LECTURE NOTES FOR MA3H5 MANIFOLDS; WARWICK

AUTUMN TERM 2020

CHRISTIAN BÖHNING

1. The general notion of a manifold

Manifolds show up in many guises in mathematics and the sciences. The fol-
lowing notion helps to introduce some order into the variety of ways that the term
“manifold” is used in different contexts.
Definition 1.1. A pseudogroup of transformations on a topological space S is a set
Γ of transformations satisfying the following axioms:
a) Each f ∈ Γ is a homeomorphism of an open set (called the domain of f ) onto
another open set (called the range of f ) of S;
b) If f ∈ Γ, then the restriction of f to an arbitrary open subset of the domain
of f is in Γ; S
c) Suppose an open subset U of S is a union U = i Ui of open subsets Ui of S.
Then a homeomorphism f of U onto an open subset of S belongs to Γ if the
restriction of f to Ui is in Γ for every i;
d) For every open set U of S, the identity transformation of U is in Γ;
e) If f ∈ Γ, then also f −1 ∈ Γ;
f ) If f ∈ Γ is a homeomorphism of U onto V and f 0 ∈ Γ is a homeomorphism
of U 0 onto V 0 and if V ∩ U 0 is non-empty, then the homeomorphism f 0 ◦ f of
f −1 (V ∩ U 0 ) onto f 0 (V ∩ U 0 ) is in Γ.
Phew! That’s a long definition! Luckily, none of its parts is too deep: b) and c)
can be paraphrased by saying that belonging to Γ is a local condition, and f) can be
expressed more simply by saying that compositions of maps in Γ give an element of
Γ whenever defined. Let’s look at some examples to make this seem less revolting:
Example 1.2. (1) Let S = Rn and let ΓC 0 be the set of all homeomorphisms
f : U → V of an open set U in Rn onto an open set V in Rn .
(2) Let S = Rn and let ΓC r be the set of all homeomorphisms f : U → V of an
open set U in Rn onto an open set V in Rn such that both f and f −1 are
differentiable of class C r , r = 1, 2, . . . , ∞, and
(3) Let S = Rn and let ΓC r ,0 be the set of all homeomorphisms f : U → V of an
open set U in Rn onto an open set V in Rn such that both f and f −1 are
differentiable of class C r , r = 1, 2, . . . , ∞ and moreover, the Jacobian matrix
of f has positive determinant everywhere on U .
(4) Let S = Rn and let ΓC ω be the set of all homeomorphisms f : U → V of an
open set U in Rn onto an open set V in Rn such that both f and f −1 are real
analytic.
(5) Let S = Cn with coordinates z1 , . . . , zn , and let Γhol be the set of all homeo-
morphisms f : U → V of an open set U in Cn onto an open set V in Cn such
that both f and f −1 can be developed in a convergent power series in the zi
locally around every point of their domains.
1
2 CHRISTIAN BÖHNING

(6) Let S = Rn and Γaf f be the set of all homeomorphisms f : U → V of an open

set U in Rn onto an open set V in Rn that locally around every point of U
can be written as
f (x) = Ax + b
for A some nonsingular n × n-matrix and b ∈ Rn . If we require in addition A
to have positive determinant, we get another pseudogroup Γ+ af f .

The good news is that now we are almost ready to define manifolds already!
Definition 1.3. Let M be a topological space. An atlas of M compatible with a
pseudogroup of transformations Γ is a family of pairs {(Ui , ϕi )}, called charts, such
that
S
(i) Each Ui is an open subset of M and M = i Ui , i.e. the Ui cover M .
(ii) Each ϕi is a homeomorphism of Ui onto an open set of S.
(iii) Whenever Ui ∩ Uj is non-empty, the mapping ϕj ◦ ϕ−1 i of ϕi (Ui ∩ Uj ) onto
ϕj (Ui ∩ Uj ) is an element of Γ.
Sometimes the ϕi are also called local Γ-coordinates, especially when S = Rn , or
Cn . For a point p ∈ Ui one then often writes
ϕi (p) = (x1 (p), . . . , xn (p))
and calls x1 (p), . . . , xn (p) the local coordinates of p, and the function x1 , . . . , xn
simply local coordinates. The atlas itself is sometimes also called a Γ-structure
on M . Basically, we would like to call such an M a Γ-manifold. However, to be
politically correct, we need to choose the atlas maximal in a certain sense:
Definition 1.4. A complete atlas on M compatible with Γ is an atlas of M com-
patible with Γ which is not contained in any other atlas of M compatible with
Γ.
Note: every atlas of M compatible with Γ is contained in a unique complete atlas
of M compatible with Γ. In fact, given an atlas A = {(Ui , ϕi )} of M compatible
with Γ, let à be the family of all pairs (U, ϕ) such that ϕ is a homeomorphism of
an open set U of M onto an open set of S and that
ϕi ◦ ϕ−1 : ϕ(U ∩ Ui ) → ϕi (U ∩ Ui )
is an element of Γ whenever U ∩ Ui is nonempty. Then à is the complete atlas
containing A.
Definition 1.5. A Γ-manifold is a Hausdorff topological space M that is second
countable, together with a fixed complete atlas compatible with Γ (sometimes also
called a Γ-structure on M ). Depending on the particular Γ, people have come up
with a plethora of special names; we keep the notation from Example 1.2.
a) If Γ = ΓC 0 , M is called a topological manifold.
b) If Γ = ΓC r , r = 1, 2, . . . , M is called a differentiable manifold of class C r . If
Γ = ΓC ∞ , M is called a smooth manifold or a differentiable manifold (without
further qualification).
c) If Γ = ΓC r ,0 , M is called an orientable differentiable manifold of class C r , r =
1, 2, . . . , and if Γ = ΓC ∞ ,0 , M is called an orientable differentiable manifold
or orientable smooth manifold.
d) If Γ = ΓC ω , then M is called a real analytic manifold.
e) If Γ = Γhol , M is called a complex manifold.
 MA3H5 MANIFOLDS 3

f ) If Γ = Γaf f , M is called an affine manifold, and if Γ = Γ+ af f , M is called an

orientable affine manifold.
In cases a), b), c), d), f), S = Rn , and this n is called the (real) dimension of M ; in
case e), S = Cn , and this n is called the (complex) dimension of M .
There are even other types in common use: symplectic manifolds etc. To avoid
chaos and to focus on a particularly rewarding and useful class we agree:

For us the term manifold, without qualification, will always mean smooth manifold.

The additional requirement that M be Hausdorff seems reasonable. The require-

ment that M be second countable is another tameness assumption on the topological
space M (it should not be too “wild” or “big”). We will never encounter a not second
countable space in the sequel and will not dwell too much on this extra assumption.
Note that if Γ0 ⊂ Γ for pseudogroups Γ0 , Γ then every atlas of M compatible with
0
Γ is also an atlas of M compatible with Γ. For example, if we identify Cn with R2n
by splitting
√ each complex coordinate zj , j = 1, . . . , n into real and imaginary parts:
zj = xj + −1yj , then we see that every complex manifold can also be viewed as a
smooth manifold, and every smooth manifold is also a topological manifold.
Definition 1.6. Let M and N be smooth manifolds. A continuous map f : M → N
is called a morphism of manifolds if for every chart (Ui , ϕi ) of M and every chart
(Vj , ψj ) of N with f (Ui ) ⊂ Vj , the composition ψj ◦ f ◦ ϕ−1
i is a C ∞ . We also
call such an f simply a smooth map or a differentiable map or even just a map.
An isomorphism f of smooth manifolds is a morphism f as above with an inverse
g : N → M which is a morphism. Recall that being an inverse means g ◦ f =
idM , f ◦ g = idN . Such an isomorphism is also called a diffeomorphism, and the
corresponding manifolds are said to be diffeomorphic.
Remark 1.7. Of course one can make similarly define morphisms for other types
of manifolds: for topological manifolds one would require the ψj ◦ f ◦ ϕ−1 i to be
merely continuous, for differentiable manifolds of class C r one would demand that
ψj ◦ f ◦ ϕ−1
i is differentiable of class C r , for complex manifolds the ψj ◦ f ◦ ϕ−1
i
should be complex analytic, i.e. locally developable into convergent power series in
the complex coordinates zi , and so on.
Exercise 1.8. Show that it is sufficient for a map f : M → N to be a morphism of
manifolds if the following holds: for every x ∈ M there exists a chart (U, ϕ), x ∈ U ,
and there exists a chart (V, ψ), f (x) ∈ V , with ψ ◦ f ◦ ϕ−1 a C ∞ map.
Cultural digression. For Γ0 ⊂ Γ pseudogroups, it is one of the most impor-
tant (and difficult) problems in geometry to determine the different Γ0 -structures
on a given M , up to isomorphism, that give rise to one and the same Γ-structure
on M (up to isomorphism). For example, it is easy to see that the spheres S n =
{x ∈ Rn+1 | x21 + · · · + x2n+1 = 1} are topological manifolds and also have a smooth
structure induced from the ambient Rn+1 (we’ll get to that in the next section). Mil-
nor (late 1950’s, early 1960’s) (and subsequently) others proved that some spheres
can be given distinct differentiable structures that give rise to the same underlying
topological manifold (the 7-sphere carries 15 such different differentiable structures;
Milnor was awarded the Fields Medal largely for this work).
Kodaira and Spencer around the same time started the investigation of different
complex structures on the same underlying differentiable manifold. For example, on
4 CHRISTIAN BÖHNING

S 2 there exists only one, on S 6 (A. Adler) there exists none, but in general, there
can exist a continuous supply of complex structures on a given (even-dimensional)
smooth manifold, and these structures themselves form a space with rich geometry.
Benzecri (1959) showed that if a Γ+
af f -structure exists on a compact surface it must
be a torus. More recently, symplectic structures have gained greater importance
placing them on equal footing with complex structures in the framework of mirror
symmetry.

2. Some examples of construction (or description) of manifolds

OK, so the big issue is how do we actually write down a manifold? There are
some recurring basic methods, all of which are important.
2.1. Basic examples. First, as hopefully no one doubts, Rn is a manifold. You can
just use the one chart (Rn , idRn ) to define an atlas! However, note that there can
be other differentiable structures on the topological manifold Rn than this standard
one (exotic ones, e.g. for n=4, found by Michael Freedman in 1982, and studied
further by Taubes). We will always consider this standard one (unless explicitly
stated otherwise).
Second, if M is a manifold and U ⊂ M an open subset, then U is a manifold in a
natural way: U is given the induced topology, and if A = {(Ui , ϕi )} is an atlas for
M , then AU = {(U ∩ Ui , (ϕi ) |U ∩Ui )} is one for U .
Definition 2.1. If M is a smooth manifold, U ⊂ M open, we call a function
f : U → R smooth or differentiable if it is a smooth map of manifolds if we endow
U and R with their smooth manifold structures as explained above.

If M is an m-dimensional manifold with atlas AM = {Ui , ϕi )}, N an n-dimensional

manifold with atlas AN = {Vj , ψj )}, then M × N is an m + n-dimensional manifold
in the following way: with the product topology, we endow M × N with the atlas
A = {(Ui × Vj , ϕi × ψj )}.

So far nothing really to knock your socks off, but there are tons of other more
interesting manifolds and constructions, which we will come to in a minute; here is
something a little more entertaining to begin with: take K = R or K = C; we will
construct a manifold PnK called (real or complex) projective space over K of (real
or complex) dimension n. As a set
PnK = (K n+1 − {0})/ ∼
where ∼ is the equivalence relation
(p0 , . . . , pn ) ∼ (q0 , . . . , qn ) ⇐⇒ ∃c ∈ K\{0} : c(p0 , . . . , pn ) = (q0 , . . . , qn ).
We give PnK the quotient topology of the usual Euclidean topology on K n+1 − {0}
(meaning a subset in PnK is declared open iff its preimage in K n+1 − {0} is open).
We denote the equivalence class of (p0 , . . . , pn ) in PnK by (p0 : · · · : pn ). We define
an atlas on PnK in the following way: For i = 0, . . . n we let
Ui = {(p0 : · · · : pn ) | pi 6= 0}
and
 
n p0 pi pn (i) (i) (i)
ϕi : Ui → K , ϕi ((p0 : · · · : pn )) = = (x0 , . . . , xi−1 , xi+1 , . . . , x(i)
c
,..., ,..., n ).
pi pi pi
 MA3H5 MANIFOLDS 5

In this case ϕi (Ui ) = K n and

ϕji := ϕj ◦ ϕ−1
i : ϕi (Ui ∩ Uj ) → ϕj (Ui ∩ Uj )
is given by
(i)
xν
x(j)
ν = (i)
, ν 6= i, j
xj
(j) (i)
xi = (xj )−1 .
Thus PnK is a smooth manifold (of real dimension 2n if K = C). By the way, this
shows that PnC is also a complex manifold in this way.

2.2. Submanifolds of known manifolds, such as Rn . If you are given a manifold

M, of dimension d, such as Rn or PnK for example, you can always make more
interesting manifolds by considering appropriate zero sets of smooth functions in M
as we will now explain.
Definition 2.2. Let us identify Re , for e ≤ d, with its image Re ⊂ Rd under the
canonical inclusion
Re → {0} × Re ⊂ Rd−e × Re = Rd .
Suppose M is a manifold of dimension d. Then a subset Y ⊂ M is called a subman-
ifold of dimension e if for every y ∈ Y there exists a chart (U, ϕ) on M with y ∈ U
such that
ϕ(U ∩ Y ) = ϕ(U ) ∩ Re .
We also say that Y has codimension c = d − e.
Proposition 2.3. A submanifold of dimension e of a manifold M is a manifold in
its own right in a canonical way. Indeed, for (U, ϕ) ranging over all charts satisfying
the conditions of Definition 2.2, the pairs (U ∩ Y, ϕ |U ∩Y ) give a smooth atlas for Y
endowed with the subspace topology.
Proof. If (U, ϕ) and (V, ψ) are two charts satisfying the condition, we have
ϕ((U ∩ Y ) ∩ (V ∩ Y )) = ϕ(U ∩ V ) ∩ Re
is open in Re and
(ψ |U ∩V ∩Y ) ◦ (ϕ |U ∩V ∩Y )−1 = (ψ ◦ ϕ−1 ) |ϕ(U ∩V )∩Re
which is smooth on ϕ(U ∩ V ) ∩ Re if ψ ◦ ϕ−1 is smooth on ϕ(U ∩ V ). We endow Y
with the topology in which a subset Ω ⊂ Y is open iff for each chart (U, ϕ) satisfying
the conditions of Definition 2.2
ϕ(Ω ∩ U )
is open in Re ; we leave it as an easy exercise to check that this is just the subspace
topology. 
There is an easy criterion to check when charts as required in Definition 2.2 exist
for a subset Y .
Proposition 2.4. Let M be a manifold of dimension d and Y ⊂ M a subset.
Suppose that for every y ∈ Y there exists a chart (U 0 , ϕ0 ), ϕ0 = (x1 , . . . , xd ), y ∈ U 0 ,
and smooth functions f1 , . . . , fc on U 0 such that
Y ∩ U 0 = {p ∈ U 0 | f1 (p) = · · · = fc (p) = 0}
6 CHRISTIAN BÖHNING

and the determinant  

∂fi
det (p) 6= 0
∂xj 1≤i,j≤c
for all p ∈ U 0 ; here we view the fi as functions of the local coordinates x1 , . . . , xd
on U 0 . Then Y is a submanifold of codimension c; indeed, a chart (U, ϕ) of the
type required in Definition 2.2 can be obtained by using as local coordinates ϕ =
(w1 , . . . , wc , wc+1 , . . . , wd ) on some open set U ⊂ U 0 containing y, where
wi (p) = fi (p), 1 ≤ i ≤ c, wc+1 = xc+1 , wc+2 = xc+2 , . . . , wd = xd .
Proof. This is really just an immediate consequence of the Inverse Function Theorem
you know from MA259 Multivariable Calculus: indeed, identifying U 0 with an open
subset of Rd via ϕ we just have to show that the map F : U 0 → Rd (coordinates in
the target Rd are w1 , . . . , wd ) given by
wi = fi (x1 , . . . , xd ), 1 ≤ i ≤ c, wc+1 = xc+1 , wc+2 = xc+2 , . . . , wd = xd
is a local diffeomorphism around p; and the Inverse Function Theorem says that for
this it is sufficient to check that the Jacobian of the mapping F in p has nonzero
determinant. But this Jacobian is of the form
  !
∂fi
∂xj (p) ∗ c×(d−c)
1≤i,j≤c
0(d−c)×c Id(d−c)×(d−c)
(the subscripts indicate the format of the matrices involved, thus 0(d−c)×c is the
(d − c) × c-matrix with all entries 0, Id(d−c)×(d−c) the (d − c) × (d − c) identity
matrix, and ∗c×(d−c) an c × (d − c)-matrix whose entries we do not care about. The
determinant of the preceding big matrix is just
 
∂fi
det (p)
∂xj 1≤i,j≤c
hence nonzero by assumption. 
The careful reader will have noted that I was being sloppy with the notation in
the preceding proof inasmuch as I tacitly identified U 0 ⊂ M and ϕ0 (U 0 ) ⊂ Rd via
ϕ0 , and did not distinguish notationally corresponding objects on U 0 and ϕ0 (U 0 ), for
example f (on U 0 ) and f ◦ (ϕ0 )−1 on ϕ0 (U 0 ). In my opinion, such sloppiness, when
used with discretion and accompanied by an explanation when first used, does not
do any harm and even helps making the main ideas of the proof more transparent.
It is also extremely common in the literature on manifolds.
Using Proposition 2.4, we can now give many more examples of manifolds in the
form of submanifolds of some known manifolds; for example, the spheres
S n = x = (x1 , . . . , xn+1 ) ∈ Rn+1 | x21 + · · · + x2n+1 − 1 = 0


are submanifolds of Rn+1 since the gradient of x21 + · · · + x2n+1 − 1 vanishes at no

point of S n . We will consider those spheres with this “standard” differentiable struc-
tures unless explicitly stated otherwise (though there are other exotic differentiable
structures on spheres).
For another example, look at the orthogonal group O(n) of n × n invertible ma-
2
trices A with At = A−1 . This is a submanifold of Rn (of dimension n(n − 1)/2).
2
Indeed, it is defined by the n(n + 1)/2 equations inside Rn given by
At A − Idn×n = 0
 MA3H5 MANIFOLDS 7

and it suffices to show that the differential/Jacobian of the map

2 2
Rn → Rn
A 7→ At A − Idn×n
has rank equal to n(n + 1)/2 (the dimension of the space of symmetric matrices) at
every point of O(n). But this Jacobian, at the point corresponding to a matrix A0 ,
is easily computed to be (Exercise!)
2 2
Rn → Rn
B 7→ B t A0 + At0 B.
Since A0 ∈ O(n) is invertible we can write B = A0 B 0 and get
B t A0 + At0 B = (B 0 )t + B 0 .
Hence the rank of the Jacobian is n(n + 1)/2 as desired.
For yet another example, we already know that S 1 ⊂ R2 is a submanifold, so
taking products
Tn = S 1 × · · · × S 1 ⊂ R2 × · · · × R2 = R2n
is a manifold, called an n-dimensional torus; we will see another method to construct
such tori in the next section.
Exercise 2.5. Consider PnC and let f = f (z0 , . . . , zn ) be a nonzero homogeneous
polynomial in z0 , . . . , zn of degree d with complex coefficients; here homogeneous
means that very monomial
z0α0 z2α2 . . . znαn
occurring in f has total degree α0 + · · · + αn equal to d. Let Mf = {(z0 : · · · : zn ) ∈
PnC | f (z0 , . . . , zn ) = 0}. Show that if at least one of the partials
∂f
, i = 0, . . . , n
∂zi
is nonzero at every point of Mf , then Mf is a submanifold of PnC .
Such Mf are very varied and interesting. For example, for n = 2, it can be shown
that Mf is homeomorphic to a compact surface with
1
g = d(d − 3) + 1
2
holes (a “g-holed torus”). To show this requires a little more sophisticated tech-
niques, though, than the ones we can get to here.
2.3. Quotient manifolds. We get a much larger supply of interesting examples of
manifolds by taking quotients of manifolds by suitable group actions.
Definition 2.6. An isomorphism of a smooth manifold M onto itself is also called a
diffeomorphism. All diffeomorphisms of M form a group Diff(M ) under composition.
Let G ⊂ Diff(M ) be a subgroup.
Definition 2.7. G is called a properly discontinuous group of diffeomorphisms of
M if for any two compact subsets K1 , K2 of M , the set
{g ∈ G | g(K1 ) ∩ K2 6= ∅}
is finite.
We say that G has no fixed points if for all g ∈ G, g 6= id, it holds that g has no
fixed points on M .
8 CHRISTIAN BÖHNING

Theorem 2.8. If G is a properly discontinuous group of diffeomorphisms of M and

has no fixed points, then the set of G-orbits M/G is a manifold in a natural way, to
be described in the course of the proof.
Proof. We endow M/G with the quotient topology, in which a set is open iff its
preimage in M under the natural quotient map M → M/G is open. We use the
notation M ∗ for M/G, the set of G-orbits in M , and for a point p ∈ M write p∗
for its orbit G · p, an element of M ∗ . We will show that for q ∈ M there exists
a neighbourhood U of q such that if p1 , p2 ∈ U , p1 6= p2 , then p∗1 6= p∗2 . In fact,
we will show that there is an open subset U 3 q such that gU ∩ U = ∅ for all
g ∈ G, g 6= id. Let U1 ⊃ U2 ⊃ U3 ⊃ . . . be a base of relatively compact open
neighbourhoods of q (relatively compact means that their closure be compact; such
a base of neighbourhoods exists since locally around q, M is homeomorphic to an
open in Rn ). Then each of the sets
Fm = {g | gUm ∩ Um 6= 0} , m = 1, 2, . . .
is finite by the assumption that G is properly discontinuous, and moreover,
F1 ⊃ F2 ⊃ F3 ⊃ . . . .
Suppose, arguing by contradiction, that each Fm contains an element gm ∈ G,
gm 6= id. Then since each Fm is finite, there will be a g ∈ G, g 6= id, with g ∈ Fm
for all m, meaning that
gUm ∩ Um 6= ∅ ∀ m
and at the same time the Um “converge to q” (more precisely, they form a basis
of the neighbourhoods of q), hence we would conclude g(q) = q, contradicting the
non-existence of fixed points which we have assumed.
This shows that we can cover M with open sets Uj such that whenever p1 , p2 ∈ Uj ,
p1 6= p2 , then also p∗1 6= p∗2 ; in other words,
Uj → Uj∗ = {p∗ | p ∈ Uj }, p 7→ p∗
is one-to-one. We now give Uj∗ the differentiable structure that Uj has (as an open
subset of M ). In other words, if (V, ψ) is a chart on U , then we let (V ∗ , ψ ∗ ) be the
chart on Uj∗ with ψ ∗ (p∗ ) = ψ(p). This defines a differentiable/smooth atlas on M/G
and endows M/G with the structure of a smooth manifold. 
Example 2.9. As our first example, consider M = Rn and choose n vectors
ω1 , . . . , ωn ∈ Rn that are R-linearly independent. Let G be Zn , and view it ias
a subgroup fo the diffeomorphisms of Rn by identifying g = (m1 , . . . , mn ), mk ∈ Z
with the map
Xn
n n
g : R → R , x 7→ x + mk ωk .
k=1
It is clear that G is without fixed points and properly discontinuous (for the latter
note that a compact subset of Rn is, in particular, bounded). The quotient manifold
Tn = Rn /G
is an n-dimensional torus (as an exercise, you should show that it is indeed diffeo-
morphic to a product of n copies of S 1 ).
Example 2.10. We let W = Cn −{0} (viewed as an open submanifold of Cn ' R2n ).
We choose complex numbers αν , ν = 1, . . . , n, with 0 < |αν | < 1 for all ν, and let
G = {g m | m ∈ Z, g(z1 , . . . , zn ) = (α1 z1 , . . . , αn zn )} .
 MA3H5 MANIFOLDS 9

It is easy to see that G is isomorphic to Z and is without fixed points and properly
discontinuous (Exercise: Show this!), and the quotient manifold W/G is called a
Hopf manifold. It can be shown that it is diffeomorphic to S 1 × S 2n−1 (you can
think about this as a challenge).
Example 2.11. Let M = S 3 and view this as the unit sphere in R4 ' C2 (note
that if z1 , z2 are complex coordinates in C2 , this is the subset z1 z̄1 + z2 z̄2 = 1. Fix
coprime integers p, q > 1 and consider the group G generated by the diffeomorphism
of S 3 given by  2πi 2πiq 
g(z1 , z2 ) = e p z1 , e p z2 .
Then G is isomorphic to Z/p and without fixed points (and properly discontinuous
since it is a finite group). The quotient manifold in this case is often denoted by
L(p, q) and called a lens space.
2.4. Surgeries. Another very useful method to produce new manifolds from old
ones is to modify a given manifold locally in a certain way we will now describe.
This is called surgery.
Theorem 2.12. Let M be a manifold, and let S ⊂ M be a compact submanifold.
Suppose we have an open neighbourhood W of S in M and another pair of manifolds
S ∗ ⊂ W ∗ with S ∗ compact and W ∗ an open neighbourhood of S ∗ . Suppose we are
given a diffeomorphism
f : W ∗ − S∗ → W − S
of W ∗ − S ∗ onto W − S. Then we can “replace” S by S ∗ and obtain a new manifold
M ∗ = (M − W ) ∪ W ∗ . More precisely, we consider
M ∗ = ((M − S) t W ∗ ) / ∼
where ∼ is the equivalence relation identifying each point z ∗ ∈ W ∗ − S ∗ with z =
f (z ∗ ). This, with the quotient topology, inherits a natural manifold structure from
M (and W ∗ ).
Proof. Exercise for the reader. 
Example 2.13. We consider M = P1C × P1C . We let ζ = (ζ0 : ζ1 ) be (homogeneous)
coordinates on the first copy of P1 , and let z = ζζ01 be a local coordinate on the open
subset U ⊂ P1C , U ' C where ζ0 6= 0. Then M contains W = U ×P1 , and W contains
S = {0} × P1 . Let W ∗ = U ∗ × (P1 )∗ be a second copy of W (on which everything
is denoted by the same letters, but with a superscript ∗), and let S ∗ = {0} × (P1 )∗ .
Fix an integer m > 0 and define f : W ∗ − S ∗ → W − S by
ζ1∗
  
∗ ∗ ∗ ∗
(z, ζ) = f (z , ζ ) := z , ζ0 : ∗ m .
(z )
Then f is a diffeomorphism and we can define Mm ∗ = (M − S) ∪ W ∗ by surgery.
∗ are homeomorphic
This is called a Hirzebruch manifold. One can show that all M2n
∗
to each other, and so are all M2n+1 , but the ones for even and odd m are not
homeomorphic to each other. This requires more techniques from topology than we
have right now, however.
Example 2.14. Let M be a complex manifold with a point S := p ∈ M and a chart
(W, ϕ = (z1 , z2 )) with ϕ : W ' U where U = {(z1 , z2 ) ∈ C2 | ||z1 | < 1, |z2 | < 1} and
ϕ(p) = (0, 0). In particular, ϕ is a smooth chart of course. Define a submanifold
W ∗ of W × P1C as follows:
W ∗ = (z1 , z2 ; (ζ1 : ζ2 )) ∈ W × P1C | z1 ζ2 − z2 ζ1 = 0

10 CHRISTIAN BÖHNING

It is easy to check this is a submanifold since if g = z1 ζ2 − z2 ζ1 we have that

∂g
, i = 1, 2
∂zi
are never simultaneously zero on W ∗ . Let F : W ∗ → U be the restriction of the
projection map. Let S ∗ = {0} × P1C ⊂ W ∗ . Clearly, F maps S ∗ onto (0, 0) and
F : W ∗ − S ∗ → U − {0, 0)}
/ S ∗ , then at least one of z1 , z2 is
is a diffeomorphism: namely, if (z1 , z2 ; (ζ1 : ζ2 )) ∈
nonzero and hence (ζ1 : ζ2 ) is determined by (z1 , z2 ). Letting f = ϕ−1 ◦F , we obtain
by surgery a manifold M ∗ = (M − p) ∪ P1C . This procedure is very important and
called the blowup of M at p. Indeed, one imagines sticking a little piece of straw
into M at p and blowing a bubble S 1 ' P1C replacing p. Note that the points of
P1C = S ∗ correspond to the complex lines in C2 passing through (0, 0).
Example 2.15. There is a real version of the preceding example, replacing C by R
everywhere. We leave it to the reader as an exercise to work out the details what
changes and what doesn’t.

3. Tangent space, vector fields, one-parameter groups of

diffeomorphisms, 1-forms, tangent and cotangent bundles, vector
bundles
Suppose we are given a smooth manifold M and p ∈ M a point. We shall now
define the notion of a tangent vector to M at the point p. By a (parametrised,
smooth) curve in M we mean a smooth map
γ : (a, b) → M
where (a, b) ⊂ R is an interval, a < b. We moreover define the R-algebra of germs
of smooth functions defined in a neighbourhood of p CM,p ∞ in the following way:

the elements of CM,p are equivalence classes of pairs (f, U ), where U 3 p is an

∞

open subset of M and f : U → R a smooth function (we also write f ∈ C ∞ (U )),

where (f, U ) and (g, V ) are considered to be equivalent if there is a smaller open
neighbourhood W of p, contained in both U and V , on which f and g agree. You
see, the name for CM,p
∞ is more intimidating than the actual thing!

We now consider a curve in M as above with γ(t0 ) = p, a < t0 < b. We define

the tangent vector to the curve γ at p to be the map
∞
X(= Xγ,p ) : CM,p →R
defined by
df (γ(t))
Xf = (t0 ).
dt
This map has the following properties:
a) X : CM,p
∞ → R is R-linear;

b) X satisfies the Leibniz rule

∞
X(f g) = (Xf )g(p) + f (p)(Xg), ∀ f, g ∈ CM,p .
The maps satisfying a) and b) form a real vector space (and are called derivations).
We will show that the set of tangent vectors to curves, i.e., derivations of the form
Xγ,p , form a vector subspace of dimension n of the vector space of all derivations,
where n = dim M . We then define a tangent vector to M at the point p to be simply
a tangent vector to some curve at p.
 MA3H5 MANIFOLDS 11

Indeed, let (U, ϕ = (x1 , . . . , xn )) be a chart at p, p ∈ U . We shall show that the

set of tangent vectors at p is nothing but the vector space with basis
   
∂ ∂
,..., .
∂x1 p ∂xn p
Given any curve γ(t) with p = γ(t0 ) let xj = γj (t) its components with respect to
the local coordinates. Then
  X  ∂f   dγj (t) 
df (γ(t))
= ·
dt t0 ∂xj p dt t0
j

which proves that every tangent vector to M at p is a linear combination of

   
∂ ∂
,..., .
∂x1 p ∂xn p
Conversely, given a linear combination
X  ∂ 
ξj
∂xj p
j

consider the curve given in local coordinates by

xj = xj (p) + ξj t, j = 1, . . . , n.
Then the tangent vector to this curve at t = 0 is the given one. We still need to
prove the linear independence of the ∂x∂ j : suppose that
p
 
X ∂
ξj = 0.
∂xj p
j

Then  
X ∂xk
0= ξj = ξk , k = 1, . . . , n.
∂xj p
j
Thus we have shown that the set of tangent vectors to curves, i.e., derivations of
the form Xγ,p , form a vector subspace of dimension n of the vector space of all
derivations, where n = dim M .
Definition 3.1. We denote the n-dimensional vector space of all tangent vectors
to M at p by
Tp (M )
and call it the tangent space to M at p.
Remark 3.2. It is known that for a smooth (C ∞ -)manifold M , Tp (M ) coincides with
the space of all derivations of CM,p
∞ (not just a subspace).

Definition 3.3. A vector field X on a manifold M is assignment to each point

p ∈ M of a tangent vector Xp to M at p. If f is a smooth function, then Xf is the
function on M defined by (Xf )(p) = Xp f . The vector field X is called smooth if
Xf is a smooth function on M for every smooth function f on M . We will denote
the smooth vector fields on M by X(M ).
In the following, we will practically exclusively deal with smooth vector fields and
therefore agree that the term “vector field” should always be understood as meaning
“smooth vector field” from now on unless explicitly stated otherwise.
12 CHRISTIAN BÖHNING

If (U, x1 , . . . , xn ) is a chart on M , a vector field X on U can be written as

X ∂
ξj
∂xj
j

with ξj smooth functions of the x1 , . . . , xn on U . If M is covered by charts

(λ)
(Uλ , x1 , . . . , x(λ)
n ),
a vector field X on all of M can be thought of in very concrete terms in the following
way: the datum of X amounts to giving oneself n smooth functions
(λ)
ξ1 , . . . , ξn(λ)
(λ) (λ)
of the local coordinates x1 , . . . , xn on Uλ , so that on Uλ
X (λ) ∂
X |Uλ = ξj (λ)
;
j ∂xj
and on Uλ ∩ Uµ we want
X (λ) ∂ X (µ) ∂
ξj (λ)
= ξj (µ)
j ∂xj j ∂xj
meaning that the system of functions need to satisfy

(λ)
X ∂x(λ)
j (µ)
ξj = ξ
(µ) k
k ∂xk
on Uλ ∩ Uµ . Although computations with these local expressions can get messy
rather quickly, this description sometimes allows one to obtain useful information
about X(M ).
The set X(M ) is obviously an R-vector space and also a module over the algebra
C (M ) of smooth functions on M , but it has another useful structure: if X, Y ∈
∞

X(M ) define their bracket by

[X, Y ]f = X(Y f ) − Y (Xf ).
This is again a vector field; indeed, if in terms of a chart (U, x1 , . . . , xn )
X ∂ X ∂
X= ξj , Y = ηj
∂xj ∂xj
j j

then
X  ∂ηj 
∂ξj ∂f
[X, Y ]f = ξk − ηk
∂xk ∂xk ∂xj
j,k
∂
so that [X, Y ] is a vector field whose components on U with respect to the ∂xj are
X  ∂ηj ∂ξj

ξk − ηk .
∂xk ∂xk
k
The bracket
[·, ·] : X(M ) × X(M ) → X(M )
is R-bilinear and satisfies
[[X, Y ]Z] + [[Y, Z], X] + [[Z, X], Y ] = 0, ∀ X, Y, Z ∈ X(M )
as a direct calculation shows. The last identity is called the Jacobi identity, and
a vector space endowed with a bilinear bracket [·, ·] satisfying the Jacobi identity
 MA3H5 MANIFOLDS 13

is called a Lie algebra. You will learn more about these in MA4E0 Lie Groups
and MA453 Lie Algebras. Suffice it to mention here that Lie algebras usually arise
as tangent spaces to the identity of some Lie group, which is a group endowed
with some kind of smooth or similar structure; indeed, X(M ) should be though
tof as the tangent space to the “manifold” Diff(M ), the diffeomorphism group of
M . Unfortunately, it is not a manifold in our sense here, but a “Fréchet manifold”,
locally modelled not on some Rd , but on some special infinite-dimensional topological
vector spaces called Fréchet spaces (sorry! The world out there is vast and cruel).
We will leave it at that here, but for cultural reasons we remark that “Lie” is
pronounced “L-ee”, after the Norwegian mathematician Sophus Lie.
If f, g are smooth functions on M and X, Y vector fields, a computation shows
that
[f X, gY ] = f g[X, Y ] + f (Xg)Y − g(Y f )X
(Exercise!)
Definition 3.4. Let
f : M → M0
be a smooth map of manifolds. The differential at p ∈ M of f is the linear map
f∗ : Tp (M ) → Tf (p) (M 0 )
defined as follows. For each tangent vector Xp ∈ Tp (M ), choose a curve γ(t) in M
such that Xp is the tangent vector to γ(t) at p = γ(t0 ). Then f∗ (Xp ) is the vector
tangent to the curve f (γ(t)) at f (p) = f (γ(t0 )).
It is easy to check that f∗ (Xp ) is independent of the choice of γ(t); in fact, if
(U, x1 , . . . , xn ) is a chart around p, and (V, y1 , . . . , ym ) is a chart around f (p) such
that f (U ) ⊂ V , it is immediate to see that the differential f∗ is the linear map given
by the Jacobian matrix  
∂fi
(p)
∂xj i=1,...,m, j=1,...,n
if we use the bases
       
∂ ∂ ∂ ∂
,..., ; ,...,
∂x1 p ∂xn p ∂y1 f (p) ∂ym f (p)
for Tp (M ) and Tf (p) (M 0 ), respectively. You should check the preceding assertions as
an exercise and also verify that if g is a smooth function defined in a neighbourhood
of f (p), then
(f∗ (Xp ))g = Xp (g ◦ f ).
If we want to specify the point p, we write (f∗ )p instead of f∗ . Another common
notation used for (f∗ )p is dfp (or Dfp ).
The following is a useful consequence of the Inverse Function Theorem that you
encounter in MA259 Multivariable Calculus:
Proposition 3.5. Let f : M → M 0 be a smooth map and p ∈ M as before.
a) If (f∗ )p is injective, there exist local coordinates x1 , . . . , xn in a neighbourhood
U of p and local coordinates y1 , . . . , ym in a neighbourhood of f (p) such that
yi (f (q)) = xi (q) for q ∈ U and i = 1, . . . , n.
This can be paraphrased by saying that f locally around p “looks like (=is
diffeomorphic to) the inclusion of Rn into Rm ”.
In particular, f is a homeomorphism of U onto f (U ).
14 CHRISTIAN BÖHNING

b) If (f∗ )p is surjective, there exists a local coordinate system x1 , . . . , xn in a

neighbourhood U of p and local coordinates y1 , . . . , ym in a neighbourhood of
f (p) such that
yi (f (q)) = xi (q) for q ∈ U and i = 1, . . . , m.
This can be paraphrased by saying that f locally around p “looks like (=is
diffeomorphic to) the projection of Rn onto Rm ”.
In particular, the mapping f : U → M 0 is open (=maps open sets to open
sets).
c) If (f∗ )p is bijective, then f defines a homeomorphism of a neighbourhood U
of p onto a neighbourhood V of f (p) and its inverse f −1 : V → U is also
differentiable/smooth.
Proof. Let us first assume that (f∗ )p is injective and let y1 , . . . , ym a system of local
coordinates around q = f (p) on M 0 . We claim that it is possible to choose from the
set of functions
y1 ◦ f, . . . , ym ◦ f
a subset of n functions that form a system of local coordinates at p in M . Indeed,
in terms of some local coordinates w1 , . . . , wn at p we can write the component
functions yi ◦ f = fi (w1 , . . . , wn ), and the hypothesis that (f∗ )p is injective means
that  
∂fi
(p), i = 1, . . . , m, j = 1, . . . , n
∂wj
has rank n; we can thus select n indices i1 , . . . , in from the set {1, . . . , m} such that
the submatrix
∂fij
 
(p),
∂wj
is a square invertible matrix with nonzero determinant; by the Inverse Function
Theorem, yi1 ◦ f, . . . , yin ◦ f give a local diffeomorphism, hence can be used as local
coordinates x1 = yi1 ◦ f, . . . , xn = yin ◦ f of the type required in a). A reordering of
the y’s (so that yi1 , . . . , yin become y1 , . . . , yn ) gives local coordinates around q such
that all properties in a) are satisfied.
For b), let us now assume that (f∗ )p is surjective and let y1 , . . . , ym be local
coordinates at q ∈ M 0 . Then we claim that
y1 ◦ f, . . . , ym ◦ f
are part of a system of local coordinates at p on M . Indeed, with the same notation
as before, the rank of
 
∂fi
(p), i = 1, . . . , m, j = 1, . . . , n
∂wj
is now m. Reordering the columns of this matrix of necessary, we can assume
without loss of generality, that the submatrix consisting of the first m columns of
this matrix is invertible. Hence again by the Inverse Function Theorem,
x1 = y1 ◦ f, . . . , xm = ym ◦ f, xm+1 = wm+1 , . . . , xn = wn
are local coordinates at p in M .
Finally part c) is simply obtained by putting parts a) and b) together. 
Definition 3.6. Let f : M → M 0 be a smooth map of manifolds.
 MA3H5 MANIFOLDS 15

a) If (f∗ )p is injective for every p ∈ M , then f is called an immersion. We then

say that M is immersed in M 0 by f , or that M is an immersed submanifold
of M 0 . We call f a submersion if (f∗ )p is surjective for every p ∈ M .
b) We call f an embedding of M into M 0 if f is an injective immersion and a
homeomorphism onto its image f (M ).
To tie this in with the notion of submanifold Y ⊂ M from Definition 2.2, you
should convince yourself that if Y is a submanifold of M , then the inclusion map
Y ⊂ M is an embedding. And, conversely, given an embedding i : N → M of
manifolds, the image Y = i(N ) ⊂ M is a submanifold.
There is a slight catch here that you should be aware of: an injective immersion
need not necessarily be an embedding. That is, if f : M → M 0 is an injective
immersion, then it does not necessarily follow that f is also a homeomorphism onto
its image f (M ). The topology of M can in general be finer than the relative topology
induced from M 0 . An example to bear in mind is the composite map
ϕ : R → L ⊂ R2 → R2 /Z2 = T2
where L ⊂ R2 is a line with irrational slope, and (m, n) ∈ Z2 acts on R2 by (x, y) 7→
(x + m, y + n). The image ϕ(R) is dense in the torus T2 .
However, if M is compact, then every injective immersion f : M → M 0 is an
embedding.
We next aim to carry over some of the theory of ordinary differential equations
to manifolds. This will also have some very useful consequences as we shall see. In
other words, we will talk about integral curves of vector fields, 1-parameter groups of
diffeomorphisms and local and global flows generated by vector fields on manifolds
next. As an application we will prove Ehresmann’s result on local triviality of
differentiable fibrations.
Let X be a vector field on a manifold M . A curve γ(t) in M is called an integral
curve of X if, for every parameter value t0 , the vector Xγ(t0 ) is tangent to the curve
γ(t) at γ(t0 ). For any point p0 of M , there is a unique integral curve x(t) of X,
defined for |t| <  for some  > 0, such that p0 = x(0). Indeed, if x1 , . . . , xn are
local coordinates in a neighbourhood U of p0 and
X ∂
X |U = ξi
∂xi
i
then an integral curve of X is a solution of the system of differential equations
dxi
= ξi (x1 (t), . . . , xn (t)), i = 1, . . . , n.
dt
The result then follows from the fundamental existence and uniqueness theorem for
solutions of ordinary differential equations (with smooth right hand side), that you
get to know in MA254 Theory of ODEs. We can view their solution theory a little
bit geometrically in terms of the concept of a 1-parameter group of diffeomorphisms,
or flow.
Definition 3.7. A 1-parameter group of diffeomorphisms or (global) flow on M
is a mapping Φ : R × M → M , Φ(t, p) = Φt (p) ∈ M , which satisfies the following
conditions:
a) For each t ∈ R, Φt : M → M is a diffeomorphism of M (=a smooth self-map
of M with a smooth inverse);
b) for all t, s ∈ R and p ∈ M , Φt+s (p) = Φt (Φs (p)).
16 CHRISTIAN BÖHNING

Property b) just says that the map R → Diff(M ), t 7→ Φt is a group homomorphism.

Each 1-parameter group induces a vector field X on M as follows: the curve
γp (t) = Φt (p) has tangent vector Xp at p = Φ0 (p). The image of γp (t) is the orbit
of p. The curve γp (t) is an integral curve of X starting at p.
Definition 3.8. A local 1-parameter group of local diffeomorphisms or a local flow
on M is morally the same as a flow except that the time parameter t is now only in
a small interval (−, ) of 0 ∈ R, and p is in an open subset of M . More precisely, let
I be the open interval (−, ) and U an open set of M . A local 1-parameter group
of local diffeomorphisms defined on I × U is a smooth map Φ : I × U → M such
that
a) for each t ∈ I , Φt = Φ(−, t) : U → Φt (U ) is a diffeomorphism;
b) if t, s, t + s ∈ I and p, Φs (p) ∈ U , then
Φt+s (p) = Φt (Φs (p)).
Obviously, such a local flow induces a vector field on U in the same manner as
before. Actually, conversely, a vector field also generates a local 1-parameter group
of local diffeomorphisms inducing the given vector field around every point where it
is defined:
Proposition 3.9. Let X be a vector field on a manifold. For each point p0 ∈ M ,
there is an open neighbourhood U of p0 and  > 0 and a local 1-parameter group of
local diffeomorphisms Φt : U → M , t ∈ I , inducing the given X on U .
Moreover, if two local 1-parameter groups of local diffeomorphisms Φ and Ψ de-
fined on I × U induce the same vector field on U , they coincide.
Proof. Strictly speaking, this is again just a rephrasement of results from MA254
Theory of ODEs in geometric terms: in fact, let x1 , . . . , xn be local coordinates in
a neighbourhood W of p0 with x1 (p0 ) = · · · = xn (p0 ) = 0, and write
X ∂
X |W = ξi (x1 , . . . , xn )
∂xi
i
as before. Then consider the system of ordinary differential equations
dfi
= ξi (f1 (t), . . . , fn (t)), i = 1, . . . , n
dt
for unknown functions f1 , . . . , fn . By the fundamental theorem for systems of or-
dinary differential equations from MA254 Theory of ODEs again, there are unique
functions
f1 (t; x), . . . , fn (t; x)
defined for all x = (x1 , . . . , xn ) with |xi | < δ1 and |t| < 1 , δ1 , 1 some small positive
real numbers, and depending smoothly on t and x, which form a system of solu-
tions of the preceding differential equations for each fixed x and satisfy the initial
conditions
fi (0; x) = xi .
We put
Φ(t, x) = (f1 (t; x), . . . , fn (t; x)) for |t| < 1 and x ∈ U1 = {x | |xi | < δ1 ∀ i}.
Then if |t|, |t| and |t + s| are all less than 1 and both x and Φs (x) = Φ(s, x) are in
U1 , then the functions
gi (t) = fi (t + s; x)
 MA3H5 MANIFOLDS 17

are a solution to the initial system of differential equations with initial conditions
gi (0) = fi (s; x), and thus by the uniqueness of the solutions,
gi (t) = fi (t; Φs (x)).
Thus Φt (Φs (x)) = Φt+s (x); moreover, there exist δ > 0 and  > 0 such that putting
U = {x | |xi | < δ ∀ i}, we have Φt (U ) ⊂ U1 for |t| < : indeed, this follows simply
because Φ0 is the identity on U1 . Since therefore
Φ−t (Φt (x)) = Φt (Φ−t (x)) = Φ0 (x) = x
for all x ∈ U and |t| < , all Φt are diffeomorphisms of U onto Φt (U ). Thus Φ is a
local 1-parameter group of local diffeomorphisms defined on I × U inducing X. 

If under the hypotheses of Proposition 3.9, there exists a global 1-parameter group
of diffeomorphisms Φ : R × M → M (hence defined for all times t) inducing X, we
say X is complete. The following is a fact that adds something genuinely new to
facts known from the (local) theory of ordinary differential equations.
Proposition 3.10. On a compact manifold M , every vector field X is complete.
Proof. For every point p, let U (p) be a neighbourhood of p and (p) > 0 such that
there exists a local 1-parameter group of local diffeomorphisms Φ defined on I(p) ×
U (p) inducing X. Since M is compact, finitely many of the U (p), U (p1 ), . . . , U (pk )
say, cover M . Putting  the minimum of the (pi ), i = 1, . . . , k, we see that Φt (p)
can be defined on I × M , hence on all of R × M . Another way to think of this is
to say that the integral curve γ of X with γ(0) = p exists for all time: after time ,
you can continue it for another time interval , and so on. 

We need one further ingredient before getting to Ehresmann’s theorem: partitions

of unity. These are in general quite useful to patch together locally defined objects
into a globally defined object on smooth manifolds. They are particular to the
context of smooth manifolds: nothing like this exists for complex manifolds for
example. Partitions of unity express in some way the fact that the geometry on a
differentiable manifold is rather “soft” and “flabby”, whereas for complex manifold it
is much “harder”and more ”rigid”. Here we also use for the first time the assumption
that manifolds be second-countable.
Theorem 3.11. Let M be a manifold, {Uα }α∈A a collection of open sets cover-
ing M . Then there exists a sequence of smooth functions {θi }i∈N on M , called a
partition of unity subordinate to the open cover {Uα }, with the following properties:
(a) For all p ∈ M and i we have 0 ≤ θi (p) ≤ 1.
(b) Each p ∈ M has a neighbourhood on which all but finitely many functions θi
are identically zero.
(c) For each i there isPan α(i) such that the support of θi is contained in Uα(i) .
(d) For each p ∈ M , i θi (p) = 1 (this sum is in effect finite by (b)).
Proof. The proof is in a sense not very enlightening and we won’t use the techniques
introduced in it again later on, but we feel a moral obligation to supply one, so bear
with us! The result itself, however, is extremely important and will be used over
and over again.
Step 1. We first prove the existence of a compact exhaustion on M , namely, a
sequence Gj , j = 1, 2, . . . of open subsets of M with compact closure Ḡj , and such
18 CHRISTIAN BÖHNING

that Ḡj ⊂ Gj+1 ,

∞
[
M= Gj .
j=1
Indeed, notice first that since M has a countable basis and is locally Euclidean, we
can find a countable basis for the topology of M , Vj , j = 1, 2, 3, . . . , such that each
Vj has compact closure (start with any countable basis and discard those elements
in it that do not have compact closure). Put G1 = V1 . Suppose inductively that
Gk = V1 ∪ · · · ∪ Vjk .
Let jk+1 be the smallest positive integer greater than jk such that
Ḡk ⊂ V1 ∪ · · · ∪ Vjk ∪ · · · ∪ Vjk+1
and put
Gk+1 = V1 ∪ · · · ∪ Vjk ∪ · · · ∪ Vjk+1 .
This gives a compact exhaustion as required.
Step 2. We prove a local result in Rn , namely: let C(r) = {(x1 , . . . , xn ) ∈
n
R | |xi | < r ∀ i} be the open cube of side length 2r and centre at the origin. Then
there exists a non-negative smooth (=C ∞ -)function ϕ on Rn that is equal to 1 on
the closed cube C(1) and zero on the complement of the open cube C(2).
It suffices to do this in one dimension, n = 1 (in the general case, we take a
product of such functions, each one depending only on one coordinate x1 , . . . , xn ).
So let us construct a non-negative smooth function h on the real line R which is
identically 1 on [−1, 1] and zero outside of (−2, 2). There are different ways of doing
this (and the existence of such an h is more than plausible on intuitive grounds of
course). One way to write down a formula is as follows: start with
(
e−1/t t > 0
f (t) =
0 t ≤ 0.
Then put
f (t)
g(t) =
f (t) + f (1 − t)
which is non-negative, smooth, and equal to 1 for t ≥ 1, and 0 for t ≤ 0. Then
setting
h(t) = g(t + 2)g(2 − t)
gives a function h as desired.
Step 3. Suppose now we are given an open cover {Uα } of M . Construct a
compact exhaustion {Gi } for M as in Step 1 and set G0 = ∅. For a point p ∈ M ,
let ip be the largest integer such that p ∈ M − Gip . We also pick an index αp ∈ A
such that p ∈ Uαp . Now let (V, τ ) be a chart around p with τ (p) = 0 and such that
V ⊂ Uαp ∩ (Gip +2 − Gip )
and such that τ (V ) contains the closed cube C(2). (As always, make a picture or
else you will be totally lost now! I am imprisoned in a word processor as they say...)
Now define (
ϕ ◦ τ on V
ψp =
0 elsewhere
 MA3H5 MANIFOLDS 19

with ϕ the smooth function constructed in Step 2. Clearly, ψp is a smooth function

on M taking the value 1 on some open neighbourhood Wp of p and with compact
support contained in V ⊂ Uαp ∩ (Gip +2 − Gip ).
Now for each i ≥ 1 let us choose a finite set of points p ∈ M whose neighbourhoods
Wp as above cover Ḡi − Gi−1 (this is certainly possible since Ḡi − Gi−1 is compact
since it is closed in a compact set). Order all functions ψp corresponding to such
points p (and i = 1, 2, . . . ) in a sequence
ψ1 , ψ2 , ψ3 , . . . .
Clearly, for every point in M only finitely many of the ψj ’s are nonzero at that point
(since the set of ψ’s with support contained in each Gi is finite by construction),
and thus
X∞
ψ= ψj
j=1
is a well-defined smooth function on M with ψ(p) > 0 everywhere. Then define
ψi
θi = .
ψ
Then these θi have all the properties stated in (a)-(d) of the Theorem by construc-
tion. 
We can finally state and prove Ehresmann’s result on fibrations; it is not entirely
clear if the attribution is accurate as the argument occurs in several places in some
guise and might be older. We say that a map of topological spaces is proper if the
preimage of every compact set in the target is compact.
Theorem 3.12. Let π : M → I be a proper submersion of manifolds, with I ⊂ R an
open interval containing 0 ∈ R. Note that all fibres Mt = π −1 (t) are automatically
compact submanifolds of M . Then for any two t1 , t2 ∈ I, the fibres Mt1 and Mt2
are diffeomorphic.
Proof. Since π is a submersion, by Proposition 3.5, b), it is locally around each
point p ∈ M diffeomorphic to the projection Rn → R, or better, we can find local
coordinates x1 , . . . , xn in a neighbourhood U (p) of p and a local coordinate t around
π(p) such that π in these local coordinates is given by (x1 , . . . , xn ) 7→ x1 = t. We
can assume that t is the standard coordinate on R (we proved in b) of Proposition
3.5 that given any local coordinate y around π(p), y ◦ π can be completed to a
system of local coordinates around p). Consider the vector field ∂/∂t on I. By the
preceding argument it is clear that on each U (p) there is a vector field XU (p) with
∂
π∗ XU (p) = .
∂t
We choose a partition of unity {θi } subordinate to the cover of M given by the
{U (p)}, p ∈ M . Write the cover {Uα }α∈A with A some index set. Then
X
X= θi XUα(i)
i

is a global vector field on M with the property π∗ X = ∂/∂t by construction.

The idea now is to look at the 1-parameter group of diffeomorphisms/flow gen-
erated by the vector field X. Suppose t1 < t2 . For a point p ∈ Mt1 consider a
local 1-parameter group of local diffeomorphisms Φ defined on I(p) × U (p). Since
20 CHRISTIAN BÖHNING

π is a submersion, π(U (p)) = V (p) ⊂ I is open (use Proposition 3.5, b)) and since
π∗ X = ∂/∂t, we have that
Φt (p) ∈ Mt1 +t
since π ◦ Φt (p) is an integral curve for the vector field ∂/∂t on R with initial value
for t = 0 equal to t1 . Now since π is proper, K := π −1 ([t1 , t2 ]) ⊂ M is compact. We
investigate the “maximum life-time” of the integral curve γp (t) = Φt (p) starting at
p ∈ Mt1 . More precisely, for any q ∈ M we look at the set of all integral curves γ of
X in a neighbourhood of q with initial condition γ(0) S = q. If Iγ ⊂ R is the domain
of definition of the map γ, then we can set J(q) = γ Iγ , the union being taken over
all such integral curves γ, and obtain an open interval J(q) together with an integral
curve γq : J(q) → M with γq (0) = q that is the maximal integral curve with initial
condition q (in the sense that J(q) is maximal among all such integral curves). Let
J(q) = (ω− , ω+ ), −∞ ≤ ω− < 0, 0 < ω+ ≤ +∞. Then a well-known fact from the
theory of ordinary differential equations, which carries over without change to the
context of manifolds, gives that (t, γq (t)) leaves every compact subset S of R × M
for t → ω− , t → ω+ : indeed, suppose that this was false. Then we can choose a
sequence of times tk → ω+ , k = 1, 2, . . . (same argument for ω− ) with xk = γq (tk ),
(tk , xk ) ∈ S, and passing to a convergent subsequence if necessary (which we can
do since S is compact), we can assume that (tk , xk ) converges to some (t0 , x0 ) ∈ S
(then clearly t0 = ω+ and this must be finite). Now picking a chart around (t0 , x0 ),
we can identify a neighbourhood of (t0 , x0 ) in M with an open subset Ω of Rn .
With our usual deliberate sloppiness of notation, we write (t0 , x0 ) ∈ Ω. By the
Picard-Lindelöf Theorem from MA254 Theory of ODEs it follows that there exist
real numbers r,  > 0 such that the initial value problem corresponding to the given
vector field on M has a solution defined for all times t with
h r ri
t ∈ t0 − , t0 +
2 2
provided the initial point is inside
h r ri  
G = t0 − , t0 + × B̄ x0 ,
2 2 2



(where B̄ x0 , 2 is the closed ball around x0 with radius /2). But for large k,
(tk , xk ) ∈ G, and is arbitrarily close to (t0 , x0 ). This is a contradiction since then
γq (t) cannot be maximal (i.e. the existence interval J(q) is not maximal). Hence
(t, γq (t)) has to leave every compact subset.
Returning to our geometric situation, the integral curve t 7→ Φt (p) cannot be
contained in K! In other words, it will exist at least on a small time interval
containing [t1 , t2 ]. The same is true for Φt (q) for any q ∈ Mt0 with t0 ∈ [t1 , t2 ].
Therefore, we can define a diffeomorphism
Φt2 −t1 : Mt1 → Mt2
using the flow of X. This is a diffeomorphism since its inverse is Φ−(t2 −t1 ) . 
Remark 3.13. As a cultural digression and to emphasise how remarkable Theorem
3.12 is, we point out that it is completely false in the context of complex manifolds:
the complex structure can vary continuously with parameters in a family of complex
manifolds! Riemann called these parameters moduli. Theorem 3.12 on the contrary
shows that differentiable manifolds do not have moduli: the differentiable structure
remains constant in a family {Mt } (over any connected base manifold). Intuitively, a
complex structure on a given differentiable manifold can be thought of geometrically
as a conformal structure, i.e. a rule to measure angles in the manifold. Thus
 MA3H5 MANIFOLDS 21

intuitively it is clear that there are different complex structures on a two-dimensional

torus (tori that are short and fat are conformally different from those that are long
and thin, though they are diffeomorphic). The theory of moduli leads to extremely
rich and interesting objects in complex geometry, called moduli spaces. The reason
why we cannot just mimic the argument in the proof of Theorem 3.12 for complex
manifolds is the absence of (complex analytic) partitions of unity in that set-up.
We return to develop the general theory of tangent spaces and related notions
just a little further for later purposes. If M is a smooth manifold, p ∈ M , and
Tp (M ) the tangent space to M at p, the dual vector space
Tp (M )∗ = HomR (Tp (M ), R)
is called the cotangent space to M at p, or the space of covectors at p (if you have
always be on a war-footing with dual vector spaces, this would be the time to shake
off this paranoia for good because they are here to stay...).
Definition 3.14. A (smooth) 1-form ω on M is an assignment of a covector ωp
at each point of p ∈ M in “a way such that ωp varies smoothly with p”; the last
phrase can be made precise as follows: for each smooth function f on M , we define
its (total) differential dfp at p as the covector satisfying
h(df )p , Xi = Xf, for X ∈ Tp (M ).
Here h·, ·i is just the natural pairing between elements of a vector space and its dual.
We point out that dfp coincides with f∗ introduced in Definition 3.4 when we view
the function as a map of manifolds f : M → R and use the canonical identification
R ' Tf (p) R.
Then, if x1 , . . . , xn is a local coordinate system around p, then
(dx1 )p , . . . , (dxn )p
form a basis for Tp (M )∗ which is just the dual basis to the basis
   
∂ ∂
,..., .
∂x1 p ∂x1 p
This in a neighbourhood of every p ∈ M , every 1-form ω can be uniquely written as
X
ω= fj dxj
j

with the fj functions in a neighbourhood of p. We say that ω is smooth, or that ωp

varies smoothly with p, if all such functions fj are smooth.
In the following 1-form will always mean smooth 1-form. We will use 1-forms
in Section 5 to manufacture more general objects on a manifold, called r-forms,
0 ≤ r ≤ dim M (a 0-form will just be a smooth function on M ). However, for
this we will need to introduce some background from multilinear algebra in Section
4 first, notably the exterior powers Λr V (and exterior algebra) of a vector space
V . An r-form on M will then be a smooth assignment of an element of Λr Tp (M )∗
to every p ∈ M . The top-dimensional forms (r = dim M ) can be used to build
an integration theory on manifolds in Section 5 (since they transform correctly
under change of local coordinates according to the well-known change of variables
formula for integrals in Rd involving the Jacobian determinant; functions on M don’t
transform correctly under change of local coordinates, so you can integrate functions
on a general manifold, sorry!)
22 CHRISTIAN BÖHNING

For the time being, we point out that 1-forms are natural objects occurring in
many applications in the sciences as well; for example, the work of a force field on
R3 is a 1-form acting on displacements (which are tangent vectors at points of R3 ).
It is a remarkable fact that the collection of all tangent vectors to a manifold M
itself forms a manifold (of dimension 2n if dim M = n), called the tangent bundle
T (M ), in a natural way, which we now explain. The same is true for covectors: they
lead to the so-called cotangent bundle T (M )∗ . As sets, as we have mentioned,
G G
T (M ) = Tp (M ), T (M )∗ = Tp (M )∗
p∈M p∈M

(we use the square union symbol to emphasise that these unions are disjoint). We
can then define the two projections
π : T (M ) → M, π(Xp ) = p for Xp ∈ Tp (M ),
∗ ∗
π : T (M ) → M, π(ωp ) = p for ωp ∈ Tp (M )∗ .
Notice that for now these are just maps of sets. M comes with a complete atlas
A = {(U, ϕ)} of charts (U, ϕ). Let ϕ = (x1 , . . . , xn ) with the xi local coordinates on
U . Then define
ϕ̃ : π −1 (U ) → R2n and ϕ̃∗ : (π ∗ )−1 (U ) → R2n
by
ϕ̃(v) = (x1 (π(v)), . . . , xn (π(v)), dx1 (v), . . . , dxn (v)) ,
    
∗ ∗ ∗ ∂ ∂
ϕ̃ (λ) = x1 (π (λ)), . . . , xn (π (λ)), λ ,...,λ
∂x1 ∂xn
for all v ∈ π −1 (U ) and λ ∈ (π ∗ )−1 (U ). Both ϕ̃ and ϕ̃∗ are one-to-one maps onto
open subsets of R2n . One can construct a topology and differentiable structure on
T (M ) (and T (M )∗ ) as follows. We do the case of T (M ) and leave the unproven
assertions as (easy, mechanical, hopefully at worst confusing) exercises:
(1) If (U, ϕ) and (V, ψ) are charts in A, then ψ̃ ◦ ϕ̃−1 is a smooth map.
(2) The collection of
 −1
ϕ̃ (W ) : W open in Rn , (U, ϕ) ∈ A
is a basis for a topology on T (M ) which makes T (M ) into a Hausdorff second-
countable topological space.
(3) The set  −1
(π (U ), ϕ̃) : (U, ϕ) ∈ A
is a smooth atlas on T (M ), and the complete atlas containing this makes
T (M ) into a differentiable manifold.
We will sometimes write the elements of T (M ) as pairs (p, Xp ) and those of T (M )∗
as (p, ωp ). If f : M → N is a smooth map of manifolds, then it is easy to check that
f∗ : T (M ) → T (N ), f∗ ((p, Xp )) = (f∗ )p (Xp )
is a smooth map of manifolds.
In fact, both T (M ) and T (M )∗ even have a little more structure to them than
just the smooth manifold structures, and this comes out automatically from the
above construction: they are vector bundles (of rank n) over M . We now define this
concept precisely. It is ubiquitous in most parts of geometry and shows up whenever
linearising certain structures locally. Intuitively, a rank r vector bundle over M is a
smoothly varying family of r-dimensional (real) vector spaces Vp , p ∈ M , on M .
 MA3H5 MANIFOLDS 23

Definition 3.15. A (smooth, real) vector bundle of rank r on a smooth manifold

M is a smooth manifold E together with a smooth map
π: E → M
onto M such that there exists an open cover {Uj }j∈J of M , where Uj are domains
(j) (j)
of charts on M with local coordinates x(j) = (x1 , . . . , xn ), with the properties:
(a) There exists a diffeomorphism fj between π −1 (Uj ) and Uj × Rr such that
fj
π −1 (Uj ) / Uj × Rr

π πj
 
Uj
id / Uj

commutes, where πj (p, ζ (j) )) = p. (This morally just means that fj preserves
fibres).
(j) (j) (k) (k)
(b) If (p, ζ1 , . . . , ζr ) ∈ Uj × Rr and (p, ζ1 , . . . , ζr ) ∈ Uk × Rr , then
(k)
fj ◦ fk−1 (p, ζ1 , . . . , ζr(k) ) = (p, fjk (p) · (ζ (k) )t )
where  (k) 
ζ1
(k) t  .. 
(ζ ) =  . 
(k)
ζr
and fjk : Uj ∩ Uk → GL(r, R) are smooth maps (here we view GL(r, R) as an
2
open submanifold of Rr ).
We call Ep = π −1 (p) the fibre of E over p. By (a) and (b) we can endow Ep with
a well-defined structure of R-vector space of dimension r for which the fj become
fibrewise linear isomorphisms.
A morphism of vector bundles, sometimes also called a (smooth) map of vector
bundles, E and E 0 of ranks r, r0 over M is a smooth map of manifolds ϕ : E → E 0
that maps the fibre Ep into the fibre Ep0 and such that ϕ |Ep is a linear map of some
constant rank s that is independent of the point p ∈ M . One then defines the notion
of isomorphism in the usual way as a morphism with a two-sided inverse.
A (smooth) section in a vector bundle E over an open subset U ⊂ M is a smooth
map σ : U → E with π ◦ σ = idU .
There are variants of the above notion: for example, to get a smooth, complex
vector bundle of rank r, one would replace Rr by Cr in the above definition etc.
Note that (smooth) sections in T (M ) and T (M )∗ over an open subset U are
nothing but (smooth) vector fields on U and (smooth) 1-forms on U as defined
earlier.
Vector bundles on a manifold M can be (and are often) (re-)constructed starting
from an open cover {Uj }j∈J as in Definition 3.15 and smooth functions
fjk : Uj ∩ Uk → GL(r, R)
satisfying
(∗) fij ◦ fjk = fik on Ui ∩ Uj ∩ Uk for all i, j, k.
The condition (∗) is called the cocycle condition and an open cover {Uj }j∈J together
with a system of functions satisfying the cocyle condition is called a GL(r, R)-cocycle
24 CHRISTIAN BÖHNING

on M . Clearly every vector bundle E gives rise to a GL(r, R)-cocycle on M , but

starting from just a GL(r, R)-cocycle on M , we can also construct a vector bundle
(which gives back E, up to isomorphism, if we take the cocycle coming from E):
indeed, take the disjoint union
G
F̃ = U j × Rr
j

and define an equivalence relation on F̃ by

(j) (k)
(p, ζ1 , . . . , ζr(j) ) ∈ Uj ×Rr ∼ (p, ζ1 , . . . , ζr(k) ) ∈ Uk ×Rr : ⇐⇒ (ζ (j) )t = fjk (p)·(ζ (k) )t .
Then the set of equivalence classes F comes with a natural projection map π : F →
M , as well as with a natural differentiable structure, making F into a vector bundle
over M . We leave it to you as an exercise to work out the details here.
This allows us to construct new bundles from old ones. First note that Definition
3.15 supplies us, given a vector bundle, with a GL(r, R)-cocycle on M associated
to a specific cover {Uj }. It is clear that this also gives a GL(r, R)-cocycle for any
refinement of the given cover, and the GL(r, R)-cocycle associated to the refined
cover determines a vector bundle on M isomorphic to E. Therefore, given two
vector bundles E, E 0 of ranks r and r0 on M , we can always assume they are
determined by a GL(r, R)-cocycle {fjk } and a GL(r0 , R)-cocycle {fjk0 } associated to

the same cover {Uj } of M .

We define the (Whitney) direct sum of E, E 0 , denoted by E ⊕E 0 , to be the vector
bundle of rank r + r0 associated to the GL(r + r0 , R)-cocycle (with cover {Uj }) given
by {gjk }
 
fjk (p) 0
gjk (p) = 0 (p) .
0 fjk
−1 t
The dual bundle E ∗ is defined by the GL(r, R)-cocycle {(fjk ) }. It is easy to
verify that T (M ) is the dual bundle of T (M ) in this sense. The fibre Ep∗ can be
∗

identified with the dual vector space Hom(Ep , R) for every p ∈ M .

We could also define the tensor product E ⊗ E 0 of two vector bundles and the
exterior powers Λr E of a vector bundle at this point explicitly via cocycles, but
this would be a bit too ad hoc and everything will be much clearer after a short
discussion of multilinear algebra.

4. Algebraic interlude: tensor, symmetric and exterior algebras

This Section feels a bit out of place here: it might fit better into a second course
on linear algebra, and may get incorporated into such a course after a curriculum
reform. We will introduce some notions from (multi-)linear algebra that you cannot
expected to be familiar with and that are indispensable to proceed further. We will
try to keep everything down to the bare minimum needed for proceeding. Princi-
pally, we need the exterior algebra of a vector space, and its functorial properties,
in subsequent Sections.
We fix a field K, which you can think of as either R or C (we certainly do not
need anything else in this course later). All vector spaces we will consider will
tacitly be assumed to be K-vector spaces. Moreover, we also assume they are finite-
dimensional over K unless otherwise stated. Much of what follows holds in much
greater generality (for example, many statements still hold if you replace K-vector
 MA3H5 MANIFOLDS 25

spaces by modules over a commutative ring). But this is not a course in algebra,
and we do not think we should strive for maximum generality here.
First, given two vector spaces U , V we define their tensor product U ⊗ V (some-
times also denoted by U ⊗K V if we want to recall the ground field) as follows. Let
F (U, V ) be the vector space which has the set U × V as a basis, i.e., the free vector
space (over K of course as always) generated by the pairs (u, v) where u ∈ U and
v ∈ V . Let R be the vector subspace of F (U, V ) spanned by all elements of the form
(u + u0 , v) − (u, v) − (u0 , v), (u, v + v 0 ) − (u, v) − (u, v 0 ),
(ru, v) − r(u, v), (u, rv) − r(u, v)
where u, u0 ∈ U , v, v 0 ∈ V , r ∈ K.
Definition 4.1. The quotient vector space
U ⊗ V := F (U, V )/R
is called the tensor product of U and V . The image of (u, v) ∈ F (U, V ) under the
projection F (U, V ) → U ⊗ V will be denoted by u ⊗ v. We define the canonical
bilinear mapping
β: U × V → U ⊗ V
by β(u, v) = u ⊗ v. Being very precise, one should refer to the pair (U ⊗ V, β) as the
tensor product of U and V , but usually people just use the term for U ⊗ V with β
tacitly understood.
Sometimes one does not need to know the construction of U ⊗ V when working
with it, but only has to use the following property it enjoys in proofs.
Proposition 4.2. Let W be a vector space with a bilinear mapping ψ : U × V → W .
We say that (W, ψ) has the universal factorisation property for U × V if for every
vector space S and every bilinear mapping f : U ×V → S there exists a unique linear
mapping g : W → S such that f = g ◦ ψ.
Then the couple (U ⊗ V, β) has the universal factorisation property for U × V .
If a couple (W, ψ) has the universal factorisation property for U × V , then (U ⊗
V, β) and (W, ψ) are canonically isomorphic in the sense that there exists a unique
isomorphism σ : U ⊗ V → W such that ψ = σ ◦ β.
Proof. Suppose we are given any bilinear mapping f : U × V → S. Since U × V is
a basis of F (U, V ) we can extend f to a unique linear mapping f 0 : F (U, V ) → S.
Now f 0 vanishes on R since f is bilinear so induces a linear mapping g : U ⊗ V → S
on the quotient. Clearly, f = g ◦ β by construction. The uniqueness of such a map
g follows from the fact that β(U × V ) spans U ⊗ V , so we have no other choice in
defining g.
Now if (W, ψ) is a couple having the universal factorisation property for U × V ,
then by the universal factorisation property of (U ⊗ V, β) (resp. of (W, ψ)), there
exists a unique linear mapping σ : U ⊗ V → W (resp. τ : W → U ⊗ V ) such that
ψ = σ ◦ β (resp. β = τ ◦ ψ). Hence
β = τ ◦ σ ◦ β, ψ = σ ◦ τ ◦ ψ.
Using the uniqueness of the g in the universal factorisation property, we conclude
that τ ◦ σ and σ ◦ τ are the identity on U × V and W respectively. 
This universal property of the tensor product can be used to prove a great many
formal properties of the tensor product in a way that is almost mechanical once one
26 CHRISTIAN BÖHNING

gets practice with it. All these proofs are boring. So we give one, and you can easily
work out the rest for some practice with this.
Proposition 4.3. The tensor product has the following properties.
a) There is a unique isomorphism of U ⊗ V onto V ⊗ U sending u ⊗ v to v ⊗ u
for all u ∈ U , v ∈ V
b) There is a unique isomorphism of K ⊗ U with U sending r ⊗ u to ru for all
r ∈ K and u ∈ U ; similarly for U ⊗ K and U .
c) There is a unique isomorphism of (U ⊗ V ) ⊗ W onto U ⊗ (V ⊗ W ) sending
(u ⊗ v) ⊗ w to u ⊗ (v ⊗ w) for all u ∈ U , v ∈ V , w ∈ W .
d) Given linear mappings
fi : Ui → Vi , i = 1, 2,
there exists a unique linear mapping f : U1 ⊗ U2 → V1 ⊗ V2 such that
f (u1 ⊗ u2 ) = f1 (u1 ) ⊗ f2 (u2 )
for all u1 ∈ U1 , u2 ∈ U2 .
e) There is a unique isomorphism from (U1 ⊕ U1 ) ⊗ V onto (U1 ⊗ V ) ⊕ (U1 ⊗ V )
sending (u1 , u2 ) ⊗ v to (u1 ⊗ v, u2 ⊗ v) for all u1 ∈ U1 , u2 ∈ U2 , v ∈ V .
f ) If u1 , . . . , um is a basis for U and v1 , . . . , vn is a basis for V , then ui ⊗ vj ,
i = 1, . . . , m, j = 1, . . . , n, is a basis for U ⊗ V . In particular, dim U ⊗ V =
dim U dim V .
g) Let U ∗ be the dual vector space to U . Then there is a unique isomorphism g
from U ⊗ V onto Hom(U ∗ , V ) such that
(g(u ⊗ v))(u∗ ) = hu, u∗ iv for all u ∈ U , v ∈ V , u∗ ∈ U ∗ .
h) There is a unique isomorphism h of U ∗ ⊗ V ∗ onto (U ⊗ V )∗ such that
(h(u∗ ⊗ v ∗ ))(u ⊗ v) = hu, u∗ ihv, v ∗ i
for all u ∈ U , u∗ ∈ U ∗ , v ∈ V , v ∗ ∈ V ∗ .
Proof. We prove a) and g) just to illustrate the method, and leave the rest as easy
exercises.
For a) let f : U × V → V ⊗ U be the bilinear mapping with f (u, v) = v ⊗ u. By
the universal property of the tensor product, there is a unique linear mapping
g: U ⊗ V → V ⊗ U
such that g(u ⊗ v) = v ⊗ u. Similarly, there is a unique linear mapping g 0 : V ⊗ U →
U ⊗V with g 0 (v⊗u) = u⊗v. Clearly, g 0 ◦g and g◦g 0 are the identity transformations.
We now prove g) (using f)). Consider the bilinear mapping f : U ×V → Hom(U ∗ , V )
given by
(f (u, v))(u∗ ) = hu, u∗ iv
and apply the universal property of the tensor product in Proposition 4.2. Thus
there exists a unique linear mapping g : U ⊗ V → Hom(U ∗ , V ) such that (g(u ⊗
v))(u∗ ) = hu, u∗ iv. To prove that g is an isomorphism, let u1 , . . . , um be a basis for
U , u∗1 , . . . , u∗m ∈ U ∗ the dual basis, and v1 , . . . , vn a basis for V . We show that
{g(ui ⊗ vj ) : i = 1, . . . , m, j = 1, . . . , n}
is a linearly independent set of vectors. Indeed
X
aij g(ui ⊗ vj ) = 0, aij ∈ K
i,j
 MA3H5 MANIFOLDS 27

gives
X X
0= aij g(ui ⊗ vj )(u∗k ) = akj vj
i,j j

hence all the aij vanish. Since dim U ⊗ V = dim Hom(U ∗ , V ) by f), g is an isomor-
phism. 

We now consider a vector space V and put V ⊗r := V ⊗ · · · ⊗ V (r-times), and set

M
T • (V ) = V ⊗r .
r≥0

If e1 , . . . , en is a basis for V , then

{ei1 ⊗ · · · ⊗ eir : 1 ≤ i1 , . . . , ir ≤ n}
is a basis for V ⊗r , applying f) of Proposition 4.3 inductively. T • (V ) has more struc-
ture than just the structure of a K-vector space (of infinite dimension in general!):
it is a graded R-algebra, associative, but not commutative, if we define the product
(v1 ⊗ · · · ⊗ vr ) · (w1 ⊗ · · · ⊗ ws ) = v1 ⊗ · · · ⊗ vr ⊗ w1 ⊗ · · · ⊗ ws
and extend by K-linearity to all of T • (V ). We call T • (V ) the tensor algebra of V .
Definition 4.4. Let I be the two-sided ideal of T • (V ) generated by all elements of
the form v ⊗ v for v ∈ V . The quotient
Λ• (V ) := T • (V )/I
inherits a natural structure of a graded algebra (since I is a homogeneous ideal),
called the exterior algebra of V .
Similarly, if J denotes the two-sided ideal of T • (V ) generated by all elements of
the form v ⊗ w − w ⊗ v for v, w ∈ V , then
Sym• (V ) = T • (V )/J
is a graded algebra, called the symmetric algebra of V .
We denote the image of v1 ⊗ · · · ⊗ vr in Λ• (V ) by v1 ∧ · · · ∧ vr , and the image in
Sym• (V ) by v1 ·. . .·vr , or simply v1 . . . vr . We will also denote the algebra product in
Λ• (V ) simply by ∧ and call it the wedge product. Similarly, we denote the algebra
product in Sym• (V ) by a dot or simply by concatenation.
The r-th graded component Λr (V ) of Λ• (V ) (resp. Symr (V ) of Sym• (V )) is called
the r-th exterior power of V (resp. r-th symmetric power of V ).
In fact, other types of important algebras can be defined in a similar way as
quotients of the tensor algebra T • (V ) by graded ideals, for example, Clifford algebra.
But in fact, only the exterior algebra
∞
M
Λ• (V ) = Λr (V )
r=0

will concern us in applications in later sections. We only mentioned the symmetric

algebra because it would have weighed too heavily on our conscience if we hadn’t- it
is so important in other contexts. In fact, it is a good exercise to convince yourself
that Sym• (V ) is simply isomorphic to a polynomial algebra K[X1 , . . . , Xn ] with one
variable Xi corresponding to each basis vector ei of V .
28 CHRISTIAN BÖHNING

We now turn to the properties of the exterior algebra we will need later. First of
all it is clear that for any v, w ∈ V we have
v ∧ v = 0, v ∧ w = −w ∧ v,
the first because v⊗v maps to zero under the quotient map T • (V ) → Λ• (V ), and the
second is implied by (v + w) ∧ (v + w) = 0. We say the wedge-product is alternating
or anti-symmetric. More generally, this implies that if ω ∈ Λr (V ) and ϕ ∈ Λs (V ),
then
ω ∧ ϕ = (−1)rs ϕ ∧ ω.
Proposition 4.5. The exterior powers and exterior algebra have the following prop-
erties.
a) If F : V ×· · ·×V → W (r copies of V ) is a multilinear alternating mapping of
vector spaces (which means F (v1 , . . . , vr ) is linear in each argument separately
and zero if two of the vi are equal), then there is a unique linear map
F̄ : Λr (V ) → W
with F̄ (v1 ∧ · · · ∧ vr ) = F (v1 , . . . , vr ).
b) If ϕ : V → W is a linear mapping, there is a unique linear mapping
Λr (ϕ) : Λr (V ) → Λr (W )
with the property
Λr (ϕ)(v1 ∧ · · · ∧ vr ) = ϕ(v1 ) ∧ · · · ∧ ϕ(vr ).
c) If e1 , . . . , en is a basis of V , then
{ei1 ∧ · · · ∧ eir : 1 ≤ i1 < · · · < ir ≤ n}
is a basis of Λr (V ). Consequently,
 
r n
dim Λ (V ) =
r
and Λi (V ) = 0 for i > n. Moreover, note that dim Λn (V ) = 1.
d) If f : V → V is an endomorphism, dim V = n, then the induced map
Λn (f ) : Λn (V ) → Λn (V )
is multiplication by det(f ).
e) For an n-dimensional vector space V we have a natural non-degenerate bilin-
ear pairing
Λr (V ∗ ) × Λr (V ) → K
mapping
(v1∗ ∧ · · · ∧ vr∗ , w1 ∧ · · · ∧ wr ) 7→ det(vi∗ (wj ))1≤i,j≤r
which induces an isomorphism
Λr (V ∗ ) ' (Λr (V ))∗ .
Proof. For a) notice that repeated application of the universal property of the tensor
product furnishes us with a linear map
F̃ : V ⊗r → W
with F̃ (v1 ⊗ · · · ⊗ vr ) = F (v1 , . . . , vr ); this factors over
Λr (V ) = V ⊗r / V ⊗r ∩ I


 MA3H5 MANIFOLDS 29

since the ideal I is generated by elements v ⊗ v that get mapped to zero since F is
alternating.
To prove b) notice that inductive application of Proposition 4.3, d), gives an
induced mapping
⊗r ϕ : V ⊗r → W ⊗r
and this maps V ⊗r ∩ I into the corresponding graded piece of the ideal we divide
out by to get Λr W , so descends to give Λr (ϕ) as desired.
For c) we first show that Λn (V ) ' K via the map induced by the determinant.
Indeed, since the elements v1 ∧ · · · ∧ vr generate Λr (V ), it is clear that, if e1 , . . . , en
is a basis of V , then
{ei1 ∧ · · · ∧ eir : 1 ≤ i1 < · · · < ir ≤ n}
is at least a generating set for Λr (V ). In particular, Λn (V ) is at most one-dimensional,
and exactly one-dimensional, generated by e1 ∧ · · · ∧ en , if we can show it is nonzero.
But by a), the determinant gives a map det : Λn (V ) → K sending e1 ∧ · · · ∧ en to 1.
Now suppose there was a linear dependence relation between the ei1 ∧ · · · ∧ eir :
X
aI e I = 0
I
where we use multi-index notation I = (i1 , . . . , ir ), 1 ≤ i1 < · · · < ir ≤ n, aI ∈
K, eI = ei1 ∧ · · · ∧ eir . For a certain multi-index J = (j1 , . . . , jr ), let J¯ be the
complimentary indices to J in {1, . . . , n}, increasingly ordered. Then
X
( aI eI ) ∧ eJ¯ = ±aJ e1 ∧ · · · ∧ en = 0.
I
Hence all coefficients aJ are zero, proving c).
The endomorphism f : V → V gives a commutative diagram
Λn (f )
Λn (V ) / Λn (V )

det det
 mult(c) 
K /K
where the lower horizontal arrow is multiplication by some constant c. We want to
show that c = det(f ) and for this it suffices to consider what happens to det(e1 ∧
· · · ∧ en ) = 1: this gets mapped to the determinant of the matrix with columns
(f (e1 ), . . . , f (en )), which is det(f ). This proves d).
For e) first note that
det((vi∗ (wj ))1≤i,j≤r
is alternating in both the v1∗ , . . . , vn∗ and the w1 , . . . , wn , and multilinear in these
sets of variables; hence by an application of a), we get a well-defined map
β : Λr (V ∗ ) × Λr (V ) → K
of the type in e). All that remains to prove is that this pairing is nondegenerate,
i.e. that for any nonzero ω ∈ Λr (V ) there is a ψ ∈ Λr (V ∗ ) with β(ψ, ω) 6= 0, and
vice versa, for any nonzero ψ 0 ∈ Λr (V ∗ ) there is an ω 0 ∈ Λr (V ) with β(ψ 0 , ω 0 ) 6= 0.
We prove the first assertion since the second is then proven completely analogously.
If e1 , . . . , en is a basis of V , write in multi-index notation
X
ω= aI eI .
I
30 CHRISTIAN BÖHNING

Since ω 6= 0, there is an aJ 6= 0. Then let ψ = e∗J = e∗j1 ∧ · · · ∧ e∗jr where e∗1 , . . . , e∗n
is the dual basis to e1 , . . . , en . We have β(ψ, ω) = aJ 6= 0 then. 

5. Differential r-forms, integration of forms, exterior derivative,

manifolds with boundary, Stokes’ theorem
Let M be a smooth manifold as usual, p ∈ M , Tp (M ) the tangent space to M at
p.
Definition 5.1. An r-form ω on M (or, more precisely, a smooth r-form, or differ-
ential form of degree r) is an assignment of an element in Λr Tp (M )∗ to each point
p ∈ M such that, if we express ω in terms of local coordinates x1 , . . . , x1 around p:
X
ω= fi1 ...ir (x1 , . . . , xn )dxi1 ∧ · · · ∧ dxir
1≤i1 <···<ir ≤n

then the component functions fi1 ...ir are smooth.

We write A r (M ) for the R-vector space of all r-forms on M (this is also a module
over the smooth functions C ∞ (M ) on M ).
So why are differential forms cool and why should you even bother about them?
There are at least 3 reasons, but more could be given.
(1) Differential forms show up naturally in numerous places in the sciences when
modelling real-life problems, e.g. in physics.
(2) They are the appropriate objects to build an integration theory on manifolds
and vastly generalise familiar theorems from vector calculus (Gauss, Green,
...) and put all of these into one conceptual framework.
(3) They allow us to measure what type of “holes” a space has, via the so-called
de Rham cohomology; in this way, we can attach quantities to manifolds that
are invariant under diffeomorphism and that, in particular, allow us to decide
in many cases that two given manifolds are not diffeomorphic.
We will get to (2) and (3) later in this Section and the next; for (1), just think of
the work of a field along a path or the flux of a fluid through a surface. We will see
more concrete examples later.
Suppose that f : M → N is a smooth map of manifolds and ω ∈ A r (N ), then we
can define the pullback f ∗ ω ∈ A r (M ) of ω by f via
(f ∗ ω)p = Λr ((f∗ )p )∗ (ωf (p) ), p∈M
where Λr ((f∗ )p )∗ is the linear map
Λr ((f∗ )p )∗ : Λr Tf (p) (N )∗ → Λr Tp (M )∗
obtained as the r-th exterior power of the map
(f∗ )∗p : Tf (p) (N )∗ → Tp (M )∗
which in turn is just the dual map of the differential
(f∗ )p : Tp (M ) → Tf (p) (N ).
We will make the pullback of forms more explicit in a second, but first note that
obviously
f ∗ (ω1 + ω2 ) = f ∗ ω1 + f ∗ ω2 , for all ω1 , ω2 ∈ A r (N ),
f ∗ (ω ∧ θ) = (f ∗ ω) ∧ (f ∗ θ), for all ω ∈ A r (N ), θ ∈ A s (N ),
(f ◦ h)∗ ω = h∗ f ∗ ω, where h : M 0 → M , f : M → N are smooth maps, ω ∈ A r (N ).
 MA3H5 MANIFOLDS 31

Let us work out what f ∗ does in local coordinates. Given a smooth map f : M →
N of manifolds, dim M = k, dim N = l, p ∈, we can choose a chart (V, y1 , . . . , yl )
around f (p) and a chart (U, x1 , . . . , xk ) around p with f (U ) ⊂ V ; in terms of these
coordinates we can write f = (f1 , . . . , fl ) with the fi smooth functions of x1 , . . . , xk
on U . Since (f∗ )q , for q ∈ U , is represented in the local coordinates by the Jacobian
matrix  
∂fi
(q)
∂xj 1≤i≤l,1≤j≤k
and ((f∗ )q )∗ is given by the transpose matrix, we get
k
∗
X ∂fi
f dyi = dxj = dfi .
∂xj
j=1

Thus for an r-form ω on V , which we can write as

X
ω= aI dyI
I
(using multi-index notation I = (i1 , . . . , ir ), 1 ≤ i1 < · · · < ir ≤ l, dyI = dyi1 ∧ · · · ∧
dyir , the aI are smooth functions on V ), we obtain
X
f ∗ω = (f ∗ aI )dfI
I
where f ∗ aI = aI ◦ f is the pullback of the function aI and dfI = dfi1 ∧ · · · ∧ dfir .
Now suppose that f : U → V is a diffeomorphism, k = l =: n, and ω = dy1 ∧ · · · ∧
dyn . Then we get
 
∗ ∂fi
f (dy1 ∧ · · · ∧ dyn ) = det dx1 ∧ · · · ∧ dxn .
∂xj 1≤i,j≤k
More specifically suppose now that (Ω, x1 , . . . , xn ) and (Ω0 , y1 , . . . , yn ) are two do-
mains of charts on the same manifold M , U = V = Ω ∩ Ω0 , and f the coordi-
nate change map expressing the y-coordinates in terms of the x-coordinates. If
a(y1 , . . . , yn ) is any function, we thus have the relation on Ω ∩ Ω0
 
∂yi
a(y1 , . . . , yn )dy1 ∧ · · · ∧ dyn = a(y1 (x), . . . , yn (x)) det dx1 ∧ · · · ∧ dxn .
∂xj
This is temptingly reminiscent of the change of variables formula in a multiple
integral in Rn :
Z Z  
∂yi
a(y1 , . . . , yn )dy1 . . . dyn = a(y1 (x), . . . , yn (x)) det dx1 . . . dxn .
∂xj
(Here a is any “nice” integrable function, for example a smooth function with com-
pact support). The only difference is the presence of the absolute value around
det(∂yi /∂xj ). This suggest that we should be able to develop a consistent integra-
tion theory on manifolds, using partitions of unity to piece together local contribu-
tions to an integral over a manifold, provided we can make consistent sign choices
for det(∂yi /∂xj ) to make the local contributions independent of the choice of local
coordinates. This requires the introduction of the concept of an orientation.
Definition 5.2. An n-dimensional manifold M is called orientable if there exists
an everywhere non-vanishing n-form ω on it (meaning ω ∈ A n (M ) and ωp 6= 0 in
Λn Tp (M )∗ for all p ∈ M ). Given an orientable manifold M , an orientation of M
is an equivalence class of non-vanishing n-forms on M where ω ∼ ω 0 if ω = f ω 0
32 CHRISTIAN BÖHNING

with f an everywhere positive smooth function on M . An oriented manifold is an

orientable manifold together with a fixed choice of orientation.
It is thus clear that a connected orientable manifold M has precisely two orien-
tations: the equivalence class of ω and of −ω.
The notion is consistent with the nomenclature introduced in Definition 1.5, c):
Proposition 5.3. A manifold M is orientable if and only if it has a covering by
coordinate charts such that  
∂yi
det >0
∂xj
on each intersection.
Proof. Suppose first that M is orientable, so we are given a nowhere vanishing n-
form ω on M . In a coordinate chart
ω = f (x1 , . . . , xn )dx1 ∧ · · · ∧ dxn
and possibly replacing x1 by −x1 , we can assume that M is covered by such coor-
dinate charts where the function corresponding to f above is positive in each chart.
Then on the overlap of two charts we have
 
∂yi
g(y1 , . . . , yn )dy1 ∧ · · · ∧ dyn = g(y1 (x), . . . , yn (x)) det dx1 ∧ · · · ∧ dxn
∂xj
= f (x1 , . . . , xn )dx1 ∧ · · · ∧ dxn
so if f > 0 and g > 0, we must have det(∂yi /∂xj ) > 0 on the intersection.
Conversely, suppose we are given such a coordinate covering. Choose a partition
of unity θi subordinate to the given cover {Uα }. Put
α(i)
X
ω= θi dy1 ∧ · · · ∧ dynα(i)
i

which is a global smooth n-form on M . Then on Uβ with coordinates y1β , . . . , ynβ we

have !
α(i)
∂yk
dy1β ∧ · · · ∧ dynβ
X
ω |Uβ = θi det β
i ∂yl
 α(i) 
∂yk
and this is nonzero because θi ≥ 0 and det β > 0 by hypothesis. 
∂yl

Suppose now that M is an orientable manifold and that we have moreover chosen
an orientation. We can then define
Z
ω
M
for any n-form ω with compact support on M , in the following way: choose a
covering {Uα } satisfying the condition in Proposition 5.3 (transition functions have
positive Jacobian on intersections of charts) and coordinates yiα on Uα such that
dy1α ∧ · · · ∧ dynα is equivalent to the given orientation form. Then on Uα we can write
ω |Uα = f (y1α , . . . , ynα )dy1α ∧ · · · ∧ dynα .
Choosing a partition of unity {θi } subordinate to {Uα }, we have
θi ω |Uα(i) = gi (y1α , . . . , ynα )dy1α ∧ · · · ∧ dynα
 MA3H5 MANIFOLDS 33

with gi a function which is smooth and has compact support in (an open subset of)
the Rn with coordinates x1 , . . . , xn . Then define
Z XZ XZ
ω := θi ω := gi (x1 , . . . , xn )dx1 dx2 . . . dxn .
M i M i Rn

The local finiteness of the family {θi }, property b) of Theorem 3.11, implies that
only finitely many θi are not identically zero on the compact support of ω, so the
above sum is finite. We leave it to you as an exercise to show this definition is
independent of the partition of unity.
We still owe you to exhibit some examples of orientable manifolds and orienta-
tions. This is rather easy. Clearly Rn is orientable; if x1 , . . . , xn are the standard
global coordinates, then the two orientations correspond to ±dx1 ∧ · · · ∧ dxn .
Suppose that M ⊂ Rn is a subset given by f (x) = c where f : Rn → Rm has
surjective derivative at all points of M , hence is a submersion in a neighbourhood
of M . Such an M is a submanifold either by Proposition 2.4 or Proposition 3.5, b).
If f = (f1 , . . . , fm ) the tangent space Ta (M ) can be identified with the subspace of
Ta (Rn ) of tangent vectors annihilated by the 1-forms df1 , . . . , dfm . Dually, Ta (M )∗
is the quotient of Ta (Rn )∗ by the subspace U spanned by df1 , . . . , dfm . Thus
Λm U ⊗ Λn−m Ta (M )∗ ' Λm U ⊗ Λn−m (Ta (Rn )∗ /U ) ' Λn (Ta (Rn )∗ )
(prove this as an exercise! It’s just linear algebra). Therefore, since df1 ∧ · · · ∧ dfm
gives a non-vanishing section in the Λm U on M , and dx1 ∧ · · · ∧ dxn gives a non-
vanishing section in T (Rn )∗ , the isomorphism furnishes us with a non-zero section
in Λn−m T (M )∗ .
In particular, it follows that all spheres S n are orientable, but obviously we get
tons of other examples as well. However, there are also compact manifolds that
are not orientable (we will give an example in a second). This gives an interesting
conclusion: it can be proved that any compact manifold M of dimension d can be
embedded into some RN for N sufficiently large (we will not do this in this course,
though); thus it follows that not every such M can be defined by N − d global
functions on RN with independent derivatives on M because such an M would
necessarily have to be orientable.
Here is a non-orientable (compact) example: consider the real projective space
PnR with the smooth quotient map
p : S n → PnR
mapping a unit vector in Rn+1 to the one-dimensional subspace it spans (this shows
that PnR is compact since S n is). We have the diffeomorphism
σ : Sn → Sn, σ(x) = −x.
Moreover, if x1 , . . . , xn+1 are global coordinates on Rn+1 and x1 6= 0 at a point of
S n , we can use x = (x2 , . . . , xn+1 ) as local coordinates on S n at that point and the
ratios
(x2 /x1 , . . . , xn+1 /x1 )
as local coordinates around the image point in PnR . With || · || the Euclidean norm
on Rn+1 one thus has in these coordinates
1
p(x) = p x
1 − ||x||2
34 CHRISTIAN BÖHNING

which is smooth with smooth inverse

1
q(y) = p y;
1 + ||y||2
therefore, we can also use x2 , . . . , xn+1 as local coordinates on PnR . We can define a
nowhere vanishing n-form ω on S n by using local coordinates x1 , . . . , xi−1 , xi+1 , . . . , xn+1
around points where xi 6= 0 and putting
1
ω = (−1)i dx1 ∧ · · · ∧ dxi−1 ∧ dxi+1 ∧ · · · ∧ dxn+1
xi
in the chart of S n where xi 6= 0. Note that
n+1
X
xi dxi = 0
i=1
on S n , so when xj 6= 0,
1
dxj = − (x1 dx1 + · · · + xj−1 dxj−1 + xj+1 dxj+1 + · · · + xn+1 dxn+1 )
xj
and substituting this into the above formula for ω on xi 6= 0 gives the analogous
formula for ω on the open set xj 6= 0, so the formula defines a global ω on S n .
If PnR is orientable, there exists a nonzero n-form θ on PnR , and then p∗ θ would be
a nonzero n-form on S n , whence p∗ θ = f ω for a nowhere vanishing smooth function
f on S n . On the other hand
1
σ ∗ ω = (−1)i d(−x1 ) ∧ · · · ∧ d(−xi−1 ) ∧ d(−xi+1 ) ∧ · · · ∧ d(−xn+1 ) = (−1)n−1 ω.
−xi
Thus, since p ◦ σ = p, we get
f ω = p∗ θ = σ ∗ p∗ θ = (f ◦ σ)(−1)n−1 ω.
If n is even, we conclude f ◦ σ = −f , and if f (a) > 0 at some point a of S n ,
f (−a) < 0. Since S n is connected, f must vanish somewhere, contradiction. Thus
P2m
R is not orientable.
On the other hand, if n is odd, we have σ ∗ ω = ω, and this σ-invariant top form
descends to the quotient S n /σ = PnR to give a nowhere vanishing n-form on PnR .
Hence P2m+1
R is orientable.
When a manifold arises as a quotient, such as the torus Tn = Rn /Zn , there is
frequently a better method to prove it is orientable: in the case of the torus, for
example, the n-form dx1 ∧ · · · ∧ dxn on Rn is invariant under all translations in Zn ,
hence descends to a nowhere vanishing n-form on Tn .
It is instructive to notice that you have already integrated over manifolds before
although you didn’t notice it back then.
Example 5.4. Suppose that
ω = f1 dx1 + f2 dx2 + f3 dx3
is a compactly supported smooth 1-form on R3 . Let γ : I → R3 be a diffeomorphism
of some open interval I ⊂ R onto C = γ(I); then C is a 1-dimensional manifold,
and Z Z
ω = γ ∗ ω.
C I
Concretely, if
γ(t) = (γ1 (t), γ2 (t), γ3 (t))
 MA3H5 MANIFOLDS 35

then
dγi
γ ∗ dxi = dγi = ,
dt
whence
Z 3 Z
X dγi
ω= fi (γ(t)) dt.
C I dt
i=1
If we define F to be the vector field (f1 , f2 , f3 ) you will recognise this as the line
integral of F over C (sometimes denoted by
I
F~ dγ

if your lecturer is into calligraphy.

Next consider a compactly supported 2-form on R3
ω = f1 dx2 ∧ dx3 + f2 dx3 ∧ dx1 + f3 dx1 ∧ dx2 .
Suppose we want to integrate ω over a surface S in R3 , which we assume to be
given as (locally) the graph of a smooth function G : R2 → R, x3 = G(x1 , x2 ). Then
h : R2 → S
h(x1 , x2 ) = (x1 , x2 , G(x1 , x2 ))
is a diffeomorphism, and
h∗ dx1 ∧ dx2 = dx1 ∧ dx2 ,
 
∗ ∂G ∂G
h dx2 ∧ dx3 = dx2 ∧ dG = dx2 ∧ dx1 + dx2
∂x1 ∂x2
∂G
=− dx1 ∧ dx2 ;
∂x1
and similarly
∂G
h∗ dx3 ∧ dx1 = − dx1 ∧ dx2 .
∂x2
Therefore Z Z
ω= (n1 f1 + n2 f2 + n3 f3 )dx1 dx2 ,
S R2
where  
∂G ∂G
~n = (n1 , n2 , n3 ) = − ,− ,1 .
∂x1 ∂x2
At any point of the surface x = (x1 , x2 , G(x1 , x2 )) the vector ~n(x) is normal to the
surface S meaning
~n(x) ⊥ Tx (S)
when we view Tx (S) as a subspace of R3 in the natural way (check this!) You will
be familiar with this formula from vector calculus: writing
~n
~u =
|~n|
for the unit normal vector, and letting F~ = (f1 , f2 , f3 ), and defining a smooth 2-form
dA = |~n|dx1 ∧ dx2
we get Z Z
ω= (F~ · ~u) dA.
S R2
36 CHRISTIAN BÖHNING

Here dA is usually called the area form of S (if you don’t recall from vector calculus,
try to figure out why!), and the integral on the right hand-side the flux of the vector
field F~ through the surface S.
We have already seen that for a 0-form on a manifold M , f ∈ A 0 (M ), which is
nothing but a smooth function on M , we can define a 1-form df , the differential of
f . This operation d can be extended in a canonical way to all r-forms and is then
called exterior differentiation. We will see that it provides a key link between the
geometry of a manifold and analysis on the manifold. The operation d is defined in
terms of the manifold structure alone without any additional choices.
Proposition 5.5. Let M be a smooth manifold, A (M ) = r A (M ). There is
r
L
a natural R-linear map d : A (M ) → A (M ), called exterior differentiation or the
exterior derivative, that has the following properties:
(i) d(A r (M )) ⊂ A r+1 (M );
(ii) For every function f ∈ A 0 (M ), df is the differential defined earlier.
(iii) For every ω ∈ A r (M ) and π ∈ A s (M ), we have
d(ω ∧ π) = dω ∧ π + (−1)r ω ∧ dπ.
(iv) d2 = d ◦ d = 0.
Moreover, if x1 , . . . , xn are local coordinates on some open subset U of M and we
write on U X
ω= fi1 ...ir dxi1 ∧ · · · ∧ dxir
i1 <···<ir
then X
dω = dfi1 ...ir ∧ dxi1 ∧ · · · ∧ dxir .
i1 <···<ir

Proof. We will define a map d having all the properties claimed by breaking up ω
as a sum of terms with support in some local coordinate system (using a partition
of unity), then define a d operator locally using a coordinate system, then showing
that the definition is independent of the choice of coordinate system. So we write
X
ω= fi1 ...ir dxi1 ∧ · · · ∧ dxir
i1 <···<ir
and define X
dω = dfi1 ...ir ∧ dxi1 ∧ · · · ∧ dxir .
i1 <···<ir
When r = 0, this is just the differential, so (ii) will be satisfied, and (i) is by
construction. Now
X ∂fi ...i
1 r
dω = dxj ∧ dxi1 ∧ · · · ∧ dxir .
∂xj
j, i1 <···<ir

whence
X ∂ 2 fi1 ...ir
d2 ω = dxk ∧ dxj ∧ dxi1 ∧ · · · ∧ dxir .
∂xk ∂xj
j,k, i1 <···<ir
Now
∂ 2 fi1 ...ir
∂xk ∂xj
is symmetric in k, j, but gets multiplied by dxk ∧ dxj , which is skew-symmetric, so
d2 ω = 0, in accordance with (iv).
 MA3H5 MANIFOLDS 37

We check (iii) for decomposable forms

ω = f dxi1 ∧ · · · ∧ dxir =: f dxI , π = gdxj1 ∧ · · · ∧ dxjs =: gdxJ
and conclude using R-linearity. Now
d(ω ∧ π) = d(f gdxI ∧ dxJ )
= d(f g) ∧ dxI ∧ dxJ
= (f dg + gdf ) ∧ dxI ∧ dxJ
= (−1)r f dxI ∧ dg ∧ dxJ + df ∧ dxI ∧ gdxJ
= (−1)r ω ∧ dπ + dω ∧ π.
So we have constructed an operator d in terms of a chosen coordinate system,
which satisfies the conditions of the Proposition. Now let us write in terms of
another coordinate system y1 , . . . , yn
X
ω= f˜i1 ...ir dyi1 ∧ · · · ∧ dyir .
i1 <···<ir
We then define in the same way
X
d0 ω = df˜i1 ...ir ∧ dyi1 ∧ · · · ∧ dyir .
i1 <···<ir

We now show that then d = d0 (on the intersection), using the properties (i)-(iv).
Now (ii) and (iii) give
X
dω = d( f˜i1 ...ir dyi1 ∧ · · · ∧ dyir )
i1 <···<ir
X X
= df˜i1 ...ir ∧ dyi1 ∧ · · · ∧ dyir + f˜i1 ...ir d(dyi1 ∧ · · · ∧ dyir )
i1 <···<ir i1 <···<ir

and from (iii)

d(dyi1 ∧ · · · ∧ dyir ) = d(dyi1 ) ∧ · · · ∧ dyir − dyi1 ∧ d(dyi2 ∧ · · · ∧ dyir ).
By (iv) the first summand is zero and continuing in the same way, the whole ex-
pression is seen to be zero. Thus
X
dω = df˜i1 ...ir ∧ dyi1 ∧ · · · ∧ dyir = d0 ω
i1 <···<ir
is independent of the coordinate system chosen and hence globally well-defined. 
The operation d commutes with pullback:
Proposition 5.6. If f : M → N is a smooth map of manifolds, and ω ∈ A r (N ),
then
d(f ∗ ω) = f ∗ (dω).
Proof. We know that f∗ : Tp (M ) → Tf (p) (N ) satisfies f∗ (Xp )(ϕ) = Xp (ϕ ◦ f ) (where
ϕ is a smooth function defined locally around f (p)). Therefore the dual map
(f∗ )∗ : Tf (p) (N )∗ → Tp (M )∗
satisfies
(f∗ )∗ (dϕ)f (p) = d(ϕ ◦ f )p

(you should recall the definition of the differential of a function dψp : h(dψ)p , Xp i =
Xp (ψ)). Thus
f ∗ (dϕ) = d(ϕ ◦ f ) = d(f ∗ ϕ)
38 CHRISTIAN BÖHNING

so our claim holds for 0-forms (functions) on N . Now in general in terms of local
coordinates y1 , . . . , yr on N
X
ω= gi1 ...ir dyi1 ∧ · · · ∧ dyir
i1 <···<ir
whence X
f ∗ω = f ∗ (gi1 ...ir )f ∗ dyi1 ∧ · · · ∧ f ∗ dyir
i1 <···<ir
and therefore
X
d(f ∗ ω) = d(f ∗ (gi1 ...ir )) ∧ f ∗ dyi1 ∧ · · · ∧ f ∗ dyir
i1 <···<ir
X
= f ∗ d(gi1 ...ir ) ∧ f ∗ dyi1 ∧ · · · ∧ f ∗ dyir
i1 <···<ir
∗
= f (dω).

Let us work out what d does on R3 ;
we will re-encounter some old friends (and if
not friends, at least acquaintances) in this way.
Example 5.7. If f is a function on R3 (with coordinates x1 , x2 , x3 ), then
df = g1 dx1 + g2 dx2 + g3 dx3
where  
∂f ∂f ∂f
(g1 , g2 , g3 ) = , , = grad (f )
∂x1 ∂x2 ∂x3
the gradient vector field of f .
If
ω = f1 dx1 + f2 dx2 + f3 dx3
is a 1-form, then
dω = df1 ∧ dx1 + df2 ∧ dx2 + df3 ∧ dx3
= g1 dx2 ∧ dx3 + g2 dx3 ∧ dx1 + g3 dx1 ∧ dx2
where
∂f3 ∂f2 ∂f1 ∂f3 ∂f2 ∂f1
g1 =− , g2 = − , g3 = − .
∂x2 ∂x3 ∂x3 ∂x1 ∂x1 ∂x2
Thus if F and G are the vector fields (f1 , f2 , f3 ) and (g1 , g2 , g3 ), then G = curl F .
For a 2-form
ω = f1 dx2 ∧ dx3 + f2 dx3 ∧ dx1 + f3 dx1 ∧ dx2
we get
 
∂f1 ∂f2 ∂f3
dω = + + dx1 ∧ dx2 ∧ dx3
∂x1 ∂x2 ∂x3
= (div (F ))dx1 ∧ dx2 ∧ dx3 .

Of course d of any 3-form is zero, and all r-forms, r > 3 are zero on R3 .
From d2 = 0 we thus get
curl (grad f ) = 0, div(curl F ) = 0
which is well-known to you (but d2 = 0 is certainly easier to remember than those
two formulas!)
 MA3H5 MANIFOLDS 39

It turns out that the connection between what you already know from vector
calculus and differential forms can be carried much further: the classical theorems
of Stokes and Green and Gauss in vector calculus are special cases of a single result
in the theory of differential forms, which is also called “Stokes’ theorem” (in the
present form it was formulated in 1945 by Élie Cartan following earlier work by
Volterra, Goursat and Poincaré on generalisations of the classical integral theorems
of vector calculus. Apparently this is an example of Arnold’s principle (named after
the famous Russian mathematician V.I. Arnold): If a famous result or notion bears
a personal name, then this name is not the name of the discoverer).
We first prove a simple version of the result and the full version afterwards.
Theorem 5.8. Let M be an oriented n-dimensional manifold (i.e., M is orientable
and we have chosen one of the two possible orientations). If ω ∈ A n−1 (M ) is a
differential form with compact support, then
Z
dω = 0.
M
Proof. Choose a coordinate covering of M (with positive Jacobian determinant of
the transition functions on intersections) and a partition of unity {θi } subordinate
to this covering to write X
ω= θi ω.
i
Then in a chart we can write
θi ω = f1 dx2 ∧ dx3 ∧ · · · ∧ dxn − f2 dx1 ∧ dx3 ∧ · · · ∧ dxn + . . .
whence  
∂f1 ∂fn
d(θi ω) = + ··· + dx1 ∧ · · · ∧ dxn .
∂x1 ∂xn
From the definition of the integral over M , we simply need to sum all contributions
Z  
∂f1 ∂fn
+ ··· + dx1 dx2 . . . dxn .
Rn ∂x1 ∂xn
Now consider Z
∂f1
dx1 dx2 . . . dxn .
Rn ∂x 1
Using Fubini’s theorem to evaluate this as a repeated integral we get that this equals
Z Z Z
∂f1

... dx1 dx2 dx3 . . . dxn .
R R R ∂x1
But since f1 has compact support, it vanishes for |x1 | ≥ N , N large, so
Z
∂f1
dx1 = [f1 ]N
−N = 0.
R ∂x 1
Continuing in the same way, we see that the entire integral is thus zero. 
As you will recall from vector calculus, Green’s theorem relates an area integral
(in the plane) to a line integral (and so does the classical Stokes’ theorem for surfaces
in 3-space), and Gauss’s theorem relates a surface integral to a volume integral. To
extend such theorems to the manifold context, we thus have to enlarge the class of
objects under study to include spaces like a closed ball in Rn+1 (with boundary the
sphere S n ), and the cylinder S 1 × [0, 1] in R3 which has boundary two copies of the
circle. We will call these manifolds with boundary and we need to integrate over
them.
40 CHRISTIAN BÖHNING

Definition 5.9. Define the upper half-space H n in Rn as a set by

H n = {(x1 , . . . , xn ) ∈ Rn | xn ≥ 0}
and endow it with the induced topology from Rn . For U ⊂ H n open, we call
f : U → Rn smooth if it is locally around every point of its domain the restriction of
a smooth function defined on some open set of Rn . A local diffeomorphism on H n
is a smooth map f : U → V with U, V ⊂ H n open, that has a smooth inverse. The
local diffeomorphisms form a pseudogroup ΓH n ,dif f on H n .
A manifold with boundary M (of dimension n) is nothing but a ΓH n ,dif f -manifold
in the sense of Definition 1.5. More concretely, for the convenience of the reader, we
recall that such a structure is determined by an atlas {(Uα , ϕα } on M that has the
following properties:
S
a) M = α Uα ;
b) ϕα : Uα → ϕα (Uα ) ⊂ H n is a homeomorphisms between Uα and an open
subset ϕα (Uα ) of H n .
c) ϕβ ◦ ϕ−1
α : ϕα (Uα ∩ Uβ ) → ϕβ (Uα ∩ Uβ ) is a smooth map between open subsets
of H n for all α, β (in the sense defined above: it is the restriction of a C ∞ -map
from a neighbourhood in Rn of ϕα (Uα ∩ Uβ ) ⊂ H n ⊂ Rn to Rn ).
We define the boundary of M by
∂ M = {p ∈ M : ϕα (p) ∈ {(x1 , . . . , xn−1 , xn ) ∈ Rn | xn = 0} for some chart (Uα , ϕα )} .
It is immediate to check that this is independent of the chart, and that {(Uα |∂M
, ϕα |∂M )} defines a smooth atlas on ∂M making it into a smooth manifold of
dimension n − 1 (not a manifold with boundary! we mean an actual manifold).
The following are some obvious examples of manifolds with boundary.
(1) H n itself, with boundary ∂H n = {xn = 0}.
(2) The closed unit ball {x ∈ Rn | ||x|| = 1} with boundary S n−1 .
(3) The cylinder [0, 1] × S 1 with boundary two copies of S 1 .
We can now define differential forms on manifolds with boundary in a straight-
forward way. Namely, in some local chart, they are just restrictions of smooth forms
on some open subset of Rn to H n . A differential form on M is then simply given
by a collection of differential forms in the local charts of some atlas, transforming
correctly under the transitions functions from one chart to the other. Moreover,
a form on M restricts to a form on the boundary ∂M . We also have the notions
of an orientable manifold with boundary (one which admits a nowhere vanishing
global differential form of top degree), and an oriented manifold with boundary in
complete analogy to the situation for manifolds.
[If a student feels uncomfortable with the few words we said here, they should
work out more details for themselves, but we also want to warn them that nothing
but boredom awaits them in that enterprise.]
Proposition 5.10. If a manifold with boundary M is orientable, then ∂M is ori-
entable.
Proof. If dim M = 1, ∂M = 0 and consists of points, thus the assertion is trivial.
So let dim M ≥ 2 in what follows. Then the same argument as in the proof of
Proposition 5.3 shows that we can choose a local coordinate systems x1 , . . . , xn
on M such that ∂M is defined by xn = 0 locally and on intersections with other
charts det(∂yi /∂xj ) > 0. Where two local charts, with coordinates x1 , . . . , xn and
 MA3H5 MANIFOLDS 41

y1 , . . . , yn overlap
yi (x1 , . . . , xn ), yn (x1 , . . . , xn−1 , 0) = 0,
so the Jacobian matrix at a point of the boundary has the form
∂y1 /∂x1 ∂y1 /∂x2 . . . ∂y1 /∂xn
 
.. .. .. ..

 . . . .


 .. .. .. .. 
.

 . . . . 
 .. .. .. .. 
 . . . . 
0 0 . . . ∂yn /∂xn
Now ϕβ ϕ−1α maps xn > 0 to yn > 0, thus if xn = 0, then yn = 0, and if xn > 0,
yn > 0. Therefore
∂yn
|x =0 > 0
∂xn n
and since the determinant of the preceding matrix is positive, the determinant of the
upper left (n − 1) × (n − 1)-matrix is positive on xn = 0. Thus the (Uα |∂M , ϕα |∂M )
give an atlas for ∂M whose transition functions have everywhere positive Jacobian,
so ∂M is orientable. 
Once we have chosen a specific orientation for M , we will always choose a certain
orientation of ∂M (one of the two possible ones) that is compatible with the orien-
tation of M in a way we now make precise. For surfaces in R3 , for example, one can
think of the convention we are about to make as the choice of an outward or inward
normal.
Definition 5.11. Suppose M is an orientable manifold with boundary for which
we have chosen a specific orientation, represented by a nowhere vanishing n-form ω.
If in local coordinates around a boundary point
ω '  dx1 ∧ · · · ∧ dxn
where  = ±1 and with xn ≥ 0 defining M locally, then we define the induced
orientation ω 0 of ∂M locally by
(−1)n  dx1 ∧ · · · ∧ dxn−1 .
Example 5.12. It is worthwhile to think about what this means, especially the
funny sign (−1)n ; for example, if we take the closed interval M = [0, 1] ⊂ R with
coordinate x, we see that ω = dx is a nowhere vanishing 1-form on [0, 1], giving it an
orientation. At 0, the local coordinate x1 = x is satisfied that M is locally given by
x1 ≥ 0 and ω ' dx1 , so the point 0 gets the induced orientation −1 (since n = 1).
Note that a choice of orientation for a point is simply the choice of +1 or −1 in R.
At the point 1, instead we can use as local coordinate y1 = 1 − x, whence locally
around 1, M is given by y1 ≥ 0, and now ω ' −dy1 whence the point 0 gets
the induced orientation +1. Note that this already indicates that our choice of
orientation of the boundary is well-suited for generalising the fundamental theorem
of differential and integral calculus: indeed,
Z 1
f 0 (x)dx = f (1) − f (0)
0
and this can be interpreted as saying that integrating the differential of a function
over M is the same as “integrating the function over the boundary”, which in this
case is just evaluating the function at the points of the boundary ∂M , with the
42 CHRISTIAN BÖHNING

boundary orientation taken into account: f (1) enters with a plus-sign, and f (0)
with a minus-sign.
More generally, if we consider H n ⊂ Rn with the standard basis e1 , . . . , en in Rn
and dual basis dx1 , . . . , dxn , we can give H n an orientation by taking the nowhere
vanishing n-form ω = dx1 ∧ · · · ∧ dxn . We call an ordered basis (v1 , . . . , vn ) in Rn
positively oriented (with respect to ω) if
ω(v1 ∧ · · · ∧ vn ) > 0.
The ω0 equivalent to ω are precisely those having positive values on a positively ori-
ented basis, and so the choice of a positively oriented basis in Rn is equivalent to the
choice of an orientation. The induced orientation on ∂H n = Rn−1 = he1 , . . . , en−1 i
in this case is given by ω∂H n = (−1)n dx1 ∧ · · · ∧ dxn−1 : this means that a basis
(w1 , . . . , wn−1 ) of ∂H n = Rn−1 is positively oriented with respect to ω∂H n if and
only if
(n, w1 , . . . , wn−1 )
is a positively oriented basis with respect to ω in Rn , where n is the outward normal
n = −en to H n at 0 ∈ Rn .
We can now prove the full version of Stokes’ theorem; in part its power
R lies in the
fact that it establishes a connection between the analytical operations and d and
the geometric operation ∂ of forming the (oriented) boundary of a manifold with
boundary.
Theorem 5.13 (Stokes’ Theorem, which, as you will recall, is not entirely due to
Stokes...). Let M be an n-dimensional oriented manifold with boundary ∂M carrying
the induced orientation. Let ω ∈ A n−1 (M ) be an (n − 1)-form of compact support.
Then Z Z
dω = ω |∂M .
M ∂M
R
Proof. We need to start with a word about how the integral M dω over the ori-
ented manifold M with boundary is defined; we completely mimic the procedure for
manifolds without boundary: if ϕ is the orientation form on M , as in the proof of
Proposition 5.3, we can write
ϕ = ϕ(x1 , . . . , xn )dx1 ∧ · · · ∧ dxn
in local charts, and provided ϕ is positive in every chart get a cover by charts of
this type such that the Jacobian of the transition functions is positive on overlaps.
We can then give a well-posed definition of the integral just as before: choose a
partition {θi } subordinate to such a cover and repeat the construction described
after the proof of Proposition 5.3. However, if you have been super vigilant, there is
one small catch here: to make ϕ positive in each chart, we may have to replace x1
by the local coordinate −x1 , and if n = 1, we cannot do that at a boundary point
because it would mess up the requirement that M be locally defined by xn ≥ 0 at a
boundary point! Thus if the dimension of M is 1 (and only then) are we seemingly
in trouble and cannot proceed in complete analogy with what we did before. It is
easy to sort out this small inconvenience and we leave it to the student to do this
as a nice and easy exercise to test his understanding of the material (maybe the
easiest fix is to allow local models for manifolds with boundary M that are either
open sets in the upper half space H n = H+ n = {x ∈ Rn | x ≥ 0} or the lower
n
n n
half-space H− = {x ∈ R | xn ≤ 0}; then one can always cover M by such charts
with positive Jacobian on overlaps and dx1 ∧ · · · ∧ dxn equivalent to the chosen
 MA3H5 MANIFOLDS 43

orientation of M in each chart). In any case, for n = 1, Theorem 5.13 is nothing

but the one-dimensional calculus assertion
Z b
f 0 (x) dx = f (b) − f (a)
a

so we proceed undeterred and give a proof now that works for all n ≥ 2 without the
need for modification.
P
As in the proof of Theorem 5.8, we write again ω = i θi ω whence
Z XZ
dω = d(θi ω)
M i M

and write
θi ω = f1 dx2 ∧dx3 ∧· · ·∧dxn −f2 dx1 ∧dx3 ∧· · ·∧dxn +· · ·+(−1)n−1 fn dx1 ∧· · ·∧dxn−1 .
There are now two types of open sets among the coordinate charts in use here: those
that do not intersect ∂M , and those which do intersect it. For those which do not
intersect it, the contribution to the integral is zero by what was shown in the proof
of Theorem 5.8. For those which do, we have
Z Z  
∂f1 ∂fn
d(θi ω) = + ··· + dx1 dx2 . . . dxn
M xn ≥0 ∂x1 ∂xn
Z
= [fn ]∞
0 dx1 . . . dxn−1
Rn−1
Z
=− fn (x1 , . . . , xn−1 , 0)dx1 . . . dxn−1
n−1
Z R
= θi ω |∂M
∂M

where the last equality follows because

θi ω |∂M = (−1)n−1 fn dx1 ∧ · · · ∧ dxn−1
and we use the induced orientation
(−1)n dx1 ∧ · · · ∧ dxn−1 .


If our endeavour here is likened to a mountain hike, you should not view Stokes’
Theorem as a peak or pinnacle that affords a beautiful view of the scenery and
everything that surrounds it, but rather as a solid pair of mountain boots. In other
words, we are after geometric applications. Here is a first one, we will give many
more in the next Section.
Theorem 5.14 (Brouwer’s fixed point theorem). Let B be the closed unit ball in
Rn and f : B → B a smooth map from B to itself. Then f has a fixed point.
Proof. Arguing by contradiction, suppose there is no fixed point, so f (x) 6= x for all
x ∈ B. Extend the segment f (x)x to a half ray starting from f (x) and intersecting
the boundary S n−1 of B in g(x). Then
g : B → ∂B
44 CHRISTIAN BÖHNING

is a smooth map with the property that g(x) = x for x ∈ ∂B. Now let ω be a
nowhere vanishing (n − 1)-form on ∂B = S n−1 (we have seen above that such an ω
exists), normalised so that Z
ω = 1.
∂B
Then Z Z
1= ω= g∗ω
∂B ∂B
since g is the identity on ∂B. By Stokes’ Theorem
Z Z Z
∗ ∗
g ω= d(g ω) = g ∗ (dω) = 0
∂B B B
using Proposition 5.6 and the fact that dω = 0 as ω is a form in dimension n − 1 on
S n−1 . This is a contradiction (1=0) and hence f must have a fixed point. 

6. Applications of Stokes’ theorem, de Rham cohomology

From the point of view of geometric imagination, it is intuitively clear that a
sphere, the surface of a solid ball, has a hole, the hollow in its interior; and likewise
the surface of a rubber tyre (or doughnut), a torus, has another kind of hole, or even
two holes, the space where the air of the tyre is, and the space you can put your
arm through if you remove the bike spokes. These are certainly different, though,
from the type of holes in a plane with two punctures. On the other hand, Rn itself
apparently does not have any holes. This can be made precise and quantified by
certain objects, called (de Rham) cohomology groups. In combination with Stokes’
Theorem these will lead us to a lot of interesting geometric insights about different
types of manifolds.
Definition 6.1. Let M be a manifold. An r-form α ∈ A r (M ) is said to be closed
if dα = 0, and exact if there exists an (r − 1)-form β such that dβ = α. We denote
the R-vector subspace of A r (M ) consisting of the closed forms by
F r (M )
(the “F” is for “fermée”, paying homage to our good friends and neighbours the
French), and the R-vector subspace of A r (M ) consisting of the exact forms by
B r (M )
(since elements in the image of d are also called boundaries and d the boundary
operator, the “B” stands for “bords”, paying homage to you can guess who). The
quotient vector space
H r (M ) = F r (M )/B r (M )
is called the r-th de Rham cohomology group of M ; note that since d2 = 0, B r (M )
is a subspace of F r (M ).
Actually H r (M ) is a real vector space, not just an additive group, but people
usually refer to it as a de Rham cohomology group since there are a lot of other
instances in mathematics where similar cohomology groups make their appearance
and in these other situations they are frequently really just abelian groups without
a vector space structure.
It is easy to figure out what H 0 (M ) signifies: indeed, this is the space of smooth
functions f on M satisfying df = 0; on each connected component Mi of M (an
open submanifold), such an f must be constant because a function on an open ball
 MA3H5 MANIFOLDS 45

in Rn with vanishing gradient is constant, hence the subset of points p ∈ Mi where

f (p) = c is both open and closed, whence equal to Mi . Hence if M is connected,
H 0 (M ) = R, and if M is a manifold with k connected components, H 0 (M ) = Rk (of
course there could also be infinitely many connected components, and then H 0 (M )
is the free R-vector space on the components of M ).
The next result summarises some basic formal and functorial properties of the de
Rham groups that are indispensable to get the theory off the ground.
Theorem 6.2. Let M be a manifold of dimension n. Then
(i) For r > n we have H r (M ) = 0.
(ii) The totality
M
H ∗ (M ) = H r (M )
r
is an R-algebra with product between a = [α] ∈ H r (M ) and b = [β] ∈ H s (M )
(where we denote by [α] is the equivalence class of the closed r-form α etc.):
ab = [α ∧ β].
The product on H ∗ (M ) thus defined is not commutative, but what is some-
times called graded commutative meaning
ab = (−1)rs ba.
(iii) If f : M → N is a smooth map of manifolds, pull-back of differential forms
induces an R-linear map
f ∗ : H ∗ (N ) → H ∗ (M )
compatible with the product (in the sense that f ∗ (cd) = f ∗ (c)f ∗ (d); moreover,
this operation is functorial in the sense that id∗M is the identity on H ∗ (M )
and for two smooth maps
f g
M /N /P

we have f ∗ ◦ g ∗ = (g ◦ f )∗ . In particular, if M and N are diffeomorphic via

f , f ∗ : H r (N ) → H r (M ) is an isomorphism for all r.
Proof. Property (i) follows since there are no nonzero r-forms on a manifold M for
r > n = dim M .
For (ii), everything follows immediately from properties of the exterior product
of forms, but we need to check that
ab = [α ∧ β]
is well-defined and really does define a cohomology class; in other words, what we
have to check is that α ∧ β is closed and that its cohomology class does not change
if we change α or β by an exact form. For the first part, if α and β are closed
d(α ∧ β) = dα ∧ β + (−1)r α ∧ dβ = 0
so α ∧ β is closed as well. For the second part, suppose we put α0 = α + dγ. Then
α0 ∧ β = (α + dγ) ∧ β = α ∧ β + d(γ ∧ β)
since dβ = 0, hence the cohomology class of α0 ∧ β is the same as that of α ∧ β.
46 CHRISTIAN BÖHNING

Finally, for (iii), one only needs to check that pullback of differential forms really
descends to cohomology groups; we need that the pull-back of a closed form α is
closed. This follows from
d(f ∗ α) = f ∗ (dα) = 0.
And we need that if we change α by an exact form, replacing it by α0 = α + dγ, the
pullback f ∗ α0 still represents the same cohomology class as f ∗ α: again, this follows
from the compatibility of f ∗ and d
f ∗ (α + dγ) = f ∗ α + d(f ∗ γ).
Since f ∗ (α ∧ β) = f ∗ (α) ∧ f ∗ (β), the operation f ∗ respects the product we defined.


We now come to a result that has more geometric content and says that the in-
duced maps ft∗ on cohomology do not vary if ft : M → N is a family of smooth maps,
smoothly varying with a parameter t. This is sometimes referred to as homotopy
invariance of de Rham cohomology. Precisely, it means the following.
Theorem 6.3. Let f : M × [0, 1] → N be a smooth map (by which of course we
mean it is the restriction of some smooth map M × (−, 1 + ) → N for some small
 > 0). Put ft (x) = f (x, t). The for the induced maps ft∗ : H r (N ) → H r (M ) we
have
f0∗ = f1∗ .
Proof. We start with a class a ∈ H r (N ) and represent it by a closed r-form α; let
us consider f ∗ α, an r-form on M × [0, 1]. We can write uniquely
f ∗ α = β + dt ∧ γ
with β an r-form on M , depending smoothly on t, and γ an (r − 1)-form on M ,
depending smoothly on t. Here β is nothing but ft∗ α. Denote by dM the exterior
derivative on M . Then from the preceding displayed formula we get
∂β
0 = dM β + dt ∧ − dt ∧ dM γ
∂t
because α is closed. Therefore we must have
∂β
= dM γ.
∂t
Since also
∂ ∗ ∂β
ft α =
∂t ∂t
integration with respect to t yields
Z 1 Z 1
∗ ∗ ∂ ∗
f1 α − f0 α = ft α = dM γ dt.
0 ∂t 0
Thus the cohomology class of f1∗ α is the same as that of f0∗ α since their difference
is exact on M . 

As an application we immediately get

Corollary 6.4. For M = Rn we have
H 0 (Rn ) = R, H r (Rn ) = 0 for r > 0.
 MA3H5 MANIFOLDS 47

Proof. Since Rn is connected, H 0 (Rn ) = R. Now consider the map

f : Rn × [0, 1] → Rn
given by f (x, t) = tx. Then f1 is the identity, and f0 is the constant map to the
origin 0 ∈ Rn . In particular, any r-form for r > 0 pulls back to 0 via f0 because
its derivative vanishes identically. Therefore Theorem 6.3 implies H r (Rn ) = 0 for
r > 0. 
Call a region Ω in Rn star-shaped (with respect to a point p ∈ Ω) if for each x ∈ Ω
the line segment px joining p and x lies entirely in Ω. Then one can use the same
argument as in the proof of Corollary 6.4 to show
H 0 (Ω) = R, H r (Ω) = 0 for r > 0
writing down a smooth family ft : Ω → Ω with f0 mapping everything to p and f1
the identity (write down the details as an exercise if you have trouble seeing it).
This result is usually called the Poincaré lemma.
Moreover, the same type of argument applied to the family of maps
ft : M × Rn → M × Rn
with ft (a, x) = (a, tx) shows that for all manifolds M
H r (M × Rn ) ' H r (M ) for all r.

Computing de Rham cohomology groups is usually not very easy, but we are
now already in a position to calculate them for some “easy” manifolds, for example,
spheres.
Theorem 6.5. Let S n be the n-sphere in Rn+1 . Then
H 0 (S n ) = H n (S n ) = R and H r (S n ) = 0 otherwise.
Proof. We start with the case n = 1 and then make an induction.
Thus consider S 1 ' R1 /Z first. The 1-form dx on R is translation-invariant, hence
descends to a nowhere vanishing (closed) 1-form, also denoted dx, on R/Z. This dx
is not the differential of a function f on S 1 because such a function would have to
have a minimum on the compact S 1 and at such a point we would have df = 0, but
dx is nowhere vanishing. Hence
H 1 (R/Z) 6= 0.
Now suppose g(x)dx is any 1-form on S 1 (automatically closed); here g can be
viewed as a periodic function on R: g(x + 1) = g(x). If we want to write α = df on
S 1 we must have f 0 (x) = g(x) on R whence we can put
Z x
f (x) = g(s) ds.
0
But we need f periodic f (x + 1) = f (x) for it to descend to S 1 whence we must
have Z 1
g(s) ds = 0.
0
Thus if α = g(x) dx with g some smooth periodic function on R (not necessarily
integral 0 over [0, 1]) we can write
Z 1 
α = g(x) dx = g(s) ds dx + df
0
48 CHRISTIAN BÖHNING
R1 Rx
because g̃ = g− 0 g(s) ds has integral zero over [0, 1], and we can put f = 0 g̃(s) ds.
Thus any 1-form on S 1 is of the form
c dx + df
with c ∈ R some constant and f a function on S 1 , whence H 1 (S 1 ) = R.
Now we assume by induction that the statement of the Theorem holds for all
spheres S j for j up to dimension n − 1 and we prove it for S n . As S n is connected,
H 0 (S n ) = R. We now prove first that H p (S n ) = 0 for all 1 < p < n. For this let α be
a closed p-form on S n representing some cohomology class in H p (S n ). The type of
argument we will employ now has been formalised in the form of the Mayer-Vietoris
exact sequence in topology, but we give a proof without the technical baggage. Let
U and V be the complements in S n of small closed balls in Rn+1 around the North
and South pole of S n . Then U and V are diffeomorphic to open balls in Rn via
stereographic projection, which are star-shaped with respect to the center, hence
they have trivial higher de Rham cohomology. Thus the restriction of α to U and
V is exact:
α = du on U and α = dv on V .
On U ∩ V , u − v is closed since d(u − v) = α − α = 0 on the intersection. Since
U ∩ V ' S n−1 × R
and H p−1 (S n−1 × R) = H p−1 (S n−1 ) = 0 by induction, u − v is exact on U ∩ V :
u − v = dw on U ∩ V .
Now U ∩ V is also diffeomorphic to S n−1 × (−2, 2) and we can choose a smooth
function ϕ that is equal to 1 on S n−1 × (−1, 1) and has support in S n−1 × (−2, 2).
Shrinking U and V slightly to get U 0 ⊂ U and V 0 ⊂ V such that U 0 ∩ V 0 =
S n−1 × (−1, 1), we can extend ϕw by 0 outside of its support to get a globally
defined (p − 2)-form on S n . Moreover we have the (p − 1)-form u on U 0 and the
(p − 1)-form
v + d(ϕw)
on V 0 that satisfy u = v + dw = v + d(ϕw) on U 0 ∩ V 0 . Thus we can patch these
together to get a (p − 1)-form β on all of S n with α = dβ globally (because this
holds on the open cover U 0 , V 0 be construction). Thus the cohomology class of α is
trivial and we have shown
H p (S n ) = 0 for 1 < p < n.

Now what about p = 1? Well, we can start the argument above as before and get
a function u − v on U ∩ V with d(u − v) = 0. Hence u − v is a constant c since U ∩ V
is connected for n > 1, which we can assume since the case n = 1 was the induction
base. Then d(v + c) = α on V as well, and v + c and u agree on U ∩ V , hence patch
together to give a globally defined function f on S n with df = α. Hence
H 1 (S n ) = 0; for n > 1
as well.
For p = n, the argument gives us an (n − 1)-form u − v on U ∩ V defining a class
in
H n−1 (S n−1 × R) ' H n−1 (S n−1 ) = R.
 MA3H5 MANIFOLDS 49

Taking an (n − 1)-form ω on S n−1 with a nontrivial cohomology class (which we

know exists by induction), we pull it back to
S n−1 × (−2, 2) ' U ∩ V
via the projection of the left-hand side on the first factor. Since H n−1 (S n−1 ) = R is
known by induction, we can conclude that H n−1 (S n−1 × (−2, 2)) is then generated
by (the pullback of) [ω] and thus
u − v = λ ω + dw
for some λ ∈ R and a (n − 2)-form w on U ∩ V . If λ = 0, the argument above
gives that the cohomology class of the n-form α is trivial, and we can conclude that
H n (S n ) is at most one-dimensional: indeed, every α determines some λ = λ(α) such
that we have an equation u − v = λ ω + dw on U ∩ V with some (n − 2)-form w,
and (n − 1)-forms u, v on U and V with α = du on U and α = dv on V ; if we
choose different u0 and v 0 with α = du0 and α = dv 0 on U and V , then d(u0 − u) = 0
(and d(v 0 − v) = 0) so we have changed u and v by some closed forms on U and V
which are exact since H n−1 (U ) = 0 and H n−1 (V ) = 0 since they are star-shaped
and n > 1. So then again
u0 − v 0 = λω + dw0
on U ∩ V for some different w0 . Thus λ does not depend on the choice of u and v.
Moreover,
λω + dw = λ0 ω + dw0
can only hold if λ = λ0 since ω gives a nontrivial class on U ∩ V . Also, λ = λ(α) is
obviously R-linear in α. For an exact form α, λ is zero, and if λ is zero, α must be
exact. Thus we get an injection H n (S n ) → R associating to an α the number λ(α),
which shows that H n (S n ) is at most one-dimensional.
Thus it only remains
R to show that H n (S n ) is exactly one-dimensional: this follows
n
because integration over S gives a well-defined linear form (by Stokes’ Theorem)
Z
: H n (S n ) → R
Sn
and we know that, by the definition of the integral, integrating a nowhere vanishing
form giving an orientation, say an everywhere positive such form, gives a nonzero
(positive) result. 
As an anecdote we mention that Leopold Vietoris was an Austrian topologist who
lived from 1891 to 2002 when he died aged 110; he was also an enthusiastic alpinist,
but the story goes that his doctor told him to stop skiing once he got past 100. He
also wrote a mathematical paper at age 103. This may prove that the theory of
manifolds and the subject of topology (or mathematics as a whole) is conducive to
long life!
Here is another very entertaining consequence of the previous calculation.
Theorem 6.6. For p, q ≥ 1, the sphere S p+q is not diffeomorphic to a product
M × N of manifolds M of dimension p and N of dimension q.
Proof. First remark that if X is any non-orientable manifold, a product X ×Y , where
Y is another manifold, can never be orientable: for if it were, X × Y contains an
open submanifold diffeomorphic to X ×Rn , which would also be orientable whence X
itself would be orientable, contradiction. Therefore, in order to prove the Theorem,
it suffices to derive a contradiction from the assumption that S p+q is diffeomorphic
50 CHRISTIAN BÖHNING

to M × N with M and N orientable; we can also assume we have chosen some

orientations for M and N and the product orientation for S p+q .
Since p, q ≥ 1, we know from Theorem 6.5 that H p (S p+q ) = H p (M × N ) = 0.
Let
π: M × N → M
be the projection and ω ∈ A (M ) inducing the orientation of M . Clearly, for fixed
p

n ∈ N the map π : M × {n} → M is an orientation-preserving diffeomorphism, so

Z Z
0< ω= π ∗ ω |M ×{n} .
M M ×{n}

Now since dω = 0 (ω is of top degree dim M = p on M ), we have d(π ∗ ω) = π ∗ (dω) =

0. Therefore, H p (M × N ) = 0 implies the existence of a (p − 1)-form α on M × N
with π ∗ ω = dα. But then
Z Z Z
∗
π ω |M ×{n} = dα |M ×{n} = d(α |M ×{n} ) = 0
M ×{n} M ×{n} M ×{n}
by Theorem 5.8, contradiction. 
Theorem 6.7. Every (smooth) vector field on an even-dimensional sphere S 2m
vanishes at some point of the sphere.
Proof. We argue by contradiction and assume there was such a nowhere vanishing
vector field. As S 2m is a submanifold of R2m+1 we can think of such a vector field
as a smooth map
v : S 2m → R2m+1
with hx, v(x)i = 0 for all x ∈ S n ; let us also normalise such a v so that ||v(x)|| = 1
for all x. Define
ft : S 2m → R2m+1 , ft (x) = (cos t)x + (sin t)v(x).
Since x, v(x) are unit length vectors and everywhere perpendicular, ft (x) is a vector
of unit length everywhere as well. Thus ft maps S 2m to itself. Now
f0 (x) = x, fπ = −x
and using our standard orientation form on S 2m defined for xi 6= 0 by
1
ω = (−1)i dx1 ∧ · · · ∧ dxi−1 ∧ dxi+1 ∧ · · · ∧ dx2m+1
xi
we find
f0∗ ω = ω, fπ∗ ω = −ω.
But by Theorem 6.3, the two maps f0∗ , f1∗ : H 2m (S 2m ) → H 2m (S 2m ) are equal; this
is a contradiction since [ω] gives a nontrivial cohomology class which cannot be equal
to its negative. Hence the vector field v has to have a zero. 
Sometimes this result is called the “Hairy Ball Theorem” since it says that one
cannot comb a hairy ball such that each hair lies flat.
We will now calculate the de Rham cohomology groups of projective spaces and
tori, which will acquaint us with further computational techniques for these groups.
Theorem 6.8. The de Rham cohomology groups of real projective spaces
H r (PnR )
are zero except for
H 0 (PnR ) and H 2m+1 (PR2m+1 )
 MA3H5 MANIFOLDS 51

which are isomorphic to R.

Proof. We use Theorem 6.5 and the description of PnR as the quotient
p : S n → PnR = S n /σ
for the involution σ : S n → S n sending x to −x. First, since PnR is the image of a
connected manifold S n , it is connected and H 0 (PnR ) = R. Now for 0 < r < n we have
H r (S n ) = 0. Take a closed r-form α ∈ F r (PnR ) ⊂ A r (PnR ). Then since d ◦ p∗ = p∗ ◦ d,
the pullback β = p∗ α is closed whence there exists a γ ∈ A r−1 (S n ) with
β = dγ.
Now σ ∗ ◦ d = d ◦ σ ∗ , so d(σ ∗ γ) == σ ∗ (dγ) = σ ∗ β = β whence
γ + σ∗γ
 
β=d .
2
But the form
γ + σ∗γ
δ=
2
is σ-invariant, hence descends to a form δ ∈ A r−1 (PnR ) with p∗ δ = δ and dδ = α by
construction. Thus for 0 < r < n we have H r (PnR ) = 0.
It remains the case r = n. In that case β = p∗ α defines a class in H n (S n ) = R.
Let ω be the standard n-form inducing the orientation of S n that we used before.
Then
β = kω + dγ
for some k ∈ R and γ ∈ A n−1 (S ). We know σ ∗ β = β and σ ∗ ω = (−1)n+1 ω, so
n

β = (−1)n+1 kω + d(σ ∗ γ)
and we can write
1 + (−1)n+1 )
β=k ω + dδ
2
with
γ + σ∗γ
δ= .
2
Then δ is invariant under σ. If n is even, β = dδ, and δ descends to a form
δ ∈ A n−1 (PnR ) with α = dδ by construction. Thus H n (PnR ) = 0 in this case.
If n is odd, β = kω + dδ with σ ∗ δ = δ. But also
σ ∗ ω = ω.
Thus there exist σ̄ and δ̄ on PnR with
α = k ω̄ + dδ̄.
Thus dim H n (PnR )
≤ 1 in this case (every class is a multiple of [ω̄]), but since ω̄ is a
volume form inducing an orientation in this case,
H n (PnR ) = R.

The case of the complex projective spaces PnC is rather different: as a real manifold,
PnC has dimension 2n and it can be shown that H r (PnC ) is equal to R precisely if
0 ≤ r ≤ 2n and r is even, and otherwise these groups are zero. In fact, we mention
here as a cultural remark that this and many other calulations become much simpler
once one recognises that for most “nice” spaces M , such as manifolds, the groups
H r (M ) cannot only be calculated with the aid of differential forms, but via a number
52 CHRISTIAN BÖHNING

of different other methods! There is morally just one “cohomology” of the space M ,
and depending on how one initially defines it (e.g. via differential forms, or so-called
singular co-chains, or...) one speaks of de Rham cohomology, singular cohomology,
Alexander-Spanier cohomology, Čech cohomology, ... and the good news is they
all essentially give the same answer! This can be proven using sheaf theory and
sheaf cohomology. There is one further remark we would like to make since we
will see some “shadow” of this phenomenon in the next Section: the cohomology
H r (M ) actually “comes from” an object over Z (i.e., with an integral structure), in
the sense that one can define natural abelian groups (=Z-modules) H r (M, Z) with
H r (M ) = H r (M, Z) ⊗Z R. One can use other “coefficients” than Z as well, but Z is
in some sense a universal choice.
We finally compute the de Rham cohomology of tori.
Theorem 6.9. Let Tn = Rn /Zn (where (m1 , . . . , mn ) ∈ Zn acts on x = (x1 , . . . , xn ) ∈
Rn by x 7→ (x1 + m1 , . . . , xn + mn )). The translation-invariant 1-forms dx1 , . . . , dxn
on Rn descend to 1-forms on Tn which we denote by the same letters. Then
 
n
dim H r (Tn ) =
r
and H r (Tn ) has a basis consisting of the classes of the r-form
dxi1 ∧ · · · ∧ dxir , 1 ≤ i1 < · · · < ir ≤ n.
Proof. Let α ∈ F r (Tn ) be a closed r-form. We will show the following:
(1) There are n natural commuting one-parameter groups of diffeomorphisms
(i)
Gt , i = 1, . . . , n, acting on Tn = Rn /Zn : indeed,
(i)
Gt : [(x1 , . . . , xn )] 7→ [(x1 , . . . , xi + t, . . . , xn )],
Q (i)
so the action of i Gt is just the action of Tn on itself by translations. We
will define an average ᾱ of α under all of these translations, and show that this
average ᾱ is closed, defines the same cohomology class as α, and is invariant
under all translations.
(2) We show that r-forms invariant under all translations are combinations of the
dxi1 ∧ · · · ∧ dxir with constant coefficients. We denote these invariant forms
by Invr (Tn ). We have Invr (Tn ) ⊂ F r (Tn ).
(3) We know at that point that Invr (Tn ) and B r (Tn ) span F r (Tn ), so we show
Invr (Tn ) ∩ B r (Tn ) = 0
to conclude that Invr (Tn ) maps isomorphically onto H r (Tn ) under the pro-
jection F r (Tn ) → H r (Tn ).
Now let us implement this program in more detail. For (1), let us start looking at
(1)
Gt := Gt . We then define
Z 1
µ1 (α) = G∗θ α dθ.
0
Since G∗t is linear and commutes with the integral sign, we get that µ1 (α) is invariant
under Gt , i.e. G∗t (µ1 (α)) = µ1 (α) for
every t. Also µ1 (α) ∈ F r (Tn ) since
Z 1  Z 1 Z 1
∗ ∗
d(µ1 (α)) = d Gθ α dθ = d(Gθ α) dθ = G∗θ dα dθ = 0.
0 0 0
 MA3H5 MANIFOLDS 53

Moreover, µ1 (α) and α define the same class in H r (Tn ) since, following the method
of proof for Theorem 6.3, we can write
G∗t α − G∗0 α = G∗t α − α = dβt
with β some (r − 1)-form depending smoothly on t. Hence
Z 1 Z 1  Z 1 
∗
µ1 (α) = Gθ α dθ = (α + dβθ ) dθ = α + d βθ dθ
0 0 0
and thus µ1 (α) and α define the same class.
We can then employ these argument inductively to conclude that
ᾱ = µn (µn−1 (. . . (µ2 (µ1 (α))) . . . )),
(i)
where µi denotes the analogous average with respect to Gt , is invariant under all
translations, closed and of the same class as α.
Now for (2), we suppose we are given an r-form γ that is invariant under all
translations. We can write, using multi-index notation again,
X
γ= aI dxI
I
(thus dxI = dxi1 ∧ · · · ∧ dxir ), with the aI functions that, when we view them as
functions on Rn , are periodic for the lattice Zn . But since the dxI are invariant
under all translations, for γ to be invariant under all translations, it must be true
that the aI are invariant under all translations, i.e. they must be constants. In
particular, such a γ is closed.
Now Step (3) can be achieved as follows: assume α ∈ Invr (Tn ) and also α = dβ
for some β, so α ∈ B r (Tn ). We want to show that α = 0. Let (−) be the averaging
operator introduced above. Then
ᾱ = dβ = dβ̄.
But β̄ is invariant under all translations, hence β̄ is closed. Hence dβ̄ = 0, and thus
ᾱ = α = 0. 
The idea used in the proof above can be used to compute the de Rham cohomology
of several other manifolds acted on by some groups that are so-called compact Lie
groups.

7. Degree theory and applications: fundamental theorem of algebra,

linking numbers, indices of singularities of vector fields, the
argument principle
In this Section we will show that the n-th de Rham group of an oriented, compact,
connected n-dimensional manifold is canonically isomorphic to R. This will allow
us to associate to any smooth map f : M → N of oriented, compact, connected
manifolds of the same dimension n a real number, called the degree of f , and
denoted by deg(f ), defined as the number such that
Z Z
∗
f ω = deg(f ) ω
M N
for any n-form ω on N . The fecundity of the concept of degree lies in the surprising
fact that we can arrive at the quantity deg(f ) in an entirely different, geometric way:
indeed, it is equal to the number of preimages, counted with appropriate signs, of a
regular value of f ; more precisely, an a ∈ N is called a regular value for a smooth
54 CHRISTIAN BÖHNING

map f : M → N if the differential (f∗ )p = dfp : Tp (M ) → Ta (N ) is surjective for

each p ∈ M with f (p) = a (if dim M = dim N this means dfp is an isomorphism).
Sard’s Theorem tells us that, in a sense to be made precise, “most” points a ∈ N
are regular values (for example, imagine taking a random point, then this will be a
regular value with probability 1). Then
X
deg(f ) = sgn (det(dfp )) .
p∈f −1 (a)

In particular, this shows that deg(f ) is always an integer. We will give several
applications of this afterwards.
We start with a crucial auxiliary result.
Lemma 7.1. Let Qn = (−1, 1)n be the open unit cube in Rn and let ω ∈ A n (Rn )
with support in Qn and Z
ω = 0.
Qn
Then there exists a β ∈ A n−1 (Rn ) with support in Qn and such that ω = dβ.
Remark that since Qn is star-shaped, Poincaré’s Lemma tells us that there exists
a β with ω = dβ (since dω = 0), but we do not know that such a β will have support
in Qn . In fact, just in case you are wondering why the Lemma is not just Poincaré’s
Lemma “+”, and why we have to R work quite a bit to get it at this point, note that
we do not have supp β ⊂ Qn if Qn ω 6= 0 (which explains the extra assumption):
indeed, by Stokes’ Theorem
Z Z Z Z
β= dβ = ω= ω
∂Qn Qn Qn Qn
(this is actually not quite the generality we proved Stokes’ Theorem in because
∂Qn has “corners”, but one can rather easily generalise it to this case), but of
supp β ⊂ Qn , the first integral would be zero.
Proof. We will prove the statement by induction on the dimension n, but we will
prove a slightly stronger assertion (which often simplifies proofs by induction because
the you can assume more in the inductive step): namely we will prove the statement
with ω and β depending smoothly on some additional parameter λ ∈ Rm , and will
also show that if ω vanishes identically for some value of λ, the same holds for β.
We start with n = 1. We can then write ω = f (x, λ)dx. Setting
Z x
β(x, λ) = f (t, λ)dt
−1
defines a function with dβ = ω; there is a δ > 0 such that f vanishes for x > 1 − δ
and x < −1 + δ since suppω ⊂ (−1, 1) and thus
Z x Z 1
f (t, λ)dt = f (t, λ)dt = 0
−1 −1
R
(by the hypothesis Qn ω = 0) for x > 1 − δ, and similarly for x < −1 + δ. So β
has support in (−1, 1) as well. Also if f (x, λ) = 0 for all x for some λ, then β(x, λ)
vanishes for all x, too, by its definition.
Now for the induction step, let us assume that the stronger statement we want
to prove is true for all dimensions less than n, and we will prove it for n. Write
ω = f (x1 , . . . , xn , λ)dx1 ∧ · · · ∧ dxn .
 MA3H5 MANIFOLDS 55

To make the induction, let us write temporarily xn = t and consider

f (x1 , . . . , xn−1 , t, λ)dx1 ∧ · · · ∧ dxn−1
as a form on Rn−1 depending smoothly on t, λ. If σ is a smooth function with
support in Qn−1 such that
Z
σdx1 ∧ · · · ∧ dxn−1 = 1
Qn−1
then for Z
g(t, λ) := f (x1 , . . . , xn−1 , t, λ)dx1 ∧ · · · ∧ dxn−1
Qn−1
we have that the form
f (x1 , . . . , xn−1 , t, λ)dx1 ∧ · · · ∧ dxn−1 − g(t, λ)σdx1 ∧ · · · ∧ dxn−1
has support in Qn−1 and integral zero over Qn−1 , so by induction there exists a
form γ = γ(t, λ) with support in Qn−1 such that the previous form in the displayed
formula is equal to dγ. Reinserting xn = t we compute
d(γ∧dxn ) = f (x1 , . . . , xn−1 , xn , λ)dx1 ∧· · ·∧dxn−1 ∧dxn −g(xn , λ)σdx1 ∧· · ·∧dxn−1 ∧dxn .
If we define
Z xn 
n−1
ξ(x1 , . . . , xn , λ) = (−1) g(t, λ)dt σdx1 ∧ · · · ∧ dxn−1
−1
we find
dξ = g(xn , λ)σdx1 ∧ · · · ∧ dxn−1 ∧ dxn
whence
f (x1 , . . . , xn−1 , xn , λ)dx1 ∧ · · · ∧ dxn−1 ∧ dxn = d(γ ∧ dxn + ξ) =: dβ.
Clearly
β = γ ∧ dxn + ξ
also has support in Qn−1 × R by induction (and the way it is constructed), so we
only need to check what happens in the xn -direction. Now we know that
f (x1 , . . . , xn−1 , t, λ)
vanishes for t < 1−δ, t > −1+δ some small δ (since ω was assumed to have support
in Qn ). Therefore, by our stronger inductive assumption (including the dependency
on parameters), we get that the same is true for γ. Therefore all we have to check
is what happens with ξ. We consider the case xn = t > 1 − δ, the case xn < −1 + δ
being similar. Then
Z t Z t Z 
g(s, λ)ds = f (x1 , . . . , xn−1 , s, λ)dx1 ∧ · · · ∧ dxn−1 ds
−1 −1 Qn−1
Z 1 Z 
= f (x1 , . . . , xn−1 , s, λ)dx1 ∧ · · · ∧ dxn−1 ds
−1 Qn−1
Z
= f (x1 , . . . , xn−1 , xn , λ)dx1 ∧ · · · ∧ dxn−1 ∧ dxn
Qn
=0
by assumption. Thus also ξ has support in Qn . To finish the inductive proof we only
need to observe that if f (x, λ) is identically zero for some λ, the same holds for β.
This holds by construction if you run through the definitions above once more. 
We are now in a position to show
56 CHRISTIAN BÖHNING

Theorem 7.2. If M is a compact, orientable, connected n-dimensional manifold,

and also suppose we have chosen a fixed orientation; then integration
Z
: H n (M ) → R
M
gives a canonical isomorphism between H n (M ) and R.
Proof. We can cover M by finitely many charts U1 , . . . , UN such that each Uj is
diffeomorphic to Qn (here we use the compactness
R of M ). Choose an n-form α0 with
support contained in U1 and such that M α0 = 1. Clearly, by Stokes’ Theorem, the
cohomology class [α0 ] is nontrivial, and we want to show that every other class in
H n (M ) is a multiple of this one.
Now take an arbitrary n-form α on M . Take a partition of unity (θi ) subordinate
to the given cover; we can modify the (θi ) given to us by Theorem 3.11 so that they
are indexed by the same finite index set J as the cover {Uj }: indeed, put ηj equal
to the sum of all θi with j(i) = j and zero if no θi has j(i) = j. Then
X
α= ηj α
j

and this is a finite sum. It suffices to show that each ηj α defines a cohomology class
proportional to [α0 ], so without loss of generality we can suppose that α has support
in one of the coordinate neighbourhoods Um , say. Since M is connected, we can
connect any p ∈ U1 and q ∈ Um by a path, and possibly by reordering the Ui and
allowing some duplications, we can assume that the path is covered by U1 , . . . , Um
such that Ui ∩ Ui+1 6= ∅. Now choose an n-form αi with support in Ui ∩ Ui+1 ,
1 ≤ i ≤ m − 1, and integral 1. Then on U1
Z
(α0 − α1 ) = 0

so by Lemma 7.1, there is a form β1 with support in U1 with α0 − α1 = dβ1 . Notice

that we precisely need the assertion in Lemma 7.1 that β1 has support in Qn to be
able to interpret β1 as a form on all of M .
Similarly, we find forms β2 , . . . , βm−1 satisfying the equations
α0 − α1 = dβ1 ,
α1 − α2 = dβ2 ,
...
αm−2 − αm−1 = dβm−1 .
Pm−1 R R
Therefore α0 − αm−1 = d( i=1 βi ). If α = c ∈ R, then (α − cαm−1 ) = 0 and we
can apply Lemma 7.1 to write
α − cαm−1 = dβ
and thus
m−1
X
α = cαm−1 + dβ = cα0 + d(β − c βi ).
i=1

Definition 7.3. Let f : M → N be a smooth map of manifolds. A point p ∈ M is
called a regular point for f if the differential
(f∗ )p : Tp (M ) → Tf (p) (N )
 MA3H5 MANIFOLDS 57

is surjective. If the differential fails to be surjective, p is called a critical point for

f . A point q ∈ N is called a critical value for f if it is the image under f of some
critical point p ∈ M for f . If q ∈ N is not a critical value, it is called a regular value
for f .
Notice some potential subtleties here: we do not define a regular value as the
image under f of a regular point; for example, if a point q is not in the image of
f at all, it is a regular value by definition. A point q ∈ N is a regular value if all
points in f −1 (q) are regular points (and this preimage may be empty whence the
requirement is vacuously satisfied), and a point q ∈ N is a critical value if some
point in f −1 (q) is a critical point.
Theorem 7.4. Let f : M → N be a smooth map between compact, oriented, con-
nected manifolds of the same dimension n. Then there exists an integer deg(f )
called the degree of f such that if ω ∈ A n (N ), then
Z Z
f ∗ ω = deg(f ) ω
M N
and if q ∈ N is a regular value for f , then
X
deg(f ) = sgn(det(f∗ )p ),
p∈f −1 (q)

where sgn(det(f∗ )p is ±1 and defined as follows: the fact that M and N are assumed
to be oriented means that we have chosen orientations for M and N , i.e. equivalence
classes of nowhere vanishing
R globalRn-forms ωM and ωN , which we can also assume
to be normalised so that M ωM = N ωN = 1; the differential
(f∗ )p : Tp (M ) → Tf (p) (N )
induces a map
Λn ((f∗ )p )∗ : Λn Tf (p) (N )∗ → Λn Tp (M )∗
whence we have (Λn ((f∗ )p )∗ )(ωN,f (p) ) = λωM,p for some λ ∈ R. Then sgn(det(f∗ )p )
is the sign of λ. More concretely, if we choose local coordinates (U, x1 , . . . , xn )
around p and (V, y1 , . . . , yn ) around q = f (p) such that (ωM ) |U is equivalent to
dx1 ∧ · · · ∧ dxn and (ωN ) |V is equivalent to dy1 ∧ · · · ∧ dyn , and we write f =
(f1 (x1 , . . . , xn ), . . . , fn (x1 , . . . , xn )) in these local coordinates, then sgn(det(f∗ )p ) is
nothing but the sign of the determinant of the Jacobian matrix:
 
∂f
det ∂xji .

Note that if f fails to be surjective, the theorem says that deg(f ) = 0.

Proof. If ω is any n-form on N , its cohomology class is a multiple of the cohomology
class of ωN , [ω] = c[ωN ], c ∈ R, by Theorem 7.2. Therefore (using Stokes’ Theorem)
Z
ω = c.
N
On the other hand f ∗ [ω N] = k[ωM ] for some number k ∈ R. So
Z Z
∗
f ω = ck = k ω.
M N
We want to show that the number k = deg(f ) is an integer and can be described
geometrically as explained in the statement of the Theorem. If q ∈ N is a regular
value, then (f∗ )p is an isomorphism at any point p ∈ f −1 (q), hence by Proposition
58 CHRISTIAN BÖHNING

2.4, f −1 (q), if nonempty, is a submanifold of codimension n, thus dimension 0 of M .

Since M is compact, f −1 (q) is also compact (since it is closed in a compact space),
hence a finite set of points. Proposition 3.5, c) implies that for a sufficiently small
open neighbourhood U of q, f −1 (U ) decomposes as a disjoint union of finitely many
opens U1 , . . . , Um , each of which maps diffeomorphically onto
R U under f .
If σ is an n-form with support in U and N σ = 1, then M f ∗ σ = k. But we can
R
compute this number in a different way now: we just have to sum the
Z
(f |Ui )∗ σ, i = 1, . . . , m.
Ui

But the change of variables formula for multiple integrals/coordinate invariance of

integrals of differential forms then implies
Z Z
(f |Ui )∗ σ = sgn(det(f∗ )pi ) σ = sgn(det(f∗ )pi )
Ui U

if pi ∈ Ui is such that f (pi ) = q. 

Before giving applications of Theorem 7.4, we feel it is our moral duty to show
you that there is always an ample supply of regular values. This is the following
(special case of) Sard’s Theorem (which we will never use in the sequel, however).
Theorem 7.5. Let M and N be manifolds of the same dimension n. Let f : M → N
be a smooth map. Then the set of critical values for f has measure zero in N .
We will omit the proof because the techniques are of a different flavour than the
ones we have used so far and we feel that it would take us too far afield here. If
you are interested, proofs can be found in lots of texts on manifolds, for example
in M. Berger, B. Gostiaux, Differential Geometry: Manifolds, Curves and Surfaces,
Springer (1988), section 4.3.
We add a few comments, however, to explain Sard’s result a little better. If M
and N have different dimensions, dim M = m, dim N = n say, it can still be proven
that the critical values of f have measure zero in N , but the proof for m > n is
more difficult.
Secondly, we have to say a few words about what we mean when we say that a
subset S ⊂ N of a manifold N has “measure zero”. We want to give a pedestrian
definition of this concept without delving too deeply into measure theory. First, we
agree that an n-cube C ⊂ Rn of egde length λ, by which we mean a product
C = I1 × · · · × In
of closed intervals Ij = [aj , aj + λ] of length λ, has measure µ(C) = µn (C) = λn .
We say that a subset X ⊂ Rn has measure zero if for every  > 0 it can be covered
by a countable family of n-cubes the sum of whose measures is less than . Thus
a countable union of sets of measure zero in Rn has measure zero. If X ⊂ N is a
subset of some manifold N , we say that X has measure zero if for every chart (U, ϕ)
the set ϕ(X ∩ U ) ⊂ Rn has measure zero. In fact, it turns out that it suffices to
check this for the charts of some atlas.
We turn to some applications of degree theory. First, consider the sphere S 2 .
We may view this as P1C = C ∪ {∞}, with coordinate z on C and homogeneous
coordinates (Z0 : Z1 ) on P1C so that z = Z1 /Z0 on the open subset of P1C where
Z0 6= 0. Let
f (z) = z k + a1 z k−1 + · · · + ak
 MA3H5 MANIFOLDS 59

be a monic polynomial with complex coefficients ai . Then

F (Z0 : Z1 ) = (Z0k : Z1k + a1 Z1k−1 Z0 + · · · + ak Z0k )
defines a smooth map from P1C to itself. Now consider the smooth family of maps
Ft : P1C → P1C
with  
Ft (Z0 : Z1 ) = Z0k : Z1k + t(a1 Z1k−1 Z0 + · · · + ak Z0k ) .
Then by Theorem 6.3, the action of Ft∗ on cohomology is independent of t, so
deg(F ) = deg(F0 ).
Here in terms of the coordinate z, F0 (z) = z k . We calculate the degree by choosing
a 2-form on C = {Z0 6= 0} ⊂ P1C with compact support in C: indeed, in terms of
z = x + iy = reiθ = r cos θ + ri sin θ we will choose
ω = f (r)dx ∧ dy = f (r)rdr ∧ dθ
with f of compact support normalised such that
Z
ω = 1.
R2
Thus
Z Z 2π Z ∞ Z
∗ k k k
deg(F0 ) = (F0 ) (ω) = f (r )r d(r ) ∧ d(kθ) = k ω = k.
R2 θ=0 r=0 R2
In particular, if k > 0, each Ft is surjective. Therefore
f (z) = z k + a1 z k−1 + · · · + ak = 0
has a solution, which is the Fundamental Theorem of Algebra.
As a second application we consider linking numbers of curves in R3 . Here we
will understand by a curve in R3 a smooth embedding
f : S 1 → R3 .
Compare Definition 3.6. A pair of curves {f, g} will consist of two curves f, g such
that f (S 1 ) ∩ g(S 1 ) = ∅.
Definition 7.6. We call pairs of curves {f, g} and {f 0 , g 0 } isotopic if there are
smooth maps
F : [0, 1] × S 1 → R3 , G : [0, 1] × S 1 → R3
such that Ft (x) := F (t, x) and Gt (x) := G(t, x) are such that
{Ft , Gt }
is a pair of curves for every t ∈ [0, 1] and F0 = f, F1 = f 0 , G0 = g, G1 = g 0 . We will
call {F, G} an isotopy between {f, g} and {f 0 , g 0 }.
The reader should at this point sketch some examples of pairs of curves: they can
be very complicated “linked” copies of S 1 ! Moreover, it would be a good idea to
think about what the notion of isotopy means from a visual informal point of view.
We now wish to define an invariant under isotopy of a pair of curves {f, g}.
60 CHRISTIAN BÖHNING

Definition 7.7. Let {f, g} be a pair of curves. The linking number of {f, g},
denoted by link(f, g), is the degree of the smooth map µ : S 1 × S 1 → S 2 given by
f (t) − g(s)
µ(t, s) =
||f (t) − g(s)||
where || · || is the Euclidean norm.
We now wish to relate this number link(f, g) to the geometric idea that the curves
can be “linked” or “unlinked”.
We will denote by {f0 , g0 } the pair of curves in R3 consisting of the standard
parametrisations of the circles
f0 : S 1 → (x, y, z) ∈ R3 : z = 0 and (x − 2)2 + y 2 = 1 ,


g0 : S 1 → (x, y, z) ∈ R3 : z = 0 and (x + 2)2 + y 2 = 1 .



Note that the images of f0 and g0 are just translations of the standard copy of S 1
of radius 1 around the origin in the (x, y)-plane, and these translations give what
we call “standard parametrisations”.
Definition 7.8. We call a pair of curves {f, g} unlinked if it is isotopic to the pair
{f0 , g0 }. Otherwise we call it linked.
We then have the following criterion.
Theorem 7.9. If {f, g} and {f 0 , g 0 } are isotopic pairs of curves, we have
link(f, g) = link(f 0 , g 0 ).
Moreover, if link(f, g) 6= 0, then the pair {f, g} is linked.
Proof. Let {F, G} be an isotopy between the isotopic pairs of curves {f, g} and
{f 0 , g 0 }. For a fixed t, we associate to the pair {Ft , Gt } the map µt : S 1 × S 1 → S 2
given by
Ft (x) − Gt (y)
µt (x, y) = .
||Ft (x) − Gt (y)||
We can then define a smooth map
µ : [0, 1] × S 1 × S 1 → S 2
by µ(t, x, y) = µt (x, y). We can apply Theorem 6.3 to conclude that deg(µt ) is
constant in the family. Hence
link(f, g) = deg(µ0 ) = deg(µ1 ) = link(f 0 , g 0 ).

We now show that if f and g are unlinked, then their linking number is zero. By
definition, unlinkedness means that {f, g} is isotopic to {f0 , g0 }, so by the first part
we only have to show that the linking number of f0 and g0 is zero. But the images
of f0 and g0 are contained in the (x, y)-plane, so the image of µ is also contained
in the (x, y)-plane and cannot be all of S 2 . Thus µ is not surjective, which means
deg(µ) = link(f0 , g0 ) = 0 by Theorem 7.4. 
Example 7.10. We should show that link(f, g) is not by chance always identically
zero. Consider the circles in Figure 1, top. The images of f and g lie in mutually
perpendicular planes, the (x, y)- and (x, z)-planes, and their centres lie one the x-
axis, the line of intersection of these two planes. We claim that link(f, g) = ±1 (the
sign depending on the chosen orientations). Consider the map
µ : S1 × S1 → S2
 MA3H5 MANIFOLDS 61

from Definition 7.7. Let e1 = (1, 0, 0) ∈ S 2 . We want to know what µ−1 (e1 ) is. Now
if (a, b) ∈ µ−1 (e1 ), then f (a) − g(b) must be parallel to the x-axis and point in the
positive direction towards +∞ on that axis: hence f (a) and g(b) must be the points
shown in the figure. Therefore, µ−1 (e1 ) is a single point.

Figure 1. Linked curves with linking number ±1 (top); the bottom

shows curves with linking number zero, but the pair on the right
hand side is linked.

We claim that e1 is indeed a regular value for µ: indeed, T(a,b) (S 1 × S 1 ) '

Ta (S 1 ) × Tb (S 1 ) maps surjectively onto Tµ(a,b) (S 2 ) because the vectors f 0 (a) and
g 0 (b) in Figure 1 are perpendicular. This implies that the two curves
µa = µ |{a}×S 1 : {a} × S 1 → S 2 , µb = µ |S 1 ×{b} : S 1 × {b} → S 2
passing through µ(a, b) have independent tangent vectors at that point. Thus by
Theorem 7.4, deg(µ) = link(f, g) = ±1.
Example 7.11. It should be noted that if link(f, g) = 0, it is not necessarily true
that the pair of curves {f, g} is unlinked. Thus the converse to the second assertion
in Theorem 7.9 is not true. Indeed, consider Figure 1, bottom. The pair of curves
pictured on the left hand side is visibly unlinked, hence has linking number zero. We
62 CHRISTIAN BÖHNING

claim that the pair of curves on the right then also has to have linking number zero
(though it is intuitively clear that this pair of curves is linked). Indeed, the proof
of the first part of Theorem 7.9 shows a little more, namely that link(f, g) is not
only invariant under isotopy, bot under what we may call (smooth) homotopy: here,
in slight modification of Definition 7.6, we call pairs of curves {f, g} and {f 0 , g 0 }
(smoothly) homotopic if there are smooth maps
F : [0, 1] × S 1 → R3 , G : [0, 1] × S 1 → R3
such that Ft (x) := F (t, x) and Gt (x) := G(t, x) satisfy F0 = f, F1 = f 0 , G0 =
g, G1 = g 0 and
Ft (S 1 ) ∩ Gt (S 1 ) = ∅
for every t ∈ [0, 1]. In other words, we relax the assumption that
{Ft , Gt }
be a pair of curves for every t ∈ [0, 1], to allow more general maps Ft , Gt , t ∈ (0, 1),
that need not be smooth embeddings anymore: for example, the image of Ft (or Gt )
could be a curve with self-intersections (so that Ft is no longer injective), or even
wilder.
Then it is clear intuitively that the pairs of curves in Figure 1, bottom are
smoothly homotopic.
Example 7.12. As an example, that the linking number link(f, g) in some sense
really measures the number of times the curve f goes around the curve g, we en-

Figure 2. This is a geometric argument why the red coil winding

around the green circle n-times gives a pair with linking number ±n
(sign depending on orientations)

courage the reader to contemplate the homotopy depicted in Figure 2, where the red
“coil” winds around the green curve n times (n = 3 is shown). Then that homotopy
 MA3H5 MANIFOLDS 63

shows that this situation is homotopic to one with a red circle being linked to the
green curve as shown, with the red circle being traversed n times (corresponding to
the map S 1 → S 1 , z 7→ z n , z ∈ C, |z| = 1), so link(f, g) = ±n.
We now leave linking numbers of curves and come to a final and very important
concept that can be accessed using degree theory: the index of an isolated singularity
of a vector field. As we will explain this has many wonderful applications and is
also intimately related to the global topological properties of a manifold.
We start with the situation in Rn , so let U ⊂ Rn be an open subset and let X
be a vector field on U . We may view X simply as a smooth map X : U → Rn . We
call a point p ∈ U a singularity of X if X(p) = 0. In addition, we call it an isolated
singularity if X(q) 6= 0 for every q 6= p in a neighbourhood V of p in U .
Proposition 7.13. Let p be an isolated singularity of the vector field X on the open
subset U in Rn , and denote by S n−1 (p, ) the sphere of radius  > 0 and centre p in
Rn . Let S n−1 = S n−1 (0, 1). If V is a neighbourhood of p that contains no further
singularities of X than p and if  is so small that S n−1 (p, ) ⊂ V , then the degree of
the map
fX : S n−1 → S n−1 ,
X(p + x)
x 7→
||X(p + x)||
does not depend on .
We call this degree the index of X at p and denote it by indp X.
Proof. Let  and δ be small real numbers such that S(p, ) and S(p, δ) are both
contained in V . Define
F : [0, 1] × S n−1 → S n−1
by
X
F (t, x) = f(1−t)+tδ (x).
Since X does not vanish on V except in p, there exist an α > 0 such that for every
t ∈ (−α, 1 + α) the denominator of ftX does not vanish for any x. Thus, since
F0 = fX and F1 = fδX , we get the result from Theorem 6.3. 
Example 7.14. You can easily calculate the following:
(1) If we put X = idRn , then the origin is an isolated singularity of index 1.
(2) For X = −idRn , the origin is an isolated singularity of index (−1)n .
(3) If X on R2 is given by X(x, y) = (−y, x), then the origin is an isolated
singularity of index one.
(4) If X(x, y) = (x2 − y 2 , 2xy) in R2 , then the origin is an isolated singularity of
index 2.
We just calculate (4) for fun (or punishment, whichever your view) and the rest is
seen similarly, but easier. We recognise (4) as the map z 7→ z 2 for z = x + iy, and
thus the preimage of any point p ∈ S 1 consists of two points ±p. Since the map
preserves orientation, ind0 X = 2.
You should look at the primitive sketches of these vector field that I made in
Figure 3, and use these to confirm that the index has the following geometric inter-
pretation: traverse a little circle around the origin in counterclockwise direction; at
each point of your way, the vector field furnishes you with an arrow that you can
view as the position of the hand of a clock. Then the index is the number of times
this hand turns on the clock as you traverse the circle and return to your starting
64 CHRISTIAN BÖHNING

Figure 3. Sketches of the vector fields (1)-(4) in R2

point (where you count each full turn in counterclockwise direction as +1, and turns
in clockwise direction as −1).
We now give a fun application of the concept of the index of a vector field, called
the argument principle in complex analysis. Suppose we want to count the number
of zeros of a complex polynomial
p(z) = z m + a1 zm−1 + · · · + am
that lie in a smooth compact region Ω of the complex plane whose boundary we
assume contains no zero of p; we will assume for simplicity and concreteness that
Ω is just some some closed ball, i.e. a disc, in the plane, but other cases would be
possible, too. We can view this as a manifold with boundary, oriented in a natural
way using the standard orientation in the ambient R2 = C.
Proposition 7.15. The number of zeros of p(z) in Ω, counted with multiplicity,
equals the degree of the map
p
: ∂Ω → S 1 .
|p|
This number is equal to the sum of the indices of singularities of the vector field
X(z) = p(z) contained in Ω.
Here the expression “counted with multiplicity” means the following: if
N
Y
p(z) = (z − αi )mi
i=1
and α1 , . . . , αt are the zeros of p(z) lying inside Ω, the number in Proposition 7.15
is by definition
m1 + · · · + mt .
 MA3H5 MANIFOLDS 65

Proof. We can find pairwise disjoint classed balls Di = B(αi , i ), i > 0 small real
numbers, i = 1, . . . , t, contained in the interior of Ω. We will now apply Stokes’
Theorem to the oriented manifold with boundary
t
[
D = Ω\ B(αi , i )
i=1
(here B(αi , i ) denotes the open disk). Note that the map
p(z)
Fp (z) :=
||p(z)||
is defined and smooth in a neighbourhood of D. It takes values in S 1 ⊂ C. Denoting
by ω the canonical volume form giving the orientation on S 1 (normalised so that its
integral over S 1 is 1), we thus have by Stokes’ Theorem
Z Z Xt Z
∗ ∗
0= d(Fp ω) = Fp ω − Fp∗ ω.
D ∂Ω i=1 S 1 (αi ,i )

Thus we have that the degree of the map

p
: ∂Ω → S 1
|p|
equals
X t
indαi X.
i=1
Let us now check that
indαi X = mi .
Write
p(z) = (z − αi )mi q(z)
with q(αi ) 6= 0. Then q is never zero in Di . Recall that i is the radius of Di . Then
g : S 1 → ∂Di , g(z) = αi + i z
is an orientation-preserving diffeomorphism of S 1 onto ∂Di . Thus the degree of
p
: ∂Di → S 1
|p|
is the same as the degree of
p◦g
: S1 → S1.
|p ◦ g|
Now define
z mi q(αi + ti z)
ht : S 1 → S 1 , ht (z) =
|q(αi + ti z)|
for t ∈ [0, 1]. Note that the division is permissible since q is never zero in Di . Then
p◦g
h1 =
|p ◦ g|
and
h0 (z) = cz mi
where c is the nonzero constant
q(αi )
.
|q(αi )|
Now by Theorem 6.3, h1 and h0 have the same degree, but the degree of h0 is clearly
equal to mi . This concludes the proof. 
66 CHRISTIAN BÖHNING

Indeed, we can do somewhat better. The following result sometimes goes under
the name argument principle.
Theorem 7.16. The number of zeros (counted with multiplicity) of the complex
polynomial p(z) in the disk Ω (with zero-free boundary) is equal to
Z
1
d(arg p(z)).
2π ∂Ω
We first make some comments so that you can parse the formula in the statement
of the Theorem. Each nonzero complex number w can be written uniquely as
w = reiθ
where r is the norm of w, and θ is, by definition, the argument. However, note
that θ = arg (w) is not really a well-defined function: all the values θ + 2πn, n any
integer, qualify equally to be the argument of w. Luckily, if we take the exteriro
derivative d, this ambiguity disappears since any two choices of argument differ by
addition of a constant. Indeed, in a suitable neighbourhood of any point in C − {0},
we can always choose values of arg (w) so that this becomes a smooth function near
the point; and any other local such choice will differ from the preceding one by some
constant 2πn. Thus we get a well-defined 1-form
d arg
on all of C − {0}, but this is not the differential of a well-defined function as the
notation may erroneously suggest.
The integrand in Theorem 7.16 is the 1-form
z 7→ d arg p(z) = p∗ (d arg)
which is well-defined and smooth on the complex plane minus the zeros of p.
(Proof of Theorem 7.16). We already know by Proposition 7.15 that the number of
zeros counted with multiplicity inside Ω is equal to the degree of the map
p
f := : ∂Ω → S 1 .
|p|
Now
p(z)
= eiarg p(z) ,
|p(z)|
and all that remains to do is to identify the integral of d(arg p(z)) over ∂Ω with
2π deg(f ).
We apply Theorem 7.4 to the restriction of the smooth 1-form d arg to S 1 . If
w = f (z) = eiarg p(z)
then arg w = arg p(z) and since arg p(z) is locally a well-defined smooth function,
we get
d arg p(z) = d(f ∗ arg w) = f ∗ (d arg w).
Thus Z Z
d(arg p(z)) = deg(f ) d arg w.
∂Ω S1
But calculating Z
d arg w
S1
 MA3H5 MANIFOLDS 67

is easy: indeed, parametrising S 1 by θ 7→ eiθ gives

Z Z 2π
d arg w = dθ = 2π.
S1 0

Suppose now that X is a vector field on a manifold M , having an isolated zero
at p ∈ M . We would like to define the index of X at p, indp X, to be the integer
indϕ(p) (ϕ∗ X)
where (U, ϕ) is any chart around p. However, for this definition to be well-posed,
we need the invariance of the index under diffeomorphism, namely:
Lemma 7.17. Let U and U 0 be open subsets of Rn and let f : U → U 0 a diffeomor-
phism. If X is a vector field on U having an isolated zero at x0 ∈ U , then f (x0 ) is
an isolated zero of f∗ X and
indx0 X = indf (x0 ) f∗ X.
Before we can prove Lemma 7.17, we need one more auxiliary result.
Lemma 7.18. Let U ⊂ Rn be open and star-shaped at the origin (meaning the line
segment connecting the origin to any point in U is entirely contained in U ). Let
f : U → f (U )
be a diffeomorphism that takes the origin 0 to itself and with det(df0 ) > 0. Then f
is isotopic to the identity, by which we mean there is a family
F : [0, 1] × U → Rn
of diffeomorphisms Ft such that F0 = f and F1 = id.
(Proof of Lemma 7.18). We define G : [0, 1] × U → Rn by
(
f (tx)
t for t ∈ (0, 1],
G(t, x) =
(df0 )(x) for t = 0.
This shows that f is isotopic to the linear map df0 . But the linear maps with positive
2
determinant form a connected open subset of Rn , hence we can connect df0 to the
identity by a path. 
We now turn to
(Proof of Lemma 7.17). We first consider the case when det(df0 ) > 0. We can then
assume that x0 = 0 and that f (0) = 0, and that U is star-shaped (by possibly
replacing U by some smaller open set). Then there is an isotopy F between f and
idU by Lemma 7.18. There is a small positive number  > 0 such that Yt = (Ft )∗ X
does not vanish on B(0, )\{0} for all t ∈ [0, 1] (use the compactness of [0, 1]). Using
Theorem 6.3 and the definition of index (and degree), we get that ind0 Yt is constant,
and in particular,
ind0 f∗ X = ind0 Y0 = ind0 Y1 = ind0 X.
If det(df0 ) < 0, we reduce to the previous case by replacing X by %∗ X and f by
f ◦ %−1 where % is the reflection in some hyperplane in Rn . Then
f∗ X = (f ◦ %−1 )∗ (%∗ X)
and ind0 (%∗ X) = ind0 X, which concludes the proof. 
68 CHRISTIAN BÖHNING

We can use the argument in Lemma 7.18 to calculate the index of a vector field
in a much simpler way in certain special cases:
Proposition 7.19. Suppose X : U → Rn is a vector field on some open subset U of
Rn , and let p ∈ U be an isolated singularity of X. If det(dXp ) 6= 0, then we have
(
1 if det(dXp ) > 0,
indp X =
−1 if det(dXp ) < 0.
Proof. Indeed, we can assume p = 0, and then X is a diffeomorphism on a neigh-
bourhood of 0, which we can assume to be star-shaped. If det(dXp ) > 0, then X is
isotopic to the identity by Lemma 7.18, and each Ft in this isotopy can be consid-
ered as a vector field having an isolated singularity at p; by the invariance of degree
(which again follows from Theorem 6.3), we get
indp X = 1.
If X has det(dXp ) < 0, we reduce to the previous case by considering X ◦ σ where
σ is the reflection in a hyperplane. 
We conclude by mentioning a celebrated result, the Poincaré-Hopf Index Theo-
rem, that relates the concept of indices of isolated singularities of vector fields on
compact manifolds with a basic topological invariant of the manifold, the Euler
characteristic (sometimes also called the Euler number ).
Theorem 7.20. If M is a compact manifold and X a vector field on M having only
isolated singularities p1 , . . . , pn , then
X n
indpi X = χ(M )
i=1
is the so-called Euler characteristic of M , which can be computed in the following
alternative way: all de Rham cohomology groups H i (M ) are finite-dimensional R-
vector spaces if M is compact, and
dim
XM
χ(M ) = (−1)i dimR H i (M ).
i=0

We do not have time in the course of this lecture to prove this result, but did
not want to pass it by in silence because it is so beautiful and important; a proof
can be found, for example, in the wonderful book by Raoul Bott and Loring W. Tu,
Differential Forms in Algebraic Topology, Springer (1982), p. 126 ff.
We just note that from our earlier computation of the de Rham cohomology of
spheres and tori, it follows that for M = S n , we get
χ(S n ) = 1 + (−1)n
and for M = Tn ,
χ(Tn ) = 0.
Moreover, for M = PnR we get
(
1 if n is even,
χ(PnR ) =
0 if n is odd.
It is easy in low dimensions to picture vector fields with isolated singularities on
these manifolds that allow us to confirm the equality stated in the Poincaré-Hopf
 MA3H5 MANIFOLDS 69

Index Theorem explicitly for these examples. The reader should compare Figure 4
and check that in each case the sum of the indices of the vector fields depicted is
equal to the Euler characteristic.

Figure 4. Vector fields with isolated singularities on some compact manifolds.

References
[BeGo88] Berger, M., Gostiaux, B., Differential Geometry: Manifolds, Curves and Surfaces, GTM
115, Springer (1988)
[BoTu82] Bott, R., Tu, L.W., Differential Forms in Algebraic Topology, GTM 82, Springer (1982)
[GuPo74] Guillemin, V., Pollack, A., Differential Topology, reprint of the 1974 edition, AMS
Chelsea Publishing, American Mathematical Society, Providence, Rhode Island (2014)
[Hitchin] Hitchin, N., Differentiable Manifolds, online notes available at
http://people.maths.ox.ac.uk/ joyce/Nairobi2019/Hitchin-DifferentiableManifolds.pdf
[KoNo63] Kobayashi, S., Nomizu, K., Foundations of Differential Geometry, Volume I, John Wiley
& Sons Inc. (1963)
[MoKo71] Morrow, J., Kodaira, K., Complex Manifolds, reprint of the 1971 edition, AMS Chelsea
Publishing, American Mathematical Society, Providence, Rhode Island (2006)
[Tu11] Tu, L.W., An Introduction to Manifolds, 2nd Edition, Springer Universitext, Springer
(2011)
[Warner83] Warner, F.W., Foundations of Differentiable Manifolds and Lie Groups, GTM 94,
Springer (1983)