arXiv++ Combinatorics

A coloring of the Hales--Jewett cube $[t]^n$ is symmetric if it is invariant under all coordinate permutations, and one-weight if it reads only an integer-weighted count of the letters. We prove that the two classes coincide -- a radix weight realizes every symmetric coloring -- so the symmetric lower-bound problem for the Hales--Jewett numbers is exactly a one-dimensional coloring problem about homothetic copies of a $t$-point set, the case $d=1$ of Gallai's theorem. Optimizing the weight yields $\mathrm{HJ}(3,3)\ge22$ and $\mathrm{HJ}(4,2)\ge14$, the latter in closed form from the new Gallai homothety numbers $G_2(\{0,2,3,5\})=67$ and $G_2(\{0,1,5,6\})=80$; new values at three colors -- $G_3(\{0,1,3\})=42$, $G_3(\{0,1,4\})=57$ and $G_3(\{0,2,5\})\ge77$ -- give $\mathrm{HJ}(3,3)\ge16$ from a one-line certificate. An anatomy of the $(4,2)$ palette locates the source of its compression: it is an extremal object of the bracket regime plus a single boundary scale. An exhaustive census shows how thin the class is: of the $1644$ line-free $2$-colorings of $[3]^3$, exactly $36$ are symmetric. For lines with at most $K$ active coordinates the same machinery gives infinite bracket numbers, $\mathrm{HJ}^{[12]}(3,3)=\mathrm{HJ}^{[12]}(4,2)=\infty$, strictly beyond the sum-type ceilings $κ_{\mathrm{sum}}(3,3)=11$ and $κ_{\mathrm{sum}}(4,2)=10$; for lines whose active set is an interval the machinery is provably blind, the interval ceiling $λ(3,r)$ is settled for every $r$ by assembling the known bounds, and a SAT computation gives the exact value $\mathrm{HJ}^{(1)}(3)=5>4=\mathrm{HJ}(3)$. We close with the Collapse, diagonal-only, and symmetric-extremality conjectures and with open problems on optimal weights. Every certificate displayed in this note has been re-verified by direct enumeration, independently of any solver.

The Hales-Jewett number $\mathrm{HJ}(t,r)$ is the least dimension $n$ in which every $r$-coloring of the cube $[t]^{n}$ contains a monochromatic combinatorial line. We prove $\mathrm{HJ}(3,3)\geq 22$ and $\mathrm{HJ}(4,2)\geq 14$, improving the previous records $\mathrm{HJ}(3,3)\geq 14$ (Farnsworth) and $\mathrm{HJ}(4,2)\geq 12$ (the van der Waerden bound). Both bounds are obtained from coordinate-symmetric colorings, which compress the cube onto the discrete simplex of letter-count vectors; a symmetric coloring is line-free if and only if no corner tuple on the simplex is monochromatic, an exact equivalence that turns line-freeness into a constraint-satisfaction problem of size polynomial in $n$. Each bound is certified by an explicit table of fewer than 600 cells together with a finite, mechanical check of the corner tuples; the SAT solver only finds the witness, while correctness rests on the published table, the reduction lemma, and a dependency-free verification that is in principle hand-auditable.