1: 

2: %

3: %

4: %

5: \documentclass{amsart}     

6: %

7: %

8: %

9: %\agtart

10: %

11: %

12: \usepackage{amssymb,amsthm,color}

13: \usepackage{epsfig}

14: \usepackage[all]{xy}

15: 

16: %

17: %

18: %

19: 

20: \title{Actions of certain arithmetic groups on Gromov hyperbolic spaces}

21: 

22: %

23: \author[J. F. Manning]{Jason Fox Manning}

24: %\givenname{Jason Fox}

25: %\surname{Manning}

26: \address{University at Buffalo, SUNY}

27: \email{j399m@buffalo.edu}

28: \urladdr{http://www.math.buffalo.edu/~j399m}

29: 

30: %

31: %

32: %

33: %

34: %

35: 

36: %\keyword{arithmetic group}

37: %\keyword{group action}

38: %\keyword{Gromov hyperbolic space}

39: %\keyword{rigidity}

40: %\subject{primary}{msc2000}{20F65}

41: %\subject{secondary}{msc2000}{22E40}

42: 

43: %

44: 

45: 

46: %

47: %

48: %

49: %

50: %

51: %

52: %

53: \newtheorem{theorem}{Theorem}[section]

54: \newtheorem{lemma}[theorem]{Lemma}

55: \newtheorem{corollary}[theorem]{Corollary}

56: \newtheorem{proposition}[theorem]{Proposition}

57: \newtheorem{question}[theorem]{Question}

58: \newtheorem{conjecture}[theorem]{Conjecture}

59: \newtheorem{qconjecture}[theorem]{``Conjecture''}

60: \newtheorem{case}{Case}

61: \newtheorem{subcase}{Case}[case]

62: \newtheorem{subsubcase}{Case}[subcase]

63: \newtheorem{claim}[theorem]{Claim}

64: \newtheorem{subclaim}[theorem]{Subclaim}

65: \newtheorem{step}{Step}

66: 

67: \theoremstyle{definition}

68: \newtheorem{remark}[theorem]{Remark}

69: \newtheorem{definition}[theorem]{Definition}

70: \newtheorem{example}[theorem]{Example}

71: \newtheorem{observation}[theorem]{Observation}

72: 

73: %

74: %

75:  \newcommand{\cd}[1]{\begin{equation*}{\xymatrix{#1}}\end{equation*}}

76:  \newcommand{\cdlabel}[2]{\begin{equation}\label{#1}{\xymatrix{#2}}\end{equation}}

77: \newcommand{\FA}{\ensuremath{(\mathrm{FA})}}

78: \newcommand{\QFA}{\ensuremath{(\mathrm{QFA})}}

79: \newcommand{\isom}[1]{\mathrm{Isom}(#1)}

80: \newcommand{\Mtwo}[4]{\ensuremath{\begin{pmatrix} {#1} & {#2} \\ {#3} &

81:     {#4}\end{pmatrix} }}

82: \newcommand{\textMtwo}[4]{\ensuremath{\left(\begin{smallmatrix} {#1} & {#2} \\ {#3} &

83:     {#4}\end{smallmatrix}\right) }}

84: \newcommand{\mc}[1]{\mathcal{#1}}

85: \newcommand{\uu}[1]{\ensuremath{\underline{#1}}}

86: \newcommand{\gp}[3]{\ensuremath{({#1}\cdot{#2})_{#3}}}

87: 

88: \def\co{\colon\thinspace}

89: \def\R{\mathbb{R}}

90: \def\Z{\mathbb{Z}}

91: \def\C{\mathbb{C}}

92: \def\H{\mathbb{H}}

93: \def\N{\mathbb{N}}

94: \def\Q{\mathbb{Q}}

95: 

96: 

97: 

98: 

99: \begin{document}

100: 

101: \begin{abstract}    

102: We study the variety of actions of a fixed (Chevalley) group on

103: arbitrary geodesic, Gromov hyperbolic spaces.  In high rank we obtain

104: a complete classification. In rank one, we obtain some partial results

105: and give a conjectural picture.

106: \end{abstract}

107: 

108: \maketitle

109: 

110: \tableofcontents

111: 

112: 

113: 

114: 

115: 

116: \section{Introduction}

117: Given a group $G$ one may ask the question: 

118: \begin{question}

119: In what ways can $G$ act non-trivially on a Gromov hyperbolic metric space?

120: \end{question}

121: Many interesting groups can be fruitfully studied via some natural action on

122: a Gromov hyperbolic space.  Examples include the action of an

123: amalgam or HNN extension on its Bass-Serre tree, the action of a

124: Kleinian group on $\H^n$, and the action of the mapping class group of

125: a surface on the curve complex of that same surface.  Alternatively, one can

126: study the space of actions of a fixed group on some (fixed or

127: varying) Gromov hyperbolic metric space.  For example, the $PSL(2,\C)$--character variety of a group

128: parameterizes the space of actions of a fixed

129: group on the hyperbolic space $\H^3$.  Analysis of the structure of

130: this variety has led to many remarkable theorems about $3$--manifolds

131: and their fundamental

132: groups (for an introduction, see \cite{shalen:handbook}).

133: A larger ``variety'' (in scare quotes because there

134: is unlikely to be any algebraic structure) would describe all

135: non-trivial actions on Gromov hyperbolic spaces, up to some

136: appropriate equivalence relation.  We give 

137: suggestions for how to define this equivalence relation and topologize the

138: variety in Section \ref{s:actions}.  Briefly, the equivalence relation

139: is that generated by coarsely equivariant quasi-isometric embeddings.

140: This equivalence is coarser than that given by quasi-conjugacy (as in

141: for instance \cite{msw:quasiactI}), but finer than that given by

142: equivariant homeomorphism of limit sets.

143: In this paper we concentrate on cases in which the set of equivalence

144: classes is particularly simple.

145: 

146: Certain equivalence classes of actions on hyperbolic spaces cannot be

147: ruled out, or 

148: even really analyzed using the tools of negative curvature.  These are

149: the \emph{actions with an invariant horoball} (see Section

150: \ref{ss:combhoro} for the definition, and Theorem \ref{t:horeq} for

151: some characterizations).  Actions with an invariant

152: horoball (in the sense used in this paper) are always elementary; they

153: include the trivial action

154: on a point and actions which preserve some horofunction.  

155: A cobounded

156: action on an unbounded space never has an invariant horoball.

157: 

158: The variety of actions of an irreducible higher rank lattice is

159: expected to be very simple.

160: \begin{conjecture}\label{c:main}

161: If $\Gamma$ is an irreducible lattice in a higher rank Lie group (or

162: in a 

163: nontrivial product of locally compact groups) $G$,

164: there there are finitely many Gromov hyperbolic $G$--spaces (up to

165: coarse equivalence) without

166: invariant horoballs.

167: \end{conjecture}

168: In the case where $G$ is a simple Lie group of rank at least $2$, the

169: only expected actions are those with an invariant horoball.  If $G$

170: has more than one direct factor, then $\Gamma$ projects densely to

171: each factor.  If the factors have rank one, then there will clearly be

172: non-elementary isometric actions of $\Gamma$ on rank one symmetric

173: spaces.  We discuss this case in Section \ref{section:rankone}.

174: 

175: For actions by lattices in nontrivial products of simple Lie

176: groups, the conjecture follows from rigidity results of Monod

177: and Monod-Shalom

178: \cite{monod:superrigid,monodshalom:jdg} 

179: if one restricts attention

180: to Gromov hyperbolic spaces which are also either CAT$(0)$ spaces,

181: proper and cocompact spaces, or bounded valence graphs.

182: (Cf.

183: \cite{gelanderkarlssonmargulis} for CAT$(0)$ spaces.)  

184: For an example

185: of an action of a lattice on a Gromov hyperbolic space which is

186: inequivalent to any action on a Gromov hyperbolic CAT$(0)$ space,

187: consider a lattice as in the Appendix to \cite{manning:qfa}, which has

188: Property (T), but admits a non-trivial pseudocharacter (or homogeneous

189: quasi-morphism).  This

190: pseudocharacter gives rise to a cobounded action on a space

191: quasi-isometric to $\R$, fixing both ends.  On the other hand, an

192: action on a CAT$(0)$ space preserving a point at infinity would give

193: (via the Busemann function) a homomorphism to $\R$, and  a cobounded such

194: action would give an unbounded homomorphism to $\R$.

195: Such a homomorphism is ruled out by Property (T).

196: 

197: \begin{theorem}\label{thm:cdr2} 

198: Suppose that $G$ is a simple Chevalley-Demazure

199: group scheme of rank at least $2$, and let $\mathcal{O}$ be the ring

200: of integers of any number field.  Then any isometric action of

201: $G(\mathcal{O})$ on a Gromov hyperbolic geodesic metric space has

202: an invariant horoball.

203: \end{theorem}

204: Some remarks:

205: \begin{enumerate}

206: \item Some closely related results are proved by Karlsson and Noskov

207:   \cite[Sections 8.2 and 8.3]{karlssonnoskov}; part of our strategy is

208:   similar to theirs, and to that of Fukunaga in \cite{fukunaga:FA}.

209: \item The rank $\geq 2$ assumption is necessary, as the

210:   action of $SL(2,\mathcal{O})$ on $\H^3$ obtained from the inclusion 

211:   $SL(2,\mathcal{O})\longrightarrow SL(2,\C)$  never has an

212:   invariant horoball.

213: \end{enumerate}

214: 

215: The rank one case is discussed in Section 

216: \ref{section:rankone}, where we apply a result of Carter, Keller and

217: Paige to show a weaker theorem for these groups.

218: \begin{theorem}\label{thm:cdr1}

219: Let $\mathcal{O}$ be the ring of integers of a number field, and

220: suppose that $\mathcal{O}$ has infinitely many units.  Suppose $X$ is

221: quasi-isometric to a tree.  Every action of $SL(2,\mathcal{O})$ on $X$

222: has a bounded orbit.

223: \end{theorem}

224: 

225: 

226: \subsection{Relative hyperbolicity and bounded generation}

227: If $G$ is a finitely generated

228: group, $S$ is a generating set for $G$, and

229: $\mc{P}=\{P_1,\ldots,P_n\}$ is a collection of

230: subgroups of $G$, then one can form the \emph{coned space}

231: $C(G,\mc{P},S)$ as follows:  Let $\Gamma(G,S)$ be the Cayley graph of

232: $G$ with respect to $S$. 

233: The coned space $C(G,\mc{P},S)$ is obtained

234: from $\Gamma(G,S)$ by coning each left coset of an element of $\mc{P}$

235: to a point.  (Here the $0$--skeleton of $\Gamma(G,S)$ is implicitly

236: identified with $G$.)

237: 

238: Recall that a graph is \emph{fine} if every edge is contained in only

239: finitely many circuits (i.e. embedded cycles) of any bounded length.

240: \begin{definition}

241: If $G$ is a group with finite generating set $S$, and a collection of

242: subgroups $\mc{P}=\{P_1,\ldots,P_n\}$, then

243: $G$ is \emph{weakly hyperbolic relative to} $\mc{P}$ if

244: $C(G,\mc{P},S)$ is $\delta$--hyperbolic for some $\delta$.  

245: 

246: If, moreover, $C(G,\mc{P},S)$ is \emph{fine}, then $G$ is

247: \emph{(strongly) hyperbolic relative to} $\mc{P}$. 

248: \end{definition}

249: 

250: A special case of weak relative hyperbolicity is bounded generation.

251: The following definition is easily seen to be equivalent to the

252: standard one:

253: \begin{definition}

254: A group $G$ is \emph{boundedly generated} by a collection of

255: subgroups $\mc{P}=\{P_1,\ldots,P_n\}$ if the Cayley graph of $G$ with

256: respect to $\cup\mc{P}$ has finite diameter.

257: 

258: If each $P_i$ is cyclic, generated by $g_i$, we say that $G$ is

259: \emph{boundedly generated by} $\{g_1,\ldots,g_n\}$.

260: \end{definition}

261: 

262: A cobounded action on an unbounded Gromov hyperbolic space does not

263: have an invariant horoball.

264: It is thus a corollary of Theorem \ref{thm:cdr2} that these higher rank

265: $G(\mathcal{O})$ are not strongly relatively hyperbolic

266: with respect to any system of proper subgroups.  

267: \begin{corollary}\label{corollary:notrelhyp}

268: If $L=G(\mc{O})$ is as in Theorem \ref{thm:cdr2}, and $L$ is weakly

269: hyperbolic relative to a system of subgroups

270: $\mc{P}$, then the coned space $C(L,\mc{P},S)$ has finite

271: diameter for any finite generating set $S$.

272: 

273: In particular, $L$ is not strongly hyperbolic relative to any system

274: of proper subgroups.

275: \end{corollary}

276: By possibly altering $\mc{P}$ to add some cyclic subgroups, the first

277: part of the corollary can be restated:  If $L$ is weakly 

278: hyperbolic relative to a system of subgroups $\mc{P}$ which generate

279: $L$, then $L$ is boundedly generated by $\mc{P}$.

280: \begin{remark}

281: Corollary \ref{corollary:notrelhyp} also can be easily deduced from

282: \cite{karlssonnoskov} in some special cases, including $L=SL(n,\Z)$

283: for $n>2$.  The second part of Corollary \ref{corollary:notrelhyp}

284: (about strong relative hyperbolicity) can be deduced from known

285: theorems in a number of ways, perhaps most straightforwardly by

286: combining a theorem of Fujiwara \cite{fujiwara:gromovhyperbolic} with

287: one of Burger and Monod \cite{burgermonod:gafa}.

288: \end{remark}

289: 

290: 

291: \subsection{Outline}

292: In Section \ref{s:preliminaries} we recall some definitions and basic

293: results, first from the theory of Gromov hyperbolic spaces, and second

294: from Chevalley groups over number rings.  In Section \ref{s:actions}

295: an equivalence relation amongst hyperbolic $G$--spaces is proposed, and

296: generalized combinatorial horoballs are introduced.  In Section

297: \ref{s:elementary}, we improve on the statement and proof of

298: Proposition 3.9 of \cite{manning:qfa}, and use the improved version to

299: characterize hyperbolic $G$--spaces with invariant horoballs.

300: In Section \ref{s:bigrank} we prove Theorem \ref{thm:cdr2}, and in

301: Section \ref{section:rankone} we prove Theorem \ref{thm:cdr1}.

302: 

303: \subsection{Acknowledgments}

304: The author thanks Benson Farb and Hee Oh for useful conversations, and

305: Nicolas Monod for helpful comments on an earlier version of this paper.

306: This work was partly supported by an NSF Postdoctoral Research

307: Fellowship (DMS-0301954). 

308: 

309: \section{Preliminaries}\label{s:preliminaries}

310: \subsection{Coarse geometry}

311: \begin{definition}

312: If $X$ and $Y$ are metric spaces, $K\geq 1$ and $C\geq 0$,

313: a \emph{$(K,C)$--quasi-isometric embedding} of $X$ into $Y$ is a function

314: $q\co X\to Y$ so that

315:  For all $x_1$, $x_2\in X$

316: \[\frac{1}{K}d(x_1,x_2)-C\leq d(q(x_1),q(x_2))\leq Kd(x_1,x_2)+C\]

317: 

318: If in addition the map $q$ is \emph{$C$--coarsely onto} -- \emph{i.e.},

319: every $y\in Y$ is

320: distance at most $C$ from some point in $q(X)$ -- then $q$ is called a

321: \emph{$(K,C)$--quasi-isometry}.

322: The two metric spaces $X$ and $Y$ are then

323: said to be \emph{quasi-isometric} to one

324: another.  This is a symmetric condition.

325: \end{definition}

326: \begin{definition}

327: A \emph{$(K,C)$--quasi-geodesic} in $X$

328: is a $(K,C)$--quasi-isometric embedding $\gamma\co\R\to X$.

329: We will occasionally abuse

330: notation by referring to the image of $\gamma$ as a quasi-geodesic.

331: \end{definition}

332: \subsection{Gromov hyperbolic spaces}

333: For more details on Gromov hyperbolic metric spaces, see

334: \cite{bridhaef:book} and \cite{gromov:wordhyperbolic}.

335: A number of equivalent definitions of Gromov hyperbolicity are known.

336: For geodesic spaces, we will use the one based on the existence of

337: \emph{comparison tripods}. 

338: Given a geodesic

339: triangle $\Delta(x,y,z)$ in

340: any metric space, there is a unique \emph{comparison tripod}, $T_\Delta$,

341: a metric tree so

342: that the distances between the three extremal points of the tree,

343: $\overline{x}$, $\overline{y}$ and $\overline{z}$ , are the same as

344: the distances between $x$, $y$ and $z$.

345: There is an obvious map $\pi\co \Delta(x,y,z)\to T_\Delta$ which takes $x$ to

346: $\overline{x}$, $y$ to $\overline{y}$ and $z$ to $\overline{z}$, and

347: which is an isometry on each side of $\Delta(x,y,z)$.

348: \begin{definition}\label{d:geodgromov}

349: A geodesic space $X$ is \emph{$\delta$--hyperbolic} if for any geodesic triangle

350: $\Delta(x,y,z)$ and any point $p$ in the comparison tripod $T_\Delta$,

351:  the diameter of $\pi^{-1}(p)$ is less than $\delta$.  If $\delta$ is

352:  unimportant we may simply say that $X$ is \emph{Gromov hyperbolic}.

353: \end{definition}

354: Gromov hyperbolicity (of geodesic spaces) is a quasi-isometry invariant.

355: 

356: The following proposition about stability of quasi-geodesics in Gromov

357: hyperbolic spaces

358: is well-known (see, e.g. \cite[III.H.1.7]{bridhaef:book}).

359: \begin{proposition}\label{p:quasistable}

360: Let $K\geq 1$, $C\geq 0$, $\delta\geq 0$.  Then there is some

361: $B=B(K,C,\delta)$, so that whenever $\gamma$ and $\gamma'$ are two

362: $(K,C)$--quasi-geodesics with the same endpoints in a

363: $\delta$--hyperbolic geodesic metric space $X$, then the Hausdorff

364: distance between $\gamma$ and $\gamma'$ is at most $B$.

365: \end{proposition}

366: 

367: 

368: We will occasionally need to deal with spaces which are not geodesic.

369: If $X$ is $\delta$--hyperbolic in the sense of Definition

370: \ref{d:geodgromov}, then it

371: satisfies the four point condition:  For all $p_1$, $p_2$, $p_3$,

372: $p_4\in X$, 

373: \begin{equation}\label{fourpoint}

374: d(p_1,p_4)+d(p_2,p_3)\leq\max\{d(p_1,p_2)+d(p_3,p_4)\mbox{, }d(p_1,p_3)+d(p_2,p_4)+2\delta\}.

375: \end{equation}

376: Conversely, if a geodesic space satisfies \eqref{fourpoint}, then it

377: is $6\delta$--hyperbolic in the sense of Definition \ref{d:geodgromov}.

378: (For both these facts, and the below definition, see

379: \cite[III.H.1.22]{bridhaef:book} or \cite{gromov:wordhyperbolic}.)

380: \begin{definition}\label{d:fourpoint}

381: A (not necessarily geodesic) metric space $X$ is \emph{$(\delta)$--hyperbolic}

382: if it satisfies the condition \eqref{fourpoint} above.  If $\delta$ is

383: unimportant, we simply say that $X$ is \emph{Gromov hyperbolic}.

384: \end{definition}

385: 

386: In order to describe the boundary of a Gromov hyperbolic space, we

387: introduce the \emph{Gromov product} notation.

388: \begin{definition}

389: If $x,y$ and $z$ are points in a metric space with a metric

390: $d(\cdot,\cdot)$, then

391: \[ \gp{x}{y}{z}:= \frac{1}{2}(d(x,z)+d(y,z)-d(x,y)).\]

392: \end{definition}

393: We should remark that in the context of Definition \ref{d:geodgromov},

394: $\gp{x}{y}{z}$ is the distance from

395: $\overline{z}$ to the central vertex of the comparison tripod for a

396: geodesic triangle with vertices $x$, $y$ and $z$.

397: \begin{definition}

398: Let $p\in X$ where $X$ is a Gromov hyperbolic space.

399: A sequence of points $\{x_i\}$ in a Gromov hyperbolic space

400: \emph{tends to infinity} if

401: $\lim_{i,j\to\infty}\gp{x_i}{x_j}{p}=\infty$.

402: Two such sequences are \emph{equivalent}, written $\{x_i\}\sim\{y_i\}$, if

403: $\lim_{i,j\to\infty}\gp{x_i}{y_j}{p}=\infty$.

404: The boundary $\partial X$ is the set of equivalence classes of

405: sequences which tend to infinity.

406: \end{definition}

407: The Gromov product extends (by taking a lim sup) to sequences which

408: tend to infinity, and this allows convergence in $\partial X$ to be

409: defined, giving $\partial X$ a natural topology.

410: 

411: \subsection{Isometries of Gromov hyperbolic spaces}

412: Isometries of geodesic hyperbolic spaces can be classified into three

413: types.

414: \begin{definition}

415: Let $f\co X\to X$ be an isometry.  If $x\in X$, we let

416: $O_x=\{f^n(x)\mid n\in \Z\}$.

417: We say that $f$ is \emph{elliptic} if $O_x$ is bounded.  We say that

418: $f$ is \emph{hyperbolic} if

419: $n\mapsto f^{n}(x)$ is a quasi-isometric embedding of $\Z$ into $X$.

420: We say that $f$ is \emph{parabolic} if $O_x$ has a unique limit point

421: in $\partial G$.

422: \end{definition}

423: The following was observed by Gromov \cite[8.1.B]{gromov:wordhyperbolic}.

424: Although it is often stated with an extra hypothesis of

425: properness, this hypothesis is unnecessary (See, for example the

426: proof in \cite[Chapitre 9]{cdp}, where the extra hypothesis is given,

427: but not used).

428: \begin{proposition}\label{p:isometryclass}

429: Every isometry of a geodesic Gromov hyperbolic space is elliptic,

430: parabolic, or hyperbolic.

431: \end{proposition}

432: 

433: 

434: 

435: \begin{lemma}\label{l:commute}

436: Suppose that $G$ acts on the geodesic Gromov hyperbolic space $X$, and that

437: $p\in G$ acts parabolically, fixing $e\in \partial X$.  If $g\in G$

438: commutes with $p$, then $g$ also fixes $e$.

439: \end{lemma}

440: \begin{proof}

441: Let $e'=g(e)$.  We have $e'=gp(e)=pg(e)=p(e')$, so $p$ fixes $e'$.

442: Since $p$ is parabolic, it fixes a unique point in $\partial X$, and

443: so $e'=e$.

444: \end{proof}

445: 

446: \begin{definition}

447: Let $G$ be a finitely generated group, and let $g\in G$.  If $n\mapsto

448: g^n$ is a quasi-isometric embedding, we say that $g$ is

449: \emph{undistorted}.  Otherwise, $g$ is \emph{distorted}.

450: \end{definition}

451: The proof of the following observation is left to the reader.

452: \begin{lemma}\label{lemma:distortion}

453: Let the finitely generated group $G$ act by isometries on a Gromov

454: hyperbolic space $X$.  If $g\in 

455: G$ acts hyperbolically on $X$, then $g$ is undistorted.

456: \end{lemma}

457: 

458: \subsection{Chevalley groups}\label{defs:chevalley}

459: In this section we recall the definition of a Chevalley group over a

460: commutative ring.

461:  (All rings are assumed to have $1 \neq 0$.)

462: The simplest example of a Chevalley group is

463: $SL(n,\Z)$.  If one thinks of $SL(n,\Z)$ as being the ``$\Z$--points

464: of $SL(n,\C)$,'' then the Chevalley-Demazure group scheme

465: identifies what the ``$R$--points of $G$'' are, where

466: $R$ is now allowed to be an arbitrary ring, and $G$ an arbitrary

467: complex semisimple Lie group.  It turns out that this idea is not entirely

468: well-formed, until one fixes an embedding of $G$ into $GL(n,\C)$ for

469: some $n$.  The following exposition is largely 

470: adapted from \cite{abe:chevalley} and \cite{tavgen:bg}.

471: 

472: Let $\rho\co G\to GL(V)$ be a representation of a connected complex semisimple

473: Lie group into the general linear group of a complex vector space $V$

474: of dimension $n$.  We will assume that

475: $d\rho\co\mathfrak{g}\to\mathfrak{gl}(V)$ is faithful (Here

476: $\mathfrak{g}$ is the Lie algebra of $G$, $\mathfrak{gl}(V)$ the Lie

477: algebra of endomorphisms of $V$.

478: Let $\mathfrak{h}$ be a Cartan subalgebra of $\mathfrak{g}$ and let

479: $\Phi$ be the (reduced) root system relative to $\mathfrak{h}$.

480: Let $\Delta$ be a

481: choice of simple roots.  Then there is a \emph{Chevalley basis} 

482: for $\mathfrak{g}$ (see \cite{steinberg:notes} or \cite{chevalley:tohoku}) of

483: the form $\mathcal{B}=\{X_\alpha\ |\ \alpha\in\Phi\}\cup\{H_\alpha\ |\

484: \alpha\in\Delta\}$, so that the $H_\alpha$ generate $\mathfrak{h}$ (and

485: thus commute) and the structure constants are all integral.

486: In other words, the $\Z$--span of $\mathcal{B}$ is actually a Lie algebra over

487: $\Z$.

488: In particular, if $\{\beta - r_{\beta,\alpha} \alpha, \ldots \beta,

489: \ldots \beta + q_{\beta,\alpha} \alpha\}$ are all the roots on the

490: line $\{\beta + t\alpha\ |\ t\in\Z\}$, then:

491: \begin{enumerate}

492: \item $[X_\alpha, X_{-\alpha}] = H_\alpha$.

493: \item $[H_\alpha, X_\beta] = A(\alpha,\beta) X_\beta$, where

494:   $A(\alpha,\beta)$ is an integer determined by $\alpha$

495:   and $\beta$.

496: \item $[X_\alpha,X_\beta] = 0$ if $\alpha + \beta$ is not in $\Phi$;

497:   otherwise $[X_\alpha,X_\beta] = \pm (r_{\beta,\alpha}+1)X_{\alpha+\beta}$.

498: \end{enumerate}

499: Notice that the $X_\alpha$ are all ad-nilpotent, and that the 

500: $H_\alpha$ are ad-semisimple.

501: 

502: It can be shown (see, e.g. \cite[\S27]{humphreys:introduction})

503: that $V$ contains 

504: an ``admissible lattice'' $V_\Z$: This is a free $\Z$--module in $V$

505: which is invariant under  

506: $(d\rho(X_\alpha))^m / m!$ for any $\alpha$ and any $m$.  Let

507: $\{v_1,\ldots,v_n\}$ be a basis for $V_\Z$.  In terms of this basis,

508: $\rho(g)$ can be written as an $n$ by $n$ matrix with complex

509: entries for any $g$.  Let $x_{ij}\co G\to \C$ be the function which

510: simply reads off the $ij$th entry of this matrix.  The $x_{ij}$

511: generate an affine complex algebra $\C(G)$.  Let $\Z(G)$ denote the

512: $\Z$--algebra with the same generators.

513: 

514: We can endow $\Z(G)$ with a Hopf algebra structure by defining

515: a comultiplication $\mu^*$ by

516: $\mu^*(x_{ij}) = \sum_k x_{ik}\otimes x_{kj}$, a counit $\epsilon$

517: by $\epsilon(x_{ij}) = \delta_{ij}$, and an antipode $s$ by

518: $s(f)(g) = f(g^{-1})$.

519: (That $s$ maps  $\Z(G)$ into itself uses the fact

520: that $\det(\rho(g))=\pm 1$ for every $g\in G$ -- this 

521:  follows from semisimplicity.)

522: Since $\Z(G)$ is a Hopf algebra over $\Z$, it defines a functor from

523: rings to groups as follows:

524: For any ring $R$, let 

525: $G(R) = \mathrm{Hom}_\Z(\Z(G), R)$.  (Note that the elements of

526: $G(R)$ are $\Z$--algebra homomorphisms, so they send $1$ to $1$.)

527: We define a group operation

528: $\bullet$ on $G(R)$ by $\rho\bullet\sigma = (\rho\otimes\sigma)\circ

529: \mu^*$.

530: (In other words, $\rho\bullet\sigma(x_{ij}) = \sum_k

531: \rho(x_{ik})\sigma(x_{kj})$.) 

532: 

533: Some observations:

534: \begin{enumerate}

535: \item $G(\C)$ can be naturally identified with the image of $\rho\co

536:   G\to GL(V)$ -- if $\rho$ is assumed to be faithful, then clearly

537:   $G(\C)\cong G$ doesn't depend on $\rho$.  On the other hand, for $R$

538:   arbitrary, $G(R)$ does  depend  on $\rho$.

539: \item  The assignment $R\mapsto G(R)$ is a covariant functor from

540:   commutative rings to groups.  This functor is often called a

541:   ``Chevalley-Demazure group scheme.''

542: \item\label{zoft}  Any Hopf algebra gives such a functor.  An

543:   important example is

544:   $\Z[t]$, with a Hopf ($\Z$--)algebra structure so that:

545: \begin{enumerate}

546: \item The comultiplication satisfies $\mu^*(t)=t\otimes 1 + 1\otimes

547:   t$.

548: \item The counit satisfies $\epsilon(t)=0$.

549: \item The antipode satisfies $S(t)=-t$

550: \end{enumerate}

551: In this case the functor $R\mapsto \mathrm{Hom}(Z[t],R)$ is the

552:   ``forgetful'' functor, which takes a ring to its underlying Abelian

553:   group.

554: \item\label{contrafunctor} Since $G(R)$ is $Hom_\Z(\Z(G),R)$, any morphism of Hopf algebras

555:  $\Z(G)\to A$ where $A$ is

556: some other Hopf algebra, will give rise to a homomorphism of groups

557:  $Hom_\Z(A,R)\to G(R)$. 

558: \end{enumerate}

559: \begin{definition}

560: Let $\alpha$ be a root.  Then since $d\rho(X_\alpha)$ is

561: nilpotent, the formal sum $\exp ( t X_\alpha ) = \sum_{m=1}^\infty

562: t^m \frac{d\rho(X_\alpha)^m}{m!} $ is a matrix with entries which are

563: polynomials in $t$.  

564: Because $\frac{d\rho(X_\alpha)^m}{m!}$ preserves

565: $V_\Z$, each $x_{ij}(\frac{d\rho(X_\alpha)^m}{m!})$ is an integer.

566: Thus we get a map

567: \[\mathrm{ev}_\alpha\co \Z(G) \to \Z[t],\]

568: which sends $x_{ij}$ to the (integral) polynomial in $t$ which appears

569: in the $ij$'th place of the matrix $\exp( t X_\alpha)$.  The map

570: $\mathrm{ev}_\alpha$ is a morphism of Hopf algebras, where $\Z[t]$ is given a

571: Hopf algebra structure as in observation \eqref{zoft} above.

572: This morphism gives rise via observation \eqref{contrafunctor} to a

573: homomorphism of the additive group underlying $R$ into the group

574: $G(R)$

575: \[x_\alpha\co R\to G(R).\]  

576: The image of this map is the \emph{root subgroup}

577: of $G(R)$ corresponding to $\alpha$.

578: \end{definition}

579: 

580: \begin{definition}

581: The subgroup of $G(R)$ generated by the root subgroups

582: is denoted $E(R)$.  

583: \end{definition}

584: 

585: In this paper, we focus mainly on the special case that $R$ is the

586: ring of integers of a number field $k$ and $\Phi$ has rank at least

587: two.  In this case, $G(R)=E(R)$ by 

588: a result of Matsumoto \cite{matsumoto:csp}.  By a result 

589: of Carter and Keller \cite{carterkeller:slno} in case $\Phi=A_n$,

590: and  Tavgen$'$ \cite{tavgen:bg} in general,

591: $G(R)$ is boundedly generated by the root subgroups.

592: 

593: It should also be noted that in this special case, $G(R)$ is actually

594: an irreducible lattice in a semisimple Lie group.  Indeed, if $s$ and

595: $t$ are the number of real 

596: and complex places of $k$, respectively, then $G(R)$ is a lattice in the

597: Lie group 

598: \[ G(\R)^s\times G(\C)^t.\]

599: 

600: \section{Equivalence of actions}\label{s:actions}

601: We wish to study the variety of actions of a group $G$ on

602: Gromov hyperbolic spaces, up to some kind of coarse equivalence.  By a

603: {\em (Gromov) hyperbolic $G$--space}, we will always mean a Gromov

604: hyperbolic geodesic metric space, equipped with an isometric

605: $G$--action.  

606: 

607: 

608: \begin{definition}

609: Let $X$ and $Y$ be Gromov hyperbolic $G$--spaces.  We say that $X$ and

610: $Y$ are {\em equivalent} if they lie in the same

611: equivalence class, under the equivalence relation generated by

612: coarsely equivariant quasi-isometric embeddings.

613: \end{definition}

614: The following proposition should serve to clarify this equivalence

615: relation. 

616: \begin{proposition}If $X_1$ and $X_2$ are equivalent Gromov hyperbolic

617:   $G$--spaces, then there is

618:   a third Gromov hyperbolic $G$--space $V$ which coarsely equivariantly

619:   quasi-isometrically embeds in both $X_1$ and $X_2$.

620: \end{proposition}

621: \begin{proof}

622: The key claim is the following:

623: \begin{claim}\label{c:diamond}

624: If $V$ and $W$ are hyperbolic $G$--spaces which coarsely equivariantly

625:   quasi-isometrically embed in a third hyperbolic $G$--space $X$, then

626:   there is a fourth hyperbolic $G$--space $A$ which coarsely

627:   equivariantly quasi-isometrically embeds into $V$ and $W$.

628: \end{claim}

629: Before proving the claim, let us see how it implies the proposition.

630: If $X_1$ and $X_2$ are equivalent hyperbolic $G$--spaces, they 

631: must be joined by a sequence of coarsely

632: equivariant quasi-isometric embeddings:

633: \cdlabel{long}{ & V_1\ar[dl]\ar[dr] & & \cdots\ar[dl]\ar[dr] & &V_n\ar[dl]\ar[dr] & &

634:   V_{n+1}\ar[dl]\ar[dr] & \\

635: X_1 & & Z_1 &\cdots & Z_{n-1} & & Z_n & & X_2}

636: By applying the claim with $V=V_n$, $W=V_{n+1}$, and $X=Z_n$, we

637: obtain a hyperbolic $G$--space $A$ which coarsely equivariantly

638: quasi-isometrically embeds into both $V_n$ and $V_{n+1}$, and hence

639: into both $Z_{n-1}$ and $X_2$.  We can thus shorten the sequence

640: \eqref{long}, unless it is a shortest possible such sequence,

641: \cd{ & V_1\ar[dl]\ar[dr] & \\

642:  X_1 & & X_2,}

643: in which case the proposition is verified.

644: 

645: \begin{proof} (of Claim \ref{c:diamond})

646: We first construct a $G$--space $A_1$ which coarsely equivariantly

647: quasi-isometrically embeds in both $V$ and $W$ and is Gromov

648: hyperbolic but not geodesic.  We then show that $A_1$ is

649: quasi-isometric to a geodesic $G$--space $A$.

650: 

651: Choose $\delta>0$ so that all of $V$, $W$, and $X$ are

652: $\delta$--hyperbolic spaces.

653: By hypothesis, we may choose $K\geq 1$ and $C\geq 0$ so that there are maps

654: $\phi\co V\to X$ and $\psi\co W\to Y$ which are $C$--coarsely equivariant

655: $(K,C)$--quasi-isometric embeddings.  We choose constants $J_0<J_1<J_2$:

656: Let

657: $J_0 = 2B(K,C,\delta)+2\delta$,

658: where $B(K,C,\delta)$ is the constant of quasi-geodesic stability from

659: Proposition \ref{p:quasistable}, let $J_1=J_0+2C$, and let $J_2=4 J_1$.

660: 

661: Let $A_0$ be the subset of $V\times

662: W$ given by $A_0=\{(v,w)\mid d(\phi(v),\psi(w))\leq J_0\}$, and let $A_1$

663: be the smallest $G$--equivariant subset of $V\times W$ containing

664: $A_0$.  We endow $A_1$ with the (pseudo)metric

665: \[

666:  d((v_1,w_1),(v_2,w_2))=d(\phi(v_1),\phi(v_2))+d(\psi(v_2),\psi(w_2)).

667: \]

668: \begin{subclaim}

669: $A_1$ is Gromov hyperbolic.

670: \end{subclaim}

671: \begin{proof}

672: Since $A_1$ is not a geodesic space, we must work with the four-point

673: definition \ref{d:fourpoint}.  We will show that

674: $A_1$ is $(2\delta+4J_1)$--hyperbolic in the sense of Definition

675: \ref{d:fourpoint}.  

676: Let $\{p_i=(v_i,w_i)\mid i=1,\ldots,4\}$ be four points in $A_1$.  

677: For each $i$, we write $\uu{v}_i$ for $\phi(v_i)$ and $\uu{w}_i$ for

678: $\psi(w_i)$. By

679: reordering the points if necessary, we can assume that

680: \[d(\uu{v}_1,\uu{v}_3)+d(\uu{v}_2,\uu{v}_4)\geq

681: d(\uu{v}_1,\uu{v}_2)+d(\uu{v}_3,\uu{v}_4),\]

682: as in Figure \ref{f:palimpsest}.

683: \begin{figure}[htbp]

684: \begin{center}

685: \begin{picture}(0,0)%

686: \epsfig{file=palimpsest.pstex}%

687: \end{picture}%

688: \setlength{\unitlength}{3947sp}%

689: %

690: \begingroup\makeatletter\ifx\SetFigFont\undefined%

691: \gdef\SetFigFont#1#2#3#4#5{%

692:   \reset@font\fontsize{#1}{#2pt}%

693:   \fontfamily{#3}\fontseries{#4}\fontshape{#5}%

694:   \selectfont}%

695: \fi\endgroup%

696: \begin{picture}(3300,1939)(2551,-2744)

697: \put(5551,-2686){\makebox(0,0)[lb]{\smash{{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\phi(v_3)$}%

698: }}}}

699: \put(2776,-961){\makebox(0,0)[lb]{\smash{{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\phi(v_2)$}%

700: }}}}

701: \put(2626,-1186){\makebox(0,0)[lb]{\smash{{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\psi(w_2)$}%

702: }}}}

703: \put(2551,-2011){\makebox(0,0)[lb]{\smash{{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\phi(v_1)$}%

704: }}}}

705: \put(2626,-2236){\makebox(0,0)[lb]{\smash{{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\psi(w_1)$}%

706: }}}}

707: \put(5851,-1411){\makebox(0,0)[lb]{\smash{{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\phi(v_4)$}%

708: }}}}

709: \put(5776,-1186){\makebox(0,0)[lb]{\smash{{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\psi(w_4)$}%

710: }}}}

711: \put(5776,-2461){\makebox(0,0)[lb]{\smash{{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\psi(w_3)$}%

712: }}}}

713: \end{picture}%

714: \caption{Points in $X$.}

715: \label{f:palimpsest}

716: \end{center}

717: \end{figure}

718: There are then two cases.

719: 

720: In case

721: \[ d(\uu{w}_1,\uu{w}_3)+d(\uu{w}_2,\uu{w}_4)\geq

722: d(\uu{w}_1,\uu{w}_2)+d(\uu{w}_3,\uu{w}_4),\]

723: then 

724: $\max\{d(p_1,p_3)+d(p_2,p_4),d(p_1,p_2)+d(p_3,p_4)\}=d(p_1,p_2)+d(p_2,p_4)$.

725: Applying $\delta$--hyperbolicity in $X$, we obtain

726: \begin{eqnarray*}

727: d(p_1,p_4)+d(p_3,p_2) & = &

728: d(\uu{v}_1,\uu{v}_4)+d(\uu{v}_2,\uu{v}_3)+d(\uu{w}_1,\uu{w}_4)+d(\uu{w}_2,\uu{w}_3)

729: \\

730:  & \leq & d(\uu{v}_1,\uu{v}_3)+d(\uu{v}_2,\uu{v}_4)+ 2\delta

731: +d(\uu{w}_1,\uu{w}_3)+d(\uu{w}_2,\uu{w}_4)+ 2\delta \\

732: & = & d(p_1,p_3)+d(p_2,p_4)+ 4\delta \\ 

733: & = & \max\{d(p_1,p_3)+d(p_2,p_4),d(p_1,p_2)+d(p_3,p_4)\} + 2

734: (2\delta) \\ 

735: & \leq & \max\{d(p_1,p_3)+d(p_2,p_4),d(p_1,p_2)+d(p_3,p_4)\} \\

736: & & + 2(2\delta+4 J_1)

737: \end{eqnarray*}

738: as required.

739: 

740: If on the other hand 

741: \[ d(\uu{w}_1,\uu{w}_3)+d(\uu{w}_2,\uu{w}_4) <

742: d(\uu{w}_1,\uu{w}_2)+d(\uu{w}_3,\uu{w}_4),\]

743: then it is still true (since $d(\uu{w_i},\uu{v_i})\leq J_1)$) that 

744: \begin{eqnarray*}

745: d(\uu{w}_1,\uu{w}_3)+d(\uu{w}_2,\uu{w}_4) & \geq &

746: d(\uu{v}_1,\uu{v}_3)+d(\uu{v}_2,\uu{v}_4) - 4J_1 \\

747: & \geq & d(\uu{v}_1,\uu{v}_2)+d(\uu{v}_3,\uu{v}_4) - 4J_1 \\

748: & \geq & d(\uu{w}_1,\uu{w}_2)+d(\uu{w}_3,\uu{w}_4) - 8J_1.

749: \end{eqnarray*}

750: We therefore obtain

751: \begin{eqnarray*}

752: d(p_1,p_4)+d(p_3,p_2) & = &

753: d(\uu{v}_1,\uu{v}_4)+d(\uu{v}_2,\uu{v}_3)+d(\uu{w}_1,\uu{w}_4)+d(\uu{w}_2,\uu{w}_3)

754: \\

755: & \leq & d(\uu{v}_1,\uu{v}_3)+d(\uu{v}_2,\uu{v}_4)+ 2\delta

756: +d(\uu{w}_1,\uu{w}_2)+d(\uu{w}_3,\uu{w}_4)+ 2\delta \\

757: & \leq & d(p_1,p_3)+d(p_2,p_4) + 2(2\delta+4 J_1) \\

758: & \leq & \max\{d(p_1,p_3)+d(p_2,p_4),d(p_1,p_2)+d(p_3,p_4)\}\\& & + 2

759: (2\delta+4 J_1),

760: \end{eqnarray*}

761: which finishes the proof of the Subclaim.

762: \end{proof}

763: Though the space $A_1$ is hyperbolic, and $G$ quasi-acts on $A_1$

764: via the diagonal action on $V\times W$, the space $A_1$ is not  geodesic,

765: and $G$ does not act by isometries.  Both of these issues can be fixed

766: at once, by replacing $A_1$ by an appropriate graph.  

767: Specifically, we let $A$ be the graph with vertex set

768: $V(A)=A_1$, and with an edge between every pair of vertices $p$ and

769: $q$ so that there exists some $g$ with $d(gp,gq)\leq J_2:=4J_1$.  Clearly

770: this graph is a geodesic $G$--space.

771: 

772: Let $\iota\co A_1\to A$ be the map which takes a point to the

773: corresponding vertex.  We claim that $\iota$ is a quasi-isometry, and

774: so $A$ is a Gromov hyperbolic space.

775: 

776: Let $a=(v_a,w_a)$ and $b=(v_b,w_b)$ be points in $A_1$ so that

777: $d(\iota(a),\iota(b))=1$.  There is some $g$ with 

778: \[d(ga,gb)=d(\phi(gv_a),\phi(gv_b))+d(\psi(gv_a),\psi(gv_b))\leq J_2.\]

779: Since $\phi$ and $\psi$ are both $C$--coarsely equivariant,

780: $d(\phi(gv_a),\phi(gv_b))$ differs  by

781: at most $2C$ from $d(\phi(v_a),\phi(v_b))$; similarly, $d(\psi(gw_a),\psi(gw_b))$ differs from $d(\psi(w_a),\psi(w_b))$  by

782: at most $2C$.  It follows that $d(ga,gb)$ differs by at most $4C$ from

783: $d(a,b)$, and so $d(a,b)\leq J_2+4C<\frac{3}{2}J_2$.  As this is true for every pair

784: of points connected by an edge, we deduce

785: \begin{equation}\label{lower}

786:  d(\iota(p),\iota(q)) \geq \frac{2}{3J_2} d(p,q) 

787: \end{equation}

788: for any pair of points $p$, $q\in A_1$.

789: 

790: We now obtain the complementary bound to \eqref{lower}.  Suppose $p=(v_1,w_1)$

791: and $q=(v_2,w_2)$ are any two points in $A_1$.  We write $\uu{v_1}$

792: for $\phi(v_1)$ and so on as before.

793: We will show that

794: \begin{equation}\label{upper}

795: d(\iota(p),\iota(q))\leq  \frac{1}{J_1}d(p,q)+2

796: \end{equation}

797: by constructing a path joining $\iota(p)$ to $\iota(q)$ in $A$.  The

798: vertices of this path will be points in $A_0\subset V\times W$ so that the first

799: coordinate lies on a geodesic between $v_1$ and $v_2$, while the

800: second lies on a geodesic between $w_1$ and $w_2$.

801: 

802: If $d(p,q)\leq 2J_2 = 8J_1$, then \eqref{upper} is automatically

803: satisfied.  We may therefore assume that $d(p,q)\geq 2J_2$.  Because

804: $d(\uu{v}_1,\uu{w}_1)$ and $d(\uu{w}_1,\uu{w}_2)$ are at most $J_0$,

805: we have

806: $\min\{d(\uu{v}_1,\uu{v}_2),d(\uu{w}_1,\uu{w}_2)\}\geq J_2-J_0 >

807: 3J_0$. It follows that

808: $\gp{\uu{v}_2}{\uu{w}_2}{\uu{v}_1}>\gp{\uu{w}_1}{\uu{w}_2}{\uu{v}_1}$, as in

809: Figure \ref{f:bar}.

810: \begin{figure}[htbp]

811: \begin{center}

812: \input{bar.pstex_t}

813: \caption{A pair of pairs of points in $X$, together with comparison tripods.

814: The dashed line on the left is the image of a geodesic in $V$; the one

815: on the right is the image of a geodesic in $W$.}

816: \label{f:bar}

817: \end{center}

818: \end{figure}

819: In fact,

820: $\gp{\uu{v}_2}{\uu{w}_2}{\uu{v}_1}-\gp{\uu{w}_1}{\uu{w}_2}{\uu{v}_1}>J_0\geq

821: J_1/2$ (since clearly $B(K,C,\delta)\geq C$).

822: We may thus choose real numbers 

823: \[ \gp{\uu{w}_1}{\uu{w}_2}{\uu{v}_1}=t_0<t_1<\cdots<t_k =

824: \gp{\uu{v}_2}{\uu{w}_2}{\uu{v}_1}\]

825: satisfying $\frac{J_1}{2}<|t_{i+1}-t_i|\leq J_1$ for each $i$.

826: Note that $k \leq \frac{1}{J_1}d(p,q)$ for such a choice.  

827: 

828: Let

829: $\gamma$ be a unit speed geodesic from $\uu{v}_1$ to $\uu{w}_2$.  For

830: each $i$ between $0$ and $k$, there are points $x_i'$ on a geodesic

831: between $\uu{v}_1$ and $\uu{w}_1$ and $y_i'$ on a geodesic between

832: $\uu{w}_1$ and $\uu{w}_2$ which are distance at most $\delta$ from

833: $\gamma(t_i)$.  Since $V$ and $W$ are geodesic spaces, there are

834: geodesics $[v_1,v_2]$ in $V$ and $[w_1,w_2]$ in $W$.  The maps $\phi$

835: and $\psi$ send these geodesics

836: to $(K,C)$--quasi-geodesics in $X$.  Applying quasi-geodesic stability

837: (Proposition \ref{p:quasistable}), there are points

838: $x_i\in [v_1,v_2]$, $y_i\in [w_1,w_2]$ so that $\phi(x_i)$ and

839: $\psi(y_i)$ lie within $B+\delta$ of $\gamma(t_i)$.  Since

840: $d(\phi(x_i),\psi(y_i))\leq 2B+2\delta=J_0$, the point $p_i=(x_i,y_i)$

841: lies in $A_0\subseteq A_1$.  Moreover, $d(p_i,p_{i+1})\leq

842: 2(2(B+\delta)+J_1)<J_2$, and so $d(\iota(p_i),\iota(p_{i+1}))\leq 1$.

843: 

844: Finally, one notes that 

845: \[

846: d(\uu{v}_1,\phi(x_0))+d(\uu{w}_1,\psi(y_0))  \leq  J_0+2\delta+2 B=2J_0<J_2

847: \]

848: and likewise for $d(\uu{v}_1,\phi(x_0))+d(\uu{w}_1,\psi(y_0))$, from

849: which it follows that $d(\iota(p),\iota(p_0))$ and

850: $d(\iota(q),\iota(p_k))$ are at most one.  Thus

851: \[ d(\iota(p),\iota(q))\leq k+2 \leq \frac{1}{J_1} d(p,q)+2.\]

852: It is obvious that $\iota$ is $1$--almost onto, and so $\iota$ is a

853: quasi-isometry.

854: 

855: It follows that $A$ is a Gromov hyperbolic geodesic $G$--space

856: which coarsely equivariantly, quasi-isometrically embeds into $V$ and

857: $W$.

858:   

859: \end{proof}

860: \end{proof}

861: 

862: \begin{remark}

863: We record some observations about this equivalence:

864: \begin{enumerate}

865: \item Let $\Gamma$ be a  group acting isometrically on $\H^2$.  This

866:   action extends in an obvious way to either $\H^3$ or $\C\H^2$.

867:   Although there is no quasi-isometric embedding either from $\H^3$ to

868:   $\C\H^2$ or vice versa, these actions are equivalent under the

869:   equivalence relation.

870: \item A Gromov hyperbolic $G$--space has a bounded orbit if and only if

871:   it is equivalent to the trivial $G$--space consisting of a single

872:   point.

873: \item If $X$ and $Y$ are equivalent Gromov hyperbolic $G$--spaces,

874:   $x\in X$ and $y\in Y$, then the

875:   limit sets $\Lambda(X)=\{e\in \partial X\cap \overline{Gx}\}$ and

876:   $\Lambda(Y)=\{e\in \partial Y\cap \overline{Gy}\}$ are equivariantly

877:   homeomorphic.

878: \item The equivalence is perfectly well-defined in the more general

879:   setting of {\em quasi-actions} on geodesic Gromov hyperbolic

880:   spaces.  Call a geodesic Gromov hyperbolic space with a

881:   $G$--quasi-action a {\em hyperbolic quasi-$G$--space}.  Every

882:   hyperbolic quasi-$G$--space is equivalent to some hyperbolic

883:   $G$--space. 

884: \end{enumerate}

885: \end{remark}

886: 

887: Given a group acting on a hyperbolic $G$--space $X$ and some basepoint

888: $x_0\in X$, one obtains a metric on $G$ given by the formula

889: \[d(g,h) = d(x_0,g^{-1}hx_0).\]

890: This metric is obviously determined by its values on $\{1\}\times G$.

891: The compact-open topology on real-valued functions on $G$ thus induces

892: a topology on the set of pointed hyperbolic $G$--spaces.

893: The quotient topology on the space of equivalence classes

894: of Gromov hyperbolic $G$--spaces is not Hausdorff.   For example, if

895: $G$ is a surface 

896: group, then all of Teichm\"uller space is identified to a single

897: point, whose closure contains many inequivalent actions of $G$ on

898: $\R$--trees.

899: 

900: 

901: \subsection{Combinatorial horoballs}\label{ss:combhoro}

902: Combinatorial horoballs of the simplest possible type were defined in

903: \cite{rhds}, and used as building blocks for complexes naturally

904: associated to relatively hyperbolic groups.  The point there as here

905: is that these spaces can be used to ``hide'' an action on a

906: non-hyperbolic space in an action on a hyperbolic space.  There is

907: some flexibility as to how this can 

908: happen which is deliberately ignored in \cite{rhds}; here we give a

909: more general construction.  

910: 

911: \begin{definition}\label{d:admissible}

912: Let $X$ be a graph, acted on by $G$, and suppose that 

913: \[\{B_i\co  X^{(0)}\to 2^{X^{(0)}}\}_{i\in \N}\]

914: is a collection of functions.  We will say that $B_*$ is

915: \emph{admissible} if it

916: satisfies the following

917: four axioms:

918: \begin{enumerate}

919: \item Connectedness:  If $v\in X^{(0)}$, then

920:   $B_1(v)=\{w\in X^{(0)}\mid d_X(v,w)\leq 1\}$.

921: \item\label{ax:exp} Exponential growth: Let $v$, $w\in X^{(0)}$ and $n\in \N$.  If

922:   $v\in B_n(w)$, then $B_n(v)\subset  B_{n+1}(w)$. 

923: \item Symmetry: Let $v$, $w\in X^{(0)}$ and $n\in \N$.

924:   If $v\in B_n(w)$, then $w\in B_n(v)$.

925: \item $G$--equivariance: If $w\in X^{(0)}$, $n\in \N$, and $g\in G$,

926:   then $g B_n(w)=B_n(gw)$.

927: \end{enumerate}

928: \end{definition}

929: 

930: \begin{definition}\label{d:choroball}

931: Let $B_*$ be a sequence of functions $\{B_i\co  X^{(0)}\to

932: 2^{X^{(0)}}\}_{i\in \N}$.  The \emph{combinatorial horoball based on

933:   $X$ and $B_*$}, or $\mc{H}(X,B_*)$, is the graph defined as follows:

934: \begin{enumerate}

935: \item $\mc{H}(X,B_*)^{(0)} = X^{(0)}\times \N$.

936: \item If $n\in \N$ and $v\in X^{(0)}$, then $(v,n)$ is connected to

937:   $(v,n+1)$ by an edge (called a \emph{vertical edge}).

938: \item If $n\in \N$ and $v\in B_n(w)$, then $(v,n)$ is connected to

939:   $(w,n)$ by an edge (called a \emph{horizontal edge}).

940: \end{enumerate}

941: \end{definition}

942: We leave the following Lemma as an exercise.

943: \begin{lemma}\label{l:horohyp}

944: Let $G$ act on the graph $X$.  If $B_*$ is admissible, then

945: $\mc{H}(X,B_*)$ is Gromov hyperbolic, and the action of $G$ on $X$

946: induces an action of $G$ on $\mc{H}(X,B_*)$.

947: \end{lemma}

948: 

949: Because of the flexibility of this construction, a group

950: typically admits many inequivalent actions on horoballs.

951: 

952: \begin{definition}\label{d:ihoroball}

953: A Gromov hyperbolic $G$--space {\em has an invariant horoball} if it is

954: equivalent to a $G$--space of the form $\mc{H}(X,B_*)$, for some graph

955: $X$ with a $G$--action, and some family $B_*$ which is admissible.

956: \end{definition} 

957: In the next section we give some other characterizations of $G$--spaces

958: with invariant horoballs (Theorem \ref{t:horeq}).

959: 

960: \section{Elementary actions and pseudocharacters}\label{s:elementary}

961: In this section we show how an elementary action by $G$ on a hyperbolic space

962: gives rise to a pseudocharacter on $G$ which ``picks out'' the

963: elements which act hyperbolically.  We then see that if no element

964: acts hyperbolically, then the action has an invariant horoball.

965: 

966: \subsection{The pseudocharacter coming from an elementary action}

967: Recall the following definitions.

968: \begin{definition}

969: An action of $G$ on a hyperbolic space $X$ is \emph{elementary} if it

970: is either equivalent to the trivial action on a point, or if the

971: induced action on $\partial X$ has a global fixed point.\footnote{This definition is slightly more restrictive than the

972:   usual one, which allows for a pair of points in $\partial X$ to be

973:   preserved.} 

974: \end{definition}

975: \begin{definition}

976: A \emph{quasicharacter} (or quasi-morphism) on a group $G$ is a

977: real valued function $q$ on $G$ satisfying 

978: \begin{equation}\label{defect}

979: |q(gh)-q(g)-q(h)|<C\mbox{, for all } g,h\in G.

980: \end{equation}

981: The \emph{defect} of a quasicharacter is the

982: smallest $C$ so that \eqref{defect} is satisfied.  A

983: \emph{pseudocharacter} (or homogeneous quasi-morphism) is a

984: quasicharacter $p$ which satisfies the additional condition

985: \begin{equation*}

986:   p(g^n)=n p(g)\mbox{, for all }g\in G, n\in \Z

987: \end{equation*}

988: \end{definition}

989: In this subsection we show how an elementary action on a hyperbolic

990: space gives rise to a pseudocharacter which is nonzero precisely on

991: the elements which act hyperbolically.

992: We begin by studying ``quasi-horofunctions'' on the space

993: $X$, corresponding to a fixed point at infinity.  A

994: quasi-horofunction restricted to an arbitrary orbit will give 

995: a quasicharacter, which can then be homogenized to give the desired

996: pseudocharacter.

997:   

998: \begin{definition}

999: (cf. \cite[7.5.D]{gromov:wordhyperbolic})

1000: Let $\mathbf{x}=\{x_i\}$ be a sequence tending to infinity in the geodesic

1001: hyperbolic space $X$.  The \emph{quasi-horofunction coming from}

1002: $\mathbf{x}$ is the function $\eta_{\mathbf{x}}\co X\to \R$ given by

1003: \[ \eta_{\mathbf{x}}(a) = \limsup_{n\to\infty}(d(a,x_n)-d(x_0,x_n)). \]

1004: \end{definition}

1005: 

1006: We use the following observation repeatedly:

1007: \begin{observation}\label{o:fourdelta}

1008: Let $A$, $B$, $C$ and $D$ be four points in the $\delta$--hyperbolic

1009: space $X$. If $\gp{C}{D}{A}$ and $\gp{C}{D}{B}$ are both larger than $d(A,B)$,

1010: then 

1011: \[ | (d(B,C) - d(A,C)) - (d(B,D) - d(A,D)) | \leq 4\delta. \]

1012: \end{observation}

1013: The observation \ref{o:fourdelta} implies in particular:

1014: \begin{lemma}\label{l:approxbyterm}

1015: If $a\in X$, and $\mathbf{x}=\{x_i\}$ tends to infinity in $X$, then

1016: for all $n$ sufficiently large, 

1017: \[| \eta_{\mathbf{x}}(a)-(d(a,x_n)-d(x_0,x_n))| \leq 4\delta.\]

1018: \end{lemma}

1019: 

1020: We now can describe the dependence of $\eta_{\mathbf{x}}$ on the sequence

1021: $\mathbf{x}$.

1022: \begin{lemma}\label{l:key}

1023: Let $\mathbf{x}=\{x_i\}$ and $\mathbf{y}=\{y_i\}$ be two sequences of

1024: points in the geodesic $\delta$--hyperbolic space $X$

1025: which tend to the same point in $\partial X$.  For any point $a\in X$,

1026: we have 

1027: \[ | \eta_{\mathbf{x}}(a)-\eta_{\mathbf{y}}(a)-\eta_{\mathbf{x}}(y_0)|

1028: \leq 16\delta. \]

1029: \end{lemma}

1030: \begin{proof}

1031: Since $\mathbf{x}$ and $\mathbf{y}$ tend to the same point at

1032: infinity, we may choose $N$ so that 

1033: $(z,z')_\alpha> 2\thinspace \mathrm{diam}\{a,x_0,y_0\}$ for every $z$, $z'$ in

1034: $\{x_i \mid i\geq N\}\cup \{y_i \mid i\geq N\}$.  Using Lemma

1035: \ref{l:approxbyterm} three times, the quantity 

1036: \[|\eta_{\mathbf{x}}(a)-\eta_{\mathbf{y}}(a)-\eta_{\mathbf{x}}(y_0)|\]

1037: differs by at most $12\delta$ from

1038: \begin{equation}\label{easy}

1039: | (d(a,x_N) - d(a,y_N)) - (d(y_0,x_N)-d(y_0,y_N))|.

1040: \end{equation}

1041: By Observation \ref{o:fourdelta}, the quantity \eqref{easy} is at most

1042: $4\delta$.  The Lemma follows.

1043: \end{proof}

1044: 

1045: Using Lemma \ref{l:key}, we deduce that an isometry of $X$ changes

1046: $\eta_{\mathbf{x}}(a)$ by approximately the same amount, independent of

1047: the $a\in X$ chosen:

1048: \begin{proposition}\label{p:isom}

1049: Let $X$ be a geodesic $\delta$--hyperbolic space, and suppose that

1050: $\mathbf{x}=\{x_i\}$ tends to $e\in \partial X$.  Let $a$ be any point

1051: in $X$.

1052: If $g\in \mathrm{Isom}(X)$ fixes $e$, then

1053: $\eta_{\mathbf{x}}(ga)$ differs from

1054: $\eta_{\mathbf{x}}(a)+\eta_{\mathbf{x}}(gx_0)$ by at most $16\delta$.

1055: \end{proposition}\label{p:trans}

1056: \begin{proof}

1057: First note that if $g\mathbf{x}$ is the sequence $\{gx_i\}$, then

1058: \[\eta_{\mathbf{x}}(a) = \eta_{g\mathbf{x}}(ga).\]

1059: But by Lemma \ref{l:key}, 

1060: \[ | \eta_{\mathbf{x}}(ga)-\eta_{g\mathbf{x}}(ga)-\eta_{\mathbf{x}}(g x_0)|

1061: \leq 16\delta. \]

1062: \end{proof}

1063: 

1064: \begin{corollary}\label{c:quasimorphism}

1065: Suppose $X$ is a $\delta$--hyperbolic space, and that $G$ acts on $X$

1066: fixing $e\in \partial X$.  Let $\mathbf{x}=\{x_i\}$ be any sequence tending to

1067: $e$.  The function $q_{\mathbf{x}}\co G\to \R$ defined by

1068: $q_{\mathbf{x}}(g)=\eta_{\mathbf{x}}(g x_0)$ is a quasicharacter of

1069: defect at most $16\delta$.

1070: \end{corollary}

1071: \begin{proof}

1072: Let $g$, $h\in G$.  Using Proposition \ref{p:isom},

1073: \begin{eqnarray*}

1074: |q_{\mathbf{x}}(gh)-q_{\mathbf{x}}(g)-q_{\mathbf{x}}(h)| & = &

1075:  |\eta_{\mathbf{x}}(ghx_0)-\eta_{\mathbf{x}}(gx_0)-\eta_{\mathbf{x}}(hx_0)| \\

1076: & \leq & |

1077:  \eta_{\mathbf{x}}(hx_0)+\eta_{\mathbf{x}}(gx_0)-\eta_{\mathbf{x}}(gx_0)-\eta_{\mathbf{x}}(hx_0)|+16\delta = 16\delta.

1078: \end{eqnarray*}

1079: \end{proof}

1080: 

1081: \begin{proposition}\label{p:pseudo}

1082: Let $X$, $e$, $\mathbf{x}$, and $q_{\mathbf{x}}$ be as in the

1083: statement of Corollary \ref{c:quasimorphism}, and let the

1084: pseudocharacter $p_{\mathbf{x}}\co G\to \R$

1085: be given by

1086: \[p_{\mathbf{x}}(g) = \lim_{n\to\infty}\frac{q_{\mathbf{x}}(g^n)}{n}.\]

1087: Then $p_{\mathbf{x}}(g)\neq 0$ if and only if $g$ acts hyperbolically on $X$.

1088: \end{proposition}

1089: 

1090: \begin{proof}

1091: First, we suppose that $p_{\mathbf{x}}(g)\neq 0$.  Without loss of generality we

1092: assume that $p_{\mathbf{x}}(g)>0$.  Since

1093: \[p_{\mathbf{x}}(g)=\lim_{n\to\infty}\frac{\eta_{\mathbf{x}}(g^n x_0)}{n}>0,\]

1094: there exists some $N$ so that

1095: $\eta_{\mathbf{x}}(g^n x_0)>\frac{1}{2}p_{\mathbf{x}}(g)n$ for all

1096: $n\geq N$.

1097: 

1098: Let $a$ and $b$

1099: be integers.  Choosing some sufficiently large $M$ and applying the triangle

1100: inequality and Lemma \ref{l:approxbyterm}, we obtain a lower bound for

1101: $d(g^a x_0,g^b x_0)$:

1102: \begin{eqnarray*}

1103: d(g^a x_0,g^b x_0) & = & d(g^N x_0,g^{N+|b-a|}x_0)\\

1104:  & \geq & d(g^{N+|b-a|} x_0, x_M)-d(g^N x_0,x_M)\\

1105:  & \geq & \eta_{\mathbf{x}}(g^{N+|b-a|})-\eta_{\mathbf{x}}(g^N) -

1106:  8\delta\\

1107:  & \geq & \frac{1}{2}p_{\mathbf{x}}(g)(N+|b-a|)-\eta_{\mathbf{x}}(g^N)-8\delta\\

1108:  & = & \frac{1}{2}p_{\mathbf{x}}(g) |b-a| - (\eta_{\mathbf{x}}g^N+8\delta-\frac{1}{2}p_{\mathbf{x}}(g)N).

1109: \end{eqnarray*}

1110: On the other hand, $d(g^a x_0,g^b x_0)\leq |b-a| d(x_0,gx_0)$, so the map 

1111: $n\mapsto g^n(x_0)$ is a quasi-isometric embedding, and $g$ acts

1112: hyperbolically.

1113: 

1114: Conversely, suppose that $g$ acts hyperbolically.  

1115: It

1116: follows that there is some $\epsilon>0$ so that

1117: \[d(g^n x_0,x_0)>\epsilon n\] for all $n$.

1118: By

1119: replacing $g$ with $g^{-1}$, we may suppose that $\{g^i x_0\}$ tends

1120: to a point in $\partial X\smallsetminus\{e\}$ as $i\to\infty$.

1121: Thus there is some $R$ so that 

1122: \[\gp{g^n x_0}{x_i}{{x_0}}<R\] for all positive $n$ and $i$.

1123: Lemma \ref{l:approxbyterm} implies that for sufficiently

1124: large $m$,

1125: \begin{eqnarray*}

1126: \eta_{\mathbf{x}}(g^n x_0) & \geq & d(g^n x_0,x_m)-d(x_0,x_m)-4\delta\\

1127: & = & d(g^n x_0,x_0)-2 \gp{g^nx_0}{x_n}{{x_0}} - 4\delta \\

1128: & \geq & \epsilon n - (2 R +4\delta).

1129: \end{eqnarray*}

1130: Since $q_{\mathbf{x}}(g^n)\geq \epsilon n - (2R+4\delta)$ for all $n>0$, we must

1131: have $p(g)\geq \epsilon>0$.

1132: \end{proof}

1133: 

1134: \begin{remark}

1135: Proposition \ref{p:pseudo} was proved in \cite[Proposition 3.9]{manning:qfa} for the case of

1136: quasi-trees.  The proof here is somewhat more efficient even in this

1137: case.

1138: \end{remark}

1139: 

1140: \subsection{Characterization of $G$--spaces with invariant horoballs}

1141: \begin{theorem}\label{t:horeq}

1142: Let $X$ be a Gromov hyperbolic $G$--space.  The following are

1143: equivalent:

1144: \begin{enumerate}

1145: \item\label{ih} $X$ has an invariant horoball.

1146: \item\label{enh} $X$ is elementary, and no element acts

1147:   hyperbolically.

1148: \item\label{onepoint} $X$ is equivalent to a hyperbolic $G$--space $Y$, so that

1149:   $\#(\partial Y)\leq 1$.

1150: \end{enumerate}

1151: \end{theorem}

1152: \begin{proof}

1153: That \eqref{ih} implies \eqref{onepoint} is trivial.

1154: 

1155: We next assume \eqref{onepoint} and show \eqref{enh}.  If \eqref{enh}

1156: holds for $Y$, it holds for $X$, so we may suppose that $X=Y$.  If

1157: $\#(\partial X) = 1$ or $G$ has a bounded orbit in $X$, then clearly

1158: $X$ is elementary.  The only case remaining is that $\partial X$ is

1159: empty, but $Gx$ is unbounded for some $x\in X$.  We show that this

1160: case does not occur.  Chose a sequence

1161: $\{g_i\}$ in $G$ so that $\lim_{i\to\infty}d(g_i x,x) = \infty$.

1162: Since $\partial X$ is empty,

1163: $\liminf_{i,j\to\infty}(g_ix,g_jx)_x<\infty$.  It follows that there

1164: are elements $g_m$, $g_n$ so that $d(g_mx,x)$ and $d(g_nx,x)$ are much

1165: larger than $(g_mx,h_nx)_x$.  It can then be shown (see, for

1166: example, \cite[Chapitre 9, Lemme 2.3]{cdp}) that $g_mg_n$ is hyperbolic.

1167: It follows that $\partial X$ contains at least two points (the fixed

1168: points of $g_mg_n$), contrary to assumption.

1169: 

1170: It remains to show that \eqref{enh} implies \eqref{ih}.

1171: Let $X$ be a hyperbolic $G$--space so that the action of $G$ is

1172: elementary.  If $X$ is equivalent to a point (i.e. if $Gx$ is bounded

1173: for $x\in X$), then $X$ is also

1174: equivalent to a ray, which is the combinatorial horoball based on a

1175: point.  We therefore may assume that $Gx$ is unbounded for any $x\in

1176: X$.  We will construct a combinatorial horoball which coarsely

1177: equivariantly quasi-isometrically embeds in $X$.

1178: Let $\delta>0$ be some number so

1179: that $X$ is $\delta$--hyperbolic.

1180: Let $e$, $\mathbf{x}$,

1181: $q_{\mathbf{x}}$ and $p_{\mathbf{x}}$ be as in the statements of

1182: Corollary \ref{c:quasimorphism} and Proposition \ref{p:pseudo}.

1183: 

1184: To build the combinatorial horoball, we first must start with a graph $Y$

1185: on which $G$ acts.  Choose a finite generating set $S$ for $G$, and

1186: let $C_0 = \mathrm{diam}(S x_0)$. 

1187: Let $V(Y)=G$, and connect $g$ to $h$ in $Y$ if $d(g x_0, h x_0)\leq C_0$.  It

1188: is clear that $G$ acts on $Y$; in fact, $Y$ is a certain Cayley graph

1189: for $G$.  We next define the functions $B_n\co V(Y)\to 2^{V(Y)}$.

1190: Let $C_1 = 2C_0+20\delta$, 

1191: and let

1192: \begin{equation}\label{bstar} 

1193: B_n(g) = \{ h\in G\mid d(hx_0,gx_0)\leq (2n+1)C_1\}.

1194: \end{equation}

1195: \begin{claim}

1196: The sequence of functions $B_*$ in \eqref{bstar} is admissible in the sense

1197: of Definition \ref{d:admissible}.

1198: \end{claim}

1199: \begin{proof}

1200: The only axiom which is not obvious is \eqref{ax:exp}.  We must show

1201: that if $a$ and $b$ are in $B_n(v)$, then $a\in B_{n+1}(b)$ (or

1202: equivalently $b\in B_{n+1}(a)$).  

1203: Put another way,

1204: we must show that if $d(ax_0,vx_0)$ and $d(bx_0,vx_0)$

1205: are bounded above by $(2n+1)C_1$ then $d(a x_0,b x_0)\leq (2n+3)C_1$.

1206: 

1207: Because no element of $G$ acts hyperbolically, the pseudocharacter

1208: $p_{\mathbf{x}}$ is identically zero.  An easy argument shows that

1209: $|q_{\mathbf{x}}(g)|\leq 16\delta$ for all $g\in G$.

1210: Using Lemma \ref{l:approxbyterm}, we can choose some large $n$ so that 

1211: \[| \eta_{\mathbf{x}}(z)-(d(z,x_n)-d(x_0,x_n))| \leq 4\delta\]

1212: for $z\in \{ax_0,bx_0,vx_0\}$.  It follows that 

1213: \begin{equation}\label{diff}

1214: |d(z_1,x_m)-d(z_2,x_m)|\leq 16\delta+4\delta\leq C_1

1215: \end{equation} for $z_1$,

1216: $z_2\in\{ax_0,bx_0,vx_0\}$.  The assertion to be proved is symmetric

1217: in $a$ and $b$, so we may assume

1218: that $\gp{x_m}{b}{v}\leq \gp{x_m}{a}{v}$.  

1219: We deduce:

1220: \begin{eqnarray*}

1221: d(a,b) & \leq & \gp{v}{x_m}{a}+[\gp{a}{x_m}{v}-\gp{b}{x_m}{v}]+\gp{v}{x_m}{b}+2\delta\\

1222: & = & d(v,a) +(d(b,x_m)-d(v,x_m))+2\delta \\

1223: & \leq & (2n+1)C_1 + C_1 + 2\delta \leq (2n+3)C_1.

1224: \end{eqnarray*}

1225: The first line follows from examining the comparison tripods

1226: for the triangles $\Delta(a,v,x_m)$ and $\Delta(b,v,x_m)$;

1227: the last follows from \eqref{diff}.

1228: \end{proof}

1229: 

1230: Since $B_*$ is admissible, the combinatorial horoball

1231: $H=\mathcal{H}_{B_*}(Y)$ is a hyperbolic $G$--space.  It remains

1232: to construct a coarsely equivariant quasi-isometric embedding from $H$

1233: to $X$.  It 

1234: suffices to define this map on the vertices of $H$.

1235: For each $g\in G = V(Y)$ and each $n\in\N$, choose some

1236: $i(g,n)$ so that $\gp{x_k}{x_l}{{gx_0}}\geq 2nC_1$ for all $k$, $l\geq

1237: i(g,n)$.  Choose also some unit speed geodesic $\gamma_{g,n}$ starting

1238: at $g x_0$ and ending at $x_{i(g,n)}$.  Any vertex of $H$ is a pair

1239: $(g,n)$ where $n\in \N$ and $g\in G$.  We define a map $\phi\co

1240: V(H)\to X$ by

1241: \begin{equation*}

1242: \phi(g,n) = \gamma_{g,n}(nC_1).

1243: \end{equation*}

1244: A number of choices were made in the definition of $\phi$ (namely, the

1245: sequence $\mathbf{x}$, the numbers $i(g,n)$, and the geodesics $\gamma_{g,n}$).

1246: However, so long as $x_0$ is unchanged, different

1247: choices lead to a function which differs by at most $\delta$

1248: from $\phi$.  In particular, we could replace $\mathbf{x}$ by

1249: $\mathbf{x}'=\{x_i'\}$, where $x_0'=x_0$ and $x_i'=hx_i$ for some

1250: fixed $h$ and for all $i\geq 1$.

1251: It follows that the distance between $\phi(hg,n)$ and $h\phi(g,n)$ is

1252: at most $\delta$ for any $h$, $g\in G$ and $n\in \N$, and so the map

1253: $\phi$ is coarsely equivariant.

1254: 

1255: 

1256: It remains to show that $\phi$ is a quasi-isometric

1257: embedding.

1258: 

1259: Note that if $v$ and $w$ are two vertices in $H$ connected by a

1260: vertical path, then 

1261: $C_1d_H(v,w)-\delta\leq d(\phi(v),\phi(w))\leq C_1d_H(v,w)+\delta$,

1262: where $d_H$ is the distance in $H$. 

1263: 

1264: We therefore assume that $v=(a,n)$ and $w=(b,m)$, where $a\neq b$.

1265: There is a unique $k$ so that $(2k-1)C_1<d(a,b)\leq (2k+1)C_1$.  If

1266: $\max\{m,n\}\geq k$, then $d(v,w)=|m-n|+1$; otherwise

1267: $d(v,w)=2k-(m+n)+\frac{1}{2}\pm\frac{1}{2}$.  Let 

1268: $I \geq \max\{i(a,n),i(b,m)\}$, and observe that the points $\phi(v)$

1269: and $\phi(w)$ lie within $\delta$ of geodesics joining $ax_0$ to $x_I$

1270: and $bx_0$ to $x_I$, respectively (see Figure \ref{f:deep}).  

1271: \begin{figure}[htbp]

1272: \begin{center}

1273: \input{deep.pstex_t}

1274: \caption{A pair of points in the image of $\phi$.}

1275: \label{f:deep}

1276: \end{center}

1277: \end{figure}

1278: Note that $\gp{x_I}{ax_0}{{bx_0}}$

1279: and $\gp{x_I}{bx_0}{{ax_0}}$ differ from $\frac{1}{2}d(ax_0,bx_0)$ by at

1280: most $\frac{C_1}{2}$.

1281: There are a couple of cases to consider.

1282: 

1283: First, assume that one or both of $n$ and $m$ is at least $k$.

1284: Without loss of generality assume that $n\geq k$.  Since 

1285: $d(ax_0,bx_0)\leq (2k+1)C_1$, we have $(x_I,bx_0)_a\leq kC_1+C_1$. 

1286: Since $d(\phi(v),ax_0)\geq k C_1$,

1287: it follows that $\phi(v)$ is at most $C_1+\delta$ from the

1288: geodesic joining $bx_0$ to $x_I$. Accounting for the possible

1289: difference between $\gp{x_I}{ax_0}{{bx_0}}$ and $\gp{x_I}{bx_0}{{ax_0}}$, we deduce that the distance between $\phi(v)$ and

1290: $\phi(w)$ differs from $|n-m|C_1=(d(v,w)+\frac{1}{2}\pm\frac{1}{2})C_1$ by at most $2C_1+\delta$.  It

1291: follows that in case one of $n$ or $m$ is at least $k$, we have

1292: \[ C_1 d(v,w)-2C_1+\delta \leq d(\phi(v),\phi(w))\leq

1293: C_1d(v,w)+3C_1+\delta.\]

1294: 

1295: In case both $n$ and $m$ are strictly less than $k$, we may argue as

1296: follows.  Since $\gp{ax_0}{x_I}{{bx_0}}$ and $\gp{bx_0}{x_I}{{ax_0}}$ are both

1297: at least $\frac{1}{2}(d(ax_0,bx_0)-C_1)\geq (k-1)C_1\geq \max\{m,n\}

1298: C_1$, it follows that both $\phi(v)$ and $\phi(w)$ are within $2\delta$

1299: of the geodesic joining $ax_0$ to $bx_0$.  From this it follows that

1300: $d(\phi(v),\phi(w))$ differs from $d(ax_0,bx_0)-(n+m)C_1$ by at most

1301: $4\delta$.  But since $d(ax_0,bx_0)$ differs by at most $C_1$ from $k

1302: C_1$, we deduce that 

1303: \[(2k-(m+n))C_1 -(C_1+4\delta) \leq d(\phi(v),\phi(w))\leq

1304: (2k-(m+n))C_1 + (C_1+4\delta),\]

1305: from which it immediately follows that

1306: \[C_1 d(v,w) - (2C_1+4\delta)\leq d(\phi(v),\phi(w))\leq C_1 d(v,w)

1307: +(2C_1+4\delta).\] 

1308: 

1309: In particular, $\phi$ is a $(C_1,2C_1+4\delta)$--quasi-isometric

1310: embedding from the combinatorial horoball $H$ into $X$, and the

1311: theorem is established.

1312: \end{proof}

1313: 

1314: 

1315: 

1316: \section{Rigidity in rank $\geq 2$}\label{s:bigrank}

1317: The purpose of

1318: this section is to establish Theorem \ref{thm:cdr2}, but we will begin

1319: with some lemmas which hold in a slightly broader context.

1320: We suppose that $G$ is a simple Chevalley-Demazure

1321: group scheme of rank at least $2$, that $\Phi$ is a root system for

1322: $G$, and that $R$ is some commutative unital ring containing $\Z$.

1323: 

1324: \begin{lemma}\label{l:phiprime}

1325: If $\alpha$ and $\beta\in \Phi$,

1326: then there is a $\Phi' = \mathrm{Span}(\Phi')\cap\Phi\subseteq \Phi$

1327: containing $\alpha$ and $\beta$ so that

1328: $\Phi'$ is 

1329: isomorphic to $A_1\times A_1$, $A_2$, $B_2$ or $G_2$.  If $\Phi'\cong

1330: A_1\times A_1$, then $\alpha\neq -\beta$.

1331: \end{lemma}

1332: \begin{proof}

1333: If $\alpha$ and $\beta$ are linearly independent, then

1334: $\Phi'=\mathrm{Span}(\{\alpha,\beta\})\cap\Phi$ is a root system of

1335: rank two, and we simply recall that such a root system is always

1336: isomorphic to one of those listed.

1337: 

1338: If $\beta=-\alpha$, we assert that there must be some $\gamma\in \Phi$

1339: so that 

1340: $\mathrm{Span}(\{\alpha,\gamma\})\cap\Phi$ is not equal to $A_1\times

1341: A_1$. 

1342: Suppose that there is no such $\gamma$.  Then either 

1343: $\Phi=\overline{\Phi}\times\langle\alpha\rangle$, or

1344: $\Phi=\langle\alpha\rangle$.

1345: Because $G$ is simple $\Phi$ cannot split as a product; because $G$

1346: has rank at least two, $\Phi\neq A_1$.

1347: \end{proof}

1348: For the following two lemmas, we refer to \cite{carter:book}.  Although

1349: the proofs there are done assuming that $R$ is a field, this

1350: assumption is unnecessary;  see also \cite{stein71} or

1351: \cite{steinberg:notes}.

1352: \begin{lemma}\label{l:steinberg}{\em (Steinberg commutator

1353:     relations)}\cite[Theorem 5.2.2]{carter:book}

1354: If $\alpha$, $\beta\in \Phi$ and $t$, $u\in R$, then 

1355: \[[x_\alpha(t),x_\beta(u)]=\prod_{i,j>0\mbox{, }i\alpha+j\beta\in \Phi }

1356: x_{i\alpha+j\beta}(N_{\alpha,\beta,i,j}t_iu_j), \]

1357: where the $N_{\alpha,\beta,i,j}\in \Z$ are integers which depend only

1358: on the order in which the product is taken.

1359: \end{lemma}

1360: \begin{lemma}\label{l:weylconj}\cite[Lemma 7.2.1]{carter:book}

1361: If $\alpha\in \Phi$, $w$ is an element of the Weyl group of $\Phi$,

1362: and $t\in R$,

1363: then 

1364: $x_{w(\alpha)}(t)$ is conjugate either to  $x_\alpha(t)$

1365: or $x_\alpha(-t)$.

1366: \end{lemma}

1367: 

1368: \begin{lemma}\label{l:distortion}

1369: Let $X$ be a hyperbolic $G(R)$--space.

1370: If $g = x_\alpha(t)$ for some $\alpha\in \Phi$ and $t\in R$, then $g$

1371: does not act hyperbolically on $X$.

1372: \end{lemma}

1373: \begin{proof}

1374: By Lemma \ref{l:phiprime}, there is a subset $\Phi'$ of $\Phi$

1375: containing $\alpha$ which is either isomorphic to $A_2$, $B_2$ or

1376: $G_2$.  In each case, we may apply Lemma \ref{l:steinberg} some number

1377: of times to show that $g$ is distorted in $G$; the details of this are

1378: left to the reader.  By Lemma

1379: \ref{lemma:distortion}, $g$ cannot act hyperbolically on $X$.

1380: \end{proof}

1381: 

1382: \begin{proposition}\label{p:allfixone}

1383: Let $X$ be a hyperbolic $G(R)$--space,

1384: let $r$, $s\in \Phi$, and let $\rho_1$, $\rho_2\in R$.

1385: Suppose that $p=x_r(\rho_1)$ acts parabolically, fixing some $e\in

1386: \partial X$.  Then $ge=e$ for any other root element $g=x_s(\rho_2)$.

1387: \end{proposition}

1388: \begin{proof}

1389: If $r=s$ or if

1390: $\langle r,s\rangle = A_1\times A_1$, then

1391: $p$ and $g$ commute.  By Lemma \ref{l:commute}, $g(e)=e$, and we are

1392: done.

1393: 

1394: Otherwise $r$ and $s$ are contained in a two-dimensional root system

1395: $\Phi'\subset \Phi$ which is isomorphic to

1396: $A_2$, $B_2$ or $G_2$, by Lemma \ref{l:phiprime}.

1397: 

1398: Each case requires a separate argument.  

1399: 

1400: \begin{case}\label{c:a2}

1401: $\Phi'\cong A_2$.

1402: \end{case}

1403: $A_2 = \{\lambda_i \mid i\in \Z_6\}$ contains six roots, arranged

1404: hexagonally; the angle between $\lambda_i$ and 

1405: $\lambda_j$ is $\frac{|i-j|}{3}\pi$.  Suppose $r=\lambda_i$ and $s=\lambda_j$.

1406: By Lemma \ref{l:weylconj}, $h = x_s(\rho_1)$ is conjugate to either

1407: $p$ or $p^{-1}$, so $h$ is a parabolic, fixing some point

1408: $f\in\partial X$.  

1409: In case $|i-j|=1$, then Lemma \ref{l:steinberg} implies that

1410: $h$ and $p$ commute, and so $f=e$ by Lemma \ref{l:commute}.

1411: In case $|i-j|>1$, one argues by induction on $|i-j|$ to the same

1412: conclusion: $f=e$.

1413: 

1414: Since $g$ commutes with $h$, we must have $ge=e$, again by Lemma

1415: \ref{l:commute}.

1416: 

1417: \begin{case}\label{c:b2}

1418: $\Phi'\cong B_2$.

1419: \end{case}

1420: 

1421: Let $\alpha$ be a short root, and $\beta$ a long root, so that

1422: $\alpha$ and $\beta$ span $\Phi'\subset \Phi$, as in

1423: Figure \ref{f:b_2}.

1424: \begin{figure}[htbp]

1425: \begin{center}

1426: \begin{picture}(0,0)%

1427: \includegraphics{b_2.pstex}%

1428: \end{picture}%

1429: \setlength{\unitlength}{4144sp}%

1430: %

1431: \begingroup\makeatletter\ifx\SetFigFont\undefined%

1432: \gdef\SetFigFont#1#2#3#4#5{%

1433:   \reset@font\fontsize{#1}{#2pt}%

1434:   \fontfamily{#3}\fontseries{#4}\fontshape{#5}%

1435:   \selectfont}%

1436: \fi\endgroup%

1437: \begin{picture}(4352,2006)(867,-3021)

1438: \put(4972,-2047){\makebox(0,0)[lb]{\smash{{\SetFigFont{6}{7.2}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\alpha$}%

1439: }}}}

1440: \put(3455,-1448){\makebox(0,0)[lb]{\smash{{\SetFigFont{6}{7.2}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\beta$}%

1441: }}}}

1442: \put(2546,-2026){\makebox(0,0)[lb]{\smash{{\SetFigFont{6}{7.2}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\alpha$}%

1443: }}}}

1444: \put(882,-1238){\makebox(0,0)[lb]{\smash{{\SetFigFont{6}{7.2}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\beta$}%

1445: }}}}

1446: \end{picture}%

1447: \caption{$B_2$ and $G_2$.}

1448: \label{f:b_2}

1449: \end{center}

1450: \end{figure}

1451: 

1452: There are two subcases, depending on whether $r$ is a short or long

1453: root.

1454: \begin{subcase}

1455: The parabolic $p=x_r(\rho_1)$, where $r$ is a short root of $\Phi'$.  

1456: \end{subcase}

1457: Without loss of generality,

1458: we may assume that $r=\alpha$.  If $s\in

1459: \{\alpha,2\alpha+\beta,-\beta\}$, then $g=x_s(\rho_2)$ commutes with

1460: $p$, by Lemma \ref{l:steinberg}. 

1461: Thus by Lemma \ref{l:commute} $ge=e$.

1462: 

1463: Suppose next that $s = \pm (\alpha+\beta)$.  Then Lemma

1464: \ref{l:steinberg} implies

1465: \begin{equation}\label{c1}

1466: p g p^{-1} g^{-1} = x_{r+s}(N \rho_1\rho_2),

1467: \end{equation}

1468: where $N=N_{r,s,1,1}$ is an integer.  Since $r+s\in

1469: \{2\alpha+\beta,-\beta\}$, we already know that 

1470: $h:=x_{r+s}(N\rho_1\rho_2)$ fixes $e$.  We rearrange \eqref{c1} to give

1471: \begin{equation}

1472:  h^{-1} p = gpg^{-1}.

1473: \end{equation}

1474: Since $p$ and $h$ both fix $e$, so must $gpg^{-1}$.  The element

1475: $gpg^{-1}$ also fixes $g(e)$, since $p$ fixes $e$.

1476: Since $p$ is

1477: parabolic, it can only fix one point in $\partial X$, and so $g(e)=e$.

1478: 

1479: If $s\in \{\beta, -\alpha\}$, let $r' = \alpha+\beta$; if

1480: $s=-2\alpha-\beta$ let $r' = -\alpha-\beta$.  In any case there is an

1481: element of the Weyl group of $\Phi$ taking $r$ to $r'$; by Lemma

1482: \ref{l:weylconj}, there is a 

1483: $p'= x_{r'}(\pm \rho_1)$ which is

1484: conjugate to $p$ in $E(\Phi,R)$.  Since $p'$ is conjugate to $p$, it

1485: is parabolic; by the argument of the previous paragraph, $p'$ has the

1486: same fixed point as $p$.  If $s\in \{\beta, -2\alpha-\beta\}$, then

1487: $g$ commutes with $p'$, and so $g(e)=e$ by Lemma \ref{l:commute}.

1488: Finally, if

1489: $s=-\alpha$, then we may apply the argument of the previous paragraph

1490: again (with $p'$ and $r'$ in place of $p$ and $r$), to deduce that $g(e)=e$.

1491: 

1492: \begin{subcase}

1493: The parabolic $p=x_r(\rho_1)$, where $r$ is a long root of $\Phi'$.  

1494: \end{subcase}

1495: In this case, we may assume for instance that $r=2\alpha+\beta$.  If

1496: $s\in\{2\alpha+\beta, \alpha+\beta, \alpha, \beta, -\beta\}$, then

1497: Lemma \ref{l:steinberg} implies that $g=x_s(\rho_2)$ commutes with $p$,

1498: and so $g(e)=e$ by Lemma \ref{l:commute}.

1499: 

1500: Suppose then that $s\in \{-\alpha,-2\alpha-\beta,-\alpha-\beta\}$.

1501: If $s\in\{-\alpha,-2\alpha-\beta\}$, let $r'=\beta$; if $s=

1502: -\alpha-\beta$, then let $r'=-\beta$.  In either case, there is an

1503: element of the Weyl group taking $r$ to $r'$, and so there is a

1504: parabolic $p' = x_{r'}(\pm \rho_1)$ conjugate to $p$ by Lemma

1505: \ref{l:weylconj}.   By the previous paragraph, $p'$ has the same fixed

1506: point as $p$ does.  Applying the previous paragraph with $p'$ and $r'$

1507: in the place of $p$ and $r$ implies that $g(e)=e$ for $g =

1508: x_s(\rho_2)$. This completes the proof of Case \ref{c:b2}.

1509: 

1510: \begin{case}\label{c:g2}

1511: $\Phi'\cong G_2$.

1512: \end{case}

1513: Let $\alpha$ be a short root, and $\beta$ a long root, so that

1514: $\alpha$ and $\beta$ span $\Phi'\subset \Phi$, as in

1515: the right half of Figure \ref{f:b_2}.

1516: 

1517: 

1518: Again there are two subcases, depending on whether $r$ is a short or long

1519: root.

1520: 

1521: \begin{subcase}

1522: The parabolic $p=x_r(\rho_1)$, where $r$ is a long root of $\Phi'$.  

1523: \end{subcase}

1524: If $s\in \{-\beta, 3\alpha+\beta, \pm(3\alpha+2\beta)\}$, then Lemma

1525: \ref{l:steinberg} implies that $g=x_s(\rho_2)$ commutes with $p$, and

1526: so $g(e)=e$ by Lemma \ref{l:commute}.

1527: 

1528: Suppose that $s\in \{2\alpha+\beta,-\alpha-\beta\}$.  Lemma

1529: \ref{l:steinberg} implies that 

1530: \begin{equation}

1531: p g p^{-1} g^{-1} = x_{r+s}(N\rho_1\rho_2)=: h,

1532: \end{equation}

1533: for some integer $N$.  Exactly as in Case \ref{c:b2}, $h$ commutes

1534: with $p$, and so $h(e)=e$.  Thus $e= h^{-1} p (e) = gpg^{-1}(e)$ and

1535: the parabolic $gpg^{-1}$ fixes $e$.  Again since $gpg^{-1}$ also fixes

1536: $g(e)$, we must have $g(e)=e$.

1537: 

1538: Using Lemma \ref{l:weylconj} repeatedly we discover that for every

1539: short root $r'$ there is a parabolic element $p'=x_{r'}(\pm \rho_1)$

1540: with $p'(e)=e$.  Since $g$ must commute with some such element,

1541: $g(e)=e$ as well.

1542: \begin{subcase}

1543: The parabolic $p=x_r(\rho_1)$, where $r$ is a long root of $\Phi'$.  

1544: \end{subcase}

1545: Without loss of generality, we may assume that $r = 3\alpha+2\beta$.

1546: 

1547: If the inner product of $s$ with $r$

1548: is nonnegative, then $s\in\{ \pm \alpha, \beta, \alpha+\beta,

1549: \alpha+2\beta,\alpha+3\beta, 3\alpha+2\beta\}$, and $g$ commutes with

1550: $p$ by Lemma \ref{l:steinberg}, and so $g(e)=e$.  

1551: 

1552: Otherwise, a (possibly

1553: repeated) application of Lemma \ref{l:weylconj} implies that $g$

1554: commutes with a parabolic $p' = x_{r'}(\pm \rho_1)$ for some long root

1555: $r$ of $\Phi'$, and with $p'(e)=e$.  This again implies via Lemma

1556: \ref{l:commute} that $g(e)=e$.  This completes the proof in Case \ref{c:g2}.

1557: \end{proof} 

1558: 

1559: We are now ready to give the proof of Theorem \ref{thm:cdr2}.  

1560: \begin{proof}

1561: A result of Tavgen$'$ \cite{tavgen:bg} shows that $G(\mathcal{O})$ is

1562: boundedly generated by its root subgroups.

1563: The ring of integers $\mathcal{O}$ is finitely generated as an Abelian

1564: group; choose generators $\mu_1,\ldots,\mu_k$.  It follows from

1565: Tavgen$'$'s result that the set

1566: \[ S=\{x_\alpha(\mu_i) \mid \alpha \in \Phi\mbox{, }1\leq i\leq k\}\]

1567: boundedly generates $G(\mathcal{O})$.

1568: 

1569: Each of these generators acts hyperbolically, elliptically or

1570: parabolically on $X$.  

1571: By Lemma \ref{l:distortion}, none can act hyperbolically.

1572: If all of the root elements act elliptically, then it follows from bounded

1573: generation that the orbit of a point under the action of $G$ must be

1574: bounded.

1575: 

1576: We therefore may assume that some $x_\alpha(\mu_i)$ acts parabolically on

1577: $X$, fixing a single point $e\in \partial X$.  It follows from

1578: Proposition \ref{p:allfixone} that all the root subgroups will fix

1579: this point $e$, and so $G(\mathcal{O})$ fixes $e$.

1580: 

1581: By Proposition \ref{p:pseudo}, the pseudocharacter 

1582: $p_\mathbf{x}\co G(\mathcal{O})\to \R$ determined by a sequence

1583: $\mathbf{x}=\{x_i\}$ tending to $e$

1584: is nonzero exactly on the hyperbolic elements.  Thus $p_\mathbf{x}(g)=0$

1585: whenever $g$ lies in a root subgroup.  

1586: 

1587: An elementary argument (see for example \cite[Proposition

1588:   5]{kotschick:quasi}) shows that a pseudocharacter $p$ on a boundedly

1589: generated group 

1590: is determined by its values on

1591: the bounded generators;  thus $p\equiv 0$ on $G(\mathcal{O})$.  It

1592: follows that no element of $G(\mathcal{O})$ acts hyperbolically on

1593: $X$. 

1594: By Theorem

1595: \ref{t:horeq}, the $G(\mathcal{O})$--space $X$ has an invariant horoball.

1596: \end{proof} 

1597: 

1598: 

1599: \section{Remarks on rank one}\label{section:rankone}

1600: One can also ask what actions rank one Chevalley groups have on

1601: hyperbolic spaces.  If $\mc{O}$ is a number ring with finitely many

1602: units, then $SL(2,\mc{O})$ is a lattice either in $SL(2,\R)$ or

1603: $SL(2,\C)$.  In particular, it has a proper non-elementary action on

1604: $\H^2$ or $\H^3$.  Moreover, such a group admits uncountably many

1605: distinct pseudocharacters (AKA homogeneous quasi-(homo)morphisms) up

1606: to scale \cite{fujiwara:gromovhyperbolic,bestvinafujiwara:mcg}.  Each

1607: such ``projective pseudocharacter'' gives rise to a 

1608: quasi-action on $\R$; no two such are equivalent.  Moreover, these

1609: often give rise to quasi-actions on more complicated trees

1610: \cite{manning:cocycles}.

1611: The groups $SL(2,\mc{O})$ where $\mc{O}$ has infinitely many units are

1612: more rigid.

1613: In this section we

1614: apply the main result of \cite{manning:qfa} to the special case of

1615: actions on quasi-trees (defined below), and

1616: speculate on the general situation.

1617: 

1618: Recall that a group $G$ is said to have 

1619: Property \FA\ if every action by $G$ on a simplicial tree $T$ has a

1620: fixed point.

1621: \begin{definition}

1622: A \emph{quasi-tree} is a graph which is quasi-isometric to a tree.

1623: \end{definition}

1624: \begin{definition}

1625: A group $G$ has property \QFA\ if every action by $G$ on a quasi-tree

1626: $X$ has a bounded orbit. 

1627: \end{definition}

1628: \begin{remark}

1629: This is differently worded than the definition in \cite{manning:qfa},

1630: but easily seen to be equivalent.  Note that \QFA\ implies \FA, but

1631: not vice versa.

1632: \end{remark}

1633: As quasi-trees are in particular Gromov hyperbolic spaces which admit

1634: no parabolic isometries (see Section 3.2 of \cite{manning:qfa}),

1635: Theorem \ref{thm:cdr2} implies that higher rank Chevalley

1636: groups over number rings have property \QFA.

1637: 

1638: We recall a definition and a theorem from \cite[Section 4]{manning:qfa}.

1639: 

1640: \begin{definition}

1641: Let $G$ be a group, and let $g$ be an element of $G$.  We will say $g$

1642: is a \emph{stubborn element of $G$} if for all

1643: $H< G$ with $[G:H]\leq 2$, there exists some integer $k_H>0$ so that

1644: $g^{k_H}\in [H,H]$.  

1645: \end{definition}

1646: 

1647: 

1648: \begin{theorem}\label{thm:qfa}\cite[Theorem 4.4]{manning:qfa}

1649: Let $G$ be a group which is boundedly generated by elements 

1650: $g_1,\ldots,g_n$, so

1651: that for each $i$, $g_i$ is a stubborn element of $B_i$ for some

1652: amenable $B_i < G$. Then $G$ has Property \QFA.

1653: \end{theorem}

1654: Note that the above theorem was misstated slightly in

1655: \cite{manning:qfa}; the

1656: word ``amenable'' was inadvertently omitted.

1657: 

1658: Here's an easy lemma:

1659: \begin{lemma}\label{lemma:finiteindex}

1660: Let $H<G$ be a subgroup of finite index.  If $H$ has Property

1661: \QFA, then so does $G$.

1662: \end{lemma}

1663: 

1664: Our methods in the higher rank case use heavily the bounded generation of

1665: $G(\mathcal{O})$ established by Tavgen$'$ in \cite{tavgen:bg}

1666: for Chevalley groups over rings of integers of algebraic

1667: number fields.  There is an analogous result of Carter, Keller, and Paige

1668: in rank $1$, at least for $SL(2,\cdot)$ and certain number

1669: fields:

1670: \begin{theorem}\label{thm:ckp2}

1671: \cite{wittemorris:ckp}  

1672: For any integer $d>1$ there is an $r=r(d)$ so that the following is true.

1673: Let $K$ be a number field of degree $d$ over $\Q$, and let

1674: $\mathcal{O}$ be the ring of integers of $K$.  If the $\mc{O}$ has

1675: infinitely many units, then:

1676: \begin{enumerate}

1677: \item every element of $E(2,\mathcal{O})$ is a product of at most $r$

1678:   elementary matrices, and

1679: \item the index of $E(2,\mathcal{O})$ in $SL(2,\mathcal{O})$ is at

1680:   most $r$.  

1681: \end{enumerate}

1682: \end{theorem}

1683: In the above statement, $E(2,\mathcal{O})$ is the subgroup of

1684: $SL(2,\mathcal{O})$ generated by the root subgroups (the strictly

1685: upper triangular and strictly lower triangular matrices).

1686: The following statement implies Theorem \ref{thm:cdr1}:

1687: \begin{theorem}\label{thm:rankone}

1688: If $\mathcal{O}$ is the ring of integers of an algebraic number field

1689: and $\mathcal{O}$ has infinitely many units, then $SL(2,\mathcal{O})$

1690: has Property \QFA.

1691: \end{theorem}

1692: \begin{proof}

1693: By Lemma \ref{lemma:finiteindex} it suffices to show that

1694: $E(2,\mathcal{O})$ has property \QFA.

1695: If $\Lambda= \{\lambda_1,\ldots,\lambda_n\}$ is an integral basis

1696: for the number field $\mathcal{O}$, then

1697: Theorem \ref{thm:ckp2} implies that $E(2,\mathcal{O})$ is boundedly

1698: generated by the $2n$ elements $\textMtwo{1}{\lambda_i}{0}{1}$ and

1699: $\textMtwo{1}{0}{\lambda_i}{1}$.  

1700: \begin{claim}

1701: For each $i\in\{1,\ldots,n\}$,

1702: $\textMtwo{1}{\lambda_i}{0}{1}$ is a stubborn element of 

1703: $B=\textMtwo{*}{*}{0}{*}\cap E(2,\mathcal{O})$.

1704: \end{claim}

1705: \begin{proof}

1706: On page 189 of \cite{carter:book}, Carter observes that

1707: $\textMtwo{t}{0}{0}{t^{-1}}$ can be written as a product of elementary

1708: matrices, for any invertible $t\in \mc{O}$, as follows.  

1709: If

1710: $\lambda$ is any invertible element, we may write

1711: \[\Mtwo{0}{\lambda}{-\lambda^{-1}}{0} =

1712: \Mtwo{1}{\lambda}{0}{1}\Mtwo{1}{0}{-\lambda^{-1}}{1}\Mtwo{1}{\lambda}{0}{1},\]

1713: and then note that

1714: \begin{equation}\label{computation}

1715: \Mtwo{t}{0}{0}{t^{-1}}=\Mtwo{0}{t}{-t^{-1}}{0}\Mtwo{0}{-1}{1}{0}.

1716: \end{equation}

1717: Carter's observation shows that $\textMtwo{\omega}{0}{0}{\omega^{-1}}$ is in

1718: $E(2,\mc{O})$ for any unit $\omega$ in $\mc{O}$.  

1719: Computing the commutator of \textMtwo{\omega}{0}{0}{\omega^{-1}} and

1720: \textMtwo{1}{\lambda}{0}{1} for $\omega$ a unit of $R$ and $\lambda\in R$

1721: yields:

1722: \begin{equation}\label{computation2}

1723: \left[\Mtwo{\omega}{0}{0}{\omega^{-1}},\Mtwo{1}{\lambda}{0}{1}\right]

1724:  = \Mtwo{1}{(1-\omega^2)\lambda}{0}{1}.

1725: \end{equation}

1726: 

1727: 

1728: By assumption, the group of units of $\mc{O}$ is infinite.

1729: Dirichlet's units theorem (see, e.g. \cite[Appendix B]{stewarttall})

1730: implies that we may

1731: choose $\omega_0\in \mc{O}^*$ a unit of infinite order. 

1732: Let $H<B$ be a subgroup of index at most two.  Then $H$ must contain

1733: $\textMtwo{\omega_0^2}{0}{0}{\omega_0^{-2}}$ and

1734: $\textMtwo{1}{2 r}{0}{1}$ for all $r\in \mc{O}$.  It follows from the

1735: computations \eqref{computation} and \eqref{computation2}

1736: that $\textMtwo{1}{i}{0}{1}\in [H,H]$ for all

1737: $i$ in the ideal $I$ generated by $2(1-\omega_0^2)$.  Let $N$ be the

1738: order of $R/I$. (The number $N$ is also called the \emph{norm} of $I$;

1739: that it is finite when $I\neq (0)$ is an elementary fact of algebraic

1740: number theory;  see, e.g., \cite[Chapter 5]{stewarttall}.)

1741: For any of the $\lambda_i$, we have

1742: $\textMtwo{1}{\lambda_i}{0}{1}^N=\textMtwo{1}{N\lambda_i}{0}{1}\in [H,H]$, and

1743: so $\textMtwo{1}{\lambda_i}{0}{1}$ is stubborn.

1744: \end{proof}

1745: It remains to observe that $B<E(2,\mathcal{O})$ is solvable, and hence

1746: amenable.

1747: We may now apply Theorem \ref{thm:qfa} to conclude that

1748: $E(2,\mathcal{O})$ has Property \QFA.

1749: \end{proof}

1750: \begin{remark}

1751: It was already known \cite[p. 68]{serre:trees} that the groups covered

1752: by Theorem \ref{thm:rankone} possessed property \FA.

1753: \end{remark}

1754: \begin{remark}

1755: Another proof of \ref{thm:rankone} may be given as follows:  First

1756: show that every unipotent is distorted.  It follows that the bounded

1757: generators cannot (quasi)-act hyperbolically.  It is shown in

1758: \cite[Corollary 3.6]{manning:qfa} that there are no parabolic isometries of

1759: quasi-trees, and so each of the bounded generators (quasi)-acts

1760: elliptically.  It then follows from bounded generation that any orbit

1761: is bounded.

1762: \end{remark}

1763: 

1764: Finally, we speculate on the variety of hyperbolic $\Gamma$-spaces, for

1765: $\Gamma=SL(2,\mathcal{O})$, where $\mathcal{O}$ is the

1766: ring of integers of a number field $k$.  

1767: We have already remarked that $\Gamma$ is a lattice in 

1768: \[\prod_{i=1}^s SL(2,\R)\times \prod_{i=1}^t SL(2,\C),\]

1769: where $s$ and $t$ are the number of real and complex places

1770: respectively.  Projection to some factor gives an isometric action

1771: either on $\H^2$ or $\H^3$.  Call a hyperbolic $\Gamma$--space

1772: \emph{standard} if it is equivalent to $\H^2$ or $\H^3$ with one of

1773: these actions.

1774: 

1775: \begin{conjecture}\label{conjecture:realquadratic}

1776: Every quasi-action by $\Gamma$ on a Gromov hyperbolic metric space

1777: either has an invariant horoball or is standard.

1778: \end{conjecture}

1779: 

1780: %

1781: %

1782: %

1783: %\bibliographystyle{gtart}

1784: %

1785: 

1786: %\bibliographystyle{abbrv}

1787: %\small

1788: %\bibliography{../../mybib.bib}

1789: \def\cprime{$'$} \providecommand\url[1]{\texttt{#1}}

1790: \begin{thebibliography}{10}

1791: 

1792: \bibitem{abe:chevalley}

1793: E.~Abe.

1794: \newblock Chevalley groups over local rings.

1795: \newblock {\em T\^ohoku Math. J. (2)}, 21:474--494, 1969.

1796: 

1797: \bibitem{bestvinafujiwara:mcg}

1798: M.~Bestvina and K.~Fujiwara.

1799: \newblock Bounded cohomology of subgroups of mapping class groups.

1800: \newblock {\em Geometry and Topology}, 6:69--89, 2002.

1801: 

1802: \bibitem{bridhaef:book}

1803: M.~R. Bridson and A.~Haefliger.

1804: \newblock {\em Metric Spaces of Non--Positive Curvature}, volume 319 of {\em

1805:   {G}rundlehren der mathematischen {W}issenschaften}.

1806: \newblock Springer--Verlag, Berlin, 1999.

1807: 

1808: \bibitem{burgermonod:gafa}

1809: M.~Burger and N.~Monod.

1810: \newblock Continuous bounded cohomology and applications to rigidity theory.

1811: \newblock {\em Geom. Funct. Anal.}, 12(2):219--280, 2002.

1812: 

1813: \bibitem{carterkeller:slno}

1814: D.~Carter and G.~Keller.

1815: \newblock Bounded elementary generation of {${\rm SL}\sb{n}({\mathcal O})$}.

1816: \newblock {\em Amer. J. Math.}, 105(3):673--687, 1983.

1817: 

1818: \bibitem{carter:book}

1819: R.~W. Carter.

1820: \newblock {\em Simple groups of {L}ie type}.

1821: \newblock John Wiley \& Sons, London-New York-Sydney, 1972.

1822: \newblock Pure and Applied Mathematics, Vol. 28.

1823: 

1824: \bibitem{chevalley:tohoku}

1825: C.~Chevalley.

1826: \newblock Sur certains groupes simples.

1827: \newblock {\em T\^ohoku Math. J. (2)}, 7:14--66, 1955.

1828: 

1829: \bibitem{cdp}

1830: M.~Coornaert, T.~Delzant, and A.~Papadopoulos.

1831: \newblock {\em G\'eom\'etrie et th\'eorie des groupes: Les groupes

1832:   hyperboliques de Gromov}, volume 1441 of {\em Lecture Notes in Mathematics}.

1833: \newblock Springer-Verlag, Berlin, 1990.

1834: 

1835: \bibitem{fujiwara:gromovhyperbolic}

1836: K.~Fujiwara.

1837: \newblock The second bounded cohomology of a group acting on a

1838:   {G}romov-hyperbolic space.

1839: \newblock {\em Proc. London Math. Soc. (3)}, 76(1):70--94, 1998.

1840: 

1841: \bibitem{fukunaga:FA}

1842: M.~Fukunaga.

1843: \newblock Fixed points of elementary subgroups of {C}hevalley groups acting on

1844:   trees.

1845: \newblock {\em Tsukuba J. Math.}, 3(2):7--16, 1979.

1846: 

1847: \bibitem{gelanderkarlssonmargulis}

1848: T.~Gelander, A.~Karlsson, and G.~A. Margulis.

1849: \newblock Superrigidity, generalized harmonic maps and uniformly convex spaces.

1850: \newblock {\em Geom. Funct. Anal.}, 17(5):1524--1550, 2008.

1851: 

1852: \bibitem{gromov:wordhyperbolic}

1853: M.~Gromov.

1854: \newblock Word hyperbolic groups.

1855: \newblock In S.~M. Gersten, editor, {\em Essays in Group Theory}, volume~8 of

1856:   {\em Mathematical Sciences Research Institute Publications}, pages 75--264.

1857:   Springer--Verlag, New York, 1987.

1858: 

1859: \bibitem{rhds}

1860: D.~Groves and J.~F. Manning.

1861: \newblock {Dehn filling in relatively hyperbolic groups}.

1862: \newblock {\em Israel Journal of Mathematics}.

1863: \newblock to appear, preprint at \url{arXiv:math/0601311v3}.

1864: 

1865: \bibitem{humphreys:introduction}

1866: J.~E. Humphreys.

1867: \newblock {\em Introduction to {L}ie algebras and representation theory},

1868:   volume~9 of {\em Graduate Texts in Mathematics}.

1869: \newblock Springer-Verlag, New York, 1978.

1870: \newblock Second printing, revised.

1871: 

1872: \bibitem{karlssonnoskov}

1873: A.~Karlsson and G.~A. Noskov.

1874: \newblock Some groups having only elementary actions on metric spaces with

1875:   hyperbolic boundaries.

1876: \newblock {\em Geom. Dedicata}, 104:119--137, 2004.

1877: 

1878: \bibitem{kotschick:quasi}

1879: D.~Kotschick.

1880: \newblock Quasi-homomorphisms and stable lengths in mapping class groups.

1881: \newblock {\em Proc. Amer. Math. Soc.}, 132(11):3167--3175 (electronic), 2004.

1882: 

1883: \bibitem{manning:cocycles}

1884: J.~F. Manning.

1885: \newblock {Geometry of pseudocharacters}.

1886: \newblock {\em Geometry and Topology}, 9:1147--1185, 2005.

1887: \newblock \url{arXiv:math.GR/0303380}.

1888: 

1889: \bibitem{manning:qfa}

1890: J.~F. Manning.

1891: \newblock {Quasi-actions on trees and Property QFA}.

1892: \newblock {\em Journal of the London Mathematical Society}, 73(1):84--108,

1893:   2006.

1894: \newblock With an appendix by Nicolas Monod and Bertrand R{\'e}my.

1895: 

1896: \bibitem{matsumoto:csp}

1897: H.~Matsumoto.

1898: \newblock Sur les sous-groupes arithm\'etiques des groupes semi-simples

1899:   d\'eploy\'es.

1900: \newblock {\em Ann. Sci. \'Ecole Norm. Sup. (4)}, 2:1--62, 1969.

1901: 

1902: \bibitem{monod:superrigid}

1903: N.~Monod.

1904: \newblock Superrigidity for irreducible lattices and geometric splitting.

1905: \newblock {\em J. Amer. Math. Soc.}, 19(4):781--814 (electronic), 2006.

1906: 

1907: \bibitem{monodshalom:jdg}

1908: N.~Monod and Y.~Shalom.

1909: \newblock Cocycle superrigidity and bounded cohomology for negatively curved

1910:   spaces.

1911: \newblock {\em J. Differential Geom.}, 67(3):395--455, 2004.

1912: 

1913: \bibitem{wittemorris:ckp}

1914: D.~W. Morris.

1915: \newblock Bounded generation of {${\rm SL}(n,A)$} (after {D}. {C}arter, {G}.

1916:   {K}eller, and {E}. {P}aige).

1917: \newblock {\em New York J. Math.}, 13:383--421 (electronic), 2007.

1918: 

1919: \bibitem{msw:quasiactI}

1920: L.~Mosher, M.~Sageev, and K.~Whyte.

1921: \newblock Quasi-actions on trees. {I}. {B}ounded valence.

1922: \newblock {\em Ann. of Math. (2)}, 158(1):115--164, 2003.

1923: 

1924: \bibitem{serre:trees}

1925: J.-P. Serre.

1926: \newblock {\em Trees}.

1927: \newblock Springer-Verlag, Berlin, 1980.

1928: \newblock Translated from the French by John Stillwell.

1929: 

1930: \bibitem{shalen:handbook}

1931: P.~B. Shalen.

1932: \newblock Representations of 3-manifold groups.

1933: \newblock In {\em Handbook of geometric topology}, pages 955--1044.

1934:   North-Holland, Amsterdam, 2002.

1935: 

1936: \bibitem{stein71}

1937: M.~R. Stein.

1938: \newblock Generators, relations and coverings of {C}hevalley groups over

1939:   commutative rings.

1940: \newblock {\em Amer. J. Math.}, 93:965--1004, 1971.

1941: 

1942: \bibitem{steinberg:notes}

1943: R.~Steinberg.

1944: \newblock {\em Lectures on {C}hevalley groups}.

1945: \newblock Yale University, New Haven, Conn., 1968.

1946: \newblock Notes prepared by John Faulkner and Robert Wilson.

1947: 

1948: \bibitem{stewarttall}

1949: I.~Stewart and D.~Tall.

1950: \newblock {\em Algebraic number theory and {F}ermat's last theorem}.

1951: \newblock A K Peters Ltd., Natick, MA, third edition, 2002.

1952: 

1953: \bibitem{tavgen:bg}

1954: O.~I. Tavgen{\cprime}.

1955: \newblock Bounded generability of {C}hevalley groups over rings of

1956:   {$S$}-integer algebraic numbers.

1957: \newblock {\em Izv. Akad. Nauk SSSR Ser. Mat.}, 54(1):97--122, 221--222, 1990.

1958: 

1959: \end{thebibliography}

1960: 

1961: 

1962: 

1963: \end{document}

1964: