CHAPTERC11

HAP T ER 1 1

Linear Functionals
and the Dual Space
11.1 Introduction
In this chapter, we study linear mappings from a vector space V into its field K of scalars. (Unless
otherwise stated or implied, we view K as a vector space over itself.) Naturally all the theorems and
results for arbitrary mappings on V hold for this special case. However, we treat these mappings
separately because of their fundamental importance and because the special relationship of V to K gives
rise to new notions and results that do not apply in the general case.

11.2 Linear Functionals and the Dual Space

Let V be a vector space over a field K. A mapping f:V ! K is termed a linear functional (or linear form)
if, for every u; v 2 V and every a; b; 2 K,
fðau þ bvÞ ¼ afðuÞ þ bfðvÞ
In other words, a linear functional on V is a linear mapping from V into K.
EXAMPLE 11.1
(a) Let pi :K n ! K be the ith projection mapping; that is, pi ða1 ; a2 ; . . . an Þ ¼ ai . Then pi is linear and so it is a linear
functional on K n .
(b) Let V be Ðthe vector space of polynomials in t over R. Let J:V ! R be the integral operator defined by
1
JðpðtÞÞ ¼ 0 pðtÞ dt. Recall that J is linear; and hence, it is a linear functional on V.
(c) Let V be the vector space of n-square matrices over K. Let T :V ! K be the trace mapping
T ðAÞ ¼ a11 þ a22 þ þ ann ; where A ¼ ½aij
That is, T assigns to a matrix A the sum of its diagonal elements. This map is linear (Problem 11.24), and so it is
a linear functional on V.
By Theorem 5.10, the set of linear functionals on a vector space V over a field K is also a vector
space over K, with addition and scalar multiplication defined by
ðf þ sÞðvÞ ¼ fðvÞ þ sðvÞ and ðkfÞðvÞ ¼ kfðvÞ
where f and s are linear functionals on V and k 2 K. This space is called the dual space of V and is
denoted by V *.
EXAMPLE 11.2 Let V ¼ K n , the vector space of n-tuples, which we write as column vectors. Then the dual space V * can
be identified with the space of row vectors. In particular, any linear functional f ¼ ða1 ; . . . ; an Þ in V * has the representation
T
fðx1 ; x2 ; . . . ; xn Þ ¼ ½a1 ; a2 ; . . . ; an ½x2 ; x2 ; . . . ; xn ¼ a1 x1 þ a2 x2 þ þ an xn
Historically, the formal expression on the right was termed a linear form.

349
 350 CHAPTER 11 Linear Functionals and the Dual Space

11.3 Dual Basis

Suppose V is a vector space of dimension n over K. By Theorem 5.11, the dimension of the dual space V *
is also n (because K is of dimension 1 over itself). In fact, each basis of V determines a basis of V * as
follows (see Problem 11.3 for the proof).

THEOREM 11.1: Suppose fv 1 ; . . . ; v n g is a basis of V over K. Let f1 ; . . . ; fn 2 V * be the linear

functionals as defined by
1 if i ¼ j
fi ðv j Þ ¼ dij ¼
0 if i 6¼ j

Then ff1 ; . . . ; fn g is a basis of V *:

The above basis ffi g is termed the basis dual to fv i g or the dual basis. The above formula, which
uses the Kronecker delta dij , is a short way of writing
f1 ðv 1 Þ ¼ 1; f1 ðv 2 Þ ¼ 0; f1 ðv 3 Þ ¼ 0; . . . ; f1 ðv n Þ ¼ 0
f2 ðv 1 Þ ¼ 0; f2 ðv 2 Þ ¼ 1; f2 ðv 3 Þ ¼ 0; . . . ; f2 ðv n Þ ¼ 0
::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
fn ðv 1 Þ ¼ 0; fn ðv 2 Þ ¼ 0; . . . ; fn ðv n 1 Þ ¼ 0; fn ðv n Þ ¼ 1
By Theorem 5.2, these linear mappings fi are unique and well defined.

EXAMPLE 11.3 Consider the basis fv 1 ¼ ð2; 1Þ; v 2 ¼ ð3; 1Þg of R2 . Find the dual basis ff1 ; f2 g.
We seek linear functionals f1 ðx; yÞ ¼ ax þ by and f2 ðx; yÞ ¼ cx þ dy such that
f1 ðv 1 Þ ¼ 1; f1 ðv 2 Þ ¼ 0; f2 ðv 2 Þ ¼ 0; f2 ðv 2 Þ ¼ 1
These four conditions lead to the following two systems of linear equations:
f1 ðv 1 Þ ¼ f1 ð2; 1Þ ¼ 2a þ b ¼ 1 f2 ðv 1 Þ ¼ f2 ð2; 1Þ ¼ 2c þ d ¼ 0
and
f1 ðv 2 Þ ¼ f1 ð3; 1Þ ¼ 3a þ b ¼ 0 f2 ðv 2 Þ ¼ f2 ð3; 1Þ ¼ 3c þ d ¼ 1
The solutions yield a ¼ 1, b ¼ 3 and c ¼ 1, d ¼ 2. Hence, f1 ðx; yÞ ¼ x þ 3y and f2 ðx; yÞ ¼ x 2y form the
dual basis.

The next two theorems (proved in Problems 11.4 and 11.5, respectively) give relationships between
bases and their duals.

THEOREM 11.2: Let fv 1 ; . . . ; v n g be a basis of V and let ff1 ; . . . ; fn g be the dual basis in V *. Then
(i) For any vector u 2 V, u ¼ f1 ðuÞv 1 þ f2 ðuÞv 2 þ þ fn ðuÞv n .
(ii) For any linear functional s 2 V *, s ¼ sðv 1 Þf1 þ sðv 2 Þf2 þ þ sðv n Þfn .

THEOREM 11.3: Let fv 1 ; . . . ; v n g and fw1 ; . . . ; wn g be bases of V and let ff1 ; . . . ; fn g and
fs1 ; . . . ; sn g be the bases of V * dual to fv i g and fwi g, respectively. Suppose P is
the change-of-basis matrix from fv i g to fwi g. Then ðP 1 ÞT is the change-of-basis
matrix from ffi g to fsi g.

11.4 Second Dual Space

We repeat: Every vector space V has a dual space V *, which consists of all the linear functionals on V.
Thus, V * has a dual space V **, called the second dual of V, which consists of all the linear functionals
on V *.
We now show that each v 2 V determines a specific element v^ 2 V **. First, for any f 2 V *, we define
v^ðfÞ ¼ fðvÞ
CHAPTER 11 Linear Functionals and the Dual Space 351

It remains to be shown that this map v^:V * ! K is linear. For any scalars a; b 2 K and any linear
functionals f; s 2 V *, we have
v^ðaf þ bsÞ ¼ ðaf þ bsÞðvÞ ¼ afðvÞ þ bsðvÞ ¼ a^
v ðfÞ þ b^
v ðsÞ
That is, v^ is linear and so v^ 2 V **. The following theorem (proved in Problem 12.7) holds.

THEOREM 11.4: If V has finite dimensions, then the mapping v 7! v^ is an isomorphism of V

onto V **.

The above mapping v 7! v^ is called the natural mapping of V into V **. We emphasize that this
mapping is never onto V ** if V is not finite-dimensional. However, it is always linear, and moreover, it is
always one-to-one.
Now suppose V does have finite dimension. By Theorem 11.4, the natural mapping determines an
isomorphism between V and V **. Unless otherwise stated, we will identify V with V ** by this
mapping. Accordingly, we will view V as the space of linear functionals on V * and write V ¼ V **. We
remark that if ffi g is the basis of V * dual to a basis fv i g of V, then fv i g is the basis of V ** ¼ V that is
dual to ffi g.

11.5 Annihilators
Let W be a subset (not necessarily a subspace) of a vector space V. A linear functional f 2 V * is called
an annihilator of W if fðwÞ ¼ 0 for every w 2 W—that is, if fðW Þ ¼ f0g. We show that the set of all
such mappings, denoted by W 0 and called the annihilator of W, is a subspace of V *. Clearly, 0 2 W 0 :
Now suppose f; s 2 W 0 . Then, for any scalars a; b; 2 K and for any w 2 W,
ðaf þ bsÞðwÞ ¼ afðwÞ þ bsðwÞ ¼ a0 þ b0 ¼ 0
Thus, af þ bs 2 W 0 , and so W 0 is a subspace of V *.
In the case that W is a subspace of V, we have the following relationship between W and its annihilator
W 0 (see Problem 11.11 for the proof).

THEOREM 11.5: Suppose V has finite dimension and W is a subspace of V. Then

ðiÞ dim W þ dim W 0 ¼ dim V and ðiiÞ W 00 ¼ W
0
Here W 00 ¼ fv 2 V :fðvÞ ¼ 0 for every f 2 W 0 g or, equivalently, W 00 ¼ ðW 0 Þ , where W 00 is viewed
as a subspace of V under the identification of V and V **.

11.6 Transpose of a Linear Mapping

Let T :V ! U be an arbitrary linear mapping from a vector space V into a vector space U . Now for any
linear functional f 2 U *, the composition f T is a linear mapping from V into K:

That is, f T 2 V *. Thus, the correspondence

f 7! f T
is a mapping from U * into V *; we denote it by T t and call it the transpose of T. In other words,
T t :U * ! V * is defined by
T t ðfÞ ¼ f T
Thus, ðT t ðfÞÞðvÞ ¼ fðTðvÞÞ for every v 2 V.
 352 CHAPTER 11 Linear Functionals and the Dual Space

THEOREM 11.6: The transpose mapping T t defined above is linear.

Proof. For any scalars a; b 2 K and any linear functionals f; s 2 U*,

T t ðaf þ bsÞ ¼ ðaf þ bsÞ T ¼ aðf T Þ þ bðs T Þ ¼ aT t ðfÞ þ bT t ðsÞ
That is, T t is linear, as claimed.
We emphasize that if T is a linear mapping from V into U , then T t is a linear mapping from U * into
V *. The same ‘‘transpose’’ for the mapping T t no doubt derives from the following theorem (proved in
Problem 11.16).

THEOREM 11.7: Let T :V ! U be linear, and let A be the matrix representation of T relative to bases
fv i g of V and fui g of U . Then the transpose matrix AT is the matrix representation of
T t :U * ! V * relative to the bases dual to fui g and fv i g.

SOLVED PROBLEMS

Dual Spaces and Dual Bases

11.1. Find the basis ff1 ; f2 ; f3 g that is dual to the following basis of R3 :
fv 1 ¼ ð1; 1; 3Þ; v 2 ¼ ð0; 1; 1Þ; v 3 ¼ ð0; 3; 2Þg

The linear functionals may be expressed in the form

f1 ðx; y; zÞ ¼ a1 x þ a2 y þ a3 z; f2 ðx; y; zÞ ¼ b1 x þ b2 y þ b3 z; f3 ðx; y; zÞ ¼ c1 x þ c2 y þ c3 z
By definition of the dual basis, fi ðv j Þ ¼ 0 for i 6¼ j, but fi ðv j Þ ¼ 1 for i ¼ j.
We find f1 by setting f1 ðv 1 Þ ¼ 1; f1 ðv 2 Þ ¼ 0; f1 ðv 3 Þ ¼ 0: This yields
f1 ð1; 1; 3Þ ¼ a1 a2 þ 3a3 ¼ 1; f1 ð0; 1; 1Þ ¼ a2 a3 ¼ 0; f1 ð0; 3; 2Þ ¼ 3a2 2a3 ¼ 0
Solving the system of equations yields a1 ¼ 1, a2 ¼ 0, a3 ¼ 0. Thus, f1 ðx; y; zÞ ¼ x.
We find f2 by setting f2 ðv 1 Þ ¼ 0, f2 ðv 2 Þ ¼ 1, f2 ðv 3 Þ ¼ 0. This yields
f2 ð1; 1; 3Þ ¼ b1 b2 þ 3b3 ¼ 0; f2 ð0; 1; 1Þ ¼ b2 b3 ¼ 1; f2 ð0; 3; 2Þ ¼ 3b2 2b3 ¼ 0
Solving the system of equations yields b1 ¼ 7, b2 ¼ 2, a3 ¼ 3. Thus, f2 ðx; y; zÞ ¼ 7x 2y 3z.
We find f3 by setting f3 ðv 1 Þ ¼ 0, f3 ðv 2 Þ ¼ 0, f3 ðv 3 Þ ¼ 1. This yields
f3 ð1; 1; 3Þ ¼ c1 c2 þ 3c3 ¼ 0; f3 ð0; 1; 1Þ ¼ c2 c3 ¼ 0; f3 ð0; 3; 2Þ ¼ 3c2 2c3 ¼ 1
Solving the system of equations yields c1 ¼ 2, c2 ¼ 1, c3 ¼ 1. Thus, f3 ðx; y; zÞ ¼ 2x þ y þ z.

11.2. Let V ¼ fa þ bt : a; b 2 Rg, the vector space of real polynomials of degree 1. Find the basis
fv 1 ; v 2 g of V that is dual to the basis ff1 ; f2 g of V * defined by
ð1 ð2
f1 ð f ðtÞÞ ¼ f ðtÞ dt and f2 ð f ðtÞÞ ¼ f ðtÞ dt
0 0

Let v 1 ¼ a þ bt and v 2 ¼ c þ dt. By definition of the dual basis,

f1 ðv 1 Þ ¼ 1; f1 ðv 2 Þ ¼ 0 and f2 ðv 1 Þ ¼ 0; fi ðv j Þ ¼ 1
Thus,
Ð1 1
) Ð1 1
)
f1 ðv 1 Þ ¼ 0 ða þ btÞ dt ¼ a þ 2 b ¼ 1 f1 ðv 2 Þ ¼ 0 ðc þ dtÞ dt ¼ c þ 2 d ¼ 0
Ð2 and Ð2
f2 ðv 1 Þ ¼ 0 ða þ btÞ dt ¼ 2a þ 2b ¼ 0 f2 ðv 2 Þ ¼ 0 ðc þ dtÞ dt ¼ 2c þ 2d ¼ 1
1 1
Solving each system yields a ¼ 2, b ¼ 2 and c ¼ 2
, d ¼ 1. Thus, fv 1 ¼ 2 2t; v 2 ¼ 2
þ tg is
the basis of V that is dual to ff1 ; f2 g.
CHAPTER 11 Linear Functionals and the Dual Space 353

11.3. Prove Theorem 11.1: Suppose fv 1 ; . . . ; v n g is a basis of V over K. Let f1 ; . . . ; fn 2 V * be

defined by fi ðv j Þ ¼ 0 for i 6¼ j, but fi ðv j Þ ¼ 1 for i ¼ j. Then ff1 ; . . . ; fn g is a basis of V *.
We first show that ff1 ; . . . ; fn g spans V *. Let f be an arbitrary element of V *, and suppose
fðv 1 Þ ¼ k1 ; fðv 2 Þ ¼ k2 ; ...; fðv n Þ ¼ kn
Set s ¼ k1 f1 þ þ kn fn . Then

sðv 1 Þ ¼ ðk1 f1 þ þ kn fn Þðv 1 Þ ¼ k1 f1 ðv 1 Þ þ k2 f2 ðv 1 Þ þ þ kn fn ðv 1 Þ

¼ k1 1 þ k2 0 þ þ kn 0 ¼ k1
Similarly, for i ¼ 2; . . . ; n,
sðv i Þ ¼ ðk1 f1 þ þ kn fn Þðv i Þ ¼ k1 f1 ðv i Þ þ þ ki fi ðv i Þ þ þ kn fn ðv i Þ ¼ ki
Thus, fðv i Þ ¼ sðv i Þ for i ¼ 1; . . . ; n. Because f and s agree on the basis vectors,
f ¼ s ¼ k1 f1 þ þ kn fn . Accordingly, ff1 ; . . . ; fn g spans V *.
It remains to be shown that ff1 ; . . . ; fn g is linearly independent. Suppose
a1 f1 þ a2 f2 þ þ a n fn ¼ 0
Applying both sides to v 1 , we obtain
0 ¼ 0ðv 1 Þ ¼ ða1 f1 þ þ an fn Þðv 1 Þ ¼ a1 f1 ðv 1 Þ þ a2 f2 ðv 1 Þ þ þ an fn ðv 1 Þ
¼ a1 1 þ a2 0 þ þ an 0 ¼ a1
Similarly, for i ¼ 2; . . . ; n,
0 ¼ 0ðv i Þ ¼ ða1 f1 þ þ an fn Þðv i Þ ¼ a1 f1 ðv i Þ þ þ ai fi ðv i Þ þ þ an fn ðv i Þ ¼ ai
That is, a1 ¼ 0; . . . ; an ¼ 0. Hence, ff1 ; . . . ; fn g is linearly independent, and so it is a basis of V *.

11.4. Prove Theorem 11.2: Let fv 1 ; . . . ; v n g be a basis

P of V and let ff1 ; .P
. . ; fn g be the dual basis in
V *. For any u 2 V and any s 2 V *, (i) u ¼ i fi ðuÞv i . (ii) s ¼ i fðv i Þfi .
Suppose
u ¼ a1 v 1 þ a2 v 2 þ þ an v n ð1Þ
Then
f1 ðuÞ ¼ a1 f1 ðv 1 Þ þ a2 f1 ðv 2 Þ þ þ an f1 ðv n Þ ¼ a1 1 þ a2 0 þ þ an 0 ¼ a1
Similarly, for i ¼ 2; . . . ; n,
fi ðuÞ ¼ a1 fi ðv 1 Þ þ þ ai fi ðv i Þ þ þ an fi ðv n Þ ¼ ai
That is, f1 ðuÞ ¼ a1 , f2 ðuÞ ¼ a2 ; . . . ; fn ðuÞ ¼ an . Substituting these results into (1), we obtain (i).
Next we prove ðiiÞ. Applying the linear functional s to both sides of (i),
sðuÞ ¼ f1 ðuÞsðv 1 Þ þ f2 ðuÞsðv 2 Þ þ þ fn ðuÞsðv n Þ
¼ sðv 1 Þf1 ðuÞ þ sðv 2 Þf2 ðuÞ þ þ sðv n Þfn ðuÞ
¼ ðsðv 1 Þf1 þ sðv 2 Þf2 þ þ sðv n Þfn ÞðuÞ
Because the above holds for every u 2 V, s ¼ sðv 1 Þf2 þ sðv 2 Þf2 þ þ sðv n Þfn , as claimed.

11.5. Prove Theorem 11.3. Let fv i g and fwi g be bases of V and let ffi g and fsi g be the respective
T
dual bases in V *. Let P be the change-of-basis matrix from fv i g to fwi g: Then ðP 1 Þ is the
change-of-basis matrix from ffi g to fsi g.
Suppose, for i ¼ 1; . . . ; n,
wi ¼ ai1 v 1 þ ai2 v 2 þ þ ain v n and si ¼ bi1 f1 þ bi2 f2 þ þ ain v n
T
Then P ¼ ½aij and Q ¼ ½bij . We seek to prove that Q ¼ ðP 1 Þ .
Let Ri denote the ith row of Q and let Cj denote the jth column of PT . Then
T
Ri ¼ ðbi1 ; bi2 ; . . . ; bin Þ and Cj ¼ ðaj1 ; aj2 ; . . . ; ajn Þ
 354 CHAPTER 11 Linear Functionals and the Dual Space

By definition of the dual basis,

si ðwj Þ ¼ ðbi1 f1 þ bi2 f2 þ þ bin fn Þðaj1 v 1 þ aj2 v 2 þ þ ajn v n Þ
¼ bi1 aj1 þ bi2 aj2 þ þ bin ajn ¼ Ri Cj ¼ dij
where dij is the Kronecker delta. Thus,
QPT ¼ ½Ri Cj ¼ ½dij ¼ I
1 T
Therefore, Q ¼ ðPT Þ ¼ ðP 1 Þ , as claimed.

11.6. Suppose v 2 V, v 6¼ 0, and dim V ¼ n. Show that there exists f 2 V * such that fðvÞ 6¼ 0.
We extend fvg to a basis fv; v 2 ; . . . ; v n g of V. By Theorem 5.2, there exists a unique linear mapping
f:V ! K such that fðvÞ ¼ 1 and fðv i Þ ¼ 0, i ¼ 2; . . . ; n. Hence, f has the desired property.

11.7. Prove Theorem 11.4: Suppose dim V ¼ n. Then the natural mapping v 7! v^ is an isomorphism of
V onto V **.
We first prove that the map v 7! v^ is linear—that is, for any vectors v; w 2 V and any scalars a; b 2 K,
d
av þ bw ¼ a^
v þ bw.
^ For any linear functional f 2 V *,
d
av þ bwðfÞ ¼ fðav þ bwÞ ¼ afðvÞ þ bfðwÞ ¼ a^ ^
v ðfÞ þ bwðfÞ ¼ ða^ ^
v þ bwÞðfÞ
Because av d þ bwðfÞ ¼ ða^v þ bwÞðfÞ
^ for every f 2 V *, we have av d þ bw ¼ a^ v þ bw.
^ Thus, the map
v 7! v^ is linear.
Now suppose v 2 V, v 6¼ 0. Then, by Problem 11.6, there exists f 2 V * for which fðvÞ 6¼ 0. Hence,
v^ðfÞ ¼ fðvÞ 6¼ 0, and thus v^ 6¼ 0. Because v 6¼ 0 implies v^ 6¼ 0, the map v 7! v^ is nonsingular and hence
an isomorphism (Theorem 5.64).
Now dim V ¼ dim V * ¼ dim V **, because V has finite dimension. Accordingly, the mapping v 7! v^
is an isomorphism of V onto V **.

Annihilators
11.8. Show that if f 2 V * annihilates a subset S of V, then f annihilates the linear span LðSÞ of S.
0
Hence, S 0 ¼ ½spanðSÞ .
Suppose v 2 spanðSÞ. Then there exists w1 ; . . . ; wr 2 S for which v ¼ a1 w1 þ a2 w2 þ þ a r wr .
fðvÞ ¼ a1 fðw1 Þ þ a2 fðw2 Þ þ þ ar fðwr Þ ¼ a1 0 þ a2 0 þ þ ar 0 ¼ 0
Because v was an arbitrary element of spanðSÞ; f annihilates spanðSÞ, as claimed.

11.9. Find a basis of the annihilator W 0 of the subspace W of R4 spanned by

v 1 ¼ ð1; 2; 3; 4Þ and v 2 ¼ ð0; 1; 4; 1Þ
By Problem 11.8, it suffices to find a basis of the set of linear functionals f such that fðv 1 Þ ¼ 0 and
fðv 2 Þ ¼ 0, where fðx1 ; x2 ; x3 ; x4 Þ ¼ ax1 þ bx2 þ cx3 þ dx4 . Thus,
fð1; 2; 3; 4Þ ¼ a þ 2b 3c þ 4d ¼ 0 and fð0; 1; 4; 1Þ ¼ b þ 4c d¼0
The system of two equations in the unknowns a; b; c; d is in echelon form with free variables c and d.
(1) Set c ¼ 1, d ¼ 0 to obtain the solution a ¼ 11, b ¼ 4, c ¼ 1, d ¼ 0.
(2) Set c ¼ 0, d ¼ 1 to obtain the solution a ¼ 6, b ¼ 1, c ¼ 0, d ¼ 1.
The linear functions f1 ðxi Þ ¼ 11x1 4x2 þ x3 and f2 ðxi Þ ¼ 6x1 x2 þ x4 form a basis of W 0 .

11.10. Show that (a) For any subset S of V ; S S 00 . (b) If S1 S2 , then S20 S10 .
0
(a) Let v 2 S. Then for every linear functional f 2 S 0 , v^ðfÞ ¼ fðvÞ ¼ 0. Hence, v^ 2 ðS 0 Þ . Therefore,
00 00
under the identification of V and V **, v 2 S . Accordingly, S S .
(b) Let f 2 S20 . Then fðvÞ ¼ 0 for every v 2 S2 . But S1 S2 ; hence, f annihilates every element of S1
(i.e., f 2 S10 ). Therefore, S20 S10 .
CHAPTER 11 Linear Functionals and the Dual Space 355

11.11. Prove Theorem 11.5: Suppose V has finite dimension and W is a subspace of V. Then
(i) dim W þ dim W 0 ¼ dim V, (ii) W 00 ¼ W.
(i) Suppose dim V ¼ n and dim W ¼ r n. We want to show that dim W 0 ¼ n r. We choose a basis
fw1 ; . . . ; wr g of W and extend it to a basis of V, say fw1 ; . . . ; wr ; v 1 ; . . . ; v n r g. Consider the dual
basis
ff1 ; . . . ; fr ; s1 ; . . . ; sn r g
By definition of the dual basis, each of the above s’s annihilates each wi ; hence, s1 ; . . . ; sn r 2 W 0 .
We claim that fsi g is a basis of W 0 . Now fsj g is part of a basis of V *, and so it is linearly
independent.
We next show that ffj g spans W 0 . Let s 2 W 0 . By Theorem 11.2,
s ¼ sðw1 Þf1 þ þ sðwr Þfr þ sðv 1 Þs1 þ þ sðv n r Þsn r
¼ 0f1 þ þ 0fr þ sðv 1 Þs1 þ þ sðv n r Þsn r
¼ sðv 1 Þs1 þ þ sðv n r Þsn r

Consequently, fs1 ; . . . ; sn r g spans W 0 and so it is a basis of W 0 . Accordingly, as required

dim W 0 ¼ n r ¼ dim V dim W :

(ii) Suppose dim V ¼ n and dim W ¼ r. Then dim V * ¼ n and, by (i), dim W 0 ¼ n r. Thus, by (i),
dim W 00 ¼ n ðn rÞ ¼ r; therefore, dim W ¼ dim W 00 . By Problem 11.10, W W 00 . Accord-
ingly, W ¼ W 00 .

0
11.12. Let U and W be subspaces of V. Prove that ðU þ W Þ ¼ U 0 \ W 0 .
0
Let f 2 ðU þ W Þ . Then f annihilates U þ W; and so, in particular, f annihilates U and W: That is,
f 2 U 0 and f 2 W 0 ; hence, f 2 U 0 \ W 0 : Thus, ðU þ W Þ0 U 0 \ W 0 :
On the other hand, suppose s 2 U 0 \ W 0 : Then s annihilates U and also W. If v 2 U þ W, then
v ¼ u þ w, where u 2 U and w 2 W. Hence, sðvÞ ¼ sðuÞ þ sðwÞ ¼ 0 þ 0 ¼ 0. Thus, s annihilates U þ W;
0 0
that is, s 2 ðU þ W Þ . Accordingly, U 0 þ W 0 ðU þ W Þ .
The two inclusion relations together give us the desired equality.

Remark: Observe that no dimension argument is employed in the proof; hence, the result holds for
spaces of finite or infinite dimension.

Transpose of a Linear Mapping

11.13. Let f be the linear functional on R2 defined by fðx; yÞ ¼ x 2y. For each of the following linear
operators T on R2 , find ðT t ðfÞÞðx; yÞ:
(a) T ðx; yÞ ¼ ðx; 0Þ, (b) Tðx; yÞ ¼ ðy; x þ yÞ, (c) T ðx; yÞ ¼ ð2x 3y; 5x þ 2yÞ
t t
By definition, T ðfÞ ¼ f T ; that is, ðT ðfÞÞðvÞ ¼ fðT ðvÞÞ for every v. Hence,
(a) ðT t ðfÞÞðx; yÞ ¼ fðT ðx; yÞÞ ¼ fðx; 0Þ ¼ x
(b) ðT t ðfÞÞðx; yÞ ¼ fðT ðx; yÞÞ ¼ fðy; x þ yÞ ¼ y 2ðx þ yÞ ¼ 2x y
(c) ðT t ðfÞÞðx; yÞ ¼ fðT ðx; yÞÞ ¼ fð2x 3y; 5x þ 2yÞ ¼ ð2x 3yÞ 2ð5x þ 2yÞ ¼ 8x 7y

11.14. Let T :V ! U be linear and let T t :U * ! V * be its transpose. Show that the kernel of T t is the
0
annihilator of the image of T—that is, Ker T t ¼ ðIm T Þ .
Suppose f 2 Ker T t ; that is, T t ðfÞ ¼ f T ¼ 0. If u 2 Im T, then u ¼ T ðvÞ for some v 2 V ; hence,
fðuÞ ¼ fðT ðvÞÞ ¼ ðf T ÞðvÞ ¼ 0ðvÞ ¼ 0
We have that fðuÞ ¼ 0 for every u 2 Im T ; hence, f 2 ðIm T Þ0 . Thus, Ker T t ðIm T Þ0 .
On the other hand, suppose s 2 ðIm T Þ0 ; that is, sðIm T Þ ¼ f0g . Then, for every v 2 V,
ðT t ðsÞÞðvÞ ¼ ðs TÞðvÞ ¼ sðT ðvÞÞ ¼ 0 ¼ 0ðvÞ
 356 CHAPTER 11 Linear Functionals and the Dual Space

We have ðT t ðsÞÞðvÞ ¼ 0ðvÞ for every v 2 V ; hence, T t ðsÞ ¼ 0. Thus, s 2 Ker T t , and so
0
ðIm TÞ Ker T t .
The two inclusion relations together give us the required equality.

11.15. Suppose V and U have finite dimension and T :V ! U is linear. Prove rankðT Þ ¼ rankðT t Þ.
Suppose dim V ¼ n and dim U ¼ m, and suppose rankðT Þ ¼ r. By Theorem 11.5,
0
dimðIm TÞ ¼ dim u dimðIm T Þ ¼ m rankðT Þ ¼ m r
t 0 t
By Problem 11.14, Ker T ¼ ðIm T Þ . Hence, nullity ðT Þ ¼ m r. It then follows that, as claimed,
t t
rankðT Þ ¼ dim U * nullityðT Þ ¼ m ðm rÞ ¼ r ¼ rankðT Þ

11.16. Prove Theorem 11.7: Let T :V ! U be linear and let A be the matrix representation of T in the
bases fv j g of V and fui g of U . Then the transpose matrix AT is the matrix representation of
T t :U * ! V * in the bases dual to fui g and fv j g.
Suppose, for j ¼ 1; . . . ; m,
T ðv j Þ ¼ aj1 u1 þ aj2 u2 þ þ ajn un ð1Þ

We want to prove that, for i ¼ 1; . . . ; n,

T t ðsi Þ ¼ a1i f1 þ a2i f2 þ þ ami fm ð2Þ
where fsi g and ffj g are the bases dual to fui g and fv j g, respectively.
Let v 2 V and suppose v ¼ k1 v 1 þ k2 v 2 þ þ km v m . Then, by (1),
T ðvÞ ¼ k1 T ðv 1 Þ þ k2 T ðv 2 Þ þ þ km T ðv m Þ
¼ k1 ða11 u1 þ þ a1n un Þ þ k2 ða21 u1 þ þ a2n un Þ þ þ km ðam1 u1 þ þ amn un Þ
¼ ðk1 a11 þ k2 a21 þ þ km am1 Þu1 þ þ ðk1 a1n þ k2 a2n þ þ km amn Þun
Pn
¼ ðk1 a1i þ k2 a2i þ þ km ami Þui
i¼1

Hence, for j ¼ 1; . . . ; n.
n
P
ðT t ðsj ÞðvÞÞ ¼ sj ðT ðvÞÞ ¼ sj ðk1 a1i þ k2 a2i þ þ km ami Þui
i¼1
¼ k1 a1j þ k2 a2j þ þ km amj ð3Þ
On the other hand, for j ¼ 1; . . . ; n,
ða1j f1 þ a2j f2 þ þ amj fm ÞðvÞ ¼ ða1j f1 þ a2j f2 þ þ amj fm Þðk1 v 1 þ k2 v 2 þ þ km v m Þ
¼ k1 a1j þ k2 a2j þ þ km amj ð4Þ

Because v 2 V was arbitrary, (3) and (4) imply that

T t ðsj Þ ¼ a1j f1 þ a2j f2 þ þ amj fm ; j ¼ 1; . . . ; n

which is (2). Thus, the theorem is proved.

SUPPLEMENTARY PROBLEMS

Dual Spaces and Dual Bases

11.17. Find (a) f þ s, (b) 3f, (c) 2f 5s, where f:R3 ! R and s:R3 ! R are defined by

fðx; y; zÞ ¼ 2x 3y þ z and sðx; y; zÞ ¼ 4x 2y þ 3z

11.18. Find the dual basis of each of the following bases of R3 : (a) fð1; 0; 0Þ; ð0; 1; 0Þ; ð0; 0; 1Þg,
(b) fð1; 2; 3Þ; ð1; 1; 1Þ; ð2; 4; 7Þg.
CHAPTER 11 Linear Functionals and the Dual Space 357

11.19. Let V be the vector space of polynomials over R of degree 2. Let f1 ; f2 ; f3 be the linear functionals on
V defined by
ð1
f1 ð f ðtÞÞ ¼ f ðtÞ dt; f2 ð f ðtÞÞ ¼ f 0 ð1Þ; f3 ð f ðtÞÞ ¼ f ð0Þ
0

Here f ðtÞ ¼ a þ bt þ ct 2 V and f 0 ðtÞ denotes the derivative of f ðtÞ. Find the basis f f1 ðtÞ; f2 ðtÞ; f3 ðtÞg of
2

V that is dual to ff1 ; f2 ; f3 g.

11.20. Suppose u; v 2 V and that fðuÞ ¼ 0 implies fðvÞ ¼ 0 for all f 2 V *. Show that v ¼ ku for some scalar k.

11.21. Suppose f; s 2 V * and that fðvÞ ¼ 0 implies sðvÞ ¼ 0 for all v 2 V. Show that s ¼ kf for some scalar k.

11.22. Let V be the vector space of polynomials over K. For a 2 K, define fa :V ! K by fa ð f ðtÞÞ ¼ f ðaÞ. Show
that (a) fa is linear; (b) if a 6¼ b, then fa 6¼ fb .

11.23. Let V be the vector space of polynomials of degree 2. Let a; b; c 2 K be distinct scalars. Let fa ; fb ; fc
be the linear functionals defined by fa ð f ðtÞÞ ¼ f ðaÞ, fb ð f ðtÞÞ ¼ f ðbÞ, fc ð f ðtÞÞ ¼ f ðcÞ. Show that
ffa ; fb ; fc g is linearly independent, and find the basis f f1 ðtÞ; f2 ðtÞ; f3 ðtÞg of V that is its dual.

11.24. Let V be the vector space of square matrices of order n. Let T :V ! K be the trace mapping; that is,
T ðAÞ ¼ a11 þ a22 þ þ ann , where A ¼ ðaij Þ. Show that T is linear.

11.25. Let W be a subspace of V. For any linear functional f on W, show that there is a linear functional s on V
such that sðwÞ ¼ fðwÞ for any w 2 W ; that is, f is the restriction of s to W.

11.26. Let fe1 ; . . . ; en g be the usual basis of K n . Show that the dual basis is fp1 ; . . . ; pn g where pi is the ith
projection mapping; that is, pi ða1 ; . . . ; an Þ ¼ ai .

11.27. Let V be a vector space over R. Let f1 ; f2 2 V * and suppose s:V ! R; defined by sðvÞ ¼ f1 ðvÞf2 ðvÞ;
also belongs to V *. Show that either f1 ¼ 0 or f2 ¼ 0.

Annihilators
11.28. Let W be the subspace of R4 spanned by ð1; 2; 3; 4Þ, ð1; 3; 2; 6Þ, ð1; 4; 1; 8Þ. Find a basis of the
annihilator of W.

11.29. Let W be the subspace of R3 spanned by ð1; 1; 0Þ and ð0; 1; 1Þ. Find a basis of the annihilator of W.

11.30. Show that, for any subset S of V ; spanðSÞ ¼ S 00 , where spanðSÞ is the linear span of S.

11.31. Let U and W be subspaces of a vector space V of finite dimension. Prove that ðU \ W Þ0 ¼ U 0 þ W 0 .

11.32. Suppose V ¼ U W. Prove that V 0 ¼ U 0 W 0.

Transpose of a Linear Mapping

11.33. Let f be the linear functional on R2 defined by fðx; yÞ ¼ 3x 2y. For each of the following linear
mappings T :R3 ! R2 , find ðT t ðfÞÞðx; y; zÞ:
(a) T ðx; y; zÞ ¼ ðx þ y; y þ zÞ, (b) T ðx; y; zÞ ¼ ðx þ y þ z; 2x yÞ

11.34. Suppose T1 :U ! V and T2 :V ! W are linear. Prove that ðT2 T1 Þt ¼ T1t T2t .
0
11.35. Suppose T :V ! U is linear and V has finite dimension. Prove that Im T t ¼ ðKer T Þ .
 358 CHAPTER 11 Linear Functionals and the Dual Space

11.36. Suppose T :V ! U is linear and u 2 U. Prove that u 2 Im T or there exists f 2 V * such that T t ðfÞ ¼ 0
and fðuÞ ¼ 1.

11.37. Let V be of finite dimension. Show that the mapping T 7! T t is an isomorphism from HomðV ; V Þ onto
HomðV *; V *Þ. (Here T is any linear operator on V.)

Miscellaneous Problems
11.38. Let V be a vector space over R. The line segment uv joining points u; v 2 V is defined by
uv ¼ ftu þ ð1 tÞv : 0 t 1g. A subset S of V is convex if u; v 2 S implies uv S. Let f 2 V *. Define

W þ ¼ fv 2 V : fðvÞ > 0g; W ¼ fv 2 V : fðvÞ ¼ 0g; W ¼ fv 2 V : fðvÞ < 0g

Prove that W þ ; W, and W are convex.

11.39. Let V be a vector space of finite dimension. A hyperplane H of V may be defined as the kernel of a nonzero
linear functional f on V. Show that every subspace of V is the intersection of a finite number of
hyperplanes.

ANSWERS TO SUPPLEMENTARY PROBLEMS

11.17. (a) 6x 5y þ 4z, (b) 6x 9y þ 3z, (c) 16x þ 4y 13z

11.18. (a) f1 ¼ x; f2 ¼ y; f3 ¼ z; (b) f1 ¼ 3x 5y 2z; f2 ¼ 2x þ y; f3 ¼ x þ 2y þ z

3 2 1
11.19. f1 ðtÞ ¼ 3t 2t ; f2 ðtÞ ¼ 2t þ 34 t2 ; f3 ðtÞ ¼ 1 3t þ 32 t2

11.22. (b) Let f ðtÞ ¼ t. Then fa ð f ðtÞÞ ¼ a 6¼ b ¼ fb ð f ðtÞÞ; and therefore, fa 6¼ fb

t2 ðb þ cÞt þ bc t2 ða þ cÞt þ ac t2 ða þ bÞt þ ab
11.23. f1 ðtÞ ¼ ; f2 ðtÞ ¼ ; f3 ðtÞ ¼
ða bÞða cÞ ðb aÞðb cÞ ðc aÞðc bÞ
11.28. ff1 ðx; y; z; tÞ ¼ 5x y þ z; f2 ðx; y; z; tÞ ¼ 2y tg

11.29. ffðx; y; zÞ ¼ x y þ zg

11.33. (a) ðT t ðfÞÞðx; y; zÞ ¼ 3x þ y 2z, (b) ðT t ðfÞÞðx; y; zÞ ¼ x þ 5y þ 3z