How do I use PEC?#

Probabilistic error cancellation (PEC) is an error mitigation technique for improving the estimation of expectation values (see What is the theory behind PEC?).

To apply PEC, you must know how to represent ideal operations as linear combinations of noisy implementable operations. This information is exploited by Mitiq to remove the bias error from noisy expectation values.

  • By ideal operations, we mean noiseless unitary gates applied to specific qubits.

  • By noisy implementable operations, we mean those noisy gates that can be applied to specific qubits by a real backend.

Since real backends are not perfect, the noisy implementable operations do not have an ideal unitary effect and typically correspond to non-unitary quantum channels.

Note: In this documentation we often interchange the terms gate and operation for simplicity. However, in the source code of Mitiq, we follow the notation in which an operation is a gate applied to specific qubits, while a gate represents an abstract (qubit-independent) operation.

As with all techniques, PEC is compatible with any frontend supported by Mitiq:

import mitiq

mitiq.SUPPORTED_PROGRAM_TYPES.keys()
dict_keys(['cirq', 'pyquil', 'qiskit', 'braket', 'pennylane'])

In the next cell you can select the frontend used in this tutorial. For example:

frontend = "qiskit"

Problem setup#

We first define the circuit of interest. For simplicity, in this example we use a single-qubit randomized-benchmarking circuit whose ideal (noiseless) execution is equivalent to the identity operation.

from mitiq import benchmarks

circuit = benchmarks.generate_rb_circuits(
  n_qubits=1, num_cliffords=2, return_type=frontend,
)[0]

print(circuit)
   ┌───┐┌────┐┌──────────┐┌───────┐┌──────────┐┌────┐
q: ┤ Y ├┤ √X ├┤ Ry(-π/2) ├┤ Rx(0) ├┤ Ry(-π/2) ├┤ √X ├
   └───┘└────┘└──────────┘└───────┘└──────────┘└────┘

In the code cell above, we used the function generate_rb_circuits(), in which num_cliffords is the number of random Clifford group elements and is proportional to the depth of the circuit.

We now define an executor which simulates a backend with depolarizing noise and computes the survival probability of the \(|00\rangle\) state.

Note: The body of the following function is supposed to be written by the user and depends on the specific frontend and backend. Here, for simplicity, we first convert the input circuit to the Mitiq internal representation (Cirq) and then simulate it.

from cirq import DensityMatrixSimulator, depolarize
from mitiq.interface import convert_to_mitiq

def execute(circuit, noise_level=0.01):
    """Returns Tr[ρ |0⟩⟨0|] where ρ is the state prepared by the circuit
    executed with depolarizing noise.
    """
    # Replace with code based on your frontend and backend.
    mitiq_circuit, _ = convert_to_mitiq(circuit)
    # We add a simple noise model to simulate a noisy backend.
    noisy_circuit = mitiq_circuit.with_noise(depolarize(p=noise_level))
    rho = DensityMatrixSimulator().simulate(noisy_circuit).final_density_matrix
    return rho[0, 0].real

The function execute can be used to evaluate noisy (unmitigated) expectation values.

# Compute the expectation value of the |0⟩⟨0| observable.
noisy_value = execute(circuit)
ideal_value = execute(circuit, noise_level=0.0)
print(f"Error without mitigation: {abs(ideal_value - noisy_value) :.5f}")
Error without mitigation: 0.03869

We now show how to use PEC to reduce this error.

Appling probabilistic error cancellation (PEC)#

Represent ideal gates as linear combinations of noisy gates#

To apply PEC, we need to represent some or all of the operations of the ideal circuit as linear combinations of noisy implementable operations. Such linear combinations are also known as quasi-probability representations (see What is the theory behind PEC?) and, in Mitiq, they correspond to OperationRepresentation objects. More information on the OperationRepresentation class can be found in the section What additional options are available in PEC?.

But how can we obtain the quasi-probability representations that are appropriate for a given backend? There are two main alternative scenarios.

  • Case 1: The noise of the backend can be approximated by a simple noise model, such that quasi-probability representations can be analytically computed [4, 7, 8]. For example, this is possible for depolarizing or amplitude damping noise.

  • Case 2: The noise of the backend is too complex and cannot be approximated by a simple noise model.

Depending on the previous two cases, the method to obtain quasi-probability representations is different.

  • Method for case 1: A simple noise model (e.g. depolarizing or amplitude damping) is typically characterized by a single noise_level parameter (or a few parameters) that can be experimentally estimated. Possible techniques for estimating the noise level are randomized-benchmarking experiments or calibration experiments. Often, gate error probabilities are reported by hardware vendors and can be used to obtain a good guess for the noise_level without running any experiments. Given the noise model and the noise_level, one can apply known analytical expressions to compute the quasi-probability representations of arbitrary gates [4, 8].

  • Method for case 2: Assuming an over-simplified noise model may be a bad approximation. In this case, the suggested approach is to perform the complete process tomography of a basis set of implementable noisy operations (e.g. the native gate set of the backend). One could also use gate set tomography (GST), a noise characterization technique which is robust to state-preparation and measurement errors. Given the superoperators of the noisy implementable operations, one can obtain the quasi-probability representations as solutions of numerical optimization problems [4, 7]. In Mitiq, this is possible through the find_optimal_representation() function as shown the section What additional options are available in PEC?.

In this example, the execute function simulates the effect of depolarizing noise of some fixed noise_level acting after each gate. Therefore, we are in the first scenario defined above and we can generate a list of OperationRepresentation objects associated to all the gates of the circuit as follows:

from mitiq.pec.representations.depolarizing import represent_operations_in_circuit_with_local_depolarizing_noise

noise_level = 0.01
reps = represent_operations_in_circuit_with_local_depolarizing_noise(circuit, noise_level)
print(f"{len(reps)} OperationRepresentation objects produced, assuming {100 * noise_level}% depolarizing noise.")
4 OperationRepresentation objects produced, assuming 1.0% depolarizing noise.

As an example, we print the first OperationRepresentation in reps, showing how an ideal gate can be expressed as a linear combination of noisy gates:

print(reps[0])
q_0: ───Rx(0)─── = 1.010*(q_0: ───Rx(0)───)-0.003*(q_0: ───Rx(0)───X───)-0.003*(q_0: ───Rx(0)───Y───)-0.003*(q_0: ───Rx(0)───Z───)

IMPORTANT: For PEC to work properly, the noise model and the noise_level associated to the execute function must correspond to those used to compute the OperationRepresentation objects.

Run PEC#

Once the necessary OperationRepresentations have been defined, probabilistic error cancellation can be applied through the execute_with_pec() function as follows:

from mitiq import pec

pec_value = pec.execute_with_pec(circuit, execute, representations=reps)

print(f"Error without PEC: {abs(ideal_value - noisy_value) :.5f}")
print(f"Error with PEC:    {abs(ideal_value - pec_value) :.5f}")
Error without PEC: 0.03869
Error with PEC:    0.00605

As printed above, PEC reduced the error compared to the unmitigated case.