[petsc-users] gamg problem

[Mark Adams]

Sorry for the delay. The problem is that the eigen estimates are bad for
the Chebyshev smoother.  Unfortunately this fails catastrophically.

First, I do not understand why we do not get an error message here:

13:57 PICell> ~/Codes/petsc/src/ksp/ksp/examples/tutorials$ mpirun -n 8
./ex54 -ne 1023 -ksp_rtol 1e-10 -ksp_monitor_true_residual -options_left
  0 KSP preconditioned resid norm 2.408779787997e+03 true resid norm
9.779947596963e-04 ||r(i)||/||b|| 1.000000000000e+00
  1 KSP preconditioned resid norm 6.605106317291e+01 true resid norm
6.550622855055e-02 ||r(i)||/||b|| 6.698014268593e+01
  2 KSP preconditioned resid norm 2.105905914541e+01 true resid norm
5.536477156386e-03 ||r(i)||/||b|| 5.661049920253e+00
  3 KSP preconditioned resid norm 2.775390521485e+00 true resid norm
1.822171753713e-03 ||r(i)||/||b|| 1.863171285579e+00
  4 KSP preconditioned resid norm 6.367933585867e-01 true resid norm
2.173992967507e-03 ||r(i)||/||b|| 2.222908605545e+00
#PETSc Option Table entries:
-ksp_monitor_true_residual
-ksp_rtol 1e-10
-ne 1023
-options_left
#End of PETSc Option Table entries
There are no unused options.

It just stops.  What is going on here?????

A problem is that GMRES is the default it is not accuare.  CG is much
better.  I will add this to all tests (54, 55, 56) that use Cheby.  Here is
what we get with CG:

14:12 PICell> ~/Codes/petsc/src/ksp/ksp/examples/tutorials$ mpirun -n 8
./ex54 -ne 1023 -ksp_rtol 1e-10 -ksp_monitor_true_residual
-mg_levels_esteig_ksp_type cg
  0 KSP preconditioned resid norm 2.397141111771e+03 true resid norm
9.779947596963e-04 ||r(i)||/||b|| 1.000000000000e+00
  1 KSP preconditioned resid norm 6.398367771703e+01 true resid norm
5.969698581458e-02 ||r(i)||/||b|| 6.104018986064e+01
  2 KSP preconditioned resid norm 2.025487152015e+01 true resid norm
4.795534114089e-03 ||r(i)||/||b|| 4.903435388118e+00
  3 KSP preconditioned resid norm 2.693141357521e+00 true resid norm
1.392786125856e-03 ||r(i)||/||b|| 1.424124323824e+00
  4 KSP preconditioned resid norm 6.147511674674e-01 true resid norm
2.801964778489e-04 ||r(i)||/||b|| 2.865010012281e-01
  5 KSP preconditioned resid norm 1.133515586579e-01 true resid norm
3.760895077288e-05 ||r(i)||/||b|| 3.845516594031e-02
  6 KSP preconditioned resid norm 1.169295614319e-02 true resid norm
5.672568123760e-06 ||r(i)||/||b|| 5.800202984239e-03
  7 KSP preconditioned resid norm 1.281609546752e-03 true resid norm
8.103035520325e-07 ||r(i)||/||b|| 8.285356787434e-04
  8 KSP preconditioned resid norm 1.898415766718e-04 true resid norm
1.250723051310e-07 ||r(i)||/||b|| 1.278864778067e-04
  9 KSP preconditioned resid norm 3.325790135426e-05 true resid norm
2.304415060174e-08 ||r(i)||/||b|| 2.356265243067e-05
 10 KSP preconditioned resid norm 8.224698701048e-06 true resid norm
4.786148212120e-09 ||r(i)||/||b|| 4.893838299917e-06
 11 KSP preconditioned resid norm 1.224414108364e-06 true resid norm
7.170267325093e-10 ||r(i)||/||b|| 7.331600966164e-07
 12 KSP preconditioned resid norm 1.760480053719e-07 true resid norm
1.005681798001e-10 ||r(i)||/||b|| 1.028310006807e-07

So all is well.

Thanks for pointing this out.
Mark

On Thu, Jul 2, 2015 at 5:38 PM, Benoit Fabrèges <benoit.fabreges at inria.fr>
wrote:

>  Dear PETSc developers,
> I updated from 3.4 to 3.6 and I have a strange behavior with the GAMG
> preconditioner. I am running ex54 with the following options:
> -ne 1023 -ksp_rtol 1e-10 -ksp_monitor_true_residual -options_left
>
> I don't have the same behavior depending on the number of process used. It
> is for example not converging with 8 and 12 process with the reason
> KSP_DIVERGED_INDEFINITE_PC but it is working fine in other cases (I didn't
> try with more than 12 process).
>
> Do you have the same behavior ? It is working fine with 3.4 version.
>
> Best,
>
> Benoit
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20150708/4bf0ea18/attachment.html>