What was already known
The authors build on an earlier differential-equation approach that reduced the problem to local inequalities. Mahdavifar and Beirami posted a separate proof in September 2026.
AI involvement
The authors credit the CUHK team with the Bellman function and a principal lower bound. Gemini led the split into parameter regions and developed many of the bounds and interval certificates. Google’s researchers directed the model work.
- Human–AI collaboration and checking accountAI work and human attribution
Importance
The proof settles the one-bit conjecture for arbitrary Boolean functions, including unbalanced ones, in every dimension.
Technical verification
The authors report an end-to-end Lean kernel check of the full theorem, with the main result collected in the pinned final check.
Formal checking
Checked the full one-bit theorem, including the computational certificates used in the argument.
- Checker
- Lean 4.33.1 kernel, as reported by Chen and colleagues
- Scope
- Chen and colleagues prove that if is uniform on the Boolean cube and is obtained by independently flipping each coordinate with probability , every Boolean function satisfies . Here is binary entropy. Reporting a single input coordinate attains equality.
- Basis
- Reported by the linked checker
- Pinned final Lean checkPinned final Lean check
Human understanding
The CUHK researchers explain how the Bellman function reduces the question to a finite-dimensional inequality. The paper credits the model with much of the subsequent regional argument.
- How the local inequality implies the information boundHow the local inequality implies the information bound
Other useful sources
Correction and publication history
Added a reviewed earlier manuscript to the research catalog.
Read this published revision