# Transient Analysis of Markov Processes Lecture 2: Algorithmic Methods for transient analysis of continuous time Markov Chains Dr. Ahmad Al Hanbali Industrial Engineering Dep University of Twente [email protected] 1 Lecture 2 This Lecture deals with continuous time Markov processes as opposed to discrete time Markov chains in Lecture 1

Objectives: 1. Find equilibrium distribution 2. Find transient probabilities 1. Matrix decomposition 2. Uniformization method 3. Find Transient measures Lecture 2: transient analysis of continuous time Markov chains 2 Background (1) Let denote a continuous time stochastic process

of state space is a Markov chain if the conditional transition probability, for every and is homogeneous (or stationary) if is irreducible if all states can communicate Lecture 2: transient analysis of continuous time Markov chains 3 Background (2)

Define (infinitesimal) transition rate from state i to j of a Markov process Let denote epochs of transition of CTMC then for (by convention ) where is the total outgoing rate of state Lecture 2: transient analysis of continuous time Markov chains 4 Background (3)

Let . The matrix is called the generator of the continuous time Markov chain (CTMC). Note: Let equilibrium probabilities. The equilibrium equations of CTMC gives in matrix equation , Idea: take advantage of Methods developed for discrete time Markov chains (in Lecture 1) Lecture 2: transient analysis of continuous time Markov chains 5 An equivalent discrete time Markov chain

Equilibrium distribution can be obtained from an equivalent Markov chain via an elementary transformation. Let be real number such that , and is a stochastic matrix, i.e., its entries are between 0 and 1, and its rows sum to 1. Further, The Markov chain with transition probability is a discretization of the Markov process of with time step Lecture 2: transient analysis of continuous time Markov chains

6 Uniformization of CTMC To have same mean sojourn time in all states per visit the uniformization of CTMC introduces fictitious transitions from states to themselves Let introduce a fictitious transition from state to itself with rate This yields: 1. Equilibrium distribution of Q doesn't change 2. Outgoing rate from state i becomes () same for all states 3. Equilibrium distribution of the uniformized Markov process of Q is same as the Markov chain of transition matrix P(=I+Q) embedded at epoch of transitions (jumps)

The transitions of the uniformized process take place according to a Poisson process with rate Lecture 2: transient analysis of continuous time Markov chains 7 Equilibrium distribution All methods developed for solving the equilibrium equation for discrete time Markov chain can be applied to the uniformized Markov chain of transition matrix P Lecture 2: transient analysis of continuous time Markov chains

8 Transient Behavior of CTMC Kolmogorov's equations are needed for the transient analysis. Let define the transient probability Then, for Kolmogorov's equations are set of differential equations for Lecture 2: transient analysis of continuous time Markov chains

9 Background (1) Let the matrix of entries Kolmogorov's forward equations are derived by letting s approaches t from below (backward equations) Hence, Truncating the infinite sum is inefficient since has

positive and negative elements Lecture 2: transient analysis of continuous time Markov chains 10 Matrix decomposition method Let be the eigenvalues of Let and be the left and right eigenvectors corresponding to , such that and for . The matrix then reads

where is the matrix whose rows are , is the diagonal matrix of entries , and is the matrix whose columns are Lecture 2: transient analysis of continuous time Markov chains 11 Matrix decomposition method (cnt'd) The transient probability matrix then reads What is the interpretation of ?

What conditions li should satisfy when t tends infinity ? Disadvantage of matrix decomposition? Lecture 2: transient analysis of continuous time Markov chains 12 Uniformization method

Let and . Conditioning on , number of transitions in which is Poisson distributed with mean , gives Truncating the latter sum on the first terms gives a good approximation, It is better to take the largest possible value of Lecture 2: transient analysis of continuous time Markov chains 13 Occupancy Time: mean

Occupancy time of a state is the sojourn time in that state during . Note that depends on state at time 0 Let denote the mean occupancy time in state during given initial state . Then, In matrix equation, Lecture 2: transient analysis of continuous time Markov chains 14 Mean occupancy time (cnt'd) Using the uniformized process (,P) then, Note , where Y is a Poisson random variable of mean

We find that Lecture 2: transient analysis of continuous time Markov chains 15 Cumulative distribution of occupancy time Let denote the total sojourn time during [0,T] in a subset of states, . Then, for , where is the probability that uniformized process visits times during given that it makes transitions. Proof, see for details Tijms 2003: 1. Condition on Poisson number of transitions of the uniformized process to be n 2. Occupancy time is smaller than x if uniformized process will visit

k times out of the n visits and at least k of these transitions happens before x. The former probability is and latter is function of a binomial distribution Lecture 2: transient analysis of continuous time Markov chains 16 Moments of occupancy time Proposition: The m-th moment of O(T) is given by: Proposition: Given that the chain starts in equilibrium the second moment of the occupancy time in the subset during [0,T] gives where is the steady state probability of the Markov chain in state , is the column vector with i-th entry equal to if and zero otherwise, and is the column vector with i-th entry equal to 1 if and zero otherwise. Lecture 2: transient analysis of continuous time Markov chains

17 References V.G. Kulkarni. Modeling, analysis, design, and control of stochastic systems. Springer, New York, 1999 Tijms, H. C. A first course in stochastic models. New York: Wiley, 2003 http://www.win.tue.nl/~iadan/algoritme/ 18