It is shown that the communicating property of Markov Decision Processes (MDPs) is equivalent to satisfaction of sets of linear equations. A mapping between the 'multichain' and 'unichain' linear programs for undiscounted MDPs is developed by applying this equivalence. The mapping is used to prove that the simpler linear program solves communicating MDPs.
- irreducible Markov chains
- linear programming
ASJC Scopus subject areas
- Management Science and Operations Research
- Industrial and Manufacturing Engineering
- Applied Mathematics