The Shared Reward Dilemma

One of the most direct human mechanisms of promoting cooperation is rewarding it. We study the effect of sharing a reward among cooperators in the most stringent form of social dilemma, namely the Prisoner’s Dilemma. Specifically, for a group of players that collect payoffs by playing a pairwise Prisoner’s Dilemma game with their partners, we consider an external entity that distributes a fixed reward equally among all cooperators. Thus, individuals confront a new dilemma: on the one hand, they may be inclined to choose the shared reward despite the possibility of being exploited by defectors; on the other hand, if too many players do that, cooperators will obtain a poor reward and defectors will outperform them. By appropriately tuning the amount to be shared a vast variety of scenarios arises, including traditional ones in the study of cooperation as well as more complex situations where unexpected behavior can occur. We provide a complete classification of the equilibria of the $n$-player game as well as of its evolutionary dynamics.

💡 Research Summary

The paper introduces a novel social‑dilemma framework called the “Shared Reward Dilemma.” A group of (n) agents repeatedly play pairwise Prisoner’s Dilemma (PD) games with each other. In addition to the usual PD payoffs, an external entity provides a fixed total reward (R) that is divided equally among all cooperators in a given round; defectors receive none. This creates a two‑fold incentive: cooperating yields the shared reward, but cooperating also exposes the player to exploitation by defectors who continue to collect the temptation payoff (T). The authors ask how the magnitude of (R) reshapes the strategic landscape and the evolutionary dynamics of cooperation.

Static game analysis.
The authors first enumerate all pure‑strategy Nash equilibria for any (R). Three regimes emerge.

1. Low‑reward regime ((R) so small that the per‑cooperator share (R/k) is below the baseline payoff of defection). The classic all‑defect equilibrium remains the unique Nash equilibrium.
2. Intermediate‑reward regime (moderate (R)). Here mixed equilibria appear in which a subset of players cooperate, share the reward, and the rest defect. The equilibrium conditions are given by the equality of expected payoffs for cooperators and defectors, leading to a nonlinear equation in the number of cooperators (k). Multiple solutions can exist, giving rise to “partial‑cooperation” equilibria.
3. High‑reward regime (large (R)). When (R/k) exceeds the PD payoff advantage of defection, a full‑cooperation equilibrium becomes feasible. However, because the per‑capita reward declines as more agents cooperate, the stability of this equilibrium is sensitive to the exact value of (R); a slight increase in (R) can destabilize full cooperation by making the marginal benefit of joining the cooperating group too small.

Evolutionary dynamics.
To capture how populations evolve toward these equilibria, the authors employ the replicator equation. Let (x) denote the fraction of cooperators. The fitness of a cooperator is (f_C = \pi_C + R/(n x)) (where (\pi_C) is the usual PD payoff) and the fitness of a defector is (f_D = \pi_D). The replicator dynamics (\dot{x}=x(1-x)(f_C-f_D)) yields fixed points that correspond exactly to the static Nash equilibria. The stability analysis shows:

• In the low‑reward regime, (x=0) (all defect) is globally stable.
• In the intermediate regime, two interior fixed points appear: a lower‑(x) unstable saddle and a higher‑(x) stable node. The basin of attraction depends on the initial cooperator density.
• In the high‑reward regime, (x=1) (all cooperate) becomes locally stable, but a new phenomenon—reward‑induced reversal—can occur. As (R) grows, the per‑capita reward (R/(n x)) shrinks, eventually making cooperation less attractive than defection even though the total reward is large. This creates a non‑monotonic relationship between (R) and the equilibrium cooperator fraction, producing a “boom‑bust” pattern where cooperation first rises, then collapses as the reward becomes excessive.

Weighted sharing extension.
The authors also explore a generalization where the reward is not split equally but proportionally to a player’s contribution history (e.g., number of past cooperative moves). This weighted sharing introduces heterogeneity among cooperators, leading to additional mixed equilibria and richer bifurcation structures. In networked populations, such weighting can generate clusters of high‑reward cooperators that are robust against invasion by defectors, suggesting that differential incentives may mitigate the reversal effect observed under equal sharing.

Policy implications.
The study demonstrates that simply increasing the size of a shared incentive does not guarantee higher cooperation. An optimal reward magnitude exists that balances the incentive to cooperate against the dilution effect when many agents cooperate. Moreover, the design of the distribution rule (equal vs. weighted) critically influences the long‑term stability of cooperative behavior. For practitioners—whether in public policy, organizational management, or online platforms—these results caution against naïve “more‑is‑better” reward schemes and advocate for calibrated, possibly differentiated, incentive structures that account for group size and strategic feedback.

In summary, the paper provides a complete classification of Nash equilibria for the (n)-player shared‑reward Prisoner’s Dilemma, derives the corresponding replicator dynamics, identifies novel non‑linear phenomena such as reward‑induced reversal, and extends the analysis to weighted reward allocations. The work bridges classical game‑theoretic equilibrium analysis with evolutionary dynamics, offering both theoretical insight and practical guidance for designing effective cooperation‑promoting mechanisms.