Skip to content

Thetis meeting agenda 2020 12 11

Jump to bottom
Tuomas Kärnä edited this page Jan 8, 2021 · 6 revisions

Date/time of meeting 2020-12-11 11:00 UTC (11:00 BST)

Previous meeting minutes

Minutes

Jenkins status

Testing on Jenkins is very flaky at the moment. Possible fixes discussed in firedrake PR1935. Proposed solution is two-fold: fix the exec permissions in mpicc etc files in firedrake/bin directory, and set global permissive umask in the container's /etc/bashrc file. Stephan takes a look.

Nonhydrostatic capability

2D version is introduced in PR225. Looks almost ready, Tuomas will review next week.

Thetis on the sphere

PR229 enables 2D shallow water equations on the sphere. Some notes:

  • Only works with bdm-dg and rt-dg element families. dg-dg could be made to work by creating a suitable 2D vector element which is forced on the tangent plane.
  • bdm-dg and rt-dg discretizations can be slow to run in practice. This is partially(?) due to the element pullbacks needed by the HDiv formulation. It may be possible to optimize the form by manually detecting cases where the pullbacks cancel out (UFL does not cancel them) Especially useful for linear terms.
  • A reasonable way to speed-up the HDiv spaces is to use better preconditioners. Slate and PCPATCH are relevant tools here. Point-of-Contact is Thomas Gibson.
  • On spherical geometry the elements are curved, so coordinate function with degree > 1 must be used. Which also adds to the cost of the model. It is unclear how much the degree affects computational cost, and what degree is needed in coastal applications where elements are typically small (and hence curvature is tiny).

Hybridized vertical diffusion

Vertical diffusion formulation could be improved, especially for P0 elements. Stephan has used hybridized diffusion in his non-hydrostatic code. Using the same approach in Thetis does not seem that straightforward, probably requires an additional aux variable.

Papers

  • With BDM elements and support for spherical geometry, we could and should write a complete 2D model description paper.
  • Earlier we've discussed a possible adjoint-modeling-for-ocean-problems review paper. Stephan will start to draft the outline.

Meander example pull request

Mariana's 2D sediment example PR228 is almost ready. Discussed the way zero boundary flux should be implemented in the Exner equation. Mariana will update the PR.

Date of next meeting

It was decided that we'll meet next on January 8th, and continue bi-weekly thereafter.

2021-01-08 11:00UTC (11:00BST)

Clone this wiki locally