A blindness property of the Min-Sum decoding for the toric code

Kitaev’s toric code is one of the most prominent models for fault-tolerant quantum computation, currently regarded as the leading solution for connectivity constrained quantum technologies. Significant effort has been recently devoted to improving the error correction performance of the toric code under message-passing decoding, a class of low-complexity, iterative decoding algorithms that play a central role in both theory and practice of classical low-density parity-check codes. Here, we provide a theoretical analysis of the toric code under min-sum (MS) decoding, a message-passing decoding algorithm known to solve the maximum-likelihood decoding problem in a localized manner, for codes defined by acyclic graphs. Our analysis reveals an intrinsic limitation of the toric code, which confines the propagation of local information during the message-passing process. We show that if the unsatisfied checks of an error syndrome are at distance greater or equal to 5 from each other, then the MS decoding is locally blind: the qubits in the direct neighborhood of an unsatisfied check are never aware of any other unsatisfied checks, except their direct neighbor. Moreover, we show that degeneracy is not the only cause of decoding failures for errors of weight at least 4, that is, the MS non-degenerate decoding radius is equal to 3, for any toric code of distance greater or equal to 9. Finally, complementing our theoretical analysis, we present a pre-processing method of practical relevance. The proposed method, referred to as stabiliser-blowup, has linear complexity and allows correcting all (degenerate) errors of weight up to 3, providing quadratic improvement in the logical error rate performance, as compared to MS only.

💡 Research Summary

The paper provides a rigorous theoretical investigation of the limitations of Min‑Sum (MS) message‑passing decoding when applied to Kitaev’s toric code, a leading quantum error‑correcting code for connectivity‑constrained quantum hardware. The authors first formalize a notion called “local blindness” of a message‑passing decoder. Given a syndrome s with several unsatisfied parity checks, they construct a fake syndrome s_c that contains only one unsatisfied check c. If, for every iteration of the decoder, the a‑posteriori belief of any qubit neighboring c is identical for s and s_c, the decoder is said to be locally blind around c. Theorem 1 proves that whenever all unsatisfied checks are at graph distance at least five from each other, the MS decoder exhibits perfect local blindness: the qubits adjacent to any unsatisfied check never receive any information about the existence of the other unsatisfied checks. Consequently such syndromes are undecodable by MS, regardless of the code distance. This result explains why increasing the toric‑code distance beyond d = 9 does not improve the logical error‑rate scaling of vanilla MP decoders.

Next, the authors distinguish between degenerate and non‑degenerate errors. A non‑degenerate error is one whose syndrome cannot be generated by any other error of equal or lower weight. By refining the classical decoding‑radius concept to a “non‑degenerate decoding radius,” they prove (Theorem 2) that for any toric code with distance d ≥ 9 the MS decoder can correctly decode all non‑degenerate errors of weight up to three, but fails on some weight‑four non‑degenerate errors. Hence the MS non‑degenerate decoding radius is exactly three. This finding clarifies the previously observed phenomenon that weight‑5 errors become undecodable even for small‑distance codes.

To overcome these intrinsic limitations, the paper introduces a linear‑time pre‑processing step called “stabiliser‑blowup” (SB). The idea is to locally modify the syndrome representation by adding auxiliary variables that “blow up” each stabiliser, thereby removing low‑weight degeneracy that blocks MS convergence. Theorem 3 shows that SB combined with MS (SB + MS) can correct every error of weight ≤ 3 on toric codes of distance d ≥ 7, irrespective of degeneracy. As a direct consequence, the logical error rate improves from a p² scaling (typical for vanilla MS) to p³, i.e., a quadratic improvement, and the number of required calls to any subsequent post‑processing routine is dramatically reduced.

The authors also discuss extensions. They conjecture that the local‑blindness property holds for the normalized MS decoder for any choice of normalization factor, and provide numerical evidence supporting this claim. Moreover, they compare SB + MS with other recent MP‑based enhancements such as belief‑propagation with memory, generalized BP, and neural‑BP, noting that those methods improve performance only for small distances (d ≤ 9) and still suffer from the same fundamental blindness for larger codes.

In summary, the paper establishes that the toric code’s geometry imposes a hard bound on how far local information can travel under MS decoding, leading to inevitable decoding failures for certain low‑weight, non‑degenerate error patterns. By introducing a simple, linear‑complexity stabiliser‑blowup pre‑processing, the authors restore the ability to correct all weight‑3 errors and achieve a quadratic gain in logical error suppression. This work not only clarifies the theoretical limits of MP decoding on topological codes but also offers a practically implementable remedy that can be integrated into existing quantum error‑correction stacks.