Kantorovich distance in the martingale CLT and quantitative homogenization of parabolic equations with random coefficients (Q466900): Difference between revisions

This paper provides a useful quantitative version of the martingale central limit theorem in terms of the Kantorovich distance in the space of probability measures and then provides an interesting application of this result in the study of homogenization of discrete parabolic equations with random coefficients. In particular the problem of the convergence of any square integrable cadlag martingale \((M_{t})_{t \in [0,1]}\) as \(t \to 1\) to the Gaussian law is considered, and an upper bound for the Kantorovich distance between \(M_1\) and \(\Phi\), the cumulative distribution function of the standard Gaussian random variable is provided in terms of \[ d_{1}(M_1,\Phi) \leq C_{p} L_{2p}^{1/(2p+1)} +2 \| \langle M\rangle_{1}-1 \|_{1}^{1/2}, \,\,\, p>1, \] where \(C_{p}\) is a constant independent of \(M\), \((\langle M \rangle_{t})_{t \in [0,1]}\) is the predictable quadratic variation of \(M\) and \[ L_{2p}={\mathbb E}\left[ \sum_{0 \leq t \leq 1} |\Delta M(t)|^{2p} \right], \] where \(\Delta M(t)=M_{t}-\lim_{s \to t^{-}} M_{s}\). These bounds are sharper than the ones obtained previously in the literature for the Kolmogorov distance. Then, this result is applied to the study of a discrete parabolic equation with random coefficients of the form \[ \frac{\partial u^{(\epsilon)}}{\partial t} = L^{\omega} u^{\epsilon}, \,\,\, \text{on} \,\,\, {\mathbb R}_{+} \times {\mathbb Z}^{d}, \] with initial condition \(u^{(\epsilon)}(0,\cdot)=f(\epsilon \, \cdot)\) on \({\mathbb Z}^{d}\), where \(L^{\omega}\) is the discrete operator corresponding to a random walk with random conductance \(L^{\omega}f(x)=\sum_{y \sim x} \omega_{x,y} (f(y)-f(x))\), \(x \in {\mathbb Z}^{d}\), and its continuous limit \[ \frac{\partial \bar{u}}{\partial t}= \frac{1}{2} \nabla \cdot (A \nabla \bar{u}), \,\,\, \text{on} \, (0,+\infty) \times {\mathbb R}^{d}, \] with initial condition \(\bar{u}(0,\cdot)=f\), in terms of homogenization theory. A number of interesting detailed asymptotic results are obtained, and in particular upper bounds on the quantity \(|{\mathbb E}[u^{(\epsilon)}(t,x)]-\bar{u}(t,x)|\) depending on the degree of smoothness of \(f\), as well as upper bounds on the error of the approximation of the heat kernel of the discrete random parabolic equation by the corresponding heat kernel for its homogenized limit. A detailed comparison with similar results obtained in the literature is provided.