Math 4330 Fall 2013 1 Dual Spaces

Dual Spaces
This section uses language and notation similar to the approach taken in the text
by Hoffman and Kunze, but gives a bit more detail and shows why we are mainly
interested in the case of finite dimensional vector spaces. There are many important
results on dual spaces that are in the exercises of this section. It is important to try
them all (as always, but especially true here), since we will use many of the concepts
in class or in future homework assignments.
Let V be a vector space over the field F . We define the dual space of V to be V ∗ =
HomF (V, F ) . The linear transformations in V ∗ are usually called linear functionals.
We begin with a few simple examples.

Example 1. Let F be any field and let V = F n×1 for a positive integer n . Given any
set of n elements of F , { a1 , . . . , an } define f : V −→ F by f (x1 , . . . , xn ) = a1
x1 +
∗
· · · + an xn . Note that the isomorphism L : F 1×n −→ HomF (F n×1 , F ) = F n×1 of
the first part of the section on the “Matrix of a Linear Transformation” asserts that
the examples f = LA , A = (a1 , . . . , an ) give all such examples.

Example 2. Let F be a field and n a positive integer. Define the trace Tr :

F n×n −→ F for a matrix A with entries Aij by Tr(A) = a11 + a22 + · · · + ann .

Example 3. Let F be an arbitrary field and let a ∈ F . Define Ea : F [x] −→ F

on the basis 1, x, x2 , . . . , xi , . . . by Ea (xi ) = ai . This linear functional is called
evaluation at a and is written Ea (f ) = f (a) .

Example 4. Let X = [0, 1] ⊆ R be the closed interval. Let C(X) denote the R 1 vector
space of all continuous functions on X with values in R . Define Int(f ) = 0 f (t)dt .
Int is a linear functional by the standard properties of definite integrals.

Let V be a vector space over the field F . Let B be a basis for V . We define the
dual set B ∗ = { b∗ | b ∈ B } where b∗ : V −→ F is defined for a ∈ B by

∗ 0 a 6= b
b (a) =
1 a=b.

That is, we have

b∗ (a) = δa,b (1)

where δa,b is the Kronecker delta function: 1 when a = b , 0 otherwise.

It will be useful to briefly study such collections of vectors and linear functionals
independently of how they might be constructed. Our applications below will be to
the pair B and B ∗ which we constructed above for B a basis of V .

09/26/19
Math 4330 Fall 2013 2 Dual Spaces

Definition 5. Let V be a vector space over a field F . Two subsets B = { bi | i ∈ I } ⊆

V and B 0 = { fi | i ∈ I } ⊆ V ∗ are said to form a dual pair if the duality equations
hold:

fi (bj ) = δi,j

for all i, j ∈ I .
Lemma 6. Let V be a vector space over F . Let B ⊆ V and B 0 ⊆ V ∗ form a dual
pair.

a) B and B 0 are linearly independent over F . Consequently |I| = |B| = |B 0 | .

b) If V is finite dimensional, then B is a basis of V if and only if B 0 is a basis of
V∗.
c) If B is a basis of V , then B 0 is a basis of V ∗ if and only if dim V is finite.
P
Proof. If i∈I αi bi = 0 in V , then applying fj to both sides of the equation yields
αj = αj fj (bj ) = 0 as all other
P terms on the left are 0 and fj (bj ) = 1 by the duality
∗
equations. Similarly, if i∈I βi fi = 0 in V , applying the linear functional to bj
yields βj = 0 . Now the function I −→ B given by i 7→ bi is onto by definition and is
one-to-one since the bi must be distinct as they are linearly independent, and a similar
argument holds for B 0 .
Now B is a basis for V if and only if B spans V . If dim V is finite, then by
our earlier computation, dim HomF (V, F ) = dim V · dim F , that is dim V ∗ = dim V .
Since we found dim V linearly independent elements in V ∗ , they must also span and
hence are a basis. Conversely, if dim V ∗ = |B 0 | , then the linearly independent set B
has the correct number of elements to be a basis of V as well.
For the last assertion, given a basis B we can define a linear functional θ : V −→ F
by θ(bi ) = 1 for all iP∈ I . We now show θ ∈ SpanF (B 0 ) if and only if V is finite
dimensional. If θ = i∈I αi fi , then by applying both sides to bj ∈ B , and arguing
as before, we obtain θ(bj ) = 1 = αj . That is, every coefficient αj = 1 for all j ∈ I .
However, the span of B 0 is the set of finite sums so if B infinite, θ ∈
/ SpanF (B 0 ) .
Remark 7. In fact, if V is a vector space with countable dimension, then V ∗ has
uncountable dimension. See Exercise 29.

We will now look a bit more carefully at the finite dimensional case. Let B =
{ e1 , . . . , en } be an ordered basis for V . Then as noted in Lemma 6, B ∗ is a ba-
sis for V ∗ .
Theorem 8. Let V be a finite dimensional vector space over F with ordered basis
B = { e1 , . . . , en } . Then the dual set B ∗ is the unique basis for V ∗ which satisfies

e∗i (ej ) = δij .

Further the following equations are valid:

09/26/19
Math 4330 Fall 2013 3 Dual Spaces

n
X
1. f = f (ei )e∗i for f ∈ V ∗ ,
i=1

n
X
2. v = e∗i (v)ei for v ∈ V .
i=1

Proof. As noted previously for dim V finite, B ∗ is a basis for V ∗ . Equation 1 was
verified in the proof of Lemma 6: the coefficient of the dual basis element b∗ is
precisely f (b) .
Pn by a similar (“dual”)∗ argument: Write v ∈ V in
The second equation is verified
terms of the basis B as v = i=1 αi ei . Now apply ej to the equation yielding

e∗j (v) = αj e∗j (ej ) + 0

as all but one term on the right hand side must be 0 with the remaining term αj · 1 .
That is, αj = ej (v) as asserted in equation 2.

Remark 9. 1. In the finite dimensional case the basis B ∗ is called the dual basis.

2. The linear functionals e∗j are nothing other than the coordinate functions with
respect to the ordered basis B .

Remark 10. The dual spaces V ∗ may seem a bit mysterious on first encounter, but
in fact there is a simple way to think about them. Let B be a basis for V . Then
there is an isomorphism
[ ]B : V −→ F (B)
given by taking coordinates with respect to B . That is, each vector v ∈ V corre-
sponds to a unique function [ v ]B which is non-zero on only finitely many elements of
B.
Similarly via the Universal Mapping Property for Bases, each linear transformation on
V , and in particular, each linear functional f : V −→ F , is uniquely determined by
what it does on the basis B : that is, by the function J f KB : B −→ F given by
J f KB (b) = f (b) for each b ∈ B or equivalently, J f KB is f restricted to B . This
yields an isomorphism
J KB : V ∗ −→ F B
given by the evaluation on the basis B . Note that if B is infinite, then F B is all
possible functions on B , so is different from F (B) .
If B is finite, then in fact we have [ f ]B∗ = J f KB by Theorem 8, first equation.
So in summary, if one fixes a basis B for V , then one may think of V is “functions
on B with values in F which have finite support” and of V ∗ as “all functions on B
with values in F ”.

Example 11 (Lagrange Interpolation). Let Pn ⊆ F [x] be the subspace of polynomials

of degree less than the positive integer n . Let c1 , . . . , cn be n distinct elements

09/26/19
Math 4330 Fall 2013 4 Dual Spaces

of F . Define π(x) = (x − c1 )(x − c2 ) · · · (x − cn ) as the product of all x − ci ,

πi (x) = π(x)/(x − ci ) and pi (x) = πi (x)/πi (ci ) . Let A = { p1 , . . . , pn } .
n o
Consider evaluation of the linear functionals E = Ec , Ec , . . . , Ecn ⊆ Pn ∗ . Then
1 2

Ec (pj ) = pj (ci )
i

= πj (ci )/πj (cj )

= δij .

Hence E = A∗ and A is linearly independent by Lemma 6. Since it has n = dim Pn

elements, it is a basis. Thus we have dual bases A∗ = E .
Now we also have the standard basis B = { 1, x, x2 , . . . , xn−1 } for Pn . What is the
change of basis matrix P = P (A, B) ? From what we did earlier we have the formulas

P = P (A, B)
 
= [ p 1 ]B [ p2 ]B · · · [ pn ]B

and

P −1 = P (B, A)
 
= [ 1 ]A [ x ]A · · · [ xn−1 ]A .

Now the inverse of P is easy to compute using the dual basis of A (the coordinate
functions with respect to A ) since A∗ = E . We have for the j -th column:

Ec (xj−1 )
   
1
c1 j−1
 E (xj−1 )   c j−1 
 j−1   c2  2
x = =  ..  .
 
A
 .. 
 .   . 
j−1
Ec (xj−1 )
n
cn

And finally the matrix of P −1

 
1 c1 c21 · · · c1n−1
 1 c c2 · · · c2n−1 
2 2
 .
 
 .. .. .. .. ..
 . . . . . 
2
1 cn cn · · · cnn−1

This invertible n × n matrix is called the Vandermonde matrix. Equation 2 in Theo-

rem 8 applied to the basis A is called the Lagrange interpolation formula: for f ∈ Pn
n
X
f = Ec (f )pi
i
i=1
Xn
= f (ci )pi .
i=1

09/26/19
Math 4330 Fall 2013 5 Dual Spaces

Annihilators

Let V be any vector space over the field F and let S ⊆ V be any subset. The
annihilator of S , S◦ , is the subset of linear functionals that vanish on every element
of S :

S◦ = { f ∈ V ∗ | f (s) = 0 for all s ∈ S } .

Lemma 12. Let S ⊆ V be a subset of the vector space V over F . Then

1. S◦ is a subspace of V ∗ .

2. S◦ = SpanF (S)◦ .

Proof. This is left as an easy exercise.

Theorem 13. Let V be a finite dimensional vector space over F . Let W ⊆ V be a

subspace. Then

dimF W + dimF W◦ = dimF V .

Proof. The basic idea is the same that was used to give the dimension formulas for
short exact sequences, quotient spaces, and the one with the kernel and image of a
linear transformation.
 for W : { v1 , . . . , v` } . Enlarge to a basis
Start with a basis
for V : B = v1 , . . . , v` , v`+1 , . . . , v`+k . So dim V = ` + k and dim W = ` . Then
B ∗ = v1∗ , . . . , v`∗ , v`+1
∗ ∗
, . . . , v`+k .
Now by the definition of the dual basis for 1 ≤◦ i ≤ ` and for ` + 1 ≤ j ≤ ` + k we
have e∗j (ei ) = 0 . Hence v`+1
∗ ∗
, . . . , v`+k ⊆W .
Next let f ∈ W◦ ⊆ V ∗ . By the first equation in Theorem 8 we have for any f ∈ V ∗
`+k
X
f= f (vi )vi∗ .
i=1

However since actually f ∈ W◦ we have f (vi ) = 0 when i ≤ ` and thus:

`+k
X
f= f (vi )vi∗ ,
i=`+1
 ∗ ∗
) showing that this set is a basis for W◦ . Finally,

that is, f ∈ Span( v`+1 , . . . , v`+k
◦
dim V = ` + k = dim W + dim W .

Remark 14. See Exercise 13 for two alternate proofs.

Proposition 15. Let V be a non-trivial vector space over F and let v ∈ V be

non-zero. Then there exists an f ∈ V ∗ so that f (v) = 1 .

09/26/19
Math 4330 Fall 2013 6 Dual Spaces

Proof. We use the Main Theorem on Bases: Let v be the non-zero vector in V .
Since { v } is linearly independent, we may enlarge it to a basis B containing v and
take the dual set B ∗ . Then v ∗ ∈ B ∗ and v ∗ (v) = 1 .

Double Dual

Next we repeat the process of taking the dual and find a striking natural relationship
with the original vector space. Let

∗ V be a vector space over the field F . We define
the double dual to be V ∗∗ = V ∗ = HomF (V ∗ , F ) . It is easy to find some elements
of V ∗∗ : for v ∈ V define

Ev (f ) = f (v)

for f ∈ V ∗ . The function Ev is called evaluation at v . It is easy to check that

Ev : V ∗ −→ F is a linear functional:

Ev (αf ) = (αf )(v)

= αf (v)
= αEv (f )

and for f, g ∈ V ∗

Ev (f + g) = (f + g)(v)
= f (v) + g(v)
= Ev (f ) + Ev (g) ,

That is, Ev is a linear functional precisely because of the definitions of addition and
scalar multiplication for functions.
Further, we can consider E itself as the function

E : V −→ V ∗∗

which sends v to Ev . We’ll keep using the notation Ev (rather than the more
common E(v) ). Note that this function E is also a linear transformation, that is, it
satisfies
Eαv = αEv
and
Eu+v = Eu + Ev
for all α ∈ F and u, v ∈ V . This is also easy to check, namely

Eαv (f ) = f (αv)
= αf (v)
= αEv (f )

09/26/19
Math 4330 Fall 2013 7 Dual Spaces

and
Eu+v (f ) = f (u + v)
= f (u) + f (v)
= Eu (f ) + Ev (f )
hold for all f ∈ V ∗ because each f is a linear transformation. Hence we have equality
of functions Eαv = αEv and Eu+v = Eu + Ev as they give the same value for each
element of their domain.
Next note that the linear transformation E is always one-to-one. To see this we need
to compute ker E . A vector v ∈ V is in ker E precisely when Ev (f ) = 0 for all
f ∈ V ∗ . But by Proposition 15 if v is not 0 , there will exist an f with f (v) = 1 .
Hence v ∈ ker E must be 0 .
For V finite dimensional, it is now clear that E : V −→ V ∗∗ is an isomomorphism
since dim V = dim V ∗ = dim V ∗∗ . We now show that this is the only case for which
E is an isomorphism.
Proposition 16. Let V be a vector space over the field F . If E : V −→ V ∗∗ is an
isomorphism, then
1. For every L ∈ V ∗∗ there exists a unique vector v ∈ V such that Ev = L .
2. For every basis A of V ∗ there exists a unique basis B of V so that B ∗ = A .
3. dim V = dim V ∗ = dim V ∗∗ .
4. V has finite dimension.

Proof. The first statement is just that E is one-to-one and onto.

The second assertion contains the real content of the proposition. Let A = { fi | i ∈ I }
be a basis for V ∗ . Consider A∗ ⊆ V ∗∗ and let B0 = E −1 (A∗ ) be the correspond-
ing subset of V . So we can write B0 = { vi | i ∈ I } where vi ∈ V is given by
Evi = fi∗ ∈ A∗ ⊆ V ∗∗ . Now A∗ is linearly independent, and as E and E −1 are
isomorphisms by our assumption, B0 is a linearly independent subset of V . Enlarge
B0 to a basis B for V and consider B ∗ ⊆ V ∗ .
Note that we have the following equations for our dual sets:
fj∗ (fi ) = δij
and
b∗ (a) = δab
for a, b ∈ B . Hence as Ev = fj∗ we have
j

fi (vj ) = Ev (fi )
j

= fj∗ (fi )
= δij
= vi∗ (vj ) .

09/26/19
Math 4330 Fall 2013 8 Dual Spaces

We show that B0 = B is a basis for V as follows: assume there exists an element

∗
b ∈ B which is not in B0 . Then b (vi ) = 0 for all vi ∈ B0 . Recall that A is a basis
for V ∗ by our original assumption. Thus there exist elements αi ∈ F , only finitely
many non-zero, so that
X
b∗ = αi f i .
i∈I

Applying this equation to any vj ∈ B0 yields

X
b∗ (vj ) = αi fi (vj )
i∈I
0 = αj fj (vj ) + 0
0 = αj
as b ∈/ B0 by assumption together with the usual argument for the right hand side.
Thus all coefficients αj are 0 and hence b = 0 , yielding a contradiction as bases never
contain 0 . Thus no b which is independent of B0 can exist. That is, B0 = B is
a basis for V . Further we see from the equations above that fi (vj ) = vi∗ (vj ) = δij
for all i, j ∈ I . Hence fi = vi∗ for all i since they agree on all basis elements in B .
That is, B ∗ = A which is the second statement.
The sets B , A and E(B) = A∗ all have the same cardinality as I , B is a basis for
V , A is a basis for V ∗ and as E is an isomorphism, E(B) = A∗ is a basis for V ∗∗ .
Thus all three vector spaces have the same dimension as given in the third statement.
Finally, the fourth assertion follows from Lemma 6 since if V has infinite dimension,
B ∗ is not a basis for V ∗ .

We summarize the final result of this discussion.

Theorem 17. Let V be a vector space over F . The evaluation map E : V −→ V ∗∗
is a one-to-one linear transformation which is an isomorphism if and only if V is
finite dimensional.

Proof. By the discussion just before Proposition 16, E is always injective. If V is

finite dimensional, this is enough to ensure that E is an isomorphism. If E is an
isomorphism, the previous proposition shows that V is finite dimensional.
Remark 18 (Identification). It is common, although perhaps slightly confusing at
first, to consider elements of V and V ∗∗ to be the same if they correspond under the
natural isomorphism E . We will do so but at the beginning will point out where the
identification is being made.
It is also clear that E preserves essentially everything. For example if X ⊆ Y are
subsets of V , then E(X) ⊆ E(Y ) . If W1 and W2 are subspaces, E(W1 + W2 ) =
E(W1 ) + E(W2 ) .
Proposition 19. Let V be a finite dimensional vector space with subspace W . Then
(W◦ )◦ = W .

09/26/19
Math 4330 Fall 2013 9 Dual Spaces

Proof. As for every element w of W and all f ∈ W◦ we have f (w) = 0 , it is clear

that we have:

W ⊆ { v ∈ V | f (v) = 0 ∀f ∈ W◦ }
= { v ∈ V | Ev (f ) = 0 ∀f ∈ W◦ }
= E −1 ((W◦ )◦ ) .

So we have (via the Remark) that E(W ) ⊆ (W◦ )◦ . Now

dim W + dim W◦ = dim V

dim W◦ + dim (W◦ )◦ = dim V ∗

but as dim V = dim V ∗ we have dim W = dim (W◦ )◦ . As E is an isomorphism, it

follows then that E(W ) = (W◦ )◦ .

Remark 20. One normally writes W◦◦ instead of (W◦ )◦ .

If the vector space V does not have finite dimension, the first computation still shows
that E(W ) ⊆ W◦◦ .

09/26/19
Math 4330 Fall 2013 10 Dual Spaces

Transpose
Let V and W be vector spaces over the field F and let T : V −→ W be a linear
transformation. We define T t : W ∗ −→ V ∗ by T t (f ) = f ◦ T for f ∈ W ∗ , that is
T t is just “compose with T ”. Further note that T t is itself a linear transformation:
T t (f1 + f2 ) = T t (f1 ) + T t (f2 )
T t (αf ) = αT t (f )
for all f, f1 , f2 ∈ W ∗ and all α ∈ F . One must really check such things. By
definition, T t (f1 + f2 ) = (f1 + f2 ) ◦ T and we need to prove a left distributive law.
But more generally one has both distributive laws
(S1 + S1 ) ◦ T = S1 ◦ T + S2 ◦ T
S ◦ (T1 + T2 ) = S ◦ T1 + S ◦ T2
for any linear transformations S, S1 , S2 , T, T1 , T2 for which these compositions are de-
fined. One easily verifies this by computing the value of each expression on a vector v
in the domain, using the definition of composition, addition of functions, and for the
second equation, the fact that S is a linear transformation.
To verify the second equation for the transpose one can verify similarly, and more
easily, the more general version
(αS) ◦ T = α(S ◦ T )
S ◦ (αT ) = α(S ◦ T )
which holds in general for the first equation, and requires that S be a linear trans-
formation for the second. The linear transformation T t is called the transpose of
T.
Thus we have shown that T t ∈ HomF (W ∗ , V ∗ ) . This now gives a map called the
transpose and denoted τ :
τ : HomF (V, W ) −→ HomF (W ∗ , V ∗ )
T 7→ T t
We next note that τ is also a linear transformation, that is, it satisfies
(T1 + T2 )t = T1t + T2t
(αT )t = α(T t ) .
But for f ∈ W ∗ the first equation is equivalent to
(T1 + T2 )t (f ) = f ◦ (T1 + T2 )
= f ◦ T1 + f ◦ T2
= T1t (f ) + T2t (f )
as observed earlier. Since this holds for all f ∈ W ∗ it follows that the two functions
are the same, yielding the first equation for τ . Proof of the second is similar via our
second earlier observation.

09/26/19
Math 4330 Fall 2013 11 Dual Spaces

Theorem 21. Let V and W be vector spaces over the field F . Then

1. The map transpose τ : HomF (V, W ) −→ HomF (W ∗ , V ∗ ) is always one-to-one.

2. If V and W are finite dimensional, then τ is an isomorphism.

Proof. Statement 1 implies statement 2 since the dimensions of the two vector spaces
are equal by our earlier results as dim V = dim V ∗ and dim W = dim W ∗ :

dim HomF (V, W ) = dim V · dim W

dim HomF (W ∗ , V ∗ ) = dim W ∗ · dim V ∗ .

So we’ll now verify the first part, which means we need to compute ker τ . For T ∈
HomF (V, W ) , T ∈ ker τ means that T t (f ) = f ◦ T = 0 for every f ∈ W ∗ . That is,
f (T (v)) = 0 for all v ∈ V or that f is 0 on im T .
But by Proposition 15 this can only happen for all f ∈ W ∗ if im T = 0 . That is,
T = 0 and thus τ is one-to-one.
Remark 22. It is also interesting to see how ker T t , im T t , (ker T)◦ (im T)◦ relate
to one another. See Exercise 18 for details.

We now give an explicit version of the transpose in the case of finite dimensional vector
spaces. Let V be a vector space with ordered basis A = { v1 . . . , vn } and let W be
a vector space with ordered basis B = { w1 , . . . , wm } . For
Pma linear transformation
T : V −→ W there exist scalars aij ∈ F so that T (vj ) = i=1 aij wi . Then we have
 
a11 a12 · · · a1n
 21 a22 · · · a2n 
 a 
[ T ]A, B =  .. .. .. ..  ∈ F m×n .
 . . . . 
am1 am2 · · · amn

Now W ∗ has dual basis A∗ and V ∗ has dual basis B ∗ . We will compute the matrix
of the transpose T t with respect to Pthese dual bases. That is, we need to determine
n
scalars b`,k ∈ F so that T (wk ) = `=1 b`,k v`∗ . Thus we compute
t ∗

T t (wk∗ ) (v` ) = wk∗ ◦ T (v` )

Xm
∗


= wk ai` wi
i=1
= 0 + ak` · 1

That is, we now have computed the coefficient of v`∗ in T (wk∗ ) as ak` . That is,
n
X
T t
(wk∗ ) = ak` v`∗
`=1

09/26/19
Math 4330 Fall 2013 12 Dual Spaces

and hence
 
a11 a21 ··· am1
 a12 a22 ··· am2 
[ T t ]B∗ , A∗ =   ∈ F n×m ;
 
.. .. .. ..
 . . . . 
a1n a2m · · · amn

that is, we have

[ T t ]B∗ , A∗ = [ T ]tA, B (2)

where the transpose on the right hand side (which we will denote by τ ), is the transpose
of matrices. In fact, this equation is the reason matrix transpose is defined as it is.
In summary, we have a commutative diagram with all horizontal and vertical arrows
isomorphisms:

τ -
HomF (V, W ) HomF (W ∗ , V ∗ )

[ ]A, B [ ]B∗ , A∗
? τ ?
F m×n - F n×m

Let T : V −→ W be a linear transformation. Then the rank of T is defined by

rank T = dim im T . In case M ∈ F m×n , then LM ; F n×1 −→ F m×1 is the linear
n×1
transformation given by LM (C)
= M C . Now im LM = { M C | C ∈ F } and if
we write M = C1 , C2 , . . . , Cn for Ci the columns of M we have
 
x1
 x 
 2 
M  ..  = x1 C1 + x2 C2 + · · · + xn Cn .
 . 
xn

Thus im LM ⊆ F m×1 is the column space of the matrix M , and rank Lm is the
dimension of the column space of M , that is the column rank of M . If we use the
standard bases A for F n×1 and B for F m×1 , then

[ LM ]A, B = M .

Applying the result from above, we have

[ LtM ]B∗ , A∗ = M t .

But transpose is an isomorphism so rank LM = rank LtM , so the column rank of M

equals the row rank of M .

09/26/19
Math 4330 Fall 2013 13 Dual Spaces

Double Transpose

Let T : V −→ W be a linear transformation. Upon taking the transpose we obtain

a linear transformation T t : W ∗ −→ V ∗ and upon taking the transpose again obtain
a linear transformation (T t )t : V ∗∗ −→ W ∗∗ . We write T tt instead of (T t )t . In the
diagram below, each evaluation map E should really be labelled to denote the vector
space on which it is defined.
Lemma 23. The diagram
T
V - W

E E
? T tt ?
V ∗∗ - W ∗∗ .
commutes, i.e., E ◦ T = T tt ◦ E.

Proof. The proof is an exercise in using the definitions (multiple times) and is left to
the reader. See Exercise 28.
Proposition 24. The map τ ◦ τ : HomF (V, W ) −→ HomF (V ∗∗ , W ∗∗ ) is always
one-to-one. It is an isomorphism if both V and W have finite dimension.

This Proposition follows immediately from the corresponding result for transpose.
If V and W are finite dimensional vector spaces the maps E are isomorphisms.
These isomorphisms allows us to consider T tt (more precisely E −1 ◦ T tt ◦ E ) as
a linear transformation from V to W . The lemma above asserts that after this
identification T tt = T .

Lattices

We will give a very brief description of what is called the lattice of subspaces of a vector
space. We begin with the notion of a partially ordered set.
Definition 25. Let X be a non-empty set. A subset O ⊆ X × X is called a partial
order on X if it satisfies the following three properties:

1. Reflexive:
For all x ∈ X , (x, x) ∈ O .

2. Anti-Symmetric:
If (x, y) ∈ O and (y, x) ∈ O , then x = y .

3. Transitive:
If (x, y), (y, z) ∈ O , then (x, z) ∈ O .

09/26/19
Math 4330 Fall 2013 14 Dual Spaces

One typically denotes the relation by some symbol, such as 4 , so that x 4 y means
the same thing as (x, y) ∈ O . The three conditions above then mean that 4 has the
same sort of properties as ≤ for real numbers or ⊆ for sets. We call a set X with a
partial order 4 a partially ordered set.

Definition 26. An element u of a partially ordered set X is an upper bound of a

subset Y if y 4 u for all y ∈ Y . An upper bound u of Y is a least upper bound if
u 4 v for every upper bound v of Y . If a least upper bound exists, then it is unique
by the anti-symmetry property. It is denoted by sup Y .
One defines lower bound, greatest lower bound and inf Y in a similar fashion.

Definition 27. A partially ordered set L is called a lattice if any two elements of L
have a least upper bound and a greatest lower bound. A lattice is called complete if
every subset has a least upper bound and a greatest lower bound.

If L is a lattice and a, b ∈ L , then one normally denotes sup { a, b } by a ∨ b (“join”)

and inf { a, b } by a ∧ b (“meet”).

Proposition 28. Let L be a lattice. The following hold:

L1. a ∨ b = b ∨ a , a ∧ b = b ∧ a .

L2. (a ∨ b) ∨ c = a ∨(b ∨ c) , (a ∧ b) ∧ c = a ∧(b ∧ c) .

L3. a ∨ a = a , a ∧ a = a .

L4. (a ∨ b) ∧ a = a , (a ∧ b) ∨ a = a .

Conversely, if L is a set with two binary operations ∨ and ∧ which satisfy L1, L2,
L3, and L4, then there is a definition of a partial order 4 on L such that a ∨ b ,
a ∧ b , are the sup and inf of a and b in this lattice.

As this proposition is not an integral part of our discussion, we will not prove it. One
can either work out the proof oneself or consult one of the many textbooks on abstract
algebra or lattice theory. Note that the axioms remain unchanged if we switch ∧
and ∨ . Thus the two binary operations satisfy the principle of duality, that is, for
any statement S which can be deduced from the axioms, then the dual statment S◦
(where ∧ and ∨ are switched) can be deduced as well.
Given a partially ordered set X one can reverse the ordering (i.e., use O◦ = { (y, x) | (x, y) ∈ O } )
to get a new partially ordered set, the dual of X denoted X◦ . Similarly, if L is a
lattice, we obtain the dual lattice, L◦ .
An isomorphism of lattices is a function ` : L1 −→ L2 from the lattice L1 to the lattice
L2 which is one-to-one, onto and preserves the lattice structure `(a ∨ b) = `(a) ∨ `(b)
and `(a ∧ b) = `(a) ∧ `(b) .
Although there are many examples of lattices, we will only consider one:

09/26/19
Math 4330 Fall 2013 15 Dual Spaces

Example 29 (The Lattice of Subspaces of a Vector Space). Let V be a vector space

over the field F . We denote by Lat(V ) the set of all subspaces of V . First note
that Lat(V ) is a partially ordered set under ⊆ . It is easy to see that


sup { Wi | i ∈ I } = Span { Wi | i ∈ I }
X
= Wi
i∈I

and
\
inf { Wi | i ∈ I } = Wi
i∈I

for any collection of subspaces { Wi | i ∈ I } of V . In particular we have for subspaces

W1 , W2 ⊆ V

W1 ∨ W2 = W1 + W2
W1 ∧ W2 = W1 ∩ W2 .

Note that Lat(V ) is a complete lattice.

Theorem 30. Let V be a finite dimensional vector space over the field F . There is
a natural duality between the vector spaces V and V ∗ :

1. Let B(V ) denote the set of bases of V . Then B 7→ B ∗ gives a bijection:

∗
: B(V ) −→ B(V ∗ ) .

2. Let Lat(V ) denote the lattice of subspaces of V . Then W 7→ W◦ gives a duality

◦
: Lat(V ) −→ Lat(V ∗ ) . That is, the inverse ◦ : Lat(V ∗ ) −→ Lat(V ) gives a
lattice isomorphism between Lat(V ∗ ) and Lat(V)◦

Proof. The first statement was proved earlier (see Theorem 8). In Proposition 19 it
was shown that W◦◦ = W , that is, ◦ is its own inverse. So the map ◦ is one-to-one
and onto.
To prove ◦ gives a lattice isomorphism with the dual (preserving ∨ and ∧ ) is the
same as saying that ◦ switches ∨ and ∧ going from Lat(V ) to Lat(V ∗ ) . This is
just

(W1 + W2 )◦ = W1◦ ∩ W2◦

(W1 ∩ W2 )◦ = W1◦ + W2◦

which is left as Exercise 20 for the reader.

◦
Remark 31. Calling “its own inverse” is a bit sloppy as there are really two different
maps.
◦
Note in particular that reverses the partial ordering, that is, it switches ⊆ and ⊇ .

09/26/19
Math 4330 Fall 2013 16 Dual Spaces

Geometry

Let V be a vector space over a field F and let f ∈ V ∗ be a non-zero linear functional.
The subspace ker f ⊆ V (which is not equal to V ) is called a hyperplane. Alterna-
tively it is referred to as having codimension 1 . For a subspace W ⊆ V , we define
codim W = dim V /W , so in the case of ker f , codim ker f = dim V / ker f = dim F =
1 . An alternate way to think of codimension 1 is that ker f has a one-dimensional
complementary subspace as is shown in the next lemma. Yet another way to think of
a hyperplane is that it is a maximal subspace of V , that is, it is not equal to V and
there are no subspace of V which properly contain it: If W ⊆ V is maximal and
W ⊆ U ⊆ V then U = W or U = V are the only possibilities.
Lemma 32. Let V be a vector space over the field F . Let f ∈ V ∗ be a non-zero
linear functional. Then for v0 ∈ V such that f (v0 ) 6= 0 , V = ker f ⊕ F v0 .

Proof. If u ∈ ker f ∩ F v0 , then u = av0 for some a ∈ F . As f (u) = af (v0 ) , it must

be the case that a = 0 , that is, u = 0 .
Let f (v0 ) = c =6 0 . If v ∈ V and f (v) = b , then v1 = v − bc−1 v0 and f (v1 ) =
b − bc−1 c = 0 . Thus v = v1 + bc−1 v0 is in ker f + F v0 .

If one wishes to think of a hyperplane W as a maximal subspace or as a subspace

having codimension 1 , then V /W is a vector space of dimension 1 (so isomorphic to
F ) and p : V −→ V /W gives W as the kernel of a non-zero linear functional. Such
a linear functional is unique up to multiplication by a non-zero scalar.
Lemma 33. Let V be a vector space over F and let f, g ∈ V ∗ . Then g is a scalar
multiple of f if and only if ker f ⊆ ker g . In particlar, if ker f = ker g 6= V , Then
g = af for some non-zero a ∈ F .

Proof. Clearly if g = af , we have ker f ⊆ ker g and if f = 0 then g = 0 as well. So

assume f 6= 0 and ker f ⊆ ker g . If g = 0 , then g = 0 · f . Assume g 6= 0 as well.
Then for g(v0 ) 6= 0 we also have f (v0 ) 6= 0 . The linear functional g(v0 )f −f (v0 )g will
be 0 on ker f ⊆ ker g as well as v0 and hence is 0 on all of V . Thus a = g(v0 )/f (v0 )
is the required scalar.
Proposition 34. Let V be a vector space over F and let f1 , . . . , fk , g ∈ V ∗ . Then
g is a linear combination of f1 , . . . , fk if and only if ki=1 ker fi ⊆ ker g .
T

Proof. Clearly if g is a linear combination of the fi , then g(v) = 0 for all v ∈ V

with fi (v) = 0 for every i .
We prove the result by induction on k with k = 1 being the preceeding lemma. Write
W for the intersection of the ker fi and U = ker fk and let g, f 1 , . . . , f k−1 denote the
restrictions of g, f1 , . . . , fk−1 to U . They are linear functionals on U . For u ∈ U
such that fi (u) = 0 for i = 1, . . . , k − 1 , then u ∈ W ⊆ ker g . Hence g(u) = 0 .
By the induction hypothesis, g is a linear combination of the f i , 1 ≤ i ≤ k − 1 :
g = a1 f 1 + · · · + ak−1 f k−1 .

09/26/19
Math 4330 Fall 2013 17 Dual Spaces

Now consider g − a1 f1 − · · · − ak−1 fk−1 . It is 0 on U and by the preceeding lemma

is a scalar multiple of fk , say ak fk , Thus g = a1 f1 + · · · + ak fk .

Proposition 35. Let V be a vector space over F of dimension n and let W ⊆ V

be a subspace of dimension m . Then W is the intersection of n − m hyperplanes.

Proof. We follow the same idea as that used in the proof of Theorem 13. Let
{ e1 , . . . , em } be a basis for W and enlarge to a basis for V : e1 , . . . , em , em1 , . . . , en .
◦ ∗ ∗ ◦

Then W = SpanF em+1 , . . . , en and dim W = n − m . By Proposition 19

W = W◦◦
= {v ∈ V | f (v) = 0∀f ∈ W◦ }
n  o
 ∗
= v∈V  ej (v) = 0, j = m + 1, . . . , n
n
\
= ker ej ∗ .
j=m+1

The basic idea geometrically is that for suitable hyperplanes (in “general position”)
each time one intersects with an additional hyperplane, the dimension of the resulting
subspace goes down by exactly 1 .

09/26/19
Math 4330 Fall 2013 18 Dual Spaces

Exercises
DualSpace
T 1. If we have a pair B ⊆ V and B 0 ⊆ V ∗ which are dual, then show that
U = f ∈B0 ker f is a complement to W = Span(B) in V , i.e., V is the internal direct
sum of U and W . Is this always true, or does it require that dim V be finite?
DualSpace 2. Let n be a positive integer.
a. Let Tr : F n×n −→ F denote the trace function on n × n matrices over the field
F . Show Tr(AB) = Tr(BA) for any A, B ∈ F n×n . Show that Tr is never the
zero linear functional.
b. Show that there do not exist A, B ∈ F n×n for F the field of complex numbers,
such that AB − BA = I . What happens for an arbitrary field?
c. Let T : V −→ V be a linear transformation where V is a finite dimensional
vector space over a field F . Choose a basis B for V and define Tr(T ) =
Tr([ T ]B, B ) . Show that the definition does not depend on the choice of basis B .
d. Let f : F n×n −→ F be a linear functional which satisfies f (AB) = f (BA) for
all A, B ∈ F n×n . Prove that f = a · Tr for some scalar a ∈ F . Further
show that if the characteristic of F is 0 , then f = Tr precisely when f (I) = n .
What happens if the characteristic is not 0 ? Can you find another way to decide
if f = Tr by computing a single value?
e. Let S = SpanF ({ AB − BA | A, B ∈ F n×n }) . Prove that S = ker Tr .
Hint: Compute the dimension of ker Tr , and find a basis for S using some
well-known matrices.
f. Let A be a fixed n×n matrix with entries in F . Define adA : F n×n −→ F n×n
by adA (B) = AB − BA . Recall that Tr is a linear functional on F n×n , and so
we may apply adA t to it. What is adA t (Tr) ?
g. Let A ∈ R n×n . Show that A = 0 if and only if At · A = 0 .
DualSpace 3. Let V be a finite-dimensional vector space over the field F and let
W be a subspace of V . If f is a linear functional on W , prove that there is a linear
functional h on V such that h(w) = f (w) for all w ∈ W .
DualSpace 4. Let F be a field with char F = 0 . Let A ∈ F n×n . Assume that
A is not similar to any matrix with 0 as its (1, 1) entry. Prove that A must be
a non-zero scalar matrix. [Hint: Use the assumption to find one vector in V ∗ and
carefully extend to a basis.]
DualSpace 5. Let F be a field with char F = 0 . Let A ∈ F n×n . Assume that
Tr(A) = 0 . Prove that there exist X, Y ∈ F n×n with A = XY − Y X . [Hint: You
may as well assume A is not 0 because . . . . Also ok if n = 1 because . . . . Induct
on n . Now A is not a scalar matrix because . . . and thus the matrix ? has 0 as its
(1, 1) entry. Apply the induction hypothesis to the lower right-hand corner A0 of ?
Further we can assume that the matrix X 0 has an inverse by replacing it by cI + X 0
for some c ∈ F (why?). Then . . . ]

09/26/19
Math 4330 Fall 2013 19 Dual Spaces

∗
DualSpace 6. Let F be a field. Let C ∈ F n×n . Define TC ∈ (F n×n ) by

TC (A) = Tr(AC)

and define
∗
T : F n×n −→ (F n×n )
by T (C) = TC .

a. Show that T is a linear transformation.

b. Prove that T is one-to-one.

c. Prove that T is an isomorphism.

∗
d. Given f ∈ (F n×n ) find an explicit formula for the unique matrix C ∈ F n×n for
which T (C) = f .
DualSpace 7. Let F be a field. Let A ∈ F n×n . Assume the characteristic polyno-
mial of A factors completely into linear factors in F [x] . Then Tr(A) is the sum of
characteristic values of A (with appropriate multiplicities).
DualSpace 8. Let F be a finite-dimensional vector space over the field F and let
T : V −→ V be a linear transformation. Let c be a scalar and suppose there is a
non-zero vector v in V such that T (v) = cv . Prove that there is a non-zero linear
functional f on V such that T t (f ) = cf .
DualSpace 9. Let V be the vector space of all polynomials over the real numbers
(considered as functions in the usual way). For real numbers a < b define the linear
functional γ on V by Z b
γ(f ) = f (t)dt .
a

Let D : V −→ V denote differentiation. What is Dt γ ?

DualSpace 10. Compute the change of basis matrix P = P (A, B) matrix in Exam-
ple 11.
DualSpace 11. Let
S T
0 −→ U −→ V −→ W −→ 0
be a short exact sequence of vector spaces over the field F . Show that
Tt St
0 −→ W ∗ −→ V ∗ −→ U ∗ −→ 0

is a short exact sequence as well.

DualSpace 12. Let F be a field and let s = ai for i = 1, . . . , ∞ be any sequence of
elements of F . Define fs : F [x] −→ F on the usual basis for formal polynomials by
fs (xi ) = ai . Show that fs ∈ F [x]∗ and every element of the dual is given by such a
sequence.

09/26/19
Math 4330 Fall 2013 20 Dual Spaces

DualSpace 13. Let V be a vector space over the field F and let W ⊆ V be a
subspace. Make no assumptions on the field F nor the dimension of V .

∗
a. Show that there is a natural isomorphism V /W −→ W◦ .

b. Show that there is a natural isomorphism V ∗ / W◦ −→ W ∗ .

c. Let V = W1 ⊕ W2 be an internal direct sum of vector spaces. Then Wi◦ ⊆ V ∗ .

Give natural isomorphisms Wi ∗ ≈ Wj◦ for i 6= j .

d. In the case V is finite dimentional, for each of the first two parts, derive an
alternate proof of Theorem 13

DualSpace 14. Verify the following statements for duals of direct sums.

a. Let V = V1 ⊕ · · · ⊕ Vn be an external direct sum of vector spaces. There is a

natural isomorphism V ∗ ≈ V1 ∗ ⊕ · · · ⊕ Vn ∗

b. Let V = ∞
L ∗
Q∞ ∗
i=1 Vi . Show that V ≈ i=1 Vi , and that dim V < dim V .

DualSpace 15. Let V be a vector space over the field F and let W ⊆ V be a
subspace. Make no assumptions on the field F nor the dimension of V . Define
Res : V ∗ −→ W ∗ to be the restriction map; for f ∈ V ∗ and w ∈ W , define
Res(f )(w) = f (w) . Show

a. Res(f ) ∈ W ∗ ,

b. Res is a linear transformation, and

c. Res is onto.

d. Compute ker Res .

DualSpace 16. Let F be a finite field (such as, for example, Fp ). Let 0 ≤ m ≤ n
be integers. Show that the number of subspaces of V = F n of dimension m is exactly
the same as the number of subspaces of dimension n − m .

DualSpace 17. Let W1 , W2 ⊆ V be subspaces of V over F . Make no assumptions

on F nor on the dimension of V .

1. If W1 6= W2 show that there is a linear functional f ∈ V ∗ such that f is 0 on

one of the Wi but not the other.

2. Show that in all cases W1 = W2 if and only if W1◦ = W2◦ .

DualSpace 18. Let V and W be vector spaces over the field F and let T : V −→ W
be a linear transformation.

a. ker T t = (im T)◦ .

09/26/19
Math 4330 Fall 2013 21 Dual Spaces

b. If V and W have finite dimension, then

i. rank T t = rank T .
ii. im T t = (ker T)◦ .

c. What happens in the previous part in the case of infinite dimensional vector
spaces?
DualSpace 19. Let F be a field.

a. Assume F is an infinite field and let V be a finite-dimensional vector space over

F . If v1 , . . . , vm are finitely many vectors in V , each non-zero, prove that there
exists a linear functional f on V such that f (vi ) 6= 0 for all i .

b. Describe precisely what happens if F is finite with q elements (i.e., relate n to

q ).
DualSpace 20. Let V be a finite dimensional vector space over the field F . Let
W1 and W2 be subspaces of V .

a. Show that (W1 + W2 )◦ = W1◦ ∩ W2◦ .

b. Show that (W1 ∩ W2 )◦ = W1◦ + W2◦ .

c. Determine what happens in each case if V has infinite dimension.

DualSpace 21. Let V be a non-zero vector space over the field F . Let f, g be two
linear functionals on V . Show that if |F | > 2 , then θ(v) = f (v) g(v) cannot be a
linear functional unless at least one of f or g is 0 . Is it possible that θ is a linear
functional in the case where F has two elements?
DualSpace 22. Let F be a field and X a non-empty set. Let V = F X denotes the
vector space of all functions from X to F . If W is any m -dimensional subspace of
V , show that there exist elements x1 , . . . , xm in X and functions f1 , . . . , fm in W
with fi (xj ) = δij .
DualSpace 23. Let n be a positive integer and let F be a field whose characteristic
is either 0 or greater than 2n .

a. Let P  n ⊆ F [x] be the vector space of polynomials of degree less than n . Let
B = E−1 , E−2 , . . . , E−n ∈ Pn ∗ be the evaluation maps for { −1, −2, . . . , −n } ∈

F . Give the basis A = { p1 , . . . , pn } ⊆ Pn which is dual to B .

 1 1 1

b. Let C = x+1 , x+2 , . . . , x+n ⊆ F (x) where F (x) is the field of fractions of
F [x] . Let Qn = SpanF (C) ⊆ F (x) . Give a natural isomorphism m : Qn −→
Pn and conclude that Qn has dimension n and that C is a basis. [Hint:
What does π(x) = (x + 1)(x + 2) · · · (x + n) = (x + n)(n) have to do with Pn ?
[ (x + n)(n) is notation of example in the section on the “Matrix of a Linear
Transformation”)].

09/26/19
Math 4330 Fall 2013 22 Dual Spaces

c. Define Ei for i ∈ { 1, 2, . . . , n } by
 
1 1
Ei = .
x+j i+j

Show that E = { E1 , E2 , . . . , En } is a basis for Qn ∗ and find the basis D =

{ f1 , f2 , . . . , fn } of Qn to which it is dual. If m(fi ) = qi ∈ Pn give a formula
for fi explicitly and determine qi . ??

d. Compute the change of basis matrix P = P (C, D) . This part should be easy.
Compute P −1 = P (D, C) . This is more interesting and should answer a hard
question you’ve seen before.

DualSpace 24. Prove Proposition 28.

DualSpace 25. Let U and V be arbitrary vector spaces over the field F . Explicitly
give a natural isomorphism (U ⊕ V )∗ −→ U ∗ ⊕ V ∗ and its inverse. Include complete
proofs.

DualSpace 26. Consider the linear transformation Der, X on the vector space F [x]
defined as follows:
Der(xn ) = nxn−1 X(xn ) = xn+1 .

a. Show that these linear transformations satisfy Der ◦ X − X ◦ Der = Id , i.e., for
any f ∈ F [x] we have Der(X(f )) − X(Der(f )) = f .

b. Construct analogs of the linear transformations Der and X , which act on the
vector space C ∞ (R) , which also satisfy the equation Der ◦ X − X ◦ Der = Id .

c. Construct analogs D̃ and X̃ of the linear transformations Der and X , which

act on the vector space C(R) and satisfy the equation D̃ ◦ X̃ − X̃ ◦ D̃ = Id .
Hint: You can not use any form of differentiation.

d. Does there exist a finite dimensional vector space V over a field F and two linear
transformations D and X from V to itself, such that D ◦ X − X ◦ D = Id .
Hint: What is char F ?

e. Does there exist a finite dimensional vector space V over a field F and two
invertible linear transformations from V to itself, such that D ◦ X −X ◦ D = Id .

DualSpace 27. A linear transformation D ∈ HomF (V, V ) is called nilpotent if there

exists some N such that DN = 0 . Here Dk denote the composition of D with itself
k times, also D0 = Id . The transformation is called locally nilpotent if for every
vector v ∈ V there exists nv such that Dnv (v) = 0 .

a. Show that if V is finite dimensional then any locally nilpotent transformation is

also nilpotent.

09/26/19
Math 4330 Fall 2013 23 Dual Spaces

b. Show that the differential operator Der : F [x] −→ F [x] from the previous exer-
cise defined by Der(xn ) = nxn−1 is locally nilpotent.

c. Show that Der : F [x] −→ F [x] (from part b.) is nilpotent if and only if
char F > 0 .

d. Let D be a locally nilpotent transformation. Let πD be the map from F [[t]]

to HomF (V, V ) defined as follows
 X  X
πD ( ai ti ) (v) = ai Di (v)

Verify that this map is well defined, i.e., all infinite sums make sense.

e. Show that πD is an algebra homomorphism, i.e.

πD (f + g) = πD (f ) + πD (g) πD (λf ) = λπD (f ) πD (f · g) = πD (f ) ◦ πD (g) .

f. Let CenD denote the set of all linear transformations T ∈ HomF (V, V ) which
commute with D , i.e.

CenD = { T ∈ HomF (V, V ) | T ◦ D=D ◦ T}

Show that CenD is a subspace of HomF (V, V ) .

g. Show that if D is locally nilpotent then im πD ⊂ CenD . Hint: This follows

immediately from part e.

h. Show im πDer ⊂ CenDer , i.e., any operator which commutes with Der can be
expressed as a formal power series. Hint: Look at the constant term of T (xn )
for T ∈ CenDer .

i. Construct an example of a locally nilpotent derivation, such that im πDer is a

proper subspace of CenDer .

j. Find a necessary and sufficient conditions for im πDer = CenDer . Hint: Look
at ker D .

DualSpace 28. . Give a proof for Lemma 23

DualSpace 29. Let V be a vector space over the field F . We demonstrate a method
to construct a set which is dual to a given linearly independent subset of V ∗ by using
Proposition 34. Assume that V has countable dimension (i.e., the basis can be put
into one-to-one correspondence with the positive integers). We prove that V ∗ has
uncountable dimension.
Here is an outline of the exercise: Assume that F is a countable basis for V ∗ .
Construct a linearly independent subset A of V which is dual to F , and show that
in fact it must be a basis for V . But this contradicts the fact (proved earlier) that
the dual ( F ) of an infinite basis ( A ) of V cannot span V ∗ .

09/26/19
Math 4330 Fall 2013 24 Dual Spaces

a. Let F = { f1 , f2 , . . . } be a sequence of linearly independent elements in V ∗ . Show

that there exists a non-zero v1 ∈ V with f1 (v1 ) = 1 .

b. Show that there exists an element v2 ∈ V with f2 (v2 ) = 1 and f1 (v2 ) = 0 .

c. Give an induction argument to show that for any positive k there exist vectors
v1 , . . . , vk ∈ V with fi (vi ) = 1 and fi (vj ) = 0 for j < i . Show that the set of vi
is linearly independent.

d. Let A be the infinite set of vi constructed inductively by this process. Assume

that F is a basis for V ∗ . Let A ⊆ B be a basis for V containing A . If v ∈ B
is not in A , then let v ∗ in V ∗ be the linear functional that is 1 on v and 0 on
the other elements of B . Show that v ∗ is not a finite linear combination of the
elements of F .

e. Conclude that A must be a basis for V . Let θ ∈ V ∗ be the linear functional that
is 1 on every element of A . Show that θ is not in the span of F . Conclude that
F is not a basis and hence that V ∗ does not have countable dimension.

f. For a fixed k show that by modifing the vi one can even construct a set u1 , . . . , uk
with fi (uj ) = δij .

DualSpace 30. Assume that F is a field whose characteristic is not 2 . Let V be the
vector space of all polynomials of degree less than or equal to three over F . Define
the linear functionals Li , i = 1, 2, 3, 4 by

L1 (f ) = f (0)
L2 (f ) = f (1)
L3 (f ) = (Df )(0)
L4 (f ) = (∆f )(1)

where D denotes the derivative and ∆ denotes the difference operator (defined by
∆(f (x)) = f (x + 1) − f (x) ).

a. Prove that these four linear functionals form a basis B ∗ for V ∗ and find the basis
B of V to which it is dual.

b. Express 1 , x , x2 , x3 in terms of B .

c. Let W be the subspace of V spanned by 1 and x . Find a basis for W◦ expressed

in terms of B ∗ .

d. Assuming that F is the field of real numbers, express the linear functional “the
definite integral of f between 0 and 1 ” in terms of B ∗ .

DualSpace 31. Let Fq be a finite field with q elements and having characteristic p .
Prove the assertions you make to answer the following questions.

09/26/19
Math 4330 Fall 2013 25 Dual Spaces

a. Let m, n be positive integers. Give a formula for the number of elements in Fnq
and Fqm×n .
b. Give a formula for the number of different ordered bases of Fnq .
c. Give a formula for the number of invertible matrices in Fqn×n .
d. Give a formula for the number of ordered, linearly independent sequences of vectors
with m elements in Fnq .
e. Give a formula for the number of matrices of Fqm×n with rank m .
f. For a non-negative integer k determine the number of subspaces of dimension k of
Fnq .
g. For a non-negative integer k determine the number of ordered sequences of k
vectors in Fnq that span.
DualSpace 32. Let U , V and W be vector spaces over the field F .

a. Let T : U −→ V and S : V −→ W be linear transformations. Give a formula

for (S ◦ T )t and verify that it is correct.
b. If the vector spaces all have finite dimension, choose bases A , B and C for U ,
V and W having m , n and p elements, respectively. Construct a commutative
diagram with HomF (V, W ) × HomF (U, V ) −→ F p×n × F n×m as its top row and
HomF (U, W ) −→ F p×m as its bottom row. The two horizontal arrows should be
isomorphisms and the two vertical arrows should be “composition” and “matrix
multiplication”, respectively. (See pages 2 and 6 of the section “The Matrix of a
Linear Transformation” for things very closely related to this.)
c. Continue with the notation of the preceeding part with finite bases A , B and C
for U , V and W . Construct the diagram of the previous part and a similar one
for the dual spaces. Construct three diagrams of the type given on page 11 of this
section. Use these three to “glue” together the two just constructed into a larger
diagram. Which faces of the diagram commute from results already proven? What
can you conclude about the remaining face(s)?
DualSpace 33. Let Pn be the vector space of all polynomials of degree less than n
over a field F of characteristic 0 . Consider the linear transformation Θ : Pn −→ Pn
defined by

Θ p(x) = p(x + 1) − 2p(x) + p(x − 1)
Construct a basis A of Pn such that the matrix [ Θ ]A of the linear transformation
Θ with respect to this basis has exactly n − 2 non-zero coefficients. Explain what
changes if the characteristic of the field F is finite.
DualSpace 34. a. Prove that the F -algebra HomF (V, V ) is simple (i.e., it does not
contain any non-trivial 2 -sided ideals) if V is finite dimensional vector space over
F.

09/26/19
Math 4330 Fall 2013 26 Dual Spaces

b. Construct a non-trivial 2 -sided ideal if dimF V is infinite.

c. Generalize to give a tower distinct of 2 -sided ideals in case dimF V > ℵ0 .

09/26/19
Math 4330 Fall 2013 27 Dual Spaces

The Notes for the course Math 4330, Honors Linear Algebra at Cornell University have
been developed over the last ten years or so mainly by the following (in chronological
order):
Gerhard O. Michler
R. Keith Dennis
Martin Kassabov
W. Frank Moore
and
Yuri Berest.
Most sections have been revised so many times the original author may no longer
recognize it. The intent is to provide a modern treatement of linear algebra using
consistent terminology and notation. Some sections are written simply to provide
a central source of information such as those on “Useful Definitions”, “Subobjects”,
and “Universal Mapping Properties” rather than as a chapter as one might find in a
traditional textbook. Additionally there are sections whose intent is to provide proofs
of some results which are not given in the lectures, but rather provide them as part
of a more thorough development of a tangential topic (e.g., Zorn’s Lemma to develop
cardinal numbers and the existence of bases and dimension in the general case).
A large number of challenging exercises from many different sources have been included.
Although most should be readily solvable by students who have mastered the material,
a few even more challenging ones still remain.
Much still remains to be done. Corrections and suggestions for additional exercises,
topics and supplements are always welcome.

09/26/19