Matrix Decomposition

This chapter deals with the following topics.

(a) Eigenvalues and eigenvectors.

(b) Diagonalization of a square matrix.

(c) Singular value decomposition.

(d) Cholesky decompostion.

1 Eigenvalues and Eigenvectors

Let A be a square matrix of order n. A scalar λ is as an eigenvalue of A if there is
a non-zero column vector v of length n such that Av = λv . The vector v is called an
eigenvector of A corresponding to the eigenvalue λ.
Equivalently, a scalar λ is an eigenvalue of A if det(A − λIn ) = 0, and the eigenvectors
corresponding to λ are the solutions of the homogeneous system (A − λIn )X = 0.

The characteristic polynomial of A is denoted by ∆(A), and is dened as ∆(A) =

det(A − λIn ). Henceforth, the eigenvalues of A are the roots of the characteristic polyno-
mial.

The following theorem gives the characteristic polynomial of A in terms of principal mi-
nors of A.

If λ is an eigenvalue of A, then the solution space of the system (A − λIn )X = 0 is called

the eigenspace of A corresponding to the eigenvalue λ and is denoted by Eλ (A).

Theorem 1.1. Let A be a square matrix of order n. Let Sk denote the sum of all principal
minors of A of order k. Then ∆(A) = λn − S1 λn−1 + S2 λn−2 − + · · · + (−1)n Sn .

Note: S1 = trace(A) and Sn = det(A).

Example 1.2. (a) If A is a square matrix of order 2, then ∆(A) = λ2 −trace(A)λ+det(A).

(b) If A is a square matrix of order 3, then ∆(A) = λ3 − trace(A)λ2 + S2 λ − det(A).

1
Problem 1.3.
 Find
 a basis for the eigenspace corresponding to each eigenvalue of the
1 2
matrix A = 2 1
.

Solution : The characteristic polynomial of A is λ2 − 2λ − 3. Therefore the eigenvalues of

A are λ = −1, 3. Let X = [x, y]T .

Eigenspace corresponding to λ = −1:

From the equation (A − λI2 )X = 0, we get x + y = 0. . Thus Eλ (A) = {(x, y) : x + y = 0}.
So, {(1, −1)} is a basis of Eλ (A).

Eigenspace corresponding to λ = 3:
From the equation (A − λI2 )X = 0, we get x − y = 0. Thus Eλ (A) = {(x, y) : x − y = 0}.
So, {(1, 1)} is a basis of Eλ (A).

Problem 1.4.
 Find a basis
 for the eigenspace corresponding to each eigenvalue of the
0 −1 1
matrix A = −1 0 1.

1 −1 0

Solution : The characteristic polynomial of A is λ3 − λ. Therefore the eigenvalues of A

are λ = −1, 0, 1. Let X = [x, y, z]T .

Eigenspace corresponding to λ = −1:

From the matrix equation (A − λI3 )X = 0, we get x − y + z = 0; −x + y + z = 0.
Employing Gauss elimination method, we have x − y + z = 0; z = 0. Thus Eλ (A) =
{(x, y, 0) : x − y = 0}. So, {(1, 1, 0)} is a basis of Eλ (A).

Eigenspace corresponding to λ = 0:
The equation (A − λI3 )X = 0 implies x = y and y = z . Thus Eλ (A) = {(x, x, x) : x ∈ R}.
So, {(1, 1, 1)} is a basis of Eλ (A).

Eigenspace corresponding to λ = 1:
The equation (A − λI3 )X = 0 implies x = z and y = 0. Thus Eλ (A) = {(x, 0, x) : x ∈ R}.
So, {(1, 0, 1)} is a basis of Eλ (A).

2
Problem 1.5.
 Find a basis
 for the eigenspace corresponding to each eigenvalue of the
5 −4 4
matrix A = 12 −11 12.
4 −4 5
Solution : The characteristic polynomial of A is λ3 +λ2 −5λ+3. Therefore the eigenvalues
of A are λ = −3, 1, 1. Let X = [x, y, z]T .

Eigenspace corresponding to λ = −3:

From the matrix equation (A − λI3 )X = 0, we get x − y + 2z = 0; 2x − y + z =
0; 3x − 2y + 3z = 0. Employing Gauss elimination method, we have x − y + 2z = 0;
y − 3z = 0. Thus Eλ (A) = {(y/3, y, y/3) : y ∈ R}. So, {(1/3, 1, 1/3)} is a basis of Eλ (A).

Eigenspace corresponding to λ = 1:
The equation (A−λI3 )X = 0 implies x−y+z = 0. Thus Eλ (A) = {(x, y, z) : x−y+z = 0}.
Substituting (y, z) = (1, 0) and (y, z) = (0, 1), we get (1, 1, 0) and (−1, 0, 1) as two
independent solutions. So, {(1, 1, 0), (−1, 0, 1)} is a basis of Eλ (A).

Properties of eigenvalues:
(1) Any square matrix and its transpose have the same eigenvalues.

(2) The eigenvalues of a triangular matrix are the diagonal elements of the matrix.

(3) The sum of the eigenvalues of a matrix is the sum of the elements of the principal
diagonal(trace of a matrix).

(4) The product of the eigenvalues of a matrix is equal to its determinant.

1
(5) If λ is an eigenvalue of a non-singular matrix A, then is an eigenvalue of A−1 .
λ
Theorem 1.6. Cayley Hamilton Theorem: Every square matrix satises its characteristic
polynomial.
Problems:
 
2 −1 2
(1) Let A =  −1 2 −1 . Using Cayley Hamilton theorem, express A4 as a linear
1 −1 2
combination of the matrices A3 , A2 , A and I .

3
  
1 1 2
(2) Find the inverse of the matrix A =  9 2 0  using Cayley-Hamilton theorem.
5 0 3
Proposition 1.7. If A is a real symmetric matrix of order n. Then all its eigenvalues
are real.

Proof. Suppose Av = λv for some non-zero column vector v. Then vT Av = λvT v. Since
A is real symmetric, v T A = (Av)T = λv T . Therefore, λv T v = λv T v . So, λ = λ because v
is a non-zero column vector.

Proposition 1.8. Let A be a square matrix of order n. If λ1 , λ2 , . . . , λk are the distinct

eigenvalues of A corresponding to the eigenvectors v1 , v2 , . . . , vk , respectively. Then the
vectors v1 , v2 , . . . , vk are linearly independent.

Proposition 1.9. Let A be a symmetric matrix of order n. If λ1 , λ2 , . . . , λk are the dis-

tinct eigenvalues of A corresponding to the eigenvectors v1 , v2 , . . . , vk , respectively. Then
the vectors v1 , v2 , . . . , vk are orthogonal.

2 Diagonalization of a Matrix
• Square matrices A and B are similar if there exists an invertible matrix P such that
A = P BP −1 .

Proposition 2.1. If A and B are similar matrices, then they share the same characteristic
polynomial.

• A square matrix A is diagonalizable if it is similar to a diagonal matrix.

Proposition 2.2. A square matrix A of order n is diagonalizable if and only if it has n

linearly independent eigenvectors.

Note: If a square matrix A of order n has ‘n′ distinct eigenvalues, then it is diagonalizable.

Diagonalization of a square matrix A of order n: Working procedure

Step 1 Find all the eigenvalues of the matrix A.

4
Step 2 Find n linearly independent eigenvectors of A.

Step 3 If v1 , v2 , . . . , vn are linearly independent eigenvectors corresponding to the eigen-

values λ1 , λ2 , . . . , λn , respectively, then set matrix P = [v1 v2 . . . vn ] and D =
diag(λ1 , λ2 , . . . , λn ).

Step 4 A = P DP −1 .

Problems:
√
 
1 1
1. Diagonalize the matrix A = . Hence nd A.
2 2
Solution : The characteristic polynomial of A is λ2 − 3λ. Thus the eigenvalues of A are
0 and 3.

Let X = [x, y]T .

Eigenspace corresponding to λ = 0:
From the matrix equation (A − λI2 )X = 0, we get x + y = 0. Thus Eλ (A) = {(x, −x) :
x ∈ R}. So, {(1, −1)} is a basis of Eλ (A).

Eigenspace corresponding to λ = 3:
From the matrix equation (A − λI2 )X = 0, we get 2x − y = 0. Thus Eλ (A) = {(x, 2x) :
x ∈ R}. So, {(1, 2)} is a basis of Eλ (A).
" #
2/3 −1/3
   
1 1 0 0
Let P = and D = . Then P −1 = .
−1 2 0 3 1/3 1/3
Hence, A = P DP −1 .
√
To nd A:
" # " √ √ #
√ √ 2/3 −1/3 (1/3) 3 (1/3) 3
 
1 1 0 √0
A = P DP −1 = = √ √ .
−1 2 0 3 1/3 1/3 (2/3) 3 (2/3) 3
 
0 −1 1
2. Diagonalize the matrix A = −1 0 1. Hence nd A100 .
1 −1 0
Solution : The characteristic polynomial of A is λ3 − λ. Therefore the eigenvalues of A
are λ = −1, 0, 1. Let X = [x, y, z]T .

5
 Eigenspace corresponding to λ = −1:
From the matrix equation (A − λI3 )X = 0, we get x − y + z = 0; −x + y + z = 0.
Employing Gauss elimination method, we have x − y + z = 0; z = 0. Thus Eλ (A) =
{(x, y, 0) : x − y = 0}. So, {(1, 1, 0)} is a basis of Eλ (A).

Eigenspace corresponding to λ = 0:
The equation (A − λI3 )X = 0 implies x = y and y = z . Thus Eλ (A) = {(x, x, x) : x ∈
R}. So, {(1, 1, 1)} is a basis of Eλ (A).

Eigenspace corresponding to λ = 1:
The equation (A − λI3 )X = 0 implies x = z and y = 0. Thus Eλ (A) = {(x, 0, x) : x ∈
R}. So, {(1, 0, 1)} is a basis of Eλ (A).
 
    1 0 −1
1 1 1 −1 0 0
Let P =  1 1 0  and D =  0 0 0 . Then P −1 =  .
 
 −1 1 1 
0 1 1 0 0 1
1 −1 0
Hence A = P DP −1 .

To nd A100 :    

1 1 1

1 0 0 
 1 0 −1 2 −1 −1
  
A100 = P D100 P −1 =  1 1 0   0 0 0  
 −1 1  =  1 0 −1  .
1   
0 1 1 0 0 1
1 −1 0 1 −1 0
 
1 1 1
3. Diagonalize the matrix A = 2 2 2.
1 1 1
Solution : The characteristic polynomial of A is λ3 − 4λ2 . Therefore the eigenvalues of A
are λ = 0, 0, 4. Let X = [x, y, z]T .

Eigenspace corresponding to λ = 0:
The equation (A−λI3 )X = 0 implies x+y+z = 0. Thus Eλ (A) = {(x, y, z) : x+y+z = 0}.
Substituting (y, z) = (1, 0) and (y, z) = (0, 1), we get (−1, 1, 0) and (−1, 0, 1) as two in-
dependent solutions. So, {(−1, 1, 0), (−1, 0, 1)} is a basis of Eλ (A).

Eigenspace corresponding to λ = 4:

6
From the matrix equation (A−λI3 )X = 0, we get −3x+y+z = 0; x−y+z = 0; x+y−3z =
0. Employing Gauss elimination method, we have x + y − 3z = 0; y − 2z = 0. Thus
Eλ (A) = {(z, 2z, z) : z ∈ R}. So, {(1, 2, 1)} is a basis of Eλ (A).
 
    −1/2 1/2 −1/2
−1 −1 1 0 0 0
Let P =  1 0 2  and D =  0 0 0 . Then P −1 =  .
 
 −1/4 −1/4 3/4 
0 1 1 0 0 4 1/4 1/4 1/4
Hence A = P DP −1 .

2.1 Spectral theorem for symmetric matrices

A square matrix A is orthogonally diagonalizable if there exists an orthogonal matrix P
such that A = P DP T , where D is a diagonal matrix.
The following is the spectral theorem for symmetric matrices.

Theorem 2.3. Let A be a symmetric matrix of order n. Then

a. A has n real eigenvalues, counting multiplicities.

b. If λ is an eigenvalue of A with multiplicity k, then the dimension of the eigenspace

Eλ (A) is k .

c. The eigenvectors corresponding to dierent eigenvalues are orthogonal.

d. A is orthogonally diagonalizable.

Problems:
 
0 1 1
(1) Find an orthogonal matrix that diagonalizes the matrix A =  1 0 1 .
1 1 0
Solution : The characteristic polynomial of A is λ3 − 3λ − 2. Therefore the eigenval-
ues of A are λ = −1, −1, 2. Let X = [x, y, z]T .

Eigenspace corresponding to λ = −1:

The equation (A − λI3 )X = 0 implies x + y + z = 0. Thus Eλ (A) = {(x, y, z) :

7
 x + y + z = 0}. Substituting (y, z) = (1, 0) and (y, z) = (0, 1), we get (−1, 1, 0)
1 1
and (−1, 0, 1) as two independent solutions. So, { √ (−1, 1, 0), √ (1, 1, −2)} is
2 6
an orthonormal basis of Eλ (A).

Eigenspace corresponding to λ = 2:
From the matrix equation (A − λI3 )X = 0, we get x + y − 2z = 0; x − 2y + z =
0; 2x − y − z = 0. Employing Gauss elimination method, we have x + y − 2z = 0;
1
y − z = 0. Thus Eλ (A) = {(z, z, z) : z ∈ R}. So, { √ (1, 1, 1)} is an orthonormal
3
basis of Eλ (A).
 √ √   
−√ 3 1 √2 −1 0 0
1
Let P = √  3 1 √2  and D =  0 −1 0 . Then A = P DP T .
6 0 −2 2 0 0 2

 
2 −1
(2) Find an orthogonal matrix that diagonalizes the matrix A = .
−1 2
Solution : The characteristic polynomial of A is λ2 −4λ+3. Therefore the eigenvalues
of A are λ = 1, 3. Let X = [x, y]T .

Eigenspace corresponding to λ = 1:
From the equation (A − λI2 )X = 0, we get x − y = 0. . Thus Eλ (A) = {(y, y) : y ∈
1
R}. So, { √ (1, 1)} is an orthonormal basis of Eλ (A).
2
Eigenspace corresponding to λ = 3:
From the equation (A−λI2 )X = 0, we get x+y = 0. Thus Eλ (A) = {(x, y)
 : x+y =

1 1 1 1
0}. So, { √ (1, −1)} is an orthonormal basis of Eλ (A). Let P = √
 2  2 1 −1
1 0
and D = . Then A = P DP T .
0 3

3 Singular Value Decomposition

3.1 Positive semi-denite matrix
A real symmetric matrix A is called positive semi-denite if v T Av ≥ 0 for all non-zero
real column vector v .

8
Theorem 3.1. For a real-symmetric matrix A, the following are equivalent:
(i) vT Av ≥ 0 for all non-zero real column vector v.

(ii) All the eigenvalues of A are non-negative.

(iii) There exists a matrix B such that A = BB T .

Theorem 3.2.
A real symmetric matrix A of order 2 is positive semi-denite if and only its diagonal
elements are non-negative and |A| ≥ 0.

Exercise: Check whether the following matrices are positive semi-denite or not.
     
1 3 1 −2 1 −2
(i) (ii) (iii) .
3 4 −2 −3 −2 5

3.2 Singular values and singular vectors

Let A be an m × n matrix. A scalar λ is called a singular value of the matrix A if there
is a pair of non-zero column vectors u and v such that Av = λu and AT u = λv . The
vectors u and v are called as singular vector pair corresponding to the singular value λ.
More precisely, the vector u is called as left singular vector and the vector v is called as
right singular vector.

Equivalently, if min{m, n} = n, then the sigular values of A are the positive square roots
of the eigenvalues of AT A. Otherwise, the singular values of A are the positive roots of
the eigenvalues of AAT .

Note:

(i) rank(A) = rank(AT A).

(ii) The matrices AT A and AAT have the same set of non-zero eigenvalues.

(iii) The left singular vectors of A are the eigenvectors of AAT and the right singular
vectors of A are the eigenvectors of AT A.

9
Example 3.3.
" # " #
1 2 5 4
(i) Let A = . Then AT A = . and its eigenvalues are 1 and 9. There-
2 1 4 5
fore the singular values of A are 1 and 3.

" #
3 6
 
1 1 1
(ii) Let A = . Then AAT = and its eigenvalues are 15 and 0.
2 2 2 6 12
√
Therefore the singular values of A are 15 and 0 .

" #
333 81
 
4 11 14
(iii) Let A = . Then AA ∗ T = and its eigenvalues are 90,
8 7 −2 81 117
√ √
360. Therefore the singular values of A are 90 and 360.

Theorem 3.4. Let A be an m × n matrix of rank r. Suppose {v1 , v2 , . . . , vn } is an

orthonormal set of eigenvectors of AT A corresponding to the eigenvalues λ1 ≥ λ2 ≥ · · · ≥
λn , respectively. Then {Av1 , Av2 , . . . , Avr } is an orthogonal set and spans the column
space of A.

3.3 Singular value decomposition

Theorem 3.5 (Singular value decomposition). Let A be a matrix of order m × n. Let
σ1 ≥ σ2 ≥ · · · ≥ σr be the non-zero singular values of A. Then A = U V T , where
P

(i) U is an m × m orthogonal matrix whose column vectors u1 , u2 , . . . , um form a set of

orthonormal eigenvectors of AT A (or orthonormal left singular vectors of A).

(ii) V is an n × n orthogonal matrix whose column vectors v1 , v2 , . . . , vn form a set of

orthonormal eigenvectors of AAT (or orthonormal right singular vectors of A).
0
 
D
(iii) is an m × n matrix of the form 0 0 where D is the diagonal matrix of order
P

r given by D = diag(σ1 , σ2 , . . . , σr ).

10
Working procedure:
To determine a singular value decomposition of a matrix A of order m × n (m ≥ n):

(i) Determine an orthonormal set of eigenvectors v1 , v2 , . . ., vn of AT A corresponding

to the eigenvalues λ1 ≥ λ2 ≥ · · · ≥ λr > 0, 0, . . . , 0.
| {z }
n−r times

(ii) Set V to be an n × n matrix whose ith column is vi .

Avi
(iii) For 1 ≤ i ≤ r, set ui := .
||Avi ||
(iv) Determine vectors ur+1 , ur+2 , . . . , um such that the set {u1 , u2 , . . . , um } is an or-
thonormal set of m vectors.

(v) Set U to be an m × m matrix whose ith column is ui .

(vi) Then, A = U V T.
P

Problems: " #
1 1 1
1. Find a singular value decomposition of the matrix A = .
1 −1 1
 
2 0 2
Solution: Consider AT A = 
 0 2 0 . Its eigenvalues are 0, 2 and 4. Therefore the
 

2 0 2
√
singular values of A are 2, 2 and 0.

Orthonormal basis of the eigenspace corresponding to λ = 4:

1
From the equation (AT A − λI3 )X = 0, we get x − z = 0 and y = 0. Thus, { √ (1, 0, 1)}
2
is an orthonormal basis of Eλ (AT A).

Orthonormal basis of the eigenspace corresponding to λ = 2:

From the equation (AT A − λI3 )X = 0, we get x = z = 0. Thus, {(0, 1, 0)} is an orthonor-
mal basis of Eλ (AT A).

11
Orthonormal basis of the eigenspace corresponding to λ = 0:
From the matrix equation (AT A − λI3 )X = 0, we get x + z = 0 and y = 0. Therefore,
1
{ √ (1, 0, −1)} is an orthonormal basis of EAT A (λ).
2
1 1 Av1 1
Set v1 = √ (1, 0, 1)T , v2 = (0, 1, 0)T and v3 = √ (1, 0, −1)T , u1 = = √ (1, 1)T
2  2  ||Av1 || 2
1 0 1
√
 
Av2 1 1 , U = √1 1 1
and u2 = = √ (1, −1)T . Let V = √  and
 
0 2 0
||Av2 || 2 2  2 1 −1
1 0 −1
 
2 √0 0
. Then A = U V T .
P P
=
0 2 0
 
0 1 0
2. Find a singular value decomposition of the matrix A = 
 1 0 0 .
 

0 0 2
 
1 0 0
Solution: Consider AT A =   0 1 0  . Its eigenvalues are 1, 1 and 4. Thus the singular
 

0 0 4
values of A are 1, 1 and 2.

Orthonormal basis corresponding to the eigenvalue λ = 4:

From the matrix equation (AT A − λI3 )X = 0, we get x = 0 and y = 0. Therefore
{(0, 0, 1)} is an orthonormal basis of Eλ (AT A).

Orthonormal basis corresponding to the eigenvalue λ = 1:

From the matrix equation (AT A − λI3 )X = 0, we get z = 0. Therefore {(1, 0, 0), (0, 1, 0)}
is an orthonormal basis of Eλ (AT A).
Av1
Set v1 = (0, 0, 1)T , v2 = (1, 0, 0)T and v3 = (0, 1, 0)T , u1 = = (0, 0, 1)T , u2 =
 ||Av1 ||  
0 1 0 0 1 0
Av2 Av3
= (0, 1, 0)T and u3 = = (1, 0, 0)T Let V =  , U = 
   
0 0 1 0 0 1 
||Av2 || ||Av3 ||    
1 0 0 1 0 0
 
2 0 0
and =  0 1 0 . Then A = U V T .
P P
0 0 1

12
  
−3 1
3. Find a singular value decomposition of the matrix A = 
 
 6 −2 
.
6 −2
 
10 −20 −20
Solution: Consider AAT =  . Its eigenvalues are 0, 0 and 90. There-
 
 −20 40 40 
−20 40 40
√
fore the singular values of A are 0 and 3 10.

Orthonormal basis corresponding to the eigenvalue λ = 90:

From the matrix equation (AAT − λI3 )X = 0, we get
 4x + y + z = 0, 2x + 5y− 4z = 0
x y z
and 2x − 4y + 5z = 0. Therefore, Eλ (AAT ) = (x, y, z) : = = . Thus,
1 −2 −2
1
{ (1, −2. − 2)} is an orthonormal basis of Eλ (AAT ).
3
Orthonormal basis corresponding to the eigenvalue λ = 0:
From the matrix equation (AAT − λI3 )X = 0, we get x − 2y − 2z = 0. Therefore,
Eλ (AAT ) = {(x, y, z) : x − 2y − 2z = 0}. Let (y, z) = (0, 1) and (y, z) = (1, 0). Then,
{(2, 0, 1), (2, 1, 0)} is a basis of Eλ (AAT ). Employing Gram-Schmidt orthogonalization
1 1
process, we get an orthonormal basis of Eλ (AAT ) as { √ (2, 0, 1), √ (2, 5, −4)}.
5 3 5
1 1 1 AT u1
Set u1 = (1, −2, −2)T , u2 = √ (2, 0, 1)T , u3 = √ (2, 5, −4)T , v1 = =
3 5 3 5 ||Au1 ||
1 1 1
√ (−3, 1)T . An unit vector orthogonal to v1 is √ (1, 3)T . Dene u2 = √ (1, 3)T .
10 10 10
 √   √ 
  5/3
√ 2 2/3 3 10 0
1 −3 1 1
Let V = √ , U = √  −2√5/3 0 5/3  and
P
=  0 0 .
10 1 3 5 −2 5/3 1 −4/3 0 0
Then A = U V .
P T

4 Cholesky decomposition
Let A = [aij ]n×n be a positive denite matrix. Then A can be written A = LLT , where
L = [lij ]n×n is a lower triangular and it is called as Cholesky decomposition of A.

13
 The entries of the matrix L can be computed as follows:
√
∗ l11 = a11 .
aj1
∗ lj1 = , j = 2, . . . , n.
l11
v
u j−1
, j = 2, 3, . . . , n.
u X
2
∗ ljj = ajj −
t ljs
s=1

j−1
!
1
ljs lps , p = j + 1, j + 2, . . . , n and j ≥ 2.
X
∗ lpj = apj −
ljj s=1

Note: The Cholesky decomposition is mainly used to solve the linear system AX = b. If
A is symmetric and positive denite, then we can solve by rst computing the Cholesky
decomposition A = LLT , then solving for Ly = b for y by forward substitution, and nally
solving for x by back substitution.

Problems:
 
4 2 14
1. Find the Cholesky decomposition of  2 17 −5 .
14 −5 83
Solution:
√
l11 = a11 = 2.
a21 a31
l21 = = 1; l31 = = 7.
l11 l11
= 4.
p
2
l22 = a22 − l21
1
l32 = (a32 − l21 l31 ) = −3.
l22
= 5.
p
2 2
l33 = a33 − l31 − l32

Therefore,
    
4 2 14 2 0 0 2 1 7
 2 17 −5  =  1 4 0   0 4 −3 
14 −5 83 7 −3 5 0 0 5

14
  
9 6 12
2. Find the Cholesky decomposition of  6 13 11 .
12 11 26
Solution:
√
l11 = a11 = 3.
a21 a31
l21 = = 2; l31 = = 4.
l11 l11
p
2
l22 = a22 − l21 = 3.
1
l32 = (a32 − l31 l21 ) = 1.
l22
p
2 2
l33 = a33 − l31 − l32 = 3.
Therefore,
    
9 6 12 3 0 0 3 2 4
 6 13 11  =  2 3 0   0 3 1 .
12 11 26 4 1 3 0 0 3

Exercise:

(1) Diagonalize the following

 matrices.
" # 1 3 3
1 1
(i) (ii)  .
 
−3 −5 −3
3 −1  
3 3 1

(2) Find an orthogonal matrix P that diagonalizes the following matrices.

   
 
 6 −2 −1   4 −1 −1 
(i)  1 −2  (ii) 
 −2 6 −1  (iii)  −1 4 −1  .
  
−2 1
−1 −1 5 −1 −1 4
" #
2 −1
(3) Find a singular value decomposition of the following matrices. (i) (ii)
2 2
 
1 1 " #
1 2 1
 0 1  (iii) .
 
  2 1 1
−1 1

15
(4) Find the Cholesky decomposition of the following matrices.
 
  1 −1 3 2
4 6 8  −1 5 −5 −2 
(i) 6 34 52  (ii) 

 3 −5 19 3 .

8 52 129
2 −2 3 21

16