Dynamic Load Balancing Using Adaptive Locally Refined Meshes

Methods for representing and processing algorithms for dynamically adaptive locally refined meshes are proposed for serial and parallel computing systems, including hybrid ones. Estimates of the complexity of the algorithms are given. New parallel algorithms for the decomposition of locally condensed meshes and dynamic load balancing, which ensure low overhead costs and reduce the overall computation time for two- and three-dimensional numerical modeling, are proposed. Time reduction is achieved by decreasing the number of cells in the computational mesh (relative to the regular mesh) and through parallel processing.