Parallel implementations of Modified Gram-Schmidt orthogonalization /

Interest in the QR factorization arises from the ographics.

Bibliographic Details
Main Author: Soma, Takako
Format: Thesis Book
Language:English
Published: [Place of publication not identified] : [publisher not identified] ; 1998.
Subjects:
Description
Summary:Interest in the QR factorization arises from the ographics.
important role it plays in a large number of
computational algorithms in linear algebra such as the
estimation of least squares and the calculation of
eigenvalues and eigenvectors. The basic idea consists
of calculating the factorization of an 'rn x n matrix
W, where m k zs. That is, finding Q and R where ,4 =
QR, Q is an orthonormal matrix (QTQ = f ), and R is
upper triangular. Gram-Schmidt orthogonalization is a
classical mathematical procedure for solving the
orthogonal bases problem. A particular important
modification of the numerically unstable classical
Gram-Schmidt is known as the Modified Gram-Schmidt
(MGS) algorithm. The importance of the QR
factorization in numerical matrix calculation and its
high algorithmic complexity, which is around O(mn2 ) ,
has motivated a huge interest in obtaining efficient
parallel implementation. Here, implementation of
parallel Gram-schmidt orthogonalization algorithms are
analyzed. Four types of column-wise partitioning
schemes including one-column, block, cyclic and block-
cyclic partitioning, and a row-wise partitioning of
O'Leary and Whitman were considered. Analytical models
for parallel execution time required by these
implementations are derived and compared with
numerical results. Threshold values of ['rrlraaz],
which is the number of rows where row partitioning
becomes better than column partitioning are found
theoretically and verified numerically. Our
theoretical analysis of column-wise algorithms, which
were implemented using pipelined fashion predicts
architecture independence. An important application of
the orthogonal factorization, the least squares
problem is discussed and other applications are
presented.
Item Description:"Major subject: Computer Science".
Vita.
Physical Description:vii, 73 leaves : illustrations ; 28 cm.
Bibliography:Includes bibliographical references: pages 61-72.