Chapter 2 Standard form for sphere systems
2.5 Consequences
In this section we explain the connection between the construction described in Section 2.2 and the one described in Section 2.4. Namely we will prove (Theorem 2.5.1) that these are two different ways of constructing the same object. This will allow us to prove the main results of this chapter, namely Theorem 2.5.4, stating that any two maximal sphere systems can be represented in standard form; and Theorem 2.5.6, stating that standard form is in some sense unique. Recall that we are still supposing that the two sphere systems do not have any sphere in common.
Let us start with some remark. Given a manifoldMg and two sphere systems
Σ1, Σ2 in standard form we have two ways of constructing a dual square complex.
The first way consists of constructing the square complex directly from the 3-manifold, as described in Section 2.2.
The second way uses the procedure described in Section 2.4 in the following way. LetT1 be the tree dual to Mfg and Σf1 and T2 be the tree dual to Mfg and Σf2
(here I use the notation introduced in Section 2.1.1). Both trees are endowed with an action by the free group Fg, coming from the action of the group on Mfg. The
productT1×T2 is therefore endowed with the diagonal action defined in Section 2.4.
The Gromov boundary ofT1 can be identified to the Gromov boundary of T2 since
they both can be identified to the Gromov boundary of the group Fg and to the
construct a square complexC(T1, T2). The quotient ofC(T1, T2) under the diagonal
action ofFg is a compact locally CAT(0) square complex. Throughout this Section
we will denote the complexC(T1, T2)/Fg by ∆(T1, T2).
We prove that these two constructions produce the same outcome, i.e::
Theorem 2.5.1. If Mg is the connected sum of g copies of S2×S1 and Σ1,Σ2 are
two embedded maximal sphere systems in standard form, then the square complex associated to Mfg, Σf1 and Σf2 is isomorphic to the core C(T1, T2), where T1 is the 3-valent tree dual to Mfg andΣf1 and T2 is the 3-valent tree dual to Mfg and Σf2.
Proof. Since no ambiguity can occur, for the remainder of the proof we will denote by ∆ the square complex associated toe Mfg, Σf1 and Σf2. Note that ∆ is endowede
with a free properly discontinuous action of the free groupFg, induced by the action
ofFg on the manifold Mfg.
Let T1 and T2 be the trees defined above and, as usual, denote the core as
C(T1, T2). As mentioned above, T1 and T2 are endowed with free properly discon-
tinuous actions of the free group Fg, and the core is endowed with the diagonal
action.
The proof will consist of three steps: we first prove that ∆ can be identifiede
to a subcomplex of the productT1×T2; then we prove that∆ is actually containede
inC(T1, T2); eventually we prove thatC(T1, T2) is contained in∆.e
In order to show the first step, we first prove the existence of two equivariant projectionsp1 :∆e →T1 andp2 :∆e →T2. We start defining these projections on the
0-skeleton of∆, then we extend them to the 1-skeleton and finally to the 2-skeleton.e
Let v be a vertex in ∆, this vertex represents a 3-piece ofe Mfg, i. e. a
complementary component ofΣf1∪Σf2 inMfg. In particular this 3-piece is contained
in a uniquely determined component ofMfg \Σf1, which is represented by a vertex
v1 inT1, and in a uniquely determined component of Mfg\Σf2, which is represented
by a vertex v2 inT2. Setp1(v) =v1 andp2(v) =v2.
Now we extent these projections to the 1-skeleton of ∆. Lete e be an edge
in∆. The edgee erepresents a 2-piece. This 2-piece belongs to a sphere σ1 in Σf1 if
eis a black edge and to a sphere σ2 in Σf2 ife is a red edge. Suppose for example
thateis a red edge. The red 2-piece it represents lies entirely in a component P1 of f
Mg\Σf1 and is contained in the boundary of two adjacent components ofMfg\Σf2.
We setp1(e) to be the vertex representing P1 andp2(e) to be the edge representing
σ2. The same holds for any black edge.
Note now that, if s is a square in ∆, then the two black (resp. red) edgese
belonging to the same sphere. Therefore the projectionsp1 andp2 naturally extend
to the 2-skeleton of∆.e
The projections p1 and p2 are clearly equivariant under the action of the
groupFg.
The existence of the projectionsp1 and p2 implies that there exists a surjec-
tive map of∆ onto a subcomplex of the producte T1×T2. To conclude the first step
we only need to show that this map is also an injection.
To see why this map is injective, note that standard form implies that a sphereσ1inΣf1and a sphereσ2inΣf2intersect at most once. Therefore, if we denote
bye1 the edge inT1 corresponding to σ1 and bye2 the edge inT2 corresponding to
σ2, there is at most one square s in∆ satisfyinge p1(s) = e1 and p1(s) = e2. This
concludes the proof of the first step. In the remainder of the proof we will consider
e
∆ as a subcomplex of the productT1×T2.
We show now the second step, i. e. we show that that∆ is contained in thee
core C(T1, T2). Since each edge in ∆ bounds a square, in order to prove thate ∆ ise
contained inC(T1, T2), it is sufficient to prove that each square of∆ is contained ine
C(T1, T2).
Letsbe a square in∆. The two black (horizontal) edges ofe sproject through
p1 onto an edge e1 in T1 and the two red (vertical) edges of s project through p2
onto an edgee2 inT2. The fact thats is a square in ∆ means that the spherese σ1
andσ2 represented bye1 ande2 intersect, and therefore, by Lemma 2.1.10, the two
partitions induced byσ1andσ2on the boundary ofMfgare not nested. Consequently
the partitions induced by the edgese1 and e2 on the boundary ofT1 andT2 are not
nested and therefores=e1×e2 is a square inC(T1, T2).
To finish the proof of Theorem 2.5.1, we only need to prove thatC(T1, T2)
is contained in∆. As above it is sufficient to prove that each square ofe C(T1, T2) is
contained in∆.e
Let s =e1×e2 be a square in C(T1, T2). The edge e1 represents a sphere
in Σf1, call this sphere σ1; the edge e2 represents a sphere in Σf2, call this sphere
σ2. The fact that sis contained in the core means that the partitions induced by
the edgese1 and e2 on the boundary of T1 and T2 are not nested; this implies that
the partitions induced by the spheresσ1 andσ2 on the space of ends ofMfg are not
nested. Therefore, by Lemma 2.1.10, the two spheres σ1 and σ2 intersect in Mfg
and their intersection consists of exactly one circle. This means that there are two 2-pieces ofσ1and two 2-pieces ofσ2 inMfg, all four of them intersecting in a 1-piece,
Remark 2.5.2. Note that Theorem 2.5.1 and Proposition 2.4.11 immediately imply Lemma 2.2.7 and Lemma 2.2.8, whose proof we had omitted in Section 2.2.
It is easy to note that, if we see ∆(Mfg,Σf1,Σf1) as a subcomplex ofT1×T2,
then theFg-action on this square complex induced by theFg-action on the manifold f
Mg coincides with the diagonal action of Fg on the productT1×T2. Therefore an
immediate consequence of Theorem 2.5.1 is the following:
Corollary 2.5.3.Under the hypothesis of Theorem 2.5.1 the square complex∆(Mg,Σ1,Σ2)
is isomorphic to the square complex ∆(T1, T2).
We have shown so far that the construction described in Section 2.2 and the one described in Section 2.4 produce the same result.
Note anyway that the construction described in Section 2.4 is much more general. Above all, we can perform this construction starting just with a three- manifoldMg and two embedded maximal sphere systems which do not contain any
sphere in common. We do not need the two sphere systems to be in minimal or standard form.
Note also that, if we have two trees T1, T2 with no edge in common, we
construct the core C(T1, T2) and quotient it by the diagonal action of the group
Fg, then we obtain a square complex ∆(T1, T2) satisfying all the properties 1-6
described in Section 2.4, and therefore we can associate to it a 3-manifoldMg with
two maximal sphere systems,QRand QB, in standard form.
Summarising, if we have a manifoldMg and two embedded maximal sphere
systems, not necessarily in standard form, then we can associate to each system a dual tree with a group action. We can construct the core of the two trees. Then, using the method described in Section 2.3 we can build a manifold Mg with two
sphere systems in standard form.
The remarks I have just made are the main ingredients for the proof of existence of standard form. In fact, we are now ready to prove the following:
Theorem 2.5.4. Let Mg be the connected sum of g copies of S2×S1 and let Σ1,
Σ2 be two embedded maximal sphere systems such that no sphere inΣ1 is homotopic
to a sphere inΣ2. Then there exist maximal sphere systems Σ01, Σ02 such that Σ0i is
homotopic toΣi for i= 1,2, and Σ01, Σ02 are in standard form.
Before starting the proof, to avoid confusion, we clarify some terminology. Given two infinite trivalent trees T and T0 endowed with (free, properly discon- tinuous and cocompact) actions by the group Fg and therefore with a boundary
simplicial isomorphism ϕ : T → T0 such that for each edge e in T its image ϕ(e) induces the same partition ase. We are now ready to prove Theorem 2.5.4.
Proof. (of Theorem 2.5.4) Let Σ1, Σ2 be two sphere systems in Mg satisfying the
hypothesis of the theorem. As usual denote by Mfg the universal cover of Mg and
by Σf1 and Σf2 the entire lifts of Σ1 and Σ2. Let T1 be the tree dual to Mfg and Σf1
and letT2 be the tree dual toMfg andΣf2. BothT1 andT2 are endowed with a (free,
properly discontinuous and cocompact) Fg-action. Let C(T1, T2) be the core of T1
andT2.
Now, applying the procedure explained in Section 2.3 to the square complex C(T1, T2) we can construct a simply connected three manifold Mf0, with two embed-
ded maximal sphere systems, QfR and gQB, in standard form with respect to each
other. The manifoldMf0 is abstractly homeomorphic to Mfg.
By construction, C(T1, T2) is the square complex dual to the manifold Mf0
and the two sphere systemsQfRandgQB, therefore, by theorem 2.5.1, it is isomorphic
to the core of the two trees associated toMf0 andgQB, and toMf0 andQfR. The trees
associated toMf0 and gQB and to Mf0 and QfR are the two projections of the square
complexC(T1, T2), namely T1 and T2. Therefore, without loss of generality, up to
permuting the labels QR and QB, we can suppose that T1 is the tree associated to f
M0 and
g
QB, andT2 is the tree associated toMf0 and QfR.
The space of ends ofMf0 can be identified to the space of ends ofMfg, since
they both can be identified to the boundaries ofT1 andT2. Moreover, since the tree
dual toMfg and Σf1 is the same as the tree dual toMf0 and gQB (they both coincide
with the treeT1), then for each sphereσ inΣf1 there is a sphere ingQB inducing the
same partition as σ and for each sphere s in QgB there is a sphere in Σf1 inducing
the same partition ass. The same holds forΣf2 and QfR.
We can choose a homeomorphism H : Mf0 → Mfg which is consistent with
the identification on the space of ends and such that, for each spheresinQfR∪gQB,
the partition induced by H(s) is the same as the partition induced by s. Denote H(gQB) byΣf0
1 andH(QfR) by Σf0 2.
The systemsΣf01 andΣf02 are maximal and are in standard form with respect
to each other, since they are homeomorphic image of two maximal sphere systems in standard form. Moreover, for each sphere in Σf1 (resp. Σf2) there is a sphere in f
Σ01 (resp. Σf02) inducing the same partition and vice versa. Therefore, for i = 1,2
the sphere systemfΣi is homotopic in Mfg to the sphere systemfΣ0i.
Let Σ10 and Σ02 in Mg be the projections of Σf01 and Σf02 through the covering
respect to each other and moreover for i= 1,2 the sphere system Σ0i is homotopic inMg to the sphere system Σi.
As an immediate consequence of Theorem 2.5.4 we can show something we had mentioned without proof in Section 2.1.2. Namely:
Remark 2.5.5. An immediate consequence of Theorem 2.5.4 is that, as promised in Section 2.1.2, two maximal sphere systems not containing any sphere in common can always be homotoped to be in strong minimal form with respect to each other. In other words a strong minimal form always exists for two maximal sphere sys- tems containing no spheres in common. This implies that the three definitions of minimality given in Section 2.1.2 are all equivalent.
To summarise what we have done, basically, in the proof of Theorem 2.5.4, we have shown a constructive way to find a standard form for two maximal sphere systems inMg. Namely, given two embedded maximal sphere systems Σ1 and Σ2 in
Mg which do not contain any sphere in common; letT1 be the tree dual toMfg and f
Σ1 and letT2be the tree dual toMfgandΣf2. LetC(T1, T2) be the core ofT1×T2and
let ∆(T1, T2) be the quotient of C(T1, T2) by the diagonal action of Fg. Applying
the procedure explained in Section 2.3 to ∆(T1, T2) we construct a 3-manifold M
homeomorphic toMg, with two embedded maximal sphere systems,QRand QB, in
standard form with respect to each other. Note that this construction is defined up to twists around spheres inQR and QB. Note also that the construction depends
only on the homotopy class of the systems Σ1 and Σ2.
We show now that a standard form for two maximal sphere systems is “in some sense”unique. More precisely:
Theorem 2.5.6. Let (Σ1,Σ2), (Σ01,Σ02) be two pairs of embedded maximal sphere
systems in Mg. Suppose that both pairs of sphere systems are in standard form
and do not contain any sphere in common (i.e. that no sphere inΣ1 (resp. Σ01) is
homotopic to a sphere in Σ2 (resp. Σ02)). Suppose also that Σi is homotopic to Σ0i
for i= 1,2.
Then there exists a homeomorphism F : Mg → Mg such that F(Σi) = Σ0i
for i= 1,2. The homeomorphism F induces the identity (up to conjugacy) on the fundamental group ofMg.
The proof of Theorem 2.5.6 is based on Lemma 2.3.13 and on the following lemma.
Lemma 2.5.7. For g ≥3, let F :Mg →Mg be a self-homeomorphism of Mg. Let
Σbe a maximal sphere system embedded inMg. Suppose that for each sphere σ inΣ
the imageF(σ)is homotopic toσ. Then the induced homomorphismF∗:π1(Mg)→
π1(Mg) is an inner automorphism of the free groupFg.
Proof. Denote as usual as Mfg the universal cover ofMg, and denote the full lift of
Σ byΣ. The manifolde Mfg is endowed with an action by the free group Fg and the
quotient of Mfg by this action is the manifold Mg. In order to prove Lemma 2.5.7
we will show that a liftFe of the homeomorphismF is equivariant under this group
action.
First note that the action ofFg onMfg induces an action ofFgon the space of
ends. This action on the space of ends determines the action onMfg up to homotopy;
in fact, since each component of Mfg \Σ is a 3-holed 3-sphere, then the action ofe
Fg on Mfg is determined by the action of Fg on Σ; and the action ofe Fg on Σ ise
determined up to homotopy by the action ofFg on the space of ends ofMfg.
Now we exhibit a particular lift Fe. Note that, since each component of f
Mg\Σ is a 3-holed 3-sphere, then the mape Feis determined, up to homotopy, by its