On the stability of splitting difference schemes with respect to boundary conditions (Q1291235): Difference between revisions

Economical splitting difference schemes are widely used for solution of multidimensional nonstationary equations. In these methods, the accuracy of computations essentially depends on the formulation of boundary conditions at fractional time moments. We say that boundary conditions of a scheme are the simplest boundary conditions if they coincide with those of the differential problem considered. It is well-known that the convergence rate of locally one-dimensional (LOD) schemes with the simplest boundary conditions can be essentially lower than the approximation order inside the region. The accuracy reduction is connected with a lack of approximation compatibility at points adjacent to the boundary. Some correction methods for the simplest boundary conditions assuring the optimal accuracy order of LOD methods were proposed and studied earlier. Such a correction, however, often cannot be economically implemented because of the complexity of either the domain in which the problem is solved, or of the differential operator. Therefore determining the accuracy of LOD schemes in the case of the simplest boundary conditions becomes an important problem. We actually get the problem of stability of splitting difference schemes with respect to boundary conditions. This article is a continuation of the authors' work [Differ. Uravn. 30, No. 3, 480-487 (1994; Zbl 0823.65086)], where it was shown that the stability of an LOD scheme depends on the stability of the corresponding one-dimensional difference scheme with respect to nonsmooth boundary conditions. In this article, we prove some results on the same type of stability of weighted one-dimensional schemes for parabolic problems with separated arguments. This enables us to get unimprovable in \(L_2\)-norm accuracy estimates for the solution of the class considered of LOD schemes in the case of the simplest boundary conditions. We give the results of a numerical experiment illustrating the accuracy estimates of LOD schemes proved in this article.