Algorithms for the reduction of matrix bandwidth and profile (Q1066617)

The reduction of matrix bandwidth and profile is one of the important objectives in the LU decomposition of sparse, symmetric matrices. A standard approach to attain such reductions is based on the growth of rooted level structures. This paper discusses some algorithms that are intended to produce good starting nodes to build up the corresponding level structures. Some of the algorithms are new and, although the starting nodes they produce are not necessarily peripheral or pseudoperipheral, the numerical tests in the paper suggest that they can be competitive with, for instance, the Gibbs-Poole-Stockmeyer algorithm.