Payoff-Scaled Prisoner’s Dilemma

• Payoff-Scaled Prisoner’s Dilemma is a variant of the classical dilemma where fixed payoffs are replaced by functions sensitive to context such as network topology, temporal modulation, or strategic controls.
• Topology-dependent scaling leverages centrality measures to adjust payoffs dynamically, potentially reversing defection-dominant outcomes and fostering cooperation.
• Temporal and iterative strategy modifications enable the control of feasible payoff regions, influencing evolutionary dynamics and the stability of cooperative behavior.

The payoff-scaled Prisoner’s Dilemma (PSPD) refers to any variant or generalization of the classical Prisoner's Dilemma (PD) in which the numerical payoffs of the game—typically fixed parameters—are replaced or modulated by functions of external context such as network topology, temporal environment, or explicit algorithmic scaling defined by player strategies. Payoff scaling can profoundly alter the evolutionary dynamics, potentially enabling outcomes such as the stable persistence of cooperation or strategic control over the achievable payoff region. Formal frameworks for PSPD encompass topology-dependent games on graphs, temporally modulated payoffs, and direct regional control in the iterated PD formalism.

1. Fundamental Definition and Conceptual Basis

In the classical PD, payoffs are defined by the matrix of rewards—Temptation ($T$), Reward ($R$), Punishment ($P$), and Sucker's payoff ($S$)—subject to $T>R>P>S$ and $2R > T+S$. These are typically treated as constants, invariant across time, topology, and strategic context. Payoff-scaled PDs break this invariance by defining one or more payoff entries as explicit functions: of agent position in a graph, temporal phase, or implemented memory-one strategy parameters. The term “payoff-scaled” thus covers both deterministic and stochastic scaling regimes, with documented manifestations in spatial network games, evolutionary dynamics with time-dependent environments, and repeated games with outcome-constrained regions.

2. Topology-Dependent Payoff Scaling

In Sinha et al. (2020), payoffs are made weakly dependent on the underlying interaction network topology, introducing the formalism of cooperator ($G_C$) and defector ($G_D$) subgraphs. Each node $i$ is assigned centrality-based weights:

• $C_i$, the (species-dependent) closeness centrality within $G_C$ or $G_D$;
• $B_i$, the betweenness centrality within the opposing-species subgraph.

The node’s accumulated payoff $\Pi_i$ is rescaled:

$\Pi'_i = \Pi_i \cdot \exp(a C_i + b B_i)$

where $(a, b)$ depend on the edge type: $a=1, b=0$ for intra-species, $a=0, b=1$ for inter-species interactions. This rewrites the payoffs for each edge as

• $R'_i = R \cdot \exp(C_i)$ for $C$–$C$,
• $P'_i = P \cdot \exp(C_i)$ for $D$–$D$,
• $T'_i = T \cdot \exp(B_i)$ for $D$–$C$,
• $S'_i = S \cdot \exp(B_i)$ for $C$–$D$.

A critical analytic result shows that for cooperators to escape the classical dominance of defection, it suffices that

$C_i - B_j > \ln(T/R)$

for a cooperator $i$ and a defector $j$ (Sinha et al., 2020). Thus, topologically central cooperators can achieve $R'_i > T'_j$, flipping the local two-player game from a PD into Harmony or Coordination, depending on $S', P'$.

3. Dynamical Implications and Analytical Results

Topology-scaled payoffs enable novel dynamical regimes:

• With conventional payoffs ($\Pi$), small levels of dispersal or high $T$ rapidly drive the population to defection.
• The $\Pi'$ scaling, by contrast, sustains significant cooperator fractions $f_C$ up to much higher temptation $T$, provided the initial cooperator density $f_{C_i}$ is sufficiently high.
• The critical temptation threshold $T_c$, above which cooperation collapses, is analytically increased for large-$C_i$ clusters; the Nash equilibrium of all-defect can be destabilized.
• Phase diagrams in $(T, f_{C_i})$ space display widened cooperation-supporting regions under payoff scaling (Sinha et al., 2020).

These outcomes hold even under moderate “random-dispersal” (fraction of strategy swaps per round), and are robust to shifts in the initial condition, but are contingent on the precise scaling parameters and centrality distributions of $G_C$ and $G_D$.

4. Temporal Scaling: Periodically Modulated Payoff Matrices

Temporal payoff-scaling generalizes the PD to scenarios where payoffs, especially the defection or punishment entries, are time-dependent. Ahmed & Safan (2012) (Ahmed et al., 2013) consider a framework in which the mutual defection payoff $U(t)$ oscillates sinusoidally:

$U(t) = U_0 - \Delta U \cos(\omega t)$

with standard PD constraints $T > R > U_0 > S$. The replicator equation reads

$\frac{dx}{dt} = x(1-x)[f_C(t) - f_D(t)]$

with $f_C(t), f_D(t)$ depending explicitly on $U(t)$. An explicit time-dependent cooperation threshold $x_*^{\max}$ is calculated:

$$x_*^{\max} = \frac{U_0 - \Delta U}{R - T + U_0 - \Delta U}$$

Numerical simulations confirm persistent cooperation when the initial cooperator density exceeds $x_*^{\max} \approx 0.6$ for typical parameters. The mean cooperator fraction remains elevated, and the system settles onto a stable periodic orbit, as verified by Floquet analysis.

5. Payoff Region Control in Iterated Prisoner’s Dilemma

In the context of repeated/iterated PD (IPD), Hao et al. (2018) (Hao et al., 2018) introduce a formalism for constraining the feasible payoff region through prescribed scaling or bounding of payoff pairs $(u_X, u_Y)$. Each player’s memory-one strategy defines a transition matrix on the four outcome states, and the stationary distribution determines the long-run payoffs. By selecting linear constraints—e.g.,

$u_Y \leq \alpha u_X + \beta$

—one constructs a “scaled” subset of the original feasible quadrilateral. Sufficient conditions on the memory-one strategy vector $\mathbf p$ ensure that, regardless of the opponent’s response, the observed payoff region remains in a specified triangle or trapezoid. This generalizes Press-Dyson zero-determinant (ZD) strategies and encompasses previously known pinning and extortion paradigms.

Concrete worked examples (e.g., enforcing $u_Y \leq 2u_X - 1$ for $(R,S,T,P) = (2,-1,3,0)$) yield explicit inequalities on the strategy parameters.

6. Significance, Extensions, and Limitations

Payoff-scaling mechanisms demonstrate that modifying payoff invariance—by infusing network topology, temporal modulation, or explicit regional control—can qualitatively restructure the evolutionary trajectories of PD-like systems. This directly addresses the challenge of explaining observed maintenance of cooperation in biological and social systems where context or structure mediates rewards.

Notable implications and limitations include:

• “Quorum sensing” phenomena, whereby local alliances or groupings enhance in-group benefits, motivate such scaling (Sinha et al., 2020).
• The robustness of cooperation is sensitive to the form of scaling (exponential and single centrality types were employed), the graph class analyzed (BA scale-free graphs), and the update protocol (synchronous imitation).
• Alternative scaling forms, network types (random-regular, small-world, multilayer), or evolutionary dynamics could yield distinct behaviors and remain areas for further research (Sinha et al., 2020).
• In temporally modulated PDs, the amplitude and frequency of oscillation critically affect the threshold for cooperation (Ahmed et al., 2013).
• The framework for regional payoff control subsumes prior regularities such as Tit-for-Tat or ZD strategies and enables systematic synthesis of new control strategies under linear constraints (Hao et al., 2018).

A plausible implication is a broadening of spatial game theory to incorporate co-evolution of both structure and payoff, and to model empirical systems (e.g., microbial communities) with dynamically realized reward landscapes.

7. Summary Table of Representative Payoff-Scaling Mechanisms

Context	Scaling Mechanism	Key Condition / Formula
Graph topology	$\Pi'_i = \Pi_i \exp(aC_i + bB_i)$	$C_i - B_j > \ln(T/R)$
Temporal modulation	$U(t) = U_0 - \Delta U \cos(\omega t)$	$x(0) > x^{\max}_*$
Iterated strategies	$u_Y \leq \alpha u_X + \beta$	Linear inequalities on $\mathbf p$

Specific outcomes (such as escape from defection dominance or confinement to prescribed payoff regions) are realized when these mechanisms are instantiated with appropriate scaling parameters and initial conditions.

---

References:

• "Topology dependent payoffs can lead to escape from prisoner's dilemma" (Sinha et al., 2020)
• "On evolutionary games with periodic payoffs" (Ahmed et al., 2013)
• "Payoff Control in the Iterated Prisoner's Dilemma" (Hao et al., 2018)