Stochastic homogenization of convolution type operators (Q2286492)

The authors consider a homogenization problem for a random convolution type operator of the form \((L_{\omega}u)(x)=\mu(x,\omega)\int_{\mathbb{R}^{d}}a(x-y)\mu(y,\omega)(u(y)-u(x))dy\). It is assumed that the random field \(\mu(x,\omega)=\mu(T_{x}\omega)\) is stationary and bounded, \(T_{x}\) is an ergodic group of measurable transformations \(T_{x}: \Omega\to\Omega\). Let \(u^{\varepsilon}\) be the solution of equation \((L_{\omega}^{\varepsilon}-m)u^{\varepsilon}=f\), where \((L^{\varepsilon}_{\omega}u)(x)=\varepsilon^{-(d+2)}\int_{\mathbb{R}^{d}}a((x-y)/\varepsilon) \mu(x/\varepsilon,\omega)\mu(y/\varepsilon,\omega)(u(y)-u(x))dy\), \(m>0\), \(f\in L^2(\mathbb{R}^{d})\). Let \(u_0(x)\) be the solution of equation \(\sum_{i,j=1}^{d}\Theta_{ij}\frac{\partial^2 u_0}{\partial x_{i}\partial x_{j}}-mu_0=f\), where \(\Theta=\{\Theta_{ij}\}, i,j=\overline{1,d}\) is a positive definite matrix. The authors prove that almost surely for any \(f\in L^2(\mathbb{R}^{d})\) and any \(m>0\) the convergence holds: \(\Vert u^{\varepsilon}-u_0\Vert_{L^2(\mathbb{R}^{d})}\to0\), as \(\varepsilon\to0\).