---
title: Erdős-Pósa Property for Graph Subdivisions
url: https://www.emergentmind.com/topics/erdos-posa-property-for-subdivisions
type: topic
---

# Erdős-Pósa Property for Graph Subdivisions

The Erdős-Pósa property for subdivisions lies at the intersection of topological graph theory and combinatorial optimization, encapsulating a min-max duality between the existence of disjoint models of a fixed graph $H$ and the size of vertex or edge sets intersecting all such models. This property, foundational for both structural theory and algorithmic applications, admits rich variant formulations—vertex, edge, induced, directed, and infinite-cardinal—each displaying subtle thresholds that depend on the pattern graph $H$. Recent advances include results on localizing hitting sets within prescribed substructures, full dichotomies for induced subdivisions, and fine structural theorems for edge and digraph analogues.

## 1. Formal Definitions and General Framework

Given a finite simple graph $H$, an **$H$-subdivision** in a host graph $G$ is a subgraph $\widetilde H \subseteq G$ equipped with a mapping $\phi: V(H)\cup E(H) \to V(\widetilde H)\cup\{\text{paths in }\widetilde H\}$ such that vertices are taken to distinct vertices (branch vertices), and each edge $e=\{u,v\}\in E(H)$ is replaced by a simple path between $\phi(u)$ and $\phi(v)$, with the property that paths for distinct edges are internally vertex-disjoint and $\widetilde H$ is the union of all images.

Let $T(H)$ denote the class of all graphs admitting an $H$-subdivision.

The **vertex-Erdős–Pósa property** for subdivisions of $H$ asserts the existence of a function $f_H:\mathbb N\to\mathbb N$ (the *gap function*) such that for every graph $G$ and $k\geq 1$, $G$ either contains $k$ vertex-disjoint $H$-subdivisions or admits a vertex-set $X\subseteq V(G)$ with $|X|\leq f_H(k)$ such that $G-X$ is $H$-subdivision-free [1603.04615].

The **edge-Erdős–Pósa property** replaces vertices by edges: for every $G$ and $k$ either $G$ has $k$ edge-disjoint $H$-subdivisions or $X\subseteq E(G)$ of $|X|\leq f(k)$ meets all subdivisions.

The **induced Erdős–Pósa property** requires the $H$-subdivisions be induced subgraphs [1803.07581].

## 2. Classical Dichotomies and Bounds

The classical Erdős–Pósa theorem concerns cycles ($H=K_3$), showing cycles have the vertex (and for some regimes, edge) Erdős–Pósa property with $f_{K_3}(k)=O(k\log k)$. Thomassen's dichotomy [1603.04615] established that for subdivisions, $T(H)$ has the vertex-Erdős–Pósa property **if and only if** $H$ is planar. In particular:
- For non-planar $H$, the packing-covering gap is unbounded; there exist $G$ with $v\text{–pack}_{T(H)}(G)=1$ but $v\text{–cover}_{T(H)}(G)$ arbitrarily large.
- For planar $H$, both existential and near-optimal bounds are known: $f_H(k)=O_h(k\log k)$ and $f_H(k)=O(h^{O(1)}k\operatorname{polylog}k)$ [1603.04615].

## 3. Localized Erdős–Pósa Theorems for Subdivisions

A significant refinement is the **localized Erdős–Pósa property** for subdivisions [2512.21530]. Let $H$ have $n$ vertices and $m$ edges and suppose $H$ has the Erdős–Pósa property for subdivisions with bounding function $f_H$. For any $G$ with no $k+1$ vertex-disjoint $H$-subdivisions, the main theorem guarantees:
- There exist $0\leq\ell\leq k$ vertex-disjoint $H$-subdivisions $H_1,\ldots,H_\ell\subseteq G$ and $X\subseteq \bigcup V(H_i)$ with
  \[
     |X|\leq 2^{f_H(k)}mk + k(m-n)
  \]
  such that $G-X$ contains no $H$-subdivision.

This *localizes* the hitting set inside the union of up to $k$ prescribed subdivisions, sharpening the classical statement where the hitting set could be supported outside any particular set of $k$ models. The proof employs an induction on $k$, data structures tracking "hitting triples" $(S,X,Y)$, an explicit score function, and iterated applications of Menger-type connectivity to bound the size of $X$.

For subcubic forests and subcubic planar $H$, sharper, polynomial bounds are possible; for example, for a subcubic tree $H$ on $n$ vertices,
\[
|X|\leq 2^{nk}(n-2)k.
\]
[2512.21530]

## 4. Edge and Induced Versions: Structure and Limitations

For **edge-disjoint** subdivisions, the edge-Erdős–Pósa property is more restrictive:
- Cycles and long cycles admit the property with $f(k,\ell)=210k^2\log k + 10\ell(k-1)$ for cycles of length at least $\ell$ [1607.01903].
- $K_4$-subdivisions (i.e., subdivisions of the complete graph $K_4$) have the edge-Erdős–Pósa property with an explicit bound $O(k^8\log k)$ [1808.10380].
- For most $H$ with complex topology (notably, planar graphs with more than three high-degree vertices or unbounded treewidth), the edge-Erdős–Pósa property fails. Techniques such as frame+hub decompositions, series-parallel reductions, and modular induction yield upper bounds only in special cases.

The **induced Erdős–Pósa property** for $H$-subdivisions admits a full classification [1803.07581]:
- For forests, induced $H$-subdivisions have the property iff every component has at most one vertex of degree at least $3$.
- For complete bipartite graphs $K_{n,m}$, the property holds iff $n\leq 1$ or $m\leq 2$.
- For cycles, only $C_\ell$-subdivisions with $\ell\leq 4$ have the induced property.
- Some small patterns admit the property (diamond, $1$-pan, $2$-pan), always with polynomial upper bounds, e.g., $f(k)=O(k^2\log k)$.

For all other cases (cycles of length $\geq 5$, $K_{2,r}, r\geq 3$, forests with two high-degree vertices in a component, non-planar $H$), "negative templates" construct graph families with maximum packing number $1$ and hitting set size growing with parameter $n$. The underlying obstacles are the forced overlap of subdivisions and the replication of many isomorphic configurations [1803.07581].

## 5. The Infinite and Directed Settings

In the context of infinite graphs and infinite-cardinal generalizations, results extend via compactness and well-quasi-ordering machinery [2411.02561]. For any (possibly infinite) tree $T$, the class $(T)$ of graphs containing $T$ as a topological minor satisfies:
- The $\kappa$-Erdős–Pósa property for all uncountable $\kappa$;
- If $T$ is rayless (contains no one-way infinite path), then also the classical and $\aleph_0$-Erdős–Pósa property (finite cover for finite packing);
- Bounds $f(k)=|G(U)|(k-1)$ are available in the finite/rayless case, with $|G(U)|$ reflecting the size of the appropriately closed portion of the underlying tree-decomposition.

The extension relies on the well-quasi-ordering of trees under topological minors (Nash-Williams/Laver) and the ability to recursively build "hordes" of embeddings or find small closures forcing the compactness argument [2411.02561].

In **directed graphs**, the first nontrivial acyclic pattern shown to admit an Erdős–Pósa property is the tripod (subdivision of a digraph consisting of two sources feeding into a common center, then out to a sink) [2408.16733]. The proof uses onion-harvesting lemmas, matroid intersection for separation lemmas, and a Ramsey-theoretic framework, yielding large (probably tower-type) bounds $f(k)$. There is, however, no such property known for more general arborescences.

## 6. Structural Techniques and Proof Schemes

Modern proofs for the Erdős–Pósa property employ tree- and branch-decomposition arguments, recursive separation schemes, and connections to grid-minor theorems:
- For planar $H$, the Chekuri–Chuzhoy grid-minor theorem yields a ceiling on treewidth sufficient to guarantee large packings, combining with induction for small treewidth cases to glue the result [1603.04615].
- Bounding functions for packings and coverings in bounded-treewidth graphs reduce to combinatorial separation and branching arguments.
- Induced and edge variants often require careful tracking of unique decomposition templates and exploit specific forbidden substructures (ears, bridges, frames, shadows, modules).

A typical structure is as follows:

| Property      | Known Positive Cases         | Gap Function Order       |
|---------------|-----------------------------|-------------------------|
| Vertex-EP     | planar $H$                  | $O(k\log k)$, $O(k\operatorname{polylog}k)$ [1603.04615] |
| Edge-EP       | cycles, long cycles, $K_4$-subdivisions        | polynomial or $O(k^8\log k)$ [1607.01903, 1808.10380] |
| Induced-EP    | forests (at most $1$ high-degree per component), $K_{n,m}$ with $n\leq1$ or $m\leq2$, small patterns | $O(k^2\log k)$ or polynomial [1803.07581] |

## 7. Open Directions and Conjectures

Several central challenges remain open:
- Characterization of all $H$ such that subdivisions or immersions of $H$ admit the edge-Erdős–Pósa property.
- Sharpening quantitative bounds for the gap function for various classes, particularly the removal (or optimality) of $\log k$ factors for outerplanar or bounded-treewidth $H$ [1603.04615].
- Directed analogues for arborescences and more complex patterns beyond the tripod [2408.16733].
- Algorithmic implications: for positive results (especially induced/vertex cases), bounding functions translate into FPT algorithms for disjoint packing or small hitting set detection, with polynomial-time implementations for specific template classes.

The emerging theme is the tightly coupled interplay between the topological properties of $H$ (planarity, treewidth, degree structure), the structural decomposition of graphs (tree- and branch-decomposition, well-quasi-ordering), and the type (vertex, edge, induced, infinite) of Erdős–Pósa duality under consideration. The property, in all its incarnations, remains a central tool for unifying probabilistic, structural, and algorithmic graph theory.

Source: https://www.emergentmind.com/topics/erdos-posa-property-for-subdivisions