1: \documentclass[a4paper]{amsart}

2: \usepackage{epsfig,a4wide}

3: 

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

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

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

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

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

9: 

10: \theoremstyle{remark}

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

12: 

13: 

14: \begin{document}

15: 

16: \title{Reducing Dehn fillings and small surfaces}

17: 

18: \author[S. Lee]{Sangyop Lee}

19: \address{School of Mathematics, Korea Institute for Advanced Study,

20:          207-43 Cheongryangri-dong, Dongdaemun-gu, Seoul 130-012, Korea}

21: \email{slee@kias.re.kr}

22: \author[S. Oh]{Seungsang Oh$^1$}

23: \address{Department of Mathematics, Korea University,

24:          1, Anam-dong, Sungbuk-ku, Seoul 136-701, Korea}

25: \email{soh@math.korea.ac.kr}

26: \author[M. Teragaito]{Masakazu Teragaito$^2$}

27: \address{Department of Mathematics and Mathematics Education,

28:          Hiroshima University, Kagamiyama 1-1-1,

29:          Higashi-Hiroshima 739-8524, Japan}

30: \email{teragai@hiroshima-u.ac.jp}

31: \keywords{Dehn filling, reducible, $\partial$-reducible, toroidal, annular}

32: \subjclass[2000]{57M50}

33: \thanks{$^1$Supported by KOSEF research project No. R05-2001-000-00015-0}

34: \thanks{$^2$Partially supported by Japan Society for the Promotion of Science,

35:         Grant-in-Aid for Scientific Research (C) 14540082, 2002.}

36: 

37: \begin{abstract}

38: In this paper we investigate the distances between Dehn fillings

39: on a hyperbolic 3-manifold that yield $3$-manifolds containing

40: essential small surfaces including non-orientable surfaces.

41: Especially we study the situations

42: where one filling creates an essential sphere or projective plane,

43: and the other creates an essential sphere, projective plane,

44: annulus, M\"{o}bius band, torus or Klein bottle,

45: all 11 pairs of such non-hyperbolic manifolds.

46: \end{abstract}

47: 

48: 

49: 

50: \maketitle

51: 

52: 

53: 

54: \section{Introduction} \label{sec:intro}

55: 

56: Let $M$ be a compact, connected, orientable $3$-manifold

57: with a torus boundary component $\partial_0 M$.

58: Let $\gamma$ be a {\it slope\/} on $\partial_0 M$, that is,

59: the isotopy class of an essential simple closed curve on $\partial_0 M$.

60: The $3$-manifold obtained from $M$ by {\em $\gamma$-Dehn filling\/}

61: is defined to be $M(\gamma)=M\cup V_{\gamma}$,

62: where $V_{\gamma}$ is a solid torus glued to $M$ along $\partial_0M$

63: in such a way that $\gamma$ bounds a meridian disk in $V_{\gamma}$.

64: 

65: By a {\it small surface\/} we mean one with non-negative Euler characteristic

66: including non-orientable surfaces.

67: Such surfaces play a special role in the theory of $3$-dimensional manifolds.

68: Thurston's geometrization theorem for Haken manifolds \cite{Th} asserts that

69: a hyperbolic $3$-manifold $M$ with non-empty boundary

70: contains no essential small surfaces.

71: Furthermore, if $M$ is hyperbolic, then the Dehn filling $M(\gamma)$ is also

72: hyperbolic for all but finitely many slopes \cite{Th},

73: and a good deal of attention has been directed towards obtaining

74: a more precise quantification of this statement.

75: 

76: Let us say that a $3$-manifold is of {\em type\/} $\mathcal{S}$,

77: $\mathcal{D}$, $\mathcal{A}$, or $\mathcal{T}$,

78: if it contains an essential orientable small surface which is

79: an essential sphere, disk, annulus or torus,

80: and of {\em type\/} $\mathcal{P}$, $\mathcal{B}$, or $\mathcal{K}$,

81: if it contains a non-orientable small surface which is

82: a projective plane, M\"{o}bius band or Klein bottle, respectively.

83: Especially $3$-manifolds of type $\mathcal{S}$, $\mathcal{D}$,

84: $\mathcal{A}$ and $\mathcal{T}$ are called reducible,

85: $\partial$-reducible, annular and toroidal, respectively.

86: The distance $\Delta(\gamma_1,\gamma_2)$ between two slopes

87: on a torus is their minimal geometric intersection number.

88: The bound $\Delta(X_1,X_2)$ is the least nonnegative number $m$

89: such that if $M$ is a hyperbolic manifold which admits two

90: Dehn fillings $M(\gamma_1)$, $M(\gamma_2)$ of type $X_1$, $X_2$,

91: respectively, then $\Delta(\gamma_1,\gamma_2)\leq m$.

92: Surveys of the known bounds of various choices $(X_1,X_2)$

93: and the maximal values realized by known

94: examples are given in \cite{G2, W3, EW}.

95: 

96: In this paper we consider $M(\gamma_i)$, $i=1,2$ of only the six types

97: $\mathcal{S}$, $\mathcal{P}$, $\mathcal{A}$, $\mathcal{B}$,

98: $\mathcal{T}$, or $\mathcal{K}$.

99: Suppose that $M(\gamma_i)$ contains such a small surface $\widehat{F}_i$.

100: Then we may assume that $\widehat{F}_i$ meets the attached solid torus

101: $V_{\gamma_i}$ in a finite collection of meridian disks,

102: and is chosen so that the number of disks $n_i$ is minimal

103: among all such surfaces in $M(\gamma_i)$.

104: The main results of this paper are the followings.

105: 

106: \begin{theorem} \label{thm:SS}

107: Suppose that $M$ is hyperbolic.

108: If $M(\gamma_1)$ and $M(\gamma_2)$ are of type $\mathcal{S}$ or $\mathcal{P}$,

109: then $\Delta(\gamma_1,\gamma_2) \leq 1$.

110: \end{theorem}

111: 

112: 

113: The original proof of the case $(\mathcal{S},\mathcal{S})$ is very complicated \cite{GL1}.

114: Our proof is remarkably short, although it is based on the analysis of intersections

115: of two surfaces as well as \cite{GL1}.

116: Recently, Hoffman and Matignon \cite{HoM} also gives such a short proof in the almost same line, but

117: ours is still simpler than it.

118: 

119: If a $3$-manifold contains a projective plane, then it is either a reducible manifold or

120: the $3$-dimensional projective space.

121: Hence type $\mathcal{P}$ breaks down into type $\mathcal{S}$ and CYC, which means

122: the class of manifolds with finite cyclic fundamental groups.

123: References are: \cite{BZ2} for $\Delta(\mathcal{S},\mbox{CYC})=1$; \cite{CGLS} for $\Delta(\mbox{CYC},\mbox{CYC})=1$.

124: See also \cite{M,Te} for $\Delta(\mathcal{P},\mathcal{P})=1$.

125: 

126: 

127: \begin{theorem} \label{thm:SA}

128: Suppose that $M$ is hyperbolic.

129: If $M(\gamma_1)$ is of type $\mathcal{S}$ or $\mathcal{P}$,

130: and $M(\gamma_2)$ is of type $\mathcal{A}$ or $\mathcal{B}$,

131: then either $\Delta(\gamma_1,\gamma_2) \leq 1$,

132: or $\Delta(\gamma_1,\gamma_2) = 2$

133: with $n_2 = 2$ when $M(\gamma_2)$ is of type $\mathcal{A}$

134: (or $n_2 = 1$ when of type $\mathcal{B}$).

135: \end{theorem}

136: 

137: For the case $(\mathcal{S},\mathcal{A})$, Wu \cite{W3} showed that $\Delta(\mathcal{S},\mathcal{A})=2$

138: by using the sutured manifold theory,

139: and he asked whether $n_2=2$ when $M(\gamma_1)$ is of type $\mathcal{S}$ and

140: $M(\gamma_2)$ is of type $\mathcal{A}$  with $\Delta(\gamma_1,\gamma_2)=2$ \cite[Question 5.8]{W3}.

141: Our Theorem \ref{thm:SA} gives the affirmative answer of this question.

142: Note that type $\mathcal{B}$ breaks down into types $\mathcal{S}$, $\mathcal{A}$ and $\mathcal{D}$.

143: 

144: 

145: \begin{theorem} \label{thm:ST}

146: Suppose that $M$ is hyperbolic.

147: If $M(\gamma_1)$ is of type $\mathcal{S}$ or $\mathcal{P}$,

148: and $M(\gamma_2)$ is of type $\mathcal{T}$ or $\mathcal{K}$,

149: then either $\Delta(\gamma_1,\gamma_2) \leq 2$,

150: or $\Delta(\gamma_1,\gamma_2) = 3$

151: with $n_2 = 2$ when $M(\gamma_2)$ is of type $\mathcal{T}$

152: (or $n_2 = 1$ when of type $\mathcal{K}$).

153: \end{theorem}

154: 

155: For the case $(\mathcal{S},\mathcal{T})$, Oh \cite{O} and Wu \cite{W2} independently showed that

156: $\Delta(\mathcal{S},\mathcal{T})=3$.

157: (See also \cite{LOT} for its short proof.)

158: Hence Theorem \ref{thm:ST} gives an improvement of this result.

159: In \cite{JLOT}, we have showed the conclusions for two cases $(\mathcal{P},\mathcal{T})$

160: and $(\mathcal{P},\mathcal{K})$.

161: 

162: If a $3$-manifold is of type $\mathcal{K}$,

163: then it is of type $\mathcal{S}$, $\mathcal{T}$ or a Seifert fibered manifold with finite fundamental group (a prism manifold).

164: Indeed, non-orientable cases are necessary to prove orientable cases in our arguments.

165: Such phenomenon is also observed in \cite{GL2} and \cite{GL3}.

166: %%%%%%%%%%%%%%%%%%%%%%%

167: 

168: 

169: 

170: We should emphasize that all 11 cases can be treated in a unified argument in this paper.

171: (Also, we have a plan to continue the study for the remaining pairs of $\mathcal{S,P,D,A,B,T,K}$.)

172: Our main tool in this paper is a two-cornered cycle, which was introduced by Hoffman \cite{H}.

173: Hoffman showed that the disk bounded by a great $x$-cycle contains a pair of

174: specific two-cornered cycles, called a seemly pair,

175: and it can be used to find a new essential sphere and projective plane

176: meeting the attached solid torus in a fewer times than an original surface, leading to a contradiction.

177: We define a slight generalization of a great $x$-cycle, called an $x$-face, and

178: show that it contains a seemly pair by a simpler argument than that in \cite{H} and that

179: such a pair is useful for the types $\mathcal{A}$ and $\mathcal{B}$ as well.

180: 

181: 

182: 

183: 

184: %%%%%%%%%%%%%

185: By virtue of Theorem \ref{thm:SS} and the fact $\Delta(\mathcal{S},\mathcal{D})=0$ \cite{S}

186: (and hence $\Delta(\mathcal{P},\mathcal{D})=0$),

187: we can put the following assumption throughout the paper to simplify the arguments greatly:

188: 

189: 

190: \medskip

191: \noindent\textbf{Assumption.} If $M(\gamma_i)$ is of type $\mathcal{A},\mathcal{B},\mathcal{T}$ or

192: $\mathcal{K}$, then we assume that $M(\gamma_i)$ is neither of type $\mathcal{S}$ nor of type $\mathcal{D}$.

193: \medskip

194: 

195: 

196: 

197: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

198: \section{Graphs of surface intersections} \label{sec:graph}

199: 

200: Hereafter $M$ is a hyperbolic $3$-manifold

201: with a torus boundary component $\partial_0 M$.

202: An orientable surface properly embedded in $M$ is called \textit{essential\/} if it is

203: either (i) incompressible, not boundary parallel and not a sphere, or (ii) a sphere

204: which does not bound a $3$-ball.

205: Note that any essential small surface is also boundary incompressible by our assumption in Section \ref{sec:intro}.

206: In this section we describe how (essential) small surfaces

207: $\widehat{F}_1$ and $\widehat{F}_2$,

208: in $M(\gamma_1)$ and $M(\gamma_2)$ respectively,

209: give rise to labelled intersection graphs $G_i \subset \widehat{F}_i$

210: for $i=1,2$ in a general context.

211: 

212: As in Section \ref{sec:intro},

213: let $\widehat{F}_i$ be a small surface in $M(\gamma_i)$

214: with $n_i = | \widehat{F}_i \cap V_{\gamma_i} |$ minimal.

215: (Recall that if $\widehat{F}_i$ is orientable then it is assumed to be essential.)

216: Then $n_i>0$.

217: Thus $F_i = \widehat{F}_i \cap M$ is a punctured surface properly embedded in $M$,

218: each of whose $n_i$ boundary components has slope $\gamma_i$ on $\partial_0M$.

219: 

220: 

221: \begin{lemma} \label{lem:incomp}

222: For each of six types, $F_i$ is incompressible and

223: boundary incompressible in $M$.

224: \end{lemma}

225: 

226: \begin{proof}

227: For types $\mathcal{S}$, $\mathcal{A}$ and $\mathcal{T}$,

228: the minimality of $n_i$ guarantees that

229: $F_i$ is incompressible and boundary incompressible in $M$.

230: 

231: 

232: For type $\mathcal{P}$, assume that $D$ is a compressing disk for $F_i$.

233: Since $\partial D$ is orientation-preserving on $F_i$,

234: $\partial D$ bounds a disk $D'$ on $\widehat{F}_i$.

235: Since $\mathrm{Int}\, D'$ meets $V_{\gamma_i}$,

236: we can create a new projective plane by replacing $D'$ with $D$,

237: which meets $V_{\gamma_i}$ fewer than $\widehat{F}_i$,

238: contradicting the minimality of $n_i$.

239: Next, assume that $E$ is a boundary compressing disk for $F_i$

240: with $\partial E = a \cup b$,

241: where $a$ is an essential arc in $F_i$ and $b$ lies in $\partial_0 M$.

242: If $a$ joins distinct components of $\partial F_i$,

243: then a compressing disk for $F_i$ is obtained from two parallel copies of $E$

244: and the disk obtained by removing a neighborhood of $b$

245: from the annulus in $\partial_0 M$ cobounded by those components of

246: $\partial F_i$ meeting $a$.

247: Hence both endpoints of $a$ lie in the same component,

248: say $\partial_1 F_i$, of $\partial F_i$.

249: If $n_i \geq 2$, then $b$ bounds a disk $D'$ in $\partial_0 M$

250: together with a subarc of $\partial_1 F_i$.

251: Then $E\cup D'$ gives a compressing disk for $F_i$ in $M$.

252: Therefore $n_i = 1$.

253: Then $M$ would contain a M\"{o}bius band, a contradiction.

254: 

255: 

256: For type $\mathcal{B}$, assume that $D$ is a compressing disk for $F_i$.

257: If $\partial D$, which is orientation-preserving on $F_i$,

258: bounds a disk $D'$ on $\widehat{F}_i$,

259: a new M\"{o}bius band obtained by replacing $D'$ with $D$

260: has fewer $n_i$, a contradiction.

261: Thus $\partial D$ is essential, and indeed separating on $\widehat{F}_i$.

262: Compressing $\widehat{F}_i$ along $D$ would give in $M(\gamma_i)$

263: a projective plane $Q_1$ and a disk $Q_2$ which are disjoint.

264: Note that the core of $V_{\gamma_i}$ must meet both $Q_1$ and $Q_2$

265: because $M$ is hyperbolic.

266: Let $\ell$ be a subarc of the core of $V_{\gamma_i}$

267: which connects $Q_1$ and $Q_2$, meeting them only on its endpoints.

268: Attaching a tube along $\ell$ to $Q_1\cup Q_2$ gives

269: a M\"{o}bius band in $M(\gamma_i)$ which intersects

270: the core of $V_{\gamma_i}$ in $n_i-2$ points, a contradiction.

271: Thus we have shown that $F_i$ is incompressible.

272: The incompressibility of $F_i$ then implies

273: the boundary incompressibility of $F_i$;

274: otherwise, $n_i = 1$ as previous,

275: and furthermore a booundary-compressing disk $E$

276: allows us to isotope the core of $V_{\gamma_i}$

277: into $\widehat{F}_i$ as an orientation-reversing loop.

278: Then $M$ would contain an essential annulus, a contradiction.

279: 

280: 

281: For type $\mathcal{K}$, assume that $D$ is a compressing disk for $F_i$.

282: As previous, $\partial D$ is

283: orientation-preserving and essential on $\widehat{F}_i$.

284: Compressing $\widehat{F}_i$ along $D$ would give in $M(\gamma_i)$

285: either a non-separating 2-sphere or two disjoint projective planes,

286: according as $\partial D$ is non-separating

287: or separating on $\widehat{F}_i$.

288: But this contradicts the irreducibility of $M(\gamma_i)$.

289: The incompressibility of $F_i$ then implies

290: the boundary incompressibility of $F_i$,

291: unless $M$ contains a M\"{o}bius band, as above.

292: \end{proof}

293: 

294: We use $i$ and $j$ to denote $1$ or $2$, with the convention that,

295: when both appear, $\{i,j\} = \{1,2\}$.

296: 

297: 

298: By an isotopy of $F_1$, say, we may assume that $F_1$ intersects

299: $F_2$ transversely. 

300: By Lemma \ref{lem:incomp} it can be assumed that no circle component of $F_1\cap F_2$

301: bounds a disk in $F_1$ or $F_2$. 

302: Let $G_i$ be the graph in $\widehat{F}_i$ obtained by taking

303: as the (fat) vertices the disks $\widehat{F}_i - \mathrm{Int}\, F_i$

304: and as edges the arc components of $F_i \cap F_j$ in $\widehat{F}_i$.

305: Thus the interior of any disk face of $G_i$ is disjoint from $F_j$.

306: We number the components of $\partial F_i\cap \partial_0M$ as $1,2,\cdots,n_i$

307: in the order in which they appear on $\partial_0 M$.

308: On occasion we will use $0$ instead of $n_i$ in short.

309: This gives a numbering of the vertices of $G_i$.

310: Furthermore it induces a labelling of the endpoints of

311: edges in $G_j$ in the usual way (see \cite{CGLS}).

312: Note that $G_1$ and $G_2$ have no trivial loops by Lemma \ref{lem:incomp}.

313: For the sake of simplicity we will say that

314: $F_i$ and $G_i$ are of type $X$ if $M(\gamma_i)$ is of type $X$.

315: 

316: 

317: Since $M$ is hyperbolic, we have the following easy lemma.

318: 

319: 

320: \begin{lemma} \label{lem:n}

321: $n_i \geq 3$ for $G_i$ of type $\mathcal{S}$, and

322: $n_i \geq 2$ for $G_i$ of type $\mathcal{P}$.

323: \end{lemma}

324: 

325: Although $F_i$ of type $\mathcal{P}$, $\mathcal{B}$ or

326: $\mathcal{K}$ is non-orientable,

327: we can establish a parity rule, which plays a crucial role.

328: In fact, this is a natural generalization of the usual parity rule \cite{CGLS}.

329: First, orient all components of $\partial F_i$

330: so that they are mutually homologous on $\partial_0 M$.

331: Let $e$ be an edge of $G_i$.

332: Since $e$ is an arc properly embedded in $F_i$,

333: a regular neighborhood $D$ of $e$ in $F_i$ is a disk in $F_i$.

334: Then $\partial D = a \cup b \cup c \cup d$,

335: where $a$ and $c$ are arcs in $\partial F_i$ with

336: induced orientations from $\partial F_i$.

337: If $a$ and $c$ are directed along $\partial D$,

338: then $e$ is called {\em positive\/}, otherwise {\em negative\/}.  See Figure \ref{fig:parity}.

339: Then we have the {\em parity rule}: {\it 

340: an edge $e$ is positive in $G_i$ if and only if

341: $e$ is negative in $G_j$.}

342: 

343: \begin{figure}[h]

344: \epsfbox{fig01.eps}

345: \caption{A sign of an edge}\label{fig:parity}

346: \end{figure}

347: 

348: The rest of this section will be devoted to

349: several definitions and well known lemmas.

350: Let $x$ be a label of $G_i$.

351: An $x$-{\em edge\/} in $G_i$ is an edge with label $x$ at one endpoint,

352: and an $xy$-edge is an edge with label $x$ and $y$ at both endpoints.

353: If an $x$-edge has an endpoint at a vertex $v$ with label $x$, then it is called

354: an $x$-edge at $v$.

355: Especially, a positive $xx$-edge is called a {\em level $x$-edge\/},

356: which means that

357: the vertex $x$ of the other graph $G_j$ is incident to a negative loop.

358: In particular, $G_i$ does not contain a level $x$-edge unless 

359: $G_j$ is of type $\mathcal{P}$, $\mathcal{B}$ or $\mathcal{K}$.

360: 

361: 

362: An $x$-{\em cycle\/} is a cycle of positive $x$-edges of $G_i$

363: which can be oriented so that the head of each edge has label $x$.

364: A {\em Scharlemann cycle\/} is an $x$-cycle that bounds a disk face of $G_i$,

365: only when $n_j \geq 2$.

366: Each edge of a Scharlemann cycle has the same label pair $\{x,x+1\}$,

367: so we refer to such a Scharlemann cycle as an $(x,x+1)$-Scharlemann cycle.

368: The number of edges in a Scharlemann cycle

369: is called its {\em length}.

370: In particular, a Scharlemann cycle of length two is called an

371: {\em $S$-cycle\/} in short.

372: 

373: Suppose that a Scharlemann cycle $\sigma$ is immediately

374: surrounded by a cycle $\kappa$, that is, each edge of $\kappa$

375: is immediately parallel to an edge of $\sigma$.

376: Then $\kappa$ will be referred to as

377: an {\em extended Scharlemann cycle\/}, only when $n_j \geq 4$.

378: A {\em generalized $S$-cycle\/} is the triple

379: $\{e_1,e_2,e_3\}$ of mutually parallel positive edges in succession

380: and $e_2$ is a level edge, only when $n_j \geq 3$.

381: 

382: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

383: \begin{lemma} \label{lem:Scharsep}

384: Assume that $M(\gamma_j)$ is either of type $\mathcal{S}$, $\mathcal{A}$ or $\mathcal{T}$.

385: If $G_i$ has a Scharlemann cycle

386: then $\widehat{F}_j$ must be separating, and so $n_j$ is even.

387: Furthermore, for cases $\mathcal{A}$ and $\mathcal{T}$,

388: the edges of a Scharlemann cycle do no lie in a disk in $\widehat{F}_j$.

389: \end{lemma}

390: 

391: \begin{proof}

392: Let $E$ be a disk face bounded by a Scharlemann cycle

393: with a label pair, say $\{1,2\}$, in $G_i$.

394: Let $V_{12}$ be the $1$-handle cut from $V_{\gamma_j}$

395: by the vertices $1$ and $2$ of $G_j$.

396: (When $n_j=2$, $V_{12}$ is chosen to meet $\partial E$.)

397: Then tubing $\widehat{F}_j$ along $\partial V_{12}$ and

398: compressing along $E$ gives a new surface $R$ in $M(\gamma_j)$, homeomorphic to $\widehat{F}_j$,

399: that intersects $V_{\gamma_j}$ fewer times than $\widehat{F}_j$.

400: If the original $\widehat{F}_j$ is non-separating, then so is $R$.

401: Thus $R$ is essential for each case, a contradiction.

402: 

403: 

404: If the edges of the Scharlemann cycle lie

405: in a disk $D$ in an annulus or torus $\widehat{F}_j$,

406: then $\mbox{nhd} (D \cup V_{12} \cup E)$ is a once punctured lens space.

407: By the irreducibility of $M(\gamma_j)$, $M(\gamma_j)$ is a lens space, so neither of type $\mathcal{A}$ nor of type $\mathcal{T}$.

408: \end{proof}

409: 

410: 

411: \begin{lemma} \label{lem:Schar}

412: \begin{itemize}

413: \item[(1)] If $G_j$ is of type $\mathcal{S},\mathcal{A}$ or $\mathcal{T}$, 

414: then $G_i$ cannot have a level edge.

415: If $G_j$ is of type $\mathcal{P}$ or $\mathcal{B}$, then

416: $G_i$ has at most one label of level edges.

417: If $G_j$ is of type $\mathcal{K}$,

418: then $G_i$ has at most two labels of level edges.

419: \item[(2)] If $G_j$ is of type $\mathcal{P},\mathcal{B}$ or $\mathcal{K}$,

420: then $G_i$ cannot have a Scharlemann cycle.

421: If $G_j$ is of type $\mathcal{S}$ or $\mathcal{A}$,

422: then any two Scharlemann cycles of $G_i$ have the same label pair.

423: %When $G_j$ is of type $\mathcal{T}$,

424: %$G_i$ has at most two label pairs of Scharlemann cycles.

425: \item[(3)] When $G_j$ is of type $\mathcal{S}$,

426: $M(\gamma_j)$ contains a projective plane if $G_i$ contains an $S$-cycle.

427: \item[(4)] If $G_j$ is of type $\mathcal{S}$, $\mathcal{A}$ or $\mathcal{T}$,

428: then $G_i$ cannot have an extended $S$-cycle.

429: \item[(5)] If $G_j$ is of type $\mathcal{P}$, $\mathcal{B}$ or $\mathcal{K}$,

430: then $G_i$ cannot have a generalized $S$-cycle.

431: \end{itemize}

432: \end{lemma}

433: 

434: 

435: \begin{proof}

436: (1) Let $e$ be a negative loop based at a vertex $x$ in $G_j$.

437: Then $\mbox{nhd} (x\cup e)$ is a M\"{o}bius band in $\widehat{F}_j$.

438: Since only projective plane, M\"{o}bius band and Klein bottle

439: can contain at most one, one and two M\"{o}bius bands respectively,

440: the conclusions follow.

441: 

442: (2) When $G_j$ is of type $\mathcal{P}$, $\mathcal{B}$ or $\mathcal{K}$,

443: assume for contradiction that $G_i$ contains a Scharlemann cycle.

444: Then the construction in the proof of Lemma \ref{lem:Scharsep} gives

445: a new surface homeomorphic to $\widehat{F}_j$ in $M(\gamma_j)$, which meets

446: $V_{\gamma_j}$ fewer times than $\widehat{F}_j$.

447: This contradicts the minimality of $n_j$.

448: 

449: 

450: When $G_j$ is of type $\mathcal{S}$ or $\mathcal{A}$, this is \cite[Theorem 2.4]{GL1} or

451: \cite[Lemma 5.4(2)]{W3}, respectively.

452: 

453: 

454: %%%%%%%%%%%%%%

455: %When $G_j$ is of type $\mathcal{T}$,

456: 

457: 

458: (3) Let $\{e_1,e_2\}$ be an $S$-cycle in $G_i$ with label pair $\{k,k+1\}$.

459: Let $v_h$ be the $h$-th vertex of $G_j$ for $h=k,k+1$, and let $H$ be the part of $V_{\gamma_j}$ between

460: $v_k$ and $v_{k+1}$.

461: Let $D$ be the disk face bounded by the $S$-cycle, and

462: let $A$ be the M\"{o}bius band obtained by taking $H\cup D$ and shrinking $H$ radially to its core.

463: Since $\partial A$ is isotopic to the curve obtained from $e_1\cup e_2\cup v_k\cup v_{k+1}$ by shrinking

464: the vertices to points, $\partial A$ bounds a disk on $\widehat{F}_j$.

465: Thus $M(\gamma_j)$ contains a projective plane.

466: 

467: %%%%%%%%%%%%%%%

468: (4) These are \cite[Lemma 2.3]{W1}, \cite[Lemma 5.4(3)]{W3} and \cite[Lemma 2.10]{BZ1}, respectively.

469: 

470: 

471: (5) Let $\{e_1,e_2,e_3\}$ be a generalized $S$-cycle in $G_i$, where $e_2$ is a level edge with label $k$,

472: and $e_1, e_3$ have the same label pair $\{k-1,k+1\}$.

473: Let $v_h$ be the $h$-th vertex of $G_j$ for $h=k-1,k,k+1$, and

474: let $H$ be the part of $V_{\gamma_j}$ between the vertices $v_{k-1}$ and $v_{k+1}$ containing $v_k$.

475: Then $C=\mbox{nhd}\,(v_k\cup e_2)$ is a M\"{o}bius band in $\widehat{F}_j$.

476: Let $D$ be the disk in $F_i$ representing the parallelism of $e_1$ and $e_3$ and containing $e_2$.

477: Then we have a M\"{o}bius band $A$ from $H\cup D$ as before.

478: Note that $\mathrm{Int}\,A$ is disjoint from $\widehat{F}_j-C$.

479: Since $\partial A$ is isotopic to the curve obtained from $e_1\cup e_3\cup v_{k-1}\cup v_{k+1}$ by shrinking

480: $v_{k\pm 1}$ to points, $\partial A$ is orientation-preserving on $\widehat{F}_j$.

481: 

482: 

483: If $G_j$ is of type $\mathcal{P}$,

484: then $\partial A$ above bounds a disk $E$ in $\widehat{F}_j-C$.

485: Thus $A\cup E$ gives a new projective plane which meets $V_{\gamma_j}$ fewer times than $n_j$, a contradiction.

486: 

487: If $G_j$ is of type $\mathcal{B}$, 

488: then either $\partial A$ bounds a disk in $\widehat{F}_j-C$ or $\partial A$ is parallel to $\partial \widehat{F}_j$.

489: In the former, $M(\gamma_j)$ contains a projective plane, and hence $M(\gamma_j)$ is reducible,

490: contradicting the irreducibility of $M(\gamma_j)$.

491: In the latter, let $A_1$ be the annulus between $\partial A$ and $\partial\widehat{F}_j$.

492: Then $A\cup A_1$ is a M\"{o}bius band in $M(\gamma_j)$ which meets $V_{\gamma_j}$ fewer times than $n_j$.

493: 

494: 

495: If $G_j$ is of type $\mathcal{K}$,

496: then $\partial A$ bounds either a disk or a M\"{o}bius band $B$ in $\widehat{F}_j-C$.

497: In the former, $M(\gamma_j)$  contains a projective plane.

498: By the irreducibility of $M(\gamma_j)$, $M(\gamma_j)$ is the lens space $L(2,1)$.

499: But it is well known that $L(2,1)$ does not contain a Klein bottle \cite{BW}.

500: In the latter, $A\cup B$ gives a new Klein bottle in $M(\gamma_j)$ which meets $V_{\gamma_j}$

501: fewer times than $n_j$ again.

502: \end{proof}

503: 

504: 

505: For simplicity

506: we will call $x$ an $sl$-{\em label\/} of $G_i$

507: (or $sl$-{\em vertex\/} of $G_j$)

508: if $x$ is a label of either a Scharlemann cycle or a level edge

509: in $G_i$, according as $\widehat{F}_j$ is orientable or not.

510: Thus Lemma \ref{lem:Schar} implies that $G_i$ contains at most 2, 1, 2, 1 or 2 $sl$-labels

511: when $G_j$ is of type $\mathcal{S}$, $\mathcal{P}$, $\mathcal{A}$,

512: $\mathcal{B}$ or $\mathcal{K}$, respectively.

513: When $G_j$ is of type $\mathcal{T}$, we do not need an upper bound for the number of $sl$-labels of $G_i$, but

514: will show that at most three labels can be labels of $S$-cycles of $G_i$ unless $M(\gamma_j)$ contains a Klein bottle.

515: (See Claim \ref{claim:Tcase}.)

516: 

517: 

518: 

519: %Let $\mathbf{P}$ (resp. $\mathbf{N}$) denote the upper bounds for the number of

520: %mutually parallel positive (resp. negative) edges in $G_i$. 

521: 

522: 

523: \begin{lemma} \label{lem:parallel}

524: Let $F$ be a family of mutually parallel positive edges in $G_i$, and let $|F|$ denote the

525: number of edges in $F$.

526: \begin{itemize}

527: \item[(1)] If $G_j$ is of type $\mathcal{S}$ or $\mathcal{A}$,

528: $|F|\le\frac{n_j}{2}+1$.

529: Furthermore if $|F|=\frac{n_j}{2}+1$, then the first two or the last two edges of $F$ form an $S$-cycle.

530: \item[(2)] If $G_j$ is of type $\mathcal{P}$ or $\mathcal{B}$,

531: $|F|\le\frac{n_j+1}{2}$.

532: Furthermore if $|F|=\frac{n_j+1}{2}$,

533: then the first or the last edge of $F$ is a level edge.

534: \item[(3)] If $G_j$ is of type $\mathcal{T}$ and $n_j\ge 3$,

535: $|F|\le\frac{n_j}{2}+2$.

536: %Furthermore if $|F|>\frac{n_j}{2}$, then

537: %$F$ contains an $S$-cycle.

538: %then the first two and the last two edges of $F$ form disjoint $S$-cycles.

539: \item[(4)] If $G_j$ is of type $\mathcal{K}$ and $n_j\ge 2$,

540: $|F|\le\frac{n_j}{2}+1$.

541: %Furthermore if $|F|=\frac{n_j}{2}+1$ (when $n_j$ is even),

542: %then the first and the last edges of $F$ are level edges with distinct labels.

543: %And if $|F|=\frac{n_j+1}{2}$ (when $n_j$ is odd),

544: %then the first or the last edge of $F$ is a level edge.

545: \end{itemize}

546: \end{lemma}

547: 

548: \begin{proof}

549: (1) is \cite[Lemma 1.5(1)]{W2} and \cite[Lemma 2.5(2)]{GW}.

550: (We remark that when $G_j$ is of type $\mathcal{A}$ and $n_j=1$, $|F|\le 1$, because

551: a pair of edges cannot be parallel in both graphs \cite[Lemma 2.1]{G1}.)

552: (3) is \cite[Lemma 1.4]{W2}.

553: For (2), if $|F|>\frac{n_j+1}{2}$ then $F$ would contain either an $S$-cycle, a generalized $S$-cycle or

554: two level edges with distinct labels, except the case where $G_j$ is of type $\mathcal{B}$ and $n_j=1$.

555: In the exceptional case, a pair of edges is parallel in both $G_i$ and $G_j$,

556: a contradiction.

557: (4) is similar to (2).

558: \end{proof}

559: 

560: 

561: \begin{lemma}\label{lem:parallel-negative}

562: Let $F$ be a family of mutually parallel negative edges in $G_i$.

563: \begin{itemize}

564: \item[(1)] When $G_j$ is of type $\mathcal{S}$ or $\mathcal{P}$,

565:            $|F|\le n_j-1$.

566: \item[(2)] When $G_j$ is of type $\mathcal{A}$,

567:            $\mathcal{B}$ or $\mathcal{K}$,

568:            $|F|\le n_j$.

569: \end{itemize}

570: \end{lemma}

571: 

572: \begin{proof}

573: (1) See \cite[Lemma 2.3]{G1}.  (This is essentially proved in \cite[Section 5]{GLi}.)

574: (2) The argument of the proof of \cite[Lemma 2.5(3)]{GW} works well.  (Also, see the proof of \cite[Lemma 4.2]{G1}.)

575: \end{proof}

576: 

577: 

578: 

579: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

580: \section{$x$-faces and non-orientable surfaces} \label{sec:Pxface}

581: 

582: In this section we assume that $G_j$ is of type $\mathcal{P}$ or $\mathcal{B}$

583: (so we may assume that $1$ is the only possible $sl$-vertex of $G_j$ by Lemma \ref{lem:Schar}),

584: and thus all level edges of $G_i$ are level $1$-edges.

585: 

586: A disk face of the subgraph of $G_i$ consisting of all the vertices

587: and positive $x$-edges of $G_i$ is called an {\em $x$-face\/}.

588: Remark that the boundary of an $x$-face $D$ may be not a circle, that is,

589: $\partial D$ may contain a double edge, and more than two edges of $\partial D$

590: may be incident to a vertex on $\partial D$

591: (see Figure 2.1 in \cite{HM}).

592: A cycle in $G_i$ is a {\em two-cornered cycle\/}

593: if it is the boundary of a disk face containing only $01$-corners,

594: $12$-corners and positive edges, only when $n_j \geq 3$.

595: Recall that $0$ denotes $n_j$.

596: A two-cornered cycle must contain both kinds of corners,

597: because $G_i$ cannot contain a Scharlemann cycle by Lemma \ref{lem:Schar}.

598: Also, it contains at least one $02$-edge.

599: For convenience, when the labels appear in anticlockwise order

600: around the boundary of a vertex $v$ of $G_i$,

601: given three distinct labels $x_1,x_2,x_3$,

602: we say $x_1 < x_2 < x_3$ if $x_1,x_2,x_3$ appear

603: in anticlockwise order on some interval in $\partial v$ containing $n_j$ edge endpoints.

604: Thus three expressions $x_1<x_2<x_3$, $x_2<x_3<x_1$ and $x_3<x_1<x_2$ are equivalent.

605: 

606: 

607: \begin{proposition} \label{prop:Ppair}

608: Suppose that $n_j \geq 3$.

609: An $x$-face, $x \ne 1$, in $G_i$ contains a pair of two-cornered

610: cycles sharing a level $1$-edge.

611: \end{proposition}

612: 

613: \begin{proof}

614: Let $\Gamma_D$ be the subgraph of $G_i$ in an $x$-face $D$.

615: There is a possibility that $\partial D$ is not a circle as mentioned before.

616: Since we will find a pair of two-cornered cycles within $D$,

617: we can cut formally the graph $G_i \cap D$ along double edges of $\partial D$

618: and at vertices to which more than two edges of $\partial D$ are incident to

619: so that $\partial D$ is deformed into a circle.

620: (See also Figure 5.1 in \cite{HM}.)

621: Thus we may assume that $\partial D$ is a circle.

622: %$\Gamma_D$ has no vertex in the interior of $D$.

623: We may assume that the labels appear in anticlockwise order

624: around the boundary of each vertex.

625: 

626: Suppose that $D$ has a diagonal edge $d$ with distinct labels

627: $\{a,b\}$, which must differ from $x$ because $D$ is an $x$-face,

628: as in Figure \ref{fig:split}(a).

629: Assume without loss of generality that $a<x<b$.

630: Formally construct a new $x$-face $D'$ as follows.

631: Keep all corners and edges of $\Gamma_D$ to the right of $d$

632: (when $d$ is directed from $a$ to $b$),

633: discard all corners and edges to the left of $d$,

634: and then insert additional edges to the left of $d$,

635: and parallel to $d$, until you first reach label $x$

636: at one or both ends of this parallel family of edges,

637: as in Figure \ref{fig:split}(b).

638: In particular, these additional edges contain no edges of two-cornered cycles

639: or Scharlemann cycles of the graph on the new $x$-face $D'$.

640: 

641: \begin{figure}[h]

642: \epsfbox{fig02.eps}

643: \caption{Split along a diagonal edge}\label{fig:split}

644: \end{figure}

645: 

646: Repeat the above process for every diagonal edge which is not a level $1$-edge,

647: then get a new $x$-face $E$ and a graph $\Gamma_E$ in $E$.

648: All diagonal edges of $\Gamma_E$ are level $1$-edges,

649: and all (and only) boundary edges are $x$-edges,

650: where the label $x$ possibly appear on both ends,

651: say level $x$-edges.

652: 

653: \begin{claim} \label{claim:level}

654: $\Gamma_E$ contains a level $1$-edge.

655: \end{claim}

656: 

657: \begin{proof}[Proof of Claim \ref{claim:level}]

658: Assume that $\Gamma_E$ contains no level $1$-edges,

659: and so no diagonal edges.

660: We first show that if for some vertex $v$ of $\Gamma_E$

661: two boundary edges are incident to $v$ with label $x$,

662: then these should be level $x$-edges.

663: For, $n_j + 1$ edges are incident to $v$ in $\Gamma_E$.

664: If $n_j$ is even, more than $\frac{n_j}{2}$ mutually parallel edges

665: are incident to $v$, and so one of these edges should

666: be a level edge different from a level $x$-edge

667: (recall that $\Gamma_E$ cannot contain a Scharlemann cycle),

668: a contradiction.

669: If $n_j$ is odd, two families of $\frac{n_j+1}{2}$ mutually parallel

670: edges are incident to $v$ by Lemma \ref{lem:parallel}(2), and the boundary edge of each family

671: should be a level $x$-edge by the same reason above.

672: 

673: Consider the cycle $\sigma$ consisting of boundary $x$-edges of $\Gamma_E$.

674: Assume that $\sigma$ has an $x$-edge which is not a level $x$-edge.

675: So the only one end has label $x$ at $v_1$, say.

676: By the fact we just proved,

677: another $x$-edge incident to $v_1$ does not have label $x$ at $v_1$.

678: Thus this edge has label $x$ at the other end $v_2$, say.

679: After repeating this process,

680: we are led to show that $\sigma$ is a great $x$-cycle

681: in the terminology of \cite{CGLS}.

682: By the same argument in the proof of Lemma 2.6.2 of \cite{CGLS},

683: $\Gamma_E$ contains a Scharlemann cycle,

684: since $\Gamma_E$ does not contain level edges in its interior,

685: a contradiction.

686: 

687: If all edges of $\sigma$ are level $x$-edges, then we have a great $x+1$-cycle just inside $\sigma$.

688: Thus we still find a Scharlemann cycle, a contradiction.

689: (See also \cite[Lemma 5.2]{HM}.)

690: \end{proof}

691: 

692: Let $e$ be a level $1$-edge.

693: So it does not belong to $\partial \Gamma_E$.

694: Let $E_1$ and $E_2$ be the faces of $\Gamma_E$ adjacent to $e$.

695: Assume for contradiction that $\partial E_1$ (or $\partial E_2$)

696: is not a two-cornered cycle.

697: Note that $\partial E_1$ may contain many level $1$-edges.

698: Let $\{a_k,a_k+1\}$, $k=1, \cdots ,n$, be the consecutive label pairs

699: of the corners between successive level $1$-edges on $\partial E_1$, which appear in order around $\partial E_1$,

700: when one runs clockwise around $\partial E_1$

701: starting at one end of $e$, as in Figure \ref{fig:cornered}(a).

702: Then some $a_k$ is neither $0$ nor $1$.

703: Since $a_1 = 1$ and $a_n = 0$,

704: there are indices $l$ and $m$ so that

705: $a_k = 0$ or $1$ when $1 \leq k < l$ or $k=m$,

706: and $a_k \neq 0,1$ when $l \leq k <m$.

707: Consider the edges of the parallelism class containing

708: each $\{a_{k-1} +1,a_k \}$-edge, $l \leq k \leq m$.

709: Note that among these edges a level $x$-edge on the boundary

710: of $\Gamma_E$ is the only possible level edge,

711: and there is no $x$-edge except on the boundary.

712: See Figure \ref{fig:cornered}(b).

713: Then we have $x \leq a_k < a_{k-1}+1 \leq x$, and so $x \leq a_k \leq a_{k-1} < x$.

714: Finally we have

715: $x \leq a_m \leq a_{m-1} \leq \cdots \leq a_l \leq a_{l-1} < x$.

716: This is impossible because $a_{l-1}, a_m = 0$ or $1$.

717: Thus we have shown that any face adjacent to a level $1$-edge is two-cornered.

718: %

719: %Choose an outermost level $1$-edge in $\Gamma_E$.

720: %Then two faces adjecent to it are a desired pair of two-cornered cycles.

721: \end{proof}

722: 

723: In fact, we can see from the argument above that 

724: $a_1=\dots=a_i=1$ and $a_{i+1}=\dots=a_n=0$ for some $i$.

725: That is, when we go along a two-cornered cycle from a level $1$-edge in some direction, 

726: $12$-corners appear successively, $23$-corners appear after them, and

727: we reach a level $1$-edge (possibly, a different one from the start).

728: 

729: 

730: \begin{figure}[h]

731: \epsfbox{fig03.eps}

732: \caption{Finding two-cornered cycles}\label{fig:cornered}

733: \end{figure}

734: 

735: Recall that $1$ is the only possible $sl$-label of $G_i$.

736: The next theorem essentially follows from the argument of \cite[Section 7]{H}.

737: 

738: \begin{theorem} \label{thm:Pxface}

739: If $G_j$ is of type $\mathcal{P}$,

740: then $G_i$ cannot contain an $x$-face for a non-$sl$-label $x$ of $G_i$.

741: \end{theorem}

742: 

743: \begin{proof}

744: Assume that $G_i$ contains such an $x$-face.

745: If $n_j = 2$, then

746: a boundary of each face in the $x$-face is a Scharlemann cycle

747: or contains both a level $1$-edge and a level $2$-edge,

748: contradicting Lemma \ref{lem:Schar}(1) and (2).

749: Thus we may assume that $n_j \geq 3$.

750: Applying Proposition \ref{prop:Ppair},

751: $G_i$ contains a pair of two-cornered cycles sharing a level $1$-edge $e$.

752: 

753: Let $\widehat{S}$ be the sphere which is the boundary of a regular neighborhood

754: of $\widehat{F}_j$ in $M(\gamma_j)$, and let $S=\widehat{S}\cap M$.

755: The arc components of $F_i \cap S$

756: give rise to a pair of labelled graphs ($G_i^S, G_S$)

757: as usual, where $G_S$ is a double cover of $G_j$.

758: Then $G_i^S$ contains an $S$-cycle $\sigma$ corresponding to the level $1$-edge $e$.

759: We may assume that $\sigma$ has label pair $\{1,2\}$.

760: Also, two cycles adjacent to $\sigma$ in $G_i^S$ contain

761: only two types of corners, $01$-corners and $23$-corners.

762: Let $E_1$ and $E_2$ be the faces of $G_i^S$ bounded by

763: these cycles respectively.

764: The edges of $\sigma$ separate $\widehat{S}$ into two disks,

765: and two vertices $0$ and $3$ of $G_S$ are contained

766: in the same disk component, say $D_0$,

767: because of the existence of $03$-edges.

768: Let $D_1$ be an expansion of $D_0$

769: containing the fat vertices $1$ and $2$ in $\widehat{S}$,

770: and regard it as properly embedded in

771: $X = \mbox{nhd}(D_1 \cup V_{01} \cup V_{23} \cup E_1 \cup E_2)$

772: where $V_{01}$ and $V_{23}$ are defined

773: as in the proof of Lemma \ref{lem:Scharsep}.

774: Since $\partial E_1$ and $\partial E_2$ are non-separating on the boundary of

775: the genus two handlebody $\mbox{nhd}(D_1 \cup V_{01} \cup V_{23})$,

776: $\partial X$ is either a $2$-sphere or

777: the disjoint union of a $2$-sphere and a torus.

778: The latter happens only when $\partial E_1$ and $\partial E_2$ are parallel on

779: the boundary of $\mbox{nhd}(D_1 \cup V_{01} \cup V_{23})$.

780: In this case, $\partial D_1$ lies in the parallelism annulus.

781: Thus $\partial D_1$ cuts the $2$-sphere component of $\partial X$

782: into two disk parts, one of which is parallel to $D_1$, in any case.

783: Let $D$ be the other disk part and

784: then extend $\partial D$ down to $\widehat {F}_j$

785: along an annulus $A = \partial D \times I$,

786: using the $I$-bundle structure of $\mbox{nhd}(\widehat {F}_j)$.

787: Note that the boundary component of $A$

788: which lies on $\widehat {F}_j$ bounds a M\"{o}bius band $B$

789: intersecting the core of $V_{\gamma_j}$ in a single point.

790: In particular,

791: $D \cup A \cup B$ is a projective plane in $M(\gamma_j)$.

792: Since $|D\cap V_{\gamma_j}|\leq |D_0\cap V_{\gamma_j}|-2=n_j -3$,

793: $|A \cap V_{\gamma_j}| = 0$ and $|B \cap V_{\gamma_j}| = 1$,

794: this projective plane intersects the core of $V_{\gamma_j}$

795: in less than $n_j$ points.

796: This contradicts the minimality of $n_j$.

797: \end{proof}

798: 

799: \begin{theorem} \label{thm:Bxface}

800: If $G_j$ is of type $\mathcal{B}$ with $n_j \geq 2$,

801: then $G_i$ cannot contain an $x$-face for a non-$sl$-label $x$ of $G_i$.

802: \end{theorem}

803: 

804: \begin{proof}

805: The same argument in the proof of Theorem \ref{thm:Pxface}

806: applies here.

807: Assume that $G_i$ contains an $x$-face for a non-$sl$-label $x$.

808: Applying Proposition \ref{prop:Ppair},

809: $G_i$ contains a pair of two-cornered cycles sharing a level $1$-edge $e$.

810: In particular, we can assume that one of the pair contains a single level $1$-edge.

811: This is guaranteed by choosing the level $1$-edge $e$ to be outermost in the $x$-face.

812: In the same notation as before, 

813: however, the edges of $\sigma$ separate $\widehat{S}$, which is an annulus,

814: into two annuli, and so $D_1$ is an annulus.

815: Also, we may assume that $\partial E_1$ contains only one $12$-edge.

816: 

817: 

818: 

819: Let $W=\mbox{nhd}(D_1\cup V_{01}\cup V_{23})$ and let

820: $X=W\cup \mbox{nhd}(E_1 \cup E_2)$ as before.

821: We may assume that $D_1$ is properly embedded in $W$ and $X$.

822: Thus $W$ consists of a solid torus $\mbox{nhd}(D_1)$ and the two $1$-handles attached there.

823: Now $\pi_1(W)=\langle x,y,t \rangle$, where taking as base \lq\lq point\rq\rq\ a

824: thick ball negihborhood of a meridian disk of $\mbox{nhd}(D_1)$, which contains

825: the $12$-edge of $\partial E_1$ and the four attaching disks

826: of two $1$-handles $V_{01}$ and $V_{23}$,

827: $x$ is represented by a core of $V_{23}$ going from vertex $2$ to vertex $3$,

828: $y$ is represented by a core of $V_{01}$ going from vertex $0$ to vertex $1$, and

829: $t$ is represented by the $12$-edge of $\sigma$ on $\partial E_2$ oriented from vertex $1$ to vertex $2$.

830: In $\pi_1(W)$, $[\partial E_2]$ (with a clockwise orientation) contains the sequence $ytx$,

831: but $[\partial E_1]$ has no such sequence by the fact that

832: $\partial E_1$ has only one level $12$-edge.

833: Hence $\partial E_1$ and $\partial E_2$ are not parallel on $\partial W$.

834: Clearly, both are non-separating on $\partial W$.

835: Thus the boundary of $W\cup \mbox{nhd}(E_1)$ is a genus two surface, but

836: $\partial X$ is either a torus or the disjoint union of two tori, according to

837: whether $\partial E_2$ is non-separating on the genus two surface or not.

838: In the former, one component of $\partial D_1$ is essential on the torus.

839: If not, $\widehat{F}_j$ is compressible, and hence

840: $M(\gamma_j)$ contains a projective plane, which contradicts the irreducibility of $M(\gamma_j)$.

841: Therefore the frontier of $X$ is an annulus.

842: In the latter, the frontier of $X$ is the disjoint union of an annulus and a torus.

843: Thus in either case, we have an annulus component among the frontier of $X$.

844: In particular, $\partial D_1$ divides it into

845: two annuli, one of which is parallel to $D_1$.

846: The rest of the proof is exactly the same as previous, and

847: we would have a new M\"{o}bius band having fewer intersection with $V_{\gamma_j}$.

848: \end{proof}

849: 

850: %Remark that a generalized $S$-cycle is one of the simplest form of $x$-faces.

851: %Thus two theorems above guarantee that

852: %if $G_j$ is of type $\mathcal{P}$ or $\mathcal{B}$,

853: %then $G_i$ cannot contain a generalized $S$-cycle.

854: 

855: 

856: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

857: \section{$x$-faces and orientable surfaces} \label{sec:Sxface}

858: 

859: In this section we assume that $G_j$ is of type $\mathcal{S}$ or $\mathcal{A}$.

860: By Lemma \ref{lem:Schar}(2) we may assume that

861: $G_i$ contains only $12$-Scharlemann cycles, if they exist.

862: 

863: An $x$-face in $G_i$ is defined as previous.

864: A cycle in $G_i$ is a {\em two-cornered cycle\/},

865: slightly different from previous,

866: if it is the boundary of a disk face containing only $01$-corners,

867: $23$-corners and positive edges, and additionally

868: it contains at least one edge of a $12$-Scharlemann cycle.

869: A two-cornered cycle must contain both types of corners and a $03$-edge.

870: A {\em cluster\/} $C$ is a connected subgraph of $G_i$ satisfying that

871: \begin{itemize}

872: \item[(i)] $C$ consists of $12$-Scharlemann cycles and two-cornered cycles,

873: \item[(ii)] every $12$-edge of $C$ belongs to both a Scharlemann cycle

874: and a two-cornered cycle, and

875: \item[(iii)] $C$ contains no cut vertex.

876: \end{itemize}

877: 

878: See Figure \ref{fig:cluster}.

879: 

880: \begin{figure}[h]

881: \epsfbox{fig04.eps}

882: \caption{A cluster and a seemly pair}\label{fig:cluster}

883: \end{figure}

884: 

885: 

886: 

887: \begin{proposition} \label{prop:cluster}

888: Suppose that $n_j \geq 3$.

889: An $x$-face, $x \neq 1,2$, in $G_i$ contains a cluster $C$.

890: \end{proposition}

891: 

892: \begin{proof}

893: Let $\Gamma_D$ be the subgraph of $G_i$ in an $x$-face $D$.

894: As in the previous,

895: we may assume that $\partial D$ is a circle.

896: %and $\Gamma_D$ has no vertex in the interior of $D$.

897: Also assume that the labels appear anticlockwisely.

898: 

899: For all diagonal edges of $D$ which are not of $12$-Scharlemann cycles

900: (also these are neither $x$-edges nor level edges),

901: apply the same argument in the proof of Proposition \ref{prop:Ppair}.

902: Then we get a new $x$-face $E$ and a graph $\Gamma_E$ in $E$

903: so that all diagonal edges are of $12$-Scharlemann cycles

904: and all (and only) boundary edges are $x$-edges.

905: Furthermore the additional edges contain no edges of

906: Scharlemann cycles or two-cornered cycles of $\Gamma_E$.

907: Remark that a level $x$-edge can appear

908: on the boundary of the graph.

909: 

910: 

911: \begin{claim} \label{claim:Schar}

912: $\Gamma_E$ contains a $12$-Scharlemann cycle,

913: so does $\Gamma_D$.

914: \end{claim}

915: 

916: \begin{proof}[Proof of Claim \ref{claim:Schar}]

917: It is clear from the proof of Claim \ref{claim:level}.

918: \end{proof}

919: 

920: This means that $\widehat{F}_j$ must be separating

921: and $n_j$ is even by Lemma \ref{lem:Scharsep}.

922: The parity rule guarantees that each edge of $\Gamma_E$

923: connects vertices with one label even and the other label odd,

924: and so there are no level $x$-edges.

925: 

926: Any $12$-edge of a Scharlemann cycle

927: does not belong to $\partial \Gamma_E$.

928: Consider the face $E_1$ of $\Gamma_E$ which is adjacent to the $12$-edge

929: and whose boundary is not the Scharlemann cycle.

930: It is possible that $E_1$ contains more than one $12$-edges

931: of Scharlemann cycles.

932: Again, let $\{a_k,a_k+1\}$, $k=1, \cdots ,n$,

933: be the consecutive label pairs of the corners

934: between two consecutive $12$-edges of Scharlemann cycles

935: when one runs clockwise around $\partial E_1$.

936: Note that $a_1 = 2$ and $a_n = 0$.

937: 

938: Assume for contradiction that

939: $\partial E_1$ is not a two-cornered cycle.

940: Since some $a_k$ then is neither $0$ nor $2$,

941: there are indices $l$ and $m$ so that

942: $a_k = 0$ or $2$ when $1 \leq k < l$ or $k=m$,

943: and $a_k \neq 0,2$ when $l \leq k <m$.

944: 

945: Consider the edges of the parallelism class containing

946: each $\{a_{k-1} +1,a_k \}$-edge for $l \leq k \leq m$.

947: Since there are neither Scharlemann cycles

948: nor level edges among these edges,

949: one finds that $x \leq a_k < a_{k-1}+1 \leq x$, and hence

950: $x \leq a_k \leq a_{k-1} < x$.

951: And so

952: $x \leq a_m \leq a_{m-1} \leq \cdots \leq a_l \leq a_{l-1} < x$.

953: This is impossible because $a_{l-1}, a_m = 0$ or $2$

954: and all $a_k$'s are even by the parity rule.

955: Hence $\partial E_1$ is a two-cornered cycle.

956: 

957: Thus we have shown that 

958: any face next to a Scharlemann cycle in $\Gamma_E$ is two-cornered.

959: Let $C$ be the union of all the Scharlemann cycles and

960: all the two-cornered cycles adjacent to each 12-edges

961: of the Scharlemann cycles.

962: If necessary, choose a block of $C$. Then it is a desired cluster

963: in $\Gamma_E$ and so in $\Gamma_D$.

964: \end{proof}

965: 

966: 

967: Furthermore, we can see from the argument that

968: $a_1=\dots=a_i=2$ and $a_{i+1}=\dots=a_n=0$ for some $i$.

969: 

970: 

971: Let $R$ be the twice-punctured sphere obtained from

972: $\widehat{F}_j$ by deleting two fat vertices $1$ and $2$

973: (if $G_j$ is of type $\mathcal{A}$, then use $\widehat{F}_j$ after capping

974: off two boundary circles by disks).

975: The family of all $12$-edges of a Scharlemann cycle in the cluster $C$

976: separates $R$ into disks,

977: and one of those disks contains both vertices $0$ and $3$ of $G_j$,

978: because of the existence of $03$-edges in $C$.

979: The two $12$-edges bounding such a disk are

980: called {\em good edges\/} of $C$.

981: Thus each Scharlemann cycle in $C$ has exactly two good edges.

982: 

983: Let $\Lambda$ be the maximal dual graph of $C$

984: whose vertices are dual to Scharlemann cycles

985: and two-cornered cycles containing good edges,

986: and edges are dual to good edges of $C$

987: as depicted in Figure \ref{fig:cluster}.

988: Thus in $\Lambda$, a vertex dual to a Scharlemann cycle has valency $2$,

989: and a vertex dual to a two-cornered cycle has valency

990: the number of good edges of the two-cornered cycle.

991: Furthermore $\Lambda$ is a forest according to the construction of $C$.

992: That is, each component of $\Lambda$ is a tree.

993: Let $\sigma$ be a two-cornered cycle with a good edge in $C$.

994: Then each $12$-edge, which is not a good one, in $\sigma$

995: contributes to the number of components of $\Lambda$ by adding $1$.

996: Consequently there is a component $\Lambda_g$ of $\Lambda$ so that

997: all $12$-edges of its dual two-cornered cycles are good.

998: 

999: 

1000: Hereafter, we consider the subgraph $C_g$ of $C$, dual to $\Lambda_g$.

1001: Say, $C_g$ contains $n$ Scharlemann cycles,

1002: and so $2n$ good edges and $n+1$ two-cornered cycles.

1003: A two-cornered cycle dual to an end vertex of the tree $\Lambda_g$

1004: has only one good edge.

1005: Choose one $e_1$ of the nearest edges to vertex $0$ (or $3$) among them,

1006: that is, there are no such good edges

1007: between $e_1$ and vertex $0$ in $R$.

1008: Let $\sigma_g$ and $\sigma_1$ be the Scharlemann cycle and

1009: two-cornered cycle adjacent to $e_1$ respectively.

1010: Note that $\sigma_1$ has only one $12$-edge.

1011: Then $\sigma_g$ has another good edge $e_2$,

1012: and $e_1$ and $e_2$ bound a disk $D_g$

1013: containing $0$ and $3$ in $R$.

1014: Note that the boundaries of the faces bounded by

1015: Scharlemann cycles are parallel on a torus obtained

1016: from $R$ by attaching an annulus $\partial V_{12}$.

1017: Thus exactly $n-1$ out of $2n$ good edges are not contained in $D_g$.

1018: Therefore we have another two-cornered cycle $\sigma_2$

1019: all of whose $12$-edges lie in $D_g$.

1020: Consequently all edges (consisting of $01$-edges, $12$-edges,

1021: $23$-edges and $03$-edges) of $\sigma_1$ and $\sigma_2$ lie in $D_g$.

1022: Furthermore if $\sigma_2$ has only one good $12$-edge,

1023: then the two good edges of $\sigma_1$ and $\sigma_2$ lie on different

1024: sides of the vertices $0$ and $3$ in $D_g$ by the choice of $\sigma_1$.

1025: Such $\sigma_1, \sigma_2$ are called a {\em seemly pair\/}.

1026: Then we can say:

1027: 

1028: \begin{proposition} \label{prop:seemlypair}

1029: There is a seemly pair of two-cornered cycles in $C$.

1030: \end{proposition}

1031: 

1032: From now we apply the argument in \cite[Section 6]{H} to get

1033: the following two theorems.

1034: Recall that $1,2$ are the only possible $sl$-labels of $G_i$

1035: 

1036: \begin{theorem} \label{thm:Sxface}

1037: If $G_j$ is of type $\mathcal{S}$,

1038: then $G_i$ cannot contain an $x$-face for a non-$sl$-label $x$ of $G_i$.

1039: \end{theorem}

1040: 

1041: \begin{proof}

1042: Suppose that $G_i$ contains such an $x$-face.

1043: Note that $n_j \geq 3$ by Lemma \ref{lem:n}.

1044: We continue the preceding argument.

1045: 

1046: Let $E$ be the disk bounded by $\sigma_g$.

1047: Then $L=\mathrm{nhd}((\widehat{F}_j-\mathrm{Int}\,D_g)\cup V_{12}\cup E)$ is a punctured lens space in $M(\gamma_j)$.

1048: Since $\partial L$ is a reducing sphere in $M(\gamma_j)$,

1049: $|\partial L\cap V_{\gamma_j}|=2|(\widehat{F}_j-\mathrm{Int}\,D_g)\cap V_{\gamma_j}|-2\ge n_j$

1050: by the minimality of $n_j$.

1051: Thus $|\mathrm{Int}\,D_g\cap V_{\gamma_j}|\le \frac{n_j}{2}-1$.

1052: Let $X_D' = \mbox{nhd} (D_g \cup V_{01} \cup V_{23})$ and

1053: $X_F' = \mbox{nhd} (\widehat{F}_j \cup V_{01}\cup V_{23})$.

1054: Then $X_D'$ is a genus two handlebody and

1055: $X_F'$ is a once-punctured genus two handlebody.

1056: The genus two torus component of $\partial X_F'$

1057: is referred to as the outer boundary of $X_F'$.

1058: Let $E_i$ be the face bounded

1059: by the two-cornered cycle $\sigma_i$ for $i=1,2$.

1060: The point is that all edges of $\sigma_i$ are contained in $D_g$.

1061: Let $X_D = X_D' \cup \mbox{nhd}(E_1\cup E_2)$

1062: and $X_F = X_F' \cup \mbox{nhd}(E_1\cup E_2)$

1063: as in Figure \ref{fig:XdXf}.

1064: Since $\partial E_1$ and $\partial E_2$ are non-separating on $\partial X_D'$ and $\partial X_F'$,

1065: both $\partial X_D$ and the outer components of $\partial X_F$ are

1066: either a $2$-sphere or the disjoint union

1067: of a $2$-sphere and a torus, simultaneously.

1068: Note that the latter case occurs

1069: only when $\partial E_1$ and $\partial E_2$ are parallel on the boundary of $X_D'$ or $X_F'$.

1070: 

1071: 

1072: \begin{figure}[h]

1073: \epsfbox{fig05.eps}

1074: \caption{$X_D$ and $X_F$}\label{fig:XdXf}

1075: \end{figure}

1076: 

1077: First, assume that $\partial X_D$ is a $2$-sphere $S_D$,

1078: and the outer component of $\partial X_F$ is a $2$-sphere $S_F$.

1079: If $X_D$ is not a $3$-ball, then $S_D$ is a reducing sphere.

1080: But $|S_D \cap V_{\gamma_j}|=2|\mathrm{Int}\,D_g \cap V_{\gamma_j}| \leq n_j-2$,

1081: contradicting the minimality of $n_j$.

1082: Thus it should be a $3$-ball,

1083: and so $X_F$ is homeomorphic to $S^2\times I$.

1084: Thus $S_F$ is isotopic to $\widehat{F}_j$,

1085: contradicting the minimality of $n_j$ again.

1086: 

1087: Next, assume that $\partial X_D$ is

1088: the disjoint union of a $2$-sphere and a torus, $S_D \cup T_D$ and

1089: the outer components of $\partial X_F$ are also

1090: the disjoint union of a $2$-sphere and a torus, $S_F \cup T_F$.

1091: Recall that this case occurs whenever $\partial E_1$ and $\partial E_2$

1092: cobound an annulus in $\partial X_D'$ and $\partial X_F'$.

1093: This is possible only when $\sigma_2$ corresponds to

1094: an end vertex of $\Lambda_g$ with the same number of

1095: $01$-corners and $23$-corners as those of $\sigma_1$.

1096: Let $D'$ be the intersection of $\partial X_D$ and

1097: the inner sphere component of $\partial X_F$.

1098: By the choice of the seemly pair,

1099: this annulus in $\partial X_D'$ contains $D'$, so does $S_D$.

1100: Similarly $S_F$ contains a pushoff of $R- D_g$.

1101: 

1102: If $S_D$ is non-separating in $M(\gamma_j)$,

1103: then it is a reducing sphere with

1104: $|S_D \cap V_{\gamma_j}| \leq n_j-2$, contradicting.

1105: Thus $S_D$ is separating in $M(\gamma_j)$.

1106: Let $X_D''$ be the manifold bounded by $S_D$

1107: containing $T_D$ in $M(\gamma_j)$.

1108: If $X_D''$ is not a $3$-ball,

1109: then $S_D$ is a reducing sphere with less intersection with $V_{\gamma_j}$.

1110: Thus it should be a $3$-ball,

1111: and so $S_F$ is isotopic to $\widehat{F}_j$, contradicting again.

1112: This completes the proof.

1113: \end{proof}

1114: 

1115: 

1116: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1117: \begin{theorem} \label{thm:Axface}

1118: If $G_j$ is of type $\mathcal{A}$ with $n_j \geq 3$,

1119: then $G_i$ cannot contain an $x$-face for a non-$sl$-label $x$ of $G_i$.

1120: \end{theorem}

1121: 

1122: \begin{proof}

1123: Suppose that $G_i$ contains such an $x$-face.

1124: Again we continue the argument stated before Proposition \ref{prop:seemlypair}.

1125: The proof is divided into three cases.

1126: Recall that $R$ is a twice-punctured sphere obtained from $\widehat{F}_j$ by deleting

1127: two vertices $1$ and $2$ and then capping off $\partial \widehat{F}_j$ by two disks,

1128: and that $D_g$ lies on $R$.

1129: 

1130: %%%%%%%%%%%%%

1131: First, suppose that $D_g$ contains $\partial \widehat{F}_j$.

1132: Then the edges of the Scharlemann cycle $\sigma_g$ lie

1133: in a disk in $\widehat{F}_j$, contradicting Lemma \ref{lem:Scharsep}.

1134: 

1135: %%%%%%%%%%%%%%

1136: Second, suppose that $D_g$ does not contain any component

1137: of $\partial \widehat{F}_j$.

1138: Then $D_g$ is contained in $\widehat{F}_j$.

1139: The proof is similar to that of the preceding theorem.

1140: Recall that $E_i$ is the face bounded

1141: by the two-cornered cycle $\sigma_i$ for $i=1,2$.

1142: Let $X_D = \mbox{nhd} (D_g \cup V_{01} \cup V_{23} \cup E_1 \cup E_2)$

1143: and $X_F = \mbox{nhd} (\widehat{F}_j \cup V_{01} \cup V_{23}\cup E_1 \cup E_2)$.

1144: Then $\partial X_D$ is either a $2$-sphere or the disjoint union

1145: of a $2$-sphere and a torus.

1146: Also, the frontier of $X_F$ has a copy of $\widehat{F}_j$, and

1147: the remaining part is referred to as the outer components.

1148: It can be seen that

1149: the outer components consist of either a single annulus or

1150: the disjoint union of an annulus and a torus by the choice of the seemly pair.

1151: ($\partial X_D$ is disconnected if and only if the outer components of the frontier of $X_F$ are disconnected.)

1152: In either case, we have an annular component $A_F$.

1153: Then $A_F$ is obtained from $\widehat{F}_j$

1154: by replacing $D_g$ with a proper disk

1155: which is a part of the $2$-sphere component of $\partial X_D$.

1156: Since $M(\gamma_j)$ is irreducible by the assumption in Section \ref{sec:intro},

1157: the $2$-sphere component of $\partial X_D$ bounds a $3$-ball.

1158: Hence $A_F$ is isotopic to $\widehat{F}_j$.

1159: This contradicts the minimality of $n_j$.

1160: 

1161: %%%%%%%%%%%%%%%%%%%%

1162: Finally suppose that $D_g$ contains exactly one component

1163: of $\partial \widehat{F}_j$.

1164: If there is a disk in $D_g$ which contains all edges of $\sigma_1$ and $\sigma_2$ and which does not contain

1165: the component of $\partial\widehat{F}_j$, then the case reduces to the previous one.

1166: Otherwise, let $A_1$ be the annulus $D_g \cap \widehat{F}_j$.

1167: Consider $Y=\mathrm{nhd}((\widehat{F}_j-\mathrm{Int}\,A_1)\cup V_{12}\cup E)$,

1168: where $E$ is the face bounded by $\sigma_g$.

1169: Then the frontier $Q$ of $Y$ in $M(\gamma_j)$ is an essential annulus by Claim on page 430 of \cite{W3}.

1170: Thus $|Q\cap V_{\gamma_j}|=2|(\widehat{F}_j-\mathrm{Int}\,A_1)\cap V_{\gamma_j}|-2\ge n_j$

1171: by the minimality of $n_j$.

1172: Hence $|\mbox{Int}\,A_1 \cap V_{\gamma_j}| \leq \frac{n_j}{2} - 1$.

1173: Let $X_A = \mbox{nhd} (A_1 \cup V_{01} \cup V_{23} \cup E_1 \cup E_2)$

1174: and $X_F = \mbox{nhd} (\widehat{F}_j \cup V_{01} \cup V_{23}

1175: \cup E_1 \cup E_2)$ again.

1176: Then the frontier of $X_A$ contains an annulus component $A_A$ by

1177: the same argument as in the proof of Theorem \ref{thm:Bxface}.

1178: (By the choice of the seemly pair, $\partial E_1$ and $\partial E_2$ are not parallel on the boundary of

1179: $\mbox{nhd} (A_1 \cup V_{01} \cup V_{23})$ and $\mbox{nhd} (\widehat{F}_j \cup V_{01} \cup V_{23})$.)

1180: Also the outer components of the frontier of $X_F$ contain an annulus $A_F$.

1181: If $A_A$ is essential in $M(\gamma_j)$,

1182: then $|A_A \cap V_{\gamma_j}| \le 2|\mbox{Int}\,A_1 \cap V_{\gamma_j}| \leq n_j-2$,

1183: contradicting the minimality of $n_j$.

1184: Remark that $A_A$ is incompressible, because

1185: the central curve of $A_A$ is isotopic to the central curve of

1186: $\widehat{F}_j$.

1187: Thus $A_A$ should be boundary parallel.

1188: That is, there is a solid torus $U$ such that $A_A$ is a longitudinal annulus on $\partial U$ and

1189: $\partial U-A_A\subset \partial M(\gamma_j)$.

1190: Then either $\widehat{F}_j$ is boundary parallel or

1191: one can isotope $\widehat{F}_j$ through $U$ to $A_F$, which intersects $V_{\gamma_j}$ fewer times than $\widehat{F}_j$.

1192: \end{proof}

1193: 

1194: 

1195: Remark that an extended Scharlemann cycle is one of the simplest form of $x$-faces.

1196: Thus two theorems above guarantee that

1197: if $G_j$ is of type $\mathcal{S}$ or $\mathcal{A}$,

1198: then $G_i$ cannot contain an extended Scharlemann cycle.

1199: We may emphasize that only an extended $S$-cycle was used

1200: in the literatures such as \cite{GW,W1} etc.

1201: 

1202: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1203: \section{Extremal block} \label{sec:extremal}

1204: 

1205: The {\em reduced graph\/} $\overline{G}_i$ of $G_i$

1206: is defined to be the graph obtained from $G_i$ by

1207: amalgamating each family of parallel edges into a single edge.

1208: Let $G_i^+$ denote the subgraph of $G_i$ consisting of all

1209: vertices and positive edges of $G_i$.

1210: When $G_i$ is of type $\mathcal{S}$, it is obvious that each component of $G_i^+$ has a disk support, that is,

1211: there is a disk in $\widehat{F}_i$ which contains the component in its interior.

1212: Even if $G_i$ is of type $\mathcal{P}$, the same is true, because any orientation-preserving loop in

1213: a projective plane is contractible.

1214: 

1215: 

1216: %The following proposition shows

1217: %how the existence of $x$-faces on $G_2$ will be applied.

1218: 

1219: \begin{proposition} \label{prop:S}

1220: If $G_1$ is of type $\mathcal{S}$ or $\mathcal{P}$

1221: and $G_2$ is any of the six types,

1222: then each non-$sl$-vertex of $G_1$ has at least

1223: $(\Delta(\gamma_1,\gamma_2)-1)n_2+\chi(\widehat{F}_2)$ positive edge endpoints.

1224: \end{proposition}

1225: 

1226: \begin{proof}

1227: Assume that there is a non-$sl$-vertex $x$ of $G_1$

1228: which has more than $n_2-\chi(\widehat{F}_2)$ negative edges.

1229: Then $G_2$ contains more than $n_2-\chi(\widehat{F}_2)$ positive $x$-edges

1230: by the parity rule.

1231: Thus the subgraph $\Gamma_x$ of $G_2$ consisting of all vertices

1232: and positive $x$-edges of $G_2$ has $n_2$ vertices and

1233: more than $n_2-\chi(\widehat{F}_2)$ edges.

1234: Then an Euler characteristic calculation

1235: shows that $\Gamma_x$ contains a disk face, which is an $x$-face.

1236: This contradicts Theorem \ref{thm:Pxface} or \ref{thm:Sxface}.

1237: \end{proof}

1238: 

1239: 

1240: Suppose that $G_1$ is of type $\mathcal{S}$ or $\mathcal{P}$, and that $\Delta(\gamma_1,\gamma_2)\ge 2$.

1241: By Lemmas \ref{lem:n} and \ref{lem:Schar}, $G_1$ has a non-$sl$-vertex.

1242: Note that each of these vertices has therefore

1243: valency at least two in $\overline{G}_1$ by Lemmas \ref{lem:parallel} and \ref{lem:parallel-negative}.

1244: Now take an innermost component $\Lambda_0$ of $G_1^+$

1245: with a disk support $D_0$, which means that $D_0 \cap G_1^+ = \Lambda_0$.

1246: The subgraph $\Lambda_0$ has therefore at most one $sl$-vertex.

1247: 

1248: Suppose that $\Lambda_0$ is a single vertex.

1249: Then only negative edges are incident there.

1250: By Proposition \ref{prop:S}, it is an $sl$-vertex.

1251: If $G_1$ is of type $\mathcal{S}$,

1252: the edges of a Scharlemann cycle of $G_2$ are incident there.

1253: Then one of the disks bounded by these edges contains non-$sl$-vertices, since $n_1\ge 3$ by Lemma \ref{lem:n}.

1254: Hence we can choose another innermost component of $G_1^+$ with more than one vertex and

1255: at most one $sl$-vertex.

1256: If $G_1$ is of type $\mathcal{P}$, a negative loop is incident there.

1257: Thus we can also choose another innermost component of $G_1^+$

1258: which contains only non-$sl$-vertices.

1259: 

1260: 

1261: 

1262: We may therefore assume that $\Lambda_0$ has more than one vertex.

1263: Then $\Lambda_0$ has no cut vertex or

1264: at least two blocks with at most one cut vertex.

1265: Thus we can choose an innermost component $\Lambda$

1266: with a disk support $D$ after splitting $\Lambda_0$ at all cut vertices,

1267: such that $\Lambda$ has more than one vertex with at most one cut vertex.

1268: In such a case that $\Lambda$ contains a cut vertex and a distinct $sl$-vertex,

1269: we can choose another innermost one containing

1270: no $sl$-vertex.

1271: 

1272: 

1273: 

1274: Such a subgraph $\Lambda$ of $G_1^+$ is called an {\em extremal block\/}

1275: with a disk support $D$.

1276: A vertex of $\Lambda$ is called a {\em ghost vertex\/} if it is either a cut vertex or an $sl$-vertex.

1277: We emphasize that $\Lambda$ has more than one vertex and at most one ghost vertex $y_0$.

1278: A vertex of $\Lambda$ is called a {\em boundary vertex\/} if there is an arc connecting it to $\partial D$ whose interior is

1279: disjoint from $\Lambda$, and an {\em interior vertex\/} otherwise.

1280: Then the preceding argument with Proposition \ref{prop:S}

1281: proves the following theorem which plays key role in this paper:

1282: 

1283: \begin{theorem} \label{thm:extremalS}

1284: Suppose that $G_1$ is of type $\mathcal{S}$ or $\mathcal{P}$.

1285: 

1286: \begin{itemize}

1287: \item[(1)] If $G_2$ is of type $\mathcal{S}$, $\mathcal{P}$,

1288: $\mathcal{A}$ or $\mathcal{B}$

1289: with the assumption that $\Delta(\gamma_1,\gamma_2) \geq 2$,

1290: then $G_1$ contains an extremal block $\Lambda$ with a disk support $D$

1291: so that each boundary vertex, except $y_0$, has at least

1292: $n_2+\chi(\widehat{F}_2)$ consecutive edge endpoints of $\Lambda$.

1293: \item[(2)] If $G_2$ is of type $\mathcal{T}$ or $\mathcal{K}$

1294: with the assumption that $\Delta(\gamma_1,\gamma_2) \geq 3$ and $n_2\ge 3$ when $G_2$ is of type $\mathcal{T}$,

1295: and $n_2\ge 2$ when $G_2$ is of type $\mathcal{K}$,

1296: then $G_1$ contains an extremal block $\Lambda$ with a disk support $D$

1297: so that each boundary vertex, except $y_0$,

1298: has at least $2 n_2$ consecutive edge endpoints of $\Lambda$.

1299: \end{itemize}

1300: \end{theorem}

1301: 

1302: 

1303: 

1304: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1305: \section{$(\mathcal{S},\mathcal{S})$ case} \label{sec:SS}

1306: 

1307: \begin{proof}[Proof of Theorem \ref{thm:SS}]

1308: Assume for contradiction that $\Delta(\gamma_1,\gamma_2) \geq 2$.

1309: Theorem \ref{thm:extremalS}(1) says that

1310: $G_1^+$ contains such an extremal block $\Lambda$ with a disk support

1311: that each boundary vertex, except $y_0$,

1312: has more than $n_2$ consecutive edge endpoints,

1313: so different all $n_2$ labels.

1314: We can choose a non-$sl$-label $x$ (at $y_0$ if it exists),

1315: unless $G_2$ is of type $\mathcal{S}$ and

1316: there is $y_0$ with only two edge endpoints which have the $sl$-labels.

1317: Since $\Lambda$ has at least the same number of $x$-edges

1318: as that of vertices, it contains an $x$-face, contradicting Theorem

1319: \ref{thm:Pxface} or \ref{thm:Sxface}.

1320: 

1321: For the exceptional case, consider $\Lambda-y_0$.

1322: Since all labels still appear on each vertex of $\Lambda-y_0$,

1323: the same argument above leads to a contradiction.

1324: \end{proof}

1325: 

1326: 

1327: 

1328: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1329: \section{$(\mathcal{S},\mathcal{A})$ case} \label{sec:SA}

1330: 

1331: \begin{proof}[Proof of Theorem \ref{thm:SA}]

1332: Assume that $\Delta=\Delta(\gamma_1,\gamma_2) \geq 2$.

1333: Recall that $M(\gamma_2)$ is irreducible and

1334: boundary irreducible.

1335: 

1336: First, assume that $n_2 \geq 3$ when $G_2$ is of type $\mathcal{A}$

1337: (or $n_2 \geq 2$ when of type $\mathcal{B}$).

1338: Theorem \ref{thm:extremalS}(1) says that

1339: $G_1^+$ contains an extremal block $\Lambda$ with a disk support

1340: such that each boundary vertex, except $y_0$, has all different $n_2$ labels.

1341: When $G_2$ is of type $\mathcal{B}$, 

1342: choose a non-$sl$-label $x$ at $y_0$, if it exists.

1343: Otherwise, $x$ is any non-$sl$-label.

1344: Then $\Lambda$ has at least the same number of $x$-edges

1345: as that of vertices, and therefore $\Lambda$ has an $x$-face,

1346: contradicting Theorem \ref{thm:Bxface}.

1347: So $G_2$ is of type $\mathcal{A}$.

1348: 

1349: 

1350: \begin{claim}\label{claim:12Schar}

1351: $G_1$ contains a Scharlemann cycle.

1352: \end{claim}

1353: 

1354: \begin{proof}[Proof of Claim \ref{claim:12Schar}]

1355: Assume not.  Choose any label $x$ (at $y_0$ if it exists).

1356: Then there is an $x$-edge at any vertex of $\Lambda$.

1357: Hence $\Lambda$ contains a great $x$-cycle, and so a Scharlemann cycle \cite[Lemma 2.6.2]{CGLS}.

1358: \end{proof}

1359: 

1360: Thus we may assume that $G_1$ contains only $12$-Scharlemann cycles.

1361: Equivalently, $sl$-labels of $G_1$ are $1$ and $2$.

1362: By Lemma \ref{lem:Scharsep}, $\widehat{F}_2$ is separating and $n_2$ is even.

1363: 

1364: \begin{claim}\label{claim:extrem}

1365: \begin{itemize}

1366: \item[(1)] $\Lambda$ has no interior vertices;

1367: \item[(2)] $\Lambda$ has a ghost vertex $y_0$ such that only two edges are incident to $y_0$ in $\Lambda$

1368: with $sl$-labels $1$ and $2$; and

1369: \item[(3)] $n_2=4$, and the two edges incident to $y_0$ are indeed a $14$-edge and a $23$-edge.

1370: \end{itemize}

1371: \end{claim}

1372: 

1373: \begin{proof}[Proof of Claim \ref{claim:extrem}]

1374: If $\Lambda$ has an interior vertex or has no ghost vertex, then let $x$ be any non-$sl$-label.

1375: Then $\Lambda$ has at least the same number of $x$-edges as that of vertices, so contains an $x$-face, contradicting

1376: Theorem \ref{thm:Axface}.

1377: Even if there is a ghost vertex $y_0$, we can choose a non-$sl$-label $x$ at $y_0$, unless (2) holds.

1378: Finally, assume $n_2\ge 6$.

1379: Then we can choose a non-$sl$-label $x$ such that $y_0$ is not incident to an $x$-edge.

1380: Thus $\Lambda-y_0$ contains an $x$-face as above again.

1381: Even if $n_2=4$, there is still such a label $x$, unless the second conclusion of (3) holds.

1382: \end{proof}

1383: 

1384: 

1385: In fact, $\Lambda-y_0$ contains a $12$-Scharlemann cycle $\sigma$, because 

1386: there is a $1$-edge at any vertex of $\Lambda$ by

1387: (3), and hence $\Lambda-y_0$ contains a great $1$-cycle.

1388: By Lemma \ref{lem:Scharsep},

1389: the edges of $\sigma$ cuts $\widehat{F}_2$ into two annuli $A_1,A_2$ and some disks.

1390: %Let $A_i'=\mathrm{cl}\,(\widehat{F}_2-A_i)$ for $i=1,2$.

1391: 

1392: 

1393: \begin{claim}\label{claim:34edge}

1394: $\Lambda$ contains a $34$-edge.

1395: \end{claim}

1396: 

1397: \begin{proof}[Proof of Claim \ref{claim:34edge}]

1398: Assume not.

1399: Then $\Lambda$ contains only $12$, $14$ and $23$-edges by the parity rule.

1400: Let $y_0,y_1,\dots,y_k$ be the vertices of $\Lambda$, numbered consecutively along $\partial \Lambda$, where 

1401: the $14$-edge $e_0$ at $y_0$ is incident to $y_1$.

1402: (Since $y_i$ $(i\ne 0)$ is incident to at least four edges in $\Lambda$, we see $k>1$.)

1403: Let $e_1$ be the $3$-edge at $y_1$, just before $e_0$.

1404: (This means that the two endpoints of $e_0$ and $e_1$ are successive around $y_1$.) 

1405: If $e_1$ goes to $y_k$, then $e_1$ is a $34$-edge.

1406: So assume $e_1$ goes to $y_i$ for $1<i<k$.

1407: Thus $e_1$ has the label $2$ at $y_i$.

1408: Split $\Lambda$ along $e_1$, and let $\Lambda'$ be the part containing $y_1,y_2,\dots,y_i$.

1409: Then $\Lambda'$ has a $3$-face, contradicting Theorem \ref{thm:Axface}.

1410: \end{proof}

1411: 

1412: 

1413: Hence the vertices $3$ and $4$ lie in the same component, and

1414: we can see that they lie in $A_1$ or $A_2$ indeed.

1415: Otherwise, the vertex $3$ or $4$ is incident to more than $n_1$ negative edges.

1416: Thus, $G_1$ contains a $3$-face or $4$-face, a contradiction.

1417: We may assume that both are in $A_2$.

1418: Let $A_1'=\mathrm{cl}(\widehat{F}_2-A_2)$, and 

1419: $Y=\mathrm{nhd}(A_1'\cup V_{12}\cup E)$, where $E$ is the face bounded by $\sigma$.

1420: Then the frontier of $Y$ gives an essential annulus (see \cite[Claim on page 430]{W3}), which

1421: meets $V_{\gamma_2}$ in two disks.

1422: This contradicts the minimality of $n_2$.

1423: 

1424: 

1425: 

1426: %%%%%%%%%%%%%%%%%%%%%%%%

1427: If $n_2 = 1$ when $G_2$ is of type $\mathcal{A}$,

1428: $G_2$ has only positive edges, and these edges are all parallel.

1429: This contradicts Lemma \ref{lem:parallel}(1-2).

1430: 

1431: %%%%%%%%%%%%%%%%%%%%%%%

1432: Finally assume $n_2 = 2$ when $G_2$ is of type $\mathcal{A}$

1433: (or $n_2 = 1$ when of type $\mathcal{B}$).

1434: Assume for contradiction that $\Delta \geq 3$.

1435: Consider the reduced graph $\overline{G}_2$.

1436: An Euler characteristic calculation shows that

1437: each vertex of $\overline{G}_2$ has valency at most 4.

1438: By Lemmas \ref{lem:parallel}(1-2) and  and \ref{lem:parallel-negative}(1) the valency must be 4,

1439: so the graph looks like that in Figure \ref{fig:annulus}

1440: ((a) for type $\mathcal{A}$ and (b) for type $\mathcal{B}$).

1441: 

1442: \begin{claim}\label{claim:annulus2}

1443: In Figure \ref{fig:annulus}(a), the two vertices have opposite signs.

1444: \end{claim}

1445: 

1446: \begin{proof}[Proof of Claim \ref{claim:annulus2}]

1447: Assume not.  Then $G_2$ has only positive edges.

1448: For a non-$sl$-label $x$ of $G_2$,

1449: let $\Gamma$ be the subgraph of $G_2$ consisting of all vertices and all $x$-edges of $G_2$.

1450: Then an Euler characteristic calculation shows that $\Gamma$ has a disk face, which

1451: is an $x$-face.

1452: This contradicts Theorem \ref{thm:Pxface} or \ref{thm:Sxface}.

1453: \end{proof}

1454: 

1455: Thus $a$ and $d$ are the families of positive edges

1456: and $b$ and $c$ are the families of negative edges.

1457: When $G_1$ is of type $\mathcal{P}$,

1458: this is impossible by Lemmas \ref{lem:parallel} and \ref{lem:parallel-negative}.

1459: When $G_1$ is of type $\mathcal{S}$,

1460: $G_2$ contains an $S$-cycle by Lemma \ref{lem:parallel},

1461: so $M(\gamma_1)$ contains a projective plane by Lemma \ref{lem:Schar}(3),

1462: a contradiction to the previous result.

1463: \end{proof}

1464: 

1465: \begin{figure}[ht]

1466: \epsfbox{fig06.eps}

1467: \caption{Annulus and M\"{o}bius band}\label{fig:annulus}

1468: \end{figure}

1469: 

1470: We now give an example realizing the case $\Delta = 2$ in Theorem \ref{thm:SA}.

1471: Theorem 2.6 of \cite{EW} shows that there is a hyperbolic manifold $M$

1472: such that $M(\gamma_1) = (S^1\times D^2)\sharp L(2,1)$ and

1473: $M(\gamma_2)$ is the union of $C(2,1)$ and $Q(2p,-2p)$,

1474: for any integer $p \geq 2$, along a torus, with $\Delta(\gamma_1,\gamma_2)=2$.

1475: We denote by $C(r,s)$ the cable space of type $(r,s)$,

1476: and by $Q(r,s)$ the Seifert fibered manifold

1477: with orbifold a disk with two cone points of index $r$ and $s$.

1478: Note that $C(2,1)$ contains a M\"{o}bius band and essential annulus

1479: with the boundaries on the outside torus, hitting the attached

1480: solid torus once and twice, respectively.

1481: 

1482: 

1483: 

1484: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1485: \section{$(\mathcal{S},\mathcal{T})$ case} \label{sec:ST}

1486: 

1487: \begin{proof}[Proof of Theorem \ref{thm:ST}]

1488: Assume that $\Delta=\Delta(\gamma_1,\gamma_2)\geq 3$.

1489: Recall that $M(\gamma_2)$ is irreducible

1490: and boundary irreducible.

1491: 

1492: First, assume that $n_2\geq 3$.

1493: %(Moreover, assume that $n_2\ne 4$ when $G_2$ is of type $\mathcal{T}$.)

1494: Theorem \ref{thm:extremalS}(2) says that

1495: $G_1^+$ contains an extremal block $\Lambda$

1496: with a disk support $D$ so that each boundary vertex, except $y_0$,

1497: has at least $2 n_2$ consecutive edge endpoints of $\Lambda$.

1498: If $G_2$ is of type $\mathcal{K}$, then we can choose a non-$sl$-label $x$ of $G_1$ by Lemma \ref{lem:Schar}.

1499: Assume that $G_2$ is of type $\mathcal{T}$.

1500: 

1501: 

1502: \begin{claim}\label{claim:Tcase}

1503: Either $G_1$ has a label $x$, which is not a label of $S$-cycles in $G_1$, or

1504: $M(\gamma_2)$ contains a Klein bottle which meets $V_{\gamma_2}$ at least two times.

1505: \end{claim}

1506: 

1507: \begin{proof}[Proof of Claim \ref{claim:Tcase}]

1508: If $n_2=3$, then $G_1$ does not have a Scharlemann cycle by Lemma \ref{lem:Scharsep}.

1509: Hence any label is a desired one.

1510: Assume $n_2\ge 4$.

1511: If $G_1$ has four labels of $S$-cycles, then

1512: there are two $S$-cycles whose label pairs are disjoint.

1513: Then $M(\gamma_2)$ contains a Klein bottle $R$ by the argument of the proof of \cite[Lemma 3.10]{GL2}.

1514: If $R$ meets $V_{\gamma_2}$ in a single meridian disk, then take a double covering torus $\widetilde{R}$ of $R$

1515: in $M(\gamma_2)$.

1516: By the assumption $n_2\ge 4$, $\widetilde{R}$ is compressible.

1517: Then $M(\gamma_2)$ is a Seifert fibered manifold, called a prism manifold,

1518: which has finite fundamental group.

1519: This contradicts the fact that $M(\gamma_2)$ is toroidal.

1520: Hence $R$ meets $V_{\gamma_2}$ at least two times.

1521: \end{proof}

1522: 

1523: In the second conclusion of Claim \ref{claim:Tcase},

1524: if the Klein bottle meets $V_{\gamma_2}$ just two times, then we jump to the case of type $\mathcal{K}$

1525: where $n_2=2$ below.

1526: Thus we have a label $x$ of $G_1$ which is not an $sl$-label when $G_2$ is of type $\mathcal{K}$,

1527: or which is not a label of $S$-cycles otherwise.

1528: 

1529: 

1530: 

1531: %%%%%%%%%%%%%%%%%%%%%%%%%

1532: Let $\Lambda^{x}$ be the subgraph of $\Lambda$

1533: consisting of all vertices and $x$-edges.

1534: Then each boundary vertex of $\Lambda^{x}$, except $y_0$,

1535: has at least two edges attached with label $x$,

1536: which cannot be parallel by Lemma \ref{lem:parallel}(3-4).

1537: Note that $\Lambda^{x}$ may not be connected.

1538: Then, apply the argument in Section \ref{sec:extremal}

1539: to the present situation;

1540: choose an extremal block $\Lambda'$ of $\Lambda^x$ with a disk support $D'$

1541: in $D$, which we can define in a similar way.

1542: 

1543: Let $v,e$ and $f$ be the numbers of vertices, edges, and disk faces

1544: of $\Lambda'$, respectively.

1545: Also let $v_i, v_{\partial}$ and $v_g$ be the numbers

1546: of interior vertices, boundary vertices and ghost vertices.

1547: Hence $v=v_i+v_{\partial}$ and $v_g=0$ or $1$.

1548: 

1549: Suppose that $\Lambda'$ has a bigon.

1550: By Lemma \ref{lem:parallel}, it contains either a generalized $S$-cycle or

1551: an extended $S$-cycle.

1552: (Recall that $x$ itself is not a label of $S$-cycle.)

1553: But this is impossible by Lemma \ref{lem:Schar}.

1554: Thus each face of $\Lambda'$ is a disk with at least $3$ sides.

1555: Hence we have $3f + v_{\partial} \leq 2e$.

1556: Since $\Lambda'$ has only disk faces, combined with $v-e+f=\chi(D')=1$,

1557: we get $e \leq 3v_i + 2v_{\partial} - 3$.

1558: On the other hand, we have $2(v_{\partial} - v_g) + \Delta v_i \leq e$,

1559: because each boundary vertex of $\Lambda'$, except $y_0$,

1560: has at least two edges attached with label $x$, and $x$ is not a label of level edges.

1561: These two inequalities give us that $3 \leq 2v_g$, a contradiction.

1562: 

1563: %%%%%%%%%%%%%%%

1564: Assume that $n_2 = 1$ when $G_2$ is of type $\mathcal{T}$.

1565: Then $G_2$ has only positive edges.

1566: Thus it contains an $x$-face for a non-$sl$-label $x$ of $G_2$,

1567: contradicting Theorems \ref{thm:Pxface} and \ref{thm:Sxface}.

1568: 

1569: 

1570: %%%%%%%%%%%%%%%

1571: Assume that $n_2 = 2$ when $G_2$ is of type $\mathcal{K}$.

1572: This case is done by Theorem \ref{thm:extremalS}(2) and the argument of \cite[Section6]{JLOT}, which can be

1573: carried over without change.

1574: 

1575: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1576: Finally assume that $n_2 = 2$ when $G_2$ is of type $\mathcal{T}$

1577: (or $n_2 = 1$ when of type $\mathcal{K}$).

1578: %The case $(\mathcal{S},\mathcal{T})$ was proved in \cite[Lemma 4.3]{BZ1}, and so

1579: %we consider only the cases $(\mathcal{P},\mathcal{T})$, $(\mathcal{S},\mathcal{K})$ and $(\mathcal{P},\mathcal{K})$.

1580: Assume for contradiction that $\Delta \geq 4$.

1581: An Euler characteristic calculation on $\overline{G}_2$

1582: shows that each vertex has valency at most 6.

1583: Then the graph looks like a subgraph of the graph

1584: shown in Figure \ref{fig:torus}

1585: ((a) for type $\mathcal{T}$ and (b) for type $\mathcal{K}$) \cite[Lemma 5.2]{G1}.

1586: Here, $p_i \geq 0$ denotes the number of edges in each parallelism class.

1587: 

1588: 

1589: \begin{claim}\label{claim:posi}

1590: When $G_2$ is of type $\mathcal{T}$, all non-loop edges of $G_2$ are negative.

1591: \end{claim}

1592: 

1593: \begin{proof}[Proof of Claim \ref{claim:posi}]

1594: This follows from the same argument as the proof of Claim \ref{claim:annulus2}.

1595: \end{proof}

1596: 

1597: 

1598: Thus $p_1 \leq \frac{n_1}{2}+1$, and $p_i \leq n_1-1$ for $i=2,3,4,5$

1599: by Lemmas \ref{lem:parallel} and \ref{lem:parallel-negative}.

1600: Since $\Delta n_1 \leq (n_1+2)+4(n_1-1)=5n_1-2$,

1601: $\Delta=4$ and all $p_i$'s are non-zero.

1602: Without loss of generality, we can assume that $p_1+p_2+p_3 \geq 2n_1$.

1603: Since $2n_1 < p_1+p_2+p_3+1 \leq (\frac{n_1}{2}+1)+2(n_1-1)+1 < 3 n_1$,

1604: we can write that $p_1+p_2+p_3+1=2n_1+r$, where $0<r<n_1$.

1605: Then $p_1=2n_1+r-(p_2+p_3+1) \geq 2n_1+r-2(n_1-1)-1=r+1$.

1606: Hence $1 \leq r \leq p_1-1$.

1607: Thus the loop family corresponding to $p_1$

1608: have two (non-level) edges with the same label $r$, and so the family contains

1609: an $S$-cycle or a generalized $S$-cycle.

1610: This is impossible when $G_1$ is of type $\mathcal{P}$.

1611: When $G_1$ is of type $\mathcal{S}$, $G_2$ contains an $S$-cycle,

1612: so $M(\gamma_1)$ must be of type $\mathcal{P}$,

1613: a contradiction again.

1614: \end{proof}

1615: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1616: 

1617: \begin{figure}[ht]

1618: \epsfbox{fig07.eps}

1619: \caption{Torus and Klein bottle}\label{fig:torus}

1620: \end{figure}

1621: 

1622: Finally we give an example realizing the case $\Delta=3$ in Theorem \ref{thm:ST}.

1623: Let $M$ be the manifold obtained from the Whitehead link exterior

1624: by Dehn filling one component with slope $6$.

1625: Then $M$ is hyperbolic \cite[p.286]{BZ1} and $M(1)=L(2,1)\sharp L(3,1)$.

1626: Since $M$ contains a once-punctured Klein bottle whose boundary has slope $4$,

1627: $M(4)$ contains a Klein bottle hitting the attached solid torus $V$ once.

1628: Also, the boundary torus of a neighborhood of the Klein bottle in $M(4)$ is known to be incompressible.

1629: Clearly, this torus meets $V$ twice.

1630: (If $M(4)$ contains a torus meeting $V$ once, then

1631: such a torus is non-separating in $M(4)$.

1632: This is impossible, because $H_1(M(4))$ is finite.)

1633: 

1634: 

1635: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1636: \begin{thebibliography}{CGLS}

1637: \bibitem[BZ1]{BZ1} S. Boyer and X. Zhang,

1638:             {\em Reducing Dehn filling and toroidal Dehn filling},

1639:             Topology Appl. \textbf{68} (1996), 285--303.

1640: \bibitem[BZ2]{BZ2} S. Boyer and X. Zhang,

1641:             {\em On Culler-Shalen seminorms and Dehn filling},

1642:             Ann. of Math. \textbf{148} (1998), 1--66.

1643: \bibitem[BW]{BW} G. Bredon and J. Wood,

1644:             {\em Non-orientable surfaces in orientable 3-manifolds},

1645:             Invent. Math. \textbf{7} (1969), 83--110.

1646: \bibitem[CGLS]{CGLS} M. Culler, C. Gordon, J. Luecke and P. Shalen,

1647:             {\em Dehn surgery on knots},

1648:             Ann. of Math. \textbf{125} (1987), 237--300.

1649: %\bibitem[DM]{DM} M. Domergue and D. Matignon,

1650: %            {\em Dehn surgeries and $P^2$-reducible 3-manifolds},

1651: %            Topology Appl. \textbf{72} (1996), 135--148.

1652: \bibitem[EW]{EW} M. Eudave-Mu\~{n}oz and Y. Q. Wu,

1653:             {\em Non-hyperbolic Dehn fillings on hyperbolic 3-manifolds},

1654:             Pacific J. Math. \textbf{190} (1999), 261--275.

1655: \bibitem[G1]{G1} C. Gordon,

1656:             {\em Boundary slopes on punctured tori in 3-manifolds},

1657:             Trans. Amer. Math. Soc. \textbf{350} (1998), 1713--1790.

1658: \bibitem[G2]{G2} C. Gordon,

1659:             {\em Small surfaces and Dehn filling},

1660:             Proceedings of the Kirbyfest,

1661:             Geom. Topol. Monogr. \textbf{2} (1999), 177--199 (electronic).

1662: %\bibitem[Go3]{Go3} C. Gordon,

1663: %            {\em Toroidal Dehn surgeries on knots in lens spaces},

1664: %            Math. Proc. Cambridge Philos. Soc. \textbf{125} (1999), 433--440.

1665: \bibitem[GLi]{GLi} C. Gordon and R. Litherland,

1666:             {\em Incompressible planar surfaces in 3-manifolds},

1667:             Topology Appl. \textbf{18} (1984), 121--144.

1668: \bibitem[GL1]{GL1} C. Gordon and J. Luecke,

1669:             {\em Reducible manifolds and Dehn surgery},

1670:             Topology \textbf{35} (1996), 385--409.

1671: \bibitem[GL2]{GL2} C. Gordon and J. Luecke,

1672:             {\em Dehn surgeries on knots creating essential tori, I},

1673:             Comm. Anal. Geom. \textbf{3} (1995), 597--644.

1674: \bibitem[GL3]{GL3} C. Gordon and J. Luecke,

1675:             {\em Toroidal and boundary-reducing Dehn fillings},

1676:             Topology Appl. \textbf{93} (1999), 77--90.

1677: \bibitem[GW]{GW} C. Gordon and Y. Q. Wu,

1678:             {\em Toroidal and annular Dehn fillings},

1679:             Proc. London Math. Soc. \textbf{78} (1999), 662--700.

1680: %\bibitem[GW2]{GW2} C. Gordon and Y. Q. Wu,

1681: %            {\em Annular and boundary reducing Dehn fillings},

1682: %            Topology \textbf{39} (2000), 531--548.

1683: %\bibitem[GW3]{GW3} C. Gordon and Y. Q. Wu,

1684: %            {\em Annular Dehn fillings},

1685: %            Comment. Math. Helv. \textbf{75} (2000), 430--456. 

1686: \bibitem[HM]{HM} C. Hayashi and K. Motegi,

1687:             {\em Only single twists on unknots can produce

1688:             composite knots},

1689:             Trans. Amer. Math. Soc. \textbf{349} (1997), 4465--4479.

1690: \bibitem[H]{H} J. Hoffman,

1691:             {\em There are no strict great $x$-cycles after

1692:             a reducing or $P^2$ surgery on a knot},

1693:             J. Knot Theory Ramifications \textbf{7} (1998), 549--569.

1694: \bibitem[HoM]{HoM} J. Hoffman and D. Matignon,

1695:             {\em Producing essential 2-spheres},

1696:             Topology Appl. \textbf{124} (2002), 435--444.

1697: \bibitem[JLOT]{JLOT} G.T. Jin, S. Lee, S. Oh and M. Teragaito,

1698:             {\em $P^2$-reducing and toroidal Dehn fillings},

1699:             to appear in Math. Proc. Cambridge Philos. Soc.

1700: \bibitem[LOT]{LOT} S. Lee, S. Oh and M. Teragaito,

1701:             {\em Dehn fillings creating essential spheres and tori},

1702:             J. Knot Theory Ramifications \textbf{11} (2002), 887--890.

1703: \bibitem[M]{M} D. Matignon,

1704:             {\em $P^2$-reducibility of 3-manifolds},

1705:             Kobe J. Math. \textbf{14} (1997), 33--47.

1706: \bibitem[O]{O} S. Oh,

1707:             {\em Reducible and toroidal 3-manifolds obtained

1708:             by Dehn fillings},

1709:             Topology Appl. \textbf{75} (1997), 93--104.

1710: %\bibitem[O2]{O2} S. Oh,

1711: %            {\em Dehn filling, reducible 3-manifolds, and Klein bottles},

1712: %            Proc. Amer. Math. Soc. \textbf{126--1} (1998), 289--296.

1713: \bibitem[S]{S} M. Scharlemann,

1714:             {\em Producing reducible 3-manifolds by surgery on a knot},

1715:             Topology \textbf{29} (1990), 481--500.

1716: \bibitem[Te]{Te} M. Teragaito,

1717:             {\em Dehn surgery and projective planes},

1718:             Kobe J. Math. \textbf{13} (1996), 203--207.

1719: \bibitem[Th]{Th} W. Thurston,

1720:             {\em Three-dimensional manifolds, Klein groups and

1721:             hyperbolic geometry},

1722:             Bull. Amer. Math. Soc. \textbf{6} (1982), 357--381.

1723: \bibitem[W1]{W1} Y. Q. Wu,

1724:             {\em The reducibility of surgered 3-manifolds},

1725:             Topology Appl. \textbf{43} (1992), 213--218.

1726: \bibitem[W2]{W2} Y. Q. Wu,

1727:             {\em Dehn fillings producing reducible manifolds and

1728:             toroidal manifolds},

1729:             Topology \textbf{37} (1998), 95--108.

1730: \bibitem[W3]{W3} Y. Q. Wu,

1731:             {\em Sutured manifold hierarchies, essential laminations,

1732:             and Dehn surgery},

1733:             J. Diff. Geom. \textbf{48} (1998), 407--437.

1734: \end{thebibliography}

1735: 

1736: 

1737: 

1738: \end{document}

1739: