TheoremComplete

Frankl-Wilson Theorem

The Frankl-Wilson theorem provides strong upper bounds on set families with restricted intersection sizes. It is proved using the linear algebra (dimension) method and has remarkable applications to geometry, coding theory, and combinatorial number theory.


Statement

Theorem6.3Frankl-Wilson Theorem

Let pp be a prime and let L={l1,…,ls}βŠ†{0,1,…,pβˆ’1}L = \{l_1, \ldots, l_s\} \subseteq \{0, 1, \ldots, p-1\}. If F\mathcal{F} is a family of subsets of [n][n], each of size kk where k≑̸li(modp)k \not\equiv l_i \pmod{p} for any ii, and ∣F∩G∣(modp)∈L|F \cap G| \pmod{p} \in L for all distinct F,G∈FF, G \in \mathcal{F}, then ∣Fβˆ£β‰€(ns).|\mathcal{F}| \leq \binom{n}{s}.


Proof Outline

Proof

Associate to each set F∈FF \in \mathcal{F} the polynomial fF(x)=∏l∈L(βˆ‘i∈Fxiβˆ’l)∈Fp[x1,…,xn].f_F(x) = \prod_{l \in L} \left(\sum_{i \in F} x_i - l\right) \in \mathbb{F}_p[x_1, \ldots, x_n].

This polynomial has degree ∣L∣=s|L| = s. For distinct F,G∈FF, G \in \mathcal{F}, evaluating fFf_F at the characteristic vector vGv_G gives fF(vG)=∏l∈L(∣F∩Gβˆ£βˆ’l)≑0(modp)f_F(v_G) = \prod_{l \in L}(|F \cap G| - l) \equiv 0 \pmod{p} since ∣F∩Gβˆ£β€Šmodβ€Šp∈L|F \cap G| \bmod p \in L.

For F=GF = G: fF(vF)=∏l∈L(kβˆ’l)≑̸0(modp)f_F(v_F) = \prod_{l \in L}(k - l) \not\equiv 0 \pmod{p} since kβ€Šmodβ€Špβˆ‰Lk \bmod p \notin L.

Now reduce each polynomial modulo the ideal (xi2βˆ’xi)(x_i^2 - x_i) (since vGv_G has 0/10/1 entries) to obtain polynomials of degree at most ss in the reduced polynomial ring. The monomials of degree at most ss in this ring span a vector space of dimension βˆ‘j=0s(nj)\sum_{j=0}^s \binom{n}{j}. But the matrix of evaluations (fF(vG))F,G(f_F(v_G))_{F,G} has nonzero diagonal and zero off-diagonal, so the corresponding vectors are linearly independent. We conclude: ∣Fβˆ£β‰€βˆ‘j=0s(nj).|\mathcal{F}| \leq \sum_{j=0}^s \binom{n}{j}.

For the uniform version (all sets of the same size), a refinement gives the stronger bound ∣Fβˆ£β‰€(ns)|\mathcal{F}| \leq \binom{n}{s}. β–‘\square

β– 

Applications

ExampleKahn-Kalai Counterexample to Borsuk's Conjecture

Borsuk conjectured (1933) that every bounded set in Rd\mathbb{R}^d can be partitioned into d+1d + 1 sets of smaller diameter. Kahn and Kalai (1993) disproved this using the Frankl-Wilson theorem: they constructed a set in Rd\mathbb{R}^d requiring (1.2)d(1.2)^{\sqrt{d}} parts, exponentially more than d+1d + 1.

RemarkRay-Chaudhuri-Wilson Theorem

The Frankl-Wilson theorem generalizes the Ray-Chaudhuri-Wilson theorem: if FβŠ†([n]k)\mathcal{F} \subseteq \binom{[n]}{k} and the intersection sizes ∣F∩G∣|F \cap G| take at most ss distinct values for Fβ‰ GF \neq G, then ∣Fβˆ£β‰€(ns)|\mathcal{F}| \leq \binom{n}{s}. This follows from a dimension argument over R\mathbb{R} using the same polynomial construction.

ExampleApplication to Chromatic Numbers

The Frankl-Wilson theorem implies that the fractional chromatic number of the Kneser graph K(n,k)K(n,k) is at least n/kn/k, complementing the integer chromatic number result from LovΓ‘sz's topological proof. This bound is tight, establishing Ο‡f(K(n,k))=n/k\chi_f(K(n,k)) = n/k.