Previous |  Up |  Next

Article

Title: Uniqueness of optimal policies as a generic property of discounted Markov decision processes: Ekeland's variational principle approach (English)
Author: Ortega-Gutiérrez, R. Israel
Author: Montes-de-Oca, Raúl
Author: Lemus-Rodríguez, Enrique
Language: English
Journal: Kybernetika
ISSN: 0023-5954 (print)
ISSN: 1805-949X (online)
Volume: 52
Issue: 1
Year: 2016
Pages: 66-75
Summary lang: English
.
Category: math
.
Summary: Many examples in optimization, ranging from Linear Programming to Markov Decision Processes (MDPs), present more than one optimal solution. The study of this non-uniqueness is of great mathematical interest. In this paper the authors show that in a specific family of discounted MDPs, non-uniqueness is a “fragile” property through Ekeland's Principle for each problem with at least two optimal policies; a perturbed model is produced with a unique optimal policy. This result not only supersedes previous papers on the subject, but it also renews the interest in the corresponding questions of well-posedness, genericity and structural stability of MDPs. (English)
Keyword: discounted Markov decision processes
Keyword: dynamic programming
Keyword: unique optimal policy
Keyword: non-uniqueness of optimal policies
Keyword: Ekeland's variational principle
MSC: 90C40
MSC: 93E20
idZBL: Zbl 1374.90407
idMR: MR3482611
DOI: 10.14736/kyb-2016-1-0066
.
Date available: 2016-03-21T17:51:53Z
Last updated: 2018-01-10
Stable URL: http://hdl.handle.net/10338.dmlcz/144863
.
Reference: [1] Bertsekas, D. P.: Dynamic Programming: Deterministic and Stochastic Models..Prentice-Hall, NJ 1987. Zbl 0649.93001, MR 0896902
Reference: [2] Bishop, E., Phelps, R. R.: The support functionals of a convex set..In: Proc. Sympos. Pure Math. Vol. VII, 1963 (V. L. Klee, ed.), Amer. Math. Soc., pp. 27-35. Zbl 0149.08601, MR 0154092, 10.1090/pspum/007/0154092
Reference: [3] Borwein, J. M., Zhu, Q. J.: Techniques of Variational Analysis..Springer, New York 2005. Zbl 1076.49001, MR 2144010
Reference: [4] Cruz-Suárez, D., Montes-de-Oca, R., Salem-Silva, F.: Conditions for the uniqueness of optimal policies of discounted Markov decision processes..Math. Methods Oper. Res. 60 (2004), 415-436. Zbl 1104.90053, MR 2106092, 10.1007/s001860400372
Reference: [5] Cruz-Suárez, D., Montes-de-Oca, R.: Uniform convergence of the value iteration policies for discounted Markov decision processes..Bol. Soc. Mat. Mexicana 12 (2006), 133-152. MR 2301750
Reference: [6] Ekeland, I.: On the variational principle..J. Math. Anal. Appl. 67 (1974), 324-353. Zbl 0286.49015, MR 0346619, 10.1016/0022-247x(74)90025-0
Reference: [7] Hernández-Lerma, O., Lasserre, J. B.: Discrete-Time Markov Control Processes: Basic Optimality Criteria..Springer-Verlag, New York 1996. Zbl 0840.93001, MR 1363487, 10.1007/978-1-4612-0729-0
Reference: [8] Lucchetti, R.: Convexity and Well-Posed Problems..CMS Books in Mathematics, Springer, New York 2006. Zbl 1106.49001, MR 2179578, 10.1007/0-387-31082-7
Reference: [9] Montes-de-Oca, R., Lemus-Rodríguez, E.: An unbounded Berge's minimum theorem with applications to discounted Markov decision processes..Kybernetika 48 (2012), 268-286. Zbl 1275.90124, MR 2954325
Reference: [10] Montes-de-Oca, R., Lemus-Rodríguez, E., Salem-Silva, F.: Nonuniqueness versus uniqueness of optimal policies in convex discounted Markov decision processes..J. Appl. Math. 2013 (2013), 1-5. Zbl 1266.90113, MR 3039713, 10.1155/2013/271279
Reference: [11] Rockafellar, R. T., Wets, R. J. B.: Variational Analysis..Springer, New York 2004. Zbl 0888.49001, MR 1491362
Reference: [12] Tanaka, K., Hosino, M., Kuroiwa, D.: On an $\varepsilon $-optimal policy of discrete time stochastic control processes..Bull. Inform. Cybernet. 27 (1995), 107-119. MR 1335274
.

Files

Files Size Format View
Kybernetika_52-2016-1_5.pdf 308.8Kb application/pdf View/Open
Back to standard record
Partner of
EuDML logo