note to report issues upstream with mwe

[unalmis]

In #1522 and #1360 it was observed that the matrix give wrong results when the tensor's size is large. Both these operations are numerically stable in these cases. For now I fixed these issues by using a different approach and/or chunking the computation. This is a note to self to simplify the examples and report to upstream repositories.

[unalmis]

Here is one example. The test_harmonic_general test fails unless chunk_size is reduced.

DESC/tests/test_integrals.py

Line 762 in 869c81b

marks=pytest.mark.xfail(

The reason it fails is because the matrix multiplication in the line linked below is performed incorrectly when chunk_size is large.

DESC/desc/integrals/singularities.py

Line 504 in 869c81b

return jnp.real(vander @ f)