Professional Documents
Culture Documents
Reuben Stern
5 February 1, 2017 18
5.1 Finishing Things from Last Time . . . . . . . . . . . . . . . . . . . . . . . 18
5.2 Dual Results: Exact Sequences . . . . . . . . . . . . . . . . . . . . . . . . 18
5.3 Relative Homotopy Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
6 February 3, 2017 21
6.1 The Action of the Fundamental Group . . . . . . . . . . . . . . . . . . . . 21
7 February 6, 2017 22
7.1 Fibrations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
8 February 8, 2017 23
8.1 Fiber Bundles and Examples . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1
Math 231br CONTENTS
16 March 1, 2017 47
16.1 Co/homology Theories from Spectra . . . . . . . . . . . . . . . . . . . . . 47
17 March 3, 2017 51
17.1 The Hurewicz Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
17.2 Spectral Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
18 March 6, 2017 54
18.1 Obstruction Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
19 March 8, 2017 55
19.1 A Complicated Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
References 80
0.1 Preliminaries
These notes were taken during the Spring semester of 2017, in Harvard’s Math 231br,
Advanced Algebraic Topology. The course was taught by Eric Peterson, and met Mon-
day/Wednesday/Friday from 2 to 3 pm. Allow me to elucidate the process for taking
these notes: I take notes by hand during lecture, which I transfer to LATEX at night. It is
an unfortunate consequence of this method that these notes do not capture the unique
lecturing style of the professor. Indeed, I take full responsibility for any errors in exposition
or mathematics, but all credit for genuinely clever remarks, proofs, or exposition will be due
to the professor (and not to the scribe). In an appendix at the end of this document, you will
find the collected homework problems (with solutions). I make no promises regarding the
correctness of these solutions; consider yourself warned. Please send any and all corrections
to reuben_stern@college.harvard.edu. They will be most appreciated.
A rough syllabus for the course is, in Eric’s own words, to “cover some initial segment
of the book.” To explain, the course plans to discuss general homology and cohomology
theories, bordism homology and cohomology, stable homotopy groups of spheres, and
spectral sequences as computational tools for homotopy groups.
3. Representability theorems (Brown, Adams) and the stable category (HZ, KU and
KO, S, HG, etc.)
As an example of the kind of analysis we can perform with these tools at our disposal,
start with your favorite simply connected space, like S n≥2 . Its homotopy groups are
notoriously difficult and important to compute.
Theorem 1.1. (Hurewicz) If X is (n − 1)-connected (that is, πk X = 0 for all k < n),
then Hn (X; Z) ∼
= πn X. We also have a decomposition
X[n + 1, ∞] X K(πn X, n)
Then we can also plug the data of H∗ X and H∗ K(πn X, n) into a gadget known as the
Serre spectral sequence to get H∗ X[n + 1, ∞].
Eric: “I can’t make it through a lecture without saying ‘category’ at least five times.”
The goal of this lecture is to give a couple of technical lemmas and some basic con-
structions, so that we don’t have to mention them explicitly ever again. There will be four
examples in this class: a sheaf condition on a decomposition of a topological space, product
spaces, gluings (or quotient spaces), and the space Y X of functions X → Y .
Aside 2.1. (Sheaves.) A sheaf is an assignment from subsets (usually taken to be open)
of a space X to arbitrary sets, together with restriction maps: if B ⊆ A ⊆ X, we get a
map resA,B : F (A) → F (B). (Example: take F (A) = {continuous functions A →
T }.) We ask that these data satisfy:
The sheaf condition is then equivalent to the leftmost map equalizing the right
hand side.
And before we move on, an aside on equalizers! As Eric keeps pointing out, this class
will be very category-heavy.
f
Aside 2.2. (Equalizers.) An equalizer (of sets) E ⊆ S g
T is the maximal subset
of S on which f and g restrict to the same function.
h
Lemma 2.3. Suppose that F −−→ S is some function such that f ◦ h = g ◦ h. Then
there is a unique factorization of h as
h
F S
i
∃!
E
Lemma 2.4. If f : X → Y is continuous such that xRx0 implies f (x) = f (x0 ), then
there is a factorization of continuous maps
f
X Y
∃!
X/R
1
This is Eric’s notation; I much prefer writing Top for this category, but I will follow his anyway because
I am the student and he is the master.
2
In the categorical sense; we will talk about this more later
Special Case: let A ⊆ X, and take the maximum equivalence relation on A. Extend
this by the identity relation on X to get a relation R where xRx0 if and only if x, x0 ∈ A.
The quotient with respect to this equivalence relation is then called X/A (edge case3 :
i f
X/∅ = X q {∗}). Suppose A X Y , and f ◦ i is constant. Then the diagram
commutes:
i f
A X Y
fe
X/A
X×Y X Y
R×S R
× S
which is always a continuous bijection. Unfortunately, this is the best we can say in
general.
Lemma 2.6. If Y is locally compact and S is the identity relation, then this map is a
homeomorphism.
Example 2.7. Take Y = [0, 1], the unit interval, and S to be the identity relation.
Then
X X ×I
×I ∼= .
α α × id
This shows up a lot.
3
This is mostly a convention
4
With the compact-open topology; the exchange went like this: me: “What’s the topology on Y X ?”
Eric: “The compact-open topology, but the short answer is it doesn’t matter.”
ev
Lemma 2.9. If X is locally compact, then Y X × X −−−−−→ Y is continuous. If X
and Z are additionally Hausdorff, then
Y Z×X ∼
= (Y Z )X .
= ∼
Corollary 2.10. There is a natural bijection Spaces(X × Z, Y ) −−→ Spaces(X, Y Z ).
g◦ g◦
∼
=
Spaces(X × Z, Y 0 ) Spaces(X, (Y 0 )Z )
commutes.
Lemma 2.11. The product of two pairs (X, A) × (Y, B) in the category of pairs is
Proof. Exercise.
Example 2.12. Let (X, {x0 }) and (Y, {y0 }) be pointed spaces. Then the product is
But this distinguished subset of the product should be familiar: it is exactly the wedge
product of X and Y at x0 and y0 !
At the end of class, we began going through a presentation of a natural bijection which
would lead well into the definition of the smash product; we didn’t have enough time to
complete it:
that sends (X, x0 ) to the pair (X, {x0 }). In the other direction, there is a functor quot :
Pairs → Spaces∗/ sending (X, A) to (X/A, ∗), where ∗ is the preferred point to which A is
collapsed. We want to use this pair of functors to study the induced adjunction. We have a
square
×
(X, {x0 }), (Y, {y0 }) (X × Y, X ∨ Y )
i quot
X×Y
(X, x0 ), (Y, y0 ) X∨Y
,∗
Spaces∗/ (X, Y Z ) ∼
= Spaces∗/ (X ∧ Z, Y ).
X ∧ Z 6= (X, x0 ) × (Y, y0 ).
Recall that π1 (X, x0 ) is the collection of based loops in X, taken up to homotopy. With
our notation, this is [S 1 , (X, x0 )]. Note that S 1 = S 0 ∧ S 1 : S 0 × S 1 is two copies of S 1 , and
then we collapse down S 0 ∨ S 1 = {∗} q S 1 to the basepoint on the other copy of S 1 , leaving
1 1
us with just the other copy. Therefore [S 1 , X] = [S 0 ∧ S 1 , X] = [S 0 , X S ], which is π0 (X S ).
1
The space X S is the function space of based loops in X. It is so frequently used that it
gets its own name: it is the loop space of X, and we denote it by Ω(X).
Recall as well that π1 (X) is a group.
Question 3.3. Expressing π1 as above, what is special about the functors π0 Ω(−) or
[S 1 , −] that makes these functors group-valued?
Eric: “Naturallya , we’re going to address this problem with category theory.”
a
No pun intended
µ : G × G → G, η : ∗ → G, and χ:G→G
G×G µ G G
∆ χ×1 1×χ ∆
G G×G G×G G×G G
µ
η η
∗ G ∗
Lemma 3.5. If G is a group, then Sets(X, G) forms a group, for any other set X.
Proof. We first have to define the “group law” on Sets(X, G): to do this, first note that there
is a canonical isomorphism (whatever that means) Sets(X, G)×Sets(X, G) ∼ = Sets(X, G×G)
(“two maps from one space into another is the same thing as a map into the cartesian
product of the codomains”). We thus consider the composition
' µ∗
µ0 : Sets(X, G) × Sets(X, G) Sets(X, G × G) Sets(X, G),
Back to Question 3.3, what does this definition have to do with our problem?
Answer 1: It turns out that ΩX is a “group” object in the homotopy category of
pointed spaces. That is, we have maps
making all the relevant diagrams commute (note that in this case, ∗ is the terminal/initial
object in the category hTop∗ ). In order to show this, we have to show that all the
diagrams commute, but we’ve done this before in 131/231a! This implies that [S 0 , Ω(X)] =
Sets(S 0 , Ω(X)) is automatically a group (“that’s cute!”). In fact, [Y, ΩX] is automatically
a group for any set Y .
Answer 2: “Adjoint-ly”, it turns out that S 1 is a cogroup object in7 hTop∗ . Eric: “It
is somewhat unlikely that you’ve heard the word ‘cogroup’ aloud,” so we’ll define it now.
Definition 3.6. A cogroup is “a group with all the arrows turned around and all the
products converted to coproducts.” That is (understanding that X ∨ Y is the coproduct of
X and Y in Spaces), we have maps
µ0 : C → C ∨ C, η 0 : C → ∗, and χ0 : C → C,
1∨µ0
C ∨C ∨C C ∨C
µ0 ∨1 µ0
C ∨C µ0
C
6
A functor applied to a commutative diagram yields a commutative diagram
7
This is the homotopy category of pointed spaces, where morphisms are homotopy classes of maps
Example 3.7. What do the cogroup operations look like for S 1 ? We have a “pinch”
map µ0 : S 1 → S 1 ∨S 1 , a collapsing map η 0 : S 1 → ∗, and a “flipping” map χ0 : S 1 → S 1 .
Eric: “I wish that I had three hands.” We then show that the diagrams commute (this
is easy but laborious; left as an exercise).
Remark 3.8. S 1 ∧(−) is also a cogroup in hTop∗ . We can see this more or less algebraically:
(Z ∧ (X ∨ Y )) = ((Z ∧ X) ∨ (Z ∧ Y )), so we can work things out. The functor S 1 ∧ (−)
is called the (reduced) suspension, and is written Σ(−). It is adjoint to the loop space
functor.
∆ f 0 ×g 0 µ
X X ×X ΩY × ΩY ΩY
In the case n = 0, 1, there are no intermediate groups. But when n ≥ 2, we have some
legitimately interesting structure: namely, the intermediate groups have two group structures
on them directly, rather than through isomorphism. The following lemma will tell us that
those structures are the same.
Lemma 4.2. (Eckmann-Hilton) Let S be a set with two products, ∗ and ◦, that share
a unit. Suppose further that (x ∗ x0 ) ◦ (y ∗ y 0 ) = (x ◦ y) ∗ (x0 ◦ y 0 ). Then ∗ and ◦ agree,
and both are associative and commutative.
Corollary 4.3. [Σn−1 S 0 , ΩX] has only one product, and it is commutative.
Proof. This is a proof by diagram chase and verbal diarrhea. The diagram is the following:
(f ∨f 0 ) fold
× ×
µ0 ×µ0 (g∨g 0 ) fold
K ×K (K ∨ K) × (K ∨ K) (L ∨ L) × (L ∨ L) L×L
∆
f
Lemma 4.7. Any map of spaces X −−→ Y extends to a coexact sequence
f
X Y Z.
Y q CX
Z = Y ∪f CX = ,
f (x) ∼ (x, 1)
where the cone on a topological space CX is defined to be X × I/(X × {0} ∪ {x0 } × I).
Technically, this is the definition of the reduced cone, but we drop words all the time
anyways, so what’s the problem?
Note that a nullhomotopy of a map X −→ T is the same data as a map CX → T 9 , so
a nullhomotopy of f ∗ γ is a map CX −→ T such that
CX T
f ∗γ
X
8
Eric: “Do you still call them air quotes when you write them on the board?”
9
Exercise: convince yourself of this.
X CW
f
Y Y ∪f CW.
f i j
X Y Y ∪f CX (Y ∪f CX) ∪i CY ((Y ∪f CX) ∪i CY ) ∪j C(Y ∪f CX) ··· .
These look nasty, but they really aren’t! We can see that by the two following lemmas,
given without proof:
is a homotopy equivalence.
f
X Y Y ∪f CX ΣX ΣY Σ(Y ∪f CX) Σ2 X ···
We get some interesting fringe phenomena at the maps [ΣX, T ] → [Cf, T ] and [Σ2 X, T ] →
[Σ(Cf ), T ], where the target set has less structure than the initial.
5 February 1, 2017
5.1 Finishing Things from Last Time
f
Recall from last time: for a map X −→ Y , we can iteratively construct Y ∪f CX =: Cf to
get a sequence
f
X Y Cf ΣX ΣY Σ(Cf ) ··· .
Applying the functor [−, T ] for some test space T , we get something interesting happening
at
HomhTop∗/ (ΣX, T ) HomhTop∗/ (Cf, T ) :
HomhTop∗/ (Cf,T )
HomhTop∗/ (ΣX,T )
CX Y
f
X
The next two lemmas follow by duality from things proved last time:
prX f
Lemma 5.3. The sequence Pf X Y is exact in Spaces∗/ .
Lemma 5.4. Iterating this construction yields a long exact sequence in Spaces∗/ :
f
··· Ω2 X Ω2 Y ΩPf ΩX ΩY Pf X Y.
π1 Pf π1 X π1 Y π 0 Pf π0 X π0 Y
Question 5.5. What is this Pf object, and how can we calculate anything about its
homotopy groups? Knowing the homotopy groups can give us a significant amount of
information about those of X and Y .
π≤n (X, A) = 0.
Remark 5.8. We can actually convert any map f : Y → X into an inclusion as follows:
→ Mf of X into the (reduced) mapping cylinder of f ,
there is a homotopy equivalence X ,−
i
→ Mf by sending Y to Y × {0}. The diagram
and an inclusion Y ,−
Mf
i
f
Y X
Each of these pairs has a long exact sequence in homotopy groups associated to it; we can
stitch these together into the following diagram:
πn A πn (X, B) πn−1 A
6 February 3, 2017
6.1 The Action of the Fundamental Group
Maybe you recall that there is an action of π1 X on πn X. If not, you may remember that a
path γ : I → X induces an isomorphism Γ : π1 (X, γ(0)) → π1 (X, γ(1)) by γ −1 · ω · γ.
(ii) The multiplication map on A is G-equivariant with the A-action. Explicitly, this
means
g · (a1 a2 ) = (g · a1 )(g · a2 ).
7 February 6, 2017
7.1 Fibrations
j f
Remember that we had this special object Pf , fitting into an exact sequence Pf X Y,
where Pf was defined as
πn Y πn (X × Y ) πn X
is exact.
8 February 8, 2017
8.1 Fiber Bundles and Examples
follows (and in the diagram): slide g1 over via H. Part of the cone is homeomorphic to make
I × S n−1 ; this “fills in” the part of the homotopy given by sliding g1 (of course, all of this is dia-
in very imprecise language). We can then “pull the top of the cone” down over the rest of gram
it, to give us the attachment on g2 .
Ultimately, our goal is to become familiar with the behavior of X ∪g CS n−1 as a homotopy
type.
Lemma 10.1. For (X, A) a relative CW complex, the pair (X, (X, A)n ), where (X, A)n
is the n-skeleton of (X, A), is n-connected: π≤n (X, (X, A)n ) = 0.
Corollary 10.2. The inclusion X n ,−→ X is n-connected (or, the pair (X, X n ) is).
Proof. Use seriously the simplicial approximation theorem from last time.
Proof. We use the CW structure of S n as one n-cell glued to a point. Then (S n )n−1 = ∗,
and S n = ∗ ∪∗ CS n−1 . By the lemma, π<n (S n , ∗) = 0.
Corollary 10.5. For X an n-connected space and Y an m-connected space, the space
X ∧ Y is (n + m + 1)-connected. As Eric says, this is “a little more connected than
you might expect.”
I: Localizing Categories
Let C be a category, and W ⊂ Mor(C ). We want to construct a category with the
morphisms in W formally inverted (i.e., become isomorphisms).
To solve these problems, we want to find an equivalent category with better properties.
By equipping (C , W ) with a “model structure”, we’ll define the “homotopy category” ho(C ),
which is better behaved.
C D
LH C
Definition 11.4. We now define the hammock localization LH C . The objects ob LH C are
just ob C . Because LH C is enriched over sSet, morphisms must form simplicial sets. We
let 0-simplices be zig-zags as before, in the morphisms of C [W −1 ]. The 1-simplices will be
“hammocks”:
∼ ∼
· · · ··· ·
X ∼ ∼ ∼ ∼ Y
∼ ∼
· · · ··· ·
C F
D
F0
ho C ho D
commutes? The naïve thing to do is to use Kan extensions, but there are issues with
that.
Model categories solve the problems, by telling us what kinds of functors you can derive,
and exactly how to do it.
C F
E C F
E
η LanK F
G α
K
γ = K
K
D D
Right Kan extensions are defined dually, with the natural transformation arrows
reversed.
X Y
We say that “i has the left lifting property with respect to p” or “p has the right lifting
property with respect to i” if the dashed arrow exists, and the whole diagram commutes.
We write this i ∈ llp(p) or p ∈ rlp(i).
(MC2) W has the “2-out-of-3” property: if f and g are composable arrows, and two of
{f, g, g ◦ f } are weak equivalences, then so is the third.
f f
· · and · ·
∼ ∼
· ·
idA
A B A
f g f
X Y X
idX
Axioms (MC4) and (MC5) make a category into a weak factorization system. Qullen’s
original formulation of model categories required only that C be finitely bicomplete and
that factorizations need not be functorial. What we give as the definition for a model
category is what Quillen originally called a closed model category.
Remark 11.11. The model category axioms are dual: if C is a model category, so is C op ,
with weak equivalences the same, and fibrations and cofibrations flipped. The upshot of
this is that we only need to prove half of the theorems!
Remark 11.12. We call the category W ∩ fib the class of acyclic fibrations, and W ∩ cof
is the class of acyclic cofibrations.
10
A wide subcategory is a subcategory that includes all objects and all identity morphisms.
Definition 11.13. Because C is bicomplete, it has an intial object ∅ and a final object ∗.
We say that A is a cofibrant object if ∅ → A is a cofibration, and a fibrant object if
A → ∗ is a fibration.
Example 11.14. (1a) Take C = Spaces. Then we let W = {weak htpy equivalences},
fib = {Serre fibrations}, cof = {retracts of cell complexes}
(1b) Again take C = Spaces, with W = {weak homotopy equivalences}, but this time
let fib = {Hurewicz fibrations} and cof = {closed Hurewicz cofibrations}. By a
theorem we will prove later, it doesn’t really matter if you know what these are
or not.
(2a) Take chain complexes ChR of R-modules. Let W = {quasiisomorphisms} (that is,
maps that induce isomorphisms on homology), fib = {degree-wise epis in positive
degrees}, and cof = {degre-wise monos in all degrees with projective cokernels}.
This is called the projective model structure.
(2b) C = ChR again, and W the same. This time, take fibrations to be degree-wise
epis in all degrees with injective kernel, and cofibrations to be degree-wise monos
in positive degrees. This is called the injective model structure.
3 The category sSets of simplicial sets has a model structure. In general, given any
abelian category, one may form the simplicial category sA , and endow it with a
model structure.
Exercise: check the model category axioms against some of these examples.
Proposition 11.15.
Proof. As an example to get a taste for how these proofs go, we show cof = llp(W ∩ fib).
Note that we only need to prove the “⊇” direction. Suppose f ∈ llp(W ∩ fib). Factor f as
i
A B
r
f p
e
X Y
A A A
f i f
r p
X B X
Proposition 11.16. The subcategories W ∩ cof and cof are stable under cobase
change (i.e., pushout) and fib and W ∩ fib are stable under base change (i.e.,
pullback).
Proof. An exercise in working with diagrams, much in the same way as the above.
A f
i0 ti1 H
AtA cyl(A) X
A g
commutes.
Lemma 11.18. With notation as above, if A is cofibrant, then i0 and i1 are cofibrations.
AtA
∅ A
η0
p
A AtA
`
Proposition 11.19. If A is cofibrant, then ∼ is an equivalence relation on maps
C (A, X).
H
AtA cyl(A) X
g
A
`
commutes. From this, it is clear that ∼ is symmetric and reflexive; the tough part is to
show transitivity.
A
i0
∼
A cyl(A)
∼ ∼
i01 ∼ p ∼
A cyl(A)0 C
∼
∼
The object C comes from taking the pushout. All maps not involving C are given in the
definitions of the cylinder objects. The universal property of pushout gives us the dotted
map, which is a weak equivalence by the two-out-of-three axiom. We also have the following
diagram commuting:
f
A cyl(A)
C X
A cyl(A)
g
`
Lemma 12.2. Suppose X is fibrant, and f ∼ g : A → X. Then if h : A0 → A is a
`
map, we have f h ∼ gh.
`
Proof. Step 1: There exists a map H : cyl(A) → X witnessing f ∼ g. Factor H (by the
axiom MC5) as
∼
AtA cyl(A) A
∼ ∼
cyl(A0 )
cyl(A0 ) ∗
hth
A0 t A0 AtA cyl(A)
k
∼
∼ h
cyl(A0 ) A0 A
h
A0 A
f
k H
cyl(A0 ) cyl(A) X
g
h
A0 A
X cocyl(X) X × X.
r
Two maps f ∼ g : A → X are right homotopic if
f X
A cocyl(X)
g
X
The two equivalence relations thus coincide if A is cofibrant and X is fibrant. In this
case, we define
π(A, X) := π ` (X, A) = π r (X, A).
With this, we can state the Whitehead theorem, one of the most important theorems in
homotopy theory. One consequence of this theorem is that a map between CW complexes is
a weak equivalence (i.e., induces isomorphisms on homotopy) if and only if it is a homotopy
equivalence. Let us say an object is bifibrant if it is both cofibrant and fibrant.
∼
Proof. Suppose f : A −−→ X. Factor f as
f
A ∼ X
∼ ∼
q p
C
∅ C
s p ∼
X X
p∗ : π ` (A, C) −→ π ` (A, X)
is a bijection.
`
Assuming the lemma, we see that sp ∼ idC . Dually, we find a homotopy inverse r for q,
and rs is a homotopy inverse for f = pq.
In the other direction, factor f as
f
A X
∼ p
q
H
cyl(X) X.
Set s = Hi
f . Then ps = id . We also have a homotopy inverse for q, say r. As pq = f ,
1 X
pqr = f r ⇒ p ∼ f r, so sp ∼ qgp ∼ qgf r ∼ qr ∼ idC .
Aside 12.7. If X is any object of a model category C , we can make X into a cofibrant
object via the factorization
∅ X
∼
QX
RX .
Remark 12.9. Note in general that RQX is not the same thing as QRX. Thus one may
ask: does using QRX instead of RQX in the definition of the homotopy category make a
different category? The next theorem will answer that: both satisfy the same universal
property, and thus are equivalent.
We have a canonical functor γ : C =⇒ ho(C ) (this is why we require our factorization
to be functorial).
C F
D
γC γD
ho(C ) LF
D
We can (more or less) obviously dualize this definition, to arrive at that of a total right
derived functor.
A natural thing to ask, then, is: what functors can we derive? One class of such functors
is Quillen functors:
Lemma 12.13. (Ken Brown’s Lemma). Let C be a model category, and D a category
with a subcategory of weak equivalences, satisfying the two-out-of-three condition. If
F : C → D is a functor that takes acyclic cofibrations between cofibrant objects to
weak equivalences, then F takes all weak equivalences between cofibrant objects to weak
equivalences.
∼
Proof. (This proof is adapted from [Hov98].) Let f : A −−→ B be a weak equivalence
between cofibrant objects. We can factor the map f t idB : A t B → B as
f tidB
AtB B
p
q ∼
∅ A
p
B AtB
and q ◦ i1 are weak equivalences, and thus acyclic cofibrations of cofibrant objects. By
assumption then, both F (q ◦i0 ) and F (q ◦i1 ) are weak equivalences. As F (p◦q ◦i1 ) = F (idB )
is a weak equivalence, the two out of three axiom again gives that F (p) is a weak equivalence,
and thus that F (f ) = F (p ◦ q ◦ i0 ) is a qeak equivalence.
The upshot of this is that a left Quillen functor preserves weak equivalences between
cofibrant objects.
Example 12.14. 1. Take the category of chain complexes, ChR . Fix some M ∈ ChR .
We have an adjunction of functors
We get the derived functors M ⊗LR − and RHomR (M, −). Taking homology gives
Tor and Ext, respectively.
3. Homology is the left derived functor of “abelianization,” whatever the hell that
means.
In summary, the homotopy category, derived adjunctions, and derived natural trans-
formations give the data of a pseudo-2-functor between
ho
ModelCat −−−−−→ Catad .
B Y
ω|∂Dn ω
S n−1 Dn
'
Lemma 14.2. If π∗<n X and π∗>n Y = 0, then [X, Y ] −−→ [πn X, πn Y ].
The idea of the second lemma is that X has no new cells of dimension less than n. From
this we conclude
∼
Corollary 14.4. ΩK(A, n) ←−− K(A, n − 1).
Collections of spaces with this property are super interesting! We should read this as
saying: the space K(A, n − 1) has a space sitting above it of which it is the loop space. We
call the space K(A, n) the delooping of K(A, n − 1).
Remark 14.7. This is sort of a converse to two claims: first, if you are given a functor, you
can construct a representing object. Thus, the second part of the theorem is a consequence
of Yoneda and the first part.
Definition 14.8. An element u ∈ F (Y ) is n-universal if [S q , Y ] → F (S q ) taking ω 7→ ω ∗ u
is onto for all q ≤ n, and an isomorphism for all q < n.
Proof of Brown Representability. Suppose that un ∈ F (Y ) is n-universal. We want to
correct un and Y to be (n + 1)-universal. Consider A = {α ∈ πn Y : α∗ un = 0} and
L = F (S n+1 ).
Form W
mapping cone
Sαn −−−α→ Y ∨ Sλn+1 −−−−−−−→ Y 0 .
_ _
α λ
Applying F , we have
Sαn ) F Y ∨ Sλn+1 F (Y 0 )
W W
F( α λ
∈
α∗ un ∃ un+1
W W
α (un , λ λ)
a
i.e., n-universal for all n
π∗ Y π∗ Y 0
∼
= ∼
=
F (S ∗ ) F (S ∗ )
j −1
Y Y0 X
Example 14.10. We like the spaces S 1 and CP ∞ , because they are K(Z, 1) and
K(Z, 2), respectively. Suppose one wanted to build the functor that S 1 represents
“from scratch.” That is, start from a point, get a natural transformation [−, ∗] → [−, S 1 ].
The Brown Representability proof says you should take a bouquet of 1-spheres, and
consider [−, Z S 1 ]. You can say a similar bunch of words for CP ∞ .
W
Proof of Brown Representability, finished. To get surjectivity (of F (X) ←− [X, Y ]), take
A = ∗ in the lemma. To get injectivity, set X 0 = X × I, and set A0 = X × ∂I. Use A0 to
support g1 and g2 . Use the lemma, so g1 = g2 .
Remark 14.11. Brown representability applied to H fn (−, A) gives K(A, n). So Eilenberg-
Maclane spaces are intimately rooted in simplicial homology.
Aside 15.1. Recall that a sheaf has (i) the sheaf condition (i.e., Mayer-Vietoris axiom)
and (ii) the wedge axiom. Furthermore, if you have a sequence of co/homology functors,
we have
he n+1 (ΣX) ∼
=he n (X),
e n (X)
h ∼
= hTop∗/ (X, hn ) induced by hn →hn+1
∼
= ∼
=
e n+1 (ΣX)
h ∼
= hTop∗/ (ΣX, hn+1 ) ∼
= hTop∗/ (X, Ωhn+1 ).
We can also read the results of Brown Representability backwards, and have the spaces
hn define a cohomology theory.
15.1 Spectra
Goal 15.2. Use this presentation of cohomology functors by sequences of spaces to blend ho-
motopy theory into the category of cohomology theories (with maps natural transformations
of cohomology functors). More precisely, we want a category where
π0 S 0 = {±1}, π1 S 1 = Z, π2 S 2 = Z, π3 S 3 = Z, . . .
fn
(S0 )n (S0 )n
fn
(S0 )n (S0 )n
Σfn−1
Σ(S0 )n−1 Σ(S0 )n−1
commutes. Because the linking (vertical) maps are surjective, fn is determined exactly by
Σfn−1 , so none of the fn≥1 matter. It follows that
F0 E
F 00 F
F 00 ∧ I+ G.
Lemma 15.11. Spectra have wedge sums and cofiber sequences/mapping cones.
Remark 15.13. We also need to account for n ∈ Z− , as seen by the following lemma.
Thus we have isomorphisms ΣSn ' Sn + 1, and can define S−1 = Σ−1 S0 .
16 March 1, 2017
16.1 Co/homology Theories from Spectra
Last time, we built this category Spectra such that the functor Σ∞ induces the diagram
Σ∞
Spaces Spectra
Σ∞
hTop hSpectra,
B.R.
and where Brown Representability gives a functor CohomThy −−−→ hSpectra such that
Question 16.1. Why do we just get cohomology theories? We’re missing half the picture
here.
where the spectrum E ∧ X is given by (E ∧ X)n = En ∧ X, and the suspension maps are
the natural ones.
Remark 16.5. The assigment E (En , ΣEn → En+1 ) presents E as lim Σ−n Σ∞ En , i.e.,
−→
E is an ind-system of desuspensions of suspension spectra. Maps between ind-systems
Xα and Yβ are generally computed by
lim lim[Xα , Yβ ],
← − −→
α β
1. (Suspension.)
i
2. (Cofibrations.) If A ,−→ X −→ X ∪i CA is a coexact sequence, then
Σ∞ A −→ Σ∞ X −→ Σ∞ (X ∪i CA) = Σ∞ X ∪i CΣ∞ A
E ∧ Σ∞ A E ∧ Σ∞ X Σ∞ (X ∪i CA) ∧ E
E∧A E∧X
3. (Wedge.)
! !
∗ ∞
Ee n n
_ _
Xα = hSpectra Σ Xα, Σ E
α α
!
∗ _
∞ n
= hSpectra Σ Xα , Σ E
α
∗
hSpectra (Σ∞ Xα , Σn E) = Ee n (Xα ).
Y Y
=
α α
Apparently, all the equalities marked with a red ∗ are easy enough to believe.
do not give an equivalence of categories. It is almost an equivalence, but fails because of the
existence of things called phantom maps of spectra: maps between spectra that cannot
are not associated to natural transformations of the corresponding cohomology theories.
However, we can get our hands on a lot of this “almost-equivalence” in one direction by
Whitehead’s theorem, and on the other side by inventing spectral sequences.
Remark 16.7. The cohomology of a union of spaces does not in general equal the limit of
the cohomologies. Instead, we can invoke Milnor Sequences; see Homework #3.
Remark 16.8. The map from CohomThy −→ hSpectra is secretly defined on objects at
the level of spectra, without passing to the homotopy category. These spectra satisfy an
adjunction:
Σhn hn+1
⊥
hn Ωhn+1
The spectra are called Ω-spectra, and they satisfy
hSpectra(Σ∞ X, Σ∞ Y ) = lim[Σn X, Σn Y ]
−→
= lim[X, Ωn Σn Y ]
−
h→ i
= X, lim Ωn Σn Y ,
−→
and we set y 0 = lim Ωn Σn Y . More generally, lim Ωn Σn (Σm Y ) = y m .
−→ −→
If we apply this recipe to the spaces in an arbitrary spectrum, we recover a sequence of
spaces in an equivalent Ω-spectrum:
E m = lim Ωn Em .
−→n
Remark 16.9. The functor − ⊗Z Z(p) , where Z(p) is Z localized at the prime ideal (p),
preserves exact sequences of abelian groups. So the assignment
also sends cofiber sequences to long exact sequences, hence is represented by a spectrum
E(p) .
The functor Hom(−, Q/Z) also does this, so
17 March 3, 2017
17.1 The Hurewicz Theorem
Recall: the natural map S n → K(Z, n) is an n-equivalence, so the induced map of spectra
S0 → HZ is a 0-equivalence. This is equivalent to saying there is a fibration of spectra
fib
F S0 HZ
where π∗≤0 F = 0
F ∧X S0 ∧ X HZ ∧ X coexact,
··· π∗+1 HZ ∧ X
∂∗
π∗ F ∧ X π∗ S0 ∧ X π∗ HZ ∧ X
∂∗
π∗−1 F ∧ X ···
πn (S0 ∧ X) ∼
= πn (HZ ∧ X).
X1 X2 X3 ··· X
A1 A2 A3
We want some sort of tool to compute H∗ X. Ideally, this wouldn’t depend so much on
H∗ (Xn ), but rather on H∗ An . If we apply homology to the iterated gluing diagram, we get
H∗ A1 H∗ A2 H∗ A3 ··· H∗ An H∗ An+1
σn σ
∈
∈
··· H∗ Xn−1 H∗ Xn H∗ Xn+1 ··· H∗ X
an−1
H∗ An−1 H∗ An H∗ An+1
Fact 17.3. The minimal n for which such a class σn appears has non-zero image an−1 .
From this, one can observe that each class σ ∈ H∗ X has a unique place where it appears in
the bottom row.
Now we want to sort out classes in H∗ An−1 from those classes without lifts to H∗ (X).
H∗ An−2 d1
H∗ An−1 d1
H∗ An d1
H∗ An+1
an−2 an−1 an
Pick an an ∈ H∗ An . We want to know: is there some σn+1 ∈ H∗ Xn+1 that hits an ? The
steps illustrated in the diagram are as follows:
Step 2: Because an−1 = 0, σn ∈ ker of the map. By exactness, we can pull back to some
σn−1 ∈ H∗ Xn−1 ; the process continues.
Continue this process until either: (a) one of the σi has nonzero image ai−1 , in which case
there cannot exist σn+1 , or (b) we get to σ0 ∈ H∗ (∗); conclude that this is zero, so there
does exist a σn+1 .
Definition 17.4. The differential d1 is the composite sending, e.g., an to an−1 : push
forward to σn , and push forward again. The r-th differential dr is the composite sending an
to an−r . The claim is that each dr is well-defined up to (co)homology with respect to dr−1 .
This isn’t super difficult; you can work it out yourself.
r r
These (co)homology groups, with respect to dr−1 , are written En,∗ . Each En,∗ is a
r−1
subquotient of En,∗ .
Theorem 17.5. The group H∗ X admits a filtration such that its n-th filtration quotient
∞
is En,∗ , a sort of “limiting page” of the spectral sequence.
Remark 17.6. Things get complicated when you deal with a descending filtration, rather
than an ascending one. You can see this by our argument with the complicated diagram:
this relied on the sequence “bottoming out” eventually. Furthermore, things get complicated
if you use cohomology instead of homology.
We claim the isomorphism of the bottom two rows of the following diagram:
1 1
d1 : E∗,∗ E∗−1,∗−1
L L
h∗+n h∗+n−2
18 March 6, 2017
18.1 Obstruction Theory
19 March 8, 2017
19.1 A Complicated Example
U U
p
Definition 24.2. Let G be a topological group. A G-bundle is a fiber bundle E −−→ B
with fiber G where G acts on E in such a way that
U ⊆ B, ϕU = p−1 (U ) ∼
=G×U
E × F/(eg, f ) ∼ (e, gf ).
Example 24.4. The unitary group U (n) acts on Cn . This gives an assignment
U (n) bundles vector bundles. In fact, is a bijection.
Explicitly, suppose given a U (n)-bundle E.
Proof. This should be evident. One can work it out if one is eager to.
id
Corollary 24.9. The universal bundle, classified by BG −→ BG has contractible total
space, which is often denoted EG.
Remark 24.11. There is a model of K(A, n), which is an actual topological group, so that
BK(A, n) ∼
= K(A, n + 1).
Example 24.12. Computing H ∗ BU (n). We have two fibrations: U (n) → ∗ → BU (n) and
U (n − 1) → U (n) → S 2n−1 .
First, we run the Serre Spectral Sequence on the second fibration: this gives that make
H U (2) ∼
∗ spec-
V
= [e1 , e2 ], the exterior algebra on two generators. Knowing this, we can run the
spectral sequence on the first fibration to compute H ∗ BU (2): one sees that tral se-
quence
H ∗ BU (2) ∼
= Z[c1 , c2 ], dia-
gram
a polynomial algebra on two generators.
One can see by an inductive argument that in general, H ∗ U (n) ∼
V
= [e1 , . . . , en ], and
H ∗ BU (n) ∼
= Z[c1 , . . . , cn ]. The classes ci are called Chern classes.
When G is a finite (discrete) group, there is an especially useful model for BG called
the bar construction. This requires some categorical definitions.
set defined by
N C0 = ob C
N C1 = mor C
N C2 = {pairs of composable maps • → • → • in C }
..
.
N Cn = {strings of n composable arrows • → • → · · · → • in C }.
The face maps di : N Cn → N Cn−1 compose the i-th and (i + 1)-st arrows when 0 < i < n,
and leaves out the first or last arrow if i = 0 or n, respectively. The degeneracy maps
si : N Cn → N Cn+1 take a string
x0 → x1 → · · · → xi → · · · → xn
of n composable arrows and inserts the identity idxi at the i-th spot.
Remark 24.14. The nerve is a functor N : Cat → sSet.
Example 24.15. There are two important examples we can construct. First, we define
the homotopy quotient G//G to be the category with objects indexed by G, and
where there is a morphism xg → xh if and only if there exists an element g 0 ∈ G such
that gg 0 = h. That is, morphisms correspond to multiplication in G.
Similarly, we define ∗//G to be the category with one object ∗, and morphisms
h
∗ −→ ∗ for each h ∈ G, which compose in the expected ways.
Lemma 24.16. The category G//G is “contractible”: any map ∗ → G//G is fully
faithful and essentially surjective.
Remark 24.19. There is a “commutative diagram” where the vertical arrows are equiva-
lences of categories:
G//G ∗//G
n o
G-torsors with
marked point {G-torsors}
A G-torsor is a G-set for which the action is free and transitive. Intuitively, a G-torsor is
“a copy of G with the identity forgotten.” For example, a C-torsor is just a complex line.
ξ
“Recall” that a map X −→ BG is the data of a G-bundle. We have that the correspon-
dence taking a map Sing X → N (∗//G) to ξ is one to one, and bijective on homotopy
classes.
Example 27.2. Take for instance the case ∗ = 1. We are then considering natural
transformations
H 1 (−; F2 ) =⇒ H 1 (−; F2 ).
The collection of all of these is isomorphic to the 1-graded bit of F2 [ι1 ], and thus
contains only
• • • •
• • • •
• • • •
• • • •
The purple classes are all in the kernel, so survive to the degree 1 part:
• • •
• • • •
• • • •
• • • •
• • • •
• • • •
• • • • • • • • •
• • • •
• • • •
• • • •
• • • •
• • • •
• • • •
1. Expand the definition of natural transformation and constant functor to reveal that a
cone is equivalent to the data of an object x ∈ C together with maps fj : x → F (j)
for each object j ∈ J such that for any map g : j → j 0 in the diagram, there is a
commuting triangle
x
fj fj 0
f (g)
F (j) F (j 0 ).
2. Now expand the definition of limit to see that a limit, expressed as an object ` together
with maps hj , has the property that any other cone point x and its maps fj factor
uniquely through a map x → `.
3. Show that the product X × Y is the limit of the diagram with objects X and Y and
no non-identity arrows.
4. Show that the equalizer E of a pair of functions X ⇒ Y of sets is indeed the limit of
that diagram.
3. Show also that your assignments respect composition of natural transformations and
of morphisms.
4. Show that your assignments are mutual inverses, i.e., a natural transformation of
representable functors is exactly the same information as a morphism of representing
objects.
Problem A.1.5. Explain why the usual recipe for forming a group structure on π1 (X, x0 )
does not apply to the relative homotopy object π1 (I, ∂I).
A path lifting function for p is a map λ : Z → E I with λ(e, γ)(0) = e and p ◦ λ(e, γ) = γ.
1. Show that p is a fibration if and only if there is a path lifting function λ for p.
g : F → Pp , f 7→ (f, ω0 )
h : Pp → F, (e, γ) 7→ [λ(e, γ −1 )](1).
Task A.2.2. Skim through Chapter 6 and look at all the proofs we skipped. Try readng a
few. Then try reading a few more. Move on to the rest of the problem set whenever you
like.
Problem A.2.3. Show that if (X, A) is a relative CW-complex, then X/A is a CW-complex.
Given CW-complexes X and Y , use this to concoct appropriate conditions so that X ∧ Y
is a CW-complex.
F
X ×I B.
Problem A.2.5. Suppose X is obtained from A by attaching n-cells {enβ |β ∈ B}. Show
that X/A ∼
= β∈B Sβn , and that the homeomorphism can be chosen so that the diagram
W
iβ W
(S n , ∗) n
β Sβ , ∗
p0
(Dn , S n−1 ) ∼
=
fβ
p
(X, A) (X/A, ∗)
X 7→ hSpectra(Σ∞ X, S) ⊗Z Z(p) .
(b) Demonstrate E(p) ' E ∧ S(p) for any spectrum E. In particular, this gives S(p) ∧ S(p) '
S(p) .
(c) Define πn,(p) E = [Sn(p) , E]. Show that πn,(p) E(p) = πn E(p) .
Problem A.4.2. The 2-adic Bockstein spectral sequence is the filtration spectral
sequence arising from the diagram
2 2
Z∧2 ··· Z Z Z
E1∗,∗ = H ∗ (X; F2 ) ∼
= F2 [w] ⊗ H ∗ (X; F2 ) =⇒ H ∗ (X; Z∧2 ),
M
where the E1 -page consists of many duplicated copies of H ∗ (X; F2 ), which we can think of
as tagged by monomials in w.
(a) Show that the differentials in this spectral sequence are “w-linear”, i.e., dBSS
r (wk x) =
wk dBSS
r (x).
(c) More generally, show that the order of w-torsion of a class on the E∞ page of the
spectral sequence is identical to the 2-primary torsion order of the corresponding
cohomology class in H ∗ (X; Z∧2 ).
Problem A.4.3. Use this spectral sequence to make a calculation of H ∗ (K(Z/2, 2); Z∧2 )
from the calculation of H ∗ (K(Q/2, 2); F2 ) given in class. You will want to know the following
mysterious formula: for any class x ∈ H even (X; F2 ) where dBSSr (x) is defined, we have
Sq 1 (x) · x + Sq |x| Sq 1 (x) for r = 2,
dBSS
r =
dBSS (x) · x for r > 2.
r−1
j p
Problem A.4.4. Let F E B be a fiber sequence, let u ∈ H n (F ; F2 ) be a class
n+1
that transgresses to τ (u) ∈ H (B; F2 ), and suppose that for some integer i ≥ 1 there is a
∗
Bockstein differential dBSS
i v = τ (u). Show that dBSS
i+1 p v is then defined and that
j ∗ dBSS ∗ BSS
i+1 p (v) = d1 (u),
Problem A.4.5. In this problem, you will reinvent one of the main results of unstable ratio-
nal homotopy. For a simply connected space X, we inductively define its rationalization
to be a space Q ⊗ X under X as follows: given a Postnikov fibration
and the rationalization map X[0, n) → (Q⊗X)[0, n), we construct a corresponding Postnikov
fibration for Q ⊗ X as the back face in
K(Q ⊗ πn X, n) K(Q ⊗ πn X, n)
K(πn X, n) K(πn X, n)
(Q ⊗ X)[0, n] ∗
X[0, n] ∗
(Q ⊗ X)[0, n) K(Q ⊗ πn X, n + 1)
X[0, n) K(πn X, n + 1)
Here the nodes X[0, n] and (Q ⊗ X)[0, n] are defined as the total spaces of the pullback
fibrations, and the map between them is induced by the universal map of fibrations. We
set Q ⊗ X to be the homotopy inverse limit
∼
π∗ X Q ⊗ π∗ X π∗ (Q ⊗ X).
2. The cohomology H ∗ (X(n, ∞); Q) as well as its ring structure are completely deter-
mined by the cohomology ring H ∗ (X[n, ∞); Q).
7. Given a Sullivan model for C ∗ (X; Q), its indecomposables compute the rational
homotopy groups of X.
Problem A.4.6. (a) The tensor product of line bundles induces a map
⊗
BU (1) × BU (1) BU (1)
on the object BU (1) representing the functor X 7→ {iso-classes of line bundles on X}.
Describe the behavior of this map in ordinary cohomology with Z coefficients.
(b) In general, the tensor product of vector bundles induces a similar map
⊗
BU (n) × BU (m) BU (nm).
Problem A.4.7. The dual Steenrod algebra is a Hopf algebra, meaning that it not only
has a multiplication map but also a diagonal map ∆ : A∗ → A∗ ⊗ A∗ and an antipode map
χ : A∗ → A∗ . In class, we deduced a formula for ∆, and we showed that as an algebra the
dual Steenrod algebra forms a polynomial ring. The antipode fits into the commutative
diagram
χ⊗1
A⊗A A⊗A
∆ µ
ε η
A F2 A
∆ µ
1⊗χ
A⊗A A⊗A
Problem A.5.2. Return to the picture of the Adams spectral sequence computing π∗ ko
described in class. At a glance, it appears that there could be a potential differential
dr h1 = hr+1
0 . Without assuming Bott periodicity, argue why this differential cannot occur.
(Hint: h0 h1 = 0.)
Problem A.5.3. Compute the first several terms (until you get tired) of a free resolution of
F2 as a module over the Steenrod algebra. Once you have the resolution, use it to compute
Ext and compare your answer with the part of the Adams spectral sequence drawn in class.
Problem A.5.4. Let E(1) denote the exterior F2 -algebra on two generators e1 and e3 , of
degrees 1 and 3 respectively. Calculate Ext∗,∗
E(1) (F2 , F2 ).
Problem A.5.6. Figure out both the statement and the proof of the 5-Lemma and the
Snake Lemma in mod-C homological algebra.
Solution. I won’t state and prove either lemmas in their greatest generality.
A1 A2 A3 A4 A5
α γ δ θ
B1 B2 B3 B4 B5
such that the top and bottom rows are C -exact sequences: ker / im ∈ C . If α, γ, , and θ
are C -isomorphisms, then δ is a C -isomorphism.
0 A B C 0
α β γ
0 A0 B0 C0 0
where the rows are C -short-exact. There is then a long C -exact sequence
δ
0 ker α ker β ker γ coker α coker β coker γ 0.
0 A1 B1 C1 0
f f f
α1 β1 γ1
0 A2 B2 C2 0
α2 β2 γ2
0 A01 B10 C10 0
f f f
δ1
0 ker α1 ker β1 ker γ1 coker α1 coker β1 coker γ1 0
f∗ f∗ f∗ f∗ f∗ f∗
δ2
0 ker α2 ker β2 ker γ2 coker α2 coker β2 coker γ2 0.
ii) for any simplex σ of K 0 , if f 0 (|σ|) meets en0 , then f 0 (|σ|) is contained in en and
f 0 is a linear map when restricted to |σ|.
Proof. The idea is to very finely subdivide K: since |K| is compact, the map g −1 f , g the
characteristic map of en , is uniformly continuous on f −1 en2 , where D2n = {x ∈ Dn : |x| ≤
3/4}, for instance. In particular, there is some δ > 0 such that d(x, y) < δ in f −1 en2 implies
that d(g −1 f (x), g −1 f (y)) < 1/4 in D2n . We can then subdivide K finely enough that no
simplex of K 0 has diameter more than δ.
References
[Hat01] Allen Hatcher, Algebraic topology, 2001.
[Hov98] Mark Hovey, Model categories, vol. 63, American Mathematical Society, 1998.
[Koc96] Stephen O. Kochmann, Bordism, stable homotopy, and adams spectral sequences,
vol. 7, American Mathematical Society, 1996.
81
Math 231br INDEX
pseudo-2-functor, 39
rationalization, 75
reduced cone, 16
relative homotopy group, 19
representable functor, 71
retract, 29
right homotopic, 35
Right Kan extensions, 28
right lifting property, 28
right Quillen functor, 38
set of n-simplices, 27
sheaf, 6, 44
sheaf condition, 6
simplex category, 27
simplicial category, 27
simplicial object, 27
simplicial set, 27
smash product, 11
spectrum, 44
stable sphere, 45
subspectrum, 45
Sullivan minimal model, 76
suspension, 14
reduced, 14
suspension spectrum, 44
universal bundle, 61
weak equivalence, 19
weak equivalences, 29
weak factorization system, 29
weak topology, 79
wedge product, 10
wide subcategory, 29