The Hellinger conjecture for Boolean functions
Let and let be obtained by passing each bit of through a binary symmetric channel with correlation parameter . Let be Boolean, and define the noise operator by
The Hellinger conjecture. Under these assumptions,
The conjecture is stronger than the most-informative Boolean function conjecture and would imply it. It is presented as open in the paper.
References
- Zijie Chen, Amin Gohari and Chandra Nair, “A Differential Equation Approach to the Most-Informative Boolean Function Conjecture”, arXiv:2502.10019 (2025).
Progress summary
Never refreshed
Nothing recorded yet. Refresh searches the literature and the public web for attempts on this problem, and writes the first summary here.
Solutions 1
Claimed by OpenAI. With H(t)=sqrt(1-t^2), the manuscript claims H(Ef)-E[H(T_rho f)] <= 1-H(rho) for every Boolean function on the uniform cube, every n >= 1 and every rho in [-1,1]. No balanced-output restriction is imposed; signed coordinate functions attain equality. This addresses the page’s Hellinger inequality and also claims its implication to the Courtade–Kumar information bound.See full solution
GitHub repository: https://github.com/openai/math
- OpenAI-119-02-Hellinger-contraction-with-arbitrary-Boolean-output-bias.pdfOpen