Common belief and common knowledge (Q690342)

A model is considered in which several agents are incompletely informed about some aspect of their environment. Each agent holds a belief (i.e. a subjective probability measure) about the true state of the environment. In addition, agents hold beliefs about other agents' beliefs, beliefs about beliefs about beliefs, etc. The authors adopt Mertens and Zamir's formal approach to such hierarchies of beliefs. Within this framework, the authors say that an agent ``knows'' that an event has occurred if and only if is true for all elements of the support of this agent's beliefs that the event has occurred. An event is ``common knowledge'' if everybody knows that it has occurred, if everybody knows that everybody knows that it has occurred, etc. The paper provides several characterisations of common knowledge in this framework. These characterisations make it simpler to determine whether an event is common knowledge, and they also clarify some conceptual problems concerning the notion of common knowledge. The most important characterisation is that an event is common knowledge if it is a ``belief subspace'' in the sense of Mertens and Zamir.