Temporal oscillations and phase transitions in the evolutionary minority game

Ehud Nakar,Shahar Hod

DOI: https://doi.org/10.1103/PhysRevE.67.016109

2003-02-04

Abstract:The study of societies of adaptive agents seeking minority status is an active area of research. Recently, it has been demonstrated that such systems display an intriguing phase-transition: agents tend to {\it self-segregate} or to {\it cluster} according to the value of the prize-to-fine ratio, $R$. We show that such systems do {\it not} establish a true stationary distribution. The winning-probabilities of the agents display temporal oscillations. The amplitude and frequency of the oscillations depend on the value of $R$. The temporal oscillations which characterize the system explain the transition in the global behavior from self-segregation to clustering in the $R<1$ case.

Condensed Matter

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to understand how the behavior of agents changes with the change of the reward - punishment ratio \(R\) in the evolutionary Minority Game. Specifically, the research aims to explain the time - series oscillation phenomenon of the agents' winning probability and its impact on the global behavior, especially the phase transition of agents from self - segregation to clustering when the reward - punishment ratio \(R\) is less than a certain critical value \(R_c\).

### Background of the Paper and Problem Description

1. **Research on Adaptive Agent Populations**
- Research on adaptive agent populations seeking minority status is an active research area.
- These systems exhibit an interesting phase transition: agents tend to self - segregate or cluster depending on the value of the reward - punishment ratio \(R\).

2. **Limitations of Existing Research**
- Early research mainly focused on simple scenarios, assuming that the reward - punishment ratio \(R = 1\).
- Existing research has shown that when the reward - punishment ratio \(R < R_c\), agents tend to cluster around \(p=\frac{1}{2}\) rather than self - segregate into two extreme groups.

3. **Main Contributions of This Paper**
- This paper reveals that these systems do not establish a true steady - state distribution but exhibit time - series oscillations in the winning probability.
- The amplitude and frequency of the oscillations depend on the reward - punishment ratio \(R\) and the agents' gene value \(p\).

### Main Findings

1. **Time - Series Oscillations in the Winning Probability**
- The agents' winning probability \(\tau(p)\) oscillates over time, and the amplitude and frequency of the oscillations depend on the reward - punishment ratio \(R\).
- When \(R\) is small, the amplitude of the oscillation is large and the frequency is also high.

2. **Performance of Different Types of Agents**
- Extreme agents (\(p = 0\) or \(p = 1\)) have a larger oscillation amplitude than central agents (\(p=\frac{1}{2}\)).
- When \(R > R_c\), extreme agents perform better, while when \(R < R_c\), central agents perform better.

3. **Phase - Transition Mechanism**
- When \(R > R_c\), the system indirectly cooperates through self - segregation to maximize the efficiency of resource utilization.
- When \(R < R_c\), agents cannot cooperate through self - segregation but tend to cluster at \(p = \frac{1}{2}\) to increase their chances of survival.

### Conclusions

- This paper explains the mechanism of changes in agent behavior in the evolutionary Minority Game by analyzing the time - series oscillations in the winning probability.
- The oscillation phenomenon explains the phase transition from self - segregation to clustering and reveals different behavior patterns of the system under different reward - punishment ratios.

### Formula Summary

The winning probability formula is:
\[ \tau(p)=\frac{1}{2}-\alpha p(1 - p)\]
where \(\alpha < 1\) is a constant that depends on the number of agents \(N\).

Through these findings, the paper provides a new perspective for understanding behavior patterns in complex adaptive systems.