MA110

Lecture 21

Saurav Bhaumik
Department of Mathematics
IIT Bombay

Spring 2025

Saurav Bhaumik, IIT Bombay MA110: Lecture 21

Linear Transformations
Definition
Let V and W be vector spaces over K. A linear
transformation or a linear map from V to W is a function
T : V → W which ‘preserves’ the operations of addition and
scalar multiplication, that is, for all u, v ∈ V and α ∈ K,

T (u + v ) = T (u) + T (v ) and T (α v ) = α T (v ).

It is clear that if T : V → W is linear, then T (0) = 0. Also,

T ‘preserves’ linear combinations of elements of V :

T (α1 v1 + · · · + αk vk ) = α1 T (v1 ) + · · · + αk T (vk )

for all k ∈ N, v1 , . . . , vk ∈ V and α1 , . . . , αk ∈ K.

Remark: A linear transformation from a vector space V to
itself is often called a linear operator on V .
Saurav Bhaumik, IIT Bombay MA110: Lecture 21
Examples
1 . Let A be an m × n matrix with entries in K. Then the
map T : Kn×1 → Km×1 defined by T (x) := A x is linear.
Similarly, the map T ′ : K1×m → K1×n defined by T ′ (y) := yA
is linear. More generally, the map

T : Kn×p → Km×p defined by T (X) := A X

is linear, and the map

T ′ : Kp×m → Kp×n defined by T ′ (Y) := YA

is linear.
2 . T : Km×n → Kn×m defined by T (A) := AT is linear.
3 . The map T : Kn×n → K defined by T (A) := trace A is
linear. But A 7−→ det A does not define a linear map.
4 . The map T : K[X ] → K defined by T (p(X )) = p(0) is
linear.
Saurav Bhaumik, IIT Bombay MA110: Lecture 21
 5 . Let V := c0 , the set of all sequences in K which converge
to 0. Then the map T : V → V defined by
T (x1 , x2 , . . .) := (0, x1 , x2 , . . .)
is linear, and so is the map T ′ : V → V defined by
T ′ (x1 , x2 , . . .) := (x2 , x3 , . . .).
Note that T ′ ◦T is the identity map on V , but T ◦T ′ is not
the identity map on V . The map T is called the right shift
operator and T ′ is called the left shift operator on V .
6 . Let V := C 1 ([a, b]), the set of all real-valued continuously
differentiable functions, and let W := C ([a, b]), the set of all
real-valued continuous functions on [a, b]. Then the map
T ′ : V → W defined by T ′ (f ) = f ′ is linear. Also, the map
Z x
T : W → V defined by T (f )(x) := f (t)dt for x ∈ [a, b],
a
is linear. [Question. What are T ◦T and T ′ ◦T ?]
′

Saurav Bhaumik, IIT Bombay MA110: Lecture 21

Let V and W be vector spaces over K, and let T : V → W be
a linear map. Two important subspaces associated with T are
(i) N (T ) := {v ∈ V : T (v ) = 0}, the null space of T , which
is a subspace of V ,
(ii) I(T ) := {T (v ) : v ∈ V }, the image space of T , which
is a subspace of W .
Suppose V is finite dimensional, and let dim V = n. Since
N (T ) is a subspace of V , it is finite dimensional and
dim N (T ) ≤ n
Let v1 , . . . , vn be a basis for V . If v ∈ V , then there are
α1 , . . . , αn ∈ K such that v = α1 v1 + · · · + αn vn , so that
T (v ) = α1 T (v1 ) + · · · + αn T (vn ). This shows that
I(T ) = span{T (v1 ), . . . , T (vn )}. Hence I(T ) is also finite
dimensional and dim I(T ) ≤ n.

Saurav Bhaumik, IIT Bombay MA110: Lecture 21

Definition
The dimension of N (T ) is called the nullity of the linear map
T , and the dimension of I(T ) is called the rank of T .

The Rank-Nullity Theorem for a matrix A that we proved

earlier is a special case of the following result.

Proposition (Rank-Nullity Theorem for Linear Maps)

Let V and W be vector spaces over K, and let T : V → W
be a linear map. Suppose dim V = n ∈ N. Then
rank(T ) + nullity(T ) = n.

Proof (Sketch): Let s := nullity(T ) and let {u1 , . . . , us } be a

basis of N (T ). Extend the linearly independent set
{u1 , . . . , us } to a basis {u1 , . . . , us , us+1 , . . . , un } of V . Check
that the set {T (us+1 ), . . . , T (un )} is a basis of I(T ).
Saurav Bhaumik, IIT Bombay MA110: Lecture 21
Corollary
Let V , W be finite dimensional vector spaces with dim V = n
and dim W = m. Also, let T : V → W be a linear map. Then

T is one-one ⇐⇒ rank(T ) = n.

In particular, if T is one-one, then n ≤ m. Further,

if m = n, then T is one-one ⇐⇒ T is onto.

Proof. The first assertion follows from the Rank-Nullity
Theorem since

T is one-one ⇐⇒ N (T ) = {0} ⇐⇒ nullity(T ) = 0.

If T is one-one, then n = rank(T ) = dim I(T ) ≤ dim W = m.

Further, if m = n, then rank(T ) = n ⇐⇒ T is onto.

Saurav Bhaumik, IIT Bombay MA110: Lecture 21

As another application of the Rank-Nullity Theorem, we find
an interesting relation between dimensions of finite
dimensional subspaces of a vector space.
Proposition
Let W1 and W2 be finite dimensional subspaces of a vector
space V . Then

dim W1 + dim W2 = dim(W1 ∩ W2 ) + dim(W1 + W2 ).

Proof. The dimension of the vector space W1 × W2 is equal

to dim W1 + dim W2 . Define T : W1 × W2 → W1 + W2 by
T (w1 , w2 ) := w1 − w2 . Then T is linear, and
N (T ) = {(w , w ) : w ∈ W1 ∩ W2 } and I(T ) = W1 + W2 .
Hence by the Rank-Nullity Theorem,
dim(W1 + W2 ) + dim(W1 ∩ W2 ) = dim(W1 × W2 ).
Saurav Bhaumik, IIT Bombay MA110: Lecture 21
To a linear map from a finite dimensional vector space to
another finite dimensional space, one can associate a matrix
exactly as before.
Let V be a vector space of dimension n, and let
E := (v1 , . . . , vn ) be an ordered basis for V . Also, let W be a
vector space of dimension m, and let F := (w1 , . . . , wm ) be an
ordered basis for W . Let T : V → W be a linear map. Then
for each k = 1, . . . , n, we can uniquely write
m
X
T (vk ) = a1k w1 + · · · + amk wm = ajk wj for some ajk ∈ K.
j=1

The m×n matrix A := [ajk ] is called the matrix of the linear

transformation T : V → W with respect to the ordered
basis E := (v1 , . . . , vn ) of V and the ordered basis
F := (w1 , . . . , wm ) of W . It is denoted by MEF (T ) .
Saurav Bhaumik, IIT Bombay MA110: Lecture 21
Examples 1. Define T : Pn → Pn−1 by T (p) = p ′ , the
derivative of p. Consider the ordered bases E := (1, t, . . . , t n )
and F := (1, t, . . . , t n−1 ) of Pn and Pn−1 respectively. Then
the n × (n + 1) matrix of the linear map T with respect to
these bases is
 
0 1 0 0 ··· 0
0 0 2 0 · · · 0
 
MEF (T ) := 0 0 0 3 · · · 0 .
 
 .. .. .. .. . . .. 
. . . . . .
0 0 0 0 ··· n

2. Let T : K2×2 → K2×2 be the linear transformation defined

by T (A) = AT . Then the matrix of T with respect  to the
1 0 0 0
0 0 1 0
basis E := {E11 , E12 , E21 , E22 } is MEE (T ) := 
0 1 0 0 .


0 0 0 1
Saurav Bhaumik, IIT Bombay MA110: Lecture 21
Eigenvalues and Eigenvectors of Linear Operators
Definition
Let V be a vector space over K, and let T : V → V be a
linear operator. A scalar λ ∈ K is called an eigenvalue of T
if there is a nonzero v ∈ V such that T (v ) = λv , and then v
is called an eigenvector or an eigenfunction of T
corresponding to λ, and the subspace N (T − λI ) is called the
eigenspace of T .

Example: Let V denote the vector space C ∞ (R) of all

real-valued infinitely differentiable functions on R. Define
T (f ) = f ′ for f ∈ V . Then T is a linear operator on V .
Given λ ∈ R, consider fλ (t) := e λt for t ∈ R. Then fλ ∈ V ,
fλ ̸= 0 and T (fλ ) = λ fλ . Thus every λ ∈ R is an eigenvalue of
T with fλ as a corresponding eigenfunction. In fact, any
eigenfunction of T corresponding to λ is a scalar multiple of fλ .
Saurav Bhaumik, IIT Bombay MA110: Lecture 21
We now consider a vector space V of finite dimension. Let E
be an ordered basis for V , and let A := MEE (T ), the matrix of
the linear operator T with respect to E . We remark that if F
is another ordered basis for V , and B := MFF (T ), the matrix
of the linear operator T with respect to F , then B is similar to
A; in fact we have seen that B = P−1 AP, where P = MEF (I ),
and where I : V → V is the identity map.
Definition
The geometric multiplicity of an eigenvalue of T is the
dimension of the corresponding eigenspace. It equals the
geometric multiplicity of λ as an eigenvalue of the associated
matrix A. The characteristic polynomial of T is defined to
be the characteristic polynomials of A. Further, T is called
diagonalizable if the matrix A is diagonalizable.

Saurav Bhaumik, IIT Bombay MA110: Lecture 21

Definition
The algebraic multiplicity of an eigenvalue of the linear
operator T is defined to be the algebraic multiplicity of the
associated matrix A.
The relationships between the geometric multiplicity and the
algebraic multiplicity of an eigenvalue of a square matrix hold
for a linear operator as well.
The above definitions do not depend on the choice of the
ordered basis E for V because if F is any other ordered basis
of V , then the matrix B := MFF (T ) is similar to the matrix
A := MEE (T ) as we have seen earlier.
Results about the linear independence of eigenvectors
corresponding to distinct eigenvalues hold in the general case.

Saurav Bhaumik, IIT Bombay MA110: Lecture 21

Inner Product Spaces
Definition
Let V be a vector space over K. An inner product on V is a
function ⟨· , ·⟩ : V ×V → K satisfying the following properties.
For u, v , w ∈ V and α, β ∈ K,
1. ⟨v , v ⟩ ≥ 0 and ⟨v , v ⟩ = 0 ⇐⇒ v = 0, (positive definite)
2. ⟨u, αv + βw ⟩ = α⟨u, v ⟩+β⟨u, w ⟩, (linear in 2nd variable)
3. ⟨v , u⟩ = ⟨u, v ⟩. (conjugate symmetric)

From the above properties, conjugate linearity in the 1st

variable follows: ⟨αu + βv , w ⟩ = α⟨u, w ⟩ + β⟨v , w ⟩.
A vector space V over K with a prescribed inner product on it
is called an inner product space.
If u, v ∈ V and ⟨u, v ⟩ = 0, then we say that u and v are
orthogonal, and we write u ⊥ v .
Saurav Bhaumik, IIT Bombay MA110: Lecture 21
For v ∈ V , we define the norm of v by ∥v ∥ := ⟨v , v ⟩1/2 .
If v ∈ V and ∥v ∥ = 1, then we say that v is a unit vector or
a unit function. The set {v ∈ V : ∥v ∥ ≤ 1} is called the
unit ball of V .
Examples
1 . We have already studied the primary example, namely
V := Kn×1 with the usual inner product ⟨x, y⟩ := x∗ y for
x, y ∈ Kn×1 . There are other inner products on Kn×1 . For
example, let w1 , . . . , wn be positive real numbers, and define
⟨x, y⟩ := w1 x 1 y1 + · · · + wn x n yn for x, y ∈ Kn×1 .
On the other hand, the function on R4×1 × R4×1 defined by
⟨x, y⟩M := x1 y1 + x2 y2 + x3 y3 − x4 y4 for x, y ∈ R4×1
is not an inner product on R4×1 . Note that for x ∈ R4×1 ,
⟨x, x⟩M = x12 + x22 + x32 − x42 . (This is used in defining the
Minkowski space)
Saurav Bhaumik, IIT Bombay MA110: Lecture 21
2 . Let V := C ([a, b]), the vector space of all continuous
K-valued functions on [a, b]. Define
Z b
⟨f , g ⟩ := f (t)g (t)dt for f , g ∈ V .
a

It is easy to check that this is an inner product on V . We shall

call this inner product the usual inner product on C ([a, b]).
R 1/2
b
In this case, the norm of f ∈ V is ∥f ∥ := a |f (t)|2 dt .
This example gives a continuous analogue of the usual inner
product on Kn×1 .
There are other inner products on V . For example, let
w : [a, b] → R be positive function, and define
Z b
⟨f , g ⟩ := w (t)f (t)g (t)dt for f , g ∈ V .
a

Saurav Bhaumik, IIT Bombay MA110: Lecture 21

Let w be a nonzero element of V . As earlier, define
⟨w , v ⟩
Pw (v ) := w for v ∈ V .
⟨w , w ⟩
It is called the (orthogonal) projection of v in the direction
of w . It is easy to see that Pw : V → V is a linear map and
its image space is one dimensional. Also, Pw (w ) = w , so that
(Pw )2 := Pw ◦ Pw = Pw .
Two important properties of the projection of a vector in the
direction of another (nonzero) vector are as follows.
Proposition
Let w ∈ V be nonzero. Then for every v ∈ V ,


(i) v − Pw (v ) ⊥ w and (ii) ∥Pw (v )∥ ≤ ∥v ∥.

The proof of (i) is an easy verification, and (ii) follows from

the formula ∥v ∥2 = ∥Pw (v )∥2 + ∥v − Pw (v )∥2 , which is a
consequence of (i).
Saurav Bhaumik, IIT Bombay MA110: Lecture 21
Theorem
Let ⟨· , ·⟩ be an inner product on a vector space V , and let
v , w ∈ V . Then
(i) (Schwarz Inequality) |⟨v , w ⟩| ≤ ∥v ∥∥w ∥.
(ii) (Triangle Inequality) ∥v + w ∥ ≤ ∥v ∥ + ∥w ∥.

Proof. (i) First, suppose w = 0. Then for any v ∈ V ,

⟨v , w ⟩ = ⟨v , 0⟩ = ⟨v , 0 + 0⟩ = 2⟨v , 0⟩, and so ⟨v , w ⟩ = 0.
Also, ∥w ∥ = 0. Hence we are done.
Now suppose w ̸= 0. Then by (ii) of the previous proposition,
⟨w , v ⟩
w = ∥Pw (v )∥ ≤ ∥v ∥,
⟨w , w ⟩
that is,
|⟨w , v ⟩|∥w ∥ ≤ ∥v ∥⟨w , w ⟩ = ∥v ∥∥w ∥2 .
Hence |⟨v , w ⟩| ≤ ∥v ∥∥w ∥.
Saurav Bhaumik, IIT Bombay MA110: Lecture 21
(ii) Since ⟨v , w ⟩ + ⟨w , v ⟩ = 2 R ⟨v , w ⟩, we see that

∥v + w ∥2 = ⟨v + w , v + w ⟩ = ∥v ∥2 + ∥w ∥2 + 2 R ⟨v , w ⟩
≤ ∥v ∥2 + ∥w ∥2 + 2 |⟨v , w ⟩|
≤ ∥v ∥2 + ∥w ∥2 + 2 ∥v ∥∥w ∥ (by (i) above)
= (∥v ∥ + ∥w ∥)2 .

Thus ∥v + w ∥ ≤ ∥v ∥ + ∥w ∥.
We observe that the norm function ∥ · ∥ : V → K satisfies the
following three crucial properties:
(i) ∥v ∥ ≥ 0 for all v ∈ V and ∥v ∥ = 0 ⇐⇒ v = 0,
(ii) ∥αv ∥ = |α|∥v ∥ for all α ∈ K and v ∈ V ,
(iii) ∥v + w ∥ ≤ ∥v ∥ + ∥w ∥ for all v , w ∈ V .

Saurav Bhaumik, IIT Bombay MA110: Lecture 21

Let V be an inner product space. Let E be a subset of V .
Define
E ⊥ := {w ∈ V : w ⊥ v for all v ∈ E }.
It is easy to see that E ⊥ is a subspace of V .
The set E is said to be orthogonal if any two (distinct)
elements of E are orthogonal (to each other), that is, v ⊥ w
for all v , w in E with v ̸= w . An orthogonal set whose
elements are unit vectors is called an orthonormal set.
If E is orthogonal and does not contain 0, then E is linearly
independent. For example, let V := C [−π, π] and
E := {cos nt : n ∈ N} ∪ {sin nt : n ∈ N}. Since E is
orthogonal and 0 ̸∈ E , the set E is linearly independent.

Saurav Bhaumik, IIT Bombay MA110: Lecture 21

If we are given a sequence of linearly independent elements of
V , then we can construct an orthogonal subset of V not
containing 0, retaining the span of the elements so
constructed at every stepby the Gram-Schmidt
Orthogonalization Process (G-S OP), just as discussed earlier.
Let (vn ) be a sequence of linearly independent elements in V .
Define w1 := v1 , and for j ∈ N, define

wj+1 := vj+1 − Pw1 (vj+1 ) − · · · − Pwj (vj+1 )

⟨w1 , vj+1 ⟩ ⟨wj , vj+1 ⟩
= vj+1 − w1 − · · · − wj .
⟨w1 , w1 ⟩ ⟨wj , wj ⟩

Then span{w1 , . . . , wj+1 } = span{v1 , . . . , vj+1 }, and the set

{w1 , . . . , wj+1 } is orthogonal.

Saurav Bhaumik, IIT Bombay MA110: Lecture 21

Now let uj := wj /∥wj ∥ for j ∈ N, then (u1 , u2 , . . .) is an
ordered orthonormal set such that for each j ∈ N,

span{v1 , . . . , vj } = span{w1 , . . . , wj } = span{u1 , . . . , uj }.

Example
Let V be the set of all real-valued polynomial functions on
[−1, 1] along with the inner product defined by
Z 1
⟨p, q⟩ := p(t)q(t)dt for p, q ∈ V .
−1

For j = 0, 1, 2, . . ., let pj (t) := t j , t ∈ [−1, 1]. Let us

orthogonalize the set {p0 , p1 , p2 , p3 }. Define q0 := p0 , and
 Z 1 
⟨q0 , p1 ⟩ 1
q1 := p1 − q0 = p1 − t dt p0 = p1 .
⟨q0 , q0 ⟩ 2 −1

Saurav Bhaumik, IIT Bombay MA110: Lecture 21

Next, define
⟨q0 , p2 ⟩ ⟨q1 , p2 ⟩
q2 := p2 − q0 − q1
⟨q0 , q0 ⟩ ⟨q1 , q1 ⟩
 Z 1   Z 1 
1 2 3 3
= p2 − t dt q0 − t dt q1
2 −1 2 −1
1
= p2 − p0 ,
3
and similarly,

⟨q0 , p3 ⟩ ⟨q1 , p3 ⟩ ⟨q2 , p3 ⟩

q3 := p3 − q0 − q1 − q2
⟨q0 , q0 ⟩ ⟨q1 , q1 ⟩ ⟨q2 , q2 ⟩
3
= p3 − p1 .
5
.

Saurav Bhaumik, IIT Bombay MA110: Lecture 21

 √ √ √ √ √
Further, ∥q
√ √0 ∥ = 2, ∥q1 ∥ = 2/ 3, ∥q2 ∥ = 2 2/3 5 and
∥q3 ∥ = 2 2/5 7.
Hence we obtain the following orthonormal subset of V having
the same span as span{p0 , p1 , p2 , p3 }, namely all real-valued
polynomial functions of degree at most 3:
√ √
2 6
u0 (t) := , u1 (t) := t,
√2 2 √
10 2 14
u2 (t) := (3t − 1), u3 (t) := (5t 3 − 3t).
4 4
The sequence of orthonormal polynomials thus obtained by
orthonormalizing the monomials by the G-S OP is known as
the sequence of Legendre polynomials. These are of much
use in many contexts.

Saurav Bhaumik, IIT Bombay MA110: Lecture 21

Let V be a finite dimensional inner product space. An
orthonormal basis for V is a basis for V which is an
orthonormal subset of V .
We have proved the following results for subspaces of Kn×1 .
Their proofs remain valid for any inner product space.
If u1 , . . . , uk is an orthonormal set in V , then we can extend it
to an orthonormal basis. As a consequence, every nonzero
vector subspace V has an orthonormal basis.
The G-S OP enables us to improve the quality of a given basis
for V by orthonormalizing it. For instance, if {u1 , . . . , un } is
an orthonormal basis for V , and v ∈ V , then it is extremely
easy to write v as a linear combination of u1 , . . . , un ; in fact

v = ⟨u1 , v ⟩u1 + · · · + ⟨un , v ⟩un .

Saurav Bhaumik, IIT Bombay MA110: Lecture 21

Orthogonal Projections
Let W be a subspace of a finite dimensional inner product
space V . The Orthogonal Projection Theorem says that
for every v ∈ V , there are unique w ∈ W and w̃ ∈ W ⊥ such
that v = w + w̃ , that is, V = W ⊕ W ⊥ . The map
PW : V → V given by PW (v ) = w is linear and satisfies
(PW )2 = PW . It is called the orthogonal projection map of
V onto the subspace W .
In fact, if u1 , . . . , uk is an orthonormal basis for W , then

PW (v ) = ⟨u1 , v ⟩u1 + · · · + ⟨uk , v ⟩uk for v ∈ V .

Given v ∈ V , its orthogonal projection PW (v ) is the unique

best approximation to v from W .
Further, PW (v ) is the unique element of W such that
v − PW (v ) is orthogonal to W .
Saurav Bhaumik, IIT Bombay MA110: Lecture 21
Definition
Suppose V is an inner product space of dimension n. For a
linear operator T : V → V , define its adjoint T ∗ : V → V by

⟨T (u), v ⟩ = ⟨u, T ∗ (v )⟩ for all u, v ∈ V .

Define T to be Hermitian or self-adjoint if T = T ∗ , and

skew-Hermitian or skew self-adjoint if T = −T ∗ .

Thus, T is Hermitian if

⟨T (u), v ⟩ = ⟨u, T (v )⟩ for all u, v ∈ V ,

and T is skew-Hermitian if

⟨T (u), v ⟩ = −⟨u, T (v )⟩ for all u, v ∈ V ,

Saurav Bhaumik, IIT Bombay MA110: Lecture 21

Note that for A ∈ Kn×n and x, y ∈ Kn×1 ,

⟨A x, y⟩ = (A x)∗ y = x∗ (A∗ y) = ⟨x, A∗ y⟩.

Hence a matrix A is self-adjoint, that is, A∗ = A if and only if

⟨A x, y⟩ = ⟨x, Ay⟩. The following result, therefore, is natural.
Proposition
Let V be a finite dimensional inner product space, and let
T : V → V be a linear operator. Then T is Hermitian if and
only if the matrix of T with respect to any ordered
orthonormal basis of V is self-adjoint.

An operator T which commutes with its adjoint T ∗ will be

called normal operator on V . One can prove the spectral
theorem for a normal operator on a finite dimensional inner
product space V just as before.

Saurav Bhaumik, IIT Bombay MA110: Lecture 21

 THE END

Saurav Bhaumik, IIT Bombay MA110: Lecture 21