Probabilistic representations for systems of elliptic equations (Q1363935)

It is well-known that various random-walk algorithms are exteremely useful for the solution of multidimensional boundary vlaue problems with different geometric features, which are often complicated by the stochastic character of some parameters. In the present paper, we propose random-walk algorithms on spheres based on a generalization of the theorem of averages in a special vector form. We discover such cases in which the variance is uniformly finite or slowly divergent, and derive complexity estimates of the form \(T_\varepsilon\sim|\ln\varepsilon|^2/\varepsilon^2\), which points to the high efficiency of the method. Finally, We isolate the cases in which the divergence of variance is exponential. The system of Lamé equations, in particular, belongs to this class. For that system we propose a new modification of the random-walk algorithm with nonisotropic distribution over the sphere, which reduces the variance.