Circuits optimised on transpile level >= 2 are not equivalent #10928
Description
Environment
- Qiskit Terra version: 0.25.1
- Python version: 3.10.13
- Operating system: MacOS Ventura
What is happening?
When optimising the circuit below using optimization_level=2 or optimization_level=3 with target gate set basis_gates=['cx', 'u3'], the result is not equivalent to the input.
Input circuit
q_0: ─■───────────────■─
│ ┌───┐ ┌───┐ │
q_1: ─■─┤ H ├─■─┤ H ├─■─
└───┘ │ └───┘
q_2: ─────────■─────────
Output circuit:
q_0: ─────────────────────────────────────
┌──────────────┐ ┌──────────────┐
q_1: ┤ U3(π/2,0,-π) ├──■──┤ U3(π/2,0,-π) ├
├─────────────┬┘┌─┴─┐├─────────────┬┘
q_2: ┤ U3(π/2,0,π) ├─┤ X ├┤ U3(π/2,0,π) ├─
└─────────────┘ └───┘└─────────────┘
How can we reproduce the issue?
The following code outputs "Circuits are not equivalent".
import numpy as np
from qiskit import transpile, QuantumCircuit
import qiskit.quantum_info as qi
circ = QuantumCircuit(3)
circ.cz(0, 1)
circ.h(1)
circ.cz(1, 2)
circ.h(1)
circ.cz(0, 1)
# optimisation
opt_circ = transpile(circ, optimization_level=2, basis_gates=["cx", "u3"])
# check equivalence
def get_unitary(circ):
op = qi.Operator(circ)
return np.matrix(op.data)
u = get_unitary(circ)
u_opt = get_unitary(opt_circ)
u_uinv = u * u_opt.H
# fix global phase
u_uinv /= u_uinv[0, 0]
if np.isclose(u_uinv, np.eye(2**circ.num_qubits)).all():
print("Circuits are equivalent")
else:
print("Circuits are not equivalent")What should happen?
The code should output Circuits are equivalent.
Any suggestions?
This only applies to gate sets including the u3 gate. Compiling to e.g. ['cx', 'rz', 'sx', 'x'] does not have this problem.