IEOR 165 Discussion 9 April 17, 2015 Stochastic Process Definition: A stochastic process is a collection of random variables indexed by time. The sequence {Xt , t ∈ T } is called a stochastic process. 1. Time index (T) (a) Discrete time index: T = {0, 1, 2, ...} (b) Continuous time index: T = [0, ∞) or T = [−∞, ∞) 2. State space (S) (a) Countable (discrete) state space (b) Uncountable state space Discrete Time Markov Chain (DTMC) Markov Property: Given the present state, the future state of the system is independent of the past. P (Xt+1 = jt+1 |X1 = j1 , ..., Xt = jt ) = P (Xt+1 = jt+1 |Xt = jt ) DTMC: A stochastic process is called a DTMC if 1. T is discrete 2. S is discrete 3. {Xt } has Markov Property One Step Transition Matrix: Each of its entries is a non-negative real number representing a transition probability. Each row of P sums to 1. p1,1 p1,2 ... p1,n p2,1 p2,2 ... p2,n P = . .. .. . . . . . . . pn,1 pn,2 1 ... pn,n where P (Xt+1 = j|Xt = i) = pi,j . Time Homogeneous MC: The transition matrix P stays same after each step, so the k-step transition probability is equal to the k th power of the transition matrix, P k . Chapman Kolmogorov Equation: The k-step transition probability satisfies: ∑ Pijk = Pirm Prjk−m ∀i, j ∈ S and 0 ≤ m ≤ k r∈S Define fj to be the probability that starting with state j, the process ever return to state ∑ will k k ) to be the probability of first return to j at step k.(fj = k fjj j and fjj Classsification of States: • j is accessible from i: ∃k ≥ 0 such that Pijk > 0. (written as i → j) • i and j are called communicative iff i → j and j → i. (written as i ↔ j) • Two states that communicate are in the same class. • MC is said to be irreducible if its state space is a single communicating class. ∑ k • A state j is said to be transient if fj < 1 or ∞ k=1 Pjj < ∞. ∑ k • A state j is said to be recurrent if fj = 1 or ∞ k=1 Pjj = ∞. • Every finite state, irreducible MC is positive recurrent. • The period of state j is defined as: (n) d(j) = gcd{n ≥ 1 : pjj > 0} where gcd means greatest common divisor. If d(j) = 1, call j aperiodic. If d(j) > 1, call j periodic with period d(j). Question 1. Classify S = 1, 2, 3, 4 by recurrence and transience: Solution 1. 2 State 1: 1 3 2 = 3 = 0 ∀k > 2 1 2 = f11 + f11 =1 1 = f11 2 f11 k f11 f1 recurrent State 3: 1 2 = 0 ∀k > 1 1 1 = f33 = 2 1 f33 = k f33 f3 transient State 4: 1 2 = 0 ∀k > 1 1 1 = f44 = 2 1 f44 = k f44 f4 transient State 2: 1 f22 = 0 2 2 f22 = 3 1 k−2 2 k f22 = ∀k > 2 3 3 ∞ ∑ 1 k−2 2 1 2 f2 = = =1 1 3 3 3 1 − 3 k=2 recurrent Proposition: If MC is aperiodic, irreducible and positive recurrent, there exists a unique stationary distribution π. It is called a stationary distribution for the MC with transition matrix P if π′ = π′P ∑ i.e., πj = r∈S πr pr,j , j ∈ S. The mean return (recurrence) time is: µj = 3 1 πj Question 2. Find the stationary distribution of MC with P: 0.45 0.48 0.07 P = 0.05 0.7 0.25 0.01 0.5 0.49 Solution 2. Sum of the stationary distribution probabilities is equal to [ ] [ ] 0.45 π1 π2 π3 = π1 π2 π3 0.05 0.01 1. i.e. π1 + π2 + π3 = 1. 0.48 0.07 0.7 0.25 0.5 0.49 π1 = 0.45π1 + 0.05π2 + 0.01π3 π2 = 0.48π1 + 0.7π2 + 0.5π3 π3 = 0.07π1 + 0.25π2 + 0.49π3 Since the row sum of P is equal to 1, one constraint is redundant. Solution is π1 = 0.07, π2 = 0.62 and π3 = 0.31. 4