INDIAN INSTITUTE OF TECHNOLOGY ROORKEE

MAI-101: MATHEMATICS-I

UADAY SINGH

DEPARTMENT OF MATHEMATICS
 About the course
The course is divided into 4 units:
• Matrix Algebra
• Differential Calculus (more than one variable)
• Integral Calculus (double and triple integrals
• Vector Calculus

The detailed syllabus and reference books can be seen on the link:
https://iitr.ac.in/Departments/Mathematics%20Department/Acade
mics/Course%20Structure.html
Lecture: Tutorial: Practical :: 3:1:0 Per Week
Preferred Books:
1. E. Kreyszig, Advanced Engineering Mathematics, 9th edition, John Wiley and Sons, Inc., U. K.
2. R.K. Jain and S.R.K. Iyenger, Advanced Engineering Mathematics, 2nd Edition, Narosa
Publishing House.

2
 Evaluation System
• Class Work Sessional (CWS): 25%
It will be based on assignment, quiz, regularity in classes etc.

• Mid-Term Examination: 25% (Subjective Exam of 1.5 Hours)

• End Term Examination (ETE): 50% (Subjective Exam of 3 Hours)

3
 MATRIX ALGEBRA
Notation: An 𝑚 × 𝑛 matrix 𝐴 = (𝑎𝑖𝑗 ) is usually written as
𝑎11 𝑎12 … 𝑎1𝑗 … 𝑎1𝑛 𝑅1
𝑎21 𝑎22 … 𝑎2𝑗 … 𝑎2𝑛 𝑅2
𝐴= ⋮ ⋮ ⋮ ⋮ ⋮ ⋮
𝑎𝑖1 𝑎𝑖2 … 𝑎𝑖𝑗 … 𝑎𝑖𝑛 𝑅𝑖
⋮ ⋮ ⋮ ⋮ ⋮ ⋮
𝑎𝑚1 𝑎𝑚2 … 𝑎𝑚𝑗 … 𝑎𝑚𝑛 𝑅 𝑚
𝐶1 𝐶2 … 𝐶𝑗 … 𝐶𝑛

• The 𝑖𝑡ℎ row of the matrix A is written as

𝑎𝑖1 𝑇
𝑎𝑖2
⋮
𝑅𝑖 = 𝑎𝑖1 , 𝑎𝑖2 , … , 𝑎𝑖𝑗 , … , 𝑎𝑖𝑛 = 𝑎 for 𝑖 = 1,2, … , 𝑚
1×𝑛 𝑖𝑗
⋮
𝑎𝑖𝑛 𝑛×1
4
• The 𝑗𝑡ℎ column of the matrix A is written as
𝑎1𝑗
𝑎2𝑗
𝑇
⋮
𝐶𝑗 = (𝑎1𝑗 , 𝑎2𝑗 ,…,𝑎𝑖𝑗 , …, 𝑎𝑚𝑗 ) = 𝑎 for 𝑗 = 1, 2, … , 𝑛
𝑖𝑗
⋮
𝑎𝑚𝑗
𝑚×1
• Often the rows and columns of a matrix are also referred as row vectors and
column vectors, respectively.
• The zero matrix or zero vector denoted by 𝑶 is a matrix of the specified
order whose all entries are 0. The matrix 𝑶 can have a suitable form
depending on the context, for example,

0
0 0 0
0
𝑶 = 0,0, … , 0 1×𝑛 or 𝑶 = or 𝑶 = 0 0 0 etc.
⋮
0 0 0
0 𝑛×1

5
 Elementary Row and Column Operations
• The following three operations on a matrix A are called the elementary row
operations:

1. Multiply a row by a non-zero constant for which we use the symbol 𝑐𝑅𝑖
or 𝑅𝑖 → 𝑐𝑅𝑖 (multiply the 𝑖 𝑡ℎ row by a non-zero constant 𝑐)
2. Interchange any two rows for which we use the symbol 𝑅𝑖𝑗 or 𝑅𝑖 ↔ 𝑅𝑗
(interchange row 𝑖 and row 𝑗)
3. Add a non-zero multiple of one row to any other row for which we use
the symbol 𝑐𝑅𝑖 + 𝑅𝑗 or 𝑅𝑗 → 𝑅𝑗 + 𝑐𝑅𝑖 (multiply the 𝑖 𝑡ℎ row by 𝑐 and add
to the 𝑗𝑡ℎ row)

• A matrix A is said to be row equivalent to a matrix B, if the matrix A can be

obtained from the matrix B by using a finite sequence of elementary row
operations. Then we usually write 𝐴~𝐵.
• Similarly, we can have the elementary column operations with similar
notations.

6
 Example: 1
2 6 1 7
Let us consider the matrix 𝐵 = 1 2 −1 −1 . Then
5 7 −4 9
2 6 1 7 1 2 −1 −1 𝑅2 → 𝑅2 − 2𝑅1 1 2 −1 −1
𝑅1 ↔ 𝑅2
1 2 −1 −1 2 6 1 7 𝑅3 → 𝑅3 − 5𝑅1 0 2 3 9
~ ~
5 7 −4 9 5 7 −4 9 0 −3 1 14
3 1 2 −1 −1 2 1 2 −1 −1
𝑅3 → 𝑅3 + 𝑅2 0 2 3 9 𝑅3 → 11 𝑅3 0 2 3 9 = 𝐴.
2
~ 0 0 11/2 55/2 ~ 0 0 1 5
Thus 𝐴~𝐵. Later we will see that this particular form of 𝐵 or 𝐴 is important.
Again, if we apply the same row operations on the 3 × 3 identity matrix 𝐼3 , we
get
0 1 0 0 1 0
1 0 0 0 1 0 0 1 0
1 −2 0 1 −2 0
𝐼3 = 0 1 0 ~ 1 0 0 ~ 1 −2 0 ~ 3 ~ 3 16 2 = 𝐼3 ′
0 0 1 0 0 1 0 −5 1 −8 1 −
2 11 11 11

7
 0 1 0 2 6 1 7 1 2 −1 −1
Now, 𝐼3′ 𝐵 = 13 −2
16
0
2
1 2 −1 −1 = 0 2 3 9 = 𝐴.
− 5 7 −4 9 0 0 1 5
11 11 11
We note that the matrix 𝐴 can be obtained by two ways:
1. by applying a sequence of elementary row operations on 𝐵.
2. pre-multiplying 𝐵 with 𝐼3′ , a matrix obtained by using the same elementary
operations on 𝐼3 .

Elementary Matrix: An 𝑚 × 𝑚 matrix 𝐸 is said to be an elementary matrix, if it is

obtained by performing an elementary (row or column) operation on the 𝑚 × 𝑚
identity matrix 𝐼𝑚×𝑚 .
0 1 0 3 0 0 1 0 0
Examples: 1 0 0 𝑅1 ↔ 𝑅2 ; 0 1 0 𝑅1 → 3𝑅1 ; 0 1 2 𝑅2 → 𝑅2 + 2𝑅3 .
0 0 1 0 0 1 0 0 1
 It can also be noted that to each elementary operation 𝑒, there is a same type
elementary row operation 𝑒′ such that 𝑒 ′ 𝑒𝐴 = 𝑒 𝑒′𝐴 = 𝐴.
Again consider the elementary matrices used in Example 1.
8
 0 1 0 1 0 0 1 0 0 1 0 0 1 0 0
𝐸1 = 1 0 0 , 𝐸2 = −2 1 0 , 𝐸3 = 0 1 0 , 𝐸4 = 0 13 0 , 𝐸5 = 0 1 02
0 0 1 0 0 1 −5 0 1 0 2 1 0 0 11
0 1 0 2 6 1 7 1 2 −1 −1
Now, 𝐸1 𝐵 = 1 0 0 1 2 −1 −1 = 2 6 1 7 (𝑅1 ↔ 𝑅2 )
0 0 1 5 7 −4 9 5 7 −4 9
1 0 0 1 2 −1 −1 1 2 −1 −1
𝐸2 𝐸1 𝐵 = −2 1 0 2 6 1 7 = 0 2 3 9 (𝑅2 → 𝑅2 − 2𝑅1 )
0 0 1 5 7 −4 9 5 7 −4 9
1 0 0 1 2 −1 −1 1 2 −1 −1
𝐸3 𝐸2 𝐸1 𝐵 = 0 1 0 0 2 3 9 = 0 2 3 9 (𝑅3 → 𝑅3 − 5𝑅1 )
−5 0 1 5 7 −4 9 0 −3 1 14
1 0 0 1 2 −1 −1 1 2 −1 −1
3
𝐸4 𝐸3 𝐸2 𝐸1 𝐵 = 0 13 0 0 2 3 9 = 0 2 3 9 (𝑅3 → 𝑅3 + 2 𝑅2 )
0 2 1 0 −3 1 14 0 0 11/2 55/2
1 0 0 1 2 −1 −1 1 2 −1 −1
2
𝐸5 𝐸4 𝐸3 𝐸2 𝐸1 𝐵 = 0 1 0 0 2 3 9 = 0 2 3 9 = 𝐴 (𝑅3 → 11 𝑅3 )
0 0 2/11 0 0 11/2 55/2 0 0 1 5

9
 1 0 0 1 0 0 1 0 0 1 0 0 0 1 0 0 1 0
Also, 𝐸5 𝐸4 𝐸3 𝐸2 𝐸1 = 0 1 0
2
0 1
3
0 0 1 0 −2 1 0 1 0 0 = 13 −216
0 .
2
0 0 0 1 −5 0 1 0 0 1 0 0 1 −
11 2 11 11 11
= 𝐼3′ = 𝑃, say.
Thus the following results hold.
1. Let e be an elementary row operation and 𝐸 be the corresponding 𝑚 × 𝑚 elementary matrix,
that is, 𝐸 = 𝑒(𝐼𝑚 ). Then for every 𝑚 × 𝑛 matrix 𝐴, we have 𝑒 𝐴 = 𝐸𝐴.
2. Let 𝐴 and 𝐵 be two 𝑚 × 𝑛 matrices. Then 𝐴 is row-equivalent to 𝐵 if and only if there is an
invertible (non-singular) 𝑚 × 𝑚 matrix 𝑃 = 𝐸𝑘 𝐸𝑘−1 … 𝐸1 (for some 𝑘) such that 𝐵 = 𝑃𝐴, where 𝑃
is the product of a finite number of 𝑚 × 𝑚 elementary matrices.
3. In case of column operations or column equivalence there is an invertible (non-singular) 𝑛 × 𝑛
matrix 𝑃 = 𝐸1 𝐸2 … 𝐸𝑘 (for some 𝑘) such that 𝐵 = 𝐴𝑃, where 𝑃 is obtained by applying the same
column operations on 𝐼𝑛 which we want to apply on the columns of 𝐴.

Row equivalence as a relation:

 𝑨~𝑨: Since there is an invertible matrix 𝐼 such that 𝐴 = 𝐼𝐴, 𝐴~𝐴.
 𝑨~𝑩 implies that 𝑩~𝑨: There is an invertible matrix 𝑃 such that 𝐴 = 𝑃𝐵 or 𝑃−1 𝐴 = 𝐵. Hence
𝐵~𝐴.
 𝑨~𝑩 and 𝑩~𝑪 implies that 𝑨~𝑪: There are two invertible matrices of the same order such that
𝐴 = 𝑃𝐵 and 𝐵 = 𝑄𝐶. Therefore, 𝐴 = 𝑃 𝑄𝐶 = 𝑃𝑄 𝐶, where 𝑃𝑄 is an invertible matrix. Hence
𝐴~𝐶.
10
Row - reduced matrix: An 𝑚 × 𝑛 matrix 𝐴 is said to be row-reduced if:
(a) the first non-zero entry in each non-zero row (called leading non-zero entry) of 𝐴 is
equal to 1.
(b) Each column of 𝐴 which contains the leading non-zero entry of some row has all its
other entries 0. Such columns are called the pivot columns.
1 0 0 0 0 2 1
Examples: 1. 𝐴 = 0 1 −1 0 and 𝐵 = 1 0 −3 are not in row-reduced form.
0 0 1 0 0 0 0
0 1 −3 0 1/2 1 0 0
2. 𝐶 = 0 0 0 1 2 and 𝐷 = 0 1 0 are in row-reduced form.
0 0 0 0 0 0 0 1

Note that two matrices are said to be row equivalent if and only if they have the same
row reduced forms.
H.W.: Prove that the following two matrices are not row-equivalent.
2 0 0 1 1 2
𝑎 −1 0 , −2 0 −1
𝑏 𝑐 3 1 3 5

11
Row-echelon form: An 𝑚 × 𝑛 matrix 𝐴 is said to be in row-echelon form, if
1) every row of 𝐴 which has all its entries 0 occurs below the non-zero rows
2) the first non-zero entry of a non-zero row is 1
3) in consecutive non-zero rows, the first entry 1 in a lower row appears to the right of
1 in the upper row
Examples: 1. The identity matrix and zero matrix are the trivial examples of row-
echelon form. The matrix 𝐵 in previous example is not in row-echelon form.
2 6 1 7
2. Let us consider the matrix 𝐵 = 1 2 −1 −1 of Example 1 on Slide 7.
5 7 −4 9
This was reduced to
1 2 −1 −1 1 2 −1 −1
0 2 3 9 ~ 0 1 3/2 9/2 ,
0 0 1 5 0 0 1 5
which is in the row-echelon form.

12
 If a matrix 𝐴 (in row-echelon form) is reduced into row-reduced form, then it
is called in row-reduced echelon form. The matrices 𝐶 and 𝐷 in Slide -11
are in row-reduced echelon form.
 The row-echelon form of a matrix can be used to find a solution of a system
of linear equation. This process is called, the Gauss Elimination Method
also.
Example 2: Let us find a solution to the system of linear equations:
2𝑥1 − 𝑥2 + 3𝑥3 + 2𝑥4 = 0
𝑥1 + 4𝑥2 − 𝑥4 = 0
2𝑥1 + 6𝑥2 − 𝑥3 + 5𝑥4 = 0
This system can be written in the following matrix form (𝐴𝑋 = 𝐵):
𝑥1
2 −1 3 2 𝑥2 0
1 4 0 −1 = 0 . (1)
𝑥3
2 6 −1 5 𝑥4 0
We know that the elimination process for solving a system of linear equations and the
elementary row operations are the same.

13
Applying 𝑅1 → 𝑅1 − 2𝑅2 and 𝑅3 → 𝑅3 − 2𝑅2 , we have
2
2 −1 3 2 9
0 −9 3 4 𝑅1 → 𝑅1 − 𝑅3 0 0 15/2 −55/2 1 15 𝑅1
𝑅 →
1 4 0 −1 ~ 1 4 0 −1 2 1 0 −2 13 1
2 6 −1 5 0 −2 −1 7 𝑅2 → 𝑅2 + 2𝑅3 0 −2 −1 7 𝑅 3 → − 𝑅
~ 2 3
~
0 0 1 −11/3 𝑅2 → 𝑅2 +2𝑅1 0 0 1 −11/3 1 0 0 17/3
1 0 −2 13 𝑅3 → 𝑅3 − 2𝑅3 1 0 0 17/3 ~ 0 1 0 −5/3 = 𝐴′(Row-
0 1 1/2 −7/2 ~ 0 1 0 −5/3 0 0 1 −11/3
interchange.
Thus the system of equations (1) and A′𝑋 = 𝑂, that is,
17 5 11
𝑥1 + 𝑥 = 0; 𝑥2 − 3 𝑥4 = 0; 𝑥3 − 𝑥 =0
3 4 3 4
are the exactly same and have the same solutions.
Finally, if we assign any arbitrary value, say 𝑡, to 𝑥4 , then solution of the system is 𝑥1 =
17 5 11
− 3 𝑡; 𝑥2 = 3 𝑡; 𝑥3 = 3 𝑡; 𝑥4 = 𝑡, where 𝑡 is an arbitrary real number. Hence the given
system of equations has infinitely many solutions.
Note that we have reduced the augmented matrix into row-echelon form and used the
back substitution.
14
Example 3: Solve the system of equations
2𝑥1 + 6𝑥2 + 𝑥3 = 7 1 2 −1 𝑥1 −1
𝑥1 + 2𝑥2 − 𝑥3 = −1 or 2 6 1 𝑥2 = 7 (𝐴𝑋 = 𝐵). (1)
5𝑥1 + 7𝑥2 − 4𝑥3 = 9 5 7 −4 𝑥3 9
The matrix [A ⋮ 𝐵] is called the augmented matrix. Here
1 2 −1⋮−1 𝑅2 → 𝑅2 − 2𝑅1 1 2 −1⋮−1
A ⋮ 𝐵 = 2 6 1 ⋮ 7 𝑅3 → 𝑅3 − 5𝑅1 0 2 3⋮9
5 7 −4⋮ 9 ~ 0 −3 1 ⋮ 14
1
𝑅2 → 2 𝑅2 1 2 −1 ⋮ −1 2 1 2 −1 ⋮ −1
𝑅3 → 11 𝑅3 0 1 3/2⋮9/2 .
𝑅3 → 𝑅3 + 3𝑅2 0 1 3/2 ⋮ 9/2 ~
~ 0 0 11/2⋮55/2 0 0 1 ⋮ 5
Thus the system (1) of equations is equivalent to the system
𝑥1 + 2𝑥2 − 𝑥3 = −1 𝑥3 = 5
3 9 9 15
𝑥2 + 2 𝑥3 = 2 which gives 𝑥2 = 2 − 2 = −3 or 𝑥1 , 𝑥2 , 𝑥3 = (10, −3,5).
𝑥3 = 5 𝑥1 = −1 − 2 −3 + 5 = 10
(a unique solution) (In other words the three planes intersect in a point).

15
Example 4: Consider the system of equations
𝑥1
𝑥1 − 2𝑥2 + 𝑥3 + 2𝑥4 = 1 1 −2 1 2 𝑥 1
2
𝑥1 + 𝑥2 − 𝑥3 + 𝑥4 = 2 or 1 1 −1 1 = 2.
𝑥3
𝑥1 + 7𝑥2 − 5𝑥3 − 𝑥4 = 3 1 7 −5 −1 𝑥 3
4
The augmented matrix is
1 −2 1 2 ⋮1 1 −2 1 2 ⋮1 1 −2 1 2 ⋮ 1
1 1 −1 1 ⋮2 ~ 0 3 −2 −1⋮1 ~ 0 1 −2/3 −1/3⋮1/3 .
1 7 −5 −1⋮3 0 9 −6 −3⋮2 0 0 0 0 ⋮ −1
The given system of equations is equivalent to
𝑥1 − 2𝑥2 + 𝑥3 + 2𝑥4 = 1
0𝑥1 + 3𝑥2 − 2𝑥3 − 𝑥4 = 1 (degenerate case).
0𝑥1 + 0𝑥2 + 0𝑥3 + 0𝑥4 = −1
There cannot be any 𝑥1 , 𝑥2 , 𝑥3 , 𝑥4 which satisfies the last equation. Hence this
system has no solution.

16
Linear independence/dependence (Rank of matrix)

A set of vectors {𝑢1 , 𝑢2 , … , 𝑢𝑛 } (row vectors or column vectors) is said to be

linearly independent (LI), if the only scalars/constants satisfying the equation
𝑘1 𝑢1 + 𝑘2 𝑢2 + ⋯ + 𝑘𝑛 𝑢𝑛 = 𝑂
(1)
(𝑙𝑖𝑛𝑒𝑎𝑟 𝑐𝑜𝑚𝑏𝑖𝑛𝑎𝑡𝑖𝑜𝑛)
are 𝑘1 = 𝑘2 = ⋯ = 𝑘𝑛 = 0.
If the set is not linearly independent, that is the homogeneous system of
equations in (1) has a non-zero solution also, is called linearly dependent
(LD). [In other words one vector can be written as a linear combination of the
others]
Example 5: (a) The rows of a 3 × 3 identity matrix, that is, 𝑢1 = (1,0,0), 𝑢2 =
(0,1,0), 𝑢3 = (0,0,1) are LI.
Let us write 𝑘1 𝑢1 + 𝑘2 𝑢2 + 𝑘3 𝑢3 = 𝑂 or 𝑘1 , 0,0 + 0, 𝑘2 , 0 + 0,0, 𝑘3 = (0,0,0).
This gives 𝑘1 , 𝑘2 , 𝑘3 = (0,0,0), that is, 𝑘1 = 𝑘2 = 𝑘3 = 0. Hence rows of the
3 × 3 identity matrix are LI.
(b) The vectors 1,1,1 , 2 − 1,4 , (5,2,7) are LD.
Here 3 1,1,1 + 2 − 1,4 − 5,2,7 = (0,0,0) or 3 1,1,1 + 2 − 1,4 = 5,2,7 .
17
 RANK OF A MATRIX
If we use the definition, 𝑘1 1,1,1 + 𝑘2 2, −1,4 + 𝑘3 5,2,7 = (0,0,0) gives
𝑘1 + 2𝑘2 + 5𝑘3 , 𝑘1 − 𝑘2 + 2𝑘3 , 𝑘1 + 4𝑘2 + 7𝑘3 = (0,0,0)
𝑘1 + 2𝑘2 + 5𝑘3 = 0 𝑘1 + 2𝑘2 + 5𝑘3 = 0 𝑘1 + 2𝑘2 + 5𝑘3 = 0
or 𝑘1 − 𝑘2 + 2𝑘3 = 0 or 0𝑘1 − 3𝑘2 − 3𝑘3 = 0 or 0𝑘1 + 𝑘2 + 𝑘3 = 0 (2).
𝑘1 + 4𝑘2 + 7𝑘3 = 0 0𝑘1 +2𝑘2 + 2𝑘3 = 0 0𝑘1 +0𝑘2 + 0𝑘3 = 0
This gives 𝑘2 = −𝑘3 and 𝑘1 = −3𝑘3 . Thus system (2) has infinitely many
solutions. Let us take 𝑘3 = 1, 𝑘2 = −1, 𝑘1 = −3. Hence the given vectors are
LD.

 Subset of a linearly independent set is linearly independent

 Superset of a linearly dependent set is linearly dependent
 A set with zero vector is always linearly dependent

The rank of an 𝑚 × 𝑛 matrix 𝐴 is the maximum number of linearly independent row (or
column) vectors in 𝐴. The rank of the matrix 𝐴 is usually denoted by rank(𝐴) or 𝜌(𝐴).

18
 1 1 −1 3
Example 6: Let us consider the 3 × 4 matrix 𝐴 = 2 −2 6 8 .
3 5 −7 8
Let us try to find LI rows of 𝐴. Let 𝑎, 𝑏, 𝑐 be three constants such that
𝑎 1,1, −1,3 + 𝑏 2, −2,6,8 + 𝑐 3,5, −7,8 = (0,0,0,0).
𝑎 + 2𝑏 + 3𝑐 = 0 Thus we have
or 𝑎 − 2𝑏 + 5𝑐 = 0 −8 1,1, −1,3 + 1 2, −2,6,8 + 2 3,5, −7,8 = 0.
−𝑎 + 6𝑏 − 7𝑐 = 0 The three rows of the matrix 𝐴 are LD. Also none of
3𝑎 + 8𝑏 + 8𝑐 = 0 the first row and second row is a scalar multiple of
or 2𝑎 + 8𝑐 = 0; 4𝑏 − 2𝑐 = 0 other rows, that is, first two rows are LI. Hence rank
of 𝐴 is 2.
or 𝑎 = −8𝑏, 𝑐 = 2𝑏.
The rank or LI rows can also be determined by using
If we set 𝑏 = 1, then 𝑎 = −8, 𝑐 = 2. row-echelon form.

1 1 −1 3 1 1 −1 3
𝐴 = 2 −2 6 8 ~ 0 1 −2 −1/2 .
3 5 −7 8 0 0 0 0
The number of non-zero rows in the row-echelon form of 𝐴 is two. Therefore, the maximum
number of LI rows in 𝐴 is two. Hence rank 𝐴 = 2.
19
Thus the rank of 𝐴 is also defined as the number of non-zero rows in the row-echelon form of
𝐴. Also rank 𝐴𝑚×𝑛 = number of LI rows ≤ 𝑚.
 The three row-vectors 𝑢1 = 1,1, −1,3 , 𝑢2 = 2, −2,6,8 , 𝑢3 = 3,5, −7,8 in the
previous Example 6 are LD as the rank of the matrix whose rows are 𝑢1 , 𝑢2 , 𝑢3 is less
than 3.
 In general, the vector 𝑢1 = 𝑎11 , 𝑎12 , … , 𝑎1𝑛 , 𝑢2 = 𝑎21 , 𝑎22 , … , 𝑎2𝑛 , … , 𝑢𝑚 =
𝑎𝑚1 , 𝑎𝑚2 , … , 𝑎𝑚𝑛 are LD or LI according as the matrix
𝑎11 𝑎12 ⋯ 𝑎1𝑛
𝑎 𝑎22 ⋯ 𝑎2𝑛
𝐴 = 21 ⋯ ⋯
⋯ ⋯
𝑎𝑚1 𝑎𝑚2 ⋯ 𝑎𝑚𝑛
has rank (number of non-zero rows in the row echelon form of 𝐴) less than 𝑚 or equal to
𝑚.
For example in order to determine whether the vectors 𝑢1 = 2,1,1 , 𝑢2 = 0,3,0 and
𝑢3 = (3,1,2) are LI, one can consider the matrix 𝐴 whose rows are 𝑢1 , 𝑢2 and 𝑢3 .
2 1 1 1 1/2 1/2 1 1/2 1/2 1 1 0 1 0 0 1 0 0
𝐴= 0 3 0 ~ 3 1 2 ~ 0 −1/2 1/2 ~ 0 −1 1 ~ 0 0 1 ~ 0 1 0 .
3 1 2 0 1 0 0 1 0 0 1 0 0 1 0 0 0 1
Thus rank 𝐴 = 3 = number of rows. This implies that given vectors are LI
20
 ROW RANK vs COLUMN RANK
Let row rank 𝐴𝑚×𝑛 = 𝑟 and column rank 𝐴𝑚×𝑛 = 𝑝. Also let 𝑢𝑗 = (𝑢𝑗1 , 𝑢𝑗2 , … , 𝑢𝑗𝑛 ), 𝑗 =
1, 2, … , 𝑚 be the row vectors of 𝐴. By definition of rank, 𝐴 has 𝑟 linearly independent
rows which we denote by 𝑣1 , 𝑣2 , … , 𝑣𝑟 (regardless their position in 𝐴). Also all the rows of
𝐴 are linear combinations of 𝑣1 , 𝑣2 , … , 𝑣𝑟 . Therefore, we can write
𝑢1 = 𝑐11 𝑣1 + 𝑐12 𝑣2 + ⋯ + 𝑐1𝑟 𝑣𝑟 = 𝑐11 𝑣11 , 𝑣12 , … , 𝑣1𝑛 + ⋯ + 𝑐1𝑟 (𝑣𝑟1 , 𝑣𝑟2 , … , 𝑣𝑟𝑛 )
𝑢2 = 𝑐21 𝑣1 + 𝑐22 𝑣2 + ⋯ + 𝑐2𝑟 𝑣𝑟 = 𝑐21 𝑣11 , 𝑣12 , … , 𝑣1𝑛 + ⋯ + 𝑐2𝑟 (𝑣𝑟1 , 𝑣𝑟2 , … , 𝑣𝑟𝑛 )
⋮ ⋮
𝑢𝑚 = 𝑐𝑚1 𝑣1 + 𝑐𝑚2 𝑣2 + ⋯ + 𝑐𝑚𝑟 𝑣𝑟 = 𝑐𝑚1 𝑣11 , 𝑣12 , … , 𝑣1𝑛 + ⋯ + 𝑐𝑚𝑟 (𝑣𝑟1 , 𝑣𝑟2 , … , 𝑣𝑟𝑛 )
In matrix form we can write
𝑢11 … 𝑢1𝑘 … 𝑢1𝑛
⋮
𝑢𝑗1 … 𝑢𝑗𝑘 … 𝑢𝑗𝑛 =
⋮
𝑢𝑚1 … 𝑢𝑚𝑘 … 𝑢𝑚𝑛 𝑚×𝑛

𝑐11 𝑣11 + 𝑐12 𝑣21 + ⋯ + 𝑐1𝑟 𝑣𝑟1 ⋯ 𝑐11 𝑣1𝑘 + 𝑐12 𝑣2𝑘 + ⋯ + 𝑐1𝑟 𝑣𝑟𝑘 ⋯ 𝑐11 𝑣1𝑛 + 𝑐12 𝑣2𝑛 + ⋯ + 𝑐1𝑟 𝑣𝑟𝑛
⋮
𝑐𝑗1 𝑣11 + 𝑐𝑗2 𝑣21 + ⋯ + 𝑐𝑗𝑟 𝑣𝑟1 ⋯ 𝑐𝑗1 𝑣1𝑘 + 𝑐𝑗2 𝑣2𝑘 + ⋯ + 𝑐𝑗𝑟 𝑣𝑟𝑘 ⋯ 𝑐𝑗1 𝑣1𝑛 + 𝑐𝑗2 𝑣2𝑛 + ⋯ + 𝑐𝑗𝑟 𝑣𝑟𝑛
⋮
𝑐𝑚1 𝑣11 + 𝑐𝑚2 𝑣21 + ⋯ + 𝑐𝑚𝑟 𝑣𝑟1 ⋯ 𝑐𝑚1 𝑣1𝑘 + 𝑐𝑚2 𝑣2𝑘 + ⋯ + 𝑐𝑚𝑟 𝑣𝑟𝑘 ⋯ 𝑐𝑚1 𝑣1𝑛 + 𝑐𝑚2 𝑣2𝑛 + ⋯ + 𝑐𝑚𝑟 𝑣𝑟𝑛 𝑚×𝑛

21
Comparing the 𝑘 𝑡ℎ column in both the sides, we have
𝑢1𝑘 𝑐11 𝑣1𝑘 + 𝑐12 𝑣2𝑘 + ⋯ + 𝑐1𝑟 𝑣𝑟𝑘 𝑐11 𝑐12 𝑐1𝑟
𝑢2𝑘 𝑐21 𝑣1𝑘 + 𝑐22 𝑣2𝑘 + ⋯ + 𝑐2𝑟 𝑣𝑟𝑘 𝑐 𝑐 𝑐
= = 𝑣1𝑘 21 + 𝑣2𝑘 22 + ⋯ +𝑣𝑟𝑘 2𝑟
⋮ ⋮ ⋮ ⋮ ⋮
𝑢𝑚𝑘 𝑐𝑚1 𝑣1𝑘 + 𝑐𝑚2 𝑣2𝑘 + ⋯ + 𝑐𝑚𝑟 𝑣𝑟𝑘 𝑐𝑚1 𝑐𝑚2 𝑐𝑚𝑟
for 𝑘 = 1,2, … , 𝑛.
Now, vector on the left is the 𝑘 𝑡ℎ column of 𝐴. We see that each of these 𝑛 columns of 𝐴
is a linear combination of the same 𝑟 columns of the right. Hence 𝐴 cannot have more
than 𝑟 LI columns. Therefore, column rank 𝐴 ≤ 𝑟 or 𝑝 ≤ 𝑟.
Also 𝑝 = column rank 𝐴 = row rank 𝐴𝑇 ≥ column rank 𝐴𝑇 = row rank 𝐴 = 𝑟.
Hence 𝑝 = 𝑟.
The following can be easily observed about a matrix 𝐴𝑚×𝑛 :
 Rank 𝐴 ≤ min{𝑚, 𝑛}
 Rank 𝐴 = Rank 𝐴𝑇
 Rank 𝐴 + 𝐵 ≤ Rank 𝐴 + Rank 𝐵 , provided 𝐴 + 𝐵 is possible
 Rank 𝐴𝐵 ≤ min{Rank 𝐴 , Rank 𝐵 }, provided 𝐴𝐵 is possible
 𝐴𝑛×𝑛 is non-singular if and only if Rank 𝐴 = 𝑛
 𝑝 vectors with 𝑛(< 𝑝) components are always LD
22
 Minors and rank of a matrix
Let 𝐴 be any 𝑚 × 𝑛 matrix. The determinant of any square submatrix of 𝐴 is
called a minor of 𝐴. The order of the submatrix is called order of the minor.
 𝐴𝑚×𝑛 has minors of order ≤ min{𝑚, 𝑛}
 A number 𝑟 is said to be rank of 𝐴, if 𝐴 has at least one non-zero minor of
order 𝑟 and every minor of order 𝑟 + 1 or more is zero
1 2 −1 0
Examples 7:1. The minors of the matrix 𝐴 = 0 1 3 1 are of order ≤ 3.
2 5 1 2
1 2 −1
Let us take the minor 0 1 3 which is equal to 1 1 − 15 + 2 6 + 1 = 0.
2 5 1
2 −1 0
Next 1 3 1 = 2 6 − 1 + 1 2 − 5 = 10 − 3 = 7 ≠ 0. Hence rank 𝐴 = 3.
5 1 2
2 1 −1
2. The matrix 𝐴 = 1 4 2 has rank 2 since det 𝐴 = 0 = minor of order 3, and
3 5 1
2 1
= 7.
1 4
23
 Normal form and rank of a matrix
The normal form of a matrix 𝐴𝑚×𝑛 is a matrix 𝐸 (obtained by elementary row and
𝐼𝑟 ⋮ 𝑂𝑟×(𝑛−𝑟)
column operations) of the form 𝐸 = ⋯ ⋮ ⋯ , where 𝐼𝑟 is the
𝑂(𝑚−𝑟)×𝑟 ⋮ 𝑂(𝑚−𝑟)×(𝑛−𝑟)
𝑚×𝑛
′
identity matrix of order 𝑟, and 𝑂 𝑠 are the zero matrices. It is trivial to observe that
rank 𝐴 = rank 𝐸 = 𝑟.
Example 8: Reduce the following matrix into normal form and find its rank.
2 −4 3 1 0 𝑅 → 𝑅 − 2𝑅 0 0 1 9 −4 𝑅 → 𝑅 + 2𝑅 0 0 1 9 −4
𝐴 = 1 −2 1 −4 2 1 1 2
1 −2 1 −4 2 2 2 4
1 0 1 20 −4
0 1 −1 3 1 𝑅4 → 𝑅4 − 4𝑅2 0 1 −1 3 1 𝑅3 → 𝑅3 − 𝑅4 0 0 −1 −9 4
4 −7 4 −4 5 ~ 0 1 0 12 −3 ~ 0 1 0 12 −3

𝑅1 → 𝑅1 + 𝑅3 0 0 0 0 0 𝐶 → 𝐶 − 11𝐶 − 12𝐶 − 9𝐶 0 0 0 0 0
1 0 0 11 0 4 4 1 2 3
1 0 0 0 0
𝑅2 → 𝑅2 + 𝑅3 0 0 −1 −9 4 𝐶5 → 𝐶5 + 4𝐶3 + 3𝐶2 0 0 −1 0 0
~ 0 1 0 12 −3 ~ 0 1 0 0 0

𝑅1 ↔ 𝑅2 ↔ 𝑅4 1 0 0 0 0 𝐼3 ⋮ 𝑂3×2
0 1 0 0 0 = ⋯ ⋯ ⋯ . Hence rank 𝐴 = 3.
𝑅3 → −𝑅3 0 0 1 0 0
~ 𝑂1×3 ⋮ 𝑂1×2
0 0 0 0 0

24
 Inverse of matrix
If an 𝑛 × 𝑛 matrix 𝐴 can be transformed into 𝐼𝑛 , then 𝐴 is non-singular, that is, invertible.
The sequence of elementary operations that transforms 𝐴𝑛×𝑛 into the identity matrix 𝐼𝑛 ,
also transforms 𝐼𝑛 into 𝐴−1
𝑛×𝑛 .
If we can convert 𝐴 = 𝐼𝐴 into 𝐼 = 𝐵𝐴, then 𝐴 is invertible and 𝐴−1 = 𝐵.
2 0 1
Example 9: To find inverse of 𝐴 = −2 3 4 .
−5 5 6
1 1
2 0 1 1 0 0 1 1 0 0 0
𝑅 → 𝑅 2 2
Let us start with −2 3 4 = 0 1 0 𝐴 1 2 1
−2 3 4 = 0 1 0 𝐴
−5 5 6 0 0 1 ~
−5 5 6 0 0 1
1 1
1 0
1 1
0 0 1 0 0 0
𝑅2 → 𝑅2 /3 2 2
𝑅2 → 𝑅2 + 2𝑅1 2 2 5 1 1
𝑅3 → 𝑅3 + 5𝑅1 0 3 5 = 1 1 0 𝐴𝑅3 → 𝑅3 − 5𝑅2 0 1 = 0 𝐴
3 3 3
~ 17 5
0 5 0 1 ~ 1 5 5
2 2 0 0 − 1
6 6 3

𝑅1 → 𝑅1 − 3𝑅3 1 0 0 −2 5 −3 −2 5 −3
𝑅2 → 𝑅2 − 10𝑅3 0 1 0 = −8 17 −1 −1
−10 𝐴. Hence 𝐴 exists, and A = −8 17 −10 .
𝑅3 → 6𝑅3 0 0 1 5 −10 6 5 −10 6
~

25
 1 −1 −2
Example 10: The matrix 𝐴 = 2 4 5 has no inverse since it cannot be reduced
6 0 −3
to an identity matrix.
1 −1 −2 1 −1 −2 1 −1 −2
2 4 5 ~ 0 6 9 ~ 0 6 9 .
6 0 −3 0 6 9 0 0 0
Also rank 𝐴 = 2 < 3.

𝑎 1 1 1
H. W.: By using elementary operations show that the matrix 𝐴 = 1 𝑏 2 −1 is
2 2 3 1
−1 1 1 1
1
invertible if and only if 𝑎 ≠ −1 and 𝑏 ≠ 2.

26
 RANK AND SYSTEM OF LINEAR EQUATIONS

A system of 𝑚 linear equations in 𝑛 unknowns/variables has the general form

𝑎11 𝑥1 + 𝑎12 𝑥2 + ⋯ + 𝑎1𝑛 𝑥𝑛 = 𝑏1 𝑎11 𝑎12 ⋯ 𝑎1𝑛 𝑥1 𝑏1
𝑎21 𝑥1 + 𝑎22 𝑥2 + ⋯ + 𝑎2𝑛 𝑥𝑛 = 𝑏2 𝑎21 𝑎22 ⋯ 𝑎2𝑛 𝑥2 𝑏
or ⋯ ⋯ ⋯ ⋯ = 2 or 𝐴𝑋 = 𝐵. (1)
………………………………….. ⋮ ⋮
𝑎𝑚1 𝑥1 + 𝑎𝑚2 𝑥2 + ⋯ + 𝑎𝑚𝑛 𝑥𝑛 = 𝑏𝑚 𝑎𝑚1 𝑎𝑚2 ⋯ 𝑎𝑚𝑛 𝑥𝑛 𝑏𝑚

• If all the constants 𝑏1 , 𝑏2 , … , 𝑏𝑛 are zero, that is, 𝐵 is zero column, then system (1) reduces to
𝐴𝑋 = 𝑂 and called the homogeneous system. Otherwise it is called a non-homogeneous
system.
• The column 𝑋 = 𝑥1 , 𝑥2 , … , 𝑥𝑛 𝑇 is called a solution of the system (1), if the numbers
𝑥1 , 𝑥2 , … , 𝑥𝑛 satisfy all the equations of the system.
• A linear system of equations is called consistent, if it has at least one solution, and
inconsistent, if it has no solution.
• If a linear system is consistent, it may has a unique solution or infinitely many solutions.
• The homogeneous system 𝐴𝑋 = 𝑂 is always consistence with zero (trivial) solution.

𝑎11 𝑎12 ⋯ 𝑎1𝑛 ⋮ 𝑏1

𝑎21 𝑎22 ⋯ 𝑎2𝑛 ⋮ 𝑏2
• The matrix 𝐴 ⋮ 𝐵 = ⋯ ⋯ ⋯ ⋯⋮ is called the augmented matrix of the system.
⋮
𝑎𝑚1 𝑎𝑚2 ⋯ 𝑎𝑚𝑛 ⋮ 𝑏𝑚

27
 Examples:
Let us recall our previous Examples 2, 3, 4.
• In Example 2, the system has infinitely many solutions and the augmented matrix
2 −1 3 2 ⋮0 1 0 0 17/3 ⋮0
is 1 4 0 −1⋮0 ~ 0 1 0 −5/3 ⋮0 .
2 6 −1 5 ⋮0 0 0 1 −11/3⋮0
Note that rank 𝐴 = rank 𝐴 ⋮ 𝐵 = 3 < number of unknowns 4. Further in the solution
we have one (4 − 3) free variable (𝑥4 ) on which other three variables depend.
• In Example 3, the system has a unique solution and the augmented matrix is
1 2 −1⋮−1 1 2 −1 ⋮ −1
A ⋮ 𝐵 = 2 6 1 ⋮ 7 ~ 0 1 3/2⋮9/2 .
5 7 −4⋮ 9 0 0 1 ⋮ 5
Note that rank 𝐴 = rank 𝐴 ⋮ 𝐵 = 3 = number of unknowns.
• In Example 4, the system has no solution and the augmented matrix is
1 −2 1 2 ⋮1 1 −2 1 2⋮ 1
1 1 −1 1 ⋮2 ~ 0 1 −2/3 −1/3⋮1/3 .
1 7 −5 −1⋮3 0 0 0 0 ⋮ −1
Note that rank 𝐴 = 2 ≠ rank 𝐴 ⋮ 𝐵 = 3.

28
Rank vs Consistency of Linear System
 A system of linear equations 𝐴𝑋 = 𝐵 is consistence if and only if the rank of the coefficient
matrix 𝐴 = rank of the augmented matrix 𝐴 ⋮ 𝐵 .
 If a linear system 𝐴𝑋 = 𝐵 with 𝑚 equations and 𝑛 unknowns is consistent with rank 𝐴 =
rank 𝐴 ⋮ 𝐵 = 𝑟, then solution of the system contains 𝑛 − 𝑟 parameters or free variables.
Let us discuss some more examples.
Example 11: Discuss the solutions of the system:
𝑥1 + 2𝑥2 + 4𝑥3 + 𝑥4 − 𝑥5 = 1
2𝑥1 + 4𝑥2 + 8𝑥3 + 3𝑥4 − 4𝑥5 = 2
𝑥1 +3𝑥2 + 7𝑥3 + 3𝑥5 = −2
1 2 4 1 −1 ⋮ 1
The augmented matrix of the system is 𝐴 ⋮ 𝐵 = 2 4 8 3 −4 ⋮ 2
1 3 7 0 3 ⋮ −2
𝑅2 → 𝑅2 − 2𝑅1 1 2 4 1 −1 ⋮ 1 1 2 4 1 −1 ⋮ 1
𝑅2 ↔ 𝑅3
𝑅3 → 𝑅3 − 𝑅1 0 0 0 1 −2 ⋮ 0 0 1 3 −1 4 ⋮ −3
~
~ 0 1 3 −1 4 ⋮ −3 0 0 0 1 −2 ⋮ 0
𝑅2 → 𝑅2 + 𝑅3 1 2 4 0 1 ⋮ 1 1 0 −2 0 −3 ⋮ 7
𝑅1 → 𝑅1 − 2𝑅2
𝑅1 → 𝑅1 − 𝑅3 0 1 3 0 2 ⋮ −3 0 1 3 0 2 ⋮ −3
~
~ 0 0 0 1 −2 ⋮ 0 0 0 0 1 −2 ⋮ 0

29
The rank 𝐴 ⋮ 𝐵 = rank 𝐴 = 3 < 5(the number of unknowns). Therefore, the system is
consistent and has infinitely many solutions.
Thus the given system has the same solution as the system:
𝑥1 − 2𝑥3 − 3𝑥5 = 7 𝑥1 − 2𝑥3 = 7 + 3𝑥5
𝑥2 + 3𝑥3 + 2𝑥5 = −3 which gives 𝑥2 + 3𝑥3 = −3 − 2𝑥5
𝑥4 − 2𝑥5 = 0 𝑥4 = 2𝑥5 .
If we assign an arbitrary value 𝑡 to the unknown 𝑥5 , then 𝑥5 = 𝑡, 𝑥4 = 2𝑡 and
𝑥1 = 7 + 3𝑡 + 2𝑥3 = 7 + 3𝑡 + 2𝑠;
𝑥2 = −3 − 2𝑡 − 3𝑥3 = −3 − 2𝑡 − 3𝑠,
where 𝑠 is an arbitrary value assigned to the variable 𝑥3 . Finally, our solution has the
form 𝑥1 , 𝑥2 , 𝑥3 , 𝑥4 , 𝑥5 = (7 + 3𝑡 + 2𝑠, −3 − 2𝑡 − 3𝑠, 𝑠, 𝑡, 2𝑡), where 𝑠 and 𝑡 are arbitrary
real numbers.
NOTE:
• A variable is a basic variable, if it corresponds to a pivot column. Otherwise, the
variable is known as a free variable.
• In Example 11, we have THREE basic variables, namely, 𝑥1 , 𝑥2 , 𝑥4 ; and TWO free
variables, namely, 𝑥3 , 𝑥5 .

30
Example 12: Investigate the values of 𝜆 and 𝜇 for which the system
2𝑥 + 3𝑦 + 5𝑧 = 9
7𝑥 + 3𝑦 − 2𝑧 = 8
2𝑥 + 3𝑦 + 𝜆𝑧 = 𝜇
has (i) no solution (ii) unique solution (iii) infinitely many solutions.
The augmented matrix of the system is

2 3 5 ⋮ 9 1 3/2 5/2 ⋮ 9/2

𝐴⋮𝐵 = 7 3 −2 ⋮ 8 ~ 0 −15/2 −39/2 ⋮ 47/2
2 3 𝜆 ⋮ 𝜇 0 0 𝜆−5 ⋮ 𝜇−9
1
(Performing the operations 𝑅1 → 2 𝑅1 , 𝑅2 → 𝑅2 − 7𝑅1 , 𝑅3 → 𝑅3 − 2𝑅1 )
(i) The system has no solution if rank 𝐴 ⋮ 𝐵 ≠ rank(A). In this case rank of none of
the matrices can be less than 2. Therefore, rank 𝐴 ⋮ 𝐵 = 3 and rank A = 2.
Hence, 𝜆 = 5 and 𝜇 ≠ 9.
(ii) The system has a unique solution if rank 𝐴 ⋮ 𝐵 = rank A = 3, i.e., 𝜆 ≠ 5 and 𝜇 is
any real number.
(iii) The system has infinitely many solutions if rank 𝐴 ⋮ 𝐵 = rank A < 3, 𝜆 = 5 and
𝜇 = 9.

31
H.W.:
𝑥2𝑧3 2
8 𝑦 𝑧
3
4 𝑥 𝑦
1. Solve = 𝑒 ; 𝑥 = 𝑒 ; 𝑧4 = 1 for 𝑥, 𝑦, 𝑧 > 0.
𝑦

Quiz:
1. If a homogeneous system of linear equations with 𝑚 equations in 𝑛 unknowns has a
non-trivial solution, then what is the relation between 𝑚 and 𝑛.
2. Let 𝐴 be a 4 × 4 non-invertible matrix. Can we find a 4 × 4 matrix 𝐵 ≠ 𝑂 such that
𝐴𝐵 = 𝑂. Can you generalize your result for general 𝑛 × 𝑛 matrices.

32
SOME SPECIAL TYPES OF REAL MATRICES
(Symmetric and Skew-symmetric)

• An 𝑛 × 𝑛 real matrix 𝐴 = [𝑎𝑖𝑗 ] is said to be symmetric if 𝑎𝑖𝑗 = 𝑎𝑗𝑖 for each 𝑖 and 𝑗,
that is, 𝐴 is symmetric, if 𝐴𝑇 = 𝐴.
• An 𝑛 × 𝑛 real matrix 𝐴 = [𝑎𝑖𝑗 ] is said to be skew-symmetric if 𝑎𝑖𝑗 = −𝑎𝑗𝑖 for each 𝑖
and 𝑗, that is, 𝐴 is skew-symmetric, if 𝐴𝑇 = −𝐴.
• In case of skew-symmetric matrices we note that 𝑎𝑖𝑖 = −𝑎𝑖𝑖 for all 𝑖 or 𝑎𝑖𝑖 =0 for all 𝑖,
that is, the diagonal entries of a skew-symmetric matrix are all 0.
1 −1 5 0 −1 5
Examples: 𝐴 = −1 2 3 is symmetric and 𝐵 = 1 0 2 is skew-symmetric.
5 3 3 −5 −2 0
PROPERTIES:
1. For any square matrix 𝐴, 𝐴 + 𝐴𝑇 is symmetric and 𝐴 − 𝐴𝑇 is skew-symmetric.
2. Sum of two symmetric (or skew-symmetric) matrices is also symmetric (or skew-
symmetric).
3. If 𝐴 is symmetric/skew-symmetric, then 𝑘𝐴, where 𝑘 is any non-zero real number, is
symmetric/skew-symmetric.
4. If 𝐴 and 𝐵 are symmetric matrices of the same order, then 𝐴𝐵 is symmetric if and
only if 𝐴𝐵 = 𝐵𝐴. Also 𝐴𝑝 is symmetric for any positive integer 𝑝.

33
5. The matrix 𝐵𝑇 𝐴𝐵 is symmetric or skew-symmetric according as 𝐴 is symmetric or
skew-symmetric.
6. If 𝐴 is skew-symmetric, then 𝐴2𝑘 is symmetric and 𝐴2𝑘−1 is skew-symmetric for any
positive integer 𝑘.
7. If 𝐴 and 𝐵 are symmetric matrices of the same order, then 𝐴𝐵 + 𝐵𝐴 is symmetric
and 𝐴𝐵 − 𝐵𝐴 is skew-symmetric.
8. Any square matrix can be expressed a the sum of a symmetric matrix and a skew-
symmetric matrix.

ORTHOGONAL MATRICES:
• The vectors 𝑢1 = (𝑎11 , 𝑎12 , … , 𝑎1𝑛 ) and 𝑢2 = (𝑎21 , 𝑎22 , … , 𝑎2𝑛 ) in ℝ𝑛 are said to be
orthogonal, if 𝑢1 . 𝑢2 = 𝑢1 𝑢2𝑇 = 𝑎11 𝑎21 + 𝑎12 𝑎22 + ⋯ + 𝑎1𝑛 𝑎2𝑛 = 0.
• A set {𝑢1 , 𝑢2 , … , 𝑢𝑚 } of vectors in ℝ𝑛 is called orthogonal, if 𝑢𝑖 . 𝑢𝑗 = 0 for 𝑖 ≠ 𝑗.
0, if 𝑖 ≠ 𝑗
• A set {𝑢1 , 𝑢2 , … , 𝑢𝑚 } of vectors in ℝ𝑛 is called orthonormal, if 𝑢𝑖 . 𝑢𝑗 = ቊ
1, if 𝑖 = 𝑗
• The length of a vector 𝑢 = (𝑥1 , 𝑥2 , … , 𝑥𝑛 ) is denoted by 𝑢 = 𝑥12 + 𝑥22 + ⋯ + 𝑥𝑛2

34
A real square matrix 𝐴 is called orthogonal, if 𝐴−1 = 𝐴𝑇 or 𝐴𝐴𝑇 = 𝐼 = 𝐴𝑇 𝐴.
1/9 8/9 −4/9
For example 𝐴 = 4/9 −4/9 −7/9 is orthogonal, since
8/9 1/9 4/9

1/9 8/9 −4/9 1/9 4/9 8/9 1 0 0

𝐴𝐴𝑇 = 4/9 −4/9 −7/9 8/9 −4/9 1/9 = 0 1 0.
8/9 1/9 4/9 −4/9 −7/9 4/9 0 0 1

𝑢1 𝑎11 𝑎12 ⋯ 𝑎1𝑛

𝑢 𝑎 𝑎22 ⋯ 𝑎2𝑛
In general suppose that 𝐴 = 2 = 21 be an orthogonal matrix.
⋮ ⋮ ⋮ ⋯ ⋮
𝑢𝑛 𝑎𝑛1 𝑎𝑛2 ⋯ 𝑎𝑛𝑛

𝑢1 𝑢12 𝑢1 . 𝑢2 ⋯ 𝑢1 . 𝑢𝑛 1 0 ⋯ 0
𝑢2 2 ⋯ 𝑢2 . 𝑢𝑛
𝑇
Then, 𝐴𝐴 = 𝑢1 , 𝑢2 , … , 𝑢𝑛 = 𝑢2 . 𝑢1 𝑢 2 = 0 1 ⋯ 0 .
⋮ ⋮ ⋮ ⋯ ⋮ ⋮ ⋮ ⋯ ⋮
𝑢𝑛 𝑢𝑛 . 𝑢1 𝑢𝑛 . 𝑢2 ⋯ 𝑢𝑛2 0 0 ⋯ 1
⇒ 𝑢12 = 𝑢22 = ⋯ = 𝑢𝑛2 = 1 and 𝑢𝑖 . 𝑢𝑗 = 0 for 𝑖 ≠ 𝑗.

35
⇒ rows of 𝐴 are orthonormal. Similarly, the columns of 𝐴 are orthonormal.

Thus the following statements are equivalent

• 𝐴 is orthogonal
• Rows of 𝐴 are orthonormal
• Columns of 𝐴 are orthonormal

H.W.: Find an orthogonal matrix whose first two rows are the multiples of
𝑢1 = 1,1,1 , 𝑢2 = 0, −1,1 .
1 1 1
3 3 3
1 1
Ans. 0 − 2 2
2 −1 −1
6 6 6

36
Normal Matrices: A real square matrix 𝐴 is said to be normal, if 𝐴𝐴𝑇 = 𝐴𝑇 𝐴.
• Every symmetric or orthogonal or skew-symmetric matrix is normal.
6 −3 45 0
• Example is normal as 𝐴𝐴𝑇 = = 𝐴𝑇 𝐴.
3 6 0 45
COMPLEX MATRICES:
 A matrix 𝐴 = [𝑎𝑖𝑗 ] is said to be a complex matrix, if entries of the matrix, that is, 𝑎𝑖𝑗
are complex numbers.
• The conjugate of 𝐴, denoted by 𝐴,ҧ is defined as 𝐴ҧ = 𝑎𝑖𝑗 .
• The conjugate transpose of 𝐴, denoted by 𝐴∗ , is defined by 𝐴∗ = 𝐴ҧ 𝑇
= (𝐴𝑇 ).
2 − 8𝑖 −6𝑖
2 + 8𝑖 5 − 3𝑖 4 − 7𝑖 ∗
• For example, if 𝐴 = , then 𝐴 = 5 + 3𝑖 1 + 4𝑖 .
6𝑖 1 − 4𝑖 3 + 2𝑖
4 + 7𝑖 3 − 2𝑖
Properties:
• 𝐴∗ ∗ = 𝐴
• 𝐴 + 𝐵 ∗ = 𝐴∗ + 𝐵∗
• 𝑘𝐴 ∗ = 𝑘𝐴ത ∗
• 𝐴𝐵 ∗ = 𝐵 ∗ 𝐴∗
37
Some Special Complex Matrices:
• A complex square matrix 𝐴 is said to be Hermitian or Skew-Hermitian according as
𝐴∗ = 𝐴ҧ 𝑇 = 𝐴 or 𝐴∗ = −𝐴.
• If 𝐴 = 𝑎𝑖𝑗 is Hermitian, then 𝑎𝑖𝑗 = 𝑎𝑗𝑖 for each 𝑖, 𝑗. In this case 𝑎𝑖𝑖 = 𝑎𝑖𝑖
⇒ 𝑎𝑖𝑖 (diagonal elements) are real.
• If 𝐴 = 𝑎𝑖𝑗 is skew-Hermitian, then 𝑎𝑖𝑗 = −𝑎𝑗𝑖 for each 𝑖, 𝑗 and then 𝑎𝑖𝑖 = −𝑎𝑖𝑖
⇒ real(𝑎𝑖𝑖 ) = 0 (diagonal elements are all purely imaginary).
• A complex matrix 𝐴 is said to be unitary if 𝐴∗ = 𝐴−1 or 𝐴∗ 𝐴 = 𝐴𝐴∗ = 𝐼.
• A complex matrix 𝐴 is said to be normal if 𝐴∗ 𝐴 = 𝐴𝐴∗ .

1 1 1
If 𝐴 is unitary, then 𝐴−1 = 𝐴
⇒ 𝐴∗ = 𝐴−1 = 𝐴
and 𝐴∗ 𝐴 = 𝐴∗ 𝐴 = 𝐴
𝐴 =1
or 𝐴𝑇 𝐴 = 𝐴 𝐴 = 1 ⇒ det 𝐴 2 = 1 ⇒ det 𝐴 = ±1.

Hence the determinant of a unitary matrix has absolute value 1.

38
 EXAMPLES

3 1 − 2𝑖 4 + 7𝑖
• 𝐴 = 1 + 2𝑖 −4 −2𝑖 is Hermitian,
4 − 7𝑖 2𝑖 5

1 −𝑖 −1 + 𝑖
1
• 𝐵= 𝑖 1 1 + 𝑖 is unitary
2
1+𝑖 −1 + 𝑖 0

2 + 3𝑖 1
• 𝐶= is normal
𝑖 1 + 2𝑖

𝑖 3 2−𝑖
• 𝐷 = −3 −𝑖 2𝑖 is skew- Hermitian.
−2 − 𝑖 2𝑖 −2𝑖

39
 EIGENVALUES AND EIGENVECTORS
Consider the system of linear equation 𝐴𝑋 = 𝜆𝑋 or 𝐴 − 𝜆𝐼 𝑋 = 𝑂, ...(1)
where 𝐴 is a square matrix.
• The system (1) always has a zero solution.
• The values of 𝜆 for which the system (1) has a non-zero (non-trivial) solution are
called eigenvalues of the matrix 𝐴 and the corresponding non-zero solution vectors
𝑋 are called the eigenvectors of 𝐴.
• The system (1) has a non-zero solution if and only if 𝐴 − 𝜆𝐼 = 0 …(2)
• The equation (2) is a polynomial equation in 𝜆 and called the characteristic equation
of 𝐴, and the polynomial 𝐴 − 𝜆𝐼 is called the characteristic polynomial of 𝐴.
1 2 1 2 1 0 1−𝜆 2
For example, if 𝐴 = , then 𝐴 − 𝜆𝐼2 = −𝜆 = .
3 2 3 2 0 1 3 2−𝜆
1−𝜆 2
The characteristic equation of 𝐴 is 𝐴 − 𝜆𝐼2 = =0
3 2−𝜆
or 𝜆2 − 3𝜆 + 2 − 6 = 0 or 𝜆2 − 3𝜆 − 4 = 0 or 𝜆 − 4 𝜆 + 1 = 0 ⇒ 𝜆 = 4, −1.
Hence the eigenvalues of 𝐴 are 4 and −1. The eigenvector corresponding to 𝜆 = 4 is
−3 2 𝑥1 0 −3 2 𝑥1 0
given by 𝐴 − 4𝐼 𝑋1 = 𝑂 ⇒ = or = or 3𝑥1 − 2𝑥2 = 0
3 −2 𝑥2 0 0 0 𝑥2 0
2
⇒ 𝑥1 = 2, 𝑥2 = 3 ⇒ 𝑋1 = (There can be many more).
3
40
 2 2 𝑥1 0
For 𝜆 = −1, the eigenvector is given by 𝐴 − 𝐼 𝑋 = 𝑂 or =
3 3 𝑥2 0
1
⇒ 𝑥1 = 1, 𝑥2 = −1, i.e., 𝑋2 = .
−1
• For given 𝑖, (𝜆𝑖 , 𝑋𝑖 ) is called an eigen pair of 𝐴.
Property: If 𝑋 is an eigenvector of 𝐴 corresponding to the eigenvalue 𝜆, then 𝑘𝑋(𝑘 ≠ 0)
is also an eigenvector of 𝐴 corresponding to 𝜆.
We have 𝐴𝑋 = 𝜆𝑋 ⇒ 𝐴 𝑘𝑋 = 𝑘 𝐴𝑋 = 𝑘 𝜆𝑋 = 𝜆(𝑘𝑋).
Hence 𝑘𝑋 is also an eigenvector of 𝐴 corresponding to 𝜆.
2 1 0
Example 13: Find the eigenvalues and eigenvectors of the matrix 𝐴 = 0 2 1 .
0 1 2
The eigenvalues are: 𝜆1 = 1, 𝜆2 = 2, 𝜆3 = 3 and the corresponding eigenvectors are:
1 1 1
𝑋1 = −1 , 𝑋2 = 0 , 𝑋3 = 1 .
1 0 1
Note that the eigenvectors are LI, and the eigenvalues are distinct.

41
 SOME MORE PROPERTIES
If 𝜆 is an eigenvalues of 𝐴, then
• 𝛼𝜆 is an eigenvalue of α𝐴, where 𝛼 is scalar.
• 𝜆 − 𝑘 is an eigenvalue of 𝐴 − 𝑘𝐼.
1
• is an eigenvalue of 𝐴−1 , 𝜆 ≠ 0.
𝜆
• 𝜆𝑚 is an eigenvalue of 𝐴𝑚 , 𝑚 is some positive integer.
• 𝜆 is an eigenvalue of 𝐴𝑇 .
Proof: (iii) Let 𝑋 be the eigenvector of 𝐴 corresponding to 𝜆. Then 𝐴𝑋 = 𝜆𝑋
1
⇒𝐴−1 𝐴𝑋 = 𝐴−1 𝜆𝑋 ⇒ 𝐼𝑋 = 𝜆 𝐴−1 𝑋 ⇒ 𝐴−1 𝑋 = 𝜆 𝑋
1
⇒ is an eigenvalue of 𝐴−1 .
𝜆
(iv). We can write
𝐴𝑚 𝑋 = 𝐴𝑚−1 𝐴𝑋 = 𝐴𝑚−1 𝜆𝑋 = 𝜆𝐴𝑚−1 𝑋 = 𝜆𝐴𝑚−2 𝐴𝑋 = 𝜆𝐴𝑚−2 𝜆𝑋
= 𝜆2 𝐴𝑚−2 𝑋 = 𝜆𝑚 𝑋
⇒ 𝜆𝑚 is an eigenvalue of 𝐴𝑚 .
(v). 𝐴 − 𝜆𝐼 = 0 ⇒ (𝐴 − 𝜆𝐼)𝑇 = 0 ⇒ 𝐴𝑇 − 𝜆𝐼 = 0 ⇒ 𝜆 is an eigenvalue of 𝐴𝑇 .
(Rest are HW)

42
Eigenvalues and Eigenvectors of some special
matrices:
1. The eigenvalues of a symmetric or Hermitian matrix are real.
Let 𝜆 be an eigenvalue of a Hermitian matrix 𝐴 corresponding eigenvector 𝑋, that is
𝑇
𝐴𝑋 = 𝜆𝑋 ⇒𝐴ҧ𝑋ത = 𝜆ҧ 𝑋ത ⇒ 𝐴ҧ𝑋ത 𝑇 = 𝜆ҧ 𝑋ത
⇒ 𝑋ത 𝑇 𝐴𝑇ҧ = 𝜆ҧ 𝑋ത 𝑇 ⇒ 𝑋ത 𝑇 𝐴∗ 𝑋 = 𝜆ҧ 𝑋ത 𝑇 𝑋 ⇒ 𝑋ത 𝑇 𝐴𝑋 − 𝜆ҧ 𝑋ത 𝑇 𝑋 = 0
⇒ 𝑋ത 𝑇 𝜆𝑋 − 𝜆ҧ 𝑋ത 𝑇 𝑋 = 0 ⇒ 𝜆𝑋ത 𝑇 𝑋 − 𝜆ҧ 𝑋ത 𝑇 𝑋 = 0 ⇒ 𝜆 − 𝜆ҧ 𝑋ത 𝑇 𝑋 = 0.
𝑥1
Since 𝑋ത 𝑇 𝑋 = 𝑥ҧ1 , 𝑥ҧ2 , 𝑥ҧ3 , … 𝑥ҧ𝑛 ⋮ = 𝑥1 2
+ 𝑥2 2
+ 𝑥3 2
+ ⋯ + 𝑥𝑛 2
≠ 0, 𝜆 = 𝜆ҧ ⇒ 𝜆 is
𝑥𝑛
real.
2. The eigenvectors corresponding to distinct eigenvalues of a symmetric matrix are
orthogonal.
Let 𝐴𝑋1 = 𝜆1 𝑋1 and 𝐴𝑋2 = 𝜆2 𝑋2 ⇒ 𝑋1 𝑇 𝐴𝑋2 = 𝜆2 𝑋1 𝑇 𝑋2 … (1)
and 𝐴𝑋1 𝑇 = (𝜆1 𝑋1 )𝑇 ⇒ 𝑋1 𝑇 𝐴 = 𝜆1 𝑋1 𝑇 ⇒ 𝑋1 𝑇 𝐴𝑋2 = 𝜆1 𝑋1 𝑇 𝑋2 … (2)
From (1) and (2),
𝜆2 − 𝜆1 𝑋1 𝑇 𝑋2 = 0 ⇒ 𝑋1 𝑇 𝑋2 = 0 as the eigenvalues are distinct.
⇒ 𝑋1 , 𝑋2 are orthogonal.

43
 0 −1 0
Example 14: Take the symmetric matrix 𝐴 = −1 −1 1 .
0 1 0
−𝜆 −1 0
The characteristic equation of 𝐴 is 𝐴 − 𝜆𝐼 = −1 −1 − 𝜆 1 = 0
0 1 −𝜆
−𝜆 0 −𝜆 1 0 1
⇒ −1 − 𝜆 −1 − 𝜆 1 − 𝜆 = 0 ⇒ 𝜆 1 + 𝜆 1 + 𝜆 𝜆 − 1 = 0
0 1 −𝜆 0 1 −𝜆
2 2
⇒ 𝜆 1 −𝜆 − 𝜆 − 𝜆 + 1 + (1 + 𝜆) = 0 ⇒ 𝜆 −𝜆 − 𝜆 + 2 = 0 ⇒ 𝜆 𝜆 − 1 𝜆 + 2 = 0
⇒ 𝜆1 = 0, 𝜆2 = 1, 𝜆3 = −2.
The eigenvector 𝑋1 corresponding to 𝜆1 = 0 is given by 𝐴 − 0𝐼 𝑋 = 0, that is,
0 −1 0 𝑥 0 0 −1 0 𝑥 0
−1 −1 1 𝑦 = 0 or −1 0 1 𝑦 = 0 .
0 1 0 𝑧 0 0 0 0 𝑧 0

⇒ 𝑦 = 0, 𝑥 − 𝑧 = 0 ⇒ 𝑥 = 𝑧, 𝑥 is arbitrary. So let 𝑥 = 1 ⇒ 𝑋1 = 1, 0,1 𝑇 .

The eigenvector 𝑋2 corresponding to 𝜆2 = 1 is given by 𝐴 − 𝐼 𝑋 = 0, that is,

44
−1 −1 0 𝑥 0 −1 −1 0 𝑥 0
−1 −2 1 𝑦 = 0 or 0 0 0 𝑦 = 0
0 1 −1 𝑧 0 0 1 −1 𝑧 0
⇒ 𝑥 + 𝑦 = 0, 𝑦 − 𝑧 = 0 ⇒ 𝑦 = 𝑧, 𝑥 = −𝑧.
−1
Let 𝑧 = 1, 𝑥 = −1, 𝑦 = 1 so that 𝑋2 = 1 .
1
1
Similarly, we can obtain 𝑋3 = 2 .
−1
−1 1
𝑇
Now 𝑋1 . 𝑋2 = 𝑋1 𝑋2 = 1 0 1 1 = 0 and 𝑋2 . 𝑋3 = 𝑋2𝑇 𝑋3 = −1 1 1 2 =𝑂
1 −1
1
𝑇
𝑋1 . 𝑋3 = 𝑋1 𝑋2 = 1 0 1 2 = 0 ⇒ 𝑋1 , 𝑋2, 𝑋3 are orthogonal.
−1

(Check whether 𝑋1 , 𝑋2, 𝑋3 are LI)

45
Sum and product of eigenvalues:
𝑎11 𝑎12
Suppose 𝐴 = 𝑎 𝑎22 . Then, 𝐴 − 𝜆𝐼 = 0 (1)
21
𝑎11 − 𝜆 𝑎12
⇒ =0
𝑎21 𝑎22 − 𝜆
⇒ 𝑎11 − 𝜆 𝑎22 − 𝜆 − 𝑎21 𝑎12 = 0
⇒ 𝜆2 − 𝑎11 + 𝑎22 𝜆 − 𝑎21 𝑎12 + 𝑎11 𝑎22 = 0
⇒ 𝜆2 − 𝑎11 + 𝑎22 𝜆 + 𝑎11 𝑎22 − 𝑎21 𝑎12 = 0
⇒ 𝜆2 − traceA 𝜆 + det 𝐴 = 0 (2)
If 𝜆1 and 𝜆2 are the roots of (1) or (2), then 𝜆1 + 𝜆2 = traceA and 𝜆1 𝜆2 = det 𝐴 .
𝑎11 𝑎12 𝑎13
In case of 𝐴= 𝑎21 𝑎22 𝑎23 , 𝐴 − 𝜆𝐼 = 0
𝑎31 𝑎32 𝑎33
⇒ 𝜆3 − trace 𝐴 𝜆2 + 𝐴11 + 𝐴22 + 𝐴33 𝜆 − det 𝐴 = 0
⇒ 𝜆1 +𝜆2 + 𝜆3 = trace 𝐴,
𝜆1 𝜆2 + 𝜆2 𝜆3 + 𝜆3 𝜆1 = 𝐴11 + 𝐴22 + 𝐴33 = sum of cofactors of diagonal elements,
𝜆1 𝜆2 𝜆3 = det 𝐴.

46
 CAYLEY HAMILTON THEOREM

Polynomials of matrices: An expression of the form

𝑓 𝐴 = 𝑎𝑛 𝐴𝑛 + 𝑎𝑛−1 𝐴𝑛−1 + ⋯ + 𝑎1 𝐴 + 𝑎0 𝐼,
where 𝐴 is a square matrix and 𝐼 is the identity matrix of the same order, is called the
polynomial of 𝐴. In particular, we say that 𝐴 is a root of 𝑓 𝑡 = 0, if 𝑓 𝐴 = 𝑂.
1 2 7 10
For example, let 𝐴 = so that 𝐴2 = .
3 4 15 22
Let 𝑓 𝑡 = 2𝑡 2 − 3𝑡 + 5 and 𝑔 𝑡 = 𝑡 2 − 5𝑡 − 2 be two polynomials. Then
7 10 1 2 1 0 16 14
𝑓 𝐴 = 2𝐴2 − 3𝐴 + 5𝐼 = 2 −3 +5 =
15 22 3 4 0 1 21 37
7 10 1 2 2 0 0 0
𝑔 𝐴 = 𝐴2 − 5𝐴 − 2𝐼 = −5 −2 = =𝑂 (1)
15 22 3 4 0 2 0 0
⇒ 𝐴 is a root of the equation 𝑡 2 − 5𝑡 − 2 = 0 but not of 2𝑡 2 − 3𝑡 + 5 = 0.
1
Also we note that 𝐴2 − 5𝐴 − 2𝐼 = 𝑂 ⇒ 𝐴 − 5𝐼 − 2𝐴−1 = 𝑂 ⇒ 𝐴−1 = 2 𝐴 − 5𝐼 .
1−𝜆 2
Also the characteristic equation of 𝐴 is 𝐴 − 𝜆𝐼 = =0
3 4−𝜆
⇒ 𝜆2 − 5𝜆 − 2 = 0 (2)
From (1), we note that 𝐴 satisfies (2), that is, 𝐴 satisfies its characteristic equation.

47
Cayley-Hamilton Theorem:
Every square matrix satisfies its characteristic equation

Proof: Let 𝑓 𝑥 = 𝑥𝐼 − 𝐴 = 𝑥 𝑛 + 𝑎𝑛−1 𝑥 𝑛−1 + ⋯ + 𝑎1 𝑥 + 𝑎0 be the characteristic

polynomial of 𝐴𝑛×𝑛 . Let 𝐵 𝑥 be the adjoint of the matrix 𝑥𝐼 − 𝐴. We know that the
elements of 𝐵 𝑥 are the cofactors of 𝑥𝐼 − 𝐴, hence are the polynomials of degree at
most 𝑛 − 1 in 𝑥. Therefore, we can write
𝐵 𝑥 = 𝐵𝑛−1 𝑥 𝑛−1 + ⋯ + 𝐵1 𝑥 + 𝐵0 , * (1)
where 𝐵𝑖′ 𝑠 are 𝑛-square matrices. By the property of adjoint of a matrix, we have
𝑥𝐼 − 𝐴 𝐵 𝑥 = 𝑥𝐼 − 𝐴 𝐼
Or 𝑥𝐼 − 𝐴 𝐵𝑛−1 𝑥 𝑛−1 + ⋯ + 𝐵1 𝑥 + 𝐵0 = 𝑥 𝑛 + 𝑎𝑛−1 𝑥 𝑛−1 + ⋯ + 𝑎1 𝑥 + 𝑎0 𝐼.
Equating the powers of 𝑥 in both sides, we have
𝐵𝑛−1 = 𝐼, 𝐵𝑛−2 − 𝐴𝐵𝑛−1 = 𝑎𝑛−1 𝐼, …, 𝐵0 − 𝐴𝐵1 = 𝑎1 𝐼, −𝐴𝐵0 = 𝑎0 𝐼.
Multiplying the above equations by 𝐴𝑛 , 𝐴𝑛−1 ,…, 𝐴, 𝐼, respectively, we get
𝐴𝑛 𝐵𝑛−1 = 𝐴𝑛 𝑎11 𝑥 2 + 𝑏11 𝑥 + 𝑐11 𝑎12 𝑥 2 + 𝑏12 𝑥 + 𝑐12 𝑎13 𝑥 2 + 𝑏13 𝑥 + 𝑐13
𝑛−1 𝑛 𝑛−1 *𝐵 𝑥 = 𝑎21 𝑥 2 + 𝑏21 𝑥 + 𝑐21 𝑎22 𝑥 2 + 𝑏22 𝑥 + 𝑐22 𝑎23 𝑥 2 + 𝑏23 𝑥 + 𝑐23
𝐴 𝐵𝑛−2 − 𝐴 𝐵𝑛−1 = 𝑎𝑛−1 𝐴 𝑎31 𝑥 2 + 𝑏31 𝑥 + 𝑐31 𝑎32 𝑥 2 + 𝑏32 𝑥 + 𝑐32 𝑎33 𝑥 2 + 𝑏33 𝑥 + 𝑐33
⋮ 𝑎11 𝑎12 𝑎13 𝑐11 𝑐12 𝑐13
𝑏11 𝑏12 𝑏13
2
𝐴𝐵0 − 𝐴 𝐵1 = 𝑎1 𝐴 𝑎
= 21 𝑎 22 𝑎 2
23 𝑥 + 𝑏 21 𝑏22 𝑏23 𝑥 + 21 𝑐22 𝑐23
𝑐
𝑎31 𝑎31 𝑎33 𝑏31 𝑏32 𝑏33 𝑐31 𝑐31 𝑐33
−𝐴𝐵0 = 𝑎0 𝐼.
Adding the above equations, 𝑂 = 𝐴𝑛 + 𝑎𝑛−1 𝐴𝑛−1 + ⋯ + 𝑎1 𝐴 + 𝑎0 𝐼 = 𝑓(𝐴).
Hence 𝐴 satisfies its characteristic equation 𝑓 𝑥 = 0.
48
 1 2 0
Example 15: Verify the Cayley-Hamilton theorem for 𝐴 = −1 1 2 and find 𝐴−1.
1 2 1
The characteristic equation of 𝐴 is given by 𝐴 − 𝜆𝐼3 = 0, that is,
1−𝜆 2 0 1−𝜆 2 0
−1 1 − 𝜆 2 = 0 or 0 3−𝜆 3−𝜆 =0
1 2 1−𝜆 1 2 1−𝜆
1−𝜆 2 0 1−𝜆 2 0
Or (3 − 𝜆) 0 1 1 = 0 or (3 − 𝜆) 0 1 1 =0
1 2 1−𝜆 1 0 −1 − 𝜆
Or (3 − 𝜆) 1 − 𝜆 −1 − 𝜆 − 0 − 2 0 − 1 + 0 = 0 or (3 − 𝜆) 𝜆2 + 1 = 0
Or 𝜆3 − 3𝜆2 + 𝜆 − 3 = 0. … (1)
1 2 0 1 2 0 −1 4 4
Now 𝐴2 = −1 1 2 −1 1 2 = 0 3 4
1 2 1 1 2 1 0 6 5
1 2 0 −4 4 4 −1 10 12
and 𝐴3 = −1 1 2 0 3 4 = 1 11 10 .
1 2 1 0 6 5 −1 16 17

49
 −1 10 12 −1 4 4 1 2 0 3 0 0 0 0 0
3 2
𝐴 − 3𝐴 + 𝐴 − 3𝐼3 = 1 11 10 − 3 0 3 4 + −1 1 2 − 0 3 0 = 0 0 0 (𝑂3×3 )
−1 16 17 0 6 5 1 2 1 0 0 3 0 0 0
⇒ 𝐴 satisfies its characteristic equation.
Also 𝐴3 − 3𝐴2 + 𝐴 − 3𝐼3 = 𝑂 ⇒ 𝐴2 − 3𝐴 + 𝐼 − 3𝐴−1 = 𝑂
−3 −2 4
1 1
⇒ 𝐴−1 = 𝐴2 − 3𝐴 + 𝐼 = 3 1 −2 .
3 3
−3 0 3
(Check why existence of 𝐴−1 is sure?)

Minimal Polynomial: The minimal polynomial 𝑔(𝑡) for a square matrix 𝐴 is the monic
polynomial of least positive degree for which 𝑔 𝐴 = 𝑂.
• Recall the division algorithm for polynomials: 𝑓 𝑥 = 𝑔 𝑥 𝑞 𝑥 + 𝑟(𝑥)
• The minimal polynomial of 𝐴 is a divisor of the characteristic polynomial of 𝐴
2 0 0
Examples: The minimal polynomial of 0 2 0 is 2 − 𝜆 as 2𝐼 − 𝐴 = 𝑂.
0 0 2
2 1 1
The minimal polynomial of 0 2 1 is 2 − 𝜆 3 since 2𝐼 − 𝐴 2 and 2𝐼 − 𝐴 are not zero.
0 0 2
50
 Computation of 𝐴𝒎
We shall discuss it by examples.
−2 4 −2 − 𝜆 4
1. Let us consider 𝐴 = whose characteristic equation is =0
−1 3 −1 3−𝜆
Or 𝜆2 − 𝜆 − 2 = 0 ⇒ 𝜆 = −1, 2.
⇒ 𝐴2 − 𝐴 − 2𝐼 = 0 ⇒ 𝐴2 = 𝐴 + 2𝐼 ⇒ 𝐴3 = 𝐴2 + 2𝐴 = 𝐴 + 2𝐼 + 2𝐴 = 3𝐴 + 2𝐼
⇒ 𝐴4 = 3𝐴2 + 2𝐴 = 3 𝐴 + 2𝐼 + 2𝐴 = 5𝐴 + 6𝐼 ⇒ 𝐴5 = 5𝐴2 + 6𝐴 = 5 𝐴 + 2𝐼 + 6𝐼
⇒ 𝐴5 = 11𝐴 + 10𝐼 and 𝐴6 = 11 𝐴 + 2𝐼 + 10𝐴 = 21𝐴 + 22𝐼.
−2 4 1 0 −20 84
𝐴6 = 21 + 22 = .
−1 3 0 1 −21 85
We note that every power of 𝐴 obtained above is a linear combination of 𝐴 and 𝐼. So,
let us assume 𝐴𝑚 = 𝑐0 𝐼 + 𝑐1 𝐴, where 𝜆𝑚 = 𝑐0 + 𝑐1 𝜆. …(1)
⇒(−1)𝑚 = 𝑐0 + 𝑐1 −1 = 𝑐0 − 𝑐1 and 2𝑚 = 𝑐0 + 𝑐1 2 ⇒ 2𝑚 − (−1)𝑚 = 3𝑐1
1 1 1
⇒𝑐1 = 2𝑚 − (−1)𝑚 and 𝑐0 = 2𝑚 − (−1)𝑚 + (−1)𝑚 = 2𝑚 + 2(−1)𝑚 .
3 3 3
Hence, from (1), we have
1 4
𝑐 − 2𝑐1 4𝑐1 (−2)𝑚 +4(−1)𝑚 2𝑚 − (−1)𝑚
𝐴𝑚 = 0 = 3
−1 1
3
, 𝑚 = 1, 2, 3, …
−𝑐1 3𝑐1 + 𝑐0 2𝑚 − (−1)𝑚 2𝑚+2 − (−1)𝑚
3 3
51
For a matrix 𝐴𝑛×𝑛 having characteristic equation of degree 𝑛, we can write
𝐴𝑚 = 𝑐0 𝐼 + 𝑐1 𝐴 + 𝑐2 𝐴2 + ⋯ + 𝑐𝑛−1 𝐴𝑛−1 , where 𝜆𝑚 = 𝑐0 + 𝑐1 𝜆 + 𝑐2 𝜆2 + ⋯ . . +𝑐𝑛−1 𝜆𝑛−1 .
1 1 −2
H.W: Compute 𝐴𝑚 for 𝐴 = −1 2 1 .
0 1 −1
Hint: First obtain −𝜆3 + 2𝜆2 + 𝜆 − 2 = 0, 𝜆 = −1,1,2 (roots are distinct)
Assume 𝐴𝑚 = 𝑐0 𝐼 + 𝑐1 𝐴 + 𝑐2 𝐴2 and 𝜆𝑚 = 𝑐0 + 𝑐1 𝜆 + 𝑐2 𝜆2 , which gives
(−1)𝑚 = 𝑐0 − 𝑐1 + 𝑐2
1 = 𝑐0 + 𝑐1 + 𝑐2
2𝑚 = 𝑐0 + 2𝑐1 + 4𝑐2 .
Solve these equations for 𝑐0 , 𝑐1 , 𝑐2 and use them in 𝑐0 𝐼 + 𝑐1 𝐴 + 𝑐2 𝐴2 to get
1 1 𝑚 1
9 − 2𝑚+1 − −1 𝑚
[2 − −1 𝑚]
−9 + 2𝑚+1 + 7 −1 𝑚
6 3 6
𝐴𝑚 = 1 − 2𝑚 2𝑚 2𝑚 − 1 .
1 1 𝑚 1
6
3 − 2𝑚+1 − −1 𝑚
3
[2 − −1 𝑚
] 6
−3 + 2𝑚+1 + 7 −1 𝑚

52
Note: In case of repeated roots, one can differentiate the relation
𝜆𝑚 = 𝑐0 + 𝑐1 𝜆 + 𝑐2 𝜆2 + ⋯ +𝑐𝑛−1 𝜆𝑛−1
as many times as required to get the coefficients.
4 1 3
H. W.: If 𝐴 = 0 2 0 , then find 𝐴14 − 3𝐴13 .
−4 1 −4
HINT: The characteristic polynomial is 2 − 𝑥 𝑥 − 2 𝑥 + 2 = − 𝑥 − 2 2 𝑥+2 .
By division algorithm 𝑥 14 − 3𝑥 13 = 𝑥 − 2 2
𝑥 + 2 𝑞 𝑥 + 𝑎0 + 𝑎1 𝑥 + 𝑎2 𝑥 2 …(1)
2
The term 𝑥 − 2 𝑥 + 2 𝑞 𝑥 is zero for 𝑥 = 2, −2. So we use only
𝑥 14 − 3𝑥 13 = 𝑎0 + 𝑎1 𝑥 + 𝑎2 𝑥 2
so that 214 − 3.213 = 𝑎0 + 2𝑎1 + 4𝑎2
and 214 + 3.213 = 𝑎0 − 2𝑎1 + 4𝑎2 .
Differentiate (1) and put 𝑥 = 2, so that 14.213 − 3.13.212 = 𝑎1 + 4𝑎2
Solving the above equations for 𝑎0 , 𝑎1 , 𝑎2 , we get
𝐴14 − 3𝐴13 = 214 + 215 𝐼3 − 3.212 𝐴 − 213 𝐴2 .
53
 DIAGONALIZATION
• Two square matrices 𝐴 and 𝐵 of the same order are said to be similar, if there is an
invertible matrix 𝑃 such that 𝐴 = 𝑃𝐵𝑃−1 or 𝐴𝑃 = 𝑃𝐵 or 𝑃−1 𝐴𝑃 = 𝐵.
𝑎11 𝑎12 𝑏11 𝑏12
• A special notation: Suppose 𝐴 = 𝑎 and 𝐵 = = 𝑋1 𝑋2 .
21 𝑎22 𝑏21 𝑏22
𝑎11 𝑎12 𝑏11 𝑏12 𝑎11 𝑏11 + 𝑎12 𝑏21 𝑎11 𝑏12 + 𝑎12 𝑏22
𝐴𝐵 = 𝑎 𝑎22 𝑏21 𝑏22 = = 𝐴𝑋1 𝐴𝑋2
21 𝑎21 𝑏11 + 𝑎21 𝑏22 𝑎21 𝑏12 + 𝑎22 𝑏22
• In general for 𝑛 × 𝑛 matrices 𝐴𝐵 = 𝐴𝑋1 𝐴𝑋2 𝐴𝑋3 … 𝐴𝑋𝑛 , where 𝑋1 , 𝑋2 , 𝑋3 , … , 𝑋𝑛 are
columns of 𝐵.
Diagonalization of a matrix: A square matrix 𝐴𝑛×𝑛 is said to be diagonalizable, if there
exists an 𝑛 × 𝑛 non-singular matrix 𝑃 such that 𝑃−1 𝐴𝑃 = 𝐷 is a diagonal matrix. Then
we say that 𝑃 diagonalizes 𝐴.
For example, consider a 3 × 3 diagonalizable matrix 𝐴. Then there exists a 3 × 3 non-
𝑑11 0 0
singular matrix 𝑃 such that 𝑃−1 𝐴𝑃 = 𝐷 or 𝐴𝑃 = 𝑃𝐷, where 𝐷 = 0 𝑑22 0 .
0 0 𝑑33
If 𝑃1 , 𝑃2 , 𝑃3 denote the columns of 𝑃, then 𝐴𝑃 = 𝑃𝐷.
⇒ 𝐴𝑃1 , 𝐴𝑃2 , 𝐴𝑃3 = 𝑑11 𝑃1 , 𝑑22 𝑃2 , 𝑑33 𝑃3 ⇒ 𝐴𝑃1 = 𝑑11 𝑃1 , 𝐴𝑃2 = 𝑑22 𝑃2 , 𝐴𝑃3 = 𝑑33 𝑃3
⇒ 𝑑11 , 𝑑22 and 𝑑33 are the eigenvalues of 𝐴 associated with the eigenvectors 𝑃1 , 𝑃2 and
𝑃3 , respectively.
54
• Further these eigenvectors are linearly independent since 𝑃 was assumed to be non-
singular.
• If 𝐴 is diagonalizable, then the columns of the diagonalizing matrix 𝑃 are linearly
independent.
We also have the following sufficient condition:
Theorem: If an 𝑛 × 𝑛 matrix 𝐴 has 𝑛 linearly independent eigenvectors 𝑃1 , 𝑃2 , … , 𝑃𝑛 ,
then 𝐴 is diagonalizable.
Proof: We can realize the proof by proving the result for a 3 × 3 matrix 𝐴. Let 𝑃1 , 𝑃2 and
𝑃3 be the 3 linearly independent eigenvectors of 𝐴 corresponding to the eigenvalues 𝜆1 ,
𝜆2 and 𝜆3 . Then 𝐴𝑃1 = 𝜆1 𝑃1 , 𝐴𝑃2 = 𝜆2 𝑃2 , 𝐴𝑃3 = 𝜆3 𝑃3 and 𝑃 = 𝑃1 𝑃2 𝑃3 .
𝜆1 0 0
Now, 𝐴𝑃 = 𝐴𝑃1 𝐴𝑃2 𝐴𝑃3 = 𝜆1 𝑃1 𝜆2 𝑃2 𝜆3 𝑃3 = 𝑃1 𝑃2 𝑃3 0 𝜆2 0 = 𝑃𝐷
0 0 𝜆3
⇒ 𝑃−1 𝐴𝑃 = 𝐷.
• Note that the diagonal entries of 𝐷 are the eigenvalues of 𝐴 in an order that
corresponds to the order of the eigenvectors placed in 𝑃.
• Thus we have the following criterion: An 𝐴𝑛×𝑛 is diagonalizable if and only if 𝐴 has 𝑛
linearly independent eigenvectors. (HOME ASSIGNEMNT)
55
Theorem: If 𝐴𝑛×𝑛 has 𝑛 distinct eigenvalues, then eigenvectors of 𝐴 are linearly
independent, that is, 𝐴 is diagonalizable.
Proof: Let 𝑋1 , 𝑋2 , … , 𝑋𝑛 be the eigenvectors of 𝐴 corresponding to the different
eigenvalues 𝜆1 , 𝜆2 , … , 𝜆𝑛 . Then 𝐴𝑋𝑘 = 𝜆𝑘 𝑋𝑘 𝑘 = 1,2,3, … , 𝑛 . Now consider the linear
combination 𝑐1 𝑋1 + 𝑐2 𝑋2 + ⋯ + 𝑐𝑛 𝑋𝑛 = 𝑂. … (1)
Multiply (1) by 𝐴 − 𝜆2 𝐼 𝐴 − 𝜆3 𝐼 … 𝐴 − 𝜆𝑛 𝐼 and use the fact that 𝐴 − 𝜆𝑘 𝐼 𝑋𝑘 = 𝑂, we
get 𝑐1 𝐴 − 𝜆2 𝐼 𝐴 − 𝜆3 𝐼 … 𝐴 − 𝜆𝑛 𝐼 𝑋1 + ⋯ +𝑐𝑛 𝐴 − 𝜆2 𝐼 𝐴 − 𝜆3 𝐼 … 𝐴 − 𝜆𝑛 𝐼 𝑋𝑛 = 𝑂
⇒ 𝑐1 𝜆1 − 𝜆2 𝜆1 − 𝜆3 𝜆1 − 𝜆𝑛 𝑋1 = 𝑂 ⇒ 𝑐1 = 0.
Similarly, multiplying (1) by 𝐴 − 𝜆1 𝐼 𝐴 − 𝜆2 𝐼 … 𝐴 − 𝜆𝑖−1 𝐼 𝐴 − 𝜆𝑖+1 𝐼 … 𝐴 − 𝜆𝑛 𝐼 , we
can show that 𝑐𝑖 = 0 for each 𝑖. Hence 𝑋1 , 𝑋2 , … , 𝑋𝑛 are LI.
1 2 1
Example 16: Show that that the matrix 𝐴 = 6 −1 0 is diagonalizble. Also find
−1 −2 −1
−1
an invertible matrix 𝑃 such that 𝑃 𝐴𝑃 is a diagonal matrix.
The characteristic equation of 𝐴 is 𝐴 − 𝜆𝐼 = 0, that is,
1−𝜆 2 1 −𝜆 0 −𝜆
6 −1 − 𝜆 0 = 0 or 6 −1 − 𝜆 0 = 0.
−1 −2 −1 − 𝜆 −1 −2 −1 − 𝜆
56
 −1 0 −1 −1 0 −1
⇒ 𝜆 6 −1 − 𝜆 0 = 0 or 6 −1 − 𝜆 0 =0
−1 −2 −1 − 𝜆 0 −2 −𝜆
⇒ 𝜆 −1 1 + 𝜆 𝜆 + 12 ⇒ 𝜆 −𝜆2 − 𝜆 + 12 = 0 ⇒ 𝜆 𝜆2 + 𝜆 − 12 = 0
⇒ 𝜆 𝜆 + 4 𝜆 − 3 = 0 ⇒ 𝜆1 = 0, 𝜆2 = −4, 𝜆3 = 3. Hence 𝐴 is diagonalizable.
1 −1 2
The corresponding eigenvector are 𝑋1 = 6 , 𝑋2 = 2 , 𝑋3 = 3 (H.W.)
−13 1 −2
Here since the eigenvalues are distinct, the eigenvectors are linearly independent.
Hence 𝑃 = 𝑋1 𝑋2 𝑋3 is non-singular. That is,
−1 −1
0
1 −1 2 12 12
−9 2 3
𝑃= 6 2 3 , where 𝑃−1 = 28 7 28
(H.W.)
−13 1 −2 8 1 2
21 7 21
−1 −1
0
1 −1 2 1 2 1 12 12 0 0 0
−9 2 3
And so 𝑃−1 𝐴𝑃 = 6 2 3 6 −1 0 28 7 28
= 0 −4 0 = 𝐷.
−13 1 −2 −1 −2 −1 8 1 2 0 0 3
21 7 21
57
The condition that 𝐴 has 𝑛 distinct eigenvalues is sufficient for diagonalization, but not
necessary. In other words, if the matrix 𝐴 does not have 𝑛 distinct eigenvalues, it may or
may not be diagonalizable.
3 4
For example, 𝐴 = has the characteristic equation 3 − 𝜆 7 − 𝜆 + 4 = 0
−1 7
or 𝜆2 − 10𝜆 + 25 = 0 ⇒ 𝜆 = 5, 5 (not distinct).
−2 4 𝑥 0
The corresponding eigenvector is = or 𝑥 − 2𝑦 = 0, 𝑥 = 2, 𝑦 = 1
−1 2 𝑦 0
2
⇒ 𝑋1 = . There are infinitely may eigenvectors but not LI.
1
0 1 0
Example 17: Check the matrix 𝐴 = 1 0 0 for diagonalization.
0 0 1
−𝜆 1 0
The characteristic equation of 𝐴 is 𝐴 − 𝜆𝐼 = 1 −𝜆 0 =0
0 0 1−𝜆
Or −𝜆 𝜆2 − 𝜆 − 1 1 − 𝜆 = 0 or 𝜆 + 1 𝜆2 − 1 = 0 or 𝜆 + 1 𝜆 − 1 𝜆 − 1 = 0
⇒ 𝜆1 = −1, 𝜆2 = 1, 𝜆3 = 1 (Two repeated root).

58
The eigenvector corresponding to 𝜆1 = −1 is given by 𝐴 + 𝐼 𝑋1 = 0
1 1 0 𝑥 0 1
Or 1 1 0 𝑦 = 0 or 𝑥 + 𝑦 = 0, 2𝑧 = 0 ⇒ 𝑧 = 0, 𝑥 = 1, 𝑦 = −1 ⇒ 𝑋1 = −1 .
0 0 2 𝑧 0 0
The eigenvector corresponding to 𝜆2 = 1 is given by (𝐴 − 𝐼)𝑋2 = 0
−1 1 0 𝑥 0 −1 1 0 𝑥 0
Or 1 −1 0 𝑦 = 0 or 0 0 0 𝑦 = 0 or −𝑥 + 𝑦 = 0, 𝑧 is arbitrary.
0 0 0 𝑧 0 0 0 0 𝑧 0
Let us choose two LI eigenvectors corresponding to the eigenvalue 𝜆2 = 1.
1 0
Take 𝑧 = 0, 𝑥 = 1 = 𝑦 or 𝑧 = 1, 𝑥 = 0 = 𝑦 ⇒ 𝑋2 = 1 or 0 .
0 1
To check linear independence of 𝑋1 , 𝑋2 and 𝑋3 , let
1 1 0 1 1 0 1 1 0
𝑃 = −1 1 0 ~ 0 2 0 ~ 0 1 0 ⇒ columns of 𝑃 are linearly independent.
0 0 1 0 0 1 0 0 1
−1 0 0
⇒ 𝑃 is non- singular. Hence 𝐴 is diagonalizable and 𝐷 = 0 1 0 so that 𝑃−1 𝐴𝑃 = 𝐷
0 0 1
59
NOTE:
• Recall from slides 28-30 that for 𝐴𝑋 = 𝑂, where 𝑋 consists of 𝑛 unknowns and 𝐴
consists of 𝑟 linearly independent rows (rank of 𝐴 = 𝑟), the number of linearly
independent solutions is 𝑛 − 𝑟.
• In Example 17, corresponding to 𝜆2 = 1, we have rank of 𝐴 − 𝐼 is 1. Therefore, there
are 2(= 3 − 1) LI solutions of 𝐴 − 𝐼 𝑋 = 𝑂.
4 𝛼 −1
H.W.: If the matrix 2 5 𝛽 is diagonalizable and has the eigenvalues 3, 3, 𝛿 ≠ 3,
1 1 𝛾
then find the values of 𝛼, 𝛽, 𝛾, 𝛿.
TWO MULTIPLICTIES ASSOCIATED WITH EIGENVALUES:
• The algebraic multiplicity of an eigenvalue is the number of times it appears as a root
of the characteristic equation.
• The geometric multiplicity of an eigenvalue is the number of LI eigenvectors
associated with the eigenvalue.
• In Example 16, every eigenvalue has geometric multiplicity and algebraic multiplicity
1. However, in Example 17, the eigenvalue 𝜆2 = 1 has the geometric multiplicity as
well as algebraic multiplicity 2.
60
 Powers of diagonalizable matrix

If 𝐴 is diagonalizable, then there exists a non-singular matrix 𝑃 such that 𝑃−1 𝐴𝑃 = 𝐷

⇒ 𝑃−1 𝐴𝑃 𝑃−1 𝐴𝑃 = 𝐷𝐷 ⇒ 𝑃−1 𝐴𝑃𝑃−1 𝐴𝑃 = 𝐷2 ⇒𝑃−1 𝐴2 𝑃 = 𝐷2 .

Continuing in this way, 𝑃−1 𝐴𝑚 𝑃 = 𝐷𝑚 ⇒ 𝐴𝑚 = 𝑃𝐷𝑚 𝑃−1 .

𝜆1 0 0 𝜆1 0 0 𝜆1 2 0 0
Also 𝐷2 = 0 𝜆2 0 0 𝜆2 0 = 0 𝜆2 2 0
0 0 𝜆3 0 0 𝜆3 0 0 𝜆3 2

𝜆1 𝑚 0 0
so that 𝐷𝑚 = 0 𝜆2 𝑚 0 .
0 0 𝜆3 𝑚
Thus if 𝑓(𝑡) is any function in powers of 𝑡, then 𝑓 𝐴 = 𝑃𝑓(𝐷)𝑃−1 .

For example, if 𝐴 is diagonalizable and 𝑃−1 𝐴𝑃 = 𝐷, then 𝑒 𝐴 = 𝑃𝑒 𝐷 𝑃−1, where

𝐷
𝐷2 𝐷3
𝑒 =𝐼+𝐷+ + +⋯
2! 3!

61