Averaging error for elliptic equations with ``layered'' random coefficients (Q1074795)

The author considers the Dirichlet problem \[ L_{\varepsilon} u_{\varepsilon}=f,\quad u_{\varepsilon}|_{\partial G}=0,\quad G\subset {\mathbb{R}}^{d+1}=(x^{(0)},x)=(x^{(0)},x^{(1)},...,x^{(d)}) \tag{1} \] in a bounded domain \(G\) with smooth boundary \(\partial G\) for a differential operator \[ L_{\varepsilon}=-\sum^{d}_{i,j=0}a_{ij}(\bar x/\varepsilon,\omega)D_i D_j+\sum^{d}_{i=0}b_i (\bar x/\varepsilon) D_i+c(\bar x/\varepsilon,\omega),\quad \bar x=(x^{(0)},x). \] Here a field of the coefficients \(\bar k(\bar x,\omega)= \{a_{ij}(\bar x,\omega),b_i(\bar x,\omega),c(\bar x,\omega)\}\) is periodic with respect to variables \(x^{(1)},\ldots,x^{(d)}\) with period 1 and is random homogeneous process with respect to variable \(x^{(0)}\): \(\bar k(x^{(0)}\pm 1,x,\omega)= \bar k(x^{(0)},x,S^{\pm 1}\omega)\) and transformation \(S\) of the corresponding probability space \((\Omega,F,P)\) satisfies the strong mixing condition \[ | P(A\cap B)-P(A)P(B)| \leq C | x|^{-\beta},\quad C,\beta >0,\quad A\in F(-\infty,a),\quad B\in F(a+x,\infty). \tag{2} \] It is well known that in rather general situation the solution \(u_{\varepsilon}\) of such problem (which describes the random walk in random environment) tends as \(\varepsilon\to 0\) to the solution \(U\) of the corresponding problem with the constant (averaged) coefficients \(<\bar k>\), which are to be calculated according to the special procedure [see e.g. \textit{G. Papanicolaou}, and \textit{S. Varadhan}, Statistics and Probability: Essays in Honor of C. R. Rao, 547--552 (1982; Zbl 0486.60076); the author, Sib. Math. J. 23, 276--287 (1982; Zbl 0518.34047)]. Theorem. If the coefficients \(\bar k\) in addition to listed properties are such that \[ a-\xi^{(i)} \xi^{(i)}\leq a_{ij} \xi^{(i)} \xi^{(j)}\leq a+\xi^{(i)} \xi^{(i)},\quad | D_ j a_{jk}(\bar x,\omega)| \leq A \] where \(a_{\pm}\), \(A\) are nonrandom constant, then \[ E\{\sup_{\bar x\in G}| u_{\varepsilon}-U|^{d+1}\}\leq C\varepsilon^{A_*} \] where \(C\) does not depend on \(\varepsilon\) and \(A^*\) is determined by \(d\) and the exponent \(\beta\) in the strong mixing condition (2). The author proves also a variety of estimates for the moments of the random variables \(\bar k-<\bar k>\) and the related quantities which may be useful in similar problems.