Parallel computation of orthogonal factors of sparse matrices
- Pennsylvania State Univ., University Park, PA (United States)
We study the solution at the least squares problem for a large and sparse overdetermined matrix A by QR factorization A and transformation the right-hand side vector b to Q{sup T}b. An efFicient method for computing Q{sup T}b on P-processor distributed memory machines is presented. In this paper, we determined the number of multiplications required to compute Q{sup T}b using the multifrontal Householder QR factorization by Lewis, Pierce, and Wah[Technical report ECA-TR-127, Boeing computer services(1989)]. Instead of storing the Q matrix itself, we store all the nonzero parts of the Householder transformation matrices Qi`s of all the multifrontal matrices Fi`s of A. We also use Schreiber and Van Loan`s Storage-Efficient-WY Representation [SIAM J. Sci. Stat. Computing, 19(1989), pp 55-57] for each Qi to introduce BLAS-2 operations. A theoretical operation count for the K by K unbordered grid model problem shows that the proposed method requires O(N{sub R}) multiplications (O(K{sup 2}log K) for the K by K grid model problems) to compute Q{sup T}b, where N{sub R} represents the number of nonzeros of the Cholesky factor of A. Some numerical results for the grid model problems as well as Harwell-Boeing problems on both sequential machines and iPSC distributed memory machines are provided.
- OSTI ID:
- 54445
- Report Number(s):
- DOE/ER/25151--1-Vol.1; CONF-930331--Vol.1; CNN: Grant CCR-9201692
- Country of Publication:
- United States
- Language:
- English
Similar Records
Parallel multifrontal solution of sparse linear least squares problems on distributed-memory multiprocessors
Computing rank‐revealing factorizations of matrices stored out‐of‐core