---
title: Sunflower Property in Extremal Combinatorics
url: https://www.emergentmind.com/topics/sunflower-property
type: topic
---

# Sunflower Property in Extremal Combinatorics

Searching arXiv for recent and foundational papers on the sunflower property in extremal combinatorics, including set-system, restricted-intersection, random-process, and finite-vector-space variants.
In extremal combinatorics, the sunflower property is the condition that a family of sets has a common pairwise intersection. More precisely, a family \(S_1,\dots,S_r\) is an \(r\)-sunflower, or \(\Delta\)-system, if there exists a set \(K\) such that \(S_i\cap S_j=K\) for all \(i\neq j\); \(K\) is the core or kernel, and the sets \(S_i\setminus K\) are the petals. An equivalent description is that every ground-set element belongs to no set, every set, or exactly one set [2307.01374]. The sunflower property is the organizing notion behind the Erdős–Rado sunflower problem, its modern tensor- and regularity-based refinements, its finite-vector-space analogues, and several related forbidden-configuration problems [1408.3671][2605.12232].

## 1. Definition and basic structural features

For set systems, the sunflower condition is
\[
S_i\cap S_j=K \qquad \text{for all } i\neq j.
\]
The common intersection \(K\) is the core or kernel, and the petals are the differences \(S_i\setminus K\). In actual set systems, if all pairwise intersections equal \(K\), then automatically every higher intersection also equals \(K\), and there is no extra analogue of a “general position” condition to impose [2605.12232].

This formulation is stronger than a mere statement about repeated intersections. The kernel determines a canonical decomposition of each member into a shared part and a petal, and the petals are pairwise disjoint. The empty-kernel case is especially important: a matching is a sunflower with empty core. At the opposite extreme, a family of identical sets is a sunflower whose kernel is the set itself; in extremal questions one therefore works with distinct sets or distinct members of a family, depending on context [2509.16355].

The same pattern appears in several neighboring settings. In complete \(k\)-uniform hypergraphs, an \(h\)-sunflower is a family of edges whose intersection has at least \(h\) elements; this variant underlies anti-Ramsey results for rainbow subhypergraphs [1505.05170]. In the Duke–Erdős problem, the forbidden configuration is a sunflower with \(s\) petals and core of size exactly \(t-1\), which interpolates between matching problems and restricted-intersection problems [2511.17142].

## 2. Classical extremal problem and its quantitative refinements

The classical Erdős–Rado sunflower lemma states that a family \(\mathcal F\) of sets, each of cardinality at most \(s\), contains a sunflower of cardinality \(k\) whenever
\[
|{\cal F}| > (k-1)^s s!.
\]
In \(k\)-uniform notation, one may equivalently define \(f(k,s)\) as the minimum integer \(m\) such that every \(k\)-uniform family of size at least \(m\) contains a sunflower of size \(s\); then Erdős and Rado proved
\[
f(k,s)\le k!(s-1)^k,
\]
and also a lower bound of order
\[
f(k,s)\ge (s-1)^k.
\]
Their conjecture is that for every \(s>2\), there is a constant \(C=C(s)\) such that
\[
f(k,s)\le C^k
\]
[2606.02667].

A first general asymptotic improvement over the Erdős–Rado threshold was proved by Fukuyama, who showed that there exists an absolute constant \(c>0\) such that
\[
|{\cal F}| \ge (\sqrt{10}-2)^2 \left[ k \cdot \min\!\left( \frac{1}{\sqrt{10}-2}, \frac{c}{\log \min(k,s)} \right) \right]^s s!
\]
forces a sunflower of cardinality \(k\). In particular, when \(k\ge s^\epsilon\) for fixed \(\epsilon\in(0,1)\), the threshold becomes
\[
(k-1)^s s! \cdot \left[ O\!\left(\frac{1}{\log s}\right) \right]^s,
\]
which is exponentially smaller than the classical bound [1408.3671].

Subsequent work continued to reduce the base of the exponential. A 2025 result established that a family \(\mathcal F\) of \(m\)-element sets contains a \(k\)-sunflower if
\[
|\mathcal{F}| \ge \left( \frac{c k^2 \ln m}{\ln \ln m} \right)^m
\]
for some absolute constant \(c>0\), replacing a logarithmic base by a sub-logarithmic base for fixed \(k\) [2510.19037]. A separate 2026 paper claims a proof of the Erdős–Rado conjecture and states the explicit theorem
\[
f(k,s)\le (3s^2)^{6s^2-2}\cdot 2^k,
\]
which is of the conjectured form \(C(s)^k\); this is best described as a claim rather than a settled theorem in the literature [2606.02667].

## 3. Nonuniform sunflower-free families and tensor methods

A complementary line of work studies the nonuniform problem in the Boolean lattice \(2^{[n]}\), where the question is how large a sunflower-free family \(\mathcal F\subseteq 2^{[n]}\) can be. Naslund and Sawin proved that any sunflower-free family satisfies
\[
|\mathcal F|\le 3(n+1)\sum_{k\le n/3}\binom{n}{k}
\]
and hence
\[
|\mathcal F| \le \left(2^{2/3}3^{1/3}\right)^n(1+o(1)).
\]
Their argument uses the slice-rank method on the tensor
\[
T(x,y,z)=\prod_{i=1}^n \bigl(2-(x_i+y_i+z_i)\bigr),
\]
which becomes diagonal after restricting to uniform layers [1606.09575].

A 2026 improvement keeps the same exponential constant but improves the polynomial prefactor. It proves that any sunflower-free family \(\mathcal F\subseteq 2^{[n]}\) satisfies
\[
|\mathcal F|=O\!\left(n^{1/6}\left(\frac{3}{2^{2/3}}\right)^n\right),
\]
improving the earlier
\[
O\!\left(n^{1/2}\left(\frac{3}{2^{2/3}}\right)^n\right).
\]
The technical innovation is a slice-rank lemma for \(i\)-triangular tensors: if \(T:A^k\to\mathbb F\) is \(i\)-triangular with nonzero diagonal entries, then \(\operatorname{sr}(T)=|A|\). This makes it possible to work with a single global tensor ordered by set size rather than summing diagonal bounds over all uniform layers [2606.30593].

These nonuniform bounds are often summarized in terms of the sunflower-free capacity
\[
\mu_3=\limsup_{n\to\infty}F_3(n)^{1/n},
\]
for which the slice-rank approach gives
\[
\mu_3 \le 2^{2/3}3^{1/3}<2
\]
[1606.09575]. A number-theoretic reformulation from 2025 connects the general fixed-\(k\) capacity
\[
\mu_k^{\mathrm S}:=\limsup_{n\to\infty}F_k(n)^{1/n}
\]
to harmonic LCM-free sets and proves
\[
(\log N)^{\log \mu_k^{\mathrm S}-o(1)} \le f_k(N) \ll (\log N)^{\mu_k^{\mathrm S}-1+o(1)},
\]
together with the equivalence
\[
\mu_k^{\mathrm S}=2 \quad\Longleftrightarrow\quad f_k(N)=(\log N)^{1-o(1)}
\]
[2512.20055].

## 4. Restricted intersections and forbidden-core variants

The sunflower property becomes more rigid when pairwise intersections are restricted in advance. If \(\mathcal F\) is \(L\)-intersecting, meaning that \(|F_i\cap F_j|\in L\) for all distinct \(F_i,F_j\), then one can force sunflowers with significantly smaller families than in the unrestricted case. For \(d\)-intersecting families, where \(L=\{0,1,\dots,d\}\), it was proved that there exists an absolute constant \(C\) such that every \(n\)-uniform \(d\)-intersecting family contains an \(r\)-sunflower whenever
\[
|\mathcal F| > (4r)^n \,[C r \log(rd)]^d.
\]
As a consequence, for any \(C>1\), there exists \(c=c(r)>0\) such that every \(n\)-uniform \(r\)-sunflower-free family of size \(>(C4r)^n\) contains two sets whose intersection has size at least \(c\,n/\log\log n\) [2307.01374].

A different refinement is the Duke–Erdős problem, which asks for the largest family \(\mathcal F\subseteq \binom{[n]}{k}\) containing no sunflower with \(s\) petals and core of size exactly \(t-1\). Frankl and Füredi showed that for fixed \(s,k,t\), \(k\ge 2t+1\), and \(n\to\infty\),
\[
|F| \le (\phi(s,t)+o(1))\binom{n}{k-t},
\]
where \(\phi(s,t)\) is the largest size of a \(t\)-uniform sunflower-free family. A 2024 extension pushes this asymptotic picture to the regime \(n>f_0(s,t)k\) with \(f_0(s,t)\) polynomial in \(s\) and \(t\), and also proves stronger results for forbidden sunflowers with core at most \(t-1\) [2410.06156].

In the case \(t=2\), odd \(s\), \(k\ge 5\), and \(n\) sufficiently large, the extremal structure is known exactly. If \(G=K_1\sqcup K_2\) is the disjoint union of two cliques of size \(s\), then the extremal family is
\[
F= \bigg\{ F\in \binom{[n]}{k} \ \big|\ |F\cap V(G)|\ge 2 \ \text{ and }\  \forall i\in\{1,2\}\text{ we have } |F\cap V(K_i)|\neq 1 \bigg\},
\]
which lifts the exact graph extremal structure for \(\phi(s,2)\) to the forbidden-sunflower setting [2511.17142].

## 5. Finite-vector-space analogues and the split between two sunflower notions

For subspaces over finite fields, the sunflower property admits two natural analogues. Let \(F_q^n\) be the ambient vector space, and let a \(k\)-space mean a \(k\)-dimensional subspace. In the stronger, general-position notion, \(k\)-spaces \(S_1,\dots,S_s\) form a sunflower with kernel \(K\) if
\[
K=S_i\cap S_j \qquad (i\ne j)
\]
and in addition the quotient petals \(S_i/K\) are in general position, equivalently
\[
\dim(S_1+\cdots+S_s)=d+s(k-d),
\]
where \(d=\dim K\). In this setting, for \(s\ge 3\) and \(k\ge 2\), every \(s\)-sunflower-free family \(\mathcal F\) of \(k\)-spaces satisfies
\[
|\mathcal F| < \prod_{i=1}^{k}[i(s-1)]_q
< \left(\frac{q}{q-1}\right)^k q^{(s-1)\binom{k+1}{2}-k},
\]
and there are constructions of size
\[
q^{(s-1)\binom{k+1}{2}-k}
\]
for \(s\ge k+1\), based on iterated lifted MRD codes [2505.03671].

A 2026 note isolates the weaker notion that is most faithful to set systems. There, a set-like \(s\)-sunflower is defined only by the pairwise-intersection condition
\[
K=S_i\cap S_j \qquad \text{for all distinct } i,j,
\]
with no condition on the span. The distinction is substantive: the constructions of Ihringer–Kupavskii for the general-position notion do not remain sunflower-free in the set-like sense. The note gives an explicit example inside a family of \(2\)-spaces in \(F_2^5\), where three distinct \(2\)-spaces \(S_1,S_2,S_3\) satisfy
\[
S_i\cap S_j = T \qquad \text{for all } i\neq j
\]
with
\[
\dim(S_1+S_2+S_3)=3<4,
\]
so they form a set-like \(3\)-sunflower but not a general-position sunflower [2605.12232].

The same paper also gives the first systematic construction tailored to the set-like problem. Let \(n\ge 2\ell+1\), \(k=n-\ell\), let \(\mathcal C\subseteq F_q^{\ell\times \ell}\) be the matrix representation of \(F_{q^\ell}\) over \(F_q\), and define
\[
\mathcal G=\left\{\operatorname{rowspace}[\,I\mid A\mid [A^2]_1\mid \mathbf 0\,]:A\in\mathcal C\right\}.
\]
Then
\[
\mathcal F=\{U^\perp:U\in\mathcal G\}
\]
is a set-like \(s\)-sunflower-free family of \(k\)-spaces of size
\[
|\mathcal F|=q^\ell=q^{\,n-k}
\]
for every \(s\ge 3\). The mechanism is that for distinct \(U,V,W\in\mathcal F\),
\[
\dim(U\cap V)=n-2\ell,\qquad \dim(U\cap V\cap W)=n-2\ell-1,
\]
so pairwise intersections are one dimension larger than triple intersections, which excludes a set-like sunflower [2605.12232].

## 6. Methods, random processes, and structural interpretations

Modern work on the sunflower property is methodologically diverse. One influential line passes through regularity and Boolean complexity. A 2019 paper introduced a structure-vs-pseudorandomness framework for set systems, defining \(\kappa\)-regularity by the condition
\[
\Pr_{S\sim D}[T\subseteq S] \le \kappa^{-|T|}
\]
for every \(T\subseteq X\), and showed that improved monotone DNF compression would imply improved sunflower theorems. Under a weaker upper-bound compression conjecture, it derives that for every \(r\ge 3\) there exists \(c_r\) such that any \(w\)-set system of size
\[
|F| \ge (\log w)^{c_r w}
\]
contains an \(r\)-sunflower [1903.00580].

A different regularity condition was developed for the 3-petal problem through the \(\Gamma(b)\)-condition
\[
|[S]| < b^{-|S|}|\mathcal F|
\]
for every nonempty \(S\). For each \(\delta\in(0,1/2)\), there exists \(c>0\) such that a family \(\mathcal F\) of \(m\)-element sets contains three mutually disjoint sets whenever it satisfies
\[
\Gamma(c m^{\frac12+\delta}),
\]
and therefore contains a 3-sunflower whenever
\[
|\mathcal F|>(c m^{\frac12+\delta})^m.
\]
The proof uses extension generators, weighted \(\Gamma_g\)-conditions, and recursive splitting arguments [1809.10318].

The random greedy viewpoint leads to the sunflower-free process. Starting from the empty family and repeatedly adding a uniformly random \(w\)-set that does not create an \(r\)-sunflower, one obtains a random maximal sunflower-free family. When \(w=n^\alpha\) with fixed \(\alpha\in(1/2,1)\), the process on an \(n\)-element universe produces a \(w\)-uniform \(r\)-sunflower-free family of size
\[
\Omega\!\left(\left(\frac{w^2}{n}\right)^{\frac1{r-1}} N D^{-\frac1{r-1}}\right),
\]
where
\[
N=\binom nw
\]
and \(D\) is the number of sunflowers containing a fixed \(w\)-set. The analysis extends Bennett–Bohman’s theory of random greedy independent-set processes to hypergraphs with a sparse set of bad pairs of very large codegree [2509.16355].

The sunflower property also appears in anti-Ramsey form. If the edges of a complete \(k\)-uniform hypergraph are colored so that any monochromatic \(h\)-sunflower has at most \(\lambda\) petals, then sufficiently large vertex sets contain rainbow complete subhypergraphs; quantitatively, there exists a constant \(c_k\) depending only on \(k\) such that
\[
|V|\geq c_k \lambda n^{\frac{2k-1}{k-h}}\log(n)^{-\frac{1}{k-h}}
\]
implies the existence of a rainbow subset of size at least \(n\) [1505.05170].

Across these developments, a recurring theme is that the sunflower property sits at the boundary between local intersection data and global structure. In the set-theoretic setting, identical pairwise intersections already force the full sunflower geometry. In vector spaces, the failure of inclusion–exclusion for sums of three or more subspaces splits the theory into general-position and set-like variants. In random, tensor, and regularity-based methods, the core problem is to quantify when local sparsity or pseudorandomness forces the emergence of a common kernel.

Source: https://www.emergentmind.com/topics/sunflower-property