Stochastic homogenization of viscous Hamilton-Jacobi equations and applications (Q2340108)

The authors consider the initial-value problem \(u_t^\varepsilon-\varepsilon\mathrm{tr}\left(A\left({x\over \varepsilon}\right)D^2u^\varepsilon\right)+H\left(Du^\varepsilon,{x\over\varepsilon}\right)=0\), in \(\mathbb R^d\times(0,\infty)\), \(u^\varepsilon=g\) on \(\mathbb R^d\times\{0\}\). Here the initial data \(g\) is given bounded and uniformly continuous function; \(A\) and \(H\) are assumed to be stationary-ergodic random fields, \(A:\mathbb R^d\to S^d\) (\(S^d\) denotes the set of \(d\)-by-\(d\) real symmetric matrices), \(H:\mathbb R^d\times\mathbb R^d\to \mathbb R\). It is assumed that there exists a function \(\sigma:\mathbb R^d\to\mathbb R^{n\times d}\) such that \(A={1\over 2}\sigma^t\sigma\), \(|\sigma(y)-\sigma(z)|\leq \Lambda_2| y-z|\); \(p\to H(p,y)\) is convex; for every \(R>0\), there exist constants \(0<a_R\leq 1\) and \(M_R\geq 1\) such that for every \(p\), \(\hat p\in\mathbb R^d\) and \(y,z\in B_R\): \(a_R| p|^q-M_R\leq H(p,y)\leq \Lambda_1(| p|^q+1)\), \(| H(p,y)-H(p,z)|\leq (\Lambda_1| p|^q+M_R)| y-z|\), \(| H(p,y)-H(\hat p,y)|\leq\Lambda_1(| p|+|\hat p|+1)^{q-1}| p-\hat p|\). The authors identify a continuous, convex \(\bar H:\mathbb R^d\to\mathbb R\) and show that, as \(\varepsilon\to 0\), the solution \(u^\varepsilon\) of considered problem converge, \(P\)-almost surely, to the unique solution of \(u\_ t+\bar H(Du)=0\) in \(\mathbb R^d\times(0,\infty)\), \(u=g\) on \(\mathbb R^d\times\{0\}\). Here \(P\) is a probability measure which satisfies stationarity, ergodicity and weak coercivity conditions. As an application, it is derived a general quenched large deviation principle for diffusions in random environments and with absorbing potentials.