[petsc-users] computing matrix exponentials

[amlan barua]

Jed,
Which one will give better performance - 1) a precomputed exp(A*t) and
dense matrix-vector multiplication exp(A*t) v  or 2) computing  exp(A*t) v
using Krylov subspace methods. My exp(A*t) will not change between time
steps.
Amlan


On Sun, Jan 19, 2014 at 12:08 PM, Jed Brown <jed at jedbrown.org> wrote:

> amlan barua <abarua at iit.edu> writes:
>
> > Hi Jose,
> > Thanks for the clarification. So if I want to compute the exponential
> > matrix just once then I should extract the columns one by one and store
> > them in a PetSc matrix.
>
> Normally when people talk about efficient/scalable methods for matrix
> exponential, they mean evaluating
>
>   y = exp(A*t) * x
>
> rather than computing the full matrix exp(A*t).  Which do you want?
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20140119/c14020ee/attachment.html>