This paper solves the inverse problem in Moebius geometry on the circle by characterizing a class of structures that correspond to hyperbolic space boundaries, showing this class is non-empty and topologically open.
Contribution
It provides a description of a class of Moebius structures on the circle that are realizable as boundaries of hyperbolic spaces, expanding understanding of boundary structures.
Findings
01
Identifies a non-empty class of Moebius structures on the circle.
02
Shows this class forms an open neighborhood of the canonical structure.
03
Establishes a link between Moebius structures and hyperbolic space boundaries.
Abstract
We give a solution to the inverse problem of Moebius geometry on the circle. Namely, we describe a class of Moebius structures on the circle for each of which there is a hyperbolic space such that its boundary at infinity is the circle, and the induced Moebius structure coincides with the given one. That class is not empty and form an open neighborhood of the canonical Moebius structure in an appropriate fine topology.
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Full text
Inverse problem for Möbius geometry on the circle
Sergei Buyalo111This work is supported by the RFBR grant 17-01-00128 and by
Presidium of RAS program “New methods of mathematical modelling in non-linear dynamical systems”
(grant 08-04)
Abstract
We give a solution to the inverse problem of
Möbius geometry on the circle. Namely, we describe a class of Möbius
structures on the circle for each of which there is a hyperbolic space
such that its boundary at infinity is the circle, and the induced Möbius
structure coincides with the given one. That class is not empty and form
an open neighborhood of the canonical Möbius structure in an appropriate
fine topology.
Any (boundary continuous) hyperbolic space induces on the boundary at infinity
a Möbius structure which reflects most essential asymptotic properties of the space.
A Möbius structureM
on a set
X
is a class of Möbius equivalent semi-metrics on
X,
where two semi-metrics are equivalent if and only if they have
the same cross-ratios on every 4-tuple of points in
X.
In other words, a Möbius structure is given by cross-ratios.
The inverse problem of Möbius geometry asks to describe Möbius structures which are
induced by hyperbolic spaces. In this paper, we give a solution to the inverse problem
in a simplest case when the space
X
is the circle,
X=S1.
The paper is a continuation of [Bu18], [Bu19], where the inverse problem is formulated,
and important steps toward its solution are done.
Various hyperbolic cone constructions (see [BoS], [BS07]) give a hyperbolic metric space with
prescribed metric at infinity. However, no one of them is equivariant with respect to Möbius
transformations of the metric. Thus one can consider the inverse problem as the existence problem
of an equivariant hyperbolic cone over a given metric.
We introduce a set of axioms describing Möbius structures on the circle, which are
induced by hyperbolic spaces. We always consider ptolemaic Möbius structures,
that is, for which every semi-metric with infinitely remote point is a metric. Our
monotonicity axiom is somewhat stronger than that in [Bu18].
Thus a Möbius structure, which satisfies it, is called strictly monotone.
As in [Bu18], we also use a key Increment axiom. For the definition and details see
sect. 2.4.
The main result of the paper is the following
Theorem 1.1**.**
Given a strictly monotone Möbius structure
M
on the circle satisfying Increment axiom, there is a complete,
proper and geodesic hyperbolic metric space
Y
with boundary at infinity
∂∞Y=S1,
for which the induced Möbius structure
MY
on
∂∞Y
is isomorphic to
M, MY=M.
Remark 1.2*.*
The class
I
of strictly monotone Möbius structures on the circle which satisfy Increment
axiom contains an open in a fine topology neighborhood of the canonical
Möbius structure
M0,
see sect. 2.4.
Structure of the paper. In section 2, we give
a brief introduction to Möbius structures, formulate basic axioms, including
Increment axiom, and discuss a fine topology on the set
M
Möbius structures satisfying our axioms.
In section 3 we recall the notions of lines and zz-paths
associated with a given Möbius structure
M∈M.
After a brief discussion in sect. 4
of the metric on the set
Harm
of harmonic 4-tuples, we consider in
sect. 5 an important notion of involutions
without fixed points and the associated notion of elliptic quasi-lines. Given
ω∈X=S1,
we consider here the set
Harmω⊂Harm
of harmonic 4-tuples containing
ω.
Such sets play a important role in the proof of the main theorem.
A key technical part of the paper is section 6,
where we give an universal upper bound for the diameter of elliptic
quasi-lines. Such estimate allows to reduce the study of geometry on
the space
Harm
to the study of its much simpler subspaces
Harmω.
In section 7, we discuss properties of
a hyperbolic cone construction over
Xω
called the hyperbolic approximation
Z
of
Xω.
We show here that
Z
is a hyperbolic geodesic metric space. This section is based on the book [BS07, Chapter 6].
Finally, in sect. 8 we show that the spaces
Harmω
and
Z
are quasi-isometric. As a corollary, we obtain that the required filling
Y=Harm
of a given Möbius structure
M∈M
on the circle is hyperbolic. The proof essentially uses Increment axiom and results
of [Bu18].
2 Möbius structures
2.1 Basic notions
Let
X
be a set. A 4-tuple
q=(x,y,z,u)∈X4
is said to be admissible if no entry occurs three or
four times in
q.
A 4-tuple
q
is nondegenerate, if all its entries are pairwise
distinct. Let
P4=P4(X)
be the set of all ordered admissible 4-tuples of
X, regP4⊂P4
the set of nondegenerate 4-tuples.
A function
d:X2→R=R∪{∞}
is said to be a semi-metric, if it is symmetric,
d(x,y)=d(y,x)
for each
x, y∈X,
positive outside the diagonal, vanishes on the diagonal
and there is at most one infinitely remote point
ω∈X
for
d,
i.e. such that
d(x,ω)=∞
for some
x∈X∖{ω}.
Moreover, we require that if
ω∈X
is such a point, then
d(x,ω)=∞
for all
x∈X, x=ω.
A metric is a semi-metric that satisfies the triangle inequality.
A Möbius structureM
on
X
is a class of Möbius equivalent semi-metrics on
X,
where two semi-metrics are equivalent if and only if they have
the same cross-ratios on every
q∈regP4.
Given
ω∈X,
there is a semi-metric
dω∈M
with infinitely remote point
ω.
It can be obtained from any semi-metric
d∈M
for which
ω
is not infinitely remote by a metric inversion,
[TABLE]
Such a semi-metric is unique up to a homothety, see [FS13],
and we use notation
∣xy∣ω=dω(x,y)
for the distance between
x, y∈X
in that semi-metric. We also use notation
Xω=X∖{ω}.
Every Möbius structure
M
on
X
determines the
M-topology
whose subbase is given by all open balls centered at finite points
of all semi-metrics from
M
having infinitely remote points.
Example 2.1**.**
Our basic example is the
canonical Möbius structure
M0
on the circle
X=S1.
We think of
S1
as the unit circle in the plane,
S^{1}=\{(x,y)\in\mathbb{R}^{2}:\,\text{x^{2}+y^{2}=1}\}.
For
ω=(0,1)∈X
the stereographic projection
Xω→R
identifies
Xω
with real numbers
R.
We let
dω
be the standard metric on
R,
that is,
dω(x,y)=∣x−y∣
for any
x,y∈R.
This generates a Möbius structure on
X
which is called canonical. The basic feature of the canonical Möbius
structure on
X=S1
is that for any 4-tuple
(σ,x,y,z)⊂X
with the cyclic order
σxyz
we have
dσ(x,y)+dσ(y,z)=dσ(x,z).
2.2 Harmonic pairs
From now on, we assume that
X
is the circle,
X=S1.
It is convenient to use unordered pairs
(x,y)∼(y,x)
of distinct points on
X,
and we denote their set by
aY=S1×S1∖Δ/∼,
where
\Delta=\{(x,x):\,\text{x\in S^{1}}\}
is the diagonal. A pair
q=(a,b)∈aY×aY
is harmonic if
[TABLE]
for some and hence any semi-metric of the Möbius structure, where
a=(x,y), b=(z,u).
The pair
a
is called the left axis of
q,
while
b
the right axis. We denote by
Harm
the set of harmonic pairs,
Harm⊂aY×aY,
of the given Möbius structure. There is a canonical involution
j:Harm→Harm
without fixed points given by
j(a,b)=(b,a).
Note that
j
permutes left and right axes. The quotient space we denote by
Hm:=Harm/j.
In other words,
Hm
is the set of unordered harmonic pairs of unordered pairs
of points in
X,
and
Harm
is its 2-sheeted covering.
Remark 2.2*.*
Sometimes, we need a 2-sheeted
covering
Harm
of
Harm,
which consists of harmonic pairs
q=(a,b)
with
a=(x,y)∈S1×S1∖Δ, b∈aY.
Note that
Harm
is homeomorphic to the tangent bundle of
H2.
2.3 Axioms
We list a set of axioms for a Möbius structure
M
on the circle
X=S1,
which needed for Theorem 1.1.
(T)
Topology: M-topology
on
X
is that of
S1.
(M(α))
Monotonicity: Fix
1>α≥2−1.
Given a 4-tuple
q=(x,y,z,u)∈X4
such that the pairs
(x,y), (z,u)
separate each other, we have
[TABLE]
for some and hence any semi-metric from
M.
(P)
Ptolemy: for every 4-tuple
q=(x,y,z,u)∈X4
we have
[TABLE]
for some and hence any semi-metric from
M.
A Möbius structure
M
on the circle
X
that satisfies axioms T, M(α), P is said to be strictly monotone.
We denote by
M
the class of strictly monotone Möbius structures on
X.
Remark 2.3*.*
Axiom M(α) is motivated by the work
[Zo18] of V. Zolotov. It is stronger than that in
[Bu19]. The lower bound for
α
is used in sect. 6.2.
Remark 2.4*.*
Axiom P is satisfied, for example,
for the Möbius structure on the boundary at infinity of any
CAT(−1)
space, see [FS12].
Remark 2.5*.*
The canonical Möbius structure
M0
on
X=S1
clearly satisfies Axioms T, M(α), P.
We recall some immediate corollaries from the axioms, see [Bu19]. It follows from axiom (P)
that any semi-metric from
M
with an infinitely remote point is a metric, i.e. it satisfies the triange inequality.
A choice of
ω∈X
uniquely determines the interval
xy⊂Xω
for any distinct
x, y∈X
different from
ω
as the arc in
X
with the end points
x, y
that does not contain
ω.
Axiom M(α) implies the following. Assume for a nondegenerate 4-tuple
q=(x,y,z,u)∈regP4
the interval
xz⊂Xu
is contained in
xy, xz⊂xy⊂Xu.
Then
∣xz∣u<∣xy∣u.
Corollary 2.7**.**
For any harmonic pair
((x,y),(z,u))∈Harm
the pairs
(x,y), (z,u)∈aY
separate each other.
2.4 Increment axiom and a fine topology on M
Increment axiom is not used explicitly in the paper.
However, it is very important in proving that lines with respect to
a Möbius structure are geodesic, see [Bu18]. We recall it here
for convenience of the reader. For more details see [Bu17], where
it has been introduced.
The following is an alternative description of a Möbius structure which
is convenient in many cases. For any semi-metric
d
on
X
we have three cross-ratios
[TABLE]
for
q=(x1,x2,x3,x4)∈regP4,
whose product equals 1, where
∣xixj∣=d(xi,xj).
We associate with
d
a map
Md:regP4→L4
defined by
[TABLE]
where
L4⊂R3
is the 2-plane given by the equation
a+b+c=0.
Two semi-metrics
d, d′
on
X
are Möbius equivalent if and only
Md=Md′.
Thus a Möbius structure on
X
is completely determined by a map
M=Md
for any semi-metric
d
of the Möbius structure, and we often identify a Möbius structure
with the respective map
M.
In this description, axioms (M(α)) and (P) are these:
M((α)) Fix
1>α≥2−1.
Given a 4-tuple
q=(x,y,z,u)∈X4
such that the pairs
(x,y), (z,u)
separate each other, we have
[TABLE]
(P) for every 4-tuple
q=(x,y,z,u)∈X4
we have
[TABLE]
We use notation
regPn
for the set of ordered nondegenerate
n-tuples
of points in
X=S1, n∈N.
For
q∈regPn
and a proper subset
I⊂{1,…,n}
we denote by
qI∈regPk, k=n−∣I∣,
the
k-tuple
obtained from
q
(with the induced order) by crossing out all entries which correspond to elements of
I.
(I) Increment Axiom: for any
q∈regP7
with cyclic order
co(q)=1234567
such that
q247
and
q157
are harmonic, we have
[TABLE]
It is proved in [Bu17, Proposition 7.10] that the canonical Möbius
structure
M0
on the circle
X=S1
satisfies Increment Axiom.
We define a fine topology on
M
as follows. Let
reg+P7⊂X7
be the subset of
regP7
which consists of all
q∈regP7
with the cyclic order. We take on
reg+P7
the topology induced from the standard topology of the 7-torus
X7.
We associate with a Möbius structure
M∈M
a section of the trivial bundle
reg+P7×R4→reg+P7
given by
[TABLE]
for
q=1234567∈reg+P7.
Taking the product topology on
reg+P7×R4,
we define the fine topology on
M
with base given by sets
[TABLE]
where
V
runs over open subsets of
reg+P7×R4.
The class
I
of (strictly) monotone Möbius structures on the circle which satisfy Axiom (I)
contains an open in the fine topology neighborhood of
M0,
see [Bu17, Proposition 7.14].
3 Lines and zigzag paths
Here we briefly recall definitions and some properties of lines and zigzag paths
from [Bu18], [Bu19].
3.1 Lines
Lemma 3.1**.**
[Bu19, Lemma 3.1]** Given
a∈aY
and
x∈X, x∈/a,
there is a uniquely determined
y∈X
such that the pair
(a,b)
is harmonic,
(a,b)∈Hm,
where
b=(x,y).
We denote by
ρa(x)=y
the point
y
from Lemma 3.1. The line with axis
a∈aY
is defined as the set
ha⊂Hm
which consists of all pairs
q=(a,b)
with
b=(x,ρa(x))
where
x
run over an arc in
X
determined by
a.
This is well defined because
ρa:X→X
is involutive,
ρa2=id
(we extend
ρa
to
a=(z,u)
by
ρa(z)=z, ρu=u). In this case, we use notation
xa:=b
and say that
xa∈ha
is the projection of
x
to the line
ha.
For more about lines see [Bu18]. In partial, every line
is homeomorphic to the real line
R,
different points on a line are in strong causal relation, that is,
either of them lies on an open arc in
X
determined by the other one, and vice versa, given
b, b′∈aY
in strong causal relation, there exists a unique line
ha
through
b, b′,
see [Bu18, Lemma 3.2, Lemma 4.2]. In this case, the pair
a∈aY
(or the line
ha)
is called the common perpendicular to
b, b′.
The segmentqq′
of a line
ha
with
q=(a,b), q′=(a,b′)∈ha
is defined as the union of
q, q′
and all
q′′=(a,b′′)∈ha
such that
b′′
separates
b, b′.
The last means that
b
and
b′
lie on different open arcs in
X
determined by
b′′.
The points
q, q′
are the ends of
qq′.
The segment
qq′⊂ha
is homeomorphic to the standard segment
[0,1].
3.2 Distance between harmonic pairs with common axis
Given two harmonic pairs in
q, q′∈Hm
with a common axis, say
q=(a,b)
and
q′=(a,b′),
we define the distance∣qq′∣
between them as
[TABLE]
for some and hence any semi-metric on
X
from
M,
where
a=(x,y), b=(z,u), b′=(z′,u′)∈aY.
One easily checks that every line
ha⊂Hm
with this distance is isometric to the real line
R
with the standard distance.
3.3 Zigzag paths
Every harmonic pair
q=(a,b)∈Hm
has two axes. Thus moving along of a line, we have a possibility
to change the axis of the line at any moment and move along the line
determined by the other axis. This leads to the notion of zig-zag path.
A zig-zag path, or zz-path,
S⊂Hm
is defined as finite (maybe empty) sequence of segments
σi
in
Hm,
where consecutive segments
σi, σi+1
have a common end
q=σi∩σi+1∈Hm
with axes determined by
σi, σi+1.
Segments
σi
are also called sides of
S,
while a vertex of
S
is an end of a side. Given
q, q′∈Hm,
there is a zz-path
S
in
Hm
with at most five sides that connects
q
and
q′
(see [Bu18, Lemma 3.3]). This notion is easily lifted to
Harm.
4 Metric on Hm and filling of M
4.1 Distance δ on Hm
Let
S={σi}
be a zz-path in
Hm.
We define the length of
S
as the sum
∣S∣=∑i∣σi∣
of the length of its sides. Now, we define a distance
δ
on
Hm
by
[TABLE]
where the infimum is taken over all zz-paths
S⊂Hm
from
q
to
q′.
One easily sees that
δ
is a finite pseudometric on
Hm,
see [Bu18, Proposition 6.2].
The following result is obtained in [Bu18], [Bu19].
Theorem 4.1**.**
Assume that a Möbius structure
M
on
X=S1
is strictly monotone, i.e., it satisfies axioms (T), (M(α)), (P). Then
(Hm,δ)
is a complete, proper, geodesic metric space with
δ-metric
topology coinciding with that induced from
X4. If, in addition
M
satisfies Increment axiom, then every line in
Hm
is a geodesic.
Remark 4.2*.*
Since
Harm
is a 2-sheeted covering of
Hm,
all of the conclusions of Theorem 4.1 hold for the space
Harm.
4.2 Filling
Now we define a filling
Y
of a strictly monotone Möbius structure
M
on
X
as the space
(Hm,δ)
of harmonic pairs in
M
with the distance
δ, Y=(Hm,δ).
Our aim is to show under the assumption that
M
in addition satisfies Increment axiom
Y
is a required in Theorem 1.1 hyperbolic space. Sometimes, we pass to
its 2-sheeted covering
Harm
and use the same notation
Y=(Harm,δ).
5 Involutions of X without fixed points
5.1 Some properties
Involution
ρ:X→X
of
X=S1
is an involutive,
ρ2=id,
homeomorphism.
Lemma 5.1**.**
Let
ρ:X→X
be an involution without fixed points. Then for any distinct
x, y∈X
the pairs
a=(x,ρ(x)), b=(y,ρ(y))
separate each other.
Proof.
Assume to the contrary that there are distinct
x, y∈X
such that the respective
a, b∈aY
do not separate each other. Let
X=a+∪a−
decomposition of
X
into (closed) arcs determined by
a.
By the assumption,
b
lies on one of these arcs, say
b⊂a+.
Since
ρ
is an involution, we have
ρ(a)=a
and
ρ(b)=b.
Therefore,
ρ
preserves
a+
permuting its ends
x, ρ(x).
But in this case we observe a fixed point of
ρ
inside of
a+.
This is a contradiction because
ρ
has no fixed points.
∎
Let
ρ:X→X
be an involution without fixed points. The factor
X/ρ
can be identified with the subset
[TABLE]
which is called an elliptic quasi-line.
Lemma 5.2**.**
Let
e=eρ
be an elliptic quasi-line in
aY.
Then for every
s∈aY
there is a unique
t∈e
such that the 4-tuple
(s,t)
is harmonic.
Proof.
First, we show that the image under the involution
ρ
of at least one of the open arcs
s+, s−,
in which
s=(x,y)
separates
X,
misses that arc. Indeed, if
ρ(x)=y,
then
ρ(y)=x.
In that case,
ρ
permutes the arcs
s+, s−
since otherwise,
ρ(s±)=s±,
and thus
ρ
has a fixed point.
By Lemma 5.1 we know that the pairs
(x,ρ(x))
and
(y,ρ(y))
separate each other. Hence,
ρ(s)
and
s
do not separate each other, and we can assume without loss of generality, that
ρ(s)⊂s−.
Then
ρ(s+)
misses
s+
since otherwise
ρ(s+)⊃s+,
and thus
ρ
has a fixed point.
We denote that arc by
s+
and define a function
f:s+→R
by
[TABLE]
where recall
x
is the infinitely remote point for the semi-metric
∣zu∣x.
By the choice of
s+,
we have
ρ(z)=y
for no
z∈s+.
Thus
f
is continuous,
f(z)→∞
as
z→x
and
f(z)→0
as
z→y.
By continuity,
f(z)=1
for some
z∈s+.
Then the 4-tuple
(s,t)
is harmonic for
t=(z,ρ(z))∈e.
If
t′∈e
is another element with harmonic
(s,t′),
then
s
is the common perpendicular to
t, t′
and thus
t, t′
are in the strong causal relation see sect. 3.1, in particular, they
do not separate each other. This contradicts the conclusion of
Lemma 5.1.
∎
Remark 5.3*.*
Let
ρ:X→X
be an involution without fixed points. Applying Lemma 5.2
to any
s∈eρ
we obtain a harmonic pair
(s,t(s))∈Harm
with both
s, t(s)∈eρ.
The set
\widehat{e}_{\rho}=\{(s,t(s)):\,\text{s\in e_{\rho}}\}\subset\operatorname{Harm}
is also called the elliptic quasi-line in
Harm
associated with the involution
ρ.
In this sense, we can lift any elliptic quasi-line
eρ⊂aY
to the uniquely determined elliptic quasi-line
eρ⊂Harm.
It follows from Lemma 5.2 and Lemma 5.1
that
eρ
is invariant under the involution
j:Harm→Harm.
Thus we can speak about elliptic quasi-lines in
Hm.
5.2 Involutions associated with a harmonic 4-tuple
Every harmonic 4-tuple
q=(a,b)∈Harm
generates a pair of involutions
ρq±:X→X
without fixed points as follows. We fix decomposition of
X∖a
into open arcs
a±
with the common ends
a, X=a+∪a−∪a,
and define maps
ρq±:X→X
by
[TABLE]
where
a±
are respective closed arcs. Since
ρb∘ρa(x)=ρa∘ρb(x)
for
x=a,
the maps
ρq±
are well defined and they are continuous involutions of
X
without fixed points. Since
ρa(b)=b
and
ρb(a)=a,
it follows from Lemma 5.2 that
q∈ρρ
for
ρ=ρq±.
Remark 5.4*.*
The maps
ρa, ρb
may not be commuting, thus
ρ+=ρ−
in general, and to define an involution
ρ
we are forced to make a choice of one of the arcs, in which
a
(or
b)
separates
X.
5.3 Canonical decomposition of Harm over X
For every
ω∈X
consider the set
Harmω
which consists of all pairs
q=(a,b)∈Harm
with
ω∈a.
Clearly,
Harm=∪ω∈XHarmω,
and for different
ω, ω′∈X
the sets
Harmω, Harmω′
intersect over the line
h(ω,ω′), Harmω∩Harmω′=h(ω,ω′).
Our aim in this section is to show that every
Harmω
is cobounded in
Harm
uniformly in
ω∈X,
see Corollary 6.11.
5.4 Virtual projection Harm→Harmω
Involutions associated with
q=(a,b)∈Harm
depend on the choice of arcs
a+, a−,
see sect. 5.2. To make that choice canonical,
we fix an orientation of the circle
X=S1
and pass to the 2-sheeted covering
Harm
of
Harm,
see Remark 2.2. Then for every
q=(a,b)∈Harm, a=(x,y)∈X2,
the arc
a+
is defined as the oriented arc from
x
to
y
with the orientation induced by the orientation of
X.
Now, we define
ρq=ρq+.
Lemma 5.5**.**
For every
ω∈X
there is a well defined retraction
hω:Harm→Harmω.
Proof.
Given
q=(a,b)∈Harm
we consider the quasi-elliptic line
e=eρ
associated with the involution
ρ=ρq+:X→X.
Then the line
hs⊂Harm
with
s=(ω,ρ(ω))∈aY
lies in fact in
Harmω
by the definition,
hs⊂Harmω.
By Lemma 5.2, there is a uniquely determined
t∈e
with
(s,t)
harmonic, that is,
(s,t)∈hs.
Now, we put
hω(q)=(s,t).
This canonically defines a retraction
hω:Harm→Harmω
which we call a virtual projection of
Harm
to
Harmω.
∎
6 Diameter of elliptic quasi-lines
In this section, we show that the diameter of any elliptic quasi-line in
Harm
is uniformly bounded above.
6.1 Width of a strip
Recall, see [Bu19, sect. 3.3], that a 4-tuple
p=(a,b)∈X4
with
a=(x,y), b=(u,z)
is a strip if
a, b
are in the strong causal relation and the pairs
(x,z), (u,y)
separate each other. Note that
p′=(b,c)∈X4
with
b=(x,u), c=(y,z)
is also a strip based on the same 4-tuple
(x,y,u,z)∈X4.
Since the pairs
a, b
are in the strong causal relation, there is uniquely determined
common perpendicular
s=(v,w)
to
a, b.
We use notation
p=(a,b,s)
for a strip with common perpendicular
s.
Note that
s
is uniquely determined by
(a,b),
and we add
s
to fix notation.
We define the width of the strip
p
as the length
l=0pt(p)
of the segment
xsus=yszs⊂hs
on the line
hs.
The following estimate has been obtained in [Bu19, Lemma 3.2].
Lemma 6.1**.**
For any strip
p=(a,b,s)
we have
[TABLE]
where
a=(x,y), b=(u,z). A similar estimate holds for the associated strip
p′=(b,c,t),
where
t
is common perpendicular to
b=(x,u), c=(y,z)
[TABLE]
in particular,
0pt(p)⋅0pt(p′)≤4.
6.2 Diameter of elliptic quasi-lines in Harm
Proposition 6.2**.**
There is a constant
D>0
such that for any involution
ρ:X→X
without fixed points we have
[TABLE]
where
eρ⊂Harm
is the elliptic quasi-line associated with
ρ,
see Remark 5.3, and
diam=diamδ
is taken with respect to the distance
δ
in
Harm,
see sect. 4.1.
In the proof, we use the construction from [Bu18, Lemma 3.3], see sect. 3.3,
which gives a zz-path in
Harm
between given
p, q∈eρ
consisting of 5 sides. We estimate the length of sides separately in
Lemmas 6.3, 6.4, 6.6,
6.8.
Let
(z,u), (s,t)∈aY
be pairs which separate each other. They separate
X
into four open arcs. We choose one of them as follows.
Assume (without loss of generality) that
∣us∣∣zt∣≥∣zs∣∣ut∣
(this does not depend of the choice of the metric from our Möbius structure
M,
in particular,
∣us∣t≥∣zs∣t
in any metric
∣∣t
from
M
with infinitely remote point
t).
Then we take the arc
us⊂X
between
u, s
that does not contain
z, t.
Next, we take a metric
∣∣t
from
M
with infinitely remote point
t,
and take points
x, y∈us
(in the order
uxys)
such that
∣ux∣t=∣xy∣t=∣ys∣t=:h.
It follows from continuity and monotonicity of the metric that
such points exist and they are uniquely determined.
Then the pairs
(x,y), (s,t)
as well as the pairs
(x,y), (z,u)
are in the strong causal relation, see sect. 3.1.
There are common perpendiculars
(c,d)
to the pairs
(x,y), (s,t),
and
(e,f)
to the pairs
(x,y), (z,u),
see Figure 1.
These common perpendiculars
are uniquely determined, see sect. 3.1.
We estimate from above the length of the segments
α=x(c,d)t(c,d)=y(c,d)s(c,d)⊂h(c,d)
and
β=x(e,f)u(e,f)=y(e,f)z(e,f)⊂h(e,f).
Lemma 6.3**.**
In notations above we have
∣α∣≤2, ∣β∣≤4.
Proof.
For the strip
p=(a,b,s),
where
a=(x,y), b=(s,t), s=(c,d),
we have
∣α∣=0pt(p).
Lemma 6.1 gives
∣α∣≤2∣xy∣∣st∣∣ys∣∣xt∣=2∣xy∣t∣ys∣t=2.
Similarly, for the strip
p′=(a,b′,s′),
where
b′=(z,u), s′=(e,f),
we have
∣β∣=0pt(p′).
Lemma 6.1 gives
∣β∣≤2∣xy∣∣uz∣xu∣∣yz∣=2∣uz∣t∣yz∣t,
because
∣xu∣t=∣xy∣t.
Let
v∈X
be the point opposite to
u
with respect to the reflection
X→X
determined by the line
h(s,t),
i.e.
u(s,t)=v(s,t)∈h(s,t).
Then
∣sv∣t=∣us∣t≤3h
and
v∈uz
for the open arc
uz⊂X,
that includes
us,
by the choice of the open arc
us⊂X.
By the triange inequality and monotonicity
∣yz∣t≤∣ys∣t+∣sz∣t<h+∣sv∣t≤4h, ∣zu∣t>∣xu∣t=h.
Hence,
∣β∣≤2h4h=4.
∎
Next, we estimate from above the length
∣γ∣
of the segment
γ=c(x,y)e(x,y)=d(x,y)f(x,y)⊂h(x,y)
on the line
h(x,y).
Lemma 6.4**.**
In notation above, we have
∣γ∣≤6.
Proof.
Using notations above, we assume that the points
d, f
lie on the segment
xy⊂Xt.
We consider, first, the case when
e≤t,
that is,
e=t,
or
e
lies on the ray
ut⊂Xt.
In this case, the points
d, f
lies in the order
xfdy
on the segment
xy⊂Xt.
Indeed, the pairs
(c,d), (e,f)
are in the strong causal relation being the perpendiculars to
(x,y).
Thus, the opposite assumption
xdfy
leads to the conclusion that the pairs
(c,d)
and
(s,t)
are in the strong causal relation. This contradicts the fact that
(c,d)
is a perpendicular to
(s,t).
Now, we have
[TABLE]
Note that
∣xd∣t<∣xy∣t=h
by monotonicity, because
d
lies in the interior of the segment
xy⊂Xt.
We have
[TABLE]
because
∣ds∣t=∣cs∣t.
By Lemma 6.3 we have
∣α∣≤2.
Thus
∣dy∣t≥∣cy∣te−2≥∣ys∣te−2=he−2.
It follows that
∣xd∣t/∣dy∣t≤h/(he−2)=e2.
Next, we estimate
∣yf∣t/∣xf∣t
from above. Since
yf⊂xy⊂Xt,
we have
∣yf∣t<∣xy∣t=h
by monotonicity.
Hence
∣xf∣≥e4∣eu∣∣uf∣∣ex∣.
By monotonicity, we have
∣uf∣t>∣ux∣t=h
and
∣ex∣t>∣eu∣t,
where the last inequality uses the assumption
e≤t,
see beginning of the proof. Therefore,
∣xf∣t≥h/e4,
and we conclude that
∣yf∣t/∣xf∣t≤e4.
Hence,
∣γ∣≤ln(e2⋅e4)=6.
Now, we consider the case
e>t,
that is,
e
lies on the ray
zt⊂Xt.
In this case, we cannot garantee that the points
d, f
lies in the order
xfdy
on the segment
xy⊂Xt.
Thus we consider two subcases
(1) The points
d, f
lies in the order
xfdy
on the segment
xy.
We represent the length
∣γ∣
as
[TABLE]
and take a metric from
M
with the infinitely remote point
u.
We have
xc⊂xe⊂Xu,
thus
∣xc∣u/∣xe∣u<1,
and hence
e∣γ∣≤∣ye∣u/∣yc∣u.
Next, we use that
[TABLE]
by Lemma 6.3.
Since
∣fz∣u=∣ze∣u,
we obtain
∣ye∣u≤e4∣fy∣u.
Since
ys⊂yc⊂Xu,
we have
∣ys∣u<∣yc∣u,
which gives
e∣γ∣≤e4∣fy∣u/∣ys∣u.
Using the metric inversion, see (1), we pass
to the metric with infinitely remote point
t
and use that
∣ys∣t=h, ∣fy∣t<∣xy∣t=h:
[TABLE]
[TABLE]
Using that
∣fu∣t>∣ux∣t=h
by monotonicity and
∣su∣t≤3h
by the triange inequality, we finally obtain
e∣γ∣≤e4∣fy∣u/∣ys∣u≤e4∣su∣t/∣fu∣t≤e4⋅3h/h=e4⋅3.
Hence,
∣γ∣≤4+ln3.
(2) The points
d, f
lies in the order
xdfy
on the segment
xy.
Recall that the pairs
(c,d)
and
(e,f)
are in the strong causal relation, and the pairs
(c,d), (s,t)
separate each other. Thus
c
lies on the ray
et⊂Xt
which does not contain
d.
Hence, this time we have
xe⊂xc⊂Xt
and
[TABLE]
By monotonicity,
∣xe∣t<∣xc∣t
and we conclude that
e∣γ∣<∣yc∣t/∣ye∣t.
To estimate
∣yc∣t
from above, we use that
[TABLE]
by Lemma 6.3. Since
∣ds∣t=∣cs∣t
and
dy⊂xy⊂Xt,
we obtain
∣yc∣t≤e2∣dy∣t≤e2∣xy∣t=e2h.
On the other hand,
ys⊂ye⊂Xt.
Thus
∣ye∣t>∣ys∣t=h
by monotonicity. Therefore
e∣γ∣≤e2
and
∣γ∣≤2.
∎
Let
p=((z,u),(z′,u′)), q=((s,t),(s′,t′))∈eρ
be given distinct harmonic pairs of pairs from
aY.
Then the pairs
(z,u), (s,t)∈aY
separate each other being different members of the elliptic quasi-line in
eρ⊂aY.
Assume as above (without loss of generality) that
∣us∣∣zt∣≥∣zs∣∣ut∣.
Then we take the arc
us⊂X
between
u, s
that does not contain
z, t.
We also assume that
t′, z′
lie on the arc in
X
between
s, t
that contains
su.
Remark 6.5*.*
In this case,
sz′⊂st′⊂su⊂Xt.
Indeed, since
z=ρ(u), s′=ρ(t′),
the pairs of points
(z,u), (s′,t′)
separate each other. Thus the opposite assumption
u∈st′
would imply
∣su∣t<∣st′∣t=∣ss′∣t<∣sz∣t,
a contradiction with our assumption
∣us∣∣zt∣≥∣zs∣∣ut∣.
To show that
z′∈st′,
we fix
q=((s,t),(s′,t′))
and move
u
from
t′
to
t
along the arc
t′t⊂st.
Then
z′
moves from
s
to
t′
along the arc
st′⊂st.
Since
u∈t′t
by the first part of the argument, we see that
z′∈st′.
Lemma 6.6**.**
In notations above, assume that
sy⊂st′⊂Xt
(recall that
st′⊂su,
see Remark 6.5). Then
∣μ∣≤ln3,
where the segment
μ=d(s,t)t(s,t)′=c(s,t)s(s,t)′
lies on the line
h(s,t).
Proof.
We have
[TABLE]
Since
sy⊂sd⊂su⊂Xt,
we estimate
h=∣sy∣t≤∣sd∣t≤∣su∣t≤3h.
Since
sy⊂st′⊂su,
we estimate
h=∣sy∣t≤∣st′∣t≤∣su∣t≤3h.
Thus
∣μ∣≤ln3.
∎
Lemma 6.7**.**
In notations above, we have
∣zz′∣t≥h.
Proof.
If
sy⊂sz′⊂Xt,
then
∣zz′∣t≥∣sz′∣≥∣sy∣t=h.
Thus we assume that
sz′⊂sy.
Then
ux⊂uz′
and hence
∣uz′∣t≥∣ux∣t=h.
Since the pair of pairs
p=((z,u),(z′,u′))
is harmonic, we have
∣zz′∣∣uu′∣=∣zu′∣∣z′u∣
in any metric of the Möbius structure
M.
Note that
t
lies on the arc in
X
between
u, u′
that does not contain
z, z′.
Thus we have
∣uu′∣t≤∣uz′∣t+∣z′z∣t+∣zu′∣t
by the triangle inequality in the metric
∣∣t
with infinitely remote point
t.
Using notations
∣zu′∣t=:a, ∣z′u∣t=:b, ∣zz′∣t=ε,
we conclude that
ab≤ε(a+b+ε).
Therefore,
[TABLE]
But
b=∣z′u∣t≥h.
Hence
∣zz′∣t≥h
also in this case.
∎
Lemma 6.8**.**
In notations above, we have
∣ν∣≤ln18,
where the segment
ν=f(u,z)z(u,z)′=e(u,z)u(u,z)′
lies on the line
h(u,z).
Proof.
We first show that
∣uz′∣t≥h.
If
sz′⊂sy,
then
xy⊂uz′
and hence
h=∣xy∣t≤∣uz′∣t.
Thus we assume that
sy⊂sz′.
Then
∣sy∣s′≤∣sz′∣s′<∣zz′∣s′.
As in Lemma 6.7 applied to a metric
∣∣s′
with infinitely remote point
s′,
we obtain
∣uz′∣s′>∣zz′∣s′>∣sy∣s′.
The metric inversion with respect to
t
gives
[TABLE]
Using that
∣sy∣t=h
and by monotonicity
∣ss′∣t<∣z′s′∣t, ∣ys′∣t<∣us′∣t,
we obtain
∣uz′∣t>h.
Now, using monotonicity, Lemma 6.7 and the first part of
the proof, we have the following two-sided estimates
for
∣uz′∣t, ∣zf∣t, ∣uf∣t
and
∣zz′∣t:
h≤∣uz′∣t≤∣us∣t≤3h,
h=∣sy∣t≤∣zf∣t≤∣zu∣t≤∣uv∣t≤6h,
h=∣ux∣t≤∣uf∣t≤∣uy∣t≤2h,
h≤∣zz′∣t≤∣uz∣t≤6h,
where the point
v∈Xt
is determined in Lemma 6.3.
Since
[TABLE]
for any metric from the Möbius structure
M,
this gives
∣ν∣≤ln18.
∎
Now, we estimate the length
of the zz-path
σ=μαγβν
in a particular case, when
sy⊂st′⊂Xt.
Lemma 6.9**.**
In notations at the beginning of
the section, assume that
sy⊂st′⊂Xt
for the zz-path
σ=μαγβν
between
p=((z,u),(z′,u′))
and
q=((s,t),(s′,t′))∈eρ.
Then
∣σ∣≤D
with
D=12+ln54<16.
Proof.
We have
∣α∣≤2, ∣β∣≤4
by Lemma 6.3,
∣γ∣≤6
by Lemma 6.4,
∣μ∣≤ln3
by Lemma 6.6
and
∣ν∣≤ln18
by Lemma 6.8. Note that the assumption
sy⊂st′
is only used in the estimate for
∣μ∣.
Thus
∣σ∣≤∣μ∣+∣α∣+∣γ∣+∣β∣+∣ν∣≤D.
∎
In notations above, assume that
the harmonic pair
q=((s,t),(s′,t′))∈eρ
is fixed. Then the harmonic pair
p=((z,u),(z′,u′))∈eρ
is uniquely determined by the point
u
on the arc
tt′⊂X
between
t, t′
that does not contain
s, s′
because
z=ρ(u)
and
(z′,u′)∈aY
is determined by
(z,u),
see Lemma 5.2. The point
u
in its own turn determines
x, y∈us.
The conclusion of Lemma 6.9 holds
for
u∈tt′
such that
sy⊂st′.
This gives an upper bound for the distance
∣us∣t,
in particular,
u
is separated from
t.
Let
u0∈tt′
by maximal with this property, i.e.
y=t′
for
y=y(u0).
At the moment, we do not have a required estimate of
∣σ∣
for
u
on the (open) arc
tu0⊂tt′.
To fill in this gap, we apply the same construction for
p=((z,u),(z′,u′))
and
q′=j(q)=((s′,t′),(s,t))
assuming without loss of generality that
∣us′∣∣zt′∣≥∣zs′∣∣ut′∣
and choosing the arc
s′u⊂X
between
s′, u
that does not contain
z, t′.
Then
u
determines as above
x′, y′∈s′u
with
∣ux′∣t′=∣x′y′∣t′=∣y′s′∣t′=:h′.
Now, the conclusion of Lemma 6.9 holds for
u∈tt′
such that
s′y′⊂s′t⊂Xt′.
Let
u1∈tt′
be maximal with this property, i.e.
y′=t
for
y′=y′(u1).
We show that the subarcs
u0t′
and
u1t
in
tt′
overlap. At this point, we need the condition
α≥2−1
in Axiom M(α).
Lemma 6.10**.**
In notations above the arcs
u0t′, u1t⊂X
overlap,
u0t′∩u1t=∅.
Proof.
By the assumption on
u0,
we have
h=∣u0x∣t=∣xy∣t=∣xt′∣t.
Thus the pair
((x,t),(u0,t′))
is harmonic. Then by Axiom M(α)
∣u0t′∣t≥2h.
Taking the metric inversion, we obtain
[TABLE]
Again, since
∣u1x′∣t′=∣x′t∣t′=h′,
the pair
((x′,t′),(u1,t))
is harmonic. By Axiom M(α),
∣u1t∣t′≥2h′.
We show that
2hh′≥1.
Note that
h=∣st′∣t=∣ss′∣t
by harmonicity of
q,
and
h′=∣s′t∣t′=∣ss′∣t′
by harmonicity of
q′=j(q).
Taking the metric inversion, we have
[TABLE]
Since
∣s′t′∣t≤∣s′s∣t+∣st′∣t
by the triange inequality, we see that
∣s′t′∣t≤2h.
Then
[TABLE]
Therefore,
2hh′≥1.
Now,
∣u1t∣t′≥2h′≥2h1≥∣u0t∣t′.
Hence
u0t′∩u1t=∅
by monotonicity.
∎
We use notations
introduced above. For
p, q∈eρ,
p=((z,u),(z′,u′)), q=((s,t),(s′,t′)),
and
x, y∈su⊂Xt
with
∣ux∣t=∣xy∣t=∣ys∣t,
if
∣ut′∣t≤∣u0t′∣t,
then
sy⊂st′
and
δ(p,q)≤D
by Lemma 6.9. In particular,
this condition is fulfilled for
p=q′=j(q)=((s′,t′),(s,t))
because then
u=t′.
Thus
δ(q′,q)≤D.
In the opposite case,
∣ut′∣t>∣u0t′∣t,
we have
∣ut∣t′≤∣u1t∣t′
by Lemma 6.10. Hence
δ(p,q′)≤D.
In this case,
δ(p,q)≤δ(p,q′)+δ(q′,q)≤2D
by the triangle inequality. Therefore,
diameρ≤2D
with
D<16.
∎
Corollary 6.11**.**
The subspace
Harmω⊂Harm
is cobounded in
Harm
uniformly in
ω∈X,
that is, for any
q∈Harm, ω∈X
we have
distδ(q,Harmω)≤D
for some universal constant
D>0.
Proof.
We take one of two involutions associated with
q∈Harm,
see sect. 5.4, and denote it by
ρ.
Let
eρ⊂Harm
be elliptic quasi-line associated with the involution
ρ:X→X.
Then
q∈eρ,
see sect. 5.2, and by Lemma 5.5,
hω(q)∈eρ∩Harmω.
Thus
distδ(q,Harmω)≤δ(q,hω(q))≤diamδ(eρ)≤D
by Proposition 6.2.
∎
7 Hyperbolic approximation of Xω
A hyperbolic approximation is a kind of a hyperbolic cone over a metric
space, see [BS07]. A specific feature of a hyperbolic approximation
of a metric space is that it is defined via families of metric balls in the space
in such a way to reflect their combinatorics.
7.1 Definition
The set
Harmω
of harmonic 4-tuples with common entry
ω
can be identified with the set of metric balls in
Xω.
Indeed, every
q=((a,b),(o,ω))∈Harmω
determines the sphere
Sr(o)=(a,b)
because
o
is the midpoint between
a, b, ∣ao∣ω=∣ob∣ω=:r,
and hence the ball
B_{r}(o)=\{x\in X_{\omega}:\,\text{|ox|_{\omega}\leq r}\}\subset X_{\omega}
with
∂Br(o)=Sr(o).
Vice versa, given a ball
Br(o)⊂Xω
of radius
r>0
centered at
o,
we have a 4-tuple
q=((a,b),(o,ω)),
where
(a,b)=∂Br(o),
which is harmonic,
q∈Harmω,
because
o
is the midpoint between
a, b.
A (finite or infinite) sequence of spheres
Sr(oi)=(ai,bi)⊂Xω
is said to be a harmonic chain of radiusr
if the pair
((ai,bi),(ai+1,bi+1))
is harmonic for every
i.
Assuming that an orientation of (and hence an order on)
Xω
is fixed, and that
ai<bi, ai+1<bi+1, ai<ai+1<bi,
we observe that
bi+1>bi
because the pairs
(ai,bi), (ai+1,bi+1)
separate each other and
ai+1<bi+1.
Moreover,
oi<ai+1
since otherwise
bi+1=ω
or
bi+1<ai+1.
Similarly,
bi<oi+1.
Speaking about harmonic chains of spheres, we mean that these
assumptions are always satisfied. Note that then the pairs
(ai,bi), (ai+2,bi+2)
are in strong causal relation. Indeed, this is equivalent to
bi<ai+2,
which is fulfilled because otherwise
ai+2≤bi
and hence
ai+2<oi+1.
But this contradicts the inequality
oi+1<ai+2.
We fix
0<σ≤1/24
and for every
k∈Z
let
Vk⊂Harmω
be an infinite in both directions harmonic chain of radius
r=σk.
We put
V=∪k∈ZVk⊂Harmω
and define a harmonic hyperbolic approximation
Z=Z(σ)
of
Xω
with parameter
σ
as a graph with the vertex set
V.
We consider vertices in
V
as spheres (balls) of respective harmonic chains. For any
v∈V
we denote
B(v)
the respective ball in
Xω.
Two vertices
v, v′∈V
are connected by an edge if and only is they lie on one and the same level
Vk
and are in this case neighboring spheres,
v=Sr(oi), v′=Sr(oj)
with
∣i−j∣=1
and
r=σk,
or
v∈Vk, v′∈Vl
with
∣k−l∣=1
and in this case the respective ball with the larger level is contained in the respective
ball with the smaller lever, i.e.
Br(oi)⊂Br′(oj)
if
r=σk+1, r′=σk.
An edge
vv′⊂Z
is called horizontal, it its vertices lie on one and the same level,
v, v′∈Vk
for some
k∈Z. Other edges are called radial. The level function
ℓ:V→Z
is defined by
l(v)=k
for
v∈Vk.
Since every level
Zk⊂Z, k∈Z,
is connected, the graph
Z
is connected. We endow
Z
with path metric assuming that the length of every edge is 1. We denote
by
∣vv′∣
the distance between points
v, v′∈V
in
Z.
Note that
Z
is geodesic because it is connected and distances between vertices take
integer values.
7.2 Geodesics in Z
The construction of the (harmonic) hyperbolic approximation
Z
here is slightly different from that in [BS07]. Thus
we basically follow [BS07, sect. 6.2] with appropriate adaptation
of the arguments.
Lemma 7.1**.**
For every
v∈V
there is a vertex
w∈V
with
ℓ(w)=ℓ(v)−1
connected with any
v′∈V, ℓ(v′)=ℓ(v), ∣vv′∣≤1,
by a radial edge.
Proof.
There are two neighbors
v′, v′′
of
v
in
Z,
sitting on the same level as
v,
∣vv′∣, ∣vv′′∣≤1.
One of them,
v′,
is on the left to
v,
the other one,
v′′
is on the right to
v.
Let
v′=(a′,b′), v′=(a′′,b′′).
Then
∣a′b′′∣ω≤6r′,
where
r′=σk+1
for
k+1=ℓ(v).
On the other hand, for every neighboring
w, w′∈Vk,
w=(c,d), w′=(c′,d′),
the pair
((c,d),(c′,d′))
is harmonic. Thus
∣c′d∣ω∣cd′∣ω=∣cc′∣ω∣dd′∣ω.
Hence
[TABLE]
for
r=σk
because
∣cc′∣ω≥r, ∣dd′∣ω≥r
and
∣cd′∣ω≤4r.
For the neighbors
v′, v′′
of
v
we have
v′∪v∪v′′=a′b′′⊂Xω.
Since
σ=r′/r≤1/24,
we have
∣a′b′′∣ω≤6r′≤r/4.
The balls
{w∈Vk}
cover
Xω.
Assume that there is
w∈Vk
such that
(v′∪v∪v′′)⊂w.
Then the vertices
v, v′, v′′∈Z
are connected with
w
by radial edges.
Otherwise
a′b′′
is covered by no
w∈Vk.
Then there are at most two neighboring
w=(c,d), w′=(c′,d′)∈Vk
which cover
a′b′′, a′b′′⊂cd∪c′d′.
Assuming that
w
is left to
w′,
we observe that the intersection
w∩w′=c′d.
Since
∣a′b′′∣ω≤∣c′d∣ω
by the estimate above, we see
that
a′b′′
is contained in one of
w, w′
in contradiction with our assumption.
∎
Lemma 7.2**.**
Any vertices
v, v′∈V
can be connected in
Z
by a geodesic
γ
which consists of at most two radial subsegments
γ′, γ′′⊂γ
and at most one horizontal edge between them. If there is such an edge,
then it lies on the lowest level of the geodesic. Otherwise the unique
common vertex
w
of
γ′, γ′′
is the lowest level vertex of
γ.
The proof proceeds exactly as in [BS07, Lemma 6.2.6]
using Lemma 7.1 and that fact that for any harmonic chain
Vk, k∈Z
two balls
v, v′∈Vk
intersect if and only if they are neighboring in
Vk.
Thus we omit it.
7.3 Hyperbolicity of Z
The Gromov product of
v, v′
with respect to
u
in a metric space
Z
is defined by
[TABLE]
A metric space
Z
is said to be
δ-hyperbolic, δ≥0,
if for any
v, v′, v′′∈Z
and a base point
u∈Z,
we have
[TABLE]
Now, we come back to our harmonic geodesic approximation
Z.
Lemma 7.3**.**
Assume that
∣vv′∣≤1
for vertices
v, v′∈Z
of one and the same level,
ℓ(v)=ℓ(v′).
Then
∣ww′∣≤1
for any vertices
w, w′∈Z
adjacent to
v, v′
respectively and sitting one level below.
Proof.
The balls
B(w), B(w′)
intersect because
B(v)⊂B(w), B(v′)⊂B(w′)
and the balls
B(v), B(v′)
intersect.
Since
w, w′
are members of a harmonic chain, they are adjacent in
Z, ∣ww′∣≤1.
∎
From this we immediately obtain.
Corollary 7.4**.**
For any two radial geodesics
γ, γ′⊂Z
with common ends, the distance in
Z
between vertices of
γ
and
γ′
of the same level is at most 1.
It is convenient to use the following terminology. Let
V′⊂V
be a subset. A point
u∈V
is called a cone point for
V′
if
ℓ(u)≤infv∈V′ℓ(v)
and every
v∈V′
is connected to
u
by a radial geodesic. A cone point of maximal level is called
a branch point of
V′.
Lemma 7.5**.**
For any two points
v, v′∈V
there is cone point and, hence, a branch point.
Proof.
By Lemma 7.2,
v, v′
can be connected in
Z
by a geodesic
γ
which contains at most one horizontal edge. If there is no horizontal edge in
γ,
then the lowest level point
w
of
γ
is a branch point of
v, v′.
Otherwise, let
uu′⊂γ
be the horizontal edge. It lies on the lowest level of
γ.
Without loss of generality, we assume that
vu, v′u′
are radial geodesics.
By Lemma 7.1, there is
w∈V
with
ℓ(w)=ℓ(γ)−1
which is connected to
u, u′
by radial edges. Taking concatenation
vuw, v′u′w
we see that
w
connected to
v, v′
by radial geodesics. Hence,
w
is a cone point of
v, v′.
∎
Note that if
u
is a cone point of
v, v′
and
w
is their branch point, then
(v∣v′)u=∣uw∣
in the case the geodesic
vv′
has no horizontal edge, and
(v∣v′)u=∣uw∣+1/2
otherwise. In particular,
∣uw∣≥(v∣v′)u−1/2
is either case.
Lemma 7.6**.**
Let
u∈V
be a cone point of
v, v′∈V,
γ=uv, γ′=uv′
radial geodesics. Then for any
y∈γ, y′∈γ′
sitting one the same level
ℓ(y)=ℓ(y′)≤ℓ(w),
where
w
is a branch point of
v, v′,
we have
∣yy′∣≤2.
Proof.
Concatenations
vwu, v′wu
are radial geodesics in
Z.
By Corollary 7.4, we have
∣yy′′∣≤1
for
y∈γ, y′′∈vwu
sitting on the same level,
ℓ(y)=ℓ(y′′),
and similarly
∣y′y′′∣≤1
for
y′∈γ′, y′′∈v′wu
with
ℓ(y′)=ℓ(y′′).
For
ℓ(y)=ℓ(y′)≤ℓ(w)
we can choose
y′′∈wu
with
ℓ(y′′)=ℓ(y)=ℓ(y′),
and thus
∣yy′∣≤∣yy′′∣+∣y′′y′∣≤2.
∎
We need the following Proposition from [BS07, Proposition 6.2.9],
for which we give a different proof.
Lemma 7.7**.**
Let
v, v′, v′′∈V
and let
w, w′, w′′
be branch points for the pairs of vertices
{v′,v′′}, {v,v′′}
and
{v,v′}
respectively. Let
u
be a cone point of
{w,w′,w′′}.
Then
[TABLE]
with
δ=5/2.
Proof.
We put
t0=min{∣uw∣,∣uw′∣}
and let
γ, γ′, γ′′
be radial geodesics between
u
and
v, v′, v′′
respectively. Assume that
y∈γ, y′∈γ′, y′′∈γ′′
satisfy
∣uy∣=∣uy′∣=∣uy′′∣=t0.
By Lemma 7.6 we have
∣yy′′∣, ∣y′y′′∣≤2.
Thus by the triangle inequality
∣yy′∣≤4.
By monotonicity of the Gromov product
[TABLE]
By the remark above
t0≥min{(v∣v′′)u,(v′∣v′′)u}−1/2.
Hence, the claim.
∎
Using argument of [BS07, Proposition 6.2.10], we obtain.
Proposition 7.8**.**
Any hyperbolic harmonic
approximation
Z
of
Xω
is a geodesic
δ-hyperbolic
space with
δ=5.
∎
8 Xω and Z are quasi-isometric
Our aim is to show that for every
ω∈X
the space
Xω
and its hyperbolic harmonic approximation
Z=Z(σ)
are quasi-isometric. Let
V
be the vertex set of
Z.
By definition, we have an inclusion
f:V↪Xω.
We show that
f
is a quasi-isometry with respect to the metric on
Z
and the
δ-metric
on
Xω.
8.1 Estimates from above
In this section we establish estimates from above, that is, we show that
there is a constant
D=D(σ)
depending only on
σ
such that for every edge
vv′
of
Z
we have
δ(v,v′)≤D.
For horizontal edges this is proven in Lemma 8.1,
and for vertical edges in Lemma 8.5.
Fix
ω∈X, r>0.
Then the sphere
Sr(o)⊂Xω
of radius
r
centered at
o∈Xω
determines the harmonic pair
((a,b),(o,ω))∈Harm,
where
Sr(o)=(a,b).
Lemma 8.1**.**
Fix
ω∈X, r>0,
and consider two spheres
Sr(o)=(a,b), Sr(o′)=(a′,b′)
in
Xω
such that the pair of pairs
((a,b),(a′,b′))
is harmonic. Then the
δ-distance
between harmonic
q=((a,b),(o,ω))
and
q′=((a′,b′),(o′,ω))
is at most
2ln4,
δ(q,q′)≤2ln4.
Proof.
We fix an orientation of
Xω
and assume without loss of generality that the ordered pairs
(a,b), (a′,b′)
agree with the orientation, and
a
precedes
b′.
Note that
b
is not on the segment
o′b′⊂Xω, b∈o′b′,
see sect. 7.1.
The harmonic pairs
q=((a,b),(o,ω))
and
q=((a,b),(a′,b′))
have the common axis
(a,b).
Thus the distance
l
between
q, q
along
h(a,b)
is computed as
[TABLE]
because
∣ao∣ω=r=∣ob∣ω.
Since
q
is harmonic, we have
∣aa′∣∣bb′∣=∣a′b∣∣ab′∣.
Thus
el=∣bb′∣ω∣ab′∣ω.
By the triange inequality and monotonicity,
∣ab′∣ω≤∣ab∣ω+∣bb′∣ω≤∣ab∣ω+∣a′b′∣ω≤4r.
By the remark above,
∣bb′∣ω≥∣o′b′∣ω=r.
Therefore,
l≤ln4.
Similarly,
q
and
q′
have the common axis
(a′,b′),
and
l′=∣q′q∣≤ln4.
Hence,
δ(q,q′)≤∣qq∣+∣q′q∣≤2ln4.
∎
Corollary 8.2**.**
For every horizontal edge
vv′⊂Z
we have
δ(v,v′)≤C
with
C≤2ln4.
Proof.
Indeed, the vertices
v, v′
of any horizontal edge in
Z
satisfy the condition of Lemma 8.1.
∎
Lemma 8.3**.**
Fix
ω∈X, 0<σ≤1/24,
and consider two spheres
Sr(o)=(a,b), Sr′(o′)=(a′,b′)
in
Xω,
where
r=σk, r′=σk+1
for some
k∈Z,
such that
o
lies in the open interval
(a′b′)⊂Xω, o∈(a′b′).
Then the spheres
(a,b), (a′,b′)
do not separate each other in
X.
Let
h⊂Harm
be the unique line that contains
(a,b)
and
(a′,b′).
Then the distance
l
between
(a,b)
and
(a′,b′)
along
h
is estimated above as
l≤2/σ.
independent of
k.
Proof.
To estimate
l
we use Lemma 6.1. We assume as in the proof of
Lemma 8.1 that the ordered pairs
(a,b), (a′,b′)
agree with a fixed orientation of
Xω.
Since both
o′,o
lies in the interval
(a′b′)⊂Xω,
we have
∣o′o∣≤∣a′b′∣≤2r′.
Then
∣a′o∣≤∣a′o′∣+∣o′o∣≤3r′<r
because
σ≤1/24.
Hence
a<a′,
similarly
b′<b,
and the pairs
(a,b), (a′,b′)⊂X
do not separate each other. Thus
p=((a,b),(a′,b′))
is a strip. By Lemma 6.1 we have
[TABLE]
Since
o∈(a′b′),
it holds
∣aa′∣ω, ∣bb′∣ω≤r.
Axiom (M(α)) gives
∣ab∣ω≥2r, ∣a′b′∣ω≥2r′.
Thus
l≤2r2/2rr′=2/σ.
∎
Lemma 8.4**.**
Fix
ω∈X, 0<σ≤1/24,
and consider two spheres
Sr(o)=(a,b), Sr′(o′)=(a′,b′)
in
Xω,
where
r=σk, r′=σk+1
for some
k∈Z,
such that
o
lies in the open interval
(a′b′)⊂Xω, o∈(a′b′).
Then the
δ-distance
between harmonic
q=((a,b),(o,ω))
and
q′=((a′,b′),(o′,ω))
is estimated above as
δ(q,q′)≤2/σ+2ln3
independent of
k.
Proof.
We fix an orientation and hence the respective order on
Xω.
If
o′=o,
then
q, q′
lie on the line
h(o,ω),
and in this case
δ(q,q′)=∣qq′∣=ln(r/r′)=ln(1/σ)<1/σ.
Thus we assume that
o′=o.
Without loss of generality, we assume that
o′<o
with respect to the order on
Xω.
We also assume that
a<b, a′<b′.
As in Lemma 8.3, the pairs
(a,b)
and
(a′,b′)
do not separate each other. Let
(c,d)
be the common perpendicular to
(a,b)
and
(a′,b′),
h=h(c,d)⊂Harm
the unique line containing
(a,b)
and
(a′,b′).
Then we have a zz-path in
Harm
between
q, q′
which consists of 3 sides.
First, one goes from
q
to
q=h(a,b)∩h
along
h(a,b).
We denote the respective distance by
m.
Then one goes along
h
from
q
to
q′=h∩h(a′,b′).
By Lemma 8.3, the respective distance
l
is estimated above as
l≤2/σ.
Finally, one goes from
q′
along
h(a′,b′)
to
q′.
We denote the respective distance by
t.
Thus we need to estimate above
m
and
t.
We assume without loss of generality that
c∈(a′b′).
Note that
c∈o′o,
since otherwise
c
is equal neither
o
nor
o′
because
o′=o,
and
d
must lie simultaneously left to
a
and right to
b′,
which is impossible.
We consider two cases
(1) c<o′
and
(2) o<c.
Case (1). We have
[TABLE]
Using that
∣bc∣ω≤∣ab∣ω≤2r
and
∣a′b′∣ω≤2r′,
we have
∣ac∣ω≥∣aa′∣ω≥r−∣a′b′∣ω≥r−2r′,
and obtain
[TABLE]
On the other hand,
[TABLE]
thus
∣bd∣ω/∣ad∣ω≤3.
Now we compute
t.
By the assumption
c<o′<o
we have
d<a.
Thus
∣b′d∣ω≤∣bd∣ω, ∣a′d∣ω≥∣ad∣ω
and we obtain
[TABLE]
Thus
t≤ln3.
Case (2). This is obtained similarly to case (1) by interchanging
a, b
and
a′, b′.
Finally,
δ(q,q′)≤m+l+t≤2/σ+2ln3.
∎
Lemma 8.5**.**
Fix
ω∈X, 0<σ≤1/24,
and for a sphere
Sr(o)=(a,b)⊂Xω
consider a maximal harmonic chain of spheres
Sr′(oi)=(ai′,bi′)⊂Xω, i=1,…,n,
that is contained in
(a,b),
where
r=σk, r′=σk+1
for some
k∈Z.
Then the
δ-distance
between harmonic
q=((a,b),(o,ω))
and
qi′=((ai′,bi′),(oi′,ω)),
is estimated above as
δ(q,qi′)≤c1/σ+c2
for every
i=1,…,n
independent of
k,
where
c1≤2+4ln4, c2=2ln3.
Proof.
The segments
ai′ai+1′, i=1,…,n
have disjoint interiors, and their union cover the union of spheres
Sr′(oi).
Thus
[TABLE]
On the other hand,
∣ai′ai+1′∣≥∣ai′oi′∣=r′
because
oi′
lies in the interval
ai′ai+1′,
see sect. 7.1. Thus
n≤2r/r′=2/σ.
There is
j∈{1,…,n}
such that
o∈(aj′,bj′).
By Lemma 8.4, we have
δ(q,qj′)≤2/σ+2ln3.
Using Lemma 8.1, we obtain
δ(q,qi′)≤δ(q,qj′)+δ(qj′,qi′)≤2/σ+2ln3+2nln4
for every
i=1,…,n.
Therefore,
δ(q,qi′)≤c1/σ+c2,
where
c1≤2+4ln4, c2=2ln3.
∎
Corollary 8.6**.**
For every vertical edge
vv′⊂Z
we have
δ(v,v′)≤C
with
C≤2/σ+2ln3.
Proof.
Indeed, vertices
v, v′
of any vertical edge in
Z
satisfy the condition of Lemma 8.5.
∎
Corollary 8.7**.**
For each pair of vertices
v, v′∈V
we have
δ(v,v′)≤C∣vv′∣Z
with
C≤2/σ+2ln3.
Proof.
Let
γ⊂Z
be a geodesic between
v, v′, γ=v0…vnv0=v, vn=v′,
with edges
vivi+1, i=0,…,n−1.
By definition, the length of
γ
is the number of edges it consists,
∣vv′∣Z=∣γ∣Z=n.
By Corollaries 8.2, 8.6 we have
δ(vi,vi+1)≤C∣vivi+1∣Z=C.
Thus
δ(v,v′)≤C∣vv′∣Z.
∎
8.2 Estimates from below
We fix an orientation of
X.
Then we have a respective order on each
Xx, x∈X,
induced by the orientation.
Lemma 8.8**.**
Fix
ω∈X, r>0,
and let
Sr(o)=(a,b), Sr(o′)=(a′,b′)⊂Xω
be separated spheres with the order
aba′b′.
Then the
δ-distance
between harmonic pairs
q=((a,b),(o,ω)), q′=((a′,b′),(o′,ω))∈Harm,
is estimated above as
δ(q,q′)≤C(r,∣ba′∣ω),
with
C(r,∣ba′∣ω)≤4ln(3∣ba′∣ωr+r∣ba′∣ω).
Proof.
Since pairs
(a,b), (a′,b′)
are in strong causal relation, there is a common perpendicular
h=h(x,y)
to them. We assume without loss of generality that
x<y
with respect to our order on
Xω.
This implies that
o<x
and
y<o′.
We have two harmonic
p=((a,b),(x,y)), p′=((x,y),(a′b′))∈Harm,
and we denote by
α
the segment of
h(a,b)
between
q
and
p,
by
γ
the segment of
h(x,y)
between
p
and
p′,
and by
β
the segment of
h(a′,b′)
between
p′
and
q′.
Then
σ=αγβ
is a zz-path between
q, q′
which consists of three sides
α, γ, β.
Since
δ(q,q′)≤∣σ∣,
we estimate above
∣σ∣=∣α∣+∣β∣+∣γ∣.
We have
[TABLE]
because
∣ao∣ω=r=∣bo∣ω.
Similarly,
[TABLE]
because
∣a′o′∣ω=r=∣b′o′∣ω.
Next
[TABLE]
Harmonicity of
p
means that
∣bx∣∣ay∣=∣ax∣∣by∣,
and harmonicity of
p′
means that
∣a′y∣∣b′x∣=∣a′x∣∣b′y∣.
Using this, we obtain
[TABLE]
Since
bx⊂ob⊂Xω
and
a′y⊂a′o′⊂Xω,
we have
∣bx∣ω≤r, ∣a′y∣ω≤r
by monotonicity. Thus by the triange inequality
∣ay∣ω≤∣ab∣ω+∣ba′∣ω+∣a′y∣ω≤3r+∣ba′∣ω.
Similarly,
∣b′x∣ω≤3r+∣ba′∣ω.
By monotonicity
∣xa′∣ω≥∣ba′∣ω, ∣by∣ω≥∣ba′∣ω,
∣b′y∣ω≥∣o′b′∣ω=r, ∣ax∣ω≥∣ao∣ω=r.
Therefore,
[TABLE]
and the required estimate follows.
∎
Lemma 8.9**.**
Fix
ω∈X, r>0,
and let
Sr(oi)=(ai,bi), i∈Z,
be a harmonic chain in
Xω.
Then for every sphere
Sr(o)=(a,b)⊂Xω
we have
δ(q,qi)≤D=4ln160,
where
q=((a,b),(o,ω)), qi=((ai,bi),(oi,ω))∈Harmω
with
i∈Z
such that
ab∩aibi=∅.
Proof.
If
o=oi
for some
i∈Z,
then
q=qi,
and there is nothing to prove. Thus we assume that
o=oi
for no
i∈Z,
and furthermore we assume without loss of generality that
i=0,
and we have the following order
aa0b
of points on
Xω.
Since
a0b0∩akbk=∅
for
∣k∣≥2,
spheres
Sr(o), Sr(ok)
are separated. For
k≥4,
the spheres
Sr(o)
and
Sr(ok)
are separated by at least the sphere
Sr(o2),
Thus
∣bak∣ω≥r
in this case. On the other hand
∣bak∣ω≤∣a0ak∣ω≤2kr
by the triangle inequality.
We let
qk=((ak,bk),(ok,ω))
be the respective harmonic pair. By Lemma 8.8, we have
δ(q,qk)≤C(r,∣bak∣ω),
where
[TABLE]
Thus
δ(q,q4)≤4ln(3+8)≤4ln10.
By Lemma 8.1,
δ(qk,q0)≤2∣k∣ln4
for every
k∈Z.
Therefore,
δ(q,q0)≤4ln10+8ln4=4ln160=D.
∎
Lemma 8.10**.**
The set
V=V(ω,σ)
is cobouded in
Harmω
uniformly in
ω∈X
with respect to the metric
δ,
that is,
δ(p,V)≤D
for every
p∈Harmω,
where
D
depends only on
σ.
Proof.
Given
p∈Harmω, p=((a,b),(o,ω)), (a,b)=Sr(o),
there is
k∈Z
such that
σk+1<r≤σk.
We take
q∈Harmω, q=((a′,b′),(o,ω))
with
(a′,b′)=Sσk(o).
Then
p, q
lie on the line
h(o,ω)
and hence
δ(p,q)≤∣pq∣=lnrσk≤lnσ1
(in fact
δ(p,q)=∣pq∣
by Theorem 4.1). By Lemma 8.9,
there is
q′∈Vk
such that
δ(q,q′)≤D1
with
D1=4ln160.
Thus
δ(p,V)≤δ(p,q′)≤lnσ1+D1=:D.
∎
Recall that by Lemma 7.2 any two vertices
p, p′∈V
are connected by a geodesic
γ
in
Z
which consists of at most two radial subsegments
γ′, γ′′⊂γ
and at most one horizontal edge
h=qq′
between them, possibly degenerated,
q=q′,
which lies on the lowest level of
γ, γ=γ′∪h∪γ′′.
We assume that
∣γ′′∣≤∣γ′∣
and consider two cases, the first is Lemma 8.11,
the second one is Lemma 8.12.
Lemma 8.11**.**
Given vertices
p, p′∈V,
assume that
∣γ′′∣≤1
for a geodesic
γ=γ′∪h∪γ′′
between
p, p′.
Then
δ(p,p′)≥C∣pp′∣Z−D
for
C=lnσ1
and a constant
D≥0
depending only on
σ.
Proof.
By our assumption,
∣γ′∣≥∣γ∣−2=∣pp′∣Z−2,
and
γ′⊂Z
is a radial geodesic between harmonic
p
and
q
in
Xω, ∣γ′∣=∣pq∣Z,
where
p=((a,b),(o,ω))=Sr(o), q=((c,d),(o′,ω))=Sr′(o′),
r=σl, r′=σk.
For the levels
l=ℓ(p)
and
k=ℓ(q)
we have
l>k
and
∣γ′∣=l−k.
The part
e∪γ′′
of
γ
consist of at most two edges between
q
and
p′,
one horizontal and one radial, thus
δ(q,p′)≤D1
by Corollaries 8.2, 8.6,
with
D1≤2/σ+2ln12.
We take the sphere
Sr′(o)=(a′,b′)⊂Xω,
and consider the harmonic
p=((a′,b′),(o,ω))∈Harmω.
Then by the triange inequality we have
∣δ(p,q)−δ(p,p)∣≤δ(q,p).
Since
q
is a vertex of the hyperbolic approximation
Z,
the sphere
Sr′(o)
is a member of a harmonic chain. Since
pq⊂Z
is a radial geodesic segment,
ab⊂cd⊂Xω.
By the choice of
Sr′(o),
we have
ab⊂a′b′,
whence
cd∩a′b′=∅.
Thus we can apply Lemma 8.9 to
q, p,
and obtain
δ(q,p)≤D2=4ln160.
Therefore,
δ(p,q)≥δ(p,p)−D2.
On the other hand,
p, p
lie on a line in
Harmω,
thus
∣pp∣=ln(r′/r)=(l−k)lnσ1
because
r′/r=1/σl−k.
By Theorem 4.1,
δ(p,p)=∣pp∣.
Furthermore,
∣pq∣Z=l−k
because
pq⊂Z
is a radial geodesic segment. Therefore,
δ(p,q)≥C∣pq∣Z−D2
with
C=ln(1/σ).
Finally,
δ(p,p′)≥δ(p,q)−δ(q,p′)≥C∣pq∣Z−(D1+D2)≥C(∣pp′∣Z−2)−(D1+D2)=C∣pp′∣Z−D
with
D=2C+D1+D2.
∎
Lemma 8.12**.**
Given vertices
p, p′∈V,
assume that
∣γ′′∣≥2
for a geodesic
γ=γ′∪h∪γ′′
between
p, p′.
Then
δ(p,p′)≥C∣pp′∣Z−D
with
C=21lnσ1
and
D
depending only on
σ.
Proof.
As in Lemma 8.11,
γ′⊂Z
is a radial geodesic between harmonic
p
and
q
in
Xω, ∣γ′∣=∣pq∣Z=l−k,
where
ℓ(p)=l, ℓ(q)=k.
Then
k
is the level of
qq′, k=ℓ(q)=ℓ(q′)
and
γ′′⊂Z
is a radial geodesic between harmonic
q′
and
p′
in
Xω, ∣γ′′∣=∣p′q′∣Z=l′−k
where
ℓ(p′)=l′.
By our assumption
∣γ′∣≥∣γ′′∣.
Thus
l≥l′,
and
∣pp′∣Z=∣γ∣≤∣γ′∣+∣γ′′∣+1≤2∣γ′∣+1=2(l−k)+1.
Let
S
be a zz-path in
Harm
between
p, p′
that approximates the distance
δ(p,p′), δ(p,p′)≥∣S∣−ε
for some
ε>0.
We fix an involution
ρ:X→X
associated with
p′=((a′,b′),t′), t′=(o′,ω),
see sect. 5.2, and let
e=eρ
be the respective elliptic quasi-line. By Lemma 5.2,
there is a unique
s∈e
such that the pair
q=(s,t)
is harmonic, where
p=((a,b),t), t=(o,ω).
Again, by Lemma 5.2, there is a unique
t′′∈e
such that the pair
(s,t′′)
is harmonic. Thus
q′′=(s,t′′)∈e
as well as
p′∈e
by definition of
e.
By Proposition 6.2,
δ(p′,q′′)≤D0
for some universal constant
D0<16.
Hence, there is a zz-path
S′
between
p′
and
q′′
with
∣S′∣≤D0+ε.
Note that
t, t′′
lie on the line
hs.
Let
S′′
be a zz-path between
q′′=(s,t′′)
and
q=(s,t)
which consists of one side,
S′′⊂hs.
Then
p=((a,b),t)
and
q
lie on the line
ht.
Thus the concatenation
S:=S∗S′∗S′′∗qp
is a closed zz-path in
Harm.
We apply [Bu18, Proposition 6.1] to conclude
∣S∣+∣S′∣+∣S′′∣>∣pq∣.
The projection
prt:S∗S′∗S′′→ht
does not increase distances, see [Bu18, Lemma 5.5 and Proposition 6.1],
and
∣prt(S′′)∣=0
because
prt(S′′)=q.
Therefore,
∣S∣≥∣pq∣−(D0+ε).
By definition of
e,
we have
t′=(o′,ω)∈e.
We denote
s=(z,u).
Since
(s,t)
is harmonic, we have
∣zo∣ω=∣ou∣ω.
We assume without loss of generality that
o<o′, z<o<u
with respect to our fixed order on
Xω.
Since
s, t′∈e,
the pairs
s=(z,u)
and
t′=(o′,ω)
separate each other, see Lemma 5.1. Hence,
∣ou∣ω>∣oo′∣ω.
We denote by
pn
the vertex of
γ′
on the level
n, ℓ(pn)=n, k≤n≤l,
and similarly by
pn′
the vertex of
γ′′
on the level
n, ℓ(pn′)=n, k≤n≤l′.
Denote by
αn
the curve in
Z
between
pn
and
pn′
consisting horizontal edges. By the assumption
∣γ′′∣≥2,
thus there is a vertex
pn′∈γ′′
with
n=k+2.
Note that
∣αk+2∣≥4
because otherwise we can shorten the geodesic
γ
between
p
and
p′.
Therefore, there is an edge
vv′⊂αk+2
with vertices
v, v′
different from the ends
pk+2, pk+2′
of
αk+2.
Thus the intersection
Bv∩Bv′
misses the balls
Bpk+2
and
Bpk+2′
by properties of harmonic chains. Here
Bv⊂Xω
is the ball corresponding to the vertex
v∈V.
Since
γ′, γ′′⊂Z
are radial geodesics, we have
Bp⊂Bpk+2, Bp′⊂Bpk+2′
for respective balls in
Xω.
Recall that
o
is the center of
Bp,
and
o′
the center of
Bp′.
It follows that the intersection
Bv∩Bv′
is a segment on
Xω
lying inside of the segment
oo′⊂Xω.
By inequality (5),
∣Bv∩Vv′∣≥r/4
for
r=σk+2,
and we obtain
∣oo′∣ω≥σk+2/4.
Thus
[TABLE]
Since
∣γ∣≤2(l−k)+1,
we have
∣pq∣≥∣γ∣/2⋅lnσ1−D1
with
D1=25lnσ1.
Therefore,
[TABLE]
where
C=21lnσ1.
Finally, we conclude
δ(p,p′)≥C∣pp′∣Z−D,
where
D=D0+D1.
∎
Proposition 8.13**.**
The inclusion
f:V↪Harmω
is a quasi-isometry with respect to the metric on
Z
and
δ-metric
on
Harmω.
Proof.
By Corollary 8.7 we have
δ(v,v′)≤C∣vv′∣Z
for every pair vertices
v, v′∈V,
where the constant
C
depends only on
σ.
By Lemmas 8.11 and 8.12
we have
δ(v,v′)≥C∣vv′∣Z−D
for every pair vertices
v, v′∈V,
where the constants
C, D
depend only on
σ.
Thus the map
f
is quasi-isometric. By Lemma 8.10, the set
V
is cobouded in
Harmω.
Thus
f
is quasi-isometry.
∎
Proposition 8.14**.**
Assume that a Möbius structure
M
on
X=S1
is strictly monotone, i.e., it satisfies axioms (T), (M(α)), (P),
and satisfies Increment axiom. Then
(Harm,δ)
is a complete, proper, hyperbolic geodesic metric space with
δ-metric
topology coinciding with that induced from
X4.
Proof.
By Theorem 4.1,
(Harm,δ)
is a complete, proper, geodesic metric space with
δ-metric
topology coinciding with that induced from
X4.
By Corollary 6.11, any its subset
Harmω, ω∈X,
is quasi-isometric
(Harm,δ).
Using Proposition 8.13, we see that
(Harmω,δ)
is quasi-isometric to its hyperbolic approximation
Z=Z(ω,σ). Thus
(Harm,δ)
is quasi-isometric to
Z.
By Proposition 7.8,
Z
is hyperbolic. Since both spaces
(Harm,δ)
and
Z
are geodesic, the space
(Harm,δ)
is hyperbolic.
∎
We define
Y=(Harm,δ).
By Proposition 8.14,
Y
is a complete, proper, hyperbolic geodesic metric space. We clearly
have
∂∞Harmω=Xω
for every
ω∈X.
Since
Harmω
is cobouded in
Y,
we have
∂∞Y=Harmω∪{ω}=X=S1.
The fact that the induced Möbius structure
MY
on
X
is isomorphic to
M
is tautological because all of the geometry of
Y
including
Y
itself is determined via
M.
In particular, given two points
x, x′∈∂∞Y,
we take
ω∈∂∞Y
different from
x, x′.
Then
x, x′∈Xω,
and we consider the line
h=h(x,ω)⊂Harmω⊂Y.
Furthermore, we fix
y∈Xω, y=x,
and observe that there are points
p, q∈h
such that
x∈p, y∈q.
Then
∣xx′∣ω=βe±∣pq∣
for some fixed constant
β(=∣xy∣ω).
In other words, the metric of
Xω
is recovered from the geometry of
Y.
∎
Bibliography8
The reference list from the paper itself. Each links out to its DOI / PubMed record.
1[Bo S] M. Bonk and O. Schramm, Embeddings of Gromov hyperbolic spaces, Geom. Funct. Anal. 10 (2000), no.2, 266–306.
2[BS 07] S. Buyalo, V. Schroeder, Elements of asymptotic geometry, EMS Monographs in Mathematics, 2007, 209 pages.
3[Bu 17] S. Buyalo, Möbius structures and timed causal spaces on the circle, Algebra i analys, 29 (2017), n.5, ar Xiv:math. MG/1705.00478
4[Bu 18] S. Buyalo, On the inverse problem of Möbius geometry on the circle, ar Xiv:math. MG/1810.03133.
5[Bu 19] S. Buyalo SRA-free condition by Zolotov for self-contracted curves and nondegeneracy of zz-distance for Möbius structures on the circle, ar Xiv:math. MG/1906.09966
6[FS 12] T. Foertsch, V. Schroeder, Ptolemy circles and Ptolemy segments. Arch. Math. (Basel) 98 (2012), no. 6, 571–581.
7[FS 13] T. Foertsch, V. Schroeder, Metric Möbius geometry and a characterization of spheres, Manuscripta Math. 140 (2013), no. 3-4, 613–620.
8[Zo 18] V. Zolotov, Subsets with small angles in self-contraced curves, ar Xiv:1804.00234 [math.MG], 2018.