Sta s cal Averages 4.

73

Closely associated to the concept of convergence in distribution is a remarkable

result known as central limit theorem, which is given below without proof.

Central Limit Theorem (Liapounoff’s Form)

If X1, X2, ..., Xn, ..., be a sequence of independent RVs with E(Xi) = mi and Var(Xi)
= si2 i = 1, 2, .., and if Sn = X1 + X2 + ... + Xn, then under certain general conditions,
n n
Sn follows a normal distribution with mean m = Â mi and variance s2 = Â s i2
as n tends to infinity. i =1 i =1

Central Limit Theorem (Lindeberg-Levy’s Form)

If X1, X2, ..., Xn, ..., be a sequence of independent identically distributed RVs with
E(Xi) = m and Var(Xi) = s2, i = 1, 2, ..., and if Sn = X1 + X2 + ... + Xn, then under
certain general conditions, Sn follows a normal distribution with mean nm and
variance ns2 as n tends to infinity.
Corollary
1 1 s2
If X = ( X1 + X 2 + + X n ) , then E ( X ) = m and Var( X ) = 2
(ns 2 ) =
n n n
Ê s ˆ
\ X follows N Á m , ˜ as n Æ •
Ë n¯

Worked Example 4(F)

Example 1
The lifetime of a certain brand of an electric bulb may be considered a RV with
mean 1200 h and standard deviation 250 h. Find the probability, using central
limit theorem, that the average lifetime of 60 bulbs exceeds 1250 h.
Let Xi represent the lifetime of the bulb.
E(Xi) = 1200 and Var (Xi) = 2502
Let X denote the mean lifetime of 60 bulbs.
Ê 250 ˆ
By corollary of Lindeberg–Levy form of CLT X follows N Á 1200, ˜
Ë 60 ¯

Ê ˆ
Á X - 1200 1250 - 1200 ˜
P( X > 1250) = P Á > ˜
Á 250 250 ˜
ÁË ˜¯
60 60
4.74 Probability, Sta s cs and Random Processes

Ê 60 ˆ
= PÁz >
Ë 5 ˜¯
= P(z > 1.55), where z is the standard normal variable
= 0.0606 (from the table of areas under normal curve)

Example 2
A distribution with unknown mean m has variance equal to 1.5. Use central limit
theorem to find how large a sample should be taken from the distribution in order
that the probability will be at least 0.95 that the sample mean will be within 0.5
of the population mean.
Let n be the size of the sample, a typical member of which is Xi.
Given: E(Xi) = m and Var(Xi) = 1.5
Let X denote the sample mean.
Ê 1.5 ˆ
By corollary under CLT, X follows N Á m , ˜
Ë n ¯
We have to find n such that
P{m – 0.5 < X < m + 0.5} ≥ 0.95
i.e., P{– 0.5 < X – m < 0.5} ≥ 0.95
i.e., P{| X – m | < 0.5} ≥ 0.95
Ï ¸
ÔÔ | X - m | 0.5 ÔÔ
i.e., PÌ < ˝ ≥ 0.95
Ô 1.5 1.5 Ô
ÔÓ n n Ô˛
i.e., P{| z | < 0.4082 n} ≥ 0.95
where z is the standard normal variable.
The least value of n is obtained from
P{| z | < 0.4082 n} = 0.95
From the table of areas under normal curve
P{|z| < 1.96} = 0.95
Therefore, least n is given by 0.4082 n = 1.96 , i.e., least n = 24.
Therefore, the size of the sample must be at least 24.
 Sta s cal Averages 4.75

Example 3
If X1, X2, ..., Xn are Poisson variates with parameter l = 2, use the central limit
theorem to estimate P(120 £ Sn £ 160), where Sn = X1, X2, ..., Xn and n = 75.
E(Xi) = l = 2 and Var(Xi) = l = 2

By CLT, Sn follows N (nm , s n )

i.e., Sn follows N (150, 150 )

Ï -30 S - 150 10 ¸
P{120 £ Sn £ 160} = P Ì £ n £ ˝
Ó 150 150 150 ˛
= P{–2.45 £ z £ 0.85}
where z is the standard normal variable
= 0.4927 + 0.2939, (from the normal tables)
= 0.7866

Example 4
Using the central limit theorem, show that, for large n,
c n n -1 - cx c 2
x e @ e - ( cx - n ) /2 n
,x>0
n -1 2p n
Let X1, X2, ..., Xn be independent RVs each of which is exponentially distributed
with parameter c.
i.e., let the pdf of Xi = c e–cx, x > 0
The characteristic function of Xi is given by
f Xi (w ) = (1 – iw/c)–1
(refer to the Note under Worked Example 7 of the characteristic function section).
By Property 4 of CFs, since X1, X2, ..., Xn are independent RVs,

f( X1 + X2 + + X n ) (w )
= [f xi (w )]n
-n
Ê iw ˆ
= Á1 - ˜
Ë c¯
= CF of Erlang distribution
(refer to Worked Example 7 of Section 4(B))
4.76 Probability, Sta s cs and Random Processes

Therefore, when n is finite, (X1 + X2 + + Xn) follows the Erlang distribution

whose pdf is given by
c n n -1 - cx
x e ,x > 0 (1)
n -1
When n tends to infinity, (X1 + X2 + + Xn) follows a normal distribution with
n n
mean nE(Xi) = and variance nVar(Xi) = 2 (Central limit theorem), i.e., when
c c
n Æ •, (X1 + X2 + + Xn) follows a normal distribution whose pdf is given by

1 ÏÔ Ê nˆ
2
2n ¸Ô
exp Ì - Á x - ˝
n ÔÓ Ë c ˜¯ c 2 Ô˛
2p
c2
c
= exp {-(cx - n)2 /2n) (2)
2np
From (1) and (2), the required result follows.

Example 5
Verify central limit theorem for the independent random variables Xk, where for
1
each k, P{Xk = ±1} = .
2
E(Xk) = 1 × 1/2 + (–1) × 1/2 = 0
Var (Xk) = 12 × 1/2 + (–1)2 × 1/2 = 1
1
Consider Yn = ( X1 + X 2 + + Xn )
n
1
E(Yn) = 0 and Var(Yn) = ¥n =1
n
Now f X k (w ) = E{eiw xk }
1
= eiw ¥ 1 / 2 + eiw ( -1) ¥
2
= cos w
\ fYn (w ) = f 1 (w )
( X1 + X 2 + + Xn )
n

= [f X k / n
(w )\n (since X1, X2, ..., Xn are independent)
n
È Ê w ˆ˘
= Ícos Á ˜˙ [since faX + b (w) = eibw fX(aw)]
Î Ë n¯˚
 Sta s cal Averages 4.77

n
È w2 1 1˘
= Í1 - + terms involving 2 and higher powers of ˙
Î 2n n n˚
n
Ê w2 ˆ 1 1
= Á1 - ˜ + term involving n and higher powers of n
Ë 2 n ¯
-1 2
w
\ lim f yn (w ) = e2
n Æ•
which is the characteristic function of N(0, 1).
\ Yn Æ N(0, 1), as n Æ •
Therefore, CLT holds good for the sequence {Xk}.

Example 6
Show that the central limit theorem holds good for the sequence {Xk}, if
1 1
P{X k = ± k a } = ¥ k -2a , P{X k = 0} = 1 - k -2a , a < .
2 2

Note Liapounoff ’s form of CLT holds good for a sequence {Xk}, if

n
 E ÈÎ X k - mk ˘
3

k =1
˚
lim 3
= 0 Condition is assumed.
n Æ•
ÏÔ n ¸2
Ô
Ì Â Var ( X k )˝
ÔÓk = 1 Ô˛

We have to verify whether this condition is satisfied by the given {Xk}.

1 1
Now mk = E(Xk) = ¥ k -a - ¥ k -a = 0
2 2
1 1
E ( X k2 ) = ¥ k 2a ¥ k -2a + k 2a ¥ k -2a = 1
2 2
\ Var(Xk) = 1
E{|Xk – mk|3 = E{|Xk|3}
1 1
= k 3a ¥ k -2a + k 3a ¥ k -2a
2 2
= ka
n
\ Â E{| X k - mk |3} = 1a + 2a + ... + na < n × na (since each term £ na)
k =1
n
 Var ( X k ) =n
k =1
4.78 Probability, Sta s cs and Random Processes

n
 E{| X k - mk |3}
k =1
\ lim 3
n Æ•
ÔÏ n Ô¸ 2
Ì Â Var ( X k )˝
ÓÔ k = 1 ˛Ô
Ê 1a + 2a + + na ˆ n ¥ na
= lim Á < lim
˜ nÆ• 3/2
n Æ• Ë n3/2 ¯ n
1
= lim 1
n Æ• -a
n 2
= 0 (since a < 1/2)
i.e., the necessary condition is satisfied. Therefore, CLT holds good for the
sequence {Xk}.

Exercise 4(F)
Part-A (Short-answer Questions)
1. What is the difference between convergence everywhere and almost
everywhere of a random sequence {Xn}?
2. Define stochastic convergence of a random sequence {Xn}.
3. State Bernoulli’s law of large numbers.
4. Define convergence of a random sequence {xn} in the mean square sense.
5. Define convergence in distribution of a random sequence {Xn}.
6. State the Liapounoff’s form of CLT.
7. State the Lindeberg-Levy’s form of CLT.
8. What is the importance of CLT?
Part-B
9. A random sample of size 100 is taken from a population whose mean is
60 and variance is 400. Using CLT, with what probability can we assert
that the mean of the sample will not differ from m = 60 by more than 4?
10. The guaranteed average life of a certain type of electric light bulb is
1000 h with a standard deviation of 125 h. It is decided to sample the
output so as to ensure that 90% of the bulbs do not fall short of the
guaranteed average by more than 2.5%. Use CLT to find the minimum
sample size.
11. If Xi, i = 1, 2, ..., 50, are independent RVs, each having a Poisson
distribution with parameter l = 0.03 and Sn = X1 + X2 + + Xn, evaluate
P(Sn ≥ 3), using CLT. Compare your answer with the exact value of the
probability.
12. If Vi, i = 1, 2, ..., 20, are independent noise voltages received in an ‘adder’
and V is the sum of the voltages received, find the probability that the
 Sta s cal Averages 4.79

total incoming voltage V exceeds 105, using CLT. Assume that each of
the random variables Vi is uniformly distributed over (0, 10).
13. 30 electronics devices D1, D1, ..., D30 are used in the following manner.
As soon as D1 fails, D2 becomes operative. When D2 fails, D3 becomes
operative and so on. If the time to failure of Di is an exponentially
distributed random variable with parameter l = 0.1/h and T is the total
time of operation of all the 30 devices, find the probability that T exceeds
350 h, using CLT.
14. Examine if the CLT holds good for the sequence {Xk}, if P{Xk = ±2k} =
2–(2k + 1), P{Xk = 0} = 1 – 2–2k.
15. Show that the CLT does not hold if the RV’s Xi have a Cauchy density.

ANSWERS
Exercise 4(A)
4. Var(X) = E{X – E(X)}2 ≥ 0
E(X2) – {E(X)}2 ≥ 0
E(X2) ≥ {E(X)}2
1
6. E(Y) = {E ( X ) - m X } = 0
sX
1
Var(Y) = 2 ¥ Var( X ) = 1
sX
SD = 1
7. mk = m¢k – kc1 m¢k – 1 D + k c2 m¢k – 2 D2 + ... + (–1)k kckDk, where D = m¢1
8. Y: 0 2 6 12
py: 0.1 0.3 0.4 0.2
E(Y) = 0.6 + 2.4 + 2.4 = 5.4
•
9. E(X) = Â j(1 - a) a j = (1 - a)(a + 2a2 + 3a3 + )
j =0

a a
= (1 - a) =
(1 - a) 2
1- a
1
Ê x3 x 4 ˆ 1
10. E(X) = 6 Ú x 2 (1 - x )dx = 6 Á - ˜=
0 Ë 3 4¯ 2
b
1 b+a
11. E(X) = Ú b-a
x dx =
2
a
b
1 1
E(X2) = Ú b-a
x 2 dx = (b2 + ba + a 2 )
3
a