--- title: Arithmetic Kakeya Sets | Combinatorics url: https://www.emergentmind.com/topics/arithmetic-kakeya-sets type: topic --- # Arithmetic Kakeya Sets | Combinatorics Arithmetic Kakeya sets are discrete sets that contain a prescribed one-dimensional configuration in every direction. In the finite-field setting, a Kakeya set \(K\subseteq \mathbb F_q^n\) contains a complete affine line in every direction and is required to have large cardinality. In the integer setting, an arithmetic Kakeya set is a finite set containing a \(k\)-term arithmetic progression for every common difference in a specified range. These formulations are related to finite-ring geometry, additive combinatorics, entropy inequalities, fractal dimension, finite geometric incidence theory, and, through discretization, the Euclidean Kakeya problem. ## 1. Definitions and principal models Let \(F=\mathbb F_q\) and \(F^n\) be \(n\)-dimensional affine space. For \(y,x\in F^n\), the affine line through \(y\) in direction \(x\) is \[ L_{y,x}=\{y+ax:a\in F\}. \] A direction is a vector \(x\in F^n\); nonzero scalar multiples determine the same geometric direction, so directions may equivalently be represented by points of \(\mathbb P^{n-1}(F)\). A finite-field Kakeya set, also called a finite-field Besicovitch set, is a subset \(K\subseteq F^n\) satisfying \[ \forall x\in F^n\ \exists y\in F^n\quad L_{y,x}\subseteq K. \] The zero direction is vacuous. The finite-field problem asks for lower bounds on \(|K|\), ideally of order \(q^n\), the size of the ambient space. Dvir’s polynomial method established this order of magnitude, and the quantitative bound \[ |K|\geq \frac{q^n}{(2-1/q)^n} \] is available in the finite-field setting [2011.11225]. Over non-archimedean local rings, a line in \(R^n\), with \(R=\mathbb F_q[[t]]\) or \(R=\mathbb Z_p\), is \[ L(x,v)=\{x+tv:t\in R\}. \] A Kakeya set contains an entire line in every nonzero direction. In \(\mathbb F_q[[t]]\), directions divisible by \(t\) are redundant because \[ L(x,v)\subseteq L(x,v/t). \] Thus reduced directions, whose coordinates are not all divisible by \(t\), suffice. In dimension two, every reduced direction has a representative of the form \((1,b)\) or \((b,1)\), with \(b\in\mathbb F_q[[t]]\) [1112.3381]. The integer arithmetic Kakeya problem is formulated using \[ F_k(N)=\min\left\{|A|:A\subset\mathbb Z,\ A\text{ contains a }k\text{-term progression of difference }d\text{ for every }d\in\{1,\ldots,N\}\right\}. \] Equivalently, for each \(d\), there must exist \(a_d\) such that \[ \{a_d,a_d+d,\ldots,a_d+(k-1)d\}\subseteq A. \] The arithmetic Kakeya conjecture is \[ \lim_{k\to\infty}\lim_{N\to\infty} \frac{\log F_k(N)}{\log N}=1. \] A variant \(F_k'(N)\) requires \(k\)-term progressions with \(N\) distinct common differences, not necessarily the prescribed differences \(1,\ldots,N\). The two formulations have the same double-logarithmic asymptotics, since \[ F_k(N)\ll k^3\log N\,F_k'(N) \] and \(F_k'(N)\leq F_k(N)\) [1712.02108]. ## 2. Finite-field polynomial methods The first polynomial-method lower bound for finite-field Kakeya sets proceeds through homogeneous polynomials. A \((\delta,\gamma)\)-Kakeya set \(K\subseteq F^n\) is one for which a set \(L\subseteq F^n\) of at least \(\delta q^n\) directions has, in each relevant direction, a line intersecting \(K\) in at least \(\gamma q\) points. Define \[ d=\left\lfloor q\min\{\delta,\gamma\}\right\rfloor-2. \] Then \[ |K|\geq \binom{d+n-1}{n-1}. \] For an ordinary Kakeya set, \(\delta=\gamma=1\), so \(d=q-2\) and \[ |K|\geq \binom{q+n-3}{n-1}, \] which has order \(q^{n-1}\) [0803.2336]. The proof assumes that \(|K|\) is smaller than the dimension of the vector space of homogeneous degree-\(d\) polynomials, \[ \dim_F F[x_1,\ldots,x_n]_d=\binom{d+n-1}{n-1}. \] Dimension counting then produces a nonzero homogeneous polynomial \(g\) vanishing on \(K\). Homogeneity implies \[ g(cx)=c^dg(x), \] so \(g\) vanishes on the cone \[ K'=\{cx:x\in K,\ c\in F\}. \] Along a line containing sufficiently many points of \(K\), scalar normalization yields at least \(d+1\) distinct roots of a univariate restriction of \(g\). The restriction therefore vanishes identically, forcing \(g\) to vanish on every relevant direction. Since the set of such directions has size at least \(\delta q^n\), the Schwartz–Zippel bound, \[ |\{x\in F^n:f(x)=0\}|\leq dq^{n-1}, \] gives a contradiction when \(d<\delta q\). A strengthened argument uses a polynomial \(P\) of total degree at most \(q-1\). If \[ |K|<\binom{q+n-2}{n}, \] dimension counting gives a nonzero \(P\) vanishing on \(K\). Writing \[ P=\sum_{i=0}^{q-1}P_i \] into homogeneous components and restricting \(P\) to a complete Kakeya line produces a univariate polynomial of degree at most \(q-1\) vanishing at all \(q\) field elements. Its coefficients vanish, successively implying \[ P_{q-1}\equiv P_{q-2}\equiv\cdots\equiv P_1\equiv0. \] The constant component also vanishes on \(K\), contradicting nonzeroness. Consequently, \[ |K|\geq \binom{q+n-2}{n} =\frac{q^n}{n!}+O_n(q^{n-1}), \] and hence \(|K|\geq C_nq^n\), with \(C_n>0\) depending only on \(n\) [0803.2336]. The finite-field conclusion is essentially optimal in its dependence on \(q\), since \(F^n\) itself is a Kakeya set. It is not, however, a proof of the Euclidean Kakeya conjecture: the finite-field argument relies on the fact that a degree-\(0\), \[ |S|\geq C_{n,\epsilon}N^{n-\epsilon}. \] This proves the square-free case of the Hickman–Wright conjecture [2011.11225]. The proof uses a line matrix \(M_S\), whose rows are line indicators and whose columns are indexed by points. Since all row supports lie in \(S\), \[ \operatorname{rank}(M_S)\leq |S|. \] The matrix is multiplied by point–hyperplane incidence matrices. Over \(\mathbb F_p\), the relevant incidence matrix has rank \[ \operatorname{rank}_{\mathbb F_p}(W_{p,n}) =\binom{p+n-2}{n-1}+1. \] For composite square-free moduli, tensor-product rank arguments combine the prime components even though the selected line in one component may depend on the complete direction, rather than solely on that component. The general-modulus theorem establishes the Kakeya conjecture over \(\mathbb Z/N\mathbb Z\) for every \(N\). If \(N=p_1^{k_1}\cdots p_r^{k_r}\), then every modular Kakeya set satisfies \[ |S|\geq N^n\prod_{i=1}^r \left(2\bigl(k_i+\lceil\log_{p_i}n\rceil\bigr)\right)^{-n}. \] If every \(p_i\geq n\), the stronger estimate is \[ |S|\geq N^n\prod_{i=1}^r (k_i+1)^{-n} \left(1+\frac{n}{p_i}\right)^{-n}. \] These imply \[ |S|\geq C_{n,\epsilon}N^{n-\epsilon} \] for every fixed \(\epsilon>0\) and sufficiently large \(N\) [2110.14889]. Prime powers require a different method because ordinary finite-field polynomial evaluation is unavailable. For \(R=\mathbb Z/p^k\mathbb Z\), the method uses roots of unity, quotient polynomial rings, Vandermonde-type matrices, Hasse derivatives, and multiplicity decoding. A key rank estimate is \[ \operatorname{rank}_{\mathbb F_p}M_{p^\ell,n} \geq \binom{\lceil p^\ell/\ell\rceil+n}{n}. \] The multiplicity method also yields stronger bounds for \((m,\epsilon)\)-Kakeya sets. If at least an \(\epsilon\)-fraction of directions contain \(m\)-rich lines, then \[ |S|\geq \epsilon\,\frac{m^n} {\left(2(k+\lceil\log_p n\rceil)\right)^n}, \] where the modulus is \(p^k\). If \(p>n\), then \[ |S|\geq \epsilon\,\frac{m^n}{(k+1)^n} \left(1+\frac np\right)^{-n}. \] The dependence on \(\epsilon\) is linear and the dependence on \(m\) is \(m^n\) [2110.14889]. ## 4. Local rings, incidence geometry, and finite geometric variants Arithmetic Kakeya phenomena also occur over non-archimedean local rings. In \(\mathbb F_q[[t]]^2\), Dummit and Hablicsek construct a measure-zero Kakeya set. Write \[ a=\sum_{i\geq0}a_it^i \] and define \[ a^*=\sum_{k\geq1}a_{2k-1}t^{2k-2}. \] The set \[ H=\{(x,y)\in\mathbb F_q[[t]]^2: ax+y=a^* \text{ for some }a\in\mathbb F_q[[t]]\} \] contains a line in every direction \((1,b)\). Symmetrizing under coordinate interchange gives a Kakeya set in all reduced directions. For \(n>2\), taking \[ K^{(n)}=K\times\mathbb F_q[[t]]^{n-2} \] preserves the Kakeya property and Haar measure zero [1112.3381]. The measure-zero assertion follows from finite-stage coefficient systems. If \(S_n(x,y)\) counts compatible coefficient tuples, the construction produces a Markov chain on \[ \{0,1,q,q^2,\ldots\} \] in which every nonzero state has a uniformly positive proportion of transitions to zero. Consequently, \[ \mu(H)=0. \] At the same time, every Kakeya set in \(\mathbb F_q[[t]]^2\) or \(\mathbb Z_p^2\) has full Minkowski dimension \(2\). Discretizing modulo \(t^k\) or \(p^k\), one obtains finite-ring Kakeya sets of size at least \[ \frac{|R_k|^2}{2k}. \] Thus, over \(\mathbb F_q[[t]]\), \[ |E_k|\geq \frac{q^{2k}}{2k}, \] which gives \[ \dim_M(E)=2. \] This separates Haar measure from Minkowski dimension: the density \(|E_k|/q^{2k}\) tends to zero, while the exponential growth rate of \(|E_k|\) remains \(q^{2k}\). A different finite-geometric variant restricts the permitted directions to a nonsingular conic \(C\subseteq\mathrm{PG}(2,q)\). In the linear representation \(T_2^*(C)\), a Kakeya line set consists of \(q+1\) affine lines, one in each direction represented by \(C\). The smallest associated point set has size \[ \left\lfloor\frac{3q^2+2q}{4}\right\rfloor. \] The extremizers arise from partitioning the selected directions between the two reguli of a hyperbolic quadric. If \(k\) lines are chosen from one regulus and \(q+1-k\) from the other, then \[ |K(\mathcal L)| =kq+(q+1-k)(q-k). \] For odd \(q\), this equals \[ \frac{3q^2+2q-1}{4} +\left(k-\frac{q+1}{2}\right)^2, \] so the minimum occurs for the most balanced partition. The classification is governed by the line-intersection graph: vertices represent selected lines, and adjacency means intersection. Maximal cliques are edge-disjoint, and the union size satisfies \[ |K(\mathcal L)| =q(q+1)-\sum_{i=1}^{q+1}k_i(i-1), \] where \(k_i\) counts maximal cliques of size \(i\). A strengthened Mantel stability argument shows that sufficiently small Kakeya sets have an intersection graph close to a balanced complete bipartite graph. Geometry then forces the two parts into the two reguli of a hyperbolic quadric. For odd \(q\), all sets below \[ \frac{3(q^2-1)}4+q \] are classified in this way; for even \(q\), the corresponding threshold is \[ \frac{3q^2}{4}+q-1. \] These results concern finite geometric configurations and do not directly imply results for integer arithmetic Kakeya sets [1601.03539]. ## 5. Additive combinatorics, entropy, and geometric dimension The arithmetic Kakeya conjecture has several equivalent formulations. For a finite \(A\subseteq\mathbb Z^2\), define \[ \pi_r(A)=\{x+ry:(x,y)\in A\}, \qquad \pi_\infty(A)=\{y:(x,y)\in A\}. \] The Katz–Tao projection formulation asserts that for every \(\varepsilon>0\), there exist rational slopes \(r_1,\ldots,r_k\neq-1\) such that \[ |\pi_{-1}(A)| \leq \left(\sup_j|\pi_{r_j}(A)|\right)^{1+\varepsilon} \] for every finite \(A\subseteq\mathbb Z^2\). Since \(\pi_{-1}(A)\) is the difference projection, this is a sum-difference inequality. The entropy formulation states that, for finitely supported real-valued random variables \(X,Y\), there are rational \(r_j\neq-1\) such that \[ H(X-Y) \leq (1+\varepsilon)\sup_j H(X+r_jY). \] Green and Ruzsa proved the equivalence of the progression, projection, entropy, finite-field progression, and prime-divisibility formulations [1712.3381]. The finite-field progression formulation defines \(f_{k,n}(p)\) as the smallest size of a set in \(\mathbb F_p^n\) containing a \(k\)-term progression in every nonzero direction. Its conjectural asymptotic is \[ \lim_{k\to\infty}\lim_{p\to\infty} \frac{\log f_{k,n}(p)}{\log p}=n. \] For complete lines, the corresponding finite-field assertion is already known by the polynomial method; the arithmetic conjecture is stronger because it concerns bounded-length progressions before \(k\) tends to infinity. The arithmetic Kakeya conjecture implies the Euclidean Kakeya conjecture for upper Minkowski dimension. Discretizing a Besicovitch set at scale \(\delta\) produces a finite set of occupied cells containing progression-like configurations in many directions. The projection inequality then yields \[ N_\delta(K)\geq \delta^{-n+o(1)}, \] and therefore \[ \overline{\dim}_{\mathrm M}(K)=n. \] The implication has been strengthened from Minkowski to packing dimension. For finite patterns, Green, Morris, and others establish equivalences between discrete arithmetic formulations and minimum Minkowski and packing dimensions. In particular, if a set contains a \(k\)-term progression for every scaling factor in \([0,1]^n\), then its minimum Minkowski and packing dimensions coincide with the corresponding finite arithmetic quantity [2011.07056]. More generally, for a finite pattern \(U\subset\mathbb R^n\), pattern problems ask for sets containing \[ x+rU \] for every basepoint or scaling parameter in a prescribed set. These formulations include arithmetic progressions, harmonic patterns, homothetic polytopes, and finite configurations whose geometry approaches that of spheres. For a convex polytope with \(m\) faces, a set containing a homothetic copy centered at every point of \([0,1]^n\) satisfies the packing-dimension lower bound \[ \dim_P B\geq n-\frac1{n+1}. \] For simplices this is sharp, and more generally there are constructions with \[ \dim_P B\leq n-\frac1m+\epsilon. \] Specially constructed polytopes converging to the sphere have minimum pattern dimensions tending to \(n\) if and only if the arithmetic Kakeya conjecture holds [2011.07056]. For higher-dimensional arithmetic Kakeya, let \(R\subset\mathbb Q^d\) be finite and define \(\beta(R)\) as the least constant satisfying \[ H(Y_1,\ldots,Y_d) \leq \beta(R)\max_{r\in R} H(X+r_1Y_1+\cdots+r_dY_d). \] Pohoata and Zakharov introduce a homogeneous form and prove that, in one dimension, the ordinary and homogeneous constants coincide. If \(R^d\subset\mathbb Q^d\), then \[ \beta(R^d) \leq \frac{d}{\beta(R)-(\beta(R)-1)/d}. \] Consequently, an \((n,d)\)-Besicovitch set \(K\subset\mathbb R^n\) satisfies \[ \dim_M K\geq \frac{d}{\beta(R)}\,n. \] Using the Katz–Tao value \(\alpha\), the bound becomes \[ \dim_M K\geq \left(\alpha-\frac{\alpha-1}{d}\right)n, \] where \(\alpha\) is the positive solution of \[ \alpha^3-4\alpha+2=0 \] [2411.13395]. ## 6. Constructions, quantitative limits, and applications Arithmetic Kakeya sets can be substantially smaller than the conjectured scale for fixed progression length. Green and Ruzsa construct sets using a product of the first \(m\) odd primes, \[ Q=\prod_{i=1}^m p_i, \] and representatives \(x_d\) satisfying \[ x_d\equiv d^2\pmod Q. \] The union \[ S=\{x_d+jd:d\in[D]\setminus\{0\},\ j\in[k]\} \] contains a \(k\)-term progression for every \(d\in[D]\). Quadratic-residue compression gives \[ |S|\ll k^2\,2^{-m}D \prod_{i=1}^m\left(1+\frac1{p_i}\right). \] After choosing parameters, this yields \[ F_k(N)\ll k\,N^{1-c/\log\log k} \] for arbitrarily large \(N\). Thus the conjectured exponent \(1\), even if correct in the double limit, is approached slowly [1712.3381]. The same construction has geometric consequences for linear hashing. For real hashing, \[ h_a(x)=\lfloor n\{ax\}\rfloor, \] the bins partition \(\mathbb T=\mathbb R/\mathbb Z\) into \(n\) equal intervals. If a set \(A\) contains a \(k\)-term progression for every difference \(d\in[D]\), with \(D\geq kn\), then every \(n\)-element set containing \(A\) has maximum load at least \(k/2\) for every seed \(a\). Indeed, among \[ a,2a,\ldots,Da \] two points have circular distance at most \(1/D\), producing a difference \(d\) for which \(\|ad\|_{\mathbb T}\leq1/D\). The associated progression lies in an interval of length less than \(1/n\), so at least half its points occupy a single bin. The Green–Ruzsa construction therefore yields, for universes of size \(u=n^{1+o(1)}\), \[ \operatorname{maxload}_X(h_a) \geq \exp\!\left(\Omega\!\left( \frac{\log n}{\log\log n} \right)\right) \] for every real seed \(a\). A reduction transfers the same lower bound to affine modular linear hashing, even pointwise over every modular slope [2608.24866]. The arithmetic Kakeya conjecture itself remains open. A weaker polynomial-length version requires that, for every \(\eta>0\), \[ \liminf_{D\to\infty} \frac{\log F_{\lfloor D^\eta\rfloor}(D)}{\log D}\geq1. \] This weaker statement already implies the Kakeya conjecture for upper Minkowski dimension. Conversely, a uniform subpolynomial upper bound for the best-seed load in real linear hashing would imply the polynomial-length arithmetic Kakeya conjecture, and therefore the upper Minkowski-dimension Kakeya conjecture. The implication does not establish the full arithmetic conjecture because the full conjecture fixes \(k\) before sending \(D\) to infinity [2608.24866]. The theory thus contains several distinct levels of result: - **Finite fields:** complete-line Kakeya sets have cardinality comparable to the ambient space, with polynomial-method proofs. - **Finite rings:** the Kakeya conjecture is proved over \(\mathbb Z/N\mathbb Z\) for every \(N\), with explicit factorization-dependent bounds. - **Local rings:** measure-zero Kakeya sets exist in \(\mathbb F_q[[t]]^n\), while planar Kakeya sets have full Minkowski dimension. - **Integer arithmetic:** the arithmetic Kakeya conjecture is equivalent to projection, entropy, finite-field progression, and prime-divisibility statements, but remains unresolved. - **Geometric consequences:** the arithmetic conjecture would imply the Euclidean Kakeya conjecture for Minkowski and packing dimensions. - **Applications:** progression-rich arithmetic Kakeya sets generate worst-case instances for real and modular linear hashing. The common structural principle is directional completeness: a set must contain a one-dimensional pattern for every admissible direction, difference, slope, or scaling parameter. Polynomial methods exploit algebraic vanishing, tensor methods exploit Chinese-remainder decompositions, incidence methods quantify overlap, entropy methods compare projections, and digit constructions pass between finite configurations and fractal sets. These mechanisms establish rigorous bridges among arithmetic combinatorics, finite geometry, local-ring analysis, fractal dimension, and hashing, while preserving the distinction between proved finite or conditional statements and the unresolved characteristic-zero arithmetic Kakeya conjecture. Source: https://www.emergentmind.com/topics/arithmetic-kakeya-sets