Sufficient Statistics for Dynamic Teams

We show that a new theory of sufficient statistics for team decision problems applies equally well to general dynamic team decision problems. In particular, we show that the sufficient statistics given in the theory are sufficient for making optimal decisions, and moreover, they can be updated with new measurements without having to scan through the old measurements. The implication is that that the process of finding, updating, and interpreting the sufficient statistics for decentralized control problems can be done relatively easily using the tools of the new theory, without resorting to complicated probabilistic or algebraic arguments.