---
title: Localization Capture Time Conjecture
url: https://www.emergentmind.com/topics/localization-capture-time-conjecture
type: topic
---

# Localization Capture Time Conjecture

Searching arXiv for recent papers on the graph-theoretic Localization Capture Time Conjecture.
The Localization Capture Time Conjecture is a conjecture in the graph-theoretic localization game, a pursuit–evasion model in which an invisible robber moves on a connected graph while cops probe vertices and receive graph-distance information. In this setting, the central parameters are the localization number $\zeta(G)$, the minimum number of cops needed to force eventual identification of the robber, and the localization capture time, the minimum number of rounds required under optimal play. The conjecture, proposed by Behague et al. in 2022, asserts that there exists a constant $C$ such that every connected $n$-vertex graph $G$ satisfies $\lcapt(G)\le Cn$; equivalently, one may ask whether even $\lcapt(G)\le n$ always holds [2508.10443]. It is framed as the localization analogue of Meyniel’s conjecture for ordinary Cops and Robber capture time and connects distance-probing games to structural graph theory, metric dimension, and pursuit–evasion [2508.10443].

## 1. Formal game model and core parameters

Let $G=(V,E)$ be a finite, connected graph. In the localization game, two players control an invisible robber and $k$ cops. The robber first chooses a start-vertex $r_0\in V$. In each round $t=1,2,\dots$, the cops simultaneously choose probe vertices $c_1^t,\dots,c_k^t\in V$, and receive the distance vector
\[
\bigl(d(c_1^t,r_{t-1}),\,\dots,\;d(c_k^t,r_{t-1})\bigr),
\]
where $d(x,y)$ denotes graph distance. The robber then optionally moves along one edge or stays put, producing a new position $r_t$. The cops win if, after some round $T$, the distances observed force the robber’s location to be a single vertex; the robber wins by avoiding capture forever. The standard assumption is that the robber is omniscient and knows the cops’ strategy [2508.10443].

The localization number $\zeta(G)$ is the minimum $k$ for which $k$ cops have a winning strategy. For $k\ge \zeta(G)$, the $k$-localization capture time $\lcapt_k(G)$ is the minimum number of rounds in which $k$ cops can force a win against a best-play robber, and the localization capture time is
\[
\lcapt(G)=\lcapt_{\zeta(G)}(G).
\]
An earlier formulation uses the notation $\capt_{\zeta,k}(G)$ and $\capt_\zeta(G)$ for the same capture-time concept [2105.09806]. The conjectural statement is therefore asymptotic in the graph order and concerns optimal play with the minimum number of cops.

The conjecture is motivated by the absence of known counterexamples with superlinear capture time. In the 2021 formulation, this was summarized as the empirical observation that no graph is known whose capture time exceeds a constant multiple of its order [2105.09806]. This suggests that the problem is not merely one of existence of winning strategies, but of quantitative efficiency.

## 2. Origin and significance of the conjecture

The explicit conjecture appears in the form: there exists a constant $c>0$ such that for every $n$-vertex graph $G$,
\[
\capt_\zeta(G)\le c\,n,
\]
or equivalently,
\[
\lcapt(G)\le Cn,
\]
depending on notation [2105.09806; 2508.10443]. In the later account, Behague et al. are identified as having proposed the conjecture in 2022 [2508.10443].

Its significance is twofold. First, a positive answer would complete, in the localization model, the analogue of Meyniel’s conjecture for ordinary Cops and Robber capture time [2508.10443]. Second, the conjecture places metric information at the center of pursuit–evasion: the cops do not occupy blocking positions but instead accumulate distance constraints, so proofs typically combine decomposition arguments, metric distinguishability, and monotonic shrinking of candidate sets.

A common misconception is to conflate the conjecture with bounds on the localization number alone. The conjecture does not assert that $\zeta(G)$ is linear in $n$; it asserts that once the minimum winning number of cops is available, the number of rounds required to force localization is $O(n)$. This distinction is visible in planar graphs, where $\zeta$ is unbounded and the capture-time conjecture is described as wide open [2508.10443]. Thus bounded or exactly known localization number is often a prerequisite for current capture-time techniques, but not itself the conjectural conclusion.

## 3. Established graph families and baseline bounds

The earliest broad evidence came from trees and interval graphs. For trees, it is known that $\zeta(T)\le 2$ for every tree $T$, and the 2021 result shows directly that if $T$ is a tree of order $n$, then $\capt_\zeta(T)\le n$; hence trees satisfy the conjecture [2105.09806]. The proof sketch separates the cases $\zeta(T)=1$ and $\zeta(T)=2$: in the former, Seager’s one-cop strategy isolates the robber in at most $n$ probes, and in the latter, one cop remains at a root while the second probes branches, with each round eliminating at least one vertex [2105.09806].

For interval graphs, the same paper states that $\zeta(G)=\pw(G)$ and $\capt_\zeta(G)\le n$ [2105.09806]. The strategy follows an optimal path decomposition, using $\pw(G)$ cops to occupy all vertices except one in each bag successively, preventing the robber from backtracking.

The later survey of known bounds identifies several additional classes. For trees, Behague et al. had previously given $\lcapt(T)\le 5n$ if $\zeta(T)=1$ and $\lcapt(T)\le n-1$ if $\zeta(T)=2$ [2508.10443]. For outerplanar graphs, Bonato–Kinnersley proved $\zeta(G)\le 2$ for all outerplanar $G$, and Behague et al. deduced $\lcapt(G)\le n-1$ when $\zeta(G)=2$ [2508.10443]. Interval graphs and complete multipartite graphs satisfy $\lcapt\le n$ or $n-1$, and for diameter-2 graphs some exact values are known [2508.10443].

The 2021 paper also develops bounds for incidence graphs of projective planes and for general graphs via treewidth [2105.09806]. If $G$ is the incidence graph of a projective plane of order $q$, then with $k\ge q+1$,
\[
\capt_{\zeta,k}(G) \le \Bigl\lceil\frac{q-1}{k-q}\Bigr\rceil+\Bigl\lceil\frac{q}{k-q+1}\Bigr\rceil,
\]
and in particular for $k=q+1$,
\[
\capt_\zeta(G)\le (q-1)+\Bigl\lceil\frac q2\Bigr\rceil,
\]
so $\capt_\zeta(G)=O(q)=O(\sqrt n)$ [2105.09806]. Since this is sublinear in $n$, it is consistent with the conjecture and shows that capture time can be substantially smaller than linear in structured families.

## 4. Improved bounds for trees

A major advance is given in “Localization game capture time of trees and outerplanar graphs” [2508.10443]. Let $T$ be a tree on $n$ vertices and let $\ell(T)$ be its number of leaves. The paper proves two refined theorems.

If $\zeta(T)=1$, equivalently $T$ contains no $T_{3,3}$, then
\[
\lcapt(T)\le \ell(T)\le n-1.
\]
If $\zeta(T)=2$, so $T$ contains $T_{3,3}$, then
\[
\lcapt(T)\le \Bigl\lfloor\frac{\ell(T)}2\Bigr\rfloor -1
\le \Bigl\lfloor\frac n2\Bigr\rfloor -3.
\]
Consequently, in all cases,
\[
\lcapt(T)\le
\begin{cases}
n-1, & \zeta(T)=1,\\
\lfloor n/2\rfloor -3, & \zeta(T)=2.
\end{cases}
\]
These results significantly improve the previously known upper bounds for trees [2508.10443].

The proof outlines are structurally informative. For $\zeta(T)=1$, the tree is rooted at a leaf $v_0$, and the single cop “drills down” one level by probing leaves or degree-2 children of the current root $v_t$, using at most as many probes as leaves in the pruned branches. Two cases, according to whether the current root has at most one or exactly two high-degree children, allow elimination of all but one branch in at most $\ell(T)$ total probes [2508.10443].

For $\zeta(T)=2$, the argument uses a pairing strategy of branches. In the first move, the cops eliminate all but three high-leaf branches around a $T_{3,3}$ core. Each subsequent probing discards at least two leaves from the candidate set, yielding capture in $\lfloor \ell(T)/2\rfloor-1$ rounds [2508.10443]. This suggests that leaf count, rather than only total order, is the operative complexity parameter for trees.

The structural role of the “double-tripod” $T_{3,3}$ is also explicit. It characterizes the transition between one-cop and two-cop trees: it is classical that $\zeta(T)\le 2$ for every tree, and equality holds exactly when $T$ contains $T_{3,3}$ [2508.10443]. Thus the improved capture-time bounds refine not just the asymptotic estimate but the dependence on the tree’s branching geometry.

## 5. Outerplanar graphs and block-based strategies

Outerplanar graphs form a second principal test bed because their localization numbers are known exactly and their sparse structure permits explicit strategies [2508.10443]. Two results are stated for the case $\zeta(G)=2$.

For an arbitrary outerplanar graph whose blocks are $B_1,\dots,B_m$,
\[
\lcapt_2(G)\le \sum_{i=1}^m (|B_i|-1)\le n-1,
\]
achieved by adding one new vertex of the cop-territory per round using the Bonato–Kinnersley strategy [2508.10443]. This gives a decomposition-sensitive bound in terms of the block structure.

For a 2-connected outerplanar graph with $c$ interior chords,
\[
\lcapt_2(G)\le c+1\le n-2.
\]
The proof sketch labels the outer cycle $v_1,\dots,v_n$ and probes midpoints of the longest induced cycle to confine the robber to an arc with fewer chords. Two endpoints $v_L,v_R$ of the current cop territory are maintained, together with known distances $d_L,d_R$. In each case, depending on whether $v_L$ and $v_R$ share a neighbor, one chord is discarded per probe by comparing distances and using inequalities of the form $d(\cdot)\le d(\cdot)\pm 1$; since there are $c$ chords, capture occurs in at most $c+1$ rounds [2508.10443].

These outerplanar results show that the conjectured linear bound is not merely a consequence of bounded order versus bounded cop number. The arguments are highly sensitive to graph decomposition: block-cut structure in general outerplanar graphs, and chord count in the 2-connected case. A plausible implication is that similar refinement by decomposition parameters may be necessary for broader sparse families such as bounded-treewidth graphs.

## 6. General frameworks: monotonicity, treewidth, and coloring structures

Beyond individual graph classes, the literature develops general-purpose frameworks for understanding capture time. One such result concerns monotonicity on induced subgraphs. In general, capture time need not be monotone under induced subgraphs, but it does hold for “special” induced subgraphs, defined by the condition that every path in $G$ between two vertices of $H$ is entirely contained in $H$ [2105.09806]. If $H$ is special in $G$, then for all $k\ge \zeta(G)$,
\[
\capt_{\zeta,k}(H)\le \capt_{\zeta,k}(G).
\]
In particular, capture time is monotone on trees: if $H$ is a subtree of $T$, then for all $k\ge \zeta(T)$,
\[
\capt_{\zeta,k}(H)\le \capt_{\zeta,k}(T)
\]
[2105.09806]. The proof uses a retraction from $V(G)$ to $V(H)$ that preserves the informational content of probes.

A second line of attack uses treewidth and decomposition radius. If $\tw(G)$ is the treewidth, $\tr(G)$ the “tree-radius,” and an optimal decomposition has $L$ leaves, then
\[
\zeta(G)\le (\tw(G)+1)\tr(G), \qquad
\capt_{\zeta,\,(\tw(G)+1)\tr(G)}(G)\le L.
\]
A complementary bound is
\[
\zeta(G)\le (\tw(G)+1)(\Delta(G)+1), \qquad
\capt_{\zeta,\,(\tw(G)+1)(\Delta(G)+1)}(G)\le \tr(G)+1
\]
[2105.09806]. These are not direct proofs of the conjecture, because they use more than $\zeta(G)$ cops, but they show that capture time can be controlled through structural width parameters.

The most abstract framework in the 2025 paper is a coloring-based generalization. Given colorings $C_1,\dots,C_k$ of $V(G)$, one defines a layered “game structure” whose rows represent sets of vertices that can be the robber’s candidate position after $i$ additional probes. Row 1 consists of all singletons. Row $i$ consists of subsets $S\subseteq V(G)$ not yet listed whose closed neighborhood $N[S]$ can be distinguished by some coloring $C_j$ into color-classes appearing in earlier rows [2508.10443].

If $\{C_S: |S|\le k\}$ are the distance-colorings of $G$ by all $k$-subsets, then the height of the reduced game structure equals $\lcapt_k(G)$ [2508.10443]. In this interpretation, each probe is an application of a coloring, and the row number measures how many further probes suffice from a given candidate set. The paper emphasizes two gains: a unifying framework that does not refer to actual distances, and a route to the conjecture via purely combinatorial bounds on the number of rows as a function of $n$ [2508.10443].

An example result illustrates the perspective: for an arbitrary pair of colorings $C_1,C_2$ on $n$ vertices, the structure has at most $\binom n2+1$ rows, because every row at least $3$ must contain a size-2 set and row 2 already contains at least one pair [2508.10443]. This is not yet linear, but it recasts the conjecture as a combinatorial row-bound problem.

## 7. Open directions and current status

The current status is that the conjecture is confirmed for several graph families but remains open in full generality. Trees and interval graphs satisfy linear bounds [2105.09806], and the later work sharpens the tree bounds and proves the conjecture for a subclass of outerplanar graphs [2508.10443]. The tree and outerplanar results confirm $\lcapt(G)=O(n)$ for these families [2508.10443].

Several open directions are stated explicitly. One is to determine the optimal constant in the conjecture, or prove the stronger form with coefficient $1$, namely whether $\lcapt(G)\le n$ always holds [2105.09806; 2508.10443]. Another is a “speed-up” question: for all $k\ge 2$, whether $\capt_{\zeta,\zeta+k}(G)=O(n/k)$ [2105.09806]. The 2025 paper further asks to extend the improved tree bound to arbitrary graphs of bounded treewidth, determine whether planar graphs with $\zeta(G)=3$ are well-localizable, tighten the general coloring-based row-bound to linear in $n$, and investigate trade-offs between $k$ and $\lcapt_k(G)$ beyond metric-dimension [2508.10443].

Planar graphs are a particularly important obstruction point. The available summary states that $\zeta$ is unbounded for planar graphs, so the capture-time conjecture is wide open there [2508.10443]. This does not refute the conjecture, but it indicates that methods relying on exact small localization number are unlikely to scale directly.

The coloring-structure program suggests a broader conceptual shift. If one can bound the height of reduced game structures linearly in $n$, independently of geometric distance formulas, then the full conjecture would follow [2508.10443]. The expectation stated in the literature is that hybridizing structural decompositions, such as block-cut trees and tree-decompositions, with coloring-based game structures may yield further constant-factor improvements and perhaps ultimately prove $\lcapt(G)\le Cn$ for all connected graphs [2508.10443]. This suggests that the conjecture sits at an interface between pursuit–evasion dynamics and purely combinatorial distinguishability.

Source: https://www.emergentmind.com/topics/localization-capture-time-conjecture