5.1 topology of the conformal boundary Σ
5.1.4. Conformal Compactifications
The definition of an almost Einstein structure (M,g,σ) does not necessarily coincide with a conformal compactification of M\Σ,σ−2g
. The following section examines the properties of special almost Einstein structures that admit a conformal compactification. In particular we will require(M,g)to have a compact subset ¯M ⊂ M that is the closure of a connected component of M\Σ. We will next label the sets of interest.
Figure 3.: schematics of a compactification
Let (M,g,σ) be an almost Einstein structure. Then the set M\Σ may have several connected components. This will be labelled by i.
Then fixing one component, we do the following notations M˚ i fixed connected component of M\Σ Mi connected component of ˚Mi∪Σc M¯i connected component ofMi∪Σd
∂M˚i := M¯i∩Σ.
With this definition we have the inclusions ˚Mi ⊂ Mi ⊂ M¯i. Provided M˚iis a conformally Einstein manifold with boundary, then∂M˚iwill be its conformal boundary.
Definition5.1.20. LetNbe a manifold. A property that is defined for everyx ∈Nwill be called openif it holds on an open subset ofN.
96 Chapter 5: almost einstein structures with vanishing almost scalar curvature
Lemma5.1.21. Let be p∈Σc∩∂topM˚i a point in the topological boundary of M˚i. Then all points q in the same connected component ofΣcalso belong to the topological boundary ofM˚i.
Proof: We will show that the propertiesq∈Σc∩∂topM˚i andq∈/Σc∩∂topM˚iare open properties inΣc. In particular for allq∈Σcwithq∈Σc∩∂topM˚ithere is a neighbourhoodUofqsuch that for allx∈U∩Σcit holdsx∈Σc∩∂topM˚i. The same can be said forq∈/Σc∩∂topM˚i.
First consider an arbitraryq ∈ Σc such that q ∈/ Σc∩ ∂t o pM˚ i. Since ∂t o pM˚ i is a closed set in M, there is an open neighbourhood U ⊂ M of q such that U∩ ∂t o pM˚ i = ∅. Hence q admits a neighbourhood U∩ Σc open in Σ such that the property x ∈/ Σc∩∂t o pM˚ i holds for allx ∈ U∩Σc. Hence it is an open property inΣc.
Now let q ∈ Σc be arbitrary such that this time q ∈ Σc∩ ∂t o pM˚ i. There is a coordinate chart ϕ = (ϕ1, . . . ,ϕn) : U → Rn for a neighbourhood of q such that ϕn(x) = 0 for all x ∈ U∩Σc. Therefore we have dϕnx(Y) = 0 for all such x and Y ∈ TxΣc and hence d ϕ1, . . . ,ϕn−1
x(Y) 6= 0 for all tangent vectors ofΣc. We define ˜ϕ := ϕ1, . . . ,ϕn−1,σ : U → Rn. Since dσx(Y) 6= 0 for transversal vectors Y in x ∈ Σc, we find kerdϕ˜x = {0} for all x ∈ U∩ Σc and in particular for x = q. By the inverse function theorem there is a neighbourhood ˜Uofqsuch that
˜
ϕ : ˜U →Rn (5.7)
is a coordinate chart and ˜ϕ(U˜) = B is an open ball. Since q ∈ ∂t o pM˚ i it holds that ˜U∩ M˚ i 6= ∅. Consider x ∈ U˜ ∩ M˚i and without loss of generality σ(x) > 0, then by con-nectedness of the open ball we get ˜ϕ−1 B∩ Rn−1×(0,∞) ⊂ M˚ i and hence ˜U∩ Σc =
˜
ϕ−1 B∩ Rn−1× {0} = U˜ ∩ Σc ⊂ ∂t o pM˚ i. Consequently q ∈ Σc∩ ∂t o pM˚ i is an open property too.
Now we have the disjoint union of setsΣc = Σ1c∪Σ2c open inΣcwhereΣ1c :={p ∈ Σc| p ∈
∂t o pM˚ i} and Σ2c := {p ∈ Σc| p ∈/ ∂t o pM˚ i}. In particular the connected components of Σc
must completely belong to one of those sets.
Lemma5.1.22. ∂M˚ icoincides with the topological boundary ∂t o pM˚ i ofM˚ i in M.
Proof: First, we will show the inclusion∂M˚ i ⊂ ∂t o pM˚ i. Consider p ∈ Σc∩∂M˚ i and therefore p ∈ Mi. By definition, Mi is connected such that there is a curve γ : [0, 1] → Mi with γ(0) = p and γ(1) =: x ∈ M˚ i. We will consider γ to be a curve with values in Mi ∪Σ.
Let bet0 := i n f {t ∈ [0, 1]|γ(t) ∈ Σc}. Since Σis a closed set and γ(t) must not be in Σd, the limit p0 := γ(t0) must be an element of Σc such that the restriction γ : [0,t0] → Σc is completely within the connected component ofΣc. For each of the neighbourhoodsUp0 of p0
the intersectionUp0 ∩ M˚ i is non-empty, since due to the definition oft0there must be at0 > t0
such thatγ(t0) ∈ Up0 ∩ M˚ i. Hence p0 is an element of the topological boundary. By Lemma 5.1.21the whole connected component of Σc and in particular p = γ(0) are in the topological boundary. Therefore we get
Mi ⊂ M˚ i. (5.8)
Now consider p ∈ Σd∩∂M˚ i. By Proposition5.1.1(ii), p is an isolated point in Σd. Therefore every neighbourhood Up of p has non-vanishing intersection with Mi, since ¯Mi is connected.
Hence p is an element of the topological boundary ∂t o pMi, which is also a subset of ∂t o pM˚ i, since Mi ⊂ M˚ i by Equation (5.8). This completes the first inclusion ∂M˚ i ⊂ ∂t o pM˚ i.
Second we show that the backward inclusion∂t o pM˚ i ⊂ ∂M˚i holds. ˚Mi is a connected compo-nent of the open set M\Σ. Now considerp ∈ ∂t o pM˚ i, thenp cannot be in the open set M\Σ and hencep ∈ Σ. First let p ∈ Σcand consider ˜ϕto be the coordinate chart in a neighbourhood of p that has been defined in Equation (5.7). Since ˜U∩M˚ i 6= ∅, there is a curve connecting p to a point in the interior ˜U∩ M˚ i such thatσ is strictly positive along the curve. Consequently p ∈ Mi and hence p ∈ ∂M˚i. Now let p ∈ Σd. Then in Morse coordinates M\Σ consists of three connected open components. Since each of them is connected to the vertex by a path, p must be an element of ¯Mi and hence of∂M˚ i. This proves the backward inclusion.
The previous observations lead to the following proposition on the topology of the level sets in a neighbourhood of a compact conformal boundaryΣ.
5.1 topology of the conformal boundary Σ 97
Proposition5.1.23. If∂M˚ i is compact and M geodesically null complete without boundary, then (i) The number of vertices in any connected component ofΣ∩∂M˚ i is finite.
(ii) There exists a pre-compact neighbourhoodU of∂M˚ iin M such that M˚ i can be written as M˚ i∩ U = [
s∈(−e,e)
Σs ∩M˚ i∩ U
withΣs∩M˚ i∩ Ubeing(n−1)-dimensional hypersurfaces not leavingU, i.e.∂U ∩ Σs∩ M˚ i
=
∅. In particular there are no critical points ofσwithin M˚ i∩ U. (iii) The gradient vector fieldgradσif restricted to∂M˚ i is complete.
Figure4.:topology near com-pact boundaries
Proof: The third claim is a consequence of Proposition5.1.18. By the requirements ∂M˚ i ⊂ Σ is compact and hence ∆σ is bounded on
∂M˚i ∩Σc. By Lemma5.1.21∂M˚ i∩Σc contains only connected parts of Σc as a whole and not just parts of it. Now applying Proposition 5.1.18gives the claim.
The first claim is a consequence of the compactness of∂M˚ i. Let be p ∈ Σd∩∂M˚ i a vertex. By Proposition5.1.1(ii), p is an isolated point such that it admits a neighbourhood Up without any other vertices.
For p ∈ Σc∩∂M˚ i choose a neighbourhood Up without any vertices in it. With these choices
∂M˚ i ⊂ [
p∈∂M˚i
Up
is a covering and by compactness admits a finite subcovering. Since the neighbourhood of every vertex must be an element of that covering, their number consequently must be finite.
The proof for the second claim uses the same idea. First we observe that for each p ∈ Σthere is a connected neighbourhoodUp such that gradσ 6= 0 for all q ∈ Up\ {p}. Moreover, we demand each neighbourhood to be such that it includes only one connected component of Σ and its closureUp is supposed to be compact. Those neighbourhoods provide an open covering of∂M˚ ithat must have a finite subcovering
U = [
j∈ {1,..,m}
Upj.
Its closureU = Si∈ {1,..,m}Upi is compact and so is its boundary∂U. By construction we have
∂U∩∂M˚ i =∅. (5.9)
Without loss of generality assume ˚Mi to be a subset of σ−1( (0,∞) ). We will now prove by contradiction that there is an e such that for all 0 < δ < e we have Σδ ∩ M¯ i ∩∂U = ∅. So assume that for alle > 0 there exists aδ ∈ (0,e] such thatΣδ∩ M¯ i∩∂U 6= ∅. Hence we can choose for everye = 1i an element of that intersection
xi ∈ Σ1i ∩ M¯ i∩∂U.
By the compactness of ¯U there is a subsequence {xij} that converges to an x ∈ ∂U. By con-struction we have σ(xij) = i1
j for that sequence. By continuity of σ we have σ(x) = 0 and hence x ∈ ∂M˚ i. This contradicts (5.9). The same argument holds for ˚Mi ⊂ σ−1( (−∞, 0) ), so the first assumption really is no loss of generality.
As a consequence there exists an e >0 such thatΣδ∩M¯ i∩∂U = ∅for all 0 <|δ| ≤ e. The connected parts ofΣδ within ˚Mi∩Uare hypersurfaces since by construction dσ 6= 0 onU. We define
U :=U∩σ−1( (−e,e) )∪M¯ ic
98 Chapter 5: almost einstein structures with vanishing almost scalar curvature
where ¯Mci is the complement of ¯Mi in M. the setU by definition is open, contains ∂M˚ i and the connected parts ofΣδwithin ˚Mi∩Uforδ ∈ (−e,e). Hence we get
U ∩ M˚ i = U ∩σ−1( (−e,e) )∩ M˚ i
and the second claim follows directly by replacingσ−1( (−e,e) )with the union of level sets.
Remark. By restrictingσto the open set ˚Mi∩ U, we are left with a smooth functionσ : ˚Mi∪ U → R. σ−1( [α0,α1] ) are compact sets for 0 < αi < e if σ is positive on ˚Mi and −e < αi < 0 else. Hence by Theorem 1.5.1 the sets σ−1(∞,α0) and σ−1(∞,α1) are diffeomorphic and the former set is a deformation retract of the latter. Sincedσ is non-singular and closed in the constructed neighbourhood of∂M˚ i, the level sets ofσrepresent a1-dimensional foliation of ˚Mi in a neighbourhood of its boundary.