Some dynamic decision process (Q2722328)

A stochastic dynamic decision process is formulated as a discrete dynamic programming problem with a one-stage non-negative unbounded loss function. Using Fenchel duality, a dual dynamic programming problem is introduced with a reward function, which is conjugate to the loss function of the original dynamic programming problem. By making use of the convexity and \(\omega^*\)-lower-semicontinuity of the reward function, weak duality results are proved. If the loss function is convex and lower semicontinuous, strong duality results are proved.NEWLINENEWLINEFor the entire collection see [Zbl 0958.00047].