(2,2)-GB Codes: Classification and Comparison with weight-4 Surface Codes

TL;DR

This paper introduces new optimal (2,2)-Generalized Bicycle codes constructed from binary circulant matrices, matching the performance of top surface codes and breaking previous limitations, with a focus on their classification and comparison.

Contribution

The paper constructs three new infinite families of optimal (2,2)-GB codes, introduces a CSS-preserving equivalence relation, and classifies all extremal codes below length 200.

Findings

Three new families of optimal (2,2)-GB codes are constructed.

The second family provides the first optimal even-distance GB codes.

The codes match or surpass the performance of existing surface codes.

Abstract

Generalized Bicycle (GB) codes offer a compelling alternative to surface codes for quantum error correction. This paper focuses on (2,2)-Generalized Bicycle codes, constructed from pairs of binary circulant matrices with two non-zero elements per row. Leveraging a lower bound on their minimum distance, we construct three novel infinite families of optimal (2,2)-GB codes with parameters [[ 2n^2, 2, n ]], [[ 4r^2, 2, 2r ]], and [[(2t + 1)^2 + 1, 2, 2t + 1 ]]. These families match the performance of Kitaev's toric code and the best 2D weight-4 surface codes, reaching known theoretical limits. In particular, the second family breaks a long-held belief by providing optimal even-distance GB codes, previously deemed impossible. All are CSS codes derived from Cayley graphs. Recognizing that standard equivalence relations do not preserve their CSS structure, we introduce a CSS-preserving equivalence relation for rigorous comparison of Cayley graph-based CSS codes. Under this framework, the first two families are inequivalent to all previously known optimal weight-4 2D surface codes, while the third family is equivalent to the best-known odd-distance 2D surface code. Finally, we classify all extremal, non-equivalent (2,2)-GB codes with length below 200 and present a comparison table with existing notable 2D weight-4 surface codes.