Roman Domination Game (RDG)

• Roman Domination Game (RDG) is a combinatorial optimization framework that applies classical domination theory with game-theoretic and distributed computation principles.
• Nash equilibria in RDG correspond to strong minimal Roman dominating functions, ensuring Pareto-optimal outcomes and robust resource allocation.
• Distributed protocols like GAA, GSA, and EGSA leverage local information to achieve efficient convergence, with proven performance on various graph types.

The Roman Domination Game (RDG) is a combinatorial optimization framework situated at the intersection of classical domination theory, algorithmic game theory, and distributed computation. It formalizes the Roman domination problem—motivated originally by military defense scenarios and now of renewed interest in cybersecurity and redundant resource allocation—as a strategic game. Each vertex of a given graph acts as a rational agent seeking to optimize local payoffs that encode both self-assignment and neighborhood “coverage” costs. Nash equilibria of this game coincide with strong minimal Roman dominating functions and are Pareto-optimal. This structure enables distributed algorithms with provable convergence and efficient, high-quality approximations of optimal Roman domination.

1. Formal Structure of the Roman Domination Game

Let $G=(V, E)$ be a simple graph with $|V|=n$ vertices. Each vertex $v_i \in V$ corresponds to a player in a strategic game $\Gamma = (V, \{S_i\}_{i=1}^n, \{u_i\}_{i=1}^n)$. Here, $S_i = \{0,1,2\}$ denotes player $v_i$’s strategy set, corresponding to assignments in a standard Roman dominating function $f: V \to \{0,1,2\}$.

A profile $C = (c_1,\dots,c_n) \in \Sigma$ ($\Sigma = S_1 \times \dots \times S_n$) fixes the joint assignment. For each vertex $v_i$:

• $N_i$: open neighborhood of $v_i$,
• $\bar{N}_i = N_i \cup \{v_i\}$: closed neighborhood,
• $N_{i,k}$: vertices within distance $1 \leq \text{dist}(v_i, v_j) \leq k$.

Define $m_j(C) = 1$ if no $v$ in $\bar{N}_j$ adopts strategy $2$ (i.e., $v_j$ is “free”/undefended), and $0$ otherwise.

For fixed constants $\lambda_1, \lambda_2$ satisfying $$\frac{2}{3}\lambda_2 < \lambda_1 < \frac{3}{4}\lambda_2$$, the payoff to player $v_i$ is: $$u_i(C) = -\lambda_1 c_i^2 - \lambda_2 \sum_{v_j \in \bar{N}_i} (2 - c_j) m_j(C)$$ This model penalizes large assignments and nearby “free” vertices. Best responses and Nash equilibria are defined as usual: $$BR(v_i, C) = \arg\max_{c'_i \in \{0,1,2\}} u_i(c'_i, C_{-i})$$ A profile $C^*$ is a Nash equilibrium (NE) if, for every $i$ and $c_i \in \{0,1,2\}$,

$u_i(c_i^*, C^*_{-i}) \geq u_i(c_i, C^*_{-i})$

This construction aligns unilateral local rationality with global and local domination objectives (Chen et al., 2023).

2. Game-Theoretic Foundations and Equilibrium Properties

The RDG has a direct correspondence between game-theoretic equilibrium and Roman domination formulations.

• Roman Dominating Function (RDF): $f: V \to \{0,1,2\}$; every vertex $v$ with $f(v)=0$ must have a neighbor $u$ with $f(u)=2$.
• Minimal RDF (M-RDF): No single decrement at any vertex produces another RDF.
• Strong Minimal RDF (S-RDF): No local “exchange” of one $2$ for multiple $1$s in the closed neighborhood strictly decreases total weight.
• Global Minimum RDF (G-RDF): Minimizes $\sum_v f(v)$.

Major theorems:

• Every NE is an RDF (Theorem 2.1): No player can strictly improve utility by covering an uncovered vertex.
• Every NE is an M-RDF (Theorem 2.2): No profitable unilateral decrement exists.
• Every NE coincides with an S-RDF (Theorem 2.3): No profitable local exchange exists.
• Every NE is Pareto-optimal (Theorem 2.4): No strictly dominating joint deviation is possible.

These properties yield the following inclusion hierarchy: $$\{ \text{G-RDF} \} \subseteq \{ \text{NE} \} \subseteq \{ \text{S-RDF} \} \subseteq \{ \text{M-RDF} \} \subseteq \{ \text{RDF} \}$$ This structure ensures that every NE yields a “good” Roman dominating function, and—if global minimality is attained—that NE is also optimally efficient (Chen et al., 2023).

3. Potential Game Formulation and Existence of Equilibria

The RDG is an exact potential game under the potential function: $$\Phi(C) = -\lambda_1 \sum_{j=1}^n c_j^2 - \lambda_2 \sum_{j=1}^n (2 - c_j) m_j(C)$$ A single-player switch $c_i \to c_i'$ yields

$$\Phi(c_i, C_{-i}) - \Phi(c_i', C_{-i}) = u_i(c_i, C_{-i}) - u_i(c_i', C_{-i})$$

Implications include:

• Existence of Nash equilibrium.
• Any sequence of strict unilateral improvements terminates at an NE.
• The number of improvement steps is bounded by $O(n)$, since each gain is lower bounded by a positive constant $\delta$ and the potential $\Phi$ has bounded range $[-(4\lambda_1+2\lambda_2)n, 0]$.

This potential-theoretic structure directly enables decentralized solution algorithms and ensures robustness to asynchronous best-response dynamics (Chen et al., 2023).

4. Distributed Algorithms for Roman Domination Game Equilibria

Several distributed protocols are formulated:

Algorithm	Dynamics	Convergence	Features
GAA	Asynchronous	$O(n)$	Sequential, single best-response
GSA	Synchronous (non-interfering)	$O(n)$	Parallel, independent acting set
EGSA	Synchronous + Contracts	$O(n^2)$	Allows local coalition “contracts”

GAA (Asynchronous): Each $v_i$ updates $c_i$ based on local data ($c_j, m_j$ for $j \in \bar{N}_i$). One round comprises $n$ updates; convergence occurs in $O(n)$ rounds.

GSA (Synchronous): Players with positive gain in $\mu_i = u_i(BR(v_i,C),C_{-i}) - u_i(c_i,C_{-i})$ act only if they have the minimum index in their 2-hop closed neighborhood $\bar{N}_{i,2}$. This enforces non-interfering parallelism and guarantees convergence to an NE in $O(n)$ rounds.

EGSA (Enhanced GSA): Introduces occasional “private contracts” among small coalitions. When GSA reaches an NE, eligible white vertices may propose simultaneous local reassignments reducing the total assignment weight. Each contract strictly decreases $\sum c_i$, and the sequence of contracts plus GSA rounds converges in $O(n^2)$ steps to an improved S-RDF/NE.

Each protocol exploits only 1–2-hop local information; all steps are fully distributed and robust against local asynchrony (Chen et al., 2023).

5. Computational Results and Benchmarking

Empirical evaluations were performed on several types of random graphs, specifically Barábasi–Albert (BA, $m_0=5$) and Erdős–Rényi (ER, $p=0.2$), with experiments averaged over 1,000 instances. Key metrics:

• Weight $\gamma_R(C)$: Sum $\sum_i c_i$ in the final RDF.
• Rounds: Iterations until convergence.
• Real time ratio $\eta$: Sequential GAA time vs. parallel GSA time.

Summary of findings:

• GSA achieves final weights comparable to GAA but with wall-clock time reduction to only $3\%$–$8\%$ of GAA.
• EGSA further reduces average RDF weight, achieving $2\%$–$5\%$ improvements on ER graphs.
• Multiple (e.g., $100$) random-restart runs of GSA sometimes outperform the greedy central algorithm (GA) of Li et al. (2022), the latter being deterministic and non-restartable.
• On BA, GA and GSA($100$) perform comparably; on ER, GA is nominally better but GSA($100$) closes the gap.

On tree graphs (random trees RT; BA-tree BAT, $m=1$), optimality benchmarks:

• GA: $0.3\%$–$2\%$ relative error on RT, $\approx0.2\%$ on BAT.
• GSA: $\approx1\%$–$1.8\%$ on RT, $\approx0.4\%$ on BAT.
• GSA($100$): $\approx0.3\%$–$0.9\%$ on RT, $0.002\%$–$0.02\%$ on BAT.
• EGSA: $0.1\%$–$0.4\%$ on RT, $0\%$–$0.02\%$ on BAT.

Thus, on tree graphs, EGSA yields solutions within $0.01\%$–$0.4\%$ of the optimum (Chen et al., 2023).

6. Connections, Implications, and Extensions

The RDG construction models the minimum Roman domination problem as an exact potential game. Its Nash equilibria are strong minimal Roman dominating functions, guaranteeing Pareto-optimality. Distributed best-response algorithms using only local information efficiently locate such equilibria in general graphs in $O(n)$ rounds. Augmentation via enhanced private contracts (EGSA) raises solution quality, especially in random and tree-like networks, approaching optimality within a small fractional error.

The RDG framework is directly relevant for applications in network resource allocation, defense, and dynamic cyber-physical systems where fully decentralized control and resilience to local disruptions are essential. The methodologies extend classical graph-theoretic notions into algorithmic game theory, offering both deeper insights and scalable, high-performance computational tools (Chen et al., 2023).