The successive approximation methods described above operate on the functional
of the DP functional equation, e.g. on function g in (16) and function f
in (12). That is, the functional of the DP functional equation is the object
of the approximation scheme. In a similar manner it is sometimes possible
and desirable to approximate the DP policy, that is the policy that determines
the optimal values of the decision variables. The update mechanisms of such
approximation schemes are therefore called policy iterations or policy improvements
(e.g. Denardo, 2003) or successive approximations in the policy space
(e.g. Sniedovich, 1992).