It is a well known fact that interleaving semantics for Markov
decission processes do not allow compositional reasoning. Some
research has been done in order to obtain compositional reasoning for
Markov decision processes. These approaches restrict the set of
possible schedulers, in order to eliminate unrealistic hehaviours.
This elimnation of schedulers is also useful to obtain better bounds
for the probability of properties. In this talk, we will show why the
model checking problem is undecidable if the set of schedulers is
restricted to obtain compositionality.