[(g,w)]-Optima in Markov Renewal Programs.
An extension of RM-5208, [Multichain Markov Renewal Programs], to cover the case of completely ergodic MRPs. Policies that are nearly optimal for small discount rates are found by linear programming. More precisely, the policies are [(g,w)]-optimal, where, for a given policy, [g] is the loss rate and [w] the bias term. Finding [(g,w)]-optima in completely ergodic programs is primarily of theoretical interest, because ordinarily there is a unique policy minimizing [g]. There are examples, however, in which several policies exist that minimize [g], but not all of these are [(g,w)]-optimal. 7 pp. Ref.