Munim Technologies

Research/Mathematics

A sticky Kakeya set over F3 that beats the classical construction

We found a sticky Kakeya set in four dimensions over the three-element field with 30 points, one fewer than the classical 31, and a lifting rule that carries the saving to every higher dimension. The classical set is the smallest sticky set exactly up to dimension three.

Nov 19, 2026 · 6 min read

A Kakeya set over a finite field contains a full line in every direction. Over the three-element field F3, a natural family of such sets is sticky: the base point of each line is built coordinate by coordinate, and coordinate j depends only on the first j coordinates of the direction. The classical example, {0,1}n together with {0,2}n, is sticky and has 2(n+1) - 1 points. Earlier notes in this project proved that it is the smallest sticky set in dimensions 1, 2 and 3. We now show that this stops in dimension 4. There is a sticky Kakeya set in F34 with 30 points instead of 31, and a simple lifting rule turns it into a smaller-than-classical sticky set in every higher dimension.

30

points in our sticky Kakeya set in F34

31

points in the classical sticky set

29

proved lower bound for dimension 4 (note a186)

The question

Let U*(n) be the least number of points in a sticky Kakeya set in F3n. Note a186 split every sticky set exactly along its first coordinate. That split gave U*(n) = 3, 7, 15 for n = 1, 2, 3, matching the classical set, and 29 <= U*(4) <= 31. It left open whether the classical set is always optimal (Question S5). A simpler relaxation, a game that keeps only the sizes of slices, had already suggested that cheaper sticky structures might exist (note a185). The open question was whether real sets can nest their pieces as tightly as the game allows.

What we did

The split of note a186 builds a sticky set in F3n from a sticky set K' one dimension lower and one tree T of base points. Its size is |K' ∪ S_h| + |S_a| + |S_b|, where S_a, S_b and S_h are the three slices of the tree. In the classical set each slice has 2(n-1) points. We looked for a tree of height 3 whose slices nest. Its two visible slices have 7 points each instead of 8. The third slice has 13 points, but it is almost entirely hidden inside a 15-point sticky set K' in F33.

The tree has three subtrees of height 2. In two of them, one slice is a single row of three points. In the third, every slice is a row plus one extra point. Placing the rows inside the row-plus-point slices makes the unions cost nothing extra. The result is 16 + 7 + 7 = 30 points. The set is listed point by point in note a187, and the tests rebuild it from its maps and check that all 80 directions have their line inside it.

How the 30 points split across the three planes y1 = 0, 1, 2

planecontentspoints
y1 = 0the 15-point set K' together with the hidden 13-point slice16
y1 = 1slice S0 of the tree7
y1 = 2slice S1 of the tree7
Plane sizes are n4.plane_sizes in results/sticky_break.json; the classical set has 15, 8 and 8.

From one dimension to all of them

To go up one dimension we colour the points of the set with three colours, one for each 2-element subset of F3. We need every line to show exactly two colours. A quadratic map on F3 can then send each line's three points to exactly the 2-element fibres its colours name. That gives a sticky set one dimension up with exactly twice as many points plus one, and a colouring of the same kind (Lemma 160). The 30-point set has such a colouring, so lifting it gives 61, 123 and 247 points in dimensions 5, 6 and 7, and 31 · 2(n-4) - 1 in general. That is always below the classical 2(n+1) - 1.

Smallest known sticky Kakeya sets over F3

points

New sizes are lifted_sizes and classical sizes are classical_sizes in the results file. Every new set is rebuilt and checked by the tests. Source: results/sticky_break.json

What it means

Stickiness does not force the classical size. The classical set is a smallest sticky set exactly in dimensions 1 to 3. From dimension 4 on, there is always a smaller one. This is the first real sticky set that follows the direction the size-only game pointed in. The game's (7, 7, 13) witness from note a186 is realised exactly, because the game's best-case merges really happen.

The saving is a constant factor, 31/32, so the growth rate of the smallest sticky sets is still somewhere between 1.90499 (proved in note a183) and 2. Whether it is below 2 is now the main open question for sticky sets over F3. Nothing changes for Kakeya sets in general: F34 has 27-point Kakeya sets, and since every sticky set there has at least 29 points, the smallest Kakeya sets in F34 are still not sticky.

Smallest sticky Kakeya sets over F3, dimension by dimension

nclassicalbest nowexact minimum
1333
2777
3151515
4313029 or 30
56361open
6127123open
7255247open
Exact values for n <= 3 and the bound 29 are from note a186; the rest is from results/sticky_break.json.

A computation pre-registered before this result, sticky-exact-v1, will decide whether 29 points are possible in dimension 4. Its rule for a sticky set below 31 is now known to fire, because our 30-point set exists.

More research