Recently, a novel concept for the computation ofessentialfeatures of thedynamics of Hamiltonian systems (such as molecular dynamics) has beenproposed. The realization of this concept had been based on subdivisiontechniques applied to the Frobenius-Perron operator for the dynamicalsystem. The present paper suggests an alternative but related conceptthat merges the conceptual advantages of the dynamical systems approachwith the appropriate statistical physics framework. This approach allowsus to define the phrase “conformation” in terms of the dynamicalbehavior of the molecular system and to characterize the dynamicalstability of conformations. In a first step, the frequency ofconformational changes is characterized in statistical terms leading tothe definition of some Markov operatorTthat describes the correspondingtransition probabilities within the canonical ensemble. In a secondstep, a discretization ofTvia specific hybrid Monte Carlo techniques isshown to lead to astochasticmatrixP. With these theoretical preparaions,an identification algorithm for conformations (to be presented in alater paper) is applicable. It is demonstrated that the discretizationofTcan be restricted to few essential degrees of freedom so that thecombinatorial explosion of discretization boxes is prevented andbiomolecular systems can be attacked. Numerical results for then-pentane molecule and the triribonucleotideadenylyl(3'-5')cytidylyl(3'-5')cytidin are given and interpreted.