Reduction of dimensionality in dynamic programming via the method of diagonal decomposition (Q2537025)

A new method of reducing the dimensionality associated with the solution of optimal control problems is introduced. The usual dynamic programming functional equation has a domain of dimension \(N\) where \(N\) is the dimension of the state sector describing the system. This functional equation is artificially decoupled into \(N\) functional equations with domains of dimension one which must be solved iteratively. The solutions are coupled by means of a pseudo-state equation which includes information about a previous guess at the optimal state trajectory in a known forcing term. Since the \(N\) functional equations are one-dimensional and since they may be solved sequentially rather than simultaneously the computational demands on rapid-access storage are dramatically reduced.