Exercises in Alg Top

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Algebraic Topology Problems

Ethan Lake
February 19, 2016

Problem 1. Construct an explicit deformation retraction of the torus with one point deleted
onto a graph consisting of two circles intersecting in a point, namely, longitude and meridian
circles of the torus.
The idea is to pull the initial hole in the torus so that it becomes as big as possible. To
do this explicitly, we write the punctured torus as a square with vertices at p˘1, ˘1q with
opposite sides identified at the point p0, 0q cut out. We enlarge the puncture by making it
a square hole that grows over time. We accomplish this with the maps ft , defined by
#
px ` tp1 ´ xq, yq if |x| ą |y|
ft px, yq “ (1)
px, y ` tp1 ´ yqq if |y| ą |x|

As a check, we see that f0 px, yq “ px, yq and fq px, yq is always somewhere on the boundary
of the square. We see that ft makes a square hole in the center of the punctured torus that
grows until t “ 1, at which point we’re left with the boundary of a square with opposite
sides identified, which is precisely a pair of circles joined at a point (which is p´1, ´1q).
Problem 2. Show that if a space X weakly deformation retracts to a subspace A Ă X via
the homotopy ft : X Ñ X, then the inclusion i : A ãÑ X is a homotopy equivalence.
By the definition of homotopy equivalence, we just need to show the existence of a map
F : X Ñ A such that F i » 1A and iF » 1X . This is easy, since we can just let F “ f1 .
Note that we are free to regard f1 as a map from A to A, since at time 1 the image of X
under f is completely contained within A.
F i is homotpic to the identity on A since we can construct the homotopy explicitly: let
ht pxq be given by ht pxq “ ft ipxq, which is just ft pxq restricted to A. Since h0 pxq “ f0 |A “
1X |A “ 1A , h establishes the homotopy between F i and 1A .
Secondly, we need to show that iF is homotopic to 1X . Let j : f1 pXq ãÑ X be an
inclusion of wherever X ends up back into X. We then introduce ht pxq by defining ht pxq “
ft j. We see that h0 “ 1X since so is f0 , and so h exhibits the homotopy equivalence we
were looking for.
Problem 3. If a space X deformation retracts to a point x0 P X, show that for each
neighborhood U of x0 in X, there exists a neighborhood V Ă U of x0 such that the inclusion
V ãÑ U is null-homotopic (i.e., homotopic to a constant map).
The idea is (presumably) to massage the problem to a point where we can use the result
of the last one. Let ft be the deformation retract of X to x0 , with f0 “ 1X and f1 “ x0 .
From the last problem, the inclusion map i : x0 ãÑ X establishes the relations f1 i » 1x0
and if1 » 1X .

1
Let U be a neighborhood of x. Also, let F be the homotopy of the (included) deformation
retract, i.e. a homotopy satisfying F0 “ 1X and F1 “ if1 (the reason for the inclusion is
to force F to map from X to X). Since the deformation retract is continuous and we can
do some smooth wiggling of Ft if needed, there is some time tin P I such that Ft´1 pU q ă U
for all times t ě tin and Ft´1 pU q “ U for all times before tin . We’re going to choose the
neighborhood V asked for by the problem statement as V “ Ft´1 in
pU q ă U .
Now, rescale time by tin by defining Frt “ Fttin |U , with t P r0, 1s. We can smoothly
wiggle this homotopy so that Frt pV q Ă V for all time. Also, we see that Fr1 pXq Ă V and
Fr0 “ 1U , so that Fr is one of these “weak” deformation retracts form the last problem.
Using the last problem, we see that the inclusion j : x0 ãÑ V is a homotopy equivalence, so
that Fr1 j » 1x0 . But of course if something is homotopic to the identity map on a single
point set, then that thing is null-homotopic.
Now we just need to move this one step up to the inclusion V ãÑ U . But that’s easy,
since the inclusion i : V ãÑ U is homotopic to 1V , exhibited by the original homotopy f . So
then we chain these two inclusions together, we see that i : V ãÑ U must be null-homotopic
as well.
Problem 4. a) Let X be the subspace of R2 consisting of the orizontal segment r0, 1s ˆ t0u
glued to the vertical segments tru ˆ r0, 1 ´ rs for r a rational number between 0 and 1. Show
that X retracts to any point in r0, 1s ˆ t0u, but not to any other point.
b) Let Y be the subspace of R2 formed by stacking an infinitely long chain of combs so
that they tile like triangles. Show that Y is contractible but does not deformation retract
onto any point.
c) Let Z be the zigzagging feature formed by the bases of all the combs in Y . Show that
there is a weak DR from Y onto Z, but not a regular one.

aq That X DRs to any point in r0, 1s ˆ t0u is pretty clear – we retract each prong of
the “comb” (which is fine, just map each prong to its base point), and then contract the
resulting line segment to a point. Totally fine. If we try to DR to some point not on the
base of the comb, we run into a problem. Pick some x0 on one of the comb’s prongs. Since
we’re thinking about X as living inside R2 (and not Q2 ), any neighborhood U of x0 must
contain many (in fact, infinitely many) different prongs of the comb, and so U must be
(very) disconnected in X.
Suppose a DR retract from X to x0 exists. Then from the last problem, we can choose
a neighborhood V Ă U of x0 such that the inclusion i : V ãÑ U is null-homotopic. Now
arguing similarly to before, we see that V must be just as badly disconnected as U , and
thus the inclusion of V into U cannot possibly be homotopic to a constant – contradiction!
bq That Y is contractible is pretty clear – we first act with an (infinite) number of
homotopies that shrink the prongs on every comb. After doing this we’re just left with a
copy of R, which we can then contract without issue.
Y doesn’t DR onto any point for the same reason that the single comb didn’t DR onto
any point on one of its prongs. If y P Y is a point on a prong, then we apply our answer in
part a directly. If y is a point on the base of a comb we can still use the disconnectedness
argument since a neighboring comb will always have lots of prongs close to y (which is clear
from the picture in Hatcher).
cq A weak retraction from Y onto Z is done simply by shrinking the prongs on every
comb. Clearly we can’t get a stronger retraction, since if we did we could then DR Z to a
point, in disagreement with part b.

2
Problem 5. Show that a retract of a contractible space is contractible.
Intuitively, this is pretty obvious. Let X be a contractible space and f : X ˆ I Ñ X be
the homotopy that squishes X to a point. Furthermore, let r : X Ñ Y be a retract. Then
g : I ˆ Y Ñ Y with g “ rf |IˆY tells us how to squish Y to a point.
Problem 6. Show that f : X Ñ Y is a homotopy equivalence if there exist maps g, h : Y Ñ
X such that f g » 1 and gf » 1. More generally, show that f is a homotopy equivalence if
f g and hf are.
Suppose there exist g, h as in the problem statement. Hit f g » 1 with h:

h » h1Y » hpf gq » phf qg » g (2)

and so 1Y » f g » f h, implying that f is a homotopy equivalence.


More generally, suppose that f g and hf are homotopy equivalences. Then there exist
maps k and l such that
kf g » f gk » hf l » lhf » 1. (3)
Now precompose lhf » 1 with gk and then use f gk » 1, getting lh » gk. But then since
f gk » 1, we have that f lh » 1, and so f is indeed a homotopy equivalence.
Problem 7. Show that S 8 is contractible.
Isn’t this easy? Let N~ be the north pole of S 8 (i.e, N
~ “ p1, 0, 0, . . . q) and let ~x be a
8
vector on the surface of S . Then form the maps
~
p1 ´ tq~x ` tN
ft p~xq “ . (4)
~
||p1 ´ tq~x ` tN ||
The denominator is there just to ensure that we stay on S 8 the whole time. f0 is the
identity on S 8 while f1 is a constant map, and so S 8 is indeed contractible.
Problem 8. Show that S m ˚ S n “ S n`m`1 (the S 1 ˚ S 1 example is familiar from physics
and Hopf stuff ).
The way to tackle this is to think about the join of X and Y as X ˆ Y ˆ I (I “ r0, 1s),
with X smashed to a point at the beginning of the interval, Y smashed to a point at the
end of the interval, and the full X ˆ Y preserved at the middle of the interval. If we start at
the middle of the interval and move backwards in time, X slowly shrinks to a point, while
Y just hangs around. So then the first half of the join looks like CX ˆ Y . Similarly, the
second half looks like X ˆ CY .
We can apply this to the problem with the spheres if we notice that CS m “ Dm`1 , so
that
S m ˚ S n “ pDm`1 ˆ S n q Y pS m ˆ Dn`1 q, glued along S n ˆ S m . (5)
Then we use S n “ Dn {BDn :
` ˘ ` ˘
S m ˚S n “ Dm`1 ˆ pDn {BDn q Y pDm {BDm q ˆ Dn`1 , glued along Dn {BDn ˆ Dm {BDm .
(6)
We then use that the product of an m-disk with an n-disk is topologically the same as an
m ` n disk, yielding

S m ˚ S n “ Dn`m`1 {BDn`m`1 “ S n`m`1 . (7)

3
Problem 9. Show that a CW complex is contractible if it is the union of two contractible
subcomplexes whose intersection is also contractible.
Let X and Y be the two contractible subcomplexes in question. First contract their
intersection. This leaves the part of X that wasn’t in Y connected at a point to the part of
Y that wasn’t in X, i.e. X{pX X Y q ^ Y {pX X Y q. Now modding out by a contractible space
is a homotopy equivalence, so both X{pX X Y q and Y {pX X Y q are still contractible. But
of course the wedge of two contractible spaces is contractible – first contract one without
touching the other (ending up with just the other), and then contract the other. After doing
this, the whole X{pX X Y q ^ Y {pX X Y q thing has been contracted, and so X Y Y must
have been contractible to start with.

Problem 10. Show that the change-of-basepoint homomorphism βh depends only on the
homotopy class of h.
Since all homotopic loops based at x0 are regarded as the same element in π1 pX, x0 q,
we need only show that if βh and βg are two change-of-basepoint homomorphisms, then
h ¨ f ¨ h̄ » g ¨ f ¨ ḡ if h » g. Assume h » g. Then we actually just need to show that
f ¨ h̄ » f ¨ ḡ. But this is true only if h̄ » ḡ, since the pre-composition by f changes the
homotopy class of both sides of the equation in the same way. But of course h̄ » ḡ if h » g
(just compose both sides with h and then precompose with g), and so h ¨ f ¨ h̄ » g ¨ f ¨ ḡ as
claimed. Am I cheating here at all? I don’t think so.
Problem 11. Show that π1 pXq (for path-connected X) is abelian iff the change-of-basepoint
homomorphisms βh only depend on the endpoints of h.
First, suppose π1 pXq is abelian. Let f P π1 pX, x0 q be some loop based at x0 , and
consider two different change-of-basepoint homomorphisms βh , βg which are maps βh , βg :
π1 pX, x0 q Ñ π1 pX, x1 q for some point x1 . Note that h̄¨g is also in π1 pX, x0 q (using the weird
first-symbol-goes-first convention). Then since π1 pXq is abelian, we have f ¨ph̄¨gq “ ph̄¨gq¨f .
Now hit both sides of this with ḡ ¨ h from the left:

ḡ ¨ h ¨ f ¨ h̄ ¨ g “ f (8)

Rearranging, we get h ¨ f ¨ h̄ “ g ¨ f ¨ ḡ (abusing notation a bit), which tells us that the


change-of-basepoint homomorphisms only care about their endpoints.
Now, suppose that the change-of-basepoint homomorphisms only care about their end-
points. Then we can assume h ¨ f ¨ h̄ “ g ¨ f ¨ ḡ. Undoing all the steps in the last paragraph,
we get that f ¨ ph̄ ¨ gq “ ph̄ ¨ gq ¨ f . But f was chosen arbitrarily, so was x1 , and both h and
g were arbitrary as well. So then with suitable choices of these guys, we can generate any
pair of loops in π1 pX, x0 q. But we just saw that these two loops commute in π1 pX, x0 q! So
π1 pX, x0 q is indeed abelian.

Problem 12. Show that for a space X, the following conditions are equivalent:
a) Every map S 1 Ñ X is homotopic to a constant
b) Every map S 1 Ñ X extends to a map D2 Ñ X
c) π1 pX, x0 q is trivial for all x0 P X.
Show that X is simply connected iff all loops are homotopic.

aq ùñ cq: Suppose every map S 1 Ñ X is homotopic to the identity map. Since two
loops based at x0 are regarded as the same element in π1 pX, x0 q if they are homotopic, then

4
all loops in X (i.e., all maps S 1 Ñ X) must correspond to the same element in π1 pX, x0 q.
Of course the constant map is invertible, so π1 pX, x0 q must be the group with only one
element. cq ùñ aq is essentially the same argument – if there is only one element in
π1 pXq, then there can only be one homotopy class of loops based at any x0 P X. This must
be the homotopy class of the identity loop, which is always in π1 pXq. So indeed, any loop
in X (i.e., a map S 1 Ñ X) must be homotopic to the constant map.
aq ùñ bq: Suppose that every map S 1 Ñ X is homotopic to the identity map. Then
define the map ρ : D2 Ñ S 1 ˆ I by ρpreiθ q “ peIθ q, takingD2 to be the unit disk. Compose ρ
with the map F ”S 1 ˆI Ñ X with F ps, 0q “ x0 and F ps, 1q “ f psq for some map f : S 1 Ñ X.
Then f 1 “ F ˝ ρ is continuous and f 1 peiθ q “ f peiθ q, i.e. f 1 and f agree on S 1 . So f 1 is the
extension we were looking for.
Now for bq ùñ aq. Assume every map f : S 1 Ñ X extends to a map D2 Ñ X. D2 is
contractible, so extend f to a map D2 Ñ X, and then compose f with the contraction of
the disk to a point. This gives us a homotopy from f to the constant map, and so since f
was arbitrary we must have that every map S 1 Ñ X is homotopic to the identity map.
Now suppose that X is simply connected. Then as established above, any map S 1 Ñ X
must be homotopic to a constant. Of course, all constant maps are homotopic in X, since
X is path-connected by assumption. So then all maps S 1 Ñ X are homotopic.
On the other hand, suppose all such maps are homotopic. A homotopy between two
constant maps is just a path between two points, and so X must be path-connected. We
showed above that π1 pXq must therefore be trivial. So since X is path-connected and has
trivial fundamental group, it must be simply connected.
Problem 13. Does Borsuk-Ulam hold for the torus?
Nope! The counterexample is the first one you think of – embedding the torus in R3 and
then projecting away its z coordinates. The point p´x, ´yq on the torus is π radians away
from px, yq on both meridians, and so since the hole in the torus has finite size, the image
of p´x, ´yq must necessarily be different from the image of px, yq (this fails for the sphere
since the north and south poles end up going to the same point under projection).
Problem 14. Since π1 pX ˆ Y, px0 , y0 qq – π1 pX, x0 q ˆ π1 pY, y0 q, it follows that loops just in
X commute with loops just in Y in π1 pX ˆY, px0 , y0 qq. Show this with an explicit homotopy.
Let fX ˆ 1y0 and 1x0 ˆ fY be loops in X ˆ ty0 u and tx0 u ˆ Y , respectively. Their
concatenation is pfX ˆ 1y0 q ¨ p1x0 ˆ fY q, which we can alternately write as pfX ¨ 1x0 q ˆ p1y0 ¨
fY q. We can homotop each term in parenthesis without changing how they commute, and
obviously fX ¨ 1x0 » 1x0 ¨ fX and fY ¨ 1y0 » 1y0 ¨ fY , which means we can write

pfX ¨ 1x0 q ˆ p1y0 ¨ fY q » p1x0 ¨ fX q ˆ pfY ¨ 1y0 q. (9)

We then rewrite the RHS as p1x0 ˆ fY q ¨ pfX ˆ 1y0 q, and so the two loops we started with
commute in π1 pX ˆ Y, px0 , y0 qq.
Problem 15. Show that there are no retractions r : X Ñ A in the following cases:
a) X “ R3 and A any subspace homeomorphic to S 1 .
b) X “ S 1 ˆ D2 with A the boundary torus S 1 ˆ S 1 .
c) X “ S 1 ˆ D2 with A the interlocking circle thing shown in the book.
d) X “ D2 _ D2 with A its boundary S 1 _ S 1 .
e) X “ D2 with two points on its boundary identified and A its boundary S 1 _ S 1 .
f ) X the Mobius strip and A its boundary circle.

5
We’re going to use proposition 1.17: if a space X retracts onto a subspace A, then the
homomorphism π1 pAq Ñ π1 pXq induced by the inclusion i : A ãÑ X is injective.
a) π1 pR3 q is trivial, but π1 pS 1 q isn’t, and obviously there is no injection Z Ñ 0.
b) π1 pS 1 ˆ S 1 q “ Z2 (two non-contractible loops) while π1 pS 1 ˆ D2 q “ Z (one non-
contractible loop), and there is no way to inject Z2 Ñ Z.
c) The two fundamental groups are the same, since π1 pXq “ Z “ π1 pAq (since A is
homeomorphic to S 1 ). However, the map i˚ : π1 pAq Ñ π1 pXq is actually trivial, not
injective. This is because i˚ maps a loop that goes around A once to a loop that goes
around the S 1 part of X once and then goes around the S 1 part of X in the opposite
direction, which is homotopic to not looping around anything. So the generator of π1 pAq
gets sent to 0, and i˚ can’t possibly be injective.
d) Since D2 is contractible, so is D2 _ D2 , and so π1 pXq “ 0. But as we mentioned in
class, π1 pS 1 _ S 1 q is the free group on 2 generators, which is huge. Obviously there is no
injection from this group into 0.
e) We can write X in R2 as the unit circle with its leftmost point rightmost point
identified, and then homotop it onto an interval with the endpoints identified by squishing
away its height in the y direction. So then π1 pXq “ Z. As mentioned in the last part,
π1 pS 1 _ S 1 q is the free group on two generators. Let g and h be these two generators,
and let n and m be their images under the homomorphism i˚ : π1 pS 1 _ S 1 q Ñ Z. Then
i˚ pg m q “ mi˚ pgq “ mn “ ni˚ phq “ i˚ phn q. But g m and hn are different elements, so i˚
can’t be an injection.
f) The Mobius strip deformation retracts onto its boundary circle, so the two have the
same fundamental group (namely Z). However, a loop that goes once around the boundary
circle once goes around the Mobius strip twice, and so the injection i˚ : π1 pS 1 q Ñ π1 pXq
must be the map i : n ÞÑ 2n. So if a retraction r existed, then we would require r˚ i˚ pnq “ n.
Since we already said that i˚ pnq “ 2n, we have i˚ p1q “ 2, and so we’d require that r˚ p2q “ 1.
But we can use 2 “ 1 ` 1 to write 2r˚ p1q “ 1, i.e. r˚ p1q “ 1{2. But 1{2 R π1 pS 1 q, and so no
retraction exists.
Problem 16. Construct infinitely many non-homotopic retractions S 1 _ S 1 Ñ S 1 .
We can do essentially anything we with one of the S 1 s (say, the left one), as long as
we don’t touch the right one. Consider the retraction that folds the left S 1 onto the right
S 1 after twisting it n times. These guys provide the infinitely many (one for each n) non-
homotopic retractions we’re looking for. To be more precise, let the left S1 be a unit circle
centered at the point p´2, 0q in R2 and the right S 1 be the unit circle centered at the origin.
Then the rectractions described above take the form
pnq
rt : p´2 ` cos θ, sin θq ÞÑ pcosptnθq, sinptnθqq. (10)
rpnq isn’t homotopic to rpmq if n ‰ m since they are identified with different elements in
π1 pS 1 q “ Z.
Problem 17. Let X Ă Rm be the union of convex open sets X1 , . . . , Xn such that Xi X
Xj X Xk ‰ H for all i, j, k. Show that X is simply-connected.
Obviously convex sets are path-connected, and so an overlapping collection of convex
sets (like X) will also be path-connected. Since simply-connected means “path-connected
+ trivial fundamental group”, we just have to compute π1 pXq. Since Xi X Xj X Xk ‰ H,
we can use Van Kampen’s theorem if we can show the existence of some x0 P Xi X ¨ ¨ ¨ X Xn
to use as the basepoint.

6
Of course, there is a basepoint common to all the Xi ’s when n “ 3. Now suppose n ą 3
and that X1 X ¨ ¨ ¨ X Xn´1 ‰ H. If X1 X ¨ ¨ ¨ X Xn “ H, then there would be at least one Xi
with i ă n such that Xn X Xi “ H, which contradicts our assumption about the Xi ’s. So
a common basepoint for use in Van Kampen’s theorem exists.
Van Kampen then tells us that π1 pXq » ˚i π1 pXi q{N , where N is the normal subgroup
generated by guys of the form ijk pωqikj pωq´1 where ω P π1 pXj XXk q and ijk : π1 pXj XXk q Ñ
π1 pXj q. But of course all the π1 ’s in the last sentence are trivial, due to the convexity of
each Xi . So then π1 pXq is trivial, and X is simply-connected.
Problem 18. Let X be a union of n lines through the origin in R3 . Compute π1 pR3 zXq.
Let Y “ R3 zX. Obviously, when n “ 1, πa pY q is the free group on a single generator.
When n “ 2, X looks like an X. At first glance, the four legs of the X would seem to imply
that π1 pY q is the free group on four generators. However, looping around the first three legs
of the X is equivalent to looping around the fourth, and so π1 pY q is actually just the free
group on three generators. This works in general: for n lines we will have 2n legs to loop
around, but looping around 2n ´ 1 legs is the same as looping around the 2nth leg, and so
the 2nth leg doesn’t give us anything new. Summarizing, this means that π1 pY q “ Z˚p2n´1q .
Problem 19. Let X be the quotient space obtained from S 2 by identifying the north and
south poles to a point. Define a cell complex structure on X, and use this to compute π1 pXq.
X is the same as a torus with one of the non-contractable loops (1-cells) squished to
a point. So we can just take the standard cell complex structure on the torus (page 5 of
Hatcher) and get rid of one of the 1-cells, giving a cell complex consisting of a 0-cell, a 1-cell,
and a 2-cell. The 2-cell is glued along the 1-cell x by gluing along x and then along x´1
(x in the opposite direction). So then π1 pXq “ xxy “ Z, which is obvious if we just think
about what X looks like.
Problem 20. Okay, I’m not going to copy it down. This is problem 1.2.9 of Hatcher.
First, let’s show that Mg doesn’t retract onto C. If Mg did retract onto C, then we
could find an injection i˚ : π1 pCq Ñ π1 pMh1 q. We need to show that this is impossible. As
suggested, we try to abelianize π1 pMh1 q. For a moment, forget about the hole in Mh1 caused
by cutting out C. Then Mh1 is just a genus h surface, which we know we can write out as a
big polygon with 4h sides, labelling them by x1 , y1 , x´1 ´1 ´1
1 , y1 , x2 , . . . , yh . Taking a trip all
the way around the polygon takes us back to the identity, so we see that π1 pMh1 q is presented
as
h
ź
π1 pMh1 q “ xx1 , y1 , . . . , xh , yh | xi yi x´1 ´1
i yi y. (11)
i“1

When we cut C out though, there’s a puncture in Mh1 that we need to include in the
fundamental group. Instead of a big polygon, Mh1 will look like a big polygon with a
śh
puncture in the middle. So then doing i“1 xi yi x´1 ´1
i yi is actually not doing the identity
– it’s doing a, where a is the loop around the puncture. So actually,
h
ź
π1 pMh1 q “ xa, x1 , y1 , . . . , xh , yh |a xi yi x´1 ´1
i yi y. (12)
i“1

Now consider abelianizing π1 pMh1 q. We can add in the loop around the puncture any time
during our journey along the polygon, so presumably a commutes with the other generators.

7
Since abelianizing is just modding out by commutators, we need to mod out π1 pMh1 q by the
relations x1 y1 x´1 ´1 ´1 ´1
1 y1 “ 1, . . . , xh yh xh yh “ 1. In light of the last display-style equation,
ab 1
this means that a “ 1 in π1 pMh q.
Now j˚ : π1ab pMh1 q Ñ π1 pMh1 q is certainly injective, and so if i˚ : π1 pCq Ñ π1 pMh1 q is
to be injective then k˚ : π1 pCq Ñ π1ab pMh1 q better be injective as well, by virtue of the
composition i˚ “ j˚ ˝ k˚ . But k˚ isn’t injective, because π1 pCq “ xay, and a gets sent to
1! Ta-da! Evidently this doesn’t work for C 1 since then we get two extra loops, neither of
which individually gets sent to 1.
The next thing to do is to show that we can still retract Mg onto C 1 . But this isn’t so
bad – first smash Mg flat so that it looks like the wedge of a bunch of S1 ’s (one of which is
C 1 ), and then contract all the S1 ’s besides C 1 . (This wouldn’t have worked for C since C
would have been squashed during this process)
Problem 21. Let X be the quotient space obtained from the cube by identifying its opposite
faces in a way such that when a face is passed through, space re-orients itself by a right-
handed π{2 rotation. Show that π1 pXq “ Q8 .

Actually, this isn’t as bad as it seems. Obviously the cell complex has one 3-cell (un-
changed during the quotient) and three 2-cells (3 “ 6{2). Now define a group operation
in this space as follows: center the cube at p0, 0, 0q and let the vertices of the cube be at
p˘1, ˘1, ˘1q{2. Let i, j, k be group elements that act on stuff inside the cube by translating
it by one unit in the x, y, or z direction, respectively. Consider the action of i on the 1-cell
that is the line segment from p1{2, ´1{2, ´1{2q to p1{2, 1{2, ´1{2q. The action of i sends
this 1-cell to an orthogonal (pointing up) line segment on the opposite side of the cube.
Doing i again sends the 1-cell to a translation of its initial position by 1 unit upwards in
the z direction. Doing it a third time gives the other orthogonally-situated 1-cell on the
opposite side of the cube, and doing it a final time returns it to where it started. By doing
this procedure with 1-cells orthogonal to the y and z axes (as opposed to the x axis) and
acting with j and k, we see that we can pick up all the original 12 1-cells of the cube in this
manner. So since we had three orbits of 1-cells, the number of 1-cells in the quotient space
is 12/3 = 4. Finally, we notice that by tracking the end of a 1-cell throughout the 1-cell’s
orbit, any given 0-cell in the original cube will get identified with any other 0-cell exactly
two steps away from it on the lattice. Since the cubic lattice can be colored in a bipartite
way, we see that there are only two distinct orbits of 0-cells under the group action, and so
there are only two 0-cells in the quotient space.
This kind of thinking actually tells us what π1 pXq is. We can imagine constructing
elements of π1 pXq by fixing the end of a hose with an asymmetric tube at the origin and
moving the mouth of the hose by translating it by integer amounts in each direction. For
example, if we acted on the mouth of the hose with i, it would loop around the x-component
of the box and the mouth would end up at the origin rotated by π{2 with respect to the
end of the hose. Of course if we did this four times, we wouldn’t have done anything since
the mouth of the hose would look the same as the end of the hose. So in π1 pXq, i4 “ 1.
The same applies to j and k. We also see that the mouth of the hose ends up in the same
orientation under i2 , j 2 , and k 2 , and so we identify each of these guys as the symbol ´1.
The remaining relations ij “ k, ji “ ´k, etc, are just derived by looking at what the mouth
of the hose looks like after repeated translations of the hose mouth through different faces
of the cube. I’ll add pictures if I have time! In any case, the fact that this ends up being
the quaternions is expected, since we know that there is a bijection between the quaternions
and the construction of Euler angles, which are essentially what we used in this dumb hose

8
argument.
Problem 22. Let p : Xr Ñ X be a covering space with p´1 pxq finite and nonempty for all
x P X. Show that X is compact Hausdorff iff X
r is compact Hausdorff.

Let’s do the compactness part first. Suppose X is compact. Choose a finite cover of
X and map it down to X with p – this gives us a finite cover of X. On the other hand,
suppose X is compact. Then choose a finite cover of X and hit it with p´1 , which maps to
up a finite cover of X,
r since the fibers of any point in X are finite.
Now suppose X is Hausdorff, and pick two points a, b P X. r If ppaq ‰ ppbq, then we can
construct two disjoint open sets around ppaq and ppbq and map them back up to get disjoint
sets around a and b (the images of these guys must be disjoint for p to be well-defined). On
the other hand, if ppaq “ ppbq then a and b are different points on the same fiber. But the
definition of a covering space tells us that I can pick disjoint open neighborhoods around a
and b, and so X r must be Hausdorff. Finally, let X r be Hausdorff. Pick a, b P X. Let tAi u
and tBi u be two collections of open sets which enumerateŞ disjoint
Ş neighborhoods around
the fibers of a and b, respectively. The claim is that i Ai and i Bi provide the disjoint
open neightborhoods of a and b that we’re looking for. This would not be true only if
there was some point c P X in all of the Ai ’s and all of the Bi ’s. The only way for this to
happen without contradicting the disjointness of the Ai ’s and Bi ’s would be if none of the
Ai ’s were contained on the same U -sheet as any of the Bi ’s, where U is some neighborhood
containing a and b. This can’t happen since the number of fibers at points in X must be
locally constant since Xr is a covering space, and so Ş Ai and Ş Bi give us the desired
i i
disjoint neighborhoods of a and b.

Problem 23. Show that if a path-connected, locally path-connected space X has finite fun-
damental group, then all maps from X to S 1 are nullhomotopic.
The assumptions about X mean that we can apply the lifting criterion. R is a cover for
S 1 , and so we are looking for a lift from some map f : X Ñ S 1 to a map fr : X Ñ R. This
will always exist, since the only subgroup of Z “ π1 pS 1 q being trivial implies that we will
always have f˚ pπ1 pXqq Ă p˚ pπ1 pRqq. But of course R is contractible, and so if we do fr and
then project down on to S 1 we get a nullhomotopy of f .
Problem 24. Construct a simply-connected covering space of the space X Ă R3 that is the
union of a sphere and a diameter. Do the same when X is the union of a sphere and a
circle intersecting it in two points.
For the sphere connected by a diameter, we obviously have π1 pXq “ Z, since this is just
a torus with one of the non-contractible loops made contractible by squeezing it to a point.
So then the covering space should be a long 1D chain thing, in keeping with the examples
from the book. We can see that one choice which works is a chain of spheres linked by
lines (like beads on a necklace). The lines represent passing through the diameter, while the
spheres ensure that after you pass through the diameter, you have an S 2 ’s worth of room
to move around in. Obviously this type of thing is simply-connected.
For the union of a sphere and a circle intersecting it in two points, I argue that for our
purposes, this is the same thing as a sphere with a diameter and a thin 1D handle, and
that the fact that the diameter and handle hit the sphere at the same points makes no
difference in terms of homotopy groups. Furthermore, if we’re just interested in homotopy
stuff, a diameter is the same thing as a thin 1D handle, so really this is just a sphere with

9
two thin 1D handles. So if we forget about the sphere part, when constructing a covering
space with trivial fundamental group we would be prompted to do something similar to the
infinite Cayley graph thing talked about in Hatcher. To take into account the sphere part,
we just need to replace each lattice point in the fractal Cayley graph thing by a sphere (see
drawing).

Problem 25. Let a and b be the two generators of π1 pS 1 _ S 1 q. Draw a picture of the
covering space corresponding to the normal subgroup generated by the relations a2 “ b2 “
pabq4 “ 1.
Since a2 “ b2 “ 1, we need to have a and b label the sides of circles a la page 58 of
Hatcher. So if we just had these relations, we’d be fine with an infinite chain of alternating
a and b circles. But this chain needs to terminate after four applications of ab, each of which
moves us two circles down the chain. So we take the infinite chain and quotient it out every
8 lattice spaces (after 4 a circles and 4 b circles), which gives a loop of 4 alternating a-circles
interspaced with 4 alternating b-circles.
Problem 26. Find all the connected covering spaces of RP 2 _ RP 2 .

According to Hatcher (page 74), π1 pRP 2 q “ Z2 . I think this means that there are only
two connected covering spaces of RP 2 (up to isomorphism), since Z2 has two subgroups.
These correspond to covering by a single copy of RP 2 and a double covering by a copy
of S 2 . When we wedge things together, the fundamental group of the wedge is equal
to the free product over the fundamental groups of each thing in the wedge (like how
π1 pS 1 _ S 1 q “ Z ˚ Z), and so π1 pRP 2 _ RP 2 q “ Z2 ˚ Z2 . Presumably this means that
to construct connected covering spaces, we just have to wedge together copies of the two
different connected covers of RP 2 , namely RP 2 itself and S 2 . But we also have to make
the points where the wedging happens look locally like the attaching point of the wedge
RP 2 _ RP 2 . This means that we can’t wedge together three RP 2 ’s in a row, since then the
wedge points would get identified with too many other points. Essentially, when we do the
wedging out of RP 2 ’s and S 2 ’s, we need to have two points of each S 2 and RP 2 wedged to
something else. But for the copies of RP 2 this means that they can only be wedged to one
other thing, since the antipodal identification means that the one wedge point is actually
two. So all the connected covering spaces can be enumerated by listing the number of ways
to wedge together a bunch of RP 2 ’s and S 2 ’s such that each S 2 is wedged to two different
things and each RP 2 is wedged to one different thing. There actually aren’t that many ways
to do this! (see back of paper)

10

You might also like