Q1 Given the function:

(
kxyz 2 0 < x, y < 1, 0 < z < 3
f (x, y, z) =
0 otherwise

a) To ensure that f (x, y, z) represents a valid probability density function

over the specified domain, we must verify that its triple integral over the domain
equals 1:
Z 3Z 1Z 1
We have: f (x, y, z) dx dy dz = 1
0 0 0
Z 3 Z 1
⇒ kxyz 2 dx dy dz = 1
0 0
1 1 Z 3 2
kyz 2
Z Z
kz
⇒ dy dz = dz
0 0 2 0 4
0  3
X 1 3 9k
⇒ k = =1
0
4 3 4
4
⇒k=
9
For the expectation E(x), we need the marginal density function fx (x):

R3R1 R3
fx (x) = 0 0 94 xyz 2 dy dz = 0 94 x2 z 2 dz
= 49 x × 21 × 9 = 2x
R1 R1
E(x) = 0 xfx (x) dx = 0 2x2 dx = 23 .

b) For the marginal densities fy (y) and fz (z):

Z 3 Z 1
4
fy (y) = xyz 2 dx dz = 2y
0 0 9
fx (z) = 2x
Z 1Z 1
4 z2
fz (z) = xyz 2 dx dy =
0 0 9 9

The joint density fXY (x, y) is found by integrating f (x, y, z) over z:

Z 3
4 2
fXY (x, y) = yz dz = 4xy.
0 9

1
 From these, we can derive the conditional densities:
Z 1
4 2
fxz (x, z) = xyz 2 dy = xz 2
0 9 9
fxy (x, y) = fx (x)fy (y)
fyz (y, z) = fy (y)fz (z)
fxz (x, z) = fx (x)fz (z)
fxyz (x, y, z) = fx (x)fy (y)fz (z)
R1
Lastly, fyz (y, z) = 0 94 xyz 2 dx = 29 yz 2 , indicating independence.
c) Marginal Probability Density Function (PDF) of X: fx (x) = 2x.
Marginal PDF of z: fz (z) = 2/9.
d) Given independence of x, y, and z, the conditional PDF fx|y=1/2 (x) is
equivalent to the marginal PDF fx (x) and equals 2x.
e) Since the variables x and z are statistically independent, their covariance
cov(x, z) is zero.
(2) The variables x1 , x2 , and x3are mutually independent and each has unit
variance. Let x⊤ = x1 x2 x3 .
 
1 0 0
The covariance matrix of x is given by Cov(x) =  0 1 0 
0 0 1
   
y1 x1 + x2 + x3
Now define  y2  =  x1 − x2 =y
y3 x1 − x3
    
y1 1 1 1 x1
we have y =  y2  =  1 −1 0   x2 
y3 1 0 −1 x3
| {z }
A
the covariance matrix of y is Cov(y) = A Cov(x)A⊤
   
1 1 1 1 0 0 1 1 1
=  1 −1 0   0 1 0   1 −1 0 
1 0 −1 0 0 1 1 0 −1
 
3 0 0
= 0 2 1 
0 1 2
 
1 0 0
1
the correlation matrix of y is Corr(y) =  0 1 2

1
0 2 1
covij
Whereas the correlation coefficient corrij is given by corrij = .
σi σj

(3) Let U ∼ N3 (0, I) and µ⊤ =

 
10 4 7 .

2
 i) Consider the matrix B given by:
 
2 1 0
B= 1 1 −1 
2 −1 3

The random vector X is defined as X = µ + BU .

Since U ∼ N3 (0, I), then X ∼ N3 B.0 + µ, BΣB ⊤ , where Σ = I3 .

   
2 1 0 1 0 0 2 1 2
BΣB ⊤ =  1 1 −1   0 1 0   1 1 −1 
2 −1 3 0 0 1 0 −1 3
 
5 3 3
=  3 3 −2 
3 −2 14
   
10 5 3 3
X ∼ N3  4  ,  3 3 −2 
7 3 −2 14

To find
 the rank of the covariance
 matrix, we reduce it to row-echelon form,
5 3 3
which is  0 6/5 −19/5 . Therefore, the matrix has full rank.
0 0 1/6
 
21
ii) Consider B = . Here, we use the last two random variables of U :
1

Let U ′ ∼ N2 (0, I) for consistency of matrices.

X = µ + BU ′ X ∼ N3 B.0 + µ, BΣB ⊤ , Σ = I2

 
2 1   
⊤ 1 0 2 1 2
BΣB = 1 1 
0 1 1 1 −1
2 −1
 
5 3 3
= 3 2 1 
3 1 5
   
10 5 3 3
X ∼ N3  4  ,  3 2 1 
7 3 1 5
To check
 the rank of thecovariance matrix, we reduce it to row echelon form,
5 3 3
which is  0 1/5 −4/5 . Hence, the rank is 2.
0 0 0
(4) Using the same µ⊤ = 10 4 7 as in the previous question.
 

3
    
10 9 −3 −3
a) i) X ∼ N3  4  ,  −3 5 1  We have X = µ + BU , where
7 −3 1 5
U ∼ N3 (0, I).
We use Mahalanobis transformation on X to get U . Hence, U = Σ−1/2 (x −
µ) ⇒ x = µ + Σ1/2 U . Therefore, B = Σ1/2 .
P3 √
Now we use eigenvector analysis to get Σ1/2 . We have Σ1/2 = i=1 λi ei e⊤
i ,
where Σ has eigenvalues and eigenvectors as follows:
 √     √ 
1/√3 0 −2/√ 6
λ1 = 3, e1 =  1/√3  ; λ2 = 4, e2 =  −y√ √  ; λ = 12, e = 
2 3 3 1/√6 
1/ 3 1/ 2 1/ 6
P3 √
Σ1/2 = i=1 λ i ei e⊤
i

√1 √1 √1
    
4 − √23 − √23

3 3 3 0 0 0 √ √ 
1 1 √1
= √ √ + 0 1 −1 + − √2 1/ 3 1/ 3 
    
3 3 3 √3 √ √
 
√1 √1 √1 0 −1 1 −2/ 3 1/ 3 1/ 3
3 33 3
√5 − √1 − √13
 
3 √ 3 √
=
 − √13 2 33+3 2 3−3 
 = B.
√ √3
− √13 2 33−3 2 3+3
3


  
10 8 −4 −4
ii) X ≃ N3  4  ,  −4 4 0 
7 −4 0 4
Similarly as (i), Σ1/2 = B. Σ has eigenvalues and eigenvectors:

,
 T
1 1 1
λ1 = 0, e1 = √ , √ , √ ,
3 3 3
 T
1 1
λ2 = 4, e2 = 0, − √ , √ ,
2 2
T
√

2 1 1
λ3 = 2 3, e3 = − √ , √ , √
6 6 6

4
 3 p
X
Σ1/2 = λ i ei e⊤
i
i=1
− √26
 
0
 
1
h i √  h i
= 0 + 2  − √2  0 − √12 √1
2
+ 2 3 √1
6

 − √26 √1
6
√1
6
√1 √1
2 6
√4 − √23 − √2
 
3 √ √ 3
3+ 3 3−3
=
 − √23 √  = B.

√3 √ 3
3−3 3+3
− √23 √
3
√
3

b) Structural relationships
  among
 x i in
(i) and (ii) 
x1 10 9 −3 −3
i) The vector x = x2  ∼ N3  4  , −3 5 1 . Here, x1 exhibits
x3 7 −3 1 5
a negative correlation with both x2 and x3 . Conversely, x2 displays a positive
correlation with x 3 .     
x1 10 8 −4 −4
ii) For X = x2  ∼ N3  4  , −4 4 0 , x1 shares a negative
x3 7 −4 0 4
correlation with both x2 and x3 . Additionally, x2 and x3 are independent,
indicated by a correlation coefficient of0 inthe
 multivariate
normal distribution.
10 5 3 3
(5) i) Distance from 3a, x ∼ N3  4  , 3 3 −2.
7 3 −2 14
 
x1
- Marginal : We need to consider the sub-covariance matrix B for x1
x2  
′ 5 3
and x2 , denoted as Σ = .
  3 3
x
- Marginal 1 : We need to consider the sub-covariance matrix B for x1 and
x2    
5 3 x
x2 , denoted as Σ′ = , and µ′ = [10, 4]T . Therefore, 1 ∼ N2 (µ′ , Σ′ ),
3 3 x2
which is non-degenerate.
Now, X̄1 :
 
48x2 + 18x3 + 83 1
N1 , .
38 38
Note that, this sub-matrix is non-degenerate
b) Structural relationships
  among
 x i in
(i) and (ii) 
x1 10 9 −3 −3
i) The vector x = x2  ∼ N3  4  , −3 5 1 . Here, x1 exhibits
x3 7 −3 1 5

5
a negative correlation with both x2 and x3 . Conversely, x2 displays a positive
correlation with x 3 .     
x1 10 8 −4 −4
ii) For X = x2  ∼ N3  4  , −4 4 0 , x1 shares a negative
x3 7 −4 0 4
correlation with both x2 and x3 . Additionally, x2 and x3 are independent,
indicated by a correlation coefficient of0 inthe
 multivariate
normal distribution.
10 5 3 3
(5) i) Distance from 3a, x ∼ N3  4  , 3 3 −2.
  7 3 −2 14
x1
- Marginal : We need to consider the sub-covariance matrix B for x1
x2  
′ 5 3
and x2 , denoted as Σ = .
  3 3
x
- Marginal 1 : We need to consider the sub-covariance matrix B for x1 and
x2
   
5 3 x
x2 , denoted as Σ = ′
, and µ = [10, 4] . Therefore, 1 ∼ N2 (µ′ , Σ′ ),
′ T
3 3 x2
which is non-degenerate.
Now, X̄1 :
 
48x2 + 18x3 + 83 1
N1 , .
38 38
Note that, this sub-matrix is non-degenerate
   
10 9 −3 −3
iii) Distribution from Q 4 (iii): X ∼ N3  4  ,  −3 5 1 .
     7 −3 1 5
x1 10 9 −3
- Marginal Distribution : N2 , . Non-degenerate.
x2 4 −3 5
 
x2
Conditional Distribution of x1 given = x̄1 :
x3
 
−x2 − x3 + 31
x̄1 ∼ N1 ,6
2
         
10 −3 −1 9 −3 −3 −1 −3
x̄12 ≈ N2 + 5 (x3 − 7) , − 5
4 1 −3 5 1 1
 3x3 +71   36 12

− 5 −5
∼ N2 x3 +13 , 512 24
5
− 5 5

[Non-degenerate]  
x1
- Conditional Distribution given x3 = x̄12
x2
[Non-degenerate]

6
    
10 8 −4 −4
iv) Distribution from Q 4 (iv): X ∼ N3  4  ,  −4 4 0 .
7 −4 0 4
     
x1 10 8 −4
- Marginal Distribution : N2 , .
x2 4 −4 4
[Non-degenerate]
 
x2
Conditional Distribution of x1 given := x̄1
x3
  !
 4 0 −1 x2 −4  4 0 −1 −4
    
 
x̄1 ∼ N1 10 + −4 −4 , 8 − −4 −4
0 4 x3 −7 0 4 −4
= N1 (−x − y + 21, 0) [As V ar = 0].
 
x1
x3 = x̄12
x2
       −1 !
10 −4 −1 8 −4 −4  
X̄12 ≈ N2 + 4 (x3 − 7), − −4 0
4 0 −4 4 0
   
−x3 + 17 4 −4
= N2 , [Non-degenerate].
4 −4 4

x .