Qiskit / qiskit-ibm-runtime / 26886199583 / 1
86%
main: 86%

DEFAULT BRANCH: main
Ran
Files 305
Run time 12s
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 84.446% (+0.002%) from 84.444%
26886199583.1

push

github

web-flow 
Fix FoldRzzAngle global phase at pi/3pi wrap boundary (#2865)

### Summary

`FoldRzzAngle` folds out-of-range `Rzz(θ)` gates into the calibrated
range `[0, π/2]` while preserving the **full** unitary, including global
phase. Wrapping `θ` into `(-π, π]` drops whole `2π` windings, and each
dropped winding flips the sign of the operator (`Rzz(θ + 2π) =
-Rzz(θ)`), so the pass re-adds a `GlobalPhaseGate(π)` for an odd number
of dropped windings.

That correction was gated on `pi < angle % (4*pi) < 3*pi` — a strict
floating-point modulo window. Odd-winding angles that wrap **exactly**
onto the `±π` boundary (e.g. `3π`, `-3π`, `7π`) fail the strict `<`
comparison and silently skip the phase, producing a folded circuit equal
to `-Rzz(θ)` instead of `Rzz(θ)`. Angles like `±5π, ±9π, ±11π` only pass
by floating-point rounding luck, so the behavior is latent and fragile.

This is unobservable for a standalone circuit measured directly
(hardware discards global phase), but becomes physically observable when
the circuit is used as a controlled subroutine, or compared/verified at
the unitary level (`Operator`, equivalence checks, process tomography).

### Fix

Derive the dropped winding parity directly from the wrap that was
already performed, so it stays consistent with `wrap_angle` regardless
of rounding:

```python
windings = round((angle - wrap_angle) / (2 * pi))
if windings % 2:
    replace.apply_operation_back(GlobalPhaseGate(pi))
```

### Verification

A sweep over 40,000+ angues across `[-8π, 8π]` plus every multiple of
`π` and `π/2`:

- **Before:** 9 failures, worst element error `2.0` (a global phase of
π), all at the `±3π, ±7π`-type boundary angles.
- **After:** 0 failures, worst error `3e-15`.

### Tests

Added boundary-angle cases (`±3π, ±5π, ±7π`) to
`test_folding_rzz_angles`, which already asserts exact `Operator ==`
equality (unlike `.equiv()`, this includes global phase and so guards
the regression directly).

Co-authored-by: Diego M. Rodríguez <diego@moreda.io>

7704 of 9123 relevant lines covered (84.45%)

0.84 hits per line

Source Files on job unit-tests_python3.11-windows-latest - 26886199583.1