Skip to main contentIBM Quantum Documentation

Synthesize unitary operations

A unitary operation describes a norm-preserving change to a quantum system. For nn qubits this change is described by a 2n×2n2^n \times 2^n dimensional, complex matrix UU whose adjoint equals the inverse, that is UU=1U^\dagger U = \mathbb{1}.

Synthesizing specific unitary operations into a set of quantum gates is a fundamental task used, for example, in the design and application of quantum algorithms or in compiling quantum circuits.

While efficient synthesis is possible for certain classes of unitaries – like those composed of Clifford gates or having a tensor product structure – most unitaries do not fall into these categories. For general unitary matrices, synthesis is a complex task with computational costs that increase exponentially with the number of qubits. Therefore, if you know an efficient decomposition for the unitary you would like to implement, it will likely be better than a general synthesis.

Note

If no decomposition is available, Qiskit® provides you with the tools to find one. However, note that this generally generates deep circuits that may be unsuitable to run on noisy quantum computers.

import numpy as np
from qiskit import QuantumCircuit
 
U = 0.5 * np.array([
    [1, 1, 1, 1],
    [-1, 1, -1, 1],
    [-1, -1, 1, 1],
    [-1, 1, 1, -1]
])
 
circuit = QuantumCircuit(2)
circuit.unitary(U, circuit.qubits)

Re-synthesis for circuit optimization

Sometimes it is beneficial to re-synthesize a long series of single- and two-qubit gates, if the length can be reduced. For example, the following circuit uses three two-qubit gates.

output

However, after re-synthesizing with the following code, it only needs a single CX gate.

from qiskit.quantum_info import Operator
 
# compute unitary matrix of target_circuit
U = Operator(target_circuit)
 
# re-synthesize
better_circuit = QuantumCircuit(2)
better_circuit.unitary(U, range(2))
better_circuit.decompose().draw()
output

Qiskit's transpile function automatically performs this re-synthesis for a sufficiently high optimization level.


Next steps

Recommendations
Was this page helpful?