Quantum Error Mitigation

Quantum error mitigation paves the way for commercial use of quantum computers, and it is an important technique for using today's quantum computers without fault-tolerant qubits. Qiskit supports quantum error mitigation through the use of a correction matrix to improve the quality of the results. A simple example of measurement error mitigation will illustrate this technique.

A simpler form of noise can be generated by the final measurement in the circuit (Listing 6). At this point, one of the possible bit strings is extracted; this is not an exact process due to the noise caused by the measurement, among other things. Again the GHZ circuit is used as an example. You have already seen the simulation of a noise-free result. When run on a real quantum computer, the noise is clearly seen in the broader distribution, which includes probabilities of the incorrect states (Figure 6).

from qiskit.visualization import plot_histogram
qdev = my_provider.get_backend( 'ibmq_lima' )
results = execute( qc, qdev, shots=10000 ).result()
noisy_counts = results.get_counts()
print( noisy_counts )
plot_histogram( noisy_counts )

Figure 6: A simpler form of noise can be generated by final measurement in the circuit.

In the simple case of an error-mitigation concept, the individual eight basic states resulting from the three qubits of the GHZ circuit are now traversed (Listing 7).

from qiskit.utils.mitigation import *
qr = QuantumRegister( 2 )
meas_calibs, state_labels = complete_meas_cal( qr=qr, circlabel='mcal' )
for circuit in meas_calibs:
  print('Circuit',circuit.name)
  print(circuit)
  print()

As an example, Figure 7 shows three of the eight circuits. The correction matrix from Figure 8 can be calculated from the individual measurement results; this takes the deviations into account. This correction is then applied to the measurement results from the sample circuit (Listing 8). Figure 9 clearly shows that the measurement error mitigation method corrects the errors caused by noise in a very good way.

from qiskit.compiler import transpile, assemble
t_qc = transpile( meas_calibs, qdev )
qobj = assemble( t_qc, shots=10000 )
cal_results = qdev.run( qobj, shots=10000 ).result()
meas_fitter = CompleteMeasFitter( cal_results, state_labels, circlabel='mcal' )
array_to_latex( meas_fitter.cal_matrix )
# Get the filter object
meas_filter = meas_fitter.filter
# Results with mitigation
mitigated_results = meas_filter.apply( results )
mitigated_counts = mitigated_results.get_counts()
noisy_counts = results.get_counts()
plot_histogram( [ noisy_counts, mitigated_counts ], legend=[ 'noisy', 'mitigated' ] )

Figure 7: In the simple case of an error-avoidance concept, the eight individual basic states resulting from the three qubits of the GHZ circuit are now traversed. Here, three of the eight circuits are shown.

Figure 8: A correction matrix can be calculated from the individual measurement results, taking the deviations into account.

Figure 9: Measurement error mitigation by means of a correction matrix helps to correct the errors caused by noise.

Because the cost of additional measurements to determine the correction matrix grows exponentially with the number of qubits, this method has been further refined, resulting in effective error mitigation even if a larger number of qubits is used.

In Qiskit, this advanced method shows up in the output resulting from Sampler primitives (quasi-probabilities). Again using the simple circuit for the GHZ state, the calculation with Sampler on the non-noise free QASM simulator is shown in Listing 9. The almost uniform probability distribution of the two correct states is shown here in the quasi-probability distribution.

service = QiskitRuntimeService()
with Sampler( circuits=[qc], service=service, options={ "backend": "ibmq_qasm_simulator" } ) as sampler:
  result = sampler( circuits=[0], shots=10000 )
  print( result )
SamplerResult( quasi_dists=[{'000': 0.4944, '111': 0.5056}], metadata=['shots': 10000}] )

One focus of current research on quantum error mitigation is on more advanced, general-purpose methods. Zero-Noise Extrapolation (ZNE) and Probabilistic Error Cancellation (PEC) are both promising techniques. In ZNE, you work directly with noise amplification in the circuit; in PEC, you identify the noise using randomly selected instances of noisy circuits. A blog post by IBM Research on the utility value of these methods provides a more detailed overview [12].

Conclusions

Qiskit is the effort of a large community to make quantum computing technology understandable and applicable and get it ready for production. Community contributions, such as meetups or an annual summer school, are essential to the project's continued development. This article is intended as another example of how you can take your first steps in quantum computing.