Pairwise independent correlation gap conjecture
For every integer , every monotone submodular function , and every marginal vector with , the correlation gap under pairwise independence is at most : . Here , while is the same maximum restricted to jointly distributed random variables that are pairwise independent.
References
Primary source
Additional references
Progress summary
A new preprint claims to settle the four-variable case and the limiting worst case, while the earlier five-variable counterexample remains valid.
Ramachandra and Natarajan (2025) conjectured that the correlation-gap ratio is at most for every dimension. A June 2026 preprint disproved this for , leaving open until the latest claimed development.
Known results
- and : the bound was known to hold.
- The bound was also known for certain marginal probabilities and specific submodular functions.
- For , a coverage-function example gives ; the example was found with assistance from GPT5.5 Pro.
September 2, 2026 claimed resolution
A new preprint claims the bound for , proves that bound is tight, and constructs asymptotic examples attaining , with an extension to -wise independence. This is a claimed resolution, not yet independently verified.
Current status (as of September 2026): The counterexample and the cases are reported, while the proof, its tightness, and the asymptotic constant are claimed by a new preprint but remain unverified.
Solutions 0
No solutions have been posted yet.