Hebron University

Faculty of Graduate Studies

On Determinant Of Non-Square Matrices

By

Shireen Azmi Sabri AL-Dweak

Supervisors

Dr. Mahmoud Shalalfeh

Submitted In Partial Fulfillment of the Requirements For The Degree of

Master of Mathematics, Faculty of Graduate Studies, At Hebron University

At Hebron, Palestine.

2021
 On Determinant Of Non-Square Matrices

By

Shireen Azmi Sabri AL-Dweak

This Thesis was defended successfully on –January – 2021 and

approved by

Committee Members Signature

1. Dr. Mahmoud Shalalfeh Supervisor ………………..

( Hebron University)
2. Dr. Bassam Manasrah Internal Examiner ………………..
( Hebron University)
3. Dr. Iyad Alhriba External Examiner ………………..
( Palestine Polytechnic University)

i
 ‫اإلهداء‬
‫إلى من أشتاق إليه بكل جوارحي ‪ ...‬إلى أبي الذي فارقنا بجسده‪ ،‬ولكن روحه ما زالت تُرفرف في‬

‫سماء حياتي‪.‬‬

‫إلى نبع المحبة واإليثار والكرم ‪ ...‬إلى من أمدَّتني بالنصح واإلرشاد والدعاء ‪ ...‬أمي أطال هللا في‬

‫عمرها وأمدَّها بالصحة والعافية لتظل عونا لي‪.‬‬

‫إلى من لم تبخل بوقت أو جهد لمساعدتي يوما ما ‪ ....‬اختي سوسن‬

‫إلى إخوتي واخواتي ‪ ....‬سندي وعضدي ومشاطري أفراحي وأحزاني‪.‬‬

‫إلى جميع أهلي داخل الوطن الغالي وخارجه‬

‫إلى كل من دعا لي بالخير‬

‫أهديكم خالصة ُجهدي العلمي‬

‫‪ii‬‬
 Acknowledgments

First of all, I thank my God for all the blessing he bestowed on me and continues
to bestow on me.

I would like to express my gratitude and I thank my kind, sympathetic mother

who prays and asks my God to help me. I ask my God to accept her prayer which
will benefit me in this life and after death.

I would also like to express my gratitude and appreciation for my sister Sawsan
Dweak for her help, patience, encouragement, supporting this project.

My sincere thanks go to my advisor Dr. Mahmoud Shalalfeh who offered me

advice and assistance of every stage of my thesis including wise supervision and
constant encouragement. His fruitful suggestions were very helpful. I am indebted
to Dr. Mahmoud for his valuable and constructive comments which substantially
improved the final product and for the time and effort he put.

I would also like to acknowledge to all my teachers in Hebron University,

department of mathematics.

My thanks are extended to Dr. Bassam Manasrah and Dr. Iyad Alhribat for
serving in my committee.

I thank my family members for the support and encouragement and for their
effective cooperation and care in preparing this thesis.

Finally it is pleasure to record my thanks to all my friends, and to all employees

in wedad naser iddeen school for their spiritual support.

iii
 Declaration

The work provided in this thesis, unless otherwise referenced, is the result of the
researcher's work, and has not been submitted elsewhere for any other degree or
qualification.

Student's name: Shireen dweak

Signature:

Date :

iv
Contents

Abstract iiv

Preface 1
1 Preliminaries 3
1. Matrices and matrix operations 3
2. Properties of algebraic operations on matrices 4
3. Determinants of Matrices 5
4. Cofactor Expansion and Adjoint 7
5. The Inverse of a Matrix 8
6. Linear Systems 9
7. Gaussian elimination, Cramer's rule 11
8. Vectors in the Euclidean space 𝑅 𝑛 13
9. Rank and Nullity 14
2 Determinant 18

1. Determinant function 18
2. Minors and cofactors 21
3. Determinant Using Cofactor Expansion 23
4. Effect of elementary row operations on determinates 26
5. Radic's determinant 33
3 Properties for non-square determinant 42
1. Properties for determinant of a rectangular matrix. 42
2. How determinant is affected by operations on columns 51
4 Applications of Determinant to Area of Polygons in 𝑹 𝟐 58
1. Areas of certain polygons in connection with determinant of 58
rectangular matrices.
2. Some properties of the determinant and their geometric 63
interpretation
5 Inverse for non-square matrix 74
1. Right inverse or left inverse of a matrix 74
2. Properties for inverse and adjoint of non- square matrices 78
3. Pseudo inverse of non-square matrices 84

v
6 Applications to linear systems of equations 𝑨 𝑿 = 𝑩 87
1. Solving a linear system Using pseudo inverse 87
2. Using adjoint to solve a system of linear equations 91
3. Cramer's rule for nonsingular 𝑚 × 𝑛 matrices 94
4. Particular Cases and Examples 99
References 105

vi
 Abstract
We study the definition of the determinant of a non-square matrix, using
cofactor definition and Radic definition, and we proved that they are identical by
proving the uniqueness of the determinant function that satisfies the four
characterizing properties of determinant function.

We also study the connection between the area of any polygon in the Cartesian
plane and determinant function for 2 × 𝑛 matrices. We used several methods to
find the mathematical isotope and prove the properties for inverse and adjoint for
a matrix as well as solving systems of equations in several ways.

vii
 ‫الملخص‬

‫قمنا بدراسة تعريف محدد المصفوفة غير المربعة باستخدام تعريف العامل المساعد وتعريف رادك ‪،‬‬

‫وأثبتنا أنهما متطابقان من خالل إثبات أن كالً منهما يحقق الشروط االربع للدالة المحددة‪.‬‬

‫ضا العالقة بين مساحة أي مضلع في المستوى الديكارتي والدالة المحددة لمصفوفات من‬
‫درسنا أي ً‬

‫𝑛×‪2‬‬ ‫الرتبة‬

‫و أستخدمنا عدة طرق إليجاد النظير الرياضي وإثبات خصائص معكوس المصفوفة ومضاد‬

‫المصفوفة باإلضافة إلى حل أنظمة المعادالت الخطية بعدة طرق‪.‬‬

‫‪viii‬‬
 Preface

In the books of a linear algebra we studied the concept of matrix and its types,
and we studied the definition of the determinant of the square matrix, its
properties and its applications. Here we will study the definition of the
determinant of the non-square matrix, its verified properties and its applications in
finding the area of polygons and finding solutions to the system of linear
equations.

My thesis consists of six chapters. Each chapter is divided into sections. A

number like 2.1.3 indicates item (definition, theorem, corollary or lemma) number
3 in section 1 of chapter 2. Each chapter begins with a clear statement of the
pertinent definition and theorems together with illustrative and descriptive
material. At the end of this thesis we present a collection of references.

In chapter (1) we introduce the basic results and definitions which shall be
needed in the following chapters. The topics include results about matrices and
matrix operations, properties of algebraic operations on matrices, Determinants of
Matrices, The Inverse of a Matrix, Cofactor Expansion, Adjoint of Matrix, Linear
Systems, Reduced Row-Echelon Form, Gauss-Jordan reduction, Cramer's rule,
Rank and Nullity and Vectors in the Euclidean space 𝑅𝑛 . This chapter is
absolutely fundamental. The results have been stated without proofs, for theory
may be looked in any text book in linear algebra. A reader who is familiar with
these topics may this chapter and refer to it only when necessary.

Chapter (2) will be devoted to give a defined determinant of a non-square

matrix. we will start by introducing the define a determinant function of non-
square matrix in terms of characterizing properties that we want it to have. In
section (2) we define minors and cofactors. In section (3 and 4), we will study the
method for finding a determinant of a non-square matrix ( 𝑚 × 𝑛, 𝑚 ≤ 𝑛) using
cofactor expansion, also study the effect of elementary row operations on
determinant. In section (5) We will study the method for finding a determinant of
a non-square matrix ( 𝑚 × 𝑛, 𝑚 ≤ 𝑛) using Radic's definition. Finally, we can
proof the cofactor definition and Radic definition are determinant function, and
the cofactor definition and Radic definition are the same.

1
 Chapter (3) we study Radic definition for determinant of a rectangular matrix
in more detailed way. We present new identities for the determinant of a
rectangular matrix. We develop some important properties of this determinant.
We generalize several classical important determinant identities, and description
how the determinant is affected by operation on columns, such as interchanging
columns, reversing columns or decomposing a single column.

Chapter (4) we will study an application for determinants of non-square

matrices in calculating the area of polygons in 𝑅2 , and proof the area of a polygon
is the determinant function, the area equals the determinant (the cofactor
definition and Radic definition).

Chapter (5) we will study existence of inverses for non-square matrices. Also,
we compute an inverse of a rectangular matrix using solution of a linear system
and an adjoint of matrices. In section (2) we study some important properties for
inverse and adjoint of non-square matrices. In section (3) we discuss Pseudo
inverse method which gives an inverse of matrices.

Chapter (6) we will discuss some results concerning the solutions of a linear
system 𝐴𝑥 = 𝑏 using inverses as well as the pseudo-inverse and adjoint of a
rectangular 𝑚 × 𝑛 matrix 𝐴, and General solution theorem. In section (3) we
study want to generalize this method of Cramer's for an 𝑚 < 𝑛 system of linear
equations. Finally in section (4) we study shall consider some particular cases and
examples to illustrate the results of what we have done in the previous sections
especially in applying pseudo inverse to some certain examples.

2
 Chapter One

Preliminaries

This chapter contains some definitions and basic results about matrices and
matrix operations, properties of algebraic operations on matrices, Determinants of
Matrices, The Inverse of a Matrix, Cofactor Expansion, Adjoint of Matrix, Linear
Systems, Reduced Row-Echelon Form, Gauss-Jordan reduction, Cramer's rule,
Rank and Nullity and Vectors in the Euclidean space 𝑅𝑛 .

1.1 Matrices and matrix operations

Definition 1.1.1 [ 9 , 𝑝. 11 ]. An 𝑚 × 𝑛 matrix 𝐴 is a rectangular array of 𝑚 𝑛 real

(or complex) numbers arranged in 𝑚 horizontal rows and 𝑛 vertical columns:

𝑎11 𝑎12 … 𝑎1𝑛

𝑎21 𝑎22 … 𝑎2𝑛
𝐴=[ ⋮ ⋱ ⋮ ]
⋮ (1)
𝑎𝑚1 𝑎𝑚1 … 𝑎𝑚𝑛

The 𝑖𝑡ℎ row of 𝐴 is [𝑎𝑖1 𝑎𝑖2 𝑎𝑖3 ⋯ 𝑎𝑖𝑛 ] (1 ≤ 𝑖 ≤ 𝑚),

𝑎1𝑗
𝑎2𝑗
The 𝑗𝑡ℎ column of 𝐴 is 𝑎3𝑗 (1 ≤ 𝑗 ≤ 𝑛).
⋮
[𝑎𝑚𝑗 ]

We shall say that 𝐴 is 𝑚 by 𝑛 (written as 𝑚 × 𝑛). If 𝑚 = 𝑛, we say that 𝐴 is a

square matrix of order 𝑛, and the numbers 𝑎11 , 𝑎22 , … , 𝑎𝑛𝑛 form the main
diagonal of 𝐴. We refer to the number 𝑎𝑖𝑗 which is in the 𝑖𝑡ℎ row and 𝑗𝑡ℎ column
of 𝐴, as the 𝑖, 𝑗 𝑡ℎ element of 𝐴, or the (𝑖, 𝑗) entry of 𝐴, and we often write the
matrix as 𝐴 = (𝑎𝑖𝑗 ).

Definition 1.1.2 [ 9 , 𝑝. 16 ]. (The Transpose of a Matrix) If 𝐴 = (𝑎𝑖𝑗 ) is an

𝑚 × 𝑛 matrix, then the 𝑛 × 𝑚 matrix 𝐴𝑇 = (𝑎𝑇𝑖𝑗 ), where 𝑎𝑇𝑖𝑗 = 𝑎𝑗𝑖

( 1 ≤ 𝑖 ≤ 𝑚, 1 ≤ 𝑗 ≤ 𝑛), is called the transpose of 𝐴. Thus the transpose of 𝐴 is
obtained by interchanging the rows and columns of 𝐴. Before operations.

3
Definition 1.1.3 [ 2 , 𝑝. 27 ]. Two matrices are defined to be equal if they have the
same size and their corresponding entries are equal.

Definition 1.1.4 [ 9 , 𝑝. 12 ]. (Diagonal Matrix) a square matrix 𝐴 = (𝑎𝑖𝑗 ) for

which every term off the main diagonal is zero, that is, 𝑎𝑖𝑗 = 0 for 𝑖 ≠ 𝑗, is called
a diagonal matrix.

Definition 1.1.5 [ 2 , 𝑝. 14 ].
a. (Matrix Addition) If 𝐴 = (𝑎𝑖𝑗 ) and 𝐵 = (𝑏𝑖𝑗 ) are 𝑚 × 𝑛 matrices, then the
sum of 𝐴 and 𝐵 is the 𝑚 × 𝑛 matrix 𝐶 = (𝑐𝑖𝑗 ), defined by 𝑐𝑖𝑗 = 𝑎𝑖𝑗 + 𝑏𝑖𝑗

( 1 ≤ 𝑖 ≤ 𝑚, 1 ≤ 𝑗 ≤ 𝑛).

That is, 𝐶 is obtained by adding corresponding elements of 𝐴 and 𝐵.

b. (Scalar Multiplication) If 𝐴 = (𝑎𝑖𝑗 ) is an 𝑚 × 𝑛 matrix, and 𝑟 is a real

number, then the scalar multiple of 𝐴 by 𝑟 , 𝑟𝐴, is the 𝑚 × 𝑛 matrix 𝐵 = (𝑏𝑖𝑗 ),
where 𝑏𝑖𝑗 = 𝑟 𝑎𝑖𝑗 , ( 1 ≤ 𝑖 ≤ 𝑚, 1 ≤ 𝑗 ≤ 𝑛).
That is, 𝐵 is obtained by multiplying each element of 𝐴 by 𝑟.

c. (Matrix Multiplication) If 𝐴 = (𝑎𝑖𝑗 ) is an 𝑚 × 𝑝 matrix, and 𝐵 = (𝑏𝑖𝑗 ) is a

𝑝 × 𝑛 matrix, then the product of 𝐴 and 𝐵, denoted 𝐴 𝐵, is the 𝑚 × 𝑛 matrix
𝐶 = (𝑐𝑖𝑗 ), defined by
𝑐𝑖𝑗 = 𝑎𝑖1 𝑏1𝑗 + 𝑎𝑖2 𝑏2𝑗 + ⋯ + 𝑎𝑖𝑝 𝑏𝑝𝑗 = ∑𝑝𝑘=1 𝑎𝑖𝑘 𝑏𝑘𝑗 , (1 ≤ 𝑖 ≤ 𝑚, 1 ≤ 𝑗 ≤ 𝑛).
The following proprieties for operations on matrices will be stated without proof.

1.2 Properties of algebraic operations on matrices

Theorem 1.2.1 [ 9 , 𝑝. 35 ].
Let 𝐴 , 𝐵 , 𝐶 𝑎𝑛𝑑 𝐷 be an 𝑚 × 𝑛 matrices. and 𝑟 and 𝑠 are real numbers, then
(1) 𝐴 + 𝐵 = 𝐵 + 𝐴.
(2) 𝐴 + (𝐵 + 𝐶 ) = (𝐴 + 𝐵) + 𝐶 .
(3) There is a unique 𝑚 × 𝑛 matrix 𝑂 such that 𝐴 + 𝑂 = 𝐴 for any 𝑚 × 𝑛
matrix 𝐴. The matrix 𝑂 is called the 𝑚 × 𝑛 additive identity or zero matrix.
(4) For each 𝑚 × 𝑛 matrix 𝐴, there is a unique 𝑚 × 𝑛 matrix 𝐷 such that
𝐴 + 𝐷 = 𝑂. (1)

4
 We shall write 𝐷 as −𝐴, so that (1) can be written as
𝐴 + (−𝐴) = 𝑂.
The matrix – 𝐴 is called the additive inverse or the negative of 𝐴.
(5) If 𝐴 , 𝐵 and 𝐶 are of the appropriate sizes, then 𝐴(𝐵 𝐶 ) = (𝐴 𝐵) 𝐶 .
(6) If 𝐴 , 𝐵 and 𝐶 are of the appropriate sizes, then 𝐴(𝐵 + 𝐶 ) = 𝐴 𝐵 + 𝐴 𝐶 .
(7) If 𝐴 , 𝐵 and 𝐶 are of the appropriate sizes, then (𝐴 + 𝐵 )𝐶 = 𝐴 𝐶 + 𝐵 𝐶 .
(8) 𝑟(𝑠 𝐴) = (𝑟 𝑠 )𝐴.
(9)( 𝑟 + 𝑠 )𝐴 = 𝑟 𝐴 + 𝑠 𝐴.
(10) 𝑟( 𝐴 + 𝐵 ) = 𝑟 𝐴 + 𝑟 𝐵.
(11) 𝐴 ( 𝑟 𝐵 ) = 𝑟 ( 𝐴 𝐵 ) = ( 𝑟𝐴 )𝐵.
For a proof for these properties [9].

Theorem 1.2.2 [ 9 , 𝑝. 41 ]. (Properties of Transposing a matrix)

If 𝑟 is a scalar, 𝐴 and 𝐵 are matrices, then

𝑇
(a) (𝐴𝑇 ) = 𝐴.

(b) ( 𝐴 + 𝐵 )𝑇 = 𝐴𝑇 + 𝐵𝑇 and ( 𝐴 − 𝐵 )𝑇 = 𝐴𝑇 − 𝐵𝑇 .
(c) ( 𝐴 𝐵 )𝑇 = 𝐵𝑇 𝐴𝑇 .
(d) ( 𝑟 𝐴)𝑇 = 𝑟 𝐴𝑇 .
For a proof for these properties [9].

1.3 Determinants of Matrices

Definition 1.3.1 [8]. (Determinant function) A determinant function assigns to

each square matrix 𝐴 a scalar associated to the matrix, denoted by 𝑑𝑒𝑡(𝐴) or |𝐴|
such that:

(1) The determinant of an 𝑛 × 𝑛 identity matrix "𝐼" is 1. |𝐼| = 1.

(2) If the matrix 𝐵 is identical to the matrix 𝐴 except the entries in one of the
rows of 𝐵 are each equal to the corresponding entries of 𝐴 multiplied by the
same scalar 𝑐, then |𝐵| = 𝑐 |𝐴|.
(3) If the matrices 𝐴, 𝐵 and 𝐶 are identical except for the entries in one row, and
for that row an entry in 𝐴 is found by adding the corresponding entries in 𝐵
and 𝐶, then |𝐴| = |𝐵| + |𝐶|.

5
 (4) If the matrix 𝐵 is the result of exchanging two rows of 𝐴, then the
determinant of 𝐵 is the negation of the determinant of 𝐴. (|𝐵| = −|𝐴|)

Theorem 1.3.2 [ 8]. A determinant function has the following four properties.
(a) The determinant of any matrix with an entire row of 0′ 𝑠 is 0.
(b) The determinant of any matrix with two identical rows is 0.
(c) If one row of a matrix is a scalar multiple of another row, then its
determinant is 0.
(d) If a scalar multiple of one row of a matrix is added to another row, then the

resulting matrix has the same determinant as the original matrix.

Theorem 1.3.3 [ 8 ]. There is at most one determinant function.

Definition 1.3.4 [ 9 , 𝑝. 92 ]. Let 𝐴 = (𝑎𝑖𝑗 ) be an 𝑛 × 𝑛 matrix. The determinant

of 𝐴 (written det(𝐴) or |𝐴|) is defined by

det(𝐴) = |𝐴| = ∑(±) 𝑎1𝑗1 𝑎2𝑗2 … 𝑎𝑛𝑗𝑛 ,

where the summation ranges over all permutations 𝑗1 𝑗2 … 𝑗𝑛 of the set

𝑆 = {1,2, … , 𝑛}. The sign is taken as + or – according to whether the

permutation 𝑗1 𝑗2 … 𝑗𝑛 is even or odd.

Theorem 1.3.5 [ 9 , 𝑝. 95 ]. (Properties of Determinants of Matrices)

(a) If 𝐴 is a square matrix. then det(𝐴) = det(𝐴𝑇 )
(b) If matrix 𝐵 results from matrix 𝐴 by interchanging two rows (columns) of 𝐴,
then 𝑑𝑒𝑡(𝐵) = − 𝑑𝑒𝑡(𝐴)
(c) If two rows (columns) of 𝐴 are equal, then 𝑑𝑒𝑡(𝐴) = 0
(d) If 𝐵 is obtained from 𝐴 by multiplying a row (column) of 𝐴 by real number 𝑐,
then 𝑑𝑒𝑡(𝐵) = 𝑐 𝑑𝑒𝑡(𝐴).
(e) If 𝐵 = (𝑏𝑖𝑗 ) is obtained from 𝐴 = (𝑎𝑖𝑗 ) by adding to each element of the
𝑟𝑡ℎ row (column) of 𝐴 a corresponding element of the 𝑠𝑡ℎ row (column)
𝑟 ≠ 𝑠 of 𝐴, then 𝑑𝑒𝑡(𝐵) = 𝑑𝑒𝑡(𝐴).
(f) The determinant of a product of two matrices is the product of their
determinants, that is, 𝑑𝑒𝑡( 𝐴 𝐵) = 𝑑𝑒𝑡 (𝐴)𝑑𝑒𝑡(𝐵).

6
 (g) If 𝐴 has a row (column) consisting of all zeros, then 𝑑𝑒𝑡(𝐴) = 0
For proofs of these properties [9].

Theorem 1.3.6 [9 , 𝑝. 93 ].
1. Let 𝐴 , 𝐵 and 𝐶 be an 𝑛 × 𝑛 matrices that differ only in a single row, say the
𝑟𝑡ℎ row, and assume that the 𝑟𝑡ℎ row of 𝐶 can be obtained by adding
corresponding entries in the 𝑟𝑡ℎ rows of 𝐴 and 𝐵. Then
𝑑𝑒𝑡( 𝐶) = 𝑑𝑒𝑡(𝐴) + 𝑑𝑒𝑡(𝐵)
The same result holds for columns.

2. (Decomposing a column). If a column 𝐾 in a square matrix 𝐴 is a sum of two

columns (eg. 𝐾 = 𝐾1 + 𝐾2 ), then the determinant | 𝐴| is a sum of tow
determinants of matrices obtained from 𝐴 by replacing 𝐾 by 𝐾1 and 𝐾2
respectively.

1.4 Cofactor Expansion and Adjoint

Definition 1.4.1 [ 9 , 𝑝. 103 ]. (Minor and Cofactor) Let 𝐴 = (𝑎𝑖𝑗 ) be an 𝑛 × 𝑛

matrix. Let 𝑀𝑖𝑗 be the (𝑛 − 1) × (𝑛 − 1) submatrix of 𝐴 obtained by deleting the

𝑖𝑡ℎ row and 𝑗𝑡ℎ column of 𝐴. The determinant 𝑑𝑒𝑡(𝑀𝑖𝑗 ) is called the minor of 𝑎𝑖𝑗 .
The cofactor 𝐴𝑖𝑗 of 𝑎𝑖𝑗 is defined as 𝐴𝑖𝑗 = (−1)𝑖+𝑗 𝑑𝑒𝑡(𝑀𝑖𝑗 ).

Theorem 1.4.2 [ 9 , 𝑝. 104 ].

Let 𝐴 = (𝑎𝑖𝑗 ) be an 𝑛 × 𝑛 matrix. Then for each 1 ≤ 𝑖 ≤ 𝑛,
𝑑𝑒𝑡(𝐴) = 𝑎𝑖1 𝐴𝑖1 + 𝑎𝑖2 𝐴𝑖2 + ⋯ + 𝑎𝑖𝑛 𝐴𝑖𝑛
(expansion of 𝑑𝑒𝑡(𝐴) about the 𝑖𝑡ℎ row).
And for each 1 ≤ 𝑗 ≤ 𝑛,
𝑑𝑒𝑡(𝐴) = 𝑎1𝑗 𝐴1𝑗 + 𝑎2𝑗 𝐴2𝑗 + ⋯ + 𝑎𝑛𝑗 𝐴𝑛𝑗

(expansion of 𝑑𝑒𝑡(𝐴) about the 𝑗𝑡ℎ column).

Definition 1.4.3 [ 9 , 𝑝. 108 ]. (Adjoint of Matrix) Let 𝐴 = (𝑎𝑖𝑗 ) be an 𝑛 × 𝑛

matrix. Then 𝑛 × 𝑛 matrix 𝑎𝑑𝑗 𝐴, called the adjoint of 𝐴, is the matrix whose
(𝑖, 𝑗)𝑡ℎ element is the cofactor 𝐴𝑗𝑖 of 𝑎𝑗𝑖 . (The transpose of the matrix of
cofactors), thus

7
 𝐴11 𝐴21 ⋯ 𝐴𝑛1
𝐴 𝐴22 ⋯ 𝐴𝑛2
𝑎𝑑𝑗(𝐴) = [ 12 ].
⋮ ⋮ ⋱ ⋮
𝐴1𝑛 𝐴2𝑛 ⋯ 𝐴𝑛𝑛

Theorem 1.4.4 [ 9 , 𝑝. 108 ]. (Properties of the Adjoint)

(a) If 𝐴 = (𝑎𝑖𝑗 ) is an 𝑛 × 𝑛 matrix. Then 𝐴(𝑎𝑑𝑗𝐴) = (𝑎𝑑𝑗𝐴)𝐴 = 𝑑𝑒𝑡(𝐴)𝐼𝑛

(b) If 𝐴 = (𝑎𝑖𝑗 ) is invertible 𝑛 × 𝑛 matrix. then det(𝑎𝑑𝑗 𝐴) = det(𝐴)𝑛−1

(c) If 𝐴 = (𝑎𝑖𝑗 ) is invertible 𝑛 × 𝑛 matrix. then 𝑎𝑑𝑗(𝑎𝑑𝑗 𝐴) = (det 𝐴 )𝑛−2 𝐴
(d) 𝑎𝑑𝑗(𝐴 𝐵) = 𝑎𝑑𝑗(𝐵) 𝑎𝑑𝑗(𝐴)
𝑇
(e) (𝑎𝑑𝑗(𝐴 )) = 𝑎𝑑𝑗(𝐴𝑇 )
(f) 𝑎𝑑𝑗( 𝑘 𝐴 ) = 𝑘 𝑛−1 𝑎𝑑𝑗(𝐴), where 𝑘 is any scalar
For proofs of these properties [9].

1.5 The Inverse of a Matrix

Definition 1.5.1 [ 9 , 𝑝. 19 ]. (Inverse of a Matrix) An 𝑛 × 𝑛 matrix 𝐴 is called

non-singular (or invertible) if there exists an 𝑛 × 𝑛 matrix 𝐵 such that

𝐴 𝐵 = 𝐵𝐴 = 𝐼𝑛 .

The matrix 𝐵 is called an inverse of 𝐴. If there exists no such matrix 𝐵, then 𝐴 is called
singular (or noninvertible).

It is easy to show that an inverse of a matrix is unique, if it exists, and so it is legitimate

to say the inverse of 𝐴 and write it as 𝐴−1 , thus 𝐴 𝐴−1 = 𝐴−1 𝐴 = 𝐼𝑛 .

Theorem 1.5.2 [ 9 , 𝑝. 71 ]. (Properties of the Inverse of a matrix)

−1
(a) If 𝐴 is a nonsingular matrix, then 𝐴−1 is a nonsingular and (𝐴−1 ) =𝐴

(b) If 𝐴 and 𝐵 are nonsingular matrices, then 𝐴 𝐵 is nonsingular and

(𝐴 𝐵)−1 = 𝐵−1 𝐴−1
−1 𝑇
(c) If 𝐴 is a nonsingular matrix, then (𝐴𝑇 ) = (𝐴−1 )
1
(d) For any nonzero scalar 𝑘, then (𝑘𝐴)−1 = 𝑘 𝐴−1
For proofs of these properties [9].

8
Corollary 1.5.3 [ 9 , 𝑝. 100 ]. If 𝐴 is nonsingular, then 𝑑𝑒𝑡(𝐴) ≠ 0 and
1
det (𝐴−1 ) = det(𝐴)

Theorem 1.5.4 [ 9 , 𝑝. 100 ]. A square matrix 𝐴 is nonsingular if and only if

𝑑𝑒𝑡(𝐴) ≠ 0

Theorem 1.5.5 [ 9 , 𝑝. 107 ]. If 𝐴 = (𝑎𝑖𝑗 ) is an 𝑛 × 𝑛 matrix. Then

𝑎𝑖1 𝐴𝑘1 + 𝑎𝑖2 𝐴𝑘2 + ⋯ + 𝑎𝑖𝑛 𝐴𝑘𝑛 = 0 𝑓𝑜𝑟 𝑖 ≠ 𝑘

𝑎1𝑗 𝐴1𝑘 + 𝑎2𝑗 𝐴2𝑘 + ⋯ + 𝑎𝑛𝑗 𝐴𝑛𝑘 = 0 𝑓𝑜𝑟 𝑗 ≠ 𝑘

Corollary 1.5.6 [ 2 , 𝑝. 106 ] If 𝐴 is an 𝑛 × 𝑛 matrix and 𝑑𝑒𝑡(𝐴) ≠ 0, then

𝐴11 𝐴21 𝐴𝑛1
⋯
det(𝐴) det(𝐴) det(𝐴)
1 𝐴 12 𝐴 22 𝐴𝑛2
−1
𝐴 = 𝑎𝑑𝑗(𝐴) = det(𝐴) det(𝐴) ⋯
det(𝐴) det(𝐴) .
⋮ ⋮ ⋮
𝐴1𝑛 𝐴2𝑛 ⋱ 𝐴𝑛𝑛
⋯
[det(𝐴) det(𝐴) det(𝐴)]

1.6 Linear Systems

A linear system (A system of linear equations) with 𝑚-equations and 𝑛-unknowns

consists of 𝑚 simultaneous equations each one is an equation in 𝑛-variables. That is, it
has the form:
𝑎11 𝑥1 + 𝑎12 𝑥2 + … + 𝑎1𝑛 𝑥𝑛 = 𝑏1
𝑎21 𝑥1 + 𝑎22 𝑥2 + … + 𝑎2𝑛 𝑥𝑛 = 𝑏2
⋮ ⋮ ⋱ ⋮ (1)
⋮
𝑎𝑚1 𝑥1 + 𝑎𝑚2 𝑥2 + … + 𝑎𝑚𝑛 𝑥𝑛 = 𝑏𝑚

A solution for the linear system is a sequence of 𝑛 real numbers s1, s2, ..., sn

which when substituted in the equations of the linear system all become true statements.

The principal question for this kind of systems is to find the set of solutions to this
system, that is to find all 𝑛-tuple (𝑥1 , 𝑥2 , . . . , 𝑥𝑛 ) that satisfy (1)

A general system of linear equations lies in one and only one of the following
categories

i. The system has no solution.

9
 ii. The system has exactly one solution.
iii. The system has infinitely many solutions.

If there are fewer equations than variables in a linear system, then the system
either has no solution or it has infinitely many solutions.

Definition 1.6.1 [ 9 , 𝑝. 3 ]. (Consistent and Inconsistent) A system of equations that has

at least one solution is called consistent. Otherwise, it is called inconsistent.

The key idea in finding the solution of a linear system is to apply elementary
row operations

Theorem 1.6.2 [ 9 , 𝑝. 7 ]. If any finitely many operations of the following is (are) applied
to the linear system (1)

1. Interchange two equations.

2. Multiply an equation by a nonzero constant.

3. Add a multiple of one equation to another.

To get a new system. Then both systems have the same solution.

The operations listed in Theorem 1.6.2 are called elementary row operations.

Definition 1.6.3 [ 9 , 𝑝. 49 ]. (Row equivalent) Let 𝐴 and 𝐵 be two 𝑚 × 𝑛 matrices.

We say that 𝐴 is row equivalent to 𝐵 if 𝐵 can be obtained by applying a finite

sequence of elementary row operations to 𝐴.

Now define the following matrices:

𝑎11 𝑎12 ⋯ 𝑎1𝑛 𝑥1 𝑏1
𝑎21 𝑎22 ⋯ 𝑎2𝑛 𝑥2 𝑏
𝑨= [ ⋮ 𝒃 = [ 2]
⋱ ⋮ ],
⋮ 𝒙 = [ ⋮ ],
⋮
𝑎𝑚1 𝑎𝑚2 ⋯ 𝑎𝑚𝑛 𝑥𝑛 𝑏𝑚
Then
𝑎11 𝑎12 ⋯ 𝑎1𝑛 𝑥1 𝑎11 𝑥1 + 𝑎12 𝑥2 + … + 𝑎1𝑛 𝑥𝑛
𝑎21 𝑎22 ⋯ 𝑎2𝑛 𝑥2 𝑎21 𝑥1 + 𝑎22 𝑥2 + … + 𝑎2𝑛 𝑥𝑛
𝑨𝒙 = [ ⋮ ⋮ ][ ⋮ ] = [ ⋮ ⋱ ⋮ ]
⋱ ⋮ ⋮
𝑎𝑚1 𝑎𝑚2 ⋯ 𝑎𝑚𝑛 𝑥𝑛 𝑎𝑚1 𝑥1 + 𝑎𝑚2 𝑥2 + … + 𝑎𝑚𝑛 𝑥𝑛

10
 The entries in the product 𝑨𝒙 are merely the left sides of the equations in (1).
Hence the linear system (1) can be written in matrix form as 𝑨𝒙 = 𝒃 .
The matrix 𝐴 is called the coefficient matrix of the linear system (1), and the
matrix
𝑎11 𝑎12 ⋯ 𝑎1𝑛 𝑏1
𝑎21 𝑎22 ⋯ 𝑎2𝑛 𝑏2
[ ⋮ ⋮ ⋱ ⋮ | ⋮ ],
𝑎𝑚1 𝑎𝑚2 ⋯ 𝑎𝑚𝑛 𝑏𝑚

obtained by adjoining b to A, is called the augmented matrix of the linear system

(1). The augmented matrix of (1) will be written as [𝑨|𝒃]. Conversely, any matrix

with more than one column can be thought of as the augmented matrix of a linear

system. The coefficient and augmented matrices play key roles in solving linear

systems.

Definition 1.6.4 [9 , 𝑝. 59 ]. (Homogenous Systems) A system of linear equations

is said to be homogenous if the constant terms are all zero. That is, the system has
form 𝑨𝒙 = 𝟎.

We note here that a homogenous system is always consistent,

in fact 𝑥1 = 𝑥2 = ⋯ = 𝑥𝑛 = 0 is a solution which is called the trivial solution.

1.7 Gaussian Elimination, Cramer's rule

Definition 1.7.1 [2 , 𝑝. 8 ]. (Reduced Row-Echelon Form) A matrix having the

following properties 1, 2 and 3 (but not necessarily 4) is said to be in row echelon
form (r.e.f). A matrix that satisfies all the 4 conditions is said to be in reduced row
echelon form (r.r.e.f)

1- If a row does not consist entirely of zeros, then the first nonzero number in
the row is a 1. (We call this a leading 1.)
2- Any row that consists entirely of zeros is placed at the bottom of the matrix.
3- In any two successive rows that do not consist entirely of zeros, the leading 1
in the lower row occurs farther to the right than the leading 1 in the higher row.
4- Each column that contains a leading 1 has zeros everywhere else.

11
 Any 𝑚 × 𝑛 matrix 𝐴 can be transformed into a reduced row echelon form 𝐴′
which is row equivalent to 𝐴, and this 𝐴′ is unique .

Definition 1.7.2 [9 , 𝑝. 54 ] (Gauss-Jordan reduction) The Gauss-Jordan reduction

procedure for solving the linear system 𝑨𝒙 = 𝒃 is as follows.

Step1. Form the augmented matrix [𝑨|𝒃].

Step 2. Transform the augmented matrix to reduced row-echelon form by using

elementary row operations.

Step 3. The linear system that corresponds to the matrix in reduced row-echelon
form that has been obtained in step 2 has exactly the same solutions as the given
linear system. For each nonzero row of the matrix in reduced row-echelon form,
solve the corresponding equation for the unknown that corresponds to the leading
entry of the row. The rows consisting entirely of zeros can be ignored, since the
corresponding equation will be satisfied for any values of the unknowns.

Definition 1.7.3 [2 , 𝑝. 50 ]. ( Elementary matrix ) An 𝑛 × 𝑛 matrix is called an

elementary matrix if can be obtained from the 𝑛 × 𝑛 identity matrix 𝐼𝑛 by
preforming a single elementary row operations.

Theorem 1.7.4 [2 , 𝑝. 19 ]. A homogeneous system of linear equations with more

unknowns than equations has infinitely many solutions.

Theorem 1.7.5 [2 , 𝑝. 109 ]. (Cramer's rule). If 𝑨𝒙 = 𝒃 is a system of 𝑛 linear

equations in 𝑛 unknowns such that det(𝐴) ≠ 0, then the system has the unique

solution, given as follows:

det(𝐴1 ) det(𝐴2 ) det(𝐴𝑛 )

𝑥1 = , 𝑥2 = , …. , 𝑥𝑛 = .
det(𝐴) det(𝐴) det(𝐴)

Where 𝐴𝑗 is the matrix obtained by replacing the entries in the 𝑗𝑡ℎ column of 𝐴
𝑏1
𝑏2
by the entries in the matrix 𝒃 = [ ] .
⋮
𝑏𝑛

12
1.8 Vectors in the Euclidean space 𝑹𝒏

In this section we state some basic definitions regarding the vector space 𝑅𝑛

Definition 1.8.1 [2 , 𝑝. 167 ]. (Vectors in 𝒏-space) If 𝑛 is a positive integer, then

an ordered 𝑛-tuple is a sequence of 𝑛 real numbers (𝑎1 , 𝑎2 , … , 𝑎𝑛 ). The set of all

ordered 𝑛-tuples is called 𝑛-space and is denoted by 𝑅𝑛 and an element

(𝑎1 , 𝑎2 , … , 𝑎𝑛 ) is called a vector.

𝑥1
A vector in the plane is a 2-vector 𝑢 = [𝑦 ], where 𝑥1 and 𝑦1 are real
1

numbers, called the components of 𝑢.

Definition 1.8.2 [9 , 𝑝. 148 ]. (Norm of a vector) The length (also called magnitude or

norm) of the vector 𝑢 = (𝑢1 , 𝑢2 , … , 𝑢𝑛 ) in 𝑅𝑛 is ‖𝑢‖ = √𝑢21 + 𝑢22 + ⋯ + 𝑢2𝑛 .

Definition 1.8.3 [9 , 𝑝. 148 ]. (Inner product) If 𝑢 = (𝑢1 , 𝑢2 , … , 𝑢𝑛 ) and

𝑣 = (𝑣1 , 𝑣2 , … , 𝑣𝑛 ) are vectors in 𝑅𝑛 , then their dot product is defined by

𝑢. 𝑣 = 𝑢1 𝑣1 + 𝑢2 𝑣2 + ⋯ + 𝑢𝑛 𝑣𝑛 . The dot product in 𝑅𝑛 is also called the

standard inner product

Definition 1.8.4 [2 , 𝑝. 174 ]. (orthogonality) Two vectors 𝑢 and 𝑣 in 𝑅𝑛 are

called orthogonal if 𝑢. 𝑣 = 0.

Definition 1.8.5 [2 , 𝑝. 136 ]. (orthogonal projection)

The vector 𝑢 is the sum of 𝑤1 and 𝑤2 , where 𝑤1 is parallel to 𝑣

and 𝑤2 is perpendicular to 𝑣. the vector 𝑤1 is called the

orthogonal projection of 𝑢 on 𝑣 or sometimes the vector

13
component of 𝑢 along 𝑣. It is denoted by 𝑝𝑟𝑜𝑗𝑣 𝑢.

The vector 𝑤2 is called the vector component of 𝑢 orthogonal to 𝑣. Since we have

𝑤2 = 𝑢 − 𝑤1 , thus the vector can be written 𝑤2 = 𝑢 − 𝑝𝑟𝑜𝑗𝑣 𝑢.

Theorem 1.8.6 [2 , 𝑝. 136 ]. If 𝑢 and 𝑣 are vectors in 2-space or 3-space and if

𝑢.𝑣
𝑣 ≠ 0, then 𝑝𝑟𝑜𝑗𝑣 𝑢 = ‖𝑣‖2 𝑣 (vector component of 𝑢 along 𝑣).

𝑢.𝑣
𝑢 − 𝑝𝑟𝑜𝑗𝑣 𝑢 = 𝑢 − ‖𝑣‖2 𝑣 (vector component of 𝑢 orthogonal to 𝑣).

1.9 Rank and Nullity

Definition 1.9.1 [2 , 𝑝. 222 ]. (subspace) A subset 𝑊 of a vector space 𝑉 is called

a subspace of 𝑉 if 𝑊 is itself a vector space under the addition and scalar

multiplication defined on 𝑉.

Definition 1.9.2 [9 , 𝑝. 207 ]. (Linear combination) Let 𝑣1 , 𝑣2 , … , 𝑣𝑛 be vectors

in a vector space 𝑉. A vector 𝑣 in 𝑉 is called a linear combination of 𝑣1 , 𝑣2 , … , 𝑣𝑛

if 𝑣 = 𝑘1 𝑣1 + 𝑘2 𝑣2 + ⋯ + 𝑘𝑛 𝑣𝑛 for some numbers 𝑘1 , 𝑘2 , … , 𝑘𝑛 .

Definition 1.9.3 [2 , 𝑝. 233 ].(Linear independence) If 𝑆 = {𝑣1 , 𝑣2 , … , 𝑣𝑛 } is a

nonempty set of vectors, then the vector equation 𝑘1 𝑣1 + 𝑘2 𝑣2 + ⋯ + 𝑘𝑛 𝑣𝑛 = 0

has at least one solution, namely 𝑘1 = 0, 𝑘2 = 0, … , 𝑘𝑛 = 0. If this is the only

solution, then S is called a linearly independent set. If there are other solutions,

then S is called a linearly dependent set.

Definition 1.9.4 [9 , 𝑝. 213 ]. ( Span ) The vector 𝑣1 , 𝑣2 , … , 𝑣𝑛 in a vector space 𝑉 are said

to span 𝑉 if every vector in 𝑉 is a linear combination of 𝑣1 , 𝑣2 , … , 𝑣𝑛 . Moreover, if these

vectors are distinct and we denote them as a set 𝑆 = {𝑣1 , 𝑣2 , … , 𝑣𝑛 }, then we also say that

the set S spans V, or that 𝑣1 , 𝑣2 , … , 𝑣𝑛 spans 𝑉, or that 𝑉 is spanned by 𝑆, or 𝑠𝑝𝑎𝑛 𝑆 = 𝑉.

14
Definition 1.9.5 [2 , 𝑝. 259 ]. (row space and column space) If 𝐴 an 𝑚 × 𝑛

matrix, then the subspace of 𝑅𝑛 spanned by the row vectors of 𝐴 is called the row

space of 𝐴, and the subspace of 𝑅𝑚 spanned by column vectors is called the

column space of A.

Definition 1.9.6 [2 , 𝑝. 273 ]. (Rank and nullity) The common dimension of the

row space and column space of a matrix 𝐴 is called the rank of 𝐴 and is denoted

by 𝑟𝑎𝑛𝑘(𝐴), the dimension of the null space which is the solution space of

𝑨𝒙 = 𝟎 of 𝐴 is called the nullity of 𝐴 and is denoted by 𝑛𝑢𝑙𝑙𝑖𝑡𝑦(𝐴).

Theorem 1.9.7 [2 , 𝑝. 275 ]. If 𝐴 is an 𝑛 × 𝑛 matrix, then

(a) 𝑟𝑎𝑛𝑘(𝐴)= the number of leading variables in the solution of 𝑨𝒙 = 𝟎.

(b) 𝑛𝑢𝑙𝑙𝑖𝑡𝑦(𝐴) = the number of parameters in the solution of 𝑨𝒙 = 𝟎.

Theorem 1.9.8 [9 , 𝑝. 249 ]. (Rank-Nullity Theorem) If 𝐴 is an 𝑚 × 𝑛 matrix, Then

𝑟𝑎𝑛𝑘(𝐴) + nullity (𝐴) = 𝑛

Theorem 1.9.9[2 , 𝑝. 275 ]. If 𝐴 is any 𝑛 × 𝑚 matrix, then 𝑟𝑎𝑛𝑘(𝐴) = 𝑟𝑎𝑛𝑘(𝐴𝑇 )

Theorem 1.9.10 [9 , 𝑝. 250 ]. if 𝐴 is an 𝑛 × 𝑛 matrix, then 𝑟𝑎𝑛𝑘(𝐴) = 𝑛 if and

only if 𝑑𝑒𝑡(𝐴) ≠ 0

Definition 1.9.11 [9 , 𝑝. 327 ] (Linear transformation from 𝑅𝑛 to 𝑅𝑚 )

A linear transformation 𝑇 from 𝑅𝑛 into 𝑅𝑚 is a function assigning a unique

vector 𝑇(𝑥) in 𝑅𝑚 to each 𝑥 in 𝑅𝑛 such that:

(a) 𝑇(𝑥 + 𝑦) = 𝑇(𝑥) + 𝑇(𝑦), for every 𝑥 and 𝑦 in 𝑅𝑛 .

(b) 𝑇(𝑘 𝑥) = 𝑘 𝑇(𝑥) for every 𝑥 in 𝑅𝑛 and every scalar 𝑘.
If 𝑛 = 𝑚, the linear transformation 𝑇: 𝑅 𝑛 → 𝑅 𝑛 is called a linear operator on 𝑅𝑛 .

15
The matrix 𝐴𝑚×𝑛 = [𝑇(𝑒1 ): 𝑇(𝑒2 ): … : 𝑇(𝑒𝑛 )] is called the standard matrix for the linear
transformation and 𝑇: 𝑅 𝑛 → 𝑅 𝑚 is a multiplication by 𝐴, that is 𝑇𝐴 (𝑥) = 𝐴𝑥.

Definition 1.9.12 [2 , 𝑝. 395 ]. (Kernel and Range) If 𝑇: 𝑉 → 𝑊 is a linear

transformation, then the set of vectors in 𝑉 that 𝑇 maps into 0 is called the kernel
of 𝑇; it is denoted by 𝑘𝑒𝑟(𝑇). The set of all vectors in 𝑊 that are images under 𝑇
of at least one vector in 𝑉 is called the range of 𝑇; it is denoted by 𝑅(𝑇).

Theorem 1.9.13 [2 , 𝑝. 397 ]. If 𝐴 is an 𝑚 × 𝑛 matrix, and 𝑇𝐴 : 𝑅𝑛 → 𝑅𝑚 is

multiplication by 𝐴 and define

𝑟𝑎𝑛𝑘(𝑇𝐴 ) = dim(𝑅(𝑇𝐴 )), nullity (𝑇𝐴 ) = dim(ker(𝑇𝐴 )), then

(i) nullity (𝑇𝐴 ) = nullity (𝐴)

(ii) 𝑟𝑎𝑛𝑘(𝑇𝐴 ) = 𝑟𝑎𝑛𝑘(𝐴)

Theorem 1.9.14 [2 , 𝑝. 281 ]. (Invertible matrix theorem) If 𝐴 is an 𝑛 × 𝑛 matrix, and

if 𝑇𝐴 : 𝑅𝑛 → 𝑅𝑛 is multiplication by 𝐴, then the following are equivalent.

(a) 𝐴 is invertible
(b) 𝑨𝒙 = 𝟎 has only the trivial solution.
(c) The reduced row-echelon form of 𝐴 is 𝐼𝑛
(d) 𝐴 is expressible as a product of elementary matrices.
(e) 𝑨𝒙 = 𝒃 is consistent for every 𝑛 × 1 matrix 𝑏.
(f) 𝑨𝒙 = 𝒃 has exactly one solution for every 𝑛 × 1 matrix 𝑏.
(g) det(𝐴) ≠ 0
(h) The range of 𝑇𝐴 is 𝑅𝑛
(i) 𝑇𝐴 is one–to–one.
(j) 𝐴 has rank 𝑛.
(k) 𝐴 has nullity 0.
For proofs of these properties [2].

16
Theorem 1.9.15 [9 , 𝑝. 365 ]. Let 𝑇: 𝑅 𝑛 → 𝑅 𝑚 be a linear transformation defined by

𝑇(𝑥) = 𝐴𝑥, 𝑥 in 𝑅 𝑛 , where 𝐴 is an 𝑚 × 𝑛 matrix.

(1) 𝑇 is one-to-one if and only if 𝑟𝑎𝑛𝑘(𝐴) = 𝑛

(2) 𝑇 is onto if and only if 𝑟𝑎𝑛𝑘(𝐴) = 𝑚

Lemma 1.9.16 [10]. Consider the vector spaces 𝑅𝑚 and 𝑅𝑛 , and let 𝑇𝐴 ∶ 𝑅𝑚 → 𝑅𝑛 and
𝑇 𝐴𝑇 : 𝑅 𝑛 → 𝑅 𝑚 the adjoint operator, then the following statement holds

)1( 𝑅𝑎𝑛𝑔(𝐴) = 𝑅 𝑛 ↔ ∃ 𝛾 > 0 such that ‖𝐴𝑇 𝑧‖ 𝑚

≥ 𝛾 ‖𝑧‖𝑅𝑛 , 𝑧 ∈ 𝑅𝑛 .
𝑅

)2( 𝑅𝑎𝑛𝑔(𝐴) = 𝑅𝑛 ↔ 𝐾𝑒𝑟 (𝐴𝑇 ) = {0} ↔ 𝐴𝑇 𝑖𝑠 1 − 1 (see theorem 1.9.15)

17
 Chapter Two

Determinant

In the history of matrices, mathematicians are interested in finding the value

of the determinant for square matrices only, actually the definition of determinant
and its properties are discussed only for square matrices. To break this we
generalize the concept of determinant from a square matrix to a non-square
matrix, and we also study their properties, methods of computation and some
application.

2.1 Determinant function

There are many ways that general 𝑚 × 𝑛 determinants can be defined. We'll
first define a determinant function in terms of characterizing properties that we
want it to have. Then we'll use the construction of a determinant following the
method given in the section, and through it we will prove that cofactor expansion
definition of the determinant and radic definition are the same.

We generalize the idea given in [8] for the case of rectangular matrices:

Definition 2.1.1 An 𝑚 × 𝑛 matrix 𝐴 = (𝑎𝑖𝑗 ) where 𝑚 ≤ 𝑛 is said to have form

𝐼̃ if i) Entries of A are 0 or 1

ii) Eeach row have exactly one non-zero entry

That is, 𝐼̃ 𝑚×𝑛 consists of all matrices of the form [𝐴1 𝐴2 ⋯ 𝐴𝑛 ] where

0
𝐴𝑗 ∈ {𝑒1 , 𝑒2 , … , 𝑒𝑚 } ∪ [0], {𝑒1 , 𝑒2 , … , 𝑒𝑚 } is the standard basis for 𝑅 𝑚
⋮
0
Such a matrix is called a permutation matrix in literature.
0 1 0 0 0
Example, 𝐴 = [0 0 1 0 0] is an example of an element in 𝐼̃
0 0 0 0 1

Definition 2.1.2 A determinant function assigns to each 𝑚 × 𝑛 (𝑚 ≤ 𝑛) matrix

𝐴 a scalar associated to the matrix, denoted 𝑑𝑒𝑡(𝐴) or |𝐴| such that

A1: The determinant of an 𝑚 × 𝑛 (𝑚 ≤ 𝑛) in 𝐼̃ 𝑚×𝑛 is (−1)𝑝+𝑞

18
where 𝑝 = 1 + 2 + ⋯ + 𝑚 and 𝑞 = 𝑗1 + 𝑗2 + ⋯ + 𝑗𝑛 ( 𝑗1 , 𝑗2 , … , 𝑗𝑛 ) represent columns

that contain 1’s.

A2: If the matrix 𝐵 is identical to the matrix 𝐴 except the entries in one of the
rows of 𝐵 are each equal to the corresponding entries of multiplied by the same
scalar 𝑐, then |𝐵| = 𝑐|𝐴|.

A3: If the matrices 𝐴, 𝐵 and 𝐶 are identical except for the entries in one row, and
for that row an entry in 𝐴 is found by adding the corresponding entries in 𝐵
and 𝐶, then |𝐴| = |𝐵| + |𝐶|.

A4: If the matrix 𝐵 is the result of exchanging two rows of 𝐴, then the determinant
of 𝐵 is the negation of the determinant of 𝐴.

These conditions are enough to characterize the determinant, but they don’t
show such a determinant function exists and is unique. We'll show both existence
and uniqueness, but start with uniqueness. First, we'll note a couple of properties
that determinant functions have that follow from the definition.

Theorem 2.1.3 A determinant function has the following four properties.

(a) The determinant of any matrix 𝐴𝑚×𝑛 ( 𝑚 ≤ 𝑛) with an entire row of 0′ 𝑠 is 0.

(b) The determinant of any matrix 𝐴𝑚×𝑛 ( 𝑚 ≤ 𝑛) with two identical rows is 0.

(c) If one row of a matrix 𝐴𝑚×𝑛 ( 𝑚 ≤ 𝑛) is a scalar multiple of another row, then
its determinant is 0.
(d) If a multiple of one row of a matrix 𝐴𝑚×𝑛 ( 𝑚 ≤ 𝑛) is added to another row,
then the resulting matrix has the same determinant as the original matrix.

Proof: Property (a) follows from the second statement (A2) in the definition.
If 𝐴 has a whole row of 0′ 𝑠, then using that row and 𝑐 = 0 in the second
statement (A2) of the definition, then 𝐵 = 𝐴.
So, |𝐴| = 0|𝐵|. Therefore, det(𝐴) = 0.

Property (b) follows from the fourth statement (A4) in the definition.

19
If you exchange the two identical rows, the result is the original matrix, but its
determinant is negated. The only scalar which is its own negation is 0.
Therefore, the determinant of the matrix is 0.
Property (c) follows from the second statement (A2) in the definition and Property (b).
Property (d) follows from the third statement (A3) in the definition and Property (c).

Now we can show the uniqueness of determinant function.

Theorem 2.1.4 There is at most one determinant function.

Proof: The four properties that determinants are enough to find the value of the

determinant of a matrix.

Suppose a matrix 𝐴𝑚×𝑛 (𝑚 ≤ 𝑛) has more than one nonzero entry in a row. Then
using the third statement (A3) in definition 2.1.2.
Now, det(𝐴) = det(𝐴1 ) + det(𝐴2 ) + ⋯ + det(𝐴𝑛 )
Where 𝐴𝑗 is the matrix that looks just like 𝐴 except in that row, all the entries are
0 expect the 𝑗𝑡ℎ one which is the 𝑗𝑡ℎ entry of that row in 𝐴.
That means we can reduce the question of evaluating determinants of general
matrices to evaluating determinants of matrices that have at most one nonzero
entry in each row.
By Property (a) in the theorem 2.1.3, if the matrix has a row of all 0′ 𝑠, its
determinant is 0. Thus, we only need to consider matrices that have exactly one
nonzero entry in each row.
Using the second statement (A2) in definition 2.1.2, we can further assume that
the nonzero entry in that row is 1.
Now, we're down to evaluating determinants that only have entries of 0′ 𝑠 and 1′ 𝑠
with exactly one 1 in each row.
If two of those rows have the 1′ 𝑠 in the same column, then by Property (b) in the
theorem 2.1.3, that matrix has determinant 0.
Now the only matrices left to consider are matrices from 𝐼̃.
Using alternation, the fourth condition (A4) in definition 2.1.2, the rows can be
interchanged until the 1′ 𝑠 only lie on the entry 𝑎𝑖𝑗

20
Finally, we're left with a matrix in 𝐼̃, but by the first condition (A1) in definition
2.1.2, its determinant is (−1)𝑝+𝑞 .
Thus, the value of the determinant of every matrix is determinant by the
definition. There can be only one determinant function.

We need some way to construct a function with those properties, and well do
that with a "cofactor construction" and " radic construction".

2.2 Minors and cofactors.

To every non-square matrix 𝐴 = (𝑎𝑖𝑗 ) of order 𝑚 × 𝑛, we can associate a

number (real or complex) called determinant of the non-square matrix 𝐴, where
𝑎𝑖𝑗 = (𝑖, 𝑗)𝑡ℎ element of 𝐴.

This may be thought of as a function which associates to each non-square

matrix over a field 𝐹 a unique number from 𝐹 (real or complex). If 𝑀 is the set of
non-square matrices, 𝐾 is the set of real numbers and 𝑓: 𝑀 → 𝐾 is defined by
𝑓(𝐴) = 𝑘, where 𝐴 ∈ 𝑀 and 𝑘 ∈ 𝐾, then 𝑓(𝐴) is called the determinant of 𝐴. It
is also denoted by |𝐴| or det( 𝐴).

The determinant can also be viewed as a function of the columns of the matrix.
Let these columns be 𝐴1 , 𝐴2 , . . , 𝐴𝑛 , then we write the determinant as

|𝐴1 , 𝐴2 , . . , 𝐴𝑛 | or 𝑑𝑒𝑡(𝐴1 , 𝐴2 , . . , 𝐴𝑛 )

Definition 2.2.1 [3] (Determinants of order 𝟏 × 𝒏 )

If 𝐴 = [𝑎11 𝑎12 𝑎13 ⋯ 𝑎1𝑛 ], then the determinant of 𝐴 is

| 𝐴 | = 𝑎11 − 𝑎12 + 𝑎13 − ⋯ + (−1)1+𝑛 𝑎1𝑛

= ∑(−1)1+𝑖 𝑎1𝑖
𝑖=1

Example 2.2.2 |1 5 9| = 1 − 5 + 9 = 5

For larger matrices, we use cofactor expansion to find the determinant of 𝐴.

First of all, let’s define a few terms.

21
Definition 2.2.3 [3] (Minor) Let 𝐴 = (𝑎𝑖𝑗 ) be an 𝑚 × 𝑛 matrix. for each entry
𝑎𝑖𝑗 of 𝐴, we define the minor 𝑀𝑖𝑗 of 𝑎𝑖𝑗 to be the determinant of the
( 𝑚 − 1) × (𝑛 − 1) matrix which remains when the 𝑖𝑡ℎ row and 𝑗𝑡ℎ column are
deleted from 𝐴.
1 3 5
Example 2.2.4 Let 𝐴 = [ ]
2 4 6

To find 𝑀11 , look at element 𝑎11 = 1, delete the entries from column 1 and row 1
that corresponding to 𝑎11 = 1, see the image below.

[
1 − −]
− 4 6
Then 𝑀11 is the determinant of remaining matrix, i.e.,

𝑀11 = |4 6| = 4 − 6 = −2

Similarly, 𝑀22 can be found by looking at the element 𝑎22 = 4 and delete the
same row and column where this element is found, i.e., deleting the second row,
second column:

[
1 − 5]
− 4 −

Then, 𝑀22 = |1 5| = 1 − 5 = −4

It is easy to see that for the matrix 𝐴 the minors of ramming elements are

− 3 −]
[
2 − 6
𝑀12 = |2 6| = 2 − 6 = −4

[− − 5]
2 4 −
𝑀13 = |2 4| = 2 − 4 = −2

− 3 5]
[
2 − −
𝑀21 = |3 5| = 3 − 5 = −2

1 3 −]
[
− − 6
𝑀23 = |1 3| = 1 − 3 = −2

22
Definition 2.2.5 [3] (Cofactor). Let 𝐴 = (𝑎𝑖𝑗 ) be an 𝑚 × 𝑛 matrix. for each
entry 𝑎𝑖𝑗 of 𝐴, we define the cofactor of 𝑎𝑖𝑗 , which is denoted by 𝐶𝑖𝑗 as

𝐶𝑖𝑗 = (−1)𝑖+𝑗 𝑀𝑖𝑗

The matrix of cofactors of 𝐴 is 𝐶 with 𝑐𝑖𝑗 is cofactor of 𝑎𝑖𝑗 and the adjoint of
𝐴 is 𝑎𝑑𝑗(𝐴) is the transpose of 𝐶.

Basically, the cofactor is either 𝑀𝑖𝑗 , or −𝑀𝑖𝑗 where the sign depends on the
location of the element in the matrix. For that reason, it is easier to know the
pattern of cofactor instead of actually remembering the formula. If you start in the
position corresponding to 𝑎11 with a positive sign, the sign of the cofactor has an
alternating pattern. you can see this by looking at a matrix containing the sign of
the cofactors:
+ − + ⋯ ⋯
[−+ − ⋯ ⋯]
⋮ ⋮ ⋮ ⋮ ⋮

The element 1 in matrix 𝐴 (example 2.2.4) has place sign + and minor -2 so its
cofactor is + (-2) = -2
The element 4 in matrix 𝐴 (example 2.2.4) has place sign + and minor -4 so its
cofactor is + (- 4) = - 4
Proceeding in this way we can find all the cofactors.
The original matrix, its matrix of minors and its matrix of cofactors are:
1 3 5 −2 −4 −2
𝐴= [ ], 𝑀=[ ],
2 4 6 −2 −4 −2

−2 2
−2 4 −2
𝐶= [ ], 𝑎𝑑𝑗(𝐴) = [ 4 −4]
2 −4 2
−2 2

2.3 Determinant Using Cofactor Expansion

In this section we shall deal with matrices of size 𝑚×𝑛 where 𝑛 ≤ 𝑚 or

𝑛 ≥ 𝑚. A matrix 𝐴𝑚×𝑛 with 𝑛≠𝑚 is called a rectangular matrix. When
𝑚 ≤ 𝑛, 𝐴 is said to be a horizontal matrix, otherwise 𝐴 is said to be a vertical
matrix.

23
Definition 2.3.1 [3] Let 𝐴 = (𝑎𝑖𝑗 ) be an 𝑚 × 𝑛 matrix with 𝑚 ≤ 𝑛. (horizontal
matrix). The determinant of 𝐴 is defined as
𝑛

𝑑𝑒𝑡(𝐴) = 𝑎11 𝐶11 + 𝑎12 𝐶12 + 𝑎13 𝐶13 + ⋯ + 𝑎1𝑛 𝐶1𝑛 = ∑ 𝑎1𝑗 𝐶1𝑗
𝑗=1

This is called cofactor expansion along the first row.

The determinant of a vertical matrix 𝐴 is defined to be the determinant of the

horizontal matrix 𝐴𝑇 .

The following theorem asserts that we can evaluate the determinant of a larger
horizontal matrix by selecting any row, multiplying each element in that row by
its corresponding cofactor, and summing the result, a result which is true in the
case of square matrices (Theorem 1.4.2).

The following theorem is from [7] but we give here another proof

Theorem 2.3.2 (The cofactor expansion theorem)

Let 𝐴 = (𝑎𝑖𝑗 ) be an 𝑚 × 𝑛 matrix. If 𝑚 ≤ 𝑛, the determinant of 𝐴 is

𝑑𝑒𝑡(𝐴) = 𝑎𝑖1 𝐶𝑖1 + 𝑎𝑖2 𝐶𝑖2 + 𝑎𝑖3 𝐶𝑖3 + ⋯ + 𝑎𝑖𝑛 𝐶𝑖𝑛

= ∑ 𝑎𝑖𝑗 𝐶𝑖𝑗
𝑗=1

This is called the determinant using cofactor expansion along the 𝑖𝑡ℎ row.

The proof of this theorem will be given after proving Theorem 2.4.1

1 3 5
Example 2.3.3. Evaluate the determinant of 𝐴 = [ ] by cofactor expansion
2 4 6

i) along the first row

ii) along the second row

Solution:

1 3 5
𝑑𝑒𝑡(𝐴) = | | = 1 |4 6| − 3 |2 6| + 5 |2 4|
2 4 6

24
 = 1 ( −2 ) − 3 (−4) + 5(−2 ) = 0

1 3 5
𝑑𝑒𝑡(𝐴) = | | = −2 |3 5| + 4 |1 5| − 6 |1 3|
2 4 6

= −2 ( −2 ) + 4 (−4) − 6(−2 ) = 0

Note that : when 𝐴 is an 𝑚 × 𝑛 matrix with 𝑚 ≥ 3 or 𝑛 ≥ 3, the cofactors 𝐶𝑖𝑗

are determinants of (𝑚 − 1 ) × (𝑛 − 1) matrices. To compute these determinants,
we apply cofactor expansion again, and obtain determinants of (𝑚 − 2 ) × 𝑛 − 2)
matrices. We keep applying cofactor expansion until we hit 1 × 𝑛 determinants,
which we know how to compute (see definition 2.2.1).

2 3 4 1
Example 2.3.4 Evaluate the determinant of 𝐴 = [1 5 0 2] using Theorem 2.3.2
2 4 1 3

Solution: Since 𝑚 < 𝑛, expanding a long any row, say first row

2 3 4 1
5 0 2| − 3 |1 0 2| + 4 |1 5 2| − 1 |1 5 0|
|1 5 0 2| = 2 |
4 1 3 2 1 3 2 4 3 2 4 1
2 4 1 3

= 2(5(−2) − 0 + 2(3)) − 3(1(−2) − 0 + 2(1))

+ 4(1(1) − 5(−1) + 2(−2)) − 1(1(3) − 5(1) + 0)

= 2(−4) − 3(0) + 4(2) − 1(−2) = 2

In calculating a determinant using cofactor expansion, it is usually a good idea

to choose a row or column containing as many zeros as possible.

Theorem 2.3.5 [3] If 𝐴 is a horizontal matrix with a row of zeros, then

det(𝐴) = 0

Proof: Since the determinant of 𝐴 can be found by a cofactor expansion along

any row, we can use the row of zeros
det(𝐴) = 0 𝐶𝑖1 + 0 𝐶𝑖2 + ⋯ + (−1)𝑖+𝑛 0 𝐶𝑖𝑛
=0
This theorem represents property (a) in the theorem 2.1.3

25
2.4 Effect of elementary row operations on determinates

The evaluation of the determinant of an 𝑚 × 𝑛 (𝑚 ≤ 𝑛) matrix using the

definition 2.3.1 involves the summation of 𝑛𝑝( 𝑚−1) , each term being a product

of 𝑚 factors. As 𝑚 , 𝑛 increases, this computation becomes too cumbersome and

so another technique has been devised to evaluate the determinant which works

quite efficiently. This technique uses row operations to put a matrix into a form in

which the determinant is easily calculated, keeping track of the row operations

used, and how they affect the determinant, we can backtrack, and determine what

the original determinant was.

We will look at the effect of each elementary row operation on the

determinant.

The following theorem is from [3], [7], but we give here another proof
Theorem 2.4.1. If 𝐴 and 𝐵 are 𝑚 × 𝑛 matrices with 𝑚 ≤ 𝑛, and 𝐵 is obtained

from 𝐴 by interchanging two rows of 𝐴, then 𝑑𝑒𝑡 (𝐴) = − 𝑑𝑒𝑡 (𝐵)

𝑎11 𝑎12 … 𝑎1𝑛

Proof: Base case : Let 𝐴 = [𝑎 𝑎22 … 𝑎2𝑛 ], and let
21

𝑎21 𝑎22 … 𝑎2𝑛

𝐵 = [𝑎 𝑎12 … 𝑎1𝑛 ]. Then 𝐵 is the only matrix that can be obtained from
11

𝐴 by swapping rows. And we see that

det(𝐵) = 𝑎21 (𝑎12 − 𝑎13 + ⋯ + (−1)𝑛 𝑎1𝑛 ) − 𝑎22 (𝑎11 − 𝑎13 + ⋯ +

(−1)𝑛 𝑎1𝑛 ) + 𝑎23 (𝑎11 − 𝑎12 + ⋯ + (−1)𝑛 𝑎1𝑛 ) + … + (−1)𝑛+1 𝑎2𝑛 (𝑎11 −

𝑎12 + ⋯ + (−1)𝑛 𝑎1𝑛−1 )

=(𝑎21 𝑎12 − 𝑎21 𝑎13 + ⋯ + (−1)𝑛 𝑎21 𝑎1𝑛 ) − (𝑎22 𝑎11 − 𝑎22 𝑎13 +

⋯ + (−1)𝑛 𝑎22 𝑎1𝑛 ) + (𝑎23 𝑎11 − 𝑎23 𝑎12 + ⋯ + (−1)𝑛 𝑎23 𝑎1𝑛 ) + … +

(−1)𝑛+1 (𝑎2𝑛 𝑎11 − 𝑎2𝑛 𝑎12 + ⋯ + (−1)𝑛 𝑎2𝑛 𝑎1𝑛−1 ) (1)

26
Also ,
det(𝐴) = 𝑎11 (𝑎22 − 𝑎23 + ⋯ + (−1)𝑛 𝑎2𝑛 ) − 𝑎12 (𝑎21 − 𝑎23 + ⋯ +

(−1)𝑛 𝑎2𝑛 ) + 𝑎13 (𝑎21 − 𝑎22 + ⋯ + (−1)𝑛 𝑎2𝑛 ) + … + (−1)1+𝑛 𝑎1𝑛 (𝑎21 −

𝑎22 + ⋯ + (−1)𝑛 𝑎2𝑛−1 )

= (𝑎11 𝑎22 − 𝑎11 𝑎23 + ⋯ + (−1)𝑛 𝑎11 𝑎2𝑛 ) − (𝑎12 𝑎21 − 𝑎12 𝑎23 +
𝑛
⋯ + (−1)𝑛 𝑎12 𝑎2𝑛 ) + (𝑎13 𝑎21 − 𝑎13 𝑎22 + ⋯ + (−1) 𝑎13 𝑎2𝑛 ) + … +

(−1)1+𝑛 (𝑎1𝑛 𝑎21 − 𝑎1𝑛 𝑎22 + ⋯ + (−1)𝑛 𝑎1𝑛 𝑎2𝑛−1 ) (2)

From (1) and (2) , clearly

𝑑𝑒𝑡 (𝐴) = − 𝑑𝑒𝑡 (𝐵)

Induction hypothesis: For all 𝑘 × 𝑛 with (𝑘 ≤ 𝑛) matrices 𝐴, if 𝐵 is obtained

from 𝐴 by swapping two rows , then 𝑑𝑒𝑡 (𝐴) = − 𝑑𝑒𝑡 (𝐵)

Induction step : Let 𝐴 be a ( 𝑘 + 1) × (𝑛) with (𝑘 + 1 ≤ 𝑛) matrix, and let 𝐵 be

a matrix obtained from 𝐴 by swapping two rows. Say row 𝑟 and 𝑟 + 1 of 𝐴 were

swapped when making 𝐵.

𝑎11 𝑎12 ⋯ 𝑎1𝑛 𝑎11 𝑎12 ⋯ 𝑎1𝑛

⋮ ⋮ ⋱ ⋮ ⋮ ⋮ ⋱ ⋮
𝑎𝑟1 𝑎𝑟2 ⋯ 𝑎𝑟𝑛 𝑎(𝑟+1)1 𝑎(𝑟+1)2 ⋯ 𝑎(𝑟+1)𝑛
𝐴 = 𝑎(𝑟+1)1 𝑎(𝑟+1)2 ⋯ 𝑎(𝑟+1)𝑛 , 𝐵= 𝑎𝑟1 𝑎𝑟2 ⋯ 𝑎𝑟𝑛
⋮ ⋮ ⋱ ⋮ ⋮ ⋮ ⋱ ⋮
[𝑎( 𝑘+1)1 𝑎( 𝑘+1)2 ⋯ 𝑎( 𝑘+1)𝑛 ] 𝑎
[ ( 𝑘+1)1 𝑎( 𝑘+1)2 ⋯ 𝑎( 𝑘+1)𝑛 ]

We may evaluate 𝑑𝑒𝑡 (𝐵) by cofactor expansion along its first row.

𝑑𝑒𝑡(𝐵) = 𝑏11 𝐵11 + 𝑏12 𝐵12 + 𝑏13 𝐵13 + ⋯ + 𝑏1𝑛 𝐵1𝑛

To compute 𝑑𝑒𝑡(𝐵), we will need to look at the submatrices 𝐵(𝑖 , 𝑗 ). Our choice

of 𝑖 = 1 means that 𝐵(1 , 𝑗 ) can be obtained from 𝐴(1 , 𝑗 ) by swapping the rows

𝑟 and 𝑟 + 1, as we swapped to get 𝐵 from 𝐴. This means that 𝐵(1 , 𝑗 ) is a 𝑘 × 𝑛

matrix that is obtained from 𝐴(1 , 𝑗 ) by swapping two rows, and thus, by our

inductive hypothesis, 𝑑𝑒𝑡𝐵(1 , 𝑗 ) = − det 𝐴(1 , 𝑗).

27
Now,

𝑑𝑒𝑡 (𝐵) = 𝑏11 (−1)1+1 𝑑𝑒𝑡 𝐵(1 , 1 ) + ⋯ + 𝑏1𝑛 (−1)1+𝑛 𝑑𝑒𝑡 𝐵(1 , 𝑛 )

= 𝑎11 (−1)1+1 𝑑𝑒𝑡𝐵(1 , 1 ) + ⋯ + 𝑎1𝑛 (−1)1+𝑛 𝑑𝑒𝑡𝐵(1 , 𝑛 )

= 𝑎11 (−1)1+1 (−1)𝑑𝑒𝑡𝐴(1 , 1 ) + ⋯ + 𝑎1𝑛 (−1)1+𝑛 (−1) 𝑑𝑒𝑡𝐴(1 , 𝑛 )

= −(𝑎11 (−1)1+1 𝑑𝑒𝑡𝐴(1 , 1 ) + ⋯ + 𝑎1𝑛 (−1)1+𝑛 𝑑𝑒𝑡𝐴(1 , 𝑛 ))

det(𝐵) = − det(𝐴)

This proves the result for the interchange of two adjacent rows in an 𝑚 × 𝑛

matrix. To see that this result holds for arbitrary row interchanges, we note that

the interchange of two rows, say row 𝑟 and 𝑠 where 𝑟 < 𝑠 can be performed by

2(𝑠 − 𝑟) − 1 interchanges of adjacent rows. As the number of interchanges is odd

and each one changes the sign of the determinant, the net effect is a change of

sign as desired.

We are now able to prove the cofactor expansion theorem 2.3.2

Proof: Let 𝐵 be the matrix obtained by moving the 𝑖𝑡ℎ row of 𝐴 to the top, using
𝑖−1 interchanges of adjacent rows. Thus 𝑑𝑒𝑡 (𝐵) = (−1)𝑖−1 det(𝐴), but
𝑏1𝑗 = 𝑎𝑖𝑗 and 𝐵1𝑗 = 𝐴𝑖𝑗 for 𝑗 ∈ [1, 𝑛] and so

𝑎𝑖1 ⋯ 𝑎𝑖𝑗 ⋯ 𝑎𝑖𝑛

𝑎11 ⋯ 𝑎1𝑗 ⋯ 𝑎1𝑛
| ⋮ ⋮ |
⋮ ⋱ ⋱
det (𝐵) = 𝑎(𝑖−1)1 ⋯ 𝑎(𝑖−1)𝑗 ⋯ 𝑎(𝑖−1)𝑛
𝑎(𝑖+1)1 ⋯ 𝑎(𝑖+1)𝑗 ⋯ 𝑎(𝑖+1)𝑛
| ⋱ ⋱ ⋮ |
⋮ ⋮
𝑎𝑚1 ⋯ 𝑎𝑚𝑗 ⋯ 𝑎𝑚𝑛

Hence,

𝑛
det(𝐴) = (−1)𝑖−1 det(𝐵) = (−1) ∑(−1)1+𝑗 𝑏
𝑖−1
1𝑗 det(𝐵1𝑗 )
𝑗=1

28
 𝑛 𝑛

= (−1)𝑖−1 ∑(−1) 1+𝑗

𝑎𝑖𝑗 det(𝐴𝑖𝑗 ) = ∑(−1)𝑖+𝑗 𝑎𝑖𝑗 det(𝐴𝑖𝑗 )
𝑗=1 𝑗=1

Giving the formula for cofactor expansion along the 𝑖𝑡ℎ row.

Corollary 2.4.2 [3] If any two rows of a horizontal matrix are identical, then the
value of its determinant is zero.

Proof: Let |𝐴| be the determinant of the horizontal matrix 𝐴. Assume that row 𝑖
and row 𝑗 in 𝐴 are identical. By Theorem 2.4.1 interchange row 𝑖 and row 𝑗, the
determinant of the resulting matrix is −|𝐴|. But the original matrix and the
resulting matrix are the same

That is |𝐴| = −|𝐴|. Hence, we obtain |𝐴| = 0.

This theorem represents property (b) in the theorem 2.1.3

The following theorem is from [3 ], [7 ], but we give here another proof

Theorem 2.4.3 Let 𝐴 and 𝐵 be 𝑚 × 𝑛 matrices with 𝑚 ≤ 𝑛, and 𝐵 is obtained

from 𝐴 by multiplying all the entries of some row of 𝐴 by a scalar 𝑘. Then

𝑑𝑒𝑡 (𝐵) = 𝑘 𝑑𝑒𝑡 (𝐴)

Proof: If we expand along the 𝑖𝑡ℎ row of 𝐵 to calculate its determinant, we get

det(𝐵) = 𝑏𝑖1 𝐵𝑖1 + ⋯ + (−1)𝑖+𝑛 𝑏𝑖𝑛 𝐵𝑖𝑛 .

But the reason we have chosen the 𝑖𝑡ℎ row of 𝐵 is that we know that 𝑏𝑖𝑗 = 𝑘 𝑎𝑖𝑗

for 𝑗 = 1 , … , 𝑛. Moreover, since the submatrices 𝐵(𝑖 , 𝑗 ) will all have row 𝑖

removed, and since this is the only place where 𝐵 differs from 𝐴, we see that

𝐴(𝑖 , 𝑗 ) = 𝐵(𝑖 , 𝑗 ). Thus, the cofactor 𝐵𝑖𝑗 for 𝑏𝑖𝑗 is the same as the cofactor 𝐴𝑖𝑗

for 𝑎𝑖𝑗 . So we have that

29
 𝑖+𝑛
det(𝐵) = 𝑘𝑎𝑖1 𝐵𝑖1 + ⋯ + (−1) 𝑘 𝑎𝑖𝑛 𝐵𝑖𝑛
= 𝑘 (𝑎𝑖1 𝐴𝑖1 + ⋯ + (−1)𝑖+𝑛 𝑎𝑖𝑛 𝐴𝑖𝑛 )
= 𝑘 det(𝐴)

If we know the determinant of matrix 𝐴, we can use this information to calculate the

determinant of the matrix 𝑘 𝐴, where k is a constant.

Corollary 2.4.4 Let 𝐴 and 𝐵 be an 𝑚 × 𝑛 matrices with 𝑚 ≤ 𝑛, and 𝐵 is obtained

from 𝐴 by multiplying all the entries of rows of 𝐴 by a scalar 𝑘. Then

det(𝑘 𝐴) = 𝑘𝑚 det(𝐴)

Proof: Since all 𝑚 rows of 𝐴 are multiplied by the scalar 𝑘 to get 𝑘 𝐴, using the
above theorem 𝑚 times gives

det(𝑘 𝐴) = (𝑘)(𝑘) … (𝑘) det(𝐴)

= 𝑘 𝑚 det(𝐴)

The following theorem is from [3 ], but we give here another proof

Theorem 2.4.5 Let 𝐴 and 𝐵 be an 𝑚 × 𝑛 matrices with 𝑚 ≤ 𝑛, and 𝐵 is obtained

from 𝐴 by adding a multiple of one row of 𝐴 to another row of 𝐴. Then

𝑑𝑒𝑡 (𝐵) = 𝑑𝑒𝑡 (𝐴)

𝑎11 𝑎12 ⋯ 𝑎1𝑛

⋮ ⋮ ⋱ ⋮
𝑎𝑠1 𝑎𝑠2 ⋯ 𝑎𝑠𝑛
Proof: Let 𝐴 = ⋮ ⋮ ⋱ ⋮ , and suppose that 𝐵 is the matrix obtained from
𝑎𝑟1 𝑎𝑟2 ⋯ 𝑎𝑟𝑛
⋮ ⋮ ⋱ ⋮
[𝑎𝑚1 𝑎𝑚2 ⋯ 𝑎𝑚𝑛 ]
𝐴 by adding 𝑘 times row 𝑠 to row 𝑟,
𝑎11 𝑎12 ⋯ 𝑎1𝑛
⋮ ⋮ ⋱ ⋮
𝑎𝑠1 𝑎𝑠2 ⋯ 𝑎𝑠𝑛
𝐵= ⋮ ⋮ ⋱ ⋮
𝑘𝑎𝑠1 + 𝑎𝑟1 𝑘𝑎𝑠2 + 𝑎𝑟2 ⋯ 𝑘𝑎𝑠2 + 𝑎𝑟𝑛
⋮ ⋮ ⋱ ⋮
[ 𝑎𝑚1 𝑎𝑚2 ⋯ 𝑎𝑚𝑛 ]

30
Then we can compute the determinant of 𝐵 by expanding along row 𝑟, getting

det(𝐵) = 𝑏𝑟1 𝐶𝑟1 + ⋯ + (−1)𝑟+𝑛 𝑏𝑟𝑛 𝐶𝑟𝑛

Our choice of 𝑟 gets us that 𝑏𝑟𝑗 = 𝑘𝑎𝑠𝑗 + 𝑎𝑟𝑗 for all 𝑗 = 1, … , 𝑛. So, now let's

consider the submatrices 𝐵(𝑟 , 𝑗 ) and 𝐴(𝑟 , 𝑗 ) are the cofactor of 𝐴 and 𝐵. This

means that 𝐵(𝑟 , 𝑗 ) can obtained from 𝐴(𝑟 , 𝑗 ) by adding 𝑘 times row 𝑠 to row 𝑟.

And since 𝐵(𝑟 , 𝑗 ) and 𝐴(𝑟 , 𝑗 ) are ( 𝑚 − 1) × (𝑛 − 1) matrices, we get

𝑑𝑒𝑡 𝐵(𝑟 , 𝑗 ) = det 𝐴(𝑟 , 𝑗 ). That is,

𝑑𝑒𝑡 (𝐵) = 𝑏𝑟1 (−1)𝑟+1 𝑑𝑒𝑡𝐵(𝑟 , 1 ) + ⋯ + 𝑏𝑟𝑛 (−1)𝑟+𝑛 𝑑𝑒𝑡𝐵(𝑟 , 𝑛 )

= (𝑘𝑎𝑠1 + 𝑎𝑟1 )(−1)𝑟+1 𝑑𝑒𝑡𝐴(𝑟 , 1 ) + ⋯ +

+(𝑘𝑎𝑠𝑛 + 𝑎𝑟𝑛 )(−1)𝑟+𝑛 𝑑𝑒𝑡𝐴(𝑟 , 𝑛 )

det(𝐵) = 𝑘𝑎𝑠1 (−1)𝑟+1 𝑑𝑒𝑡𝐴(𝑟 , 1 ) + 𝑎𝑟1 (−1)𝑟+1 𝑑𝑒𝑡𝐴(𝑟 , 1 ) + ⋯

+ 𝑘𝑎𝑠𝑛 (−1)𝑟+𝑛 𝑑𝑒𝑡𝐴(𝑟 , 𝑛 ) + 𝑎𝑟𝑛 (−1)𝑟+𝑛 𝑑𝑒𝑡𝐴(𝑟 , 𝑛 )

= (𝑘𝑎𝑠1 (−1)𝑟+1 𝑑𝑒𝑡𝐴(𝑟 , 1 ) + ⋯ + k 𝑎𝑠𝑛 (−1)𝑟+𝑛 𝑑𝑒𝑡𝐴(𝑟 , 𝑛 ))

+ 𝑎𝑟1 (−1)𝑟+1 𝑑𝑒𝑡𝐴(𝑟 , 1 ) + ⋯ + 𝑎𝑟𝑛 (−1)𝑟+𝑛 𝑑𝑒𝑡𝐴(𝑟 , 𝑛 )

𝑎11 𝑎12 ⋯ 𝑎1𝑛 𝑎11 𝑎12 ⋯ 𝑎1𝑛

⋮ ⋮ ⋱ ⋮ ⋮ ⋮ ⋱ ⋮
| 𝑎𝑠1 𝑎𝑠2 ⋯ 𝑎𝑠𝑛 | | 𝑎𝑠1 𝑎𝑠2 ⋯ 𝑎𝑠𝑛 |
det(𝐵) = ⋮ ⋮ ⋱ ⋮ + ⋮ ⋮ ⋱ ⋮ ,
|
𝑘𝑎 𝑠1 𝑘𝑎 𝑠2 ⋯ 𝑘𝑎𝑠2 | |
𝑎𝑟1 𝑎𝑟2 ⋯ 𝑎𝑟𝑛 |
⋮ ⋮ ⋱ ⋮ ⋮ ⋮ ⋱ ⋮
𝑎𝑚1 𝑎𝑚2 ⋯ 𝑎𝑚𝑛 𝑎𝑚1 𝑎𝑚2 ⋯ 𝑎𝑚𝑛

any matrix in which one row is a multiple of another has determinant

zero, thus, det(𝐵) = 0 + det(𝐴)
det(𝐵) = det(𝐴)

This theorem represents property (d) in the theorem 2.1.3

31
 Since we know how elementary row operations affect the determinant, we can
compute the determinant of a given matrix by computing the determinant of its
r.r.e.f (see definition 1.7.1) and taking into account the effect of the row
operations. The same procedure that we used in books of linear algebra for
determinant of square matrices.
The following table describes the effect of applying row operations on
computing the determinant of a horizontal matrix.

Type of ERO Effect on determinant

1 Add a multiple of one row to another row No effect
2 Multiply a row by a constant k Determinate is multiplied by k
3 Interchange two rows Determinant changes sign

We mention here that these properties correspond to their counter parts for
determinants of square matrices (See Theorem 1.3.5)
2 4 6
Example 2.4.6 Find the determinant of 𝐴 = [ ]
1 3 5

Solution: We use row reduction until 𝐴 is in reduced row echelon form. At each

step we keep track of the effect on the determinant.

2 4 6 𝑅1 ↔𝑅2 : det×(−1) 1 3 5
| | → −| |
1 3 5 2 4 6

1 1
R2 −2𝑅1 →R2 ∶det 𝑢𝑛𝑐ℎ𝑎𝑛𝑔𝑒𝑑 (− )𝑅2 : 𝑑𝑒𝑡(− )
− |1 3 5 1 3 5
2 2
→ | → −| |
0 −2 −4 0 1 2

R1 − 3𝑅2 →R1 ∶det 𝑢𝑛𝑐ℎ𝑎𝑛𝑔𝑒𝑑 1 1 0 −1

→ | |
2 0 1 2
1 1 0 −1 1
| | = 2 (1|1 2| − |0 1|)
2 0 1 2
1
= ((−1) − (−1) ) = 0
2

32
Note that, the definition of the determinant (see theorem 2.4.1, 2.4.3, 3.1.1) satisfies the

axioms of determinant function 2.1.2 (*).

Also, the cofactor definition of the determinant (see theorem 2.3.5, 2.4.2, 2.4.5) satisfies

the properties of determinant function 2.1.3.

2.5 Radic's determinant

Many definitions have been proposed for the determinant of non-square

matrices. Earlier works have been mainly focused on utilizing the determinant of
square blocks to define the determinant of the non-square matrix. They studied
many useful properties of this determinant. Radic (1969) proposed the following
efficient definition that has some of the major properties of the determinants of
square matrices.

Definition 2.5.1 [14 ]. Let 𝐴 = (𝑎𝑖𝑗 ) be an 𝑚 × 𝑛 matrix with 𝑚 ≤ 𝑛. The

determinant of 𝐴 is defined as

𝑎1𝑗1 ⋯ 𝑎1𝑗𝑚
𝑟+𝑆
det(𝐴) = ∑ (−1) 𝑑𝑒𝑡 [ ⋮ ⋱ ⋮ ] (2.5.1)
1≤𝑗1 <⋯<𝑗𝑚 ≤𝑛 𝑎𝑚𝑗1 ⋯ 𝑎𝑚𝑗𝑚

Where 𝑗1 , 𝑗2 , … , 𝑗𝑚 ∈ 𝑁, 𝑟 = 1 + 2 + ⋯ + 𝑚 and 𝑠 = 𝑗1 + 𝑗2 + ⋯ + 𝑗𝑚 .

If 𝑚 > 𝑛, then 𝑑𝑒𝑡(𝐴) = 𝑑𝑒𝑡(𝐴𝑇 ) .

The determinant of a square matrix and the determinant 2.5.1 of a 𝑚 × 𝑛 matrix,

where 𝑚 ≤ 𝑛, have several common standard properties, including the following:

(1) If a row of matrix 𝐴 is a linear combination of some other rows, then

𝑑𝑒𝑡(𝐴) = 0
(2) If a row of A is multiplied by a number 𝑘, then the determinant of the
resulting matrix is equal to 𝑘. 𝑑𝑒𝑡(𝐴).
(3) Interchanging two rows of 𝐴 results in changing the sign of the
determinant.
(4) If the matrix 𝐴 has two identical rows, then 𝑑𝑒𝑡(𝐴) = 0.

33
Proof: Let 𝐴 be an 𝑚 × 𝑛 matrix with 𝑚 ≤ 𝑛, by Radic's definition

𝑎1𝑗1 ⋯ 𝑎1𝑗𝑚
det(𝐴) = ∑ (−1)𝑟+𝑆 𝑑𝑒𝑡 [ ⋮ ⋱ ⋮ ]
1≤𝑗1 <⋯<𝑗𝑚 ≤𝑛 𝑎𝑚𝑗1 ⋯ 𝑎𝑚𝑗𝑚

Where 𝑗1 , 𝑗2 , … , 𝑗𝑚 ∈ 𝑁, 𝑟 = 1 + 2 + ⋯ + 𝑚 and 𝑠 = 𝑗1 + 𝑗2 + ⋯ + 𝑗𝑚 .

(1) If a row of matrix 𝐴 is a linear combination of some other rows, then all

𝑎1𝑗1 ⋯ 𝑎1𝑗𝑚
𝑑𝑒𝑡 [ ⋮ ⋱ ⋮ ] contains a row is a linear combination of some other rows, and
𝑎𝑚𝑗1 ⋯ 𝑎𝑚𝑗𝑚

𝑎1𝑗1 ⋯ 𝑎1𝑗𝑚
therefore, all 𝑑𝑒𝑡 [ ⋮ ⋱ ⋮ ] = 0, where 𝑗1 , 𝑗2 , … , 𝑗𝑚 ∈ 𝑁, are square matrices,
𝑎𝑚𝑗1 ⋯ 𝑎𝑚𝑗𝑚

hence 𝑑𝑒𝑡(𝐴) = 0

𝑎11 𝑎12
⋯ 𝑎1𝑛
⋮ ⋮
⋱ ⋮
𝑎
(2) Let 𝐴 = 𝑖1 𝑎𝑖2
⋯ 𝑎𝑖𝑛 , and suppose that 𝐵 is the matrix obtained
⋮ ⋮
⋱ ⋮
[𝑎𝑚1 𝑎𝑚2
⋯ 𝑎𝑚𝑛 ]
𝑎11 𝑎12 ⋯ 𝑎1𝑛
⋮ ⋮ ⋱ ⋮
from 𝐴 by multiplying row 𝑖 by 𝑘 , 𝐵 = 𝑘𝑎𝑖1 𝑘𝑎𝑖2 ⋯ 𝑘𝑎𝑖𝑛 then,
⋮ ⋮ ⋱ ⋮
[ 𝑎𝑚1 𝑎𝑚2 ⋯ 𝑎𝑚𝑛 ]

𝑎1𝑗1 𝑎1𝑗2 ⋯ 𝑎1𝑗𝑚

⋮ ⋮ ⋱ ⋮
det(𝐵) = ∑ (−1)𝑟+𝑆 𝑑𝑒𝑡 𝑘𝑎 𝑖𝑗1 𝑘𝑎 𝑖𝑗2 ⋯ 𝑘𝑎 𝑖𝑗𝑚
1≤𝑗1 <⋯<𝑗𝑚 ≤𝑛 ⋮ ⋮ ⋱ ⋮
[ 𝑎 𝑎 ⋯ 𝑎
𝑚𝑗 1 𝑚𝑗 2 𝑚𝑗 𝑚]

𝑎1𝑗1 𝑎1𝑗2 ⋯ 𝑎1𝑗𝑚

⋮ ⋮ ⋱ ⋮
Since all 𝑑𝑒𝑡 𝑘𝑎 𝑖𝑗1 𝑘𝑎𝑖𝑗2 ⋯ 𝑘𝑎𝑖𝑗𝑚 where 𝑗 , 𝑗 , … , 𝑗 ∈ 𝑁 are square matrices,
1 2 𝑚
⋮ ⋮ ⋱ ⋮
[ 𝑎𝑚𝑗1 𝑎𝑚𝑗2 ⋯ 𝑎𝑚𝑗𝑚 ]

34
 𝑎1𝑗1 𝑎1𝑗2 ⋯ 𝑎1𝑗𝑚 𝑎1𝑗1 𝑎1𝑗2 ⋯ 𝑎1𝑗𝑚
⋮ ⋮ ⋱ ⋮ ⋮ ⋮ ⋱ ⋮
𝑑𝑒𝑡 𝑘𝑎 𝑖𝑗1 𝑘𝑎𝑖𝑗2 ⋯ 𝑘𝑎𝑖𝑗𝑚 = 𝑘𝑑𝑒𝑡 𝑎𝑖𝑗1 𝑎𝑖𝑗2 ⋯ 𝑎𝑖𝑗𝑚 , 𝑗1 , 𝑗2 , … , 𝑗𝑚 ∈ 𝑁
⋮ ⋮ ⋱ ⋮ ⋮ ⋮ ⋱ ⋮
[ 𝑎𝑚𝑗1 𝑎𝑚𝑗2 ⋯ 𝑎𝑚𝑗𝑚 ] 𝑎
[ 𝑚𝑗1 𝑎𝑚𝑗2 ⋯ 𝑎𝑚𝑗𝑚 ]

can be taken 𝑘 out from all determinant square in common to produce 𝑑𝑒𝑡(𝐵) = 𝑘𝑑𝑒𝑡(𝐴).

𝑎11 𝑎12 ⋯ 𝑎1𝑛

⋮ ⋮ ⋱ ⋮
𝑎𝑟1 𝑎𝑟2 ⋯ 𝑎𝑟𝑛
(3) Let 𝐴 = ⋮ ⋮ ⋱ ⋮ , and suppose that 𝐵 is the matrix
𝑎𝑠1 𝑎𝑠2 ⋯ 𝑎𝑠𝑛
⋮ ⋮ ⋱ ⋮
[𝑎𝑚1 𝑎𝑚2 ⋯ 𝑎𝑚𝑛 ]
𝑎11 𝑎12 ⋯ 𝑎1𝑛
⋮ ⋮ ⋱ ⋮
𝑎𝑠1 𝑎𝑠2 ⋯ 𝑎𝑠𝑛
obtained Interchanging two rows of 𝐴, 𝐵= ⋮ ⋮ ⋱ ⋮
𝑎𝑟1 𝑎𝑟2 ⋯ 𝑎𝑟𝑛
⋮ ⋮ ⋱ ⋮
[𝑎𝑚1 𝑎𝑚2 ⋯ 𝑎𝑚𝑛 ]

then,
𝑎1𝑗1 𝑎1𝑗2 ⋯ 𝑎1𝑗𝑚
⋮ ⋮ ⋱ ⋮
𝑎𝑠𝑗1 𝑎𝑠𝑗2 𝑎𝑠𝑗𝑚
⋯
det(𝐵) = ∑ (−1)𝑟+𝑆 𝑑𝑒𝑡 ⋮ ⋮ ⋱ ⋮ ,
𝑎𝑟𝑗1 𝑎𝑟𝑗2 ⋯ 𝑎𝑟𝑗𝑚
1≤𝑗1 <⋯<𝑗𝑚 ≤𝑛
⋮ ⋮ ⋱ ⋮
𝑎 𝑎 ⋯ 𝑎𝑚𝑗𝑚 ]
[ 𝑚𝑗1 𝑚𝑗2

𝑎1𝑗1 𝑎1𝑗2 ⋯ 𝑎1𝑗𝑚

⋮ ⋮ ⋱ ⋮
𝑎𝑠𝑗1 𝑎𝑠𝑗2 ⋯ 𝑎𝑠𝑗𝑚
Since all 𝑑𝑒𝑡 ⋮ ⋮ ⋱ ⋮ where 𝑗1 , 𝑗2 , … , 𝑗𝑚 ∈ 𝑁 are square matrices,
𝑎𝑟𝑗1 𝑎𝑟𝑗2 ⋯ 𝑎𝑟𝑗𝑚
⋮ ⋮ ⋱ ⋮
[𝑎𝑚𝑗1 𝑎𝑚𝑗2 ⋯ 𝑎𝑚𝑗𝑚 ]

𝑎1𝑗1 𝑎1𝑗2 ⋯ 𝑎1𝑗𝑚 𝑎1𝑗1 𝑎1𝑗2 ⋯ 𝑎1𝑗𝑚

⋮ ⋮ ⋱ ⋮ ⋮ ⋮ ⋱ ⋮
𝑎𝑠𝑗1 𝑎𝑠𝑗2 ⋯ 𝑎𝑠𝑗𝑚 𝑎𝑟𝑗1 𝑎𝑟𝑗2 ⋯ 𝑎𝑟𝑗𝑚
𝑑𝑒𝑡 ⋮ ⋮ ⋱ ⋮ = −𝑑𝑒𝑡 ⋮ ⋮ ⋱ ⋮ , 𝑗1 , 𝑗2 , … , 𝑗𝑚 ∈ 𝑁
𝑎𝑟𝑗1 𝑎𝑟𝑗2 ⋯ 𝑎𝑟𝑗𝑚 𝑎𝑠𝑗1 𝑎𝑠𝑗2 ⋯ 𝑎𝑠𝑗𝑚
⋮ ⋮ ⋱ ⋮ ⋮ ⋮ ⋱ ⋮
[𝑎𝑚𝑗1 𝑎𝑚𝑗2 ⋯ 𝑎𝑚𝑗𝑚 ] [𝑎𝑚𝑗1 𝑎𝑚𝑗2 ⋯ 𝑎𝑚𝑗𝑚 ]

35
 can be taken sign negative out from all determinant square in common to produce

𝑑𝑒𝑡(𝐵) = −𝑑𝑒𝑡(𝐴).

𝑎11 𝑎12 ⋯ 𝑎1𝑛

⋮ ⋮ ⋱ ⋮
𝑎𝑟1 𝑎𝑟2 ⋯ 𝑎𝑟𝑛
(4) Let 𝐴 = ⋮ ⋮ ⋱ ⋮ , then
𝑎𝑟1 𝑎𝑟2 ⋯ 𝑎𝑟𝑛
⋮ ⋮ ⋱ ⋮
[𝑎𝑚1 𝑎𝑚2 ⋯ 𝑎𝑚𝑛 ]
𝑎1𝑗1 𝑎1𝑗2 ⋯ 𝑎1𝑗𝑚
⋮ ⋮ ⋱ ⋮
𝑎𝑟𝑗1 𝑎𝑟𝑗2 𝑎𝑟𝑗𝑚
⋯
det(𝐴) = ∑ (−1)𝑟+𝑆 𝑑𝑒𝑡 ⋮ ⋮ ⋱ ⋮
𝑎𝑟𝑗1 𝑎𝑟𝑗2 ⋯ 𝑎𝑟𝑗𝑚
1≤𝑗1 <⋯<𝑗𝑚 ≤𝑛
⋮ ⋮ ⋱ ⋮
𝑎 𝑎 ⋯ 𝑎𝑚𝑗𝑚 ]
[ 𝑚𝑗1 𝑚𝑗2

𝑎1𝑗1 𝑎1𝑗2 ⋯ 𝑎1𝑗𝑚

⋮ ⋮ ⋱ ⋮
𝑎𝑟𝑗1 𝑎𝑟𝑗2 ⋯ 𝑎𝑟𝑗𝑚
Since all 𝑑𝑒𝑡 ⋮ ⋮ ⋱ ⋮ where 𝑗1 , 𝑗2 , … , 𝑗𝑚 ∈ 𝑁 are square matrices,
𝑎𝑟𝑗1 𝑎𝑟𝑗2 ⋯ 𝑎𝑟𝑗𝑚
⋮ ⋮ ⋱ ⋮
[𝑎𝑚𝑗1 𝑎𝑚𝑗2 ⋯ 𝑎𝑚𝑗𝑚 ]

𝑎1𝑗1 𝑎1𝑗2 ⋯ 𝑎1𝑗𝑚

⋮ ⋮ ⋱ ⋮
𝑎𝑟𝑗1 𝑎𝑟𝑗2 ⋯ 𝑎𝑟𝑗𝑚
then all 𝑑𝑒𝑡 ⋮ ⋮ ⋱ ⋮ = 0, hence 𝑑𝑒𝑡(𝐴) = 0
𝑎𝑟𝑗1 𝑎𝑟𝑗2 ⋯ 𝑎𝑟𝑗𝑚
⋮ ⋮ ⋱ ⋮
[𝑎𝑚𝑗1 𝑎𝑚𝑗2 ⋯ 𝑎𝑚𝑗𝑚 ]

(5) We need to show that radic definition of the determinant satisfies the axioms
(A3) of determinant function
𝑎11 𝑎12 ⋯ 𝑎1𝑛 𝑏11 𝑏12 ⋯ 𝑏1𝑛
𝑎21 𝑎22 ⋯ 𝑎2𝑛 𝑎21 𝑎22 ⋯ 𝑎2𝑛
Let 𝐴 = [ ⋮ ⋱ ⋮ ],
⋮ 𝐵=[ ],
⋮ ⋮ ⋱ ⋮
𝑎𝑚1 𝑎𝑚2 ⋯ 𝑎𝑚𝑛 𝑎𝑚1 𝑎𝑚2 ⋯ 𝑎𝑚𝑛
𝑎11 + 𝑏11 𝑎12 + 𝑏12 ⋯ 𝑎1𝑛 + 𝑏1𝑛
𝑎21 𝑎22 ⋯ 𝑎2𝑛
𝐶=[ ]
⋮ ⋮ ⋱ ⋮
𝑎𝑚1 𝑎𝑚2 ⋯ 𝑎𝑚𝑛

36
Then,
𝑎1𝑗1 + 𝑏 𝑎1𝑗2 + 𝑏 ⋯ 𝑎1𝑗 + 𝑏1𝑗
1𝑗1 1𝑗2 𝑚 𝑚
𝑎2𝑗1 𝑎2𝑗2 ⋯ 𝑎2𝑗𝑚
det(𝐶) = ∑ (−1)𝑟+𝑆 𝑑𝑒𝑡
⋮ ⋮ ⋱ ⋮
1≤𝑗1 <⋯<𝑗𝑚 ≤𝑛
[ 𝑎𝑚𝑗1 𝑎𝑚𝑗2 ⋯ 𝑎 𝑚𝑗𝑚 ]

𝑎1𝑗1 + 𝑏1𝑗1 𝑎1𝑗2 + 𝑏1𝑗2 ⋯ 𝑎1𝑗𝑚 + 𝑏1𝑗𝑚

𝑎2𝑗1 𝑎2𝑗2 ⋯ 𝑎2𝑗𝑚
Since all 𝑑𝑒𝑡 where 𝑗1 , 𝑗2 , … , 𝑗𝑚 ∈ 𝑁 are
⋮ ⋮ ⋱ ⋮
[ 𝑎𝑚𝑗1 𝑎𝑚𝑗2 ⋯ 𝑎𝑚𝑗𝑚 ]
square matrices,

𝑎1𝑗1 + 𝑏1𝑗1 𝑎1𝑗2 + 𝑏1𝑗2 ⋯ 𝑎1𝑗𝑚 + 𝑏1𝑗𝑚

𝑎2𝑗1 𝑎2𝑗2 ⋯ 𝑎2𝑗𝑚
Therefore, 𝑑𝑒𝑡
⋮ ⋮ ⋱ ⋮
[ 𝑎𝑚𝑗1 𝑎𝑚𝑗2 ⋯ 𝑎𝑚𝑗𝑚 ]

𝑎1𝑗1 𝑎1𝑗2 ⋯ 𝑎1𝑗𝑚 𝑏1𝑗1 𝑏1𝑗2 ⋯ 𝑏1𝑗𝑚

𝑎2𝑗1 𝑎2𝑗2 ⋯ 𝑎2𝑗𝑚 𝑎2𝑗1 𝑎2𝑗2 ⋯ 𝑎2𝑗𝑚
= 𝑑𝑒𝑡 [ ⋮ ⋮ ⋮ ] + 𝑑𝑒𝑡 ⋮
⋱ ⋮ ⋱ ⋮
𝑎𝑚𝑗1 𝑎𝑚𝑗2 ⋯ 𝑎𝑚𝑗𝑚 [𝑎𝑚𝑗1 𝑎𝑚𝑗2 ⋯ 𝑎𝑚𝑗𝑚 ]

Hence, det(𝐶) =
𝑎1𝑗1 𝑎1𝑗2 ⋯ 𝑎1𝑗𝑚 𝑏1𝑗1 𝑏1𝑗2 ⋯ 𝑏1𝑗𝑚
𝑎2𝑗1 𝑎2𝑗2 ⋯ 𝑎2𝑗𝑚 𝑎2𝑗1 𝑎2𝑗2 ⋯ 𝑎2𝑗𝑚
∑ (−1)𝑟+𝑆 ( 𝑑𝑒𝑡 + 𝑑𝑒𝑡 )
⋮ ⋮ ⋱ ⋮ ⋮ ⋮ ⋱ ⋮
1≤𝑗1 <⋯<𝑗𝑚 ≤𝑛
[𝑎𝑚𝑗1 𝑎𝑚𝑗2 ⋯ 𝑎𝑚𝑗𝑚 ] [𝑎𝑚𝑗1 𝑎𝑚𝑗2 ⋯ 𝑎𝑚𝑗𝑚 ]

= det(𝐴) + det(𝐵)

Note that, the definition of the determinant (see properties 2,3,5) satisfies the
axioms of determinant function 2.1.2 (**).

Also, the Radic definition of the determinant (see properties 1.4 ) satisfies the

properties of determinant function 2.1.3.

37
Corollary 2.5.2 The determinant obtained by cofactor expansion and Radic
definition are the same.

Proof: the cofactor definition and Radic definition are determinant function, since
the four characterizing properties of determinant listed in definition 2.1.2 are
satisfied by the cofactor definition 2.3.1 and Radic definition 2.5.1 of
determinants, and because of uniqueness of determinant function (see theorem
2.1.4), the cofactor definition and Radic definition are the same. ( see * , **)

Example 2.5.3 Evaluate the determinant of 𝐴 = [𝑎1 𝑎2 𝑎3 ] using Radic definition

Solution: |𝐴| = (−1)1+1 𝑎1 + (−1)1+2 𝑎2 + (−1)1+3 𝑎3

= 𝑎1 − 𝑎2 + 𝑎3

𝑎1 𝑎2 𝑎3
Example 2.5.4 Evaluate the determinant of 𝐴 = [𝑏 𝑏2 𝑏3 ] using Radic definition
1

Solution :

𝑎1 𝑎2 𝑎1 𝑎3 𝑎2 𝑎3
|𝐴| = (−1)(3)+(1+2) |𝑏 | (−1) (3)+(1+3)
| | (−1) (3)+(2+3)
|
1 𝑏2 + 𝑏1 𝑏3 + 𝑏2 𝑏3 |

𝑎1 𝑎2 𝑎1 𝑎3 𝑎2 𝑎3
= |𝑏 𝑏2 | − | 𝑏1 𝑏3 | + | 𝑏2 𝑏3 |
1

The evaluation of the determinant of an 𝑚×𝑛 matrix (𝑚 ≤ 𝑛) using Radic's

𝑛
definition reduces to evaluation of (𝑚 ) determinant of 𝑚 × 𝑚 matrices.

Note, Let A1 , A2 , … , An be columns of the matrix 𝐴𝑚×𝑛 (𝑚 ≤ 𝑛). Then Radic’s

determinant of 𝐴 is a function in the columns of 𝐴 and can be written in the form

𝑑𝑒𝑡(𝐴) = 𝑑𝑒𝑡(𝐴1 , … , 𝐴𝑛 ) = |𝐴1 , … , 𝐴𝑛 |

38
Theorem 2.5.5 [13 ] Let 𝐴 = [𝐴1 , … , 𝐴𝑛 ] be 2 × 𝑛 matrix with 𝑛 ≥ 2. Then

𝑑𝑒𝑡(𝐴1 , … , 𝐴𝑛 ) = 𝑑𝑒𝑡(𝐴1 , 𝐴2 − 𝐴3 + 𝐴4 … + (−1)𝑛 𝐴𝑛 ) +

𝑑𝑒𝑡(𝐴2 , 𝐴3 − 𝐴4 + … + (−1)𝑛−1 𝐴𝑛 ) + ⋯ +

𝑑𝑒𝑡(𝐴𝑛−1 , 𝐴𝑛 )

Proof: By principle of mathematical induction, (P.M.I)

Base case, 𝑛 = 3,

𝑑𝑒𝑡(𝐴1 , 𝐴2 , 𝐴3 ) = 𝑑𝑒𝑡(𝐴1 , 𝐴2 ) − 𝑑𝑒𝑡(𝐴1 , 𝐴3 ) + 𝑑𝑒𝑡(𝐴2 , 𝐴3 )

Since 𝑑𝑒𝑡(𝐴1 , 𝐴2 ), 𝑑𝑒𝑡(𝐴1 , 𝐴3 ) are square then

𝑑𝑒𝑡(𝐴1 , 𝐴2 ) − 𝑑𝑒𝑡(𝐴1 , 𝐴3 ) = 𝑑𝑒𝑡(𝐴1 , 𝐴2 − 𝐴3 )

Then, 𝑑𝑒𝑡(𝐴1 , 𝐴2 , 𝐴3 ) = 𝑑𝑒𝑡(𝐴1 , 𝐴2 − 𝐴3 ) + 𝑑𝑒𝑡(𝐴2 , 𝐴3 )

Induction hypothesis: We assume that it is true for 𝑛 = 𝑘

𝑑𝑒𝑡(𝐴1 , … , 𝐴𝑘 ) = 𝑑𝑒𝑡(𝐴1 , 𝐴2 − 𝐴3 + 𝐴4 … + (−1)𝑘 𝐴𝑘 ) +

𝑑𝑒𝑡(𝐴2 , 𝐴3 − 𝐴4 + … + (−1)𝑘−1 𝐴𝑘 ) + ⋯ + 𝑑𝑒𝑡(𝐴𝑘−1 , 𝐴𝑘 )

We will show that the identity holds for 𝑛 = 𝑘 + 1

𝑑𝑒𝑡(𝐴1 , … , 𝐴𝑘+1 ) = (−1)(3)+(3) 𝑑𝑒𝑡(𝐴1 , 𝐴2 ) + (−1)(3)+(4) 𝑑𝑒𝑡(𝐴1 , 𝐴3 ) +

(−1)(3)+(5) 𝑑𝑒𝑡(𝐴1 , 𝐴4 ) + (−1)(3)+(6) 𝑑𝑒𝑡(𝐴1 , 𝐴5 ) + ⋯ +

(−1)(3)+(1+𝑘+1) 𝑑𝑒𝑡(𝐴1 , 𝐴𝑘+1 ) + (−1)(3)+(5) 𝑑𝑒𝑡(𝐴2 , 𝐴3 ) +

(−1)(3)+(6) 𝑑𝑒𝑡(𝐴2 , 𝐴4 ) + (−1)(3)+(7) 𝑑𝑒𝑡(𝐴2 , 𝐴5 ) + ⋯ +

(−1)(3)+(2+𝑘+1) 𝑑𝑒𝑡(𝐴2 , 𝐴𝑘+1 ) + ⋯ + (−1)(3)+(𝑘+𝑘+1) 𝑑𝑒𝑡(𝐴𝑘 , 𝐴𝑘+1 )

= 𝑑𝑒𝑡(𝐴1 , 𝐴2 − 𝐴3 + 𝐴4 … + (−1)𝑘+1 𝐴𝑘+1 )

+ 𝑑𝑒𝑡(𝐴2 , 𝐴3 − 𝐴4 + … + (−1)𝑘 𝐴𝑘+1 ) + ⋯ + 𝑑𝑒𝑡(𝐴𝑘 , 𝐴𝑘+1 )

39
 Theorem 2.5.5 converts the computations of the determinant of an 2 × 𝑛
matrix according to Radic definition which needs computing (𝑛2) determinants to
computation of 𝑛 − 1 determinants of size 2 × 2.

𝑛(𝑛−1)
( (𝑛2) = > 𝑛−1 )
2

1 3 5
Example 2.5.6 Evaluate the determinant of 𝐴= [ ] using theorem 2.5.5
2 4 6

1 −2 3 5
Then, 𝑑𝑒𝑡(𝐴) = | |+ | | = (−2 + 4 ) + ( 18 − 20 ) = 0 .
2 −2 4 6

Theorem 2.5.7 [13 ] Let 𝐴 = [𝐴1 , … , 𝐴𝑛 ] be 2 × 𝑛 matrix with 𝑛 ≥ 2. Then

𝑑𝑒𝑡(𝐴1 , 𝐴2 , … , 𝐴𝑛−1 , 𝐴𝑛 ) = 𝑑𝑒𝑡(𝐴1 , 𝐴2 , … , 𝐴𝑛−1 ) +

(−1)𝑛 𝑑𝑒𝑡(𝐴1 − 𝐴2 + … + (−1)𝑛 𝐴𝑛−1 , 𝐴𝑛 )

Proof: By principle of mathematical induction, (P.M.I)

Base case: 𝑛 = 3,

𝑑𝑒𝑡(𝐴1 , 𝐴2 , 𝐴3 ) = 𝑑𝑒𝑡(𝐴1 , 𝐴2 ) − 𝑑𝑒𝑡(𝐴1 , 𝐴3 ) + 𝑑𝑒𝑡(𝐴2 , 𝐴3 )

= 𝑑𝑒𝑡(𝐴1 , 𝐴2 ) − 𝑑𝑒𝑡(𝐴1 − 𝐴2 , 𝐴3 )

Induction hypothesis: We assume that it is true for 𝑛 = 𝑘

𝑑𝑒𝑡(𝐴1 , … , 𝐴𝑘 )

= (−1)3+3 𝑑𝑒𝑡(𝐴1 , 𝐴2 ) + (−1)3+4 𝑑𝑒𝑡(𝐴1 , 𝐴3 ) + (−1)3+5 𝑑𝑒𝑡(𝐴1 , 𝐴4 )

+ ⋯ + (−1)4+𝑘 𝑑𝑒𝑡(𝐴1 , 𝐴𝑘 ) + (−1)3+5 𝑑𝑒𝑡(𝐴2 , 𝐴3 )

+ (−1)3+6 𝑑𝑒𝑡(𝐴2 , 𝐴4 ) + (−1)3+7 𝑑𝑒𝑡(𝐴2 , 𝐴5 ) + ⋯

+ (−1)5+𝑘 𝑑𝑒𝑡(𝐴2 , 𝐴𝑘 ) + ⋯ + (−1)2+2𝑘 𝑑𝑒𝑡(𝐴𝑘−1 , 𝐴𝑘 )

= 𝑑𝑒𝑡(𝐴1 , 𝐴2 , 𝐴3 , … , 𝐴𝑘−1 ) + (−1)𝑘 𝑑𝑒𝑡(𝐴1 − 𝐴2 + … + (−1)𝑘 𝐴𝑘−1 , 𝐴𝑘 )

40
We will show that the equality holds for 𝑛 = 𝑘 + 1

𝑑𝑒𝑡(𝐴1 , … , 𝐴𝑘+1 ) = (−1)3+3 𝑑𝑒𝑡(𝐴1 , 𝐴2 ) + (−1)3+4 𝑑𝑒𝑡(𝐴1 , 𝐴3 ) +

(−1)3+5 𝑑𝑒𝑡(𝐴1 , 𝐴4 ) + ⋯ + (−1)4+𝑘 𝑑𝑒𝑡(𝐴1 , 𝐴𝑘 ) + (−1)3+𝑘 𝑑𝑒𝑡(𝐴1 , 𝐴𝑘+1 ) +

+(−1)3+5 𝑑𝑒𝑡(𝐴2 , 𝐴3 ) + (−1)3+6 𝑑𝑒𝑡(𝐴2 , 𝐴4 ) + (−1)3+7 𝑑𝑒𝑡(𝐴2 , 𝐴5 ) +

⋯ + (−1)5+𝑘 𝑑𝑒𝑡(𝐴2 , 𝐴𝑘 ) + (−1)6+𝑘 𝑑𝑒𝑡(𝐴2 , 𝐴𝑘+1 ) + ⋯ +

(−1)4+2𝑘 𝑑𝑒𝑡(𝐴𝑘 , 𝐴𝑘+1 )

= 𝑑𝑒𝑡(𝐴1 , 𝐴2 , 𝐴3 , … , 𝐴𝑘−1 ) + (−1)𝑘 𝑑𝑒𝑡(𝐴1 − 𝐴2 + … + (−1)𝑘 𝐴𝑘−1 , 𝐴𝑘 )

+ (−1)𝑘+1 𝑑𝑒𝑡(𝐴1 , 𝐴𝑘+1 ) + (−1)𝑘+1 𝑑𝑒𝑡(𝐴2 , 𝐴𝑘+1 ) + ⋯

+ 𝑑𝑒𝑡(𝐴𝑘 , 𝐴𝑘+1 )

= 𝑑𝑒𝑡(𝐴1 , 𝐴2 , 𝐴3 , … , 𝐴𝑘−1 ) + (−1)𝑘 𝑑𝑒𝑡(𝐴1 − 𝐴2 + … + (−1)𝑘 𝐴𝑘−1 , 𝐴𝑘 )

+ (−1)𝑘+1 𝑑𝑒𝑡(𝐴1 − 𝐴2 + … + (−1)𝑘+1 𝐴𝑘 , 𝐴𝑘+1 )

𝑑𝑒𝑡(𝐴1 , … , 𝐴𝑘+1 ) = 𝑑𝑒𝑡(𝐴1 , 𝐴2 , 𝐴3 , … , 𝐴𝑘 )

+(−1)𝑘+1 𝑑𝑒𝑡(𝐴1 − 𝐴2 + … + (−1)𝑘+1 𝐴𝑘 , 𝐴𝑘+1 )

1 3 5
Example 2.5.8. Evaluate the determinant of 𝐴 = [ ], using theorem 2.5.7
2 4 6

Solution :

|𝐴| = |1 3| + (-1)(3) |−2 5| = ( 4 – 6 ) – (-12 + 10 ) = 0

2 4 −2 6

41
 Chapter Three

Properties for non-square determinant

In this chapter we study Radic definition for determinant of a rectangular

matrix in more detailed way. We present new identities for the determinant of a
rectangular matrix. We develop some important properties of this determinant.
We generalize several classical important determinant identities, and describe
how the determinant is affected by operation on columns such as interchanging
columns, reversing columns or decomposing a single column.

Although we present here properties of Radic determinant but we have proved

in chapter 2 that Radic determinant and the determinant by cofactor expansion
give the same value. So, we may use the term determinant to mean any of the
common values.

3.1 Properties for determinant of a rectangular matrix.

In this section we will be mainly concerned with the properties of the

determinants of square matrices (theorem 1.3.5) that are still valid when one goes
to rectangular matrices.

The following theorem is from [3 ], but we give here another proof

Theorem 3.1.1 If every element in any fixed row of a horizontal matrix can be
expressed as the sum of tow quantities then the given horizontal matrix
determinant can be expressed as the sum of tow horizontal matrix determinant of
the same order with the elements of the remaining rows of the both being the
same.

𝛼 + 𝑎11 𝛽 + 𝑎12 … 𝛿 + 𝑎1𝑛

𝑎 𝑎22 … 𝑎2𝑛
Proof : Let 𝐴 = [ 21 ],
⋮ ⋮ ⋱ ⋮
𝑎𝑚1 𝑎𝑚2 … 𝑎𝑚𝑛

By cofactor expansion along first row of det(𝐴)

42
 𝑎22 … 𝑎2𝑛 𝑎21 … 𝑎2𝑛
| 𝐴 | = (𝛼 + 𝑎11 ) | ⋮ ⋱ ⋮ | − (𝛽 + 𝑎12 ) | ⋮ ⋱ ⋮ |+
𝑎𝑚2 … 𝑎𝑚𝑛 𝑎𝑚1 … 𝑎𝑚𝑛
𝑎21 … 𝑎2𝑛 𝑎21 … 𝑎2(𝑛−1)
1+𝑛
(𝛾 + 𝑎13 ) | ⋮ ⋱ ⋮ | + ⋯ +(−1) (𝛿 + 𝑎1𝑛 ) | ⋮ ⋱ ⋮ |
𝑎𝑚1 … 𝑎𝑚𝑛 𝑎𝑚1 … 𝑎𝑚(𝑛−1)

det(𝐴) =
𝑎22 … 𝑎2𝑛 𝑎21 … 𝑎2𝑛 𝑎21 … 𝑎2𝑛
(𝛼 | ⋮ ⋱ ⋮ |− 𝛽 | ⋮ ⋱ ⋮ |+ 𝛾| ⋮ ⋱ ⋮ |+
𝑎𝑚2 … 𝑎𝑚𝑛 𝑎𝑚1 … 𝑎𝑚𝑛 𝑎𝑚1 … 𝑎𝑚𝑛
𝑎21 … 𝑎2(𝑛−1) 𝑎22 … 𝑎2𝑛 𝑎21 … 𝑎2𝑛
⋯ +(−1)1+𝑛 𝛿| ⋮ ⋱ ⋮ |) + (𝑎11 | ⋮ ⋱ ⋮ | − 𝑎12 | ⋮ ⋱ ⋮ |+
𝑎𝑚1 … 𝑎𝑚(𝑛−1) 𝑎𝑚2 … 𝑎𝑚𝑛 𝑎𝑚1 … 𝑎𝑚𝑛

𝑎21 … 𝑎2𝑛 𝑎21 … 𝑎2(𝑛−1)

𝑎13 | ⋮ ⋱ ⋮ | + ⋯ +(−1)1+𝑛 𝑎1𝑛 | ⋮ ⋮ |)
⋱
𝑎𝑚1 … 𝑎𝑚𝑛 𝑎𝑚1 … 𝑎𝑚(𝑛−1)

𝛼 𝛽 … 𝛿 𝑎11 𝑎12 … 𝑎1𝑛

𝑎 𝑎22 … 𝑎2𝑛 𝑎21 𝑎22 … 𝑎2𝑛
= | 21 | + | ⋮ ⋮ ⋮ |
⋮ ⋮ ⋱ ⋮ ⋱
𝑎𝑚1 𝑎𝑚2 … 𝑎𝑚𝑛 𝑎𝑚1 𝑎𝑚2 … 𝑎𝑚𝑛

Note: 1- is those axiom A3 in definition 2.1.2

2- This property is valid for square matrices as well ( see theorem 1.3.6)

Theorem 3.1.2 [1 ]. Let 1 ≤ 𝑚 ≤ 𝑛, and 𝐴 be an 𝑚 × 𝑚 matrix, and 𝐵 be

an 𝑚 × 𝑛 matrix, then 𝑑𝑒𝑡(𝐴 𝐵) = det(A) det(B)

Proof: Let 𝐵 = [𝐵1 , … , 𝐵𝑛 ], then

𝑑𝑒𝑡(𝐴 𝐵) = det(A [𝐵1 , … , 𝐵𝑛 ] ) = 𝑑et ([A B1 , … , A B𝑛 ])

By Radic definition

det(𝐴 𝐵) = ∑ (−1)𝑟+𝑆 𝑑𝑒𝑡 ([A Bj1 , … , A Bj𝑚 ])

1≤𝑗1 <⋯<𝑗𝑚 ≤𝑛

Where 𝑟 = 1 + 2 + ⋯ + 𝑚 and 𝑠 = 𝑗1 + 𝑗2 + ⋯ + 𝑗𝑚

43
 𝑟+𝑆
det(𝐴 𝐵) = ∑ (−1) det(A) 𝑑𝑒𝑡 ([Bj1 , … , Bj𝑚 ])
1≤𝑗1 <⋯<𝑗𝑚 ≤𝑛

Since Radic definition gives square matrix (see theorem1.3.5.f)

= det(A) ∑ (−1)𝑟+𝑆 𝑑𝑒𝑡 ([Bj , … , Bj ])

1 𝑚
1≤𝑗1 <⋯<𝑗𝑚 ≤𝑛

= det(A)det(B)

2 2 3 5 7
Example 3.1.3 Prove theorem 3.1.2 for 𝐴 = [ ], 𝐵=[ ]
−1 3 2 1 4

10 12 22
Then, 𝐴 𝐵 = [ ],
3 −2 5

det(A) = 8, det(B) = 8, det(𝐴. 𝐵) = 64

𝑑𝑒𝑡(𝐴 𝐵) = det(A) det(B) = 8 × 8 = 64

A sufficient condition for the equation is that 𝐴 is square and 𝐴 𝐵 is defined.

We note here that 𝐵 𝐴 is not defined.

In fact there is no determinant function that satisfies 𝑑𝑒𝑡(𝐴𝐵) = 𝑑𝑒𝑡(𝐴)𝑑𝑒𝑡(𝐵) for all
matrices 𝐴, 𝐵.

1 2 3
1 0 2
Example 3.1.4 Let 𝐴=[ ], 𝐵 = [3 0 −2]
−1 1 1
2 1 0

5 3 3
Then, 𝐴𝐵= [ ],
4 0 −5

det(A) = −4, det(B) = 7, det(𝐴𝐵) = 10

det(A) det(B) = −28 ≠ 𝑑𝑒𝑡(𝐴 𝐵)

We notice from this example that the determinant is not distributed in the case
that the first matrix is rectangular, so for the theorem to be true, the first matrix
must be the square.

44
 0 1
1 1 0
Example 3.1.5 Let 𝐴=[ ], 𝐵 = [ 2 0]
2 3 0
−4 2

2 3 0
2 1
Then, 𝐴 𝐵 = [ ], 𝐵 𝐴 = [2 2 0 ]
6 2
0 2 0

det(A) = 1, det(B) = −2, det(𝐴 𝐵) = −2, det(𝐵 𝐴) = 0

Now, det(𝐴 𝐵) = det(A) det(B) = det(B) det(A) ≠ 𝑑𝑒𝑡(𝐵 𝐴)

−2 = (1)(−2) = (−2)(1) ≠ 0

Lemma 3.1.6 [1 ] Let 𝐴 be an 𝑚 × 𝑛 matrix, 1 ≤ 𝑚 < 𝑛, and

𝑑𝑒𝑡
𝑚 + 𝑛 odd, then 1≤𝑖≤𝑚 [(𝑎𝑖𝑗 )] = 0, where 𝑎1𝑗 = 1 for all 𝑗, 1 ≤ 𝑗 ≤ 𝑛 .
1≤𝑗≤𝑛

Proof: by induction on even integer 𝑛 for all odd integer numbers 𝑚,

1 ≤ 𝑚 < 𝑛.

Base case : If 𝑛 = 2 , then 𝑚 = 1 we have det([1 , 1]) = 1 − 1 = 0

Induction hypothesis : We assume that it is true for even 𝑛 and odd 𝑚,

1 ≤ 𝑚 < 𝑛,

𝑎2,𝑗
1 1 … 1
𝐷 = 𝑑𝑒𝑡[𝐴1 , 𝐴2 , … , 𝐴𝑛 ] = 𝑑𝑒𝑡 [ ] where 𝐵𝑗 = [ ⋮ ] , (1 ≤ 𝑗 ≤ 𝑛)
𝐵1 𝐵2 … 𝐵𝑛 𝑎𝑚,𝑗

Expanding the determinant with respect to the first row yields

𝐷 = 𝑑𝑒𝑡[𝐵2 , 𝐵3 , … , 𝐵𝑛 ] − 𝑑𝑒𝑡[𝐵1 , 𝐵3 , … , 𝐵𝑛 ] + ⋯ + (−1)𝑛+1 𝑑𝑒𝑡[𝐵1 , 𝐵2 , … , 𝐵𝑛−1 ] = 0

Induction step : We will show that the identity holds for 𝑛+2

( which is even number ) and all odd 𝑚, 1≤𝑚 <𝑛+2

1 1 … 1 1 1
𝐷 = 𝑑𝑒𝑡[𝐴1 , 𝐴2 , … , 𝐴𝑛 , 𝐴𝑛+1 , 𝐴𝑛+2 ] = 𝑑𝑒𝑡 [ ]
𝐵1 𝐵2 … 𝐵𝑛 𝐵𝑛+1 𝐵𝑛+2

Expanding the determinant with respect to the first row yields

45
𝐷 = 𝑑𝑒𝑡[𝐵2 , 𝐵3 , … , 𝐵𝑛+2 ] − 𝑑𝑒𝑡[𝐵1 , 𝐵3 , … , 𝐵𝑛+2 ] + ⋯

+ (−1)𝑛+1 𝑑𝑒𝑡[𝐵1 , 𝐵2 , … , 𝐵𝑛−1 , 𝐵𝑛+1 , 𝐵𝑛+2 ]

+ (−1)1+𝑛+1 𝑑𝑒𝑡[𝐵1 , 𝐵2 , … , 𝐵𝑛 , 𝐵𝑛+2 ] + (−1)1+𝑛+2 𝑑𝑒𝑡[𝐵1 , 𝐵2 , … , 𝐵𝑛+1 ]

By theorem 2.5.7, all the resulting determinants are deleted from each other.

Then , 𝐷 = 0.

Another case can be established in the same way(odd 𝑛 and even 𝑚).

This property does not apply if the matrices are square ( 𝑛 × 𝑛) because the
sum of the order is even (whether 𝑛 is odd or even)

1 1 1 1 1
Example 3.1.7 Evaluate the determinant of 𝐴 = [ ] and 𝐵 = [ ]
2 3 4 5 6

Then, det(A) = 1 ≠ 0, and 𝑑𝑒𝑡(𝐵) = 0

Lemma 3.1.8 [1 ] .

1. det [𝐴1 , 𝐴2 , … , 𝐴𝑛 , 0𝑚 ] = 𝑑𝑒𝑡 [𝐴1 , 𝐴2 , … , 𝐴𝑛 ], and

𝟐. det[𝐴1 , 𝐴2 , … , 𝐴𝑗−1 , 0𝑚 , 𝐴𝑗+1 , … , 𝐴𝑛 ] = 𝑑𝑒𝑡[𝐴1 , 𝐴2 , … , 𝐴𝑗−1 , −𝐴𝑗+1 , … , −𝐴𝑛 ]

𝑇
Where 𝑚 ≤ 𝑛, 𝐴𝐾 = [𝑎1,𝑘 , … , 𝑎𝑚,𝑘 ] for 𝑘 ∈ {1 , … , 𝑛} − {𝑗} and 0𝑚 is an 𝑚
by 1 zero vector .

Proof : proof of the first formula,

𝑎1,𝑗
𝑎2,𝑗
𝑎2,𝑗 𝑎1,𝑗
Let 𝐴𝑗 = [ ⋮ ] , 𝐴𝑗 = [ 𝐵 ] where 𝐵𝑗 = [ ⋮ ] , (1 ≤ 𝑗 ≤ 𝑛)
𝑗 𝑎𝑚,𝑗
𝑎𝑚,𝑗

By expanding the determinant with respect to the first row, we get

𝑎 𝑎 ⋯ 𝑎1𝑛 0
D = det [𝐴1 , 𝐴2 , … , 𝐴𝑛 , 0𝑚 ] = 𝑑𝑒𝑡 [ 𝐵11 𝐵12 ⋯ 𝐵 ]
1 2 𝑛 0𝑚

46
𝐷 = (−1)2 𝑎1,1 𝑑𝑒𝑡[𝐵2 , 𝐵3 , … , 𝐵𝑛 , 0𝑚 ] + (−1)3 𝑎12 𝑑𝑒𝑡[𝐵1 , 𝐵3 , … , 𝐵𝑛 , 0𝑚 ] + ⋯

+ (−1)1+𝑛 𝑎1𝑛 𝑑𝑒𝑡[𝐵1 , 𝐵2 , … , 𝐵𝑛−1 , 0𝑚 ]

+ (−1)2+𝑛 (0) 𝑑𝑒𝑡[𝐵1 , 𝐵2 , … , 𝐵𝑛 ]

𝐷 = 𝑎1,1 𝑑𝑒𝑡[𝐵2 , 𝐵3 , … , 𝐵𝑛 , 0𝑚 ] − 𝑎12 𝑑𝑒𝑡[𝐵1 , 𝐵3 , … , 𝐵𝑛 , 0𝑚 ] + ⋯

+ (−1)1+𝑛 𝑎1𝑛 𝑑𝑒𝑡[𝐵1 , 𝐵2 , … , 𝐵𝑛−1 , 0𝑚 ]

det [𝐴1 , 𝐴2 , … , 𝐴𝑛 , 0𝑚 ] = 𝑑𝑒𝑡 [𝐴1 , 𝐴2 , … , 𝐴𝑛 ]

Now, proof the second formula, by (P.M.I)

Base case : For 𝑛 = 2 , 𝑚 = 1 , we have 𝑑𝑒𝑡[𝑎1 0 ] = 𝑎1 − 0 = 𝑎1

Induction hypothesis: Assume that for all 𝑛 and 𝑚, 1 ≤ 𝑚 < 𝑛, it is true that

det[𝐴1 , 𝐴2 , … , 𝐴𝑗−1 , 0𝑚 , 𝐴𝑗+1 , … , 𝐴𝑛 ] = 𝑑𝑒𝑡[𝐴1 , 𝐴2 , … , 𝐴𝑗−1 , −𝐴𝑗+1 , … , −𝐴𝑛 ]

Induction step: We will show that the identity holds for 𝑛 + 1, 1 ≤ 𝑚 < 𝑛 + 1 ?

𝑎1,𝑗
𝑎2,𝑗
𝑎2,𝑗 𝑎1,𝑗
Let 𝐴𝑗 = [ ⋮ ] , 𝐴𝑗 = [ 𝐵 ] where 𝐵𝑗 = [ ⋮ ] , (1 ≤ 𝑗 ≤ 𝑛)
𝑗 𝑎𝑚,𝑗
𝑎𝑚,𝑗

By expanding the determinant with respect to the first row, we get

D = det[𝐴1 , 𝐴2 , … , 𝐴𝑗−1 , 0𝑚 , 𝐴𝑗+1 , … , 𝐴𝑛 , 𝐴𝑛+1 ]

𝑎1,1 … 𝑎1,𝑗−1 0 𝑎1,𝑗+1 … 𝑎1,𝑛 𝑎1,𝑛+1
= 𝑑𝑒𝑡 [ 𝐵𝑗+1 … 𝐵𝑛 𝐵𝑛+1 ]
𝐵1 … 𝐵𝑗−1 0𝑚

= 𝑎1,1 𝑑𝑒𝑡[𝐵2 , … , 𝐵𝑗−1 , 𝑜𝑚 , 𝐵𝑗+1 … , 𝐵𝑛 , 𝐵𝑛+1 ] + ⋯

+ (−1)𝑗−1+1 𝑎1,𝑗−1 𝑑𝑒𝑡[𝐵1 , … , 𝐵𝑗−2 , 0𝑚 , 𝐵𝑗+1 … , 𝐵𝑛 , 𝐵𝑛+1 ] + 0

+ (−1)𝑗+1+1 𝑎1,𝑗+1 𝑑𝑒𝑡[𝐵1 , … , 𝐵𝑗−1 , 0𝑚 , 𝐵𝑗+2 … , 𝐵𝑛 , 𝐵𝑛+1 ] + ⋯

+ (−1)𝑛+2 𝑎1,𝑛+1 𝑑𝑒𝑡[𝐵1 , … , 𝐵𝑗−1 , 0𝑚+1 , 𝐵𝑗+2 … , 𝐵𝑛 ]

47
By inductive hypothesis

𝐷 = 𝑎1,1 𝑑𝑒𝑡[𝐵2 , … , 𝐵𝑗−1 , −𝐵𝑗+1 … , −𝐵𝑛+1 ] + ⋯

+ (−1)𝑗−1+1 𝑎1,𝑗−1 𝑑𝑒𝑡[𝐵1 , … , 𝐵𝑗−2 , − 𝐵𝑗+1 … , −𝐵𝑛+1 ] + ⋯

+ (−1)𝑗+1+1 𝑎1,𝑗+1 𝑑𝑒𝑡[𝐵1 , … , 𝐵𝑗−1 , − 𝐵𝑗+2 … , −𝐵𝑛+1 ] + ⋯

+ (−1)𝑛+2 𝑎1,𝑛+1 𝑑𝑒𝑡[𝐵1 , … , 𝐵𝑗−1 , − 𝐵𝑗+2 … , −𝐵𝑛 ]

𝑎1,1 … 𝑎1,𝑗−1 − 𝑎1,𝑗+1 … −𝑎1,𝑛+1

= 𝑑𝑒𝑡 [ 𝐵 … 𝐵𝑗−1 − 𝐵𝑗+1 … −𝐵𝑛+1 ]
1

= 𝑑𝑒𝑡[𝐴1 , 𝐴2 , … , 𝐴𝑗−1 , −𝐴𝑗+1 , … , −𝐴𝑛+1 ]

This property does not apply if the matrices are square since if all the elements
of a column are zeros, then the value of the determinant is zero. (see theorem
1.3.5)

Example 3.1.9 An example that illustrates Theorem 3.1.8 is

3 5 7
Let 𝐴 = [ ], det(𝐴) = 8
2 1 4

So, |3 5 7 0| = |3 5 7| = 8,
2 1 4 0 2 1 4

and |3 0 5 7| = |3 −5 −7 | = 8
2 0 1 4 2 −1 −4

Theorem 3.1.10 [1 ]. Suppose 1≤𝑚<𝑛, and 𝑚+𝑛 be an odd integer,

𝐴 = (𝑎𝑖,𝑗 ) = [𝐴1 , … , 𝐴𝑛 ] be an 𝑚 × 𝑛 matrix, and 𝑋 be an arbitrary 𝑚×1
column vector, then

det[𝐴1 + 𝑋 , … , 𝐴𝑛 + 𝑋] = 𝑑𝑒𝑡[𝐴1 , … , 𝐴𝑛 ]

Proof: by (P.M.I )

Base case : For 𝑛 = 2 , 𝑚 = 1, we have

det[𝑎1,1 + 𝑋 𝑎1,2 + 𝑋] = 𝑎1,1 − 𝑎1,2 = 𝑑𝑒𝑡[𝑎1,1 𝑎1,2 ]

48
Induction hypothesis: Assume the assertion is true for all 𝑛 even and 𝑚 odd with
1≤𝑚<𝑛,

det[𝐴1 + 𝑋 , … , 𝐴𝑛 + 𝑋] = 𝑑𝑒𝑡[𝐴1 , … , 𝐴𝑛 ]

We have to prove that is true for 𝑛 + 2 even and all 𝑚 odd ?

With 1 ≤ 𝑚 < 𝑛 + 2 ,

𝑎2𝑗 𝑥2
𝑎1𝑗 𝑥1
Let 𝐴𝑗 = [ 𝐵 ] where 𝐵𝑗 = [ ⋮ ] , (1 ≤ 𝑗 ≤ 𝑛 + 2) and 𝑋 = [ / ] where 𝑋/ = [ ⋮ ]
𝑗 𝑎𝑚𝑗 𝑋 𝑥𝑚

By expanding the determinant with respect to the first row , we get

𝑎1,1 + 𝑥1 … 𝑎1,𝑛+2 + 𝑥1
det[𝐴1 + 𝑋 , … , 𝐴𝑛+2 + 𝑋] = 𝑑𝑒𝑡 [ ]
𝐵1 + 𝑋/ … 𝐵𝑛+2 + 𝑋/

= (−1)1+1 (𝑎1,1 + 𝑥1 )𝑑𝑒𝑡 [ 𝐵2 + 𝑋/ , … , 𝐵𝑛+2 + 𝑋/ ] + ⋯

+ (−1)𝑛+2+1 (𝑎1,𝑛+2 + 𝑥1 )𝑑𝑒𝑡 [ 𝐵1 + 𝑋/ , … , 𝐵𝑛+1 + 𝑋/ ]

By induction hypothesis for each component we obtain:

𝑑𝑒𝑡[𝐴1 + 𝑋 , … , 𝐴𝑛+2 + 𝑋] = (−1)1+1 (𝑎1,1 + 𝑥1 )𝑑𝑒𝑡[ 𝐵2 , … , 𝐵𝑛+2 ]

+ ⋯ +(−1)𝑛+2+1 (𝑎1,𝑛+2 + 𝑥1 )𝑑𝑒𝑡[ 𝐵1 , … , 𝐵𝑛+1 ]

= (−1)1+1 (𝑎1,1 )𝑑𝑒𝑡[ 𝐵2 , … , 𝐵𝑛+2 ] + ⋯ + (−1)𝑛+2+1 (𝑎1,𝑛+2 )𝑑𝑒𝑡[ 𝐵1 , … , 𝐵𝑛+1 ]

+ 𝑥1 [(−1)1+1 𝑑𝑒𝑡[ 𝐵2 , … , 𝐵𝑛+2 ] + ⋯ (−1)𝑛+2+1 𝑑𝑒𝑡[ 𝐵1 , … , 𝐵𝑛+1 ]]

𝑎1,1 … 𝑎1,𝑛+2 1 … 1
𝑑𝑒𝑡[𝐴1 + 𝑋 , … , 𝐴𝑛+2 + 𝑋] = 𝑑𝑒𝑡 [ 𝐵 … 𝐵𝑛+2 ] + 𝑥1 𝑑𝑒𝑡 [ ]
1 𝐵1 … 𝐵𝑛+2

Applying lemma 3.1.6

𝑎1,1 … 𝑎1,𝑛+2
𝑑𝑒𝑡[𝐴1 + 𝑋 , … , 𝐴𝑛+2 + 𝑋] = 𝑑𝑒𝑡 [ 𝐵 … 𝐵𝑛+2 ] + 0
1

= 𝑑𝑒𝑡[𝐴1 , … , 𝐴𝑛+2 ]

The second case (𝑛 odd and 𝑚 even) can be treated similarly.

49
 3 5 7 1
Example 3.1.11 Let 𝐴= [ ], take 𝑋 = [ ],
2 1 4 2

Then, 𝑑𝑒𝑡(𝐴) = det(𝐴1 , 𝐴2 , 𝐴3 ) = 8

Also, det(𝐴1 + 𝑋 , 𝐴2 + 𝑋 , 𝐴3 + 𝑋) = |4 6 8 | = 8
4 3 6

Corollary 3.1.12 [1] For 𝑚 + 𝑛 odd, 1 ≤ m < n and for all 𝑘,

1 ≤ k ≤ n , we have

det[𝐴1 , … , 𝐴𝑘−1 , 𝐴𝑘 , 𝐴𝑘+1 , … , 𝐴𝑛 ]

= 𝑑𝑒𝑡[𝐴𝑘 − 𝐴1 , … , 𝐴𝑘 − 𝐴𝑘−1 , 𝐴𝑘+1 − 𝐴𝑘 , … , 𝐴𝑛 − 𝐴𝑘 ]

Proof: det(A) = det[𝐴1 , … , 𝐴𝑘−1 , 𝐴𝑘 , 𝐴𝑘+1 , … , 𝐴𝑛 ]

Applying theorem 3.1.10 with X = −Ak and lemma 3.1.8

= 𝑑𝑒𝑡[𝐴1 − 𝐴𝑘 , … , 𝐴𝑘−1 − 𝐴𝑘 , 𝐴𝑘 − 𝐴𝑘 , 𝐴𝑘+1 − 𝐴𝑘 , … , 𝐴𝑛 − 𝐴𝑘 ]

= 𝑑𝑒𝑡[𝐴1 − 𝐴𝑘 , … , 𝐴𝑘−1 − 𝐴𝑘 , 0𝑚 , 𝐴𝑘+1 − 𝐴𝑘 , … , 𝐴𝑛 − 𝐴𝑘 ]

= 𝑑𝑒𝑡[𝐴1 − 𝐴𝑘 , … , 𝐴𝑘−1 − 𝐴𝑘 , 𝐴𝑘 − 𝐴𝑘+1 , … , 𝐴𝑘 − 𝐴𝑛 ]

= 𝑑𝑒𝑡[𝐴𝑘 − 𝐴1 , … , 𝐴𝑘 − 𝐴𝑘−1 , 𝐴𝑘+1 − 𝐴𝑘 , … , 𝐴𝑛 − 𝐴𝑘 ]

3 5 7
Example 3.1.13 let 𝐴 = [ ] , then det(𝐴 ) = 8
2 1 4

Applying Corollary 3.1.12 with 𝐴𝑘 = [5]

1

det(𝐴 ) = | 2 2| = 8
−1 3

Note, This properties (theorem 3.1.10 and corollary 3.1.12) do not apply if the
matrices are square ( 𝑛 × 𝑛) because the sum of the order is even (Whether 𝑛 is
odd or even )

50
3.2 How determinant is affected by operations on columns.

In this section we describe how the determinant is affected by operations on

columns such as interchanging columns, reversing columns or decomposing a
single column.

(1) Decomposing column in a square matrix if

𝐴𝑛×𝑛 = [𝐴1 , 𝐴2 , … , 𝐴𝑘 , … , 𝐴𝑛 ] and 𝐴𝑘 = 𝐵𝑘 + 𝐶𝑘 , 𝑘 𝜖 {1,2, … , 𝑛} then

| 𝐴 | = | 𝐴1 , 𝐴2 , … , 𝐴𝑘−1 , 𝐵𝑘 , 𝐴𝑘+1 , … , 𝐴𝑛 | + | 𝐴1 , 𝐴2 , … , 𝐴𝑘−1 , 𝐶𝑘 , 𝐴𝑘+1 , … , 𝐴𝑛 |.

In non-square matrices case what happens, we have

Theorem 3.2.1 [ 12 ]. Let 𝐴 = [𝐴1 , 𝐴2 , … , 𝐴𝑘 , … , 𝐴𝑛 ] be a 𝑚 ×𝑛 matrix,

𝑚 ≤ 𝑛, and 𝐴𝑘 = 𝐵𝑘 + 𝐶𝑘 for some 𝑘 𝜖 {1,2, … , 𝑛} . Then

| 𝐴 | = | 𝐴1 , 𝐴2 , … , 𝐴𝑘−1 , 𝐵𝑘 , 𝐴𝑘+1 , … , 𝐴𝑛 | + | 𝐴1 , 𝐴2 , … , 𝐴𝑘−1 , 𝐶𝑘 , 𝐴𝑘+1 , … , 𝐴𝑛 |

+ ∑ (−1)𝑟+𝑗1 +𝑗2 +⋯+𝑗𝑚+1 |𝐴𝑗1 , 𝐴𝑗2 , … , 𝐴𝑗𝑚 |

1≤𝑗1 <⋯<𝑗𝑚 ≤𝑛
𝑘 {𝑗1 ,… ,𝑗𝑚 }

Where 𝑟 = 1 + 2 + ⋯ + 𝑚

Proof : After applying (2.5.1)

|𝐴| = ∑ (−1)𝑟+𝑗1 +𝑗2 +⋯+𝑗𝑚 |𝐴𝑗1 , 𝐴𝑗2 , … , 𝐴𝑗𝑚 |

1≤𝑗1 <⋯<𝑗𝑚 ≤𝑛

We separate the sum of determinants into tow sums: the first one consisting of the
determinants of matrices which contain the column 𝐴𝑘 = 𝐵𝑘 + 𝐶𝑘 and the
second one consisting of other determinants.

|𝐴|= ∑ (−1)𝑟+𝑗1 +𝑗2 +⋯+𝑗𝑚 |𝐴𝑗1 , … 𝐴𝑘 , … , 𝐴𝑗𝑚 |

1≤𝑗1 <⋯<𝑗𝑚 ≤𝑛
𝑘∈ {𝑗1 ,… ,𝑗𝑚 }

+ ∑ (−1)𝑟+𝑗1 +𝑗2 +⋯+𝑗𝑚 |𝐴𝑗1 , 𝐴𝑗2 , … , 𝐴𝑗𝑚 |

1≤𝑗1 <⋯<𝑗𝑚 ≤𝑛
𝑘 {𝑗1 ,… ,𝑗𝑚 }

51
 = ∑ (−1)𝑟+𝑗1 +𝑗2 +⋯+𝑗𝑚 |𝐴𝑗1 , … , 𝐵𝑘 , … , 𝐴𝑗𝑚 |
1≤𝑗1 <⋯<𝑗𝑚 ≤𝑛
𝑘∈ {𝑗1 ,… ,𝑗𝑚 }

+ ∑ (−1)𝑟+𝑗1 +𝑗2 +⋯+𝑗𝑚 |𝐴𝑗1 , . . , 𝐶𝑘 , … , 𝐴𝑗𝑚 |

1≤𝑗1 <⋯<𝑗𝑚 ≤𝑛
𝑘∈ {𝑗1 ,… ,𝑗𝑚 }

+ ∑ (−1)𝑟+𝑗1 +𝑗2 +⋯+𝑗𝑚 |𝐴𝑗1 , 𝐴𝑗2 , … , 𝐴𝑗𝑚 |

1≤𝑗1 <⋯<𝑗𝑚 ≤𝑛
𝑘 {𝑗1 ,… ,𝑗𝑚 }

Now the third sum is added and subtracted so that it can be included into both the
first and the second sum :

| 𝐴 | = | 𝐴1 , 𝐴2 , … , 𝐴𝑘−1 , 𝐵𝑘 , 𝐴𝑘+1 , … , 𝐴𝑛 | + | 𝐴1 , 𝐴2 , … , 𝐴𝑘−1 , 𝐶𝑘 , 𝐴𝑘+1 , … , 𝐴𝑛 |

− ∑ (−1)𝑟+𝑗1 +𝑗2 +⋯+𝑗𝑚 |𝐴𝑗1 , 𝐴𝑗2 , … , 𝐴𝑗𝑚 |

1≤𝑗1 <⋯<𝑗𝑚 ≤𝑛
𝑘 {𝑗1 ,… ,𝑗𝑚 }

= | 𝐴1 , 𝐴2 , … , 𝐴𝑘−1 , 𝐵𝑘 , 𝐴𝑘+1 , … , 𝐴𝑛 | + | 𝐴1 , 𝐴2 , … , 𝐴𝑘−1 , 𝐶𝑘 , 𝐴𝑘+1 , … , 𝐴𝑛 |

+ ∑ (−1)𝑟+𝑗1 +𝑗2 +⋯+𝑗𝑚+1 |𝐴𝑗1 , 𝐴𝑗2 , … , 𝐴𝑗𝑚 |

1≤𝑗1 <⋯<𝑗𝑚 ≤𝑛
𝑘 {𝑗1 ,… ,𝑗𝑚 }

Example 3.2.2 Let [𝐴1 , 𝐴2 , 𝐴3 ] be a 2 × 3 matrix and 𝐴1 = 𝐵1 + 𝐶1 . Then

according to theorem 2.7.2 we have

|𝐵1 + 𝐶1 , 𝐴2 , 𝐴3 |

(−1) 1+2 +𝑗1+𝑗2+1 |𝐴𝑗1 , 𝐴𝑗2 |

( )
= |𝐵1 , 𝐴2 , 𝐴3 | + |𝐶1 , 𝐴2 , 𝐴3 | + ∑
1≤𝑗1 <𝑗2 ≤3
1 {𝑗1 ,𝑗2 }

= |𝐵1 , 𝐴2 , 𝐴3 | + |𝐶1 , 𝐴2 , 𝐴3 | + (−1)3+2+3+1 |𝐴2 , 𝐴3 |

= |𝐵1 , 𝐴2 , 𝐴3 | + |𝐶1 , 𝐴2 , 𝐴3 | − |𝐴2 , 𝐴3 |.

(2) Interchanging columns in a square matrix results in changing the sign of the

determinant. Non-square matrices in which the number of columns is equal to the

number of rows increased by one have the same property.

52
Theorem 3.2.3 [12 ]. Let 𝐴 = [𝐴1 , 𝐴2 , … , 𝐴𝑚 , 𝐴𝑚+1 ] be a 𝑚 × (𝑚 + 1)

matrix. Then for each 𝑖 , 𝑗 ∈ {1,2, … , 𝑚 + 1} such that 𝑖 < 𝑗, we have

𝑑𝑒𝑡(𝐴) = −𝑑𝑒𝑡(𝐴1 , 𝐴2 , … , 𝐴𝑖−1 , 𝐴𝑗 , 𝐴𝑖+1 , … , 𝐴𝑗−1 , 𝐴𝑖 , 𝐴𝑗+1 , … , 𝐴𝑚 , 𝐴𝑚+1 )

Proof: Let 𝑟 = 1 + 2 + ⋯ + 𝑚. Fix each 𝑖 , 𝑗 ∈ {1,2, … , 𝑚 + 1} such that 𝑖 < 𝑗.

From all the determinants in the right–hand side of

|𝐴| = ∑ (−1)𝑟+𝑗1 +𝑗2 +⋯+𝑗𝑚 |𝐴𝑗1 , 𝐴𝑗2 , … , 𝐴𝑗𝑚 |

1≤𝑗1 <⋯<𝑗𝑚 ≤𝑛

We distinguish the determinants in the expression which contain either 𝐴𝑖 or 𝐴𝑗

but not both of them. Thus, we have

(𝑚+1)(𝑚+2)
|𝐴| = (−1)(𝑟+ 2
−𝑖)
|𝐴1 , 𝐴2 , … , 𝐴𝑖−1 , 𝐴𝑖+1 , … , 𝐴𝑗−1 , 𝐴𝑗 , 𝐴𝑗+1 , … , 𝐴𝑚+1 |

(𝑚+1)(𝑚+2)
+ (−1)(𝑟+ 2
−𝑗)
|𝐴1 , 𝐴2 , … , 𝐴𝑖−1 , 𝐴𝑖 , 𝐴𝑖+1 , … , 𝐴𝑗−1 , 𝐴𝑗+1 , … , 𝐴𝑚+1 |

+ ∑ (−1)𝑟+𝑗1 +𝑗2 +⋯+𝑗𝑚 |𝐴𝑗1 , … , 𝐴𝑖 , … , 𝐴𝑗 , . . , 𝐴𝑗𝑚 |

1≤𝑗1 <⋯<𝑗𝑚 ≤𝑛
𝑖 ,𝑗 ∈ {𝑗1 ,,… ,𝑗𝑚 }

Notice that exactly 𝑗 − 𝑖 − 1 inversions are needed to move the column 𝐴𝑗 to the
position between 𝐴𝑖−1 and 𝐴𝑖+1 in the first summand. Similarly, in the second
summand, also 𝑗 − 𝑖 − 1 inversions are needed to move the column 𝐴𝑖 to the
position between 𝐴𝑗−1 and 𝐴𝑗+1 .

In other summands we can simply interchange columns 𝐴𝑖 and 𝐴𝑗 with the sign
change ( square matrix 𝑚 × 𝑚). Thus, we have

(𝑚+1)(𝑚+2)
(𝑟+ −𝑖+(𝑖−𝑗−1))
|𝐴| = (−1) 2

× |𝐴1 , 𝐴2 , … , 𝐴𝑖−1 , 𝐴𝑗 , 𝐴𝑖+1 , … , 𝐴𝑗−1 , 𝐴𝑗+1 , … , 𝐴𝑚+1 |

(𝑟+ 𝑚+1 2 𝑚+2 −𝑗+(𝑗−𝑖−1))

( )( )
+ (−1) |𝐴 1 , 𝐴2 , … , 𝐴𝑖−1 , 𝐴𝑖+1 , … , 𝐴𝑗−1 , 𝐴𝑖 , 𝐴𝑗+1 , … , 𝐴𝑚+1 |

53
 𝑟+𝑗1 +𝑗2 +⋯+𝑗𝑚
− ∑ (−1)(−1) |𝐴𝑗1 , … , 𝐴𝑗 , … , 𝐴𝑖 , . . , 𝐴𝑗𝑚 |
1≤𝑗1 <⋯<𝑗𝑚 ≤𝑛
𝑖 ,𝑗 ∈ {𝑗1 ,… ,𝑗𝑚 }

(𝑚+1)(𝑚+2)
det(𝐴) = −(−1)(𝑟+ 2 −𝑗)
|𝐴1 , 𝐴2 , … , 𝐴𝑖−1 , 𝐴𝑗 , 𝐴𝑖+1 , … , 𝐴𝑗−1 , 𝐴𝑗+1 , … , 𝐴𝑚+1 |

(𝑚+1)(𝑚+2)
− (−1)(𝑟+ 2 −𝑖)
|𝐴1 , 𝐴2 , … , 𝐴𝑖−1 , 𝐴𝑖+1 , … , 𝐴𝑗−1 , 𝐴𝑖 , 𝐴𝑗+1 , … , 𝐴𝑚+1 |

− ∑ (−1)𝑟+𝑗1 +𝑗2 +⋯+𝑗𝑚 |𝐴𝑗1 , … , 𝐴𝑗 , … , 𝐴𝑖 , . . , 𝐴𝑗𝑚 |

1≤𝑗1 <⋯<𝑗𝑚 ≤𝑛
𝑖 ,𝑗 ∈ {𝑗1 ,… ,𝑗 }
𝑚

det(𝐴) = −𝑑𝑒𝑡(𝐴1 , 𝐴2 , … , 𝐴𝑖−1 , 𝐴𝑗 , 𝐴𝑖+1 , … , 𝐴𝑗−1 , 𝐴𝑖 , 𝐴𝑗+1 , … , 𝐴𝑚 , 𝐴𝑚+1 )

Remark 3.2.4 Consider an 𝑚 × 𝑛 matrix 𝐴 with 𝑚 rows and 𝑛 columns, 𝑚 ≤ 𝑛.

Let 𝐴/ be a matrix obtained from 𝐴 by interchanging two columns. Theorem 3.2.3

tells us that det(𝐴) + det (𝐴/ ) = 0 when 𝑛 − 𝑚 = 1. However, in general, if

𝑛 − 𝑚 > 1 the sum det(𝐴) + det (𝐴/ ) is not zero as explained in the following

example.

1 2 7 4 3
Example 3.2.5 Let 𝐴 = [ ] , det(𝐴) = 5
3 0 1 −1 2

And 𝐴/ = [1 2 4 7 3] , det (𝐴/ ) = 5

3 0 −1 1 2

det(𝐴) + det (𝐴/ ) is not zero, since 𝑛 − 𝑚 > 1

Theorem 3.2.6 [1] (Cyclic). If 1 ≤ 𝑚 < 𝑛, and 𝑚 + 𝑛 is an odd integer, then for all
𝑖 ∈ {1 , … , 𝑛}. We have

(−1) 𝑖+1 𝑚 𝑑𝑒𝑡[𝐴𝑖 , … , 𝐴𝑛 , … , 𝐴𝑖−1 ] = 𝑑𝑒𝑡[𝐴1 , … , 𝐴𝑛 ]

( )

Proof: It is sufficient to prove

(−1)𝑚 𝑑𝑒𝑡[𝐴𝑛 , 𝐴1 , … , 𝐴𝑛−1 ] = 𝑑𝑒𝑡[𝐴1 , … , 𝐴𝑛 ]

54
Applying Theorem 3.1.10 with 𝑋 = −𝐴𝑛 and Lemma 3.1.8 we have

(−1)𝑚 𝑑𝑒𝑡[𝐴𝑛 , 𝐴1 , … , 𝐴𝑛−1 ] = (−1)𝑚 𝑑𝑒𝑡[0𝑚 , 𝐴1 − 𝐴𝑛 , … , 𝐴𝑛−1 − 𝐴𝑛 ]

= (−1)𝑚 (−1)𝑚 𝑑𝑒𝑡[ 𝐴1 − 𝐴𝑛 , … , 𝐴𝑛−1 − 𝐴𝑛 ]

= 𝑑𝑒𝑡[ 𝐴1 − 𝐴𝑛 , … , 𝐴𝑛−1 − 𝐴𝑛 ]

= 𝑑𝑒𝑡[ 𝐴1 − 𝐴𝑛 , … , 𝐴𝑛−1 − 𝐴𝑛 , 𝐴𝑛 − 𝐴𝑛 ]

Applying Theorem 3.1.10 with 𝑋 = 𝐴𝑛

(−1)𝑚 𝑑𝑒𝑡[𝐴𝑛 , 𝐴1 , … , 𝐴𝑛−1 ] = 𝑑𝑒𝑡[𝐴1 , … , 𝐴𝑛 ]

Theorem 3.2.7 [1 ] (Semi-Cyclic). If 1 ≤ 𝑚 < 𝑛 , and 𝑚 + 𝑛 even, then

for all 𝑖 ∈ {1 , … , 𝑛}. We have

(−1) 𝑛−𝑖 𝑚 𝑑𝑒𝑡[𝐴𝑖 , … , 𝐴𝑛 , −𝐴1 , … , −𝐴𝑖−1 ] = 𝑑𝑒𝑡[𝐴1 , … , 𝐴𝑛 ]

( )

Proof: It is sufficient to prove

𝑑𝑒𝑡[𝐴𝑛 , − 𝐴1 , … , − 𝐴𝑛−1 ] = 𝑑𝑒𝑡[𝐴1 , … , 𝐴𝑛 ]

Applying Lemma 3.1.8 and Theorem 3.1.10 with 𝑋 = −𝐴𝑛 we have

𝑑𝑒𝑡[𝐴𝑛 , − 𝐴1 , … , − 𝐴𝑛−1 ] = 𝑑𝑒𝑡[𝐴𝑛 , − 𝐴1 , … , − 𝐴𝑛−1 , 0𝑚 ]

= 𝑑𝑒𝑡[𝐴𝑛 − 𝐴𝑛 , − 𝐴1 − 𝐴𝑛 , … , − 𝐴𝑛−1 − 𝐴𝑛 , 0𝑚 − 𝐴𝑛 ]

= 𝑑𝑒𝑡[0𝑚 , − 𝐴1 − 𝐴𝑛 , … , − 𝐴𝑛−1 − 𝐴𝑛 , −𝐴𝑛 ]

= 𝑑𝑒𝑡[ 𝐴1 + 𝐴𝑛 , … , 𝐴𝑛−1 + 𝐴𝑛 , 𝐴𝑛 ]

= 𝑑𝑒𝑡[ 𝐴1 + 𝐴𝑛 , … , 𝐴𝑛−1 + 𝐴𝑛 , 𝐴𝑛 , 0𝑚 ]

= 𝑑𝑒𝑡[ 𝐴1 , … , 𝐴𝑛−1 , 0𝑚 , 0𝑚 − 𝐴𝑛 ]

= 𝑑𝑒𝑡[ 𝐴1 , … , 𝐴𝑛−1 , 𝐴𝑛 ]

55
 1 2 7 4 3
Example 3.2.8 Let 𝐴 = [ ] , then det(𝐴 ) = 5
3 0 1 −1 2

let 𝑖 = 2, by theorem 3.2.6

det(𝐴2 ) = (−1)(3)(2) |2 7 4 3 1| = 5
0 1 −1 2 3

let 𝑖 = 3, by theorem 3.2.6

det(𝐴3 ) = (−1)(4)(2) |7 4 3 1 2| = 5
1 −1 2 3 0

1 2 7 2
Let 𝐵 = [ ], then det(𝐵 ) = 26
3 0 1 3

Let 𝑖 = 2 , by theorem 3.2.7

det(𝐵2 ) = (−1)(2)(2) |2 7 2 −1| = 26

0 1 3 −3

(3) Reversing columns in a 𝑛 × 𝑛 square matrix results in changing the sign of

its determinant if and only if 𝑛 is congruent to 2 or 3 (mod 4). Surprisingly, the
determinant of non-square matrix also either change or does not change the sign
after column reversing, depending on the number of rows and the number of
columns of the matrix.

Theorem 2.7.13 [12 ]. Let [𝐴1 , 𝐴2 , … , 𝐴𝑛 ] be a 𝑚 × 𝑛 matrix, 𝑚 ≤ 𝑛. Then we

𝑚(2𝑛+𝑚+1)
have |𝐴𝑛 , 𝐴𝑛−1 , … , 𝐴2 , 𝐴1 | = |𝐴1 , 𝐴2 , … , 𝐴𝑛−1 , 𝐴𝑛 |. (−1) 2

|𝐴1 , 𝐴2 , … , 𝐴𝑛−1 , 𝐴𝑛 | 𝑖𝑓 𝑚 ≡ 0(𝑚𝑜𝑑4),

|𝐴1 , 𝐴2 , … , 𝐴𝑛−1 , 𝐴𝑛 | . (−1)𝑛+1 𝑖𝑓 𝑚 ≡ 1(𝑚𝑜𝑑4),
=
|𝐴1 , 𝐴2 , … , 𝐴𝑛−1 , 𝐴𝑛 |. (−1) 𝑖𝑓 𝑚 ≡ 2(𝑚𝑜𝑑4),
{ 1 , 𝐴2 , … , 𝐴𝑛−1 , 𝐴𝑛 | . (−1)𝑛
|𝐴 𝑖𝑓 𝑚 ≡ 3(𝑚𝑜𝑑4),

𝑚(𝑚+1)
Proof: Let 𝑟 = 1 + 2 + ⋯ + 𝑚 = and 𝐵𝑘 = 𝐴𝑛+1−𝑘 ,
2

(𝑚−1)𝑚
𝑘 ∈ {1,2, … , 𝑛}. Since exactly (𝑚 − 1) + (𝑚 − 2) + ⋯ + 1 =
2
inversions
of (adjacent) columns are needed to reverse the columns of a 𝑚 × 𝑚 matrix, we
have
|𝐵1 , 𝐵2 , … , 𝐵𝑛 | = ∑1≤𝑖1<⋯<𝑖𝑚 ≤𝑛(−1)𝑟+𝑖1+𝑖2+⋯+𝑖𝑚 |𝐵𝑖1 , 𝐵𝑖2 , … , 𝐵𝑖𝑚 |

56
 (𝑚−1)𝑚
= ∑ (−1)𝑟+𝑖1 +𝑖2+⋯+𝑖𝑚+ 2 |𝐵𝑖𝑚 , 𝐵𝑖𝑚−1 , … , 𝐵𝑖1 |
1≤𝑖1 <⋯<𝑖𝑚 ≤𝑛

(𝑚−1)𝑚
= ∑ (−1)𝑟+𝑖1 +𝑖2+⋯+𝑖𝑚+ 2 |𝐴𝑛+1−𝑖𝑚 , 𝐴𝑛+1−𝑖𝑚−1 , … , 𝐴𝑛+1−𝑖1 |.
1≤𝑖1 <⋯<𝑖𝑚 ≤𝑛

Applying the following change of variables: 𝑗𝑘 = 𝑛 + 1 − 𝑖𝑚−𝑘+1 for each

𝑘 ∈ {1,2, … , 𝑚}, we get

|𝐴𝑛 , 𝐴𝑛−1 , … , 𝐴2 , 𝐴1 | = |𝐵1 , 𝐵2 , … , 𝐵𝑛 |

(𝑚−1)𝑚
= ∑ (−1)𝑟+(𝑛+1)−(𝑗1 +𝑗2 +⋯+𝑗𝑚)+ 2 |𝐴𝑗1 , 𝐴𝑗2 , … , 𝐴𝑗𝑚 |
1≤𝑗1 <⋯<𝑗𝑚 ≤𝑛

(𝑚−1)𝑚
= ∑ (−1)𝑟+𝑚(𝑛+1)−(𝑗1 +𝑗2 +⋯+𝑗𝑚)+ 2 |𝐴𝑗1 , 𝐴𝑗2 , … , 𝐴𝑗𝑚 |
1≤𝑗1 <⋯<𝑗𝑚 ≤𝑛

(𝑚−1)𝑚
= |𝐴1 , 𝐴2 , … , 𝐴𝑛−1 , 𝐴𝑛 |. (−1)𝑚(𝑛+1)+ 2

𝑚(2𝑛+𝑚+1)
= |𝐴1 , 𝐴2 , … , 𝐴𝑛−1 , 𝐴𝑛 |. (−1) 2

Finally, we state that

1 𝑖𝑓 𝑚 ≡ 0(𝑚𝑜𝑑4),
𝑚 (−1)𝑛+1 𝑖𝑓 𝑚 ≡ 1(𝑚𝑜𝑑4),
(−1) 2 (2𝑛+𝑚+1) =
(−1) 𝑖𝑓 𝑚 ≡ 2(𝑚𝑜𝑑4),
𝑛
{ (−1) 𝑖𝑓 𝑚 ≡ 3(𝑚𝑜𝑑4),
Which is easy to verify.
1 2 7 4 3
Example 2.7.14 Let 𝐴 = [ ]
3 0 1 −1 2

Then det(𝐴 ) = 5, by theorem 2.7.13,

3 4 7
|
2 1| = (−1)22(10+2+1) det(𝐴) = (−1)(5) = −5
2 −1 1 0 3

𝑚 ≡ 2(𝑚𝑜𝑑4) → 2 𝑚𝑜𝑑 4 = 2

3 4 7 2 1 1 2 7 4 3 (−1)
| |=| | = −5
2 −1 1 0 3 3 0 1 −1 2

57
 Chapter Four

Applications of Determinant to Area of Polygons in 𝑹 𝟐

In this Chapter we will study the application for determinants of non-square

matrices in calculating the area of polygons in 𝑹𝟐 .

4.1 Areas of certain polygons in connection with determinant of rectangular

matrices.

First we state and prove the following result that relates the determinant of a

2 × 2 matrix with the area of the parallelogram spanned by its columns

Theorem 4.1.1 [9] The absolute value of the determinant

𝑢1 𝑣1
det [𝑢 𝑣2 ]
2

is equal to the area of the parallelogram in 2-space determined by the vectors

𝑢 = (𝑢1 , 𝑢2 ) 𝑎𝑛𝑑 𝑣 = (𝑣1 , 𝑣2 )

Proof: The parallelogram defined by the columns of the above matrix is the one

with vertices at (0, 0), (𝑢1 , 𝑢2 ), (𝑢1 + 𝑣1 , 𝑢2 + 𝑣2 ) 𝑎𝑛𝑑 (𝑣1 , 𝑣2 ), as shown in

the accompanying diagram.

area of the parallelogram 𝐴 = 𝐵 𝐻

𝐵 = ‖𝑢
⃗ ‖, and 𝐻 2 + ‖𝑝𝑟𝑜𝑗𝑢 𝑣‖2 = ‖𝑣‖2

𝐻 2 = ‖𝑣‖2 − ‖𝑝𝑟𝑜𝑗𝑢 𝑣‖2

⃗ .𝑢
𝑣 ⃗ 2
𝐻2 = 𝑣. 𝑣 − ‖ ⃗ ⃗
𝑢‖ (see theorem 1.9.6)
𝑢.𝑢

58
 𝑣. 𝑢
⃗ 𝑣. 𝑢
⃗
𝐻2 = 𝑣. 𝑣 − ( 𝑢
⃗ . 𝑢
⃗)
𝑢
⃗ .𝑢
⃗ 𝑢
⃗ .𝑢
⃗

(𝑣. 𝑢
⃗ )(𝑣. 𝑢
⃗)
𝐻2 = 𝑣. 𝑣 − ( (𝑢 ⃗ ))
⃗⃗⃗ . 𝑢
(𝑢 ⃗ )(𝑢
⃗ .𝑢 ⃗)
⃗ .𝑢

2
(𝑣. 𝑢⃗ )2
𝐻 = 𝑣. 𝑣 − ( )
(𝑢
⃗ .𝑢⃗)

Now, 𝐴2 = 𝐵 2 𝐻 2

2
⃗ )2
(𝑣. 𝑢
𝐴 =𝑢
⃗ .𝑢
⃗ (𝑣. 𝑣 − )
(𝑢 ⃗)
⃗ .𝑢

𝐴2 = (𝑢 ⃗ )2
⃗ )(𝑣. 𝑣) − (𝑣. 𝑢
⃗ .𝑢

But, 𝑢 = (𝑢1 , 𝑢2 ) 𝑎𝑛𝑑 𝑣 = (𝑣1 , 𝑣2 )

So,

𝐴2 = (𝑢12 + 𝑢22 )(𝑣12 + 𝑣22 ) − (𝑢1 𝑣1 + 𝑢2 𝑣2 )2 (see definition 1.8.3)

𝐴2 = 𝑢12 𝑣12 + 𝑢12 𝑣22 + 𝑢22 𝑣12 + 𝑢22 𝑣22 − 𝑢12 𝑣12 − 2𝑢1 𝑣1 𝑢2 𝑣2 − 𝑢22 𝑣22

𝐴2 = 𝑢12 𝑣22 − 2𝑢1 𝑣1 𝑢2 𝑣2 + 𝑢22 𝑣12

𝑢1 𝑣1 2
𝐴2 = (𝑢1 𝑣2 − 𝑢2 𝑣1 )2 = (det [𝑢 𝑣2 ])
2

𝑢1 𝑣1
Area = |det [𝑢 𝑣2 ]|
2

The area of the triangle whose heads are 𝑢 = (𝑢1 , 𝑢2 ) 𝑎𝑛𝑑 𝑣 = (𝑣1 , 𝑣2 ) is

1 𝑢1 𝑣1
Area = 2 |det [𝑢 𝑣2 ]|
2

𝑑𝑒𝑡(𝑢, 𝑣) = −𝑑𝑒𝑡(𝑣, 𝑢) and orientation

59
Also, the area of the quadrilateral can be found by dividing the shape into four

triangles, as in the adjacent figure

area 𝐴1 𝐴2 𝐴3 𝐴4 = area of triangle 𝐴1 𝐴2 𝑍 + area of triangle 𝐴2 𝐴3 𝑍 + area of

triangle 𝐴3 𝐴4 𝑍 + area of triangle 𝐴4 𝐴1 𝑍

1 1 1 1
area 𝐴1 𝐴2 𝐴3 𝐴4 = |𝐴1 , 𝐴2 | + |𝐴2 , 𝐴3 | + |𝐴3 , 𝐴4 | + |𝐴4 , 𝐴1 |
2 2 2 2

This division of quadrilateral into triangles gives us the idea of calculating the
area of any polygon in 𝑅 2 .

Theorem 4.1.2 The area of a polygon with vertices

(𝑥1 , 𝑦1 ), (𝑥2 , 𝑦2 ), … , (𝑥𝑛 , 𝑦𝑛 ) listed counter-clockwise around the perimeter is

given by

1 𝑥1 𝑥2 𝑥2 𝑥3 𝑥𝑛 𝑥1
𝐴= (| | + | | + ⋯ + | 𝑦1 |)
2 𝑦1 𝑦2 𝑦2 𝑦3 𝑦𝑛

From the adjacent figure it is clear that the

area of the polygon is equal to the sum of the
areas of triangles resulting from two
consecutive vertices and the center of the
polygon

It is clear that every real 𝑚 × 𝑛 matrix 𝐴 = [𝐴1 , … , 𝐴𝑛 ] determines a polygon

in 𝑅 𝑚 (the columns of the matrix correspond to the vertices of the polygon) and
vice versa. The polygon which corresponds to the matrix [𝐴1 , … , 𝐴𝑛 ] will be
denoted by 𝐴1 … 𝐴𝑛 .

60
 1
area of 𝐴1 … 𝐴𝑛 = 2 (|𝐴1 , 𝐴2 | + |𝐴2 , 𝐴3 | + ⋯ + |𝐴𝑛−1 , 𝐴𝑛 | + |𝐴𝑛 , 𝐴1 |).

In the following we shall restrict ourselves to the case when 𝑚 = 2 (polygons in 𝑅 2 ).

Now, if 𝑛 = 3. The area 𝐴 with vertices (𝑥1 , 𝑦1 ), (𝑥2 , 𝑦2 ), (𝑥3 , 𝑦3 ) listed counter-
clockwise around the perimeter is given by

1 𝑥1 𝑥2 𝑥2 𝑥3 𝑥3 𝑥1
𝐴= (| 𝑦2 | + |𝑦2 𝑦3 | + |𝑦3 𝑦1 |)
2 𝑦1

1
= ((𝑥1 𝑦2 − 𝑥2 𝑦1 ) + (𝑥2 𝑦3 − 𝑥3 𝑦2 ) + (𝑥3 𝑦1 − 𝑥1 𝑦3 ))
2

1
= 2 (𝑥1 (𝑦2 − 𝑦3 ) − 𝑥2 (𝑦1 − 𝑦3 ) + 𝑥3 (𝑦1 − 𝑦2 ))

1 𝑥1 𝑥2 𝑥3
= 2 |𝑦 𝑦2 𝑦3 |
1

Theorem 4.1.3 The area 𝐴 of the polygon determined by 𝐴1 … 𝐴𝑛 is given by

1
𝐴 = (|𝐴1 , 𝐴2 | + |𝐴2 , 𝐴3 | + ⋯ + |𝐴𝑛−1 , 𝐴𝑛 | + |𝐴𝑛 , 𝐴1 |)
2

= |𝐴1 , 𝐴2 , … , 𝐴𝑛 |

Proof: What we need here is to prove the second equality, we shall proceed by
1
showing that the formula (|𝐴1 , 𝐴2 | + |𝐴2 , 𝐴3 | + ⋯ + |𝐴𝑛−1 , 𝐴𝑛 | + |𝐴𝑛 , 𝐴1 |) is
2

indeed a determinant function

1 1 0 0 1
A1: The area of a triangle with vertices (1,0), (0,1), (0,0) is | |=2
2 0 1 0
A2: It is clear that multiplying a single row by a constant 𝑐 means the dialation or
contraction of one coordinate while keeping the other coordinate fixed and in this
case new area equals |𝑐| times old area.

A3: A Let 𝐴 a polygon with vertices (𝑥1 , 𝑦1 ), (𝑥2 , 𝑦2 ), … , (𝑥𝑛 , 𝑦𝑛 ) such that
𝑥𝑖 = 𝑢𝑖 + 𝑣𝑖 , and 𝐵 a polygon with vertices (𝑢1 , 𝑦1 ), (𝑢2 , 𝑦2 ), … , (𝑢𝑛 , 𝑦𝑛 ), and
𝐶 a polygon with vertices (𝑣1 , 𝑦1 ), (𝑣2 , 𝑦2 ), … , (𝑣𝑛 , 𝑦𝑛 )

𝑢 + 𝑣 𝑢2 + 𝑣2 ⋯ 𝑢𝑛 + 𝑣𝑛
Suppose that, area of a polygon 𝐴 = [𝐴1 , … , 𝐴𝑛 ] = [ 1 𝑦 1 𝑦2 ⋯ 𝑦𝑛 ],
1

61
 𝑢1 𝑢2 ⋯ 𝑢𝑛
but, area of a polygon 𝐵 = [𝐵1 , … , 𝐵𝑛 ] = [ 𝑦 𝑦2 ⋯ 𝑦𝑛 ],
1

𝑣1 𝑣2 ⋯ 𝑣𝑛
area of a polygon 𝐶 = [𝐶1 , … , 𝐶𝑛 ] = [ 𝑦 𝑦2 ⋯ 𝑦𝑛 ],
1

1
that is, area of 𝐴1 … 𝐴𝑛 = 2 (|𝐴1 , 𝐴2 | + |𝐴2 , 𝐴3 | + ⋯ + |𝐴𝑛−1 , 𝐴𝑛 | + |𝐴𝑛 , 𝐴1 |)

1 𝑥1 𝑥2 𝑥2 𝑥3 𝑥𝑛 𝑥1
= (| 𝑦2 | + |𝑦2 𝑦3 | + ⋯ + |𝑦𝑛 𝑦1 |)
2 𝑦1

1 𝑢1 + 𝑣1 𝑢2 + 𝑣2 𝑢2 + 𝑣2 𝑢3 + 𝑣3 𝑢𝑛 + 𝑣𝑛 𝑢1 + 𝑣1
= (| 𝑦 𝑦2 | + | 𝑦2 𝑦3 | + ⋯ + | 𝑦𝑛 𝑦1 |)
2 1

(use theorem 3.1.1)

1 𝑢1 𝑢2 𝑣1 𝑣2 𝑢2 𝑢3 𝑣2 𝑣3 𝑢𝑛 𝑢1 𝑣𝑛 𝑣1
= ( |𝑦 𝑦2 | + | 𝑦1 𝑦2 | + | 𝑦2 𝑦3 | + | 𝑦2 𝑦3 | + ⋯ + | 𝑦𝑛 𝑦1 | + | 𝑦𝑛 𝑦1 |)
2 1

1 𝑢1 𝑢2 𝑢2 𝑢3 𝑢𝑛 𝑢1 1 𝑣1 𝑣2 𝑣2 𝑣3
= ( | 𝑦 𝑦2 | + | 𝑦2 𝑦3 | + ⋯ + | 𝑦𝑛 𝑦1 |) + (| | + | 𝑦3 | +
2 1 2 𝑦1 𝑦2 𝑦2
𝑣𝑛 𝑣1
|𝑦 𝑦1 |)
𝑛

area of 𝐴1 … 𝐴𝑛

1 1
= (|𝐵1 , 𝐵2 | + |𝐵2 , 𝐵3 | + ⋯ + |𝐵𝑛 , 𝐵1 |) + (|𝐶1 , 𝐶2 | + |𝐶2 , 𝐶3 | + ⋯ + |𝐶𝑛 , 𝐶1 |)
2 2

area of 𝐴1 … 𝐴𝑛 = area of 𝐵1 … 𝐵𝑛 + area of 𝐶1 … 𝐶𝑛

A4: Exchanging two rows geometrically means the reflection of the heads of the
polygon about in the straight line 𝑦 = 𝑥, That is, the image of each point (𝑥, 𝑦)
0 1 𝑥 𝑦
under the transformation is [ ] [𝑦] = [ ], which keeps the area fixed.
1 0 𝑥

Since the area of a polygon is a determinant function, which is unique by

Theorem 2.1.4, we obtain that the area equals the previously defined determinant
( the cofactor definition and Radic definition ).

62
Example 4.1.4 Use matrices to find the area of triangle with vertices:

(4, 0), (7, 2), (2, 3)

1 4 7 2 13
area of triangle = | |= 2
2 0 2 3

4.2 Some properties of the determinant and their geometric interpretation

Theorem 4.2.1 [13 ]. Let 𝐴1 … 𝐴𝑛 be a polygon in 𝑅 2 . Then

1
area of 𝐴1 … 𝐴𝑛 = |𝐴1 + 𝐴2 , 𝐴2 + 𝐴3 , … , 𝐴𝑛−1 + 𝐴𝑛 , 𝐴𝑛 + 𝐴1 |
2

Proof: We need to show that

|𝐴1 + 𝐴2 , 𝐴2 + 𝐴3 , … , 𝐴𝑛−1 + 𝐴𝑛 , 𝐴𝑛 + 𝐴1 | =

|𝐴1 , 𝐴2 | + |𝐴2 , 𝐴3 | + ⋯ + |𝐴𝑛−1 , 𝐴𝑛 | + |𝐴𝑛 , 𝐴1 |

The proof will use the method of mathematical induction.

First, we have the theorem holds for 𝑛 = 3, that is

From the adjacent figure it appears that the triangular shape resulted from the vertices shift

process and therefore,

1
area of 𝐴1 𝐴2 𝐴3 = |𝐴1 + 𝐴2 , 𝐴2 + 𝐴3 , 𝐴3 + 𝐴1 |
2

1
= 2 (|𝐴1 + 𝐴2 , 𝐴2 + 𝐴3 | + |𝐴2 + 𝐴3 , 𝐴3 + 𝐴1 | + |𝐴3 + 𝐴1 , 𝐴1 + 𝐴2 |)

63
Applying Theorem 3.1.10 with 𝑋1 = −𝐴2 , 𝑋2 = −𝐴3 , 𝑋3 = −𝐴1 on the

determinants in order

1
area of 𝐴1 𝐴2 𝐴3 = 2 (|𝐴1 , 𝐴3 | + |𝐴2 , 𝐴1 | + |𝐴3 , 𝐴2 |)

1
= (|𝐴1 , 𝐴2 | + |𝐴2 , 𝐴3 | + |𝐴3 , 𝐴1 |)
2

Second, for 𝑛 ≥ 3 . Assume that is true for 𝑛 = 𝑘, by theorem 2.5.7

|𝐴1 + 𝐴2 , 𝐴2 + 𝐴3 , … , 𝐴𝐾−1 + 𝐴𝐾 , 𝐴𝑘 + 𝐴1 |

= |𝐴1 + 𝐴2 , 𝐴2 + 𝐴3 , … , 𝐴𝑘−1 + 𝐴𝑘 | + (−1)𝑘 |𝐴1 + (−1)𝑘 𝐴𝑘 , 𝐴𝑘 + 𝐴1 |

= |𝐴1 + 𝐴2 , 𝐴2 + 𝐴3 , … , 𝐴𝑘−1 + 𝐴𝑘 | + (−1)𝑘 |𝐴1 , 𝐴𝑘 | + |𝐴𝑘 , 𝐴1 |

= |𝐴1 , 𝐴2 | + |𝐴2 , 𝐴3 | + ⋯ + |𝐴𝑘−1 , 𝐴𝑘 | + |𝐴𝑘 , 𝐴1 | + (−1)𝑘 |𝐴1 , 𝐴𝑘 | + |𝐴𝑘 , 𝐴1 |

We shall prove the result holds for 𝑘 + 1

𝑑𝑒𝑡[𝐴1 + 𝐴2 , 𝐴2 + 𝐴3 , … , 𝐴𝑘−1 + 𝐴𝑘 , 𝐴𝑘 + 𝐴𝑘+1 , 𝐴𝑘+1 + 𝐴1 ]

= det[𝐴1 + 𝐴2 , 𝐴2 + 𝐴3 , … , 𝐴𝑘−1 + 𝐴𝑘 ]

+ (−1)𝑘 det[𝐴1 + (−1)𝑘 𝐴𝑘 , 𝐴𝑘 + 𝐴1 ] + det[𝐴𝑘 + 𝐴𝑘+1 , 𝐴𝑘+1 + 𝐴1 ]

= det[𝐴1 + 𝐴2 , 𝐴2 + 𝐴3 , … , 𝐴𝑘−1 + 𝐴𝑘 ] + (−1)𝑘 det[𝐴1 , 𝐴𝑘 ] − det[𝐴𝑘 , 𝐴1 ]

+ det[𝐴𝑘 , 𝐴𝑘+1 ] + det[𝐴𝑘 , 𝐴1 ] + det[𝐴𝑘+1 , 𝐴1 ]

= det[𝐴1 + 𝐴2 , 𝐴2 + 𝐴3 , … , 𝐴𝑘−1 + 𝐴𝑘 ] + (−1)𝑘 det[𝐴1 , 𝐴𝑘 ] + det[𝐴𝑘 , 𝐴𝑘+1 ]

+ det[𝐴𝑘+1 , 𝐴1 ]

= det[𝐴1 , 𝐴2 ] + det[𝐴2 , 𝐴3 ] + ⋯ + det[𝐴𝑘 , 𝐴𝑘+1 ] + det[𝐴𝑘+1 , 𝐴1 ]

64
Example 4.2.2 Use Theorem 4.2.1 to compute the area of the polygon in 𝑅 2 with
vertices (0, 0), (1, 0), (1, 1), (0 ,1)

1
Now, Area 𝐴1 𝐴2 𝐴3 𝐴4 = 2 |𝐴1 + 𝐴2 , 𝐴2 + 𝐴3 , 𝐴3 + 𝐴4 , 𝐴4 + 𝐴1 |

1 1 2 1 0
= | |=1
2 0 1 2 1

Corollary 4.2.3 [13 ]. If 𝑛 is odd, then for every point 𝑋 in 𝑅 2 it holds

|𝐴1 + 𝑋, … , 𝐴𝑛 + 𝑋| = |𝐴1 , … , 𝐴𝑛 |

Proof: Since 𝑚 = 2, if 𝑛 is odd, then 𝑚 + 𝑛 is odd, by theorem 3.1.10 it holds

that is, the area of the polygon 𝐴1 𝐴2 … 𝐴𝑛 will not be changed when a fixed
vector 𝑋 = (𝑥, 𝑦) is subtracted from all the vertices (heads) of the polygon.

Example 4.2.4 The area of a polygon with vertices

(1 , 3), (2 , 0), (7 , 1), (4 , −1), (3 , 2)

when all the heads are shifted by an amount 𝑋 = (1,3)

1 2 7 4 3 2 3 8 5 4
| |=| | = −3
3 0 1 −1 2 6 3 4 2 5

area of a polygon = 3

65
Theorem 4.2.5 [13]. Let 𝐴1 … 𝐴𝑛 be a polygon in 𝑅 2 with odd 𝑛.

Then

|𝐴1 , … , 𝐴𝑛 | = |𝐴𝑛 , 𝐴1 , … , 𝐴𝑛−1 |

Proof: Since 𝑚 = 2 , if 𝑛 is odd, then 𝑚 + 𝑛 is odd.

by Cyclic theorem, (−1)(𝑖+1)𝑚 𝑑𝑒𝑡[𝐴𝑖 , … , 𝐴𝑛 , … , 𝐴𝑖−1 ] = 𝑑𝑒𝑡[𝐴1 , … , 𝐴𝑛 ]

When 𝑖 = 𝑛, then (𝑖 + 1)𝑚 is even

So, |𝐴1 , … , 𝐴𝑛 | = |𝐴𝑛 , 𝐴1 , … , 𝐴𝑛−1 |.

Example 4.2.6 Use Theorem 4.2.5 to find the area of a polygon with vertices:

𝐴1 = (3,2), 𝐴2 = (5,1), 𝐴3 = (7,4)

Then, area of a polygon 𝐴1 𝐴2 𝐴3 =

3 5 7 7 3 5
| |=| |=8
2 1 4 4 2 1

Note, we can calculate the area by starting

from any vertex of the polygon paying attention to the direction and arrangement
of the vertices.

Area of a polygon 𝐴1 𝐴2 𝐴3 = det(𝐴1 , 𝐴2 , 𝐴3 ) = det(𝐴2 , 𝐴3 , 𝐴1 ) = det(𝐴3 , 𝐴1 , 𝐴2 )

Theorem 4.2.7 [13 ]. Let 𝐴1 … 𝐴𝑛 be a polygon in 𝑅 2 and let 𝑛 be an even integer.

Then for any point 𝑋 in 𝑅 2 it holds

|𝐴1 + 𝑋, … , 𝐴𝑛 + 𝑋| = |𝐴1 , … , 𝐴𝑛 |

Only if ∑𝑛𝑖=1(−1)𝑖 𝐴𝑖 = 0.

Proof: This theorem is incompatible with the theorem 3.1.10 and the reason is
𝑚 = 2, if 𝑛 is even, then 𝑚+𝑛 is even, This does not meet the theorem 3.1.10
requirement

66
Now, by Theorem 2.5.7, it is clear that for any point 𝑝 in 𝑅 2 holds

|𝐴1 , 𝐴2 , … , 𝐴𝑛 , 𝑝| = |𝐴1 , 𝐴2 , … , 𝐴𝑛 | + (−1)𝑛+1 |𝐴1 − 𝐴2 + ⋯ + (−1)𝑛+1 𝐴𝑛 , 𝑝|

= |𝐴1 , 𝐴2 , … , 𝐴𝑛 | (1)

Only if 𝐴1 − 𝐴2 + ⋯ − 𝐴𝑛 = 0.

Now, by theorem 3.1.10, since 𝑚 + 𝑛 + 1 is odd, taking 𝑋 = −𝑝, we can write

|𝐴1 , 𝐴2 , … , 𝐴𝑛 , 𝑝| = |𝐴1 + (−𝑝), 𝐴2 + (−𝑝), … , 𝐴𝑛 + (−𝑝), 𝑝 + (−𝑝)|

= |𝐴1 + (−𝑝), 𝐴2 + (−𝑝), … , 𝐴𝑛 + (−𝑝), 02 |

By Lemma 3.1.8

= |𝐴1 + (−𝑝), 𝐴2 + (−𝑝), … , 𝐴𝑛 + (−𝑝)|

Putting 𝑋 = −𝑝, we get

|𝐴1 , 𝐴2 , … , 𝐴𝑛 , 𝑝| = |𝐴1 + 𝑋, 𝐴2 + 𝑋, … , 𝐴𝑛 + 𝑋|

since |𝐴1 , 𝐴2 , … , 𝐴𝑛 , 𝑝| = |𝐴1 , 𝐴2 , … , 𝐴𝑛 | ( from 1 )

We get,
|𝐴1 , 𝐴2 , … , 𝐴𝑛 | = |𝐴1 + 𝑋, 𝐴2 + 𝑋, … , 𝐴𝑛 + 𝑋|

That is, the area of the polygon 𝐴1 𝐴2 … 𝐴𝑛 will not be changed when a fixed
vector 𝑋 = (𝑥, 𝑦) is subtracted from all the vertices (heads) of the polygon.

When for example 𝑋 = −𝐴1 , we obtain

|𝐴1 , 𝐴2 , … , 𝐴𝑛 | = |0, 𝐴2 − 𝐴1 , … , 𝐴𝑛 − 𝐴1 |

= |𝐴1 − 𝐴2 , … , 𝐴1 − 𝐴𝑛 |

= |𝐵1 , 𝐵2 , … , 𝐵𝑛−1 |

1
= |𝐵 , 𝐵 | + |𝐵2 , 𝐵3 | + ⋯ + |𝐵𝑛−1 , 𝐵1 |
2 1 2

67
Example 4.2.8 Use Theorem 4.2.7 to find the area of a polygon with vertices:

𝐴1 = (1,2), 𝐴2 = (3,5), 𝐴3 = (4,2), 𝐴4 = (2, −1)

Since −𝐴1 + 𝐴2 − 𝐴3 + 𝐴4 = 0

area of a polygon
1 3 4 2
𝐴1 𝐴2 𝐴3 𝐴4 = || || = |−9 | = 9
2 5 2 −1

We note that when we move on the vertices of the polygon in opposite

direction but in the same order, we obtain the same area but opposite value of the
determinant.

While the original space recedes itself, meaning

1 2 4 3
area of a polygon 𝐴1 𝐴4 𝐴3 𝐴2 = | |=9
2 −1 2 5

Now, by an amount 𝑋 = (1,2)

area of a polygon (𝐴1 + 𝑋)(𝐴2 + 𝑋) (𝐴3 + 𝑋)(𝐴4 + 𝑋)

2 4 5 3
= || || = |−9 | = 9
4 7 4 1

Theorem 4.2.9 [13]. Let 𝐴1 … 𝐴𝑛 be a polygon in 𝑅 2 with even 𝑛 and let

∑𝑛𝑖=1(−1)𝑖 𝐴𝑖 = 0. Then

|𝐴1 , … , 𝐴𝑛 | = |𝐴𝑛 , 𝐴1 , … , 𝐴𝑛−1 |

Proof: If 𝑛 is even, by theorem 4.1.10, taking 𝑋 = −𝐴𝑛 , we can write

|𝐴1 , … , 𝐴𝑛 | = |𝐴1 − 𝐴𝑛 , … , 𝐴𝑛−1 − 𝐴𝑛 , 𝐴𝑛 − 𝐴𝑛 |

= |𝐴1 − 𝐴𝑛 , … , 𝐴𝑛−1 − 𝐴𝑛 , 02 | (see lemma 3.1.8)

= |02 , 𝐴𝑛 − 𝐴1 , … , 𝐴𝑛 − 𝐴𝑛−1 |

68
Adding −𝐴𝑛 to each column

𝑑𝑒𝑡(𝐴1 , … , 𝐴𝑛 ) = 𝑑𝑒𝑡(−𝐴𝑛 , −𝐴1 , … , −𝐴𝑛−1 )

Take out a common negative signal factor from the two rows

𝑑𝑒𝑡(𝐴1 , … , 𝐴𝑛 ) = 𝑑𝑒𝑡(𝐴𝑛 , 𝐴1 , … , 𝐴𝑛−1 )

Note that, this says that we can calculate the area by starting from any vertex of
the polygon paying attention to the direction and arrangement of the vertices.

Example 4.2.10 Use Theorem 4.2.9 to find the area of a polygon with vertices:

𝐴1 = (1,2), 𝐴2 = (3,5), 𝐴3 = (4,2), 𝐴4 = (2, −1)

Since −𝐴1 + 𝐴2 − 𝐴3 + 𝐴4 = 0

1 3 4 2 2 1 3 4
|| || = || || = 9
2 5 2 −1 −1 2 5 2

We note that when we move on the vertices of the polygon in opposite

direction but in the same order, we obtain the same area but opposite value of the
determinant.

Theorem 4.2.11 [13]. Let 𝐴1 … 𝐴𝑛 be a polygon in 𝑅 2 with even 𝑛 and let

∑𝑛𝑖=1(−1)𝑖 𝐴𝑖 = 0. Then

|𝐴1 , … , 𝐴𝑛 | = |𝐴1 , … , 𝐴𝑛−1 |

Proof: By Theorem 2.5.7,

|𝐴1 , 𝐴2 , … , 𝐴𝑛 | = |𝐴1 , 𝐴2 , … , 𝐴𝑛−1 | + (−1)𝑛 |𝐴1 − 𝐴2 + ⋯ + 𝐴𝑛−1 , 𝐴𝑛 |

= |𝐴1 , 𝐴2 , … , 𝐴𝑛−1 |

since ∑𝑛𝑖=1(−1)𝑖 𝐴𝑖 = 0, we get 𝐴1 − 𝐴2 + ⋯ + 𝐴𝑛−1 = 𝐴𝑛 .

That means, |𝐴1 , 𝐴2 , … , 𝐴𝑛 | = |𝐴1 , 𝐴2 , … , 𝐴𝑛−1 |

69
 We notice that if we displace the shape to the original point and calculate the
area by dividing the shape into such based on the origin point, the area does not
change.

Example 4.2.12 If 𝐴1 𝐴2 𝐴3 𝐴4 is parallelogram, then

𝑑𝑒𝑡[ 𝐴1 , 𝐴2 , 𝐴3 , 𝐴4 ] = 𝑑𝑒𝑡[ 𝐴1 , 𝐴2 , 𝐴3 ]

Now, | 𝐴1 , 𝐴2 , 𝐴3 | = | 𝐴1 , 𝐴2 | − | 𝐴1 , 𝐴3 | + | 𝐴2 , 𝐴3 |

= | 𝐴1 , 𝐴2 | + |−𝐴1 + 𝐴2 , 𝐴3 |

Since 𝑛 is even, we get −𝐴1 + 𝐴2 − 𝐴3 + 𝐴4 = 0

| 𝐴1 , 𝐴2 , 𝐴3 | = | 𝐴1 , 𝐴2 | + |𝐴3 − 𝐴4 , 𝐴3 |

= | 𝐴1 , 𝐴2 | + |− 𝐴4 , 𝐴3 | = | 𝐴1 , 𝐴2 | + | 𝐴3 , 𝐴4 |

| 𝐴1 , 𝐴2 , 𝐴3 , 𝐴4 | = | 𝐴1 , 𝐴2 | + | 𝐴3 , 𝐴4 |

Example 4.2.13 Use Theorem 4.2.11 to find the area of a polygon with vertices:

𝐴1 = (1,2), 𝐴2 = (3,5), 𝐴3 = (4,2), 𝐴4 = (2, −1)

Since −𝐴1 + 𝐴2 − 𝐴3 + 𝐴4 = 0
|𝐴1 , 𝐴2 , 𝐴3 , 𝐴4 | = |𝐴1 − 𝐴4 , 𝐴2 − 𝐴4 , 𝐴3 − 𝐴4 , 𝐴4 − 𝐴4 |

= |𝐴1 − 𝐴4 , 𝐴2 − 𝐴4 , 𝐴3 − 𝐴4 , 0| = |𝐴1 − 𝐴4 , 𝐴2 − 𝐴4 , 𝐴3 − 𝐴4 |

−1 1 2 0 −1 1 2
=| |=| | = −9
3 6 3 0 3 6 3

area of a polygon 𝐴1 𝐴2 𝐴3 𝐴4 = 9

70
Note that the point can be considered as the center on which to divide triangles to
calculate the area

then,

1 3 4 2 1 3 4
area of a polygon 𝐴1 𝐴2 𝐴3 𝐴4 = || || = || || = |−9| = 9
2 5 2 −1 2 5 2

Theorem 4.2.14 [13 ]. Let 𝐴1 … 𝐴𝑛 be a polygon in 𝑅 2 with even 𝑛 and let

∑𝑛𝑖=1(−1)𝑖 𝐴𝑖 = 0. Then

|𝐴1 , … , 𝐴𝑛 | = |𝐴1 , … , 𝐴𝑘 | + |𝐴𝑘+1 , … , 𝐴𝑛 |,

Where 𝑘 may be any integer such that 1 < 𝑘 < 𝑛.

Proof:

area of 𝐴1 … 𝐴𝑛

1
= 2 (|𝐴1 , 𝐴2 | + |𝐴2 , 𝐴3 | + ⋯ + |𝐴𝑘−1 , 𝐴𝑘 | + |𝐴𝑘 , 𝐴𝑘+1 | + |𝐴𝑘+1 , 𝐴𝑘+2 | + ⋯ +

|𝐴𝑛−1 , 𝐴𝑛 | + |𝐴𝑛 , 𝐴1 |)

1
= 2 (|𝐴1 , 𝐴2 | + |𝐴2 , 𝐴3 | + ⋯ + |𝐴𝑘−1 , 𝐴𝑘 | + |𝐴𝑘+1 , 𝐴𝑘+2 | + ⋯ + |𝐴𝑛−1 , 𝐴𝑛 |) +
1
(|𝐴𝑘 , 𝐴𝑘+1 | + |𝐴𝑛 , 𝐴1 |)
2

1
|𝐴1 , … , 𝐴𝑛 | = |𝐴1 , … , 𝐴𝑘 | + |𝐴𝑘+1 , … , 𝐴𝑛 | + (|𝐴𝑘 , 𝐴𝑘+1 | + |𝐴𝑛 , 𝐴1 |) −
2

1
(|𝐴𝑘 , 𝐴1 | + |𝐴𝑛 , 𝐴𝑘+1 |)
2

1
let ∆ = 2 (|𝐴𝑘 , 𝐴𝑘+1 | + |𝐴𝑛 , 𝐴1 | − |𝐴𝑘 , 𝐴1 | − |𝐴𝑛 , 𝐴𝑘+1 |)

|𝐴1 , … , 𝐴𝑛 | = |𝐴1 , … , 𝐴𝑘 | + |𝐴𝑘+1 , … , 𝐴𝑛 | + ∆

But, if ∑𝑛𝑖=1(−1)𝑖 𝐴𝑖 = 0, then ∆= 0 ( see figure)

Then, |𝐴1 , … , 𝐴𝑛 | = |𝐴1 , … , 𝐴𝑘 | + |𝐴𝑘+1 , … , 𝐴𝑛 |

71
Example 4.2.15 Use Theorem 4.2.14 to find the area of a polygon with vertices:

𝐴1 = (1,2), 𝐴2 = (0,1), 𝐴3 = (1,0), 𝐴4 = (2, −1), 𝐴5 = (3, −1), 𝐴6 = (3,1)

Since

−𝐴1 + 𝐴2 − 𝐴3 + 𝐴4 + 𝐴5 − 𝐴6 = 0

Then,

area of a polygon 𝐴1 𝐴2 𝐴3 𝐴4 𝐴5 𝐴6 =
1 0 1 2 3 3
|| || = 4
2 1 0 −1 −1 1

take 𝑘 = 3 then

area of a polygon 𝐴1 𝐴2 𝐴3 + area of a polygon 𝐴4 𝐴5 𝐴6

1 0 1 2 3 3
= || || + || || = 2 + 2 = 4
2 1 0 −1 −1 1

take 𝑘 = 2 then

area of a polygon 𝐴1 𝐴2 + area of a polygon 𝐴3 𝐴4 𝐴5 𝐴6

1 0 1 2 3 3
= || || + || || = 1 + 3 = 4
2 1 0 −1 −1 1

Corollary 4.2.16 [13]. Let 𝐵1 , … , 𝐵𝑛 be given by

𝐴1 +𝐴2 𝐴2 +𝐴3 𝐴𝑛 +𝐴1

𝐵1 = , 𝐵2 = , … . , 𝐵𝑛 = .
2 2 2

Then

4|𝐵1 , … , 𝐵𝑛 | = |𝐴1 , 𝐴2 | + |𝐴2 , 𝐴3 | + ⋯ + |𝐴𝑛 , 𝐴1 |

72
 𝐴1 +𝐴2 𝐴2 +𝐴3 𝐴𝑛 +𝐴1
Proof: 4|𝐵1 , … , 𝐵𝑛 | = 4 | , ,…, |.
2 2 2

By theorem 4.2.1

1 1 1
4|𝐵1 , … , 𝐵𝑛 | = 4 ( |𝐴1 , 𝐴2 | + |𝐴2 , 𝐴3 | + ⋯ + |𝐴𝑛 , 𝐴1 |)
4 4 4

= |𝐴1 , 𝐴2 | + |𝐴2 , 𝐴3 | + ⋯ + |𝐴𝑛 , 𝐴1 |

Example 4.2.17 From the adjacent figure we find that

1
area of a polygon 𝐵1 𝐵2 𝐵3 𝐵4 = area of a polygon 𝐴1 𝐴2 𝐴3 𝐴4
2

Use Theorem 4.1.3 to find the area of a polygon with vertices:

𝐴1 = (1,2), 𝐴2 = (3,5), 𝐴3 = (4,2), 𝐴4 = (2, −1)

Since −𝐴1 + 𝐴2 − 𝐴3 + 𝐴4 = 0

Then, area of a polygon 𝐴1 𝐴2 𝐴3 𝐴4 = 9

area of a polygon with vertices

𝐵1 = (2,3.5), 𝐵2 = (3.5,3.5), 𝐵3 = (3,0.5), 𝐵4 = (1.5,0.5)

Since −𝐵1 + 𝐵2 − 𝐵3 + 𝐵4 = 0

2 3.5 3 1.5
Then, area of a polygon 𝐵1 𝐵2 𝐵3 𝐵4 = || || = 4.5
3.5 3.5 0.5 0.5

That means, by Corollary 4.2.16

1
area of a polygon 𝐵1 𝐵2 𝐵3 𝐵4 = area of a polygon 𝐴1 𝐴2 𝐴3 𝐴4 .
2

73
 Chapter Five

Inverse for non-square matrix

In linear algebra, the inverse of a matrix is defined only for square matrices,
and if a matrix is singular, it does not have an inverse.

The aim of this chapter is the discussion of existence of inverses for non-
square matrices. We know that the fundamental idea for existence of inverse of
matrix it must be nonsingular. (it has non-zero determinant).

5.1 Right inverse or left inverse of a matrix

Although non-square matrices do not have inverses (both sides inverse),

some of them have one side inverses. For this reason we introduce the concepts of
"left inverse" and "right inverse"

Definition 5.1.1 [3 , 𝑝. 397 ].

(i) A non-singular non-square matrix 𝐴 has a left inverse if there exists a matrix
𝐴−1
𝐿 such that 𝐴−1
𝐿 𝐴 = 𝐼, where 𝐼 denote the identity matrix.

(ii) A non-singular non-square matrix 𝐴 has a right inverse if there exists a matrix
𝐴−1
𝑅 such that 𝐴 𝐴−1
𝑅 = 𝐼, where 𝐼 denote the identity matrix.

We note here a right inverse of 𝐴𝑚×𝑛 (𝑚 < 𝑛) is an 𝐴−1

𝑅 matrix, where a left

inverse of 𝐴𝑚×𝑛 (𝑚 > 𝑛) is an 𝐴−1

𝐿 matrix.

1 3 0
Example 5.1.2 Let A = [ ] Find a right inverse of A
2 2 1

Solution: Let 𝐴−1

𝑅 be a a right inverse of 𝐴, then

𝑥1 𝑦1
𝐴 𝐴−1
𝑅 = [
1 3 0 𝑥
] [ 2 𝑦2 ] = [1 0]
2 2 1 𝑥 𝑦3 0 1
3

𝑥1 + 3𝑥2 𝑦1 + 3𝑦2 1 0
[ ]=[ ]
2𝑥1 + 2𝑥2 + 𝑥3 2𝑦1 + 2𝑦2 + 𝑦3 0 1

74
Gauss-Jordan reduction Using we obtain the following

𝑥1 + 34 𝑥3 = −1
2
, 𝑥2 − 14 𝑥3 = 12 , 𝑦1 + 34 𝑦3 = 34 , 𝑦2 − 14 𝑦3 = −1
4
(*)

This system has infinitely many solutions, one solution gives the following

1 1 0 0
−1⁄ 13
𝐴−1
𝑅 = [0 3] , another solution is 𝐴−1
𝑅 = [
⁄32 0]
−2 −1⁄ −3⁄ 1
3 8

Computing a right inverse of horizontal matrix always can be transformed to

finding a solution of a linear system with 𝑚 equations and 𝑛 variable (𝑚 < 𝑛).

Note: When a right inverse or a left inverse for a non-singular non-square matrix
exists, it is not unique.

Remark 5.1.3 For any non-singular square matrix 𝐴, left inverse and right inverse
exists and it is equal to inverse of 𝐴, that is

1
𝐴−1 −1
𝑅 = 𝐴𝐿 = 𝐴
−1
= det(𝐴) 𝑎𝑑𝑗(𝐴).

The following theorem is from [3], but we give here another proof

Theorem 5.1.4 Every non-singular horizontal matrix 𝐴 has a right inverse 𝐴−1
𝑅 given

by

1
𝐴−1
𝑅 = 𝑎𝑑𝑗(𝐴)
det(𝐴)

Proof:
𝑎11 𝑎12 ⋯ 𝑎1𝑛
𝑎21 𝑎22 ⋯ 𝑎2𝑛 𝑀11 𝑀21 ⋯ 𝑀𝑗1 ⋯ 𝑀𝑚1
⋮ ⋮ ⋱ ⋮ 𝑀 𝑀22 … 𝑀𝑗2 ⋯ 𝑀𝑚2
Let 𝐴 = 𝑎𝑖1 𝑎𝑖2 ⋯ 𝑎𝑖𝑛 , 𝑎𝑑𝑗(𝐴) = [ 12 ]
⋮ ⋮ ⋱ ⋮ ⋱ ⋮
⋮ ⋮ ⋱ ⋮ 𝑀1𝑛 𝑀2𝑛 … 𝑀𝑗𝑛 ⋯ 𝑀𝑚𝑛
[𝑎𝑚1 𝑎𝑚2 ⋯ 𝑎𝑚𝑛 ]
Since 𝐴 is non-singular, 𝑑𝑒𝑡(𝐴) is non-zero the matrix

75
 𝑀11 𝑀21 ⋯ 𝑀𝑗1 ⋯ 𝑀𝑚1
1 1 𝑀 𝑀22 … 𝑀𝑗2 ⋯ 𝑀𝑚2
𝐵= 𝑎𝑑𝑗(𝐴) = [ 12 ]
det(𝐴) det(𝐴) ⋮ ⋮ ⋱ ⋮ ⋱ ⋮
𝑀1𝑛 𝑀2𝑛 … 𝑀𝑗𝑛 ⋯ 𝑀𝑚𝑛

satisfies

𝑎11 𝑎12 ⋯ 𝑎1𝑛

𝑎21 𝑎22 ⋯ 𝑎2𝑛 𝑀11 𝑀21 ⋯ 𝑀𝑗1 ⋯ 𝑀𝑚1
1 ⋮ ⋮ ⋱ ⋮ 𝑀12 𝑀22 … 𝑀𝑗2 ⋯ 𝑀𝑚2
𝐴𝐵 = 𝑎 𝑎𝑖2 ⋯ 𝑎𝑖𝑛 [ ⋮ ]
det(𝐴) 𝑖1 ⋮ ⋱ ⋮ ⋱ ⋮
⋮ ⋮ ⋱ ⋮ 𝑀1𝑛 𝑀2𝑛 … 𝑀𝑗𝑛 ⋯ 𝑀𝑚𝑛
[𝑎𝑚1 𝑎𝑚2 ⋯ 𝑎𝑚𝑛 ]

The (𝑖 , 𝑗 )𝑡ℎ element in the product matrix 𝐴 𝐵 is (**)

𝑖𝑓 𝑖 = 𝑗, 𝑎𝑖1 𝑀𝑗1 + 𝑎𝑖2 𝑀𝑗2 + ⋯ + 𝑎𝑖𝑛 𝑀𝑗𝑛 = det(𝐴) ( theorem 2.3.2)

and, 𝑖𝑓 𝑖 ≠ 𝑗, we need to show that 𝑎𝑖1 𝑀𝑗1 + 𝑎𝑖2 𝑀𝑗2 + ⋯ + 𝑎𝑖𝑛 𝑀𝑗𝑛 = 0

Consider the matrix 𝐵 obtained from 𝐴 by replacing the 𝑘 𝑡ℎ row of 𝐴 by its

𝑖 𝑡ℎ row. Thus 𝐵 is a matrix having two identical rows the 𝑖 𝑡ℎ and 𝑘 𝑡ℎ rows. Then
𝑑𝑒𝑡(𝐵) = 0. Now expand 𝑑𝑒𝑡(𝐵) about the 𝑘 𝑡ℎ row, the elements of the 𝑘 𝑡ℎ row
of 𝐵 are 𝑎𝑖1 , 𝑎𝑖2 , … , 𝑎𝑖𝑛 . The cofactors of the 𝑘 𝑡ℎ row are 𝑀𝑘1 , 𝑀𝑘2 , … , 𝑀𝑘𝑛 .

We have, 0 = det(𝐵) = 𝑎𝑖1 𝑀𝑘1 + 𝑎𝑖2 𝑀𝑘2 + ⋯ + 𝑎𝑖𝑛 𝑀𝑘𝑛

This means that

det(𝐴) 0 ⋯ 0
1 0
𝐴𝐵 = [ 0 det(𝐴) ⋯
⋮ ] = 𝐼𝑚
det(𝐴) ⋮ ⋮ ⋱
0 0 ⋯ det(𝐴)
Hence, the matrix 𝐵 is a right inverse of 𝐴.

1 3 0
In example 5.1.2 we have seen that a right inverse for A= [ ] is
2 2 1
1 1
−1
𝐴−1
𝑅 = [0 ⁄3] here we compute an inverse of the same 𝐴 by theorem 5.1.4
−2 −1⁄3

−1⁄ 3⁄
2 2
which is 𝐴−1
𝑅 = [ 1⁄ −1⁄ ] . it is easy to see that the resulting inverse is
2 2
0 1
another solution to the system (*).

76
 1 1 2 0
Example 5.1.5 Let 𝐴 = [ 1 2 1 2]
3 4 1 2

to find a right inverse of 𝐴, 𝐴−1

𝑅 , by applying theorem 5.1.4 we first compute

cofactors of 𝐴 , det(A) = −4 and the result is

−1⁄ −5⁄ 1⁄
2 4 4
2 5 −1 1 3⁄ 1⁄
1 1 −2 −3 −1 ⁄2 4 4
𝐴−1
𝑅 = 𝑎𝑑𝑗(𝐴) = [ ]=
det(𝐴) −4 −2 −1 1 1⁄ 1⁄ −1⁄
2 −1 1 2 4 4
−1⁄ 1⁄ −1⁄
[ 2 4 4]

Theorem 5.1.6 [3 ] Every non-singular vertical matrix 𝐴 has a left inverse 𝐴−1
𝐿 , such

that

1
𝐴−1
𝐿 = 𝑎𝑑𝑗(𝐴)
det(𝐴)

Proof: similar to the proof of theorem 5.1.4

The following theorem describes the case that 𝐴 has both a right inverse and a left inverse.

Theorem 5.1.7 If 𝐴 is an 𝑚 × 𝑛 matrix such that both 𝐴−1

𝑅 and 𝐴−1
𝐿 exist, then
𝑚 = 𝑛 (so 𝐴 is square). Moreover, 𝐴 is invertible and 𝐴 −1 = 𝐴−1 −1
𝑅 = 𝐴𝐿 .

Proof: Let 𝐴 be an 𝑚 × 𝑛 matrix

If 𝐴 𝐴−1
𝑅 = 𝐼𝑚 , then the equation 𝑨𝒙 = 𝒃 has a solution for every possible 𝒃 in 𝑅
𝑚

(given b, just let 𝒙 = 𝐴−1 −1

𝑅 𝒃, then 𝑨𝒙 = 𝑨(𝐴𝑅 𝒃) = 𝐼𝑚 𝒃 = 𝒃.

Since b is arbitrary, in particular, for all 𝑖 = 1, … , 𝑚, the system 𝑨𝒙 = 𝑒𝑖

(where 𝑒𝑖 = (0, … ,1, … ,0)𝑇 , the 1 in the 𝑖th component) has a solution,
say 𝑟𝑖 for every 𝑖, that is,
𝐴𝑟1 = 𝑒1 , 𝐴𝑟2 = 𝑒2 , … , 𝐴𝑟𝑚 = 𝑒𝑚 .

Let 𝐴−1
𝑅 = [𝑟1 ⋯ 𝑟𝑚 ]. Then

𝐴𝐴−1
𝑅 = 𝐴 ⋅ [𝑟1 ⋯ 𝑟𝑚 ] = [𝐴 𝑟1 ⋯ 𝐴 𝑟𝑚 ] = [𝑒1 ⋯ 𝑒𝑚 ] = 𝐼𝑚 .

77
 Therefore 𝐴 has a pivot position in every row. This force 𝑚 ≤ 𝑛, since every
pivot position must be in a different column.

If 𝐴−1 −1 −1
𝐿 𝐴 = 𝐼𝑛 , consider the equation 𝑨𝒙 = 0. Then 𝐴𝐿 𝑨 𝒙 = 𝐴𝐿 0 = 0.

But 𝐴−1
𝐿 𝑨 𝒙 = 𝐼𝑛 𝒙 = 𝒙, so 𝒙 = 𝟎. In other words, 𝑨𝒙 = 𝟎 has a unique solution
and therefore the columns of 𝐴 must be linearly independent (see definition 1.9.3)
and therefore each column must be a pivot position column. Since each pivot
position must be in a different row, these forces 𝑛 ≤ 𝑚.

So, combining the two paragraphs gives that 𝑚 = 𝑛. Since 𝐴 is now known to be
square, 𝐴 is invertible and 𝐴 −1 = 𝐴−1 −1
𝑅 = 𝐴𝐿 .

5.2 Properties for inverse and adjoint of non- square matrices.

In Chapter one we have known that taking inverse and transposing a matrix
commute. (theorem 1.5.2). Here we write an example that asserts this fact for a
non-square matrix.

−1⁄ 3⁄
1 3 0 2 2
Example 5.2.1 Let A = [ ], a right inverse of A is A−1
𝑅 = [ 1⁄ −1⁄ ],
2 2 1 2 2
0 −1

−1⁄ 1⁄ 0 1 2
(A−1 𝑇 2 2 AT = [3 2] ,
𝑅 ) = [
3⁄ −1⁄ −1
], also
2 2 0 1

−1⁄ 1⁄ 0
(AT )−1 2 2 ] = (A−1 𝑇
𝐿 =[
3⁄ −1⁄ 𝑅 )
2 2 −1

The following Corollaries says that this is true in general.

Corollary 5.2.2 If 𝐴 = (𝑎𝑖𝑗 ) is non-singular 𝑚 × 𝑛 matrix, then

(AT )−1 −1 𝑇
𝐿 = (A𝑅 ) and (AT )−1 −1 𝑇
𝑅 = (A𝐿 )

Proof: If 𝑚 < 𝑛, then 𝐴𝑇 is an 𝑛 × 𝑚

A−1 −1
𝑅 is a right inverse of A, we have 𝐴 𝐴𝑅 = I𝑚

78
Taking transposes, we obtain

(𝐴 𝐴−1 𝑇
𝑅 ) = (I𝑚 )
𝑇

(A−1 𝑇 𝑇
𝑅 ) 𝐴 = I𝑚

These equations imply that

(AT )−1 −1 𝑇
𝐿 = (A𝑅 )

The proof for the case 𝑛 < 𝑚 is similar.

Corollary 5.2.3 Let 𝐴 be a non-singular 𝑚 × 𝑚 matrix, and 𝐵 be non-singular

𝑚 × 𝑛 matrix, then

(A B)−1 −1 −1
𝑅 = B𝑅 𝐴 and (A B)−1 −1 −1
𝐿 = B𝐿 𝐴

Proof: If 𝑚 < 𝑛, since 𝐴 is an 𝑚 × 𝑚 matrix and 𝐵 is an 𝑚×𝑛 matrix,

we have 𝐴−1 is an 𝑚 × 𝑚 matrix and B𝑅−1 is an 𝑛 × 𝑚 matrix

Now,

(A B)(B𝑅−1 𝐴−1 ) = A(B B𝑅−1 )A−1 = A I𝑚 A−1 = A A−1 = I𝑚

When 𝑛 < 𝑚 we obtain

(B𝐿−1 𝐴−1 )(A B) = B𝐿−1 ( A −1 𝐴)B = B𝐿−1 I𝑚 B = B𝐿−1 B = I𝑛

Therefore, (A B)−1 −1 −1
𝑅 = B𝑅 𝐴 and (A B)−1 −1 −1
𝐿 = B𝐿 𝐴

1 3 1 3 0
Example 5.2.4 Prove theorem 5.2.3 for the matrices 𝐴 = [ ] , 𝐵= [ ]
2 1 2 2 1

7 −6
7 9 3
Then, 𝐵 = [ ], 𝑎𝑑𝑗( 𝐴 𝐵 ) = [−3 4]
4 8 1
−4 2

7⁄ −6⁄
10 10 −1⁄ 3⁄
(A B)−1 = −3 ⁄10 4⁄ and A−1
= [ 5 5]
𝑅 10 2⁄ −1⁄
−4 2⁄ 5 5
[ ⁄10 10 ]

−1⁄ 3⁄
1 −3 2 2
𝑎𝑑𝑗(𝐵) = [−1 1 ], B𝑅−1 = [ 1⁄ −1⁄ ]
0 2 2 2
0 −1

79
 7⁄ −6⁄
10 10
B𝑅−1 𝐴−1 −3 4⁄ −1
= ⁄10 10 = (A B)𝑅
−4 2⁄
[ ⁄10 10 ]

Corollary 5.2.5 If k is nonzero real number and 𝐴 = (𝑎𝑖𝑗 ) is a non-singular

𝑚 × 𝑛 matrix, then

1 −1
(k A)−1 = A
𝑘

where the inverse is a right or a left inverse according to 𝑚 ≤ 𝑛 or 𝑚 > 𝑛

Proof: If 𝑚 ≤ 𝑛, since 𝐴 is an 𝑚 × 𝑛 matrix, 𝐴−1

𝑅 is an 𝑛 × 𝑚 matrix
and we have

1
(k A) ( 𝐴−1 ) = A 𝐴−1
𝑅 = 𝐼𝑚 , which gives
𝑘 𝑅

1 −1
(k A)−1
𝑅 = 𝐴
𝑘 𝑅

1
Also, if 𝑛 < 𝑚, (𝑘 𝐴−1 −1
𝐿 ) (k A) = 𝐴𝐿 A = 𝐼𝑛 which gives

1
(k A)−1 −1
𝐿 = 𝑘 𝐴𝐿

We notice here that the properties of the inverse that are satisfied in square
matrices (see theorem 1.5.2.b-c-d) are also satisfied in non-square matrices (see
theorem 5.2.2, 5.2.3, 5.2.5)

The following theorem is given in [8], but we are providing our own proof.

Theorem 5.2.6 If 𝐴 = (𝑎𝑖𝑗 ) is an 𝑚 × 𝑛 ( 𝑚 < 𝑛) matrix, then

A adj(A) = det(𝐴) 𝐼𝑚

𝑎11 𝑎12 ⋯ 𝑎1𝑛

𝑎21 𝑎22 ⋯ 𝑎2𝑛 𝑀11 𝑀21 ⋯ 𝑀𝑗1 ⋯ 𝑀𝑚1
⋮ ⋮ ⋱ ⋮ 𝑀12 𝑀22 … 𝑀𝑗2 ⋯ 𝑀𝑚2
Proof: Let 𝐴 = 𝑎𝑖1 𝑎𝑖2 ⋯ 𝑎𝑖𝑛 , 𝑑𝑗(𝐴) = ⋮ ⋱
⋮ ⋱ ⋮ ⋮
⋮ ⋮ ⋱ ⋮ ⋯
[𝑀1𝑛 𝑀2𝑛 … 𝑀𝑗𝑛 𝑀𝑚𝑛 ]
[𝑎𝑚1 𝑎𝑚2 ⋯ 𝑎𝑚𝑛 ]

80
 𝑎11 𝑎12 ⋯ 𝑎1𝑛
𝑎21 𝑎22 ⋯ 𝑎2𝑛 𝑀11 𝑀21 ⋯ 𝑀𝑗1 ⋯ 𝑀𝑚1
⋮ ⋮ ⋱ ⋮ 𝑀12 𝑀22 … 𝑀𝑗2 ⋯ 𝑀𝑚2
𝐴 𝑎𝑑𝑗(𝐴) = 𝑎𝑖1 𝑎𝑖2 ⋯ 𝑎𝑖𝑛 ⋱
⋮ ⋮ ⋱ ⋮ ⋮
⋮ ⋮ ⋱ ⋮ [𝑀1𝑛 𝑀2𝑛 … 𝑀𝑗𝑛 ⋯ 𝑀𝑚𝑛 ]
[𝑎𝑚1 𝑎𝑚2 ⋯ 𝑎𝑚𝑛 ]

The (𝑖 , 𝑗 )𝑡ℎ element in the product matrix 𝐴 𝑎𝑑𝑗(𝐴) is

𝑖𝑓 𝑖 = 𝑗, 𝑎𝑖1 𝑀𝑗1 + 𝑎𝑖2 𝑀𝑗2 + ⋯ + 𝑎𝑖𝑛 𝑀𝑗𝑛 = det(𝐴)

and, 𝑖𝑓 𝑖 ≠ 𝑗, 𝑎𝑖1 𝑀𝑗1 + 𝑎𝑖2 𝑀𝑗2 + ⋯ + 𝑎𝑖𝑛 𝑀𝑗𝑛 = 0 )It was pre-established in
theorem 5.1.4 **)

det(𝐴) 0 ⋯ 0
0
Hence, 𝐴 𝑎𝑑𝑗(𝐴) = [ 0 det(𝐴) ⋯
⋮ ] = det(𝐴) 𝐼𝑚
⋮ ⋮ ⋱
0 0 ⋯ det(𝐴)
We note that this property is valid for square matrices except that the switch

here is not permissible. (see theorem 1.4.4.a).

i.e, 𝐴 𝑎𝑑𝑗(𝐴) = det(𝐴) 𝐼𝑚 ≠ (𝑎𝑑𝑗(𝐴))𝐴

1 3 0
Example 5.2.7 Prove theorem 5.2.6 for the matrices 𝐴 = [ ]
2 1 −1

2 3
Then, 𝑎𝑑𝑗(𝐴) = [3 1 ] , and 𝑑𝑒𝑡(𝐴) = −7
1 −2

so, 𝐴 𝑎𝑑𝑗(𝐴) = −7 𝐼2 but (𝑎𝑑𝑗(𝐴))𝐴 ≠ −7 𝐼2

All of the following corollaries in this section are proved by using theorem 5.2.6
as follows:

Corollary 5.2.8 If 𝐴 = (𝑎𝑖𝑗 ) is singular 𝑚 × 𝑛 matrix, then 𝑎𝑑𝑗(𝐴) is singular

Proof: by theorem 5.2.6, A adj(A) = det(𝐴) 𝐼𝑚

Since, 𝐴 is singular, then det(𝐴) = 0

So, A adj(A) = 0,

If 𝐴 = 0 (null and singular), then 𝑎𝑑𝑗(𝐴) = 0 and hence 𝑎𝑑𝑗(𝐴) is singular too.

81
If 𝐴 ≠ 0 (non-null and singular), then 𝐴 contains a non-null row, say the 𝑖 𝑡ℎ row
𝑎 ′𝑖 it follows that 𝑎 ′𝑖 𝑎𝑑𝑗(𝐴) = 0

Which implies that the rows of 𝑎𝑑𝑗(𝐴) are linearly dependent (see definition
1.9.3), and hence 𝑎𝑑𝑗(𝐴) is singular.

This property is satisfied in the case of square matrices.

1 1 1
Example 5.2.9 Prove theorem 5.2.8 for the matrices 𝐴 = [ ]
2 3 4

−1 0
then 𝑎𝑑𝑗(𝐴) = [ 2 0] , and 𝑑𝑒𝑡(𝐴) = 0, det(𝑎𝑑𝑗(𝐴)) = 0
−1 0

Corollary 5.2.10 If 𝐴 = (𝑎𝑖𝑗 ) is an 𝑚 × 𝑛 (𝑚 < 𝑛) non-singular matrix, then

(adj(A))T = adj(AT )

Proof: By theorem 5.1.4, if 𝑚 < 𝑛, , then adj(A) = det(𝐴) 𝐴−1

𝑅

Taking transposes, we obtain

𝑇
( adj(A)) = (det(𝐴) 𝐴−1
𝑅 )
𝑇

(adj(A))T = det(𝐴) (𝐴−1

𝑅 )
𝑇
(1)

And, adj(A𝑇 ) = det(𝐴) (𝐴𝑇 )−1

𝐿

adj(A𝑇 ) = det(𝐴) (𝐴−1

𝑅 )
𝑇
(2)

From (1) and (2), we obtain

(adj(A))T = adj(AT )

This property is valid for singular matrices of all size ( square or non-square)

1 3 0
Example 5.2.11 Let 𝐴 = [ ]
2 1 −1

2 3
(𝑎𝑑𝑗(𝐴) )𝑇 = [2 3 1
then 𝑎𝑑𝑗(𝐴) = [3 1 ] , ]
3 1 −2
1 −2

82
 1 2
2 3 1
A𝑇 = [3 1 ], 𝑎𝑑𝑗(A𝑇 ) = [ ] = (𝑎𝑑𝑗(𝐴) )𝑇
3 1 −2
0 −1

Corollary 5.2.12 If k is a scalar and 𝐴 = (𝑎𝑖𝑗 ) is non-singular 𝑚 × 𝑛 matrix,

then

adj(k A) = k 𝑚−1 adj(A)

Proof: By theorem 5.1.4, if 𝑚 ≤ 𝑛

(from corollaries 2.4.4, 5.2.5)

1 −1
adj(k A) = 𝑑𝑒𝑡(k A)(𝑘 𝐴)−1 𝑚
𝑅 = 𝑘 det(𝐴) 𝐴
𝑘 𝑅

= 𝑘 𝑚−1 det(𝐴) 𝐴−1

𝑅 =𝑘
𝑚−1
adj(A)

Similarly, if 𝑛<𝑚 .

This property is valid for singular matrices of all size ( square or non-square)

Corollary 5.2.13 Let 𝐴 be an 𝑚 × 𝑚 non-singular matrix, and 𝐵 be an 𝑚 × 𝑛

non-singular matrix, then

𝑎𝑑𝑗(𝐴 𝐵) = adj(B) adj(A)

Proof: By theorem 5.1.4 , if 𝑚 ≤ 𝑛

𝑎𝑑𝑗(𝐴 𝐵) = det( A B)( A B)−1

R

= det(𝐴) . det(𝐵) . 𝐵𝑅−1 𝐴−1 (from theorem 3.1.2 , corollary 5.2.3)

= det(𝐵) . 𝐵𝑅−1 . det(𝐴) . 𝐴−1

= adj(B). adj(A)

Similarly, if 𝑛<𝑚

1 3 1 3 0
Example 5.2.14 Let 𝐴 = [ ] , 𝐵= [ ]
2 1 2 2 1

83
 7 −6
7 9 3
then 𝐴 𝐵 = [ ], 𝑎𝑑𝑗( 𝐴 𝐵 ) = [−3 4 ]
4 8 1
−4 2

1 −3
1 −3
𝑎𝑑𝑗(𝐵) = [−1 1 ], 𝑎𝑑𝑗(𝐴) = [ ]
−2 1
0 2

1 −3 7 −6
1 −3
𝑎𝑑𝑗(𝐵) 𝑎𝑑𝑗(𝐴) = [−1 1 ][ ] = [−3 4 ] = 𝑎𝑑𝑗( 𝐴 𝐵)
−2 1
0 2 −4 2

1 2 2 5 8
Example 5.2.15 Let 𝐴= [ ], 𝐵= [ ]
2 4 3 2 1

8 9 10
Then, det(𝐴) = 0, 𝑑𝑒𝑡(𝐵) = 0, 𝐴𝐵 =[ ], det(𝐴𝐵) = −4,
16 18 20

−2 1 1 3
) 4 −2
𝑎𝑑𝑗( 𝐴 𝐵 = [ 4 −2], 𝑎𝑑𝑗(𝐵) = [−2 −6], 𝑎𝑑𝑗(𝐴) = [ ]
−2 1
−2 −1 1 3

det(𝑎𝑑𝑗(𝐵)) = 0, 𝑑𝑒𝑡(𝑎𝑑𝑗(𝐴)) = 0, 𝑑𝑒𝑡(𝑎𝑑𝑗(𝐴𝐵)) = −12,

−2 1
𝑎𝑑𝑗(𝐵) 𝑎𝑑𝑗(𝐴) = [ 4 −2] = 𝑎𝑑𝑗( 𝐴 𝐵)
−2 −1

We notice here that the properties of the adjoint of a matrix that are satisfied
in square matrices (see theorem 1.4.4.d-e-f) are also satisfied in non-square
matrices (see theorem 5.2.10, 5.2.12, 5.2.13).

5.3 Pseudo inverse of non-square matrices.

In section 5.2 we compute an inverse of a rectangular matrix using solution

of a linear system and an adjoint of 𝐴, here we discuss another method which
gives an inverse of 𝐴 .

Theorem 5.3.1 [11, 𝑝. 216 ] Let 𝐴 be an 𝑚 × 𝑛 matrix, the null space of 𝐴 is

denoted by Ɲ(𝐴). The dimension of the null space of 𝐴 is called the nullity of 𝐴.

84
 (1) Ɲ(𝐴) = Ɲ(𝐴𝑇 𝐴)
(2) 𝑟𝑎𝑛𝑘(𝐴) = 𝑟𝑎𝑛𝑘(𝐴𝑇 𝐴)

Proof: (1) show Ɲ(𝐴) ⊂ Ɲ(𝐴𝑇 𝐴)

Consider any 𝑥 ∈ Ɲ(𝐴), Then we have 𝑨𝒙 = 𝟎 ( see theorem 1.9.6)

Multiplying it by 𝐴𝑇 from the left, we obtain

𝐴𝑇 𝑨𝒙 = 𝐴𝑇 0 = 0

Thus 𝑥 ∈ Ɲ(𝐴𝑇 𝐴)

and hence Ɲ(𝐴) ⊂ Ɲ(𝐴𝑇 𝐴) (𝑖)

show Ɲ(𝐴𝑇 𝐴) ⊂ Ɲ(𝐴)

let 𝑥 ∈ Ɲ(𝐴𝑇 𝐴), thus we have 𝐴𝑇 𝑨𝒙 = 0

Multiplying it by 𝑥 𝑇 from the left, we obtain

𝑥 𝑇 𝐴𝑇 𝑨𝒙 = 𝑥 𝑇 0 = 0

This implies that we have 0 = (𝑨𝒙)𝑇 𝑨𝒙 = ‖𝑨𝒙‖2

and the length of the vector 𝑨𝒙 is zero, thus the vector 𝑨𝒙 = 𝟎. Hence 𝑥 ∈ Ɲ(𝐴)

hence Ɲ(𝐴𝑇 𝐴) ⊂ Ɲ(𝐴) (𝑖𝑖)

from (𝑖) and (𝑖𝑖)

Hence, Ɲ(𝐴) = Ɲ(𝐴𝑇 𝐴)

(2) We use the rank-nullity theorem and obtain (see theorem 1.9.8)

𝑟𝑎𝑛𝑘(𝐴) = 𝑛 − 𝑑𝑖𝑚(Ɲ(𝐴))

= 𝑛 − 𝑑𝑖𝑚(Ɲ(𝐴𝑇 𝐴))

= 𝑟𝑎𝑛𝑘(𝐴𝑇 𝐴)

(Note that the size of the matrix 𝐴𝑇 𝐴 is 𝑛 × 𝑛)

85
Definition 5.3.2 [4] The matrix 𝐴+ 𝑇 𝑇 −1
𝑅 = 𝐴 (𝐴 𝐴 ) , when 𝐴 is 𝑚 × 𝑛 (𝑚 ≤ 𝑛)

and (𝐴 𝐴𝑇 )−1 exists, and 𝑟𝑎𝑛𝑘 (𝐴) = 𝑚, is called the right Pseudo inverse of 𝐴.

Definition 5.3.3 [4 ] The matrix 𝐴+ 𝑇 −1 𝑇

𝐿 = ( 𝐴 𝐴) 𝐴 , when 𝐴 is 𝑚 × 𝑛 (𝑚 > 𝑛)

and ( 𝐴𝑇 𝐴)−1 exists, and 𝑟𝑎𝑛𝑘 (𝐴) = 𝑛, is called the left Pseudo inverse of 𝐴

Remark 5.3.4 The matrix 𝐴𝑚×𝑛 ( 𝑚 > 𝑛) has 𝑟𝑎𝑛𝑘 = 𝑛, and therefore 𝐴𝑇 also

has 𝑟𝑎𝑛𝑘 = 𝑚, 𝐴𝑇 𝐴 is also of 𝑟𝑎𝑛𝑘 = 𝑛 ,(see theorem 5.3.1) since 𝐴𝑇 𝐴 is a

𝑛 × 𝑛 matrix. it therefore has full rank and its inverse exists. (see theorem 1.9.14)

Note that 𝐴 𝐴𝑇 is a 𝑚 × 𝑚 matrix but its inverse does not exist.

1 1 1
Example 5.3.5 Let A = [ ], 𝑟𝑎𝑛𝑘 (𝐴) = 2 = 𝑚 = 𝑟𝑎𝑛𝑘 (𝐴𝐴𝑇 )
3 −1 1

11 −3
𝐴𝐴𝑇 is invertible, |𝐴 𝐴𝑇 | = 33 − 9 = 24, and (𝐴 𝐴𝑇 )−1 = 24
1
[ ]
−3 3

1 1 3
11 −3 1 2 6
𝐴−1 + 𝑇 𝑇 −1
𝑅 = 𝐴𝑅 = 𝐴 (𝐴 𝐴 ) = [1 −1] [ ]= [14 −6]
24 −3 3 24
1 1 8 0

1 1 1 1 2 6
𝐴 𝐴−1
𝑅 = [ ] [14 −6] = 𝐼2
24 3 −1 1
8 0

Since det(𝐴 ) = 0, we cannot use adjoint method

1 −2
Example 5.3.6 Let A = [−2 1 ], 𝑟𝑎𝑛𝑘 (𝐴) = 2 = 𝑛 = 𝑟𝑎𝑛𝑘 (𝐴𝑇 𝐴)
1 1
6 3
𝐴𝑇 𝐴 is invertible, | 𝐴𝑇 𝐴 | = 36 − 9 = 27, and ( 𝐴𝑇 𝐴)−1 = 27
1
[ ]
3 6
−1 1
1 6 0
3 1 −2 1 3 3]
𝐴−1
𝐿 = 𝐴+ 𝑇 −1 𝑇
𝐿 = ( 𝐴 𝐴) 𝐴 = [ ][ ][
27 3 6 −2 1 1 −1 1
0
3 3
−1 1
0 1 −2
3 3
𝐴−1
𝐿 𝐴 = [−1 1] [−2 1 ] = 𝐼2
3
0 3 1 1

86
 Chapter Six

Applications to linear systems of equations 𝑨 𝑿 = 𝑩

In this chapter we discuss some results concerning the solutions of a linear

system 𝑨𝒙 = 𝒃 using inverses as well as the pseudo-inverse and adjoint of a
rectangular 𝑚 × 𝑛 matrix 𝐴.

The solution of the system can be expressed as 𝒙 = 𝐴−1 𝒃 where 𝐴−1 is the

inverse of 𝐴. when matrix 𝐴 is of order 𝑚 × 𝑛 (𝑚 < 𝑛) because if 𝐴−1

𝑅 is the

right inverse of 𝐴 then we have 𝑨𝐴−1 −1

𝑅 𝒙 = 𝒃𝐴𝑅 which yields 𝐼𝑚 𝒙 = 𝒃𝐴−1
𝑅

implies that 𝒙 = 𝒃𝐴−1

𝑅 but 𝒃𝐴−1
𝑅 is not defined. Since we can't find an actual

solution to the system, we will now try to find solution to the system.

6.1 Solving a linear system Using pseudo inverse

If 𝐴 is an 𝑚 × 𝑛 matrix, then the linear system 𝑨 𝒙 = 𝒃 is a system of 𝑚

equations in 𝑛 unknowns.

Theorem 6.1.1[5 ] Let 𝑨 𝒙 = 𝒃 be a linear system with 𝐴 an 𝑚 × 𝑛 ( 𝑚 > 𝑛) matrix.

If 𝐴𝑇 𝐴 is invertible, then the solution can be given by 𝑥 = 𝑨𝑳−𝟏 𝒃 where 𝐴−1

𝐿 is a left

pseudo-inverse of 𝐴

Proof: Let 𝐴 be an 𝑚 × 𝑛 ( 𝑚 > 𝑛) and 𝐴𝑇 𝐴 is invertible, then (𝐴𝑇 𝐴)−1 exists,

and we can multiply 𝑨 𝒙 = 𝒃 by 𝐴𝑇 on both sides. Obtaining 𝐴𝑇 𝑨 𝒙 = 𝐴𝑇 𝒃

So, multiply AT 𝐀 𝐱 = AT 𝐛 by ( AT A)−1 on the both sides,

(𝐴𝑇 𝐴)−1 (𝐴𝑇 𝑨 𝒙)=(𝐴𝑇 𝐴)−1 (𝐴𝑇 𝒃)

𝐼𝑛 𝒙 =(𝐴𝑇 𝐴)−1 (𝐴𝑇 𝒃)

𝒙 = 𝐴−1 𝑇 −1 𝑇
𝐿 (𝐴 ) 𝐴 𝒃
𝒙 = 𝐴−1
𝐿 𝒃

87
Example 6.1.2 Find a solution to the system
𝑥1 + 3𝑥2 = −2
3 𝑥1 − 𝑥2 = 4
2 𝑥1 + 2 𝑥2 = 0

Solution: Let 𝒙 be a solution to 𝑨𝒙 = 𝒃, (if exists)

1 3 −2
𝑨=[3 −1] is the coefficient matrix, and 𝒃 = [ 4 ] is the constant vector
2 2 0

𝒙 = ( 𝐴𝑇 𝐴)−1 𝐴𝑇 𝐛

1 3
1 3 2 14 4
𝐴𝑇 𝐴 = [ ][ 3 −1] = [ ]
3 −1 2 4 14
2 2

1 7 −2
So, (𝐴𝑇 𝐴)−1 = 90 [ ]
−2 7

1 7 −2 1 3 2 1 2 23 10
( 𝐴𝑇 𝐴)−1 𝐴𝑇 = [ ][ ]= [ ]
90 −2 7 3 −1 2 90 19 −13 10

1 2 −2 1 90
𝑇 −1 𝑇 23 10 1
𝒙 = ( 𝐴 𝐴) 𝐴 𝐛 = [ ][ 4 ] = [ ]= [ ]
90 19 −13 10 90 −90 −1
0

Note that x = ( 1 , −1) represents the intersection point of the three straight lines
shown in the graph.

We note that 𝑟𝑎𝑛𝑘(𝐴) = 2, which means that there

is a unique solution to the original system occurs at
the intersection of these three lines

Note that : if 𝐴𝑇 𝐴 is invertble, then the only possibility for 𝑨𝒙 = 𝒃 are either unique
solution or no solution.

88
Example 6.1.3 Find a solution to the system
𝑥1 + 3 𝑥2 = 5
𝑥1 − 𝑥2 = 1
𝑥1 + 𝑥2 = 0

Solution: Let x be a solution to 𝑨𝒙 = 𝒃, (if exists)

1 3 5
𝐴 = [1 −1] is the coefficient matrix and 𝒃 = [1] is the constant vector
1 1 0

𝒙 = ( 𝐴𝑇 𝐴)−1 𝐴𝑇 𝐛

1 3
1 1 1 3 3
We calculate 𝐴𝑇 𝐴 = [ ][ 1 −1] = [ ]
3 −1 1 3 11
1 1

1 11 −3
Next, (𝐴𝑇 𝐴)−1 = 24 [ ]
−3 3

1 11 −3 1 1 1 1 2 14 8
( 𝐴𝑇 𝐴)−1 𝐴𝑇 = [ ][ ]= [ ]
24 −3 3 3 −1 1 24 6 −6 0

5
𝑇 −1 𝑇 1 2 14 8 1 24 1
𝒙 = ( 𝐴 𝐴) 𝐴 𝐛 = [ ] [1 ] = [ ]= [ ]
24 6 −6 0 24 24 1
0

Note that x = (1 , 1) doesn’t satisfy the three

equations, that is, it doesn’t represent the intersection

point of the three straight lines. This coincides with the

second case in the note.

Theorem 6.1.4 [5] Let 𝑨 𝒙 = 𝒃 be a linear system with 𝐴 an 𝑚 × 𝑛 ( 𝑚 < 𝑛) matrix.

If 𝐴 𝐴𝑇 is invertible, then the solution can be given by 𝒙 = 𝐴−1 −1

𝑅 𝒃 where 𝐴𝑅 is

aright pseudo-inverse of 𝐴

Proof: Let 𝐴 be an 𝑚 × 𝑛 ( 𝑚 < 𝑛) and 𝐴 𝐴𝑇 is invertible, then ( 𝐴 𝐴𝑇 )−1 exists,

89
and we can multiply 𝑨 𝒙 = 𝒃 by (𝐴 𝐴𝑇 )−1 on both sides, obtaining

(𝐴 𝐴𝑇 )−1 𝑨 𝒙 = (𝐴 𝐴𝑇 )−1 𝒃
(𝐴𝑇 )−1 −1 𝑇 −1
𝑅 𝐴𝐿 𝑨 𝒙 = (𝐴 𝐴 ) 𝒃
𝑇 −1
(𝐴 )𝑅 𝐼𝑛 𝒙 = 𝑇 −1
(𝐴 𝐴 ) 𝒃 (1)

So, multiply (1) by 𝐴𝑇 on both sides, we obtain

𝒙 = 𝐴𝑇 (𝐴 𝐴𝑇 )−1 𝐛

Example 6.1.5 Find a solution to the system

2𝑥1 + 3 𝑥2 − 2𝑥3 = 4
−6 𝑥1 − 8 𝑥2 + 6𝑥3 = 1

Solution: Let x be a solution to 𝑨𝒙 = 𝒃, (if exists)

𝒙 = ( 𝐴𝑇 𝐴)−1 𝐴𝑇 𝐛

2 −6
2 3 −2 17 −48
We calculate 𝐴 𝐴𝑇 = [ ] [ 3 −8] = [ ]
−6 −8 6 −48 136
−2 6

1 136 48
Next, (𝐴𝑇 𝐴)−1 = 8 [ ]
48 17

1 2 −6 1 −16 −6
136 48
𝐴𝑇 (𝐴 𝐴𝑇 )−1 = [ 3 −8] [ ] = [ 24 8]
8 48 17 8
−2 6 16 6

−70
−16 −6 −70 8
1 4 1
𝒙 = 𝐴𝑇 (𝐴 𝐴𝑇 )−1 𝐛 = 8 [ 24 8 ][ ] = [ 104 ] = [ 13 ] is a solution
1 8 70
16 6 70
8

−35
𝑥1 4 1
The general solution in vector form as [ 𝑥2 ] = [ 13 ] + 𝑡 [0]
𝑥3 35 1
4

This is a parameter vector equation of the line of intersection 𝐿 of the three lines. The
coordinates of each of 𝐿′𝑠 points make one of the infinitely many solutions of the system.

90
Theorem 6.1.6 [6 ]. (General solution)

Let 𝑨 𝒙 = 𝒃 be a full rank underdetermined system (𝐴 an 𝑚 × 𝑛 ( 𝑚 < 𝑛)

matrix.). then the solution set is given by

𝒙 = 𝐴𝑇 (𝐴 𝐴𝑇 )−1 𝒃 + (𝐼 − 𝐴𝑇 (𝐴 𝐴𝑇 )−1 𝐴)𝑦

𝒙 = 𝐴−1 −1
𝑅 𝒃 + (𝐼𝑛 − 𝐴𝑅 𝐴)𝑦, (1)

where 𝑦 is an arbitrary vector in 𝑅 𝑛×1 .

Proof: To verify that (1) is a solution, pre-multiply by 𝐴

𝑨𝒙 = 𝐴 𝐴−1 −1
𝑅 𝒃 + 𝐴 (𝐼 − 𝐴𝑅 𝐴)𝑦

= 𝐼𝑚 𝒃 + (𝐴 − 𝐴 𝐴−1
𝑅 𝐴)𝑦 by hypothesis

= 𝒃 , since 𝐴 𝐴−1
𝑅 𝐴 = 𝐼𝑚 𝐴𝑚×𝑛 = 𝐴

That all solutions are of this form can be seen as follows.

Let 𝒛 be an arbitrary solution of 𝑨 𝒙 = 𝒃 , i.e., 𝑨 𝒛 = 𝒃. Then we can write

𝑧 = 𝐴−1 −1 −1 −1
𝑅 𝑨 𝒛 + (𝐼 − 𝐴𝑅 𝐴)𝑧 = 𝐴𝑅 𝑨 𝒛 + 𝑧 − 𝐴𝑅 𝑨 𝒛

So that any solution 𝒙 of 𝑨 𝒙 = 𝒃 is given by (1) with 𝒚 = 𝒛

Remark: When 𝐴 is square and nonsingular, 𝐴−1

𝑅 =𝐴
−1
and so (𝐼 − 𝐴−1
𝑅 𝐴) = 0

Thus, there is no “arbitrary” component, leaving only the unique solution

𝑥 = 𝐴−1 𝑏

Note that: when 𝐴 is not full rank, then the above theorem cannot be used.

6.2 Using adjoint to solve a system of linear equations

Let 𝐴 be a non-singular 𝑚 × 𝑛 matrix. In this section we use adjoint of 𝐴 is

find a solution for 𝑨𝒙 = 𝒃.

91
Let 𝐴𝑚×𝑛 , 𝑚 ≤ 𝑛, be nonsingular, then 𝐴−1
𝑅 exists and we can multiply 𝑨 𝒙 = 𝒃

by 𝐴−1 −1 −1
𝑅 on both sides, obtaining 𝑨 𝐴𝑅 𝒙 = 𝐴𝑅 𝒃

1
Then 𝐼𝑚 𝒙 = 𝐴−1
𝑅 𝒃 , but 𝐴−1
𝑅 = det(𝐴) 𝑎𝑑𝑗(𝐴) ( see theorem 5.1.4)

1
So, 𝒙 = 𝑎𝑑𝑗(𝐴) 𝒃
det(𝐴)

assume 𝑚 > 𝑛, and 𝐴 is nonsingular, then 𝐴−1

𝐿 exists and we can multiply both

sides of 𝑨 𝒙 = 𝒃 by 𝐴−1
𝐿 , obtaining 𝐴−1 −1
𝐿 𝑨 𝒙 = 𝐴𝐿 𝒃

1
That is 𝐼𝑛 𝒙 = 𝐴−1
𝐿 𝒃, but 𝐴−1
𝐿 = det(𝐴) 𝑎𝑑𝑗(𝐴) (see theorem 5.1.6)

1
So, 𝒙 = 𝑎𝑑𝑗(𝐴) 𝒃
det(𝐴)

Example 6.2.1 Find a solution to the system

𝑥1 + 3 𝑥2 = −2
3 𝑥1 − 𝑥2 = 4
2 𝑥1 + 2 𝑥2 = 0

Solution: Let 𝒙 be a solution to 𝑨𝒙 = 𝒃 , (if exists)

1 3 −2
𝐴 = [3 −1] is the coefficient matrix, and 𝒃 = [ 4 ] is the constant vector
2 2 0

1
𝑑𝑒𝑡(𝐴) = 2, since 𝐴 is 3 × 2, 𝐴−1
𝐿 exists, 𝒙 = 𝑎𝑑𝑗(𝐴) 𝒃
det(𝐴)

−3 −1 4
So, we calculate 𝑎𝑑𝑗(𝐴) = [ ]
−1 −1 2

1 1 −3 −1 4
Next, 𝐴−1
𝐿 = det(𝐴) 𝑎𝑑𝑗(𝐴) = 2 [ ]
−1 −1 2

1 −3 −1 4 −2 1 2 1
𝒙 = 𝐴−1
𝐿 𝒃= [ ] [ 4 ]= [ ]=[ ]
2 −1 −1 2 2 −2 −1
0

92
 We note that in this example, adjoint method and pseudo method give the same
solution (see example 6.1.2).

Example 6.2.2 Find a solution to the system

2𝑥1 + 3 𝑥2 − 2𝑥3 = 4
−6𝑥1 − 8 𝑥2 + 6𝑥3 = 1

Solution: Let 𝒙 be a solution to 𝑨𝒙 = 𝒃 , (if exists)

2 3 −2 4
Hold by 𝐴 = [ ] is the coefficient matrix, and 𝒃 = [ ] is the constant
−6 −8 6 1
1
vector. 𝑑𝑒𝑡(𝐴) = 4, since 𝐴 is 2 × 3, 𝐴−1
𝑅 exists, 𝒙 = 𝑎𝑑𝑗(𝐴) 𝒃
det(𝐴)

−14 −5
But, 𝑎𝑑𝑗(𝐴) = [ 12 4]
2 1

−14 −5
1 1
Next, 𝐴−1
𝑅 = 𝑎𝑑𝑗(𝐴) = [ 12 4]
det(𝐴) 4
2 1

1 1 −14 −5 4 1 −61
𝒙= 𝑎𝑑𝑗(𝐴) 𝒃 = [ 12 4 ] [ ] = [ 52 ]
det(𝐴) 4 1 4
2 1 9

−71 52 9
That is, x = ( , , )
4 4 4

Here the solution we have obtained using adjoint method is different from the

pseudo method solution (see example 6.1.5) and both of them are members from

the general solution given by

𝑥 = 𝐴𝑇 (𝐴 𝐴𝑇 )−1 𝑏 + (𝐼 − 𝐴𝑇 (𝐴 𝐴𝑇 )−1 𝐴)𝑦 = 𝐴−1 −1

𝑅 𝑏 + (𝐼 − 𝐴𝑅 𝐴)𝑦

−35
4 1 0 0 −16 −6 𝑡1
1 2 3 −2
𝑥 = [ 13 ] + ([0 1 0] − 8 [ 24 8 ][ ]) [𝑡2 ]
35 −6 −8 6 𝑡3
0 0 1 16 6
4

−35
4 1
1 −71 52 9
𝑥 = [ 13 ] + 2 𝑡 [ 0], Choose 𝑡 = −13, x = ( , , ).
35 4 4 4
1
4

93
6.3 Cramer's rule for nonsingular 𝒎 × 𝒏 matrices

In linear algebra, Cramer's rule (see theorem 1.7.4) gives an explicit formula for the

solution of a system of linear equations with as many equations as unknowns. That is, for

the solution of a system with a square matrix and provided that the coefficient matrix is

invertible, Cramer’s rule offers a simple and a convenient formula for the solution. In this

section we want to generalize this method for an 𝑚 < 𝑛 system of linear equations. As in

the usual method of Cramer's, the result for rectangular matrices uses the minors of a

matrix. We also use the results in order to solve a matrix equation. in the case of systems

with an infinite number of solutions, get the final formula for calculating the unknowns by

the minors of the augmented matrix of the system.

The key to Cramer’s Rule is replacing the variable column of interest with the
constant column and calculating the determinants.

Theorem 6.3.1 Consider the following linear system of 𝑚 equations in 𝑛 unknowns

𝑎11 𝑥1 + 𝑎12 𝑥2 + … + 𝑎1𝑛 𝑥𝑛 = 𝑏1

𝑎21 𝑥1 + 𝑎22 𝑥2 + … + 𝑎2𝑛 𝑥𝑛 = 𝑏2
⋮ ⋮ ⋱ ⋮ ⋮
𝑎𝑚1 𝑥1 + 𝑎𝑚2 𝑥2 + … + 𝑎𝑚𝑛 𝑥𝑛 = 𝑏
𝑚

𝑏1
𝑏
𝐴 = [𝑎𝑖𝑗 ] be the coefficient matrix, 𝑏 = [ 2 ] . If det(𝐴) ≠ 0. Then
⋮
𝑏𝑚

det(𝐴1 ) det(𝐴2 ) det(𝐴𝑛 )

𝑥1 = , 𝑥2 = , …. , 𝑥𝑛 = .
det(𝐴) det(𝐴) det(𝐴)

Is a solution for the system 𝑨𝒙 = 𝒃, where 𝐴𝑗 is the matrix obtained from 𝐴 by

replacing the 𝑗 𝑡ℎ column of 𝐴 by 𝒃.

94
proof: We look at the linear system 𝑨𝒙 = 𝒃.

𝒙 = 𝐴−1 𝒃

𝑥1
𝑥2 1
[ ⋮ ] = det(𝐴) (𝑎𝑑𝑗 𝐴) 𝒃
𝑥𝑛

𝑥1
𝐴11 𝐴21 ⋯ 𝐴𝑚1
𝑥2 1
[ ⋮ ] = det(𝐴) [ ⋮ ⋮ ⋱ ⋮ ] 𝒃
𝑥𝑛 𝐴1𝑛 𝐴2𝑛 ⋯ 𝐴𝑚𝑛

𝐴11 𝐴21 𝐴𝑚1

⋯
𝑥1 det(𝐴) det(𝐴) det(𝐴)
⋮ ⋮ ⋮ ⋱ ⋮ 𝑏1
𝐴1𝑗 𝐴2𝑗 𝐴𝑚𝑗 𝑏
𝑥𝑗 = [ ⋮2 ]
det(𝐴) det(𝐴) ⋯ det(𝐴)
⋮
⋮ ⋮ ⋱ ⋮ 𝑏𝑚
[𝑥𝑛 ] 𝐴1𝑛 𝐴2𝑛 𝐴𝑚𝑛
…
[ det(𝐴) det(𝐴) det(𝐴)]

This means that

𝐴1𝑗 𝐴2𝑗 𝐴𝑚𝑗
𝑥𝑗 = 𝑏1 + 𝑏2 + ⋯ + 𝑏𝑚 (1 ≤ 𝑗 ≤ n)
det(𝐴) det(𝐴) det(𝐴)

Where

𝑎11 𝑎12 … 𝑎1𝑗−1 𝑏1 𝑎1𝑗+1 … 𝑎1𝑛

𝑎21 𝑎22 … 𝑎2𝑗−1 𝑏2 𝑎2𝑗+1 … 𝑎2𝑛
𝐴𝑗 = ⋮ ⋮ ⋮
⋱ ⋮ ⋮ ⋮ ⋱
𝑎
[ 𝑚1 𝑎𝑚2 … 𝑎𝑚𝑗−1 𝑏𝑚 𝑎𝑛𝑗+1 … 𝑎𝑚𝑛 ]

If we evaluate det(𝐴𝑗 ) be expanding about the 𝑗 𝑡ℎ column, we find that

det(𝐴𝑗 ) = 𝐴1𝑗 𝑏1 + 𝐴2𝑗 𝑏2 + ⋯ + 𝐴𝑚𝑗 𝑏𝑚

Hence

det(𝐴𝑗 )
𝑥𝑗 = .
det(𝐴)

95
For 𝑗 = 1, 2, … , 𝑛. In this expression for 𝑥𝑗 , the determinant of 𝐴𝑗 , det(𝐴𝑗 ), can be

calculated by any method. It was only in the derivation of the expression for 𝑥𝑗 that we

had to evaluate it by expanding about the 𝑗 𝑡ℎ column.

Example 6.3.2. Find a solution to the system

2𝑥1 + 3 𝑥2 − 2𝑥3 = 4
−6𝑥1 − 8 𝑥2 + 6𝑥3 = 1

Solution: Let 𝒙 be a solution to 𝑨𝒙 = 𝒃 , (if exists)

2 3 −2 4
Here the coefficient matrix is 𝐴=[ ], 𝑑𝑒𝑡(𝐴) = 4, and 𝒃 = [ ] is the
−6 −8 6 1

constant vector.

4 3 −2 det(𝐴1 )
𝐴1 = [ ] , det(𝐴1 ) = −59  𝑥1 = = −59⁄4
1 −8 6 det(𝐴)

2 4 −2 det(𝐴2 )
𝐴2 = [ ] , det(𝐴2 ) = 52  𝑥2 = = 52⁄4
−6 1 6 det(𝐴)

2 3 4 det(𝐴3 )
𝐴3 = [ ] , det(𝐴3 ) = 11  𝑥3 = = 11⁄4
−6 −8 1 det(𝐴)

A solution given by Cramer's rule is 𝑥 = (−59⁄4 , 13 , 11⁄4)

We note that the general solution for this example given by Theorem 6.1.6 is

𝑥 = 𝐴𝑇 (𝐴 𝐴𝑇 )−1 𝑏 + (𝐼 − 𝐴𝑇 (𝐴 𝐴𝑇 )−1 𝐴)𝑦 = 𝐴−1 −1

𝑅 𝑏 + (𝐼 − 𝐴𝑅 𝐴)𝑦

−35
4 1 1
𝑥 = 13 + 𝑡 [ 0]
35 2
1
[ 4 ]

Choose 𝑡 = −12, 𝑥 = (−59⁄4 , 13 , 11⁄4)

96
Theorem 6.3.3 [10 ]. (Generalization of Cramer's rule )

For all 𝑏 ∈ 𝑅 𝑛 and 𝐴 an 𝑚 × 𝑛 the system 𝑨𝒙 = 𝒃 is solvable if and only if

𝑑𝑒𝑡(𝐴𝐴𝑇 ) ≠ 0.

Moreover, one solution for this equation is given by 𝑥 = 𝐴𝑇 (𝐴 𝐴𝑇 )−1 𝑏 , where 𝐴𝑇

is the transpose of 𝐴.

Also, this solution coincides with the Cramer's rule formula when 𝑛 = 𝑚. In

fact, this formula is given as follows:

𝑚
𝑑𝑒𝑡((𝐴 𝐴𝑇 ))𝑖
𝑥𝑗 = ∑ 𝑎𝑖𝑗 , 𝑗 = 1, 2, 3 , … , 𝑛,
𝑑𝑒𝑡(𝐴 𝐴𝑇 )
𝑖=1

where (𝐴 𝐴𝑇 )𝑖 is the matrix obtained by replacing the entries in the 𝑗 𝑡ℎ column of

𝑏1
𝑏
𝐴 𝐴𝑇 by the entries in the matrix [ 2 ]
⋮
𝑏𝑚

Proof: The matrix 𝐴 gives a linear transformation 𝑇𝐴 ∶ 𝑅 𝑚 → 𝑅 𝑛 (see definition

1.9.11) and its transpose 𝐴𝑇 gives the linear transformation, adjoint operator, such
that 𝑇𝐴𝑇 : 𝑅 𝑛 → 𝑅 𝑚 .

The system 𝑨𝒙 = 𝒃 is solvable for all 𝑏 ∈ 𝑅 𝑚 , if and only if, the operator 𝐴 is
onto (see theorem 1.9.15).

i.e, 𝑅𝑎𝑛𝑔𝑒(𝐴) = 𝑅 𝑛 Hence, from the lemma 1.8.15 there exists 𝛾 > 0 such that
‖𝐴𝑇 𝑧‖𝑅𝑚 ≥ 𝛾 ‖𝑧‖𝑅𝑛 , 𝑧 ∈ 𝑅 𝑛 .

Therefore,

〈𝐴 𝐴𝑇 𝑧 , 𝑧〉 ≥ 𝛾 2 ‖𝑧‖2 𝑅𝑛 , 𝑧 ∈ 𝑅 𝑛 .

This implies that 𝑇𝐴 𝐴𝑇 : 𝑅 𝑛 → 𝑅 𝑛 , and 𝑅𝑎𝑛𝑔(𝐴 𝐴𝑇 ) = 𝑅 𝑛 and 𝐾𝑒𝑟 (𝐴𝐴𝑇 ) = {0}

From the Theorem 1.9.15, lemma 1.9.16 then 𝐴 𝐴𝑇 is one to one.

Since 𝐴 𝐴𝑇 is a 𝑛 × 𝑛 matrix, from the Theorem 1.9.14 then 𝑑𝑒𝑡(𝐴 𝐴𝑇 ) ≠ 0.

97
Suppose now that 𝑑𝑒𝑡(𝐴 𝐴𝑇 ) ≠ 0. Then (𝐴 𝐴𝑇 )−1 exists and given 𝑏 ∈ 𝑅𝑛 we

can see that 𝑥 = 𝐴𝑇 (𝐴 𝐴𝑇 )−1 𝑏 is a solution of 𝑨 𝒙 = 𝒃.

Now, since 𝑧 = (𝐴 𝐴𝑇 )−1 𝑏 is the only solution of the equation (𝐴 𝐴𝑇 ) 𝑧 = 𝑏,

then from theorem 6.3.1 (Cramer's rule ) we obtain that :

det((𝐴 𝐴𝑇 )1 ) det((𝐴 𝐴𝑇 )2 ) det((𝐴 𝐴𝑇 )𝑛 )

𝑧1 = , 𝑧2 = , …. , 𝑧𝑛 = .
det(𝐴 𝐴𝑇 ) det(𝐴 𝐴𝑇 ) det(𝐴 𝐴𝑇 )

Where (𝐴 𝐴𝑇 )𝑖 is the matrix obtained by replacing the entries in the 𝑗 𝑡ℎ column

𝑏1
𝑏
of 𝐴 𝐴𝑇 by the entries in the matrix [ 2 ].
⋮
𝑏𝑚

Then, the solution 𝑥 = 𝐴𝑇 (𝐴 𝐴𝑇 )−1 𝑏 = 𝐴𝑇 𝑧 of 𝑨 𝒙 = 𝒃 can be written as

follows

𝑑𝑒𝑡((𝐴 𝐴𝑇 )𝑗 )
det((𝐴 𝐴𝑇 )1 )
∑𝑛𝑗=1 𝑎𝑗,1
det(𝐴 𝐴𝑇 ) 𝑑𝑒𝑡(𝐴 𝐴𝑇 )
𝑎11 𝑎21 ⋯ 𝑎𝑛1
det((𝐴 𝐴𝑇 )2 ) 𝑑𝑒𝑡((𝐴 𝐴𝑇 )𝑗 )
𝑎12 𝑎22 ⋯ 𝑎𝑛2 ∑𝑛𝑗=1 𝑎𝑗,2
𝑥𝑗 = [ ⋮ ⋱ ⋮ ]
⋮ det(𝐴 𝐴𝑇 ) = 𝑑𝑒𝑡(𝐴 𝐴𝑇 )
𝑎1𝑚 𝑎21 ⋮ ⋮
… 𝑎𝑛𝑚 det((𝐴 𝐴𝑇 )𝑚 )
𝑑𝑒𝑡((𝐴 𝐴𝑇 )𝑗 )
[ det(𝐴 𝐴𝑇 ) ] 𝑛
[∑𝑗=1 𝑎𝑗,𝑚 𝑑𝑒𝑡(𝐴 𝐴𝑇 ) ]

Example 6.3.4 Find a solution to the system

2𝑥1 + 3 𝑥2 − 2𝑥3 = 4
−6𝑥1 − 8 𝑥2 + 6𝑥3 = 1

Solution: Let 𝑥 be a solution to 𝑨𝒙 = 𝒃 , (if exists)

2 3 −2 4
Here the coefficient matrix is 𝐴=[ ], 𝑑𝑒𝑡(𝐴) = 4, and 𝑏 = [ ] is the
−6 −8 6 1

constant vector.

2 −6
𝑇 2 3 −2 17 −48
𝐴𝐴 =[ ] [ 3 −8] = [ ] → 𝑑𝑒𝑡(𝐴 𝐴𝑇 ) = 8
−6 −8 6 −48 136
−2 6

98
 4 −48
(𝐴 𝐴𝑇 )1 = [ ] → 𝑑𝑒𝑡(𝐴 𝐴𝑇 )1 = 592
1 136

(𝐴 𝐴𝑇 )2 = [ 17 4
] → 𝑑𝑒𝑡(𝐴 𝐴𝑇 )2 = 209
−48 1

𝑎11 𝑑𝑒𝑡(𝐴 𝐴𝑇 )1 𝑎21 𝑑𝑒𝑡(𝐴 𝐴𝑇 )2 2 × 592 −6 × 209 −70

𝑥1 = + = + =
𝑑𝑒𝑡(𝐴 𝐴𝑇 ) 𝑑𝑒𝑡(𝐴 𝐴𝑇 ) 8 8 8

𝑎12 𝑑𝑒𝑡(𝐴 𝐴𝑇 )1 𝑎22 𝑑𝑒𝑡(𝐴 𝐴𝑇 )2 3 × 592 −8 × 209 104

𝑥2 = 𝑇
+ 𝑇
= + = = 13
𝑑𝑒𝑡(𝐴 𝐴 ) 𝑑𝑒𝑡(𝐴 𝐴 ) 8 8 8

𝑎13 𝑑𝑒𝑡(𝐴 𝐴𝑇 )1 𝑎23 𝑑𝑒𝑡(𝐴 𝐴𝑇 )2 −2 × 592 6 × 209 70

𝑥3 = 𝑇
+ 𝑇
= + =
𝑑𝑒𝑡(𝐴 𝐴 ) 𝑑𝑒𝑡(𝐴 𝐴 ) 8 8 8

We find that the solution of the system in this way is the same as pseudo
solution given in example 6.2.2 which comes from the general solution (see
theorem 6.1.6 )

−35
4 1
1
𝑥 = [ 13 ] + 2 𝑡 [ 0], at 𝑡=0
35
1
4

That is both a Generalization of Cramer's rule and pseudo method give the

same solution.

6.4 Particular Cases and Examples

In this section we shall consider some particular cases and examples to

illustrate the results of what we have done in the previous sections especially in

applying pseudo inverse (theorem 6.1.1) to some certain examples.

Example 6.4.1 [10 ]. Consider the following particular case of the system 𝑨𝒙 = 𝒃

𝑎11 𝑥1 + 𝑎12 𝑥2 + ⋯ . +𝑎1𝑛 𝑥𝑛 = 𝑏. (1)

In this case 𝑚 = 1 and 𝐴 = [𝑎11 , 𝑎12 , … . , 𝑎1𝑛 ].

99
 𝑎11
𝑎12
Then, if we define the column vector 𝐼1 = [ ⋮ ],
𝑎1𝑛

𝑎11
𝐴 𝐴𝑇 = [𝑎11 𝑎12 … 𝑎1𝑛 ] [𝑎12 ] = ‖𝐼1 ‖2 . (see definition 1.8.2)
⋮
𝑎1𝑛

1
Then, (𝐴 𝐴𝑇 )−1 𝒃 = ‖𝐼1 ‖2
𝒃 and

𝑎11 𝑎11 𝑏 ‖𝐼1 ‖−2

1 𝑎12 ‖𝐼 ‖−2
𝒙 = 𝐴𝑇 (𝐴 𝐴𝑇 )−1 𝒃 = 2
𝒃 [ ⋮ ] = 𝑎12 𝑏 1 .
‖𝐼1 ‖ ⋮
𝑎1𝑛 [ 𝑎1𝑛 𝑏 ‖𝐼1 ‖−2 ]

Therefore, a solution of the system (1) is given by:

𝑎1𝑗 𝑏 𝑎1𝑗 𝑏
𝑥𝑗 = 2
= 𝑛 2 , 𝑗 = 1,2, … , 𝑛
‖𝐼1 ‖ ∑𝑗=1 𝑎1𝑗

Example 6.4.2 [10 ].

Here we apply pseudo inverse to the case 𝑚 = 2, for any natural 𝑛 in 𝑨𝒙 = 𝒃

𝑎11 𝑥1 + 𝑎12 𝑥2 + ⋯ . +𝑎1𝑛 𝑥𝑛 = 𝑏1

(2)
𝑎21 𝑥1 + 𝑎22 𝑥2 + ⋯ . +𝑎2𝑛 𝑥𝑛 = 𝑏2

𝑎11 𝑎12 ⋯ 𝑎1𝑛

𝐴 = [𝑎 𝑎22 ⋯ 𝑎2𝑛 ]
21

Then, let 𝐼1 , 𝐼2 be the column vectors

𝑎11 𝑎21
𝑎12 𝑎22
𝐼1 = [ ⋮ ], 𝐼2 = [ ⋮ ].
𝑎1𝑛 𝑎2𝑛

Then,
𝑎11 𝑎21
𝑎11 𝑎12 ⋯ 𝑎1𝑛 𝑎12 𝑎22 ‖𝐼1 ‖2 〈𝐼1 , 𝐼2 〉
𝐴 𝐴𝑇 = [𝑎 𝑎22 ⋯ 𝑎2𝑛 ] [ ⋮ ⋮ ] = [ ].
21 〈𝐼2 , 𝐼1 〉 ‖𝐼2 ‖2
𝑎1𝑛 𝑎2𝑛

100
 1 ‖𝐼2 ‖2 −〈𝐼1 , 𝐼2 〉
(𝐴 𝐴𝑇 )−1 = [ ].
‖𝐼1 ‖2 ‖𝐼2 ‖2 − |〈𝐼1 , 𝐼2 〉|2 −〈𝐼2 , 𝐼1 〉 ‖𝐼1 ‖2

Hence, from the formula 𝒙𝒋 = 𝐴𝑇 (𝐴 𝐴𝑇 )−1 𝒃 we obtain that:

𝑥1 𝑎11 𝑎21
𝑥2 1 𝑎12 𝑎22 ‖𝐼2 ‖2 −〈𝐼1 , 𝐼2 〉 𝑏1
[ ⋮]= [ ] [ ] [ ].
‖𝐼1 ‖2 ‖𝐼2 ‖2 − |〈𝐼1 , 𝐼2 〉|2 ⋮ ⋮ −〈𝐼2 , 𝐼1 〉 ‖𝐼1 ‖2 𝑏2
𝑥𝑛 𝑎1𝑛 𝑎2𝑛

Therefore, a solution of the system (2) is given by :

𝑏1 ‖𝐼2 ‖2 − 𝑏2 〈𝐼1 , 𝐼2 〉 𝑏2 ‖𝐼1 ‖2 − 𝑏1 〈𝐼2 , 𝐼1 〉

𝑥1 = 𝑎11 + 𝑎21
‖𝐼1 ‖2 ‖𝐼2 ‖2 − |〈𝐼1 , 𝐼2 〉|2 ‖𝐼1 ‖2 ‖𝐼2 ‖2 − |〈𝐼1 , 𝐼2 〉|2

𝑏1 ‖𝐼2 ‖2 − 𝑏2 〈𝐼1 , 𝐼2 〉 𝑏2 ‖𝐼1 ‖2 − 𝑏1 〈𝐼2 , 𝐼1 〉

𝑥2 = 𝑎12 + 𝑎22
‖𝐼1 ‖2 ‖𝐼2 ‖2 − |〈𝐼1 , 𝐼2 〉|2 ‖𝐼1 ‖2 ‖𝐼2 ‖2 − |〈𝐼1 , 𝐼2 〉|2

𝑏1 ‖𝐼2 ‖2 − 𝑏2 〈𝐼1 , 𝐼2 〉 𝑏2 ‖𝐼1 ‖2 − 𝑏1 〈𝐼2 , 𝐼1 〉

𝑥𝑛 = 𝑎1𝑛 + 𝑎2𝑛
‖𝐼1 ‖2 ‖𝐼2 ‖2 − |〈𝐼1 , 𝐼2 〉|2 ‖𝐼1 ‖2 ‖𝐼2 ‖2 − |〈𝐼1 , 𝐼2 〉|2

Example 6.4.3. Find a solution of the following system

𝑥1 + 𝑥2 = 1
−𝑥1 + 𝑥2 + 𝑥3 = −1

Solution: With the above notations

1 −1
1
𝐼1 = [1] , 𝐼2 = [ 1 ] , 𝑏=[ ]
−1
0 1

Then, ‖𝐼1 ‖2 = 1 + 1 = 2 , ‖𝐼2 ‖2 = 1 + 1 + 1 = 3, |〈𝐼1 , 𝐼2 〉| = −1 + 1 + 0 = 0

𝑑𝑒𝑡(𝐴 𝐴𝑇 ) = ‖𝐼1 ‖2 ‖𝐼2 ‖2 − |〈𝐼1 , 𝐼2 〉|2 = 2 × 3 − 0 = 6

101
 𝑏1 ‖𝐼2 ‖2 − 𝑏2 〈𝐼1 , 𝐼2 〉 𝑏2 ‖𝐼1 ‖2 − 𝑏1 〈𝐼2 , 𝐼1 〉
𝑥1 = 𝑎11 + 𝑎 21
‖𝐼1 ‖2 ‖𝐼2 ‖2 − |〈𝐼1 , 𝐼2 〉|2 ‖𝐼1 ‖2 ‖𝐼2 ‖2 − |〈𝐼1 , 𝐼2 〉|2

1(1 × 3 − −1 × 0) (−1)(−1 × 2 − 1 × 0) 3 2 5
= + = + =
6 6 6 6 6

𝑏1 ‖𝐼2 ‖2 − 𝑏2 〈𝐼1 , 𝐼2 〉 𝑏2 ‖𝐼1 ‖2 − 𝑏1 〈𝐼2 , 𝐼1 〉

𝑥2 = 𝑎12 + 𝑎22
‖𝐼1 ‖2 ‖𝐼2 ‖2 − |〈𝐼1 , 𝐼2 〉|2 ‖𝐼1 ‖2 ‖𝐼2 ‖2 − |〈𝐼1 , 𝐼2 〉|2

1(3 − 0) (1)(−2 − 0) 3 −2 1
= + = + =
6 6 6 6 6

𝑏1 ‖𝐼2 ‖2 − 𝑏2 〈𝐼1 , 𝐼2 〉 𝑏2 ‖𝐼1 ‖2 − 𝑏1 〈𝐼2 , 𝐼1 〉

𝑥3 = 𝑎13 + 𝑎 23
‖𝐼1 ‖2 ‖𝐼2 ‖2 − |〈𝐼1 , 𝐼2 〉|2 ‖𝐼1 ‖2 ‖𝐼2 ‖2 − |〈𝐼1 , 𝐼2 〉|2

(0)(3) (1)(−2) −2
= + =
6 6 6

We note that if ‖𝐼1 ‖2 ‖𝐼2 ‖2 = |〈𝐼1 , 𝐼2 〉|2 this means that the angle between

𝐼1 , 𝐼2 the equals zero and this indicates that the system of equations are identical,

meaning that there is an infinite number of solutions and in this case it is the

same as in example 6.4.1.

But if the angle between 𝐼1 , 𝐼2 is not equal to zero, then this means that the

solution of the system has an infinite number of solutions.

We note that the general solution for this example given by Theorem 6.1.6 is

𝑥 = 𝐴𝑇 (𝐴 𝐴𝑇 )−1 𝑏 + (𝐼 − 𝐴𝑇 (𝐴 𝐴𝑇 )−1 𝐴)𝑦 = 𝐴−1 −1

𝑅 𝑏 + (𝐼 − 𝐴𝑅 𝐴)𝑦

1 5 1 1
𝑥 = [ 1 ] + 𝑡 [ −1]
6 6
−2 2

Choose 𝑡 = 0, 𝑥 = (5⁄6 , 1⁄6 , −2⁄6)

102
Example 6.4.4 [10 ]. Consider the following general case of system 𝑨𝒙 = 𝒃

𝑎11 𝑥1 + 𝑎12 𝑥2 + … + 𝑎1𝑛 𝑥𝑛 = 𝑏1

𝑎21 𝑥1 + 𝑎22 𝑥2 + … + 𝑎2𝑛 𝑥𝑛 = 𝑏2
⋮ ⋮ ⋱ ⋮ (3)
⋮
𝑎𝑚1 𝑥1 + 𝑎𝑚2 𝑥2 + … + 𝑎𝑚𝑛 𝑥𝑛 = 𝑏
𝑛

Then, if {𝐼1 , 𝐼2 , … , 𝐼𝑚 } is an orthogonal set in 𝑅 𝑛 , (see definition 1.9.4) we obtain

‖𝐼1 ‖2 0 0 ⋯ 0
‖𝐼2 ‖2 ⋯ 0
𝐴 𝐴𝑇 = [ 0 0
⋮ ]
⋮ ⋮ ⋮ ⋱
0 0 0 ⋯ ‖𝐼𝑛 ‖2

The solution of the system 𝑨 𝒙 = 𝒃 is simple and is given by:

𝑥𝑗 = ∑ 𝑎𝑖𝑗 𝑏𝑗𝑖 ‖𝐼𝑖 ‖−2 , 𝑗 = 1,2, … , 𝑛

𝑖=1

Example 6.4.5. Find the solution of the following system

−𝑥1 − 𝑥2 + 𝑥3 + 𝑥4 = 1
−𝑥1 + 𝑥2 − 𝑥3 + 𝑥4 = 1
𝑥1 − 𝑥2 − 𝑥3 + 𝑥4 = 1

Solution:

−1 −1 1
1
𝐼1 = [−1] , 𝐼2 = [ 1 ], 𝐼3 = [−1] , 𝑏 = [1]
1 −1 −1
1
1 1 1

Then, ‖𝐼1 ‖2 = 4 , ‖𝐼2 ‖2 = 4, ‖𝐼3 ‖2 = 4

𝑑𝑒𝑡(𝐴 𝐴𝑇 ) = ‖𝐼1 ‖2 ‖𝐼2 ‖2 ‖𝐼3 ‖2 = 4 × 4 × 4 = 64

𝑎11 𝑏1 𝑎21 𝑏2 𝑎31 𝑏3 (−1)(1) (−1)(1) (1)(1) −1

𝑥1 = + + = + + =
‖𝐼1 ‖2 ‖𝐼2 ‖2 ‖𝐼3 ‖2 4 4 4 4

𝑎12 𝑏1 𝑎22 𝑏2 𝑎32 𝑏3 (−1)(1) (1)(1) (−1)(1) −1

𝑥2 = + + = + + =
‖𝐼1 ‖2 ‖𝐼2 ‖2 ‖𝐼3 ‖2 4 4 4 4

103
 𝑎13 𝑏1 𝑎23 𝑏2 𝑎33 𝑏3 (1)(1) (−1)(1) (−1)(1) −1
𝑥3 = + + = + + =
‖𝐼1 ‖2 ‖𝐼2 ‖2 ‖𝐼3 ‖2 4 4 4 4

𝑎14 𝑏1 𝑎24 𝑏2 𝑎34 𝑏3 (1)(1) (1)(1) (1)(1) 3

𝑥4 = + + = + + =
‖𝐼1 ‖2 ‖𝐼2 ‖2 ‖𝐼3 ‖2 4 4 4 4

We note that the general solution for this example given by Theorem 6.1.6 is

𝑥 = 𝐴𝑇 (𝐴 𝐴𝑇 )−1 𝑏 + (𝐼 − 𝐴𝑇 (𝐴 𝐴𝑇 )−1 𝐴)𝑦 = 𝐴−1 −1

𝑅 𝑏 + (𝐼 − 𝐴𝑅 𝐴)𝑦

−1 1
1 −1 1 1]
𝑥 = [ ]+ 𝑡[
4 −1 4 1
3 1

−1 −1 −1 3
Choose 𝑡 = 0, 𝑥=( , , , )
4 4 4 4

Finally we note that if a system of linear equations has an infinite number of

solutions, then we can use a suitable method discussed in this chapter to find a
fixed solution and construct the general solution mentioned in Theorem 6.1.6.
Certainly, it will be the same general solution given by Gauss elimination(see
definition 1.7.2).

104
References: -

[1] Amiri, M., Fathy, M., Bayat, M., Generalization of some determinantal identities for
non-square matrices based on Radics definition, TWMS J. Pure Appl. Math. 1, no 2(2010),
163-175.

[2] Anton, H., Ilementary Linear Algebra with application, Seventh edition.

[3] Arunkumar, M., Murthy, S., and Ganapath, G., Determinant for non-square
matrices, IJMSEA, no 5 (2011), 389-401.

[4] Barata, J.C.A.,. Hussein, M.S; The Moore-Penrose Pseudoinverse. A Tutorial

Review of the Theory. John Hopkins University Press,(2013).

[5] Jamshidvanda,S., Ghalandarzadeha,S., Amiraslanib,A., Oliaa,F., On the

Maximal Solution of A Linear System over Tropical Semirings, STEM
Department, The University of Hawaii-Maui College, Kahului, Hawaii, USA,
(June 26, 2019).

[6] Jimmy, The Moore-Penrose Pseudoinverse, Math 33A: Laub, University of

California - Los Angeles(2011).

[7] Joshi, V.N., A determinant for rectangular matrices, no. 21 (1980) , 137-146.

[8] Joyce,D., Determinants, part II , Math 130 Linear Algebra, Clark

University(2015).

[9] Kolman, B., Introductory Linear Algebra with application, Six edition.

[10] Leiva, H., A Generalization of Cramer's Rule, Advances in linear Algebra

and Matrix Theory, no 5 (2015), 156-166.

[11] Leon,S.,J., Linear Algebra with Applications, Fifth edition.

[12]Makarewicz,A., Pikuta, P., and Szalkowski, D., Properties of the determinant

of a rectangular matrix, AUMC-S L-P, LXVIII, no 1(2014), 31-41.

105
[13] Radic, M.,A bout a determinant of rectangular 2 × 𝑛 matrix and its
geometric interpretation, Beitrage Algebra Geom. 46,no.1(2005),321-349.

[14] Radić, M., A definition of determinant of rectangular matrix, Glas. Mat. Ser.
III 1(21) (1966), 17–22.

106