A threshold separation between maximum-likelihood and minimum-weight decoding with uniform tie-breaking
We construct a family of two-logical-qubit toric CSS codes with bounded check weights for which maximum-likelihood decoding has a strictly higher threshold than either of two uniform minimum-weight rules. Under independent bit flips and perfect syndrome measurements, maximum-likelihood decoding succeeds with probability tending to one for every fixed error rate in [13/25000,3/2500], while minimum-weight decoding that samples uniformly among the tied logical classes fails with probability tending to 3/4. The same failure limit holds on [11/10000,3/2500] when sampling uniformly among the minimum-weight errors. Every minimum-weight decoder succeeds with probability tending to one at rates up to 1/5000. An exact reduction to Ising models on the undecorated torus gives different effective couplings for posterior class masses and minimum weights. At the channel temperature, a bound on winding dual cycles controls the posterior mass of incorrect classes. At zero temperature, the vanishing of many effective energy differences makes the four logical sectors tie, and a signed bond expansion controls the ratios of their ground-state counts.
Authors
- Yan Ru Pei (ORCID: https://orcid.org/0000-0002-6999-3691)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-17
- DOI
- https://doi.org/10.5281/zenodo.24302026
- Primary Topic
- Quantum Computing Algorithms and Architecture
- Type
- preprint