<div dir="ltr">Petsc does not support parallel LU. You can use Superlu_dist or mumps for parallel LU<div>via petsc interface.</div><div><br></div><div>Hong</div></div><div class="gmail_extra"><br><br><div class="gmail_quote">
On Wed, May 22, 2013 at 3:15 PM, Heikki Virtanen <span dir="ltr"><<a href="mailto:heikki.a.virtanen@hotmail.com" target="_blank">heikki.a.virtanen@hotmail.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div><div dir="ltr">Hi, I have been trying to solve a generalized eigenvalue problem using<br>SLEPc's EPS object. I have <span>tried to parallelize my solver, so I should <br>use PETSc's mpiaij matrices instead of seqaij matrices. <br>
Unfortunately, </span><span><span>PETSc's </span>LU preconditioner does not support MPI matrices. <br>(this is said in the manual and if I use it with mpiaij matrices I <br>get an error message) I get the eigenvalue problem solved with <br>
LU preconditioner and seqaij matrices b</span><span><span>ut, I also have to use a <br>spectral transformation ( shift-invert ) to improve convergence. <br><br>But, as I mentioned before I cannot use LU preconditioner with mpiaij <br>
matrices.</span> Thus, I have changed preconditioner to Hypre's BoomerAMG, <br>for example. ( which supports mpiaij matrices ) When I use this <br>combination ( BoomerAMG/shift-invert transformation/<br>Krylov-Schur solver) I get an early convergence failure after a couple <br>
of iterations. I have also tried other solvers and preconditioners, <br>(Hypre's pilut, euclid, bjacobi and Jacobi-Davidson solver) but the result is <br>the same. Without any preconditioner I also get the early convergence<br>
failure and without the spectral transformation convergence is<br>too slow. Any comments or suggestions?<span class="HOEnZb"><font color="#888888"><br><br>-Heikki<br></font></span></span> </div></div>
</blockquote></div><br></div>