On modeling punishment in multi-agent systems

In this paper we study isolation as a form of punishment. Although an isolated violator is punished as it can not benefit from the interactions with other agents, compliant agents may also suffer from not engaging with the violators. In this paper we analyze such problems. Certain modifications of multi agent systems are needed to solve this problem. These modifications are aimed to make the violator redundant so that it can be ignored and hence isolated. Deciding on these modifications is NP-complete and approximation algorithms exist.