Computational Infrastructure for Geodynamics Wiki

Rajesh reported on the CitcomS benchmarking progress. The “A1” benchmark is running on Stampede, and preliminary results are expected in the next couple of days.

Nicola expressed an interest in timing the Gaia code on Stampede. This code has already been benchmarked. The question is should we create an account on Stampede so that Nicola can run this code himself, or will he provide the code to us (CIG staff) and we run the code on Stampede.

It was suggested that a mantle convection focussed meeting of Citcom and Aspect users should be organized, perhaps in Davis before the AGU Fall meeting. We will discuss this in further detail at a later date

The following CitcomS work plan was reiterated:

Rajesh reported on the Cookbook 4 “solver not converging” warnings. The problem has been identified and further testing is in progress.

CitcomS manual is being updated. It was agreed that adding a 'Multigrid HOWTO' section would be a good idea. Scott,Margarette and Katrina agreed to provide config files, MATLAB scripts, etc. to faciliate this

Identify and fix the source of the warning in cookbook 4.

Update manual to include the Multigrid versions of the cookbooks and the benchmarks.

Validating CitcomS 3.2 using cookbooks and benchmarks.

Rajesh reported on the conversion of cookbooks 1-8 using the Python to C configuration file conversion tool, and the testing of cookbooks 1-4. The conversion tool and the converted config files have been checked into new CitcomS branch, called “python_removal”

Cookbook 4 has several “solver not converging” warnings and Scott suggested running this cookbook with a uniform mesh to check if the warnings are eliminated.

We disucssed if only the multigrid versions of the cookbooks should be presented in the manual. While this would be useful in introducing the concepts of multigrid to new Citcom users, there are many subtleties with setting up a multigrid problems which need to be carefully presented.

Scott suggested that after adding topography to Aspect, we need to check if topography results can be improved by higher order elements.

Margarete brought up the technique of “manufactured solutions” and how they can be used to verify code that can't be tested analytically. She provided the following papers on this subject

Shijie noted that we need to be careful regarding the initial condition spherical harmonics for CitcomS

Identify and fix the source of the warning in cookbook 4 [Rajesh]

Update the manual to refer to the new C based versions of the config files. Eliminate all references to Python from the CitcomS manual [Rajesh]

Conver the benchmark configuration files to use multigrid, rather than conjugate gradient. [Rajesh]

Continue work on the spherical convection benchmarks for CitcomS and Aspect.

The priority list for CitcomS was discussed. Specifically, the status of the energy solver was discussed. To this end we discussed Entropy Viscosity method, Discontinuous Galerkin methods, Finite Volume methods, etc. The issue of cleanup of tracer code in CitcomS was also discussed. It was agreed that the initial focus should be on improving the Stokes solver by incorporating PETSc into Citcom, and eliminating Python dependencies from CitcomS

Spherical convection benchmarks for Citcom and Aspect were discussed. It was agreed that the initial focus should be on the benchmarks from Zhong et. al (2008) paper. The target for the initial benchmarking is the Banff meeting in early May.

The C-based version of CitcomS, and the tool for converting Python-based .cfg files to C-based files, need to output the full parameter list, indicating if each parameter has the default value, or if the value has been overriden via the config file. [Rajesh]

Test the consistency of the cookbooks in CitcomS. This will be accomplished by comparing the results of running the cookbooks under CitcomS 3.1 and 3.2 [Rajesh]

Start collecting the materials needed for benchmarking CitcomS and Aspect. Eric will setup a Google doc to start gathering the information about spherical convection benchmarks for Citcom and Aspect. [Eric]