Qiskit / qiskit-fermions / 30749101080

98%

Build Type

push

github

Committed by web-flow

Commit Message

feat(operators): contract generators in Edge/Transfer normal_ordered (#257)

`normal_ordered` only ever *reordered* generators, so its result was not a
canonical form: `V_j V_j`, `E_jk E_kj` and friends survived as-is. Two
representations of one operator could therefore normal-order to different terms,
and `is_hermitian` -- which tests `self - self.adjoint()` against zero --
reported false negatives whenever that cancellation needed a contraction first.

Both families now contract to a fixed point, gated behind a `reduce` flag that
defaults to `true`, mirroring `MajoranaOperator::normal_ordered(reduce=true)`.
`FermionOperator` deliberately gains no such flag: there the
`a_i a^dag_j = delta_ij - a^dag_j a_i` contraction is what *permits* each swap
rather than a separable cleanup pass, so disabling it would return an operator
not equal to the input.

The rules were derived from dense matrices built straight from the Majorana
definitions rather than from Eq. (5)/(7) of arXiv:2512.11418v1, which are stated
only for pairwise-distinct indices and so say nothing about the coinciding-index
cases that contraction is entirely about -- the same blind spot that produced the
two preceding sign bugs. That oracle is committed alongside as
`majorana_matrix_oracle.py` so the derivation stays reproducible.

Edge additionally gains `ascending` to fix the `E_kj = -E_jk` orientation
convention, closing the TODO that asked for exactly this. It is not merely
cosmetic: recognizing `E_10 E_21 -> E_20` requires matching on indices
independently of which orientation happens to be stored, so canonicalizing is
what makes fusion findable at all. Both directions fuse identically.

The two families end up deliberately asymmetric, because the algebra is:

- Edge fuses. `E_ab E_bc = -i E_ac` collapses two generators into one, and
  together with `E_kj = -E_jk` that single rule generates all 96 shared-mode
  products.
- Transfer does not. Writing `T_jk = (i/2) gamma_2j gamma_2k-1`, can... (continued)

Coverage Stats

253 of 266 new or added lines in 2 files covered. (95.11%)

125 existing lines in 5 files now uncovered.

11940 of 12181 relevant lines covered (98.02%)

276.44 hits per line