What was already known
Grassl’s tables record lower and upper bounds for the distance of qubit codes at given lengths and dimensions. The comparison uses the 4 November 2026 lower bounds.
AI involvement
Girard chose the constructions to try and directed the project. AI agents wrote and ran the searches that found the retained codes and produced the original distance certifiers.
Importance
The constructions raise lower bounds across a large range of parameters. At fixed length and distance, the largest reported gain is nine logical qubits.
Technical verification
The release reports a weight-d logical operator and an exhaustive lower-distance check for each of the 192 codes. The original certifiers covered 190, a separately written certifier covered 32, and 30 passed both.
Formal checking
The release reports a weight-d logical operator and an exhaustive lower-distance check for each of the 192 codes. The original certifiers covered 190, a separately written certifier covered 32, and 30 passed both.
- Checker
- Girard’s original Aurora certifiers and the separately written distance checker
- Scope
- Girard reports 192 explicit binary stabilizer codes with lengths from 53 to 254 and distances from 5 to 13. At their own lengths and dimensions, 161 raise the listed lower bound by one and 31 by two. Standard lengthening and dimension reduction propagate these gains to 1,873 entries with positive dimension in the 4 November 2026 Grassl table snapshot.
- Basis
- Reported by the linked checker
- Pinned codes, distance certificates and comparison scriptsTechnical checking and scope
Human understanding
Girard explains how the search promotes selected logical operators to stabilizers so that every low-weight logical operator anticommutes with one of them. A counting argument explains where the method gains logical qubits.
Correction and publication history
New curated research with documented central AI work.
Read this published revision