MA5158 ENGINEERING MATHEMATICS I

UNIT I - MATRICES
Section 5. Diagonalization using an orthogonal
transformation

Faculty
Department of Mathematics
Anna University, Chennai

Strictly for University Departments only 1

Section 5. Diagonalization using an
orthogonal transformation

Contents
Preliminaries. Orthogonal vectors
Orthogonal reduction / diagonalization using an
orthogonal matrix
Examples
Practice Problems and MCQ’s

Strictly for University Departments only 2

Preliminaries: Orthogonal vectors
   
x1 y1
 x2   y2 
n
   
Definition: Two vectors X =   .  and Y =  .  in R are said
  
 .   . 
xn yn
to be orthogonal if their dot product X · Y = x1 y1 + x2 y2 + · · · + xn yn is
zero.
Geometrically, the vector X and the vector Y are orthogonal if they are
perpendicular in Rn .
   
1 −5
Example: The vectors X =  −2  and Y =  11  are orthogonal
9 3
in R3 as their dot product
X · Y = 1.(−5) + (−2).11 + 9.3 = −5 − 22 + 27 = 0.

Strictly for University Departments only 3

Orthogonal vectors vs linearly independent vectors

Note: Orthogonal vectors are linearly independent. For example if

X1 , X2 ∈ Rn are orthogonal, they are perpendicular to each other. So
they do not lie on the same line. Therefore they are independent. But
the converse need not be true.
   
Example: The vectors X = and Y =
1 0
are linearly
1 1
independent in R2 but not orthogonal in R2 as their dot product
X · Y = 1.1 + 1.0 = 1 6= 0.
Remark: Geometrically, if two vectors in Rn are linearly independent,
then they do not lie on the same line. But the angle between the vectors
need not be 90◦ . Therefore they need not be perpendicular.

Strictly for University Departments only 4

Orthonormal system
Definition: A set of vectors X1 , X2 , X3 , . . . , Xn in Rn is said to be
orthonormal if
(i) Xi · Xj = 0 for i 6= j (i.e. Xi ’s are pairwise orthogonal),

(ii) kXi k = 1, i = 1, 2, . . . , n (i.e. each Xi has norm 1).

   
1/3 2/3
Example: The vectors X1 =  2/3 , X2 =  1/3 ,
2/3 −2/3
 
2/3
X3 =  −2/3  form an orthonormal system in R3 .
1/3
Check: X1 · X2 = X2 · X3 = X1 · X3 = 0 and kX1 k = kX2 k = kX3 k = 1.

Strictly for University Departments only 5

Diagonalization of symmetric matrices using an
orthogonal matrix/orthogonal transformation
We know how to diagonalize a square matrix A of order n, when it has n
linearly independent eigen-vectors.

When the matrix A is a symmetric matrix of order n with n linearly

independent eigen-vectors, we can diagonalize it using an orthogonal
matrix.

This is due to the following property of the eigen vectors of a symmetric

matrix which we have already studied.

Property 9: The eigen-vectors corresponding to distinct eigen-values of

a symmetric matrix A are orthogonal.

Strictly for University Departments only 6

Orthogonal reduction

Theorem: Let A be a symmetric matrix of size n with n orthonormal

eigen-vectors. Then we can diagonalize A using an orthogonal matrix.
Proof:Take n λ1 , λ2
= 3. Let , λ3 be the
 eigen-values
 of A. Let
x11 x12 x13
X1 =  x21 , X2 =  x22 , X3 =  x23  be the corresponding
x31 x32 x33

eigen-vectors of A which form an orthonormal system.

Strictly for University Departments only 7

  
x11 x12 x13
Let P = [X1 X2 X3 ] =  x21 x22 x23  be the matrix whose column
x31 x32 x33
vectors are the eigen-vectors of A which are pairwise orthogonal and
each eigen vector has norm 1.
  
x11 x21 x31 x11 x12 x13
T
Consider P P =  x12 x22 x32   x21 x22 x23 
x x x33 x31 x32 x33
 13 232   
kX1 k X1 · X2 X1 · X3 1 0 0
=  X2 · X1 kX2 k2 X2 · X3  =  0 1 0 ,
X3 · X1 X3 · X2 kX3 k2 0 0 1
the identity matrix. Similarly one can check PP T = I. Therefore P is an
orthogonal matrix. P is called the normalized modal matrix.

Strictly for University Departments only 8

Use this P for diagonalizing A.
Note that since PP T = I, we have P −1 = P T .
 
λ1 0 0
You will get P T AP = D =  0 λ2 0  which is a diagonal matrix
0 0 λ3
whose diagonal elements are the eigen values of A.
This method of diagonalizing a symmetric matrix using an orthogonal

matrix is called orthogonal reduction of A.

Strictly for University Departments only 9

Example 1 - A has distinct eigen values
Example 1. (A has distinct eigen values)
 
1 1 3
Diagonalize the matrix A =  1 5 1  using an orthogonal
3 1 1
transformation.
Solution:

 equation of A is |A − λI| = 0.
Step 1: The characteristic
 1−λ 1 3 

 1
 5−λ 1  = 0 i.e. λ3 − 7λ2 + 36 = 0 is the characteristic
 3 1 1−λ 
equation of A.

Step 2:The eigen values of A (roots of this characteristic equation) are

λ = −2, 3, 6.

Strictly for University Departments only 10

Step 3: To
 findthe eigen vectors:
x1
Let X =  x2  be an eigen vector corresponding to the eigen value λ.
x3
Then (A − λI)X = 0.
    
1−λ 1 3 x1 0
 1 5−λ 1  x2  = 0 (1)
3 1 1−λ x3 0
When λ = −2 in equation 1, we have the following system of equations:

3x1 + x2 + 3x3 = 0 (2)

x1 + 7x2 + x3 = 0 (3)
3x1 + x2 + 3x3 = 0 (4)

Strictly for University Departments only 11

Equations (2) and (4) are the same. Therefore, from equations (2) and (3)

x1 x2 x3
we have
1 − 21 −(3 − 3) 21 − 1
= =

x1 x2 x3
or
−20
= =
0 20
x1 x2 x3
or
−1
= =
0 1
 an eigen vector corresponding to the eigen value λ = −2 is
Therefore

−1
 0 .
1
Any non-zero multiple of this vector is also an eigen vector corresponding
to the eigen value λ = −2.

Strictly for University Departments only 12

When λ = 3 in equation (1), we have the following system of equations:

−2x1 + x2 + 3x3 = 0 (5)

x1 + 2x2 + x3 = 0 (6)
3x1 + x2 − 2x3 = 0 (7)

As the determinant of the coefficient matrix is zero, at the most only two
of these equations are linearly independent. Therefore, from equations
(6) and (7) we have
x1 x2 x3
(−4 − 1) −(−2 − 3) 1−6
= =

x1 x2 x3
or
−5 −5
= =
5
x1 x2 x3
or
−1
= =
1 1

Strictly for University Departments only 13

  an eigen vector corresponding to the eigen value λ = 3 is
Therefore

1
−1.
1
When λ = 6 in equation (1), we have the following system of equations:

−5x1 + x2 + 3x3 = 0 (8)

x1 − x2 + x3 = 0 (9)
3x1 + x2 − 5x3 = 0 (10)

As the determinant of the coefficient matrix is zero, at the most only two
of these equations are linearly independent. Therefore, from equations
(9) and (10) we have
x1 x2 x3
(5 − 1) −(−5 − 3)
= =
1+3

Strictly for University Departments only 14

 x1 x2 x3
or
= =
4 8 4
x1 x2 x3
or
= =
1 2 1
 
1
Therefore an eigen vector corresponding to the eigen value λ = 6 is 2

1
Therefore,
  −2, 3, 6 are
 the eigen values
  and
−1 1 1
0
X1 =   0   0
0 , X2 = −1 , X3 = 2 are the corresponding

1 1 1
eigen vectors of A.
Step 4: Form the normalized modal matrix
Note that these eigen vectors X10 , X20 , X30 are pairwise orthogonal as they
correspond to distinct eigen values −2, 3, 6 of the symmetric matrix A.

Strictly for University Departments only 15

  √ 
Let us normalize these vectors.
 √ 
0 −1/ 2 0 1/ √3
X1  0  , Let X2 = X2 = −1/ 3, Let
Let X1 = √ √
kX10 k kX20 k
=
√
1/ 2 1/ 3
 
X30
1/ √6
X3 = = 2/ √6 .
kX30 k
1/ 6
X1 , X2 , X3 are orthonormal.
 √ √ √ 
−1/ 2 1/ √3 1/ √6
Take P = [X1 X2 X3 ] =  0√ −1/√ 3 2/ √6 .
1/ 2 1/ 3 1/ 6

Then P is an orthogonal matrix. (Therefore, P −1 = P T ).

P is the normalized modal matrix.

Strictly for University Departments only 16

Step 5: Diagonalization by an orthogonal matrix
Now √
 consider P −1 AP =√ P T
AP
  √ √ √ 
−1/√ 2 0√ 1/ √2 1 1 3 −1/ 2 1/ √3 1/ √6
= 1/ √3 −1/√ 3
 1/ √31 5 1  0√ −1/√ 3 2/ √6
1/ 6 2/ 6 1/ 6 3 1 1 1/ 2 1/ 3 1/ 6
 
−2 0 0
= 0 3 0=D
0 0 6

which is a diagonal matrix whose diagonal entries are the eigen values of
A. Thus we have diagonalized A using an orthogonal matrix P.

Strictly for University Departments only 17

Example 2 - A has repeated eigen values
 
1 −2 2
Example 2: Diagonalize the matrix A =  −2 1 −2  using an
2 −2 1
orthogonal transformation.
Solution:
 equation of A is |A − λI| = 0.
Step 1: The characteristic
 1 − λ −2 2 

 −2 1 − λ −2  = 0 i.e. λ3 − 3λ2 − 9λ − 5 = 0 is the
 
 2 −2 1 − λ 
characteristic equation of A.

Step 2:The eigen values of A (roots of this characteristic equation) are

λ = −1, −1, 5. Note that the eigen value −1 is repeated two times.

Strictly for University Departments only 18

Step 3: To
 findthe eigen vectors:
x1
Let X =  x2  be an eigen vector corresponding to the eigen value λ.
x3
Then AX = λX or (A − λI)X = 0.
    
1 − λ −2 2 x1 0
 −2 1 − λ −2  x2  = 0 (11)
2 −2 1 − λ x3 0
When λ = 5 in equation (1), we have the following system of equations:

−4x1 − 2x2 + 2x3 = 0 (12)

−2x1 − 4x2 − 2x3 = 0 (13)
2x1 − 2x2 − 4x3 = 0 (14)

Strictly for University Departments only 19

From equations (12) and (13) we have
x1 x2 x3
−(8 + 4) 21 − 1
= =
4+8
x1 x2 x3
or
−1
= =
1 1
Therefore
 an eigen vector corresponding to the eigen value λ = 5 is
1
X10 = −1
1
Any non-zero multiple of this vector is also an eigen vector corresponding

to the eigen value λ = 5.

Strictly for University Departments only 20

When λ = −1 in equation (11), we have the following system of
equations:

2x1 − 2x2 + 2x3 = 0 (15)

−2x1 + 2x2 − 2x3 = 0 (16)
2x1 − 2x2 + 2x3 = 0 (17)

All these equations are the same. They are equal to

x1 − x2 + x3 = 0. (18)

In equation (18), take x3 = 0. We get x1 = x2 . if we take x2 = 1 then

x1 = 1.  
1
Let X20 = 1. This is one linearly independent eigen vector
0
corresponding to the eigen value λ = −1.

Strictly for University Departments only 21

  
x
We have to find another linearly independent eigen vector X30 = y
z
corresponding to λ = −1 such that X20 · X30 = 0.
i.e X30 should satisfy equation (18) and
 
x
X20 · X30 = (1 1 0) y = 1.x + 1.y + 0.z = 0.
z
Therefore, we have to solve the equations x − y + z = 0 and x + y = 0
simultaneously. From these two equations we have,
x y z
−1
= =
1 2

Strictly for University Departments only 22

Therefore the second
  eigen vector corresponding to the eigen value
−1
λ = −1 is X30 =  1 .
2
These eigen vectors satisfy X10 · X20 = 0, X20 · X30 = 0, X30 · X10 = 0.

They are pairwise orthogonal.

Step 4: Form the normalized modal matrix

Let us normalize these eigen vectors.

Strictly for University Departments only 23

  √   √ 
X10 0
1/ √3 1/ √2
X
Let X1 = −1/ 3, X2 =
√
2 1/ 2,
kX10 k kX20 k
= =
√ 
1/ 3 0

−1/√ 6
X30
X3 = √
 1/ 6  .
kX30 k
=
2/ 6
X1 , X2 , X3 are orthonormal.
 √ √ √ 
1/ √3 1/ √2 −1/√ 6
Take P = [X1 X2 X3 ] = −1/√ 3 1/ 2 1/ √6  .
1/ 3 0 2/ 6

Then P is an orthogonal matrix. (Therefore, P −1 = P T ).

P is the normalized modal matrix.

Strictly for University Departments only 24

Step 5: Diagonalization by an orthogonal matrix
Now consider P −1 AP = P T AP
 √ √ √  
1/ √3 −1/√ 3 1/ 3 1 −2 2
= 1/ √2
 1/ √2 0√   −2 1 −2
−1/ 6 1/ 6 2/ 6 −2 1
 √ √ √  
2

1/ √3 1/ √2 −1/√ 6 5 0 0
−1/ 3 1/ 2 1/ 6  =0
√ √ −1 0 =D
1/ 3 0 2/ 6 0 0 −1

which is a diagonal matrix whose diagonal entries are the eigen values of
A. Thus we have diagonalized A using an orthogonal matrix P.

Remark. This procedure has immediate application in the next section.

Strictly for University Departments only 25

Practice Problems

Diagonalize the following symmetric matrices using orthogonal

transformations:
 
1 1 3
(i) A =  1 5 1 
3 1 1
 
6 −2 2
(ii) A =  −2 3 −1 
2 −1 3

Strictly for University Departments only 26

Multiple Choice Questions

1. A set of linearly independent vectors are mutually orthogonal.

(a) True
(b) False

2. If P is orthogonal then P −1 is
(a) P
(b) P T
(c) P −1 does not exist
(d) None of the above

Strictly for University Departments only 27

Answers to MCQ’s

1. (b)
2. (b)

Strictly for University Departments only 28

THANK YOU

Strictly for University Departments only 29