Hackathon 2024 and ASPECT 3.0 changes

Dear all,
Thanks again for all the great contributions during this years hackathon! I enjoyed our time in Colorado a lot and I think we made a lot of progress and also a lot of new friends!

As part of the changes during and after the hackathon you may notice that some input parameters have changed and will continue to change until we release the next version of ASPECT later this year. I just wanted to remind everyone that we keep a script around in contrib/utilities/update_prm_files.sh that exists to update ASPECT input files to the latest version (use of the script is explained here. We try to keep this script up-to-date so that if you encounter problems with your input files after a change in ASPECT you can fix most (if not all) of them by running the update script.

That being said I wanted to summarize two changes that will probably affect many input files and model results:

  • ASPECT’s default Stokes solver has changed to the matrix-free GMG-preconditioned Stokes solver described in https://doi.org/10.1002/nla.2375. This change should generally result in faster model runtimes as long as you have compiled deal.II and ASPECT with the native optimizations as described here. The solver should also show more robust convergence behavior, but for some problems it may behave worse. If you encounter convergence issues with your models after updating ASPECT, you can always switch back to the previous default solver by setting the parameter set Stokes solver type = block AMG in the subsection Solver parameters -> Stokes solver parameters. Because the GMG-preconditioned Stokes solver does not support all model features yet, ASPECT’s new default behavior is to select the GMG-solver whenever possible, and fall back to our previous AMG-preconditioned matrix based solver whenever necessary.
  • If you are using particles in your computations be aware that most of the particle parameters have moved from the subsection Postprocess -> Particles to the new global subsection Particles. The subsection Postprocess -> Particles now only contains the parameters that are directly related to the output of particles (such as output format, or interval between output files).

It is possible that more things change until our next release, so if you like to avoid updating your parameter files I would suggest to wait for a bit until things stabilize before our next release this fall. We try to keep parameter files compatible with newer versions of ASPECT as much as possible, but it is time to clean up some things that have accumulated since ASPECT 2.0 and this release cycle will see some bigger changes than usual.

Thank you all for your understanding and continued support of ASPECT.


1 Like