Maximally-Matched Mixture Decoding for Correlated Atom Loss in Quantum LDPC Codes

Abstract

Heralded atom loss is a dominant noise channel in neutral atom quantum computing platforms, and losses at a two-qubit gate are strongly correlated: when one atom is lost, its gate partner is often lost too. This work shows how to exploit this correlation to construct a better decoder, focusing on [[144,12,12]] Gross code syndrome extraction circuits. Given a loss readout, the possible pairings of the lost qubits according to shared gates form a graph whose maximum matchings carry almost all of the conditional probability mass. Our maximal-matching-mixture (MMM) decoder builds a per-readout decoding hypergraph as a weighted mixture over exactly these loss scenarios. This decoding hypergraph can be used by an ordinary belief-propagation decoder. On the [[144,12,12]] Gross code, MMM lowers the logical error rate by as much as ~25x relative to a loss-unaware decoder and ~2x relative to a loss-aware uniform-mixture baseline from Baranes et al. (2026).

Abstract: arXiv descriptive metadata (CC0). Thank you to arXiv for use of its open access interoperability. This service was not reviewed or approved by, nor does it necessarily express or reflect the policies or opinions of, arXiv.