[Paper Review] A New Mathematical Model for Evolutionary Games on Finite Networks of Players

Dario Madeo, Chiara Mocenni|arXiv (Cornell University)|Jul 5, 2013

Evolutionary Game Theory and Cooperation42 references19 citations

TL;DR

This paper proposes a novel replicator equation on finite, networked populations with arbitrary topology, where each player (vertex) evolves via internal replicator dynamics influenced by games with neighbors. The model extends classical evolutionary game theory by allowing heterogeneous strategies, weighted/directed links, and player-specific payoff matrices, enabling coexistence of strategies and richer dynamics than the classical replicator equation, especially in structured populations like small-world or scale-free networks.

ABSTRACT

A new mathematical model for evolutionary games on graphs is proposed to extend the classical replicator equation to finite populations of players organized on a network with generic topology. Classical results from game theory, evolutionary game theory and graph theory are used. More specifically, each player is placed in a vertex of the graph and he is seen as an infinite population of replicators which replicate within the vertex. At each time instant, a game is played by two replicators belonging to different connected vertices, and the outcome of the game influences their ability of producing offspring. Then, the behavior of a vertex player is determined by the distribution of strategies used by the internal replicators. Under suitable hypotheses, the proposed model is equivalent to the classical replicator equation. Extended simulations are performed to show the dynamical behavior of the solutions and the potentialities of the developed model.

Motivation & Objective

• To extend classical evolutionary game theory to finite populations organized on complex networks with generic topologies.
• To model players as vertices in a graph, each hosting an internal population of replicators that evolve based on pairwise games with connected neighbors.
• To allow player-specific payoff matrices and weighted/directed connections to reflect individual perceptions and interaction strengths.
• To generalize the classical replicator equation to finite, structured populations while preserving key equilibria under homogeneous conditions.
• To investigate how network topology and heterogeneous initial conditions influence strategy coexistence and evolutionary dynamics.

Proposed method

• Each player is modeled as a vertex in a graph, with internal replicators that follow standard replicator dynamics based on payoff matrices.
• Pairwise games are played between replicators in connected vertices, and outcomes influence replication rates via payoff aggregation (WA or WS models).
• The overall behavior of a vertex player is determined by the distribution of strategies among its internal replicators, aggregated using weighted (WA) or unweighted (WS) sum of neighbor payoffs.
• The model reduces to the classical replicator equation under homogeneous initial conditions, identical payoff matrices, and WA payoff aggregation.
• Graph topology is arbitrary—undirected, directed, weighted—allowing analysis of diverse network structures including stars, small-world, and scale-free networks.
• Simulations are conducted using various topologies (e.g., star, asymmetric weighted) and initial conditions (homogeneous, outlayer-based) to study dynamical behavior.

Experimental results

Research questions

• RQ1 How does network topology influence the emergence of strategy coexistence in evolutionary games on finite populations?
• RQ2 Can the classical replicator equation be generalized to finite, structured populations with heterogeneous payoff matrices and weighted connections?
• RQ3 What happens to evolutionary dynamics when initial conditions are non-homogeneous in a networked game setting?
• RQ4 How does the presence of mixed strategies affect stability and convergence in networked evolutionary games?
• RQ5 In what ways does the proposed model extend the concept of evolutionary stability in structured populations?

Key findings

• The proposed model generalizes the classical replicator equation under homogeneous initial conditions, identical payoff matrices, and WA payoff aggregation.
• In the absence of pure Nash equilibria, such as in the anti-coordination game with payoff matrix [[0,1],[1,0]], the system converges to a unique mixed Nash equilibrium at [0.5, 0.5] under homogeneous initial conditions.
• With non-homogeneous initial conditions, the dynamics become highly dependent on network topology and initial strategy distribution, leading to heterogeneous strategy adoption across the network.
• In the prisoner’s dilemma game, the model allows coexistence of cooperative and non-cooperative strategies across different vertices, even when full cooperation is not evolutionarily stable in the classical sense.
• When a strictly dominant strategy exists, the model shows partial adoption: some vertices converge to the dominant strategy while others maintain alternative strategies, demonstrating mixed behavior.
• The model reveals more complex dynamics than the classical replicator equation, including stable coexistence and non-uniform convergence patterns, especially in asymmetric or weighted networks.