Schrödinger's control and estimation paradigm with spatio-temporal distributions on graphs

The problem of reconciling a prior probability law with data was introduced by E. Schr\"odinger in 1931/32. It represents an early formulation of a maximum likelihood problem. The specific formulation can also be seen as the control problem to modify the law of a diffusion process so as to match specifications on marginal distributions at given times. Thereby, in recent years, this so-called {\em Schr\"odinger Bridge problem} has been at the center of the development of uncertainty control. However, an unstudied facet of this program has been to address uncertainty in space and time, modeling the effect of tasks being completed, instead of imposing specifications at fixed times. The present work is a first study to extend Schr\"odinger's paradigm on such an issue. It is developed in the context of Markov chains and random walks on graphs. Specifically, we study the case where one marginal distribution represents the initial state occupation of a Markov chain, while others represent first-arrival time distributions at absorbing states signifying completion of tasks. We establish that when the prior is Markov, a Markov policy is once again optimal with respect to a likelihood cost that follows Schr\"odinger's dictum.