Short answer: Randomized compiling is a technique that inserts random single-qubit gates into a circuit so that coherent errors behave like stochastic Pauli noise, which is easier to predict and suppress. A 2021 PRX paper reports performance gains on a four-qubit quantum Fourier transform and random circuits on a superconducting processor, and a 2022 arXiv preprint proves the effect also holds for non-Markovian errors and reports a drastic improvement on public IBM devices. These are the authors' reported results on small circuits.
The problem it addresses
The PRX abstract explains that in the noisy intermediate-scale era, systematic miscalibrations, drift and crosstalk in qubit control can produce a coherent form of error with no classical analog. The authors say coherent errors limit algorithm performance in unpredictable ways. They add that average error rates from randomized benchmarking and related protocols are not sensitive to the full impact of coherent errors, so they do not reliably predict global algorithm performance.
What randomized compiling does
According to the PRX abstract, randomized compiling is a protocol designed to overcome these limits by converting coherent errors into stochastic noise. The authors say this dramatically reduces unpredictable errors and allows accurate predictions of algorithmic performance from error rates measured with cycle benchmarking. The 2022 preprint describes the same idea as tailoring arbitrary Markovian errors into stochastic Pauli noise.
2021: an experiment
In the PRX paper, the authors demonstrate significant performance gains under randomized compiling for the four-qubit quantum Fourier transform algorithm and for random circuits of variable depth on a superconducting quantum processor. They also accurately predict algorithm performance using experimentally measured error rates. They conclude that the technique can be used to leverage and predict the capabilities of modern noisy processors. The conclusion is their own.
2022: theory beyond Markovian errors
The arXiv preprint, which is not peer reviewed in the version read here, proves that randomized compiling also tailors non-Markovian errors into local stochastic Pauli noise, and it investigates the technique's limitations. The authors say it changes errors in three ways. It prevents the coherent accumulation of errors across gate cycles, including hard-to-remove crosstalk, by destroying intercycle coherent correlations. It converts individual gate cycle errors into Pauli noise. And it reduces the variability inherent to noisy devices. They confirm these predictions on the IBM Quantum Experience platform and describe experimental data showing a drastic performance improvement across public devices.
A 2016 paper
A 2016 Physical Review A paper on randomized compiling was listed as a source, but only its section headings were readable, so its results are not described.
Limits
Randomized compiling converts error types and reduces variance, but the abstracts do not say it removes errors. The experiments are on small circuits, four qubits for the Fourier transform and variable-depth random circuits, and results on public cloud devices depend on those devices at the time. The preprint itself says it investigates limitations, which this post does not detail.
FAQ
What is a coherent error?
A systematic error from miscalibration, drift or crosstalk that has no classical analog, per the PRX abstract.
Why convert coherent errors to stochastic noise?
Stochastic Pauli noise is more predictable, and error rates measured with cycle benchmarking can then predict algorithm performance.
Does it reduce crosstalk accumulation?
The 2022 authors say it prevents coherent accumulation across gate cycles, including crosstalk.
What was tested in 2021?
A four-qubit quantum Fourier transform and random circuits on a superconducting processor.
Does it fix all noise?
No. The papers describe tailoring and reduced variability, not removal of error.
All field notes