1: 

2: 

3: \input amstex

4: \documentstyle{amsppt}

5: \nologo

6: \NoBlackBoxes

7: %%\NoRunningHeads

8: %\mag=1200

9: %\hsize=31 pc  

10: %\vsize=44 pc  

11: %\hcorrection{5mm}

12: %\baselineskip 15pt

13: 

14: \def\journame{COMMENT.\ MATH.\ HELV.}

15: \def\issueinfo{(to appear).  Preprint 1999.}

16: 

17: \def\jourlogo{%

18:   \vbox to\headlineheight{%

19:     \baselineskip 7pt 

20:     \noindent{\sixrm{\journame} \hfill{$ $}\newline

21:     \issueinfo \hfill{$ $} }

22: }  }

23: \def\flheadline{\jourlogo} \let\frheadline\flheadline

24: \hcorrection{19mm}

25: 

26: \topmatter

27: \title    Annular Dehn fillings

28: \endtitle

29: \author    Cameron McA. Gordon$^{1}$ and Ying-Qing Wu$^2$

30: \endauthor

31: \leftheadtext{C.~M{\fiverm c}A.~Gordon and Y-Q.~Wu}

32: \rightheadtext{Annular Dehn fillings}

33: \address Department of Mathematics, University of Texas at Austin,

34: Austin, TX 78712

35: \endaddress

36: \email  gordon\@math.utexas.edu

37: \endemail

38: \address Department of Mathematics, University of Iowa, Iowa City, IA

39: 52242

40: \endaddress

41: \email  wu\@math.uiowa.edu

42: \endemail

43: \keywords 3-manifolds, Dehn filling, essential annuli

44: \endkeywords

45: \subjclass  Primary 57N10

46: \endsubjclass

47: \thanks  $^1$ Partially supported by NSF grant \#DMS 9626550.

48: \endthanks 

49: \thanks  $^2$ Partially supported by NSF grant \#DMS 9802558.

50: \endthanks

51: 

52: \abstract

53: We show that if a simple 3-manifold $M$ has two Dehn fillings at

54: distance $\Delta \geq 4$, each of which contains an essential annulus,

55: then $M$ is one of three specific 2-component link exteriors in

56: $S^3$.  One of these has such a pair of annular fillings with $\Delta

57: = 5$, and the other two have pairs with $\Delta = 4$.  

58: \endabstract

59: 

60: \endtopmatter

61:  

62: \document

63: \define\proof{\demo{Proof}}

64: \define\endproof{\qed \enddemo}

65: \define\nequiv{\not \equiv}

66: \redefine\rga{\hat{\Gamma}_1}

67: \redefine\rgb{\hat{\Gamma}_2}

68: \redefine\rgi{\hat{\Gamma}_i}

69: \define\a{\alpha}

70: \redefine\b{\beta}

71: \redefine\tilde{\widetilde}

72: \redefine\hat{\widehat}

73: \redefine\lll{\lambda}

74: \redefine\LLL{\Lambda}

75: \define\rrr{\gamma}

76: \redefine\d{\delta}

77: \redefine\D{\Delta}

78: \redefine\bdd{\partial}

79: \define\em{\it}

80: \define\Int{\text{\rm Int}}

81: \define\val{\text{\rm val}}

82: 

83: \input epsf.tex

84: 

85: 

86: \head \S 1.  Introduction

87: \endhead

88: 

89: Let $M$ be a (compact, connected, orientable) 3-manifold with a torus

90: boundary component $T_0$.  If $r$ is a {\it slope\/} (the isotopy

91: class of an essential unoriented simple loop) on $T_0$, then as usual

92: we denote by $M(r)$ the 3-manifold obtained from $M$ by {\it $r$-Dehn

93: filling}, that is, attaching a solid torus $J$ to $M$ along $T_0$ in

94: such a way that $r$ bounds a meridian disk in $J$.  

95: 

96: We shall say that a compact, connected, orientable $3$-manifold $M$ is

97: {\it simple\/} if it contains no essential surface of non-negative

98: Euler characteristic, i.e., sphere, disk, annulus or torus.  If $M$

99: has non-empty boundary and is not the 3-ball, then $M$ is simple if

100: and only if $M$ with its boundary tori removed has a complete

101: hyperbolic structure of finite volume with totally geodesic boundary

102: [Th1, Th2].  If $M$ is closed, then the geometrization conjecture

103: asserts that $M$ is simple if and only if $M$ is either hyperbolic or

104: belongs to a certain small class of Seifert fiber spaces [Th1, Th2].

105: 

106: If $M$ is hyperbolic, then Dehn fillings on $M$ are hyperbolic if we

107: exclude finitely many slopes from each torus boundary component [Th1,

108: Th2].  By doubling $M$ along its non-torus boundary components, we see

109: that if $M$ is simple then $M(r)$ is simple for all but finitely many

110: slopes $r$ on any given torus boundary component $T_0$, and a good

111: deal of attention has been directed towards obtaining a more precise

112: quantification of this statement.  Denote by $\Delta (r_1, r_2)$ the

113: distance, or minimal geometric intersection number, between two slopes

114: $r_1, r_2$ on a torus.  Define a 3-manifold to be of type $S$, $D$,

115: $A$ or $T$ if it contains an essential sphere, disk, annulus or torus,

116: respectively.  For $X_i \in \{S, D, A, T\}$, $i=1,2$, define

117: $\Delta(X_1, X_2)$ to be the maximum of $\Delta(r_1, r_2)$, where

118: $r_1$ and $r_2$ are Dehn filling slopes of some simple manifold $M$

119: such that $M(r_i)$ is of type $X_i$.  These numbers $\Delta (X_1,

120: X_2)$ are now known in all ten cases; see [GW2] for more details.

121: 

122: Except when $(X_1, X_2) = (A,A)$, $(A,T)$ or $(T,T)$, it is also

123: known that $\Delta(X_1, X_2)$ is realized by infinitely many

124: simple manifolds $M$; see [EW].  On the other hand, $\Delta(T, T)

125: = 8$, and there are exactly two simple manifolds $M$ admitting

126: toroidal fillings $M(r_1)$, $M(r_2)$ with $\Delta = \Delta(r_1, r_2) =

127: 8$, exactly one with $\Delta = 7$, exactly one with $\Delta = 6$, and

128: infinitely many with $\Delta = 5$ [Go].  Similarly, $\Delta(A, T) =

129: 5$ [Go, GW1], and there is exactly one simple manifold $M$ having

130: an annular filling $M(r_1)$ and a toroidal filling $M(r_2)$ with

131: $\Delta = 5$, exactly two with $\Delta = 4$, and infinitely many with

132: $\Delta = 3$ [GW1].  In the present paper we complete the picture by

133: dealing with the case $(A, A)$.  In this case, $\Delta(A, A) = 5$ [Go,

134: GW1], and there are infinitely many simple manifolds $M$ admitting

135: annular fillings $M(r_1), M(r_2)$ with $\Delta = 3$ [GW1].  Here we

136: show that there is exactly one such manifold $M$ with $\Delta = 5$,

137: and exactly two with $\Delta = 4$.  More precisely, we have the

138: following theorem.

139: 

140: \proclaim{Theorem 1.1} Suppose $M$ is a compact, connected,

141: orientable, irreducible, $\bdd$-irreducible, anannular $3$-manifold

142: which admits two annular Dehn fillings $M(r_1)$, $M(r_2)$ with $\Delta

143: = \Delta(r_1, r_2) \geq 4$.  Then one of the following holds.

144: 

145: (1)  $M$ is the exterior of the Whitehead link, and $\Delta = 4$.

146: 

147: (2)  $M$ is the exterior of the $2$-bridge link associated to the

148: rational number $3/10$, and $\Delta = 4$.

149: 

150: (3)  $M$ is the exterior of the $(-2,3,8)$ pretzel link, and $\Delta =

151: 5$.  

152: \endproclaim

153: 

154: The three manifolds listed in the theorem are the exteriors of the

155: links in $S^3$ shown in Figure 1.1.

156: 

157: \bigskip

158: \leavevmode

159: 

160: \epsfxsize=5in      % adjust the width  of the figure

161: %\epsfysize=3in      % adjust the height of the figure

162: \centerline{\epsfbox{Fig1.1.ps}} 

163: \bigskip

164: \centerline{Figure 1.1}

165: \bigskip

166: 

167: That each of these link exteriors does have a pair of annular fillings

168: with $\Delta = 4$, $4$ and $5$ respectively is proved in [GW1].  The

169: fillings in question are also toroidal [GW1], so in fact these are

170: exactly the same manifolds which admit an annular and a toroidal

171: filling with $\Delta \geq 4$ [GW1, Theorem 1.1].  Using [GW1, Theorem

172: 1.1], Qiu has independently proved Theorem 1.1 in the special case

173: where $M$ is the exterior of a knot in a solid torus [Q].

174: 

175: According to the proof of [GW1, Theorem 7.5], the annular fillings on

176: the three manifolds listed in Theorem 1.1 are non Seifert fibered

177: graph manifolds.  If $M$ admits some Seifert fibered surgery, then

178: $\bdd M$ consists of tori, in which case $M$ is hyperbolic if and only

179: if it is simple.  Hence the following corollary is an immediate

180: consequence of Theorem 1.1.

181: 

182: \proclaim{Corollary 1.2} Suppose $M$ is a compact orientable

183: hyperbolic 3-manifold with at least two torus boundary components, and

184: suppose $M(r_1), M(r_2)$ are Seifert fibered manifolds.  Then $\Delta

185: (r_1, r_2) \leq 3$.  \endproclaim

186: 

187: The condition that $M$ has at least two boundary components cannot be

188: removed.  For example, if $M$ is the figure 8 knot complement, then

189: $M(3)$ and $M(-3)$ are Seifert fibered, and $\Delta (-3, 3) = 6$.  It

190: is not known whether the bound 3 in the corollary is the best possible.

191: 

192: The proof of Theorem 1.1 proceeds as follows.  For $\alpha = 1,2$, let

193: $A_{\a}$ be an essential annulus in $M(r_{\a})$, meeting the Dehn

194: filling solid torus $J_{\a}$ in $n_{\a}$ meridian disks, with $n_{\a}$

195: minimal over all choices of $A_{\a}$.  This gives rise to a punctured

196: annulus $F_{\a} = A_{\a} \cap M$ in $M$, such that the boundary

197: components of $F_{\a}$ which lie on $T_0$ have slope $r_{\a}$, $\a =

198: 1,2$.  The arcs of intersection of $F_1$ and $F_2$ then define labeled

199: graphs $G_{\a}$ in $A_{\a}$ with $n_{\a}$ vertices, $\a = 1,2$.  We

200: assume that $\Delta = \Delta(r_1, r_2) = 4$ or $5$, and show by a

201: detailed analysis that there are only three such pairs of graphs,

202: corresponding to the three examples listed in the theorem.  

203: 

204: The paper is organized as follows.  In Section 2 we give some

205: definitions and establish some basic properties of the graphs

206: $G_{\a}$.  In Section 3 we show that any graph in an annulus with no

207: trivial loops or parallel edges must satisfy one of four

208: possibilities; if the reduced graph $\hat G_{\a}$ is of the fourth

209: type we say that $G_{\a}$ is {\it special}.  Section 4 is devoted to

210: showing that if one of the graphs $G_1, G_2$ is special then they both

211: are, and (up to relabeling) $n_1 = 1$, $n_2 = 2$.  Section 5 considers

212: the generic case, $n_1, n_2 > 2$.  This is shown to be impossible, by

213: eliminating in turn the first three possibilities of Section 3 for the

214: reduced graphs $\hat G_{\a}$.  Section 6 shows that the case $n_1 =

215: 2$, $n_2 > 2$ is also impossible.  In Section 7 we show that if $G_1$

216: and $G_2$ are special, so $n_1 = 1$ and $n_2 = 2$, then there is

217: exactly one possible pair $G_1, G_2$, with $\Delta = 4$, corresponding

218: to case (1) of Theorem 1.1.  Finally in Section 8, we show that if

219: $G_1, G_2$ are not special and $n_1, n_2 \leq 2$, then there are

220: exactly two possible pairs $G_1, G_2$, one with $\Delta = 4$, $n_1 =

221: n_2 = 2$, and one with $\Delta = 5$, $n_1 = n_2 = 2$, corresponding to

222: cases (2) and (3) of Theorem 1.1.

223: 

224: We would like to thank the referee for his/her careful reading and

225: helpful comments.

226: 

227: 

228: \head \S 2.  Preliminary Lemmas

229: \endhead

230: 

231: Throughout this paper, we will always assume that $M$ is a compact,

232: connected, irreducible, $\bdd$-irreducible, anannular 3-manifold, with

233: a torus boundary component $T_0$.  We use $\a, \b$ to denote the

234: numbers $1$ or $2$, with the convention that if they both appear, then

235: $\{\a, \b\} = \{1,2\}$.  Let $r_1, r_2$ be slopes on $T_0$ such that

236: $M(r_1), M(r_2)$ are annular, and let $A_{\a}$ be an essential annulus

237: in $M(r_{\a})$ such that $n_{\a}$, the number of components of

238: intersection of $A_{\a}$ with the Dehn filling solid torus $J_{\a}$,

239: is minimal among all essential annuli in $M(r_{\a})$, $\a = 1,2$.

240: Denote by $F_{\a}$ the punctured annulus $A_{\a } \cap M$.  Denote by

241: $\Delta = \Delta(r_1, r_2)$ the minimal geometric intersection number

242: between $r_1$ and $r_2$.  By [Go, Theorem 1.3] we have $\Delta \leq

243: 5$.  Throughout this paper, we will always assume $\Delta = 4$ or $5$,

244: unless otherwise stated.

245: 

246: Minimizing the number of components of $F_1 \cap F_2$ by an isotopy,

247: we may assume that $F_1\cap F_2$ consists of arcs and circles which

248: are essential on both $F_{\a}$.  Let $u_1,\ldots, u_{n_{\a}}$ be the

249: disks that are the components of $A_{\a}\cap J_{\a}$, labeled

250: successively when traveling along $J_{\a}$.  Similarly let $v_1,

251: \ldots, v_{n_{\b}}$ be the disks in $A_{\b}\cap J_{\b}$.  Let $G_{\a}$

252: be the graph on $A_{\a}$ with the $u_i$'s as (fat) vertices, and the

253: arc components of $F_1\cap F_2$ with at least one endpoint on $T_0$ as

254: edges.  Note that we do not regard an edge endpoint on the boundary of

255: the annulus as a vertex, so we are abusing terminology somewhat in

256: that our graphs may have edge endpoints that do not lie on vertices.

257: The minimality of the number of components in $F_1\cap F_2$ and the

258: minimality of $n_{\a}$ imply that $G_{\a}$ has no trivial loops, and

259: that each disk face of $G_{\a}$ in $A_{\a}$ has interior disjoint from

260: $F_{\b}$.

261: 

262: An edge $e$ of a graph $G$ on an annulus $A$ is a {\it boundary

263: edge\/} if it has one endpoint on the boundary of $A$, otherwise it is

264: an {\it interior edge}.  A vertex $v$ of $G$ is a {\it boundary vertex

265: \/} if it is incident to a boundary edge, otherwise it is an {\it

266: interior vertex}.  Similarly, a face of $G$ is a {\it boundary face\/}

267: if it contains a boundary edge.

268: 

269: If $e$ is an edge of $G_{\a}$ with an endpoint $x$ on a fat vertex

270: $u_i$, then $x$ is labeled $j$ if $x$ is in $\bdd u_i\cap \bdd v_j$.

271: When going around the boundary of a vertex in

272: $G_{\a}$, the labels of the edge endpoints appear as $1, 2, \ldots,

273: n_{\b}$ repeated $\Delta$ times.

274: 

275: An edge $e$ at a vertex $u_i$ of $G_{\a}$ is called a {\it $j$-edge\/}

276: at $u_i$ if it has an endpoint at $u_i$ labeled $j$.  Dually, a

277: $j$-edge at $u_i$ is also an $i$-edge at $v_j$ in $G_{\b}$.  We say

278: that $e$ is an {\it $(i,k)$-edge\/} if it has labels $i$ and $k$ at

279: its two endpoints.

280: 

281: Each vertex of $G_{\a}$ is given a sign according to whether $J_{\a}$

282: passes $A_{\a}$ from the positive side or negative side at this

283: vertex.  Two vertices of $G_{\a}$ are {\it parallel\/} if they have

284: the same sign, otherwise they are {\it antiparallel.}  Note that if

285: $A_{\a}$ is a separating surface in $M(r_{\a})$, then $n_{\a}$ is

286: even, and $u_i, u_j$ are parallel if and only if $i, j$ have the same

287: parity.  An interior edge of $G_{\a}$ is a {\it positive edge\/} if it

288: connects parallel vertices.  Otherwise it is a {\it negative edge}.

289: We use $\val(v, G)$ to denote the valency of a vertex $v$ in a graph

290: $G$.

291: 

292: By considering each family of parallel edges of $G_{\a}$ as a single

293: edge $E$, we get the {\it reduced graph\/} $\hat{G}_{\a}$ on $A_{\a}$.

294: It has the same vertices as $G_{\a}$.  Denote by $|E|$ the number of

295: edges in $G_{\a}$ represented by $E$.

296: 

297: A cycle in $G_{\a}$ consisting of positive edges is a {\it Scharlemann

298: cycle\/} if it bounds a disk with interior disjoint from the graph,

299: and all the edges in the cycle have the same pair of labels $(i, i+1)$

300: at their two endpoints.  ($i+1 = 1$ if $i=n_{\b}$.)  The pair $(i,

301: i+1)$ is called the {\it label pair\/} of the Scharlemann cycle.  In

302: particular, a pair of adjacent parallel positive edges with the same

303: label pair is a Scharlemann cycle.  The boundary of the disk $D$

304: bounded by a Scharlemann cycle consists of edges of the Scharlemann

305: cycle and some arcs on the annulus $C_i$ on $T_0$ between $\bdd v_i$

306: and $\bdd v_{i+1}$.  When $n_{\b} = 2$, the two annuli $C_1$ and $C_2$

307: are still distinct, allowing one to differentiate between a

308: $(1,2)$-Scharlemann cycle and a $(2,1)$-Scharlemann cycle.  A pair of

309: edges $\{e_1, e_2\}$ is an {\it extended Scharlemann cycle\/} if there

310: is a Scharlemann cycle $\{e'_1, e'_2\}$ such that $e_i$ is parallel

311: and adjacent to $e'_i$.

312: 

313: A subgraph $G'$ of a graph $G$ on a surface $F$ is {\it essential\/}

314: if it is not contained in a disk in $F$. 

315: 

316: \proclaim{Lemma 2.1} 

317: (1) {\rm (The Parity Rule)} An edge $e$ is a positive edge in $G_1$ if

318: and only if it is a negative edge in $G_2$.

319: 

320: (2) A pair of edges cannot be parallel on both $G_1$ and $G_2$.

321: 

322: (3) If $G_{\a}$ has a set of $n_{\b}$ parallel negative edges, then on

323: $G_{\b}$ they form mutually disjoint essential cycles of equal length.

324: 

325: (4) If $G_{\a}$ has a Scharlemann cycle, then $A_{\b}$ is

326: separating, and $n_{\b}$ is even.  Moreover, the edges of the

327: Scharlemann cycle and the vertices at their endpoints form an

328: essential subgraph of $G_{\b}$.

329: 

330: (5)  $G_{\a}$ contains no extended Scharlemann cycle.

331: \endproclaim

332: 

333: \proof See [GW1, Lemma 2.2], except for (2) in the case that the pair

334: of edges $e_1, e_2$ are boundary edges.  If $e_1, e_2$ are boundary

335: edges parallel on both $G_1, G_2$, then they cut off bands $B_1, B_2$

336: on the punctured annuli $F_1, F_2$, which can be glued together to get

337: an annulus in the manifold $M$, which intersects the Dehn filling

338: torus $T_0$ in an essential circle.  This contradicts the assumption

339: that $M$ is $\bdd$-irreducible and anannular.  \endproof

340: 

341: Let $E$ be an edge of $\hat G_{\a}$ representing $n_{\b}$ parallel

342: negative edges on $G_{\a}$, connecting $u_i$ to $u_j$.  Then $E$

343: defines a permutation $\varphi: \{1, \ldots, n_{\b}\} \to \{1, \ldots,

344: n_{\b}\}$, such that an edge $e$ in $E$ has label $k$ at $u_i$ if and

345: only if it has label $\varphi(k)$ at $u_j$.  Call $\varphi$ the {\it

346: permutation associated to $E$}.  Because of the ambiguity in the order

347: of $u_i, u_j$, the permutation is only well defined up to inverse.  An

348: {\it $E$-orbit\/} is an orbit of $\varphi$.  Such an orbit determines

349: a cycle in $G_{\b}$ consisting of the edges of $E$ with endpoint

350: labels in this orbit, called the cycle of this orbit.  Note that all

351: the vertices in a cycle are parallel.  Topologically each such cycle

352: is a circle.  Lemma 2.1(3) says that these circles are mutually

353: disjoint, mutually parallel, essential circles on the annulus

354: $A_{\b}$.

355: 

356: 

357: \proclaim{Lemma 2.2} 

358: (1) Any two Scharlemann cycles on $G_{\a}$ have the same label pair.

359: 

360: (2) If $E$ is a positive edge in $\hat G_{\a}$, then $|E| \leq

361: n_{\b}/2 + 1$.  Moreover, if $|E| = n_{\b}/2 + 1$, then the

362: corresponding edges of $G_{\a}$ contain a Scharlemann cycle.

363: 

364: (3) Any family of parallel interior edges in $G_{\a}$ contains at

365: most $n_{\b}$ edges.  \endproclaim

366: 

367: \proof 

368: See [GW1, Lemma 2.5].

369: \endproof

370: 

371: 

372: \proclaim{Lemma 2.3}

373: (1)  If some vertex of $G_{\a}$ has more than $n_{\b}$ negative

374: edge endpoints, then $G_{\b}$ contains a Scharlemann cycle.

375: 

376: (2)  No vertex of $G_{\a}$ has more than $2n_{\b}$ negative edge

377: endpoints.  

378: \endproclaim

379: 

380: \proof 

381: For any label $i$ of $G_{\b}$, let $G_{\b}^+(i)$ be the subgraph of

382: $G_{\b}$ consisting of all vertices of $G_{\b}$ and all positive

383: $i$-edges of $G_{\b}$.  The edges of $G_{\b}^+(i)$ correspond to

384: the negative edges of $G_{\a}$ incident to the vertex $u_i$.  Let

385: the number of such edges be $k$.  Then if $f$ denotes the sum of the

386: Euler characteristics of the faces of $G_{\b}^+(i)$, we have 

387: $$ 0 = \chi(A_{\b}) = n_{\b} - k + f.$$

388: Therefore, if $k > n_{\b}$, $G_{\b}^+(i)$ has a disk face $D$.  Then

389: there is a Scharlemann cycle of $G_{\b}$ in $D$ by [HM, Proposition

390: 5.1].  This proves (1).

391: 

392: To prove (2), assume $k > 2n_{\b}$.  Then by the above we have $f = k

393: - n_{\b} > n_{\b}$, so $G_{\b}^+(i)$ has more than $n_{\b}$ disk

394: faces, and by [HM, Proposition 5.1] each such face contains a

395: Scharlemann cycle of $G_{\b}$.  Hence $G_{\b}$ contains $s > n_{\b}$

396: Scharlemann cycles, all on the same label pair, say $(1,2)$, by

397: Lemma 2.2(1).  Define a graph $H$ in $A_{\b}$ as follows; see [GL,

398: Proof of Theorem 2.3].  The vertices of $H$ consist of the vertices of

399: $G_{\b}$, together with a vertex $v_{D}$ in the interior of each disk

400: face of $G_{\b}$ bounded by a Scharlemann cycle.  The edges of $H$ are

401: defined by joining each vertex $v_{D}$, within $D$, to the vertices of

402: $G_{\b}$ in $\bdd D$.  Thus $H$ has $n_{\b} + s$ vertices and at least

403: $2s$ edges.  An Euler characteristic argument then shows that $H$ has

404: a disk face $E$.  This disk $E$ contains a 1-cycle of $G_{\b}$ (see

405: [CGLS, p.\ 279] for definition), and hence a Scharlemann cycle [CGLS,

406: Lemma 2.6.2].  But this contradicts the fact that $E$ is a face of

407: $H$, because by definition $H$ would have a vertex in the disk bounded

408: by this Scharlemann cycle.  \endproof

409: 

410: 

411: Let $P, Q$ be two edge endpoints on the boundary of a vertex $u$ in

412: $G_{\a}$.  Let $P_0 = P, P_1, \ldots, P_{k-1}, P_k = Q$ be the edge

413: endpoints encountered when traveling along $\bdd u$ in the direction

414: induced by the orientation of $u$.  Then the {\it distance\/} from $P$

415: to $Q$ (at the vertex $u$) is defined as $\rho_u (P,Q) = k$.  Notice

416: that since the valency of $u$ is $\Delta n_{\b}$, we have $\rho_u(Q,

417: P) =\Delta n_{\b} - \rho_u (P, Q)$.  If $e_1, e_2$ is a pair of edges,

418: each having a single endpoint $P_i$ on the vertex $u$ in $G_{\a}$,

419: then define $\rho_u(e_1, e_2) = \rho_u (P_1, P_2)$.

420: 

421: A pair of edges $e_1, e_2$ connecting two vertices $u, v$ in $G_{\a}$

422: is an {\it equidistant pair\/} if $\rho_u(e_1, e_2) =\rho_v(e_2,

423: e_1)$.  In particular, one can check that if $e_1, e_2$ are a pair of

424: parallel edges connecting a pair of parallel vertices in $G_{\a}$,

425: then $e_1, e_2$ is an equidistant pair in $G_{\a}$.  

426: 

427: \proclaim{Lemma 2.4} {\rm (The Equidistance Lemma.)}  Let $e_1, e_2$

428: be a pair of edges connecting the same vertices on $G_1$ and the same

429: vertices on $G_2$.  Then $e_1, e_2$ is an equidistant pair in $G_1$ if

430: and only if it is an equidistant pair in $G_2$.  \endproclaim

431: 

432: \proof

433: See [GW1, Lemma 2.8].

434: \endproof

435: 

436: Given two slopes $r_1, r_2$ on the torus $T_0$, let $l$ be a curve

437: intersecting $r_1$ at a single point.  Choosing $l$ and the

438: orientations of the curves properly, we may assume that homologically

439: $r_2 = q r_1 + \Delta l$, where $1\leq q < \Delta/2$.  The number $q$

440: is called the {\it jumping number\/} of $r_1, r_2$.  Note that if

441: $\Delta = 4$ then $q = 1$, and if $\Delta = 5$ then $q=1$ or $2$.

442: 

443: \proclaim{Lemma 2.5} (1) If the jumping number $q=1$, in particular if

444: $\Delta = 4$, then a pair of $j$-edges at a vertex $u_i$ in $G_{\a}$

445: are adjacent among all the $j$-edges if and only if on $G_{\b}$ they

446: are also adjacent at $v_j$ among all $i$-edges.  

447: 

448: (2) If $q=2$, then a pair of $j$-edges at a vertex $u_i$ in $G_{\a}$

449: are adjacent among all $j$-edges if and only if on $G_{\a}$ they are

450: not adjacent among all the $i$-edges at $v_j$.  \endproclaim

451: 

452: \proof

453: This is essentially [GW1, Lemma 2.10].  It was shown that if $P_1,

454: ..., P_{\Delta}$ are the consecutive $j$-edge endpoints at $u_i$, then

455: on $\bdd v_j$ they appear in the order $P_q, P_{2q}, ..., P_{\Delta

456: q}$, hence the result follows.

457: \endproof

458: 

459: A graph $G$ on an annulus $A$ is {\it special\/} if every vertex has

460: at least two nonparallel boundary edges.  Note that $G$ is special if

461: and only if the corresponding reduced graph $\hat G$ is special.

462: 

463: \proclaim{Lemma 2.6} 

464: (1) If $G$ is special then every vertex has exactly two boundary edges

465: in $\hat G$, going to distinct boundary components of $A$.

466: 

467: (2) If $G_{\a}$ has $2n_{\b}$ parallel boundary edges, then $G_{\b}$

468: is special.  $G_{\a}$ cannot have more than $2n_{\b}$ parallel

469: boundary edges.

470: 

471: (3) If some edge $E$ of $\hat G_{\a}$ represents $n_{\b}$ negative

472: edges, and if $G_{\a}$ has some positive edges, then $G_{\a}$ has at

473: most $n_{\b}$ parallel boundary edges, and each vertex of $\hat

474: G_{\b}$ has at most one boundary edge.  \endproclaim

475: 

476: \proof

477: (1) Otherwise there would be a pair of edges of $\hat G$ at some 

478: vertex $v$ 

479: going to the same boundary component of $A$.  By looking at an 

480: outermost such pair one can see that some vertex $u$ of $\hat G$ has a 

481: single boundary edge in $\hat G$, contradicting the definition of 

482: a special graph.

483: 

484: (2) If $G_{\a}$ has $2n_{\b}$ parallel boundary edges, then for any

485: label $i$ it has two parallel $i$-edges.  Since no two edges are

486: parallel on both graphs, these two edges are non-parallel on $G_{\b}$,

487: hence $G_{\b}$ is special.  If $G_{\a}$ has more than $2n_{\b}$

488: parallel boundary edges, then there is a label $i$ such that $G_{\a}$

489: has three parallel boundary $i$-edges.  Since by (1) the vertex $v_i$

490: in $\hat G_{\b}$ has only two boundary edges, two of these edges would

491: be parallel on both graphs, contradicting Lemma 2.1(2).

492: 

493: (3)  Since $G_{\a}$ has some positive edges, the vertices of $G_{\b}$

494: cannot all be parallel, so there are at least two $E$-orbits, which 

495: form parallel essential cycles on $G_{\b}$.  Hence all boundary edges 

496: at a vertex of $G_{\b}$ must be parallel to each other.  If $G_{\a}$ has

497: more than $n_{\b}$ parallel boundary edges then two of them would be

498: parallel on both graphs, contradicting Lemma 2.1(2).

499: \endproof

500: 

501: \proclaim{Lemma 2.7} Suppose all vertices of $\hat G_{\a}$ are

502: boundary vertices, and suppose there are two boundary edges $E_1, E_2$

503: of $\hat G_{\a}$ incident to the same vertex $v$ and going to the same

504: boundary component of $A_{\a}$.  Then $\hat G_{\a}$ has a vertex $v'$

505: of valency at most 3 which is incident to a single boundary edge,

506: and $G_{\b}$ is special.  \endproclaim

507: 

508: \proof Let $D$ be the disk on $A_{\a}$ cut off by $E_1\cup E_2$.

509: Since $E_1, E_2$ are nonparallel, $D$ contains a vertex $v_1 \neq v$,

510: hence by adding an edge if necessary we may assume that there is an

511: edge incident to $v$ other than $E_1, E_2$.  Let $\tilde D$ be the

512: double of $D$ along $E_1\cup E_2$, and let $\tilde G$ be the double of

513: $\hat G_{\a} \cap D$.  Then each vertex of $\tilde G$ has a boundary

514: edge.  By [CGLS, Lemma 2.6.5] $\tilde G$ has a vertex $v'$ of valency

515: at most 3 and incident to at most one boundary edge.  Since $v$

516: has valency at least $4$ in $\tilde G$, $v' \neq v$.  Hence $\val (v',

517: \hat G_{\a}) = \val (v', \tilde G) \leq 3$.  By Lemma 2.2(3) each

518: interior edge of $\tilde G_{\a}$ represents at most $n_{\b}$ edges.

519: Since $\Delta \geq 4$, this implies that the unique boundary edge at

520: $v'$ represents at least $2n_{\b}$ edges.  By Lemma 2.6(2) in this

521: case $G_{\b}$ is special.  \endproof

522: 

523: 

524: \head  \S 3.  Reduced graphs on annuli      \endhead

525: 

526: By a {\it reduced graph\/} on a surface we mean one with no trivial

527: loops or parallel edges; in other words, no faces of the graph are

528: monogons or bigons.

529: 

530: \definition{Definition 3.1}  Let $G$ be a reduced graph on an annulus

531: $A$.  Then $G$ is said to be {\it triangular\/} if 

532: 

533: (i)  every vertex has at most one boundary edge; 

534: 

535: (ii)  every interior vertex has valency 6;

536: 

537: (iii)  every boundary vertex has valency 5;

538: 

539: (iv)  every face of $G$ is a disk with three edges.

540: \enddefinition

541: 

542: We remark that the only properties of a triangular graph that we will

543: use are (i), (iii), and the fact that the graph has at least one

544: boundary vertex (which follows from (iv)).

545: 

546: \proclaim{Proposition 3.1}  Let $G$ be a reduced graph in an annulus $A$.

547: Then either 

548: 

549: (1)  $G$ contains an interior vertex of valency at most $5$; or

550: 

551: (2)  $G$ contains a boundary vertex of valency at most $4$ with

552: exactly one boundary edge; or 

553: 

554: (3)  $G$ is triangular; or

555: 

556: (4)  $G$ is special.

557: \endproclaim

558: 

559: \proof Let $G_1$ be a graph obtained from $G$ by adding extra edges so

560: as to make each face of $G_1$ a disk with three edges.  In particular,

561: in $G_1$, each boundary face has three edges, and if some vertex $v$

562: has two boundary edges $e_1, e_2$, then $v$ has an edge on each side

563: of $e_1 \cup e_2$, so it has valency at least 4.

564: 

565: Let $G_2$ be the union of $G_1$ and $\bdd A$, with the obvious graph

566: structure.  Thus the points of $G_1\cap \bdd A$ are now considered

567: vertices, and the segments of $\bdd A$ cut by these vertices are

568: considered edges of $G_2$.  Note that $\val (v, G_2) = 3$ for all

569: vertices $v$ on $\bdd A$, and each boundary face now has four edges.

570: Let $G_3$ be obtained from $G_2$ by adding a diagonal edge in each

571: boundary face of $G_2$, all sloping in the same direction; in other

572: words, no two edges added have a common vertex on $\bdd A$.  We have

573: $\val (v, G_3) = 4$ if $v \in \bdd A$.  One can see that if we remove

574: all edges and vertices on $\bdd A$ then we get a graph that is

575: obtained from $G_1$ by adding an extra copy of each boundary edge.

576: Hence if $\val (v, G_1) = p$ and $v$ has $q$ boundary edges in $G_1$,

577: then $\val(v, G_3) = p+q$.  In particular, if $v$ has two boundary

578: edges in $G_1$, then its valency in $G_3$ is at least $4+2 = 6$.  Each

579: face of $G_3$ is now a triangle.

580: 

581: The double of $A$ along $\bdd A$ is a torus $T$, and the corresponding

582: double of $G_3$ is a reduced graph $\tilde G_3$ on $T$ with triangular

583: faces.  By an Euler characteristic argument, one can show that the

584: number of edges in $\tilde G_3$ is three times the number of vertices

585: of $\tilde G_3$.  Thus either (i) some vertex $v$ of $\tilde G_3$ has

586: valency at most 5, or (ii) all vertices of $\tilde G_3$ have valency

587: 6.  All vertices on $\bdd A$ have valency 4 in $G_3$, hence valency 6

588: in $\tilde G_3$, and we have shown that if $v$ has two boundary edges

589: in $G_1$ then it has valency at least $6$ in $G_3$; therefore (i)

590: implies that either $v$ is an interior vertex of $G$ with valency at

591: most 5, or it is a boundary vertex of $G$ with valency at most $5 - q

592: \leq 4$ and incident to at most one boundary edge, so the graph is of

593: type (1) or (2) in the proposition.  Hence we may assume that all

594: vertices of $G_3$ in the interior of $A$ have valency 6.

595: 

596: If no vertex of $G_1$ has two boundary edges then each boundary vertex

597: of $G_1$ has valency $6 - q = 6 - 1 = 5$.  Since each interior vertex

598: of $G_1$ has valency 6, it follows that $G_1$ is triangular.  Since

599: $G$ is a subgraph of $G_1$ with the same vertices, either $G = G_1$

600: and hence $G$ is of type (3), or $G$ has a vertex $v$ with $\val (v,

601: G) < \val (v, G_1)$, in which case $G$ is of type (1) or (2).

602: 

603: Now assume some vertex $v$ of $G_1$ has two boundary edges $e_1, e_2$

604: going to different boundary components.  Then the valency of $v$ in

605: $G_1$ is at most $6 - 2 = 4$.  Since each face of $G_1$ has three

606: edges, there is exactly one interior edge $e'$ on each side of

607: $e_1\cup e_2$.  Let $v'$ be the other endpoint of $e'$.  Since each

608: face has three edges, $v'$ must also have two boundary edges going to

609: different boundary components of $A$.  Repeating this process, we see

610: that $G_1$ is a special graph such that each vertex has valency 4.

611: Since $G$ is a subgraph of $G_1$, either it is special, hence of type

612: (4), or it has a vertex of valency at most 3 and incident to at most

613: one boundary edge, in which case it is of type (1) or (2).

614: 

615: Finally, assume $G_1$ has a vertex $v$ which has two boundary edges

616: going to the same boundary component.  Then they cut off a disk $D$

617: from the annulus, which we may assume to be outermost.  However,

618: arguing as in the previous paragraph, we see that the vertex on the

619: other end of an edge $e'$ in $D$ incident to $v$ must have two

620: boundary edges, which is a contradiction.  Therefore this case does

621: not happen.  \endproof

622: 

623: 

624: \head   \S 4.   Special graphs

625: \endhead

626: 

627: Recall that a graph $G$ on an annulus $A$ is special if every vertex

628: has two nonparallel boundary edges.  By Lemma 2.6(1) this implies that

629: every vertex of $G$ has exactly two boundary edges in $\hat G$, going

630: to different boundary components of $A$.

631: 

632: To simplify notation, denote $n_{\b}$ by $n$.

633: 

634: 

635: \proclaim{Lemma 4.1}  

636: If $G_{\a}$ is a special graph, then $G_{\b}$ is also special.

637: \endproclaim

638: 

639: \proof First notice that since each vertex $u_i$ of $G_{\a}$ is

640: incident to at most two families of interior edges and each such

641: family contains at most $n$ edges (Lemma 2.2(3)), there are at most

642: two interior $j$-edges at $u_i$ for any $j$.  Hence there are at least

643: $\Delta -2$ boundary $j$-edges at $u_i$.  Since this is true for all

644: $i, j$, we see that each vertex $v_j$ of $G_{\b}$ has at least $2

645: n_{\a}$ ($3n_{\a}$ if $\Delta = 5$) boundary edges.  Since each

646: parallel family contains at most $2n_{\a}$ edges (Lemma 2.6(2)), the

647: lemma follows immediately when $\Delta = 5$.

648: 

649: Now assume $\Delta = 4$, and assume $G_{\b}$ is not special.  Then it

650: has a vertex $v_i$ such that all boundary edges are parallel.  By

651: Lemma 2.6(2) and the above, $v_i$ has exactly $2n_{\a}$ boundary

652: edges, all parallel to each other.  In particular, there are only two

653: boundary $1$-edges $e_1, e'_1$ at $v_i$.  Dually this means that $e_1,

654: e'_1$ are the only boundary $i$-edges at $u_1$.  Since they are

655: parallel on $G_{\b}$, they cannot be parallel on $G_{\a}$, so they

656: belong to different families of boundary edges.  Since these two edges

657: are adjacent among all $1$-edges at $v_i$, by Lemma 2.5(1) they must

658: also be adjacent among all $i$-edges at $u_1$.  This implies that the

659: two interior $i$-edges at $u_1$ are on the same side of $e_1\cup

660: e'_1$, so they belong to the same edge $E$ in $\hat G_{\a}$ because

661: there is only one interior edge of $\hat G_{\a}$ on each side of

662: $e_1\cup e'_1$.  Since by Lemma 2.2(3) $E$ contains at most $n$

663: edges, this is impossible.  \endproof

664: 

665: In the remainder of this section we will assume that both $G_1$ and

666: $G_2$ are special.

667: 

668: The sign of a vertex $u$ in $G_{\a}$ induces an orientation on $\bdd

669: u$, called its {\it preferred orientation}.  Thus the preferred

670: orientations of the $\bdd u$'s are all in the same direction on $T_0$.

671: Let $e_1, e_2$ be a pair of adjacent boundary edges at some vertex $u$

672: of $G_{\a}$.  When traveling on $\bdd u$ along the preferred

673: orientation, the labels at the endpoints of $e_1, e_2$ appear as $i,

674: i+1$ for some $i$ ($i+1=1$ if $i = n$).  They cut off a band $B$ on

675: the surface $F_{\a}$, called an {\it $i$-band} at $u$ (of $G_{\a}$).

676: Note that the label $i$ is determined by the pair $e_1, e_2$ even if

677: $n = 2$.  The edge labeled $i$ at $u$ is called the {\it initial

678: edge\/} of $B$, the other the {\it terminal edge.}  Two $i$-bands of

679: $G_{\a}$ are of different {\it types\/} if their initial edges are

680: nonparallel on $G_{\b}$; otherwise they are of the same type.

681: 

682: If $e_1, ..., e_k$ are all the edges of a parallel family $E$ at a

683: vertex $u$, appearing in this order when traveling along the preferred

684: orientation of $\bdd u$, then $e_k$ is called the {\it ending edge\/}

685: of $E$, and the label of $e_k$ at $u$ is called the {\it ending

686: label\/} of $E$.  Note that if a boundary $i$-edge $e$ is not an

687: ending edge, then it is the initial edge of an $i$-band.

688: 

689: \proclaim{Lemma 4.2}  There is a label $i$ such that all $i$-bands of

690: $G_{\a}$ are of the same type.

691: \endproclaim

692: 

693: \proof Assuming otherwise, then there are two $i$-bands $B_i^1, B_i^2$

694: of different types for each $i$.  Since the graph $G_{\b}$ is special,

695: there are only two families of parallel boundary edges for each vertex

696: $v_i$ in $G_{\b}$, so each family contains the initial edge of some

697: $B_i^j$.  Therefore, the terminal edge of each $B_i^j$ is parallel to

698: the initial edge of some $B_{i+1}^k$, so there is a band $D_i^j$ on

699: $F_{\b}$ connecting these two edges.  Note that $D_i^j$ degenerates to

700: a single edge if these two edges coincide.  

701: 

702: Consider the 2-complex $Q = \cup (B_i^j \cup D_i^j)$.  Then $Q \cap

703: T_0 = \cup (e_i^j \cup d_i^j)$ is a graph $G$ on $T_0$, where $e_i^j =

704: B_i^j \cap T_0$ and $d_i^j = D_i^j \cap T_0$.  We have $Q \cong G

705: \times I$.  Shrinking each $d_i^j$ to a point, and orienting $e_i^j$

706: so that its endpoint is on $d_i^j$, we get an oriented graph $G'$ in

707: which each vertex $d_i^j$ is the tail of some edge $e_{i+1}^k$.  Hence

708: $G'$ contains an embedded oriented cycle.  The corresponding cycle $C$

709: in $G$ is then an embedded loop in $T_0$.  Let $\gamma$ be a parallel

710: copy of some boundary component of $F_{\b}$ on $T_0$, intersecting

711: some $e_i^j$ in $C$ transversely at a single point.  The definition of

712: $B_i^j$ and the orientation of $e_i^j$ implies that $C$ intersects

713: $\gamma$ always in the same direction; hence $C$ is an essential

714: curve.  Thus $A = C\times I \subset Q$ is an annulus properly embedded

715: in $M$ intersecting $T_0$ in the essential curve $C$, which

716: contradicts the assumption that $M$ is $\bdd$-irreducible and

717: anannular.  \endproof

718: 

719: \proclaim{Lemma 4.3}  Each family of boundary edges in $G_{\a}$

720: contains at least $n$ edges.  \endproclaim

721: 

722: \proof Let $E_1, ..., E_4$ be the four edges of $\hat G_{\a}$ at

723: $u_1$, with $E_1, E_3$ the boundary edges.  If $|E_1| < n$ then there

724: is a label $i$ which does not appear at the endpoints of edges in

725: $E_1$.  If $\Delta = 5$ then we would have $|E_3| = 5n - |E_1| - |E_2|

726: - |E_4| \geq 2n + 1$, contradicting Lemma 2.6(2).  Hence $\Delta = 4$.

727: Since $|E_3| \leq 2n$, $E_3$ contains at most two $i$-edges, so each

728: of $E_2, E_4$ contains one $i$-edge.

729: Let $e_1, e_2$ be the $i$-edges of $E_2, E_4$ at $u_1$, and let $e_3,

730: e_4$ be the $i$-edges of $E_3$.  Since $e_3, e_4$ are adjacent

731: $i$-edges at $u_1$, by Lemma 2.5(1) they are adjacent $1$-edges at

732: $v_i$.  On $G_{\b}$ the two edges $e_3, e_4$ belong to different

733: families of boundary edges at $v_i$, because they cannot be parallel

734: on both graphs.  Therefore the two edges $e_1, e_2$ belong to the same

735: family of interior edges.  Since they both have label $1$ at $v_i$,

736: this would imply that the interior family containing them has

737: at least $n_{\a} + 1$ edges, contradicting Lemma 2.2(3).  \endproof

738: 

739: \proclaim{Lemma 4.4}  The jumping number $q = 1$.

740: \endproclaim

741: 

742: \proof This is automatically true if $\Delta = 4$.  Hence assume

743: $\Delta = 5$.  First assume that there is a vertex $u_j$ of $G_{\a}$

744: which has two interior $i$-edges $e_1, e_2$ for some $i$.  Since each

745: interior family contains at most $n$ edges, $e_1, e_2$ are nonparallel

746: on $G_{\a}$.  By Lemma 4.3 each boundary family contains an $i$-edge,

747: hence $e_1, e_2$ are non adjacent among the $i$-edges at $u_j$.

748: Dually on $G_{\b}$ these are $j$-edges at the vertex $v_i$.  For the

749: same reason, they are non adjacent among all $j$-edges at $v_i$.

750: Therefore by Lemma 2.5 the jumping number $q =1$.

751: 

752: Now assume that $u_j$ has at most one interior $i$-edge for all $i$.

753: Then it has at most $n$ interior edge endpoints.  On the other hand,

754: since each boundary family contains at most $2n$ edges and the valency

755: of $u_j$ is $\Delta n = 5n$, we see that it cannot have less than $n$

756: interior edge endpoints; therefore it has exactly $n$ interior edge

757: endpoints, and each boundary family contains exactly $2n$ edges.  If

758: $u_j$ has two interior families, so each family contains less than $n$

759: edges, then the two boundary families have different ending labels.

760: In this case for each label $i$ there are three $i$-bands, which

761: cannot all be of the same type because each boundary family of $v_i$

762: has at most two $j$-edges.  This contradicts Lemma 4.2.  Therefore

763: $u_j$ has only one family of interior edges, which contains $n$ edges.

764: For the same reason, each vertex of $G_{\b}$ has only one family of

765: interior edges, containing $n_{\a}$ edges.  By the parity rule one of

766: these families is negative, and by Lemma 2.1(3) they form cycles on

767: the other graph, so each vertex of that graph would then have two

768: families of interior edges, contradicting the above conclusion.

769: \endproof

770: 

771: 

772: \proclaim{Lemma 4.5}  Suppose all $i$-bands at a vertex $u_j$ of

773: $G_{\a}$ are of the same type.  Then 

774: 

775: (1) there are $n$ parallel interior edges at $u_j$, and

776: 

777: (2) each family of $n$ parallel interior edges at $u_j$ has $i$

778: as its ending label.  \endproclaim

779: 

780: \proof Let $E_1, ..., E_4$ be the edges at $u_j$ of $\hat G_1$,

781: appearing in this order around $\bdd u_j$ along its preferred

782: orientation, with $E_1, E_3$ the boundary edges.  Let $e_1, ..., e_4$

783: be four $i$-edges at $u_j$, appearing successively along the preferred

784: orientation of $\bdd u_j$.

785: 

786: First assume that all $e_i$ are boundary edges.  Then we may assume

787: that $e_1, e_2 \in E_1$, and $e_3, e_4 \in E_3$.  Thus $e_1, e_3$ are

788: not ending edges, so they are initial edges of some $i$-bands $B_1,

789: B_3$.  Since the jumping number $q=1$ (Lemma 4.4), and since $e_1,

790: e_3$ are non adjacent among $i$-edges at $u_j$, by Lemma 2.5 they are

791: non adjacent among $1$-edges at $v_i$ in $G_{\b}$, hence they are non

792: parallel boundary edges on $G_{\b}$.  Therefore $B_1, B_3$ are of

793: different type.

794: 

795: Now assume that $E_2$ contains an $i$-edge $e_2$, say.  Since each of

796: $E_1, E_3$ contains at least $n$ edges, we must have $e_1 \in E_1$ and

797: $e_3 \in E_3$.  Assume that either $e_2$ is not the ending edge of

798: $E_2$ or $|E_2| < n$.  Then $e_1$ is not an ending edge of $E_1$, and

799: there is an $i$-band $B_1$ with $e_1$ as the initial edge.  If $e_3$

800: is not an ending edge either, then there is an $i$-band $B_3$ with

801: $e_3$ as initial edge.  For the same reason as above, $B_1, B_3$ are

802: of different type, and we are done.  So assume that $e_3$ is the

803: ending edge of $E_3$.  Now we must have $|E_4| < n$ as otherwise $E_4$

804: would have $n$ edges and have the $i$-edge $e_4$ as its ending edge,

805: contradicting the assumption.  Hence $e_4$ is in $E_1$, and so there

806: is an $i$-band with $e_4$ as an initial edge.  Since $e_1, e_4$ are

807: parallel on $G_{\a}$, they are nonparallel on $G_{\b}$, so again $B_1,

808: B_4$ are of different type.  This proves (1).  To prove (2), notice

809: that if $|E_2| = n$ but $e_2$ is not the ending edge, then $e_1, e_3$

810: are not ending edges of $E_1, E_3$, so from the above the two $i$-bands

811: $B_1, B_3$ are of different type.  \endproof

812: 

813: \proclaim{Lemma 4.6} Suppose $n > 2$.  Then each positive edge of

814: $\hat G_{\a}$ represents at most $n/2$ edges.  \endproclaim

815: 

816: \proof When $n > 2$, the special graph $\hat G_{\b}$ has at most

817: one edge connecting any two vertices.  If $G_{\a}$ has $n/2+1$

818: parallel positive edges, then it has a Scharlemann cycle $e_1\cup e_2$

819: with label pair $(1,2)$, say.  So the two edges $e_1, e_2$ would be

820: parallel on $G_{\b}$, contradicting Lemma 2.1(4).  \endproof

821: 

822: \proclaim{Proposition 4.7} If $G_{\a}$ is special then up to

823: relabeling we have $n_1=1$, $n_2 = 2$, and $G_1$ has exactly two

824: interior edges.  \endproclaim

825: 

826: \proof First assume $n_{\a}\geq 2$ for $\a = 1,2$.  By Lemma 4.2 for

827: each graph $G_{\a}$ there is a label $i$ such that all $i$-bands of

828: $G_{\a}$ are of the same type.  Let $u_j$ be a vertex of $G_{\a}$.  By

829: Lemma 4.5(1), it has a set of $n$ parallel interior edges $E$ with $i$

830: as its ending label at $u_j$.  Let $u_k$ be the vertex on the other

831: endpoint of $E$, then by Lemma 4.5(2), $E$ also has ending label $i$

832: at $u_k$.  If $E$ is negative, then the ending edge $e$ of $E$ at

833: $u_j$ is the same as that at $u_k$, so $e$ would have the same label

834: $i$ on its two endpoints, and hence is a loop on $G_{\b}$.  Since $n

835: \geq 2$ and $G_{\b}$ is special, this is absurd.  If $E$ is positive,

836: then the two ending edges would give rise to two negative edges at

837: $v_i$ in $G_{\b}$, which must be nonparallel because they cannot be

838: parallel on both graphs.  Thus both families of interior edges at

839: $v_i$ are negative.  Replacing $u_j$ by $v_i$ in the above argument,

840: we get a contradiction because now $E$ must be negative.  Therefore up

841: to relabeling we must have $n_1 = 1$.

842: 

843: By Lemma 4.5(1) the only vertex $u_1$ of $G_1$ has $n_2$ parallel

844: interior edges, which by the parity rule must be negative edges on

845: $G_2$, hence $n_2 \geq 2$.  If $n_2 > 2$, then by Lemma 4.6 $G_1$ has

846: at most $n_2/2$ interior edges, which is a contradiction.  Hence the

847: result follows.  \endproof

848: 

849: 

850: \head  \S 5.  The generic case

851: \endhead

852: 

853: In this section except for Lemma 5.1, we assume $n_{\a}, n_{\b} > 2$.

854: By Proposition 4.7, $G_{\a}$ and $G_{\b}$ are not special.  Again

855: denote $n_{\b}$ by $n$.

856: 

857: \proclaim{Lemma 5.1} Suppose $n_{\a} \geq 2$, $n > 2$, and suppose

858: $\hat G_{\a}$ has a negative edge $E$ with $|E| = n$, and a positive

859: edge $E'$ with $|E'| = n/2+1$.  Let $(1,2)$ be the label pair of the

860: Scharlemann cycle in $E'$.  Then

861: 

862: (1) $G_{\b}$ has at most $n/2$ boundary vertices; 

863: 

864: (2) when $n=4$, the two vertices $v_3, v_4$ of $G_{\b}$ cannot

865: both be boundary vertices; and

866: 

867: (3) when $n=4$, $G_{\a}$ cannot have both a $(1,4)$-edge and a

868: $(2,3)$-edge.  \endproclaim

869: 

870: \proof Let $k$ be the number of $E$-orbits.  Since $E'$ contains more

871: than $n/2$ edges, hence contains a Scharlemann cycle, the annulus

872: $A_{\b}$ is separating, so $G_{\b}$ has the same number of positive

873: and negative vertices.  Each $E$-orbit contains the same number

874: ($n/k$) of vertices, all of the same sign, so the number of orbits

875: containing positive vertices is the same as the number of those

876: containing negative ones, and hence $k$ must be even.  Recall that

877: each $E$-orbit forms an essential cycle on $G_{\b}$, so only the

878: vertices on the two cycles adjacent to the two boundary components of

879: $A_{\b}$ could be boundary vertices.  Hence the number of boundary

880: vertices is at most $2(n/k)$, and since $k$ is even, (1) follows 

881: unless $k = 2$.

882: 

883: Assume $k=2$.  Let $C_1, C_2$ be the two cycles of $E$-orbits on

884: $G_{\b}$, and let $e_1, e_2$ be the edges of the Scharlemann cycle in

885: $E'$.  By Lemma 2.1(4) $e_1\cup e_2$ is an essential cycle on

886: $G_{\b}$.  The two vertices $v_1, v_2$ of $e_1\cup e_2$ are on

887: different $C_1, C_2$ because they are antiparallel, so the cycle

888: $e_1\cup e_2$ lies between $C_1$ and $C_2$, separating the vertex

889: $v_3$ on the first orbit from the vertex $v_n$ on the second.  On the

890: other hand, since $E'$ contains more than two edges, there is an edge

891: adjacent to the Scharlemann cycle which is a $(3, n)$-edge, so on

892: $G_{\b}$ there would be an edge connecting $v_3$ to $v_n$.  This is a

893: contradiction, showing that $k=2$ is impossible.  In particular, this

894: proves (1).

895: 

896: Now assume $n=4$.  Since we have shown that $k$ is even and $k\neq 2$,

897: we must have $k=4$.  In this case each vertex $v_i$ of $G_{\b}$ has an

898: essential loop $C_i$ coming from the $n$ parallel negative edges in

899: $G_{\a}$.  These loops and their vertices form essential circles on

900: $A_{\b}$ which are parallel to each other.  As above, there is an edge

901: in $E'$ which connects $v_3$ to $v_4$.  Hence the circles $C_3$ and

902: $C_4$ are adjacent to each other, so $v_3, v_4$ cannot both be

903: boundary vertices.  This proves (2).  Since the edges in the

904: Scharlemann cycle connect $v_1$ to $v_2$, $C_1$ is adjacent to $C_2$.

905: Thus either $C_3$ separates $v_4$ from $v_1, v_2$, so there is no edge

906: connecting $v_4$ to $v_1$, or $C_4$ separates $v_3$ from $v_1,v_2$, so

907: there is no edge connecting $v_3$ to $v_2$.  This proves (3).

908: \endproof

909: 

910: \proclaim{Lemma 5.2} Suppose $E_1, ..., E_5$ are the edges of $\hat

911: G_{\a}$ at a vertex $u$ of valency 5.  If $E_1, E_2, E_3$ are

912: positive, and $E_4$ is an interior edge, then $|E_5| > n$; in

913: particular, $E_5$ is a boundary edge.  \endproclaim

914: 

915: \proof Assume $|E_5| \leq n$.  By Lemma 2.2(2) we have $|E_i| \leq n/2

916: + 1$ for $i \leq 3$, and by Lemma 2.2(3) $|E_4| \leq n$.  Since $n>2$,

917: we must have $\Delta = 4$, and

918: $$ 4n = \Delta n = |E_1| + \cdots + |E_5| \leq 3 (\frac {n}{2} + 1) +

919: 2n = \frac 72 n + 3,$$ which implies that $n \leq 6$.  Moreover, $n$

920: must be even, otherwise by Lemmas 2.2(2) and 2.1(4) we would have

921: $|E_i| \leq n/2$ for $i=1,2,3$, hence $ 4n \leq 3 (n/2)+ 2n $, which

922: is absurd.

923: 

924: If $n = 6$, then all the above inequalities are equalities.  In

925: particular, $\Delta = 4$, $|E_4|=|E_5|= 6$, and $|E_i| = 4$ for $i =

926: 1, 2,3$, so each of $E_1, E_2, E_3$ contains a Scharlemann cycle, and

927: by Lemma 2.2(1) they all have the same label pair, say $(1,2)$.

928: But since $|E_4|= | E_5| = n$, these labels also appear in $E_4$ and

929: $E_5$.  Thus the label $1$ appears 5 times, contradicting the fact

930: that $\Delta = 4$.

931: 

932: Now assume $n = 4$.  If each of $E_1, E_2, E_3$ contains a Scharlemann

933: cycle with label pair $(1,2)$, say, (in particular, if $|E_i| = 3$

934: for $i=1,2,3$), then again the labels $\{1,2\}$ appear three times

935: among the endpoints of $E_1\cup E_2 \cup E_3$ at $u$.  Also, since

936: $E_4\cup E_5$ has at least $16 - 3\times 3 = 7$ edge endpoints at $u$,

937: one of the labels $\{1,2\}$ appears at least twice among the endpoints

938: of $E_4\cup E_5$ at $u$, so it appears 5 times at $u$, contradicting

939: the fact that $\Delta = 4$.  Hence we may assume that $|E_1| = |E_2| =

940: 3$, $|E_3| = 2$, $E_3$ contains no Scharlemann cycle, and $|E_4|=

941: |E_5| = 4$.  Since the two edges of $E_3$ have labels $3, 4$ at $u$,

942: they must have label sets $\{1,4\}$ and $\{2,3\}$.  Since $|E_4| = 4$,

943: the edges in $E_4$ are negative.  This contradicts Lemma 5.1(3),

944: completing the proof of the lemma.  \endproof

945: 

946: \proclaim{Lemma 5.3}  $\hat G_{\a}$ has no interior vertex of valency

947: at most $5$.

948: \endproclaim

949: 

950: \proof Let $E_1, \ldots, E_5$ be the edges of $\hat G_{\a}$ incident

951: to $u$.  Since all these edges are interior edges, by Lemma 5.2 they

952: can have at most two positive edges, say $E_1, E_2$.  By Lemma

953: 2.3(2), $u$ has at most $2n$ negative edges in $G_{\a}$, hence $E_1

954: \cup E_2$ represents at least $2n$ positive edges.  By Lemma 2.2(2)

955: we have $2n \leq 2(n/2+1)$, which contradicts the assumption that $n

956: \geq 3$.  \endproof

957: 

958: 

959: \proclaim{Lemma 5.4} $\hat G_{\a}$ cannot have a boundary vertex $u$

960: of valency at most 4 with a single boundary edge.  \endproclaim

961: 

962: \proof Let $E_0$ be the boundary edge, and $E_1, E_2, E_3$ the

963: interior edges of $\hat G_{\a}$ at $u$.  By Lemma 2.6(2) and

964: Proposition 4.7 we have $|E_0| < 2n$.  By Lemma 2.3(2), $u$ can have

965: at most $2n$ negative edges in $G_{\a}$, so one of the interior edges,

966: say $E_1$, must be positive, and by Lemma 2.2(2) $|E_1| \leq n/2+1 <

967: n$.  Since each of $E_2, E_3$ represents at most $n$ edges, we have

968: $|E_0| > n$.

969: 

970: We claim that {\it either $G_{\a}$ or $G_{\b}$ contains a Scharlemann

971: cycle.}  If $u$ has more than $n$ negative edges, then by Lemma

972: 2.3(1)  $G_{\b}$ contains a Scharlemann cycle.  So assume $u$ has at

973: most $n$ negative edges.  Since $u$ has less than $2n$ boundary edges,

974: it must have more than $n$ positive edges.  If at most two of the

975: $E_1, E_2, E_3$ are positive, then one of them represents more than

976: $n/2$ positive edges; if all the three interior edges at $u$ are

977: positive, then since they represent more than $2n$ edges, again one

978: of them represents more than $n/2$ edges.  In either case these

979: parallel edges contain a Scharlemann cycle.  This completes the proof

980: of the claim.

981: 

982: Now $|E_0| > n$ implies that some vertex of $G_{\b}$ has two

983: nonparallel boundary edges.  In particular, $\hat G_{\b}$ cannot be

984: triangular.  It follows from Lemma 5.3 and Proposition 3.1 that $\hat

985: G_{\b}$ must also have a boundary vertex $v$ of valency at most 4 with

986: a single boundary edge.  Since one of $G_{\a}$ and $ G_{\b}$ has a

987: Scharlemann cycle, by considering $v$ instead of $u$ if necessary, we

988: may assume without loss of generality that $G_{\a}$ contains a

989: Scharlemann cycle with label pair $(1, 2)$.

990: 

991: We claim that $|E_0| \leq n+2$.  Otherwise each vertex of $G_{\b}$ has

992: a boundary edge, and some vertex $v_i$ other than $v_1, v_2$ has two

993: such edges $e_1, e_2$.  Since the edges of the Scharlemann cycle form

994: an essential subgraph of $G_{\b}$ (Lemma 2.1(4)), separating the two

995: boundary components of $A_{\b}$, the edges $e_1, e_2$ must go to the

996: same boundary component.  Applying Lemma 2.7, we see that $G_{\a}$ is

997: special, a contradiction.

998: 

999: Since $|E_0| > n$ and $u$ has some positive edges, by Lemma 2.6(3) the

1000: graph $G_{\a}$ cannot have $n$ parallel negative edges.  Thus if $k$

1001: of the $E_1, E_2, E_3$ are positive, then

1002: $$ 4n \leq (n+2) + k (\frac n2 +1) + (3-k)(n-1) = (4-\frac k2)n + (2k

1003: -1)$$ which implies that $n < 4$.  But since $G_{\a}$ contains a

1004: Scharlemann cycle, $n$ is even.  This contradicts the assumption that

1005: $n > 2$.  \endproof

1006: 

1007: 

1008: \proclaim{Lemma 5.5} If both $\hat G_1, \hat G_2$ are triangular, then

1009: each boundary vertex has exactly two positive and two negative edges

1010: in $\hat G_{\a}$.  \endproclaim

1011: 

1012: \proof Let $E_0, ..., E_4$ be the edges of $\hat G_{\a}$ at a boundary

1013: vertex $v$, with $E_0$ the boundary edge.  Since $\hat G_{\b}$ is also

1014: triangular, $E_0$ represents at most $n$ edges.  Therefore by Lemma

1015: 5.2 at most two of the $E_i$ are positive.  On the other hand, by

1016: Lemma 2.3(2) $v$ has at most $2n$ negative edges, hence at least $n$

1017: positive edges.  Since each $E_i$ represents at most $n/2 + 1 < n$

1018: positive edges, $v$ must have two positive edges in $\hat G_{\a}$.

1019: \endproof

1020: 

1021: \proclaim{Lemma 5.6} Suppose both $\hat G_1, \hat G_2$ are triangular.

1022: Then all vertices of $G_1, G_2$ are boundary vertices.  \endproclaim

1023: 

1024: \proof Let $E_0, ..., E_4$ be the edges of $\hat G_{\a}$ at a boundary

1025: vertex $v$, with $E_0$ the boundary edge.  By Lemma 5.5 we may

1026: assume that $E_1, E_2$ are negative edges, and $E_3,E_4$ are positive

1027: edges.

1028: 

1029: Suppose $G_{\b}$ has some interior vertices.  Then $|E_0| < n$.  Since

1030: $v$ has at most $2n$ negative edges, it has more than $n$ positive

1031: edges, so $E_3\cup E_4$ contains a Scharlemann cycle with label pair

1032: $(1,2)$, say, and $n$ is even.  Also, one of $E_1, E_2$ must

1033: represent $n$ parallel edges, for otherwise $E_0\cup E_1 \cup E_2$

1034: would contain at most $3n-3$ edges, so one of $E_3, E_4$ would

1035: contain at least $n/2+2$ edges, contradicting Lemma 2.2(2).  Now we

1036: can apply Lemma 5.1(1) and conclude that $G_{\b}$ has at most $n/2$

1037: boundary vertices.  Thus $|E_0| \leq n/2$.  We have the inequality

1038: $$ 4n \leq |E_0| + ... + |E_5| \leq \frac n2 + 2n + 2(\frac n2 + 1) \leq

1039: \frac 72 n + 2.$$ Since $n$ is even, this implies $n=4$, $|E_0| = 2$,

1040: $|E_1|=E_2|=4$, and $|E_3| = |E_4| = 3$.  Now each of $E_3, E_4$

1041: contains a Scharlemann cycle on label pair $(1,2)$, so these labels

1042: appear 4 times among the interior edge endpoints at $v$.  Thus the

1043: labels of $E_0$ must be $3,4$.  This contradicts Lemma 5.1(2).

1044: \endproof

1045: 

1046: \proclaim{Lemma 5.7}  $\hat G_1, \hat G_2$ cannot both be triangular.

1047: \endproclaim

1048: 

1049: \proof Assume $\hat G_1, \hat G_2$ are triangular.  Then by Lemma 5.6

1050: all vertices of $G_1, G_2$ are boundary vertices, and by Lemma 5.5

1051: each vertex $v$ of $G_{\a}$ has exactly two positive edges and two

1052: negative edges in $\hat G_{\a}$.  Since a positive edge in $G_{\a}$ is

1053: a negative edge in $G_{\b}$, it follows that either (i) some vertex

1054: $v$ of one of the graphs, say $G_1$, has more positive edge endpoints

1055: than negative ones, or (ii) all vertices of $G_1$ and $G_2$ have the

1056: same number of positive and negative edge endpoints.

1057: 

1058: In case (i), (writing $n = n_2$), $v$ has at most $2(n/2+1) = n + 2$

1059: positive edges, at most $n + 1$ negative edges, and at most $n$

1060: boundary edges.  From the inequality

1061: $$4n \leq (n+2) + (n+1) + n$$ we see that $n\leq 3$.  But if $n=3$

1062: then $v$ has at most $2(n/2) = n$ positive edges, at most $n-1$

1063: negative edges, and at most $n$ boundary edges, which would lead to the

1064: contradiction that $4n \leq n + (n-1) + n$.

1065: 

1066: In case (ii), any vertex $v$ of $G_{\a}$ has at most $n+2$ positive

1067: edges, the same number of negative edges, and at most $n$ boundary

1068: edges; so from $4n \leq (n+2) + (n+2) + n$ we see that $n=4$, $|E| =

1069: 3$ for all positive interior edges of $\hat G_{\a}$, and $|E| = 4$ for

1070: all boundary edges of $G_{\a}$.  Each label appears three times among

1071: the interior edges endpoints at any vertex $v$ of $G_{\a}$, but since

1072: each of the two families of positive edges at $v$ contains a

1073: Scharlemann cycle, which must all have the same label pair $(1,2)$,

1074: it follows that these labels appear only once among the negative edge

1075: endpoints at $v$, so the label $3$ appears twice among the negative

1076: edge endpoints at $v$.  Since this is true for all vertices $v$ in

1077: $G_{\a}$, it means that the vertex $v_3$ on $G_{\b}$ has $2n_{\a}$

1078: positive edge endpoints, and $n_{\a}$ negative ones, a contradiction.

1079: \endproof

1080: 

1081: \proclaim{Proposition 5.8} One of the graphs $G_{\a}$ has at most two

1082: vertices.  \endproclaim

1083: 

1084: \proof Assume $n_1, n_2 \geq 3$.  By Proposition 4.7, $G_{\a}$ is not

1085: special, by Lemma 5.3 $\hat G_{\a}$ does not have an interior vertex

1086: of valency at most 5, and by Lemma 5.4 it cannot have a boundary

1087: vertex of valency at most 4 with a single boundary edge.  Thus by

1088: Proposition 3.1 both $\hat G_1, \hat G_2$ are triangular, which

1089: contradicts Lemma 5.7.  \endproof

1090: 

1091: 

1092: 

1093: \head  \S 6. Nonspecial graphs with $n_1 = 2$ and $n_2 > 2$

1094: \endhead

1095: 

1096: Throughout this section we will assume that $G_1, G_2$ are not special

1097: graphs.  We will show that the case $n_1=2$ and $n_2 = n > 2$ does not

1098: happen.  Together with Propositions 4.7 and 5.8, this shows that

1099: $n_{\a}$ must be at most 2 for both $\a = 1$ and $2$.

1100: 

1101: \proclaim{Lemma 6.1} If $n_1 = 2$ then $\hat G_1$ is a subgraph of

1102: that shown in Figure 6.1.  \endproclaim

1103: 

1104: \proof Since $\hat G_1$ is not a special graph, one of the vertices

1105: $u_1, u_2$ has at most one boundary edge.  If either $u_1$ or $u_2$

1106: does not have a loop, then one can find a vertex $u$ of valency at

1107: most 3 in $\hat G_1$, with at most one boundary edge.  Since each

1108: interior edge represents at most $n$ edges, $u$ would have at least

1109: $2n$ boundary edges, which would imply that $G_2$ is a special graph,

1110: a contradiction.  Hence each vertex $u_i$ has a loop.  It is now easy

1111: to see that $\hat G_1$ must be a subgraph of that in Figure 6.1.

1112: \endproof

1113: 

1114: \bigskip

1115: \leavevmode

1116: 

1117: %\epsfxsize=5in      % adjust the width  of the figure

1118: %\epsfysize=3in      % adjust the height of the figure

1119: \centerline{\epsfbox{Fig6.1.ps}} 

1120: \bigskip

1121: \centerline{Figure 6.1}

1122: \bigskip

1123: 

1124: Label the edges of $\hat G_1$ as in Figure 6.1.  Denote by $m$ the

1125: number of non-loop interior edges of $G_1$, i.e.\ $m = |E_3| + |E_4|$.

1126: 

1127: 

1128: \proclaim{Lemma 6.2}  Suppose $n_1 = 2$, and $n > 2$.

1129: 

1130: (1) Either $m = 2n$, or $m = 2n -2$ and $E_2$

1131: contains a Scharlemann cycle.  

1132: 

1133: (2) The two vertices of $G_1$ are antiparallel.  

1134: \endproclaim

1135: 

1136: \proof (1) If no label appears twice among the endpoints of edges in

1137: $E_2$, then from the labeling on $\bdd u_1$ one can see that either $m

1138: \geq 2n$ or $|E_1| \geq 2n$.  But the second possibility does not

1139: occur because then by Lemma 2.6(2) the graph $G_2$ would be special.

1140: Hence in this case we have $m\geq 2n$.  Since each of $E_3, E_4$

1141: represents at most $n$ edges, we conclude that $m = 2n$.

1142: 

1143: Now assume that some label appears twice among the endpoints of edges

1144: in $E_2$.  Then $E_2$ contains a Scharlemann cycle $e_1, e_2$, with

1145: label pair $(1,2)$, say.  Since $n > 2$, $E_2$ contains no extended

1146: Scharlemann cycle (Lemma 2.1(5)), so one of these two edges, say

1147: $e_1$, must be an outermost edge among those in $E_2$.  Thus the

1148: endpoints of $e_1$ are either adjacent to those in $E_3\cup E_4$ or to

1149: those in $E_1$.  In the first case, the label sequence of $E_3\cup

1150: E_4$ at $u_1$ is $3, 4, ..., n$, so $m \equiv n-2$ mod $n$.  If $m =

1151: 2n-2$ then we are done.  If $m\neq 2n-2$, then since $|E_3|, |E_4|

1152: \leq n$, we must have $m = n-2$.  Thus $|E_1| = \Delta n - m - 2|E_2|

1153: \geq 2n$, which by Lemma 2.6(2) would imply that $G_2$ is special, a

1154: contradiction.  Therefore $e_1$ must be adjacent to $E_1$.  As above,

1155: we have either $|E_1| = 2n-2$, or $|E_1| = n-2$ and $m \geq 2n$.  In

1156: the second case we have $m = 2n$ because $|E_3|, |E_4| \leq n$.  It

1157: remains to show that $|E_1| = 2n-2$ is impossible.

1158: 

1159: Assume $|E_1| = 2n-2$.  Notice that this happens only if $E_2$

1160: contains a Scharlemann cycle.  Moreover, if $(1,2)$ is the label pair

1161: of the Scharlemann cycle then all labels other than $1,2$ would appear

1162: twice among endpoints of edges in $E_1$.  Thus on $G_2$ each vertex

1163: other than $v_1, v_2$ would have two boundary $1$-edges.  But since

1164: the edges in the Scharlemann cycle and the vertices $v_1, v_2$ form an

1165: essential subgraph of $G_2$, these two parallel $1$-edges must go to

1166: the same boundary component of $A_2$.  By looking at an outermost

1167: vertex one can see that there is a vertex $v_i$ with $i \neq 1,2$, at

1168: which the two boundary $1$-edges are parallel, so they are parallel on

1169: both graphs, contradicting Lemma 2.1(2).  

1170: 

1171: (2) If $u_1, u_2$ are parallel then $2n-2 \leq |E_3| + |E_4| \leq

1172: 2(n/2+1)$, implying that $n=4$ and $|E_3| = |E_4| = 3$.  In this case

1173: both $E_3, E_4$ contain Scharlemann cycles, and by Lemma 2.2(1) they

1174: must have the same label pair $(1,2)$ as the one in $E_2$.  But since

1175: each of the labels $1,2$ appears only once among the endpoints at

1176: $u_1$ of edges in $E_3\cup E_4$, this is impossible.  \endproof

1177: 

1178: 

1179: \proclaim{Lemma 6.3} Suppose $n_1 = 2$, and $n > 2$.  Then $G_1$

1180: cannot have $2n$ negative edges.  \endproclaim

1181: 

1182: \proof We must have $|E_2| > 0$, otherwise $|E_1| \geq 2n$, so $G_2$

1183: would be special, contradicting our assumption.  Assume that $G_1$ has

1184: $2n$ negative edges.  Then $|E_3| = |E_4| = n$, and by Lemma 2.6(3) we

1185: have $|E_1| \leq n$, hence $|E_2| \geq n/2$.  On the other hand, by

1186: Lemma 2.2(2) $|E_2|\leq n/2+1$.  Hence $E_2$ contains either $n/2$ or

1187: $n/2+1$ edges.  We want to show that $|E_2| \neq n/2+1$.  Assuming

1188: otherwise, then since $E_3$ contains $n$ parallel negative edges, by

1189: Lemma 5.1 the graph $G_1$ has at most $n/2$ parallel boundary edges.

1190: On the other hand, we have $|E_1| = \Delta n - m - 2|E_2| \geq n - 2$,

1191: and since $E_2$ contains a Scharlemann cycle with label pair

1192: $(1,2)$, say, $n$ is even.  Therefore we must have $n=4$.  Now in

1193: this case the labels of the edges in $E_1$ are $3,4$, contradicting

1194: Lemma 5.1(2).

1195: 

1196: \bigskip

1197: \leavevmode

1198: 

1199: %\epsfxsize=5in      % adjust the width  of the figure

1200: %\epsfysize=3in      % adjust the height of the figure

1201: \centerline{\epsfbox{Fig6.2.ps}} 

1202: \bigskip

1203: \centerline{Figure 6.2}

1204: \bigskip

1205: 

1206: We have shown that $|E_2| = n/2$, and $|E_1| = n$.  For the same

1207: reason, we have $|E_5| = n/2$ and $|E_6| = n$.  Without loss of

1208: generality we may assume that $u_1$ is a positive vertex, $u_2$ is

1209: negative, and the edges of $E_1$ have label sequence $1, ..., n$ at

1210: $u_1$.  See Figure 6.2.  Let $t = n/2$.  Since $|E_2| = n/2$, the

1211: label sequence of the endpoints at $u_1$ of the edges of $E_3$ is

1212: $t+1, ..., n, 1, ..., t$.  There is a number $k$ such that the label

1213: sequence at the other end of $E_3$ is $t+k, t+k+1, ..., t+k -1$.  The

1214: number $k \neq 1$, otherwise these edges would be loops in $G_2$, so

1215: $n>2$ would imply that some vertex of $G_2$ does not have a boundary

1216: edge, contradicting the fact that $|E_1| = n$.  Now from Figure 6.2 we

1217: can see that the label sequence of $E_6$ is $k, ..., n, 1, ..., k-1$,

1218: hence the two edges $e_3, e_4$ in $E_6$ labeled $n$ and $1$

1219: respectively, are adjacent (because $k\neq 1$).  Let $e_1, e_2$ be the

1220: edges of $E_1$ labeled $1$ and $n$ respectively.  By Lemma 2.6(3),

1221: each vertex of $G_2$ has at most one family of parallel boundary

1222: edges, so $e_2$ is parallel to $e_3$, and $e_4$ parallel to $e_1$ in

1223: $G_2$.  Let $B(e_1, e_2)$ be the band on $F_1$ between $e_1$ and

1224: $e_2$, and let $B(e_3, e_4)$ be that between $e_3$ and $e_4$.

1225: Similarly, let $B(e_2, e_3)$ and $B(e_4, e_1)$ be the bands on $F_2$

1226: between $e_2, e_3$ and $e_4, e_1$, respectively.  Now we can form an

1227: annulus $A = B(e_1,e_2) \cup B(e_2,e_3) \cup B(e_3,e_4) \cup

1228: B(e_4,e_1)$ in the manifold $M$.  Since the boundary curve $C$ of $A$

1229: on $T_0$ intersects the circle $\bdd v_2$ transversely at a single

1230: point (on the arc $B(e_1, e_2) \cap \bdd u_1$), it is an essential

1231: curve.  This contradicts the fact that the manifold $M$ is

1232: $\bdd$-irreducible and anannular.

1233: \endproof

1234: 

1235: 

1236: \proclaim{Lemma 6.4} Suppose $n_1 = 2$, and $n > 2$.  Then $G_1$

1237: cannot have exactly $2n -2$ negative edges.  \endproclaim

1238: 

1239: \proof If $G_1$ has $2n - 2$ negative edges, then (up to symmetry)

1240: either $|E_3| = |E_4| = n-1$, or $|E_3| = n$ and $|E_4| = n-2$.

1241: Looking at the labeling, one can see that the two loops of $E_2$ near

1242: $E_3\cup E_4$ form a Scharlemann cycle, with label pair $(1,2)$, say.

1243: If $|E_3| =n$ then by Lemma 2.6(3) we have $|E_1| \leq n$, hence

1244: $|E_2| \geq n/2+1$.  Now by Lemma 5.1(1) the graph $G_2$ has at

1245: most $n/2$ boundary vertices, which contradicts the fact that $|E_1| =

1246: n$.  Therefore we must have $|E_3| = |E_4| = n-1$.

1247: 

1248: For the same reason, the two loops in $E_5$ near $E_3\cup E_4$ form a

1249: Scharlemann cycle, which by Lemma 2.2(1) must have the same label pair

1250: $(1,2)$.  Now we can see that $E_3$ has label sequence $3, 4, ...,

1251: n, 1, $ at $u_1$, and has label sequence $2, 3, ..., n$ at $u_2$.

1252: However, in this case $E_3$ has only one orbit, containing all the

1253: labels, so all the vertices of $G_2$ are parallel to each other, hence

1254: all edges of $G_1$ are negative.  But since $G_1$ contains some

1255: loops, this is a contradiction.  \endproof

1256: 

1257: \proclaim{Proposition 6.5}  If

1258: $M(r_1), M(r_2)$ are annular, and $\Delta \geq 4$, then $n_{\a} \leq

1259: 2$ for $\a = 1,2$.  

1260: \endproclaim

1261: 

1262: \proof By Proposition 4.7 this is true if one of the $G_{\a}$ is

1263: special.  By Proposition 5.8 one of the graphs, say $G_1$, has at most

1264: two vertices.  Since the two possibilities in Lemma 6.2(1) have been

1265: ruled out by Lemmas 6.3 and 6.4, the case $n_1\leq 2$ and $n_2 > 2$

1266: cannot happen.  \endproof

1267: 

1268: 

1269: 

1270: \head   \S 7.  Special graphs with $n_1=1$ and $n_2 = 2$

1271: \endhead

1272: 

1273: \proclaim{Proposition 7.1} If $G_{\a}$ is special, then $\Delta = 4$,

1274: up to relabeling $n_1 = 1$, $n_2 = 2$, and the manifold $M$ is the

1275: exterior of the Whitehead link.  \endproclaim

1276: 

1277: \proof By Lemma 4.1, both graphs must be special.  By Proposition 4.7,

1278: up to relabeling we must have $n_1=1$, $n_2 = 2$, and $G_1$ has

1279: exactly two interior edges $e_1,e_2$.

1280: 

1281: Assume $\Delta = 5$.  By Lemma 4.4 the jumping number $q=1$.  There is

1282: a pair of adjacent boundary $1$-edges $e_1, e_2$ at $v_1$ in $G_2$,

1283: which by Lemma 2.5(1) should also be adjacent at $u_1$ in $G_1$ among

1284: all $1$-edges; but since the two families of boundary edges at $u_1$

1285: are separated by two interior edges, $e_1, e_2$ must be in the same

1286: family, so they are parallel on both graphs, a contradiction.

1287: Therefore we must have $\Delta = 4$.

1288: 

1289: Now the Whitehead link exterior $W$ does admit two annular Dehn

1290: fillings $W(r_1), W(r_2)$ with $\Delta(r_1,r_2) =4$, $n_1 = 1$, and

1291: $n_2 = 2$, see [GW1, Theorem 7.5].  It remains to show that the

1292: manifold satisfying these conditions is unique.

1293: 

1294: Each vertex of $G_2$ has two boundary edges, which are nonparallel

1295: because $G_2$ is special.  Thus the graph $G_2$ must be as shown in

1296: Figure 7.1(b).  Similarly, since $G_1$ is special it has two

1297: families of parallel boundary edges.  The loops have different labels

1298: at their two endpoints, so each family of boundary edges of $G_1$

1299: contains an even number of edges.  Hence $G_1$ must be as shown in

1300: Figure 7.1(a).

1301: 

1302: 

1303: \bigskip

1304: \leavevmode

1305: 

1306: %\epsfxsize=5in      % adjust the width  of the figure

1307: %\epsfysize=3in      % adjust the height of the figure

1308: \centerline{\epsfbox{Fig7.1.ps}} 

1309: \bigskip

1310: \centerline{Figure 7.1}

1311: \bigskip

1312: 

1313: Label the six edges of $G_1$ as in the figure.  Orient $e_3, e_4$ so

1314: that on $G_1$ they have label $1$ at their tails.  Up to symmetry we

1315: may assume that the edge $e_1$ on $G_2$ is as shown in Figure 7.1(b).

1316: The label $1$ endpoints of edges $e_1, e_3, e_5, e_4$ appear

1317: successively on $\bdd u_1$, hence by Lemma 2.5(1) they also appear in

1318: this order on $\bdd v_1$ in $G_2$, so these edges must be as shown in

1319: the figure.  Similarly by looking at the label $2$ endpoints of $e_2,

1320: e_4, e_6, e_3$ one can determine the edges $e_4$ and $e_6$.  Therefore

1321: up to symmetry the graphs $G_{\a}$ are exactly as shown in the figure.

1322: We need to show that these graphs uniquely determine the manifold $M$.

1323: 

1324: Recall that $F_{\a}$ denotes the punctured annulus $A_{\a} \cap M$.

1325: Let $X = N(F_1 \cup T_0)$, and let $Y = N(F_1\cup F_2 \cup T_0)$,

1326: where the regular neighborhoods are taken in $M$.  The frontier of $X$

1327: in $M$, i.e.\ $X \cap \overline{M-X}$, is a surface $F$, which is a

1328: four punctured sphere.  Note that $Y$ is obtained from $X$ by adding

1329: regular neighborhoods of the faces of $G_2$.  Each of the four faces

1330: of $G_2$ is a disk $D_i$ with $\bdd D_i = c_i \cup c_i'$, where $c'_i$

1331: is an arc on $\bdd M$, and $c_i$ an arc on $F_1\cup T_0$.  Let $\tilde

1332: c_i$ be the arc $D_i \cap F$.  Then the frontier of $Y = X \cup (\cup

1333: N(D_i))$ in $M$ is a properly embedded surface $F'$, homeomorphic to

1334: the surface obtained by cutting $F$ along the arcs $\tilde c_i$.  Thus

1335: $Y$ and $X$ are homeomorphic, but they are embedded in $M$

1336: differently.  Note that $Y$ is uniquely determined by the graphs

1337: $G_1$ and $G_2$.

1338: 

1339: It is easy to see that all the $\tilde c_i$ are essential arcs on $F$.

1340: Since each boundary component of $F$ meets $\cup \tilde c_i$ twice,

1341: after cutting along all these $\tilde c_i$, the remnant, and hence

1342: $F'$, consists of either two disks, or two disks and an annulus.  In

1343: fact, by examining the graphs, one can see that $F'$ indeed consists

1344: of two disks and an annulus.  Since $M$ is irreducible and

1345: $\bdd$-irreducible, the disk components of $F'$ are boundary parallel.

1346: If the annular component $A$ of $F'$ is incompressible in $M$ then $A$

1347: is also boundary parallel because $M$ is anannular and irreducible, so

1348: $M$ would be homeomorphic to $Y$, which in turn is homeomorphic to

1349: $X$.  Let $C$ be an essential curve on $T_0$ disjoint from $\bdd F_1$.

1350: Then $C \times I$ in $T_0 \times I$ would be an essential annulus in

1351: $X$, contradicting the fact that $M$ is anannular.  Therefore $A$ must

1352: be compressible.  Let $D$ be a compressing disk of $A$ in $M$.  Then

1353: $D$ lies in either $Y$ or $M- \Int Y$.  We show that the first case is

1354: impossible.  

1355: 

1356: 

1357: First notice that the surface $F_1$ cuts $X$ into a manifold $F \times

1358: I$, in which both $F$ and the two copies of $F_1$ are incompressible.

1359: By an innermost circle argument one can show that $F$ is

1360: incompressible in $X$.  Under the homeomorphism $Y \cong X$, $A$ can

1361: be considered as a subsurface of $F$, hence $A$ is also incompressible

1362: unless the core of $A$ is a trivial curve on $F$.  On the other hand,

1363: notice that $A$ is a component of $\bdd Y - \Int F''$, where $F'' = Y

1364: \cap (\bdd M - T_0)$ is a neighborhood of $\bdd A_1 \cup \bdd A_2$ on

1365: $\bdd M$, which is connected.  Since $\bdd A \subset F''$, it follows

1366: that the core of $A$ is nonseparating on $\bdd Y$, hence it is

1367: nontrivial on $F$.  This completes the proof that $A$ is

1368: incompressible in $Y$.

1369: 

1370: Hence the compressing disk $D$ of $A$ lies in $M-\Int Y$.  Let $M'$ be

1371: the union of $Y$ and a regular neighborhood of $D$.  Then the frontier

1372: of $M'$ in $M$ is a set of disks, which must be boundary parallel

1373: because $M$ is irreducible and $\bdd$-irreducible.  Therefore $M'$ is

1374: homeomorphic to $M$.  It follows that $M$ is obtained from $Y$ by

1375: adding a 2-handle along the core of $A$, and hence is uniquely

1376: determined by the graphs $G_1$ and $G_2$.  \endproof

1377: 

1378: \head  \S 8.  Nonspecial graphs with $n_{\a} \leq 2$

1379: \endhead

1380: 

1381: First note that if $n_{\a} = 1$ and $G_{\a}$ is not special, then the

1382: unique vertex of $G_{\a}$ has valency at most $3$ in $\hat G_{\a}$,

1383: and hence by Lemma 2.2(3) $G_{\a}$ has at least $2n_{\b}$ parallel

1384: boundary edges.  By Lemma 2.6(2) this implies that $G_{\b}$, and

1385: therefore (by Lemma 4.1) $G_{\a}$, is special, a contradiction.  Hence

1386: if $G_1, G_2$ are not special and $n_1, n_2 \leq 2$, we must have $n_1

1387: = n_2 = 2$.

1388: 

1389: \proclaim{Lemma 8.1} Suppose that $n_1 = n_2 = 2$ and $G_1, G_2$ are

1390: not special.  Then for $\a = 1,2$, the two vertices of $G_{\a}$ are

1391: antiparallel, $\hat G_{\a}$ is a subgraph of the graph $\hat G$ in

1392: Figure 6.1, and one of the following holds.

1393: 

1394: (i) $\Delta = 4$, each interior edge of $\hat G_{\a}$ represents two

1395: edges of $G_{\a}$, and $G_{\a}$ has no boundary edges.

1396: 

1397: (ii) $\Delta = 5$, each edge of $\hat G_{\a}$ represents two edges of

1398: $G_{\a}$, and the jumping number $q=2$.  \endproclaim

1399: 

1400: \proof By Lemma 6.1, $G_{\a}$ is a subgraph of the graph $\hat G$

1401: shown in Figure 6.1.  Each vertex $v$ of $G_{\a}$ must have a loop,

1402: otherwise some vertex would have valency $3$ in $\hat G_{\a}$ with a

1403: single boundary edge, so by Lemma 2.6(2) $G_{\b}$ would be special,

1404: contradicting the assumption.

1405: Since a loop in $G_{\a}$ is a non-loop negative edge of $G_{\b}$, it

1406: follows that each graph $G_{\b}$ has some negative edges, hence the

1407: two vertices of $G_{\b}$ must be antiparallel, $\b=1,2$.  By Lemma

1408: 2.2(3) each interior edge of $\hat G_{\a}$ represents at most two

1409: edges of $G_{\a}$.  Similarly, each boundary edge of $\hat G_{\a}$

1410: also represents at most two edges of $G_{\a}$, by Lemma 2.1(2).

1411: 

1412: First assume $\Delta = 4$.  Notice that a vertex of $G_{\a}$ has

1413: either no boundary edge or two boundary edges, for if it has exactly

1414: one boundary edge then the loops based at that vertex would have the

1415: same label at their two endpoints, which contradicts the parity rule.

1416: Since two boundary edges at a vertex of $G_{\a}$ correspond to

1417: boundary edges at different vertices of $G_{\b}$, it follows that

1418: either both vertices of $G_{\a}$ have two boundary edges, or they both

1419: have no boundary edges.  The second possibility gives rise to

1420: conclusion (i) in the lemma.

1421: 

1422: Assume that each vertex of $G_{\a}$ has two boundary edges.  Then

1423: there are a total of 6 interior edges in each graph.  Note that an

1424: interior edge is a loop on $G_{\a}$ if and only if it is a non-loop on

1425: $G_{\b}$ because of the parity rule, hence one of the graphs, say

1426: $G_1$, has at least three loops.  Without loss of generality we may

1427: assume that there are two loops $e_1, e_2$ based at the vertex $u_1$.

1428: Consider their label $1$ endpoints.  Because there are two boundary

1429: edges at $u_1$, these two endpoints are non adjacent among all label

1430: $1$ endpoints at $u_1$.  Now look at the graph $G_2$.  By Lemma 2.5(1)

1431: $e_1, e_2$ are non adjacent $1$-edges at $v_1$ among all $1$-edges.

1432: However, since they are non-loops in $G_2$, they are contained in the

1433: two adjacent families $E_3, E_4$ in Figure 6.1.  Since $E_3\cup E_4$

1434: contains a total of at most four edges, $e_1, e_2$ are adjacent among

1435: all $1$-edges at $v_1$.  This contradiction completes the proof of the

1436: lemma for the case $\Delta = 4$.

1437: 

1438: Now assume $\Delta = 5$.  Since each vertex of $\hat G$ has valency 5,

1439: and since each edge of $\hat G_{\a}$ represents at most two edges of

1440: $G_{\a}$, $\Delta = 5$ implies that each edge of $\hat G_{\a}$

1441: represents exactly two edges.  By the same argument as above one can

1442: show that the jumping number $q$ cannot be $1$, so we are in case (ii).

1443: \endproof

1444: 

1445: \proclaim{Lemma 8.2}  There is a unique irreducible, 

1446: $\bdd$-irreducible, anannular manifold $M$ which

1447: admits two annular Dehn fillings $M(r_1), M(r_2)$ with $\Delta(r_1,

1448: r_2) = 5$.

1449: \endproclaim

1450: 

1451: \proof By Lemma 8.1, the graphs must be as shown in Figure 8.1.

1452: We first show that the edge correspondence and the labelings of the 

1453: vertices are unique up to symmetry.

1454: 

1455: Reflecting the annuli vertically and changing their orientations if

1456: necessary, we may assume that the vertices $u_1, v_1$ are positive,

1457: and the labeling of edge endpoints at $\bdd u_1, \bdd v_1$ are as

1458: shown.  Any non-loop edge has the same label on its two endpoints,

1459: because it is a loop edge on the other graph.  Thus the labeling on

1460: $\bdd u_2, \bdd v_2$ is determined by that on $\bdd u_1, \bdd v_1$,

1461: respectively.  Orient the edges so that a non-loop edge goes from

1462: $u_1$ to $u_2$ (resp.\ $v_1$ to $v_2$).  Then dually the orientation

1463: of a loop edge must go from label $1$ to label $2$.  Label the edges

1464: of $G_1$ as in Figure 8.1(a).

1465: 

1466: If $P_1, ..., P_5$ are the points of $u_1\cap v_1$, appearing in this

1467: order on $\bdd u_1$ along its orientation, then since the jumping

1468: number $q=2$, they appear in the order $P_1, P_3, P_5, P_2, P_4$ on

1469: $\bdd v_1$ either along or against the orientation of $\bdd v_1$.  In

1470: other words, along the orientation of $\bdd v_1$ they either appear in

1471: this order, or in the order $P_1, P_4, P_2, P_5, P_3$.  In the second

1472: case, write $(Q_1, Q_2, ..., Q_5 ) = (P_1, P_4, P_2, P_5, P_3)$; then

1473: $(P_1, ..., P_5) = (Q_1, Q_3, Q_5, Q_2, Q_4)$.  Hence by interchanging

1474: the roles of $G_1$ and $G_2$ if necessary, we may assume that the

1475: points appear as $(P_1, P_3, P_5, P_2, P_4)$ on $\bdd v_1$ along the

1476: orientation of $\bdd v_1$.

1477: 

1478: Now we can see that the labeling of the edges on $G_2$ is completely

1479: determined by that of $G_1$: The $1$-edges at $u_1$ appear in the

1480: order $a, c, e, k, d$ in the positive direction, so at $v_1$ they appear

1481: in the order $a, e, d, c, k$, where $a$ is the unique boundary edge at

1482: $v_1$ labeled $1$.  The order of the 2-edges at $u_1$ is $b,d,f,l,c$,

1483: so dually the $1$-edges at $v_2$ are in the order $b, f, c, d, l$.

1484: Similarly by looking at $u_2$ one can determine the labeling of the

1485: remaining edges in $G_2$.  See Figure 8.1(b).

1486: 

1487: \bigskip

1488: \leavevmode

1489: 

1490: %\epsfxsize=5in      % adjust the width  of the figure

1491: %\epsfysize=3in      % adjust the height of the figure

1492: \centerline{\epsfbox{Fig8.1.ps}} 

1493: \bigskip

1494: \centerline{Figure 8.1}

1495: \bigskip

1496: 

1497: It remains to show that the manifold $M$ is uniquely determined by

1498: these graphs.  As in the proof of Proposition 7.1, consider the

1499: submanifold $X = N(A_1 \cup J_1)$ of $M(r_1)$.  Since $J_1$ intersects

1500: $A_1$ in two meridian disks of opposite sign, the frontier $F$ of $X$

1501: consists of two components $F_b, F_w$, each being a twice punctured

1502: torus, called the {\it black surface\/} and the {\it white surface\/}

1503: respectively.  A face of $G_2$ is black or white according to whether

1504: it intersects the black surface or the white surface.  Note that each

1505: face of $G_2$ intersects $F$ in a circle or an arc, so it is either

1506: black or white, but not both.

1507: 

1508: Let $D_1$ be a face of $G_2$ bounded by a pair of parallel loops, and

1509: let $D_2$ be the triangular interior face of $G_2$ adjacent to $D_1$.

1510: Since they have an edge in common, they are of different colors, so we

1511: may assume that $D_1$ is black and $D_2$ is white.  The boundary of

1512: $D_1$ intersects a meridian of $J_1$ twice in the same direction,

1513: hence $\bdd D_1$ is a nonseparating curve on $F_b$.  After adding a

1514: neighborhood of $D_1$ to $X$, the black frontier is homeomorphic to

1515: the surface obtained by 2-surgery on $F_b$ along $\bdd D_1$, hence is

1516: an annulus $A_b$.  Since its boundary components are essential curves

1517: on $\bdd M$, and since $M$ is $\bdd$-irreducible, $A_b$ is

1518: incompressible in $M$, and hence is boundary parallel in $M$.

1519: Similarly, since the boundary of $D_2$ intersects a meridian of $J_1$

1520: three times, $\bdd D_2$ is a nonseparating curve on $F_w$, so after

1521: adding $N(D_2)$ the white frontier becomes an annulus $A_w$, which for

1522: the same reason must be boundary parallel in $M$.  It follows that $M$

1523: is homeomorphic to $N(F_1 \cup T_0 \cup D_1 \cup D_2)$, where $F_1$ is

1524: the punctured annulus $A_1 \cap M$.  The boundary curves of $D_i$ are

1525: determined by the graphs, which have been determined (up to symmetry)

1526: as above.  Hence the manifold $M$ is uniquely determined.  \endproof

1527: 

1528: \proclaim{Lemma 8.3} There is a unique irreducible,

1529: $\bdd$-irreducible, anannular manifold $M$ which admits two annular

1530: Dehn fillings $M(r_1), M(r_2)$ with $\Delta(r_1, r_2) = 4$ and $n_1 =

1531: n_2 = 2$.  \endproclaim

1532: 

1533: \proof The proof is similar to that of Lemma 8.2. In this case the

1534: jumping number is $1$, and one can show that up to symmetry the graphs

1535: must be as shown in Figure 8.2.  The proof that $M$ is determined by

1536: the graphs is the same as in the proof of Lemma 8.2.

1537: \endproof

1538: 

1539: \bigskip

1540: \leavevmode

1541: 

1542: %\epsfxsize=5in      % adjust the width  of the figure

1543: %\epsfysize=3in      % adjust the height of the figure

1544: \centerline{\epsfbox{Fig8.2.ps}} 

1545: \bigskip

1546: \centerline{Figure 8.2}

1547: \bigskip

1548: 

1549: We now prove Theorem 1.1, which we restated here for the reader's

1550: convenience.  

1551: 

1552: \proclaim{Theorem 1.1} Suppose $M$ is a compact, connected,

1553: orientable, irreducible, $\bdd$-irreducible, anannular $3$-manifold

1554: which admits two annular Dehn fillings $M(r_1)$, $M(r_2)$ with $\Delta

1555: = \Delta(r_1, r_2) \geq 4$.  Then one of the following holds.

1556: 

1557: (1)  $M$ is the exterior of the Whitehead link, and $\Delta = 4$.

1558: 

1559: (2)  $M$ is the exterior of the $2$-bridge link associated to the

1560: rational number $3/10$, and $\Delta = 4$.

1561: 

1562: (3)  $M$ is the exterior of the $(-2,3,8)$ pretzel link, and $\Delta =

1563: 5$.  

1564: \endproclaim

1565: 

1566: \proof

1567: By Proposition 6.5, we must have $n_{\a} \leq 2$ for $\a = 1,2$.  If

1568: $G_{\a}$ is special, then by Proposition 7.1 the manifold $M$ is the

1569: exterior of the Whitehead link.  If $G_{\a}$ is nonspecial, then by

1570: Lemma 8.1 the graphs $G_{\a}$ must be as in Figure 8.1 or 8.2, and by

1571: Lemmas 8.2 and 8.3, in each case the manifold $M$ is uniquely

1572: determined by the graphs; hence there are at most three manifolds $M$

1573: which may admit two annular Dehn fillings of distance at least $4$

1574: apart.  On the other hand, it has been shown in [GW1, Theorem 7.5]

1575: that each of these manifolds admits two such fillings.  Hence the

1576: result follows.

1577: \endproof

1578: 

1579: \Refs

1580: \widestnumber\key{BFLW}

1581: 

1582: \ref \key CGLS \by  M. Culler, C. Gordon, J. Luecke and P. Shalen

1583: \paper Dehn surgery on knots \jour Annals Math. \vol 125 \yr 1987 \pages 

1584: 237--300 

1585: \endref

1586: 

1587: \ref \key EW \by M. Eudave-Mu\~noz and Y-Q. Wu \paper Nonhyperbolic

1588: Dehn fillings on hyperbolic 3-manifolds \jour Pac. J. Math. \vol 190

1589: \yr 1999 \pages 261--275

1590: \endref

1591: 

1592: \ref \key Go \by C. Gordon \paper Boundary slopes of punctured tori in

1593: 3-manifolds \jour Trans. Amer. Math. Soc.  \vol 350 \yr 1998 \pages

1594: 1713--1790 

1595: \endref

1596: 

1597: \ref \key GL \by C. Gordon and J. Luecke \paper Reducible manifolds

1598: and Dehn surgery \jour Topology \vol 35 \yr 1996 \pages 385--409

1599: \endref

1600: 

1601: \ref \key GW1 \by C. Gordon and Y-Q. Wu \paper Toroidal and annular

1602: Dehn fillings \jour Proc. London Math. Soc. \vol 78 \yr 1999 \pages

1603: 662--700

1604: \endref

1605: 

1606: \ref \key GW2 \bysame \paper Annular and boundary reducing Dehn

1607: fillings \jour Topology \toappear \endref

1608: 

1609: \ref \key HM \by C.~Hayashi and K.~Motegi \paper Only single twists on

1610: unknots can produce composite knots \jour Trans. Amer. Math. Soc. 

1611: \vol 349 \yr 1997 \pages 4465--4479

1612: \endref

1613: 

1614: \ref \key Q \by R. Qiu \paper Annular Dehn surgery on knots in a solid

1615: torus \jour preprint 

1616: \endref

1617: 

1618: \ref \key Th1 \by W. Thurston \paper Three dimensional manifolds,

1619: Kleinian groups and hyperbolic geometry \jour Bull.\ Amer.\ Math.\

1620: Soc. \vol 6  \yr 1982 \pages 357--381

1621: \endref

1622: 

1623: \ref \key Th2 \bysame \book The Geometry and Topology

1624: of 3-manifolds \yr 1978 \bookinfo Princeton University

1625: \endref

1626: 

1627: \endRefs

1628: \enddocument

1629: 

1630: 

1631: