Computing the Singular-Value Decomposition on the ILLIAC IV
Franklin T. Luk · ACM Transactions on Mathematical Software · 1980
This paper presents the computation of the singular-value decomposition of a matrix on the ILLIAC IV computer.The architecture of the machme is described and it is explained why the standard Golub-Remsch algorithm is not apphcable to this problem.Then a one-sided orthogonalization method is presented.This method makes very efficmnt use of the parallel-computing abilities of the [LLIAC machine The method is shown to be Jacobi-hke and numerically stable.Finally, a comparison of the method on the ILLIAC IV computer with the Golub-Reinsch algorithm on a conventional machme demonstrates the great potentml of parallel computers m the important area of matrLx computations Key Words and Phrases ILLIAC IV computer, singular-value decomposmon, Golub-Remsch algorithm, Jacobl-hke method, parallel matrix computations CR Categories 5 14