Sveshnikov V.M.  

Parallel algorithms and technologies solving boundary value problems on quasi-structured grids for modeling electron-optical systems

The effectiveness of modeling complex nonlinear electron-optical systems is largely determined by the algorithms and techniques for solving boundary value problems in each step of the linearized iterative process. Multiscale geometry of the computational domain, and strong nonlinearity and heterogeneity of the physical processes that require the introduction of cost-adaptive grids. Quasistructured grid examined in the report meet these requirements. They are built of rectangular uniform locally modified subgrids that provides, first, ease of use and, second, to adapt to the solution. Pairing the solutions in the subdomains are using by the proposed decomposition method based on the approximation of the Poincare-Steklov equation using discrete Green's function. Considered algorithms are basically parallel and, therefore, are useful for implementation on multiprocessor supercomputers. This report describes the parallelization and gives examples of numerical calculations.