Large-scale regularity of nearly incompressible elasticity in stochastic homogenization (Q2138648)

The authors consider a bounded and Lipschitz domain \(D\subset \mathbb{R}^{d}\) occupied by a non-homogeneous, anisotropic and nearly incompressible material, thus leading to the system \(\nabla \cdot (A(x)\nabla u)+\nabla (\lambda (x)\nabla \cdot u)=F\) in \(D\), where \(A\) is a tensor-valued function which satisfies the usual ellipticity, continuity and symmetry properties, and \( \lambda \) is a scalar function which satisfies the compressibility condition \(\lambda _{0}\leq \lambda (x)\leq \lambda _{0}+\Lambda \), with \(\lambda _{0}\geq 0\).\ Dirichlet boundary conditions are imposed \(u=f\) on \(\partial D\) . Introducing \(\lambda (x)=\lambda _{0}+b(x)\), this system may be written as \(\nabla \cdot (A(x)\nabla u_{\lambda })+\lambda \nabla (\nabla \cdot u_{\lambda })=F\) in \(D\), \(u_{\lambda }=f\) on \(\partial D\). The authors will consider the problem in a homogenization context \(\nabla \cdot (A^{\varepsilon }(x)\nabla u_{\lambda }^{\varepsilon })+\nabla (\lambda ^{\varepsilon }\nabla \cdot u_{\lambda }^{\varepsilon })=0\) in \(D\), with \( A^{\varepsilon }(x)=A(x/\varepsilon )\) and \(\lambda ^{\varepsilon }(x)=\lambda (x/\varepsilon )\). They consider the \(\sigma \)-algebra \( \mathcal{F}_{D}\) generated by the random elements \((A,\lambda )\rightarrow (\int_{\mathbb{R}^{d}}a_{ij}^{\alpha \beta }(x)\phi (x),\int_{\mathbb{R} ^{d}}\lambda (x)\psi (x))\) with \(\phi ,\psi \in C_{0}^{\infty }(D)\), and a probability measure \(\mathbb{P}\) which satisfies a stationarity property with respect to \(\mathbb{Z}^{d}\)-translations and a unit range of dependence property, that is \(\mathcal{F}_{D}\) and \(\mathcal{F}_{E}\) are \(\mathbb{P}\)-independent for every Borel subsets \(D,E\subset \mathbb{R}^{d}\) satisfying \( dist(D,E)\geq 1\). The first main result of the paper proves that for any \( s\in (0,d)\) and \(\lambda _{0}\in \lbrack 0,\infty )\), there exist a constant \( C_{0}\), which only depends on \(s,d,\Lambda \) and a random variable \(\mathcal{X }=\mathcal{X}_{s,\lambda }:\Omega \rightarrow \lbrack 1,\infty )\) satisfying \(\mathbb{E}[\exp ((\mathcal{X}/\theta )^{s})]\leq 2\), for some \(s>0\), such that if \(u_{\lambda }^{\varepsilon }\in H^{1}(B_{2}(0);\mathbb{R}^{d})\) is a weak solution to the last preceding problem in \(B_{2}(0)\) then for every \( r\in \lbrack \varepsilon \mathcal{X},1]\), the interior estimate \[ \left( \frac{1}{\left\vert B_{r}\right\vert }\int_{B_{r}}\left\vert \nabla u_{\lambda }^{\varepsilon }\right\vert ^{2}\right) ^{1/2}+\left( \frac{1}{ \left\vert B_{r}\right\vert }\int_{B_{r}}\left\vert \lambda ^{\varepsilon }\nabla \cdot u_{\lambda }^{\varepsilon }-\frac{1}{\left\vert B_{2}\right\vert }\int_{B_{2}}\lambda ^{\varepsilon }\nabla \cdot u_{\lambda }^{\varepsilon }\right\vert ^{2}\right) ^{1/2}\leq C\left( \frac{1}{ \left\vert B_{2}\right\vert }\int_{B_{2}}\left\vert \nabla u_{\lambda }^{\varepsilon }\right\vert ^{2}\right) ^{1/2} \] holds true. The second main result proves the boundary estimate \[ \left( \frac{1}{\left\vert D_{r}\right\vert }\int_{D_{r}}\left\vert \nabla u_{\lambda }^{\varepsilon }\right\vert ^{2}\right) ^{1/2}+\left( \frac{1}{\left\vert D_{r}\right\vert } \int_{D_{r}}\left\vert \lambda ^{\varepsilon }\nabla \cdot u_{\lambda }^{\varepsilon }-\frac{1}{\left\vert D_{2}\right\vert }\int_{D_{2}}\lambda ^{\varepsilon }\nabla \cdot u_{\lambda }^{\varepsilon }\right\vert ^{2}\right) ^{1/2}\leq C\left( \frac{1}{\left\vert D_{2}\right\vert } \int_{D_{2}}\left\vert \nabla u_{\lambda }^{\varepsilon }\right\vert ^{2}\right) ^{1/2}, \] where \(D_{r}=D\cap B_{r}\). Here, the authors assume that the domain \(D\) satisfies a \(\varepsilon \)-scale \(C^{1,\alpha }\) condition at 0. For the proof, the authors first recall the homogenized problem associated with the last preceding problem, and the rate of convergence of the solutions. They recall the existence of a unique weak solution to this problem and a Caccioppoli inequality for this weak solution. They introduce the Stokes problem and recall Meyers' inequality for a weak solution to this Stokes problem and to the problem under consideration. Finally, they prove energy decay estimates using the above ingredients and an iteration process.