[rogeriojorge]

Round each coil pair’s absolute Gauss integral to its integer linking number, matching SIMSOPT. Stop gradients through this topological diagnostic and mask padded quadrature entries without changing the physical denominator. Optional downsampling reduces pairwise work.

Tests cover Hopf links, unlinked circles, orientation, symmetry copies, uneven quadrature blocks, and zero gradients. Mandatory NumPy reference tests and optional SIMSOPT parity tests verify each pair and the aggregate. Linking number remains undefined for intersecting curves.

Recorded validation: reference/SIMSOPT parity and the 99-test suite passed (one expected failure). Python 3.10–3.12 CI passes. No production dependency is added.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines	Coverage Δ
essos/objective_functions.py	99.08% <100.00%> (+0.33%)	⬆️

... and 1 file with indirect coverage changes

[rogeriojorge]

Added SIMSOPT-compatible downsample support in 0fd8379. The integration step is scaled by the factor, invalid/non-divisible values are rejected, and the existing blockwise path operates on the reduced point set. Tests compare raw per-pair integrals and exact integer classifications against both the NumPy translation of SIMSOPT C++ and installed SIMSOPT for factors 1, 2, 3, 4, 5, and 6, covering expanded unlinked circles and a Hopf link. Full local suite: 99 passed, 1 xfailed. Benchmark for 56 expanded coils / 80 source points: 6.61 ms at downsample=1, 1.33 ms at 2, and 0.45 ms at 8; all report linking number 0.

[eduardolneto]

The changes to add restructure and add extra compatibility with simsopt equivalent calculations are good. However, should we find a way to get an alternative loss function that is differentiable?