arXiv:2610.39067
Individually Addressed Logical Pauli Measurements by QLDPC Surgery on a Trapped Ion Quantum Computer
Surabhi Luthra, Alex Owens
Abstract
Quantum low-density parity check (QLDPC) codes may lead to order-of-magnitude reductions in the qubit overhead required for fault-tolerant quantum computation. Achieving this improved overhead requires logical operations that selectively address individual logical qubits encoded within multi-qubit code blocks. Recently proposed architectures achieve this through QLDPC surgery, in which non-destructive Pauli measurements are performed using ancillary ''gadget'' systems transiently entangled with the code block, generalising lattice surgery. Here we implement QLDPC surgery on a [[10,2,3]] generalised bicycle code using Quantinuum's H2-2 trapped ion processor. A single-round surgery measurement using seven gadget qubits measures the $\bar{X}_1$ operator with a 0.72 [0.32, 1.41]% (95% credible interval) error rate and agrees with subsequent transversal measurement with 1.03 [0.60, 1.65]% error. Decoding across three rounds of syndrome extraction suppresses the gadget measurement error, reducing the error rate to below 0.26%. The syndrome after surgery confirms return to the code space and the unaddressed ''spectator'' qubit is preserved with error 0.21 [0.04, 0.67]% during the three-round measurement. We further demonstrate composition of conjugate basis gadgets by sequentially measuring $\bar{X}_1$ then $\bar{Z}_2$, reusing the gadget qubits, and obtain a 1.12 [0.60, 1.94]% error in the $\bar{Z}_2$ measurement. These results constitute a hardware validation of the generalised surgery primitive underlying Pauli-based computing for QLDPC architectures.
AI Summary
QLDPC architectures require non-destructive measurements that address individual logical qubits within multi-qubit code blocks. This paper experimentally tests a generalised-surgery construction for this purpose on Quantinuum’s H2-2 trapped-ion processor, using a $[[10,2,3]]$ generalised bicycle code, seven gadget qubits, and 27 physical qubits overall. A single-round $\bar X_1$ measurement has a decoded error rate of $0.72\%$ (95% credible interval $[0.32,1.41]\%$) relative to verified eigenstate preparation, and disagrees with subsequent destructive logical readout at $1.03\%$. Conjugate-basis preparations produce random outcomes correlated with subsequent readout, testing measurement-induced projection. Three-round syndrome decoding gives zero measurement errors in 957 verified shots, bounding the error below $0.26\%$. Compiled-circuit fault searches establish measurement circuit distance increasing from one to three; removing inter-round gadget-check detectors yields seven errors on the same dataset. Fusion-corrected syndrome statistics test return to the code space. Spectator readout has two errors in 957 three-round shots. Sequential $\bar X_1$ then $\bar Z_2$ measurements reuse the gadget qubits, with the second measurement error at $1.12\%$ ($[0.60,1.94]\%$). The three-round bound depends on initial-syndrome verification: without it, ten errors occur in 1,000 shots, all during a multi-shot hardware disturbance. The demonstration uses minimal distance, offline decoding, and only $|++\rangle_L$ for the three-round test; the code also admits a rotated-toric interpretation. Higher-distance operation, real-time feed-forward, and Pauli-product measurements remain future steps. Hardware compilation details are unavailable, limiting the noise model.
Error found: The abstract's three-round gadget error bound below approximately 0.26% is conditional on initial trivial-syndrome verification, which removes all ten failures seen in 1000 unfiltered shots, but the abstract does not disclose that condition.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.