Firedrake meeting 2020 07 07

Date and time 2020-07-07 15:00UTC (16:00BST)

Action Items

1. Choose someone to minute and chair
2. ALL: (ongoing) triage the open issues and confirm if they are indeed still open (and perhaps provide labels)
3. KS, (DH, LM): Document describing what we think the mixed domain interface should look like (and hence what is needed in UFL, and whether it matches the existing Fenics efforts). Try an alternative description and make previously agreed changes.
4. **: Think about the correct mathematical formulation for Filtered
5. ALL: Please review complex.
6. **: Add --remove-build-files to make install smaller; convert this to an issue #1771

Minutes

Present: Reuben Hill, Lawrence Mitchell, David Ham, Koki Sagiyama, Paul Kelly, Robert Kirby, Jack Betteridge, Matthew Kan, Mohammad Usman, Nacime Bouziani, Sofia Simola, Sophia Vorderwuelbecke

Apologies: Dan Shapero

RCK: discuss FIAT progress for dual spaces, what to do in FInAT dual evaluation?

(Note: busy discussion - below notes may contain inaccuracies)

Element Oracle

• Problem: Practical implementation has moved well beyond initial design constraints of FIAT and elements are all over the place and complicated interdependencies. Lots of pullbacks in UFL which suggest leaky interface.
• Proposed Solution: Create an element oracle that would replace FIAT, FinAT, element parts of UFL, and some bits of TSFC. (Rob Kirby can expand on this)
• Raised issue: UFL needs symbolic pullbacks e.g. for shape derivatives.
  • Response: Have element oracle return legal UFL as result of a pullback. Make element oracle a registered package to avoid circular dependencies (David Ham can expand on this).
  • Alternative: Have coordinate derivative callback in API of element oracle.
• Making this happen:
  • David Ham will have an UG student able to put in 8 weeks work on element oracle from the end of July.
  • Lawrence Mitchell suggests straw-manning the interface to get it right before putting in serious programming work.

Dual Spaces in FIAT

• Rob Kirby has worked on adding dual space representations to FIAT with new concept of point dual bases from which dual functions can be represented as linear sums. (Rob Kirby can expand on this)
• Could make certain interpolations work out of the box which currently do not.
• Rob Kirby to make PR.

LM: reworking MG interpolation

• Want to make use of new technology for FIAT.
• Lawrence open to suggestions on how to move forwards

LM: Vectorisation status

• Only 1 kernel test failing at time of writing. Close to solving.

JB: SIAM-CSE

Thoughts on running minisyposia as well as minisymposterium? Chased up Chris Richardson and he's very keen to contribute for possible Exascale simulation themed mini. (Also happy to contribute a poster).

• Suggested that JB go ahead with Exascale simulation themed minisymposium.
  • DH will provide names of who to contact

JB: Randomgen

Is Firedrake currently using any features of randomgen that aren't now in numpy?

• Apparently parallel safe RNG has been ported to numpy.
• Aim to try and drop the dependency if possible.

KS: ufl.Filtered/ufl.Transformed -> ufl.Masked

Status: https://github.com/firedrakeproject/firedrake/issues/1648 (see main.pdf in edited issue description)

Summary:

0. Viewing ufl.Transformed as a map is awkward.

1. Introduce V0 = ufl.Subspace(V, some_constraint)

2. Introduce v0 = ufl.Masked(v, V0) (v0 = v if v in V0 else 0)

• After reviewing main.pdf:
  • Still have issues with concepts being basis-dependent which is not allowed in UFL.
  • Make concepts like FunctionSpaceModifier non-basis-dependent by making them "masks".
  • More design work needed.

DH Absense

DH will be away for several of the next firedrake meetings.

Date of next meeting

2020-07-14 15:00UTC (16:00BST)