Normal Distribution

Definition 1 A random variable X with pdf

1 (x−µ)2
f (x) = √ e− 2σ2 , −∞ < x < ∞
σ 2π
is called normal distribution with parameters µ and σ, denoted N (µ, σ 2 ). ¦
A normal distribution with µ = 0 and σ = 1, usually denoted Z ; N (0, 1), is called
standard normal distribution, its pdf is:
1 z2
g(z) = √ e− 2 , −∞ < z < ∞
2π
Z ∞
1 2
- To check that g is a pdf, we need to compute the integral I = √ e−x /2 dx;
Z ∞ −∞ 2π
1 −y2 /2
we first consider J = √ e dy, then
−∞ 2π
Z ∞Z ∞
2 1 −(x2 +y2 )/2
I =I ×J = e dxdy,
−∞ −∞ 2π
Z 2π Z ∞
1 −r2 /2
= re drdθ = 1 (in polar coordinates), hence I = 1
0 0 2π
Now we can check that f is a pdf:
Z ∞ Z ∞
1 −
(x−µ)2 1 z2
√ e 2σ dx =2 √ e− 2 dz = 1 (by a changing variable z = x−µ σ
)
−∞ σ 2π −∞ 2π
Z +∞
1 z2 1 h − z2 i+∞
- E(Z) = √ z e− 2 dz = √ −e 2 =0
−∞ 2π 2π −∞
Z +∞ i+∞ Z +∞ 1
1 2 − z2 1 h − z2 z2
- V ar(Z) = √ z e 2 dz = √ −ze 2 + √ e− 2 dz = 1
−∞ 2π 2π −∞ −∞ 2π
0 −z 2 /2
(integration by parts u = z and v = ze )
Z +∞ Z +∞ Z +∞
1 tz − z2 1 − (z2 −2tz) t 2 /2 1 (z−t)2
- MZ (t) = √ e e 2 dz = √ e 2 dz = e √ e− 2 dz =
−∞ 2π −∞ 2π −∞ 2π
t2 /2
e
(the last integral is the pdf of a normal distribution with µ = t and σ = 1)
σ 2 t2
Exercise 1 Show that E(X) = µ, V ar(X) = σ 2 , and MX (t) = eµt+ 2 .
x−µ
(hint: consider the change of variable z = σ
) 5
X −µ
Proposition 1 If X ; N (µ, σ 2 ), then Z = ; N (0, 1).
σ
(equivalently, If Z ; N (0, 1), then X = σZ + µ ; N (µ, σ 2 )). 2
Proof: FZ (z) = P (Z ≤ z) = P ( X−µ
σ
≤ z) = P (X ≤ σz + µ) = FX (σz + µ), hence
2 /2
fZ (z) = FZ0 (z) = FX0 (σZ + µ) = σfX (σz + µ) = √1 e−z , −∞<z <∞
2π

1
Proposition 2 If Z1 , Z2 , . . . , Zn are n independent N (0, 1), then
X = Z12 + Z22 + . . . + Zn2 ; χ2 (n) 2

Proof: Z12 ; χ2 (1) (why?), and Z12 + Z22 + . . . + Zn2 ; χ2 (n) (why?)
Sum of independent normal distributions:

Proposition 3 Let Xi ; N (µi , σi2 ) , i = 1, 2, . . . , n be n independent normal distri-

butions. Then Y = a1 X1 + . . . + an Xn ; N (µ, σ 2 ),
where µ = a1 µ1 + . . . + an µn , and σ 2 = a21 σ12 + . . . + a2n σn2 2

Proof: MY (t) = MX1 (a1 t) × . . . × MXn (an t)

a2 2 2
1 σ1 t a2 2 2
n σn t
= ea1 µ1 t+ 2 × . . . × ean µn t+ 2

(a2 2 2 2 2
1 σ1 +...+an σn )t
= e(a1 µ1 +...+an µn )t+ 2

hence Y ; N (a1 µ1 + . . . + an µn , a21 σ12 + . . . + a2n σn2 )

Example 1 Suppose that the length of life in hours of a light bulb manufactured by
company A is N (800, 14400) and the length of life in hours of a light bulb manufactured
by company B is N (850, 2500). One bulb is selected from each company and is burned
until death. Find the probability that length life of the bulb from company A exceeds
the length of life of the bulb from company B by at least 15 hours.
P (A > B + 15) = P (A − B > 15); and A − B 7→ N (−50, 16900)
hence P (A − B > 15) = P (Z > 15+50
130
) = P (Z > 0.5) = 1 − 0.6915 = 0.3085 (from the
standard normal table) 4

The Central Limit Theorem:

Definition 2 The mean of n random variables X1 , . . . , Xn , denoted X, is defined by

Pn
Xi
X = i=1 ¦
n
Proposition 4 Let X1 , . . . , Xn be n independent random variables with E(Xi ) = µ,
and V ar(Xi ) = σ 2 , for i = 1, 2, . . . , n. Then
σ2
E(X) = µ , and V ar(X) = n
2

The proof is obvious.

Proposition 5 Let X1 , . . . , Xn be a random sample of size n with normal distribution

N (µ, σ 2 ), then

2 X −µ
X ; N (µ, σn ) (or equivalently √ ; N (0, 1)) 2
σ/ n

Proof: Combine propositions 3 and 4.

2
Proposition 6 (Central limit theorem)
Let X1 , . . . , Xn be a random sample of size n with E(Xi ) = µ and V ar(Xi ) = σ 2 , then
2
X is approximately N (µ, σn ) 2

Example 2 Let X1 , . . . , X10 be a random sample of size 10 with exponential distribu-

tion with mean 2, i.e.
1
f (x) = e−x/2 , 0 < x < ∞
2
a. Find P (12.44 < X1 + . . . + X10 < 28.41)

b. Approximate P (12.44 < X1 + . . . + X10 < 28.41)

Solution: a) X1 + . . . + X10 ; χ2 (20), and P (12.44 < X1 + . . . + X10 < 28.41) =

P (12.44 < χ2 (20) < 28.41) = P (χ2 (20) < 28.41) − P (χ2 (20) < 12.44) = 0.9 − 0.1 = 0.8
(from the Chi-square table).

X1 +...+X10 4
b) 10
∼ N (2, 10 ) by the central limit theorem.

X1 +...+X10
P (12.44 < X1 + . . . + X10 < 28.41) ' P (1.244 < 10
< 2.841)
= P (−0.765 < X1 +...+X
10
10
− 2 < 0.841)
X1 +...+X10
−2
= P (−1.21 < 10√
2/ 10
< 1.33)
= P (−1.21 < Z < 1.33) = 0.795 ' 0.8

Example 3 Fifty numbers are rounded off to the nearest integer and then summed. If
the individual round-off errors are uniformly distributed over the interval (−1/2, 1/2).
Find the probability that the resultant sum differs from the exact sum by more than
3. 4

Solution: Let X1 , X2 , . . . , X50 be the errors for the 50 numbers; Xi 7→ U (−1/2, 1/2),
E(Xi ) = 0, and V ar(Xi ) = 1/12

X1 +X2 +...+X50 1
50
∼ N (0, 600 ) by the central limit theorem.
√
P (|X1√+ X2 + . . . + X50 | ≥ 3) = P (−3 < X1 + X2 + . . . + X50 < 3) = 1 − P (− 3 50600 <
Z < 3 50600 ) = 1 − P (−1.47 < Z < 1.47) = 1 − (0.9292 − (1 − 0.9292)) = 1 − 0.86 = 0.14
(from the normal table)

3
Normal approximation to Binomial:

Let X1 , . . . , Xn be a random sample of size n with Bernouilli distribution b(p), and let
X = X1 + . . . + Xn

(we know that X has a Binomial distribution b(n, p))

By the Central limit theorem,

X1 +...+Xn
n
−p
q ∼ N (0, 1)
p(1−p)
n
X1 + . . . + Xn − np
or equivalently p ∼ N (0, 1)
np(1 − p)
or X1 + . . . + Xn ∼ N (np, np(1 − p))

Thus, we have the proposition:

Proposition 7 If X ; b(n, p), then X can be approximated by a normal distribution

N (np, np(1 − p)). 2

Example 4 In the casino game roulette, the probability of winning with a bet on red
is 18/38. Let Y equal the number of winning bets out of 1000 independent bets that
are placed. Approximate P (Y > 500).

Let Xi be the winning at the ith bet; Xi 7→ b( 18

38
), hence
18 9 18 20 90
E(Xi ) = 38
= 19
and V ar(Xi ) = .
38 38
=361

by the central limit theorem, Y 7→ N ( 9000

19
; 90000
361
)
³ ´
Y −(9000/19) 500−(9000/19)
P (Y > 500) = P √ > √ = P (Z > 1.67) = 0.0475 (from the
90000/361 90000/361
normal table) 4