Chapter 4: Integration

4.1: Stokes’ Theorem on Chains

If U ⊂ Rn is an open subset and g : U → R a real-valued function on U , then the
“n-tuple” Riemann integral of g over U can be defined through a straightforward gener-
alisation of the definitions given in Chapters 1.1 and 1.3, which deal with cases
R n = 2
and n = 3, respectively. Provided this integral exists, we will denote it by U g. We
note that the various versions of Fubini’s Theorem (for example Theorem 1.18) discussed
for triple integrals generalise to arbitrary dimensions, and that these are what we will
use for calculations and many other applications to treat the n-dimensional integral as
iterated integration in 1-dimension. The Theorem for calculating integrals via a change
of variables (Theorem 1.21) also generalises, and can be written as:

Theorem 4.1 (Change of Variables): Let U, V ⊂ Rn be open subsets, and f : U → V

a diffeomorphism. Then for any integrable scalar function g : V → R on V , the function
g ◦ f : U → R is integrable and their integrals are related by
Z Z
g = (g ◦ f )|det(Df )|.
V U

Definition 4.2: For U ⊂ Rn an open subset, let ω ∈ Ωn (U ) be a differential n-form on

U . Then we have
ω = gdx1 ∧ . . . ∧ dxn
for a uniquely determined, differentiable scalar function g : U → R. If g is integrable, we
say that ω is integrable and define
Z Z
ω := g.
U U

Before using this to define integrals of differential forms over chains, note the behaviour
of this integral with respect to pullbacks:

Lemma 4.3 (Integration and Pullback in Rn ): Let U, V ⊂ Rn be open subsets, and

f : U → V a diffeomorphism. If ω ∈ Ωn (V ) is an integrable differential n-form on V ,
then its pullback to U is also integrable. Moreover, if det(Df (p)) > 0 for all p ∈ U , then
Z Z
∗
f ω= ω.
U V

If this is the case, we say that the diffeomorphism f is orientation-preserving.

Proof: Write ω = gdx1 ∧. . .∧dxn for g ∈ C ∞ (V ). By hypothesis, ω ∈ Ωn (V ) is integrable,

which by definition just means that g is integrable over V . Thus, Theorem 4.1 implies
that g ◦ f : U → R is integrable and that
Z Z
g = (g ◦ f )|det(Df )|.
V U

1
However, by Question 3(b) of the Chapter 3.3 TUT Problems, we have

f ∗ ω = f ∗ (gdx1 ∧ . . . ∧ dxn ) = (g ◦ f )det(Df )dx1 ∧ . . . ∧ dxn ∈ Ωn (U ),

and thus by definition 4.2 we have

Z Z
∗
f ω = (g ◦ f )det(Df ).
U U

This shows that f ∗ ω ∈ Ωn (U ) is integrable, and if det(Df (p)) > 0 for all p ∈ U , then
combining the two identities (since |det(Df )| = det(Df )) shows:
Z Z Z
∗
f ω= g= ω,
U V V

as claimed. (QED)

Note: By the assumption that f : U → V is a diffeomorphism, we must have either

det(Df (p)) > 0 or det(Df (p)) < 0 for each p ∈ U . If we assume that U ⊂ Rn is a
connected subset (in this context, that means we assume that any two points in U can
be connected by a continuous path γ : [a, b] → U ), then the continuity of the function
det(Df ) : U → R in fact implies that one of these two alternatives is true for all of
U : either det(Df (p)) > 0 for all p ∈ U (in which case f is orientation-preserving) OR
det(Df (p)) < 0 for all p ∈ U (then we say f is orientation-reversing). So Lemma 4.3
tells us that the integral of differential n-forms on open subsets of Rn is invariant un-
der pullback orientation-preserving diffeomorphisms, and changes sign under pullback by
orientation-reversing diffeomorphisms.

Definition 4.4: Let M ⊂ Rn+r be a n-manifold, and k = 0, 1, 2, . . . a non-negative

integer. Denote by [0, 1]k ⊂ Rk the unit cube of dimension k, i.e.

[0, 1]k = {x ∈ Rk | 0 ≤ xi ≤ 1 for all i = 1, . . . , k}.

(a) A k-cube in M is any differentiable map c : [0, 1]k → M .

(b) A k-chain in M is a (finite) formal sum of k-cubes in M with integer coefficients,

with addition and multiplication by integers formally defined according to the usual rules.

Example 4.5: (a) For each k = 0, 1, 2, . . ., the standard k-cube in Rk , denoted I k , is

given by the identity map I k : [0, 1]k → Rk . In other words, I k (x) = x ∈ Rk for all
x ∈ [0, 1]k ⊂ Rk . This seems silly, but it will turn out to be very useful.

(b) For M ⊂ Rn+r any manifold, a 0-cube in M is simply a point in M , since a map
c : [0, 1]0 = {0} → M just picks out a point. Hence, a 0-chain in M is simply a collection
of points in M , each with an integer coefficient. Again, this may seem trivial but it has
some use and makes intuitive sense, because the 0-chains are the objects over which we
intend to define integration of 0-forms. Since a 0-form on M is just a scalar-valued func-
tion on M , it is clear how to integrate it over a 0-chain to get a number: simply evaluate

2
the function at each of the points given by the 0-chain, multiply by the corresponding
integer coefficient, and take the sum.

(c) For M ⊂ Rn+r again a manifold, a 1-cube in M is a differentiable map c : [0, 1] →

M , i.e. a differentiable path. If ω ∈ Ω1 (M ) is a differential 1-form on M , it is not too hard
to guess how we should defined the integral of ω over the 1-sube c: Write the pullback
c∗ ω ∈ Ω1 ([0, 1]) as c∗ ω = gdt for some scalar function g : [0, 1] → R (this can always be
done, for example using Lemma 3.18 for the case k = n = 1). From the calculation done
in Example 3.7 of Chapter 3.1, we see that this generalises the definition of a vector path
integral from MC.

(d) Consider the standard k-cube in Rk , I k , as defined in (a). Then if k > 0, for each
i = 1, . . . , k and each fixed value α ∈ [0, 1], we can define a (k − 1)-cube in Rk , denoted
k k
I(i,α) , as the map I(i,α) : [0, 1]k−1 → Rk given by
k
I(i,α) (x) = (x1 , . . . , xi−1 , α, xi , . . . , xk−1 ), x = (x1 , . . . , xk−1 ) ∈ [0, 1]k−1 .

In other words, we just insert α as the ith coordinate. The case where α = 0 or α = 1 are
k
special cases, and then we refer to the (k − 1)-cube I(i,α) as the (i, α)-face of I k . Similarly,
if c : [0, 1]k → M is a k-cube in a manifold M , we can define its (i, α)-face to be the
k
(k − 1)-cube c(i,α) = c ◦ I(i,α) if c : [0, 1]k → M .

Definition 4.6 (Boundary of a Chain): (a) For I k the standard k-cube in Rk , its
boundary, denoted ∂I k , is the (k − 1)-chain in Rk given by the following formal sum of
faces:
X k X
k
∂I := (−1)i+α I(i,α)
k
.
i=1 α=0,1

(b) If c : [0, 1]k → M is a k-cube in a manifold M , its boundary, denoted ∂c, is the
(k − 1)-chain in M given by
k X
X
∂c := (−1)i+α c(i,α)
i=1 α=0,1
k X
X
= (−1)i+α c ◦ I(i,α)
k
.
i=1 α=0,1

(c) Finally, if c = N A A k
P
A=1 gA c is a general k-chain in M (meaning c : [0, 1] → M and
gA ∈ Z for each A = 1, . . . , N ), then we define the boundary by extending it linearly:
N
X
∂c = gA ∂cA ,
A=1

which is a (k − 1)-chain in M .

The boundary of chains has a number of interesting and important properties, in-
cluding the fact that ∂(∂c) = 0 for any k-chain c. However, for our purposes, the most

3
important thing is the relationship to integration, which gives us what is called “Stokes’
on Chains”. First we define integration of a differential k-form on M over a k-chain in
M:

Definition 4.7 (Integration over Chains): Let M ⊂ Rn+r be a n-manifold, 0 ≤ k ≤ n,

and ω ∈ Ωk (M ) a differential k-form. If c = A=1 gA cA is a k-chain in M , then we define
P
the integral of ω over c as
Z N
X Z
ω := gA ω
c A=1 cA

XN Z
:= gA (cA )∗ ω,
A=1 [0,1]k

where the last integral denotes the integral of (cA )∗ ω ∈ Ωk ([0, 1]k ) as determined by Def-
inition 4.2.

Example: Let M = S 1 ⊂ R2 − {(0, 0)}, and let ω ∈ Ω1 (M ) be the differential 1-form

given by restricting to M the differential 1-form, also denoted by ω,
−y x
ω= dx + 2 dy,
x2 +y 2 x + y2

(see Question 1 of Chapter 3.1 TUT Problems), which is defined everywhere on R2 except
at (0, 0). For any integer A ∈ Z, let cA : [0, 1] → M be the 1-cube given by cA (t) =
(cos(2Aπt), sin(2Aπt)) ∈ M for t ∈ [0, 1]. Then using a calculation as in the solution of
that TUT question, we see that

(cA )∗ ω = 2Aπdt ∈ Ω1 ([0, 1]),

and hence
Z Z
ω= 2Aπdt
cA [0,1]

= 2Aπ.

Theorem 4.8 (Stokes’ on Chains): Let M ⊂ Rn+r be a n-manifold, and k any integer
between 1 and n. If c is a k-chain in M and ω ∈ Ωk−1 (M ) is a differential (k − 1)-form,
then Z Z
dω = ω.
c ∂c

Proof: Because of how our definitions have been set up, the main calculation we need to do
is proving this identity in the special case where M = U is an open subset of Rk containing
[0, 1]k and c = I k is the standard k-cube in Rk . In this case, using linearity it suffices to
consider a differential (k − 1)-form ω ∈ Ωk−1 (U ) of the form ω = gdxi1 ∧ . . . ∧ dxik−1 for
some g ∈ C ∞ (U ) and some strictly increasing set of indices 1 ≤ i1 < . . . < ik−1 ≤ k. Any
such index set must be of the form (i1 , . . . , ik−1 ) = (1, . . . , î, . . . , k) for some i ∈ {1, . . . , k},

4
where the “hat” over the number i denotes the fact that it has been omitted. Hence, using
the same notation, we compute,

dω = dg ∧ dx1 ∧ . . . ∧ dx
ci ∧ . . . ∧ dxk
∂g i
= dx ∧ dx1 ∧ . . . ∧ dx
ci ∧ . . . ∧ dxk
∂xi
∂g
= (−1)i−1 i dx1 ∧ . . . ∧ dxi ∧ . . . ∧ dxk .
∂x
Therefore, Z Z Z
k ∗ i−1 ∂g
dω = (I ) dω = (−1) ,
Ik [0,1]k [0,1]k ∂xi
since (I k )∗ dω = dω for the standard k-cube I k . The integral over [0, 1]k can be carried
out using iterated integration in any order we want to, by Fubini’s Theorem. Using the
Fundamental Theorem of Calculus, the integral with respect to the variable xi is
Z 1
∂g i
i
dx = g(x1 , . . . , xi , 1, xi+1 , . . . , xk ) − g(x1 , . . . , xi−1 , 0, xi+1 , . . . , xk )
0 ∂x
= g(y 1 , . . . , y i−1 , 1, y i , . . . , y k−1 ) − g(y 1 , . . . , y i−1 , 0, y i , . . . , y k−1 )
k k
= g(I(i,1) (y)) − g(I(i,0) (y), y ∈ [0, 1]k−1 .

Hence, we see that

Z Z Z
i−1 k i k
dω = (−1) g◦ I(i,1) + (−1) g ◦ I(i,0) . (1)
Ik [0,1]k−1 [0,1]k−1

On the other hand, using definitions 4.6 and 4.7, we have

Z k X
X Z
ω= (−1) j+α k
(I(j,α) )∗ ω.
∂I k j=1 α=0,1 [0,1]k−1

For each (j, α), we calculate the pullback of ω = gdx1 ∧ . . . ∧ dx

ci ∧ . . . ∧ dxk :

k
(I(j,α) )∗ ω = (I(j,α)
k
)∗ g(I(j,α)
k
)∗ (dx1 ) ∧ . . . ∧ dx
ci ∧ . . . (I k )∗ (dxk ), by Thm 3.12(e)
(j,α) (2)
k 1 k ci ∧ . . . ∧ d(xk ◦ I k ), by Thm 3.28.
= (g ◦ I(j,α) )d(x ◦ I(j,α) ) ∧ . . . ∧ dx (j,α) (3)

But for each l = 1, . . . , k, we have


l
x
 if l < j
l k
x ◦ I(j,α) = α if l = j

 l−1
x if l > j.
k
This implies that (I(j,α) )∗ ω = 0 whenever j 6= i, since on of the factors in the wedge
product (3) will be
d(xk ◦ I(j,α)
k
) = dα = 0.
Hence, the only non-zero terms are those for j = i, α = 0, 1, in which case we have
k
(I(i,α) )∗ ω = (g ◦ I(i,α)
k
)dx1 ∧ . . . ∧ dxk−1 ∈ Ωk−1 ([0, 1]k−1 ).

5
This shows that Z X Z
i+α k
ω= (−1) g ◦ I(i,α) ,
∂I k α=0,1 [0,1]k−1
R R
which proves that ∂I k
ω= Ik
dω by comparison with (4).

Now, if M is a manifold, c = A gA cA is a k-chain in M (so each cA : [0, 1]k → M is

P
a k-cube), and ω ∈ Ωk−1 (M ), note that
Z X Z
dω = gA (cA )∗ dω, by Def. 4.7
c A Ik
X Z
= gA d(cA )∗ ω, by Thm. 3.28.
A Ik

On the other hand, by Def. 4.6,

X
∂c = gA ∂cA
A
k X
!
X X
= gA (−1)i+α (cA ◦ I(i,α)
k
) .
A i=1 α=0,1

and so, by Definition 4.7,

Z k X Z !
X X
ω= gA (−1) i+α A
(c ◦ k
I(i,α) )∗ ω
∂c A i=1 α=0,1 [0,1]k−1

k X Z !
X X
= gA (−1)i+α k
(I(i,α) )∗ ((cA )∗ ω)
A i=1 α=0,1 [0,1]k−1
X Z
= gA (cA )∗ ω.
A ∂I k

(Here we used the fact that (cA ◦ I(i,α)

k
)∗ = (I(i,α)
k
)∗ ◦ (cA )∗ , which is a general property of
the pullback given in the Lemma below.) Using the calculation we did above,
Z Z
A ∗
d(c ) ω = (cA )∗ ω
Ik ∂I k
R R
for each A, and hence c
dω = ∂c
ω, as claimed. (QED)

Lemma 4.9: If M, N and P are manifolds, and f : M → N , g : N → P are differentiable

maps, then for any ω ∈ Ωk (P ) we have
(g ◦ f )∗ ω = f ∗ (g ∗ ω) ∈ Ωk (M ).

Proof: First consider vector spaces V, W and U , and linear maps L : V → W , T : W → U .

For ω ∈ Λk (U ) a linear k-form, and v1 , . . . , vk ∈ V any vectors, we have
[(T ◦ L)∗ ω](v1 , . . . , vk ) = ω(T (Lv1 ), . . . , T (Lvk ))
= T ∗ ω(Lv1 , . . . , Lvk )
= [L∗ (T ∗ ω)](v1 , . . . , vk )

6
which shows that for linear maps we have (T ◦ L) = L∗ ◦ T ∗ .

Now let p ∈ M be an arbitrary point. Then by definition we have

[(g ◦ f )∗ ω](p) = [D(g ◦ f )(p)]∗ (ω(g ◦ f (p)))

= [Dg(f (p)) ◦ Df (p)]∗ (ω(g(f (p)))), by the Chain Rule
= Df (p)∗ (Dg(f (p))∗ (ω(g(f (p)))), by the linear case above
= Df (p)∗ [g ∗ ω(f (p))]
= [f ∗ (g ∗ ω)](p)

which shows that (g ◦ f )∗ ω = f ∗ (g ∗ ω) ∈ Ωk (M ).

7
 Chapter 4.2: Stokes’ Theorem on Manifolds
If M ⊂ Rn+r is a manifold of dimension n, we need to explain how to integrate a
differential n-form ω ∈ Ωn (M ) over M . The simplest case is when all the points of M
can be described with a single parametrisation:

Definition 4.10: Let M ⊂ Rn+r be a (connected) n-manifold, and suppose that r : U →

Rn+r is a regular parametrisation of M with M = r(U ). Then for a differential n-form
ω ∈ Ωn (M ), we define its integral over M as
Z Z
ω := r∗ ω,
M U

where the right-hand integral, assuming it converges, is given by Definition 4.2.

R
The following Lemma explains the dependence of M ω on the choice of parametrisa-
tion r, and also leads naturally to the notion of orientation, which is needed to define the
integral of differential forms over manifolds requiring more than one local parametrisation
to describe all points. Note that here, and for the rest of this section, we will assume that
all our manifolds, and the subsets used to parametrise them, are connected.

Lemma 4.11: Let rα : Uα → Rn+r and rβ : Uβ → Rn+r be regular parametrisations of

M with rα (Uα ) = rβ (Uβ ) = M . If the map

hαβ := r−1α ◦ rβ : Uβ → Uα

is orientation-preserving, then Uα r∗α ω = Uβ r∗β ω. If hαβ is orientation-reversing, then the

R R

two integrals have opposite signs.

Proof: First, note that by the definition of hαβ , we have rβ = rα ◦ hαβ . Using this (and
Lemma 4.9), we calculate

r∗β ω = (rα ◦ hαβ )∗ ω = h∗αβ (r∗α ω).

Now, hαβ : Uβ → Uα is a diffeomorphism (see Lemma 2.16, Chapter 2.4). If it is

orientation-preserving, then using Lemma 4.3,
Z Z Z
∗ ∗ ∗
rβ ω = hαβ (rα ω) = r∗α ω.
Uβ Uβ Uα

This proves the first statement of the lemma. Note that if hαβ is not orientation-
preserving, connectedness of Uβ implies that it must be orientation-reversing, and in
that case the second assertion follows by a similar argument, using the Note following the
proof of Lemma 4.3.

Definition 4.12: Let M ⊂ Rn+r be a n-manifold. We say that M is orientable iff it is

possible to find a collection {rα : Uα → Rn+r }α∈I of regular local parametrisations of M

8
such that:
S
(i) M = α∈I rα (Uα ); and

(ii) For any α, β ∈ I, if rα (Uα ) ∩ rβ (Uβ ) 6= ∅, then the change of coordinates map
hαβ = r−1
α ◦ rβ is orientation-preserving.

If M is orientable, then an orientation is fixed by choosing a collection of regular local

parametrisations satisfying these conditions (abbreviated as {(rα , Uα )}α∈I ). Once an ori-
entation has been fixed, we say a regular local parametrisation r : U → Rn+r of M is
positively-oriented iff adding it to the collection would not disturb condition (ii). In other
words, for any α ∈ I, if r(U ) ∩ rα (Uα ) 6= ∅, then hα := r−1
α ◦ r is orientation-preserving.

Definition 4.13 (Integration on Oriented Manifolds): Suppose M ⊂ Rn+r is an

orientable manifold of dimension n, and let an orientation on M be fixed. If ω ∈ Ωn (M )
is a differential n-form, then its integral over M , provided it converges, is defined via the
following steps:

(i) If there is a positively-oriented regular parametrisation r : U → M such that

supp(ω) := {p ∈ M | ω(p) 6= 0} ⊂ r(U ),

then we define Z Z
ω= r∗ ω
M U
for any such local parametrisation.

(ii) If it is possible, find a collection {φα }α∈I of differentiable, scalar-valued functions

φα ∈ C ∞ (M ), such that: (a) φα ≥ 0 everywhere and supp(φ P α ) ⊂ rα (Uα ); and (b) for
every p ∈ M , φα (p) = 0 for all but finitely many α ∈ I, and α∈I φα (p) = 1. Then define
Z XZ XZ
ω := φα ω = r∗α (φα ω).
M α∈I M α∈I Uα

Note 4.14: RFirst, it follows immediately from Lemma 4.10 that in case (i) the value of
the integral M ω, provided it converges, does not depend on which positively-oriented
regular local parametrisation satisfying the condition is chosen. To justify the definition
in (ii) takes more work. We would need to prove that collections {φα }α∈I with these prop-
erties can actually be found, and that the value of the integral is independent of which
such collection is chosen. Both things are true, but proving them is beyond our scope.
However, for those familiar with the notion of compactness (from Topology and/or Real
Analysis), it is a good challenge (optional) to try to prove this for a compact manifold
M . A collection {φα } with these properties is called a partition of unity.

Before we prove Stokes’ Theorem for Manifolds, note that if M ⊂ Rn+r is a n-manifold
with non-empty boundary ∂M , then the boundary ∂M ⊂ Rn+r is a manifold of dimension
(n − 1). Since we want to be able to integrate differential (n − 1)-forms over ∂M , we need

9
to define the orientation that is induced on ∂M by an orientation of M . The following
definition does this, and with some reflection we see that it is a generalisation of the
“right-hand rule” for determining the orientation on the boundary of a surface S ⊂ R3
(see Definition 1.7(c), Ch. 1.1):

Definition 4.15 (Induced Orientation): Let M ⊂ Rn+r be an orientable n-manifold

with non-empty boundary ∂M , and suppose an orientation {(rα , Uα )}α∈I has been cho-
sen. The induced orientation on ∂M is determined as follows: For any p ∈ ∂M , note that
Tp ∂M ⊂ Tp M is a linear subspace of dimension (n − 1), and so we can choose a non-zero
tangent vector ν ∈ Tp M that is perpendicular to Tp ∂M and “outward-pointing”, meaning
that we can choose a differentiable curve γ : (−ε, ε) → Rn+r with γ(0) = p, γ 0 (0) = ν,
and γ(t) ∈ M for t ≤ 0 but γ(t) ∈/ M for t > 0. We call ν an outward-pointing normal to
∂M . Then we say that a regular local parametrisation s : V → ∂M of ∂M (in particular,
V ⊂ Rn−1 ) near p is positively-oriented iff s can be obtained from a positively-oriented
regular local parametrisation r : U → M of M near p with D1 r(q) = ν(r(q)) for all
q ∈ ∂U , by letting V = {y ∈ Rn−1 | (0, y) ∈ ∂U } and s = r ◦ j, where j : V → ∂U ⊂ U is
the map j : y 7→ (0, y).

Note: The way to think about this is as follows: Think of a local parametrisation
s : V → ∂M as defining local coordinates (y 1 , . . . , y n−1 ) on the boundary ∂M . Then to get
local coordinates on M we need to add one “extra coordinate”, which we denote as y 0 , and
we adopt the convention that the coordinate y 0 is going from “inside of” M to “outside
of” M . This means that y 0 is defined in such a way that y 0 ≤ 0 iff (y 0 , y 1 , . . . , y n−1 ) ∈ M ;
y 0 = 0 iff (y 0 , y 1 , . . . , y n−1 ) ∈ ∂M , and for y 0 > 0, (y 0 , y 1 , . . . , y n−1 ) ∈
/ M . In this way
of looking at things, Definition 4.14 just says that the local coordinates (y 1 , . . . , y n−1 ) on
∂M are positively-oriented if and only if the local coordinates (y 0 , y 1 , . . . , y n−1 ) on M are
positively oriented.

Theorem 4.16 (Generalised Stokes’ Theorem): Let M ⊂ Rn+r be a compact, ori-

ented n-dimensional manifold with boundary ∂M , which is given the induced orientation.
If ω ∈ Ωn−1 (M ) is a differential form of degree n − 1, then
Z Z
ω= dω.
∂M M

Proof: We first prove the identity in two special cases (we denote by int(M ) = M − ∂M
the interior of M ):

Case I: There is a positively-oriented local parametrisation r : U → Rn+r of M such that

[0, 1]n ⊂ U and supp(ω) ⊂ r([0, 1]n ) ⊂ int(M ).

Case II: There is a positively-oriented local parametrisation r : U → Rn+r of M such

that [0, 1]n ⊂ U , supp(ω) ⊂ r([0, 1]n ), and

r([0, 1]n ) ∩ ∂M = r([0, 1]n−1 × {0}) (4)

1 n−1 i
= {r(y , . . . , y , 0) | 0 ≤ y ≤ 1}. (5)

10
Proof in Case I: We let c : [0, 1]n → M be the n-cube defined by restricting the local
parametrisation r to the unit cube [0, 1]n ⊂ U . Then, by assumption, supp(ω) ⊂ c([0, 1]n ).
It follows that supp(dω) ⊂ c([0, 1]n ) as well. Thus, by Definition 4.12(i),
Z Z Z
∗
dω = r (dω) = dω.
M U c

Now, if supp(ω) ⊂ c([0, 1]n ) ⊂ int(M ), it must be that ω(p) = 0 for all p ∈ c(∂[0, 1]n ). To
see this, note that for any p ∈ c(∂[0, 1]n ), the fact that pint(M ) means that we can find
a sequence of points (pn ) ⊂ M − c([0, 1]n ) that converges to p. Then ω(pn ) = 0 for all
n = 1, 2, . . ., which implies ω(p) = 0 by continuity of ω.
This means that c∗(i,α) ω = 0 for any face c(i,α) of c, and so ∂c ω = 0. On the other
R
R
hand, since supp(ω) ⊂ int(M ), we have ω(p) = 0 for all p ∈ ∂M , so ∂M ω = 0. Finally,
using these identities and Stokes’ on Chains (Thm 4.8), we get:
Z Z
dω = dω
M Zc
= ω
∂c
=0
Z
= ω.
∂M

n
R II: LetR c : [0, 1] → M be the n-cube defined by restricting r, as before. We
Proof in Case
still have M dω = c dω, as in Case I. The induced orientation on ∂M (Def. 4.14) was
chosen in such a way to ensure that the (n − 1)-cube

(−1)n c(n,0) : [0, 1]n−1 → ∂M

is positively-oriented. Also, since supp(ω) ⊂ c([0, 1]n ), an argument similar to the one
in the proof for Case I shows that ω(p) = 0 for all p ∈ c(∂[0, 1]n ) ∩ int(M ). But the
assumption (4) means that c([0, 1]n ) ∩ ∂M = c(n,0) ([0, 1]n−1 ), so if p ∈ c(∂[0, 1]n ), then
either p ∈ c(n,0) ([0, 1]n−1 ) or ω(p) = 0. Hence, c∗(i,α) ω = 0 for any face c(i,α) with (i, α) 6=
(n, 0). Using this, we calculate:
Z Z
n
ω = (−1) ω
∂M c(n,0)
n
X X Z
i+α
= (−1) ω
i=1 α=0,1 c(i,α)
Z
= ω
Z∂c
= dω
Zc
= dω.
M

Proof in general case: For a general differential (n − 1)-form ω, we will assume that
{rα : Uα → M }α∈I is a collection of positively-oriented local parametrisations of M ,

11
chosen so that for each α ∈ I, we have [0, 1]n ⊂ Uα and either rα ([0, 1]n ) ⊂ int(M ) or
rα ([0, 1]n )∩∂M = rα ([0, 1]n−1 ×{0}). Furthermore, we assume that a collection {φα }α∈I of
differentiable scalar-valued functions can be found that satisfy the conditions of Definition
4.12(ii), and with supp(φα ) ⊂ rα ([0, 1]n ) for all α ∈ I. (I am going to just ASSUME that
these exist for the sake of simplicity; in fact, it can be proven that these conditions can
always be satisfied, and in fact since M is compact the index set I can always be taken
to be finite.)
These assumptions mean that for each α ∈ I, the differential (n − 1)-form φα ω ∈
n−1
Ω (M ) fits into either Case I or Case II above, and hence we have
Z Z
φα ω = d(φα ω), α ∈ I.
∂M M
P P
On the other hand, since α∈I φα (p) = 1 for all p ∈ M , the function φ = α∈I φα is a
constant function, and therefore
X
dφ = dφα = 0 ∈ Ω1 (M ).
α∈I

From this, we derive:

X X
d(φα ω) = dφα ∧ ω + φα dω
α∈I α∈I
!
X X
= dφα ∧ω+ φα dω
α∈I α∈I
X
=0+ φα dω.
α∈I

Using the above, we calculate:

Z XZ
dω = φα dω, (Def. 4.12(ii))
M α∈I M
XZ
= d(φα ω)
α∈I M
XZ
= φα ω, (Cases I and II)
α∈I ∂M
Z
= ω, (Def. 4.12(ii)).
∂M

(QED)

4.1-4.2 TUT Problems

1. Let D = [0, 1]2 ⊂ R2 , let ω ∈ Ω1 (D) be a differential 1-form on D, which can be

uniquely expressed as ω = f dx + gdy for f, g ∈ C ∞ (D), and let I 2 : D → R2 be the
standard 2-cube (so I 2 (x, y) = (x, y) for all (x, y) ∈ D). Give an explicit proof of Stokes’
Theorem on Chains in this case:
Z Z
ω= dω,
∂I 2 I2

12
and show that in this case Stokes’ Theorem on Chains is equivalent to Green’s Theorem:
Z Z Z  
∂g ∂f
f dx + gdy = − dxdy.
∂D D ∂x ∂y

2. Let D = [0, 1]2 ⊂ R2 as in Question 1, let r : D → U be a 2-cube in U , for U some

open subset of R3 , and let ω ∈ Ω1 (U ) be a differential 1-form on U . Using Question 1
and the properties of the exterior derivative and the pullback, prove Stokes’ Theorem on
Chains in this case: Z Z
ω = dω.
∂r r

Now write ω = F dx + F dy + F dz for some uniquely determined F 1 , F 2 , F 3 ∈ C ∞ (U ).

1 2 3

Show that if we let S = r(D) be the surface parametrised by the 2-cube r, and ∂S = r(∂D)
its boundary, then in this case Stokes’ Theorem on Chains is equivalent to the Classical
Stokes’ Theorem: For any vector field F = (F 1 , F 2 , F 3 ) ∈ Vect(U ),
Z Z Z
F · dγ = (∇ × F ) · da.
∂S S

3. For this question, let U = {(x, y) | (x, y) 6= (0, 0)} ⊂ R2 , and let ω ∈ Ω1 (U ) be the
differential 1-form defined by
−y x
ω= dx + 2 dy.
x2
+y 2 x + y2
Given a differentiable (or even piecewise differentiable) curve γ : [0, 1] → U , define its
winding number, wind(γ), by Z
1
wind(γ) := ω.
2π γ
We say that two curves γ, ρ : [0, 1] → U are homotopic iff there is a differentiable (or even
piecewise differentiable) map H : [0, 1]2 → U satisfying the following:
ˆ H(t, 0) = γ(t) and H(t, 1) = ρ(t) for all t ∈ [0, 1];

ˆ H(0, s) = γ(0) = ρ(0) and H(1, s) = γ(1) = ρ(1) for all s ∈ [0, 1].

(a) For a given integer n, let γn : [0, 1] → U be the curve defined by

γn (t) = (cos(2nπt), sin(2nπt)), t ∈ [0, 1].

Show that wind(γn ) = n, and use Stokes’ Theorem on Chains to explain why this identity
shows that there is no scalar-valued function φ ∈ C ∞ (U ) = Ω0 (U ) such that dφ = ω.

(b) Prove that if γ and ρ are two homotopic curves, then wind(γ) = wind(ρ). In
particular, if n 6= m are two distinct integers, then the curves γn and γm as defined in (a)
are not homotopic.

4*. Let f (z) = z n + an−1 z n−1 + . . . + a1 z + a0 be a complex polynomial of degree n > 0.

This question outlines a proof of the Fundamental Theorem of Algebra using Stokes’ The-
orem on Chains. Throughout, we identify the complex numbers C with the Euclidean

13
plane R2 in the usual way, writing any complex number z ∈ C as z = x + iy, for x, y ∈ R,
and identifying z = (x, y). We let U = C − {0} = R2 − {(0, 0)} as in Question 3.

(a) For a given positive real number R > 0, define the curves γR,f , γR,n : [0, 1] → R2
by the formulas

γR,f (t) = f (R cos(2πt) + iR sin(2πt)), t ∈ [0, 1];

γR,n (t) = Rn cos(2πnt) + iRn sin(2πnt), t ∈ [0, 1].

Show that for the 2-cube H : [0, 1]2 → C given by

H(t, s) = (1 − s)γR,f (t) + sγR,n (t),

we have ∂H = γR,f − γR,n .

(b) Show that if R > 0 is chosen sufficiently large, then H([0, 1]2 ) ⊂ U . Therefore, the
winding numbers of γR,f and γR,n are both defined, and

wind(γR,f ) = wind(γR,n ).

(c) Show that if the polynomial f (z) has no roots in C (i.e. no z ∈ C such that
f (z) = 0), then defining

e s) = γ(1−s)R,f (t), (t, s) ∈ [0, 1] × [0, 1],

H(t,

e s) ∈ U for all (t, s) ∈ [0, 1]2 , and hence H

we have H(t, e is a 2-cube in U . Moreover, show
that its boundary is
∂He = γR,f − γ0 ,

where γ0 : [0, 1] → U is the constant curve given by γ0 (t) = a0 for all t ∈ [0, 1].

(d) Conclude that if f (z) has no roots, then

wind(γR,n ) = wind(γ0 ),

which is a contradiction. Hence, f (z) must have a root, which is the main result needed
to prove the Fundamental Theorem of Algebra.

14