Chapter 9

Laws of Large Numbers and Central Limit Theorem

9.1 INTRODUCTION

Let Ö\8 × be a sequence of random variables. Suppose that W8 œ \"  \#  â  \8 ß

8 œ "ß #ß âÞ In this chapter we will study the following questions: Do there exists two
sequences of constants Ö-8 × and Ö.8 ×ß .8 Ä ∞ as 8 Ä ∞ß such that the sequence of
random variables W8.-8
8
converges in probability to !, or W8.-
8
8
converges in +Þ=Þ to !ß
W8 -8
or .8 converges to a non-degenerate r.v. Here the numbers -8 are called location
constants, and the numbers .8 are called scale constants. The convergence of W8.- 8
8
in
W8 -8
probability gives us the weak law of large numbers (WLLN). The convergence of .8
in +Þ=Þ provide us the strong law of large numbers (SLLN). The convergence of W8.- 8
8
to
a non-degenerate r.v. gives us the central limit theorem (CLT).

9.2 WEAK LAW OF LARGE NUMBERS

Theorem 1. (WLLN) Suppose that Ö\8 × is a sequence of pair-wise uncorrelated

random variables with mean IÐ\3 Ñ œ .3 and variance VarÐ\3 Ñ œ 53#  ∞ß
3 œ "ß #ß âÞ Let  53# Ä ∞ as 8 Ä ∞Þ Then
8

3œ"

\" \# â\8  .3 W8  .3

8 8
:
 53#  53#
3œ" 3œ"
8 œ 8 Ä! as 8 Ä ∞Þ
3œ" 3œ"

Proof. By Chebyshev's inequality, we have

 W8 3œ"
 .3  VarW8  .3 
T  8   % œ T W8   .3   % 53#  Ÿ
8 8
8 8

 3œ"
 53# 
3œ"

  53 
8 #
3œ" 3œ" %# #
3œ"

I   \3 .3 
#
 5#
8 8
3 "
3œ" 3œ"

%#   53# 
œ œ œ
%#   53#  %#   53# 
8 # 8 # 8

3œ" 3œ" 3œ"

Ä! as 8 Ä ∞,

which implies

1
 W8  .3
8
:
 53#
3œ"
8 Ä! as 8 Ä ∞Þ
3œ"

Corollary 1. If the \3 's are identically distributed and pair-wise uncorrelated random
variables with IÐ\3 Ñ œ . and VarÐ\3 Ñ œ 5#  ∞, then -8 œ  .3 œ 8. and .8 œ
8

 53#
3œ"
8
#
œ 85 and we have
3œ"

W8  .3
8

\" \# â\8 8. :

 53#
3œ"
8 œ 85# Ä! as 8 Ä ∞ß
3œ"

that is,

\" \# â\8 :

\œ 8 Ä. as 8 Ä ∞.

In the above Theorem 1 and in the Corollary 1 the pair-wise uncorrelated r.v.'s part can
be replace by a stronger condition independent r.v.'s, because independent r.v.'s implies
pair-wise uncorrelated r.v.'s but not otherway.

Example 1. Let \" ß \# ß âß \8 be i.i.d. Bernoulli random variables with parameter :Þ

:
Show that W88 œ \ Ä : as 8 Ä ∞Þ

Solution. Note that IÐ\" Ñ œ :ß VarÐ\" Ñ œ :Ð"  :Ñ and both are finite. Thus using
WLLN we have

W8 :
8 œ \ Ä : as 8 Ä ∞Þ 

Suppose that Ö\8 × is a sequence of i.i.d. random variables. In fact, in this case, we do
not need the finiteness of the variance in order to have WLLN. In the following theorem
we state result:

Theorem 2. Suppose that \" ß \# ß âß \8 are i.i.d. random variables with common mean
IÐ\" Ñ œ .  ∞. Then

\" \# â\8 W8 :

8 œ 8 Ä. as 8 Ä ∞.

This theorem is due to Khintchine. It requires only Il\" l  ∞à but no condition is

imposed on the variance.

2
Example 2. Let \" ß \# ß âß \8 be i.i.d. random variables with Il\" l5  ∞ for some
positive integer 5Þ Then

\"5 \#5 â\85 :

8 Ä IÐ\"5 Ñ as 8 Ä ∞Þ

Solution. Since \" ß \# ß âß \8 are i.i.d. random variables, so are \"5 ß \#5 ß âß \85 Þ Also
it is given that Il\" l5  ∞. Thus, by WLLN, we have

\"5 \#5 â\85 :

8 Ä IÐ\"5 Ñ as 8 Ä ∞Þ 

Example 3. Let \" ß \# ß âß \8 be i.i.d. random variables with Il\" l#  ∞Þ Show

that

 \3#
 Ð\3  \ Ñ# œ  \  Ä VarÐ\" Ñ œ 5# Þ
8
8 # :
"
=#8 œ 8
3œ"
8
3œ"

Solution. Since \" ß \# ß âß \8 are i.i.d. random variables, so are \"# ß \## ß âß \8# Þ Since
Il\" l#  ∞ß thus by WLLN we have

 \3#
8
: :
\ Ä IÐ\" Ñ œ . and 3œ"
8 Ä IÐ\"# Ñ as 8 Ä ∞Þ

Also using the properties of continuous function, we have

\ # Ä .# Þ
:

Now it follows that

 \3#
 \  Ä IÐ\"# Ñ  .# œ 5# Þ
8

# :
3œ"
8 

Example 4. Let \" ß \# ß âß \8 be i.i.d. random variables with common density

function
"α
0\ ÐBÑ œ B#α ß B "ß α  !Þ

W8 \" \# â\8 : "α

Show that 8 œ 8 Ä α as 8 Ä ∞Þ

Solution. Note that \" ß \# ß âß \8 are i.i.d. random variables with common mean

IÐ\Ñ œ Ð"  αÑ" B B#α

∞ " "α
.B œ α  ∞Þ

Now, using WLLN, we have

3
 \" \# â\8 : "α
8 Ä α as 8 Ä ∞Þ

Example 5. Let \" ß \# ß âß \8 be i.i.d. Cauchy random variables with common density
function
"
0\" ÐBÑ œ 1Ð"B# Ñ ß  ∞  B  ∞Þ

In this case WLLN does not hold. Mean does not exists in this case. It is known that W88
follows for every 8 the same Cauchy distribution, so W88 does not converge in probability
to !Þ 

9.3 STRONG LAW OF LARGE NUMBERS

First we will discuss some important tools that are commonly used in almost sure
convergence. We will begin with the proof of Borel-Cantelli lemma.
Theorem 1 (Borel-Cantelli Lemma). Let ÖE8 × be sequence of events. If  T ÐE8 Ñ  ∞ß
∞

8œ"
then T Ðlim sup E8 Ñ œ !Þ
8
∞ ∞
Proof. Recall that lim sup E8 œ ∩ ∪ E5 Þ Now by Bool's inequality, we have
8 8 œ "5 œ 8

T Ðlim sup E8 Ñ œ T  ∩ ∪ E5  Ÿ T  ∪ E5 
∞ ∞ ∞
8 8 œ "5 œ 8 5œ8

Ÿ  T ÐE5 Ñ Ä ! as 8 Ä ∞, by hypothesis,
∞

since  T ÐE8 Ñ  ∞ means tail of the series converges to zero. Thus,

5œ8
∞

8œ"
T Ðlim sup E8 Ñ œ !Þ 
8

In view of the definition of the limit supremum of a sequence of events, it is customary to

write

Ölim sup E8 × œ ÖE8 i.o.×,

8

where i.o. is an abbreviation for "infinitely often". That is, ÖE8 i.o.× means infinitely
many of E8 occur. Therefore, ÖE8 i.o.×- will mean that only finite number of E8 occur.

Theorem 2 (Borel Zero-One Law). If ÖE8 × is a sequence of independent events, then

if and only if  T ÐE8 Ñ  ∞ß and

∞
(i) T Ðlim sup E8 Ñ œ !
8 8œ"

4
 if and only if  T ÐE8 Ñ œ ∞Þ
∞
(ii) T Ðlim sup E8 Ñ œ 1
8
 T ÐE8 Ñ œ ∞ implies
8œ"
∞
Proof. Because of the Theorem 1 we only need to prove that

 T ÐE8 Ñ œ ∞ . Now,
8œ"
∞
that T Ðlim sup E8 Ñ œ 1. So assume that
8 8œ"

∞ ∞ ∞
T ÐÖE8 i.o.×- Ñ œ T Ð ∪ ∩ E5- Ñ œ lim T Ð ∩ E5- Ñ
8 œ "5 œ 8 8Ä∞ 5œ8

∩ E5- Ñ œ lim lim  T ÐE5- Ñ

7 7
œ lim lim T Ð
8Ä∞ 7Ä∞ 5œ8 8Ä∞ 7Ä∞ 5œ8

 T ÐE5 Ñ
œ lim lim  Ð"  T ÐE5 ÑÑ Ÿ lim lim /
7
7
5œ8
8Ä∞ 7Ä∞ 5œ8 8Ä∞ 7Ä∞

since "  B Ÿ /B Þ Again,  T ÐE8 Ñ œ ∞ implies

∞

8œ"

 T ÐE5 Ñ
7

-
T ÐÖE8 i.o.× Ñ Ÿ lim lim / 5œ8 Ä !ß
8Ä∞ 7Ä∞

This implies T ÐÖE8 i.o.×- Ñ œ !ß and hence T ÐÖE8 i.o.×Ñ œ "Þ 

Example 1. Let \" ß \# ß â be a sequence of independent random variables. Then

\8 Ä ! Í  T Ðl\8 l  %Ñ  ∞, for all %  !Þ

∞
+Þ=Þ
8œ"

Proof. Let F8 œ Öl\8 l  %×Þ Clearly, ÖF8 × is a sequence of independent events. If

+Þ=Þ ∞
\8 Ä !, then lim T Ð ∪ Öl\5 l  %×Ñ œ 0 (see Chapter 8, Section 8.4). That
8Ä∞ 5œ8
is,
∞ ∞ ∞
T Ð lim ∪ Öl\5 l  %×Ñ œ 0 or T Ð ∩ ∪ Öl\5 l  %×Ñ œ 0.
8Ä∞ 5 œ 8 8 œ "5 œ 8

In turn, we have T Ðlim sup F8 Ñ œ !Þ Now by Borel-Cantelli lemma we must have

8

 T ÐF8 Ñ  ∞, i.e.  T Ðl\8 l  %Ñ  ∞, for all %  !Þ

∞ ∞

8œ" 8œ"

5
Now if  T Ðl\8 l  %Ñ  ∞, then T Ðlim sup F8 Ñ œ ! by Borel-Cantelli lemmaÞ This
∞

8œ" 8
means
∞ ∞
TÐ ∩ ∪ Öl\5 l  %×Ñ œ 0,
8 œ "5 œ 8
∞ +Þ=Þ
which implies lim T Ð ∪ Öl\5 l  %×Ñ œ 0. This implies \8 Ä !. 
8Ä∞ 5œ8

Remark. Thus, if E8 are independent events, T ÐE8 i.o.Ñ œ ! if and only if

 T ÐE8 Ñ  ∞. Similarly, if Ö\8 × is a sequence of independent random variables,
∞

8œ"
+Þ=Þ
\8 Ä ! if and only if T ÐÖl\8 l  %× i.o.Ñ œ ! a %  !Þ

Truncation
The random variable \ is said to be truncated at - if \ is replaced by \ - Ð!  -  ∞Ñß
where

\- œ 
\ if l\l  -
! if l\l -Þ

Thus, we are replacing all values of \ by zero when ever l\l - . By truncation we are
replacing an arbitrary random variable \ by an almost surely bounded random variable
\ - Þ Evidently all moments of \ - exists and finite. In fact

T Ð\ Á \ - Ñ œ T Ðl\l -Ñß IÐ \ - Ñ œ lBl- B.J\ Ÿ -ß and

IÐ \ - Ñ# œ lBl- B# .J\ Ÿ - # Þ

By choosing - large, T Ð\ Á \ - Ñ can be made arbitrarily small.

Equivalent sequences

Definition 1. Two sequences Ö\8 × and Ö]8 × of random variables are said to be tail-
equivalent if

T Ð\8 Á ]8 i.o.Ñ œ !Þ

In other words, they differ by a finite number of terms. By Borel-Cantilli lemma, if

 T Ð\8 Á ]8 Ñ  ∞ß then the sequences Ö\8 × and Ö]8 × are tail equivalent. We say
∞

8œ"
that the sequences Ö\8 × and Ö]8 × are convergence equivalent if they converge on the
same set except for a null set.

6
Theorem 3. Suppose that Ö\8 × and Ö]8 × are two sequences of random variables such
that  T Ð\8 Á ]8 Ñ  ∞. Then the following statements hold:
∞

8œ"

(i) The sequences Ö\8 × and Ö]8 × are tail equivalent.

(ii) The series  \5 and  ]5 are convergence equivalent.

8 8

5œ" 5œ"

 \5  ]5
8 8

(iii) The sequences  5œ"-8  and  5œ"-8 ß where -8 Å ∞ as 8 Ä ∞ß

converge on the same set and to the same limit, except possibly for a null
set.

Proof. (i) Clearly,

T Ð\8 Á ]8 i.o.Ñ œ lim T  ∪ Ö\5 Á ]5 ×

∞
8Ä∞ 5œ8

Ÿ lim  T  Ö\5 Á ]5 × œ !Þ
∞

8Ä∞ 5œ8

Hence, (i) follows.

(ii) Let W8 œ  \5 and W8w œ  ]5 Þ We noted in (i) that Ö\8 × and Ö]8 × are
8 8

5œ" 5œ"
tail equivalent. Therefore, for a certain =, and 8  8Ð=Ñß

W8 Ð=Ñ  W8w Ð=Ñ œ constant.

If lim W8 exists for a given =, so is lim W8w Þ Thus the set of convergence of W8 and of W8w
are almost surely the same (even though the limits may be different).

(iii) See that

W8 W8w constant
-8 œ -8 Ä !ß as 8 Ä ∞Þ

W8 W8w
This implies that -8 and -8 converges almost surely to the same limit. 

Lemma 1 (Kolmogorov's Inequality-I). Suppose that \" ß \# ß âß \8 are independent

random variables with common mean ! and variances 55# ß 5 œ "ß #ß âß 8ß respectively.
Then for any %  !

7
  55#
8

T max lW5 l  % Ÿ VarÐW8 Ñ

%# œ 5œ"
%# ß
"Ÿ5Ÿ8

where W8 œ \"  \#  â  \8 Þ

Proof. Let

F" œ ÖlW" l  %×ß

F# œ ÖlW" l Ÿ %ß lW# l  %×ß

ã
F8 œ ÖlW" l Ÿ %ß âß lW8" l Ÿ %ß lW8 l  %×Þ
8
Let E œ Ö max lW5 l  %×Þ Clearly, F5 's are disjoint events, and E œ ∪ F5 Þ Let
"Ÿ5Ÿ8 5œ"
MF5 be the indicator function of the event F5 Þ Then

IÐW8 MF5 Ñ# œ IÖÐW8  W5 ÑMF5  W5 MF5 ×#

œ IÖÐW8  W5 Ñ# MF5  W5# MF5  #ÐW8  W5 ÑW5 MF5 ×

since MF# 5 œ MF5 Þ Again note that W8  W5 œ \5"  \5#  â  \8 and W5 MF5 are
independent, and IÐ\5 Ñ œ ! for all 5Þ Thusß it follows that

IÖÐW8  W5 ÑW5 MF5 × œ IÐW5 MF5 ÑIÐW8  W5 Ñ œ !Þ

Therefore, we have

IÐW8 MF5 Ñ# œ IÖÐW8  W5 Ñ# MF5 ×  IÖW5# MF5 ×

IÖW5# MF5 × %# T ÐF5 Ñß

the last inequality holds from the fact that in F5 ß lW5 l  %Þ Moreover,

 IÐW8 MF5 Ñ# œ  IÐW8# MF5 Ñ

8 8

5œ" 5œ"

œ IÐW8# ME Ñ Ÿ IÐW8# Ñ œ VarÐW8 Ñ œ  55# ß

8

5œ"
so that
VarÐW8 Ñ œ  55# %#  T ÐF5 Ñ œ %# T Ð ∪ F5 Ñ œ %# T ÐEÑÞ
8 8 8

5œ" 3œ" 5œ"

Thus,
 55#
8

VarÐW8 Ñ
T ÐEÑ Ÿ %# œ 5œ"
%# Þ 

8
Lemma 2 (Kolmogorov's Inequality-II). Suppose that \" ß \# ß âß \8 are independent
random variables with l\5 l Ÿ - (constant) for " Ÿ 5 Ÿ 8. Then for any %  !

T max lW5 l  % " Ð%#-Ñ#

VarÐW8 Ñ ß
"Ÿ5Ÿ8

where W8 œ \"  \#  â  \8 Þ

Theorem 4. Let \" ß \# ß â be independent random variables with zero and variance 58# ß
8 œ "ß #ß âÞ If  58#  ∞ß then  \8 converges almost surely. If \3 's are a.s.
∞ ∞

bounded, converse is also true, that is  \8 converges almost surely implies 

8œ" 8œ"
∞ ∞

8œ" 8œ"
58#  ∞.

Proof. Since WR 8  W8 œ \8"  \8#  â  \R 8 it follows from Kolmogorov's

inequality that for any %  !

T  max lWR 8  W8 l  % Ÿ  VarÐ\3 Ñ  53# Þ

85 85
" "
%# œ %#
"ŸR Ÿ5
3œ8" 3œ8"

Now letting 5 Ä ∞ this becomes

T max lWR 8  W8 l  % œ T max

87
lW7  W8 l  %
"ŸR

 53# Þ
∞
"
Ÿ %#
3œ8"

 53# œ !ß since  53#  ∞ß which implies

∞ ∞
But lim
8Ä∞ 3œ8" 3œ"

lim T max
87
lW7  W8 l  % œ !Þ
8Ä∞

Since %  ! is arbitrary this in turn implies

T  lim l  \3 l Ÿ % œ "
∞

8Ä∞ 3œ8

and consequently  \3 converges almost surely.

∞

3œ"

Conversely, let W8 œ  \5 Þ Using Lemma 2, the Kolmogorov's Inequality-II, we have

8

5œ"

9
 Ð%#-Ñ#
Ÿ T max lW75  W7 l  %Þ
 5#
" 78
"Ÿ5Ÿ8
5
5œ7"

Now the almost sure convergence of W8 implies that

T max lW75  W7 l  % Ä !ß
"Ÿ5Ÿ8

as 8ß 7 Ä ∞Þ Hence, this implies that

lim   55#  œ lim  55# ß

78 ∞
∞  Ð%  #-Ñ#
7ß8 5œ7" 7 5œ7"

establishing the converse part.



Example 2. Let Ö\8 × be a sequence of independent random variables such that

T Ð\8 œ 8" Ñ œ "

# œ T Ð\8 œ  8" Ñß 8 œ "ß #ß â

Show that D\8 converges almost surely.

Solution. Note that IÐ\8 Ñ œ "# 8"  #" Ð  8" Ñ œ ! and VarÐ\8 Ñ œ #" Ð 8" Ñ  #" Ð  8" Ñ# œ "
8# Þ
Clearly, VarÐ\8 Ñ œ  8"#  ∞ß hence D\8 converges almost surely. 
8 8

Theorem 5. (Kolmogorov Three Series Theorem). Let \8- be \8 truncated at -  !Þ

The series D\8 of independent random variables converges almost surely to a random
variable if and only if for a fixed -  !ß the three series

(i) DT Ðl\8 l -Ñß (ii) DVarÐ\8- Ñß (iii) DIÐ\8- Ñ

converge.

Proof. Suppose that the three series (i)-(iii) converge. If (a) converges, then D\8 and

D\8- are convergence equivalent, that is D\8 converges almost surely if and only if
D\8- converges almost surely. Convergence of (ii) implies DÐ\8-  I\8- Ñ converges
almost surely (by Theorem 4). Since (iii) converges, DÐ\8-  I\8- Ñ  I\8- œ D\8-

converges almost surely, implying that D\8 converges almost surely.

10
 +Þ=Þ
Conversely, suppose that D\8 converges almost surely. Then \8 Ä !Þ This implies

that (i) holds for some -  !Þ Convergence of (i) implies that D\8 and D\8- are

convergence equivalent and hence D\8- converges almost surely. Since \8- is uniformly

bounded, (by Theorem 4) this implies the convergence of DVarÐ\8- Ñ. Hence

DÐ\8-  I\8- Ñ converges almost surely. Since D\8- converges almost surely, thus
DIÐ\8- Ñ converges almost surely. 

Lemma 3.

T Ðl\l 8Ñ Ÿ Il\l Ÿ "  T Ðl\l

∞ ∞
8ÑÞ
8œ" 8œ"

Proof. We will proof the lemma in the continuous case. The will be similar in the discrete
case. Let \ be a continuous random variable with density 0\ ÐBÑÞ Thus,

Il\l œ ∞ lBl0\ ÐBÑ.B œ  

∞
∞
lBl0\ ÐBÑ.BÞ
5œ! 5 Ÿ lBl  5  "

Now it follows that

5  0\ ÐBÑ.B Ÿ Il\l Ÿ Ð5  "Ñ 

∞ ∞
0\ ÐBÑ.B,
5œ! 5 Ÿ lBl  5  " 5œ! 5 Ÿ lBl  5  "

hence

5T Ð5 Ÿ l\l  5  "Ñ Ÿ Il\l Ÿ Ð5  "ÑT Ð5 Ÿ l\l  5  "ÑÞ

∞ ∞

5œ! 5œ!

Now

5T Ð5 Ÿ l\l  5  "Ñ œ   T Ð5 Ÿ l\l  5  "Ñ

∞ ∞ ∞

5œ! 8œ! 5œ8

œ T Ðl\l
∞
8Ñ
8œ"

and

Ð5  "ÑT Ð5 Ÿ l\l  5  "Ñ œ 5T Ð5 Ÿ l\l  5  "Ñ  "

∞ ∞

5œ! 5œ!

11
 œ T Ðl\l
∞
8Ñ  "Þ
8œ"

This completes the proof. 

Example 3. Let \" ß \# ß âß \8 be independent and identically distributed random

variables. If ] is a random variable and

\" \# â\8 W8 +Þ=Þ

8 œ 8 Ä ]Þ

Show that Il\" l  ∞Þ

Solution. Note that

\8 W8 8" W8" +Þ=Þ

8 œ 8  8 8" Ä ]  ] œ !Þ

This implies that

T Ðl \88 l " i.o.Ñ œ !ß

that is

T Ðl \ 8 l 8 i.o.Ñ œ !Þ

Now by Borel-cantelli lemma, we have

T Ðl\8 l
∞
8Ñ  ∞Þ
8œ"

Therefore, the previous lemma, the Lemma 3, implies that Il\" l  ∞Þ 

Lemma 4 (Toeplitz Lemma). Let Ö,8 × be a sequence of real numbers such that ,8 Ä ,
as 8 Ä ∞Þ Then ," ,# â,
8
8
Ä , as 8 Ä ∞Þ

Lemma 5 (Kronecker Lemma). Let Ö,8 × be a sequence of real numbers such that
,8 converges. Then  5,5 Î8 Ä ! as 8 Ä ∞Þ
∞ 8

8œ" 5œ"

Theorem 6. Let Ö\8 × be sequence of independent random variables with IÐ\8 Ñ œ !

and VarÐ\8 Ñ œ IÐ\8# Ñ œ 58#  ∞ß 8 œ "ß #ß âÞ Suppose that  Var5Ð\ œ  5# 5
∞ ∞ #
5Ñ IÐ\ Ñ
#

 ∞ and W8 œ  \5 Þ Then
5œ" 5œ"
8
W8 +Þ=Þ
8 Ä !Þ
5œ"

12
  IÐ\# 5 Ñ œ  I Ð \5 Ñ#  ∞ implies that  \5 converges
∞ # ∞ ∞
Proof. By Theorem 4, 5 5 5
5œ" 5œ" 5œ"
almost surely. Now using Kronecker Lemma, we have

5
 \5
8
8
W8 " \5 +Þ=Þ
8 œ 5œ"
8 œ 8 5 Ä !Þ 
5œ"

Theorem 7. (Kolmogorov strong law of large numbers): Let \" ß \# ß â be

independent and identically distributed random variables with IÐ\" Ñ œ .. Let
W8 œ  \5 Þ If Il\" l  ∞ß then
8
W8 +Þ=Þ
8 Ä .Þ
5œ"

T Ðl\" l
∞
Proof. Since Il\" l  ∞, thus 8Ñ  ∞. Now define a sequence of
8œ"
truncated random variables \5w as follows:

\5w œ  5
\ for l\5 l Ÿ 5
! otherwise.

Let W8w œ  \5w Þ Now note that

8

5œ"

 T Ð\5 Á \5w Ñ œ  T Ðl\5 l 5Ñ œ T Ðl\" l

∞ ∞ ∞
8Ñ  ∞ß
5 œ" 5 œ" 8œ"

since \3 's are identical and Il\" l  ∞. Thus Ö\5 × and Ö\5w × are tail equivalent
w
sequence. By Theorem 3 it follows that W88 and W88 converge on the same set and to the
w +Þ=Þ
same limit, except for a null set. Therefore it is sufficient to show that W88 Ä .Þ To show
this by Theorem 6, it is enough to show that that  Var8Ð\
∞ w
8Ñ
 ∞Þ Now

l\lŸ8 B# .J ÐBÑ
#

 Ÿ œ 
8œ"
∞ ∞ ∞
VarÐ\8w Ñ " w # "
8# 8# IÐ\8 Ñ 8#
8œ" 8œ" 8œ"

ÐJ is the distribution function of \" Ñ

œ   B# .J ÐBÑ œ   
∞ ∞ ∞ ∞
" "
8# 8# B# .J ÐBÑ
8œ" 5œ" 5  "  lBl Ÿ 5 5œ" 8œ5 5  "  lBl Ÿ 5

Ÿ  B# .J ÐBÑ œ #  
∞ ∞
# B#
5 5 .J ÐBÑ
5œ" 5  "  lBl Ÿ 5 5œ"5  "  lBl Ÿ 5

13
 ( since    Ð 8"
∞ ∞
" " "
8# Ÿ 5#  8" Ñ œ "
5#  "
5 Ÿ #
5 Ñ
8œ5 8œ5"

Ÿ # 
∞
lBl.J ÐBÑ œ #Il\" l  ∞Þ
5œ"5  "  lBl Ÿ 5

W8w IÐW8w Ñ +Þ=Þ

Hence, by Theorem 6 it follows that 8 Ä !Þ Since IÐ\5w Ñ Ä IÐ\" Ñ as
5 Ä ∞ß it follows by Lemma 4 that

IÐW8w Ñ IÐ\"w ÑIÐ\#w ÑâIÐ\8w Ñ +Þ=Þ

8 œ 8 Ä IÐ\" ÑÞ
W8w +Þ=Þ W8 +Þ=Þ
Thus, we may conclude that 8 Ä IÐ\" Ñ, hence 8 Ä IÐ\" Ñ œ .Þ 

9.4 CENTRAL LIMIT THEOREM

The central limit theorem (CLT) plays an important role in statistical theory. In fact it is
one of the most remarkable results in probability theory that helps explain the fact that
the empirical data of many population exhibit bell-shaped curves. Loosely speaking, it
says that the sum of a large number of independent random variables follow
approximately a normal distribution. Thus it provides a simple method of computing
approximate probabilities for sums of independent random variables. We will discuss in
this section number of different forms of the CLT. The most simplest form of the CLT is
as follows:

Theorem 1.(Lindeberg-Lévy): Let \" ß \# ß â be a sequence of independent and

identically distributed random variables with IÐ\3 Ñ œ . and VarÐ\3 Ñ œ 5# for all
3 œ "ß #ß âÞ Let

8 5 Þ
s8 œ W8 8.
W8 œ \"  \#  â  \8 and W

s 8 tends to R Ð!ß "Ñ as 8 Ä ∞Þ

Then the distribution W

s 8 converges to the
Proof. Here it is enough to show the moment generating function of W
moment generating function of R Ð!ß "ÑÞ Let Q Ð>Ñ denote the moment generating
function of the random variable \35. Þ Since first two moments of \3 's exist, thus we
may write Q Ð>Ñ œ IÐ/>Ð\3 .ÑÎ5 Ñ as

>#
Q Ð>Ñ œ "  #  9Ð># Ñ,

0 ÐBÑ
where 0 ÐBÑ œ 9Ò1ÐBÑÓ is such that lim œ !Þ
BÄ! 1ÐBÑ

s 8 can be written as
Note that W

14
 8 5 8 5 8 5
s8 œ \ " . \#  . \8 .
W  â Þ

Since \3 's are independent and identically distributed random variables, thus the moment
s 8 may be written as
generating function of W

QWs 8 Ð>Ñ œ IÐ/>W 8 Ñ œ Q Ð>Î8 Ñ œ Ò " 

s 8 >#
#8  9Ð># Î8ÑÓ8 .

As 8 Ä ∞ß 9Ð># Î8Ñ Ä ! uniformly in > so that as 8 Ä ∞

>#
QWs 8 Ð>Ñ Ä / # ß

which is the moment generating function of R Ð!ß "ÑÞ Thus the theorem is established.


8 5 5 Î 8
s8 œ W8 8. \.
Remark 1. Note that W œ Þ Therefore, \ is asymptotically normal
#
5
with mean . and variance 8 Þ

Example 1. Let \" ß \# ß â be Bernoulli trial with probability of success :Þ Show that
W8 œ \"  \#  â  \8 tends to normal with mean 8: and variance 8:Ð"  :ÑÞ

Solution. Note that IÐ\3 Ñ œ : and VarÐ\3 Ñ œ :Ð"  :Ñ Ÿ "% Þ Hence, by Lindeberg-
8:Ð":Ñ
Lévy theorem we may conclude that W s 8 œ W8 8: is asymptotically R Ð!ß "ÑÞ
Therefore, W8 is asymptotically normal with mean 8: and variance 8:Ð"  :ÑÞ 

In the next theorem the condition identical distribution part is relaxed.

Theorem 2.(Liapounov): Let \" ß \# ß â be a sequence of independent random variables

with IÐ\5 Ñ œ .5 ß VarÐ\5 Ñ œ 55# and Il\5  .5 l#$  ∞Þ Let

W8 œ \"  \#  â  \8 ß .8 œ ."  .#  â  .8 and ,8# œ 5"#  5##  â  58# Þ

s8 œ W8 .8
Then W ,8 is asymptotically R Ð!ß "Ñß provided

 Il\5  .5 l#$ œ !
8
"
lim ,8#$
8Ä∞ 5œ"

for $  !Þ

Proof. With out loss of generality, assume that IÐ\5 Ñ œ ! for all 5Þ Let

s
<5 Ð>Ñ œ IÐ/>\5 Ñ and 98 Ð>Ñ œ IÐ/>W 8 ÑÞ

15
  \5
Ñ œ IÐ /
8
> 8 >
s8
Ñ œ I Ð/ Ñ
>W ,8 \5
Then 98 Ð>Ñ œ IÐ/ 5œ" ,8

œ  I Ð/ ,8 \5 Ñ œ  <5 Ð>Î,8 ÑÞ
5œ"
8 > 8

5œ" 5œ"

Since Il\5 l#$  ∞ß

55# ># "$ l>l#$

<5 Ð>Ñ œ IÐ/>\5 Ñ œ "  ,8# #x  )5 Ð"#$ ÑÐ#$ Ñ ,8#$
Il\5 l#$ ß for l)5 l  "Þ

The second and third terms go to zero since

#$
Max Ð 5,85 Ñ#$ Ÿ Max
5Ÿ8 5Ÿ8
Ð Il\, l Ñß by Liaponov's inequality
5
#
8

 Il\5 l#$ Ä !ß by assumptionÞ

8
"
Ÿ ,8#$
5œ"

Thus <5 Ð>Ñ Ä 1, as 8 Ä ∞Þ That is Ð<5 Ð>Ñ  1Ñ Ä !ß as 8 Ä ∞Þ Therefore, for

sufficiently large 8ß since lnÐ"  DÑ œ D  9ÐDÑ œ DÐ"  9Ð"ÑÑß for lDl  "ß ( here
9Ð"Ñ Ä !ß D Ä !Ñß and  55# œ ,8# ß
8

5œ"

ln 98 Ð>Ñ œ  ln <5 Ð>Î,8 Ñ œ  ln Ò"  Ð<5 Ð>Î,8 Ñ  "ÑÓ

8 8

5œ" 5œ"

œ  Ò ,#5 >#x  )5 Ð"#$ ÑÐ#$ Ñ

8
5# # "$ l>l#$
,8#$
Il\5 l#$ ÓÐ"  9Ð"ÑÑ
8
5œ"

#
Ä ># Þ
>#
That is 98 Ð>Ñ Ä / # as 8 Ä ∞ß which is the moment generating function of R Ð!ß "ÑÞ
Thus the proof is completed. 

Theorem 3. (Lindeberg-Feller): Let \" ß \# ß â be a sequence of independent random

variables with IÐ\5 Ñ œ .5 ß VarÐ\5 Ñ œ 55#  ∞ and J5 be the distribution function of
\5 Þ Let

W8 œ \"  \#  â  \8 ß .8 œ ."  .#  â  .8 and ,8# œ 5"#  5##  â  58# Þ

Then

s8 œ W8 .8
(i) W ,8 is asymptotically R Ð!ß "Ñ and

16
 55
(ii) lim max œ ! holds
8Ä∞ 5 Ÿ 8 ,8

if and only if, for every %  !

 
8
"
1 8 Ð% Ñ œ ,8# ÐB  .5 Ñ# .J5 Ä !
5œ" lB  .5 l % ,8

as 8 Ä ∞Þ

Proof of Lindeberg-Feller theorem is omitted. In the following two corollaries we will

show that Lindeberg-Lévey and Liapounov both theorems imply the Lindeberg-Feller
theorem.

Corollary 1. Without loss of generality assume that IÐ\5 Ñ œ ! for all 5Þ For the
independent and identically distributed random variables with finite variance 5# ß

  B# .J5 œ 
8
" 8
1 8 Ð% Ñ œ B# .J
% 8 5
,8# 85#
5œ" lBl % ,8 lBl

Ä ! as 8 Ä ∞Þ

Thus Lindeberg-Lévey implies Lindeberg-Feller theorem. 

Corollary 2. Without loss of generality assume that IÐ\5 Ñ œ ! for all 5 . Suppose that
the conditions of Liapounov theorem are satisfied. Now

  B# .J5 Ÿ  
8 8
" " lBl#$
1 8 Ð% Ñ œ ,8# ,8# %$ ,8$
.J5
5œ" lBl % ,8 5œ" lBl % ,8

  lBl#$ .J5
8
"
œ %$ ,8#$
5œ" lBl % ,8

Il\5 l#$ Ä ! as 8 Ä ∞Þ
8
"
Ÿ %$ ,8#$
5œ"

Thus Liapounov theorem implies Lindeberg-Feller theorem. 

Example 2. Let \" ß \# ß â be independent random variables with density

 -5 Ÿ B Ÿ -5 Þ Suppose that l-5 l  - and  -5# Ä ∞ as 8 Ä ∞Þ Show
∞
"
0\5 ÐBÑ œ #-5 ß
5œ"
that the Lindeberg-Feller condition is satisfied (i.e. 18 Ð%Ñ Ä ! as 8 Ä ∞ÑÞ

17
 Since  -5# Ä ∞ thus ,8# Ä ∞Þ
∞
-5#
Solution. Note that IÐ\5 Ñ œ ! and VarÐ\5 Ñ œ $Þ
5œ"
Thus,

  B# .J5 œ   B# 0\5 ÐBÑ.B

8 8
" "
1 8 Ð% Ñ œ ,8# ,8#
5œ" lBl % ,8 5œ" lBl % ,8

   
8 8
" -# -# "
Ÿ ,8# #-5 .B œ ,8# #-5 .B
5œ" lBl % ,8 5œ" lBl % ,8

 T Ðl\5 l  Var#Ð\# 5 Ñ ß
8 8
-# -#
œ ,8# %,8 Ñ Ÿ ,8# % ,8
5œ" 5œ"

by Chebyshev's inequality,

Ä ! as 8 Ä ∞Þ 

Example 3. Let \4 be independent random variables defined as for some α  "

"
„ 4α
\4 œ 
with prob. '4#Ðα"Ñ
"
! with prob. "  $4#Ðα"Ñ
Þ

Show that the Lindeberg-Feller condition is satisfied if α  $# Þ

Solution. Here note that

" "
IÐ\5 Ñ œ 5 α Þ '5 #Ðα"Ñ
 5α Þ '5 #Ðα"Ñ
œ!

#5 #α
VarÐ\5 Ñ œ IÐ\5# Ñ œ '5 #α# œ $" 5 #

#5 α " "
IÐl\5 lÑ œ '5 #α# œ $ 5 α#

,8# œ  VarÐ\5 Ñ œ  5# œ
8 8
" 8Ð8"ÑÐ#8"Ñ
$ ") Þ
5œ" 5œ"


Now

8
"
1 8 Ð% Ñ œ ,8# B# J5
5œ"l\5 l  %,8

 5 #α 
8
"
Ÿ ,8# J5 , since l\5 l Ÿ 5 α
5œ" l\5 l  %,8

18
  5 #α T Ðl\5 l  %,8 Ñ
8
"
œ ,8#
5œ"

 5 #α IÐl\5 lÑ ß by Markov's inequality

8
"
Ÿ ,8# %,8
5œ"

"  5 α# ¸  8 Bα# .B

8 8
" 5 #α " "
œ %,8$ $ 5 α# œ $%,8$ $%,8$ !
5œ" 5œ"

Ð")Ñ$Î# 8α$
œ $%Ðα$ÑSÐ8*Î# Ñ
Ä ! iff *
#  α  $  ! i.e. iff α  $# Þ 

Example 4. Let Ö\4 ß 4 "× be a sequence of independent and identically distributed

 8  \4
8

.
4œ"
Ä
 \4#
(i.i.d.) random variables with mean ! and variance "Þ Show that 8 R Ð!ß "ÑÞ
4œ"

Solution. Here IÐ\5 Ñ œ !, VarÐ\5 Ñ œ IÐ\5# Ñ œ " and ,8# œ  VarÐ\5 Ñ œ 8Þ Since
8

5œ"
\" ß \# ß â are i.i.d.ß so are \"# ß \## ß âÞ Therefore by strong law of large numbers
"  # +Þ=Þ
implies that 8"  \5#
8 8 :
8 \5 Ä IÐ\5# Ñ œ "Þ This Ä "Þ Now by Lindeberg-Lévy's
5œ" 5œ"
 \4
8

. .
8 8
W8
CLT, we have œ 4œ"
Ä R Ð!ß "ÑÞ By Slutsky's theorem we know that if \8 Ä \
: \8 .
and ]8 Ä "ß then ]8 Ä \Þ Now using this result, we have

 \4  8  \4
 \5# œ
8 8
8
.
8
"
4œ" 4œ"
Ä 
 \4#
ƒ 8 8 R Ð!ß "Ñ as 8 Ä ∞Þ
5œ"
4œ"

Example 5. Let \5 be independent random variables such that

T Ð\5 œ "Ñ œ T Ð\5 œ  "Ñ œ "# Þ

Suppose that Y8 œ  \#55 Þ Show that Y8

8 .
Ä Yß where Y is uniform over Ð  "ß "ÑÞ
5œ"

Solution. The characteristics function of Y is given by

9Y Ð>Ñ œ IÐ/3>Y Ñ œ "# " /3>? .? œ

" /3> /3> sin >
#3> œ > ß

since /3> œ cos >  3 sin >Þ The characteristics function of \5 is given by

19
 9Y5 Ð>Ñ œ IÐ/3>Y5 Ñ œ "# /3>  "# /3> œ cos >Þ

Thus, the characteristics function of Y8 is given by

9Y8 Ð>Ñ œ IÐ/3>Y8 Ñ œ  IÐ/3>\5 Î# Ñ œ  cosÐ #>5 Ñ

8 5
8

5œ" 5œ"

Ä  cosÐ #> Ñ œ
∞
sin >
5 > as 8 Ä ∞ß
5œ"

> > > > > > > > >

since sin > œ # cos # sin # œ # cos # cos ## sin ## œ â œ # cos # cos ## âcos #8 sin #8

œ  cosÐ #>5 Ñ † œ  cosÐ #>5 Ñ. Since

8 ∞
sin > sinÐ>Î#8 Ñ sin > sin >
that is, > Ð>Î#8 Ñ and 8 Ä ∞ß we have > >
5œ" 5œ"
is characteristics function uniform distribution over Ð  "ß "Ñß thus we may conclude that

.
Y8 Ä Y where Y is uniform over Ð  "ß "ÑÞ 

Exercises

Section 9.2

1. For the following sequence of independent random variables Ö\8 × does WLLN hold?

(i) T Ð\8 œ #8 Ñ œ T Ð\8 œ  #8 Ñ œ "# Þ

(ii) T Ð\8 œ 8 Ñ œ T Ð\8 œ  8 Ñ œ "# Þ
(iii) T Ð\8 œ 8Ñ œ T Ð\8 œ  8Ñ œ #" 8 ß T Ð\8 œ !Ñ œ "  8 Þ
"

(iv) T Ð\8 œ #8 Ñ œ T Ð\8 œ  #8 Ñ œ #Þ#" 8 ß

T Ð\8 œ "Ñ œ T Ð\8 œ  "Ñ œ "# Ð"  #"8 ÑÞ
"
(v) T Ð\8 œ #8 Ñ œ T Ð\8 œ  #8 Ñ œ ##8" ß T Ð\8 œ !Ñ œ "  #"#8 Þ
(vi) T Ð\8 œ 8α Ñ œ T Ð\8 œ  8α Ñ œ #8"#α ß T Ð\8 œ !Ñ œ "  8"#α Þ

2. Let \" ß \# ß â be a sequence of independent random variables with IÐ\5 Ñ œ .5

 55# Ä ! as 8 Ä ∞Þ Show that the
8
"
and VarÐ\5 Ñ œ 55#  ∞ß 5 œ "ß #ß â and 8#
5œ"
WLLN holds.
:
3. Let \" ß \# ß â be any sequence of random variables. Suppose that max l\5 l Ä !Þ
"Ÿ5Ÿ8
W8 :
Show that 8 Ä !Þ

20
4. Let \" ß \# ß â be independent and identically distributed random variables with
common density 0\ ÐBÑ œ "ß ! Ÿ B Ÿ "Þ Let ^8 œ Ð  \5 Ñ"Î8 Þ Show that
8 :
^8 Ä -ß
5œ"
where - is a constant. Determine -Þ

5. Let \" ß \# ß â be independent and identically distributed random variables with

5\5 Þ Show that
8
#
IÐ\" Ñ œ . and VarÐ\" Ñ œ 5#  ∞Þ Suppose that ^8 œ 8Ð8"Ñ
5œ"
:
^8 Ä ..

Section 9.3

1. For the following sequence of independent random variables Ö\8 × does SLLN hold?

(i) T Ð\8 œ #8 Ñ œ T Ð\8 œ  #8 Ñ œ "# Þ

(ii) T Ð\8 œ 8 Ñ œ T Ð\8 œ  8 Ñ œ "# Þ
(iii) T Ð\8 œ 8Ñ œ T Ð\8 œ  8Ñ œ #" 8 ß T Ð\8 œ !Ñ œ "  8 Þ
"

(iv) T Ð\8 œ #8 Ñ œ T Ð\8 œ  #8 Ñ œ #Þ#" 8 ß

T Ð\8 œ "Ñ œ T Ð\8 œ  "Ñ œ "# Ð"  #"8 ÑÞ
" "
(v) T Ð\8 œ #8 Ñ œ T Ð\8 œ  #8 Ñ œ ##8" ß T Ð\8 œ !Ñ œ "  ##8 Þ

2. Let Ö\8 × be a sequence of random variables defined as follows:

(i) T Ð\8 œ 8α Ñ œ T Ð\8 œ  8α Ñ œ #8"#α ß T Ð\8 œ !Ñ œ "  "

8# α Þ
(ii) T Ð\8 œ 8α Ñ œ T Ð\8 œ  8α Ñ œ #" .

For what values of α does the SLLN hold?

3. Let \" ß \# ß â be a sequence of independent random variables such that

 VarÐ\  VarÐ\5 Ñ Ä ! as 8 Ä ∞Þ Is converse true?
∞ 8
5Ñ "
5#  ∞Þ Show that 8#
5œ" 5œ"

4. Let \" ß \# ß â be independent and identically distributed random variables with

Il\" l œ ∞Þ Show that, for every positive number -ß T Ð l\88 l  - i.o.) œ " and
T Ð lW88 l  - i.o.) œ "Þ

5. Let Ö\8 × be a sequence of independent random variables such that

T Ð\8 œ "Ñ œ T Ð\8 œ  "Ñ œ "# Þ

Show that  \88 converges.

8

21
Section 9.4

1. For the following sequence of independent random variables Ö\8 × does CLT hold?

(i) T Ð\8 œ #8 Ñ œ T Ð\8 œ  #8 Ñ œ "# Þ

(ii) T Ð\8 œ 8 Ñ œ T Ð\8 œ  8 Ñ œ "# Þ
(iii) T Ð\8 œ 8Ñ œ T Ð\8 œ  8Ñ œ #" 8 ß T Ð\8 œ !Ñ œ "  8 Þ
"

(iv) T Ð\8 œ #8 Ñ œ T Ð\8 œ  #8 Ñ œ #Þ#" 8 ß

T Ð\8 œ "Ñ œ T Ð\8 œ  "Ñ œ "# Ð"  #"8 ÑÞ
"
(v) T Ð\8 œ #8 Ñ œ T Ð\8 œ  #8 Ñ œ ##8" ß T Ð\8 œ !Ñ œ "  #"#8 Þ
8 log 8
(vi) T Ð\8 œ log8 8 Ñ œ T Ð\8 œ  log8 8 Ñ œ log
#8 ß T Ð\8 œ !Ñ œ "  8 Þ
" "
(vii) T Ð\8 œ #8" Ñ œ T Ð\8 œ  #8" Ñ œ #8$ ß T Ð\8 œ !Ñ œ "  #8# Þ

2. Let Ö\8 × be a sequence of independent random variables defined as follows:

" "
T Ð\8 œ 8α Ñ œ T Ð\8 œ  8α Ñ œ #8#α ß T Ð\8 œ !Ñ œ "  8# α Þ

Show that Liapounov's condition holds for α  "# ß but not for α "
#.

3. Let \ be a chi-square random variable with 8 d.f. Using moment generating function
.
#8 Ä R Ð!ß "Ñ as 8 Ä ∞Þ
show that \8
.
-
\-
4. Let \ be a Poisson random variable with parameter -. Show that Ä R Ð!ß "Ñ as
- Ä ∞Þ
I Ð8"Ñ <"
5. Let \ be a beta random variable with density 0\ ÐBÑ œ I Ð<ÑI Ð8<"Ñ B Ð"  BÑ8< ß
.
!  B  "Þ Show that ] œ 8\ Ä ^ß where ^ is a gamma random variable with
density 0^ ÐDÑ œ I "Ð<Ñ /D D <" ß D  !Þ

6. Let \" ß \# ß â be a sequence of independent and identically random variables with

 \3 and W # œ " 
8 8
"
mean . and finite variance 5# . Let \ œ 8 8" Ð\3  \Ñ# Þ Show that
3œ" 3œ"
.
WÎ8
\.
Ä R Ð!ß "Ñ as 8 Ä ∞Þ

7. Let \" ß \# ß â be a sequence of independent and identically random variables with

mean . and finite variance 5# . Let ^8 œ W
8 8.
85
and Q8 Ð>Ñ denote the mgf of ^8 Þ Show
#
that Q8 Ð>Ñ Ä /> Î# as 8 Ä ∞Þ

8. Let \" ß \# ß â be a sequence of independent R Ð!ß 55# Ñß 5 œ "ß #ß â random variables

and  55#  ∞Þ Show that W
∞
.
s8 Ä R (0,1) but the Lindeberg-Feller condition is not
5œ"
satisfied (i.e. 18 Ð%Ñ Ä
Î ! as 8 Ä !ÑÞ

22
9. Show that semi-inter-quartile range "# Ð^!Þ(&  ^!Þ#& Ñ of a random sample of size 8 has
limiting normal distribution with mean "# Ð'!Þ(&  '!Þ#& Ñ and variance

"
Ò $
'%8 0 # Ð'!Þ(& Ñ  #
0 Ð'!Þ#& Ñ0 Ð'!Þ(& Ñ  $
0 # Ð'!Þ#& Ñ ÓÞ
10. Prove that a necessary and sufficient condition for the Lindeberg-Feller theorem to
hold is

lim max #
"  ÐB  .5 Ñ# .J5 œ !Þ
8Ä∞ 5 Ÿ 8 55 lB  .
5l %,8

11. Show that the Lindeberg-Feller condition implies ,8#  ∞Þ Is converse true? Provide a
counter example.

12. Let Ö\8 × be a sequence of uniformly bounded, independent random variables, i.e.
there exists a constant - such that l\8 l Ÿ - for all 8Þ Suppose that ,8# Ä ∞ as 8 Ä ∞Þ
Show that Liapounov-Feller condition is satisfied.
"
13. Let \" ß \# ß â be independent random variables with density 0\5 ÐBÑ œ #-5 ß

 -5 Ÿ B Ÿ -5 Þ Suppose that  -5#  ∞Þ Show that the Lindeberg-Feller condition is

∞

5œ"
not satisfied (i.e. 18 Ð%Ñ Ä
Î ! as 8 Ä ∞ÑÞ

14. For each 4ß let \4 have the uniform distribution over Ò  4ß 4ÓÞ Show that the
Lindeberg-Feller condition is satisfied and state the resulting central limit theorem.

15. It is important to understand that failure of Lindeberg-Feller condition means only the
failure of either (i) or (ii) in Theorem 3 (Lindeberg-Feller) with the specified constants
,8 Þ A central limit theorem may very well hold with a different sequence of constants. Let
Ö\8 × be a sequence of independent random variables such that


 „ 8# "
with prob. "#8
\8 œ  „ 8
#


"
with prob. "#
! with prob. "  "
'  "
'8# Þ

$ W8
Show that Lindeberg-Feller condition is not satisfied. But if we take ,8# œ 8") ß then ,8
converges to R Ð!ß "ÑÞ The point is that abnormally large values may not count.

16. Let Ö\4 ß 4 "× be a sequence of independent random variables with mean ! and
 \4
8

.
variance "Þ Show that 4œ"
Ä R Ð!ß "ÑÞ
 #
4œ"\4
8

23
17. Let \5 be independent random variables such that

T Ð\5 œ "Ñ œ T Ð\5 œ !Ñ œ "# Þ

Suppose that Y8 œ  \#55 Þ Show that Y8

8 .
Ä Yß where Y is uniform over Ð!ß "ÑÞ
5œ"

18. Let \" ß \# ß â be i.i.d. random variables with common density

0\ ÐBÑ œ "# /lBl ß  ∞  B  ∞Þ


8 .
Suppose that P8 œ "
,8
\5
5 Þ Show that P8 Ä Pß where P has the following Logistic
5œ"
distribution


' Ð"/1BÎ' Ñ# ß
1 /1BÎ '
0P ÐBÑ œ  ∞  B  ∞Þ

24