Hi, I'm interested in this as well. Could you post or link to your solution? Thanks.
On Friday, 19 February 2016 17:21:22 UTC+1, Michael Harmon wrote: > > If you want to use Trilinos, I refactored their direct solver so that it > can be used the way you want, I will submit the patch, but if you want can > send you the changes as well. > > > On Tuesday, December 1, 2015 at 4:05:29 AM UTC-6, Pascal Kraft wrote: >> >> Hello Wolfgang, >> >> does the same hold true for other direct solvers? I would like to use >> Mumps (PETSc) to get a LU-factorization and reuse that a couple of times. >> The PETSc wrappers only provide a constructor and solve. If I reuse solve >> withe the same matrix argument, will the factorization be done again or >> does it detect that the inverse is already present? The code for both the >> PETSc wrappers as well as the Trilinos Wrappers does not look as if that >> was supported. >> >> Greetings, >> Pascal Kraft >> >> Am Sonntag, 4. Oktober 2015 20:14:30 UTC+2 schrieb bangerth: >>> >>> >>> Zhen, >>> >>> Jean-Paul already gave the correct answer, namely use a sparse direct >>> solver >>> and factorize only once. >>> >>> > Therefore my following questions are: >>> > >>> > What is the limit of the size of matrix system can I do a lapack LU >>> > factorization on a single machine? >>> >>> For a sparse direct solver, I think you can probably go to around 1M >>> unknowns. >>> If you use a LAPACK matrix, you will quickly be limited by memory. For >>> example, a 10,000 x 10,000 matrix already requires almost a GB of >>> memory. >>> >>> >>> > Are there any better way to get LU factorization in deal.ii? >>> > Can we incorporate "history experience" to achieve more efficient >>> solvers? >>> >>> People have tried that and there is a substantial amount of literature >>> on the >>> topic of Krylov subspace recycling. I always think that it is a great >>> idea and >>> want to investigate using it, but it has not found its way into >>> mainstream >>> applications to the best of my knowledge. >>> >>> Best >>> W. >>> >>> -- >>> ------------------------------------------------------------------------ >>> Wolfgang Bangerth email: bang...@math.tamu.edu >>> www: >>> http://www.math.tamu.edu/~bangerth/ >>> >>> -- The deal.II project is located at http://www.dealii.org/ For mailing list/forum options, see https://groups.google.com/d/forum/dealii?hl=en --- You received this message because you are subscribed to the Google Groups "deal.II User Group" group. To unsubscribe from this group and stop receiving emails from it, send an email to dealii+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
