• Tidak ada hasil yang ditemukan

9 Properties of Expectation

N/A
N/A
Protected

Academic year: 2024

Membagikan "9 Properties of Expectation"

Copied!
12
0
0

Teks penuh

(1)

SEOUL NATIONAL UNIVERSITY

School of Mechanical & Aerospace Engineering

446.358

Engineering Probability

9 Properties of Expectation

Recall

E[X] = X

x

xp(x) : descrete random variable E[X] =

Z

−∞

xf(x)dx : continuous random variable

If

P{a≤X≤b} = 1 then a≤E[X]≤b Proof. (Discrete random variable)

E[X] = X

x:p(x)>0

xp(x) X

x:p(x)>0

ap(x) =a X

x:p(x)>0

p(x) =a

Similarly, E[X] b Proposition

If X & Y have a joint probability mass function p(x, y), then E[g(X, Y)] = X

y

X

x

g(x, y) p(x, y)

If X & Y have a joint probability density function f(x, y), then E[g(X, Y)] =

Z

−∞

Z

−∞

g(x, y) f(x, y)dxdy

Whenever E[X] & E[Y] are finite,

E[X + Y] = E[X] + E[Y]

For random variables X & Y such that XY,

E[X - Y] 0 E[X]E[Y]

If E[Xi] is finite for alli= 1,2, ...., n, then

E[X1+...+Xn] = E[X1] +...+ E[Xn].

(2)

Example .

X1, ...., Xn: i.i.d. random variables having distribution function F and expected valueµ Such a sequence of random variables is said to constitute a sample from the

distribution F.

X : = Xn

i=1

Xi

n : Sample mean Then

E[X] =E

" n X

i=1

Xi n

#

= 1 nE[P

Xi] = 1 n

XE[Xi] = µ

∴The expected value of the sample mean = the mean of the distribution.

Whenµ is unknown, the sample mean is often used in statistics to estimate it.

Example .

A1, ..., An: Some events

indicator variablesXi =

½ 1 ifAi occurs 0 else

X: = Xn

i=1

Xi : # of the events Ai that occurs

Y: =

½ 1 if X1 0 else

X Y E[X]

Pn|{z}

i=1E[Xi]

| {z }

Pn i=1P[Ai]

E[Y]

|{z}

P{at least one of the Ai occure}

| {z }

P(Fn i=1Ai)

∴ P Ã n

G

i=1

Ai

!

Xn

i=1

P(Ai) : Boole’s inequality

(3)

Example . [A random walk in the plane]

Starting from a given point in the plane, taken a sequence of steps of fixed length but in a completely random direction and uniform over (0, 2π).

The expected square of the distance from the origin aftern steps ? Solution.

Let (Xi, Yi) denote the change in position at theith step. then Xi = cosθi

Yi = sinθi

θi ,i= 1, ..., n : indept, uniform (0, 2π) random variables Aftern steps,

the position = ÃXn

i=1

Xi, Xn

i=1

Yi

!

We are interested in

D2 = ³X Xi

´2

+ ³X

Yi

´2

= X ¡

Xi2 +Yi2¢

+ ³X X

i6=jXiXj +YiYj

´

= n + X X

i6=j(cosθicosθj + sinθisinθj) θi&θj (i6=j) : independent, and

E[cosθi] = 1 2π

Z 2π

0

cosu du= 0.

E[sinθi] = 1 2π

Z 2π

0

sinu du= 0.

∴ E[D2] = n.

(4)

Example . [Analyzing the quick-sort algorithm]

We are given a set ofndistinct valuesx1, ..., xn, and we want to sort them in increasing order.

Quick-Sort Algorithm

•n=2 Compares two values & puts them in an appropriate order.

•n >2 One of the element is randomly chosen (say, xi), then all the other values are compared withxi.

{...}

| {z }

smallar than xi

xi {...}

| {z }

larger than xi

repetition on these brackets continuous until all the values have been sorted.

Example .

5, 9, 3, 10, 11, 14, 8, 4, 17, 6 Choose one of them at random.

Suppose that 10 is chosen {5 9 3 8 4 6}

| {z }

say6is chosen

10 {11 14 17}

| {z }

say11is chosen

{5 3 4}

| {z }

say4is chosen

6 {9 8}

| {z }

say ... chosen

1011 {14 17}

| {z }

say ... chosen

{3}4 {5} 6 8 9 10 11 14 17

X : # of comparisons that it takes that algorithm to sortndistinct numbers.

E[X] : a measure of the effectiveness of this algorithm.

E[X] = ?

Let 1 stands for the smallest value to be sorted Let 2 stands for the next smallest value to be sorted

For 1≤i < j ≤n, letI(i, j) =

½ 1 ifi&j are ever directly compared 0 else

(5)

Then, X =

n−1X

i=1

Xn

j=i+1

I(i, j)

E[X] = E

n−1X

i=1

Xn

j=i+1

I(i, j)

=

n−1X

i=1

Xn

j=i+1

E [I(i, j)]

= X X

P{i&j are ever compared}

| {z }

?

Initially,i, i+ 1, ..., j−1, jwill be in the same bracket, and they will remain in the same bracket if the number chosen for the first comparison is not betweeni&j.

If one ofi+ 1, ..., j−1 is chosen for the comparison, i

will go into a left bracket andjwill go into a right bracket. →i&jwill never be compared.

Ifiorj is chosen, then there will be a direct comparison between iorj.

Probability thatiorj is chosen among i, i+ 1, ..., j−1, j is j−i+12 . P{i&j are ever compared}

E[X] =

n−1X

i=1

Xn

j=i+1

2 j−i+ 1. Whenn is large,

Xn

j=i+1

2 j−i+ 1

Z n

i+1

2 x−i+ 1dx

= 2 log(x−i+ 1)|ni+1

= 2 log(n−i+ 1)2 log 2

2 log(n−i+ 1)

∴ E[X] 2

n−1X

i=1

log(n−i+ 1)

2 Z n−1

1

log(n−x+ 1)dx

(6)

= 2 Z n

2

log(y)dy

= 2(ylogy−y)|n2

2nlogn.

9.1 Covariance Proposition

If X & Y are independent then for any functionsh&g, E[g(X)h(Y)] = E[g(X)]E[h(Y)]

Proof.

Suppose thatX&Y are jointly continuous with f(x, y).

E[g(X)h(Y)] = Z

−∞

Z

−∞

g(x)h(y)f(x, y)dxdy

= Z Z

g(x)h(y)fX(x)fY(y)dxdy

= Z

....dx Z

....dy

= E[h(Y)]E[g(X)]

Definition: The Covarience between X & Y is

Cov (X, Y) = E[(X - E[X])(Y - E[Y])]

Cov (X, Y) = E[XY - E[X]Y - E[Y]X + E[X]E[Y]]

= E[XY] - E[X]E[Y]]

IfX&Y are independent then Cov(X, Y) = 0.

(Please: refereed for counter example to text page 328)

(7)

9.1.1 Properties of Covariance (i) Cov (X, Y) = Cov (Y, X)

(ii) Cov (X, X) = Var (X)

(iii) Cov (aX, Y) = a Cov (Y, X) (iv) Cov (Pn

i=1Xi,Pm

j=1Yj) =P

i

P

j Cov (Xi, Yj) Proof. of (iv)

Letµi =E[(Xi)], νi =E[(Yj)]

Then

E

"

X

i

Xi

#

= X

i

µi , E

X

j

Yj

 = X

j

νj

Cov

X

i

Xi, X

j

Yj

= E

 ÃX

i

Xi - X

i

µi

! 

X

j

Yj - X

j

νj

= E

X

i

(Xi - µi)X

j

(Yj -νj)

= E

X

i

X

j

(Xi - µi) (Yj -νj)

= X

i

X

j

E [(Xi - µi) (Yj - νj)].

From (ii) & (iv), Var

ÃXn

i=1

Xi

!

= Cov

Xn

i=1

Xi, Xn

j=1

Xj

= Xn

i=1

Xn

j=1

Cov (Xi,Xj)

= Xn

i=1

Var (Xi) + X X

i6=jCov (Xi,Xj)

∴ Var à n

X

i=1

Xi

!

= Xn

i=1

Var (Xi) + 2X X

i<jCov (Xi,Xj)

(8)

If Xi, ...., Xnare pairwise independent (i.e. Xi&Xj are independent fori6=j), then

Var à n

X

i=1

Xi

!

= Xn

i=1

Var (Xi) ... ()

Example .

X1, ..., Xn : i.i.d. random variables, expected value µ, varianceσ2. X = 1

n Xn

i=1

Xi : sample mean Xi−X (i= 1, ...., n) : deviations

S2 = Xn

i=1

(Xi−X)2

n−1 : sample variance Then,

Var(X) = µ1

n

2Xn

i=1

Var(Xi) by ()

= σ2 n (n- 1)S2 =

Xn

i=1

(Xi−µ+µ−X)2

= X

i

(Xi−µ)2+X

i

(X−µ)22(X−µ)X

i

(Xi−µ)

= Xn

i=1

(Xi−µ)2−n(X−µ)2. Take Expectations:

(n−1)E[S2] = Xn

i=1

E(Xi−µ)2−nE(X−µ)2

= 2−nV ar(X)

(9)

= (n−1)σ2

∴ E[S2] =σ2

Example .

The variance of a binomial random variable X with parametersn&p.

Solution.

X : # of successes innindependent trials, each with success probability p.

X=X1+....+Xn. Xi =

½ 1 ifith trial is a success (bernoulli) 0 else

Var(X) = Var(Xi) +...+ Var(Xn) Var(Xi) = E[Xi2](E[Xi])2

by replacing Xi=Xi2 Var(X) = E[Xi](E[Xi])2

=p−p2

∴ Var(X) =np(1−p).

Definition

The Correlation of two random variablesX&Y is ρ(X, Y) = Cov(X, Y)

pV ar(X)V ar(Y) (as long as Var(X) Var(Y)> 0)

(10)

-1 ≤ρ(X, Y)1

Proof.

Suppose thatX&Y have variances σx2 & σy2 Then

0 Var µX

σx + Y σy

= V ar(X)

σ2x +V ar(Y)

σ2y +2Cov(X, Y) σxσy

= 2 [1 + ρ(X, Y)]

ρ(X, Y)≥ −1 0 Var

µX σx Y

σy

= V ar(X)

σ2x +V ar(Y)

σ2y 2Cov(X, Y) σxσy

= 2 [1 - ρ(X, Y)]

ρ(X, Y)1

Remarks

Var (Z) = 0 Z is constant with probability 1 (to be proved in chapter-8) ρ(X, Y) = 1 Y = a + bX, b = σσyx >0

ρ(X, Y) =1 Y = a + bX, b = -σσy

x <0

(11)

Remarks

The correlation coefficient is a measure of the degree of linearity betweenX&Y.

-1 0 1

Y tend to decrease when X increases

X, Y are Uncorrelated

Y tend to increase when X does

High degree of linearity between X & Y

Example .

IA, IB : indicator variables for the events A & B.

Then

E[IA] = P(A) E[IB] = P(B) E[IAIB] = P(AB) So,

Cov(IA, IB) = P(AB) - P(A) P(B)

= P(B)[P(A|B) - P(A)]

Positively Correlateddepending onP(A|B)>P(A) UnCorrelateddepending onP(A|B) = P(A) Negatively Correlateddepending onP(A|B)<P(A)

(12)

Example .

X1, ..., Xn : i.i.d., variance σ2 Then,

Cov (Xi - X,X) = Cov (Xi,X)Cov (X,X)

= Cov (Xi, 1 n

X

j

Xj) - Var(X)

= 1 n

X

j

Cov(Xi, Xj)−σ2 n

= σ2 n −σ2

n = 0

Remarks

X and Xi−X are uncorrelated, but in general they are not independent.

Referensi

Dokumen terkait

The projection of this Gaussian random walk onto any fixed line is a Gaussian random walk in one dimension, and by Theorem 13.4 the probability that it has a point of increase tends

A central limit theorem for random walk in a random environment on marked Galton-Watson trees.. Gabriel

In the following we give a short description of the standard branching random walk, its intrinsic martingales and an associated multiplicative random walk.. Consider a

It was proved in (7) that SLE(2) is the scaling limit of the corresponding loop-erased random walk (LERW), and SLE(8) is the scaling limit of some uniform spanning tree (UST)

We prove a strong approximation for the spatial Kesten–Spitzer random walk in random scenery by a Wiener process.. All

After which, in-plane tensile tests and out-of-plane compressive tests are conducted to determine the three fundamental Young’s moduli for the honeycomb structures.. The core was a

Since sentiment increases an investor’s overconfidence, which can result in a stronger self-attribution bias, we expect that an agent’s sensitivity to past performance decreases in

[7M] b A random process is defined as Xt=A coswct+ θ where θ is a uniform random variable over0,2π .Verify the process is ergodic in the mean sense and auto correlation sense [7M] 8..