MATH 5010–001 SUMMER 2003 SOLUTIONS TO ASSIGNMENT 10 Problems from pp. 379–383: 3. Note that f (x, y) = 1 if x and y are both between zero and one; else, f (x, y) = 0. Therefore, as long as α > −1, Z 1Z 1 α E {|X − Y | } = |x − y|α dx dy 0 0 Z 1 Z 1 α =2 (y − x) dy dx (by symmetry) 0 Z 1 x 1−x Z =2 α z dz 0 2 = 1+α Z 0 1 0 dx (1 − x)1+α dx = (z = y − x) 2 . (1 + α)(2 + α) 16. Note that X = g(Z), where g(z) = z if z > x and g(z) = 0 if z ≤ x. So, Z ∞ −z2 /2 2 e−z /2 e g(z) √ z √ dz = dz E[X] = 2π 2π −∞ x Z ∞ −u e √ du = (u = z 2 /2) 2π x2 /2 Z ∞ 2 e−x /2 . = √ 2π 30. Because X and Y have the same mean µ and the same variance σ 2 , n o 2 E (X − Y )2 = E ((X − µ) − (Y − µ)) = E (X − µ)2 + E (Y − µ)2 − 2E{X − µ}E{Y − µ}. The last term is zero, and the first two terms are both σ 2 . So E{(X − Y )2 } = 2σ 2 . 38. Cov(X, Y ) = E(XY ) − E(X)E(Y ). So we have three computations to make. First, note that Z ∞Z x Z ∞ −2x E(X) = 2e dy dx = 2xe−2x dx 0 0 Z 0 1 ∞ −z = ze dz (z = 2x) 2 0 1 = . 2 1 (1) Similarly, Z ∞ Z x 2y −2x E(Y ) = dy dx = e x 0 0 Z ∞ 1 = xe−2x dx = . 4 0 Finally, Z ∞ Z x E(XY ) = −2x 2ye 0 Z = 0 ∞ Z ∞ 0 ∞ dy dx = xe−2x dx = Z x y dy dx 0 (2) Z 0 2 −2x e x −2x Z 2e 0 x y dy dx 0 1 . 4 Therefore, this and equatinos (1) and (2), together yield: Cov(X, Y ) = 1 4 − 1 2 · 1 4 = 18 . 75. Because MY (t) = exp(2et − 2) and MX (t) = ( 14 + 34 et )10 (note the typo in some of your texts), you can recognize the distribution of X as binomial with n = 10 and p = 34 , and that of Y as Poisson with λ = 2. Now to our regular programming · · · 75. (a) X and Y are positive integers (albeit random variables), so that X + Y can only be 2 if {X = 0, Y = 2}, {X = 1, Y = 1}, or {X = 2, Y = 0}. Therefore, P {X + Y = 2} = P {X = 0} · P {Y = 2} + P {X = 1} · P {Y = 1} + P {X = 2} · P {Y = 2} 0 10 1 9 2 10 3 10 3 1 1 21 −2 2 = + ·e · e−2 0 1 4 4 2! 4 4 1! 2 8 10 3 1 20 · e−2 + 2 4 4 0! 467 = 10 2 = 0.000059. 4 e 76. We are asked to find MX,Y (s, t) = E{esX+tY } for all s and t. Let Z denote the outcome of the second die and note that Y = X + Z. Therefore, o n o n sX+t(X+Z) (s+t)X tZ =E e e MX,Y (s, t) = E e = MX (s + t)MZ (t). But X and Z have the same distribution, so MX,Y (s, t) = MX (s + t)MX (t). (3) It remains to find the function MX . But this is easy: For any real number r, 6 rX er + e2r + · · · + e6r 1X r j = = (e ) . MX (r) = E e 6 6 j=1 2 (4) This is a geometric sum, which Pn Pn I evaluate next. Recall that for any u, j=0 uj = (un+1 − 1)/(u − 1). Therefore, j=1 uj = (un+1 − 1)/(u − 1) − 1 = (un+1 − u)/(u − 1) = u(un − 1)/(u − 1). Thanks to (4) (with u = er ), 1 er (enr − 1) MX (r) = . 6 er − 1 (5) Use (3), and apply (5), once with r = s + t and once with r = t, to obtain: " # et (ent − 1) 1 es+t en(s+t) − 1 MX,Y (s, t) = . 36 es+t − 1 et − 1 77. (a) We want MX,Y (s, t) = E esX+tY Z ∞Z ∞ −y −(x−y)2 /2 e sx+ty e √ = e dx dy 2π 0 −∞ Z ∞Z ∞ −y −z 2 /2 e s(z+y)+ty e √ = dz dy (z = x − y) e 2π 0 −∞ ! Z ∞ Z ∞ −z 2 /2 −[1−(s+t)]y sz e = dz dy. e e √ 2π 0 −∞ Now, the inside integral equals E(esZ ) where Z is N (0, 1). Therefore, this term is equal to exp(s2 /2). Thus, s2 /2 Z ∞ MX,Y (s, t) = e e−[1−(s+t)]y dy 0 2 es /2 , if s + t < 1, = 1 − (s + t) +∞, otherwise. Theoretical Problem from p. 391: 13 (a). Let Ij = 1Pif a record occurs at j, and else Ij = 0. We are asked to show that n E[I1 + · · · + In ] = j=1 (1/j). I will show that E[Ij ] = 1/j; this is equivalent to showing that E(Ij ) = 1/j which I will prove next. Note that E(Ij ) is the probability that Xj > max(X1 , . . . , Xj−1 ); the seemingly innocent strict inequality is due to the fact that the X’s are continuous (why does this follow?), and is very important in what is about to happen. 3 Note that X1 , . . . , Xj are exchangeable; i.e., the joint density of (X1 , · · · , Xj ) is the same as that of any (and all) of the following: • (X1 , · · · , Xj−2 , Xj , Xj−1 ); • (X1 , · · · , Xj−3 , Xj , Xj−2 , Xj−1 ); .. . • (Xj , X1 , · · · , Xj−1 ). In particular, P {Xj > max(X1 , . . . , Xj−1 )} is the probability that, in any of the above items, the last random variable is greater than the maximum of the first j − 1 rv’s. There are j equal probabilities here. Moreover, they represent disjoint events whose union is everything (since one of the X’s must be maximum–the continuity of the X’s is coming in here; sort the logic out). This means that jE(Ij ) = 1, which is equivalent to saying that E(Ij ) = 1/j, as asserted. 4