Math 554 Introduction to Stochastic Processes Instructor: Alex Roitershtein Iowa State University Department of Mathematics Fall 2015 Test #1 (solutions to the sample) 1 [20 Points]. (a) Solve Exercise 1.5 in the textbook. (b) A fair die is thrown repeatedly. Let Xn denote the sum of the first n throws. Find lim P(Xn is a multiple of 13). n→∞ Solution: (a) There are three communication classes. S1 = {0, 1} and S2 = {2, 4} are recurrent, and S3 = {3, 5} is transient. Let π = {π0 , π1 } be the stationary distribution of the chain within the recurrent communication class. Then π0 = 0.5π + 0.7π1 , and hence π0 = 3/5π1 . Thus π0 = 3/8 and π1 = 5/8. Since P restricted to S1 is aperiodic, it follows that lim P n (0, 0) = n→∞ 8 1 = . π0 3 Furthermore, using the notation of Section 1.5 (see pp. 29-30), we obtain 0.25 0.25 0 0.25 0 0.25 1 −0.25 S= , Q= , I −Q= , 0 0.2 0 0.2 0.2 0.4 −0.2 0.6 1 12 5 1 3 4 0 4 −1 M = (I − Q) = , A = MS = , 11 4 20 11 1 5 0 5 and lim P n (5, 0) = α(5, 0) + α(5, 1) = A(2, 1) + A(2, 2) = n→∞ 6 . 11 (b) Let Yn be the reminder of Xn after dividing by 13. That is Xn − Yn is a multiple of 13 and 0 ≤ Yn ≤ 12. Then Yn is a Markov chain with transition kernel P defined as follows: P (i, j) = 1/6 if i + 1 ≤ j ≤ i + 6 or i + 1 − 13 ≤ j ≤ i + 6 − 13, and P (i, j) = 0 otherwise. In words, if the states from 0 to 12 are arranged around a circle in the natural order clockwise, then P (i, j) = 1/6 as long as j is located within the distance between 1 to 6 from i, moving clockwise. It is not hard to check that the chain is an aperiodic and irreducible and that the uniform distribution on {0, . . . , 12} 1. is invariant for it. Hence, the limit is equal to 13 1 2 [20 Points]. (a) Solve Exercise 2.6 (a) in the textbook. (b) Solve Exercise 2.6 (b) in the textbook. (b) Solve Exercise 2.6 (c) in the textbook. Solution: (a) p(i, i + 1) = p and p(i, 0) = 1 − p. (b) We have: π0 = (1 − p) ∞ X πi = 1 − p, i=0 and, for i ∈ N, πi = pπi−1 . Using induction, πi = pi (1 − p). (c) Since the chain is positive recurrent, E(Tk ) < ∞. Once the state k − 1 reached, in the next step the chain moves to k with probability p and returns to zero with probability q. Therefore, E(Tk ) = E(Tk−1 ) + E(Tk − Tk−1 ) = E(Tk−1 ) + p · 1 + q · 1 + E(Tk ) = E(Tk−1 ) + 1 + qE(Tk ). Hence, E(Tk ) = 1 1 + E(Tk−1 ). p p Iterating the last identity, we obtain 1 1 1 1 11 1 1 1 + E(Tk−1 ) = + + E(Tk−2 ) = + 2 + 2 E(Tk−2 ) p p p p p p p p p k−1 X 1 = ... = p−j + k−1 E(T1 ). p j=1 E(Tk ) = Since T1 is a geometric random variable with parameter p (i. e., P(T1 = m) = q m−1 p for m ≥ 1, we have E(T1 ) = p−1 . Thus E(Tk ) = k X j=1 p−j = 1 1 − p−k 1 −k = p − 1 . p 1 − p−1 q 2 2 [20 Points]. [20 points] (a) Solve Exercise 2.7 (a) in the textbook. (b) Solve Exercise 2.7 (c) in the textbook. Solution: To establish whether or not the underlying Markov chain is transient, it is convenient to use the criterion stated as the Fact on p. 50. Let z = 0 and assume that α(x) defined by Equations (2.2)–(2.4) on p. 49 exists. In particular, for x > 0 we have: α(x) = ∞ X p(x, y)α(y) = p(x, 0)α(0) + p(x, x + 1)α(x + 1) y=0 = p(x, 0) + p(x, x + 1)α(x + 1), and hence α(x + 1) = α(x) · p(x, 0) 1 − . p(x, x + 1) p(x, x + 1) Using induction (iterating the above inequality), it is not hard to verify that α(2) = α(1) p(1, 0) − p(1, 2) p(1, 2) and, for x > 1, x−1 x x X Y Y p(x, 0) p(y, 0) 1 1 α(x + 1) = − − + α(1) · p(x, x + 1) y=1 p(y, y + 1) t=y+1 p(t, t + 1) p(t, t + 1) t=1 Using the probabilistic meaning of α(x) = Px (Xn = 0 for some n ≥ 0), one conclude that α(x) is a non-increasing function of x given by α(x) = 1 − ∞ Y p(t, t + 1). (1) t=x (a) It follows from (1) that α(x) ≡ 1 and hence the chain is recurrent. To see whether the chain is positive- or null-recurrent, let T = inf{k > 0 : Xk = 0} and compute E0 (T ) = = ∞ X n=2 ∞ X n=2 ∞ n−2 X Y nP0 (T = n) = n p(t, t + 1) · p(n − 1, 0) n=2 n n−2 Y t=0 t=0 ∞ X n t+1 · = t+2 n+1 Thus the chain is null-recurrent. 3 n=2 n = +∞. n+1 (c) It follows from (1) that α(x) = 1 − ∞ Y 1− t=x 1 , t2 + 2 and hence limx→∞ α(x) = 0. Thus the chain is transient. 4 [20 Points]. (a) Solve Exercise 2.8 (a) in the textbook. (b) Solve Exercise 2.8 (c) in the textbook. Solution: (a) a is the minimal solution of φ(a) = a in (0, 1]. We have: a = p0 + p1 a + p2 a2 ⇐⇒ 35a2 − 60a + 25 = 0 Hence a = 5/7. (c) We have, µ = 0.05 + 0.01(2 + 3 + 6 + 13) = 0.29 < 1. Thus the extinction is certain. 5 [20 Points]. Solve Exercise 2.16 in the textbook. Solution: (a) By the definition of the invariant measure, π(n) = ∞ X π(m)p(m, n) = m=0 ∞ X π(m)pn−m , n ≥ 1. (2) m=0 Since pn−m = 0 if n − m > 1, it follows that π(n) = P∞ m=n−1 π(m)pn−m . (b) Let Y be a random variable valued in the set of non-negative integers, such that P(Y = n) = qn , n ≥ 0. Then E(Y ) = ∞ X qn n = n=0 = − ∞ X p1−n n = − n=0 n=0 X ∞ X p1−n (1 − n) + ∞ X p1−n n=0 pk k + 1 = −µ + 1 > 0, k since it is given that µ < 0. To conclude that the equation α = E(αY ) has a root within the interval (0, 1), one can use the same argument as for the super-critical branching processes. (c) This involves a guess. The correct answer isPπ(n) = cαn which can be validated directly using (2) together with the fact that ∞ n=0 π(n) = 1. The latter also implies c = (1 − α)−1 . 4