Asymptotic threshold versus finite-size effects in concatenated-code decoding

Determine whether the gap between the observed pseudo-thresholds of the neural message-passing decoder and the zero-rate hashing bounds arises from finite-size effects at currently accessible concatenation levels or from inherent limitations of the decoder, by tracking threshold crossings at deeper concatenation levels.

Background

The paper reports pseudo-thresholds of approximately 6.8% for bit-flip noise and 12.3% for depolarizing noise, which remain below the corresponding zero-rate hashing bounds of approximately 11.0% and 18.9%. Because optimal decoding can produce logical error rates that initially increase before decreasing as the concatenation level grows, threshold crossings measured at low levels may reflect finite-size behavior rather than the true asymptotic threshold.

The authors state that their present data cannot distinguish finite-size effects from intrinsic decoder limitations. They identify deeper-level simulations using simpler, highly scalable decoding architectures as necessary to resolve whether the observed gap is fundamental.