optimal_givens_decomposition has bad side effects

[kevinsung]

trafficstars

If you call optimal_givens_decomposition and then construct another circuit using the same qubits, the latter circuit can be incorrect. For example, below I create two circuits that should be identical, but the second one is different and incorrect:

import cirq
from openfermion import optimal_givens_decomposition

n_qubits = 3
qubits = cirq.LineQubit.range(n_qubits)
mat = cirq.testing.random_unitary(n_qubits)

circuit1 = cirq.Circuit(optimal_givens_decomposition(qubits, mat))
circuit2 = cirq.Circuit(optimal_givens_decomposition(qubits, mat))

circuit1

0: ──────────────PhISwap(0.25)──────────────────────────────────────────────PhISwap(0.25)──────────Z^-0.577───
                 │                                                          │
1: ───Z^(-1/3)───PhISwap(0.25)^(-5/16)───PhISwap(0.25)───────────Z^0.351────PhISwap(0.25)^-0.435───Z^(1/7)────
                                         │
2: ───Z^0.255────────────────────────────PhISwap(0.25)^(-8/15)───Z^-0.284─────────────────────────────────────

circuit2

0: ───PhISwap(0.25)───────────────────────PhISwap(0.25)─────Z^-0.577───
      │                                   │
1: ───PhISwap(0.25)^0───PhISwap(0.25)─────PhISwap(0.25)^0───Z^(1/7)────
                        │
2: ─────────────────────PhISwap(0.25)^0───Z^-0.284─────────────────────

[ncrubin]

This is because optimal_givens_decomposition is mutating the variable mat I think. Can you confirm this?

[kevinsung]

You're right, the issue is that mat is being mutated.