Turán exponent of Latin square hypergraphs

Conjecture 6.4 · arXiv:2401.00359

arXiv Conjecture high confidence— first stated 2023-12-31

Status open high confidence

Conjecture 6.4 predicts that for any d×d Latin square L, the Turán number satisfies ex(n, H_L) = n^{3-Θ(1/d)}, which would match the bound predicted by the skeletal degeneracy d_max(H_L) = Θ(d) via Conjecture 6.2. The source paper only achieves the weaker upper bound O(n^{3-c/d^2}) via Theorem 1.4. No follow-up work addressing this conjecture was found in a search of the literature through May 2026.

Reviewer notes. No follow-up found in the indexed literature. The four papers citing arXiv:2401.00359 (per Semantic Scholar) address unrelated topics: hypergraph universality, subhypergraph counting, canonical Ramsey numbers, and blowups of triangle-free graphs. The conjecture is recent (posted 2023-12-31) and the gap between the proven bound O(n^{3-c/d^2}) and the conjectured exponent n^{3-Θ(1/d)} remains open.

Auto-reviewed 2026-05-15 with claude-sonnet-4-6 (web search enabled).

Conjecture. Let $L$ be a $d\times d$ Latin square. Define $H_{L}$ to be the 3-uniform 3-partite hypergraph on vertex set $[d]\sqcup[d]\sqcup[d]$ where $(i,j,k)\in E(H_{L})$ if and only if $L(i,j)=k$. Then $\mathrm{ex}(n,H_{L})=n^{3-\Theta(1/d)}$.

Context

For every $d\times d$ Latin square one has $d_{1}(H_{L}),d_{2}(H_{L})=\Theta(d)$, so $d_{\mathsf{max}}(H_{L})=\Theta(d)$, yet Theorem 1.4 only yields $\mathrm{ex}(n,H_{L})\leq O(n^{3-c/d^{2}})$. The 1-skeleton of $H_{L}$ is $K_{d,d,d}$, the same as that of the complete 3-partite hypergraph $K^{(3)}_{d,d,d}$, so current techniques cannot distinguish $H_{L}$ from $K^{(3)}_{d,d,d}$ at the level of the Turán exponent. Conjecture 6.4 predicts that the exponent is $3-\Theta(1/d)$, matching $d_{\mathsf{max}}(H_{L})=\Theta(d)$ via Conjecture 6.2.

Source paper

Ramsey and Turán numbers of sparse hypergraphs
Jacob Fox, Maya Sankar, Michael Simkin, Jonathan Tidor, Yunkun Zhou · 2023-12-31
https://arxiv.org/abs/2401.00359