arXiv:2609.35535
Analysis of asymmetric errors in NISQ experiments
Gil Kalai, Tomer Shoham, Carsten Voelkmann
Abstract
This paper continues our analysis of NISQ experiments and especially the Google 2019 ``quantum supremacy" experiment. In arXiv:2008.05177 we proposed statistical tools to analyze symmetric and asymmetric readout errors, and the symmetric case was extensively developed and studied in arXiv:2404.00935, where we used Fourier methods combined with statistical tools. Here we extend and complement our earlier study and analyze asymmetric readout errors for data from experiments and simulations. Our study was further motivated by a paper by Fefferman et al. arXiv:2306.16659, who showed that the effect of amplitude damping gate errors is to have larger fractions of zeroes in the samples produced by the quantum computer. We show that this effect of gate errors is not observed in the Google experimental data, where, for the entire range of the number of qubits, $12 \le n \le 53$, the fraction of zeroes and ones reflects only the asymmetry of reported readout errors, and report on simulations with $n=12,14$ using IBM's and Google's simulators.
AI Summary
The paper asks whether non-unital gate noise, particularly amplitude damping, produces an excess of measured zeroes beyond that explained by asymmetric readout errors in random-circuit sampling experiments. It extends earlier statistical analyses using the model $R(q_{1\to0},q_{0\to1})[s\mathcal P_C+(1-s)/2^n]$, comparing observed bit frequencies and maximum-likelihood parameter estimates with reported readout calibrations. For Google’s 2019 data across $12\le n\le53$, the observed zero–one imbalance is generally below the calibration-based readout imbalance, with no additional zero bias identified. For 12 qubits, mean fitted rates are $q_{1\to0}=0.046$ and $q_{0\to1}=0.022$, versus calibration values approximately $0.050$ and $0.021$; for 14 qubits, fitted rates are $0.042$ and $0.020$. In 12-qubit IBM Fake Guadalupe simulations, fitted rates are instead $0.081$ and $0.048$, compared with physical readout rates $0.0278$ and $0.0072$. The increased asymmetry is interpreted as potentially reflecting gate noise. Additional analyses cover five-qubit IBM experiments and temporal drift in bit frequencies in Sycamore and Zuchongzhi datasets. An appendix sketches an extension of the Fourier treatment to asymmetric errors using biased Fourier–Walsh bases and tensor-product transforms. The paper notes discrepancies between two Google readout-calibration datasets and ambiguities in the Google simulator error settings, leaving that simulation comparison inconclusive. It calls for further experimental samples and controlled simulations, including larger circuits. The empirical finding concerns the absence of additional measured zero bias, rather than establishing the absence of amplitude damping.
Central claim may not hold: The abstract's claim that the observed bias reflects only readout asymmetry is not established because the analysis does not separate gate-induced bias from uncertainty or mismatch in the readout calibrations. Table t:google-readout-data-av contains empirical zero–one differences above the original calibration at 40, 41, and 43 qubits, contradicting the stated universal inequality. The description of empirical bias as roughly 10%–20% below calibration is inconsistent across calibration sets and qubit counts, including a 34.5% deficit for the original 12-qubit calibration and excesses at 43 qubits. Appendix B and Table t:googleexp misreport the original 12-qubit readout calibration and its derived difference and symmetric average, which should be 0.0335, 0.0383, and 0.05265, respectively.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 29 pages, 6 figures, 10 tables