1. Chapter 6 Problems P (a) Because P{X < k} = k−1 j=0 f(j) for k > 1 and P{X < 0} = 0, ∞ ∞ ∞ ∞ k−1 X X X X X sk f(j) P{X < k}sk = f(j) sk = k=0 k=1 = ∞ X j=0 (b) P{X > k} = j=0 j=0 k=j+1 sj+1 s f(j) = G(s). 1−s 1−s P∞ f(j) for all k > 1; therefore, ! j ∞ ∞ ∞ ∞ X X X X X k k k P{X > k}s = f(j) s = s f(j) j=k k=0 k=0 = 2. j=0 ∞ X 1 − sj+1 j=0 = j=k 1−s k=0 f(j) = 1 1−s ∞ X f(j) − s j=0 ∞ X sj f(j) j=0 1 (1 − sG(s)) . 1−s (a) Here, n is a fixed integer > 1: n X sk 1 G(s) = = n n k=1 n X ! k s −1 k=0 1 = n 1 − sn+1 −1 1−s = s − sn+1 . n(1 − s) (b) Again, n is a fixed integer > 1: G(s) = n X k=−n n n X X sk 1 1 = sk = s−n sk+n . 2n + 1 2n + 1 2n + 1 k=−n k=−n A change of variables [j := k + n] shows that 2n s−n s−n X j 1 − s2n+1 s−n + sn+1 G(s) = s = = . 2n + 1 2n + 1 1−s (2n + 1)(1 − s) j=0 (c) Write f(k) = k−1 − (k + 1)−1 to find that Z ∞ ∞ ∞ Zs ∞ X X X sk X sk 1 s k G(s) = − = xk−1 dx − x dx k k+1 s 0 k=1 k=1 k=1 0 k=1 Z ∞ Z Zs X Zs ∞ 1 s 1 1 sX k dx k−1 − − 1 dx = x dx − x dx = s 0 s 0 1−x 0 k=1 0 1−x k=1 1 1 1 1 1 = ln − ln −s =1− − 1 ln . 1−s s 1−s s 1−s 1 (d) This one is nonsense, sadly, because f(k) < 0 if k < 0; therefore, f is not a probability mass function. (e) Recall that Z denotes the collection of all integers [positive as well as negative]. Therefore, ! ∞ −1 ∞ X X 1 − c X |k| k 1 − c k k 1+ (cs) + (s/c) c s = G(s) = 1+c 1+c k=1 k=−∞ k=−∞ ∞ X 1−c cs cs 1−c c/s j = 1+ . 1+ + (c/s) = + 1+c 1 − cs 1+c 1 − cs 1 − (c/s) j=1 Is it really the case that wise all is lost. 3. P k∈Z f(k) = 1? This is important, for other- (a) Let N be a Poisson(λ) r.v., and recall that GN (s) = exp{−λ(1 − s)} (Example 12, p. 238). Therefore, exp {−λ(1 − GX (s))} = GN (GX (s)). This and Theorem 13 (p. 242) together show that exp{−λ(1 − GX (s))} P is the generating function of N i=0 Xi , where the Xi ’s are independent random variable, all independent of Y; it is convergent if and only if GX (s) < ∞. [Question: Does it matter that Theorem 13 is P about N i=1 Xi , that is the sum starts from one there? The answer is “no,” but you should sort this out.] (b) Since 0 < s < 1, G(s) := sin(πs/2) is positive increasing [as a function of s], and at most one. The Taylor series for sin is ∞ sin x = x − X (−x)j+1 x3 x5 + ··· = . 3! 5! (2j + 1)! j=1 Therefore, G(s) = ∞ X (−πs/2)j+1 j=1 (2j + 1)! . If this were a probability generating function, then the coefficient of sj+1 is f(j+1); but that produces nonsense; for example plug in j = 1 to obtain f(2) = −π/(2 × 5!) < 0, which cannot be. (c) Throughout, I assume that r is a positive integer. If p = 0 then G(s) = 1 for all s. Therefore, G is the generating function for X ≡ 0 [that is, f(0) = 1 and f(s) = 0 for s 6= 0]. If p = 1, then q = 0 and so G(s) = 0 for all s. In that case, G(1) = 0 6= 1, and so G is not a p.g.f. 2 Next consider the interesting case that 0 < p < 1. Let X1 , . . . , Xr be independent random variables each with p.g.f., q GXj (s) = . 1 − ps According to Example 4 (p. 233), if |s| < p−1 , then GXj are well defined. Let Y = X1 + · · · + Xr to find that as long as |s| < p−1 , GY (s) = E sX1 +···+Xr = E sX1 × · · · × E sXr r q = GX1 (s) × · · · × GXr (s) = . 1 − ps See also Example 6 (p. 245). (d) This is a Binomial(r , p) example; see Example 11, p. 238. (e) Yes as long as |s| < 1; see Example 2 (p. 259). (f) According to Example 16 (p. 242), if P{X = k} = pk k log(1/q) (k > 1), and 0 < p < 1 with q := 1 − p, then GX (s) = log(1 − sp) . log q And of course this is defined only when 1 − sp > 0; i.e., when s < p−1 . Therefore, if α log(1 + βs) is a p.g.f., then β = −p and α = 1/ log q, with s < p−1 . Chapter 7 Problems 1. I will assume that α < β; this can be done without incurring a loss in generality. If α < x < β, then we want Zβ Zβ 1 = c(α , β) (x − α)(β − x) dx = c(α , β) −αβ + (α + β)x − x2 dx α α β2 − α2 β3 − α3 − = c(α , β) −αβ(β − α) + (α + β) 2 3 2 1 β + αβ + α2 2 = c(α , β)(β − α) −αβ + (α + β) − 2 3 2 α + β2 β2 + αβ + α2 − = c(α , β)(β − α) 2 3 (α − β)2 1 = c(α , β)(β − α) × = c(α , β)(β − α)3 . 6 6 Therefore, c(α , β) = 6(β − α)−3 . If x is not between α and β then f ≡ 0. 3 6. We want Za exp −x − e−x dx F(a) = for all real numbers a. −∞ Change variables [z := e−x ]: For all a ∈ R, Z exp(−a) e−w dw = exp(−e−a ). F(a) = −∞ 9. First of all, a P X>x+ x P{X > x + ax−1 } X>x = . P{X > x} The numerator and the denominator both tend to zero as x → ∞. Therefore, by the l’Hôpital’s rule, a lim P X > x + x→∞ x d −1 } dx P{X > x + ax . d x→∞ dx P{X > x} X > x = lim Now, d d 1 2 (1 − FX (x)) = −fX (x) = − √ e−x /2 . P{X > x} = dx dx 2π Also, d d a a a 1 − FX x + = −1 + 2 fX x + . P{X > x + ax−1 } = dx dx x x x Since 2 a 1 a2 x + 2a + (a/x)2 fX x + = √ exp − = fX (x) exp −a − 2 , x 2 2x 2π it follows that a lim P X > x + x→∞ x −fX (a + ax−1 ) a2 X > x = lim = lim exp −a − 2 = e−a . x→∞ x→∞ −fX (x) 2x 4