Stochastic homogenization of quasilinear Hamilton-Jacobi equations and geometric motions (Q1746196)

This paper deals with time-dependent, quasilinear, viscous Hamilton-Jacobi equations of the form \(\partial_{t}u^{\varepsilon}-\varepsilon\;\text{tr}(A(Du^{\varepsilon}/| Du^{\varepsilon}|,x/\varepsilon)D^2 u^{\varepsilon})+H(Du^{\varepsilon},x/\varepsilon)=0\) in \(\mathbb R^{d}\times(0,\infty)\). The diffusion matrix \(A(e,x)\) is assumed to be nonnegative definite for each \((e,x)\in\partial B_1\times \mathbb R^{d}\); the Hamiltonian \(H(\xi,x)\) is assumed to be positively homogeneous of order \(p\in [1,\infty)\) in \(\xi\), but is not necessarily convex in \(\xi\). Both \(A(e,\cdot)\) and \(H(\xi, \cdot)\) are assumed to be stationary random fields sampled by a probability measure \(P\) which satisfies a finite range of dependence. It is shown that \(u^{\varepsilon}\) converges locally uniformly, as \(\varepsilon\to0\), with \(P\)-probability one, to the solution \(u\) of a deterministic equation of the form \(\partial_{t} u+\bar H(Du)=0\) in \(\mathbb R^{d}\times (0,\infty)\), with an effective Hamiltonian \(\bar H:\;\mathbb R^{d}\to \mathbb R\) which has sublevel sets that are star-shaped with respect to the origin. The quantitative error estimates which give an algebraic rate of homogenization are presented.