Parallel orthogonal factorizations of large sparse matrices on distributed-memory multiprocessors
- Cornell Univ., Ithaca, NY (United States)
We describe the issues involved in the design and implementation of an efficient parallel multifrontal algorithm for computing the QR factorization of a large sparse matrix on distributed-memory multiprocessors. The proposed algorithm has the following novel features. First, a supernodal tree computed from the sparsity structure of R is used to organize the numerical factorization. Second, a new algorithm has been designed for the most crucial task in this context-the QR factorization of two upper trapezoidal matrices in parallel. Third, the overall factorization is accomplished by a sequence of Householder and Givens transformations. Experimental results on an Intel iPSC/860 are included.
- OSTI ID:
- 54441
- Report Number(s):
- DOE/ER/25151--1-Vol.1; CONF-930331--Vol.1
- Country of Publication:
- United States
- Language:
- English
Similar Records
Parallel computation of orthogonal factors of sparse matrices
Supernodal symbolic Cholesky factorization on a local-memory multiprocessor