Differentiable Manifolds

Eckhard Meinrenken

Lecture Notes, University of Toronto, Fall 2001

The main references for these lecture notes are the first volume of Greub-HalperinVanstone, and the book by Bott-Tu (second edition!).

Contents
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.

Manifolds
Partitions of unity
Vector fields
Differential forms
De Rham cohomology
Mayer-Vietoris
Compactly supported cohomology
Finite-dimensionality of de Rham cohomology
Poincare duality
Mapping degree
Kuenneth formula
De Rham theorem
Fiber bundles
The Thom class
Intersection numbers

4
14
17
23
38
40
44
46
47
53
54
57
62
66
71

CONTENTS

1. Manifolds
1.1. Definition of manifolds. A n-dimensional manifold is a space that locally
looks like Rn . To give a precise meaning to this idea, our space first of all has to come
equipped with some topology (so that the word local makes sense). Recall that a
topological space is a set M , together with a collection of subsets of M , called open
subsets, satisfying the following three axioms: (i) the empty set and the space M itself
are both open, (ii) the intersection of any finite collection of open subsets is open, (iii)
the union of any collection of open subsets is open. The collection of open subsets of M
is also called the topology of M . A map f : M1 M2 between topological spaces is
called continuous if the pre-image of any open subset in M2 is open in M1 . A continuous
map with a continuous inverse is called a homeomorphism.
One basic ingredient in the definition of a manifold is that our topological space
comes equipped with a covering by open sets which are homeomorphic to open subsets
of Rn .
Definition 1.1. Let M be a topological space. An n-dimensional chart for M is a
pair (U, ) consisting of an open subset U Rn and a continuous map : U Rn such
that is a homeomorphism onto its image (U ). Two such charts (U , ), (U , ) are
C -compatible if the transition map
1
: (U U ) (U U )
is a diffeomorphism (a smooth map with smooth inverse). A covering A = (U )A of
M by pairwise C -compatible charts is called a C -atlas.
Some people define a C -manifold to be a topological space with a C atlas. It is
more common, however, to restrict the class of topological spaces.
Definition 1.2 (Manifolds). A C -manifold is a Hausdorff topological space M ,
with countable basis, together with a C -atlas.
Remarks 1.3.
(a) We recall that a topological space is called Hausdorff if any
two points have disjoint open neighborhoods. A basis for a topological space M
is a collection B of open subsets of M such that every open subset of M is a
union of open subsets in the collection B. For example, the collection of open
balls B (x) in Rn define a basis. But one already has a basis if one takes only
all balls B (x) with x Qn and Q>0 ; this then defines a countable basis. A
topological space with countable basis is also called second countable.
(b) The Hausdorff axiom excludes somewhat pathological examples, such as following: Let M = R {p}, where p is a point, with the topology given by open
sets in R, together with sets of the form (U \{0}) {p}, for open sets U R
containing 0. An open covering of M is given by the two sets U+ = R and
U = R\{0} {p}. The natural projection from M to R, taking p to 0, descends to smooth maps + : U+ R and : U R. Then M with atlas

1. MANIFOLDS

(U , ) satisfies all the axioms of a 1-dimensional manifold except that it is not

Hausdorff: Every neighborhood of 0 intersects every neighborhood of p.
(c) It is immediate from the definitions that any open subset of a manifold is a
manifold and that the direct products of two manifolds is again a manifold.
The definition of a manifold can be generalized in many ways. For instance, a manifold with boundary is defined in exactly the same way as a manifold, except that the
charts take values in a half space {x Rn | x1 0}. For this to make sense, one needs
to define a notion of smooth maps between open subsets of half-spaces of Rn , Rm : Such
a map is called smooth if it extends to a smooth map of open subsets of Rn , Rm . Even
more generally, one defines manifolds with corners modeled on open subset of the positive
orthant {x Rn | xj 0} in Rn .
A complex manifold is a manifold where all charts take values in Cn
= R2n , and all
transition maps 1
are holomorphic.
1.2. Examples of manifolds.
(a) Spheres. Let S n Rn+1 be the unit sphere of radius 1. Let N = (1, 0, . . . , 0)
be the north pole and S = (1, 0, . . . , 0) the south pole. Let U1 = S n \{S} and
U2 = S n \{N }. Define maps j : Uj Rn by
x (x S)S
x (x N )N
x (x N )N
, 2 (x) =
=
.
1xN
1xS
1+xN
Then j : Uj Rn define the structure of an oriented manifold on S n . Both
charts are onto Rn , and 1 (U1 U2 ) = 2 (U1 U2 ) = Rn \{0}. The inverse map
to 1 reads,
(||y||2 1)N + 2y
1
(y)
=
.
1
1 + ||y||2
Thus,
y
,
2 1
1 (y) =
||y||2
a global diffeomorphism from Rn \{0} onto itself. The 2-sphere S 2 is in fact a
complex manifold: Identify R2
= C in the usual way, so that 1 , 2 take values in
C. Replace 2 by its complex conjugate, 2 (x) = 2 (x). In complex coordinates,
1
2 1
1 (z) = z , which is a holomorphic function. A different complex structure
is obtained by replacing 1 by it complex conjugate. For n 6= 2, the spheres S n
are not complex manifolds.
(b) Projective spaces. Let RP (n) be the quotient S n / under the equivalence
relation x x. Let : S n RP (n) be the quotient map. For any chart
: V Rn of S n with the property x V x 6 V , let U = (V ), and
: U Rn the unique map such that = . The collection of all such
charts defines an atlas for RP (n); the compatibility of charts follows from that
for S n .
1 (x) =

CONTENTS

(c) Grassmannians. The set GrR (k, n) of all k-dimensional subspaces of Rn is called
the Grassmannian of k-planes in Rn . A C -atlas may be constructed as follows.
For any subset I {1, . . . , n} of cardinality #I = k, let RI Rn be the subspace
consisting of all x Rn with xi = 0 for i 6 I. Thus each RI GrR (k, n).
Let UI Gr(k, n) be the set of all k-dimensional subspaces E Rn with
E (RI ) = {0}. There is a bijection I : UI
= L(RI , (RI ) )
= Rk(nk) of UI
with the space of linear maps AI : RI (RI ) , where each such A corresponds
to the subspace E = {x + AI (x)| x RI }.
To check that the charts are compatible, let I denote orthogonal projection
Rn RI . We have to show that for all intersections, UI UI, the map taking
AI to AI is smooth. The map AI is determined by the equations
AI (xI ) = (1 I )x, xI = I x
for x E, and x = xI + AI xI . Thus
AI(xI) = (I I)(AI + 1)xI , xI = I(AI + 1)xI .

The map I(AI + 1) restricts to an isomorphism S(AI ) : RI RI . The above

equations show,
AI = (I I)(AI + 1)S(AI )1 .
The dependence of S on the matrix entries of AI is smooth, by Cramers formula
for the inverse matrix. It follows that the collection of all I : UI Rk(nk)
defines on GrR (k, n) the structure of a manifold of dimension k(n k). Later,
we will give an alternative description of the manifold structure for the Grassmannian as a homogeneous space.
The discussion above can be repeated by replacing R with C everywhere.
The space GrC (k, n) is a complex manifold of complex dimension k(n k), i.e.
real dimension 2k(n k). In particular, GrC (1, n) = CP (n) is the complex
projective space.
(d) Flag manifolds. A (complete) flag in Rn is a sequence of subspaces {0} = V0
V1 Vn = Rn where dim Vk = k for all k. Let Fl(n) be the set of all flags.
It is a manifold of dimension (n2 n)/2, as one can see from the following rough
argument. Any real flag in Rn determines, and as determined by a sequence
of 1-dimensional subspaces L1 , , Ln , where each Lj is orthogonal to the sum
of Lk with k 6= j. Indeed the flag is recovered from this by putting E1 = L1 ,
E2 = L1 L2 and so on. There is an RP (n 1) of choices for L1 . Given L1
there is an RP (n 2) of choices for L2 (since L2 is orthogonal to L1 ). Given
L1 , . . . , Lj there is an RP (n j 1) of possibilities for Ln+1 . Hence we expect,
dim FlR (n) =

n1
X
j=1

(n j) = n(n 1)/2.

1. MANIFOLDS

It is possible to construct an atlas for FlR (n) using this idea. Below we will give
an alternative approach, by showing that the flag manifold is a homogeneous
space (see below). Similarly, one can define a complex flag manifold FlC (n), consisting of flags of subspaces in Cn . Also, one can define spaces FlR (k1 , . . . , kl , n)
of partial flags, consisting of subspaces {0} = E0 E1 El El+1 = Rn
of given dimensions k1 , . . . , kl , n. Note that FlR (k, n) = Gr(k, n).
(e) Klein Bottle. Let M be the manifold obtained as a quotient [0, 1] [0, 1]/
under the equivalence relation (x, 0) (x, 1), (0, x) (1, 1 x). Exercise: The
quotient space has natural manifold structure. Hint: Write M as a quotient of
R2 rather than [0, 1]2 . Then use charts for R2 to define charts for M .
A manifold M is called orientable if it admits an atlas such that the Jacobians of all
transition maps 1
have positive determinants. An manifold M with such an atlas
is called an oriented manifold.
Exercise 1.4. Show that GrR (1, n + 1) = RP (n), and GrR (k, n) = GrR (n k, n).
Exercise 1.5. Construct a manifold structure on the space M = Gror
R (k, n) of orin
ented k-planes in R .
Exercise 1.6. Show that RP (n) is orientable if and only if n is odd. Any idea for
which k, n the Grassmannian GrR (k, n) is orientable? (Answer: If and only if n is odd.)
Show that the Klein Bottle is non-orientable. Show that any complex manifold
(viewed as a real manifold) is oriented.
1.3. Smooth maps between manifolds.
Definition 1.7. A map F : N M between manifolds is called smooth (or C )
if for all charts (U, ) of N and (V, ) of M with F (U ) V , the composite map
F 1 : (U ) (V )
is smooth. The space of smooth maps from N to M is denoted C (N, M ). A smooth
map F : N M with smooth inverse F 1 : M N is called a diffeomorphism.
In the special case where the target space is the real line we write C (M ) :=
C (M, R). The space C (M ) is an algebra under pointwise multiplication, called the
algebra of functions on M . For any f C (M ), one defines the support of f to be the
closed set
supp(f ) = {x M | f (x) 6= 0}.
Clearly, the composition of any two smooth maps is again smooth. In particular, any
F C (N, M ) defines an algebra homomorphism
F : C (M ) C (N ), f 7 F f = f F

CONTENTS

called the pull-back.

> RO

F f

f
F

/N

In fact, a given map F : N M is smooth if and only if for all f C (M ), the pulled
back map F f = f F is smooth. (Exercise.)
If M is a manifold, we say that a coordinate chart : U Rm is centered at x M
if (x) = 0.
Definition 1.8. Let F C (N, M ) be a smooth map between manifolds of dimensions n, m, and x N . The rank of F at x, denoted rankx (F ), is the rank of the
Jacobian
D(x) ( f 1 ) : Rn Rm ,
for any choice of charts : U Rn centered at x and : V Rm centered at F (x).
The point x is called regular if rankx (F ) = m, and singular (or critical) otherwise. A
point y M is called a regular value if rankx (F ) = m for all x F 1 (y), singular value
otherwise.
Note that the rank of the map F does not depend on the choice of coordinate charts.
According to our definition, points that are not in the image of F are regular values.
Lemma 1.9. The map M Z, x 7 rankx (F ) is lower semi-continuous: That is, for
any x0 M there is an open neighborhood U around x0 such that rankx (F ) rankx0 (F )
for x U . In particular, if r = maxxM rankx (F ), the set {x M | rankx (F ) = r} is
open in M .
Proof. Choose coordinate charts : U Rn centered at x0 and : V Rm
centered at F (x0 ). By assumption, the Jacobian D(x0 ) : Rn Rm has rank r =
rankx0 (F ). Equivalently, some r r-minor of the matrix representing D(x0 ) has nonzero determinant. By continuity, the same r r-minor for any D(x) , x U has nonzero
determinant, provided U is sufficiently small. This means that the rank of F at x must
be at least r.

Definition 1.10. Let F C (N, M ) be a smooth map between manifolds of dimensions n, m. The map F is called a
submersion if rankx (F ) = m for all x M .
immersion if rankx (F ) = n for all x M .
local diffeomorphism if dim M = dim N and F is a submersion (equivalently, an
immersion).
Thus, submersions are the maximal rank maps if m n, and immersions are the
maximal rank maps if m n.

1. MANIFOLDS

Theorem 1.11 (Local normal form for submersions). Suppose F C (N, M ) is a

submersion, x0 N . Given any coordinate chart (V, ) centered at F (x0 ), one can find a
coordinate chart (U, ) centered at x0 such that the map F = F 1 : (U ) (V )
is given by
F (x1 , . . . , xn ) = (x1 , . . . , xm ).
Proof. The idea is simply to take the components of F as the first m components
j (x) of the coordinate function near x0 .
Using the given coordinate chart around F (x0 ) and any coordinate chart around x0 ,
we may assume that M is an open neighborhood of 0 Rm and N an open neighborhood of 0 Rn . We have to find a smaller neighborhood U of 0 M Rn and a
diffeomorphism : U (U ) Rn such that F = F 1 has the desired form.
i
By a linear transformation of Rn , we may assume that F
(0) = ij for i, j m. Let
xj
: M Rn be the map


(x1 , . . . , xn ) = F 1 (x1 , . . . , xn ), . . . , F m (x1 , . . . , xn ), xm+1 , . . . , xn .

The Jacobian of at x = 0 is just the identity matrix. Hence the inverse function
theorem applies: There exists some smaller neighborhood U of 0 M Rn , such that
is a diffeomorphism U (U ). Then (U, ) is the desired coordinate system.


Theorem 1.12 (Local normal form for immersions). Suppose F C (N, M ) is an

immersion, x0 N . Given any coordinate chart (U, ) centered at x0 , one can find a
coordinate chart (V, ) centered at F (x0 ) such that the map F = F 1 : (U )
(V ) is given by
F (x1 , . . . , xn ) = (x1 , . . . , xn , 0, . . . , 0).
Proof. The idea is to use the given coordinates on U as coordinates on F (U ),
near F (x0 ), and supplementary m n coordinates in transversal directions to get
coordinates on M near F (x0 ).
Using the given coordinates chart around x0 and any coordinate chart around F (x0 ),
we may assume that M is an open neighborhood of 0 Rm and N an open neighborhood
of 0 Rn . We have to find a smaller neighborhood V M , and a diffeomorphism :
V (V ) Rm such that F = F has the form F (x1 , . . . , xn ) = (x1 , . . . , xn , 0, . . . , 0).
Let x1 , . . . , xn be the coordinates on Rn and y 1 , . . . , y m the coordinates on Rm . By
i
assumption, the matrix F
(x) has maximal rank n for all x U . By a linear change of
xj
i

(x0 ))i,jn = ij . Consider the map Consider

coordinates on V , we may assume that ( F
xj
the map
: N Rmn Rm , (x, s) 7 F (x) + (0, . . . , 0, sn+1 , . . . , sm )
The Jacobian of at 0 is just the identity matrix. Hence, the inverse function theorem
applies, and we can find an open neighborhood V around 0 M Rm such that 1 is a
well-defined diffeomorphism over V . The map = 1 : V (V ) U Rmn Rm
is the desired coordinate chart.


10

CONTENTS

Exercise 1.13. Let : N M be a surjective submersion. Suppose F : M X

is any map such that F lifts to a smooth map F : N X, i.e. F = F . Then F is
smooth.
Definition 1.14. Let M be a manifold. A subset S M is called an embedded
submanifold if for each x0 S there exists a coordinate system (U, ) centered at x0 ,
such that
(U S) = {(x1 , . . . , xm ) (U )| xk+1 = . . . = xm = 0}
It is obvious from the definition that submanifolds inherit a manifold structure from
the ambient space: Given a covering of S by coordinate charts (U, ) as above, one
simply takes (U S, |U S ) to define an atlas for S.
The following two Theorems are immediate consequences of the normal form theorems
for submersions and immersions, respectively.
Theorem 1.15. Let F C (N, M ) be a submersion. Then each level set S =
F 1 (y) for y M is an embedded submanifold of dimension n m.
Theorem 1.16. Let F C (N, M ) be an immersion. For each x0 N there exists
a neighborhood U of x0 such that the image S = F (U ) is an embedded submanifold of
dimension m n.
These theorems provide many new examples of manifolds. Often, manifolds are
obtained as level sets for a smooth function on a Euclidean space RN . For example, we
see again that S n Rn+1 is a manifold. Another example is the 2-torus, for 0 < r < R
the radii of the small and big circles, given as a level set G1 (r2 ) where G C (R3 )
is the function,
p
G(x1 , x2 , x3 ) = (x3 )2 + ( (x1 )2 + (x2 )2 R)2 .

The 2-torus can also be described as the image of an immersion F : R2 R3 , where

F (, ) = (x1 , x2 , x3 ) is given by
x1 = (R + r cos ) cos ,
x2 = (R + r cos ) sin ,
x3 = r sin

In fact, it is clear that this map descends to an embedding R2 /(2Z)2 R3 as a

submanifold.
We will show below that any compact manifold can be smoothly embedded into some
N
R . (In fact, compactness is not necessary but we wont prove this harder result.)

1. MANIFOLDS

11

Exercise 1.17. Construct an explicit embedding of the Klein bottle into R4 . Solution: Given 0 < r < R define F (, ) = (x1 , x2 , x3 , x4 ) where
x1
x2
x3
x4

=
=
=
=

(R + r cos ) cos ,
(R + r cos ) sin ,
r sin cos /2,
r sin sin /2.

for 0 2 and 0 2.
1.4. Tangent vectors. There is a number of equivalent coordinate-free definitions
for the tangent space Tx M of a manifold x M at some point x M . Our favorite
definition defines Tx M as the space of directional derivatives.
Definition 1.18. Let M be a manifold. A tangent vector at x M is a linear map
v : C (M ) R satisfying the Leibnitz rule (product rule)
v(f1 f2 ) = v(f1 )f2 (x) + f1 (x)v(f2 ).
The vector space of tangent vectors at x is denoted Tx M , and called the tangent space
at x.
It follows immediately from the definition that any tangent vector vanishes on constant functions. Indeed, if 1 denotes the constant function f (x) = 1, the product rule
gives
v(1) = v(1 1) = v(1) 1 + 1 v(1) = 2v(1)
thus v(1) = 0. Furthermore, the product rule shows that for any two functions g, h with
g(x) = h(x) = 0, v(gh) = 0.
Lemma 1.19. If U Rn is an open subset and x0 U , the tangent space Tx0 U is
isomorphic to Rn , with basis the derivatives in coordinate directions,

f
|x0 : f 7
(x0 )
i
x
xi

Proof. We may assume x0 = 0. Let v TP

0 U . Given f C (U ), use Taylors
Pm
m f
i
theorem with remainder to write f (x) = f (0) + i=1 xi (0)x + i=1 gi (x)xi , where gi
vanishes at 0. v vanishes on the constant f (0), and also on the products xi gi (x). Thus
v(f ) =

m
m
X
X
f
f
i
(0)v(x
)
=
ai i |x=0
i
x
x
i=1
i=1

where ai = v(xi ).

Lemma 1.20. Let M be a manifold of dimension m. If : U M is any open

neighborhood of x, the map
: Tx U Tx M, v(f ) = v(f |U )

12

CONTENTS

is an isomorphism. In particular, any coordinate chart (U, ) around x gives an isomorphism Tx M

= Rm .
Proof. We first show that for any v Tx M , v(f ) depends only on the restriction
of f to an arbitrary open neighborhood of x. Equivalently, we have to show that if f
vanishes on a neighborhood of x then v(f ) = 0. Using a coordinate chart around x
construct C (M ) with = 1 on a neighborhood of x and = 0 on a neighborhood
of the support of f . Let g = 1 . Then f g = f since g = 1 on supp(f ). Since both
f, g both vanish at x, v(f ) = v(f g) = 0, as required.
This result can be re-interpreted as follows: Let V M be an open neighborhood
of x, and FV (M ) C (M ) be the functions supported in V . Then Tx M can also be
defined as the space of linear maps F R satisfying the Leibnitz rule. Indeed, if is
supported on V and = 1 near x, then any function f coincides with f FV (M ) near
x. In particular, choose V with V U . Then FV (U ) = FV (M ), and it follows directly
that Tx M = Tx U .

Definition 1.21. Let F C (N, M ) be a smooth map, and x N . The tangent
map
dx F : Tx N TF (x) M
is defined as follows:
(dx F (v))(f ) = v(F f ), f C (M ).
It is immediate from the definition that dx F is a linear map. We often write F :
Tx N TF (x) M if the base point is understood. The map F is also called push-forward.
Under composition of maps, (F1 F2 ) = (F1 ) (F2 ) .
Exercise 1.22. Let U Rm and V Rn be open subsets and F C (U, V ). For
all x U , the isomorphisms Tx U = Rm and TF (x) V = Rn identify the tangent map
dx F : Tx U TF (x) V with the Jacobian Dx F : Rm Rn . That is,
X F j

,
(dx F )( i ) =
x
xi y j
j
where xi , y j are the coordinates on U, V .
Thus dx F is just the coordinate-free definition of the Jacobian: any choice of charts
(U, ) around x and (V, ) around F (x) identifies dx F with the Jacobian of the map
F 1 : (U ) (V ) at (x). In particular,
rankx (F ) = rank(dx F ).
F is an immersion if dx F is injective everywhere, and a submersion if dx F is surjective
everywhere.
Definition 1.23. A map F C (N, M ) is called an embedding if F is an injective
immersion and F is a homeomorphism onto F (N ) (with the subspace topology).

1. MANIFOLDS

13

Thus, a 1:1 immersion is an embedding if and only if the map N F (N ) is open

for the subset topology. That is, one has to verify that for each open subset U of N , the
image F (U ) can be written F (U ) = F (N ) V where V is open in M .
Example 1.24. Consider the curve
: R R2 , t 7 (sin(2t), cos(t)).
Then is an immersion, with image a figure 8. Let F be the restriction of to the
open interval (/2, 3/2). Then F is a 1-1 immersion, but is not an embedding. For
instance, the image of the open interval (0, ) is not open in the subspace topology. Note
that the image of F is still the full figure 8, so it is not an embedded submanifold.
Exercise 1.25. Show that the image of an embedding is an embedded submanifold.
Conversely, if N M is an embedded submanifold with the induced structure of a
manifold on N , then the inclusion map N M is an embedding.
If F is an embedding, then dx F : Tx N TF (x) M is injective, so Tx N is identified
with a subspace of TF (x) M . In particular, if N is an embedded submanifold of Rm , the
tangent spaces to N are canonically identified with subspaces of Rm .
Exercise 1.26. Suppose F C (N, M ) has a M as a regular value, so F 1 (a) is
an embedded submanifold of N . Show that
Tx (F 1 (a)) = ker(dx F ) Tx N
for all x F 1 (a). (Hint: Use the normal form theorem for submersions.)
1.5. Velocity vectors for curves. If I R is an open interval, a map
C (I, M ) is called a parametrized curve in M . For any t I, one can define the
velocity vector
(t)

T(t) M

by (t)

= () ( t
). The action of the velocity vector on functions is, by definition of
push-forward,
d
(t)(f

) = f ((t)).
dt

Exercise 1.27. Show that every v Tx M is of the form v = (0)

for some curve

: (, ) M with (0) = x. (Hint: Use a chart.)
Exercise 1.28. Let F C (N, M ) and C (I, N ) a smooth curve. Show that
F ((t))

=
the tangent vector for the curve F .

d
F ((t)),
dt

14

CONTENTS

1.6. Jet spaces. Let Cx (M ) be the space of functions vanishing at x, and Cx (M )2

linear combinations of products of such functions. Thus Cx (M )2 consists of
functions on M that vanish to second order at x. From the definition, it is clear that
any tangent vector is determined by its value on Cx (M ), and is zero on Cx (M )2 . Thus
one has a natural linear map Tx M (Cx (M )/Cx (M )2 ) .
Cx (M )

Proposition 1.29. The map

Tx M
= (Cx (M )/Cx (M )2 )
is a vector space isomorphism.
Proof. Clearly, this map is injective. To show surjectivity, we have to show that
if v : C (M ) R is any linear map vanishing on constants and on Cx (M )2 , then
v satisfies the Leibnitz rule. Given f, g write f = a + (f a) where f (x) = a, and
g = b + (g b) where g(x) = b. Then f a and g b vanish at x. Thus
v(f g) = v(ab) + av(g b) + v(f a)b + v((f a)(g b))
= av(g b) + v(f a)b
= av(g) + v(f )b.

Exercise 1.30. Show that Cx (M ) is a maximal ideal in the algebra C (M ), and
that every maximal ideal is of this form, for some x.
Exercise 1.31. Give the definition of the tangent map dx F from this point of view.
This second definition suggests a definition of higher tangent bundles: One defines,
for all k = 1, 2, . . .
Jxk M := C (M )/Cx (M )k ,
the kth jet space. For any function f Cx (M ), its image Jxk (f ) in Jxk M is called the
kth order jet at x. In local coordinates, Jxk (f ) represents the kth order Taylor expansion
at x.
Thus Jx0 M = R, Jx1 M = R Tx M . For larger k one still has projection maps
k+1
Jx M Jxk M but these maps no longer split: There is no coordinate invariant meaning
of the terms of order exactly k + 1 of a Taylor expansion, unless the terms of order k
vanish.
2. Partitions of unity
Before developing the theory of vector fields, differential forms etc., we need an
important technical tool known as partitions of unity.
We recall a few notions from topology. An open cover (V )B of a topological space
M is called a refinement of a given open cover (U )A if each V is contained in some
U . A cover (U )A is called locally finite if each point in M has an open neighborhood
that intersects only finitely many U s. A topological space is called paracompact if

2. PARTITIONS OF UNITY

15

every open cover has a locally finite refinement. We will show now that manifolds are
paracompact. First we need:
Lemma 2.1. Every manifold has an open covering (Si )N
i=1 where each S i is compact,
and S i Si+1 , for all i.
Proof. Let (Ui )
i=1 be a countable basis of the topology. Already the Ui s with
compact closure are a basis for the topology, so passing to a subsequence we may assume
that each Ui has compact closure. Let S1 = U1 . Let k(1) > 1 be an integer such that
U1 , . . . , Uk(1) cover S 1 . Put S2 := U1 . . . Uk(1) . Let k(2) > k(1) be an integer such
that U1 , . . . , Uk(2) cover S 2 . Put S3 := U1 . . . Uk(2) . Proceeding in this fashion, one
produces a sequence Si with the required properties.

Theorem 2.2. Every manifold M is paracompact. In fact, every open cover admits
a countable, locally finite refinement consisting of open sets with compact closures.
Proof. Let (U )A be any given cover of M . Here A is any indexing set. Let
S1 , S2 , . . . be the sequence from Lemma 2.1. Each S i is compact, and is therefore covered
by finitely many U s. It follows that there exists a countable subset A A such that
(U )A is a covering of M . Replacing A with A , we may assume that our indexing set
is A = {1, 2, . . .} (possibly finite). For each j, let k(j) be an integer such that the sets
Ui with i k(1) cover S j . For 1 k k(j) define
Vkj = Uk (Sj+1 \S j1 )
where j is the integer such that k(j 1) < k k(j) (we put k(0) = 0 and S0 = ).
Then the collection Vkj s are a locally finite refinement of the given cover, and each Vkj
has compact closure.

Lemma 2.3. Let C M be a compact subset of some manifolds M . For any open
neighborhood U of C there exists a smooth function f C (M ) with 0 f 1, such
that f = 1 on C and supp(f ) U .
Proof. For each x C, choose a function fx C (M ) with supp(fx ) U and
fx = 1 on a neighborhood of x. (Such a function is easily constructed using a local
coordinate chart.)Let Cx = fx1 (1). Then {int(Cx )}xC are an open cover of C. By
compactness, there exists a finite subcover. Thus we can choose a finite collection of
points xi such that the interiors of the sets Ci = Cxi cover C. Write fi = fxi . Then
N
Y
f = 1 (1 fi )
i=1

has all the required properties.

Lemma 2.4 (Shrinking Lemma). Let (U )A be a locally finite covering of a manifold

M by open sets with compact closure. Then there exists a covering (V )A such that
V U for all A.

16

CONTENTS

Proof. Since the covering is locally finite, it is in particular countable. (Taking an

open cover Si of M as above, where each S i is compact and contained in Si+1 , only
finitely many U s meet each Si . This implies countability.) Thus we may assume that
A = N or a finite sequence, and write the given cover as (Ui )N
i=1 , N N. Inductively,
we now construct open subsets Vi with Vi Ui , such that V1 , . . . , Vi , Ui+1 , Ui+2 , . . . still
form a cover of M . Given V1 , . . . , Vi , we construct Vi+1 as follows: Choose 0 f 1
supported in Ui+1 with f = 1 on the compact set
[
[


C = M\
Vi
Ui Ui+1 .
ki

ki+2

Then take Vi+1 to be the open set where f > 0. Then V1 , . . . , Vi+1 , Ui+2 , . . . is an open
cover. Since the original cover was locally finite, (Vi )N

i=1 is a cover of M .
Theorem 2.5. Let (U )A be an open covering of a manifold M . Then there exists
a partition of unity subordinate to the cover U , that is, a collection of functions
such that
(a) Each point x M has an open neighborhood U meeting the support of only
finitely many s.
(b) 0 1,
(c) P
supp( ) U ,
(d) = 1
(The sum is well-defined, since near each point only finitely many are non-zero.)
Proof. Suppose first that the cover is locally finite and that each U is compact.
Choose a shrinking (V )A as in the Lemma. Choose functions 0 f 1 supported
P
in U with f = 1 on V . Since the covering is locally finite, the sum f =
f
exists, and clearly f > 0 everywhere. Put = f /f . This proves the Theorem for
locally finite covers with compact closures. In the general case, choose a locally finite
refinement (U )B consisting of open sets with compact closures. There is a function
P
j : B A, 7 = j() such that U Uj() , and define = j()= .

Exercise 2.6. Show that Lemma 2.3 holds for any closed, not necessarily compact,
subset of M .
Here is a typical application of partitions of unity.
Theorem 2.7. Every manifold M can be embedded into Rk , for k sufficiently large.
Proof. We will prove this only under the additional assumption that M admits a
finite atlas. The existence of a finite atlas is obvious if M is compact. It can be shown
that in fact, every manifold admits a finite atlas but the proof is not so easy (see e.g. the
book Greub-Halperin-Vanstone). Let (Uk , k )N
k=1 be a finite atlas. Choose a partition of
unity subordinate k , subordinate to the cover Uk . Then k k : Uk Rm extends by
zero to a function fk C (M, Rm ). Let
F : M R(m+1)N , x 7 (f1 (x), . . . , fN (x), 1 (x), . . . , N (x)).

3. VECTOR FIELDS

17

We claim that F is an embedding: That is, F is a 1-1 immersion that is a homeomorphism

onto its image. First, F is 1-1. For suppose F (x) = F (y). Thus i (x) = i (y) for all i.
Choose i with i (x) > 0. In particular, x, y Ui . Dividing the equation fi (x) = fi (y)
by i (x) = i (y) we find i (x) = i (y), thus x = y. More generally, for any x N and
any sequence yi N , F (yi ) F (x) implies yi x. This shows that the inverse map is
continuous, so F is a homeomorphism onto its image. Finally, let us show that F has
maximal rank at x: For suppose v ker(dx F ). Then v(i ) = 0 and v(fi ) = 0 for all i.
Choose i such that i (x) > 0, thus x Ui . Writing fi = i i the product rule gives
0 = v(i i ) = i (x)v(i ),
thus v(i ) = 0. But i is a diffeomorphism onto its image, so v = 0.

A famous theorem of Whitney (1944) says that any manifold of dimension dim M =
m can be embedded into R2m , and immersed into R2m1 . See Smale, Bull.Am.Math.Soc.
69 (1963), 133-145 for a survey of results on embeddings and immersions.
3. Vector fields
3.1. Vector fields. Suppose A is an algebra over R. A derivation of A is a linear
map D : A A satisfying the Leibnitz rule
D(ab) = aD(b) + D(a)b.
If D1 , D2 are derivations, then so is their commutator [D1 , D2 ] = D1 D2 D2 D1 . Recall
that the commutator satisfies the Jacobi identity,
[D1 , [D2 , D3 ]] + [D2 , [D3 , D1 ]] + [D3 , [D1 , D2 ]] = 0.
Thus Der(A) is a Lie subalgebra of the algebra End(A). If A is commutative, the space
Der(A) is an A-submodule of End(A): If D is a derivation and x A, then xD is also
a derivation.
Definition 3.1. A vector field on M is a derivation X Der(C (M )). That is, X
is a linear map X : C (M ) C (M ) satisfying the Leibnitz rule
X(f g) = (Xf )g + f (Xg), f, g C (M ).
The space of vector fields will be denoted X(M ). If X, Y X(M ), the vector field
[X, Y ] = X Y Y X is called the Lie bracket of X and Y .
Thus, the space X(M ) of vector fields is a Lie algebra. They are also a C (M )module, that is, f X X(M ) for all f C (M ) and X X(M ). Evaluation at any
point x M defines a linear map
X Tx M, X 7 Xx , Xx (f ) = (Xf )(x).
Conversely X can be recovered from the collection of tangent vectors Xx with x M .
We write
supp(X) = {x M | Xx 6= 0}.

18

CONTENTS

One can think of a vector field as a family of tangent vectors depending smoothly on the
base point.
Exercise 3.2 (Locality). Show that vector fields are local, in the following sense: If
U is an open subset of a manifold M , and X X(M ), there exists a unique vector field
XU X(U ) (called the restriction of X to U ) such that (XU )x = Xx for all x U .
The action of X on functions f C (M ) supported in U is given in terms of
the restriction by Xf |U = XU (f |U ). Using a partition of unity, this means that the
restrictions of X to coordinate charts determine the vector field X. In local coordinates,
vector field have the following form:
Lemma 3.3. Suppose U Rn is an open subset. Then any X X(U ) has the form
n
X

X=
ai
xi
i=1
where ai C (U ).
Proof. From the description of tangent vectors v Tx U , we know that
n
X

Xx =
ai (x)
xi
i=1
for some function ai (x). We have to show that ai is smooth. But this is clear since
ai = X(f ) where f C (U ) is the coordinate function f (x) = xi .

Any diffeomorphism F C (N, M ) induces a map F : X(M ) X(N ) of vector
fields, with F (F (X)(f )) = X(F f ). Thus
F (Xx ) = (F X)F (x) .
For a general map F , this equation does need not define a vector field F X, for two
reasons: 1) If F is not surjective, the equation does not specify F X at points away from
the image of F , 2) If F is not injective, the equation assigns more than one value to
points y M having more than one pre-image.
Definition 3.4. Let F C (N, M ). Two vector fields X X(N ) and Y X(M )
are called F -related if
F (Y (f )) = X(F f )
for all f C (M ). One writes X F Y .
Proposition 3.5. X F Y if and only if
F (Xx ) = YF (x)
for all x N . If X1 F Y1 and X2 F Y2 then
[X1 , X2 ] F [Y1 , Y2 ].
Proof. Exercise.

3. VECTOR FIELDS

19

For example, if F is the embedding of a submanifold, then X X(N ) is F -related

to a vector field Y X(M ) if and only if Y is tangent to F (N ) everywhere. The vector
field X is then just the restriction of Y to N .
Definition 3.6. A Riemannian metric on a manifold M is a symmetric C (M )bilinear form g : X(M ) X(M ) C (M ) which is positive definite in the sense that
for all x M , g(X, X)(x) > 0 for Xx 6= 0.
Exercise 3.7. Every manifold M admits a Riemannian metric g. (Hint: Define
such a metric in charts, and then use a partition of unity to patch the local solutions
together.) If g is a Riemannian metric and x M , the value g(X, X)(x) depends
only on Xx . If U M is open, there exists a unique Riemannian metric gU such that
gU (XU , YU ) = g(X, Y )|U .
3.2. The flow of a vector field. Suppose A is a finite dimensional algebra, and
Aut(A) its group of algebra automorphisms. A 1-parameter subgroup of automorphisms
is a group homomorphism : R Aut(A), t 7 t , i.e. a map such that t1 +t2 =
t1 t2 for all t1 , t2 R. For any 1-parameter group of automorphisms, the derivative
D = dtd |t=0 t : A A is well defined. By taking the derivative of the equation
t (ab) = t (a)t (b)
one finds that D Der(A). Conversely, every derivation gives ride to a 1-parameter
group t = exp(tD) (exponential of matrices), and one check that each t is an automorphism. Thus, if dim A < one can identify derivations of an algebra with 1-parameter
groups of automorphisms.
For the infinite-dimensional algebra A = C (M ), this does not directly apply. We
will see that so-called complete vector fields X X(M ) define a 1-parameter group of
diffeomorphisms t of M , where the algebra automorphism t of C (M ) plays the role
of exp(tX). This flow of X is constructed using integral curves.
Let X be a vector field on a manifold M . A curve : I M is called an integral
curve of X if for all t I,
(t)

= X(t) .
If U Rm is an open subset and X X(U ) is a vector field on U , this has the following
interpretation. Letting xi be the local coordinates on Rm , write
X
X=
ai i
x
i
where ai C (U ). Also write (t) = ( 1 (t), . . . , m (t)). Then the tangent vector
(t)

T(t) U is just
X

(t)

=
i i  .
x (t)
i

20

CONTENTS

To see this, apply (t)

to a function f C (U ):
X d i f 

(t)(f

) = f ((t)) =
 .
t
dt xi (t)
i

Thus is an integral curve if and only its components satisfy the following system of
first order ordinary differential equation:
d i
= ai ((t))
dt
From ODE theory, it follows that for any x U , there exists a unique maximal solution
x of this system: That is, there is an open interval Ix around t = 0, and a curve
x : Ix U with
x (0) = x, xi (t) = ai (x (t))
such that any other solution of this initial value problem is obtained by restriction to
a subinterval. Moreover, this solution depends smoothly on the initial value x. By
applying this result in manifold charts, one obtains:
Theorem 3.8. Let X be a vector field on a manifold M . For each x M there
exists a unique maximal solution x : Ix M to the initial value problem
x (0) = x, x (t) = Xx (t) .
The solution depends smoothly on the initial value x, in the following sense:
[
U=
({x} Ix ) M R.
xM

Then U is an open neighborhood of M {0} in M R, and the map

: U M, (x, t) 7 x (t)
is smooth.
Here maximal solution means that any other solution is obtained by restriction to
some subinterval.
One calls the flow of the vector field X. If U = M R, that is if Ix = R for
all x, the vector field is called complete, and the flow is called a global flow. A simple

on M = (0, 1) R. By choosing a
example for an incomplete vector field is X = t

diffeomorphism (0, 1) = R, one obtains an incomplete vector field on R. Let us write

t (x) = (x, t) for t Ix .
Exercise 3.9. Show that on a compact manifold, any vector field is complete. More
generally, any compactly supported vector field on a manifold is complete.
Theorem 3.10. Suppose X X(M ) is a complete vector field, and t its flow. Then
each t is a diffeomorphism, and the map
R Diff(M ), t 7 t

3. VECTOR FIELDS

21

is a group homomorphism. In particular,

0 = Id, t1 t2 = t1 +t2 .
Conversely, if t 7 t is any such group homomorphism and if the map (t, x) = t (x)
is smooth, then t is the flow of a uniquely defined complete vector field X, called the
generating vector field for the flow. For all f C (M ) one has
X(f ) = LX (f ) :=

d
f.
dt t=0 t

Proof. We first prove the identity t1 t2 = t1 +t2 for t1 , t2 R. Let x M .

Given t2 , both (s) = s (t2 (x)) and (s) = s+t2 (x) are integral curves for X. Indeed,
d
d
(s) =
u (x)|u=s+t2 = Xu (x) |u=s+t2 = X(s) .
ds
du
Since , have the same initial value, they coincide on their domain of definition. In
particular, (t1 ) = (t1 ) which proves t1 (t2 (x)) = t1 +t2 (x). In particular, if X is
complete, this equation holds for all t1 , t2 . Setting t1 = t, t2 = t we see that t is a

smooth inverse to t . Conversely, if t is a global flow, define X by Xx = t

|t=0 t (x).
Using local coordinates, one checks that this defines a smooth vector field.

Exercise 3.11. How much of this theorem goes though for incomplete vector fields?
Exercise 3.12. Show that the map t : Rm Rm given by multiplication by et is
a global flow, and give a formula for the generating vector field. More generally, if A is
any m m matrix, show that the map
t (x) = etA x
is a global flow, and find its generating vector field.
Exercise 3.13. Let X X(N ), Y X(M ) be vector fields and F C (N, M ) a
Y
smooth map. Show that X F Y if and only if it intertwines the flows X
t , t : That is,
Y
F X
t = t F.

Exercise 3.14. Show that for any vector field X X(M ) and any x M with
Xx 6= 0, there exists a local chart around x in which X is given by the constant vector
field x 1 . Hint: Show that if S is an embedded codimension 1 submanifold, with x S
and Xx 6 Tx S, the map U (, ) M is a diffeomorphisms onto its image, fo some
open neighborhood U of x in S. Use the time parameter t and a chart around x U to
define a chart near x.
For any vector field X X(M ) and any diffeomorphism F C (N, M ), we define
F X X(N ) by
F X(F f ) = F (X(f )).

22

CONTENTS

Thus F : X(M ) X(N ) is just the inverse map to F : X(N ) X(M ). Any complete
vector field X X(M ) with flow t gives rise to a map t : X(M ) X(M ). One
defines the Lie derivative LX of a vector field Y X(M ) by
d 
LX (Y ) =  t Y X(M ).
dt t=0
In fact, this definition also makes sense for incomplete X: (To define the restriction of
LX Y to an open set U M with compact closure, let > 0 be small enough such
[, ] Ix for all x U . Then t : U 7 t (U ) is defined for |t| < , and the equation
above makes sense.)
Theorem 3.15. For any X, Y X(M ), the Lie derivative LX Y is just the Lie bracket
[X, Y ]. One has the identity
[LX , LY ] = L[X,Y ] .

Proof. Let t = X
t be the flow of X. For f C (M ) we calculate,

(LX Y )(f ) =
=
=
=
=

d
|t=0 (t Y )(f )
dt
d
|t=0 t (Y (t (f )))
dt
d
|t=0 (t (Y (f )) Y (t (f ))
dt
X(Y (f )) Y (X(f ))
[X, Y ](f ).

The identity [LX , LY ] = L[X,Y ] just rephrases the Jacobi identity for the Lie bracket. 
The definition of Lie derivative gives the formula
d X

(1)
(t ) Y = (X
t ) (LX Y )
dt
d
d
X

X d
X
|u=0 (X
by the calculation, dt (t ) Y = du
u+t ) Y = (t ) du |u=0 (u ) Y . From this we
obtain:
Y
Theorem 3.16. Let X, Y be two complete vector fields, with flows X
t and s . Then
X
Y
X
Y
Y
X
[X, Y ] = 0 if and only if t and s commute for all t, s: t s = s t .

Proof. Suppose [X, Y ] = 0. Then

d X

X
( ) Y = (X
t ) LX Y = (t ) [X, Y ] = 0
dt t

X
for all t. Hence (X
t ) Y = Y for all t, which means that Y is t -related to itself.
Y
It follows that X
t takes the flow s of Y to itself, which is just the desired equation
X
Y
Y
X
Y
Y
X
t s = s t . Conversely, by differentiating the equation X
t s = s t
with respect to s, t, we find that [X, Y ] = 0.


4. DIFFERENTIAL FORMS

23

Y
Thus [X, Y ] measures the extent to which X
t and t fail to commute. This can be
made more precise:

Exercise 3.17. Suppose X, Y are complete vector fields, and f C (M ).

a) Prove the formula for the kth order Taylor expansion of (X

t ) f:

(X
t ) f

k
X
tj
j=0

j!

(X)j (f ) + O(tk+1 ).

Formally, one writes (X

t ) = exp(tX).
b) Let Ft be the family of diffeomorphisms,
Y
X
Y
Ft = X
t t t t .

Show that
Ft f = f + t2 [X, Y ](f ) + O(t3 ).
4. Differential forms
4.1. Super-algebras. A super-vector space is a vector space E with a Z2 -grading.
Elements of degree 0 mod 2 are called even, and elements of degree 1 mod 2 are called
odd. Thus a super-vector space is simply a vector space with a decomposition E =
E 0 E 1 . Elements in E 0 are called even and those in E 1 are called odd. We will also
write E 0 = E even and E 1 = E odd , in particularly if E carries other gradings that might
be confused with the Z2 -grading. The space End(E) of endomorphisms has a splitting
End(E) = End(E)0 End(E)1 ,
where End(E)0 consists of endomorphisms preserving the splitting E = E 0 E 1 , and
End(E)1 consists of endomorphisms taking E 0 to E 1 and E 1 to E 0 . A = End(E) is a
first example of a super-algebra, that is, a Z2 -graded algebra 1 A = Aeven Aodd such
that
Ak Al Ak+l mod 2
for k, l {0, 1}. The sign conventions of supermathematics decrees:
Super-sign convention: A minus sign appears whenever two odd elements interchange their position.
Physicists thinks of elements of odd degree as fermions, and those of even degree
as bosons. Her are some examples of the super-sign convention. The tensor product
of super-algebras A, B is defined to be the usual tensor product A B, with Z2 -grading
M
(A B)k =
Al B m ,
l+m=k mod 2

and algebra structure

(a1 b1 )(a2 b2 ) = (1)m1 l2 (a1 a2 ) (b1 b2 )
1In

this course, algebra always means an associative algebra over R with unit 1.

24

CONTENTS

for ai Ali , bj Bmj . Also, if A is any super-algebra, the super-commutator of two

elements a1 Ak1 and a2 Ak2 is defined by
[a1 , a2 ] = a1 a2 (1)k1 k2 a2 a1 Ak1 +k2 .
Proposition 4.1. The super-commutator is super-skew symmetric,
[a1 , a2 ] = (1)k1 k2 [a2 , a1 ]
and satisfies the super-Jacobi identity,
[a1 , [a2 , a3 ]] + (1)k1 (k2 +k3 ) [a2 , [a3 , a1 ]] + (1)k3 (k1 +k2 ) [a3 , [a1 , a2 ]] = 0.
Proof. Exercise.

More generally, a super-space E with bracket [, ] satisfying these identities is called

a super-Lie algebra.
An derivation of a superalgebra A of degree r {0, 1} is a linear map D : A A
such that
D(ab) = D(a)b + (1)kr aD(b)
for all a Ak and b Al . Note that any derivation is determined by its values on
generators for the algebra, and that D(1) = 0 for any derivation D. We denote
Der(A) = Der0 (A) Der1 (A)
the space of super-derivations.
Exercise 4.2. Show that Der(A), with bracket the super-commutator of endomorphisms, is super-Lie algebra. If A is super-commutative, show that Der(A) is also an
A-module (by multiplication from the left).
Exercise 4.3. Suppose A is a super-algebra. Show that the map : A End(A)
taking a A to the operator (a) of multiplication (from the left) by a is a homomorphism of super-algebras, i.e. it preserves products and degrees.
Exercise 4.4. Let E, F be super-vector spaces. Define the tensor product of two
endomorphisms of finite degree, in such a way that it respects the sign convention of
supermathematics. Show that with this definition,
End(E F ) = End(E) End(F ).
L
Many super-vector spaces arise as Z-graded algebras E = kZ E k , by reducing the
degree mod 2:
M
M
E even =
E 2k , E odd =
E 2k+1 .
kZ

kZ

In this case, we will call E a Z-graded super-vector space. Ordinary vector spaces E can
be viewed as super-vector spaces by putting E even = E, E odd = 0.

4. DIFFERENTIAL FORMS

25

4.2. Exterior algebra. We now give our main example of a graded algebra. Let
E be a finite dimensional real vector space. (The example to keep in mind is E = Tx M ,
the tangent space to a manifold. But other examples will appear as well.)
For any finite dimensional vector space E over R, one defines 0 E = R, 1 E = E ,
and more generally
k E = {antisymmetric k-linear maps : E
E} R}.
| {z
k times

Thus E if it is linear in each argument and satisfies

(v(1) , . . . , v(k) ) = sign()(v1 , . . . , vk )

for any v1 , . . . , vk E and any permutation of the index set 1, . . . , k. Each k E is
a finite dimensional vector space. If e1 , . . . , en is a basis for E, any is determined by
its values on ei1 , . . . , eik for any i1 < . . . < ik . Since the number of ordered k-element
n!
subsets of {1, . . . , n} is k!(nk)!
, it follows that
dim k E =

n!
k!(n k)!

for k n, and k E = 0 for k > n. Note dim nk E = k E . Non-zero elements of

the 1-dimensional vector space n E are called volume elements. The direct sum of the
vector space k E is denoted2
n
M
E =
k E ,
k=0

its dimension is dim E = 2 , the number of subsets of {1, . . . , n}.

The graded algebra structure on the vector space E is the so-called called wedge
product.
For 1 k1 E and 2 k2 E one defines 1 2 k E , k = k1 + k2 by
anti-symmetrization
1 X
sign()1 (v(1) , . . . , v(k1 ) )2 (v(k1 +1) , . . . , v(k) )
(1 2 )(v1 . . . , vk ) =
k1 !k2 ! S
k

where runs over all permutations. The wedge product extends to all of E by linearity.
For example, if f1 , f2 E then
(f1 f2 )(v1 , v2 ) = f1 (v1 )f2 (v2 ) f1 (v2 )f2 (v1 ).
More generally, if f1 , . . . , fk E ,
(f1 fk )(v1 , . . . , vk ) = det(fi (vj )).

(2)

The following properties of the wedge product are left as an exercise.

2In

the infinite dimensional case dim E = , one defines the exterior algebra somewhat differently.
Fortunately, we will only be concerned with finite dimensions.

26

CONTENTS

Proposition 4.5.
(a) The wedge product is associative and super-commutative.
(b) Let e1 , . . . , en be a basis of E and f 1 , . . . , f n the dual basis of E . For each
ordered subset I = {i1 , . . . , ik } {1, . . . , n} let f I = f i1 . . . f ik . Then
{f I | #I = k} is a basis for k E .
Exercise 4.6. If E1 , E2 are two finite dimensional vector space, show that
(E1 E2 ) = E1 E2
(tensor product of graded algebras).
There are two important operations on E called exterior multiplication and contraction. Exterior multiplication is just the algebra homomorphism
: E End(E )
for the graded commutative algebra E . Thus ()() = . For v E, one defines
the contraction operator
v : k E k1 E
by v ()(v2 , . . . , vk ) = (v, v2 , . . . , vk ). Sometimes we write v = (v).
Thus v End1 (E ). Clearly v v = 0. Hence also
[v , w ] = v+w v+w = 0.
If f1 , . . . , fk are vectors in E ,
(3)

v (f1 fk ) =

X
(1)j+1 fj (v)f1 fbj fk .
j

This follows from the definition of the wedge product, or equivalently by expanding the
determinant on the right hand side of (2) at the first column.
Theorem 4.7. The contraction operator v is a super-derivation of degree 1 of
E . One has
[v , ()] = (v ).
Proof. Let f1 , . . . , fr be vectors in E . From (3) we read off that for any l r,
v (f1 fr ) = v (f1 fl ) (fl fr ) + (1)l (f1 fl ) v (fl fr ).
By linearity, it follows that
v ( ) = v + (1)l v
for l E and E , proving that v is a super-derivation. The identity [v , ()] =
(v ) is just re-phrasing the same condition.

For any linear map A : F E the dual map A : E F extends uniquely to an
algebra homomorphism A : E F . Thus if f1 , . . . , fk E ,
A (f1 fk ) = (A f1 A fk ).

4. DIFFERENTIAL FORMS

27

In particular, any endomorphism of E gives rise to an endomorphism of E . For any

1-parameter group At = exp(tL) of automorphisms, we obtain a 1-parameter group
At = exp(tL) of automorphisms of E . It follows that L : E E extends to
a derivation DL of degree 0 of E . One has
X
DL (f1 fk ) =
(1)j+1 L (fj )f1 fbj fk .
j

(Dont confuse DL with the algebra homomorphism (L) : E E !).

Exercise 4.8. a) Show that the map End(E) Der0 (E ), L 7 DL is a Lie

algebra homomorphism.
b) For v E, L End(E), show that [DL , v ] = L(v) . (Hint: Since both sides are
derivations, it suffices to check on generators.)
After this lengthy discussion of linear algebra, let us return to manifolds.
4.3. Differential forms.
Definition 4.9. For any manifold M and any x M , the dual space Tx M :=
(Tx M ) of the tangent space at x is called the cotangent space at x. The elements of
Tx M are called covectors. Elements of k T M are called k-forms on Tx M .
Any function f C (M ) determines a covector dx f Tx M , called its exterior
differential at x, by
(df )x (v) := v(f ), v Tx M.
If U Rm is an open subset with coordinates x1 , . . . , xm (viewed as functions on U ), we
can thus define 1-forms
(dxi )x Tx U.


This 1-forms are the dual basis to the basis xi  of Tx U : Indeed,
x





i 
=
(x
)
= ij .
(dxi )x
x
xj x
xj

A differential k-form on M is a family of k-forms on tangent spaces Tx M depending

smoothly on the base point, in the following sense:
Definition 4.10. A differential k-form on M is a C (M )-linear map
: X(M ) . . . X(M ) C (M )
|
{z
}
k times

that is anti-symmetric and C (M )-linear in each argument. The space of k-forms is

denoted k (M ), the direct sum of all k (M ) is denoted (M ). In particular, 0 (M ) =
C (M ).

28

CONTENTS

For any x M there is a natural evaluation map k (M ) k Tx M, 7 x such

that
(X1 , . . . , Xk )|x = x ((X1 )x , . . . , (Xk )x ).
Similarly, if U M is open, one can define the restriction |U k (U ) such that
(X1 , . . . , Xk )|U = |U ((X1 )|U , . . . , (Xk )|U ).
If U, V are two open subsets, and U , V are forms on U and V with U |U V = V |U V
then there is a unique form U V (U V ) restricting to U and V , respectively.
One defines the wedge product on (M ) by anti-symmetrization, similar to the wedge
product on E . It is uniquely defined by requiring that all the evaluation maps (M )
Tx M are algebra homomorphisms. Wedge product makes (M ) into a graded supercommutative super-algebra. For X X(M ), one denotes by X = (X) Der1 ((M ))
the operator of contraction by X, and for (M ) one denotes by () the operator
of left multiplication by . As before, we have [X , Y ] = 0 and [X , ()] = (X ).
4.4. The exterior differential. The magic fact about the algebra of differential
forms is the existence of a canonical derivation d of degree 1. For any function f
C (M ) one defines df 1 (M ) by the equation,
(df )(X) = X(f ).
It has the property d(f g) = f dg + gdf . In particular, if U Rm is an open subset, we
have 1-forms dx1 , . . . , dxm 1 (U ). Writing dxI = dxi1 dxik for any k-element
subset i1 < . . . < ik , the most general k-form on U reads
X
=
I dxI
I

where C (U ) are recovered as

,
.
.
.
,
).
xi1
xik
Theorem 4.11. The map d : 0 (M ) 1 (M ), f 7 df extends uniquely to a
super-derivation of degree 1 on (M ), called exterior differential, with d(df ) = 0. The
exterior differential has property
d d = 0.
I = (

Proof. It suffices to prove existence and uniqueness for the restrictions to elements
of an open cover (U )A . Indeed, once we know that d|U exist and are unique, then
the forms d|U agree on overlaps U U by uniqueness, so they define a global form
d. In particular, we may take (U )A to be a covering by coordinate charts. This
reduces the problem to open subsets U Rm .
The derivation property of d forces us to define d : k (U ) k+1 (U ) as follows,
X
X
X X I
dxj dxI .
d(
I dxI ) =
dI dxI =
j
x
j
I
I
I

4. DIFFERENTIAL FORMS

We have

29

X
X X 2I
d2 (
I dxI ) =
dxk dxj dxI = 0,
k xj
x
I
I
jk

by equality of mixed partials. It remains to check d is a derivation. For = f dxI k

and = gdxJ l we have
d( ) = d(f dxI gdxJ )
= d(f g) dxI dxJ
= (f dg + gdf ) dxI dxJ
= df dxI gdxJ + (1)k f dxI dg dxJ
= (d) + (1)k d.

Let us write out the exterior differential on M = R3 , with coordinates x, y, z. On
functions,
f
f
f
dx +
dy +
dz,
df =
x
y
z
so d is essentially the gradient. On 1-forms = f dx + gdy + hdz,
g f

h

h g

f
d =

dx dy +
dy dz +
dz dx,
x y
y z
z
x
so d is essentially the curl, and on 2-forms = f dy dz + gdz dx + hdx dy,
g h

f
+
+
dx dy dz,
d =
x y z

so d is essentially the divergence. In general, one should think of d as the proper abstract
setting for grad, curl, div.
4.5. Functoriality. Any smooth map F C (N, M ) between manifolds gives rise
to an algebra homomorphism
F : (M ) (N )
such that (F )x = F (x ).
Theorem 4.12. F commutes with exterior differential:
F d = d F .
Proof. The algebra (M ) is generated by all functions f C (M ) together with
all differentials df . That is, any differential form can be written as a finite linear combination of expressions
f0 df1 dfk ,

30

CONTENTS

and finally by linearity it holds everywhere. This follows by choosing a finite atlas of M
and a subordinate partition of unity. 3 Hence, it suffices to check the the identity on
functions f and differentials df .
For f C (M ), and any x M, v Tx M , we have
(F (df ))x (v) = (df )F (x) (dx F (v)) = (dx F (v))(f ) = v(F f ) = d(F f )x (v).
Thus F d = d F on functions. On differentials df :
F (d(df )) = 0, d(F (df ) = d(d(F f )) = 0.

For instance, if U Rm and V Rn are open subsets with coordinates xi , y j , and
F C (U, V ), with components F j = F y j , we have
X F j
F dy j = dF y j = dF j =
dxi .
i
x
i
If dim U = dim V = m, we obtain in particular
F j
) (dx1 dxm ).
xi
Suppose X X(M ) is a complete vector field, with flow t . Then we have a 1parameter group of automorphisms t of (M ). We define the Lie derivative to be its
generator:
d
LX = |t=0 t .
dt
The definition extends to incomplete vector fields: If U is any open subset with compact
closure, the map t : U M exists for |t| < , and the equation defines LX |U .
We have now defined three kinds of derivations of (M ): The contraction X , the
Lie derivative LX , and the exterior differential d. Recall that the graded commutator of
two derivations is again a derivation.
F (dy 1 dy m ) = det(

Theorem 4.13. One has the following identities:

[d, d]
[LX , d]
[X , d]
[X , Y ]
[LX , LY ]
[LX , Y ]
3We

=
=
=
=
=
=

0,
0,
LX ,
0,
L[X,Y ] ,
[X,Y ] .

mentioned earlier that any manifold admits a finite atlas, although we never proved this. The
following proof can be easily modified to close this gap: it is enough to check the identity on functions
supported in coordinate charts, where the fi for i > 0 can be taken as coordinate functions.

4. DIFFERENTIAL FORMS

31

Proof. The first equation is just [d, d] = 2d2 = 0. Since (tX ) and d commute,
[LX , d] = 0 by definition of LX . The other identities can be checked on generators f, df
of (M ) (using that two derivations are equal if and only if they agree on generators).
For instance, the third equation is verified by the calculations,
[X , d](f ) = X df = (df )(X) = X(f ) = LX f
and
[X , d](df ) = X ddf + dX df = dLX f = LX df.
The fourth equation is obvious since both sides vanish on generators. The fifth equation,
on functions f , is just the definition of the Lie bracket, and on df follows since Lie
derivatives and d commute. In the last equation, both sides vanish on functions, and on
df we have
[LX , Y ](df ) = [[LX , Y ], d](f ) = [LX , [Y , d]](f ) = [LX , LY ](f ) = L[X,Y ] (f ) = [X,Y ] df.
(Alternatively, the last equation follows by direct application of the definition of the Lie
derivative.)

These identities are of fundamental importance in the Cartans calculus of differential
forms. It is somewhat remarkable that contractions, Lie derivatives and d form a graded
Lie subalgebra of the graded Lie algebra of derivations!
Proposition 4.14. For any k (M ) and any X0 , . . . , Xk X(M ),
X
cj , . . . , Xk ))
(d)(X0 , . . . , Xk ) =
(1)j Xj ((X0 , . . . , X
j

X
i<j

ci , . . . , X
cj , . . . , Xk )
(1)i+j ([Xi , Xj ], X0 , . . . , X

Proof. We give the proof for k = 2, which will show the relevant pattern. The left
hand side is X2 X1 X0 d. Using the commutation relations between d and contractions,
we find
X2 X1 X0 d = X2 X1 LX0 X2 X1 dX0
= X2 X1 LX0 X2 LX1 X0 + LX2 X1 X0 .
The desired identity now follows by permuting the Lie derivatives in each term all the
way to the left. For example, the first term gives
X2 X1 LX0 = X2 LX0 X1 X2 [X0 ,X1 ]
= LX0 X2 X1 [X0 ,X2 ] X1 X2 [X0 ,X1 ] .

For example, if is a 1-form,
d(X0 , X1 ) = X0 ((X1 )) X1 ((X0 )) ([X0 , X1 ]).

32

CONTENTS

P
Exercise 4.15. Let U be an open subset of Rm , and X =
Xi x i a vector field.
Let = dx1 dxm be the standard volume form. Show that
X Xi
LX = (
) .
i
x
i
Conclude
P Xi that the flow of X is volume preserving if and only if the divergence div (X) =
( i xi ) vanishes everywhere.

4.6. Orientation of manifolds. A diffeomorphism F : U V between open

subsets of Rm is called orientation preserving if for all x U , the Jacobian Dx F has
positive determinant.
An atlas (U , ) for a manifold M is said to be oriented if all transition functions
1
are orientation preserving. If such an atlas exists, M is called orientable.

An orientable manifold M with an oriented atlas is called an oriented manifold. A

Map F C (N, M ) between oriented manifolds of the same dimension are orientation
preserving if its expressions in oriented charts are orientation preserving.
Orientability of manifolds M is closely related to the existence of volume forms. A
volume form is an m-form m (M ), with x 6= 0 for all x M .
Theorem 4.16. A manifold M is orientable if and only if it admits a volume form.
Any volume form on a manifold M defines a unique orientation, with the property
that for all oriented charts (U , ),
(dx1 dxn ) = f |U
where f > 0 everywhere. Two volume forms , define the same orientation if and
only if = f with f > 0.
Proof. Suppose M is orientable, and let (U , )A be an oriented atlas. Choose a
partition of unity subordinate to the cover (U )A . The form = (dx1 dxm )
is a volume form on U . Set
X
=
.

Then is a volume form on M . Indeed, if v1 . . . , vm is an oriented basis of Tm M ,

each of the forms ( )x with x U takes positive values on v1 . . . , vm . Hence so does
P

(x) ( )x .

Exercise 4.17 (Oriented double cover). Let M be a connected manifold. The tangent space at any point x M has exactly two orientations. The choice of an orientation
M be the map with
on x also induces orientations on nearby points. Let : M
has a natural structure of
fibers 1 (x) the two possible orientations on x. Show that M
an oriented manifold, with a local diffeomorphism. Show that M is orientable if and
is disconnected, and the choice of an orientation is equivalent to the choice of
only if M
. One calls M
the oriented double cover of M . What is the oriented
a component of M
double cover for RP (2)? For the Klein bottle?

4. DIFFERENTIAL FORMS

33

If M, N are two oriented manifolds, with orientations defined by volume forms

M , N , the direct product M N carries an orientation defined by the volume form

M
M N
N , where M , N are the projections from M N to the two factors.
Exercise 4.18. Let F : N M be a smooth map between oriented manifolds, and
suppose a M is a regular value. Then any choice of volume forms N , M defines a
volume form S on the level set S = F 1 (a). In particular, S is oriented.
Suppose M is an oriented manifold with boundary M . We define an orientation
on M as follows: Given x M , let v1 Tx M \Tx M be an outward pointing vector.
That is, v1 (f ) 0 for any function f C (M ) with f 0 on M . Call a basis v2 , . . . , vm
of Tx M oriented if v1 , . . . , vm is an oriented basis of Tx M .
4.7. Integration on manifolds. Let U Rm be open. The integral of a compactly
supported differential form k (U ) is defined to be 0 unless k = m. For
= f dx1 dxm m (U ),
with f a compactly supported smooth function on U , one defines
Z
Z
Z
= f (x1 , . . . , xm ) dx1 dx2 dxm
U

as a Riemann integral.
Recall the change of variables formula for the Riemann integral: If F : V U is a
diffeomorphism, then
Z
Z
Z
F i
= (F f )(y 1 , . . . , y m ) det( j )dy 1 dy 2 dy m
y
U
i

)dy 1 dy 2 dy m is just the pull-back under F of the form dx1 dx2 dxm ,
Since det( F
y j
this can be written in more compact form,
Z
Z
=
F .
U

F 1 (U )

This formula shows that integration is independent of the choice of coordinates, and is
used to extend integration to manifolds:
Theorem
R 4.19. Let M be an oriented manifold of dimension m. There is a unique
linear map M : comp (M ) R such that for all oriented charts (U, ) of M , and any
form with compact support in U ,
Z
Z
(4)
=
(1 ) .
M

(U )

Proof. Choose an atlas (U , ) and a partition of unity subordinate to the

cover {U }. For any compactly supported form on M , define
Z
XZ

:=
(1
) ( ).
M

(U )

34

CONTENTS

This has the required property (4), since if is supported in a chart (U, ), the change
of variables formula shows
Z
XZ
XZ
1
1
( ) ( ) =
( ) ( ) =
(1 ) .

(U )

(U )

(U )

Conversely, this equation also shows uniqueness of an integration map satisfying (4). 
We will often use the following notation: If M is any manifold, and S M is an
oriented embedded submanifold, and : S M the inclusion map, one defines
Z
Z
Z
: (M ) R,
:=
.
S

4.8. Integration over the fiber. Let M, B be manifolds of dimension m, b, and

let
: M B
be a submersion. We have proved that each fiber 1 (y) M of is a smooth embedded
submanifold of dimension f = m b. Moreover, for any x M there exists a local chart
(U, ) around x such, in the coordinates x1 , . . . , xm defined by , the map is just
projection onto the last b coordinates.
To define the operation integration over the fibers we need to assume that each
fiber 1 (y) is oriented, and that the orientation depends smoothly on y.
Definition 4.20. A fiberwise orientation for the submersion : M B is an
equivalence class of forms mb (M ) such that for each y B, the pull-back of
to the fiber 1 (y) is a volume form for 1 (y). Two such forms , are called
equivalent if their pull-backs to each fiber 1 (y) differ by a positive function on 1 (y).
For example, if M, B are oriented by volume forms M , B , one obtains a fiberwise
orientation by letting any mb-form with M = B . The form is not uniquely
defined by this property, but its pull-back to the fibers is.)
Theorem 4.21. Let M, B be manifolds of dimension m, b, and : M B be a
submersion with smoothly oriented fibers. There exists a unique linear map
: comp (M ) comp (B)
of degree b m with the following properties:
a) For all y B the diagram
comp (M )

/ comp ( 1 (y))
R

comp (B)


/ ({y}) = R

commutes. Here the horizontal maps are pull-back under the inclusion.

4. DIFFERENTIAL FORMS

35

b) For all comp (M ) and all (B),

= ( ).
The map is called integration over the fiber,
Proof. Since has degree f = (m b), it vanishes on forms of degree less
then f . On forms of degree f it is determined by property (a). In general, any form on
M can be written as a locally finite sum of f -forms on M , wedged with pull-backs of
forms on B. (This is easily seen locally, and the global statement follows by choosing a
partition of unity.) Hence, uniqueness is clear and it suffices to prove existence. Again,
by partition of unity it suffices to prove existence in charts. By the normal form theorem
for submersions, we can cover M by charts U , with image a direct product of open sets
(U ) = V W , such that becomes projection to the first f m coordinates. This
reduces the theorem to the product situation.

4.9. Stokes theorem. Integration can be extended to manifolds M with boundary.
Recall that manifolds with boundary are defined similar to manifolds, but taking the half
space x1 0 as the target space for the coordinate charts. The boundary M of M is
a manifold of dimension m 1. It inherits an orientation from a given orientation on
M , as follows: (...) Let x M , and v1 TX M an outward-pointing tangent vector at
x. A basis v2 , . . . , vn a basis of Tx (M ) is oriented if and only if v1 , . . . , vn is oriented.
That is, in local coordinates identifying U with an open subset of {x1 0} and U with
an open subset of the subspace {x1 = 0}, the orientation is given by the volume form
dx2 . . . dxm .
Theorem 4.22 (Stokes). If M is an oriented manifold with boundary, and is a
compactly supported form, then
Z
Z
d =
.
M

Proof. We may assume that the support of is contained in some coordinate chart
(U, ), where (U ) is an open subset of the half space {x1 0}. Write
=

m
X
i=1

ci dxm
fi dx1 dx

where each fi has compact support. Then

d =

m
X
i=1

(1)i+1

fi 1
dx dxm .
xi

Using Fubinis theorem, we can change the order of integration, and in the ith term
integrate over the xi -variable first. Thus all terms, except the term for i = 1, vanish.

36

CONTENTS

We find

Z

f1 1 
dx dx2 dxm
1
x

Z
Z
2
m
=
f1 dx dx =
.

d =


One can view Stokes theorem as a geometric version of Greens formula from calculus.
Two maps F0 , F1 C (S, M ) are called (smoothly) homotopic if there exists a map
F C ([0, 1] S, M ) such that F (0, ) = F0 and F (1, ).
Corollary 4.23. Suppose S is an oriented manifold, and F0 , F1 : S M two
smooth maps which are smoothly homotopic. For any closed form on M ,
Z
Z

F1 =
F0 .
S

Proof. Using Stokes theorem,

Z
Z

F1 F0 =
S

[0,1]S

d(Ft )

[0,1]S

Ft d = 0.


The following generalizes Stokes theorem to submersions.

Theorem 4.24. If : M B is a submersion between oriented manifolds (without
boundary),
(1)mb d = d.
More generally, if M has a boundary, and := |M is also a submersion, then
(1)mb d( ) = (d) () .
4.10. Homotopy operators. Suppose F0 , F1 C (N, M ) are called (smoothly)
homotopic. Let
h := F : (M ) (N )
be the operator of degree 1, defined as a composition of pull-back by F and integration
over fibers of : [0, 1] N N .
Theorem 4.25. The map h is a homotopy operator between F0 , F1 . That is,
h d + d h = F1 F0 : (M ) (N ).
Proof. Let 0 , 1 denote the inclusion of the two boundary components N {0}, N
{1}. Since the fibers of are 1-dimensional, we have
d + d = () = 1 0 .

4. DIFFERENTIAL FORMS

37

Therefore
F d + d F
( d + d )F
() F
(F 1 ) (F 0 )
F1 F0 .

hd + dh =
=
=
=
=


Corollary 4.26 (Poincare Lemma). Suppose U Rm is an open ball of radius R
around 0, possibly R = . If k (U ), with k > 0, is closed (i.e. d = 0) then is
exact, i.e. = d for some .
Proof. Let F : [0, 1] U U be the map F (t, x) = tx. Then F1 is the identity
map, and F0 is the constant map taking everything to the orgin. Thus F0 = 0 since
we assume the degree of is positive, and of course F1 = . Let h be the homotopy
operator for F . Then
= F1 F0 = dh + hd = dh = d
where = h.

Note that the homotopy operator provides an explicit primitive for the closed form
.
Examples 4.27. Let h be the homotopy operator for Rm , corresponding to the
retraction F (t, x) = tx. Consider the volume form = dx1 dxm on Rm . Then
X
X
dj dxm + . . .
(1)j+1 tm1 dt
(1)j xj dx1 dx
F =
j

where . . . denotes terms not involving dt. Thus

1 X
dj dxm
h =
(1)j+1 xj dx1 dx
m j

The right hand sides is proportional to the volume form on S m1 . More precisely, one
has
h = c rm
where r = ||x|| and c is a certainP
constant.
Next, consider a 1-form = i fi dxi . Then
X
F =
xi fi (tx)dt + . . .
i

so that
h =

X
i

fi (tx)dt.

38

CONTENTS

If is closed, one checks that dh = .

P
Exercise 4.28. Let E = j xj x j be the Euler vector field on Rm . Show that if
P
is a closed k-form = I I dxI where the coefficients are polynomials of degree l, we
have
1
(E).
h =
k+l
5. De Rham cohomology
5.1. Definition. Let M be a manifold. The exterior differential d on (M ) gives
us a complex
d
d
d
0 0 (M ) 1 (M ) m (M ) 0.
called the de Rham complex. The subspace of closed forms
Z k (M ) = { k (M )| d = 0}
is called the space of cocycles, and the space of exact forms
B k (M ) = d(k1 (M )) k (M )
L
Lm
k
k
is the space of coboundaries. One writes B(M ) = m
k=0 B (M ) and Z(M ) =
k=0 Z (M ).
Since d squares to zero, every k-coboundary is a k-cocycle. The converse is not true in
general, and the obstruction is measured by the cohomology
H k (M ) = Z k (M )/B k (M )
L
Lm
k
k
We write Z(M ) = m
k=0 Z (M ) and B(M ) =
k=0 B (M ) and call
H(M ) =

m
M

H k (M )

k=0

the the de Rham cohomology of M . The numbers

bk = dim H k (M )
are called the Betti numbers of the manifold M . (We will see later that bk < if M is
compact.) The polynomial
X
p(t) =
bk tk
k

is called the Poincare polynomial of M . Note that Z(M ) is a graded algebra under wedge
product. Moreover, B(M ) is an ideal in Z(M ), since (d) = d( ) for (M )
and Z(M ). Thus H(M ) becomes a super-commutative graded algebra.
We have seen that on Rm (and more generally, on star-shaped open subset fo Rm ),
every closed form of degree k > 0 is also exact. Thus

R for k = 0,
k
m
H (R ) =
.
0 otherwise

5. DE RHAM COHOMOLOGY

39

Examples 5.1.
(a) Since B 0 (M ) = 0, the zeroth cohomology space is just
0
0
H (M ) = Z (M ). A function f 0 (M ) is a cocycle if and only is f is
constant on connected components of M . Thus Z 0 (M ) is the space of locally
constant functions, and b0 is simply the number of connected components.
(b) Suppose M is a manifold of dimension m, d vanishes on m-forms since there are
no m + 1-forms. Thus
Z m (M ) = m (M ).
When is such a form exact? Suppose M is connected, orientable and
R compact,
and let M be a volume form on M defining the orientation.
R
R Then M > 0. If
= d for some m1-form , then we would have M = M d = 0 by Stokes
theorem. This contradiction shows that represents a non-trivial cohomology
class in H m (M ). We will show later that indeed, if M is compact, orientable
and connected H m (M ) is 1-dimensional, and is spanned by the class of . If M
is non-compact or non-orientable this is usually false.
(c) Let us try toR get a feeling for H 1 (M ). If 1 (M ) is exact, = df then
the integral S 1 along any closed path : S 1 M vanishes, by Stokes
theorem:
Z
Z
Z

=
df =
d f = 0.
1
1
1
S
S
S
R
1
The converse is true as well: If (M ) is such that S 1 is always 0, then
the integral of along a path : [0, 1] M depends only on the endR points.
One can then define a function f by fixing x0 M and setting f (x) = [0,1]
for any path from x0 to x. It is not hard to see that the function f obtained in
this way is smooth and satisfies df = . (Indeed, in a coordinate chart around
x0 , the function f is just the image under the homotopy operator from Poincares
Lemma.)
Similarly, is closed if and only if the integral along closed paths does not
change under smooth homotopies of the path. We has already seen one direction
of this statement. The opposite direction can be seen as follows: Suppose is
not closed. then there exists x0 M with d 6= 0 at x0 . Choose X X(M ) such
that (X)d 6= 0 near x0 , and a loop 0 never tangent to X. Using the flow of X
1
(with > 0 sufficiently
R small), we can construct a homotopy : [0, 1] S M
of this loop. Then [0,1]S 1 d 6= 0, showing by Stokes theorem that the
integral of along t changes.
Exercise 5.2. Show more generally that a k-form is closed if and only if for all
maps F : S k M , the integral of F depends only on the homotopy class of F .
5.2. Homotopy invariance. For any F C (N, M ), the pull-back map F :
(M ) (N ) is a homomorphism of graded algebras commuting with d. Hence it
defines an algebra homomorphism F : H(M ) H(N ).

40

CONTENTS

Theorem 5.3 (Homotopy invariance). If F0 , F1 C (N, M ) are smoothly homotopic, then the induced maps F0 , F1 : H(M ) H(N ) are equal.
Proof. We have to show that the map F1 F0 : (M ) (N ) induces the zero
map in cohomology. Thus, we have to show that if Z(M ) then F1 F0 B(N ).
Let h : (N ) (M ) be the homotopy operator for a smooth homotopy between
F0 , F1 :
F1 F0 = h d + d h : (M ) (N ).
Apply this equation to , using d = 0:
F1 F0 = d(h).

Two maps F : N M and G : M N are called homotopy inverses if F G is
homotopic to IdM and G F homotopic to IdN . In this case, the theorem shows:
Corollary 5.4. Suppose F : N M and G : M N are homotopy inverses.
Then F : H(M ) H(N ) is an isomorphism with inverse G : H(N ) H(M ).
Proof. According to the theorem, the maps G F : H(M ) H(M ) and F G :
H(N ) H(N ) are the identity maps.

Definition 5.5. Let N be a submanifold of a manifold M . A smooth map F :
[0, 1] M M such that F (0, ) = IdM and F (1, x) N for all x M is called a
deformation retraction from M onto N . It is called a strong deformation retraction if
F (t, x) = x for all t [0, 1] and x N .
As a special case, one has:
Proposition 5.6. If M admits a strong deformation retraction onto an embedded
submanifold N , then the inclusion map : N M gives an isomorphism : H(M )
H(N ).
Proof. Let F : [0, 1]M M be a strong deformation retraction. Let : M N
be the map thus obtained, i.e. = F (1, ) : M N . Then F gives a homotopy
between IdM and . But = IdN . Thus and are homotopy inverses; in particular
is an isomorphism in cohomology.

This Proposition once again explains Poincares Lemma: Since Rm admits a strong
deformation retraction onto a point, its cohomology is trivial.
6. Mayer-Vietoris
6.1. Exact sequences. A (finite or infinite) sequence of vector spaces and maps
E E E

6. MAYER-VIETORIS

41

is called exact at E if the kernel of the map E E is equal to the image of the map
E E . The sequence is called exact if it is exact everywhere. An exact sequence of
the form

0 F E G 0
is called short exact. This has the following interpretation: Exactness at F means that
the map : F E is injective. That is, we can think of F as a subspace of E. Exactness
at G means that the map : E G is surjective. Thus we may think of G as a quotient
space G = E/ ker(). Exactness at E tells us that ker() = im()
= F . Putting all this
together, we conclude that F is a subspace of E and G = E/F .
Exercise 6.1. Let A : F E be any linear map. Show that the sequence
A

0 ker(A) F E E/ im(A) 0
is exact.
Suppose the sequence

0 E E E
is exact. Then E can be thought of as a subspace of E , and the map k : E E
vanishes exactly on E E . That is, k descends to an injective map E /E E . Thus
we obtain a new exact sequence
k

0 E /E E
Exercise 6.2. Show that if the sequence 0 E 0 E k 0 is exact, then
the alternating sum of dimensions vanishes:
X
(1)i dim E i = 0.
i

6.2. Differential complexes. The de Rham complex ((M ), d) is an example of

a differential complex. Since we will encounter many more examples of differential complexes later on,
L it is useful to treat them in generality.
Let E = kZ E k be a graded vector space, and d : E E a linear map of degree
1. That is, d is a collection of maps
(5)

E k E k+1 E k+2

One calls (E, d) a differential complex if d squares to 0. The kernel of the map d :
E k E k+1 is called the space of k-cocycles, and is denoted Z k (E). The image of the
map d : E k1 E k is called the space of k-coboundaries and is denoted B k (E). Since
d d = 0, B k (E) is a subspace of Z k (E). The quotient space
H k (E) = Z k (E)/B k (E)
is called the kth cohomology of the complex (E, d). We denote H(E) =
Note that H(E) = 0 if and only if the sequence (5) is exact.

kZ

H k (E).

42

CONTENTS

(In the special case E = (M ), we will continue to write B(M ), Z(M ), H(M ) rather
than B((M )), Z((M ), H((M )).
Definition 6.3. Let (E, d) and (E , d) be two differential complexes. A linear map
: E E of degree 0 is called a homomorphism of differential complexes or cochain
map, if d = d .
A cochain map : E E takes cocycles to cocycles, hence it gives a map in
cohomology, which we denote by the same letter:
: H(E) H(E ).
For example, if F : M N is a smooth map of manifolds, it defines a cochain map
= F : (N ) (M ), and therefore H(N ) H(M ).
Definition 6.4. A homotopy operator (or chain homotopy) between two cochain
maps 0 , 1 : E E is a linear map h : E E of degree 1, with property
h d + d h = 1 0 .
If such a map h exists, 0 , 1 are called chain homotopic. Two cochain maps : E E
and : E E are called homotopy inverse if : F F and : E E are
both homotopic to the identity, if such maps exist, E, E are called homotopy equivalent.
For example, if F0 , F1 : N M are smoothly homotopic, we had constructed a
homotopy operator between 1 = F1 and 0 = F0 . As in this special case, we see:
Proposition 6.5. Chain homotopic cochain maps induce equal maps in cohomology.
If two differential complexes E, E are homotopy equivalent, their cohomologies are equal.
Suppose now that E, F, G are differential complexes, and let
j

0 E F G 0
be a short exact sequence where all maps are homomorphisms of differential complexes.
We will construct a map : H(G) H(E) of degree +1 called the connecting homomorphism. This is done as follows: Let [] H k (G) be a given cohomology class,
represented by Z k (G). Since k : F G is surjective, we can choose, F k with
k() = . Then
k(d) = d(k()) = d = 0.
Thus, d ker(k) = im(j). Since j is injective, there is a unique E k+1 with
j() = d. Then j(d) = dj() = dd = 0, so d = 0. Set
([]) = [].
One has to check that this definition does not depend on the choices of and . For
example, if F k is another choice with k( ) = , then k( ) = 0, so
ker(k) = im(j). Since j is injective, there is a unique element E k with j() = .
Thus if E k+1 is the unique element such that j( ) = d , we have = d,

6. MAYER-VIETORIS

43

so [ ] = []. Similarly, one checks that the definition does not depend on the choice of
representative for [].
Theorem 6.6. Let

0 E F G 0
be a short exact sequence of homomorphisms of differential complexes with connecting
homomorphism . Then there is a long exact sequence in cohomology,
j

H k (E) H k (F ) H k (G) H k+1 (E)

where the connecting homomorphism : H k (G) H k+1 (E) is defined as follows: For
Z k (G), ([]) = [] where Z k+1 (E) is an element with j() = d for k() = .
Proof. Let us check exactness at H k (G). We have to check two inclusions, (i)im(k)
ker() and (ii) ker() im(k).
(i) Suppose Z k (F ). We have to show that [k()] = 0. By construction of the
connecting homomorphism, [k()] = [] where j() = d = 0. Thus = 0.
(ii) Suppose Z k (G) with [] = 0. We will show that there exists a closed element
Z k (F ) such that = k(). Start by choosing any F k with k() = , and let
E k+1 be the unique element with j() = d. By definition of , 0 = [] = [], thus
= d for some E k . Define = j(). Then is closed:
d = d j(d) = d j() = 0,
and k( ) = k() = since k j = 0. Exactness at H k (F ) and H k (E) is checked
similarly.

Theorem 6.7 (Mayer-Vietoris). Suppose M = U V where U, V are open, and let
U , V be a partition of unity subordinate to the cover of U, V . The sequence
j

0 p (U V ) p (U ) p (V ) p (U V ) 0
where j() = (|U , |V ) and k(, ) = |U V |U V is exact. It induces a long exact
sequence

0 H p (U V ) H p (U )H p (V ) H p (U V ) H p+1 (U V ) 0
The connecting homomorphism takes the cohomology class of Z p (U V ) to the class
of () Z p+1 (U V ), where
()|U = d(V ), ()|V = d(U ).
The formula for () makes sense, since on the overlap U V ,
d(V ) (d(U )) = d((U + V )) = d = 0.
Proof. Exactness at p (M ) is obvious since the map p (M ) p (U ) p (V ) is
clearly injective. The kernel of the map p (U ) p (V ) p (U V ) consists of forms
, with |U V = |U V . These are exactly the forms which patch together to a global
form on M , with = |U and = |V . This shows exactness at p (U ) p (V ). It

44

CONTENTS

remains to check surjectivity of the last map : p (U ) p (V ) p (U V ). Let

p (U V ). Choose a partition of unity U , V subordinate to the cover U, V . Then
V is zero in a neighborhood of U \U V , hence it extends by 0 to a form on U .
Similarly U is zero in a neighborhood of V \U V , hence it extends by 0 to a form
on V .
On U V we have |U V |U V = . This shows that k is surjective. Any short
exact sequence of chain complexes induces a long exact sequence in cohomology.

As a first application of the Mayer-Vietoris sequence, we can now compute the cohomology of the sphere.
Example 6.8. Let M = S m , and U, V the covering by the complements of the
south pole and north pole, respectively. The intersection U V smoothly retracts onto
the equatorial S m1 . Since H k (U ) = H k (V ) = 0 for k > 0, Mayer-Vietoris gives
isomorphisms
H k (S n )
= H k1 (U V ) = H k1 (S n1 )
for k > 1. By induction, we get H k (S n ) = H 1 (S nk+1 ). In low degrees, if n > 1,
Mayer-Vietoris is an exact sequence
0 R R R R H 1 (S n ) 0.
Here the map R R R is easily seen to be surjective, so R H 1 (S n ) must be the
zero map, so H 1 (S n ) = 0. For n = 1 we have H 1 (S 1 ) = 0. We conclude,

R for k = 0, n,
k
n
H (S ) =
.
0 otherwise
Thus p(t) = 1 + tn is the Poincare polynomial.
7. Compactly supported cohomology
If M is non-compact, one can also study the compactly supported de Rham cohomology: Working with the complex of compactly supported forms
d

0 0comp (M ) 1comp (M ) m
comp (M ) 0
p
one defines Hcomp
(M ) as the cohomology of this complex,
p
Hcomp
(M ) = H p (m
comp (M )).

If M is compact, Hcomp (M ) agrees with the usual cohomology, but for non-compact
manifolds it is quite different. For instance, if M has no compact component,
0
Hcomp
(M ) = 0.
0
This follows since Zcomp
(M ) consists of locally constant functions of compact support,
and there are no such if M has no compact component. It is also different in higher
degree: For instance,

(6)

1
(R) = R
Hcomp

7. COMPACTLY SUPPORTED COHOMOLOGY

45

in contrast to H 1 (R) = 0. To see (6), let 1comp (R). Write = f dt, where t is

the coordinate on R, and f Ccomp

(R). Suppose = dF . Then, by the fundamental
Rt
theorem of calculus, F (t) = f (s)ds + c, where c is a constant. Let T > 0 so that
R
supp(f ) [T, T ]. Then F (t) is equal to F () = c + for t > T and equal to
F () = c for t R< T . This shows that = dF for a compactly supported function
F , if and only if R = 0. If we take f to be a bump function, i.e. a compactly
supported function of integral 1, it follows that the class of 0 = f dt 1comp (R)
1
1
generates Hcomp
(R). Indeed, [0 ] Hcomp
(R) is non-zero, and for any 1comp (R), we
1
have c0 Bcomp
(R) (thus [] = c[0 ]) since its integral is zero.
This example shows that in general, the map
k
Hcomp
(M ) H k (M ).
k
induced by the inclusion Zcomp
(M ) Z k (M ) is neither injective nor surjective.

Theorem 7.1. Let M be a compact manifold, and x M . Then

k
Hcomp
(M \{x}) = H k (M )

for all k > 0.

Proof. Let k (M ) be closed. On any contractible open neighborhood U of x,
|U = d for some form (U ). Choose a function C (M ) supported on U with
= 1 near x. Then d() is cohomologous to and vanishes near x. This shows
k
that the natural map Hcomp
(M \{x}) H k (M ) is surjective. Let us now show that it is
k
injective. Suppose comp (M \{x}) is closed as a form on M , that is = d where
(M ). We have to show that can be chosen to be 0 near x.
Let U be a contractible open neighborhood of x such that vanishes near x. Then
d = 0 on U , i.e. is closed on U . If k > 1, we can argue as before: Write |U = d
and replace by d(). This shows that is also the differential of a compactly
supported form. If k = 1, has degree 0 so this argument doesnt work. But in this
case, f = is just a function which equals a constant a = f (x) near x. We may replace
f by f a to produce a compactly supported function with differential .

Theorem 7.2.
k
Hcomp
(Rm )

R
0

for k = m,
otherwise

Proof. This follows from the cohomology of S m , since Rm is diffeomorphic to S m

minus a point.

Let us discuss some of the properties of Hcomp . If F C (M, N ) is a proper map,
pull-back F induces a chain map comp (N ) comp (M ) hence a map Hcomp (N )
Hcomp (M ). (A map is called proper if the pre-image of any compact set is compact.)
One proves as before that Hcomp is homotopy invariant, but only under proper homotopies. This is why, for example, the compactly supported cohomology of Rm is not
the cohomology of a point.

46

CONTENTS

Similarly, if U M is an open subset, one has a natural restriction map

(M ) (U ),
but this map does not take comp (M ) into comp (U ) (unless U is a connected component
of M , the inclusion map is not proper.) On the other hand, one has a natural extension
by 0 map
comp (U ) comp (M ),
but there is no such map from (U ) to (M ). As a consequence, the Mayer-Vietoris
sequence for comp () looks different from that for ().
Theorem 7.3 (Mayer-Vietoris for cohomology with compact support). Suppose M =
U V where U, V are open, and let U , V be a partition of unity subordinate to the cover
{U, V }. The sequence
j

0 pcomp (U V ) pcomp (U ) pcomp (V ) pcomp (U V ) 0,

where j() = (, ) and k(, ) = + , is exact for all p. Hence there is a long
exact sequence in compactly supported cohomology,

p
p
p
p
p+1
Hcomp
(U V ) Hcomp
(U ) Hcomp
(V ) Hcomp
(U V ) Hcomp
(U V ) .
p
p+1
The connecting homomorphism : Hcomp
(U V ) Hcomp
(U V ) takes the class of
p
Zcomp (U V ) to the class of

() := dU = dV .
Note that the formula for () is well-defined since dU = dV 1comp (U V ).
Proof. This is very similar to the proof of Mayer-Vietoris for (), and is left as an
exercise.

Again, the direct sum Hcomp (M ) is an algebra under wedge product. Since the wedge
product of a compactly supported form with any form is compactly supported, we see
that Hcomp (M ) is a module for the algebra H(M ).
8. Finite-dimensionality of de Rham cohomology
Using the Mayer-Vietoris sequence, we will now show that H(M ) is for any compact
manifold M , or more generally for all manifolds of so-called finite type.
Definition 8.1. A cover {U }A is called a good cover if all finite non-empty intersections of the sets U are diffeomorphic to Rm . A manifold admitting a finite good
cover is called of finite type.
Theorem 8.2. Every given cover of a manifold M has a refinement which is a good
cover.

9. POINCARE DUALITY

47

Sketch of proof. We wont give the full proof here, which requires some elements
from Riemannian geometry. The idea is as follows. Choose a Riemannian metric g on
M . That is, g is a symmetric C (M )-bilinear form X(M ) X(M ) C (M ), with
the property that g(X, X)x > 0 if Xx 6= 0. Using g one can define the length of curves
Rb
)
1/2 dt. An open set U M is called geodesically
: [a, b] M as length() = a g(,
convex, that is any two points in U are joined by a unique (up to re-parametrization)
curve : I U of shortest length.
Geodesically convex open sets are diffeomorphic to Rm , and the intersection of two
geodesically convex sets is again geodesically convex. Thus any cover consisting of
geodesically convex open sets is a good cover. It can be shown that for any point
x M , there exists > 0 such that the set U of points that can be joined by a curve of
length < is a geodesically convex open neighborhood of x. This shows that any cover
can be refined to a good cover.

In particular, every compact manifold is of finite type.
Theorem 8.3. For any manifold of finite type, the de Rham cohomology H(M ) and
the compactly supported de Rham cohomology Hcomp (M ) are finite dimensional.
Proof. Suppose U, V are open subsets of M , and suppose H(U ), H(V ), H(U V )
are all finite dimensional. Exactness of the Mayer-Vietoris sequence

H k1 (U V ) H k (U V ) H k (U ) H k (V )
shows
dim(H k (U V )) = dim(ker ) + dim(im )
= dim(im ) + dim(im )
dim H k1 (U V ) + dim H k (U ) + dim H k (V ) < .
Now let U1 , . . . , UN be a finite good cover ofSM . For each l N , the open subsets
U1 . . . Ul and (U1 . . . Ul ) Ul+1 = jl (Uj Ul+1 ) have a finite good cover
by l open sets. Hence, by induction on l they all have finite dimensional de Rham
cohomology, and by the Mayer-Vietoris argument it follows that U1 . . . Ul Ul+1
has finite dimensional de Rham cohomology. A similar argument applies to cohomology
with compact supports.

Note that the theorem is not true, in general, for manifolds of infinite type: For
instance, any manifold with an infinite number of components certainly has infinitedimensional H 0 .
9. Poincare duality
Let M be an oriented manifold. Recall that we have defined a linear map
Z
: comp (M ) R
M

48

CONTENTS

equal to zero on forms of degree k < dim M . Since the wedge product of a compactly
supported form and any form is compactly supported, we can use this to define a bilinear
form
Z
(M ) comp (M ) R, (, ) 7
.
M
R
Since M d = 0 (Stokes theorem), integration descends to a linear map
Z
: Hcomp (M ) R.
M

Thus we have a bilinear pairing

H(M ) Hcomp (M ) R, ([, ]) 7

Let P : H(M ) Hcomp (M ) be the map defined by

Z
P ([])([]) =
.
M

Note that P takes H (M ) to

mp
Hcomp
(M ).

The main theorem of this section is:

mp
Theorem 9.1 (Poincare duality). The map P : H p (M ) Hcomp
(M ) is an isomorphism for all p.

The proof is based on the following

Lemma 9.2. Let M = U V where U, V are open subsets of M . Then there is a
commutative diagram
H p (U V )

/ H p (U ) H p (V )

(1)p1 D


mp
Hcomp
(U V )

/ H p (U V )


mp
mp
/ Hcomp
(U ) Hcomp
(V )


mp
/ Hcomp
(U V )

/ H p+1 (U V )
(1)p D


mp1
/ Hcomp
(U V )

Here the upper horizontal map is the Mayer-Vietoris sequence in cohomology, and the
lower horizontal map is the dual of the Mayer-Vietoris sequence in compactly supported
cohomology.
Proof. Let Z(U V ) and (, ) Zcomp (U ) Zcomp (V ). We have
Z
Z
Z

|U +
|V =
( + )
U

U V

which shows commutativity of the first square. Commutativity of the second square is
obtained similarly. To prove commutativity of the third square, let U , V be a partition
mp1
of unity for U, V . Let Z p (U V ) and Zcomp
(U V ). We have to show
Z
Z
() = (1)p
().
U V

U V

9. POINCARE DUALITY

49

Recall that
()|U = d(V ), ()|V = d(U ).
and that
() = d(U ) = dU .
We have
(() )|U = d(V ) |U |U = (1)p |U dV |U .
Since dV |U is compactly supported in U V , we see that (() )|U is supported
in U V . Switching the roles of U, V , we find that () is supported in U V , and
compute
Z
Z
() = (1)p

U V

(),

U V

as desired.

Proof of Poincare duality. We give the argument for finite type manifolds;
the general case is covered in Greub-Halperin-Vanstone. As in the proof of finite dimensionality of H(M ), the proof is basedSon induction on the number of elements in a good
cover (Ui )N
i=1 . For l N , let U =
il Ui , V = Ul+1 . Then the induction hypothesis
applies to U, V, U V . Hence the corresponding Poincare duality maps are all isomorphisms. The following algebraic fact implies that H(U V ) Hcomp (U V ) is then
an isomorphism as well.

Lemma 9.3 (Five-Lemma). Let
E1

j1

/ E2

j2

F1

k1

j3

/ E3

/ F2

j4

k2

/ E4

/ F3

k3

/ E5

/ F4

k4


/ F5

be a commutative diagram of vector spaces and linear maps. Suppose the rows are exact,
and that 1 , 2 , 4 , 5 are isomorphisms. Then 3 is also an isomorphism.
Proof. We will only show injectivity of the map 3 . The proof of surjectivity is
very similar and is left as an exercise. Suppose x3 E3 is in the kernel of 3 . We have
to show x3 = 0. Since
4 (j3 (x3 )) = k3 (3 (x3 )) = 0,
and 4 is injective, we have j3 (x3 ) = 0. Thus x3 ker(j3 ) = im(j2 ). Thus we can write
x3 = j2 (x2 ). We have
k2 (2 (x2 )) = 3 (j2 (x2 )) = 3 (x3 ) = 0.
Thus 2 (x2 ) ker(k2 ) = im(k1 ). Thus we can write 2 (x2 ) = k1 (y1 ). Since 1 is
surjective, we can write y1 = 1 (x1 ). Then
2 (j1 (x1 )) = k1 (1 (x1 )) = k1 (y1 ) = 2 (x2 ).

50

CONTENTS

Since 2 is injective, this shows j1 (x1 ) = x2 . Thus x2 im(j1 ) = ker(j2 ). It follows that
x3 = j2 (x2 ) = 0.

Poincare duality has many applications. As a first consequence, we can finally get a
handle on the top degree cohomology.
m
Corollary 9.4. For every connected oriented manifold M , one has Hcomp
(M ) = R,
and

R
if M is compact,
m
H (M ) =
0 if M is non-compact.
m
If M is connected but non-orientable, one has H m (M ) = Hcomp
(M ) = 0.

Proof. For the case M oriented, all the facts are immediate by Poincare duality,
0
and using that Hcomp
(M ) = 0 if M is non-compact, and equal to R if M is compact. Note
that if M is compact, the non-trivial generator of H m (M ) is represented by a volume
be the oriented double cover.
form. Suppose now that M is non-orientable. Let M

Let : M M be the orientation-reversing diffeomorphism inducing the identity on

) = (M
)+ (M
) splits into the direct sum of M . The de-Rham complex (M
)+
invariant and anti-invariant forms. Both summands are invariant under d, and (M

is identified with (M ). Similarly H(M ) = H(M )+ H(M ) , where H(M )+ = H(M ).

) = 0. If M is compact,
If M is non-compact, we conclude H m (M ) = 0 since H m (M
). Since the volume form is anti-invariant, its class is in H m (M
) . Thus
then so is H(M
m
m
H (M ) = H (M )+ = 0.

Suppose M is a manifold, and S M a compact, Roriented, embedded submanifold of
dimension k. Integration over S defines a linear map S : H k (M ) R, in other words it
mp
is an element of H k (M ) . Let [ ] Hcomp
(M ) be the class defined by Poincare duality.
It is called the Poincare dual class to S. By definition, if Z k (M ), we have
Z
Z
=
.
S

One may think of as some kind of delta measure along S, keeping in mind however
that this equation only holds true for closed forms .
In fact, it is not important that S is an embedded submanifold: If S is any compact,
oriented manifold of dimension k m and C (S, M ), we can define a Poincare dual
mk
class [ ] Hcomp
(M ) by the condition
Z
Z

for all Z(M ). Since the left side depends only on the homotopy class of , it
follows that [ ] is invariant under homotopies of . Note furthermore that if U M is a
neighborhood of the image (S), the Poincare dual of S in U also serves as a Poincare dual
in M . In other words, one may choose the Poincare dual to be supported in arbitrary
small neighborhoods of (S).

9. POINCARE DUALITY

51

Example 9.5 (Poincare dual of a circle inside an annulus). The Poincare dual of
S 1 R2 inside the annulus U = {x R2 | 1/2 < |x| < 2} is given by the 1-form ,
written in polar coordinates as
= F (r)dr
where F (r) = 0 for r > 1 + and F (1) = 0 for r < 1 . To see this, we have to check
the defining equation for the Poincare dual. Let = f (r, )dr + g(r, )d be a closed
f
form on U . Since is closed,
= g
. We calculate, using integration by parts,
r
Z
Z
g(r, )F (r)d dr
=
2
2
R
R
Z
g(r, )F (r)dr d
=
2
Z R
Z

g
=
(g(r, )F (r))drd
(r, )F (r)drd
2 r
R2 r
Z
ZR
f
(r, )F (r)ddr
g(1, )d
=
1
R2
ZS
=
g(1 , )d
S1
Z
=

r=1
Z
=
.
r=1

Note that is exact in (U ) and also in comp (R2 ), since = dF .

Example 9.6 (Intersection number). Let M be a manifold, and S, S two compact
submanifolds manifolds of complementary dimension, dim S + dim S = dim M . Say
mp
p
dim S = p and dim S = m p. Let [ ] Hcomp
(S) and [ ] Hcomp
(S ) be the two
cohomology classes. The intersection number of S, S is defined as the integral,
Z

i(S, S ) :=
.
M

We will see later (?) that this is always an integer. Already at this stage, one can see
that i(S, S ) = 0 if S S = {0}. (Why?)
Example 9.7 (Linking number). Let M be any 3-manifold with H 1 (M ) = 0. (E.g.,
M simply connected.) Let , : S 1 M be two smooth loops with (S 1 ) (S 1 ) = .
Let U, U be disjoint open neighborhoods of the images of , . Let 2comp (U )
represent the Poincare dual of in U and 2comp (U ) represent the Poincare dual
2
of in U . Extend , by 0 to forms on M . Since Hcomp
(M ) = H 1 (M ) = 0, we can
write = d and = d , where , have compact (but usually not disjoint) support.

52

CONTENTS

One defines the linking number of , by the equation

Z

(7)
lk(, ) =
.
M

Integration by parts shows that the linking number is anti-symmetric in , . We claim

that lk(, ) is independent of all the choices involved in its definition. Suppose =
+ d, where 1comp (U ). Then the right hand side of (7) does not change, since
Z
Z
Z
d =
d =
=0
M

(since , have disjoint support.) Furthermore, the choice of U, U does not matter, since
the Poincare duals may be represented by a form supported in an arbitrarily small open
neighborhood of the images of , .
The linking number is invariant under smooth isotopies of , , provided the curves
remain disjoint:
Finally, note that lk(, ) = 0 if it is possible to contract one loop to a point while
keeping it disjoint from the other loop during the retraction.
Since , have disjoint support, it follows that is closed on the support of . That
is, we have
Z
Z
lk(, ) =

( ) =

S1

S1

It turns out that the linking number is always an integer. Let us prove this for the
special case : S 1 R3 is the unit circle in the x y plane. Let > 0 be small. After
applying some isotopy to in M \(S 1 ), we may assume that meets the region |z| <
in a number of line segments, and does not meet the region |z| < , |r 1| < . Let n+
(resp. n ) be the number of line segments in the region r < 1 going from z = to
z = , (resp from z = to z = ). We claim that
lk(, ) = n+ n .
Indeed, the Poincare dual of may be represented by a form
= dF dG = d(F dG)
where F = F (r) is equal to 1 for r < 1 and F (r) = 0 for r > 1 + , and G = G(z) is
equal to 1 for z > and equal to 0 for z < .
R Let = F dG. Note that is supported
the region r 1 + , |z| < . The integral S 1 ( ) is a sum of integrals over the line
segments described above. But = dG on the region U {|r 1| < }. Thus each
integral over a line segment is calculated by Stokes, and gives 1 depending on whether
the line segment travels from z = to z = or vice versa.
Exercise 9.8. Generalize the above concept of linking number to disjoint, immersed
compact submanifolds S, S of a manifold M , with dimensions dim S +dim S = dim M
1. What are the conditions on H(M ) so that the definition makes sense?

10. MAPPING DEGREE

53

10. Mapping degree

Let M, N be compact, connected, oriented manifolds of equal dimension m, and
F C (N, M ) a smooth map. Let M , N be volume forms defining the orientation,
normalized to have total
equal to 1. Thus [M ] H m (M ) corresponds to 1
R integral
under the isomorphism M : H m (M ) R. The mapping degree of F is defined to be
Z
deg(F ) =
F M .
N

Basic properties of the mapping degree, which follow immediately from the definition,
are: (1) deg(F ) Rdepends only on Rthe smooth homotopy class of F . (2) or any
m (M ), one has N F = deg(F ) M . (3) Under composition of maps, deg(F G) =
deg(F ) deg(G). (4) If F is a diffeomorphism, then deg(F ) = 1 if F preserves orientation
and deg(F ) = 1 if F reverses orientation.
Example 10.1. Let F : S m S m be the diffeomorphism x 7 x. The standard
volume form on S m transforms according to F = (1)m+1 . Thus F preserves
orientation if and only if m is odd. We conclude deg(F ) = (1)m+1 . Thus F cannot be
homotopic to the identity map if m is even.
We will now show that the mapping degree is always an integer, and also give an
alternative interpretation of the degree. Without proof we will use Sards theorem,
which implies that for every smooth map between compact manifolds, the set of regular
values is open and dense. (Recall that points that are not in the image are regular
values, according to our definition.) Thus let x M be a regular value of F . Since
dim M = dim N , the pre-image F 1 (x) is zero dimensional, i.e. is a finite collection of
points y1 , . . . , yd . For each i = 1, . . . , d, let i = 1, according to whether or not the
tangent map Tyi N Tx M preserves or reverses orientation.
Theorem 10.2. The mapping degree is given by the formula,
deg(F ) =

d
X

i ,

i=1

for any regular value x with pre-images y1 , . . . , yd , and i defined as above. In particular,
deg(F ) is an integer.
Proof. Let U be a connected open neighborhood of x, with the property that U is
contained in the set of
values of F . Let m
comp (U ) be a form of total integral
R regular

1. Thus deg(F ) = N F . Choosing U sufficiently small, the pre-image F 1 (U ) is

`
a disjoint union di=1 Vi where Vi is an open neighborhood of yi . Then F restricts to
diffeomorphisms Vi U , which are orientation preserving if i = 1 and orientation
reversing if i = 1. Thus
Z
Z
d Z
d
d
X
X
X

deg(F ) =
F =
(F |Vi ) =
i
=
i .
N

i=1

Vi

i=1

i=1

54

CONTENTS


Corollary 10.3. Suppose deg(F ) 6= 0. Then F is surjective.
Proof. Apply the Theorem to any x in the complement of the image of F .

Pd
Example 10.4 (Fundamental theorem of algebra). Let p(z) = i=0 ai z i be a polynomial of degree d > 0, with ad = 1. The map p : C C takes to , so it extends to
a map F p : S 2 S 2 where we view S 2 as the one point compactification of C. The map
F p is smooth. Replacing ai with tai for i < d, we see that F p is homotopic to the map
defined by the polynomial q(z) = z d . The equation q(z) = 1 has exactly d solutions,
given by the dth roots of unity. Since holomorphic maps are orientation preserving at
all regular points, it follows that
deg(F p ) = deg(F q ) = d.
Thus F p is surjective. This shows that the equation p(z) = a has at least one solution,
for all a C. (In fact, we see that for an open dense subset of values, it has exactly d
solutions.)
Example 10.5. View S 1 as the unit circle in R2 = C, and let the kth power map
Pk : S 1 S 1 be the restriction of the map C C, z 7 z k . If = (2)1 d is the
standard volume form on S 1 , we have Pk = k. Thus deg(Pk ) = k.
Theorem 10.6. Two maps F0 , F1 : S 1 S 1 are smoothly homotopic if and only if
they have the same mapping degree.
Proof. Since homotopy is an equivalence relation, it suffices to show that any
smooth map F : S 1 S 1 of degree k is homotopic to Pk . In fact, since F Pk
has mapping degree 0, it suffices to consider the case k = 0. In this case, F has
integral 0. Locally, F = (2i)1 d log(F ). But since integration is an isomorphism
H 1 (S 1 ) R, we know that F1 = df for some globally defined function f C (S 1 ).
That is,
F (z) = e2if (z)
for all z S 1 . Let Ft (z) := e2it f (z) . Then Ft defines a smooth homotopy between F1
and F0 = 0 .

11. Kuenneth formula
Suppose the manifold M is a direct product M = B F . Let B , F denote the
projections from M to the two factors. The bilinear map
(B) (F ) (M ), (, ) 7 B F
is a chain map, hence it induces a map in cohomology,
(8)

H(B) H(F ) H(B F ) = H(M ).

11. KUENNETH FORMULA

55

Theorem 11.1 (Kuenneth theorem). Suppose B (or F ) is of finite type. Then the
map (8) is an isomorphism. That is,
p

H (B F ) =

p
M

H j (B) H pj (F ).

j=0

Put differently, the Kuenneth theorem says that any class in H(B F ) has a representative of the form,
N
X
=
B i F i .
i=1

Proof. Let us temporarily introduce the notation

H p (B; F ) :=

p
M

H j (B) H pj (F )

j=0

(since the following long exact sequences would get too long otherwise.) We want to
show that the natural map H p (B; F ) H p (B F ) is an isomorphism. The idea is once
again to use induction on the number l of open sets in a good cover. Note first that the
Kuenneth theorem holds for B = Rn , since B F retracts onto {0} F in this case.
Thus the Theorem is true for l = 0. For the induction step, we use the Mayer-Vietoris
sequence. Suppose B = U V . We have the Mayer-Vietoris sequence
H j (U V ) H j (U ) H j (V ) H j (U V ) H j+1 (U V ) .
Tensor with H pj (F ), and sum over j to obtain a new exact sequence,
H p (U V ; F ) H p (U ; F )H p (V ; F ) H p (U V ; F ) H j+1 (U V ; F ) .
Consider the diagram
H p (U V ; F )

/ H p (U ; F ) H p (V ; F )

/ H p (U V ; F )


/ H p (U F ) H p (V F )


/ H p ((U V ) F )

H p ((U V ) F )

/ H p+1 (U V ; F )

/ H p+1 ((U V ) F )

We claim that this diagram commutes. As usual, this is fairly easy, except for the square
involving the connecting homomorphism. Thus let Z j (U V ) and Z pj (F ). Let
U , V be a partition of unity for U, V , and U V U , U V V the corresponding partition
of unity for U F , V F . Then is realized on the level of cocycles. We have to show
that the two forms
(U V F ),
U V () F

56

CONTENTS

are cohomologous. Recall that ()|U = d(V ), ()|V = d(U ) so that is

actually supported in U V . Similarly, (U V F ) is supported in (U V ) F .
We calculate, on (U V ) F ,
(U V F ) = d(U V V F )
= U V d(V ) F
= U V () F .
The Theorem now follows by induction on the number of elements in a good cover,
together with the Five-lemma.

As an application, we obtain the cohomology of an m-torus
1
T =S
S}1 .
| {z
m times

As an algebra, H(S 1 ) is an exterior algebra in one generator of degree 1. That is,

H(S 1 ) = R. (This just reflects the obvious fact that the volume form squares to 0.)
Consequently
H(T ) = H(S 1 ) H(S 1 )
is an exterior algebra in M generators, i.e.
H(T ) = R R
= RN
as graded algebras.
It is convenient to introduce the following notation. For any manifold M , one calls
bk (M ) = dim(H k (M )) the kth Betti number and one defines a polynomial in one variable
t,
dim
XM
p(M )(t) =
bk (M ) tk ,
k=0

called the Poincare polynomial. For instance, p(Rm )(t) = 1 while p(S m )(t) = 1 + tm .
Poincare duality for a compact connected oriented m-dimensional manifold M says
that
tm p(M )(t1 ) = p(M )(t),
and the Kuenneth formula implies that
p(M1 M2 ) = p(M1 )p(M2 ).
For instance, the Poincare polynomial for an m-torus T = S 1 S 1 is p(T )(t) =
(1 + t)m .
There is also a Kuenneth theorem for compactly supported cohomology. It can
be derived using the Mayer-Vietoris sequence for compactly supported cohomology, or
simply by Poincare duality. One introduce Betti numbers bk (M )comp and a Poincare
polynomial p(M )comp as before.

12. DE RHAM THEOREM

57

12. De Rham theorem

12.1. Cech cohomology. Let A be an abelian group, written additively. We are
mostly interested in the cases A = R, Z, Zp . Let M be a manifold, and U = {U } an
open cover of M . For any collection of indices 0 , . . . , p such that U0 . . . Up 6= ,
let
U0 ...p = U0 . . . Up
One defines the Cech complex

C 0 (U, A) C 1 (U, A) C 2 (U, A)

as follows. A Cech-p-cochain f C p (U, A) is a function

a
a
f=
f0 ...p :
U0 ...p A,
0 ...p

where each f0 ...p : U0 ...p A. is locally constant, and anti-symmetric in its indices.
The differential is defined by the formula,
(f )0 ,...,p+1 =

p+1
X

(1)i f0 ,...,c
i ,...,p+1

i=0

where the hat means that the entry is to be omitted. For example, if f C 0 (U, A),
(f )0 1 = f1 f0
if g C 1 (U, A),
(g)0 1 2 = g1 2 g0 2 + g0 1 .
We thus see that
(f ) = (f2 f1 ) (f2 f0 ) + (f1 f0 ) = 0.
Exercise 12.1. Verify that = 0 in general.
p (U, A) := H p (C(U,

The cohomology groups H

A)) are called Cech cohomology groups
with coefficients in A.
A nice feature of the Cech cohomology groups is that they are purely combinatorial,
reflecting the intersections of elements in our open cover. On the other hand, if we want
to use them to define invariants of a manifold, there is a problem that they do depend
on the cover.
If A A is a homomorphism of abelian groups, one has an induced homomor p (U, A)
phism of Cech cochain, and hence a homomorphism of cohomology groups, H
p (U, A ).
H
Example 12.2. Suppose each U is connected. A Cech 0-cocycle is just a collection
0 (U, A) = A.
of elements f A that agree on overlaps. That is, if M is connected, H

58

CONTENTS

Example 12.3. Suppose M is connected and all U are diffeomorphic to Rm . If

Z 1 (M ), we can choose 0 (U ) such that d = |U . Then f := is a
constant f R, and clearly defines a 1-cocycle in C 0 (U, R). A different choice differs
from by a constant c , thus f = c c = (c) . Hence, the Cech cohomology
class of f does not change. Similarly, if is cohomologous to , the difference is the
differential dg of a global function g 0 (M ). But replacing by = + g|U , does
not change f . The upshot is that there is a well-defined map
1 (U, R).
H 1 (M ) H
Suppose [] is in the kernel of this map. This means that the can be chosen such
that f becomes a coboundary, i.e. f = c c for some constants c . Replacing
by c takes f to 0. That is the new functions agree on overlaps, and
define a function 0 (M ) with d = . Thus [] = 0. This shows that the map
1 (U, R) is injective. Conversely, suppose f is a given Cech 1-cocycle.
H 1 (M ) H
P
Choose a partition of unity , and define = f . Then
X
X
(f f ) =
f = f
=

and therefore
d d = 0
Hence, there is a unique form Z 1 (M ) with |U = d . This shows that the map
1 (U, R) is an isomorphism.
H 1 (M ) H
12.2. De Rhams theorem.
Theorem 12.4 (De Rham). If U is a good cover, the Cech cohomology groups
p

H (U, R) are canonically isomorphic to the de Rham cohomology groups H p (M ). In

particular, they are independent of the choice of good cover.
Note that in the case of a good cover, the open sets are in particular connected. Thus
the locally constant functions f0 ...p are really just elements of the group A.
Below we will present A. Weils proof of de Rhams theorem. A first step is to
introduce, for each q, another type of Cech complex

C 0 (U, q ) C 1 (U, q ) C 2 (U, q )

as follows. A Cech-p-cochain C p (U, q ) is a collection of q-forms 0 ...p q (U0 ...p ),
p (U, q ) :=
anti-symmetric in its indices. The differential is defined as before. We define H

H p (C(U,
q )).

p (U, q ) = 0 for all p > 0.

Theorem 12.5. Suppose U is a good cover. Then H
Proof. Let be a partition of unity subordinate to the given cover. Define an
operator
h : C p (U, q ) C p1 (U, q )

12. DE RHAM THEOREM

59

by
(h)0 ...p1 =

0 ...p1

We verify that h is a homotopy operator: Indeed

(h)0 ...p =

X
(1)i 0 ...c
i ...p1
i,

and
h()0 ...p =

()0 ...p

0 ...p

(1)i 0 ...c
i ...p ,

thus h + h = Id on forms of degree p > 0.

We now arrange C p,q := C p (U, q ) as a double complex, as follows:

O

C 0 (U, 2 )
O

/C
1 (U, 2 )
O

C 0 (U, 1 )
O

/C
1 (U, 2 )
O

/C
1 (U, 1 )
O
/

/C
1 (U, 1 )
O

C 0 (U, 0 )

/C
1 (U, 0 )

/C
1 (U, 0 )
/

Notice that the cohomology groups are trivial in both horizontal and vertical directions,
except in degree 0. One can make the double complex into a single complex, by letting
K p :=

p
M

C j (U, pj ),

j=0

with differential, D = +(1)p d. The factor (1)p is necessary so that D2 = 0. One calls

H(K, D) = H(C(U,
), D) the total cohomology of the double complex. A D-cocycle is
(0)
a collection ( , . . . , (p) ) of cochains (j) C j (U, pj ) with D( (0) + . . . + (p) ) = 0.

60

CONTENTS

This gives a set of equations,

d (0) = 0
d (1) = (1)p (0)
d (2) = (1)p1 (1)

(p)
d
= (p1) ,
0 = (p)
Lemma 12.6. The restriction map p (M ) C 0 (U, p ) K p induces an isomorphism in cohomology,

H p (M ) = H p (C(U,
), D).
Proof. Surjectivity: Let ( (0) , . . . , (p) ) be a D-cocycle. We have to show that it
is cohomologous to a cocycle in the image of the restriction map p (M ) C 0 (U, p ).
Since (p) = 0 and the -cohomology is trivial, we can write (p) = (p1) , where
(p1) C p1 (U, 0 ). Subtracting from ( (0) , . . . , (p) ) the coboundary D (p1) , we thus
achieve (p) = 0. Then (p1) = d (p) = 0. Thus we can write (p1) = (p2) , and
subtracting D (p2) we achieve (p1) = 0. Proceeding in this manner, we successively
subtract D-coboundaries and arrange that (j) = 0 for all j > 0. The remaining (0)
C 0 (U, 0 ) satisfies d (0) = 0 and (0) = 0. Thus it is the restriction of a closed form
p (M ).
Injectivity: Let Z p (M ) be a cocycle, and (0) C 0 (U, p ) be defined by restriction. Suppose (0) is the D-coboundary of some ( (0) , . . . , (p1) ) K p1 . Then
(p1) = 0, so by adding a D-coboundary we can arrange (p1) = 0. Proceeding in
this manner, we can arrange that all (j) = 0, except maybe (0) . The remaining form
(0) satisfies (0) = 0, which means that it is the restriction of a global form , and
(1)p d (0) = (0) , meaning that (1)p d = . Thus [] = 0 and we are done.

Notice that the proof was purely algebraic. It involved that the -cohomology is
trivial in positive degree, and that the kernel of the map : C 0 (U, q ) C 1 (U, q ) is
exactly the image of the map q (M ) C 0 (U, q ).
Hence, the proof works equally well with the roles of , d reversed. That is, we also
have
Lemma 12.7. The restriction map C p (U, R) C p (U, 0 ) K p induces an isomorphism in cohomology,
p (U, R) = H p (C(U,

H
), D).
Putting the two Lemmas together, we have proved de Rhams theorem:
p (U, R) = H p (C(U,

H
), D) = H p (M ).

12. DE RHAM THEOREM

61

As one application, one can introduce the notion of integral de Rham classes: Indeed, the
inclusion Z R induces a homomorphism of Cech cochains, hence a map in cohomology
p (U, Z) H
p (U, R) = H p (M ). Classes in the image of this map are called integral.
H
One important consequence of de Rhams theorem is that they are topological invariants (since the Cech cohomology groups are). Hence they cannot be used to distinguish
different manifold structures on a given topological space.
12.3. Relative cohomology. Let C (M, N ) be a smooth map. Define a
complex
p () := p1 (M ) p (N )
with differential, d(, ) = ( d, d). It is straightforward to check that d squares
to 0. A class in H p () is called a relative de Rham cohomology class for the map . Thus
classes in H p (M ) are represented by closed p-forms on N together with a primitive
p1 (M ) for the pull-back . The sequence
0 p1 (M ) p () p (N ) 0,
where the first map takes to (1)p (, 0) and the second map takes (0, ) to , is
an exact sequence of differential complexes. Hence it induces a long exact sequence in
cohomology:
H p1 (M ) H p () H p (N ) H p (M )
An immediate consequence is that if M is contractible, then H p () = H p (N ) for p > 0,
while if N is contractible, then H p () = H p1 (M ) for p > 0.
Suppose A is an abelian group, and Ui , Vi are good covers of M, N such that (Ui )
Vi . Such covers exist: Start with a good cover Vi of N , and replace the cover 1 (Vi ) by
a good refinement Ui . (We may use the same index set, if we allow the same Vi to appear
several times.) Let C (M, A), C (N, A) be the Cech complexes for the given covers. We
then have a pull-back map : C (N, A) C (M, A) which one can use to define
C p (, A) = C p1 (M, A) C p (N, A)
with differential
(, ) = ( (), ()).
Again, we have a long exact sequence
p1 (M, A) H
p (, A) H
p (N, A) H
p (M, A)
H
De Rhams theorem extends to relative cohomology:
p (, R).
Theorem 12.8. There is a canonical isomorphism H p ()
=H
Proof. As in the proof of de Rhams theorem, it is useful to introduce auxiliary
Cech cohomology groups. Consider the double complex
C p,q := C p (, q ) = C p1 (M, q ) C p (N, q )

62

CONTENTS

with the obvious differentials d, . Since we are working with good covers, and since
the complex p () is acyclic if M, N are contractible, it follows that the columns of the
double complex are acyclic. We claim that the rows are acyclic as well. To this end, we
need to generalize Theorem 12.5 to our setting. Indeed, let hN , hM denote the homotopy
operators for the Cech complexes of M, N , cf. Theorem 12.5. Then
h : C p (, q ) C p1 (, q ), h(, ) = (hM ( (hN ) ), hN )
is a homotopy operator for the complex C (, q ). We verify:
h(, ) =
=
=
=

h( , )
(hM ( (hN ) + ), hN )
(hM ( (hN ) + ), hN )
(hM ( (hN )), hN )

while
h(, ) = (hM ( (hN ) ), hN )
= ( (hN ) + hM ( (hN )), hN ).
Adding the two expressions, we obtain
h(, ) + h(, ) = (, )
as desired. Hence the proof of de Rhams theorem goes through as before.

13. Fiber bundles

13.1. Fiber bundles and vector bundles. In first approximation, a fiber bundle
with fiber F is a smooth map : E B with fibers diffeomorphic to a given fiber F ,
in such a way that E is locally the product of the base and the fiber:
Definition 13.1. A fiber bundle over with standard fiber F is a manifold E (called
the total space) together with a map : E B to another manifold B (called the base),
with the following property: There exists an open covering U of B, and diffeomorphisms
: 1 (U ) U F
such that
prU = |1 (U ) .
A smooth map : B E is called a section of E if = IdB . The space of sections
is denoted (B, E).
Note that it follows from the definition that is smooth and that it is a surjective
submersion. The maps are called local trivializations of the fiber bundle.
For every subset S B, we denote ES = 1 (S). In particular, taking S to be a
point b B, Eb is the fiber 1 (b). If U B is an open neighborhood, EU is again a
fiber bundle.

13. FIBER BUNDLES

63

Definition 13.2. Two fiber bundles : E B and : E B with standard

fiber F are called isomorphic if there exists a diffeomorphism : E E such that
= .
The direct product E = B F with : E B the projection to the first factor, is
a fiber bundle called the trivial bundle. A bundle is called trivializable if it isomorphic
to the trivial bundle. Clearly, existence of sections is a necessary (but not sufficient)
condition for a fiber bundle to be isomorphic to the trivial bundle.
Exercise 13.3. a) View S 1 R2 = C as the unit circle in the complex plane, with
coordinates z C. Let k be a non-zero integer. The map C C, z 7 z k restricts to
a map : S 1 S 1 . Construct local trivializations to show that is non-trivial fiber
bundle with fiber Zk = Z/kZ.
b) Classify the fiber bundles : E S 1 with fiber Z2 , up to isomorphism. How
about fiber Zk for arbitrary integers k 2?
Exercise 13.4. Show that the map : S m RP (m) makes S m into a fiber bundle
with fiber Z2 .
Exercise 13.5 (Hopf fibration). Complex projective space may e defined as a quotient of S 2n+1 Cn+1 by the relation, z z if and only if z = cz, where c S 1 C
is a complex number of absolute value 1. Let : S 2n+1 CP (n) be the quotient map.
Show that that is a fiber bundle with fiber F = S 1 . For n = 1, this becomes a fiber
bundle : S 3 S 2 = CP (1) called the Hopf fibration.
Lemma 13.6. If B is smoothly contractible, then any fiber bundle E over B is isomorphic to the trivial bundle.
Sketch of proof. Let S : I B B be a smooth retraction onto b B. Let
F = Eb be the fiber over b. Given a Riemannian metric on E, there exists a unique
smooth lift S : I E E such that for all t I,
x)) = S(t, (x))
(S(t,
and such that
S
(t, x) (Tx F(x) ) .
t
For each y B, the restriction of S to 1 (y) {1} is a diffeomorphism Ey Eb = F .
The map
1))
: E B F, x 7 ((x), S(x,
is the desired isomorphism with the trivial bundle.

Definition 13.7. Let : E B be a fiber bundle. A smooth map : B E is

called a section of if = idB . The set of sections is denoted (B, E).

64

CONTENTS

13.2. Vector bundles. A vector bundle is a fiber bundle whose fibers have the
structure of vector spaces. More precisely:
Definition 13.8. Let M be a manifold. A real (resp. complex) vector bundle of
rank k over M is a fiber bundle : E M with standard fiber F = Rk (resp. Ck ),
such that all fibers Ex (x M ) carry the structure of real (resp. complex) vector spaces,
and such that the local trivializations EU
= U F can be chosen fiberwise R-linear
(resp. C-linear). An isomorphism of vector bundles : E M and : E M is a
fiber bundle isomorphism that is fiberwise linear.
One can think of a vector bundle as a family of vector spaces smoothly parametrized
by the base. Suppose E is a real vector bundle with local trivializations : EU
=
U Rk . On U the map
k
k
1
: U R U R

has the form

1
(x, v) = (x, g (x).v)
where g (x) GL(k, R), the group of invertible kk-matrices. The transition functions
g : U GL(k, R) have the cocycle property,
g g = g
on U . Conversely, given any cover U of a manifold M , and any collection of functions
g : U GL(k, R) with the cocycle property, there is a unique vector bundle having
the g s as transition functions. Indeed, E may be defined as a quotient
a
E=
(U Rk )/

where for any x U and v Rk , (x, v) U Rk is declared equivalent to (x, g (v))

U Rk .
13.3. Examples of vector bundles.
Examples 13.9.
(a) Let M = S n Rn+1 be the unit sphere. Define E
n
n+1
S R
to be the set of pairs (x, v) with x v = 0. Then E is a vector bundle
of rank n. The map (x, v) 7 (x, v) is a vector bundle homomorphism.
(b) Let M = RP (n). Each point x RP (n) represents a 1-dimensional subspace of
Ex Rn . Let
E RP (n) Rn+1
be the set of all (x, v) such that v is in the 1-dimensional subspace determined
by x. Then E is a vector bundle of rank 1 (also called a (real) line bundle). It
is called the tautological bundle over projective space.

13. FIBER BUNDLES

65

(c) This example generalizes: Let M = GrR (k, n) be the Grassmannian of k-planes
in Rn . By definition, each x M represents a k-dimensional subspace of Ex
Rn . Let
E GrR (k, n) Rn
be the set of al (x, v) such that v is contained in the k-dimensional subspace
parametrized by x. The total space of this vector bundle is called the Stiefel
manifold.
(d) Let M be a manifold with atlas {(U , )}. For each U , the map 1
:
(U ) (U ) is a diffeomorphism. For x U let g (x) GL(k, R) be
the Jacobian of the map 1
at (x). Clearly, the g satisfy the cocycle
condition. The resulting vector bundle is the tangent bundle T M of M . Its
fibers (T M )x are canonically identified with the tangent spaces Tx M . (Recall
that any choice of chart identifies Tx M
= Rk , and a change of chart changes
the identification by the Jacobian of the change of coordinates.) The space of
sections is just the space of vector fields:
X(M ) = (M, T M ).
Note that it seems somewhat miraculous from this perspective that X(M ) carries
a natural Lie bracket.
(e) Let S be an embedded submanifold of a manifold M . Then the restriction T M |S
is a vector bundle over S, containing T S. The quotient bundle NS := T M |S /T S
with fibers (NS )x := Tx M/Tx S is again a vector bundle called the normal bundle
of S in M .
All standard constructions for vector spaces carry over to vector bundles. Thus,
if E M is a vector bundle, one can form the dual bundle E M with fibers
(E )x = Ex , and the exterior powers p E M with fibers (p E)x = p Ex . If E E is
a vector subbundle, one can form the quotient bundle E/E with fibers (E/E )x = Ex /Ex .
Similarly, one defines tensor products, direct sums, ...
Exercise `
13.10. Work out the details of all these claims. E.g., show that the disjoint
union E := xM Ex carries a natural structure of a vector bundle. What are the
transition functions?
Starting from the tangent bundle T M one can form the dual bundle T M = (T M ) ,
with sections (M, T M ) = 1 (M ). One can also form k T M , with sections
(M, k T M ) = k (M ). If S M is any submanifold, one can consider the restriction T M |S S; sections of T M |S are called vector fields along S. The quotient
bundle NS := T M |S /T S is called the normal bundle to S in M . Given a Riemannian
metric on M , one identifies NS with the orthogonal complement of T S in T M |S , that is
one has a splitting T M |S = T S NS .
We will need the following fact about the normal bundle.

66

CONTENTS

Theorem 13.11. Suppose M is a manifold, S a compact embedded submanifold, and

NS the normal bundle of S inside M . Then there exist open neighborhoods V of S in
M and U of S in NS , and a diffeomorphism : U V such that restricts to the
identity map on S.
That is, the normal bundle NS serves as a local model for a neighborhood of S in
M.
Sketch of proof. Put a Riemannian metric on M . Suppose for simplicity that
M is geodesically complete. (That is, all geodesics exist for all time.) Then NS gets
identified with the set of all vectors in T MS that are orthogonal to T S. Geodesic flow
defines a map NS R M . Let be the restriction of this map to N S {1}. One
verifies that the tangent map along S NS is invertible , so by the implicit function
theorem is a diffeomorphism on a neighborhood U of S.

14. The Thom class
14.1. Thom isomorphism. Let : E M be a rank k vector bundle over a
compact manifold M of dimension m. Since E retracts onto M , the pull-back map
induces an isomorphism
: H p (M )
= H p (E).
Let us suppose that both M and E carry orientations. Then Poincare duality tells us
that we have a dual isomorphism,
q
Hcomp
(E) H qk (M ).

where q = m + k p. This isomorphism has a simple meaning: Suppose Z(M ) and

Zcomp (E). Then
Z
Z
Z

=
( ) =
,
E

where is integration over the fibers.

Theorem 14.1 (Thom isomorphism). Let : E M be a rank k vector bundle
over a compact manifold M , with a fiberwise orientation. Then integration over the
fibers defines an isomorphism,
p
(E) H pk (M ).
: Hcomp

Remark 14.2. The argument given above only applies to the case where M is also
oriented (and E carries the induced orientation.) However, one can show that this
assumption is not necessary. One can also drop the assumption that M be compact,
p
(E) with the cohomology of the complex of forms with compact
if one replaces Hcomp
support in fiber direction.

14. THE THOM CLASS

67

k
In particular, there is a unique class [Th(E)] Hcomp
(E), called the Thom class,
with property
Th(E) = 1.
A representative of the Thom class is called a Thom form. Now let : M E be the
inclusion of the zero section.

Lemma 14.3. The Thom class has the property

[ (Th(E) )] = [ ]
for all [] H(M ).
Proof. Let = . The inclusion is a homotopy inverse to the projection ,
since = IdM and since is homotopic to IdE . Therefore, the pull-back maps
, are inverse to each other in cohomology, and [] = [ ]. We have
(Th(E) ) = Th(E) = ,
by the properties of the fiber integration map.

Remark 14.4. The definition of the Thom for extends to non-compact manifolds
M that need not be orientable. All that is required is a fiberwise orientation of the
vector bundle : E M . A differential form (E) is said to have fiberwise
compact support if for all compact subsets K M , the intersection 1(K) supp()
is compact. The space f.c. (E) of differential forms with fiberwise compact support
is a differential complex, and one defines the corresponding cohomology with fiberwise
compact support Hf.c. (E). One can prove that the fiber integration map defines an
isomorphism : Hf.c. (E) H(M ), and one defines the Thom class to be the inverse
image of 1 H 0 (M ) under this isomorphism. The space f.c. (E) is a module for (E)
under wedge product, and in cohomology one once again has Lemma 14.3.
Definition 14.5. The class Eul(E) := Th(E) H k (M ) is called the Euler class
of the oriented rank k vector bundle E. If M be a compact oriented manifold, one calls
Eul(T M ) the Euler class of the manifold M .
Note that e(M ) = 0 if the tangent bundle can be trivialized.
L
Definition 14.6. Suppose K = p0 K p is a differential complex with finite dimenL
P
sional cohomology H(K) = pZ H p (K). The polynomial p(K)(t) = p tp dim H p (K)
is called the Poincare polynomial of (K, d), and p(K)(1) Z is the Euler characteristic.
In particular, if K = (M ), one calls p(M ) := p((M )) the Poincare polynomial of M
and p(M )(1) the Euler characteristic of M .
Theorem 14.7 (Gauss-Bonnet-Chern). Let M be a compact oriented manifold. Then
the Euler characteristic of M equals the integral of its Euler class,
Z
X
Eul(T M ) =
(1)i dim H i (M ).
M

In particular, this integral is an integer.

68

CONTENTS

Theorem 14.7 is quite remarkable, in that it expresses a topological quantity (the

alternating sum of de Rham cohomology groups) in local terms, as the integral of a
differential form over the manifold.
Remarks 14.8. a) The full Gauss-Bonnet-Chern theorem gives, in addition, a concrete prescription how to represent Eul(T M ) in terms of curvature invariants. (The
classical Gauss-Bonnet theorem says that for any compact connected oriented 2-manifold
, the integral of the so-called Gauss curvature equals the Euler characteristic 2 2g,
where g is the genus = number of handles.
b) Of course, Eul(T M ) = 0 whenever the tangent bundle of M is trivial. For instance,
it can be shown (see e.g. the book on characteristic classes by Milnor-Stasheff) that the
tangent bundle of any compact oriented 3-manifold is trivial.
c) Note that pM (1) = 0 if dim M is odd. This follows by putting t = 1 in the
equation for Poincare duality, pM (t1 ) = tm pM (t).
The proof of Theorem 14.7 requires some preparation. Let : M M be the
diagonal, and [ ] its Poincare dual. Now let N denote the normal bundle of
in M M .
Lemma 14.9. There is an isomorphism N = T .
Proof. Choose a Riemannian metric on M , and let M M be equipped with the
product metric. For any x M , the tangent space to at (x) consists of vectors (v, v),
and its normal space of vectors (v, v) where v Tx M .

By the tubular neighborhood theorem, a small neighborhood of in M M is
modeled by a neighborhood of in N . This gives
Z
e(M ) =
Eul(T M )
M
Z
=
Eul(T )

Z
=
Eul(N )

Z
=
.

We calculate the integral of as follows. Let j Z pj (M ) be cocycles such that [j ]

are a basis for the vector space H(M ). Let j Z mpj (M ) be forms representing the
Poincare duals, i.e.
Z
j k = jk .

Let pr1 , pr2 : M M M denote the projections to the two factors. By the Kuenneth
theorem, pr1 j pr2 k represent a basis of H(M M ).

14. THE THOM CLASS

69

Lemma 14.10. The expansion of [ ] in the basis [pr1 j pr2 k ] reads,

X
[ ] =
(1)pj [pr1 j pr2 j ].
j

Proof. To find the coefficient cjk of [pr1 j pr2 k ], consider the integral
Z
Z

(9)
(pr1 j pr2 k ) = (pr1 j pr2 k ).
M M

The left hand side is given by

Z
X
cab
ab

M M

(pr1 j pr2 k ) (pr1 a pr2 b )

pa (pk +mpj )

cab (1)

M M

ab

pa (pk +mpj )

cab (1)

pr1 (a j ) pr2 (k b )

(a j )

ab

(k b )

= cjk (1)pj (pk +mpj ) .

The right hand side of (9) can be written as an integral over M , using the diagonal
embedding : M M M
Z
Z

(pr1 j pr2 k ) =
j k = (1)pk (mpj ) jk .
M

Comparing the two results, we have found

cjk = (1)pj pk jk = (1)pj jk ,

Using this result we calculate,
e(M ) =

pj

(1)

(1)pj

pr1 j pr2 j
j j

(1)pj

X
p

(1)p dim H p (M ),

70

CONTENTS

proving Theorem 14.7. The calculation just given has the following generalization. For
any smooth map F : M M from M to itself, let : = {x, F (x)| x M } M M
be its graph, and the Poincare dual.
UsingPour basis j for H(M ), introduce the components of F : H(M ) H(M ) by
F j = k Akj k . Thus
Z
Akj =
(F j ) k
M

Again, one expand [ ] in terms of our basis for H(M M ):

X
[ ] =
cjk [pr1 j pr2 k ].
jk

We have

M M

(pr1

pr2

k ) =

pr1 j pr2 k .

The left hand side is cjk (1)pj (pk +mpj ) as before. For the right hand side we obtain,
Z
Z

pr1 j pr2 k =
(pr1 j pr2 k )

ZM
=
j F k
M

= (1)pk (mpj ) Ajk .

Comparing the two results we obtain,
cjk = (1)pj pk Ajk .
If we integrate over the diagonal , we obtain, therefore
Z
Z
X
pj pk
=
(1) Ajk
j k

jk

X
(1)pj pk Ajk jk
=
jk

X
=
(1)pj Ajj
j

(10)

X
(1)p tr(F | H p (M )).
=
p

the alternating sum of the traces of the linear maps F : H p (M ) H p (M ). This

integral has a nice geometric interpretation, as we shall explain in the following section.

15. INTERSECTION NUMBERS

71

15. Intersection numbers

Let M be a manifold, and S, S two embedded submanifolds with dim S + dim S
dim M . One says that S, S intersect transversally if for all points x S S ,
Tx M = Tx S Tx S .
Exercise 15.1. Suppose S, S M are submanifolds of dimensions k, k which intersect transversally. Given x S S , show that there exists a coordinate chart (U, )
around x, with (x) = 0, such that (U S) is an open subset of {(x1 , . . . , xk , 0 . . . , 0)}

and (U S ) is an open subset of {(0, . . . , 0, xmk +1 , . . . , xm )}. Conclude that S S

is an embedded submanifold of dimension k + k m, with Tx (S S ) = Tx S Tx S .
From the exact sequence
0 Tx (S S ) Tx S Tx S Tx M 0
one sees that orientations on S, S , M naturally induce an orientation on S S . In
particular, if S, S have complementary dimensions, S S is a discrete set of points, and
an orientation is given by assigning a sign to each element of S S . The plus sign
appears if and only if an oriented basis of Tx S, followed by an oriented basis of Tx S , is
an oriented basis for Tx M .
Theorem 15.2. Suppose S, S are compact oriented submanifolds of the oriented
manifold M , and that S, S intersect transversally. Then the Poincare dual SS of
S S in M can be represented by the product of Poincare duals of S, S :
SS = S S .
Sketch of proof. The key point is that the pull-back of S to S is the Poincare
dual of S S inside S . (To see this, note that the restriction to S S of the normal
bundle NS of S is the normal bundle of S S inside S . Hence, the Thom class Th(NS )
pulls back to the Thom class of NS |SS , since Thom forms are chratcerized by the
property that theirt integral over the fibers is 1. But the Thom class of the normal bundle
equals the Poincare dual for a tubular neighborhood. ). We can therefore compute,
Z
Z
Z

S =
S S =
M

SS

for all cocycles , which shows that S S represents the Poincare dual of S S .

In the special case that S, S have complementary dimension and intersect transversally. the Theorem shows that the intersection number
Z
Z
Z
(mk)(mk )

S
S = (1)
i(S, S ) :=
S S =
M

is an integer, in fact

i(S, S ) =

SS =

xSS

(x),

72

CONTENTS

where (x) = 1 according to whether the decomposition Tx M = Tx S Tx S preserves or reverses orientation. More generally, i(S, S ) is an integer whenever S can be
perturbed (i.e. homotoped) to have transversal intersection, since i(S, S ) is invariant
under homotopies. (It is a result from differential
R topology that this is always possible.)
In the last section, we considered integrals , where = {(x, F (x)} is the graph
of a smooth map F : M M . The intersection with the diagonal is in 1-1 correspondence with the fixed points x = F (x) of F . When is the intersection transversal?
Lemma 15.3. and are transversal, if and only if all fixed points x of F are
non-degenerate, that is,
det(dx F I) 6= 0.
The sign (x) := (x, x) at any point of intersection is equal to the sign of the determinant
det(dx F I).
Proof. Suppose x = F (x), so that (x, x) . We have
T(x,x) = {(v, v)| v Tx M }, T(x,x) = {(v, dx F (v))| v Tx M }.
The map Tx M Tx M
= T(x,x) T(x,x) T(x,x) M = Tx M Tx M
(v, w) 7 (v + w, v + dx F (w))
is described by a block matrix,

1 1
1 dx F

This has determinant,





1
0
1 1
= det(dx F 1).
= det
det
1 dx F 1
1 dx F
Thus, and are transversal if and only if dx F I is invertible, and the sign of the
determinant gives (x, x).

Putting this together with the formula (10), we obtain:
Theorem 15.4 (Lefschetz fixed point formula). Let F : M M be a smooth map
from a compact oriented manifold to itself, with the property that all fixed points of F
are non-degenerate. For each fixed point x, let (x) = sign(det(dx F I)). Then
X
X
(x) =
(1)p tr(f | H p (M )).
x=F (x)

Note that the right hand side of this equation depends only on the smooth homotopy
class of the map F .
R
ROne can obtain a similar description for the integral . The integral is equal
to , where is the graph of a smooth map F homotopic to the identity map. To
obtain such a map, pick a vector field X X(M ), and let F t : M M be its flow. We
want that for t sufficiently small, all fixed points of F t are non-degenerate. For small t,

15. INTERSECTION NUMBERS

73

the fixed points of F t are exactly the zeroes of the vector field X. Let x M be such
a zero, Xx = 0. Since F t preserves x, we obtain a 1-parameter group of linear maps

dx F t : Tx M Tx M . Let A : Tx M Tx M be the linear map A := t

|t=0 dx F t . Then
t
dx F = exp(tA). One calls A the linearization of X at x. One calls x a non-degenerate
zero of X if det(A) 6= 0. Let (x) := sign(det(A))
We have
det(dx F t I) = det(tA + . . .) = tm det(A) + . . .
where . . . denotes higher order terms in the Taylor expansion. Thus x is a non-degenerate
zero for X if and only if it is a non-degenerate fixed point for F t , for t sufficiently small.
Let t > 0. Since sign(det((dx F t I)) = sign(det(A)) = (x) we obtain:
Theorem 15.5 (Poincare-Hopf). Let X be a vector field on a compact-oriented manifold M with non-degenerate zeroes. Then
X
X
(x) =
(1)p dim H p (M ).
x: Xx =0

where (x) is the sign of the determinant of the linearization of X at x.

One way of constructing vector fields with non-degenerate zeroes comes from Morse
theory, cf. Milnors book.
The Poincare-Hopf and Lefschetz formula can sometimes be used to obtain a lot of
information on the Betti numbers, without any serious calculation.