Stochastic homogenisation of free-discontinuity problems (Q2418422)

The authors prove a stochastic homogenization result for free-discontinuity functions of the type $E_{\varepsilon }(\omega )(u,A)=\int_{A}f(\omega , \frac{x}{\varepsilon },\nabla u)dx+\int_{S_{u}\cap A}g(\omega ,\frac{x}{ \varepsilon },u^{+}-u^{-},\nu _{u})d\mathcal{H}^{n-1}$, if $u\in GSBV(A, \mathbb{R}^{m})$ and $E_{\varepsilon }(\omega )(u,A)=+\infty $ otherwise in $ L^{0}(\mathbb{R}^{n},\mathbb{R}^{m})$, where $A$ is an open subset of $ \mathbb{R}^{n}$, $S_{u}$ is the discontinuity set of $u$, $u^{+}$ and $u^{-}$ are the traces of $u$ on both sides of $S_{u}$, $\nu _{u}$ is the unit normal to $S_{u}$, $\nabla u$ is an approximate differential of $u$ and $ \mathcal{H}^{n-1}$ is Hausdorff measure in $\mathbb{R}^{n}$. Here $\omega $ is a random parameter in the probability space $(\Omega ,\mathcal{T},P)$. For the main convergence result, the authors assume that $f$ and $g$ are stationary random volume and surface integrands with respect to a group $ (\tau _{z})_{z\in \mathbb{R}^{n}}$ of $P$-preserving transformations in $ (\Omega ,\mathcal{T},P)$ and which satisfy different measurability, continuity and growth properties. The main result of the paper proves that $ E_{\varepsilon }(\omega )(\cdot ,A)$ $\Gamma $-converges to $E_{\hom }(\omega )(\cdot ,A)$ in $L^{0}(\mathbb{R}^{n},\mathbb{R}^{m})$ and that $ E_{\varepsilon }^{p}(\omega )(\cdot ,A)$ $\Gamma $-converges to $E_{\hom }^{p}(\omega )(\cdot ,A)$ in $L_{\mathrm{loc}}^{p}(\mathbb{R}^{n},\mathbb{R}^{m})$ for every $\omega \in \Omega ^{\prime }$ with $\Omega ^{\prime }\in \mathcal{ T}$ and $P(\Omega ^{\prime })=1$ and $A\in \mathcal{A}$ the collection of bounded and open subsets of $\mathbb{R}^{n}$. The superscript $p$ means the restriction of the function to $L_{\mathrm{loc}}^{p}(\mathbb{R}^{n},\mathbb{R} ^{m})\times \mathcal{A}$. Here $E_{\hom }(\omega )(u,A)=\int_{A}f_{\hom }(\omega ,\nabla u)dx+\int_{S_{u}\cap A}g_{\hom }(\omega ,[u],\nu _{u})d \mathcal{H}^{n-1}$, if $u\mid _{A}\in GSBV^{p}(A,\mathbb{R}^{n})$ and $ E_{\varepsilon }(\omega )(u,A)=+\infty $ otherwise in $L^{0}(\mathbb{R}^{n}, \mathbb{R}^{m})$ and $f_{\hom }$ and $g_{\hom }$ are given explicit expressions which involve infimum over appropriate cubes. The authors first recall the notion of stationary random integral and the subadditive ergodic theorem by \textit{M. A. Akcoglu} and \textit{U. Krengel} [J. Reine Angew. Math. 323, 53--67 (1981; Zbl 0453.60039)]. The main part of the proof of the $\Gamma $-convergence result is devoted to the construction of the expressions of $f_{\hom }$ and $g_{\hom }$. In the case of the volume integrand, the derivation of the expression of $f_{\hom }$ is rapidly obtained. The expression of $g_{\hom }$ is first proved in the case where $ x=0$, the authors using the subadditive ergodic theorem and building appropriate groups of $P$-preserving transformations on the completion $ (\Omega ,\hat{\mathcal{T}},\hat{P})$ of $(\Omega ,\mathcal{T},P)$, which leads to a subadditive process. Then the authors establish the expression for a general $x\in \mathbb{R}^{n}$. They here prove an invariance property for the homogenized functional associated to a stationary random surface integrand with respect to a group of $P$-preserving transformations on $(\Omega ,\mathcal{T},P)$ and they use Birkhoff's ergodic theorem and a dominated convergence theorem.