Math 136 / Stat 219 - Stochastic Processes Homework Set 2, Autumn 2011, Due: October 12 1. Exercise 1.3.21. ANS: (i) The triangle inequality for the Lq norm is ∥X + Y ∥q ≤ ∥X∥q + ∥Y ∥q (X, Y ∈ Lq ). So q.m. |∥Xn ∥q − ∥X∥q | ≤ ∥Xn − X∥q , and since Xn → X implies ∥Xn − X∥q → 0, it can be concluded that ∥Xn ∥q → ∥X∥q , which is equivalent to E|Xn |q → E|X|q . 1.m (ii)Since q ≥ 1, corollary 1.3.19 indicates Xn → X, thus E|Xn − X| → 0. Applying Jensen’s inequality with g(x) = |x| to the R.V. Xn − X, we obtain that E|Xn − X| ≥ |E(Xn − X)|, so EXn → EX. (iii)Let Xn have the distribution P (Xn = −1) = P (Xn = 1) = 0.5 for all n, X = 0. It is clear that EXn = 0 = EX, but E|Xn − X| = 1. 2. Exercise1.4.12 ANS: For any bounded function g that is continuous on the range of f (X), i.e., g is continuous on {f (ω) : ω ∈ Ω}. Let h(x) = g(f (x)). Then it is clear that h is bounded and h is a continuous function on the range of X. By Proposition 1.4.11 and the assumption that Xn → X in law, we can obtain that E(h(Xn )) → E(h(X)). That is to say, Eg(f (Xn )) → Eg(f (X)) for any bounded g that is continuous on the range of f (X). Applying Proposition 1.4.11 again, we conclude that f (Xn ) → f (X) in law, as required. 3. Exercise 1.4.14. ANS: Note that the event {Mn ≤ x} is equivalent with the event {Ti ≤ x : 1 ≤ i ≤ n}. Therefore, ∏ P(Mn ≤ x) = P(Ti ≤ x , for all i = 1, 2, . . . n) = ni=1 P(Ti ≤ x), where in the last step we used ∏ independence of the Ti ’s. Thus, P(Mn ≤ x) = ni=1 FTi (x) = (1 − e−λx )n . So, we have P(Mn − an ≤ y) = [1 − e−λ(y+an ) ]n = exp{n × log[1 − e−λ(y+an ) ]} (if the limit exists, an → +∞) = exp{−ne−λ(y+an ) (1 + o(1))} −λy So if eλan = n, P(Mn − an ≤ y) → e−e possible solution is an = log(n) λ , , which is a distribution function on the real line. Therefore a and in this case M∞ has the distribution function e−e −λy . Question: What is EMn ? 4. Exercise 1.4.17. d ANS:(a)Let X be a bernoulli random variable, P(X = 1) = 0.5, and Y = 1 − X, therefore X = Y , but 1 P(X = Y ) = P(X = 0.5) = 0. (b)Let Xn have the uniform distribution on [0, n1 ] and X = 0, then Xn → X, but it is obvious that Xn has density but X doesn’t. (c)(In this part, let [x] denote the Gaussian function, whose value is the greatest integer which doesn’t exceed x.) t P(pZp > t) = P(Zp > [ ]) p [ pt ] = (1 − p) → e−t (p → 0) Thus P(pZp ≤ t) → 1 − et for every t ∈ [0, +∞) , which is the cdf of Exp(1). This means the cdf of pZp converges to the cdf of T pointwise, and so the statement of the problem can be concluded. ∫ (d)Since both fn (x) and f∞ (x) are probability density functions, it must be true that R fn (x)dx = ∫ ∫ ∫ R f∞ (x)dx = 1, which means R (f∞ (x) − fn (x))I{f∞ >fn } dx = R (fn (x) − f∞ (x))I{fn >f∞ } dx. Therefore ∫ ∫ |f∞ (x) − fn (x)|dx = 2 (f∞ (x) − fn (x))I{f∞ >fn } dx R ∫ R = gn (x)f∞ (x)dx R Since the RHS of the last equation is equal to Egn (X∞ ), gn (x) → 0 pointwise when n → ∞ and ∫ gn (x) ∈ [0, 2], it follows from Corollary 1.4.28 that R gn (x)f∞ (x)dx = Egn (X∞ ) → 0 as n → ∞. Let Fn (x) n ∈ N ∪ {∞} denote the cdf of Xn , then it follows that for every x ∈ R ∫ |Fn (x) − F∞ (x)| = | fn (x) − f∞ (x)dx| (−∞,x] ∫ ≤ |fn (x) − f∞ (x)|dx (−∞,x] ∫ ≤ |fn (x) − f∞ (x)|dx → 0 R Thus Xn → X∞ in law. 5. Exercise 1.4.44. ANS: We first show that if N is a R.V. that takes non-negative integer value, then EN = j). In order to see this, note that EN = ∞ ∑ i=1 iP(N = i) = ∞ ∑ i ∑ P(N = i) = i=1 j=1 ∞ ∑ ∞ ∑ j=1 i=j 2 P(N = i) = ∞ ∑ j=1 P(N ≥ j) . ∑∞ j=1 P(N ≥ Since ξi is positive, the monotone convergence theorem indicates that EX (M on) ∞ ∑ (indep.) ∞ ∑ = Eξi I{N ≥i} i=1 = Eξi EI{N ≥i} i=1 ∞ ∑ = Eξi P (N ≥ i) i=1 Eξ1 = ∞ ∑ P (N ≥ i) i=1 Eξ1 EN = Question: Suppose that the assumption is not N is independent with ξi . Instead, we assume that {N ≤ k} ∈ σ(X1 , . . . , Xk ) for all k ∈ IN. Would the same result hold? 3