[CIG-SHORT] Intermediate Results

[Brad Aagaard]

On 3/11/13 9:42 AM, Matthew Knepley wrote:
> On Mon, Mar 11, 2013 at 12:33 PM, <BOK10 at pitt.edu
> <mailto:BOK10 at pitt.edu>> wrote:
>
>     So, I just checked and both the linear and nonlinear solutions are
>     converging. I'm not sure I know what you mean by different solution
>     settings? Do you mean the following:
>
>     # Preconditioner settings.
>     pc_type = asm
>     sub_pc_factor_shift_type = nonzero
>
>     # Convergence parameters.
>     ksp_rtol = 1.0e-20
>
>
> These kinds of tolerances usually mean that something is not scaled
> right. You
> cannot really get meaningful information below 1.0e-12.
>
> Brad, does this have something to do with the friction solve?

When using friction, we want the KSP and SNES rtol to be really small to 
force the absolute tolerance to be used so PyLith can differentiate 
between roundoff errors and initiation of slip. The relative sizes of 
the absolute tolerances seems okay. It should be fine to increase the 
absolute tolerances.

>     ksp_atol = 1.0e-13
>     ksp_max_it = 1000000
>     ksp_gmres_restart = 50
>
>
> This restart is too small. If you have more memory, increase it to 100-200.

In many of our benchmarks convergence is fastest with a restart between 
50 and 100.

>     # Linear solver monitoring options.
>     ksp_monitor = true
>     ksp_view = true
>     ksp_converged_reason = true
>
>     # Nonlinear solver monitoring options.
>     snes_rtol = 1.0e-20
>
>
> Again, this seems way too small.
We want a really small rtol with friction as mentioned above.

 > Running both simulations, I got wildly different results.

This doesn't tell us much at all. What is different (be as specific as 
you can)? When do the differences appear? Are the differences localized 
in space?

Brad