[Help-gsl] Re: gsl vs ARPACK++

[Rodney Sparapani]

On 08/24/10 04:08 PM, Rafal Topolnicki wrote:
> Hi,
>
> I need to compute 10% of smallest eigenvalues of huge (3432x3432,
> 12870x12870, 184756x184756) real symmetric sparse matrix (over 99,8% of
> elements =0).
> At first I tried to do it using GSL and in first case it take ~160s to
> capure all eigenvalues. Because matrix is extremely sparse I thought
> that it will be faster if I use ARPACK++ instead of gsl. I use function
>
> ARluSymStdEig<double> dprob(ilosc_wartosci, matrix, "SM");
> dprob.ChangeMaxit(10000000000);
> dprob.FindEigenvectors();
>
> where ilosc_wartosci is number of eigenvalues I want to compute.
> It works, but I takes ~129s to compute 10% of all eigenvalues. I thought
> I would be much faster.
> My question is, if gsl is so good, ARPACK++ so bad or am I doing sth wrong.
>
> BTW, ARPACK++ use sparse matrix in CSC format. Is it possible to write
> matrix with columns consisting only zeros using this code?
>
> Rafal

Hi Rafal:

I seem to remember that ARPACK++ is a sparse matrix package.
As such, it may incur some overhead for smaller examples.
If you try your larger cases, then you may find it beating
GSL (if you can even do it with GSL for those).

Rodney