Fixed Pattern Densities in Combinatorics

• Fixed pattern densities are defined as normalized counts of specific small patterns in combinatorial objects like sequences, permutations, and graphs.
• Key methodologies include entropy maximization, flag-algebra products, and variational principles to capture structure and phase transitions in constrained systems.
• Applications span from enumeration and limit shape analysis to quasirandomness characterizations, opening avenues in extremal combinatorics and geometric analysis.

A fixed pattern density quantifies the frequency of occurrence of a prescribed small pattern—typically a finite ordered tuple, subgraph, or geometric configuration—within a larger combinatorial or geometric object, under the constraint that the count or asymptotic density of the given pattern is specified. The study of fixed pattern densities is foundational in extremal combinatorics, permutation and graph limits (permuton and graphon theory), additive and geometric combinatorics, and statistical physics-inspired models, enabling both enumeration and structure theory for constrained systems.

1. Definitions and Model Frameworks

For a combinatorial object (sequence, permutation, graph), the pattern density is defined as the (normalized) number of occurrences of a specified pattern $P$:

• Binary sequences: For $s\in\{0,1\}^n$ and $P\in\{0,1\}^k$, the (consecutive) pattern density is

$$d_n(P,s) = \frac{\#\{\,1\le i\le n-k+1:\;(s_i,\dots,s_{i+k-1})=P\}}{n-k+1}.$$

For patterns as subsequences $w$,

$$\rho_w(X) = \frac{N_w(X)}{\binom n m}, \quad N_w(X) = \#\{\,(1\le i_1 < \cdots < i_m \le n): (x_{i_1},\dots,x_{i_m}) = w\}$$

(Kenyon, 7 Jan 2026).

• Permutations: For $\sigma\in S_n$ and pattern $\pi\in S_k$,

$\rho_\pi(\sigma) = \frac{\#\{\text{$k$-subsets forming$\pi$}\}}{\binom n k}.$

In the limit ($n\to\infty$), pattern densities determine the weak closure of permutation sequences (permutons) (Kenyon et al., 2015, Khovanova et al., 2018, Garbe et al., 2023).

• Graphs: For simple graphs $G$ on $n$ vertices and a $k$-vertex graph $K$,

$$t_K(G) = \frac{\#\{\text{homomorphisms } K \to G\}}{n^{|V(K)|}}$$

and for the dense limit, pattern densities correspond to subgraph densities of graphons (Neeman et al., 2021).

The feasible region—the set of all attainable pattern density vectors for a given size—encodes structural constraints.

2. Geometry and Algebraic Structure of Feasible Regions

For permutations, the feasible region of pattern densities up to size $k$ is

$\mathcal{R}_k = \left\{ x \in [0,1]^{\,_k} : \exists\text{ permuton } \Pi\text{ with } x_\sigma = d(\sigma,\Pi), \ \forall \sigma \in {_k} \right\},$

with total probability constraints $\sum_{\sigma\in {_m}} d(\sigma,\Pi) = 1$ for $1\le m\le k$ (Garbe et al., 2023).

The affine dimension of $\mathcal{R}_k$ equals the number of non-trivial Lyndon permutations of size at most $k$:

$\dim \mathcal{R}_k = |\ ^L_k |,$

where $^L_k$ denotes the set of non-trivial Lyndon permutations. Any additional pattern density is an algebraic (flag-algebraic) function of Lyndon pattern densities; there are no further linear relations beyond normalization.

In graph analogues, the basis is given by non-trivial connected graphs (Garbe et al., 2023). In permutations, the naive guess that indecomposable patterns suffice fails: for $k=3$, there are only 3 indecomposable permutations, but $\dim\mathcal{R}_3=5$, matching the count for non-trivial Lyndon permutations.

Key technical elements:

• Unique direct sum decomposition into Lyndon blocks for permutations.
• The flag-algebra product:

$$d(\pi_1 \times \cdots \times \pi_n, \Pi) = \prod_{i=1}^n d(\pi_i, \Pi).$$

• Shuffle lemma for Lyndon words to guarantee algebraic independence.

3. Extremal and Variational Principles for Fixed Pattern Densities

Imposing constraints on pattern densities leads to extremal and variational problems:

• Permutations: Maximizing entropy over the space of permutons subject to fixed pattern densities captures both typical structure and large deviation asymptotics (Kenyon et al., 2015). The entropy functional is

$$H(\mu) = -\iint_\mathbf{Q} g(x, y) \log g(x, y)\,dx\,dy$$

for $g$ the Lebesgue density of $\mu$, with variational constraints $\rho_{\pi_i}(\mu) = d_i$.

• Sequences: For binary sequences and a finite set of subsequence constraints, the maximizing measure $f(x)$ is determined via Lagrange multipliers and can be explicitly characterized for many patterns (Kenyon, 7 Jan 2026).
• Graphs: The graphon variational problem for fixed subgraph densities and its Lagrangian duality have a direct combinatorial-analytic formulation (Neeman et al., 2021).

These variational problems frequently exhibit uniqueness of optimizers in the interior of feasible regions and phase transitions (singularities) on the boundary.

4. Pattern Densities in High-Density Sets and Geometric Combinatorics

In Euclidean settings and high-density sets, fixed pattern densities translate to structural guarantees for large configurations:

• For $E\subset \mathbb{R}^d$ with positive upper density, the classical unpinned results assert affine copies of every fixed $k$-pattern at large scales (Durcik et al., 2018). For pinned pattern densities, precise distinctions arise:
  • For $k=2$ (pinned distances), there exist dense sets $E$ such that no single $x\in E$ sees all large distances, but every $x$ sees a positive upper density of distances (Wang, 1 Sep 2025).
  • For $k\ge3$, no single $k$-pattern can be forced at every pin for all $E$, but there always exists a small catalog of patterns $\mathcal{V}$ such that, for every $x$ in $E$, a member $V\in\mathcal{V}$ appears with positive density at $x$ (Wang, 26 Oct 2025).

These geometric density results rely on a combination of harmonic analytic, probabilistic, and combinatorial (particularly Gowers-norm) machinery.

5. Special Constructions: Permuton Inflations and Quasirandomness

Inflation/tensor-product constructions of permutations provide fine control over induced pattern densities:

• The limit density of a pattern $\pi$ in the inflation $\tau[\sigma_m]$ can be decomposed via block-decomposition sums involving pattern densities in both $\tau$ and $\sigma_m$ (Khovanova et al., 2018).
• $k$-Inflatable permutations: For $k=2$, all permutations with $t(12, \tau) = 1/2$ are 2-inflatable. For $k\ge4$, only the fully random case is possible. For $k=3$, there is a nontrivial classification via a system of equations on small pattern densities that have been algorithmically enumerated (Khovanova et al., 2018).

Quasirandomness (uniformity) characterizations: In the context of permutations, certain linear combinations (or sums) of fixed $k$-pattern densities correspond precisely to the property of full asymptotic randomness. For $k=4$, only ten special subsets of patterns suffice to force quasirandomness by sum constraints, aligned with independence test statistics in nonparametric statistics (Chan et al., 2019).

6. Applications, Explicit Constructions, and Open Problems

Applications include enumeration of large constrained structures, analytic determination of limit shapes, identification of phase transitions, and precise understanding of the degree of freedom for the feasible region of densities (Kenyon et al., 2015, Kenyon, 7 Jan 2026, Garbe et al., 2023).

Small-$k$ examples and generating series are explicitly given for Lyndon permutations, enabling concrete enumeration of independent pattern densities:

• For $k=3$: 5 Lyndon permutations ($21, 132, 231, 312, 321$), so $\dim \mathcal{R}_3=5$.
• For $k=4$: 21 Lyndon permutations, so $\dim \mathcal{R}_4=21$, with the remainder of the 33 pattern densities functions thereof (Garbe et al., 2023).

Open problems (see (Kenyon, 7 Jan 2026, Kenyon et al., 2015, Garbe et al., 2023)):

• Analytic structure and algebraicity of feasible regions.
• Extension to larger alphabets and higher dimensions.
• Complete classification of phase transition phenomena and boundary singularities in the variational problems.
• Optimal catalogs for geometric pattern densities in Euclidean space at fixed density thresholds.

7. Summary Table: Algebraic Basis for Pattern-Density Regions

Structure	Free coordinates	Linear dependencies
Graphs	Non-trivial connected subgraphs	Subgraph count normalizations
Permutations	Non-trivial Lyndon permutations	Pattern-count sum-to-one (sizes)
Sequences	Conjectural: certain patterns/words	Known normalization constraints

The algebraic basis provided by non-trivial Lyndon permutations in the permutation setting, or by connected graphs in the graph setting, canonically parameterizes the feasible region of pattern densities up to size $k$ and enables an explicit global description of allowable pattern statistics (Garbe et al., 2023).