Re: Problem with DGESVD - Singular Value Decomposition (LAPACK)
- From: "Beliavsky" <beliavsky@xxxxxxx>
- Date: 5 Apr 2007 08:59:36 -0700
On Apr 5, 11:56 am, Udo Grabowski <udo.grabow...@xxxxxxxxxx> wrote:
mich...@xxxxxxxxxxxxxxxx wrote:
For the matrix
a(1,:)=(/2,5,3,9/)
a(2,:)=(/1,1,9,1/)
a(3,:)=(/1,-1,1,251/)
calling DGESVD with argument JOBU='N' and JOBVT='N' the Compaq Visual
Fortran compiler goes into an infinite loop, while the g95 compilers
returns division by zero.
calling DGESVD with arguments JOBU='A' and JOBVT='N' the Compaq
Visual
Fortran compiler returns log: -SING error, while the g95 compiler
returns the correct singular values that also also match MATLAB
function svd results.
Anyone has experience with the DGESVD routine or any ideas?
Hmm, I think I was the one that recommended using it, so I feel
responsible for
answering here...
I assume A is declared as double precision. Double and triple check the
M and N ,
LDA, LDU and LDVT parameters and read four times through the function
description,
I always get those parameters wrong on the first try...
Yes, and once you understand the calling conventions, write a Fortran
95 wrapper that is easy to use and safe (checking consistency of
dimensions). For Lapack, this has already been done in Lapack95
http://www.netlib.org/lapack95/ .
.
- References:
- Problem with DGESVD - Singular Value Decomposition (LAPACK)
- From: michael
- Re: Problem with DGESVD - Singular Value Decomposition (LAPACK)
- From: Udo Grabowski
- Problem with DGESVD - Singular Value Decomposition (LAPACK)
- Prev by Date: papers on memory management for allocatable components of user-defined types
- Next by Date: Re: Reference Information Manager (RIM) Source
- Previous by thread: Re: Problem with DGESVD - Singular Value Decomposition (LAPACK)
- Next by thread: Re: Problem with DGESVD - Singular Value Decomposition (LAPACK)
- Index(es):
Relevant Pages
|
