Quantum Computers: Fewer Measurements, Higher Accuracy

The University of California, Santa Barbara has unveiled MoSAIC, a new quantum error mitigation framework that sharply improves upon existing probabilistic error cancellation techniques. Until now, unbiased error mitigation methods suffered from exponential increases in sampling cost as circuit size grew, limiting their application to smaller quantum systems. Maya Ma and colleagues have created MoSAIC, a framework that enhances the precision of calculations on current quantum computers.

Existing methods for minimising errors become increasingly challenging to implement as quantum processors become larger, however, MoSAIC addresses this limitation, enabling more dependable results from complex quantum systems. Validated on IBM’s 156-qubit processor, MoSAIC partitions circuits into blocks to learn and correct errors more efficiently than previous techniques. Maya Ma and colleagues at the University of California, Santa Barbara have developed MoSAIC, a framework for improving the reliability of calculations performed on quantum computers. Current quantum processors, while rapidly advancing, are susceptible to errors that accumulate during complex calculations; unbiased error mitigation techniques, like probabilistic error cancellation, which works by repeatedly running a calculation and averaging the results to create a clearer outcome, become increasingly difficult to implement as systems grow. MoSAIC overcomes this limitation by dividing a quantum calculation into smaller, manageable sections, akin to breaking down a large task into sub-tasks, and learning to correct errors within these ‘noise-aligned blocks’. This approach, validated on IBM’s 156-qubit processor, achieves better accuracy than existing methods, but the precise details of how MoSAIC scales with increasingly complex systems remain to be explored.

Significant error reduction in quantum computation via modular error mitigation

Error rates dropped to two orders of magnitude lower using MoSAIC compared to standard probabilistic error cancellation, under the same sampling conditions. Standard PEC’s exponential scaling previously limited its use to smaller circuits, but this breakthrough enables accurate quantum computations on systems previously too large for existing error mitigation techniques. Scientists at the University of California, Santa Barbara demonstrated this improvement by preparing the ground state of the one-dimensional transverse-field Ising model for systems of up to 50 qubits, representing a sharp increase in scale.

MoSAIC, or Modular Spatio-temporal Aggregation for Inverted Channels, achieves this by partitioning circuits into ‘noise-aligned blocks’ and applying error correction at this block level, rather than after each layer of computation. Simulations utilising CUDA-Q acceleration validated these trends across diverse noise models, with the performance boost demonstrated on IBM’s 156-qubit Heron processor. The method divides quantum circuits into noise-aligned blocks, learns a simplified noise model for each block, and applies error correction at a coarser level, thereby reducing the overall computational burden. While these results demonstrate sharp scalability, achieving truly fault-tolerant quantum computation still requires substantial reductions in underlying qubit error rates.

Modular error mitigation substantially improves performance on the transverse-field Ising model

Error rates fell to 0.46%, 0.81%, and 1.95% for 14-, 30-, and 50-qubit systems respectively on the TFIM ground state preparation, demonstrating a one to two order of magnitude improvement over standard probabilistic error cancellation (PEC) under comparable sampling conditions. PEC previously suffered from exponentially increasing computational cost as circuit size grew, restricting its application to smaller quantum systems, but this advance directly addresses that vital limitation. The framework mitigates this by partitioning circuits into noise-aligned blocks and applying error correction at the block level, rather than after each layer of computation.

This experimental validation is significant, having been achieved on IBM’s 156-qubit Heron processor, representing the largest PEC-based mitigation demonstration to date on quantum hardware. The current work focuses on a specific problem, preparing the transverse-field Ising model (TFIM) ground state, and broader testing across diverse quantum algorithms is necessary to confirm its general applicability. The computational expense associated with the classical variational optimisation required to learn the block noise model remains a consideration for scalability.

MoSAIC distinguishes itself from alternative error mitigation techniques like zero-noise extrapolation by maintaining an unbiased estimator, a property that approximate methods often sacrifice for efficiency. The blockwise approach reduces both sampling and circuit-depth overhead, enabling mitigation of larger circuits than previously possible with PEC. However, the effectiveness of this technique relies on accurately partitioning circuits into noise-aligned blocks, a process that may require careful tuning and characterisation of the underlying hardware. Real-world impact hinges on streamlining the classical variational optimisation step and demonstrating consistent performance across a wider range of quantum algorithms and hardware platforms. Further research should also investigate the sensitivity of MoSAIC to imperfect block partitioning and the potential for automated block scheduling to maximise mitigation performance and minimise classical computational overhead.

Block-level error mitigation enhances accuracy and scalability in quantum computation

A new quantum error mitigation framework, MoSAIC, has been unveiled by scientists, demonstrably improving upon existing probabilistic error cancellation (PEC) techniques. The framework details accuracy gains of one to two orders of magnitude, achieved by partitioning quantum circuits into noise-aligned blocks and applying error correction at the block level. This contrasts with PEC, which traditionally applies correction after each layer of computation, and directly addresses a vital limitation of PEC: its exponential increase in sampling cost as circuit size grows.

Previous methods required an impractical number of measurements to maintain accuracy on larger quantum systems. The team validated MoSAIC on IBM’s 156-qubit Heron processor, representing the largest PEC-based mitigation demonstrated on quantum hardware to date. The framework employs classical variational optimisation, a machine learning technique, to learn the noise characteristics within each block, though the computational cost of this optimisation step remains an area for future investigation.

Zero-noise extrapolation represents a competitive field for quantum error mitigation, being an approximate technique trading accuracy for efficiency. If MoSAIC scales effectively, it could unlock practical applications for quantum computing in areas like materials science and drug discovery. The ability to run larger, more complex circuits with improved accuracy accelerates the path towards achieving quantum advantage.

The demonstrated deployment on existing IBM hardware suggests relatively rapid translation to real-world quantum computing platforms. MoSAIC represents a significant step forward in managing errors on quantum computers, extending the scale of reliable calculations. This new framework allows for more efficient error correction than previous methods which addressed errors layer by layer. By learning noise characteristics within these blocks, it reduces the computational burden associated with unbiased error mitigation techniques like probabilistic error cancellation, enabling larger, more complex quantum computations. As a result, scientists can now tackle problems previously inaccessible due to the limitations of existing error mitigation strategies, opening avenues for exploring more intricate quantum systems.

The research successfully demonstrated MoSAIC, a new quantum error mitigation framework achieving one to two orders of magnitude better accuracy than standard probabilistic error cancellation techniques under the same sampling conditions. This matters because reducing errors is crucial for building practical quantum computers capable of solving complex problems in fields like materials science and drug discovery. Using IBM’s 156-qubit Heron processor, the team mitigated errors while preparing the transverse-field Ising model ground state for systems up to 50 qubits, representing the largest demonstration of its kind. Future work will likely focus on optimising the classical variational optimisation step and exploring MoSAIC’s scalability on even larger quantum processors.