Dual Spaces

Denition 1 (Dual Space) Let V be a nite dimensional vector space.

(a) A linear functional on V is a function u

: V IR that is linear in the sense that

u

(v + w) = u

(v) +u

( w) and u

(v) = u

(v)
for all u, w V and all IR.
(b) The dual space V

of the vector space V is the set of all linear functionals on V . It is itself a
vector space with the operations
(u

+v

)( w) = u

( w) +v

( w)
(v

)( w) = v

( w)
(c) If V has dimension d and has basis
_
e
1
, , e
d
_
then, by Problem 1, below, V

also has
dimension d and a basis for V

(called the dual basis of
_
e
1
, , e
d
_
) is
_
e

1
, , e

d
_
where, for
each 1 j d
e

j
_

1
e
1
+ +
d
e
d
_
=
j
That is,
e

j
(e
k
) =
_
1 if j = k
0 if j = k
Problem 1 Let
_
e
1
, , e
d
_
be a basis for the vector space V and dene, for each 1 j d, e

j
as in Denition 1.c. Prove that
_
e

1
, , e

d
_
is a basis for V

.
Problem 2 Let W be a proper subspace of the nite dimensional vector space V and let e be
an element of V that is not in W. Prove that there exists a linear functional w

V

such that
w

(e) = 1 and w

( w) = 0 for all w W.
Problem 3 Let V be a nite dimensional vector space. Each v V has naturally associated to
it the linear functional v

(V

)

(i.e. v

is a linear functional on the dual space V

) dened by
v

( w

) = w

(v) for all w

V

Prove that the map v v

is an isomorphism (i.e. 11, onto and linear) between the vector

spaces V and (V

)

. Consequently, for nite dimensional vector spaces, one usually thinks of V

and (V

)

as being the same.

Lemma 2 Let V be a vector space of dimension d < . Let v

V

. Its kernel
K =
_
v V

(v) = 0
_
is a linear subspace of V . If v

is not the zero functional, then the dimension of K is exactly d1.

October 9, 2008 Dual Spaces 1
Proof: That K is a linear subspace of V is obvious from the linearity of v

. Assume that v

is not the zero functional. Then there is some zector e

d
V with v

(e
d
) = 0. Choose any d 1
additional vectors

f
1
, ,

f
d1
such that

f
1
, ,

f
d1
, e
d
is a basis for V . For each 1 d 1,
set
e

=

f

)
v

(e
d
)
e
d
Then e
1
, , e
d1
, e
d
is again a basis for V and v

(e

) = 0 for all 1 d 1. Indeed K is

exactly the subspace spanned by e
1
, , e
d1
and hence has dimension d 1.
Denition 3 (transpose) Let V and W be nite dimensional vector spaces with dimensions d
V
and d
W
, respectively.
(a) A linear map (or linear transformation) from V to W is a function A : V W that is linear
in the sense that
A(v + w) = A(v) + A( w) and A(v) = A(v)
for all u, w V and all IR. Often, one writes Av in place of A(v).
(b) If
_
e
j
_
1jd
V
and
_

f
j
_
1jd
W
are bases for V and W respectively, then the matrix
_
A
i,j

1id
W
1jd
V
of the linear map A : V W is determined by
Ae
j
=
d
W

i=1
A
i,j

f
i
for all 1 j d
V
If v =

d
V
j=1
v
j
e
j
then
Av =
d
V

j=1
v
j
Ae
j
=
d
V

j=1
d
W

i=1
v
j
A
i,j

f
i
so that the components w
i
of the output vector Av = w =

d
W
i=1
w
i

f
i
are determined by
w
i
=
d
V

j=1
A
i,j
v
j
for each 1 i d
W
(c) The transpose of a linear transformation A : V W is the linear transformation A
t
: W

V

determined by
(A
t
w

)(v) = w

(Av)
Note that
for each xed w

and A, the right hand side is real valued and linear in v so that A
t
w

is a
legitimate element of V

and
the right hand side is linear in w

, so that A
t
is a linear transformation from W

to V

.
October 9, 2008 Dual Spaces 2
if v =

d
V
i=1
v
i
e
i
,
(A
t

j
)(v) =

f

j
(Av) =
d
V

i=1
v
i

j
(Ae
i
) =
d
V

i=1
d
W

=1
v
i
A
,i

j
(

) =
d
V

i=1
A
j,i
v
i
=
d
V

i=1
A
j,i
e

i
(v)
we have A
t

j
=

d
V
i=1
A
j,i
e

i
, so the matrix for the transpose is A
t
i,j
= A
j,i
, the usual matrix
transpose.
Lemma 4 Let V and W be nite dimensional spaces and let A : V W be linear. Denote the
transpose of A by A
t
: W

V

. Then
A
t
is 11 A is onto
A
t
is onto A is 11
Proof 1: We rst prove that A
t
is 11 A is onto.
A
t
w

is the zero element of V

if and only if (A
t
w

)(v) = w

(Av) is zero for all v V .

In the event that A is onto, then
_
Av

v V
_
= W, and w

(Av) is zero for all v V if and

only if w

( w) is zero for all w W which is the case if and only if w

is the zero element of

V

. This proves that A
t
is 11 when A is onto.
On the other hand if A is not onto, its range W

is some proper subspace of W. Let w

1
, , w
d
W
be a basis for W whose rst d
W
< d
W
vectors form a basis for W

. Then w

= w

d
W
(the
last linear functional in the dual basis) takes the value 1 when applied to w
d
W
and takes
the value 0 when applied to any element of W

. Thus when A is not onto, we can have

(A
t
w

d
W
)(v) = w

d
W
(Av) =

0 for all v V so that A
t
w

d
W
is the zero element of V

even
though w

d
W
is not the zero element of W

. So A
t
is not 11.
We now prove that A is 11 A
t
is onto.
Av is the zero element of W if and only if w

(Av) = (A
t
w

)(v) is zero for all w

.
In the event that A
t
is onto, then
_
A
t
w

_
= V

and (A
t
w

)(v) is zero for all

w

if and only if v

(v) is zero for all v

V

, which, in turn, is the case if and only if
v is the zero element of V . This proves that A is 11 when A
t
is onto.
On the other hand if A
t
is not onto, its range is some proper subspace of V

. Let v

1
, , v

be a basis for the range of A

t
. We are assuming that d

< d
V
= d
V
, the dimension of V .
Let v
1
, , v
d
V
be any basis for V . Then v =
1
v
1
+ +
d
V
v
d
V
has (A
t
w

)(v) = 0 for all

w

if and only if
0 = v

j
(
1
v
1
+ +
d
V
v
d
V
) =
d
V

=1
v

j
(v

for all 1 j d

This is a system of d

< d
V
homogeneous linear equations in the d
V
unknowns
1
, ,

d
V
. It necessarily has a solution
1
, ,
d
V
with at least one

nonzero. Then v =

1
v
1
+ +
d
V
v
d
V
is a nonzero vector with w

(Av) = (A
t
w

)(v) = 0 for all w

and
hence with Av = 0. So A is not 11.
October 9, 2008 Dual Spaces 3
Proof 2: Let
_
A
i,j

be the matrix of A in some bases for V and W. Denote by a

j
, 1 j d
V
the column vectors of that matrix. Then A is 1-1 if and only if the column vectors are all linearly
independent. That is, if the dimension of the space spanned by the column vectors (i.e. the rank
of A) is d
V
. The space spanned by the column vectors is exactly the range of A (expressed in the
basis of W). So A is onto if and only if the rank of A is d
W
. But the rank of a matrix is the same
as the rank of its transpose (i.e. the column rank of a matrix is the same as its row rank), so
A injective rank(A) = d
V
rank(A
t
) = d
V
A
t
surjective
A surjective rank(A) = d
W
rank(A
t
) = d
W
A
t
injective
October 9, 2008 Dual Spaces 4