§ Covering spaces
§ Covering spaces: Intuition
- Consider the map p(z)=z2:C×→C×. This is a 2-to-1 map. We can try to define an inverse regardless.
- We do define a "square root" if we want. Cut out a half-line [0,−infty)called B for branch cuts. We get two functions on q+,q−:C−B→C×, such that p(q+(z))=z. Here, we have q−=−q+.
- The point of taking the branch cut is to preserve simply connectedness. C×is not simply connected, while C/B is simply connected! (This seems so crucial, why has no one told me this before?!)
- Eg 2: exponential. Pick exp:C→C×. This is surjective, and infinite to 1. ez+2πn=eiz.
- Again, on C/B, we have qn≡log+2πin, such that exp(qn(z))=z.
- A covering map is, roughly speaking, something like the above. It's a map that's n-to-1, which has n local inverse defined on simply connected subsets of the target.
- So if we have p:Y→X, we have q:U→Y (for U⊆X) such that p(q(z))=z,∀z∈U.
§ Covering spaces: Definition
- A subset U⊂X is a called as an elementary neighbourhood if there is a discrete set F and a homeomorphism h:p−1(U)→U×Fsuch that p∣p−1(U)(y)=fst(h) or p∣p−1(U)=pr1∘h.
- Alternative definition A subset U⊂X is called as evenly covered/elementary nbhd if p−1(U)=⊔αVα where the Vα are disjoint and open, and p∣Vα:Vα→U is a homeomorphism for all α.
- An elementary neighbourhood is the region where we have the local inverses (the complement of a branch cut).
- We get for each i∈F , a map qi:U→U×F;qi(x)=(x,i)and then along h−1 sending h−1(x,i)∈p−1(U).
- We say p is a covering map if X is covered by elementary neighbourhoods.
- We say V⊆Y is an elementary sheet if it is path connected and p(V)is an elementary neighbourhood.
- So, consider p(x)=eix:R→S1. If we cut the space at (0,0), then we will have elementary neighbourhood S1−{(0,0)}and elementary sheets (2πk,2π+1).
- The point is that the inverse projection p−1 takes U to some object of the form U×F: a local product! So even though the global covering space R does not look like a product of circles, it locally does. So it's some sort of fiber bundle?
Slogan: Covering space is locally disjoint copies of the original space.
§ Path lifting and Monodromy
- Monodromy is behaviour that's induced in the covering space, on moving in a loop in a base.
- Etymology: Mono --- single, drome --- running. So running in a single loop / running around a single time.
- Holonomy is a type of monodromy that occurs due to parallel transport in a loop, to detect curvature
- Loop on the base is an element of π1(X).
- Pick some point x∈X. Consider F≡π−1(x) ( F for fiber).
- Now move in a small loop on the base, γ. The local movement will cause movement of the elements of the fiber.
- Since γ(1)=γ(0), the elements of the fiber at the end of the movement are equal to the original set F.
- So moving in a loop induces a permutation of the elements of the fiber F.
- Every element of π1(X) induces a permutation of elements of the fiber F.
- This lets us detect non-triviality of π1(X). The action of π1(X) on the fiber lets us "detect" what π1(X) is.
- We will define what is means to "move the fiber along the path".
§ Path lifting lemma
Theorem :Suppose p:y→X is a covering map. Let δ:[0,1]→X
be a path such that δ(0)=x, and let y∈p−1(x) [ y is in the fiber of x].
Then there is a unique path γ:[0,1]→Y which "lifts" δ.
That is, δ(p(y))=γ(y), such that γ(0)=Y.
Slogan: Paths can be lifted. Given how to begin the lift, can be extended all the way.
- Let N be a collection of elementary neighbourhoods of X.
- {δ−1(U):U∈N} is an open cover (in the compactness sense) of [0,1].
- By compactness, find a finite subcover. Divide interval into subintervals 0=t0<t1<…tn=1such that δ∣k=δ∣[tk,tk+1] lands in Uk, an elementary neighbourhood.
- Build γ by induction on k.
- We know that γ(0) should be y.
- Since we have an elementary neighbourhood, it means that there are a elementary sheets living over U0, indexed by some discrete set F. y lives in one of thse sheets. We have local inverses qm. One of them lands on the sheet of y, call it q. So we get a map q:U0→Y such that q(x)=y.
- Define γ(0)≡q(δ(0))=q(x)=y.
- Extend γ upto t1.
- Continue all the way upto tk.
- To get γ from (tk,tk+1, there exists a qk:Uk→Ysuch that qk(δ(tk))=γ(tk). Define γ(tk≤t≤tk1)≡qk(δ(tk)).
- This is continuous because δ continuous by definition, qk continuous by neighbourhood, γ is pieced together such that endpoints fit, and is thus continuous.
- Can check this is a lift! We get p∘γ=p∘qk∘δk. Since qk is a local inverse of p, we get p∘γ=δkin the region.
§ 7.03: Path lifting: uniqueness
If we have a space X and a covering space Y, for a path gamma that
starts at x, we can find a path γ′ which starts at y∈p−1(x)
and projects down to γ: γ(t)=p(γ′(t)). We want to show
that this path lift is unique
§ Lemma
Let p:Y→X be a covering space. Let T be a connected space
Let F:T→X be a continuous map (for us, T≃[0,1]).
Let F1,F2:T→Y be lifts of F ( p∘F1=F, p∘F2=F).
We will show that F1=F2 iff the lifts are equal for some t∈T.
Slogan: Lifts of paths are unique: if they agree at one point, they agree at all points!
- We just need to show that if F1 and F2 agree somewhere in Y, they agree everywhere. It is clear that if they agree everywhere, they must agree somewhere.
- To show this, pick the set S where F1,F2 agree in Y: S≡{t∈T:F1(t)=F2(t)}.
- We will show that S is open and closed. Since T is connected, S must be either the full space or the empty set. Since S is assumed to be non-empty, S=T and the two functions agree everywhere.
- (Intuition: if both S and Sc are open, then we can build a function that colors T=S∪Scin two colors continuously; ie, we can partition it continuously; ie the spaces must be disconnected. Since T is connected, we cannot allow that to happen, hence S=∅ or S=T.)
- Let t inT. Let U be an evenly covered neighbourhood/elementary neighbourhood of F(t) downstairs (in X). Then we have p−1(U)=⊔αVα such that p∣Vα:Vα→Uis a local homeomorphism.
- Since F1,F2 are continuous, we will have opens V1,V2 in Vα, which contain F1(t),F2(t) upstairs (mirrroring U containing F(t) downstairs).
- The pre-images of V1, V2 along F1,F2 give us open sets t∈T1,T2⊆T.
- Define T∗=T1∩T2. If F1(t)=F2(t), then V1=V2and thus F1=F2 on all of T∗. So, Sc=T∗ is open.
- If F1(t)=F2(t), then V1=V2 and thus F1=F2 on T∗(since p∘F1=F=p∘F2, and pis injective within U, ie within V1,V2). So S is open.
- Hence we are done, as S is non-empty and clopen and is thus equal to T. Thus, the two functions agree on all of T.
§ Homotopy lifting, Monodromy
- Given a loop γ in X based at x , the monodromy around γ is a permutation σγ:p−1(x)→p−1(x), where σγ(y)≡γy(1)where γy is the unique lift of γ staring at y. We have that σγ∈Perm(p−1(x)).
- Claim: if γ1≃γ2 then σγ1=σγ2.
- We need a tool: homotopy lifting lemma.
Slogan: permutation of monodromy depends only on homotopy type
§ Homotopy lifting lemma/property of covering spaces
Suppose p:Y→X is a covering map and γs is a homotopy
of paths rel. endpoints ( γs(0) and γs(1) are independent of s /
endpoints are fixed throughout the homotopy). Then there exists for each
lift γ0′:[0,1]→Y of γ0:[0,1]→X
(ie, p∘γ0′=gamma0), a completion
of the lifted homotopy γs′:[0,1]→Y (ie, p∘gammas′=gammas).
Moreover, this lifted homotopy is rel endpoints: ie, the endpoints of gamma′ are
independent of s.
Slogan: homotopy lifted at 0 can be lifted for all time
- Let H:[0,1]×[0,1]→X be the homotopy in X such that H(s,t)=γs(t). Subdivide the square into rectangles Rij such that H(Rij) is contained in Uij for some elementary neighbourhood Uij. We build H′:[0,1]×[0,1]→Y by building local inverses qij:Uij→Y such that p∘qij=Rij. We then set H′∣Rij=qij∘H.