[CIG-SHORT] Speed up nonlinear solver for faults with friction

[Brad Aagaard]

Jiangzhi,

You need to investigate what the linear and nonlinear solvers are doing. 
You should turn on the residual and convergence flags:

[pylithapp.petsc]
ksp_monitor = true
ksp_converged_reason = true
ksp_error_if_not_converged = true

snes_monitor = true
snes_linesearch_monitor = true
snes_converged_reason = true
snes_error_if_not_converged = true

This will help you understand if the linear solver is converging, and 
then what the nonlinear solver is doing. You want to see the residuals 
decreasing with each solver converging.

You may want to review Step03 in the debugging tutorial from the June 
workshop as well as the friction tutorial. 
http://geodynamics.org/cig/events/calendar/2014-cdm-workshop/meeting-info/agenda/
See PyLith: Debugging simulation problems (Step03) and PyLith: 
Introduction to fault friction.

Regards,
Brad

On 12/22/14, 1:52 PM, Jiangzhi Chen wrote:
> Hi Brad,
>
>     Is there some way to speed up the nonlinear solver when calculating
> the interseismic fault deformation with friction on the fault? In my
> current simulation, when I use kinematic constraints on faults, the
> whole calculation takes about 10 hours, but if I use friction instead of
> fault slip, the calculation seems to run forever (it has been running on
> server during the whole AGU week but not even the first time step is
> completed). I attached the configuration file.
>
> cheers,
> Jiangzhi
>
>
> _______________________________________________
> CIG-SHORT mailing list
> CIG-SHORT at geodynamics.org
> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>