Revised Course Notes

DIFFERENTIAL GEOMETRY COURSE NOTES
KO HONDA
1. R EVIEW OF TOPOLOGY AND LINEAR ALGEBRA

1.1. Review of topology.
Definition 1.1. A topological space is a pair (X, T ) consisting of a set X and a collection T =
{Uα } of subsets of X, satisfying the following:
(1) ∅, X ∈ T ,
(2) if Uα , Uβ ∈ T , then Uα ∩ Uβ ∈ T ,
(3) if Uα ∈ T for all α ∈ I, then ∪α∈I Uα ∈ T . (Here I is an indexing set, and is not
necessarily finite.)
T is called a topology for X and Uα ∈ T is called an open set of X.
Example 1: Rn = R × R × · · · × R (n times) = {(x1 , . . . , xn ) | xi ∈ R, i = 1, . . . , n}, called

real n-dimensional space.
How to define a topology T on Rn ? We would at least like to include open balls of radius r about
y ∈ Rn :
Br (y) = {x ∈ Rn | |x − y| < r},
where p
|x − y| = (x1 − y1 )2 + · · · + (xn − yn )2 .
Question: Is T0 = {Br (y) | y ∈ Rn , r ∈ (0, ∞)} a valid topology for Rn ?

No, so you must add more open sets to T0 to get a valid topology for Rn .
T = {U | ∀y ∈ U, ∃Br (y) ⊂ U }.
Example 2A: S 1 = {(x, y) ∈ R2 | x2 + y 2 = 1}. A reasonable topology on S 1 is the topology

induced by the inclusion S 1 ⊂ R2 .
Definition 1.2. Let (X, T ) be a topological space and let f : Y → X. Then the induced topology
f −1 T = {f −1 (U ) | U ∈ T } is a topology on Y .
Example 2B: Another definition of S 1 is [0, 1]/ ∼, where [0, 1] is the closed interval (with the
topology induced from the inclusion [0, 1] → R) and the equivalence relation identifies 0 ∼ 1. A
reasonable topology on S 1 is the quotient topology.
1
2 KO HONDA
Definition 1.3. Let (X, T ) be a topological space, ∼ be an equivalence relation on X, X = X/ ∼

be the set of equivalence classes of X, and π : X → X be the projection map which sends x ∈ X
to its equivalence class [x]. Then the quotient topology T of X is the set of V ⊂ X for which
π −1 (V ) is open.
Definition 1.4. A map f : X → Y between topological spaces is continuous if f −1 (V ) = {x ∈
X|f (x) ∈ V } is open whenever V ⊂ Y is open.
HW: Show that the inclusion S 1 ⊂ R2 is a continuous map. Show that the quotient map [0, 1] →
S 1 = [0, 1]/ ∼ is a continuous map.
More generally,
(1) Given a topological space (X, T ) and a map f : Y → X, the induced topology on Y is the
“smallest”1 topology which makes f continuous.
(2) Given a topological space (X, T ) and a surjective map π : X Y , the quotient topology
on Y is the “largest” topology which makes π continuous.
When are two topological spaces equivalent? The following gives one notion:
Definition 1.5. A map f : X → Y is a homeomorphism is there exists an inverse f −1 : Y → X
for which f and f −1 are both continuous.
HW: Show that the two incarnations of S 1 from Examples 2A and 2B are homeomorphic.
Zen of mathematics: Any world (“category”) in mathematics consists of spaces (“objects”) and
maps between spaces (“morphisms”).
Examples:
(1) (Topological category) Topological spaces and continuous maps.
(2) (Groups) Groups and homomorphisms.
(3) (Linear category) Vector spaces and linear transformations.
1.2. Review of linear algebra.
Definition 1.6. A vector space V over a field k = R or C is a set V equipped with two operations
V × V → V (called addition) and k × V → V (called scalar multiplication) s.t.
(1) V is an abelian group under addition.
(a) (Identity) There is a zero element 0 s.t. 0 + v = v + 0 = v.
(b) (Inverse) Given v ∈ V there exists an element w ∈ V s.t. v + w = w + v = 0.
(c) (Associativity) (v1 + v2 ) + v3 = v1 + (v2 + v3 ).
(d) (Commutativity) v + w = w + v.
(2) (a) 1v = v.
(b) (ab)v = a(bv).
(c) a(v + w) = av + aw.
1
Figure out what “smallest” and “largest” mean.
DIFFERENTIAL GEOMETRY COURSE NOTES 3
(d) (a + b)v = av + bv.
Note: Keep in mind the Zen of mathematics — we have defined objects (vector spaces), and now
we need to define maps between objects.
Definition 1.7. A linear map φ : V → W between vector spaces over k satisfies φ(v1 + v2 ) =
φ(v1 ) + φ(v2 ) (v1 , v2 ∈ V ) and φ(cv) = c · φ(v) (c ∈ k and v ∈ V ).
Now, when are two vector spaces equivalent in the linear category?
Definition 1.8. A linear map φ : V → W is an isomorphism if there exists a linear map ψ : W →

V such that φ ◦ ψ = id and ψ ◦ φ = id. (We often also say φ is invertible.)
If V and W are finite-dimensional? ,2 then we may take bases? {v1 , . . . , vn } and {w1 , . . . , wm } and
represent a linear map φ : V → W as an m × n matrix A. φ is then invertible if and only if m = n
and det(A) 6= 0.?
Examples of vector spaces: Let φ : V → W be a linear map of vector spaces.
(1) The kernel ker φ = {v ∈ V | φ(v) = 0} is a vector subspace of V .
(2) The image im φ = {φ(v) | v ∈ V } is a vector subspace of W .
(3) Let V ⊂ W be a subspace. Then the quotient W/V = {w + V | w ∈ W } can be given the
structure of a vector space. Here w + V = {w + v | v ∈ V }.
(4) The cokernel coker φ = W/ im φ.
2
? means you should look up its definition.
4 KO HONDA
2. R EVIEW OF DIFFERENTIATION
2.1. Definitions. Let f : Rm → Rn be a map. The discussion carries over to f : U → V for open
sets U ⊂ Rm and V ⊂ Rn .
Definition 2.1. The map f : Rm → Rn is differentiable at a point x ∈ Rm if there exists a linear
map L : Rm → Rn satisfying
|f (x + h) − f (x) − L(h)|
(1) lim = 0,
h→0 |h|
where h ∈ Rm − {0}. L is called the derivative of f at x and is usually written as df (x).
HW: Show that if f : Rm → Rn is differentiable at x ∈ Rm , then there is a unique L which
satisfies Equation (1).
Fact 2.2. If f is differentiable at x, then df (x) : Rm → Rn is a linear map which satisfies
f (x + tv) − f (x)
(2) df (x)(v) = lim .
t→0 t
We say that the directional derivative of f at x in the direction of v exists if the right-hand side
of Equation (2) exists. What Fact 2.2 says is that if f is differentiable at x, then the directional
derivative of f at x in the direction of v exists and is given by df (x)(v).
2.2. Partial derivatives. Let ej be the usual basis element (0, . . . , 1, . . . , 0), where 1 is in the jth
∂f
position. Then df (x)(ej ) is usually called the partial derivative and is written as ∂x j
(x) or ∂j f (x).
More explicitly, if we write f = (f1 , . . . , fn )T (here T means transpose), where fi : Rm → R,
then T
∂f ∂f1 ∂fn
(x) = (x), . . . , (x) ,
∂xj ∂xj ∂xj
and df (x) can be written in matrix form as follows:
 ∂f1 ∂f1

∂x1
(x) . . . ∂x m
(x)
df (x) = 
 .. .. .. 
. . . 
∂fn ∂fn
∂x1
(x) ... ∂xm
(x)
The matrix is usually called the Jacobian matrix.
Facts:
(1) If ∂i (∂j f ) and ∂j (∂i f ) are continuous on an open set 3 x, then ∂i (∂j f )(x) = ∂j (∂i f )(x).
∂fi
(2) df (x) exists if all ∂x j
(y), i = 1, . . . , n, j = 1, . . . , m, exist on an open set 3 x and each
∂fi
∂xj
is continuous at x.
Shorthand: Assuming f is smooth, we write ∂ α f = ∂1α1 ∂2α2 . . . ∂kαk f where α = (α1 , . . . , αk ).

Definition 2.3.
(1) f is smooth or of class C ∞ at x ∈ Rm if all partial derivatives of all orders exist at x.

(2) f is of class C k at x ∈ Rm if all partial derivatives up to order k exist on an open set 3 x
and are continuous at x.
2.3. The Chain Rule.
Theorem 2.4 (Chain Rule). Let f : R` → Rm be differentiable at x and g : Rm → Rn be
differentiable at f (x). Then g ◦ f : R` → Rn is differentiable at x and
d(g ◦ f )(x) = dg(f (x)) ◦ df (x).
Draw a picture of the maps and derivatives.
Definition 2.5. A map f : U → V is a C ∞ -diffeomorphism if f is a smooth map with a smooth
inverse f −1 : V → U . (C 1 -diffeomorphisms can be defined similarly.)
One consequence of the Chain Rule is:
Proposition 2.6. If f : U → V is a diffeomorphism, then df (x) is an isomorphism for all x ∈ U .
Proof. Let g : V → U be the inverse function. Then g ◦ f = id. Taking derivatives, dg(f (x)) ◦
df (x) = id as linear maps; this give a left inverse for df (x). Similarly, a right inverse exists and
hence df (x) is an isomorphism for all x.
6 KO HONDA
3. M ANIFOLDS
3.1. Topological manifolds.
Definition 3.1. A topological manifold of dimension n is a pair consisting of a topological space
X and a collection A = {φα : Uα → Rn }α∈I of maps (called an atlas of X) such that:
(1) Uα is an open set of X and ∪α∈I Uα = X,
(2) φα is a homeomorphism onto an open subset φα (Uα ) of Rn .
(3) (Technical condition 1) X is Hausdorff.
(4) (Technical condition 2) X is second countable.
Each φα : Uα → Rn is called a coordinate chart.
Definition 3.2. A topological space X is Hausdorff if for any x 6= y ∈ X there exist open sets Ux
and Uy containing x, y respectively such that Ux ∩ Uy = ∅.
Definition 3.3. A topological space (X, T ) is second countable if there exists a countable sub-
collection T0 of T and any open set U ∈ T is a union (not necessarily finite) of open sets in
T0 .
HW: Show that S 1 from Example 2A or 2B from Day 1 (already shown to be homeomorphic from
an earlier exercise) is a topological manifold.
Non-example. A topological manifold which satisfies all the axioms except for the Hausdorff
condition: Take R × {0, 1}/ ∼, where (x, 0) ∼ (x, 1) for all x 6= 0, with the quotient topology.
For any open sets Ui , i = 0, 1, containing (0, i), U0 ∩ U1 6= ∅.
Non-example. The long line is a topological manifold which satisfies all the axioms except for
the second countable condition. It is ω1 × [0, 1) with the smallest element deleted, where ω1 is the
“first uncountable ordinal” (in particular it is uncountable and has an ordering), and the topology
is the order topology coming from the lexicographic ordering on ω1 × [0, 1). For more details see
Wikipedia or Munkres, “Topology”.
HW: Give an example of a Hausdorff, second countable topological space X which is not a topo-
logical manifold. (You may have trouble proving that it is not a topological manifold, though. You
may also want to find several different types of examples.)
Observe that in the land of topological manifolds, a square and a circle are the same, i.e., they are
homeomorphic! That is not the world we will explore — in other words, we seek a category where
squares are not the same as circles. In other words, we need derivatives!
3.2. Differentiable manifolds.
Definition 3.4. A smooth manifold is a topological manifold (X, A = {φα : Uα → Rn }) satisfying
the following: For every Uα ∩ Uβ 6= ∅,
φβ ◦ φ−1
α : φα (Uα ∩ Uβ ) → φβ (Uα ∩ Uβ )
is a smooth map. The maps φβ ◦ φ−1
α are called transition maps.
Note: In the rest of the course when we refer to a “manifold”, we mean a “smooth manifold”,
unless stated otherwise.
8 KO HONDA
4. E XAMPLES OF SMOOTH MANIFOLDS

Today we give some examples of smooth manifolds. For each of the examples, you should also
verify the Hausdorff and second countable conditions!
(1) Rn is a smooth manifold.3 Atlas: {id : U = Rn → Rn } consisting of only one chart.
(2) Any open subset U of a smooth manifold M is a smooth manifold. Given an atlas {φα : Uα →
Rn }) for M , an atlas for U is {φα |U ∩Uα : Uα ∩ U → Rn }.
(3) Let Mn (R) be the space of n × n matrices with real entries, and let
GL(n, R) = {A ∈ Mn (R) | det(A) 6= 0}.
2
GL(n, R) is an open subset of Mn (R) ' Rn , hence is a smooth n2 -dimensional manifold.
GL(n, R) is called the general linear group of n × n real matrices.
(4) If M and N are smooth m- and n-dimensional manifolds, then their product M × N can
naturally be given the structure of a smooth (m + n)-dimensional manifold. Atlas: {φα × ψβ :
Uα × Vβ → Rm × Rn }, where {φα : Uα → Rm } is an atlas for M and {ψβ : Vβ → Rn } is an atlas
for N .
(5) S 1 = {x2 + y 2 = 1} is a smooth 1-dimensional manifold.
(i) One possible atlas: Open sets U1 = {y > 0}, U2 = {y < 0}, U3 = {x > 0}, U4 =
{x < 0}, together with projections to the x-axis or the y-axis, as appropriate. Check the
transition maps!
(ii) Another atlas: Open sets U1 = {y 6= 1} and U2 = {y 6= −1}, together with stereographic
projections from U1 to y = −1 and U2 to y = 1. The map φ1 : U1 → R is defined
as follows: Take the line L(x,y) which passes through (0, 1) and (x, y) ∈ U1 . Then let
φ1 be the x-coordinate of the intersection point between L(x,y) and y = −1. The map
φ2 : U2 → R is defined similarly by projecting from (0, −1) to y = 1. Check the transition
maps!
(6) S n = {x21 + · · · + x2n+1 = 1} ⊂ Rn+1 . Generalize the discussion from (5).

(7) In dimension 2, S 2 , T 2 , genus g surface.
(8) (Real projective space) RPn = (Rn+1 − {(0, . . . , 0)})/ ∼, where
(x0 , x1 , . . . , xn ) ∼ (tx0 , tx1 , . . . , txn ), t ∈ R − {0}.
RPn is called the real projective space of dimension n. The equivalence class of (x0 , . . . , xn ) is
denoted by [x0 , . . . , xn ].
3
Strictly speaking, this should say “can be given the structure of a smooth manifold”. There may be more than one
choice and we have not yet discussed when two manifolds are the same.
Consider U0 = {x0 6= 0} with the coordinate chart φ0 : U0 → Rn given by

x1 xn x1 xn
[x0 , x1 , . . . , xn ] = 1, , . . . , 7→ ,..., .
x0 x0 x0 x0
Similarly, take Ui = {xi 6= 0} and define φi : Ui → Rn . What about transition maps φj ◦ φ−1
i ?
(Explain this in detail.)
(9) (Group actions) The 2-torus T 2 = R2 /Z2 . The discrete group Z2 acts on R2 by translation:
Z2 × R2 → R2
((m, n), (x, y)) 7→ (m + x, n + y).
Note that for each fixed (m, n), we have a diffeomorphism
R2 → R2 ,
(x, y) 7→ (m + x, n + y).
R2 /Z2 is the set of orbits of R2 under the action of Z2 . (One orbit is (x, y) + Z2 .)
Equivalently, the 2-torus is obtained from the “fundamental domain” [0, 1] × [0, 1] by identifying
(0, y) ∼ (1, y) and (x, 0) ∼ (x, 1), i.e., the sides and the top and the bottom. The assignment
R2 → R2 /Z2 , x 7→ [x], is injective when restricted to the interior of the fundamental domain.
The n-torus T n = Rn /Zn is defined similarly.
Next time: Try to answer the question of what it means for two atlases of the same M to be “the
same”.
10 KO HONDA
5. S MOOTH FUNCTIONS AND SMOOTH MAPS

Today we discuss smooth functions on a manifold and smooth maps between manifolds.
5.1. Choice of atlas. Let (M, T ) be the underlying topological space of a manifold, and A1 =
{(Uα , φα )}, A2 = {(Vβ , ψβ )} be two atlases.
Question: When do they represent the same smooth manifold?
Definition 5.1. Two atlases A1 and A2 on M are compatible if
ψβ ◦φ−1
α
φα (Uα ∩ Vβ ) −→ ψβ (Uα ∩ Vβ )
is a smooth map for all pairs Uα ∩ Vβ 6= ∅.
If A1 and A2 are compatible, then we can take A = A1 ∪ A2 which is compatible with both A1
and A2 .
Definition 5.2. Given a smooth manifold (M, A), its maximal atlas Amax = {(Uα , φα )} is an atlas
which is compatible with A and contains every atlas A0 ⊃ A which is compatible with A.
5.2. Smooth functions.
Some more zen: You can study an object (such as a manifold) either by looking at the object itself
or by looking at the space of functions on the object. In the topological category, the space of
functions would be C 0 (M ), the space of continuous functions f : M → R. The function space
perspective has been especially fruitful in algebraic geometry.
Question: What is the appropriate space of functions for a smooth manifold (M, A)?
Definition 5.3. Given a smooth manifold (M, A), a function f : M → R is smooth if
f ◦ φ−1
α : φα (Uα ) → R
is smooth for each coordinate chart (Uα , φα ) of A.

Note that the definition of a smooth function on M depends on the atlas A.
The space of smooth functions f : M → R with respect to A is written as CA∞ (M ). When A is
understood, we write C ∞ (M ).
Lemma 5.4. Two atlases A1 and A2 are compatible if and only if CA∞1 (M ) = CA∞2 (M ).
Proof. Suppose A1 = {(Uα , φα )} and A2 = {(Vβ , ψβ )} are compatible. It suffices to show that
CA∞1 (M ) ⊃ CA∞2 (M ). If f ∈ CA∞2 (M ), then f ◦ ψβ−1 : ψβ (Vβ ) → R is smooth for all β. Now
(3) f ◦ φ−1
α : φα (Uα ∩ Vβ ) → R
can be written as (f ◦ ψβ−1 ) ◦ (ψβ ◦ φ−1 −1 −1

α ), and each of f ◦ ψβ and ψβ ◦ φα is smooth (the latter is
smooth because A1 and A2 are compatible); hence (3) is smooth for all α and β. This implies that
f ◦ φ−1
α : φα (Uα ) → R is smooth for all α.
Suppose CA∞1 (M ) = CA∞2 (M ). We use the existence of bump functions, i.e., smooth functions
h : R → [0, 1] such that h(x) = 1 on [a, b] and h(x) = 0 on R − [c, d], where c < a < b < d. (The
construction of bump functions is an exercise.)
In order to show that the transition maps
ψβ ◦ φ−1
α : φα (Uα ∩ Vβ ) → ψβ (Uα ∩ Vβ ) ⊂ R
n
are smooth, we postcompose with the projection πj : Rn → R to the jth R factor and show that
πj ◦ ψβ ◦ φ−1
α is smooth. Given x ∈ ψβ (Uα ∩ Vβ ), let Bε (x) ⊂ B2ε (x) ⊂ ψβ (Uα ∩ Vβ ) be small
open balls around x. Using the bump functions we can construct a function f on Uα ∩ Vβ such that
f ◦ ψβ−1 equals πj on Bε (x) and 0 outside B2ε (x); f can be extended to the rest of M by setting
−1
f = 0. f is clearly in CA∞2 (M ). Since CA∞1 (M ) = CA∞2 (M ), f ◦ φ−1 −1
α = (f ◦ ψβ ) ◦ (ψβ ◦ φα ) is
smooth. This is sufficient to show the smoothness of πj ◦ ψβ ◦ φ−1 −1
α and hence of ψβ ◦ φα .
Pullback: Let φ : X → Y be a continuous map between topological spaces. Then there is a

naturally defined pullback map
φ∗ : C 0 (Y ) → C 0 (X)
given by f 7→ f ◦ φ. Note that pullback is contravariant, i.e., the direction is from Y to X, which
is the opposite from the original map φ.
∼
Consider the smooth manifold (M, A). If ψ : M → M is a homeomorphism, then ψ ∗ : C 0 (M ) →
∼
C 0 (M ). Although CA∞ (M ) → ψ ∗ (CA∞ (M )), in general CA∞ (M ) 6= ψ ∗ (CA∞ (M )).
Definition 5.5. Two C ∞ -structures CA∞1 (M ) and CA∞2 (M ) are equivalent if there exists a homeo-
morphism of M which takes CA∞1 (M ) ' CA∞2 (M ).
Amazing fact: (Milnor) S 7 has several inequivalent smooth structures! (Not amazingly, S 1 has
only one smooth structure.)
Major open question: (Smooth Poincaré Conjecture) How many smooth structures does S 4 have?
5.3. Smooth maps. In the category of smooth manifolds, we need to define the appropriate maps,
called smooth maps.
Definition 5.6. A map φ : M → N between manifolds is smooth if for any p ∈ M there exist
coordinate charts (Uα , φα ), (Vβ , ψβ ) such that Uα 3 p, Vβ 3 f (p), and the composition
φ−1
α φ ψβ
φα (Uα ) → Uα → Vβ → ψβ (Vβ )
is smooth.
Remark 5.7. For the above definition, we need to take Uα 3 p which is “sufficiently small” so that
φ(Uα ) ⊂ Vβ . So this means that we should be using a maximal atlas (or at least a “large enough”
atlas).
Lemma 5.8. φ : M → N is smooth if and only if φ∗ (C ∞ (N )) ⊂ C ∞ (M ).
12 KO HONDA
The proof is similar to that of Lemma 5.4.

Definition 5.9. A smooth map φ : M → N is a diffeomorphism if there exists a smooth inverse
φ−1 : M → N .
Upshot: Smooth maps between smooth manifolds can be “reduced” to smooth maps from Rn to
Rm .
6. T HE I NVERSE F UNCTION T HEOREM

6.1. Inverse function theorem.
Definition 6.1. A smooth map f : M → N between two manifolds is a diffeomorphism if there is
a smooth inverse f −1 : N → M .
The inverse function theorem, given below, is the most important basic theorem in differential
geometry. It says that an isomorphism in the linear category implies a local diffeomorphism in the
differentiable category. Hence we can move from “infinitesimal” to “local”.
Theorem 6.2 (Inverse function theorem). Let f : U → V be a C 1 map, where U and V are open
sets of Rn . If df (x) : Rn → Rn is an isomorphism, then f is a local diffeomorphism near x, i.e.,
there exist open sets Ux 3 x and Vf (x) 3 f (x) such that f |Ux : Ux → Vf (x) is a diffeomorphism.
Partial proof. Refer to Spivak, Calculus on Manifolds for a complete proof.
Assume without loss of generality that x = 0 and f (0) = 0. We will only show that for all
y ∈ V near 0 there exists x0 ∈ U near 0 such that f (x0 ) = y. First pick x1 such that df (0)(x1 ) = y;
this is possible since df (0) is an isomorphism. We then compare f (x1 ) and df (0)(x1 ) = y: By
the differentiability of f , for any sufficiently small ε > 0 there exists δ > 0 such that whenever
|x1 | < δ we have:
|f (x1 ) − f (0) − df (0)(x1 )| = |f (x1 ) − y| ≤ ε|x1 |.
In other words, the error |f (x1 ) − y| is much smaller than |x1 |. Next we take x2 such that
df (x1 )(x2 ) = y − f (x1 ). Then we have:
|f (x1 + x2 ) − f (x1 ) − df (x1 )(x2 )| = |f (x1 + x2 ) − y| ≤ ε|x2 |.
Now, since f is in the class C 1 , df (x̃) is invertible for all x̃ near 0 and there exists a constant C > 0
such that the norm of (df (x̃))−1 is < C. Hence
|x2 | < C|df (x1 )(x2 )| = C|y − f (x1 )| ≤ Cε|x1 |.
We then repeat the process to obtain x1 , x2 , . . . , and f (x1 + x2 + . . . ) = y. (This process is usually
called Newton iteration.)
6.2. Illustrative example. Let f : R2 → R, (x, y) 7→ x2 + y 2 . We would like to analyze the level
sets f −1 (a), where a > 0. To that end, we consider
F : R2 → R2 , (x, y) 7→ (f (x, y), y).
Let us use coordinates (x, y) for the domain R2 and coordinates (u, v) for the range R2 . We
compute:
2x 2y
dF (x, y) = .
0 1
Let us restrict our attention to the portion x > 0. Since det(dF (x, y)) = 2x > 0, the inverse
function theorem applies and there is a local diffeomorphism between a neighborhood U(x,y) ⊂ R2
14 KO HONDA
of a point (x, y) on the level set f (x, y) = a and a neighborhood VF (x,y) of F (x, y) on the line
u = a.
In particular, f −1 (a)∩U(x,y) is mapped to {u = a}∩VF (x,y) ; in other words, F is a local diffeomor-
phism which “straightens out” f −1 (a). Hence f −1 (a), restricted to x > 0, is a smooth manifold.
Check the transition functions!
Interpreted slightly differently, the pair f, y can locally be used as coordinate functions on R2 ,
provided x > 0.
6.3. Rank. Recall that the dimension of a vector space V is the cardinality of a basis for V . If V
is finite-dimensional, then V ' Rm for some m, and dim V = m.
Definition 6.3. The rank of a linear map L : V → W is the dimension of im(L).
Definition 6.4. The rank of a smooth map f : Rm → Rn at x ∈ Rm is the rank of df (x) : Rm →
Rn . The map f has constant rank if the rank of df (x) is constant.
We can similarly define the rank of a smooth map f : M → N at a point x ∈ M by using local
coordinates.
Claim 6.5. The rank at x ∈ M is constant under change of coordinates.
−1
Proof. We compare the ranks of d(ψα ◦ f ◦ φ−1α ) and d(ψβ ◦ f ◦ φβ ), where φα : Uα → R ,
m
ψα : Vα → Rn , Uα ⊂ M , Vβ ⊂ N , and φβ , ψβ are defined similarly. The invariance of rank is due

to the chain rule:
d(ψβ ◦ f ◦ φ−1 −1 −1 −1
β ) = d((ψβ ◦ ψα ) ◦ (ψα ◦ f ◦ φα ) ◦ (φα ◦ φβ ))
= d(ψβ ◦ ψα−1 ) ◦ d(ψα ◦ f ◦ φ−1 −1

α ) ◦ d(φα ◦ φβ ),
and by observing that d(ψβ ◦ ψα−1 ) and d(φ−1
α ◦ φβ ) are linear isomorphisms.
7. S UBMERSIONS AND REGULAR VALUES

7.1. Submersions.
Definition 7.1. Let U ⊂ Rm and V ⊂ Rn be open sets. A smooth map f : U → V is a submersion
if df (x) is surjective for all x ∈ U . (Note that this means that m ≥ n and that f has full rank.)
Definition 7.2. Let M be a manifold with maximal atlas A = {φα : Uα → Rm } and let N be
a manifold with maximal atlas B = {ψβ : Vβ → Rn }. Then a smooth map f : M → N is a
submersion if all ψβ ◦ f ◦ φ−1
α are submersions, where defined.
Prototype: f : Rm → Rn , (x1 , . . . , xm ) 7→ (x1 , . . . , xn ), where m ≥ n.

Theorem 7.3 (Implicit function theorem, submersion version). Let f : U → V be a submersion,
where U ⊂ Rm and V ⊂ Rn are open sets with m ≥ n. Then for each p ∈ U there exist
∼
U ⊃ Up 3 p and a diffeomorphism F : Up → W ⊂ Rm such that
f ◦ F −1 : W → Rn
is given by
(x1 , . . . , xm ) 7→ (x1 , . . . , xn ).
Proof. Write f = (f1 , . . . , fn ) where fi : U → R, and define the map
F : U → V × Rm−n ,
(x1 , . . . , xm ) 7→ (f1 , . . . , fn , xn+1 , . . . , xm ).
Here we choose the appropriate xn+1 , . . . , xm (after possibly permuting some variables) so that
dF (x) is invertible. Then F is a local diffeomorphism by the inverse function theorem,
f ◦ F −1 (f1 , . . . , fn , xn+1 , . . . , xm ) = (f1 , . . . , fn ),
and F satisfies the conditions of the theorem.
Carving manifolds out of other manifolds: The implicit function theorem, submersion version,
has the following corollary:
Corollary 7.4. If f : M → N is a submersion, then f −1 (y), y ∈ N , can be given the structure of
a manifold of dimension dim M − dim N .
Proof. The implicit function theorem above gives a coordinate chart with coordinate functions of
the form (xi1 , . . . , xin−m ) about each point in f −1 (y). HW: Check the transition functions and
verify the Hausdorff and second countable conditions!!
Example: The easy way to prove that the circle S 1 = {x2 + y 2 = 1} ⊂ R2 can be given the
structure of a manifold is to consider the map
f : R2 − {(0, 0)} → R, f (x, y) = x2 + y 2 .
16 KO HONDA
The Jacobian is df (x, y) = (2x, 2y). Since x and y are never simultaneously zero, the rank of df
is 1 at all points of R2 − {(0, 0)} and in particular on S 1 . Using the implicit function theorem, it
follows that S 1 is a manifold.
7.2. Regular values and Sard’s theorem.
Definition 7.5. Let f : M → N be a smooth map.
(1) A point y ∈ N is a regular value of f if df (x) is surjective for all x ∈ f −1 (y).
(2) A point y ∈ N is a critical value of f if df (x) is not surjective for some x ∈ f −1 (y).
(3) A point x ∈ M is a critical point of f if df (x) is not surjective.
The implicit function theorem implies that f −1 (y) can be given the structure of a manifold if y
is a regular value of f .
Example: Let M = {x3 + y 3 + z 3 = 1} ⊂ R3 . Consider the map
f : R3 → R, (x, y, z) 7→ x3 + y 3 + z 3 .
Then M = f −1 (1). The Jacobian is given by df (x, y, z) = (3x2 , 3y 2 , 3z 2 ) and the rank of
df (x, y, z) is one if and only if (x, y, z) 6= (0, 0, 0). Since (0, 0, 0) 6∈ M , it follows that 1 is a
regular value of f . Hence M can be given the structure of a manifold.
HW: Prove that S n ⊂ Rn+1 is a manifold.
Example: Zero sets of homogeneous polynomials in RPn . A polynomial f : Rn+1 → R is
homogeneous of degree d if f (tx) = td f (x) for all t ∈ R − {0} and x ∈ Rn+1 . The zero set
Z(f ) of f is given by {[x0 , . . . , xn ] | f (x0 , . . . , xn ) = 0}. By the homogeneous condition, Z(f ) is
well-defined. We can check whether Z(f ) is a manifold by passing to local coordinates.
For example, consider the homogeneous polynomial f (x0 , x1 , x2 ) = x30 + x31 + x32 of degree 3
on RP2 . Consider the open set U = {x0 6= 0} ⊂ RP2 . If we let x0 = 1, then on U ' R2 we have
f (x1 , x2 ) = 1 + x31 + x32 . Check that 0 is a regular value of f (x1 , x2 )! The open sets {x1 6= 0} and
{x2 6= 0} can be treated similarly.
More involved example: Let SL(n, R) = {A ∈ Mn (R) | det(A) = 1}. SL(n, R) is called the
special linear group of n × n real matrices. Consider the determinant map
2
f : Rn → R, A 7→ det(A).
We can rewrite f as follows:
f : Rn × · · · × Rn → R, (a1 , . . . , an ) 7→ det(a1 , . . . , an ),
where ai are column vectors and A = (a1 , . . . , an ) = (aij ).
First we need some properties of the determinant:
(1) f (e1 , . . . , en ) = 1.
(2) f (a1 , . . . , ci ai + c0i a0i , . . . , an ) = ci · f (a1 , . . . , ai , . . . , an ) + c0i · f (a1 , . . . , a0i , . . . , an ).
(3) f (. . . , ai , ai+1 , . . . ) = −f (. . . , ai+1 , ai , . . . ).
(1) is a normalization, (2) is called multilinearity, and (3) is called the alternating property. It turns
out that (1), (2), and (3) uniquely determine the determinant function.
We now compute df (A)(B):
f (A + tB) − f (A)
df (A)(B) = lim
t→0 t
det(a1 + tb1 , . . . , an + tbn ) − det(a1 , . . . , an )
= lim
t→0 t
det(a1 , . . . , an ) + t[det(b1 , a2 , . . . , an ) + det(a1 , b2 , . . . , an )
= lim
t→0 t
+ · · · + det(a1 , . . . , an−1 , bn )] + t2 (. . . ) − det(a1 , . . . , an )
t
= det(b1 , a2 , . . . , an ) + · · · + det(a1 , . . . , an−1 , bn )
It is easy to show that 1 is a regular value of df (it suffices to show that df (A) is nonzero for any
A ∈ SL(n, R)). For example, take b1 = ca1 where c ∈ R and bi = 0 for all i 6= 1.
Theorem 7.6 (Sard’s theorem). Let f : U → V be a smooth map. Then almost every point y ∈ Rn
is a regular value.
The notion of almost every point will be made precise later. But in the meantime:
Reality Check: In Sard’s theorem what happens when m < n?
18 KO HONDA
8. I MMERSIONS AND EMBEDDINGS

8.1. Some more point-set topology. We first review some more point-set topology. Let X be a
topological space.
(1) A subset V ⊂ X is closed if its complement X − V = {x ∈ X | x 6∈ V } is open.
(2) The closure V of a subset V ⊂ X is the smallest closed set containing V .
(3) A subset V ⊂ X is dense if U ∩ V 6= ∅ for every nonempty open set U . In other words,
V = X.
(4) A subset V ⊂ X is compact if it satisfies the following finite covering property: any open
cover {Uα }α∈I of V (i.e., the Uα are open and ∪α∈I Uα = V ) admits a finite subcover.
(5) A metric space is compact if and only if every sequence has a convergent subsequence.
(6) A subset of Euclidean space is compact if and only if it is closed and bounded, i.e., is a
subset of some Br (y).
(7) A map f : X → Y is proper if the preimage f −1 (V ) of every compact set V ⊂ Y is
compact. (Remark: If f : X → Y is continuous, then the image of every compact set is
compact.)
8.2. Immersions.
Definition 8.1. Let U ⊂ Rm and V ⊂ Rn be open sets. A smooth map f : U → V is an immersion
if df (x) is injective for all x ∈ U . (Note this means n ≥ m.) A smooth map f : M → N between
manifolds is an immersion if f is an immersion with respect to all local coordinates.
Prototype: f : Rm → Rn , n ≥ m, (x1 , . . . , xm ) 7→ (x1 , . . . , xm , 0, . . . , 0).
Theorem 8.2 (Implicit function theorem, immersion version). Let f : U → V be an immersion,
where U ⊂ Rm and V ⊂ Rn are open sets. Then for any p ∈ U , there exist open sets U ⊃ Up 3 p,
∼
V ⊃ Vf (p) 3 f (p) and a diffeomorphism G : Vf (p) → W ⊂ Rn such that
G ◦ f : Up → Rn
is given by
(x1 , . . . , xm ) → (x1 , . . . , xm , 0, . . . , 0).
Proof. The proof is similar to that of the submersion version. Define the map
F : U × Rn−m → Rn ,
(x1 , . . . , xm , ym+1 , . . . , yn ) 7→ (f1 (x), . . . , fm (x), fm+1 (x) + ym+1 , . . . , fn (x) + yn ),

where x = (x1 , . . . , xm ) and f (x) = (f1 (x), . . . , fn (x)). We can check that dF is nonsingular
after possibly reordering the f1 , . . . , fn and that F −1 ◦ f : Up → Rn is given by
(x1 , . . . , xm ) 7→ (x1 , . . . , xm , 0, . . . , 0).
We then set G = F −1 .
Zen: The implicit function theorem tells us that under a constant rank condition we may assume
that locally we can straighten our manifolds and maps and pretend we are doing linear algebra.
Examples of immersions:
(1) Circle mapped to figure 8 in R2 .
(2) The map f : R → C, t 7→ eit , which wraps around the unit circle S 1 ⊂ C infinitely many
times.
(3) The map f : R → R2 /Z2 , t 7→ (at, bt), where b/a is irrational. The image of f is dense in
R2 /Z2 .
8.3. Embeddings and submanifolds. We upgrade immersions f : M → N as follows:
Definition 8.3. An embedding f : M → N is an immersion which is one-to-one and proper. The
image of an embedding is called a submanifold of N .
The “pathological” examples above are immersions but not embeddings. Why? (1) and (2) are not
one-to-one and (3) is not proper.
Proposition 8.4. Let M and N be manifolds of dimension m and n with topologies T and T 0 . If
f : M → N is an embedding, then f −1 (T 0 ) = T .
Proof. It suffices to show that f −1 (T 0 ) ⊃ T , since a continuous map f satisfies f −1 (T 0 ) ⊂ T . Let
x ∈ M and U be a small open set containing x. Then by the implicit function theorem f can be
written locally as U → Rn , x0 7→ (x0 , 0) (where we are using x0 to avoid confusion with x). We
claim that there is an open set V ⊂ Rn such that V ∩ f (M ) = f (U ): Arguing by contradiction,
suppose there exist y ∈ f (U ) and a sequence {xi }∞ i=1 ⊂ M such that f (xi ) → y but f (xi ) 6∈ f (U ).
The set {y} ∪ {f (xi )}∞ i=1 is compact, so {f −1
(y)} ∪ {xi }∞
i=1 is compact by properness, where we
are recalling that f is one-to-one. By compactness, there is a subsequence of {xi } which converges
to f −1 (y). This implies that xi ∈ U and f (xi ) ∈ f (U ) for sufficiently large i, a contradiction.
20 KO HONDA
9. TANGENT SPACES , DAY I

9.1. Concrete example. Consider S 2 = {x2 + y 2 + z 2 = 1} ⊂ R3 . We recall the defini-
tion/computation of the tangent plane T(a,b,c) S 2 from multivariable calculus. We use the fact that
S 2 is the preimage of the regular value 1 of f , where
f : R3 → R, f (x, y, z) = x2 + y 2 + z 2 .
The derivative of f at the point (a, b, c) is:
df (a, b, c)(x, y, z)T = (2a, 2b, 2c)(x, y, z)T .
The tangent directions are directions (x, y, z) where df (a, b, c)(x, y, z)T = 0.
If you want to think of the tangent plane as a vector space, then T(a,b,c) S 2 = {ax + by + cz = 0}.
If you want to think of it as an affine space, then the tangent plane is the plane through (a, b, c)
which is parallel to ax + by + cz = 0, i.e., T(a,b,c) S 2 = {ax + by + cz = a2 + b2 + c2 = 1}.
Definition 9.1. Let M be a (codimension m) submanifold of Rn . Then we can define Tp M as
follows: Pick a small neighborhood Up ⊂ Rn of p and a submersion f : Up → Rm such that
M ∩ Up is a level set of f . Then Tp M is the set of vectors v ∈ Rn such that df (p)(v) = 0.
Some issues with this definition:
(1) Need to verify that Tp M does not depend on the choice of f : Up → Rm . (Not so serious.)
(2) The definition seems to depend on how M is embedded in Rn . In other words, the definition
is not intrinsic.
We will give several definitions of Tp M which are intrinsic, in increasing order of abstraction!!
9.2. First definition. Let M be a smooth n-dimensional manifold. If U ⊂ M is an open set, then
let C ∞ (U ) be the set of smooth functions f : U → R.
Notation: Let f, g : U → R where 0 ∈ U ⊂ R. Then f = O(g) if there exists a constant C such
that |f (t)| ≤ C|g(t)| for all t sufficiently close to 0. For example, t = sin t + O(t3 ) near t = 0.
(1)
Definition 9.2 (First definition). The tangent space Tp M (here (1) is to indicate that it’s the first
definition) to M at p is the set of equivalence classes
Tp(1) (M ) = {smooth curves γ : (−εγ , εγ ) → M, γ(0) = p}/ ∼,
where γ1 ∼ γ2 if f ◦γ1 (t) = f ◦γ2 (t)+O(t2 ) for all pairs (f, U ) where U is an open set containing
p and f ∈ C ∞ (U ). Here εγ > 0 is a constant which depends on γ.
Let x1 , . . . , xn be coordinate functions for an open set U ⊂ M .
Theorem 9.3 (Taylor’s Theorem). Let f : U ⊂ Rn → R be a smooth function and 0 ∈ U . Then
we can write X X
f (x) = a + ai x i + aij (x)xi xj ,
i i,j
on an open rectangle (−a1 , b1 ) × · · · × (−an , bn ) ⊂ U which contains 0, where a, ai are constants
and aij (x) are smooth functions.
Proof. We prove the theorem for one variable x. By the Fundamental Theorem of Calculus,
Z 1
g(1) − g(0) = g 0 (t)dt.
0
Substituting g(t) = f (tx) and integrating by parts (i.e., udv = uv − vdu) with u = dtd f (tx)
R R
and v = t − 1 we obtain:
Z 1
d
f (x) − f (0) = f (tx)dt
0 dt
Z 1
0 t=1
= x · f (tx) · (t − 1)|t=0 − (t − 1)x2 · f 00 (tx)dt
0
Z 1
= −xf 0 (0)(−1) − x2 (t − 1)f 00 (tx)dt
0
= f 0 (0) · x + h(x) · x2 .
R1
Here we write h(x) = − 0
(t − 1)f 00 (tx)dt. This gives us the desired result
f (x) = f (0) + f 0 (0) · x + h(x) · x2
for one variable.
Corollary 9.4. γ1 ∼ γ2 if and only if xi (γ1 (t)) = xi (γ2 (t)) + O(t2 ) for i = 1, . . . , n.
Corollary 9.5. If M is a submanifold of Rm , then γ1 ∼ γ2 if and only if γ10 (0) = γ20 (0).
(1)
Remark: At this point it’s not clear whether Tp M has a canonical vector space structure. Here’s
one possible definition: Choose coordinates x = (x1 , . . . , xn ) about p = 0. If we write γ1 , γ2 ∈
(1)
Tp M with respect to x, then we can do addition x ◦ γ1 + x ◦ γ2 and scalar multiplication cx ◦ γ1 .
However, the above vector space structure depends on the choice of coordinates. Of course we
can show that the definition does not depend on the choice of coordinates, but the vector space
(2)
structure is clearly canonical in the definition of Tp M from next time.
22 KO HONDA
10. TANGENT SPACES , DAY II

10.1. Sheaf-theoretic notions. We first discuss some sheaf-theoretic notions.
The set C ∞ (U ) of smooth functions f : U → R is an algebra over R, i.e., it has operations c · f ,
f · g, f + g, where c ∈ R and f, g ∈ C ∞ (U ).
Let V ⊂ M be any set, not necessarily open. Then we define
C ∞ (V ) = {(f, U ) | U ⊃ V, f : U → R smooth}/ ∼ .
Here (f1 , U1 ) ∼ (f2 , U2 ) if there exists V ⊂ U ⊂ U1 ∩ U2 for which f1 |U = f2 |U . When we refer
to a smooth function on V , what we really mean is an element of C ∞ (V ), since we need open sets
to define derivatives.
Given open sets U1 ⊂ U2 , there exists a natural restriction map ρUU21 : C ∞ (U2 ) → C ∞ (U1 ),
f 7→ f |U1 . Then C ∞ (V ) is the direct limit? of C ∞ (U ) for all U containing V .
Example: When V = {p}, C ∞ ({p}) (written simply as C ∞ (p)), is called the stalk at the point p
or the set of germs of smooth functions at p.
10.2. Second definition.
Definition 10.1 (Second definition). A derivation is an R-linear map X : C ∞ (p) → R which
satisfies the Leibniz rule:
X(f g) = X(f ) · g(p) + f (p) · X(g).
(2)
The tangent space Tp M is the set of derivations at p.
(2)
By definition, Tp M is clearly a vector space over R.
Remark: It does not matter whether M is a manifold — it could have been Euclidean space
instead, since C ∞ (p) only depends on a small neighborhood of p.
HW: If X is a derivation, then X(c) = 0, c ∈ R.
Examples.
∂f
(1) Xi = ∂x∂ i . Take local coordinate functions x1 , . . . , xn near p = 0. Then let X(f ) = ∂xi
(0).
∂
Check: this is indeed a derivation and ∂xi (xj ) = δij .
(2) Given γ : (−ε, ε) → M with γ(0) = p, define X(f ) = (f ◦ γ)0 (0). This is usually called
the directional derivative of f in the direction γ. Easy check: two γ ∼ γ 0 give rise to the
same directional derivative.
(2)
Lemma 10.2. If M is an n-dimensional manifold, then dim Tp M = n.
∂
Proof. Take local coordinates x1 , . . . , xn so that p = 0. Let Xi = ∂xi
. Then Xi (xj ) = δij
(2)
and clearly the Xi are linearly independent. Thus dim Tp M P≥ n. Now, given some derivation
X, suppose X(xi ) = bi . Then the derivation Y = X − j j Xj satisfies Y (xi ) = 0 for all
b
xi . By Taylor’s Theorem, any f ∈ C ∞ (p) can be written as a +

P P
ai x i + aij xi xj . By the
derivation property, all the quadratic and higher terms vanish, and hence Y (f ) = 0. Therefore
(2)
dim Tp M = n.
Lemma 10.3. The first two definitions of Tp M are equivalent.

Proof. Given γ : (−ε, ε) → M with γ(0) = p, we define X(f ) = (f ◦ γ)0 (0) as in the example.
Since we already calculated dim Tp M = n for the first and second definitions, we see this map is
surjective and hence an isomorphism.
10.3. Third definition. Define Fp ⊂ C ∞ (p) to be germs of functions which are 0 at p. Fp is an

ideal of C ∞ (p). This means that if f ∈ Fp and g ∈ C ∞ (p), then f g ∈ Fp . Let Fp2 ⊂P
Fp be the
∞
ideal of C (p) generated by products of elements of Fp , i.e., consisting of elements fi φj φk ,
where φj , φk ∈ Fp .
(3) (3)
Definition 10.4 (Third definition). Tp M = (Fp /Fp2 )∗ , i.e., Tp M is the dual vector space (the
space of R-linear functionals) of Fp /Fp2 .
(2) (3)
Equivalence of Tp M and Tp M : Recall that a derivation X : C ∞ (p) → R satisfies X(const) =
0. Show that X : Fp → R factors through Fp2 . (Pretty easy, since it’s a derivation.) Hence we have
(2) (3)
a linear map Tp M → Tp M . Now note that dim(Fp /Fp2 ) = n by Taylor’s Theorem.
Fp /Fp2 is called the cotangent space at p, and is denoted Tp∗ M . If f ∈ C ∞ (p), then f − f (p) ∈ Fp ,
and is denoted df (p), when viewed as an element [f − f (p)] ∈ Fp /Fp2 .
24 KO HONDA
11. T HE TANGENT BUNDLE

11.1. The tangent bundle. Let T M = tp∈M Tp M . This is called the tangent bundle. We explain
how to topologize (i.e., give a topology) and give a smooth structure on the tangent bundle.
Consider the projection π : T M → M which sends any q ∈ Tp M to p. Let Pn U ⊂ ∂M be an open
set with coordinates x1 , . . . , xn . Since an element q ∈ Tp M is written as i=1 ai ∂xi , we identify
π −1 (U ) ' U × Rn by sending the point q ∈ T M to (x1 , . . . , xn , a1 , . . . , an ). By this identification
we can induce a topology on π −1 (U ) from U × Rn ; at the same time, we also obtain a local chart
for π −1 (U ).
Check the transition functions. Let x = (x1 , . . . , xn ) be coordinates on U and y = (y1 , . . . , yn )
be coordinates on V . We need to show that the induced topologies on π −1 (U ) and π −1 (V ) are
consistent and that the transition functions on π −1 (U ∩ V ) are smooth.
Let (x, a) be coordinates on π −1 (U ) and (y, b) be coordinates on π −1 (V ). Think of y as a
∂y ∂yi
function of x on U ∩ V . Write ∂x = ( ∂x j
). In terms of y coordinates,
X ∂ X ∂yj ∂
ai = ai .
∂xi ∂xi ∂yj
P ∂y
This is easily verified by thinking of evaluation on functions. Thus, bj = i ai ∂xji .
∂y
Then q ∈ T M corresponds to (x, a) or (y, ∂x a), where a is viewed as a column matrix.
Computation of the Jacobian of the transition function. The Jacobian matrix of the transition
∂y
map (x, a) 7→ (y, ∂x a) is:
!
∂y ∂y ∂y
0
∂x
∂b
∂a
∂b = P ∂x ∂ 2 yi ∂y .
∂x ∂a k ∂xj ∂xk ak ∂x
P ∂yi
The two terms on the bottom are obtained by differentiating bi = k ∂xk ak .
π
Thus we obtain a smooth manifold T M and a C ∞ -function T M → M .
11.2. Examples of tangent bundles.

Example: The tangent bundle of U ⊂ Rn . T U ' U × Rn ⊂ Rn × Rn . A tangent vector
n
Pby a∂ point (x1 , . . . , xn ) ∈ U , together with a tangent direction (a1 , . . . , an ) ∈ R ,
is described
written as i ai ∂xi .
Example: S 2 ⊂ R3 , S 2 = {x21 + x22 + x23 = 1}. A multivariable calculus definition of T S 2

is the following: Think of T S 2 ⊂ R3 × R3 with coordinates (x, y) where x = (x1 , x2 , x3 ) and
y = (y1 , y2 , y3 ). At x ∈ S 2 , Tx S 2 is the set of points y such that x · y = 0. Therefore:
T S 2 = {(x, y) ∈ R3 × R3 | |x| = 1, x · y = 0}.
HW: Show that the tangent bundle T S 2 defined in this way is diffeomorphic to our “official defi-
nition” of the tangent bundle.
11.3. Complex manifolds.
More abstract example: S 2 defined by gluing coordinate charts. Let U = R2 and V = R2 , with
coordinates (x1 , y1 ), (x2 , y2 ), respectively. Alternatively, think of R2 = C. Take U ∩ V = C − {0}.
The transition function is:
φU V
U − {0} −→ V − {0},
1
z 7→ ,
z
with respect to complex coordinates z = x + iy. With respect to real coordinates,

x y
(x, y) 7→ ,− .
x2 + y 2 x2 + y 2
S 2 has the structure of a complex manifold.
Definition 11.1. A function φ : C → C is holomorphic (or complex analytic) if
dφ φ(z + h) − φ(z)
= lim
dz h→0 h
exists for all z ∈ C. Here h ∈ C − {0}.
A function φ : Cn → Cm is holomorphic if
∂φ φ(z1 , . . . , zi + h, . . . , zn ) − φ(z1 , . . . , zn )
= lim
∂zi h→0 h
exists for all z = (z1 , . . . , zn ) and i = 1, . . . n.
Definition 11.2. A complex manifold is a topological manifold with an atlas {Uα , φα }, where
φα : Uα → Cn and φβ ◦ φ−1 n n
α : C → C is a holomorphic map.
Remark: A holomorphic map f : C → C, when viewed as a map f : R2 → R2 , is a smooth map.

Hence a complex manifold is automatically a smooth manifold.
Compute the Jacobians. Rewriting as a map φU V : R2 − {0} → R2 − {0}, we compute:
 
∂ x ∂ x 2 2

∂x x2 +y 2 ∂y x2 +y 2 1 y − x −2xy
JφU V =  = .
y 2 − x2

∂ −y
2 2
∂ −y
2 2
(x2 + y 2 )2 2xy
∂x x +y ∂y x +y
Remark: It is not a coincidence that a11 = a22 and a21 = −a12 .

Explain that T S 2 is obtained by gluing two copies of (R2 − {0}) × R2 together via a map which
sends (a1 , a2 )T over (x1 , y1 ) to J(a1 , a2 )T over (x2 , y2 ).
26 KO HONDA
12. C OTANGENT BUNDLES AND 1- FORMS

12.1. The cotangent bundle. An element of the cotangent space Tp∗ M = Fp /Fp2 is df (p) =
[f − f (p)], which we often write without brackets. It is not hard to see that if x1 , . . . , xn are local
coordinates near p, then dx1 (p), . . . , dxn (p) are linearlyPindependent and hence form a basis for
Tp∗ M . Therefore, an element of Tp∗ M can be written as ai dxi .
We now “topologize” the cotangent bundle T ∗ M = tp Tp∗ M . Again we have a projection
π : T ∗ M → M . Given a coordinate chart U ⊂ M , we identify π −1 (U ) ' U × Rn . This induces
the topology and smooth structure on π −1 (U ).
−1 −1
P U, V ⊂ M , and coordinatize
Transition functions. Take charts P π (U ) and π (V ) by (x, a) and
(y, b), where a corresponds to ai dxi and b corresponds to bi dyi .
P ∂yi
Lemma 12.1. On the overlap U ∩ V , dyi = j ∂x j
dxj .
P ∂f
Similarly, if f is a smooth function on U , then df = j ∂x j
dxj .
Proof. If p ∈ U ∩ V , then, using Taylor’s Theorem,
X ∂yi X ∂yi
dyi (p) = yi − yi (p) = (p)(xj − xj (p)) = (p)dxj (p).
j
∂xj j
∂xj
∂y ∂y −1 ∂y −1
P P
We denote this more simply as dy = ∂x dx. Then dx = ∂x dy. Hence ai dxi = ai [( ∂x ) ]ij dyj
∂y −1 T
and (x, a) 7→ (y, (( ∂x ) ) a), if we write a as a column matrix.
HW: Compute the Jacobian of the transition function.
12.2. Functoriality. Let f : M → N be a smooth map between manifolds. Then we can define
the following natural maps:
f∗ f∗
Contravariant functor. f induces the map C ∞ (f (p)) → C ∞ (p) which restricts to Ff (p) → Fp .
(Check this!) This descends to the quotient
f ∗ : Tf∗(p) N → Tp∗ M,
dg 7→ dg ◦ f.
(The functor takes the category of “pointed smooth manifolds”, i.e., pairs (M, p) consisting of a
smooth manifold and a point p ∈ M and smooth maps f : (M, p) → (N, f (p)), to the category of
R-vector spaces. The functor is contravariant, i.e., reverses directions.)
Covariant functor. f also induces the map
f∗ : Tp M → Tf (p) N,
(3)
given by X 7→ X ◦ f ∗ . (Here we are using Tp M = Tp M .) This makes sense:
f∗ X
Tf∗(p) N → Tp∗ M → R.
(1)
HW: Define the derivative map in terms of Tp M and show the equivalence with the definition
just given.
Derivative map. The maps f∗ : Tp M → Tf (p) N can be combined into the derivative map
f∗ : T M = tp∈M Tp M → T N = tr∈N Tr N.
HW: Show that the derivative map is a smooth map between tangent bundles.
12.3. Properties of 1-forms.
π
Definition 12.2. Let T ∗ M → M be the cotangent bundle. A 1-form over U ⊂ M is a smooth map
s : U → T ∗ M such that π ◦ s = id.
Note that a 1-form assigns an element of Tp∗ M to a given p ∈ M in a smooth manner. The space
of 1-forms on U is denoted by Ω1 (U ) and is an R-vector space.
1. We often write Ω0 (M ) = C ∞ (M ). Then there exists a map d : Ω0 (M ) → Ω1 (M ), g 7→ dg.
HW: Verify that dg is a 1-form on M , i.e., check that dg is a smooth map M → T ∗ M . Hint: in
∂g ∂g
local coordinates dg can be written as x = (x1 , . . . , xn ) 7→ (x, ∂x 1
(x), . . . , ∂x n
(x)).
2. Given φ : M → N , there is no natural map T ∗ M → T ∗ N unless φ is a diffeomorphism.
However, there exists φ∗ : Ω1 (N ) → Ω1 (M ), θ 7→ φ∗ θ. HW: check that the map is well-defined,
i.e., indeed takes a smooth section of T ∗ N to a smooth section of T ∗ M . The 1-form φ∗ θ is called
the pullback of θ under φ.
3. Let ψ : L → M and φ : M → N be smooth maps between smooth manifolds and let θ be a
1-form on N . Then (φ ◦ ψ)∗ θ = ψ ∗ (φ∗ θ). HW: check this. Note however that the order of pulling
back is reasonable.
4. There exists a commutative diagram:
φ∗
Ω0 (N ) −→ Ω0 (M )
d↓ ↓d ,
φ∗
Ω1 (N ) −→ Ω1 (M )
i.e., d ◦ φ∗ = φ∗ ◦ d. HW: Check this by unwinding the definitions.
5. d(gh) = gdh + hdg. HW: Check this.
Example: Let θ = f (x, y)dx + g(x, y)dy on R2 . If i : R → R2 , t 7→ (a(t), b(t)), then
i∗ θ = (f (a(t), b(t))a0 (t) + g(a(t), b(t))b0 (t))dt.
28 KO HONDA
13. L IE GROUPS
13.1. Lie groups.
Definition 13.1. A Lie group G is a smooth manifold together with smooth maps µ : G × G → G
(multiplication) and i : G → G (inverse) which make G into a group.
Definition 13.2. A Lie subgroup H ⊂ G is a subgroup of G such that the inclusion map H → G
is a 1-1 immersion.4 A Lie group homomorphism φ : H → G is a homomorphism which is also a
smooth map of the underlying manifolds.
Examples of Lie groups:
(1) Rn with the usual addition; the quotient T n = Rn /Zn with the usual addition.
(2) The general linear group GL(n, R) = {A ∈ Mn (R) | det(A) 6= 0}. We already showed
that this is a manifold. The product AB is defined by a formula which is polynomial in the
matrix entries of A and B, so µ is smooth. Similarly one can show that i is smooth.
(3) More invariantly, given a finite-dimensional R-vector space V , let GL(V ) be the group of
∼
R-linear automorphisms V → V .
(4) The special linear group SL(n, R) = {A ∈ Mn (R) | det(A) = 1}.
(5) The orthogonal group O(n) = {A ∈ Mn (R) | AAT = I}.
(6) The special orthogonal group SO(n, R) = SL(n, R) ∩ O(n).
Examples of Lie subgroups:

(1) SL(n, R), O(n), and SO(n, R) are Lie subgroups of GL(n, R); SO(n, R) is a Lie sub-
group of both SL(n, R) and O(n).
(2) Let V be a subspace of Rn . Then the image of V under the quotient map
Rn → T n = Rn /Zn
is a Lie subgroup. For example, we can take V = Rh(a, b)i ⊂ R2 . If b/a is irrational,
then the image of V is dense in T 2 . See Section 8.2. This is why we want to define a Lie
subgroup as the image of a 1-1 immersion.
Definition 13.3. A Lie group representation is a Lie group homomorphism φ : G → GL(V ) for
some finite-dimensional R-vector space V .
13.2. Extended example: O(n).
1. AAT = I implies det(AAT ) = det(I) ⇒ det(A) = ±1. Here we are using det(AB) =
det(A) · det(B) and det(AT ) = det(A).
2. Note that O(n) ⊂ GL(n, R) but O(n) is not quite a subgroup of SL(n, R). O(n) has two
connected components
SO(n) = O(n) ∩ {det(A) = 1} = O(n) ∩ SL(n, R)
4
In some texts a submanifold is the image of a 1-1 immersion; such a definition would be more natural in Lie theory.

−1 0
and A0 · SO(n) = O(n) ∩ {det(A) = −1}, where A0 = ∈ O(n) ∩ {det(A) = −1}.
0 In−1
3. Show O(n) is a submanifold of GL(n, R). Consider the map φ : GL(n, R) → Sym(n) given
by A 7→ AAT , where Sym(n) is the symmetric n × n matrices with real coefficients. We compute
that
dφ(A)(B) = AB T + BAT = AB T + (AB T )T .
Since for any symmetric matrix C there is a solution to the equation AB T = C (here A is fixed
and we are solving for B), it follows that dφ(A) is surjective.
n(n+1) n(n−1)
4. dim O(n) = dim GL(n, R) − dim Sym(n) = n2 − 2
= 2
.
5. SO(n) is the group of “rigid rotations” of S n−1 ⊂ Rn . To see this, write A = (a1 , . . . , an )T ,
where ai are column vectors. Then AAT = id implies that aTi aj = δij , and {a1 , . . . , an } forms
an orthonormal basis for Rn with the usual inner product. (The extra ingredient of det = 1 is
necessary for A to be a “rigid rotation”.)
6. Show compactness of O(n). Since O(n) = φ−1 (I), it follows that O(n) is closed. It is also
bounded by the previous paragraph.

cos θ sin θ
7. HW: Show that SO(2) = | θ ∈ R and that SO(2) is diffeomorphic to S 1 .
− sin θ cos θ
8. Consider SO(3). These are the rigid rotations of S 2 = {x2 + y 2 + z 2 = 1} ⊂ R3 . Prove that
every element A of SO(3) which is not the identity has a unique axis of rotation in R3 . In other
words, there is a unique line through the origin in R3 which is fixed by A, and A is given by a
rotation about this axis.
30 KO HONDA
14. V ECTOR BUNDLES

π π
14.1. Vector bundles. The tangent bundle T M → M and the cotangent bundle T ∗ M → M are
examples of vector bundles.
Definition 14.1. A real vector bundle of rank k over a smooth manifold M is a pair (E, π : E →
M ), where:
(1) E is a smooth manifold and π : E → M is a smooth map.
(2) π −1 (p) has a structure of an R-vector space of dimension k.
∼
(3) There exists an open cover {Uα } of M and identifications π −1 (Uα ) → Uα × Rk which
∼
restrict to vector space isomorphisms π −1 (p) → Rk for all p ∈ M .
A rank 1 vector bundle is often called a line bundle.
∼
The identifications π −1 (Uα ) → Uα × Rk in (3) are called local trivializations.
Definition 14.2. A section of a vector bundle π : E → M over an open set U ⊂ M is a smooth
map s : U → E such that π ◦ s = id. A section over M is called a global section.
We write Γ(E, U ) for the space of sections of E over U ; we also write Γ(E) if U = M . Note that
Γ(E, U ) has an R-vector space structure.
Sections of T M are called vector fields and we often write X(M ) = Γ(T M ). Sections of T ∗ M
are called 1-forms and we often write Ω1 (M ) = Γ(T ∗ M ).
14.2. Transition functions, reinterpreted. Consider π : T M → M and local trivializations

π −1 (U ) ' U × Rn , π −1 (V ) ' V × Rn . Let x = (x1 , . . . , xn ) be the coordinates for U and
y = (y1 , . . . , yn ) be the coordinates for V . We already computed the transition functions
φU V : (U ∩ V ) × Rn → (U ∩ V ) × Rn
∂y
(x, a) 7→ (y(x), ∂x (x)a),
where the domain is viewed as a subset of U × Rn , the range is viewed as a subset of V × Rn , and
a = (a1 , . . . , an )T . Alternatively, think of φU V as
ΦU V : U ∩ V → GL(n, R),
∂y
x 7→ ∂x
(x).
(1) For double intersections U ∩ V , we have ΦU V (p) · ΦV U (p) = id.

(2) For triple intersections Uα ∩ Uβ ∩ Uγ , we have
ΦUα Uγ (p) = ΦUβ Uγ (p) · ΦUα Uβ (p).
∂z ∂z ∂y
This is called the cocycle condition. [If Uα , Uβ , Uγ have coordinates x, y, z, then ∂x
= ∂y
· ∂x
by
the chain rule.]
What’s this cocycle condition? This cocycle condition (triple intersection property) is clearly
necessary if we want to construct a vector bundle by patching together {Uα × Rn }. It guarantees
that the gluings that we prescribe, i.e., ΦUα Uβ from Uα to Uβ , etc. are compatible.
HW: On the other hand, if we can find a collection {ΦUα Uβ } (for all Uα , Uβ ), which satisfies
Conditions (1) and (2), then we can construct a vector bundle by gluing {Uα × Rn } using this
prescription.
Consider π : T ∗ M → M . In a previous lecture we computed that the transition functions ΦU V :
∂y
U ∩ V → GL(n, R) are given by x 7→ (( ∂x (x))−1 )T . The inverse and transpose are both necessary
for the cocycle condition to be met.
14.3. Constructing new vector bundles out of T M . Let M be a manifold and {Uα } an atlas
for M . View T M as being constructed out of {Uα × Rn } by gluing using transition functions
ΦUα Uβ : Uα ∩ Uβ → GL(n, R) that satisfy Conditions (1) and (2).
Consider a representation ρ : GL(n, R) → GL(m, R).
Examples of representations:
(1) ρ : GL(n, R) → GL(1, R) = R× , A 7→ det(A).
(2) ρ : GL(n, R) → GL(n, R), A 7→ BAB −1 .
(3) ρ : GL(n, R) → GL(n, R), A 7→ (A−1 )T .
We can use ρ and glue Uα × Rm together using:

ρ ◦ ΦUα Uβ : Uα ∩ Uβ → GL(n, R) → GL(m, R).
Observe that Conditions (1) and (2) are satisfied since ρ is a representation. Therefore we obtain a
new vector bundle T M ×ρ Rm , called T M twisted by ρ.
Examples of bundles obtained by twisting:
(1) Gives rise to a line bundle which is usually denoted ∧n T M .
(3) Gives the cotangent bundle T ∗ M .
32 KO HONDA
15. M ORE ON VECTOR BUNDLES ; ORIENTABILITY

15.1. Orientability. Let GL+ (k, R) ⊂ GL(k, R) be the Lie subgroup of k × k matrices with
positive determinant. Observe that GL(k, R), like O(k), is not connected, and has two connected
components GL+ (k; R) and diag(−1, 1, . . . , 1) · GL+ (k; R).
π
Let E → M be a rank k vector bundle, {Uα } be a maximal open cover of M on which
π −1 (Uα ) ' Uα × Rk , and
ΦUα Uβ : Uα ∩ Uβ → GL(k, R),
be the transition functions.
Definition 15.1. E → M is an orientable vector bundle if there exists a subcover such that every
ΦUα Uβ factors through GL+ (k, R) (i.e., the image of ΦUα Uβ is in GL+ (k, R)).
Definition 15.2. A manifold M is orientable if T M → M is an orientable vector bundle.
It is not easy to prove, directly from the definition, that the following examples are not orientable.
Example: The Möbius band B is obtained from [0, 1] × R by identifying (0, t) ∼ (1, −t). By
following an oriented basis along the length of the band, we see that the orientation is reversed
when we cross {1} × R. Hence B is not orientable.
Example: The Klein bottle K is obtained from [0, 1] × [0, 1] by identifying (0, t) ∼ (1, 1 − t) and
(s, 0) ∼ (s, 1). It is not orientable.
Example: RP2 = R3 − {0}/ ∼, where x ∼ tx, t ∈ R − {0}. This is the set of lines through
the origin of R3 . Take the unit sphere S 2 . Then RP2 is the quotient of S 2 , obtained by identifying
x ∼ −x. By following an oriented basis from x to −x, we find that RP2 is not orientable. Observe
that there is a two-to-one map π : S 2 → RP2 , which is called the orientation double cover.
Classification of compact 2-manifolds (surfaces). The oriented ones are: S 2 , T 2 , surface of
genus g. The nonorientable ones are: RP2 , Klein bottle, and one whose orientation double cover
is an orientable surface of genus g.
Remark 15.3. Recall ∧n T M from last time. The orientability of M is equivalent to the existence
of a global section s ∈ Γ(∧n T M, M ) which is nonvanishing (i.e., never zero).
15.2. Complex manifolds. Let GL(n, C) = {A ∈ Mn (C) | det A 6= 0}.
Claim 15.4. GL(n, C) is a Lie subgroup of GL(2n, R).
Proof. We’ll explain ρ : GL(1, C) ,→ GL(2, R) and leave the general case as an exercise. Con-
sider z ∈ GL(1, C). If we write z = x + iy, then ρ maps

x −y
z = x + iy 7→ .
y x
It is easy to verify that ρ is a homomorphism (i.e., a representation).
Example: Recall S 2 as a complex manifold, obtained by gluing together U = C and V = C via

the map z 7→ z1 . The transition function, written in real coordinates (x, y), was:
ΦU V : U ∩ V → GL(2, R),
2
1 y − x2 −2xy
(x, y) 7→ 2 .
(x + y 2 )2 2xy y 2 − x2
By Claim 15.4 exists a factorization:
ΦU V : U ∩ V → GL(1, C) → GL(2, R).
Also note that the determinant is positive, so S 2 is oriented.
HW: Show that GL(n, C) ⊂ GL+ (2n, R). This implies that complex manifolds are always ori-
entable.
34 KO HONDA
16. I NTEGRATING 1- FORMS ; TENSOR PRODUCTS

16.1. Integrating 1-forms. Let C be an embedded arc in M , i.e., it is the image of some em-
bedding γ : [c, d] → M . In addition, we assume that C is oriented. In this case, the direc-
tion/orientation on C arises from the usual ordering on the real line. Let ω be a 1-form on M .
Then we define the integral of ω over C to be:
Z Z d
def
ω = γ ∗ ω.
C c
∗
If t is the coordinate on [c, d], then γ ω (in fact any 1-form) will have the form f (t)dt.
Lemma 16.1. The definition does not depend on the particular orientation-preserving parametriza-
tion γ : [c, d] → M .
Proof. Take a different parametrization γ1 : [a, b] → M of C. Then there exists an orientation-
preserving diffeomorphism g : [a, b] → [c, d] such that γ1 = γ ◦ g. (In our case, orientation-
preserving means g(a) = c and g(b) = d.) Now, γ1∗ ω = (γ ◦ g)∗ ω = g ∗ (γ ∗ ω), and
Z d Z d Z b Z b
∗
γ ω= f (t)dt = f (g(s))dg(s) = γ1∗ ω.
c c a a
Now we know how to integrate 1-forms. Over the next few weeks we will define objects that we
can integrate on higher-dimensional submanifolds (not just curves), called k-forms. For this we
need to do quite a bit of preparation.
16.2. Tensor products. We first review some notions in linear algebra and then define the tensor
product. Let V , W be vector spaces over R. (The vector spaces do not need to be finite-dimensional
or over R, but you may suppose they are if you want.)
1. (Direct sum) V ⊕ W . As a set, V ⊕ W = V × W . The addition is given by (v1 , w1 ) + (v2 , w2 ) =
(v1 + v2 , w1 + w2 ) and the scalar multiplication is given by c(v, w) = (cv, cw). dim(V ⊕ W ) =
dim(V ) + dim(W ).
2. Hom(V, W ) = {R-linear maps φ : V → W }. In particular, we have V ∗ = Hom(V, R).
dim(Hom(V, W )) = dim(V ) · dim(W ).
We now define the tensor product V ⊗ W of V and W .
Informal definition. Suppose V and W are finite-dimensional and let {v1 , . . . , vm }, {w1 , . . . , wn }
be bases for V and W , respectively. Then the tensor produce V ⊗ W is a vector space which has
{vi ⊗ wj | i = 1, . . . , m; j = 1, . . . , n}
P
as a basis. Elements of V ⊗ W are linear combinations ij aij vi ⊗ wj .
Definition 16.2. Let V1 , . . . , Vk , U be vector spaces. A map φ : V1 × · · · × Vk → U is multilinear

if φ is linear in each Vi separately, i.e., φ : {v1 } × · · · × Vi × · · · × {vk } → U is linear for each
vj ∈ Vj , j 6= i. If k = 2, we say φ is bilinear.
Formal definition. The tensor product V ⊗ W of V and W is a vector space Z together with a
bilinear map i : V × W → Z, which satisfies the following universal mapping property: Given
any bilinear map φ : V × W → U , there exists a unique linear map φ̃ : Z → U such that φ = φ̃ ◦ i.
Actual construction. Start with the free vectorPspace F (S) generated by a set S. By this we mean
F (S) consists of finite linear combinations i ai si , where si ∈ S, ai ∈ R, and we are treating
distinct si ∈ S as linearly independent. (In other words, S is a basis for F (S).) In our case we take
F (V × W ).
In V ⊗ W , we want finite linear combinations of things that look like v ⊗ w. We also would like
the following:
(v1 + v2 ) ⊗ w = v1 ⊗ w + v2 ⊗ w,
v ⊗ (w1 + w2 ) = v ⊗ w1 + v ⊗ w2 ,
c(v ⊗ w) = (cv) ⊗ w = v ⊗ (cw).
We therefore consider R(V, W ) ⊂ F (V × W ), the vector space generated by the “bilinear rela-
tions”
(v1 + v2 , w) − (v1 , w) − (v2 , w),
(v, w1 + w2 ) − (v, w1 ) − (v, w2 ),
(cv, w) − c(v, w),
(v, cw) − c(v, w).
Then the quotient space F (V × W )/R(V, W ) is V ⊗ W .
Verification of universal mapping property. With V ⊗ W defined as above, define
i : V × W → V ⊗ W, (v, w) 7→ v ⊗ w.
The bilinearity of i follows from the construction of V ⊗ W . (For example, i(v1 + v2 , w) =
(v1 + v2 ) ⊗ w = v1 ⊗ w + v2 ⊗ w = i(v1 , w) + i(v2 , w).) We can define
X X
φ̃ : V ⊗ W → U, ai vi ⊗ wi 7→ ai φ(vi , wi ).
i i
This map is well-defined because all the elements of R(V, W ) are mapped to 0. The uniqueness of
φ̃ is clear.
36 KO HONDA
17. T ENSOR AND EXTERIOR ALGEBRA

17.1. More on tensor products. Recall the definition of the tensor product as V ⊗ W = F (V ×
W )/R(V, W ) and the universal property. The universal property is useful for the following reason:
If we want to construct a linear map V ⊗W → U , it is equivalent to check the existence of a bilinear
map V × W → U .
Dimension of V ⊗ W . Suppose V and W are finite-dimensional. Then we claim that
dim(V ⊗ W ) = dim V · dim W.
To see this, consider the map
τ : V ∗ ⊗ W → Hom(V, W ), f ⊗ w 7→ f w,
where f w : v 7→ f (v)w. The universal mapping property guarantees the well-definition of τ . In
order to verify that τ is an isomorphism, we define its inverse η. Let {v1 , . . . , vm } and {w1 , . . . , wn }
P for V and WPand let {f1 , . . . , fm } be dual to {v1 , . . . , vm }, i.e., fi (vj ) = δij . Then we
be bases
set η( aij fi wj ) = aij fi ⊗ wj . τ and η are clearly inverses. Then dim(Hom(V, W )) =
dim(V ) · dim(W ), which implies that dim(V ⊗ W ) = dim V · dim W .
Properties of tensor products.
(1) V ⊗ W ' W ⊗ V.
(2) (V ⊗ W ) ⊗ U ' V ⊗ (W ⊗ U ).
Proof. (1) It is difficult to directly get a well-defined map V ⊗W → W ⊗V , so start with a bilinear
map V × W → W × V → W ⊗ V , where (v, w) 7→ w ⊗ v. The universal property gives a map
V ⊗ W → W ⊗ V which sends v ⊗ w 7→ w ⊗ v. The map is an isomorphism since there is an
inverse w ⊗ v 7→ v ⊗ w. (2) is left as an exercise.
(2) ensures us that we do not need to write parentheses when we take a tensor product of several
vector spaces.
Let A : V → V and B : W → W be linear maps. Then we have
A ⊕ B : V ⊕ W → V ⊕ W, A ⊗ B : V ⊗ W → V ⊗ W.
We denote V ⊗k for the k-fold tensor product of V . Then we have a representation
ρ : GL(V ) → GL(V ⊗k ), A 7→ A ⊗ · · · ⊗ A.
17.2. Tensor and exterior algebra. The tensor algebra T (V ) of V is
T (V ) = R ⊕ V ⊕ V ⊗2 ⊕ V ⊗3 ⊕ . . . ,
where the multiplication is given by
(v1 ⊗ · · · ⊗ vs )(vs+1 ⊗ · · · ⊗ vt ) = v1 ⊗ · · · ⊗ vt .
Here s or t may be zero, in which case 1 ∈ R acts as the multiplicative identity.
The exterior algebra ∧V of V is T (V )/I, where I is the 2-sided ideal generated by elements
of the form v ⊗ v, v ∈ V , i.e., elements of I are finite sums of terms that look like η1 ⊗ v ⊗ v ⊗ η2 ,
where η1 , η2 ∈ T (V ) and v ∈ V . Elements of ∧V are linear combinations of terms of the form
v1 ∧ · · · ∧ vk , where k ∈ {0, 1, 2 . . . } and vi ∈ V . By definition, in ∧V we have
(4) v ∧ v = 0.
Also note that (v1 + v2 ) ∧ (v1 + v2 ) = v1 ∧ v1 + v1 ∧ v2 + v2 ∧ v1 + v2 ∧ v2 . The first and last terms
are zero, so
(5) v1 ∧ v2 = −v2 ∧ v1 .
∧V is clearly an algebra, i.e., there is a multiplication ω ∧ η given elements ω, η in ∧V .
We define ∧k V to be the degree k terms of ∧V , i.e., linear combinations of terms of the form
v1 ∧ · · · ∧ vk where vi ∈ V .
Alternating multilinear forms. A multilinear form φ : V × · · · × V → U is alternating if
φ(v1 , . . . , vi , vi+1 , . . . , vk ) = −1 · φ(v1 , . . . , vi+1 , vi , . . . vk ).
Recall that transpositions generate the full symmetric group Sk on k letters. If σ ∈ Sk maps
(1, . . . , k) 7→ (σ(1), . . . , σ(k)) and sgn(σ) is the number of transpositions mod 2 that is needed to
write σ as a product of transpositions, then
φ(v1 , . . . , vk ) = (−1)sgn(σ) φ(vσ(1) , . . . , vσ(k) ).
Universal property. ∧k V and i : V × · · · × V → ∧k V satisfy the following: Given an alternating

multilinear map φ : V × · · · × V → U , there is a unique linear map φ̃ : ∧k V → U such that
φ = φ̃ ◦ i.
38 KO HONDA
18. D IFFERENTIAL k- FORMS

18.1. Basis for ∧k V .
Proposition 18.1. Given a basis {e1 , . . . , en } for V , a basis for ∧k V consists of degree k mono-
mials ei1 ∧ · · · ∧ eik with i1 < · · · < ik . Therefore, dim ∧k V = 0 for k > n and dim ∧k V = nk
for k ≤ n.
Example. If V = R3 with basis {e1 , e2 , e3 }, then ∧0 V = Rh1i, ∧1 V = Rhe1 , e2 , e3 i, ∧2 V =
Rhe1 ∧ e2 , e1 ∧ e3 , e2 ∧ e3 i, ∧3 V = Rhe1 ∧ e2 ∧ e3 i, and ∧k V = 0 for k > 3.
Proof. The proof is given in several steps. It is easy to see that {ei1 ∧ · · · ∧ eik | 1 ≤ i1 , . . . , ik ≤ n}
spans ∧k V . Moreover, using Equations (4) and (5), we can shrink the spanning set to:
{ei1 ∧ · · · ∧ eik | 1 ≤ i1 < · · · < ik ≤ n}.
1. If k > n, ∧k V = 0. This is clear since it is impossible to find 1 ≤ i1 < · · · < ik ≤ n.

2. If k = n, ∧k V = R, and the basis is given by e1 ∧ · · · ∧ en . Since e1 ∧ · · · ∧ en spans ∧k V , it
remains to show that e1 ∧ · · · ∧ en is nonzero! This is done by defining a (nontrivial!) alternating
multilinear form V × · · · × V → R, where we are taking n copies of V . Then, by the universal
property, ∧n V cannot be zero and hence must be R. The details are left as an exercise.
3. If 1 ≤ k < n, then we show P that {ei1 ∧ · · · ∧ eik | 1 ≤ i1 < · · · < ik ≤ n} is linearly
independent. Indeed, suppose ω = i1 <···<ik ai1 ,...,ik ei1 ∧ · · · ∧ eik = 0. For each summand, there
is a unique term ej1 ∧ · · · ∧ ejn−k which, when multiplied to ω, kills all the other summands and
gives ±ai1 ,...,ik e1 ∧ · · · ∧ en . Hence ai1 ,...,ik = 0, which proves the linear independence.
Remark 18.2. For any v1 , . . . , vk ∈ V , v1 ∧ · · · ∧ vk 6= 0 in ∧k V if and only if v1 , . . . , vk are

linearly independent.
18.2. Tensor calculus on manifolds. We have now constructed V ⊗k and ∧k V , given a finite-
dimensional vector space V . There exists a natural representation
ρ0 : GL(V ) → GL(V ⊗k ), A 7→ A ⊗ · · · ⊗ A,
and an analogously defined representation
ρ1 : GL(V ) → GL(∧k V ), A 7→ A ∧ · · · ∧ A.
Example. dim V = 2. Basis {v1 , v2 }. ∧V has basis {1, v1 , v2 , v1 ∧ v2 }. If A : V → V is linear

and sends vi 7→ a1i v1 + a2i v2 , i = 1, 2, then A(v1 ∧ v2 ) = Av1 ∧ Av2 = det(A)v1 ∧ v2 .
Thus we can form T M ×ρ0 V ⊗k = ⊗k T M and T M ×ρ1 ∧k V = ∧k T M . We can also form ⊗k T ∗ M
and ∧k T ∗ M .
We will focus on ∧k T ∗ M in what follows. Sections of ∧k T ∗ M are called k-forms and can be
written in local coordinates x1 , . . . , xm as:
X
ω= fi1 ,...,ik dxi1 ∧ · · · ∧ dxik .
i1 <···<ik
Denote by Ωk (M ) the sections of ∧k T ∗ M .

Pullback. Let φ : M → N be a smooth map between manifolds, x1 , . . . , xm and y1 , . . . , yn
∗
P coordinates on M and N , and ω a k-form. Then we can define the pullback φ ω of ω =
local
i1 <···<ik fi1 ,...,ik dyi1 ∧ · · · ∧ dyik as follows, using the definition for 1-forms:
X
φ∗ ω = (fi1 ,...,ik ◦ φ) · d(yi1 ◦ φ) ∧ · · · ∧ d(yik ◦ φ).
i1 <···<ik
HW: Check that the above definition is independent of choice of coordinates.

18.3. The exterior derivative. We can define the extension dk : Ωk (M ) → Ωk+1 (M ) of d0 = d :
Ω0 (M ) → Ω1 (M ) as follows (in local coordinates x1 , . . . , xn ):
P ∂f
(1) If f ∈ Ω0 (M ), then d0 f = df = i ∂x dxi .
i P
k
P
(2) If ω = I fI dxI ∈ Ω (M ), then dk ω = I dfI ∧ dxI .
Here I = (i1 , . . . , ik ) is an indexing set and dxI = dxi1 ∧ · · · ∧ dxik . We will often suppress the k
in dk .
HW: Check that dk is independent of the choice of local coordinates.
Example on R3 . Consider R3 with coordinates (x, y, z). Consider
d d d
Ω0 (R3 ) −→
0
Ω1 (R3 ) −→
1
Ω2 (R3 ) −→
2
Ω3 (R3 ).
d0 is the gradient
∂f ∂f ∂f
d0 : f 7→ dx + dy + dz.
∂x ∂y ∂z
d1 is the curl
∂h ∂g
d1 : f dx + gdy + hdz 7→ − dy ∧ dz + . . . .
∂y ∂z
d2 is the divergence

∂f1 ∂f2 ∂f3
d2 : f1 dy ∧ dz + f2 dz ∧ dx + f3 dx ∧ dy 7→ + + dx ∧ dy ∧ dz.
∂x ∂y ∂z
Remark 18.3. In what follows we will often omit ∧.
40 KO HONDA
19. D E R HAM COHOMOLOGY

This material is nicely presented in Bott & Tu.
Lemma 19.1. The exterior derivative d satisfies the (skew-commutative) Leibniz rule:
(6) d(α ∧ β) = (dα) ∧ β + (−1)k α ∧ (dβ),
where α ∈ Ωk (M ) and β ∈ Ωl (M ), and k or l may be zero (in which case we ignore the ∧).
Proof. HW.
Lemma 19.2. d2 = 0.
Proof. For d2 : Ω0 (M ) → Ω2 (M ), we compute:
!
X ∂f X ∂ 2f
d ◦ df = d dxi = dxj ∧ dxi = 0.
i
∂xi i,j
∂xi ∂xj
When α = dxI , we verify that dα = d(dxI ) = 0.
Now if α = fI dxI , then
dα = dfI ∧ dxI + fI d(dxI ) = dfI ∧ dxI ,
d2 α = (d2 fI ) ∧ dxI − dfI ∧ d(dxI ) = 0,
which proves the lemma.
Example: If M = R3 , then d0 = grad, d1 = curl, d2 = div. Then d2 = 0 is equivalent to

div(curl) = 0, curl(grad) = 0.
Consider the sequence
d d d dn−1
(7) 0 → Ω0 (M ) −→
0
Ω1 (M ) −→
1
Ω2 (M ) −→
2
. . . −→ Ωn (M ) → 0,
where n = dim M . Since dk ◦ dk−1 = 0, we have Im dk−1 ⊂ Ker dk . This leads to the following
definition:
Definition 19.3. The kth de Rham cohomology group of M is given by:
k def
HdR (M ) = Ker dk / Im dk−1 .
Definition 19.4. Let ω ∈ Ωk (M ). Then ω is closed if dω = 0 and is exact if ω = dη for some
η ∈ Ωk−1 (M ).
Fact 19.5. The de Rham cohomology groups are diffeomorphism invariants of the manifold M
∼
(this means that if there is a diffeomorphism φ : M → N , then the de Rham cohomology groups
of M and N are isomorphic), and are finite-dimensional if M is compact.
d dk+1
Definition 19.6. A sequence of vector spaces · · · −→ C k −→
k
C k+1 −→ C k+2 −→ . . . is said to
be exact if Im dk−1 = Ker dk for all k.
The de Rham cohomology groups measure the failure of Equation (7) to be exact.
We will often write H k (M ) instead of HdR
k
(M ).
Examples:
1. M = {pt}. Then Ω0 (pt) = R and Ωk (pt) = 0 for i > 0. Hence H 0 (pt) = R and H k (pt) = 0
for k > 0.
2. M = R. Then Ω0 (M ) = C ∞ (R) and Ω1 (M ) ' C ∞ (R) because every 1-form is of the form
df
f dx. Now, d : f 7→ dx dx can be viewed as the map
d : C ∞ (R) → C ∞ (R),
f 7→ f 0 .
It is easy to see that Ker d = {constant functions} and hence H 0 (R) = R. Next, Im d is all of
x
C ∞ (R), since given any f we can take its antiderivative 0 f (t)dt. Therefore, H 1 (R) = 0. Also
R
H k (R) = 0 for k > 1 since Ωk (R) = 0 for k > 1.

3. M = S 1 . View S 1 as R/Z with coordinates x. Then
Ω0 (S 1 ) = {periodic functions on R with period 1}.
Ω1 (S 1 ) is also the set of periodic functions on R by identifying f (x)dx 7→ f (x). H 0 (S 1 ) = R as
Rbefore. Now, H 1 (S 1 ) = Ω1 (S 1 )/ Im d and Im d is the space of all C ∞ -functions f (x) with integral
1
0
f (x)dx = 0. Hence H 1 (S 1 ) = R. We also have an exact sequence:
R
0 d 1
0 → R → Ω → Ω → R → 0.
Basic properties.
1. H 0 (M ) ' R if M is connected. Proof: df = 0 if and only if f is locally constant.
2. H k (M ) = 0 if dim M < k. Proof: Ωk (M ) = 0.
3. If M = M1 t M2 , then H k (M ) ' H k (M1 ) ⊕ H k (M2 ) for all k ≥ 0. Proof: HW.
42 KO HONDA
20. M AYER -V IETORIS SEQUENCES ; P OINCAR É LEMMA

20.1. Pullback. Let φ : M → N be a smooth map between manifolds.
Lemma 20.1. d ◦ φ∗ = φ∗ ◦ d.
HW: Verify the lemma. This follows easily by computing in local coordinates.
Corollary 20.2. There is an induced map φ∗ : H k (N ) → H k (M ) on the level of cohomology.
Proof. If ω is a closed k-form on N , i.e., ω ∈ Ωk (N ) satisfies dω = 0, then φ∗ ω satisfies dφ∗ ω =
φ∗ (dω) = 0. If ω is exact, i.e., ω = dη, then φ∗ ω = φ∗ dη = d(φ∗ η) is exact as well.
20.2. Mayer-Vietoris sequences. This is a method for effectively decomposing a manifold and
computing its cohomology from its components.
Let M = U ∪ V , where U and V are open sets. Then we have natural inclusion maps
i ,i
U V i
(8) U ∩ V −→ U t V → M.
Here iU and iV are inclusions of U ∩ V into U and into V .
Example: M = S 1 , U = V = R, U ∩ V = R t R.
Theorem 20.3. We have the following long exact sequence:
i∗ i∗ −i∗
0 → H 0 (M ) → H 0 (U ) ⊕ H 0 (V ) −→
U V
H 0 (U ∩ V ) →
i∗ i∗ −i∗
→ H 1 (M ) → H 1 (U ) ⊕ H 1 (V ) −→
U V
H 1 (U ∩ V ) →
→ ....
Remark 20.4. 0 → A → B exact means A → B is injective. A → B → 0 exact means A → B
is surjective. Hence 0 → A → B → 0 exact means A → B is an isomorphism.
The proof of Theorem 20.3 will be given over the next couple of lectures, but for the time being
we will apply it:
Example: Compute H k (S 1 ) using Mayer-Vietoris using the above decomposition of S 1 as R ∪ R.
20.3. Poincaré lemma. The following lemma is an important starting point when using the Mayer-
Vietoris sequence to compute cohomology groups.
Lemma 20.5 (Poincaré lemma). Let ω ∈ Ωk (Rn ) for k ≥ 1. Then ω is closed if and only if it is
exact.
In other words, H k (Rn ) = 0 for k ≥ 1. We will give the proof later, when we discuss homotopy-
theoretic properties of de Rham cohomology.
20.4. Partitions of unity.

Definition 20.6. Let {Uα } be an open cover of M . Then a collection of real-valued functions
{fα ≥ 0} on M is a partition of unity subordinate to {Uα } if:
(1) Supp(fα ) ⊂ Uα . Here the support Supp(fα ) of fα is the closure of {x ∈ M | fα (x) 6= 0}.
(2) At every point x ∈ M , there exists a neighborhood N (x) of x such that the set
{fα | fα |N (x) 6= 0}
Pk
is finite. If we write f1 , . . . , fk for the nonzero functions, then i=1 fi (x) = 1.
Proposition 20.7. For any open cover {Uα } of M , there exists a partition of unity subordinate to
{Uα }.
Proof. The proof is done in stages.
Step 1. Consider the function f : R → R:

e−1/x for x > 0
f (x) =
0 for x ≤ 0
It is easy to show that f ≥ 0 and f is smooth.
Step 2. Let a < b. Then the function
g = gab : R → R, x 7→ f (x − a) · f (b − x)
is a bump function, i.e., satisfies g ≥ 0; g > 0 on (a, b); and hence Supp(g) = [a, b].
Step 3. Construct a bump function on [a1 , b1 ] × · · · × [an , bn ] ⊂ Rn with coordinates (x1 , . . . , xn )
by letting φ(x) = ga1 b1 (x1 ) . . . gan bn (xn ). Then φ is supported on [a1 , b1 ] × · · · × [an , bn ] and is
positive on the interior.
Step 4. Suppose M is compact. For each p ∈ M , choose an open neighborhood Up of p of the
form (a1 , b1 ) × · · · × (an , bn ) whose closure is contained inside some Uα . For each Up , construct
φp as in Step 3. Now, since M is compact, there exists a finite collection of {p1 , . . . , pk } where
{Upi } cover M . Note that ki=1 φp > 0 everywhere on M . If we let ψp = Pφpφp , then
P P
ψp = 1.
Finally, we associate to each ψp an open set Uα for which Up ⊂ Uα . Then ψα is the sum of all the
ψp associated to Uα .
Step 5. Not so easy exercise: prove the general case!!
44 KO HONDA
21. S OME HOMOLOGICAL ALGEBRA

21.1. Short exact sequence. Suppose M = U ∪ V . Then we have
U V i ,i i
U ∩ V −→ U t V → M,
where iU and iV are two inclusions, one into U and the other into V .
Lemma 21.1. The following is a short exact sequence:
i∗ i∗ −i∗
(9) 0 → Ωk (M ) → Ωk (U ) ⊕ Ωk (V ) −→
U V
Ωk (U ∩ V ) → 0.
Proof. We prove that i∗U − i∗V is surjective. Let {ρU , ρV } be a partition of unity subordinate to
{U, V }; in particular ρU + ρV = 1. Given ω ∈ Ωk (U ∩ V ), we have ρV ω ∈ Ωk (U ) and −ρU ω ∈
Ωk (V ). Then i∗U (ρV ω) − i∗V (−ρU ω) = ω. The rest of the exact sequence is left as an exercise.
21.2. Short exact sequences to long exact sequences. Getting from the short exact sequence to
the long exact sequence is a purely algebraic operation.
d dk+1
Define a cochain complex (C, d): · · · → C k −→ k
C k+1 −→ C k+1 → . . . to be a sequence of
vector spaces and maps with dk+1 ◦ dk = 0 for all k. (C, d) gives rise to H k (C) = Ker dk / Im dk−1 ,
the kth cohomology of the complex.
A cochain map φ : A → B is a collection of maps φk : Ak → B k such that the squares of the
following diagram commute (i.e., dk ◦ φk = φk+1 ◦ dk ):
dk−2 dk−1 d dk+1
−−−→ Ak−1 −−−→ Ak −−−
k
→ Ak+1 −−−→
  

φk−1 y

φk y

φk+1 y
dk−2 dk−1 d dk+1

−−−→ B k−1 −−−→ B k −−−
k
→ B k+1 −−−→
A cochain map φ : A → B induces a map φ : H k (A) → H k (B) on cohomology. The

verification is identical to that of the special case of de Rham.
φ ψ
Given an exact sequence 0 → A → B → C → 0 of cochain maps, i.e., we have collections of
φ ψ
0 → Ak →k B k →k C k → 0 and commuting squares
x x x

dk+1 

dk+1 

dk+1 
φk+1 ψk+1
0 −−−→ Ak+1 −−−→ B k+1 −−−→ C k+1 −−−→ 0
x x x

dk 

dk 

dk 
φ
k k ψ
0 −−−→ Ak −−−→ Bk −−−→ Ck −−−→ 0
x x x

dk−1 

dk−1 

dk−1 
we obtain a long exact sequence:

k φ k ψ δ
. . . −−−→ H k (A) −−−→ H k (B) −−−→ H k (C) −−−
k
→
φk+1 ψk+1 δk+1
−−−→ H k+1 (A) −−−→ H k+1 (B) −−−→ H k+1 (C) −−−→ . . . .
Verification of Ker ψk ⊃ Im φk . Suppose [b] ∈ Im φk . Then b = φk a + db0 , where a ∈ Ak and

b0 ∈ B k−1 . Now, ψk b = ψk (φk a) + ψk (db0 ) = d(ψk−1 b0 ). Therefore, [ψk b] = 0 ∈ H k (C).
Verification of Ker ψk ⊂ Im φk . Suppose [b] ∈ Ker ψk . Then ψk b = dc0 , c0 ∈ C k−1 . Next, use the
fact that B k−1 → C k−1 → 0 to find b0 ∈ B k−1 such that c0 = ψk−1 b0 . Then ψk b = d(ψk−1 b0 ) =
ψk (db0 ). Hence, by the exactness, b − db0 = φk (a) for some a ∈ Ak . (Check that da = 0.) Thus,
φk [a] = [b].
Definition of δk : H k (C) → H k+1 (A). Let [c] ∈ H k (C). Then dc = 0. Also we have b ∈ B k
with ψk b = c by the surjectivity of B k → C k . Consider db. Since ψk+1 (db) = d(ψk b) = dc = 0,
there exists an a ∈ Ak+1 such that φk+1 a = db. Let [a] = δk [c]. Here, da = 0, since φk+2 (da) =
d(φk+1 a) = d(db) = 0, and Ak+2 → B k+2 is injective. We need to show that this definition is
independent of the choice of c, choice of b, and choice of a. This is left as an exercise.
HW: Verify the rest of the exactness.
46 KO HONDA
22. I NTEGRATION
n
Let M beR an n-dimensional manifold and ω ∈ Ω (M ). The goal of this lecture is to motivate
and define M ω.
22.1. Review of integration on Rn . For more details see Spivak.

Given a function f : R = [a1 , b1 ] × [an , bn ] → R, take a partition P = (P1 , . . . , Pk ) of R into
small rectangles. Consider the upper and lower bounds
X X
U (f, P ) := sup(f |Pi ) · vol(Pi ), L(f, P ) := inf(f |Pi ) · vol(Pi ).
i i
Definition 22.1. The function f : R → R is Riemann-integrable (or simply integrable) if for any
ε > 0 there exists a partition P such that U (f, P ) − L(f, P ) < ε. If f is integrable, then we define
Z
f dx1 . . . dxn := lim U (f, P ) = lim L(f, P ).
R P P
R
Similarly we define U
f dx1 . . . dxn , if U ⊂ Rn is an open set and f : U → R is a function with
compact support.
Theorem 22.2. If f is a continuous function with support on a compact set, then f is integrable.
∼
Change of variables formula. If g : [a, b] → [c, d] is a smooth reparametrization and x, y are
coordinates on [a, b], [c, d], then
Z g(b) Z b
f (y)dy = (f ◦ g)(x) · g 0 (x)dx.
g(a) a
This can be rewritten as:

Z Z
f (y)dy = (f ◦ g)(x) · |g 0 (x)|dx.
g([a,b]) [a,b]
More generally, let U and V ⊂ Rn be open sets with coordinates x = (x1 , . . . , xn ) and y =
∼
(y1 , . . . , yn ), and φ : U → V a diffeomorphism. Then:
Z Z
f (φ(x)) det ∂φ

f (y)dy1 . . . dyn = ∂x
dx1 . . . dxn .
V U
R
Remark 22.3. In light of the change of variables formula, M ω makes sense only when M is
orientable, since the change of variables for an n-form does not have the absolute value. At any
rate, n-forms have the wonderful property of having the correct transformation property (modulo
sign) under diffeomorphisms.
22.2. Orientation. Recall that M is orientable if there exists a subatlas {φα : Uα → Rn } such
that the Jacobians Jαβ = d(φβ ◦ φ−1
α ) have positive determinant. We will refer to such an atlas as
an oriented atlas.
Proposition 22.4. M is orientable if and only if there exists a nowhere zero n-form ω on M .
Proof. Suppose M is orientable. Let {φα : Uα → Rn } be an oriented atlas, i.e., a subatlas whose
transition functions have Jacobians Jαβ with positive determinant. Take a partition of unity {fα }
subordinate to Uα and let xα1 , . . . , xαn be the coordinates on Uα . Construct ωα = fα dx α α
P1 ∧ · · · ∧ dxn ,
which can be viewed as a smooth n-form on M with support in Uα . Let ω = α ωα . This is
nowhere zero, since
(φβ ◦ φ−1 ∗ −1 α α
α ) ωβ = fβ ◦ (φβ ◦ φα ) det(Jαβ )dx1 ∧ · · · ∧ dxn .
Since det(Jαβ ) is positive, fβ ◦ (φα ◦ φ−1 β ) det(Jαβ ) ≥ 0. At any point p ∈ M , at least one fα is
positive and all the terms that are added up contribute nonnegatively with respect to coordinates
dxα1 , . . . , dxαn , so ω is nowhere zero on M .
On the other hand, suppose there exists a nowhere zero n-form ω on M . We choose a subatlas
{φα : Uα → Rn } whose coordinate functions xα1 , . . . , xαn satisfy the condition that dxα1 ∧ · · · ∧ dxαn
is a positive function times ω. Then clearly Jαβ has positive determinant on this subatlas.
If M is orientable, then there is a nowhere zero n-form ω, which in turn implies that ∧n T ∗ M is
isomorphic to M × R as a vector bundle, i.e., is a trivial vector bundle.
On a connected manifold M , any two nowhere zero n-forms ω and ω 0 differ by a function, i.e.,
∃ a positive (or negative) function f such that ω = f ω 0 . We define an equivalence relation ω ∼ ω 0
if ω = f ω 0 and f > 0. Then there exist two equivalence classes of ∼ and each equivalence class is
called an orientation of M .
The standard orientation on Rn is dx1 ∧ · · · ∧ dxn .
Equivalent definition of orientation. Let F r(V ) be the set of ordered bases (or frames) of a
finite-dimensional R-vector space V of dimension n. It can be given a smooth structure so that it
is diffeomorphic to GL(V ), albeit not naturally: Fix an ordered basis (v1 , . . . , vn ). Then any other
basis (w1 , . . . , wn ) can be written as (Av1 , . . . , Avn ), where A ∈ GL(V ). Hence there is a bijection
F r(V ) ' GL(V ) and we induce a smooth structure on F r(V ) from GL(V ) via this identification.
The non-naturality comes from the fact that there is a distinguished point id ∈ GL(V ) but no
distinguished basis in F r(V ).
Since GL(V ) has two connected components, F r(V ) also has two connected components, and
each component is called an orientation for V . An orientation for M is a choice of orientation for
each Tp M which is smooth in p ∈ M .
We can also construct the frame bundle F r(M ) = tp∈M F r(Tp M ) by topologizing as follows:
Identify a neighborhood of p ∈ M with Rn and identify tp∈Rn F r(Tp Rn ) = F r(Rn ) × Rn . The
frame bundle is a fiber bundle over M whose fibers are diffeomorphic to GL(V ).
48 KO HONDA
22.3. Definition of the integral. Suppose M is orientable. Choose an oriented atlas {φα : Uα →
Rn } for M . We then define:
Z XZ
ω= (φ−1 ∗
α ) (fα ω),
M α φα (Uα )
R
where {fα } is a partition of unity subordinate to {Uα }. We will often write Uα fα ω instead of
(φ−1 ∗
R
φα (Uα ) α ) (fα ω).
R
HW: Check that the definition of M ω does not depend on the choice of oriented atlas {φα : Uα →
Rn } as well as on the choice of {fα } subordinate to {Uα }.
23. S TOKES ’ T HEOREM

23.1. Manifolds with boundary. We now enlarge the class of manifolds by allowing those “with
boundary”. These are locally modeled on the half-plane Hn = {x1 ≤ 0} ⊂ Rn with coordinates
x1 , . . . , xn . Let ∂Hn = {x1 = 0}.
Definition 23.1. An n-dimensional manifold with boundary is a pair (M, A = {φα : Uα → Hn }),
where M is a Hausdorff, second countable topological space. φα is a homeomorphism of an open
set Uα ⊂ M onto an open subset of Hn , and the transition functions φβ ◦ φ−1
α are smooth. The
boundary ∂M of M is the set of points of M that are mapped to ∂Hn under some φα .
Example: The n-dimensional unit ball B n = {(x1 , . . . , xn ) ∈ Rn |x21 + · · · + x2n ≤ 1}. ∂B n =
S n−1 .
Lemma 23.2. If M is a manifold with boundary, then ∂M is a manifold of dimension dim M − 1.
Proof. Let {(Uα , φα )} be an atlas for M . The key point that you should verify is that if any p ∈ M
is mapped to ∂Hn under some φα , then p cannot be mapped to the interior of Hn under any other
coordinate chart φβ . We then take the atlas {Vα } for ∂M where Vα = Uα ∩ φ−1 α ({x1 = 0}).
Lemma 23.3. If M is an orientable manifold with boundary, then ∂M is an orientable manifold.

Proof. Suppose that {(Uα , φα )} is an oriented atlas for M . If ( ∂x∂ 1 , ∂x∂ 2 , . . . , ∂x∂n ) is an oriented
basis for T M at each point of Uα , then let ( ∂x∂ 2 , . . . , ∂x∂n ) be an oriented basis for T (∂M ) at each
point of Uα ∩∂M . Since an outward-pointing vector ∂x∂ 1 will go to another outward-pointing vector
under a change of coordinates we have consistency.
Remark 23.4. The choice of orientation in the above lemma is the boundary orientation of ∂M
induced from the orientation of M .
23.2. Stokes’ Theorem.
Theorem 23.5 (Stokes’ Theorem).
R RLet ω be an (n−1)-form on an oriented manifold with boundary
M of dimension n. Then M dω = ∂M ω.
Zen: The significance of Stokes’ Theorem is that a topological operation ∂ is related to an analytic
operation d.
Proof. Take an open cover {Uα } where Uα is diffeomorphic to (i) (0, 1) × · · · × (0, 1) (i.e., Uα does
not intersect ∂M ) or (ii) (0, 1] × (0, 1) × · · · × (0, 1) (i.e., Uα ∩ ∂M = {x1 = 1}). (Note that in the
definition of a manifold with boundary, we could have allowed Hn to be {x1 ≤ c}, where c ∈ R
is a constant. Here we are taking R c = 1.) Let {f R α } be a partition of unity subordinate to {Uα }. By
linearity, it suffices to compute Uα d(fα ω) = ∂M ∩Uα fα ω, i.e., assume ω is supported on one Uα .
We will treat the n = 2 case. Let ω be an (n − 1)-form of type (ii). Then on [0, 1] × [0, 1] we
can write ω = f1 dx1 + f2 dx2 .
Z Z Z 1
ω= f1 dx1 + f2 dx2 = f2 (1, x2 )dx2 .
∂M ∂M 0
50 KO HONDA
On the other hand,

Z Z 1Z 1
∂f1 ∂f2
dω = − + dx1 dx2
M 0 0 ∂x2 ∂x1
Z 1 Z 1 Z 1 Z 1
∂f2 ∂f1
= dx1 dx2 + − dx2 dx1
0 0 ∂x1 0 0 ∂x2
Z 1 Z 1
= (f2 (1, x2 ) − f2 (0, x2 ))dx2 + (f1 (x1 , 0) − f1 (x1 , 1))dx1
0 0
Z 1
= f2 (1, x2 )dx2
0
HW: Check that n > 2 also works in the same way.
Example: (Green’s Theorem) Let Ω ⊂ R2 be a compact domain with smooth boundary, i.e., Ω is
a 2-dimensional manifold with boundary ∂Ω = γ. Then
Z Z
∂g ∂f
f dx + gdy = − dxdy.
γ Ω ∂x ∂y
Example: Consider ω defined on R2 − {0}:

−y x
ω(x, y) = dx + dy.
x2 + y 2 x2 + y 2
Let C = {x2 + y 2 = R2 }. Then x = R cos θ, y = R sin θ, and we compute
Z
ω = 2π.
C
One can directly verify that ω is closed. We claim that ω is not exact. Indeed, if ω = dη, then
Z Z
0= η= dη = 2π,
∂C C
which is a contradiction.
24. A PPLICATIONS OF S TOKES ’ T HEOREM

24.1. The Divergence Theorem.
Theorem 24.1. Let Ω ⊂ R3 be a compact domain with smooth boundary. Let F = (F1 , F2 , F3 ) be
a vector field on Ω. Then Z Z
div F dxdydz = hn, F idA,
Ω ∂Ω
where n = (n1 , n2 , n3 ) is the unit outward normal to ∂Ω,
dA = n1 dy ∧ dz + n2 dz ∧ dx + n3 dx ∧ dy,
and h·i is the standard inner product.
R Let ω R= F1 dydz + F2 dzdx + F3 dxdy. Then dω = (div F )dxdydz. It remains to see why
∂Ω
ω = ∂Ω hn, F idA.
Evaluating forms. We explain what it means to take ω(v1 , . . . , vk ), where ω is a k-form and vi are
tangent vectors. Let V be a finite-dimensional vector space. There exists a map:
(∧k V ∗ ) × (V × · · · × V ) → R
X
(f1 ∧ · · · ∧ fk , (v1 , . . . , vk )) 7→ (−1)σ f1 (vi1 ) . . . fk (vik ),
where the sum ranges over all permutations of (1, . . . , k) and σ is the number of transpositions
required for the transposition (1, . . . , k) 7→ (i1 , . . . , ik ). Note that this alternating sum is necessary
for the well-definition of the map.
∂ ∂

Example: Let ω = F1 dydz + F2 dzdx + F3 dxdy. Then ω ∂x , ∂z = −F2 .
Interior product. We can define the interior product as follows: iv : ∧k V ∗ → ∧k−1 V ∗ , iv ω =
ω(v, ·, ·, . . . , ·). (Insert v into the first slot to get a (k − 1)-form.)
Example: On R3 , let η = dxdydz. Also let n be the unit normal vector to ∂Ω. Then, along ∂Ω we
can define in η = n1 dydz + n2 dzdx + n3 dxdy.
Why is this dA? At any point of p ∈ ∂Ω, take tangent vectors v1 , v2 of ∂Ω so that n, v1 , v2
is an oriented orthonormal basis. Then the area form dA should evaluate to 1 on v1 , v2 . Since
η(n, v1 , v2 ) = 1 (since η is just the determinant), we see that dA = in η.
Explanation of hn, F idA = F1 dydz + F2 dzdx + F3 dxdy. Also note that iF η = F1 dydz + . . . .
But now, iF η(v1 , v2 ) = η(F, v1 , v2 ) = η(hn, F in, v1 , v2 ) = hn, F idA (by Gram-Schmidt).
24.2. Evaluating cohomology classes. In this section let M be a compact oriented n-manifold
without boundary.
R
Lemma
R 24.2. There exists a well-defined, nonzero map : H n (M ) → R which sends [ω] →
M
ω.
52 KO HONDA
R
Proof. Given ω ∈ Ωn (M R ), wen map ω 7→ M ω. Note that every n-form ω is closed. To show the
map descends to a map : H (M ) → R, let ω be an exact form, i.e., ω = dη. Then
Z Z Z
ω= dη = η = 0.
M M ∂M
R
R Next we prove the nontriviality of : if ω is an orientation form (i.e., ω is nowhere zero), then
M
ω > 0 or < 0, since on each oriented coordinate chart ω is some positive function times
dx1 . . . dxn .
The lemma implies that dim H n (M ) ≥ 1. In fact, we have the following, which will be proved
in a later lecture.
Theorem 24.3. H n (M ) ' R.
Example. If M = S n , then H i (S n ) = R for i = 0 or n and H i (S n ) = 0 for all other i.

25. E VALUATING COHOMOLOGY CLASSES ; DEGREE

25.1. Evaluating cohomology classes. Let φ : M → N be a smooth map between compact
oriented manifolds of dimensions m and n, respectively, and φ∗ :R H k (N ) → H k (M ) the induced
map on cohomology. If ω ∈ Ωm (N ) is a closed m-form such that M φ∗ ω 6= 0, then φ∗ ω represents
a nonzero element in H m (M ). This implies that [ω] is a nonzero cohomology class in H m (N ).
Example: On R2 − {0}, consider the closed 1-form

−y x
ω(x, y) = dx + dy.
x2 + y 2 x2 + y 2
We computed S 1 φ∗ ω = 2π, where φ : S 1 → R2 − {0} mapped θ 7→ (R cos θ, R sin θ). Since
R
[φ∗ ω] is a nonzero cohomology class in H 1 (S 1 ), so is [ω] ∈ H 1 (R2 − {0}).

Definition 25.1. Two maps φ0 , φ1 : M → N are (smoothly) homotopic if there exists a map
Φ : M × [0, 1] → N where Φ(x, t) = φt (x) for t = 0, 1.
Lemma 25.2. If φ0 , φ1 : M → N are homotopic and ω ∈ Ωm (N ), m = dim M , is closed, then
φ∗ ω = M φ∗1 ω.
R R
M 0
Proof. Z Z Z Z Z
φ∗1 ω − φ∗0 ω = ∗
Φω= ∗
d(Φ ω) = Φ∗ (dω) = 0,
M M ∂(M ×[0,1]) M M
since ω is closed.
Example, continued: Since ω is a closed 1-form on N = R2 − {0}, C ω = C 0 ω if C and C 0 are

R R
homotopic. That’s why the integral did not depend on the radius R of the circle.
25.2. Definition of degree. This material can be found in Guillemin & Pollack.
Let φ : M → N be a smooth map between compact oriented n-manifolds M and N . Let y ∈ N
be a regular value of φ. (Recall y ∈ N is a regular value if, for all x ∈ φ−1 (y), df (x) is surjective.
y ∈ N which is not a regular value is a critical value.)
Claim 25.3. φ−1 (y) is a finite set {x1 , . . . , xk }.
Proof. Suppose φ−1 (y) is infinite. By the compactness of M , there exists a sequence x1 , x2 , . . . of
distinct points in φ−1 (y) and an accumulation point x = limi→∞ xi , which itself must also be in
φ−1 (y). However, for every x ∈ φ−1 (y) there exists an open set Ux which maps diffeomorphically
onto an open set around y. Therefore, x could not have been the limit of xi ∈ φ−1 (y).
The claim implies that for a sufficiently small open set Vy containing y, φ−1 (Vy ) is a finite
disjoint union of open sets Ux1 , . . . , Uxk , each of which is diffeomorphic to Vy .
Definition 25.4. The degree of a smooth map φ : M → N between compact oriented n-manifolds
is the sum of orientation numbers ±1 for each xi in the preimage of a regular value y. Here the
54 KO HONDA
sign is +1 if the map from a neighborhood of xi to a neighborhood of y is orientation-preserving

and −1 otherwise.
The following theorem will be explained in a couple of lectures.
Theorem 25.5 (Degree Theorem). The degree of a mapping φ : M → N is well-defined.
26. S ARD ’ S T HEOREM

26.1. Sard’s Theorem and measure zero.
Theorem 26.1 (Sard). Let f : M → N be a smooth map. Then the set of critical values of f (i.e.,
the set of points y ∈ N such that df (x) is not surjective for some x ∈ f −1 (y)) has measure zero.
Definition 26.2. A set S ⊂ N has measure zero if there exists a countable atlas {(Ui , φi )}∞
i=1 such
that, for each φi (Ui ) ⊂ Rn (here dim N = n) and ε > 0, φi (Ui ∩ S) can be covered by a countable
union of rectangles [a1 , b1 ] × · · · × [an , bn ] with total volume ≤ ε.
From now on we will not distinguish between Ui and its image φi (Ui ) ⊂ Rn .
Remark 26.3. The measure in “measure zero” refers to the Lebesgue measure µ when restricted to
Rn . Note that the Lebesgue measure µ is multiplied by a positive smooth function under a change
of coordinates. Hence “measure zero” is easier to define than a measure on all of N .
Remark 26.4. If S ⊂ N has measure zero, then S itself can be covered by a countable union of
rectangles with total volume ≤ ε (here to each rectangle we need to assign the open set Ui with
i
respect to which we compute the volume): For Ui ,take rectangles so that the total volume ≤ ε 12 .
Adding up over all the Ui , we get ε 21 + 14 + . . . = ε.
Remark 26.5. Open subsets of Rn have nonzero Lebesgue measure. Hence the set of regular
values of a smooth map f : M → N is dense in N .
26.2. Proof of Sard’s Theorem. The proof closely follows that of Milnor, Topology from the
Differentiable Viewpoint.
It suffices to prove Sard’s Theorem in the following local situation.
Theorem 26.6. Let f : Rm → Rn be a smooth map. If we set C = {p | rank df (p) < n}, then
f (C) has measure zero in Rn .
Proof. We will prove the theorem for n = 1, i.e., f : Rm → R. The general case is left to the
reader.
Define the following subsets of Rm :
n o
m ∂f
C1 = p ∈ R ∂xi (p) = 0, ∀i = 1, . . . , m ,
Ck = {p ∈ Rm | all partial derivatives of f up to and including order k vanish at p}.
Then clearly C = C1 ⊃ C2 ⊃ C3 ⊃ . . . .
Strategy. We prove the theorem by induction on m. For m = 0 the theorem is clear. Assuming
the theorem holds for smooth maps Rm−1 → R, we show that for f : Rm → R:
(1) f (C1 − C2 ) has measure zero.
(2) f (Ck − Ck+1 ) has measure zero for any k ≥ 1.
(3) For k large enough (k ≥ m), f (Ck ) has measure zero.
56 KO HONDA
Step 1. Let p ∈ C1 − C2 . We claim there exists a neighborhood V of p for which f ((C1 − C2 ) ∩ V )

has measure zero. (This suffices because if we can cover C1 − C2 with countably many such V ’s,
the total measure of f (C1 − C2 ) is zero by the argument of Remark 26.4.) If p ∈ C1 − C2 , then
∂f ∂f 2f
∂x1
(p) = · · · = ∂x m
(p) = 0, but some ∂x∂i ∂x j
(p) 6= 0. Without loss of generality assume that
∂2f
∂x1 ∂xj
(p) 6= 0. Then consider the map
∂f
h : Rm → Rm , (x1 , . . . , xm ) 7→ ( ∂x j
, x2 , . . . , xm ).
Since the Jacobian is invertible at p, by the Inverse Function Theorem the map h restricts to a
∼
diffeomorphism V → V 0 , where V, V 0 are open sets containing p, f (p). Let (x̃1 , . . . , x̃m ) =
∂f
( ∂xj
, x2 , . . . , xm ) be coordinates on V 0 . Observe that:
• {critical values of f : V → R} = {critical values of f ◦ h−1 : V 0 → R}; and
• {critical values of f ◦ h−1 : V 0 → R} ⊂ {critical values of f ◦ h−1 : {x̃1 = 0} → R}.
We can then apply the inductive assumption to f ◦ h−1 whose domain is (m − 1)-dimensional.
Step 2. Similar to Step 1.
Step 3. Let p ∈ Ck , k ≥ m. In local coordinates we assume f : [− 12 , 12 ]m → R and p = 0. Taylor’s
Theorem with remainder gives us
f (x + h) = f (x) + R(x; h),
where |R(x; h)| ≤ C|h|k+1 , for all x ∈ Ck ∩ [− 21 , 12 ]m and |h| < δ, where δ > 0 is small.
We cover [− 21 , 12 ]m by cubes of length δ > 0 small. For this we need roughly δ1m cubes. Consider
one such cube Q which nontrivially intersects Ck . Then its volume is δ m , whereas its image under
f has length on the order of magnitude of δ k+1 by Taylor’s Theorem. Adding up the total volume of
the image, we obtain an upper bound of C δ1m δ k+1 , which can be made arbitrarily small by choosing
δ small.
27. D EGREE
27.1. H n (M ) of a n-dimensional manifold.
R
Theorem 27.1. If M is an oriented, compact n-manifold (without boundary), then : H n (M ) →
R is an isomorphism.
R
Proof.
R By Lemma 24.2, : H n (M ) → R is well-defined and nonzero. It remains to show that if
ω = 0, then ω is exact. Let {Ui } be a cover of M which is finite and has the property that every
Ui is diffeomorphic
P to Rn . Using a partition of unity {fα } Rsubordinate to {Ui }, we can split ω into
the sum i ωi , where ωi is supported inside Ui . Note that Ui ωi may not be zero.
Lemma 27.2. If ω is an n-form with compact support and zero integral on Rn , then ω = dη, where
η has compact support.
R∞
Proof. We will prove this for n = 2. Then ω = f (x, y)dxdy. Define g(x) = −∞ f (x, y)dy. By
R R∞
Fubini’s theorem and the hypothesis that R2 ω = 0, we have −∞ g(x)dx = 0. Define G(x, y) =
ε(y)g(x), where ε(y) is a bump function with total area 1. Then write:
Z y Z x
η(x, y) = − [f (x, t) − G(x, t)]dt dx + G(t, y)dt dy.
−∞ −∞
Clearly, dη = [f (x, y) − G(x, y)]dxdy + G(x, y)dxdy and η has compact support.
By Lemma 27.2 we can replace ωi by a cohomologically equivalent n-form which is supported

on a small neighborhood of a point xi ∈ M , i.e., we may assume that ωi is a bump n-form.
The total volume of the ωi is still zero. Now, engulf all the xi in an open set U ⊂ M which is
diffeomorphic to Rn so that ω is compactly supported in U and has total area zero. We use the
lemma again to complete the proof of Theorem 27.1.
27.2. Relationship to the degree. Recall the definition of degree: Let φ : M → N be a smooth
map between compact, oriented manifolds without boundary of dimension n. By Sard’s Theorem,
the set of regular values of φ has full measure in N . Let y ∈ N be a regular value and let φ−1 (y) =
Pk
{x1 , . . . , xk }. Then deg(φ) = i=1 ±1, where the contribution is +1 when φ is orientation-
preserving near xi and −1 is otherwise.
We will explain why the degree is well-defined. The map φ : M → N induces the map
φ∗ : H n (N ) → H n (M ) and we have a commutative diagram:
φ∗
H n (N ) −−−→ H n (M )
 
R R
y y
c
R −−−→ R
where the map R → R is multiplication by some real number c. The following proposition calcu-
lates this constant c to be deg φ:
58 KO HONDA
Proposition 27.3. deg φ satisfies

Z Z
∗
(10) φ ω = deg φ ω.
M N
Proof. Once we can prove Equation (10) for a suitable ω of our choice, the proposition R follows.
Take ω to be supported on a small open disk Vy about y with positive integral. Then M
φ∗ ω will
be the sum of Ux φ∗ ω, where Uxi are connected components of the preimage of Vy . Noting that
R
i
φ is a diffeomorphism from Uxi to Vy , we have Ux φ∗ ω = ± Vy ω, depending on whether the
R R
i
orientations agree or not. This proves Equation (10).
Corollary 27.4. The degree of φ is independent of the choice of regular value y.

28. L IE DERIVATIVES
We define the interior product on the linear algebra level as follows:5 If v ∈ V , then iv :
∧k V ∗ → ∧k−1 V ∗ is given by:
X
f1 ∧ · · · ∧ fk 7→ (−1)l+1 f1 ∧ . . . fl (v) · · · ∧ fk .
l
Check this is well-defined!

Let X be a vector field on M . Then the we can globalize the above pointwise definition to
obtain interior product iX : Ωk (M ) → Ωk−1 (M ) such that (iX ω)(x) = iX(x) ω(x). It satisfies the
following properties:
(1) For 1-forms ω, iX (ω) = ω(X).
(2) In general, we obtain the relation:
iX (α ∧ β) = (iX α) ∧ β + (−1)deg(α) α ∧ iX β.
Convention: We are taking Ω−1 (M ) = 0 and iX : Ω0 (M ) → Ω−1 (M ) to be the zero map.
Given ω ∈ Ωk (M ) and X1 , . . . , Xk ∈ X(M ), we define:
ω(X1 , . . . , Xk ) := iXk ◦ · · · ◦ iX1 ω.
In particular, if ω = df1 ∧ · · · ∧ dfk , then
X
ω(X1 , . . . , Xk ) = (−1)sgn(σ) df1 (Xσ(1) ) . . . dfk (Xσ(k) ).
σ∈Sk
Definition 28.1. The Lie derivative with respect to X is the map

LX := d ◦ iX + iX ◦ d : Ωk (M ) → Ωk (M ).
Lemma 28.2.
(1) If f ∈ Ω0 (M ), then LX f = d(iX f ) + iX (df ) = df (X) = X(f ). Hence LX : Ω0 (M ) →
Ω0 (M ) satisfies the (unsigned) Leibniz rule.
(2) LX (dω) = d(LX ω).
(3) LX : Ωk (M ) → Ωk (M ) satisfies LX (α ∧ β) = LX (α) ∧ β + α ∧ LX (β), i.e., the unsigned
Leibniz rule.
Proof. The proof is a simple computation and is left as an exercise.
By Lemma 28.2, LX : Ωk (M ) → Ωk (M ) naturally extends the derivation X : Ω0 (M ) →

Ω0 (M ). (We will usually call anything that satisfies the unsigned Leibniz rule a “derivation”.)
∼
The following is also a source of derivations Ωk (M ) → Ωk (M ): Let φt : M → M be a 1-
parameter family of diffeomorphisms, i.e., there exists a smooth map Φ : M × [−1, 1] → M such
5
This looks slightly different from the previous definition.
60 KO HONDA
that φt (x) := Φ(x, t), t ∈ [−1, 1], is a diffeomorphism. Assume in addition that φ0 = id. Then
d ∗
φ ω|t=0
dt t
is a derivation (verification is easy). If f ∈ Ω0 (M ), then
d ∗ d
φt f |t=0 = f (φt )|t=0 = df (X) = X(f ),
dt dt
where X is the vector field such that X(x), x ∈ M , is viewed as the arc φt (x), t ∈ [−1, 1], that
passes through x. We will often write X = dφ t
| .
dt t=0
∼
Proposition 28.3 (Cartan formula). If φt : M → M is a 1-parameter family of diffeomorphisms
such that φ0 = id, then dtd φ∗t |t=0 : Ωk (M ) → Ωk (M ) is given by d◦iX +iX ◦d, where X = dφ t
| .
dt t=0
Proof. It suffices to check the following:

• dtd φ∗t |t=0 and LX both satisfy the Leibniz rule. (Already verified!)
• dtd φ∗t |t=0 and LX agree on Ω0 (M ). (Yes, they are both vector fields.)
• d commutes with dtd φ∗t |t=0 and with LX .
The above three properties allow us to do an induction on the degree.
∼
Remark 28.4. We can generalize the Cartan formula slightly as follows: Let φt : M → M ,
t ∈ [a, b], be a 1-parameter family of diffeomorphisms and let Xt0 , t0 ∈ (a, b), be the vector field
on M where the arc φt (x) : [t0 − ε, t0 + ε] → M is assigned at the point φt0 (x) (note NOT at x).
Then
d ∗
φt ω|t=t0 = φ∗t0 (d ◦ iXt0 + iXt0 ◦ d)ω.
dt
29. H OMOTOPY PROPERTIES

29.1. Homotopy properties of de Rham cohomology.
∼
Proposition 29.1. Let φt : M → M , t ∈ [0, 1], be a 1-parameter family of diffeomorphisms. Then
φ∗t induce the same map H k (M ) → H k (M ) for all t ∈ [0, 1].
Proof. Consider a closed k-form ω on M . In view of Remark 28.4,
d ∗
φt ω|t=t0 = φ∗t0 (d ◦ iXt0 + iXt0 ◦ d)ω = d(φ∗t0 iXt0 ω).
dt
Rt
Therefore it is exact. Now, φ∗t ω − ω = 0 dtd φ∗t ω|t=s ds, and the difference is exact as well. (This is
evident by thinking of the integral as a limit of Riemann sums.)
Next we say two maps φ0 , φ1 : M → N are (smoothly) homotopic if there exists a smooth map
Φ : M × [0, 1] → N with φt (·) = Φ(·, t), t = 0, 1. φt is said to be the homotopy from φ0 to φ1 .
Proposition 29.2 (Homotopy invariance). If φt : M → N , t ∈ [0, 1], is a homotopy from φ0 to φ1 ,
then φ∗t : H k (N ) → H k (M ) is independent of t.
Proof. Consider Φ : M ×R → N . (There exists an extension Φ : M ×[0, 1] → N to Φ : M ×R →
N ; this is easy for example when M is compact.) We have inclusions it : M → M ×R, x 7→ (x, t),
and clearly φt = Φ ◦ it . Now take a diffeomorphism Ψt : M × R → M × R, (x, s) 7→ (x, s + t).
Since it = Ψt ◦ i0 , i∗t = i∗0 ◦ Ψ∗t . By the previous proposition, Ψ∗t is independent of t on the level
of cohomology. Hence so are i∗t and φ∗t .
29.2. Homotopy equivalence. We say φ : M → N is a homotopy equivalence if there exists

ψ : N → M such that φ ◦ ψ : N → N and ψ ◦ φ : M → M are homotopic to id : N → N and
id : M → M . Using Proposition 29.2, it is easy to show:
Proposition 29.3 (Homotopy equivalence). A homotopy equivalence φ : M → N induces an
isomorphism φ∗ : H k (N ) → H k (M ).
Proof. This is because φ∗ ◦ ψ ∗ = id and ψ ∗ ◦ φ∗ = id by homotopy invariance. This proves that φ∗
and ψ ∗ are left and right inverses (as linear maps) and are isomorphisms.
k
Corollary 29.4 (Poincaré lemma). HdR (Rn ) = 0 if k > 0.
Proof. We will show that Rn is homotopy equivalent to R0 = {pt}. Consider maps φ : Rn → R0 ,
(x1 , . . . , xn ) 7→ 0, and ψ : R0 → Rn , 0 7→ (0, . . . , 0). Clearly, φ ◦ ψ : R0 → R0 , 0 7→ 0, is the
identity map. Next, ψ ◦ φ : Rn → Rn , (x1 , . . . , xn ) 7→ 0 is homotopic to the identity map. In fact,
consider F : Rn × [0, 1] → Rn , ((x1 , . . . , xn ), t) 7→ (tx1 , . . . , txn ).
Example: Consider a band S 1 × (−1, 1). It is homotopy equivalent to S 1 . We have maps φ :

S 1 × (−1, 1) → S 1 , (θ, t) 7→ θ and ψ : S 1 → S 1 × (−1, 1), θ 7→ (θ, 0). φ ◦ ψ : S 1 → S 1
62 KO HONDA
is id. ψ ◦ φ : S 1 × (−1, 1) → S 1 × (−1, 1) is (θ, t) 7→ (θ, 0) is homotopic to id. In fact, take

F : S 1 × (−1, 1) × [0, 1] → S 1 × (−1, 1), (θ, t, s) 7→ (θ, ts). Therefore, we have:
H k (S 1 × (−1, 1)) ' H k (S 1 ).
Example: Similarly, H k (M × Rn ) ' H k (M ). More generally, if E is a vector bundle over M ,

then H k (E) ' H k (M ).
29.3. Extended example: surface of genus g. Consider a surface Σ of genus g. If you remove a
disk from Σ, you are left with a bouquet of 2g bands. You can now use Mayer-Vietoris with U a
disk and V a bouquet of 2g bands.
29.4. Euler characteristic. Let M be an n-dimensional manifold. Then we define the Euler
characteristic of M to be: n
X
χ(M ) = (−1)i dim HdR
i
(M ).
i=0
Examples:
(1) χ(Rn ) = 1.
(2) χ(S 2 ) = 1 + 0 + 1 = 2.
(3) χ(T 2 ) = 1 − 2 + 1 = 0.
(4) χ(genus g surface) = 2 − 2g.
Remark 29.5. For compact surfaces, the Euler characteristic is given by the classical formula
V − E + F , where V is the number of vertices, E is the number of edges, and F is the number of
faces of a polyhedron representing the surface.
HW: Prove that if 0 → C1 → C2 → · · · → Ck → 0 is an exact sequence, then
k
X
(−1)i dim Ci = 0.
i=1
Lemma 29.6. If M = U ∪ V , then χ(M ) = χ(U ) + χ(V ) − χ(U ∩ V ).

Proof. Use the Mayer-Vietoris sequence and add up the dimensions, using the above exercise.
30. V ECTOR FIELDS

Recall a vector field on U ⊂ M is a section of T M defined over U .
30.1. Lie brackets. Given two vector fields X and Y on M viewed as derivations at each p ∈ M ,
we can define its Lie bracket [X, Y ] = XY − Y X, i.e., for f ∈ C ∞ (M ),
[X, Y ](f ) = X(Y f ) − Y (Xf ).
Proposition 30.1. [X, Y ] is also a derivation, hence is a vector field.
Proof. This is a local computation. Take X = i ai ∂x∂ i and Y = j bj ∂x∂ j . Then:
P P
! !
X ∂ X ∂f X ∂ X ∂f
[X, Y ](f ) = ai bj − bj ai
i
∂x i j
∂x j j
∂x j i
∂xi
X ∂bj ∂f X ∂ai ∂f
= ai − bj
ij
∂x i ∂x j ij
∂xj ∂xi
In other words,
" #
X ∂ X ∂ X ∂bi ∂ai

∂
ai , bj = aj − bj .
i
∂xi j
∂xj ij
∂xj ∂xj ∂xi
Properties of Lie brackets.

(1) (Anticommutativity) [X, Y ] = −[Y, X].
(2) (Jacobi identity) [X, [Y, Z]] + [Y, [Z, X]] + [Z, [X, Y ]] = 0.
(3) [f X, gY ] = f g[X, Y ] + f X(g)Y − gY (f )X.
These properties are easy to verify, and are left as exercises.
30.2. Fundamental Theorem of Ordinary Differential Equations. Let X ∈ X(M ).

dγ dγ
Definition 30.2. A curve γ : (a, b) → M is an integral curve of X if dt
(t) = X(γ(t)). Here dt
(t)
∂
means γ∗ ( ∂t (t)).
Pn
With respect to local coordinates x = (x1 , . . . , xn ), if we write X = i=1 ai (x) ∂x∂ i and γ(t) =
x(t), then
dx
(t) = (a1 (x(t)), . . . , an (x(t))).
dt
Theorem 30.3 (Fundamental Theorem of ODEs). Given X ∈ X(M ) and p ∈ M , there exist an
open set U 3 p, ε > 0, and a smooth map Φ : U ×(−ε, ε) → M such that if we set γx (t) = Φ(x, t),
x ∈ U , then γx (0) = x and γx (t) is an integral curve of X. Moreover, if γ : (−δ, δ) → M is
another integral curve of X with γ(0) = x, then γ(t) = γx (t) on (−ε, ε) ∩ (−δ, δ).
64 KO HONDA
For the Fundamental Theorem of ODEs we need X to be at least C 1 . The proof of this theorem
will be omitted. We also write φt (x) := Φ(x, t) and refer to it as the time-t flow of X. Note that
φ0 (x) = id.
Remarks.
(1) If M is compact without boundary and X is a vector field on M , then there exists a global
flow Φ : M × R → M with φ0 = id by the uniqueness of the flow of X, where defined.
Since M is compact, the finite covering property ensures that we may choose ε to work for
all the open sets U . If we know that there is a flow for a short time ε, we can repeat the
flow and obtain a flow for an arbitrarily long time.
(2) However, if M is not compact, then there are vector fields X which do not admit global
short-time flows Φ : M × (−ε, ε) → M . (See the example below.)
(3) φs ◦ φt = φs+t and φ−1
t = φ−t . In particular, on a compact M , {φt }t∈R forms a 1-parameter
group of diffeomorphisms isomorphic to the additive group R.
∂
Example. On M = R − {0} consider X = ∂x . The vector field X, considered as a vector field on
R, clearly integrates to Φ : R × R → R, (x, t) 7→ x + t. However, when {0} is removed, no matter
how small an ε you take, there is no Φ : (R − {0}) × (−ε, ε) → R − {0}.
∂
Corollary 30.4. If X(p) 6= 0, then there exist coordinates (x1 , . . . , xn ) near p such that X = ∂x1
.
Proof. If M is n-dimensional, choose an (n − 1)-manifold Σ, defined in a neighborhood of p,
which is transverse to X. Here Σ is transverse to X if Tq Σ and X(q) span Tq M at all q ∈ Σ.
(Why does such a Σ exist?) Now take ψ : Σ × (−ε, ε) → M given by the flow Φ of X restricted
to Σ. Since Σ is transverse to X, ψ is a diffeomorphism near p by the inverse function theorem. If
x2 , . . . , xn are coordinates on Σ and x1 is the coordinate for (−ε, ε), then X can be written as ∂x∂ 1 .

31. V ECTOR FIELDS AND L IE DERIVATIVES

31.1. Pullback.
Lemma 31.1. Let f : M → N be a smooth map, ω ∈ Ωk (N ), and X1 , . . . , Xk ∈ X(M ). Then we
have:
(f ∗ ω)(x)(X1 , . . . , Xk ) = ω(f (x))(f∗ X1 , . . . , f∗ Xk ).
Proof. It suffices to show the lemma for 1-forms dg. (Why is that?) Then:
(f ∗ dg)(X) = d(g ◦ f )(X) = X(g ◦ f ) = f∗ X(g) = dg(f∗ X),
by definition of the pushforward of X.
AlternatePproof. If we write this in local coordinates x1 , . . . , xm and y1 , . . . , yn for M and N ,
∂g
then dg = i ∂y i
dyi and
X ∂g ∂yi
f ∗ dg = dxk .
i,k
∂yi ∂xk
Writing X = j bj ∂x∂ j , we obtain
P
!
X ∂ X ∂g ∂yi
f ∗ dg bj = bj ,
j
∂xj i,j
∂yi ∂xj
whereas
!! !
X ∂ X ∂g X ∂yl ∂ X ∂g ∂yi
dg f∗ bj = dyi bj = bj .
j
∂xj i
∂yi j,l
∂xj ∂yl i,j
∂yi ∂xj
31.2. Lie derivatives. Let X be a vector field on M . Assume there exists a flow Φ : M ×
(−ε, ε) → M , φt (x) = Φ(x, t), such that φ0 (x) = x. (Since we will be considering derivatives,
local flows suffice, but we will assume a global flow for ease of notation.)
The Lie derivative LX on forms ω is given by:
d
LX ω = φ∗t ω|t=0 .
dt
Lie derivatives can be defined on vector fields Y as well:
d
LX Y := (φ−t )∗ Y |t=0 .
dt
Here, vector fields cannot usually be pulled back, but for a diffeomorphism φ, there is a suitable
substitute, namely (φ−1 )∗ .
More generally, it is easy to see that LX can be defined on any section of ∧k T ∗ M ⊗ ∧l T M .
Properties of LX .
(1) LX f = Xf .
66 KO HONDA
(2) LX ω = (d ◦ iX + iX ◦ d)ω P
(3) LX (ω(X1 , . . . , Xk )) = (LX ω)(X1 , . . . , Xk ) + i ω(X1 , . . . , LX Xi , . . . , Xk ).
(4) LX Y = [X, Y ].
Proof. (1), (2) are already proven. (3) is left for homework. (For example, you can do this in local
coordinates; unwinding the definition needs to be done carefully.) We will prove (4), assuming (1),
(2), (3). We compute:
X(Y (f )) = LX (Y f )
= LX (df (Y ))
= (LX df )(Y ) + df (LX Y )
= (d ◦ LX f )Y + df (LX Y )
= d(X(f ))Y + (LX Y )(f )
= Y (X(f )) + (LX Y )(f ).
Therefore, (LX Y )f = X(Y (f )) − Y (X(f )) = [X, Y ](f ).
31.3. Interpretation of LX Y = [X, Y ]. As before, X, Y may not have global flows, but for
∼
simplicity let us assume they do. Let φs : M → M , s ∈ R, be the 1-parameter group of diffeomor-
∼
phisms generated by X and ψt : M → M , t ∈ R, be the 1-parameter group of diffeomorphisms
generated by Y . Noting that Y (x) = limt→0 ψt (x)−x
t
, we have
((φ−s )∗ Y )(x) − Y (x)

LX Y (x) = lim
s→0 s
(φ−s ◦ ψt ◦ φs (x) − x) − (ψt (x) − x)
= lim
s,t→0 st
φ−s ◦ ψt ◦ φs (x) − ψt (x)
= lim
s,t→0 st
−1
ψt ◦ φ−1

s ◦ ψt ◦ φs (x) − x
= lim ψt
s,t→0 st
−1
ψ ◦ φ−1 s ◦ ψt ◦ φs (x) − x
= lim t .
s,t→0 st
Hence, the Lie bracket [X, Y ] measures the infinitesimal discrepancy when you flow s units along
X, t units along Y , −s units along X and finally −t units along Y .
32. R ELATIONSHIP BETWEEN d AND [, ]; DISTRIBUTIONS

32.1. Relationship between d and [, ].
Proposition 32.1. If θ ∈ Ω1 (M ) and X, Y ∈ X(M ), then
dθ(X, Y ) = X(θ(Y )) − Y (θ(X)) − θ([X, Y ]).
Proof.
dθ(X, Y ) = iY iX dθ = iY (LX − d ◦ iX )θ
= iY (LX θ − d(θ(X)))
= (LX θ)Y − Y (θ(X))
= X(θ(Y )) − Y (θ(X)) − θ([X, Y ]).
Here we are using LX (θ(Y )) = (LX θ)Y + θ([X, Y ]) to get from the third line to the fourth.
More generally, for ω ∈ Ωk (M ) and X1 , . . . , Xk+1 ∈ X(M ):

X
dω(X1 , . . . , Xk+1 ) = (−1)i+1 Xi (ω(X1 , . . . , X
ci , . . . , Xk+1 ))
i
X
+ (−1)i+j ω([Xi , Xj ], X1 , . . . , X
ci , . . . , X
cj , . . . , Xk+1 ).
i<j
Here X
ci means omit the term with Xi . The proof is an exercise.
32.2. Distributions. Recall from Corollary 30.4 that if X is a vector field with X(p) 6= 0, then
there exist local coordinates (x1 , . . . , xn ) near p such that X = ∂x∂ 1 . Can we generalize this? If
X, Y are two vector fields such that X(p), Y (p) span a 2-dimensional subspace of Tp M , then the
span of X(x) and Y (x) is a 2-plane field for every x in a neighborhood of p.
Let M be an n-dimensional manifold.
Definition 32.2.
(1) A k-dimensional distribution D is a smooth choice of a k-dimensional subspace Dp ⊂ Tp M
at every point p ∈ M . By a smooth choice we mean that for each p ∈ M there exists a
small neighborhood Up of p and k linearly independent vector fields X1 , . . . , Xk on Up
which span Dq for each q ∈ Up . Alternatively, D is a rank k subbundle of T M .
(2) An integral submanifold N of M is a submanifold where Tp N ⊂ Dp at every p ∈ N .
dim N is not necessarily equal to dim D, but dim N ≤ dim D.
(3) A distribution D is integrable if M is covered by local coordinate charts (xα1 , . . . , xαn ) such
that D = Rh ∂x∂α , . . . , ∂x∂α i. Equivalently, D is integrable if there locally exist functions
1 k
f1 , . . . , fn−k such that {f1 = const, . . . , fn−k = const} are integral submanifolds of D
and the fi are independent, i.e., df1 ∧ · · · ∧ dfn−k 6= 0.
68 KO HONDA
Dually, we can define a k-dimensional distribution on M locally by prescribing n − k linearly

independent 1-forms ω1 , . . . , ωn−k .
∂ ∂
Example: On R3 with coordinates (x, y, z), consider ω = dz. Then D = ker ω = Rh ∂x , ∂y i. The
integral surfaces (i.e., 2-dimensional integral submanifolds) are z = const and D is an integrable
2-plane distribution.
Example: On R3 , consider ω = dz + (xdy − ydx). Then
∂ ∂ ∂ ∂
D = ker ω = Rh ∂x + y ∂z , ∂y − x ∂z i.
D is called a contact 2-plane distribution. We claim that D is not integrable. First calculate
ω ∧ dω = 2dx ∧ dy ∧ dz 6= 0.
If D = Rh ∂x∂ 1 , ∂x∂ 2 i for some coordinates (x1 , x2 , x3 ), then ω is of the form f dx3 . Now, dω =
df ∧ dx3 and ω ∧ dω = f dx3 ∧ df ∧ dx3 = 0. This is a contradiction.
33. F ROBENIUS ’ T HEOREM

Let M be an n-dimensional manifold. A distribution D of rank k is a rank k subbundle of
T M . Locally, D is defined as the span of independent vector fields X1 , . . . , Xk or as the kernel of
independent 1-forms ω1 , . . . , ωn−k .
Theorem 33.1 (Frobenius’ Theorem). A distribution D ⊂ T M of rank k is integrable if and only

if for all X, Y ∈ Γ(D), [X, Y ] ∈ Γ(D).
33.1. Proof of Frobenius’ Theorem. Suppose D ⊂ T MP is integrable. Then there

Pexist coordi-
nates x1 , . . . , xn so that D = Rh ∂x∂ 1 , . . . , ∂x∂ k i. Hence X = ki=1 ai (x) ∂x∂ i and Y = kj=1 bj (x) ∂x∂ j ,
and
k X k
X ∂bj ∂ ∂ai ∂
[X, Y ] = ai − bj ∈ Γ(D).
i=1 j=1
∂x i ∂x j ∂x j ∂x i
Suppose for all X, Y ∈ Γ(D), [X, Y ] ∈ Γ(D). We will find coordinates x1 , . . . , xn so that
D = Rh ∂x∂ 1 , . . . , ∂x∂ k i. Note that all our computations are local, so we restrict to M = Rn . We will
first do a slightly easier situation.
Proposition 33.2. Let X1 , . . . , Xk be independent vector fields with D = RhX1 , . . . , Xk i. If

[Xi , Xj ] = 0 for all i, j, then D is integrable.
Proof. We will deal with the case where dim D = 2 and M = R3 . Suppose P3 [X,∂ Y ] = 0. Using the
∂
fundamental theorem of ODE’s, we can write X = ∂x1 . Then Y = i=1 bi ∂xi , and [X, Y ] = 0
∂bi
implies that ∂x 1
= 0, i.e., bi = bi (x2 , x3 ) (there is no dependence on x1 ). Now take Y 0 = Y −
b1 X = b2 (x2 , x3 ) ∂x∂ 2 + b3 (x2 , x3 ) ∂x∂ 3 . If we project to R2 with coordinates x2 , x3 , then Y 0 can be
integrated to ∂x∂ 0 , after a possible change of coordinates. Therefore, D = Rh ∂x∂ 1 , ∂x∂ 0 i.
2 2
Still assuming dim D = 2 and M = R3 , suppose [X, Y ] = AX + BY . Without loss of

generality, X = ∂x∂ 1 and Y = b2 ∂x∂ 2 + b3 ∂x∂ 3 . Then,
∂b2 ∂ ∂b3 ∂ ∂ ∂ ∂
[X, Y ] = + =A + Bb2 + Bb3 .
∂x1 ∂x2 ∂x1 ∂x3 ∂x1 ∂x2 ∂x3
∂b2 ∂b3
This implies: A = 0, ∂x1
= Bb2 , ∂x1
= Bb3 . Hence,
R t=x1 R t=x1
B(t,x2 ,x3 )dt B(t,x2 ,x3 )dt
b2 = f (x2 , x3 )e t=0 , b3 = g(x2 , x3 )e t=0 .
R
Therefore, Y = e B (f (x2 , x3 ) ∂x∂ 2 +g(x2 , x3 ) ∂x∂ 3 ), and by rescaling Y we get Y 0 = f (x2 , x3 ) ∂x∂ 2 +
g(x2 , x3 ) ∂x∂ 3 . As before, now Y 0 can be integrated to give ∂x∂ 0 .
2
HW: Write out a general proof.

70 KO HONDA
33.2. Restatement in terms of forms. If D has rank k on M of dimension n, then locally there
exist 1-forms ω1 , . . . , ωn−k such that D = {ω1 = · · · = ωn−k = 0}.
Pn−k
Proposition 33.3. D is integrable if and only if dωi = j=1 θij ∧ ωj , where θij are 1-forms.
Proof. We use the identity
(11) dω(X, Y ) = Xω(Y ) − Y ω(X) − ω([X, Y ]).
Let X, Y ∈ Γ(D). Then the identity simplifies to:
(12) dω(X, Y ) = −ω([X, Y ]).
Pn−k
Suppose dωi = j=1 θij ∧ ωj . Then dωi (X, Y ) = 0. Hence ωi ([X, Y ]) = 0 for all i by
Equation (12), which implies that [X, Y ] ∈ Γ(D).
Suppose D is integrable. Complete ω1 , . . . , ωn−k to a (pointwise) basis by adding η1 , . . . , ηk .
Then
X Xk X n−k X
dωi = aij ωi ∧ ωj + bij ηi ∧ ωj + cij ηi ∧ ηj .
i<j i=1 j=1 i<j
By Equation (12), dωi (X, Y ) = 0 for X, Y ∈ Γ(D). Taking X1 , . . . , Xk ∈ Γ(D) dual to η1 , . . . , ηk ,
we find that dωi (Xr , Xs ) = crs (or −csr ). This proves that all the cij are zero.
34. C ONNECTIONS
34.1. Definition. Let E be a rank k vector bundle over M and let s be a section of E. s may
be local (i.e., in Γ(E, U )) or global (i.e., in Γ(E, M )). Also let X be a vector field. We want to
differentiate s at p ∈ M in the direction of X(p) ∈ Tp M .
Definition 34.1. A connection or covariant derivative ∇ assigns to every vector field X ∈ X(M )
a differential operator ∇X : Γ(E) → Γ(E) which satisfies:
(1) ∇X s is R-linear in s, i.e., ∇X (c1 s1 + c2 s2 ) = c1 ∇X s1 + c2 ∇X s2 if c1 , c2 ∈ R.
(2) ∇X s is C ∞ (M )-linear in X, i.e., ∇f X+gY s = f ∇X s + g∇Y s.
(3) (Leibniz rule) ∇X (f s) = (Xf )s + f ∇X s.
Note: The definition of connection is tensorial in X (condition (2)), so (∇X s)(p) depends on s
near p but only on X at p.
34.2. Flat connections. We will now present the first example of a connection.
A vector bundle E of rank k is said to be trivial or parallelizable if there exist sections s1 , . . . , sk ∈
Γ(E, M ) which span Ep at every p ∈ M . Although not every vector bundle is parallelizable,
locally every vector bundle is trivial since E|U ' U × Rk . We will now construct connections on
trivial bundles.
Write any section s as s = i fi si , where fi ∈ C ∞ (M ). Then define
P
X
∇X s = (Xfi )si = (Xf1 )s1 + · · · + (Xfk )sk ∈ Γ(E).
i
This connection is usually called a flat connection.

HW: Check that this satisfies the axioms of a connection.
Note that ∇X si = 0 for all X ∈ X(M ). Sections s satisfying such a property are said to be
covariant constant.
Important remark: We can define a connection ∇ for each trivialization E|U = U ×Rk , and there
is nothing canonical about the connection ∇ above. (It depends on the choice of trivialization.)
The space of connections is a large space (to be made more precise later).
Proposition 34.2. Any two covariant constant frames s1 , . . . , sk and s1 , . . . , sk differ by an element
of GL(k, R).
Proof. Let s1 , . . . , sk be another covariant constant frame, i.e., ∇X si = 0. Since we can write
X
si = fij sj ,
j
72 KO HONDA
with fij ∈ C ∞ (M ), we have:

X
0 = ∇ X si = ∇X (fij sj )
j
X
= [(Xfij )sj + fij ∇X sj ]
j
X
= (Xfij )sj .
j
This proves that Xfij = 0 for all X and hence fij = const.
Therefore, a flat connection determines a covariant constant frame {s1 , . . . , sk } up to an element

of GL(k, R).
35. M ORE ON CONNECTIONS

35.1. Preliminaries on vector bundles. Let E be a vector bundle over M and φ : N → M be a
smooth map. Then we can define the pullback bundle φ−1 E over N as follows:
(1) The fiber (φ−1 E)n over n ∈ N is the fiber Eφ(n) over φ(n) ∈ M .
∼
(2) There exist sufficiently small open sets V ⊂ N , so that φ(V ) ⊂ U and ϕU : E|U → U ×Rk .
The trivialization φ−1 E|V ' V × Rk is induced from this.
Next, if E and F are vector bundles over M , then we can define E ⊕ F as follows:
(1) The fiber (E ⊕ F )m over m ∈ M is Em ⊕ Fm .
∼ ∼
(2) Take U ⊂ M small enough so that E|U → U × Rk and F |U → U × Rl . Then we get
(E ⊕ F )|U ' U × (Rk ⊕ Rl ).
E ⊗ F is defined similarly.
35.2. Existence. Let M be an n-dimensional manifold and E be a rank k vector bundle over M .
Recall a connection ∇ is a way of differentiating sections of E in the direction of a vector field X.
∇X : Γ(E) → Γ(E),
∇X (f s) = (Xf )s + f ∇X s.
Definition 35.1. A connection ∇ on E is flat if there exists an open cover {Uα } of M such that
E|Uα admits a covariant constant frame s1 , . . . , sk .
Proposition 35.2. Connections exist on any vector bundle E over M .
Note that if E is parallelizable we have already defined connections globally on E. The key point
is to pass from local to global when E is not globally trivial.
Let ∇0 and ∇00 be two connections on E|U . Let us see whether ∇0 + ∇00 is a connection.
(∇0X + ∇00X )(f s) = ∇0X (f s) + ∇00X (f s)
= (Xf )s + f ∇0X s + (Xf )s + f ∇00X s
= 2(Xf )s + f (∇0X + ∇00X )s.
This is not quite a connection, since 2(Xf ) should be Xf instead. However, a simple modification
presents itself:
Lemma 35.3. Suppose λ1 , λ2 ∈ C ∞ (U ) satisfies λ1 + λ2 = 1. Then λ1 ∇0 + λ2 ∇00 is a connection
on E|U .
Proof. HW.
Proof of Proposition 35.2. Let {Uα } be an open cover such that E|Uα is trivial. Let ∇α be a flat
connection on E|Uα associated P to some trivialization. Next let {fα } be a partition of unity sub-
ordinate to {Uα }. Then form α fα ∇α . By the previous lemma, the Leibniz rule is satisfied.

74 KO HONDA
Remark: Although each of the pieces ∇α is flat before patching, the patching destroys flatness.
There is no guarantee that (even locally) there exist sections s1 , . . . , sk which are covariant con-
stant. In fact, for a generic connection, there is not even a single covariant constant section. One
way of measuring the failure of the existence of covariant constant sections is the curvature.
35.3. The space of connections. Given two connections ∇ and ∇0 , we compute their difference:
(∇X − ∇0X )(f s) = f (∇X − ∇0X )s.
Therefore, the difference of two connections is tensorial in s.
Locally, take sections s1 , . . . , sk (not necessarily covariant constant). Then (∇X − ∇0X )si =
0
P
j aij sj , where (aij ) is a k × k matrix of functions. In other words, ∇ − ∇ can be represented
by a matrix A = (Aij ) of 1-forms Aij . Here aij = Aij (X). Hence, locally it makes sense to write:
∇ = d + A.
fi si corresponds to (f1 , . . . , fk )T and more precisely
P
Here s =
∇(f1 , . . . , fk )T = d(f1 , . . . , fk )T + A(f1 , . . . , fk )T .
Globally, ∇−∇0 is a section of T ∗ M ⊗End(E). Here End(E) = Hom(E, E). The space of such
sections is often written as Ω1 (End(E)) and a section is called a “1-form with values in End(E)”.
This proves:
Proposition 35.4. The space of connections on E is an affine space Ω1 (End(E)).
Remark: We view Ω1 (End(E)) not as a vector space (which has a preferred zero element) but
rather as an affine space, which is the same thing except for the lack of a preferred zero element.
36. C URVATURE
Let E → M be a rank r vector bundle and ∇ be a connection on E.
Definition 36.1. The curvature R∇ (or simply R) of a connection ∇ is given by:
R(X, Y ) = [∇X , ∇Y ] − ∇[X,Y ] = ∇X ∇Y − ∇Y ∇X − ∇[X,Y ] ,
or
R(X, Y )s = [∇X , ∇Y ]s − ∇[X,Y ] s.
Proposition 36.2.
(1) R(X, Y )s is tensorial, i.e., C ∞ (M )-linear, in each of X, Y , and s.
(2) R(X, Y ) = −R(Y, X).
Proof. (2) is easy. For (1), we will prove that R(X, Y ) is tensorial in s and leave the verification
for X and Y as an exercise.
R(X, Y )(f s) = (∇X ∇Y − ∇Y ∇X )(f s) − ∇[X,Y ] (f s)
= ∇X ((Y f )s + f ∇Y s) − ∇Y ((Xf )s + f ∇X s) − (([X, Y ]f )s + f ∇[X,Y ] s)
= f R(X, Y )s

P
P The flat connection ∇X s =
Proposition 36.3. (Xfk )sk has R = 0. (Here s1 , . . . , sr trivializes
E|U and s = fk sk .)
Proof. We use X = ∂x∂ i , Y = ∂x∂ j . Since R(X, Y ) is tensorial, it suffices to compute it for our
choices.
X
∂ ∂
R , s = ∇ ∂ ∇ ∂ − ∇ ∂ ∇ ∂ f k sk
∂xi ∂xj k
∂xi ∂xj ∂xj ∂xi

X ∂fk ∂fk
= ∇ ∂ sk − ∇ ∂ sk
l
∂xi ∂x j ∂xj ∂x i
X ∂ 2 fk ∂ 2 fk

= − sk = 0.
∂xi ∂xj ∂xj ∂xi

36.1. Interpretations of curvature. Think of ∇ as d + A in local coordinates if necessary. We

have a sequence:
∇ ∇
Ω0 (E) → Ω1 (E) → Ω2 (E) → . . . .
76 KO HONDA
The first map is covariant differentiation (interpreted slightly differently). It turns out that this
sequence is not a chain complex, i.e., ∇ ◦ ∇ = 6 0 usually. In fact the obstruction to this being a
chain complex is the curvature. Let us locally write:
∇ ◦ ∇s = (d + A)(d + A)s = (d2 + Ad + dA + A ∧ A)s = (dA + A ∧ A)s.
Proposition 36.4. R = dA + A ∧ A, i.e., R(X, Y )s = (dA + A ∧ A)(X, Y )s.
Proof. It suffices to prove the proposition for X = ∂x∂ i , Y = ∂x∂ j , and s = sk , where s1 , . . . , sr is a
local frame for E|U . A is an r ×r matrix of 1-forms (Atij dxt ). (We will use the Einstein summation
convention – if the same index appears twice we assume it is summed over this index.) Then we
compute:
∂Ajmk
(13) ∇ ∂ ∇ ∂ sk = ∇ ∂ (sm Ajmk ) = sm + sn Ainm Ajmk
∂xi ∂xj ∂xi ∂xi
The computation of the rest is left as an exercise.
37. R IEMANNIAN METRICS , L EVI -C IVITA

37.1. Leftovers from last time. Last time we defined the curvature R∇ of a connection ∇. Lo-
cally, if ∇ is given by d + A, then R = dA + A ∧ A.
Theorem 37.1. ∇ is a flat connection if and only if R∇ ≡ 0.
We have already shown the easy direction: If ∇ is flat, then R∇ ≡ 0. The other direction will
be omitted for now (probably will be given next semester), since a “good proof” will take us a bit
far afield. Our only comment is that R = dA + A ∧ A = 0 or dA = A ∧ A looks a lot like the
Frobenius integrability condition given in terms of forms....
Corollary 37.2. Let E be a rank r vector bundle over R and ∇ be a connection on E. Then ∇ is
flat.
Proof. This is because all 2-forms on R are zero.
Remark: There are lots of connections which are not flat, since it is easy to find A so that dA +
A ∧ A 6= 0.
37.2. Riemannian metrics.

Definition 37.3. A Riemannian metric h, i or g on M is a positive definite symmetric bilinear form
(or inner product) g(x) : Tx M × Tx M → R which is smooth in x ∈ M .
Recall: A bilinear form h, i : V × V → R is positive definite if hv, vi ≥ 0 and hv, vi = 0 if and
only if v = 0. h, i is symmetric if hv, wi = hw, vi.

n ∂ ∂
Example: On R take the standard Euclidean metric g ∂xi , ∂xj = δij . This is usually written as
g = i dxi ⊗ dxi . Any other Riemannian metric on Rn can be written as g(x) = ij gij (x)dxi ⊗
P P
dxj , where gij (x) = gji (x).

Proposition 37.4. Every manifold M admits a Riemannian metric.
Proof. Let {Uα } be an open cover so that Uα ' Rn . On each Uα , we take P the standard Euclidean
metric gα . Now let {fα } be a partition of unity subordinate to {Uα }. Then α fα gα is the desired
metric.
The pair (M, g) of a manifold M together with a Riemannian metric g on M is called a Riemannian
manifold.
Let i : N → (M, g) be an embedding or immersion. Then the induced Riemannian metric i∗ g on
N is defined as follows:
i∗ g(x)(v, w) = g(i(x))(i∗ v, i∗ w),
where v, w ∈ Tx N . The injectivity of i∗ is required for the positive definiteness.
78 KO HONDA
37.3. Levi-Civita connections. Connections on T M → M have extra structure because X and

Y are the same type of object in the expression ∇X Y . In fact, we can define the torsion:
T∇ (X, Y ) = ∇X Y − ∇Y X − [X, Y ].
Proposition 37.5. T∇ (X, Y ) is tensorial in X and Y .
This is an easy exercise and is left as an exercise. (Note that the notion of torsion does not depend
at all on the Riemannian metric.) We say ∇ is torsion-free if T∇ = 0.
Definition 37.6. ∇ is compatible with g if XhY, Zi = h∇X Y, Zi + hY, ∇X Zi. Here X, Y, Z ∈
X(M ).
Theorem 37.7. Let (M, g) be a Riemannian manifold. Then there exists a unique torsion-free
connection which is compatible with g.
Proof. For any vector fields X, Y, Z, we have:
(14) XhY, Zi = h∇X Y, Zi + hY, ∇X Zi,
(15) Y hX, Zi = h∇Y X, Zi + hX, ∇Y Zi,
(16) ZhX, Y i = h∇Z X, Y i + hX, ∇Z Y i.
Taking (14) + (15) − (16), we get:
(17) 2h∇X Y, Zi + h[Y, X], Zi + h[X, Z], Y i + h[Y, Z], Xi = XhY, Zi + Y hX, Zi − ZhX, Y i,
and solving for h∇X Y, Zi, we get:
1
(18) h∇X Y, Zi = (h[X, Y ], Zi + h[Z, X], Y i + h[Z, Y ], Xi + XhY, Zi + Y hX, Zi − ZhX, Y i).
2
It is clear that the values of Equation 18 determine ∇. It remains to show that Equation 18 indeed
defines a connection which satisfies the torsion-free and compatibility conditions. It is clear that
h∇X Y, Zi = h∇Y X, Zi when X, Y, Z are of the form ∂x∂ i , and that Equation 14 can be recovered
from Equation 18. The details are left as an exercise.
The unique torsion-free, compatible connection is called the Levi-Civita connection for (M, g).
38. S HAPE OPERATOR

Let Σ be a surface embedded in the standard Euclidean (R3 , g), and let g be the induced metric
on Σ. We will denote the Levi-Civita connection on (R3 , g) by ∇ and the Levi-Civita connection
on (Σ, g) by ∇.
∂
Claim: ∇ satisfies ∇ ∂
∂xj
= 0.
∂xi
d
The verification is easy. The claim implies that ∇X Y is simply dt
Y (γ(t))|t=0 , where γ(t) is the
arc representing X at a given point.
What we will do today is valid for hypersurfaces ((n − 1)-dimensional submanifolds) M inside
(N, g) of dimension n, but we will restrict our attention to N = R3 for simplicity.
Definition 38.1. Let X, Y be vector fields of R3 which are tangent to Σ, and let N be the unit
normal vector field to Σ inside R3 . The shape operator is S(X, Y ) = h∇X Y, N i. In other words,
S(X, Y ) is the projection in the N -direction of ∇X Y .
Proposition 38.2. S(X, Y ) is tensorial in X, Y and is symmetric.
Proof. S(X, Y ) = S(Y, X) follows from the torsion-free condition and the fact that [X, Y ] is
tangent to Σ. Now,
S(f X, Y ) = h∇f X Y, N i = hf ∇X Y, N i = f S(X, Y ).
Tensoriality in Y is immediate from the symmetric condition.
Remark: The shape operator is usually called the second fundamental form in classical differential
geometry and measures how curved a surface is. (In case you are curious what the first fundamental
form is, it’s simply the induced Riemannian metric.)
Also observe that S(X, Y ) = h∇X Y, N i = h∇X N, Y i, by using the fact that hY, N i = 0 (since
N is a normal vector and Y is tangent to Σ).
38.1. Induced connection vs. Levi-Civita. If X, Y ∈ X(M ), we can write:
∇X Y = ∇hX Y + S(X, Y )N,
where ∇hX Y denotes the projection of ∇X Y onto T Σ.
Proposition 38.3. ∇h = ∇, i.e., ∇h is the Levi-Civita connection of (Σ, g).
Proof. We have defined ∇hX Y = ∇X Y − S(X, Y )N . It is easy to verify that ∇h satisfies the
properties of a connection on Σ.
∇h is torsion-free:
∇hX Y − ∇hY X = (∇X Y − S(X, Y )N ) − (∇Y X − S(Y, X)N )
= ∇X Y − ∇ Y X
= [X, Y ]
80 KO HONDA
∇h is compatible with g:
XhY, Zi = h∇X Y, Zi + hY, ∇X Zi
= h∇hX Y, Zi + hY, ∇hX Zi,
since hN, Xi = 0 for any vector field N on Σ.
It seems miraculous that somehow the induced connection is a Levi-Civita connection. Classically,
the induced covariant derivative came first, and Levi-Civita came as an abstraction of the covariant
derivative.
39. G AUSS ’ T HEOREMA E GREGIUM

Let (Σ, g) be a 2-dimensional Riemannian submanifold of the standard Euclidean (R3 , g). The
shape operator is a symmetric bilinear form:
S : Tx Σ × Tx Σ → R,
S(X, Y ) = h∇X Y, N i
where N is a unit normal to Σ, X, Y are vectors in Tx Σ which are extended to an arbitrary vector
field tangent to Σ, and ∇ is the Levi-Civita connection for (R3 , g). We can represent S(x), x ∈ Σ,
as a matrix by taking an orthonormal basis {e1 , e2 } at Tx Σ and taking the entries S(ei , ej ). The
trace of this matrix is called the mean curvature and the determinant is called the scalar curvature
or the Gaußian curvature.
Denote by ∇ the Levi-Civita connection for g and ∇ the Levi-Civita connection for g. Also
write R = R∇ and R for R∇ .
Theorem 39.1 (Gauß’ Theorema Egregium). If X, Y are vector fields on Σ, then
hR(X, Y )Y, Xi = S(X, X)S(Y, Y ) − S(X, Y )2 .
What this says is that the right-hand side, an extrinsic quantity (depends on the embedding into 3-
space) is equal to the left-hand side, an intrinsic quantity (only depends on the Riemannian metric
g and not on the particular embedding into R3 ). Therefore, the scalar curvature is expressed purely
in terms of the curvature of the induced metric.
Proof. Let N be the unit normal vector to Σ.
h∇X ∇Y Y, Xi = Xh∇Y Y, Xi − h∇Y Y, ∇X Xi
= Xh∇Y Y − S(Y, Y )N, Xi − h∇Y Y − S(Y, Y )N, ∇X X − S(X, X)N i
= Xh∇Y Y, Xi − h∇Y Y, ∇X Xi + hS(X, X)N, ∇Y Y i
= h∇X ∇Y Y, Xi + S(X, X)S(Y, Y ).
Similarly,
h∇Y ∇X Y, Xi = h∇Y ∇X Y, Xi − S(X, Y )2 ,
h∇[X,Y ] Y, Xi = h∇[X,Y ] Y, Xi.
Finally,
hR(X, Y )Y, Xi = hR(X, Y )Y, Xi + S(X, X)S(Y, Y ) − S(X, Y )2
= S(X, X)S(Y, Y ) − S(X, Y )2 .

82 KO HONDA
40. E ULER CLASS

40.1. Compatible connections. Let E be a rank k vector bundle over a manifold M . A fiber
metric is a family of positive definite inner products h, ix : Ex × Ex → R which varies smoothly
with respect to x ∈ M . A connection ∇ is compatible with h, i if Xhs1 , s2 i = h∇X s1 , s2 i +
hs1 , ∇X s2 i, for all vector fields X and s1 , s2 ∈ Γ(E).
Remark: We can view the Riemannian metric g on M as a fiber metric of T M → M . When we
think of T M as a vector bundle over M , we forget the fact that T M was derived from M .
Let U ⊂ M be an open set over which E is trivializable, and let {s1 , s2 , . . . , sk } be an orthonormal
frame of E over U . An orthonormal frame can be obtained by starting from some frame of E over
U and applying the Gram-Schmidt orthogonalization process.
With respect to {s1 , . . . , sk } we can write ∇ = d + A, where A is a k × k matrix with entries which
are 1-forms on U .
Lemma 40.1. A is a skew-symmetric matrix, i.e., AT = −A.
Proof. If we write A = (Akij dxk ), then we have ∇ ∂ sj = si Akij .
∂xk
∂
hsi , sj i = h∇ ∂ si , sj i + hsi , ∇ ∂ sj i,
∂xk ∂xk ∂xk
so we have
Akij = −Akij .

Lemma 40.2. Let {s01 , . . . , s0k } be another orthonormal frame for E over U . If g : U → SO(k) is
the transformation sending coordinates with respect to si to coordinates with respect to s0i (by left
multiplication), then the connection matrix transforms as: A 7→ g −1 dg + g −1 Ag.
Proof.
g −1 (d + A)g = g −1 dg + g −1 gd + g −1 Ag
= d + (g −1 dg + g −1 Ag).
You may want to check that if A is skew-symmetric and g is orthogonal, then g −1 dg + g −1 Ag is
also skew-symmetric.
40.2. Rank 2 case. Suppose from now on that E has rank 2 over M of arbitrary dimension. Then
AU (the connection matrix over U with respect to some trivialization) is given by

0 A21
AU = .
−A21 0
Then the curvature matrix RU is

0 ωU
RU = dAU + AU ∧ AU = ,
−ωU 0
where ωU is the 2-form dA21 .
Theorem 40.3. There is a global closed 2-form ω which coincides with ωU on each open set U .
2
Hence a connection ∇ on E gives rise to an element [ω] ∈ HdR (M ). This cohomology class is
independent of the choice of connection ∇ compatible with h, i, and hence is an invariant of the
vector bundle E. It is called the Euler class of E and is denoted e(E).
Proof. We need to show that on overlaps U ∩ V , ωU = ωV . If g : U ∩ V → SO(2) is the
orthogonal transformation taking from U to V , then we compute R with respect to the connection
1-form g −1 dg + g −1 AU g. It is not hard to see that we still get

0 ωU
R= .
−ωU 0
Now, two different connections ∇ and ∇0 have difference in Ω1 (End(E)). (Moreover, they have
values in 2 × 2 skew-symmetric matrices.) It is not hard to see that if we pick out the upper right
hand corner of the matrix on each local coordinate chart U , then they coincide and yield a global
1-form α, and the difference between R∇ and R∇0 will be the exact form dα.
Example: For the Levi-Civita connection ∇ on a surface (Σ, g) ,→ (R3 , g), we have, locally,

0 κθ1 ∧ θ2
RU = ,
−κθ1 ∧ θ2 0
where κ is the scalar curvature, {e1 , e2 } is an orthonormal frame, and {θ1 , θ2 } is dual to the frame
(called the dual coframe). (The fact that κ is the scalar curvature is the content of the Theorema
Egregium!)
40.3. The Gauß-Bonnet Theorem. Let (M, g) be an oriented Riemannian manifold of dimension
n. Then there exists a naturally defined volume form ω which has the following property: At
x ∈ M , let e1 , . . . , en be an oriented orthonormal basis for Tx M . Then ω(x)(e1 , . . . , en ) = 1. If
we change the choice of orthonormal basis by multiplying by A ∈ SO(n), then we have a change
of det(A), which is still 1. Therefore, ω is well-defined.
For surfaces (Σ, g), we have an area form dA.
Theorem 40.4 (Gauß-Bonnet). Let Σ be a compact submanifold of Euclidean space (R3 , g). Then,
for one of the orientations of Σ, Z
κdA = 2πχ(Σ).
Σ
Here κ is the scalar curvature, dA is the area form for g induced from (R3 , g), and χ(Σ) is the
Euler characteristic of Σ.
84 KO HONDA
The Euler characteristic of a compact manifold M of dimension n is:

Xn
χ(M ) = (−1)i dim HdR
i
(M ).
i=0
Note that a compact surface Σ (without boundary) of genus g has χ(Σ) = 2 − 2g.
Proof. Notice that κdA is simply κθ1R∧ θ2 above in the Example, and hence R the Euler class is
e(T M ) = [κdA]. In order to evaluate Σ κdA, we therefore need to compute Σ ω for the connec-
tion of our choice on T Σ compatible with g, by using Theorem 40.3.
In what follows we will frequently identify SO(2) with the unit circle S 1 = {eiθ |θ ∈ [0, 2π]} in
C via
cos θ − sin θ
↔ eiθ .
sin θ cos θ
We will do a sample computation in the case of the sphere S 2 . Let S 2 be the union of two
regions U = {|z| ≤ 1} and V = {|w| ≤ 1} identified via z = 1/w along their boundaries.
Here z, w are complex coordinates. (Note that U and V are not open sets, but it doesn’t really
matter....) If we trivialize T Σ on U and V using the natural trivialization from T C, then the gluing
map g : U ∩ V → SO(2) is given by θ 7→ e2iθ . If we set AV to be identically zero, then
0 2dθ
AU = g −1 dg + g −1 AV g = g −1 dg = along ∂U (after transforming via g). No
−2dθ 0
matter how we extend AU to the interior of U , we have the following by Stokes’ Theorem:
Z Z
ωU = 2dθ = 4π = 2πχ(S 2 ).
U ∂U
Now let Σ be a compact surface of genus g (without boundary). Then we can remove g annuli
S 1 × [0, 1] from Σ so that Σ becomes a disk Σ0 with 2g − 1 holes. We make A flat on the annuli,
and see what this induces on Σ0 . A computation similar to the one above gives the desired formula.
(Check this!!)

Revised Course Notes

Uploaded by

Copyright:

Available Formats

Revised Course Notes

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Revised Course Notes

Uploaded by

Copyright:

Available Formats

DIFFERENTIAL GEOMETRY COURSE NOTES

1. R EVIEW OF TOPOLOGY AND LINEAR ALGEBRA

Example 1: Rn = R × R × · · · × R (n times) = {(x1 , . . . , xn ) | xi ∈ R, i = 1, . . . , n}, called

Question: Is T0 = {Br (y) | y ∈ Rn , r ∈ (0, ∞)} a valid topology for Rn ?

Example 2A: S 1 = {(x, y) ∈ R2 | x2 + y 2 = 1}. A reasonable topology on S 1 is the topology

Definition 1.3. Let (X, T ) be a topological space, ∼ be an equivalence relation on X, X = X/ ∼

(d) (a + b)v = av + bv.

Definition 1.8. A linear map φ : V → W is an isomorphism if there exists a linear map ψ : W →

Shorthand: Assuming f is smooth, we write ∂ α f = ∂1α1 ∂2α2 . . . ∂kαk f where α = (α1 , . . . , αk ).

(1) f is smooth or of class C ∞ at x ∈ Rm if all partial derivatives of all orders exist at x.

4. E XAMPLES OF SMOOTH MANIFOLDS

(6) S n = {x21 + · · · + x2n+1 = 1} ⊂ Rn+1 . Generalize the discussion from (5).

Consider U0 = {x0 6= 0} with the coordinate chart φ0 : U0 → Rn given by

5. S MOOTH FUNCTIONS AND SMOOTH MAPS

is smooth for each coordinate chart (Uα , φα ) of A.

can be written as (f ◦ ψβ−1 ) ◦ (ψβ ◦ φ−1 −1 −1

Pullback: Let φ : X → Y be a continuous map between topological spaces. Then there is a

The proof is similar to that of Lemma 5.4.

6. T HE I NVERSE F UNCTION T HEOREM

ψα : Vα → Rn , Uα ⊂ M , Vβ ⊂ N , and φβ , ψβ are defined similarly. The invariance of rank is due

= d(ψβ ◦ ψα−1 ) ◦ d(ψα ◦ f ◦ φ−1 −1

7. S UBMERSIONS AND REGULAR VALUES

Prototype: f : Rm → Rn , (x1 , . . . , xm ) 7→ (x1 , . . . , xn ), where m ≥ n.

8. I MMERSIONS AND EMBEDDINGS

(x1 , . . . , xm , ym+1 , . . . , yn ) 7→ (f1 (x), . . . , fm (x), fm+1 (x) + ym+1 , . . . , fn (x) + yn ),

9. TANGENT SPACES , DAY I

10. TANGENT SPACES , DAY II

xi . By Taylor’s Theorem, any f ∈ C ∞ (p) can be written as a +

Lemma 10.3. The first two definitions of Tp M are equivalent.

10.3. Third definition. Define Fp ⊂ C ∞ (p) to be germs of functions which are 0 at p. Fp is an

11. T HE TANGENT BUNDLE

11.2. Examples of tangent bundles.

Example: S 2 ⊂ R3 , S 2 = {x21 + x22 + x23 = 1}. A multivariable calculus definition of T S 2

Remark: A holomorphic map f : C → C, when viewed as a map f : R2 → R2 , is a smooth map.

Remark: It is not a coincidence that a11 = a22 and a21 = −a12 .

12. C OTANGENT BUNDLES AND 1- FORMS

Examples of Lie subgroups:

14. V ECTOR BUNDLES

14.2. Transition functions, reinterpreted. Consider π : T M → M and local trivializations

(1) For double intersections U ∩ V , we have ΦU V (p) · ΦV U (p) = id.

We can use ρ and glue Uα × Rm together using:

15. M ORE ON VECTOR BUNDLES ; ORIENTABILITY

Example: Recall S 2 as a complex manifold, obtained by gluing together U = C and V = C via

16. I NTEGRATING 1- FORMS ; TENSOR PRODUCTS

Definition 16.2. Let V1 , . . . , Vk , U be vector spaces. A map φ : V1 × · · · × Vk → U is multilinear

17. T ENSOR AND EXTERIOR ALGEBRA

Universal property. ∧k V and i : V × · · · × V → ∧k V satisfy the following: Given an alternating

18. D IFFERENTIAL k- FORMS

1. If k > n, ∧k V = 0. This is clear since it is impossible to find 1 ≤ i1 < · · · < ik ≤ n.

Remark 18.2. For any v1 , . . . , vk ∈ V , v1 ∧ · · · ∧ vk 6= 0 in ∧k V if and only if v1 , . . . , vk are

Example. dim V = 2. Basis {v1 , v2 }. ∧V has basis {1, v1 , v2 , v1 ∧ v2 }. If A : V → V is linear

Denote by Ωk (M ) the sections of ∧k T ∗ M .

HW: Check that the above definition is independent of choice of coordinates.

19. D E R HAM COHOMOLOGY

Example: If M = R3 , then d0 = grad, d1 = curl, d2 = div. Then d2 = 0 is equivalent to

H k (R) = 0 for k > 1 since Ωk (R) = 0 for k > 1.

20. M AYER -V IETORIS SEQUENCES ; P OINCAR É LEMMA

20.4. Partitions of unity.

21. S OME HOMOLOGICAL ALGEBRA