Quantum Circuits Become Three Times More Reliable with Learned Error Correction

Christopher Tong and colleagues at Massachusetts Institute of Technology, in collaboration with IBM Thomas J. Watson Research Centre and IBM Quantum, have developed a new learning framework that optimises dynamical decoupling sequences for circuits incorporating mid-circuit measurement and feedforward. The approach achieves a three-fold reduction in dynamic circuit error rates, validated through randomized benchmarking and demonstrated on a 20-qubit implementation of the quantum Fourier transform with measurement. Empirically optimised dynamical decoupling is a superior method to theoretical sequences for mitigating errors in these circuits, with key implications for advancing quantum error correction and measurement-based quantum algorithms.

Empirical optimisation of dynamical decoupling sequences reduces errors in multi-qubit circuits

Error rates dropped threefold in dynamic quantum circuits after the application of empirically optimised dynamical decoupling sequences. This improvement surpasses the performance of theoretically derived sequences, unlocking the potential for reliable quantum computations previously hampered by measurement-induced errors. Standard error correction methods, designed for static qubit errors, struggle with errors arising from mid-circuit measurements, where a qubit’s state is read during processing. These measurements inherently disturb the quantum state, introducing correlated errors that are difficult to address with conventional techniques. Dynamical decoupling, a technique traditionally used to suppress low-frequency noise, involves applying a series of carefully timed pulses to qubits to effectively ‘average out’ environmental disturbances. However, designing optimal decoupling sequences for dynamic circuits, those with mid-circuit measurements and feedforward, is a complex task, as the measurement process itself introduces new error pathways.

Researchers at Massachusetts Institute of Technology and IBM have achieved nontrivial process fidelity on connected chains of up to 20 qubits, a key threshold for scaling quantum algorithms. This represents a significant step forward, as maintaining coherence and control over a larger number of qubits is crucial for implementing complex quantum computations. High signal-to-noise quantum Fourier transforms were achieved immediately after preparing a 10-qubit entangled state, paving the way for more complex and strong quantum operations. The quantum Fourier transform (QFT) is a fundamental building block in many quantum algorithms, including Shor’s algorithm for factoring and quantum phase estimation. Achieving a high-fidelity QFT on 20 qubits demonstrates the potential for performing meaningful computations with this architecture. Applying the optimised dynamical decoupling (DD) sequences to a quantum Fourier transform with measurement successfully demonstrated nontrivial process fidelity across chains of up to 20 qubits, indicating that the error suppression techniques are effective even in the presence of measurement-induced errors.

Conventional methods often struggle with errors introduced by mid-circuit measurements, a process where a qubit’s state is read during computation. The act of measurement collapses the superposition state of the qubit, introducing a probabilistic error that can propagate through the circuit. A high signal-to-noise quantum Fourier transform was achieved immediately after preparing a 10-qubit entangled state, confirming the enhanced stability of the system. This initial entangled state serves as a benchmark for assessing the performance of the error suppression techniques. Empirically learned DD sequences consistently outperformed theoretically derived alternatives, highlighting the benefits of hardware-specific optimisation. Theoretical sequences are often based on simplified models of noise and qubit behaviour, whereas the empirical approach learns directly from the characteristics of the specific quantum hardware. However, these results do not yet indicate sustained performance across increasingly complex algorithms or prolonged computational durations, and substantial engineering remains to address coherence limitations. Maintaining qubit coherence, the ability of a qubit to maintain its superposition state, is a major challenge in quantum computing, and longer computations require improved coherence times.

Empirically optimised decoupling significantly reduces errors in larger quantum systems

Researchers are refining techniques to shield quantum bits from errors, a vital step towards practical quantum computers. Quantum bits, or qubits, are susceptible to various sources of noise, including electromagnetic interference, temperature fluctuations, and imperfections in the control electronics. Dynamical decoupling is one approach to mitigating these errors, but its effectiveness depends on the specific noise environment and the characteristics of the qubits. While empirically optimised dynamical decoupling outperforms existing theoretical methods, its computational demands present a challenge; the learning process itself becomes more complex as the number of qubits increases. The optimisation process involves repeatedly running the quantum circuit with different DD sequences and measuring the resulting error rates. This requires significant computational resources, and the time required for optimisation can become prohibitive as the number of qubits grows. This creates a tension between optimising error suppression and the resources needed to achieve that optimisation, potentially limiting scalability.

Establishing a demonstrably superior method for dynamic quantum circuits is significant, despite the computational intensity of optimising these error suppression techniques as systems grow. A three-fold reduction in error rates, achieved via empirically optimised dynamical decoupling, directly improves the viability of complex quantum algorithms. This reduction in error rates translates to a higher probability of obtaining the correct result from a quantum computation. Successful application to a 20-qubit quantum Fourier transform demonstrates practical benefit beyond theoretical gains, paving the way for more reliable quantum computation with measurement and feedback. The ability to perform a high-fidelity QFT on 20 qubits is a significant milestone, as it demonstrates the potential for scaling quantum algorithms to larger problem sizes.

Empirically optimised dynamical decoupling achieved a threefold reduction in error rates within dynamic quantum circuits. This advance surpasses existing theoretical methods, improving the reliability of complex quantum algorithms and strengthening progress towards scalable quantum computation. The framework offers substantial gains for dynamic quantum circuits, exceeding the performance of theoretically designed sequences. This framework tackles measurement-induced errors, a specific challenge in circuits integrating computation with mid-circuit measurement and feedback, by optimising error suppression at the level of circuit subintervals and qubit groupings. This granular approach allows for more precise control over the error suppression process, tailoring the DD sequences to the specific characteristics of each circuit segment. Nontrivial process fidelity on 20-qubit systems, validated via a quantum Fourier transform, enables high signal-to-noise results immediately after entangled state preparation. This achievement establishes a new, efficient method for error mitigation, prompting investigation into scaling this learning framework to larger, more complex quantum processors and diverse hardware platforms. Future work will focus on developing more efficient optimisation algorithms and exploring the applicability of this approach to different types of quantum hardware, such as superconducting qubits, trapped ions, and photonic qubits.

Empirically optimised dynamical decoupling reduced average error rates in dynamic quantum circuits threefold. This improvement enhances the reliability of quantum computations that incorporate mid-circuit measurement and feedback, a key feature for advanced algorithms and quantum error correction. Researchers successfully applied this technique to a quantum Fourier transform implemented on chains of up to 20 qubits, demonstrating a practical benefit beyond theoretical predictions. The authors intend to extend this learning framework to larger quantum processors and explore its use with different qubit technologies.