Uniform witness conjecture

For every admissible parameter triple (n,d,s)(n,d,s) and every relevant witness family F\mathcal{F}, the uniform witness conjecture asserts that ∣F∣≤(n−1d)\lvert\mathcal{F}\rvert\leq\binom{n-1}{d}; equivalently, the extremal quantity for such witness families is at most (n−1d)\binom{n-1}{d}.

References

Primary source

GitHub

Additional references

Progress summary

Refreshed
Claimed solved

A June 2026 preprint claims the conjecture is false, and a Lean formalization is advertised, but neither has been independently checked.

The uniform witness conjecture, proposed by Chao, Xu, Yip, and Zhang in 2025, predicts the bound ∣F∣≤(n−1d)|\mathcal{F}|\leq\binom{n-1}{d} for the relevant witness families. A counterexample would disprove it.

Known results

  • Chao, Xu, Yip, and Zhang (2025): proved the cases s=ds=d and s=1s=1 for sufficiently large nn.
  • Chao, Xu, and Zakharov: proved s≤d/2s\leq d/2 for sufficiently large nn.
  • The remaining thin question is whether W⁡2s−1,s(n)>(n−12s−1)\operatorname{W}_{2s-1,s}(n)>\binom{n-1}{2s-1} can hold for fixed s≥2s\geq2 and infinitely many nn.

June 2026 disproof claim and September 2026 formalization

Zixiang Xu’s June 2026 preprint claims a construction exceeding (n−1d)\binom{n-1}{d} for broad middle ranges of dd and ss, which would refute the conjecture. A September 17, 2026 record additionally reports that the construction has been encoded and proved in Lean, but supplies no independent inspection.

Current status (as of September 2026): The conjecture is claimed to be refuted, while the preprint and Lean formalization remain independently unverified.

Sources

Solutions 0

No solutions have been posted yet.