Auditing an AI-Generated Mathematical Proof: A Correction to a Greedy Conditioning Lemma in Quantum Parallel Repetition

Abstract:Chapter 6 of OpenAI's *Ten Advances in Mathematics and Theoretical Computer Science* claims an exponential parallel-repetition theorem for all finite two-player, one-round entangled games. Early in the proof, the chapter uses a quantitative greedy conditioning lemma. The lemma is meant to select a small set of coordinates (D) such that, after conditioning on winning every coordinate in (D), a randomly chosen remaining coordinate is won with average probability at least (1-\delta). The statement is correct, but the proof as printed contains a polarity error. Its continuation test is written in terms of average success, while the next step requires a coordinate with a large conditional failure probability. That implication is false, and even simple examples can leave the printed procedure without a valid next move. This note gives an explicit counterexample, identifies the intended continuation condition, and supplies a complete corrected proof. The repair is local: it leaves the statement of the lemma and the parameters used later in the chapter unchanged. It should not, however, be read as an independent verification of the main parallel-repetition theorem. More broadly, the example shows how a mathematically plausible AI-generated argument can hide a small but decisive reversal between complementary events.

Artificial Intelligence (cs.AI); Quantum Physics (quant-ph)

arXiv:2608.14673 [cs.AI]

(or arXiv:2608.14673v1 [cs.AI] for this version)

https://doi.org/10.48550/arXiv.2608.14673 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.