Munim Technologies

Research/Mathematics

Ruzsa's bound is exact for sticky Kakeya configurations over F3

We prove that the conjectured exact Kakeya constant of the three-element field holds for every 'sticky' configuration, in which directions that agree in their first coordinates get lines that agree there too. Any counterexample must have no invariant subspace at all.

Nov 19, 2026 · 7 min read

A Kakeya set in F3^n contains a full line in every direction. Over the field with three elements, the smallest such sets known grow like 1.889882^n, and the best proved lower bound is 1.82074^n (MV2's note a180). The previous note, a181, reduced the gap to one entropy inequality in characteristic 3, Conjecture R3, and showed it holds near its extremal example. The new note, a182, proves R3 globally for a large natural class, the triangular or 'sticky' configurations, in every dimension. It also shows that any counterexample must be irreducible. One consequence: every sticky Kakeya set in F3^n has at least (3^n - 1)^0.57938 points, which is exactly the growth rate of the best known construction.

1.889882

proved lower base for sticky Kakeya sets, equal to the best known upper base for all Kakeya sets

1.725982

exact sums-differences constant for triangular configurations in characteristic 3

0.55%

share of maps of F3^2 with an invariant line, now covered by the proof

The question

Over F3 a line is three points y, y + d, y + 2d. Choosing one line per direction d gives a configuration: a random direction D and three slices Y0, Y1, Y2, the points of its line. Every slice lies in the Kakeya set, so the entropy of each slice is at most log |K|. Conjecture R3 says the entropy of the direction is at most R0/3 times the sum of the three slice entropies, with R0 = log 27 / log(27/4). If true, it gives the exact Kakeya growth base of F3, (27/4)^(1/3) = 1.889882, because an example of Imre Ruzsa, the map d to d squared, attains R0 and matches the Kopparty-Lev-Saraf-Sudan construction.

What we did

Dimension one. On the line F3 each pair of directions has equal values in exactly one slice. So a map is either affine (one slice collapses completely) or a 'triangle' in which each slice merges a different pair. For triangles R3 becomes a two-variable inequality about binary entropies. a181 had checked it only on a grid. We prove it: at an interior minimum two of the three weights must be equal, and on that segment the derivative changes sign exactly twice, while the function is zero at both ends of the range where it could dip (Theorem 70).

Building up. Next we show that R3 passes through extensions. Suppose some linear quotient of the space carries the configuration to a configuration, meaning the map respects a subspace. If R3 holds for the quotient and for every fibre, it holds for the whole (Theorem 71). The proof is a chain-rule argument. Applied coordinate by coordinate, it gives R3 for every triangular configuration, in which coordinate j of the chosen point depends only on the first j coordinates of the direction. Ruzsa's example is of this kind, so for triangular configurations the constant R0 is exact.

Sums-differences constants

ConstantValueSetting
Ruzsa, R01.725982exact for triangular configurations in F3^N (this note)
Lemm, lower bound1.77898real numbers; does not survive characteristic 3 (a181)
Farnamfar, upper bound1.826348groups without 2-torsion (unrefereed)
Katz-Tao, upper bound1.833333all abelian groups
Values from results/a182.json. Conjecture R3 says R0 is exact for every configuration in characteristic 3, not only the triangular ones.

What we found

Call a Kakeya set sticky if its lines can be chosen so that directions agreeing in their first j coordinates get lines whose chosen points also agree in their first j coordinates. Then the configuration is triangular, and Theorem 71 gives |K| at least (3^n - 1)^(1/R0), base 1.889882. For sticky sets the lower bound therefore reaches the value that all Kakeya sets are conjectured to have. The simple sticky set {0,1}^n together with {0,2}^n has 2^(n+1) - 1 points, so the sticky growth base lies between 1.889882 and 2.

Kakeya growth bases over F3

For sticky sets the new lower bound equals the best upper bound known for all Kakeya sets. Whether sticky sets can actually be that small is open. Source: results/a182.json

The extension theorem also says where a counterexample to R3 would have to live. A smallest counterexample has no invariant subspace. In the plane F3^2 that means no invariant line. Only 2,121,471 of the 387,420,489 maps of F3^2, about 0.55%, have one. So the theorem covers the structured maps, including every coordinatewise construction, but most maps of the plane remain open.

Status of Conjecture R3

CaseStatus
Dimension one, every law and every mapproved (Theorem 70)
Triangular configurations, every dimensionproved (Theorem 71)
Small perturbations of Ruzsa's exampleproved (a181, Theorem 69)
Irreducible configurations, starting with F3^2open; the only place a counterexample can be
Counts of maps of F3^2 are in results/a182.json and checked by test_sd3_sticky.py.

What it means

In the Euclidean Kakeya problem, sticky sets were the first case resolved on the way to Wang and Zahl's 2025 proof in three dimensions. Here the finite-field analogue of the sticky case reaches the conjectured constant, and the rest of Conjecture R3 is reduced to irreducible configurations. Two questions follow: whether minimal Kakeya sets over F3 can be taken sticky, and whether R3 holds for irreducible maps of the plane. Our pre-registered computer search for counterexamples will now concentrate on the irreducible maps.

More research