Chapter 3: Linear Maps

Linear Algebra Done Right (4th Edition), by Sheldon Axler

Last updated: September 4, 2024

Contents
3A: Vector Space of Linear Maps 2
3A Problem Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

3B: Null Spaces and Ranges 7

3B Problem Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

3C: Matrices 13
3C Problem Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

3D: Invertibility and Isomorphisms 23

3D Problem Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3E: Products and Quotients of Vector Spaces 30

3E Problem Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

3F: Duality 40
3F Problem Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

1
3A: Vector Space of Linear Maps
Definition 1 (Linear Map). A linear map from V to W is a function T : V → W
with the following properties:
ˆ Additivity: T (u + v) = T u + T v for all u, v ∈ V .
ˆ Homogeneity: T (λv) = λT (v) for all λ ∈ F and v ∈ V .
Notation: L(V, W ), L(V )

ˆ The set of linear maps from V to W is denoted by L(V, W ).

ˆ The set of linear maps from V to V is denoted by L(V ).

Lemma 2 (linear map basis lemma). Suppose v1 , . . . , vn is a basis of V and

w1 , . . . , wn ∈ W . Then there exists a unique linear map T : V → W such that

T vk = wk

for each k = 1, . . . , n.

Definition 3 (additional and scalar multiplication on L(V, W )). Suppose S, T ∈

L(V, W ) and λ ∈ F. The sum S + T and the product λT are the linear maps
from V to W defined by

(S + T )(v) = Sv + T v and (λT )(v) = λ(T v)

for all v ∈ V .
Remark 4. L(V, W ) is a vector space.
Definition 5 (product of linear maps). If T ∈ L(U, V ) and S ∈ L(V, W ), then
the product ST ∈ L(U, W ) is defined by

(ST )(u) = S(T u)

for all u ∈ U .
Remark 6 (algebraic properties of product of linear maps). We have associa-
tivity, identity, and distributive properties whenever such properties are defined.

Theorem 7 (linear maps take 0 to 0). Suppose T is a linear map from V to W .

Then T (0) = 0.

2
 Problem 1
Suppose b, c ∈ R. Define T : R3 → R2 by

T (x, y, z) = (2x − 4y + 3z + b, 6x + cxyz)

Show that T is linear if and only if b = c = 0.

Proof. T is linear ⇐⇒ T (x, y, z = 0) ⇐⇒ b = 0

In addition, let’s only consider the second coordinate, then we have

T (x1 + x2 , y1 + y2 , z1 + z2 )2 = 6(x1 + x2 ) + c(x1 + x2 )(y1 + y2 )(z1 + z2 )

which only equals T (x1 , y1 , z1 ) + T (x2 , y2 , z2 ) if c = 0.

Problem 3
Suppose that T ∈ L(Fn , Fm ). Show that there exist scalars Aj,k ∈ F for
j = 1, . . . , m and k = 1, . . . , n such that

T (x1 , . . . , xn ) = (A1,1 x1 + · · · + A1,n xn , . . . , Am,1 x1 + · · · + Am,n xn )

for every (x1 , . . . , xn ) ∈ Fn .

Proof. Let {u1 , . . . , un } denote the standard basis of Fn . We have that

T ui = (A1,i , . . . , Am,i )

Take arbitrary (x1 , . . . , xn ) ∈ Fn , we have that

T xi ui = xi (A1,i , . . . , Am,i )

thus we have that

T (x1 , · · · , xn ) = (A1,1 x1 + · · · + A1,n xn , · · · , Am,1 x1 + · · · + Am,n xn )

Problem 4
Suppose T ∈ L(V, W ) and v1 , . . . , vm is a list of vectors in V such that
T v1 , . . . , T vm is a linearly independent list in W . Prove that v1 , . . . , vm
is linearly independent.

Pm Pm
Proof. This means that the only solution to i=1 ai T (vi ) = i=1 T (ai vi ) = 0
is all ai = 0. However, as we know T (0) = 0 so ai vi = 0 and thus we’ve proved
the claim.

3
 Problem 7
Show that every linear map from a one-dimensional vector space to itself
is multiplication by some scalar. More precisely, prove that if dim V = 1
and T ∈ L(V ), then there exists λ ∈ F such that T v = λv for all v ∈ V .

Proof. Since dim V = 1, every v ∈ V can be expressed as λv ′ for some other

v ′ ∈ V . As T v ∈ V , we have T v = βv. Then T v = T λv ′ = βλv ′ = βv
Problem 8
Give an example of a function ϕ : R2 → R such that

ϕ(av) = aϕ(v)

for all a ∈ R and all v ∈ R2 but ϕ is not linear.

Proof. Consider f (x1 , x2 ) = x21 /x2 if x2 ̸= 0 o.w. 0, then f (a(x1 , x2 )) =

ax21 = af (x1 , x2 ). However, f ((x1 + y1 , x2 + y2 )) = (x1 + y1 )2 /(x2 + y2 ) ̸=
f (x1 , x2 ) + f (y1 , y2 ).

Problem 11
Suppose V is finite-dimensional and T ∈ L(V ). Prove that T is a scalar
multiple of the identity if and only if ST = T S for all S ∈ L(V ).

Proof. ⇒ We can express T = λI. Then S(λI) = λS = λIS = PT S.

m
⇐ Let v1 , . . . , vm be a basis of V . Pick Si such that Si ( i=1 ai vi ) = ai vi ,
which is clearly a linear operator. Then we have that
Si T (v) = T Si (v)
m
X
Si bj vj = T (ai vi )
j=1

bi vi = ai T (vi )
This shows that for all vi , there exists λi such that T (vi ) = λi vi . Next
we P
show that such λi does not depend on i. Construct Sij subtly such that
n
Sij k=1 ak vk = aj vi + ai vj .
Then we have that
Sij T v = T Sij v
 
n
X
Sij  λk ak vk  = T (aj vi + ai vj )
k=1

λj aj vi + λi ai vj = λi aj vi + λj ai vj
This shows that λi = λj for all i, j and thus we’ve shown that T = λI for
some λ.

4
 Problem 12
Suppose U is a subspace of V with U ̸= V . Suppose S ∈ L(U, W ) and
S ̸= 0. Define T : V → W by
(
Sv if v ∈ U
Tv =
0 if v ∈ V and v ∈
/U
Prove that T is not a linear map on V .

Proof. Take u ∈ U such that u ∈

/ Null(S). Take v ∈ V \ U , then u + v ∈ V \ U .
This means that

T (u + v) = S(u + v) = 0 ̸= T (u) + T (v) = Su

Problem 13
Suppose V is finite-dimensional. Prove that every linear map on a
subspace of V can be extended to a linear map on V . In other words,
show that if U is a subspace of V and S ∈ L(U, W ), then there exists
T ∈ L(V, W ) such that T u = Su for all u ∈ U .

Proof. Note that there exists subspace P s.t. V = P ⊕U . For all v ∈ V , v = p+u
for some p, u. Then define T (v) = Su + p. Clearly we have that T (u) = Su for all
u ∈ U . It now only suffices to prove T is a linear map. Homogeneity is trivial to
show. For additivity, T (v1 + v2 ) = T (u1 + u2 + p1 + p2 ) = S(u1 + u2 ) + p1 + p2 =
(Su1 + p1 ) + (Su2 + p2 ) = T v1 + T v2 .

Problem 14
Suppose V is finite-dimensional with dim V > 0, and suppose W is
infinite-dimensional. Prove that L(V, W ) is infinite-dimensional.

Proof. Recall the definition of infinite-dimension from ch2 P17:

V is infinite-dimensional if and only if there is a sequence v1 , v2 , . . . of vectors in
V such that v1 , . . . , vm is linearly independent for every positive integer m.

W being infinite-dimensional implies this. Denote the sequence by w1 , w2 , . . .

Let v1 , . . . , vm be the basis for V . Define a sequence of linear operators as follows:
Tk ∈ L(V, W ) such that Tk (v) = vk . Then we have that for every positive integer
m, the solution for the following equation is all ai = 0.

a1 T1 (v) + · · · + am Tm (v) = a1 v1 + · · · + am vm = 0

This shows that L(V, W ) is infinite-dimensional.

5
 Problem 15
Suppose v1 , . . . , vm is a linearly dependent list of vectors in V . Suppose
also that W ̸= {0}. Prove that there exist w1 , . . . , wm ∈ W such that no
T ∈ L(V, W ) satisfies T vk = wk for each k = 1, . . . , m.

Proof. Take w1 , . . . , wm to be linearly independent list of vectors in W . Then

we have that
T a1 v1 + · · · T am vm = a1 w1 + · · · am wm = 0
The only solution is that ai = 0 for all i, but this contradicts that v1 , . . . , vm
is linearly dependent.

Problem 16
Suppose V is finite-dimensional with dim V > 1. Prove that there exist
S, T ∈ L(V ) such that ST ̸= T S.

Pm
Proof.
Pm Let v1 , . . . , vm be the basis of V . Define S(v) = S( i=1 ai vi ) =
i=1 am−i vi and T (v) = a1 v1 . Then we have that

ST v = Sa1 v1 = a1 v1

but
m
X
T Sv = T am−i vi = am v1
i=1

Problem 17
Suppose V is finite-dimensional. Show that the only two-sided ideas
of L(V ) are {0} and L(V ), where we define that a subspace E of L(V )
is called two-sided ideal if T E ∈ E and ET ∈ E for all E ∈ E and all
T ∈ L(V ).

Proof. It’s easy to verify that {0} and L(V ) are two-sided ideal of L(V ). Suppose
for the sake of contradiction that such E exists. Let e1 , . . . , em beP its basis and
n
let
Pn e 1 , . . . em , em+1 , . . . , e n be the basis for V . Define T (v) = T (
Pmi=1 ai ei ) =
i=1 ai en−i to be a linear map in V (ez to verify), and Ej (u) = Ej ( k=1 bk ek ) =
bj ej . We have reached the contradicting example such that T Ej (v) = aj en−j ∈
V \ E.

6
3B: Null Spaces and Ranges
Definition 8 (null space). For T ∈ L(V, W ), the null space of T , denoted by
null T , is the subset of V consisting of those vectors that T maps to 0:
null T = {v ∈ V : T v = 0}
Corollary 9. Suppose T ∈ L(V, W ), then null T is a subspace of V .
Definition 10 (injective). A function T : V → W is called injective if T u = T v
implies u = v.
Theorem 11. Let T ∈ L(V, W ). Then T is injecive if and only if null T = {0}.
Definition 12 (range). For T ∈ L(V, W ), the range of T is the subset of W
consisting of those vectors that are equal to T v for some v ∈ V :
range T = {T v : v ∈ V }
Corollary 13. Suppose T ∈ L(V, W ), then range T is a subspace of W .
Definition 14 (surjective). A function T : V → W is called surjective if its
range equals W .
Theorem 15 (fundamental theorem of linear map). Suppose V is finite-
dimensional and T ∈ L(V, W ). Then range T is finite-dimensional and
dim V = dim null T + dim range T
Proof. Let u1 , . . . , um be a basis of null T and let u1 , . . . , um , v1 , . . . , vn be a
basis of V . It now suffices to prove dim range T = n. Let v ∈ T , then
m
X n
X
v= ai ui + bj v j
i=1 j=1

We apply T on both sides to get that

n
X
Tv = bj T vj
j=1

which shows that T v1 , . . . , T vn spans range T and thus it’s finite-dimensional.

To show they are linearly independent, we have
n
X n
X
bj T vj = T bj v j = 0
j=1 j=1

The only solution is that all bj = 0.

Corollary 16 (linear map to a lower-dimen space is not injective). Suppose V
and W are finite-dimensional vector spaces such that dim V > dim W . Then no
linear map from V to W is injective.
Corollary 17 (linear map to a higher-dimen space is not surjective). Suppose
V and W are finite-dimensional vector spaces such that dim V < dim W . Then
no linear map from V to W is surjective.

7
Application Consider the system of linear equation defined by the map
T : Fn → Fm :
 
n
X n
X
T (x1 , . . . , xn ) =  A1,k xk , . . . , Am,k xk 
k=1 k=1

Corollary 18 (homogeneous systems of linear equations). A homogeneous sys-

tem of linear equations with more variables than equations has nonzero solutions.
Corollary 19 (inhomogeneous system fo linear equations). An inhomogeneous
system of linear equations with more equations than variables has no solution
for some choice of the constant terms.

Problem 1
Give an example of a linear map T with dim null T = 3 and dim range
T = 2.

Proof. Consider T (x1 , x2 , x3 , x4 , x5 ) = (0, 0, 0, x4 , x5 ).

It’s easy to verify that dim null T = 3 and applying the theorem of linear
map solves the problem.

Problem 2
Suppose S, T ∈ L(V ) are such that range S ⊆ null T . Prove that
(ST )2 = 0.

Proof. This means that take x ∈ V , T (S(x)) = 0. We have (ST )2 = ST ST =

0.

Problem 3
Suppose v1 , . . . , vm is a list of vectors in V . Define T ∈ L(Fm , V ) by

T (z1 , . . . , zm ) = z1 v1 + · · · + zm vm

(a) what property of T corresponds to v1 , . . . , vm spanning V ?

(b) what property of T corresponds to v1 , . . . , vm being linearly indepen-
dent?

Proof. (a) surjective (range = V )

(b) injective (z1 , . . . , zm is identically zero if and only if z1 v1 + · · · + zm vm =
0)

Problem 4
Show that {T ∈ L(R5 , R4 ) : dim null T > 2} is not a subspace of
L(R5 , R4 ).

8
Proof. Try to come up with a counterexample. Consider f (e1 , e2 , e3 , e4 , e5 ) =
(e1 , 0, 0, e4 ) and g(e1 , e2 , e3 , e4 , e5 ) = (0, e2 , e3 , 0). Then we have that
null f = {0, e2 , e3 , 0, e5 } and null g = {e1 , 0, 0, e4 , e5 } so both in {T ∈
L(R5 , R4 ) : dim null T > 2}. However, f + g(e1 , e2 , e3 , e4 , e5 ) = {(e1 , e2 , e3 , e4 )}
and this means their dim null = 5 - 4 = 1 ¡ 3.

Problem 7
Suppose V and W are finite-dimensional and 2 ≤ dim V ≤ dim W . Show
that {T ∈ L(V, W ) : T is not injective} is not a subspace of L(V, W ).

Proof. T is not injective means that dim V > dim W , which contradicts the
assumption in the question.

Problem 9
Suppose T ∈ L(V, W ) is injective and v1 , . . . , vn is linearly independent
in V . Prove that T v1 , . . . , T vn is linearly independent in W .

Proof.
n
X n
X
zi vi = 0 ⇐⇒ zi T (vi ) = 0
i=1 i=1

Problem 10
Suppose v1 , . . . , vn spans V and T ∈ L(V, W ). Show that T v1 , . . . , T vn
spans range T .

Pn
Proof. Take v in V , then we know v = i=1 ai vi . Take win range Pn T . Then
we
Pn know there exists v ∈ V such that w = T (v) and thus w = T ( i=1 ai vi ) =
i=1 ai T (vi ).

Problem 11
Suppose that V is finite-dimensional and that T ∈ L(V, W ). Prove that
there exists a subspace U of V such that

U ∩ null T = {0} and range T = {T u : u ∈ U}

Proof. Since we know null T is a subspace of V , then there exists U such that
U ⊕ null T = V . We can define U to be such case. To finish the proof, let
v = u + t, u ∈ U, t ∈ nullT , then rangeT = {T (u + t) = T u : u ∈ U}.

Problem 16
Suppose V and W are both finite-dimensional. Prove that there exists
an injective linear map from V to W if and only if dim V ≤ dim W .

9
Proof. ⇒ T ∈ L(V, W ) is injective. Then this means that dim V =
dim range T ≤ dim W .
⇐ By assumption, we can define a linear map T such that T vi = wi where
vi , wi are the respective basis of V and W . Then we have that
n
X n
X
ai T vi = ai wi = 0 ⇐⇒ ai are identically zero
i=1 i=1

This means that null T is {0} and thus it is injective.

Problem 17
Suppose V and W are both finite-dimensional. Prove that there exists a
surjective linear map from V onto W if and only if dim V ≥ dim W .

Proof. ⇒ dim V ≥ range T = dim W .

⇐ By assumption we can define a linear map T such that T vi = wi for
1 ≤ i ≤ dim W and T vi = 0 for dim W ≤ i dim V . This mean that take w ∈ W ,
Pdim W
then w = i=1 ai wi . At the same time, we know that for all wi , there exists vi
Pdim W
s.t. T vi = wi , therefore we have w = i=1 ai T vi and thus T is surjective.

Problem 18
Suppose V and W are finite-dimensional and that U is a subspace of V .
Prove that there exists T ∈ L(V, W ) such that null T = U if and only if
dim U ≥ dim V − dim W .

Proof. We know dim V = null T + range T and range T ≤ dim W . Therefore,

dim U = dim null T ≥ dim V − dim W .

Problem 19
Suppose W is finite-dimensional and T ∈ L(V, W ). Prove that T is
injective if and only if there exists S ∈ L(W, V ) such that ST is the
identity operator on V .

Proof. T is injective ⇐⇒ T (v) is unique for v ∈ V . ⇐⇒ We can define S : ST v =

v.

Problem 21
Suppose V is finite-dimensional, T ∈ L(V, W ), and U is a subspace of W .
Prove that T = {v ∈ V : T v ∈ U} is a subspace of V and

dim T = dim null T + dim(U ∩ range T )

Proof. To prove subspace, we can simply follow by definition. We can define

S ∈ L(T , U) such that Sv = T v for all v ∈ T . Then we have that range(S) ∈ U∩
range(T ) and that null T = null S. Hence we have proved the claim.

10
 Problem 22
Suppose U and V are finite-dimensional vector spaces and S ∈ L(V, W )
and T ∈ L(U, V ). Prove that

dim null ST ≤ dim null S + dim null T

Proof. We can write that

null ST = {u ∈ U : T (u) ∈ null (S)}
By the previous question, we know that
dim null ST = dim null T + dim null S ∩ range T ≤ dim null T + dim null S

Problem 23
Suppose U and V are finite-dimensional vector spaces and S ∈ L(V, W )
and T ∈ L(U, V ). Prove that

dim range ST ≤ min{dim range S, dim range T }

Proof. First note that dim range ST ≤ dim range S. This because range ST ⊆
range S.
To prove dim range ST ≤ dim range T , we have that dim U = dim null T +
dim range T = dim null ST + dim range ST . Since we know dim null T ≥
dim null ST , dim range ST ≤ dim range T .

Problem 24
(a) Suppose dim V = 5 and S, T ∈ L(V ) are such that ST = 0. Prove
that dim range T S ≤ 2.
(b) Give an example of S, T ∈ L(F5 ) with ST = 0 and dim range T S = 2.

Proof. (a) dim null ST = 5 ≤ dim null T + dim null S

We also know that 5 = dim null T + dim range T = dim range S + dim null S
and thus we have that dim range T + dim range S ≤ 5. This implies that
min{dim range T, dim range S} ≤ 2. Hence, by applying P23, we finish the
proof.
(b) Consider T (x1 , x2 , x3 , x4 , x5 ) = (x3 , x4 , 0, 0, 0), S(x1 , x2 , x3 , x4 , x5 ) =
(0, 0, x3 , x4 , x5 ). Then we have that ST (x1 , x2 , x3 , x4 , x5 ) = S(x3 , x4 , 0, 0, 0) =
(0, 0, 0, 0, 0) while T S(x1 , x2 , x3 , x4 , x5 ) = T (0, 0, x3 , x4 , x5 ) = (x3 , x4 , 0, 0, 0).

Problem 25
Suppose that W is finite-dimensional and S, T ∈ L(V, W ). Prove that
null S ⊆ null T if and only if there exists E ∈ L(W ) such that T = ES.

11
Proof. null S ⊆ null T ⇒ dim range T ≤ dim range S ⇒ let s1 , . . . , sn be basis
of range S and t1 , . . . , tm be basis of range T where m ≤ n. We can always
define E ∈ L(W ) such that E(si ) = ti for all 1 ≤ i ≤ m and 0 otherwise.
The other direction is trivial.

Problem 26
Suppose that V is finite-dimensional and S, T ∈ L(V, W ). Prove that
range S ⊆ range T if and only if there exists E ∈ L(V ) such that S = T E.

Proof. range S ⊆ range T ⇒ Take v1 , . . . , vm to be the basis of V , define the

linear map Svi = si for si ∈ range S ⊆ range T . Then there exists u1 , . . . , um
such that T ui = si for all i. Then we can define the linear map Evi = ui for all
i such that S = T E.
The other direction is trivial.

Problem 27
Suppose P ∈ L(V ) and P 2 = P . Prove that V = null P ⊕ range P .

Proof. Take v ∈ null (P ) ∩ range (P ). Then this means that P (v) = 0 and there
exists u s.t. v = P (u). At the same time, P (v) = P (P (u)) = P (u) = 0 = v. So
we have that null P ∩ range P = {0}. Since P is defined on L(P ), we have that
V = null P ⊕ range P .

12
3C: Matrices
Definition 20 (matrix, Aj,k ). Suppose m and n are nonnegative integers. An
m-by-n matrix matrix A is a rectangular array of elements of F with m rows
and n columns:  
A1,1 · · · A1,n
 . .. 
A=  .. . 

Am,1 ··· Am,n

where the notation Aj,k denotes the entry in row j and column k.
Definition 21 (matrix of a linear map, M(T )). Suppose T ∈ L(V, W ) and
v1 , . . . , vn is a basis of V and w1 , . . . , wm is a basis of W . The matrix of T with
respect to these bases is the m-by-n matrix M(T ) whose entries Aj,k are defined
by
T vk = A1,k w1 + · · · + Am,k wm
.
If the bases v1 , . . . , vn and w1 , . . . , wm are not clear from the context, then
the notation M(T, (v1 , . . . , vn ), (w1 , . . . , wm )) is used.
Remark 22. The k-th column of M(T ) consists of the scalars needed to write
T vk as a linear combination of w1 , . . . , wm :
m
X
T vk = Aj,k wj
j=1

Remark 23. If T is a linear map from n-dimensional vector space to an m-

dimensional vector space, then M(T ) is an m-by-n matrix.
Corollary 24 (Matrix addition and scalar multiplication). Suppose S, T ∈
L(V, W ) and λ ∈ F. Then M(λS + T ) = λM(S) + M(T ).

For m and n positive integers, the set of all m-by-n matrices with entries in
F is denoted by Fm,n .
Theorem 25. Suppose m and n are positive integers. Fm,n is a vector space of
dimension mn.
Definition 26 (matrix multiplication). Suppose A is an m-by-n matrix and B
is an n-by-p matrix. Then AB is defined to be the m-by-p matrix whose entry in
row j, column k, is given by the equation
n
X
(AB)j,k = Aj,r Br,k
r=1

In words, the entry in row j, column k of AB is computed by taking row j of A

and column k of B.

13
 Motivations. Let v1 , . . . , vn to be the basis of V , w1 , . . . , wm to be the basis
of W , and u1 , . . . , up to be the basis of U . Consider linear maps T : U → V and
S : V → W . Suppose M(T ) = A and M(S) = B. For 1 ≤ k ≤ p, we have

 
Xn
(ST )uk = S  Br,k vr 
r=1
n
X
= Br,k (Svr )
r=1
Xn m
X
= Br,k Aj,r wj
r=1 j=1
 
m
X Xn
=  Aj,r Br,k  wj
j=1 r=1

Theorem 27 (matrix of product of linear maps). If T ∈ L(U, V ) and S ∈

L(V, W ), then M(ST ) = M(S)M(T ).
Definition 28 (Aj,· , A·,k ). Suppose A is an m-by-n matrix
ˆ If 1 ≤ j ≤ m, then Aj,· denotes the 1-by-n matrix consisting of row j of
A.
ˆ If 1 ≤ k ≤ n, then A·,k denotes the m-by-1 matrix consisting of column k
of A.
Corollary 29 (entry of matrix product equals row times column). Suppose A
is an m-by-n matrix and B is an n-by-p matrix. Then

(AB)j,k = Aj,· B·,k

if 1 ≤ j ≤ m and 1 ≤ k ≤ p. In other words, the entry in row j, column k, of

AB equals (row j of A) times (column k of B).
Corollary 30 (column of matrix product equals matrix times column). Suppose
A is an m-by-n matrix and B is an n-by-p matrix. Then

(AB)·,k = AB·,k

if 1 ≤ k ≤ p. In other words, column k of AB equals A times column k of B.

Corollary
 31(linear combination of columns). Suppose A is an m-by-n matrix
b1
.
 ..  is an n-by-1 matrix. Then
and b =  
bn

Ab = b1 A·,1 + · · · + bn A·,n .

14
In other words, Ab is a linear combination of the columns of A, with the scalars
that multiply the columns coming from b.
Theorem 32 (matrix multiplication as linear combination of columns). Suppose
C is an m-by-c matrix and R is a c-by-n matrix.
ˆ If k ∈ {1, . . . , n}, then column k of CR is a linear combination of the
columns of C, with the coefficients of this linear combination coming from
columns k of R.
ˆ If j ∈ {1, . . . , m}, then row j of CR is a linear combination of the rows of
R, with the coefficients of this linear combination coming from row j of C.

Definition 33 (column rank, row rank). Suppose A is an m-by-n matrix with

entries in F.
ˆ The column rank of A is the dimension of the span of the columns of A in
Fm,1 .
ˆ The row rank of A is the dimension of the span of the rows of A in F1,n .
Definition 34 (transpose, A⊤ ). The transpose of a matrix A, denoted by A⊤ ,
is the matrix obtained from A by interchanging rows and columns. Specifically,
if A is an m-by-n matrix, then A⊤ is the n-by-m matrix whose entries are given
by the equation
(A⊤ )k,j = Aj,k
Lemma 35 (column-row factorization). Suppose A is an m-by-n matrix with
entries in F and column rank c ≥ 1. Then there exists an m-by-c matrix C and
a c-by-n matrix R, both with entries in F, such that A = CR.
Proof. Each column of A is an m-by-1 matrix. The list A·,1 , . . . , A·,n of columns
of A can be reduced to a basis of the span of the columns of A. This basis has
length c. The c columns in this basis can be put together to form an m-by-c
matrix C.
If k ∈ {1, . . . , n}, then column k of A is a linear combination of the columns
of C. Make the coefficients of this linear combination into column k of a c-by-n
matrix that we call R. Then A = CR.
Theorem 36 (column rank equals row rank). Suppose A ∈ Fm,n . Then the
column rank of A equals the row rank of A.
Proof. Let c denote the column rank of A. Let A = CR be the column-row
factorization of A given by the proof before, where C is m-by-c and R is c-by-n.
Then the column-row factorization lemma tells us that every row of A is a linear
combination of the rows of R. Because R has c rows, this implies that the row
of A is less than or equal to the column rank c of A.
To prove the other direction, we can do the same thing to A⊤ and then we
can get that column rank of A = row rank of A⊤ ≤ column rank of A⊤ = row
rank of A which we proved above.

15
 Problem 1
Suppose T ∈ L(V, W ). Show that with respect to each choice of bases of
V and W , the matrix of T has at least dim range T nonzero entries.

Proof. Let v1 , . . . , vn be the basis of V and w1 , . . . , wm be the basis of W .

Suppose there exists a matrix A = M(T ) of T has less than dim range T nonzero
entries. This means that A has at most dim range T − 1 nonzero columns so
this implies that dim range T < dim range − 1 which forms a contradiction.

Problem 2
Suppose V and W are finite-dimensional and T ∈ L(V, W ). Prove that
dim range T = 1 if and only if there exists a basis of V and a basis of W
such that with respect to these bases, all entries of M(T ) equal 1.

Proof. ⇒ We solve this problem through careful construction of the basis. That
is, we will construct basis v1 , . . . , vn of V and w1 , . . . , wm of W such that
T vi = w1 + · · · + wm for all i. We can achieve this again mainly because
dim range T = 1. let u1 , . .P . , um be a set of arbitrary basis of W . Take w ∈
m
range T so we have w = i=1 ai ui . We consider all the coefficients ai as
follows: take an index set I such that for all i ∈ I, ai = ̸ 0 and we have
|I| = c. Take arbitrary j ∈ I and we construct the basis as follows: let
wj = (aj − (m − r))uj , wk = ak uk P for all k ∈ I, k = ̸ j, and wk = uk + uj for
m
all k ∈
/ I. Then we have that w = l=1 wl . Since we know w ∈ range T , then
there exists v1 ∈ V s.t. T (v1 ) = w. Let v2 , . . . , vn be the basis of null T . Since
we know dim V = null T + 1 so v1 , v2 , . . . , vn constitutes a basis of V . In this
way, we successfully constructs the basis such that all entries of M(T ) equal 1.
⇐ There exists basis v1 , . . . , vn of V and w1 , . . . , wm of W such that all entries
of M(T ) equal 1. This means that T v1 = · · · = T vn = w1 + · · · + wm . So we first
have that dim range T ̸= 0. To proveP dim range T = P1, take arbitrary u1 , u2 ∈
n n
range T . ThenP we know that u1 = T ( i=1 ai vi ) = i=1 ai (w1 + · · · + wn ) and
n
similarly u2 = i=1 bi (w1 + · · · + wn ). Here, we have that
   
Xn Xn
u1 =  ai  /  bi  u2
i i

and thus dim range T has to be 1-dimensional since every two arbitrary vector
is simply a scalar multiple of each other.

Problem 3
Suppose v1 , . . . , vn is a basis of V and w1 , . . . , wm is a basis of W .
Show that if S, T ∈ L(V, W ) and λ ∈ F, then M(λS + T ) = λM(S) +
M(T ).

16
Proof.

(λS + T )vk = λS(vk ) + T (vk )

Xm X m
=λ Ai,k wi + Bi,k wi
i=1 i=1

Problem 5
Suppose V and W are finite-dimensional and T ∈ L(V, W ). Prove that
there exists a basis of V and a basis of W such that with respect to these
bases, all entries of M(T ) are 0 except that the entries in row k, column
k equal 1 if 1 ≤ k ≤ dim range T .

Proof. Let dim V = n, dim W = m, dim range T = k. The main idea of the
proof is to construct basis v1 , . . . , vn and w1 , . . . , wm such that T vi = wi for all
1 ≤ i ≤ k and zero otherwise. (So the constructed T satisfies the requirement).
We proceed with obtaining the basis w1 , . . . , wk of range T and extend this
w1 , . . . , wk , . . . , wm to the basis of W . We know for all 1 ≤ i ≤ k, there exists
vi ∈ V s.t. T vP i = wi . We claimPthat v1 , . . . , vk P
are linearly independent. To
k k k
prove this, see j=1 aj vj = 0 ⇒ j=1 aj T (vj ) = j=1 aj wj = 0. Similar to the
proof in the fundamental theorem of linear map, we extend the basis to V through
considering the null space. We further claim that (let K = span (v1 , . . . , vk ))
V = K ⊕ null T (which is easy to prove). Hence, extending the basis from K
with the basis from null T completes the proof.

Problem 6
Suppose v1 , . . . , vm is a basis of V and W is finite-dimensional. Suppose
T ∈ L(V, W ). Prove that there exists a basis w1 , . . . , wn of W such that
all entries in the first column of M(T ) [with respect to these bases] are 0
except for possibly a 1 in the first row, first column.

Proof. Let w1 = T v1 and extend w1 to basis of W . Then we automatically

obtains T that gets the desired property.

Problem 7
Suppose w1 , . . . , wn is a basis of W and V is finite-dimensional. Suppose
T ∈ L(V, W ). Prove that there exists a basis v1 , . . . , vm of V such that
all entries in the first of M(T ) [wrt. these bases] are 0 except for possibly
a 1 in the first row, first column.

Pn
Proof. Take arbitrary u1 ∈ V Pand we have that T u1 = i=1 ai,1 wi . Take
n
v1 = u1 /a1,1 , then T v1 = w1 + j=2 bj,1 wj . We can extend v1 , u2 , . . . , um to be

17
the basis of V . Then consider
n
X
T uj = bk,j wk
k=1

Consider to let b1,j = 0 for all 2 ≤ j ≤ n.PTo do this, let vj = uj − b1,j v1 ,

n
then we have that T (vj ) = T (uj − b1,j v1 ) = k=2 ck,j wk . It now left to verify
that v1 , v2 , . . . , vm is the basis.

q1 v1 + · · · + qm vm = 0
q1 v1 + q2 (u2 − b1,2 v1 ) + · · · + qm (um − b1,m v1 ) = 0
(q1 − (q2 b1,2 + · · · + qm b1,m ))v1 + q2 u2 + · · · + qm um = 0

The only solution is that q1 − (q2 b1,2 + · · · + qm b1,m ) = q2 = · · · = qm = 0

and thus we have completed the proof.

Problem 8
Suppose A is an m-by-n matrix and B is an n-by-p matrix. Prove that

(AB)j,· = Aj,· B

for each 1 ≤ j ≤ m. In other words, show that row j of AB equals (row

j of A) times B.

Proof. We know that

n
X
(AB)j,k = Aj,i Bi,k
i=1
Pn
For instance (AB)j,1 = i=1 Aj,i Bi,1 = Aj,· B·,1 . Similarly, (AB)j,k =
Aj,· B·,k and thus by treating this we have

(AB)j,· = Aj,· B

Problem 9
Suppose a = (a1 , · · · , an ) is a 1-by-n matrix and B is an n-by-p matrix.
Prove that
aB = a1 B1,· + · · · + an Bn,·
In other words, show that aB is a linear combination of the rows of B,
with the scalars that multiply the rows coming from a.

Proof. By definition of matrix multiplication, the entry in the column k of the

l.h.s. equals
(aB)1,k = a1 B1,k + · · · + an Bn,k
which equals the right-hand side.

18
 Problem 11
Prove that the distributive property holds for matrix addition and multi-
plication. In other words, suppose A, B, C, D, E, F are matrices whose
sizes are such that A(B + C) and (D + E)F make sense. Explain this
and prove that

A(B + C) = AB + AC (D + E)F = DF + EF

Proof. We let A, D, E to be m-by-n matrix and B, C, F to be n-by-p matrix.

Then all the matrix multiplication make sense. We proceed to prove the equality:
n
X
(A(B + C))j,k = Aj,i (B + C)i,k
i=1
Xn
= Aj,i (Bi,k + Ci,k )
i=1
Xn
= (Aj,i Bi,k ) + (Aj,i Ci,k )
i=1
= AB + AC

For the other one, we have that

n
X
((D + E)F)j,k = (D + E)j,i Fi,k
i=1
n
X
= (Dj,i Fi,k ) + (Ej,i Fi,k )
i
= DF + EF

Problem 13
Suppose A is an n-by-n matrix and 1 ≤ j, k ≤ n. Show that the entry in
row j, column k of A3 (which is defined to mean AAA) is
n X
X n
Aj,p Ap,r Ar,k
p=1 r=1

19
Proof.
n
X
(A(AA))j,k = Aj,p (AA)p,k
p
n
X n
X
= Aj,p Ap,r Ar,k
p r=1
n X
X n
= Aj,p Ap,r Ar, k
p r

Problem 14
Suppose m and n are positive integers. Prove that the function A 7→ A⊤
is a linear map from Fm,n to Fn,m .

Proof. Let T denote the linear map such that T (A) = A⊤ . Then we have that

T (λA + B) = (λA + B)⊤ = λA⊤ + B⊤ = λT (A) + T (B)

Problem 15
Prove that if A is an m-by-n matrix and C is an n-by-p matrix, then

(AC)⊤ = C⊤ A⊤

Proof.

((AC)⊤ )j,k = (AC)k,j

Xn
= Ak,i Ci,j
i=1

On the other hand,

n
X
(C⊤ A⊤ )j,k = C⊤ ⊤
j,i Ai,k
i=1
n
X
= Ak,i Ci,j
i=1

20
 Problem 16
Suppose A is an m-by-n matrix with A ̸= 0. Prove that the rank of A is
1 if and only if there exist (c1 , . . . , cm ) ∈ Fm and (d1 , . . . , dn ) ∈ Fn such
that Aj,k = cj dk for every j = 1, . . . , m and every k = 1, . . . , n.

Proof. ⇒ If rank of A is 1, then this means that the span of the columns of
A is 1-dimensional, take A·,1 ∈ Fm from the span. (Note that here we assume
̸ 0 ow. take the first nonzero one) Then we can let (c1 , . . . , cm ) = vec(A·,1 ).
A·,1 =
and let For every other columns, they are the scalar multiple of the the first one,
so there exists d2 , . . . , dn s.t. Aj,k = cj dk (we take d1 = 1).
⇐ If we denote c = (c1 , . . . , cm ), then


A = d1 c · · · dn c
so each column is a scalar multiple of the other and thus A has rank 1.
Problem 17
Suppose T ∈ L(V ), and u1 , . . . , un and v1 , . . . , vn are bases of V . Prove
that the following are equivalent:
(a) T is injective.
(b) The column of M(T ) are linearly independent in Fn,1 .
(c) The columns of M(T ) span Fn,1 .
(d) The rows of M(T ) span F1,n .
(e) The rows of M(T ) are linearly independent in F1,n .
Here M(T ) means M(T, (u1 , . . . , un ), (v1 , . . . , vn )).

Proof. Let A represents the matrix of M(T ). Pn

(a) ⇒ (b) : a1 A·,1 + · · · + an A·,n = 0. We also know that T vi = j=1 Aj,i uj .
Then we have that
 
n
X n
X n
X n
X Xn
T (a1 v1 +· · ·+an vn ) = ai T (vi ) = ai Aj,i uj =  ai Aj,i  uj = 0
i=1 i=1 j=1 j=1 i=1

To prove all ai is zero, we note that a1 v1 + · · · + an vn ∈ null T and since T

is injective, the only possible case is that all ai = 0 and thus we prove the claim.
(b) ⇒ (c) The columns of M(T ) forms a basis and thus span the space.
(c) ⇒ (d) Column rank of M(T ) is n and thus row rank is also n and thus
the rows span F1,n .
(d) ⇒ (e) Same as above.
(e) ⇒ (a) The only solution to a1 A1,· + · · · + an An,· = 0 is all ai = 0.
Suppose for the sake of contradiction P that T is not injective so P there exists
n n
nonzero v ∈ null T . We know that v = i=1 ai vi and that T vi = j=1 Aj,i uj .
Then
 
Xn n
X Xn Xn Xn
Tv = ai T (vi ) = ai Aj,i uj =  ai Aj,i  uj = 0
i=1 i=1 j=1 i=1 j=1

21
 Pn
and thus j=1 ai Aj,i = 0 for all j. We know not all ai = 0 since v is non-
trivial and thus the solution contradicts that the rows are all linearly independent.
Proof completed.

22
3D: Invertibility and Isomorphisms
Definition 37 (invertible, inverse). ˆ A linear map T ∈ L(V, W ) is called
invertible if there exists a linear map S ∈ L(W, V ) such that ST equals the
identity operator on V and T S equals the identity operator on W .
ˆ A linear map S ∈ L(W, V ) satisfying ST = I and T S = I is called an
inverse of T .
Theorem 38. An invertible linear map has a unique inverse.
If T is invertible, then its inverse is denoted by T −1 .
Theorem 39. A linear map is invertible if and only if it is injective and
surjective.
Proof. ⇒ We have an invertible linear map T that T u = T v. Then we have

u = T −1 T u = T −1 T v = v

To show T is surjective, we have that for any w ∈ W, w = T (T −1 w).

⇐ Define S such that for each w ∈ W , S(w) is the unique element s.t.
T (S(w)) = w. (we can do this due to injectivity and surjectivity). Then we
have that T (ST )v = (T S)T v = T v and thus ST v = v and thus ST = I. So
we suffices the identity operator condition. It’s easy to show that S is a linear
map.
Theorem 40. Suppose that V and W are finite-dimensional vector spaces,
dim V = dim W , and T ∈ L(V, W ). Then

T is invertible ⇐⇒ T is injective ⇐⇒ T is surjective.

Corollary 41. Suppose V and W are finite-dimensional vector spaces of the
same dimension, S ∈ L(W, V ), and T ∈ L(V, W ). Then ST = I if and only if
T S = I.
Definition 42 (Isomorphism). An Isomorphism is an invertible linear map.
Two vector spaces are called isomorphic if there is an isomorphism from one
vector space onto the other one.
Theorem 43 (dimension shows whether vector spaces are isomorphic). Two
finite-dimensional vector spaces over F are isomorphic if and only if they have
the same dimension.
Theorem 44. Suppose v1 , . . . , vn is a basis of V and w1 , . . . , wm is a basis of
W . Then M is an isomorphism between L(V, W ) and Fm,n .
Corollary 45. Suppose V and W are finite-dimensional. Then L(V, W ) is
finite-dimensional and

dim L(V, W ) = (dim V )(dim W )

23
Definition 46 (matrix of a vector, M(v)). Suppose v ∈ V and v1 , . . . , vn is a
basis of V . Then matrix of v with respect to the basis is the n-by-1 matrix
 
b1
.
M(v) =   ..  ,

bn

where b1 , . . . , bn are the scalars such that

v = b1 v1 + · · · + bn vn
Remark 47. The matrix M(v) of a vector v ∈ V depends on the basis v1 , . . . , vn
and v. We can think of elements of V as relabeled to be n-by-1 matrices, i.e.
V 7→ Fn,1 .
Corollary 48. Suppose T ∈ L(V, W ) and v1 , . . . , vn is a basis of V and
w1 , . . . , wm is a basis of W . Let 1 ≤ k ≤ n. Then kth column of M(T ),
which is denoted by M(T )·,k , equals M(T vk )
Theorem 49 (linear maps act like matrix multiplication). Suppose T ∈ L(V, W )
and v ∈ V . Suppose v1 , . . . , vn is a basis of V and w1 , . . . , wm is a basis of W .
Then
M(T v) = M(T )M(v)
Proof.
M(T v) = b1 M(T v1 ) + · · · + bn M(T vn )
= b1 M(T )·,1 + · · · + bn M(T )·,n
= M(T )M(v)

Remark 50. Each m-by-n matrix A induces a linear map from Fn,1 to Fm,1 ,
namely the matrix multiplication function that takes x ∈ Fn,1 to Ax ∈ Fm,1 . We
can think of every linear map (from a finite vector space to another) as a matrix
multiplication map after suitable relabeling via the isomorphisms given by M.
Specifically, if T ∈ L(V, W ) and we identify v ∈ V with M(v) ∈ Fn,1 , then the
result above says that we can identify T v with M(T )M(v).
Theorem 51 (dimension of range T equals column rank of M(T )). Suppose V
and W are finite-dimensional and T ∈ L(V, W ). Then dim range T equals the
column rank of M(T ).
Theorem 52 (change-of-basis-formula). Suppose T ∈ L(V ). Suppose u1 , . . . , un
and v1 , . . . , vn are bases of V . Let
A = M(T, (u1 , . . . , un )) and B = M(T, (v1 , . . . , vn ))
and C = M(I, (u1 , . . . , un ), (v1 , . . . , vn )). Then
A = C−1 BC.

24
Proof.

Problem 1
Suppose T ∈ L(V, W ) is invertible. Show that T −1 is invertible and

(T −1 )−1 = T.

Proof. T −1 is invertible because there exists T such that T T −1 = T −1 T = I.

T −1 T = T T −1 = I
so (T −1 )−1 = T .

Problem 2
Suppose T ∈ L(U, V ) and S ∈ L(V, W ) are both invertible linear maps.
Prove that ST ∈ L(U, W ) is invertible and that (ST )−1 = T −1 S −1 .

Proof. (ST )(T −1 S −1 ) = S(T T −1 )S −1 = I = T −1 S −1 ST .

Problem 3
Suppose V is finite-dimensional and T ∈ L(V ). Prove that the following
are equivalent.
(a) T is invertible.
(b) T v1 , . . . , T vn is a basis of V for every basis v1 , . . . , vn of V .
(c) T v1 , . . . , T vn is a basis of V for some basis v1 , . . . , vn of V .

Proof. (a) ⇒ (b) It only suffice to prove linear independence. We can show this

a1 T v1 + · · · + an T vn = 0 ⇐⇒ a1 v1 + · · · + an vn = 0

since T is injective and thus the only solution is all ai are identically zero.
(b) ⇒ (c) Trivial.
(c) ⇒ (a) By the linear map lemma, there exists S ∈ L(V ) such that
S(T vi ) = vi for all i. Such S is the inverse of T (one can verify) and thus T is
invertible.

Problem 5
Suppose V is finite-dimensional, U is a subspace of V , and S ∈ L(U, V ).
Prove that there exists an invertible linear map T from V to itself such
that T u = Su for every u ∈ U if and only if S is injective.

25
Proof. ⇒ Since T is invertible, we can define T −1 restricted to U , where S −1 u =
T −1 u for all u ∈ U . Then S has an inverse and thus is injective.
⇐ Let v1 , . . . vm be the basis of U and v1 , . . . , vm , x1 . . . , vn be the extended
basis of V . Define T vi = Svi and T xi = xi . Then T is injective and thus
invertible.

Problem 6
Suppose that W is finite-dimensional and S, T ∈ L(V, W ). Prove that
null S = null T if and only if there exists an invertible E ∈ L(W ) such
that S = ET .

Proof. ⇒ Suppose V is finite-dimensional for simplicity (one can do more work

for relaxing this assumption). Let V = null S ⊕ C = null T ⊕ C for some C.
Then we know that T |C : C → range (T (C)) and S|C : C → range (S(C)) is
invertible. Then there exists an invertible map Ê : range (T (C)) → range (S(C))
as dim range T (C) = range (S(C)) = dim V . Extending this map to W solves
the problem (take bases from range T (C) and extend to W with “identity” map).
⇐ Take v ∈ null T , then S(v) = E(T v) = E(0) = 0 and thus v ∈ null S.
Conversely, take v ∈ null S, then T (v) = E −1 S(v) = 0 so v ∈ null T . Thus
null S = null T .

Problem 7
Suppose that V is finite-dimensional and S, T ∈ L(V, W ). Prove that
range S = range T if and only if there exists an invertible E ∈ L(V ) such
that S = T E.

Proof. ⇒ By 3B P26, we know that there exists Ẽ ∈ L(V ) s.t. S = T Ẽ as

range S ⊆ range T . range S = range T implies that dim null S = dim range T .
Then we can define an invertible map Ē from null S to null T . We can define
the map E as follows: Denote V = null S ⊕ P , then E(v) = Ẽ(v1 ) + Ē(v2 ) for
v = v1 + v2 , where v1 ∈ P, v2 ∈ null S and now Ẽ = Ẽ|P . Then we have that
T E(v) = T (Ẽ(v1 ) + Ē(v2 )) = S(v1 ) + S(v2 ) = S(v). Now, it left to verify that
our proposed E is invertible.
To show this, we mainly aim at proving range Ẽ ∩ range Ē = {0}, as if we
have this, we know that E(v) = Ẽ(v1 ) + Ē(v2 ) = 0 ⇒ Ẽ(v1 ) = Ē(v2 ) = 0. We
know Ē is invertible so v2 = 0. To see Ẽ is injective, we can see that for any
v1 ∈ null Ẽ, we have Sv1 = T Ẽv1 = T 0 = 0 and thus v1 ∈ null S ∩ P = 0, so
v1 = 0 and thus nul(E) = {0}, E is injective and therefore invertible.
To complete the proof, let u ∈ range Ẽ ∩ range Ē, then we know there exists
v1 ∈ P, v2 ∈ null S s.t. u = Ẽ(v1 ) = Ē(v2 ). We also know that u ∈ null T as
range Ē = null T . Now, we have that S(v1 ) = T Ẽ(v1 ) = T Ē(v2 ) = 0 and thus
v1 ∈ null S. So v1 ∈ P ∩ null S = {0} and thus we have that u = {0}, finishing
the proof.
⇐ Take any s ∈ range S, then there exists E(s) s.t. T (E(s)) = s. Conversely,
take any t ∈ range T , then there exists v ∈ V s.t. t = T u = T EE −1 v = S(E −1 v).

26
Thus range S = range T .

Problem 8
Suppose V and W are finite-dimensional and S, T ∈ L(V, W ). Prove that
there exist invertible E1 ∈ L(V ) and E2 ∈ L(W ) such that S = E2 T E1
if and only if dim null S = dim null T .

Proof. ⇒ We know from 3B exercises that dim range ST ≤

min(dim range S, dim range T ). Applying here gets that dim range S ≤
dim range T . Consider that T = E2−1 SE1−1 , then we can also derive
dim range T ≤ dim range S and thus dim range S = dim range T . By the
fundamental theorem of linear map, we can get that dim null S = dim null T .
⇐ We know dim null S = dim null T , so there exists an isomorphism
Ẽ1 : null S 7→ null T . Extending Ẽ1 : null S 7→ V still preserves its injectiv-
ity. Therefore, by P5, we have an invertible E1 : V 7→ V such that E1 u = Ẽ1 u
for all u ∈ null S. Now we intend to show that null S = null T E1 , as proving
so would imply there exists an invertible E2 ∈ L(W ) such that S = E2 T E1
by P6. To see this, take s ∈ null S, then T E1 (s) = T (Ẽ1 s) = 0 and thus
s ∈ null T E1 . To see the other direction, take t ∈ null T E1 , then we know
E1 t ∈ null T = range Ẽ1 . Thus there exists v ∈ null S s.t. E1 t = Ẽ1 v = E1 v.
As E1 is invertible, t = v ∈ null S and thus we have shown null S = null T E1 ,
completing the proof.

Problem 9
Suppose V is finite-dimensional and T : V → W is a surjective linear map
of V onto W . Prove that there is a subspace U of V such that T |U is an
isomorphism of U onto W .

Proof. T being surjective means that dim V ≥ dim W . Take any subspace of
V that has dim W , then there is an isomorphism between this subspace and
W.

Problem 10
Suppose V and W are finite-dimensional and U is a subspace of V . Let

E = {T ∈ L(V, W ) : U ⊆ null T }.

(a) Show that E is a subspace of L(V, W ).

(b) Find a formula for dim E in terms of dim V, dim W, and dim U .

Proof. (a) Take T1 , T2 ∈ E and λ ∈ F, then consider null (λT1 + T2 ). Let u ∈ U ,

then λT1 + T2 (u) = λT1 (u) + T2 (u) = 0 + 0 = 0 and thus we have λT1 + T2 ∈ E
and thus E is a subspace of L(V, W ).
(b) Following the hint, define Φ : L(V, W ) → L(U, W ) by Φ(T ) = T |U . Then
Φ ∈ L(L(V, W ), L(U, W )). We try to show the range and null space of Φ.

27
 We first claim null Φ = E. To see this, take T ∈ null Φ, then we know
Φ(T ) = T |U = 0 ∈ L(U, W ), meaning that U is the subset of null T , so
null Φ ⊆ E. Conversely, take T ∈ E, then we have that for all u ∈ U , T u = 0,
then we have that T |U = 0. This shows that E ⊆ null Φ.
Next we claim range Φ = L(U, W ). Take any S ∈ L(U, W ), then we can
naturally always extend S to T ∈ L(V, W ) by setting T u = Su for all u ∈ U .
The converse direction is trivial.
Thus we have that dim E = dim null Φ and dim range Φ = dim L(U, W ) =
dim U dim W and dim L(V, W ) = dim V dim W = dim null Φ + dim range Φ,
thus we derive that

dim V dim W = dim E + dim U dim W.

Problem 14
Prove or give a counterexample: If V is finite-dimensional and R, S, T ∈
L(V ) are such that RST is surjective, then S is injective.

Proof. We know that

dim range RST = dim range V ≤ min{dim range R, dim range S, dim range T }

This means that dim RST ≤ dim range S. If dim null S > 0, then dim range S <
dim V and thus contradicts the assumption that RST is surjective.

Problem 15
Suppose T ∈ L(V ) and v1 , . . . , vm is a list in V such that T v1 , . . . , T vm
spans V . Prove that v1 , . . . , vm spans V .

Proof. We can reduce the list to basis and use previous conclusions. Since
T v1 , . . . , T vm spans V , we can reduce the list to T v1 , . . . , T vn such that the list
is the basis of V . By previous results, v1 , . . . , vn is also linearly independent
and its dimension equals the dim V and thus is the basis of V . It thus spans V .
Extending the list to v1 , . . . , vm also spans V .

Problem 16
Prove that every linear map from Fn,1 to Fm,1 is given by a matrix
multiplication. In other words, prove that if T ∈ L(Fn,1 , Fm,1 ), then
there exists an m-by-n matrix A such that T x = Ax for every x ∈ Fn,1 .

Proof. We can simply consider the standard basis of Fn,1 and Fm,1 . Then Let
A be the matrix of T wrt these bases. Then we have that by definition,

T x = M(T x) = M(T )M(x) = Ax

28
 Problem 17
Suppose V is finite-dimensional and S ∈ L(V ). Define A ∈ L(L(V )) by

A(T ) = ST

for T ∈ L(V ).
(a) Show that dim null A = dim V dim null S.
(b) Show that dim range A = dim V dim range S.

Proof. (a) We prove this by showing null A = L(V, null S). Take A ∈ null A,
then we know that A(A) = SA = 0 and thus Av ∈ null S for all v ∈ V . For
the other direction, take A ∈ L(V, null S), then SA(v) = A(v) = 0 and thus
A ∈ null A.
(b) We know dim(L(V )) = dim null A + dim range A. Hence

dim range A = dim V (dim V − dim null S) = dim V dim range S.

Problem 18
Show that V and L(F, V ) are isomorphic vector spaces.

Proof. For simplicity, assume V is finite-dimensional, then dim V =

dim F dim V = dim L(F, V ) and thus they are isomorphic. One might do a
more careful construction for infinite-dimensional V .

Problem 19
Suppose V is finite-dimensional and T ∈ L(V ). Prove that T has the
same matrix with respect to every basis of V if and only if T is a scalar
multiple of the identity operator.

Pn Pn
Proof. ⇒ We know T vk = j=1 Aj,k vk = j=1 2Aj,k vk = T (2vk ). This means
that Aj,k = 0 for j ̸= k. When j = k, we have that T vk = Ak,k vi for some
arbitrary ordering of the same basis. Then this means that Ai,i = Aj,j for all
i, j. Thus, T is a scalar multiple of the identity operator.
⇐ This follows by the definition of the identity operator.

29
3E: Products and Quotients of Vector Spaces
Definition 53 (product of vector spaces). Suppose V1 , . . . , Vm are vector spaces
over F.
ˆ The product V1 × · · · × Vm is defined by

V1 × · · · × Vm = {(v1 , . . . , vm ) : v1 ∈ V1 , . . . , vm ∈ Vm }.

ˆ Addition on V1 × · · · × Vm is defined by

(u1 , . . . , um ) + (v1 , . . . , vm ) = (u1 + v1 , . . . , um + vm ).

ˆ Scalar multiplication on V1 × · · · × Vm is defined by

λ(v1 , . . . , vm ) = (λv1 , . . . , λvm ).

Theorem 54 (product of vector spaces is a vector space). Suppose V1 , . . . , Vm

are vector spaces over F. Then V1 × · · · × Vm is a vector space over F.
Theorem 55 (dimension of a product is the sum of dimensions). Suppose
V1 , . . . , Vm are finite-dimensional vector spaces. Then V1 × · · · × Vm is finite-
dimensional and

dim(V1 × · · · × Vm ) = dim V1 + · · · + dim Vm .

Lemma 56 (products and direct sums). Suppose that V1 , . . . , Vm are subspaces

of V . Define a linear map Γ : V1 × · · · × Vm → V1 + · · · + Vm by

Γ(v1 , . . . , vm ) = v1 + · · · + vm .

Then V1 + · · · + Vm is a direct sum if and only if Γ is injective.

Theorem 57 (a sum is a direct sum if and only if dimensions add up). Suppose
V is finite-dimensional and V1 , . . . , Vm are subspaces of V . Then V1 + · · · + Vm
is a direct sum if and only if

dim(V1 + · · · + Vm ) = dim V1 + · · · + dim Vm .

Definition 58 (v + U , translation). Suppose v ∈ V and U ⊂ V . Then v + U is

the subset of V defined by

v + U = {v + u : u ∈ U }.

The set v + U is said to be a translate of U .

Definition 59 (quotient space, V /U ). Suppose U is a subspace of V . Then
the quotient space V /U is the set of all translates of U . Thus

V /U = {v + U : v ∈ V }.

30
Remark 60. If U = {(x, 2x) ∈ R2 : x ∈ R}, then R2 /U is the set of all lines in
R2 that have slope 2.
Lemma 61 (two translates of a subspace are equal or disjoint). Suppose U is a
subspace of V and v, w ∈ V . Then

v − w ∈ U ⇐⇒ v + U = w + U ⇐⇒ (v + U ) ∩ (w + U ) = ∅
Definition 62 (addition and scalar multiplication on V /U ). Suppose U is a
subspace of V . Then addition and scalar multiplication are defined on V /U
by
(v + U ) + (w + U ) = (v + w) + U
λ(v + U ) = (λv) + U
for all v, w ∈ V and all λ ∈ F.
Theorem 63 (quotient space is a vector space). Suppose U is a subspace of V .
Then V /U , with the operations of addition and scalar multiplication as defined
above, is a vector space.
Definition 64 (quotient map). Suppose U is a subspace of V . The quotient
map π : V → V /U is the linear map defined by
π(v) = v + U
for each v ∈ V .
Theorem 65 (dimension of quotient space). Suppose V is finite-dimensional
and U is a subspace of V . Then
dim V /U = dim V − dim U
Definition 66. Suppose T ∈ L(V, W ). Define T̃ : V /(null T ) → W by
T̃ (v + null T ) = T v.
Another interpretation of quotient space is on congruence of subspaces.
The definition goes as follows:
If Y is a subspace of X, then two vectors x1 , x2 ∈ X are congruent modulo
Y , denoted x1 ∼ = x2 ( mod Y ) if x1 − x2 ∈ Y .
The quotient space X/Y denotes the set of equivalence classes in X, modulo
Y by defining that
{x} + {z} := {x + z} a{x} := {ax}
So here one can think of the zero vector of v + U defined as the equivalence
class that contains that zero vector 0 ∈ V :
0 + U = {0 + u : u ∈ U } = U.
This means that the zero element in the quotient space V /U is simply the
subspace U , and this equivalence class contains all vectors differ from 0 by an
element of U , meaning it is precisely U itself, so π(v) = 0 if and only if v ∈ U .

31
Remark 67. The output of π(v) is the equivalence class v + U , which is the set
of all vectors in V that are equivalent to v under the subspace U .
For T̃ , the input to it is an equivalent class v + null (T ) in the quotient space
V /null T . It represents all vectors in V that differ from v by a vector in null T .
Its output is T v, which is an element in W .
The null space is

null (T̃ ) = {v + null (T ) : T̃ (v + null (T )) = 0}.

Since we know that T̃ (v + null T ) = T v, we have that

T̃ (v + null T ) = 0 ⇐⇒ T v = 0.

so v ∈ null T and thus null T̃ = {null T }. We have that range T̃ = range T .

Theorem 68. Suppose T ∈ L(V, W ). Then
(a) T̃ ◦ π = T , where π is the quotient map of V onto V /null T ;
(b) T̃ is injective;
(c) range T̃ = range T ;
(d) V /null T and range T are isomorphic vector spaces.
Remark 69. One might think of the domain of T̃ as the non-trivial domain of
T and the range as the normal image of T .

32
 Problem 1
Suppose T is function from V to W . The graph of T is the subset of
V × W defined by

graph of T = {(v, T v) ∈ V × W : v ∈ V }.

Prove that T is a linear map if and only if the graph of T is a subspace

of V × W .

Proof. The two properties closely relate to each other:

(v1 , T v1 ) + (v2 , T v2 ) ∈ V × W : v ∈ V ⇐⇒ T v1 + T v2 = T (v1 + v2 )

and
(λv, T λv) ∈ V × W : v ∈ V ⇐⇒ T λv = λT v

Problem 2
Suppose that V1 , . . . , Vm are vector spaces such that V1 ×· · ·×Vm is finite-
dimensional. Prove that Vk is finite-dimensional for each k = 1, . . . , m.

Proof. Recall the definition of infinite-dimension from ch2 P17:

V is infinite-dimensional if and only if there is a sequence v1 , v2 , . . . of vectors in
V such that v1 , . . . , vm is linearly independent for every positive integer m.
Suppose for the sake of contradiction that there exists Vk to be infinite-
dimensional. Then this means that there is a sequence vk,1 , vk,2 , . . . of vectors
in Vk such that vk,1 , . . . , vk,m is linearly independent for every positive integer
m. We can build such sequence for the product space as well. More specifically,
consider the sequence 0, . . . , vk,j , . . . , 0 where we fill 0 for each of other Vj ’s and
only leave the sequence from Vk . In this case, we can see that the product space
is infinite-dimensional, contradicting the claim.

Problem 3
Suppose V1 , . . . , Vm are vector spaces. Prove that L(V1 × · · · × Vm , W )
and L(V1 , W ) × · · · × L(Vm , W ) are isomorphic vector spaces.

Proof. We define a permutation invariant set function S : L(V1 , W ) × · · · ×

L(Vm , W ) → L(V1 × · · · × Vm , W ) such that

S(T1 , . . . , Tm ) = T1 + · · · + Tm

We will show that S is an isomorphism (it’s linear). For injectivity, null S

means that for any arbitrary input T1 + · · · + Tm = 0, which implies that
Ti = 0 for all i. For surjectivity, take ϕ ∈ L(V1 × · · · × Vm , W ), we can define

33
Ti (vi ) = ϕ((0, . . . , vi , . . . , 0)) where i means
Pm for the P space Vi . Then we can
m
naturally get that ϕ((v1 , . . . , vm )) = ϕ( i=1 e⊤ v
i i ) = i=1 Ti (vi ) and thus we
finish the proof.

Problem 4
Suppose W1 , . . . , Wm are vector spaces. Prove that L(V, W1 × · · · × Wm )
and L(V, W1 ) × · · · × L(V, Wm ) are isomorphic vector spaces.

We can simply define a concatenate operator S : L(V, W1 )×· · ·×L(V, Wm ) →

L(V, W1 × · · · × Wm ) such that

S(T1 , . . . , Tm ) = (T1 , . . . , Tm )

We prove this linear map is an isomorphism. The injective property can be proved
using the same logic as above. For surjectivity, take ϕ ∈ Lc(V, W1 × · · · × Wm ),
then we can naturally obtain Tk = πk ϕ where πk means projecting the k-th
component of the input. Note that Tk ∈ L(V, Wk ) and therefore we finish the
proof.

Problem 5
For m a positive integer, define V m by

V m = V × ··· × V .
| {z }
m times

Prove that V m and L(Fm , V ) are isomorphic vector spaces.

Proof.
dim(V m ) = m dim V = dim(L(Fm , V )).

Problem 6
Suppose that v, x are vectors in V and that U, W are subspaces of V
such that v + U = x + W . Prove that U = W .

Proof. Note that v = x + w for some w ∈ W if we take u = 0 ∈ U and similarly

x = v + u for some u ∈ U . Hence v − x ∈ W and x − v ∈ U . Then we
have that take u ∈ U, u = (x + w) − v = (x − v) + w ∈ W . Similarly, take
w ∈ W, w = (v + u) − x = (v − x) + u ∈ U . Hence W = U .

Problem 9
Prove that a nonempty subset A of V is a translate of some subspace of
V if and only if λv + (1 − λ)w ∈ A for all v, w ∈ A and all λ ∈ F.

34
Proof. ⇒ Suppose A = x + U for some subspace U ⊆ V . Then take a1 =
x + u1 ∈ A and a2 = x + u2 ∈ A. We have that

λa1 + (1 − λ)a2 = λx + λu1 + (1 − λ)x + (1 − λ)u2 = x + (λu1 + (1 − λ)u2 ) ∈ A

⇐ take any x ∈ A and define the subspace such that

U := (−x) + A

first 0 = −x + x ∈ U . Next, take v1 − x and v2 − x ∈ U . We have that

v1 − x + v2 − x = (−x) + (2v1 − x)/2 + (2v2 − x)/2 ∈ A as 2vi − x ∈ A by taking
λ = 2. Finally, λ(−x + v) = (−x) + (λv + (1 − λ)x) ∈ A.

Problem 10
Suppose A1 = v + U1 and A2 = w + U2 for some v, w ∈ V and some
subspaces U1 , U2 of V . Prove that the intersection A1 ∩ A2 is either a
translate of some subspace of V or is the empty set.

Proof. If A1 ∩ A2 is not empty, then S = A1 ∩ A2 is a subspace such that

v + U1 |S = w + U2 |S , so we know that U1 |S = U2 |S by Problem 6. Hence it is a
translate of some subspace of V . Otherwise, it is the empty set.

Problem 11
Suppose U = {(x1 , x2 , . . .) ∈ F∞ : xk ̸= 0 for only finitely many k}.
(a) Show that U is a subspace of F∞ .
(b) Prove that F∞ /U is infinite-dimensional.

Proof. (a) all zero, where xk = ̸ 0 for zero k’s (finitely many) is an element of
U . Next, take (x1 , x2 , . . .) and (y1 , y2 , . . .) ∈ U . Then (x1 + y1 , x2 + y2 , . . .) will
have finitely many nonzero since each of them has only finitely many nonzero
entries. Same holds for λ(x1 , x2 , . . .).
(b) Let’s consider a “standard basis” {vi }∞ i=1 that only has 0 on i-th spot and
all 1 otherwise. We can see that for all m, v1 , . . . , vm is linearly independent.
And so does v1 + U, . . . , vm + U since each vi ∈ / U,

Problem 12
Suppose v1 , . . . , vm ∈ V . Let

A = {λ1 v1 + · · · + λm vm : λ1 , . . . , λm ∈ F and λ1 + · · · + λm = 1}.

(a) Prove that A is a translate of some subspace of V .

(b) Prove that if B is a translate of some subspace of V and {v1 , . . . , vm } ⊆
B, then A ⊆ B.
(c) Prove that A is a translate of some subspace of V of dimension less
than m.

35
Proof. (a)P
We try to show thisPthrough usingPthe conclusion
Pm from Problem 9.
m m m
Let a1 = i=1 αi vi and a2 = i=1 γi vi s.t. α
i=1 i = i=1 γi = 1. We have
that
m
X
λa1 + (1 − λ)a2 = (λαi + (1 − λ)γi )vi
i=1

We can show that

m
X m
X m
X
(λαi + (1 − λ)γi ) = λ αi + (1 − λ) γi = 1
i=1 i=1 i=1

Thus λa1 + (1 − λ)a2 ∈ A and thus A is a translate of some subspace of V .

(b) Let B = w + Y for some subspace Y ⊆ V . Then we have that vk = w + yk .
Take x ∈ A, then we know that
m
X m
X m
X
x= λi vi = λi (w + yi ) = w + λi yi ∈ B
i=1 i=1 i=1

(c) Write A = w + U . Denote B = span{v1 , . . . , vm }. Then by (b) we know

that A ⊆ B. We will show that dim U < m. The statement is trivial if v1 , . . . , vm
is not linearly independent. Let’s consider the case that it’s indeed linearly
independent. HerePwe claim that w ∈ / U and thus dim U < m. To see this, the
m
only way to write i=1 λi vi = 0 is to let all λi = 0, but in this case the vectors
will not be in A, so 0 ∈
/ A. Suppose for the sake of contradiction that w ∈ U .
This implies that 0 = w + (−w) ∈ w + U = A, forming a contradiction. Hence
we finish the proof.

Problem 13
Suppose U is a subspace of V such that V /U is finite-dimensional. Prove
that V is isomorphic to U × (V /U ).

Proof.

dim V = dim U + (dim V − dim U ) = dim U + dim(V /U ).

Problem 14
Suppose U and W are two subspaces of V and V = U ⊕ W . Suppose
w1 , . . . , wm is a basis of W . Prove that w1 + U, . . . , wm + U is a basis of
V /U .

Proof. First we show that the list is linearly independent. We consider the
following system:

36
 0 + U = λ1 (w1 + U ) + · · · + λm (wm + U ) = (λ1 w1 + · · · + λm wm ) + U

This means that λ1 w1 +· · ·+λm wm ∈ U and since we know that w1 , . . . , wm ∈

/
U , we have that the only solution is λi = 0 for all i.
Next we show that the list spans VP /U . Take arbitrary v + U ∈ V /U , then
m
we have that v = u + w for some w = i=1 λi wi . Then we have that
m
X m
X
v+U = λi wi + U = λi (wi + U ).
i=1 i=1

Thus the list spans the quotient space.

Problem 15
Suppose U is a subspace of V and v1 + U, . . . , vm + U is a basis of V /U
and u1 , . . . , un is a basis of U . Prove that v1 , . . . , vm , u1 , . . . , un is a basis
of V .

Proof. We first know that

dim V = dim V /U + dim U.

So it only suffices to prove the list either spans V or is linearly independent.

Take any v ∈ V . Then we have that

v + U = λ1 v1 + · · · + λm vm + U.
Pm
Thus v − i=1 λi vi ∈ U . So
m
X n
X
v= λ i vi + αj ui .
i=1 j=1

Hence, we finish the proof.

Problem 16
Suppose φ ∈ L(V, F) and φ ̸= 0. Prove that dim V /(null φ) = 1.

Proof. We know that dim range φ = 1 and thus

dim V /null φ = dim V − dim null φ = dim range φ = 1.

Problem 17
Suppose that U is a subspace of V such that dim V /U = 1. Prove that
there exists φ ∈ L(V, F) such that null φ = U .

37
Proof. We make construction as follows: (1) there exists a natural isomorphism
T : V /U → F. We further define S : V → V /U by S(v) = v + U . Then we show
the map
φ = TS
satisfies the requirement.
First take u ∈ U . Then we have that

φ(u) = T S(u) = T (u + U ) = T (0) = 0

and thus u ∈ null φ.

Conversely take u ∈ null φ, then

0 = φ(u) = T (S(u)) = T −1 T (S(u)) = S(u).

Thus u ∈ null S = U . We’ve finished the proof.

Problem 18
Suppose U is a subspace of V such that V /U is finite-dimensional.
(a) Show that if W is a finite-dimensional subspace of V and V = U + W ,
then dim W ≥ dim V /U .
(b) Prove that there exists a finite-dimensional subspace W of V such
that dim W = dim V /U and V = U ⊕ W .

Proof. (a) We know that

dim W = dim V + dim(U ∩ V ) − dim U ≥ dim V − dim U = dim V /U.

(b) We will construct the space W through some manipulation with the basis.
Since we are given V /U , we start with the basis w1 + U, . . . , wm + U of that.
Note that {w1 , . . . , wm } is linearly independent as the only solution is all-zero.
We can now define
W = span{w1 , . . . , wm }.
Then we have that dim W = dim V /U and now it suffices to verify V = U ⊕W .
To see this, one only needs to show W ∩U = {0} as we’ve proved that V = W +U
in previous questions (of similar construction). Take arbitrary v ∈ W ∩ U . Then
we have that
Xm
v= λi wi
i=1

and that
m
X
v+U = λi wi + U = 0 + U
i=1

The only solution is all λi = 0.

38
 Problem 19
Suppose T ∈ L(V, W ) and U is a subspace of V . Let π denote the
quotient map from V onto V /U . Prove that there exists S ∈ L(V /U, W )
such that T = S ◦ π if and only if U ⊆ null T.

Proof. ⇒ Take u ∈ U , then we have that

T (u) = Sπ(u) = S(u + U ) = S(0) = 0

and thus u ∈ null T .

⇐ We define S : V /U → W by

S(v + U ) = T (v)

It now only suffices to show that this map is valid for the equivalence mapping.
Take v1 + U = v2 + U . Then we have that v1 − v2 ∈ U ⊆ null T and thus
T (v1 − v2 ) = 0 = T v1 − T v2 and thus S(v1 + U ) = T v1 = T v2 = S(v2 + U ).
This map is clearly linear and thus we have T = S ◦ π.

39
3F: Duality
Definition 70 (linear functional). A linear functional on V is a linear map
from V to F. In other words, a linear functional is an element of L(V, F).
Definition 71 (dual space, V ′ ). The dual space of V , denoted by V ′ , is the
vector space of all linear functionals on V . In other words, V ′ = L(V, F).
Lemma 72. Suppose V is finite-dimensional. Then V ′ is also finite-dimensional
and
dim V ′ = dim V.
Definition 73 (dual basis). If v1 , . . . , vn is a basis of V , then the dual basis
of v1 , . . . , vn is the list φ1 , . . . , φn of elements of V ′ , where each φj is the linear
functional on V such that
(
1 if k = j,
φj (vk ) =
0 if k ̸= j.

Theorem 74 (dual basis gives coefficients for linear combination). Suppose

v1 , . . . , vn is a basis of V and φ1 , . . . , φn is the dual basis. Then

v = φ1 (v)v1 + · · · + φn (v)vn

for each v ∈ V .

Theorem 75 (dual basis is a basis of the dual space). Suppose V is finite-

dimensional. Then the dual basis of a basis of V is a basis of V ′ .
Definition 76 (dual map, T ′ ). Suppose T ∈ L(V, W ). The dual map of T is
the linear map T ′ ∈ L(W ′ , V ′ ) defined for each φ ∈ W ′ by

T ′ (φ) = φ ◦ T.

Corollary 77 (algebraic properties of dual map). Suppose T ∈ L(V, W ). Then

(a) (S + T )′ = S ′ + T ′ for all S ∈ L(V, W );
(b) (λT )′ = λT ′ for all λ ∈ F;
(c) (ST )′ = T ′ S ′ for all S ∈ L(W, U ).

The goal of this section is to describe null T ′ and range T ′ in terms of range T
and null T .
Definition 78 (annihilator, U 0 ). For U ⊆ V , the annihilator of U , denoted
by U 0 , is defined by

U 0 = {φ ∈ V ′ : φ(u) = 0 for all u ∈ U }.

Remark 79. U 0 is a subspace of V ′ .

40
Theorem 80 (dimension of the annihilator). Suppose V is finite-dimensional
and U is a subspace of V . Then

dim U 0 = dim V − dim U.

Lemma 81. Suppose V is finite-dimensional and U is a subspace of V . Then

(a) U 0 = 0 ⇐⇒ U = V ;
(b) U 0 = V ′ ⇐⇒ U = {0}.
Lemma 82 (the null space of T ′ ). Suppose V and W are finite-dimensional
and T ∈ L(V, W ). Then
(a) null T ′ = (range T )0 ;
(b) dim null T ′ = dim null T + dim W − dim V .
Theorem 83 (T surjective is equivalent to T ′ injective). Suppose V and W are
finite-dimensional and T ∈ L(V, W ). Then

T is surjective ⇐⇒ T ′ is injective.

Lemma 84 (the range of T ′ ). Suppose V and W are finite-dimensional and

T ∈ L(V, W ). Then
(a) dim range T ′ = dim range T ;
(b) range T ′ = (null T )0 .
Theorem 85 (T surjective is equivalent to T ′ injective). Suppose V and W are
finite-dimensional and T ∈ L(V, W ). Then

T is surjective ⇐⇒ T ′ is injective.

Theorem 86 (matrix of T ′ is transpose of matrix of T ). Suppose V and W are

finite-dimensional and T ∈ L(V, W ). Then

M(T ′ ) = (M(T ))⊤

Proof. Let v1 , . . . , vn be basis of V and w1 , . . . , wm be basis of W . Let φ1 , . . . , φn

be the dual basis of V ′ and ψ1 , . . . , ψm be the dual basis of W ′ .
Let A = M(T ) and C = M(T ′ ). Suppose 1 ≤ j ≤ m and 1 ≤ k ≤ n. From
the definition of M(T ′ ) we have
n
X
T ′ (ψj ) = Cr,j φr
r=1

Then applying both sides of the equation above to vk gives that

n
X
(ψj ◦ T )(vk ) = Cr,j φr (vk )
r=1
= Ck,j

41
At the same time, we have that

(ψj ◦ T )(vk ) = ψj (T vk )
 
X m
= ψj  Ar,k wr 
r=1
m
X
= Ar,k ψj (wr )
r=1
= Aj,k

Here we have that Ck,j = Aj,k and thus C = A⊤ . Hence, M(T ′ ) = (M(T ))⊤ ,
as desired.

We can use duality to provide an alternative proof for the rank of matrix:
Theorem 87 (column rank equals row rank). Suppose A ∈ Fm,n . Then the
column rank of A equals the row rank of A.

42
 Problem 3
Suppose V is finite-dimensional and v ∈ V with v =
̸ 0. Prove that there
exists φ ∈ V ′ such that φ(v) = 1.

Proof. The main requirement is to ensure that the constructed φ is valid. Note
that there exists subspace W ⊆ V such that V = W ⊕ span{v}. Thus we can
define φ(u) = i where u = w + iv for u ∈ V, w ∈ W, i ∈ F. Here φ is a valid
linear map and φ(v) = 1.

Problem 6
Suppose φ, β ∈ V ′ . Prove that null φ ⊆ null β if and only if there exists
c ∈ F such that β = cφ.

Proof. ⇒ If φ = 0, then it trivially holds. If φ = ̸ 0, then we can first take

v0 ∈ V s.t. ψ(v0 ) ̸= 0, then define c = β(v0 )/φ(v0 ), then we have that β(v) =
β(v0 + u) = β(v0 ) = cφ(v0 ) for some u ∈ null β.
⇐ Take v ∈ null φ, then β(v) = cφ(v) = 0. Thus null φ ⊆ null β.

Problem 8
Suppose v1 , . . . , vn is a basis of V and φ1 , . . . , φn is the dual basis of V ′ .
Define Γ : V → Fn and Λ : Fn → V by

Γ(v) = (φ1 (v), . . . , φn (v)) and Λ(a1 , . . . , an ) = a1 v1 + · · · + an vn .

Explain why Γ and Λ are inverses of each other.

Proof. Take a1 , . . . , an ∈ Fn , then

Γ(Λ(a1 , . . . , an )) = Γ(a1 v1 + · · · + an vn )
= (φ1 (a1 v1 ), . . . , φn (an vn ))
= (a1 , . . . , an )

Similarly, take v = a1 v1 + · · · + an vn , we can get that

Λ(Γ(v)) = v

Problem 9
Suppose m is a positive integer. Show that the dual basis of the basis
1, x, . . . , xm of Pm (R) is φ0 , φ1 , . . . , φm , where

p(k) (0)
φk (p) = .
k!

43
Proof. We consider different cases. If j = k, then

(xj )(k) k!
φk (xj ) = = =1
k! k!
If j < k, then

(xj )(k) 0j−k j!/(j − k)!

φk (xj ) = = =0
k! k!
If j > k, then
(xj )(k) 0
φk (xj ) = = =0
k! k!
Hence, we have that
(
1 if j = k
φk (xj ) =
0 o.w.

Problem 11
Suppose v1 , . . . , vn is a basis of V and φ1 , . . . , φn is the corresponding
dual basis of V ′ . Prove that

ψ = ψ(v1 )φ1 + · · · + ψ(vn )φn .

Proof. By property of dual basis, we have that

ψ = a1 φ1 + · · · + an φn

Apply vk on both sides give that

ψ(vk ) = a1 φ1 (vk ) + · · · + an φn (vk )

= ak

Substitute this back gives the desired equality.

Problem 13
Show that the dual map of the identity operator on V is the identity
operator on V ′ .

Proof. Take arbitrary f ∈ V ′ , then we have that

I ′ (f )(v) = f ◦ I(v) = f (v)

for all v ∈ V .

44
 Problem 16
Suppose W is finite-dimensional and T ∈ L(V, W ). Prove that

T ′ = 0 ⇐⇒ T = 0.

Proof. This is obvious from dim range T ′ = dim range T .

Problem 19
Suppose U ⊆ V . Explain why

U 0 = {φ ∈ V ′ : U ⊆ null φ}

Proof. This follows from definition.

Problem 20
Suppose V is finite-dimensional and U is a subspace of V . Show that

U = {v ∈ V : φ(v) = 0 for every φ ∈ U 0 }.

Proof. Denote K = {v ∈ V : φ(v) = 0 for every φ ∈ U 0 }.

Take u ∈ U and φ ∈ U 0 , then by definition φ(u) = 0, and thus u ∈ K.
Conversely, take φ ∈ U 0 . Suppose for the sake of contradiction that there
exist v ∈
/ U but v ∈ K, then φ(v) ̸= 0, contradicting that v ∈ K, thus completing
the proof.

Problem 21
Suppose V is finite-dimensional and U and W are subspaces of V .
(a) Prove that W 0 ⊆ U 0 if and only if U ⊆ W .
(b) Prove that W 0 = U 0 if and only if U = W .

Proof. (a) ⇒ We know that there exists ψ ∈ V ′ such that null ψ = W . Then
ψ ∈ W 0 and thus ψ ∈ U 0 . This means that U ⊆ W = null ψ.
⇐ Take ψ ∈ W 0 . So we know ψ(w) = 0 for all w ∈ W . Suppose for the sake
/ U 0 , then there exists u ∈ U s.t. ψ(u) ̸= 0. However,
of contradiction that ψ ∈
as u ∈ W , we have reached a contradiction.
(b) W 0 = U 0 ⇐⇒ W0 ⊆ U0 and U0 ⊆ W0 ⇐⇒ U ⊆ W and W ⊆ W ⇐⇒
U = W.

Problem 22
Suppose V is finite-dimensional and U and W are subspaces of V .
(a) Show that (U + W )0 = U 0 ∩ W 0 .
(b) Show that (U ∩ W )0 = U 0 + W 0 .

45
Proof. Note that we have

(U + W )0 = {φ ∈ V ′ : φ(v) = 0 for every v ∈ U + W }

(U ∩ W )0 = {φ ∈ V ′ : φ(v) = 0 for every v ∈ U ∩ W }

(a) Take φ ∈ (U + W )0 , then we know that φ(u) = 0 and φ(w) = 0 for all
u ∈ U, w ∈ W . Thus (U + W )0 ⊆ U 0 ∩ W 0 . Conversely, take φ ∈ U 0 ∩ W 0 ,
then we know that for every v = u + w, φ(v) = φ(u) + φ(w) = 0, therefore
U 0 ∩ W 0 ⊆ (U + W )0 .
(b) Take φ1 ∈ U 0 , φ2 ∈ W 0 , then we have that φ1 +φ2 (v) = φ1 (v)+φ2 (v) = 0
for v ∈ U ∩ W , therefore U 0 + W 0 ⊆ (U ∩ W )0 . It suffices now to show that the
dimension equal each other.

dim((U ∩ W )0 ) = dim(V ) − dim(U ∩ W )

Note that

dim(U 0 + W 0 ) = dim(U 0 ) + dim(W 0 ) − dim(U 0 ∩ W 0 )

= (dim(V ) − dim(U )) + (dim(V ) − dim(W )) − dim((U + W )0 )
= 2 dim(V ) − (dim(U ) + dim(W )) − (dim(V ) − dim(U + W ))
= dim(V ) − (dim(U ) + dim(W ) − dim(U + W ))
= dim(V ) − dim(U ∩ W )

Thus we have completed the proof.

Problem 23
Suppose V is finite-dimensional and φ1 , . . . , φm ∈ V ′ . Prove that the
following three sets are equal to each other.
(a) span(φ1 , . . . , φm ).
(b) ((null φ1 ) ∩ · · · ∩ (null φm ))0 := A.
(c) {φ ∈ V ′ : (null φ1 ) ∩ · · · ∩ (null φm ) ⊆ null φ} := B.

Pm
Proof. (a) ⇒ (b) take φ = i=1 ai φi ∈ span(φ1 ,P . . . , φm ). Then we know that
m
for every v ∈ (null φ1 ) ∩ · · · ∩ (null φm ), φ(v) = i=1 ai φi (v) = 0.
(b) ⇒ (c) Take φ ∈ A, then by P19 this holds.
(c) ⇒ (a) Take φ ∈ B, then we know by P3 there exists c ∈ F such that
φ = cφi for all 1 ≤ i ≤ m. This means that φ ∈ span(φ1 , . . . , φm ).

Problem 24
Suppose V is finite-dimensional and v1 , . . . , vm ∈ V . Define a linear map
Γ : V ′ → Fm by Γ(φ) = (φ(v1 ), . . . , φ(vm )).
(a) Prove that v1 , . . . , vm spans V if and only if Γ is injective.
(b) Prove that v1 , . . . , vm is linearly independent if and only if Γ is
surjective.

46
 Pm
Proof. (a) ⇒ Take v = i=1 ai vi and φ ∈ null Γ. We aim to prove that φ = 0.
We have that Γ(φ) =P (φ(v1 ), . . . , φ(vm )) = 0 so φ(v1 ) = · · · = φ(vm ) = 0. So
m
for all v ∈ V , φ(v) = i=1 ai φ(vi ) = 0. Thus φ is the zero map and thus Γ is
injective.
⇐ Suppose for the sake of contradiction that v1 , . . . , vm does not span V , then
this means there exists some subspace W such that span(v1 , . . . , vm ) ⊕ W = V .
Then there exists nonzero φ ∈ V ′ such that φ(vk ) = 0 for all k (one can set
the basis in W to be nonzero mapping). Then this means that φ ∈ null Γ ̸= 0,
contradicting that Γ is injective.
(b) ⇒ Let K be the subspace spanned by the linearly independent list of
vectors v1 , . . . , vm . Then by the linear map lemma, for all (a1 , . . . , am ) ∈ Fm ,
there exists a unique mapping T ′ : K → F such that T ′ vi = ai . Extending T ′ to
V ensures that Γ is surjective.
⇐ Suppose for the sake of contradiction that v1 , . . . , vm are not linearly
independent, then we know there exists nonzero a′i s such that a1 v1 +· · ·+am vm =
0. Applying any linear function φ ∈ V ′ to this linear combination yields that
 
Xm
φ ai vi  = φ(0) = 0
i=1

This equals that

m
X m
X
ai φ(vi ) = ai Γ(φ)i = 0
i=1 i=1

Since there is nonzero ai ’s, this means that the image of Γ is a strict subspace
of Fm , therefore contradicting the hypothesis that it’s surjective.

Problem 25
Suppose V is finite-dimensional and φ1 , . . . , φm ∈ V ′ . Define a linear
map Γ : V → Fm by Γ(v) = (φ1 (v), . . . , φm (v)).
(a) Prove that φ1 , . . . , φm spans V ′ if and only if Γ is injective.
(b) Prove that φ1 , . . . , φm is linearly independent if and only if Γ is
surjective.

Pm
Proof. (a) ⇒ Take any φ ∈ V ′ , then φ = i=1 ai φi . Take any v ∈ V and let
Γ(v) = 0. Then this means that

Γ(v) = (φ1 (v), . . . , φm (v)) = 0

So we have that φi (v) = P0mfor all i. By the spanning assumption, we have that
for any φ ∈ V ′ , φ(v) = i=1 ai φi (v) = 0. Thus v = 0 and then null (Γ) = {0}.
⇐ Suppose φ1 , . . . , φm does not span V ′ , then this means there exists some
subspace W such that span(φ1 , . . . , φm )⊕W = V ′ . Let φ1 , . . . , φn be the basis of
span(φ1 , . . . , φm ), φn+1 , . . . , φl to be the basis of W . Let v1 , . . . , vn , vn+1 , . . . , vl
be the corresponding basis for V . Then there exists nonzero v ∈ V such that

47
 Pn Pl
φk (v) = 0 for all k as v = i=1 φi (v)vi + j=n+1 φj (v)vj , where one can obtain
nonzero φj (v) for some n + 1 ≤ j ≤ l but zero φi (v) for all 1 ≤ i ≤ n. Such
v ∈ null Γ, contradicting the hypothesis that Γ is injective.
(b) ⇒ Let K be the subspace spanned by the linearly independent Pm list of
vectors φ1 , . . . , φm . Then for all (a1 , . . . , am ) ∈ Fm , take v = i=1 ai vi for the
corresponding basis v1 , . . . , vm . Then we have that φi (v) = ai so Γ is surjective.
⇐ Suppose for the sake of contradiction that φ1 , . . . , φm are not linearly
independent, then we know there exists nonzero a′i s such that a1 φ1 +· · ·+am φm =
0. Applying any vector v ∈ V to this linear functional yields that
m
X
ai φi (v) = 0(v) = 0
i=1

This equals that

m
X m
X
ai φi (v) = ai Γ(v)i = 0
i=1 i=1
Since there is nonzero ai ’s, this means that the image of Γ is a strict subspace
of Fm , therefore contradicting the hypothesis that it’s surjective.

Problem 26
Suppose V is finite-dimensional and Ω is a subspace of V ′ . Prove that

Ω = {v ∈ V : φ(v) = 0 for every φ ∈ Ω}0

Proof. Denote U = {v ∈ V : φ(v) = 0 for every φ ∈ Ω}. So we have that

U 0 = {ψ ∈ V ′ : ψ(v) = 0 for every v ∈ U }

⇒ Take φ ∈ Ω and v ∈ U , then by definition φ(v) = 0 and thus φ ∈ U 0 .
⇐ Take ψ ∈ U 0 , suppose ψ ∈ / Ω. Take v ∈ U , then we know that for every
φ ∈ Ω, φ(v) = 0. Since ψ ∈ / Ω, there exists v, s.t. ψ(v) ̸= 0. However, this
contradicts that ψ(v) = 0 for all v ∈ U , the proof is completed.

Problem 28
Suppose V is finite-dimensional and φ1 , . . . , φm is a linearly independent
list in V ′ . Prove that

dim((null φi ) ∩ · · · ∩ (null φm )) = (dim V ) − m

Tm 0
Proof. Note
Tm that ( i=1 null φi ) = span(φ1 , . . . , φm ) by P23. Then we have
that dim i=1 null φi = dim V − span(φ1 , . . . , φm ) = dim V − m.

Problem 30
Suppose V is finite-dimensional and φ1 , . . . , φn is a basis of V ′ . Show
that there exists a basis of V whose dual basis is φ1 , . . . , φn .

48
Proof. We know that dim V ′ = n and that dim null φi + 1(dim range φi ) = n.
Thus we can construct that V = null φi ⊕ Ui for all 1 ≤ i ≤ n. Here dim Ui = 1.
Here we extract all vi′ s from those Ui . By linear map lemma, there exists vi ∈ Ui
such that φi (vi ) = 1. We claim that v1 , . . . , vn is the corresponding basis of V .
We first verify that it is the “corresponding basis”. Consider φj (vi ) for j ̸= i.
Suppose for contradiction that there exists nonzero w ∈ Ui ∩ Uk . Since we
know all Ui are 1-d, Ui = Uk . Here we have that null φi = null φk , then this
means that φi = cφk for some c ∈ F, but this contradicts the assumption that
φ1 , . . . , φn is the basis of V ′ .
Next we verify it is basis. ItP only suffices to verify that the list is linearly
m
independent. To Pm see this, 0 = i=1 ai vi . Apply φi on both side yield that
0 = φi (0) = φ( i=1 ai vi ) = ai . Thus all coefficients are zero.

Problem 32
The double dual space of V , denoted by V ′′ , is defined to be the dual
space of V ′ . In other words, V ′′ = (V ′ )′ . Define Λ : V → V ′′ by

(Λv)(φ) = φ(v)

for each v ∈ V and φ ∈ V ′ .

(a) Show that Λ is a linear map from V to V ′′ .
(b) Show that if T ∈ L(V ), then T ′′ ◦ Λ = Λ ◦ T , where T ′′ = (T ′ )′ .
(c) Show that if V is finite-dimensional, then Λ is an isomorphism from
V onto V ′′ .

Proof. (a) First, Λ(0)(φ) = φ(0) = 0. Next, take v1 , v2 ∈ V , then Λ(λv1 +

v2 )(φ) = λφ(v1 ) + φ(v2 ) = λΛ(v1 ) + Λ(v2 ).
(b) By definition, the double dual map T ′′ : V ′′ → W ′′ is defined as T ′′ (ψ) =
ψ(T ′ (φ)) for ψ ∈ V ′′ and for all φ ∈ W ′ . This further equals that T ′′ (ψ) =
ψφ ◦ T (v) for v ∈ V .

T ′′ ◦ (Λv)(φ) = (Λv) ◦ T ′ (φ)

= (Λv)φ ◦ T
=Λ◦T

(c) First we have that dim V = dim V ′ = dim V ′′ . Note that if we take
v ∈ null Λ, then this means for all linear functional φ ∈ V ′ , we have that Λv(φ) =
φ(v) = 0. So v = 0 and thus Λ is injective and therefore an isomorphism.

49