Generation of divergence and gradient forms fails

Hi,

I tried to regenerate the forms for divergence/gradient for the P1-P0 combination, but this does not (or no longer?) work. Running e.g.

python3 generate_all_hyteg_forms.py -f p0_to_p1 -o FOO

gives me

[MainProcess] — [2025-02-26 18:19:30,391] — [     generate_all_hyteg_forms.py] — [line 1029] - [    INFO] ################################
[MainProcess] — [2025-02-26 18:19:30,392] — [     generate_all_hyteg_forms.py] — [line 1030] - [    INFO] ### HyTeG Operator Generator ###
[MainProcess] — [2025-02-26 18:19:30,392] — [     generate_all_hyteg_forms.py] — [line 1031] - [    INFO] ################################
[MainProcess] — [2025-02-26 18:19:30,392] — [     generate_all_hyteg_forms.py] — [line 1064] - [    INFO] - selected geometries: triangle, tetrahedron
[MainProcess] — [2025-02-26 18:19:30,392] — [     generate_all_hyteg_forms.py] — [line 1067] - [    INFO] Generating 2 forms:
[MainProcess] — [2025-02-26 18:19:30,392] — [     generate_all_hyteg_forms.py] — [line 1071] - [    INFO] - p0_to_p1_divt_affine_q0
[MainProcess] — [2025-02-26 18:19:30,392] — [     generate_all_hyteg_forms.py] — [line 1071] - [    INFO] - p0_to_p1_divt_blending_q0
[MainProcess] — [2025-02-26 18:19:30,392] — [     generate_all_hyteg_forms.py] — [line 1073] - [    INFO] Patience you must have young padawan
[MainProcess] — [2025-02-26 18:19:30,392] — [     generate_all_hyteg_forms.py] — [line 1075] - [    INFO] p0_to_p1_divt_affine_q0, trial: (Lagrange, degree 0), test: (Lagrange, degree 1), dimension: 3 x 1, qua\
drature schemes/degree (dimension): {2: 0, 3: 0}, blending: IdentityMap
[MainProcess] — [2025-02-26 18:19:30,392] — [     generate_all_hyteg_forms.py] — [line 1099] - [    INFO]   - Generating code for class p0_to_p1_divt_0_affine_q0, 2D ...
[MainProcess] — [2025-02-26 18:19:30,484] — [     generate_all_hyteg_forms.py] — [line 1099] - [    INFO]   - Generating code for class p0_to_p1_divt_0_affine_q0, 2D ... done (took 0.09 s)
Traceback (most recent call last):
  File "[...]/hog/generate_all_hyteg_forms.py", line 1181, in <module>
    main()   
  File "[...]/hog/generate_all_hyteg_forms.py", line 1111, in main
    mat = form_func(
          ^^^^^^^^^^
  File "[...]/hog/generate_all_hyteg_forms.py", line 822, in form_func
    return gradient(
           ^^^^^^^^^
  File "[...]/hog/hog/forms.py", line 615, in gradient
    return process_integrand(
           ^^^^^^^^^^^^^^^^^^
  File "[...]/hog/hog/integrand.py", line 494, in process_integrand
    mat[data.row, data.col] = integrand(**asdict(s))
                              ^^^^^^^^^^^^^^^^^^^^^^
  File "[...]/hog/hog/recipes/integrands/volume/gradient.py", line 32, in integrand
    -(jac_b_inv.T * tabulate(jac_a_inv.T * grad_v)).trace()
     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "[...]/python-for-hog/lib/python3.11/site-packages/sympy/matrices/common.py", line 2443, in trace
    raise NonSquareMatrixError()

The problem actually seems to affect all gradient and divergence forms.

added bug user experience labels

assigned to @pponkumar

mentioned in merge request !41 (merged)

I am not sure if the form generation was planned to be supported since operator generation came along. AFAIK it is not even tested in the pipeline.

You are right that we moved away from forms to full operator generation due to the enhanced possibilities for performance optimisation. However, some thoughts on this:

• I still find being able to generate forms helpful, since they are easier to verify. Additionally the underlying functionality is also needed for operator generation.
• There are (at least in main) no operators for a P1-P0 pairing.
• It is not looking good to users, when you offer functionality that does not work.

Since I am working on the

P_2^+-P_1^\text{disc}

element, I was curious to see how the P1-P0 pairing is implemented. But, obviously, there is an issue with this at the moment.

changed title from Generation of mixed P1-P0 forms fails to Generation of svi and divt forms fails

changed the description

changed title from Generation of svi and divt forms fails to Generation of dvi and divt forms fails

I updated the issue, as the problem is not specific to the mixed P0-P1 form, but seems to affect all gradient/divergence forms.

changed title from Generation of dvi and divt forms fails to Generation of divergence and gradient forms fails

created branch 48-generation-of-divergence-and-gradient-forms-fails to address this issue

mentioned in merge request !42 (merged)

mentioned in commit 34726b52

closed via merge request !42 (merged)