exploit H 2 ( G j ) ⋐ H 1 ( G j ) ↪ L 2 ( G j ) H^2(G_j)\Subset H^1(G_j)\hookrightarrow L^2(G_j) H 2 ( G j ) ⋐ H 1 ( G j ) ↪ L 2 ( G j ) , gives d n → d d^n\to d d n → d strongly in L 2 ( I j ; H 1 ( G j ) ) L^2(I_j;H^1(G_j)) L 2 ( I j ; H 1 ( G j )) . By (A.12), the spacetime ( 0 , T ) × F ~ σ ( t ) (0,T)\times\widetilde F_\sigma(t) ( 0 , T ) × F σ ( t ) is contained in the union of these fixed cylinders. Summing over the finite covering proves d n → d d^n\to d d n → d strongly in L 2 ( 0 , T ; H 1 ( F ~ σ ( t ) ) ) L^2(0,T;H^1(\widetilde F_\sigma(t))) L 2 ( 0 , T ; H 1 ( F σ ( t ))) , which is the assertion of Lemma 3.10 . □
We next discuss the strong convergence of velocity fields ( v j ) (v_j) ( v j ) .
Lemma A.6. Let ( v j ) (v_j) ( v j ) be bounded in L ∞ ( 0 , T ; L 2 ( Ω ; R 3 ) ) L^\infty(0,T;L^2(\Omega;\mathbb{R}^3)) L ∞ ( 0 , T ; L 2 ( Ω ; R 3 )) with v j ⇀ v v_j\rightharpoonup v v j ⇀ v weakly in L 2 ( ( 0 , T ) × Ω ; R 3 ) L^2((0,T)\times\Omega;\mathbb{R}^3) L 2 (( 0 , T ) × Ω ; R 3 ) . Suppose that for every η > 0 \eta>0 η > 0 and γ ∈ ( 0 , T / 2 ) \gamma\in(0,T/2) γ ∈ ( 0 , T /2 ) there are a partition γ = s 0 < s 1 < ⋯ < s N = T − γ \gamma=s_0<s_1<\cdots<s_N=T-\gamma γ = s 0 < s 1 < ⋯ < s N = T − γ with intervals I ℓ : = ( s ℓ − 1 , s ℓ ) I_\ell:=(s_{\ell-1},s_\ell) I ℓ := ( s ℓ − 1 , s ℓ ) , connected open sets A ℓ ⋐ Ω A_\ell\Subset\Omega A ℓ ⋐ Ω with C 2 C^2 C 2 -boundary, an index j η j_\eta j η , a constant C η C_\eta C η , and fields z j , ℓ ∈ L 2 ( I ℓ ; K ( A ℓ ) ) z_{j,\ell}\in L^2(I_\ell;\mathcal K(A_\ell)) z j , ℓ ∈ L 2 ( I ℓ ; K ( A ℓ )) , 1 ≤ ℓ ≤ N 1\le\ell\le N 1 ≤ ℓ ≤ N , such that, for all j ≥ j η j\ge j_\eta j ≥ j η ,
∑ ℓ = 1 N ∫ I ℓ ∥ v j − z j , ℓ ∥ L 2 ( Ω ) 2 d t ≤ η , ∑ ℓ = 1 N ∫ I ℓ ∥ z j , ℓ ∥ H 1 ( Ω ) 2 d t ≤ C η , (161) \sum_{\ell=1}^{N}\int_{I_\ell}\|v_j-z_{j,\ell}\|_{L^2(\Omega)}^2\,\mathrm{d}t\le\eta,\qquad
\sum_{\ell=1}^{N}\int_{I_\ell}\|z_{j,\ell}\|_{H^1(\Omega)}^2\,\mathrm{d}t\le C_\eta, \tag*{(161)} ℓ = 1 ∑ N ∫ I ℓ ∥ v j − z j , ℓ ∥ L 2 ( Ω ) 2 d t ≤ η , ℓ = 1 ∑ N ∫ I ℓ ∥ z j , ℓ ∥ H 1 ( Ω ) 2 d t ≤ C η , (161) and such that, for every ℓ \ell ℓ and every e ∈ K ( A ℓ ) ∩ C c ∞ ( Ω ; R 3 ) e\in\mathcal K(A_\ell)\cap C^\infty_c(\Omega;\mathbb{R}^3) e ∈ K ( A ℓ ) ∩ C c ∞ ( Ω ; R 3 ) , there is a constant C η , ℓ , e C_{\eta,\ell,e} C η , ℓ , e with
∣ ∫ I ℓ ( v j ( t ) , e ) L 2 ( Ω ) ϑ ′ ( t ) d t ∣ ≤ C η , ℓ , e ∥ ϑ ∥ L ∞ ( I ℓ ) \left|\int_{I_\ell}(v_j(t),e)_{L^2(\Omega)}\vartheta'(t)\,\mathrm{d}t\right|\le C_{\eta,\ell,e}\|\vartheta\|_{L^\infty(I_\ell)} ∫ I ℓ ( v j ( t ) , e ) L 2 ( Ω ) ϑ ′ ( t ) d t ≤ C η , ℓ , e ∥ ϑ ∥ L ∞ ( I ℓ ) for ϑ ∈ C c ∞ ( I ℓ ) \vartheta\in C^\infty_c(I_\ell) ϑ ∈ C c ∞ ( I ℓ ) and j ≥ j η j\ge j_\eta j ≥ j η . Then v j → v v_j\to v v j → v strongly in L 2 ( ( 0 , T ) × Ω ; R 3 ) L^2((0,T)\times\Omega;\mathbb{R}^3) L 2 (( 0 , T ) × Ω ; R 3 ) .
Proof. We define V ‾ : = sup j ∥ v j ∥ L ∞ ( 0 , T ; L 2 ( Ω ) ) < ∞ \overline V:=\sup_j\|v_j\|_{L^\infty(0,T;L^2(\Omega))}<\infty V := sup j ∥ v j ∥ L ∞ ( 0 , T ; L 2 ( Ω )) < ∞ . Next, we fix γ ∈ ( 0 , T / 2 ) \gamma\in(0,T/2) γ ∈ ( 0 , T /2 ) and η > 0 \eta>0 η > 0 , and choose the partition, sets, and approximating fields from the hypotheses.
For each ℓ \ell ℓ , the space K ( A ℓ ) ∩ C c ∞ ( Ω ; R 3 ) \mathcal K(A_\ell)\cap C^\infty_c(\Omega;\mathbb{R}^3) K ( A ℓ ) ∩ C c ∞ ( Ω ; R 3 ) is dense in K ( A ℓ ) \mathcal K(A_\ell) K ( A ℓ ) in the H 1 H^1 H 1 -norm. Indeed, choose a function χ ℓ ∈ C c ∞ ( Ω ) \chi_\ell\in C^\infty_c(\Omega) χ ℓ ∈ C c ∞ ( Ω ) that equals one near A ℓ A_\ell A ℓ . For an affine rigid field R ( x ) = a + ω × x R(x)=a+\omega\times x R ( x ) = a + ω × x , set
E ℓ R : = ∇ × [ χ ℓ ( 1 2 a × x − 1 2 ∣ x ∣ 2 ω ) ] . E_\ell R:=\nabla\times\left[\chi_\ell\left(\frac{1}{2}a\times x-\frac{1}{2}|x|^2\omega\right)\right]. E ℓ R := ∇ × [ χ ℓ ( 2 1 a × x − 2 1 ∣ x ∣ 2 ω ) ] .
This field is smooth, solenoidal, compactly supported in Ω \Omega Ω , and equal to R R R near A ℓ ‾ \overline{A_\ell} A ℓ . For z ∈ K ( A ℓ ) z\in\mathcal K(A_\ell) z ∈ K ( A ℓ ) , we invoke that A ℓ A_\ell A ℓ is connected to identify the restriction there with a rigid field R z R_z R z . As a consequence, z − E ℓ R z z-E_\ell R_z z − E ℓ R z vanishes in A ℓ A_\ell A ℓ and belongs to the solenoidal subspace of H 0 1 ( Ω ∖ A ℓ ‾ ; R 3 ) H^1_0(\Omega\setminus\overline{A_\ell};\mathbb{R}^3) H 0 1 ( Ω ∖ A ℓ ; R 3 ) . It readily follows from a density argument on this C 2 C^2 C 2 open set that there are approximations in C c ∞ ( Ω ∖ A ℓ ‾ ; R 3 ) C^\infty_c(\Omega\setminus\overline{A_\ell};\mathbb{R}^3) C c ∞ ( Ω ∖ A ℓ ; R 3 ) . Extending them by zero and adding E ℓ R z E_\ell R_z E ℓ R z proves the claimed density.
We then orthonormalize a countable subset that is dense in H 1 H^1 H 1 to find an L 2 L^2 L 2 -orthonormal basis ( e ℓ , m ) m ≥ 1 (e_{\ell,m})_{m\ge1} ( e ℓ , m ) m ≥ 1 of the L 2 L^2 L 2 -closure of K ( A ℓ ) \mathcal K(A_\ell) K ( A ℓ ) , with e ℓ , m ∈ K ( A ℓ ) ∩ C c ∞ ( Ω ; R 3 ) e_{\ell,m}\in\mathcal K(A_\ell)\cap C^\infty_c(\Omega;\mathbb{R}^3) e ℓ , m ∈ K ( A ℓ ) ∩ C c ∞ ( Ω ; R 3 ) . Let P ℓ , M P_{\ell,M} P ℓ , M be the orthogonal projection onto Span { e ℓ , 1 , … , e ℓ , M } \operatorname{Span}\{e_{\ell,1},\ldots,e_{\ell,M}\} Span { e ℓ , 1 , … , e ℓ , M } . Applying the Rellich–Kondrachov compactness theorem and by strong convergence of P ℓ , M P_{\ell,M} P ℓ , M , we have
α ℓ , M : = sup z ∈ K ( A ℓ ) ∥ z ∥ H 1 ( Ω ) ≤ 1 ∥ ( I − P ℓ , M ) z ∥ L 2 ( Ω ) ⟶ 0 for M → ∞ . \alpha_{\ell,M}:=\sup_{\substack{z\in\mathcal K(A_\ell)\\ \|z\|_{H^1(\Omega)}\le1}}\|(I-P_{\ell,M})z\|_{L^2(\Omega)}\longrightarrow0\quad\text{for }M\to\infty. α ℓ , M := z ∈ K ( A ℓ ) ∥ z ∥ H 1 ( Ω ) ≤ 1 sup ∥ ( I − P ℓ , M ) z ∥ L 2 ( Ω ) ⟶ 0 for M → ∞.
For every ℓ \ell ℓ and m m m , the scalar functions t ↦ ( v j ( t ) , e ℓ , m ) L 2 ( Ω ) t \mapsto(v_j(t),e_{\ell,m})_{L^2(\Omega)} t ↦ ( v j ( t ) , e ℓ , m ) L 2 ( Ω ) are uniformly bounded in L ∞ ( I ℓ ) L^\infty(I_\ell) L ∞ ( I ℓ ) and, by (A.14), in B V ( I ℓ ) BV(I_\ell) B V ( I ℓ ) for j ≥ j η j \ge j_\eta j ≥ j η . Compactness in B V ( I ℓ ) BV(I_\ell) B V ( I ℓ ) , together with the uniform L ∞ L^\infty L ∞ -bound, yields relative compactness in L 2 ( I ℓ ) L^2(I_\ell) L 2 ( I ℓ ) . Consequently, for fixed M M M , the fields
V j , M ( t ) : = { P ℓ , M v j ( t ) , t ∈ I ℓ , 1 ≤ ℓ ≤ N , 0 , t ∈ ( 0 , γ ) ∪ ( T − γ , T ) , V_{j,M}(t):=
\begin{cases}
P_{\ell,M}v_j(t), & t\in I_\ell,\quad1\le\ell\le N,\\
0, & t\in(0,\gamma)\cup(T-\gamma,T),
\end{cases} V j , M ( t ) := { P ℓ , M v j ( t ) , 0 , t ∈ I ℓ , 1 ≤ ℓ ≤ N , t ∈ ( 0 , γ ) ∪ ( T − γ , T ) ,
form a relatively compact family in L 2 ( ( 0 , T ) × Ω ; R 3 ) L^2((0,T)\times\Omega;\mathbb{R}^3) L 2 (( 0 , T ) × Ω ; R 3 ) .
Since I − P ℓ , M I-P_{\ell,M} I − P ℓ , M is a contraction and z j , ℓ ( t ) ∈ K ( A ℓ ) z_{j,\ell}(t)\in K(A_\ell) z j , ℓ ( t ) ∈ K ( A ℓ ) , for almost every t ∈ I ℓ t\in I_\ell t ∈ I ℓ ,
∥ v j − P ℓ , M v j ∥ L 2 ( Ω ) ≤ ∥ v j − z j , ℓ ∥ L 2 ( Ω ) + α ℓ , M ∥ z j , ℓ ∥ H 1 ( Ω ) . \lVert v_j-P_{\ell,M}v_j\rVert_{L^2(\Omega)}
\le\lVert v_j-z_{j,\ell}\rVert_{L^2(\Omega)}
+\alpha_{\ell,M}\lVert z_{j,\ell}\rVert_{H^1(\Omega)}. ∥ v j − P ℓ , M v j ∥ L 2 ( Ω ) ≤ ∥ v j − z j , ℓ ∥ L 2 ( Ω ) + α ℓ , M ∥ z j , ℓ ∥ H 1 ( Ω ) .
Squaring, integrating, and using (A.13) and the definition of V ‾ \overline{V} V , we obtain
∥ v j − V j , M ∥ L 2 ( ( 0 , T ) × Ω ) 2 ≤ 2 η + 2 C η max 1 ≤ ℓ ≤ N α ℓ , M 2 + 2 γ V ‾ 2 for j ≥ j η . \lVert v_j-V_{j,M}\rVert_{L^2((0,T)\times\Omega)}^2
\le2\eta+2C\eta\max_{1\le\ell\le N}\alpha_{\ell,M}^2+2\gamma\overline{V}^2
\quad\text{for }j\ge j_\eta. ∥ v j − V j , M ∥ L 2 (( 0 , T ) × Ω ) 2 ≤ 2 η + 2 C η 1 ≤ ℓ ≤ N max α ℓ , M 2 + 2 γ V 2 for j ≥ j η .
Thus, by first choosing γ \gamma γ and η \eta η sufficiently small and then M M M sufficiently large, the tail of ( v j ) (v_j) ( v j ) can be approximated arbitrarily closely by a relatively compact family. Adding the finitely many omitted fields shows that ( v j ) (v_j) ( v j ) is totally bounded, hence relatively compact, in L 2 ( ( 0 , T ) × Ω ; R 3 ) L^2((0,T)\times\Omega;\mathbb{R}^3) L 2 (( 0 , T ) × Ω ; R 3 ) . Every strong cluster point equals v v v by the assumed weak convergence. Therefore the whole sequence converges strongly to v v v . □
Proof of Lemma 3.11 . We combine approximate rigidity with temporal compactness in fixed spaces of solenoidal fields. The argument follows the restriction-operator strategy of [45 Theorem 3.2] and [16 ] . Throughout the proof, C C C is a constant that is independent of n n n , the comparison-set index j j j , and sufficiently small σ \sigma σ , whereas constants denoted by C σ C_\sigma C σ may depend on σ \sigma σ . All pointwise-in-time estimates hold for almost every t t t .
We first show that the sets S n ( t ) S_n(t) S n ( t ) have a uniformly controlled geometry and that v n v_n v n is close to a rigid velocity field on them. By (84) and Morrey’s embedding, the sets S n ( t ) S_n(t) S n ( t ) enjoy a uniform C 1 , α C^{1,\alpha} C 1 , α -Hölder regularity for every 0 < α < 1 / 2 0<\alpha<1/2 0 < α < 1/2 . Let R n ( t , x ) = a n ( t ) + ω n ( t ) × x R_n(t,x)=a_n(t)+\omega_n(t)\times x R n ( t , x ) = a n ( t ) + ω n ( t ) × x be the L 2 ( S n ( t ) ) L^2(S_n(t)) L 2 ( S n ( t )) -orthogonal projection of v n ( t ) v_n(t) v n ( t ) onto the space of affine rigid velocities, and set q n : = v n − R n q_n:=v_n-R_n q n := v n − R n . Since the Gram matrices associated with the rigid motion are uniformly coercive, we get the estimate ∣ a n ( t ) ∣ + ∣ ω n ( t ) ∣ ≤ C ∥ v n ( t ) ∥ L 2 ( S n ( t ) ) |a_n(t)|+|\omega_n(t)|\le C\lVert v_n(t)\rVert_{L^2(S_n(t))} ∣ a n ( t ) ∣ + ∣ ω n ( t ) ∣ ≤ C ∥ v n ( t ) ∥ L 2 ( S n ( t )) . By Korn’s inequality, with the rigid motion chosen by this projection, we obtain
∥ q n ( t ) ∥ H 1 ( S n ( t ) ) ≤ C ∥ D ( v n ( t ) ) ∥ L 2 ( S n ( t ) ) . (162) \lVert q_n(t)\rVert_{H^1(S_n(t))}
\le C\lVert\mathbb{D}(v_n(t))\rVert_{L^2(S_n(t))}. \tag*{(162)} ∥ q n ( t ) ∥ H 1 ( S n ( t )) ≤ C ∥ D ( v n ( t )) ∥ L 2 ( S n ( t )) . (162)
To justify the choice of projection, one may first apply Korn’s inequality with any minimizing rigid motion and then use the L 2 L^2 L 2 -minimality of R n R_n R n and the uniform equivalence of norms on the six-dimensional rigid-motion space.
On this uniformly Lipschitz family, the Sobolev extension operators are bounded simultaneously at the H 1 H^1 H 1 - and H 3 H^3 H 3 -levels, with uniform constants. By interpolation of the extensions, the embedding H 2 ( R 3 ) → L ∞ ( R 3 ) H^2(\mathbb{R}^3)\to L^\infty(\mathbb{R}^3) H 2 ( R 3 ) → L ∞ ( R 3 ) , and (A.15), we have
∥ q n ( t ) ∥ L ∞ ( S n ( t ) ) ≤ C ∥ q n ( t ) ∥ H 1 ( S n ( t ) ) 1 / 2 ∥ q n ( t ) ∥ H 3 ( S n ( t ) ) 1 / 2 ≤ C ∥ D ( v n ( t ) ) ∥ L 2 ( S n ( t ) ) 1 / 2 ∥ v n ( t ) ∥ H 3 ( Ω ) 1 / 2 . \lVert q_n(t)\rVert_{L^\infty(S_n(t))}
\le C\lVert q_n(t)\rVert_{H^1(S_n(t))}^{1/2}\lVert q_n(t)\rVert_{H^3(S_n(t))}^{1/2}
\le C\lVert\mathbb{D}(v_n(t))\rVert_{L^2(S_n(t))}^{1/2}\lVert v_n(t)\rVert_{H^3(\Omega)}^{1/2}. ∥ q n ( t ) ∥ L ∞ ( S n ( t )) ≤ C ∥ q n ( t ) ∥ H 1 ( S n ( t )) 1/2 ∥ q n ( t ) ∥ H 3 ( S n ( t )) 1/2 ≤ C ∥ D ( v n ( t )) ∥ L 2 ( S n ( t )) 1/2 ∥ v n ( t ) ∥ H 3 ( Ω ) 1/2 .
Here the rigid-coefficient bound controls ∥ q n ( t ) ∥ H 3 \lVert q_n(t)\rVert_{H^3} ∥ q n ( t ) ∥ H 3 by C ∥ v n ( t ) ∥ H 3 ( Ω ) C\lVert v_n(t)\rVert_{H^3(\Omega)} C ∥ v n ( t ) ∥ H 3 ( Ω ) . By the same bound, together with Sobolev embedding on the fixed container, we have
∥ q n ( t ) ∥ L ∞ ( Ω ) + ∥ q n ( t ) ∥ C 0 , 1 / 2 ( Ω ‾ ) ≤ C ∥ v n ( t ) ∥ H 3 ( Ω ) . \lVert q_n(t)\rVert_{L^\infty(\Omega)}
+\lVert q_n(t)\rVert_{C^{0,1/2}(\overline{\Omega})}
\le C\lVert v_n(t)\rVert_{H^3(\Omega)}. ∥ q n ( t ) ∥ L ∞ ( Ω ) + ∥ q n ( t ) ∥ C 0 , 1/2 ( Ω ) ≤ C ∥ v n ( t ) ∥ H 3 ( Ω ) .
Applying the penalty estimate (86) and Hölder’s inequality in time, together with the uniform L t 2 H x 3 L_t^2H_x^3 L t 2 H x 3 -bound recorded in (85) , we obtain
∫ 0 T ∥ D ( v n ( t ) ) ∥ L 2 ( S n ( t ) ) 2 d t ≤ C n − 1 , ∫ 0 T ∥ D ( v n ( t ) ) ∥ L 2 ( S n ( t ) ) ∥ v n ( t ) ∥ H 3 ( Ω ) d t ≤ C n − 1 / 2 . (163) \int_0^T\lVert\mathbb{D}(v_n(t))\rVert_{L^2(S_n(t))}^2\,dt\le Cn^{-1},
\qquad
\int_0^T\lVert\mathbb{D}(v_n(t))\rVert_{L^2(S_n(t))}
\lVert v_n(t)\rVert_{H^3(\Omega)}\,dt\le Cn^{-1/2}. \tag*{(163)} ∫ 0 T ∥ D ( v n ( t )) ∥ L 2 ( S n ( t )) 2 d t ≤ C n − 1 , ∫ 0 T ∥ D ( v n ( t )) ∥ L 2 ( S n ( t )) ∥ v n ( t ) ∥ H 3 ( Ω ) d t ≤ C n − 1/2 . (163)
These estimates will control the integrated restriction error.
We now fix a sufficiently small σ > 0 \sigma> 0 σ > 0 and take the intervals I j I_j I j and comparison sets S j S_j S j supplied by the hypotheses. Upon refining the covering, we may assume that the I j I_j I j are consecutive intervals with disjoint interiors. Increasing n σ n_\sigma n σ if necessary, we also find that d H ( S n ( t ) ‾ , S ( t ) ) ≤ σ d_H(\overline{S_n(t)},S(t)) \le\sigma d H ( S n ( t ) , S ( t )) ≤ σ uniformly in t t t . Since the volume of the fluid is preserved, we have ∣ S n ( t ) ∣ = ∣ S ( t ) ∣ = ∣ S 0 ∣ |S_n(t)|=|S(t)|=|S_0| ∣ S n ( t ) ∣ = ∣ S ( t ) ∣ = ∣ S 0 ∣ . On the other hand, we invoke that S n ( t ) ⊂ S j ⊂ S ( t ) σ S_n(t)\subset S_j\subset S(t)_\sigma S n ( t ) ⊂ S j ⊂ S ( t ) σ and ( S j ) σ ⊂ S ( t ) 2 σ (S_j)_\sigma\subset S(t)_{2\sigma} ( S j ) σ ⊂ S ( t ) 2 σ for t ∈ I j t\in I_j t ∈ I j ,
S n ( t ) ⊂ S j ⊂ S ( t ) σ , ( S j ) σ ⊂ S ( t ) 2 σ ( t ∈ I j ) , S_n(t)\subset S_j\subset S(t)_\sigma,\qquad(S_j)_\sigma\subset S(t)_{2\sigma}\qquad(t\in I_j), S n ( t ) ⊂ S j ⊂ S ( t ) σ , ( S j ) σ ⊂ S ( t ) 2 σ ( t ∈ I j ) ,
as well as the associated volume estimates
∣ S j ∖ S n ( t ) ∣ ≤ C σ , ∣ ( S j ) σ ∖ S n ( t ) ∣ ≤ C σ . (164) |S_j\setminus S_n(t)|\le C\sigma,\qquad|(S_j)_\sigma\setminus S_n(t)|\le C\sigma. \tag*{(164)} ∣ S j ∖ S n ( t ) ∣ ≤ C σ , ∣ ( S j ) σ ∖ S n ( t ) ∣ ≤ C σ . (164)
Moreover, every point of ∂ S j \partial S_j ∂ S j is within distance 2 σ 2\sigma 2 σ of S n ( t ) ‾ \overline{S_n(t)} S n ( t ) . The preceding L ∞ L^\infty L ∞ and Hölder estimates, and the uniform surface-area bound for ∂ S j \partial S_j ∂ S j , thus yield
∥ q n ( t ) ∥ L 2 ( ∂ S j ) ≤ C ∥ D ( v n ( t ) ) ∥ L 2 ( S n ( t ) ) 1 / 2 ∥ v n ( t ) ∥ H 3 ( Ω ) 1 / 2 + C σ 1 / 2 ∥ v n ( t ) ∥ H 3 ( Ω ) . (165) \|q_n(t)\|_{L^2(\partial S_j)}\le C\|\mathbb{D}(v_n(t))\|_{L^2(S_n(t))}^{1/2}\|v_n(t)\|_{H^3(\Omega)}^{1/2}+C\sigma^{1/2}\|v_n(t)\|_{H^3(\Omega)}. \tag*{(165)} ∥ q n ( t ) ∥ L 2 ( ∂ S j ) ≤ C ∥ D ( v n ( t )) ∥ L 2 ( S n ( t )) 1/2 ∥ v n ( t ) ∥ H 3 ( Ω ) 1/2 + C σ 1/2 ∥ v n ( t ) ∥ H 3 ( Ω ) . (165)
Next, on each interval I j I_j I j , we replace v n v_n v n by a field that is exactly rigid on the fixed set S j S_j S j and remains close to v n v_n v n . Choose a smooth cutoff η j , σ \eta_{j,\sigma} η j , σ that is zero near S j ‾ \overline{S_j} S j , equals one outside ( S j ) σ (S_j)_\sigma ( S j ) σ , and satisfies 0 ≤ η j , σ ≤ 1 0\le\eta_{j,\sigma}\le1 0 ≤ η j , σ ≤ 1 and ∣ ∇ η j , σ ∣ ≤ C σ − 1 |\nabla\eta_{j,\sigma}|\le C\sigma^{-1} ∣∇ η j , σ ∣ ≤ C σ − 1 . Set z n , j 0 : = R n + η j , σ q n z^0_{n,j}:=R_n+\eta_{j,\sigma}q_n z n , j 0 := R n + η j , σ q n and div z n , j 0 = ∇ η j , σ ⋅ q n \operatorname{div}z^0_{n,j}=\nabla\eta_{j,\sigma}\cdot q_n div z n , j 0 = ∇ η j , σ ⋅ q n . Then z n , j 0 z^0_{n,j} z n , j 0 equals R n R_n R n on S j S_j S j and v n v_n v n outside ( S j ) σ (S_j)_\sigma ( S j ) σ . Its divergence is compactly supported in the fixed-width collar D j = ( S j ) r ∗ ∖ S j ‾ D_j=(S_j)_{r_*}\setminus\overline{S_j} D j = ( S j ) r ∗ ∖ S j , provided σ < r ∗ \sigma<r_* σ < r ∗ . The integral of div z n , j 0 \operatorname{div}z^0_{n,j} div z n , j 0 vanishes on every connected component of D j D_j D j . Indeed, the field equals R n R_n R n on the inner boundary and v n v_n v n on the outer boundary, and both have zero flux through each closed boundary surface. For v n v_n v n , this follows by extending the H 0 1 ( Ω ) H^1_0(\Omega) H 0 1 ( Ω ) -field by zero to R 3 \mathbb{R}^3 R 3 , which preserves its divergence-free property.
On D j D_j D j , choose a Bogovskiĭ right inverse B j \mathcal{B}_j B j . Besides the usual L 0 2 → H 0 1 L^2_0\to H^1_0 L 0 2 → H 0 1 -bound, we use its negative-order estimate, see [23 ] , with s = − 1 s=-1 s = − 1 and p = 2 p=2 p = 2 , namely
∥ B j f ∥ L 2 ( D j ) ≤ C ∣ f ∣ ∗ , D j , ∣ f ∣ ∗ , D j : = sup φ ∈ H 1 ( D j ) , ∫ D j φ d x = 0 ∥ ∇ φ ∥ L 2 ( D j ) ≤ 1 ∣ ∫ D j f φ d x ∣ . (166) \|\mathcal{B}_j f\|_{L^2(D_j)}\le C|f|_{*,D_j},\qquad|f|_{*,D_j}:=\sup_{\substack{\varphi\in H^1(D_j),\\ \int_{D_j}\varphi\,dx=0\\ \|\nabla\varphi\|_{L^2(D_j)}\le1}}\left|\int_{D_j}f\varphi\,dx\right|. \tag*{(166)} ∥ B j f ∥ L 2 ( D j ) ≤ C ∣ f ∣ ∗ , D j , ∣ f ∣ ∗ , D j := φ ∈ H 1 ( D j ) , ∫ D j φ d x = 0 ∥∇ φ ∥ L 2 ( D j ) ≤ 1 sup ∫ D j f φ d x . (166)
This is the case s = − 1 s=-1 s = − 1 , p = 2 p=2 p = 2 of [23 Theorem 2.5] . There, the source space is the dual of the full space H 1 ( D j ) H^1(D_j) H 1 ( D j ) , restricted to functionals that annihilate constants (and not H − 1 ( D j ) = ( H 0 1 ( D j ) ) ′ H^{-1}(D_j)=(H^1_0(D_j))' H − 1 ( D j ) = ( H 0 1 ( D j ) ) ′ ), and Poincaré’s inequality identifies its norm with ∣ ⋅ ∣ ∗ , D j |\cdot|_{*,D_j} ∣ ⋅ ∣ ∗ , D j . If D j D_j D j is disconnected, the operator and the zero-mean normalization are understood componentwise. The constants in both Bogovskiĭ estimates, and in the Poincaré and trace inequalities below, are uniform, since the fixed-width collars considered here admit a uniformly controlled finite atlas (see also [5 ] and [21 Chap. III.3] ).
Let n j n_j n j be the outward normal to S j S_j S j . Since div q n = 0 \operatorname{div}q_n=0 div q n = 0 and η j , σ = 1 \eta_{j,\sigma}=1 η j , σ = 1 near the outer boundary of D j D_j D j , integration by parts gives
∫ D j div z n , j 0 φ d x = − ∫ D j ( η j , σ − 1 ) q n ⋅ ∇ φ d x + ∫ ∂ S j ( q n ⋅ n j ) φ d σ . \int_{D_j}\operatorname{div}z^0_{n,j}\varphi\,dx=-\int_{D_j}(\eta_{j,\sigma}-1)q_n\cdot\nabla\varphi\,dx+\int_{\partial S_j}(q_n\cdot n_j)\varphi\,d\sigma. ∫ D j div z n , j 0 φ d x = − ∫ D j ( η j , σ − 1 ) q n ⋅ ∇ φ d x + ∫ ∂ S j ( q n ⋅ n j ) φ d σ .
Consequently, retaining the inner-boundary term and applying the trace and Poincaré inequalities, we obtain
∣ div z n , j 0 ∣ ∗ , D j ≤ C ( ∥ q n ∥ L 2 ( ( S j ) σ ∖ S j ‾ ) + ∥ q n ∥ L 2 ( ∂ S j ) ) . |\operatorname{div}z^0_{n,j}|_{*,D_j}\le C\left(\|q_n\|_{L^2((S_j)_\sigma\setminus\overline{S_j})}+\|q_n\|_{L^2(\partial S_j)}\right). ∣ div z n , j 0 ∣ ∗ , D j ≤ C ( ∥ q n ∥ L 2 (( S j ) σ ∖ S j ) + ∥ q n ∥ L 2 ( ∂ S j ) ) .
The first term is at most C σ 1 / 2 ∥ v n ( t ) ∥ H 3 ( Ω ) C\sigma^{1/2}\|v_n(t)\|_{H^3(\Omega)} C σ 1/2 ∥ v n ( t ) ∥ H 3 ( Ω ) by (A.17) and the global L ∞ L^\infty L ∞ -bound for q n q_n q n . Combining this with (A.18) and (A.19) yields
∥ B j div z n , j 0 ( t ) ∥ L 2 ( D j ) ≤ C ∥ D ( v n ( t ) ) ∥ L 2 ( S n ( t ) ) 1 / 2 ∥ v n ( t ) ∥ H 3 ( Ω ) 1 / 2 + C σ 1 / 2 ∥ v n ( t ) ∥ H 3 ( Ω ) . (167) \|\mathcal{B}_j\operatorname{div}z^0_{n,j}(t)\|_{L^2(D_j)}\le C\|\mathbb{D}(v_n(t))\|_{L^2(S_n(t))}^{1/2}\|v_n(t)\|_{H^3(\Omega)}^{1/2}+C\sigma^{1/2}\|v_n(t)\|_{H^3(\Omega)}. \tag*{(167)} ∥ B j div z n , j 0 ( t ) ∥ L 2 ( D j ) ≤ C ∥ D ( v n ( t )) ∥ L 2 ( S n ( t )) 1/2 ∥ v n ( t ) ∥ H 3 ( Ω ) 1/2 + C σ 1/2 ∥ v n ( t ) ∥ H 3 ( Ω ) . (167)
Extend B j div z n , j 0 \mathcal{B}_j\operatorname{div}z^0_{n,j} B j div z n , j 0 by zero to Ω \Omega Ω and define z n , j : = z n , j 0 − B j div z n , j 0 z_{n,j}:=z^0_{n,j}-\mathcal{B}_j\operatorname{div}z^0_{n,j} z n , j := z n , j 0 − B j div z n , j 0 . The zero trace of the correction ensures z n , j ( t ) ∈ K ( S j ) z_{n,j}(t)\in\mathcal{K}(S_j) z n , j ( t ) ∈ K ( S j ) . The ordinary Bogovskiĭ estimate joint with the estimate
∥ div z n , j 0 ( t ) ∥ L 2 ( D j ) ≤ C σ − 1 ∥ v n ( t ) ∥ H 3 ( Ω ) gives ∥ z n , j ( t ) ∥ H 1 ( Ω ) ≤ C σ ∥ v n ( t ) ∥ H 3 ( Ω ) . (168) \|\operatorname{div} z_{n,j}^{0}(t)\|_{L^{2}(D_j)} \le C\sigma^{-1}\|v_n(t)\|_{H^{3}(\Omega)} \text{ gives}
\qquad
\|z_{n,j}(t)\|_{H^{1}(\Omega)} \le C_{\sigma}\|v_n(t)\|_{H^{3}(\Omega)}. \tag*{(168)} ∥ div z n , j 0 ( t ) ∥ L 2 ( D j ) ≤ C σ − 1 ∥ v n ( t ) ∥ H 3 ( Ω ) gives ∥ z n , j ( t ) ∥ H 1 ( Ω ) ≤ C σ ∥ v n ( t ) ∥ H 3 ( Ω ) . (168)
Furthermore, v n − z n , j 0 = ( 1 − η j , σ ) q n v_n-z_{n,j}^{0}=(1-\eta_{j,\sigma})q_n v n − z n , j 0 = ( 1 − η j , σ ) q n is supported in ( S j ) σ (S_j)_{\sigma} ( S j ) σ . On S n ( t ) S_n(t) S n ( t ) , its L 2 L^{2} L 2 -norm is controlled by (A.15). On the remaining set, (A.17) and the global L ∞ L^{\infty} L ∞ -bound give C σ 1 / 2 ∥ v n ( t ) ∥ H 3 ( Ω ) C\sigma^{1/2}\|v_n(t)\|_{H^{3}(\Omega)} C σ 1/2 ∥ v n ( t ) ∥ H 3 ( Ω ) . Together with the above bound (A.20), for t ∈ I j t\in I_j t ∈ I j and n ≥ n σ n\ge n_{\sigma} n ≥ n σ , this proves
∥ v n ( t ) − z n , j ( t ) ∥ L 2 ( Ω ) ≤ C ∥ D ( v n ( t ) ) ∥ L 2 ( S n ( t ) ) + C ∥ D ( v n ( t ) ) ∥ L 2 ( S n ( t ) ) 1 / 2 ∥ v n ( t ) ∥ H 3 ( Ω ) 1 / 2 + C σ 1 / 2 ∥ v n ( t ) ∥ H 3 ( Ω ) . (169) \begin{aligned}
\|v_n(t)-z_{n,j}(t)\|_{L^{2}(\Omega)}
&\le C\|D(v_n(t))\|_{L^{2}(S_n(t))}
+C\|D(v_n(t))\|_{L^{2}(S_n(t))}^{1/2}\|v_n(t)\|_{H^{3}(\Omega)}^{1/2}\\
&\qquad+C\sigma^{1/2}\|v_n(t)\|_{H^{3}(\Omega)}. \tag*{(169)}
\end{aligned} ∥ v n ( t ) − z n , j ( t ) ∥ L 2 ( Ω ) ≤ C ∥ D ( v n ( t )) ∥ L 2 ( S n ( t )) + C ∥ D ( v n ( t )) ∥ L 2 ( S n ( t )) 1/2 ∥ v n ( t ) ∥ H 3 ( Ω ) 1/2 + C σ 1/2 ∥ v n ( t ) ∥ H 3 ( Ω ) . (169)
It remains to verify the hypotheses of Lemma A.6 with A j : = S j A_j:=S_j A j := S j and the fields z n , j z_{n,j} z n , j constructed above. Since the hypotheses of the present lemma hold on every subinterval, they may be restricted to ( γ , T − γ ) (\gamma,T-\gamma) ( γ , T − γ ) for a given γ ∈ ( 0 , T / 2 ) \gamma\in(0,T/2) γ ∈ ( 0 , T /2 ) . Squaring (A.22), integrating over the intervals I j I_j I j , which have disjoint interiors, and using the integrated estimates (A.16) together with (85) and (A.21), we obtain, for n ≥ n σ n\ge n_{\sigma} n ≥ n σ , the bounds
∑ j ∫ I j ∥ v n − z n , j ∥ L 2 ( Ω ) 2 d t ≤ C ( n − 1 + n − 1 / 2 + σ ) and ∑ j ∫ I j ∥ z n , j ∥ H 1 ( Ω ) 2 d t ≤ C σ . \sum_j\int_{I_j}\|v_n-z_{n,j}\|_{L^{2}(\Omega)}^{2}\,\mathrm{d}t
\le C(n^{-1}+n^{-1/2}+\sigma)
\quad\text{and}\quad
\sum_j\int_{I_j}\|z_{n,j}\|_{H^{1}(\Omega)}^{2}\,\mathrm{d}t
\le C_{\sigma}. j ∑ ∫ I j ∥ v n − z n , j ∥ L 2 ( Ω ) 2 d t ≤ C ( n − 1 + n − 1/2 + σ ) and j ∑ ∫ I j ∥ z n , j ∥ H 1 ( Ω ) 2 d t ≤ C σ .
Given η > 0 \eta>0 η > 0 , we first choose σ \sigma σ with C σ ≤ η / 2 C_{\sigma}\le\eta/2 C σ ≤ η /2 and then enlarge n σ n_{\sigma} n σ such that C ( n − 1 + n − 1 / 2 ) ≤ η / 2 C(n^{-1}+n^{-1/2})\le\eta/2 C ( n − 1 + n − 1/2 ) ≤ η /2 for n ≥ n σ n\ge n_{\sigma} n ≥ n σ . This gives (A.13). Finally, for e ∈ K ( S j ) ∩ C c ∞ ( Ω ; R 3 ) e\in K(S_j)\cap C_{c}^{\infty}(\Omega;\mathbb{R}^{3}) e ∈ K ( S j ) ∩ C c ∞ ( Ω ; R 3 ) and ϑ ∈ C c ∞ ( I j ) \vartheta\in C_{c}^{\infty}(I_j) ϑ ∈ C c ∞ ( I j ) , the assumed temporal estimate applied to ζ ( t , x ) = ϑ ( t ) e ( x ) \zeta(t,x)=\vartheta(t)e(x) ζ ( t , x ) = ϑ ( t ) e ( x ) gives
∣ ∫ I j ( v n ( t ) , e ) L 2 ( Ω ) ϑ ′ ( t ) d t ∣ ≤ C σ , j ( 1 + T 1 / 2 ) ∥ e ∥ H 3 ( Ω ) ∥ ϑ ∥ L ∞ ( I j ) , \left|\int_{I_j}(v_n(t),e)_{L^{2}(\Omega)}\vartheta'(t)\,\mathrm{d}t\right|
\le C_{\sigma,j}(1+T^{1/2})\|e\|_{H^{3}(\Omega)}\|\vartheta\|_{L^{\infty}(I_j)}, ∫ I j ( v n ( t ) , e ) L 2 ( Ω ) ϑ ′ ( t ) d t ≤ C σ , j ( 1 + T 1/2 ) ∥ e ∥ H 3 ( Ω ) ∥ ϑ ∥ L ∞ ( I j ) ,
which is (A.14). Hence Lemma A.6 yields v n → v v_n\to v v n → v strongly in L 2 ( 0 , T ; L 2 ( Ω ) ) L^{2}(0,T;L^{2}(\Omega)) L 2 ( 0 , T ; L 2 ( Ω )) .
Proof of Lemma 3.12. The argument follows the compactness strategy of [45 Theorem 3.2] and [16 ] . To estimate the temporal functional on I j I_j I j , we test (42) with functions compactly supported in time in I j I_j I j , which gives
∫ I j ∫ Ω u n ⋅ ∂ t ζ d x d t = ∫ I j ∫ Ω [ − ( u n ⊗ u n ) : ∇ ζ + ( 2 + n φ n ) D ( u n ) : D ( ζ ) ] d x d t + δ ∫ I j a u ( u n , ζ ) d t + ∫ I j ∫ Ω ( ζ ⋅ ∇ ) d n ⋅ Δ d n d x d t . (170) \int_{I_j}\int_{\Omega}u^n\cdot\partial_t\zeta\,\mathrm{d}x\,\mathrm{d}t
=
\int_{I_j}\int_{\Omega}\left[-(u^n\otimes u^n):\nabla\zeta+(2+n\varphi^n)D(u^n):D(\zeta)\right]\mathrm{d}x\,\mathrm{d}t
+\delta\int_{I_j}a_u(u^n,\zeta)\,\mathrm{d}t
+\int_{I_j}\int_{\Omega}(\zeta\cdot\nabla)d^n\cdot\Delta d^n\,\mathrm{d}x\,\mathrm{d}t. \tag*{(170)} ∫ I j ∫ Ω u n ⋅ ∂ t ζ d x d t = ∫ I j ∫ Ω [ − ( u n ⊗ u n ) : ∇ ζ + ( 2 + n φ n ) D ( u n ) : D ( ζ ) ] d x d t + δ ∫ I j a u ( u n , ζ ) d t + ∫ I j ∫ Ω ( ζ ⋅ ∇ ) d n ⋅ Δ d n d x d t . (170)
The director-dependent term will be treated in stress form, so that only the energy-level H 1 H^1 H 1 -bound is used. We estimate the terms on the right-hand side of (A.23) for
ζ ∈ C c 1 ( I j × Ω ; R 3 ) ∩ L 2 ( I j ; K ( A j ) ∩ H 3 ( Ω ; R 3 ) ) . \zeta\in C_{c}^{1}(I_j\times\Omega;\mathbb{R}^{3})\cap L^{2}(I_j;K(A_j)\cap H^{3}(\Omega;\mathbb{R}^{3})). ζ ∈ C c 1 ( I j × Ω ; R 3 ) ∩ L 2 ( I j ; K ( A j ) ∩ H 3 ( Ω ; R 3 )) .
Then φ n D ( ζ ) = 0 \varphi^nD(\zeta)=0 φ n D ( ζ ) = 0 for all sufficiently large n n n , because S ( φ n ( t ) ) ⋐ A j S(\varphi^n(t))\Subset A_j S ( φ n ( t )) ⋐ A j , so the penalized viscosity term drops out. A global H 2 H^2 H 2 -bound for d n d^n d n is not available uniformly in n n n . The elastic term is therefore rewritten in stress form. From div ζ = 0 \operatorname{div}\zeta=0 div ζ = 0 and ζ = 0 \zeta=0 ζ = 0 on ∂ Ω \partial\Omega ∂ Ω , it follows that
− ∫ Ω ( ζ ⋅ ∇ ) d n ⋅ ( Δ d n + 1 ε 2 ( 1 − ∣ d n ∣ 2 ) d n ) d x = ∫ Ω ( ∇ d n ) T ∇ d n : ∇ ζ d x . -\int_{\Omega}(\zeta\cdot\nabla)d^n\cdot\left(\Delta d^n+\frac{1}{\varepsilon^{2}}(1-|d^n|^{2})d^n\right)\,\mathrm{d}x
=
\int_{\Omega}(\nabla d^n)^{\mathsf T}\nabla d^n:\nabla\zeta\,\mathrm{d}x. − ∫ Ω ( ζ ⋅ ∇ ) d n ⋅ ( Δ d n + ε 2 1 ( 1 − ∣ d n ∣ 2 ) d n ) d x = ∫ Ω ( ∇ d n ) T ∇ d n : ∇ ζ d x .
The contribution from the Ginzburg–Landau regularization is a spatial gradient and therefore vanishes. The energy bound allows us to infer that
∣ ∫ I j ∫ Ω ( ∇ d n ) T ∇ d n : ∇ ζ d x d t ∣ ≤ C ∥ ζ ∥ L 2 ( I j ; W 1 , ∞ ( Ω ) ) ≤ C ∥ ζ ∥ L 2 ( I j ; H 3 ( Ω ) ) . \left|\int_{I_j}\int_{\Omega}(\nabla d^n)^{\mathsf T}\nabla d^n:\nabla\zeta\,\mathrm{d}x\,\mathrm{d}t\right|
\le C\|\zeta\|_{L^{2}(I_j;W^{1,\infty}(\Omega))}
\le C\|\zeta\|_{L^{2}(I_j;H^{3}(\Omega))}. ∫ I j ∫ Ω ( ∇ d n ) T ∇ d n : ∇ ζ d x d t ≤ C ∥ ζ ∥ L 2 ( I j ; W 1 , ∞ ( Ω )) ≤ C ∥ ζ ∥ L 2 ( I j ; H 3 ( Ω )) .
The convective term is bounded in the same norm by ∥ u n ∥ L t ∞ L x 2 2 \|u^n\|_{L^\infty_tL^2_x}^2 ∥ u n ∥ L t ∞ L x 2 2 , the baseline viscous term by ∥ ∇ u n ∥ L 2 ( I j × Ω ) \|\nabla u^n\|_{L^2(I_j\times\Omega)} ∥∇ u n ∥ L 2 ( I j × Ω ) , and the penalization term vanishes because φ n D ( ζ ) = 0 \varphi^n\mathbb{D}(\zeta)=0 φ n D ( ζ ) = 0 . Finally,
∣ δ ∫ I j a u ( u n , ζ ) d t ∣ ≤ δ ∥ δ a n ∥ L 2 ( I j ; X u ) ∥ ζ ∥ L 2 ( I j ; X u ) . \left|\delta\int_{I_j}a_u(u^n,\zeta)\,\mathrm{d}t\right|\leq\sqrt{\delta}\|\sqrt{\delta}a^n\|_{L^2(I_j;X_u)}\|\zeta\|_{L^2(I_j;X_u)}. δ ∫ I j a u ( u n , ζ ) d t ≤ δ ∥ δ a n ∥ L 2 ( I j ; X u ) ∥ ζ ∥ L 2 ( I j ; X u ) .
Thus, for 0 < δ ≤ 1 0<\delta\leq1 0 < δ ≤ 1 ,
∣ ∫ I j ∫ Ω u n ∂ t ζ d x d t ∣ ≤ C ( u 0 , d 0 , ε , T ) ∥ ζ ∥ L 2 ( I j ; H 3 ( Ω ) ) , \left|\int_{I_j}\int_{\Omega}u^n\partial_t\zeta\,\mathrm{d}x\,\mathrm{d}t\right|\leq C(u_0,d_0,\varepsilon,T)\|\zeta\|_{L^2(I_j;H^3(\Omega))}, ∫ I j ∫ Ω u n ∂ t ζ d x d t ≤ C ( u 0 , d 0 , ε , T ) ∥ ζ ∥ L 2 ( I j ; H 3 ( Ω )) ,
with a constant independent of n n n (and bounded for 0 < δ ≤ 1 0<\delta\leq1 0 < δ ≤ 1 ). The nested localization construction (A.12), with the fixed comparison set S j : = A j S_j:=A_j S j := A j on each I j I_j I j , gives the fixed comparison sets required in Lemma 3.11 . Moreover, the artificial-viscosity estimate gives u n u^n u n uniformly bounded in L 2 ( 0 , T ; X u ) L^2(0,T;X_u) L 2 ( 0 , T ; X u ) at fixed δ \delta δ . Since X u ↪ W 1 , ∞ ( Ω ) X_u\hookrightarrow W^{1,\infty}(\Omega) X u ↪ W 1 , ∞ ( Ω ) ,
sup n ∫ 0 T ∥ ∇ u n ( t ) ∥ L ∞ ( Ω ) d t < ∞ . \sup_n\int_0^T\|\nabla u^n(t)\|_{L^\infty(\Omega)}\,\mathrm{d}t<\infty. n sup ∫ 0 T ∥∇ u n ( t ) ∥ L ∞ ( Ω ) d t < ∞.
Applying Grönwall’s inequality for the forward and inverse flow equations we get uniform bi-Lipschitz bounds for Φ n ( t ) : = X 0 , t n \Phi_n(t):=X^n_{0,t} Φ n ( t ) := X 0 , t n and Φ n ( t ) − 1 = Y n ( t ) \Phi_n(t)^{-1}=Y^n(t) Φ n ( t ) − 1 = Y n ( t ) . The maps are volume preserving because div u n = 0 \operatorname{div}u^n=0 div u n = 0 . In addition, X u ↪ H 3 ( Ω ) ↪ W 2 , 6 ( Ω ) X_u\hookrightarrow H^3(\Omega)\hookrightarrow W^{2,6}(\Omega) X u ↪ H 3 ( Ω ) ↪ W 2 , 6 ( Ω ) , so the differentiated-flow estimate established at the beginning of § 3.2 gives
sup n , t ( ∥ Φ n ( t ) ∥ W 2 , 6 ( Ω ) + ∥ Φ n ( t ) − 1 ∥ W 2 , 6 ( Ω ) ) < ∞ . \sup_{n,t}\left(\|\Phi_n(t)\|_{W^{2,6}(\Omega)}+\|\Phi_n(t)^{-1}\|_{W^{2,6}(\Omega)}\right)<\infty. n , t sup ( ∥ Φ n ( t ) ∥ W 2 , 6 ( Ω ) + ∥ Φ n ( t ) − 1 ∥ W 2 , 6 ( Ω ) ) < ∞.
Thus, the transported sets S ( φ n ( t ) ) = Φ n ( t , S 0 ) S(\varphi^n(t))=\Phi_n(t,S_0) S ( φ n ( t )) = Φ n ( t , S 0 ) satisfy the full geometric hypothesis (84) . The sets A j A_j A j can be chosen as small tubular neighborhoods of a rigid copy of S 0 S_0 S 0 . From the fixed positive distance of the limiting body from the wall, it follows that the comparison-set collar hypothesis of Lemma 3.11 is also satisfied. The energy inequality also yields the uniform bound
n ∫ 0 T ∫ Ω φ n ∣ D ( u n ) ∣ 2 d x d t ≤ C , n\int_0^T\int_{\Omega}\varphi^n|\mathbb{D}(u^n)|^2\,\mathrm{d}x\,\mathrm{d}t\leq C, n ∫ 0 T ∫ Ω φ n ∣ D ( u n ) ∣ 2 d x d t ≤ C ,
so all hypotheses of Lemma 3.11 are satisfied. An application of Lemma 3.11 then implies that u n → u u^n\to u u n → u strongly in L 2 ( 0 , T ; L 2 ( Ω ) ) L^2(0,T;L^2(\Omega)) L 2 ( 0 , T ; L 2 ( Ω )) , as claimed. □
Proof of Lemma 3.13 . Because the limiting solid stays a positive distance from ∂ Ω \partial\Omega ∂ Ω on the interval under consideration and moves by a rigid motion, there exists a family of bi-Lipschitz diffeomorphisms Ψ ( t ) : Ω → Ω \Psi(t):\Omega\to\Omega Ψ ( t ) : Ω → Ω , with Ψ ( t ) ( S 0 ) = S ( t ) \Psi(t)(S_0)=S(t) Ψ ( t ) ( S 0 ) = S ( t ) . This map equals the rigid map h ( t ) + O ( t ) ⋅ h(t)+O(t)\cdot h ( t ) + O ( t ) ⋅ in a fixed neighborhood of S 0 S_0 S 0 , equals the identity in a fixed neighborhood of ∂ Ω \partial\Omega ∂ Ω , and is uniformly bounded in W 1 , ∞ ( 0 , T ; W 1 , ∞ ( Ω ) ) W^{1,\infty}(0,T;W^{1,\infty}(\Omega)) W 1 , ∞ ( 0 , T ; W 1 , ∞ ( Ω )) together with its inverse. For the construction of such a map, we extend the instantaneous rigid velocity smoothly from a fixed collar of the moving solid to an Eulerian vector field that vanishes in a fixed collar of ∂ Ω \partial\Omega ∂ Ω , with a uniform spatial Lipschitz bound. As the separation is positive, this extension can be chosen uniformly in time. The flow of this ALE field equals the prescribed rigid motion near the solid and the identity near the container wall. Since the rigid translational and angular velocities are essentially bounded in time, the flow and its inverse have the stated W t 1 , ∞ W x 1 , ∞ W^{1,\infty}_tW^{1,\infty}_x W t 1 , ∞ W x 1 , ∞ -bounds.
For ψ ∈ X d ( φ ) \psi\in\mathcal{X}_d(\varphi) ψ ∈ X d ( φ ) , set ψ ^ ( t , y ) = ψ ( t , Ψ ( t , y ) ) \widehat{\psi}(t,y)=\psi(t,\Psi(t,y)) ψ ( t , y ) = ψ ( t , Ψ ( t , y )) on F 0 \mathcal{F}_0 F 0 . By the Sobolev chain rule, the following identities hold a.e. ∇ y ψ ^ = ( ∇ x ψ ) ∘ Ψ ∇ y Ψ \nabla_y\widehat{\psi}=(\nabla_x\psi)\circ\Psi\,\nabla_y\Psi ∇ y ψ = ( ∇ x ψ ) ∘ Ψ ∇ y Ψ and ∂ t ψ ^ = ( ∂ t ψ ) ∘ Ψ + ( ∇ x ψ ) ∘ Ψ ∂ t Ψ \partial_t\widehat{\psi}=(\partial_t\psi)\circ\Psi+(\nabla_x\psi)\circ\Psi\,\partial_t\Psi ∂ t ψ = ( ∂ t ψ ) ∘ Ψ + ( ∇ x ψ ) ∘ Ψ ∂ t Ψ . The corresponding inverse formulas are obtained by composing with Ψ − 1 \Psi^{-1} Ψ − 1 . Uniform bounds for Ψ , Ψ − 1 \Psi,\Psi^{-1} Ψ , Ψ − 1 and their first derivatives, together with uniform upper and lower Jacobian bounds, show that pullback is an isomorphism, with equivalent norms, between X d ( φ ) \mathcal{X}_d(\varphi) X d ( φ ) and H 1 ( 0 , T ; L 2 ( F 0 ) ) ∩ L 2 ( 0 , T ; H 0 1 ( F 0 ) ) H^1(0,T;L^2(\mathcal{F}_0))\cap L^2(0,T;H^1_0(\mathcal{F}_0)) H 1 ( 0 , T ; L 2 ( F 0 )) ∩ L 2 ( 0 , T ; H 0 1 ( F 0 )) with ψ ^ ( T ) = 0 \widehat{\psi}(T)=0 ψ ( T ) = 0 . By construction, Ψ ( t , ⋅ ) \Psi(t,\cdot) Ψ ( t , ⋅ ) maps the boundary components of F 0 \mathcal{F}_0 F 0 onto those of F ( t ) \mathcal{F}(t) F ( t ) , this composition also preserves the variational zero-trace condition defining H 0 1 H^1_0 H 0 1 . Next, we observe that C ∞ ( [ 0 , T ) × F 0 ) C^\infty([0,T)\times F_0) C ∞ ([ 0 , T ) × F 0 ) is dense in this fixed cylinder space. Choose ψ ^ m \widehat{\psi}_m ψ m converging to ψ ^ \widehat{\psi} ψ in the indicated norm. Pushing forward gives functions supported a positive spatial distance from the moving boundary and vanishing on a terminal time interval. Note that these pushforwards only have Sobolev regularity in time because the map Ψ \Psi Ψ is only Lipschitz in time. To smooth them without losing the value at t = 0 t=0 t = 0 , we start by using the standard bounded extension operator for H 1 ( 0 , T ; L 2 ( Ω ) ) ∩ L 2 ( 0 , T ; H 1 ( Ω ) ) H^1(0,T;L^2(\Omega))\cap L^2(0,T;H^1(\Omega)) H 1 ( 0 , T ; L 2 ( Ω )) ∩ L 2 ( 0 , T ; H 1 ( Ω )) across t = 0 t=0 t = 0 and extend by zero across the terminal interval. Because the support at t = 0 t=0 t = 0 is compactly contained in F 0 F_0 F 0 and the rigid trajectory is continuous, the extension can be chosen on a short negative-time interval with the same positive spatial clearance. A spacetime mollification with radius smaller than that clearance then produces ψ m ∈ D F \psi_m\in\mathcal{D}_F ψ m ∈ D F without changing the limit. The continuity of the trace map H 1 ( 0 , T ; L 2 ( Ω ) ) ↪ C ( [ 0 , T ] ; L 2 ( Ω ) ) H^1(0,T;L^2(\Omega))\hookrightarrow C([0,T];L^2(\Omega)) H 1 ( 0 , T ; L 2 ( Ω )) ↪ C ([ 0 , T ] ; L 2 ( Ω )) implies that ψ m ( 0 ) → ψ ( 0 ) \psi_m(0)\to\psi(0) ψ m ( 0 ) → ψ ( 0 ) in L 2 ( Ω ) L^2(\Omega) L 2 ( Ω ) .
Proof of Lemma 3.14 . The task is essentially to verify the hypotheses of Lemma A.6 . For this purpose, we fix γ ∈ ( 0 , T / 2 ) \gamma\in(0,T/2) γ ∈ ( 0 , T /2 ) and σ > 0 \sigma>0 σ > 0 smaller than one quarter of the uniform distance to ∂ Ω \partial\Omega ∂ Ω . The limiting rigid trajectory is uniformly continuous, and the rigid motions converge uniformly, so there exist a partition γ = s 0 < s 1 < ⋯ < s N = T − γ \gamma=s_0<s_1<\cdots<s_N=T-\gamma γ = s 0 < s 1 < ⋯ < s N = T − γ and fixed smooth connected sets A ℓ ⋐ Ω A_\ell\Subset\Omega A ℓ ⋐ Ω such that, with I ℓ : = ( s ℓ − 1 , s ℓ ) I_\ell:=(s_{\ell-1},s_\ell) I ℓ := ( s ℓ − 1 , s ℓ ) , for all large j j j and all t ∈ I ℓ t\in I_\ell t ∈ I ℓ , we have S j ( t ) ⋐ A ℓ S_j(t)\Subset A_\ell S j ( t ) ⋐ A ℓ , S ( t ) 2 σ ⊂ A ℓ ⊂ S ( t ) 3 σ S(t)_{2\sigma}\subset A_\ell\subset S(t)_{3\sigma} S ( t ) 2 σ ⊂ A ℓ ⊂ S ( t ) 3 σ as well as A ℓ ⊂ ( S j ( t ) ) C σ A_\ell\subset(S_j(t))_{C\sigma} A ℓ ⊂ ( S j ( t ) ) C σ . The sets can be chosen as tubular enlargements of the limiting solid at a time in I ℓ I_\ell I ℓ . After refining the partition, ∂ A ℓ \partial A_\ell ∂ A ℓ and ∂ S j ( t ) \partial S_j(t) ∂ S j ( t ) are uniformly normal graphs over one another, with graph height between c σ c\sigma c σ and C σ C\sigma C σ . All constants below are independent of j j j , t t t , and small σ \sigma σ .
For a.e. t ∈ I ℓ t\in I_\ell t ∈ I ℓ , let R j ( t ) R_j(t) R j ( t ) be the unique affine rigid velocity agreeing with v j ( t ) v_j(t) v j ( t ) on S j ( t ) S_j(t) S j ( t ) and put q j : = v j − R j q_j:=v_j-R_j q j := v j − R j . Thus q j = 0 q_j=0 q j = 0 in S j ( t ) S_j(t) S j ( t ) . On the other hand, the rigid coefficients of R j R_j R j are bounded by ∥ v j ∥ L 2 ( S j ( t ) ) \|v_j\|_{L^2(S_j(t))} ∥ v j ∥ L 2 ( S j ( t )) . This implies that
∥ q j ∥ H 1 ( Ω ) ≤ C ∥ v j ∥ H 1 ( Ω ) . (171) \|q_j\|_{H^1(\Omega)}\le C\|v_j\|_{H^1(\Omega)}. \tag*{(171)} ∥ q j ∥ H 1 ( Ω ) ≤ C ∥ v j ∥ H 1 ( Ω ) . (171)
Uniform tubular coordinates for the rigid copies of S 0 S_0 S 0 , followed by the one-dimensional fundamental theorem of calculus along the normal fibers, give the thin-layer Poincaré and trace estimates
∥ q j ∥ L 2 ( ( S j ( t ) ) C σ ∖ S j ( t ) ) ≤ C σ ∥ ∇ q j ∥ L 2 ( ( S j ( t ) ) C σ ) , ∥ q j ∥ L 2 ( ∂ A ℓ ) ≤ C σ 1 / 2 ∥ ∇ q j ∥ L 2 ( ( S j ( t ) ) C σ ) . (172) \|q_j\|_{L^2((S_j(t))_{C\sigma}\setminus S_j(t))}\le C\sigma\|\nabla q_j\|_{L^2((S_j(t))_{C\sigma})},\qquad\|q_j\|_{L^2(\partial A_\ell)}\le C\sigma^{1/2}\|\nabla q_j\|_{L^2((S_j(t))_{C\sigma})}. \tag*{(172)} ∥ q j ∥ L 2 (( S j ( t ) ) C σ ∖ S j ( t )) ≤ C σ ∥∇ q j ∥ L 2 (( S j ( t ) ) C σ ) , ∥ q j ∥ L 2 ( ∂ A ℓ ) ≤ C σ 1/2 ∥∇ q j ∥ L 2 (( S j ( t ) ) C σ ) . (172)
These estimates extend from smooth fields to H 1 H^1 H 1 by density. Choose a cutoff η ℓ , σ \eta_{\ell,\sigma} η ℓ , σ equal to zero on A ℓ A_\ell A ℓ , equal to one outside ( A ℓ ) σ (A_\ell)_\sigma ( A ℓ ) σ , and satisfying ∣ ∇ η ℓ , σ ∣ ≤ C σ − 1 |\nabla\eta_{\ell,\sigma}|\le C\sigma^{-1} ∣∇ η ℓ , σ ∣ ≤ C σ − 1 . Set z j , ℓ 0 : = R j + η ℓ , σ q j z^0_{j,\ell}:=R_j+\eta_{\ell,\sigma}q_j z j , ℓ 0 := R j + η ℓ , σ q j . Take a fixed-width smooth collar D ℓ D_\ell D ℓ of A ℓ A_\ell A ℓ , independent of the transition width, and let B ℓ \mathcal{B}_\ell B ℓ be its Bogovskiĭ right inverse. The flux of div z j , ℓ 0 \operatorname{div}z^0_{j,\ell} div z j , ℓ 0 is zero. As in (A.19) in the proof of Lemma 3.11 , the negative-order estimate
∥ B ℓ div z j , ℓ 0 ∥ L 2 ( D ℓ ) ≤ C ∣ div z j , ℓ 0 ∣ ∗ , D ℓ (173) \|\mathcal{B}_\ell\operatorname{div}z^0_{j,\ell}\|_{L^2(D_\ell)}\le C|\operatorname{div}z^0_{j,\ell}|_{*,D_\ell} \tag*{(173)} ∥ B ℓ div z j , ℓ 0 ∥ L 2 ( D ℓ ) ≤ C ∣ div z j , ℓ 0 ∣ ∗ , D ℓ (173)
and integration by parts give
∣ div z j , ℓ 0 ∣ ∗ , D ℓ ≤ C ( ∥ q j ∥ L 2 ( ( A ℓ ) σ ∖ A ℓ ) + ∥ q j ∥ L 2 ( ∂ A ℓ ) ) ≤ C σ 1 / 2 ∥ q j ∥ H 1 ( Ω ) . (174) |\operatorname{div}z^0_{j,\ell}|_{*,D_\ell}\le C\left(\|q_j\|_{L^2((A_\ell)_\sigma\setminus A_\ell)}+\|q_j\|_{L^2(\partial A_\ell)}\right)\le C\sigma^{1/2}\|q_j\|_{H^1(\Omega)}. \tag*{(174)} ∣ div z j , ℓ 0 ∣ ∗ , D ℓ ≤ C ( ∥ q j ∥ L 2 (( A ℓ ) σ ∖ A ℓ ) + ∥ q j ∥ L 2 ( ∂ A ℓ ) ) ≤ C σ 1/2 ∥ q j ∥ H 1 ( Ω ) . (174)
The usual L 0 2 → H 0 1 L^2_0\to H^1_0 L 0 2 → H 0 1 Bogovskiĭ estimate gives an H 1 H^1 H 1 -bound for the correction for each fixed σ \sigma σ . Extending the correction by zero and setting z j , ℓ : = z j , ℓ 0 − B ℓ div z j , ℓ 0 z_{j,\ell}:=z^0_{j,\ell}-\mathcal{B}_\ell\operatorname{div}z^0_{j,\ell} z j , ℓ := z j , ℓ 0 − B ℓ div z j , ℓ 0 , we obtain
z j , ℓ ( t ) ∈ K ( A ℓ ) : = { z ∈ V ( Ω ) : D ( z ) = 0 on A ℓ } . z_{j,\ell}(t)\in\mathcal{K}(A_\ell):=\{z\in\mathcal{V}(\Omega):\mathbb{D}(z)=0\text{ on }A_\ell\}. z j , ℓ ( t ) ∈ K ( A ℓ ) := { z ∈ V ( Ω ) : D ( z ) = 0 on A ℓ } .
Thus, invoking that v j − z j , ℓ = q j − η ℓ , σ q j + B ℓ div z j , ℓ 0 v_j-z_{j,\ell}=q_j-\eta_{\ell,\sigma}q_j+\mathcal{B}_\ell\operatorname{div}z^0_{j,\ell} v j − z j , ℓ = q j − η ℓ , σ q j + B ℓ div z j , ℓ 0 , and using the estimates (A.24)–(A.27), for a.e. t ∈ I ℓ t\in I_\ell t ∈ I ℓ , we find that
∥ v j ( t ) − z j , ℓ ( t ) ∥ L 2 ( Ω ) ≤ C σ 1 / 2 ∥ v j ( t ) ∥ H 1 ( Ω ) . \|v_j(t)-z_{j,\ell}(t)\|_{L^2(\Omega)}\le C\sigma^{1/2}\|v_j(t)\|_{H^1(\Omega)}. ∥ v j ( t ) − z j , ℓ ( t ) ∥ L 2 ( Ω ) ≤ C σ 1/2 ∥ v j ( t ) ∥ H 1 ( Ω ) .
Squaring, integrating over I ℓ I_\ell I ℓ , and summing over ℓ \ell ℓ , we obtain the estimate
∑ ℓ ∫ I ℓ ∥ v j − z j , ℓ ∥ L 2 ( Ω ) 2 d t ≤ C σ sup i ∥ v i ∥ L 2 ( 0 , T ; H 1 ( Ω ) ) 2 . \sum_\ell\int_{I_\ell}\|v_j-z_{j,\ell}\|_{L^2(\Omega)}^2\,dt\le C\sigma\sup_i\|v_i\|_{L^2(0,T;H^1(\Omega))}^2. ℓ ∑ ∫ I ℓ ∥ v j − z j , ℓ ∥ L 2 ( Ω ) 2 d t ≤ C σ i sup ∥ v i ∥ L 2 ( 0 , T ; H 1 ( Ω )) 2 .
For fixed σ \sigma σ , the L 0 2 → H 0 1 L^2_0 \to H^1_0 L 0 2 → H 0 1 Bogovskiĭ estimate also gives ∑ ℓ ∫ I ℓ ∥ z j , ℓ ∥ H 1 ( Ω ) 2 d t ≤ C σ \sum_\ell\int_{I_\ell} \|z_{j,\ell}\|_{H^1(\Omega)}^2 \,\mathrm{d}t \le C_\sigma ∑ ℓ ∫ I ℓ ∥ z j , ℓ ∥ H 1 ( Ω ) 2 d t ≤ C σ , uniformly in j j j . Thus, (A.13) holds with η = C σ \eta=C_\sigma η = C σ . We also argue that every e ∈ K ( A ℓ ) ∩ C ∞ ( Ω ; R 3 ) e \in K(A_\ell) \cap C^\infty(\Omega;\mathbb{R}^3) e ∈ K ( A ℓ ) ∩ C ∞ ( Ω ; R 3 ) is rigid on A ℓ ⊃ S ( t ) 2 σ A_\ell\supset S(t)_{2\sigma} A ℓ ⊃ S ( t ) 2 σ for t ∈ I ℓ t \in I_\ell t ∈ I ℓ , so that ζ ( t , x ) = ϑ ( t ) e ( x ) \zeta(t,x)=\vartheta(t)e(x) ζ ( t , x ) = ϑ ( t ) e ( x ) with ϑ ∈ C c ∞ ( I ℓ ) \vartheta\in C^\infty_c(I_\ell) ϑ ∈ C c ∞ ( I ℓ ) is admissible in (105) with 2 σ 2\sigma 2 σ in place of σ \sigma σ , showing (A.14). As σ \sigma σ can be chosen arbitrarily small, Lemma A.6 yields the claim. □ \square □
Department of Mathematics, University of Southern California, Los Angeles, CA 90089, USA. Email address: fbrandt@usc.edu
Department of Mathematics, University of California, Berkeley, CA 94720, USA. Email address: vnguyen26@berkeley.edu