[CIG-LONG] Gale: Problems with benchmarks

[Walter Landry]

Hi Edie,

Cc'ing the list.


Edie Miglio <edie.miglio at polimi.it> wrote:
> Dear Walter,
>    we are starting using a yielding criteria in Gale.
>    In order to develop some feeling in using such a
> rheology we have tried to run the benchmarks: in
> particular we have tried geomod_2004_extension.xml
> and geomod_2004_shortening.xml with Gale version
> 1.6.2.

1.6.2 is an older version.  You should upgrade to 2.0.0.  However, I
have not done all of the geomod benchmarks with 2.0.0, only the
extension benchmark.

>   In the extension simulation 5000 time steps are
> set in the xml file but at iteration 78 the code does not
> converge anymore.
>   As for the shortening we have non-convergence since
> the very first iteration.
>   We have also tried to use Von Mises criterium (we attach
> the xml file along with the error file) but we are even not able to
> start the simulation and we get a segmentation fault error.
>   Do you have some suggestions concerning the optimal
> values of minimumViscosity and maxStrainRate (related to
> the DrugerPrager rheology) ? We have a lot of problems
> running all the benchmarks with this kind of rheology.

The benchmarks should work with the input files as given.  You should
only have to adjust the resolution to get the exact benchmark results.

However, when you run it, you need to use a direct solver.  Section
4.2.2 of the Gale 2.0.0 manual covers how to do that.

>   Finally we have observed that the amount of memory
> required increases during the run and for long time
> simulation it is a severe problem since we are not able to
> reach the end. We fear that there is some memory leak
> somewhere in the code.

That does sound like a problem.  Does it happen in 2.0.0?

Edie Miglio <edie.miglio at polimi.it> wrote:
>   I forgot to ask you which are the differences between Gale and
> Underworld.

The biggest differences are that Gale can use Q2 P-1 elements.  I
believe that Underworld defaults to Q1 P0 elements, which have
numerical issues especially when non-linear rheologies come into play.
Gale and Underworld can both use Q1-Q1 elements, but those have a
number of difficulties with realistic geologic models.

Gale can use JSON as an input format.  Gale incorporates an equation
parser making it easy to specify initial and boundary conditions.
There are other minor changes.  If you look at the file RELEASE_NOTES,
all of the changes in 2.0.0 are specific to Gale.

Cheers,
Walter Landry