[CIG-SHORT] PyLith Convergence issue - DIVERGED_FUNCTION_COUNT and CONVERGED_SNORM_RELATIVE

[Brad Aagaard]

Josimar,

Please send the pylith_parameters.json file that PyLith generates.

Thanks,
Brad


On 06/22/2017 02:19 PM, Josimar Alves da Silva wrote:
> Matt,
> 
> Thanks for the help. Here is the log file with more information: 
> https://www.dropbox.com/s/qnpk055o9xg38j1/log.out?dl=0
> 
> Let me know if you need anything else.
> 
> thank you,
> Josimar
> 
> On Thu, Jun 22, 2017 at 3:21 PM, Matthew Knepley <knepley at rice.edu 
> <mailto:knepley at rice.edu>> wrote:
> 
>     On Thu, Jun 22, 2017 at 12:10 PM, Josimar Alves da Silva
>     <jsilva.mit at gmail.com <mailto:jsilva.mit at gmail.com>> wrote:
> 
>         Matt,
> 
>         Thank you so much for the help. I followed up on your suggestions:
> 
>         1) Using snes_max_funcs=10000 did not help. The simulation does
>         not go beyond iteration #5000. It stopped at the same time step
>         as the previous one.
> 
> 
>     Note this has to be in the [petsc] section. Please add snes_view.
>     Then we can see exactly what the setting were.
>     It would also be good to add snes_converged_reason
> 
>     I looked at the log. It makes as much as progress as it will ever
>     make after 2 iterates, thus I think there
>     is a problem with interplay of tolerances here.
> 
>         2) Can you comment a little bit on how can I improve my initial
>         guess, as you mentioned here : "if your initial guess is not
>         good enough they can fail". What you mean about it ?
> 
> 
>     I mean if you are taking time steps so big that the last solution is
>     not close to the next one, the solves will be harder.
> 
>        Matt
> 
>         The log file is attached.
> 
>         thank you so much in advance,
>         Josimar
> 
> 
> 
> 
>         On Thu, Jun 22, 2017 at 11:33 AM, Matthew Knepley
>         <knepley at rice.edu <mailto:knepley at rice.edu>> wrote:
> 
>             On Wed, Jun 21, 2017 at 6:25 PM, Josimar Alves da Silva
>             <jsilva.mit at gmail.com <mailto:jsilva.mit at gmail.com>> wrote:
> 
>                 Dear Brad, Matt and Charles,
> 
>                 I am working on a 2-D quasi-static, elastic, simulation
>                 using rate and state friction model.
> 
>                 The simulation runs fine for many time steps, then at
>                 some point I get the error below related to
>                 DIVERGED_FUNCTION_COUNT.
> 
>                 Nonlinear solve did not converge due to
>                 DIVERGED_FUNCTION_COUNT iterations 5000
> 
>                 All the log files, along with the .cfg files, model
>                 geometry and boundary conditions as well as a
>                 convergence plot for the SNES residual can be found on
>                 the attached .zip file.
> 
>                 I have some questions that I would like to have your input:
> 
>                 1) Is it possible to extend the maximum value for
>                 DIVERGED_FUNCTION_COUNT  ? Note that I have snes_max_it
>                 =10000  but SNES still stops at 5000 value.
> 
> 
>             snes_max_funcs 10000
> 
>                 2) Looking at the convergence plots that I have
>                 attached, I noticed that the time step that fails
>                 contains iterations that actually increase the
>                 residual,instead of decreasing it. Would you know what
>                 is causing this behavior ?
> 
> 
>             There is no guarantee of decrease with a general nonlinear
>             problem.
> 
>                 3) There are several time steps that converge with
>                 tolerances smaller than what I ask for using the
>                 snes_atol (see figure attached). At these time steps I
>                 get the following
> 
>                 Line search: Aborted due to ynorm < stol*xnorm
>                 (9.933794721748e-10 < 1.046238089915e-09) and inadequate
>                 full step.
> 
>                 Nonlinear solve converged due to
>                 CONVERGED_SNORM_RELATIVE iterations 694
> 
> 
>             This means that even a very short step length gave no
>             residual decrease, also known as "stagnation". Here is the
>             nonlinear
>             solver is just not working.
> 
>             We have no convergence theory for the friction problems, and
>             if your initial guess is not good enough they can fail. We
>             want to explore
>             more powerful solvers for these, but after we get a revised
>             formulation in place.
> 
>               Thanks,
> 
>                  Matt
> 
>                 Note that I am using snes_error_if_not_converged =true.
> 
>                 Could you please comment on why this is happening and
>                 how I would fix this issue ? Does it affect the next
>                 time steps after it ?
> 
>                 Let me know if you need any further information.
> 
>                 Thank you so much for you help in advance,
>                 Best
>                 Josimar
> 
> 
> 
>                 _______________________________________________
>                 CIG-SHORT mailing list
>                 CIG-SHORT at geodynamics.org <mailto:CIG-SHORT at geodynamics.org>
>                 http://lists.geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>                 <http://lists.geodynamics.org/cgi-bin/mailman/listinfo/cig-short>
> 
> 
> 
>             _______________________________________________
>             CIG-SHORT mailing list
>             CIG-SHORT at geodynamics.org <mailto:CIG-SHORT at geodynamics.org>
>             http://lists.geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>             <http://lists.geodynamics.org/cgi-bin/mailman/listinfo/cig-short>
> 
> 
> 
>         _______________________________________________
>         CIG-SHORT mailing list
>         CIG-SHORT at geodynamics.org <mailto:CIG-SHORT at geodynamics.org>
>         http://lists.geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>         <http://lists.geodynamics.org/cgi-bin/mailman/listinfo/cig-short>
> 
> 
>