A weak convergence approach to hybrid LQG problems with infinite control weights (Q1608646)

The paper deals with the LQG problem: a linear quadratic Gaussian regulator problem modulated by continuous-time Markov chains with \(m\) states. At any time the system follows an Ito type stochastic differential equation in which the coefficients take one of \(m\) possible configurations. Thus the system depends on a continuous dynamics and also on discrete events. The main attention is paid to models where the Markov chain is singularly perturbed with a small parameter \(\varepsilon\). The cost functional may depend on an indefinite control weight. The important problem is to study the weak convergence of the system when \(\varepsilon\to 0\). Several results are presented to show the weak convergence of the system, i.e. of the stochastic processes involved, as well as of the corresponding cost functionals. The advantage of this approach is that the limit system has a reduced complexity which is achieved by aggregating the states of the Markov chain in each weakly irreducible sub-class of the state space.