---
title: Isolation Game in Graph Theory
url: https://www.emergentmind.com/topics/isolation-game
type: topic
---

# Isolation Game in Graph Theory

Searching arXiv for the cited paper and closely related work on the graph isolation game.
arXiv_search("2409.14180 isolation game on graphs")
arXiv_search query: "2409.14180 Isolation game on graphs"
arXiv_search query: "2507.08503 Bounds on the game isolation number and exact values for paths and cycles"
The isolation game in graph theory is the special case $\mathcal F=\{K_2\}$ of the $\mathcal F$-isolation game introduced by Brešar, Dravec, Johnston, Kuenzel, and Rall. In this framework, two players alternately select vertices of a graph while jointly constructing an $\mathcal F$-isolating set, but with opposite optimization goals: Dominator seeks to minimize the final number of selected vertices, and Staller seeks to maximize it. For a graph $G=(V,E)$ and a family of graphs $\mathcal F$, a set $S\subseteq V(G)$ is $\mathcal F$-isolating if $G-N_G[S]$ contains no member of $\mathcal F$ as a subgraph. When $\mathcal F=\{K_2\}$, the residual graph is required to be edgeless, so the game terminates with an isolating set in the usual sense [2409.14180].

## 1. Formal setting and basic parameters

Let $G=(V,E)$ be a graph and let $\mathcal F$ be a finite family of forbidden graphs. For $X\subseteq V$, the closed neighborhood is
$$
N_G[X]=\bigcup_{x\in X}(\{x\}\cup N_G(x)).
$$
A set $S\subseteq V$ is $\mathcal F$-isolating if the subgraph induced by the remaining vertices,
$$
G-N_G[S],
$$
contains no member of $\mathcal F$ as a subgraph. The minimum size of such a set is the ordinary $\mathcal F$-isolation number,
$$
\iota(G,\mathcal F)=\min\{\,|S|:S\subseteq V(G),\ G-N_G[S]\text{ is }\mathcal F\text{-free}\,\}.
$$

The game version asks how large the selected set becomes under adversarial play. In the original notation, the Dominator-start and Staller-start game lengths are denoted by $g(G,\mathcal F)$ and $g'(G,\mathcal F)$, respectively [2409.14180]. In later work on the special case $\mathcal F=\{K_2\}$, the same two quantities are denoted by $\iota_g(G)$ and $\iota'_g(G)$ [2507.08503]. This difference is terminological rather than conceptual.

The special case $\mathcal F=\{K_2\}$ is called the isolation game. Here the final residual graph must be independent, so the selected vertices isolate all edges. This makes the game a vertex-selection analogue of domination-style processes, but with a stopping condition defined by the disappearance of nontrivial residual components rather than by full domination alone [2409.14180].

## 2. Game dynamics, legality, and termination

The $\mathcal F$-isolation game is played by Dominator and Staller, who alternately build a chosen set $S$. The game state is tracked through a marked set $M$, initially empty, although the theory also allows partially marked graphs. At any stage, a vertex $x\notin M$ is playable if it dominates at least one vertex of a component of $G-M$ that is not yet $\mathcal F$-free; equivalently, $x$ is not playable precisely when $N[x]\subseteq M$ [2409.14180].

After a legal move $x$, the marked set is updated by
$$
M \leftarrow M\cup N[x]
$$
together with all vertices of any newly $\mathcal F$-forbidden components of $G-M$. The game ends when no playable vertex remains; in the formulation of the original paper this is equivalent to $M=V(G)$. By construction, the final chosen set $S$ is then $\mathcal F$-isolating [2409.14180].

This marking formalism is important because the chosen set and the marked set are not identical. A move can cause many vertices to become marked even though only one new vertex is chosen. A plausible implication is that much of the strategic complexity lies in controlling the residual component structure rather than merely the cardinality of $S$.

## 3. Structural theory of the $\mathcal F$-isolation game

A central result is the Continuation Principle. If $A,B\subseteq V(G)$ and $B\subseteq A$, then playing on $G$ with $A$ already marked can only be no longer than playing on $G$ with $B$ marked:
$$
g(G|A,\mathcal F)\le g(G|B,\mathcal F),\qquad
g'(G|A,\mathcal F)\le g'(G|B,\mathcal F).
$$
The proof proceeds by an imagination argument with a real game and an imagined game played in parallel, copying Staller’s real moves into the imagined game and Dominator’s imagined moves into the real game whenever legal [2409.14180].

An immediate consequence is the start-player inequality
$$
|\,g(G,\mathcal F)-g'(G,\mathcal F)\,|\le 1.
$$
Thus, for the original $\mathcal F$-isolation game, the first-move advantage is tightly controlled [2409.14180].

The theory also compares different forbidden families. If for every $H\in\mathcal F$ there exists $H'\in\mathcal F'$ with $H$ a subgraph of $H'$, then for every graph $G$,
$$
g(G,\mathcal F')\le g(G,\mathcal F),\qquad
g'(G,\mathcal F')\le g'(G,\mathcal F).
$$
This monotonicity formalizes the idea that forbidding a structurally larger family can only make isolation easier. In particular, the paper gives the general bounds
$$
\iota(G,\mathcal F)\le g(G,\mathcal F)\le 2\,\iota(G,\mathcal F)-1,
$$
and
$$
\iota(G,\mathcal F)\le g'(G,\mathcal F)\le 2\,\iota(G,\mathcal F).
$$
These inequalities locate the game parameters between the ordinary isolation number and a factor-of-two enlargement [2409.14180].

## 4. The isolation game proper: the $\frac12 n$ bound and the $\frac37 n$ conjecture

For $\mathcal F=\{K_2\}$, the notation simplifies to $g(G)=g(G,\{K_2\})$, $g'(G)=g'(G,\{K_2\})$, and $\iota(G)=\iota(G,\{K_2\})$. The foundational upper bound is
$$
g(G)\le \frac{|V(G)|}{2}.
$$
The proof assigns weight $1$ to every vertex initially and lets the weight drop to $0$ when a vertex becomes marked. Dominator’s strategy is to avoid playing inside an isolated $K_2$ residual and, in any other component, to play a vertex of degree at least $2$ whenever possible. Under this strategy each Dominator move reduces the total remaining weight by at least $3$, each Staller move by at least $1$, and once only isolated edges remain each move reduces the total by exactly $2$ [2409.14180].

The original paper conjectures that the actual sharp upper bound is smaller. If $G$ has no isolated edge to start, then
$$
g(G)\le \Bigl\lceil \frac{3|V(G)|}{7}\Bigr\rceil,\qquad
g'(G)\le \Bigl\lceil \frac{3|V(G)|}{7}\Bigr\rceil.
$$
The same work gives a family $G^*$ showing that this bound is attained if true: each vertex of an arbitrary base graph is “blown up” to the center of a path on $7$ vertices, yielding
$$
g(G^*)=g'(G^*)=\frac{3|V(G^*)|}{7}.
$$
This suggests that the conjectured $\frac37$ law, if established in full generality, would be best possible [2409.14180].

Later work sharpens the global $\frac12 n$ picture. It proves that there are only two graphs attaining equality in the bound $\iota_g(G)\le \frac12|V(G)|$, namely $K_2$ and $C_6$, and that there are precisely eleven graphs attaining equality in the bound $\iota'_g(G)\le \frac12|V(G)|$. The same paper constructs a new infinite family $\widehat G$ with
$$
n(\widehat G)=7k,\qquad \iota_g(\widehat G)=\iota'_g(\widehat G)=3k=\frac37\,n(\widehat G),
$$
again confirming the sharpness of the conjectured ratio at the level of infinite examples [2507.08503].

## 5. Forests, paths, cycles, and exact evaluations

Forests occupy a special position. In any partially marked forest $F$, the original paper proves by induction that
$$
g(F)\le g'(F).
$$
In particular, for any forest $T$, the Dominator-start game never lasts longer than the Staller-start game. This forest monotonicity is stronger than the general $|g-g'|\le 1$ bound and depends on the acyclic residual structure [2409.14180].

For paths, the original paper establishes
$$
g'(P_n)\le \Bigl\lceil \frac{2n+2}{5}\Bigr\rceil \qquad (n\ge 3),
$$
using a strategy in which Dominator responds at distance exactly $4$ from Staller whenever possible. It also proves
$$
\Bigl\lfloor \frac{2n}{5}\Bigr\rfloor-1 \le g(P_n)\le g'(P_n)\qquad (n\ge 6),
$$
and derives the exact values
$$
g(P_n)=g'(P_n)=\Bigl\lfloor\frac{2n+2}{5}\Bigr\rfloor
$$
when $n\equiv i\pmod 5$ and $i\in\{1,2,3\}$ [2409.14180].

These partial results were later completed. For every $n\ge 1$,
$$
\iota_g(P_n)=
\begin{cases}
\lfloor(2n+1)/5\rfloor-1,&n\equiv0\pmod5,\\
\lfloor(2n+1)/5\rfloor,&\text{otherwise},
\end{cases}
\qquad
\iota'_g(P_n)=\lfloor(2n+2)/5\rfloor.
$$
For every $n\ge 4$,
$$
\iota_g(C_n)=
\begin{cases}
2\,\lceil n/5\rceil,&n\equiv0\pmod5,\\
2\,\lceil n/5\rceil-1,&\text{otherwise},
\end{cases}
$$
and
$$
\iota'_g(C_n)=
\begin{cases}
2\,\lfloor n/5\rfloor+1,&n\equiv4\pmod5,\\
2\,\lfloor n/5\rfloor,&\text{otherwise}.
\end{cases}
$$
The proof methods include “runs-and-gaps” lower bounds for Staller and periodic Dominator strategies that select every fifth new vertex around a cycle [2507.08503].

The same later paper also improves the general tree bound: if $T$ is a tree of order $n\ge 3$, then
$$
\iota_g(T)\le \frac{5}{11}n.
$$
Its proof uses a refined weight-counting method on the residual tree, with vertices colored white, blue, or red and a three-stage strategy ensuring that in every two-move pair the total weight decreases by at least $22$ from an initial total of $5n$ [2507.08503].

## 6. Related variants and terminological distinctions

A common source of confusion is that several graph games use closely related terminology while differing substantially in rules, state variables, and objectives. The total isolation game is one such variant. It is played on an isolate-free graph, and a legal move must be adjacent either to a vertex in a nontrivial component of $G-N_G(S)$ or to a vertex that is isolated in $G-N_G(S)$ and belongs to $S$. The resulting parameter $\iota_{\rm gt}(G)$ satisfies
$$
\iota_{\rm gt}(G)<\frac56 n
$$
for connected graphs of order $n\ge 3$,
$$
\iota_{\rm gt}(G)\le \frac34 n
$$
when $\delta(G)\ge 2$, more generally
$$
\iota_{\rm gt}(G)\le \Bigl(\frac{2\delta-1}{3\delta-2}\Bigr)n,
$$
and
$$
\iota_{\rm gt}(G)\le \frac23 n
$$
when $\operatorname{diam}(G)\le 2$. Unlike the original isolation game, the Continuation Principle fails here: for $P_5$, one has $|\iota_{\rm gt}(P_5)-\iota'_{\rm gt}(P_5)|=2$ [2601.03363].

Another distinct family is the Toucher–Isolator game, which is played on edges rather than vertices. Toucher and Isolator alternately claim previously unclaimed edges, Toucher moves first, and the parameter $u(G)$ counts the number of vertices untouched by Toucher under optimal play. In this setting, Dowden, Kang, Mikalački, and Stojaković introduced the game and obtained general bounds, while Räty proved for every tree $T$ on $n\ge 3$ vertices that
$$
u(T)\ge \Bigl\lfloor\frac{n+3}{5}\Bigr\rfloor,
$$
which is sharp because
$$
u(P_n)=\Bigl\lfloor\frac{n+3}{5}\Bigr\rfloor.
$$
These results concern a Maker-Breaker type edge-claiming game and should not be conflated with the vertex-selection isolation game on graphs [1903.11411; 2001.10498].

Within graph game theory, the isolation game is therefore best understood as a residual-component game governed by closed neighborhoods, partial marking, and isolating-set formation. Its current theory combines abstract comparison principles, sharp exact values on classical families, and a still-active search for the definitive $\frac37$ upper bound beyond presently known classes and constructions [2409.14180].

Source: https://www.emergentmind.com/topics/isolation-game