Spectral measure and approximation of homogenized coefficients (Q1930857): Difference between revisions

This paper deals with the numerical approximation of effective coefficients in stochastic homogenization of discrete linear elliptic equations. The authors consider a discrete elliptic operator \(-\nabla^{*}\cdot A\nabla\), where \(\nabla^{*}\) and \(\nabla\) are the discrete backward divergence and forward gradient, respectively. For all \(z\in Z^{d}, A(z)\) is the diagonal matrix whose entries are the conductances \(\omega(z,z+e_{i})\) of the edges \((z,z+e_{i})\) starting at \(z\), where \(e_{i}, i=1,\ldots,d\) denotes the canonical basis of \(Z^{d}\). The values of the conductance are independent and identically distributed random variables. The authors first describe the strategy on the case of periodic homogenization of continuous elliptic equations. It is introduced the spectral decomposition formula for the homogenized coefficients. The binomial formula then provides with natural approximations of the homogenized coefficients in terms of the associated spectral measure. The authors rewrite these formulas in physical space using solutions to the modified corrector equation, which yields new computable approximations of the homogenized coefficients. For the stochastic homogenization of discrete elliptic equations the authors recall the spectral decomposition of the generator of the environment viewed by the particle. Then it is shown that the asymptotic behavior of the systematic error is driven by the behavior of the edge of the spectrum of the generator. The authors obtain estimates on the edge of this spectrum, which show that the systematic error is effectively reduced in high dimensions (obtained bounds are not optimal when \(d>6\)).