A new algorithm for multi-dimensional dynamic programming problems

The new algorithm presented here solves medium size multi-dimensional dynamic programming problems in a relatively short computational time with no fast-memory restraints. The algorithm converges to the global optimal solution under some differentiability and convexity assumptions.The procedure is to solve a succession of dynamic programming problems, the state sets of which are limited to only a very small subset of the original state space. The interrelated definition of state sets for successive subproblems facilitates an algorithmic convergence while moving the subsets to contain the optimal states at the end.