[borisdevos]

Use the sector testsuite from TensorKitSectors.

• The sector tests run SectorTestSuite.test_sector for SU3Irrep, SU4Irrep, SU5Irrep and SU3Irrep ⊠ SU3Irrep; test/sectors.jl keeps only the SU(N)-specific checks.
• The F-, R-, A- and B-symbols are computed by specialising the *_from_fusiontensor methods of TensorKitSectors (avoiding views); SUN_Fsymbol and SUN_Rsymbol stay cached and call these.
• isone → isunit in SUN_CGC.

Artin braid / hexagon failures

The failures for higher-dimensional irreps were a bug in TensorKitSectors' hexagon_equation: the underbraiding used conj(R) instead of the adjoint, which only matters for non-symmetric R-symbols such as SU(3) with fusion multiplicities. This is fixed in TensorKitSectors v0.3.12 (QuantumKitHub/TensorKitSectors.jl#119), which is now the lower compat bound.

Keeping the tests within CI limits

The pentagon test scales steeply with the dimensions of the irreps, and the intermediate CGCs filled up the RAM caches (> 140 GB with the previous sets). Therefore:

• smallset is reduced to 5 SU(3) irreps of dim ≤ 11, 4 SU(4) irreps of dim ≤ 16, 3 SU(5) irreps of dim ≤ 16, and 5 products of dim ≤ 30 for SU(3) ⊠ SU(3). Each sector takes about 25–50 s and at most ~2.5 GiB.
• During the tests the RAM caches are bounded in bytes (1 GiB for the CGCs, 256 MiB for the F-symbols, 64 MiB for the R-symbols).
• The CGC disk-cache test now also empties the RAM cache first, since CGCs left in RAM by the sector tests would otherwise not be written to disk.
• The fusion tree tests draw their sectors from the 5 smallest nontrivial irreps, as before. The testsuite's randsector draws from the larger smallset above, and trees of five such irreps made these tests exceed the CI time limit.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines	Coverage Δ
src/clebschgordan.jl	97.89% <100.00%> (ø)
src/sector.jl	100.00% <100.00%> (ø)

... and 2 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[lkdvos]

When you say these tests fail, do you mean the tolerance is just set too high? Because that might be fair for higher-dimensional irreps

[borisdevos]

No, it can already mismatch at the 2nd decimal. Here a quick example:
isapprox([0.49357987913169243; 0.3608970552458627; 0.12256062712019984;;;;], [0.4796164601864595; 0.3801689443717772; 0.11970177000507069;;;;]; atol = 1.0e-12, rtol = 1.0e-12)
As far as I've seen it's also always the underbraiding path which fails. I'm curious if this also gets picked up in the fusion tree braid tests

[lkdvos]

possibly related to failures: #61