Quantum Noise Characterisation Enables Accurate Prediction and Error Mitigation

Researchers demonstrate that unidentifiable components of quantum hardware noise do not impede accurate prediction of system behaviour or error mitigation. Employing gate set Pauli noise learning on up to 92 qubits, they show optimising noise characterisation reduces computational demands without affecting final results.

Accurate characterisation of noise is paramount to realising the potential of quantum computation, yet fundamental limitations hinder the precise identification of all noise sources. Researchers have now demonstrated that, despite these inherent ambiguities, reliable prediction of noisy quantum dynamics and effective error mitigation are still achievable through self-consistent learning of noise parameters. This work, detailed in a new study by Edward H. Chen (IBM Quantum, Research Triangle Park) and colleagues, including Senrui Chen (Pritzker School of Molecular Engineering, University of Chicago), Laurin E. Fischer (IBM Quantum, IBM Research Europe – Zürich), and others, utilises a framework termed ‘gate set Pauli noise learning’ to characterise and mitigate noise across a complete set of quantum operations. The team, spanning multiple IBM Quantum locations and the University of Chicago, present their findings in the article “Disambiguating Pauli noise in quantum computers”, validating their approach with experiments involving up to 92 qubits and demonstrating that strategic optimisation of unidentifiable noise parameters can reduce computational overhead.

Noise Parameter Learning Enables Effective Quantum Error Mitigation at Scale

Quantum computation relies on the precise manipulation of quantum states, but is inherently susceptible to noise arising from hardware imperfections. Complete characterisation of this noise is fundamentally limited by the complexity of quantum systems and the sheer number of potential error sources. Recent research indicates that despite these limitations, effective error mitigation is achievable through self-consistent learning of noise parameters, rather than attempting a complete noise specification.

Researchers have validated this approach through experiments utilising up to 92 qubits. They employed a framework called ‘gate set Pauli noise learning’. This technique characterises and mitigates noise across all fundamental quantum operations – including initial state preparation, final measurements, and the application of both single- and multi-qubit gates (the basic building blocks of quantum algorithms).

The core principle involves identifying a set of learnable parameters within a noise model. This model describes how errors affect quantum computations. Crucially, the research demonstrates that aspects of the noise model which are fundamentally unlearnable – termed ‘gauge’ degrees of freedom – do not hinder the ability to predict system behaviour or effectively mitigate errors. This finding offers insight into the intrinsic nature of quantum noise. The team restricted their analysis to two-local Pauli errors – errors affecting at most two qubits at a time – to improve scalability and simplify analysis. Pauli errors represent a fundamental basis for describing quantum errors, encompassing bit-flip and phase-flip errors.

Experimental results confirm that the final, error-mitigated results are independent of the specific choice of gauge. However, optimisation of this gauge significantly reduces the computational resources – specifically, the number of samples required – to achieve a given level of accuracy. This reduction in sampling overhead is critical for scaling quantum computations to larger, more complex problems.

The study establishes a new paradigm for quantum error mitigation, demonstrating that learning the characteristics of noise, rather than attempting to fully define it, is sufficient for achieving robust and reliable quantum computations. This approach offers a pathway towards building practical quantum computers capable of tackling computationally challenging problems.