Social Optimum of a Linear Quadratic Collective Choice Model under Congestion

Abstract

This paper investigates the social optimum for a dynamic linear quadratic collective choice problem where a group of agents choose among multiple alternatives or destinations. The agents' common objective is to minimize the average cost of the entire population. A naive approach to finding a social optimum for this problem involves solving a number of linear quadratic regulator (LQR) problems that increases exponentially with the population size. By exploiting the problem's symmetries, we first show that one can equivalently solve a number of LQR problems equal to the number of destinations, followed by an optimal transport problem parameterized by the fraction of agents choosing each destination. Then, we further reduce the complexity of the solution search by defining an appropriate system of limiting equations, whose solution is used to obtain a strategy shown to be asymptotically optimal as the number of agents becomes large. The model includes a congestion effect captured by a negative quadratic term in the social cost function, which may cause agents to escape to infinity in finite time. Hence, we identify sufficient conditions, independent of the population size, for the existence of the social optimum. Lastly, we investigate the behavior of the model through numerical simulations in different scenarios.