Competing Fault Histories in Lattice-Surgery Decoding
Abstract
Different physical faults can produce the same detector record while carrying opposite logical labels. We investigate the consequence of this ambiguity for a lattice surgery parity measurement. In a distance-three merge--split circuit, single-gate faults compete with reset/readout pairs on three complete syndromes. Full-syndrome probabilities give a simultaneous 95\% lower confidence bound of $16.64$ parts per million on the cost of retaining the selected single-fault decisions at one noise point. Changing those decisions is harmful at low noise and beneficial at higher noise. To test the physical explanation, we attenuate eight implicated channels and compare the change in override effect with a matched non-target suppression. The two interventions change the decoder effect differently, resolving a contribution associated with those locations. Likelihood predictions fixed before sampling then select useful decisions under mixed Pauli noise and after extending preparation and recovery. The latter test gives a reduction of $13.20$ parts per million, with simultaneous 95\% bounds $[5.55,20.84]$. The mechanism is limited in scope: the revised decisions give up certain single-fault corrections, and under the same witness-selection rule no candidate is identified in the fully expanded distance-five circuit. These results connect a finite-noise decoding trade-off to specific physical channels and test how far that explanation can be used to predict logical-parity performance.
Related
- Patchwork: automated lattice surgery and space-time partitioning for real-time decoding
- Adaptive error management for fault-tolerant quantum computing across hardware-level suppression, system-level diagnosis, and real-time decoding
- Competing lattice structures induced by Sn substitution in CsV3Sb5
- Competing delocalisation and localisation induced by the dipolar tail in fragmented lattice gases
- Floquet engineering of competing antiferromagnetism and d-wave superconductivity on the square lattice
BibTeX
@misc{verma2026competing,
title = {{Competing Fault Histories in Lattice-Surgery Decoding}},
author = {Niharika Verma},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.33591},
eprint = {2610.33591},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.33591},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Competing Fault Histories in Lattice-Surgery Decoding
AU - Niharika Verma
PY - 2026
DA - 2026/10/24
PB - arXiv
DO - 10.48550/arxiv.2610.33591
UR - https://arxiv.org/abs/2610.33591
AB - Different physical faults can produce the same detector record while carrying opposite logical labels. We investigate the consequence of this ambiguity for a lattice surgery parity measurement. In a distance-three merge--split circuit, single-gate faults compete with reset/readout pairs on three complete syndromes. Full-syndrome probabilities give a simultaneous 95\% lower confidence bound of $16.64$ parts per million on the cost of retaining the selected single-fault decisions at one noise point. Changing those decisions is harmful at low noise and beneficial at higher noise. To test the physical explanation, we attenuate eight implicated channels and compare the change in override effect with a matched non-target suppression. The two interventions change the decoder effect differently, resolving a contribution associated with those locations. Likelihood predictions fixed before sampling then select useful decisions under mixed Pauli noise and after extending preparation and recovery. The latter test gives a reduction of $13.20$ parts per million, with simultaneous 95\% bounds $[5.55,20.84]$. The mechanism is limited in scope: the revised decisions give up certain single-fault corrections, and under the same witness-selection rule no candidate is identified in the fully expanded distance-five circuit. These results connect a finite-noise decoding trade-off to specific physical channels and test how far that explanation can be used to predict logical-parity performance.
ER -