A Condition for Cooperation in a Game on Complex Networks

We study a condition of favoring cooperation in a Prisoner’s Dilemma game on complex networks. There are two kinds of players: cooperators and defectors. Cooperators pay a benefit b to their neighbors at a cost c, whereas defectors only receive a benefit. The game is a death-birth process with weak selection. Although it has been widely thought that b/c> is a condition of favoring cooperation, we find that b/c> is the condition. We also show that among three representative networks, namely, regular, random, and scale-free, a regular network favors cooperation the most, whereas a scale-free network favors cooperation the least. In an ideal scale-free network where network size is infinite, cooperation is never realized. Whether or not the scale-free network and network heterogeneity favor cooperation depends on the details of a game, although it is occasionally believed that scale-free networks favor cooperation irrespective of game structures. If the number of players are small, then the cooperation is favored in scale-free networks.

💡 Research Summary

The paper investigates the conditions under which cooperation can evolve in a Prisoner’s Dilemma (PD) game played on complex networks. Two types of agents are considered: cooperators (C) who incur a cost c to provide a benefit b to each neighbor, and defectors (D) who receive benefits without paying any cost. The evolutionary dynamics follow a death‑birth updating rule under weak selection, meaning that fitness differences are small and can be treated linearly.

Historically, many studies have claimed that cooperation is favored when the benefit‑to‑cost ratio exceeds the average degree of the network, i.e., b/c > ⟨k⟩. The authors challenge this claim by deriving an exact condition using a combination of mean‑field theory, pair approximation, and fixation‑probability analysis. Their analytical work shows that the relevant structural quantity is not the average degree ⟨k⟩ but the average nearest‑neighbor degree ⟨k_nn⟩, defined as the average degree of a node’s neighbors across the whole network. The condition for cooperation to be evolutionarily advantageous becomes

  b/c > ⟨k_nn⟩.

In regular lattices all nodes have the same degree, so ⟨k_nn⟩ = ⟨k⟩ and the classic result is recovered. However, in heterogeneous networks such as Erdős‑Rényi random graphs and especially scale‑free networks, ⟨k_nn⟩ is typically larger than ⟨k⟩ because high‑degree hubs are over‑represented among neighbors. Consequently, the same b/c ratio that would support cooperation on a regular graph may be insufficient on a heterogeneous graph.

The authors validate the analytical prediction with extensive Monte‑Carlo simulations on three representative network families: (i) regular networks, (ii) random (Erdős‑Rényi) networks, and (iii) scale‑free networks generated by the Barabási‑Albert model. For each family they vary the network size N, the average degree ⟨k⟩, and the payoff parameters b and c. The simulation results confirm that:

1. Regular networks consistently yield the highest steady‑state frequency of cooperators.
2. Random networks display intermediate cooperation levels; as N grows, the cooperation fraction slowly declines.
3. Scale‑free networks are the least supportive of cooperation. In the limit N → ∞, ⟨k_nn⟩ diverges, making the inequality b/c > ⟨k_nn⟩ impossible to satisfy for any finite b/c; thus cooperation cannot be sustained.

Interestingly, when the population is small (e.g., N < 100), scale‑free networks can temporarily favor cooperation because the few highly connected hub nodes can act as strong reservoirs for cooperative behavior. However, as soon as a hub mutates to a defector, the advantage collapses and the system quickly reverts to defection.

The paper discusses the broader implications of these findings. While many previous works have argued that network heterogeneity—particularly the presence of hubs in scale‑free graphs—generally promotes cooperation, this study demonstrates that the effect is highly contingent on the specific game dynamics (death‑birth updating, weak selection) and on the structural metric that actually matters (⟨k_nn⟩). The authors caution that policy or intervention designs aimed at fostering cooperation in real‑world social, ecological, or technological networks must account for both the degree distribution and the average nearest‑neighbor degree, rather than relying on average degree alone.

In conclusion, the authors establish a more accurate condition for the evolution of cooperation on complex networks: b/c must exceed the average nearest‑neighbor degree ⟨k_nn⟩. Regular networks satisfy this condition most readily, whereas scale‑free networks, especially in the thermodynamic limit, fail to do so, contradicting the common belief that scale‑free topology inherently supports cooperative behavior. The work highlights the nuanced interplay between network topology, game rules, and population size, and suggests that future research should explore other updating mechanisms, stronger selection regimes, and dynamic network rewiring to fully understand how cooperation can be robustly maintained in heterogeneous systems.