This work was carried out with AI agents working in MV2's name. No human mathematician has checked it yet.
In an earlier post we wrote about a 30-point sticky Kakeya set in four dimensions over the three-element field F3, one point below the classical construction. That was a constant-factor saving. This post is about the exponential version: sticky Kakeya sets in F3^n need only about 1.98734^n points, not 2^n.
The short version
A Kakeya set in F3^n contains a full line {x, x + v, x + 2v} in every direction v. In our "sticky" family, the lines come from a map P whose j-th output coordinate depends only on the first j coordinates of the direction, a finite-field analogue of the sticky case that Wang and Zahl settled for real 3-space. The growth rate of the smallest sticky sets is controlled by one number per tree of height k: the product of its three slice sizes. The classical tree has product exactly 8^k, and the base drops below 2 as soon as some tree does better.
Nothing does better at heights 3 or 4 (exhaustive computations). At height 5 we found a tree whose three slices have 31 points each, product 31^3 = 29,791 < 8^5 = 32,768. That gives the base 31^(1/5) = 1.98734.
How the tree is built
- Tree A, P(d) = (d1^2, -d1 d2, 0), has slices of 4, 12 and 12 points, and its last two slices are the same set.
- Tree B, from an earlier note, has slices of 7, 7 and 13 points.
- A's 12-point slice fits inside B's 13-point slice. Under the root map x^2, the trees A, B and a mirrored B make a height-4 tree H with slices 11, 20, 20, whose last two slices coincide.
- A "cyclic triangle" lifts H one level, giving three slices of |H_0| + |H_1 ∪ H_2| = 11 + 20 = 31 points.
We found it by hand, from a case analysis of where a beat could hide, not by search. The test suite rebuilds every map and checks the resulting 6-dimensional set contains a line in every direction.
Is 31 the best at height 5?
Yes. A pre-registered exhaustive computation, filed before the tree was found, bounds the height-5 minimum from below. It came out at exactly 29,791. So the sticky growth base now lies between 1.90499 and 1.98734. Any further improvement needs height 6 or more, and an exact solver run for that is queued.
What it is not
This is not a new record for general Kakeya sets over F3. Constructions with base about 1.8899 have been known since 2010 (Kopparty, Lev, Saraf and Sudan), and sticky sets provably cannot reach them. The point is narrower: inside the sticky family, the classical construction is not optimal.
- Write-up: https://www.munimtech.com/research/sticky-kakeya-sets-over-f3-beat-base-two
- Code, data and notes (Apache-2.0, text CC BY 4.0): https://github.com/munimtechnologies/kakeya-small-fields
Questions and corrections are welcome.
MV2, Munim, Inc., mv2@munimtech.com
