Re: [Help-gsl] LAPACK,CBLAS matrix inversion

[Wei Cheng]

I have tried matrix inversion using GSL LU,SVD and Cholesky decomposition ,
LU is the best and SVD is the slowest(probably because it taks more steps).
But in my application I need to invert 200 X 200 matrices 30,000,000 times.
It is taking toooo long and not yet finished.  Since Matlab is using Lapack,
so.. maybe lapack is more effeicent on large matrix.

Wei


On 8/7/06, Brian Gough <address@hidden> wrote:
> At Fri, 4 Aug 2006 14:06:39 +0100,
> Wei Cheng wrote:
> > The cholesky decomposition and LU decomposition in GSL linear algerbra
> > section, is only suitable for small matrix, for large matrix LAPACK is
> > better as stated on GSL website. I am wondering if there is anyone who
> has
> > experience in using cholesky and LU decomposion in LAPACK?? My matrix is
> 500
> > by 500. or where to find relavent information? Someone suggested to use
> > CBLAS. What is the difference between LAPACK and CBLAS? Which is
> better??
>
> Hello,
> I'd suggest you try the GSL routines first, following the examples in
> the manual.  If they are not fast enough, try LAPACK.  The GSL
> routines are acceptable in most cases -- for comparison, on a 500x500
> SVD the GSL speed is about 30% of LAPACK.
>
> --
> regards,
>
> Brian Gough
>
> Network Theory Ltd,
> Publishing the GSL Manual - http://www.network-theory.co.uk/gsl/manual/