Memory-based strategy reputation and adaptive learning in spatial evolutionary games: A robust agent-based model for...

This paper studies cooperation in spatial evolutionary games by combining local memory-based reputation with heterogeneous Fermi learning. Agents are placed on a two-dimensional lattice and repeatedly play the Prisoner's Dilemma with their neighbors. Each agent records the recent cooperative behavior of neighboring agents within a finite memory window and uses this information to form a local reputation assessment. This reputation then biases the choice of reference neighbors for imitation. Strategy adoption is modeled through a Fermi-type update rule with agent-specific learning sensitivity, allowing agents to differ in how strongly they respond to payoff differences. Numerical experiments show that reputation-biased reference selection improves cooperation by increasing the influence of historically cooperative neighbors. When combined with heterogeneous learning sensitivity, the mechanism produces more stable cooperative clusters within the tested parameter ranges and delays the decline of cooperation under stronger temptation to defect. Additional comparisons with baseline models, sensitivity analyses, structural-disorder experiments, and system-size checks are used to examine the stability of these findings. The results suggest that cooperation in structured populations depends not only on local interaction structure, but also on how agents use past behavior to select social references and respond to payoff differences.