Covering Spaces

Some pictures to keep in mind when it comes to covers and path lifting:

Picture to keep in mind

A more complicated situation

Useful Facts

remark:

When covering spaces are involved in any way, try computing Euler characteristics - this sometimes yields nice numerical constraints.

fact (Euler characteristics are multiplicative on covering spaces):

For p:AB an n-fold cover, χ(A)=nχ(B).

fact:

Covering spaces of orientable manifolds are orientable.

fact:

The preimage of a boundary point under a covering map must also be a boundary point

fact:

Normal subgroups correspond to normal/regular coverings, where automorphisms act freely/transitively. These are “maximally symmetric”.

Universal Covers

definition (Galois/normal/regular covers):

A covering ˜XpX is Galois (or normal/regular) over (X,x0) iff Deck(˜X) acts transitively on the fibers: for any two lifts ˜x1,˜x2˜X of x0X, there is a ψDeck(˜X) with ψ(˜x1)=˜x2.

proposition (Fundamental theorem of covering spaces, Hatcher 1.39):

If X is

  • Path-connected,
  • Locally path-connected, and
  • Semilocally simply connected,

then X admits a universal cover ˆXX: if CqX is any other covering map with C connected, then there exists a covering map ˜p:ˆXC making the following diagram commute:

Link to diagram

That is, any other cover C of X is itself covered by ˆX. Note that by the universal property, ˆX is unique up to homeomorphism when it exists.

Moreover, letting ˜XX be an arbitrary path-connected cover and H:=pπ1(˜X;˜x0),G:=π1(X;x0),

  • The ˜XX is Galois iff H  G,
  • Deck(˜XX)NG(H)/H where N is the normalizer in G of H,
  • Deck(˜XX)G/H if ˜XX is Galois,
  • Deck(ˆXX)G.
remark:

A Galois cover ˜XX with Deck(˜X)=G is equivalently a principal G-bundle for a discrete G:

remark:

Covering spaces of X are classified by subgroups of π1(X): {Covering spaces ˜XX}/Isomorphisms over X{Subgroups Hπ1(X)}/Conjugacy{Galois covering spaces ˜XX}{Normal subgroups H  π1(X)}

theorem (Homotopy lifting property for covers, Hatcher 1.30):

Let p:˜XX be any covering space, F:Y×IX be any homotopy, and ˜F0:Y˜X be any lift of F0. Then there exists a unique homotopy ˜F:Y˜X of ˜F0 that lifts F:

Link to diagram

theorem (Lifting criterion for covers, Hatcher 1.33):

If f:YX with Y path-connected and locally path-connected, then there exists a unique lift ˜f:Y˜X if and only if f(π1(Y))π(π1(˜X)):

Link to diagram

Moreover, lifts are unique if they agree at a single point.

remark (Automatic lifts):

Note that if Y is simply connected, then π1(Y)=0 and this holds automatically!

proposition (Covering spaces induce injections on π1, Hatcher 1.31):

Given a covering space ˜XpX, the induced map p:π1(˜X)π1(X) is injective. The image consists of classes [γ] whose lifts to ˜X are again loops.

theorem (Fundamental theorem of covering spaces, Hatcher 1.39):

For ˜XpX a covering space with

  • ˜X path-connected,
  • X path-connected and locally path-connected,

letting H be the image of π1(˜X) in π1(X), we have

  • ˜X is normal if and only if H  π1(X),

  • For the normalizer NG(H) where G:=π1(X), G(˜X):=AutCov(X)(˜X)NG(H)H.

In particular, ˜X normalG(˜X)π1(X)/HˆX universalG(ˆX)π1(X).

fact:

There is a contravariant bijective correspondence {Connected covering spacesp:˜XX}/{Conjugacy classes of subgroupsof π1(X)}. If one fixes ˜x0 as a basepoint for π1(˜X), this yields {Connected covering spacesp:˜XX}/{Subgroups of π1(X)}.

proposition (Number of sheets in a covering space, Hatcher 1.32):

For X,˜X both path-connected, the number of sheets of a covering space is equal to the index [p(π1(˜X)):π1(X)].

Note that the number of sheets is always equal to the cardinality of p1(x0).

Examples

fact (Some common covering spaces):

Any subgroup Hπ1(S1;1)=Z is of the form H=nZ, so intermediate covers are obtained from ^S1/nZ=R/nZS1 by pulling back the universal bundle:

Link to Diagram

given by the n-valent Cayley graph covering a wedge of circles.

For n=2

  • T2×2K
example (The circle S1):

Identify S1C, then every map pn:S1S1 given by zzn a yields a covering space ˜Xn. The induced map can be described on generators as pn:π1(S1)π1(S1)[ω1][ωn]=n[ω1] and so the image is isomorphic to nZ and thus pn(π1(S1))=AutCov(˜Xn)=Z/nZ. where the deck transformations are rotations of the circle by 2π/n. The universal cover of S1 is R; this is an infinitely sheeted cover, and the fiber above x0 has cardinality |Z|.

example (Projective n-space RPn):

The universal cover of RPn is Sn; this is a two-sheeted cover. The fiber above x0 contains the two antipodal points.

example (The torus S1×S1):

The universal cover of T=S1×S1 is ˜X=R×R. The fiber above the base point contains every point on the integer lattice Z×Z=π1(T)=Aut(˜X)

proposition (General construction for wedge products):

For a wedge product X=ni˜Xi, the covering space ˜X is constructed as a infinite tree with n-colored vertices:

  • Each vertex corresponds to one of the universal covers ˜Xi,
  • The color corresponds to which summand ˜Xi appears,
  • T The neighborhood of each colored vertex has edges corresponding (not bijectively) to generators of π1(Xi).
example (Wedge of circles):

The fundamental group of S1S1 is ZZ by van Kampen, and the universal cover is the following 4-valent Cayley graph:

The universal cover of \S^1 \vee S^1

See Hatcher p.58 for other covers.

corollary (Every subgroup of a free group is free):

Idea for a particular case: use the fact that π1(kS1)=Zk, so if GZk then there is a covering space X such that \pi_1(X) = G. Since X can be explicitly constructed as a graph, i.e. a CW complex with only a 1-skeleton, \pi_1(X) is free on the edges in the complement of a maximal tree.

example (Wedge of projective spaces):

The fundamental group of {\mathbf{RP}}^2 \vee {\mathbf{RP}}^2 is {\mathbf{Z}}_2 \ast {\mathbf{Z}}_2, corresponding to an infinite string of copies of 2-valent S^2s:

Another universal cover.

example (\RP^2 \wedgeprod T^2):

The fundamental group of {\mathbf{RP}}^2 \vee T^2 is {\mathbf{Z}}_2 \ast {\mathbf{Z}}, and the universal cover is shown in the following image. Each red vertex corresponds to a copy of S^2 covering {\mathbf{RP}}^2 (having exactly 2 neighbors each), and each blue vertex corresponds to {\mathbf{R}}^2 cover {\mathbb{T}}^2, with {\left\lvert {{\mathbf{Z}}^2} \right\rvert} many vertices as neighbors.

Universal cover of {\mathbb{T}}^2 \vee {\mathbf{RP}}^2

Applications

theorem (Maps into contractible spaces are always nullhomotopic):

If X is contractible, every map f: Y \to X is nullhomotopic.

proof (?):

If X is contractible, there is a homotopy H: X\times I \to X between \operatorname{id}_X and a constant map c: x \mapsto x_0. So construct \begin{align*} H': Y\times I &\to X \\ H'(y, t) &\coloneqq \begin{cases} H(f(y), 0) = (\operatorname{id}_X \circ f)(y) = f(y) & t=0 \\ H(f(y), 1) = (c \circ f)(y) = c(y) = x_0 & t=1 \\ H(f(y), t) & \text{else}. \end{cases} \end{align*} Then H' is a homotopy between f and a constant map, and f is nullhomotopic.

corollary (Factoring through a contractible space implies nullhomotopic):

Any map f:X\to Y that factors through a contractible space Z is nullhomotopic.

proof (?):

We have the following situation where f = p \circ \tilde f:

Link to diagram

Since every map into a contractible space is nullhomotopic, there is a homotopy \tilde H: Y\times I \to Z from \tilde f to a constant map c: Y\to Z, say c(y) = z_0 for all y. But then p\circ \tilde H: X \times I \to Y is also a homotopy from f to the map p\circ c, which satisfies (p\circ c)(y) = p(z_0) = x_0 for some x_0 \in X, and is in particular a constant map.

proposition (Showing one space can not cover another):

There is no covering map p: {\mathbf{RP}}^2 \to {\mathbb{T}}^2.

proof (?):

    
  • Use the fact that \pi_1({\mathbb{T}}^2) \cong {\mathbf{Z}}^2 and \pi_1({\mathbf{RP}}^2) = {\mathbf{Z}}/2{\mathbf{Z}} are known.
  • The universal cover of {\mathbb{T}}^2 is {\mathbf{R}}^2, which is contractible.
  • Using the following two facts, p_* is the trivial map:
    • By the previous results, p is thus nullhomotopic.
    • Since p is a covering map, p_*: {\mathbf{Z}}/2{\mathbf{Z}}\hookrightarrow{\mathbf{Z}}^2 is injective.
  • Since p was supposed a cover, this can be used to imply that \operatorname{id}_{{\mathbb{T}}^2} is nullhomotopic.
  • Covering maps induce injections on \pi_1, and the only way the trivial map can be injective is if \pi_1(T^2) = 0, a contradiction.
theorem (When actions yield covering maps onto their quotients, Hatcher 1.40):

If G\curvearrowright X is a free and properly discontinuous action, then

  • The quotient map p:X \to X/G given by p(y) = Gy is a normal covering space,

  • If X is path-connected, then G = \mathop{\mathrm{Aut}}_{\mathrm{Cov}} (X) is the group of deck transformations for the cover p,

  • If X is path-connected and locally path-connected, then G\cong \pi_1(X/G) / p_*(\pi_1(X)).

fact:

If f:X\to Y is a covering map of degree 1, then f is necessarily a homeomorphism.

Exercises