The winner's curse in hardware VQE: drift-differenced remeasurement of finite-shot selection bias

Abstract:Finite-shot optimization can make noisy variational quantum eigensolver energies look artificially accurate. We studied six molecules on two IBM Heron r3 processors in eight replicated cells. Seven optimizer-selected values fell below the exact active-space eigenvalue by up to mHa. Independent later evaluation of the selected and final parameter points, differenced to remove a common epoch shift, showed that most of the selected advantage disappeared: the primary recovery was (approximate 95% session-level interval). Across the reported aggregation choices, recovery estimates ranged from 0.892 to 0.989. The measured contrast was consistent in scale with finite-shot selection, while calibration drift and other sources of variation remain possible. Accuracy claims based on best-observed hardware VQE values therefore require an independent check.

arXiv:2609.27704 [quant-ph]

(or arXiv:2609.27704v1 [quant-ph] for this version)

https://doi.org/10.48550/arXiv.2609.27704 Focus to learn more arXiv-issued DOI via DataCite (pending registration)

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.