Stability of the Courtade–Kumar inequality

Vu Khac Ky, Tuan Tran

A Boolean function whose mutual information is close to the coordinate value is close in probability to a signed coordinate, with a bound independent of the dimension. The correlation dependence has the optimal order near zero and, for increasing functions, near the noiseless endpoint.

What was already known

Ky and Tran’s qualitative proof establishes that a coordinate attains the largest mutual information. It announces a companion theorem measuring the distance from near-optimal functions to coordinates.

AI involvement AI-centered

The authors formulated the approach and credit ChatGPT with developing the proofs, preparing the verification code and revising the exposition.

Importance Notable

The stability theorem bounds how far a near-optimal Boolean function can be from an optimizer and quantifies how this changes with the noise level.

Technical verification Partial

Appendix B describes rational series bounds, outward-rounded Arb intervals and exact rational and algebraic domain-coverage checks for finite inequalities.

Formal checking

Checked the stated finite scalar inequalities and domain coverage using exact rational arithmetic and enclosing Arb intervals.

Checker
The authors’ certificate routines
Scope
Finite inequalities and domain coverage listed in Appendix B.
Basis
Reported by the linked checker

Human understanding Few

The authors identify their entropy-barrier and Fourier strategy. Their disclosure leaves human reconstruction of the AI-assisted quantitative inequalities undocumented.

Other useful sources

Correction and publication history