Greedy capsets
Oliver Dawson, Oleg Shuvaev, José Felipe Voloch
TL;DR
This work addresses the capset problem in ${\mathbb F}_3^n$ by examining a greedy removal procedure that starts from the full ambient space and repeatedly deletes the point with the most incident lines in the current set. Grounded in affine-space geometry, it formalizes lines via $P+Q+R=0$ and proves a structural rigidity result: greedy capsets are recursively built by removing a hyperplane $H_0$ with empty intersection and then applying the same procedure inside the two parallel hyperplanes $H_1$ and $H_2$, yielding capsets of size $|C|=2^n$ (with base case $n=1$ matching). This explains why the greedy approach does not produce larger capsets and identifies the canonical example $C=\{0,1\}^n$ as a prototypical greedy capset. The paper also discusses variants and computational observations, noting that alternative greedy strategies did not improve capset sizes.
Abstract
A capset is a subset $C \subset \mathbb{F}_3^n$ with no three points on a line. We characterise the capsets produced by successively removing points from the ambient space such that the removed point has the maximum number of lines contained in the set of remaining points and passing through it until the set of remaining points contains no lines.
