0% found this document useful (0 votes)
67 views63 pages

18.952 Differential Forms

This document provides an introduction and notes for a course on the theory of differential forms. It begins with examples of integrals that involve differential forms in multivariable calculus. The notes then review relevant linear algebra concepts like vector spaces, subspaces, bases, linear maps, and the dual space. It introduces the idea of tensors and discusses how linear maps between vector spaces induce maps between their dual spaces. The notes aim to lay the groundwork for understanding differential forms by reviewing key linear algebra concepts.

Uploaded by

Angelo Oppio
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
67 views63 pages

18.952 Differential Forms

This document provides an introduction and notes for a course on the theory of differential forms. It begins with examples of integrals that involve differential forms in multivariable calculus. The notes then review relevant linear algebra concepts like vector spaces, subspaces, bases, linear maps, and the dual space. It introduces the idea of tensors and discusses how linear maps between vector spaces induce maps between their dual spaces. The notes aim to lay the groundwork for understanding differential forms by reviewing key linear algebra concepts.

Uploaded by

Angelo Oppio
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 63

18.

952: Theory of Differential Forms


Lecturer: Professor Victor Guillemin
Notes by: Andrew Lin

Spring 2022

Introduction
This class will follow the textbook “Differential Forms,” published by World Scientific and authored by Professor
Guillemin and Peter Haine. We can reach out at [email protected] or stop by room 2-270 with any questions –
regular office hours will be Mondays and Wednesdays from 1-2pm.
The prerequisites for the course are 18.100B or 18.901, and (not necessary but helpful) 18.101. And background
in linear algebra (18.700 or helpfully 18.702) will be very essential – many of the concepts in the theory of differential
forms are basically linear algebra in various aspects. For now, we won’t have any exams, and grading will be done using
biweekly problem sets. Hopefully, a grader will be recruited for the course who will also hold regular office hours.

1 January 31, 2022


We’ll begin with a simplistic description of what differential forms look like, looking at a few concrete examples that
occur in multivariable calculus (though they aren’t called differential forms in that context):

Example 1
R
Line integrals of the form γ f1 dx1 +f2 dx2 +f3 dx3
for functions f1 , f2 , f3 in the three variables x1 , x2 , x3 , where γ is a
R
curve in 3D space, are differential forms. Similarly, surface integrals of the form S f1 dx2 dx3 +f2 dx1 dx3 +f3 dx1 dx3
R
with respect to some compact surface S are differential forms, and so are volume integrals D f dx1 dx2 dx3 over
some domain D.

We should be familiar already with how we compute such integrals, but one question we can ask is what the
integrands in these expressions mean intrinsically. In a similar vein, we may remember expressions like grad, curl,
and ÷ from vector calculus – we may want to ask how to naturally generalize such operations to n-dimensional space,
without having to explicitly write down complicated expressions. And that’s going to lead us to the study of differential
forms, but we’ll start by thinking about multilinear algebra explicitly. (We’ll want to generalize these notions when
we extend our definitions later, so it’s good to have everything in one place.)

Definition 2
A vector space V over R is a set with two basic operations (v , w ) 7→ v + w (vector addition) and (x, v ) 7→ xv
(scalar multiplication) for any v , w ∈ V and x ∈ R, containing a zero vector, additive inverses, and also satisfying
commutativity and associativity of addition, distributivity, x1 (x2 v ) = (x1 x2 )v , and 1v = v .

1
Definition 3
A set of vectors v1 , · · · , vn span a vector space V if every v ∈ V can be written as v = c1 v1 + · · · + cn vn for
some c1 , · · · , cn ∈ R, and it is linearly independent if whenever c1 v1 + · · · + cn vn = 0, we have c1 , · · · , cn = 0.
If v1 , · · · , vn is both linearly independent and spanning, then we call the set a basis.

Fact 4
For any finite-dimensional vector space, there always exists a basis (in other words, the definitions above are
legitimate).

Remark 5. The first week or two of the class will revolve around a lot of linear algebra, and we’ll be exclusively
discussing finite-dimensional vector spaces (so we don’t need to worry about any nuances with infinite-dimensional
spaces).

Definition 6
A subset W ⊆ V is a subspace of V if it is closed under addition and scalar multiplication.

Definition 7
Let V and W be two vector spaces. A map A : V1 → W is a linear mapping if for all v1 , v2 ∈ V1 and c1 , c2 ∈ R,
we have A(c1 v1 + c2 v2 ) = c1 A(v1 ) + c2 A(v2 ). The kernel of A is the set ker A = {v ∈ V1 : Av = 0}, and the
image of A is the set Im A = {w ∈ W : w = Av for some v ∈ V1 }.

We can verify that ker A and Im A are subspaces of V1 and W , respectively, and we also have the elementary result
dim ker A + dim Im A = dim V1 . And we also have a more concrete way of writing down what A looks like when V and
W are finite-dimensional: if we let v1 , · · · , vn be a basis of V and let w1 , · · · , wm be a basis of W , then we can write
P
Avi = aji wj for some constants aij ∈ R, and then the matrix formed by these aij s gives us a matrix representation
of the linear map A.
P
Remark 8. The convention is often to use Avi = aij wj instead to define the matrix of A, but we’ll use this definition
for consistency with the book.

Fact 9
The map between linear maps and matrices is a bijective correspondence, so we can use matricial identities to
learn a lot of facts about linear maps. In particular, any map defined by a matrix A = [aij ] is a linear map.

Definition 10
Let V be a vector space. A bilinear form is a map B : V × V → R which is bilinear in each variable, meaning that
B(c1 v1 + c2 v2 , v ) = c1 B(v1 , v ) + c2 B(v2 , v ) and similar in the other variable.

We can refer to Chapter 1 of the textbook for more linear algebra details if we’d like, but we’ll finish this lecture
with a few other useful definitions:

2
Definition 11
Let W be a subspace of a vector space V . A subset of the form v + W = {v + w : w ∈ W } is called a W -coset
of V .

In particular, note that two W -cosets v1 + W and v2 + W are either the same set or completely disjoint, so that
V is a disjoint union of its W -cosets. Thus, we can make the following definition:

Definition 12
Let W be a subspace of a vector V . Then V /W is the set of W -cosets of V .

It’s left as an exercise for us to check that V /W is indeed a vector space (meaning that it satisfies the axioms of
Definition 2), under the addition operation (v1 + W ) + (v2 + W ) = (v1 + v2 ) + W .

2 February 2, 2022
Last time, we did a review of linear algebra – in particular, for a subspace W of a vector space V , we defined the
W-coset v + W = {v + w : w ∈ W } for any v ∈ V , and we mentioned that v1 + W and v2 + W are always disjoint or
identical, so that V can always be written as a disjoint union of its W -cosets.
We also mentioned that this set of cosets itself has a vector space structure: we can define a quotient space V /W
of the set of W -cosets, with addition given by (v1 + W ) + (v2 + W ) = (v1 + v2 ) + W and scalar multiplication given
by c(v + W ) = cv + W . (Then the zero vector is 0 + W = W itself, and we can check that all of the axioms of a
vector space are satisfied.) We then have dim V /W = dim V − dim W whenever V is finite-dimensional.
Finally, we also discussed the concept of functoriality: suppose V and U are two finite-dimensional vector spaces,
and A : V → U is a linear map. If W = ker A, then we get an injective linear map V /W → U, given by v + W 7→ Av .
Also, for any subspace W of V , the projection map π : V → V /W maps any v ∈ V to the corresponding coset v + W .
We can now turn to the topic of today’s lecture, tensors. Before we get to those definitions, we’ll need to discuss
a few more important linear algebra notions:

Definition 13
Let V be an n-dimensional vector space. The dual space of V , denoted V ∗ , is the set of linear maps {ℓ : V →
R, ℓ linear}.

If we let e1 , · · · , en be a basis of V , then we can consider the corresponding dual basis e1∗ , · · · , en∗ , where ei∗ is the
linear map that sends ei to 1 and all of the other ej s to 0: in other words,

v = a1 e1 + · · · + an en =⇒ ei∗ v = ai .

Proposition 14
These maps ei∗ form a basis of V ∗ , so choosing a basis for V automatically gives a basis for V ∗ .

Proof. To show that the ei∗ s span V ∗ , suppose we have some linear map ℓ ∈ V ∗ in the dual space. Then we can define
au eu∗ , since
P
au = ℓ(eu ), and we can write our map as the linear combination ℓ =
X  X
ℓ(ei ) = au eu∗ (ei ) = au eu∗ ei = ai .

3
au eu∗ is the zero map. Then 0 = 0ei = ( au eu∗ ) ei = ai , so all ai must
P P
To show independence, suppose some map
be zero.

With this, if we have a linear map A : V → W , we can always define a linear map A∗ : W ∗ → V ∗ in the following
way: for any ℓ ∈ W ∗ (meaning that ℓ is a linear map W → R), we define A∗ ℓ to be the map V → R which applies A
and then ℓ to any vector in v .
To understand what this looks like more concretely, we can think about this in terms of coordinates: let e1 , · · · , en
be a basis of V , and let f1 , · · · , fm be a basis of W . Then we can characterize A with the numbers aij given by
X
Aei = aji fj .

If we define e1∗ , · · · , en∗ and f1∗ , fm∗ to be the dual bases of V ∗ and W ∗ , we now wish to check that
X
A∗ fi ∗ = aij ej∗ .

This is left as an exercise to us, but essentially we should use the fact that A∗ fi ∗ (ej ) = fi ∗ Aej . And what this tells us
is that if [aij ] is the matrix for the map A, then its transpose [aji ] is the matrix for the map A∗ .

Problem 15
Show that the double dual of V satisfies (V ∗ )∗ = V . (As a hint, for any v ∈ V , we can define the map
uv (ℓ) = ℓ(v ), and that will give us a way to map the two spaces to each other.

Problem 16
Let W be a subspace of the vector space V , and let W ⊥ be the set of ℓ ∈ V ∗ such that ℓ(w ) = 0 for all w ∈ W .
Show that W ⊥ is a subspace of V ∗ and that (V /W )∗ = W ⊥ .

We can now think about functoriality in this context: if A : V → W is a linear map, and A∗ : W ∗ → V ∗ is its dual
map, then we claim that
ker A∗ = (Im A)⊥ , Im A∗ = (ker A)⊥ .

Definition 17
Let V k denote the k-fold product V × V × · · · × V . A map T : V k → R is linear in its i th slot if for any fixed
vectors v1 , · · · , vi−1 , vi+1 , · · · , vn , the map v → T (v1 , · · · , vi−1 , v , vi+1 , · · · , vn ) is a linear map. T is a k-tensor if
it is linear in all k slots. Let Lk (V ) denote the set of k-tensors.

We can define a vector space structure on Lk (V ) as follows: if we have two k-tensors T1 : V k → R and T2 : V k → R,
then we can check that c1 T1 + c2 T2 also gives us a valid k-tensor (by checking linearity in each slot).

Example 18
A 1-tensor is a map from V → R, so L1 (V ) = V ∗ . Meanwhile, a 2-tensor is a map V × V → R, so L2 (V ) is the
set of bilinear forms on V . We’ll use the convention that L0 (V ) = R for convenience.

Definition 19
A multi-index of length k is a sequence of integers I = (i1 , · · · , ik ), where 1 ≤ i1 , · · · , ik ≤ n.

4
In particular, for a basis e1 , · · · , en of an n-dimensional vector space V , a k-tensor T , and a multi-index I =
(i1 , · · · , ik ), we can define
TI = T (ei1 , · · · , eik ).

As an exercise, we can check that these numbers (across all multi-indices I) determine the k-tensor T .

Problem 20
Let I denote the set of multi-indices, and let Lk (I) be the set of maps from I to R. Show that dim Lk (I) =
|I| = nk , and show that the map from to L̃(I), mapping T 7→ TI , is bijective. Thus, conclude that the dimension
of the k-tensors is dim Lk = nk .

We’ll finish by defining the tensor product operation:

Definition 21
If T1 ∈ Lk (V ) and T2 ∈ Lℓ (V ) are two tensors, we can define the tensor product T1 ⊗ T2 , a (k + ℓ)-tensor, via

(T1 ⊗ T2 )(v1 , · · · , vk+ℓ ) = T1 (v1 , . . . , vk )T2 (vk+1 , · · · , vk+ℓ ).

As an exercise, we should check that this is indeed a valid tensor – we’ll be using it throughout this class.

3 February 4, 2022
Last lecture, we introduced the concept of a k-tensor, which is a map T : V k → R which is linear in each of
the k copies of V . (In other words, if we fix vectors v1 , · · · , vi−1 , vi+1 , · · · , vk , then for any v ∈ V , the map v 7→
T (v1 , · · · , vi−1 , v , vi+1 , · · · , vk ) is linear in v .) Recall that Lk (V ) denote the set of k-tensors; this is in fact a vector
space because any linear combination of k-tensors is itself a k-tensor.
In our definitions last time, we also introduced multi-index notation: a multi-index of length k is some sequence
I = (u1 , · · · , uk ) of integers 1 ≤ ui ≤ n, and we can define a k-tensor using a multi-index description by defining the
numbers TI = T (eu1 , eu2 , · · · , euk ) for each I. Those numbers then give us, by linearity, a concrete description of the
k-tensor: as we’ll describe later, if vi = kj=1 aij ej , then we have
P

X
T (v1 , · · · , vk ) = aI TI .
I

We’ll spend today discussing some more algebraic properties of these k-tensors. Recall that given two tensors T1 ∈
Lk1 (V ) and T2 ∈ Lk2 (V ), we can define the tensor product T1 ⊗ T2 ∈ Lk1 +k2 (V ) via

T1 ⊗ T2 (v1 , v2 , · · · , vk1 +k2 ) = T1 (v1 , · · · , vk1 )T2 (vk1 +1 , · · · , vk1 +k2 ).

It’s left as an exercise to use to check this is a valid tensor, and we can also verify some other properties of this product
operation:

1. (Associativity) For any three tensors T1 , T2 , T3 , we have (T1 ⊗ T2 ) ⊗ T3 = T1 ⊗ (T2 ⊗ T3 ), so we do not need
to worry about parentheses when doing tensor product computations.

2. (Left and right distributivity) For any two tensors T1 , T2 ∈ Lk1 of the same order and any T3 ∈ Lk2 , we have
(T1 + T2 ) ⊗ T3 = T1 ⊗ T3 + T2 ⊗ T3 , as well as T3 ⊗ (T1 + T2 ) = T3 ⊗ T1 + T3 ⊗ T2 .

5
Note that we do not have commutativity (that is, T1 ⊗ T2 = T2 ⊗ T1 ) in general. But still, being able to take these
kinds of products allows us to consider an important special class of tensors:

Definition 22
For all 1 ≤ i ≤ k, let ℓi ∈ V ∗ be linear maps. Then ℓ1 ⊗ · · · ⊗ ℓk is a decomposable k-tensor.

Theorem 23
Let e1 , · · · , en be a basis of a (finite-dimensional) vector space V , and let e1∗ , · · · , en∗ be the corresponding dual basis.
Let I = (u1 , · · · , uk ) be an arbitrary multi-index of indices. Then the (decomposable) k-tensors eI∗ = eu∗1 ⊗· · ·⊗eu∗k
form a basis of Lk (V ).

P
Proof sketch. This is essentially a reformulation of the boxed equation T (v1 , · · · , vk ) = I aI TI from above. For
any k-tensor V and any v1 , · · · , vk satisfying vi = kj=1 aij ej , if we define aI = a1u1 · · · akuk , then plugging in the vi
P

expressions and using linearity of the tensor gives us


X
T (v1 , · · · , vk ) = aI TI ,

so we’ve written our tensor T as a linear combination of these TI s. From here, it just remains to show that the TI s
are linearly independent, which will be left for us to show.

Being able to uniquely express T as linear combinations of tensors of the form eu∗1 ⊗ · · · ⊗ eu∗k will be useful for
computations in the future!

4 February 7, 2022
We’ll discuss the theory of permutations today as they relate to some of the linear algebra objects that we’ve been
studying in this class:

Definition 24
A permutation of order k is a bijective map σ : {1, 2, · · · , k} → {1, 2, · · · , k}. The set of all permutations of
order k is denoted Sk .

This set of permutations has a natural group structure, because bijective maps can be composed and inverted: if
σ, τ are elements of Sk , then στ is the permutation that sends i to σ(τ (i )). (We can then check the group axioms
ourselves from here as an exercise.)

Definition 25
Let 1 ≤ i , j ≤ k. The transposition permutation τij is the map sending i to j, j to i , and fixing all other integers.

It is a fact (that we might learn in an abstract algebra class) that every permutation is a finite product of trans-
positions – this can be proved by induction.

Definition 26
The elementary transpositions are the transposition permutations of the form τi,i+1 .

6
It turns out that we can also write every transposition as a finite product of elementary permutations – this is
again proved by induction. Essentially, notice that τi,i+2 = τi+1,i+2 τi,i+1 τi+1,i+2 , and then τi,i+3 = τi+2,i+3 τi,i+2 τi+2,i+3
(from which we can substitute in the expression for τi,i+2 ), and so on. Thus, combining these two facts gives us the
important result:

Proposition 27
Every transposition is a finite product of elementary transpositions.

Definition 28
Let x1 , · · · , xn be coordinate functions on Rn , and let σ ∈ Sn be a permutation. The sign of the permutation σ,
denoted (−1)σ , is given by Q
i<j xσ(i) − xσ(j)
σ
(−1) = Q .
i<j (xi − xj )

Notice that each set of indices {i , j} will show up in both the numerator and denominator in some order, so the
right-hand side will just be one of ±1.

Lemma 29
If τ is any transposition permutation, then (−1)τ = −1.

(This is easily verified by plugging the form of τ back into the formula.)

Lemma 30
If σ, τ ∈ Sn are two permutations, then (−1)στ = (−1)σ (−1)τ .

Proof. Plugging into the formula, we have


Q
στ i<j xστ (i) − xστ (j)
(−1) = Q .
i<j (xi − xj )

We can rewrite this product as Q Q


i<j xστ (i) − xστ (j) i<j (xτ (i) − xτ (j) )
= Q · Q ,
i<j (xτ (i) − xτ (j) ) i<j (xi − xj )

so that the second factor is (−1)τ by definition. But now replacing i and j with τ (i ) and τ (j) in the numerator
and denominator is essentially a relabeling of Q
the indices (alternatively, a reordering of the terms in the product). So
xσ(i # ) −xσ(j # )
replacing i # = τ (i ) will make the first term i<j
Q , and these products each sum over each unordered pair
i<j (xi # −xj # )
# #
i ,j exactly once, so this fraction is exactly the definition of (−1)σ . This finishes the proof.

Corollary 31
If σ ∈ Sn is a product of k transpositions, then (−1)σ = (−1)k .

Definition 32
Let T ∈ Lk (V ) be a k-tensor, and let σ ∈ Sk . We define T σ to be the k-tensor such that

T σ (v1 , · · · , vk ) = T (vσ−1 (1) , · · · , vσ−1 (k) ).

7
(The reasons for using the inverse permutation will become clear soon.)

Example 33
Let ℓ1 , · · · , ℓk ∈ V ∗ be linear maps, and let T be the decomposable k-tensor ℓ1 ⊗ · · · ⊗ ℓk . Then T σ =
ℓσ(1) ℓσ(2) · · · ℓσ(k) .

Proposition 34
Let T be a k-tensor, and let σ, τ ∈ Sk . Then T στ = (T σ )τ (meaning that we apply the σ permutation first to T ,
and then τ to that result).

Proof. By linearity, it suffices to show that this result holds for decomposable k-tensors, so we can apply Example 33
and verify that the result holds there.

We’ll finish this lecture by introducing two important objects that we’ll be using for the rest of this course:

Definition 35
A k-tensor T ∈ Lk (V ) is a symmetric k-tensor if for every permutation σ ∈ Sk , T σ = T .

But for the theory of differential forms, the even more important object is the following:

Definition 36
A k-tensor T ∈ Lk (V ) is an alternating k-tensor if for every permutation σ ∈ Sk , T σ = (−1)σ T .

These alternating k-tensors will turn out to be the basic building blocks of differential forms, and we’ll be studying
them a lot in the coming lectures.

5 February 9, 2022
Last lecture, we introduced some properties of permutations, which are bijective maps σ : {1, · · · , k} → {1, · · · , k}.
Treating composition of these maps as multiplication, the set of permutations Sk has a group structure (where the
inverse σ −1 of a permutation σ is the inverse map).
A very relevant property of permutations is their sign, which is either 1 or −1 and is given by the definition
Q x −xσ(j)
(−1)σ = i<j σ(i) k σ
xi −xj . In particular, if we take a k-tensor T ∈ L (V ), we defined the k-tensor T given by

T σ (v1 , v2 , · · · , vk ) = T (vσ−1 (1) , vσ−1 (2) , · · · , vσ−1 (k) ),

and we said that a k-tensor is alternating if T σ = (−1)σ T (and symmetric if T σ = T ) for all permutations σ ∈ S k .
These definitions become more clear if we look at the decomposable tensors of the form T = ℓ1 ⊗ ℓ2 ⊗ · · · ⊗ ℓk (where
ℓi s are linear maps) – for such tensors, we have T σ = ℓσ(1) ⊗ ℓσ(2) ⊗ · · · ⊗ ℓσ(k) . Looking at such decomposable
k-tensors, which form a basis of all k-tensors, allows us to prove that T (στ ) = (T σ )τ .
We’ll let S k (V ) denote the space of symmetric k-tensors, and we’ll let Ak (V ) denote the space of alternating k-
tensors. The latter set will be essential for the theory of differential forms, and here we’ll describe a way of constructing
such alternating tensors:

8
Definition 37
Let T ∈ Lk (V ) be an arbitrary tensor. The alternation operation is defined via
X
Alt(T ) = (−1)τ T τ .
τ ∈Sk

Proposition 38
The facts below follow from definitions and properties of the sign of a permutation:
1. For any T ∈ Lk (V ), we have Alt(T ) ∈ Ak (V ).

2. For any σ ∈ Sk , we have Alt(T σ ) = (−1)σ Alt(T ).

3. If T ∈ Alt(T ), then Alt(T ) = k!T .

We will now construct a basis for Ak (V ). First, let e1 , · · · , en be a basis of V , and let e1∗ , · · · , en∗ be the corresponding
dual basis of V ∗ . For any multi-index I = (i1 , . . . , ik ), we define eI∗ = ei∗1 ⊗ · · · ⊗ ei∗k , and we may recall from previous
lectures that these eI∗ form a basis of Lk (V ). Given this fact, we now define

φI = Alt(eI∗ )

for all multi-indices I; because the eI∗ form a basis and Alt is surjective, we know that the φI s span Ak (V ). But to
avoid having linear dependence, we need to restrict the set of Is that we use:

Definition 39
A multi-index I is strictly increasing if i1 < i2 < · · · < ir .

Theorem 40
The set {φI : I strictly increasing} is a basis for the set of alternating k-tensors Ak (V ).

Beginning of the proof. First of all, call a multi-index I repeating if ir = is for some r ̸= s. Notice that φI = 0 for
any repeating multi-index I – we can see this by breaking Sk up into the two cosets formed by the subgroup {1, σr s },
or equivalently saying that TI = TIσr s , so that

Alt(TI ) = Alt(TIσr s ) = − Alt(TI ) =⇒ Alt(TI ) = 0.

Thus we do not want to include φI for repeating I. Now if we consider a non-repeating multi-index I, let (ir1 , ir2 , · · · , irk )
be the reordering of the indices of I so that r1 < r2 < · · · < rk , and let σ ∈ Sk be the permutation that takes iℓ
to irℓ , then I σ is strictly increasing, and Alt(eI∗σ ) = (−1)σ Alt(eI∗ ) (because (eI∗ )σ = eI∗σ ). Thus any φI is ±φI σ for
some increasing multi-index I σ , so the set of φI formed by just the increasing multi-indices I also spans the whole set
Ak (V ∗ ). The rest of the proof will be shown next time!

6 February 11, 2022


Our first homework assignment will be on Canvas today, and it will be due two weeks from today.

9
First, let’s do some review. We’ve been studying the “permuted” versions of tensors in the last few lectures: for
any tensor T ∈ Lk (V ) and any permutation σ ∈ Sk , we define T σ via T σ (v1 , · · · , vk ) = T (vσ−1 (1) , · · · , vσ−1 (k) ). (This
definition is motivated by the fact that a decomposable tensor T = ℓ1 ⊗ · · · ⊗ ℓk becomes T σ = ℓσ(1) ⊗ · · · ⊗ ℓσ(k) .) The
fundamental objects important for the theory of differential forms are the alternating tensors satisfying T = (−1)σ T σ
(or equivalently T σ = (−1)σ T ) – we can construct the tensor
X
Alt(T σ ) = (−1)σ T σ .
σ∈Sk

It turns out that Alt is a surjective map from Lk (V ) to Ak (V ), and Alt(T ) = k!T if T is an alternating tensor. It then
makes sense to ask about the kernel of the map Alt:

Definition 41
A decomposable k-tensor T = ℓ1 ⊗ · · · ⊗ ℓk is redundant if ℓr = ℓr +1 . The linear span of all redundant k-tensors
is denoted I k (elements of I k will be called redundant as well).

Proposition 42
If T ∈ I k , then Alt(T ) = 0.

Proof. Suppose T = ℓ1 ⊗ · · · ⊗ ℓk is decomposable with ℓr = ℓr +1 . Let σ be the transposition permutation swapping


r and r + 1, so that
Alt(T ) = Alt(T σ ) = (−1)σ Alt(T ) = − Alt(T ),

so that Alt(T ) = 0. Since these redundant tensors span I k , Alt(T ) = 0 for any element of I k .

We aim to show that these redundant k-tensors span the kernel of Alt, which is the converse result. This is
essentially showing the other part of the theorem from last lecture, but first, we mention some important preliminary
results:

Lemma 43
If T1 and T2 are elements of I r (V ) and I s (V ), respectively, then T1 ⊗ T2 and T2 ⊗ T1 are element sof I r +s (V ).

Proof. It suffices to consider the case where T1 and T2 are both decomposable redundant k-tensors, so that T1 has
some repeat ℓi = ℓi+1 . Then whether we are tensoring T1 ⊗ T2 or T2 ⊗ T1 , we will have two adjacent slots where the
linear maps are the same (either ℓi = ℓi+1 or ℓi+s = ℓi+s+1 , respectively).

Lemma 44
If T ∈ Lk (V ) and σ ∈ Sk , then T = (−1)σ T σ + T ′ for some T ′ ∈ I k .

Proof. By linearity, it suffices to show the result when T is decomposable. Let T = ℓ1 ⊗ · · · ⊗ ℓk . We know that any
permutation σ can be written as a product of elementary transpositions of the form τi = σi,i+1 ; if we consider the
case where σ is a single transposition τi , then

T + T σ = ℓ1 ⊗ · · · ⊗ (ℓi ⊗ ℓi+1 + ℓi+1 ⊗ ℓi ) ⊗ · · · ⊗ ℓk .

10
But this right-hand side is in the linear span of redundant decomposable tensors, because

(ℓi ⊗ ℓi+1 + ℓi+1 ⊗ ℓi ) = (ℓi + ℓi+1 ) ⊗ (ℓi + ℓi+1 ) − ℓi ⊗ ℓi − ℓi+1 ⊗ ℓi+1 .

Thus T + T σ is in I k , and because (−1)σ = −1 in this case that’s exactly what we want to show.
From here, we use induction: suppose σ is a product of the form τ1 τ2 . . . τm−1 , and we know that T − (−1)σ T σ
is in I k . Then by the inductive hypothesis we have (abusing notation a little here)

T στ = (T σ )τ = (−1)τ T σ + I k ,

and again applying the inductive hypothesis we get

= (−1)τ (−1)σ T + I k = (−1)στ T + I k ,

showing the desired result.

Corollary 45
For all T ∈ Lk (V ), we have
1
T = Alt(T ) + I
k!
for some I ∈ I k (V ).

Proof. By Lemma 44, we have


X X
Alt(T ) = (−1)σ T σ = (−1)σ (−1)σ T + Sσ
σ σ

for some Sσ ∈ I k for each σ ∈ Sk . Since (−1)σ (−1)σ = 1, this simplifies to


X
Alt(T ) = k!T + Sσ .
σ

Dividing through by k! and rearranging gives the result – because each Sσ is in I ∗ , so is their average.

That finally gives us the result about the kernel of the map Alt:

Corollary 46
For any k-tensor T , if Alt(T ) = 0, then T ∈ I k . Also, any k-tensor T can be uniquely written as T = T ′ + S ′ ,
where T ′ is alternating and S ′ is redundant.

Proof. The first part follows by setting Alt(T ) = 0 in Corollary 45. For the second part, suppose we have T =
T ′ + S ′ = T1′ + S1′ , where T ′ , T1′ are alternating and S ′ , S1′ are redundant. Taking Alt of both sides, we find that
Alt(T ) = k!T ′ = k!T1′ , so T ′ = T1′ and thus S ′ = S1′ .

We thus can think of alternating k-tensors as a quotient space, and that will be foundational for our future study.

7 February 14, 2022


Last time, we introduced the set of redundant k-tensors I k (V ), which are the linear span of the decomposable k-
tensors ℓ1 ⊗ · · · ⊗ ℓk with ℓi = ℓi+1 for some i . We then proved that if T is redundant, then Alt(T ) = 0, and in fact

11
I k (V ) is the kernel of the map Alt : Lk (V ) → Ak (V ).
This allows us to make the definition of the most important object of the class:

Definition 47
The space of exterior k-forms on a vector space V is the quotient space Λk (V ∗ ) = Lk (V )/I k (V ).

In particular, here we’re making use of the exact sequence


Alt
0 → I k (V ) → Lk (V ) → Ak (V ) → 0

(where exactness follows from the result we showed last lecture). And furthermore, we now get a natural bijection

Λk (V ∗ ) → Ak (V ),

where the idea is that the two descriptions of the same object (either as a quotient space, or as a subspace) can be
useful in conjunction. (Since we’re looking at the space of tensors over V ∗ , we can imagine these spaces as being
spanned by vi1 ⊗ · · · ⊗ vik , where the vij s are now basis elements of the original space V .)

Definition 48
Let π : Lk (V ) → Lk (V )/I k (V ) = Λk (V ∗ ) be the natural projection map. The wedge product operation is
defined as follows: for ω1 ∈ Λk1 (V ∗ ) and ω2 ∈ Λk2 (V ∗ ), pick some Ti ∈ Lki (V ) so that π(Ti ) = ωi . Then define
ω1 ∧ ω2 = π(T1 ⊗ T2 ).

This definition basically tells us to go back into the full space of k-tensors and do a tensor product there, so that
the wedge product is a “factored version” of the tensor product. We can check that this is well-defined – in particular,
if we have T1 or T2 redundant, then T1 ⊗ T2 is redundant, so that after quotienting out we’ll still have 0. Thus our
choice of T1 (which is up to a redundant tensor defined) will not change the end result π(T1 ⊗ T2 ).

Definition 49
Let V be an n-dimensional vector space, and let T ∈ Lk (V ). The interior product operation is defined as follows:
for v ∈ V , ιv (T ) is the (k − 1)-tensor given by
k
X
ιv (T )(v1 , · · · , vk−1 ) = (−1)r −1 T (v1 , · · · , vr −1 , v , vr +1 , · · · , vk−1 ).
r =1

We’ll talk more about this next time!

8 February 16, 2022


Last time, we wrote down the short exact sequence
Alt
0 → I k (V ) → Lk (V ) → Ak (V ) → 0,

which is basically a cleaner way of explaining that I ∗ is the kernel of the Alt map Lk (V ) → Ak (V ). This enabled us
to define the space of exterior k-forms Λk (V ∗ ) = Lk (V )/I k (V ), in such a way that we have a bijective map between
Λk (V ∗ ) and Ak (V ). From there, the projection operation Lk (V ) → Lk (V )/I k (V ) = Λk (V ∗ ) (which we can think

12
about as basically applying Alt) allows us to define the wedge product: if ω1 ∈ Λk1 (V ∗ ) and ω2 ∈ Λk2 (V ∗ ), then
we can pick T1 ∈ Lk1 (V ) and T2 ∈ Lk2 (V ) such that π(T1 ) = ω1 and π(T2 ) = ω2 (this is always possible because the
projection map is onto). Then ω1 ∧ ω2 = π(T1 ⊗ T2 ), and we mentioned last time that this indeed well-defined.
We can now turn to considerations of functoriality:

Definition 50
If we have a linear map A : V → W between vector spaces, then we also get a linear map A∗ : Lk (W ) → Lk (V ),
given by the pullback operation: for any k-tensor T ∈ Lk (W ), we have

A∗ T (v1 , · · · , vk ) = T (Av1 , · · · , Avk ).

In particular, we can check that if T ∈ I k (W ), then A∗ T ∈ I k (V ) (start with a decomposable k-tensor, noticing
that if ℓr = ℓr +1 , then A∗ ℓr = A∗ ℓr +1 ), and we can also verify the relation

A∗ (T1 ⊗ T2 ) = A∗ (T1 ) ⊗ A∗ (T2 ).

Thus, we can also define an induced map on the quotient spaces

A∗ : Lk (W )/I k (W ) → Lk (V )/I k (V ),

and therefore this pullback operation takes a linear map A : V → W and gives us a map

A∗ : Λk (W ∗ ) → Ak (V ).

(The reason we write Ak (V ) instead of Λk (V ∗ ) here is that it’s more in line with the naturality of quotienting by
the kernel and ending up with the image, but it’s essentially the same thing.) In particular, bringing our definitions
together, we find that for ω1 ∈ Λk1 (V ∗ ) and ω2 ∈ Λk2 (V ∗ ), we have

A∗ (ω1 ∧ ω2 ) = Aω1 ∧ Aω2 ,

and thus in the special case where we’re wedging together linear maps, we get

A∗ (ℓ1 ∧ · · · ∧ ℓk ) = A∗ ℓ1 ∧ · · · ∧ A∗ ℓk

(we’ll be using this a lot throughout the rest of the course!). With this in mind, we can apply our discussion to the
notion of a determinant from linear algebra:

Proposition 51
Let V, W be two n-dimensional vector spaces, and let e1 , · · · , en and f1 , · · · , fn be bases of V and W , respectively
with corresponding dual bases e1∗ , · · · , en∗ and f1∗ , · · · , fn∗ . Let A : V → W be a linear map with corresponding
matrix [aij ] with respect to these bases, so that Aej = aij fi and A∗ fi ∗ = aij ej∗ . Then
P P

X  X  X
A∗ (f1∗ ∧ · · · ∧ fn∗ ) = A∗ f1∗ ∧ · · · ∧ A∗ fn∗ = a1j1 ej∗1 ∧ · · · ∧ anjn ej∗n = a1j1 a2j2 · · · anjn eJ∗ ,

where eJ∗ = ej∗1 ej∗2 · · · ej∗n .

In particular, for any repeating multi-index J, eJ∗ = 0, and for any non-repeating multi-index, there is a permutation
σ ∈ Sn such that σ(i ) = ji for all i , meaning that eJ∗ = (−1)σ e1∗ ∧ · · · ∧ en∗ . And thus, our familiar above reduces to

13
something more familiar, since we have a summation of the form
X
(−1)σ a1j1 · · · anjn e1∗ ∧ e2∗ ∧ · · · ∧ en∗ :

Corollary 52
For any linear map A : V → W , we have

A∗ (f1∗ ∧ · · · ∧ fn∗ ) = (det A)e1∗ ∧ · · · ∧ en∗ .

This functoriality allows us to prove some nice results from linear algebra without needing too much computation:
for example, we can see quickly that det(AB) = det(A) det(B), using that

det(AB)ω = (AB)∗ ω = B ∗ (A∗ ω) = B ∗ (det A · ω) = det AB ∗ (ω) = det(A) det(B)ω.

We can also see that the identity map A : V → V has det(A) = 1, and that whenever A : V → V is not onto,
det(A) = 0

9 February 18, 2022

Fact 53
We’re assigned to read Section 1.9 and Section 2.1 of the textbook on our own over the weekend, which essentially
covers the concepts of orientations, vector fields, and 1-forms. But we’ll mention a few of the facts here:

Definition 54
Let ℓ be a line through the origin in R2 . Then ℓ − {0} has two connected components, and an orientation of ℓ is
a choice of one of these two components.

(This is equivalent to essentially choosing a direction for the line.) We also have a natural generalization of this
which will connect back to the material in the class:

Definition 55
Let L be a one-dimensional vector space. Then L−{0} has two components; specifically, if we fix some v ∈ L−{0},
we have the component L+ = {λv : λ > 0} and L− = {λv : λ < 0}. An orientation of L is a choice of either
L+ or L− .
Even more generally, if V is an n-dimensional vector space, an orientation of V is an orientation of the space
n
Λ (V ). If (e1 , · · · , en ) form an ordered basis of V , then the basis is said to be positively oriented if e1 ∧ · · · ∧ en
is in the positive part of Λn (V ).

n n
 
(Note that the dimension of the space Λk (V ) is k , so Λn (V ) indeed has dimension n = 1.)

14
Proposition 56
Let V be an n-dimensional vector space, and let W ⊆ V be a k-dimensional subspace. If we are given orientations
on V and W , then there is a natural orientation of the quotient space V /W given as follows: choose an oriented
basis v1 , · · · , vn of V such that the first k of these vectors form an oriented basis of W . Then we can orient V /W
by projecting the remaining n − k vectors onto V /W .

With that, we’ll start making a few remarks that will start us on the next section of the class next week (Chapter
2 of the book). The idea is that Section 2.1 looks at simple but pivotal objects in the study of differential forms.

Definition 57
Let p ∈ Rn be a point. The tangent space at p, denoted Tp (Rn ), is the set

Tp (Rn ) = {(p, v ) : v ∈ Rn }.

The point p can be called a base point.

We can form the obvious vector space structure on Tp (Rn ), keeping the same base point, via

(p, v1 ) + (p, v2 ) = (p, v1 + v2 ), λ(p, v ) = (p, λv ).

The idea is to think of v as an arrow originating from p and pointing in the direction of v .

Definition 58
Let U be an open subset of Rn . A vector field v on U is a function which assigns to each point p ∈ U a
corresponding v (p) ∈ Tp (U) = Tp (Rn ). A one-form on U is a function which assigns to each point p ∈ U an
element ω(p) ∈ Tp∗ (Rn ).

Starting next week, we’ll talk more about these definitions, generalize to k-forms, and get into the main topic of
this course!

10 February 22, 2022


Today, we’ll fully begin discussing the concept of differential forms. Last lecture, we introduced the concept of a
tangent space (the set of pairs (p, v ) for some fixed p ∈ Rn ) – we convert this to a vector space by using the usual
vector space structure on Rn through v , keeping the base point p fixed:

(p, v1 ) + (p, v2 ) = (p, v1 + v2 ), λ(p, v ) = (p, λv ).

We can now think about functoriality in the following way: suppose we have open sets U, V ∈ Rn , Rm and we have
ah C ∞ (smooth)
i map φ : U → V . Then we can define the derivative map Dφ(r ) : Rn → Rm encoded by the matrix
∂φi
∂xj (r ) . This leads us to the base pointed version of the derivative definition:

15
Definition 59
Let φ : U → V be a C ∞ map, and let q = φ(p). The map (dφ)p between tangent spaces Tp U → Tq V is given by

(dφ)p (p, v ) = (q, Dφ(p)(v )).

In other words, we map the base point to the new base point, and we apply the derivative map to the vector v .
One result from calculus is the familiar chain rule: if W is additionally an open set in Rℓ and ψ : V → W is a C ∞ map,
then we know that
(dψ ◦ φ)p = dψq ◦ dφp .

Definition 60
The cotangent space to U at p is the vector space dual of the corresponding tangent space:

Tp∗ U = (Tp U)∗ .

In particular, if f ∈ C ∞ (U) is a smooth real-valued function, p ∈ U is some point, and u = f (p). Then we can
think of f as a map (U, p) → (R, u). Taking its derivative, we then know that dfp is a map Tp U → Tu R = ∼ R, so it’s
fundamentally an element of the cotangent space: dfp ∈ Tp∗ U .

Problem 61
Let x1 , · · · , xn be coordinate functions on U ⊆ Rn . Then (dx1 )p , · · · , (dxn )p form a basis of the cotangent space
Tp∗ U.

This basis can be described in an alternative way as well: if we let ei be the standard basis vector with a 1 in the
i th spot and 0s in the others, and we use the notation
 

= (p, ei ) ∈ Tp U,
∂xi p

then these elements form a basis of Tp , and then (dxi )p is the corresponding dual basis on Tp∗ U.
We can now return to the definitions of vector fields and one-forms that we started last lecture: recall that a
vector field is a map v : U → Tp U, and we can write that as

v (p) = (p, v(p))

for some map v : U → Rn .

Example 62  
∂ ∂
Let ei again be the standard basis vectors. Then the vector field p 7→ (p, ei ) is denoted ∂xi . Since ∂xi = (p, ei )
p
form a basis for Tp U, we can write every vector field as
 
X ∂ X ∂
v (p) = fi (p) =⇒ v = fi .
∂xi p ∂xi

for real-valued functions (the coefficients for our basis vectors) fi .

16
Definition 63
A vector field v is C ∞ if v is C ∞ , or equivalently if the corresponding fi s are functions in C ∞ (U).

The duals to these vector fields are the one-forms: as we defined last time, a one-form is functions u which send
points p ∈ U to elements of Tp∗ U. In particular, if f ∈ C ∞ (U) is a smooth function, then df is the one-form on U
mapping
df : p 7→ dfp .

The coordinate functions x1 , · · · , xn on U then give us one-forms dx1 , · · · , dxn . Since we’ve shown that these are basis
elements of Tp∗ U, we find that every one-form u on U can be written uniquely as

u = f1 dx1 + · · · + fn dxn

for some real-valued functions fi : U → R.

Definition 64
A one-form is C ∞ if the corresponding fi s are C ∞ functions. The space of C ∞ one-forms is denoted Ω1 (U).

Example 65
If f ∈ C ∞ (U) is any smooth function, then the one-form df defined above is C ∞ . Indeed, as we might expect,
P ∂f
df = ∂xi dxi .

We’ll finish this lecture by defining the pullback operation on one-forms:

Definition 66
Let U ∈ Rn and V ∈ Rm be open sets, and let f : U → V be a C ∞ map. Then given a one-form ν on V , we can
define the one-form fν∗ via
fν∗ (p) = (dfp )∗ ν(f (p)),

where (dfp )∗ : Tq∗ V → Tp∗ U is the map we defined earlier as being given by the transpose of dfp : Tp U → Tq V .

Problem 67
Suppose y1 , · · · , ym are the standard coordinates on V , and fi = fy∗i . Then any C ∞ one-form on V of the form
ν = ai dyi has corresponding pullback f ∗ ν =
P ∗
f ai dfi , so that f ∗ ν is a C ∞ one-form if ν is.
P

In particular, f ∗ is a map Ω1 (V ) → Ω1 (U).

11 February 23, 2022

Fact 68
Our last lecture was just yesterday, and today’s lecture is a lot of review of that material (because of the new
definitions).

17
Last lecture, we made a lot of important definitions relevant to our eventual introduction of differential forms.
Specifically, we discussed the tangent space Tp U of a point p in an open set U, and we mentioned that when we have
a smooth map f : U → W between open sets, we can also define a base-pointed version of the differential map Df (p),
which we call dfp (this just maps base point p to the new base point f (p) and applies Df to the “tangent” vector v ).
We then have a chain rule (which is essentially a matrix multiplication statement) for these base-pointed differential
maps, given by
(dg ◦ f )p = dgq ◦ dfp .

We next defined the dual space Tp∗ U; specifically, base-pointed differential maps dfp can be thought of as maps
(U, p) → (R, q) ∼
= R, which are linear maps on the tangent space and thus elements of that dual space. And with all
of these new vector spaces, we can construct associated linear bases: for example, the (dxi )p for 1 ≤ i ≤ n form a
∗ ∂
basis of Tp U if we’re in n-dimensional space, and the dual basis of Tp U consists of the elements ∂xi = (p, ei ) for
p
1 ≤ i ≤ n.
 

Remark 69. The reason for the definition ∂xi = (p, ei ) is essentially that an element (p, v ) of the tangent space
p
encodes a vector v rooted at p, so a vector pointed in the ei direction is inherently connected to the notion of taking
the derivative along the xi coordinate.
Additionally, we defined one-forms, which are maps from U to Tp∗ U (assigning to each point p a corresponding
element up ∈ Tp∗ U). Today, we’ll start by discussing operations that we can do on these one-forms:

1. If u1 , u2 are one-forms on U, then u1 + u2 is the one-form u such that u(p) = u1 (p) + u2 (p) for all p ∈ U.

2. If u is a one-form on U and φ ∈ C ∞ (U) is a smooth function, then φu is the one-form such that (ℓu)(p) =
φ(p)u(p) for all p. (But we cannot just multiply two one-forms together because u(p) is not a number.)

3. If ρ ∈ C ∞ (U) is a smooth function, then we can define the one-form dρ which sends p 7→ dρp . In particular,
each coordinate function xi is a smooth function, so dxi (for 1 ≤ i ≤ n) are all one-forms, and in fact every
one-form is “locally” a linear combination of this nature: we have a class of C ∞ one-forms, denoted Ω1 (U), given
by
u = φ1 dx1 + · · · + φn dxn

for some φi ∈ C ∞ (U) (and more generally, every one-form is of this form but for arbitrary real-valued functions
φi ). In particular, Ω1 (U) is closed under the first two operations of addition and multiplication by φ.

The last object that we defined last lecture is the pullback operation: if we have a smooth map U → V , then we
get a corresponding map f ∗ : Ω1 (V ) → Ω1 (U) mapping one-forms via ν 7→ f ∗ ν. Specifically (changing the notation
slightly from last lecture), we have
(f ∗ ν)p = (dfp )∗ νq ,

where (as usual) q = f (p). As mentioned last time, one way to understand why this pullback operation is defined in
this way is that if ν = dxi and our smooth function is f = (f1 , · · · , fm ), then f ∗ dxi = dfi . (And more generally, if ν is
a C ∞ one-form and f is a smooth mapping, then f ∗ ν is also a C ∞ one-form.)
We’ll finish this lecture by reviewing a few operations on vector fields, analogous to the ones on one-forms before:
if v1 , v2 are vector fields (maps U → Tp U), then we can define v1 + v2 by pointwise addition

(v1 + v2 )(p) = v1 (p) + v2 (p).

We can also define multiplication by a smooth function φ : U → R

φv (p) = φ(p)v (p).

18
 
∂ ∂
Finally, the objects ∂xi are vector fields on U which send p to ∂xi , and all vector fields are linear combinations of
p
these fundamental vector fields of the form
X ∂
v= φi .
∂xi
In particular, if φi are C ∞ functions, then we call v a C ∞ vector field.

12 February 28, 2022


(Friday’s class did not occur because of a snow day, so we’re having that lecture instead.) We’ll discuss the theory of
integral curves of vector fields today.

Definition 70
Let U be an open set in Rn , and let v be a vector field on U. A function γ : (a, b) → U is an integral curve of v
if for all a < t < b, if we define p = γ(t), we have
 

v (p) = p, (t) .
dt

More explicitly, if our curve is written out as γ(t) = (x1 (t), · · · , xn (t)), then this equation has the more explicit
form
dxi
(t) = vi (x(t)),
dt
P ∂
where we’re writing out our vector field explicitly as v = vi ∂xi . With this, we can do some basic ODE theory. We’ll
start by citing some relevant local results:

Fact 71 (Existence of integral curves)


Let v be a vector field, and let p0 ∈ U and a ∈ R. Then there exists an interval I = (a − ε, a + ε) for some ε > 0,
an open set U0 ⊂ U containing p0 , and a C ∞ map F : U0 × I → U, such that γp (t) is an integral curve of v .

Fact 72 (Uniqueness of integral curves)


Suppose γ1 : I1 → U and γ2 : I2 → U are two integral curves of v . Then if γ1 (t0 ) = γ2 (t0 ) for some t ∈ I1 ∩ I2 ,
then γ1 (t) = γ2 (t) for all t ∈ I1 ∩ I2 , and patching the two curves

γ1 (t) t ∈ I1 ,
γ(t) =
γ (t) t ∈ I ,
2 2

also results in an integral curve of v .

Fact 73
If we let I = (a, b) and Ic = (a − c, b − c), then given an integral curve γ(t) : I → U, we also have the integral
curve γ(t + c) : Ic → U.

There are also some relevant global results:

19
Definition 74
A vector field v on U is complete if for every p ∈ U, there exists an integral curve γp (t) : R → U such that
γp (0) = p, and the map F : U × (−∞, ∞) → U defined by F (p, t) = γp (t) is a C ∞ map.

In other words, there is an integral curve of v going through any point p in our open set which exists for all time.
(And these notions will come up when we generalize from Euclidean space to manifolds as well.) This means that the
map ft : U → U defined by ft (p) = F (t, p) is a C ∞ map. Furthermore, f0 is the identity map on U, and ft ◦ fa = ft+a
(so that ft and f−t are inverses). While this condition looks strong, it turns out that there is a large collection of
vector fields which satisfy this property:

Definition 75
vi ∂x∂ i on U is compactly supported if vi ∈ C0∞ (U) for all i .
P
A vector field v =

Proposition 76
If a vector field v is compactly supported, then v is complete.

Proof sketch. Suppose we have v (p0 ) = 0. Then the curve γ0 (t) = p0 for t ∈ R (not to be confused with the γp (t)
maps above) satisfies
d
0=γ0 (t) = v (p0 ),
dt
so γ0 is an integral curve for v and is the unique integral curve through the point p0 . Now defining the set A = {p ∈
U : v (p) = 0}, for any p ∈ A, the ODE uniqueness tells us that the unique integral curve of v through p is the constant
curve γ0 (t) = p0 .

Definition 77
If an integral curve γ on [0, T ) cannot be extended to a larger interval [0, T1 ), then it is maximal.

We claim that if γp (t) is maximal, then T = ∞. Indeed, if p ∈ A and γp (t) ∈ A for all 0 ≤ t < T , then
γp (t) → q ∈ A as t approaches T because of compactness of A. Then using local existence at uniqueness at q,
we can show that γp (t) can be extended to an interval 0 ≤ t ≤ T + ε (so we can always extend if T is finite). In
particular, γp (t) is well-defined for −∞ < t ≤ 0 and 0 ≤ t < ∞, and we’ve shown completeness.

Definition 78
A function φ ∈ C ∞ (U) is an integral of motion for the dynamical system generated by a vector field v if for every
d
integral curve γ(t), we have dt φ(γ(t)) = 0.

Theorem 79
A necessary and sufficient condition for φ to have this property is that the Lie derivative of φ is Lv φ = 0.

Proof. We have  
d d
φ(γ(t)) = (dφ)p γp (0) = (dφ)p v (p) = 0
dt t=0 dt
by the definition of an integral of motion.

20
Proposition 80
Suppose φ is proper (meaning that preimages of compact sets are compact). Then φ−1 ([−a, a]) is compact for
all a ∈ R.

We can prove this by noting that if φ(γ(t)) is constant for a < t < b, then γ(t) can’t go off to ∞ as t → a or
t → b.

Example 81
Let U = R2 , and let v = x1 ∂x∂ 2 − x ∂ be our integral curve. We can check that the function φ(x1 , x2 ) = x12 + x22
∂x1
is an integral of motion for v , so v is complete.

Next lecture, we’ll connect this back to our definition of differential forms and the definition of the space Λk (V ).

13 March 2, 2022
We’ll start by reviewing some material from previous lectures about the exterior algebra: recall that if Lk (V ) is
the space of k-tensors on V , and I k (V ) is the space of redundant k-tensors, spanned by elements ℓ1 ⊗ · · · ⊗ ℓk
where ℓi = ℓi+1 , we can define the space Λk (V ∗ ) = Lk (V )/I k (V ). Then because the tensor product of redundant
k-tensors is itself redundant, we can also define the wedge product operation as follows: letting πk be the projection
Lk 7→ Lk (V )/I k (V ), it is well-defined to let the wedge product of ω1 = πk1 T1 and ω2 = πk2 T2 be

ω1 ∧ ω2 = πk1 +k2 (T1 ⊗ T2 ) = Λk (V ∗ ).

In particular, if ℓ1 , · · · , ℓn form a basis of V ∗ , then the wedge products ωI = ℓi1 ∧ ℓi2 · · · ∧ ℓik with increasing indices
i1 < i2 < · · · < ik form a basis of Λk (V ∗ ), and the dimension of Λk (V ∗ ) is kn (so Λk (V ∗ ) = 0 for k > n).


Fact 82
n
We will set Λ0 (V ∗ ) = R for convention, and this is consistent with having an

0 = 1-dimensional vector space.

Today, we’ll now connect this back to the definition of k-forms and differential forms. If we let U ⊆ Rn be an open
subset and p ∈ U be a point, recall that the tangent space of U is the set of points {(p, v ) : v ∈ Rn }.

Definition 83
A k-form on an open set U ⊆ Rn is a “function” A, which assigns to each p ∈ U an element ωp ∈ Λk (Tp∗ U).

Example 84
If ω1 , · · · , ωk are 1-forms on U, then for any p ∈ U, ω1 (p) ∈ Tp∗ U, so ω1 (p) ∧ · · · ∧ ωk (p) ∈ Λk (Tp∗ U). Therefore,
ω1 ∧ · · · ∧ ωk is a k-form on U, assigning p ∈ U to ω1 (p) ∧ · · · ∧ ωk (p).

Example 85
If fi ∈ C ∞ (U) are smooth functions, then dfi is a 1-form for each 1 ≤ i ≤ k, so df1 ∧ · · · ∧ dfk is a k-form. More
specifically, letting I = (i1 , · · · , ik ) be a multi-index of length k, we can define dxI = dxi1 ∧ · · · ∧ dxik .

21
Since (dx1 )p , · · · , (dxn )p form a basis of the cotangent space Tp∗ U, we find that the elements

(dxI )p = (dxi1 )p ∧ · · · ∧ (dxik )p , i1 < i2 < · · · < ik

form a basis for the kth exterior power of the cotangent space, Λk (Tp∗ U). Therefore, given any k-form ω on U, we
can write
X
ωp = fI (p)dxI ,
I

where the sum I goes over increasing multi-indices I and fI (p) are each real numbers. Therefore, we are really saying
that the k-form can be represented as
X
ω= fI dxI ,
I

where fI : U → R is a function mapping each p ∈ U to fI (p).

Remark 86. As a check, notice that if I = (i1 , · · · , ik ) is a repeating multi-index with ir = ir +1 , then (dxi1 )p ∧ · · · ∧
(dxik )p = 0 for all p, and thus dxI = 0. On the other hand, for any non-repeating multi-index I, I σ is strictly increasing
for some permutation σ, and then (dxI )p = (−1)σ (dxI σ )p so we do not need to include the non-increasing dxI s in our
sum for ω. (This is why our sum only needs to go over increasing multi-indices.)

Definition 87
With the notation above, a k-form ω is C ∞ if the fI ’s are each in C ∞ (U), and we let Ωk (U) denote the space
(linear span) of C ∞ k-forms on U.

We can now take our discussion above about the wedge product into consideration:

Definition 88
Let ω1 ∈ Ωk1 (U) and ω2 ∈ Ωk2 (U). The wedge product ω1 ∧ ∧2 is the (k1 + k2 )-form which sends p ∈ U to
ω1 (p) ∧ ω2 (p) ∈ Ωkp1 +k2 (U).

We can verify from the definition directly that the wedge product is indeed C ∞ . This is one of the two fundamental
operations that we’ll be using for k-forms, and the other we’ll now discuss (the d− operation). We define Ω0 (U) =
C ∞ (U), and we’ll start by considering a 0-form f (which is just a function). Motivated by the differential statement
n
X ∂fi
df = dxi ,
∂xi
i=1

we make the following definition:

Definition 89
P
For a general k-form uniquely written as ω = I fI dxI (with sum over increasing multi-indices I), we define
X
dω = dfI ∧ dxI .

We are now ready to discuss a few important properties based on these definitions:

22
Theorem 90
We have the following properties of k-forms:
1. For ω1 , ω2 ∈ Ωk (U), we have d(ω1 + ω2 ) = dω1 + dω2 .

2. For ω1 ∈ Ω1 (U) and ω2 ∈ Ω2 (U), d(ω1 ∧ ω2 ) = dω1 ∧ ω2 + (−1)k1 ω1 ∧ dω2 .

3. For any k-form ω, we have d(dω) = 0.

Looking back at Remark 86, we’re saying that permutations do not really cause us issues up to a sign, and in fact
for all multi-indices I we have
d(fI ∧ dxI ) = dfI ∧ dxI .

To prove (2), notice that for f , g ∈ C ∞ (U), we can use the product rule to find the special result for one-forms
X ∂f X ∂g
d(f g) = g dxi + f dxi =⇒ d(f g) = gdf + f dg.
∂xi ∂xi

More generally, if we take ω1 = fI dxI ∈ Ωk1 (U) and ω2 = gJ dxJ ∈ Ωk2 (U), we find that

d(ω1 ∧ ω2 ) = d(fI gJ )dxI ∧ dxJ ,

and then using that product rule above gives us point (2) from the theorem.

14 March 4, 2022

Fact 91
Because Wednesday’s lecture covered a lot of material, most of this class is review of that content.

We’ll start today’s lecture by reviewing the interior product operation from a few lectures ago in the context of
the exterior algebra: recall that for any v ∈ V and any ω = ℓ1 ∧ · · · ∧ ℓk ∈ Λk (V ∗ ), we can define the interior product
ι(v )ω, an element of Λk−1 (V ∗ ), via
n
X
ι(v )ω = (−1)r −1 ℓr (v )ℓ1 ∧ · · · ∧ ℓr −1 ∧ ℓr +1 · · · ∧ ℓk ,
r =1

and one fact we can prove as an exercise is that ι(v )ι(v )ω = 0 for any v .
Last lecture, we also defined k-forms, which take in a point p ∈ U ⊆ Rn and output an element of Λk (Tp∗ U).
(In particular, the wedge product of k 1-forms on U is a k-form, and those are easy to think about because 1-forms
assign to each point p ∈ U an element ωp ∈ Tp∗ U). Specifically, if ωi = dfi for C ∞ functions fi , we get k-forms that
look like df1 ∧ · · · ∧ dfk , and more specifically, we can define the k-forms dxI = dxi1 ∧ · · · ∧ dxik for any multi-index
I = (i1 , · · · , ik ). And last time, we described that because (dxi )p form a basis for the cotangent space Tp∗ U, the (dxI )p
for increasing I form a basis for Λk (Tp∗ ), meaning any k-form sends p to some ωp =
P P
fI (p)dxI , and thus ω = fI dxI
for some functions fI : U → R.
From here, we can add k-forms ω1 , ω2 together to get another k-form ω1 + ω2 , and we can wedge any k1 -form ω1
and k2 -form ω2 together (by sending p to ω1 (p) ∧ ω2 (p) – this gives us a C ∞ (k1 + k2 )-form as long as ω1 , ω2 were
P ∂f
C ∞ ). And we also have the important d operation, motivated by df = i ∂x i
xi from calculus (taking a 0-form f to a

23
1-form): we have
X X
ω= fI dxI =⇒ dω = dfI ∧ dxI .
I I

The important properties of this d operation are then that d(ω1 + ω2 ) = dω1 + dω2 , d(dω) = 0, and also that

d(ω1 ∧ ω2 ) = dω1 ∧ ω2 + (−1)k1 ω1 ∧ dω2 .

In particular, because dxI σ = (−1)σ dxI , combining this fact with the “product rule” above tells us that d(fI dxI ) =
dfI ∧ dxI for all multi-indices I, which will make later calculations easier. And we can check that dd = 0 by looking at
k-forms of the form ωfI dxI and plugging in

ddω = d(dfI ∧ dxI ) = d(dfI ) ∧ dxI − dfI ∧ d(dxI ),

and noticing that d(dfI ) = 0 and d(dxI ) = 0 (left as an exercise by writing out the double sum and thinking about
reversing the indices in the sum).

15 March 7, 2022
Last time, we started by talking about the interior product operation, sending a k-form ω ∈ Λk (V ∗ ) to a (k − 1)-form
ι(v )ω ∈ Λk−1 (V ∗ ): by linearity we can just consider the action on ω = ℓ1 ∧ · · · ∧ ℓk for ℓi ∈ V ∗ , and we have
k
X
(−1)r −1 ℓr (v ) ℓ1 ∧ · · · ∧ ℓr −1 ∧ ℓr +1 ∧ · · · ∧ ℓk .

ι(v )ω =
r =1

We’ll mention a few more properties of that operation today: notice from the definition that ι(v )ω is linear in both
v and in ω, and also for any one-form ω, we have ι(v )ω = ω(v ) ∈ R. Additionally, connected to the fact that
ι(v )ι(v )ω = 0 for any v , we also have ι(v1 )ι(v2 )ω = −ι(v2 )ι(v1 )ω. Finally, combining the properties of the interior
and wedge product gives us (again being careful about the sign change)

ι(v )(ω1 ∧ ω2 ) = ι(v )ω1 ∧ ω2 + (−1)k ω1 ∧ ι(v )ω2 .

We’ll now generalize this definition to vector fields:

Definition 92
Let U ⊆ Rn be an open set, v a vector field on U, and ω ∈ Ωk (U) a differential k-form. Define ι(v )ω to be the
differential (k − 1)-form given by
(ι(v )ω)p = ι(vp )ωp .

(It’s left as an exercise for us to check that if v is a C ∞ vector field, and ω ∈ Ωk (U), then ι(v )ω is indeed in
Ωk−1 (U).) Our properties from before now generalize – in particular, linearity in v and ω of the interior product
operation imply that for any k-forms ω1 , ω2

ι(v )(ω1 + ω2 ) = ι(v )ω1 + ι(v )ω2 , ι(v1 + v2 )ω = ι(v1 )ω + ι(v2 )ω,

and the wedge product identity also gives us (for any k1 -form ω1 and k2 -form ω2 )

ι(v )(ω1 ∧ ω2 ) = ι(v )ω1 ∧ ω2 + (−1)k1 ω1 ∧ ι(v )ω2 .

24
Also, we still have ι(v )ι(v )ω = 0 and thus ι(v )ι(w )ω = −ι(w )ι(v )ω like before. And finally, we can write out the
interior product operation more explicitly as
k
X
ι(v )(dxi1 ∧ · · · ∧ dxik ) = (−1)ℓ−1 vℓ dxi1 ∧ · · · ∧ dxiℓ−1 ∧ dxiℓ+1 ∧ dxik
ℓ=1

(where vℓ is the ℓth coordinate of the vector field, and it outputs a number if we evaluate it at a point p ∈ U).
We’re now ready to talk about a more complicated operation on forms:

Definition 93
Let ω ∈ Ωk (U), and let v be a C ∞ vector field. The Lie differentiation operation is defined via

Lv ω = ι(v )dω + d(ι(v )ω).

Proposition 94
This Lie differentiation operation commutes with the d operation:

Lv dω = dLv ω.

Additionally, when interacting with the wedge product, we have

Lv (ω1 ∧ ω2 ) = Lv ω1 ∧ ω2 + ω1 ∧ Lv ω2 .

Proof. For the first property, note that (by definition)

Lv dω = ι(v )ddω + dι(v )dω,

while
d(Lv ω) = dι(v )dω + ddι(v )ω.

But the terms with dd now vanish (because applying the d operation twice gives us 0), and thus the two terms are
indeed both equal (to dι(v )dω). The second property is left as an exercise to us, but we can use the fact that by
definition,
Lv ω1 ∧ ω2 = dι(v )(ω1 ∧ ω2 ) + ι(v )d(ω1 ∧ ω2 )

and now simplifying with properties of d and ι gives us

= (dι(v )ω1 ) ∧ ω2 ) + (−1)k ω1 ∧ dι(v )ω2 + ι(v )(dω1 ∧ ω2 + (−1)k ω1 ∧ dω2 ).

We’re now going to turn to the divergence formula. Notice that for any vector field which we write in the form
P ∂
v= vi ∂xi , we have
Lv dxi = dι(v )dxi + ι(v )ddxi = dvi .

In other words, if we have specifically an n-form ω = dx1 ∧ · · · ∧ dxn , we find that


X
Lv ω = dx1 ∧ · · · ∧ dxi−1 ∧ Lv dxi ∧ dxi+1 ∧ · · · ∧ dxn .
i

25
P ∂vi
Now because Lv dxi = dvi and dvi = i ∂xj dxj , and notice that dxi ∧ dxj = 0 for i ̸= j, meaning all of the contribution
to the blue term that is not from dxi goes away. So in fact our expression simplifies to
X ∂vi
Lv ω = dx1 ∧ · · · ∧ dxi−1 ∧ dxi ∧ dxi+1 ∧ · · · ∧ dxn ,
∂xi
i

and thus we can understand the action of the Lie differentiation operation as it connects to calculus:
!
X ∂vi
Lv ω = ω,
∂xi
i

and this term in parentheses is the divergence of the vector field v which we might remember from calculus.
We’ll finish this lecture by starting to look again at functorial properties of differential forms (and continue this
next time): recall from early on in 18.952 that for any linear map A : W1 → W2 of vector spaces, we get the dual map
A∗ : W2∗ → W1∗ . Recall that we get a linear map A∗ : Λk (W2∗ ) → Λk (W1∗ ) as well, with the additional properties that
(this is how we can define the map)
A∗ (ω1 ∧ ω2 ) = A∗ ω1 ∧ A∗ ω2 ,

and for any linear map B : W2 → W3 we also get

(BA)∗ ω = A∗ B ∗ ω.

We can now generalize this by letting U be an open set in Rn and V an open set in Rm : for any map f : U → V such
that p 7→ q, we get the map dfp : Tp U → Tq V and also the corresponding dual map dfp∗ : Λk (Tq∗ V ) → Λk (Tp∗ U). The
assertion is now that for any ω ∈ Ωk (V ), we have an element ωq ∈ Λk (Tq∗ V ), and then we can define the “pullback”

(f ∗ ω)p = (dfp )∗ ωq ,

allowing us to define the k-form f ∗ ω. In particular, we have f ∗ dφ = df ∗ φ for any φ ∈ C ∞ (V ), and for coordinates
x1 , · · · , xn and a function f = (f1 , · · · , fn ), we have f ∗ dxi = dfi (so that the pullback of a general k-form
P
aI dxi1 ∧
· · · ∧ dxik is obtained by replacing aI with f ∗ aI and dxi1 with dfi1 ). But we’ll talk more about this next time!

16 March 9, 2022
Last lecture, we generalized the interior product operation to vector fields: for a C ∞ vector field on U and a differential
k-form ω ∈ Ωk (U), we get the differential (k − 1)-form ι(v )ω defined as

(ι(v )ω)p = ι(vp )ωp ,

where recall that ωp is an element of Λk (Tp∗ U) and vp is an element of Tp U (so that this interior product makes sense).
We also introduced the Lie differentiation operation

Lv ω = ι(v )dω + dι(v )ω,

and we can also consider the following alternate definition:

26
Definition 95
Suppose ft : U → U is a one-parameter group of diffeomorphisms generated by a complete vector field v , meaning
that we have paths γp for each p ∈ U satisfying γp (t) = ft (p). (Recall that then in fact γp is the unique integral
curve of v satisfying γ0 (p) = p.) Then we can also define Lie differentiation via

d ∗
Lv ω = f ω .
dt t t=0

Proposition 96
With the definition above, we also have (for all time t)

d ∗
f ω = fta stLv ω.
dt t

Proof. We have (by change of variables in ordinary calculus)



d ∗ d ∗
f ω= f (ω) ,
dt t ds s+t s=0

and because we have a one-parameter group, we can rewrite this as


d
= (f ∗ ω ◦ fs∗ ω)s=0 ,
ds t
and pulling out the ft∗ and applying the definition gives us what we want.

Thus, the relation for the Lie derivative can be written


d ∗
f ω = ft∗ (ι(v )dω + dι(v )ω) = ft∗ ι(v )dω + dft∗ ω,
dt t
which we will define to be
dQt ω + Qt dω, Qt = ft∗ ι(v )ω.

If we then integrate this equation from a to b, we find that

fb∗ ω = fa∗ ω = Q̃a,b dω + d Q̃ab ω,

where Q̃a,b is the integral of Qt from a to b. This motivates the following definition:

Definition 97
Suppose ω1 , ω2 ∈ Ωk (U) are differential k-forms on U, and suppose they are closed (meaning that dω1 = 0 and
dω2 = 0). Then ω1 and ω2 are cohomologous, denoted ω1 ∼ ω2 , if ω1 − ω2 = du for some u ∈ Ωk−1 (U).

In particular, the calculation above for fb∗ ω − fa∗ ω shows the following result:

Theorem 98
Suppose ω ∈ Ωk (U) is a closed k-form. Then fb∗ ω ∼ fa∗ ω for all a, b.

Since f0 is always the identity map in a one-parameter family, this tells us that

fb∗ ω − ω = d Q̃b ω

27
for Q̃b = Q̃b,0 . In other words, the pullback operation does not change the “cohomology class” – the difference between
ω and fb∗ ω is d of some differential form.
We’ll now move on to the integration operation in the theory of differential forms. Recall that for any n-
dimensional vector space, the nth exterior power of V , Λn (V ∗ ), is a one-dimensional space, so Λn (V ∗ ) − {0} has two
components, and an orientation of V is the choice of one of these two components which we denote Λn (V ∗ )+ (the
other is then denoted Λn (V ∗ )− ). Furthermore, if we let v1 , · · · , vn be a basis of V and v1∗ , · · · , vn∗ be the dual basis,
then (v1 , · · · , vn ) is positively oriented if v1∗ ∧ · · · ∧ vn∗ is in Λn (V ∗ )+ and negatively oriented otherwise.

Definition 99
Let A : V → W be a bijective linear map between n-dimensional vector spaces. Then A is orientation-preserving
if A∗ : Λn (W ∗ ) → Λn (V ∗ ) maps Λn (W ∗ )+ onto Λn (V ∗ )+ .

This can be written more explicitly using coordinates: if (v1 , · · · , vn ) is a basis of V and (w1 , · · · , wn ) is our basis
P
of W , then we can represent A with the matrix [aij ] where Avj = i aij wi , and furthermore

A∗ (w1∗ ∧ · · · ∧ wn∗ ) = det[aij ]v1∗ ∧ · · · ∧ vn∗ .

We can thus say that A is orientation-preserving if det[aij ] > 0 (which might look familiar from other linear algebra
classes we’ve taken).
We can generalize these concepts to open subsets of Rn as well:

Definition 100
Let U be an open set in Rn , and let p ∈ U. We can orient the space Tp U by requiring that

(dx1 ∧ · · · ∧ dxn )p ∈ Λn (Tp∗ U)+ .


   
In other words, we can alternatively choose that ∂x∂ 1 , · · · , ∂x∂ n form a positively oriented basis for Tp U.
p p

Definition 101
Suppose f : U → V is a C ∞ diffeomorphism between open sets U, V of Rn . Then f is orientation-preserving if
for all p ∈ U and q = f (p), dfp : Tp U → Tq V is orientation-preserving.

We can apply this to the change of variables formula from calculus: recall that for any C ∞ diffeomorphism
f : U → V , we can make a “u-substitution”
Z Z
φdx = f ∗ φ |det Js | dy ,
V U

f ∗ det Js if f is orientation-preserving, and


R R
where Js denotes the Jacobian matrix. In other words, we have V φ= U
we have a negative sign in that equality otherwise. We can now rewrite this standard calculus result in a nicer way:
let f : U → V be a diffeomorphism. Then if p ∈ U and q = f (p), then
   
∂ X ∂fi ∂
(dfp ) = (p) .
∂xj p ∂xj ∂xi p
i

We then get the dual result

dfp∗ (dx1 ∧ · · · ∧ dxn )p = det(Jf )p (dx1 )p ∧ · · · ∧ (dxn )p

28
h i
∂fi
for the Jacobian matrix (Jf )p = ∂xj (p) , so that our change-of-variables formula now reads

f ∗ (dx1 ∧ · · · ∧ dxn ) = det(Jf )dx1 ∧ · · · ∧ dxn .

And next lecture, we’ll see a more intrinsic version of this formula that doesn’t involve matrices at all!

17 March 11, 2022


Remark 102. I was not able to attend this lecture in person, so these notes are transcribed from Paige Bright and
Jeffery Yu’s class notes.

Last lecture, we mentioned the change of variables formula from integral calculus, which states that for an
orientation-preserving diffeomorphism f : U → V between open subsets of Rn and a compactly supported differ-
ential form ω ∈ Ωkc (V ), we have Z Z
ω= f ∗ ω.
V U
We’ll be proving that result next week, but we’ll work on some other results towards that goal for now:

Theorem 103 (Poincaré lemma for rectangles)


Let U = I1 × · · · × In ⊂ Rn be the open rectangle with Ir = (ar , br ) for all r . Suppose ω ∈ Ωkc (U). Then ω is
R
exact, meaning that ω = dµ for µ ∈ Ωk−1c (U), if and only if Rn ω = 0.

Pk
Proof of forward direction. By the definition of the d operation, we know that if we write µ = i=1 fi dxi ∧ ··· ∧
dxi−1 ∧ dxi+1 ∧ · · · ∧ dxn , then
X ∂fi
ω = dµ = (−1)i−1 dx1 ∧ · · · ∧ dxn .
∂xi
i

In particular, this means that if ω = dµ, we have


Z Z Z
∂fi
ω= ω = (−1)i−1 dx1 ∧ · · · ∧ dxn
R n U U ∂x i
X Z ∂fi 
= dxi dx1 dxi ∧ · · · ∧ dxi−1 ∧ dxi+1 ∧ · · · ∧ dxn
Ir ∂xi
X
= (fi (br ) − fi (ar )) dx1 dxi ∧ · · · ∧ dxi−1 ∧ dxi+1 ∧ · · · ∧ dxn

(slightly abusing notation here because fi (br ) and fi (ar ) depend on the other xi values too, but the logic still works)
because all fi s are compactly supported and thus fi vanishes on the boundary.

For the other direction, we’ll need to do some more work, but it follows directly from the following result:

Theorem 104
R
Let U ⊆ Rm be open, and let A ⊂ R be an open interval. Suppose U satisfies that for all ω ∈ Ωm
c (U) with ω = 0,
we have ω ∈ dΩm−1
c (U). Then A × U also satisfies this condition.

Proof. We proceed by induction on m.The base case m = 1 is where U = (a, b) ⊂ R. Indeed, if f ∈ C0∞ (A) and
Rb Rx ∞
a f (t)dt = 0, then we can use the function g(x) = 0 f (t)dt ∈ C0 (U), and we indeed have dg = f (this is the
ordinary fundamental theorem of calculus).

29
Say that an open set that satisfies the condition in this theorem satisfies CPL. We then want to prove that if U
is CPL, then A × U is also CPL. For this, let x1 , · · · , xn be coordinates on U and t, x1 , · · · , xn be the corresponding
coordinates on A × U. We can then write any ω ∈ Ωm+1 (A × U) as

ω = f (x, t)dt ∧ dx1 ∧ · · · ∧ dxn ,


R R
with corresponding integral ω= A×U f (x, t)dt ∧ d⃗
x . Define
Z 
θ= f (x, t)dt d⃗
x,
A
R R R R
so that Rn ω = Rn f (x, t)dt = Rn−1 dx (in other words, integrate out the variable t in advance). Then if ω = 0, then
R
θ = 0 as well. So now suppose U is CPL (as we have in our assumption), so that θ = dν for some ν ∈ Ωcn−1 (U). Let
ρ ∈ C ∞ (R) be a bump function (in particular, compactly supported) on A, such that A ρ = 1. Defining K = ρdt ∧ ν,
R

we then have
dK = ρdt ∧ dν = ρ(t)dt ∧ θ,

so that
ω − dν = dt ∧ (ν − ρ(t)θ) = dt ∧ u(x, t)dx1 ∧ · · · ∧ dxn−1 ,

where we define the function Z


u(x, t) = f (x, t) − ρ(t) f (x, t)dt.
A
R
In particular, because ρ integrates out to 1, we find that u(x, t)dt = 0. So for our interval A = (a, b), if we define
Z t
v (x, t) = u(x, s)ds,
a

we’ll have v ∈ C0∞ (U × A) and ∂t v = u by the ordinary fundamental theorem of calculus. We’re now ready to finish:
letting φ = v (x, t)dx1 ∧ · · · ∧ dxn−1 , we then have that

dφ = dt ∧ u(x, t)dx1 ∧ · · · ∧ dxn−1 = ω − dK,

so that ω = d(φ − K), and we’ve proved CPL for A × U.

We’ve now proved both directions of the Poincaré lemma for rectangles, and next we’ll generalize this to compactly
supported forms on open subset of Rn .

Theorem 105 (Poincaré lemma)


Let U be a connected open subset of Rn , and let ω ∈ Ωnc (U) (so that the support of ω is contained in U). Then
R
ω = 0 if and only if there is some µ ∈ Ωn−1
c (U) with ω = dµ.

The backwards direction follows from the Poincaré lemma for rectangles. Indeed, if µ ∈ Ωn−1
c (U), then the support
R
of µ is contained within some rectangle and thus dµ = 0.
For the forwards direction, suppose ω1 , ω2 ∈ Ωnc (U). Recall that we write ω1 ∼ ω2 if ω1 − ω2 = dµ for some
compactly supported (n − 1)-form µ. It’s equivalent to prove the following:

30
Theorem 106
Let Q0 ⊂ U be a rectangle, and let ω0 be an n-form with support contained in Q0 and total integral 1. Let
R
ω ∈ Ωnc (U) be a compactly supported n-form with support contained in U and total integral c = ω. Then we
have ω ∼ cω0 .

In particular, setting c = 0 proves the forward direction of the Poincaré lemma, since it implies that ω ∼ 0 and
thus ω = dµ for some µ.

Proof. Construct a collection of rectangles Qi ⊂ U such that U is the union of the interiors of Qi . Let φi be a partition
of unity (meaning we have compactly supported functions which in total add to the function 1 on U) such that supp(φi )
is contained within the interior of Qi for each i . Then for large enough m, ω = m
P
i=1 φi ω by compactness, so we can
reduce the statement to the case where we prove this result for each φi ω. In other words, it suffices to prove this
result when the support of ω is contained within the interior of some open rectangle Qi , which we call Q.
We claim that we can connect Q0 and Q with a sequence of rectangles Q0 = R0 , · · · , RN+1 = R such that the
interior of any two consecutive rectangles is non-empty. Indeed, the set

A = {x ∈ U : there exists a sequence {Ri } with x ∈ int(RN+1 )}

is an open set, and its complement is also open. Thus because U is connected, U = A. Now to finish the proof,
R
for each i , choose a compactly supported n-form vi contained within int(Ri ) ∩ int(Ri+1 ) such that vi = 0. Since
vi − vi+1 is supported in int(Ri+1 ), by the Poincaré lemma for rectangles we have vi ∼ vi+1 . The same logic says that
because we also have ω0 ∼ v0 , we have cω) ∼ cv0 , and finally ω ∼ cvN . Thus this chain of equivalences tells us that
ω ∼ cω0 , as desired, finishing the proof.

18 March 14, 2022


We’ll start today by going over yesterday’s proof of the Poincaré lemma in more detail. Recall that we started by
proving the rectangle version of the lemma, which states for any open rectangle Q = I1 × · · · × In and any ω ∈ Ωkc (Q)
(that is, ω is a compactly-supported k-form), if dω = 0, then ω ∈ dΩck−1 (Q) (because d 2 = 0). And a more involved
R
result is that we have Q ω = 0 if and only if ω is exact (meaning that ω = dµ for some µ).
Generalizing this to arbitrary connected open sets U ⊆ Rn then involved a lemma which proved that for any
p, q ∈ U, we have a sequence of rectangles Qi such that Qi ∩ Qi+1 ̸= ∅, and p is in the first rectangle and q is in
the last rectangle. We showed this by constructing the set of points q for which this is true, mentioning that it is
open (because there is always an neighborhood of q within a rectangle it’s contained in) and so is its complement (for
basically the same reason – otherwise we could find arbitrarily close points to q which can be reached, meaning q can
also be reached), so connectivity of U implies that this set of points must be all of U (since it is nonempty).
That lemma was useful because it then allows us to repeatedly use the rectangle Poincaré lemma. Specifically,
we are then able to prove a slight generalization of the Poincaré lemma, which is that if ω0 is a compactly supported
R
n-form on an open rectangle Q0 ⊆ U, then ω ∼ cω0 for c = U ω. The argument is to use compactness to break up
ω into a finite sum of forms supported on rectangles, and to note that the result we are trying to prove holds for finite
sums if it holds for the individual parts. From there, we use our connected set of rectangles {Qi }, showing repeated
equivalence of our n-forms by looking at compactly supported forms on the intersections Qi ∩ Qi+1 .
As of this result, we can now talk about the change of variables formula. The version that we’ve already talked about
R ∗ R
states that if f : U → V is a diffeomorphism between connected subsets of Rn , then U f ω = V ω if f is orientation-

31
preserving, and similar with a negative sign if f is orientation-reversing. Today, we’ll replace “diffeomorphism” with a
much weaker assumption, but first we’ll review a bit of point-set topology:

Proposition 107
Let X, Y be topological spaces, and let f : X → Y be an arbitrary continuous map. Then if C is a compact subset
of X, then f (C) is compact.

The converse is not true as stated (that is, the preimage of a compact set is not always compact), but there are
many maps for which the converse does hold:

Definition 108
A continuous map f : X → Y is proper if for every compact subset C, f −1 (C) is compact.

We can check that if f : X → Y and g : Y → Z are proper maps, then g ◦ f is also proper. We’ll discuss proper
maps more next lecture, but we’ll connect this back to differential forms for now:

Proposition 109
Let f : U → V be a proper map between open subsets U, V of Rn . Then if ω ∈ Ωkc (V ), then f ∗ ω ∈ Ωkc (U).

Proof. Let C be the support of ω. Then the support of the pullback form, f ∗ ω, is contained in the preimage f −1 (C),
which is compact.

Theorem 110
Let U, V be open subsets of Rn , and let f : U → V be a proper C ∞ map. Then for all ω ∈ Ωnc (V ), we have
Z Z

f ω = deg(f ) ω.
U V

Here, deg(f ) denotes the degree of the map f . We’ll give a full definition of what it is in the proof, but provisionally
R
we’ll use the following definition: if we fix (forever) an ω0 ∈ Ωnc (V ) with V ω0 = 1, then we set
Z
deg(f ) = f ∗ ω0 .
U

(This integral indeed makes sense because f ∗ ω0 is compactly supported.)


R
Proof. For any ω ∈ Ωnc (V ), define c = V ω. Then
Z Z Z
ω − cω0 = ω−c ω0 = c − c = 0,
V V V

so by the Poincaré lemma, ω − cω0 = dν for some ν ∈ Ωcn−1 (V ). Applying the pullback map, we find that

f ∗ ω − cf ∗ ω0 = f ∗ dν = df ∗ ν,

and because f ∗ ν is compactly supported integrating it out over U gives us 0. Thus


Z Z Z
∗ ∗
f ω= cf ω0 = cdeg(f ) = deg(f ) ω,
U U V

as desired.

32
We’ll convert this wacky definition of degree into something more fundamental next time (which is essentially about
“counting the number of points in the preimage”), instead of needing to rely on this choice of ω0

19 March 16, 2022


Last time, we discussed the Poincaré lemma, which states that any compactly supported n-form ω ∈ Ωnc (U) on a
R
connected open subset U of Rn has U ω = 0 if and only if ω ∈ dΩn−1
c (U). We then started analyzing the connections
of this result to the change of variables formula – in particular, if f : U → V is a proper map (meaning that preimages
of compact sets are compact), then ω ∈ Ωkc (V ) implies that f ∗ ω ∈ Ωkc (U), and this allows us to use the Poincaré
lemma to prove that Z Z
f ∗ ω = deg(f ) ω.
U V

However, last time, we gave a sketchy definition of the degree deg(f ), where we fix a compactly supported n-form ω0
with V ω0 = 1 and defined deg(f ) = U f ∗ ω0 . Specifically, with this definition and setting c = V ω, we found that
R R R
R
V ω − cω0 = 0, so that ω − cω0 = dν for some compactly supported (n − 1)-form ν. Applying the pullback map and
integrating then gave us
Z Z Z Z Z Z
f ∗ ω − cf ∗ ω0 = df ∗ ν = 0 =⇒ f ∗ ω = c f ∗ ω0 = deg(f ) ω.
U U V

We can now talk more about the properties of this degree map and start doing some applications:

Proposition 111
Suppose U, V, W are connected open subsets of Rn , and suppose f : U → V and g : V → W are proper C ∞ maps.
Then deg(g ◦ f ) = deg(f ) deg(g).

Proof. By functoriality, we have


(g ◦ f )∗ ω = f ∗ g ∗ ω,

so that Z Z Z Z Z
deg(g ◦ f ) ω (g ◦ f )∗ ω = f ∗ g ∗ ω = deg(f ) g ∗ ω = deg f deg g ω.
U W W V U

Equating the first and last expressions in this chain of equalities gives us the result.

We’ll now get this result to look more like the usual language of calculus: suppose U and V are connected
open subsets of Rn , and let f : U → V be a diffeomorphism and φ be a scalar function. Then we know that
R R ∗
V φ = U f φ| det Jf |. In the language of this course, for a differential form of the form

ω = φ(x)dx1 ∧ · · · ∧ dxn ,

f ∗ω = ±
R R R
we can then show that V ω = U V ω, depending on whether f : U → V is orientation-preserving or
orientation-reversing. Here’s that statement in alternative words:

Proposition 112
Let U and V be connected open subsets of Rn , and let f : U → V be a C ∞ diffeomorphism. Then deg(f ) = 1 if
f is orientation-preserving and −1 if orientation-reversing.

We’ll prove this topologically, but first we need preliminary results:

33
Lemma 113
Suppose U, V, W are connected open subsets of Rn , and suppose f : U → V and g : V → W are C ∞ diffeomor-
phisms. Suppose that Proposition 112 holds for both f and g. Then it also holds for g ◦ f .

(As a hint toward proving this, fix some ω ∈ Ωnc (W ). Then because g ◦ f is also a diffeomorphism, we know that
Z Z
∗ ∗ ∗
(g ◦ f ) ω = f g ω =⇒ (g ◦ f ) ω = ± g ∗ ω

because the degrees of f and g are both ±1.) We will also need the following (which we will prove as exercises):

Lemma 114
Let a ∈ Rn , and consider the translation map Ta : x 7→ x + a. Then P r oposi ti on 112 does indeed hold for the
map Ta . Also, if A : Rn → Rn is a bijective linear map, then P r oposi ti on 112 does hold for A.

Beginning of proof of Proposition 112. From our preliminary results, we may assume by applying translation maps
(before and after applying f ) that 0 ∈ U ◦ V and that f (0) = 0. This means that df0 is the identity map, so we can
then write f (x) = x + g(x), where g(0) = 0 and g ′ (0) = 0 so that g(x) ≤ C|x|2 for some C. If we now choose δ > 0
so that Cδ ≤ 12 , and we pick a function ρ(x) ∈ C0∞ (Rn ) such that 0 ≤ ρ(x) ≤ 1 and also satisfying

1 |x| ≤ 2δ ,
ρ(x) =
0 |x| ≥ δ,

we then have ρ(x)g(x) = 0 for |x| ≥ δ, so that


1
|ρ(x)g(x)| ≤ ρ(x)Cδ|x| ≤ |x|
2

and ρ(x) = g(x) for |x| ≤ 2δ . From here, we can look at the function f˜(x) = f (x) + ρ(x)g(x) and show that
deg(f ) = deg(f˜), and the remainder of the proof is topological.

20 March 18, 2022


After spring break, we’ll be shifting gears from differential forms on Rn to differential forms on manifolds. But for now,
we’ll focus on the concepts we’ve been discussing so far – recall that last time, we related the differential forms ω and
f ∗ ω via the change of variables formula, which made use of the notion of degree. Specifically, if we fix some compactly
supported n-form ω0 ∈ Ωnc (V ) with V ω0 = 1, then for any proper map f : U → V , we defined deg(f ) = U f ∗ ω0 .
R R

The idea is that this is actually an intrinsic definition which doesn’t depend on the choice of ω0 : specifically, for all
ω ∈ Ωnc (V ), we have that Z Z

f ω = deg(f ) ω,
U V
so the degree of f is indeed going to yield the same answer no matter which ω0 we choose.

Theorem 115
Let U, V be connected open subsets of Rn , and let f : U → V be a proper C ∞ map. If f is not onto, then
deg(f ) = 0.

34
Proof. To prove this result, we’ll need the following topological fact:

Fact 116
Let f : U → V be a proper C ∞ map between open subsets of Rn . Then f (U) is a closed subset of V , so if f is
not onto and q ∈ V \ f (U), there is an open neighborhood W of q in V , such that W ∩ f (U) = ∅.

R
We can then pick a compactly supported form ω0 ∈ Ωnc (W ) satisfying W ω0 = 1. Applying the pullback map, we
know that ωq = 0 for all q ∈ f (U), so 0 = U f ∗ ω0 and thus the degree of f is zero.
R

Our main goals for today’s lecture are to discuss how to compute degree geometrically and show homotopy
invariance of degree, again showing that the degree of a map f is very topological. First, the following is an exercise
in our textbook:

Proposition 117
Let U and V be open sets of Rn , and let f : U → V be a proper C ∞ map. Suppose we have some q ∈ V and
f −1 (q) ⊆ U0 for some open set U0 of U. Then there exists a neighborhood V0 of q in V , such that f −1 (V0 ) ⊆ U0 .

In other words, we can expand around a given point’s preimage by an “arbitrarily small amount.”

Definition 118
Let f : U → V be a proper C ∞ map. A point p ∈ U is a critical point if the map dfp : Tp → Tq (where q = f (p))
is not bijective. Let Cf be the set of critical points, also known as the critical set; we call f (Cf ) the set of critical
values.

We may notice that Cf is closed, and if f is proper, then f (Cf ) is also closed. Furthermore, the set of regular
values Vreg = V \ f (C) is open.

Lemma 119
For any regular value q of f , f −1 (q) is a finite set.

Proof. Since q is a regular value, any p ∈ f −1 (q) is a regular point, meaning that dfp : Tp → Tq is bijective. Thus, by
the inverse function theorem, f maps a neighborhood Up of p diffeomorphically to a neighborhood Vq of q.
We thus know that the Up s cover f −1 (q), and since f is proper, f −1 (q) is compact. By compactness, we can thus
find a finite subcover Up1 , · · · , UpN of f −1 (q). Because we have a bijective mapping from these neighborhoods Upi ,
the only point in each neighborhood Upi mapping to the point q is pi itself. Thus the preimage of q is the finite set
{q1 , · · · , qN }.

We can then choose the neighborhoods to be disjoint (by choosing small enough neighborhoods), so that we have
a neighborhood Vq of q and a neighborhood Upi of each pi satisfying
N
[
f −1 (Vq ) = Upi , Upi ∩ Upj = ∅ ∀i , j, f : Upi → Vq diffeomorphism.
i=1

(This is known as the stack of records theorem.) We can then finally understand another way to define the degree:

35
Theorem 120 (Degree theorem)
Let q be a regular value of f , and let {p1 , · · · , pN } be its preimage. Then we have
N
X
deg(f ) = σ pi ,
i=1

where each σpi is 1 or −1, depending on whether dfpi : Tpi → Tq is orientation-preserving or orientation-reversing,
respectively.

R
Proof. Choose some ω0 ∈ Ωnc (V ) with V ω0 = 1. By the definition of degree, we know that
Z XZ

deg(f ) = f ω0 = f ∗ ω0 .
V i Upi

And by our work last lecture, because f : Upi → V is a diffeomorphism, each term in this sum will be 1 if f is
orientation-preserving and −1 otherwise.

This then connects back to Theorem 115: if f : U → V is a proper C ∞ map and q is not in the image of f , then
q ∈ Cf must be a critical point. Additionally, f −1 (q) will be empty and the degree of f is zero.
We’ll now move on quickly to the topic of homotopy:

Definition 121
Let f0 , f1 : U → V be proper C ∞ maps. A homotopy between f0 and f1 is a C ∞ map F : U × [0, 1] → V such
that F (x, 0) = f0 (x) and F (x, 1) = f1 (x). If F is proper, then we call it a proper homotopy.

(In other words, a homotopy involves continuously deforming f0 into f1 with a smooth map.)

Theorem 122
If f0 and f1 are properly homotopic, then they have the same degree.

ft∗ ω
R R
Proof. Fix a compactly supported ω0 ∈ Ωnc (V ) such that V ω0 = 1. Then for each t ∈ [0, 1], deg(ft ) =
depends smoothly on t, but the degree is always an integer by Theorem 120. Thus deg(ft ) must be constant in t and
deg(f0 ) = deg(f1 ).

21 March 28, 2022


We’ll start by reviewing the degree theorem from before spring break, as well as the related definitions. If U and V
are open subsets of Rn and f : U → V is a proper C ∞ map, then the degree is an invariant of the map f satisfying
Z Z

f ω = deg(f ) ω
U V


whenever ω ∈ Ωnc (V ) (so that both f ω and ω are compactly supported and we can have both integrals make sense).
For another perspective on this degree, recall that a point q ∈ V is a regular value of f if dfp : Tp U → Tq V is bijective
for any p ∈ f −1 (q). (We previously mentioned Sard’s theorem, wihch states that the set of regular values forms an
open dense subset of U.) Then for any regular value q, we proved that f −1 (q) is a finite set {p1 , · · · , pk }, and then

36
we found the formula
k
X
deg(f ) = σ pi ,
i=1

where σpi is 1 if dfpi : Tpi → Tq is orientation-preserving and −1 if it is orientation-reversing (this makes sense because
the map is basically Rn → Rn ). So if f is not surjective, then deg(f ) = 0. Finally, we discussed the homotopy
invariance of degree, which tells us that two proper C ∞ maps f0 , f1 : U → V that are homotopic also have the same
degree (because the degree must be continuous and integer-valued).
Today, we’ll talk about some applications, starting with a familiar result:

Theorem 123 (Fundamental theorem of algebra)


Consider the monic polynomial p(z) = z n + an−1 z n−1 + · · · + a0 (where a0 , · · · , an−1 ∈ C). The equation p(z) = 0
has at least one complex solution z.

Proof. Notice that p is a C ∞ map C → C. Identify C with R2 (by sending x + i y to (x, y )) and think of p as a map
from R2 to R2 . Then notice that
n−1
X
|p(z)| ≥ |z|n + |ai ||z i |,
i=0

so |p(z)| → ∞ as |z| → ∞ because the leading term here, and thus p(z) is a proper map. Now define

p(z , t) = (1 − t)z n + tp(z)

(we can think of this as a homotopy between p0 (z) = z n and p(z)). Plugging in the expression for p(z), we know
that |p(z , t)| ≥ |z|n − t n−1 i
P
i=0 |ai ||z |, so again |p(z , t)| is unbounded as |z| → ∞ and we have a proper homotopy
between z n and p(z). Since the degree of p0 (z) is n, so is the degree of p(z). (This is left as an exercise for us: as
a hint, let φ be a C0∞ (R) function, and consider ω = φ(x 2 + y 2 )dxdy for calculation purposes. In polar coordinates,
this is φ(r 2 )r dr dθ, so we can check that p0∗ ω = n2 φ(r 2n )r 2n−1 dr dθ and R2 p0∗ ω = n R2 ω.) Thus the degree of p(z)
R R

is nonzero and p is surjective; in particular, there is at least one point where p(z) = 0.

Theorem 124 (Brouwer fixed point theorem)


Let B n be the unit ball {x ∈ Rn : |x| ≤ 1}, and let S n−1 = Bd(B n ) be the unit (n − 1)-sphere which bounds that
ball. If f : B n → B n is a C ∞ map, then there exists some x0 ∈ B n such that f (x0 ) = x0 (that is, we always have
a fixed point).

Proof. Suppose for the sake of contradiction that there is no fixed point. Consider the map γ : B n → S n−1 defined
in the following way: for each x ∈ B n , draw a ray originating at f (x) and pointing in the direction of x (because
f (x) ̸= x), and let γ(x) be the point on the boundary S n−1 that the ray intersects.
Notice that γ(x) = x if |x| = 1, so |γ(x) − x| = 0 for all x ∈ S n−1 . But because γ is a C ∞ map, it extends to
an open subset U of B n , and for every ε > 0, there is some δ > 0 such that B1+δ
n
⊇ U and |γ(x) − x| ≤ ε for all
1 ≤ |x| ≤ 1 + δ. If we now pick some ρ ∈ C0∞ (Int(B1+δ
n
) such that ρ is identically 1 for |x| ≥ 1, we can define the map
γ̃ : Rn → Rn such that 
ρ(x)γ(x) + (1 − ρ(x))x |x| ≤ 1 + δ,
γ̃(x) =
x |x| ≥ 1 + δ,
and consider the homotopy
g(x, t) = t γ̃(x) + (1 − t)x.

37
This is a proper homotopy, so γ̃(x) has the same degree as the identity map, which is 1. But γ̃(x) ≥ 1 − ε everywhere,
so for any |x0 | ≤ 1 − ε, γ̃ −1 is empty, which is a contradiction with the degree being nonzero.

22 March 30, 2022

Fact 125
As a reading assignment, we should read about the implicit function theorem in Appendix B of the textbook,
because it will have applications to what we’ll do in the coming weeks.

We’re currently transitioning from talking about differential forms on open subsets of Rn to differential forms on
manifolds – in particular, we’ll soon explain what a differentiable manifold is and start exploring relevant properties.
But today, we’ll first go back to some previous topics that we’ve discussed, starting with some multivariable calculus
results:

Proposition 126 (Inverse function theorem)


Let U and V be open subsets of Rn , and suppose f : U → V is a C ∞ map such that f (p) = q. If dfp : Tp U → Tq V
is bijective, then it maps a neighborhood Up of p diffeomorphically to a neighborhood Vq of q.

This result is actually the special case of two more general results:

Proposition 127 (Canonical submersion theorem)


Let k < n, and define the canonical submersion π : Rn → Rk to be the map (x1 , · · · , xn ) 7→ (x1 , · · · , xk ). Let U
be an open subset of Rn , and suppose f : U → Rk is a C ∞ map with f (p) = 0. If dfp : Tp U → T0 Rk is surjective,
then there exists a neighborhood W of 0 in Rk , a neighborhood Up of U, and a diffeomorphism φ : W → Up
sending 0 to p such that π = f ◦ φ.

Proposition 128 (Canonical immersion theorem)


Let k < n, and define the canonical immersion ι : Rk → Rn to be the map (x1 , · · · , xk ) 7→ (x1 , · · · , xn ). Let U
be an open neighborhood of 0 in Rk , and suppose f : U → Rn is a C ∞ map with f (0) = p. If dfp : T0 U → Tp Rn
is injective, then there exists a neighborhood W of 0 in Rn , a neighborhood Vp of p in Rn , and a diffeomorphism
φ : Vp → W sending p to 0 such that ι = φ ◦ f .

We can now turn to a study of manifolds, starting with a preliminary definition:

Definition 129
Let X be a subset of RM , Y be a subset of RN , and f : X → Y be a continuous map. Then f is a C ∞ map if for
every p ∈ X, there exists a neighborhood U of p in RM and a C ∞ map f˜ : U → RN such that f˜|U∩X = f |U∩X .

Definition 130
Let N ≥ n. A set X ⊂ RN is an n-dimensional manifold if for every p ∈ X, there exists an open set U of Rn , an
open neighborhood V of p in RN , and a diffeomorphism φ : U → V ∩ X. (We call φ a parameterization of X at
p.)

38
In other words, neighborhoods of p look locally like neighborhoods of Rn , and we’ll now try to define differential
forms on manifolds. In particular, if we let a = φ−1 (p), and we let Ta : Rn → Rn be the translation map x 7→ x + a,
we can compose φ with Ta and always assume that φ−1 (p) = 0. (In this case, we say that the parameterization is
centered at zero.) Manifolds may seem like an abstract concept, but it’s worthwhile to think about them because
they come up as solutions to systems of equations:

Proposition 131
Let U be an open subset of RN , and let f : U → Rk be a C ∞ mapping. (Recall that a point a is a regular value
if for all p ∈ f −1 (a), dfp : Tp U → Ta Rk is surjective. If a is a regular value, then X = f −1 (a) is an n-dimensional
manifold, where n = N − k.

In fact, many of the examples of manifolds we’ll be seeing from now on are of this form X = f −1 (a).

Proof. We can replace f with f − a, so that we can assume a = 0. If 0 is a regular value of f , that means that for all
p ∈ f −1 (0), dfp : Tp RN → T0 Rk is surjective. By the canonical submersion theorem, there are open neighborhoods U
of 0 and V of P in RN , as well as a diffeomorphism φ : U → V satisfying π = f ◦ φ (for the canonical submersion π).
But any x ∈ π −1 (0) is of the form (0, · · · , 0, xk+1 , · · · , xN ), so φ maps a neighborhood of 0 in Rn diffeomorphically
onto a neighborhood of p ∈ f −1 (0), and thus f −1 (0) is an n-dimensional manifold.

23 April 1, 2022
Last lecture, we introduced the concept of a manifold by explaining what it means to be a C ∞ map between subsets
X and Y of RM and RN , namely that for every point p ∈ X, there is a neighborhood U of p and a C ∞ map f˜ : U → RN
which agrees with f on U ∩ X. We say that f is a diffeomorphism if it is bijective and both f and f −1 is C ∞ , and
that allows us to define a subset X ⊆ RN to be an n-dimensional manifold if there is a neighborhood Vp around any
point p ∈ X such that we have a diffeomorphism φ (called a parameterization at p, centered at p if φ(0) = p) from
U → Vp ∩ X for some open set U of Rn (so that X looks locally like an open subset of Rn ).

Example 132
The n-sphere S n ⊆ Rn+1 , defined as

{x ∈ Rn+1 : x12 + · · · + xn+1


2
= 1}

is an n-dimensional manifold. To check this, let Hi± be the set of all points (x1 , · · · , xn+1 ) ∈ S n with xi > 0 or
xi < 0, respectively. We can verify that these are open sets which cover S n , and furthermore the map Hi± → Rn
sending (x1 , · · · , xn+1 ) to (x1 , · · · , xi−1 , xi+1 , · · · , xn+1 ) is a diffeomorphism. So any point in S n has a neighborhood
(one of the Hi± s that contains it) which looks like an open set of Rn , and we can choose our parameterization φ
to be the inverse of one of the Hi± s.

Example 133
Let Xi ⊆ RNi be C ∞ manifolds of dimension ni for 1 ≤ i ≤ k. Then the product manifold X1 × · · · × Xk is a
manifold of dimension n = n1 + · · · + nk , because any point in the manifold is of the form p = (p1 , · · · , pk ) where
pi ∈ RNi . Each of these pi s come along with a neighborhood Vpi , a subset Ui of Rni , and a map φi : Ui → Vpi ∩ Xi ,
and then we can check that the product map (φ1 , · · · , φk ) is a parameterization of the product manifold at p.

39
Last time, we mentioned that one important source of manifolds is the following: if U is an open subset of RN ,
f : U → Rk is a C ∞ map, and a ∈ Rk is a regular value, then X = f −1 (a) is an n-dimensional manifold. (As an
example of this, if we let f : Rn+1 → R be the map f (x1 , · · · , xn+1 ) = x12 + · · · + xn+1
2
, then Df (x) = 2(x1 , · · · , xn+1 ).
1 is then a regular value, and f −1 (1) is an n-dimensional manifold, namely S n . We’ll now see a more substantial
example:

Example 134
Let Mn×n be the set of all n × n real-valued matrices [aij ], and let Sn×n be the set of symmetric n × n real-valued
matrices (satisfying aij = aji for all i , j). We can verify that Mn×n ∼
2
= Rn and that there is a natural map
φ : Mn×n → Sn×n given by φ(A) = AT A. Then because I is a regular value of φ, φ−1 (In ) is a manifold, namely
the set of orthogonal n × n matrices (satisfying AT A = I).

Example 135
Let G(k, n) be the set of all symmetric matrices P ∈ Sn×n such that the rank of P is k (the image of the linear
mapping has dimension k) and P 2 = P . We can show that G(n, k) is a k(n − k)-dimensional manifold using the
methods from above, known as the Grassmannian (we can think of it as the set of k-dimensional linear subspaces
of Rn ).

(This will likely be on our next, and possibly last, homework assignment.) From here, we can now move on and
generalize the notion of a tangent space that we introduced originally on Rn :

Proposition 136
Let X ⊆ Rn be a k-dimensional manifold, let p be a point in X, and let W be a neighborhood of p in RN . Suppose
φ : U → V = W ∩ X is a parameterization of X centered at p (for some open set U of Rk ). If we think of φ as a
C ∞ map of U into W , then its derivative dφ0 : T0 U → Tp W is injective.

Proof. By definition, there is a C ∞ extension ψ : W → U of φ−1 , such that dψp ◦ dφ0 is the identity map, which can
only happen if dφ0 is injective.

Definition 137
With the notation in the result above, we let the tangent space of p at X be Tp X = Im(dφ0 ).

Proposition 138
The tangent space Tp X has an intrinsic definition not dependent on the particular choice of φ.

Proof. If φ1 : U1 → V1 and φ2 : U2 → V2 are two parameterizations centered at p, replacing V1 , V2 by V = V1 ∩ V2 and


replacing U1 , U2 by φ−1
i (V ) allows us to assume that φ1 : U1 → V and φ2 : U2 → V are both diffeomorphisms. Then
we have a diffeomorphism ψ = φ−1
2 ◦ φ1 . Since (dφ1 )0 and (dφ2 )0 are injective and (dψ)0 is bijective, the images of
(dφ1 )0 and (dφ2 )0 must be the same (since they have the same dimension, and if (dφ1 )0 = (dφ2 )0 ◦ (dψ)0 then the
image of (dφ1 )0 is contained in the image of (dφ2 )0 ).

We’ll finish by mentioning a functoriality property for Tp X: if X ⊆ Rm and Y ⊆ Rℓ are C ∞ manifolds, and
f : X → Y is a C ∞ map sending p to q, then dfp : Tp X → Tq Y is well-defined. But we’ll talk more about this next
time!

40
24 April 4, 2022
Last time, we introduced the concept of a tangent space on a manifold. Specifically, recall that a manifold X ⊆ RN
is a set that comes with a parameterization centered at p (for any p ∈ X), in which we take a neighborhood U of 0 in
Rn and a neighborhood W of p in RN , and we have a diffeomorphism φ : U → W ∩ X sending 0 to p. In particular, if
we think of φ actually as a C ∞ map from U to W (which contains W ∩ X), the map dφ0 : T0 U → Tp W (this maps
an n-dimensional space to an N-dimensional space) is injective. (The proof of this was that because φ was originally
a diffeomorphism from U to W ∩ X, it has a C ∞ inverse map ψ which extends to some open neighborhood W ′ of p.
Then using W ′ ∩ W instead of W , the chain rule on Euclidean space tells us dψp ◦ dφ0 = id, which implies that dφ0
is injective.) We then proved that this gives us an intrinsic definition Tp X = Im(dφ0 ).
We’ll now explore the functoriality concepts that we started discussing at the very end of last lecture. In particular,
let X be a submanifold of RN1 (that is, a manifold in RN1 ) and Y be a submanifold of RN2 . If we have a C ∞ map
X → Y sending p to q, then we can define the derivative of f at p, dfp : Tp X → Tq Y in two different ways:

Definition 139
Taking the notation above, suppose W is a neighborhood of p in RN1 , and g : W → RN2 is a C ∞ map such that
f = g restricted to X ∩ W . Then dfp is defined to be dgp restricted to Tp X.

This definition makes sense intuitively (we look at derivatives on Euclidean space instead of on the manifold), but
it depends on the choice of the function g. So we can try something else:

Definition 140
Again with the notation above, let φ1 : U1 → V1 and φ2 : U2 → V2 be parameterizations of X and Y centered
at p and q, respectively. Without loss of generality, assume V1 ⊆ f −1 (V2 ) (otherwise we can just make V1
smaller because φ1 is a diffeomorphism). Then dfp is the the map that completes the following diagram, where
ψ0 = φ−1
2 ◦ f ◦ φ1 :

dfp
Tp X Tq Y
(dφ1 )0 (dφ2 )0
dψ0
T0 U1 T0 U2

Unfortunately, this definition may look like it instead depends on the extension mappings φ1 and φ2 . So we’ll
reconcile that with the following diagram and the use of the 18.101 chain rule on f˜ ◦ φ1 = φ2 ◦ ψ0 :

d f˜p
Tp W T q RN 2
(dφ1 )0 (dφ2 )0
dψ0
T0 U1 T0 U2

In particular, the tangent space Tp X of an n-dimensional manifold X is n-dimensional. As an exercise, we can now
prove the chain rule on manifolds, which says that if f1 : X1 → X2 and f2 : X2 → X3 are C ∞ maps on manifolds and
f1 (p1 ) = p2 , then
(df2 ◦ f1 )p1 = (df2 )p2 ◦ (df1 )p1 .

An application of the chain rule is the following:

41
Proposition 141
Recall that for a map U → V between open sets of RN and Rk , and for a regular value q ∈ V of f , X = f −1 (q)
is an n-dimensional manifold, where N = N − k. Then Tp X is the kernel of the map dfp : Tp U → Tq V .

Proof. Let W be a neighborhood of p in RN , such that φ : U → W ∩ X is a parameterization of X centered at p.


Then f ◦ φ is the constant map U → X → q, so dfp ◦ dφ0 = 0. But because dφ0 maps T0 U bijectively to Tp X, Tp X
must be a subset of the kernel of dfp . On the other hand, Im(dfp ) is k-dimensional, so the kernel of dfp must be
(N − k = n)-dimensional. Since Tp X is n-dimensional, this means the kernel of dfp must coincide with it.

Next time, we’ll talk more about the canonical immersion and submersion theorems and generalize them to mani-
folds!

25 April 6, 2022
We’ve now spent a few lectures understanding the notation and relevant definitions behind manifolds X, maps X → Y
between them, and their tangent spaces Tp X. Last time, for a C ∞ map f : X → Y sending p to q, we defined the
derivative map dfp at p between tangent spaces Tp X → Tq Y in two ways, either extending f to an open subset of
RN1 (the Euclidean space that X lives in) and using the Euclidean space definition of dfp , or by making use of the
parameterizations U1 → W1 ∩ X and U2 → W2 ∩ Y that come along with the manifolds X and Y and defining dfp in
terms of the corresponding dψ0 map from the map on the Euclidean subsets ψ : U1 → U2 . It can be shown that these
definitions are both legitimate and coincide with each other (though as a sidenote, the second definition works even if
we have manifolds that aren’t subsets of Rn for some n), so that we have a valid definition of the derivative dfp and
thus of the map df . In particular, we also have a chain rule just like for ordinary Euclidean space.
Today, we’ll show that our results about differential forms on Euclidean space apply for manifolds too.

Theorem 142 (Inverse function theorem for manifolds)


Let X and Y be n-dimensional manifolds, and let f : X → Y be a C ∞ map sending p to q. If dfp : Tp X → Tq Y is
bijective, then f maps a neighborhood Up of p diffeomorphically onto a neighborhood Vq of q in Y .

We may recall from a few lectures ago that we also stated the canonical submersion and canonical immersion
theorem for open subsets of Rn , and there are analogous versions of those too. But because the canonical submersion
(for example) is defined from Rn → Rk (for k < n) as sending (x1 , · · · , xn ) to (x1 , · · · , xk ), we’re going to need to do
that “submersion” on the open sets U coming with the parameterizations:

Theorem 143 (Canonical submersion theorem for manifolds)


Let X and Y be manifolds such that f : X → Y is a C ∞ map sending p to q. If dfp : Tp X → Tq Y is surjective,
then there exist neighborhoods Up and Vq of p and q in X and Y , respectively, and parameterizations φ : U → Up
sending 0 to p and ψ : V → Vq sending 0 to q, such that f (Up ) ⊆ Vq and the following commutative diagram is
followed, with bottom arrow being the canonical submersion:
f
Up Vq
φ ψ
π
U V

42
In other words, every submersion looks locally like the canonical submersion. A similar property holds for the
canonical immersion ι : (x1 , · · · , xk 7→ (x1 , · · · , xk , 0, · · · , 0):

Theorem 144 (Canonical immersion theorem for manifolds)


Let X and Y be manifolds such that f : X → Y is a C ∞ map sending p to q. If dfp : Tp X → Tq Y is injective,
then there exist neighborhoods Up and Vq of p and q in X and Y , respectively, and parameterizations φ : U → Up
sending 0 to p and ψ : V → Vq sending 0 to q, such that f (Up ) ⊆ Vq and the following commutative diagram is
followed, with bottom arrow being the canonical immersion:
f
Up Vq
φ ψ
ι
U V

We may read the proofs on our own, but the idea is to choose coordinates so that the manifolds actually become
locally open subsets of Euclidean space and applying the canonical submersion and immersion theorems that we’ve
already seen. (And we should note that these are local results – everything here has to do with picking a particular
point p and q.)

26 April 8, 2022
We’ll start to talk about differential forms on manifolds today – just like on Euclidean space, if X is an n-dimensional
manifold, recall that we can define the tangent space Tp X at any point p ∈ X. We can then define the analogous
cotangent (dual) space Tp∗ X, and we can define Λk (Tp∗ X) in the same way using the standard linear algebra techniques.

Definition 145
Let X be a manifold. A k-form on X is a “function” ω which takes in a point p ∈ X and produces an element
ωp ∈ Λk (Tp∗ X).

Many of the properties of k-forms that we’ve already studied still hold – we’ll talk first about functoriality. If X1 and
X2 are manifolds, and f : X1 → X2 is a C ∞ map sending p1 to p2 , then we have a mapping df : Tp1 X1 → Tp2 X2 , and
accordingly we get a transpose (pullback) map (dfp )∗ : Λk (Tp∗2 X2 ) → Λk (Tp∗1 X1 ). Extending this pullback operation to
all points p, we find that given any k-form ω on X2 , we can define a k-form f ∗ ω2 on X1 via

(f ∗ ω)p1 = dfp∗1 ωp2 .

1 2 f f
We also have the usual “chain rule:” if X1 −
→ X2 −
→ X3 is a sequence of maps between manifolds, we then have

(f2 ◦ f1 )∗ ω = f1∗ f2∗ ω.

We may recall that we like to restrict our attention to C ∞ k-forms when we first studied everything over Euclidean
space, and that will be relevant here as well:

43
Definition 146
Let ω be a k-form on X. Then ω is C ∞ if for every p ∈ X, there exists a parameterization φ : U → V ∩ X (for
U, V open subsets of Euclidean space) such that φ∗ ω restricted to X is a C ∞ k-form on U. The space of C ∞
k-forms on X is denoted Ωk (X).

In particular, if ω is a C ∞ k-form, then f ∗ ω is also a C ∞ k-form. We do need to show that this definition is
independent of φ, but this is true because if we have diffeomorphisms φ1 : U1 → V ∩ X and φ2 : U2 → V ∩ X, then
ψ = φ−1 ∗ ∗ ∗ ∗
2 ◦ φ1 is a diffeomorphism as well, and φ1 ω = ψ φ2 ω2 (so φ1 ω is C

if and only if φ∗2 ω is).
Returning to operations on k-forms, just like over Euclidean space, if we have ω1 , ω2 ∈ Ωk (X), then their sum
ω1 + ω2 (defined pointwise) is also in Ωk (X). Also, we have a d operation just like over Euclidean space, defined in
the following way:

Definition 147
For any ω ∈ Ωk (X) and p ∈ X, let φ : U → V ∩ X be a parameterization of X at p (sending 0 to p). Then we
define
dω|X = (φ−1 )∗ d(φ∗ ω).

In other words, we use the definition of the d operation on Euclidean space, and this is a legitimate definition
because of the functoriality argument we just made (left as an exercise to us).

Proposition 148
Let f : X1 → X2 be a C ∞ map. Then for any ω ∈ Ωk (X2 ), we have f ∗ dω = df ∗ ω.

We showed this result for open subsets of Euclidean space, but they also hold for differentiable manifolds.

Proof. Let f (p1 ) = p2 for p1 ∈ X1 and p2 ∈ X2 . Suppose we have parameterizations φ1 : U1 → V1 and φ2 : U2 → V2


centered at p1 and p2 (from now on we’ll suppress the “intersection with X1 ” and so on). If we pick our neighborhoods
so that f −1 (V2 ) ⊃ V1 , then we have the following diagram:
f
V1 V2
φ1 φ2
ψ
U1 U2

By the chain rule and the commutativity of the diagram, we know that φ∗1 f ∗ dω = ψ ∗ φ∗2 dω, which leads us to
dψ ∗ φ∗2 ω = dφ∗1 f ∗ ω. Some more manipulation (using the definition of d and the pullback operation) gives us the
result.

From the definition of d, we can essentially read off the usual properties

d(ω1 + ω2 ) = dω1 + dω2 , d(ω1 ∧ ω2 ) = dω1 ∧ dω2 + (−1)k1 ω1 ∧ dω2 , d(dω) = 0

(where ω1 is a k1 -form), because of the versions that we proved earlier on in the class.

44
27 April 11, 2022
Last time, we discussed the definition of a (C ∞ ) differential k-form on an n-dimensional manifold (by looking at the
tangent space Tp X, the corresponding exterior algebra Λk (Tp∗ X), and then having ω assign to each p ∈ X an element
ωp ∈ Λk (Tp∗ ) in a way such that the pullback φ∗ ω of the parameterization at any p is a C ∞ k-form on the corresponding
domain, an open subset of Rn ). And as a reminder, in the Euclidean case, a k-form ω = φI dxI is said to be C ∞ if
P

each φI is a C ∞ function on U.
We also described the pullback operation, which gives us a differential form f ∗ ω given a differential form ω and a C ∞
φI dxI , then f ∗ ω =
P P ∗
map. More explicitly, recall that on Euclidean space, we did this by saying that if ω = f φI dfI ,
where f ∗ φI is still a C ∞ function and dfI = f ∗ dxI . Parameterizations then let us generalize that definition (as we
P P
mentioned last time). Similarly, we saw that the d operation, which takes ω = fI dxI to dω = dfI ∧ dxI , can be
extended to manifolds as well. And while these definitions appear to depend on the diffeomorphism φ that comes with
the manifold X, we can verify that the objects we define are in fact intrinsic. Furthermore, we verified that the usual
functorial properties of these operations that we proved over Euclidean space still hold more generally.
We’ll now talk about vector fields on manifolds, generalizing our discussion on Euclidean space. the pullback

Definition 149
Let X be a manifold. A vector field v on X assigns to each p ∈ X an element v (p) ∈ Tp X, and a vector field is
C ∞ if its pullback is C ∞ .

Definition 150
Let X be a manifold, ω ∈ Ωk (X) be a differential k-form, and v be a C ∞ vector field on X. Like over Euclidean
space, let the interior product ι(v )ω be the differential (k − 1)-form given by (ι(v )ω)p = ι(vp )ωp .

The only thing we must check is that this indeed gives us a C ∞ k-form:

Proposition 151
If ω, v are C ∞ , then so is ι(v )ω.

Proof. Let φ : U → Vp be a parameterization of our manifold X at p. Define v # to be the pullback φ∗ (v ) restricted


to X, and similarly let ω # = φ∗ ω|V . Then we can check that φ∗ (ι(v )ω|V ) = ι(v ♯ )ω ♯ .

With this verification, we are now permitted to extend the Lie differentiation to differential forms on manifolds:

Definition 152
Let v be a C ∞ vector field on X, and let ω ∈ Ωk (X). Then the Lie differentiation operation is defined (just like
on Euclidean space) as
Lv ω = dι(v )ω + ι(v )dω

Proposition 153
Just like over Euclidean space, we have

dLv ω = Lv dω, Lv (ω1 ∧ ω2 ) = Lv ω1 ∧ ω2 + ω1 ∧ Lv ω2 .

45
(The same proofs carry over verbatim.) We also have an alternative definition of the Lie derivative.

Proposition 154
Suppose that a vector field v is complete (recall that this means there is a one-parameter family of diffeomorphisms
ft such that for each p ∈ X, ft (p) is an integral curve of v with f0 (p) = p). Then

d
Lv ω = (f ∗ ω) .
dt t t=0

This proof is left as an exercise to us, but the idea is that for any φ ∈ C ∞ (X), we have

ft∗ φ(p) = φ(γt (p))

for an integral curve γt (p) with γ0 (p) = p. Then because ft∗ dω = ft∗ ω and ft∗ (ω1 ∧ ω2 ) = ft∗ ω ∧ f2∗ ω2 (discussed
previously in the class), we can differentiate the first boxed equation at t = 0 to conclude the result for differential
0-forms (that is, functions), differentiate the second boxed equation at t = 0 to conclude the result for dω given the
result for ω, and differentiate the third boxed equation at t = 0 to conclude the result for wedge products. Since every
k-form is a sum of products of 1-forms, this proves the result.

28 April 13, 2022


We’ll be talking about integration of differential forms on manifolds in the next section of this course. Our first result
is one about “partitions of unity:”

Theorem 155
Let X ⊆ RN be an n-dimensional manifold, and let U = {Uα : α ∈ I} be an open covering of X (where I is some
index set). Then there exists a family of functions ρi ∈ C0∞ (X), such that
• each ρi is nonnegative on X,

• for every compact subset C ⊆ X, there is some M > 0 such that suppφi ∩ C = ∅for all i > M,
P
• ρi = 1 on all of X,

• for every i ∈ α, there is some α such that supp(ρi ) ⊆ Uα .

Proof. For every p ∈ X, there is some Uα with p ∈ Uα , so that we can choose an open set Op in RN containing p
such that Op ∩ X ⊆ Uα . If we define O = p Op , and we let ρ̃i ∈ C0∞ (O) be a partition of unity subordinate to the
S

covering by the Op s (here we’re using the partition of unity result on Euclidean space), we can let ρi be ρ̃i restricted
to X. Then the properties in the theorem statement can be readily verified.

We’ll also define the concept of an “orientation” on a manifold, just like over Euclidean space. Recall that for any
one-dimensional vector space V , an orientation is a labeling of the connected components of V −{0} as V+ and V− , and
thus for any n-dimensional vector space V , an orientation is a labeling of the connected components of Λn (V ) − {0}
as Λn (V )+ and Λn (V )− .

46
Definition 156
Let X be an n-dimensional manifold. An orientation of X is a function which assigns to each p ∈ X an orientation
of the tangent space Tp X. The orientation is smooth if for every p ∈ X, there is an open neighborhood U of p
in X and a C ∞ n-form ω ∈ Ωn (U), such that ωq ∈ Λn (Tp∗ )+ for all q ∈ U.

Example 157
Let U be an open subset of Rn . The standard orientation of U is the canonical orientation defined by the
Euclidean volume element dx1 ∧ · · · ∧ dxn .

If we let X and Y be oriented n-dimensional manifolds and let f : X → Y be a diffeomorphism, we can again make
a functoriality argument: we again call f orientation-preserving if for every p ∈ X (and corresponding q = f (p)), the
map (dfp )∗ sends Λn (Tq∗ )+ to Λn (Tp∗ )+ (and otherwise orientation-reversing).

Definition 158
Let X be an oriented manifold, U be an open subset of Rn , V an open subset of X, and φ : U → V a diffeomorphism.
φ is an oriented parameterization if φ is orientation-preserving.

One important note is that if U is connected but φ is not orientation-preserving (that is, it is orientation-reserving),
we can replace the open set U with U ′ = {(x1 , · · · , xn ) : (x1 , · · · , xn−1 , −xn ) ∈ U}. Then replacing φ with φ′ , defined
via φ′ (x1 , · · · , xn ) = φ(x1 , · · · , xn−1 , −xn ), gives us an orientation-preserving map φ′ .
R
Our goal is now to define the integral X ω for any compactly supported ω ∈ Ωnc (X), and we’ll be doing so under
the assumption that X is oriented. We’ll first do the case where ω ∈ Ωnc (V ), where V is a parameterizable open set
of X. In that case, we choose an oriented parameterization φ : U → V , we can then define
Z Z Z
ω= ω= φ∗ ω,
X V U

where we can use the definition of differentiation of forms over Euclidean space by multivariable calculus.

Lemma 159
R
The definition of V ω is intrinsic (not dependent on the choice of parameterization φ).

Proof. Suppose φ1 : U1 → V and φ2 : U2 → V be two oriented parameterizations. Because parameterizations are


diffeomorphisms, we get a diffeomorphism g = φ−1
2 ◦ φ1 , and then
Z Z Z
φ∗1 ω = (φ2 ◦ g)∗ ω = g ∗ φ∗2 ω.
U1 U1 U

But because g ∗ is orientation-preserving (since φ1 , φ2 are orientation-preserving), this right-hand side is equal to

R
U2 φ2 ω. So the integral is indeed independent of our choice of parameterization.

We can now define integration of forms in the general case by using Theorem 155:

47
Definition 160
Let {Vα : α ∈ U} be an open covering of X by parameterizable open sets, and let ρi ∈ C ∞ (X) be a partition of
unity subordinate to the Vα . Then for any compactly supported n-form ω ∈ Ωnc (X), we may define
Z XZ
ω= ρi ω,
X X

where the right-hand side is well-defined because each ρi ω is supported within a parameterizable open set.

By the last point of Theorem 155, because ω is compactly supported, this sum eventually vanishes, and thus there
are no convergence issues. And furthermore, we can check that this definition is independent of our choice of partition
of unity – to see this, suppose U ′ = {Uβ′ } is a different open covering of X by parameterizable open sets and ρ′i is a
corresponding partition of unity. Then by swapping the order of summation,
XZ XZ XZ XZ
ρi ω = ρi ρj ω = ρj ρi ω = ρj ω,
i X i,j X j,i X j X

showing that the definitions agree. And we can show functoriality as well: if X1 and X2 are n-dimensional manifolds,
and f : X1 → X2 is an orientation-preserving diffeomorphism, then for any ω ∈ Ωnc (X2 ), we have X1 f ∗ ω = X2 ω.
R R

29 April 15, 2022


Last lecture, we started discussing integration of forms on a manifold X. Specifically, we said that if V is a parame-
terizable open set in X, ω ∈ Ωc (V ) is compactly supported, and φ : U → V is an oriented parameterization of V , then
we can define Z Z
ω= φ∗ ω
X U
via the definition of integration of forms on Euclidean space and the definition of the pullback map. More generally,
for an arbitrary form ω, we can choose a covering of X by parameterizable open sets and then forming a partition of
unity subordinate to that covering, which then allows us to define
Z XZ
ω= ρi ω,
X i X

in which we justified why this sum is well-defined last lecture. Finally, we mentioned that the functoriality properties
of integration still hold over oriented manifolds: for any orientation-preserving diffeomorphism f : X1 → X2 between
n-dimensional manifolds, we have X1 f ∗ ω = X2 ω for any compactly supported n-form ω ∈ Ωnc (X2 ).
R R

Today, we’ll look some fundamental results in multivariable calculus, namely Stokes’ theorem and the divergence
theorem, and generalize them to manifold versions. We’ll start with a “model case” of Stokes’ theorem on Euclidean
space:

Proposition 161
Let Hn = {(x1 , · · · , xn ) ⊆ Rn : x1 < 0} be a half-space of Rn , then Rn−1 = {(x1 , · · · , xn ) : x1 = 0} (identified
with the inclusion map ιRn−1 : (x1 , · · · , xn−1 ) → (x1 , · · · , xn−1 , 0) is its boundary bd(Hn ). Then for a compactly
supported (n − 1)-form ω ∈ Ωn−1
c (Rn ), we have
Z Z
dω = ι∗Rn−1 ω
Hn Rn−1

48
∂f
R
This proof is left as an exercise for us – the idea is that for any i > 1, Hn ∂xi dx1 · · · dxn = 0 by integration by
parts, while for i = 1 we have
Z
∂f
dx1 · · · dxn = f (0, x2 , · · · , xn )dx2 · · · dxn .
Hn ∂x1

We’ll now generalize this as promised:

Definition 162
Let X be an n-dimensional manifold. An open subset D of X is a smooth domain if for every p ∈ bd(D), there
exists a parameterization φ : U → V centered at p such that φ maps U ∩ Hn diffeomorphically onto V ∩ D. (This
is called a D-adapted parameterization.)

In other words, every point on the boundary of D looks locally like a point on the boundary of the half-space Hn .
We can notice a few important facts:

• If D is a smooth domain and Z = bd(D), then because Bd(Hn ) is isomorphic to Hn−1 , from the parameterization
φ we have a diffeomorphism U ∩ Rn−1 onto V ∩ Z. In other words, Z is a submanifold of X, and φ : U ∩ Rn−1 →
V ∩ Z is a parameterization of Z at p.

• Assuming now that X is an oriented manifold, we can assume that φ is an oriented parameterization (otherwise
do the trick from last time, where we replace U with U ′ = {(x1 , · · · , xn ) : (x1 , · · · , −xn ) ∈ U} and use the map
φ′ (x1 , · · · , xn ) = φ(x1 , · · · , xn−1 , −xn ) in place of φ). We then find that the map φ : U ∩ Rn−1 → V ∩ Z is
orientation-preserving. This is in fact an intrisic property: if φ1 : U1 → V and φ2 : U2 → V are two oriented
D-adapted parameterizations, and f = φ−1 n
2 ◦ φ1 , then we get an orientation-preserving map of U1 ∩ H onto
U2 ∩ Hn , so if φ1 : U1 ∩ Hn is orientation-preserving then so is φ2 : U2 ∩ Hn .

Theorem 163 (Stokes’ theorem on manifolds)


Let X be an n-dimensional manifold, and let ω ∈ Ωcn−1 (X). Let D be a smooth domain, and let Z be the boundary
of D. Then Z Z
ι∗Z ω = dω.
Z D

Proof. Recalling the definition of integration of forms on manifolds, by a partition of unity we may assume that
ω ∈ Ωn−1
c (W ), where W is a parameterizable open set satisfying one of the following three properties:
• W ∩ D = ∅,

• W ⊆ Int(D),

• there exists a D-adapted parameterization φ : U → W , in particular meaning that φ(U ∩ Hn−1 ) → W ∩ Z.


In the first case, both sides of Stokes’ theorem are zero. In the second case, if φ : U → W is an oriented
parameterization, then because d commutes with the pullback map,
Z Z Z Z
dω = ω= φ∗ dω = dφ∗ ω = 0.
D Int(D) U U

Finally, in the third case, we have Z Z Z


dω = φ∗ dω = dφ∗ ω,
D Hn Hn

49
and on the other hand we have Z Z
ι∗Z ω = ι∗Rn−1 φ∗ ω.
Z Rn−1

So now we can just apply the Euclidean version of Stokes’ theorem to get the desired result.

We also get a simple but important corollary:

Theorem 164 (Divergence theorem)


Let X be an n-dimensional manifold, and let v be a C ∞ vector field on X. If ω ∈ Ωnc (X), then ι(v )ω ∈ Ωn−1
c (X)
and Lv ω ∈ Ωnc (X) are well-defined. Then for any smooth domain D, we have
Z Z
Lv ω = ι(v )ω.
D Z

R R
Proof. Recall the definition of Lv ω = ι(v )dω + dι(v )ω. Applying Stokes’ theorem gives us D dι(v )ω = Z ι(v )ω,
which implies the result.

In particular, formulating this result over Euclidean space gives us the ordinary divergence theorem.

30 April 20, 2022


Today, we’ll extend the discussion of degree theory to oriented manifolds, starting with an important generalization:

Theorem 165 (Poincaré lemma for manifolds)


Let X be an oriented, connected n-dimensional manifold, and let ω ∈ Ωnc (X). Then the following are equivalent:
R
1. X ω = 0,

2. ω = dµ for some µ ∈ Ωn−1


c (X).

The above result can be restated in an alternate formulation: if ω1 , ω2 are both compactly supported n-forms on
X, say that ω1 ∼ ω2 if ω1 − ω2 ∈ dΩn−1
c (X). Then Theorem 165 is equivalent to saying that ω1 ∼ ω2 if and only if
R R
X ω1 = X ω2 .
Notice that Stokes’ theorem on manifolds already tells us that the second property implies the first, so we just
R
need to prove that ifX ω = 0, then we can write ω = dµ. Also, we can fix some parameterizable open set U0 ⊆ X,
and let ω0 ∈ Ωnc (U0 ) be an n-form compactly supported on U0 . Then we may equivalently prove the following:

Theorem 166
Let X be an oriented manifold, let ω ∈ Ωnc (X), and let ω0 ∈ Ωnc (U0 ) be compactly supported on a parameterizable
R
open set U0 of X. If ω ∈ Ωnc (X) and X ω = c, then ω ∼ cω0 .

Proof. By the definition of integration of forms, we can (using a partition of unity) assume that ω ∈ Ωnc (U) for some
R
parameterizable open set. Additionally, if c = 0, then U ω = 0 implies that ω ∈ dΩn−1 c (U) (here we’re using the
Poincaré lemma on Rn by applying the pullback map), meaning that dω ∼ 0. Otherwise, if c ̸= 0, replacing ω
with ωc allows us to assume without loss of generality that U ω = 1.
R

The main step of the proof from here (which should look familiar) is to choose a sequence of parameterizable
open sets U1 , · · · , UN , so that UN = U and Ui ∩ Ui−1 = ∅ for all i (including i = 1). Then for each i , pick an

50
R
ωi ∈ Ωnc (Ui−1 ∩ Ui ) such that ωi = 1. Since Ui−1 is always a parameterizable open set, ωi−1 ∼ ωi for all i (since
they have the same integral), and additionally ω ∼ ωN . Thus ω0 ∼ ω1 ∼ ω2 ∼ · · · ∼ ωN ∼ ω as desired.

Definition 167
Let X and Y be oriented connected n-dimensional manifolds, and let f : X → Y be a proper C ∞ map. Define the
degree of f to be the topological invariant such that for every ω ∈ Ωnc (Y ), we have X f ∗ ω = deg(f ) Y ω.
R R

R
To check why such a topological invariant exists, suppose we choose some ω0 ∈ Ωnc (Y ) such that Y ω0 = 1,
R ∗ R
and define deg(f ) to be X f ω0 . Now suppose that Y ω = c. Then by the variant of the Poincaré lemma above,
ω ∼ cω0 , meaning that ω − cω0 = dµ for some µ ∈ Ωcn−1 (Y ). Applying the pullback map, we find that
Z Z
f ∗ ω = c f ∗ ω0 ,

f ∗ ω = deg f
R R
which leads us to the assertion ω.

Proposition 168
Let X, Y, Z be oriented n-dimensional manifolds, and let f : X → Y and g : Y → Z be proper maps. Then
deg(g ◦ f ) = deg(f ) deg(g).

Proof. We’ve seen this result before for Euclidean space, and the same proof works: for any ω ∈ Ωnc (Z), we have
(g ◦ f )∗ ω = f ∗ g ∗ ω, so deg(g ◦ f ) Z ω = X (g ◦ f )∗ ω = X f ∗ g ∗ ω = deg(f ) Y g ∗ ω = deg(f ) deg(g) Z ω.
R R R R R

We’ll now explain the stack of records theorem, which should again look familiar. Let f : X → Y be a proper
map between oriented manifolds, and let q ∈ Y be a regular value of f , meaning that the preimage f −1 (q) is a set
{p1 , · · · , pN } and there exist disjoint neighborhoods Ui of pi and V of q such that each fi : Ui → V is a diffeomorphism
and f −1 (V ) = N
S n
R
i=1 Ui . Choose ω ∈ Ωc (V ) with V ω = 1. Then
Z XZ

f ω= f a sti ω,
X i Ui


R
and because each fi : Ui → V is a diffeomorphism, Ui fi ω will be either 1 or −1 depending on whether f is orientation-
preserving or orientation-reversing. Thus we again get a formula for the degree:

Theorem 169
Let f : X → Y be a proper map between oriented manifolds, and let q be a regular value of f with preimage
P
{p1 , · · · , pN }. Then deg(f ) = i σpi , where σpi is 1 (resp. −1) if the map fi above is orientation-preserving (resp.
orientation-reversing).

Furthermore, we have the same homotopy invariance that we previously discussed (which explains that we indeed
have a topological invariant):

Proposition 170
Let f0 , f1 : X → Y be two C ∞ proper maps, and let F : X × [0, 1] → Y be a proper homotopy between f0 and f1
(meaning that F (x, 0) = f0 (x), F (x, 1) = f1 (x) and we have a proper C ∞ map). Then if the map ft is defined
via ft (x) = F (x, t) for all x, then ft is proper, and for a compact set A ⊆ Y , there exists a compact set B ⊆ X
with ft−1 (A) ⊂ B.

51
(The relevant ideas here are the following: let π : X × [0, 1] → X be the projection (x, t) 7→ x. Then there exists
a compact set B in X such that ft−1 (A) ⊆ B, because the ft s are proper maps.) And because the result above tells us
that the degree of any proper map ft is always an integer, and the degree is continuous, we have the following result:

Theorem 171
If f0 and f1 are properly homotopic, then deg(f0 ) = deg(f1 ).

31 April 22, 2022


We reviewed and generalized the definition of degree last lecture – for a proper C ∞ map f : X → Y between oriented
connected n-dimensional manifolds, we found that there was an invariant deg(f ) such that X f ∗ ω = deg(f ) Y ω for
R R

all compactly supported n-forms ω ∈ Ωnc (Y ). We also established some familiar properties of this degree, namely that
deg(g ◦ f ) = deg(f ) deg(g), that the degree is homotopy invariant, and that we can calculate the degree of a map
f via the “stack of records theorem.” The latter result basically states that if q is a regular value of the map f with
preimage {p1 , · · · , pN }, then we can associate to it diffeomorphisms fi : Ui → V between (disjoint) neighborhoods of
pi and q for each i (with the condition that f −1 (V ) = N
S
i=1 Ui for the purposes of the proof). We then have
X
deg(f ) = σ pi ,
i

where each σpi is 1 or −1 depending on whether fi is orientation-preserving or orientation-reversing (and recall that
this proof came from the fact that for any ω ∈ Ωnc (V ) with V ω = 1, we have deg(f ) = X f ∗ ω = i Ui f ∗ ω). We
R R P R

can apply these discussions of degree to the extendability problem (left as an exercise to us):

Proposition 172
Suppose M is an oriented n-dimensional manifold, D ⊆ M be a smooth domain with compact closure, and
X = Bd(D). Then if f : X → X is a proper map, then f cannot be extended to a proper map on D, because any
extendable map has deg(f ) = 0.

We’ll discuss another application of degree theory today, index theory for vector fields.

Definition 173
Let D be a smooth domain in Rn such that D is connected with compact closure. Let X = Bd(D), and let v be
a C ∞ vector field on Rn such that v (p) ̸= 0 for all p ∈ X. From this map v , we get a map X → S n−1 sending
v (p)
p to |v (p)| . (Notice that X and S n−1 are both compact (n − 1)-dimensional manifolds). The index of v over D,
denoted ind(v , D), is then the degree of the map g.

Theorem 174
Let D1 be a smooth domain in Rn whose closure is contained in D. Then if v has no zeros in D \ D1 , then
ind(v , D) = ind(v , D1 ).

Proof. Let W = D \ D1 , and notice that the map fv : ∂D → S n−1 extends to a map F : W → S n−1 (given by
v (p)
p 7→ |v (p)| . We may write ∂W = X ∪ X1 , where X = Bd(D) and X1 = Bd(D1 ) with its orientation reversed (it’s good

52
R
to draw a diagram for this). Now choose a differential (n − 1) form ω ∈ Ωn−1 (S n−1 ) such that S n−1 ω = 1. Define
the maps f0 = fv and f1 = (f1 )v to be the restrictions of F to the two boundaries X and X1 . By Stokes’ theorem (in
the last step below), and the fact that dω = 0 (it’s an n-form on an (n − 1)-dimensional space), we find that
Z Z Z Z
∗ ∗
0= F dω = dF ω = fω− f ω = deg(f ) − deg(f1 ).
W W X X1

Thus the degrees of the maps f and f1 are equal, meaning that the index of v over D and over D1 are identical.

Using this result, if we now assume that v only has a finite set of zeros {p1 , · · · , pk } over its domain D, then we
may let B(p, ε) denote an ε-ball around p, and we can let D0 = D \ ki=1 B(pi , ε). Our map g above will then extend
S
0 v (p) 0
to a map G : D → S n−1 (on which p still maps to |v (p)| , and the boundary of D is the union X and the boundaries
of the balls B(pi , ε). Thus, putting everything together, Theorem 174 gives us the following index theorem:
X X
deg(g) = ind(v , B(pi , ε)) =⇒ ind(v , D) = ind(v , pi ) .
i i

32 April 25, 2022


We’ll spend the last few lectures of the class on the last chapter of the textbook, discussing de Rham theory. We’ll
start today with some basic definitions – recall that for a manifold X, Ωk (X) is the set of C ∞ k-forms on X, and
Ωkc (X) is the set of compactly supported C ∞ k-forms on X.

Definition 175
Let X be an n-dimensional manifold. Define the vector spaces

Z k (X) = {ω ∈ Ωk (X) : dω = 0}, B k (X) = {ω ∈ Ωk (X) : µ ∈ dΩk−1 (X).

Analogously, we may define Zck (X) = {ω ∈ Ωkc (X) : dω = 0} and Bck (X) = {ω ∈ Ωkc (X) : µ ∈ dΩk−1
c (X)}.

In particular, we can note that B k (X) ⊆ Z k (X) (because of exactness, coming from the fact that d 2 = 0), so the
next definition also makes sense:

Definition 176
The kth de Rham cohomology group of X is given by H k (X) = Z k (X)/B k (X), and analogously we also define
Hck (X) = Zck (X)/Bck (X).

Our goal will be to understand these objects and to understand methods for computing the cohomology groups.
We’ll use the notation that for any ω ∈ Z k (X), [ω] is the image of ω in H k (X) (and similarly for the same statement
for compactly supported forms).

Proposition 177
We have the following facts:
1. If X is connected, then H 0 (X) = R.

2. If X is n-dimensional, then H k (X) = Hck (X) = 0 if k > n or k < 0.

3. If X is connected and oriented, then Zcn (Ω) = Ωnc (X).

53
Proof. For (1), because a 0-form on X is a C ∞ function and d of a function on a connected X is only zero if it is
constant, H 0 (X) must consist of the set of constant functions. (2) follows from the fact that we only have nontrivial
k-forms for 0 ≤ k ≤ n. Finally, (3) holds because dω of any n-form is zero, so all n-forms work.
R R
If we now think about the integration operation : Ωnc (X) → R, sending a form ω to X ω, then recall that we’ve
previously proved that Z
ω∈ dΩcn−1 (X) ⇐⇒ ω = 0.
X

Quotienting out by Bcn (X), we thus get a bijective map between Hcn (X) and R in this case (by associating to each
form its integral on X).

Proposition 178
Let U ⊆ Rn be an open rectangle. Then for all k > 0, H k (U) = {0}.

(This was on our homework, in which we proved that every form ω ∈ Z k (U) is exact.)

Proposition 179
Let U ⊆ Rn be an open rectangle. Then for all k < n, Hck (U) = {0}.

(This was actually also on our homework – we proved that if ω is closed, then ω = dµ for some µ ∈ Ωk−1
c (U), so
in fact Z k (U) = B k (U).) We’ll next discuss the Poincaré lemma for manifolds: let X be an n-dimensional manifold,
and at a point p ∈ X let φ : U → V be a parameterization sending 0 to p. We may assume (by restriction of φ) that
U is an open rectangle. Then if α ∈ Z k (X) for some k > 0, and α0 is α restricted to V , then

dα = 0 =⇒ dα0 = 0 =⇒ dφ∗ α0 = 0 =⇒ α0 = (φ−1 )∗ dβ = d(φ−1 ∗


1 ) β,

and thus α restricted to V is an element of B k (X), because it’s d of a (k − 1)-form. Thus H k (V ) = 0 for all k > 0,
and thus every closed k-form, restricted to an open set, is exact. Now that we’ve discussed some local results,
we’ll turn to global ones:

Theorem 180
For the sphere S n , we have H 0 (S n ) = H n (S n ) = R and H k (S n ) = 0 for 0 < k < n.

Proof. Because S n is compact and connected, we’ve already proved that H 0 (S n ) = H n (S n ) = R from our earlier
remarks. It now suffices to consider the situation for 0 < k < n. Let p0 be the point (0, 0, · · · , 0, 1) ∈ S n , and let
γ : S n − {p0 } → Rn be the stereographic projection, meaning that for any point q ∈ S n , we draw a line between p0
and q and consider where it intersects the hyperplane Rn × {0} in Rn+1 . This is a bijective map (since for any point
in Rn we can draw a line between it and p0 , and it will intersect the sphere at one other point q besides p0 ).
Now for any α ∈ Z k (S n ), there exists a neighborhood V of p and some β ∈ Ωk−1 (V ) such that α = dβ on V (by
our discussion above for parameterizable open sets V ). Letting ρ ∈ C0∞ (V ) be a “bump function” such that ρ = 1 on
a neighborhood of p, we can then replace α with α̃ = α − dρβ, meaning that

α̃ ∈ Ωkc (S n − {p0 }) = Ωk0 (Rn ).

In particular, α̃ = d β̃ for some β̃ ∈ Ωk−1


c (S n − {p0 }). If we then write β̃˜ = ρβ + β̃, we find that d β̃˜ = α, so
α ∈ B k (S n ). Therefore we mod out Z k (S n ) by itself and thus H k (X) = 0.

54
We’ll close by thinking about functoriality of these groups: if X and Y are C ∞ manifolds, and f : X → Y is a C ∞
map, then we have a corresponding map f ∗ : Ωk (Y ) → Ωk (X). Since f ∗ dω = df ∗ ω< we find that we also get maps
f ∗ : Z k (Y ) → Z k (X) and B k (Y ) → B k (X). We thus get an induced map

f ♯ : H k (Y ) → H k (X)

between the kth de Rham cohomology groups of X and Y : specifically, the map sends f ♯ [ω] to [f ∗ ω]. In addition, if
f : X → Y is a proper map, then preimages of compact sets are compact, so ω ∈ Ωkc (Y ) =⇒ f ∗ ω ∈ Ωkc (X). Thus,
even within the compactly supported forms, we also get a map f ♯ : Hck (Y ) → Hck (X), again sending [ω] to [f ∗ ω].
We’ll be making use of these facts in the coming lectures, and next time we’ll show that these cohomology groups are
finite-dimensional as long as X is comacp.

33 April 27, 2022


Last lecture, we started introducing de Rham theory: recall that the de Rham cohomology groups of X are defined in
terms of the d map d : Ωk (X) → Ωk+1 (X) on k-forms of X. Specifically, if Z k (X) is the kernel of the d map, and
B k+1 (X) is the image, then we define H k (X) = Z k (X)/B k (X) (and also similarly define a version of this for compactly
supported forms). The notation we’ll use, going forward, is that [ω] denotes the image of a form ω ∈ Z k (X) (resp.
Zck (X)) in H k (X) (resp Hck (X)).
We’ll start to understand some techniques that can be used for computing these groups, starting with the Mayer–
Vietoris theorem.

Definition 181
Let C 1 , C 2 , · · · be vector spaces, and let di : C i → C i+1 be maps between those sequences (we will often just
d d
denote these maps as d to simplify notation). We call the sequence C 0 −→
0
C 1 −→
1
C 2 · · · is called a complex if
di+1 ◦ di = 0 for all i .

Example 182
Because d 2 = 0 for the d map we’ve been discussing, we have the de Rham complex
d d
0 → Ω0 (X) −
→ Ω1 (X) −
→ Ω2 (X) · · · ,
d d
as well as a similar complex for compactly supported forms 0 → Ω0c (X) −
→ Ω1c (X) −
→ Ω2c (X) · · · .

We may also discuss morphisms between complexes:

Definition 183
α
If C1k , d and C2k , d form two different complexes, then a morphism between the complexes is a map α : C1k −
→ C2k
such that dα = αd.

In other words, we can imagine writing out our complexes in two rows and having α be maps from the top to the
bottom. Then the “squares” in the diagram will commute:

55
d d d
C10 C11 C12 ···
α α α

d d d
C20 C21 C22 ···

We’ll now explain the general abstract definition of a cohomology group:

Definition 184
If C, d is a complex, then the kth cohomology group H k (C) of the complex C is given by defining Z i = {c ∈
C i : dc = 0} and Bi+1 = dC i and setting H k (C) = Z i /B i .

In particular, a morphism of complexes will induce a map α : H k (C1 ) → H k (C2 ) (we can check that this is indeed
well-defined even though we have quotient groups because d and α commute).

Definition 185
Let V0 , V1 , · · · be a sequence of vector spaces, and let αi : Vi → Vi+1 be maps between those spaces. The sequence
α0 1 α
V0 −→ V1 −→ V2 → · · · is exact if ker αi = im αi−1 for all i . A short exact sequence is an exact sequence of the
1 α 2 α
form {0} → V1 −→ V2 −→ V3 → 0 (meaning that α1 is injective, α2 is surjective, and ker(α2 ) = im (α1 )).

With all of this notation, we’re now ready to discuss the main result of today’s lecture. Here’s the setup: let
d d d i j
→ Cr1 −
{0} − → · · · be a complex for r = 1, 2, 3 (meaning that d 2 = 0), and suppose that 0 → C1k →
→ Cr2 − − C2k →
− C3k → 0
is a short exact sequence for each k, meaning that the image of each i map is the kernel of the j map. Then we may
draw the following commutative diagram (which extends further both up and down):

i j
0 C1k+1 C2k+1 C3k+1 0
d d d

i j
0 C1k C2k C3k 0
d d d

i j
0 C1k−1 C2k−1 C3k−1 0

Then defining Zrk = {c ∈ Crk : dc = 0} and Brk = dCrk−1 for each k, r , we find that i (B1k ) ⊆ B2k and i (Z1k ) ⊆ Z1k
(because i and d commute), giving us an induced map i♯ : H k (C1 ) → H k (C2 ). We similarly get an induced map
i♯ j♯
j♯ : H k (C2 ) → H k (C3 ). We may check that ker j♯ = im i♯ , meaning that H k (C1 ) −
→ H k (C2 ) −
→ H k (C3 ) is exact. It
turns out to not be short exact (we can’t append a 0 to the beginning and end of it and still have exactness), but we
instead have that the following diagram commutes:

0 0
d d

i j
C1k+1 C2k+1 C3k+1
d d

i
C2k C3k

We may check (by exploring the implications of this diagram) that we then get a correspondence between an
element c3k ∈ Z3k and an element c1k+1 ∈ Z1k+1 , so we induce a map δ : H k (C3 ) → H k+1 (C1 ), known as the coboundary
map. We then get our result:

56
Theorem 186 (Mayer–Vietoris)
With the notation above, we have an exact sequence of the form
i♯ j♯ δ i♯ j♯ δ
· · · → H k−1 (C1 ) −
→ H k−1 (C2 ) −
→ H k−1 (C3 ) →
− H k (C1 ) −
→ H k (C2 ) −
→ H k (C3 ) →
− H k+1 (C1 ) → · · · .

The proof of this requires a lot of commutative diagram chasing (keeping track of exactness and images and
kernels), but we can read about it on our own. And while this sequence does, in principle, extend infinitely, recall
that we’ve demonstrated that the H k cohomology groups are eventually zero for finite-dimensional manifolds, so the
sequence is only interesting within a finite range.

Corollary 187
Let X be a manifold, and let U1 and U2 be open sets in X. If we let C1k = Ωk (U1 ∪ U2 ), C2k = Ωk (U1 ) ⊕ Ωk (U2 ),
i j
and C3k = Ωk (U1 ∩ U2 ), then we (may check that we) get short exact sequences 0 → C1k → − C3k → 0, where
− C2k →
i is essentially restriction of a k-form on U1 ∪ U2 to U1 and U2 and j sends ω1 ⊕ ω2 to the difference of ω1 and
ω2 on U1 ∩ U2 . Then we get along exact sequence of cohomology groups
δ i♯ j♯ δ i♯
− H k (U1 ∪ U2 ) −
··· → → H k (U1 ) ⊕ H k (U2 ) −
→ H k (U1 ∩ U2 ) →
− H k+1 (U1 ∪ U2 ) −
→ ···

and also a similar result for compactly supported cohomology groups.

In particular, this often allows us to calculate the cohomology groups for U1 ∪ U2 in terms of the cohomology
groups of U1 , U2 , and U1 ∩ U2 .

34 April 29, 2022


Last lecture, we described the Mayer-Vietoris theorem, which in particular gives us an exact sequence involving de
Rham cohomology groups of U1 , U2 , U1 ∩ U2 , and U1 ∪ U2 (where U1 and U2 are open subsets of an n-dimensional
manifold X). More explicitly, if we let i : Ωk (U1 ∪ U2 ) → Ωk (U1 ) ⊕ Ωk (U2 ) be the map which sends a k-form ω on
U1 ∪ U2 to the direct sum of the ω restricted to U1 and U2 , and we let j : Ωk (U1 ) ⊕ Ωk (U2 ) → Ωk (U1 ) ∩ Ωk (U2 ) be the
map that sends (ω1 , ω2 ) to ω1 |U1 ∩u2 − ω2 |U1 ∩U2 , then we satisfy the requirements on the large commutative diagram
that we drew last time, namely that dj(ω1 , ω2 ) = j(dω1 , dω2 ) (that is, d and j commute), di (ω) = i dω (that is, d
and i commute), and composing the i and j maps gives us a short exact sequence. We’ll reproduce that diagram here,
but now in terms of the Ωi spaces instead of vector spaces C k from general complexes:
i j
0 Ωk+1 (U1 ∪ U2 ) Ωk+1 (U1 ) ⊕ Ωk+1 (U2 ) Ωk+1 (U1 ∩ U2 ) 0
d d d

i j
0 Ωk (U1 ∪ U2 ) Ωk (U1 ) ⊕ Ωk (U2 ) Ωk (U1 ∩ U2 ) 0
d d d

k−1 i k−1 k−1 j k−1


0 Ω (U1 ∪ U2 ) Ω (U1 ) ⊕ Ω (U2 ) Ω (U1 ∩ U2 ) 0

From here, “diagram chasing” lets us convert this commutative diagram into a statement about the H k cohomology
groups, namely that we have a long exact sequence
δ i♯ j♯ δ i♯
− H k (U1 ∪ U2 ) −
··· → → H k (U1 ) ⊕ H k (U2 ) −
→ H k (U1 ∩ U2 ) →
− H k+1 (U1 ∪ U2 ) −
→ ··· ,

57
where the sequence continues on between cohomology groups of U1 ∪ U2 , then of the direct sum of the groups for U1
and U2 , then of the groups of U1 ∩ U2 , incrementing the index k.
Today, we’ll see some applications of the Mayer-Vietoris theorem:

Theorem 188
Suppose U is a convex open set in Rn . Then H 0 (U) = R and H k (U) = 0 for all k > 0.

Proof. Fix some point p0 ∈ U, and for each t ∈ [0, 1], let ft : U → U be the map

ft (p) = (1 − t)p + tp0 .

This map is well-defined because U is convex, and we may notice that f0 (p) = p is the identity map, while f1 (p) = p0
is the constant map. This gives us a homotopy between the two maps, and thus by homotopy invariance, we get an
induced map ft♯ : H k (U) → H k (U) independent of t. Thus the cohomology groups of a convex set must be the same
as those of a point, which can be checked easily to be H 0 (U) = R and H k (U) = 0 for all k > 0.

Definition 189
Let X be an n-dimensional manifold, and suppose U = {Uα : α ∈ I} is a covering of X by open sets (for some
index set I). Call U a good cover if for every finite subset α1 , · · · , αk ∈ I , Uα1 ∩ · · · ∩ Uαk is either empty or
diffeomorphic to a convex subset of Rn .

This is basically a “niceness” condition, and it in particular requires the Uα s to each be diffeomorphic to a convex
subset of Rn .

Example 190
Suppose X is an open subset of Rn , and for every p ∈ X, let Up be a convex neighborhood of p. Then {Up : p ∈ X}
is a good cover (because the intersection of convex sets is convex).

We’ll now generalize the notion of convexity on Euclidean space to an analogous version for manifolds, remembering
that we always assume in this class our manifolds are subsets of RN for some large N:

Definition 191
Let X ⊆ RN be an n-dimensional manifold. For any p ∈ X, let Tp X be the tangent space of X at p, where we
view Tp X as a subset of Tp RN = RN . Then let Lp be the set

Lp = {p + v : v ∈ Tp X}.

Let πp : X → Lp be the orthogonal projection of X onto the (affine) hyperplane Lp . Then an open set U in X is
convex if for every p ∈ U, πp maps U bijectively onto a convex open set in Lp (which we can view as equivalent
to Euclidean space).

(In particular, in the definition above, if X is a one-dimensional manifold, we can think of Lp as the literal tangent
line to X at p, and more generally we can imagine the hyperplane formed by all tangent vectors at p in RN .)

58
Proposition 192
Let X be a manifold as in the above definition, and for each p ∈ X, let Bε (p) be an open ball of radius ε in Lp .
Then for any p ∈ X, we may always pick a small enough ε such that π(Bε (p)) is convex.

This result is left as an exercise to us, but the idea is to let Up = πp−1 (Bε (p)) and notice that πq (Up ) is very
close (for small ε) to the original copy Bε (p). And the idea is that a small enough perturbation of a convex set is still
convex, but we won’t write out the details. We also have the following result:

Proposition 193
If U1 and U2 are two convex open sets of X, then U1 ∩ U2 is also convex.

(This basically follows from the fact that if πp (U1 ) and πp (U2 ) are convex in Lp , then so is their intersection.)
This discussion of convexity can then lead us to the following result (by taking the Bε (p)s for each point p of small
enough ε so that each one is convex in X):

Corollary 194
Every manifold admits a good cover.

Definition 195
A manifold has finite topology if it admits a finite good cover {U1 , · · · , Um }.

Our next few lectures will be to explore the following idea: suppose we have a finite good cover {U1 , · · · , Um } of
a manifold X. We’ll see that the cohomology groups of X can be read off from intersection properties of the Ui s,
and that will give us a concrete way to compute the groups H k (X). (This is known as Čech cohomology, and it will
turn out that that theory will be isomorphic to de Rham cohomology theory.)

Theorem 196
If X admits a finite good cover, then dim H k (X) < ∞ for all k (all cohomology groups have finite dimension).

To show this, we first mention a lemma which we can prove as an exercise:

Lemma 197
α β
Suppose V1 −
→ V2 −
→ V3 are linear maps such that im α = ker β. Then if V1 and V3 are finite-dimensional, then so
is V2 .

Proof of Theorem 196. We induct on the size of the good cover. The base case m = 1 is clear, because then we can
use our convexity results from above. For the inductive step, if we have a good cover consisting of {U1 , · · · , Um−1 } and
Um , the inductive hypothesis tells us that the cohomology groups of U1 ∪ · · · ∪ Um−1 , Um , and their intersection are all
finite. Now we can apply Mayer-Vietoris and Lemma 197 to show the result (each cohomology group of U1 ∪ · · · ∪ Un
is surrounded by two finite-dimensional spaces, and we have exactness at that group because it’s part of a long exact
sequence). More explicitly, we have exactness

H k−1 ((U1 ∪ · · · ∪ Un−1 ) ∩ Un ) → H k (U1 ∪ · · · ∪ Un ) → H k (U1 ∪ · · · ∪ Un−1 ) ⊕ H k (Un ),

59
and now because U1 ∪ · · · ∪ Un−1 , Un , and (U1 ∪ · · · ∪ Un−1 ) ∩ Un all admit good covers by at most (n − 1) open sets (for
the latter using the open sets Ui ∩ Un for 1 ≤ i ≤ n − 1), all of the other cohomology groups except H k (U1 ∪ · · · ∪ Un )
are finite-dimensional and thus the middle one is as well.

In the coming lectures, we’ll see how to use Mayer-Vietoris more explicitly and get a formula for these cohomology
groups!

35 May 2, 2022
Today’s lecture will work towards connecting de Rham cohomology (involving the exterior d operation) to the new
theory that we started exploring last time, Čech cohomology. Recall from last time that this connection was introduced
in the following way: a good cover of an n-dimensional manifold X by a family of open sets U = {Uα : α ∈ I} is
a covering where any finite intersection of the Uα s is either empty or diffeomorphic to a convex open set of Rn .
We showed last time (by Mayer-Vietoris) that if X admits a good cover by a finite collection of open sets, then all
cohomology groups H k (X) are finite-dimensional. This then motivated us to think about studying cohomology via
intersection properties of the Uα s, and we’ll start that construction today.

Definition 198
Let U = {U1 , U2 , · · · , UN } be a good cover of X, and let N k be the set of all multi-indices I = (i0 , · · · , ik ) ∈
{1, · · · , N}k such that UI = Ui1 ∩ · · · ∩ Uik is nonempty. The nerve of a cover U is the collection of sets
N 0, N 1, N 2, · · · .

Definition 199
The set of Čech cochains Č k (U, R) is the set of maps c : N k → R, and the Čech coboundary map is a map
δ : Č k−1 (U, R) → Č k (U, R) defined as follows: for any c ∈ Č k−1 (U, R) and any I ∈ N k , we have
k
X
δc(I) = (−1)r c(Ir ),
r =0

where Ir is the multi-index obtained from I by deleting the index ir (that is, the (r + 1)th index).

Proposition 200
The map δδ : Č k−1 → Č k+1 is the zero map.

Proof. By applying the definition twice, we have (carefully keeping track of indices, having c ∈ Č k−1 and I ∈ N k+1 )
k+1 k+1 X
!
X X X
r r s r s−1
δδc(I) = (−1) δc(Ir ) = (−1) (−1) c(Ir,s ) + (−1) (−1) c(Ir,s ) ,
r =0 r =0 s<r s>r

with the two cases coming from the fact that indices after r get shifted by one in Ir compared to I. But then each
Ir,s shows up once in each subsum with opposite (−1)m prefactors, meaning that everything cancels and we indeed
get 0.

This implies that we have a complex

0 → Č 0 (U, R) → Č 1 (U, R) → Č 2 (U, R) → · · · ,

60
and we can now finally define Čech cohomology groups:

Definition 201
The kth Čech cohomology group of an open covering U of X is given by

Ȟ k (U, R) = ker δ : Č k → Č k+1 / im δ : Č k−1 → Č k .


 

Our goal in this last section of 18.952 is essentially to prove that

Ȟ(U, R) ∼
= H k (X)

(in other words, that the Čech cohomology group and the de Rham cohmology groups are isomorphic). Before that,
we’ll need to make a generalization of these cochains:

Definition 202
A Čech cochain of degree k with values in Ωℓ is a map c which assigns to each I ∈ N k a ℓ-form c(I) ∈ Ωℓ (UI ).

The set of Čech cochains of degree k with values in Ωℓ is the vector space Č k (U, Ωℓ ), with addition given by
pointwise addition: for any c1 , c2 ∈ Č k (U, Ωℓ ), c1 + c2 is the cochain which sends I to c1 (I) + c2 (I). We also get a
generalized coboundary map:

Definition 203
Let I ∈ N k . For any 0 ≤ r ≤ k, define the restriction map

γr : Ωℓ (UIr ) → Ωℓ (UI )

(this makes sense because UI is contained in UIr ). Then define the coboundary map (mimicking the definition
above) by setting
k
X
δc(I) = (−1)r γr c(Ir )
r =0
k−1 ℓ k
for any c ∈ Č (U, Ω ) and I ∈ N .

We may check that δδ = 0 for this more general definition, using basically the same proof as Proposition 200. We
can now start computing cohomology groups by looking at small k: a c ∈ Č 0 (U, Ωℓ ) is a map that assigns to each
1 ≤ i ≤ N an element ωi ∈ Ωℓ (Ui ). From the definition, we see that δc = 0 if and only if

ωi |Ui ∩Uj − ωj |Ui ∩Uj = 0 ∀(i , j) ∈ N 1

In other words, the kernel of the map δ : Č 0 (U, Ωℓ ) → Č 1 (U, Ωℓ ) is Ωℓ (X) itself (the set of forms that can be defined
consistently on all of the Ui s at once). Next time, we’ll extend this to see that we in fact get an exact sequence of
the form
0 → Ωℓ (X) → Č 0 (U, Ωℓ ) → Č 1 (U, Ωℓ ) → · · ·

and see how that leads us to the isomorphism we’re after.

61
36 May 4, 2022
Last lecture, we defined the Čech cohomology groups Ȟ k (U, R) (where U = {U0 , · · · , Uk } is a finite open cover of
a manifold X). The motivation for this cohomology theory is that it is easier to compute the Čech cohomology
groups from the definition than the de Rham cohomology groups – recall that if N k is the set of all multi-indices
(i0 , · · · , ik ) ∈ {1, · · · , N}k such that Ui0 ∩ · · · ∩ Uik is nonempty, then Čech cohomology is defined in terms of the Čech
cochains Č k (U, R) (the maps from N k to R) and the Čech coboundary map δc(I) = kr=0 (−1)r c(Ir ). We showed
P

that δδ = 0, forming a complex 0 → Č 0 (U, R) → Č 1 (U, R) → · · · , and that gives us the Čech cohomology groups
Ȟ k (U, R) = (ker δ : Č k → Č k+1 )/(im δ : Č k−1 → Č k ).
Today, we’ll explain why the H k and Ȟ k cohomology groups are in fact the same. Last time, we generalized
Čech cochains to take values in Ωℓ (for some integer ℓ) – such cochains are maps c that assign to each multi-index
I ∈ N k an element c(I) ∈ Ωℓ (UI ). The sets Č k (U, Ωℓ ) are vector spaces, and we defined a generalized δ map
δc(I) = kr=0 (−1)r γr c(Ir ) (where γr is the restriction map of a form from UIr to UI ). Since δ 2 = 0 here as well, we
P

are now motivated to consider the cohomology groups associated to this complex (of cochains with values in Ωℓ ).
Last time, we mentioned that an element c ∈ Č 0 (U, Ωℓ ) is a map that assigns to each index 1 ≤ i ≤ N an ℓ-form
ωℓ ∈ Ωℓ (Ui ). Furthermore, δc = 0 for such a map if and only if for every (i , j) ∈ N 1 , γi c(j) − γj c(i ) = 0, meaning
that there is some unique ℓ-form ω ∈ Ωℓ (X) such that ω restricted to each Ui is ωi . Thus the kernel of the map from
Č 0 (U, Ωℓ ) → Č 1 (U, Ωℓ ) is the set of all differential k-forms on X, Ωℓ (X). Inserting this into the complex, we now
arrive at the result from the end of last lecture, which is that we have a sequence
δ
0 → Ωℓ (X) → Č 0 (U, Ωℓ ) →
− ··· .

Theorem 204
The sequence above is exact for any ℓ.

Proof. Let φi ∈ C0∞ (X) be a partition of unity, such that the support of φi is contained in Ui for each i . Define the
map Q : Č k+1 (U, Ωℓ ) → Č k (U, Ωℓ ) (notice that this lowers the order k instead of raising it like δ does) via
N
X
Qc(I) = φi c(i , I),
i=1

where we extend each φi c to be zero outside Ui ∩ UI . We may check that (Qδ + δQ)c(I) = c(I) for any I, and now
if we fix k, we can define the map d : C k (U, Ωℓ ) → C k (Ωℓ+1 ) via dc(I) = d(c(I)). Since d 2 = 0 on forms, we thus
get a valid complex
Č k (U, Ω0 ) → Č k (U, Ω1 ) → Č k (U, Ω2 ) → · · ·

via the d map. Additionally, because 0-forms c(I) are elements of C ∞ (UI ), dc(I) = 0 if and only if c(I) is a constant,
and thus dc = 0 if and only if c ∈ Č k (U, R). Thus we get an exact sequence

0 → Č k (U, R) → Č k (U, Ω0 ) → Č k (U, Ω1 ) → · · · ,

because if dc = 0 for some c ∈ Č k (U, Ωℓ ), then c(I) ∈ Ωℓ (UI ) must be closed for any I, and because UI is diffeomorphic
to a convex open set (by definition) we have c(I) = dc(I)′ for some c(I)′ ∈ Ωℓ−1 (UI ). (This implies that the image
of one d map is the kernel of the next.)

It is now left as an exercise to check that the d and the δ maps in fact commute, meaning that for any c ∈ Č k (U, Ωℓ )

62
we have dδc = δdc (this can be checked from the definitions). From all of the properties we’ve verified, we thus get
a commutative diagram of the following form (where empty arrows are basically inclusion maps):

.. .. .. ..
. . . .
d d d d

δ δ δ
0 Ω2 (X) Č 0 (U, Ω2 ) Č 1 (U, Ω2 ) Č 2 (U, Ω2 ) ···
d d d d

δ δ δ
0 Ω1 (X) Č 0 (U, Ω1 ) Č 1 (U, Ω1 ) Č 2 (U, Ω1 ) ···
d d d d

δ δ δ
0 Ω0 (X) Č 0 (U, Ω0 ) Č 1 (U, Ω0 ) Č 2 (U, Ω0 ) ···

δ δ δ
0 Č 0 (U, R) Č 1 (U, R) Č 2 (U, R) ···

0 0 0

In this picture, everywhere except the blue column and row is exact, and those columns and rows are the de Rham
and Čech complexes that we’ve defined in the last few lectures. Such a diagram therefore allows us to convert a form
c ∈ Ωk+1 (X) such that dc = 0 to a cochain č ∈ Č k+1 (U, R) such that δ č = 0, by basically making a sequence of
right-down-right-down moves between the left row of Ωk s and the bottom column of Č k s. This shows that whenever
X admits a good cover, H k and Ȟ k indeed agree.

63

You might also like