Vector Spaces and Linear Transformations

Beifang Chen
Fall 2006

1 Vector spaces
A vector space is a nonempty set V , whose objects are called vectors, equipped with two operations,
called addition and scalar multiplication: For any two vectors u, v in V and a scalar c, there are unique
vectors u + v and cu in V such that the following properties are satisfied.
1. u + v = v + u,
2. (u + v) + w = u + (v + w),
3. There is a vector 0, called the zero vector, such that u + 0 = u,
4. For any vector u there is a vector −u such that u + (−u) = 0;
5. c(u + v) = cu + cv,
6. (c + d)u = cu + du,
7. c(du) = (cd)u,
8. 1u = u.
By definition of vector space it is easy to see that for any vector u and scalar c,

0u = 0, c0 = 0, −u = (−1)u.

For instance,
(3) (4) (2)
0u = 0u + 0 = 0u + (0u + (−0u)) = (0u + 0u) + (−0u)
(6) (4)
= (0 + 0)u + (−0u) = 0u + (−0u) = 0;
(7)
c0 = c(0u) = (c0)u = 0u = 0;
−u = −u + 0 = −u + (1 − 1)u = −u + u + (−1)u = 0 + (−1)u = (−1)u.

Example 1.1. (a) The Euclidean space Rn is a vector space under the ordinary addition and scalar
multiplication.
(b) The set Pn of all polynomials of degree less than or equal to n is a vector space under the ordinary
addition and scalar multiplication of polynomials.
(c) The set M(m, n) of all m × n matrices is a vector space under the ordinary addition and scalar
multiplication of matrices.
(d) The set C[a, b] of all continuous functions on the closed interval [a, b] is a vector space under the
ordinary addition and scalar multiplication of functions.
Definition 1.1. Let V and W be vector spaces, and W ⊆ V . If the addition and scalar multiplication in
W are the same as the addition and scalar multiplication in V , then W is called a subspace of V .

1
 If H is a subspace of V , then H is closed for the addition and scalar multiplication of V , i.e., for any
u, v ∈ H and scalar c ∈ R, we have
u + v ∈ H, cv ∈ H.
For a nonempty set S of a vector space V , to verify whether S is a subspace of V , it is required to check
(1) whether the addition and scalar multiplication are well defined in the given subset S, that is, whether
they are closed under the addition and scalar multiplication of V ; (2) whether the eight properties (1-8) are
satisfied. However, the following theorem shows that we only need to check (1), that is, to check whether
the addition and scalar multiplication are closed in the given subset S.
Theorem 1.2. Let H be a nonempty subset of a vector space V . Then H is a subspace of V if and only if
H is closed under addition and scalar multiplication, i.e.,
(a) For any vectors u, v ∈ H, we have u + v ∈ H,
(b) For any scalar c and a vector v ∈ H, we have cv ∈ H.

Example 1.2. (a) For a vector space V , the set {0} of the zero vector and the whole space V are subspaces
of V ; they are called the trivial subspaces of V .
(b) For an m × n matrix A, the set of solutions of the linear system Ax = 0 is a subspace of Rn . However,
if b 6= 0, the set of solutions of the system Ax = b is not a subspace of Rn .
(c) For any vectors v1 , v2 , . . . , vk in Rn , the span Span {v1 , v2 , . . . , vk } is a subspace of Rn .
(d) For any linear transformation T : Rn → Rm , the image

T (Rn ) = {T (x) : x ∈ Rn }

of T is a subspace of Rm , and the inverse image

T −1 (0) = {x ∈ Rn : T (x) = 0}

is a subspace of Rn .

2 Some special subspaces

Lecture 15

Let A be an m × n matrix. The null space of A, denoted by Nul A, is the space of solutions of the linear
system Ax = 0, that is,
Nul A = {x ∈ Rn : Ax = 0}.
The column space of A, denoted by Col A, is the span of the column vectors of A, that is, if A =
[a1 , a2 , . . . , an ], then
Col A = Span {a1 , a2 , . . . , an }.
The row space of A is the span of the row vectors of A, and is denoted by Row A.

Let T : Rn → Rm , T (x) = Ax, be a linear transformation. Then Nul A is the set of inverse images of 0
under T and Col A is the image of T , that is,

Nul A = T −1 (0) and Col A = T (Rn ).

2
3 Linear transformations
Let V and W be vector spaces. A function T : V → W is called a linear transformation if for any vectors
u, v in V and scalar c,
(a) T (u + v) = T (u) + T (v),
(b) T (cu) = cT (u).
The inverse images T −1 (0) of 0 is called the kernel of T and T (V ) is called the range of T .
Example 3.1. (a) Let A is an m × m matrix and B an n × n matrix. The function

F : M(m, n) → M(m, n), F (X) = AXB

is a linear transformation. For instance, for m = n = 2, let

· ¸ · ¸ · ¸
1 2 2 1 x1 x2
A= , B= , X= .
1 3 2 3 x3 x4

Then F : M(2, 2) → M(2, 2) is given by

· ¸· ¸· ¸ · ¸
1 2 x1 x2 2 1 2x1 + 2x2 + 4x3 + 4x4 x1 + 3x2 + 2x3 + 6x4
F (X) = = .
1 3 x3 x4 2 3 2x1 + 2x2 + 6x3 + 6x4 x1 + 3x2 + 3x3 + 9x4

(b) The function D : P3 → P2 , defined by

¡ ¢
D a0 + a1 t + a2 t2 + a3 t3 = a1 + 2a2 t + 3a3 t2 ,

is a linear transformation.
Proposition 3.1. Let T : V → W be a linear transformation. Then T −1 (0) is a subspace of V and T (V )
is a subspace of W . Moreover,
(a) If V1 is a subspace of V , then T (V1 ) is a subspace of W ;
(b) If W1 is a subspace of W , then T −1 (W1 ) is a subspace of V .
Proof. By definition of subspaces.
Theorem 3.2. Let T : V → W be a linear transformation. Given vectors v1 , v2 , . . . , vk in V .
(a) If v1 , v2 , . . . , vk are linearly dependent, then T (v1 ), T (v2 ), . . . , T (vk ) are linearly dependent;
(b) If T (v1 ), T (v2 ), . . . , T (vk ) are linearly independent, then v1 , v2 , . . . , vk are linearly independent.

4 Independent sets and bases

Definition 4.1. Vectors v1 , v2 , . . . , vp of a vector space V are called linearly independent if, whenever
there are constants c1 , c2 , . . . , cp such that

c1 v1 + c2 v2 + · · · + cp vp = 0,

we have
c1 = c2 = · · · = cp = 0.
The vectors v1 , v2 , . . . , vp are called linearly dependent if there exist constants c1 , c2 , . . . , cp , not all zero,
such that
c1 v1 + c2 v2 + · · · + cp vp = 0.

3
 Any family of vectors that contains the zero vector 0 is linearly dependent. A single vector v is linearly
independent if and only if v 6= 0.
Theorem 4.2. Vectors v1 , v2 , . . . , vk (k ≥ 2) are linearly dependent if and only if one of the vectors is a
linear combination of the others, i.e., there is one i such that
vi = a1 v1 + · · · + ai−1 vi−1 + ai+1 vi+1 + · · · + ak vk .
Proof. Since the vectors v1 , v2 , . . . , vk are linearly dependent, there are constants c1 , c2 , . . . , ck , not all zero,
such that
c1 v + c2 v2 + · · · + ck vk = 0.
Let ci 6= 0. Then
µ ¶ µ ¶ µ ¶ µ ¶
c1 ci−1 ci+1 ck
vi = − v1 + · · · + − vi−1 + − vi+1 + · · · + − vk .
ci ci ci ci

Note 1. The condition v1 6= 0 can not be omitted. For instance, the set {0, v2 } (v2 6= 0) is a dependent
set, but v2 is not a linear combination of the zero vector v1 = 0.
Theorem 4.3. Let S = {v1 , v2 , . . . , vk } be a subset of independent vectors in a vector space V . If a vector
v can be written in two linear combinations of v1 , v2 , . . . , vk , say,
v = c1 v1 + c2 v2 + · · · + ck vk = d1 v1 + d2 v2 + · · · + dk vk ,
then
c1 = d1 , c2 = d 2 , ..., ck = dk .
Proof. If (c1 , c2 , . . . , ck ) 6= (d1 , d2 , . . . , dk ), then one of the entries in (c1 − d1 , c2 − d2 , . . . , ck − dk ) is nonzero,
and
(c1 − d1 )v1 + (c2 − d2 )v2 + · · · + (ck − dk )vk = 0.
This means that the vectors v1 , v2 , . . . , vk are linearly dependent. This is a contradiction.
Definition 4.4. Let H be a subspace of a vector space V . An ordered set B = {v1 , v2 , . . . , vp } of vectors
in V is called a basis for H if
(a) B is a linearly independent set, and
(b) B spans H, that is, H = Span {v1 , v2 , . . . , vp }.
Example 4.1. (a) The set {1, t, t2 } is basis of P2 .
(b) The set {1, t + 1, t2 + t} is basis of P2 .
(c) The set {1, t + 1, t − 1} is not a basis of P2 .
Proposition 4.5 (Spanning Theorem). Let H be a nonzero subspace of a vector space V and H =
Span {v1 , v2 , . . . , vp }.
(a) If some vk is a linear combination of the other vectors of {v1 , v2 , . . . , vp }, then
H = Span {v1 , . . . , vk−1 , vk+1 , . . . , vp }.

(b) If H 6= {0}, then some subsets of S = {v1 , v2 , . . . , vp } are bases for H.

Proof. It is clear that Span {v1 , . . . , vk−1 , vk+1 , . . . , vp } is contained in H. Write
vk = c1 v1 + · · · + ck−1 vk−1 + ck+1 vk+1 + · · · + cp vp .
Then for any vector v = a1 v1 + · · · + ak vk + · · · + ap vp in H, we have
v = (a1 + ak c1 )v1 + · · · + (ak−1 + ak ck−1 )vk−1
+(ak+1 + ak ck+1 )vk+1 + · · · + (ap + ak cp )vp .
This means that v is contained in Span {v1 , . . . , vk−1 , vk+1 , . . . , vp }.

4
 © ª
Example
© 4.2. The
ª vector space P(t) of polynomials of degree ≤ 2 has a basis B = 1, t, t2 . The set
B1 = 1, t + 1, t2 is also a basis of P(t). However, {1, t + 1, t − 1} is not a basis of P(t).
Example 4.3. The vector space M(2, 2) of 2 × 2 matrices has a basis
½· ¸ · ¸ · ¸ · ¸¾
1 0 0 1 0 0 0 0
B= , , ,
0 0 0 0 1 0 0 1

The following set ½· ¸ · ¸ · ¸ · ¸¾

1 0 1 1 1 1 1 1
C= , , ,
0 0 0 0 1 0 1 1
is also a basis of M(2, 2).

5 Bases of null and column spaces

Example 5.1. Consider the matrix
 
1 4 0 2 4
 −1 −4 1 −3 −2 
A=  2
 = [a1 , a2 , a3 , a4 , a5 ].
8 1 3 10 
1 4 1 1 6

Its reduced row echelon form is the matrix

 
1 4 0 2 4
 0 0 1 −1 2 
B=  0 0
 = [b1 , b2 , b3 , b4 , b5 ].
0 0 0 
0 0 0 0 0

Since Ax = 0 is equivalent to Bx = 0, that is,

x1 a1 + x2 a2 + x3 a3 + x4 a4 + x5 a5 = 0 ⇐⇒ x1 b1 + x2 b2 + x3 b3 + x4 b4 + x5 b5 = 0.

This means that the linear relations among the vectors a1 , a2 , a3 , a4 , a5 are the same as the linear relations
among the vectors b1 , b2 , b3 , b4 , b5 . For instance,

b2 = 4b1 ←→ a2 = 4a1
b4 = 2b1 − b3 ←→ a4 = 2a1 − a3
b5 = 4b1 + 2b3 ←→ a4 = 4a1 + 2a3 .

This shows that row operations do not change the linear relations among the column vectors of a matrix.
Note 2. Let A and B be matrix such that A ∼ B, that is, A is equivalent to B. Then

Nul A = Nul B, Row A = Row B, but Col A 6= Col B.

Theorem 5.1 (Column Space Theorem). The column vectors of a matrix A corresponding to its pivot
positions form a basis of Col A.
Proof. Let B = [b1 , b2 , . . . , bn ] denote the reduced row echelon form of A = [a1 , a2 , . . . , an ]. Let bi1 , bi2 , . . . , bik
be the column vectors of B containing the pivot positions. It is clear that bi1 , bi2 , . . . , bik are linearly inde-
pendent and every column vector of B is a linear combination of the vectors bi1 , bi2 , . . . , bik .
Let ai1 , ai2 , . . . , aik be the corresponding column vectors of A. It suffices to prove that a linear relation
for b1 , b2 , . . . , bn is also a linear relation for a1 , a2 , . . . , an , and vice versa. Notice that a linear relation
among the vectors b1 , b2 , . . . , bn is just a solution of the system Bx = 0; and the systems Ax = 0 and
Bx = 0 have the same solution set. Thus ai1 , ai2 , . . . , aik are linearly independent and every column vector
of A is a linear combination of ai1 , ai2 , . . . , aik . So ai1 , ai2 , . . . , aik form a basis of Col A.

5
6 Coordinate systems
Lecture 17
Theorem 6.1. Let S = {v1 , v2 , . . . , vn } be a basis of a vector space V . Then for each vector v in V , there
exists a unique set of scalars c1 , c2 , . . . , cn such that
v = c1 v1 + c2 v2 + · · · + cn vn .
Proof. Trivial.
Let B = {v1 , v2 , . . . , vn } be a basis of a vector space V . Then every vector v of V has a unique expression
v = c1 v1 + c2 v2 + · · · + cn vn .
The scalars c1 , c2 , . . . , cn are called the coordinates of v relative to the basis B (or B-coordinates of
v); and the vector  
c1
 c2 
 
[v]B =  . 
 .. 
cn
is called the coordinate vector of v relative to B (or the B-coordinate vector of v). We may write
 
c1
£ ¤ 
 c2 
v = c1 v1 + c2 v2 + · · · + cn vn = v1 , v2 , . . . , vn  .  = B[v]B .
 .. 
cn

Example 6.1. Any two linearly independent vectors of R2 form a basis for R2 . For instance, the set
½· ¸ · ¸¾
1 1
B= ,
1 −1
· ¸ · ¸
1 2
is basis of R2 . The vector has the B-coordinate vector . However, the coordinate vector of
· ¸ 3 −1
1
is just itself under the standard basis
3
½· ¸ · ¸¾
1 0
E= , .
0 1
Theorem 6.2. Let B = {v1 , v2 , . . . , vp } be a basis of a subspace H of Rn . Let PB be the matrix
PB = [v1 , v2 , . . . , vp ].
Then for any vector v in Rn ,
v = PB [v]B .
The matrix PB , which transfers the B-coordinate vector [v]B of v to its standard coordinate vector v = [v]E ,
is called the change-of-coordinate matrix from B to E.
Proof. Let v = c1 v1 + c2 v2 + · · · + cn vn . Then
 
c1
 c2 
 
v = [v1 , v2 , . . . , vp ]  ..  = PB [v]B .
 . 
cp

6
Theorem 6.3. Let B = {v1 , v2 , . . . , vn } be a basis of a vector space V . Then the coordinate transformation,

V → Rn , v 7→ [v]B ,

is linear, one-to-one, and onto.

Proof. For vectors v, w of V and scalar a, if

v = c1 v1 + c2 v2 + · · · + cn vn ,

w = d1 v1 + d2 w2 + · · · + dn vn ,
then
v + w = (c1 + d1 )v1 + (c2 + d2 )v2 + · · · + (cn + dn )vn ,
av = a(c1 v1 + c2 v2 + · · · + cn vn ) = (ac1 )v1 + (ac2 )v2 + · · · + (acn )vn .
Thus      
c1 + d1 c1 d1
 c2 + d2   c2   d2 
     
[v + w]B =  .. = .. + .. ,
 .   .   . 
cn + dn cn dn
   
cc1 c1
   ..  = c[v] .
[cv]B =  ...  = c  .  B
ccn cn
So the coordinate transformation is a linear transformation.
Now for any vector  
c1
 c2 
 
 ..  in Rn ,
 . 
cn
consider the vector v = c1 v1 + c2 v2 + · · · + cn vn in V . The coordinate vector of v relative to B is
 
c1
 c2 
 
[v]B =  .  .
 .. 
cn

Thus the transformation is onto. The linear independence of {v1 , v2 , . . . , vn } implies that the transformation
is also one-to-one.
A one-to-one and onto linear transformation from a vector space V to a vector space W is called an
isomorphism.
Example 6.2. The vector space P3 of polynomials of degree at most 3 in variable t is isomorphic to the
vector space R4 , and {1, t, t2 , t3 } is a basis of P3 .

Proof. The map F : P3 → R4 , defined by

 
c0
¡ ¢  c1 
F [p(t)] = F c0 + c1 t + c2 t2 + c3 t3 =  
 c2  ,
c3

is a one-to-one linear transformation from P3 onto R4 .

7
Example 6.3. The vector space M(2, 2) of 2 × 2 matrices is isomorphic to R4 , and the matrices
· ¸ · ¸ · ¸ · ¸
1 0 0 1 0 0 0 0
, , ,
0 0 0 0 1 0 0 1

form a basis of M(2, 2). In fact, the map F : M(2, 2) → R4 , defined by

 
µ· ¸¶ a1
a1 a2  a2 
F (M ) = F = 
 a3  ,
a3 a4
a4

is a one-to-one linear transformation from M(2, 2) onto R4 .

Theorem 6.4. Let B = {v1 , v2 , . . . , vn } be a basis of a vector space V . Then any set of V consisting more
than n vectors are linearly dependent.
Proof. Let {u1 , u2 , . . . , up } be a set of vectors with p > n. Since any set of more than n vectors of Rn is
linearly dependent, the vectors [u1 ]B , [u2 ]B , . . . , [up ]B of Rn must be linearly dependent. Then there exist
constants c1 , c2 , . . . , cp , not all zero, such that

c1 [u1 ]B + c2 [u2 ]B + · · · + cp [up ]B = 0.

Thus
[c1 u1 + c2 u2 + · · · + cp up ]B = c1 [u1 ]B + c2 [u2 ]B + · · · + cp [up ]B = 0 = [0]B .
Note that the coordinate transformation is one-to-one. It follows that

c1 u1 + c2 u2 + · · · + cp up = 0.

This means that the vectors u1 , u2 , . . . , up are linearly dependent by definition.

Theorem 6.5. If B1 = {u1 , u2 , . . . , un } and B2 = {v1 , v2 , . . . , vp } are bases of a vector space V , then n = p.
Proof. Suppose n < p. By Theorem 6.4, {v1 , v2 , . . . , vp } is linearly dependent, contrary to the properties
for a basis. Thus n ≥ p. A similar argument shows that n ≤ p. Hence n = p.

7 Dimensions of vector spaces

Lecture 18

A vector space V is said to be finite dimensional if it can be spanned by a set of finite number of
vectors. The dimension of V , denoted by dim V , is the number of vectors of a basis of V . The dimension of
the zero vector space {0} is zero. If V cannot be spanned by any finite set of vectors, then V is said to be
infinite dimensional.
Theorem 7.1. Let H be a subspace of a finite dimensional vector space V . Then any linearly independent
subset of H can be expanded to a basis of H. Moreover, H is finite dimensional and dim H ≤ dim V .
Proof. Let S = {v1 , v2 , . . . , vk } be a set of linearly independent vectors of H. If Span S = H, then S is a
basis of H by definition. Otherwise, there exists a vector vk+1 in H such that vk+1 is not in Span S. Then
{v1 , v2 , . . . , vk , vk+1 } is a linearly independent set of H. Now set S = {v1 , v2 , . . . , vk+1 }. If Span S = H,
then S is a basis of H. Otherwise, continue to add one vector of H −Span S to S in this way until Span S = H.
Since H is of finite dimensional, the extension ends in finite number of steps.
Theorem 7.2 (Basis Theorem). Given a set S = {v1 , v2 , . . . , vn } of n vectors of an n-dimensional vector
space V .
(a) If {v1 , v2 , . . . , vn } is linearly independent, then {v1 , v2 , . . . , vn } is a basis of V .

8
 (b) If Span {v1 , v2 , . . . , vn } = V , then {v1 , v2 , . . . , vn } is a basis of V .
Proof. (a) By Theorem 7.1, S can be extended to a basis of V . Since S has n vectors and all bases have the
same number of vectors. It follows that no vectors were added to S to be extended a basis of V . Hence S
itself must be a basis.
(b) We need to show that S is linearly independent. Note that if S is not a basis, then S is linearly
dependent. Thus S contains a linearly independent proper subset S 0 such that Span S 0 = V . So S 0 is a basis
of V ; therefore #(S 0 ) ≥ n, contradict to #(S 0 ) < n.

8 Rank
Theorem 8.1. For any rectangular matrix A,

dim Row A = dim Col A = #(pivot positions of A).

Definition 8.2. The rank of a rectangular matrix A is the number pivot positions of A, that is, the
dimension of the row space and the column space of A. For a linear transformation T : V → W , the rank
of T is the dimension of the subspace T (V ).
Theorem 8.3 (Rank Theorem). For any m × n matrix A,

rank A + dim Nul A = n.

Proof. The rank of A is the number of pivot positions of A and the dimension of the null space of A is the
number of free variables of the system Ax = 0. It is clear that

#(pivot positions) + #(free variables) = n.

Theorem 8.4. Let A be an n × n invertible matrix. Then

dim Row A = dim Col A = rank A = n,

dim Nul A = 0.
Proof. The invertibility of A implies that the number of pivot positions of A is n. So rank A = n and
dim Nul A = 0.

9 Matrices of linear transformations

Definition 9.1. Let T : V → W be a linear transformation from a vector space V with basis B =
{v1 , v2 , . . . , vn } to a vector space W basis C = {w1 , w2 , . . . , wm }. Let


 T (v1 ) = a11 w1 + a21 w2 + · · · + am1 wm

 T (v2 ) = a12 w1 + a22 w2 + · · · + am2 wm
..

 .


T (vn ) = a1n w1 + a2n w2 + · · · + amn wm

writing more compactly,

 
a11 a12 ··· a1n
£ ¤ £ ¤
 a21 a22 ··· a2n  £
 ¤
T (v1 ), T (v2 ), . . . , T (vn ) = w1 , w2 , . . . , wm  .. .. .. ..  = w1 , w2 , . . . , wm A.
 . . . . 
am1 am2 ··· amn

9
The m × n matrix A is called the matrix of T relative to the basis B of V and the basis C of W .
Alternatively, the matrix A can be defined as
h£ £ £ i
A = T (v1 )]C , T (v2 )]C , . . . , T (vn )]C .

Consider the isomorphisms

T1 : V −→ Rn by T1 (v) = [v]B ,
(9.1)
T2 : W −→ Rm by T2 (w) = [w]C .
The composition
T −1 T T
T2 ◦ T ◦ T1−1 : Rn −→
1 2
V −→ W −→ Rm
is a linear transformation. For any vector x of Rn , let v be the vector of V whose coordinate vector is x,
i.e., [v]B = x, or £
v = x1 v1 + x2 v2 + · · · + xn vn = v1 , v2 , . . . , vn ]x.
Then
T (v) = T (x1 v1 + x2 v2 + · · · + xn vn )
= x1 T (v1 ) + x2 T (v2 ) + · · · + xn T (vn )
£ ¤
= T (v1 ), T (v2 ), . . . , T (vn ) x
£ ¤
= w1 , w2 , . . . , wm Ax.
The coordinate vector of T (v) relative to the basis C is Ax. Thus
¡ ¢ ³ ¡ ¢´ ¡ ¢ £ ¤
T2 ◦ T ◦ T1−1 (x) = T2 T T1−1 (x) = T2 T (v) = T (v) C = Ax.

This means that A is the standard matrix of the linear transformation T2 ◦ T ◦ T1−1 : Rn −→ Rm .
In particular, for a linear transformation T : Rn → Rm , T (x) = Ax. Let En = {e1 , e2 , . . . , en } be the
standard basis of Rn , and let Em = {e1 , e2 , . . . , em } be the standard basis of Rm . Then
£ ¤ £ ¤
T (e1 ), T (e2 ), . . . , T (en ) = e1 , e2 , . . . , em A.
Example 9.1. Let T : P3 → P2 be a linear transformation defined by
¡ ¢
T a0 + a1 t + a2 t2 + a3 t3 = (a0 + a3 ) + (a1 + a2 )t + (a0 + a1 + a2 + a3 )t2 .
It is clear that P3 has a basis © ª
B = 1, t, t(t + 1), t(t + 1)(t + 2)
and P2 has a basis © ª
C = 1, t, t(t − 1)
The matrix of T relative to the bases B and C can be found as follows:


 T (1) = 1 + t2 = 1 + t + t(t − 1)

¡ T (t)¢ = t + t2 = 2t + t(t − 1)
 ¡ T (t(t + 1)¢ =
 2t + 2t2 = 4t + 2t(t − 1)

T t(t + 1)(t + 2) = 1 + 5t + 6t2 = 1 + 11t + 6t(t − 1)
which is equivalent to
 
h ¡ ¢ ¡ ¢i £ ¤ 1 0 0 1
T (1), T (t), T t(t + 1) , T t(t + 1)(t + 2) = 1, t, t(t − 1)  1 2 4 11  .
1 1 2 6
Thus the matrix of T relative to the bases B and C is
 
1 0 0 1
 1 2 4 11  .
1 1 2 6

10
Theorem 9.2. Let V be an l-dimensional subspace of Rn with a basis B = {v1 , v2 , . . . , vl }, and let W be a
k-dimensional subspace of Rm with a basis C = {w1 , w2 , . . . , wk }. Let T : V → W be a linear transformation.
Then the matrix A of T relative to the bases B and C is given by
h ¯ i · I A ¸
w1 , w2 , . . . , wk ¯ T (v1 ), T (v2 ), . . . , T (vl ) ∼ k
.
0 0
n m
© ª
Corollary 9.3.
© Let T : R →
ª R be a linear transformation. Let B = v1 , v2 , . . . , vn be a basis of Rn ,
and let C = w1 , w2 , . . . , wm be a basis of Rm . If
h ¯ i h ¯ i
w1 , w2 , . . . , wm ¯ T (v1 ), T (v2 ), . . . , T (vn ) ∼ Im ¯ A ,

then A is the matrix of T relative to the bases B and C.

Example 9.2. Let V be the subspace of R4 defined by the linear equation x1 + x2 + x3 + x4 = 0, and let
W be the subspace of R3 defined by y1 + 2y2 + y3 = 0. Let T : R4 −→ R3 be defined by
¡ ¢
T (x1 , x2 , x3 , x4 ) = 3x1 + x2 , x2 + x3 , x3 + 3x4 .

(a) Show that T is a linear transformation from V to W .

(b) Verify that       

 1 1 1 
  −1   0   
B = v1 =     , v3 =  0 
  0  , v2 =  −1   0 

 

0 0 −1
is basis of V , and     
 0 1 
C = w1 =  1  , w2 =  −1 
 
−2 1
is basis of W .
(c) Find the matrix of T relative to the basis B of V and the basis C of W .
Solution. (a) Let [x1 , x2 , x3 , x4 ]T ∈ V , i.e., x1 + x2 + x3 + x4 = 0. We have

(2x1 + x2 ) + 2(x2 + x3 ) + (x3 + 2x3 ) = 3(x1 + x2 + x3 + x4 ) = 0.

So T is a well-defined linear transformation from V to W . (b) Trivial. (c) Consider the matrix
 
£ ¤ 0 1 2 3 3
w1 , w2 | T (v1 ), T (v2 ), T (v3 ) =  1 −1 −1 −1 0 ∼
−2 1 0 −1 −3
     
1 −1 −1 −1 0 1 −1 −1 −1 0 1 0 1 2 3
 0 1 2 3 3 ∼ 0 1 2 3 3 ∼ 0 1 2 3 3 .
−2 1 0 −1 −3 0 −1 −2 −3 −3 0 0 0 0 0
The matrix of T relative to the basis B and the basis C is
· ¸
1 2 3
.
2 3 3

11
10 Matrices of linear operators
Let V be an n-dimensional vector space with a basis B = {v1 , v2 , . . . , vn }. A linear transformation T : V → V
is called a linear operator on V . The matrix
h£ ¤ £ ¤ £ ¤ i
A = T (v1 ) B , T (v2 ) B , . . . , T (vn ) B
is called the matrix of T relative to the basis B.
Theorem 10.1. Let T : V → V be a linear transformation from a finite dimensional vector space V to
itself. Let B = {v1 , v2 , . . . , vn } and B 0 = {v10 , v20 , . . . , vn0 } be bases of V with the transition matrix P , that is,
£ 0 0 ¤ £ ¤
v1 , v2 , . . . , vn0 = v1 , v2 , . . . , vn P.
Let A be the matrix of T relative to the basis B, and let A0 be the matrix of T relative to the basis B 0 . Then
A0 = P −1 AP.
Corollary 10.2. Let T : Rn → Rn , T x = Ax, be a linear transformation. Let E = {e1 , e2 , . . . , en } be the
standard basis of Rn , and let B = {v1 , v2 , . . . , vn } be another basis. Then
£ ¤ £ ¤
v1 , v2 , . . . , vn = e1 , e2 , . . . , en P,
where P = [v1 , v2 , . . . , vn ], and the matrix of T relative to the basis B is
P −1 AP.
Example 10.1. Let T : R2 → R2 be the linear transformation given by
· ¸· ¸
7 −10 x1
T (x) = .
5 −8 x2
Find the matrix of T relative to the basis
½ · ¸ · ¸¾
1 2
B= v1 = , v2 = .
1 1
Solution. Since
· ¸· ¸ · ¸ · ¸
7 −10 1 −3 £ ¤ −3
T (v1 ) = = = −3v1 = v1 , v2 ,
5 −8 1 −3 0
· ¸· ¸ · ¸ · ¸
1 −10 2 4 £ ¤ 0
T (v2 ) = = = 2v2 = v1 , v2 ,
5 −8 1 2 2
then · ¸
£ ¡ ¤ £ ¤ −3 0
T v1 ), T (v2 ) = v1 , v2 .
0 2
· ¸
−3 0
The matrix of T relative to the basis B is .
0 2 · ¸
£ ¤ £ ¤ £ ¤ 1 2
Let E = {e1 , e2 }. Then v1 , v2 = e1 , e2 P , where P = v1 , v2 = . One verifies
1 1
· ¸−1 · ¸· ¸ · ¸
1 2 7 −10 1 2 −3 0
= .
1 1 5 −8 1 1 0 2
· ¸ · ¸
1 2 7 −10
We say that the matrix the matrix diagonalizes the matrix .
1 1 5 −8
Definition 10.3. An n × n matrix A is said to be similar to an n × n matrix B if there is an invertible
matrix P such that
P −1 AP = B.
Theorem 10.4. For a finite dimensional vector space V and a linear transformation T : V → V , the matri-
ces of T relative to various bases are similar. In other words, the matrices of the same linear transformation
from a vector space to itself under different bases are similar.

12
11 Change of basis
Let B = {u1 , u2 , . . . , un } and C = {v1 , v2 , . . . , vn } be bases of a vector space V . Then

v1 = p11 u1 + p21 u2 + · · · + pn1 un ,

v2 = p12 u1 + p22 u2 + · · · + pn2 un ,
..
.
vn = p1n u1 + p2n u2 + · · · + pnn un .

We may write this more compactly as

 
p11 p12 ··· p1n
£ ¤ £ ¤
 p21 p22 ··· p2n  £
 ¤
v1 , v2 , . . . , vn = u1 , u2 , . . . , un  .. .. ..  = u1 , u2 , . . . , un P.
 . . . 
pn1 pn2 ··· pnn
The matrix P is called the transition matrix from the basis B to the basis C. Clearly, the transition matrix
from the basis C to the basis B is the inverse matrix P −1 .
Let B = {u1 , u2 , . . . , uk } and C = {v1 , v2 , . . . , vk } be bases for a k-dimensional subspace V of Rn . Then
there is a k × k invertible matrix P such that
£ ¤ £ ¤
v1 , v2 , . . . , vk = u1 , u2 , . . . , uk P.

The matrix P is called the transition matrix from the basis B to the basis C. For a vector v in V , let x
and y be the coordinate vectors of v relative to the bases B and C, respectively, i.e.,
£ ¤ £ ¤
v = u1 , u2 , . . . , uk x = v1 , v2 , . . . , vk y.

Then £ ¤ £ ¤
u1 , u2 , . . . , uk x = u1 , u2 , . . . , uk P y.
It follows that
x = P y or y = P −1 x.
The matrix P −1 is also called the change-of-coordinate matrix from the basis B to the basis C.
To find the invertible matrix P , we set

P = [p1 , p2 , . . . , pk ], B = [u1 , u2 , . . . , uk ], C = [v1 , v2 , . . . , vk ]

Then
Bp1 = v1 , Bp2 = v2 , ..., Bpk = vk .
This means that p1 , p2 , . . . , pk are solutions of the linear systems

Bx = v1 , Bx = v2 , ..., Bx = vk

respectively. The linear systems can be solved simultaneously as follows:

· ¸
£ ¤ Ik P
u1 , u2 , . . . , uk | v1 , v2 , . . . , vk ∼ .
0 0

Example 11.1. Let V be a subspace of R5 with two bases

           

 6 5 3   
 1 2 2 


  7   0   
 6 
    

 

 2  
  1  
  1 

B=   2  ,  3  ,  −1  , C = 
    
 1 ,
  0 ,
  0  .


  


 4   3   2  




 2   1   1 


   
5 4 3 0 2 2

13
  
1
Let v be a vector of V whose coordinate vector relative to the basis B is  2 . Find the coordinate vector
3
of v relative to the basis C.
Solution. Performing the row operations to the n-by-n matrix [A|B], we have
 
1 2 1 6 5 3
 2 1 2 6 7 R2 − 2R1
 0 

 1 0 1 2 3 −1  R 3 − R1
  →
 2 1 0 4 3 2 
R4 − 2R1
0 2 1 5 4 3
 
1 2 1 6 5 3
 0 −3 0 −6 −3 −6  R2 /(−3)
 
 0 −2 0 −4 −2 −4  →
 
 0 −3 −2 −8 −7 −4  R3 ↔ R5
0 2 1 5 4 3
 
1 2 1 6 5 3
 0 R3 − 2R2
 1 0 2 1 2 
 R4 + 3R2
 0 2 1 5 4 3 
  →
 0 −3 −2 −8 −7 −4 
R5 + 2R2
0 −2 0 −4 −2 −4
 
1 2 1 6 5 3
 0 R4 + 2R3
 1 0 2 1 2 

 0 R1 − R 3
 0 1 1 2 −1  
 0 →
0 −2 −2 −4 2 
R1 − 2R2
0 0 0 0 0 0
 
1 0 0 1 1 0
 0 1 0 2 1 2 
 
 0 0 1 1 2 −1 
 
 0 0 0 0 0 0 
0 0 0 0 0 0
Then B = CP , where  
1 1 0
P = 2 1 2 
1 2 −1
So the coordinate vector of v relative to C is
    
1 1 0 1 3
 2 1 2  2  =  4 .
1 2 −1 3 1

Example 11.2. Let T : R3 → R3 be defined by

    
0 1 1 x1 x1
T (x) =  1 0 1   x2  , x =  x2  .
1 1 0 x3 x3

Consider two bases       

 1 1 0 
B = u1 =  1  , u2 =  0  , u3 =  −1  ,
 
0 1 2

14
       
 2 1 3 
C = v1 =  0  , v2 =  0  , v3 =  −1  .
 
3 2 6
Then  
£ ¤ £ ¤ 1 1 1
v1 , v2 , v3 = u1 , u2 , u3  1 0 2 .
1 1 2
Theorem 11.1. Let B = {v1 , v2 , . . . , vn } and B 0 = {v10 , v20 , . . . , vn0 } be bases of V with the transition matrix
P , that is, £ 0 0 ¤ £ ¤
v1 , v2 , . . . , vn0 = v1 , v2 , . . . , vn P.
Let C = {w1 , w2 , . . . , wm } and C 0 = {w10 , w20 , . . . , wm0
} be bases of W with the connection matrix Q, that is,
£ 0 0 0
¤ £ ¤
w1 , w2 , . . . , wm = w1 , w2 , . . . , wm Q.

Let T : V → W be a linear transformation with the matrix A relative to the bases B and C, then the matrix
of T relative to the bases B 0 and C 0 is given by

Q−1 AP.

Proof. Let A0 denote the matrix of T relative to the bases B 0 and C. Then
£ ¤ £ ¤ 0 £ ¤
T (v10 ), T (v20 ), . . . , T (vn0 ) = w10 , w20 , . . . , wm
0
A = w1 , w2 , . . . , wm QA0 .

On the other hand, since

v10 = p11 v1 + p21 v2 + · · · + pn1 vn

v20 = p12 v1 + p22 v2 + · · · + pn2 vn
..
.
vn0 = p1n v1 + p2n v2 + · · · + pnn vn

then

T (v10 ) = p11 T (v1 ) + p21 T (v2 ) + · · · + pn1 T (vn )

T (v20 ) = p12 T (v1 ) + p22 T (v2 ) + · · · + pn2 T (vn )
..
.
0
T (vn ) = p1n T (v1 ) + p2n T (v2 ) + · · · + pnn T (vn )

Thus £ ¤ £ ¤ £ ¤
T (v10 ), T (v20 ), . . . , T (vn0 ) = T (v1 ), T (v2 ), . . . , T (vn ) P = w1 , w2 , . . . , wm AP.
Therefore QA0 = AP , that is,
A0 = Q−1 AP.

Example 11.3. Let T : P3 → P2 be a linear transformation defined by

T (a0 + a1 t + a2 t2 + a3 t3 ) = (a0 + a3 ) + (a1 + a2 )t + (a0 + a1 + a2 + a3 )t2 .

Then the matrix of T relative to the bases

B = {1, t, t(t + 1), t(t + 1)(t + 2)} and C = {1, t, t(t − 1)}

is the matrix  
1 0 0 1
 1 2 4 11  .
1 1 2 6

15
Given new bases © ª © ª
B 0 = 1, t, t(t − 1), t(t − 1)(t − 2) and C 0 = 1, t, t(t + 1) .
Since 

 T (1) = 1 + t2 = 1 − t + t(t + 1)

¡ T (t)¢ = t + t2 = t(t + 1)
,

 T t(t − 1)¢ = 0 = 0
 ¡
T t(t − 1)(t − 2) = 1−t = 1−t
the matrix of T relative to the bases B 0 and C 0 is
 
1 0 0 1
 −1 0 0 −1  .
1 1 0 0

Note that
 
h ¡ ¢ ¡ ¢i £ ¤ 1 0 0 1
T (1), T (t), T t(t + 1) , T t(t + 1)(t + 2) = 1, t, t(t − 1)  1 2 4 11  ,
1 1 2 6
 
1 0 0 0
£ ¤ £ ¤  0 1 −2 6 
1, t, t(t − 1), t(t − 1)(t − 2) = 1, t, t(t + 1), t(t + 1)(t + 2)  0 0
,
1 −6 
0 0 0 1
 
£ ¤ £ ¤ 1 0 0
1, t, t(t + 1) = 1, t, t(t − 1)  0 1 2  .
0 0 1
One verifies that
 
   −1   1 0 0 0
1 0 0 1 1 0 0 1 0 0 1  0 1 −2 6 
 −1 0 0 −1  =  0 1 2   1 2 4 11  
 0 0
.
1 −6 
1 1 0 0 0 0 1 1 1 2 6
0 0 0 1

16