Ch5. Homework Solution P173 #1 Let X 1 , X 2 ,... be a sequence of independent random variables with E ( X i ) and Var ( X i ) i2 . Show that if n 2 i 1 i2 0 , then n X in probability. sol VarX Var ( 0 n 1 n 1 1 xi ) 2 Var ( xi ) 2 n i 1 n n i 1 n 2 i 1 as n ( xi , x j independen t for i j , Cov( xi , x j ) 0 i j ) 由 Chebyshev’s inequality: >0 n P( X ) VarX 2 i 1 2 2 i n 2 0 as n X n P173 #2 Let X i be as in problem 1 but with E(X i ) μ i and n 1 i 1 μ i μ. n Show that X μ in probabilit y. sol 令X i be a Problem 1. but with EX i μ i and μ n 1 n μ i , 假定 μ n μ n i 1 as 由Chebyshev' s inequality : ε 0 n 4 σ i2 Var X n ε P( X n μ n ) i21 2 0 as n ε 2 2 n ε ( ) 2 ε ε 0, δ 0, n 1 Ν s.t. n n 1 P( X n μ n ) δ 2 μ n μ as n n 2 Ν s.t. n n 2 μ n μ ε 2 ε 故若能証出n n 2 X n μ ε X n μ n , 則由n max n 1 , n 2 2 ε P( X n μ ε) P( X n μ n ) δ 2 P 得証 X n n μ ε claim : n n 2 X n μ ε X n μ n 2 假定 X n μ ε, 則n n 2 ε X n μ X n μ n X n μ X n μ n Xn μn ε 2 ε 2 ε 因此, n n 2 X n μ ε X n μ n 2 P173 #3 Suppose that the number of insurance claims, N, filed in a year is Poisson distributed with E(N)=10,000 . Use the normal approximation to the Poisson to approximate P(N>10,200). sol Let N be a Poisson random var iable with mean 10000, We find P( N 10200)by s tan dardizing : N 10000 P( N 10200) P( 10000 10200 10000 10000 ) 1 (2) 0.0228 where is the s tan dard normal c.d . f . 6.Using moment-generating functions, show that α the gamma distribution with parameters α and λ, properly standardized, tends to the standard normal distribution. pf﹕X~gamma(α , λ ) M x (t ) =( t E(X)= Var(X)= 2 t<λ )2 X E( X ) = X Let Z= = Var ( X ) 2 X M z (t ) = e = e t t Mx( ( t )= e t t ( t ) ) 1 1 log M z (t ) t [log 2 log( 2 t )] 1 1 = t log log( 2 t ) 2 = t 1 3 1 t2 2 log [ log 2 t 1 2 ( t 3 ) ] 2 2 2 3 1 = t2 2 2 ( t3) 2 3 → t2 2 M z (t ) → e as t2 2 所以 Z ~ U ( 0 ,1) 14.Answer Problem 13 under the assumption that the drunkard has some idea of where he wants to go so that he steps north with probability 2/3 and south with probability 1/3. Ans:令 X i 為在第 i 分鐘內的位移(以北為正向以南為負向) P( X i =50)=2/3 P( X i = - 50)=1/3 E( X i ) =50×(2/3)+(- 50)×(1/3)=50/3 E( X i ) 50 2 (2 / 3) (50) 2 (1 / 3) 2500 Var( X i )= E ( X i2 ) [ E ( X i )] 2 20000 / 9 60 S 60 = X i i 1 令 Zn = Sn i n = S 60 (50 / 3) 60 60 (20000 / 9) ~ N (0 , 1) by C.L.T. 所以 S 60 ~(1000 , 400000/3) 最有可能在原位置北方 1000cm 1 19.Use the Monte Carlo method with n=100and n=1000 to estimate cos(2X )dx . 0 Compare the estimates to the exact answer. Ans: Iˆ( f ) 1 1000 cos( 2xi) 1000 i 1 1 I ( f ) c o s2(x)dx 0 Assume X i ~U(0 ,1) iid f( xi ) iid I ( f ) =E(f(x)) Iˆ 1 1000 f ( xi ) 1000 i 1 1 1 1 1 1 1 f ( x)dx c o s2(x)dx c o s2(x)d (2x) s i n2(x) 2 0 2 0 0 E ( f ( x)) 0 =0 1 1 0 0 1 c o s4x dx 2 0 1 Var ( f ( x)) ( f ( x) 0) 2 dx c o 2s(2x)dx 1 x 1 = sin( 4x) =1/2 2 8 0 P( Iˆ( f ) I ( f ) ) P( Iˆ( f ) I ( f ) ) P( 1000 / 2 Iˆ( f ) I ( f ) 1000 / 2 1000 / 2 22.Using th central limit theorem to find △ such that P(∣ Iˆ (f)-I(f)∣≦ △ )= 0.05, where Iˆ (f) is the Monte Carlo estimate of 1 cos(2X )dx based on 1000 0 points. Ans: X 1 , X 2 ,... X 1000 ~ U (0 ,1) 且獨立 令 Y=f(x)=cos(2πx) and Y1,Y2,……,Y1000 獨立 1 n Iˆ( f ) Yi Yn n i 1 1 E (Y ) E ( f ( x)) cos( 2x)dx I ( f ) 0 E (Y ) E (Y ) 0 c o s4(x) 1 1 dx 2 2 0 1 1 E (Y ) E[( f ( x)) ] c o s 2(x)dx 2 2 2 0 Var(X i )=1/2 By C.L.T Zn= Yn E (Yn ) Var (Yn ) Iˆ( f ) I ( f ) Var (Y ) / n Iˆ( f ) I ( f ) 1 / 2000 ~ N (0 ,1) 故 P( Iˆ( f ) I ( f ) ) P( Iˆ( f ) I ( f ) ) P( 1 / 2000 Iˆ( f ) I ( f ) 1 / 2000 ( 2000) ( 2000) 2 ( 2000 ) 1 0.05 ( 2000) 0.525 2000 0.0 6 2 7 5 0.0 0 1 4 0 3 1 3 3 27.Prove that if a n a ,then (1+ a n / n) n e a . Pf:令 f n (x) (=(1+( x / n)) 2 e x f ( x)為一連續函 數 1 / 2000 ) ) P an a 0 P( a n a ) ) 0 f 在 x a 連續 0, 0, st 0 x a f ( x) f (a) { f ( x) f (a) } { x a 故 0. P( f (a n ) f (a) ) P( a n a ) 0 P P 即 f (a n ) f ( a ) e an ea (1 an n P ) e a n