A belief-based evolutionarily stable strategy

Xinyang Deng, Zhen Wang, Qi Liu, Yong Deng, Sankaran MahadevanarXiv 2014 · arXiv

Introduction

In evolutionary game theory (EGT), the concept of the Evolutionarily Stable Strategy (ESS) is fundamental to understanding how specific behaviors or traits persist within a population. An ESS is defined as a strategy that, if adopted by a population, cannot be invaded by any rare alternative strategy. Classical models typically represent these strategies as either pure actions or mixed strategies with fixed, precise probabilities. However, real-world interactions—whether in biological ecosystems or human social structures—are rarely characterized by such precision. Environmental noise, varying degrees of individual rationality, and internal fluctuations introduce a level of "fuzziness" that fixed-probability models struggle to capture.

This paper proposes a novel framework to address this limitation by introducing a "belief strategy" based on Dempster-Shafer (D-S) evidence theory. By allowing strategies to be defined by a range of probabilities rather than a single point, the authors provide a more robust model for evolutionary stability under uncertainty.

Theoretical Framework: Belief Strategies and Evidence Theory

The core issue addressed is the gap between idealized mathematical models and the fluctuating reality of agent interactions. To bridge this, the authors leverage Dempster-Shafer evidence theory, a mathematical framework designed for reasoning with uncertain or incomplete information.

Defining the Belief Strategy

Unlike a standard mixed strategy that assigns a fixed probability to each action, a belief strategy uses a "mass function" over the power set of all pure strategies. This allows for the assignment of probability not just to individual actions (e.g., "Hawk" or "Dove"), but also to the uncertainty or "ignorance" between them.

Two critical metrics are derived from this mass function:

  1. Belief ($Bel$): Represents the lower bound of support for a particular strategy.
  2. Plausibility ($Pl$): Represents the upper bound of support.

The interval $[Bel, Pl]$ defines the range of possible probabilities for a strategy. The width of this interval represents the "level of ignorance." Crucially, if the uncertainty is zero ($Bel = Pl$), the model collapses back into a standard mixed strategy, making it a true generalization of classical EGT.

The Belief-Based Evolutionarily Stable Strategy

The authors extend the classical ESS definition into a belief-based ESS. A key part of this extension involves adapting the Bishop-Canning theorem. In classical theory, this theorem states that all pure strategies within a mixed ESS must yield the same expected payoff. In the belief-based framework, this condition is expanded to account for the range of probabilities defined by the belief and plausibility bounds.

Geometrically, while a traditional mixed ESS in a 2x2 game is represented as a single point on a probability line, a belief-based ESS is represented as a segment. This segment defines the range of behavioral fluctuations the population can tolerate while remaining stable against invasion. The authors introduce a parameter, $\delta$, to quantify this uncertainty, allowing for a formal analysis of how "fuzzy" a stable system can be.

Application: The Hawk-Dove Game

To demonstrate the practical application of this theory, the authors analyze the Hawk-Dove game, a canonical model for animal competition over a resource of value $V$ with a cost of injury $C$.

The payoff matrix of the Hawk-Dove game, serving as the primary model for demonstrating belief-based ESS.

In the classical version of this game, where $C > V$, the unique mixed ESS is to play "Hawk" with a fixed probability $P = V/C$. Applying the belief-based framework, the authors derive that the stable strategy is no longer a single point but a range: $$t \in [V/C - \delta, V/C + \delta]$$ where $t$ is the fluctuating probability of choosing the "Hawk" action. Through analytical derivation, the paper proves that the expected payoff of this belief-based strategy against itself remains higher than that of any invading strategy, provided the average behavior remains within the calculated bounds. This confirms that the population remains stable even in the presence of continuous fluctuations.

Discussion and Conclusion

The introduction of belief-based ESS offers a significant step toward more realistic models of evolutionary dynamics. By accounting for uncertainty and fluctuations, it provides a mathematical language for the "fuzziness" observed in nature.

However, several areas for future research remain. The current model assumes a uniform distribution of probabilities within the belief-plausibility interval, which may not reflect all real-world scenarios. Furthermore, the application has so far been limited to simple 2x2 symmetric games. Future work should explore more complex interactions, including asymmetric games and multi-strategy systems. Finally, empirical studies are needed to determine if the behavioral variations observed in biological populations actually correspond to the theoretical bounds predicted by the belief-based ESS model.

Summary updated

BibTeX
@misc{deng2014belief,
  author = {Xinyang Deng and Zhen Wang and Qi Liu and Yong Deng and Sankaran Mahadevan},
  title = {{A belief-based evolutionarily stable strategy}},
  year = {2014},
  eprint = {1406.0045},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1406.0045},
}
Figure