Fewer edges in error models aid hardware for quantum error correction

Stefano Tinelli of the Institute of Communications and Navigation, German Aerospace Center, and colleagues have developed detector error models with up to five times fewer edges, potentially simplifying the physical construction of quantum error correction hardware. The team introduced non-binary detector error models, which model each fault location with a single random variable representing its full range of possible errors, for example, all 16 outcomes after a two-qubit CNOT gate.

Unlike current binary models that accurately represent only factorizable channels, these non-binary models can exactly represent any Pauli channel. Numerical results using surface, color, and bivariate bicycle codes demonstrate improved performance with this approach, alongside a speed increase of up to 9.4 times on CPUs when using optimized decoding.

Non-Binary Detector Error Models for Quantum Error Correction

The construction of more efficient quantum error correction systems benefits from a new approach to modeling errors in syndrome extraction circuits, streamlining the connections needed for hardware implementation. This advancement addresses a limitation of existing binary detector error models, which are only accurate when channels factorize into independent parts. The new non-binary models use variables over a larger alphabet, such as the 16 two-qubit Pauli errors following a CNOT gate, to represent the full range of potential errors.

This is achieved through a “split-merge-regroup” transformation that builds a compact model by merging error components affecting the same detectors and observables. While binary models decompose errors, the non-binary approach represents circuit-level noise in a more detailed way, potentially leading to more robust error correction. A key advantage of these non-binary models lies in their impact on the complexity of the resulting Tanner graphs, which are used in belief propagation decoding.

Their Tanner graphs have four to five times fewer edges, which is advantageous for hardware implementations. Despite the use of non-binary variables, the belief propagation method still exchanges a single scalar message per edge, maintaining computational efficiency. The team’s work defines a non-binary detector error model as a list of “columns,” each representing a random variable with a prior distribution and matrices defining its impact on detectors.

For example, a quaternary variable, representing four possibilities, connected to a check node will flip the detector if the error “anticommutes” with a specific Pauli operator. This principle extends to 16-ary nodes, allowing for a comprehensive representation of error possibilities. The researchers note in their description of the construction process that the focus is on the underlying error model itself.

Split-Merge-Regroup Transformation Builds Compact Non-Binary DEMs

The construction of detector error models (DEMs) benefits from a transformation that reduces the number of connections required for hardware implementation, a key step toward scalable quantum error correction. This reduction in graph complexity stems from a fundamental shift in how errors are modeled. Current DEMs decompose each error into independent binary mechanisms, while the new method uses a single random variable to represent the original error alphabet.

The team proved this construction is exact when channels factorize into independent binary parts, like depolarizing channels, meaning the binary and non-binary DEMs produce the same decoding results under degenerate quantum maximum-a-posteriori (MAP) decoding. However, the equivalence between binary and non-binary models breaks down under full quantum MAP decoding, suggesting the non-binary approach offers a more accurate representation of error behavior. This is significant because it avoids the increase in complexity that often accompanies more accurate modeling techniques.

Analysis of the resulting Tanner graphs reveals that 16-ary variable nodes are the most prevalent, as the split-merge-regroup procedure allows larger locations to absorb parts of smaller ones. The performance gains are further amplified by optional post-processing, such as ordered search decoding (OSD), which is applied to the binary image of the DEM when BP fails to converge.

The team’s tests show that the non-binary DEM achieves a lower logical error rate (LER) at every tested physical error rate, both with and without OSD. The researchers note that for both DEMs, the LER is dominated by non-convergence, so the gain comes from an improved convergence rate, highlighting the practical benefit of the new approach.

Equivalence of Binary and Non-Binary DEMs Under Degenerate MAP Decoding

Analysis of surface, color, and bivariate bicycle codes reveals these models require between 4.3 and 5.3 times fewer edges in their associated Tanner graphs compared to standard binary DEMs, a reduction that directly translates to less complex physical construction. While binary DEMs traditionally decompose errors into independent components, the newly introduced non-binary DEMs represent errors in a more detailed way, exactly representing any Pauli channel regardless of its factorizability.

This is an important distinction, as binary models are limited to accurately portraying only those channels that break down into independent parts. Numerical results demonstrate that this streamlined structure, coupled with the improved accuracy, yields a lower logical error rate across the tested codes.

The performance gains are particularly pronounced with bivariate bicycle (BB) codes, where the non-binary DEM achieves error rate reductions ranging from 11 to 40 times better than its binary counterpart at varying error probabilities. Decoding speeds are also improved, with BP running up to 9.4 times faster on BB codes using a CPU implementation.

Reduced Tanner Graph Complexity with Non-Binary DEM Implementation

A key aspect of this improvement lies in how these models represent error locations within a quantum circuit. This allows for a more compact representation of the error landscape, directly impacting the physical construction of quantum systems.

For example, analysis of a specific three-location error scenario demonstrates a shift from 15 mechanisms in a binary DEM to a non-binary DEM with only three variable nodes, eight edges and two 4-cycles after regrouping. This simplification isn’t merely aesthetic; fewer edges translate to fewer physical connections required in a quantum computer’s control system.

However, the two approaches diverge in their behavior under different decoding methods. The paper states that BP, which approximates quantum MAP, is also expected to behave differently on the two DEMs, noting that the cycle structure of the Tanner graph significantly influences the performance of belief propagation. The researchers found that splitting an error location into binary parts creates four-cycles within that location, cycles that do not exist in the non-binary graph.

This is because parts of different locations frequently exhibit equal symptoms, leading to redundant connections in the binary representation. The paper explains that in their example, M and the three different Paulis of I are degenerate copies of parts of C, highlighting how the regrouping procedure eliminates these unnecessary cycles.

Performance Gains of Non-Binary DEMs with Belief Propagation Decoding

The resulting decrease in connections translates to fewer physical components needed to implement the decoding process, potentially easing the challenges of scaling up quantum computers. This means the non-binary approach offers greater accuracy without sacrificing performance in specific decoding scenarios, and provides a more complete picture of potential errors.

The team’s transformation splits error locations into binary parts, merges those parts that affect the same detectors, and then regroups them into non-binary variables, streamlining the representation of complex error patterns. This is a surprising result, as one might expect the increased accuracy of non-binary models to come at the cost of greater computational complexity.

Specifically, the analysis shows a reduction in variable nodes of 5.9 to 8 times and a decrease in edges of 4.3 to 5.3 times, leading to a speed increase of up to 9.4 times for BP with ordered statistics decoding on a CPU. This streamlined structure, coupled with the ability to accurately represent any Pauli channel, offers a promising pathway to address the challenge of building scalable and reliable quantum computers.

Error Modeling of Syndrome Extraction Circuits with Pauli Noise

Non-binary detector error models reduce the number of edges in the associated Tanner graphs by a factor of four to five, a structural change that directly benefits hardware implementations of quantum error correction. This reduction in connections simplifies the physical construction of these systems, potentially lowering manufacturing complexity and improving signal routing.

Stay current

See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.

Avatar of Dr. Donovan

Latest Posts by Dr. Donovan: