Finite Generalized Markov Programming.
Martin L. Puterman, P.J. Weeda · Journal of the American Statistical Association · 1982
will be called the gain of policy f.A policy f E F such that v_ 1 (f) ~ ~ v_ 1 (f) for f E F exists and is called gain-optimal.A gain-optimal policy which maximizes also v 0 (f) over the set of gain-optimal policies is called bias-optimal.The computation of the gain v_ 1 (fl for a policy f E F requires the matrix Q 0 (f) and the vectors T(f) ~ Q 1 (f)7 and c 0 (f).Hence if only the data (1.4;6) for i,j EM, k EK(i) ljEM JtEJR+ tQ(dt):j for iEM, kEK(i) for i EM, k E K(i) *