Multiscale stochastic control with state constraints
Glauco Valle (UFRJ)

This talk concerns stochastic optimal control problems in which a slow state variable is constrained to remain in a bounded domain while interacting with fast periodic dynamics. Motivated by applications involving distinct time scales, such as hydrothermal systems subject to short-term renewable fluctuations and long-term hydrological uncertainty, we first discuss a single-scale reflected control problem. The control may affect both the drift and a possibly degenerate diffusion. Under suitable geometric and regularity assumptions, the discounted value function is characterized as the unique viscosity solution of a Hamilton–Jacobi–Bellman equation with a Neumann-type boundary condition induced by reflection and the boundary cost. We then consider a multiscale extension in which the fast variable evolves periodically on a torus. By averaging the fast-dependent coefficients with respect to the invariant density of the frozen fast dynamics, we obtain an effective Hamiltonian and a homogenized reflected control problem. We describe the convergence of the multiscale value functions to the effective value function and conclude with an explicit two-dimensional example in which the invariant density and the homogenized HJB equation can be computed in closed form. Joint work with Anderson O. Calixto and Bernardo Freitas Paulo da Costa.