Fix Statevector.evolve modifying input state in-place (Fixes 15750)

[conradhaupt]

Summary

Statevector.evolve(gate) should leave the input state untouched. However, the input state is currently modified with DiagonalGate as an argument (see #15750). This is caused by a shallow copy and an in-place-array modification when gate has a global phase. This PR adds failing tests, a fix for the failing tests, and a reno release note for the bug fix.

Details and comments

Proof and cause of bug

Statevector.evolve calls Statevector._evolve_instruction where the following code is run (lines 1029-1030)

        if obj.definition.global_phase:
            statevec._data *= np.exp(1j * float(obj.definition.global_phase))

This is an in-place modification of statevec._data. However, when Statevector.evolve is called, a shallow copy is made (line 461). This means that a gate with a global phase modifies the original input state. This is verified with the following short script:

from qiskit.quantum_info import Statevector
from qiskit.circuit.library import DiagonalGate
import numpy as np

gate = DiagonalGate([1, -1])
print(f"Global Phase of instruction is {gate.definition.global_phase:0.3f} rad")

sv = Statevector.from_label("1")
sv_original = sv.copy()
_ = sv.evolve(gate)

np.testing.assert_allclose(sv_original.data, sv.data)

Possible fixes and implemented fix

1. (Implemented fix) Reassign statevec._data when applying the global phase, instead of an in-place multiplication. This comes at the cost of copying the _data array but only when the instruction has a global phase.
2. Deep copy instead of a shallow copy. Instead of _copy.copy (shallow copy) at the beginning of Statevector.evolve, we could call self.copy(). However, this causes test.python.primitives.test_statevector_estimator.TestStatevectorEstimator.test_reset to fail. I tracked this down to the internal rng being copied and the behaviour for resets changing. Instead of fixing the deep copy of the rng, the first fix was chosen.

Implemented Tests

I implemented four tests which verify that five different instructions do not modify the input state to Statevector.evolve. Only the one with DiagonalGate failed. With the fix it passes.

1. XGate: Passes.
2. CXGate: Passes.
3. Operator(XGate()): Passes.
4. QuantumCircuit (Bell-state preparation): Passes.
5. DiagonalGate([1.0, -1.0, -1.0, 1.0]): Failed but now passes.

[qiskit-bot]

Thank you for opening a new pull request.

Before your PR can be merged it will first need to pass continuous integration tests and be reviewed. Sometimes the review process can be slow, so please be patient.

While you're waiting, please feel free to review other open PRs. While only a subset of people are authorized to approve pull requests for merging, everyone is encouraged to review open pull requests. Doing reviews helps reduce the burden on the core team and helps make the project's code better for everyone.

One or more of the following people are relevant to this code:

• @Qiskit/terra-core

[CLAassistant]

All committers have signed the CLA.

[Cryoris]

Thanks for the fix -- I would've expected a deepcopy to be the right solution, but it seems that explicit steps were taken to not deepcopy here, so I think your solution works for fixing the bug now.