Probability: Solutions to Select Problems Davar Khoshnevisan Department of Mathematics University of Utah Salt Lake City, UT 84112–0090 davar@math.utah.edu http://www.math.utah.edu/˜davar October 7, 2013 Chapter 6 Independence 6.1. First consider the case that f 0 > 0. For all ! 2 ⌦, Z X(!) Z1 f(X(!)) = f 0 (x) dx = 1{06x6X} (!)f 0 (x) dx. 0 -1 Assuming that we can handle product-measurability issues, it follows from Fubini–Tonelli that Z 1 Z1 E[f(X)] = E 1{06x6X} (!)f 0 (x) dx = P{X > x}f 0 (x) dx. 0 -1 0 0 0 0 This is the desired result Rx 0when f > 0. In general, we write f = f+ - fand define f(±) (x) = 0 f± (z) dz. The preceding development yields, Z1 ⇥ ⇤ 0 E f(±) (X) = f± (x)P{X > x} dx. 0 But f+ - f- = f (why?), whence the Problem. It suffices to prove the asserted product measurability. Evidently, f 0 is a product-measurable function of (x, !); so is x 7! 1[0,1) (x). So it suffices to prove that (x, !) 7! 1{X>x} (!) is product-measurable. Define In (x, !) = X 1[i/n,(i+1)/n) (x)1[j/n,(j+1)/n) (X(!)). 16i<j Each In is manifestly product-measurable. Therefore, so is 1{X>x} (!) = limn!1 In (x, !). Therefore, 1{y<X6x} = 1{y<X} - 1{x<X} is measurable for all x > y; hence, so is 1{X=x} = limy"x 1{y<X6x} . Finally, we see that 1{X>x} (!) = 1{X>x} (!) + 1{X=x} (!) is measurable. 6.3. Note that Z 1 Z 1 Z1 Z1 P{X > x , Y > y} dx dy = E 1{X>x}\{Y>y} dx dy = E[XY]. 0 0 0 13 0 14 CHAPTER 6. INDEPENDENCE The result follows from this and Lemma 6.9. 6.5. According to Problem 1.3 of Chapter 1 we can construct a probability space (⌦, F, P) on which there are events A, B, C that are pairwise independent but not independent. Define X := 1A , Y := 1B , and Z := 1Z to finish. 6.8. To prove Cor. 6.19 we apply Lemma 6.13 to find that if X and Y are independent then E[XY] = EXEY. Equivalently, Cov(X , Y) = 0. Next, suppose E[Xi ] = 0, and Xi 2 L2 (P). Then, 2 ! !2 3 n n X X Var Xi = E 4 Xi 5 i=1 2 = E4 = n X i=1 n X i=1 X2i + XX 16i6=j6n Var(Xi ) + i=1 3 Xi Xj 5 XX Cov(Xi , Xj ). 16i6=j6n If E[Xi ] 6= 0, then replace Xi by Yi := Xi - E[Xi ] to find that ! n n X X XX Var Yi = Var(Yi ) + Cov(Yi , Yj ). i=1 i=1 16i6=j6n Now let Y and Z be random variables, and a, b 2 R. Then, you can check that: Var(Y + a) = Var(Y), and Cov(Y + a , Z + b) = Cov(Y , Z). Together, these remarks prove Corollary 6.20. 6.9. Let Z denote an independent N(0, 1), and define Xn = X + (Z/n). The variable Xn has all of the desired properties. P P1 1/p p } = 6.10. (i))(ii) If X1 2 Lp (P) then 1 n=1 P{|X1 | > n=1 P{|Xn | > "n p p p " n} 6 E{|X1 | }/" . So by the Borel–Cantelli lemma, with probability one, |Xn | 6 "n1/p for all n large. This proves that |Xn |/n1/p ! 0 a.s. (ii)) (iii) This follows from the following real-variable Fact. If an /n⇢ ! 0 for some ⇢ > 0, then max16j6n (aj /n⇢ ) ! 0. Proof: If not, then maxj6n aj > "n⇢ infinitely often. But there exists n0 such that for all n > n0 |an | 6 ("/2)n⇢ . So maxj6n0 aj > "n⇢ for infinitely-many n’s, which is patently nonsense. (iii))(i) Because (iii) implies (ii), it suffices to prove that (ii) implies (i). But this too is Borel–Canelli (as in i)ii).