1: %%%%%%%%%%%%LaTeX file%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2: \documentclass[11pt]{amsart}
3: \usepackage{latexsym}
4: \usepackage{amsfonts}
5: %\setlength{\textheight}{20cm}
6: %\setlength{\textwidth}{141mm}
7:
8: \usepackage{amsmath,amsthm,amssymb}
9:
10: \usepackage{epsf,psfrag}
11:
12:
13:
14: \newtheorem{definition}{Definition} \newtheorem{question}{Question}
15: \newtheorem{proposition}{Proposition}[section]
16: \newtheorem{theorem}[proposition]{Theorem}
17: \newtheorem{example}{Example}
18:
19: \newtheorem{theor}{Theorem} \renewcommand{\thetheor}{\Alph{theor}}
20:
21: \newtheorem{corol}[theor]{Corollary}
22: \renewcommand{\thecorol}{\Alph{corol}}
23:
24:
25: \newtheorem{theore}{Theorem} \renewcommand{\thetheore}{\Alph{theore}}
26:
27:
28: \newtheorem{coroll}[theore]{Corollary}
29: \renewcommand{\thecoroll}{\Alph{coroll}}
30:
31: \newtheorem{thm}{Theorem}[section] \newtheorem{lem}[thm]{Lemma}
32: \newtheorem{cor}[thm]{Corollary} \newtheorem{conject}[thm]{Conjecture}
33: \newtheorem{prop}[thm]{Proposition} \theoremstyle{definition}
34: \newtheorem{defn}[thm]{Definition}
35: \newtheorem{notation}[thm]{Notation}
36: \newtheorem{conv}[thm]{Convention} \newtheorem{rem}[thm]{Remark}
37: \newtheorem{exmp}[thm]{Example}
38: \newtheorem{propdfn}[thm]{Proposition-Definition}
39: \newtheorem{lede}[thm]{Lemma-Definition}
40: \newtheorem{warning}[thm]{Warning}
41: \newtheorem{history}[thm]{Historical Note}
42: \newtheorem{openp}[thm]{Open Problem} \newtheorem{prob}[thm]{Problem}
43:
44:
45:
46: \begin{document}
47:
48: \title[Translation equivalence in free groups]{Translation equivalence
49: in free groups}
50:
51: \author[I.~Kapovich]{Ilya Kapovich}
52:
53: \address{\tt Department of Mathematics, University of Illinois at
54: Urbana-Champaign, 1409 West Green Street, Urbana, IL 61801, USA
55: \newline http://www.math.uiuc.edu/\~{}kapovich/} \email{\tt
56: kapovich@math.uiuc.edu}
57:
58: \thanks{The first author acknowledges the support of the Max Planck
59: Institute of Mathematics in Bonn. The first and the third author
60: were supported by the NSF grant DMS\#0404991 and the NSA grant
61: DMA\#H98230-04-1-0115. The fourth author was supported by the NSF
62: grant DMS\#0405105}
63:
64: \author[G.~Levitt]{Gilbert Levitt} \address{\tt Laboratoire de
65: Mathematiques Nicolas Oresme, CNRS UMR 6139, Universite de Caen, BP
66: 5186, 14032 Caen Cedex, France} \email{\tt
67: Gilbert.Levitt@math.unicaen.fr}
68:
69: \author[P.~Schupp]{Paul Schupp}
70:
71: \address{\tt Department of Mathematics, University of Illinois at
72: Urbana-Champaign, 1409 West Green Street, Urbana, IL 61801, USA
73: \newline http://www.math.uiuc.edu/People/schupp.html}
74: \email{schupp@math.uiuc.edu}
75:
76: \author[V.~Shpilrain]{Vladimir Shpilrain} \address{\tt Department of
77: Mathematics, The City College of New York, New York, NY 10031, USA
78: \newline http://www.sci.ccny.cuny.edu/\~{}shpil} \email{\tt
79: shpil@groups.sci.ccny.cuny.edu}
80:
81: \begin{abstract}
82: Motivated by the work of Leininger on hyperbolic equivalence of
83: homotopy classes of closed curves on surfaces, we investigate a
84: similar phenomenon for free groups. Namely, we study the situation
85: when two elements $g,h$ in a free group $F$ have the property that
86: for every free isometric action of $F$ on an $\mathbb{R}$-tree $X$
87: the translation lengths of $g$ and $h$ on $X$ are equal.
88: \end{abstract}
89:
90:
91: \subjclass[2000]{ Primary 20F36, Secondary 20E36, 57M05}
92:
93:
94:
95: \maketitle
96:
97:
98:
99:
100: \section{Introduction}\label{intro}
101:
102:
103: Let $S$ be a closed oriented surface of negative Euler characteristic
104: and let $\gamma$ be a free homotopy class of essential closed curves
105: on $S$. If $\rho$ is a hyperbolic metric on $S$ then $\gamma$ contains
106: a unique curve $c$ of minimal $\rho$-length. We denote this length by
107: $\ell_{\rho}(\gamma)$. The curve $c$ is a closed geodesic on $(S,
108: \rho)$ and $\ell_{\rho}(\gamma)$ is the translation length of any
109: representative of $\gamma$ in the action corresponding to $\rho$ of
110: $G=\pi_1(S)$ on $\mathbb{H}^2 = \tilde{S}$. There is an obvious
111: identification between the set of nontrivial conjugacy classes $C$ of
112: $G$ and the set of free homotopy classes of essential closed curves on
113: $S$ and we shall not distinguish between the two. Thus each marked
114: hyperbolic structure on $S$ defines a so-called \textit{marked length
115: spectrum} $l:C \to \mathbb{R}$. It is well-known and easy to see
116: that a marked hyperbolic structure on $S$, considered as a point of
117: the Teichm\"uller space of $S$, is uniquely determined by its marked
118: length spectrum.
119:
120: The dual situation, however, is different. For $\gamma_1, \gamma_2
121: \in C$ we say that $\gamma_1$ is \emph{hyperbolically equivalent to}
122: $\gamma_2$, denoted $\gamma_1 \equiv_h \gamma_2$, if for every
123: hyperbolic structure $\rho$ on $S$ we have $\ell_{\rho}(\gamma_1)=
124: \ell_{\rho}(\gamma_2)$. In more algebraic terms, for two conjugacy
125: classes $\gamma_1,\gamma_2 \in C$ we have $\gamma_1 \equiv_h \gamma_2$
126: if for every discrete and co-compact isometric action of $G$ on
127: $\mathbb{H}^2$ the translation lengths of $\gamma_1$ and $\gamma_2$
128: are equal. It can happen that $\gamma_1 \ne \gamma_2^{\pm 1}$ and yet
129: $\gamma_1 \equiv_h \gamma_2$. The main source of hyperbolic
130: equivalence comes from ``trace identities" in $SL(2, \mathbb{C})$. A
131: number of interesting new results about hyperbolic equivalence were
132: recently obtained by Chris Leininger~\cite{Le}.
133:
134: We are interested in investigating a similar phenomenon for free
135: groups. In this context the Teichm\"uller space is replaced by the
136: Culler-Vogtmann outer space~\cite{CV}, so that instead of actions on
137: $\mathbb{H}^2$ we consider free and discrete actions on
138: $\mathbb{R}$-trees. Recall that if $G$ is a group acting by isometries
139: on an $\mathbb{R}$-tree $X$ and $g\in G$ then the \emph{translation
140: length} $\ell_X(g)$ is defined as:
141: \[
142: \ell_X(g) = \inf_{x \in X} d (x,gx) .
143: \]
144: It is easy to see that $\ell_X(g)$ only depends on the conjugacy class
145: of $g$ and that in the definition above the infimum can be replaced by
146: a minimum. Thus $\ell_X(g)=0$ if and only if $g$ fixes a point of $X$.
147: This discussion naturally leads us to the following definition:
148:
149: \begin{defn}
150: Let $F$ be a finitely generated free group and let $g, h \in F$ be
151: elements of $F$. We say that $g$ and $h$ are \emph{translation
152: equivalent in $F$}, denoted $g \equiv_{t} h$, if for every free
153: and discrete isometric action of $F$ on an $\mathbb{R}$-tree $X$ we
154: have $\ell_X(g) = \ell_X(h)$.
155: \end{defn}
156:
157: It is obvious that $\equiv_t$ is an equivalence relation on $F$.
158: Applying the above definition to the case where $X$ is the Cayley
159: graph of $F$ with respect to some free basis of $F$ implies that every
160: $\equiv_t$-equivalence class is the union of finitely many conjugacy
161: classes in $F$. Clearly, if $g \equiv_{t} h$ in $F$ and $\phi:F \to
162: F_1$ is an injective homomorphism to a free group $F_1$ then $\phi(g)
163: \equiv_{t} \phi(h)$ in $F_1$. Indeed, suppose $F_1$ acts freely and
164: discretely by isometries on an $\mathbb R$-tree $X$. Then, by
165: restriction, we get a free and discrete action of $\phi(F)$ on $X$,
166: and via a twist by $\phi$, we also get a free and discrete action of
167: $F$ itself on $X$. Namely, $f\cdot p:=\phi(f) p$, where $p\in X, f\in
168: F$. Since $u$ and $v$ are translation equivalent in $F$, it follows
169: that $\ell_X(u)=\ell_X(v)$, that is,
170: $\ell_X(\phi(u))=\ell_X(\phi(v))$.
171:
172:
173:
174: A phenomenon related to but different from translation equivalence was
175: studied by Smillie and Vogtmann~\cite{SV} who, given an arbitrary
176: finite set of conjugacy classes in a free group, constructed
177: multi-parametric families of free discrete actions where the
178: translation length of each conjugacy class from this set remains
179: constant through the family. Results similar in spirit to those of
180: \cite{SV} were also obtained by Cohen, Lustig and Steiner~\cite{CLS}.
181:
182:
183: The notion of translation equivalence is also related to the space of
184: geodesic currents on a free group. In~\cite{Ka04} the first author
185: studies the properties of an \emph{intersection form}
186: \[
187: I: FLen(F)\times Curr(F)\to \mathbb R.
188: \]
189: Here $FLen(F)$ is the
190: (non-projectivized) space of hyperbolic length functions corresponding
191: to free and discrete isometric actions of a $F$ on $\mathbb R$-trees
192: and $Curr(F)$ is the space of \emph{geodesic currents} on $F$, that is
193: the space of $F$-invariant positive Borel measures on the set of all
194: pairs $(x,y)$, where $x,y\in \partial F$ and $x\ne y$. Similarly to
195: Bonahon's notion~\cite{Bo} of the intersection number between geodesic
196: currents on hyperbolic surfaces, it turns out that if $\eta_g$ is the
197: ``counting'' current corresponding to a nontrivial $g\in F$ and if
198: $\ell\in FLen(F)$ is a length function then $I(\ell,\eta_g)=\ell(g)$.
199: Thus $g\equiv_t h$ in $F$ if and only if for every $\ell\in FLen(F)$
200: we have $I(\ell,\eta_g)=I(\ell,\eta_h)$. Therefore the notion of
201: translation equivalence, in a sense, measures the degeneracy of the
202: intersection form $I$ with respect to its second argument. We refer
203: the reader to~\cite{Ka04} for a detailed discussion on this topic.
204:
205:
206: The first natural problem is to demonstrate that there are nontrivial
207: instances of translation equivalence in free groups. We provide two
208: different sources of translation equivalence: one based on trace
209: identities in $SL(2,\mathbb C)$ and another based on power
210: redistribution for certain products of translation equivalent
211: elements. Both methods can be iterated and used to produce arbitrarily
212: large finite collections of distinct conjugacy classes in $F$ that are
213: pairwise translation equivalent. Both of these sources can also be
214: used to produce hyperbolic equivalence in the context of
215: $\mathbb{H}^2$-actions, although some distinctions do arise as will be
216: pointed out later. (See Example~\ref{distinct}.)
217:
218:
219:
220: Another natural question is to give a more algebraic and combinatorial
221: characterization of translation equivalence.
222:
223: Recall that an isometric action of a group $G$ on an $\mathbb{R}$-tree
224: $X$ is \emph{very small}~\cite{CL} if the following conditions hold:
225: \begin{enumerate}
226: \item The action is \emph{small}, that is, arc stabilizers do not
227: contain free subgroups of rank two.
228:
229: \item Stabilizers of tripods are trivial.
230:
231: \item For any $g \in G$ and for each $n\ne 0$ the fixed sets of $g$
232: and $g^n$ are equal.
233: \end{enumerate}
234: In particular, every free action and, more generally, an action with
235: trivial arc stabilizers, is very small. Results of
236: Cohen-Lustig~\cite{CL} and Bestvina-Feighn~\cite{BF93} imply that an
237: action of $F_n$ on an $\mathbb{R}$-tree is very small if and only if
238: this action represents a point in the standard length functions
239: compactification of the outer space. Thus a very small action can
240: always be approximated in the sense of length functions by a sequence
241: of free simplicial actions.
242:
243:
244: \begin{notation} If $A$ is a basis of a free group $F$ any element $g\in F$
245: is represented by a unique reduced word $w_g$ over the alphabet
246: $A^{\pm 1}$. The \emph{length of $g$ with respect to the basis
247: $A$}, denoted $|g|_A$, is the number of letters in $w_g$. A word
248: is \emph{cyclically reduced} if all its cyclic permutations are
249: reduced. Any reduced word $w$ can be uniquely factored as $w = c u
250: c^{-1} $ where $u$ is cyclically reduced. If $w_g = c u c^{-1}$ is
251: such a factorization, then $|u|_A$ is called \emph{the cyclically
252: reduced length of $g$ with respect to $A$} is denoted by
253: $||g||_A$. Note that $||g||_A=\ell_X(g)$ where $X$ is the Cayley
254: graph of $F$ with respect to $A$. A \emph{cyclic word} in $A^{\pm
255: 1}$ is the set of all cyclic permutations of a cyclically reduced
256: word. There is a canonical identification between the set of cyclic
257: words in $A^{\pm 1}$ and the set of conjugacy classes in $F$.
258:
259: If $w$ is a cyclic word consisting of all cyclic permutations of a
260: nontrivial word $u$, and if $v$ is a word in $A^{\pm 1}$, we define
261: the \emph{number of occurrences of $v$ in $w$} as the number of
262: those $i$, $0\le i<|u|$, such that the infinite word $uuu\dots$
263: begins with $u_iv$, where $u_i$ is the initial segment of $u$ of
264: length $i$. If $w$ is a cyclic word and $x,y \in A^{\pm 1}$ we use
265: $n_A(w;x,y)$ (or just $n(w;x,y)$ if $A$ is fixed) to denote the
266: total number of occurrences of the subwords $x y$ and $y^{-1}
267: x^{-1}$ in $w$. Thus $n(w;x,y) = n(w;y^{- 1},x^{- 1}) =n(w^{-
268: 1};x,y)$. Similarly, in this case if $w$ is a cyclic word and
269: $x\in A^{\pm 1}$, we denote by $n(w;x)$ the total number of
270: occurrences of $x$ and $x^{-1}$ in $w$. Thus again
271: $n(w;x)=n(w;x^{-1}) = n(w^{-1};x)$. If $[g]$ is a nontrivial
272: conjugacy class in $F$ and $w$ is the unique cyclic word over $A$
273: representing $[g]$, we denote $n_A([g];x):=n(w;x)$, where $x\in
274: A^{\pm 1}$.
275: \end{notation}
276:
277:
278: In studying automorphisms of free groups, Whitehead automorphisms and
279: the Whitehead graph of a cyclically reduced word play a major role.
280: See Lyndon-Schupp \cite{LS} for a detailed discussion. Note that in
281: \cite{LS} Whitehead graphs are called \emph{star graphs}.
282:
283: \begin{defn}[Whitehead graph]
284: Let $w$ be a nontrivial cyclic word in $F(A)$. The \emph{Whitehead
285: graph $\mathcal{W}_A(w)$ of $w$ with respect to $A$} is the
286: labelled undirected graph defined as follows. The vertex set of
287: $\mathcal{W}_A(w)$ is $A^{\pm 1}$. If $x, y \in A^{\pm 1}$ are such
288: that $x \ne y$, there is an edge in $\mathcal{W}_A(w)$ between $x$
289: to $y$ with label $n(w;x,y^{-1})$.
290:
291:
292: If $[g]$ is a nontrivial conjugacy class in $F$, then $[g]$ is
293: represented by a unique cyclic word $w$ in $F(A)$. Then the
294: \textit{Whitehead graph $\mathcal{W}_A ([g])$ of $[g]$ with respect
295: to $A$} is defined as $\mathcal{W}_A(w)$. Note that
296: $\mathcal{W}_A (w) = \mathcal{W}_A (w^{- 1})$ for any nontrivial
297: cyclic word $w$.
298:
299: \end{defn}
300:
301:
302:
303: We obtain the following result:
304:
305: \begin{theor}\label{A}
306: Let $F$ be a finitely generated free group and let $g, h \in F$ be
307: nontrivial elements.
308:
309: Then the following statements are equivalent:
310: \begin{enumerate}
311: \item $\ell_X(g) = \ell_X(h)$ for every very small action of $F$ on an
312: $\mathbb{R}$-tree $X$.
313:
314:
315: \item $\ell_X(g) = \ell_X(h)$ for every free action of $F$ on an
316: $\mathbb{R}$-tree $X$.
317:
318:
319: \item $g \equiv_{t} h$ in $F$.
320:
321: \item $||g||_A = ||h||_A$ for every free basis $A$ of $F$.
322:
323: \item $\mathcal{W}_A ([g]) = \mathcal{W}_A ([h])$ for every free basis
324: $A$ of $F$. That is, the conjugacy classes $[g]$ and $[h]$ have the
325: same Whitehead graphs with respect to $A$.
326:
327: \item $n_A([g];x)=n_A([h];x)$ for every free basis $A$ of $F$ and for
328: every $x\in A$.
329: \end{enumerate}
330: \end{theor}
331:
332:
333: Theorem~\ref{A} immediately yields a more combinatorial version of
334: translation equivalence:
335:
336:
337: \begin{cor}
338: Let $F$ be a finitely generated free group with a free basis $A$ and
339: let $g, h \in F$. Then the following conditions are equivalent:
340: \begin{enumerate}
341: \item $g \equiv_{t} h$ in $F$.
342:
343: \item $||\phi(g)||_A =||\phi(h)||_A$ for every automorphism $\phi$ of
344: $F$.
345:
346: \item $||\phi(g)||_A = ||\phi(h)||_A$ for every injective endomorphism
347: $\phi$ of $F$.
348:
349: \item $||\phi(g)||_B =||\phi(h)||_B$ for every free group $F_1$ with a
350: free basis $B$ and for every injective homomorphism $\phi : F \to
351: F_1$.
352:
353: \end{enumerate}
354: \end{cor}
355:
356: The a priori weakest condition in Theorem~\ref{A} is condition (4)
357: which deserves further comment. Let $S$ be a closed surface as before.
358: If $\gamma$ and $\delta$ are free homotopy classes of closed curves on
359: $S$, we use $i(\gamma, \delta)$ to denote the \emph{geometric
360: intersection number} of $\gamma$ and $\delta$. We say that free
361: homotopy classes $\gamma_1, \gamma_2$ of closed curves on $S$ are
362: \emph{simple intersection equivalent} if
363: $i(\gamma_1,[c])=i(\gamma_2,[c])$ for every essential \textit{simple
364: closed curve} $c$ on $S$. It is easy to see that hyperbolic
365: equivalence implies simple intersection equivalence. Surprisingly
366: however, the converse is not true as was recently proved by Chris
367: Leininger~\cite{Le}.
368:
369: It is therefore natural to consider the analogue of simple
370: intersection equivalence for free groups. If $G = \pi_1(S)$ and $c$ is
371: an essential simple closed curve on $S$, then $c$ defines a splitting
372: of $G$ as either an amalgamated product or as an HNN-extension over a
373: cyclic subgroup. Let $X$ be the Bass-Serre tree corresponding to that
374: splitting. It is easy to see that for any $g \in G$ we have $i([g],
375: [c])= \ell_X(g)$. The difficulty is that for a free group $F$, a
376: single element of $F$, even if it is a primitive one, does not define
377: a splitting of $F$. A free basis $A$ of $F$ does, however, define
378: such a splitting. Namely, the splitting of $F$ as the multiple
379: HNN-extension of the trivial group with stable letters corresponding
380: to the elements of $A$ and the Bass-Serre tree $X$ of this splitting
381: is precisely the Cayley graph of $F$ with respect to $A$. Then for any
382: $g \in F$ we have $\ell_X(g) =||g||_A$, the cyclically reduced length
383: of $g$ with respect to $A$. For a free basis $A$ of $F$ and an
384: element $g \in F$ we can therefore define the \textit{intersection
385: number} $i(g,A):=||g||_A$. By analogy with the surface group case
386: we say that $g_1,g_2 \in F$ are \emph{simple intersection equivalent}
387: in $F$ if for every free basis $A$ of $F$ we have $i(g_1,A) =
388: i(g_2,A)$. Unlike in the surface group case, Theorem~\ref{A} says that
389: in free groups simple intersection equivalence is the same as
390: translation equivalence.
391:
392: We can now state the main sources of translation equivalence in free
393: groups that we have discovered so far.
394:
395: If $F$ is a finitely generated free group and $u,v\in F$ are
396: nontrivial elements, we say that $u$ and $v$ are \emph{trace
397: equivalent} or \emph{character equivalent} in $F$, denoted $u
398: \equiv_{c} v$, if for every representation $\alpha: F\to SL(2,\mathbb
399: C)$ we have $tr(\alpha(u))=tr(\alpha(v))$. Character equivalent words
400: come from the so-called ``trace identities'' in $SL(2,\mathbb C)$ and
401: are quite plentiful (see, for example, \cite{Ho}). A corollary of
402: Theorem~\ref{A} together with a result of Horowitz~\cite{Ho} is:
403:
404: \begin{theor}\label{B}
405: Let $F$ be a finitely generated free group and suppose that
406: $u\equiv_c v$ in $F$. Then $u\equiv_t v$ in $F$.
407: \end{theor}
408:
409: A particularly interesting example of character equivalence comes from
410: two-variable ``palindromic reversing''. Let $w(x,y) \in F(x,y)$ be a
411: freely reduced word. We denote by $w^R(x,y)$ the word $w(x,y)$ read
412: backwards, but without inverting the letters. Thus $w^R(x,y) = (w(x^{-
413: 1}, y^{- 1}))^{- 1}$. We prove:
414:
415: \begin{theor}\label{C}
416: Let $F$ be a free group of rank $k \ge 2$ and let $w(x,y) \in
417: F(x,y)$ be a freely reduced word. Then for any $g, h \in F$ we have
418: \[
419: w(g,h) \equiv_{t} w^R(g,h) \quad \text{ in } F.
420: \]
421: \end{theor}
422:
423: We shall give a direct proof of the above statement as well as a proof
424: via character equivalence.
425:
426:
427: A very different source of translation equivalence is given by:
428:
429: \begin{theor}\label{D}
430: Let $F$ be a free group of rank $k \ge 2$ and let $g,h \in F$ be
431: such that $g \equiv_{t} h$ but $g \ne h^{-1}$. Then for any
432: positive integers $p, q, i, j$ such that $p+q = i+j$ we have
433: \[
434: g^p h^q \equiv_{t} g^i h^j \quad \text{ in } F.
435: \]
436: \end{theor}
437: Theorem~\ref{B} states that $\equiv_c$ implies $\equiv_t$. However, it
438: turns out that Theorem~\ref{D} does not hold for character
439: equivalence, as demonstrated by Example~\ref{distinct} below. This
440: example shows that $\equiv_t$ does not imply $ \equiv_c$ and that,
441: although character equivalence and translation equivalence are closely
442: related phenomena, they are not the same and translation equivalence
443: is more general.
444:
445:
446:
447: The following is an analogue of a result of Randol~\cite{Ra} about
448: hyperbolic surfaces.
449:
450: \begin{cor}\label{cor:C}
451: Let $F$ be a free group of rank $k \ge 2$. Then for any integer $M
452: \ge 1$ there exist elements $g_1, \dots, g_M \in F$ such that $g_i
453: \equiv_{t} g_j$ in $F$ and such that for $i \ne j$ $g_i$ is not
454: conjugate to $g_j^{\pm 1}$.
455: \end{cor}
456:
457: \begin{proof}
458: Let $M \ge 1$ be an integer and let $(a, b, \dots )$ be a free basis
459: of $F$. Consider $g=a$ and $h = b a^{-1} b^{-1}$. Put $g_i = g^i
460: h^{2M-i} = a^i b a^{i-2M} b^{-1}$ for $i = 1, \dots, M$. Then by
461: Theorem~\ref{D} $g_i \equiv_{t} g_j$ in $F$. On the other hand $g_i$
462: is not conjugate to $g_j^{\pm 1}$ for $i \ne j$. This can be seen,
463: for example, by observing that $g_i$ and $g_j^{\pm 1}$ have distinct
464: images in the abelianization of $F$.
465: \end{proof}
466:
467: \begin{rem}
468: Theorem~\ref{C} can be iterated to produce other examples with the
469: same properties as in Corollary~\ref{cor:C}. Namely, let
470: $\phi_i:F(x,y) \to F(x,y)$ be injective endomorphisms of $F(x,y)$
471: for $i = 1, \dots, N$. Let $\phi := \phi_N \circ \dots \circ
472: \phi_1$. For $i = 1, \dots, N-1$ put $\psi_i : = \phi_N \circ \dots
473: \circ \phi_{i+1}$ and $\theta_i = \phi_i \circ \dots \circ \phi_1$.
474: Then $\phi = \psi_i \circ \theta_i$. Let $\psi_i (x) = u_i(x,y)$,
475: $\psi_i(y)=v_i(x,y)$, $\theta_i(x)=r_i(x,y)$ and $\theta_i(y)
476: =s_i(x,y)$. Let $w(x,y) = \phi(x)$. Then $w = \psi_i(\theta_i(x)) =
477: \psi_i(r_i(x,y)) = r_i(u_i(x,y),v_i(x,y))$. Let $w_i =
478: r_i^R(u_i(x,y), v_i(x,y))$. Theorem~\ref{C} implies that, $w(g,h)
479: \equiv_{t} w_i(g,h)$ for any $g, h \in F$ for $i =1, \dots, N-1$. It
480: is possible, for each $N \ge 1$, to choose the endomorphisms
481: $\phi_i$ and then elements $g, h \in F$ so that $w_1, \dots,
482: w_{N-1}$ are pairwise non-conjugate in $F$.
483:
484: Corollary~\ref{cor:C} also follows from Theorem~\ref{B} and the
485: result of Horowitz~\cite{Ho} establishing (via a more complicated
486: family of words) a similar result for character equivalence.
487: Moreover, as we observe later in Corollary~\ref{cor:C'}, it is
488: possible to generalize the proof of Corollary~\ref{cor:C} to many
489: non-free groups.
490: \end{rem}
491:
492:
493: Using $SL_2$ trace identities, we show that Theorem~\ref{C} and a
494: version of Theorem~\ref{D} also hold for standard hyperbolic
495: equivalence. Via a limiting argument we conclude that these statements
496: also apply to the tree actions that occur in the Thurston boundary of
497: the Teichm\"uller space of a closed hyperbolic surface, either
498: orientable or non-orientable. Recall that each point $\mu$ in the
499: Thurston boundary of the Teichm\"uller space is a measured lamination.
500: There is an $\mathbb{R}$-tree $X_{\mu}$, dual to the lift of this
501: lamination, that comes equipped with a small isometric action of the
502: fundamental group of the surface. In the case of a non-orientable
503: surface not all such actions are very small.
504:
505: \begin{theor}\label{E} Let $S$ be a possibly non-orientable closed surface of negative Euler
506: characteristic and let $G =\pi_1(S)$. Then the following hold:
507: \begin{enumerate}
508: \item For any $g, h \in G$ and for any $w(x,y) \in F(x,y)$ and for any
509: tree action $\mu$ of $G$ in the Thurston boundary of the
510: Teichm\"uller space of $S$ we have
511: \[
512: \ell_{X_{\mu}}(w(g,h)) = \ell_{X_{\mu}}(w^R(g,h)).
513: \]
514: \item For any conjugate elements $g,h \in G$, for any $p, q > 0$ and
515: for any tree action $\mu$ of $G$ in the Thurston boundary of the
516: Teichm\"uller space of $S$ we have
517: \[
518: \ell_{X_{\mu}}( g^p h^q) = \ell_{X_{\mu}}(g^q h^p).
519: \]
520: \item If $S$ is orientable then for any conjugate elements $g,h \in
521: G$, for each point $\mu\in \overline{\mathcal T(S)}-\mathcal T(S)$
522: and for any positive integers $p,q,i,j$ such that $p+q=i+j$ we have
523: \[
524: \ell_{X_{\mu}}(g^ph^q)=\ell_{X_{\mu}}(g^ih^j).
525: \]
526: \end{enumerate}
527: \end{theor}
528:
529: The paper is organized as follows. In Section~2 we discuss Whitehead
530: graphs and prove Theorem~A. In Section~2 we obtain a direct geometric
531: proof of Theorem~C about the palindromic sources of translation
532: equivalence. In Section~4 we use the analysis of possible axis
533: configurations for compositions of isometries of $\mathbb R$-trees to
534: provide a geometric proof of Theorem~D. Section~5 contains a
535: discussion of $SL_2$ trace identities and a proof of Theorem~B. In
536: Section~6 we analyze how our results apply to tree actions occurring
537: in the boundary of the Teichm\"uller space and prove Theorem~E. In
538: Section~7 we discuss various examples and counter-examples, and in
539: Section~8 we list a number of interesting open problems.
540:
541:
542: The authors are grateful to the organizers of the ``Geometric Groups
543: on the Gulf'' Conference in Mobile in February 2004, where the
544: conversations and discussions eventually leading to the writing of
545: this paper took place. The authors also thank David Berg, Brian
546: Bowditch, Enric Ventura and Victor Pan for useful conversations. The
547: authors also thank the referee for his careful reading of the paper
548: and for many helpful comments and suggestions.
549:
550:
551: \section{Whitehead graphs and a characterization of translation equivalence}
552:
553: Let $F = F(A) = F(a_1, \dots, a_k)$ be a free group with basis $A = \{
554: a_1, \dots, a_k \}$ where $k \ge 2$.
555:
556:
557: Let $x, y \in \{ a_1, \dots, a_k \}^{\pm 1}$ be such that $x \ne
558: y^{\pm 1}$. Denote by $\phi_{x, y}$ the Nielsen automorphism of $F$
559: that sends $x$ to $x y$ and fixes each generator $a_i \ne x^{\pm 1}$.
560: For simplicity, if $i \ne j, 1 \le i, j \le k$ we use $\phi_{i, j}$ to
561: denote $ \phi_{a_i, a_j}$.
562:
563: The following lemma is obvious:
564:
565: \begin{lem}\label{l1}
566: For any cyclic word $w$ and any $x, y \in \{a_1, \dots, a_k \}^{\pm
567: 1}$ such that $x \ne y^{\pm 1}$ we have
568: \[
569: ||\phi_{x, y}(w)|| - ||w|| = n(w;x) - 2 n(w;x, y^{- 1}).
570: \]
571: \end{lem}
572:
573: \begin{lem}\label{l2}
574: Let $w$ be a cyclic word. Then:
575: \[
576: n(w;a_i) = \frac{1}{k} \big( ||w|| + \sum_{j \ne i} (||\phi_{i,
577: j}(w)|| - ||\phi_{j, i}(w)||) \big).
578: \]
579: \end{lem}
580:
581: \begin{proof}
582: Note that $n(w; x, y^{- 1}) = n(w; y, x^{- 1})$. Therefore
583:
584: \begin{gather*}
585: ||{\phi}_{x,y}(w)||-||w||=n(w;x)-2n(w;x,y^{-1})\\
586: ||{\phi}_{y,x}(w)||-||w||=n(w;y)-2n(w;y,x^{-1})=n(w;y)-2n(w;x,y^{-1}),
587: \end{gather*}
588:
589:
590: and so
591: \[
592: ||\phi_{x, y}(w)|| - ||\phi_{y, x}(w)|| = n(w;x) - n(w; y).
593: \]
594:
595: Let $x = a_i$ and $y$ vary over $a_j, j \ne i$. Then summing up the
596: instances of the above equality for $x = a_i, y = a_j$ we get
597: \[
598: (k-1) n(w;a_i) - \sum_{j \ne i} n(w,a_j) = \sum_{j \ne i}
599: (||\phi_{i, j}(w)|| - ||\phi_{j, i}(w)||)
600: \]
601: On the other hand
602: \[
603: n(w;a_i) + \sum_{j \ne i} n(w,a_j) = ||w||.
604: \]
605: Adding the above formulas we get
606: \[
607: k \, n(w;a_i) = ||w|| +
608: \sum_{j \ne i}^n (||\phi_{i,j}(w)|| - ||\phi_{j, i}(w)||),
609: \]
610: which yields the statement of the lemma.
611: \end{proof}
612:
613: \begin{prop}\label{prop:impl}
614: Let $F$ be a finitely generated free group and let $u,v \in F$ be
615: nontrivial elements such that for every free basis $A$ of $F$ we
616: have $||u||_A = ||v||_A$. Then for every free basis $A$ of $F$ the
617: Whitehead graphs of $[u]$ and $[v]$ with respect to $A$ are equal.
618: \end{prop}
619: \begin{proof}
620: Let $A$ be a free basis of $F$. We consider the conjugacy classes
621: $[u]$, $[v]$ as cyclic words over $A$. The assumptions of the
622: proposition imply that for every automorphism $\phi$ of $F$
623: \[
624: || \phi(u)||_A = ||\phi(v)||_A.
625: \]
626:
627: By Lemma~\ref{l2} it follows that for each $x \in A$ we have $n([u];x)
628: =n([v],x)$. Therefore by Lemma~\ref{l1} for every $x, y \in A^{\pm 1}$
629: such that $x \ne y^{\pm 1}$ we have
630: \[
631: n([u];x, y) = n([v]; x, y ).
632: \]
633: For a fixed $x \in A$ and an arbitrary cyclic word $w$
634: \[
635: n(w;x) =n(w; x, x) + \sum_{y \ne x^{\pm 1}} n(w; x, y ).
636: \]
637: Since $n([ u ]; x) = n([v];x)$ and $n([u]; x, y) = n([v]; x, y)$ for
638: any $y \ne x^{\pm 1}$, $y \in A^{\pm 1}$, it follows that $n([u]; x,
639: x) = n([v]; x, x)$.
640:
641: Thus we have shown that for any $x, y \in A^{\pm 1}$ such that $x \ne
642: y^{-1}$ we have $n([u] ; x, y) = n ([v]; x, y)$. This means that $[u]$
643: and $[v]$ have equal Whitehead graphs with respect to $A$, as claimed.
644: \end{proof}
645:
646: We can now establish Theorem~\ref{A} from the Introduction:
647:
648: \begin{theore}
649: Let $F$ be a finitely generated free group and let $g, h \in F$ be
650: nontrivial elements.
651:
652: Then the following statements are equivalent:
653: \begin{enumerate}
654: \item $\ell_X(g) = \ell_X(h)$ for every very small action of $F$ on an
655: $\mathbb{R}$-tree $X$.
656:
657:
658: \item $\ell_X(g) = \ell_X(h)$ for every free action of $F$ on an
659: $\mathbb{R}$-tree $X$.
660:
661:
662: \item $g \equiv_{t} h$ in $F$.
663:
664: \item $||g||_A = ||h||_A$ for every free basis $A$ of $F$.
665:
666: \item $\mathcal{W}_A ([g]) = \mathcal{W}_A ([h])$ for every free basis
667: $A$ of $F$. That is, the conjugacy classes $[g]$ and $[h]$ have the
668: same Whitehead graphs with respect to $A$.
669:
670: \item $n_A([g];x)=n_A([h];x)$ for every free basis $A$ of $F$ and for
671: every $x\in A$.
672: \end{enumerate}
673: \end{theore}
674:
675:
676: \begin{proof}
677: Let $k$ be the rank of $F$. We may assume that $k\ge 2$ since for
678: $k=1$ the statement of the theorem is obvious.
679:
680: The implications $(1)\Rightarrow (2)$, $(2)\Rightarrow (3)$,
681: $(3)\Rightarrow (4)$, $(5) \Rightarrow (6)$ and $(6) \Rightarrow
682: (4)$ are obvious. Moreover, $(3)$ implies $(1)$ since by the results
683: of \cite{CL} every very small action is the limit (in the sense of
684: length functions) of free discrete actions. Thus $(1), (2), (3)$ are
685: equivalent. The implication $(4)\Rightarrow (6)$ follows from
686: Lemma~\ref{l2} and the implication $(4)\Rightarrow (5)$ is
687: Proposition~\ref{prop:impl}.
688:
689:
690: We now show that $(5)$ implies (3). Indeed, suppose that $F$ is
691: acting freely, discretely and isometrically on an $\mathbb R$-tree
692: $X$. Let $Y=X/F$ be the quotient graph of $X$. Since $X$ is a metric
693: tree, the edges $e$ of $Y$ come equipped with the lengths $l(e)$.
694: Thus every edge-path in $Y$ has a \emph{length} which is the sum of
695: the lengths of the edges of this path. There is an obvious canonical
696: identification between $F$ and $\pi_1(Y,y)$ where $y$ is a vertex of
697: $Y$. Choose an orientation $EY=E^+Y\sqcup E^-Y$ on $Y$. Then for
698: every maximal tree in $Y$ there is a canonically associated free
699: basis of $\pi_1(Y,y)$.
700:
701:
702: Given a conjugacy class $[g]$ of $g\in F$, represent it by an
703: immersed loop $\gamma$ in $Y$. Then $\ell_X(g)=\sum_{e\in E^+Y}
704: n(\gamma; e )l(e)$, where $n(\gamma; e )$ is the number of times
705: that $\gamma$ traverses $e$ (in either direction). We prove that (5)
706: implies (3) by showing that, given $e\in E^+Y$, the number
707: $n(\gamma; e )$ is completely determined by the Whitehead graphs of
708: $[g]$ with respect to free bases of $F$.
709:
710: There are two cases.
711:
712: First, suppose that $e$ does not separate $Y$. Choose a maximal tree
713: not containing $e$, and consider the associated free basis $A$ of
714: $F=\pi _1(Y,y)$. Then $n_A(\gamma;e )=n_A([g];a_e)$, where $a_e\in
715: A$ is the generator corresponding to $e$.
716:
717: Suppose next that $e$ separates $Y$, so that $Y=Y_1\cup e \cup Y_2$,
718: say with $y \in Y_1$. Choose a basis $B$ of $F$ associated to any
719: maximal tree in $Y$. This basis is partitioned as $B=B_1\sqcup B_2$,
720: with $b\in B_i$ if and only if it corresponds to an edge in $Y_i$.
721: Then $\displaystyle n(\gamma; e )= \sum_{x\in B_1, y\in
722: B_2}n_B([g];x,y)$.
723:
724: \end{proof}
725:
726: \section{Palindromic sources of translation equivalence}
727:
728: Let $F$ be a free group of rank $k \ge 2$ and let $A = \{a_1, \dots,
729: a_k\}$ be a free basis of $F$. Let $u = u(a_1, \dots, a_k) \in F$ be a
730: freely reduced word over $A$. We define the \textit{palindromic
731: reverse of $u$ with respect to $A$}, denoted $u^R$, as:
732: \[ u^R:=u(a_1^{- 1}, \dots, a_k^{- 1})^{- 1}. \]
733: Thus $u^R$ is the word $u$ read backwards without inverting the
734: letters.
735:
736: Similarly, we define the palindromic reverse $w^R$ of a cyclic word
737: $w$ over $A$. Thus $w^R$ is again a cyclic word. Namely, if $w$ is
738: represented by a cyclically reduced word $u$ then $w^R$ is represented
739: by the cyclically reduced word $u^R$.
740:
741: \begin{prop}\label{prop:use}
742: Let $F = F (a,b)$ be free of rank two. Then for any cyclic word $w$
743: in $F$ over $\{a,b\}^{\pm 1}$ and for any $\phi \in Aut(F)$ we have:
744: \[
745: (\phi(w))^R = \phi(w^R).
746: \]
747: \end{prop}
748: \begin{proof}
749: For any $\psi \in Aut(F)$ denote by $\overline{\psi}$ the image of
750: $\psi$ in $Out(F)$. For a free basis $A =(a,b)$ of $F$ denote by
751: $\tau_A$ the automorphism of $F$ defined as $\tau_A(a) = a^{- 1},
752: \tau_A(b) = b^{-1}$. It is easy to see that in $Out(F)$ the element
753: $\overline{\tau_A}$ commutes with all the elementary Nielsen
754: automorphisms with respect to $A$ and hence $\overline{\phi_A}$ is
755: central in $Out(F)$. Therefore for any other free basis $B$ of $F$
756: we have $\overline{\tau_A} = \overline{\tau_B}$.
757:
758: Suppose now that $w(a,b)$ is a cyclic word in $F(a,b)$ and $\phi \in
759: Aut(F)$. Let $u(a, b) = \phi(w)$. Then by the above observation
760: $\phi(w(a^{-1}, b^{-1})) = u(a^{- 1}, b^{- 1})$. Since $w^R =(w(a^{-
761: 1}, b^{-1}))^{- 1}$ and $u^R =(u(a^{- 1}, b^{- 1}))^{- 1}$, this
762: implies that $(\phi(w))^R = \phi(w^R)$.
763: \end{proof}
764:
765:
766: \begin{rem}
767: It is well-known that $Out(F_2) \cong GL(2, \mathbb{Z})$ and that
768: the center of $GL(2, \mathbb{Z})$ is cyclic of order two. Thus in
769: fact the outer automorphism $\overline{\tau_A}$ is the only
770: nontrivial element of the center of $Out(F_2)$, although we did not
771: need this fact in the above proof.
772: \end{rem}
773:
774: We can now prove Theorem~\ref{C} from the Introduction:
775: \setcounter{theore}{2}
776: \begin{theore}\label{palindr}
777: Let $F$ be a free group of rank $k \ge 2$ and let $w(x,y) \in
778: F(x,y)$ be a freely reduced word. Then for any $g, h \in F$ we have
779: \[
780: w(g,h) \equiv_{t} w^R(g,h) \quad \text{ in } F. \]
781: \end{theore}
782:
783:
784:
785: \begin{proof}
786: If $g$ and $h$ commute in $F$ then the statement is obvious. Suppose
787: now that $g$ and $h$ do not commute and hence $F_1 = \langle g, h
788: \rangle$ is a free group of rank two. Let $F$ act freely and
789: discretely by isometries on an $\mathbb{R}$-tree $X$. Let $X_1$ be
790: the minimal $F_1$-invariant subtree and let $Y = X_1/F_1$ be the
791: quotient graph. Then topologically $Y$ is either a wedge of two
792: circles or is a $\theta$-graph or $Y$ consists of two disjoint
793: circles joined by an edge. In each case $Y$ possesses an involution
794: isometry $\sigma$ that leaves a maximal subtree $T$ of $Y$
795: invariant, fixes some point $y\in T$ and takes every edge $e$
796: outside of a maximal tree to $e^{-1}$. This is shown in
797: Figure~\ref{Fi:inv}.
798:
799: \begin{figure}[here]
800: \epsfxsize=4in \epsfbox{inv.eps}
801: \caption{``Hyper-elliptic involution"}\label{Fi:inv}
802: \end{figure}
803:
804:
805:
806: Thus if $c_1, c_2$ is the basis of $\pi_1(Y,y)$ corresponding to $T$
807: and $\sigma_{\#}$ denotes the isomorphism of $\pi_1(Y,y)$ induced by
808: $\sigma$, then for any cyclic word $u(c_1,c_2)$ we have $\sigma_{\#}
809: u(c_1,c_2) =u(c_1^{-1},c_2^{-1})$. Since $\sigma$ is an isometry of
810: $Y$, this means that $\ell_X(u(c_1,c_2)) = \ell_X(u(c_1^{- 1}, c_2^{-
811: 1}))$. On the other hand $u(c_1^{- 1}, c_2^{- 1}) = (u^R(c_1,c_2
812: ))^{-1}$ and so
813: \[
814: \ell_X(u( c_1, c_2 )) = \ell_X(u^R(c_1,c_2 )).
815: \]
816:
817: The pair $(c_1, c_2)$ is a free basis of $F_1 = F(g,h)$. Write $w(g,h)
818: =u(c_1, c_2 )$ and $w^R(g,h) = u'(c_1,c_2)$. By
819: Proposition~\ref{prop:use} $u'= u^R$. Hence
820: \[
821: \ell_X(w(g, h)) = \ell_X(u(c_1, c_2)) = \ell_X(u^R(c_1, c_2)) =
822: \ell_X(w^R(g, h)),
823: \]
824: as required.
825: \end{proof}
826:
827:
828: \section{Axis diagrams for free actions}
829:
830: Let $G$ be a group acting by isometries on an $\mathbb{R}$-tree $X$.
831: Recall that $g\in G$ is called \emph{elliptic} if $\ell_X(g)=0$ and
832: $g$ is called \emph{hyperbolic} if $\ell_X(g) > 0$. Thus $g$ is
833: elliptic if and only if it fixes a point of $X$.
834:
835: For a hyperbolic $g \in G$ put
836: \[
837: L_g = \{ x \in X : d(x, g x) = \ell_X(g)\} .
838: \]
839: Then $L_g$ is the smallest $g$-invariant subtree of $X$ which is
840: isometric to a line and on which $g$ acts by a translation of
841: magnitude $\ell_X( g )$. The set $L_g$ is called the \emph{axis} of
842: $g$. In this section if an $\mathbb R$-tree $X$ is fixed, we will omit
843: the subscript and denote the translation length of an isometry $g$ of
844: $X$ by $\ell(g)$.
845:
846: The following simple proposition enumerating all the possibilities for
847: the configuration of $L_{g h}$ with respect to $L_g, L_h$ for two
848: hyperbolic isometries $g$ and $h$ is essentially a restatement of
849: Proposition~1.6 of Paulin~\cite{Pau}.
850:
851:
852: \begin{prop}\label{pau}
853: Let $X$ be an $\mathbb R$-tree and let $g, h \in Isom(X)$ be two
854: hyperbolic isometries of $X$. Then the following hold:
855: \begin{enumerate}
856:
857:
858: \item Suppose that $|L_g \cap L_h| \le 1$ and let $D = d(L_g, L_h)$.
859: Then
860: \[
861: \ell(g h) = \ell(g) + \ell(h) + 2D.
862: \]
863: \item Suppose that $L_g \cap L_h$ is a nondegenerate segment $[x,y]$.
864:
865: \noindent (a) If the translation directions of $g$ and $h$ on $[x,y]$ coincide then
866: \[
867: \ell(g h) = \ell(g) + \ell(h).
868: \]
869:
870: \noindent (b) If the translation directions of $g$ and $h$ on $[x,y]$ are opposite then
871: \[
872: \ell(g h) =
873: \begin{cases}
874: \ell(g) + \ell(h) - 2 d(x, y) \quad \text{ if } \ell(g)\ge d(x,y), \ell (h)\ge d(x,y) \\
875: |\ell(g) - \ell(h)| \quad \text{otherwise.}
876: \end{cases}
877: \]
878: \item If $L_g$ and $L_h$ are equal or intersect in a ray, then
879: \[
880: \ell(gh)=
881: \begin{cases}
882: \ell(g) +\ell(h) \quad \text{ if $g$ and $h$ translate
883: in the same direction } \\
884: |\ell(g) - \ell(h)| \quad \text{ otherwise.}
885: \end{cases}
886: \]
887: \end{enumerate}
888: \end{prop}
889:
890:
891: \begin{rem}
892: The case where $L_g\cap L_h$ consists of a single point is omitted
893: in Proposition~1.6 of \cite{Pau}. Only the cases where $L_g\cap L_h$
894: is empty or contains a nondegenerate segment are explicitly covered
895: there. However the proofs for the cases where $L_g\cap L_h$ is empty
896: and where $L_g\cap L_h$ is a single point are completely analogous.
897: \end{rem}
898:
899: We need another simple fact (see Proposition~1.8 of Paulin~\cite{Pau})
900:
901:
902: \begin{prop}\label{pau1}
903: Let $g,h$ be two elliptic isometries of an $\mathbb R$-tree $X$.
904:
905: Then
906: \[
907: \ell_X(gh)=2 \min \{ d(x,y)| gx=x, hy=y\}.
908: \]
909: \end{prop}
910:
911: The following statement, together with the definition of translation
912: equivalence, immediately implies Theorem~\ref{D} from the
913: introduction.
914:
915: \begin{thm}\label{thm-pq}
916: Let $G$ be a group acting isometrically with trivial arc stabilizers
917: on an $\mathbb{R}$-tree $X$. Let $g, h \in G$ be nontrivial
918: elements of $G$ such that $g\ne h^{-1}$.
919:
920: \begin{enumerate}
921: \item Suppose that $\ell_X(g) =\ell_X(h)>0$. Then for any positive
922: integers $p,q,i,j$ such that $p + q = i + j$ we have
923: \[
924: \ell_X(g^p h^q) = \ell_X(g^i h^j).
925: \]
926:
927: \item Suppose that $\ell_X(g)=\ell_X(h)=0$.
928:
929: Then for any integers $p,q, i,j$ such that $g^p,h^q, g^i, h^j$ are
930: nontrivial we have
931: \[
932: \ell_X(g^ph^q)=\ell_X(g^ih^j)
933: \]
934: \end{enumerate}
935: \end{thm}
936:
937: \begin{proof}
938:
939: Part (2) follows directly from Proposition~\ref{pau1}. Indeed, since
940: the action of $G$ has trivial arc stabilizers, there are some
941: $x,y\in X$ such that $Fix(g)=\{x\}$ and $Fix(h)=\{y\}$. Then for any
942: $p,q,i,j$ such that $g^p,g^i,h^q,h^j$ are nontrivial, we have
943: $Fix(g^p)=Fix(g^i)=\{x\}$ and $Fix(h^q)=Fix(h^j)=\{y\}$ and hence by
944: Proposition~\ref{pau1} $\ell(g^ph^q)=d(x,y)=\ell(g^ih^j)$.
945:
946: Suppose now that the assumptions of part (1) of Theorem~\ref{thm-pq}
947: are satisfied.
948:
949: Denote $a =\ell(g) = \ell(h)$. If $g = h$ then the statement is
950: obvious. Suppose now that $g \ne h$, so that $g \ne h^{\pm 1}$. Let
951: $p,q,i,j \ge 1$ be integers such that $p+q =i+j$.
952:
953:
954: Observe that $L_g = L_{g^n}$ and $L_h = L_{h^n}$ for any $n \ge 1$.
955: Moreover, in this case $\ell(g^n) = \ell(h^n) = n \ell(h) = n
956: \ell(g) = na$.
957:
958: Suppose first that $L_g \cap L_h$ consists of at most one point. Put
959: $D = d(L_g, L_h)$. Then by part (1) of Proposition~\ref{pau}
960: \begin{gather*}
961: \ell(g^p h^q) = \ell(g^p) + \ell(h^q) + 2D = p \ell(g) + q \ell(h)
962: + 2 D =\\
963: p a + q a + 2 D = (p+q) a + 2 D .
964: \end{gather*}
965: Thus we see that $\ell(g^p h^q)$ depends only on $p+q$ and hence
966: $\ell(g^p h^q) = \ell(g^i h^j)$, as required.
967:
968:
969: If the intersection of $L_g$ and $L_h$ contains a ray then either $g
970: h$ or $g h^{- 1}$ fixes a segment of that ray. This is impossible
971: since the action of $G$ on $X$ has trivial arc stabilizers.
972:
973: Suppose now that $L_g \cap L_h =[x,y]$ and that $d(x,y) > 0$. If the
974: translation directions of $g$ and $h$ on $[x,y]$ coincide then by part
975: (2a) of Proposition~\ref{pau} we have:
976: \[
977: \ell(g^p h^q) = \ell(g^p) + \ell(h^q) = (p+q) a = (i+j)a = \ell(g^i
978: h^j).
979: \]
980: Assume now that $g$ and $h$ translate on $[x,y]$ in the opposite
981: directions.
982:
983: If $a < d (x,y)$ then $g h$ fixes an arc contained in $[x,y]$,
984: yielding a contradiction. Hence $\ell(g) = \ell(h)= a \ge d(x,y)$.
985: Then by part (2b) of Proposition~\ref{pau} we have
986: \begin{gather*}
987: \ell(g^ph^q)=\ell(g^p)+\ell(h^q)-2d(x,y)=\\
988: (p+q)a-2d(x,y)=(i+j)a-2d(x,y)=\ell(g^ih^j).
989: \end{gather*}
990: \end{proof}
991:
992:
993: The following lemma is an elementary exercise, but we provide a proof
994: for completeness.
995:
996: \begin{lem}\label{ab}
997: Let $A$ be a finite abelian group and let $g\in A$ be an element of
998: order bigger than four. Then for any $u\in A$ there is an integer
999: $i$ such that $g^iu$ has order bigger than four.
1000: \end{lem}
1001: \begin{proof}
1002: Note that if $a=bc$ in $A$ then the order of any of these three
1003: elements divides the least common multiple of the orders of the
1004: other two.
1005:
1006: If $u\in \langle g\rangle$, then $u=g^n$ for some $n$ and the
1007: conclusion of the lemma holds with $i=-n-1$. Suppose now that $u$ is
1008: not a power of $g$ and that for every $i$ the order of $g^iu$ is at
1009: most four. Hence for each integer $i$ the order of $g^iu$ is two,
1010: three or four. If $i$ is an integer then by the observation above
1011: the orders of $g^iu$ and $g^{i+1}u$ cannot be both even or both
1012: equal to three. Indeed, in that case the order of $g$ would divide
1013: either three or four, contrary to the assumption that the order of
1014: $g$ is bigger than four. Choose $i$ such that the order of $g^iu$
1015: is three. Then the order of $g^{i+2}u$ is also three and the orders
1016: of $g^{i+1}u, g^{i+3}u$ are both even. Hence the order of
1017: $g^2=g^{i+2}u(g^iu)^{-1}=g^{i+3}u(g^{i+1}u)^{-1}$ divides both three
1018: and four, yielding a contradiction.
1019: \end{proof}
1020:
1021: The following is a generalization of Corollary~\ref{cor:C} from the
1022: Introduction.
1023:
1024: \begin{cor}\label{cor:C'}
1025: Let $G$ be a group.
1026: \begin{enumerate}
1027: \item Suppose that $G$ is nonabelian and that the abelianization of
1028: $G$ contains an element of infinite order. Then for every integer
1029: $M\ge 1$ there exist nontrivial elements $g_1, \dots, g_M \in G$
1030: such that $g_i$ is not conjugate to $g_j^{\pm 1}$ for $i \ne j$ and
1031: such that for every action of $G$ on an $\mathbb{R}$-tree $X$ with
1032: trivial arc stabilizers we have $\ell_X(g_i) = \ell_X(g_j)$.
1033:
1034: \item Suppose that $G$ is nonabelian and that the abelianization of
1035: $G$ contains an element of order bigger than four. Then there exist
1036: nontrivial elements $g_1, g_2\in G$ such that $g_1$ is not conjugate
1037: to $g_2^{\pm 1}$ and such that for every action of $G$ on an
1038: $\mathbb{R}$-tree $X$ with trivial arc stabilizers we have
1039: $\ell_X(g_1) = \ell_X(g_2)$.
1040: \end{enumerate}
1041: \end{cor}
1042: \begin{proof}
1043: Let $\overline{G}$ be the abelianization of $G$. For an element $x
1044: \in G$ we denote by $\overline{x}$ the image of $x$ in
1045: $\overline{G}$.
1046:
1047:
1048: (1) Suppose first that $G$ is nonabelian and that $\overline G$
1049: contains an element of infinite order. Then there exists a
1050: noncentral element $g$ of $G$ whose image has infinite order in
1051: $\overline{G}$. Indeed, suppose not. Then all noncentral elements of
1052: $G$ have finite order images in $\overline{G}$. Take $g_1\in G$ such
1053: that $\overline{g_1}$ has infinite order in $\overline G$. Then by
1054: assumption $g_1$ is central in $G$. Since $G$ is nonabelian, there
1055: exists a noncentral element $u\in G$. Again, by assumption,
1056: $\overline u$ has finite order. But then $g=g_1u$ is noncentral and
1057: has infinite order in $\overline G$, yielding a contradiction.
1058:
1059: Thus we can choose a noncentral element $g\in G$ such that
1060: $\overline g$ has infinite order. Then there is some $f \in G$ such
1061: that $h:=f^{- 1} g^{- 1} f \ne g^{- 1}$. Choose $M \ge 1$ and put
1062: $g_i:= g^i h^{2M+1-i} = g^i f^{- 1} g^{i-2M-1} f$ for $i = 1, \dots,
1063: M$. Then $\overline{g}_i = \overline{g}^{2i - 2M-1}\ne 1$ for $1\le
1064: i\le M$. Since $\overline{g}$ has infinite order in $G$, for $i \ne
1065: j$ we have $\overline{g_i} \ne \overline{g_j}^{\pm 1}$ and therefore
1066: $g_i$ is not conjugate to $g_j^{\pm 1}$ in $G$. Also, since
1067: $\overline{g_i} \ne 1$, it follows that $g_i \ne 1$ for $1 \le i \le
1068: M$. Theorem~\ref{thm-pq} implies that for every action of $G$ on an
1069: $\mathbb{R}$-tree $X$ with trivial arc stabilizers we have
1070: $\ell_X(g_i) = \ell_X(g_j)$. This establishes part (1) of the
1071: corollary.
1072:
1073:
1074: (2) Suppose now that $G$ is nonabelian and that $\overline{G}$ has
1075: an element of order bigger than four. We may assume that
1076: $\overline{G}$ is torsion by part (1).
1077:
1078: We claim that in this case there exists a noncentral element $g$ of
1079: $G$ whose image has order bigger than four in $\overline{G}$. Let
1080: $g_0\in G$ be such that $\overline{g_0}$ has order bigger than four.
1081: If $g_0$ is noncentral, put $g=g_0$. Otherwise, choose a noncentral
1082: $u\in G$ and note that $\langle \overline{g_0}, \overline{u}\rangle$
1083: is a finite abelian group. By Lemma~\ref{ab} there is an integer $i$
1084: such that $\overline{g_0^iu}$ has order at least five and hence
1085: $g=g_0^iu$ is the desired noncentral element.
1086:
1087:
1088:
1089:
1090: Thus let $g\in G$ be a noncentral element such that $\overline{g}$
1091: has order bigger than four. Hence there exists $f\in F$ such that
1092: $h:=f^{-1}g^{-1}f\ne g^{-1}$.
1093:
1094: Put $g_1=g^4h=g^4f^{- 1} g^{-1} f$, $g_2=g^3h^2=g^3f^{- 1} g^{-2}
1095: f$. Then $\overline{g_1}=\overline{g^3}$ and
1096: $\overline{g_2}=\overline{g}$. Since $\overline g$ has order bigger
1097: than four, the elements $\overline{g}, \overline{g}^{-1},
1098: \overline{g^3}$ are nontrivial and pairwise distinct in
1099: $\overline{G}$. Hence $g_2\ne 1$ and $g_1$ is not conjugate to
1100: $g_2^{\pm 1}$ in $G$. Theorem \ref{thm-pq} again implies that for
1101: every action of $G$ on an $\mathbb{R}$-tree $X$ with trivial arc
1102: stabilizers we have $\ell_X(g_1) = \ell_X(g_2)$.
1103: \end{proof}
1104:
1105: \section{Trace identities}
1106:
1107: The following statement is well-known and probably goes back to the
1108: work of Klein in the late 19-th century (see, for
1109: example,~\cite{Ho,Ma} for a proof).
1110:
1111: \begin{lem}\label{klein}
1112: For any freely reduced word $w(a,b)\in F(a,b)$ there exists a
1113: polynomial $f_w\in {\mathbb Z}[x,y,z]$ such that for any field
1114: $\mathbb K$ and any matrices $A,B\in SL(2,\mathbb K)$
1115:
1116: \[
1117: tr \, w(A,B)= f_w(tr \, A, tr \, B, tr\, AB).
1118: \]
1119: \end{lem}
1120:
1121:
1122: We now obtain Theorem~\ref{B} from the Introduction:
1123:
1124: \setcounter{theore}{1}
1125: \begin{theore}
1126: Let $F$ be a finitely generated free group and suppose that
1127: $u\equiv_c v$ in $F$. Then $u\equiv_t v$ in $F$.
1128: \end{theore}
1129:
1130:
1131: \begin{proof}
1132: Choose an arbitrary free basis $A$ of $F$. By Lemma~6.8 of
1133: Horowitz~\cite{Ho}, established via a careful analysis of traces
1134: under an explicit family of representations in $SL(2,\mathbb C)$,
1135: the assumption that $u\equiv_c v$ implies that $||u||_A=||v||_A$.
1136: Since $A$ was an arbitrary free basis of $F$, Theorem~\ref{A}
1137: implies that $u\equiv_{t} v$ in $F$.
1138: \end{proof}
1139:
1140: Lemma~6.8 of Horowitz~\cite{Ho} was strengthened by Southcott (see
1141: Theorem~6.6 of \cite{So}) who proved that character equivalent
1142: elements have essentially the same ``syllable structure'' with respect
1143: to every free basis of $F$.
1144:
1145:
1146: \begin{cor}
1147: Let $F$ be a finitely generated free group and let $u,v\in F$ be
1148: trace equivalent elements. Then for any word $w(x,y)\in F(x,y)$ we
1149: have $tr(w(u,v))=tr(w(v,u))$ and hence, by Theorem~\ref{B}, $w(u,v)$
1150: is translation equivalent to $ w(v,u)$ in $F$.
1151: \end{cor}
1152: \begin{proof}
1153: By Lemma~\ref{klein} there is a polynomial $f(r,s,t)\in \mathbb
1154: Z[r,s,t]$ such that for every $A,B\in SL(2,\mathbb C)$ we have $tr
1155: (w(A,B))=f( tr(A), tr(B), tr(AB))$.
1156:
1157: Let $\alpha: F\to SL(2,\mathbb C)$ be an arbitrary representation.
1158: Put $A=\alpha(u)$ and $B=\alpha(v)$. Thus $tr(A)=tr(B)$ and, of
1159: course, $tr(AB)=tr(BA)$. Therefore $f(tr(A), tr(B), tr(AB))=f(tr(B),
1160: tr(A), tr(BA)$ and hence $tr (w(A,B))= tr (w(B,A))$. Since $\alpha$
1161: was arbitrary, this implies that $tr(w(u,v))=tr(w(v,u))$, as
1162: required.
1163:
1164: \end{proof}
1165:
1166:
1167:
1168: \begin{prop}\label{prop:trace}
1169: Let $w(x,y)\in F(x,y)$ be a freely reduced word, let $\mathbb K$ be
1170: any field and let $A,B\in GL(2,\mathbb K)$ be arbitrary matrices.
1171: Then
1172:
1173: \[
1174: tr\, w(A,B)= tr\, w^R(A,B) \tag{$\dag$}
1175: \]
1176: \end{prop}
1177:
1178: \begin{proof}
1179:
1180: We first will prove $(\dag)$ under the assumption that both $A,B\in
1181: SL(2,\mathbb K)$.
1182:
1183: Let $f_w(x,y,z)$ be the polynomial provided by Lemma~\ref{klein}.
1184: Let $A,B\in SL(2,\mathbb K)$ be arbitrary matrices. Note that
1185: $tr(A)=tr(A^{-1})$, $tr(B)=tr(B^{-1})$, and $tr(AB)=tr (B^{-1}
1186: A^{-1})=tr (A^{-1}B^{-1})$. Then by Lemma~\ref{klein}
1187: \begin{gather*}
1188: tr (w(A,B))=f_w(tr(A), tr(B), tr(AB))=\\
1189: f_w(tr(A^{-1}), tr(B^{-1}), tr(A^{-1}B^{-1}))=tr (w(A^{-1},
1190: B^{-1}))= tr (w^R(A,B)),
1191: \end{gather*}
1192: where the last equality holds because
1193: $w^R(a,b)=[w(a^{-1},b^{-1})]^{-1}$.
1194:
1195: Consider now the general case of $GL(2,\mathbb K)$. Since every field
1196: embeds in an algebraically closed field, it suffices to prove $(\dag)$
1197: for algebraically closed field. So let $\mathbb K$ be an
1198: algebraically closed field, let $w\in F(x,y)$ be a freely reduced word
1199: and let $X,Y\in GL(2,\mathbb K)$ be arbitrary. Since $\mathbb K$ is
1200: algebraically closed, there exist nonzero $r,q\in \mathbb K$ such that
1201: $r^2=\det(X)$ and $q^2=\det(Y)$. Put $X_1:=X/r$ and $Y_1:=Y/q$. Then
1202: $\det(X_1)=\det(Y_1)=1$, so that $X_1,Y_1\in SL(2,\mathbb K)$.
1203: Therefore, by the already established fact, we have
1204:
1205: \[
1206: tr\, w(X_1,Y_1)=tr\, w^R(X_1,Y_1).
1207: \]
1208:
1209: Let $\sigma$ be the exponent sum on $x$ in $w$ (and hence in $w^R$)
1210: and let $\tau$ be the exponent sum on $y$ in $w$ (and hence in $w^R$).
1211: Then $w(X,Y)=w(rX_1,qY_1)=r^{\sigma}q^{\tau}w(X_1,Y_1)$ and, similarly
1212: $w^R(X,Y)=w^R(rX_1,qY_1)=r^{\sigma}q^{\tau}w^R(X_1,Y_1)$.
1213:
1214: Therefore
1215:
1216: \[
1217: tr\, w(X,Y)= r^{\sigma}q^{\tau} tr\, w(X_1,Y_1)=r^{\sigma}q^{\tau}
1218: tr\, w^R(X_1,Y_1)=tr\, w^R(X,Y),
1219: \]
1220: as required.
1221:
1222: \end{proof}
1223:
1224: We refer the reader to~\cite{A} for a detailed discussion on the above
1225: proposition.
1226:
1227: \begin{rem}
1228: Theorem~\ref{B} and Proposition~\ref{prop:trace} immediately imply
1229: Theorem~\ref{C} established earlier by a direct argument.
1230: \end{rem}
1231:
1232:
1233: \begin{prop}\label{tr-powers}
1234: For any field $\mathbb K$, for any integers $p,q$ and for any
1235: conjugate matrices $A,B\in GL(2,\mathbb K)$ we have
1236: \[
1237: tr\, A^p B^q= tr \, A^qB^p.
1238: \]
1239: \end{prop}
1240:
1241: \begin{proof}
1242: Again, we may assume that $\mathbb K$ is algebraically closed. For
1243: any matrices $A,B\in SL(2,\mathbb K)$ conjugate in $GL(2,\mathbb K)$
1244: we have $tr\, A =tr \, B$, and hence $tr\, A^p B^q= tr \, B^pA^p= tr
1245: \, A^qB^p$ by Lemma~\ref{klein} applied to $w(x,y)=x^py^q$.
1246:
1247: Suppose now $A,B\in GL(2,\mathbb K)$ are conjugate. Since $\mathbb
1248: K$ is algebraically closed, there exists $s\in \mathbb K$ such that
1249: $s^2=\det(A)=\det(B)$. Then $A_1=A/s$ and $B_1=B/s$ have determinant
1250: $1$ and are still conjugate in $GL(2,\mathbb K)$. Therefore $tr\,
1251: A_1^p B_1^q= tr \, A_1^qB_1^p$. However $A^p B^q= s^{p+q} A_1^p
1252: B_1^q$ and $A^qB^p=s^{p+q} A_1^q B_1^p$ which implies that $tr\, A^p
1253: B^q= tr \, A^qB^p$, as required.
1254: \end{proof}
1255:
1256:
1257:
1258: \section{Tree actions in the boundary of the Teichm\"uller space}
1259:
1260: Let $S$ be a closed surface of negative Euler characteristic, possibly
1261: non-orientable, and let $\mathcal T(S)$ be the Teichm\"uller space of
1262: $S$. We think of $\mathcal T(S)$ as the set of (isotopy classes of)
1263: marked hyperbolic structures on $S$ or, equivalently, as the set of
1264: (conjugacy classes of) free discrete and cocompact isometric actions
1265: of $G=\pi_1(S)$ on $\mathbb H^2$. Let $\overline{\mathcal T(S)}$ be
1266: the Thurston compactification of $\mathcal T(S)$. The points of
1267: $\mu\in \overline{\mathcal T(S)}-\mathcal T(S)$ are measured
1268: laminations of $S$. Each such lamination $\mu$ defines a small action
1269: of $G$ on an $\mathbb R$-tree dual $X_{\mu}$ to the lift of this
1270: lamination to $\mathbb H^2$~\cite{MS84,MS88}. We refer the reader to
1271: {\cite{Be,Ka}} for a detailed discussion of this topic.
1272:
1273: We next show that our results regarding two-variable palindromes also
1274: apply to the elements of $\overline{\mathcal T(S)}$.
1275:
1276:
1277: \begin{thm}\label{pal-t}
1278: Let $S$ be a closed surface (possibly non-orientable) of negative
1279: Euler characteristic and let $G=\pi_1(S)$. Then for any $g,h\in G$
1280: and for any $w(x,y)\in F(x,y)$ we have:
1281:
1282: \begin{enumerate}
1283: \item For each point $p\in \mathcal T(S)$, thought of as an action of
1284: $G$ on $\mathbb H^2$, the elements $w(g,h)$ and $w^R(g,h)$ have
1285: equal translation lengths.
1286: \item For each point $\mu\in \overline{\mathcal T(S)}-\mathcal T(S)$
1287: we have
1288: \[
1289: \ell_{X_{\mu}}(w(g,h))=\ell_{X_{\mu}}(w^R(g,h)).
1290: \]
1291: \end{enumerate}
1292:
1293: \end{thm}
1294:
1295: \begin{proof}
1296: Let $p\in \mathcal T(S)$ and consider the corresponding action
1297: $\phi:G\to Isom({\mathbb H}^2)$ of $G$ by isometries on $\mathbb
1298: H^2$. Recall that, when $\mathbb H^2$ is considered in the
1299: upper-half space model, there is a canonical isometric action of
1300: $GL(2,\mathbb R)$ on $\mathbb H^2$ whose image is the full isometry
1301: group of $\mathbb H^2$. Namely, let $A=\begin{pmatrix} a & b\\ c &
1302: d\end{pmatrix}\in GL(2,\mathbb R)$. If $\det(A)>0$ then
1303: $Az=\frac{az+b}{cz+d}$ for any $z\in \mathbb H^2$. If $\det(A)<0$
1304: then $Az=\frac{a\bar z+b}{c\bar z+d}$ for any $z\in \mathbb H^2$.
1305: The first case gives us an orientation-preserving isometry of
1306: $\mathbb H^2$ and the second case gives an orientation-reversing
1307: isometry. It is well-known that for $A\in GL(2,\mathbb R)$ the
1308: trace $tr(A)$ and the determinant $\det(A)$ uniquely determine the
1309: translation length of $A$ as an isometry of the hyperbolic plane.
1310:
1311:
1312:
1313: Let $w(x,y)\in F(x,y)$ and let $g,h\in G$. Let $A\in GL(2,\mathbb
1314: R)$ represent $\phi(g)$ and let $B\in GL(2,\mathbb R)$ represent
1315: $\phi(h)$. Then by Proposition~\ref{prop:trace} $tr\, w(A,B)=tr \,
1316: w^R(A,B)$. Moreover, $\det w(A,B)= \det w^R(A,B)$. Therefore we
1317: have $\ell_{\mathbb H^2}(w(g,h))=\ell_{\mathbb H^2}(w^R(g,h))$, as
1318: claimed.
1319:
1320: Thus part (1) of the theorem is established.
1321:
1322:
1323: Now, part (1) immediately implies part (2). Indeed, recall that each
1324: element $p\in \mathcal T(S)$ determines a \emph{marked length
1325: spectrum} $\ell_p: G\to \mathbb R$ where $\ell_p(g)$ is the
1326: translation length of $g$ for the isometric action of $G$ on
1327: $\mathbb H^2$ corresponding to $p$. Then it is well-known that for
1328: any $\mu\in \overline{\mathcal T(S)}-\mathcal T(S)$ the
1329: length-function $\ell_{X_{\mu}}:G\to \mathbb R$ is projectively the
1330: limit of marked length spectra $\ell_{p_n}$ for some sequence of
1331: $p_n\in \mathcal T(S)$. That is, there exists a sequence of scalars
1332: $\lambda_n>0$ such that for every $f\in G$
1333: \[
1334: \ell_{X_{\mu}}(f)=\lim_{n\to\infty} \lambda_n \ell_{p_n}(f).
1335: \]
1336: Since for $f=w(g,h)$ and $f'=w^R(g,h)$ we know by (1) that
1337: $\ell_{p_n}(f)=\ell_{p_n}(f')$ for all $n$, it follows that
1338: $\ell_{X_{\mu}}(f)=\ell_{X_{\mu}}(f')$, as required.
1339: \end{proof}
1340:
1341: Together with Theorem~\ref{pal-t} the following result immediately
1342: implies Theorem~\ref{E} from the Introduction:
1343:
1344: \begin{thm}\label{surface}
1345: Let $S$ be a closed surface of negative Euler characteristic,
1346: possibly non-orientable, and let $G=\pi_1(S)$. Then for any
1347: conjugate $g,h\in G$ and for any integers $p,q$ we have:
1348:
1349: \begin{enumerate}
1350: \item For each point $p\in \mathcal T(S)$, thought of as an action of
1351: $G$ on $\mathbb H^2$, the elements $g^ph^q$ and $g^qh^p$ have equal
1352: translation lengths.
1353: \item For each point $\mu\in \overline{\mathcal T(S)}-\mathcal T(S)$
1354: we have
1355: \[
1356: \ell_{X_{\mu}}(g^ph^q)=\ell_{X_{\mu}}(g^qh^p).
1357: \]
1358: \item If $S$ is orientable then for any conjugate elements $g,h \in
1359: G$, for each point $\mu\in \overline{\mathcal T(S)}-\mathcal T(S)$
1360: and for any positive integers $p,q,i,j$ such that $p+q=i+j$ we have
1361: \[
1362: \ell_{X_{\mu}}(g^ph^q)=\ell_{X_{\mu}}(g^ih^j).
1363: \]
1364: \end{enumerate}
1365:
1366: \end{thm}
1367: \begin{proof}
1368: Parts (1) and (2) are established exactly as Theorem~\ref{pal-t},
1369: but using Proposition~\ref{tr-powers} instead of
1370: Proposition~\ref{prop:trace}.
1371:
1372: To see that part (3) holds observe that, by well-known results, if
1373: $S$ is orientable then every orbit of the action of the mapping
1374: class group of $S$ on $\overline{\mathcal T(S)}-\mathcal T(S)$ is
1375: dense in $\overline{\mathcal T(S)}-\mathcal T(S)$. In particular,
1376: this applies to the orbit of a point corresponding to the stable
1377: foliation of a pseudo-Anosov homeomorphism of $S$. Therefore the set
1378: of those $\mu\in \overline{\mathcal T(S)}-\mathcal T(S)$, such that
1379: the action of $G$ on the tree $X_{\mu}$ is free, is dense in
1380: $\overline{\mathcal T(S)}-\mathcal T(S)$. Since
1381: Theorem~\ref{thm-pq} applies to free actions, part (3) of
1382: Theorem~\ref{surface} follows.
1383: \end{proof}
1384:
1385: \begin{rem}
1386: The argument used in the proof of part (3) of Theorem~\ref{surface}
1387: does not work in the case of non-orientable surfaces, as follows
1388: from the results of Danthony and Nogueira~\cite{DN}. At the moment
1389: we do not know how to prove part (3) of Theorem~\ref{surface} in the
1390: non-orientable case. The problem is that if $S$ is nonorientable
1391: and $\mu\in \overline{\mathcal T(S)}-\mathcal T(S)$ then the action
1392: of $G=\pi_1(S)$ on $X_{\mu}$ need not be very small. Note, however,
1393: that $G$ has an index two subgroup whose action on $X_{\mu}$ is very
1394: small. Note also that, as Example~\ref{distinct} below shows, part
1395: (1) of Theorem~\ref{thm-pq} does not have a precise analogue for
1396: $SL_2$ trace identities. Hence it is not possible to argue as in the
1397: proof of Theorem~\ref{pal-t} to establish part (3) of
1398: Theorem~\ref{surface} for non-orientable surfaces.
1399: \end{rem}
1400:
1401:
1402: \section{Examples}
1403:
1404:
1405: \begin{exmp}
1406: The palindrome and the $g^ph^q$ phenomena described above no longer
1407: hold for the class of small actions (as opposed to free or very
1408: small actions).
1409:
1410: \begin{figure}[here]
1411:
1412: \psfrag{<a^2>}{$\langle a^2\rangle$} \psfrag{<a>}{$\langle
1413: a\rangle$} \psfrag{<1>}{$\langle 1\rangle$} \psfrag{e}{$e$}
1414: \psfrag{t}{$t$} \psfrag{f}{$f$} \epsfbox{ex.eps}
1415: \caption{Counterexample}\label{Fi1}
1416:
1417: \end{figure}
1418:
1419:
1420: Consider the graph of groups $\mathbb A$ shown in Figure~\ref{Fi1}.
1421: Let $v$ be the vertex of the graph incident to all three marked up
1422: edges $e,f,t$. Put $G=\pi_1({\mathbb A},v)$ and let
1423: $X=\widetilde{({\mathbb A},v)}$ be the Bass-Serre universal covering
1424: tree. It is easy to see that the action of $G$ on $X$ is minimal and
1425: that $G=F(x,y)$ is a free group of rank two with free basis $x=eae,
1426: y=tf$. Consider now the elements $g=xyx^2y^{-1}$ and $h=y^{-1}x^2yx$.
1427: Thus $h$ is the palindromic reverse of $g$ in $F(x,y)$. Moreover,
1428: with $a=x$ and $b=yxy^{-1}$ we see that $g=ab^2$ and $h$ is conjugate
1429: to $a^2b$.
1430:
1431:
1432: In $G$ we have:
1433: \begin{gather*}
1434: g=xyx^2y^{-1}=eae^{-1}tfea^2e^{-1}f^{-1}t^{-1}=eae^{-1}tfa^2f^{-1}t^{-1}
1435: =eaeta^2t^{-1}\\
1436: h=y^{-1}x^2yx=f^{-1}t^{-1}ea^2e^{-1}tfeae^{-1}=f^{-1}t^{-1}a^2tfeae^{-1}.
1437: \end{gather*}
1438: Both $eaeta^2t^{-1}$ and $f^{-1}t^{-1}a^2tfeae^{-1}$ are cyclically
1439: reduced closed paths in $\mathbb A$. Therefore $\ell_X(g)=4$ while
1440: $\ell_X(h)=6$.
1441: \end{exmp}
1442:
1443: \begin{rem}
1444: We believe that the palindrome phenomenon also holds for small
1445: actions of a free group $F$ such that, if $g\in F$ is nontrivial and
1446: fixes more than one point, then $Fix(g)$ is a compact segment
1447: containing no branch point in its interior.
1448: \end{rem}
1449:
1450: \begin{exmp}
1451: We have remarked earlier that translation equivalence is preserved
1452: by injective homomorphisms, and hence by passing to larger free
1453: groups containing the given free group as a subgroup. However,
1454: passing to subgroups, even of finite index, no longer preserves
1455: translation equivalence in general.
1456:
1457: Let $F=F(a,b)$ and let $H=\langle a, bab^{-1}, b^2\rangle\le F$. It
1458: is not hard to see that $H$ is a subgroup of index two in $F$ and
1459: that $x=a, y=bab^{-1}, z=b^2$ is a free basis of $H$, so that
1460: $H=F(x,y,z)$. Consider the elements $g=aba^2b^{-1}$ and
1461: $h=a^2bab^{-1}$ in $F$. Then $g\equiv_{t} h$ in $F$ by, for
1462: example, Theorem~\ref{thm-pq}. We also have $g,h\in H$ and $g=xy^2,
1463: h=x^2y$. Clearly, $g$ and $h$ have different Whitehead graphs in
1464: $F(x,y,z)$ and therefore $g\not\equiv_{t} h$ in $H$.
1465:
1466:
1467: On the other hand, it is easy to see that if $H$ is a free factor of
1468: $F$ and $g,h\in H$ then $g\equiv_{t} h$ in $F$ if and only if
1469: $g\equiv_{t} h$ in $H$.
1470: \end{exmp}
1471:
1472: \begin{exmp}
1473: Theorem~\ref{A} states, in particular, that two elements of $F$ are
1474: translation equivalent in $F$ if and only if they have equal
1475: Whitehead graphs with respect to each free basis of $F$. This
1476: statement no longer holds if we consider the obvious ``higher rank"
1477: analogues of Whitehead graphs where symmetrized numbers of
1478: occurrences of subwords of length $m > 2$ are recorded. Again
1479: consider $g=aba^2b^{-1}$ and $h=a^2bab^{-1}$ in $F(a,b)$. We
1480: already know that $g\equiv_{t} h$ in $F$. However, consider the
1481: cyclic words $w$ and $u$ defined by $g$ and $h$ accordingly. The
1482: number of occurrences of $b^{-1}ab$ in $w$ is equal to $1$, while
1483: neither $b^{-1}ab$ nor its inverse $b^{-1}a^{-1}b$ occur in $u$.
1484: \end{exmp}
1485:
1486:
1487: \begin{exmp}\label{distinct}
1488: Theorem~\ref{B} shows that in free groups character equivalence
1489: implies translation equivalence. However, the converse implication
1490: does not hold and these two phenomena are different. For example, we
1491: know from Theorem~\ref{D} that if $g$ and $h$ are conjugate in $F$
1492: and $g\ne h^{-1}$ then \newline $g^3h\equiv_{t} g^2h^2$ in $F$. On
1493: the other hand, let
1494: \[
1495: A=\begin{bmatrix}
1496: 2 & 1\\
1497: 0 & 1/2
1498: \end{bmatrix},\quad
1499: B=\begin{bmatrix} 1 & 0\\ 2 & 1\end{bmatrix},\quad
1500: C=BAB^{-1}=\begin{bmatrix} 0 & 1\\ -1 & 5/2\end{bmatrix}
1501: \]
1502: be matrices in $SL(2,\mathbb R)$. A direct computation shows that
1503: $tr\, A^3C=-79/16$ while $tr \, A^2C^2=-143/16$.
1504:
1505:
1506: Thus $a^3bab^{-1}\not\equiv_c a^2 ba^2b^{-1}$ while
1507: $a^3bab^{-1}\equiv_t a^2 ba^2b^{-1}$ in $F(a,b)$ and there is no
1508: precise analogue of Theorem~\ref{D} for character equivalence.
1509: \end{exmp}
1510:
1511: \section{Open Problems}
1512:
1513:
1514: \begin{prob}
1515: Is there an algorithm which, when given two elements in a finitely
1516: generated free group, decides whether or not they are translation
1517: equivalent?
1518: \end{prob}
1519:
1520: It can be deduced from results of Leininger~\cite{Le} that hyperbolic
1521: equivalence in surface groups is algorithmically decidable by using
1522: standard commutative algebra techniques applied to the representation
1523: variety of the surface group. Similarly, one can algorithmically
1524: decide whether or not two elements of a free group are character
1525: equivalent.
1526:
1527: \begin{prob}
1528: Find other sources of translation equivalence in free groups,
1529: different from those discussed in this paper.
1530: \end{prob}
1531:
1532:
1533: \begin{prob}
1534: Is it true that whenever $g\equiv_t h$ in $F$ and $w(x,y)\in F(x,y)$
1535: is arbitrary then $w(g,h)\equiv_t w(h,g)$ in $F$? It easily follows
1536: from Lemma~\ref{klein} that $g\equiv_c h$ (e.g. $g$ is conjugate to
1537: $h$) implies $w(g,h)\equiv_c w(h,g)$ in $F$ and hence
1538: $w(g,h)\equiv_t w(h,g)$ in $F$.
1539: \end{prob}
1540:
1541: The notion of translation equivalence has several natural
1542: generalizations.
1543:
1544: \begin{prob}[Bounded translation equivalence]
1545: For nontrivial $g,h\in F$ we say that $g$ is \emph{boundedly
1546: translation equivalent} to $h$, denoted $g\equiv_b h$, if there is
1547: $C>0$ such that for every free and discrete action of $F$ on an
1548: $\mathbb R$-tree $X$ we have
1549: \[
1550: \frac{1}{C}\le \frac{\ell_X(g)}{\ell_X(h)}\le C.
1551: \]
1552:
1553: What are the sources of bounded translation equivalence in free groups
1554: and how much more general is it compared to translation equivalence?
1555: What are the sources of the failure of bounded translation
1556: equivalence? Is bounded translation equivalence, in some natural
1557: sense, generic? Is bounded translation equivalence algorithmically
1558: decidable?
1559: \end{prob}
1560: A similar notion can be defined in the context of hyperbolic metrics
1561: on closed surfaces and the above questions make sense there as well.
1562:
1563: \begin{prob}[Volume equivalence of subgroups]
1564: If $G$ is a finitely generated group acting discretely isometrically
1565: and without a global fixed point on an $\mathbb R$-tree $X$, we
1566: denote by $vol_X(G)$ the sum of the lengths of the edges of the
1567: metric graph $X_G/G$, where $X_G$ is the unique minimal
1568: $G$-invariant subtree.
1569:
1570: We will say that nontrivial finitely generated subgroups $H,K$ of
1571: $F$ are \emph{volume equivalent} in $F$, denoted $H\equiv_v K$, if
1572: for every free and discrete action of $F$ on an $\mathbb R$-tree $X$
1573: we have $vol_X(H)=vol_X(K)$.
1574:
1575: Thus $g\equiv_t h$ in $F$ iff $\langle g\rangle \equiv_v \langle h
1576: \rangle$ in $F$. Moreover, if $H,K$ have the same finite index $n$
1577: in $F$ then it is easy to see that $H\equiv_v K$ in $F$.
1578:
1579: Are there any other sources of volume equivalence in free groups? If
1580: $H\equiv_v K$ in $F$, does this imply that $H$ and $K$ are free
1581: groups of the same rank? Is volume equivalence algorithmically
1582: decidable?
1583: \end{prob}
1584:
1585: Again, the notion of volume equivalence and the above questions also
1586: make sense in the context of hyperbolic surfaces. In that situation
1587: one should consider free discrete cocompact isometric actions of
1588: $G=\pi_1(S)$ on $\mathbb H^2$. For every such action $\phi$ and for a
1589: finitely generated non-cyclic subgroup $H$ of $G$ we define
1590: $vol_{\phi}(H)$ as the hyperbolic volume of the compact surface
1591: $Conv(\Lambda H)/H$ where $Conv(\Lambda H)$ is the convex hull in
1592: $\mathbb H^2$ of the limit set $\Lambda H$ of $H$.
1593:
1594: \begin{thebibliography}{ABC}
1595:
1596: \bibitem{A} J.~Anderson, \emph{Variations on a theme of Horowitz.}
1597: Kleinian groups and hyperbolic 3-manifolds (Warwick, 2001),
1598: 307--341, London Math. Soc. Lecture Note Ser., 299, Cambridge Univ.
1599: Press, Cambridge, 2003
1600:
1601:
1602: \bibitem{Be} M.~Bestvina, \emph{Degenerations of the hyperbolic
1603: space.} Duke Math. J. \textbf{56} (1988), no. 1, 143--161
1604:
1605: \bibitem{BF93} M. Bestvina and M. Feighn, \emph{Outer Limits},
1606: preprint, 1993.
1607:
1608: \bibitem{Bo}
1609: F.~Bonahon, \emph{The geometry of Teichm\"uller space via geodesic
1610: currents.} Invent. Math. \textbf{92} (1988), no. 1, 139--162
1611:
1612: \bibitem{CL} M. Cohen and M. Lustig, \emph{Very small group actions on
1613: $R$-trees and Dehn twist automorphisms.} Topology \textbf{34}
1614: (1995), no. 3, 575--617
1615:
1616: \bibitem{CLS} M.~Cohen, M.~Lustig and M.~Steiner, \emph{$R$-tree
1617: actions are not determined by the translation lengths of finitely
1618: many elements.} Arboreal group theory (Berkeley, CA, 1988),
1619: 183--187, Math. Sci. Res. Inst. Publ., \textbf{19}, Springer, New
1620: York, 1991
1621:
1622:
1623: \bibitem{CV} M. Culler and K. Vogtmann, \emph{Moduli of graphs and
1624: automorphisms of free groups.} Invent. Math. \textbf{84} (1986),
1625: no. 1, 91--119
1626:
1627: \bibitem{DN} C.~Danthony and A.~Nogueira, \emph{Measured foliations on
1628: nonorientable surfaces.} Ann. Sci. \'Ecole Norm. Sup. (4)
1629: \textbf{23} (1990), no. 3, 469--494
1630:
1631: \bibitem{Ho} R. Horowitz, \emph{Characters of free groups represented
1632: in the two-dimensional special linear group.} Comm. Pure Appl.
1633: Math. \textbf{25} (1972), 635--649
1634:
1635: \bibitem{Ka04}
1636: I.~Kapovich,
1637: \emph{Currents on free groups}, preprint, 2004;\\
1638: http://arxiv.org/abs/math.GR/0412128
1639:
1640: \bibitem{Ka} M. Kapovich, \emph{Hyperbolic manifolds and discrete
1641: groups.} Progress in Mathematics, 183. Birkh\"auser Boston, Inc.,
1642: Boston, MA, 2001
1643:
1644: \bibitem{Le} C. J. Leininger, \emph{Equivalent curves in surfaces.}
1645: Geom. Dedicata \textbf{102} (2003), 151--177
1646:
1647: \bibitem{LS} R. C. Lyndon and P. Schupp, \emph{Combinatorial Group
1648: Theory}, Springer-Verlag, 1977. Reprinted in the Classics in
1649: Mathematics series, Springer-Verlag, 2000.
1650:
1651: \bibitem{Ma} W. Magnus, \emph{Rings of Fricke characters and
1652: automorphism groups of free groups.} Math. Z. \textbf{170} (1980),
1653: no. 1, 91--103.
1654:
1655: \bibitem{MS84} J. Morgan and S. Shalen, \emph{Valuations, trees, and
1656: degenerations of hyperbolic structures. I.} Ann. of Math. (2) 120
1657: (1984), no. 3, 401--476.
1658:
1659: \bibitem{MS88} J. Morgan and S. Shalen, \emph{Degenerations of
1660: hyperbolic structures. III. Actions of $3$-manifold groups on
1661: trees and Thurston's compactness theorem.} Ann. of Math. (2)
1662: \textbf{127} (1988), no. 3, 457--519
1663:
1664: \bibitem{SV} J. Smillie and K. Vogtmann, \emph{Length functions and
1665: outer space.} Michigan Math. J. \textbf{39} (1992), no. 3,
1666: 485--493
1667:
1668: \bibitem{So} J.~B.~Southcott, \emph{Trace polynomials of words in
1669: special linear groups.} J. Austral. Math. Soc. Ser. A \textbf{28}
1670: (1979), no. 4, 401--412
1671:
1672: \bibitem{Pau} F. Paulin, \emph{The Gromov topology on $R$-trees.}
1673: Topology Appl. \textbf{32} (1989), no. 3, 197--221
1674:
1675: \bibitem{Ra} B. Randol, \emph{The length spectrum of a Riemann surface
1676: is always of unbounded multiplicity.} Proc. Amer. Math. Soc.
1677: \textbf{78} (1980), no. 3, 455--456
1678:
1679: \end{thebibliography}
1680:
1681: \end{document}
1682: