MOVES Seminar 22 Dec, 2011, 11:00

Towards faster numerical solution of Continuous Time Markov Chains stored by symbolic data structures


In my research, I am interested in methods for alleviating some well-known problems in the numerical analysis of Continuous Time Markov Chains. All algorithms presented rely on an MTBDD-based storage of the state graph. Our approach is threefold: 1. For obtaining smaller state graphs in the model generation phase (which usually are easier to solve) a symbolic algorithm for the elimination of vanishing states is developed. 2. For the calculation of steady-state probabilities of Markov chains a multilevel algorithm is developed in order to speed up the convergence of the numerical solution. 3. To calculate the most probable paths in a state graph and for calculating the mean time to the first failure of a system, a path-based solver is developed. In the talk we will focus on the last two topics.