6

Differential forms

6.1 Review: Differential forms on Rm

A differential k-form on an open subset U ✓ Rm is an expression of the form

w= Â wi1 ...ik dxi1 ^ · · · ^ dxik

i1 ···ik

where wi1 ...ik 2 C• (U) are functions, and the indices are numbers

1  i1 < · · · < ik  m.

Let W k (U) be the vector space consisting of such expressions, with pointwise addi-
tion. It is convenient to introduce a short hand notation I = {i1 , . . . , ik } for the index
set, and write w = ÂI wI dxI with

wI = wi1 ...ik , dxI = dxi1 ^ · · · ^ dxik .

m!
Since a k-form is determined by these functions wI , and since there are k!(m k)! ways
of picking k-element subsets from {1, . . . , m}, the space W k (U) can be identified with
vector-valued smooth functions,
m!
W k (U) = C• (U, R k!(m k)! ).

The dxI are just formal expressions; at this stage they do not have any particular
meaning. They are used, however, to define an associative product operation

W k (U) ⇥ W l (U) ! W k+l (U)

by the ‘rule of computation’

dxi ^ dx j = dx j ^ dxi

for all i, j; in particular dxi ^ dxi = 0. In turn, using the product structure we may
define the exterior differential
102 6 Differential forms
⇣ ⌘ m
∂ wI
d : W k (U) ! W k+1 (U), d  wI dxI =   i dxi ^ dxI . (6.1)
I i=1 I ∂ x

The key property of the exterior differential is the following fact:

Proposition 6.1. The exterior differential satisfies

d d = 0,

i.e. ddw = 0 for all w.

Proof. By definition,
m m
∂ 2 wI
ddw = Â Â Â ∂ x j ∂ xi dx j ^ dxi ^ dxI ,
j=1 i=1 I

which vanishes by equality of mixed partials ∂ ∂xiw∂Ix j = ∂ ∂x jw∂Ixi . (We have dxi ^ dx j =
dx j ^ dxi , but the coefficients in front of dxi ^ dx j and dx j ^ dxi are the same.) t u

Exercise 75.
(a) A 0-form on R3 is simply a smooth function f 2 W 0 (R3 ). Use the definition
of the exterior differential above to compute the resulting 1-form d f .
(b) A general 1-form w 2 W 1 (R3 ) is an expression

w = f dx + gdy + hdz

with smooth functions f , g, h 2 C• (R3 ). Use the definition of the exterior

differential above to compute the resulting 2-form dw.
(c) A general 2-form w 2 W 2 (R3 ) may be written

w = a dy ^ dz + b dz ^ dx + c dx ^ dy,

with A = (a, b, c) : U ! R3 . Use the definition of the exterior differential

above to compute the resulting 3-form dw.
(d) Relate your results from the previous parts to familiar vector-calculus op-
erators, and conclude that the usual properties

curl(grad( f )) = 0, div(curl(F)) = 0

are both special cases of d d = 0.

(e) Write an expression for a general 3-form r 2 W 3 (R3 ). What is dr?
(f) Write an expression for a general element of W 4 (R3 ). Generalize to W m (Rn )
for m > n (prove your assertion by using the ‘rule of computation’).

The support supp(w) ✓ U of a differential form is the smallest closed subset such
that w vanishes on U\supp(w). Suppose w 2 W m (U) is a compactly supported form
of top degree k = m. Such a differential form is an expression
 6.2 Dual spaces 103

w = f dx1 ^ · · · ^ dxm

where f 2 C• (U) is a compactly supported function. One defines the integral of w

to be the usual Riemann integral:
Z Z
w= f (x1 , . . . , xm )dx1 · · · dxm . (6.2)
U Rm

Note that we can regard w as a form on all of Rm , due to the compact support condi-
tion.
Our aim is now to define differential forms on manifolds, beginning with 1-
forms. Even though 1-forms on U ✓ Rm are identified with functions U ! Rm ,
they should not be regarded as vector fields, since their transformation properties
under coordinate changes are different. In fact, while vector fields are sections of the
tangent bundle, the 1-forms are sections of its dual, the cotangent bundle. We will
therefore begin with a review of dual spaces in general.

6.2 Dual spaces

For any real vector space E, we denote by E ⇤ = L(E, R) its dual space, consisting of
all linear maps a : E ! R. We will assume that E is finite-dimensional. Then the
dual space is also finite-dimensional, and dim E ⇤ = dim E. ⇤ It is common to write
the value of a 2 E ⇤ on v 2 E as a pairing, using the bracket notation:†

ha, vi := a(v).

This pairing notation emphasizes the duality between a and v. In the notation a(v)
we think of a as a function acting on elements of E, and in particular on v. However,
one may just as well think of v as acting on elements of E ⇤ by evaluation: v(a) =
a(v) for all a 2 E ⇤ . This symmetry manifests notationally in the pairing notation.
Let e1 , . . . , er be a basis of E. Any element of E ⇤ is determined by its values on
these basis vectors. For i = 1, . . . , r, let ei 2 E ⇤ (with upper indices) be the linear
functional such that ⇢
i i 0 if i 6= j,
he , e j i = d j =
1 if i = j.
The elements e1 , . . . , er are a basis of E ⇤ ; this is called the dual basis. The element
a 2 E ⇤ is described in terms of the dual basis as
r
a= Â a j e j, a j = ha, e j i.
j=1

⇤ For possibly infinite-dimensional vector spaces, the dual space E ⇤ is not isomorphic to E,
in general.
† In physics, one also uses the Dirac bra-ket notation ha| vi := a(v); here a = ha| is the
‘bra’ and v = |vi is the ‘ket’.
104 6 Differential forms

Similarly, for vectors v 2 E we have

r
v = Â vi ei , vi = hei , vi.
i=1

Notice the placement of indices: In a given summation over i, j, . . ., upper indices are
always paired with lower indices.
Remark 6.1. As a special case, for Rr with its standard basis, we have a canonical
identification (Rr )⇤ = Rr . For more general E with dim E < •, there is no canonical
isomorphism between E and E ⇤ unless more structure is given.

Exercise 76. Let V be a finite dimensional real vector space equipped with an
inner product h·, ·i. Every vector v 2 V determines a linear transformation Av by

v 7! hv, ·i.

(a) Show that if V = Rn , then Av = v⇤ .

(b) For a general V , let {e1 , . . . , en } be an orthonormal basis. Show that
{Ae1 , . . . , Aen } is the corresponding dual basis.
(c) Conversely, for some element A 2 V ⇤ , show how to use the inner product
in order to recover a v 2 V such that A = Av . Show that this v is unique.
We see that equipping a finite-dimensional real vector space with an inner prod-
uct gives it an additional structure that allows for a canonical isomorphism be-
tween V and V ⇤ .

Given a linear map R : E ! F between vector spaces, one defines the dual map
R⇤ : F ⇤ ! E ⇤
(note the direction), by setting
hR⇤ b , vi = hb , R(v)i
for b 2 F ⇤ and v 2 E. This satisfies (R⇤ )⇤ = R, and under the composition of linear
maps,
(R1 R2 )⇤ = R⇤2 R⇤1 .
In terms of basis e1 , . . . , er of E and f1 , . . . , fs of F, and the corresponding dual bases
(with upper indices), a linear map R : E ! F is given by the matrix with entries
Ri j = h f j , R(ei )i,
while R⇤ is described by the transpose of this matrix (the roles of i and j are re-
versed). Namely,‡
‡ In bra-ket notation, we have Ri j = h f j |R |ei i, and
|Rei i = R|ei i = Â | f j ih f j |R |ei i, hR⇤ ( f j )| = h( f j )|R = h f j |R |ei ihei |
j
 6.3 Cotangent spaces 105
s r
R(ei ) = Â Ri j f j , R⇤ ( f j ) = Â Ri j f i .
j=1 i=1

Thus,
(R⇤ ) j i = Ri j .

6.3 Cotangent spaces

Definition 6.1. The dual of the tangent space Tp M of a manifold M is called the
cotangent space at p, denoted

Tp⇤ M = (Tp M)⇤ .

Elements of Tp⇤ M are called cotangent vectors, or simply covectors. Given a smooth
map F 2 C• (M, N), and any p 2 M we have the cotangent map

Tp⇤ F = (Tp F)⇤ : TF(p)

⇤
N ! Tp⇤ M

defined as the dual to the tangent map.

Thus, a co(tangent) vector at p is a linear functional on the tangent space, as-

signing to each tangent vector at p a number. The very definition of the tangent
space suggests one such functional: Every function f 2 C• (M) defines a linear map,
Tp M ! R, v 7! v( f ). This linear functional is denoted (d f ) p 2 Tp⇤ M.§

Definition 6.2. Let f 2 C• (M) and p 2 M. The covector

(d f ) p 2 Tp⇤ M, h(d f ) p , vi = v( f ).

is called the differential of f at p.

Lemma 6.1. For F 2 C• (M, N) and g 2 C• (N),

d(F ⇤ g) p = Tp⇤ F((dg)F(p) ).

Exercise 77. Prove the lemma.

vectors; so it suffices to show that the pairings are the same.)
(Hint: every element of the dual space is completely determined by its action of

Consider an open subset U ✓ Rm , with coordinates x1 , . . . , xm . Here TpU ⇠

= Rm ,
with basis
∂ ∂
, . . . , m 2 TpU (6.3)
∂ x1 p ∂x p

§ Note that this is actually the same as the tangent map Tp f : Tp M ! T f (p) R = R.
106 6 Differential forms

The basis of the dual space Tp⇤U, dual to the basis (6.3), is given by the differentials
of the coordinate functions:

(dx1 ) p , . . . , (dxm ) p 2 Tp⇤U.

Indeed,
D ∂ E ∂
(dxi ) p , = (xi ) = d i j
∂xj p ∂xj p
as required. For f 2 C• (M), the coefficients of (d f ) p = Âi h(d f ) p , ei iei are deter-
mined as D ∂ E ∂ ∂f
(d f ) p , = (f) = .
∂xj p ∂xj p ∂xj p
Thus,
m
∂f
(d f ) p = Â i (dxi ) p .
i=1 ∂ x p

Let U ✓ Rm and V ✓ Rn be open, with coordinates x1 , . . . , xm and y1 , . . . , yn . For

F 2 C• (U,V ), the tangent map is described by the Jacobian matrix, with entries

∂F j
(D p F)i j = (p)
∂ xi
for i = 1, . . . , m, j = 1, . . . , n. We have:
n
∂ ∂
(Tp F)(
∂ xi p
)= Â (D p F)i j ∂yj F(p)
,
j=1

hence dually
m
(Tp F)⇤ (dy j )F(p) = Â (D p F)i j (dxi ) p . (6.4)
i=1
Thought of as matrices, the coefficients of the cotangent map are the transpose of the
coefficients of the tangent map.

Exercise 78. Consider R3 with coordinates x, y, z, and R2 with coordinates u, v.

Let F : R3 ! R2 be given by

(x, y, z) 7! (x2 y + ez , yz x).

✓ ◆
Let p = (1, 1, 1). What is Tp F ∂∂x ? What is Tp⇤ F(du)F(p) ?
p

6.4 1-forms
Similar to the definition of vector fields, one can define co-vector fields, more com-
monly known as 1-forms: Collections of covectors a p 2 Tp⇤ M depending smoothly
 6.4 1-forms 107

on the base point. One approach of making precise the smooth dependence on the
base point is to endow the cotangent bundle
G
T ⇤M = Tp⇤ M.
p

(disjoint union of all cotangent spaces) with a smooth structure, and require that the
map p 7! a p be smooth. The construction of charts on T ⇤ M is similar to that for
the tangent bundle: Charts (U, j) of M give cotangent charts (T ⇤U, T ⇤ j 1 ) of T ⇤ M,
using the fact that T ⇤ (j(U)) = j(U)⇥Rm canonically (since j(U) is an open subset
of Rm ). Here T ⇤ j 1 : T ⇤U ! T ⇤ j(U) is the union of inverses of all cotangent maps
Tp⇤ j : Tj(p)
⇤ j(U) ! T ⇤U.
p

Exercise 79. Carry out this construction to prove that T ⇤ M is naturally a 2m-
dimensional manifold.
A second approach is observe that in local coordinates, 1-forms are given by
expressions Âi fi dxi , and smoothness should mean that the coefficient functions are
smooth.
We will use the following (equivalent) approach.
Definition 6.3. A 1-form on M is a linear map
a : X(M) ! C• (M), X 7! a(X) = ha, Xi,
which is C• (M)-linear in the sense that
a( f X) = f a(X)
for all f 2 C• (M), X 2 X(M). The space of 1-forms is denoted W 1 (M).
Let us verify that a 1-form can be regarded as a collection of covectors:
Lemma 6.2. Let a 2 W 1 (M) be a 1-form, and p 2 M. Then there is a unique covector
a p 2 Tp⇤ M such that
a(X) p = a p (X p )
for all X 2 X(M).
(We indicate the value of the function a(X) at p by a subscript, just like we did for
vector fields.)
Proof. We have to show that a(X) p depends only on the value of X at p. By consid-
ering the difference of vector fields having the same value at p, it is enough to show
that if X p = 0, then a(X) p = 0. But any vector field vanishing at p can be written as a
finite sum X = Âi fiYi where fi 2 C• (M) vanish at p. ¶ By C• -linearity, this implies
that
a(X) = a(Â fiYi ) = Â fi a(Yi )
i i
vanishes at p. t
u
¶ ∂
For example, using local coordinates, we can take the Yi to correspond to ∂ xi
near p, and
the fi to the coefficient functions.
108 6 Differential forms

The first example of a 1-form is described in the following definition.

Definition 6.4. The exterior differential of a function f 2 C• (M) is the 1-form

d f 2 W 1 (M),

defined in terms of its pairings with vector fields X 2 X(M) as hd f , Xi = X( f ).

Clearly, d f is the 1-form defined by the family of covectors (d f ) p . Note that critical
points of f may be described in terms of this 1-form: p 2 M is a critical point of f if
and only if (d f ) p = 0.
Similar to vector fields, 1-forms can be multiplied by functions; hence one has
more general examples of 1-forms as finite sums,

a = Â fi dgi
i

where fi , gi 2 C• (M).
Let us examine what the 1-forms are for open subsets U ✓ Rm . Given a 2 W 1 (U),
we have
m
a = Â ai dxi
i=1
⌦ ↵
with coefficient functions ai = a, ∂∂xi 2 C• (U). (Indeed, the right hand side takes
on the correct values at any p 2 U, and is uniquely determined by those values.)
General vector fields on U may be written
m
∂
X= Â X j ∂xj
j=1

(to match the notation for 1-forms, we write the coefficients as X i rather than ai , as
we did in the past), where the coefficient functions are recovered as X j = hdx j , Xi.
The pairing of the 1-form a with the vector field X is then
m
ha, Xi = Â ai X i .
i=1

Lemma 6.3. Let a : p 7! a p 2 Tp⇤ M be a collection of covectors. Then a defines a

1-form, with
a(X) p = a p (X p )
for p 2 M, if and only if for all charts (U, j), the coefficient functions for a in the
chart are smooth.

Exercise 80. Prove Lemma 6.3.

 6.5 Pull-backs of function and 1-forms 109

6.5 Pull-backs of function and 1-forms

Recall again that for any manifold M, the vector space C• (M) of smooth functions is
an algebra, with product the pointwise multiplication. Any smooth map F : M ! M 0
between manifolds defines an algebra homomorphism, called the pull-back

F ⇤ : C• (M 0 ) ! C• (M), f 7! F ⇤ ( f ) := f F.

Exercise 81. Show that the pull-back is indeed an algebra homomorphism by

showing that it preserves sums and products:

F ⇤ ( f ) + F ⇤ (g) = F ⇤ ( f + g) ; F ⇤ ( f )F ⇤ (g) = F ⇤ ( f g).

Next, show that if F, F 0 : M ! M 0 are two smooth maps between manifolds,

then
(F 0 F)⇤ = F ⇤ (F 0 )⇤ .
(With order reversed.)

Let F 2 C• (M, N) be a smooth map. Recall that for vector fields, there is no
general ‘push-forward’ or ‘pull-back’ operations, unless F is a diffeomorphism. For
1-forms the situation is better. Indeed, for any p 2 M one has the dual to the tangent
map
Tp⇤ F = (Tp F)⇤ : TF(p)
⇤
N ! Tp⇤ M.
For a 1-form b 2 W 1 (N), we can therefore define

(F ⇤ b ) p := (Tp⇤ F)(bF(p) ).

This gives us a collection of covectors in Tp⇤ M at each point p 2 M. The following

Lemma shows that these form a 1-form.
Lemma 6.4. The collection of co-vectors (F ⇤ b ) p 2 Tp⇤ M depends smoothly on p,
defining a 1-form F ⇤ b 2 W 1 (M).
Proof. We shall use Lemma 6.3. By working on local coordinates, we may assume
that M is an open subset U ✓ Rm , and N is an open subset V ✓ Rn . Write
n
b= Â b j (y)dy j .
j=1

By (6.4), the pull-back of b is given by

m ⇣ n
∂F j ⌘
F ⇤b = Â Â b j (F(x)) ∂ xi dxi .
i=1 j=1

In particular, the coefficients are smooth. t

u
110 6 Differential forms

Lemma 6.4 shows that we have a well-defined pull-back map

F ⇤ : W 1 (N) ! W 1 (M), b 7! F ⇤ b .
Note that with respect to composition of two maps
(F1 F2 )⇤ = F2⇤ F1⇤
with order reversed.
A nice property of the pull-back of forms is its relation to the pull-back of func-
tions. Lemma 6.1 shows that for g 2 C• (N),
F ⇤ (dg) = d(F ⇤ g)
(Note that on the left we are pulling-back a form, and on the right a function.)
Exercise 82. In this exercise we shall use coordinates x, y, z on the domain and
u, v, w on the target space. Consider the maps F : R3 ! R2 and g : R2 ! R given
by

F(x, y, z) = (x3 eyz , sin x)

g(u, v) = (u + v)2 euv .

(a) Compute F ⇤ (du) and F ⇤ (v cos udv). Compute F ⇤ (v cos udu + sin udv) by
using Lemma 6.1.
(b) Verify Lemma 6.1 by computing dg, F ⇤ (dg), as well as F ⇤ g and d(F ⇤ g).

Recall once again that while F 2 C• (M, N) induces a tangent map T F 2 C• (T M, T N),
there is no natural push-forward operation for vector fields. By contrast, for cotan-
gent bundles there is no naturally induced map from T ⇤ N to T ⇤ M (or the other way),
yet there is a natural pull-back operation for 1-forms!
In the case of vector fields, rather than working with ‘F⇤ (X)’ one has the notion
of related vector fields, X ⇠F Y . We know that 1-forms act on vector fields, how do
they act on related vector fields?

Exercise 83. Show that for any related vector fields X ⇠F Y , and b 2 W 1 (N),

(F ⇤ b )(X) = F ⇤ (b (Y )).

(Notice once again the different notions of pull-back that we are using.)

Exercise 84. Recall that for a given vector field on a manifold X 2 X(M), the
smooth curve g 2 C• (J, M) (where J ✓ R) is a solution curve iff

∂
gX.
∂t
Let g : R ! R2 be given by
 6.6 Integration of 1-forms 111

t 7! (cost, sint).

Let X 2 X(M) be given by

∂ ∂
x y .
∂y ∂x
Finally, let b 2 W 1 (M) be given by dx dy.
∂
(a) Show by computing directly that ∂t is F-related to X.
(b) Verify the conclusion of Exercise 83 by computing each of b (X), F ⇤ b ,
∂
F ⇤ b ( ∂t ), and F ⇤ (b (X)).

6.6 Integration of 1-forms

Given a curve g : J ! M in a manifold, and any 1-form a 2 W 1 (M), we can consider
the pull-back g ⇤ a 2 W 1 (J). By the description of 1-forms on R, this is of the form

g ⇤ a = f (t)dt

for a smooth function f 2 C• (J).

To discuss integration, it is convenient to work with closed intervals rather than
open intervals. Let [a, b] ✓ R be a closed interval. A map g : [a, b] ! M into a
manifold will be called smooth if it extends to a smooth map from an open interval
containing [a, b]. We will call such a map a smooth path.

Definition 6.5. Given a smooth path g : [a, b] ! M, we define the integral of a 1-form
a 2 W 1 (M) along g as
Z Z b
a= g ⇤ a.
g a

The fundamental theorem of calculus has the following consequence for mani-
folds. It is a special case of Stokes’ theorem.
Proposition 6.2. Let g : [a, b] ! M be a smooth path, with g(a) = p, g(b) = q. For
any f 2 C• (M), we have Z
d f = f (q) f (p).
g

In particular, the integral of d f depends only on the end points of the path, rather
than the path itself.

Proof. We have
∂ ( f g)
g ⇤ d f = dg ⇤ f = d( f g) = dt.
∂t
Integrating from a to b, we obtain, by the fundamental theorem of calculus, f (g(b))
f (g(a)). tu
112 6 Differential forms

Exercise 85. Given a diffeomorphism k : [c, d] ! [a, b] one defines the corre-
sponding reparametrization

g k : [c, d] ! M.

The diffeomorphism (or the reparametrization) is called orientation preserving

if k(c) = a, k(d) = b, orientation reversing if k(c) = b, k(d) = a.
Prove that the integral is invariant under orientation preserving reparametriza-
tion Z Z
a= a,
g g k
R R
while an orientation reversing reparametrization gives g a = g k a.

Exercise 86. Consider the 1-form

a = y2 ex dx + 2yex dy 2 W (R2 ).

Find the integral of a along the path

g : [0, 1] ! M, t 7! (sin(pt/2),t 3 ).

A 1-form a 2 W 1 (M) such that a = d f for some function f 2 C• (M) is called

exact. Proposition 6.2 gives a necessary condition for exactness: The integral of a
along paths should depend only on the end points. This condition is also sufficient,
since we can define f on the connected components R
of M, by fixing a base point p0
on each such component, and putting f (p) = g a for any path from p0 to p.
If M is an open subset U ✓ Rm , so that a = Âi ai dxi , then a = d f means that
ai = ∂∂ xfi . A necessary condition is the equality of partial derivatives,

∂ ai ∂aj
= ,
∂xj ∂ xi
In multivariable calculus one learns that this condition is also sufficient, provided U
is simply connected (e.g., convex). Using the exterior differential of forms in W 1 (U),
this condition becomes da = 0. Since a is a 1-form, da is a 2-form. Thus, to obtain
a coordinate-free version of the condition, we need higher order forms.

6.7 k-forms
To get a feeling for higher degree forms, and constructions with higher forms, we
first discuss 2-forms.
 6.7 k-forms 113

6.7.1 2-forms.

Definition 6.6. A 2-form on M is a C• (M)-bilinear skew-symmetric map

a : X(M) ⇥ X(M) ! C• (M), (X,Y ) 7! a(X,Y ).

Here skew-symmetry means that a(X,Y ) = a(Y, X) for all vector fields X,Y , while
C• (M)-bilinearity means

a( f X,Y ) = f a(X,Y ) = a(X, fY )

for f 2 C• (M), as well as a(X 0 + X 00 ,Y ) = a(X 0 ,Y ) + a(X 00 ,Y ), and similarly in the

second argument. (Actually, by skew-symmetry it suffices to require C• (M)-linearity
in the first argument.) By the same argument as for 1-forms, the value a(X,Y ) p
depends only on the values X p ,Yp . Also, if a is a 2-form then so is f a for any
smooth function f .
Our first examples of 2-forms are obtained from 1-forms: Let a, b 2 W 1 (M).
Then we define a wedge product a ^ b 2 W 2 (M), as follows:

(a ^ b )(X,Y ) = a(X)b (Y ) a(Y )b (X). (6.5)

Exercise 87. Show that Equation (6.5) indeed defines a 2-form.

For an open subset U ✓ Rm , a 2-form w 2 W 2 (U) is uniquely determined by its

values on coordinate vector fields. By skew-symmetry the functions
⇣ ∂ ∂ ⌘
wi j = w ,
∂ xi ∂ x j
satisfy wi j = w ji ; hence it suffices to know these functions for i < j. As a conse-
quence, we see that the most general 2-form on U is

1 m
2 i,Â
w= wi j dxi ^ dx j = Â wi j dxi ^ dx j .
j=1 i< j

Exercise 88. Compute the following 2-forms on R3 with coordinates (x, y, z)

(that is, write them in normal form as above):
(a) (3dx) ^ ( 7dy) + (xdy) ^ (dx).
(b) (dx dy) ^ (dx + dz).
3 2
(c) d(x2 + xyz) ^ d(zex y ).

We now generalize to forms of arbitrary degree.

114 6 Differential forms

6.7.2 k-forms

Definition 6.7. Let k be a non-negative integer. A k-form on M is a C• (M)-multilinear,

skew-symmetric map

a : X(M) ⇥ · · · ⇥ X(M) ! C• (M).

| {z }
k times

The space of k-forms is denoted W k (M); in particular W 0 (M) = C• (M).

Here, skew-symmetry means that a(X1 , . . . , Xk ) changes sign under exchange of any
two of its elements. For example, a(X1 , X2 , X3 , . . .) = a(X2 , X1 , X3 , . . .). More gen-
erally, if Sk is the group of permutations of {1, . . . , k}, and sgn(s) is the sign of a
permutation s 2 Sk (+1 for an even permutation, 1 for an odd permutation) then

a(Xs(1) , . . . , Xs(k) ) = sgn(s)a(X1 , . . . , Xk ).

The C• (M)-multilinearity means C• (M)-linearity in each argument, similarly to the

condition for 2-forms. It implies, in particular, a is local in the sense that the value of
a(X1 , . . . , Xk ) at any given p 2 M depends only on the values X1 | p , . . . , Xk | p 2 Tp M.
One thus obtains a skew-symmetric multilinear form

a p : Tp M ⇥ · · · ⇥ Tp M ! R,

for all p 2 M.
If a1 , . . . , ak are 1-forms, then one obtains a k-form a =: a1 ^ . . . ^ ak by ‘wedge
product’.

(a1 ^ · · · ^ ak )(X1 , . . . , Xk ) = Â sign(s)a1 (Xs(1) ) · · · ak (Xs(k) ).

s2Sk

(More general wedge products will be discussed below.)

Exercise 89. Show that the wedge-product above indeed defines a k-form.

Using C• -multilinearity, a k-form on U ✓ Rm is uniquely determined by its val-

ues on coordinate vector fields ∂∂x1 , . . . , ∂ ∂xm , i.e. by the functions
⇣ ∂ ∂ ⌘
ai1 ...ik = a , . . . , .
∂ xi1 ∂ x ik
Moreover, by skew-symmetry we only need to consider ordered index sets I =
{i1 , . . . , ik } ✓ {1, . . . , m}, that is, i1 < · · · < ik . Using the wedge product notation,
we obtain
a = Â ai1 ...ik dxi1 ^ · · · dxik .
i1 <···<ik
 6.7 k-forms 115

Exercise 90. Compute the following 3-forms on R5 with coordinates (x1 , . . . , x5 ):

(a) (dx1 + dx2 ) ^ (dx2 + dx3 ) ^ (dx4 + dx5 ).
(b) d(x1 x23 ex5 + sin x3 cos x4 ) ^ d(cos(x12 x5 ) ex3 ) ^ d(x4 ).

6.7.3 Wedge product

We next turn to the definition of a wedge product of forms of arbitrary degree a 2
W k (M) and b 2 W l (M). A permutation s 2 Sk+l is called a k, l shuffle if it satisfies
s(1) < · · · < s(k), s(k + 1) < · · · < s(k + l).

Definition 6.8. The wedge product of a 2 W k (M), b 2 W l (M) is the element

a ^ b 2 W k+l (M)
given as
(a ^ b )(X1 , . . . , Xk+l ) = Â sgn(s)a(Xs(1) , . . . , Xs(k) ) b (Xs(k+1) , . . . , Xs(k+l) )
where the sum is over all k, l-shuffles.

Exercise 91.
(a) Show that Definition 6.8 is consistent with our previous definition of the
wedge-product of two 1-forms (Equation (6.5)).
(b) Show that Definition 6.8 indeed defines a (k + l)-form.

Exercise 92.
(a) For a, b , r 2 W 2 (M), and T,U,V,W, X,Y, Z 2 X(M), compute

(a ^ b )(W, X,Y, Z)

and
(a ^ b ) ^ r(T,U,V,W, X,Y, Z).
(b) For a, b 2 W 3 (M), and T,U,V,W, X,Y, Z 2 X(M), compute

(a ^ b )(T,U,V,W, X,Y, Z).

Exercise 93.
(a) Prove that the wedge product is graded commutative: If a 2 W k (M) and
b 2 W l (M) then
a ^ b = ( 1)kl b ^ a.
116 6 Differential forms

(b) Prove that the wedge product is associative: Given ai 2 Wki (M) we have

(a1 ^ a2 ) ^ a3 = a1 ^ (a2 ^ a3 ).

So, we may in fact drop the parentheses when writing wedge products.

6.7.4 Exterior differential

Recall that we have defined the exterior differential on functions by the formula

(d f )(X) = X( f ). (6.6)

we will now extend this definition to all forms.

Theorem 6.1. There is a unique collection of linear maps d : W k (M) ! W k+1 (M),
extending the map (6.6) for k = 0, such that d(d f ) = 0 and satisfying the graded
product rule,
d(a ^ b ) = da ^ b + ( 1)k a ^ db (6.7)
for a 2 W k (M) and b 2 W l (M). This exterior differential satisfies d d = 0.

Proof. Suppose first that such an exterior differential is given. Then d is local, in the
sense that for any open subset U ✓ M the restriction (da)|U depends only on a|U ,
or equivalently (da)|U = 0 when a|U = 0. Indeed, if this is the case and p 2 U, we
may choose f 2 C• (M) = W 0 (M) such that f vanishes on M\U and f | p = 1. Then
f a = 0, hence the product rule (6.7) gives

0 = d( f a) = d f ^ a + f da.

Evaluating at p we obtain (da) p = 0 as claimed. Using locality, we may thus work

in local coordinates. If a 2 W 1 (M) is locally given by

a= Â ai1 ···ik dxi1 ^ · · · ^ dxik ,

i1 <···<ik

then the product rule together with ddxi = 0 forces us to define

m ∂ ai1 ···ik l
da = Â dai1 ···ik ^ dxi1 ^ · · · ^ dxik = Â Â ∂ x l
dx ^ dxi1 ^ · · · ^ dxik .
i1 <···<ik l=1 i1 <···<ik

Conversely, we may use this explicit formula (cf. (6.1)) to define da|U for a coor-
dinate chart domain U; by uniqueness the local definitions on overlas of coordinate
chart domains agree. Proposition 6.1 shows that (dda)|U = 0, hence it also holds
globally. tu
 6.8 Lie derivatives and contractions 117

Exercise 94. Find the exterior differential of each of the following forms on
R3 (with coordinates (x, y, z)).
(a) a = y2 ex dy + 2yex dx.
(b) b = y2 ex dx + 2yex dy.
2
(c) r = ex y sin zdx ^ dy + 2 cos(z3 y)dx.
sin exy cos sin z3 x
(d) w = 1+(x+y+z) 4 +(7xy)6 dx ^ dy ^ dz.

Definition 6.9. A k-form w 2 W k (M) is called exact if w = da for some a 2

W k 1 (M). It is called closed if dw = 0.
Since d d = 0, the exact k-forms are a subspace of Rthe space of closed k-forms. For
the case of 1-forms, we had seen that the integral g a of an exact 1-form a = d f
along a smooth path g : [a, b] ! M is given by the difference of the values at the end
points; a necessary condition for a to be exact is that it is closed. An example of a
1-form that is closed but not exact is
ydx xdy
a= 2 W 1 (R2 \{0}).
x 2 + y2
Remark 6.2. The quotient space (closed k-forms modulo exact k-forms) is a vector
space called the k-th (de Rham) cohomology

{a 2 W k (M)| a is closed }
H k (M) = .
{a 2 W k (M)| a is exact }

It turns out that whenever M is compact (and often also if M is non-compact), H k (M)
is a finite-dimensional vector space. The dimension of this vector space

bk (M) = dim H k (M)

is called the k-th Betti number of M; these numbers are important invariants of M
which one can use to distinguish non-diffeomorphic manifolds. For example, if M =
CPn one can show that

bk (CPn ) = 1 for k = 0, 2, . . . , 2n

and bk (CPn ) = 0 otherwise. For M = SN the Betti numbers are

bk (Sn ) = 1 for k = 0, n

while bk (Sn ) = 0 for all other k. Hence CPn cannot be diffeomorphic to S2n unless
n = 1.

6.8 Lie derivatives and contractions

Given a vector field X, and a k-form a 2 W k (M), we can define a (k 1)-form
118 6 Differential forms

iX a 2 W k 1
(M)

by contraction: Thinking of a as a multi-linear form, one simply puts X into the first
slot:
(iX a)(X1 , . . . , Xk 1 ) = a(X, X1 , . . . , Xk 1 ).
Contractions have the following compatibility with the wedge product, similar to that
for the exterior differential:

iX (a ^ b ) = iX a ^ b + ( 1)k a ^ iX b , (6.8)

for a 2 W k (M), b 2 W l (M),

Exercise 95. Prove Equation (6.8).

Another important operator on forms is the Lie derivative:

Theorem 6.2. Given a vector field X, there is a unique collection of linear maps
LX : W k (M) ! W k (M), such that

LX ( f ) = X( f ), LX (d f ) = dX( f ),

and satisfying the product rule,

LX (a ^ b ) = LX a ^ b + a ^ LX b (6.9)

for a 2 W k (M) and b 2 W l (M).

Proof. As in the case of the exterior differential, we can use the product rule to show
that LX is local: (LX a)|U depends only on a|U and X|U . Since any differential form
is a sum of wedge products of 1-forms, LX is uniquely determined by its action on
functions and differential of functions. This proves uniqueness. For existence, we
give the following formula:

LX = d iX + iX d.

On functions, this gives the correct result since

LX f = iX d f = X( f ),

and also on differentials of functions since

LX d f = diX d f = dLX f = 0.

t
u
 6.8 Lie derivatives and contractions 119

Exercise 96. For each of the following vector-fields X and differential forms a
on R3 (with coordinates (x, y, z)) compute LX a
(a) X = y ∂∂x + x ∂∂y + z ∂∂z and a = ydx xdy zdz.
(b) The Wikipedia example: X = sin x ∂∂y y2 ∂∂x and a = x2 sin(y).
(c) The Wikipedia example: X = sin x ∂∂y y2 ∂∂x and a = (x2 + y2 )dx ^ dz.

To summarize, we have introduced three operators

d : W k (M) ! W k+1 (M), LX : W k (M) ! W k (M), iX : W k (M) ! W k 1

(M).

These have the following compatibilities with the wedge product: For a 2 W k (M)
and b 2 W l (M) one has

d(a ^ b ) = (da) ^ b + ( 1)k a ^ db ,

LX (a ^ b ) = (LX a) ^ b + a ^ LX b ,
iX (a ^ b ) = (iX a) ^ b + ( 1)k a ^ iX b .

One says that LX is an even derivation relative to the wedge product, whereas d, iX
are odd derivations. They also satisfy important relations among each other:

Exercise 97. Prove the following relations:

d d=0
LX LY LY LX = L[X,Y ]
iX iY + iY iX = 0
d LX LX d = 0
LX iY iY LX = i[X,Y ]
iX d + d iX = LX .

Again, the signs are determined by the even/odd parity of these operators; one should
think of the left hand side as ‘graded’ commutators, where a plus sign appears when-
ever two entries are odd. Writing [·, ·] for the graded commutators (with the agree-
ment that the commutator of two odd operators has a sign built in) the identities
become

[d, d] = 0,
[LX , LY ] = L[X,Y ] ,
[iX , iY ] = 0,
[d, LX ] = 0,
[LX , iY ] = i[X,Y ] ,
[d, iX ] = LX .
120 6 Differential forms

This collection of identities is referred to as the Cartan calculus, after Élie Cartan
(1861-1951), and in particular the last identity (which certainly is the most intrigu-
ing) is called the Cartan formula. Basic contributions to the theory of differential
forms were made by his son Henri Cartan (1906-1980), who also wrote a textbook
on the subject.

Exercise 98.
(a) As an illustration of the Cartan identities, let us prove the following formula
for the exterior differential of a 1-form a 2 W 1 (M):

(da)(X,Y ) = LX (a(Y )) LY (a(X)) a([X,Y ]).

(In the Cartan Calculus, we prefer to write LX f instead of X( f ) since ex-

pressions such as X(a(Y )) would look too confusing.)
(b) Prove a similar formula for the exterior differential of a 2-form, and try to
generalize to arbitrary k-form.

Exercise 99. Show that iX iX = 0.

Exercise 100. Prove the Jacobi-identity for the Lie derivative: for any X,Y, Z 2
M we have
L[X,[Y,Z]] + L[Y,[Z,X]] + L[Z,[X,Y ]] = 0.

6.8.1 Pull-backs

Similar to the pull-back of functions (0-forms) and 1-forms, we have a pull-back

operation for k-forms,
F ⇤ : W k (N) ! W k (M)
for any smooth map between manifolds, F 2 C• (M, N). Its evaluation at any p 2 M
is given by
(F ⇤ b ) p (v1 , . . . , vk ) = bF(p) (Tp F(v1 ), . . . , Tp F(vk )).
The pull-back map satisfies d(F ⇤ b ) = F ⇤ db , and for a wedge product of forms,

F ⇤ (b1 ^ b2 ) = F ⇤ b1 ^ F ⇤ b2 .

In local coordinates, if F : U ! V is a smooth map between open subsets of Rm and

Rn , with coordinates x, y, the pull-back just amounts to ‘putting y = F(x)’.

Exercise 101. If F : R3 ! R2 is given by (u, v) = F(x, y, z) = (y2 z, x), compute

F ⇤ (du ^ dv).

The next example is very important, hence we state it as a proposition. It is the

‘key fact’ toward the definition of an integral.
 6.8 Lie derivatives and contractions 121

Proposition 6.3. Let U ✓ Rm with coordinates xi , and V ✓ Rn with coordinates y j .

Suppose m = j, and F 2 C• (U,V ). Then

F ⇤ (dy1 ^ · · · ^ dyn ) = J dx1 ^ · · · ^ dxn

where J(x) is the determinant of the Jacobian matrix,

⇣ ∂ F i ⌘n
J(x) = det .
∂xj i, j=1

Proof.

F ⇤ b = dF 1 ^ · · · ^ dF n
∂ F1 ∂ Fn
= Â · · · i dxi1 ^ · · · ^ dxin
i1 ...in ∂ x ∂x n
i 1

∂ F1 ∂ F n s(1)
= Â ∂ x s(1)
· · ·
∂ x s(n)
dx ^ · · · ^ dxs(n)
s2Sn
∂ F1 ∂ Fn
= Â sign(s) ∂ xs(1) · · · ∂ xs(n) dx1 ^ · · · ^ dxn
s2Sn

= J dx1 ^ · · · ^ dxn ,

Here we noted that the wedge product dxi1 ^ · · · ^ dxin is zero unless i1 , . . . , in are a
permutation of 1, . . . , n. t
u

One may regard this result as giving a new, ‘better’ definition of the Jacobian deter-
minant.

Remark 6.3. The Lie derivative LX a of a differential form with respect to a vector
field X has an important interpretation in terms of the flow Ft of X. Assuming for
simplicity that X is complete (so that Ft is a globally defined diffeomorphism), one
has the formula
d
LX a = F ⇤ a.
dt t=0 t
(If X is incomplete, the flow Ft is defined only locally, but the definition still works.)
To prove this identity, it suffices to check that the right hand side satisfies a product
rule with respect to the wedge product of forms, and that it takes on the correct values
on functions and on differentials of functions. The formula shows that LX measures
to what extent a is invariant under the flow of X.

Exercise 102. Prove the identity by following the suggestion above.

122 6 Differential forms

6.9 Integration of differential forms

Differential forms of top degree can be integrated over oriented manifolds. Let M be
an oriented manifold of dimension m, and w 2 W m (M). Let supp(w) be the support
of w. k
If supp(w) is contained in an oriented coordinate chart (U, j), then one defines
Z Z
w= f (x)dx1 · · · dxm
M Rm

where f 2 C• (Rm ) is the function, with supp( f ) ✓ j(U), determined from

1 ⇤
(j ) w = f dx1 ^ · · · ^ dxm .

This definition does not depend on the choice of oriented chart. Indeed, suppose
(V, y) is another oriented chart with supp(w) ✓ V , and write
1 ⇤
(y ) w = g dy1 ^ · · · ^ dym .

where we write y1 , . . . , ym for the coordinates on V . Letting F = y j 1 be the

change of coordinates y = F(x), Proposition 6.3 says that

F ⇤ (dy1 ^ · · · ^ dym ) = J(x)dx1 ^ · · · ^ dxm ,

where J(x) = det(DF(x)) is the determinant of the Jacobian matrix of F at x. Hence,

f (x) = g(F(x))J(x), and we obtain
Z Z Z
g(y)dy1 · · · ym = g(F(x))J(x)dx1 · · · dxm = f (x)dx1 · · · dxm ,
y(U) j(U) j(U)

as required.
Remark 6.4. Here we used the change-of-variables formula from multivariable cal-
culus. It was very important that the charts are oriented, so that J > 0 everywhere.
Indeed, for general changes of variables, the change-of-variables formula involves
|J| rather than J itself.
If w is not necessarily supported in a single oriented chart, we proceed as follows.
Let (Ui , ji ), i = 1, . . . , r be a finite collection of oriented charts covering supp(w).
Together with U0 = M\supp(w) this is an open cover of M.
Lemma 6.5. Given a finite open cover of a manifold there exists a partition of unity
subordinate to the cover, i.e. functions ci 2 C• (M) with supp(ci ) ✓ Ui and Âri=0 ci =
1.
k The support of a form is defined similar to the support of a function, or support of a vec-
tor field. For any differential form a 2 W k (M), we define the support supp(a) to be the
smallest closed subset of M outside of which a is zero. (Equivalently, it is the closure of
the subset over which a is non-zero.)
 6.11 Stokes’ theorem 123

Indeed, partitions of unity exists for any open cover, not only finite ones. A proof is
given in the appendix on ‘topology of manifolds’.
Let c0 , . . . , cr be a partition of unity subordinate to this cover. We define
Z r Z
w=Â ci w
M i=1 M

where the summands are defined as above, since ci w is supported in Ui for i 1. (We
didn’t include the term for i = 0, since c0 w = 0.) We have to check that this is well-
defined, independent of the choices. Thus, let (V j , y j ) for j = 1, . . . , s be another
collection of oriented coordinate charts covering supp(w), put V0 = M supp(w),
and let s0 , . . . , ss a corresponding partition of unity subordinate to the cover by the
Vi ’s.
Then the Ui \V j form an open cover, with the collection of ci s j as a partition of
unity. We obtain
s Z s Z r s r Z
 s jw =  (  ci ) s j w =  s j ci w.
j=1 M j=1 M i=1 j=1 i=1 M
R
This is the same as the corresponding expression for Âri=1 M ci w.

6.10 Integration over oriented submanifolds

Let M be a manifold, not necessarily oriented, and S is a k-dimensional oriented
submanifold, with inclusion i : S ! M. We define the integral over S, of any k-form
w 2 W k (M) such that S \ supp(w) is compact, as follows:
Z Z
w= i⇤ w.
S S

Of course, this definition works equally well for any smooth map from S into M. For
example, the integral of compactly supported 1-forms along arbitrary paths g : R !
M is defined. Note also that M itself does not have to be oriented, it suffices that S is
oriented.

6.11 Stokes’ theorem

Let M be an m-dimensional oriented manifold.

Definition 6.10. A region with (smooth) boundary in M is a closed subset D ✓ M
with the following property: There exists a smooth function f 2 C• (M, R) such that
0 is a regular value of f , and

D = {p 2 M| f (p)  0}.
124 6 Differential forms

We do not consider f itself as part of the definition of D, only the existence of f

is required. The interior of a region with boundary, given as the largest open subset
contained in D, is int(D) = {p 2 M| f (p) < 0, and the boundary itself is

∂ D = {p 2 M| f (p) = 0},

a codimension 1 submanifold (i.e., hypersurface) in M.

Example 6.1. The region with bounday defined by the function f 2 C• (R2 ), given
by f (x, y) = x2 + y2 1, is the unit disk D ✓ R2 ; its boundary is the unit circle.

Example 6.2. Recall that for 0 < r < R, zero is a regular value of the function on R3 ,
p
f (x, y, z) = z2 + ( x2 + y2 R)2 r2 .

The corresponding region with boundary D ✓ R3 is the solid torus, its boundary is
the torus.

Recall that we are considering D inside an oriented manifold M. The boundary

∂ D may be covered by oriented submanifold charts (U, j), in such a way that ∂ D is
given in the chart by the condition x1 = 0, and D by the condition x1  0: ⇤⇤

j(U \ D) = j(U) \ {x 2 Rm | x1  0}.

(Indeed, given an oriented submanifold chart for which D lies on the side where
x1 0, one obtains a region chart by composing with the orientation-preserving co-
ordinate change (x1 , . . . , xm ) 7! ( x1 , x2 , x3 . . . , xm ).) We call oriented submanifold
charts of this kind ‘region charts’.††
Lemma 6.6. The restriction of the region charts to ∂ D form an oriented atlas for
∂ D.

Proof. Let (U, j) and (V, y) be two region charts, defining coordinates x1 , . . . , xm
and y1 , . . . , ym , and let F = y j 1 : j(U \ V ) ! y(U \ V ), x 7! y = F(x). It
restricts to a map

F1 : {x 2 j(U \V )| x1 = 0} ! {y 2 y(U \V )| y1 = 0}.

Since y1 > 0 if and only if x1 > 0, the change of coordinates satisfies

∂ y1 ∂ y1
> 0, = 0, for j > 0.
∂ x1 x1 =0 ∂xj x1 =0

⇤⇤ Note that while we originally defined submanifold charts in such a way that the last m
k coordinates are zero on S, here we require that the first coordinate be zero. It doesn’t
matter, since one can simply reorder coordinates, but works better for our description of
the ‘induced orientation’.
†† This is not a standard name.
 6.11 Stokes’ theorem 125

Hence, the Jacobian matrix DF(x)|x1 =0 has a positive (1, 1) entry, and all other en-
tries in the first row equal to zero. Using expansion of the determinant across the first
row, it follows that

∂ y1
det(DF(0, x2 , . . . , xm )) = det(DF 0 (x2 , . . . , xm )).
∂ x1 x1 =0

which shows that det(DF 0 ) > 0.

In particular, ∂ D is again an oriented manifold. To repeat: If x1 , . . . , xm are local

coordinates near p 2 ∂ D, compatible with the orientation and such that D lies on the
side x1  0, then x2 , . . . , xm are local coordinates on ∂ D. This convention of ‘induced
orientation’ is arranged in such a way that the Stokes’ theorem holds without extra
signs.
For an m-form w such that supp(w) \ D is compact, the integral
Z
w
D

is defined similar to the case of D = M: One covers D \ supp(w) by finitely many

submanifold charts (Ui , ji ) with respect to ∂ D (this includes charts that are entirely
in the interior of D), and puts
Z Z
w =Â ci w
D D\Ui

where the ci are supported in Ui and satisfy Âi ci over D \ supp(w). By the same
argument as for D = M, this definition of the integral is independent of the choice
made.
Theorem 6.3 (Stokes’ theorem). Let M be an oriented manifold of dimension m,
and D ✓ M a region with smooth boundary ∂ D. Let a 2 W m 1 (M) be a form of
degree m 1, such that supp(a) \ D is compact. Then
Z Z
da = a.
D ∂D
R
∂ D i a, where i : ∂ D ,! M is the
As explained above, the right hand side means ⇤

inclusion map.
Proof. We will see that Stokes’ theorem is just a coordinate-free version of the fun-
damental theorem of calculus. Let (Ui , ji ) for i = 1, . . . , r be a finite collection of re-
gion charts covering supp(a) \ D. Let c1 , . . . , cr 2 C• (M) be functions with ci 0,
supp(ci ) ✓ Ui , and such that c1 + . . . + cr is equal to 1 on supp(a) \ D. (E.g., we
may take U1 , . . . ,Ur together with U0 = M\supp(w) as an open covering, and take
the c0 , . . . , cr 2 C• (M) to be a partition of unity subordinate to this cover.) Since
Z r Z Z r Z
da = Â d(ci a), a=Â ci a,
D i=1 D ∂D i=1 ∂ D
126 6 Differential forms

it suffices to consider the case that a is supported in a region chart.

Using the corresponding coordinates, it hence suffices to prove Stokes’ theorem
for the case that a 2 W m 1 (Rm ) is a compactly supported form in Rm , and D = {x 2
Rm |x1  0}. That is, a has the form
m
a = Â fi dx1 ^ · · · dx
ci ^ · · · ^ dxm ,
i=1

with compactly supported fi where the hat means that the corresponding factor is to
be omitted. Only the i = 1 term contributes to the integral over ∂ D = Rm 1 , and
Z Z
a= f1 (0, x2 , . . . , xm ) dx2 · · · dxm .
Rm 1

On the other hand,

⇣ m
∂ fi ⌘ 1
da = Â( 1)i+1
∂ xi
dx ^ · · · ^ dxm
i=1

Let us integrate each summand over the region D given by x1  0. For i > 1, we have
Z • Z •Z 0
∂ fi
··· (x1 , . . . , xm )dx1 · · · dxm = 0
• • • ∂ xi
where we used Fubini’s theorem to carry out the xi -integration first, and applied the
fundamental theorem of calculus to the xi -integration (keeping the other variables
fixed, the integrand is the derivative of a compactly supported function). It remains
to consider the case i = 1. Here we have, again by applying the fundamental theorem
of calculus,
Z Z • Z •Z 0
∂ f1
da = ··· (x1 , . . . , xm )dx1 · · · dxm
D • • • ∂ x1
Z • Z • Z
= ··· fm (0, x2 , . . . , xm )dx2 · · · dxm = a
• • ∂D

t
u
As a special case, we have
Corollary 6.1. Let a 2 W m 1 (M) be a compactly supported form on the oriented
manifold M. Then Z
da = 0.
M
Note that it does not suffice that da has compact support. For example, if f (t) is a
function
R
with f (t) = 0 for t < 0 and f (t) = 1 for t > 0, then d f has compact support,
but R d f = 1.
A typical application of Stokes’ theorem shows that for a closed form w 2
W k (M), the integral of w over an oriented compact submanifold does not change
with smooth deformations of the submanifold.
 6.11 Stokes’ theorem 127

Theorem 6.4. Let w 2 W k (M) be a closed form on a manifold M, and S a compact,

oriented manifold of dimension k. Let F 2 C• (R ⇥ S, M) be a smooth map, thought
of as a smooth family of maps

Ft = F(t, ·) : S ! M.

Then the integrals Z

Ft⇤ w
S
do not depend on t.
If Ft is an embedding, then this is the integral of w over the submanifold Ft (S) ✓ M.
Proof. Let a < b, and consider the domain D = [a, b] ⇥ S ✓ R ⇥ S. The boundary ∂ D
has two components, both diffeomorphic to S. At t = b the orientation is the given
orientation on S, while at t = a we get the opposite orientation. Hence,
Z Z Z Z Z
0= F ⇤ dw = dF ⇤ w = F ⇤w = Fb⇤ w Fa⇤ w.
D D ∂D S S
R R
S Fb w S Fa w.
Hence ⇤ = ⇤ t
u
Remark 6.5. Note that if one member of this family of maps, say the map F1 , takes
values in a k 1-dimensional submanifold (for instance, if F1 is a constant map), then
F1⇤ w = 0. (Indeed, the assumption means that F1 = j F10 , where j is the inclusion of
a k 1-submanifold and F10 takes
R ⇤
values in that submanifold. But j⇤ w = 0 for degree
reasons.) It then follows that S Ft w = 0 for all t.
Given a smooth map j : S ! M, one refers to a smooth map F : R ⇥ S ! M
with F0 = j as an ‘smooth deformation’ (or ‘isotopy’) of j. We say that j can be
smoothly deformed into j 0 if there exists a smooth isotopy F with j = F0 and j 0 =
F1 . The theorem shows that if S is oriented, and if there is a closed form w 2 W k (M)
with Z Z
j ⇤ w 6= (j 0 )⇤ w
S S
then j cannot be smoothly deformed into j 0 . This observation has many applica-
tions; here are some of them. ‡‡
Example 6.3. Suppose j : S ! M R
is a smooth map, where S is oriented of dimension
k, and w 2 W k (M) is closed. If S j ⇤ w 6= 0, then j cannot smoothly be ‘deformed’
into a map taking values in a lower-dimensional submanifold. (In particular it cannot
be deformed into a constant map.) Indeed, if j 0 takes values in a lower-dimensional
submanifold, then j 0 = j j10 where j is the inclusion of that submanifold. But then
j⇤ w = 0, hence (j 0 )⇤ w = 0. For instance, the inclusion j : S2 ! M = R3 \{0} cannot
be smoothly deformed inside M so that j 0 would take values in R2 \{0} ✓ R3 \{0}.
‡‡ You may wonder if it is still possible to find a continuous deformation, rather than smooth.
It turns out that it doesn’t help: Results from differential topology show that two smooth
maps can be smoothly deformed into each other if and only if they can be continuously
deformed into each other.
128 6 Differential forms

Example 6.4 (Winding number). Let w 2 W 2 (R2 \{0}) be the 1-form

1
w= (xdy ydx)
x2 + y2
In polar coordinates x = r cos q , y = r sin q , one has that w = dq . Using this fact
one sees that w is closed (but not exact, since q is not a globally defined function on
R2 \{0}.) Hence, if
g : S1 ! R2 \{0}
is any smooth map (a ‘loop’), then the integral
Z
g ⇤w
S1

does not change under deformations (isotopies) of the loop. In particular, g cannot
be deformed into a constant map, unless the integral is zero. The number
Z
1
w(g) = g ⇤w
2p S1

is the winding number of g. (One can show that this is always an integer, and that
two loops can be deformed into each other if and only if they have the same winding
number.)
Example 6.5 (Linking number). Let f , g : S1 ! R3 be two smooth maps whose im-
ages don’t intersect, that is, with f (z) 6= g(w) for all z, w 2 S1 (we regard S1 as the
unit circle in C). Define a new map
f (z) g(w)
F : S1 ⇥ S1 ! S2 , (z, w) 7! .
|| f (z) g(w)||

On S2 , we have a 2-form w of total integral 1. It is the pullback of

1
xdy ^ dz ydx ^ dz + zdx ^ dy 2 W 2 (R3 )
4p
to the 2-sphere. The integral
Z
L( f , g) = F ⇤w
S1 ⇥S1

is called the linking number of f and g. (One can show that this is always an integer.)
Note that if it is possible to deform one of the loops, say f , into a constant loop
through loops that are always disjoint from g, then the linking number is zero. In his
case, we consider f , g as ‘unlinked’.
Example 6.6. Let M be a compact,R
connected oriented manifold. There exists a dif-
ferential form w on M such that M w = 1. (Note that w cannot be exact, since other-
wise the integral would be zero, by Stokes.) Given another compact, oriented mani-
fold N of the same dimension (e.g., M itself), and a smooth map F : N ! M, we can
define the degree of F
 6.12 Volume forms 129
Z
deg(F) = F ⇤ w.
N
The degree is invariant under deformations of F. It turns out that it is independent
of the choice of w, and is always an integer. In the preceding example, the linking
number was defined as the degree of a map S1 ⇥ S1 ! S2 obtained from f , g.

6.12 Volume forms

A top degree differential form G 2 W m (M) is called a volume form if it is non-
vanishing everywhere: Gp 6= 0 for all p 2 M. In a local coordinate chart (U, j), this
means that
(j 1 )⇤G = f dx1 ^ · · · ^ dxm
where f (x) 6= 0 for all x 2 j(U).

Example 6.7. The Euclidean space Rn has a standard volume form G0 = dx1 ^ · · · ^
dxn . Suppose S ✓ Rn is a submanifold of dimension n 1, and X a vector field that
is nowhere tangent to S. Let i : S ! Rn be the inclusion. Then

G := i⇤ iX G0 2 W n 1 (S)

is a volume form. For instance, if S is given as a level set f 1 (0), where 0 is a regular

value of f , then the gradient vector field

n
∂f ∂
 ∂ xi ∂ xi
i=1

has this property.

Exercise: Verify the claim that G := i⇤ iX G0 is a volume form.

Example 6.8. Let i : Sn ! Rn+1 be the inclusion of the standard n-sphere. Let X =
Âni=0 xi ∂∂xi . Then
n
iX (dx0 ^ · · · ^ dxn ) = Â ( 1)i xi dx1 ^ · · · dxi 1
^ dxi+1 ^ · · · ^ dxn
i=0

pulls back to a volume form on Sn .

Lemma 6.7. A volume form G 2 W m (M) determines an orientation on M, by taking

as the oriented charts those charts (U, j) such that (j 1 )⇤G = f dx1 ^ · · · ^ dxm with
f > 0 everywhere on F(U).
130 6 Differential forms

Proof. We have to check that the condition is consistent. Suppose (U, j) and (V, y)
are two charts, where (j 1 )⇤G = f dx1 ^ · · · ^ dxm and (y 1 )⇤G = g dy1 ^ · · · ^ dym
with f > 0 and g > 0. If U \ V is non-empty, let F = y j 1 : j(U) ! y(V ) be
the transition function. Then
1 ⇤ 1 ⇤
F ⇤ (y ) G |U\V = (j ) G |U\V ,

hence
g(F(x)) J(x) dx1 ^ · · · ^ dxm = f (x) dx1 ^ · · · ^ dxm .
where J is that Jacobian determinant of the transition map F = y j 1 . Hence f =
J (g F) on j(U \V ). Since f > 0 and g > 0, it follows that J > 0. Hence the two
charts are oriented compatible. tu

Theorem 6.5. A manifold M is orientable if and only if it admits a volume form. In

this case, any two volume forms compatible with the orientation differ by an every-
where positive smooth function:

G 0 = fG , f > 0.

Proof. As we saw above, any volume form determines an orientation. Conversely, if

M is an oriented manifold, there exists a volume form compatible with the orienta-
tion: Let {(Ua , ja )} be an oriented atlas on M. Then each

Ga = ja⇤ (dx1 ^ · · · ^ dxm ) 2 W m (Ua )

is a volume form on Ua ; on overlaps Ua \Ub these are related by the Jacobian deter-
minants of the transition functions, which are strictly positive functions. Let {ca } be
a locally finite partition of unity subordinate to the cover {Ua }, see Appendix A.4.
The forms ca Ga have compact support in Ua , hence they extend by zero to global
forms on M (somewhat imprecisely, we use the same notation for this extension).
The sum
G = Â ca Ga 2 W m (M)
a

is a well-defined volume form. Indeed, near any point p at least one of the summands
is non-zero; and if other summands in this sum are non-zero, they differ by a positive
function.

For a compact manifold M with a given volume form G 2 W m (M), one can define
the volume of M, Z
vol(M) = G.
M
Here the orientation used in the definition of the integral is taken to be the orientation
given by G . Thus vol(M) > 0.
Note that volume forms are always closed, for degree reasons (since W m+1 (M) =
0). But on a compact manifold, they cannot be exact:
 6.12 Volume forms 131

Theorem 6.6. Let M be a compact manifold with a volume form G 2 W m (M). Then
G cannot be exact.
R
Proof. We Rhave vol(M) = MG > 0. But if G were exact, then Stokes’ theorem
would give M G = 0.

Of course, the compactness of M is essential here: For instance, dx is an exact volume

form on R.