MA110:

Lecture 10

Saurav Bhaumik
Department of Mathematics
IIT Bombay

Spring 2025

Saurav Bhaumik, IIT Bombay MA110: Lecture 10

Matrix of a Linear Transformation
General Case: Let V and W be vector subspaces of dimension
n and m, and let T : V → W be a linear map. Suppose
E := (v1 , . . . , vn ) and F := (w1 , . . . , wm ) are ordered bases of
V and W respectively. Then the matrix of T with respect to
E and F is the m × n matrix A = (ajk ) determined by
Xm
T (vk ) = ajk wj for k = 1, . . . , n.
j=1

This matrix is denoted by MEF (T ) . Note that the kth column

of this matrix corresponds to the coefficients of T (vk ) when
expressed as a linear combination of w1 , . . . , wm . Also, for
v = α1 v1 + · · · + αn vn ∈ V , w = β1 w1 + · · · + βm wm ∈ W ,
   
α1 β1
 ..   .. 
T (v) = w ⇐⇒ A  .  =  .  .
αn βm
Saurav Bhaumik, IIT Bombay MA110: Lecture 10
Composites of Linear Transformation
Suppose U is another vector space of dimension p and
D = (u1 , . . . , up ) is an ordered basis of U. If S : U → V is a
linear map, then we can form the composite of S and T , i.e.,
T ◦S :U →W
Now if B = (bkℓ ) is the n × p matrix of S w.r.t. D and E , i.e.,
B = MD E (S), then for ℓ = 1, . . . , p,
n
X n
X
S(uℓ ) = bkℓ vk and hence (T ◦ S)(uℓ ) = bkℓ T (vk );
k=1 k=1

substituting for T (vk ) from the earlier equation, we see that

n m m n
!
X X X X
(T ◦ S)(uℓ ) = bkℓ ajk wj = ajk bkℓ wj
k=1 j=1 j=1 k=1

Thus we see that MD E D

F (T ◦ S) = AB = MF (T )ME (S).
Saurav Bhaumik, IIT Bombay MA110: Lecture 10
Remark: Let V be a subspace of Rn×1 , W be a subspace of
Rm×1 , and let T : V → W be a linear map. Two important
subspaces associated with T are as follows.
(i) N (T ) := {x ∈ V : T (x) = 0}, called the null space of T ,
(ii) I(T ) := {T (x) : x ∈ V }, called the image space of T .
We note that
a linear map T is one-one ⇐⇒ N (T ) = {0}, and
a linear map T is onto ⇐⇒ I(T ) = W .
Further, if V := Rn×1 , W := Rm×1 , and A ∈ Rm×n , then

N (TA ) = {x ∈ Rn×1 : Ax = 0} = N (A),

I(TA ) = {Ax : x ∈ Rn×1 } = C(A).
 
The last equality follows by noting that if A = c1  · · · cn ,
then Ax = x1 c1 + · · · + xn cn for x := x1 · · · xn ∈ Rn×1 .
Saurav Bhaumik, IIT Bombay MA110: Lecture 10
Example
 
1 −1
Let A := −1 2  ∈ R3×2 . Then TA : R2×1 → R3×1 .
0 1
 T  T
In fact, TA : x1 x2 7−→ x1 − x2 −x1 + 2x2 x2 for all
 T
x1 x2 ∈ R2×1 . Clearly, N (TA ) = {0}. Also,
T
y1 y2 y3 ∈ R3×1 : y1 + y2 − y3 = 0 .

I(TA ) =

To see this, note that (x1 − x2 ) + (−x1 + 2x2 ) − x2 = 0 for all

 T  T
x1 x2 ∈ R2×1 , and if y1 y2 y3 ∈ R3×1 satisfies
y1 + y2 − y3 = 0, then we may let x1 := y1 + y3 , x2 := y3 , so
that x1 − x2 = y1 , −x1 + 2x2 = y2 and x2 = y3 , that is,
 T  T
TA ( x1 x2 ) = y1 y2 y3 .
 T
Note: I(TA ) is a plane through the origin 0 0 0 in R3×1 .
Saurav Bhaumik, IIT Bombay MA110: Lecture 10
Complex Numbers
In our development of matrix theory, we have so far used real
numbers as scalars, and we have considered matrices whose
entries are real numbers. It turns out that all the concepts
extend easily when the scalars allowed to be complex
numbers instead of real numbers. We shall assume familiarity
with the basics of complex numbers and just recall a few
properties. The set of all complex numbers is denoted by C.
Let z ∈ C. Then z = x + iy for unique x, y ∈ R. We call x
the real part of z, and denote it by R(z). Also, y is called
the imaginary part of z, and it is denoted by I(z). The
complex number x − iy is called the conjugate of z = x + iy ,
and it is denoted by z. Let a, b, c, d ∈ R. The addition and
multiplication of complex numbers a + ib, c + id is defined by
(a + ib) + (c + id) = (a + c) + i(b + d)
(a + ib)(c + id) = (ac − bd) + i(ad + bc).
Saurav Bhaumik, IIT Bombay MA110: Lecture 10
Recall also that the absolute value of a complex number
z = x + iy ∈ C is defined by by
√ p
|z| := zz = x 2 + y 2 .
Note that the triangle inequality
|z1 + z2 | ≤ |z1 | + |z2 |
holds for all z1 , z2 ∈ C. In terms of the norm
p
∥(x, y )∥ := x 2 + y 2 of a vector (x, y ) ∈ R2 ,
this corresponds to the inequality
∥(x1 , y1 ) + (x2 , y2 )∥ ≤ ∥(x1 , y1 )∥ + ∥(x2 , y2 )∥
for all (x1 , y1 ), (x2 , y2 ) ∈ R2 . Also, note that
max {|R(z)|, |I(z)|} ≤ |z| ≤ |R(z)|+|I(z)| for all z ∈ C.
Saurav Bhaumik, IIT Bombay MA110: Lecture 10
We shall use complex numbers as scalars and consider
matrices whose entries are complex numbers.
The set of all m × n matrices with entries in C is denoted by
Cm×n . In particular, C1×n is the set of all row vectors of length
n, while Cm×1 is the set of all column vectors of length m.
For A := [a ] ∈ Cm×n , define A∗ := [a ]. Then A∗ ∈ Cn×m .
jk kj
It is called the conjugate transpose or the adjoint of A.
Note: (αA + βB)∗ = αA∗ +βB∗ for A, B ∈ Cm×n and
α, β ∈ C. In case m = n, then (AB)∗ = B∗ A∗ .
A square matrix A = [ajk ] is called Hermitian or
self-adjoint if A∗ = A, that is, if ajk = akj for all j, k.
A square matrix A = [ajk ] is called skew-Hermitian or
skew self-adjoint if ajk = −akj for all j, k.
Note: Every diagonal entry of a self-adjoint matrix is real since
ajj = ajj =⇒ ajj ∈ R for j = 1, . . . , n. On the other hand, the
real part of every diagonal entry of a skew self-adjoint matrix
is equal to 0 Saurav ajj =IIT−a
sinceBhaumik, jj =⇒MA110:
Bombay jj ) =
R(aLecture 10 0 for j = 1, . . . , n.
  T
Note: If x := x1 · · · xn ∈ Cn×1 is a column vector, then
x∗ = x1 · · · xn ∈ C1×n is a row vector, and
x∗ x = |x1 |2 + · · · + |xn |2 . It follows that x∗ x = 0 ⇐⇒ x = 0.
A matrix A ∈ Cm×n defines a linear transformation from Cn×1
to Cm×1 , and every linear transformation from Cn×1 to Cm×1
can be represented by a matrix A ∈ Cm×n (with respect to an
ordered basis for Cn×1 and an ordered basis for Cm×1 ).
Similarly, we can consider vector subspaces of Cn×1 , and the
concepts of linear dependence of vectors and of the span of a
subset carry over to Cn×1 . The Fundamental Theorem for
Linear Systems remains valid for matrices with complex entries.
Having thus completed our discussion of solution of a linear
system, we shall turn to solution of an ‘eigenvalue problem’
associated with a matrix. In this development, the role of
complex numbers will turn out to be important.
Saurav Bhaumik, IIT Bombay MA110: Lecture 10
Matrix Eigenvalue Problem

The German word ‘eigen’ means ‘belonging to itself’. The

eigenvalue problem for a matrix consists of finding a nonzero
vector which is sent to a scalar multiple of itself by the linear
transformation defined by the matrix.
Eigenvalue problems come up frequently in many engineering
branches, quantum mechanics, physical chemistry, biology, and
even in economics and psychology.
Please refer to Section 8.2 of Kreyszig’s book for applications
of eigenvalue problems to stretching of elastic membranes, to
vibrating mass-spring systems, to Markov processes and to
population control models.

Saurav Bhaumik, IIT Bombay MA110: Lecture 10

In the development that follows, we shall use either real
numbers or complex numbers as scalars. To facilitate a
general discussion which applies to both types of scalars, we
shall write K to mean either R or C. When we want to switch
to a special treatment valid for only the real scalars, or only
for the complex scalars, we shall specify K := R or K := C.
Definition
Let A be an n × n matrix with entries in K, that is, let
A ∈ Kn×n . A scalar λ ∈ K is called an eigenvalue of A if

Ax = λ x for some x ∈ Kn×1 with x ̸= 0.

Any nonzero vector x ∈ Kn×1 satisfying Ax = λ x is called an

eigenvector of A corresponding to the eigenvalue λ. Further,

{x ∈ Kn×1 : Ax = λ x} = N (A − λI).

is called the eigenspace of A associated with λ.

Saurav Bhaumik, IIT Bombay MA110: Lecture 10
How to find eigenvalues
Let A = [ajk ] ∈ Kn×n and let λ ∈ K. Clearly,
λ is an eigenvalue of A ⇐⇒ N (A − λI) ̸= {0}
⇐⇒ rank(A − λI) < n
⇐⇒ det(A − λI) = 0.
The last condition suggests that we consider the polynomial
 
a11 − t · · · a1n
pA (t) := det(A − t I) = det  ... .. ..  .

. . 
an1 · · · ann − t
This is called the characteristic polynomial of A. It is a
polynomial of degree n with coefficients in K and for λ ∈ K,
λ is an eigenvalue of A ⇐⇒ λ is a root of pA , i.e., pA (λ) = 0.
In particular, an n × n matrix with entries in K has at most n
eigenvalues in K.
Saurav Bhaumik, IIT Bombay MA110: Lecture 10
Algebraic and Geometric Multiplicities
Definition
Let A = [ajk ] ∈ Kn×n and let λ ∈ K be an eigenvalue of A.
The algebraic multiplicity of λ (as an eigenvalue of A)
is the order m of the root λ of pA (t), i.e., m is the largest
positive integer such that (t − λ)m divides pA (t).
The geometric multiplicity of λ (as an eigenvalue of A)
is the dimension of its eigenspace, i.e., dim N (A − λI).

Observe that if λ ∈ K be an eigenvalue of A, then

geometric multiplicity of λ = nullity(A−λI) = n−rank(A−λI).
This can be calculated by solving the homogeneous system
(A − λI)x = 0 using, for instance, Gaussian elimination. In
fact, GEM and back substitution will also give the basic
solutions, i.e., a set of eigenvectors of A which forms a basis
of the eigenspace of A associated to λ.
Saurav Bhaumik, IIT Bombay MA110: Lecture 10
Examples
(i) Let A = diag(a1 , . . . , an ), i.e., let A be a diagonal matrix
with diagonal entries a1 , . . . , an in that order. Clearly
pA (t) = (a1 − t)(a2 − t) · · · (an − t).
Thus the eigenvalues of A are precisely a1 , . . . , an . Note that
this can also be seen directly since Aek = ak ek for each
k = 1, . . . , n, where e1 , . . . , en are the basic column vectors in
Kn×1 . Observe that in this case the algebraic multiplicity of
each eigenvalue is equal to its geometric multiplicity. Indeed, if
λ ∈ {a1 , . . . , an }, then the algebraic multiplicity of λ equals
m := the number of i ∈ {1, . . . , n} such that ai = λ.
To find the geometric multiplicity of λ, consider A − λI and
note that this is a diagonal matrix with exactly m rows of zeros
and n − m nonzero rows. So rank(A − λI) = n − m and hence
geometric multiplicity of λ = nullity(A − λI) = m.
Saurav Bhaumik, IIT Bombay MA110: Lecture 10
Examples Contd.
On the other hand, let λ ∈ K and λ ∈ / {a1 , . . . , an }. Then
(A − λI)x = 0 ⇐⇒ (a1 − λ)x1 = · · · = (an − λ)xn = 0.
Thus the linear system (A − λI)x = 0 has only the zero
solution, and so λ is not an eigenvalue of A.
If λ ∈ K appears g times on the diagonal of the matrix A,
then the geometric multiplicity of the eigenvalue λ is equal to
g since the number of zero rows in an REF of A − λI is g .

(ii) Let A be an upper triangular matrix with diagonal entries

a11 , . . . , ann . Again, the characteristic polynomials factors as
pA (t) = (a11 − t)(a22 − t) · · · (ann − t).
So the eigenvalues of A are precisely a11 , . . . , ann . The
algebraic multiplicities can be found as in the previous
example. However, they may not always coincide with the
corresponding geometric multipliciities.
Saurav Bhaumik, IIT Bombay MA110: Lecture 10
Examples Contd.

 
3 1
For example, consider the 2 × 2 matrix A := . Clearly 3
0 3
is the only eigenvalue of A and its algebraic multiplicity is 2.
On the other hand, the homogeneous linear system
(A − 3I)x = 0 comprises of the equations x2 = 0 and 0 = 0.
So the eigenspace of A associated with the eigenvalue 3 has
[1 0]T as its basis. Thus the geometric multiplicity of the
eigenvalue 3 of A is 1.

Saurav Bhaumik, IIT Bombay MA110: Lecture 10

Examples Contd.

Let A be an upper triangular matrix with diagonal entries

a11 , . . . , ann . Now a11 is an eigenvalue of A since Ae1 = a11 e1 .
In fact, each a11 , . . . , ann is an eigenvalue of A. To see this, let
λ ∈ {a11 , . . . , ann }, and let k be the smallest positive integer
such that akk = λ. Then A − λI is an upper triangular matrix
with the kth diagonal entry zero and earlier diagonal entries
nonzero, and so in an REF of A − λI, the kth column cannot
be pivotal. Hence rank(A − λI) < n. This shows that λ is an
eigenvalue of A; eigenvectors of A corresponding to λ can be
found by backward substitution.
On the other hand, if λ ̸∈ {a11 , . . . , ann }, then backward
substitution shows that the linear system (A − λI)x = 0 has
only the zero solution, and so λ is not an eigenvalue of A.

Saurav Bhaumik, IIT Bombay MA110: Lecture 10

(iii) Let A be a lower triangular matrix.Then AT is an upper
triangular matrix with the same diagonal elements as A. Also,
rank(A − λI) = rank(AT − λI) for any λ ∈ K. It follows that
each diagonal element of A is an eigenvalue of A, and no
other λ ∈ K is an eigenvalue of A. Eigenvectors of A
corresponding to λ can be found by forward substitution.

Saurav Bhaumik, IIT Bombay MA110: Lecture 10

In general, solving an eigenvalue problem Ax = λ x is much
harder than finding solutions of a linear system Ax = b. In the
latter case, the matrix A and the right side b are given. On
the other hand, in the eigenvalue problem, the ‘unknown’
vector x appears on both sides of the equation, and
additionally, an ‘unknown’ scalar λ appears on the right side.
We need to find an eigenvalue λ of A and a corresponding
eigenvector x of A simultaneously. It is tough, but if one of
them is known beforehand, then the other can be found easily.
Suppose a scalar λ is known to be an eigenvalue of A. Then
all eigenvectors of A corresponding to λ can be obtained by
finding the general solution of the homogeneous linear system
(A − λI) x = 0. Next, suppose a nonzero vector x is known to
be an eigenvector of A. Then one only needs to calculate Ax
and observe that it is a scalar multiple of x. This scalar is the
corresponding eigenvalue of A.
Saurav Bhaumik, IIT Bombay MA110: Lecture 10
Examples  
−2 2 −3
(i) Let A := 2
 1 −6, and suppose somehow we know
−1 −2 0
that λ := −3 is an eigenvalue of A. Let
 
1 2 −3
B := A − (−3)I = A + 3I =  2 4 −6 .
−1 −2 3
 
1 2 −3
By EROs, we can transform B to B′ := 0 0 0  . Now the
0 0 0
′ 3×1
solution space of B x = 0 is {x ∈ R : x1 + 2x2 − 3x3 = 0},
which is also the solution space of Bx := (A + 3I)x = 0, the
 T  T
basic solutions being s2 := −2 1 0 and s3 := 3 0 1 .
Thus the eigenvectors of A corresponding to the eigenvalue
λ = −3 are the nonzero linear combinations of s2 and s3 . The
geometric multiplicity of the eigenvalue −3 is equal to 2.
Saurav Bhaumik, IIT Bombay MA110: Lecture 10
  
29 6 −1
1 
(ii) Let A := 29 16 −11, and suppose somehow we
10
25 10 15
 T
know that x := 1 −1 3 is an eigenvector of A. We easily
 T
find that Ax = 2 1 −1 3 . Hence 2 is the corresponding
eigenvalue of A.
We saw that eigenvalue problems for diagonal matrices are the
easiest to solve. We wonder when a nondiagonal matrix would
‘behave like a diagonal matrix’. To make this precise, we
introduce the following notion.
Let A, B ∈ Kn×n . We say that A is similar to B (over K) if
there is an invertible P ∈ Kn×n such that B = P−1 AP, that is,
AP = PB. In this case, we write A ∼ B.

Saurav Bhaumik, IIT Bombay MA110: Lecture 10

Similarity of Square Matrices
Definition
Let A, B ∈ Kn×n . We say that A is similar to B (over K) if
there is an invertible P ∈ Kn×n such that B = P−1 AP, that is,
AP = PB. In this case, we write A ∼ B.
One can easily check (i) A ∼ A, (ii) if A ∼ B then B ∼ A,
and (iii) if A ∼ B and B ∼ C, then A ∼ C.
Examples:    
4 −3 3 1
(i) Let A := . It is easily seen that P := is
2 −1   2 1
1 −1
invertible and . P−1 = . Hence A is similar to
−2 3
        
1 −1 4 −3 3 1 1 −1 6 1 2 0
B := = = ,
−2 3 2 −1 2 1 −2 3 4 1 0 1
which is a diagonal matrix.
Saurav Bhaumik, IIT Bombay MA110: Lecture 10
More Examples and a Characterization of Similarity
(ii) Let A ∈ Kn×n . Then A ∼ I ⇐⇒ A = I.
(iii) Let A ∈ Kn×n , and let E be n×n an elementary matrix.
Then B := EAE−1 is similar to A. Note: EA is obtained from
A by an elementary row operation on A, and B is obtained
from EA by the ‘reverse column operation’ on EA.
Similarity of matrices has the following characterization.
Proposition
Let A, B ∈ Kn×n . Then A ∼ B if and only if there is an
ordered basis E for Kn×1 such that B is the matrix of the
linear transformation TA : Kn×1 → Kn×1 with respect to E .
In fact, B = P−1 AP if and only if the columns of P form an
ordered basis, say E , for Kn×1 and B = MEE (TA ).

A proof of this proposition will be outlined later.

Saurav Bhaumik, IIT Bombay MA110: Lecture 10
Diagonalizability
Definition
A matrix A ∈ Kn×n is called diagonalizable (over K) if A is
similar to a diagonal matrix (over K).

Here is a useful characterization of diagonalizability.

Proposition
A matrix A ∈ Kn×n is diagonalizable if and only if there is a
basis for Kn×1 consisting of eigenvectors of A. In fact,
P−1 AP = D, where P, D ∈ Kn×n are of the form
 
P = x1 · · · xn and D = diag(λ1 , . . . , λn )
⇐⇒ {x1 , . . . , xn } is a basis for Kn×1 and
Axk = λk xk for k = 1, . . . , n.

We shall see a proof of this result as well as many examples of

diagonalizable and non-diagonalizable matrices later.
Saurav Bhaumik, IIT Bombay MA110: Lecture 10