2020

edition

KWAME NKRUMAH UNIVERSITY

DEPARTMENT OF MATHEMATICS AND STATISTICS

MAT 260- LINEAR ALGEBRA

MODULE 3: ORTHOGONALITY, CHARACTERISTIC ROOTS AND

VECTORS AND QUADRATIC FORMS

BY

SR. HORTENSIA ZULU

(2020)

1
 2020
edition

TABLE OF CONTENT

CONTENT PAGE

About the course 3

About this module 3

Unit 1: Orthogonality 7

Unit 2: Eigenvalues and Eigenvectors 15

Unit 3: Matrix Diagonalisation 23

Readings

2
 2020
edition

ABOUT THE COURSE

Introduction

Linear Algebra, MAT 260, is a second year course. The course comprise several topics. To
complete the course you need to study through the two modules:

Module 1: Linear Equations, Matrices and Determinants 25 hours

Module 2: Vector Spaces, Linear Transformations and Orthogonality 70 hours

Characteristic Roots and Vectors and Quadratic Forms

Course Out Comes

A student who successfully completes this course should be able to proceed to third year to
study advanced mathematics courses with less difficulty.

Course assessment

To successfully complete the course a student is required to be assessed through tests,

assignments and examination as tabulated below:

Three tests 30%

Three assignments 10%
Final examination 60%

ABOUT THIS MODULE

Time

This module may take you a minimum of 70 hours to work through. The recommended time
is for you to read through the module and do all the given activities. You may need more
hours to read through other suggested readings and websites. Time- tabling yourself may
help you to work through this module easily.

3
 2020
edition

Icons

In this module, you will be exposed to the following icons.

Learning Outcomes

Readings

Activity

Terminology

Reflections

Counselling and academic help

In the event that you would like help from the college, you may contact the
Distance Education Director on +260 979 85421. You may request for personal
contact phone numbers from the Director. Remember also that, our Library is
open every day except on weekends. The Library personnel shall be more than
glad to help you.

4
 2020
edition

PROLOGUE: What Is Linear Algebra?

Linear Algebra is about Vector spaces and transformations on them that are linear. It involves
systems of linear equations, Matrices and Vectors.

The use of Linear Algebra is almost in everything. It is used for;

a) Projections
b) Shadows
c) Computer graphics
d) Map – making
e) Forest management
f) Computer vision
g) Model – making
h) Representation molecules
i) Building bridges, churches etc
j) Atomic physics
k) Economic planning and
l) Color vision and many more.

5
 2020
edition

UNIT 1: UNIT STRUCTURE

Unit 1: Orthogonality
1.1 Learning Outcomes
1.2 Introduction
1.2.0 Inner Product Space
1.2.1 Length and Distance in inner Product Space
1.3 Orthogonality, Bases Gram – Schmidt Process
1.4 Co-ordinates relative to Orthonormal Bases
1.5 Co-ordinates relative to Orthogonal Bases
1.6 Change of Basis of Orthogonal Matrices
1.6.1 Orthogonal Matrices
1.7 Orthogonal Projections
Activity 1
Answers to some selected questions

6
 2020
edition

UNIT 1: ORTHOGONALITY
You are welcome to unit three of this module. This unit will introduce you to the

concept of orthogonality.

1.1 LEARNING OUTCOMES

By the end of this unit, you should be able to

• Define orthogonality
• Understand the Gram-schmidt process and use it to find an orthonormal basis.

1.2 INTRODUCTION:

In many problems involving vector spaces, the problem solver is free to choose any basis
for the vector space that seems appropriate. In inner product spaces the solution of a
problem is often greatly simplified by choosing a basis in which the vectors are orthogonal
to one another In this section, we shall show how such bases ca

1.2.0 Inner Product Spaces

Definition3.2.0: Let 𝑉 be a (real or complex) vector space over 𝐾(field). Suppose to each pair
of vectors 𝑢, 𝑣 ∈ 𝑉 there is assigned a scalar 〈𝑢 𝑣〉 ∈ 𝐾

This mapping is called an Inner Product in 𝑉 if it satisfies

1. 〈𝑢, 𝑣〉 = (𝑣 𝑢) (symmetry axiom)

2. 〈𝑢 + 𝑣, 𝑤〉 = 〈𝑢, 𝑤〉 + (𝑣, 𝑤) [additivity axiom]
3. 〈𝐾𝑢, 𝑣〉 = 𝐾〈𝑢, 𝑣〉 (Homogeneity property)
4. 〈𝑢, 𝑣〉 ≥ 0 (positivity property) 〈𝑢, 𝑢〉 = 0 if and only if 𝑢 = 0

The vector space 𝑉 with the inner product is called an inner product space. A real inner
product space is called a Euclidean space and a complex inner product space is called a
unitary space.

Example 1

If 𝑢 = (𝒖𝟏 , 𝒖𝟐 … … 𝒖𝒏 ) and 𝑣 = (𝒗𝟏 , 𝒗𝟐 … … 𝒗𝒏 ) are vectors in 𝑅 𝑛 , then

〈𝒖, 𝒗〉 = 𝑼. 𝑽 = 𝑢1 𝑣1 + 𝑢2 𝑣2 + … … + 𝑢𝑛 𝑣𝑛 is a Euclidean inner product on 𝑅 𝑛 .

7
 2020
edition

1.2.1 Length and Distance in Inner Product Spaces

Definition [Link]: If 𝑉 is an inner product space, then the norm (or length) of a vector u in𝑉
is denoted by ‖𝒖‖and defined by;
𝟏
‖𝒖‖ = √〈𝒖. 𝒖〉 = 〈𝒖, 𝒖〉𝟐

The distance between two points (vectors) 𝒖 and 𝒗 is denoted by 𝒅(𝒖𝒗) and is defined by
𝒅(𝒖, 𝒗) = ‖𝒖 − 𝒗‖

Example 1

If 𝑢 = (𝑢1 , 𝑢2 … … 𝑢𝑛 ) and 𝑣 = (𝑣1 , 𝑣2 … … 𝑣𝑛 ) are vectors in 𝑅 𝑛 with the Euclidean inner

product then.

1
‖𝑢‖ = 〈𝑢, 𝑢〉2 = √𝑢21 + 𝑢22 … … + 𝑢𝑛2

1 1
and 𝑑(𝑢𝑣) = ‖𝑢 − 𝑣‖〈𝑢 − 𝑣, 𝑢 − 𝑣〉2 = [(𝑢 − 𝑣). (𝑢 − 𝑣)]2

= √(𝑢1 − 𝑣1 )2 + (𝑢2 − 𝑣2 )2 + … … + (𝑢𝑛 − 𝑣𝑛 )2

Example 2

Find the norm and distance for the vectors 𝑢 = (1 0) and 𝑣 = (0 1)in ℝ2 .

Answer ‖𝑢‖ = 1. 𝑑(𝑢, 𝑣) = √2

Definition: If 𝑉 is an inner product space, then the set of points in 𝑉 that satisfy ‖𝑢‖ = 1 is
called the unit sphere or the unit cir cle in 𝑉.

1.2.2 Angle and Orthogonality in inner Product Spaces

If 𝑢 and 𝑣 are non-zero vectors in 𝑅 2 or 𝑅 3 and 𝜃 is the angle between them, then
𝑈.𝑉
𝑈. 𝑉 = ‖𝑢‖‖𝑣‖𝑐𝑜𝑠𝜃 (1) or 𝑐𝑜𝑠𝜃 = ‖𝑢‖‖𝑣‖ (2)

Squaring both sides of 1 we have ‖𝑢‖2 = 𝑢. 𝑢‖𝑣‖2 = 𝑣. 𝑣 and𝑐𝑜𝑠 2 𝜃 ≤ 1 we obtain the

inequality;

(𝑢. 𝑣)2 ≤ (𝑢. 𝑢)(𝑣. 𝑣) = 〈𝑢, 𝑣〉2 ‖𝑢‖2 ‖𝑣‖2 which is known as the (Cauchy – Schwarz
Inequality) or 〈𝑢, 𝑣 〉 ≤ ‖𝑢‖‖𝑣‖

Definition1.2.1.1 : In an inner product space, two vectors 𝒖 and 𝒗 are called Orthogonal if
〈𝒖 𝒗〉 = 𝟎. If 𝒖 is orthogonal to each vector in a set 𝑊, we say that 𝒖 is orthogonal to 𝑾.

8
 2020
edition

Example:

Let 𝑃2 have the inner product

1

〈𝑝 𝑞 〉 = ∫ 𝑃(𝑥)𝑞(𝑥)𝑑𝑥
−1

and let 𝑝 = 𝑥, 𝑞 = 𝑥 2

show that 𝑝 and 𝑞 are orthogonal relative to the given inner product.

Solution
1 1
〈𝑃 𝑞〉 = ∫ 𝑥𝑥 2 𝑑𝑥 = ∫ 𝑥 3 𝑑𝑥 = 0
−1 −1

〈𝑃 𝑞〉 = 0 ∴the vector 𝑝 = 𝑥 and 𝑞 = 𝑥 2 are orthogonal relative to the given inner product.
(Example 1)

If 𝑈 and 𝑉 are orthogonal vectors in an inner product space then ‖𝑢 + 𝑣‖2 = ‖𝑢‖2 + ‖𝑣‖2 .

Proof: ‖𝑢 + 𝑣‖2 = 〈(𝑢 + 𝑣)(𝑢 + 𝑣)〉 = ‖𝑢‖2 + 2〈𝑢, 𝑣) + ‖𝑣‖2

= ‖𝑢‖2 + ‖𝑣‖2

1.3 ORTHOGONAL BASES GRAM – SCHMIDT PROCESS

Definition3.3.0: A set of vectors in an inner product space is called orthogonal set if all pairs
of distinct vectors in the set are orthogonal. An orthogonal set in which each vector has a
norm 1 is called orthogonal set.

Example 1:

1. Let 𝑢1 = (0 1 0)𝑢2 = (1 0 1)𝑢3 = (1 0 −1) and assume that 𝑅 3 has the

Euclidean Inner Product.
The set of vectors 𝑆 = {𝑢1 𝑢2 𝑢3 } is orthogonal since
〈𝑢1 𝑢2 〉 = 〈𝑢1 𝑢3 〉 = 〈𝑢2 𝑢3 〉 = 0

− An orthogonal set of nonzero vectors can always be converted to an orthonormal set

by normalizing each of its vectors.

If 𝑉 is a non-zero vector in an inner product space, then it can be multiplied by the

reciprocal of its length to obtain a vector of norm 1. This is known as normalizing.e.g.
𝟏 𝟏
‖‖𝑽‖‖ = ‖𝑽‖ . ‖𝑽‖ = 𝟏

9
 2020
edition

Example 2:

Euclidean norms of the vectors in previous example are

‖𝑢1 ‖ = 1 ‖𝑢2 ‖ = √2‖𝑢3 ‖ = √2

Normalizing 𝑢1 𝑢2 𝑢3 we have

𝑢1 (0 1 0)
𝑣1 = = (0 1 0) =
‖𝑢‖ 1

𝑢2 (1 0 1) 1 1
𝑣2 = = (0 1 0) = =( , 0, )
‖𝑢1 ‖ √2 √2 √2
𝑢3 (1 0 −1) 1 −1
𝑣3 = = (1 0 −1) = = ( , 0, )
‖𝑢3 ‖ √2 √2 √2
〈𝑣1 , 𝑣2 〉 = 〈𝑣1 , 𝑣3 〉 = 〈𝑣2 , 𝑣3 〉 = 0

‖𝑣1 ‖ = ‖𝑣2 ‖ = ‖𝑣3 ‖ = 1 ∴ 𝑆 = {v1 , v2 , v3 }is an orthonormal set.

In an inner product space, a basis consisting of orthonormal vectors is called an orthonormal

basis and a basis consisting of orthogonal vectors is called an orthogonal basis. An
orthonormal set is an orthogonal set of unit vectors

1.4 CO-ORDINATES RELATIVE TO ORTHONORMAL BASES

Theorem 3.4.0: If 𝑆 = {v𝟏 , v𝟐 , … … v𝒏 }is an orthonormal basis for inner product space, 𝑽 and
𝒖 is any vector in 𝑉, then

𝒖 = 〈𝒖 𝒗〉, 𝒗𝟏 + 〈𝒖, 𝒗𝟐 )𝒗𝟐 + … … + 〈𝒖, 𝒗𝒏 〉𝒗𝒏 This formula expresses 𝑢 as linear

combination of the vectors in the orthonormal basis.

(〈𝑢𝑣1 〉, 〈𝑢𝑣2 〉 … … ⟨𝑢𝑣𝑛 ⟩is the co-ordinate of 𝑢 relative to the basis.

Example1 : Let 𝑆 = {𝑣1 𝑣2 𝑣3 }

−4 3 3 4
Let 𝑣1 = (0 1 0) 𝑣2 = ( 5 , 0, 5
) 𝑣3 = (5 , 0, )
5

Express the vector 𝑢 = (1 1 1) as a linear combination of the vectors in 𝑆 and find the co-
ordinate vector (𝑢)𝑠 .

Solution

10
 2020
edition

〈𝑢𝑣1 〉 = 1 = 0 + 1 + 0

−4 3 −1
〈𝑢𝑣2 〉 = +0+ =
5 5 5
3 4 7
〈𝑢𝑣3 〉 = +0+ =
5 5 5

𝑢 = 〈𝑢𝑣1 〉𝑣1 + 〈𝑢𝑣2 〉𝑣2 + 〈𝑢𝑣3 〉𝑣3

1 7
= 𝑣1 − 𝑣2 + 𝑣3
5 5
Express 𝑢 = (1 1 1) as a linear combination of the vectors in is

1 −4 3 7 3 4
𝑆(1 1 1) = (0 1 0) − ( , 0, ) + ( , 0, )
5 5 5 5 5 5

The co-ordinate vector of 𝑢 relative to 𝑆 is

1 7
(𝒖)𝑺 = (〈𝑢, 𝑣1 〉, 〈𝑢, 𝑣2 〉, 〈𝑢, 𝑣3 〉 = (1, − , )
5 5

1 7 2 75
√
‖𝑢‖ = 1 + (− ) + ( ) = √ = √3
2
5 5 25

1.5 CO-ORDINATES RELATIVE TO ORTHOGONAL BASES

If 𝑆 = {v1 , v2 , … … v𝑛 } is an orthogonal basis for a vector space 𝑉, then normalizing each of these
vectors yields the orthogonal basis.
𝑣1 𝑣2 𝑣𝑛
𝑆′ = { , ,……. }
‖𝑣1 ‖ ‖𝑣2 ‖ ‖𝑣𝑛 ‖

Then if 𝑢 is any vector in 𝑉, then

𝑣 𝑣 𝑣 𝑣 𝑣 𝑣
𝑢 = 〈𝑢, ‖𝑣1‖〉 ‖𝑣1‖ + 〈𝑢, ‖𝑣2 ‖〉 ‖𝑣2 ‖ + … … + 𝑢 〈‖𝑣𝑛 ‖〉 ‖𝑣𝑛 ‖which can be written as
1 1 2 2 𝑛 𝑛

11
 2020
edition

〈𝑢𝑣1 〉 〈𝑢𝑣2 〉 〈𝑢𝑣𝑛 〉

𝑢= 𝑣1 + 𝑣2 + … … + 𝑣
‖𝑣1 ‖2 ‖𝑣2 ‖2 ‖𝑣𝑛 ‖2 𝑛

This formula expresses 𝒖 as a linear combination of the vectors in the orthogonal basis 𝑺.

1.6 CHANGE OF BASIS; ORTHOGONAL MATRICES.

1.6.0 Change of Basis

If we change the basis for a vector space 𝑉 from some old basis 𝐵 = {𝑢1 𝑢2 … … 𝑢𝑛 } to some new
basis 𝐵′ = {𝑢1′ , 𝑢2′ , … … 𝑢𝑛′ } then the old co-ordinate matrix [𝑣]𝐵 of the same vector 𝑣 is related
to the new co-ordinate matrix [𝑣]𝐵′ of the same vector 𝑣 by the equation.

[𝑣]𝐵 = 𝑃[𝑣]𝐵′ where the columns of 𝑃 are the co-ordinate matrices for the basis vectors relative
to the old basis, that is, the column vectors of 𝑃 are [𝑢1′ ]𝐵 ,[𝑢2′ ]𝐵 … … [𝑢𝑛′ ]𝐵

The matrix 𝑃 is called the transition matrix from 𝐵 ′ to 𝐵.

Example:

Let 𝐵 = {𝑢1 , 𝑢2 }and 𝐵 ′ = {𝑢1′ , 𝑢2′ } be bases for ℝ2 where 𝑢1 = (1 0)

𝑢2 = (0 1) 𝑢1′ = (1, 1) 𝑢2′ = (2, 1)

a) Find the transition matrix from 𝐵 ′ to 𝐵

−3
b) Use (1) to find [𝑣]𝐵 if [𝑣]𝐵′ = [ ]
5

Solution

1 2
a) Find the co-ordinate matrices for the new basis vectors [𝑢1′ ]𝐵 = [ ] and [𝑢2′ ]𝐵 = [ ] so
1 1
1 2
the transition matrix from𝐵 ′ to 𝐵 is 𝑃 = [ ]
1 1
1 2 −3 7
b) [𝑣]𝐵 = [ ][ ] = [ ]
1 1 5 2
Theorem 1.6.0

If 𝑃 is a transition matrix from a basis to 𝐵 ′ to a basis 𝐵, then for every vector 𝑣, the following
relationships hold

12
 2020
edition

[𝑣]𝐵 = 𝑃[𝑣]𝐵′

[𝑣]𝐵′ = 𝑃−1 [𝑣]𝐵

If 𝑃 is the transition matrix from one orthogonal basis to another orthogonal basis for an inner
product space then

𝑃−1 = 𝑃𝑡

Example 1.

cos 𝜃 − sin 𝜃 −1 cos 𝜃 sin 𝜃

𝑃=[ ]𝑃 = [ ] = 𝑃𝑡
sin 𝜃 cos 𝜃 − sin 𝜃 cos 𝜃

1.6.1 Orthogonal Matrices

Definition1.6.0 : A square matrix 𝐴 with the property 𝑨−𝟏 = 𝑨𝒕 is said to be an orthogonal

matrix. Matrix 𝐴 is orthogonal 𝑨𝑨𝒕 = 𝑨𝒕 𝑨 = 𝑰.

Theorem: The following are equivalent for an 𝑛 x 𝑛matrix 𝐴.

a) 𝐴 is orthogonal
b) The row vectors form an orthogonal set in ℝ𝑛 with Euclidean Inner Product
c) The column vectors of 𝐴 form an orthogonal set in ℝ𝑛 with the Euclidean Inner Product.

1 1
0
√2 √2
𝐴= 0 0 1
1 1
− 0
[√2 √2 ]

Row vectors are

1 1 1 1
𝑟1 = ( , , 0) 𝑟2 = (0, 0, 1)𝑟3 = ( , , 0)
√2 √2 √2 √2
Relative to the Euclidean Inner Product, we have

‖𝑣1 ‖ = ‖𝑣2 ‖ = ‖𝑣3 ‖ = 1and𝑟1 . 𝑟2 = 𝑟2 . 𝑟3 = 𝑟1 . 𝑟3 = 0

13
 2020
edition

So the row vectors of 𝐴 form an orthogonal set i𝑛 ℝ𝑛 Thus𝑨 is orthogonal.

1 1
0
√2 √2
𝐴−1 = 𝐴𝑡 = 1 1
0
√2 √2
[0 1 −0]

An equation of the form 𝒙 = 𝑷𝒙𝟏 or 𝒙𝟏 = 𝑷−𝟏 𝒙 where 𝑃 is an orthogonal matrix is called an

orthogonal co-ordinate transformation.

1.7 ORTHOGONAL PROJECTIONS

In ℝ2 or ℝ3 with Euclidean Inner Product.

If 𝑊 is a line or plane through the origin, then each vector u in the space can be expressed as a
sum

𝒖 = 𝒘𝟏 + 𝒘𝟐 where𝒘𝟏 is in 𝑾 and 𝒘𝟐 is perpendicular to 𝑾.

𝑢 𝑤2
𝑢
𝑤2

𝑊
𝑊 𝑤1
𝑤1

Projection theorem 1.7.0:- If 𝑊 is a Finite – dimensional subspace of an Inner Product Space V,

then every vector 𝒖 in 𝑉 can be expressed in exactly one way as

𝒖 = 𝒘𝟏 + 𝒘𝟐 where𝒘𝟏 is in 𝑊 and 𝒘𝟐 is orthogonal to 𝑊.

14
 2020
edition

The vector 𝒘𝟏 is called the orthogonal projection of 𝒖 on 𝑊 and is denoted by 𝒑𝒓𝒐𝒋𝒘 𝒖.

The vector 𝒘𝟐 = 𝒖 − 𝒑𝒓𝒐𝒋𝒘 𝒖 is called the component of 𝒖 orthogonal to 𝑊.

The following theorems provide formulas for calculating orthogonal projections.

Theorem3.7.1 : Let 𝑊 be a finite – dimensional subspace of an Inner Product Space 𝑉,

a) If {𝒗𝟏 , 𝒗𝟐 , … … 𝒗𝒏 } is an orthogonal basis, for 𝑊, and 𝒖 is any vector in 𝑉 then

𝒑𝒓𝒐𝒋𝒘 𝒖 = 〈𝒖, 𝒗𝟏 〉𝒗𝟏 + 〈𝒖,𝒗𝟐 )𝒗𝟐 + … … + 〈𝒖, 𝒗𝒓 )𝒗𝒓

b) If {𝒗𝟏 , 𝒗𝟐 , … … 𝒗𝒏 } is an orthogonal basis for 𝑊, and 𝒖 is any vector in 𝑉 then

〈𝒖, 𝒗𝟏 〉 〈𝒖,𝒗𝟐 ) 〈𝒖, 𝒗𝒏 〉
𝒑𝒓𝒐𝒋𝒘 𝒖 = 𝟐
𝑽𝟏 + 𝟐
𝒗𝟐 + … … +
‖𝑽𝟏 ‖ ‖𝑽𝟐 ‖ ‖𝒗𝒏 ‖

Example

Let ℝ3 have the Euclidean Inner Product, and let 𝑊 be the subspace spanned by the
4 3
orthonormal vectors 𝑣1 = (0 1 0) and 𝑣2 = (− 5 , 0, 5)

Find the

i) The orthogonal project of (1 1 1) on 𝑊 is

ii) Component of 𝑢 orthogonal to 𝑊.

Solution

i) 𝑝𝑟𝑜𝑗𝑤 𝑢 = 〈𝑢, 𝑣1 〉𝑣1 + 〈𝑢2 , 𝑣2 〉𝑣2

1 4 3
= 1(0 1 0) + − 5 (− 5 , 0, 5)
4 −3 4 3
= (0 1 0) + (25 , 0, 25 ) = (25 , 1, − 25)

4 −3
ii) 𝑤2 = 𝑢 − 𝑝𝑟𝑜𝑗𝑤 𝑢 = (1 1 1) − (25 , 1, 25 )
21 28
= ( , 0, )
25 25
Theorem 3.7.2: Every non zero finite – dimensional Inner product space has an orthogonal
basis.

15
 2020
edition

Proof: Let 𝑣 be a non-zero finite – dimensional inner product space.

Let {𝑢1 𝑢2 … … 𝑢𝑛 } be any basis for 𝑉, we have to show that 𝑉 has an orthogonal basis, then
we normalise the vectors in the orthogonal basis. To do this we follow some steps.

STEP 1: Let 𝑣1 = 𝑢1
〈𝑣1 𝑢1 〉
STEP 2: Obtain a vector 𝑣2 = 𝑢2 − 𝑝𝑟𝑜𝑗𝑤 𝑢 = 𝑢2 − ‖𝑣1 ‖
𝑣1

STEP 3: To construct a vector 𝑣3 that is orthogonal to both 𝑣1 and 𝑣2 , compute the component
of 𝑢3 orthogonal to the space 𝑤2 spanned by 𝑣1 and 𝑣2 .

〈𝑢3 , 𝑣1 〉 〈𝑢3 , 𝑣2 〉
𝑣3 = 𝑢3 − 𝑝𝑟𝑜𝑗𝑤 𝑢 = 𝑢3 − 2
𝑣1 −
‖𝑣1 ‖ ‖𝑣2 ‖2

STEP 4: To determine a vector 𝑣4 that is orthogonal to 𝑣1 𝑣2 and 𝑣3 compute the component of

𝑢4 orthogonal to the space 𝑤3 spanned by 𝑣1 𝑣2 and 𝑣3 .

⟨𝑢3 , 𝑣1 ) 〈𝑢4 𝑣2 〉
𝑣4 = 𝑢4 − 𝑝𝑟𝑜𝑗𝑤3 𝑢4 = 𝑢4 − 𝑣1 −
‖𝑣‖2 ‖𝑣2 ‖2

Continuing this way, we will obtain after 𝑛 – steps an orthogonal set of vectors {𝑣1 , 𝑣2 , … … 𝑣𝑛 }.

Since 𝑉 is 𝑛 – dimensional and every orthogonal set is linearly independent the set
{𝑣1 , 𝑣2 , … … 𝑣𝑛 }will be an orthogonal basis for 𝑉.

The above step by step construction for converting an arbitrary basis into an orthogonal basis is
called the Gram – Schmidt process.

Example

Find an orthogonal basis for the subspace

𝑤 = Span {𝑢1 𝑢2 … … 𝑢𝑛 } of ℝ3 where

𝑢1 = (1 1 1)𝑢2 = (0 1 1)and𝑢3 = (0 0 1)

Solution
𝑣1 = 𝑢1 = (1 1 1)
〈𝑢2 𝑣1 〉
𝑣2 = 𝑢2 − 𝑝𝑟𝑜𝑗𝑤1 𝑢2 = 𝑣
‖𝑣1 ‖2 1

2 2 1 1
= (0 1 1) − (1 1 1) = (− , , )
3 3 3 3

16
 2020
edition

〈𝑢3 𝑣1 〉 〈𝑢3 𝑣2 〉
𝑣3 = 𝑢3 − 𝑝𝑟𝑜𝑗𝑤2 𝑢3 = 𝑢3 − 𝑣1 −
‖𝑣1 ‖2 ‖𝑣2 ‖2

1
1 2 1 1
= (0 0 1) − 3 (1 1 1) − 32 (− 3 , 3 , 3)
3

1 1
= (0, − , )
2 2

Thus
2 1 1 1 1
𝑣1 = (1 1 1) 𝑣2 = (− 3 , 3 , 3) 𝑣3 = (0, − 2 , 2) 𝑖𝑠 𝑡ℎ𝑒 𝑜𝑟𝑡ℎ𝑜𝑔𝑜𝑛𝑎𝑙 𝑏𝑎𝑠𝑖𝑠.

Normalise

Norms are

√6 1
‖𝑣1 ‖ = √3 ‖𝑣2 ‖ = ‖𝑣3 ‖ =
9 √2

𝑣1 (1 1 1) 1 1 1
∴ = =( , , )
‖𝑣1 ‖ √3 √3 √3 √3

2 1 1
𝑣2 (− 3 , 3 , 3) 2 1 1
= = (− , , )
‖𝑣2 ‖ √6 √6 √6 √6
9
1 1
𝑣3 (0, − 2 , 2) 1 1
= = (0, − , )
‖𝑣3 ‖ 1 √2 √2
√2

17
 2020
edition

ACTIV TY 1
1. 𝐿𝑒𝑡 𝑢 = (𝑢1 , 𝑢2 ) 𝑣 = (𝑣1 , 𝑣2 ) 𝑆ℎ𝑜𝑤 𝑡ℎ𝑎𝑡 𝑡ℎ𝑒 𝑓𝑜𝑙𝑙𝑜𝑤𝑖𝑛𝑔 𝑎𝑟𝑒 𝑖𝑛𝑛𝑒𝑟 𝑝𝑟𝑜𝑑𝑢𝑐𝑡𝑠

𝑜𝑛 𝑅 2 𝑏𝑦 𝑣𝑒𝑟𝑖𝑓𝑦𝑖𝑛𝑔 𝑡ℎ𝑎𝑡 𝑡ℎ𝑒 𝑖𝑛𝑛𝑒𝑟 𝑝𝑟𝑜𝑑𝑢𝑐𝑡 𝑎𝑥𝑖𝑜𝑚𝑠 ℎ𝑜𝑙𝑑.

𝑎) 〈𝑢, 𝑣〉 = 3𝑢1 𝑣1 + 5𝑢2 𝑣2 𝑏) 〈𝑢, 𝑣〉 = 4𝑢1 𝑣1 + 𝑢2 𝑣1 + 𝑢1 𝑣2 + 4𝑢2 𝑣2

2. 𝐿𝑒𝑡 𝑅 3 ℎ𝑎𝑣𝑒 𝑡ℎ𝑒 𝐸𝑢𝑐𝑙𝑖𝑑𝑒𝑎𝑛 𝑖𝑛𝑛𝑒𝑟 𝑝𝑟𝑜𝑑𝑢𝑐𝑡. 𝐹𝑜𝑟 𝑤ℎ𝑖𝑐ℎ 𝑣𝑎𝑙𝑢𝑒𝑠 𝑜𝑓 𝑘 𝑎𝑟𝑒 𝒖 𝑎𝑛𝑑 𝒗
𝑜𝑟𝑡ℎ𝑜𝑔𝑜𝑛𝑎𝑙.
𝑎) 𝑢 = (2,1,3) 𝑣 = (1,7, 𝑘) 𝑏) 𝑢 = (𝑘, 𝑘, 1) 𝑣 = (𝑘, 5,6)

3 𝐿𝑒𝑡 𝑅 4 ℎ𝑎𝑣𝑒 𝑡ℎ𝑒 𝐸𝑢𝑐𝑙𝑖𝑑𝑒𝑎𝑛 𝑖𝑛𝑛𝑒𝑟 𝑝𝑟𝑜𝑑𝑢𝑐𝑡. 𝑈𝑠𝑒 𝑡ℎ𝑒 𝐺𝑟𝑎𝑚 𝑠𝑐ℎ𝑚𝑖𝑑𝑡 𝑝𝑟𝑜𝑐𝑒𝑠𝑠 𝑡𝑜 𝑡𝑟𝑎𝑛𝑠𝑓𝑜𝑟𝑚 𝑡ℎ𝑒

𝑏𝑎𝑠𝑖𝑠 {𝑢1 , 𝑢2 , 𝑢3 , 𝑢4 } 𝑖𝑛𝑡𝑜 𝑎𝑛 𝑜𝑟𝑡ℎ𝑜𝑛𝑜𝑟𝑚𝑎𝑙 𝑏𝑎𝑠𝑖𝑠.

ANSWERS TO SOME SELECTED QUESTIONS

2a) 𝑘 = −3 𝑏) 𝑘 = 2, 𝑘 = −3
2 1 5 1 2 1 1 2 2 1 1 2 3
3. (0, , , 0) , ( ,− , , 0) , ( , ,− ,− ),( , ,− , )
√5 √5 √30 √30 √30 √10 √10 √10 √10 √15 √15 √15 √15

18
 2020
edition

UNIT 2: UNIT STRUCTURE

Unit 2: Eigenvalues and Eigenvectors
2.1 Learning Outcomes
2.2 Introduction
2.3 Eigenvalues and Eigenvectors
2.4 Eigenvalues of the power of a matrix
Activity 2
Answers to selected questions

19
 2020
edition

UNIT 2: EIGENVALUES AND EIGENVECTORS

You are welcome to unit one of this module. This unit will introduce you to the concept of
eigenvalues and eigenvectors or bases for eigenspaces.

2.1 LEARNING OUTCOMES

By the end of this unit, you should be able to

• Define an eigenvalue and eigenvector

• Find eigenvalues and their corresponding bases for eigenspaces

20
 2020
edition

• Find eigenvalues of Linear Operator.

2.2 INTRODUCTION
In this section, we will seek to understand eigenvalues and eigen vectors. Let us now define
eigenvalues and eigenvectors.

2.3 EIGEN VALUES AND EIGEN VECTORS

Definitions: If 𝐴 is an 𝑛 × 𝑛 matrix, then a nonzero vector 𝑥 in ℝ𝑛 is called an Eigen vector of

𝐴 if 𝐴𝑥 is a scalar multiple of 𝑥, that is 𝐴𝑥 = 𝜆𝑥 some scalar 𝜆.

The scalar 𝜆 is called an Eigen value of 𝐴, and 𝑥 is said to be an Eigen vector of 𝐴 corresponding
to 𝜆.

− The set of all such vectors is a sub space of a vector space called the Eigen space of 𝜆.
− The terms characteristics value and characteristic vector or proper value and proper
vector are frequently used instead of Eigen value and Eigen vectors.

FINDING EIGEN VALUES

− To find the Eigen values of an 𝑛 × 𝑛 matrix 𝐴, we rewrite;

𝐴𝑥 = 𝜆𝑥 as 𝐴𝑥 = 𝜆𝐼𝑥 or (𝜆𝐼 − 𝐴) = 0.
− For 𝜆 to be an Eigen value, there must be a non-zero solution of this equation.
− Equation 1 will have a non-zero solution if and only if det(𝜆𝐼 − 𝐴) = 0 → this is called
the characteristic equation of 𝐴.

Where expanded, the determinant det(𝜆𝐼 − 𝐴) is a polynomial in 𝜆 called the characteristic

polynomial of 𝐴.

EXAMPLE

Find the Eigen values of

3 2
1. 𝐴 = [ ]
−1 0

21
 2020
edition

1 0 3 2 𝜆−3 −2
Solution:- 𝜆𝐼 − 𝐴 = 𝜆 [ ]−[ ]=[ ]
0 1 −1 0 1 𝜆

Characteristic polynomial of 𝐴 is 𝜆2 − 3𝜆 + 2

Characteristic Equation of 𝐴 is 𝜆2 − 3𝜆 + 2 = 0

Solutions are 𝜆 = 1 and 𝜆 = 2. are Eigen values of 𝐴.

2. Find the Eigen values of the matrix

−2 −1
𝐴=[ ]
5 2
𝜆+2 1
=[ ]
−5 𝜆 − 2
= 𝜆2 + 1

𝜆2 + 1 = 0 solutions are imaginary numbers 𝜆 = 𝑖 and 𝜆 = −𝑖. Since we are assuming

that all our scalars are real numbers ∴ 𝐴 has no Eigen values.

0 1 0 1 0 0 0 1 0
3. 𝐴 = [0 0 1 ] − 𝜆 [0 1 0 ] [0 0 1] =
4 −17 8 0 0 1 4 −17 8

𝜆 −1 0
(𝜆𝐼 ) = [0 𝜆 −1 ] = 𝜆3 − 𝜆2 + 17𝜆 − 4 = 0
4 −17 𝜆−8

⇒ (𝜆 − 4)(𝜆2 − 4𝜆 + 1 = 0)
(𝜆 − 4) = 0 ⇒ 𝜆 = 4
(𝜆2 − 4𝜆 + 1) = 0 ⇒ 𝜆 = 2 + √3 and 𝜆 = 2 − √3

If 𝐴 is an 𝑛 × 𝑛 matrix and 𝜆 is a real number, then the following are equivalent;

a) 𝜆 is an Eigen value of 𝐴
b) The system of equations (𝜆𝐼 − 𝐴)𝑥 = 0 has nontrivial solutions.
c) There is a nonzero vector 𝑥 in ℝ𝑛 such that 𝐴𝑥 = 𝜆𝑥.
d) 𝜆 is a solution of the characteristic equation 𝑑𝑒𝑡(𝜆𝐼 − 𝐴) = 0

22
 2020
edition

FINDING BASES FOR EIGEN SPACE

The Eigen vectors of 𝐴 corresponding to an Eigen value 𝜆, are non zero vectors 𝑥 that satisfy
𝐴𝑥 = 𝜆𝑥.

Equivalently, the Eigen vectors corresponding to 𝜆 are the nonzero vectors in the solution space
of (𝜆𝐼 − 𝐴)𝑥 = 0. We call this solution space the Eigen space of a corresponding to 𝜆.

Example 4. Find bases for the Eigen spaces of

0 0 −2
𝐴 = [1 2 1]
1 0 3

Solution:- characteristic equation of 𝐴 is 𝜆3 − 5𝜆2 + 8𝜆 − 4 = 0 or (𝜆 − 1)(𝜆 − 2)2 = 0.

Eigen values of 𝐴 are 𝜆 = 1 and 𝜆 = 2. So there are two Eigen spaces of 𝐴.

By definition;

𝑥1
𝑥
𝑥 = [ 2 ] is an Eigen vector of 𝐴 corresponding to 𝜆 𝑖𝑓𝑓 𝑥 is a nontrivial solution of (𝜆𝐼 − 𝐴)𝑥 =
𝑥3
0

That is of

𝜆 0 2 𝑥1 0
𝑥
[−1 𝜆 − 1 −1 ] [ 2 ] = [0] (1 )
−1 0 𝜆 − 3 𝑥3 0

If 𝜆 = 2. Then (1) becomes,

2 0 2 𝑥1 0
𝑥
[−1 0 −1] [ 2 ] = [0]
−1 0 −1 𝑥3 0

23
 2020
edition

Solving the system we get

𝑥1 = −𝑠 𝑥2 = 𝑡 𝑥3 = 𝑠

So Eigen vectors of 𝐴 corresponding to 𝜆 = 2 are non zero vectors of the form

−𝑠 −𝑠 0 −1 0
𝑡 0
𝑥 = [ ] = [ ] + [ 𝑡 ] = 𝑠 [ 0 ] + 𝑡 [1]
𝑠 𝑠 0 1 0

−1 0
Which are [ 0 ] and [1] which form the Eigen space when 𝜆 = 2
1 0

1 0 2 𝑥1 0
If 𝜆 = 1 [−1 −1 −1] [𝑥2 ] = [0]
−1 0 −2 𝑥3 0

Solving the system we get

𝑥1 = −2𝑠 𝑥2 = 𝑠 𝑥3 = 𝑠

So Eigen vectors corresponding to 𝜆 = 1 are non zero vectors of the form

−2𝑠 −2 −2
[ 𝑠 ] = 𝑠 [ 1 ] and so [ 1 ] is a basis for the Eigen space corresponding to 𝜆 = 1
𝑠 1 1

2.4 EIGENVALUES OF THE POWERS OF A MATRIX

EIGENVALUES OF TRIANGULAR MATRICES AND THE POWERS OF A MATRIX.

Theorem: If 𝐴 is an 𝑛 x 𝑛 triangular matrix (upper, lower or diagonal) then the Eigen values of 𝐴
are the entries on the main diagonal of 𝐴.

24
 2020
edition

E.g.

1
0 0
2
1 2 1 2 1
𝐴 = −2 3
0 𝜆 = 2 , 3 and − 4 are the Eigen values.
1
[5 −8 − 4]

Once the Eigen values and Eigen vectors of a matrix 𝐴 are found, it is easy to find the Eigen values
and Eigen vectors of any positive integer power of 𝐴.

e.g. if 𝜆 is an Eigen value of 𝐴 and 𝑥 is a corresponding Eigen vector then

𝐴2 𝑥 = 𝐴(𝐴 𝑥) = 𝐴(𝜆𝑥) = 𝜆(𝐴 𝑥) = 𝜆(𝜆𝑥) = 𝜆2 𝑥

This shows that 𝜆2 is an Eigen value of 𝐴2 and 𝑥 is the corresponding Eigen vector.

Theorem: If 𝐾 is a positive integer, 𝜆 is an Eigen value of a matrix 𝐴, and 𝑥 is a corresponding

Eigen vector, then 𝜆𝐾 is an Eigen value of 𝐴𝐾 and 𝑥 is a corresponding Eigen vector.

Example:

0 0 −2
𝐴 = [1 2 1]
1 0 3

𝜆 = 2 and 𝜆 = 1 both 𝜆 = 27 = 128 and 𝜆 = 17 = 1 are Eigen values of 𝐴7

The Eigen vectors that we found in example 4 are also Eigen vectors of 𝐴7 .

EIGEN VALUES OF A LINEAR OPERATOR

Eigen values and Eigenvectors can be defined for linear operations. A scalar 𝜆 is called an Eigen
linear operator 𝑇: 𝑉 → 𝑉 if there is a nonzero vector 𝑥 in 𝑉 𝑠𝑡 𝑇𝑥 = 𝜆𝑥

The linear operator 𝑥 is called an Eigenvector of 𝑇 corresponding to 𝜆 is equivalently the Eigen

vectors of 𝑇 corresponding to 𝜆 are the non zero vectors in the Kernel of 𝜆𝐼 − 𝑇. This kernel is
called the Eigen space of 𝑇 corresponding to 𝜆.

25
 2020
edition

So we have that

If 𝑉 is a finite – dimensional vector space and 𝐵 is any basis for 𝑉, then

1) The Eigen values of 𝑇 are the same as the Eigen values of [𝑇]𝐵
2) A vector 𝑥 is an Eigen vector of 𝑇 corresponding to 𝜆 if and only if its co-ordinate matrix
[𝑥]𝐵 is an Eigen vector of [𝑇]𝐵 corresponding to 𝜆.

Find the Eigen values and bases for the Eigen spaces of the linear operator 𝑇: 𝑃2 → 𝑃2 defined by

𝑇(𝑎 + 𝑏𝑥 + 𝑐𝑥 2 ) = −2𝑐 + (𝑎 + 2𝑏 + 𝑐)𝑥 + (𝑎 + 3𝑐)𝑥 2

Solution: Matrix of 𝑇 to the standard basis 𝐵 = {1, 𝑥, 𝑥 2 } is

0 0 −2
[𝑇]𝐵 = [1 2 1]
1 0 3

Eigen values of 𝑇 are 𝜆 = 1 and 𝜆 = 2

Eigen space of [𝑇]𝐵 corresponding to 𝜆 = 2 has the basis {𝑢1 𝑢2 }
−1 0
Where 𝑢1 = [ 0 ] 𝑢2 = [1]
1 0

Eigen space for [𝑇]𝐵 corresponding to 𝜆 = 1 has the basis {𝑢3 } where

−2
𝑢3 = [ 1 ]
1

The matrices 𝑢1 , 𝑢2 and 𝑢3 are the co-ordinating matrices relative to 𝐵 of

𝑃1 = −1 + 𝑥 2 , 𝑃2 = 𝑥 𝑃3 = −2 + 𝑥 + 𝑥 2

Thus the Eigenspace of 𝑇 corresponding to 𝜆 = 2 has the basis

{𝑃1 𝑃2 } = {−1 + 𝑥 2 , 𝑥}

Eigenspace 𝜆 = 1 has the basis

26
 2020
edition

{𝑃3 } = {−2 + 𝑥 + 𝑥 2 }

ACTIVITY 2

1. 𝐹𝑜𝑟 𝑒𝑎𝑐ℎ 𝑜𝑓 𝑡ℎ𝑒 𝑔𝑖𝑣𝑒𝑛 𝑚𝑎𝑡𝑟𝑖𝑐𝑒𝑠 𝑓𝑖𝑛𝑑 𝑡ℎ𝑒 𝑒𝑖𝑔𝑒𝑛𝑣𝑎𝑙𝑢𝑒𝑠 𝑎𝑛𝑑 𝑒𝑖𝑔𝑒𝑛𝑣𝑒𝑐𝑡𝑜𝑟𝑠.

−3 1 −1 4 1 −1
𝐴 = [−7 5 −1] 𝑖𝑖. 𝐴 = [2 5 −2]
−6 6 −2 1 1 2

3 0 0
𝑖𝑖𝑖. 𝐴 = [0 2 −5] 𝑖𝑣. 𝐴 = [ ]
0 1 −2
2. Show that similar matrices have the same characteristic equation.
3. Suppose 𝜆 𝑖𝑠 𝑎𝑛 𝑒𝑖𝑔𝑒𝑛𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑎𝑛 𝑖𝑛𝑣𝑒𝑟𝑡𝑖𝑏𝑙𝑒 𝑜𝑝𝑒𝑟𝑎𝑡𝑜𝑟 𝑇.
𝑆ℎ𝑜𝑤 𝑡ℎ𝑎𝑡 𝜆−1 𝑖𝑠 𝑎𝑛 𝑒𝑖𝑔𝑒𝑛𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑇 −1

𝟒. 𝐿𝑒𝑡 𝐴 𝑎𝑛𝑑 𝐵 𝑏𝑒 𝑠𝑞𝑢𝑎𝑟𝑒 𝑚𝑎𝑡𝑟𝑖𝑐𝑒𝑠. 𝑆ℎ𝑜𝑤 𝑡ℎ𝑎𝑡 𝐴𝐵 𝑎𝑛𝑑 𝐵𝐴 ℎ𝑎𝑣𝑒 𝑡ℎ𝑒 𝑠𝑎𝑚𝑒 𝑒𝑖𝑔𝑒𝑛𝑣𝑎𝑙𝑢𝑒𝑠.

𝟓. 𝑆ℎ𝑜𝑤 𝑡ℎ𝑎𝑡 0 𝑖𝑠 𝑎𝑛 𝑒𝑖𝑔𝑒𝑛𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑇 𝑖𝑓 𝑎𝑛𝑑 𝑜𝑛𝑙𝑦 𝑖𝑓 𝑇 𝑖𝑠 𝑠𝑖𝑛𝑔𝑢𝑙𝑎𝑟.

27
 2020
edition

UNIT 3: UNIT STRUCTURE

Unit 3: Matrix Diagonalisation
3.1 Learning Outcomes

3.2 Matrix Diagonalisation

3.3 Orthogonal Diagonalisation

3.4 Norm and Distance in C

3.5 Unitary, Normal and Hermitian Matrices

3.6 Unitary Diagonalisation

3.7 Bilinear, Hermitian and Quadratic forms

3.8 Quadratic Forms

3.9 Diagonalising Quadratic forms- Conic section

28
 2020
edition

3.10 Conic section

Readings

UNIT 3: MATRIX DIAGONALISATION

You are welcome to unit two of this module. This unit will introduce you to a concept of Matrix
Diagonalisation.

3.1 LEARNING OUTCOMES

By the end of this unit, you should be able to

• Define Matrix Diagonalisation

• Understand and follow the procedure for Matrix Diagonalisation
• Define Unitary diagonalization and solve problems involving it.

3.2 MATRIX DIAGONALISATION

Definition: A square matrix 𝐴 is called diagonalizable if there is an invertible matrix 𝑃 which

is said to diagonalise 𝐴.
Theorem: If 𝐴 is an 𝑛 × 𝑛 matrix, then the following are equivalent

(i) 𝐴 is diagonalizable
(ii) 𝐴 has a linearly independent Eigenvector

We have that the Eigen vector problem and diagonalisation problem are the same.

Also Similarity is the same with diagonalisation.

Procedure for Diagonalising a Matrix.

To diagonalise an 𝑛 × 𝑛 matrix 𝐴 we follow the following steps;

29
 2020
edition

(i) Find 𝑛 linearly independent Eigen vectors of 𝐴 say 𝑃1 , 𝑃2 … … 𝑃𝑛 .

(ii) Form the matrix 𝑃 having 𝑃1 , 𝑃2 … … 𝑃𝑛 as its column vectors.
(iii) The matrix 𝑃−1 𝐴𝑃 where 𝜆𝑖 is the Eigen value corresponding to 𝑃𝑖 = 1,2 … … 𝑛

Example:

Find the invertible matrix 𝑃 such that 𝑃 −1 𝐴𝑃 is diagonal if

1 −3 3
𝐴=[ 3 −5 3]
−6 6 4

𝜆 − 1 −3 3
[ 3 𝜆+5 3 ]
−6 6 𝜆−4

Roots are −2 and 4 = eigenvalues

Characteristic equations = (𝑡 + 2)2 (𝑡 − 4)

For 𝜆 − 2 we get
1 3 1
−2 −2 1 −3 3 1 1 1 −2 0 0
2
−1
Then 𝑃 𝐴𝑃 = ( 1 −1 0 ) ( 3 −5 3 ) (1 0 1) = ( 0 −2 0)
1 1 1
−2 6 −6 4 0 −1 2 0 0 4
2 2

3.3 ORTHOGONAL DIAGONALIZATION

Definition: A square matrix 𝐴 is called orthogonaly diagonalizable if there is an orthogonal matrix
𝑃 is said to orthogonally diagonalise 𝐴.
Example;

30
 2020
edition

1 4 5 1 4 5
𝑡
If 𝐴 = [ 4 −3 0] then 𝐴 = [4 −3 0] = 𝐴
−5 0 7 5 0 7
Orthogonal diagonalization problem can only be solved when 𝐴 satisfies a special
condition namely 𝐴 = 𝐴𝑡

Definition: A square matrix 𝐴 is called symmetric if 𝐴 = 𝐴𝑡

1 4 5 1 4 5
e.g. 𝐴 = [4 −3 0] then 𝐴𝑡 = [4 −3 0] = 𝐴
5 0 7 5 0 7
Conditions for Orthogonal diagonalizability
1) Which matrices are orthogonally diagonalizable?
2) How do we find an orthogonal matrix to carry out the diagonalization?
The following theorem addresses the first question.
1) If 𝐴 is an 𝑛 × 𝑛 matrix, then the following are equivalent.
a) 𝐴 is orthogonally diagonalizable.
b) 𝐴 has an orthonormal set of 𝑛 eigenvectors.
c) 𝐴 is symmetric
2) Theorem: If 𝐴 is a symmetric matrix, then eigenvectors from different eigenspaces are
orthogonal.
As a consequence of this theorem we obtain the following procedure for orthonogally
diagonalizing a symmetric matrix.
Step 1. Find a basis for each eigenspace 𝐴
Step 2. Apply the Gram – Schimdt process to each of these bases to obtain an
orthonormal basis for each eigenspaces.
Step 3. Form the matrix 𝑃 whose columns are the basis vectors constructed in step 2.
This matrix orthogonally diagonalises 𝐴.
This theorem ensures that eigen vectors from distinct eigenspaces are orthogonal which the
application of the Gram – Schmidt process, ensures that the eigenvectors obtained within the
same eigenspace are orthonormal.
Thus the entire set of eigenvectors obtained by this procedure is orthornormal.

31
 2020
edition

Example
Find an orthogonal matrix 𝑃 that diagonolises
4 2 2
𝐴 = [2 4 2]
2 2 4
Solution
Characteristics equation of 𝐴 is
𝜆 − 4 −2 −2
det(𝜆𝐼 − 𝐴) = det [ −2 𝜆 − 4 −2 ] = (𝜆 − 2)2 (𝜆 − 8) = 0
−2 −2 𝜆 − 4
Eigenvalues are

𝜆 = 2 and 𝜆 = 8

For 𝜆 = 2

−2 −2 −2 𝑥 0 −1 −1
[−2 −2 −2] [𝑦] = [0] so we get 𝑢1 = [ 1 ] and 𝑢2 = [ 0 ]
−2 −2 −2 𝑧 0 0 1

Which form the basis for the eigen space corresponding to 𝜆 = 2

Applying the Gram – Schmidt process to 𝑢1 , 𝑢2 and 𝑢3 , we get the orthogonal eigenvectors
1 1
𝑉2 = (− 2 , 2 , 1) 𝑉1 = (−1 1 0)

1
−
1 − ‖𝑢1 ‖ = √2 ‖𝑉1 ‖
√6
√2 1
𝑃1 = [ 1 ] and 𝑃2 = − √6 6
√2 2
‖𝑢2 ‖ = √ ‖𝑉2 ‖
2
0 [ √6 ]

For 𝜆 = 8 we have

−2 −2 𝑥 0
𝑦
[−2 −4 −2] [ ] = [0]
−2 −2 −4 𝑧 0

So we get

32
 2020
edition

1
𝑢3 = [1] as a basis. Apply Gram – Schmidt process to 𝑢3
1 𝑉 = (1 1 1) 3

Yield
‖𝑉3 ‖ = √3
1
√3
1
𝑃1 = √3
1
[− √3]

Using 𝑃1 𝑃2 and 𝑃3 as column vectors we obtain

1 1 1
− −
√2 √6 √3
1 1 1
𝑃= − which orthogonally diagonalises 𝐴.
√2 √6 √3
2 1
[ 0 √6 √3]

As a check verify that 𝑃−1 𝐴𝑃 is a diagonal matrix

Some properties of symmetric matrices

a) The characteristic equation of a symmetric matrix 𝐴 has only real roots.

b) If an eigenvalue λ of symmetric matrix 𝐴 is repeated 𝐾 times as a root of the
characteristic equation, then the eigen space corresponding to the eigen value λ is 𝐾 −
dimensional.

e.g.

3 1 0 0 0
1 3 0 0 0
𝐴 0 0 2 1 1
0 0 1 2 1
[0 0 1 1 2]

Characteristic equation is (𝜆 − 4)2 (𝜆 − 1)2 (𝜆 − 2) = 0 so the eigen values are 𝜆 = 4, 𝜆 = 1,

and 𝜆 = 2 where 𝜆 = 4 and 𝜆 = 1 are repeated twice and 𝜆 = 2 occurs once.

33
 2020
edition

Thus the Eigen spaces corresponding to 𝜆 = 4 and 𝜆 = 1 are 2 – dimensional and the Eigen
space corresponding to 𝜆 = 1 is 1 – dimensional.

Complex Inner Product.

Definition: If 𝑢 = {𝑢1 𝑢2 … … 𝑢𝑛 } and 𝑣 = {𝑣1 𝑣2 … … 𝑣𝑛 } are vectors in ℂ𝑛 , then their complex

Euclidean Inner Product 〈𝑢. 𝑣〉 is defined by 〈𝑢. 𝑣〉 = 𝑢1 ̅̅̅
𝑣1 + 𝑢2 ̅̅̅
𝑢2 + … … 𝑢𝑛 ̅̅̅
𝑣𝑛 where 𝑣̅1 𝑣̅2 𝑣̅𝑛
are the conjugates of 𝑣1 𝑣2 … … 𝑣𝑛

Example

Let 𝑢 = (−𝑖, 21 + 3𝑖) and 𝑣 = (1, −𝑖, 01 + 3𝑖)

̅̅̅̅̅̅̅̅̅
𝑢. 𝑣 = (−𝑖)(1 ̅̅̅̅ + (1 + 3𝑖)(1
− 𝑖) + 2(0) ̅̅̅̅̅̅̅̅̅̅̅
+ 3𝑖)

= (𝑖)(1 + 2) + 2(0) + (1 + 3𝑖)(1 − 3𝑖)

= −𝑖 − 𝑖 2 + 1 − 9𝑖 2 = 11 − 𝑖

Properties of the complex Euclidean Inner Product.

Theorem: If 𝑢 𝑣 and 𝑤 are vectors. In ℂ𝑛 ,and 𝐾 is any complex number then

i) 𝑢. 𝑣 = ̅̅̅̅̅
𝑣. 𝑢
ii) (𝑢 + 𝑣). 𝑤 = 𝑢. 𝑤 + 𝑣. 𝑤
iii) (𝐾𝑢). 𝑣 = 𝐾(𝑢. 𝑣). ̅ (𝑢. 𝑣)
……….. 𝑢. (𝐾𝑣) = 𝐾
iv) 𝑣. 𝑣 ≥ 0 further 𝑣. 𝑣 = 0 if and only if 𝑣 = 0

3.4 NORM AND DISTANCE IN ℂ𝒏 .

1
‖𝑢‖ = (𝑢. 𝑢)2 = √|𝑢1 |2 + |𝑢2 |2 + … … + |𝑢𝑛 |2

𝑑(𝑢𝑣) = √

𝑑(𝑢𝑣) = ‖𝑢. 𝑣‖ = √|𝑢1 − 𝑣1 |2 + |𝑢2 − 𝑣2 |2 + … … + |𝑢𝑛 − 𝑣𝑛 |2

Example; If 𝑢 = (𝑖, 1 + 𝑖, 3) and 𝑣 = (1 − 𝑖, 2, 4𝑖) then

34
 2020
edition

‖𝑢‖ = √|𝑢1 − 𝑣1 |2 + |𝑢2 − 𝑣2 |2

= √1 + 2 + 9

= √12 √3 × √4 = 2√3

𝑑(𝑢 𝑣) = √|𝑖 − (1 − 𝑖)2 +|(1 + 𝑖) − 2|2 + |3 − 4𝑖|2

= √5 + 2 + 25

= √32

= 4√2

A complex vector space with an inner product is called a complex inner product unitary space.

3.5 UNITARY, NORMAL AND HERMITIAN MATRICES.

Unitary Matrices.

If 𝐴 is a matrix with complex entries, then the conjugate transpose of 𝐴 denoted by 𝐴∗ is defined
by 𝐴∗ = 𝐴̅ 𝑡 . Where 𝐴̅ is the matrix whose entries are the compex conjugates of the
corresponding entries in 𝐴 and 𝐴̅ 𝑡 is the transpose of 𝐴̅.

1+𝑖 −𝑖 0 1−𝑖 𝑖 0
Example: If 𝐴 = [ ] then 𝐴̅ = [ ]
2 3 − 2𝑖 𝑖 2 3 + 2𝑖 −𝑖

1−𝑖 2
So 𝐴∗ = 𝐴̅ 𝑡 = [ 𝑖 3 + 2𝑖 ]
0 −𝑖

The basic properties of the conjugate transpose operation are similar to those of a transpose.

Theorem: If 𝐴 and 𝐵 are matrices with complex entries and 𝐾 is any complex number, then

35
 2020
edition

a) (𝐴∗ )∗ = 𝐴
b) (𝐴 + 𝐵)∗ = 𝐴∗ + 𝐵 ∗
̅ 𝐴∗
c) (𝐾𝐴)∗ = 𝐾
d) (𝐴𝐵)∗ = 𝐵 ∗ 𝐴∗

UNITARY MATRICES

Definition: A square matrix 𝐴 with complex entries is called unitary if 𝐴−1 = 𝐴∗

Theorem: If 𝐴 is an 𝑛 × 𝑛 matrix with complex entries, then the following are equivalent.

a) 𝐴 is unitary 𝐴−1 = 𝐴∗ 𝐴∗ = 𝐴̅ 𝑡
b) The row vectors of 𝐴 form an orthogonal set in 𝐶 𝑛 with the Euclidean Inner product.
c) The column vectors of 𝐴 form an orthogonal set in 𝐶 𝑛 with the Euclidean Inner product.

Example: The matrix

1+𝑖 1+𝑖
2 2 1+𝑖 1+𝑖
𝐴= [1−𝑖 −1+𝑖
] (1) has row vectors 𝑟1 = ( , )
2 2
2 2

1−𝑖 −1+𝑖
𝑟2 = ( , )
2 2

Relative to the Euclidean Inner Product on ℂ𝑛𝑛 we have

1+𝑖 2 1+𝑖 2 1 1
‖𝑟1 ‖ = √| | +| | = √2 + 2 = 1.
2 2

1−𝑖 2 −1+𝑖 2 1 1
‖𝑟2 ‖ = √| | +| | = √2 + 2 = 1.
2 2

and

1+𝑖 ̅̅̅̅̅̅̅
1−2 1+𝑖 ̅̅̅̅̅̅̅̅
−1+𝑖
𝑟1 . 𝑟2 = ( )( 2 ) + ( 2 )( 2 )
2

1+𝑖 1+𝑖 1+𝑖 1−𝑖

=( )( )+( ) (− )
2 2 2 2

36
 2020
edition

= 𝑖 − 𝑖 = 0. So the row vectors form an orthonormal set in ℂ𝑛 . Thus 𝐴 is unitary and

1−𝑖 1+𝑖
2 2
𝐴−1 = 𝐴∗ = [1−𝑖 −1−𝑖
] (2) matrix 2 is the inverse on matrix 1. Verify by taking 𝐴𝐴∗ = 𝐴∗ 𝐴 =
2 2

3.6 UNITARY DIAGONALISATION

A square matrix 𝐴 with complex entries is called Unitarily diagonalisable if there is a unitary 𝑃
such that 𝑃−1 𝐴𝑃(= 𝑃∗ 𝐴𝑃) is diagonal. The matrix 𝑃 is said to unitarily diagonalise 𝐴 .

1) Which matrices are unitarily diagonalizable.

2) How do we find a unitary matrix 𝑃 to carry out the diagonalization .

Recall that symmetric matrices played a fundamental role in the problem of orthogonally
diagonalising a matrix with real entries for complex entries, Hermitian matrices are like
symmetric matrices.

Hermitian Matrices
Definition: A square matrix 𝐴 with complex entries is called Hermitian. If 𝐴 = 𝐴∗

1 𝑖 1+𝑖
[ −𝑖 5 2 − 𝑖]
1−𝑖 2+𝑖 3

By inspection we can know a Hermitian matrix

i) Entries on main diagonal are real numbers

ii) Minor image of each entry across the main diagonal is its complex conjugate.

Example;
−2 1−𝑖 −1 + 𝑖
𝐴 = [ 1+𝑖 0 3 ]
−1 − 𝑖 3 5

37
 2020
edition

−2 1+𝑖 −1 − 𝑖
̅
𝐴 = [ 1−𝑖 0 3 ]
−1 + 𝑖 3 5

−2 1−𝑖 −1 + 𝑖
∗ −1
𝐴 =𝐴 =[ 1+𝑖 0 3 ]
−1 − 𝑖 3 5

Just as the real symmetric matrices are orthogonally diagonalizable, Hermitian matrices are also
unitarily diagonalizable, the Hermitian matrices do not constitute the entire class of unitary
matrices.

Normal Matrices
Definition: A Square 𝐴 with complex entries is called normal if

𝐴𝐴∗ = 𝐴∗ 𝐴

Every Hermitian matrix 𝐴 is normal since 𝐴𝐴∗ = 𝐴𝐴 = 𝐴∗ 𝐴

Every unitary matrix 𝐴 is normal since 𝐴𝐴∗ = 𝐼 = 𝐴∗ 𝐴

Diagonalisation Procedure

1) Find the basis for each Eigen space of 𝐴 .

2) Apply the Gram Schmidt to process to each of these bases to obtain an orthonormal for
each eigen space
3) Form the matrix 𝑃 whose columns are the basis vectors constructed in 2. This matrix
unitarily diagonalises 𝐴.

Example

Find a matrix 𝑃 that unitarily diagonalises

2 1+𝑖
𝐴=( )
1−𝑖 3

Solution: Characteristic polynomial of 𝐴 is

38
 2020
edition

𝜆−2 −1 − 𝑖
det | | = (𝜆 − 2)(𝜆 − 3) − 2
−1 + 𝑖 𝜆−3

= 𝜆2 − 5𝜆 + 4 = (𝜆 − 1)(𝜆 − 4) = 0

Eigen values are 𝜆 − 1 and 𝜆 − 4.

EIGEN VECTORS

𝜆−2 −1 − 𝑖 −1 −1 − 𝑖 𝑥 0
𝜆−1[ ]=[ ] [𝑦 ] = [ ]
−1 + 𝑖 𝜆−3 −1 + 𝑖 −2 0

Solve by Gauss – Jordan elimantion and get

−1 − 𝑖
𝑥−1−𝑖 𝑦=1 𝑢1 = ( ) = ‖𝑢1 ‖ = √3
1

Normalise

−1−𝑖
√3
𝑉1 = ( 1 )
√3

𝜆−4

2 −1 − 𝑖 𝑥 0
[ ] [𝑦] = [ ]
−1 + 𝑖 1 0

1+𝑖
Solve and get 𝑥1 = ( ) 𝑦=1
2

1+𝑖
𝑢2 = ( 2 )
1

Normalise and get

1+𝑖 2
𝑉2 = ( , )
√6 √6

39
 2020
edition

1+𝑖 2

𝑃= √3 √6
1 2
( √3 √6)

3.7 BILINEAR, HERMITIAN AND QUADRATIC FORM

1) Bilinear Form
Let 𝑉 be a vector space of finite dimension over a field 𝐾, a Bilinear form on 𝑉 is a
transformation (mapping).
𝑇: 𝑉 × 𝑉 → 𝐾 which satisfies
(i) 𝑇(𝑎𝑢1 , +𝑏𝑢2 , 𝑣) = 𝑎𝑇(𝑢1 , 𝑣) + 𝑏𝑇(𝑢2 , 𝑣) 𝑇 is linear in the 1st variable
(ii) 𝑇(𝑢1 , 𝑎𝑣1 + 𝑏𝑣2 ) = 𝑎𝑇(𝑢, 𝑣1 ) + 𝑏𝑇(𝑢, 𝑣2 ) 𝑎, 𝑏 ∈ 𝐾 and all 𝑢𝑖 , 𝑣𝑖 ∈ 𝑉. ( 𝑇 is linear
in the second variable)

Examples

Let 𝑓 be a dot product on ℝ𝑛 ,

𝑓(𝑢, 𝑣) = 𝑢, 𝑣 = 𝑎1 𝑏1 + 𝑎2 𝑏2 + … … + 𝑎𝑛 𝑏𝑛 . Where 𝑢 = (𝑎𝑖 ) and 𝑣 = (𝑏𝑖 )

Then 𝑓 is bilinear form on ℝ𝑛

The bilinear form is denoted by 𝐵(𝑣)

Definition: A matrix 𝐵 is said to be congruent to a matrix 𝐴 if there exits an invertible matrix

𝑃 such that 𝐵 = 𝑃𝑡 𝐴𝑃.

Matrixes representing the same bilinear form are congruent. Congruent matrices have the same
rank because 𝑃 and 𝑃𝑡 are invertible.

Definition: The rank of a bilinear form 𝑓 on 𝑉, written rank (𝑓), is defined to be the rank of any
matrix representation.

40
 2020
edition

We say that 𝑓 is degenerate if rank (𝑓) < dim 𝑉 and 𝑓 is non degenerate if rank (𝑓) < dim 𝑉.
and 𝑓 is non degenerate if rank (𝑓) = dim 𝑉.

Symmetric Bilinear Forms

A bilinear form 𝑓 on 𝑉 is said to be symmetric if 𝑓(𝑢 𝑣) = 𝑓(𝑣, 𝑢). 𝑢, 𝑣 ∈ 𝑉

If 𝐴 is a matrix representation of 𝑓, we write

𝑓(𝜒, 𝑦) = 𝜒 𝑡 𝐴𝑦 = (𝜒 𝑡 𝐴𝑦)𝑡 = 𝑦 𝑡 𝐴𝑡 𝜒.

We use the fact that 𝜒 𝑡 𝐴𝑦 is a scalar and therefore equals its transpose.

We have 𝑦 𝑡 𝐴𝑡 𝑥 = 𝑓(𝑥 𝑦) = 𝑓(𝑦 𝑥) = 𝑦 𝑡 𝐴𝑥

Theorem: Let 𝐴 be a symmetric matrix over 𝐾. Then there exists an invertible matrix 𝑃 such
that 𝑃𝑡 𝐴𝑃 is diagonal. i.e. 𝐴 is congruent matrix.

Example

Let

1 2 −3
𝐴=( 2 5 −4) be a symmetric matrix. Find a diagonal matrix 𝑃𝑡 𝐴𝑃
−3 −4 8

Solution

Form the block matrix (𝐴, 𝐼)

1 2 −3 1 0 0 −2𝐶
→
1 +𝐶2 1 2 −3 1 0 0
(𝐴, 𝐼) = ( 2 5 −4|0 1 0) −2𝑅1+𝐶2 0 1 2 | −2 1 0 =
→
−3 −4 8 0 0 1 3𝑅1+𝑅3 0 2 −1 3 0 1
3𝐶1 +𝐶3
→
−2𝐶1 + 𝐶2 1 0 0 1 0 0
(0 1 2 |−2 1 0)
3𝐶1 + 𝐶3 0 2 −1 3 0 1

41
 2020
edition

1 0 0 1 0 0 1 0 0 1 0 0
−2𝐶1 + 𝐶3 (0 1 2 |−2 1 0) −2𝐶1 + 𝐶3 (0 1 2 |−2 1 0)
−2𝑅1 + 𝑅3 0 2 −5 7 −2 1 0 2 −5 7 −2 1

𝐴 has been diagonalised

1 −2 7 1 0 0
𝑡
𝑃 = (0 1 −2) 𝑃 𝐴𝑃 = (0 1 0 )
0 0 1 0 0 −5

3.8 QUADRATIC FORMS

A quadratic form in two variables 𝑥 and 𝑦 is defined to be an expression that can be written as

𝑎𝑥 2 + 2𝑏𝑥𝑦 + 𝐶𝑦 2 (1) where 𝑎 = 1 𝑏 = 1 𝐶 = 1

𝑎𝑥 2 + 2𝑏𝑥𝑦 + 𝐶𝑦 2 can be written in matrix form as

[𝑥 𝑦] [𝑎 𝑏 𝑥
] [ ] (2)
𝑏 𝑐 𝑦

This matrix (2) is symmetric, the diagonal entries are the coefficients of squared terms and the
entries off the main diagonal are each half the coefficient of the product term 𝑥𝑦 . 𝑥𝑦 is called
the cross product term.

𝑦] [2 3 𝑥
2𝑥 2 + 6𝑥𝑦 − 7𝑦 2 = [𝑥 ][ ]
3 −7 𝑦

Quadratic forms are not limited to the variables.

Definition: A quadratic form in the 𝑛 - variables 𝑥1 𝑥2 … … 𝑥𝑛 is an expression that can be written

as

𝑥1
𝑥2
[𝑥1 𝑥2 … … 𝑥𝑛 ]𝐴 [ ⋮ ] where 𝐴 is a symmetric 𝑛 × 𝑛 matrix
𝑥𝑛

42
 2020
edition

𝑥1
𝑥2
If we let 𝑥 = [ ⋮ ] then (1) can be written as 𝑥 𝑡 𝐴𝑥.
𝑥𝑛

When we denote a quadratic form by 𝑥 𝑡 𝐴𝑥, it will be understood that 𝐴 is symmetric i.e. 𝐴 =
𝐴𝑡 .

Example; quadratic form in 𝑥1 𝑥2 … … 𝑥3 of

1 2 −1 𝑥1
𝑥12 + 7𝑥22 − 3𝑥32 + 4𝑥1 𝑥2 − 2𝑥1 𝑥3 + 6𝑥2 𝑥3 = [𝑥1 𝑥2 𝑥3 ] [ 2 7 3 ] [𝑥2 ]
−1 3 −3 𝑥3

Note the coefficients of the squared terms appear on the main diagonal of 3 × 3 matrix,
coefficients of the cross product term are each split in half and appear in the off – diagonal
positions as follows.

Coefficients 𝑥1 𝑥2 appear in 𝑎12 , 𝑎21 , 𝑥1 𝑥3 in 𝑎13 , 𝑎31 and 𝑥2 𝑥3 in 𝑎23 and 𝑎32

PSOITIVE DEFINITE MATRICES AND QUADRATIC FORMS

𝑦] [2 3 𝑥
2𝑥 2 + 6𝑥𝑦 − 7𝑦 2 = [𝑥 ][ ]
3 −7 𝑦

Quadratic forms are not limited to two variables. A general quadratic form is

Definition: A quadratic form in the 𝑛 - variables 𝑥1 𝑥2 … … 𝑥𝑛 is an expression that can be written

𝑥1
𝑥2
as [𝑥1 𝑥2 … … 𝑥3 ]𝐴 [ ⋮ ] where 𝐴 is a symmetric 𝑛 × 𝑛 matrix (1)
𝑥𝑛

𝑥1
𝑥2
If we let 𝑥 = [ ⋮ ] then (1) can be written as 𝑥 𝑡 𝐴𝑥
𝑥𝑛

When we denote a quadratic form by 𝑥 𝑡 𝐴𝑥, it will be understood that 𝐴 is symmetric i.e. 𝐴 =
𝐴𝑡 .

43
 2020
edition

POSITIVE DEFINITE MATRICES AND QUADRATIC FORMS

Definition: A quadratic form 𝑥 𝑡 𝐴𝑥 is called positive definite if 𝑥 𝑡 𝐴𝑥 > 0 ∀ 𝑥 ≠ 0.

A symmetric matrix 𝐴 is called a positive definite matrix if 𝑥 𝑡 𝐴𝑥 is a positive definite quadratic

form.

Theorem: A symmetric 𝐴 is positive definite if and only if all the eigen values of 𝐴 are positive.

- We can also determine whether a symmetric matrix is positive definite without finding its
eigen values.

e.g.
𝑎11 𝑎12 ⋯ 𝑎1𝑛
𝑎21 𝑎22 ⋯ 𝑎2𝑛
𝐴=[ ⋮ ] is a square matrix , then the principle sub matrices of 𝐴 are
⋮
𝑎𝑛1 𝑎𝑛2 ⋯ 𝑎𝑛𝑛
the submatrices formed from the first 𝑟 rows and 𝑟 columns of 𝐴 for 𝑟 = 1,2 … … . 𝑛.
e.g.
𝑎11 𝑎12 𝑎13
𝑎11 𝑎12
𝐴1 = [𝑎11 ] 𝐴2 = [𝑎 𝑎22 ] 𝐴 = [𝑎21 𝑎22 𝑎23 ]
21
𝑎31 𝑎32 𝑎33

Theorem

A symmetric matrix 𝐴 is positive definite if and only if the determinant of every principal sub
matrix is positive

e.g.

2 −1 −3
𝐴 = [−1 2 4]
−3 4 9

|𝐴11 | = |2| = 2

44
 2020
edition

𝑎11 𝑎12 2 −1 −3
2 −1
𝐴2 = |𝑎 𝑎22 | = |−1 2 | = 3 𝐴3 = |−1 2 4 |=1
21
−3 4 9

∴ 𝐴 is positive definite

Remark; A symmetric matrix 𝐴 and the quadratic form 𝑥 𝑡 𝐴𝑥 are called positive semi definite if
𝑥 𝑡 𝐴𝑥 ≥ 0 ∀ 𝑥.

Negative definite if 𝑥 𝑡 𝐴𝑥 < 0 ∀ 𝑥 ≠ 0.

Negative semi definite if 𝑥 𝑡 𝐴𝑥 ≤ 0 ∀ 𝑥.

Indefinite if 𝑥 𝑡 𝐴𝑥 has both positive and negative values.

3.9 DIAGONALIZING QUADRATIC FORMS CONIC SECTION

Theorem: Let 𝑥 𝑡 𝐴𝑥 be a quadratic form in 𝑛 variables 𝑥1 𝑥2 … … 𝑥𝑛 where 𝐴 is symmetric. If 𝑃

orthogonally diagonalises 𝐴 and if the new variables 𝑦1 𝑦2 … … 𝑦𝑛 are defined by the equation
𝑥 = 𝑃𝑦, then substituting this equation in 𝑥 𝑡 𝐴𝑥 yields;

𝑥 𝑡 𝐴𝑥 = 𝑦 𝑡 𝐷𝑦 = 𝜆1 𝑦12 + 𝜆2 𝑦22 + … … + 𝜆𝑛 𝑦𝑛𝑛 where 𝜆1 𝜆2 … … 𝜆𝑛 are the eigen values of 𝐴

and

𝜆1 0 ⋯ 0
0 𝜆2 ⋯ 0
𝐷 = 𝑃𝑡 𝐴𝑃 = [ ]
⋮ ⋮
0 0 ⋯ 𝜆𝑛

The matrix 𝑃 is said to orthogonally diagonalise the quadratic form or reduce the quadratic
form to a sum of squares .

Example

45
 2020
edition

Find a change of variables that will reduce the quadratic form 𝑥12 − 𝑥32 − 4𝑥1 𝑥2 + 4𝑥2 𝑥3 to a
sum of squares and express the quadratic form in terms of the new variables.

Solution

Quadratic form can be written as

1 −2 0 𝑥1
[𝑥1 𝑥2 𝑥3 ] [−2 0 2 ] [𝑥2 ]
0 2 −1 𝑥3

Char Equation is

𝜆−1 2 0
| 2 𝜆 −2 | = 𝜆3 − 9𝜆 = 𝜆(𝜆 + 3)(𝜆 − 3) = 0
0 −2 𝜆 + 1

Eigen values are 𝜆 = 0 𝜆 = −3 𝜆=3

Find the eigen vectors and normalise

2 1 2
−3 −3
3
1 2 2
𝜆=0 𝜆 = −3 −3 𝜆=3
3 3
2 2 1
[3] [ 3 ] [ 3 ]

Thus the substitution 𝑥 = 𝑃𝑦 that eliminates cross – product terms is

2 1 2
2 1 2
−3 −3 𝑥1 = 𝑦1 − 𝑦2 − 𝑦3
𝑥1 3 𝑦1 3 3 3
1 2 2
[𝑥2 ] = 3
−3 3
[𝑦2 ] or 1 2 2
𝑥3 2 2 1 𝑦3 𝑥2 = 𝑦1 − 𝑦2 + 𝑦3
3 3 3
[3 3 3 ]
2 2 1
𝑥3 = 𝑦1 + 𝑦2 + 𝑦3
3 3 3

The new quadratic form is

0 0 0 𝑦1
[𝑦1 𝑦2 𝑦3 ] [0 −3 0] [𝑦2 ] or −3𝑦22 + 3𝑦32
0 0 3 𝑦3

46
 2020
edition

3.10 CONIC SECTION

We shall apply our work quadratic forms to the study of equations of the form

𝑎𝑥 2 + 2𝑏𝑥𝑦 + 𝐶𝑦 2 + 𝑑𝑥 + 𝑒𝑦 + 𝑓 = 0

Where 𝑎, 𝑏 … … 𝑓 are real numbers and at least one of the numbers 𝑎, 𝑏, 𝑐 is not zero. An
equation of this type is called a quadratic equation in 𝑥 and 𝑦 and 𝑎𝑥 2 + 2𝑏𝑥𝑦 + 𝐶𝑦 2 is called
the associated quadratic form.

ELIMINATING THE CROSS PRODUCT TERM

𝑎𝑥 2 + 2𝑏𝑥𝑦 + 𝐶𝑦 2 + 𝑑𝑥 + 𝑒𝑦 + 𝑓 = 0 in matrix form can written as

[𝑥 , 𝑦 ] [𝑎 𝑏 𝑥 𝑥
] [𝑦] + [𝑑, 𝑒] [𝑦] + 𝑓 = 0 or 𝑥 𝑡 𝐴𝑥 + 𝐾𝑥 + 𝑓 = 0 where
𝑏 𝑐

𝑥 𝑎 𝑏
𝑥 = [𝑦] 𝐴=[ ] 𝐾 = [𝑑 𝑒]
𝑏 𝑐

Rotate the 𝑥𝑦 co – ordinate axes so that the equation of the conic in the new 𝑥1 𝑦1

co – ordinates system has no cross product term;

𝑃11 𝑃12
(i) Find matrix 𝑃 = [ ] that orthogonally diagonalises the quadratic form 𝑥 𝑡 𝐴𝑥.
𝑃21 𝑃22
(ii) Interchange the columns of 𝑃 if necessary to make det(𝑃) = 1. This assures that the
orthogonal transformation is a rotation.
(iii) Substitute (ii) into 𝑥 𝑡 𝐴𝑥 + 𝐾𝑥 + 𝑓 = 0

Theorem: Principal Axes theorem for ℝ2 .

47
 2020
edition

Let 𝑎𝑥 2 + 2𝑏𝑥𝑦 + 𝐶𝑦 2 + 𝑑𝑥 + 𝑒𝑦 + 𝑓 = 0 be the equation of a conic 𝐶, and let 𝑥 𝑡 𝐴𝑥 = 𝑎𝑥 2 +

2𝑏𝑥𝑦 + 𝐶𝑦 2 be the associated quadratic form. Then the co – ordinate axes can be rotated so that
the equation for 𝐶 in the new 𝑥1 𝑦 1 co – ordinates system has the form

𝜆1 𝑥12 + 𝜆2 𝑦 12 + 𝑑1 𝑥1 + 𝑒 1 𝑦 1 + 𝑓 = 0 where 𝜆1 and 𝜆2 are the eigen values of 𝐴.

The rotation can be accomplished by the substitution 𝑥 = 𝑃𝑥1 where 𝑃 orthogonally
diagonalises 𝑥 𝑡 𝐴𝑥 and det(𝑃) = 1.

Identify the conic whose equation is

24 6
5𝑥 2 − 6𝑥𝑦 + 5𝑦 2 − 𝑥+ 𝑦 + 56 = 0
√2 √2

Matrix form for this equation is

𝑥 𝑡 𝐴𝑥 + 𝐾𝑥 + 56 = 0

5 −3 24 8
𝐴=[ ] 𝐾 = [− √2 + ]
−3 5 √2

Eigen values

𝜆=2 𝜆=8

Eigen space

1 1 1
𝜆=2=[ ] −
1
normalise 𝑃 = √2 √2
1 1
−1
𝜆=8=[ ] [√2 √2 ]
1

𝑃 orthogonally diagonalises 𝑥 𝑡 𝐴𝑥

Substitute 𝑥 = 𝑃𝑥1 in 𝑥 𝑡 𝐴𝑥 + 𝐾𝑥 + 56 = 0

48
 2020
edition

= 𝑃(𝑥1 )𝑡 𝐴(𝑃𝑥1 ) + 𝐾(𝑃𝑥1 ) + 56 = 0

1 1
−
𝑡 2 0 24 8 √2 √2
𝑃 𝐴𝑃 = [ ] 𝐾𝑃 [− √2 + ][ 1 1 ] = [−16,32]
0 8 √2
√2 √2

2 2
2𝑥 1 + 8𝑦1 − 16𝑥1 + 32𝑦 1 + 56 = 0

2 2
𝑥 1 + 4𝑦1 − 8𝑥1 + 16𝑦1 + 28 = 0

Compute the square to complete the graph.

2 2
(𝑥 1 −4) (𝑦 1 +2)
+ =1
4 1

Let 𝑥11 = (𝑥1 − 4) 𝑦11 = (𝑦 11 + 2)

𝑥 11 𝑦 11
Equation becomes + = 1= ellipse in standard position with respect to 𝑥11 𝑦 11 co –
4 1

ordinate axes.

ACTIVITY 3
1. For matrices in Activity 4. Show if any are diagonalizable. If so find an invertible matrix P
such that 𝑃−1 𝐴𝑃 = 𝐼.

READINGS
Anton, H and Rorres, C.(1991) Elementary Linear Algebra. John
Wiley and Sons.
Lipschutz, S. (1987) Linear Algebra. McGraw-Hill Book Company.

49
 2020
edition

Kolman, B and Hill, D.R. (2008) Elementary Linear Algebra With

Applications. Pearson Education, Inc.
Poole, D. (2011) Linear Algebra: A Modern Introduction.
Brooks/Cole, Cengage Learning.

50
 2020
edition

51