Risk-sensitive control of an ergodic diffusion over an infinite horizon (Q1600581)

This paper deals with the ergodic control of diffusion processes with small risk. Consider the stochastic differential equation in \(\mathbb{R}^n\), \[ dx_t= f(x_t)dt+ h(x_t, v_t) dt+ \sigma(x_t) dw_t, \] where \((v_t)\) is a control process with a given compact control region \(U\). To study ergodic risk sensitive control, the authors exploit the discrete-time technique as follows. Let \(\gamma\) and \(\Gamma\) be concentric spheres with radii \(r< R\), respectively. Put \(T_B=\) hitting time to \(B\), \(\Theta_t=\) time shift operator and \(\tau_1= T_\Gamma+ T_\gamma\circ\Theta_{T_\Gamma}\), \(\tau_{i+1}= \tau_i+ \tau_1\circ \Theta_{\tau_i}\). Putting \(v_s= u_i(x_s)\), on \(s\in [\tau_i, \tau_{i+1})\), choose \(\{x_s, s\leq \tau_i\}\) measurable \(U\)-valued functions \(u_i\), \(i= 0,1,\dots\), successively. The problem is to minimize the following risk-sensitive cost functional, by suitable \(u_i\), \(i= 0,1,\dots\); \[ {1\over\delta}\limsup_{n\to\infty} n^{-1}\log E_x\Biggl[\exp \int^{\tau_n}_0 \delta c(x_s, u_{n-1}(x_s)) ds\Biggr], \] where \(c\) is a given cost function and \(\delta\) is small and positive. As \(\delta\to 0\), applying the normalization of the optimal value of the risk-sensitive cost functional, the authors prove that its limit is independent of \((\gamma,\Gamma)\) and corresponds to the optimal value of the risk neutral average cost per unit time; \[ \limsup_{t\to\infty} {1\over t} E_x \int^t_0 c(x_s,u(x_s)) ds. \]