Reproduce “Evolutionary Discovery of Bivariate Bicycle Codes with LLM-Guided Search” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Juan Cruz-Benito, Andrew W. Cross, David Kremer, Ismael Faro
Researchers at IBM Research developed an LLM-guided evolutionary search framework to discover 465 novel bivariate bicycle (BB) and perturbed bivariate bicycle (PBB) quantum codes at practical block lengths (n <= 360). This framework, employing a multi-stage verification pipeline, significantly expands the catalog of known structured qLDPC codes and provides crucial insights into rate-distance tradeoffs and the limitations of heuristic distance estimation methods.
Key insights
- LLM-guided program evolution is an effective and feasible approach for discovering complex algebraic structures like quantum codes, outperforming classical genetic algorithms in finding high-distance codes.
- There exists a consistent empirical tradeoff between encoding rate (k) and minimum distance (d) for BB and PBB codes within the explored parameter space, with high-k codes generally having low distances.
- Heuristic decoders like BP-OSD systematically and severely overestimate code distance, particularly for high-rate codes (up to 12x), necessitating exact MILP verification for reliable claims.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.