kakeya

The Kakeya question

A Besicovitch set in the plane is a set of points containing a unit line segment in every direction — a rotation of a needle through every angle, with the needle's track left behind. In the 1920s Abram Besicovitch shocked the mathematical community by constructing such a set with arbitrarily small Lebesgue measure (a "Besicovitch set of measure zero"). The Kakeya question that grew out of this: how small can such a set be in higher dimensions?

The Kakeya conjecture, sharpened by Tom Wolff in the 1990s, asks: every Besicovitch set in ℝⁿ has Hausdorff dimension n. Wolff proved (5n+3)/4 in 1995. Improvements have crawled forward via the polynomial method, decoupling, and elaborate inductive arguments. Open in ℝ³ until very recently; ℝⁿ for n ≥ 4 still partially open.

The finite-field version

Wolff (1999) posed the discrete analogue: let 𝔽_q be the finite field with q elements. A set K ⊂ 𝔽_qⁿ is Besicovitch if it contains a line in every direction (a line in 𝔽_qⁿ is a set of q points {a + tb : t ∈ 𝔽_q} for some direction vector b). The finite-field Kakeya conjecture says every Besicovitch set has at least c_n · qⁿ points for an absolute constant.

This stood as a hard combinatorial problem for a decade. Then:

Dvir's 2008 proof

Zeev Dvir, then a graduate student, settled it in five pages. The argument is breathtakingly short:

1. Suppose K has fewer than ⌈qⁿ/n!⌉ points.
2. The space of polynomials in n variables of total degree < q has dimension binomial(q+n−1, n) ≈ qⁿ/n!. That's strictly larger than |K|.
3. Linear algebra: the constraint "P(x) = 0 for every x ∈ K" is |K| linear conditions on the coefficients. With more variables than conditions, a non-zero polynomial P exists that vanishes on K.
4. Restrict P to any line L ⊂ K: this gives a polynomial of one variable, of degree < q, vanishing at all q points of the line. The fundamental theorem of algebra (finite-field version) forces it to be identically zero.
5. So P vanishes identically on every line in every direction. Looking at the highest-degree homogeneous part of P, this means P's leading term vanishes on every direction in 𝔽_qⁿ — i.e. everywhere — so P is zero. Contradiction.

Therefore |K| ≥ qⁿ/n!, up to lower-order terms. In the plane (n = 2) this gives |K| ≥ q(q+1)/2, the bound shown in the stats panel.

The polynomial method

Dvir's argument is the prototype of what's now called the polynomial method. Three ingredients keep recurring:

• Counting: dimension of a polynomial space exceeds the number of constraints, so a non-zero polynomial exists with the desired vanishing.
• Algebraic restriction: the polynomial, restricted to a line, curve, or low-degree variety, becomes one-variable and is forced to be zero by degree count.
• Global propagation: many local "this polynomial is zero" facts force the polynomial to be globally zero, contradicting non-zeroness.

The technique has since resolved or made progress on a long list of problems:

• 2010–2015: Guth–Katz solve the Erdős distinct-distances problem (see guthkatz).
• 2016: Croot–Lev–Pach + Ellenberg–Gijswijt bound the cap-set problem.
• 2017: Naslund's improvement on the Hales–Jewett theorem in low dimensions.
• 2018+: Maynard, Tao, and others apply variants in additive combinatorics and harmonic analysis.

It now sits in nearly every combinatorial-geometer's toolbox. Dvir's five pages started a movement.

What's still open

The real-Euclidean Kakeya conjecture — the one Besicovitch's measure-zero set provoked, almost a century ago — is not resolved. The polynomial method gave the finite-field analogue, but real ℝⁿ is harder: the field has infinite cardinality, you don't get to substitute the fundamental theorem of algebra, and Hausdorff dimension is a more delicate quantity than cardinality. As of 2024 the dimension of Besicovitch sets in ℝ³ was nailed down to 3 by Wang–Zahl (a deep paper using the polynomial method and incidence geometry and decoupling); ℝⁿ for n ≥ 4 remains the open frontier.

What's in this site

• plane — interactive 𝔽_q² for q ∈ {3, 5, 7, 11}. Click cells to toggle membership in your candidate set K. A compass tracks the q+1 directions; each direction's tick fills in when some line in that direction lies inside K. Stats show |K|, directions covered, and Dvir's lower bound q(q+1)/2. The reveal Besicovitch button drops in a known construction (q+1 parabola tangents plus the vertical line at infinity) — a Besicovitch set of size near the bound.
• docs — this page.

Why this sits next to guthkatz

The polynomial method that proves guthkatz's distinct-distances lower bound is, in retrospect, a direct descendant of Dvir's argument here. The shape is the same: assume the set is small, build a polynomial of bounded degree that vanishes on it, restrict to a line, force the polynomial to be zero, contradict. Six years apart. One in a discrete plane over a tiny field; one in real geometry's hardest open problem.

sources

• Dvir, Z. (2008). On the size of Kakeya sets in finite fields. Journal of the American Mathematical Society 22(4): 1093–1097.
• Wolff, T. (1999). Recent work connected with the Kakeya problem. Prospects in mathematics. American Mathematical Society. — the paper that posed the finite-field reformulation.
• Guth, L. (2016). Polynomial methods in combinatorics. AMS University Lecture Series. — the canonical exposition of the technique.
• Terry Tao — exposition of Dvir's proof on his blog.
• Wang & Zahl (2024). Volume estimates for unions of convex sets, and the Kakeya set conjecture in three dimensions. The ℝ³ resolution.