[Paper Review] Kakeya-type sets in finite vector spaces

Swastik Kopparty, Vsevolod F. Lev+2 more|arXiv (Cornell University)|Mar 19, 2010

Limits and Structures in Graph Theory2 references4 citations

TL;DR

This paper studies Kakeya-type sets in finite vector spaces, introducing rank-$r$ Kakeya sets that contain a translate of every $r$-dimensional subspace. Using the polynomial method and novel constructions, it establishes tight bounds on the minimal size of such sets, showing $|V \setminus K| = \Theta(nq^{n-r+1})$ when $r$ is bounded and $n \leq rq^{r-1}$, improving prior estimates and providing nearly optimal constructions for $q$ odd and $r=1$. The results advance the finite field Kakeya problem and open new directions in extremal combinatorics and additive combinatorics.

ABSTRACT

For a finite vector space $V$ and a non-negative integer $r\le\dim V$ we estimate the smallest possible size of a subset of $V$, containing a translate of every $r$-dimensional subspace. In particular, we show that if $K\subset V$ is the smallest subset with this property, $n$ denotes the dimension of $V$, and $q$ is the size of the underlying field, then for $r$ bounded and $r

Motivation & Objective

• To determine the minimal size of a subset $K \subseteq \mathbb{F}_q^n$ that contains a translate of every $r$-dimensional subspace, known as a Kakeya set of rank $r$.
• To improve upon existing lower and upper bounds for the size of such sets, particularly in the regime where $r$ is bounded and $n \leq rq^{r-1}$.
• To develop new constructions of Kakeya sets using lifting, tensoring, and quadratic residue techniques, extending prior work in finite field Kakeya problems.
• To establish the asymptotic tightness of bounds in key parameter regimes, especially for $r=1$ and odd $q$, and to identify open problems in higher-rank settings.

Proposed method

• The paper employs the polynomial method in the spirit of Dvir and others, using degree-$r$ polynomials to derive lower bounds on the size of Kakeya sets via non-vanishing arguments.
• It introduces a lifting technique to construct higher-rank Kakeya sets from lower-rank ones, generalizing a construction by Ellenberg, Oberlin, and Tao.
• A key construction uses the product of rank-1 Kakeya sets and applies the tensor power trick to amplify size efficiency in higher dimensions.
• It leverages projections and subspace decompositions to build universal sets that contain translates of all $k$-tuples, enabling upper bounds via probabilistic and combinatorial arguments.
• The authors use a lifting lemma to show that if $K_1$ is a rank-$r_1$ Kakeya set in a subspace, then $K = K_1 \cup (\mathbb{F}_q^n \setminus \mathbb{F}_q^{n-(r-r_1)})$ is a rank-$r$ Kakeya set.
• For upper bounds, it combines explicit constructions (e.g., for $q=3$) with asymptotic estimates, showing $|K| \leq \left(1 - \frac{q - \delta_q}{2q^r}\right)^{\lfloor n/(r+1) \rfloor} q^n$ with $\delta_q$ depending on $q$.

Experimental results

Research questions

• RQ1 What is the minimal size of a subset of $\mathbb{F}_q^n$ that contains a translate of every $r$-dimensional subspace, for fixed $r$ and growing $n$?
• RQ2 How do the best-known lower and upper bounds for Kakeya sets of rank $r$ compare in the regime $n \leq rq^{r-1}$, and are they essentially tight?
• RQ3 Can the polynomial method be refined to yield stronger lower bounds than previously known, particularly for $r \geq 2$?
• RQ4 Is the universal set construction optimal for $r \gtrsim q / \log q$ when $n$ grows, or can better constructions be found?
• RQ5 What is the exponential growth rate $\lim_{n \to \infty} \frac{1}{n} \ln \kappa_q^{(n)}(r)$ of the minimal Kakeya set size for fixed $q$ and $r$?

Key findings

• The paper establishes a new lower bound: $|K| \geq \left(1 + (q-1)q^{-r}\right)^{-n} q^n$, which improves upon the previous bound $|K| \geq (1 - q^{1-r})^{\binom{n}{2}} q^n$ for large $q$.
• For $r$ bounded and $n \leq rq^{r-1}$, the paper shows $|V \setminus K| = \Theta(nq^{n-r+1})$, tightening the gap between prior bounds of $\Omega(q^{n-r+1})$ and $O(n^2 q^{n-r+1})$.
• A new construction yields $|K| \leq \left(1 - \frac{q - \delta_q}{2q^r}\right)^{\lfloor n/(r+1) \rfloor} q^n$ with $\delta_q = 3$ for odd $q \geq 5$, $1$ for even $q$, and $5/3$ for $q=3$, providing nearly optimal size for $r=1$ and odd $q$.
• The authors show that for $n \leq (1 - \varepsilon)q^{r-1}$, the lower and upper bounds are essentially equivalent, indicating tightness in this regime.
• For $q=3$, a refined construction yields $|K| \leq 3^{r+1} - 2$ in $\mathbb{F}_3^{r+1}$, improving on the naive bound and showing non-vacuous results even in small fields.
• The paper proves that the limit $\lim_{n \to \infty} \frac{1}{n} \ln \kappa_q^{(n)}(r)$ exists for fixed $q$ and $r$, and poses the open problem of computing it explicitly, especially for $q=3, r=1$.