Controlled equilibrium selection in stochastically perturbed dynamics (Q1800819)

The model under consideration is a general dynamical system in $\mathbb{R}^d$ determined by the vector field $m$ with finitely many critical points perturbed by a Wiener process of intensity $\epsilon^\nu$ and by an additive control with a small multiplier $\epsilon$. The running cost is a sum of a function of a state variable and of a quadratic function of the control; the cost is ergodic, i.e., the average in time and state in the long run, and the objective is to minimize it. The vector field which defines the non-perturbed system satisfies a Lyapunov type hypothesis among other usual technical assumptions. The control is of a feedback relaxed form; a weakly unique solution of the corresponding SDE is assumed. \par Three regimes are studied as $\epsilon\to 0$ depending on the value of the constant $\nu$ with a critical value of $\nu=1$ (Theorem 1.11); in some sense there is another critical value $\nu=2$ (Remark 1.15). Theorem 1.11, in particular, shows the concentration properties of the optimal invariant (limiting) measure for small $\epsilon$ depending on the value of the parameter $\nu$. The bounds for the discrepancy between the cost-related component $\beta_*^\epsilon$ of solution of the HJB equation from Theorem 1.4 and one of three limiting functionals determined by the vector field $m$ and the running cost function $\ell$ are stated, depending on the value of $\nu$. In Theorem 1.12 concentration bounds for the optimal stationary measure is given in terms of moments. According to Theorem 1.13 a scaled limit of the optimal distribution density approaches a Gaussian density. Theorem 1.19 addresses a linear-quadratic case.