A metric sphere not a quasisphere but for which every weak tangent is Euclidean

# A metric sphere not a quasisphere but for which every weak tangent is Euclidean

Angela WU Angela Wu
Department of Mathematics
University of California
Los Angeles
CA90095
USA
June 7, 2018
###### Abstract.

We show that for all , there exists a doubling linearly locally contractible metric space that is topologically a -sphere such that every weak tangent is isometric to but is not quasisymmetrically equivalent to the standard -sphere. The same example shows that -Ahlfors regularity in Theorem 1.1 of [BK02] on quasisymmetric uniformization of metric -spheres is optimal.

The author was partially supported by NSF grants DMS-1506099 and DMS-1266164

## 1. Introduction

Let and be two metric spaces. A homeomorphism is a quasisymmetry if there exists a homeomorphism such that for all , with , we have

 dY(f(x),f(y))dY(f(x),f(z))≤η(dX(x,y)dX(x,z)).

In this case we say that and are quasisymmetrically equivalent.

In [Kin17] Kinneberg characterized metric circles that are quasisymmetric to the standard circle in terms of weak tangents (defined in Section 4 below):

###### Theorem 1.1 (Kinneberg, [Kin17]).

A doubling metric circle is quasisymmetrically equivalent to the standard circle if and only if every weak tangent of quasisymmetrically equivalent to the real line based at .

Here a metric space is said to be doubling if for every , every ball of radius can be covered by balls of radius , where is a positive integer that does not depend on the choice of the ball . In this paper we prove that Kinneberg’s result cannot be extended to higher dimensions:

###### Theorem 1.2.

For every , there exists a doubling, linearly locally contractible metric space that is topologically an -sphere such that every weak tangent is isometric to but is not quasisymmetrically equivalent to the standard -sphere.

When , one can compare our result with the following Theorem:

###### Theorem 1.3 (Bonk and Kleiner, [Bk02]).

Let be an 2-Ahlfors regular metric space homeomorphic to . Then is quasisymmetric to if and only if is linearly locally contractible.

Our Theorem 1.2 shows that the conclusion of Theorem 1.3 is false if we replace -Ahlfors regularity with -Ahlfors regularity for .

Our study is also related to the following theorem, proven in [Lin17]:

###### Theorem 1.4.

Let be a doubling metric space homeomorphic to . The following are equivalent:

1. is quasisymmetrically equivalent to the standard -sphere.

2. For every , there exists an open neighborhood of in such that is quasisymmetrically equivalent to .

An alternative proof of Theorem 1.4 can be given using ideas in [GW18]. Roughly speaking, Theorem 1.4 says local geometry properties promote to global property. Since weak tangents are local, one could ask the following question:

###### Question 1.5.

Suppose is doubling and linearly locally connected. Are the following two statements equivalent?

1. is quasisymmetrically equivalent to the standard -sphere

2. Every weak tangent of is quasisymmetrically equivalent to .

When is a doubling and linearly locally connected metric sphere, statement (i) implies statement (ii). However, our construction shows that statement (ii) does not imply statement (i).

###### Acknowledgement.

The author would like to thank Mario Bonk and John Garnett for their support and many conversations. The author thanks Jeff Lindquist for helpful discussions and Wenbo Li for his comments and corrections.

## 2. The Modification of the Metric on R

We first construct a metric on so that every weak tangent of is isometric to , where is the Euclidean metric on , and so that there are segments of so that . Here, and in the remaining of the paper, denotes the length of the curve with respect to the metric . Once the metric is constructed, we show that for all , the weak tangents of the product space are isometric to the Euclidean space , but because , the product space cannot be quasisymmetric to the Euclidean space .

To construct , we glue together segments that look like snowflake curves of Hausdorff dimension i.e. intervals equipped with the metric . Such snowflake curves will have infinite length, so we modify the metric on the segment to get a metric so that looks like the Euclidean metric when two points are close together, but it looks like when the two points are far apart.

Note that our construction of only modifies the geometry of a bounded subset of , and therefore we can embed our -dimensional construction into a topological -sphere.

To begin the construction let and and define

 φα,c(x)={L(α,c)x,x∈[0,c](x−c(1−α)1−c(1−α))α,x∈[c,1],

where

 L(α,c)=1c(cα1−c(1−α))α.

The function is the only function that has the following properties:

1. ,

2. ,

3. is linear on ,

4. There exists , with , such that when ,

5. is continous and differentiable at .

In addition to the above properties that uniquely define , we observe that has the following extra properties:

1. is a homeomorphism,

2. is concave on .

###### Lemma 2.1.

For any , , and , we have

 φα,c(ab)φα,c(a)φα,c(b).
###### Proof.

The function is decreasing in to , therefore if , then

 φα,c1≥φα,c2.

Let

 φ∗(x)=φα,c(ax)φα,c(a).

The function meets condition (1)-(5) listed above for some , therefore . This implies that for all . Taking , we have

 φα,c(ab)≤φα,c(a)φα,c(b).

###### Lemma 2.2.

For any fixed , for any , and whenever , we have

 0≤φα,c(t)+φα,c(x−t)−φα,c(x)≤(2−2α)φα,c(x).
###### Proof.

When , for all , we have . When is fixed, is maximized when

 φ′α,c(t)−φ′α,c(x−t)=0,

which is possible only if i.e. . Then we have

 φα,c(t)+φα,c(x−t)−φα,c(x) ≤2φα,c(x/2)−φα,c(x) ≤2φα,c(x/2)−2αφα,c(x/2) =(2−2α)φα,c(x/2).

When , we have

 φα,c(t)+φα,c(x−t)≤L(α,c)x=xcφα,c(c).

By the concavity of , we have

 φα,c(x)x≥φα,c(2c)2c.

Therefore

 φα,c(t)+φα,c(x−t)−φα,c(x) ≤xcφα,c(c)−xφα,c(2c)2c=x2c(2φα,c(c)−φα,c(2c)).

But

 2φα,c(c)−φα,c(2c)≤(2−2α)φα,c(c)=(2−2α)cL(α,c).

Therefore

 φα,c(t)+φα,c(x−t)−φα,c(x)≤x2c(2−2α)cL(α,c)=(2−2α)φα,c(x/2).

In any case, we have

 0≤φα,c(t)+φα,c(x−t)−φα,c(x)≤(2−2α)φα,c(x/2)≤(2−2α)φα,c(x).

## 3. The One Dimensional Construction

For any , we have

 (1) limc→0+L(α,c)=+∞.

Let be an increasing sequence in such that . By (1), we can choose such that , and

 limn→∞L(αn,cn)=+∞.

Let . Choose a sequence so that for all , , such that is decreasing and .

For all , let . Let , and equip with the metric , where is the usual Euclidean metric on . Note that

1. The distance between the two endpoints of is .

2. is rectifiable and the length of is .

We construct a metric on so that if ,

1. when restricted to ,

2. when ,

3. when and ,

4. when and , and

5. when , .

## 4. The Weak Tangents of (R,δ)

A pointed metric space is a a triplet , where is a metric space and is a point in . Let . A map between two pointed metric spaces and is a -rough isometry if

1. ,

2. , and

3. for all ,

 |dY(ψ(x1),ψ(x2))−dX(x1,x2)|≤ε.

Note that a -rough isometry may not be continuous. The pointed Gromov-Hausdorff distance between 2 pointed metric spaces, denoted , is defined as the infimum of for which we can find a -rough isometry .

A pointed metric space is called a weak tangent of another metric space if there exist points and positive integers such that converges to in pointed Gromov-Hausdorff sense, i.e. for all , and for all there exists such that for all ,

 dGH(¯¯¯¯BλndX(xn,R+ε),¯¯¯¯BdT(p,R))≤ε.

In particular, we get if for all , and for all there exists such that for all ,

 dGH(¯¯¯¯BλndX(xn,R),¯¯¯¯BdT(p,R))≤ε.

Our notion of pointed Gromov-Hausdorff convergence is adopted from [HKST15, Definition 11.3.1]. Also see [BBI01] for detailed discussion on Gromov-Hausdorff distance and Gromov-Hausdorff convergence.

Let M be a set of separable, uniformly doubling, and uniformly linearly locally contractible pointed metric spaces. The pointed Gromov Hausdorff convergence on the set induces a topology on that is metrizable[Don11]. In particular, if a sequence of pointed metric space converges to a Gromov-Hausdorff limit, then the limit is unique.

The goal of this section is to prove the following proposition that describe all the weak tangents of .

###### Proposition 4.1.

For all , and for all positive integers , converges in pointed Gromov-Hausdorff sense to .

Note that Proposition 4.1 guarentees the existence of weak tangents of . To prove the above proposition we will use the following three lemmas:

###### Lemma 4.2.

Suppose are three points so that . Suppose . Then

 0≤δ(x,y)+δ(y,z)−δ(z,x)≤(2−2αN)min{sN,δ(x,y)}
###### Proof.

If , then . Otherwise, let

 a =sup{an≤y:n∈N}∨sup{bn≤y:n∈N}∨x b =inf{an≥y:n∈N}∧inf{bn≥y:n∈N}∧z

We have . By definition of , we have

 δ(x,y)+δ(y,z)−δ(x,z) =(δ(x,a)+δ(a,y))+(δ(y,b)+δ(b,z)) −(δ(x,a)+δ(a,b)+δ(b,z)) =δ(a,y)+δ(y,b)−δ(a,b).

Bu our choice of and , either for some , or . In the latter case, we have

 δ(x,y)+δ(y,z)−δ(z,x)=δ(a,y)+δ(y,b)−δ(a,b)=0.

In the former case, Lemma 2.2 gives

 0 ≤δ(x,y)+δ(y,z)−δ(z,x)=δ(a,y)+δ(y,b)−δ(a,b) ≤(2−2αn)δ(a,b)=(2−2αN)min{sN,δ(x,y)}.

Since , we have our desired conclusion. ∎

###### Lemma 4.3.

Let and be arbitrary. Suppose . Let and . The map

 ψ:([a,b],p,δ)→([−r,r],0,d) ψ(x)={−δ(p,x),x≤pδ(p,x),x>p.

is a -rough isometry, where .

###### Proof.

Since is surjective and fixes , it remains to check that for all , we have

 (d(ψ(x),ψ(y))−δ(x,y))≤ε.

Suppose . If , then

 |d(ψ(x),ψ(y))−δ(x,y)| =||δ(p,x)−δ(p,y)|−δ(x,y)| =|δ(p,y)−δ(p,x)−δ(x,y)| ≤(2−2αN)min{sN,δ(p,x)} =(2−2αN)min{sN,2r}.

The second to last inequality is a consequence of Lemma 4.2. If , then

 |d(ψ(x),ψ(y))−δ(x,y)| =||δ(p,x)−δ(p,y)|−δ(x,y)| =|δ(p,y)+δ(p,x)−δ(x,y)| ≤(2−2αN)min{sN,δ(x,y)} =(2−2αN)min{sN,2r}.

If , then following a similar arguement as when , we get

 |d(ψ(x),ψ(y))−δ(x,y)|≤(2−2αN)min{sN,2r}

This verifies that is a -rough isometry. ∎

###### Lemma 4.4.

For , we have

 supp∈RdGH((¯¯¯¯Bδ(p,r),p,δ),(¯¯¯¯Bd(0,r),0,d))=o(r)

as .

###### Proof.

Let and be arbitrary. Let . If , then on . We have

 dGH((¯¯¯¯Bδ(p,r),p,δ),(¯¯¯¯Bd(0,r),0,d))=0.

If , we consider 2 cases:

Case 1: . As , , therefore . In this case Lemma 4.3 implies

 dGH((¯¯¯¯Bδ(p,r),p,δ),(¯¯¯¯Bd(0,r),0,d))≤2(2−2αN)r=o(r).

Case 2: . In this case, is a length metric on , and is an isometry between two length spaces. We have

 dGH((¯¯¯¯Bδ(p,r),p,δ),(¯¯¯¯Bd(0,r),0,d))=0.

###### Proof of Proposition 4.1.

Let be a sequence in and be a sequence of positive numbers that diverges to . By Lemma 4.4,

 dGH((¯¯¯¯Bλnδ(an,R),an,λnδ),(¯¯¯¯Bλnd(0,R),0,λnd)) =λndGH((¯¯¯¯Bδ(an,λ−1nR),an,δ),(¯¯¯¯Bd(0,λ−1nR),0,d))=λno(λ−1nR).

But by symmetry of . As , . We have

 (¯¯¯¯Bλnδ(an,R),an,δ)→¯¯¯¯Bd(0,R),0,d).

This is true for all . We conclude that . ∎

## 5. Linear Local Contractibility and Assouad Dimension of (R,δ)

In this section we establish two properties of the space . These properties often appear in the study of quasisymmetry classes of metric spheres. Both properties are discussed in detail in [Hei01].

###### Definition 5.1.

Let be a constant. A metric space is -linearly locally contractible if every small ball is contractible inside a ball whose radius is times larger. A metric space is linearly locally contractible if it is -linearly locally contractible for some .

###### Definition 5.2.

Let . A metric space is -doubling if for all , every open ball of can be covered by balls of radius . A metric space is doubling if it is -doubling for some .

Both doubling and linear local contractibility are preserved under quasisymmetry. The Euclidean spaces like or are doubling and linearly locally contractible. The doubling property also ensures the existence of weak tangents.

###### Proposition 5.3.

The space is -linearly locally contractible.

###### Proof.

Any open ball in is an open interval . Denote the center of (in .) Note that the map is increasing on , and decreasing on . Therefore the map is a homotopy of to in . This proves that is -linearly locally contractible. ∎

###### Proposition 5.4.

The space is doubling.

If a metric space is doubling, then there exists and such that for all and , any set of diameter in can be covered by at most subsets of diameter at most . The function is called the covering function of . The Assouad dimension of is defined to be the infimum of all so that a covering function of the form of exists. Conversely, any metric space of finite Assouad dimension is doubling. Proposition 5.4 will follow from the stronger proposition below.

###### Lemma 5.5.

For each , the function is a covering function of .

###### Proof.

Every subinterval of of -diameter has -diameter . Thus our goal is to show that for every , and every , every subset of of -diameter can be covered by no more than -many subintervals of of -diameter at most . The number of subintervals we need can be bounded from above by

 snφ−1n(s−1nr)snφ−1n(s−1nεr)+1≤supy∈(0,1]φ−1n(y)φ−1n(εy)+1=supy∈[φn(cn),1]φ−1n(y)φ−1n(εy)+1.

We claim that the last supremum is attained when . This is equivalent to

 (2) φn(φ−1n(εy))φn(φ−1n(y))≤φn(φ−1n(εy)φ−1n(y))

Inequality (2) is true by Lemma 2.1.

Suppose . When , we have , and

 εα−1n=x0−c(1−αn)1−c(1−αn)≤x0.

When , we have and . Since , we have

 1x0=φ(c)cε=ε−α−1n(εφ(c))α−1n−1φ(c)α−1nc≤ε−α−1n.

In any case, we can take the covering function of to be

 ε−α−1n+1≤2ε−α−1n.

###### Proposition 5.6.

The Assouad dimension of is .

###### Proof.

Let be arbitrary. There exists such that when , . Let . Then the function , where , is a covering function of for all . Thus is a covering function of .

Since has Assouad dimension , the proposition follows. ∎

## 6. Higher Dimension Construction

Let . We will denote by the Euclidean metric on . Let

 Xd=(R×Rd−1,√δ2+d2Euclid)

be the product of and . Write . Here are some facts about .

###### Proposition 6.1.
1. Every weak tangent of is isometric to .

2. is doubling and linearly locally contractible.

###### Proof.
1. Every weak tangent of is of the form , where is a weak tangent of .

2. Recall that is doubling (Proposition 5.4). is product of doubling metric space, hence doubling. By Proposition 5.3, is -linearly localy contractible for some . Let be any point in , and be arbitrary. The ball in can first be contracted to within a , which can then be contracted to the point .

Every finite segment in is rectifiable. Let be a the measure on given by length. For , let be the product measure on , where is the -dimensional Lebesgue measure on .

In the remaining of this section we show that is not quasisymmetrically equivalent to . To do that we consider a geometric quantity that is roughly preserved under quasisymmetry called modulus. Given a family of curves in a measured metric space , we say that a function is admissible if for all ,

 ∫γρ(x)ds≥1.

Let . We define the -modulus of as

Let be two disjoint nondegerate continua in . Let to be the collection of all rectifiable curves joining and . We write .

Moduli behave nicely under quasisymmetry, as illustrated by the following theorem.

###### Theorem 6.2 (Tyson, [Tys98]).

Let be locally compact, connected, -Ahlfors regular metric spaces, where , and be a quasisymmetric homeomorphism. Then there exists such that for all curve family , we have

 1CmodQ(Γ)≤modQ(f(Γ))≤CmodQ(Γ).

See the next section for the definition of -Ahlfors regularity.

We can now prove that is not quasisymmetrically equivalent to . The idea is that if the two spaces are quasisymmetrically equivalent, then for any curve family , and should have comparable moduli. We know that has the property that any disjoint nondegerate continua in satisfy

 (3) modd(E,F)≤ϕ(d(E,F)diam(E)∧diam(F)).

for some non-increasing function . However, Proposition 6.3 shows that some sequence disjoint nondegerate continua in do not behaviour as in (3). The only problem is that is not -Ahlfors regular, so we cannot apply Theorem 6.2 directly. In Proposition 6.4, however, we will show that the inequality

 modd(En,Fn)≤Cmodd(f(En),f(Fn))

holds for some independent of .

From now on, we will denote .

###### Proposition 6.3.

There exists such that and .

###### Proof.

Take . Then , and , therefore

 Δ(En,Fn)=d(En,Fn)diamEn∧diamFn=1.

For each , and for and , let be the path

 t↦(x,t,v2,…,vd−1),t∈[0,sn].

Let

 Γn={γx,v2,…,vd−1:x∈In,(v2,…,vd−1