Large-Scale Modal Analysis on Multi-Core Architectures
Krishnan Suresh, Praveen Yadav · 2012
We propose here a subspace augmented Rayleigh-Ritz conjugate gradient method (SaRCG) for solving large-scale eigen-value problems. The method is highly scalable and well suited for multi-core architectures since it only requires sparse matrix-vector multiplications (SpMV). As a specific application, we consider the modal analysis of geometrically complex structures that are discretized via non-conforming voxels. The voxelization process is robust and relatively insensitive to geometric complexity, but it leads to large eigen-value problems, that are difficult to solve via standard eigen-solvers such as block-Lanczos. Such problems are easily solved via the proposed SaRCG, where one can, in addition, exploit the voxelization structure to render the SpMV assembly-free. As the numerical experiments indicate, the resulting implementation on multi-core CPUs, and graphics-programmable-units is a practical solution to automated eigen-value estimation during early stages of design.