problems / schur

Also in this folder: compute/ · lean/

Improve the lower bound for the seventh Schur number

In general

The Schur number S(k)S(k) is the largest nn such that {1,…,n}\{1,\dots,n\} can be kk-colored with no monochromatic x+y=zx+y=z (including 2x=z2x=z). Exact values are known only through S(5)S(5). Rowley (2021) gave S(7)≥1696S(7)\ge 1696 via a fully symmetric 7-coloring of [1696]. A coloring of [1697] would raise the bound by one.

Precise statement

A kk-coloring of [n][n] is sum-free if there are no x,y∈[n]x,y\in[n] (not necessarily distinct) with x+y≤nx+y\le n monochromatic. Tonight: find a 7-coloring of [1697] that is sum-free, or record a precise incomplete SAT/search. Do not claim a new upper bound on S(7)S(7).

What happened (recovered)

q1: no 1697 coloring. Near-miss with 2 violations (file compute/near_1697_two_violations.txt recovered empty). Search log in compute/q1-search-log.json. Memory: commit 61e0369. q2: m=144 seed, still no 1697 coloring; still 2 violations. Alternate-template / CEGAR / exact SAT scripts recovered under compute/. q3: recovered and verified Rowley's published 1696 specimen, then reconstructed and preserved a new 1697 near-coloring with exactly two violations. Unrestricted repair, lazy/full SAT repair, and color-twisted reflection all timed out. This is residue, not a lower bound. q4: exact fixed-outside repairs through 132 mutable vertices were UNSAT; finite defect-vertex swaps and one-cut suffix color exchanges found no coloring; the 537-step color-twisted translation that forbids both residue sums was exactly UNSAT. These are named-family walls, not an upper bound.