Quantitative version of the Kipnis-Varadhan theorem and Monte Carlo approximation of homogenized coefficients (Q363858)
From MaRDI portal
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | Quantitative version of the Kipnis-Varadhan theorem and Monte Carlo approximation of homogenized coefficients |
scientific article |
Statements
Quantitative version of the Kipnis-Varadhan theorem and Monte Carlo approximation of homogenized coefficients (English)
0 references
5 September 2013
0 references
The purpose of the paper is to describe a stochastic homogenization result for discrete elliptic equations using a Monte Carlo method. Let \(\nabla \) (resp. \(\nabla ^{\ast }\)) be the forward gradient (resp. backward divergence) operator; \(A(x)\) be the diagonal matrix whose entries are the conductances \(\omega _{x,x+e_{i}}\) of the edges \((x,x+e_{i})\) starting at \(x\) and where \((e_{i})_{i=1,\dots ,d}\) denotes the canonical basis of \(\mathbb{R }^{d}\). If \(\mathbb{B}\) is the set of unoriented edges of \(\mathbb{Z}^{d}\), the set \((\omega _{e})_{e\in \mathbb{B}}\) is called the environment which is symmetric and random with distribution \(\mathbb{P}\) and expectation \(\mathbb{ E}\). It is assumed that \(\mathbb{P}\) is invariant under translations, that the conductances are i.i.d. and that there exists \(0<\alpha <\beta \) such that \(\alpha \leq \omega _{e}\leq \beta \) almost surely. The operator \( \nabla ^{\ast }\cdot A\nabla \) is the infinitesimal generator of a stochastic process whose jump rate from a site \(x\) to a neighboring site \(y\) is given by \(\omega _{x,y}\). \(\mathbf{P}_{0}^{\omega }\) is the law of this process starting at \(x\). Let \(Y(t)\) be a random walk. The authors introduce \(\sigma _{t}^{2}=t^{-1}\widetilde{\mathbb{E}}\mathbf{E}_{0}^{\omega }[(\xi \cdot Y(t))^{2}]\) for some fixed \(\xi \in \mathbb{R}^{d}\) with \(\left| \xi \right| =1\). It is known that \(\sigma _{t}^{2}\) converges to \(\sigma ^{2}=\frac{2\xi \cdot A_{\hom }\xi }{\mathbb{E}[p]}\), where \(A_{\hom }\) is the homogenized operator associated to this homogenization process. The main result of the paper considers the quantity \[ \widehat{A}_{n}(t)=\frac{ p(\omega ^{(1)})(\xi \cdot Y^{(1)}(t))^{2}+\dots +p(\omega ^{n1)})(\xi \cdot Y^{(n)}(t))^{2}}{nt\mathbb{E}[p]}. \] It proves that there exists \(C,c>0 \) such that \(\mathbb{P}^{\otimes }\mathbf{P}_{0}^{\overline{\omega } }\biggl[\left| \widehat{A}_{n}(t)-\sigma ^{2}\right| \geq \frac{C\mu _{d}(t)+\varepsilon }{t}\biggl]\leq \exp (-\frac{n\varepsilon ^{2}}{ct^{2}})\), where \(\mu _{d}(t)=\ln ^{q}t\) if \(d=2\) and \(\mu _{d}(t)=1\) if \(d>2\) for some \(q>0\) depending only on \(\alpha \) and \(\beta \). Here \(Y^{(1)},\dots ,Y^{(n)} \) are independent random walks evolving on the environments \(\omega ^{(1)},\dots ,\omega ^{(n)}\), describing a Monte Carlo method. The proof of this result gathers probabilistic arguments (martingale description, Kipnis-Varadhan theory, large deviation estimates) and arguments of elliptic theory (Harnack inequality, De Giorgi-Nash-Moser theory and \(L^{p}\)-theory).
0 references
random walk
0 references
random environment
0 references
stochastic homogenization
0 references
effective coefficients
0 references