Better error-correcting codes inch quantum computers toward usable scale

Quantum computers will not be useful until their physical qubits are reliable enough to run long calculations without the noise drowning out the answer. The standard fix is quantum error correction: bundle noisy physical qubits into a single, more stable logical qubit, and run the computation on the logical one. A new arXiv preprint (paper) pushes that bundle closer to engineering scale with five new explicit quantum Tanner code instances and an open-source Julia library that lets other groups build on the work.

For two decades, the field has known good quantum codes exist. What it has not had is a way to write them down at a size engineers can actually fabricate. Most useful results in quantum coding theory are asymptotic: they say error rates can be driven arbitrarily low as blocklength grows, without giving a construction anyone can implement at a specific size. The new paper, Quantum Tanner Codes at Moderate Blocklength, works in the moderate regime of 500 to 1000 physical qubits per logical qubit, where the codes are still tractable to analyze and the constructions are explicit enough to build.

The authors use two complementary descriptions of the same code family. One is a left-right Cayley complex, the other a "lifting" approach that takes a small seed code and extends it using a finite group's commuting actions. They searched non-abelian groups from GAP's SmallGrp library and surfaced five new code instances whose distance upper bounds exceed 20: [[480,8,≤21,≤21]], [[504,4,≤36,≤27]], [[672,4,≤48,≤28]], [[720,6,≤30,≤30]], and [[864,8,≤39,≤31]], with check weights between 9 and 20. In the standard [n, k, d] notation, n is the number of physical qubits, k the number of logical qubits, and d the code distance, the minimum number of errors the code can guarantee to catch.

The bounds come from sQetch, a randomized distance estimator run for up to 350 million trials per code. sQetch gives upper bounds, not proven lower bounds, so the true distances may be smaller. The paper is honest about that gap, and the rest of the field should be too.

The authors also report pseudo-thresholds for their Tesseract decoder: 3.6%–4.6% under phenomenological noise and 0.14%–0.27% under circuit-level noise. A pseudo-threshold is the error rate at which adding the code starts to help rather than hurt, and the gap between the two numbers is the cost of modeling real hardware. The authors say the results are "comparable to prior results at shorter code lengths," which is the right register: incremental, not breakthrough.

The bigger engineering contribution may be the open-source artifact. QuantumExpanders.jl is a Julia library that builds quantum Tanner codes and the Ramanujan graphs they sit on, and the Error Correction Zoo entry for quantum Tanner codes places the construction in the broader landscape. Until now, the field has had asymptotic theory and a handful of one-off constructions; the library turns the search into something other groups can rerun, extend, and stress-test.

That matters because the bottleneck for fault-tolerant quantum computing is no longer the existence of codes. It is the translation from math to layout, and a tool that lets theorists and hardware groups iterate against the same construction is the kind of artifact that compounds. The closest prior baseline is Radebold et al.'s "Explicit Instances of Quantum Tanner Codes", and the new paper extends that line of work further into the regime where engineers can ask serious questions.

The five new code instances are a step, not a landing. Distance upper bounds above 20 at hundreds of physical qubits, paired with an open-source tool, are exactly the shape of progress the field needs: a buildable artifact with honest limits. The watch item is whether the Tesseract pseudo-thresholds hold up when other groups run the codes on their own decoders and noise models.