Roman Dominating Function (RDF)

Updated 2 March 2026

RDF is a combinatorial labeling on graphs where every vertex labeled 0 must have a neighbor labeled 2, establishing the Roman domination number by minimizing the total weight.
It captures key structural properties and bounds, displaying sensitivity to vertex and edge modifications, and connects to graph products, reconfiguration, and complexity analysis.
Algorithmic approaches, including FPT, distributed, and game-theoretic methods, are employed to compute the Roman domination number, highlighting both optimization challenges and practical applications.

A Roman Dominating Function (RDF) on a graph formalizes a combinatorial optimization principle inspired by the allocation of mobile defensive units in networked structures. Introduced in connection with classical domination but enriched with a labeling that encodes both redundancy and local defense, RDFs have become a central object in domination theory, parameterized complexity, optimization algorithms, and graph product theory. The minimum-weight realization of such a function, termed the Roman domination number, exhibits deep connections with graph modification operations, criticality, product and iterative constructions, and parameterized complexity.

1. Definitions and Structural Properties

Let $G=(V,E)$ be a simple graph. An RDF is a function $f:V\rightarrow\{0,1,2\}$ such that every $v\in V$ with $f(v)=0$ has a neighbor $u$ with $f(u)=2$ ; $V_0,V_1,V_2$ denote, respectively, the label classes. The total weight is $w(f)=\sum_{v\in V}f(v)=|V_1|+2|V_2|$ . The Roman domination number, $\gamma_R(G)$ , is the minimum weight of an RDF. Any minimum weight RDF is called a $\gamma_R$ -function (Samodivkin, 2016, Martinez et al., 2021, Yero et al., 2011, Kazemi, 2011, Samodivkin, 2017, Ashok et al., 2024, Chen et al., 2023).

Global properties include:

$\gamma(G)\leq\gamma_R(G)\leq 2\gamma(G)$ , where $\gamma(G)$ is the domination number.
$|V_2|\leq \gamma_R(G)-\gamma(G)$ and $|V_1|\geq 2\gamma(G)-\gamma_R(G)$ for any $\gamma_R$ -function (Yero et al., 2011).
For standard classes, $\gamma_R(P_n)=\gamma_R(C_n)=\lceil2n/3\rceil$ for paths and cycles of order $n$ (Yero et al., 2011).

A vertex $v$ is called $\gamma_R$ -good if some $\gamma_R$ -function assigns it a nonzero label; a graph is $\gamma_R$ -excellent if every vertex is $\gamma_R$ -good (Samodivkin, 2016).

2. Operations, Criticality, and Changing Classes

Roman domination is sensitive to vertex and edge modifications. The following lemmas and classification results are central:

Vertex Deletion Principle: $\gamma_R(G-v)<\gamma_R(G)$ if and only if there is a $\gamma_R$ -function $f$ with $f(v)=1$ , in which case $\gamma_R(G-v)=\gamma_R(G)-1$ .
Edge Addition Principle: For nonadjacent $x,y$ , $\gamma_R(G) \geq \gamma_R(G+xy)\geq\gamma_R(G)-1$ , with equality $\gamma_R(G+xy)=\gamma_R(G)-1$ exactly when a $\gamma_R$ -function assigns $\{1,2\}$ to $\{x,y\}$ (Samodivkin, 2016, Samodivkin, 2017).

Six classes of graphs arise concerning $\gamma_R$ 's response to edge/vertex removal or addition. Notably, $\mathcal R_{UVR}$ comprises graphs where $\gamma_R$ is invariant under any single vertex removal; every tree in this class is $\gamma_R$ -excellent (Samodivkin, 2016, Samodivkin, 2017).

A $k$ -vertex Roman-critical graph is one where removal of any $k$ -vertex set strictly decreases $\gamma_R$ (Samodivkin, 2017).

3. Roman Domination in Product and Iterated Constructions

The behavior of $\gamma_R$ under graph products and iterative constructions is governed by sharp inequalities and, in specific cases, structural trichotomies.

Direct Product: For graphs $G,H$ without isolated vertices,

$\max\{p(G)\gamma_R(H),p(H)\gamma_R(G)\} \leq \gamma_R(G\times H) \leq \min\{2\gamma(G)\gamma_{tR}(H),2\gamma(H)\gamma_{tR}(G)\}$

with $p(\cdot)$ being the packing number, $\gamma_{tR}(\cdot)$ the total Roman domination number, and $\gamma(\cdot)$ the domination number (Martinez et al., 2021).

Rooted Product: For $G$ of order $n(G)$ , $H$ rooted at $v$ ,

$\gamma_R(G\circ H)\in\{n(G)\gamma_R(H),~\gamma_R(G)+n(G)(\gamma_R(H)-1),~\gamma(G)+n(G)(\gamma_R(H)-1)\}$

with precise structural criteria for which value is attained (Martinez et al., 2021).

Cartesian and Strong Products:

$\gamma_R(G\square H) \ge \gamma(G)\gamma_R(H)$ . If $G$ has an efficient dominating set, $\gamma_R(G\square H) = \gamma(G)\gamma_R(H)$ (Yero et al., 2011).
For strong product $G\boxtimes H$ ,

$\gamma_R(G\boxtimes H) \leq \gamma_R(G)\gamma_R(H)-2|A_2||B_2|$

where $A_2,B_2$ are the sets with label 2 in $\gamma_R$ -functions of $G,H$ (Yero et al., 2011).

Mycieleskian: For the Mycieleskian $\mu(G)$ ,

$\gamma_R(G)+1\le\gamma_R(\mu(G))\le\gamma_R(G)+2$

$\gamma_R(\mu(G))=\gamma_R(G)+1$ if and only if $G$ is special Roman (i.e., $G$ is Roman with a $\gamma_R$ -function using only labels $0,2$ and $G[V_2]$ has no isolated vertex) (Kazemi, 2011).

4. Algorithmic and Parameterized Complexity

Computing $\gamma_R(G)$ is generally hard; however, fixed-parameter tractable (FPT) and distributed approaches are established under structural restrictions.

FPT for Cluster Deletion Distance: For $G$ with cluster-vertex-deletion size $k$ , $\gamma_R(G)$ and the independent Roman domination number $i_R(G)$ can be computed in time $O(4^k n^{O(1)})$ (Ashok et al., 2024).
Lower Bounds: No $2^{\varepsilon k}n^{O(1)}$ -time algorithm exists for any $0<\varepsilon<1$ (unless SETH fails); no polynomial-size kernel unless NP $\subseteq$ coNP/poly (Ashok et al., 2024).
Distributed/Game-theoretic Algorithms: Nash equilibria of the Roman Domination Game (RDG) correspond to strong minimal RDFs. The game-based synchronous algorithm (GSA) converges in $O(n)$ rounds, and the enhanced GSA (EGSA) in $O(n^2)$ rounds, where $n=|V|$ . EGSA achieves high-quality approximations, often outperforming classic greedy approaches on random graph classes (Chen et al., 2023).

5. $\gamma_R$ -Excellent Graphs, Recognition, and Construction

A graph $G$ is $\gamma_R$ -excellent if every vertex $x$ is $\gamma_R$ -good, i.e., for each $x$ , there is a $\gamma_R$ -function $h_x$ with $h_x(x)\ne 0$ . For trees, there is a constructive labeling characterization using operations (O1–O4) starting from base trees $H_1,\dots,H_5$ :

Label vertices by statuses $A,B,C,D$ where $S_A=V^{01}(T)$ , $S_D=V^{012}(T)$ , $S_B$ collects certain $V^{02}$ vertices of degree 2, $S_C=V^{02}\setminus S_B$ .
Recognition algorithm on trees is linear-time via dynamic programming, checking local degree/status patterns, and systematically peeling off substructures guided by the labeling (Samodivkin, 2016).

A prominent subclass are UVR-trees, for which the domination number is invariant under any vertex deletion; every UVR-tree is $\gamma_R$ -excellent, and they correspond to the cases with $V^{01}(T)=\emptyset$ (Samodivkin, 2016).

6. Partition and Reconfiguration Structures

The $\gamma_R$ -partition of $V(G)$ groups vertices by the set of labels they attain in all $\gamma_R$ -functions: $V^0,V^1,V^2,V^{01},V^{02},V^{12},V^{012}$ . $G$ is $\gamma_R$ -excellent if $V^0=\emptyset$ (Samodivkin, 2016).

The set of all $\gamma_R$ -functions, $\mathscr{D}_R(G)$ , forms the vertex set of various reconfiguration graphs, with adjacency defined by local relabelings such as $(2,0)$ -swaps or $(2,1)$ -swaps. These $\gamma_R$ -graphs can model complex combinatorial structures, and the structure of these graphs is closely connected to path and cycle decompositions and universal graph constructions (Samodivkin, 2017).

7. Connections to Variants and Open Problems

Variants include the independent Roman domination number $i_R(G)$ , where an RDF must have $V_1\cup V_2$ as an independent set (Ashok et al., 2024). Other extensions include total Roman domination and further refinements involving packing numbers and product graph decompositions (Martinez et al., 2021).

Open problems concern the tightness of bounds in general product graphs, characterization of $\gamma_R$ -excellent graphs beyond trees, the universal structure of $\gamma_R$ -graphs, and complexity status outside the regime of known FPT or distributed approaches (Martinez et al., 2021, Yero et al., 2011, Samodivkin, 2017).