1: \documentclass[12pt]{amsart}
2:
3: \usepackage{graphicx}
4: \usepackage{amsthm}
5: \usepackage{amsmath, amssymb}
6: \swapnumbers
7:
8: \theoremstyle{plain}
9: \newtheorem{Thm}{Theorem}
10: \newtheorem{MThm}[Thm]{Main Theorem}
11: \newtheorem{Coro}[Thm]{Corollary}
12: \newtheorem{Lem}[Thm]{Lemma}
13: \newtheorem{Claim}[Thm]{Claim}
14: \newtheorem{Prop}[Thm]{Proposition}
15: \newtheorem{Que}[Thm]{Question}
16:
17: \theoremstyle{definition}
18: \newtheorem{Def}[Thm]{Definition}
19:
20: %\linespread{1.6}
21:
22: \begin{document}
23:
24: \title{Mapping class groups of Heegaard splittings}
25:
26: \author{Jesse Johnson and Hyam Rubinstein}
27: \address{\hskip-\parindent
28: Department of Mathematics \\
29: Yale University \\
30: PO Box 208283 \\
31: New Haven, CT 06520 \\
32: USA}
33: \email{jessee.johnson@yale.edu}
34: \address{\hskip-\parindent
35: Dept. of Mathematics and Statistics \\
36: The University of Melbourne \\
37: Parkville, VIC, 3053 \\
38: AUSTRALIA}
39:
40: \email{H.Rubinstein@ms.unimelb.edu.au}
41: \subjclass{Primary 57M}
42: \keywords{Heegaard splitting, mapping class group}
43:
44: \thanks{Research supported by NSF MSPRF grant 0602368 and ARC grant DP020849}
45:
46: \begin{abstract}
47: The mapping class group of a Heegaard splitting is the group of connected components in the set of automorphisms of the ambient manifold that map the Heegaard surface onto itself. We find examples of elements of the mapping class group that are periodic, reducible and pseudo Anosov on the Heegaard surface, but are isotopy trivial in the ambient manifold. We prove structural theorems about the first two classes, in particular showing that if a periodic element is trivial in the mapping class group of the ambient manifold, then the manifold is a Seifert fibered space.
48: \end{abstract}
49:
50:
51: \maketitle
52:
53: \section{Introduction}
54: Given a 3-manifold $M$, define $Aut(M)$ to be the group of automorphisms (self homeomorphisms) of $M$. The set of connected components of $Aut(M)$ forms a group $Mod(M)$ called the mapping class group of $M$. Given a Heegaard splitting $(\Sigma, H_1, H_2)$ of $M$, we will define $Aut(M, \Sigma)$ to be the subset of $Aut(M)$ consisting of all maps that send $\Sigma$ onto itself. The set of connected components of $Aut(M, \Sigma)$ forms a group, the \textit{mapping class group of $\Sigma$}, which we will denote $Mod(M, \Sigma)$. We will give a more detailed explanation of these groups and a review of the relevant literature in Section~\ref{mcgsect}.
55:
56: Automorphisms of Heegaard splittings can be separated into three classes as follows: An element of $Mod(M, \Sigma)$ is periodic, reducible or pseudo Anosov if its restriction to $\Sigma$ is periodic, reducible or pseudo Anosov, respectively. We will deal with each case separately, in order of decreasing understanding, beginning with periodic.
57:
58: One can construct a Heegaard splitting with finite order elements in its mapping class group as follows: Let $N$ be a 3-manifold and $M$ a finite branched or unbranched cover of $N$. Let $(\Sigma', H'_1, H'_2)$ be a Heegaard splitting of $N$. If the covering of $N$ is branched, assume the branch set is in bridge position with respect to $\Sigma$. This condition on the branch set implies that the Heegaard splitting $(\Sigma', H'_1, H'_2)$ lifts to a Heegaard splitting $(\Sigma, H_1, H_2)$ of $M$.
59:
60: The deck transformations of the covering define a finite group of automorphisms of $M$. These automophisms take $\Sigma$ onto itself so they define a finite subgroup of $Mod(M, \Sigma)$. In Sections~\ref{hbodysect} and~\ref{extendsect}, we prove the following:
61:
62: \begin{Thm}
63: \label{mainthm1}
64: Let $G \subset Mod(M, \Sigma)$ be a finite subgroup of the mapping class group of a Heegaard splitting $(\Sigma, H_1, H_2)$. Then $M$ is a (possibly branched) finite cover of a manifold $N$ such that $(\Sigma, H_1, H_2)$ is the lift of a Heegaard splitting $(\Sigma', H'_1, H'_2)$ for $N$. The branch set of the covering is in bridge position with respect to $\Sigma'$ and the group $G$ is induced by the deck transformations of the covering.
65: \end{Thm}
66:
67: Each connected component of $Aut(M, \Sigma)$ is a subset of a connected component of $Aut(M)$, so the inclusion map in $Aut(M)$ determines a canonical homomorphism $i : Mod(M, \Sigma) \rightarrow Mod(M)$. In general, this map need be neither injective nor surjective. If $\Sigma$ is stabilized, the kernel of $i$ will contain all Dehn twists along the intersection of $\Sigma$ with a sphere around the stabilization. If there are alternate Heegaard splittings of $M$ that are homeomorphic to $\Sigma$, but not isotopic (this is common in toroidal manifolds) then $i$ will not be onto.
68:
69: The kernel elements of $Mod(M, \Sigma)$ that come from spinning around a stabilization are all reducible as automorphisms of $\Sigma$, and therefore have infinite order. Heegaard splittings with finite order elements in the kernel are much less common. Theorem~\ref{mainthm1} allows us to pose the question of finite order elements in terms of the deck transformations of a finite cover and this point of view leads to the following Theorem:
70:
71: \begin{Thm}
72: \label{mainthm2}
73: If $M$ is a closed hyperbolic 3-manifold then the kernel of the canonical homomorphism $i : Mod(M,\Sigma) \rightarrow Mod(M)$ contains no finite order elements.
74: \end{Thm}
75:
76: This is proved in Section~\ref{decksect}. Combining Theorem~\ref{mainthm2} with Namazi's results~\cite{nam:mcg} on mapping class groups of high distance splittings, we conclude (assuming the geometrization conjecture) the following:
77:
78: \begin{Coro}
79: \label{maincoro}
80: For every genus $g$, there is a constant $K$ such that if $\Sigma$ is a genus $g$ Heegaard splitting of distance at least $K$ then the canonical homomorphism is an isomorphism.
81: \end{Coro}
82:
83: The distance refered to here is that defined by Hempel~\cite{hemp:comp} via the curve complex.
84:
85: In the case of reducible elements of $Mod(M, \Sigma)$, there appears to be no simple classification. This is because reducible elements fall into at least two classes with very different behavior.
86:
87: In Section~\ref{wrsect}, we show that a simple combinatorial condition on disks in $H_1$ and $H_2$ implies the existence of ``local'' reducible automorphisms in which many of the disks of the handlebodies are fixed. In particular, we show that every weakly reducible Heegaard splitting contains such an element in the kernel of $i$. In Section~\ref{opensect}, on the other hand, we demonstrate reducible elements in the kernel with a more global behavior, arising from open book decompositions.
88:
89: In Section~\ref{fixsect}, we use Oertel's characterization of reducible automorphisms of handlebodies to prove two lemmas which we use in Section~\ref{redsect} to characterize reducible automorphisms of Heegaard splittings. As a result of this characterization, we show the following:
90:
91: \begin{Thm}
92: \label{redthm}
93: If $Mod(M, \Sigma)$ contains a reducible element then the distance of $\Sigma$ is at most 4.
94: \end{Thm}
95:
96:
97: Finally, in Section~\ref{stabsect}, we consider the case of pseudo Anosov maps. In this case, we are only able to show that every stabilized Heegaard splitting has pseudo Anosov elements in its kernel. The problem of characterizing in general Heegaard splitings with pseudo Anosov automorphism remains open.
98:
99:
100:
101:
102: \section{The Mapping Class Group}
103: \label{mcgsect}
104:
105: A Heegaard splitting for a 3-manifold $M$ is a triple $(\Sigma, H_1, H_2)$ where $H_1, H_2 \subset M$ are handlebodies (manifolds homeomorphic to regular neighborhoods of graphs) and $\Sigma$ is a compact, connected, closed and orientable surface embedded in $M$ such that $H_1 \cup H_2 = M$ and $\partial H_1 = \Sigma = \partial H_2$.
106:
107: As noted above, $Mod(M, \Sigma)$ is the group of equivalence classes of automorphisms of $M$ that take $\Sigma$ onto itself. Two automorphisms are equivalent if there is an isotopy from one to the other by automorphisms that take $\Sigma$ onto itself. In this paper, we will restrict our attention to elements of $Mod(M, \Sigma)$ that are orientation preserving on $M$ and orientation preserving on $\Sigma$. Such automorphisms take the handlebody $H_1$ onto itself and $H_2$ onto itself. If automorphisms that swap the handlebodies exist, the automorphisms that preserve the handlebodies form an index two subgroup.
108:
109: Throughout the paper, $i$ will refer to the homomorphism from $Mod(M, \Sigma)$ to $Mod(M)$ induced by ``forgetting'' $\Sigma$ and considering each element of $Mod(M, \Sigma)$ as an automorphism of $M$. The two immediately obvious questions to ask are when $i$ will be onto and when $i$ will be one-to-one. The first of these questions is equivalent to two classical questions from the field of Heegaard splittings.
110:
111: Two Heegaard surfaces, $\Sigma$ and $\Sigma'$ of a manifold $M$ are called \textit{homeomorphic} if there is an automorphism of $M$ taking $\Sigma$ onto $\Sigma'$. The surfaces are called \textit{isotopic} if there is an automorphism that is isotopic to the identity and sends $\Sigma$ onto $\Sigma'$.
112:
113: For any Heegaard surface $\Sigma'$ homeomorphic to $\Sigma$, there is an automorphism $\phi$ such that $\phi(\Sigma) = \Sigma'$. If the homomorphism $i$ is onto then there is an automorphism $\psi$ that is isotopic to $\phi$ and sends $\Sigma$ onto itself. The map $\phi \circ \psi^{-1}$ is isotopic to the identity on $M$ and sends $\Sigma$ onto $\Sigma'$, so $\Sigma$ is in fact isotopic to $\Sigma'$. Combining this argument with its converse, we get the following:
114:
115: \begin{Prop}
116: \label{ontoprop}
117: The homomorphism $i$ is onto if and only if every Heegaard splitting that is homeomorphic to $\Sigma$ is isotopic to $\Sigma$.
118: \end{Prop}
119:
120: Li~\cite{li:wald} showed that an atoroidal manifold has only finitely many isotopy classes of strongly irreducible Heegaard splittings. This implies that for strongly irreducible splittings of atoroidal manifolds, the image of $i$ has finite index in $Mod(M)$.
121:
122: For toroidal manifolds, Morimoto and Sakuma~\cite{ms:tunnels} have found a number of tunnel number one knot complements and Bachman and Derby-Talbot~\cite{bdt:seif} have found Seifert fibered spaces with an infinite number of homeomorphic, non-isotopic Heegaard splittings. For all these Heegaard splittings, the image of $i$ has infinite index.
123:
124: The question of the kernel of $i$ appears to have no classical analogue. It can be broken into three cases as follows: For each element $\phi$ of $Mod(M, \Sigma)$, the restriction of $\phi$ to $\Sigma$ is an element of the mapping class group of $\Sigma$. We can classify each element of $Mod(M, \Sigma)$ as periodic, reducible or pseudo Anosov by whether the restriction to $\Sigma$ is periodic, reducible or pseudo Anosov.
125:
126: If the automorphism of $\Sigma$ is trivial isotopic then it is trivial isotopic on each handlebody, so $\phi$ is trivial in $Mod(M, \Sigma)$. Thus the restriction map from $Mod(M, \Sigma)$ to $Mod(\Sigma)$ is an injection. An element of $Mod(M, \Sigma)$ will have finite order if and only if its restriction to $\Sigma$ has finite order.
127:
128: The literature on automorphisms of Heegaard splittings is varied and disconnected. As part of a study of tunnel number one knots in $S^3$, Goeritz~\cite{goer:s3} showed that the mapping class group of a genus two Heegaard splitting of $S^3$ is finitely generated. Scharlemann~\cite{schar:mcg} recently published a new proof of this result.
129:
130: As an application of a number of results on the geometry of the curve complex, Namazi~\cite{nam:mcg} showed that for sufficiently high distance Heegaard splittings, the mapping class group of the Heegaard splitting is finite. One can deduce the same result for Heegaard splittings that satisfy Lustig and Moriah's double rectangle condition~\cite{lm:dblrect}. If a Heegaard splitting has one or more diagrams satisfying the double rectangle condition then these are permuted by the mapping class group, and Lustig and Moriah have shown that every Heegaard splitting has at most finitely many such diagrams.
131:
132: Futer~\cite{fu:invol} showed that the only tunnel number one knots with an automorphism of the complement that reverses the orientation of the knot and preserves a genus two Heegaard splitting are two bridge knots. The genus two Heegaard splittings are those coming from level tunnels joining the two maxima or the two minima.
133:
134: In order to study automorphisms of 3-manifolds, Birman and Rubinstein~\cite{rb:homeo} have classified the mapping class groups of a number of one sided Heegaard splittings. Zimmermann~\cite{zim:Hur} used automorphisms of Heegaard splittings to find hyperbolic 3-manifolds with large mapping class groups and Oertel~\cite{oer:hbdy} has given a nice characterization of automorphisms of handlebodies. Oertel and Carvalho~\cite{oc:auts} used this to characterize automorphisms of reducible manifolds.
135:
136: Casson and Long~\cite{cl:alg} found an algorithm to determine if a pseodo Anosov map on the boundary of a handlebody extends to the entire handlebody. They did this not in the context of Heegaard splittings, but to determine when a knot is not homotopically ribbon. Long later found the first (and so far only) example of an irreducible Heegaard splitting with a pseudo Anosov automorphism~\cite{long:twoh}.
137:
138:
139:
140:
141: \section{Isometries and Handlebodies}
142: \label{hbodysect}
143:
144: Given a finite subgroup $G$ of $Mod(M, \Sigma)$ we would like to find an isomorphic subgroup $G'$ of $Aut(M, \Sigma)$ that maps onto $G$ by the inclusion map. The Nielsen Realization Theorem ~\cite{ker:niels} implies that there is a subgroup $G''$ of $Aut(\Sigma)$ with this property. To extend this into the handlebodies, we will find systems of disks that behaves nicely under $G''$. Futer~\cite{fu:invol} uses a similar method in his analysis of automorphisms of genus two Heegaard splittings of knot complements.
145:
146: Given a collection $L$ of pairwise disjoint, simple closed curves in a surface $\Sigma$, a simple closed curve $\ell$ that is disjoint from $L$ will be called \textit{essential relative to $L$} if $\ell$ does not bound a planar surface in the complement of $L$ with boundary on $L$. If $\Sigma$ is the boundary of a handlebody $H$, we will say that an isometry $\phi : \Sigma \rightarrow \Sigma$ extends to an automorphism of $H$ if $\phi$ is the restriction to $\Sigma$ of an automorphism of $H$.
147:
148: \begin{Lem}
149: \label{mindistlem}
150: Let $H$ be a handlebody whose boundary $\Sigma = \partial H$ is endowed with a hyperbolic structure. Let $L$ be a (possible empty) collection of simple closed geodesics and let $\ell$ be a simple closed geodesic in $\Sigma$ that bounds a disk in $H$ and is essential relative to $L$. Assume the length of $\ell$ is minimal over all the simple closed geodesics that bound disks and are essential relative to $L$. If $\phi : \Sigma \rightarrow \Sigma$ is an isometry of $\Sigma$ that extends to an automorphism of $H$ and sends $L$ onto itself then $\phi$ sends $\ell$ onto itself or onto a disjoint loop.
151: \end{Lem}
152:
153: \begin{proof}
154: Given $H$, $\Sigma$, $L$, $\ell$ and $\phi$ as above, let $\ell'$ be the image $\phi(\ell)$. This loop is disjoint from $L$ because $\phi$ takes the set $L$ onto itself. Moreover, $\ell'$ is essential relative to $L$. If $\ell'$ is disjoint from $\ell$ or equal to $\ell$ then we're done. Otherwise, because $\ell$ and $\ell'$ are geodesics, they are transverse and $\ell' \cap \ell$ is a non-empty, finite collection of points.
155:
156: Because $\phi$ extends to an automorphism of $H$, the loop $\ell'$ bounds a disk in $H$. Let $D$, $D'$ be properly embedded disks in $H$ whose boundaries are $\ell$, $\ell'$, respectively. We can choose $D$ and $D'$ transverse to each other. If $D \cap D'$ contains a loop then we can compress $D$ along an innermost loop in $D'$, removing a loop of intersection. Thus by minimizing $D \cap D'$, we can assume that $D \cap D'$ consists of a collection of arcs.
157:
158: An arc $\alpha$ in $D \cap D'$ is \textit{outermost} in $D$ if a component of $D \setminus \alpha$ is disjoint from $D'$. The component of $D \setminus \alpha$ disjoint from $D'$ is called an \textit{outermost disk} in $D$. An outermost disk in $D'$ is defined similarly.
159:
160: The boundary of an outermost disk $E$ in $D$ consists of the outermost arc $\alpha$ and the arc $\partial E \cap \partial D$ in $\Sigma$. For each outermost disk $E$ in $D$ or $D'$, the \textit{length} of $E$ is the length of the arc $\partial E \cap \partial D$ in the hyperbolic metric on $\Sigma$. Let $E$ be an outermost disk of minimal length over all the outermost disks in $D$ and $D'$.
161:
162: First assume $E$ is contained in $D'$. The outermost arc $\alpha = E \cap D$ cuts $D$ into two disks, $F$ and $F'$. If the boundaries of both the disks $E \cup F$ and $E \cup F'$ bound planar surfaces in the complement of $L$ then the boundary of $F \cup F' = D$ bounds a planar surface in the complement of $L$. This contradicts the assumption that $\ell$ is essential relative to $L$. Thus the boundary of $E \cup F$ or $E \cup F'$ must be essential relative to $L$.
163:
164: Without loss of generality, assume the boundary of $E \cup F$ is essential relative to $L$. If the disk $F'$ is not outermost in $D$ then it contains an outermost disk of $D$. Because $E$ has minimal length over outermost disks in $D$ and $D'$, the arc $F' \cap \partial D$ has length greater than or equal to the length of $E \cap \partial D'$. Thus the length of the boundary of the disk $E \cup F$ is at most the length of $\partial D$.
165:
166: Minimal length geodesics in surfaces are smooth. Because the loop $\partial (E \cup F)$ has corners, it is not minimal, so it is isotopic to a geodesic of strictly shorter length. Because geodesics intersect minimally and $\partial (E \cup F)$ is disjoint from $L$, this shorter loop is disjoint from $L$. It bounds a disk and has length strictly less than that of $\ell$, contradicting the minimality assumption on $\ell$.
167:
168: Next assume the outermost disk $E$ is contained in $D$ rather than $D'$. As noted above, $\ell'$ bounds a disk in $H$ and is essential relative to $L$. Because $\phi$ is an isometry on $\Sigma$, the length of $\ell'$ is equal to that of $\ell$, so $\ell'$ has minimal length over simple closed geodesics that bound disks and are essential relative to $L$. Thus we can use the outermost disk $E$ with $D'$, as we did with $D$, to find a simple closed geodesic that bounds a disk and has strictly shorter length. This again contradicts the minimality assumption and completes the proof.
169: \end{proof}
170:
171:
172: \section{Extending automorphisms}
173: \label{extendsect}
174:
175: We will now use Lemma~\ref{mindistlem} to construct a collection of disks that cut $H$ into balls and whose boundaries are permuted by each element of $G''$.
176:
177: \begin{Lem}
178: \label{permutediskslem}
179: Let $G'' \subset Isom(\Sigma)$ be a finite group of isometries of $\Sigma = \partial H$ such that each isometry extends to an automorphism of $H$. Then there is a collection $L$ of essential, simple closed geodesics in $\Sigma$, bounding pairwise disjoint, properly embedded disks such that $G''$ permutes the loops in $L$ and the complement in $\Sigma$ of $L$ is planar.
180: \end{Lem}
181:
182: \begin{proof}
183: Let $\ell_1 \subset \Sigma$ be a simple closed geodesic that bounds an essential, properly embedded disk in $H$ and whose length is minimal over all simple closed geodesics that bound disks. Such a loop is essential relative to the empty set, so by Lemma~\ref{mindistlem}, every isometry of $\Sigma$ that extends to an automorphism of $H$ sends $\ell_1$ onto itself or onto a disjoint loop. Thus the set $L_1 = G''(\ell_1)$ of images of $\ell_1$ under isometries in $G''$ is a collection of pairwise disjoint, simple closed geodesics.
184:
185: Because each element of $G''$ extends to an automorphism of $H$, each loop of $L_1$ bounds a disk in $H$. If two of these disks have non-empty intersection, they must intersect in a number of loops, which can be removed by compressing. Thus the loops in $L_1$ bound a collection $\mathbf{D}_1$ of pairwise disjoint, properly embedded, essential disks.
186:
187: The complement in a handlebody of pairwise disjoint, properly embedded, essential disks is a collection of balls and handlebodies. If a component $\Sigma \setminus L_1$ is non-planar then the corresponding component $H'$ of $H \setminus \mathbf{D}$ is not a ball. Let $D \subset H'$ be a disk that is properly embedded in $H$, disjoint from $\mathbf{D}_1$ and essential in $H'$.
188:
189: The boundary of $D$ is essential relative to $L_1$ and if we isotope $\partial D$ to a geodesic in $\partial H$, it will remain disjoint from $L_1$. Thus there is a simple closed geodesic that is essential relative to $L_1$ and bounds a disk in $H$. Let $\ell_2$ be a minimal length such geodesic.
190:
191: The loops of $L_1$ are permuted by each isometry of $G''$ so by Lemma~\ref{mindistlem}, the images of $\ell_2$ in the isometries of $G''$ are pairwise disjoint. Let $L_2 = L_1 \cup G''(\ell_2)$ be the result of adding these loops into $L_1$
192:
193: This process can continue until we have constructed a collection $L = L_n$ of loops whose complement is planar. These loops are simple closed geodesics which bound pairwise disjoint, properly embedded disks in $H$ and are permuted by $G''$.
194: \end{proof}
195:
196: \begin{Coro}
197: \label{extendhlem}
198: Let $G'' \subset Isom(\Sigma)$ be a finite group of isometries of $\Sigma = \partial H$ such that each isometry extends to an automorphism of $H$. Then there is an isomorphic subgroup $G' \subset Aut(H)$ such that each element of $G'$ restricts to an element of $G''$ and the set of fixed points form boundary parallel arcs and graphs in $H$.
199: \end{Coro}
200:
201: \begin{proof}
202: We will construct the group $G'$ in the piecewise linear category. By Lemma~\ref{permutediskslem}, let $L$ be a collection of simple closed curves that are permuted by $G''$ and bound a collection $\mathbf{D}$ of disks in $H$. Choose a triangulation of $\Sigma$ such that $L$ is contained in the 1-skeleton, each isometry of $G''$ is simplicial and each fixed point is a vertex of the triangulation.
203:
204: Extend this triangulation to each disk by placing a vertex in the center of the disk and extending edges out to the vertices on the boundary. Then for each $g \in G''$, we can extend $g$ to the disks by the unique simplicial map that agrees with $g$ on the boundary of the disk.
205:
206: The complement of the disks is a collection of balls whose boundaries are now triangulated. Extend this triangulation to the balls by placing a vertex in the center of each and extending edges and faces to the boundary. We can thus extend $g$ to $H$ by the unique simplicial map that agrees with $g$ on the boundaries of the balls.
207:
208: For each element $g$ of $G''$, there is a unique simplicial map of $H$ that agrees with $g$ on the boundary. The set $G' \subset Aut(H)$ consisting of these maps is a subgroup such that each element of $G'$ restricts to an element of $G''$. The fixed point set in $H$ consists of the edges that have one vertex at a fixed point in $\partial H$ and the other vertex in the interior of $H$, or vertices at the center of a disk and a ball in the interior. Any properly embedded arcs and graphs consisting of these edges are boundary parallel.
209: \end{proof}
210:
211:
212: \begin{proof}[Proof of Theorem~\ref{mainthm1}]
213: Let $G \subset Mod(M, \Sigma)$ be a finite subgroup of the mapping class group of a Heegaard splitting $(\Sigma, H_1, H_2)$. The restriction of $G$ to $\Sigma$ is a subgroup of the mapping class group $Mod(\Sigma)$. By the Nielsen realization Theorem~\cite{ker:niels}, there is a hyperbolic structure on $\Sigma$ and a subgroup $G''$ of $Isom(\Sigma)$ such that the restriction map from $G$ to $G''$ is an isomorphism.
214:
215: By Corollary~\ref{extendhlem}, the group $G''$ extends to a finite subgroup of $Aut(H_1)$ and to a finite subgroup of $Aut(H_2)$. Thus $G''$ extends to a subgroup $G'$ of $Aut(M)$.
216:
217: Let $N$ be the quotient of $M$ by orbits of $G'$. Because $G''$ is piecewise-linear and finite, the inclusion map $M \rightarrow N$ is a (possibly branched) finite cover and $G''$ is the group of deck transformations for this cover. Because the set of fixed points is boundary parallel in each handlebody, the preimage in $N$ of each handlebody is a handlebody, the Heegaard surface descends to a Heegaard surface $\Sigma'$ and the branch set is in bridge position with respect to $\Sigma'$.
218: \end{proof}
219:
220:
221:
222:
223:
224: \section{Deck transformations}
225: \label{decksect}
226:
227: Theorem~\ref{mainthm1} allows us to pose the question of periodic elements of the kernel in terms of deck transformations of finite covers. Or, equivalently, we can think of the deck transformations as a group action on $M$ by the finite subgroup of $Mod(M,\Sigma)$. This allows us to employ a recent Theorem of Dinkelbach and Leeb. The following is Theorem 1.2 from~\cite{dl:ricci}:
228:
229: \begin{Thm}[Dinkelbach and Leeb~\cite{dl:ricci}]
230: \label{trivisolem}
231: Any smooth action by a finite group on a closed hyperbolic 3-manifold is smoothly conjugate to an isometric action.
232: \end{Thm}
233:
234:
235: \begin{proof}[Proof of Theorem~\ref{mainthm2}]
236: Let $(\Sigma, H_1, H_2)$ be a Heegaard splitting of a hyperbolic 3-manifold $M$ and let $\phi$ be a periodic element of $Mod(M, \Sigma)$. Let $G$ be the (finite) subgroup of $Mod(M, \Sigma)$ generated by $\phi$. By Theorem~\ref{mainthm1}, $M$ is a (possibly branched) finite cover of a manifold $N$ such that $\phi$ is induced by a deck transformation of the covering. The deck transformations induce a smooth action of $M$ by $G$, so Theorem~\ref{trivisolem} implies that the action by $G$ is conjugate to an action by isometries. No isometry of a hyperbolic 3-manifold is isotopy trivial, so $\phi$ is conjugate to an isotopy non-trivial automorphism of $M$. This implies that $\phi$ is isotopy non-trivial, and therefore $\phi$ is not in the kernel of the homomorphism $i : Mod(M,\Sigma) \rightarrow Mod(M)$.
237: \end{proof}
238:
239: \begin{proof}[Proof of Corollary~\ref{maincoro}]
240: Scharlemann and Tomova~\cite{st:dist} showed if the distance of a genus $g$ Heegaard splitting $\Sigma$ is greater than $2g$ then every genus $g$ Heegaard splitting of $M$ is isotopic to $\Sigma$. By Proposition~\ref{ontoprop}, this implies that $i$ is onto.
241:
242: Hempel~\cite{hemp:comp} showed that if the distance of a splitting is strictly greater than two then the manifold is atoroidal and not a Seifert fibered space. Namazi~\cite{nam:mcg} showed that for every genus $g$, there is a constant $K_g$ such that every Heegaard splitting of distance greater than $K_g$ has a finite mapping class group. In particular for such a Heegaard splitting, the kernel of $i$ is finite.
243:
244: Assuming the geometrization conjecture holds, if $d(\Sigma) \geq max \{2g+1, K_g\}$ then $M$ is atoroidal and Seifert fibered so $M$ is hyperbolic and Theorem~\ref{mainthm2} implies the kernel of $i$ contains no non-trivial finite order elements. However, Namazi's Theorem implies that $Mod(M,\Sigma)$ is finite so no non-trivial element of $Mod(M,\Sigma)$ can be in the kernel of $i$. Since $i$ is onto and one-to-one, the homomorphism $i$ is in fact an isomorphism.
245: \end{proof}
246:
247:
248:
249: \section{Weakly Reducible Splittings}
250: \label{wrsect}
251:
252: A Heegaard splitting $(\Sigma, H_1, H_2)$ is \textit{weakly reducible} if there are properly embedded, essential disks $D_1$, $D_2$ in $H_1$, $H_2$, respectively such that the boundaries of $D_1$ and $D_2$ are disjoint in $\Sigma$. In this section we show that for every weakly reducible Heegaard splitting, the kernel of the induced homomorphism $i$ contains infinite order, reducible elements. However, we begin by showing that a slightly weaker condition is enough to imply this.
253:
254: \begin{Lem}
255: \label{twopointslem}
256: Let $(\Sigma, H_1, H_2)$ be a Heegaard splitting of genus at least 3. If there are
257: essential, properly embedded disks $D_1 \subset H_1$ and $D_2 \subset H_2$ whose boundaries intersect transversely in precisely two points then the kernel of $i$ contains an infinite order reducible element or the boundary loops of $D_1$ and $D_2$ are isotopic.
258: \end{Lem}
259:
260: \begin{proof}
261: Let $(\Sigma, H_1, H_2)$ be a Heegaard splitting of a manifold $M$ of genus at least 3 and let $D_1 \subset H_1$, $D_2 \subset H_2$ be properly embedded, essential disks whose boundaries are not isotopic in $\Sigma$. Assume $\partial D_1 \cap \partial D_2$ consists of two points. If we orient $\partial D_1$ and $\partial D_2$ then at each intersection the orientations of the loops will induce an orientation of $\Sigma$. We have two cases to consider: when the induced orientations agree and when they disagree.
262:
263: Let $N$ be the closure of a regular neighborhood of $D_1 \cup D_2$. Because there are two points of intersection, $N$ is a solid torus, as in Figure~\ref{localfig}. If the orientations at the two intersections agree then the surface $\Sigma \cap N$ will be a four punctured sphere whose boundary consists of four simple closed curves in $\partial N$, as shown on the left side of Figure~\ref{disksfig}. If the orientations disagree then $\Sigma \cap N$ will be a twice punctured torus, whose boundary consists of two simple closed curves in $\Sigma$, shown on the right side of Figure~\ref{disksfig}. The loops $\Sigma \cap \partial N$ are parallel longitudes in $\partial N$.
264: \begin{figure}[htb]
265: \begin{center}
266: \includegraphics[width=1.5in]{local.eps}
267: \caption{A neighborhood of the two disks is a solid torus.}
268: \label{localfig}
269: \end{center}
270: \end{figure}
271: \begin{figure}[htb]
272: \begin{center}
273: \includegraphics[width=2.5in]{disks.eps}
274: \caption{If disks intersect in two points, the induced orientations at the intersections may agree or disagree.}
275: \label{disksfig}
276: \end{center}
277: \end{figure}
278:
279: In both cases, ``spinning'' $N$ along its longitude induces an automorphism of the solid torus $N$ that fixes the boundary of $N$ and induces an automorphism of $\Sigma \cap N$ consisting of Dehn twists along loops parallel to $\Sigma \cap \partial N$. If the induced automorphism is trivial then each loop of $\Sigma \cap \partial N$ either bounds a disk in $\Sigma$ or cobounds an annulus with a second loop in $\Sigma \cap \partial N$.
280:
281: In the case when the orientations disagree, there are four loops. If two of the loops in $\Sigma \cap \partial N$ are trivial in $\Sigma$ then either one of the loops $\partial D_1$, $\partial D_2$ is trivial in $\Sigma$ or the two loops are parallel. Because we assumed both loops are essential, we conclude that in this case, the two loops are parallel.
282:
283: If each loop in $\Sigma \cap \partial N$ is parallel to a second loop in $\Sigma \cap \partial N$ then $\Sigma$ is the union of the four punctured sphere $\Sigma \cap N$ and two annuli. In this case $\Sigma$ has genus two. We assumed that $\Sigma$ has genus at least 3 so in the case when the orientations at the intersections agree, we conclude that the induced automorphism is non-trivial on $\Sigma$ or $\partial D_1$ and $\partial D_2$ are isotopic.
284:
285: Next consider the case when the orientations at the intersections disagree and $\Sigma \cap N$ is a twice punctured torus. If the two loops in $\Sigma \cap \partial N$ are parallel then $\Sigma$ is the union of a twice punctured torus and an annulus. This surface has genus two. If both loops bound disks then the surface is a closed torus. Because of the assumption that $\Sigma$ has genus at least three, we conclude that the induced automorphism is non-trivial.
286: \end{proof}
287:
288: \begin{Coro}
289: If $(\Sigma, H_1, H_2)$ is a weakly reducible Heegaard splitting then the kernel of $i$ contains a reducible element.
290: \end{Coro}
291:
292: \begin{proof}
293: Let $(\Sigma, H_1, H_2)$ be a weakly reducible Heegaard splitting. A weakly reducible, genus two Heegaard splitting is reducible. We will deal with this case separately at the end, so for now assume $\Sigma$ has genus at least three.
294:
295: Because $\Sigma$ is weakly reducible, there are properly embedded, essential disks $D_1 \subset H_1$, $D_2 \subset H_2$ with disjoint boundaries. If their boundaries are parallel then replace $D_1$ with a second disk in $H_1$ that is disjoint from the original $D_1$ (and therefore from $\partial D_2$) and not parallel to $D_1$.
296:
297: We can thus assume that $\partial D_1$ and $\partial D_2$ are disjoint, but not parallel in $\Sigma$. Because $\Sigma$ is connected, there is an arc $\alpha$ with one endpoint in $\partial D_1$, the second endpoint in $\partial D_2$ and interior disjoint from both loops. Dragging the boundary of $D_1$ along this arc produces a disk $D'_1$ that intersects $\partial D_2$ in exactly two points. The disks $D'_1$, $D_2$ satisfy the conditions of Lemma~\ref{twopointslem}, so we conclude that the kernel of the homomorphism $i$ contains an infinite order, reducible element.
298:
299: In the case when $\Sigma$ has genus two, we note that a weakly reducible, genus two Heegaard splitting is always reducible. Let $D_1 \subset H_1$ be a separating disk whose boundary also bounds a disk in $H_2$. Let $D_2 \subset H_2$ be a non-separating disk whose boundary is disjoint from $\partial D_1$. As in the proof of Lemma~\ref{twopointslem}, we can isotope $\partial D_2$ to intersect $\partial D_1$ in two points. Let $N$ be a regular neighborhood in $\Sigma$ of these two loops. The boundary of $N$ consists of a trivial loop, two parallel loops and one non-trivial, separating loop. Thus a composition of Dehn twists along these loops will not be trivial in $Mod(\Sigma)$. As in Lemma~\ref{twopointslem}, the automorphism is trivial in $Mod(M)$, so this automorphism is in the kernel of $i$.
300: \end{proof}
301:
302: Note that in the case of weakly reducible splittings, the orientations defined by the two intersections agree, so the automorphism of $\Sigma$ consists of Dehn twists along the boundary loops of a four punctured sphere. However, one of these loops is trivial in $\Sigma$, so we are in fact twisting along the boundary loops of a pair of pants, specifically a regular neighborhood of $\partial D_1 \cup \alpha \cup \partial D_2$. (Here $\alpha$ is the arc constructed in the proof.)
303:
304:
305:
306: \section{Open book decompositions}
307: \label{opensect}
308:
309: Let $M$ be a 3-manifold. An open book decomposition for $M$ is an ordered pair $(L, \phi)$ where $L \subset M$ is a link and $\phi : (M \setminus L) \rightarrow S^1$ is a fibration of the complement of $L$. One can construct a Heegaard splitting for $M$ from the open book decomposition $(L,\phi)$ as follows:
310:
311: Parametrize $S^1$ as the interval $[0,1]$ with its endpoints identified. The preimage $\phi^{-1}((0,\frac{1}{2}))$ is homeomorphic to the interior of a punctured surface cross an interval. Its closure in $M$ is a handlebody $H_1$ which is the union of $L$ and $\phi^{-1}([0,\frac{1}{2}])$. The closure of the set $\phi^{-1}((\frac{1}{2},1))$ is a second handlebody $H_2$ such that $H_1 \cap H_2$ is a surface $\Sigma$ which is the union of $L$ and the surfaces $\phi^{-1}(\{0,\frac{1}{2}\})$. The triple $(\Sigma, H_1, H_2)$ is the Heegaard splitting for $M$ induced from $(L, \phi)$.
312:
313: We will find automorphisms of a Heegaard splitting coming from an open book decomposition using the same method employed by Birman and Rubinstein~\cite{rb:homeo} for one sided Heegaard splittings.
314:
315: \begin{Lem}
316: If $(\Sigma, H_1, H_2)$ is induced by an open book decomposition of the ambient manifold then the kernel of $i$ contains a reducible element.
317: \end{Lem}
318:
319: \begin{proof}
320: Let $(L, \phi)$ be an open book decomposition of a 3-manifold $M$. Let $N$ be a closed regular neighborhood of the link $L$. The complement of $N$ is homeomorphic to the result of gluing the surface product $F \times [0,1]$ by a map $\psi : (F \times \{0\}) \rightarrow (F \times \{1\})$.
321:
322: There is a family $\{h_t\}$ of maps from the complement of $N$ to itself such that $h_0$ is the identity and $h_t$ sends each leaf $F \times \{x\}$ onto $F \times \{x + t\}$ where $x + t$ is taken modulo 1. Each $h_t$ extends to an automorphism of $M$ by rotating the solid tori of $N$.
323:
324: For each integer $n$, $h_n$ takes the surface $\Sigma$ to itself and each handlebody onto itself. The induced automorphism on $\Sigma$ fixes the loops $L \subset \Sigma$ and restricts to $\psi_n$ on the complement of $L$.
325: \end{proof}
326:
327: Note that all Heegaard splittings induced by open book decompositions have distance at most two: If $\Sigma$ is induced by an open book decomposition, with leaf $S$, let $\alpha$ be a properly embedded, essential arc in $S$ and $\beta$ a disjoint, essential, simple closed curve in $S$. We can construct $M$ by taking $S \times [0,1]$, gluing the top and bottom surfaces together, then capping off the torus boundary.
328:
329: The Heegaard splitting is determined by the handlebodies $S \times [0,\frac{1}{2}]$ and $S \times [\frac{1}{2}, 1]$ in which $\alpha \times [0,\frac{1}{2}]$ and $\alpha \times [\frac{1}{2}, 1]$, respectively, are properly embedded, essential disks. Both disks are disjoint from the loop $\beta \times \{\frac{1}{2}\}$, which is essential in the Heegaard surface. Thus the splitting has distance at most two.
330:
331: Both the constructions we have described of splittings with reducible automorphisms produce Heegaard splittings with distance at most two. There are, however, two constructions that potentially produce higher distance splittings:
332:
333: For the first construction, let $S$ be a once punctured surface and let $\phi_0$, $\phi_1$ be fixed point free, orientation reversing, order two automorphisms of $S$. Let $M'$ be the result of taking $S \times [0,1]$ and gluing $S \times \{0\}$ to itself according to $\phi_0$ and $S \times \{1\}$ to itself according to $\phi_1$. If there is an automorphism $\phi$ that commutes with both $\phi_0$ and $\phi_1$ then the product map $\phi \times id$ acts on $M'$, where $id$ is the identity on the interval $[0,1]$.
334:
335: Let $M$ be the result of gluing a solid torus to $\partial M$ such that $S \times \{\frac{1}{2}\}$ is a longitude of the solid torus. The map $\phi \times id$ extends into this solid torus, defining an automorphism $\psi : M \rightarrow M$. Let $H_1$ be the image in $M$ of $S \times [0,\frac{1}{2}]$ and $H_2$ the image of $S \times [0,\frac{1}{2}]$. Each of these sets is a handlebody, so they define a Heegaard splitting that is preserved by the automorphism $\phi$. We will say that such a splitting is induced by a \textit{twisted book decomposition} for $M$.
336:
337: For the second construction, let $(\Sigma, H_1, H_2)$ be a Heegaard splitting with a reducible automorphism $\psi$ fixing a single, non-separating loop $\ell \subset \Sigma$. Moreover, assume there is an order two involution $\phi$ of $M$ that takes $H_1$ onto $H_2$, $H_2$ onto $H_1$, commutes with $\psi$ and whose fixed point set is precisely $\ell$.
338:
339: Let $M'$ be the quotient of $M$ by the automorphism $\phi$. The image in $M'$ of $\Sigma$ is a one sided, surface $S'$ whose boundary is the image of $\ell$. The image of $H_1$ and $H_2$ is a single handlebody of the same genus. A regular neightborhood of $S'$ is a handlebody $H'_1$. The intersection of $H'_1$ with the image of $H_1$ and $H_2$ is a regular neighborhood of the boundary, so the closure of the complement of $H'_1$ is a second handlebody. These handlebodies form a Heegaard splitting $(\Sigma', H'_1, H'_2)$ of $M'$. Moreover, because $\phi$ commutes with $\psi$, the automorphism $\psi$ induces an automorphism of $M$ that preserves $\Sigma$, fixing $\ell$. We will say that $(\Sigma', H'_1, H'_2)$ is a \textit{quotient by a commuting involution}.
340:
341:
342:
343:
344: \section{Fixed loops in handlebody boundaries}
345: \label{fixsect}
346:
347: Given a reducible automorphism $\psi$ of a surface $\Sigma$, there is a collection $L$ of simple closed curves such that $\psi$ takes $L$ onto itself and the restriction of $\psi$ to each component of $\Sigma \setminus L$ is either periodic or pseudo Anosov. If $L$ is minimal (i.e. no proper subset has this property) then any other minimal such collection of loops is isotopic to $L$. This $L$ is called the \textit{canonical reducing curve system} for $\psi$.
348:
349: A simple closed curve $\ell$ in the boundary of a handlebody $H$ is called \textit{disk busting} if for every essential disk $D$, properly embedded in $H$, the boundary of $D$ intersects $\ell$. The following Lemma follows from work of Oertel~\cite{oer:hbdy}:
350:
351: \begin{Lem}
352: \label{oerlem}
353: Let $\psi$ be an automorphism of a handlebody $H$ whose restriction to $\partial H$ is reducible (and not periodic). Let $L$ be the canonical reducing curve system in $\partial H$. At most one loop in $L$ is disk busting in $H$.
354: \end{Lem}
355:
356: \begin{proof}
357: The map $\psi$ is periodic or pseudo Anosov on each component of the complement of $L$. By taking an appropriate power of $\psi$, we can assume that it is in fact pseudo Anosov or isotopic to the identity on each component.
358:
359: If there is a properly embedded, essential disk in $H$ that is disjoint from $L$ then no loop of $L$ is disk busting. Assume this is not the case, i.e. that $\partial H$ is incompressible in the complement of $L$.
360:
361: Lemma 2.13 of Oertel's paper~\cite{oer:hbdy} states that if $\psi$ is pseudo Anosov on a component $C$ of the complement of $L$ then there is an embeddeing of an I-bundle into $H$, sending the horizontal boundary into $\partial H$ such that the restriction of $\psi$ to $C$ is induced by a bundle automorphism. The vertical boundary of the I-bundle is a collection of properly embedded annuli in $H$.
362:
363: Assume an annulus $A$ in the vertical boundary of such an I-bundle is boundary parallel in $H$. Then the two boundary components of $A$ are parallel to a loop $\ell \subset L$. Let $\alpha$ be an essential arc in the base space of the I-bundle with both its endpoints in the boundary component that lifts to $A$. Then $\alpha$ lifts to a disk in $H$ which, after pushing across $A$, becomes properly embedded. This disk intersects $\ell$ in two points, but is disjoint from the remaining loops of $L$. Thus $\ell$ is the only loop in $L$ that can be disk busting.
364:
365: In the case when none of the annuli are boundary parallel, they must also be incompressible because compressing an annulus produces two disks whose boundaries can be made disjoint from $L$, contradicting incompressibility of $\partial H \setminus L$. We will say that an annulus is \textit{essential} if it is incompressible and not boundary parallel. Let $\mathbf{A}$ be the collection of all essential annuli that come from pseudo Anosov components.
366:
367: The complement in a handlebody of a collection of essential annuli is a collection of handlebodies. Because $\psi$ preserves $\mathbf{A}$, it acts on each component of the complement in $H$ of $\mathbf{A}$. For each component $C$ of $H \setminus \mathbf{A}$, either $\psi$ preserves an I-bundle structure on $C$ or $\psi$ consists of Dehn twists along loops of $L$. In the latter case, Oertel's Theorem 1.11~\cite{oer:hbdy} implies that each component of $L$ in $C$ is a boundary component of an essential annulus in $H$ that has both boundary loops isotopic into $L$.
368:
369: In this case, every loop $\ell$ of $L$ is a boundary component of an essential annulus in $H$. (Some loops may be the boundary of more than one such annulus.) Such an annulus is boundary compressible in $H$ and the compression produces an essential, properly embedded disk disjoint from $\ell$. Thus in this case, no loop of $L$ is disk busting.
370: \end{proof}
371:
372: Given a Heegaard splitting $(\Sigma, H_1, H_2)$, if there is an automorphism of $\Sigma$ with canonical fixed set $L$, at most one loop of $L$ is disk busting in each handlebody. If $L$ has three of more components then there is a loop in $L$ that is disjoint from both a disk in $H_1$ and a disk in $H_2$. Thus $\Sigma$ has distance at most two. If $\Sigma$ has higher distance then $L$ must have one or two components. For these cases, we employ the following Lemma:
373:
374: \begin{Lem}
375: \label{oerlem2}
376: Let $\psi$ be an automorphism of a handlebody $H$ whose restriction to $\partial H$ is reducible. Let $L$ be the canonical system of reducing curves in $\partial H$. If $L$ has two components then neither loop is disk busting. If $L$ consists of a single disk busting loop then $\psi$ preserves an I-bundle structure on $H$.
377: \end{Lem}
378:
379: \begin{proof}
380: We saw in the proof of Lemma~\ref{oerlem} that if $\partial H$ is incompressible in the complement of $L$ then every loop of $L$ either cobounds an essential annulus in $H$ with a second loop of $L$ or comes from a cuff of an I-bundle embedded in $H$. If $L$ consists of a single loop $\ell$ then there is no loop to provide a second boundary of an annulus. Thus either $\partial H$ is compressible away from $\ell$ (so $\ell$ is not disk busting) or $\ell$ is the cuff of an I-bundle embedded in $H$. In the second case, because there are no other loops in $L$, this I-bundle structure extends over all of $H$.
381:
382: If $L$ consists of two loops then either the loops cobound an essential annulus in $H$ or both come from cuffs of one or more I-bundles embedded in $H$. In the first case, boundary compressing the annulus produces a disk disjoint from both loops, so neither is disk busting. In the second case, lifting an arc in the base surface with its endpoints on one loop produces a disk disjoint from the other arc. Switching the roles of the two arcs shows that neither loop is disk busting.
383: \end{proof}
384:
385:
386:
387: \section{Reducible automorphisms}
388: \label{redsect}
389:
390: \begin{Thm}
391: \label{caseredthm}
392: If a Heegaard splitting $(\Sigma, H_1, H_2)$ admits a reducible automorphism $\psi$ and $\Sigma$ has distance greater than two then $\Sigma$ is induced by a twisted book decomposition or is a quotient by a commuting involution on a splitting of distance at most two.
393: \end{Thm}
394:
395: \begin{proof}
396: Let $L$ be the canonical system of reducing curves in $\Sigma$. By Lemma~\ref{oerlem} if $L$ has three or more loops then at least one of these loops is not disk busting in either handlebody, so $\Sigma$ has distance at most two. By the first part of Lemma~\ref{oerlem2}, if $L$ consists of two loops then both loops fail to be disk busting in both handlebodies, and again $\Sigma$ has distance at most 2. Thus if the distance is greater than two then $L$ is a single loop.
397:
398: If the loop $\ell = L$ is not disk busting in either handlebody then once again, we have that $\Sigma$ has distance at most two. Otherwise, by Lemma~\ref{oerlem2} $\psi$ must preserve an I-bundle structure on one of the handlebodies. There are two cases to consider: when the I-bundle is a product, and when it is twisted.
399:
400: First assume that $\psi$ preserves a product structure on $H_1$. In other words, there is a homeomorphism from $H$ to $S \times [0,1]$ for a once punctured surface $S$ such that for some automorphism $h : S \rightarrow S$, $\psi$ is the product map $h \times id$. (Here, $id$ is the identity on the interval $[0,1]$.)
401:
402: If $\partial H_2$ is compressible in the complement of $\ell$, let $D_2 \subset H_2$ be a properly embedded, essential disk with boundary disjoint from $\ell$. Then $\partial D_2$ is a simple closed curve in $S \times \{0\}$ or $S \times \{1\}$. In either case, there is an essential arc $\alpha \subset S$ such that $\alpha \times [0,1]$ is disjoint from $\partial D_2$. The disk $\alpha \times [0,1]$ is properly embedded and essential in $H_1$ so $\Sigma$ has distance at most one.
403:
404: Otherwise, by Lemma~\ref{oerlem2}, $\psi$ must also preserve an I-bundle structure on $H_2$. An I-bundle structure is twisted if and only if its horizontal boundary is connected. Because the structure on $H_1$ is a product, $\partial H_1 \setminus \ell$ is disconnected, as is $\partial H_2 \setminus \ell$, so $H_2$ has a product structure. This implies that $(\Sigma, H_1, H_2)$ is induced from an open book decomposition for $M$ and has distance at most two.
405:
406: Next assume the I-bundle structure on $H_1$ is twisted. Then the loop $\ell$ is non-separating so if $\psi$ preserves an I-bundle structure on $H_2$, it is also a twisted bundle structure. Let $S_1$ be the base space for the structure on $H_1$ and $S_2$ for $H_2$. If we define $S = \Sigma \setminus \ell$ then the complement in $M$ of $S_1$, $S_2$ and a neighborhood of $\ell$ is homeomorphic to $S \times (0,1)$. We can reconstruct $M$ and $\Sigma$ be gluing along $S \times \{0\}$ to itself and $S \times \{1\}$ to itself by an involution, then gluing in a neighborhood of $\ell$. In other words, $\Sigma$ is induced by a twisted book decomposition.
407:
408: If $\psi$ does not preserve an I-bundle structure on $H_2$ then by Lemma~\ref{oerlem2}, $\partial H_2$ is compressible in the complement of $\ell$. Let $S_1$ be the interior of the base surface for the I-bundle structure on $H_1$. The complement in $H_1$ of $S_1$ is an open neighborhood of $\Sigma \setminus \ell$ and the complement in $M$ of $S_1$ is a neighborhood of $H_2$, homeomorphic to the interior of a handlebody.
409:
410: Let $H$ be a handlebody and let $f : H \rightarrow M$ send the interior of $H$ onto $M \setminus S_1$. The preimage in $H$ of $\ell$ is a single loop in $\partial H$ and the map $f$ is two to one on boundary of $H$ away from this loop. Thus points in $\partial H \setminus f^{-1}(\ell)$ are identified in pairs. Let $M'$ be the result of taking two copies, $H'_1$ and $H'_2$ of $H$ and identifying each point in $\partial H'_1$ with the point in $H'_2$ with which it is identified by $f$. The two handlebodies define a Heegaard splitting $(\Sigma', H'_1, H'_2)$ of $M'$.
411:
412: We assumed that there is a compressing disk for $H_2$ disjoint from $\ell$. This implies that there are compressing disks in $H'_1$ and $H'_2$ disjoint from $f^{-1}(\ell)$. Thus $\Sigma'$ has distance at most 2. The automorphism $\psi$ determines an automorphism of $\Sigma'$. Moreover, because the construction of $\Sigma'$ is symmetric in $H'_1$ and $H'_2$, there is an automorphism of $M'$ that swaps $H'_1$ and $H'_2$. This automorphism preserves $\ell$ and commutes with $\psi$ so $(\Sigma, H_1, H_2)$ is a quotient by a commuting involution on a Heegaard splitting of distance at most two.
413: \end{proof}
414:
415:
416: \begin{proof}[Proof of Theorem~\ref{redthm}]
417: Let $(\Sigma, H_1, H_2)$ be a Heegaard splitting that admits a reducible automorphism. By Theorem~\ref{caseredthm}, if $\Sigma$ has distance greater than two then it falls into one of two classes. We will show that both of these classes have distance at most four.
418:
419: Assume $\Sigma$ is a quotient by a commuting automorphism. In particular, assume that $H_1$ has the structure of a twisted I-bundle over $S$ with $\ell$ the boundary of $S$ and that $\partial H_2$ is compressible in the complement of $\ell$. Let $\alpha$ be a properly embedded, essential arc in $S$. It lifts to a properly embedded, essential disk $D$ in $H_1$ that intersects $\ell$ in two points. Thus there is a simple closed curve $\ell'$ that is disjoint from both $\ell$ and $\partial D$. Finally, because $\partial H_2$ is compressible in the complement of $\ell$, there is a properly embedded, essential disk in $H_2$ with boundary disjoint from $\ell$. Thus the distance of $\Sigma$ is at most three.
420:
421: Next assume that $\Sigma$ is induced by a twisted book decomposition. Then there are I-bundle structures for $H_1$ and $H_2$ over $S_1$ and $S_2$, respectively, such that $\ell$ is both the boundary of $S_1$ and of $S_2$. By lifting an arc in $S_1$, as in the previous case, we find a disk in $H_1$ that is distance two from $\ell$. The same construction in $H_2$ produces a disk there that is distance two from $\ell$. Thus the distance of $\Sigma$ is at most four.
422: \end{proof}
423:
424: Given a twisted book decomposition of a manifold $M$, the complement of the top and bottom leaves and the spine is homeomorphic to surface $\times (0,1)$ If we take two copies of the complement and glue them together to form a double cover, the resulting manifold is a surface bundle. Thus a twisted book decomposition has a double branched cover that is a standard open book decomposition. The monodromy of this open book decomposition is the composition of the two orientation reversing maps that define the twisted book decomposition.
425:
426: If $(\Sigma, H_1, H_2)$ is a quotient by a commuting involution and has distance greater than two then there is a one sided, incompressible surface $S$ in $H_1$ with boundary on the fixed loop $\ell$. The restriction of the automorphism $\psi$ of $\Sigma$ to $S$ is pseudo Anosov. The complement in $M$ of $\ell$ must therefore be atoroidal or Seifert fibered because otherwise the canonical incompressible tori would meet $S$ in a collection of essential loops that must be permuted by $\psi$.
427:
428: If the complement is Seifert fibered then $\ell$ is the spine of an open book decomposition for $M$. Otherwise, $M \setminus \ell$ is hyperbolic so the automorphism group of $M \setminus \ell$ is finite and some power $\psi^n$ is isotopy trivial in $M \setminus \ell$. The surface $\Sigma \setminus \ell$ compresses into $H_2$ (but not into $H_1$). Let $\Sigma'$ be the result of compressing $\Sigma$ as far as possible. This surface is incompressible in $M \setminus \ell$ and the action of $\psi$ takes $\Sigma'$ onto itself. If the restriction of $\psi^n$ is not the identity on the incompressible surface $\Sigma'$ then $M \setminus \ell$ must have a finite cover that is a surface bundle. Thus in the case of quotients by a commuting involution, either $\psi$ acts periodically on a properly embedded surface in $H_2$ with boundary parallel to $\ell$ or $\ell$ lifts to the spine of an open book decomposition of a finite branched cover of $M$ (branched over $\ell$).
429:
430:
431:
432:
433: \section{Stabilized Splittings}
434: \label{stabsect}
435:
436: The goal of this section is to show that pseudo Anosov automorphisms of Heegaard splittings are quite common, occurring in particular in every stabilized Heegaard splitting of genus three or more. The same methods can be used to find pseudo Anosov automorphisms of stabilized genus two splittings. The pseudo Anosov automorphisms are compositions of reducible automorphisms.
437:
438: It is an open question whether there are Heegaard splittings with only pseudo Anosov automorphisms. Such a splitting would need to be strongly irreducible. Pseudo Anosov automorphisms do occur in irreducible Heegaard splittings, but there are no general methods known for constructing such splittings. The only known example was found by Long~\cite{long:twoh}.
439:
440: \begin{Thm}
441: If $(\Sigma, H_1, H_2)$ is a stabilized Heegaard splitting of genus at least three then $Mod(M, \Sigma)$ contains a pseudo Anosov element in the kernel of $i$.
442: \end{Thm}
443:
444: In fact, this method implies the same result for stabilized genus two Heegaard splittings, but we will allow the reader to modify the proof for those cases.
445:
446: \begin{proof}
447: Let $(\Sigma, H_1, H_2)$ be a stabilized Heegaard splitting of a manifold $M$. Then there is an embedded sphere $S \subset M$ bounding a ball $B$ such that $S \cap \Sigma$ is a single loop and $\Sigma \cap B$ is a punctured torus. If we replace the subsurface $\Sigma \cap B$ with one of the disks in the complement $S \setminus \Sigma$, the result is a Heegaard surface $\Sigma'$ of a Heegaard splitting $(\Sigma', H'_1, H'_2)$.
448:
449: We can reconstruct $\Sigma$ from $\Sigma'$ by choosing a ball $B \subset M$ such that the intersection $B \cap \Sigma$ is a disk $D \subset \Sigma$, then replacing $D$ with a standardly embedded punctured torus in $B$. The sphere $\partial B$ becomes a reducing sphere for the new surface (which is isotopic to $\Sigma$.)
450:
451: Assume there is a properly embedded, essential disk $D_1 \subset H_1$ such that the intersection $D \cap \partial D_1$ consists of a single arc. Let $\alpha \subset D_1$ be a properly embedded arc whose endpoints are in $D$. Stabilizing $\Sigma'$ at $D$ is equivalent to attaching an annulus to $\Sigma'$ along $\alpha$. The reducing sphere $\partial B$ is isotopic to the union of the boundary of a regular neighborhood of $\alpha$ and one of the disk components of $D_1 \setminus \alpha$. There is a second reducing sphere defined by the union of a regular neighborhood of $\alpha$ and the second component of $D_1 \setminus \alpha$. The loops of intersection between $\Sigma$ and these spheres are shown in Figure~\ref{diskstabfig}. As shown on the right of the figure, the loop determined by the second reducing sphere intersects $\Sigma \cap B$ in two arcs and intersects $\Sigma'$ in a pair of arcs parallel to $\partial D_1 \setminus D$.
452: \begin{figure}[htb]
453: \begin{center}
454: \includegraphics[width=3in]{diskstab.eps}
455: \caption{After adding a stabilization along an arc in a disk $D$ in $H_1$, there is a second reducing sphere defined by $D$.}
456: \label{diskstabfig}
457: \end{center}
458: \end{figure}
459:
460: Because $\Sigma$ has genus at least three, $\Sigma'$ has genus at least two. Hempel~\cite{hemp:comp} showed that the diameter of the set of curves in the curve complex bounding loops in a handlebody of genus at least two is infinite. Thus there are disks $D'_1 \subset H'_1$ and $D'_2 \subset H'_2$ whose boundaries are distance at least three apart. This implies that the complement in $\Sigma'$ of the boundaries of these disks is a collection of disks.
461: \begin{figure}[htb]
462: \begin{center}
463: \includegraphics[width=3.5in]{stab.eps}
464: \caption{By placing the stabilization at the intersection of a pair of disks in opposite handlebodies, we find two reducing spheres that cut the surface into disks.}
465: \label{stabfig}
466: \end{center}
467: \end{figure}
468:
469: Let $x$ be a point in the intersection $\partial D'_1 \cap \partial D'_2$ and let $D \subset \Sigma'$ be a small disk containing $x$ and such that $D$ intersects the boundary of each disk $D'_1$, $D'_2$ in a single arc. Construct a surface isotopic to $\Sigma$ by gluing a punctured torus to $\Sigma$ along $D$.
470:
471: As described above, there is a reducing sphere $S_1$ for $\Sigma$ such that $S_1 \cap \Sigma'$ consists of two arcs parallel to $\partial D_1$ and a second reducing sphere $S_2$ that intersects $\Sigma'$ in a pair of arcs parallel to $\partial D_2$, as shown in Figure~\ref{stabfig}. The complement in $\Sigma'$ of $S_1 \cup S_2$ is a subset of the complement of $\partial D_1 \cup \partial D_2$. This is a collection of disks. The complement in $\Sigma \cap B$ of $S_1 \cup S_2$ consists of three disks, as in Figure~\ref{stabfig}.
472:
473: There is an automorphism of $(\Sigma, H_1, H_2)$ that restricts to a Dehn twist along $S_1 \cap \Sigma$ and a second automorphism that restricts to a Dehn twist along $S_2 \cap \Sigma$. Because the complement in $\Sigma$ of the loops $S_1 \cap \Sigma$ and $S_2 \cap \Sigma$ is a collection of disks, a result of Penner~\cite{pen:pseudo} implies that the composition of these two Dehn twists is pseudo Anosov.
474: \end{proof}
475:
476:
477:
478:
479: \bibliographystyle{amsplain}
480: \bibliography{mcg}
481:
482: \end{document}
483:
484: