[CIG-LONG] Cumulative strain history in Gale et alia

[Walter Landry]

"Bradford H. Hager" <bhhager at mit.edu> wrote:
> Christoph and others-
> 
> I'd also be interested in any benchmarking, as well as learning some
> "tricks of the trade" running GALE.
> 
> I am running some simple wedge models with viscous and Coulomb
> rheologies.  I find that viscous problems run quickly, and problems
> where the deformation is all by plastic deformation converge, but
> take a lot longer.  However, problems where I'd expect the
> deformation to be mixed viscous and plastic do not converge before
> my patience expires, contrary to my expectation that run times
> should be intermediate between the two extremes.
> 
> Also, increasing the tolerance by a factor of 3 can also increase
> the run time by a lot.

There was a bug in MohrCoulomb that I fixed a week and a half ago that
fixed a lot of these kinds of problems.  The input file format has not
changed, so you should be able to try it out with your existing
models.  I plan to make a release in October once I get all of the
benchmarking stuff done.

> Can anyone point me to an explanation of what effects run times?

Everything ;) But the main culprits are detailed in Section 4.3 of the
manual: linear and non-linear tolerance, resolution, particle number,
and time scaling factor.

Also, for serial runs, direct solvers tend to be more predictable.
Section 4.1 of the manual has the syntax

  ./Gale-1_2_1 input/extension.xml -pc_type lu -ksp_type preonly

> Thanks,
> 
> Brad Hager
> 
> On Oct 5, 2007, at 3:10 PM, Christoph Schrank wrote:
> 
> > 3) I would be curious to know what is going on in terms of
> > benchmarking the code. 

I have just finished implementing everything needed to run the Geomod
benchmarks.  I am still in the process of tweaking the model, but if
you check the code out of the repository, the file

  input/benchmarks/extension.xml

should be a good approximation of the extension model.  You may have
to increase the resolution, but it looks much like the other examples
in Buiter's paper.

> > Will there be any presentations on the next Geomod meeting?

That has not been decided yet.  We are certainly keeping an eye on it.

> > Or would other users perhaps be interested in sharing their
> > experiences? I started to run a few free-falling slab models,
> > sandwedge shortening experiments, Stokes flow tests, and RT
> > instabilities.

I would certainly be interested in your experiences.  Good and bad
experiences are both appreciated.

Cheers,
Walter Landry
walter at geodynamics.org