Computing efficient steady state policies for deterministic dynamic programs. II (Q1260897)

[For part I see the author, ibid. 166, No. 1, 175-191 (1992; Zbl 0771.90097).] The author applies his results discussed in part I to his fractional flow model with \(n\)-dimensional state space [see the author, Management Sci., Theory 21, 753-764 (1975; Zbl 0311.90036)]. The various results for the construction of a first order approximation \(\pi_ F\) having finite opportunity cost and the second order improvements on \(\pi_ F\) that reduce transient costs are reviewed. The author claims that the above model possesses a Lagrangian saddle-point thus ensuring the existence of a steady state policy (SSP) with finite opportunity cost. The above results are used for the efficient computation of SSP and constructing a first order approximation \(\pi_ F\) that guides the system to the target set \(s^*\) in \(n\) stages. The author also evaluates the performance of SSPs for the fractional flow model focusing attention on \(\pi_ F\) and the two policies produced by second order improvements. The test bed consists of 16,800 problems with dimension ranging from 2 to 8. The author claims that his approach produces suboptimal policies that show promise.