I agree that LAPACK is slow since it computes all eigenvalues, but I found that alternative eigensolvers (like SLEPc) generally failed to converge for the SCM eigenproblems, so in my experience LAPACK is the only reliable choice. I don't have any other suggestion that I can offer on this, unfortunately, since I have never found a better option than LAPACK for this. Also, I don't think there is a way to get only the min or max eigenvalues from LAPACK, I think it automatically computes all eigenvalues (but you could check the LAPACK documentation if you want to be sure about that).
In practice I have rarely used the SCM since I generally don't compute rigorous error bounds, as I explained in some earlier emails. But if you do want rigorous error bounds then I agree that the SCM is necessary. Best regards, David On Thu, Apr 4, 2019 at 8:52 PM <ss.k...@pusan.ac.kr> wrote: > Hello, David. > > > > I have applied SCM in my 3-D elasticity problem with the fine mesh. > > Although the SCM process works, It took too long to compute eigenvalues > using LAPACK. > > > > The LAPACK yielded in almost all eigenvalues, which was a time-consuming > cause. > > I tried to find ways to reduce the eigenvalue computation time, but I > could not obtain a solution. > > Therefore, I would like to ask you for help. > > > > My questions are: > > 1. Can we use only the LAPACK in SCM? I tried to use Krylov-Schur > solver, but there was an error that failed to convergence the eigenvalue in > the “compute_SCM_bounding_box()” process. > > 2. Is there a way to derive only the minimum or maximum eigenvalue > in LAPACK? > > > > I am always grateful for your help. > > > > Best regards, > > > > SKang > _______________________________________________ Libmesh-users mailing list Libmesh-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/libmesh-users
