Optimal control: coarse control and fine dynamics

138 Views Asked by At

Consider the time evolution of a probability distribution $P(\alpha_{t})$ as follows.

\begin{equation} P(\alpha_{t}) = F[P(\alpha_{t-1}), U_t(y_{t-1})], \end{equation} where $y_{t-1} = \mathcal{P}\alpha_{t-1}$, with $\mathcal{P}$ being some projection matrix, and $U$ being a control variable.

I wish to find a set of functions $\{U_t(y_{t-1})\}_{t=1}^{t=T}$, such that $\langle f(\alpha_T)\rangle_{P(\alpha_T)} \to \max$, for some known function $f$.

My questions:

  1. Is such a scenario already explored? If so, can you share with me the literature?
  2. If it is trivial, can you share with me the Bellman equation that needs to be solved?

Additional info.:

  1. The initial conditions of $\alpha$ (and hence $y$) are known.
  2. $\alpha$'s belong to a countable and bounded space.