PARTITIONED ITERATIVE SOLVER FOR VERY LARGE THERMAL SIMULATIONS

Abstract

To enhance the capabilities of solving very large homogeneous and heterogeneous heat conduction simulations, a partitioned iterative solver is developed, lite methodology involves the use of a multiply gauged, preconditioned, conjugate gradient procedure. In particular, local substructural level gauging is introduced for both the steepest descent and orthogonalization-conjugation phases of operation, lb improve convergence noes for very-large-scale thermal simulations, the scheme combines partitioned initialization and “bias control” steps. Due to the use of local gauging, quicker penetration toward the solution is realized. This is verified by an investigation of formal properties, as well as from the speed-up results of several benchmark examples involving a wide range of problem sizes. These include both homogeneous and heterogeneous conduction properties. Additionally, various substructures formats are considered, namely, stripped, checkerboard, and layered shells. Such problems point to the robustness of the multiply gauged iterative scheme.