RIGIDITY OF QUASICONFORMAL MAPS ON CARNOT GROUPS
Mark Medwid
A Dissertation
Submitted to the Graduate College of Bowling Green State University in partial fulfillment of the requirements for the degree of
DOCTOR OF PHILOSOPHY
August 2017
Committee:
Xiangdong Xie, Advisor
Alexander Tarnovsky, Graduate Faculty Representative
Mihai Staic
Juan Bes´ Copyright c August 2017 Mark Medwid All rights reserved iii ABSTRACT
Xiangdong Xie, Advisor
Quasiconformal mappings were first utilized by Grotzsch¨ in the 1920’s and then later named by Ahlfors in the 1930’s. The conformal mappings one studies in complex analysis are locally angle-preserving: they map infinitesimal balls to infinitesimal balls. Quasiconformal mappings, on the other hand, map infinitesimal balls to infinitesimal ellipsoids of a uniformly bounded ec- centricity. The theory of quasiconformal mappings is well-developed and studied. For example, quasiconformal mappings on Euclidean space are almost-everywhere differentiable. A result due to Pansu in 1989 illustrated that quasiconformal mappings on Carnot groups are almost-everywhere (Pansu) differentiable, as well. It is easy to show that a biLipschitz map is quasiconformal but the converse does not hold, in general. There are many instances, however, where globally defined quasiconformal mappings on Carnot groups are biLipschitz. In this paper we show that, under cer- tain conditions, a quasiconformal mapping defined on an open subset of a Carnot group is locally biLipschitz. This result is motivated by rigidity results in geometry (for example, the theorem by Mostow in 1968). Along the way we develop background material on geometric group theory and show its connection to quasiconformal mappings. iv ACKNOWLEDGMENTS
I would like to acknowledge, first and foremost, my wife, Heather. She’s put up with the long nights and anxiety that accompany graduate school; she constantly cheered me on, told me to be more confident, and agreed to put her life on hold for a number of years while I finished my degree. I’m not sure my finishing graduate school would be possible without her support. Next, I would like to thank my advisor, Xiangdong Xie, who exposed me to a new field of mathematics. He very patiently put up with my na¨ıve questions, ignorance and incorrect proofs; he slowly helped me to become a better mathematician. Beyond academics, he gave me lots of helpful career and life advice that I’ll be sure to apply in the future. He is also one of the first people to encourage me to graduate early – this lead to me accepting a new tenure-track job. I would also like to acknowledge and thank the other members of my dissertation commit- tee (Mihai Staic, Juan Bes´ and Alexander Tarnovsky) for their thoughtful commentary on my manuscript and insightful questions during my defense. Furthermore, I would like to acknowledge my fellow colleagues (both former and present) for their many helpful discussions on life, graduate school and mathematics. I formed many valuable friendships during graduate school. There are too many to be included here, but I will attempt to do so: Jake Laubacher, David Walmsley, Todd Romutis, John Haman and Sam Carolus shared an office with me and kept my day-to-day life more interesting. This was a room filled with exceedingly promising young mathematicians and I was honored to be a part of it. I’d also like to thank Robert Kelvey, Jeff Norton and Leo Pinheiro (who are all now on to greener pastures) for their many helpful conversations with me. My family, in particular, my parents, deserve special mention for all their support over the ten consecutive years I’ve been in school since graduating from high school in 2007. They’ve always had my back, no matter what. Finally, I feel that the Math & Stats Department at BGSU deserves special mention. The department supported my education expenses and my livelihood, and there are so many stellar faculty and staff members that helped form me both as a mathematician and as a teacher. v
TABLE OF CONTENTS Page
CHAPTER 1 INTRODUCTION ...... 1 1.1 The Word Metric ...... 2 1.2 Quasi-Isometries ...... 3 1.3 Quasi-Isometric Rigidity ...... 9
CHAPTER 2 PRELIMINARIES ...... 11 2.1 BiLipschitz Mappings ...... 11 2.2 Quasiconformality ...... 12 2.3 Quasisymmetric Mappings ...... 17
CHAPTER 3 QUASI-ISOMETRIES ON HYPERBOLIC SPACE ...... 24 3.1 The Morse Lemma ...... 24 3.2 The Gromov Boundary ...... 35
CHAPTER 4 CARNOT GROUPS ...... 42 4.1 Lie Algebras and Lie Groups ...... 42 4.2 The Exponential Map & Carnot-Caratheodory´ Distance ...... 46 4.3 Pansu’s Differentiation Theorem ...... 51
CHAPTER 5 MAIN RESULTS ...... 55 5.1 Rigidity of Quasisymmetric Mappings on Fibered Metric Spaces ...... 55 5.2 Rigidity of Quasiconformal Maps on Two-Step Carnot Groups ...... 66
BIBLIOGRAPHY ...... 76 vi
LIST OF FIGURES Figure Page
2.1 An artist’s rendering of what a quasiconformal map might look like...... 13
3.1 The Morse Lemma...... 25 3.2 Two cases for triangles in trees...... 31
3.3 Asymptotic rays in the upper-half plane model of H2...... 36 1
CHAPTER 1 INTRODUCTION
The purpose of this dissertation is to develop the theory accompanying the main result of the author’s research efforts while studying at Bowling Green State University, as well as the proof of the result itself. The author assumes knowledge of basic group theory, analysis, and point-set topology; anything beyond this will be developed in the pages contained hereafter. The following is the author’s main result:
Theorem. Let G = V1 V2 be a two-step Carnot group and let W1 V1 be a subspace and set ✓ W := [W ,W ]. Let W = W W ; W is a Lie sub-algebra of G and so corresponds to a Lie 2 1 1 1 2 subgroup. Suppose that E G is an open set. If f : E G is a quasiconformal mapping that ✓ ! permutes the left cosets of W , then f is locally biLipschitz.
Many of the terms contained in the above theorem are ostensibly foreign to the reader. The basic idea is the following: Carnot groups are spaces that are “nice” enough to carry some notion of differentiability. The definition is fairly similar in flavor to the familiar one from calculus. One then becomes interested in a certain type of (almost everywhere) differentiable function f that we name “quasiconformal.” As in calculus, f needs only to be defined on an open subset for the purposes of the theorem. If f moves a particular subset in a nice enough way, then it turns out that f is locally biLipschitz – to wit, inside a smaller open set, f distorts distances by no more than some bounded factor. This is perhaps enough to remind one of the famed Mean Value Theorem of first year calculus:
Theorem. Let f :[a, b] R be continuous and differentiable on (a, b). Then there exists some ! number c in (a, b) so that
f(b) f(a) = f 0(c) (b a). | | ·
When f has a bounded derivative, then this implies f distorts the lengths of intervals by a bounded factor. The main result of this dissertation is, of course, different in many ways from the 2 Mean Value Theorem, requiring a different setting and assumptions, but the punchline to this quick corollary carries a similar spirit. With the above context, the main theorem is an analysis theorem. However, the study of such theorems also has some deep ties to the world of geometric group theory that will be fleshed out in the body of the dissertation. While quasiconformal analysis is used in several branches of math- ematics, it is through geometric group theory that the author was first exposed to the study of quasiconformal mappings. Details for theorems and definitions will be provided (when useful) so as to make this dissertation a stand-alone document for the readers; however, the introductory material is, for the most part, well studied. Many excellent papers and books exist for provid- ing copious details on these subjects – recommendations will be provided when relevant for the interested reader.
1.1 The Word Metric
Recall that a group is finitely generated if there exists a finite set S so that G = S . In what h i follows, unless specified, G denotes a finitely generated group. One can endow G with a metric space structure as follows.
Definition 1.1.1. (Kapovich, 2013) Suppose that G is a finitely generated group with generating set S = s1,s2,...,sd . The Cayley graph of G with respect to S is a graph (G, S) with vertex { } set G and edges are constructed according to the rule
there is an edge connecting g and h g = h s✏ for some 1 i d, ✏ 1, 1 . () · i 2{ }
Definition 1.1.2. (Kapovich, 2013) Let G be a finitely generated group with generating set S, and suppose (G, S) is the Cayley graph of G with respect to S. Endow with the shortest path metric d – the restriction of d to G, dS, is called the word metric on G with respect to S and is denoted
(G, dS).
One should note that, if e is the identity element of G, then for any g G, d (e, g) is 2 S the reduced word length of g with letters in S. Suppose h, g G; write both h and g as re- 2 3 ✏ ✏ ✏ ✏ duced words with letters in S, i.e. h = s 1h s kh ,g = s 1g s kg . Then the sequence i(1h) ··· i(kh) i(1g) ··· i(kg) ✏1g ✏1g ✏kg 1 (h, hs ,...,hs s ,hg) h hg i(1g) i(1g) i(kg 1) gives an edge-path joining and , so this implies that ··· d(e, g) d(h, hg). Conversely, suppose (h = g ,g ,...,g,g = hg) gives an edge-path joining 0 1 l l+1 1 1 h and hg – then the sequence (e, h g1,...,h gl,g) gives an edge-path joining e and g as gj and g only differ by multiplication on the right by some s", where s S, ✏ 1, 1 . This j+1 2 2{ } illustrates that d(h, hg) d(e, g) and so the two quantities are equal. This means that the word metric on G with respect to S is always left-invariant. With G as a metric space one may talk about the usual notions of limits and continuity of functions. However, one must note that the above definitions depend on the generating set S – different generating sets will produce different Cayley graphs, which in turn will induce different word metrics. One cannot hope for a well-defined “canonical” metric on G, so studying the small- scale metric geometry of G is a futile endeavor. Studying the coarse geometry of G, however, gives a partial solution to the well-definedness issue.
1.2 Quasi-Isometries
Recall that an isometry between metric spaces X and Y is a surjective distance-preserving map f : X Y . If S, T are two different generating sets of G, then there is no reason to suspect that !
(G, dS) and (G, dT ) are isometric. However, they will be “coarsely” isometric in the following sense.
Definition 1.2.1. (Kapovich, 2013) Let (X, d) and (Y,⇢) be two metric spaces. Let L 1 and A 0. A function f : X Y is an (L, A) quasi-isometry if the following hold: !
1. For any p, q X, 2
1 d(p, q) A ⇢(f(p),f(q)) L d(p, q)+A. L ·
2. For any y Y , d(f(X),y) A. 2
Two metric spaces are said to be quasi-isometric if there exists a quasi-isometry between them. 4 A way to summarize the conditions needed for a function to be a quasi-isometry is to say a quasi-isometry is both coarsely biLipschitz and quasi-surjective. A map satisfying the first condi- tion is also called an “(L, A) quasi-isometric embedding”. These conditions seem rather innocuous at first glance, but in reality a quasi-isometry is poorly behaved – quasi-isometries are not even con- tinuous in general.
Example 1.2.2.
(a) The natural inclusion map Z , R is a (1,1/2) quasi-isometry. !
(b) Let X be any bounded metric space. Then the one-point metric space is quasi-isometric to X. In particular, any finite group endowed with a word metric is quasi-isometric to the trivial group.
(c) The grid space (R Z) (Z R) is quasi-isometric to the plane R2. ⇥ [ ⇥
Though quasi-isometries are poorly behaved, they are at least strong enough to carry some structure.
Proposition 1.2.3. The relation given by X Y if and only if there exists a quasi-isometry ⇠ ⇠ from X to Y is an equivalence relation.
Proof. First observe that if X is a metric space, the identity map X X is an isometry. In ! particular, it is a (1, 0) quasi-isometry, so X X. Next, suppose that f : X Y is an (L ,A ) ⇠ ! 1 1 quasi-isometry and g : Y Z is an (L ,A ) quasi-isometry. Then, the composition g f : X Z ! 2 2 ! is also a quasi-isometry. Suppose p, q X. Then, 2
d (g(f(p)),g(f(q))) L d (f(p),f(q)) + A L L d (p, q)+L A + A Z 2 · Y 2 1 2 · X 1 2 1 and the other inequality is similar. Furthermore, if z Z, there exists y Y so that 2 2 d (g(y),z) A , and there exists x X so that d (f(x),y) A . Then, by the triangle Z 2 2 Y 1 5 inequality and the coarse biLipschitz condition,
d (g(f(x)),z) d (g(f(x),g(y)) + d (g(y),z) Z Z Z L d (f(x),y)+A + A 2 · Y 2 2 L A +2A 2 1 2
Hence, g f is a quasi-isometry and thus X Z. Finally, suppose f : X Y is a (L, A) ⇠ ! quasi-isometry. By quasi-surjectivity, we may construct a “quasi-inverse” to f. Let y Y ; by 2 quasi-surjectivity, there exists x X so that d (f(x ),y) A. Define g : Y X so that y 2 Y y ! g(y)=x . Now suppose x X. Since f(x) Y , we have: y 2 2
d (x, g(f(x))) = d (x, x ) L d (f(x),f(x )) + LA 2LA. X X f(x) · Y f(x)
This means that g is quasi-surjective. Now, if p, q Y , we have 2
d (g(p),g(q)) = d (x ,x ) L d (f(x ),f(x )) + LA X X p q · Y p q L(d (f(x ),p)+d (p, q)+d (q, f(x ))) + LA Y p Y Y q L d (p, q)+3LA. · Y
For the other inequality, we have
1 A d (g(p),g(q)) = d (x ,x ) d (f(x ),f(x )) . X X p q L Y p q L
Now d (f(x ),f(x )) d (f(x ),p) d (f(x ),p) and d (f(x ),p) d (p, q) d(f(x ),q) Y p q Y q Y p Y q Y q by the triangle inequality, so finally
1 A 1 3A d (g(p),g(q)) (d (p, q) d (f(x ),p) d (f(x ),q)) d (p, q) . X L Y Y p Y q L L Y L
Hence, g is a quasi-isometry and Y X. ⇠ 6 The above proposition implies that one can consider a metric space up to quasi-isometry. This is not a completely natural notion of equivalence (as, for example, Z, Q, R are all in the same quasi-isometry class). The upside is that considering quasi-isometry classes fixes the dilemma of a well-defined word metric, as it turns out that a (finitely generated) group G equipped with two different word metrics results in metric spaces quasi-isometric to one another. This fact is not entirely trivial. To prove it, we must adopt some basic notions about group actions. We let Sym X denote the set of all bijections X X and Isom X denote the set of all self-isometries of X. ! Definition 1.2.4. Let X be a metric space and G a group and suppose :G Sym X is a group action of G on X. If (G) Isom X, we say that G acts by ! ✓ isometries on X.
1. (Kapovich, 2013) The action of G on X by isometries is said to be properly discontinuous if for any C X compact, the set g G : g(C) C = is finite. ⇢ { 2 \ 6 ;}
2. (Bridson and Haefliger, 1999) The action of G on X by isometries is said to be cocompact if there exists some compact set C X so that for any x X, there exists some g G so ⇢ 2 x 2 that x g (C). 2 x We say that G acts geometrically on X if the action is both properly discontinuous and cocompact.
Definition 1.2.5. Let X be a metric space and p, q X. 2 1. If closed balls in X are compact, we say that X is proper.
2. A geodesic joining p and q is an isometric embedding ↵ :[0,d(p, q)] so that ↵(0) = p and ↵(d(p, q)) = q. If any two points may in X may be joined by a geodesic, we say that X is a
geodesic space.
With these definitions in hand, we may prove the following theorem.
Theorem 1.2.6. (Kapovich, 2013) (The Milnor-Schwarz Lemma) Let G be a group and X be a proper geodesic metric space. If G acts geometrically on X, then G is finitely generated and quasi-isometric to X when considered as a metric space with a word metric. 7 Proof. Let C be the compact subset of X satisfying the cocompactness condition. Let x C 0 2 and consider B := BR(x0), the closed ball of radius R about x0, with R large enough so that C B (x ). Since X is proper B is compact, and since the action of G is properly discontinuous, ⇢ R 0 the set S = s G : s(B) B = is finite; write S = s ,s ,...,s . Note here that the identity { 2 \ 6 ;} { 1 2 k} element of G is in S so S must be nonempty. We claim that S generates G. If G = S there is nothing to prove. Consider the quantity
r := inf d(g(B),B):g G S . { 2 \ }
We must have that r>0, since B is compact (in particular, closed) and g(B) is disjoint from B for g S. Now let g G. Connect x to g(x ) by a geodesic and let m be the smallest integer such 62 2 0 0 that d(x ,g(x )) mr + R. Choose points x ,x ,...,x = g(x ) on the geodesic joining x 0 0 1 2 m+1 0 0 and g(x ) so that x B and d(x ,x ) r for 1 j m. Now, for 1 j m, there exists 0 1 2 j j+1 g G so that x g (B), since C B; g is the identity element, as x B. For 1 j m, j 2 j 2 j ⇢ 1 1 2 1 g (x ) B. Furthermore, j j 2
1 1 1 d(g (g (B)),B) d(g (x ),g (x )) = d(x ,x ) 1 1 In particular this must mean that B and gj (gj+1(B)) intersect. This implies that gj gj+1 = si(j) for some si(j) in S. This gives: gm+1 = gmsi(m) = gm 1si(m 1)si(m) = = g1si(1)si(2) si(m) = si(1)si(2) si(m). ··· ··· ··· Hence the set S generates G, and so G is finitely generated. Now consider G equipped with the word metric relative to S. Then, because of the above calculation, d(g, e) m, where e is the identity of G. Furthermore by definition of m, d(x ,g(x )) R d(x ,g(x )) R 0 0 m 0 0 +1, r r 8 so we have d(x ,g(x )) R d(g, e) 0 0 +1 r for all g G. Now since the word metric on G is left-invariant, for any h G, 2 2 d(x ,g(x )) R d(h(x ),h(g(x ))) R d(h, hg) 0 0 +1= 0 0 +1, r r and hence for g ,g G, 1 2 2 d(g (x ),g (x )) R d(g (x ),g (x )) d(g ,g ) 1 0 2 0 +1 1 0 2 0 +1. 1 2 r r Now if h = s s s G, then by the triangle inequality, 1 2 ··· k 2 d(x0,h(x0)) d(x0,s1(x0)) + d(s1(x0),s1(s2(x0))) + + d(s1 sk 1(x0))),s1 sk(x0)))) ··· ··· ··· = d(x ,s (x )) + + d(x ,s (x )) , 0 1 0 ··· 0 k 0 2Rd(h, e) and left invariance again implies for any g ,g G, 1 2 2 d(g (x ),g (x )) 1 0 2 0 d(g ,g ). 2R 1 2 This implies that the map f : G X sending g to g(x ) is coarsely biLipschitz. Since every point ! 0 in X lies in some h(B), we get that every point in X is at most distance R from a point in the image of f, so f is actually quasi-surjective and hence a quasi-isometry. Corollary 1.2.7. Let G be a finitely generated group with two different finite generating sets S and T . Then (G, dS) and (G, dT ) are quasi-isometric. Proof. We first consider G as the metric space (G, dS). Since the metric dS is left-invariant, G acts on itself (through left-multiplication) by isometries. Furthermore, the action is geometric. The set e may be moved to any element of G simply by acting on e with the appropriate element, so the { } { } 9 action is cocompact. We next show the action is properly discontinuous. Of course, if G is finite, there is nothing to prove, so assume G is infinite. Suppose the action of G on itself is not properly discontinuous – that is, there exists a compact set C G so that the set g G : g(C) C = ⇢ { 2 \ 6 ;} is infinite. If C is compact, then it is necessarily finite. For any g G, the set g(C) is also finite 2 and g(C) = C . By the finite Pigeonhole Principle there must be one x C that is in g(C) for | | | | 2 infinitely many g G, and again by the finite Pigeonhole Principle, there must be some y C 2 2 so that gy = x for infinitely may g G. This is clearly absurd, and so a contradiction. Hence, 2 the action of G on itself is properly discontinuous – therefore, it is geometric. By Milnor-Schwarz (and its proof) we may obtain a finite generating set for G, S0, so that (G, dS) and (G, dS0 ) are quasi-isometric. However the above proof works the same way for the metric space (G, dT ) and we may even obtain the same generating set S0 (using the same R in the proof), so it follows that (G, dT ) and (G, dS) are quasi-isometric. The corollary above solves the well-definedness issue of the group G as a metric space. That is, one may discuss G as a metric space without referencing a generating set as long as we consider equivalence up to quasi-isometry. However, we do lose some information: Corollary 1.2.8. A finitely generated group G is quasi-isometric to any of its finite index sub- groups. Corollary 1.2.9. A finite group is quasi-isometric to the trivial group. In particular quasi-isometries do not “see” finite groups or finite subgroups, so we restrict our attention to those groups that are infinite and finitely generated. 1.3 Quasi-Isometric Rigidity Studying metric spaces up to quasi-isometry means being concerned with properties of the metric space in the large scale. One may wonder in fact what useful properties are preserved up to quasi-isometry. If a property is preserved by quasi-isometry, then it must be “robust” in some sense as (in general) the quasi-isometry is poorly behaved. This is the essence of what is called quasi-isometric rigidity. 10 Definition 1.3.1. (Kapovich, 2013) Let G1,G2 be groups. We say G1 is virtually isomorphic to G if there exist finite-index subgroups H G and finite normal subgroups N /H so that H /N 2 i ⇢ i i i 1 1 is isomorphic to H2/N2. In light of the Milnor-Schwarz Lemma and its corollaries, one may easily see that two virtually isomorphic groups will be quasi-isometric. The interest is then to reverse this implication. Definition 1.3.2. (Kapovich, 2013) Let be a class of groups. We say that the class is quasi- G G isometrically rigid (QI rigid) if whenever a group G is quasi-isometric to some H , there 2G exists G0 so that G is virtually isomorphic to G0. 2G A way to summarize the above definition is that the class of groups is QI rigid if whenever a G group G is quasi-isometric to some group in , it follows that G is virtually in . Finding classes G G of groups that are QI rigid is ostensibly a tall order, but it turns out there are many (nice) types of groups that enjoy some sort of QI rigidity. For example, the class of nilpotent groups was shown to be QI rigid by Gromov (1981) as a consequence to his famed theorem on groups of polynomial growth. On the other side, Dyubina (2000) showed that the class of solvable groups is not QI rigid. The proof itself is quite accessible. The study of QI rigidity is just one branch of geometric group theory, but nonetheless, a variety of tools from many areas of mathematics are employed in pursuit of QI rigidity. Furthermore, the concept of the quasi-isometry can be applied in a more analytic setting, as we will see in the subsequent chapters of this dissertation. To appropriately tackle the statement and proof of the author’s main result we need also develop some tools of basic quasiconformal analysis and Pansu differentation. The connection of quasi-isometries to the more famous rigidity theorems, for example the theorems proved by Mostow (1968) or Tukia (1986), will also be explored. 11 CHAPTER 2 PRELIMINARIES 2.1 BiLipschitz Mappings The main result of this dissertation is concerned with certain types of mappings being locally biLipschitz. In this chapter, we aim to refresh the reader on the basic theory of biLipschitz map- pings as well as introducing the concepts of quasiconformality and quasisymmetry. Definition 2.1.1. Let (X, dX ) and (Y,dY ) be metric spaces. We say that f : X Y is L-Lipschitz ! (L 1) if, for any x ,x X, we have 1 2 2 d (f(x ),f(x )) L d (x ,x ). Y 1 2 · X 1 2 Roughly speaking, Lipschitz functions distort distances by no more than some uniform factor. We immediately get the following: Proposition 2.1.2. Let (X, dX ), (Y,dY ) be metric spaces and suppose f : X Y is L-Lipschitz. ! Then f is uniformly continuous. Proof. Let ">0. Pick = " . Then, if x ,x X are such that d (x ,x ) <