Exponential constant in r(n,m) for 𝔽₃ⁿ subspaces

Informal Question (correct exponential constant in r(n,m)) · arXiv:1708.08486

arXiv Informal medium confidence— first stated 2019-11-21

Authors: Jacob Fox, Huy Tuan Pham

Status open high confidence

The question asks whether the exponential base in the exponent of r(n,m) (the maximum size of a subset of F_3^n containing no affine m-dimensional subspace) is 3 (as in the random lower bound r(n,m) >= N^{1-(m+1)3^{-m}}) rather than C≈13.901 (as in the cap-set-supersaturation upper bound). Fox and Pham's paper itself provides evidence that the popular-differences approach cannot close this gap, and no subsequent work resolving or substantially narrowing it was found in the literature search.

Reviewer notes. No follow-up paper resolving or substantially advancing the question was found in five web searches. The paper itself proves that the popular-differences/density-increment approach via Green's arithmetic regularity lemma cannot settle the correct exponential base, leaving the gap between 3 and C≈13.901 open. The question is closely tied to sharp bounds on n_3(alpha,beta) for beta=alpha^{3+o(1)}, which also remains open.

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

Informal. Is $3$, the lower bound on $r(n,m)$ given by considering a random set, the right exponential constant? That is, does $r(n,m) = N^{1-\varepsilon_m}$ hold with the same exponential base $3$ as in the random lower bound $r(n,m) \geq N^{1-(m+1)3^{-m}}$, rather than the base $C \approx 13.901$ appearing in the upper bound $r(n,m) \leq (1+o(1))N^{1-C^{-m}}$?

Context

The lower and upper bounds for $r(n,m)$ (the maximum size of a subset of $\mathbb{F}_3^n$ containing no affine $m$-dimensional subspace) both have the form $N^{1-\varepsilon_m}$ with $\varepsilon_m \to 0$ exponentially fast in $m$, but with different exponential constants: $3$ in the random lower bound and $C \approx 13.901$ in the upper bound derived from the cap set supersaturation result. The authors note that resolving this question is closely tied to whether $n_3(\alpha,\beta)$ is small enough for $\beta = \alpha^{3+o(1)}$, and their main result provides evidence that the approach via popular differences likely cannot resolve it.

Notes. Stated as an open question in running prose in the introduction without a formal labelled environment. PDF source — exponent notation in r(n,m) bounds may be garbled.

Source paper

Popular progression differences in vector spaces II
Jacob Fox, Huy Tuan Pham · 2019-11-21
https://arxiv.org/abs/1708.08486 PDF source