A portable, scalable implementation of a multifrontal Cholesky algorithm on a distributed memory machine

A large class of linear systems have a coefficient matrix that is sparse, sym-metric, and positive definite. Many of these sparse linear systems are large, and a direct solution using Cholesky factorization may require extensive time and memory on serial computers. Therefore, it is desirable to solve these systems in parallel. This thesis presents a scalable multifrontal sparse Cholesky factoriza-tion scheme for distributed memory architectures. One of the advantages of the work is the use of “off the shelf”, dense, library kernels. This thesis develops the algorithm and presents performance results on the Intel Paragon.