Lecture 26 1. Moment generating functions Let X be a continuous random variable with density f . Its moment generating function is defined as ! ∞ tX M (t) = E[e ] = etx f (x) dx, (19) provided that the integral exists. −∞ Example 26.1 (Uniform(0 , 1)). If X = Uniform(0 , 1), then ! 1 et − 1 M (t) = E[etX ] = etx dx = . t 0 Example 26.2 (Gamma). If X = Gamma(α , λ), then ! ∞ λα α−1 −λx M (t) = etx x e dx Γ(α) 0 ! ∞ λα = xα−1 e−(λ−t)x dx. Γ(α) 0 If t ≥ λ, then the integral is infinite. On the other hand, if t < λ, then ! ∞ z α−1 dz λα (z = (λ − t)x) M (t) = e−z Γ(α) 0 (λ − t)α−1 λ−t ! ∞ λα = z α−1 e−z dz Γ(α) × (λ − t)α 0 #$ % " Γ(α) = λα (λ − t)α . 91 92 Thus, 26 *α ) λ λ−t M (t) = ∞ if t < λ, otherwise. Example 26.3 (N (0 , 1)). If X = N (0 , 1), then ! ∞ 1 2 M (t) = √ etx e−x /2 dx 2π −∞ ) 2 * ! ∞ x − 2tx 1 exp − dx =√ 2 2π −∞ ) 2 * ! 2 et /2 ∞ x − 2tx + t2 =√ exp − dx 2 2π −∞ * ) ! 2 et /2 ∞ (x − t)2 =√ dx exp − 2 2π −∞ ! 2 et /2 ∞ −u2 /2 =√ e du (u = x − t) 2π −∞ 2 /2 = et . 2. Relation to moments Suppose we know the function M (t) = E exp(tX). Then, we can compute the moments of X from the function M by successive differentiation. For instance, suppose X is a continuous random variable with moment generating function M and density function f , and note that ! d ∞ tx d + tX , E[e ] = e f (x) dx. M # (t) = dt dt −∞ Now, if the integral converges absolutely, then a general fact states that we can take the derivative under the integral sign. That is, ! ∞ . xetx f (x) dx = E XetX . M # (t) = −∞ The same end-result holds if X is discrete with mass function f , but this time, / . M # (t) = xetx f (x) = E XetX . Therefore, in any event: x M # (0) = E[X]. In general, this procedure yields, . M (n) (t) = E X n etX . 93 2. Relation to moments Therefore, M (n) (0) = E [X n ] . Example 26.4 (Uniform). We saw earlier that if X is distributed uniformly on (0 , 1), then for all real numbers t, M (t) = Therefore, M # (t) = whence tet − et + 1 , t2 et − 1 . t M ## (t) = t2 et − 2tet + 2et − 2 , t3 tet 1 tet − et + 1 = lim = , 2 t$0 2t t$0 t 2 EX = M # (0) = lim by l’Hopital’s rule. Similarly, - . t2 et − 2tet + 2et − 2 t2 et 1 E X 2 = lim = lim = . t$0 t$0 3t2 t3 3 Alternatively, 0these can be checked by direct computation, using the fact 1 that E[X n ] = 0 xn dx = 1/(n + 1). A discussion of physical CLT machines . . . .