NPTEL- Probability and Distributions

MODULE 6
RANDOM VECTOR AND ITS JOINT DISTRIBUTION
LECTURE 35
Topics
6.10.2 Transformation of Variables Technique
6.10.3 Moment Generating Function Technique

Example 10.2.11

(i) Let 𝑋1 and 𝑋2 be independent random variables such that 𝑋𝑖 ~ 𝐺 𝛼𝑖 , 𝜃 , 𝛼𝑖 > 0, 𝜃 >
𝑋1
0, 𝑖 = 1, 2 . Define 𝑌1 = 𝑋1 + 𝑋2 and 𝑌2 = 𝑋 . Show that 𝑌1 and 𝑌2 are
1 +𝑋2
independently distributed with

𝑌1 ~ 𝐺 𝛼1 + 𝛼2 , 𝜃 and 𝑌2 ~ Be 𝛼1 , 𝛼2 .

(ii) If 𝑋1 ~ Exp 𝜃 and 𝑋2 ~ Exp 𝜃 are independently distributed then show that
𝑋1
𝑌=𝑋 ~ 𝑈(0,1).
1 +𝑋2

Solution.

(i) The p.d.f.s of 𝑋𝑖 and 𝑋 = 𝑋1 , 𝑋2 are given by

1 𝑥
𝛼 𝑖 −1 −𝜃
𝑓𝑋 𝑖 𝑥 = 𝑥 𝑒 𝐼 0,∞ 𝑥 , 𝑖 = 1,2,
Γ(𝛼𝑖 )𝜃 𝛼 𝑖

and
2
1 𝑥 1 +𝑥 2
𝑓𝑋 𝑥1 , 𝑥2 = 𝑓𝑋 𝑖 𝑥𝑖 = 𝛼 +𝛼
𝑥1 𝛼 1 −1 𝑥2 𝛼 2 −1 𝑒 − 𝜃 𝐼 0,∞ 2 𝑥 ,
Γ(𝛼1 )Γ(𝛼2 ) 𝜃 1 2
𝑖=1

respectively.

Clearly 𝑆𝑋 = 𝑥 ∈ ℝ2 : 𝑓𝑋 𝑥1 , 𝑥2 > 0 = 0, ∞ 2 . Consider the transformation 𝑕 =

𝑕1 , 𝑕2 ∶ ℝ2 → ℝ2 defined by

Dept. of Mathematics and Statistics Indian Institute of Technology, Kanpur 1

NPTEL- Probability and Distributions

𝑥1
, if 𝑥1 + 𝑥2 ≠ 0
𝑕1 𝑥1 , 𝑥2 = 𝑥1 + 𝑥2 and 𝑕2 𝑥1 , 𝑥2 = 𝑥1 + 𝑥2 .
0, if 𝑥1 + 𝑥2 = 0

Then 𝑃 𝑌1 , 𝑌2 = 𝑕1 𝑋1 , 𝑋2 , 𝑕2 𝑋1 , 𝑋2 = 1 and therefore

𝑑
𝑌1 , 𝑌2 = 𝑕1 𝑋1 , 𝑋2 , 𝑕2 𝑋1 , 𝑋2 .

Also the transformation 𝑕 = 𝑕1 , 𝑕2 : 𝑆𝑋 → ℝ2 is one-to-one with inverse transformation

𝑕−1 = 𝑕1−1 , 𝑕2−1 , where for 𝑦1 , 𝑦2 ∈ 𝑕 𝑆𝑋 ,

𝑕1−1 𝑦1 , 𝑦2 = 𝑦1 𝑦2 and 𝑕2−1 𝑦1 , 𝑦2 = 𝑦1 1 − 𝑦2 .

The Jacobian determinant of the transformation is

𝜕𝑕1−1 𝜕𝑕1−1
𝜕𝑦1 𝜕𝑦2 𝑦2 𝑦1
J= = 1 − 𝑦2 −𝑦1 = −𝑦1 .
𝜕𝑕2−1 𝜕𝑕2−1
𝜕𝑦1 𝜕𝑦2

Also

𝑦 = 𝑦1 , 𝑦2 ∈ 𝑕 𝑆𝑋 ⇔ 𝑕1−1 𝑦 , 𝑕2−1 𝑦 ∈ 𝑆𝑋

⇔ 𝑦1 𝑦2 > 0, 𝑦1 1 − 𝑦2 > 0

⇔ 𝑦1 > 0, 0 < 𝑦2 < 1.

Therefore 𝑕 𝑆𝑋 = 0, ∞ × (0, 1) and the joint p.d.f. of 𝑌 is given by

𝑓𝑌 𝑦1 , 𝑦2 = 𝑓𝑋 𝑕1−1 𝑦 , 𝑕2−1 𝑦 𝐽 𝐼𝑕 𝑆𝑋 𝑦

= 𝑓𝑋 𝑦1 𝑦2 , 𝑦1 1 − 𝑦2 −𝑦1 𝐼 0,∞ ×(0,1) 𝑦1 , 𝑦2

𝑦1
𝛼 +𝛼 −1
𝑦1 1 2 𝑒 − 𝜃 1
= 𝐼 𝑦 𝑦2 𝛼 1 −1 1 − 𝑦2 𝛼 2 −1
𝐼 0,1 𝑦2 .
Γ(𝛼1 + 𝛼2 ) 0,∞ 1 𝐵 𝛼1 , 𝛼2

It follows that 𝑌1 and 𝑌2 are independent random variables, 𝑌1 ~𝐺 𝛼1 + 𝛼2 , 𝜃 and

𝑌2 ~ Be 𝛼1 , 𝛼2 .

(ii) Follows from (a) by taking 𝛼1 = 𝛼2 = 1. ▄

Dept. of Mathematics and Statistics Indian Institute of Technology, Kanpur 2

 NPTEL- Probability and Distributions

Example 10.2.12

(i) Let 𝑋 = 𝑋1 , 𝑋2 be a random vector of absolutely continuous type with joint p.d.f.

𝑓𝑋 𝑥1 , 𝑥2 = 𝑔 𝑥12 + 𝑥22 , 𝑥 = 𝑥1 , 𝑥2 ∈ ℝ2 ,

where 𝑔: [0, ∞) → ℝ is a non-negative function such that

∞
1
𝑥𝑔 𝑥 𝑑𝑥 = .
2π
0

Let (𝑅, 𝛩) be the polar coordinate of the point 𝑋 = 𝑋1 , 𝑋2 in the Cartesian plane, so
that, 𝑋1 = 𝑅 cos 𝛩 , 𝑋2 = 𝑅 Sin𝛩, 𝑅 > 0, 𝛩 ∈ [0,2𝜋), 𝑅 = 𝑋12 + 𝑋12 and one may take

0, if 𝑋1 = 0, 𝑋2 = 0
𝜋
, if 𝑋1 = 0, 𝑋2 > 0
2
3𝜋
, if 𝑋1 = 0, 𝑋2 < 0
2
Θ = tan−1 𝑋2 , if 𝑋1 > 0, 𝑋2 ≥ 0
𝑋1
X2
𝜋 + tan−1 , if 𝑋1 < 0
𝑋1
X2
2π + tan−1 , if 𝑋1 > 0, 𝑋2 < 0
𝑋1
𝜋 𝜋
where tan−1 𝜃 ∈ − 2 , 2 denotes the principal value. Show that 𝑅 and 𝛩 are
independently distributed with p.d.f.s

𝑓𝑅 𝑟 = 2π𝑟𝑔 𝑟 𝐼 0,∞ (𝑟)

and

1
𝑓𝛩 𝜃 = 𝐼 𝜃 ,
2𝜋 0,2𝜋
respectively.

(ii) Let 𝑋1 and 𝑋2 be independent and identically distributed 𝑁(0, 1) random variables.
𝑋
Show that the distribution of random variable 𝑌 = 𝑋2 has p.d.f.
1
1 1
𝑓𝑌 𝑦 = ∙ , −∞ < 𝑦 < ∞.
π 1 + 𝑦2
(iii) Let 𝑋 = 𝑋1 , 𝑋2 have the joint p.d.f.

Dept. of Mathematics and Statistics Indian Institute of Technology, Kanpur 3

 NPTEL- Probability and Distributions

1 2 2
𝑓𝑋 𝑥1 , 𝑥2 = 𝜋 , if 0 < 𝑥1 + 𝑥2 < 1.
0, otherwise

Find 𝐸 𝑋12 + 𝑋22 and 𝐸 𝑋1 + 𝑋2 .

Solution.

(i) Let 𝑆𝑋 = 𝑥 ∈ ℝ2 : 𝑓𝑋 𝑥 > 0 = 𝑥 ∈ ℝ2 : 𝑔 𝑥12 + 𝑥22 > 0 . Consider the

transformation 𝑕 = 𝑕1 , 𝑕2 : ℝ2 → ℝ2 , defined by 𝑕1 𝑥1 , 𝑥2 = 𝑥12 + 𝑥22 and`

0, if 𝑥1 = 0, 𝑥2 = 0
𝜋
, if 𝑥1 = 0, 𝑥2 > 0
2
3𝜋
, if 𝑥1 = 0, 𝑥2 < 0
2
𝑕2 𝑥1 , 𝑥2 = tan−1 X2 , if 𝑥1 > 0, 𝑥2 ≥ 0 .
𝑋1
X2
𝜋 + tan−1 , if 𝑥1 < 0
𝑋1
X2
2π + tan−1 , if 𝑥1 > 0, 𝑥2 < 0
𝑋1

Then 𝑅, Θ = 𝑕1 𝑥1 , 𝑥2 , 𝑕2 𝑥1 , 𝑥2 . The transformation 𝑕 = 𝑕1 , 𝑕2 : 𝑆𝑋 → ℝ2 is

one-to-one with inverse transformation 𝑕−1 𝑦1 , 𝑦2 = 𝑕1−1 𝑦1 , 𝑦2 , 𝑕2−1 𝑦1 , 𝑦2 , where
for 𝑟, 𝜃 ∈ 𝑕 𝑆𝑋 ,

𝑕1−1 𝑟, 𝜃 = 𝑟 cos 𝜃 and 𝑕2−1 𝑟, 𝜃 = 𝑟 sin𝜃.

The Jacobian determinant of the transformation is

𝜕𝑕1−1 𝜕𝑕1−1
J = 𝜕𝑟−1 𝜕𝜃 = cos θ −r sin θ
= 𝑟.
𝜕𝑕2 𝜕𝑕2−1 sin θ r cos θ
𝜕𝑟 𝜕𝜃

Also 𝑕 𝑆𝑋 = 𝑟, 𝜃 ∈ ℝ2 : 𝑟 ∈ 0, ∞ , 𝜃 ∈ 0, 2𝜋 and 𝑔 𝑟 > 0 = 𝐴1 × 𝐴2 , where

𝐴1 = 𝑟 ∈ [0, ∞): 𝑔 𝑟 > 0 and 𝐴2 = [0,2𝜋). The joint p.d.f. of (𝑅, Θ) is given by

𝑓𝑅,𝛩 𝑟, 𝜃 = 𝑓𝑋 𝑕1−1 𝑟, 𝜃 , 𝑕2−1 𝑟, 𝜃 𝐽 𝐼𝑕 𝑆𝑋 𝑟, 𝜃

= 𝑓𝑋 𝑟 cos 𝜃 , 𝑟sin𝜃 𝑟 𝐼𝐴1 ×𝐴2 𝑟, 𝜃

Dept. of Mathematics and Statistics Indian Institute of Technology, Kanpur 4

 NPTEL- Probability and Distributions

= 𝑟𝑔 𝑟 𝐼𝐴1 (𝑟) 𝐼𝐴2 (𝜃)

1
= 2𝜋𝑟𝐼𝐴1 (𝑟) 𝐼 (𝜃)
2𝜋 0,2𝜋
1
= 2𝜋𝑟𝐼 0,∞ (𝑟) 𝐼 (𝜃) .
2𝜋 0,2𝜋
It follows that 𝑅 and Θ are independent random variables with respective p.d.f.s

𝑓𝑅 𝑟 = 2𝜋𝑟𝑔 𝑟 𝐼 0,∞ (𝑟)

and

1
𝑓𝛩 𝜃 = 𝐼 𝜃 .
2𝜋 0,2𝜋
X
(ii) Note that 𝑌 = 𝑋2 is not defined if 𝑋1 = 0 . However 𝑃 𝑋1 = 0 = 0 i. e. , 𝑃 𝑋1 ≠
1
X
0 = 1 and therefore 𝑌 = 𝑋2 is well defined with probability one. In fact, since
1
𝑋 = 𝑋1 , 𝑋2 is of absolutely continuous type, we may, without loss of generality, take
𝑆𝑋 = ℝ2 − 𝑥1 , 𝑥2 ∈ ℝ2 : 𝑥1 = 0 . Define

𝜋 3𝜋
𝑌, if 𝑥1 , 𝑥2 ∈ 𝑆𝑋 tan Θ , if Θ ∈ [0, 2π) − 0, ,
𝑍= = 2 2 .
0, otherwise 0, otherwise

𝑑
Then 𝑃 𝑍 = 𝑌 = 1 and therefore 𝑌 = 𝑍. Thus we will find the distribution of random
variable 𝑍.
𝜋 3𝜋
tan Θ , if Θ ∈ [0, 2π) − 0, ,
𝑍= 2 2 .
0, otherwise

The p.d.f. of Θ is given by

1
𝑓Θ θ = 2π , if 0 ≤ θ ≤ 2π .
0, otherwise

Consider the transformation 𝑕: ℝ → ℝ defined by

𝜋 3𝜋
𝑕 𝑥 = tan 𝑥 , if 𝑥 ∈ [0, 2π) − 0, 2 , 2 .
0, otherwise

Dept. of Mathematics and Statistics Indian Institute of Technology, Kanpur 5

NPTEL- Probability and Distributions

Note that the transformation 𝑕: ℝ → ℝ is not one-to–one. Since Θ is of absolutely

continuous type we may, without loss of generality, take

𝑆Θ = 𝜃 ∈ ℝ: 𝑓Θ (𝜃) > 0

𝜋 3𝜋
= 0, 2π − 0, ,
2 2

= 𝑆1 ∪ 𝑆2 ∪ 𝑆3 , say,
𝜋 𝜋 3𝜋 3𝜋
where 𝑆1 = 0, 2 , 𝑆2 = , and 𝑆3 = , 2𝜋 . On each of the sets 𝑆1 , 𝑆2 and 𝑆3 , 𝑕
2 2 2
is strictly increasing with inverse transformations

𝑕1−1 𝑧 = tan−1 𝑧, 𝑧 ∈ 0, ∞ ,

𝑕2−1 𝑧 = 𝜋 + tan−1 𝑧, 𝑧 ∈ −∞, ∞

and

𝑕3−1 𝑧 = 2𝜋 + tan−1 𝑧, 𝑧 ∈ −∞, 0 .

Also 𝑕 𝑆1 = 0, ∞ , 𝑕 𝑆2 = −∞, ∞ and 𝑕 𝑆3 = −∞, 0 . Therefore the p.d.f. of 𝑍 is

given by
3
𝑑 −1
𝑓𝑍 𝑧 = 𝑓𝛩 𝑕𝑗−1 𝑧 𝑕 𝑧 𝐼𝑕 𝑧
𝑑𝑧 𝑗 𝑆𝑗
𝑗 =1

1 1
= 𝑓𝛩 tan−1 𝑧 𝐼 0,∞ 𝑧 + 𝑓𝛩 𝜋 + tan −1
𝑧 𝐼 𝑧
1 + 𝑧2 1 + 𝑧 2 −∞,∞
1
+𝑓𝛩 2𝜋 + tan−1 𝑧 𝐼 𝑧
1 + 𝑧 2 −∞,0
1 1 1 1 1 1
= ∙ 𝐼 0,∞ 𝑧 + ∙ 𝐼 −∞,∞ 𝑧 + . 𝐼 𝑧
2𝜋 1 + 𝑧 2 2𝜋 1 + 𝑧 2 2𝜋 1 + 𝑧 2 −∞,0
1 1
. 2
, if 𝑧 ∈ ℝ − 0
= 𝜋 1+𝑧 .
1
, if 𝑧 = 0
2𝜋
Since the random variable 𝑍 is of absolutely continuous type we may take the p.d.f. of 𝑍
as

1 1
𝑓𝑍 𝑧 = . , −∞ < 𝑧 < ∞.
𝜋 1 + 𝑧2

Dept. of Mathematics and Statistics Indian Institute of Technology, Kanpur 6

 NPTEL- Probability and Distributions

𝑋
It follows that the random variable 𝑍 and hence 𝑌 = 𝑋2 has the Cauchy distribution
1
(see Definition 11.1 (ii)).

(iii) We have

𝐸 𝑋12 + 𝑋22 = 𝐸(𝑅)

𝐸 𝑋1 + 𝑋2 = 𝐸 𝑅 cos Θ + sin Θ
= 𝐸 𝑅 𝐸(cos Θ + sin Θ) (since 𝑅 and Θ are independent).

Under the notation of (i), we have

1
𝑔 𝑥 = π , if 0 < 𝑥 < 1.
0, otherwise

Moreover

2𝑟, if 0 < 𝑟 < 1

𝑓𝑅 𝑟 =
0, otherwise

and

1
𝑓𝛩 𝜃 = 𝐼 𝜃 .
2𝜋 0,2𝜋
Therefore
1
2
𝐸 𝑋12 + 𝑋22 = 𝐸(𝑅) = 2𝑟 2 𝑑𝑟 = ,
3
0

and

𝐸 𝑋1 + 𝑋2 = 𝐸 𝑅 𝐸 cos Θ + sin Θ
2𝜋
2 cos 𝜃 + sin 𝜃
= 𝑑𝜃
3 2𝜋
0

= 0. ▄

Dept. of Mathematics and Statistics Indian Institute of Technology, Kanpur 7

 NPTEL- Probability and Distributions

6.10.3 Moment Generating Function Technique

Let 𝑋 = 𝑋1 , … , 𝑋𝑝 be a random vector with p.d.f./p.m.f. 𝑓𝑋 (∙) and let 𝑔: ℝ𝑝 → ℝ𝑞 be a

Borel function. Suppose that we seek the probability distribution of 𝑌 = 𝑔(𝑋). Under the
m.g.f. technique we try to identify the m.g.f. 𝑀𝑌 (𝑡) of random vector 𝑌 with the m.g.f. of
some known distribution. Then the uniqueness of m.g.f.s (Theorem 7.3) ascertains that
the random vector 𝑌 has that known distribution. Various usages of this technique are
illustrated in Examples 7.1, 7.2, 7.3, 7.4, 7.5 and 7.6.

Theorem 10.3.1

Let 𝑋1 , … , 𝑋𝑛 (𝑛 ≥ 2) be a random sample from 𝑁 𝜇, 𝜎 2 distribution, where 𝜇 ∈

1 𝑛 1 𝑛 2
−∞, ∞ and 𝜎 > 0. Let 𝑋 = 𝑛 𝑖=1 𝑋𝑖 and 𝑆 2 = 𝑛−1 𝑖=1 𝑋𝑖 − 𝑋 denote the sample
mean and the sample variance respectively. Then
𝜎2
(i) 𝑋 ~ 𝑁 𝜇, ;
𝑛
(ii) 𝑋 and 𝑆 2 are independent random variables;
𝑛 −1 2
(iii) 𝑆 2 ~ 𝜒𝑛−1 ;
𝜎2
𝑛
2𝜎 4 2 Γ( )
2 2 2
(iv) 𝐸 𝑆 = 𝜎 , Var 𝑆 = and 𝐸 𝑆 = 2
𝑛 −1 𝜎.
𝑛−1 𝑛−1 Γ( )
2

Solution.

(i) Follows from Example 7.1.

𝑛 𝑛
(ii) Let 𝑌𝑖 = 𝑋𝑖 − 𝑋, 𝑖 = 1, … , 𝑛 and let 𝑌 = 𝑌1 , … , 𝑌𝑛 . Then 𝑖=1 𝑌𝑖 = 𝑖=1 𝑋𝑖 − 𝑛𝑋 = 0
𝑛 2 𝑛 2
and 𝑛 − 1 𝑆 2 = 𝑖=1 𝑋𝑖 − 𝑋 = 𝑖=1 𝑌𝑖 , a function of 𝑌. The joint m.g.f. of 𝑌, 𝑋
is given by
𝑛
𝑀𝑌,𝑋 𝑢, 𝑣 = 𝐸 𝑒 𝑖=1 𝑢 𝑖 𝑌𝑖 +𝑣𝑋 , 𝑢 = 𝑢1 , … , 𝑢𝑛 ∈ ℝ𝑛 , 𝑣 ∈ ℝ.

Let us fix 𝑢 = 𝑢1 , … , 𝑢𝑛 ∈ ℝ𝑛 and 𝑣 ∈ ℝ. Then

𝑛 𝑛

𝑢𝑖 𝑌𝑖 + 𝑣 𝑋 = 𝑢𝑖 𝑋𝑖 − 𝑋 + 𝑣𝑋
𝑖=1 𝑖=1

𝑛 𝑛 𝑛
𝑣− 𝑖=1 𝑢𝑖
= 𝑢𝑗 𝑋𝑗 + 𝑋𝑗
𝑛
𝑗 =1 𝑗 =1

Dept. of Mathematics and Statistics Indian Institute of Technology, Kanpur 8

NPTEL- Probability and Distributions

𝑛
𝑣
= 𝑢𝑗 − 𝑢 + 𝑋
𝑛 𝑗
𝑗 =1

= 𝑡𝑗 𝑋𝑗 ,
𝑗 =1

1 𝑛 𝑣 𝑛
where 𝑢 = 𝑛 𝑖=1 𝑢𝑖 and 𝑡𝑗 = 𝑢𝑗 − 𝑢 + 𝑛 , 𝑗 = 1, … , 𝑛 . Note that 𝑗 =1 𝑢𝑗 − 𝑢 = 0,
and therefore,
𝑛 𝑛
𝑣
𝑡𝑗 = 𝑢𝑗 − 𝑢 + = 𝑣,
𝑛
𝑗 =1 𝑗 =1

and
𝑛 𝑛 𝑛
𝑣 2 2 𝑣2
𝑡𝑗2 = 𝑢𝑗 − 𝑢 + = 𝑢𝑗 − 𝑢 + .
𝑛 𝑛
𝑗 =1 𝑗 =1 𝑗 =1

Consequently,
𝑛
𝑀𝑌,𝑋 𝑢, 𝑣 = 𝐸 𝑒 𝑗 =1 𝑡 𝑗 𝑋 𝑗

= 𝐸 𝑒 𝑡𝑗 𝑋𝑗
𝑗 =1

= 𝑀𝑋 𝑗 𝑡𝑗
𝑗 =1

𝑛
𝜎 2𝑡𝑗 2
𝜇 𝑡𝑗 +
= 𝑒 2

𝑗 =1

𝑛 𝑛
𝜇
𝜎2
=𝑒 𝑡𝑗 + 𝑡𝑗 2
2
𝑗 =1 𝑗 =1

𝑛
𝜇𝑣 +
𝜎2 2 𝑣2
=𝑒 2 𝑢𝑗 − 𝑢 +
𝑛
𝑗 =1

2
𝜎 2 𝑣2 𝜎2 𝑛
𝑗 =1 𝑢 𝑗 −𝑢
𝜇𝑣 +
=𝑒 2𝑛 𝑒 2 , 𝑢 ∈ ℝ𝑛 , 𝑣 ∈ ℝ.

Dept. of Mathematics and Statistics Indian Institute of Technology, Kanpur 9

 NPTEL- Probability and Distributions

The joint m.g.f. of 𝑌 = 𝑌1 , … , 𝑌𝑛 is given by

𝜎2 𝑛 2
𝑢 𝑗 −𝑢
𝑀𝑌 𝑢 = 𝑀𝑌,𝑋 𝑢, 0 = 𝑒 2 𝑗 =1 , 𝑢 ∈ ℝ𝑛 ,

and the m.g.f. of 𝑋 is given by

𝜎 2 𝑣2
𝑀𝑋 𝑣 = 𝑀𝑌,𝑋 0, 𝑣 = 𝑒 𝜇𝑣 + 2𝑛 , 𝑣 ∈ ℝ.

Clearly

𝑀𝑌,𝑋 𝑢, 𝑣 = 𝑀𝑌 𝑢 𝑀𝑋 𝑣 , ∀ 𝑢, 𝑣 ∈ ℝ𝑛 +1 .

Now using Theorem 7.4 it follows that 𝑌 = 𝑋1 − 𝑋, … , 𝑋𝑛 − 𝑋 and 𝑋 are

independent. This in turn implies that, for any Borel functions Ψ1 ∙ and Ψ2 ∙ , Ψ1 𝑌
and Ψ2 𝑋 are independent. In particular, it follows that 𝑆 2 (a function of 𝑌) and 𝑋 are
independent.
𝑋 𝑖 −𝜇
(iii) Let 𝑍𝑖 = , 𝑖 = 1, … , 𝑛. Then 𝑍1 , … , 𝑍𝑛 are independent and identically distributed
𝜎
𝑁(0,1) random variables. Furthermore, by (i) and Theorem 4.1 (i)-(b), Module
2
𝑛 𝑋−𝜇 𝑛 𝑋−𝜇 𝑛−1 𝑆 2
5,𝑍 = ~𝑁 0,1 . Let 𝑊 = 𝑍 2 = and 𝑌 = . Then, by (ii), 𝑊 and
𝜎 𝜎2 𝜎2
𝑌 are independent random variables. Also, by Example 7.6 (ii), 𝑊 ~ 𝜒12 and 𝑇 =
𝑛 2 2
𝑖=1 𝑍𝑖 ~ 𝜒𝑛 . Thus the m.g.f.s of 𝑊 and 𝑇 are

−
1 1
𝑀𝑊 𝑡 = 1 − 2𝑡 2 , 𝑡< ,
2
and
𝑛
− 1
𝑀𝑇 𝑡 = 1 − 2𝑡 2 , 𝑡< .
2
Also
𝑛

𝑇= 𝑍𝑖2
𝑖=1

𝑛
2
𝑋𝑖 − 𝜇
=
𝜎2
𝑖=1

Dept. of Mathematics and Statistics Indian Institute of Technology, Kanpur 10

 NPTEL- Probability and Distributions

𝑛 2
𝑋𝑖 − 𝑋 + 𝑋 − 𝜇
=
𝜎2
𝑖=1

𝑛 2 2
𝑋𝑖 − 𝑋 𝑛 𝑋−𝜇
= +
𝜎2 𝜎2
𝑖=1

= 𝑌 + 𝑊.

Since 𝑌 and 𝑊 are independent random variables, we have

𝑀𝑇 𝑡 = 𝑀𝑌 𝑡 𝑀𝑊 (𝑡)

𝑀𝑇 𝑡
⇒ 𝑀𝑌 𝑡 =
𝑀𝑊 (𝑡)
𝑛
−
1 − 2𝑡 2
= 1
−
1 − 2𝑡 2

−
𝑛 −1 1
= 1 − 2𝑡 2 , 𝑡< ,
2
2
which is the m.g.f. of 𝜒𝑛−1 distribution. Now, by uniqueness of m.g.f.s it follows that
𝑛 −1 𝑆 2 2
𝑌= ~ 𝜒𝑛−1 .
𝜎2

𝑛 −1 𝑆 2
(iv) We have 𝑌 = ~ 𝜒𝑛2−1 . Therefore
𝜎2

𝑟
𝜎𝑟 𝑟
𝐸 𝑆 = 𝑟 𝐸 𝑌2
𝑛−1 2
∞ 𝑦 𝑛 −1
𝜎𝑟 𝑟 𝑒 −2 𝑦 2
−1
= 𝑟 𝑦 2
𝑛 −1 𝑑𝑦
𝑛−1
𝑛−1 2
0 2 2 Γ( )
2

∞
𝜎𝑟 1 𝑦 𝑛 −1+𝑟
= 𝑟 𝑛 −1 𝑒 −2 𝑦 2
−1
𝑑𝑦
𝑛−1
𝑛−1 2 2 2 Γ( )0
2

𝑛 −1+𝑟 𝑛−1+𝑟
2 2 Γ( ) 𝜎𝑟
2
= 𝑛 −1 𝑛−1 𝑟 , 𝑟 > −(𝑛 − 1)
2 2 Γ( ) 𝑛−1 2
2
𝑟 𝑛−1+𝑟
2 2 Γ( )
2
= 𝑛−1 𝜎𝑟 , 𝑟 > − 𝑛 − 1 .
𝑛−1 Γ( )
2

Dept. of Mathematics and Statistics Indian Institute of Technology, Kanpur 11

NPTEL- Probability and Distributions

Therefore
𝑛
2 Γ 2
𝐸 𝑆 = ∙ 𝑛−1 𝜎,
𝑛−1 Γ
2

𝑛−1
2 Γ( 2 + 1) 2
𝐸 𝑆2 = 𝜎 = 𝜎2,
𝑛 − 1 Γ(𝑛−1)
2

𝑛−1
2 2 Γ( + 2) 𝑛+1 4
4 2
𝐸 𝑆 = 𝑛−1 𝜎4 = 𝜎
𝑛−1 Γ( ) 𝑛−1
2

and

2 2𝜎 4
Var 𝑆 2 = 𝐸 𝑆 4 − 𝐸 𝑆 2 = .▄
𝑛−1

Dept. of Mathematics and Statistics Indian Institute of Technology, Kanpur 12