sparce matrices

[Peter Schmitteckert]

Salut,

On Wednesday 03 October 2001 20:31, Edwin Robert Tisdale wrote:
> Matthew J. Doller wrote:
> > I am about to begin a project at my school using gsl
> > and my advising professor seems to be very interested in gsl.
> > one question he did have was,
> > "Are there any constructs to store sparces matrices?"
> > It might be beneficial to have such a way to store them
> > especially when dealing with matrices that are 10e8 x 10e8 in size.
>
> First, let me point out that a matrix -- even a sparse matrix --
> of that size has no physical relevance.
> Even a matrix of size 10,000 by 10,000 is on shaky numerical ground.
> More typically, large sparse matrices are on the order of 1,000 by
> 1,000.
> If they are much larger than that,
> the user is probably doing something very, very wrong.

Please let me note that I diagonalized 10e7x10e7 matrices as
a sub-problem of a larger calculation (Density matrix renormalization
group calculation) on high end workstation about 5 years ago, with the
aim of getting more than 5 Digits accuracy of the lowest eigen state of a
10e30x10e30 symmetric matrix, and it worked very well. If you're interested,
just search my name on the cond-mat preprint server.

Best wishes,
Peter