Professional Documents
Culture Documents
Roberto Pignatelli
Copyright
c 2016 Roberto Pignatelli
P UBLISHED ONLINE
Licensed under the Creative Commons Attribution-NonCommercial 3.0 Unported License (the
License). You may not use this file except in compliance with the License. You may ob-
tain a copy of the License at http://creativecommons.org/licenses/by-nc/3.0. Unless
required by applicable law or agreed to in writing, software distributed under the License is
distributed on an AS IS BASIS , WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND,
either express or implied. See the License for the specific language governing permissions and
limitations under the License.
Contents
3 De Rham theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.1 De Rham cohomology and compact support cohomology 65
3.2 Differential complexes and cohomologies 67
3.3 Mayer-Vietoris 70
3.4 The Poincar lemma 72
3.5 The Poincar lemma for the compact support cohomology 75
3.6 Manifolds of finite type 78
3.7 The Cech-De Rham complex 79
: Vi K
R The tensors of degree 1 are the elements of the dual of V1 . The tensors of degree 2 are
bilinear maps as, e.g., the standard scalar product on R2 or R3 (in which case V1 = V2 ).
Indeed, qlinearity is the natural generalization of the idea of bilinearity to the case of
more than two (but still finitely many) factors. If you know the cross product in R3 you
may prove that the map (v1 , v2 , v3 ) 7 (v1 v2 ) v3 is a tensor of degree 3.
6 Chapter 1. The de Rham cohomology on Kn
Definition 1.1.2 The space of multilinear maps from V1 V2 Vq to K is a vector space
(see Homework 1.4), which is the tensor product of V1 , . . ., Vq and is denoted by
V1 V2 Vq
R Note that the last definition, in the case n = 1, gives the vector space V1 of all linear maps
from V1 to K: the dual space of V1 .
v() := (v)
Definition 1.1.3 If V1 , . . . ,Vq are finitely dimensional vector spaces, we define then
V1 Vq := V1 Vq
We fix bases {ei1 , . . . , ein1 } of each space Vi , and we consider its dual basis {i1 , . . . , in } of
Vi . They are uniquelly determined by the formula i j (ei j0 ) = j j0 .
V1 = = Vq =: V.
Definition 1.1.5 A tensor (V )q is symmetric if its value does not depend on the order
of the vectors. In other words, if i 6= j,
(. . . , vi , . . . , v j , . . .) = (. . . , v j , . . . , vi , . . .).
(. . . , vi , . . . , v j , . . .) = (. . . , v j , . . . , vi , . . .).
The symmetric tensors form a vector subspace of (V )q usually denoted Symq V . The
skew tensors form a vector subspace of usually denoted qV .
For later convenience we define conventionally (V )0 = Sym0 V = 0V = K.
1 2 : V V K
1 1 (v 1 ) 1 (v2 )
(v1 , v2 ) 7 2 det (v ) (v )
2 1 2 2
a The constant 12 is a convention which is not universaly accepted; some authors use 1 instead there. It does not
make any "significant" difference (just change some constant in few formulas). The same holds for the constant
1
q! in Proposition 1.1.2.
: 1V 1V 2V
(1 , 2 ) 7 1 2
: q1 V q2 V q1 +q2 V
(1 , 2 ) 7 1 2
as followsa :
Note that the definition makes sense also when q1 = 0 (and/or q2 = 0), in which case
1 = K and 1 2 = 2 .
The wedge product has useful properties (see Homework 1.14). For example, it is associative,
(k1 ) 2 = k(1 2 ) = 1 k2 , 1 2 = (1)q1 q2 2 1 . In particular we can write
k1 j without ambiguity. When all the i are 1forms this has a nice expression.
Proposition 1.1.2 Assume 1 , , q V .
Then
q
1 1
1 q (v1 , . . . , vq ) = ( ) i (v (i) ) = det(i (v j )).
q! q i=1 q!
(1 q1 ) q (v1 , . . . , vq ) =
1
= ()1 q1 (v(1) , . . . , v(q1) )q (v(q) ) =
q! q
!
q1
1 1 0
= () (q 1)! 0 ( ) i (v( 0 (i)) ) q (v(q) ) =
q! q q1 i=1
!
q1
1
= ()( 0 ) i (v (i) ) q (v(q) ). (1.1)
q!(q 1)! q ,
0
0
q1 i=1
Note that each summand in the right-hand term do not really depend on and 0 , but just
on := 0 . Varying (, 0 ) q q1 we obtain each q exactly (q 1)! times, and
therefore
!
q
1
1 q (v1 , . . . , vq ) = (q 1)!( ) i (v (i) )
q!(q 1)!
q i=1
!
q
1
= ) i (v (i) )
q!
(
q i=1
1.1 Multilinear algebra 9
so v ji = i (v j ).
We have already noticed that V = 1 (V ), and therefore the i are 1-forms.
The next Theorem shows that all skew forms may be expressed by using these "determinants".
Proof. We first show that it is a set of linearly independent forms. Take constants a j1 jq R
such that
a j1 jq j1 jq = 0.
1 j1 << jq n
R From the properties in Homework 1.14 follow few very useful rules:
i j = j i ;
i i = 0;
i1 iq j = (1)q j i1 iq .
Linear applications between vector spaces induce naturally linear applications among their
spaces of tensors, mapping symmetric tensors to symmetric tensors and skew tensors in skew
tensors. Since we are mostly interested in skew tensors, we consider only them.
Definition 1.1.9 Let L : V W be a linear application. It naturally induces linear applications
(pull-backs)
L : qW qV
defined by (L )(v1 , . . . , vq ) = (L(v1 ), . . . , L(vq )).
R To ease the notation we have given the same name, L , to many different maps, so making
an abuse of notation; many others similar abuses will follow. This is a standard choice in
differential geometry: the student must get used to it.
R By definition (L1 L2 ) = L2 L1 .
L = (det L).
1.1 Multilinear algebra 11
Homework 1.1 Prove that the map (v1 , v2 , v3 ) 7 (v1 v2 ) v3 belongs to (R3 ) (R3 ) (R3 ) .
Homework 1.2 Prove that the dual basis is a basis. In other words, prove that if {ei1 , . . . , eiq }
is a basis of Vi , the formula i j (ei j0 ) = j j0 determines uniquelly the elements {i1 , . . . , iq } and
they form a basis of Vi .
Homework 1.3 Show that the map V V defined by v( ) = (v) is injective (even if the
dimension of V is not finite).
Show that, if we assume further that the dimension of V is finite, then this map is an
isomorphism.
Homework 1.4 Show that V1 V2 Vq is a vector space over K with sum and product
with scalar given by
1 , 2 V1 V2 Vq , vi Vi ,
K, V1 V2 Vq , vi Vi , ( )(v1 , , vq ) = (v1 , , vq ).
Homework 1.5 Prove that the functions 1 q in Definition 1.1.4 are multilinear. In
other words, prove that 1 q V1 V2 Vq .
Homework 1.6 Show that
1 (i + i0 ) q =
= 1 i q + 1 i0 q
and that
1 ( i ) q = = (1 i q )
Homework 1.7 Consider the map
det : (Rn )n R
associating to each n vectors in Rn , the determinant of the matrix whose columns are those
vectors in the natural order. Show that det is an element of ((Rn ) )n . Show that if det is
decomposable, then n = 1. Show that det is skew.
Homework 1.8 In the notation of Theorem 1.1.1 give (and prove) an explicit formula for
1i1 2i2 qiq (v1 , . . . vq ) when each vi is a vector of the given basis of Vi , so vi = ei j for
some j.
[Hint: the formula will need some Kronecker symbols; if you have difficulties, try first the
case of (R2 ) (R2 ) with the canonical basis.]
Homework 1.9 Use the formula in Homework 1.8 to prove that the decomposable tensors in
Theorem 1.1.1 are linearly independent.
[Hint: write a general linear combination of them and assume that the combination produces
the zero function. Choose suitable elements in the domain to prove the vanishing of each
coefficient; every coefficient will need a different element in the domain]
Homework 1.10 1) Use the formula in Homework 1.8 to produce explicitely, for every tensor
V1 V2 Vn , a linear combination 0 of the decomposable tensors in Theorem 1.1.1
such that and 0 assume the same value on each n-tuple (v1 , . . . vn ) with the property that each
vi is a vector of the given basis of Vi .
2) Use multilinearity to conclude that = 0 .
3) Notice that, having solved this homework and the two previous homeworks, you have
written a proof of Theorem 1.1.1. Congratulations.
Homework 1.11 Prove that, assuming i V , j K,
1. 1 = 2 1 2 = 0.
12 Chapter 1. The de Rham cohomology on Kn
2. 1 2 = 2 1 .
3. (1 1 + 2 2 ) = 1 (1 ) + 2 (2 ).
4. (1 1 + 2 2 ) = 1 ( 1 ) + 2 ( 2 ).
Homework 1.12 Prove that assuming i V , j K,
1. 1 2 (v, v) = 0.
2. 1 2 (v1 , v2 ) = 1 2 (v2 , v1 )
3. 1 2 (1 v1 + 2 v2 , w) = 1 1 2 (v1 , w) + 2 1 2 (v2 , w).
4. 1 2 (w, 1 v1 + 2 v2 ) = 1 1 2 (w, v1 ) + 2 1 2 (w, v2 ).
Homework 1.13 Let qV . Prove using only the definition of qV that if (v1 , . . . , vq ) 6=
0, then the vi are pairwise distinct.
Homework 1.14 Let 1 , 1 q1 V , 2 q2 (V ), 3 q3 (V ), k R. Show that
i) 1 2 q1 +q2 (V );
ii) (1 + 1 ) 2 = 1 2 + 1 2 ; 2 (1 + 1 ) = 2 1 + 2 1 ;
iii) (k1 ) 2 = k(1 2 ) = 1 (k2 );
iv) (1 2 ) 3 = 1 (2 3 ).
v) 1 2 = (1)q1 q2 2 1 ;
Homework 1.15 Prove that all maps L are linear.
Homework 1.16 Show that L ( ) = L L .
Exercise 1.2 Show that, for each decomposable tensor (V )q different from 0, the set
{v V |(v, v, . . . , v) = 0}
Exercise 1.4 To each bilinear form (Kn ) (Kn ) associate a square matrix A so that
(v, w) = wT Av. Show that this gives an isomorphism among (Kn ) (Kn ) and gl(K, n).
Show that this induces two isomorphisms
among Sym2 (Kn ) and the space of symmetric n n matrices;
among 2 (Kn ) and the space of skewsymmetric n n matrices.
Symq (V ) q (V ) 6= {0} q 1.
Exercise 1.6 Show that (V )2 = Sym2 (V ) 2V . Is there a similar relation when q 3?.
1.2 Vector fields and differential forms 13
Exercise 1.7 Show that there is a canonical (so you cant use a basis to construct it) isomor-
phism (V ) W
= (V W ) .
2 (V W )
= 2V 2W (V W ).
Exercise 1.9 Assume V = Rn , i V . Prove that 2|1 2 (v1 , v2 )| is the area of the
parallelogram spanned by the vectors (1 , 2 )(v1 ) and (1 , 2 )(v2 ).
Exercise 1.13 Assume V = R3 . Compute explicitly the wedge product of two general
1-forms. Compare the result with the usual definition of cross product on R3 .
where the equivalence relation is the following: two functions f , g are equivalent if there
exists an open set W 3 p contained in the domain of both functions such that f|W = g|W . An
equivalence class for this relation is a germ of smooth function at p. E p is the stalk at p of
the sheaf of smooth functions.
If K = C one usually considers instead the stalk at p of the sheaf of holomorphic
functions
O p := { f : U C holomorphic |U is open and p U}/
with the analogous equivalence relation.
Note that, given a germ f E p or O p , f (p) K is well defined since all the functions in
the same equivalence class have the same value at p. On the contrary, q 6= p, f (q) is not well
defined.
14 Chapter 1. The de Rham cohomology on Kn
Definition 1.2.2 A tangent vector or derivation at p is a linear application v : E p R (in
the real case) or v : O p C (in the complex case) such that
f , g, v( f g) = f (p)v(g) + g(p)v( f )
Theorem 1.2.1 The set
xi |1 i n is a basis for Tp Kn . In particular dim Tp Kn = n.
p
We will not prove it (most students have probably already seen a proof of it in the real case, the
proof in the complex case is similar).
The next definition is the definition of vector field. Roughly speaking, a vector field on U is
the datum, for every point p U, of a tangent vector v p Tp Kn . A natural way to do that is by
writing something of the form ni=1 fi xi for some functions fi : U K: this associates to each
point p U the vector ni=1 fi (p) xi . Now we make this idea into a precise definition.
p
For sake of simplicity, in the remaining part of this section, we will restrict to the case
K = R, although everything has a complex version. We leave to the reader to state the analogous
definitions/results in the complex case.
Consider a vector field v and a function f . Then, p U, v(p) acts on the germ off atp,
giving a real number v(p)( f ). This gives a map v( f ) : U K. More precisely, if v = vi xi ,
p
f
v( f ) = vi . (1.2)
xi
From (1.2) follows that, if v and f are smooth, then v( f ) is smooth too: we have defined a map
To define the 1-forms we work with the cotangent space (Tp Rn ) , the dual of Tp Rn .
1.2 Vector fields and differential forms 15
Roughly speaking, a 1-form on U is the datum, for every point p U, of a p (Tp Rn ) . We will
write ni=1 i dxi to associate to each point p U the covector ni=1 i (p)(dxi ) p .
Definition 1.2.5 Let U be an open set in Rn . The cotangent bundle of U is T U := U Rn
Rn Rn = R2n .
p U we may consider each cotangent space (Tp Rn ) as a subset of T U by identifying
the covector ni=1 i (dxi ) p with the point (p, (1 , . . . , n )) T U: this gives a bijection
between T U and the disjoint union of all cotangent spaces (Tp Rn ) for p U.
There is natural map : T U U, the projection on the first factor. If we consider T U
as the union of the cotangent spaces, this is the map associating to every covector the point
on which tangent space it acts.
A differential 1-form on U is a map : U T U such that = idU (so p U,
(p) (Tp Rn ) ). For every differential 1-form there are functions i : U R such
that (p) = (p, (1 (p), . . . , n (p)). We will write as ni=1 i dxi : indeed for all p U,
(p) = ni=1 i (p)(dxi ) p . A differential form is smooth if it is smooth as function among
open sets of euclidean spaces. The space of the smooth differential 1-forms is denoted by
1 (U). The usual notation for its complex analogous (the space of the holomorphic 1-forms)
is 1,0 (U).
Consider a differential 1-form and a vector field v. Then, p U, (p) acts on v(p),giving
a real number (p)(v(p)). This gives a map (v) : U R. More precisely, if v = vi xi ,
p
= i (dxi ) p ,
(v) = i vi . (1.3)
From (1.3) follows that, if and v are smooth, then (v) is smooth too: we have defined a map
1 (U) X(U) C (U)
(, v) 7 (v)
We can now give an intrinsic definition of the differential q-forms. The idea is to repeat
the same game played for the 1-forms, by using q (Tp Kn ) instead of (Tp Kn ) . For sake of
simplicity we just treat the case K = R.
Definition 1.2.6 Let U be an open set in Rn . The bundle of the q-forms of U (in the complex
case: bundle of holomorphic q-forms) is q T U := U q (Rn ) Rn q (Rn ) . Note that
n
q (Rn ) is a real vector space of dimension nq , so q T U is an open set in Rn+(q) .
A differential q-form (in the complex case: holomorphic q-form) or differential form
16 Chapter 1. The de Rham cohomology on Kn
= i1 iq dxi1 dxiq .
The wedge product of smooth forms is smooth, since sums of products of smooth functions
are smooth. So we get bilinear maps
which inherits all the properties of the wedge product of alternating forms. Namely (see
Homework 1.14)
i) Let 1 , 1 q1 (U), 2 q2 (U); then (1 + 1 ) 2 = 1 2 + 1 2 ; 2 (1 +
1 ) = 2 1 + 2 1 ;
ii) Let 1 , q1 (U), 2 q2 (U), f C (U). Then ( f 1 ) 2 = f (1 2 ) = 1
( f 2 );
iii) Let 1 , q1 (U), 2 q2 (U), 3 q3 (U), then (1 2 ) 3 = 1 (2 3 ).
1.3 Operations on differential forms 17
(d f ) p : Tp Rn R
d f : U T U
d : C (U) 1 (U)
Now we extend the differential operator d : 0 1 (the one considered in the last section,
defined by the formula d f (v) = v( f )) to an operator
d : q q+1 .
Note that (Exercise 1.17) the differential of the coordinate function xi is the 1-form dxi (this
motivates the notation for the basis of the cotangent space).
Theorem 1.3.1 There is a unique linear operator d : (U) (U) of degree 1 (which
means d(q (U)) q+1 (U)), such that
i) f 0 (U), v X(U), d f (v) = v( f ).
18 Chapter 1. The de Rham cohomology on Kn
d(1 2 ) = d1 2 + (1)q1 1 d2 ;
iii) d d = 0.
If
= i1 iq dxi1 dxiq ,
1i1 <i2 <<iq n
then
Proof. The existence is easy: we just need to consider the formal expression given in the
statement, d = di1 iq dxi1 dxiq , and check that it has the required properties. We
only check iii), leaving the other checks to the reader.
By the linearity of d it is enough if we prove the statement for = f dxi1 dxiq . Then,
writing (for sake of simplicity) dxI for dxi1 dxiq and using that dxi dx j = dx j dxi
!
n
f
d(d( f dxI )) = d dxi dxI
i=1 xi
n
2 f
= dx j dxi dxI
i, j=1 x j xi
2 f
= dx j dxi dxI
i6= j x j xi
2 f 2 f
= dx j dxi dxI + dx j dxi dxI
i< j x j xi i> j x j xi
2
2 f
f
= dx j dxi + dxi dx j dxI
i< j x j xi xi x j
2
2 f
f
= dx j dxi dx j dxi dxI = 0
i< j x j xi xi x j
We need to show that every linear operator with the properties i), ii), and iii) coincides with
it.
By linearity
d = d i1 iq dxi1 dxiq .
1i1 <i2 <<iq n
Proof. We denote by Mi, j the (i, j)-entry (entry in the ith row and jth column) of the matrix
of dFp in the given bases. By definition
dFp = Mi, j ,
xj p i yi F(p)
and therefore
Mi, j = Mk, j (yi ) = dFp (yi ) = (yi F)
k yk F(p) xj p xj p
Definition 1.3.4 The differential of F is the map dF : TU TV obtained by gluing all dFp ;
note that, since F is assumed smooth, dF is smooth too. Equivalently we can define dF by
asking that v TU, f EF((v))
dF(v)( f ) = v( f F).
d(F G) = dF dG.
Comparing (1.5) and (1.6) we conclude then that we have given the same name to two
objects which are different (even if we consider both as maps from TU, the targets are
different), but substantially the same object (the difference being just ( dtd ) f (p) ). This is a
further standard abuse of notations in differential geometry.
F : q (V ) q (U);
(F ) p = dFp (F(p) ).
Conventionally, if f 0 (V ) = E (V ), then F f := f F.
We will show later that if F and are smooth, then F is smooth.
R From the analogous properties of the alternating forms all F are linear and F (1 2 ) =
F 1 F 2 , so they are ring homeomorphisms.
Moreover
(F G) = G F ;
Proposition 1.3.4 Let F : U V be a smooth function, and consider a general smooth differen-
tial qform
= i1 iq dyi1 dyiq q (V ).
1i1 <...<iq n
Then
F = (i1 iq F)dFi1 dFiq ,
1i1 <...<iq n
where Fk := yk F.
F i1 iq dyi1 dyiq =
=
1i1 <...<iq n
i1 iq F F dyi1 F dyiq .
=
1i1 <...<iq n
We have then only to check F dyk = dFk . Indeed, by lemma 1.3.3, F dyk = d(yk F) =
dFk .
Now we can write the pull-back of a form explicitly. For example, if F is the function
F(x1 , x2 ) = (x1 x2 , x12 + x22 ), then
F (y1 dy2 ) = (x1 x2 )d(x12 + x22 ) = 2x12 x2 dx1 + 2x1 x22 dx2 .
Proof. Since
F = (i1 iq F)dFi1 dFiq ,
1i1 <...<iq n
Fi
and dFi j = xkj dxk , then we obtain that F = gI dxI where all gI are sums of products of
the smooth functions (i1 iq F) and of the partial derivatives xFki .
Proof. We write
= i1 iq dyi1 dyiq ,
1i1 <i2 <<iq n
which yields
F = (i1 iq F)dFi1 dFiq .
1i1 <i2 <<iq n
and therefore
( f 1 1 + f 2 2 ) = f 1 1 + f 2 2
( f 1 1 + f 2 2 ) = f 1 1 + f 2 2
Homework 1.20 Let f C (U), = I fI dxI (U). Define f = I ( f fI )dxI . Show that
f = f = f.
Homework 1.21 Assume 1 q1 , 2 q2 (U). Then 1 2 q1 +q2 (U).
Homework 1.22 Prove that , , (U)
( ) = ( ).
Homework 1.23 Assume that , are homogeneous forms (possibly of different degree). Then
= (1)(deg )(deg )
Homework 1.24 Show that if q (U), and if v1 , . . . , vq X(U), then the function
(v1 , . . . , vq ) : U R
is smooth.
x1 x22
Exercise 1.15 Check that (x2 dx1 dx2 ) x1 x , x2 x = 2 .
1 2
Exercise 1.21 Compute explicit formulas for the differential of a general 0-form, 1-form resp.
2-form on R3 and relate the result with the usual definition of gradient, curl and divergence.
Whats the differential of a 3form?
Exercise 1.25 Let U, V be open sets of Rn and assume that they are diffeomorphic, i.e. that
there is a smooth map F : U V which is invertible and such that F 1 is also smooth. Then,
q, F : q (V ) q (U) is an isomorphism.
1 2 = (10 + d1 ) (20 + d2 ) =
= 10 20 + d1 (20 + d2 ) + d1 d2 =
= 10 20 + d(1 (20 + d2 )) + d(1 d2 )
so [1 2 ] = [10 20 ].
Then HDR (U) with the three operations (multiplication by scalar, sum, wedge product) is a
q q
graded Ralgebra with HDR (U) graded piece of degree q. Note that HDR (U) is defined for all
q Z, but it is interesting only for 0 q n since it vanishes for any other value of q.
The forthcoming exercise 1.27 shows that HDR 0 (U) only depends on the topology of U; more
precisely it counts the connected components of U. Some similar interpretations hold true also
for the other cohomology groups; we will discuss some of them later.
A straighforward but very important consequence of Proposition 1.3.6 is the following
It follows
Proposition 1.4.3 Let F : U V be a smooth map. Then there is a ring homomorphism
(V ) H (U) of degree zero (meaning F (H q (V )) H q (U)) such that, for each
F : HDR DR DR DR
closed form qDR (V ), F ([]) = [F ].
q
Here and in the following, [] is the class of in HDR (for short: the cohomology class of ).
q
Proof. We need to prove that F ([]) = [F ] is a good definition of a map F : HDR (V )
q
HDR (U).
First of all we need that F is closed, to be able to consider its cohomology class. But
is the representative of a cohomology class, so is closed, and then F is closed too by
Proposition 1.3.6.
1.4 The de Rham cohomology 25
Then we need to show that the cohomology class of F only depends on the cohomology
class of . Indeed, if [] = [ 0 ], then with d = 0 . It follows by the proposition 1.3.6
F F 0 = F ( 0 ) = F d = d(F )
[F F 0 ] = 0 [F ] = [F 0 ].
q q
Then we have a well-defined map F : HDR (V ) HDR (U). We leave to the reader the now
straightforward check that F is a ring homomorphism,
R Note that we have given the same name F to the map F : (V ) (U) and to
the induced map F : HDR (V ) H (U). This is a standard abuse of notation. Indeed,
DR
since the latter are induced by the former by the formula F ([]) = [F ()], most of the
properties of the first map are clearly shared also by the second one.
For example the formula
(F G) = G F
holds also in cohomology.
(U) and
Homework 1.25 Show that there is an isomorphism of vector spaces between HDR
L q
qZ HDR (U).
Exercise 1.26 Show that f 0 (U) = C (U) is closed if and only if it is locally constant,
i.e. p U there exists an open neighbourhood V of p, V U, such that f|V is constant.
Determine all exact form in 0 (U)
Exercise 1.27 Show that, for any U open set in Rn , dim H 0 (U) equals the number of con-
nected components of U.
1 (R).
Exercise 1.28 Use the definition to compute HDR
q
Exercise 1.29 Compute HDR (U) for all q Z when
U is a point (U = R0 );
U = R;
U R is a disjoint union of k open intervals.
Exercise 1.30 Let U,V be open sets of Rn and assume that they are diffeomorphic. Show
q
that their de Rham cohomologies are isomorphic: q HDR (U) q
= HDR (V ).
Introduction
Topological manifolds
Construction
Differentiable structures
Tangent spaces
Fibre bundles
Vector bundles
The tangent bundle
Regular values and a method to construct
manifolds
Differential forms
Pull-back and differential of forms
Orientability
Integration
Stokes theorem and applications
2.1 Introduction
Most of the students of this course have met the differentiable manifolds in the previous years of
their undergraduate studies. We will develop in this chapter this theory, considering at the same
time the complex manifolds and the real manifolds with boundary. To be precise, we will mostly
discuss the slightly more complicated real case, where the boundary can be considered, and give
indications on how to rewrite everything in the complex case.
Before setting the first formal definition, let us try to give some general idea. A topological
manifold without boundary is a topological space which is locally Euclidean: in other words
something which "locally" cant be distinguished by Rn . The surface of a sphere, S2 , is a typical
example: we know that the surface of the Earth is approximatively a sphere, locally we cant
distinguish a sphere from a plane and indeed our ancestors were convinced that the Earth was
flat.
We are interested in a slightly more general class of objects: a typical example is the closed
ball B3 = {(x, y, z) R3 |x2 + y2 + z2 1}.
B3 is not locally euclidean because of its boundary. Indeed, if we consider a point p B3 of
norm 1 there is no neighborhood of p (in the topology of B3 ) homeomorphic to an open set of
R3 . To include B3 in our class of objects we need to modify the definitions to allow a boundary.
Note that B3 may be decomposed as disjoint union of its boundary B3 and its interior B3 as
follows
B3 := {(x, y, z) R3 |x2 + y2 + z2 < 1}
B3 := {(x, y, z) R3 |x2 + y2 + z2 = 1}.
We remark that B3 is a topological manifold without boundary (is an open set of R3 !), and B3 is
the sphere S2 , therefore it is also a topological manifold without boundary, although of different
dimension. Similarly we will decompose every manifold with boundary as disjoint union of two
manifolds without boundary: its interior and its boundary.
Notation 2.1. We will denote by Rn+ the halfspace of the points of Rn whose last coordinate is
nonnegative:
Rn+ = {(x1 , . . . , xn ) Rn |xn 0}.
Similarly Rn = {(x1 , . . . , xn ) Rn |xn 0}.
A topological manifold with boundary (sometimes just topological manifold for short)
of dimension n is a topological space M which
is locally homeomorphic to Rn+ (that is: p M, U open set containing p homeomorphic
to an open set of Rn+ )1 ;
is Hausdorff;
is connected2 ;
admits a countable basis of open sets3 .
We see few examples
Example 2.1 Every open set of Rn+ or Rn is a topological manifold with boundary.
Example 2.2 Rn is a topological manifold with boundary. Indeed it is isomorphic to the open
set of Rn+ of the points whose last coordinate is strictly positive: {(x1 , . . . , xn )|xn > 0}. The
homeomorphism is the map
Example 2.3 The map expn induces homeomorphisms among any open set U of Rn and
the open set expn (U) Rn+ . Therefore every open set U Rn is a manifold with boundary
(indeed this shows that every manifold in the classical sense, as for example the spheres Sn , is a
topological manifold with boundary).
Recall that an open covering of a topological space M is a family U : = {Ui }iI of open sets
S
of M with the property that iI Ui = M.
Example 2.4 The closed interval B1 := [1, 1] R is a topological manifold with boundary
of dimension 1. B1 is connected, Hausdorff, and has a countable basis of open sets, so to prove
our statement we need only to construct a covering of M made of open sets homeomorphic to
open sets of R1+ = [0, +). The easiest choice is B1 = [1, 12 ) ( 12 , 1]: note that [1, 21 )
=
( 12 , 1]
= [0, 32 ).
"manifold" with more than a connected component; for our definitions it will not be a manifold, but a disjoint union
of manifolds.
3 Some authors remove this assumption too. We prefer to add this condition, restricting a bit the set of objects we
are considering (but keeping all examples we are interested in). If we remove this assumption, the theory becomes
much more complicated, because two important results fail in this greater generality: the existence of partitions of
unity and Whitney Embedding Theorem
2.2 Topological manifolds 29
coordinate function on Rn
ui : Rn R
(w1 , . . . , wn ) 7 wi
Each chart {(U, )} induces local coordinates (x1 , . . . , xn ) defined by
xi := ui : U R.
If you have traveled by car, you had probably to "move" from a map to another. To follow
your path you need to find the coordinates, in both maps, of the same point, your position in that
moment.
Consider for example M = B1 = [1, 1]. The example 2.4 suggests two charts (U1 , 1 ) and
(U2 , 2 ) on B1 :
U1 := [1, 12 ), 1 : [1, 21 ) [0, 23 ) given by 1 (t) = t + 1;
U2 := ( 12 , 1], 2 : ( 21 , 1] [0, 23 ) given by 2 (t) = 1 t.
The point p = 14 B1 has coordinates (coordinate: B1 has dimension 1) 54 for (U1 , 1 ) and
3
4 for (U2 , 2 ).
How do the coordinates change? For every ordered pair of charts (U , ) and (U , ) we
define the associated transition function
:= ( )|U U ( )1
| (U U ) : (U U ) (U U )
In our example U1 U2 = ( 12 , 12 ) and the transition functions 21 and 12 are easily com-
puted: 21 = 12 : ( 21 , 32 ) ( 12 , 23 ) is given by 21 (t) = 12 (t) = 2 t. These functions allow to
compute the coordinates of a point in a chart from the coordinates in the other chart: 21 ( 54 ) = 34
and 12 ( 34 ) = 54 .
Notation 2.2. The definition of is heavy, because we had to restrict the domains of all
functions to be able to compose them.
From now on we will use the following convention. Let f and g be functions such that the
image of f and the domain of g do not coincide, but are subset of a "common universe". Then by
g f we mean the composition of the restriction of f and g to the biggest possible subsets such
that the composition is possible.
With this convention, the definition "reduces" to the easier := 1 .
Similarly, if we write an inequality among two functions which do not share the same domain,
we mean that the two functions coincide on the points were both are defined.
By definition every topological manifold with boundary M may be covered by a set of charts
{(U , )}I ; in other words U := {U }I is an open covering of M.
Assume you have charts of the whole surface of the Earth, and some glue (I mean, anything
one can use to glue two sheets of paper). Then start gluing all your charts in such a way that two
points are glued if and only if they represent the same point on the Earth. You will end up with a
paper-made sphere: you have constructed something homeomorphic to the surface of the Earth,
and the drawings on it make the homeomorphism explicit.
Similarly, we can reconstruct any manifold (well, something homeomorphic to it), by taking
the images of the charts, and gluing them using the transition functions. This gives a very
concrete method to construct manifolds.
2.2.1 Construction
Take a family {D }I of open sets of Rn Rn+ or Rn , and denote by N the topological space
obtained as disjoint union a
N := D .
I
30 Chapter 2. Manifolds (with boundary)
We may finally introduce the smooth (resp. holomorphic) functions. Let M be a manifold
with atlas {(U , )}I and N a manifold with atlas {(V , )} I .
A function f : M N is smooth (resp. holomorphic) in a point p M if, given a chart
(U , ) with p U , and a chart (V , ) with f (p) V , the function f 1 is smooth
(resp. holomorphic) in (p). Note that this definition do not depend on the choice of the charts:
32 Chapter 2. Manifolds (with boundary)
here we need the assumption that all transition functions are smooth (resp. holomorphic).
A function f : M N is smooth (resp. holomorphic) if it is smooth (resp. holomorphic) in
every point p M. A diffeomorphism (resp. biholomorphism) is a smooth (resp. holomorphic)
function which is invertible and whose inverse function is smooth (resp. holomorphic) .
Example 2.6 Let M be a manifold, : U Kn a chart, D := (U).
Then U and D are open sets of the manifolds M and Kn , and therefore they have a natural
structure of manifold (see Exercise 2.4): an atlas for U is given by the single chart : U Kn ;
the differential structure of D has an atlas given also by a single chart, the inclusion i : D , Kn .
Then we can consider as function among two manifolds. It is easy to check that it is a
diffeomorphism.
An important case is given by the smooth (resp. holomorphic) functions from a manifold M
to K. In the real case denote it by
C (M) = { f : M K| f is smooth}.
In the complex case the susual notation is O(M). Note that it is a real (resp. complex) vector
space, with the operations induced by the target space R (resp. C).
Example 2.7 Let M be a manifold, : U D Kn a chart. Consider the local coordinates
xi := ui . Then xi C (U) (resp. O(U)).
R We will never consider two different differentiable structure (neither two different complex
structures) on the same topological manifold, but the student should be aware that it is
possible.
It is indeed easy to construct two different differentiable structures on S2 , but one can
prove (although this is not always easy) that the two resulting manifold are diffeomorphic.
Note that the diffeomorphisms will not be the identity: if we consider two different
differentiable structures on the source and on the target, the identity map is not smooth!
The situation in higher dimension is much more complicated. Mumford constructed 28
different differentiable structure of S7 , which give 28 differentiable manifolds which are
pairwise not diffeomorphic. We will not further investigate this problem in these lectures.
From this point of view the complex case is simpler, since it is not difficult to construct
infinitely many pairwise distinct complex structures on the "donut" S1 S1 , but we will
not show it in these lectures.
- the boundary of M is M := M \ M .
M is without boundary if M = 0.
/
Note that in particular, if U0 is not empty, then also V0 is not empty. Let now M be a manifold,
p M. Then assume that there is a chart (U, ) such that (p) (U) Rn . Then, since
every transition function is a diffeomorphism, by Lemma 2.3.1 for each other chart (V, ) with
p V , (p) Rn . It follows
Exercise 2.3 Write explicitely an atlas for S2 formed by two charts obtained considering the
stereographic projections S2 \ P R2 where P is either the north pole (0, 0, 1) or the south
pole (0, 0, 1). Write all the transition functions and show that they provide a differentiable
structure on S2 .
Consider now your charts as complex charts identifying R2 with C in the natural way.
Show that this does NOT give a complex structure on S2 . Modify these charts (in a simple
way) to get a complex atlas for S2 .
Exercise 2.4 Let M be a manifold. Then any connected open subset U M has a natural
induced differentiable structure.
Exercise 2.5 Let M be a manifold, and let (U, ) be a chart on it, D = (U). Then by
Exercise 2.4 both U and D are manifolds with the differentiable structure respectively induced
by M and by Rn+ . Show that is a diffeomorphism among them.
Exercise 2.6 Prove that if G : M N and F : N N 0 are smooth, then also F G is smooth.
Exercise 2.7 Let M, N be two real (resp. complex) manifolds, and assume M = 0.
/ Then
show that M N has a natural induced real (resp. complex) structure, by constructing an atlas
for M N using an atlas of M and an atlas of N. What goes wrong if both manifolds have a
boundary?
34 Chapter 2. Manifolds (with boundary)
Note that, given a germ f E p , f (p) R is well defined. On the contrary, q 6= p, f (q) is
not well defined.
Definition 2.4.2 A tangent vector or derivation at p is a linear application v : E p R
which is linear and such that
f , g E p , v( f g) = f (p)v(g) + g(p)v( f )
d(F G) p = dFG(p) dG p .
It follows
Theorem 2.4.1 The set xi |1 i n is a basis for Tp M. In particular dim Tp M = n.
p
Tp M which is a complex vector space of dimension n. Since (as noticed in the previous
sections) M has an induced real structure of dimension 2n, it has also a tangent space as
real manifold, also denoted (by abuse of notation) Tp M of dimension 2n.
It is important to distinguish them, for example by denoting the second as TpR M. It is indeed
tempting, since both are real vector spaces of dimension 2n, to think that there is a good
way to identify them, but there is no canonical (= independent on the coordinate choice)
way to do that. Indeed there is no natural way to connect real and complex derivations;
recall that the real derivations act on real valued smooth maps and produce real numbers,
whence the complex derivations act on complex valued holomorphic maps and produce
complex numbers.
Note that
xi : E p R is by definition given by
p
( f 1 )
f
(p) := f= ( f 1 ) = ((p))
xi xi p ui (p) ui
Why did we choose the notation xi ? Recalling that the chart (U, ) induces coordinates
x1 , . . . , xn on U by xi := ui , we note
1
(x j ) = (x j ) = (u j ) = i j .
xi p ui p ui p
We are now able to compute xfi (p) for every function f C (U) which we can explicitly
write "in coordinates near p".
This means, given a function, we choose a chart (U, ) in p and consider
the induced
coordinates x1 , . . . , xn . If we can express f as combination of the xi , since xi is a derivation
p
(linear+Leibniz rule) and xi x j = i j we can compute xi f formally as if xi were
p p
coordinates in R . For example, if f = x1 x2 , x1 f = (2x1 x2 )(p), x2 f = x12 (p).
n 2
p p
Now that we have given bases to every Tp M, so we can associate a matrix to every dFp . The
matrix can be computed by exactly the same method used for Proposition 1.3.2. The result is the
following.
Proposition 2.4.2 Let M and N be manifolds of respective dimensions n and m, and let F : M
N be a smooth function. Let p M. Choose charts (U, ) in p for M and (V, ) in F(p) for N,
and the respective associated local coordinates x1 , . . . , xn and y1 , . . . , ym .
Then the matrix of the linear application dFp respect to the bases
( ) ( )
,..., and ,...,
x1 p xn p y1 F(p) ym F(p)
By definition 2.4.4, M1 and M3 are both identity matrices (resp. m m and n n), and
therefore the matrix we are looking for equals M2 , which we have computed in proposition
1.3.2.
n n
v = vi = v j
i=1 xi p i=1 xi p
Then
v1 v1
.. ..
. = J( ) (p) .
vn vn
where J( ) (p) denotes the Jacobi matrixa of the application at the point (p).
a We recall that the Jacobi matrix of an application f in a point p is the matrix having as (i, (ith row
j)-entry
fi
and jth column) the ith component of f respect to the jth coordinate, computed at p: uj
.
p
Proof. Obviously
v1 v1
.. ..
. = M .
vn vn
for the matrix M representing the identity map of Tp M in the bases in the source and xi
p
x in the target. Since IdTp M = d(IdM ) p for the identity map IdM : M M, we can
j p
compute M by the proposition 2.4.2: M is the Jacobi matrix of the map IdM 1 = in
(p).
We could now define the smoothness of our "roughly defined" vector field using the expres-
sion in coordinates, and use this corollary to prove that our definition does not depend on the
coordinate choice. We will instead follow the longer way to define the vector bundles and use
their theory; this choice will be more convenient later.
2.5 Fibre bundles 37
Homework 2.7 Let M be a manifold, U M an open set with the differentiable structure
induced by the differentiable structure of M. Let i : U M be the inclusion map, and fix a point
p U. Prove that di p : TpU Tp M is an isomorphism.
Homework 2.8 Let M, N be manifolds, M without boundary. Consider M N with the differ-
entiable structure induced as in exercise 2.7. Then
The projections map 1 : M N M and 2 : M N N are smooth.
Consider, q N the inclusion map iq : N M N defined by iq (p) = (p, q). Similarly
consider, p M the inclusion map i p : N M N defined by i p (q) = (p, q). Prove that
both maps i p , iq are smooth.
Show that p M, q N, d(1 )(p,q) , d(1 )(p,q) are surjective whence d(i p )q , d(iq ) p are
injective
Show that p M, q N, the image of d(i p )q equals ker d(1 )(p,q) . Similarly the image
of diq equals ker d(2 )(p,q) .
Show that p M, q N, d(1 )(p,q) (diq ) p = IdTp M , d(2 )(p,q) (di p )q = IdTq N
Show that T(p,q) (M N) = d(iq ) p (Tp M) d(i p )q (Tq N).
It is a usual abuse of notation to identify Tp M with its image d(iq ) p (Tp M) T(p,q) (M N); the
identification is possible since, by Exercise 2.8, the map d(iq ) p is injective. With this abuse of
notation the last equality can be written T(p,q) (M N) = Tp M Tq N.
Homework 2.9 Prove that {(TU, d)|(U, ) is a chart for M} is an atlas for T M, by perform-
ing all necessary checks. [Hint: use the corollary of Proposition 2.4.2].
As in every category, once determined the objects, we have to determine which maps among
them we want to consider.
Definition 2.5.2 Let E, E 0 two bundles with respective bases B, B0 and respective fibres F,
F 0 , and let : E B, 0 : E 0 B be the respective projections. A morphism of bundles
among E and E 0 is a pair of continous maps f : E E 0 , g : B B0 such that g = 0 f .
If f is also an homeomorphism, we say that f is an isomorphism of bundles, and that E is
isomorphic to E 0 . A bundle is trivial if it is isomorphic to the bundle 1 : B F B.
A first remark is that the cocycle determines the bundle up to isomorphisms; this follows by
the same argument used for the similar property of the transition functions of a differentiable
structure.
38 Chapter 2. Manifolds (with boundary)
If E, B and F are all real (resp. complex) manifolds, it is usual, unless differently specified, to
move all the above definitions to the corresponding category: so all continous map are supposed
instead smooth (resp. holomorphic).
We conclude this section by an important construction, the base change, also known as fibre
product.
Definition 2.5.5 Consider two functions with the same target space f : A C, g : B C.
The fiber product of f and g, usually denoted by A C B is the subset of the product
A B of the elements which "agree on C" in the following sense:
f0
A C B /B
g0 g
f
A /C
Exercise 2.8 Show that the cylinder and the Moebius band are bundles over S1 with fibre an
open interval. Write a trivialization of both and the corresponding cocycle.
Exercise 2.9 Show that, if (E, ) is a trivial bundle, then f 1 E is trivial (for every function
f ).
Exercise 2.10 Consider the Moebius band M as bundle on S1 as in the previous exercise. Let
f : S1 S1 be defined by f (cos , sin ) = (cos 2 , sin 2 ). Show that f 1 M is isomorphic
to the cylinder as fibre bundle over S1 .
2.6 Vector bundles 39
We will always consider vector bundles that are manifolds: so real vector bundles on real
manifolds (there are real manifolds) or complex vector bundles on complex manifolds (these are
complex manifolds). Note that, in both cases, dim E = dim B + rk E = dim B + r.
For all p B the fibre Fp is mapped isomorphically to the vector space Kr by any . This
gives a vector space structure on Fp which, since we have assumed that the cocycles have values
in GL(r, K), does not depend on the choice of . So we can see a vector bundle as a way to
attach to each point of B a vector space of fixed dimension r in a "smooth (resp. holomorphic)
way".
In particular we can consider the 0 of each E p , and the map associating to each point p B
the 0 of the corresponding E p is a smooth section, the zero section s0 : B E.
Definition 2.6.2 Let E, E 0 two vector bundles on respective bases B and B0 . A morphism of
vector bundles is a morphism of bundles f : E E 0 such that p B, the map f|E p : E p
0
Eg(p) (here g is the underlying map g : B B0 ) is a linear application. If f is bijective and
f 1 is a morphism of vector bundles we will say that f is an isomorphism of vector bundles
and that E is isomorphic to E 0 as vector bundles. A trivial vector bundle is a vector bundle
isomorphic as vector bundle to a product B Kr B.
If f is injective, B = B0 and g = IdB : B B then we will say that E is a subbundle of E 0 .
We have already seen that every fibre bundle is determined by its cocycle. For vector bundles
we can be more precise. First of all we notice that for every vector bundle E over B the maps
g : U U GL(r, K) are smooth (resp. holomorphic). On the other hand one can prove
that
Proposition 2.6.1 Let B be a manifold, let U := {U }I be an open cover of B, r N. Assume
we have , I, a smooth (resp. holomorphic) map g : U U GL(r, R) (resp. GL(r, C ))
such that the set {g } verifies the three cocycle conditions. Then there is a (unique up to
isomorphism) vector bundle E of rank r over B (so a manifold of dimension r + dim B) having a
trivialization with cocycle {g }.
By Proposition 2.6.1 we can construct a vector bundle simply by giving a cocycle. This is
very helpful to lift all standard constructions in linear algebra from the category of vector bundles
to the category of vector bundles.
Direct sum. For any pair of vector spaces V,V 0 we can consider their direct sum V V 0
whose dimension is dimV + dimV 0 .
Let us then consider two vector bundles E and E 0 of respective ranks r and r0 on the same
base B. We need a vector bundle E E 0 on B of rank r + r0 such that, for every p B, the fibre
(E E 0 ) p is canonically isomorphic to E p E p0 .
To construct E E 0 we choose respective cocycles {g } and {g0 } for E and E 0 relative to
the same open cover of B (we can always do it, see Homework 2.11), and we define E E 0 by
the cocycle {g g0 } where
g 0
g g0 :=
0 g0
40 Chapter 2. Manifolds (with boundary)
It is now not difficult to use the and 0 to write isomorphisms from each (E E 0 ) p to E p E p0
not depending on the choice of (see Exercise 2.15).
Dual. Let E be a vector bundle on a base B with cocycle {g }. Then we define E to be the
bundle with cocycle {(g1 t t
) } (here stands for "transpose"). Then each fibre E p is isomorphic
to (E p ) .
Tensor product. For any pair of vector spaces V,V 0 we have introduced their tensor product
V (V 0 ) and V V 0 (see Homework 2.12) whose dimension is (dimV )(dimV 0 ).
As in the previous case, we want, for every two vector bundles E and E 0 of respective ranks
r and r0 on the same base B, to construct a vector bundle E E 0 on B of rank rr0 such that, for
every p B, the fibre (E E 0 ) p is E p E p0 .
We give its cocycle: choosing again cocycles {g } and {g0 } relative to the same cover,
we define E E 0 through the cocycle {g g0 } (Homework 2.13).
Alternating powers. For every vector space V we have defined qV .
If {g } is a cocycle for E, the cocycle for (q )E is given by the maps g : q (Kr )
q (Kr ) .
Hom. Since (Exercise 2.14) Hom(V,W ) = W V , then we define Hom(E, E 0 )
= E0 E
and isomorphisms Hom(E, E 0 )P = Hom(EP , EP0 ) follow immediately.
Quotient. If V is a subspace of V 0 we can consider the quotient V 0 /V of dimension dimV 0
dimV . Similarly, when E is a subbundle of E 0 , we want to define E 0 /E so that for all P,
(E 0 /E)P = EP0 /EP . This is simple: it is enough to notice that by definition g is the restriction
of g0 to an invariant subspace, and consider the map induced by g0 onto this quotient subspace.
Homework 2.12 Consider a vector space V and define a map e : V V , as follows:
v V, V e(v)() = (v).
Show that e is linear injective and, if we further assume that V is of finite dimension, an
isomorphism. Note that we can define, given two (or more) vector spaces V1 , V2 and vectors
v1 V1 , v2 V2 , the decomposable element v1 v2 V1 V2 (which is V1 V2 ).
Homework 2.13 Consider 4 vector spaces V1 , V2 , W1 , W2 , and maps Li : Vi V j . Show that
there is a unique linear map
L1 L2 : V1 V2 W1 W2
Exercise 2.11 Show that a line bundle E over B is trivial if and only if it has a section
s : B E which never vanishes: in other words p B, s(p) 6= 0 E p .
Exercise 2.12 Show that a vector bundle E of rank r over B is trivial if and only if it has r
sections s : B E forming, p B, a basis of E p .
Exercise 2.14 Use Proposition 2.6.1 to construct a vector bundle of rank 1 on S1 which is
not trivial.
Exercise 2.15 Let E, E 0 be two vector bundles over B, p B. Show that there are canonical
isomorphisms from (E E 0 ) p to E p E p0 .
Exercise 2.16 Let E, E 0 be two vector bundles over B, p B. Show that there are canonical
isomorphisms from (E E 0 ) p to E p E p0 , from each (q E ) p to q (E p ) , from each (E/E 0 ) p
to E p /E p0 .
So we can just define the tangent bundle as the bundle given by the cocycle (2.1).
We can then define the vector fields.
Definition 2.7.1 A vector field on a manifold M is a map v : M T M such that v = IdM .
A vector field is smooth (resp. holomorphic) if it is smooth (resp. holomorphic) as a map
among manifolds. The smooth vector fields form the vector space X(M).
For every vector field v, every chart (U, ) for M may be used to represent the vector field v on
U: v|U . If x1 , . . . , xn are the local coordinates given by the chart, there are functions vi : U R
such that p U, v(p) = vi (p) xi . We will write v|U as vi xi .
p
As in the case of Rn (resp. Cn ), the vector fields act on C (M) (resp. O(M): if v is a vector
field and f C (U) (resp. O(U)), the function v( f ) is naturally
defined by v( f )(p) := v p f . In
local coordinates, if v|U = vi xi , then v( f )(p) = vi (p)
xi f , which we shortly write (on
p
U)
f
v( f ) = vi
xi
By Exercise 2.16 it follows that, if v and f are smooth (resp. holomorphic), then v( f ) is smooth
(resp. holomorphic): we have defined a map X(U) C (U) C (U) (resp. X(U) O(U)
O(U)).
Homework 2.15 Consider the natural map : T M M defined by v T(v) M T M. Prove
that is smooth.
Homework 2.16 Prove that the given definition of smooth vector field corresponds to the naive
idea we started with. In other words, prove that v = vi xi is smooth at a point p U if and
only if i, vi is smooth at p.
42 Chapter 2. Manifolds (with boundary)
Note that by definition every point which is not in the image of F, is a regular value. Indeed,
Sards Lemma shows that the set of regular values Reg(F) is very big, and more precisely it is
an open dense subset of N.
A very important special case is the case when Reg(F) = N, that is when no point is a
critical point. This is the case of the immersions (when dim M dim N), the submersions (when
dim M dim N) and the embeddings (when moreover the map is an homeomorphism among of
M with its image).
Definition 2.8.2 Let M, N be manifolds and let F : M N be a smooth function. Then
F is an immersion if p M, dFp is injective.
F is a submersion if p M, dFp is surjective.
F is an embedding if F is an immersion and an homeomorphism among M and F(M),
where F(M) is considered with the topology induced by N.
Example 2.8 We have already seen two examples of embeddings, the inclusions M , M
and X , M when X is a connected component of M.
F 1 (x1 , . . . , xn ) = (x1 , . . . , xm ).
A couple of (not trivial) consequences of the last corollary can be used to construct manifolds.
The following holds only in the real case.
Note that, since composition of embeddings is an embedding, by Example 2.8, in the situation
of Theorem 2.8.4 every connected component of f 1 (y) is embedded in M.
The complex version of Theorem 2.8.4 is
Theorem 2.8.5 Regular Value Theorem 2. Let M be a complex manifold, and choose
a function f O(M). Let y C be a regular value of f . Then every connected component
X of f 1 (y) has a complex structure such that the inclusion X , M is an embedding and
dim X = dim M 1.
Example 2.9 The function f = xi2 C (Rn ) has only one critical point, the origin, so it
has only one critical value, zero. Theorem 2.8.4 induces then a differentiable structure on each
closed ball of positive radius.
Choosing y = 1 we obtain then differential structures on Bn and Sn1 such that the respective
inclusion maps in Rn are embeddings.
If we construct a manifold in this way, then we can represent easily the tangent spaces of the
components of N as subspaces of the corresponding tangent spaces of M.
/ Let f C (M)
Proposition 2.8.6 Let M be a manifold. In the real case we assume M = 0.
(in the complex case: O(M)), y Reg( f ). Let X be a connected component of f 1 (y) with
the differentiable structure induced by Theorem 2.8.4 (in the complex case: 2.8.5), i : X , M
the corresponding embedding and choose a point p X. Then di p is injective and di p (Tp X) =
ker d f p .
Proof. The function f i C (X) is the constant function, assuming in each point the same
value y. Therefore d f di = d( f i) = 0, so the image of di is contained in the kernel of d f :
di p (Tp X) ker d f p .
Since i is an embedding, di p is injective. Since dim X = dim M 1, di p (Tp X) has codimen-
sion 1. On the other hand, since y is a regular value, p is not a critical point, and therefore d f p
has maximal rank 1, so ker d f p has codimension 1 too. Since the first space is contained in the
second one, they must coincide.
There is a different version of the regular value theorem, which applies to manifolds with
boundary.
Theorem 2.8.7 Regular Value Theorem 3. Let M, N be real manifolds, dim N < dim M,
F : M N a smooth function, y Reg(F) Reg(F| M ). Then every connected component X
of F 1 (y) has a differentiable structure such that the inclusion X , M is an embedding, and
X = M X.
It is not difficult to show, exactly as in the other case, that the differential of the inclusion
identify Tp X with ker dFp . In particular dim X = dim M dim N.
Exercise 2.17 Construct a smooth function F C (R) such that Reg(F) is exactly the
complement of the image of F.
Exercise 2.18 Show that the map F : R R2 defined by F(t) = (cost, sint) is an immersion
and is not an embedding.
Exercise 2.19 Consider the function F : (0, 2) R2 defined by F(t) = (sint, sin 2t). Show
that it is injective immersion but it is not an embedding.
Exercise 2.20 Show that {(x12 + x22 + x32 + 3)2 16(x12 + x22 ) = 0} is a manifold without
boundary embedded in R3 . Can you recognize the underlying topological manifold?
Exercise 2.21 Prove that ki=1 u2i u u2i1 u
defines a smooth vector field on S2k1
2i1 2i
which never vanishes. We have combed the odd dimensional spheres.
As in the case of the euclidean spaces, the q-forms act on X(M)q ; that is we can see every
q-form as a map : X(M)q C (M) (resp. O(M)) as follow. For every choice of q
smooth vector fields v1 , . . . , vq , (v1 , . . . , vq ) is the function defined by p (v1 , . . . , vq )(p) :=
p (v1 (p), . . . , vq (p)).
For every q-form , every chart (U, ) for M may be used to represent the restriction of
to U |U as follows.
Letx1 , . .. , xn be the local coordinates induced by the chart. p U we have an induced
basis { xi } of Tp M. We denote by {(dxi ) p } the corresponding dual basis of (Tp M) . For
p
every multiindex (i1 , . . . , iq ), we write (dxI ) p for (dxi1 ) p (dxiq ) p .
2.10 Pull-back and differential of forms 45
R Again
F (1 2 ) = F 1 F 2 ,
and therefore F is a morphism of Ralgebras. Moreover
(F G) = G F .
In particular, if F is a diffeomorphism, then F is invertible with inverse (F 1 ) .
Note that if (U, ) is a chart, with coordinates x1 , . . . xn then (Exercise 2.24) dui = dxi and
therefore I duI = (I )dxI .
We need an operator d : 0 (M) 1 (M). For a general f 0 (M), we do not have defined
a 1-form d f yet, but we have a map
d f : TM TR
= RR
whose restrictions d f p = (d f )|Tp M : Tp M T f (p) R are linear.
46 Chapter 2. Manifolds (with boundary)
d
" (v dtd p 7 v), d f p can be con-
Definition 2.10.1 Identifying T f (p) R with R by "forgetting dt
sidered as an element in (Tp M) = 1 (Tp M) T M = 1 T M. This describes a differential
1form which we denote by d f .
f
R d f 1 (M). Indeed, using local coordinates x1 , . . . , xn , computing d f
xi = xi
p p
we deduce d f = xfi dxi , which is obviously smooth. Note that the differential of the
coordinate function xi equals dxi .
= i1 iq dyi1 dyiq q (V ).
1i1 <...<iq n
Then
F = (i1 iq F)dFi1 dFiq ,
1i1 <...<iq n
where Fk := yk F.
Proof.
!
F = F i1 iq dyi1 dyiq
1i1 <...<iq n
F i1 iq dyi1 dyiq
=
1i1 <...<iq n
i1 iq F F dyi1 F dyiq .
=
1i1 <...<iq n
We have then only to check F dyk = dFk . which follows since F dyk = dyk dF = d(yk
F) = dFk .
Definition 2.10.2 There is one case which is rather important, it is the case when F is an
embedding. In this case we will write |M for F ().
d : (M) (M),
2.10 Pull-back and differential of forms 47
d(1 2 ) = d1 2 + (1)q1 1 d2 ;
iii) d d = 0.
If (U, ) is a chart with coordinates x1 , . . . , xn and on U
= i1 iq dxi1 dxiq ,
1i1 <i2 <<iq n
then
= d(( 1 ) )
Proof. The uniqueness follows repeating word-by-word the proof of the analogous theorem 1.3.1.
To prove the existence we need to show that the local expression given for d is independent on
the choice of the chart.
Then let (U , ) and (U , ) be two charts in p. We need to show d((1 ) ) =
d((1 ) ), which may be rewritten, setting := (1 ) 1 (D ) as
d = d(1 )
which is equivalent to
d = d
which follows from proposition 1.3.6.
It follows
Corollary 2.10.3 Let F : M N be a smooth function, (N). Then
F d = dF .
Proof. Choose p M and charts (U, ) in M and (V, ) in N such that p U, F(U) V . Then
F d = F d(( 1 ) )
= ( 1 ) F d(( 1 ) )
= ( F 1 ) d( 1 ) )
= d( F 1 ) ( 1 ) )
= d( 1 ) F ( 1 )
= d( 1 ) F
= dF
48 Chapter 2. Manifolds (with boundary)
Exercise 2.24 Let (U, ) be a chart for a manifold M, and let x1 , . . . xn be the corresponding
local coordinates.
Proves dui = dxi .
F : q (N) q (M).
ZM () := {p M| p = 0}
ZX () := p X|(|X ) p = 0
1) Show that ZM () X ZX ().
2) Consider the 1form dx1 1 (R2 ). Show that ZR2 (dx1 ) S1 6= ZS1 (dx1 ).
Exercise 2.30 Consider the following two open subsets of S1 : Ui = {p S1 |xi 6= 0} for
i = 1, 2. Consider the 1form on S1 defined by
dx
x1 |U if p U1
2
1
p = p
dx 1
if p U2
x2 |U2 p
Show that this gives a well defined 1-form 1 (S1 ) which is a volume form on S1 .
Exercise 2.31 Consider a function f C (Rn ), y Reg( f ), M = f 1 (y). Prove that the
canonical bundle of M is trivial.
Hint: consider the open subsets Mi := {p M| xfi (p) 6= 0}]. Try to define i n1 (M)
2.11 Orientability 49
so that p Mi ,
2.11 Orientability
The orientability of a real manifold is an interesting geometrical property which can be defined
only for real manifolds. Indeed (as we see in the next definition) to consider it we need to be
able to distinguish "positive" and "negative" numbers.
Definition 2.11.1 A matrix A GL(n, R) preserves the orientation if det A > 0; A reverses
the orientation if det A < 0.
Let , 0 be two open subsets of Rn and let F : 0 be a smooth function. F pre-
serves the orientation if p , dFp preserves the orientation. F reverses the orientation
if p , dFp reverses the orientation.
L|Rn : Rn Rn
is a diffeomorphism.
Assume now M orientable. Let {(U , )}I be an oriented atlas for M, and consider the
atlas {(U , L )}I .
The new atlas is not orientedly compatible with the first one, since , L 1 reverses
the orientation. Therefore every orientable manifolds has at least two orientations, and it remains
only to show that every further orientable atlas {(V , )} J for M is compatible with one ot
these two.
For every point p M we choose I, J with p U V . we define
| det J( 1 ) (p) |
(p) := {1} R.
det J( 1 ) (p)
50 Chapter 2. Manifolds (with boundary)
Since both atlases {(U , )}I and {(V , )} J are oriented, (p) do not depend on
the choice of and . Moreover is smooth, therefore continous. But M is connected, {1} is
discrete, so is constant. We have then two cases: either 1 or 1.
If 1, a straightforward computation shows that {(U , )}I and {(V , )} J are
compatible. Else, 1, and similarly {(U , L )}I and {(V , )} J are compatible.
Notation 2.3. If M is an oriented manifold, we will denote by M the same manifold taken with
the other orientation, the opposite orientation.
R There is no natural way to extend the definition of orientability of the category of complex
manifolds, since we cant decide if a complex number is "positive" or "negative" in a
reasonable way. On the other hand, we know that every complex manifold of dimension
n has a natural differentiable structure of real manifold without boundary of dimension
2n, sometimes denoted as the underlying real manifold. It is then natural to ask, for every
complex manifold, if its underlying real manifold is orientable or not. This natural question
has a surprisingly simple answer.
Assume for sake of simplicity that M is a complex manifold of dimension 1, with atlas
{(U , )}I . Then the : C C are holomorphic functions in one variable. The
underlying real manifold has atlas {(U , )}I , and the transition function are
obtained by the removing the complex structure from its source and its target: in other
words = (a , b ) is exactly the map where we are considering its sources and
targets as open sets of R2 (instead of C).
By the Cauchy-Riemann relations the Jacobi matrix of is
a b
!
x x
b a
x x
a 2 b 2
whose determinant is x + x > 0. Therefore the real atlas induced by the comlex
atlas is already an oriented atlas.
A similar (although more complicated) computation works also in higher dimension,
showing that the underlying real manifold of any complex manifold is orientable, and we
will always consider it with the natural orientation obtained by considering any complex
atlas as real atlas as above.
Now we need to introduce an important and very powerful technical tool in the theory of real
manifolds: the partitions of unity.
Definition 2.11.3 Let X be a topological space. A family S := {S }I P(X) of subsets
of X is locally finite if p X there exists an open set U 3 p such that U S 6= 0/ only for
finitely many I.
The definition is for a general S := {S }I P(X), but we will only use it for families of
open sets U := {U }I T (X) P(X) (here T (X) is the topology of X).
Definition 2.11.4 Let U := {U }I be an open covering of a manifold M. A partition
of unity subordinate to U is a family of smooth functions i : M [0, 1], i varying in a
countable set of indices J, such that
a) i, the support supp(i ) := {p M|i (p) 6= 0} is compact.
b) i J, (i) I such that supp(i ) U(i) ;
c) {supp(i )}iJ P(M) is locally finite.
d) p M, iJ i (p) = 1;
2.11 Orientability 51
Note that the sum at the point d) is meaningful because, by c), it reduces to a finite sum on a
suitable small neighbourhood of every point.
We will use the next result without proving it. We only mention that the proof uses the fact
that M has a countable basis of open subsets.
Theorem 2.11.2 Let U := {U } be an open covering of a real manifold M. Then there exists
a partition of unity subordinate to U.
Proposition 2.11.3 Let M be a manifold of dimension n > 0. Then M is orientable if and only
if there exists a volume form on M, i.e. if and only if the canonical bundle is trivial (compare
Exercise 2.11).
Then we can consider the form = i i n (M). Indeed, since the support of each i is
supp i := {p M|(i ) p 6= 0} = supp i , then the family {supp i } is locally finite, and therefore
i i is locally a finite sum.
We show that, p M, p 6= 0.
First of all choose i with i (p) 6= 0. Let x1 , . . . xn be the coordinates induced by a chart
(U , ) with supp i = supp i U . Then i = i dx1 xn .
For every j 6= i, j = j dx1 xn for the coordinates x1 , . . . , xn induced by a chart
(U , ). Since dxi = dui , dxi = dui ,
and therefore, since our atlas is supposed oriented, j, j 0 such that ( j ) p = j (dx1
xn ) p . Since i (p) = i (p) > 0, p 6= 0.
() Take an atlas {(U , )}I for M such that all U are connected. We construct a further
atlas for M which is oriented, by using the same open sets: an atlas of the form {(U , )}I .
Fix I, and let x1 , . . . , xn be the local coordinates induced by the chart (U , ). Then
we may write |U = f dx1 dxn with f C (U ).
By assumption f never vanishes. Since U is assumed connected, then the function f is
either strictly positive or strictly negative. In the former case we take = ; in the latter case
we take = L for the map L introduced in the proof of proposition 2.11.1.
We show that the atlas {(U , )}I is oriented. Denoting by y1 , . . . , yn the local coordi-
nates of the chart (U , ), we write |U = g dy1 dyn with g C (U ), obtaining
g (p) > 0 for all p. Indeed, if we had f > 0, then g = f . Else f < 0, and then dyn = dxn
whence for i < n dyi = dxi . In particular f dx1 dxn = f dy1 dyn and
therefore g = f .
Arguing as before (dy1 dyn ) p = det J( ) (p) (dy1 dyn ) p , and therefore
g (p)
det J( ) (p) = g (p) 0.
52 Chapter 2. Manifolds (with boundary)
The proof of Proposition 2.11.3 shows a bit more than the statement. Fix a volume form
, a point p in M, and a chart in p, if x1 , . . . , xn are the corresponding local coordinates, then
clearly p = (dx1 dxn ) p , for some 6= 0. The proof of Proposition 2.11.3 shows that, if
we choose the chart in one oriented atlas, the sign of p does not depend neither from the chart
nor from the point, but only from the chosen of the orientation. Then we can give the following
definition.
Definition 2.11.5 Let M be an oriented manifold and let be a volume form on .
We will say that M is positively oriented respect to if for every choice of a chart, in
the given local coordinates = dx1 dxn with p (p) > 0.
Similarly we will say that M is negatively oriented respect to if for every choice of a
chart, in the given local coordinates = dx1 dxn with p (p) < 0.
The proof of Proposition 2.11.3 shows that, for each volume form n (M), one of the
two orientations of M is positively oriented respect to , the other one is negatively oriented
respect to (and positively oriented respect to ).
We conclude this section by few important definitions.
Definition 2.11.6 Let M, N be oriented manifolds, and let F : M N be a smooth map.
We say that F preserves, resp. reverses the orientation if, p M, given local coor-
dinates x1 , . . . , xm around p induced by a chart of an atlas of the orientation of M and local
coordinates y1 , . . . , yn around
F(p) induced by a chart of an atlas of the orientation of N, then
Fi
the Jacobi matrix x j (p) preserves, resp. reverses the orientation.
A similar argument shows that, if M is orientable, every connected component of its boundary
is orientable too. Anyway, the usual conventions for the induced orientation are not the natural
ones.
Definition 2.11.8 Let M be an oriented manifold, and let X be a connected component of
M. We define an orientation on X, the one induced by M, as follows.
If dim M = 0, then M = 0, / and there is nothing to do.
If dim M = 1, then M is discrete, so X is a point. To orient it we must choose a sign.
We choose the opposite sign respect to the one induced by the target of any chart in
this point. In particular, if p M, we pick an oriented chart (U, ) in M with p U:
if (U) R1 we choose the +, if (U) R1+ we choose the . This do not depend
on the choice of the chart, see Exercise 2.38.
if dim M 2 is even, we choose an atlas {(U , )}I such that I, (U ) is
open either in Rn or in Rn+ (see Exercise 2.39). Then we take on X the orientation of
the atlas {(U X, ( )|UX )}I 0 ; we ask the student to check that it is oriented in
Exercise 2.40.
if dim M 2 is odd, we take the orientation opposite to the one of the atlas {(U
X, ( )|UX )}I 0 induced by {(U , )}I of M.
Homework 2.18 The cylinder C is the quotient of [0, 1][0, 1] R2 by the equivalence relation
y [0, 1], (0, y) (1, y). We denote by : [0, 1] [0, 1] C the projection map.
2.11 Orientability 53
i) Compute all transition functions. Notice that the atlas is not oriented, but all transition
functions either preserve or reverse the orientation
ii) Prove that the cylinder is orientable by producing an oriented atlas {(Ui , i )}i{1,2,3,4} .
Exercise 2.32 Let M be a manifold and assume that there exist two charts (U1 , 1 ) and
(U2 , 2 ) such that U1 and U2 are connected, U1 U2 6= 0/ and the transition function 12
neither preserves nor reverses the orientation. Show that then M is not orientable.
Exercise 2.33 The Moebius band M is the quotient of the square [0, 1] [0, 1] R2 by the
equivalence relation y [0, 1], (0, y) (1, 1 y). We denote by : [0, 1] [0, 1] M also
the projection on this quotient.
We give an atlas for M: {(Ui , i )}i{1,2,3,4} where
if x< 32
h h i i (x,y)
U1 = 0, 32 0, 23 5 1
6 ,1 3 ,1 1 ((x,y))=
(x1,1y) if x> 56
if x< 61
h i i h (x,1y)
U2 = 0, 61 13 ,1 1 2
3 ,1 0, 3 2 ((x,y))=
(x1,y) if x> 13
if x< 32
h i i h (x,1y)
U3 = 0, 32 13 ,1 5 2
6 ,1 0, 3 3 ((x,y))=
(x1,y) if x> 56
if x< 61
h h i i (x,y)
U4 = 0, 61 0, 23 1 1
3 ,1 3 ,1 4 ((x,y))=
(x1,1y) if x> 13 .
Exercise 2.35 Show that the real projective plane P2R is not orientable, and deduce that there
is no complex structure on P2R ; in other words, no complex manifold has P2R as underlying
real manifold.
Exercise 2.36 Interpretation of the relation among a volume form and the induced orien-
tation on the manifold. Let M be an oriented manifold, (U, ) a chart in a corresponding
54 Chapter 2. Manifolds (with boundary)
oriented atlas, and let as usual x1 , . . . , xn be the induced local coordinates on U. Let be a
volume form on M.
1) Show that M is positively oriented respect to if and only if p U,
!
p ,..., > 0.
x1 p xn p
Exercise 2.37 Prove that the sign of the Jacobi matrix in Definition 2.11.6 do not depend on
the choice of the local coordinates.
Exercise 2.39 Let M be an oriented manifold of dimension at least 2. Show that there is an
atlas {(U , )}I for the chosen orientation such that I, (U ) is open in Rn or Rn+ .
Show that the previous statement fails if we suppose dim M = 1.
Exercise 2.40 Show that the atlas given for X in definition 2.11.8 is oriented.
Exercise 2.41 Consider the identity map of an orientable manifold, taking two different
orientations in the source and in the target: IdM : M M. Show that, with this choice of the
orientations, Id reverses the orientation.
Exercise 2.42 Assume that a map F : M N preserves the orientation. Prove that the map
F considered as a map F : M N or as a map F : M N, reverses the orientation.
What can be said on the map F : M N?
What if we assume instead that F reverses the orientation?
2.12 Integration
We know how to integrate smooth functions on open subsets of Rn ; the classical Riemanns
integration theory is sufficient for this class of functions.
Every idea on Rn which is sufficiently independent from the choice of the coordinates
may be lifted to the larger category of the real manifolds. Unfortunately, the integration is not
independent from this choice. Already the area of an open subsets U (which is the integral on U
of the function 1) depends on the chosen coordinates: if you "double" all coordinates the area is
2.12 Integration 55
multiplied by 2n .
The students have for sure met in their bachelor studies the following result, describing the
action of coordinate changes on integrals.
Theorem 2.12.1 Let U and V be two open subsets of Rn and let : V U be a diffeomor-
phism. Let F : U R be a smooth function with compact support. Then
Z Z
F= (F )| det J()|.
U V
Here and in the following we assume compact support only to avoid convergence problems.
Indeed, most of the things which follows may be extended to a much bigger class of functions.
The theorem 2.12.1 shows that the action of a coordinate change on an integral depends only
on the determinant of the Jacobi matrix of the coordinate change. Exercise 1.24 of chapter one
suggest then to consider n-forms where n = dim M.
We first restrict our attention to the forms with compact support, again just to avoid conver-
gence problems.
q
Definition 2.12.1 The space of qforms with compact support c (M) is the vector sub-
space of q (M)
qc (M) := { q (M)| supp is compact}
Is this definition independent from the choice of the coordinates? Not completely. The
precise result is the following
Proposition 2.12.2 Let U,V be two open subsets of Rn , nc (U), and let : V U be a
diffeomorphism.
If preserves the orientation, then
Z Z
= .
U V
Proof. Assume that preserves the orientation; in other words, assume that det J() is always
positive.
Write = Fdu1 dun . Then, by Theorem 2.12.1 and by the exercise 1.24 of chapter
one,
Z Z Z Z
= F= (F )| det J()| = ( F) det J() =
U U V V
Z Z Z
= ( F) det J()du1 dun = ( F) (du1 dun ) =
V V V
56 Chapter 2. Manifolds (with boundary)
If follows that, to have a definition of integral which is independent from the coordinates, we
have to ensure that all transition functions preserve the orientation: we have to fix an orientation.
Indeed we define in a natural way the integral of a n-form on an oriented manifold, when the
support is contained in a chart.
Definition 2.12.3 Let M be an oriented manifold of dimension n, and let nc (M). Assume
that there exists (U, ) in the oriented atlas of M such that supp U. Then we define
Z Z
:= ( 1 ) (2.3)
M (U)
Proposition 2.12.2 ensures (because we have fixed an oriented atlas) that Definition 2.12.3 is
well posed, since the right-hand term of (2.3) is independent from the choice of the chart.
More precisely, if supp U U , since preserves the orientation, then
Z Z
(1 ) = (1 ) =
(U ) (U U )
Z Z Z
= (1 ) = (1 ) = (1 )
(U U ) (U U ) (U )
To extend Definition 2.12.3 to any nc (M) we need to use the partitions of unity.
Definition 2.12.4 Let M be an oriented manifold and choose one of the corresponding
oriented atlases {(U , )}I . Choose a partition of unity subordinate to the cover U :=
{U }I . For every i N choose (i) with supp i U(i) and define i := i .
Then we define Z Z
:= i .
M iN M
Apparently the right-hand term is an infinite sum. One can prove that since {supp i } is locally
finite and supp is compact, then there are only finitely many indices such that i is not
identically 0. So all but finitely many addenda of the right-hand term are zero: it is a finiteRsum.
Anyway, at a first glance this is still not a good definition, since the formula defining M
appears to be dependent on the chosen atlas and on the chosen partition of unity. This problem is
solved by the next proposition.
Proposition 2.12.3 Definition 2.12.4 do not depend neither on the choice of the partition nor on
the choice of the atlas, but only on the orientation of M.
More precisely, if M is the same manifold taken with the opposite orientation, then
Z Z
=
M M
Proof. A partition of unity may be subordinate to many different atlases. Obviously if we change
atlas (for the same orientation) without changing the partition of unity, the i do not change, and
therefore Definition 2.12.4 do not depend on the choice of the atlas.
Consider now the general case of two different partitions of unity {i }iN and { j } jN ,
subordinate to two different atlases {(U , )} and {(U , )}.
First of all, we notice that {(U , )} {(U , )} is an atlas orientedly compatible with
both, and such that both partitions of unity are subordinate to it. So we can assume {(U , )} =
{(U , )}.
Second, we note that also the family of functions {i j }(i, j)NN is a partition of unity!
Indeed N N is countable and all other properties follow from the analogous properties of {i }
and { j }.
2.12 Integration 57
We define i j := i j . If we prove, i N,
Z Z
i = i j .
M j M
R R R
then i M i = i, j M i j , and similarly it equals j M j concluding our proof.
This is simple to prove: take a chart (U, ) containing supp i , and compute
Z Z Z
i j = j i = ( j 1 )( 1 ) i =
j M j M j (U)
Z
! Z Z
1
= j ( 1 ) i = ( 1 ) i = i .
(U) j (U) M
Definition 2.12.5 If dim M = 0, then M = p is a point, and its orientation is a sign, (p)
{} . The objects to integrate
R
are the functions F : p R, which are naturally identified
with R. Then we define M F := (p)F(p).
In the definition of partition of unity we have requested the supports of the i to be compact,
which was convenient in some point of our theories. Anyway, we notice that in the proof of
Proposition 2.12.3, we havent used the compactness of the supports of the j . It follows
that, when computing the integral of a form, we can also use a "partition of unity" with
noncompact supports.
This is important for solving Homework 2.22 and then Homework 2.23, which are very
important to compute integrals explicitly. Indeed, nobody computes integrals using directly
the definition, since partitions of unity produce functions usually very hard to integrate.
Anyway, most manifolds contains a chart whose complement is a union of one or more
embedded manifolds of smaller dimension. Then by the above mentioned homeworks,
the integral of a form does not change when we restrict to such a chart, and then we can
reduce the computation to a single "classical" integral.
R We can now define an integration theory on orientable manifolds for smooth functions as
follows.
Choose a volume form on M; we know that there is one by Proposition 2.11.3.
R R
Then, for every F Cc (M), we define M F := M F where in the right-hand term M is
taken with the positive orientation respect to .
The main example of volume form is the form du1 dun on Rn . If M = f 1 (y) Rn is
obtained by the regular value theorem, then Exercise 2.31 produces a volume form on M.
More precisely consider a function f 0 (Rn ), y Reg( f ), M a connected component of
1
f (y) Rn .
To ease the notation we write du1 duci dun for the form du1 dui1
dui+1 dun . Then by Exercise 2.31 the expression
(du1 du
ci dun ) p
p = (1)n+i f
(2.4)
dui (p)
gives a well-defined volume form on the whole M, since it does not depend on i as soon as the
denominator does not vanish, which happens for each p for at least one value of i. Be careful:
this is true only on M; the expression above gives different forms on Rn when changing i, they
differ even in any point p M, since they are different operators on (Tp (Rn ))n1 that coincide
on (Tp M)n1 .
Note however that
(du1 du
ci dun ) p
(1)n+i f
d f p = (du1 dun ) p
dui (p)
( )(1)n+i (du1 du
ci dun ) p
f
(v (1) , . . . , v (n1) )d f p (v (n) ) =
Sn n! (p) dui
( )(1)n+i (du1 du
ci dun ) p
f
(v (1) , . . . , v (n1) )d f p (vn ) =
n!
Sn1 dui (p)
d f p (vn )
n! S ( ) p (v (1) , . . . , v (n1) ) =
n1
p M.
Note that we need y Reg( f ) to ensure that we are not dividing by zero. It is now clear that
does not depend on the choice of f at least up to a sign. Indeed a change of the sign of f
changes the sign of and therefore it changes the orientation of M.
Let us see an example.
Example 2.10 Let f C (R2 ), y Reg( f ), M a connected component of f 1 (y), and assume
that M is compact: a compact closed regular plane curve. Let 1 (M) be the volume form
induced by f as in Definition 2.12.7.
Consider a regular parametrization of M, that is a surjective immersion : [0, 1] M such
that [0,1) is injective and (0) = (1). Set 0 (t0 ) := dt0 dtd p .
For later use, it will be useful to be able to integrate forms on objects which are disjoint
union of few manifolds. The natural way to do it is by summing the result on each component.
Definition 2.12.8 Let M be the disjoint union of manifolds Mi , all of the same dimension n.
Then q (M) := i q (Mi ). The support of a form q (M) is the union of the supports of
the components i q (Mi ) of . Consequently qc (M) is the set of the forms q (M)
with compact supports, and so equals i qc (Mi ) i qc (Mi ). If all Mi are oriented we
L
define Z Z
= i
M i Mi
Note that in our definition do not assume that M is a finite union of manifolds. Indeed,
we want to apply it to the boundary M = N of a manifold, which can have countably many
components, all of the same dimension and oriented by M.
Anyway, by definition a form Rqc (M) vanish on all but finitely many components Mi . In
particular the sum in the definition of M is a finite sum.
Homework 2.19 If F : M N is a diffeomorphism which preserves the orientation, and
nc (N) then
Z Z
F = .
M N
60 Chapter 2. Manifolds (with boundary)
Homework 2.20 Let M1 , M2 oriented manifolds, assume M1 = 0/ and consider the manifold
M1 M2 with the orientation induced by the orientations of the Mi . Let i : M1 M2 Mi be
the natural projections and consider two forms i dim
c
Mi (M ).
i
Prove that Z Z Z
(1 1 2 2 ) = 1 2
M1 M2 M1 M2
Exercise 2.45 Consider a parametrized plane curve : [0, 1] R2 , and assume that is an
embedding in a submanifold M of R2 as in Example 2.10. Endowe := ([0, 1]) with the
volume form pull-back of the volume form of M.
Prove that the volume of (say the length) equals 01 | 0 |.
R
Exercise 2.46 Find a form 2 (R2 ) whose restriction to S1 is the volume form induced
by f = x12 + x22 as in Definition 2.12.7.
2.13 Stokes theorem and applications 61
Proof. Consider an oriented atlas {(U , )}I for M, a partition of unity {i }iN subordinate
to {U }I , and define i := i . Then = i i and therefore
Z Z Z
= i = i .
M M i i M
Therefore if the theorem holds for each i , then it holds for . We may then assume that
supp U for an (oriented) chart (U, ).
We assume for simplicity : U Rn+ (the proof for the case in which the target space of
is Rn is almost identical).
We write
n
( 1 ) = ai (u1 , . . . , un )du1 dui1 dui+1 dun ,
i=1
for the ai some smooth functions whose compact support is contained in the open set (U) of
Rn+ . We extend these functions to functions aiR C (RRn+ ) setting them zero out of (U); this
extends ( ) to a form in c R+ . Since M d = (U) ( ) d = (U) d( 1 ) , we
1 n1 n 1
R
get
!
Z Z n
M
d =
Rn+
d ai du1 dui1 dui+1 dun =
i=1
!
Z n
=
Rn+
dai du1 dui1 dui+1 dun =
i=1
!
n n
ai
Z
=
Rn+
u j du j du1 dui1 dui+1 dun
i=1 j=1
!
n
ai
Z
= dui du1 dui1 dui+1 dun
Rn+ i=1 ui
n
ai
Z
= (1)i1 du1 dun
i=1 Rn+ ui
n
ai
Z
= (1)i1 du1 dun .
i=1 Rn+ ui
To compute this last integral we start integrating respect to the variable ui , to use the
fundamental theorem of the calculus. We are integrating on Rn+ , so all variables vary from
62 Chapter 2. Manifolds (with boundary)
to but the last one, un , which varies from 0 to +. We need to compute separately these two
cases.
n
ai
Z Z
d = (1)i1 du1 dun
M i=1 Rn+ ui
n1
an ai
Z Z
n1
= (1) du1 dun + (1)i1 du1 dun
Rn+ un i=1 Rn+ ui
Z + n1
an ai
Z Z Z
= (1)n1 du1 dun1 dun + (1)i1 dui
0 un i=1 ui
R ai R + an
Finally we note that, since all the ai have compact support, u dui = 0 and 0
i u dun = n
an (u1 , . . . , un1 , 0). Therefore
Z Z
d = (1)n an (u1 , . . . , un1 , 0)du1 dun1 . (2.5)
M
To compute M = Rn+ ( 1 ) we recall that the orientation of Rn+ coincides with the
R R
The restriction of every form dui , i < n to Rn1 is the namesake form dui . On the contrary the
restriction of the form dun to Rn1 , is the zero form! Therefore all summands vanish but the last
one (for i = n) and
Z Z
n
= (1) an (u1 , . . . , un1 , 0)du1 dun1 . (2.6)
X Rn1
The statement follows by comparing (2.5) and (2.6).
R
/ n1
Corollary 2.13.2 Let M be oriented of dimension n and M = 0, c (M). Then
M d =0
We conclude this section by showing some classical application of the Stokes theorem.
Example 2.11 Fundamental Theorem of Calculus. Take M = [a, b] R, with the natural
orientation, = F C ([a, b]).
The boundary is M = {a, b} oriented by taking the + in b and the in a. Therefore
0
R
M F = F(b) F(a). By dF = F (t)dt Stokes theorem in this case is just the fundamental
theorem of the calculus Z
F 0 (t)dt = F(b) F(a),
[a,b]
Similarly, suppose that M is an arc, that is the image of an embedding i : [a, b] Rn , and
consider the arc with the orientation making i an orientation preserving diffeomorphism. Take a
function f C (Rn ).
Then, as in the previous case
Z
d f = f (i(b)) f (i(a)).
M
R
More generally, if dim M = 1, then Mdf is the sum (with suitable signs) of the values of f
on the boundary points (if any) of M.
2.13 Stokes theorem and applications 63
Example 2.12 Green formula. Let A R2 be an open set with regular boundary (which
means that A is a manifold embedded in R2 whose interior
is A). We consider a 1form
1c (A), = P(x, y)dx + Q(x, y)dy, so d = Qx Py dx dy.
Then Stokes theorem in this case gives
Z Z
Q P
= Pdx + Qdy,
A x y
3. De Rham theory
where conventionally q (M) = 0 when q < 0 or q > dim M, and let the differential d : (M)
(M) be the linear map which coincides on each graded piece q (M) with the differential in
Theorem 2.10.2.
Definition 3.1.1 Let (M).
is closed if d = 0, i.e. if ker d.
is exact if there is (M) such that = d, i.e. if Im d.
By the property iii) of Theorem 2.10.2, every exact form is closed, and we can take the
quotient closed/exact.
Definition 3.1.2 The de Rham cohomology of M is e
[1 ] [2 ] := [1 2 ].
66 Chapter 3. De Rham theory
(M) with this further operation we get a structure of Ralgebra on H (M).
Considering HDR DR
Thats why we will denote sometimes HDR (M) as the cohomology ring or De Rham coho-
mology ring of M.
q
Note that HDR (M) is defined for all q Z, but interesting only for 0 q n since it vanishes
for any other value of q.
Most of the properties which were valid on open sets of Rn hold in this more general case,
see the homeworks 3.2, 3.3, 3.4 and Exercise 3.5.
We can develop a similar theory by considering only the forms which have compact support.
Since supp d supp the differential of a form with compact support has compact support
too. Therefore the restriction of d maps
whose graded pieces are the qth cohomology group with compact support
As in the previous case, considering Hc (M) with the further operation induced by we get a
structure of Ralgebra which allows us to denote it as the compact support cohomology
ring of M.
As in the case of the De Rham cohomology, Hcq (M) is defined for all q Z, but interesting
only for 0 q n since it vanishes for any other value of q.
Note that, if M is compact, q q (M) = qc (M) from which it follows HDR
(M) = H (M).
c
Definition 3.1.4 Let M be a manifold such that all De Rham cohomology groups are finitely
dimensional.
Then the Euler number of M is e(M) := (1)q hqDR (M).
This number has many different geometrical interpretations, the most famous being the one
in Exercise 3.19.
q (M)
Homework 3.1 Show that there are isomorphisms of vector spaces qZ HDR (M) HDR
L
q
and qZ Hc (M) Hc (M).
L
Homework 3.2 Let F : M N be a smooth function, and let q (N). Show that
if is closed, then F is closed;
if is exact, then F is exact.
q q
Homework 3.3 Let F : M N be a smooth map. There is a linear map F : HDR (N) HDR (M)
q
such that, for each closed form DR (N), F ([]) = [F ()].
3.2 Differential complexes and cohomologies 67
[1 2 ] = [1 ] [2 ]
2) State and prove the analogous result for the cohomology w1th compact support.
Exercise 3.1 Show that the restrictions to S1 of the forms xdy and xdy ydx are closed but
not exact.
Exercise 3.3 Compute the De Rham cohomology ring and the compact support cohomology
ring of the intervals [0, 1) and [0, 1].
Exercise 3.4 Show that h0DR (M) equals the number of connected components of M. Find a
similar description for h0c (M).
Exercise 3.5 Let M, N be diffeomorphic manifolds. Then their cohomologies are isomorphic
(M) H (N), H (M) H (N).
graded Ralgebras: HDR = DR c = c
Note that to give a differential complex we can equivalently give all its graded pieces
{V q }qZ and linear applications d|V q : V q V q+1 with d|V q+1 d|V q = 0. We have attached to
every manifold M two complexes: the de Rham complex (M) := q (M) and its subcomplex
L
c (M).
1F is proper if K N compact, then F 1 (K) M is compact too; note that diffeomorphisms are proper maps.
68 Chapter 3. De Rham theory
ker d : V q V q+1
H q (V ) := .
Im d : V q1 V q
q
Note that H q ( (M)) = HDR (M) and H q (c (M)) = Hcq (M).
Definition 3.2.3 An exact sequence is a complex with trivial cohomology.
In other words, an exact sequence is a complex with ker d = Im d. if you consider a
complex as a sequence
V q1 V q V q+1
then the complex is an exact sequence if and only if the image of each map coincides with
the kernel of the next one.
A short exact sequence is an exact sequence of the form
0ABC 0
where 0 stands for the 0dimensional vector space {0}. Note that it implies that the map
A B is injective and the map B C is surjective.
In other words, it is a list of linear applications f : V q W q such that the diagram below
commutes
.. ..
. .
d d
f
V q1 / W q1
d d
f
Vq / Wq
d d
f
V q+1 / W q+1
d d
.. ..
. .
For example, if F : M N is a smooth map, then the map F : (N) (M) is a chain map.
If F is also proper (see Homework 3.5) then F : c (N) c (M) is a chain map.
A special role is played by the short exact sequences of chain maps
f g
0 A B C 0
3.2 Differential complexes and cohomologies 69
.. .. ..
. . . (3.1)
d d d
f
g
0 / Aq1 / Bq1 / Cq1 /0
d d d
f g
0 / Aq / Bq / Cq /0
d d d
f
g
0 / Aq+1 / Bq+1 / Cq+1 /0
d d d
.. .. ..
. . .
whose rows are exact, and whose columns are differential complexes. Therefore in the diagram
(3.1)
all maps f , g and d are linear;
d d = 0;
all f are injective;
all g are surjective;
Im f = ker g;
d f = f d and d g = g d.
The key result is the following
f g
0 A B C 0.
Proof. We have to define the maps f , g and d in the sequence (3.2) and then prove that the
image of each map is the kernel of the next one. We only give the definitions, leaving to the
reader many little computation which may be done playing with the diagram (3.1) using the
properties listed right below it: it takes some time, but it is a very interesting computation, and
fun!
The map f is the map induced by f exactly as a map in cohomology was induced by
F : (N) (M): we define f by writing
f ([a]) = [ f (a)].
We leave to the reader to show that this is a good definition (that is: the class [ f (a)], which
apparently depends on the choice of the representative a, really depends only on the cohomology
class [a]) of a linear application. Similarly g ([b]) = [g(b)].
To define d : H q1 (C ) H q (A ) we proceed as follows. By the surjectivity of g, for every
c Cq1 we can pick an element b Bq1 such that g(b) = c. If c is a representative of a
cohomology class, then d(c) = 0 and g(d(b)) = d(g(b)) = d(c) = 0. Then d(b) ker g = Im f
70 Chapter 3. De Rham theory
and therefore there is an element a Aq such that f (a) = d(b). We define then
d ([c]) = [a]
Homework 3.7 Let f : A B be a chain map among differential complexes. Then there
exists maps f : H q (A ) H q (B ) such that f [a] = [ f (a)].
Homework 3.8 Run all details of the Proof of Theorem 3.2.1.
Exercise 3.7 Let 0 V0 Vk 0 be an exact sequence of vector spaces. Then
(1)i dimVi = 0
f g
Exercise 3.8 Let A B C be an exact sequence. Prove that
g f
C B A
3.3 Mayer-Vietoris
To apply Theorem 3.2.1 to our cohomology theories (De Rham and compact support) we need
to construct suitable short exact sequences of complexes. Recall that we have defined, for each
union of manifolds M, the differential complex (M) with differential d, and the q-th De Rham
q
cohomology group HDR (M) is its cohomology group H q ( (M)). Similarly we defined the
compact support cohomology Hcq (X) using the forms with compact support qc (X).
Definition 3.3.1 Let M be a disjoint union of manifolds, U M be an open subset with the
induced structure. Then we consider the restriction map
UM : (M) (U).
Note that, since the restriction map is a pull-back (for the inclusion), and pull-back and
differential commute, UM is a chain map.
3.3 Mayer-Vietoris 71
By the property a), both functions are locally the sum of finitely many smooth functions;
in particular fU , fV C (M). Moreover fU + fV = 1, and one can easily prove supp fU U,
supp fV V .
Let now q (U V ). Consider the form fU . We extend it to a form on V by setting
p V \U, ( fU ) p = 0. one can assume that this form (which we keep calling fU ) is smooth:
indeed this is obvious on U V , whereas for every p V \ U there is a neighbourhood of p,
namely V \ supp fU , where fU = 0, and therefore fU is smooth at these points too.
Similarly for fV : we have constructed two forms fU q (V ), fV q (U) with fU +
fV = in U V . We conclude by computing
g( fV , fU ) = fU + fV = .
Proof. It follows immediately applying Theorem 3.2.1 to the exact sequence in Theorem 3.3.1.
The same construction does not work for forms with compact support because the restriction
of a form with compact support to an open subset may have a support which is not compact.
Still, there is a construction which works.
Definition 3.3.2 Let M be a disjoint union of manifolds and let U M be an open subset.
Consider the inclusion iUM : U M smooth. Then we define
jU
M : c (U) c (M)
so that jU
M is the form which coincides with on the points of U, and is 0 elsewhere. Notice
that jM is smooth because by assumption supp compact. Note that jU
U
M is a chain map.
72 Chapter 3. De Rham theory
Proof. If you have understood the Proof of Theorem 3.3.1, this should be for you a simple
exercise, do it!
Hcq1 (M)
Hcq (U V ) Hcq (U) Hcq (V ) Hcq (M)
Hcq+1 (U V )
Exercise 3.11 Let M be a manifold, and U,V M be open subsets such that all De Rham
cohomology groups of U, V and U V are finitely dimensional.
Prove that then all De Rham cohomology groups of M are finitely dimensional and
moreover
e(U V ) + e(U V ) = e(U) + e(V ).
K : q (M R) q1 (M R)
such that
t
Z
K() := f (x1 , . . . , xn , u)du dxi1 dxiq1 .
c
Since every form in q (M R) is a sum of forms of the two considered types, there is only
a linear operator K : q (M R) q1 (M R) acting as above (in other words: we extend K
to the whole q by linearity).
We skip the proof that this definition is independent from the choice of the coordinates xi .
By linearity, we only need to check (3.3) for forms of type f dxi1 dxiq and of type
f dxi1 dxiq1 dt.
In the first case, = f dxi1 dxiq ,
and
In the second case, = f dxi1 dxiq1 dt, since pull-back and differential commute, s dt
is the differential of the function with constant value c, and therefore it vanishes. So s = 0
and (Idq (MR) s ) = Idq (MR) 0 = . Moreover
and
(d K) = d(K( f dxi1 dxiq1 dt))
Z t
=d f dxi1 dxiq1
c
Z t
=d f dxi1 dxiq1
c
Z t
f
= f dt dxi1 dxiq1 + dxi dxi1 dxiq1
i c xi
Z t
f
= (1)q1 + dxi dxi1 dxiq1 .
i c xi
Theorem 3.4.2 Extended Poincar Lemma. For every manifold M, the cohomology
rings of M and M R are isomorphic. More precisely, the maps
: H (M) H (M R)
and
s : H (M R) H (M)
are isomorphisms and s = ( )1 .
The classical Poincar Lemma, claiming that every closed form on Rn is exact, follows then
immediately.
A striking application of the extended Poincar lemma is that the cohomology do not
distinguish varieties which have the same homotopy type. To state it properly we need few
definitions.
Definition 3.4.1 Let M, N be manifolds, and let F, G : M N be smooth maps. We say that
F and G are smoothly homotopic if there exists a smooth map
H : MR N
such that
3.5 The Poincar lemma for the compact support cohomology 75
t 0, p M, H(p,t) = F(p)
t 1, p M, H(p,t) = G(p)
H is the smooth homotopy among F and G.
F = (H s0 ) = s0 H = s1 H = (H s1 ) = G .
Definition 3.4.2 Two manifolds M, N have the same homotopy type if there exist smooth
maps F : M N and G : N M such that both F G and G F are smoothly homotopic to
the identities of the respective manifold.
Corollary 3.4.5 If two manifolds have the same homotopy type, then their cohomology rings
are isomorphic.
Homework 3.10 Prove that the operator K of the Lemma 3.4.1 is well defined, i.e. that its
definition is independent on the coordinates xi .
Homework 3.11 Prove that the existence of a smooth homotopy defines an equivalence relation
on the space of smooth functions from M to N.
Exercise 3.12 Let : E B be a vector bundle. Show that the De Rham cohomology ring of
E is isomorphic to the De Rham cohomology ring of B. Compute the De Rham cohomology
rings of the interior of the cylinder and of the Moebius band.
Exercise 3.14 Compute the De Rham cohomology groups of the complex projective spaces
PnC .
Hcq (M R)
= Hcq1 (M)
76 Chapter 3. De Rham theory
Proof. Arguing as in the proof of the Theorem 3.4.2, the statement follows from the construction,
q Z, of two linear maps
e : q1 q
c (M) c (M R)
: qc (M R) q1
c (M)
e 1 (M R) be the pull back ( 0 ) (e0 (t)dt) via the projection map 0 : M R R and finally
define
e := e
Since every form in qc (M R) is a sum of two forms as above, there is only a linear operator
: q1 q
c (M) c (M R) acting on the forms above as prescribed. We skip the proof that
this definition is independent from the choice of the coordinates.
We check now the claimed properties 1), 2) and 3).
1) For e : de = d( e) = d e de = d e + 0 = e d.
For , if = f dxi1 dxiq , d = d0 = 0 and
d = (d f dxi1 dxiq )
!
f f
= dt dxi1 dxiq + xi dxi dxi1 dxiq
t i
q f
= (1) dxi1 dxiq dt
t
Z
q f
= (1) dt dxi1 dxiq
R t
= 0.
3.5 The Poincar lemma for the compact support cohomology 77
If = f dxi1 dxiq1 dt
d = d ( f dxi1 dxiq1 dt)
Z
=d f dt dxi1 dxiq1
R
Z
=d f dt dxi1 dxiq1
R
Z !
= f dt dxi dxi1 dxiq1
i xi R
Z
f
= dt dxi dxi1 dxiq1
i R xi
and
d = (d f dxi1 dxiq1 dt)
!
f f
= dt dxi1 dxiq1 dt + xi dxi dxi1 dxiq1 dt
t i
!
f
= dxi dxi1 dxiq1 dt
i xi
!
f
Z
= dt dxi dxi1 dxiq1
R i xi
Z
f
= dt dxi dxi1 dxiq1
i R xi
Since all forms are sum of forms as above, the statement follows.
2) e = ( e) = (e0 (t)( ) dt) = ( R e0 (t)dt) = .
R
Exercise 3.15 Compute the compact support cohomology of the following (non compact)
spaces
Rn
Rn+
the interior of the cylinder
78 Chapter 3. De Rham theory
S n Rm
It is not difficult to construct a good cover in every concrete case (try it wit your favourite
manifold!). Indeed
We skip the proof of this theorem, which needs some Riemannian Geometry.
Definition 3.6.2 A manifold is of finite type if it admits a good cover of finite cardinality.
Anyway, the category of manifolds of finite type is much larger than the category of compact
manifolds: all the examples of manifolds we have considered up to now are of finite type.
Proposition 3.6.3 All De Rham cohomology groups of a manifold of finite type have finite
dimension.
The idea of this proof is very important, since the same inductive procedure will be used in
many other proofs in the next sections.
Proof. Let M be a manifold of finite type and let U = {U1 , . . . ,Uk } be a finite good cover of M.
We prove the statement by induction on k.
If k = 1 then M is Rn or Rn+ , whose cohomology groups have finite dimension.
Assume then the statement true for all manifolds of finite type admitting a good cover of
cardinality strictly smaller than k.
We define U := U1 Uk1 , V := Uk . We note that
- {U1 , . . . ,Uk1 } is a good cover of U of cardinality k 1;
- {Uk } is a good cover of V of cardinality 1
- {U1 Uk , . . . ,Uk1 Uk } is a good cover of U V of cardinality k 1;
Then the statement holds for (every connected component of) U, V and U V .
By the Mayer-Vietoris exact sequence
q1 q f q g q
HDR (U V ) HDR (M) HDR (U) HDR (V )
3.7 The Cech-De Rham complex 79
we compute
Homework 3.12 Take your favorite manifold and construct a good cover of it.
Homework 3.13 Construct a (connected) manifold not of finite type, and compute its cohomol-
ogy groups for both the De Rham cohomology and the compact support cohomology.
Homework 3.14 State and prove the analogous of Proposition 3.6.3 for the cohomology with
compact support.
.. .. .. .. ..
.O .O .O .O .O (3.4)
d d d d d
K 0,4
/ K 1,4 / K 2,4 / K 3,4 / K 4,4 /
O O O O O
d d d d d
K 0,3
/ K 1,3 / K 2,3 / K 3,3 / K 4,3 /
O O O O O
d d d d d
K 0,2
/ K 1,2 / K 2,2 / K 3,2 / K 4,2 /
O O O O O
d d d d d
K 0,1
/ K 1,1 / K 2,1 / K 3,1 / K 4,1 /
O O O O O
d d d d d
K 0,0
/ K 1,0
/ K 2,0
/ K 3,0
/ K 4,0
/
with differential
D := + (1) p d, (3.5)
in the sense that D is defined as the only linear operator D : K n K n+1 such that for each
K p,q K n , D = + (1) p d K p,q+1 K p+1,q K n+1 .
Lemma 3.7.1 K is a differential complex.
Proof. The only nontrivial check is D2 = 0, which indeed is the point where the sign (1) p in
(3.5) becomes relevant.
It is clearly enough if we prove DD = 0 for any K p,q . And indeed in this case
q
Definition 3.7.2 Given any double complex {K p,q } we will denote by HD (K ) the q-th
cohomology group of the associated differential complex.
U0 p := U0 U1 . . . U p
and, 0 i p, the inclusion maps (please forgive the several natural abuse of notations from
now on)
i : U0 i1 i i+1 p , U0 i1 i+1 p
induced (via pull-back) maps
0 p1 : q (U0 p1 ) q (U0 p )
0 <<k
: C p1 (U , q ) C p (U , q )
3.7 The Cech-De Rham complex 81
Lemma 3.7.2 = 0
This key Lemma is the reason for the choice of the sign ((1)i i ) in the definition of
few lines above. If you change the definition just by removing the (1)i , this Lemma fails.
To introduce the relations among the Cech-De Rham complex and the De Rham complex,
we have to give the following
Definition 3.7.4 An augmented double complex is given by a double complex K p,q with
differentials d, , a differential complex Aq with differential d and linear maps r : Aq K 0,q ,
such that rd = dr; we will represent it by the diagram
.. .. .. .. .. ..
.O .O .O .O .O .O (3.6)
d d d d d d
which is commutative and whose rows and columns are differential complexes.
We will say that the augmented double complex (3.6) (resp. the double complex (3.4))
has exact rows if all the rows in (3.6) (resp. (3.4)) are exact sequences.
Of course, if an augmented double complex has exact rows, the "smaller" double complex {K p,q }
has exact rows too. Note that, by the form we chose for the diagram (3.6), if it has exact rows
then all r are injective.
Lemma 3.7.3 Assume to have an augmented double complex with exact rows as in (3.6). Then
K n with D = 0, 0 K 0,n and 00 K n1 such that = 0 + D00 .
82 Chapter 3. De Rham theory
= 0 + . . . + k
D = 0 + . . . + k D = 0 + . . . + (k1 d)
and since (k1 d) K k1,nk+1 we can apply the inductiv hypothesis to D and
conclude
D = 0 + D00 = 0 + D(00 + ).
r : H q (A ) HDq (K );
ii) if the augmented double complex has exact rows, then all r are isomorphisms.
Proof. i) follows by the commutativity of the diagram (3.6), since then r is a chain map among
the complexes A and K .
To prove ii), we show first the surjectivity and then the injectivity.
By lemma 3.7.3 every class in HDq (K ) is represented by some K 0,q . The assumption
D = 0 is then equivalent to the vanishing of both d K 0,q+1 and K 1,q . The latter
vanishing = 0 implies, by the exactness of the rows, that there exists a Aq such that ra = .
The former vanishing d = 0 and the commutativity of the diagram (3.6) give then rda = 0
which implies, by the injectivity of r, da = 0. So we can consider [a] H q (A ) and by definition
r [a] = [ra] = []. We have proved the surjectivity of r .
Assume now r [] = 0. Then r = D K 0,q . Arguing as in the proof of Lemma 3.7.3
we can assume K 0,q1 and then D K 0,q = 0. The latter condition gives by the
exactness of the rows the existence of Aq1 such that r = . Then
r = D = d = dr = rd,
We now construct an augmented double complex joining, for each open cover U of a
manifold M, the Cech-De Rham complex {C p (U , q )} with the De Rham differential complex
3.7 The Cech-De Rham complex 83
{q (M)} as follows
.. .. .. ..
.O .O .O .O (3.7)
d d d d
0 / 3 (M) r / C0 (U , 3 ) / C1 (U , 3 ) / C2 (U , 3 ) /
O O O O
d d d d
0 / 2 (M) r / C0 (U , 2 ) / C1 (U , 2 ) / C2 (U , 2 ) /
O O O O
d d d d
0 / 1 (M) r / C0 (U , 1 ) / C1 (U , 1 ) / C2 (U , 1 ) /
O O O O
d d d d
0 / 0 (M) r / C0 (U , 0 ) / C1 (U , 0 ) / C2 (U , 0 ) /
O
where the linear maps r : q (M) C0 (U , q ) = q (U ) are given by the usual restriction
maps.
Proposition 3.7.5 (3.7) has exact rows.
Proof. Notice that, if U = {U,V } is a covering made by exactly two open sets, the statement is
exactly Theorem 3.3.1. This proof follows indeed exactly the same line of the proof of Theorem
3.3.1, starting from a partition of unity subordinate to U . We leave the details to the reader.
By Proposition 3.7.4 and Proposition 3.7.5, the cohomology of the Cech-De Rham double
complex equals the De Rham cohomology. Can we use it to compute the De Rham cohomology?
The answer is yes.
The idea here is to exchange the role of the rows and of the columns. Indeed, we can
"augment" a double complex by adding a row at the bottom. Repeating the same proof we obtain
that, if all columns are exact, the cohomology of the double complex (which is invariant by the
operation of exchanging rows and columns) is isomorphic also to the cohomology of the added
row. We define now the differential complex which will provide us of the "additional" row.
Definition 3.7.5 Let U be an open covering of a manifold M. The Cech complex of the
constant sheaf R relative to U , is the differential complex C (U , R) whose spaces are
so one copy of R for each connected component of each nonempty intersection: note that
C p (U , R) is contained C p (U , 0 ).
The differential : C p1 (U , R) C p (U , R) is the restriction of the above defined
: C p1 (U , 0 ) C p (U , 0 ) (see Exercise 3.17).
84 Chapter 3. De Rham theory
.. .. .. ..
.O .O .O .O (3.8)
d d d d
0 / 3 (M) r / C0 (U , 3 ) / C1 (U , 3 ) / C2 (U , 3 ) /
O O O O
d d d d
0 / 2 (M) r / C0 (U , 2 ) / C1 (U , 2 ) / C2 (U , 2 ) /
O O O O
d d d d
0 / 1 (M) r / C0 (U , 1 ) / C1 (U , 1 ) / C2 (U , 1 ) /
O O O O
d d d d
0 / 0 (M) r / C0 (U , 0 ) / C1 (U , 0 ) / C2 (U , 0 ) /
O O O O
d d d
0 / C0 (U , R) / C1 (U , R) / C2 (U , R) /
O O O
0 0 0
Note that all good covers are acyclic, and therefore acyclic covers always exist.
Proof. (3.8) has exact rows by Proposition 3.7.5 and exact columns by Remark 3.7. Appllying
twice Proposition 3.7.4 we obtaine then that bith cohomology group equals the q-th cohomology
group of the Cech-De Rham double complex, and therefore they are equal.
Note that the augmenting column (the De Rham complex) does not depend on the cover,
and the augmenting row has nothing to do with differential forms, so the cohomology of the
double complex (assuming the cover acyclic) does not depend on both things. This shows that
indeed the De Rham cohomology can be computed without using the differential forms (which
is a rather surprising conclusion for these notes). Indeed one can show (and we are not far from
that) that the De Rham cohomology is a topological invariant.
Homework 3.15 Complete the proof of Proposition 3.7.5.
Exercise 3.18 Choose a cover of S1 made by two connected open subsets, construct the Cech
complex of the constant presheaf R relative to it, and compute explicitely its cohomology.
Do the analogous computation with a cover made by three connected open subsets, and
finally with a cover made by four connected open subsets.
Exercise 3.19 Let S be a compact manifold of dimension 2, and assume that S is topologically
union of f polygons (possibly with different number of sides) such that the intersection of
two polygons is either empty or an edge of both.
Let e be the total number of edges (then we count every edge that belongs to two polygons
only once), and v the number of vertices (counted in the analogous way). Prove that
e(S) = f e + v.
State and prove a similar result in higher dimension.
The Poincar duality
The orientation covering
The Knneth formula
The Poincar dual of a closed submanifold
The Lefschetz number
Orientable vector bundles
The Thom class
Transversality and the intersection form
The Lefschetz fixed point formula
The Euler class and the Euler number
A /B /C /D /E
fA fB fC fD fE
A0 / B0 / C0 / D0 / E0
and assume that both rows are exact sequence and that the "external" vertical maps fA , fB , fD
and fE are isomorphisms. Then also fC is an isomorphism.
The proof is similar to other proofs of homological algebras in these notes: we leave it to the
reader.
Theorem 4.1.2 Poincar duality. Let M be an oriented manifold with M = 0.
/ Then
there exists, q, isomorphisms
q
PM : HDR (M) Hcnq (M)
such that Z
q
[] HDR (M) [] Hcnq (M) PM ([])([]) = .
M
Proof. By sake of simplicity, we prove the statement only for manifolds of finite type. We first
show that the map PM : H q (M) Hcnq (M)
R
given by the formula PM ([])([]) = M
88 Chapter 4. The Poincar duality and its applications
is well defined; in other words, we show that M depends only on the classes [] H q (M)
R
= + d( ) d(( + d) )
ZM ZM ZM
= + ( ) (( + d) )
ZM M M
=
M
PU PV
H q1 (U) H q1 (V ) / Hcnq+1 (U) Hcnq+1 (V ) (4.1)
PUV
H q1 (U V ) / Hcnq+1 (U V )
PM
H q (M) / Hcnq (M)
PU PV
H q (U) H q (V ) / Hcnq (U) Hcnq (V )
PUV
H q (U V ) / Hcnq (U V )
here
- the column at left is the Mayer-Vietoris exact sequence for the cohomology of M = U V ;
- the column at right is the dual of the Mayer-Vietoris exact sequence for the cohomology
with compact support of M = U V .
The dual of an exact sequence is an exact sequence (cf. Exercise 3.8), so both columns of
the diagram are exact sequences. By induction the maps PU PV and PUV are isomorphism.
Therefore, if there is a choice of the signs so that the diagram commutes, we can apply the five
lemma and conclude that also the map PM is an isomorphism, which is our goal. We need then
only to prove that there is a choice of the signs in the diagram (4.1) making it commutative.
4.1 The Poincar duality 89
We have to check the commutativity of four squares; the one at the bottom is
PU PV
H q (U) H q (V ) / Hcnq (U) Hcnq (V )
PUV
H q (U V ) / Hcnq (U V )
We check it, by taking a general element ([1 ], [2 ]) H q (U) H q (V ) and by computing the
two images of it in Hcnq (U V ) ; the one from the "top" way (through Hcnq (U) Hcnq (V ) )
and the one from the "bottom" way (through H q (U V )).
Top:
Z Z
([1 ], [2 ]) 7 ([1 ], [2 ]) 7 1 1 + 2 2
U V
Z
7 [] 7 (2 1 )
UV
Bottom:
Then the bottom square commutes for a suitable choice of the signs. The same proof shows
that the same holds also for the top square.
We study the commutativity of the square
PM
H q (M) / Hcnq (M)
PU PV
H q (U) H q (V ) / Hcnq (U) Hcnq (V )
in a similar way.
Top:
Z
[] 7 [] 7
M
Z
U V
7 ([1 ], [2 ]) 7 ( jM 1 + jM 2 )
M
Bottom:
[] 7 ([|U ], [|V ])
Z Z
7 ([1 ], [2 ]) 7 1 + 2
U V
Since clearly M ( jU V
R R R
M 1 + jM 2 ) = U 1 + V 2 also this square commutes for
a suitable choice of the signs.
90 Chapter 4. The Poincar duality and its applications
PM
H q (M) / Hcnq (M)
Here we need the coboundary maps of a long exact sequence induced by the short exact sequences
of Mayer-Vietoris, and then the function fV in the proof of Theorem 3.3.1.
Bottom:
([]) 7 [ jUV UV
M d( fV )] = [ jM (d fV )]
Z
7 [] 7 d fV
UV
Top:
Z
[] 7 [] 7
UV
Z
7 [] 7 d( fV )
UV
The Poincar duality has many interesting consequences. We discuss some of them.
Proof. We know that H 0 (M) = R is the space of constant functions and Theorem 4.1.2 for q = 0
gives first of all that Hcn (M) , and therefore also Hcn (M), have dimension 1.
Theorem 4.1.2 gives also an explicit isomorphism, the map
Z
H (M) 3 c 7 [] 7 c Hcn (M)
0
M
This gives for each c a different (as PM is an isomorphism) linear map among two vector spaces
of dimension 1, Hcn (M) and R. Since for c = 0 it is the zero map, for c = 1 it is a different map,
and a linear map among vector spaces of dimension 1 is either zero or an isomorphism.
This corollary allows the definition of the degree of a proper map (see Homework 3.5),
generalization of the degree of a polynomial.
Let now M, N be oriented manifolds of the same dimension n,and let F : M N be a smooth
proper map. By the properness assumption, the support of the pull-back of a form with compact
support is still compact. Then the pull-back induces a map F : Hcn (N) Hcn (M).
The Poincar duality implies Rthat both spaces have dimension 1, and more precisely the
linear maps M : Hcn (M) R and N : Hcn (N) R are isomorphisms. Composing F with these
R
Z
deg F := F
M
The argument above shows that this definition do not depend on the choice of . In the next
proposition we show that it is an integer which counts (in some sense) the cardinality of the fibre.
Recall that by Sards Lemma a smooth map among manifolds of the same dimension has always
at least a regular value.
Proposition 4.1.4 Let F : M N be a smooth proper map among oriented manifolds of the
same dimension, and let q N be a regular value of F. Then
deg F = (p)
pF 1 (q)
Proof. Since q is regular, then p F 1 (q), dFp is invertible and then (p) is well defined. It
follows that F 1 (q) is discrete: by the properness of F, F 1 (q) is also compact, and therefore
finite. Then pF 1 (q) (p) is a finite sum of 1s and 1s, an integer.
We write F 1 (q) = {p1 , . . . , pk }. By the local diffeomorphism theorem, there are open
neighborhoods Ui of pi such that F|Ui is a diffeomorphism onto a neighborhood of q. By
restricting the Ui we may assume that they are pairwise disjoint and that i F(Ui ) = V for a fixed
open neighborhood of q. We may also assume that V is contained in a chart (V 0 , ) inducing
local coordinates x1 , . . . , xn .
Finally, we may also assume, up to shrinking V , that F 1 (V ) = Ui . Indeed, if this were
S
false, there would be a sequence {zi } in M \ Ui such that {F(zi )} converges to q. Then, since
S
{ f (zi )} {q} is compact, its preimage is a compact containing the sequence {zi }. Therefore, up
to passing to a subsequence, {zi } converges to some z M and by continuity of F, F(z) = q, so
z is one of the pi . In particular {zi } intersects Ui , a contradiction.
We choose a nonnegative function 0 6= f Cc (N) with supp f V . Then N f dx1 xn 6=
R
R If F is an holomorphic proper map among complex manifolds of the same dimension, then
we can consider F as a smooth map among real oriented manifolds without boundary. In
this case the Cauchy-Riemann equations force (p) = 1, and therefore the cardinality of
the preimage of a regular value equals the degree and the map, and therefore does not
depend on the chosen regular value.
92 Chapter 4. The Poincar duality and its applications
This is not true in general for a smooth map among real oriented manifolds (not obtained
by an holomorphic map as above). For example, the map f : R R given by f (x) = x3 x
has two critical values, diviving Reg(F) in three connected components. The reader may
check that this map has degree 1 but the cardinality of the preimage of a regular value may
be in this example either 1 (a single point with = 1) or 3 (two points with = 1 and
one point with = 1). Anyway, we notice that by Proposition 4.1.4 the cardinality of
each fibre has the same parity of the degree (and indeed in the example all fibres have odd
cardinality), and is not smaller than | deg F|.
Exercise 4.3 Compute the De Rham cohomology groups of a torus with g holes (the Riemann
surface of genus g).
Exercise 4.4 Compute the De Rham cohomology groups of a torus with g holes minus n
points.
Exercise 4.5 Compute the De Rham cohomology groups of a torus with g holes minus n
small open discs pairwise disjoint (be careful, this manifold has a boundary).
Exercise 4.6 Let P be a real polynomial of degree d, and consider it as a map from R to
itself.
1) Prove that P is smooth.
2) Prove that P is proper if and only if d > 0.
3) Prove that if d is even, then the degree of P as a smooth function (cf. Definition 4.1.1)
is 0.
4) Prove that if d is odd, then the degree of P as a smooth function is 1.
Exercise 4.7 Let P be a complex polynomial of degree d, and consider it as a map from R2
to itself.
1) Prove that P is smooth.
2) Prove that P is proper if and only if d > 0.
3) Use the Cauchy-Riemann equations to prove that, if p is not a critical point, then
(p) = 1.
4) Use Proposition 4.1.4 to prove that deg F 6= 0.
5) Use 4) to deduce that P is surjective (this is the fundamental theorem of algebra).
6) Prove that the degree of P as a smooth function equals its degree as a polynomial, d.
4.2 The orientation covering 93
Proof. We consider the natural projection map : S2 P2R and the antipodal map A : S2 S2
defined by A(p) = p. Note that
A A = IdS2 ;
A is a diffeomorphism which reverses the orientation: this follows easily from A =
(IdB3 )| B3 .
all fibres of have cardinality two, pair of points exchanged by A: in particulary A = ;
is a local diffeomorphism: p S2 , d p is an isomorphisms;
Pick any 2form 2 (P2R ). Consider := 2 (S2 ). Then A = A =
( A) = = : in other words is A-invariant. Since A reverses the orientation
Z Z Z Z
= A = = 0.
S2 S2 S2 S2
Then, by corollary 4.1.3, is exact. Pick 0 such that d 0 = , and average it respect to A
0 0 0 +A 0 d 0 +A d 0
by defining := +A2 . Note that d
= d 2 = 2 = and is Ainvariant.
We use the A-invariance of to define a form such that = as follows. For each p
P choose an open neighborhood U of p small enough to have 1 (U) disjoint union of two open
2
We have then seen that the Poincar duality does not extend to P2 . Indeed, we will see that it
does not extend to any nonorientable manifold. In the proof of Proposition 4.2.1, a key role was
played by S2 , and the maps A and . We give now an analogous of that for every manifold.
Definition 4.2.1 Let V be a finite dimensional real vector space. Then we will say that
two bases of V are orientation equivalent if the determinant of the corresponding base
change matrix is positive. This is an equivalence relation partitioning the bases of V in two
equivalence classes, the two orientations of V .
:= {(p, o)|p
Definition 4.2.2 Let M be a manifold. The orientation covering of M is M
M, o is an orientation of Tp M} with the following topology and differentiable structure.
We will denote by : M M the projection (p, o) = p. Let {(U , )}I be the
maximal atlas of M. Every chart (U , ) gives local coordinates x1 , . . . , xn ; x1 , . . . , xn
determines, for each p U an orientation of Tp (M). This gives a subset, say V , of M, such
that maps V bijectively onto U . We take on M the topology generated by the V ; note
94 Chapter 4. The Poincar duality and its applications
It follows the following interesting theorem, which gives the promised simple criterion for
orientability.
Proof. When M is orientable, this is just Corollary 4.1.3. The proof of the other case is obtained
by the proof of Proposition 4.2.1 by substituting S2 with the orientation covering M of M; we
have seen in the remark above that M and the maps A : M M and : M M have all the
necessary properties.
Exercise 4.8 Compute the De Rham cohomology ring of the real projective plane.
Exercise 4.9 Compute the De Rham cohomology rings of all real projective spaces PnR =
Sn /x x.
Exercise 4.10 Compute the De Rham cohomology ring of the Klein bottle R2 / where the
equivalence relation is given by (x0 , y0 ) (x1 , y1 ) (x0 x1 , y0 (1)x0 x1 y1 ) Z2 .
We can also take (compare Homework 2.13 of Chapter 2) the tensor product of two maps.
Definition 4.3.2 Let V1 ,V2 ,W1 ,W2 be vector spaces on the same field K, and let, i {1, 2},
fi : Vi Wi be a linear map. Then
f1 f2 : V1 V2 W1 W2
is the only linear map such that (v1 , v2 ) V1 V2 , ( f1 f2 )(v1 v2 ) = f1 (v1 ) f2 (v2 ).
fi1 fi fi+1
Ai1
Ai
is exact too.
Theorem 4.3.2 Knneth formula. Let M, N be manifolds of finite type, and assume
M = 0.
/
k (M N) L
Then, k, HDR p q
= p+q=k HDR (M) HDR (N), and the isomorphism is given by
p q p+q
the maps HDR (M) HDR (N) HDR (M N) mapping [] [] to the class [1 2 ].
Here 1 : M N M and 2 : M N N are the natural projections.
Proof. Let U := {U1 , . . . ,Uh } be a finite good cover of M. We prove the statement by induction
on h.
If h = 1, then M = Rn and
p q
(N)
M
0
HDR (M) HDR (N) = HDR (Rn ) HDR
k k
= HDR (N)
p+q=k
where the last isomorphism is given (identifying as usual HDR 0 (Rn ) with R identifying the class
of a costant function with the constant) by [] 7 [ ]. Then the statement for h = 1 claims
that the map HDRk (N) H k (N Rn ) mapping [] to [ ] ( : N Rn N being the standard
DR
projection) is an isomorphism: this is the Poincar lemma.
Assume then h > 1. Arguing as in the proof of Proposition 3.6.3 we can find two open sets
U, V such that U V = M and U, V and U V have finite good covers of cardinality strictly
smaller than h: by induction, we may assume that the statement holds when substituting any of
the three to M.
96 Chapter 4. The Poincar duality and its applications
Fix two integers p and q and consider the following diagram of linear maps
KU KV
p1
(HDR p1
(U) HDR q
(V )) HDR (N) / H p+q1 (U N) H p+q1 (V N) (4.2)
DR DR
KUV
p1
HDR (U q
V ) HDR (N) / H p+q1 ((U V ) N)
DR
KM
p
HDR q
(M) HDR (N) / H p+q (M N)
DR
KU KV
p
(HDR p
(U) HDR q
(V )) HDR (N) / H p+q (U N) H p+q (V N)
DR DR
KUV
p
HDR q
(U V ) HDR (N) / H p+q ((U V ) N)
DR
where
- the column at right is the Mayer-Vietoris exact sequence induced by the decomposition
M N = (U N) (V N);
- the column at left is the exact sequence obtaining tensoring the Mayer-Vietoris exact
sequence induced by the decomposition M = U V by H q (N) as in Lemma 4.3.1;
- the maps KX are defined by KX ([] []) = [1 2 ].
We fix k and consider all diagrams (4.2) for p, q with p + q = k; they have the same right column.
They induce a diagram
p1 p1 q / H k1 (U N) H k1 (V N)
p+q=k (HDR (U) HDR (V )) HDR (N)
L
DR DR (4.3)
p1 q / H k1 ((U V ) N)
p+q=k HDR (U V ) HDR (N)
L
DR
p q / H k (M N)
p+q=k HDR (M) HDR (N)
L
DR
p p q / H k (U N) H k (V N)
p+q=k (HDR (U) HDR (V )) HDR (N)
L
DR DR
p q / H k ((U V ) N)
p+q=k HDR (U V ) HDR (N)
L
DR
such that
- the columns are exact sequences;
- by the inductive hypothesis, the first two horizontal maps and the last two horizontal maps
are isomorphisms.
A simple computation shows that the diagram (4.2) commutes. Then (4.3) commutes and the
statement follows by the five lemma.
The natural generalization of the costruction "cartesian product" is the "fibre bundle", which
we have defined in Chapter 1. Indeed, the product of two manifolds is a trivial bundle (in two
4.3 The Knneth formula 97
Homework 4.3 Show that the definition 4.3.2 is well posed by showing that there is a linear map
f1 f2 : V1 V2 W1 W2 such that (v1 , v2 ) V1 V2 , ( f1 f2 )(v1 v2 ) = f1 (v1 ) f2 (v2 ),
and that this map is unique.
Homework 4.4 Prove Lemma 4.3.1
Homework 4.5 Prove that there is a canonical isomorphism
(A B) C
= (A C) (B C).
Exercise 4.11 Compute the De Rham cohomology groups of (S1 )k , and compare the result
with Pascals triangle.
Exercise 4.12 Compute the De Rham cohomology ring of (S1 )k , and compare the result
with Pascals triangle.
Exercise 4.13 Prove that Sm1 Sn1 is diffeomorphic to Sm2 Sn2 if and only if {m1 , n1 } =
{m2 , n2 }.
Exercise 4.14 Let M1 , . . . , Mk be manifolds without boundary of finite type. Prove that
e(M1 Mk ) = e(Mi ).
i
Exercise 4.16 Let : E B be a fibre bundle with fibre PrR on a manifold B of finite type
without boundary.
Prove that if r is even then the map : HDR
(B) H (E) is a ring isomorphism.
DR
in (Hck (M)) nk
= HDR (M).
nk
Definition 4.4.1 By Poincar duality, there is a unique cohomology class [] HDR (M)
R nk k
representing S ; in other words there is a unique [] HDR (M) such that [] Hc (M)
Z Z
= .
S M
We will say that [] is the Poincar dual or the closed Poincar dual of S.
Z Z Z
(F 1 i) = i (F 1 ) = (F 1 ) S =
S S M
Z Z
1
= F ((F ) S ) = F S .
M M
So PM0 is injective. Since its source and its target are by assumption finitely dimensional, and
of the same dimension by Poincar duality, PM0 is an isomorphism.
So, if we assume M of finite type (or, slightly weaker assumption, just that all its de Rham
cohomology groups are finitely dimensional), then we can exchange the role of de Rham
cohomology and compact support cohomology in the discussion above.
Definition 4.4.2 Let M be an oriented manifold of dimension n without boundary, and assume
that all cohomology group are finitely dimensional. Let S be a compact oriented manifold of
dimension k without boundary embedded in M.
By Poincar duality, there is a unique cohomology class [] Hcnk (M) representing S ;
R
Of course, if M is compact, there is no distinction among the Poincar dual and the compact
Poincar dual. The proof of Proposition 4.4.1 gives in this case the following analogous statement.
Proposition 4.4.4 Let M, S be oriented manifolds without boundary of respective dimension
n, k. Assume S compact, M of finite type, and let i : S , M be an embedding. Set [S ] for the
compact Poincar dual of S.
Let F : M M be a diffeomorphism.
100 Chapter 4. The Poincar duality and its applications
Proof. Consider S as a compact manifold embedded in the manifold W . Then S has a compact
Poincar dual in Hcnk (W ); choose a representative 0 nk 0
c (W ) of it. Since has compact
0 to a smooth form nk (M) vanishing on M \W . Since
R
support, we can extend c S =
0 =
R R
W M , is a representative of the compact Poincar dual of S in M and its
support is contained in W .
Exercise 4.17 For the following manifolds M and for every q Z find embedded submani-
q
folds whose Poincar duals form a basis of HDR (M), and compact embedded submanifolds
whose compact Poincar duals form a basis of Hcq (M).
- Rn ;
- Rn \ {0};
- Sn ;
- the torus S1 S1 (this is harder).
Exercise 4.18 Show that if S is the boundary of a closed orientable manifold T embedded in
M, then its Poincar dual is 0.
In this section we will show that, if we further assume M compact orientable and without
boundary, there is a surprising relation among L and the fixed points of f .
Consider the manifold M M with the natural orientation. Denote the projections on the
two factors M M M respectively by 1 and 2 , so that p, q M 1 (p, q) = p, 2 (p, q) = q.
Consider the diagonal := {(p, p)|p M} M M; (1 )| = (2 )| is a diffeomorphism
of onto M. Let us compute the Poincar dual of in M M.
4.5 The Lefschetz number 101
Since by assumption HDR (M) is finite dimensional, we can choose a basis { } of H (M).
i DR
We can assume that each i is an homogeneous form of degree qi := deg i .
Poincar duality sets an isomorphism among H (M) and its dual, so that there is a further
basis
R
{i } which is dual to {i } in the R
sense of Poincar. In other words, deg i = n qi ,
M i i = 1, and, if qi = q j , i 6
= j M i j = 0. Note that if qi 6= q j , deg(i j ) 6= n, so
we cant integrate i j on M.
Lemma 4.5.1 Let M be a compact oriented manifold without boundary, fix a basis {i } of
(M ) and consider its dual (respect to Poincar duality) basis { } of H (M ).
HDR i i DR i
Then the Poincar dual of the diagonal in M M is
= (1)qi 1 i 2 i .
Z Z Z Z
1 k 2 l = 1 k 1 l = 1 (k l ) = k l = (1)qk (nqk ) kl ,
M
Proof.
Z Z
=
MM
Z
n
= (1)
ZMM
= (1)n (1)q 1 i 2 i
i
Z
= (1)n+qi 1 i 2 i
Z
= (1)deg i i f i .
M
It follows that the Lefschetz number is related to the fixed points of f ; indeed, if it does not
vanish, there is at least a fixed point somewhere!
We can say more: the Lefschetz number counts, in some sense, the fixed points of f . To
understand in which sense we need some more tools, which we will explain in the next sections.
L( f ) = (1)q hq (M)
Exercise 4.21 Let f : M M be a smooth map from a compact oriented manifold without
boundary to itself. Show that hdim M ( f ) is the multiplication by deg f .
Exercise 4.22 Show that the antipodal map of Sn preserves the orientation if and only if n is
odd.
4.6 Orientable vector bundles 103
Exercise 4.23 Show that every holomorphic map from P1C to itself has at least a fixed point.
Exercise 4.24 Let T be complex torus, that is a complex manifold of dimension 1 whose
underlying real manifold is S1 S1 .
Let f : T T be a biholomorphism such that
f f = id
f has no fixed points
Prove that H 1 ( f ) is the identity.
[Hint: compute first its trace, and then its eigenvalues!]
Hcq (E)
= (HDR (E))
n+rq
= (HDR (B))
n+rq
= Hcqr (B).
104 Chapter 4. The Poincar duality and its applications
Homework 4.8 Show that an orientable vector bundle on an orientable manifold is an orientable
manifold.
Exercise 4.26 Show that the normal bundle of Sn1 in Rn (by the usual embedding) is trivial.
Exercise 4.28 Show that, if M is orientable, then the tangent bundle T M M is an orientable
bundle.
dim M T M M
is an orientable bundle.
Exercise 4.31 Let M be an oriented manifold, and let S be a manifold embedded in M. Show
that T M |S is orientable
Exercise 4.32 Let S, M be oriented manifolds, and assume that S is embedded in M. Show
that NS|M is orientable
4.7 The Thom class 105
R
Note that if rcv (E), p B, supp |E p is compact. Therefore, if E is oriented, then
E p is well defined. In this case we define the integration along the fibre as the maps
q+r q
: Hcv (E) HDR (B)
: q+r q
cv (E) (B).
To conclude that it induces a map in cohomology we only need the usual condition: we have to
show that is a chain map.
Proposition 4.7.1 d = d
Proof. We just need to write explicitely d and d in the two cases = f (xi ,t j )
dti1 . . . dtis and and = f (xi ,t j ) dt1 . . . dtr , and check the equality in both cases.
We leave this check to the reader.
Please note that, if the bundle is the trivial rank 1 bundle B R R, then the map
coincides exactly with the namesake map considered in the proof of the Poincar Lemma for the
cohomology with compact support.
We will later need the following result
Lemma 4.7.2 [Projection formula] If : E B is an oriented vector bundle, q (B),
0
qcv (E), then
( ) = .
Moreover, if B is oriented of dimension n, supp is compact and r + n = q + q0 , then
Z Z
=
E B
Proof. The first statement is local on B. Since every bundle is locally trivial, it is enough if we
prove it for the trivial bundle U Rr U where U is a chart with coordinates x1 , . . . , xn . Then,
setting t1 . . . ,tr for the vertical coordinates, by linearity it is enough if we prove our statement in
the two cases = f (xi ,t j ) dti1 . . . dtis , s < r and and = f (xi ,t j ) dt1 . . . dtr .
106 Chapter 4. The Poincar duality and its applications
This is a straightforward computation: more precisely in the first case both ( ) and
vanish, whereas in the second case both are equal to
Z
f dt1 dtr .
The second statement, comparing the value of two integrals, is global. We consider a trivialization
{ } related to a cover {U } made by charts, and a partition of unity i subordinate to it. Setting
i = i then i = i i and
Z Z
= i ,
E i E|U
(i)
Z Z
= i
B i U (i)
and it suffices to prove E|U (i) i = U (i) i . In other words, we can assume that the
R R
bundle is trivial and the base is a chart. We can now conclude by two explicit computations as in
the previous case.
Integration along the fibres provides a generalization of the Poincar lemma for forms with
compact support as follows
An useful property of the Thom class is its good behavior respect to the direct sum of line
bundles.
4.7 The Thom class 107
Proposition 4.7.5 Let E, F be two oriented vector bundles over the same base B and consider
the vector bundle E F. Consider the natural projections E : E F E, F : E F F. Let
0
E rcv (E), F rcv (F) be representatives of the respective Thom classes of E and F. Then
E E F F
is a representative of the Thom class of E F.
Proof. First of all we show that E E F F is closed. Indeed
d(E E F F ) = dE E F F E E dF F =
= E dE F F E E F dF = 0 0 = 0.
Moreover
Z Z Z Z
E E F F = E E F F = E F = 11 = 1
(EF) p E p Fp Ep Fp
If R and S are transversal dim R + dim S dim M. We will be mainly interested in the case
dim R + dim S = dim M: in this special case the transversality condition can be more clearly
written as Tx M = Tx R Tx S. One can easily prove
Proposition 4.8.1 If R and S are transversal submanifolds of M, then every connected component
of R S has a structure of manifold embedded in R, in S and in M so that x R S, Tx (R S) =
Tx R Tx S as vector subspaces of Tx M. In particular
By the proof of Proposition 4.8.1 one easily deduce that the intersection of transversal
manifolds behaves well respect the orientations. In other words, if R and S are transversal in M,
and all the three manifolds are orientable, then every component of R S is orientable too. As in
other similar situations, since there are two possible orientations, it is convenient to choose once
and for all the orientation induced from the orientations of R, S and M.
Let us start with the case dim R + dim S = dim M. In this case R S is a discrete set, each
point x R S is a component and to orient it we have to choose a sign. We choose it as follows:
we pick an oriented basis v1 , . . . , vr of Tx R and an oriented basis v01 , . . . , v0s of Tx S and
if v1 , . . . , vr , v01 , . . . , v0s is an oriented basis of Tx M we choose the sign +;
else we choose the .
The general case is the natural generalization of this idea: the induced orientation on R S is
the one such that, x R S
if v1 , . . . , va is an oriented basis of Tx (R S)
completed it to an oriented basis v1 , . . . , va , va+1 , . . . , vr of Tx R
and to an oriented basis v1 , . . . , va , v0a+1 , . . . , v0s of Tx S
then v1 , . . . , vr , v0a+1 , . . . , v0s is an oriented basis of Tx M,
This gives an orientation of Tx (R S) which only depends on the orientations of Tx R, Tx S and
Tx M, and then is globally "coherent", producing an orientation on R S.
R Note that R S and S R have the same orientation if and only if (dim M dim R)(dim M
dim S) is even. In particular if M is a compact complex manifold and R, S are manifolds
holomorphically embedded in M transversally, then R S = S R as real oriented manifolds.
Indeed one can prove that in this case R S has a structure of complex manifold holomor-
phically embedded in R, S and M, and the orientation we have obtained is exactly the one
induced by this complex structure.
Note that v0a+1 , . . . , v0s map to a basis of (NR|M )x , giving a natural orientation on the normal
bundle. Following the same criterion, the natural orientation on the normal bundle NS|M is
given by the images of va+1 , . . . , vr1 , (1)(ra)(sa) vr (can you see where do (1)(ra)(sa)
come from?), while the natural orientation on the normal bundle NRS|M is given by va+1 , . . . , vr ,
v0a+1 , . . . , v0s which is the same orientation of v0a+1 , . . . , v0s , va+1 , . . . , vr1 , (1)(ra)(sa) vr . So we
have an isomorphism of oriented vector bundles
It follows that the wedge product is the Poincar dual of the intersection of transversal manifolds
in the following sense:
4.8 Transversality and the intersection form 109
Theorem 4.8.2 Let M, R, S be oriented manifolds without boundary with R and S compact
and transversal in M. Then it holds the following relation among the compact Poincar duals
of R, S and R S (this last with the induced orientation)
RS = R S
Proof. (Sketch) From the proof of the tubular neighbourhood theorem (which we did not discuss)
we can choose a tubular neighbourhood TR of R in M such that the isomorphism among TR and
NR|M maps TR S onto (NR|M )|RS . We choose an analogous tubular neighbourhood TS = NS|M
of S in M.
We set T := TR TS and denote by R : T R TS R the map induced by the bundle
structure NR|M R. Similarly we define S : T S TR R.
Up to shrink the tubes a bit we can assume that T is a tubular neighbourhood of R S in M,
so, by (4.4), T
= (NR|M )|RS (NS|M )|RS , in such a way that R resp. S are the projections
on the two factors. In particular the first factor coincides with R TS , and the second factor
coincides with S TR .
By the characterizing property of the Thom class, we can choose a representative R of the
Thom class of NR|M such that (R )|T = S (R )|ST . Similarly, we can choose a representative
S of the Thom class of NS|M such that (S )|T = R (S )|RT . Then, by Proposition 4.7.5, a
representative for the Thom class of NRS|M is
:= S (R )|ST R (S )|RT = R S .
Let us now consider the case dim R + dim S = dim M. In this case (even if R and S are not
transversal), R S dim
c
M (M), and we can then integrate it.
Definition 4.8.2 If R and S are compact oriented manifolds without boundary embedded in an
oriented manifold without boundary M of dimension dim R+dim S we define the intersection
number of R and S to be Z
RS = R S
M
R Note that:
R S = S R unless both submanifolds have odd codimension in M, in which case
R S = S R.
If R and S are transversal, then by Theorem 4.8.2, R S = pRS p where p equals
1 or 1 according to the orientation of p. In particular R S Z.
Exercise 4.33 Let M be a compact complex manifold and R, S are manifolds holomorphically
embedded in M, transversal. Show that R S equals the cardinality of R S (and therefore is
nonnegative).
Exercise 4.34 Compute the De Rham cohomology rings of the complex projective spaces
PnC .
Exercise 4.37 The first Hirzebruch surface F1 . Consider P1C P2C with coordinates
((t0 : t1 ), (x0 : x1 : x2 )), F1 P1C P2C defined as {t0 x1 = t1 x0 }.
1) Show that F1 is a complex manifold of complex dimension 2 embedded in P1C P2C ;
2) Show that the formula f (t0 : t1 ), (x0 : x1 : x2 )) = ((t0 : t1 ), (t0 x2 : t1 x2 : t0 x0 + t1 x1 ))
defines a (real) smooth map f : F1 F1 of degree 1.
Consider E P1C P2C defined as {x0 = x1 = 0}. Then
3) Show that E is a complex manifold of complex dimension 1 embedded R
in F1 .
4) Show that the self intersection (as submanifold of F1 ) E E = F1 E E (so E
HDR2 (F )) equals 1.
1
Hint: Use 2) to answer 4)
is surjective (so invertible), where I is the identity matrix with n rows and n columns, and J( f ) p
is the Jacobi matrix of f in the given coordinates. By a standard Gauss elimination transversality
is the equivalent to the nonvanishing of the determinant of the matrix
I I
J( f ) p I 0
which happens if and only if J( f ) p I is invertible, that is if and only if 1 is not in the spectrum
of d f p .
Note that the condition on the eigenvalues is necessary to the definition, since otherwise
det(IdTp M d f p ) = 0.
Theorem 4.9.2 Lefschetz fixed point formula. Let M be a compact oriented manifold
without boundary, and let f : M M be a smooth map such that, p Fix( f ), 1 is not an
eigenvalue of d f p . Then
L( f ) = p
p
Proof. The assumption on the eigenvalues of d f p ensured that and are transversal. Then,
each point p in comes with an induced orientation, which is, by the proof of the Lemma
I I
4.9.1, the sign of det . By elementary operations
J( f ) p I
I I I I n I I
det = det = (1) det =
J( f ) p I J( f ) p I 0 0 J( f ) p I
I I
= det = det (I J( f ) p ) .
0 I J( f ) p
112 Chapter 4. The Poincar duality and its applications
Please note that it follows that L( f ) Z, which cant be deduced directly from the definition.
If M is a complex manifold and f is holomorphic, det(IdTp M d f p ) cant be negative by the
standard argument we have already used several times. Therefore in this special case, always
under the assumption that 1 be not in the spectrum of d f p , the number of fixed points of f equals
exactly L( f ).
Exercise 4.38 Let f be a diffeomorphism from S2 to itself such that for every fixed point of
f , 1 is not an eigenvalue of d f p .
Show that, if f preserves the orientation, then f has exactly 2 fixed points. Construct
an example of a preserving orientation diffeomorphism f : S2 S2 with exactly two fixed
points.
Exercise 4.39 Let f be a biholomorphism of PnC such that for every fixed point of f , 1 is not
an eigenvalue of d f p .
Show that f has n + 1 fixed points. Construct an example of a biholomorphism of PnC
with n + 1 fixed points.
q
and the result follows since the number of i in each HDR (M) equals its dimension.
As the Leftschetz number is an obstruction to the existence of smooth self maps of M without
fixed points, the Euler number is an obstruction to the existence of vector field without zeroes on
M. More precisely
Proof. The vector field is a smooth section of the tangent bundle, so it is an embedding v : M
T M, and the condition about the zeroes ensures v(M) s0 (M) = 0, where as usual s0 denotes
the zero section. We write v for the compact Poincar dual of v(M), and 0 for the compact
4.10 The Euler class and the Euler number 113
Poincar dual of s0 (M): v and 0 are both elements in Hcdim M (T M) which equals, M being
compact, Hcvdim M (T M).
As in the case of the Lefschetz fixed point formula, the statement can be refined by consid-
ering the case when s0 (M) and v(M) intersect trasversally (in that case we say that the vector
field has nondegenerate zeroes). In this assumption the same argument shows that e(M) equals a
sum on the zeroes of v of 1s and 1s, where the sign is orientation of the point as transversal
intersection of s0 (M) and v(M).
More generally, Let v X(M) be a vector field with only isolated zeroes; in other words
we are assuming that s0 (M) (M) is a discrete set. This is then a condition weaker than the
trasversality of s0 (M) (M). Then one defines an integer i(v) p , called the index of the vector
field v at p which is in {1, 1} if and only if the intersection at p is transversal, and then prove
following the same ideas (with some work) the following
Theorem 4.10.3 Hopfs Theorem. Let v X(M) be a vector field with only isolated
zeroes. Then
e(M) = i(v) p
p|v(p)=0
We do not give here the definition of the index., neither we prove the Hopfs Theorem.
Exercise 4.40 Let M M be the diagonal. Show that the tangent bundle of is
isomorphic to the normal bundle of in M M.
Exercise 4.41 Prove that every compact orientable manifold of odd dimension has Euler
number zero.
Exercise 4.42 Construct a compact manifold with Euler number zero and one with Euler
number different from zero for every possible even dimension.
Exercise 4.44 Show that every real projective space of even dimension cannot be combed
(Warning: you cant use Hopf theorem on the real projective plane, as it is not orientable!).