[petsc-users] SLEPC solvers not converging in latest version
Jose E. Roman
jroman at dsic.upv.es
Thu Jul 13 06:28:57 CDT 2017
Sidd,
I don't know why you are getting different behaviour in the current and older installations. There is no substantial difference between slepc-3.7.1 and slepc-3.7.4 in the parts you are using.
Anyway, computing smallest_magnitude eigenvalues in a generalized eigenproblem is a bad idea, it is much better to compute largest eigenvalues, either swapping the matrices, or using shift-and-invert with sigma=0. Below you can see the output I get. This strategy should be more robust and I would bet the problem you report will not appear.
$ mpiexec -n 1 ./ex7 -f1 A_mat_binary -f2 B_mat_binary -eps_gen_hermitian -eps_nev 50 -eps_ncv 100 -eps_tol 1e-12 -st_pc_factor_mat_solver_package mumps -eps_monitor -st_type sinvert -eps_target 0
Generalized eigenproblem stored in file.
Reading COMPLEX matrices from binary files...
1 EPS nconv=11 first unconverged value (error) 0.0131612+0i (8.75138656e-12)
2 EPS nconv=11 first unconverged value (error) 0.0131612+0i (1.12729823e-12)
3 EPS nconv=31 first unconverged value (error) 0.0197448+0i (6.83423520e-03)
4 EPS nconv=40 first unconverged value (error) 0.0244434+0i (1.36991619e-12)
5 EPS nconv=40 first unconverged value (error) 0.0244434+0i (1.64341819e-11)
6 EPS nconv=49 first unconverged value (error) 0.0272636+0i (2.53225772e-12)
7 EPS nconv=73 first unconverged value (error) 0.0338452+0i (1.02615547e-07)
Number of iterations of the method: 7
Number of linear iterations of the method: 312
Solution method: krylovschur
Number of requested eigenvalues: 50
Stopping condition: tol=1e-12, maxit=212
Linear eigensolve converged (73 eigenpairs) due to CONVERGED_TOL; iterations 7
---------------------- --------------------
k ||Ax-kBx||/||kx||
---------------------- --------------------
0.002820 1.11732e-12
0.005640 8.2307e-13
0.005640 6.28498e-13
0.005640 1.33386e-12
0.008461 1.66237e-12
0.008461 6.73195e-13
0.008461 5.51759e-13
0.010341 4.572e-13
0.010341 4.71274e-13
0.010341 5.66704e-13
0.011281 5.64244e-13
0.013161 6.26089e-13
0.013161 5.68378e-13
0.013161 6.06859e-13
0.013161 5.14892e-13
0.013161 4.86807e-13
0.013161 7.12961e-13
0.015981 7.21151e-13
0.015981 6.6669e-13
0.015981 6.95718e-13
0.016923 7.27843e-13
0.016923 5.9206e-13
0.016923 9.19785e-13
0.017862 6.79618e-13
0.017862 6.86632e-13
0.017862 8.48137e-13
0.019743 1.73846e-12
0.019743 1.03895e-12
0.019743 7.29832e-13
0.019743 9.52146e-13
0.019743 6.08032e-13
0.019743 8.11897e-13
0.020682 8.97699e-13
0.020682 8.84601e-13
0.020682 5.55719e-13
0.022563 5.86552e-13
0.022563 5.27907e-13
0.022563 5.43232e-13
0.024443 4.99962e-13
0.024443 9.92323e-13
0.024443 6.195e-13
0.024443 5.15713e-13
0.024443 7.09306e-13
0.024443 6.86883e-13
0.025382 9.61258e-13
0.025388 8.24511e-13
0.025388 6.1544e-13
0.025388 3.64722e-13
0.027264 8.3364e-11
0.027264 4.71596e-13
0.027264 8.76059e-13
0.027264 5.89152e-13
0.027264 1.38562e-12
0.027264 7.34739e-13
0.028208 8.70508e-13
0.028208 5.33376e-13
0.028208 1.41842e-10
0.028208 7.31921e-13
0.028208 9.3135e-13
0.028208 1.3238e-12
0.031025 8.70416e-13
0.031025 4.034e-13
0.031025 5.21825e-13
0.031029 5.07974e-13
0.031029 5.93828e-13
0.031029 8.67516e-13
0.031964 6.06749e-13
0.031964 1.26228e-12
0.031964 4.21629e-13
0.032909 4.91244e-12
0.032909 5.36743e-13
0.032909 1.46225e-12
0.032909 4.99251e-13
---------------------- --------------------
> El 12 jul 2017, a las 23:15, Pandey, Siddhant <spandey2 at wpi.edu> escribió:
>
> I have attached the matrices and the 'eps_view_pre' ouput in the zip file.
>
> The versions of the older installation were: petsc-3.7.5, slecp-3.7.1;
> and the current versions are: petsc-3.7.6, slecp-3.7.4.
>
>
> Thanks,
> Sidd (WPI)
> From: Jose E. Roman <jroman at dsic.upv.es>
> Sent: Wednesday, July 12, 2017 4:29:57 PM
> To: Pandey, Siddhant
> Cc: Hong Zhang; petsc-users at mcs.anl.gov
> Subject: Re: [petsc-users] SLEPC solvers not converging in latest version
>
>
> > El 12 jul 2017, a las 22:05, Pandey, Siddhant <spandey2 at wpi.edu> escribió:
> >
> > I've attached the B matrix (binary) in a zip file. To give more context to the problem, the previous installation of petsc/slecp doesn't give this error. So the matrix being singular shouldn't be the issue since my matrix assembly routines are independent of the petsc/slepc installation I use. Also, diagonalization happens for smaller matrices.
> >
> > Unfortunately, since EPSsolve never got completed, -eps_viewer didn't return anything. Is there any other way I can show the settings used?
> >
> > Sidd
>
> Use -eps_view_pre (it should view settings at the beginning of solve), or run a smaller case (settings should be the same).
>
> I will also need the other matrix, otherwise I cannot run EPS.
>
> Also, tell me which version are you using now, and which version worked before.
>
> Jose
>
> <A,B,eps.zip>
More information about the petsc-users
mailing list