Axis-aware S³ compression: emit rx/ry/rz for Cartesian-aligned fused rotations

[Copilot]

Single-qubit gate runs that fuse to a Cartesian-axis rotation were emitting a generic U gate. They should emit a named rx/ry/rz — lower semantic complexity, directly hardware-aligned (notably, rz is virtual/"free" on IBM backends).

What changed

geometry.py — axis_aligned_rotation

Pure-math helper operating on the S³ quaternion. Returns ("x"/"y"/"z", theta) if the rotation axis is within tolerance of a Cartesian unit vector; None for generic or near-identity rotations.

qiskit_adapter.py — emit_axis_aligned_gate + two-pass build_optimized_circuit

emit_axis_aligned_gate converts the fused SU(2) matrix to its quaternion, runs the axis check, and appends rx/ry/rz (returns 1) or does nothing (returns 0).

build_optimized_circuit now uses a two-pass approach per fused run:

1. Axis-aware pass — fires first; emits named gate if axis-aligned and skips to next segment
2. Euler decomposer — existing fallback for generic rotations, unchanged

tests/test_axis_aware.py — 26 new tests

Unit tests for axis_aligned_rotation (per-axis detection, generic → None, near-identity → None, custom atol) and integration tests through optimize().

Example

from qiskit import QuantumCircuit
from rqm_optimize import optimize

qc = QuantumCircuit(1)
qc.rz(0.3, 0)
qc.rz(0.5, 0)

opt = optimize(qc)
print([i.operation.name for i in opt.data])
# Before: ['u']
# After:  ['rz']

Run	Before	After
rz(a) rz(b)	U(0, 0, a+b)	rz(a+b)
rx(a) rx(b)	U(θ, φ, λ)	rx(a+b)
ry(a) ry(b)	U(θ, φ, λ)	ry(a+b)
rx(a) ry(b) rz(c)	U(θ, φ, λ)	U(θ, φ, λ) (unchanged)

PyPI auto-publish

publish.yml already triggers on v*.*.* tag pushes (created by GitHub releases). No changes needed.

---

🔒 GitHub Advanced Security automatically protects Copilot coding agent pull requests. You can protect all pull requests by enabling Advanced Security for your repositories. Learn more about Advanced Security.