Diffusion approximations for controlled stochastic networks: an asymptotic bound for the value function (Q997413): Difference between revisions

The authors consider a broad family of unitary networks (studied in detail by \textit{M. Bramson}, and \textit{R. J. Williams} [Queueing Syst. 45, 191--221 (2003; Zbl 1131.60305)]) under heavy traffic with general interarrival, service times and probabilistic routing with an infinite horizon discounted linear cost. For an admissible network control problem, the authors establish that the associated cost is asymptotically bounded below by the value function of the corresponding diffusion control problem. They develop a general approach to argue the convergence function which does not rely on the tightness of the various processes describing the dynamics, but rather on the tightness of suitable time-rescaled versions of the same. This time rescaling has been effectively used by \textit{H. J. Kushner} and \textit{L. F. Martins}, [SIAM J. Control Optimization 34, 1781--1797 (1996; Zbl 0857.90041)] in stochastic control problems.