Firedrake meeting 2023 02 08

Date and time 2023-02-08 16:00UTC

Action Items

Pick Chair and Minuter (RNH to pick minuter)
ALL: (ongoing) triage the open issues and confirm if they are indeed still open (and perhaps provide labels)
JB: A Firedrake manual
JB: Move PyOP2 and FInAT to firedrakeproject
ALL: do things with SV's branches
KS: Fix checkpointing error with pickling elements

Present: CW (minuter), RK, KS, DH, JB, IM, DS, RNH, PB, NB, FA, PF

Apologies:

RK: Previously assumed a 1-to-1 relationship assumed between global and local basis functions. Makes zany, splines and other bits hard/impossible. This is solved by extraction operators. "Global basis supported on cell i - how do I represent this in a local basis?". Can be encoded as a sparse matrix.
DH: Would call this "packing" and "unpacking" operators. Need to be careful about data movement.
DH: Some complex cases of this include Serendipity or H-curl with 2 DoFs per face (can't be represented by permutations).
DH: This is a property of a function space. Could be computed eagerly or lazily.
DH: We assume that the element kernel operates in reference/canonical orientation. This means that zany/complex transformations would be happening outside of the element kernel.
DH: We are effectively proposing a system where for some elements the kernel does not lower to reference space (instead local space). Moves the pullback operation (and possibly other transformations) into the "pack", "unpack" space.
RK: Need to attach extraction operator to the function space not the finite element as some information belongs to the mesh.

Often CI is failing with:

Error: Process completed with exit code 137.

DH: Processes never return memory to the OS so tests consume a huge amount of memory by the end. Current plan is to mpiexec every single test. This guarantees that tests do not pollute the existing environment.