Chapter 22

Tensor Algebras, Symmetric Algebras

and Exterior Algebras

22.1 Tensors Products

We begin by defining tensor products of vector spaces over a field and then we investigate
some basic properties of these tensors, in particular the existence of bases and duality. After
this, we investigate special kinds of tensors, namely, symmetric tensors and skew-symmetric
tensors. Tensor products of modules over a commutative ring with identity will be discussed
very briefly. They show up naturally when we consider the space of sections of a tensor
product of vector bundles.
Given a linear map, f : E → F , we know that if we have a basis, (ui )i∈I , for E, then f
is completely determined by its values, f (ui ), on the basis vectors. For a multilinear map,
f : E n → F , we don’t know if there is such a nice property but it would certainly be very
useful.
In many respects, tensor products allow us to define multilinear maps in terms of their
action on a suitable basis. The crucial idea is to linearize, that is, to create a new vector space,
E ⊗n , such that the multilinear map, f : E n → F , is turned into a linear map, f⊗ : E ⊗n → F ,
which is equivalent to f in a strong sense. If in addition, f is symmetric, then we can define
a symmetric tensor power, Symn (E), and every symmetric multilinear map, f : E n → F , is
turned into a linear map, f⊙ : Symn (E) → F , which is equivalent to f in a strong �n sense.
Similarly, if f is alternating, then we can define a skew-symmetric tensor � power, (E), and
every alternating multilinear map is turned into a linear map, f∧ : n (E) → F , which is
equivalent to f in a strong sense.
Tensor products can be defined in various ways, some more abstract than others. We
tried to stay down to earth, without excess!
Let K be a given field, and let E1 , . . . , En be n ≥ 2 given vector spaces. For any vector
space, F , recall that a map, f : E1 × · · · × En → F , is multilinear iff it is linear in each of

585
586 CHAPTER 22. TENSOR ALGEBRAS

its argument, that is,

f (u1 , . . . ui1 , v + w, ui+1 , . . . , un ) = f (u1 , . . . ui1 , v, ui+1 , . . . , un )

+ f (u1 , . . . ui1 , w, ui+1 , . . . , un )
f (u1 , . . . ui1 , λv, ui+1 , . . . , un ) = λf (u1 , . . . ui1 , v, ui+1 , . . . , un ),

for all uj ∈ Ej (j �= i), all v, w ∈ Ei and all λ ∈ K, for i = 1 . . . , n.

The set of multilinear maps as above forms a vector space denoted L(E1 , . . . , En ; F ) or
Hom(E1 , . . . , En ; F ). When n = 1, we have the vector space of linear maps, L(E, F ) or
Hom(E, F ). (To be very precise, we write HomK (E1 , . . . , En ; F ) and HomK (E, F ).) As
usual, the dual space, E ∗ , of E is defined by E ∗ = Hom(E, K).
Before proceeding any further, we recall a basic fact about pairings. We will use this fact
to deal with dual spaces of tensors.

Definition 22.1 Given two vector spaces, E and F , a map, (−, −) : E × F → K, is a

nondegenerate pairing iff it is bilinear and iff (u, v) = 0 for all v ∈ F implies u = 0 and
(u, v) = 0 for all u ∈ E implies v = 0. A nondegenerate pairing induces two linear maps,
ϕ : E → F ∗ and ψ : F → E ∗ , defined by

ϕ(u)(y) = (u, y)
ψ(v)(x) = (x, v),

for all u, x ∈ E and all v, y ∈ F .

Proposition 22.1 For every nondegenerate pairing, (−, −) : E × F → K, the induced maps
ϕ : E → F ∗ and ψ : F → E ∗ are linear and injective. Furthermore, if E and F are finite
dimensional, then ϕ : E → F ∗ and ψ : F → E ∗ are bijective.

Proof . The maps ϕ : E → F ∗ and ψ : F → E ∗ are linear because u, v �→ (u, v) is bilinear.

Assume that ϕ(u) = 0. This means that ϕ(u)(y) = (u, y) = 0 for all y ∈ F and as our
pairing is nondegenerate, we must have u = 0. Similarly, ψ is injective. If E and F are finite
dimensional, then dim(E) = dim(E ∗ ) and dim(F ) = dim(F ∗ ). However, the injectivity of ϕ
and ψ implies that that dim(E) ≤ dim(F ∗ ) and dim(F ) ≤ dim(E ∗ ). Consequently dim(E) ≤
dim(F ) and dim(F ) ≤ dim(E), so dim(E) = dim(F ). Therefore, dim(E) = dim(F ∗ ) and ϕ
is bijective (and similarly dim(F ) = dim(E ∗ ) and ψ is bijective).
Proposition 22.1 shows that when E and F are finite dimensional, a nondegenerate pairing
induces canonical isomorphims ϕ : E → F ∗ and ψ : F → E ∗ , that is, isomorphisms that do
not depend on the choice of bases. An important special case is the case where E = F and
we have an inner product (a symmetric, positive definite bilinear form) on E.

Remark: When we use the term “canonical isomorphism” we mean that such an isomor-
phism is defined independently of any choice of bases. For example, if E is a finite dimensional
22.1. TENSORS PRODUCTS 587

vector space and (e1 , . . . , en ) is any basis of E, we have the dual basis, (e∗1 , . . . , e∗n ), of E ∗
(where, e∗i (ej ) = δi j ) and thus, the map ei �→ e∗i is an isomorphism between E and E ∗ . This
isomorphism is not canonical.
On the other hand, if �−, −� is an inner product on E, then Proposition 22.1 shows that
the nondegenerate pairing, �−, −�, induces a canonical isomorphism between E and E ∗ .
This isomorphism is often denoted � : E → E ∗ and we usually write u� for �(u), with u ∈ E.
Given any basis,
� (e1 , . . . , en ), of E (not necessarily orthonormal), if we let gij = (ei , ej ), then
for every u = ni=1 ui ei , since u� (v) = �u, v�, for all v ∈ V , we get
n
� n
�
u� = ωi e∗i , with ωi = gij uj .
i=1 j=1

If we �
use the convention that coordinates of vectors are written using superscripts
n i
(u = � i=1 u ei ) and coordinates of one-forms (covectors) are written using subscripts
n ∗
(ω = i=1 ωi ei ), then the map, �, has the effect of lowering (flattening!)
�n indices. The
∗ ∗ ∗
inverse
�of � is denoted � : E → E. If we write ω ∈ E as ω = i=1 ωi ei and ω � ∈ E as
n
ω = j=1 (ω � )j ej , since
�

n
�
�
ωi = ω(ei ) = �ω , ei � = (ω � )j gij , 1 ≤ i ≤ n,
j=1

we get
n
�
� i
(ω ) = g ij ωj ,
j=1

where (g ij ) is the inverse of the matrix (gij ). The inner product, (−, −), on E induces an
inner product on E ∗ also denoted (−, −) and given by

(ω1 , ω2 ) = (ω1� , ω2� ),

for all ω1 , ω2 ∈ E ∗ . Then, it is obvious that

(u, v) = (u� , v � ), for all u, v ∈ E.

If (e1 , . . . , en ) is a basis of E and gij = (ei , ej ), as

n
�
(e∗i )� = g ik ek ,
k=1

an easy computation shows that

(e∗i , e∗j ) = ((e∗i )� , (e∗j )� ) = g ij ,

588 CHAPTER 22. TENSOR ALGEBRAS

that is, in the basis (e∗1 , . . . , e∗n ), the inner product on E ∗ is represented by the matrix (g ij ),
the inverse of the matrix (gij ).
The inner product on a finite vector space also yields a natural isomorphism between
the space, Hom(E, E; K), of bilinear forms on E and the space, Hom(E, E), of linear maps
from E to itself. Using this isomorphism, we can define the trace of a bilinear form in an
intrinsic manner. This technique is used in differential geometry, for example, to define the
divergence of a differential one-form.

Proposition 22.2 If �−, −� is an inner product on a finite vector space, E, (over a field,
K), then for every bilinear form, f : E × E → K, there is a unique linear map, f � : E → E,
such that
f (u, v) = �f � (u), v�, for all u, v ∈ E.
The map, f �→ f � , is a linear isomorphism between Hom(E, E; K) and Hom(E, E).

Proof . For every g ∈ Hom(E, E), the map given by

f (u, v) = �g(u), v�, u, v ∈ E,

is clearly bilinear. It is also clear that the above defines a linear map from Hom(E, E) to
Hom(E, E; K). This map is injective because if f (u, v) = 0 for all u, v ∈ E, as �−, −� is
an inner product, we get g(u) = 0 for all u ∈ E. Furthermore, both spaces Hom(E, E) and
Hom(E, E; K) have the same dimension, so our linear map is an isomorphism.
If (e1 , . . . , en ) is an orthonormal basis of E, then we check immediately that the trace of
a linear map, g, (which is independent of the choice of a basis) is given by
n
�
tr(g) = �g(ei ), ei �,
i=1

where n = dim(E). We define the trace of the bilinear form, f , by

tr(f ) = tr(f � ).

From Proposition 22.2, tr(f ) is given by

n
�
tr(f ) = f (ei , ei ),
i=1

for any orthonormal basis, (e1 , . . . , en ), of E. We can also check directly that the above
expression is independent of the choice of an orthonormal basis.
We will also need the following Proposition to show that various families are linearly
independent.
22.1. TENSORS PRODUCTS 589

Proposition 22.3 Let E and F be two nontrivial vector spaces and let (ui )i∈I be any family
of vectors ui ∈ E. The family, (ui )i∈I , is linearly independent iff for every family, (vi )i∈I , of
vectors vi ∈ F , there is some linear map, f : E → F , so that f (ui ) = vi , for all i ∈ I.

Proof . Left as an exercise.

First, we define tensor products, and then we prove their existence and uniqueness up to
isomorphism.

Definition 22.2 A tensor product of n ≥ 2 vector spaces E1 , . . . , En , is a vector space T ,

together with a multilinear map ϕ : E1 × · · · × En → T , such that, for every vector space F
and for every multilinear map f : E1 ×· · ·×En → F , there is a unique linear map f⊗ : T → F ,
with
f (u1 , . . . , un ) = f⊗ (ϕ(u1 , . . . , un )),
for all u1 ∈ E1 , . . . , un ∈ En , or for short

f = f⊗ ◦ ϕ.

Equivalently, there is a unique linear map f⊗ such that the following diagram commutes:

ϕ
E1 × · · · × En � T
f⊗
f
� �
F

First, we show that any two tensor products (T1 , ϕ1 ) and (T2 , ϕ2 ) for E1 , . . . , En , are
isomorphic.

Proposition 22.4 Given any two tensor products (T1 , ϕ1 ) and (T2 , ϕ2 ) for E1 , . . . , En , there
is an isomorphism h : T1 → T2 such that

ϕ2 = h ◦ ϕ1 .

Proof . Focusing on (T1 , ϕ1 ), we have a multilinear map ϕ2 : E1 × · · · × En → T2 , and thus,

there is a unique linear map (ϕ2 )⊗ : T1 → T2 , with

ϕ2 = (ϕ2 )⊗ ◦ ϕ1 .

Similarly, focusing now on on (T2 , ϕ2 ), we have a multilinear map ϕ1 : E1 × · · · × En → T1 ,

and thus, there is a unique linear map (ϕ1 )⊗ : T2 → T1 , with

ϕ1 = (ϕ1 )⊗ ◦ ϕ2 .

But then, we get

ϕ1 = (ϕ1 )⊗ ◦ (ϕ2 )⊗ ◦ ϕ1 ,
590 CHAPTER 22. TENSOR ALGEBRAS

and
ϕ2 = (ϕ2 )⊗ ◦ (ϕ1 )⊗ ◦ ϕ2 .
On the other hand, focusing on (T1 , ϕ1 ), we have a multilinear map ϕ1 : E1 × · · · × En → T1 ,
but the unique linear map h : T1 → T1 , with

ϕ1 = h ◦ ϕ1

is h = id, and since (ϕ1 )⊗ ◦ (ϕ2 )⊗ is linear, as a composition of linear maps, we must have

(ϕ1 )⊗ ◦ (ϕ2 )⊗ = id.

Similarly, we must have

(ϕ2 )⊗ ◦ (ϕ1 )⊗ = id.
This shows that (ϕ1 )⊗ and (ϕ2 )⊗ are inverse linear maps, and thus, (ϕ2 )⊗ : T1 → T2 is an
isomorphism between T1 and T2 .
Now that we have shown that tensor products are unique up to isomorphism, we give a
construction that produces one.

Theorem 22.5 Given n ≥ 2 vector spaces E1 , . . . , En , a tensor product (E1 ⊗ · · · ⊗ En , ϕ)

for E1 , . . . , En can be constructed. Furthermore, denoting ϕ(u1 , . . . , un ) as u1 ⊗ · · · ⊗ un ,
the tensor product E1 ⊗ · · · ⊗ En is generated by the vectors u1 ⊗ · · · ⊗ un , where u1 ∈
E1 , . . . , un ∈ En , and for every multilinear map f : E1 × · · · × En → F , the unique linear
map f⊗ : E1 ⊗ · · · ⊗ En → F such that f = f⊗ ◦ ϕ, is defined by

f⊗ (u1 ⊗ · · · ⊗ un ) = f (u1 , . . . , un ),

on the generators u1 ⊗ · · · ⊗ un of E1 ⊗ · · · ⊗ En .

Proof . Given any set, I, viewed as an index set, let K (I) be the set of all functions, f : I → K,
such that f (i) �= 0 only for finitely many i ∈ I. As usual, denote such a function by (fi )i∈I ,
it is a family of finite support. We make K (I) into a vector space by defining addition and
scalar multiplication by

(fi ) + (gi ) = (fi + gi )

λ(fi ) = (λfi ).

The family, (ei )i∈I , is defined such that (ei )j = 0 if j �= i and (ei )i = 1. It is a basis of
the vector space K (I) , so that every w ∈ K (I) can be uniquely written as a finite linear
combination of the ei . There is also an injection, ι : I → K (I) , such that ι(i) = ei for every
i ∈ I. Furthermore, it is easy to show that for any vector space, F , and for any function,
f : I → F , there is a unique linear map, f : K (I) → F , such that

f = f ◦ ι,
22.1. TENSORS PRODUCTS 591

as in the following diagram:

ι
I � K (I)
f
f
� �
F
This shows that K (I) is the free vector space generated by I. Now, apply this construction
to the cartesian product, I = E1 × · · · × En , obtaining the free vector space M = K (I) on
I = E1 ×· · ·×En . Since every, ei , is uniquely associated with some n-tuple i = (u1 , . . . , un ) ∈
E1 × · · · × En , we will denote ei by (u1 , . . . , un ).
Next, let N be the subspace of M generated by the vectors of the following type:

(u1 , . . . , ui + vi , . . . , un ) − (u1 , . . . , ui , . . . , un ) − (u1 , . . . , vi , . . . , un ),

(u1 , . . . , λui , . . . , un ) − λ(u1 , . . . , ui , . . . , un ).

We let E1 ⊗· · ·⊗En be the quotient M/N of the free vector space M by N , π : M → M/N
be the quotient map and set
ϕ = π ◦ ι.
By construction, ϕ is multilinear, and since π is surjective and the ι(i) = ei generate M ,
since i is of the form i = (u1 , . . . , un ) ∈ E1 ×· · ·×En , the ϕ(u1 , . . . , un ) generate M/N . Thus,
if we denote ϕ(u1 , . . . , un ) as u1 ⊗ · · · ⊗ un , the tensor product E1 ⊗ · · · ⊗ En is generated by
the vectors u1 ⊗ · · · ⊗ un , where u1 ∈ E1 , . . . , un ∈ En .
For every multilinear map f : E1 × · · · × En → F , if a linear map f⊗ : E1 ⊗ · · · ⊗ En → F
exists such that f = f⊗ ◦ ϕ, since the vectors u1 ⊗ · · · ⊗ un generate E1 ⊗ · · · ⊗ En , the map
f⊗ is uniquely defined by
f⊗ (u1 ⊗ · · · ⊗ un ) = f (u1 , . . . , un ).
On the other hand, because M = K (E1 ×···×En ) is free on I = E1 × · · · × En , there is a unique
linear map f : K (E1 ×···×En ) → F , such that
f = f ◦ ι,
as in the diagram below:
ι
E1 × · · · × En � K (E1 ×···×En )
f
f
� �
F
Because f is multilinear, note that we must have f (w) = 0, for every w ∈ N . But then,
f : M → F induces a linear map h : M/N → F , such that
f = h ◦ π ◦ ι,
592 CHAPTER 22. TENSOR ALGEBRAS

by defining h([z]) = f (z), for every z ∈ M , where [z] denotes the equivalence class in M/N
of z ∈ M :
π◦ι �
E1 × · · · × En K (E1 ×···×En ) /N
h
f
� �
F
Indeed, the fact that f vanishes on N insures that h is well defined on M/N , and it is clearly
linear by definition. However, we showed that such a linear map h is unique, and thus it
agrees with the linear map f⊗ defined by

f⊗ (u1 ⊗ · · · ⊗ un ) = f (u1 , . . . , un )

on the generators of E1 ⊗ · · · ⊗ En .
What is important about Theorem 22.5 is not so much the construction itself but the
fact that it produces a tensor product with the universal mapping property with respect to
multilinear maps. Indeed, Theorem 22.5 yields a canonical isomorphism,

L(E1 ⊗ · · · ⊗ En , F ) ∼
= L(E1 , . . . , En ; F ),

between the vector space of linear maps, L(E1 ⊗ · · · ⊗ En , F ), and the vector space of
multilinear maps, L(E1 , . . . , En ; F ), via the linear map − ◦ ϕ defined by

h �→ h ◦ ϕ,

where h ∈ L(E1 ⊗ · · · ⊗ En , F ). Indeed, h ◦ ϕ is clearly multilinear, and since by Theorem

22.5, for every multilinear map, f ∈ L(E1 , . . . , En ; F ), there is a unique linear map f⊗ ∈
L(E1 ⊗ · · · ⊗ En , F ) such that f = f⊗ ◦ ϕ, the map − ◦ ϕ is bijective. As a matter of fact,
its inverse is the map
f �→ f⊗ .
Using the “Hom” notation, the above canonical isomorphism is written

Hom(E1 ⊗ · · · ⊗ En , F ) ∼
= Hom(E1 , . . . , En ; F ).

Remarks:

(1) To be very precise, since the tensor product depends on the field, K, we should subscript
the symbol ⊗ with K and write

E1 ⊗K · · · ⊗ K En .

However, we often omit the subscript K unless confusion may arise.

22.2. BASES OF TENSOR PRODUCTS 593

(2) For F = K, the base field, we obtain a canonical isomorphism between the vector
space L(E1 ⊗ · · · ⊗ En , K), and the vector space of multilinear forms L(E1 , . . . , En ; K).
However, L(E1 ⊗ · · · ⊗ En , K) is the dual space, (E1 ⊗ · · · ⊗ En )∗ , and thus, the vector
space of multilinear forms L(E1 , . . . , En ; K) is canonically isomorphic to (E1 ⊗· · ·⊗En )∗ .
We write
L(E1 , . . . , En ; K) ∼
= (E1 ⊗ · · · ⊗ En )∗ .

The fact that the map ϕ : E1 × · · · × En → E1 ⊗ · · · ⊗ En is multilinear, can also be

expressed as follows:

u1 ⊗ · · · ⊗ (ui + vi ) ⊗ · · · ⊗ un = (u1 ⊗ · · · ⊗ ui ⊗ · · · ⊗ un )
+ (u1 ⊗ · · · ⊗ vi ⊗ · · · ⊗ un ),
u1 ⊗ · · · ⊗ (λui ) ⊗ · · · ⊗ un = λ(u1 ⊗ · · · ⊗ ui ⊗ · · · ⊗ un ).

Of course, this is just what we wanted! Tensors in E1 ⊗ · · · ⊗ En are also called n-tensors,
and tensors of the form u1 ⊗ · · · ⊗ un , where ui ∈ Ei , are called simple (or indecomposable)
n-tensors. Those n-tensors that are not simple are often called compound n-tensors.
Not only do tensor products act on spaces, but they also act on linear maps (they are
functors). Given two linear maps f : E → E � and g : F → F � , we can define h : E × F →
E � ⊗ F � by
h(u, v) = f (u) ⊗ g(v).
It is immediately verified that h is bilinear, and thus, it induces a unique linear map

f ⊗ g : E ⊗ F → E � ⊗ F �,

such that
(f ⊗ g)(u ⊗ v) = f (u) ⊗ g(u).

If we also have linear maps f � : E � → E �� and g � : F � → F �� , we can easily verify that

the linear maps (f � ◦ f ) ⊗ (g � ◦ g) and (f � ⊗ g � ) ◦ (f ⊗ g) agree on all vectors of the form
u ⊗ v ∈ E ⊗ F . Since these vectors generate E ⊗ F , we conclude that

(f � ◦ f ) ⊗ (g � ◦ g) = (f � ⊗ g � ) ◦ (f ⊗ g).

The generalization to the tensor product f1 ⊗ · · · ⊗ fn of n ≥ 3 linear maps fi : Ei → Fi

is immediate, and left to the reader.

22.2 Bases of Tensor Products

We showed that E1 ⊗ · · · ⊗ En is generated by the vectors of the form u1 ⊗ · · · ⊗ un . However,
there vectors are not linearly independent. This situation can be fixed when considering
bases, which is the object of the next proposition.
594 CHAPTER 22. TENSOR ALGEBRAS

Proposition 22.6 Given n ≥ 2 vector spaces E1 , . . . , En , if (uki )i∈Ik is a basis for Ek ,

1 ≤ k ≤ n, then the family of vectors

(u1i1 ⊗ · · · ⊗ unin )(i1 ,...,in )∈I1 ×...×In

is a basis of the tensor product E1 ⊗ · · · ⊗ En .

Proof . For each k, 1 ≤ k ≤ n, every v k ∈ Ek can be written uniquely as

�
vk = vjk ukj ,
j∈Ik

for some family of scalars (vjk )j∈Ik . Let F be any nontrivial vector space. We show that for
every family
(wi1 ,...,in )(i1 ,...,in )∈I1 ×...×In ,
of vectors in F , there is some linear map h : E1 ⊗ · · · ⊗ En → F , such that

h(u1i1 ⊗ · · · ⊗ unin ) = wi1 ,...,in .

Then, by Proposition 22.3, it follows that

(u1i1 ⊗ · · · ⊗ unin )(i1 ,...,in )∈I1 ×...×In

is linearly independent. However, since (uki )i∈Ik is a basis for Ek , the u1i1 ⊗ · · · ⊗ unin also
generate E1 ⊗ · · · ⊗ En , and thus, they form a basis of E1 ⊗ · · · ⊗ En .
We define the function f : E1 × · · · × En → F as follows:
� � �
f( vj11 u1j1 , . . . , vjnn unjn ) = vj11 · · · vjnn wj1 ,...,jn .
j1 ∈I1 jn ∈In j1 ∈I1 ,...,jn ∈In

It is immediately verified that f is multilinear. By the universal mapping property of the

tensor product, the linear map f⊗ : E1 ⊗ · · · ⊗ En → F such that f = f⊗ ◦ ϕ, is the desired
map h.
In particular, when each Ik is finite and of size mk = dim(Ek ), we see that the dimension
of the tensor product E1 ⊗· · ·⊗En is m1 · · · mn . As a corollary of Proposition 22.6, if (uki )i∈Ik
is a basis for Ek , 1 ≤ k ≤ n, then every tensor z ∈ E1 ⊗ · · · ⊗ En can be written in a unique
way as
�
z= λi1 ,...,in u1i1 ⊗ · · · ⊗ unin ,
(i1 ,...,in ) ∈ I1 ×...×In

for some unique family of scalars λi1 ,...,in ∈ K, all zero except for a finite number.
22.3. SOME USEFUL ISOMORPHISMS FOR TENSOR PRODUCTS 595

22.3 Some Useful Isomorphisms for Tensor Products

Proposition 22.7 Given 3 vector spaces E, F, G, there exists unique canonical isomor-
phisms

(1) E ⊗ F � F ⊗ E

(2) (E ⊗ F ) ⊗ G � E ⊗ (F ⊗ G) � E ⊗ F ⊗ G

(3) (E ⊕ F ) ⊗ G � (E ⊗ G) ⊕ (F ⊗ G)

(4) K ⊗ E � E

such that respectively

(a) u ⊗ v �→ v ⊗ u
(b) (u ⊗ v) ⊗ w �→ u ⊗ (v ⊗ w) �→ u ⊗ v ⊗ w
(c) (u, v) ⊗ w �→ (u ⊗ w, v ⊗ w)
(d) λ ⊗ u �→ λu.

Proof . These isomorphisms are proved using the universal mapping property of tensor prod-
ucts. We illustrate the proof method on (2). Fix some w ∈ G. The map

(u, v) �→ u ⊗ v ⊗ w

from E × F to E ⊗ F ⊗ G is bilinear, and thus, there is a linear map fw : E ⊗ F → E ⊗ F ⊗ G,

such that fw (u ⊗ v) = u ⊗ v ⊗ w.
Next, consider the map
(z, w) �→ fw (z),
from (E ⊗ F ) × G into E ⊗ F ⊗ G. It is easily seen to be bilinear, and thus, it induces a
linear map
f : (E ⊗ F ) ⊗ G → E ⊗ F ⊗ G,
such that f ((u ⊗ v) ⊗ w) = u ⊗ v ⊗ w.
Also consider the map
(u, v, w) �→ (u ⊗ v) ⊗ w
from E × F × G to (E ⊗ F ) ⊗ G. It is trilinear, and thus, there is a linear map

g : E ⊗ F ⊗ G → (E ⊗ F ) ⊗ G,

such that g(u ⊗ v ⊗ w) = (u ⊗ v) ⊗ w. Clearly, f ◦ g and g ◦ f are identity maps, and thus,
f and g are isomorphisms. The other cases are similar.
596 CHAPTER 22. TENSOR ALGEBRAS

Given any three vector spaces, E, F, G, we have the canonical isomorphism

Hom(E, F ; G) ∼
= Hom(E, Hom(F, G)).

Indeed, any bilinear map, f : E × F → G, gives the linear map, ϕ(f ) ∈ Hom(E, Hom(F, G)),
where ϕ(f )(u) is the linear map in Hom(F, G) given by

ϕ(f )(u)(v) = f (u, v).

Conversely, given a linear map, g ∈ Hom(E, Hom(F, G)), we get the bilinear map, ψ(g),
given by
ψ(g)(u, v) = g(u)(v),
and it is clear that ϕ and ψ and mutual inverses. Consequently, we have the important
corollary:

Proposition 22.8 For any three vector spaces, E, F, G, we have the canonical isomorphism,

Hom(E ⊗ F, G) ∼
= Hom(E, Hom(F, G)),

22.4 Duality for Tensor Products

In this section, all vector spaces are assumed to have finite dimension. Let us now see how
tensor products behave under duality. For this, we define a pairing between E1∗ ⊗· · ·⊗En∗ and
E1 ⊗ · · · ⊗ En as follows: For any fixed (v1∗ , . . . , vn∗ ) ∈ E1∗ × · · · × En∗ , we have the multilinear
map,
lv1∗ ,...,vn∗ : (u1 , . . . , un ) �→ v1∗ (u1 ) · · · vn∗ (un ),
from E1 × · · · × En to K. The map lv1∗ ,...,vn∗ extends uniquely to a linear map,
Lv1∗ ,...,vn∗ : E1 ⊗ · · · ⊗ En −→ K. We also have the multilinear map,

(v1∗ , . . . , vn∗ ) �→ Lv1∗ ,...,vn∗ ,

from E1∗ × · · · × En∗ to Hom(E1 ⊗ · · · ⊗ En , K), which extends to a linear map, L, from
E1∗ ⊗ · · · ⊗ En∗ to Hom(E1 ⊗ · · · ⊗ En , K). However, in view of the isomorphism,

Hom(U ⊗ V, W ) ∼
= Hom(U, Hom(V, W )),

we can view L as a linear map,

L : (E1∗ ⊗ · · · ⊗ En∗ ) ⊗ (E1 ⊗ · · · ⊗ En ) → K,

which corresponds to a bilinear map,

(E1∗ ⊗ · · · ⊗ En∗ ) × (E1 ⊗ · · · ⊗ En ) −→ K,

22.4. DUALITY FOR TENSOR PRODUCTS 597

via the isomorphism (U ⊗ V )∗ ∼

= L(U, V ; K). It is easy to check that this bilinear map is
nondegenerate and thus, by Proposition 22.1, we have a canonical isomorphism,

(E1 ⊗ · · · ⊗ En )∗ ∼
= E1∗ ⊗ · · · ⊗ En∗ .

This, together with the isomorphism, L(E1 , . . . , En ; K) ∼= (E1 ⊗ · · · ⊗ En )∗ , yields a canonical

isomorphism
L(E1 , . . . , En ; K) ∼
= E1∗ ⊗ · · · ⊗ En∗ .
We prove another useful canonical isomorphism that allows us to treat linear maps as
tensors.
Let E and F be two vector spaces and let α : E ∗ × F → Hom(E, F ) be the map defined
such that
α(u∗ , f )(x) = u∗ (x)f,
for all u∗ ∈ E ∗ , f ∈ F , and x ∈ E. This map is clearly bilinear and thus, it induces a linear
map,
α⊗ : E ∗ ⊗ F → Hom(E, F ),
such that
α⊗ (u∗ ⊗ f )(x) = u∗ (x)f.

Proposition 22.9 If E and F are vector spaces with E of finite dimension, then the linear
map, α⊗ : E ∗ ⊗ F → Hom(E, F ), is a canonical isomorphism.

Proof . Let (ej )1≤j≤n be a basis of E and, as usual, let e∗j ∈ E ∗ be the linear form defined by

e∗j (ek ) = δj,k ,

where δj,k = 1 iff j = k and 0 otherwise. We know that (e∗j )1≤j≤n is a basis of E ∗ (this is
where we use the finite dimension of E). Now, for any linear map, f ∈ Hom(E, F ), for every
x = x1 e1 + · · · + xn en ∈ E, we have

f (x) = f (x1 e1 + · · · + xn en ) = x1 f (e1 ) + · · · + xn f (en ) = e∗1 (x)f (e1 ) + · · · + e∗n (x)f (en ).

Consequently, every linear map, f ∈ Hom(E, F ), can be expressed as

f (x) = e∗1 (x)f1 + · · · + e∗n (x)fn ,

for some fi ∈ F . Furthermore, if we apply f to ei , we get f (ei ) = fi , so the fi are unique.

Observe that
n
� n
�
(α⊗ (e∗1 ⊗ f1 + · · · + e∗n ⊗ fn ))(x) = (α⊗ (e∗i ⊗ fi ))(x) = e∗i (x)fi .
i=1 i=1
598 CHAPTER 22. TENSOR ALGEBRAS

Thus, α⊗ is surjective. As (e∗j )1≤j≤n is a basis of E ∗ , �

the tensors e∗j ⊗ f , with f ∈ F , span
E ∗ ⊗ F . Thus, every element of E ∗ ⊗ F is of the form ni=1 e∗i ⊗ fi , for some fi ∈ F . Assume
n
� n
�
α⊗ ( e∗i ⊗ fi ) = α⊗ ( e∗i ⊗ fi� ) = f,
i=1 i=1

for some fi , fi� ∈ F and some f ∈ Hom(E, F ). Then for every x ∈ E,

n
� n
�
e∗i (x)fi = e∗i (x)fi� = f (x).
i=1 i=1

Since the fi and fi� are uniquely determined by the linear map, f , we must have fi = fi� and
α⊗ is injective. Therefore, α⊗ is a bijection.
Note that in Proposition 22.9, the space F may have infinite dimension but E has finite
dimension. In view of the canonical isomorphism
Hom(E1 , . . . , En ; F ) ∼
= Hom(E1 ⊗ · · · ⊗ En , F )
and the canonical isomorphism (E1 ⊗ · · · ⊗ En )∗ ∼
= E1∗ ⊗ · · · ⊗ En∗ , where the Ei ’s are finite-
dimensional, Proposition 22.9 yields the canonical isomorphism
Hom(E1 , . . . , En ; F ) ∼
= E1∗ ⊗ · · · ⊗ En∗ ⊗ F.

22.5 Tensor Algebras

The tensor product
� ⊗ ·��
V · · ⊗ V�
m
is also denoted as m
�
V or V ⊗m
and is called the m-th tensor power of V (with V ⊗1 = V , and V ⊗0 = K). We can pack all
the tensor powers of V into the “big” vector space,
�
T (V ) = V ⊗m ,
m≥0

also denoted T • (V ), to avoid confusion with the tangent bundle. This is an interesting object
because we can define a multiplication operation on it which makes it into an algebra called
the tensor algebra of V . When V is of finite dimension n, this space corresponds to the
algebra of polynomials with coefficients in K in n noncommuting variables.
Let us recall the definition of an algebra over a field. Let K denote any (commutative)
field, although for our purposes, we may assume that K = R (and occasionally, K = C).
Since we will only be dealing with associative algebras with a multiplicative unit, we only
define algebras of this kind.
22.5. TENSOR ALGEBRAS 599

Definition 22.3 Given a field, K, a K-algebra is a K-vector space, A, together with a

bilinear operation, · : A × A → A, called multiplication, which makes A into a ring with
unity, 1 (or 1A , when we want to be very precise). This means that · is associative and that
there is a multiplicative identity element, 1, so that 1 · a = a · 1 = a, for all a ∈ A. Given
two K-algebras A and B, a K-algebra homomorphism, h : A → B, is a linear map that is
also a ring homomorphism, with h(1A ) = 1B .

For example, the ring, Mn (K), of all n × n matrices over a field, K, is a K-algebra.
There is an obvious notion of ideal of a K-algebra: An ideal, A ⊆ A, is a linear subspace
of A that is also a two-sided ideal with respect to multiplication in A. If the field K is
understood, we usually simply say an algebra instead of a K-algebra.
We would like to define a multiplication operation on T (V ) which makes it into a K-
algebra. As �
T (V ) = V ⊗i ,
i≥0

for every i ≥ 0, there is a natural injection ιn : V ⊗n → T (V ), and in particular, an injection

ι0 : K → T (V ). The multiplicative unit, 1, of T (V ) is the image, ι0 (1), in T (V ) of the unit,
1, of the field K. Since every v ∈ T (V ) can be expressed as a finite sum

v = ιn1 (v1 ) + · · · + ιnk (vk ),

where vi ∈ V ⊗ni and the ni are natural numbers with ni �= nj if i �= j, to define multiplica-
tion in T (V ), using bilinearity, it is enough to define multiplication operations,
· : V ⊗m × V ⊗n −→ V ⊗(m+n) , which, using the isomorphisms, V ⊗n ∼ = ιn (V ⊗n ), yield multi-
⊗m ⊗n ⊗(m+n)
plication operations, · : ιm (V ) × ιn (V ) −→ ιm+n (V ). More precisely, we use the
canonical isomorphism,
V ⊗m ⊗ V ⊗n ∼
= V ⊗(m+n) ,
which defines a bilinear operation,

V ⊗m × V ⊗n −→ V ⊗(m+n) ,

which is taken as the multiplication operation. The isomorphism V ⊗m ⊗ V ⊗n ∼

= V ⊗(m+n)
can be established by proving the isomorphisms

V ⊗m ⊗ V ⊗n ∼
= V ⊗m ⊗ V
� ⊗ ·��
· · ⊗ V�
n

V ⊗m
⊗V · · ⊗ V� ∼
� ⊗ ·�� = V ⊗(m+n) ,
n

which can be shown using methods similar to those used to proved associativity. Of course,
the multiplication, V ⊗m × V ⊗n −→ V ⊗(m+n) , is defined so that

(v1 ⊗ · · · ⊗ vm ) · (w1 ⊗ · · · ⊗ wn ) = v1 ⊗ · · · ⊗ vm ⊗ w1 ⊗ · · · ⊗ wn .
600 CHAPTER 22. TENSOR ALGEBRAS

(This has to be made rigorous by using isomorphisms involving the associativity of tensor
products, for details, see see Atiyah and Macdonald [9].)

Remark: It is important to note that multiplication in T (V ) is not commutative. Also, in

all rigor, the unit, 1, of T (V ) is not equal to 1, the unit of the field K. However, in view
of the injection ι0 : K → T (V ), for the sake of notational simplicity, we will denote 1 by 1.
More generally, in view of the injections ιn : V ⊗n → T (V ), we identify elements of V ⊗n with
their images in T (V ).
The algebra, T (V ), satisfies a universal mapping property which shows that it is unique
up to isomorphism. For simplicity of notation, let i : V → T (V ) be the natural injection of
V into T (V ).

Proposition 22.10 Given any K-algebra, A, for any linear map, f : V → A, there is a
unique K-algebra homomorphism, f : T (V ) → A, so that
f = f ◦ i,
as in the diagram below:
i
V � T (V )
f
f
� �
A
Proof . Left an an exercise (use Theorem 22.5).
Most algebras of interest arise�as well-chosen quotients of the tensor algebra T (V ). This
is true for the exterior algebra, (V ) (also called Grassmann algebra), where we take the
quotient of T (V ) modulo the ideal generated by all elements of the form v ⊗ v, where v ∈ V ,
and for the symmetric algebra, Sym(V ), where we take the quotient of T (V ) modulo the
ideal generated by all elements of the form v ⊗ w − w ⊗ v, where v, w ∈ V .
Algebras such as T (V ) are graded, in the sense that there is a sequence of subspaces,
V ⊗n ⊆ T (V ), such that �
T (V ) = V ⊗n
k≥0

and the multiplication, ⊗, behaves well w.r.t. the grading, i.e., ⊗ : V ⊗m × V ⊗n → V ⊗(m+n) .
Generally, a K-algebra, E, is said to be a graded algebra iff there is a sequence of subspaces,
E n ⊆ E, such that �
E= En
k≥0

(E = K) and the multiplication, ·, respects the grading, that is, · : E m × E n → E m+n .

0

Elements in E n are called homogeneous elements of rank (or degree) n.

In differential geometry and in physics it is necessary to consider slightly more general
tensors.
22.5. TENSOR ALGEBRAS 601

Definition 22.4 Given a vector space, V , for any pair of nonnegative integers, (r, s), the
tensor space, T r,s (V ), of type (r, s), is the tensor product

T r,s (V ) = V ⊗r ⊗ (V ∗ )⊗s = V
� ⊗ ·��
· · ⊗ V� ⊗ V ∗
· · ⊗ V �∗ ,
� ⊗ ·��
r s

with T 0,0 (V ) = K. We also define the tensor algebra, T •,• (V ), as the coproduct
�
T •,• (V ) = T r,s (V ).
r,s≥0

Tensors in T r,s (V ) are called homogeneous of degree (r, s).

Note that tensors in T r,0 (V ) are just our “old tensors” in V ⊗r . We make T •,• (V ) into an
algebra by defining multiplication operations,

T r1 ,s1 (V ) × T r2 ,s2 (V ) −→ T r1 +r2 ,s1 +s2 (V ),

in the usual way, namely: For u = u1 ⊗ · · · ⊗ ur1 ⊗ u∗1 ⊗ · · · ⊗ u∗s1 and

v = v1 ⊗ · · · ⊗ vr2 ⊗ v1∗ ⊗ · · · ⊗ vs∗2 , let

u ⊗ v = u1 ⊗ · · · ⊗ ur1 ⊗ v1 ⊗ · · · ⊗ vr2 ⊗ u∗1 ⊗ · · · ⊗ u∗s1 ⊗ v1∗ ⊗ · · · ⊗ vs∗2 .

Denote by Hom(V r , (V ∗ )s ; W ) the vector space of all multilinear maps from V r × (V ∗ )s

to W . Then, we have the universal mapping property which asserts that there is a canonical
isomorphism
Hom(T r,s (V ), W ) ∼= Hom(V r , (V ∗ )s ; W ).
In particular,
(T r,s (V ))∗ ∼
= Hom(V r , (V ∗ )s ; K).
For finite dimensional vector spaces, the duality of Section 22.4 is also easily extended to the
tensor spaces T r,s (V ). We define the pairing

T r,s (V ∗ ) × T r,s (V ) −→ K

as follows: If
v ∗ = v1∗ ⊗ · · · ⊗ vr∗ ⊗ ur+1 ⊗ · · · ⊗ ur+s ∈ T r,s (V ∗ )
and
∗ ∗
u = u1 ⊗ · · · ⊗ ur ⊗ vr+1 ⊗ · · · ⊗ vr+s ∈ T r,s (V ),
then
(v ∗ , u) = v1∗ (u1 ) · · · vr+s
∗
(ur+s ).
This is a nondegenerate pairing and thus, we get a canonical isomorphism,

(T r,s (V ))∗ ∼
= T r,s (V ∗ ).
602 CHAPTER 22. TENSOR ALGEBRAS

Consequently, we get a canonical isomorphism,

T r,s (V ∗ ) ∼
= Hom(V r , (V ∗ )s ; K).

Remark: The tensor spaces, T r,s (V ) are also denoted Tsr (V ). A tensor, α ∈ T r,s (V ) is
said to be contravariant in the first r arguments and covariant in the last s arguments.
This terminology refers to the way tensors behave under coordinate changes. Given a basis,
(e1 , . . . , en ), of V , if (e∗1 , . . . , e∗n ) denotes the dual basis, then every tensor α ∈ T r,s (V ) is
given by an expression of the form
� i ,...,i
α= aj11 ,...,jrs ei1 ⊗ · · · ⊗ eir ⊗ e∗j1 ⊗ · · · ⊗ e∗js .
i1 ,...,ir
j1 ,...,js

The tradition in classical tensor notation is to use lower indices on vectors and upper indices
on linear forms and in accordance to Einstein summation convention (or Einstein notation)
the position of the indices on the coefficients is reversed. Einstein summation convention is
to assume that a summation is performed for all values of every index that appears simul-
taneously once as an upper index and once as a lower index. According to this convention,
the tensor α above is written

α = aij11,...,i j1 js
,...,js ei1 ⊗ · · · ⊗ eir ⊗ e ⊗ · · · ⊗ e .
r

An older view of tensors is that they are multidimensional arrays of coefficients,

� i1 ,...,ir �
aj1 ,...,js ,

subject to the rules for changes of bases.

Another operation on general tensors, contraction, is useful in differential geometry.

Definition 22.5 For all r, s ≥ 1, the contraction, ci,j : T r,s (V ) → T r−1,s−1 (V ), with 1 ≤ i ≤
r and 1 ≤ j ≤ s, is the linear map defined on generators by

ci,j (u1 ⊗ · · · ⊗ ur ⊗ v1∗ ⊗ · · · ⊗ vs∗ )

= vj∗ (ui ) u1 ⊗ · · · ⊗ u�i ⊗ · · · ⊗ ur ⊗ v1∗ ⊗ · · · ⊗ v�j∗ ⊗ · · · ⊗ vs∗ ,

where the hat over an argument means that it should be omitted.

Let us figure our what is c1,1 : T 1,1 (V ) → R, that is c1,1 : V ⊗ V ∗ → R. If (e1 , . . . , en )

is a basis of V and (e∗1 , . . . , e∗n ) is the dual basis, every h ∈ V ⊗ V ∗ ∼
= Hom(V, V ) can be
expressed as
�n
h= aij ei ⊗ e∗j .
i,j=1
22.5. TENSOR ALGEBRAS 603

As
c1,1 (ei ⊗ e∗j ) = δi,j ,
we get
n
�
c1,1 (h) = aii = tr(h),
i=1

where tr(h) is the trace of h, where h is viewed as the linear map given by the matrix, (aij ).
Actually, since c1,1 is defined independently of any basis, c1,1 provides an intrinsic definition
of the trace of a linear map, h ∈ Hom(V, V ).

Remark: Using the Einstein summation convention, if

α = aij11,...,i j1 js
,...,js ei1 ⊗ · · · ⊗ eir ⊗ e ⊗ · · · ⊗ e ,
r

then
i ,...,i
ck,l (α) = aj11 ,...,jk−1
,i,ik+1 ...,ir
e ⊗ · · · ⊗ e� j1 �j js
ik ⊗ · · · ⊗ e ir ⊗ e ⊗ · · · ⊗ e l ⊗ · · · ⊗ e .
l−1 ,i,jl+1 ,...,js i1

If E and F are two K-algebras, we know that their tensor product, E ⊗ F , exists as a
vector space. We can make E ⊗ F into an algebra as well. Indeed, we have the multilinear
map
E × F × E × F −→ E ⊗ F
given by (a, b, c, d) �→ (ac) ⊗ (bd), where ac is the product of a and c in E and bd is the
product of b and d in F . By the universal mapping property, we get a linear map,

E ⊗ F ⊗ E ⊗ F −→ E ⊗ F.

Using the isomorphism,

E⊗F ⊗E⊗F ∼
= (E ⊗ F ) ⊗ (E ⊗ F ),

we get a linear map,

(E ⊗ F ) ⊗ (E ⊗ F ) −→ E ⊗ F,
and thus, a bilinear map,
(E ⊗ F ) × (E ⊗ F ) −→ E ⊗ F,
which is our multiplication operation in E ⊗ F . This multiplication is determined by

(a ⊗ b) · (c ⊗ d) = (ac) ⊗ (bd).

One immediately checks that E ⊗ F with this multiplication is a K-algebra.

We now turn to symmetric tensors.
604 CHAPTER 22. TENSOR ALGEBRAS

22.6 Symmetric Tensor Powers

Our goal is to come up with a notion of tensor product that will allow us to treat symmetric
multilinear maps as linear maps. First, note that we have to restrict ourselves to a single
vector space, E, rather then n vector spaces E1 , . . . , En , so that symmetry makes sense.
Recall that a multilinear map, f : E n → F , is symmetric iff

f (uσ(1) , . . . , uσ(n) ) = f (u1 , . . . , un ),

for all ui ∈ E and all permutations, σ : {1, . . . , n} → {1, . . . , n}. The group of permutations
on {1, . . . , n} (the symmetric group) is denoted Sn . The vector space of all symmetric
multilinear maps, f : E n → F , is denoted by Sn (E; F ). Note that S1 (E; F ) = Hom(E, F ).
We could proceed directly as in Theorem 22.5, and construct symmetric tensor products
from scratch. However, since we already have the notion of a tensor product, there is a more
economical method. First, we define symmetric tensor powers.

Definition 22.6 An n-th symmetric tensor power of a vector space E, where n ≥ 1, is a

vector space S, together with a symmetric multilinear map ϕ : E n → S, such that, for every
vector space F and for every symmetric multilinear map f : E n → F , there is a unique linear
map f⊙ : S → F , with
f (u1 , . . . , un ) = f⊙ (ϕ(u1 , . . . , un )),
for all u1 , . . . , un ∈ E, or for short
f = f⊙ ◦ ϕ.
Equivalently, there is a unique linear map f⊙ such that the following diagram commutes:
ϕ
En � S
f⊙
f � �
F

First, we show that any two symmetric n-th tensor powers (S1 , ϕ1 ) and (S2 , ϕ2 ) for E,
are isomorphic.

Proposition 22.11 Given any two symmetric n-th tensor powers (S1 , ϕ1 ) and (S2 , ϕ2 ) for
E, there is an isomorphism h : S1 → S2 such that

ϕ2 = h ◦ ϕ1 .

Proof . Replace tensor product by n-th symmetric tensor power in the proof of Proposition
22.4.
We now give a construction that produces a symmetric n-th tensor power of a vector
space E.
22.6. SYMMETRIC TENSOR POWERS 605

Theorem 22.12 Given a vector space E, a symmetric n-th tensor power (Symn (E), ϕ)
for E can be constructed (n ≥ 1). Furthermore, denoting ϕ(u1 , . . . , un ) as u1 ⊙ · · · ⊙ un ,
the symmetric tensor power Symn (E) is generated by the vectors u1 ⊙ · · · ⊙ un , where
u1 , . . . , un ∈ E, and for every symmetric multilinear map f : E n → F , the unique linear
map f⊙ : Symn (E) → F such that f = f⊙ ◦ ϕ, is defined by

f⊙ (u1 ⊙ · · · ⊙ un ) = f (u1 , . . . , un ),

on the generators u1 ⊙ · · · ⊙ un of Symn (E).

Proof . The tensor power E ⊗n is too big, and thus, we define an appropriate quotient. Let
C be the subspace of E ⊗n generated by the vectors of the form

u1 ⊗ · · · ⊗ un − uσ(1) ⊗ · · · ⊗ uσ(n) ,

for all ui ∈ E, and all permutations σ : {1, . . . , n} → {1, . . . , n}. We claim that the quotient
space (E ⊗n )/C does the job.
Let p : E ⊗n → (E ⊗n )/C be the quotient map. Let ϕ : E n → (E ⊗n )/C be the map

(u1 , . . . , un ) �→ p(u1 ⊗ · · · ⊗ un ),

or equivalently, ϕ = p ◦ ϕ0 , where ϕ0 (u1 , . . . , un ) = u1 ⊗ · · · ⊗ un .

Let us denote ϕ(u1 , . . . , un ) as u1 ⊙ · · · ⊙ un . It is clear that ϕ is symmetric. Since the
vectors u1 ⊗ · · · ⊗ un generate E ⊗n , and p is surjective, the vectors u1 ⊙ · · · ⊙ un generate
(E ⊗n )/C.
Given any symmetric multilinear map f : E n → F , there is a linear map f⊗ : E ⊗n → F
such that f = f⊗ ◦ ϕ0 , as in the diagram below:
ϕ0
En � E ⊗n
f⊗
f
� �
F

However, since f is symmetric, we have f⊗ (z) = 0 for every z ∈ E ⊗n . Thus, we get an

induced linear map h : (E ⊗n )/C → F , such that h([z]) = f⊗ (z), where [z] is the equivalence
class in (E ⊗n )/C of z ∈ E ⊗n :
p◦ϕ0
En � (E ⊗n )/C
h
f
� �
F
However, if a linear map f⊙ : (E ⊗n )/C → F exists, since the vectors u1 ⊙ · · · ⊙ un generate
(E ⊗n )/C, we must have
f⊙ (u1 ⊙ · · · ⊙ un ) = f (u1 , . . . , un ),
606 CHAPTER 22. TENSOR ALGEBRAS

which shows that h and f⊙ agree. Thus, Symn (E) = (E ⊗n )/C and ϕ constitute a symmetric
n-th tensor power of E.
Again, the actual construction is not important. What is important is that the symmetric
n-th power has the universal mapping property with respect to symmetric multilinear maps.

Remark: The notation ⊙ for the commutative multiplication of symmetric tensor powers is
not standard. Another notation commonly used is ·. We often abbreviate “symmetric tensor
power” as “symmetric power”. The symmetric power, Symn (E), is also denoted n
�n Sym E or
S(E). To be consistent with the use of ⊙, we could have used the notation E. Clearly,
1 ∼ 0
Sym (E) = E and it is convenient to set Sym (E) = K.
The fact that the map ϕ : E n → Symn (E) is symmetric and multinear, can also be
expressed as follows:

u1 ⊙ · · · ⊙ (ui + vi ) ⊙ · · · ⊙ un = (u1 ⊙ · · · ⊙ ui ⊙ · · · ⊙ un )
+ (u1 ⊙ · · · ⊙ vi ⊙ · · · ⊙ un ),
u1 ⊙ · · · ⊙ (λui ) ⊙ · · · ⊙ un = λ(u1 ⊙ · · · ⊙ ui ⊙ · · · ⊙ un ),
uσ(1) ⊙ · · · ⊙ uσ(n) = u1 ⊙ · · · ⊙ un ,

for all permutations σ ∈ Sn .

The last identity shows that the “operation” ⊙ is commutative. Thus, we can view the
symmetric tensor u1 ⊙ · · · ⊙ un as a multiset.
Theorem 22.12 yields a canonical isomorphism

Hom(Symn (E), F ) ∼
= S(E n ; F ),

between the vector space of linear maps Hom(Symn (E), F ), and the vector space of sym-
metric multilinear maps S(E n ; F ), via the linear map − ◦ ϕ defined by

h �→ h ◦ ϕ,

where h ∈ Hom(Symn (E), F ). Indeed, h ◦ ϕ is clearly symmetric multilinear, and since by

Theorem 22.12, for every symmetric multilinear map f ∈ S(E n ; F ), there is a unique linear
map f⊙ ∈ Hom(Symn (E), F ) such that f = f⊙ ◦ ϕ, the map − ◦ ϕ is bijective. As a matter
of fact, its inverse is the map
f �→ f⊙ .
In particular, when F = K, we get a canonical isomorphism

(Symn (E))∗ ∼
= Sn (E; K).

Symmetric tensors in Symn (E) are also called symmetric n-tensors, and tensors of the
form u1 ⊙ · · · ⊙ un , where ui ∈ E, are called simple (or decomposable) symmetric n-tensors.
22.7. BASES OF SYMMETRIC POWERS 607

Those symmetric n-tensors that are not simple are often called compound symmetric n-
tensors.
Given two linear maps f : E → E � and g : E → E � , we can define h : E × E → Sym2 (E � )
by
h(u, v) = f (u) ⊙ g(v).
It is immediately verified that h is symmetric bilinear, and thus, it induces a unique linear
map
f ⊙ g : Sym2 (E) → Sym2 (E � ),
such that
(f ⊙ g)(u ⊙ v) = f (u) ⊙ g(u).

If we also have linear maps f � : E � → E �� and g � : E � → E �� , we can easily verify that

(f � ◦ f ) ⊙ (g � ◦ g) = (f � ⊙ g � ) ◦ (f ⊙ g).

The generalization to the symmetric tensor product f1 ⊙ · · · ⊙ fn of n ≥ 3 linear maps

fi : E → E � is immediate, and left to the reader.

22.7 Bases of Symmetric Powers

The vectors u1 ⊙ · · · ⊙ un , where u1 , . . . , un ∈ E, generate Symn (E), but they are not linearly
independent. We will prove a version of Proposition 22.6 for symmetric tensor powers. For
this, recall that a (finite) multiset over a set I is a function M : I → N, such that M (i) �= 0
for finitely many i ∈ I, and that the set of all multisets over I is denoted as N(I) . We let
� any multiset M ∈ N , note
(I)
dom(M ) = {i ∈ �I | M (i) �= 0}, which is a finite �set. Then, for
that the sum i∈I M (i) makes sense, since i∈I M (i) = i∈dom(M ) M (i), and dom(M )
is finite. For every multiset M ∈ N(I) , for any n ≥ 2, we define the set JM of functions
η : {1, . . . , n} → dom(M ), as follows:
�
JM = {η | η : {1, . . . , n} → dom(M ), |η −1 (i)| = M (i), i ∈ dom(M ), M (i) = n}.
i∈I

�
In other words, if i∈I M (i) = n and dom(M ) = {i1 , . . . , ik },1 any function η ∈ JM specifies
a sequence of length n, consisting of M (i1 ) occurrences of i1 , M (i2 ) occurrences of i2 , . . .,
M (ik ) occurrences of ik . Intuitively, any η defines a “permutation” of the sequence (of length
n)
�i1 , . . . , i1 , i2 , . . . , i2 , . . . , ik , . . . , ik �.
� �� � � �� � � �� �
M (i1 ) M (i2 ) M (ik )

1
Note that must have k ≤ n.
608 CHAPTER 22. TENSOR ALGEBRAS

Given any k ≥ 1, and any u ∈ E, we denote

� ⊙ ·��
u · · ⊙ u�
k

as u⊙k .
We can now prove the following Proposition.

Proposition 22.13 Given a vector space E, if (ui )i∈I is a basis for E, then the family of
vectors � �
⊙M (i1 ) ⊙M (ik )
ui 1 ⊙ · · · ⊙ ui k �
M ∈N(I) , i∈I M (i)=n, {i1 ,...,ik }=dom(M )

is a basis of the symmetric n-th tensor power Symn (E).

Proof . The proof is very similar to that of Proposition 22.6. For any nontrivial vector space
F , for any family of vectors
(wM )M ∈N(I) , �i∈I M (i)=n ,
we show the existence
� of a symmetric multilinear map h : Symn (E) → F , such that for every
M ∈ N(I) with i∈I M (i) = n, we have
⊙M (i1 ) ⊙M (ik )
h(ui1 ⊙ · · · ⊙ ui k ) = wM ,

where {i1 , . . . , ik } = dom(M ). We define the map f : E n → F as follows:

� � � �� �
1 1 n n 1 n
f( v j 1 uj 1 , . . . , v j n uj n ) = vη(1) · · · vη(n) wM .
j1 ∈I jn ∈I � M ∈N
(I) η∈JM
i∈I M (i)=n

It is not difficult to verify that f is symmetric and multilinear. By the universal mapping
property of the symmetric tensor product, the linear map f⊙ : Symn (E) → F such that
f = f⊙ ◦ ϕ, is the desired map h. Then, by Proposition 22.3, it follows that the family
� �
⊙M (i1 ) ⊙M (ik )
ui 1 ⊙ · · · ⊙ ui k �
M ∈N(I) , i∈I M (i)=n, {i1 ,...,ik }=dom(M )

is linearly independent. Using the commutativity of ⊙, we can also show that these vectors
generate Symn (E), and thus, they form a basis for Symn (E). The details are left as an
exercise.

As a consequence, when I is finite, say of size p = dim(E), the dimension of Symn (E) is
� that j1 + · · · + jp = n, jk ≥ 0. We leave as
the number of finite multisets (j1 , . . . , j�p ), such
an exercise to show that this number is p+n−1 n
. Thus, if dim(E) = p, then the dimension of
22.8. SOME USEFUL ISOMORPHISMS FOR SYMMETRIC POWERS 609
� �
Symn (E) is p+n−1
n
. Compare with the dimension of E ⊗n , which is pn . In particular, when
p = 2, the dimension of Symn (E) is n + 1. This can also be seen directly.
� �
Remark: The number p+n−1 n
is also the number of homogeneous monomials

X1j1 · · · Xpjp
of total degree n in p variables (we have j1 + · · · + jp = n). This is not a coincidence!
Symmetric tensor products are closely related to polynomials (for more on this, see the next
remark).
Given a vector space E and a basis (ui )i∈I for E, Proposition 22.13 shows that every
symmetric tensor z ∈ Symn (E) can be written in a unique way as
� ⊙M (i ) ⊙M (i )
z= λM u i 1 1 ⊙ · · · ⊙ ui k k ,
(I)
� M ∈N
i∈I M (i)=n
{i1 ,...,ik }=dom(M )

for some unique family of scalars λM ∈ K, all zero except for a finite number.
This looks like a homogeneous polynomial of total degree n, where the monomials of total
degree n are the symmetric tensors
⊙M (i1 ) ⊙M (ik )
ui 1 ⊙ · · · ⊙ ui k ,
in the “indeterminates” ui , where i ∈ I (recall that M (i1 ) + · · · + M (ik ) = n). Again, this
is not a coincidence. Polynomials can be defined in terms of symmetric tensors.

22.8 Some Useful Isomorphisms for Symmetric Powers

We can show the following property of the symmetric tensor product, using the proof tech-
nique of Proposition 22.7:
n
�
Symn (E ⊕ F ) ∼
= Symk (E) ⊗ Symn−k (F ).
k=0

22.9 Duality for Symmetric Powers

In this section, all vector spaces are assumed to have finite dimension. We define a nonde-
generate pairing, Symn (E ∗ ) × Symn (E) −→ K, as follows: Consider the multilinear map,
(E ∗ )n × E n −→ K,
given by �
(v1∗ , . . . , vn∗ , u1 , . . . , un ) �→ ∗
vσ(1) ∗
(u1 ) · · · vσ(n) (un ).
σ∈Sn
610 CHAPTER 22. TENSOR ALGEBRAS

Note that the expression on the right-hand side is “almost” the determinant, det(vj∗ (ui )),
except that the sign sgn(σ) is missing (where sgn(σ) is the signature of the permutation
σ, that is, the parity of the number of transpositions into which σ can be factored). Such
an expression is called a permanent. It is easily checked that this expression is symmetric
w.r.t. the ui ’s and also w.r.t. the vj∗ . For any fixed (v1∗ , . . . , vn∗ ) ∈ (E ∗ )n , we get a symmetric
multinear map, �
∗ ∗
lv1∗ ,...,vn∗ : (u1 , . . . , un ) �→ vσ(1) (u1 ) · · · vσ(n) (un ),
σ∈Sn

from E to K. The map lv1∗ ,...,vn∗ extends uniquely to a linear map, Lv1∗ ,...,vn∗ : Symn (E) → K.
n

Now, we also have the symmetric multilinear map,

(v1∗ , . . . , vn∗ ) �→ Lv1∗ ,...,vn∗ ,

from (E ∗ )n to Hom(Symn (E), K), which extends to a linear map, L, from Symn (E ∗ ) to
Hom(Symn (E), K). However, in view of the isomorphism,

Hom(U ⊗ V, W ) ∼
= Hom(U, Hom(V, W )),

we can view L as a linear map,

L : Symn (E ∗ ) ⊗ Symn (E) −→ K,

which corresponds to a bilinear map,

Symn (E ∗ ) × Symn (E) −→ K.

Now, this pairing in nondegenerate. This can be done using bases and we leave it as an exer-
cise to the reader (see Knapp [89], Appendix A). Therefore, we get a canonical isomorphism,

(Symn (E))∗ ∼
= Symn (E ∗ ).

Since we also have an isomorphism

(Symn (E))∗ ∼
= Sn (E, K),

we get a canonical isomorphism

Symn (E ∗ ) ∼
= Sn (E, K)

which allows us to interpret symmetric tensors over E ∗ as symmetric multilinear maps.

Remark: The isomorphism, µ : Symn (E ∗ ) ∼ = Sn (E, K), discussed above can be described
explicity as the linear extension of the map given by
�
µ(v1∗ ⊙ · · · ⊙ vn∗ )(u1 , . . . , un ) = ∗
vσ(1) ∗
(u1 ) · · · vσ(n) (un ).
σ∈Sn
22.10. SYMMETRIC ALGEBRAS 611

Now, the map from E n to Symn (E) given by (u1 , . . . , un ) �→ u1 ⊙ · · · ⊙ un yields a

surjection, π : E ⊗n → Symn (E). Because we are dealing with vector spaces, this map has
some section, that is, there is some injection, ι : Symn (E) → E ⊗n , with π ◦ι = id. If our field,
K, has characteristic 0, then there is a special section having a natural definition involving
a symmetrization process defined as follows: For every permutation, σ, we have the map,
rσ : E n → E ⊗n , given by

rσ (u1 , . . . , un ) = uσ(1) ⊗ · · · ⊗ uσ(n) .

As rσ is clearly multilinear, rσ extends to a linear map, rσ : E ⊗n → E ⊗n , and we get a map,

Sn × E ⊗n −→ E ⊗n , namely,
σ · z = rσ (z).
It is immediately checked that this is a left action of the symmetric group, Sn , on E ⊗n and
the tensors z ∈ E ⊗n such that

σ · z = z, for all σ ∈ Sn

are called symmetrized tensors. We define the map, ι : E n → E ⊗n , by

1 � 1 �
ι(u1 , . . . , un ) = σ · (u1 ⊗ · · · ⊗ un ) = uσ(1) ⊗ · · · ⊗ uσ(n) .
n! σ∈S n! σ∈S
n n

As the right hand side is clearly symmetric, we get a linear map, ι : Symn (E) → E ⊗n .
Clearly, ι(Symn (E)) is the set of symmetrized tensors in E ⊗n . If we consider the map,
S = ι ◦ π : E ⊗n −→ E ⊗n , it is easy to check that S ◦ S = S. Therefore, S is a projection and
by linear algebra, we know that

E ⊗n = S(E ⊗n ) ⊕ Ker S = ι(Symn (E)) ⊕ Ker S.

It turns out that Ker S = E ⊗n ∩ I = Ker π, where I is the two-sided ideal of T (E) generated
by all tensors of the form u ⊗ v − v ⊗ u ∈ E ⊗2 (for example, see Knapp [89], Appendix A).
Therefore, ι is injective,

E ⊗n = ι(Symn (E)) ⊕ E ⊗n ∩ I = ι(Symn (E)) ⊕ Ker π,

and the symmetric tensor power, Symn (E), is naturally embedded into E ⊗n .

22.10 Symmetric Algebras

As in the case of tensors, we can pack together all the symmetric powers, Symn (V ), into an
algebra, �
Sym(V ) = Symm (V ),
m≥0
612 CHAPTER 22. TENSOR ALGEBRAS

called the symmetric tensor algebra of V . We could adapt what we did in Section 22.5 for
general tensor powers to symmetric tensors but since we already have the algebra, T (V ),
we can proceed faster. If I is the two-sided ideal generated by all tensors of the form
u ⊗ v − v ⊗ u ∈ V ⊗2 , we set
Sym• (V ) = T (V )/I.
Then, Sym• (V ) automatically inherits a multiplication operation which is commutative and
since T (V ) is graded, that is, �
T (V ) = V ⊗m ,
m≥0

we have �
Sym• (V ) = V ⊗m /(I ∩ V ⊗m ).
m≥0

However, it is easy to check that

Symm (V ) ∼
= V ⊗m /(I ∩ V ⊗m ),

so
Sym• (V ) ∼
= Sym(V ).
When V is of finite dimension, n, T (V ) corresponds to the algebra of polynomials with
coefficients in K in n variables (this can be seen from Proposition 22.13). When V is of
infinite dimension and (ui )i∈I is a basis of V , the algebra, Sym(V ), corresponds to the
algebra of polynomials in infinitely many variables in I. What’s nice about the symmetric
tensor algebra, Sym(V ), is that it provides an intrinsic definition of a polynomial algebra in
any set, I, of variables.
It is also easy to see that Sym(V ) satisfies the following universal mapping property:
Proposition 22.14 Given any commutative K-algebra, A, for any linear map, f : V → A,
there is a unique K-algebra homomorphism, f : Sym(V ) → A, so that

f = f ◦ i,

as in the diagram below:

i�
V Sym(V )
f
f
� �
A

Remark: If E is finite-dimensional, recall the isomorphism, µ : Symn (E ∗ ) −→ Sn (E, K),

defined as the linear extension of the map given by
�
µ(v1∗ ⊙ · · · ⊙ vn∗ )(u1 , . . . , un ) = ∗
vσ(1) ∗
(u1 ) · · · vσ(n) (un ),
σ∈Sn
22.11. EXTERIOR TENSOR POWERS 613

Now, we have also a multiplication operation, Symm (E ∗ )×Symn (E ∗ ) −→ Symm+n (E ∗ ). The

following question then arises:
Can we define a multiplication, Sm (E, K) × Sn (E, K) −→ Sm+n (E, K), directly on sym-
metric multilinear forms, so that the following diagram commutes:
⊙
Symm (E ∗ ) × Symn (E ∗ ) � Symm+n (E ∗ )
µ×µ µ
� �
·
Sm (E, K) × Sn (E, K) � Sm+n (E, K).

The answer is yes! The solution is to define this multiplication such that, for f ∈ Sm (E, K)
and g ∈ Sn (E, K),
�
(f · g)(u1 , . . . , um+n ) = f (uσ(1) , . . . , uσ(m) )g(uσ(m+1) , . . . , uσ(m+n) ),
σ∈shuffle(m,n)

where shuffle(m, n) consists of all (m, n)-“shuffles”, that is, permutations, σ, of {1, . . . m+n},
such that σ(1) < · · · < σ(m) and σ(m + 1) < · · · < σ(m + n). We urge the reader to check
this fact.
Another useful canonical isomorphim (of K-algebras) is
Sym(E ⊕ F ) ∼
= Sym(E) ⊗ Sym(F ).

22.11 Exterior Tensor Powers

We now consider alternating (also called skew-symmetric) multilinear�n maps and exterior
tensor powers (also called alternating tensor powers), denoted (E). In many respect,
alternating multilinear maps and exterior tensor powers can be treated much like symmetric
tensor powers except that the sign, sgn(σ), needs to be inserted in front of the formulae valid
for symmetric powers. Roughly speaking, we are now in the world of determinants rather
than in the world of permanents. However, there �nare also some fundamental differences, one
of which being that the exterior tensor power, (E), is the trivial vector space, (0), when
E is finite-dimensional and when n > dim(E). As in the case of symmetric tensor powers,
since we already have the tensor algebra, T (V ), we can proceed rather quickly. But first, let
us review some basic definitions and facts.
Definition 22.7 Let f : E n → F be a multilinear map. We say that f alternating iff
f (u1 , . . . , un ) = 0 whenever ui = ui+1 , for some i with 1 ≤ i ≤ n − 1, for all ui ∈ E,
that is, f (u1 , . . . , un ) = 0 whenever two adjacent arguments are identical. We say that f is
skew-symmetric (or anti-symmetric) iff
f (uσ(1) , . . . , uσ(n) ) = sgn(σ)f (u1 , . . . , un ),
for every permutation, σ ∈ Sn , and all ui ∈ E.
614 CHAPTER 22. TENSOR ALGEBRAS

For n = 1, we agree that every linear map, f : E → F , is alternating. The vector

space of all multilinear alternating maps, f : E n → F , is denoted Altn (E; F ). Note that
Alt1 (E; F ) = Hom(E, F ). The following basic proposition shows the relationship between
alternation and skew-symmetry.

Proposition 22.15 Let f : E n → F be a multilinear map. If f is alternating, then the

following properties hold:

(1) For all i, with 1 ≤ i ≤ n − 1,

f (. . . , ui , ui+1 , . . .) = −f (. . . , ui+1 , ui , . . .).

(2) For every permutation, σ ∈ Sn ,

f (uσ(1) , . . . , uσ(n) ) = sgn(σ)f (u1 , . . . , un ).

(3) For all i, j, with 1 ≤ i < j ≤ n,

f (. . . , ui , . . . uj , . . .) = 0 whenever ui = uj .

Moreover, if our field, K, has characteristic different from 2, then every skew-symmetric
multilinear map is alternating.

Proof . (i) By multilinearity applied twice, we have

f (. . . , ui + ui+1 , ui + ui+1 , . . .) = f (. . . , ui , ui , . . .) + f (. . . , ui , ui+1 , . . .)

+ f (. . . , ui+1 , ui , . . .) + f (. . . , ui+1 , ui+1 , . . .).

Since f is alternating, we get

0 = f (. . . , ui , ui+1 , . . .) + f (. . . , ui+1 , ui , . . .),

that is, f (. . . , ui , ui+1 , . . .) = −f (. . . , ui+1 , ui , . . .).

(ii) Clearly, the symmetric group, Sn , acts on Altn (E; F ) on the left, via

σ · f (u1 , . . . , un ) = f (uσ(1) , . . . , uσ(n) ).

Consequently, as Sn is generated by the transpositions (permutations that swap exactly two

elements), since for a transposition, (ii) is simply (i), we deduce (ii) by induction on the
number of transpositions in σ.
(iii) There is a permutation, σ, that sends ui and uj respectively to u1 and u2 . As f is
alternating,
f (uσ(1) , . . . , uσ(n) ) = 0.
22.11. EXTERIOR TENSOR POWERS 615

However, by (ii),
f (u1 , . . . , un ) = sgn(σ)f (uσ(1) , . . . , uσ(n) ) = 0.
Now, when f is skew-symmetric, if σ is the transposition swapping ui and ui+1 = ui , as
sgn(σ) = −1, we get
f (. . . , ui , ui , . . .) = −f (. . . , ui , ui , . . .),
so that
2f (. . . , ui , ui , . . .) = 0,
and in every characteristic except 2, we conclude that f (. . . , ui , ui , . . .) = 0, namely, f is
alternating.
Proposition 22.15 shows that in every characteristic except 2, alternating and skew-
symmetric multilinear maps are identical. Using Proposition 22.15 we easily deduce the
following crucial fact:
Proposition 22.16 Let f : E n → F be an alternating multilinear map. For any families of
vectors, (u1 , . . . , un ) and (v1 , . . . , vn ), with ui , vi ∈ E, if
n
�
vj = aij ui , 1 ≤ j ≤ n,
i=1

then
� �
�
f (v1 , . . . , vn ) = sgn(σ) aσ(1),1 · · · aσ(n),n f (u1 , . . . , un ) = det(A)f (u1 , . . . , un ),
σ∈Sn

where A is the n × n matrix, A = (aij ).

Proof . Use property (ii) of Proposition 22.15. .
We are now ready to define and construct exterior tensor powers.
Definition 22.8 An n-th exterior tensor power of a vector space, E, where n ≥ 1, is a
vector space, A, together with an alternating multilinear map, ϕ : E n → A, such that, for
every vector space, F , and for every alternating multilinear map, f : E n → F , there is a
unique linear map, f∧ : A → F , with
f (u1 , . . . , un ) = f∧ (ϕ(u1 , . . . , un )),
for all u1 , . . . , un ∈ E, or for short
f = f∧ ◦ ϕ.
Equivalently, there is a unique linear map f∧ such that the following diagram commutes:
ϕ
En � A
f∧
f � �
F
616 CHAPTER 22. TENSOR ALGEBRAS

First, we show that any two n-th exterior tensor powers (A1 , ϕ1 ) and (A2 , ϕ2 ) for E, are
isomorphic.

Proposition 22.17 Given any two n-th exterior tensor powers (A1 , ϕ1 ) and (A2 , ϕ2 ) for E,
there is an isomorphism h : A1 → A2 such that

ϕ2 = h ◦ ϕ1 .

Proof . Replace tensor product by n exterior tensor power in the proof of Proposition 22.4.

We now give a construction that produces an n-th exterior tensor power of a vector space
E.
�
Theorem 22.18 Given a vector space E, an n-th exterior tensor power ( n (E), ϕ) for E
� (n ≥ 1). Furthermore, denoting ϕ(u1 , . . . , un ) as u1 ∧· · ·∧un , the exterior
can be constructed
tensor power n (E) is generated by the vectors u1 ∧ · · · ∧ un , where u1 , . . .�
, un ∈ E, and for
every alternating multilinear map f : E → F , the unique linear map f∧ : n (E) → F such
n

that f = f∧ ◦ ϕ, is defined by

f∧ (u1 ∧ · · · ∧ un ) = f (u1 , . . . , un ),
�
on the generators u1 ∧ · · · ∧ un of n (E).

Proof sketch. We can give a quick proof using the tensor algebra, T (E). let Ia be the
two-sided ideal of T (E) generated by all tensors of the form u ⊗ u ∈ E ⊗2 . Then, let
n
�
(E) = E ⊗n /(Ia ∩ E ⊗n )

⊗n
�n
�n π : E →
and let π be the projection, (E). If we let u1 ∧ · · · ∧ un = π(u1 ⊗ · · · ⊗ un ), it
is easy to check that ( (E), ∧) satisfies the conditions of Theorem 22.18.

Remark: We can also define

� n
��
(E) = T (E)/Ia = (E),
n≥0

the exterior algebra of E. This is the skew-symmetric counterpart of Sym(E) and we will
study it a little later.
� �
For simplicity of notation, we may write� n E for n (E). We also abbreviate � “exterior
tensor power” as “exterior power”. Clearly, 1 ∼
(E) = E and it is convenient to set 0 (E) =
K.
22.11. EXTERIOR TENSOR POWERS 617
�n
The fact that the map ϕ : E n → (E) is alternating and multinear, can also be expressed
as follows:
u1 ∧ · · · ∧ (ui + vi ) ∧ · · · ∧ un = (u1 ∧ · · · ∧ ui ∧ · · · ∧ un )
+ (u1 ∧ · · · ∧ vi ∧ · · · ∧ un ),
u1 ∧ · · · ∧ (λui ) ∧ · · · ∧ un = λ(u1 ∧ · · · ∧ ui ∧ · · · ∧ un ),
uσ(1) ∧ · · · ∧ uσ(n) = sgn(σ) u1 ∧ · · · ∧ un ,
for all σ ∈ Sn .
Theorem 22.18 yields a canonical isomorphism
n
�
Hom( (E), F ) ∼
= Altn (E; F ),
�
between the vector space of linear maps Hom( n (E), F ), and the vector space of alternating
multilinear maps Altn (E; F ), via the linear map − ◦ ϕ defined by
h �→ h ◦ ϕ,
�n
where h ∈ Hom( (E), F ). In particular, when F = K, we get a canonical isomorphism
�n �∗
�
(E) ∼ = Altn (E; K).

�
Tensors α ∈ n (E) are called alternating n-tensors or alternating tensors of degree n
and we write deg(α) = n. Tensors of the form u1 ∧ · · · ∧ un , where ui ∈ E, are called simple
(or decomposable) alternating n-tensors. Those alternating n-tensors that are�not simple are
often called compound alternating
�n n-tensors. Simple tensors u1 ∧ · · · ∧ un ∈ n (E) are also
called n-vectors and tensors in (E ∗ ) are often called (alternating) n-forms.
�
Given two linear maps f : E → E � and g : E → E � , we can define h : E × E → 2 (E � ) by
h(u, v) = f (u) ∧ g(v).
It is immediately verified that h is alternating bilinear, and thus, it induces a unique linear
map
2
� �2
f ∧ g: (E) → (E � ),
such that
(f ∧ g)(u ∧ v) = f (u) ∧ g(u).

If we also have linear maps f � : E � → E �� and g � : E � → E �� , we can easily verify that

(f � ◦ f ) ∧ (g � ◦ g) = (f � ∧ g � ) ◦ (f ∧ g).

The generalization to the alternating product f1 ∧· · ·∧fn of n ≥ 3 linear maps fi : E → E �

is immediate, and left to the reader.
618 CHAPTER 22. TENSOR ALGEBRAS

22.12 Bases of Exterior Powers

Let E be any vector space. For any basis, (ui )i∈Σ , for E, we assume that some total ordering,
≤, on Σ, has been chosen. Call the pair ((ui )i∈Σ , ≤) an ordered basis. Then, for any nonempty
finite subset, I ⊆ Σ, let
u I = ui 1 ∧ · · · ∧ u i m ,
where I = {i1 , . . . , im }, with i1 < · · · < im .
�
Since n (E) is generated by the tensors of the form v1 ∧ · · · ∧ vn , with
�n vi ∈ E, in view of
skew-symmetry, it is clear that the tensors uI , with |I| = n, generate (E). Actually, they
form a basis.

Proposition 22.19 Given any vector space, � E, if E has finite �dimension, d = dim(E),
then for all n > d, the exterior power n (E) is trivial, that is� n (E) = (0). Otherwise,
for every ordered basis, ((ui )i∈Σ , ≤), the family, (uI ), is basis of n (E), where I ranges over
finite nonempty subsets of Σ of size |I| = n.

Proof�. First, assume that E has finite dimension, d = dim(E) and that n > d. We know
that n (E) is generated by the tensors of the form v1 ∧ · · · ∧ vn , with vi ∈ E. If u1 , . . . , ud
is a basis of E, as every vi is a linear combination of the uj , when we expand v1 ∧ · · · ∧ vn
using multilinearity, we get a linear combination of the form
�
v1 ∧ · · · ∧ vn = λ(j1 ,...,jn ) uj1 ∧ · · · ∧ ujn ,
(j1 ,...,jn )

where each (j1 , . . . , jn ) is some sequence of integers jk ∈ {1, . . . , d}. As n > d, each sequence
(j1 , . . . , jn ) must contain two identical
� elements. By alternation, uj1 ∧ · · · ∧ ujn = 0 and so,
v1 ∧ · · · ∧ vn = 0. It follows that n (E) = (0).
Now, assume that either dim(E) = d and that n ≤ d or that E is infinite dimensional.
The argument below shows that the uI are nonzero and linearly independent. As usual, let
u∗i ∈ E ∗ be the linear form given by

u∗i (uj ) = δij .

For any nonempty subset, I = {i1 , . . . , in } ⊆ Σ, with i1 < · · · < in , let lI be the map given
by
lI (v1 , . . . , vn ) = det(u∗ij (vk )),
�
for all vk ∈ E. As lI is alternating multilinear, it induces a linear map, LI : n (E) → K.
Observe that for any nonempty finite subset, J ⊆ Σ, with |J| = n, we have
�
1 if I = J
LI (uJ ) =
0 if I �= J.
22.12. BASES OF EXTERIOR POWERS 619

Note that when dim(E) = d and n ≤ d, the forms u∗i1 , . . . , u∗in are all distinct so, the above
does hold. Since LI (uI ) = 1, we conclude that uI �= 0. Now, if we have a linear combination,
�
λI uI = 0,
I

where the above sum is finite and involves nonempty finite subset, I ⊆ Σ, with |I| = n, for
every such I, when we apply LI we get
λI = 0,
proving linear independence.
As a corollary, if E is finite dimensional, say dim(E) = d and if 1 ≤ n ≤ d, then we have
�n � �
n
dim( (E)) =
d
�n
and if n > d, then dim( (E)) = 0.
�
Remark: When n = 0, if we set u∅ = 1, then (u∅ ) = (1) is a basis of 0 (V ) = K.
It follows from Proposition
� 22.19 �nthe family, (uI )I , where I ⊆ Σ ranges over finite
� that
subsets of Σ is a basis of (V ) = n≥0 (V ).
As a corollary of Proposition 22.19 we obtain the following useful criterion for linear
independence:
Proposition 22.20 For any vector space, E, the vectors, u1 , . . . , un ∈ E, are linearly inde-
pendent iff u1 ∧ · · · ∧ un �= 0.
Proof . If u1 ∧ · · · ∧ un �= 0, then u1 , . . . , un must be linearly independent. Otherwise, some
ui would be a linear combination of the other uj ’s (with j �= i) and then, as in the proof
of Proposition 22.19, u1 ∧ · · · ∧ un would be a linear combination of wedges in which two
vectors are identical and thus, zero.
Conversely, assume that u1 , . . . , un are linearly independent. Then, we have the linear
forms, u∗i ∈ E ∗ , such that
u∗i (uj ) = δi,j 1 ≤ i, j ≤ n.
�
As in the proof of Proposition 22.19, we have a linear map, Lu1 ,...,un : n (E) → K, given by
Lu1 ,...,un (v1 ∧ · · · ∧ vn ) = det(u∗j (vi )),
�n
for all v1 ∧ · · · ∧ vn ∈ (E). As,
Lu1 ,...,un (u1 ∧ · · · ∧ un ) = 1,
we conclude that u1 ∧ · · · ∧ un �= 0.
Proposition 22.20 shows that, geometrically, every nonzero wedge, u1 ∧ · · · ∧ un , corre-
sponds to some oriented version of an n-dimensional subspace of E.
620 CHAPTER 22. TENSOR ALGEBRAS

22.13 Some Useful Isomorphisms for Exterior Powers

We can show the following property of the exterior tensor product, using the proof technique
of Proposition 22.7:
n
� �n � k n−k
�
(E ⊕ F ) =∼ (E) ⊗ (F ).
k=0

22.14 Duality for Exterior Powers

In this section, all�vector spaces
�n are assumed to have finite dimension. We define a nonde-
n ∗
generate pairing, (E ) × (E) −→ K, as follows: Consider the multilinear map,

(E ∗ )n × E n −→ K,

given by
�
(v1∗ , . . . , vn∗ , u1 , . . . , un ) �→ ∗
sgn(σ) vσ(1) ∗
(u1 ) · · · vσ(n) (un ) = det(vj∗ (ui )).
σ∈Sn

It is easily checked that this expression is alternating w.r.t. the ui ’s and also w.r.t. the vj∗ .
For any fixed (v1∗ , . . . , vn∗ ) ∈ (E ∗ )n , we get an alternating multinear map,

lv1∗ ,...,vn∗ : (u1 , . . . , un ) �→ det(vj∗ (ui )),

from E n to K. By the argument used in the symmetric case, we get a bilinear map,
n
� n
�
∗
(E ) × (E) −→ K.

Now, this pairing in nondegenerate. This can be done using bases and we leave it as an
exercise to the reader. Therefore, we get a canonical isomorphism,
n
� n
�
( (E))∗ ∼
= (E ∗ ).

Since we also have a canonical isomorphism

n
�
( (E))∗ ∼
= Altn (E; K),

we get a canonical isomorphism

n
�
(E ∗ ) ∼
= Altn (E; K)

which allows us to interpret alternating tensors over E ∗ as alternating multilinear maps.

22.14. DUALITY FOR EXTERIOR POWERS 621
�
The isomorphism, µ : n (E ∗ ) ∼
= Altn (E; K), discussed above can be described explicity
as the linear extension of the map given by

µ(v1∗ ∧ · · · ∧ vn∗ )(u1 , . . . , un ) = det(vj∗ (ui )).

Remark: Variants of our isomorphism, µ, are found in the literature. For example, there
is a version, µ� , where
1
µ� = µ,
n!
1
with the factor n! added in front of the determinant. Each version has its its own merits
and inconvenients. Morita [114] uses µ� because it is more convenient than µ when dealing
with characteristic classes. On the other hand, when using µ� , some extra factor is needed
in defining the wedge operation of alternating multilinear forms (see Section 22.15) and for
exterior differentiation. The version µ is the one adopted by Warner [147], Knapp [89],
Fulton and Harris [57] and Cartan [29, 30].
If f : E → F is any linear map, by transposition we get a linear map, f � : F ∗ → E ∗ ,
given by
f � (v ∗ ) = v ∗ ◦ f, v∗ ∈ F ∗.
Consequently, we have

f � (v ∗ )(u) = v ∗ (f (u)), for all u ∈ E and all v ∗ ∈ F ∗ .

For any p ≥ 1, the map,

(u1 , . . . , up ) �→ f (u1 ) ∧ · · · ∧ f (up ),

� � � �
from E n to p F is multilinear alternating, so it induces a linear map, p f : p E → p F ,
defined on generators by
��
p �
f (u1 ∧ · · · ∧ up ) = f (u1 ) ∧ · · · ∧ f (up ).
�p �p �p �p
Combining and duality, we get a linear map, f� : F∗ → E ∗ , defined on gener-
ators by
��
p �
f � (v1∗ ∧ · · · ∧ vp∗ ) = f � (v1∗ ) ∧ · · · ∧ f � (vp∗ ).

Proposition 22.21 If f : E → F is any linear map between two finite-dimensional vector

spaces, E and F , then
���
p � � p
�
µ f � (ω) (u1 , . . . , up ) = µ(ω)(f (u1 ), . . . , f (up )), ω∈ F ∗ , u1 , . . . , up ∈ E.
622 CHAPTER 22. TENSOR ALGEBRAS

Proof . It is enough to prove the formula on generators. By definition of µ, we have

���
p � �
µ f � (v1∗ ∧ · · · ∧ vp∗ ) (u1 , . . . , up ) = µ(f � (v1∗ ) ∧ · · · ∧ f � (vp∗ ))(u1 , . . . , up )
= det(f � (vj∗ )(ui ))
= det(vj∗ (f (ui )))
= µ(v1∗ ∧ · · · ∧ vp∗ )(f (u1 ), . . . , f (up )),

as claimed.
�
The map p f � is often denoted f ∗ , although this is an ambiguous notation�since p is
dropped. Proposition 22.21 gives us the behavior of f ∗ under the identification of p E ∗ and
Altp (E; K) via the isomorphism µ.
�
As in the case of symmetric powers, the�map from E n to n (E) given by (u1 , . . . , un ) �→
u1 ∧ · · · ∧ un yields a surjection,
�n π : E ⊗n → n (E). Now, this map has some section so there
is some injection, ι : (E) → E ⊗n , with π ◦ ι = id. If our field, K, has characteristic 0,
then there is a special section having a natural definition involving an antisymmetrization
process.
Recall that we have a left action of the symmetric group, Sn , on E ⊗n . The tensors,
z ∈ E ⊗n , such that
σ · z = sgn(σ) z, for all σ ∈ Sn
are called antisymmetrized tensors. We define the map, ι : E n → E ⊗n , by

1 �
ι(u1 , . . . , un ) = sgn(σ) uσ(1) ⊗ · · · ⊗ uσ(n) .
n! σ∈S
n

�n
As the right
�nhand side is clearly an alternating map, we get a linear map, ι : (E) → E ⊗n .
Clearly, ι( (E)) is the set of antisymmetrized tensors in E ⊗n . If we consider the map,
A = ι ◦ π : E ⊗n −→ E ⊗n , it is easy to check that A ◦ A = A. Therefore, A is a projection
and by linear algebra, we know that
n
�
⊗n ⊗n
E = A(E ) ⊕ Ker A = ι( (A)) ⊕ Ker A.

It turns out that Ker A = E ⊗n ∩ Ia = Ker π, where Ia is the two-sided ideal of T (E)
generated by all tensors of the form u ⊗ u ∈ E ⊗2 (for example, see Knapp [89], Appendix
A). Therefore, ι is injective,
n
� n
�
⊗n ⊗n
E = ι( (E)) ⊕ E ∩ I = ι( (E)) ⊕ Ker π,
�n
and the exterior tensor power, (E), is naturally embedded into E ⊗n .
22.15. EXTERIOR ALGEBRAS 623

22.15 Exterior Algebras

�n
As in the case of symmetric tensors, we can pack together all the exterior powers, (V ),
into an algebra,
� �� m
(V ) = (V ),
m≥0

called the exterior algebra (or Grassmann algebra) of V . We mimic the procedure used
for symmetric powers. If Ia is the two-sided ideal generated by all tensors of the form
u ⊗ u ∈ V ⊗2 , we set
�•
(V ) = T (V )/Ia .
�
Then, • (V ) automatically inherits a multiplication operation, called wedge product, and
since T (V ) is graded, that is, �
T (V ) = V ⊗m ,
m≥0

we have •
� �
(V ) = V ⊗m /(Ia ∩ V ⊗m ).
m≥0

However, it is easy to check that

m
�
(V ) ∼
= V ⊗m /(Ia ∩ V ⊗m ),

so •
� �
(V ) ∼
= (V ).
When V has finite dimension, d, we actually have a finite coproduct

� d �
� m
(V ) = (V ),
m=0
�m �d�
and since each (V ) has dimension, m , we deduce that
�
dim( (V )) = 2d = 2dim(V ) .

�m �n �m+n
The multiplication, ∧ : (V ) × (V ) → (V ), is skew-symmetric in the following
precise sense:
�m �n
Proposition 22.22 For all α ∈ (V ) and all β ∈ (V ), we have

β ∧ α = (−1)mn α ∧ β.
624 CHAPTER 22. TENSOR ALGEBRAS

Proof . Since v ∧ u = −u ∧ v for all u, v ∈ V , Proposition 22.22 follows by induction.

Since α ∧ α = 0 for every simple
� tensor, α = u1 ∧ · · · ∧ un , it seems natural to infer that
α ∧ α = 0 for every tensor α ∈ (V ). If we consider the case where dim(V ) ≤ 3, we can
indeed prove the above assertion. However, if dim(V ) ≥ 4, the above fact is generally false!
For example, when dim(V ) = 4, if u1 , u2 , u3 , u4 are a basis for V , for α = u1 ∧ u2 + u3 ∧ u4 ,
we check that
α ∧ α = 2u1 ∧ u2 ∧ u3 ∧ u4 ,
which is nonzero.
The above discussion suggests that it might be useful to know when an� alternating tensor
is simple, that is, decomposable. It can be shown that for tensors, α ∈ 2 (V ), α ∧ α = 0 iff
α is simple. A general criterion for decomposability can be given in terms of some operations
known as left hook and right hook (also called interior products), see Section 22.17.
�
It is easy to see that (V ) satisfies the following universal mapping property:
Proposition 22.23 Given any K-algebra, A, for any linear map, f � : V → A, if (f (v))2 = 0
for all v ∈ V , then there is a unique K-algebra homomorphism, f : (V ) → A, so that
f = f ◦ i,
as in the diagram below:
i �
V � (V )
f
f
� �
A
�n
When E is finite-dimensional, recall the isomorphism, µ : (E ∗ ) −→ Altn (E; K), de-
fined as the linear extension of the map given by
µ(v1∗ ∧ · · · ∧ vn∗ )(u1 , . . . , un ) = det(u∗j (ui )).
� � �
Now, we have also a multiplication operation, m (E ∗ ) × n (E ∗ ) −→ m+n (E ∗ ). The
following question then arises:
Can we define a multiplication, Altm (E; K) × Altn (E; K) −→ Altm+n (E; K), directly on
alternating multilinear forms, so that the following diagram commutes:
�m ∗ � ∧ �m+n ∗
(E ) × n (E ∗ ) � (E )
µ×µ µ
� �
∧
Altm (E; K) × Altn (E; K) � Altm+n (E; K).
As in the symmetric case, the answer is yes! The solution is to define this multiplication
such that, for f ∈ Altm (E; K) and g ∈ Altn (E; K),
�
(f ∧ g)(u1 , . . . , um+n ) = sgn(σ) f (uσ(1) , . . . , uσ(m) )g(uσ(m+1) , . . . , uσ(m+n) ),
σ∈shuffle(m,n)
22.15. EXTERIOR ALGEBRAS 625

where shuffle(m, n) consists of all (m, n)-“shuffles”, that is, permutations, σ, of {1, . . . m+n},
such that σ(1) < · · · < σ(m) and σ(m+1) < · · · < σ(m+n). For example, when m = n = 1,
we have
(f ∧ g)(u, v) = f (u)g(v) − g(u)f (v).
When m = 1 and n ≥ 2, check that
m+1
�
(f ∧ g)(u1 , . . . , um+1 ) = (−1)i−1 f (ui )g(u1 , . . . , u�i , . . . , um+1 ),
i=1

where the hat over the argument ui means that it should be omitted.
As a result of all this, the coproduct
�
Alt(E) = Altn (E; K)
n≥0

�
is an algebra under the above multiplication and this algebra is isomorphic to (E ∗ ). For
the record, we state
�n
Proposition 22.24 When E is finite dimensional, the maps, µ : (E ∗ ) −→ Altn (E; K),
induced by the linear extensions of the maps given by

µ(v1∗ ∧ · · · ∧ vn∗ )(u1 , . . . , un ) = det(u∗j (ui ))

�
yield a canonical isomorphism of algebras, µ : (E ∗ ) −→ Alt(E), where the multiplication
in Alt(E) is defined by the maps, ∧ : Altm (E; K) × Altn (E; K) −→ Altm+n (E; K), with
�
(f ∧ g)(u1 , . . . , um+n ) = sgn(σ) f (uσ(1) , . . . , uσ(m) )g(uσ(m+1) , . . . , uσ(m+n) ),
σ∈shuffle(m,n)

where shuffle(m, n) consists of all (m, n)-“shuffles”, that is, permutations, σ, of {1, . . . m+n},
such that σ(1) < · · · < σ(m) and σ(m + 1) < · · · < σ(m + n).

�
Remark: The algebra, (E) is a graded algebra. Given two graded algebras, E and F , we
� F , where E ⊗
can make a new tensor product, E ⊗ � F is equal to E ⊗ F as a vector space,
but with a skew-commutative multiplication given by

(a ⊗ b) ∧ (c ⊗ d) = (−1)deg(b)deg(c) (ac) ⊗ (bd),

where a ∈ E m , b ∈ F p , c ∈ E n , d ∈ F q . Then, it can be shown that

� � �
(E ⊕ F ) ∼= (E) ⊗ � (F ).
626 CHAPTER 22. TENSOR ALGEBRAS

22.16 The Hodge ∗-Operator

In order to define a generalization of the Laplacian that will apply to differential forms on a
Riemannian manifold, we need to define isomorphisms,
k
� n−k
�
V −→ V,

for any Euclidean vector space, V , of dimension n and any k, with �k 0 ≤ k ≤ n. If �−, −�
denotes the inner product on V , we define an inner product on V , also denoted �−, −�,
by setting
�u1 ∧ · · · ∧ uk , v1 ∧ · · · ∧ vk � = det(�ui , vj �),
for all ui , vi ∈ V and extending �−, −� by bilinearity.
�
It is easy to show that if (e1 , . . . , en ) is an orthonormal basis of V , then the basis of k V
consisting �kof the eI (where I = {i1 , . . . , ik }, with 1 ≤ i1 < · · · < ik ≤ n) is an∗orthonormal
basis of V . Since the inner product on V induces an inner product � on V (recall that
�ω1 , ω2 � = �ω1 , ω2 �, for all ω1 , ω2 ∈ V ), we also get an inner product on k V ∗ .
� � ∗

Recall that an orientation of a vector space, V , of dimension n is given by the choice

of some basis, (e1 , . . . , en ). We say that a basis, (u1 , . . . , un ), of V is positively oriented iff
det(u1 , . . . , un ) > 0 (where det(u1 , . . . , un ) denotes the determinant of the matrix whose jth
column consists of the coordinates of uj over the basis (e1 , . . . , en )), otherwise it is negatively
oriented . An oriented vector space is a vector space, V , together with an orientation of V .
If V is oriented by the basis (e1 , . . . , en ), then V ∗ is oriented by the dual basis, (e∗1 , . . . , e∗n ).
If V is an oriented vector space of dimension n, then we can define a linear map,
k
� n−k
�
∗: V → V,

called the Hodge ∗-operator , as follows: For any choice of a positively oriented orthonormal
basis, (e1 , . . . , en ), of V , set

∗(e1 ∧ · · · ∧ ek ) = ek+1 ∧ · · · ∧ en .

In particular, for k = 0 and k = n, we have

∗(1) = e1 ∧ · · · ∧ en
∗(e1 ∧ · · · ∧ en ) = 1.

It is easy to see that the definition of ∗ does not depend on the choice of positively oriented
orthonormal basis.
�k �
The Hodge
� � ∗-operators, ∗ : V → n−k V , induces � a linear bijection,
�
∗ : (V ) → (V ). We also have Hodge ∗-operators, ∗ : k V ∗ → n−k V ∗ .
The following proposition is easy to show:
22.17. TESTING DECOMPOSABILITY; LEFT AND RIGHT HOOKS 627

Proposition 22.25 If V is any oriented vector space of dimension n, for every k, with
0 ≤ k ≤ n, we have

(i) ∗∗ = (−id)k(n−k) .
�k
(ii) �x, y� = ∗(x ∧ ∗y) = ∗(y ∧ ∗x), for all x, y ∈ V.

If (e1 , . . . , en ) is an orthonormal basis of V and (v1 , . . . , vn ) is any other basis of V , it is

easy to see that �
v1 ∧ · · · ∧ vn = det(�vi , vj �) e1 ∧ · · · ∧ en ,
from which it follows that
1
∗(1) = � v1 ∧ · · · ∧ vn
det(�vi , vj �)

(see Jost [83], Chapter 2, Lemma 2.1.3).

22.17 Testing Decomposability; Left and Right Hooks

In this section, all vector spaces are assumed to have finite dimension. Say dim(E) = n.
Using our nonsingular pairing,
p p
� �
∗
�−, −� : E × E −→ K (1 ≤ p ≤ n),

defined on generators by

�u∗1 ∧ · · · ∧ u∗p , v1 ∧ · · · ∧ up � = det(u∗i (vj )),

we define various contraction operations,

p p+q q
� � �
�: E× ∗
E −→ E∗ (left hook)

and
p+q p q
� � �
�: ∗
E × E −→ E∗ (right hook),
as well as the versions obtained by replacing E by E ∗ and E ∗∗ by E. We begin with the left
interior product or left hook, �.
�
Let u ∈ p E. For any q such that p + q ≤ n, multiplication on the right by u is a linear
map
q p+q
� �
∧R (u) : E −→ E,
628 CHAPTER 22. TENSOR ALGEBRAS

given by
v �→ v ∧ u
�q
where v ∈ E. The transpose of ∧R (u) yields a linear map,
p+q q
� �
∗
t
(∧R (u)) : ( E) −→ ( E)∗ ,
� �p+q ∗ � �q ∗
which, using the isomorphisms ( p+q E)∗ ∼ = E and ( q E)∗ ∼
= E can be viewed as
a map
p+q q
� �
(∧R (u))t : E ∗ −→ E ∗,
given by
z ∗ �→ z ∗ ◦ ∧R (u),
�p+q
where z ∗ ∈ E ∗.
We denote z ∗ ◦ ∧R (u) by
u � z∗.
In terms of our pairing, the q-vector u � z ∗ is uniquely defined by
� � �
�u � z ∗ , v� = �z ∗ , v ∧ u�, for all u ∈ p E, v ∈ q E and z ∗ ∈ p+q E ∗ .

It is immediately verified that

(u ∧ v) � z ∗ = u � (v � z ∗ ),

so � defines a left action

p p+q q
� � �
�: E× E ∗ −→ E ∗.
By interchanging E and E ∗ and using the isomorphism,
k
� k
�
( F )∗ ∼
= F ∗,

we can also define a left action

p p+q q
� � �
�: ∗
E × E −→ E.

In terms of our pairing, u∗ � z is uniquely defined by

� � �
�v ∗ , u∗ � z� = �v ∗ ∧ u∗ , z�, for all u∗ ∈ p E ∗ , v ∗ ∈ q E ∗ and z ∈ p+q E.

�p In order to proceed any further, we need some combinatorial properties of the basis of
E constructed from a basis, (e1 , . . . , en ), of E. Recall that for any (nonempty) subset,
I ⊆ {1, . . . , n}, we let
e I = e i1 ∧ · · · ∧ e ip ,
22.17. TESTING DECOMPOSABILITY; LEFT AND RIGHT HOOKS 629

where I = {i1 , . . . , ip } with i1 < · · · < ip . We also let e∅ = 1.

Given any two subsets H, L ⊆ {1, . . . , n}, let
�
0 if H ∩ L �= ∅,
ρH,L =
(−1)ν if H ∩ L = ∅,

where
ν = |{(h, l) | (h, l) ∈ H × L, h > l}|.

Proposition 22.26 For any basis, (e1 , . . . , en ), of E the following properties hold:

(1) If H ∩ L = ∅, |H| = h, and |L| = l, then

ρH,L ρL,H = (−1)hl .

(2) For H, L ⊆ {1, . . . , m}, we have

eH ∧ eL = ρH,L eH∪L .

(3) For the left hook,

p p+q q
� � �
�: E× ∗
E −→ E ∗,

we have

eH � e∗L = 0 if H �⊆ L
eH � e∗L = ρL−H,H e∗L−H if H ⊆ L.

�p �p+q �q
Similar formulae hold for � : E∗ × E −→ E. Using Proposition 22.26, we
have the

Proposition 22.27 For the left hook,

p p+q q
� � �
�: E× ∗
E −→ E ∗,

for every u ∈ E, we have

u � (x∗ ∧ y ∗ ) = (−1)s (u � x∗ ) ∧ y ∗ + x∗ ∧ (u � y ∗ ),
�s
where y ∈ E ∗.
630 CHAPTER 22. TENSOR ALGEBRAS

Proof . We can prove the above identity assuming that x∗ and y ∗ are of the form e∗I and e∗J us-
ing Proposition 22.26 but this is rather tedious. There is also a proof involving determinants,
see Warner [147], Chapter 2.
Thus, � is almost an anti-derivation, except that the sign, (−1)s is applied to the wrong
factor.
It is also possible to define a right interior product or right hook , �, using multiplication
on the left rather than multiplication on the right. Then, � defines a right action,
p+q p q
� � �
�: ∗
E × E −→ E ∗,

such that
�p �q �p+q
�z ∗ , u ∧ v� = �z ∗ � u, v�, for all u ∈ E, v ∈ E, and z ∗ ∈ E ∗.

Similarly, we have the right action

p+q p q
� � �
�: E× E ∗ −→ E,

such that
�p �q �p+q
�u∗ ∧ v ∗ , z� = �v ∗ , z � u∗ �, for all u∗ ∈ E ∗, v∗ ∈ E ∗ , and z ∈ E.
�p �p+q �q
Since the left hook, � : E× E ∗ −→ E ∗ , is defined by
�p �q �p+q
�u � z ∗ , v� = �z ∗ , v ∧ u�, for all u ∈ E, v ∈ E and z ∗ ∈ E ∗,

the right hook,

p+q p q
� � �
�: E∗ × E −→ E ∗,

by
�p �q �p+q
�z ∗ � u, v� = �z ∗ , u ∧ v�, for all u ∈ E, v ∈ E, and z ∗ ∈ E ∗,
and v ∧ u = (−1)pq u ∧ v, we conclude that

u � z ∗ = (−1)pq z ∗ � u,
�p �p+q
where u ∈ E and z ∈ E ∗.
Using the above property and Proposition 22.27 we get the following version of Proposi-
tion 22.27 for the right hook:
 22.17. TESTING DECOMPOSABILITY; LEFT AND RIGHT HOOKS 631

Proposition 22.28 For the right hook,

p+q p q
� � �
�: E∗ × E −→ E ∗,

for every u ∈ E, we have

(x∗ ∧ y ∗ ) � u = (x∗ � u) ∧ y ∗ + (−1)r x∗ ∧ (y ∗ � u),

�r
where x∗ ∈ E ∗.

Thus, � is an anti-derivation.
For u ∈ E, the right hook, z ∗ � u, is also denoted, i(u)z ∗ , and called insertion operator or
interior
� product. This operator plays an important role in differential geometry. If we view
z ∗ ∈ n+1 (E ∗ ) as an alternating multilinear map in Altn+1 (E; K), then i(u)z ∗ ∈ Altn (E; K)
is given by
(i(u)z ∗ )(v1 , . . . , vn ) = z ∗ (u, v1 , . . . , vn ).

� Note that certain authors, such as Shafarevitch [138], denote our right hook z ∗ � u (which
is also the right hook in Bourbaki [21] and Fulton and Harris [57]) by u � z ∗ .
� �
Using the�two versions of �, we can define linear maps γ : p E → n−p E ∗ and
�
δ : p E ∗ → n−p E. For any basis (e1 , . . . , en ) of E, if we let M = {1, . . . , n}, e = e1 ∧· · ·∧en ,
and e∗ = e∗1 ∧ · · · ∧ e∗n , then

γ(u) = u � e∗ and δ(v) = v ∗ � e,

� �
for all u ∈ p E and all v ∗ ∈ p E ∗ . The following proposition is easily shown.
� � � �
Proposition 22.29 The linear maps γ : p E → n−p E ∗ and δ : p E ∗ → n−p E are
isomorphims. The isomorphisms �p γ and δ map decomposable vectors to decomposable vectors.
�
Furthermore, if z ∈ E is decomposable, then��γ(z), z��= 0, and similarly
� for z ∈� p E ∗ .
If (e�1 , . . . , e�n ) is any other basis of E and γ � : p E → n−p E ∗ and δ � : p E ∗ → n−p E
are the corresponding isomorphisms, then γ � = λγ and δ � = λ−1 δ for some nonzero λ ∈ Ω.

Proof . Using Proposition 22.26, for any subset J ⊆ {1, . . . , n} = M such that |J| = p, we
have
γ(eJ ) = eJ � e∗ = ρM −J,J e∗M −J and δ(e∗J ) = e∗J � e = ρM −J,J eM −J .
Thus,
δ ◦ γ(eJ ) = ρM −J,J ρJ,M −J eJ = (−1)p(n−p) eJ .
A similar result holds for γ ◦ δ. This implies that

δ ◦ γ = (−1)p(n−p) id and γ ◦ δ = (−1)p(n−p) id.

632 CHAPTER 22. TENSOR ALGEBRAS
�
Thus, γ and δ are isomorphisms. If z ∈ p E is decomposable, then z = u1 ∧ · · · ∧ up where
u1 , . . . , up are linearly independent since z �= 0, and we can pick a basis of E of the form
(u1 , . . . , un ). Then, the above formulae show that
γ(z) = ±u∗p+1 ∧ · · · ∧ u∗n .
Clearly
�γ(z), z� = 0.
�
If (e�1 , . . . , e�n ) is any other basis of E, because m E has dimension 1, we have
e�1 ∧ · · · ∧ e�n = λe1 ∧ · · · ∧ en
for some nonnull λ ∈ Ω, and the rest is trivial.
We are now ready to tacke the problem of finding criteria for decomposability. We need
a few preliminary results.
�p
Proposition
� 22.30 Given z ∈ E, with z �= 0, the smallest vector space W ⊆ E such
that z ∈ p W is generated by the vectors of the form
�
u∗ � z, with u∗ ∈ p−1 E ∗ .
�
Proof . First, let W be any subspace such that z ∈ p (E) and let �(e1 ,∗ . . . , er , er+1 , . . . , en ) be
∗
a basis of E such that (e� 1 , . . . , e r ) is a basis of W . Then, u = I eI , where I ⊆ {1, . . . , n}
and |I| = p−1, and z = J eJ , where J ⊆ {1, . . . , r} and |J| = p ≤ r. It follows immediately
from the formula of Proposition 22.26 (3) that u∗ � z ∈ W .
�
Next, we prove that if W is the smallest subspace of E�such that z ∈ p (W ), then W
is generated by the vectors
� of the form u∗ � z, where u∗ ∈ p−1 E ∗ . Suppose not, then the
vectors u∗ � z with u∗ ∈ p−1 E ∗ span a proper subspace, U , of W . We prove that for every
subspace, W ��, of W , with dim(W � ) = dim(W ) − 1 = r − 1, it is not possible that u∗ � z ∈ W �
for all u∗ ∈ p−1 E ∗ . But then, as U is a proper subspace of W , it is contained in some
subspace, W � , with dim(W � ) = r − 1 and we have a contradiction.
Let w ∈ W − W � and pick a basis of � W formed by a�basis (e1 , . . . , er−1 ) of W � and w. We
can write z = z � + w ∧ z �� , where z � ∈ p W � and z �� ∈ p−1 W � , and since � W is the smallest
subspace containing z, we have z �� �= 0. Consequently, if we write z �� = I eI in terms of
the basis (e1 , . . . , er−1 ) of W � , there is some eI , with I ⊆ {1, . . . , r − 1} and |I| = p − 1, so
that the coefficient λI is nonzero. Now, using any basis of E containing (e1 , . . . , er−1 , w), by
Proposition 22.26 (3), we see that
e∗I � (w ∧ eI ) = λw, λ = ±1.
It follows that
e∗I � z = e∗I � (z � + w ∧ z �� ) = e∗I � z � + e∗I � (w ∧ z �� ) = e∗I � z � + λw,
with e∗I � z � ∈ W � , which shows that e∗I�� z ∈
/ W � . Therefore, W is indeed generated by the
vectors of the form u � z, where u ∈
∗ ∗ p−1 ∗
E .
22.17. TESTING DECOMPOSABILITY; LEFT AND RIGHT HOOKS 633
�p
Proposition 22.31 Any nonzero z ∈ E is decomposable iff
�
(u∗ � z) ∧ z = 0, for all u∗ ∈ p−1 E ∗ .
�p
Proof
�p . Clearly, z ∈ E is decomposable iff the smallest vector space, W , such that z ∈
W has dimension p. If dim(W ) = p, we have�z = e1 ∧ · · · ∧ ep where e1 , . . . , ep form a
basis of W . By Proposition 22.30, for every u∗ ∈ p−1 E ∗ , we have u∗ � z ∈ W , so each u∗ � z
is a linear combination of the ei ’s and (u∗ � z) ∧ z = (u∗ � z) ∧ e1 ∧ · · · ∧ ep = 0.
�
Now, assume that (u∗ � z) ∧ z = 0 for all u� ∗
∈ p−1 E ∗ and that dim(W ) = n > p. If
e1 , . . . , en is a basis of W , then we have z = I λI eI , where I ⊆ {1, . . . , n} and |I| = p.
Recall that z �= 0, and so, some λI is nonzero. By Proposition 22.30, each
�p−1 �p−1ei can be written
as u � z for some u ∈
∗ ∗
E and since (u � z) ∧ z = 0 for all u ∈
∗ ∗ ∗ ∗
E , we get

ej ∧ z = 0 for j = 1, . . . , n.
�
By wedging z = I λI eI with each ej , as n > p, we deduce λI = 0 for all I, so z = 0, a
contradiction. Therefore, n = p and z is decomposable.
�
In Proposition 22.31, we can let u∗ range over a basis of p−1 E ∗ , and then, the conditions
are
(e∗H � z) ∧ z = 0
�
for all H ⊆ {1, . . . , n}, with |H| = p − 1. Since (e∗H � z) ∧ z ∈ p+1 E, this is equivalent to

e∗J ((e∗H � z) ∧ z) = 0

for all H, J ⊆ {1, . . . , n}, with |H| = p − 1 and |J| = p + 1. Then, for all I, I � ⊆ {1, . . . , n}
with |I| = |I � | = p, we can show that

e∗J ((e∗H � eI ) ∧ eI � ) = 0,

unless there is some i ∈ {1, . . . , n} such that

I − H = {i}, J − I � = {i}.

In this case, � �
e∗J (e∗H � eH∪{i} ) ∧ eJ−{i} = ρ{i},H ρ{i},J−{i} .
If we let
�i,J,H = ρ{i},H ρ{i},J−{i} ,
we have �i,J,H = +1 if the parity of the number of j ∈ J such that j < i is the same as the
parity of the number of h ∈ H such that h < i, and �i,J,H = −1 otherwise.
Finally, we obtain the following criterion in terms of quadratic equations (Plücker’s equa-
tions) for the decomposability of an alternating tensor:
634 CHAPTER 22. TENSOR ALGEBRAS
� �p
Proposition 22.32 (Grassmann-Plücker’s Equations) For z = I λI e I ∈ E, the condi-
tions for z �= 0 to be decomposable are
�
�i,J,H λH∪{i} λJ−{i} = 0,
i∈J−H

for all H, J ⊆ {1, . . . , n} such that |H| = p − 1 and |J| = p + 1.

�Using these criteria, it is a good exercise to prove that if dim(E) = n, then every tensor
in n−1 (E) is decomposable. This can also be shown directly.
It should be noted that the equations given by Proposition 22.32 are not independent.
For example, when dim(E) = n = 4 and p = 2, these equations reduce to the single equation

λ12 λ34 − λ13 λ24 + λ14 λ23 = 0.

When the field, K, is the field of complex numbers, this is the homogeneous equation of a
quadric in CP5 known as the Klein quadric. The points on this quadric are in one-to-one
correspondence with the lines in CP3 .

22.18 Vector-Valued Alternating Forms

In this section, the vector space, E,
�n is ∗assumed to have finite dimension. We know that
there is a canonical isomorphism, ∼ n
(E ) = Alt (E; K), between alternating n-forms and
alternating multilinear maps. As in the case of general tensors, the isomorphisms,
n
�
Alt (E; F ) ∼
n
= Hom( (E), F )
n
� n
�
Hom( (E), F ) ∼
= ( (E))∗ ⊗ F
n
� n
�
( (E)) ∗ ∼
= (E ∗ )

yield a canonical isomorphism

� n
�
�
Altn (E; F ) ∼
= (E ∗ ) ⊗ F.

Note
�n that F may have infinite dimension. This isomorphism allows us to view the tensors in
∗
(E ) × F as vector valued alternating forms, a point of view � that is useful in differential
geometry. If (f1 , . . . , fr ) is a basis of F , every tensor, ω ∈ n (E ∗ ) × F can be written as
some linear combination r
�
ω= αi ⊗ fi ,
i=1
22.18. VECTOR-VALUED ALTERNATING FORMS 635
�n
with αi ∈ (E ∗ ). We also let
�n �
� � � �� �
(E; F ) = (E ∗ ) ⊗F = (E) ⊗ F.
n=0

Given three vector spaces, F, G, H, if we have some bilinear map, Φ : F ⊗ G → H, then

we can define a multiplication operation,
� � �
∧Φ : (E; F ) × (E; G) → (E; H),

as follows: For every pair, (m, n), we define the multiplication,

� m � � n �
�� � �� � �m+n
� �
∗ ∗
∧Φ : (E ) ⊗ F × (E ) ⊗ G −→ (E ∗ ) ⊗ H,

by
(α ⊗ f ) ∧Φ (β ⊗ g) = (α ∧ β) ⊗ Φ(f, g).
As in Section 22.15 (following H. Cartan [30]) we can also define a multiplication,
∧Φ : Altm (E; F ) × Altm (E; G) −→ Altm+n (E; H),
directly on alternating multilinear maps as follows: For f ∈ Altm (E; F ) and g ∈ Altn (E; G),
�
(f ∧Φ g)(u1 , . . . , um+n ) = sgn(σ) Φ(f (uσ(1) , . . . , uσ(m) ), g(uσ(m+1) , . . . , uσ(m+n) )),
σ∈shuffle(m,n)

where shuffle(m, n) consists of all (m, n)-“shuffles”, that is, permutations, σ, of {1, . . . m+n},
such that σ(1) < · · · < σ(m) and σ(m + 1) < · · · < σ(m + n).
In general, not much can be said about ∧Φ unless Φ has some additional properties. In
particular, ∧Φ is generally not associative. We also have the map,
�n �
�
µ: (E ∗ ) ⊗ F −→ Altn (E; F ),

defined on generators by
µ((v1∗ ∧ · · · ∧ vn∗ ) ⊗ a)(u1 , . . . , un ) = (det(vj∗ (ui ))a.
Proposition 22.33 The map
� n
�
�
µ: (E ∗ ) ⊗ F −→ Altn (E; F ),

defined as above is a canonical isomorphism for every n ≥ 0. Furthermore,�given any three

n ∗
�n ∗F, G, H, and any bilinear map, Φ : F × G → H, for all ω ∈ ( (E )) ⊗ F and
vector spaces,
all η ∈ ( (E )) ⊗ G,
µ(α ∧Φ β) = µ(α) ∧Φ µ(β).
636 CHAPTER 22. TENSOR ALGEBRAS
�n ∗
Proof . Since we already know that ( � (E ))⊗F and Altn (E; F ) are isomorphic, it is enough
to show that µ maps some basis of ( n (E ∗ )) ⊗ F to linearly independent elements. Pick
some bases, (e1 , . . . , ep ) in E and (fj )j∈J in F�. Then, we know that the vectors, e∗I ⊗fj , where
I ⊆ {1, . . . , p} and |I| = n form a basis of ( n (E ∗ )) ⊗ F . If we have a linear dependence,
�
λI,j µ(e∗I ⊗ fj ) = 0,
I,j

applying the above combination to each (ei1 , . . . , ein ) (I = {i1 , . . . , in }, i1 < · · · < in ), we
get the linear combination �
λI,j fj = 0,
j

and by linear independence of the fj ’s, we get λI,j = 0, for all I and all j. Therefore, the
µ(e∗I ⊗ fj ) are linearly independent and we are done. The second part of the proposition is
easily checked (a simple computation).
A special case of interest is the case where F = G = H is a Lie algebra and Φ(a, b) �= [a, b],
bracket of F . In this case, using a base, (f1 , . . . , fr ), of F if we write ω = i αi ⊗ fi
is the Lie�
and η = j βj ⊗ fj , we have
�
[ω, η] = αi ∧ βj ⊗ [fi , fj ].
i,j

Consequently,
[η, ω] = (−1)mn+1 [ω, η].

The following proposition will be useful in dealing with vector-valued differential forms:
�
Proposition 22.34 If (e1 , . . . , ep ) is any basis of E, then every element, ω ∈ ( n (E ∗ )) ⊗ F ,
can be written in a unique way as
�
ω= e∗I ⊗ fI , fI ∈ F,
I

where the e∗I are defined as in Section 22.12.

�
Proof . Since,�by Proposition 22.19, the e∗I form a basis of n (E ∗ ), elements of the form
e∗I ⊗ f span ( n (E ∗ )) ⊗ F . Now, if we apply µ(ω) to (ei1 , . . . , ein ), where I = {i1 , . . . , in } ⊆
{1, . . . , p}, we get

µ(ω)(ei1 , . . . , ein ) = µ(e∗I ⊗ fI )(ei1 , . . . , ein ) = fI .

Therefore, the fI are uniquely determined by f .

Proposition can also be formulated in terms of alternating multilinear maps, a fact that
will be useful to deal with differential forms.
22.19. TENSOR PRODUCTS OF MODULES OVER A COMMMUTATIVE RING 637

Define the product, · : Altn (E; R) × F → Altn (E; F ), as follows: For all ω ∈ Altn (E; R)
and all f ∈ F ,
(ω · f )(u1 , . . . , un ) = ω(u1 , . . . , un )f,
�
for all u1 , . . . , un ∈ E. Then, it is immediately verified that for every ω ∈ ( n (E ∗ )) ⊗ F of
the form
ω = u∗1 ∧ · · · ∧ u∗n ⊗ f,
we have
µ(u∗1 ∧ · · · ∧ u∗n ⊗ f ) = µ(u∗1 ∧ · · · ∧ u∗n ) · f.
Then, Proposition 22.34 yields

Proposition 22.35 If (e1 , . . . , ep ) is any basis of E, then every element, ω ∈ Altn (E; F ),
can be written in a unique way as
�
ω= e∗I · fI , fI ∈ F,
I

where the e∗I are defined as in Section 22.12.

22.19 Tensor Products of Modules over a

Commmutative Ring
If R is a commutative ring with identity (say 1), recall that a module over R (or R-module)
is an abelian group, M , with a scalar multiplication, · : R × M → M , and all the axioms of
a vector space are satisfied.
At first glance, a module does not seem any different from a vector space but the lack of
multiplicative inverses in R has drastic consequences, one being that unlike vector spaces,
modules are generally not free, that is, have no bases. Furthermore, a module may have
torsion elements, that is, elements, m ∈ M , such that λ · m = 0, even though m �= 0 and
λ �= 0.
Nevertheless, it is possible to define tensor products of modules over a ring, just as in
Section 22.1 and the results of this section continue to hold. The results of Section 22.3
also continue to hold since they are based on the universal mapping property. However, the
results of Section 22.2 on bases generally fail, except for free modules. Similarly, the results
of Section 22.4 on duality generally fail. Tensor algebras can be defined for modules, as
in Section 22.5. Symmetric tensor and alternating tensors can be defined for modules but
again, results involving bases generally fail.
Tensor products of modules have some unexpected properties. For example, if p and q
are relatively prime integers, then

Z/pZ ⊗Z Z/qZ = (0).

638 CHAPTER 22. TENSOR ALGEBRAS

It is possible to salvage certain properties of tensor products holding for vector spaces by
restricting the class of modules under consideration. For example, projective modules, have
a pretty good behavior w.r.t. tensor products.
A free R-module, F , is a module that has a basis (i.e., there is a family, (ei )i∈I , of
linearly independent vectors in F that span F ). Projective modules have many equivalent
characterizations. Here is one that is best suited for our needs:

Definition 22.9 An R-module, P , is projective if it is a summand of a free module, that

is, if there is a free R-module, F , and some R-module, Q, so that

F = P ⊕ Q.

Given any R-module, M , we let M ∗ = HomR (M, R) be its dual . We have the following
proposition:

Proposition 22.36 For any finitely-generated projective R-modules, P , and any R-module,
Q, we have the isomorphisms:

P ∗∗ ∼
= P
HomR (P, Q) ∼
= P ∗ ⊗R Q.

Sketch of proof . We only consider the second isomorphism. Since P is projective, we have
some R-modules, P1 , F , with
P ⊕ P1 = F,
where F is some free module. Now, we know that for any R-modules, U, V, W , we have
�
HomR (U ⊕ V, W ) ∼= HomR (U, W ) HomR (V, W ) ∼
= HomR (U, W ) ⊕ HomR (V, W ),

so
P ∗ ⊕ P1∗ ∼
= F ∗, HomR (P, Q) ⊕ HomR (P1 , Q) ∼
= HomR (F, Q).
By tensoring with Q and using the fact that tensor distributes w.r.t. coproducts, we get

(P ∗ ⊗R Q) ⊕ (P1∗ ⊗ Q) ∼
= (P ∗ ⊕ P1∗ ) ⊗R Q ∼
= F ∗ ⊗R Q.

Now, the proof of Proposition 22.9 goes through because F is free and finitely generated, so

α⊗ : (P ∗ ⊗R Q) ⊕ (P1∗ ⊗ Q) ∼
= F ∗ ⊗R Q −→ HomR (F, Q) ∼
= HomR (P, Q) ⊕ HomR (P1 , Q)

is an isomorphism and as αα maps P ∗ ⊗R Q to HomR (P, Q), it yields an isomorphism between

these two spaces.
The isomorphism α⊗ : P ∗ ⊗R Q ∼
= HomR (P, Q) of Proposition 22.36 is still given by

α⊗ (u∗ ⊗ f )(x) = u∗ (x)f, u∗ ∈ P ∗ , f ∈ Q, x ∈ P.

22.20. THE PFAFFIAN POLYNOMIAL 639

It is convenient to introduce the evaluation map, Evx : P ∗ ⊗R Q → Q, defined for every

x ∈ P by
Evx (u∗ ⊗ f ) = u∗ (x)f, u∗ ∈ P ∗ , f ∈ Q.

In Section 11.2 we will need to consider a slightly weaker version of the universal mapping
property of tensor products. The situation is this: We have a commutative R-algebra, S,
where R is a field (or even a commutative ring), we have two R-modules, U and V , and
moreover, U is a right S-module and V is a left S-module. In Section 11.2, this corresponds
to R = R, S = C ∞ (B), U = Ai (ξ) and V = Γ(ξ), where ξ is a vector bundle. Then, we can
form the tensor product, U ⊗R V , and we let U ⊗S V be the quotient module, (U ⊗R V )/W ,
where W is the submodule of U ⊗R V generated by the elements of the form

us ⊗R v − u ⊗R sv.

As S is commutative, we can make U ⊗S V into an S-module by defining the action of S via

s(u ⊗S v) = us ⊗S v.

It is immediately verified that this S-module is isomorphic to the tensor product of U and
V as S-modules and the following universal mapping property holds:

Proposition 22.37 For every, R-bilinear map, f : U × V → Z, if f satisfies the property

f (us, v) = f (u, sv), for all u ∈ U, v ∈ V, s ∈ S,

then f induces a unique R-linear map, f�: U ⊗S V → Z, such that

f (u, v) = f�(u ⊗S v), for all u ∈ U, v ∈ V.

Note that the linear map, f�: U ⊗S V → Z, is only R-linear, it is not S-linear in general.

22.20 The Pfaffian Polynomial

Let so(2n) denote the vector space (actually, Lie algebra) of 2n × 2n real skew-symmetric
matrices. It is well-known that every matrix, A ∈ so(2n), can be written as

A = P DP � ,

where P is an orthogonal matrix and where D is a block diagonal matrix

 
D1
 D2 
 
D= . . 
 . 
Dn
640 CHAPTER 22. TENSOR ALGEBRAS

consisting of 2 × 2 blocks of the form

� �
0 −ai
Di = .
ai 0

For a proof, see see Horn and Johnson [79], Corollary 2.5.14, Gantmacher [61], Chapter IX,
or Gallier [58], Chapter 11.
Since det(Di ) = a2i and det(A) = det(P DP � ) = det(D) = det(D1 ) · · · det(Dn ), we get

det(A) = (a1 · · · an )2 .

The Pfaffian is a polynomial function, Pf(A), in skew-symmetric 2n × 2n matrices, A, (a

polynomial in (2n − 1)n variables) such that

Pf(A)2 = det(A)

and for every arbitrary matrix, B,

Pf(BAB � ) = Pf(A) det(B).

The Pfaffian shows up in the definition of the Euler class of a vector bundle. There is a
simple way to define the Pfaffian using some exterior algebra. Let (e1 , . . . , e2n ) be any basis
of R2n . For any matrix, A ∈ so(2n), let
�
ω(A) = aij ei ∧ ej ,
i<j
�n
where A = (aij ). Then, ω(A) is of the form Ce1 ∧ e2 ∧ · · · ∧ e2n for some constant, C ∈ R.

Definition 22.10 For every skew symmetric matrix, A ∈ so(2n), the Pfaffian polynomial
or Pfaffian is the degree n polynomial, Pf(A), defined by
n
�
ω(A) = n! Pf(A) e1 ∧ e2 ∧ · · · ∧ e2n .

Clearly, Pf(A) is independent of the basis chosen. If A is the block diagonal matrix D,
a simple calculation shows that

ω(D) = −(a1 e1 ∧ e2 + a2 e3 ∧ e4 + · · · + an e2n−1 ∧ e2n )

and that n
�
ω(D) = (−1)n n! a1 · · · an e1 ∧ e2 ∧ · · · ∧ e2n ,
and so
Pf(D) = (−1)n a1 · · · an .
Since Pf(D)2 = (a1 · · · an )2 = det(A), we seem to be on the right track.
22.20. THE PFAFFIAN POLYNOMIAL 641

Proposition 22.38 For every skew symmetric matrix, A ∈ so(2n) and every arbitrary ma-
trix, B, we have:
(i) Pf(A)2 = det(A)

(ii) Pf(BAB � ) = Pf(A) det(B).

Proof . If we assume that (ii) is proved then, since we can write A = P DP � for some
orthogonal matrix, P , and some block diagonal matrix, D, as above, as det(P ) = ±1 and
Pf(D)2 = det(A), we get

Pf(A)2 = Pf(P DP � )2 = Pf(D)2 det(P )2 = det(A),

which is (i). Therefore, it remains to prove (ii).

�
Let fi = Bei , for i = 1, . . . , 2n, where (e1 , . . . , e2n ) is any basis of R2n . Since fi = k bki ek ,
we have � �� �
τ= aij fi ∧ fj = bki aij blj ek ∧ el = (BAB � )kl ek ∧ el ,
i,j i,j k,l k,l

and so, as BAB � is skew symmetric and ek ∧ el = −el ∧ ek , we get

τ = 2ω(BAB � ).

Consequently,
n
� n
�
τ =2 n
ω(BAB � ) = 2n n! Pf(BAB � ) e1 ∧ e2 ∧ · · · ∧ e2n .

Now,
n
�
τ = C f1 ∧ f2 ∧ · · · ∧ f2n ,
for some C ∈ R. If B is singular, then the fi are linearly dependent which implies that
f1 ∧ f2 ∧ · · · ∧ f2n = 0, in which case,

Pf(BAB � ) = 0,

as e1 ∧ e2 ∧ · · · ∧ e2n �= 0. Therefore, if B is singular, det(B) = 0 and

Pf(BAB � ) = 0 = Pf(A) det(B).

� �
If B is invertible, as τ = i,j aij fi ∧ fj = 2 i<j aij fi ∧ fj , we have
n
�
τ = 2n n! Pf(A) f1 ∧ f2 ∧ · · · ∧ f2n .

However, as fi = Bei , we have

f1 ∧ f2 ∧ · · · ∧ f2n = det(B) e1 ∧ e2 ∧ · · · ∧ e2n ,

 642 CHAPTER 22. TENSOR ALGEBRAS

so n
�
τ = 2n n! Pf(A) det(B) e1 ∧ e2 ∧ · · · ∧ e2n
and as n
�
τ = 2n n! Pf(BAB � ) e1 ∧ e2 ∧ · · · ∧ e2n ,
we get
Pf(BAB � ) = Pf(A) det(B),
as claimed.

Remark: It can be shown that the polynomial, Pf(A), is the unique polynomial with integer
coefficients such that Pf(A)2 = det(A) and Pf(diag(S, . . . , S)) = +1, where
� �
0 1
S= ,
−1 0
see Milnor and Stasheff [110] (Appendix C, Lemma 9). There is also an explicit formula for
Pf(A), namely:
n
1 � �
Pf(A) = n sgn(σ) aσ(2i−1) σ(2i) .

�
2 n! σ∈S i=1
2n

Beware, some authors use a different sign convention and require the Pfaffian to have
the value +1 on the matrix diag(S � , . . . , S � ), where
� �
� 0 −1
S = .
1 0
For example, if R2n is equipped with an inner product, �−, −�, then some authors define
ω(A) as �
ω(A) = �Aei , ej � ei ∧ ej ,
i<j

where A = (aij ). But then, �Aei , ej � = aji and not aij , and this Pfaffian takes the value +1
on the matrix diag(S � , . . . , S � ). This version of the Pfaffian differs from our version by the
factor (−1)n . In this respect, Madsen and Tornehave [100] seem to have an incorrect sign in
Proposition B6 of Appendix C.
We will also need another property of Pfaffians. Recall that the ring, Mn (C), of n × n
matrices over C is embedded in the ring, M2n (R), of 2n × 2n matrices with real coefficients,
using the injective homomorphism that maps every entry z = a + ib ∈ C to the 2 × 2 matrix
� �
a −b
.
b a
If A ∈ Mn (C), let AR ∈ M2n (R) denote the real matrix obtained by the above process.
�
Observe that every skew Hermitian matrix, A ∈ u(n), (i.e., with A∗ = A = −A) yields a
matrix AR ∈ so(2n).
 22.20. THE PFAFFIAN POLYNOMIAL 643

Proposition 22.39 For every skew Hermitian matrix, A ∈ u(n), we have

Pf(AR ) = in det(A).

Proof . It is well-known that a skew Hermitian matrix can be diagonalized with respect to a
unitary matrix, U , and that the eigenvalues are pure imaginary or zero, so we can write

A = U diag(ia1 , . . . , ian )U ∗ ,

for some reals, ai ∈ R. Consequently, we get

AR = UR diag(D1 , . . . , Dn )UR� ,

where � �
0 −ai
Di =
ai 0
and
Pf(AR ) = Pf(diag(D1 , . . . , Dn )) = (−1)n a1 · · · an ,
as we saw before. On the other hand,

det(A) = det(diag(ia1 , . . . , ian )) = in a1 · · · an ,

and as (−1)n = in in , we get

Pf(AR ) = in det(A),
as claimed.

� Madsen and Tornehave [100] state Proposition 22.39 using the factor (−i)n , which is
wrong.
644 CHAPTER 22. TENSOR ALGEBRAS
Bibliography

[1] Ralph Abraham and Jerrold E. Marsden. Foundations of Mechanics. Addison Wesley,
second edition, 1978.

[2] George E. Andrews, Richard Askey, and Ranjan Roy. Special Functions. Cambridge
University Press, first edition, 2000.

[3] Vincent Arsigny. Processing Data in Lie Groups: An Algebraic Approach. Application
to Non-Linear Registration and Diffusion Tensor MRI. PhD thesis, École Polytech-
nique, Palaiseau, France, 2006. Thèse de Sciences.

[4] Vincent Arsigny, Olivier Commowick, Xavier Pennec, and Nicholas Ayache. A fast
and log-euclidean polyaffine framework for locally affine registration. Technical report,
INRIA, 2004, route des Lucioles, 06902 Sophia Antipolis Cedex, France, 2006. Report
No. 5865.

[5] Vincent Arsigny, Pierre Fillard, Xavier Pennec, and Nicholas Ayache. Geometric means
in a novel vector space structure on symmetric positive-definite matrices. SIAM J. on
Matrix Analysis and Applications, 29(1):328–347, 2007.

[6] Vincent Arsigny, Xavier Pennec, and Nicholas Ayache. Polyrigid and polyaffine trans-
formations: a novel geometrical tool to deal with non-rigid deformations–application
to the registration of histological slices. Medical Image Analysis, 9(6):507–523, 2005.

[7] Michael Artin. Algebra. Prentice Hall, first edition, 1991.

[8] Andreas Arvanitoyeogos. An Introduction to Lie Groups and the Geometry of Homo-
geneous Spaces. SML, Vol. 22. AMS, first edition, 2003.

[9] M. F. Atiyah and I. G. Macdonald. Introduction to Commutative Algebra. Addison

Wesley, third edition, 1969.

[10] Michael F. Atiyah. K-Theory. Addison Wesley, first edition, 1988.

[11] Michael F Atiyah, Raoul Bott, and Arnold Shapiro. Clifford modules. Topology, 3,
Suppl. 1:3–38, 1964.

645
646 BIBLIOGRAPHY

[12] Sheldon Axler, Paul Bourdon, and Wade Ramey. Harmonic Function Theory. GTM
No. 137. Springer Verlag, second edition, 2001.

[13] Andrew Baker. Matrix Groups. An Introduction to Lie Group Theory. SUMS. Springer,
2002.

[14] Ronen Basri and David W. Jacobs. Lambertian reflectance and linear subspaces. IEEE
Transactions on Pattern Analysis and Machine Intelligence, 25(2):228–233, 2003.

[15] Marcel Berger. Géométrie 1. Nathan, 1990. English edition: Geometry 1, Universitext,
Springer Verlag.

[16] Marcel Berger. A Panoramic View of Riemannian Geometry. Springer, 2003.

[17] Marcel Berger and Bernard Gostiaux. Géométrie différentielle: variétés, courbes et
surfaces. Collection Mathématiques. Puf, second edition, 1992. English edition: Dif-
ferential geometry, manifolds, curves, and surfaces, GTM No. 115, Springer Verlag.

[18] William M. Boothby. An Introduction to Differentiable Manifolds and Riemannian

Geometry. Academic Press, second edition, 1986.

[19] Raoul Bott and Tu Loring W. Differential Forms in Algebraic Topology. GTM No. 82.
Springer Verlag, first edition, 1986.

[20] Nicolas Bourbaki. Algèbre, Chapitre 9. Eléments de Mathématiques. Hermann, 1968.

[21] Nicolas Bourbaki. Algèbre, Chapitres 1-3. Eléments de Mathématiques. Hermann,

1970.

[22] Nicolas Bourbaki. Elements of Mathematics. Lie Groups and Lie Algebras, Chapters
1–3. Springer, first edition, 1989.

[23] Nicolas Bourbaki. Topologie Générale, Chapitres 1-4. Eléments de Mathématiques.

Masson, 1990.

[24] Nicolas Bourbaki. Topologie Générale, Chapitres 5-10. Eléments de Mathématiques.

CCLS, 1990.

[25] T. Bröcker and T. tom Dieck. Representation of Compact Lie Groups. GTM, Vol. 98.
Springer Verlag, first edition, 1985.

[26] R.L. Bryant. An introduction to Lie groups and symplectic geometry. In D.S. Freed and
K.K. Uhlenbeck, editors, Geometry and Quantum Field Theory, pages 5–181. AMS,
Providence, Rhode Island, 1995.

[27] N. Burgoyne and R. Cushman. Conjugacy classes in linear groups. Journal of Algebra,
44:339–362, 1977.
BIBLIOGRAPHY 647

[28] Élie Cartan. Theory of Spinors. Dover, first edition, 1966.

[29] Henri Cartan. Cours de Calcul Différentiel. Collection Méthodes. Hermann, 1990.

[30] Henri Cartan. Differential Forms. Dover, first edition, 2006.

[31] Roger Carter, Graeme Segal, and Ian Macdonald. Lectures on Lie Groups and Lie
Algebras. Cambridge University Press, first edition, 1995.

[32] Sheung H. Cheng, Nicholas J. Higham, Charles Kenney, and Alan J. Laub. Approx-
imating the logarithm of a matrix to specified accuracy. SIAM Journal on Matrix
Analysis and Applications, 22:1112–1125, 2001.

[33] Shiing-shen Chern. Complex Manifolds without Potential Theory. Universitext.

Springer Verlag, second edition, 1995.

[34] Claude Chevalley. Theory of Lie Groups I. Princeton Mathematical Series, No. 8.
Princeton University Press, first edition, 1946. Eighth printing.

[35] Claude Chevalley. The Algebraic Theory of Spinors and Clifford Algebras. Collected
Works, Vol. 2. Springer, first edition, 1997.

[36] Yvonne Choquet-Bruhat and Cécile DeWitt-Morette. Analysis, Manifolds, and

Physics, Part II: 92 Applications. North-Holland, first edition, 1989.

[37] Yvonne Choquet-Bruhat, Cécile DeWitt-Morette, and Margaret Dillard-Bleick. Anal-

ysis, Manifolds, and Physics, Part I: Basics. North-Holland, first edition, 1982.

[38] Morton L. Curtis. Matrix Groups. Universitext. Springer Verlag, second edition, 1984.

[39] James F. Davis and Paul Kirk. Lecture Notes in Algebraic Topology. GSM, Vol. 35.
AMS, first edition, 2001.

[40] Anton Deitmar. A First Course in Harmonic Analysis. UTM. Springer Verlag, first
edition, 2002.

[41] C. R. DePrima and C. R. Johnson. The range of A−1 A∗ in GL(n, C). Linear Algebra
and Its Applications, 9:209–222, 1974.

[42] Jean Dieudonné. Sur les Groupes Classiques. Hermann, third edition, 1967.

[43] Jean Dieudonné. Special Functions and Linear Representations of Lie Groups. Regional
Conference Series in Mathematics, No. 42. AMS, first edition, 1980.

[44] Jean Dieudonné. Éléments d’Analyse, Tome V. Groupes de Lie Compacts, Groupes de
Lie Semi-Simples. Edition Jacques Gabay, first edition, 2003.
648 BIBLIOGRAPHY

[45] Jean Dieudonné. Éléments d’Analyse, Tome VI. Analyse Harmonique. Edition Jacques
Gabay, first edition, 2003.
[46] Jean Dieudonné. Éléments d’Analyse, Tome VII. Équations Fonctionnelles Linéaires.
Première partie, Opérateurs Pseudo-Différentiels. Edition Jacques Gabay, first edition,
2003.
[47] Jean Dieudonné. Éléments d’Analyse, Tome II. Chapitres XII à XV. Edition Jacques
Gabay, first edition, 2005.
[48] Dragomir Djokovic. On the exponential map in classical lie groups. Journal of Algebra,
64:76–88, 1980.
[49] Manfredo P. do Carmo. Differential Geometry of Curves and Surfaces. Prentice Hall,
1976.
[50] Manfredo P. do Carmo. Riemannian Geometry. Birkhäuser, second edition, 1992.
[51] B.A. Dubrovin, A.T. Fomenko, and S.P. Novikov. Modern Geometry–Methods and
Applications. Part I. GTM No. 93. Springer Verlag, second edition, 1985.
[52] B.A. Dubrovin, A.T. Fomenko, and S.P. Novikov. Modern Geometry–Methods and
Applications. Part II. GTM No. 104. Springer Verlag, first edition, 1985.
[53] J.J. Duistermaat and J.A.C. Kolk. Lie Groups. Universitext. Springer Verlag, first
edition, 2000.
[54] Gerald B. Folland. A Course in Abstract Harmonic Analysis. CRC Press, first edition,
1995.
[55] Joseph Fourier. Théorie Analytique de la Chaleur. Edition Jacques Gabay, first edition,
1822.
[56] William Fulton. Algebraic Topology, A first course. GTM No. 153. Springer Verlag,
first edition, 1995.
[57] William Fulton and Joe Harris. Representation Theory, A first course. GTM No. 129.
Springer Verlag, first edition, 1991.
[58] Jean H. Gallier. Geometric Methods and Applications, For Computer Science and
Engineering. TAM, Vol. 38. Springer, first edition, 2000.
[59] Jean H. Gallier. Logarithms and square roots of real matrices. Technical report,
University of Pennsylvania, Levine Hall, 3330 Walnut Street, Philadelphia, PA 19104,
2008. Report No. MS-CIS-08-12.
[60] S. Gallot, D. Hulin, and J. Lafontaine. Riemannian Geometry. Universitext. Springer
Verlag, second edition, 1993.
BIBLIOGRAPHY 649

[61] F.R. Gantmacher. The Theory of Matrices, Vol. I. AMS Chelsea, first edition, 1977.
[62] Christopher Michael Geyer. Catadioptric Projective Geometry: Theory and Applica-
tions. PhD thesis, University of Pennsylvania, 200 South 33rd Street, Philadelphia,
PA 19104, 2002. Dissertation.
[63] André Gramain. Topologie des Surfaces. Collection Sup. Puf, first edition, 1971.
[64] Robin Green. Spherical harmonic lighting: The gritty details. In Archives of the Game
Developers’ Conference, pages 1–47, 2003.
[65] Marvin J. Greenberg and John R. Harper. Algebraic Topology: A First Course. Addi-
son Wesley, first edition, 1981.
[66] Phillip Griffiths and Joseph Harris. Principles of Algebraic Geometry. Wiley Inter-
science, first edition, 1978.
[67] Cindy M. Grimm. Modeling Surfaces of Arbitrary Topology Using Manifolds. PhD
thesis, Department of Computer Science, Brown University, Providence, Rhode Island,
USA, 1996. Dissertation.
[68] Cindy M. Grimm and John F. Hughes. Modeling surfaces of arbitrary topology using
manifolds. In Proceedings of the 22nd ACM Annual Conference on Computer Graphics
and Interactive Techniques (SIGRAPH’95), pages 359–368. ACM, August 6-11 1995.
[69] Victor Guillemin and Alan Pollack. Differential Topology. Prentice Hall, first edition,
1974.
[70] Brian Hall. Lie Groups, Lie Algebras, and Representations. An Elementary Introduc-
tion. GTM No. 222. Springer Verlag, first edition, 2003.
[71] Allen Hatcher. Algebraic Topology. Cambridge University Press, first edition, 2002.
[72] Sigurdur Helgason. Groups and Geometric Analysis. Integral Geometry, Invariant
Differential Operators and Spherical Functions. MSM, Vol. 83. AMS, first edition,
2000.
[73] Sigurdur Helgason. Differential Geometry, Lie Groups, and Symmetric Spaces. GSM,
Vol. 34. AMS, first edition, 2001.
[74] Nicholas J. Higham. The scaling and squaring method of the matrix exponential
revisited. SIAM Journal on Matrix Analysis and Applications, 26:1179–1193, 2005.
[75] D. Hilbert and S. Cohn-Vossen. Geometry and the Imagination. Chelsea Publishing
Co., 1952.
[76] Morris W. Hirsch. Differential Topology. GTM No. 33. Springer Verlag, first edition,
1976.
650 BIBLIOGRAPHY

[77] Friedrich Hirzebruch. Topological Methods in Algebraic Geometry. Springer Classics

in Mathematics. Springer Verlag, second edition, 1978.

[78] Harry Hochstadt. The Functions of Mathematical Physics. Dover, first edition, 1986.

[79] Roger A. Horn and Charles R. Johnson. Matrix Analysis. Cambridge University Press,
first edition, 1990.

[80] Roger Howe. Very basic Lie theory. American Mathematical Monthly, 90:600–623,
1983.

[81] James E. Humphreys. Introduction to Lie Algebras and Representation Theory. GTM
No. 9. Springer Verlag, first edition, 1972.

[82] Dale Husemoller. Fiber Bundles. GTM No. 20. Springer Verlag, third edition, 1994.

[83] Jürgen Jost. Riemannian Geometry and Geometric Analysis. Universitext. Springer
Verlag, fourth edition, 2005.

[84] Charles S. Kenney and Alan J. Laub. Condition estimates for matrix functions. SIAM
Journal on Matrix Analysis and Applications, 10:191–209, 1989.

[85] A.A. Kirillov. Spinor representations of orthogonal groups. Technical report, University
of Pennsylvania, Math. Department, Philadelphia, PA 19104, 2001. Course Notes for
Math 654.

[86] A.A. Kirillov. Lectures on the Orbit Method. GSM, Vol. 64. AMS, first edition, 2004.

[87] A.A. Kirillov (Ed.). Representation Theory and Noncommutative Harmonic Analysis.
Encyclopaedia of Mathematical Sciences, Vol. 22. Springer Verlag, first edition, 1994.

[88] Wilhelm Klingenberg. Riemannian Geometry. de Gruyter & Co, second edition, 1995.

[89] Anthony W. Knapp. Lie Groups Beyond an Introduction. Progress in Mathematics,

Vol. 140. Birkhäuser, second edition, 2002.

[90] Shoshichi Kobayashi and Katsumi Nomizu. Foundations of Differential Geometry, II.
Wiley Classics. Wiley-Interscience, first edition, 1996.

[91] Wolfgang Kühnel. Differential Geometry. Curves–Surfaces–Manifolds. Student Math-

ematical Library, Vol. 16. AMS, first edition, 2002.

[92] Jacques Lafontaine. Introduction Aux Variétés Différentielles. PUG, first edition, 1996.

[93] Serge Lang. Real and Functional Analysis. GTM 142. Springer Verlag, third edition,
1996.

[94] Serge Lang. Undergraduate Analysis. UTM. Springer Verlag, second edition, 1997.
BIBLIOGRAPHY 651

[95] Serge Lang. Fundamentals of Differential Geometry. GTM No. 191. Springer Verlag,
first edition, 1999.

[96] Blaine H. Lawson and Marie-Louise Michelsohn. Spin Geometry. Princeton Math.
Series, No. 38. Princeton University Press, 1989.

[97] N. N. Lebedev. Special Functions and Their Applications. Dover, first edition, 1972.

[98] John M. Lee. Introduction to Smooth Manifolds. GTM No. 218. Springer Verlag, first
edition, 2006.

[99] Pertti Lounesto. Clifford Algebras and Spinors. LMS No. 286. Cambridge University
Press, second edition, 2001.

[100] Ib Madsen and Jorgen Tornehave. From Calculus to Cohomology. De Rham Cohomol-
ogy and Characteristic Classes. Cambridge University Press, first edition, 1998.

[101] Paul Malliavin. Géométrie Différentielle Intrinsèque. Enseignement des Sciences, No.
14. Hermann, first edition, 1972.

[102] Jerrold E. Marsden and T.S. Ratiu. Introduction to Mechanics and Symmetry. TAM,
Vol. 17. Springer Verlag, first edition, 1994.

[103] William S. Massey. Algebraic Topology: An Introduction. GTM No. 56. Springer
Verlag, second edition, 1987.

[104] William S. Massey. A Basic Course in Algebraic Topology. GTM No. 127. Springer
Verlag, first edition, 1991.

[105] Yukio Matsumoto. An Introduction to Morse Theory. Translations of Mathematical

Monographs No 208. AMS, first edition, 2002.

[106] John W. Milnor. Morse Theory. Annals of Math. Series, No. 51. Princeton University
Press, third edition, 1969.

[107] John W. Milnor. On isometries of inner product spaces. Inventiones Mathematicae,

8:83–97, 1969.

[108] John W. Milnor. Topology from the Differentiable Viewpoint. The University Press of
Virginia, second edition, 1969.

[109] John W. Milnor. Curvatures of left invariant metrics on lie groups. Advances in
Mathematics, 21:293–329, 1976.

[110] John W. Milnor and James D. Stasheff. Characteristic Classes. Annals of Math. Series,
No. 76. Princeton University Press, first edition, 1974.
652 BIBLIOGRAPHY

[111] R. Mneimné and F. Testard. Introduction à la Théorie des Groupes de Lie Classiques.
Hermann, first edition, 1997.

[112] Jules Molk. Encyclopédie des Sciences Mathématiques Pures et Appliquées. Tome I
(premier volume), Arithmétique. Gauthier-Villars, first edition, 1916.

[113] Mitsuo Morimoto. Analytic Functionals on the Sphere. Translations of Mathematical

Monographs No 178. AMS, first edition, 1998.

[114] Shigeyuki Morita. Geometry of Differential Forms. Translations of Mathematical

Monographs No 201. AMS, first edition, 2001.

[115] James R. Munkres. Topology, a First Course. Prentice Hall, first edition, 1975.

[116] James R. Munkres. Analysis on Manifolds. Addison Wesley, 1991.

[117] Raghavan Narasimham. Compact Riemann Surfaces. Lecture in Mathematics ETH

Zürich. Birkhäuser, first edition, 1992.

[118] Mitsuru Nishikawa. On the exponential map of the group O(p, q)0 . Memoirs of the
Faculty of Science, Kyushu University, Ser. A, 37:63–69, 1983.

[119] Barrett O’Neill. Semi-Riemannian Geometry With Applications to Relativity. Pure

and Applies Math., Vol 103. Academic Press, first edition, 1983.

[120] Xavier Pennec. Intrinsic statistics on Riemannian Manifolds: Basic tools for geometric
measurements. Journal of Mathematical Imaging and Vision, 25:127–154, 2006.

[121] Peter Petersen. Riemannian Geometry. GTM No. 171. Springer Verlag, second edition,
2006.

[122] L. Pontryagin. Topological Groups. Princeton University Press, first edition, 1939.

[123] L. Pontryagin. Topological Groups. Gordon and Breach, second edition, 1960.

[124] Ian R. Porteous. Topological Geometry. Cambridge University Press, second edition,
1981.

[125] M.M. Postnikov. Geometry VI. Riemannian Geometry. Encyclopaedia of Mathematical

Sciences, Vol. 91. Springer Verlag, first edition, 2001.

[126] Marcel Riesz. Clifford Numbers and Spinors. Kluwer Academic Press, first edition,
1993. Edited by E. Folke Bolinder and Pertti Lounesto.

[127] Wulf Rossmann. Lie Groups. An Introduction Through Linear Groups. Graduate Texts
in Mathematics. Oxford University Press, first edition, 2002.
BIBLIOGRAPHY 653

[128] Joseph J. Rotman. Introduction to Algebraic Topology. GTM No. 119. Springer Verlag,
first edition, 1988.

[129] Arthur A. Sagle and Ralph E. Walde. Introduction to Lie Groups and Lie Algebras.
Academic Press, first edition, 1973.

[130] Takashi Sakai. Riemannian Geometry. Mathematical Monographs No 149. AMS, first
edition, 1996.

[131] Hans Samelson. Notes on Lie Algebras. Universitext. Springer, second edition, 1990.

[132] Giovanni Sansone. Orthogonal Functions. Dover, first edition, 1991.

[133] Hajime Sato. Algebraic Topology: An Intuitive Approach. Mathematical Monographs

No 183. AMS, first edition, 1999.

[134] D.H. Sattinger and O.L. Weaver. Lie Groups and Algebras with Applications to Physics,
Geometry, and Mechanics. Applied Math. Science, Vol. 61. Springer Verlag, first
edition, 1986.

[135] Laurent Schwartz. Analyse II. Calcul Différentiel et Equations Différentielles. Collec-
tion Enseignement des Sciences. Hermann, 1992.

[136] Jean-Pierre Serre. Lie Algebras and Lie Groups. Lecture Notes in Mathematics, No.
1500. Springer, second edition, 1992.

[137] Jean-Pierre Serre. Complex Semisimple Lie Algebras. Springer Monographs in Math-
ematics. Springer, first edition, 2000.

[138] Igor R. Shafarevich. Basic Algebraic Geometry 1. Springer Verlag, second edition,
1994.

[139] Richard W. Sharpe. Differential Geometry. Cartan’s Generalization of Klein’s Erlan-

gen Program. GTM No. 166. Springer Verlag, first edition, 1997.

[140] Marcelo Siqueira, Dianna Xu, and Jean Gallier. Construction of C ∞ -surfaces from
triangular meshes using parametric pseudo-manifolds. Technical report, University
of Pennsylvania, Levine Hall, Philadelphia, PA 19104, 2008. pdf file available from
http://repository.upenn.edu/cis reports/877.

[141] Norman Steenrod. The Topology of Fibre Bundles. Princeton Math. Series, No. 14.
Princeton University Press, 1956.

[142] Elias M. Stein and Guido Weiss. Introduction to Fourier Analysis on Euclidean Spaces.
Princeton Math. Series, No. 32. Princeton University Press, 1971.

[143] S. Sternberg. Lectures On Differential Geometry. AMS Chelsea, second edition, 1983.
654 BIBLIOGRAPHY

[144] Michael E. Taylor. Noncommutative Harmonic Analysis. Mathematical Surveys and

Monographs, No. 22. AMS, first edition, 1986.

[145] Loring W. Tu. An Introduction to Manifolds. Universitext. Springer Verlag, first

edition, 2008.

[146] N.J. Vilenkin. Special Functions and the Theory of Group Representations. Transla-
tions of Mathematical Monographs No 22. AMS, first edition, 1968.

[147] Frank Warner. Foundations of Differentiable Manifolds and Lie Groups. GTM No. 94.
Springer Verlag, first edition, 1983.

[148] André Weil. Foundations of Algebraic Topology. Colloquium Publications, Vol. XXIX.
AMS, second edition, 1946.

[149] André Weil. L’Intégration dans les Groupes Topologiques et ses Applications. Hermann,
second edition, 1979.

[150] R.O. Wells. Differential Analysis on Complex Manifolds. GTM No. 65. Springer Verlag,
second edition, 1980.

[151] Hermann Weyl. The Classical Groups. Their Invariants and Representations. Prince-
ton Mathematical Series, No. 1. Princeton University Press, second edition, 1946.