Kahn's equal-size upward-closed systems conjecture

From papers

Let QnQ_n be the power set of [n]={1,2,,n}[n]=\{1,2,\dots,n\} ordered by inclusion, and let X,Y,ZQnX,Y,Z\subset Q_n be upward closed set systems of equal size. Define the set of points belonging to exactly one of the three systems by

S1=(XYcZc)(XcYZc)(XcYcZ).S_1=(X\cap Y^c\cap Z^c)\sqcup (X^c\cap Y\cap Z^c)\sqcup (X^c\cap Y^c\cap Z).

Kahn's conjecture. If X=Y=Z|X|=|Y|=|Z|, then

S12n49.\frac{|S_1|}{2^n}\le \frac49.

The conjecture is motivated by the product construction for three independent upward-closed systems. The source gives an upper bound of 3ρ(1ρ)/(1+ρ)3\rho(1-\rho)/(1+\rho) for equal-sized systems and notes that the conjectured bound is false in the preceding stronger formulation for n=5n=5; it does not state whether this equal-size formulation is resolved.

Progress summary

Solved

A 2025 preprint claims a counterexample, so the conjecture appears false, but the claim has not been independently verified.

Kahn's conjecture asserts that three upward-closed systems X,Y,ZQnX,Y,Z\subset Q_n of equal size have density at most 4/94/9 of points belonging to exactly one system.

February 2025 counterexample claim

A preprint dated February 20, 2025, claims a counterexample in dimension n=21n=21, with exactly-one density exceeding 0.4470.447, hence exceeding 4/90.4444/9\approx 0.444. Its construction starts from a weighted-hypercube example in dimension 77 and lifts it while preserving equal densities. The result is presented as a disproof, but the source is an unverified arXiv preprint rather than an independently corroborated proof. The stronger bound 3ρ(1ρ)3\rho(1-\rho) was already known to fail for n=5n=5; that is a different formulation.

Current status (as of August 2026): A preprint claims to disprove Kahn's equal-size conjecture in dimension 2121, but absent independent verification the conjecture is not settled; related componentwise questions remain open.

Sources
Sources & referencesView supporting material

Primary source

Kada Williams, “A Correlation Inequality on Three Functions”, arXiv:2502.14857 (2025).

Solutions 1

Counterexample

Disproven by Kada Williams in 2025, see "A Correlation Inequality on Three Functions" (https://doi.org/10.48550/arXiv.2502.14857).

0 endorsements
Kada Williams ·