BANDWIDTH REDUCTION ON SPARSE MATRICES BY INTRODUCING NEW VARIABLES.

A sparse matrix bandwidth reduction method is analyzed. It consists of equation splitting, substitution and introducing new variables, similar to the substructure decomposition in the finite element method (FEM). It is especially useful when the bandwidth cannot be reduced by strategically interchanging columns and rows. In such cases, equation splitting and successive reordering can further reduce the bandwidth, at cost of introducing new variables. While the substructure decomposition is carried out before the system matrix is built, the given approach is applied afterwards, independently on the origin of the linear system. It is successfully applied to a sparse matrix, the bandwidth of which cannot be reduced by reordering. For the exemplary FEM simulation, an increase of performance of the direct solver is obtaine. [ABSTRACT FROM AUTHOR]