Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Theorem The beta(b, b) distribution converges to the normal distribution when b → ∞. Proof (by Professor Robin Ryder in the CEREMADE at Université Paris Dauphine) Let the random variable X have the beta(b, b) distribution with probability density function fX (x) = Γ(2b)xb−1 (1 − x)b−1 Γ(b)Γ(b) 0 < x < 1, where b is a real, positive parameter. The mean of X is E[X] = 1/2 and the variance of X is V [X] = 1/4(2b + 1). Substract the mean and divide by the √ standard deviation before taking the limit. So consider the transformation Y = g(X) = 2 2b√+ 1(X − 1/2),√which is a one-to-one transformation from√A = {x|0 < x < 1} to B = {y| − 2b + 1 < y < 2b + 1} with inverse X = g −1 (Y ) = X/2 2b + 1 + 1/2 and Jacobian 1 dX = √ . dY 2 2b + 1 Then the probability density function of Y is 1 1 y √ fY (y) = √ fX + 2 2b + 1 2 2b + 1 2 b−1 b−1 y y 1 1 Γ(2b) 1 + √ − √ = √ · 2 2 2b + 1 2 2b + 1 Γ(b)2 2 2 2b + 1 b−1 √ √ y2 Γ(2b) 1 1 − − · = √ 2b + 1 < y < 2b + 1. 2 2b + 1 Γ(b)2 4 4(2b + 1) p We now apply Stirling’s formula Γ(z) = 2π/z(z/e)z (1 + O(1/z)) and get q b−1 2π 2b 2b 1 y2 1 1 2b e fY (y) = √ 1+O − · 2π b 2b b 4 4(2b + 1) 2 2b + 1 b e b−1 2b b 1 1 y2 2 1+O = √ √ − · √ b 4 4(2b + 1) 2b 2b + 1 2 2π b−1 1 y2 1 1 b−1 1+O − = √ 4 b 4 4(2b + 1) 2π b−1 1 1 y2 1+O = √ 1− 2b + 1 b 2π 2 √ √ 1 y 1 = √ exp − 1+O − 2b + 1 < y < 2b + 1. 2 b 2π Thus, in the limit b → ∞, fY (y) converges pointwise to the probability density function of a standard normal random variable. By Scheffé’s theorem, Y converges in distribution to the standard normal distribution. 1