This work considers Markov chains with finite state space. It is supposed that the process has a single recurrent class, but the set of transient states is not necessarily empty. In this context, a Varadhan

s function, giving the exponential grow rate of an aggregated cost function, is studied. The main result establishes that this functional is the optimal value of a minimization problem on the Euclidean space whose dimension equals the number of states.
Keywords Exponential grow rate - Single recurrent class - Transient states - Risk–sensitive average cost
AMS Subject Classification 60J05 - 60F10
Manuscript received: August 2003 / Final version received: January 2004This work was supported by the PSF Organization under Grant No. 010/300/01-5, and Conacyt Grant No. 37643–E.