LQ Decomposition and Null Space
  Home FAQ Contact Sign in
comp.lang.fortran only
 
Advanced search
POPULAR GROUPS

more...

comp.lang.fortran Profile…
 Up
LQ Decomposition and Null Space         


Author: michael
Date: Mar 29, 2007 13:12

Given a matrix A(m,n) where m<=n, LAPACK subroutine DGELQF calculates
the LQ decomposition A=(L 0)Q where L(m,m) and Q(n,n). The subroutine
DORGLQ forms the first m rows of the matrix Q. These rows correspond
to the Transpose(Range) space of the matrix A. The null space
transpose corresponds to the n-m rows of Q. Does anyone know of an
efficient way to form this null space?

Michael
2 Comments
Re: LQ Decomposition and Null Space         


Author: Pierre Asselin
Date: Mar 31, 2007 07:50

michael@athenavisual.com wrote:
> Given a matrix A(m,n) where m<=n, LAPACK subroutine DGELQF calculates
> the LQ decomposition A=(L 0)Q where L(m,m) and Q(n,n). The subroutine
> DORGLQ forms the first m rows of the matrix Q. These rows correspond
> to the Transpose(Range) space of the matrix A. The null space
> transpose corresponds to the n-m rows of Q. Does anyone know of an
> efficient way to form this null space?

See the part on Householder matrices in section 5.2 of Gilbert
Strang, "Introduction to Applied Mathematics", starting on p. 392
(Wellesley-Cambridge Press 1985, ISBN 0-9614088-0-4). It gives an
LQ decomposition (QR, actually, but take the transpose) with a
square matrix Q, including your missing rows.

I don't know if there is a Lapack for that.

--
pa at panix dot com
no comments
Re: LQ Decomposition and Null Space         


Author: Udo Grabowski
Date: Apr 1, 2007 06:48

michael@athenavisual.com wrote:
> Given a matrix A(m,n) where m<=n,
>...
> Does anyone know of an efficient way to form this null space?

DGESVD from LAPACK gives an SVD decomposition A = U * S * V^t,
where the elements of the columns of U corresponding to nonzero
singular values in S span the range and the columns of V corresponding
to zero singular values in S form the nullspace of A.

See also
<http://www.uwlax.edu/faculty/will/svd/index.html>
for a very nice visual introduction to SVD.
no comments