Keep the given order of Matsubara sampling points

[shinaoka]

Summary

MatsubaraSampling with custom points given in any order but ascending returned evaluate results and expected fit data in a different order from sampling_points. Results were silently wrong for the caller's data order.

The C library orders Matsubara points ascending internally, while the wrapper kept sampling_points in the caller's order.

This follows up #86, found while fixing the tutorial notebooks.

Changes

• Order. The wrapper hands the C library sorted points and permutes the data between the two orders in evaluate and fit, on any axis. TauSampling already follows the given order since Enrich the tests, fix the wrapper defects they expose, and document the conventions #86.
  • Default points are sorted already and take no extra work.
  • cond is unchanged, since a row permutation does not change it.
• Docstrings.
  • MatsubaraSampling states that points may be given in any order.
  • The unfolded default τ points were said to be symmetric about 0. They come in pairs ±τ, plus β/2 when their number is odd. So reversing them maps τ to −τ only for an even number, while reversing the folded default points maps τ to β − τ in every case.

Tests

tests/test_boundary.py checks that shuffled default points give evaluate equal to gl @ basis.uhat(points) and a fit round trip, along axis 0 and axis 1. It covers:

• fermions and bosons, with and without positive_only;
• an augmented basis;
• a DLR.

All five cases failed before the fix. The full suite gives 350 passed, 5 xfailed (was 345 passed, 5 xfailed).

🤖 Generated with Claude Code