[cig-commits] [commit] master: Increase limit for the number of inner iterations. (1acb52a)

[cig_noreply at geodynamics.org]

Repository : https://github.com/geodynamics/aspect

On branch  : master
Link       : https://github.com/geodynamics/aspect/compare/36e4420c87fda8026b8a3d1f639e798f6c59f3d0...01f19ef36177d8c77f84b59ece4918ba5f3e7661

>---------------------------------------------------------------

commit 1acb52ab7eda8192611b8c256c65088c45910096
Author: Wolfgang Bangerth <bangerth at math.tamu.edu>
Date:   Mon Dec 1 11:06:27 2014 -0600

    Increase limit for the number of inner iterations.
    
    This is apparently necessary for some very ill-conditioned models. Specifically,
    I have been running a model by Sarah Stamps with very large variations of the
    viscosity, and it takes more than 2000 inner iterations to converge in this
    place. Maybe not surprisingly, the run time profile looks rather unfamiliar:
    
    +---------------------------------------------+------------+------------+
    | Total wallclock time elapsed since start    |  4.37e+05s |            |
    |                                             |            |            |
    | Section                         | no. calls |  wall time | % of total |
    +---------------------------------+-----------+------------+------------+
    | Assemble Stokes system          |         2 |      82.4s |     0.019% |
    | Assemble temperature system     |         1 |      44.5s |      0.01% |
    | Build Stokes preconditioner     |         1 |      97.3s |     0.022% |
    | Build temperature preconditioner|         1 |      2.77s |   0.00063% |
    | Solve Stokes system             |         2 |  4.35e+05s |     1e+02% |
    | Solve temperature system        |         1 |      6.15s |    0.0014% |
    | Initialization                  |         2 |        18s |    0.0041% |
    | Postprocessing                  |         1 |  1.71e+03s |      0.39% |
    | Setup dof systems               |         1 |      60.4s |     0.014% |
    +---------------------------------+-----------+------------+------------+
    
    Note that we spend 99.5% of the time in the Stokes solver.


>---------------------------------------------------------------

1acb52ab7eda8192611b8c256c65088c45910096
 source/simulator/solver.cc | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/source/simulator/solver.cc b/source/simulator/solver.cc
index 5d66526..0c3cc90 100644
--- a/source/simulator/solver.cc
+++ b/source/simulator/solver.cc
@@ -320,7 +320,7 @@ namespace aspect
       // iterations of our two-stage outer GMRES iteration)
       if (do_solve_A == true)
         {
-          SolverControl solver_control(1000, utmp.l2_norm()*1e-2);
+          SolverControl solver_control(10000, utmp.l2_norm()*1e-2);
 #ifdef ASPECT_USE_PETSC
           SolverCG<LinearAlgebra::Vector> solver(solver_control);
 #else