Re: [Boost-Users] uBLAS - Calculation of Inverse and/or Least Squares

From: Kresimir Fresl (fresl_at_[hidden])
Date: 2003-01-07 09:21:33

Hi Jon,

jhr.walter_at_[hidden] wrote:

> From: jonny5cents <jon_nichols95_at_[hidden]>

>>Hello All,
>>
>>   Can anyone suggest the best way to use uBLAS to calculate the
>>inverse of a matrix??
>>
>>   One of the reasons I want to do this is to solve a linear least
>>squares problem, which I know can be done without explicitly solving
>>for the inverse.  Does anyone know if this is possible??

> There are two ways to go: either you could try to port the corresponding
> LAPACK routine(s) to use uBLAS or you could consider to use and extend the
> bindings in the boost sandbox written by Toon Knapen and Kresimir Fresl.

I just added functions `getri()' and `potri()' to ATLAS
(http://math-atlas.sourceforge.net/) bindings.

`getri()' computes the inverse of a general matrix A using
the LU factorization previously computed by `getrf()'.

`potri()' computes the inverse of a symmetric or Hermitian
positive definite matrix A using the Cholesky factorization
previously computed by `potrf()'.

But please note that `bindings' in boost sandbox are not updated
yet. Therefore you must download `bindings_2003_01_07.zip' from
`ublas-dev' mailing list file section.

There's no documentation, but some simple examples
of use with uBLAS matrices can be found in
`libs/numeric/bindings/atlas/ublas_getri.cc' and
`libs/numeric/bindings/atlas/ublas_potri.cc'.

Sincerely,

fres