Riverlane researchers achieved a 20 percent improvement in quantum error correction performance by using a new data-driven method to model errors. The team’s approach discovers entirely new error mechanisms in quantum systems, moving beyond simply refining the probabilities of those already known. This method generates these mechanisms by combining existing ones, then validates them using detector correlations, and provides each with a defined logical effect for use in decoding. The work demonstrates increased performance both in simulations and on experimental data from Google’s superconducting device.

Data-Driven DEM Augmentation Improves Quantum Error Correction

Riverlane researchers achieved a 9 percent increase in the Λ parameter, a key metric for assessing quantum error correction performance, using data collected from Google’s superconducting devices across code distances 5-7. This improvement stems from a novel data-driven method for augmenting detector error models (DEMs) by identifying previously unaddressed error mechanisms directly from experimental detector-firing statistics. The method generates candidate error mechanisms by pairing existing mechanisms as building blocks, then filters and confirms them using detector correlations beyond simple pairwise interactions.

Each newly discovered mechanism is assigned a well-defined logical effect, derived from its constituent errors, which allows for its direct use in decoding algorithms. This is particularly valuable for correlated graph-based decoders such as correlated matching, as it preserves compatibility with the original code and noise model. Simulations using SI1000 noise, supplemented with additional correlated noise, complement the experimental validation on Google’s hardware.

These simulations, extending to code distances exceeding 10, offer insight into the regime where quantum error correction is most effective, when codes are large enough to suppress logical errors and the impact of unmodeled noise is most pronounced. “To our knowledge, this is the first data-driven DEM augmentation method that produces decomposed hyperedges with assigned logical effects,” the researchers state, demonstrating the advantage of their approach in both simulated and real-world quantum systems.

Correlated Error Mechanisms Limit Standard QEC Models

Riverlane researchers are addressing limitations in standard quantum error correction models by focusing on previously unmodeled error mechanisms. Their work reveals that theoretical noise models, while useful, consistently omit correlated errors that appear in experimental data, hindering performance gains. These correlated errors, such as crosstalk between qubits, introduce dependencies that traditional models treat as independent, leading to inaccuracies in decoding. The team’s approach centers on augmenting detector error models (DEMs) with these newly discovered mechanisms, derived directly from measurement data collected during quantum error correction.

This differs from existing methods that primarily refine the probabilities of known errors. The process begins by combining pairs of existing error mechanisms to generate candidate mechanisms, then filtering and validating them using detector correlations.

A key aspect of this method is maintaining compatibility with existing decoder infrastructure. Each new mechanism is decomposed into lower-order components, preserving functionality within correlated graph-based decoders like correlated matching. Simulations using code distances up to 15, alongside experimental validation, demonstrate the effectiveness of this approach. The researchers observed that the advantage of a correlation-aware DEM becomes more pronounced at larger code distances, as standard noise mechanisms are suppressed and the impact of unmodeled correlated noise increases.

Method Generates Candidate Errors from Existing Mechanisms

The process of identifying new error sources in quantum systems relies on combining existing, understood mechanisms as fundamental building blocks, rather than attempting an exhaustive search of all possibilities. This approach circumvents a computationally intractable problem, allowing the team to efficiently generate candidate error mechanisms from pre-existing models. Initial steps involve calculating the correlation between detector pairs using experimental data, then combining pairs of established error mechanisms to form these candidates.

A rapid pre-filter then discards combinations lacking sufficient correlation, streamlining the process before more intensive validation. This method’s efficiency stems from its focus on pairwise combinations, though the underlying principles extend to incorporating any number of existing mechanisms. By constructing candidates from known errors, the approach can reveal correlations between any pair of noise sources within the system. This is a key advantage, as it moves beyond simply identifying individual errors and begins to characterize how they interact.

The method’s five-step process culminates in a validation stage that uses higher-order detector correlations, ensuring the discovered mechanisms are not spurious artifacts of the data. The researchers note, highlighting the method’s ability to capture complex interactions.

Pairwise & Higher-Order Correlations Validate New Errors

Initial stages of the error augmentation process involve calculating correlations between detectors using experimental data, specifically employing a method detailed in Section II. The researchers note this cutoff represents a trade-off, balancing the risk of discarding genuine signals against the need to reduce computational load. This filtering step is important, as it prevents the more computationally expensive explicit validation stage from being overwhelmed by spurious candidates. Explicit higher-order validation confirms the presence of new error mechanisms. The process uses correlations beyond the initial pairwise analysis, providing a robust check on the candidate mechanisms.

Logical Effects Enable Direct Decoder Use

Decoder performance relies on assigning logical effects to each identified error, a step often missing in current models. This contrasts with approaches that introduce new errors without establishing how they impact the logical qubits, hindering practical application. The construction of these mechanisms also prioritizes compatibility with existing quantum computing infrastructure. This is particularly valuable because it allows smooth integration into established decoding schemes without requiring extensive re-engineering of existing systems.

The team’s approach avoids disrupting the foundational structure of current decoders while enhancing their ability to address complex error patterns. Detection events, representing deviations in expected detector values, is the basis for inferring error corrections. These triggered detectors provide a sparse yet informative representation of underlying error processes, and the decoder uses a detector error model (DEM) to interpret these events.

The DEM contains independent error mechanisms, each defined by the detectors it affects, the logical observables it flips, and its probability of occurrence. The work demonstrates that augmenting these decoder priors with newly discovered mechanisms, alongside their logical effects, improves the accuracy of the decoding process and ultimately reduces logical error rates. The logical error probability, denoted $p_{L} (d, r)$ , is measured by comparing the decoder’s predicted logical outcome to the true outcome after repeated sampling of noisy syndromes.

Decomposition Preserves Compatibility with Graph-Based Decoders

Maintaining decoder functionality is critical when improving quantum error correction, and Riverlane researchers addressed this directly by ensuring each newly discovered error mechanism included a natural decomposition into lower-order components. This improvement was achieved relative to the unaugmented DEM, and nearly matched the performance of an “oracle DEM,” a theoretical ideal. the augmented DEM reduced the logical error rate by more than a factor of four at code distance 15, indicating a substantial gain in reliability. These results were obtained using surface-code simulations with correlated noise, a realistic scenario for near-term quantum devices.

The ability to decompose new error mechanisms into their constituent parts is not merely an academic exercise. It addresses a limitation of previous methods. “Both are essential for ensuring that the augmentation does not impose additional restrictions on the decoder and, in particular, that compatibility with correlated graph-based decoders is preserved when augmenting graph-compatible input DEMs,” the researchers state in their work. This focus on decoder compatibility distinguishes their method and enables practical implementation within current quantum computing systems.

20% Λ Parameter Increase in Surface-Code Simulations

This improvement, observed across code distances 7 through 15, demonstrates a measurable lift in performance within realistic, near-term noise scenarios. The researchers validated their method using both simulated data and experimental results from Google’s superconducting devices. These results were generated by adding a correlated two-qubit XX error mechanism to an SI1000 noise model, then building a baseline detector error model for decoding. Higher Λ values indicate more effective error suppression as the code grows, signifying improved decoder accuracy and a step toward more robust quantum computation.

75% Logical Error Rate Reduction at Distance 15

Simulations incorporating a newly developed error model achieved a 75% reduction in logical error rate at a code distance of 15, demonstrating substantial gains in quantum error correction performance. This improvement, realized by Riverlane researchers, stems from an augmented detector error model (DEM) that surpasses standard approaches by discovering previously unknown error mechanisms directly from experimental data. Experimental validation on Google’s superconducting hardware further confirmed these gains, with the augmented DEM increasing the Λ parameter by 9% across code distances 5 to 7.

This translated to a 14% reduction in logical error rate at distance 7, even with a limited dataset of 50,000 samples per circuit. The researchers addressed data scarcity by employing a recurrence edge method, treating data from multiple rounds as equivalent to additional samples.

The success of the method is tied to its ability to assign a logical effect to each newly discovered error mechanism, facilitating smooth integration with existing decoders. The logical error rate is expected to scale with code distance according to the equation ε(d) = C Λ^(-(d + 1)/2), where a higher Λ value signifies faster suppression of errors and improved decoder accuracy at larger distances. These results demonstrate a pathway toward more robust and reliable quantum computation through data-driven refinement of error correction strategies.

👉 More information
🗞 Discovering correlated error mechanisms: data-driven augmentation of detector error models improving decoding
✍️ Zohar Schwartzman-Nowik, Matthew P. Stafford, Boris M. Varbanov and Joan Camps (Riverlane, Cambridge, UK)
🧠 ArXiv: https://arxiv.org/abs/2611.06605

More like this