1: \input amstex
2: \input psfig.sty
3: \documentstyle{amsppt}
4: \topmatter
5: \title
6: Injectivity Radius and Fundamental Groups of Hyperbolic 3-Manifolds
7: \endtitle
8: \rightheadtext{Injectivity Radius and Fundamental Groups}
9: \author
10: Matthew E. White
11: \endauthor
12: \abstract
13: It is shown that for each integer $n>1$ there exists a constant
14: $R_n>0$ such that if $M$ is a closed hyperbolic 3-manifold with
15: Rank $\pi_1(M) = n$, then the injectivity radius of $M$ is bounded above
16: by $R_n$.
17: \endabstract
18: \date
19: 26 October 1998
20: \enddate
21: \endtopmatter
22: \document
23: \head 1. Introduction
24: \endhead
25:
26: Thurston's Geometrization conjecture implies that every closed irreducible
27: 3-manifold with infinite fundamental group containing no ${\Bbb Z}\oplus
28: {\Bbb Z}$
29: subgroup is hyperbolic. Thus, it is of interest to obtain information about the
30: hyperbolic metric from purely topological data. For example, the Gromov norm,
31: which is defined homologically, turns out to be a certain scalar multiple of
32: volume if the manifold is hyperbolic [Th]. Similarly, there is also an upper
33: bound on the volume in terms of length of any presentation of its fundamental
34: group [C]. The {\bf injectivity radius} of a closed hyperbolic 3-manifold $M$,
35: denoted $inj(M)$, is one half the minimum length of all essential loops in $M$.
36: This is equivalent to the usual definition from differential geometry (see
37: [DC]). Recall that the {\bf rank} of a finitely generated group is the minimum
38: number of elements required to generate the group. In this paper we give an
39: upper bound on the injectivity radius in terms of fundamental group rank. It
40: is clear that no such lower bound exists since Dehn-filling of a hyperbolic
41: manifold provides examples of bounded rank with injectivity radius approaching
42: zero. In a subsequent paper, we hope to provide a lower bound on injectivity
43: radius in terms of other group theoretic data. The basic outline of the proof
44: in our present setting is the following: given a closed hyperbolic 3-manifold
45: $M$, we construct a graph of minimal length which carries $\pi_1(M)$. By
46: considering the graph's preimage in hyperbolic space, we show that if the
47: injectivity radius is sufficiently large, any relation in the fundamental group
48: forces some of the graph's edges into close proximity. This in turn allows
49: modification of the graph to reduce its length while still carrying fundamental
50: group. We therefore show that, in fact, $\pi_1(M)$ is a free group of finite
51: rank. A closed hyperbolic 3-manifold is irreducible. Therefore, its
52: fundamental group cannot be a free product. This contradiction proves:
53: \proclaim{Theorem 4.4} For each integer $n>1$ there exists a universal constant
54: $R_n>0$ such that if $M$ is a closed hyperbolic 3-manifold and Rank $\pi_1(M) =
55: n$, then $inj(M) < R_n$. \endproclaim \noindent Residual finiteness of closed
56: hyperbolic 3-manifolds then provides us with: \proclaim{Corollary 4.5} Given a
57: closed hyperbolic 3-manifold $M$ and an integer $n>0$, there exists a finite
58: sheeted cover $$p:\tilde M\longrightarrow\ M$$ with $Rank\ \pi_1 (\tilde M) \ge
59: n$. \endproclaim
60:
61: We shall use $M$ to denote a closed, connected hyperbolic 3-manifold.
62: In particular, $M$ is isometric to a quotient of ${\Bbb H}^3$ by a group
63: of loxodromic isometries. We say that a connected finite graph $\Gamma$
64: is an {\bf $n$-graph} if all vertices of $\Gamma$ are trivalent and
65: $Rank\ \pi_1(\Gamma) = n$. A {\bf carrier $n$-graph} for a manifold $M$
66: is an $n$-graph $\Gamma$ together with a map $f:\Gamma\longrightarrow M$
67: such that $f_*:\pi_1(\Gamma)\longrightarrow \pi_1(M )$ is an
68: epimorphism. In this case, the {\bf length} of $\Gamma$ in $M $ is
69: defined by
70: $$\ell(f(\Gamma)) = \sum_{e\ \text{an edge of}\ \Gamma} |e| $$
71: where $|e|$ is the length of $e$ measured by pulling back the path
72: metric of $M $ to $\Gamma$. When we refer to a
73: carrier $n$-graph for a given manifold $M$, we
74: implicitly assume that $n = Rank\ \pi_1(M)$.
75:
76: This paper is organized as follows: In section 2, we prove some
77: necessary technical results using trivalent graphs. Section 3 contains
78: the $n=2$ case of the main theorem; this was previously known to C.
79: Adams. The main theorem and its corollary are
80: proved in section 4. We remark here that our
81: results extend naturally to the case of bounded
82: negative curvature.
83:
84: \head 2. Trivalent Graphs of Minimal Length
85: \endhead
86: \medskip
87: \proclaim{Proposition 2.0}
88: For each integer $n>1$, the set of $n$-graphs is finite. Each $n$-graph
89: has $3(n-1)$ edges.
90: \endproclaim
91: \medskip
92: \noindent {\bf Proof} If $\Gamma$ is an $n$-graph, then $$\chi(\Gamma) = 1-n =
93: (\text{number of vertices}) - (\text{number of edges})$$. Note that twice the
94: number of edges equals three times the number of vertices since all vertices of
95: $\Gamma$ are trivalent. Therefore, there are $3(n-1)$ edges. $\square$
96: \medskip
97: \noindent Let $e_1\dots e_{3n-1}$ be the oriented edges of $\Gamma$.
98: Given a loop in $\Gamma$, its homotopy class may be represented by a {\bf
99: directed edge path}. This is a loop in $\Gamma$ which is a product of
100: homeomorphisms $\gamma_i :[0,1]\longrightarrow e_i$ with the property that for
101: each $i\ge 2$, we have $\gamma_{i-1}(1)=\gamma_i(0)$. In our case, we shall
102: also assume that the maps $\gamma_i$ are parameterized proportionally to arc
103: length. Although a slight abuse of notation, the standard convention is to
104: denote a directed edge path as $$(e_{i_1}^{\pm 1},e_{i_2}^{\pm
105: 1},\dots,e_{i_k}^{\pm 1})$$ with $e_i^{-1}$ denoting $e_i$ with the opposite
106: orientation. A directed edge path that is a closed loop is called {\bf
107: reduced} if $e_{i_m} \ne e_{i_{m-1}}^{-1}$ for any index $i$ and $e_{i_1}\ne
108: e_{n_k}^{-1}$. For further details, see [M].
109:
110: \medskip
111: \proclaim{Proposition 2.1} Let $f:\Gamma\longrightarrow M $ be a carrier $n$-graph. If $\alpha$ is a simple closed curve in
112: $\Gamma$, then $f_*($[$\alpha$]$) \ne 0$. In particular,
113: $\ell(f(\alpha)) > inj(M)$.
114: \endproclaim
115: \medskip\noindent {\bf Proof} Suppose $\alpha$ is a simple closed curve
116: in $\Gamma$. Then $\alpha$ represents a generator of $\pi_1(\Gamma)$.
117: Therefore, $f_*($[$\alpha$]$) \ne 0$ as otherwise we could generate
118: $\pi_1(M)$ with fewer than $n$ elements. Moreover, the length of any
119: essential loop in $M $ is at least $2 inj(M)$. $\square$
120:
121: \proclaim{Lemma 2.2}
122: A closed hyperbolic 3-manifold $M$ has a carrier $n$-graph $\Gamma$ of
123: minimal length. The edges of $\Gamma$ map to geodesic segments in $M$.
124: \endproclaim
125: \medskip\noindent {\bf Proof}
126: We shall omit much detail since the proof closely follows the standard proof
127: of the existence of closed geodesics in free homotopy classes (see [DC]). Fix
128: $M$ and let $\Gamma$ be an $n$-graph. We may regard $\Gamma$ as a compact
129: subset of $\Bbb R^k$ with each edge smoothly embedded. Put
130: $$d = \text{inf }\lbrace\ell (f(\Gamma))\ \vert \
131: f:\Gamma\longrightarrow M \ ,\ f_*\text{ is an epimorphism}\rbrace.$$
132: \noindent
133: Now suppose that $f:\Gamma\longrightarrow M$ is a carrier $n$-graph. Suppose
134: also that there is a loop in $\Gamma$ containing exactly one edge. By
135: Proposition 2.1, the restriction of $f$ to this edge must lift to a path in
136: ${\Bbb H}^3$ with distinct endpoints. Therefore, if the restriction of $f$ to
137: an arbitrary edge of $\Gamma$ is not constant, it must lift to a path in ${\Bbb
138: H}^3$ with distinct endpoints. This means we may construct a family of maps
139: $F=\lbrace f_i:\Gamma\longrightarrow M \rbrace_{i=1}^\infty$ such that for each
140: $i\ge 1$, $f_{i*}$ is an epimorphism, $f_i$ is geodesic on each edge of
141: $\Gamma$, and $\ell(f_i(\Gamma)) \longrightarrow d$. Since $M$ is compact, a
142: straightforward Arzela-Ascoli argument then shows that the closure of $F$ is
143: compact in ${\Cal C}[\Gamma,M]$. Thus, there exists a subsequence ${f_j}$ of
144: $F$ which converges uniformly to a continuous map $f:\Gamma\longrightarrow M$.
145: Uniform convergence implies that $f$ maps every edge to a geodesic segment and
146: that $\ell(f(\Gamma)) = d$. We show that $f_*$ is an epimorphism. Choose a
147: vertex $x_0\in\Gamma$. Given $\epsilon$ small enough, cover each edge of
148: $f(\Gamma)$ with finitely many $\epsilon$ balls such that each is isometric to
149: a ball in ${\Bbb H}^3$. Since the convergence of the $f_j$ is uniform, by
150: picking $j$ large enough, we can ensure that $f_j$ maps into this covering.
151: Now $f_{j*}$ is an epimorphism. Thus, given $[\beta]\in\pi_1(M,f_j(x_0))$, let
152: $[\alpha]\in\pi_1(\Gamma,x_0)$ be such that $f_{j*}([\alpha])=[\beta]$. Note
153: that every homotopy class in $\pi_1(\Gamma,x_0)$ is represented by a directed
154: edge path in $\Gamma$. Hence, we assume that $\alpha$ is a directed edge path.
155: In each ball, we may homotop $f_j$ to $f$. This means $f(\alpha)$ is freely
156: homotopic to $f_j(\alpha)$. In particular, we have that $f_*$ is an
157: epimorphism. By proposition 2.0, there are finitely many $n$-graphs. Hence we
158: many repeat the above proof for each distinct $n$-graph and then select the
159: graph of minimal length.$\square$ \medskip Minimal length carrier graphs have a
160: very nice symmetry property which we shall later put to good use. \medskip
161: \proclaim{Definition 2.3} A subset of a hyperbolic 3-manifold is a {\bf
162: Y-subset} if it is isometric to a subset ${\Cal S}$ of ${\Bbb H}^3$ with the
163: following properties.\newline \noindent (i) ${\Cal S}$ is composed of three
164: geodesic arcs which meet only at a common endpoint.\newline \noindent (ii) Each
165: pair of geodesic arcs in ${\Cal S}$ meets the common endpoint with incidence
166: angle $120^\circ$. \endproclaim \noindent It follows at once that a Y-subset is
167: planar. \medskip \proclaim{Lemma 2.4} Given a manifold $M $ with Rank
168: $\pi_1(M)\ge 2$, there exists a minimal length carrier $n$-graph
169: $f:\Gamma\longrightarrow M $ with the following property: each vertex $v$ of
170: $\Gamma$ has a neighborhood N($v$) such that $f$ maps $N(v)$ isometrically to a
171: Y-subset of $M$. In particular, $f(v)$ is the common endpoint of the geodesic
172: edges in $f(N(v))$. Furthermore, every edge has non-zero length.\endproclaim
173: \medskip \noindent {\bf Proof} We assume that $f:\Gamma\longrightarrow M $ is a
174: fixed minimal length carrier $n$-graph with the property that $f$ is an
175: arc-length parameterization on each edge of $\Gamma$. Let $v$ be a vertex of
176: $\Gamma$. Suppose first that no edge of $\Gamma$ has length 0 under $f$.
177: Choose a small neighborhood $W$ of $f(v)$ such that $W$ is isometric to an open
178: ball in ${\Bbb H}^3$. The edges of $\Gamma$ incident at $v$ map to geodesic
179: edges in $M$. If any pair of these edges intersect in $W$ at a point other
180: than $f(v)$, then the image of one edge is contained in that of the other edge.
181: It is easy to see that in this case we can modify the graph to reduce its
182: length. Therefore, the images of the three edges intersect only at $f(v)$ in
183: $W$. It also follows that these three edges are planar inside $W$, since
184: otherwise orthogonal projection to a hyperbolic plane reduces the length of
185: $\Gamma$. We now suppose there exists a pair of edges with incidence angle
186: less than $120^\circ$. We modify the map $f$ near $v$ as pictured in Figure
187: 2.0. A short hyperbolic trigonometry calculation then shows that this reduces
188: total length.
189:
190: \medskip
191: \psfig{file=fig20h.eps,height=1.5in,width=5.0in}
192: Figure 2.0. Adjustment near a vertex: $2c > a + 2b$.
193: \medskip
194:
195: \noindent Thus, every incidence angle is at least $120^\circ$. Of course, the
196: trivalent angle sum is exactly $360^\circ$. To complete the proof in the
197: general case, we show that if there exists an edge $e$ of $\Gamma$ with length
198: 0, then we may modify $f$ to reduce total length. Suppose such an edge $e$
199: exists. Form a maximal connected subgraph $T$ such that $T$ contains $e$ and
200: $\ell(f(T)) = 0$. We note that Proposition 2.1 implies that $T$ must be a
201: tree. Hence there exist two vertices $v_1$ and $v_2$ in $T$ such that
202: $f(v_1)=f(v_2)$ and each vertex has exactly one incident edge of length $0$.
203: \medskip
204: \psfig{file=fig21a.eps,height=1.5in,width=5.0in}
205: Figure 2.1. Adjustment at identified vertex.
206: \medskip
207:
208: \noindent There are two possibilities. In case one, there are four edges
209: incident at $f(v_1)=f(v_2)$. Then we can modify the image of $f$ using the
210: trivalent structure exactly as shown in Figure 2.1. This move separates the
211: vertices by introducing a new edge. The idea is to stretch the edge
212: of length $0$ to have small positive length. Then we modify the map using the
213: move in Figure 2.0. This produces a new map of a (possibly different) carrier
214: $n$-graph of strictly lower length. In case two, one of the edges incident at
215: $f(v_2)$ is a subset of an edge incident at $f(v_1)$ and all other incidence
216: angles are $120^\circ$. In this case, a short calculation shows that we may
217: reduce length by sliding $f(v_1)$ along the identified edges while
218: straightening the remaining two. $\square$ \medskip
219:
220: To simplify the exposition, we note the following fact. A minimal length carrier
221: $n$-graph need not be embedded; however, intersections are transverse. Thus,
222: given any $\epsilon > 0$ we can adjust the graph by pushing apart intersecting
223: edges inside an $\epsilon$ ball. This allows us to produce an embedded graph
224: with arbitrarily small total increase in length. Of course, some edges will
225: now fail to be geodesic within finitely many $\epsilon$ neighborhoods. But the
226: distortion can be made as small as desired. It is our aim to show that given an
227: integer $n>0$, there is a constant $R_n$ such that if a manifold $M $ has
228: $inj(M ) > R_n$, then the minimal length carrier $n$-graph
229: $f:\Gamma\longrightarrow M$ provided in Lemma 2.2 has $f_*$ an isomorphism.
230: This will imply that $\pi_1(M)$ is a free group of rank $n$ which contradicts
231: that $M$ is a closed hyperbolic 3-manifold. To do this, we must prove that
232: $f_*$ is injective. Hence, we try to understand the generic properties of a
233: non-trivial class $[\alpha]\in \pi_1(\Gamma)$ for which $f_*([\alpha]) = 0$.
234: \proclaim{Definition 2.5} Fix a universal covering $p:{\Bbb H}^3\longrightarrow
235: M$ and let $f:\Gamma\longrightarrow M$ be a minimal length carrier $n$-graph.
236: \newline (a) We say that $\gamma :S^1\longrightarrow \Gamma$ is a {\bf
237: compressing loop} if $\gamma$ is essential in $\Gamma$ and $f\gamma$ is null
238: homotopic in $M $.\newline (b) A {\bf shortest compressing loop} is a
239: compressing loop of shortest length in $\Gamma.$\newline (c) A {\bf standard
240: lift} for $kernel(f_*)$ is a lift to $\Bbb H^3$ of $f\gamma :S^1\longrightarrow
241: M $ where $\gamma :S^1\longrightarrow \Gamma$ is a shortest compressing loop.
242: \endproclaim \medskip
243:
244: \noindent {\bf Remarks} It follows easily that we may represent a
245: shortest compressing loop by a reduced directed edge path. Thus, Lemma
246: 2.4 and the remarks above show we may assume that a standard lift for
247: $kernel(f_*)$ is a piecewise geodesic embedding of $S^1$ in $\Bbb H^3$
248: in which the geodesic pieces meet at $120^\circ$ angles. We shall call
249: the geodesic pieces in the image of a standard lift {\bf edges}.
250: Notice that these are lifted edges of the minimal length carrier graph.
251: We now abstract the important properties of standard lifts.
252:
253: \proclaim {Definition 2.6}
254: A path $h:[0,1]\longrightarrow {\Bbb H}^3$ is a {\bf Geodesic-120 path}
255: if the following conditions hold: \newline
256: (i) $h$ is a (possibly closed) piecewise geodesic path. \newline
257: (ii) edges in the image of $h$ meet at $120^\circ$ angles. \newline
258: \endproclaim
259: \medskip
260: The key point is that any standard lift we construct for a given closed
261: hyperbolic 3-manifold will be a closed geodesic-120 path in $\Bbb H^3$.
262: For our purposes, it will be helpful to think of the image of a standard
263: lift as a tractable geometric object in ${\Bbb H}^3$. Frequently, we
264: shall also need to consider an important type of subpath in a
265: geodesic-120 path.
266: \medskip
267: \proclaim {Definition 2.7}
268: Let $h:[0,1]\longrightarrow {\Bbb H}^3$ be a geodesic-120 path. Let
269: $[a,b]\subset [0,1]$ where $h(a)$ and $h(b)$ are endpoints of edges
270: in $H = h([0,1])$. A {\bf segment} of $h$ joining $h(a)$ to
271: $h(b)$ is a subpath $s=h|_{[a,b]}:[a,b]\longrightarrow {\Bbb H}^3$.
272: \endproclaim
273: \medskip
274: Naturally, if a geodesic-120 path is an embedding, given two endpoints
275: there is exactly one segment joining them.
276: \proclaim{Lemma 2.8}
277: Let $f:\Gamma\longrightarrow M $ be a minimal length carrier $n$-graph.
278: If $h:S^1\longrightarrow\Bbb H^3$ is a standard lift, then in any
279: segment of $h(S^1)$ containing at least $3(n-1)$ edges,
280: there is a edge with hyperbolic length $\ge {inj(M)\over {3(n-1)}}$.
281: \endproclaim
282: \medskip
283: \noindent {\bf Proof} Given a manifold $M $, Proposition 2.0 shows that
284: an $n$-graph $\Gamma$ in $M $ has $3(n-1)$ edges.
285: Thus, a simple closed curve in $\Gamma$ has less than $3(n-1)$ edges.
286: By Proposition 2.1, the length of such a curve is at least $inj(M)$.
287: Now since $h$ is a standard lift, $h=f\gamma$ where $\gamma$ is a shortest
288: compressing loop. We consider how $\gamma$ behaves in $\Gamma$. We may view
289: $\gamma$ as a product of directed closed edge paths
290: $\alpha_1\alpha_2\ldots\alpha_k$ with the following properties: (a) each path
291: $\alpha_i$ has at most $3(n-1)$ edges (b) each path $\alpha_i$ contains an edge
292: $e$ such that the path defined by removing the edge $e$ from $\alpha_i$ is not
293: a loop (c) each path $\alpha_i$ contains a simple closed curve
294: with at most $3(n-1)$ edges. Thus, each $\alpha_i$ contains an edge of
295: length at least $inj(M )\over {3(n-1)}$. The proof is complete since this
296: argument is independent of the starting vertex for the path $\gamma$. $\square$
297:
298: \bigskip
299: \head 3. The Case $Rank(\pi_1(M))=2$.
300: \endhead
301: In the remainder of this paper, we use the Poincare Disc model for $\Bbb
302: H^3$. We shall also let ${\Cal W}$ denote the horoball with diameter
303: given by the geodesic joining 0 to (-1,0,0). In this section, we shall
304: prove the main theorem in the special case of Rank $\pi_1(M)$ = 2. Note
305: that there are precisely two 2-graphs (see figure 3.0).
306: \bigskip
307:
308: \psfig{file=fig30a.eps,height=1.5in,width=5.0in}
309: Figure 3.0. The two possible 2-graphs.
310: \medskip
311:
312: \proclaim {Lemma 3.0} There exists $R_2>0$ such that if $M$ is a closed,
313: connected, hyperbolic 3-manifold with Rank $\pi_1(M)$ = 2 then $inj(M)
314: \le R_2$.
315: \endproclaim
316: \medskip
317: \noindent {\bf Proof} The proof is established by considering
318: Geodesic-120 paths in $\Bbb H^3$. Given a closed, connected hyperbolic
319: 3-manifold $M$ and a minimal length carrier $2$-graph
320: $f:\Gamma_2\longrightarrow M$ we may choose a standard lift
321: $h:S^1\longrightarrow {\Bbb H}^3$ for $kernel (f_*)$. Let $H$ denote
322: $h(S^1)$. Since $H$ is compact, there exist points $v$,$w \in H$ of
323: maximal distance apart. By an isometry of $\Bbb H^3$, we may assume that
324: $w=0$ and that $v$ lies on the geodesic joining 0 to (-1,0,0). Now
325: $H\subset \bar B(v,d(v,0))$ which implies that $H$ is contained in the
326: horoball ${\Cal W}$ . Note also that since $H$ is a Geodesic-120
327: path, 0 and $v$ must be endpoints of edges of $H$ . Moreover, since
328: $H$ is a loop, there must be two edges of $H$ incident at 0.
329: Consider these two edges. Geodesics through 0 in the disc model
330: correspond to Euclidean diameters of the unit ball in $\Bbb R^3$. Since
331: the angle between the edges is $120^\circ$, there exists $L_0>0$ so that if
332: one of the edges of $H$ incident at 0 has length at least $L_0$, then $H$ is
333: not contained in ${\Cal W}$. There are precisely two 2-graphs (see figure
334: 3.0) and each of these graphs has three edges. By using Proposition 2.1, we see
335: that a 2-graph may have at most one ``short'' edge. More precisely, if
336: $inj(M) > 2L_0$, then the image $H$ of any standard lift has the following
337: property: given any pair of edges of $H$ with a common endpoint, at least one
338: edge has length greater than $L_0$. In other words, put $R_2=2L_0$. If
339: $inj(M) > R_2$, (after isometry of ${\Bbb H}^3$) any standard lift for
340: $kernel(f_*)$ must have a edge which is not contained in ${\Cal W}$. This
341: contradiction shows that $kernel(f_*)$ is trivial, so that $\pi_1(M)$ is
342: actually a free group of rank 2. Since this is impossible, we conclude that
343: $inj(M) \le R_2$. $\square$
344:
345: \medskip
346: \psfig{file=fig31a.eps,height=4.0in,width=5.0in}
347: Figure 3.1 Escape of Standard Lift in Rank 2 Case.
348: \medskip
349:
350: \medskip\noindent {\bf Remark} It is possible to explicitly compute an upper
351: bound for injectivity radius in the $n=2$ case. One finds using a very
352: simple geometry calculation that $L = 2.6338\ldots$ so that $inj(M) \le
353: 5.2676\ldots$ \bigskip
354:
355: \head 4. Proof of Main Theorem
356: \endhead
357: The general case is more subtle because $n$-graphs with $n>2$ have more
358: edges. Although large injectivity radius will ensure that minimal
359: length carrier graphs have some long edges, if $n>2$ these graphs may
360: have subpaths of very short edges. This prevents us from using the
361: simple geometric approach in Lemma 3.0 since the image of a standard
362: lift can have very large geodesic edges while remaining inside the
363: horoball ${\Cal W}$ . Instead, we will argue that, for sufficiently
364: large injectivity radius, if the standard lift is contained in ${\Cal
365: H}$, then it is possible to reduce the length of the minimal length
366: carrier $n$-graph. We will produce a {\bf short cut arc} in ${\Bbb
367: H}^3$ which allows us to make a shorter carrier graph using
368: ``cut-and-paste'' in the manifold below. To do this, we need two
369: technical lemmas which we now describe. Recall that ${\Bbb H}^3$ enjoys
370: ``thin triangles.'' This means there is a universal constant $\Delta$, the
371: {\bf thin triangles constant}, such that given a hyperbolic triangle with
372: geodesic sides $X$, $Y$ and $Z$ and a point $x\in X$, there exists a point
373: $y\in Y\cup Z$ with $d(x,y)<\Delta.$ The idea is to use this fact together
374: with the geometry of the horoball to show that sufficiently large injectivity
375: radius forces a standard lift to be ``thin.'' Our first lemma is motivated by
376: this fact:
377:
378: \medskip
379: \proclaim {Proposition 4.0} Given $L>0$, if $M$ is a closed hyperbolic
380: 3-manifold with $n =$ Rank $(\pi_1(M))$, and $inj(M) > [3(n-1)]^2 L$, then
381: every standard lift for $\pi_1(M)$ has two edges of length at least $L$.
382: \endproclaim \medskip\noindent {\bf Proof} Let $M$ satisfy the hypothesis, and
383: let $h:S^1\longrightarrow\Bbb H^3$ be a standard lift with image $H$. Lemma
384: 2.8 shows that any standard lift has one edge of length at least $3(n-1)L$. But
385: if $H$ has exactly one edge of length at least $3(n-1) L$, then it has at most
386: $3(n-1)$ edges. Therefore, $H$ cannot be a closed loop in ${\Bbb H}^3$ unless
387: there is another edge of length at least $L$. $\square$ \medskip \noindent This
388: proposition suggests we should try to produce our short cut arc between two
389: ``long'' edges. The lemma below says that, if $A$ and $B$ are two such edges
390: in ${\Cal W}$ that start at points ``close'' to the origin, then there is a
391: short cut arc between them.
392:
393: \medskip
394: \proclaim {Lemma 4.1 (Short Cut Lemma)}
395: For any $\delta>0$ there exists $\bar L(\delta)>0$ such that if $A=[a$,$a^\prime]$
396: and $B=[b$,$b^\prime]$ are geodesic edges contained in ${\Cal W}$
397: with:\newline
398: (i) $a$,$b\in \bar B(0,\delta),$\newline
399: (ii) $|A|$,$|B| \ge \bar L$,\newline
400: \noindent then there exist points $e\in A$ and $f\in B$ such
401: that:\newline
402: (i) $d(e$,$a) < {1 \over 3}\bar L\ \text{and}\ d(f$,$b) < {1 \over
403: 3}\bar L$\newline
404: (ii) $d(e$,$f) < d(e$,$a)-\Delta$ \text{and}\ $d(e$,$f) < d(f$,$b)-\Delta$.
405: \endproclaim
406: \medskip\noindent {\bf Proof} See Figure 4.0. Given $A$ and $B$ as in the
407: hypothesis, let $l_A=[0$,$a^\prime]$, $l_B=[0$,$b^\prime]$, and $\phi$ be the
408: angle between $l_A$ and $l_B$. Let $\delta >0$. Choose $L_1 >
409: 2(\delta+\Delta)$. This choice assures that given $A,B$ as in hypothesis
410: with $|A|$,$|B| > L_1$, we have $a^\prime$,$b^\prime \not \in \bar
411: B(0$,$\delta)$. Now notice that as $L_1\rightarrow\infty$, we have
412: for all $A,B$ as in the hypothesis with $|A|$,$|B| > L_1$, that
413: $a^\prime$,$b^\prime\rightarrow$ (-1,0,0) in the Euclidean metric.
414: Hence, we may choose $L_2>L_1$ so that given any $A$ and $B$ as in the
415: hypothesis with $|A|$,$|B| > L_2$, we are guaranteed to have
416: $|l_A|$,$|l_B|$ sufficiently large and $\phi$ sufficiently small that
417: there exist points $x\in l_A$ and $y\in l_B$ with $d(x,\partial \bar
418: B(0$,$\delta))$, $d(y,\partial \bar B(0$,$\delta)) = 5\Delta$ and
419: $d(x$,$y)<\Delta$. Using thin triangles, there exist points $e\in A$
420: and $f\in B$ such that $d(e$,$x)$,$d(f$,$y) < \Delta$. This implies
421: $d(e$,$f) < 3\Delta$. Now choose $\bar L(\delta) = 3max(L_2,6\Delta + 2\delta)$.
422: Then $d(e$,$a)$,$d(f$,$b) > 4\Delta$ which proves that
423: $d(e$,$f)<d(e$,$a)-\Delta$ and $d(e$,$f)< d(f$,$b)-\Delta$. Also, $d(e$,$a)\le
424: 6\Delta + 2\delta$ so that $d(e$,$a)< {1\over 3}\bar L(\delta)$. $\square$
425: \medskip
426: \noindent {\bf Remark} In subsequent work, we shall cut out one of the
427: geodesics $[e,a]$ or $[f,b]$ and paste in $[e,f]$. It is worth mentioning at
428: this point that conclusion $(ii)$ tells us $\ell ([e,a])-\ell ([e,f])>\Delta$
429: and $\ell ([f,b])-\ell ([e,f])>\Delta$, so the short cut reduces length by at
430: least $\Delta$. The small perturbation needed to make $\Gamma$ embedded can be
431: chosen to change lengths less than $\Delta$.
432:
433: \medskip
434: \psfig{file=fig40a.eps,height=3.5in,width=5.0in}
435: Figure 4.0. Trapping of geodesic segments.
436: \medskip
437:
438: \medskip The next two results show that we can actually achieve the conditions
439: in the hypothesis of Lemma 4.1 for embedded Geodesic-120 paths provided we
440: attach the right assumptions. Suppose that $n=Rank(\pi_1(M))$. If $inj(M)$ is
441: very large, Lemma 2.8 and Proposition 4.0 imply that there are two ``very
442: long'' edges in a standard lift for $\pi_1(M)$. Denoting these two edges $A$
443: and $B$, we then notice that by Lemma 2.8 there is a Geodesic-120 path of at
444: most $6n-6$ edges joining $A$ to $B$. We show that either the starting points
445: of $A$ and $B$ are sufficiently close or there is another (not quite as)
446: ``long'' edge between $A$ and $B$. Taking innermost ``long'' edges then
447: provides the requirement for the short cut lemma. The basis for the procedure
448: is the following fact about finite sequences of numbers: \medskip \proclaim
449: {Proposition 4.2} Let $L(0)<L(1)<L(2)<\dots$ be a given sequence of positive
450: numbers. Let $x_0,\dots x_n$ and $y_0,\dots y_m$ be sequences of positive
451: numbers with $k=n+m$ such that:\newline (i) $x_n$ and $y_m$ are both greater
452: than $max\lbrace x_{n-1},...,x_0,y{m-1},...,y_0\rbrace$. \newline (ii) $x_n$
453: and $y_m$ are both greater than $L(k)$\newline Then one of the following
454: conditions must hold:\newline (1) There exist $r\ge 0$ and $s\ge 0$ such that
455: for $q = r+s$, \newline \indent (i) $x_r$ and $y_s$ are both greater than
456: $max\lbrace x_{r-1},...,x_0,y{s-1},...,y_0\rbrace$. \indent (ii) $x_r$ and
457: $y_s$ are both greater than $L(q)$\newline \indent (iii) $\sum_{i=0}^{r-1} x_i
458: + \sum_{i=0}^{s-1} y_i \le q L(q-1)$ \newline
459: (2) Either $x_0 > L(0)$ or $y_0 > L(0)$.
460: \endproclaim
461: \medskip
462: {\bf Proof} Let $x_0,\dots x_n$ and $y_0,\dots y_m$ be sequences which
463: satisfy the hypothesis. We may assume that $n\ge 1$ and $m\ge 1$ as
464: otherwise $(2)$ holds. We may also assume that $\sum_{i=1}^n x_i +
465: \sum_{i=1}^m y_i > k L(k-1)$.
466: Then without loss of generality there exists $x_r$ with $r< n$ such
467: that $x_r > L(k-1)$. By choosing $s=m$ if necessary, there also exists
468: $y_s>L(k-1)$. We choose $r$ and $s$ to be as small as possible. If
469: $r=0$ or $s=0$, the proof is finished since $(2)$ holds. Otherwise we
470: have that $x_r$ and $y_s$ are both greater than $max\lbrace
471: x_{r-1},...,x_0,y{s-1},...,y_0\rbrace$. Note also that $r+s\le n-1+m = k-1$.
472: Hence, for $q=r+s$ we have $x_r>L(k-1)\ge L(q)$ and $y_s>L(k-1)\ge
473: L(q)$. Now if $1(iii)$ holds, the proof is done. Suppose otherwise.
474: Repeat the above procedure on the sequences $x_r,\dots x_n$ and
475: $y_s,\dots y_m$; continuing in this manner, we either obtain sequences
476: $x_{r'},\dots x_0$ and $y_{s'},\dots y_0$ which satisfy $(1)$ or we run
477: out of sequence elements so that we obtain $(2)$. $\square$
478: \medskip
479: \proclaim {Lemma 4.3}
480: For each integer $k\ge 0$ there is an $L(k)>0$ such that if
481: $g:[0,1]\longrightarrow {\Bbb H}^3$ is an embedded Geodesic-120 path
482: with $G=g([0,1])$ which satisfies:\newline
483: (i) $0\in G$ \newline
484: (ii) There exist two geodesic edges $A$ and $B$ in $G$ with
485: $|A|$,$|B|>L(k)$\newline
486: (iii) Let $S$ be the segment of $g$ which joins $A$ to $B$.
487: Then $0\in S$ and $S$ has at most $k$ edges. \newline
488: (iv) Every edge of $S$ has length at most $L(k)$.
489: \smallskip\noindent Then one of the following holds:\newline
490: \noindent (1) $G$ is not contained in the horoball ${\Cal W}$ .
491: \newline
492: \noindent (2) There exist edges $E_1$ and $E_2$ in $G$ such
493: that:\newline
494: (i) $g^{-1}(E_1)$ and $g^{-1}(E_2)$ are in different components of
495: $[0,1]-{g^{-1}(0)}$\newline
496: (ii) Every edge in the segment joining $E_1$ to $0$ has length
497: less than $|E_1|$ and $|E_2|$ \newline
498: (iii) Every edge in the segment joining $E_2$ to $0$ has length
499: less than $|E_1|$ and $|E_2|$ \newline
500: (iv) Let $e_1\in E_1$ and $e_2\in E_2$ be the endpoints of these
501: edges contained in the segment joining $E_1$ to $E_2$. Then
502: there is a short cut arc $F$ joining $e \in E_1$ to $f\in E_2$ such that
503: $d(e_1$,$e)>|F|+\Delta$, $d(e_1$,$e) < {1\over 3} L(k)$, and $d(e_2$,$f) <
504: {1\over 3} L(k)$.
505: \endproclaim
506:
507:
508: {\bf Proof} Define the following sequence: set $L(0)$ equal to the $L_0$ in
509: the proof of Lemma 3.0. Recall that this is the length required to force
510: one of the edges at $0$ to leave the horoball ${\Cal W}$. Put $L(1)=2L(0)$ and
511: for $k\ge 2$ define $L(k)=\bar L(kL(k-1))$ with $\bar L$ provided by Lemma 4.1.
512: This implies that for $k\ge 2$, $L(k)\ge 2(k L(k-1)+\Delta)$
513: so that evidently $L(0)<L(1)<L(2) ...$ is an increasing sequence of positive
514: numbers. Now fix $k\ge 0$ and let $G$ be the image of a geodesic-120 path which
515: satisfies the hypothesis. Let $x_n = |A|$, $y_m = |B|$, and let $x_{n-1},\dots
516: x_0$ denote the lengths of the edges joining $A$ to $0$ labelled in decreasing
517: order as we proceed from $A$ to $0$. Likewise, let $y_{m-1},\dots y_0$ denote
518: the lengths of the edges joining $B$ to $0$. By hypothesis $(iii)$, $n+m =
519: k_1\le k$. By $(ii)$, $x_n>L(k)\ge L(k_1)$ and $y_m>L(k)\ge L(k_1)$. Also, by
520: $(iv)$ we have that $x_n$ and $y_m$ are both greater than $max\lbrace
521: x_{n-1},...,x_0,y{m-1},...,y_0\rbrace$. This means these sequences satisfy the
522: hypothesis of Proposition 4.2. Conclusion $(2)$ of Proposition 4.2 gives that
523: $x_0>L(0)$ or $y_0>L(0)$. This means that either $x_0$ or $y_0$ leaves the
524: horoball ${\Cal W}$; hence, $G$ is not contained in ${\Cal W}$. Otherwise,
525: conclusion $(1)$ of Proposition 4.2 gives us subsequences $x_r,\dots x_0$ and
526: $x_s,\dots x_0$ with these properties: $x_r>L(q)$, $y_s>L(q)$, $x_r$ and $y_s$
527: are both greater than $max\lbrace x_{r-1},...,x_0,y{s-1},...,y_0\rbrace$, and
528: $\sum_{i=0}^{r-1} x_i + \sum_{i=0}^{s-1} y_i \le q L(q-1)$. Refer to Figure
529: 4.1. Geometrically, this means the following. Let $E_1$ and $E_2$ be the edges
530: of $G$ which correspond respectively to the lengths $x_r$ and $y_s$. It follows
531: at once that statements $2(i)$ through $2(iii)$ hold. To see that $2(iv)$ is
532: also true, note that the segment joining $E_1$ to $0$ has length at most
533: $qL(q-1)$ since $\sum_{i=0}^{r-1} x_i \le q L(q-1)$. The analogous statement
534: holds for the segment joining $E_2$ to $0$. Thus, $e_1,e_2\in \bar
535: B(0,qL(q-1))$. Since $|E_1| = x_r > L(q)$ and $|E_2| = y_s > L(q)$, Lemma 4.1
536: applies to give us the required short cut arc $F$. $\square$
537:
538: \bigskip
539: \psfig{file=fig41a.eps,height=3.5in,width=5.0in}
540: Figure 4.1.
541: \medskip
542:
543: We now show how to use Lemma 4.3 to establish the main result in this
544: paper. By the remark following the proof of Lemma 4.1, we can assume our
545: graph is embedded without affecting the validity of our ``cut-and
546: paste'' argument described below. This means we can think of a standard
547: lift as a geodesic-120 embedding of $S^1$.
548: \medskip
549: \proclaim {Theorem 4.4}
550: For each integer $n>1$ there exists an $R_n > 0$ such that if M is a
551: closed, connected, hyperbolic 3-manifold and Rank $\pi_1$(M) = n, then
552: $inj(M) < R_n$.
553: \endproclaim
554: \medskip
555: \noindent {\bf Proof of Main Theorem} Given an integer $n>1$, we will
556: show that there exists $R_n>0$ such that if a closed hyperbolic
557: 3-manifold $M $ has Rank $\pi_1(M )=n$ and $inj(M ) \ge R_n$, then
558: $\pi_1(M)$ is a free group. This will establish the theorem since the
559: fundamental group of such a manifold cannot be a free group. Fix
560: $n>0$. We shall use the notation $L(k)$ as in the statement of Lemma
561: 4.3. Notice that we may choose $R>0$ so that if $n=Rank(\pi_1(M))$ and
562: $inj(M)>R$ then the following holds: if $f:\Gamma\longrightarrow M $ is
563: a minimal length carrier graph, then every standard lift for $kernel(f_*)$
564: has two edges of length
565: at least $L=L(3(n-1))$. Set $R_n$ = max(R, (3n-3)L(3n-4)).
566: Thus, suppose $M$ has $inj(M)>R_n$ and Rank $\pi_1(M )=n$. Let $h$ be a
567: standard lift for $kernel(f_*)$ to $\Bbb H^3$. As in the proof of lemma 3.0,
568: $h$ is a Geodesic-120 path through $0$ with $H = h(S^1)$ contained in the
569: horoball ${\Cal W}$. Also, using the remark following Definition 2.5, we
570: assume $h$ is an embedding of $S^1$. Notice that there are two edges $A$ and
571: $B$ in $H$ with $|A|$,$|B|> L(3n-3)$ and at most $3n-3$ edges in the segment
572: joining them through $0$. Consider the Geodesic-120 path $g$ defined by
573: $A$,$B$ and the segment of $h$ containing $0$ which joins them. By the
574: proof of Lemma 3.0, since $|A|$ and $|B|$ are greater than $L(1)$, $g$
575: is not a closed loop. Thus, $g$ satisfies the hypothesis of Lemma 4.3.
576: Therefore, the assumption that $H$ is contained in ${\Cal W}$ means
577: there exist edges $E_1$,$E_2$ in $G = g([0,1])$ which satisfy $2.(i)
578: - 2.(iv)$ in Lemma 4.3. Let $e_1$,$e_2$ denote the respective endpoints
579: of $E_1$,$E_2$ contained in the segment of $g$ which joins
580: $E_1$ to $E_2$. This segment contains $0$. Moreover, since
581: $H$ is contained in ${\Cal W}$, the proof of Lemma 4.3 shows that we may
582: assume that $E_1$ and $E_2$ are innermost with respect to $0$ in the
583: following sense: There is an integer $i>0$ with $|E_1|$,$|E_2| > L(i)$
584: and $e_1$,$e_2\in\bar B(0,iL(i-1))$ where $iL(i-1)<(3n-3)L(3n-4)$. To
585: see this, note that if no such $i$ exists, then we may repeat the
586: inductive argument in the proof of Proposition 4.2 to obtain a edge
587: $E$ incident at $0$ with $|E|>L(0)$. This contradicts the assumption
588: that $H\subset {\Cal W}$ .
589:
590: \medskip
591: \psfig{file=fig42a.eps,height=2.75in,width=5.0in}
592: Figure 4.2.
593: \medskip
594:
595: \noindent Refer to figure 4.2. We complete the proof using the arc $F$
596: joining $E_1$ to $E_2$ provided by Lemma 4.3. We let $e\in E_1$ and
597: $e^\prime\in E_2$ denote the endpoints of $F$. Let $C$ denote the
598: subarc of $E_1$ which joins $e_1$ to $e$. We wish to show that we can remove
599: $C$ and attach $F$ such that the projection to $M$ gives a shorter carrier
600: n-graph. Let $\rho : {\Bbb H}^3\longrightarrow M$ denote the universal cover
601: of $M$. We first show that the modified image $(f(\Gamma) - \rho (int(C)))
602: \cup \rho (F)$ defines a map of an $n$-graph into $M$. We can work with the
603: image $f(\Gamma)$ since $\Gamma$ is embedded in $M$. Now, $\rho (e_1)$ is a
604: vertex of $f(\Gamma)$. By removing $\rho (int(C))$, we convert $\rho(e_1)$
605: into a bivalent vertex. Hence, we can amalgamate the remaining two edges
606: incident at $\rho (e_1)$ into a single edge. This removes the vertex $\rho
607: (e_1)$. We then view $\rho ((E_1 - int(C)) \cup F$ as a single edge so that
608: $\rho (e^\prime)$ becomes a trivalent vertex. This new subset of $M$ is an
609: embedded, closed, trivalent graph $f^\prime : \Gamma^\prime\longrightarrow M$
610: with the same number of vertices as $\Gamma$; therefore, $f^\prime :
611: \Gamma^\prime\longrightarrow M$ is an $n$-graph. Note also that $\ell (f^\prime
612: (\Gamma^\prime)) < \ell (f(\Gamma))$. Therefore, we need only show that
613: $\Gamma^\prime$ carries $\pi_1 (M)$. Consider again the carrier $n$-graph
614: $f:\Gamma\longrightarrow M $. Let $P$ denote the path defined by the edges
615: joining $e_2$ to $e_1$ together with the subarc of $E_2$ joining $e'$ to $e_2$.
616: The idea is then to show that $P$ can be used to replace $C$ in any loop of
617: $\Gamma$. This follows if we show we have not cut edges in $P$ by
618: removing $C$. Choose a basepoint $x_0\in f(\Gamma) - (\rho(E_1)\cup\rho(E_2))$
619: and let $y_0\in f^{-1}(x_0)$. Let $[\alpha]\in\pi_1(M,x_0)$. Since $f_*$ is an
620: epimorphism, there exists $[\beta]\in\pi_1(\Gamma,y_0)$ with
621: $f_*([\beta])=[\alpha]$. We shall construct a corresponding loop $\beta^\prime$
622: in $\Gamma^\prime$ so that $f_*^\prime[\beta^\prime] = [\alpha]$. It follows
623: from $2.(i)$ through $2.(iv)$ of Lemma 4.3 that every edge in the segment of
624: $g$ which joins $E_1$ to $E_2$ has length less than both $E_1$ and $E_2$.
625: Therefore, none of these edges projects to $\rho(E_1)$ or to $\rho(E_2)$. Now
626: suppose $\rho(E_1)=\rho(E_2)$. We must have $\rho(e_1) \ne \rho(e_2)$ since
627: otherwise $d(e_1,e_2) < (3n-3)L(3n-4)$ implies the geodesic edge with endpoints
628: $e_1$ and $e_2$ projects to a nontrivial loop in $M$ of length less than $R_n$.
629: This guarantees that the arc of $E_2$ joining $e'$ to $e_2$ does not project to
630: $\rho(C)$. Thus, $\rho(C)$ is not a subset of $\rho(P)$. In particular,
631: $\rho(F\cup P)$ and $\rho(C)$ are paths in $M$ that are homotopic with
632: endpoints fixed. These facts imply that if we imagine $\beta$ as a directed
633: edge path in $\Gamma$, we can use the word for $\beta$ to build $\beta^\prime$
634: in $\Gamma^\prime$. That is, $\rho(E_1)$ corresponds to an edge $W_1$ in
635: $\Gamma$ and $\rho((E-1-int(C)) \cup F \cup P)$ corresponds to an edge path
636: $W_2$ in $\Gamma^\prime$. In the word for $\beta$, replace $W_1$ (resp.
637: $W_1^{-1}$) by $W_2$ (resp. $W_2^{-1}$). This gives a loop $\beta^\prime$ in
638: $\Gamma^\prime$. By construction, we may homotop $f\beta$ to
639: $f^\prime\beta^\prime$ fixing $x_0$, which shows that $f^\prime_*$ is onto.
640: This completes the proof of the main theorem. $\square$
641:
642: \proclaim{Corollary 4.5} Given a closed hyperbolic 3-manifold $M$ and an
643: integer $n>0$, there exists a finite sheeted cover $$p:\tilde M\longrightarrow\
644: M$$ with $Rank\ \pi_1 (\tilde M) \ge n$. \endproclaim \medskip \noindent {\bf
645: Proof}. Given $M$ and $n>0$, let $\tilde M_0$ be a covering space of $M$.
646: Suppose we have $Rank\ \pi_1 (\tilde M_0) < n$. Then Theorem 4.3 shows
647: $inj(\tilde M_0)< R_n$. $\tilde M_0$ has finitely many closed geodesics
648: $\gamma_1,\dots,\gamma_k$ of length less than $R_n$. By residual finiteness,
649: there exists a a covering space $\tilde M$ of $\tilde M_0$ such that the loops
650: $\gamma_1,\dots,\gamma_k$ do not lift. This implies $inj(\tilde M) > R_n$ so
651: that $Rank\ \pi_1 (\tilde M_0) \ge n$. $\square$ \bigskip \head References
652: \endhead
653:
654: \bigskip
655: \noindent {\bf Acknowledgements} The author thanks Professor Daryl Cooper for
656: his encouragement and for many helpful conversations.
657: \bigskip
658:
659: \noindent [C] D. Cooper, ``The Volume of a Closed Hyperbolic
660: 3-Manifold is Bounded by $\pi$ Times the Length of any Presentation of its
661: Fundamental Group'', To appear P.A.M.s. \smallskip
662:
663: \noindent [DC] M. Do Carmo, ``Riemannian Geometry'', Birkhauser, Boston,
664: 1993.
665: \smallskip
666: \noindent [E] P. Eberlein, ``Structure of Manifolds of Nonpositive
667: Curvature'' in {\it Global Geometry and Global Analysis 1984, Proceedings
668: 1984}, D. Ferus et.al. editors, Lecture Notes in Math. 1156, Springer, New
669: York, 1985, pp. 86-153. 1993. \smallskip \noindent [G] M. L. Gromov,
670: ``Hyperbolic Groups'' in {\it Essays in Group Theory}, S. Gersten, MSRI
671: Publications 8, Springer, New York, 1987, pp. 75-263. 1993. \smallskip
672: \noindent [M] W. Massey, ``Algebraic Topology: An Introduction'',
673: Harbrace, New York, 1967.
674: \smallskip
675: \noindent [Pr] A. Preissman, {\it Quelques proprietes globales des espaces de
676: Riemann}, Comm. Math. Helv. 15 (1943), pp. 175-216. \smallskip
677: \noindent [Th] W.P. Thurston, ``The Geometry and Topology of
678: Three-Manifolds'', Princeton University, 1979.
679: \enddocument
680: \end
681: