Qiskit / qiskit-fermions / 33873559976

98%

push

github

fix(circuit): terminate the Evolution decomposition at atomic factors (#312)

The decomposition of an `Evolution` gate had no base case. A gate holding a single
operator term decomposed into an identical single-term `Evolution`, so expanding it
again made no progress. `inverse()` recurses through the definition, so it hit the
interpreter's recursion limit instead of returning.

The same missing base case had a second, quieter consequence. Splitting an operator
that carries groups descended past the groups and split each one term by term. An
individual term is generally not Hermitian even when the group containing it is --
a UCC cluster generator holds conjugate pairs precisely so that each *group* is
Hermitian -- and the exponential of a non-Hermitian factor is not unitary. The
fermion-to-qubit stage rejected such factors with a complaint about complex
coefficients, but `expm_multiply` applied them without objection, returning a state
vector of norm ~1.8 with nothing raised anywhere.

Both follow from the same thing: a factor a synthesis method emits is finished.
Splitting it again either makes no progress or discards the very partition that was
chosen. Marking those factors atomic terminates the recursion and, with it, stops
the group-splitting that manufactured the non-Hermitian factors.

An atomic gate deliberately leaves its definition unset rather than assigning an
empty circuit: an empty definition makes `decompose()` *drop* the evolution rather
than keep it, which would silently replace it with the identity.

`inverse()` is also implemented directly now, negating the time and preserving the
operator and synthesis method. The inherited version rebuilt the gate from its
definition and returned a plain `Gate`, losing both.

Refs #298

Co-authored-by: Claude Opus 5 <noreply@anthropic.com>

8 of 8 new or added lines in 1 file covered. (100.0%)

13455 of 13702 relevant lines covered (98.2%)

481400.67 hits per line