Proof of Central Limit Theorem

H. Krieger, Mathematics 157, Harvey Mudd College

Spring, 2005
Preliminary Inequalities: In order to utilize the result (sometimes called
the continuity theorem) that convergence in distribution is equivalent to point-
wise convergence of the corresponding characteristic functions, we need the fol-
lowing estimates about Taylor expansions of exponential functions.
1. If u 0, then
0 e
u
1 +u u
2
/2.
2. If t is real, then
|e
it
1 it| |t|
2
/2 and |e
it
1 it (it)
2
/2| |t|
3
/6.
Central Limit Theorem: Let {X
n
} be a sequence of i.i.d. (independent
identically distributed) random variables with common mean 0 and common
variance 1. Then, if Z N(0, 1) and S
n
= X
1
+ X
2
+ + X
n
, we have
S
n
/

n Z in distribution as n . In other words, for every x R,

lim
n
P
_
X
1
+X
2
+ +X
n

n
x
_
=
1

2
_
x

e
u
2
/2
du.
Proof: Let

F be the characteristic function of the common distribution of the
{X
n
}. Then for every t R, the characteristic function of S
n
/

n is given by
E
_
e
itS
n
/

n
_
= E
_
e
it

n
k=1
X
k
/

n
_
=
_

F
_
t/

n
_
_
n
.
Consequently, our task is to prove that, for each t R,
lim
n
_

F
_
t/

n
_
_
n
= e
t
2
/2
.
Note that there is nothing to prove if t = 0.
We begin our estimation by noting that

_

F
_
t/

n
_
_
n
e
t
2
/2

_

F
_
t/

n
_
_
n

_
e
t
2
/2n
_
n

F
_
t/

n
_
e
t
2
/2n

since |

F (t/

n) | 1 and 0 e
t
2
/2n
1.
1
For the next step, we use the triangle inequality to see that
n

F
_
t/

n
_
e
t
2
/2n

F
_
t/

n
_
(1 t
2
/2n)

+n

(1 t
2
/2n) e
t
2
/2n

.
By our rst estimate, letting u = t
2
/2n 0, we see that
n

(1 t
2
/2n) e
t
2
/2n

n(t
2
/2n)
2
/2 = t
4
/8n
and this approaches 0 as n .
In the rst term we note that
n

F
_
t/

n
_
(1 t
2
/2n)

= n

E
_
e
itX/

n
(1 +itX/

n +i
2
t
2
X
2
/2n)
_

nE
_

e
itX/

n
(1 +itX/

n +i
2
t
2
X
2
/2n)

_
,
where X is a random variable with characteristic function

F, since E(X) = 0
and V ar(X) = E(X
2
) = 1. Now, on the one hand,

e
itX/

n
(1 +itX/

n +i
2
t
2
X
2
/2n)

e
itX/

n
(1 +itX/

n)

+t
2
X
2
/2n
t
2
X
2
/2n +t
2
X
2
/2n
= t
2
X
2
/n,
using the triangle inequality and the rst of the complex exponential estimates.
On the other hand,

e
itX/

n
(1 +itX/

n +i
2
t
2
X
2
/2n)

|t|
3
|X|
3
/6n
3/2
,
using the second of the complex exponential estimates.
For any > 0 and positive integer n, let A = A(, n) = {|X| >

n}. Then

e
itX/

n
(1 +itX/

n +i
2
t
2
X
2
/2n)

(t
2
X
2
/n)I
A
+ (|t|
3
|X|
3
/6n
3/2
)I
A
c .
Consequently,
nE
_

e
itX/

n
(1 +itX/

n +i
2
t
2
X
2
/2n)

_
nE
_
(t
2
X
2
/n)I
A

+nE
_
(|t|
3
|X|
3
/6n
3/2
)I
A
c
_
t
2
E(X
2
I
{|X|>

n}
) +|t|
3
/6
for every positive integer n, since E(X
2
) = 1. Therefore, given > 0, we rst
choose > 0 so that |t|
3
/6 /2 and then for this we choose the positive
integer N so that if n N we have
t
2
E(X
2
I
{|X|>

n}
) = t
2
[1 E(X
2
I
{|X|

n}
)] /2.
The last inequality is a consequence of either the monotone or dominated con-
vergence theorems.
2