Work

A threshold separation between maximum-likelihood and minimum-weight decoding with uniform tie-breaking

Year: 2026
Type: preprint
Abstract: For a family of toric CSS codes with two logical qubits and bounded check weights, we prove a threshold separation between maximum-likelihood decoding and specified uniform tie rules, both with a perfect syndrome and with repeated rounds of selectively noisy measurements. Under independent bit flips and a single perfect syndrome, maximum-likelihood decoding succeeds with probability tending to one at each fixed bit-flip rate up to 3/2500. Minimum-weight decoding fails with probability tending to 3/4 when sampling uniformly among tied logical classes on [13/25000, 3/2500], and when sampling uniformly among minimum-weight errors on [11/10000, 3/2500]. Every minimum-weight rule succeeds at rates up to 1/5000. With L rounds on a torus of side length L, independent measurement flips of rate 0 < q ≤ 1/2000 on the extended square checks before the final round, and exact inner checks and final readout, maximum-likelihood decoding still succeeds at bit-flip rates up to 3/2500. Choosing uniformly among the final classes of maximum-probability histories gives failure probability tending to 3/4 for rates in [13/25000, 1/2), and every maximum-probability-history rule succeeds up to 1/9000. Exact inner readouts reduce each replacement graph to one parity per round, with distinct effective posterior weights and minimum history costs on a space-time dual graph. Winding cycles bound posterior failure, and zero-cost changes in a single round tie all four final classes. For the perfect syndrome, a signed bond expansion also controls the ratios of minimum-weight counts.
Author: Pei, Yan Ru
Language: English
Open Access status: green
OpenAlex ID: W7235315544
Cites:
Cited by:
Related to: 10