Towards a theory of stochastic hybrid systems (Q4953390)

The authors define a stochastic hybrid system essentially as (a) a family of SDEs, \(dX_t = f(q,X_t) dt + g(q,X_t) dB_t\) on \(U_q\), (\(U_q\) open in \({\mathbb R}^d\), for \(t \in [\tau_n, \tau_{n+1}), \tau_n\) stopping times) labeled by \(q\) from a countable set \({\mathbb Q}\), together with (b) a transition mechanism yielding a new index \(q' \in {\mathbb Q}\) and a new initial condition \(x=X_{\tau_{n+1}}\) in \(U_{q'}\) for the differential equation indexed by \(q'\). Naturally associated is the embedded Markov chain of the system, \((X_n,Q_N) = (X_{\tau_n},q) \in U_q \times \{q\}\). Modelling highway traffic situations (distance of cars) is given as an example. A further application is the simulation of 1-dimensional SDEs, \(dX_t = f(X_t) dt + g(X_t) dB_t\). Given the state discretizastion parameter \(\delta > 0\), the authors consider the (discretized) stochastic hybrid system associated with the SDE which is defined via drift and diffusion vector fields that are kept constant on suitable intervals \(U_k(\delta)\), \(d\overline{X}_t^{\delta} = f^{\delta}(k) dt + g^{\delta}(k) dB_t\), and the corresponding embedded Markov chain \((\overline{X}_n^{\delta},k_n)\). They discuss recurrence properties for the embedded Markov chain and convergence of invariant measures for \(\delta \rightarrow 0\) (if existing) for the case of additive white noise, \(g(x)=g^{\delta}(k) \equiv 1\). They consider, in particular, special drift vector fields such as in gradient systems. For the entire collection see [Zbl 0934.00029].