Re: large deviation principle for Markov processes

Correction: I have forgotten to tell you the end of the
compactification trick above! Sorry!
The RW is killed at the boundary of the torus! This is why I wonder why
the inf is not taken over the defective prob. measures because the
empirical distribution at time t of the killed RW has of course total
mass less than 1, if the RW reaches the boundary before t and thus will
stay for the rest of the time in the coffin state not returning to the
torus!
BR
pkg

.