Analysis on Manifolds

Luis Florit ([email protected], office 404)

Version: 20201116.2250
Download the last version from here: http://luis.impa.br/aulas/anvar/aulas.pdf
Bibliography: [Tu], [Sp], [Le], [Ha], [Hi], [Hr]...

Clickable index

1. Manifolds 17. Stokes Theorem, version 1.0

2. Differentiable functions 18. Stokes Theorem, version 2.0
3. The moduli space 19. The de Rham cohomology
4. Quotients 20. Homotopy invariance
5. The tangent space 21. Integrating cohomology: degree
6. Submanifolds 22. Application: Winding number
7. Tangent and vector bundles 23. The birth of exact sequences
8. Partitions of unity 24. Complexes
9. Orientation 25. The Mayer–Vietoris sequence
10. Differential 1–forms 26. The Euler characteristic
11. Multilinear algebra 27. Mayer–Vietoris: compact support
12. k–forms and tensor fields 28. Mayer–Vietoris for pairs
13. Orientation and n–forms 29. Application: Jordan’s theorem
14. Exterior derivative: VIP!! 30. Poincaré duality
15. Manifolds with boundary 31. The de Rham Theorem
16. Integration (Riemann) 32. References

1
§1. Manifolds
We want to extend calculus: object needs to be locally a vector
space. Example: Sn.
Topological space, neighborhood, covering.
Countable basis.
Hausdorff (T2).
REM: Countable basis and Hausdorff are inherited by subspaces.
Locally Euclidean Topological space: charts and coordinates.
Dimension, notation: dim M n = n.
Topological manifold = Topological space + Locally Euclidean +
Countable basis + Hausdorff.
Examples: Rn, graph, cusp. Not a manifold: ‘ × ’ (⊂ R2).
Compatible (C ∞–)charts, transition functions, atlas (C ∞).
Example: Sn.
Differentiable structure = maximal atlas.
From now on, for us: Manifold = differentiable manifold = smooth
manifold = Topological manifold + maximal (C ∞) atlas.
Examples: Rn, Sn, U ⊂ M n, GL(n, R), graphs, products.

§2. Differentiable functions between manifolds

Definition, composition, diffeomorphism, local diffeomorphism.
Examples: Function from/to a product; every chart is a diffeo
with its image.
Partial derivatives, Jacobian matrix, Jacobian.
Lie Groups, examples: Gl(n, R), S1, S3.
Right and left translations: Lg , Rg for g ∈ G.

2
§3. The moduli space
2
As you know, Rn and the set of square matrices Rn×n are iso-
morphic as vector spaces. This means that, although they are
different as sets, they are indistinguishable as vector spaces:
every inherent property of vector spaces is satisfied by both.
In fact, the dimension is the only vector space property that
distinguishes between vector spaces (of finite dimension).
Now, regard M := R as a topological manifold, and N := R
as a smooth manifold. Consider the map τ : M → N given
by τ (t) = t3. Since τ is a homeomorphism, the topologies and
therefore the sets of continuous functions on M and N agree:
C 0(M ) = C 0(N ). On the other hand, since τ is a bijection, there
is a unique differentiable structure on M such that τ is a diffeo-
morphism, that is, the one induced by {τ } as an atlas. Let M̂ be
M with this differentiable structure. Now, although M̂ = N as
sets (and as topological manifolds), M̂ 6= N as smooth manifolds,
since τ is not even an immersion when we regard on M = R the
standard differentiable structure of R. In fact, F(M̂ ) 6= F(N ).
However, τ : M̂ → N is a diffeomorphism by definition (hence
F(M̂ ) = {g ◦ τ : g ∈ F(N )}), and thus, by the above discussion,
as smooth manifolds they should be indistinguishable! Huh????
Answer: As a general fact in math, when studying a mathemat-
ical structure as such, we should distinguish them only up to
the isomorphism of the category. That is, we should not really
study the set Mn of differentiable n-manifolds, but its moduli
space Mn/∼, where two manifolds are identified if they are dif-
2
feomorphic. So we finally obtain M̂ ∼ N , as we got Rn ∼ Rn×n.

3
In fact, every topological manifold of dimension n ≤ 3 has a dif-
ferentiable structure, which is also unique (in the above sense).
Yet, in 1956 John Milnor showed that the topological 7-sphere S7
has more than one differentiable structure! We now know exactly
how many smooth structures exist on Sn... except for n = 4
for which almost nothing is known. See here. (Don’t worry, you will
understand more of this Wiki article by the end of the course).

§4. Quotients
Exercise: Show that on any topological space quotient there is a unique minimal topo-
logical structure, called quotient topology, such that the projection π is continuous (i.e.,
the final topology of π). But the quotient of a manifold is not necessarily a manifold...

Examples: Möbius strip, R2/Z2, [0, 1]/{0, 1} = S1.

Open equivalence relations: X has countable basis ⇒ X/∼ has,
and {(x, y) ∈ X × X : x ∼ y} is closed ⇒ X/∼ is Hausdorff.
Example: RPn.
A properly discontinuous action ϕ : G × M → M satisfies:
1) ∀p ∈ M, ∃ Up ⊂ M such that (g·Up)∩Up = ∅, ∀g ∈ G\{e},
2) ∀p, q ∈ M in different orbits, ∃ Up, Uq ⊂ M such that
(G · Up) ∩ Uq = ∅ (this is necessary to ensure Hausdorff!).
In this situation, M/ ∼ (= M/ϕ) is a manifold.
Exercise: Consider S3 as the unit quaternions, and define the map P : S3 → SO(3)
by Pu x = uxu−1 , where x ∈ R3 is identified with the imaginary quaternions. Prove that
this map is well defined, a homomorphism and a 2-1 surjective local diffeomorphism.
Conclude that SO(3) ∼
= S3 /{±I}.

4
 §5. The tangent space
Germs of functions: Fp(M ) = {f : U ⊂ M → R : p ∈ U }/ ∼
TpM , x : Up ⊂ M n → Rn chart ⇒ ∂x∂ i |p ∈ TpM , 1 ≤ i ≤ n.
Differential of functions ⇒ chain rule.
f local diffeomorphism ⇒ f∗p isomorphism ⇒ dimension is pre-
served by local diffeomorphisms.
Converse: Inverse function Theorem (it has to hold!).
Since every chart x is a diffeomorphism with its image and since
x∗p(∂/∂xi|p) = ∂/∂ui|x(p) ∀1 ≤ i ≤ n,
then { ∂x∂ 1 |p, . . . , ∂x∂ n |p} is a basis of TpM ⇒ dim TpM = dim M .
Local expression of the differential.
Curves: speed, local expression.
Differential using curves: every vector is the derivative of a curve.
REM: TpRn = Rn. Therefore, if f ∈ Fp(U ), v ∈ TpM , then
f∗p(v) = v(f ).
Differential of curves, and computation of differentials using curves.
Immersion, submersion, embedding. Rank.
Examples: projections and injections in product manifolds.
Identification of the tangent space of a product manifold:
Tp M × Tp 0 M 0 ∼
= T(p,p0)(M × M 0).
Definition 1. The point p ∈ M is a critical point of f : M → N
if f∗p is not surjective. Otherwise, p is a regular point. The point
q ∈ N is a critical value of f if it the image of some critical point.
Otherwise, q is a regular value of f (in particular, q ∈ N, q 6∈ Im (f ) ⇒ q is a
regular value of f ).

5
 §6. Submanifolds
Regular submanifolds S ⊂ M , adapted charts ϕS .
Codimension. Topology.
Examples: sin(1/t) ∪ I; points and open sets.
The ϕS give an atlas of S.
Differentiable functions from and to regular submanifolds.
Level sets: f −1(q). Regular level sets.
Examples: Sn, SL(n, R): use the curve t 7→ det(tA) !!
Exercise: S ⊂ M is a submanifold ⇐⇒ ∃ covering C of S such that S ∩ U is a
submanifold of U , for all U ∈ C.

Theorem 2. If q ∈ Im (f ) ⊂ N n is a regular value of

f : M m → N n, then f −1(q) ⊂ M m is a regular submanifold
of M m with dimension m − n.

Proof: Let p ∈ M m with f (p) = q and local charts (x, U ) and

(y, V ) in p and q. We can assume that y(q) = 0, f (U ) ⊂ V and
that span{f∗p( ∂x∂ i |p) : i = 1, . . . , n} = Tq N . Define ϕ : U → Rm
by ϕ = (y ◦ f, xn+1, . . . , xm). Then, since ϕ∗p is a isomorphism,
∃U 0 ⊂ U such that x0 = ϕ|U 0 : U 0 → Rm is a chart of M m in p.
Moreover, since y ◦ f ◦ x0−1 = πn, we have that f −1(q) ∩ U 0 =
{r ∈ U 0 : x01(r) = · · · = x0n(r) = 0}. Therefore, x0 is an adapted
chart to f −1(q).
Exercise: Adapting the proof of Theorem 2, prove the following: Let f : M m → N n a
function whose rank is a constant k in a neighborhood of p ∈ M . Then, there are charts
in p and f (p) such that the expression of f in those coordinates is given by

πk := (x1 , . . . , xm ) 7→ (x1 , . . . , xk , 0, . . . , 0) ∈ Rn .

Conclude from this the normal form of immersions and submersions as particular cases.

6
 Exercise: Conclude for the previous exercise that, if f has constant rank = k in a
neighborhood U of f −1 (q) 6= ∅, then U ∩ f −1 (q) is a regular submanifold of M m with
dimension m − k.

Example: f : GL(n, R) → GL(n, R), f (A) = AAt has constant

rank n(n + 1)/2 (since f ◦ LC = LC ◦ RC t ◦ f ∀C) ⇒ O(n) is a
submanifold of dimension n(n−1)/2 (no needed for constant rank: enough
to see that I is a regular value of f thought the Im (f ) ⊂ Sim(n, R)).

REM: Since “having maximal rank” is an open condition, if a

function f is an immersion (or a submersion) at point p, then it
is an immersion (or a submersion) at a neighborhood of p.
SL(n, R), SO(n), O(n), S3, U (n),... are all Lie groups.
Immersed and embedded submanifolds. Figure 8.
Identify: p ∈ S ⊂ M ⇒ TpS ⊂ TpM ; S ⊂ Rn ⇒ TpS ⊂ Rn.
Exercise: Read (and understand!) the proof of Sard’s Theorem (see here: aqui).

§7. Tangent and vector bundles (see [Zi])

Topological and differentiable structure of T M .
π : T M → M . Vector fields over M :
X (M ) = {X : M → T M : π ◦ X = IdM }.
Differentiability, module structure of X (M ).
Vector fields on M ∼ = Derivations on M :
D(M ) = {X ∈ End(F(M )) : X(f g) = X(f )g + f X(g)}.
Lie bracket: X (M ) is a Lie algebra: [ · , · ] is bilinear, skewsym-
metric and satisfy Jacobi identity.
7
Given f : M → N ⇒ f -related vector fields: Xf . Ex.: X|U .
Xi ∼f Yi ⇒ [X1, X2] ∼f [Y1, Y2] ⇒ [X|U , X 0|U ] = [X, X 0]|U .
Fields along f : local expression.
Integral curves, local flux and Fundamental Theorem ODE.
Vector bundles, local trivializations, transition functions. T M .
Trivial vector bundle, product vector bundle.
Whitney sum of of vector bundles.
Pull-back of vector bundles: f ∗(E).
Bundle maps, isomorphism. Example: f∗ : T M → T N .
Sections. Frames. Differentiability.
Exercise: a vector bundle is trivial if and only if exists a frame global.
Cotangent bundle: T ∗M , {dxi, i = 1, . . . , n}.
General bundles and G-bundles. Reduction.

§8. Partitions of unity

Support of functions. Bump functions.
Global extensions of locally defined C ∞ fields and functions.
Partitions of unity subordinated to coverings.
Existence of partitions of unity for compact manifolds.
Application: Existence of Riemannian metrics.
Application: Whitney’s embedding theorem (proof here).
Exercise: Read (and understand!) the proof of the existence of partitions of unity in
general (better than in Tu, see here).

§9. Orientation
Orientability: bundle! Example: T M is orientable as manifold.
Moebius strip: paper trick, knot: intrinsic vs extrinsic topology.

8
 §10. Differential 1–forms
Ω1(M ) = Γ(T ∗M ) = {w : X (M ) → F(M )/w is F(M )−linear}:
Local operator ⇒ point-wise operator ⇒ F(M )-linear.
f ∈ F(M ) ⇒ df ∈ Ω1(M ), and df ∼ = f∗ .
∂ ∂
(x, U ) chart ⇒ { ∂x1 |p, . . . , ∂xn |p} is basis of TpM whose dual basis
is {dx1|p, . . . , dxn|p} (i.e., basis of Tp∗M ).
{dx1, . . . , dxn} are then a frame of T ∗U : local expression.
Example: Liouville form on T ∗M : λ(w) := w ◦ π∗w .
Pull-back: ϕ ∈ End(V , W ) ⇒ ϕ∗ ∈ End(W ∗, V ∗);
f : M → N ⇒ f ∗ : F(N ) → F(M ); f ∗ : Ω1(N ) → Ω1(M ).
Importance of pull-back!
Restriction of 1-forms to a submanifold i : S → M : w|S = i∗w.

§11. Multilinear algebra

Let V and V 0 R–vector spaces. V ∗ = Hom(V , R).
Bi/tri/multi linear functions on vector spaces: V ⊗ V .
Tensors and k–forms on V : Bil(V ) = (V ⊗ V )∗ = V ∗ ⊗ V ∗.
V ⊗ V 0, V ∧ V , ∧0 V = V ⊗0 := R,
V ⊗k := V ⊗ · · · ⊗ V , dim V ⊗k = (dim V )k
 
dim V
∧k V := V ∧ · · · ∧ V ⊂ V ⊗k , dim ∧k V =
k
Operators ⊗ and ∧ (bil. and assoc.) over multilinear maps:
1
σ ∈ ∧k V , ω ∈ ∧s V ⇒ ω ∧ σ := A(ω ⊗ σ) ∈ ∧(k+s) V
k!s!
REM: ω ∧ σ = (−1)ks σ ∧ ω.
9
 §12. Differential k – forms and tensor fields
ALL the multilinear algebra extends to vector bundles: Hom(E, E 0)
Examples: T ∗M ; Riemannian metric: h , i|U =
P
gij dxi ⊗ dxj
Tensor (field) and (differential) k-form:
X k (M n), Ωk (M n)
are simply the sections of the bundles (T ∗M )⊗k , Λk (T ∗M ).
Tensors = F(M )-multilinear maps (bump-functions).
REM: Ω0(M ) = X 0(M ) = F(M ), Ω1(M ) = X 1(M ).
Notation: Jk,n := {(i1, . . . , ik ) : 1 ≤ i1 < · · · < ik ≤ n}, and for
I = (i1, . . . , ik ) ∈ Jk,n, we set dxI := dxi1 ∧ · · · ∧ dxik .
Local expression:
df1 ∧ · · · ∧ dfn = det([∂fi/∂xj ]1≤i,j≤n) dx1 ∧ · · · ∧ dxn , (1)
and, for J = (j1, . . . , jk ) ∈ Jk,n and y1, . . . , yn ∈ F(M ),
 
X ∂yjr
dyJ = det(AJI ) dxI , onde AJI = .
∂xis 1≤r,s≤k
I∈Jk,n

Wedge operator ∧ : Ωk (M ) × Ωs(M ) → Ωk+s(M ) bilinear, ten-

sorial n
M
•
Ω (M ) := Ωk (M )
k=0
is a graded algebra with ∧.
Pull-back of tensors and forms: linear, tensorial, respects ∧:
F ∗f := f ◦ F, ∀f ∈ F(M ),
F ∗(ω ∧ σ) = F ∗ω ∧ F ∗σ,
(F ◦ G)∗ = G∗ ◦ F ∗.
10
 §13. Orientation and n – forms
Recall: if B = {v1, . . . , vn}, B 0 = {v10 , . . . , vn0 } are bases of
V n ⇒ β(v1, . . . , vn) = det C(B, B 0)β(v10 , . . . , vn0 ), ∀ β ∈ Λn(V n).
We say that β determines an orientation [B] if β(v1, ..., vn) > 0.
REM: M n orientable ⇔ exists β ∈ V, where
V = {σ ∈ Ωn(M n) : σ(p) 6= 0, ∀ p ∈ M n}.
Orientations of M ∼
= V/F+(M ).
Diffeomorphisms that preserve/revert orientation.

§14. Exterior derivative: VIP!!

Definition 3. The exterior derivative on Ω•(M ) is the linear
map d : Ω•(M ) → Ω•(M ) that satisfies the following properties:
1. d(Ωk (M )) ⊂ Ωk+1(M );
2. f ∈ F(M ) = Ω0(M ) ⇒ df (X) = X(f ), ∀ X ∈ X (M );
3. ∀ω ∈ Ωk (M ), σ ∈ Ω•(M ) ⇒ d(ω ∧σ) = dω ∧σ +(−1)k ω ∧dσ;
4. d2 = 0.

• Props (2) + (3) + bump functions: ω|U = 0 ⇒ dω|U = 0.

• Then, dω|U = d(ω|U ), and we can carry local computations.
• Props (3) + (4) + induction ⇒ d(df1 ∧ · · · ∧ dfk ) = 0.
• d exists and is unique: coordinate local expression.
For every F : M → N we have that (see first for Ω0):

F∗ ◦ d = d ◦ F∗

11
i.e., F ∗ : Ω•(N ) → Ω•(M ) is a morphism of differential graded
algebras (i.e., preserves degree and commutes with d).
REM: This also explains why dω|U = d(ω|U ) via inc∗.
Exercise: ∀ k, ∀ ω ∈ Ωk (M ), ∀ Y0 , . . . , Yk ∈ X (M ), it holds that dw(Y0 , . . . , Yk ) =
k
X k
X
(−1)i Yi ω(Y0 , . . . , Ŷi , . . . , Yk ) + (−1)i+j ω([Yi , Yj ], Y0 , . . . , Ŷi , . . . , Ŷj , . . . , Yk ).
i=0 0≤i<j≤k

Given X ∈ X (M ) we define the interior multiplication

iX : Ωk+1(M ) → Ωk (M )
by (iX ω)(Y1, . . . , Yk ) = ω(X, Y1, . . . , Yk ).
• iX ω is tensorial (= F(M )-bilinear) on X and on ω.
• ∀ ω ∈ Ωk (M ), σ ∈ Ωr (M ),
iX (ω ∧ σ) = (iX ω) ∧ σ + (−1)k ω ∧ (iX σ).
• iX ◦ iX = 0.

§15. Manifolds with boundary

C ∞ functions and diffeos over arbitrary subsets S ⊂ M n.

Proposition 4. Let U ⊂ M n open, S ⊂ M̂ n arbitrary, and

f : U → S a diffeomorphism. Then, S is open on M̂ n.

Corolary 5. Let U and V open of Hn := Rn+ = {xn ≥ 0} and

f : U → V a diffeomorphism. Then f takes interior (resp.
boundary) points to interior (resp. of boundary) points.

12
Manifold with boundary: definition. (Rough idea of orbifold).
Interior points.
The boundary of M n = ∂M n is a manifold of dimension n − 1.
∂M versus topological boundary.
If p ∈ ∂M : Fp(M ), Tp∗M , v ∈ TpM (yet, it could be no curve
with α0(0) = v), T M , orientation: SAME as before.
If p ∈ ∂M : v ∈ TpM interior and exterior.
REM: In any manifold with boundary M there exists an ex-
terior vector field X along ∂M (i.e., considering the inclusion
inc : ∂M → M we have that X ∈ Xinc). Then, ∂M is ori-
entable if M is, with the induced orientation inc∗iX ω.
Examples: Hn, [a, b]; B n, B n.
Example: If j = inc : Sn−1 = ∂B n → B n, Z(p) = p ∈ Xinc is
exterior ⇒ orientation σ in Sn−1 ⊂ B n via B n ⊂ Rn and dvRn :
X
∗
σ = j (iZ dvRn ) = (−1)i−1 xi dx1 ∧ · · · ∧ dx
ci ∧ · · · ∧ dxn. (2)
i

§16. Integration (Riemann)

Forms with compact support = Ω•c (M ): preserved by pull-backs.
If ω ∈ Ωnc(U ), U ⊂ Hn, we have ω = f dx1 ∧ · · · ∧ dxn and define
Z Z Z
ω= ω := f dx.
U Hn Hn

Same for w n-form continuous on U , A ⊂ R U bounded with mea-

sure zero boundary (e.g., A = cube) ⇒ A ω.
Given a diffeo ξ : U ⊂ Hn → V ⊂ Hn with  = 1 (resp. -1) if
ξ preserves (resp. reverses) orientation, we get from (1) and the

13
Change of Variables Theorem (CVT) that
Z Z
ξ ∗ω = ξ ∗(f dx1 ∧ · · · ∧ dxn)
U ZU
= f ◦ ξ (ξ ∗dx1 ∧ · · · ∧ ξ ∗dxn)
ZU
= f ◦ ξ (dξ1 ∧ · · · ∧ dξn)
ZU Z
= f ◦ ξ det(Jξ ) dx1 ∧ · · · ∧ dxn =  ω.
U V
Def.: If M n is oriented, ϕ : UR⊂ M n → Hn chart oriented, and
w ∈ Ωnc(U ), we define U ω = M ω := ϕ(U )(ϕ−1)∗w. Linear!
R R
n n n
R P R
Def.: RM oriented, w ∈ Ωc (M ) ⇒ M ω := α M ρα w.
CVT: N ϕ∗ω = M ω, ∀ ϕ ∈ Dif+(N, M ), ∀ w ∈RΩnc(M n).
R

M n oriented ⇒ linear Roperator: ω ∈ Ωnc(M n) 7→ M ω.

P P
The
R dim M R= 0 case: M f := i f (p i ) − j f (qj ).
−M ω = − M ω.

§17. Stokes Theorem 1.0

...which was not proved by Stokes, but by Klein (dim 2) and E.Cartan in general... :o/

Theorem 6 (Stokes v.1.0). M n oriented, w ∈ Ωn−1 n

c (M ) ⇒
Z Z
dω = ω.
M ∂M

Underlying idea: Sum integrals over small cubes, since the inte-
rior faces cancel down due to orientation
R (dim 1 and 2 pictures).
Cor.: M n compact oriented ⇒ M dω = 0, ∀ω ∈ Ωn−1(M ).
Exercise: The classical calculus theorems all follow from Stokes.

14
OBS (!!): i : N k ⊂ M , N k compact oriented regular sub-
k k k
manifold,
R and
R ∗ ω ∈ Ω (M ) (or N oriented and ω ∈ Ω c (M ))
⇒ N ω (= N i ω). R
If ρ ∈ Dif f+(N k ) ⇒ N ρ∗ω = N ω ⇒ we only care about the
R

image i(N R), nor really

R ∗on the map i.
Notation: i w := N i ω. R
It makes sense for anyR differentiable
R function i: i w (even for M
not orientable!), and i◦ρ w = i w (we only care about i(N )).
Curiosity: Palais’ Theorem. Let D : Ωk → Ωr such that Df ∗ = f ∗ D, ∀f : M → N . Then,
R
either k = l and D = cId, or r = k + 1 and D = c d, or k = dimM , r = 0, and D = c M .

§18. Stokes Theorem 2.0 (Spivak vol.1 chap.8)

k-cube: I k : [0, 1]k ,→ Rk . SingularR k-cube:R c: [0, 1]k → M .

c singular k-cube, ω ∈ Ωk (M ) ⇒ c ω := [0,1]k c∗ω.
Ck (M ) = Ck (M ; G) := k-chains of M = free G-module over
singular
R cubes, for G = R (or Z or Q or Z2 or...).
: Ck (M ) × Ωk (M ) → R is defined ∀ M and is bilinear!
n
Ii,α (x1, . . . , xn−1) := I n(x1, . . . , xi−1, α, xi, . . . , xn−1)), α = 0,1.
n
, ∂c = ni=1 1α=0(−1)i+α ci,α (dim 2 picture).
P P
ci,α := c ◦ Ii,α
Extend linearly ∂: Ck (M ) → Ck−1(M ): ∂c = boundary of c.
Defs: c ∈ Ck (M ) is closed if ∂c = 0; c is um boundary if c = ∂c̃.
Examples: c1, c2 1-cubes. c1 closed ⇔ c1(0) = c1(1); c = c1 −c2 is
closed ⇔ c1(0) = c2(0) and c1(1) = c2(1), or c1 and c2 closed.
n n
Since (Ii,α )j,β = (Ij+1,β )i,α ∀ 1 ≤ i ≤ j ≤ n−1 ⇒ ∂ 2 = 0 .
What we proved in Theorem 6 is, in fact, the following:

Theorem 7 (Stokes v.2.0). For every differentiable mani-

15
fold M , w ∈ Ωk−1(M ), and c ∈ Ck (M ), we have that
Z Z
dω = ω.
c ∂c
R
In other words, ∂ (over R) is the dual (with respect to ) of d.
Everything works exactly the same considering k-simplex instead
of k-cubes.

DO ALL EXERCISES IN CHAP. 8 AND 11 OF SPIVAK!!

16
§19. De Rham cohomology (Spivak, vol.1 chap.8)

If w ∈ Ω1(Rn), when w = df for certain f ∈ F(Rn)? Necessary

condition: dw = 0. Is it enough?? YES: R taking singular 1-cube
c, c(0) = 0, c(1) = p, define f (p) = c w. It is well defined by
Stokes(!), since every closed curve on Rn is a boundary. In fact,
cs(t) = sc1(t) + (1 − s)c0(t). That is: solutions of certain PDEs
are related to the topology of the space.
Poincaré’s Lemma (seen later): Z k (Rn) = B k (Rn).
That is, locally we can always solve the problem, but globally...
depends on the topology!
System of linear PDEs: integrability condition.
Obstructions to solve PDEs, or globalize certain local objects.
Z k (M ) := Ker dk = closed forms (local condition)
B k (M ) := Im dk−1 = exact forms (global condition!)
Definition: The k-th de Rham cohomology of the manifold M
(with or without boundary) is given by
H k (M ) := Z k (M )/B k (M ).
H 0(M ) = Rr , where r is the number of connected comp. of M .
H n(M n) 6= 0 if M n is a compact orientable manifold (Stokes).
H n+k (M n) = 0, ∀ k ≥ 1.
n

R
Ex: dimH (T ) ≥ k : if ωI := [dθi1 ∧· · ·∧dθik ] ⇒ TJ wI = δJI .
k n

Pull-back: F : M → N ⇒ F ∗(= F #) : H k (N ) → H k (M ).
(F ◦G)∗ = G∗ ◦F ∗ ⇒ H k (M ) is an invariant of the differentiable
structure (!), and invariant under diffeomorphisms.
∧ : H k (M ) × H r (M ) → H k+r (M ), [ω] ∧ [σ] := [ω ∧ σ] (well!).
H •(M ) := ⊕k∈ZH k (M ) is the de Rham cohomology ring of M .

17
In fact, H •(M ) is a anticommutative graded algebra, and F ∗ is
a homomorphism of graded algebras.

§20. Homotopy invariance (Spivak, vol.1 chap.8)

Definition 8. Given two manifolds (with or without boundary)

M and N , we say that f, g : M → N are (differentiably) ho-
motopic if there is a smooth function T : M × [0, 1] → N such
that T0 := T ◦ i0 = f , T1 := T ◦ i1 = g, where is(p) = (p, s).
This is an equivalence relation on F(M, N ): f ∼ g.
Example: M is contractible ⇔ IdM ∼ cte.
Proposition 9. If M is a manifold with or without bound-
ary, for all k there is a linear map τ : Ωk (M × [0, 1]) →
Ωk−1(M ) (called cochain homotopy) such that
i∗1 ω − i∗0 ω = dτ ω + τ dω, ∀ ω ∈ Ωk (M × [0, 1]).
R1 ∗
Proof: Define τ (ω) = 0 is (i∂/∂t(ω))ds. It is enough to check
two cases (identify via π1∗ and π2∗). If ω = f dxI , dω = · · · +
(∂f /∂t)dt ∧ dxI , and therefore it is just the Fundamental Theo-
rem of Calculus. If ω = f dt ∧ dxI , then i∗1 ω = i∗0 ω = 0, and an
easy computation gives ⇒ dτ ω + τ dω = 0.
More than a differential invariant H •(M ) is a homotopic invariant:
Theorem 10 (!!!!!!). f ∼ g ⇒ f ∗ = g ∗ (in H •(M )).
Proof: Immediate from Proposition 9. (The same holds true for
the singular homology: see Theorem 2.10 on [Ha] and its proof).
Corolary 11. M contractible ⇒ H k (M ) = 0, ∀ k ≥ 1.
18
Corolary 12. (Poincaré’s Lemma) Z k (Rn) = B k (Rn) ∀k ≥ 1.
Corolary 13. M n compact orient. ⇒ M n not contractible.
Definition 14. f : M → N is a homotopic equivalence if
there exists g : N → M such that g ◦ f ∼ IdM and f ◦ g ∼ IdN .
In this case, we say that M and N are homotopically equivalent,
or that M and N have the same homotopy type: M ∼ N .
Example: M contractible ⇐⇒ M ∼ point.
Exercise. The “letters” X and Y as subsets of R2 are homotopically equivalent but
not homomorphic.

REM: Whitehead’s Theorem states that, if a continuous function (between CW complexes)

induces isomorphisms between all homotopy groups, then f is a homotopy equivalence. Yet,
it is not enough to assume that all homotopy groups are isomorphic: RP2 × S3 6∼ S2 × RP3
since they are covered by S2 × S3 and π1 = Z2 . By Hurewicz Theorem, this implies that a
continuous function f between simply connected CW complexes that induces isomorphisms
between the singular homologies with integer coefficients is also a homotopy equivalence.

Corolary 15 (!!!!!). Let f : M → N be a homotopy equiva-

lence between manifolds with or without boundary.
Then f ∗ : H •(M ) → H •(N ) is a isomorphism.
Corolary 16. If M has boundary, then H •(M ) = H •(M ◦).
Definition 17. A retract of M to a submanifold S ⊂ M is a
function f : M → S such that f |S (= f ◦ incS ) = IdS . S is
called a retract of M (⇒ f ∗ is injective and inc∗S is surjective).
Theorem 18 (Brouwer’s fixed point Theorem). If B ⊂ Rn is
a closed ball (or a compact convex subset), then every contin-
uous function f : B → B has fixed points.
Exercise. If M is compact and orientable, then there is no retraction f : M → ∂M .

19
Definition 19. A deformation retract from M to S ⊂ M is a
function T : M × [0, 1] → M such that T0 = IdM , Im (T1) ⊆ S,
and T1|S = IdS (i.e., retract T1 ∼ T0 = IdM ⇒ T1∗ and inc∗S are
isomorphisms).

In other words, a deformation retract is a homotopy between a

retract from M to S and the identity of M . In particular, if S is
a deformation retract of M , then M ∼ S.

Corolary 20. If E is a vector bundle over M , then H •(E) =

H •(M ).

Application: tubular neighborhoods. Given an embedded

compact submanifold N ⊂ M , for each 0 <  < 0 there exists
an open subset N ⊂ V ⊂ M , such that N is a deformation
retract of V, V ⊂ V0 if  < 0, and ∩V = N . (Proof: use
Whitney’s Theorem for M , or Riemannian metrics; see Theorem
5.2 on [Hr]). In particular, H •(V) = H •(N ).

Definition 21. A strong deformation retract is a deformation

retract T as in Definition 19 such that Tt|S = IdS , ∀ t ∈ [0, 1]
(e.g, H below).

Example: Rn \ {0} ∼ Sn−1 6∼ Rn: H(x, t) = ((1 − t) + t/kxk)x.

Example: Möbius strip F ∼ S1 (⇒ H 2(F ) = 0).

§21. Integrating cohomology: degree (Spivak, vol.1 chap.8)

For noncompact M (without boundary) we also work with
Hck (M ) := Zck (M )/Bck (M ), k ∈ Z.
20
 R
REM: If M is orientable, then : Hcn(M n) → R is of course
n

a well defined linear map. And more:

Theorem
R 22. If M n is a connected orientable manifold, then
: Hcn(M n) → R is a isomorphism (⇒ dim Hcn(M n) = 1).
R
Proof: We only need to check that, if M ω = 0, then ω = dβ
with β with compact support. Rt
(a) It is true for M = R. Se g(t) = −∞ ω ⇒ ω = dg.
(b) If it holds for Sn−1, then it holds for Rn. If ω ∈ Ωnc(Rn) ⊂
Ωn(Rn), since Rn is contractible we know that ω = dη for some
η ∈ Ωn−1(Rn) (but η not necessarily with compact support!).
Now, since ω has compact support (say, inside the ball B1n) and
∗ 0 ∗
R R R R
Rn ω = 0, we have Sn−1 j η = Sn−1 i η = Rn ω = 0 by Stokes,
where i : Sn−1 → Rn and j : Sn−1 → Rn \ {0} are the inclusions,
and η 0 = η|Rn\{0}. Then, by hypothesis, j ∗[η 0] = 0. But j ∗ is a
isomorphism since Sn−1 is deformation retract of Rn \ {0}. We
conclude that η 0 = dλ for some λ ∈ Ωn−2(Rn \{0}). In particular,
if h : Rn → R satisfies h ≡ 1 outside of B1n and h ≡ 0 inside Bn,
then β = η − d(hλ) ∈ Ωn−1(Rn) has compact support on B1n,
and ω = dβ.
Another, more explicit proof of (b): If ω = f dvRn ∈ Ωn (Rn ) has compact sup. on B1n , then
R1
define g : Rn → R by g(p) = 0 tn−1 f (tp)dt, r : Rn \ {0} → Sn−1 , r(x) = x/kxk (retract),
i : Sn−1 → Rn the inclusion and σ = iX dvRn ∈ Ωn−1 (Rn ) as in (2).
• RComputation ⇒ Rw = d(gσ) (yet R gσ not necessarily with compact support!)
∗ ∗
• Sn−1 (g ◦ i)i σ = B n f dvR = Rn ω = 0 ⇒ i (gσ) = dλ, by hypothesis.
n

• gσ = r∗ (i∗ (gσ)) = d(r∗ λ) outside B1n , since (i ◦ r)∗p = kpk−1 Πp⊥ , (i ◦ r)∗ σ(p) = kpk−n σ(p),
and g(p) = kpk−n (g ◦ i ◦ r)(p), if kpk ≥ 1.
• If β := gσ − d(hr∗ λ) ⇒ w = d(gσ) = dβ, with sup(β) ⊆ B1n .

(c) (!!!) If it holds for Rn it holds for every MR n. Fix any

w0 ∈ Ωnc(U0) with U0 ⊂ M n diffeo to Rn, with M w0 6= 0.
Let w ∈ Ωnc(M n). It is enough to see that there is a ∈ R and
21
η ∈ Ωn−1 n
c (M ) such that w = aw0 + dη. Taking partitions of
unity we can assume that sup(w) ⊂ U , U diffeo a Rn. Since M n
is connected, there exists a sequence {Ui, 1 ≤ i ≤ m}, Ui diffeo
a Rn, with Um = U and Ui ∩ Ui+1 6= ∅. Let R wi with compact
support, sup(wi) ⊂ Ui ∩ Ui+1, and such that M wi 6= 0. Since it
holds for Rn ∼
= Ui+1, wi+1 − ci+1wi = dηi+1. Done! :)
Theorem 23. M n connected not orientable ⇒ Hcn(M n) = 0.
Proof: Use the idea in (c) above.
Exercise. Prove Theorem 23 using the orientable double cover.

Theorem 24. M n connected non compact, with or without

boundary ⇒ H n(M n) = 0.
Proof: Use the idea in (c). Suppose first M n orientable and
use exhaustion by compact sets (or by Theorem 51). For non
orientable M n, prove that π ∗ : H n(M n) → H n(M̃ n) is injective.
By Theorem 22, for any proper differentiable function between
connected orientable manifolds, f : M n → N n (same dimen-
sion!), there exists deg(f ) ∈ R, the degree of f , such that
Z Z
f ∗ω = deg(f ) ω, ∀ ω ∈ Ωnc(N n).
M N
Theorem 25. Under the above hypothesis, if q ∈ N n is a
regular value of f and f (p) = q, set sgnf (p) = ±1, according
to f∗p preserving or reversing orientation. Then,
X
deg(f ) = sgnf (p).
p∈f −1 (q)

In particular, deg(f ) ∈ Z, and deg(f ) = 0 for f not surjective.

22
Proof: If {p1, . . . , pk } = f −1(q), choose small disjoint neigh-
borhoods Ui of pi and V of q such that f : URi → V is diffeo.
R ω∗ with compactR support on V such that N ω 6= 0. Then,
Let
Ui f ω = sgnf (pi ) V ω. So, the result is immediate... if it only
holds that sup(f ∗ω) ⊂ U1 ∪ · · · ∪ Uk . But we fix it like this:
Let K ⊂ V compact such that q ∈ K o. Then, K 0 = f −1(K) \
(U1 ∪ · · · ∪ Uk ) is compact, and thus f (K 0) is closed not contain-
ing q. Now just change V by any V 0 ⊂ K o \ f (K 0) ⊂ K, with
q ∈ V 0, that automatically satisfies f −1(V 0) ⊂ U1 ∪ · · · ∪ Uk .
REM: The set of regular values is open and dense, and the sum
in Theorem 25 is finite.
REM: Hcn(M n) 6⊂ H n(M n) in general: Hcn(Rn) = R, yet
H n(Rn) = 0, n ≥ 1. In fact, f ∼ g 6⇒ f ∗ = g ∗ on Hc•. But:

Corolary 26. f, g : M n → N n as above, f ∼ g (properly

homotopic) ⇒ deg(f ) = deg(g).

Example: deg(−IdSn ) = (−1)n+1.

Corolary 27. Hairy even dimensional dog Theorem.

REM: We can always comb odd dimensional dogs!

Corolary 28. Fundamental Theorem of Algebra.

Proof: Extend g(z) = z k + a1z k−1 + · · · + ak to C ∪ ∞ = S2 via

zk
g(∞) = ∞. It is smooth since 1/g(1/z) = 1+a z+···a k , and it is
1 kz
homotopic to h(z) = z k via gt(z) = z k + t(a1z k−1 + · · · + ak ).

23
Let w = f (r)dx R ∧ dy = fR(r)rdr ∧ dθ with f with compact
support. Then, R2 h∗w = k R2 w ⇒ deg(g) = deg(h) = k > 0 ⇒
g is surjective.
Another proof: h is a local diffeo that preserves orientation on
C \ {0}, and ∀u ∈ C \ {0}, h−1(u) has k points ⇒ deg(h) = k.

§22. Application: winding number (video 25)

f : M n → Rn+1 an immersion of a compact connected orientable

manifold, p ∈ Rn+1 \ M n, r > 0 such that Br (p) ∩ M n = ∅ ⇒
π ◦ f : M n → ∂Br (p) ∼ = Sn ⇒ w(p) := deg(π ◦ f ) ∈ Z is
the winding number of M n around p (independent on r) ⇒
w is constant on each connected component of Rn+1 \ M n.
See for curves, in particular, the effect of the orientation.
M n is not orientable? Theorem 25 ⇒ winding number mod 2:
exercises 23 to 26 Spivak chap.8: f : M n × I → N n homotopy,
y ∈ N n regular value de f, f0, f1 ⇒ #f0−1(y) = #f1−1(y) mod 2.
Picture ⇒ w is never constant and jumps at M n ⇒
Corolary 29. M n orientable or not, b0(Rn+1 \ M n) ≥ 2.

24
 §23. The birth of exact sequences
Let U, V ⊂ M open such that M = U ∪ V , k ∈ Z ⇒ iU : U ,→
M , jU : U ∩ V ,→ U ⇒ i∗U : Ωk (M ) → Ωk (U ), jU∗ : Ωk (U ) →
Ωk (U ∩ V ). Idem for iV , jV . We then have:
i = i∗U ⊕ i∗V : Ωk (M ) → Ωk (U ) ⊕ Ωk (V ),
j = jV∗ ◦ π2 − jU∗ ◦ π1 : Ωk (U ) ⊕ Ωk (V ) → Ωk (U ∩ V ),
i.e., i(ω) = (ω|U , ω|V ), j(σ, ω) = jV∗ ω − jU∗ σ = ω|U ∩V − σ|U ∩V .
Joining, we get
i j
0 → Ωk (M ) → Ωk (U ) ⊕ Ωk (V ) → Ωk (U ∩ V ) → 0, (3)
with each image contained in the kernel of the next. Now, the
fundamental point is that, in fact, they equal! (the only not
obvious is that j is surjective, but, if {ρU , ρV } is a partition of
unity subordinated to {U, V } and ω ∈ Ωk (U ∩ V ), then ωU :=
ρV ω ∈ Ωk (U ), ωV := ρU ω ∈ Ωk (V ), and j(−ωU , ωV ) = ω).

§24. Complexes (Spivak, vol.1, chap.11)

Exact sequences of abelian groups: short, long.

Exercise. The dual of an exact sequence is an exact sequence.
f
A → B → 0 ⇔ f epimorphism
f
0 → A → B ⇔ f monomorphism
f
0 → A → B → 0 ⇔ f isomorphism
f
A→B→C →0⇒C∼ = B/Im f
0→A→B→C →0⇒C∼ = B/A

25
Proposition 30. (General linear algebra dimension Theorem)
α β
If 0 → V 1 → V 2 → · · · → V k → 0 is exact ⇒ i(−1)i dim V i = 0.
P

β[ ]
Proof: Induction on k, changing to 0 → V 2/Im α → V 3 → · · ·
Cochain complex: C = {C k }k∈Z + ‘differentials’ {dk }k∈Z:
d−1 d d
· · · C −1 → C 0 →0 C 1 →1 C 2 · · · , dk ◦ dk−1 = 0.
Direct sum of cochain complexes.
a ∈ C k is a k−cochain of C.
a ∈ Z k (C) := Ker dk ⊂ C k is a k−cocycle of C.
a ∈ B k (C) := Im dk−1 ⊂ C k is a k−coboundary of C.
The k-th cohomology of C is given by
H k (C) := Z k (C)/B k (C).
If a ∈ Z k (C) ⇒ [a] ∈ H k (C) is the cohomology class of a.
Um cochain map ϕ : A → B is a sequence {ϕk : Ak → B k }k∈Z
such that d ◦ ϕk = ϕk+1 ◦ d. This gives maps ϕ∗ : H •(A) →
i j
H •(B). The sequence 0 → A → B → C → 0 is said to be short
exact if at each level k is exact. In this situation,
i∗ j∗
H (A) → H (B) → H k (C)
k k

is exact for all k. But it is NOT exact with 0 at the left or at the
right... BUT:

26
 i j
Theorem 31 (!!!!!!!). If 0 → A → B → C → 0 is short ex-
act, then there exist homomorphisms (explicit and natural)
δ ∗ : H k (C) → H k+1(A),
called connection homomorphisms, that induce the following
long exact sequence in cohomology:

Proof: (“Diagram chasing”: make with students) Given c ∈ Z k (C), there

exists b ∈ B k such that jb = c. But then db ∈ Ker j (jdb =
djb = dc = 0), and, since Ker j = Im i, there is a ∈ Ak+1 such
that db = ia (given b, a is unique since i is injective). Now,
ida = dia = d2b = 0 ⇒ da = 0. Define then δ ∗[c] := [a]
(independent of the choice of b and c).
Let’s check, e.g., that the long sequence is exact on H k (C).
• Im j ∗ ⊂ Ker δ ∗: for [b] ∈ H k (B), we have δ ∗j ∗[b] = δ ∗[jb]. By
definition of δ ∗, we can choose as b itself the element that goes to
c = jb. But b is a cocycle: db = 0. Therefore, in the definition of
δ ∗, ia = db = 0 ⇒ a = 0 ⇒ δ ∗[jb] = [0] = 0. (Idem i∗δ ∗ = 0).
• Ker δ ∗ ⊂ Im j ∗: if δ ∗[c] = 0, the a in the definition of δ ∗ is a

27
coboundary and the b is a cocycle: a = da0. Thus db = ida0 =
dia0, i.e., d(b − ia0) = 0. So j ∗[b − ia0] = [jb − jia0] = [jb] = [c].

§25. The Mayer–Vietoris sequence

As we saw, (3) is exact for all k, hence we conclude:

Theorem 32 (!!!!). The following long sequence of coho-

mology, called the sequence of Mayer–Vietoris, is exact:
0 i∗ 0 0 j∗ 0 δ∗
0 → H (M ) → H (U ) ⊕ H (V ) → H (U ∩ V ) → · · ·
···
δ∗ i∗ j∗ δ∗
· · · → H k (M ) → H k (U ) ⊕ H k (V ) → H k (U ∩ V ) →
δ∗ k+1 i∗ k+1 k+1 j∗ k+1 δ∗
→H (M ) → H (U ) ⊕ H (V ) → H (U ∩ V ) → · · ·

E, for the same price we got the recipe to construct δ ∗:

• If ω ∈ Ωk (U ∩ V ), with part. of unity we get forms ωU and ωV
on U and V such that j(−ωU , ωV ) = ωV |U ∩V + ωU |U ∩V = ω;
• Now, if ω is closed, −dωU and dωV agree on U ∩ V (!!!), since
j(−dωU , dωV ) = dj(−ωU , ωV ) = dω = 0;
• Therefore, −dωU and dωV define a form σ ∈ Ωk+1(M ), that
is clearly closed (yet not necessarily exact!). We conclude that
δ ∗[ω] = [σ] ∈ H k+1(M ).
REM: If U, V and U ∩ V are connected we begin at k = 1, i.e.,
i∗ j∗
0
0 → H (M ) → H (U ) ⊕ H (V ) → H 0(U ∩ V ) → 0,
0 0

i∗ j∗
0 → H 1(M ) → H 1(U ) ⊕ H 1(V ) → · · ·
28
 δ∗
are exact (since M is connected, and H (U ∩ V ) → H 1(M ) is
0

the zero function, since j ∗ : H 0(U ) ⊕ H 0(V ) → H 0(U ∩ V ) is

surjective).
Examples: M = i Mi disjoint ⇒ H k (M ) = ⊕iH k (Mi). H •(Sn).
S

H •(T 2).

§26. The Euler characteristic

In this section we assume that all cohomologies of M have finite
dimension (we will see that this is always the case for M compact).

Definition 33. The Euler characteristic of M is the homo-

topic invariant
X
χ(M ) := (−1)ibi(M ) ∈ Z,
i

where bk (M ) := dim H k (M ) is the k-th Betti number of M .

Mayer–Vietoris + Proposition 30 ⇒
χ(M ) = χ(U )+χ(V )−χ(U ∩ V ). (4)
Simplex ⇒ triangulations: always exist (by countable basis).

Theorem 34. For any triangulation of M n it holds that

n
X
χ(M n) = (−1)iαk ,
i=0

where αk = αk (T ) is the number of k-simplexes in T .

29
Proof: For each n-simplex σi of T , choose pi ∈ σio and pi ∈
Bpi ⊂ σio (think about pi as a small ball too). Let U1 be the
disjoint union of these αn balls, and Vn−1 = M \ {p1, . . . , pαn }.
Then, (4) ⇒ χ(M n) = χ(Vn−1) + (−1)nαn.
For each (n − 1)-face τj of T , pick the “long” ball Bτj joining
the two Bpi ’s of each n-simplex touching τj . Call U2 the union
of these disjoint αn−1 balls. Pick an arc (inside Bτj ) joining the
boundaries of the two Bpi ’s , and let Vn−2 be the complement of
these αn−1 arcs. Again, (4) ⇒ χ(Vn−1) = χ(Vn−2)+(−1)n−1αn−1.
Inductively we obtain Vn−3, · · · , V0, the last one being the union
of α0 contractible sets (each one a neighborhood of a vertex of T ),
so that χ(V0) = α0 and χ(Vk ) = χ(Vk−1) + (−1)k αk .

Corolary 35. (Descartes-Euler) If a convex polyhedron has

V vertices, F faces, and E edges, then V − E + F = 2.

Corolary 36. There are only 5 Platonic solids.

Proof: If r ≥ 3 is the number of edges (= vertices) on each

face, and s ≥ 3 is the number of edges (= faces) that arrive at
each vertex, we have that rF = 2E = sV . But V − E + F =
2 ⇒ 1/s + 1/r = 1/E + 1/2 > 1/2, or (r − 2)(s − 2) < 4.
Since F = 4s/(2s + 2r − sr) we get (r, s) = (3,3) = tetrahedron
= Fire, (4,3) = cube = Earth, (3,4) = octahedron = Air, (3,5)
= icosahedron = Water, and (5,3) = dodecahedron... which,
according to Plato, was “...used by God to distribute the (12!)
Constellations in the Universe” (I was unable to prove this last
assertion).

30
Platonic model of the solar system by Kepler; Circogonia icosahedra; Stones from 2000 AC

STRONG advice: Watch this video about Kepler’s life, from

the spectacular Cosmos TV series (the one from the 80s!).
REM: On dimension n = 4 there are 6 regular solids (there is
one with 24 faces), and for n ≥ 5 there are only 3: the simplex
(tetrahedron), the hypercube (of course), and the hyperoctahe-
dron, that is the convex hull of {±ei}.

§27. Mayer–Vietoris compact support

We cannot simply switch H k by Hck in Mayer–Vietoris, since
ω ∈ Ωkc (M ) 6⇒ i∗U (ω) ∈ Ωkc (U ). However, if ω ∈ Ωkc (U ), the
extension as 0 of ω, îU (ω), satisfies îU (ω) ∈ Ωkc (M ). And this
works! (j := ĵU ⊕ ĵV , i := îU − îV ):
Lemma 37. The following sequence is exact ∀k (exercise):
j i
0 → Ωkc (U ∩ V ) → Ωkc (U ) ⊕ Ωkc (V ) → Ωkc (U ∪ V ) → 0.
Then, Theorem 31 + Lemma 37 ⇒
Theorem 38. The following long sequence is exact:
δ∗ j∗ i∗ δ∗
··· → Hck (U ∩V) → Hck (U ) ⊕ Hck (V )→ k
Hc (M ) →
δ∗ j∗ i∗ δ∗
→ Hck+1(U ∩ V ) → Hck+1(U ) ⊕ Hck+1(V ) → Hck+1(M ) → · · ·
31
REM: Compare both Mayer–Vietoris.
REM: BEWARE not to mix them!!!
REM: Theorem 31 is a factory of theorems!

§28. Mayer–Vietoris for pairs

Let i: N ,→ M be a compact embedded submanifold, and k ∈ Z.
Then, W = M \ N is a manifold and thus
ĵ i∗
Ωkc (M \ N ) →
W
Ωkc (M ) → Ωk (N ).
But this is not exact on Ωkc (M ): the kernel of i∗ are the forms
that vanish on N , while the image of ĵW are the ones that vanish
on a neighborhood of N . But we fix this with a standard trick:
Let V be a tubular neighborhood with compact closure of N ,
j : N ,→ V the inclusion, and π : V → N a deformation retract,
i.e., π ◦ j = idN , j ◦ π ∼ idV . We construct now a sequence of
such V , V = V1 ⊃ V2 ⊃ · · ·, such that ∩iVi = N . Then, we
say that ω and ω 0 on Ωk (U ) for some open U ⊂ M containing
N are equivalent if there is r > i, j such that ω|Vr = ω 0|Vr . The
set of these classes is a vector space G k (N ), that of “germs of
k-forms defined in a neighborhood of N ”, which has an obvious
differential induced by d, and is therefore a cochain complex G =
• k î∗
(G (N ), d). This gives a cochain map Ωc (M ) → G k (N ), where
î∗(ω) = class of ω|V1 .
Lemma 39. The following sequence is exact (exercise):
ĵW î∗
0→ Ωkc (M \ N) → k
Ωc (M ) → G k (N ) → 0.

32
Now, since j ∗ : H k (Vi) → H k (N ) is an isomorphism for all i
and for all k, H k (N ) is isomorphic to H k (G) (exercise). Then,
Theorem 31 + Lemma 39 ⇒

Theorem 40. There is a long exact sequence:

δ∗
··· → Hck (M \N ) → Hck (M ) → H (N ) → Hck+1(M \N ) → · · ·
k

In a completely analogous way to Theorem 40 we conclude:

Theorem 41. Let M be a compact manifold with boundary.

Then there exists a long exact sequence:
k k k δ∗
· · · → Hc (M\∂M ) → Hc (M ) → H (∂M ) → Hck+1(M\∂M ) → · · ·

Corolary 42. Hck (Rn) ∼

= H n−k (Rn) ∼
= (H n−k (Rn))∗, ∀ k.
Proof: By Corolary 16, if B ⊂ Rn is an open ball, Hck (Rn) =
Hck (B) ∼
= Hck (B) = H k (B) = H k (B) = 0, ∀ k 6= n.
Exercise: Compute H • (Sn × Sm ). Suggestion: Sn × Sm = ∂(B × Sm ).

§29. Application: Jordan’s theorem

Theorem 43 (Jordan generalized). Let M n ⊂ Rn+1 be a
connected embedded compact hypersurface. Then, M n is ori-
entable, Rn+1 \ M n has exactly 2 connected components, one
bounded and one not, and M n is the boundary of each one.
Proof: By Theorem 40 and Corolary 42 we have that
0∼
= Hcn (Rn+1 ) → H n (M n ) → Hcn+1 (Rn+1 \ M ) → Hcn+1 (Rn+1 ) ∼
= R → 0.

33
That is, dim H n(M n) + 1 = b0(Rn+1 \ M n) ≥ 2 (Corolary 29).
Hence, by Theorem 22 and Theorem 23, H n(M n) ∼ = R, M n is
orientable, and #{connected components of Rn+1 \ M n} = 2.
By the same argument for winding numbers, each point of M n is
arbitrarily close to points in both connected components.
Corolary 44. Neither the Klein bottle nor the projective
plane can be embedded in R3.

§30. Poincaré duality

Let U ⊂ Rn open bounded and star shaped with respect to 0,
i.e.,
U = Uρ = {tx : 0 ≤ t < ρ(x), x ∈ Sn−1}
for some bounded function ρ : Sn−1 → R>0.
Lemma 45. If ρ ∈ C ∞, U is diffeomorphic to Rn.
Proof: Clearly we can assume ρ ≥ 1, so just choose the diffeo-
morphism h: B1 → U as h(tx) = (t + (ρ(x) − 1)f (t))x, for any
smooth function f with f = 0 on [0, ), f 0 ≥ 0, f (1) = 1.
But ρ does not even need to be continuous... yet, it is semicon-
tinuous:
Lemma 46. Given x ∈ Sn−1 and  > 0, there exist a neigh-
borhood Vx = V (x, ) of x such that ρ|Vx > ρ(x) − .
Proof: U is open.
Lemma 47. H •(U ) ∼ = H •(Rn) and Hc•(U ) ∼
= Hc•(Rn).
(In fact, U is diffeomorphic to Rn even if ρ is not C ∞, but
this is a difficult result).
34
Proof: The first is obvious since U is contractible. By Corolary 42
we thus only need to verify that Hck (U ) = 0 for k < n. But
if [ω] ∈ Hck (U ), suppose that there is ρ ∈ C ∞(R) such that
K = sup(ω) ⊂ Uρ ⊂ U (i.e., ρ < ρ). Then Uρ ∼ = Rn and
[ω] ∈ Hck (Uρ) = 0. So there is η ∈ Ωk−1 k−1
c (Uρ ) ⊂ Ωc (U ) with
ω = dη.
To show that there exists such a ρ, let 2 = d(K, Rn \ U ) > 0
and, for x ∈ Sn−1, t(x) := max{t : tx ∈ K} ≤ ρ(x) − 2. At
a neighborhood Vx of x we have that t|Vx < ρ(x) −  < ρ|Vx by
Lemma 46 and the definition of . Pick a finite subcover {Vxi } of
Sn−1 and a partition of unity {ϕi} subordinated to it, and define
P
ρ = i(ρ(xi) − )ϕi. Then, t < ρ < ρ, and K ⊂ Uρ ⊂ U .

Definition 48. We say that M n is of finite type if there is a

finite covering U of M n such that every nonempty intersection V
of elements of U satisfies that H •(V ) = H •(Rn) and Hc•(V ) =
Hc•(Rn). Such a covering U is called good.

Lemma 49. Every compact manifold has a good covering.

Proof: Totally convex neighborhoods (Riemannian geometry).

Proposition 50. If M is of finite type (e.g. M compact),

then H •(M ) and Hc•(M ) have finite dimension.

Proof: Induction on # U using Mayer–Vietoris.

Now, observing that H k (M ) ∧ Hcr (M ) ⊂ Hck+r (M ) we obtain:

35
Theorem 51 (Poincaré duality). If M n is connected and
orientable, the linear function P D: H k (M ) → (Hcn−k (M ))∗,
Z
P D([ω])([σ]) := ω∧σ
M

is an isomorphism, for all k.

Proof: The proof for manifolds of finite type follows by induction

in the number of elements of a good covering by the next lemma.

Lemma 52. If U and V are open such that P D is an iso-

morphism for all k in U , V and U ∩ V , then P D is an iso-
morphism for all k in U ∪ V .

Proof: Let M = U ∪ V and l = n − k. Mayer–Vietoris gives

H k−1 (U ) ⊕ H k−1 (V ) → H k−1 (U ∩ V ) → H k (M ) → H k (U ) ⊕ H k (V ) → H k (U ∩ V )

↓ PD ⊕ PD ↓ PD ↓ PD ↓ PD ⊕ PD ↓ PD
(Hcl+1 (U ) ⊕ Hcl+1 (V ))∗ → Hcl+1 (U ∩ V )∗ → Hcl (M )∗ → (Hcl (U ) ⊕ Hcl (V ))∗ → Hcl (U ∩ V )∗

where all vertical maps are isomorphisms, except maybe the middle
one. Moreover, all squares commute up to signs (exercise), and
hence up to some signs in the P D’s everything commutes. The
lemma follows now from the five Lemma (prove), which says
precisely that the middle one must also be an isomorphism.

Corolary 53. If M n is compact, connected and orientable,

then bk (M n) = bn−k (M n). In particular χ(M n) = 0 if n is odd.

Corolary 54. Theorem 24 follows from Poincaré duality.

36
30.1 The Poincaré sphere

Henri Poincaré conjectured that a 3-manifold with the homology of a sphere must be
homeomorphic to the 3-sphere S3 . Poincaré himself found a counterexample, essentially
creating the concept of fundamental group. Indeed, by Hurewicz theorem, it would be
enough to take S3 /G, with G ⊂ SO(4) a nontrivial perfect group (i.e., G = [G, G])
acting freely. The simplest such example that we can think of is G = A5 ⊂ SO(3) as
the order 60 icosahedral group since A5 is simple. This almost works, except that G
has to be extended to the binary icosahedral group G = 2A5 of order 120, which is still
perfect, though not simple (or work with A5 but on SO(3) ∼
= S3 /{±I} instead) . Then,
H1 (S3 /G, Z) = G/[G, G] = 0, and H2 (S3 /G) = H1 (S3 /G) = 0 by e.g. Poincaré duality.
Thus, H∗ (S3 /G) = H∗ (S3 ), yet S3 /G is not simply connected. It is remarkable that
this is the only example with finite fundamental group (there are plenty with infinite
fundamental group). After Poincaré found this counterexample to his own conjecture,
he made another one: the 3-sphere is the only simply connected homology 3-sphere.
This is of course the very famous Poincaré conjecture, proved (among other things!) by
G.Perelman in 2002. Notice that, by Perelman’s result, any homology 3-sphere with
finite fundamental group must be S3 /G, with G ⊂ SO(4) perfect, reducing the original
problem to a group one: find the finite perfect subgroups of SO(4) that act freely. It
turns out that 2A5 is the only one!

§31. Singular homology and de Rham Theorem

As seen in Section 18, we have the boundary operator between
chains (of simplex) with any abelian group G as coefficients,
∂k : Ck (M ) → Ck−1(M ), that satisfies ∂ 2 = 0. That is, chains
form a complex (for any topological space). The homology of this
complex is called the singular homology of M :
Hk (M ) = Hk (M ; G) := Ker ∂k /Im ∂k+1.
37
Now, if M = U ∪V , the composition of chains with the inclusions
gives the next (obviously exact) Mayer–Vietoris sequence:
0 → Ck (U ∩ V ) → Ck (U ) ⊕ Ck (V ) → Ck (U + V ) → 0,
where Ck (U + V ) are the k-chains of M that decompose as sum
of k-chains on U and V . By Theorem 31 we get then the cor-
responding long exact sequence on homology. But, with an idea
conceptually similar to the one used to construct G (“barycentric
decomposition”) we prove (with a bit of work) that
H•(U ∪ V ) ∼
= H•(U + V ).
Therefore we have the long exact sequence of singular homology:
· · · Hk+1 (M ) → Hk (U ∩ V ) → Hk (U ) ⊕ Hk (V ) → Hk (M ) → Hk−1 (U ∩ V ) → · · · (5)

Compare with Theorem 38 and use Theorem 7!

For the singular (differentiable) homology H•(M ; R), by Stokes
and in an analogous way to Poincaré duality (Lemma 52 in the
proof of Theorem 51), we prove the following (see Section 29 and
Section 18):

Theorem 55 (deRham). For every manifold M , the linear

function DR : H k (M ) → (Hk (M ; R))∗,
Z
DR([ω])([c]) = ω
c

is an isomorphism, for all k.

Proof: See here for a general argument, even for manifolds that
are not of finite type.

38
 The End. :o)

References
[Ha] Hatcher, A.: Algebraic topology. Cambridge University
Press, 2002.
[Hi] Hitchin, N.: Differentiable manifolds. Lecture notes here.
[Hr] Hirsch, M.: Differential topology. Graduate text in Math-
ematics 33, Springer-Verlag, New York, 1972.
[Le] Lee, J.: Introduction to smooth manifolds. University of
Washington, Washington, 2000.
[Tu] Tu, L: An introduction to manifolds. Second edition. Uni-
versitext. Springer, New York, 2011.
[Sp] Spivak, M.: The comprehensive introduction to differen-
tial geometry.. Vol. III. Third edition. Publish or Perish,
Inc., Wilmington, Del., 1979.
[Zi] Zinger, A: Notes on vector bundles.. Lecture notes here.

39