1: 

2: 

3: 

4: \documentclass[12pt]{amsart}  

5:  

6: \usepackage{amssymb} 

7: \usepackage{enumerate, amsfonts, latexsym,psfig,epsfig, color} 

8: 

9: \usepackage{epstopdf}

10: 

11: 

12: 

13: %%% THIS IS FOR THE PICS ON TEXSHOP %%% (Comment inserted Dec. 13, 2004)

14: % Use xfig to get files with .eps_t suffixes (exporting on `Combined PS/Latex (Both Parts)'

15: % Then (for example)

16: 

17: 

18: %\begin{figure}[htb!]

19: %\begin{center}

20: 

21: %\input{RelThin.eps_t}

22: 

23: %\caption{A triangle which is thin relative to the flat $E$.}

24: %\label{RelThinPic}

25: %\end{center}

26: %\end{figure}

27:  

28:  

29: %%% THIS IS FOR THE PICS %%%%  -- Older than the above, doesn't work on TexShop...

30: % Use psfig and epsfig and then the following command with the figures 

31: % Loop.pstex and Loop.pstex_t (found by exporting from xfig with the

32: % option `Combined PS/Latex (Both parts)'.

33:  

34: %\begin{figure}[htbp] 

35: %\begin{center} 

36:   

37: %\input{Loop.pstex_t} 

38:   

39: %\caption{The loop picture.} 

40: %\label{figure:Loop} 

41: %\end{center} 

42: %\end{figure} 

43:  

44: %%%%% end-figures 

45:  

46:  

47:  

48: % Alpha labelled theorems 

49: \newtheorem{thmA}{Theorem} 

50: \renewcommand{\thethmA}{\Alph{thmA}} 

51: \newtheorem{corA}[thmA]{Corollary} 

52:  

53:   

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

55: \newtheorem{thm}[theorem]{Theorem} 

56: \newtheorem{lem}[theorem]{Lemma} 

57: \newtheorem{cor}[theorem]{Corollary} 

58: \newtheorem{prop}[theorem]{Proposition} 

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

60:  

61: 

62: % theorems with special labels 

63: \newtheorem{thmspec}{Main Theorem}

64: %% add a line like the following before each use 

65: %\renewcommand{\thethmspec}{1-2-3 Theorem \kern-.3em} 

66:  

67: 

68:  

69:  

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

71: %\newtheorem{example} [theorem] {Example} 

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

73: %\newtheorem{definition} [theorem]{Definition} 

74: \newtheorem{problem}    {Problem} 

75: \newtheorem{conjecture} {Conjecture} 

76: \newtheorem{observation} {Observation} 

77: \newtheorem{notation}[theorem]{Notation} 

78:  

79: \theoremstyle{definition} 

80: \newtheorem{definition}[theorem]{Definition} 

81: \newtheorem{example}[theorem]{Example} 

82:  

83: \theoremstyle{remark} 

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

85: \newtheorem{RIproof}[theorem]{}%Proof of Regular Implosions}

86:  

87: \newtheorem{remarks}[theorem]{Remarks} 

88:  

89: \numberwithin{equation}{section} 

90:  

91: \newfont{\msb}{msbm10 scaled 1200} 

92: \newfont{\euf}{eufm10 scaled 1200} 

93: \def\Prf:       {\it Proof. \rm} 

94: \def\ad {\mbox{ad}} 

95: \def\Gthree{\mathcal G_3}

96: \def\bt {\begin{theorem}} 

97: \def\et         {\hfill $\square$\end{theorem}} 

98: \def\bl {\begin{lemma}} 

99: \def\el         {\hfill $\square$\end{lemma}} 

100: \def\bd         {\begin{definition}} 

101: \def\ed{\end{definition}} 

102: \def\bc         {\begin{corollary}} 

103: \def\ec         {\hfill $\square$ \end{corollary}} 

104: \def\bea {\begin{eqnarray}} 

105: \def\eea        {\end{eqnarray}} 

106: \def\be {\begin{eqnarray*}} 

107: \def\ee {\end{eqnarray*}} 

108: \def\bp {\begin{proposition}} 

109: \def\ep {\end{proposition}}  

110: \def\bex {\begin{example}} 

111: \def\eex {\end{example}} 

112: \def\height{\text{\rm{time}}} 

113: \def\Pin{\Pi} 

114:   

115: \hoffset -1cm 

116: \voffset 1cm 

117: \textheight 21cm \textwidth 14cm  

118:  

119: \def \Cal{\mathcal} 

120: \def \Bbb{\mathbb} 

121: \def \S{\Sigma} 

122: \def\D{\Delta} 

123: \def\E{\cal E} 

124: \def\ssm{\smallsetminus} 

125:  

126: \def\area{\text{\rm{Area}}} 

127: \def\Area{\text{\rm{Area}}}  

128:  

129: \def\<{\langle} 

130: \def\>{\rangle} 

131: \def\|{{\,|\! |\, }} 

132: \def\iff{{\leftrightarrow}} 

133: \def\onto{\to}  

134: \def\ker{\text{\rm{ker }}} 

135: \def\a{\alpha} 

136: \def\b{\beta} 

137: \def\P{\Cal P} 

138: \def\cal{\Cal} 

139: \def\e{\varepsilon} 

140:  

141: \def\inv{^{-1}} 

142: \def\-{\underline} 

143: \def\wt{\widetilde} 

144: \def\N{\Bbb N} 

145: \def\Z{\Bbb Z} 

146: \def\Q{\Bbb Q} 

147: \def\G{\Gamma} 

148: \def\A{\Cal A} 

149:  

150: \def\time{\text{\rm{time}}} 

151:  

152: \def\An{\text{\rm{Aut}} (F_n)} 

153: \def\OF{\text{\rm{Out}} (F)} 

154: \def\On{\text{\rm{Out}} (F_n)} 

155: \def\slnz{\text{\rm{SL}}(n,\Bbb Z)} 

156: \def\fix{\text{\rm{Fix}}} 

157: \def\FR{\text{\rm{F}}\Bbb R} 

158: \def\isom{\text{\rm{Isom}}} 

159: \def\san{\text{\rm{SA}}_n} 

160:  

161: \def\iso{\cong} 

162: \def\supp{\text{\rm{Supp}}} 

163: \def\pre{\text{\rm{Pre}}} 

164: \def\lpl{left para-linear } 

165: \def\rpl{right para-linear } 

166: \def\n  {\,|\partial\Delta|} 

167: \def\vecZ{\mathcal Z} 

168: \def\tz{{T_0}} % A1A5Lemma 

169: \def\life{\text{\rm Life}} 

170: \def\ttt{T_1} 

171: \def\mess  {\text{\rm Mess}} 

172: \def\R{\Cal R} 

173: \def\AFourC     {2C_1 + 6\ll + 2B(5T_0 + 6\ttt + 2) + 2MC_4(6\ttt +

174: 8T_0 + 3) + (B+3)(3\ttt + 2T_0)M +5M+2} %The length of sum |A4|

175: 

176: \def\T{\mathcal T} 

177: \def\cmm{C_{(\mu,\mu')}(2)} 

178: \def\vin        {\in_v}  % `is a virtual member of' 

179: \def\subT{\chi(\Pin_{\T})} 

180: \def\CT{\chi_c(\T)} %this used to be \chi(\T) 

181: \def\down{\text{\rm{down}}} 

182: \def\ageT{\age\,(\T)}%$\age_{\plT}(t_2({\mathcal T}))$ 

183: \def\age{\text{\rm{Age}}} 

184: \def\bonus{\text{\rm{bonus}}} 

185: \def\dlong{\text{\rm{D}}\Lambda} 

186: \def\ulong{\text{\rm{U}}\Lambda} 

187: \def\plT{p_l({\mathcal T})} 

188: \def\tplT{\tilde p_l(\T)}

189: \def\Bb         {(B+3)(3\ttt + 2T_0)M + 6B\ttt + 4BT_0 + 2\ll + 2B +

190: 5M + 1} % The total contribution of all bonus' (without the $n$).

191: \def\F{\mathcal F} 

192: \def\ptmm{\hat t_1(\mu,\mu')} 

193: \def\prmm{\hat\rho(\mu,\mu')} 

194: \def\pEmm{\hat{\text{\euf T}}  (\mu,\mu')} 

195: 

196: \def\pTmm{\hat\T(\mu,\mu')} 

197: \def\tmm{t_1(\mu,\mu')}

198: \def\pT{\hat\T} 

199: \def\pl{p_l} 

200: \def\pr{p_r} 

201: \def\xT{x({\mathcal T})} %first non-constant edge 

202: \def\prT{p_r({\mathcal T})}  

203: \def\tprT{\tilde p_r(\T)}

204: \def\ttwo{t_2(\T)} 

205: \def\ET{{\text{\euf T}}}

206: \def\Emm{{\text{\euf T}}(\mu,\mu')}

207: \def\tone{t_1(\T)} 

208: \def\I{\mathcal I} 

209: \def\r{\rho} 

210: \def\rT{\rho_\T} 

211: \def\mm{(\mu,\mu')} 

212: \def\bonusT{\text{\rm{bonus}}(\T)} 

213: \def\K  { 2C_0 + 2K_1 + 2B + 1} % |S_0| \leq Kn... 

214: 

215: \def\ll {\lambda_0} 

216:  

217: \def\L{\Cal L} 

218: \def\R{\Cal R}

219: \def\QT{Q(\T)} 

220:  

221:  

222: % this gets rid of the AMS series logo and copyright 

223: \catcode`\@=11 

224: \def\serieslogo@{\relax} 

225: \def\@setcopyright{\relax} 

226: \catcode`\@=12 

227:  

228:  

229: \begin{document} 

230:  

231:  

232:  

233:  

234:  

235:  

236:  

237: \title[Mapping tori of free group automorphisms]

238: {The quadratic isoperimetric inequality

239: for mapping tori of free group automorphisms I:

240: Positive automorphisms\\

241: }

242:  

243: % author  information 

244: \author[Martin R. Bridson]{Martin R.~Bridson} 

245: \address{Martin R.~Bridson\\

246: Mathematics Department\\

247: Imperial College of Science, Technology and Medicine\\

248: 180 Queen's Gate \\

249: London SW7 2BZ\\

250: UK} 

251: \email{m.bridson@ic.ac.uk}  

252:  

253: % author  information 

254: \author[Daniel Groves]{Daniel Groves}

255: \address{Daniel Groves\\

256: Merton College\\

257: Oxford OX1 4JD\\

258: UK} 

259: \email{grovesd@maths.ox.ac.uk} 

260:  

261: \date{29 July, 2003}

262:  

263: \subjclass[2000]{20F65, (20F06, 20F28, 57M07)} 

264:  

265: \keywords{free-by-cyclic groups, automorphisms

266:  of free groups, isoperimetric

267: inequalities, Dehn functions} 

268:  

269: \thanks{The first author's work was supported in part

270: by an EPSRC Advanced Fellowship} 

271:  

272: \begin{abstract} {If $F$ is a finitely generated free group

273: and $\phi$ is a positive automorphism of $F$

274: then  $F\rtimes_\phi\mathbb Z$

275: satisfies a quadratic isoperimetric inequality.} 

276: \end{abstract} 

277:  

278:  

279: \maketitle 

280: 

281:   

282: Associated to an   automorphism $\phi$

283: of any group $G$ one has the algebraic {\em mapping torus}

284: $G\rtimes_\phi\mathbb Z$. In this paper we shall be concerned

285: with the case where $G$ is a finitely generated

286: free group, denoted $F$. We seek to understand the complexity

287: of  word problems for the

288:  groups $F\rtimes_\phi\mathbb Z$ as measured

289: by their Dehn functions. 

290:  

291: The class of groups of the form $F\rtimes_\phi\mathbb Z$

292: has been the subject of intensive investigation  

293: in recent years and a rich structure has begun to

294: emerge  in keeping with the subtlety of the

295: classification of free group automorphisms \cite{BFH}, \cite{BFH2}

296: \cite{BH2}, \cite{FH}, \cite{Lu}, \cite{Sela}.  (See \cite{BestICM} and

297: the references therein.) 

298: Bestvina--Feighn and Brinkmann proved that if $F\rtimes_\phi\mathbb Z$

299: doesn't contain

300: a free abelian subgroup of rank two then

301: it is hyperbolic \cite{BF}, \cite{Brink},

302: i.e. its Dehn function

303: is linear. Epstein and Thurston \cite{E+}

304: proved that if $\phi$ is induced by a surface automorphism

305: (in the sense discussed below) then $F\rtimes_\phi\mathbb Z$ is automatic

306: and hence 

307: has

308: a quadratic Dehn function. The question

309: of whether or not all non-hyperbolic groups of the form

310: $F\rtimes_\phi\mathbb Z$  have quadratic Dehn functions has

311: attracted a good deal of attention. 

312: 

313:  

314: Recall that an automorphism $\phi$ of a finitely generated free

315: group $F$ is called {\em positive} if there is a 

316: basis $a_1,\dots,a_n$ for $F$ such that the reduced

317: word representing each $\phi(a_i)\in F$ contains no inverses $a_j^{-1}$. 

318:  

319:  

320:  

321: 

322: \renewcommand{\thethmspec}{{\bf{Main Theorem. \kern-.3em}}}

323: 

324:  

325: \smallskip

326: 

327: \noindent

328: \begin{thethmspec} \label{MainThmPos}

329: {\em Let $F$ be a finitely generated free group.

330: If $\phi$ is a positive  automorphism of $F$, then

331: $F\rtimes_\phi\mathbb Z$ satisfies a  quadratic isoperimetric inequality.}

332: \end{thethmspec} 

333:  

334: \smallskip

335:  

336:  

337: Modulo a simple change in the interpretation of the symbols used, the

338: proof of this theorem extends {\em verbatim}  to 

339: automorphisms $\phi$ that have a  power that admits a train track

340: representative. Not all automorphisms of free groups admit such

341: representatives. Nevertheless, in a subsequent article

342: \cite{BGrovesII} we use the relative train-track technology developed

343: by Bestvina, Feighn and Handel (\cite{BH2} and \cite{BFH}) and the

344: architecture of the proof of the Main Theorem to establish the

345: quadratic isoperimetric inequality for all groups of the form $F

346: \rtimes \Bbb Z$.

347:  

348:  

349: Much of our modern understanding of the automorphisms of

350: free groups has been guided by the  analogy with

351: automorphisms of surface groups, i.e. mapping classes

352: of surfaces of finite type. This analogy  provides a

353: useful reference point when considering the word problems of

354: mapping tori. 

355:  

356: A self-homeomorphism of a compact surface $S$ defines an outer

357: automorphism of $\pi_1S$ and hence a semidirect product

358: $\pi_1S\rtimes_\phi\mathbb Z$. This group  is the fundamental

359: group of a compact   3-manifold, namely the

360: mapping torus $M_\phi$ of the homeomorphism. By using

361: Thurston's Geometrization

362: Theorem for Haken manifolds, Epstein and Thurston \cite{E+} were

363: able to prove that

364: $\pi_1S\rtimes_\phi\mathbb Z$

365:  is an automatic group and hence its Dehn function is

366: either linear or quadratic. If $S$ has boundary then

367: only the quadratic case arises. A more geometric explanation for the existence of 

368: a quadratic isoperimetric inequality in the bounded case

369: comes from the fact that

370:  $M_\phi$ supports a

371: metric of non-positive curvature, as does any irreducible

372: 3-manifold with non-empty boundary \cite{mb-shs}, \cite{leeb}. 

373:  

374: If $S$ has boundary, then $\pi_1S$ is free. Thus the

375: foregoing considerations give many examples of free-by-cyclic

376: groups that have quadratic Dehn functions. But

377: there are many types of  free group automorphisms that

378: do not arise from surface automorphisms, for example

379: those $\phi$ that do not have a power leaving any

380: non-trivial conjugacy class invariant, and those $\phi$

381: for which there is a word $w\in F$ such that the function $n\mapsto |\phi^n(w)|$

382: grows like a super-linear polynomial. 

383:  

384: The non-automaticity of certain  $F\rtimes_\phi\mathbb Z$ provides

385: a more subtle obstruction to realising $\phi$ as a surface automorphism:

386: in contrast to the

387: Epstein-Thurston Theorem, Brady, Bridson and Reeves  \cite{BB},

388: \cite{BR} showed that certain mapping tori $F_3\rtimes \mathbb Z$ are

389: not automatic, for example that associated to the  automorphism

390: $[a\mapsto a,\, b\mapsto ab,\, c\mapsto a^2c]$.

391: Such examples show that one cannot proceed  via automaticity in order

392: to prove the Main Theorem. Nor can one rely on non-positive

393: curvature,  because Gersten \cite{Ge} showed that the above example

394: $F_3\rtimes\mathbb Z$ is not the fundamental group of any compact

395: non-positively curved space. Thus one needs a new approach to the

396: quadratic isoperimetric inequality.

397: 

398: A technique for dealing with classes of linearly growing automorphisms is described by

399: Brady and Bridson in \cite{BB}, and Macura  \cite{Mac} developed techniques for

400: dealing with polynomially growing automorphisms.  But these techniques apply only

401: to  restricted classes of automorphisms and do not speak to the core 

402: problem of establishing the quadratic isoperimetric inequality

403: for mapping tori of general free group automorphisms. 

404: In the present article and its sequel

405: we attack  this core problem

406: directly, undertaking a detailed analysis of the geometry of van Kampen diagrams over the

407: natural presentations of free-by-cyclic groups.   

408:  

409:  

410: This paper is organised as follows.

411: In Section \ref{vanKampSection} we

412: recall some basic definitions associated to Dehn functions.

413: In Sections \ref{BCSection} and \ref{time}

414:  we record some simple but important observations

415: concerning the large-scale behaviour of  

416: the van Kampen

417: diagrams associated to free-by-cyclic groups and in

418: particular the geometry of {\em corridor} subdiagrams. (The automorphisms

419: considered up to this point are not assumed to be positive.)

420: These observations lead us to a strategy for proving the Main Theorem based

421: on the geometry of the {\em time flow of corridors}. In Section  \ref{StrategySection}

422: we state a sharper version of the Main Theorem adapted to this strategy

423: and reduce to the study of automorphisms with  stability

424: properties  that regulate the evolution of corridors. 

425:   In Section \ref{PrefFutSec} we develop the notion of

426:  {\em preferred future} which allows

427:  us to trace the trajectory of

428:   $1$-cells in the corridor flow.

429:   

430: The estimates that we establish in Sections 5 and 6 reduce us to the

431: nub of the difficulties that one faces in trying to prove the Main

432: Theorem, namely

433: the possible existence of  large blocks of ``constant letters". A

434: sketch of the strategy that we shall use to overcome this problem is presented

435: in Section 7. The three main ingredients in this strategy are

436: the elaborate global cancellation arguments in Section 8, the machinery

437: of {\em teams} developed in Section 9, and the {\em bonus scheme}

438: developed in Section 10 to accommodate a final tranche of cancellation

439: phenomena whose quirkiness eludes the grasp of teams. In a brief final

440: section we gather our many estimates to establish the bound required for

441: the Main Theorem. A glossary of constants is included for the reader's convenience. 

442: 

443: \section{Van Kampen Diagrams} \label{vanKampSection} 

444:  

445:  

446: We recall some basic definitions and facts concerning 

447: Dehn functions and van Kampen diagrams. 

448:  

449:  

450: \subsection{Dehn Functions and Isoperimetric Inequalities} 

451:  

452: Given a finitely presented group $G=\langle \mathcal A \mid 

453: \mathcal R \rangle$

454: and a word $w $ in the generators $\mathcal A^{\pm 1}$ that

455: represents $1\in G$,

456: one defines

457: $$

458: \area (w)  = \\

459:  \min\big\{ N \in {\mathbb N}^+ \; | \;

460: \exists\text{ equality }

461: w = \prod^N_{j=1}u_j^{-1}r_j u_j \text{ in $F(\mathcal A)$

462: with } r_j \in \mathcal R^{\pm 1} \big\}\, .

463: $$

464:  

465: 

466: The {\it Dehn function} $\delta(n)$ of the finite 

467: presentation $\langle \mathcal A \mid \mathcal R\rangle$ is defined by 

468: $$ 

469: \delta(n) \; = \; \max\{\text{\rm{Area}}(w) \; 

470: |\; w \in \text{\rm{ker}}(F(\mathcal A) \twoheadrightarrow G), 

471: \; |w| \leq n \, \} \, , 

472: $$ 

473: where $|w|$ denotes the length of the   word $w$. 

474: Whenever two presentations 

475: define isomorphic (or indeed  quasi-isometric) 

476: groups, the Dehn functions of 

477: the finite presentations   

478: are equivalent under the relation 

479: $\simeq$ that identifies functions  

480: $[0,\infty)\to [0,\infty)$ that only differ by a quasi-Lipschitz 

481: distortion of their domain and their range. 

482:  

483:  

484: For any constants $p,q\ge 1$, one sees that 

485: $n\mapsto n^p$ is $\simeq$ equivalent to $n\mapsto n^q$ 

486: only if $p=q$. Thus it makes sense to say that the 

487: ``Dehn function of a group" is $\simeq n^p$. 

488:  

489: A group $\G$ is said to {\em satisfy a quadratic isoperimetric 

490: inequality} if its Dehn function is $\simeq n$ or 

491: $\simeq n^2$. A result of  Gromov \cite{Gromov}, detailed 

492: proofs of which were given by several authors, states that if  

493: a  Dehn function is subquadratic, then it is linear --- 

494: see \cite[III.H]{BH} for a discussion, proof and references. 

495:  

496:  

497: See \cite{steer} for a thorough and  elementary account of 

498: what is known about  Dehn functions and an  

499: explanation of their connection  with filling 

500: problems in Riemannian geometry.  

501: \smallskip  

502:  

503: \subsection{Van Kampen diagrams}\label{vkD} 

504:  

505: According to van Kampen's lemma (see \cite{vK}, 

506: \cite{LS} or \cite[I.8A]{BH})   

507: an equality $w = \prod^N_{j=1}u_jr_ju_j^{-1}$ in the 

508: free group $\mathcal A$, with $N=\area(w)$, 

509: can be portrayed by  

510: a finite, 1-connected, 

511: combinatorial 2-complex with basepoint, embedded in $\mathbb R^2$. Such a complex is 

512: called a {\em van Kampen diagram} for $w$; its oriented 1-cells   

513: are labelled by elements of $\mathcal A^{\pm 1}$; 

514: the boundary label on each 2-cell (read with clockwise 

515: orientation from one of its vertices) is an element  

516: of $\mathcal R^{\pm 1}$;  and the boundary cycle of the 

517: complex (read with positive orientation from the basepoint) 

518: is the word $w$; 

519: the number 

520: of 2-cells in the  diagram   is $N$.  Conversely, any van Kampen diagram with $M$ 

521: 2-cells gives rise 

522: to an equality in $F(\mathcal A)$ expressing the word 

523: labelling the boundary cycle 

524: of the diagram as a product of $M$  

525: conjugates of the defining relations.  

526: Thus 

527: $\text{\rm{Area}}(w)$ is the minimum number of 2-cells among all 

528: van Kampen diagrams  

529: for $w$. If a van Kampen diagram $\Delta$ for $w$ has $\text{\rm{Area}}(w)$ 

530: 2-cells, then $\Delta$ is a called a {\em least-area} diagram. If 

531: the underlying 2-complex is homeomorphic to a 2-dimensional 

532: disc, then the van Kampen diagram is called a {\em disc diagram}. 

533:  

534: We use the term {\em area} to describe the number of 2-cells in a 

535: van Kampen diagram, and write $\text{\rm{Area }} \Delta$. We write 

536: $\partial \Delta$ to denote the boundary cycle of the diagram; we write 

537: $|\partial\Delta|$ to denote the length of this cycle. 

538:  

539:  

540: Note that associated to a van Kampen diagram $\Delta$ with basepoint $p$ 

541: one has a morphism of 

542: labelled, oriented graphs $h_\Delta: (\Delta^{(1)},p)\to (\mathcal C_\A, 1)$,  where 

543: $\mathcal C_\A$ is the Cayley graph associated to the choice of 

544: generators $\A$ for $G$. The map $h_\Delta$ takes $p$ to the 

545: identity vertex  $1\in \mathcal C_\A$ and preserves the labels on oriented edges. 

546:  

547: We shall need the following simple observations. 

548:  

549: \begin{lemma} If a van Kampen diagram 

550: $\Delta$ is least-area, then every simply-connected 

551: subdiagram of $\Delta$ is also least-area. 

552: \end{lemma} 

553:  

554: Recall that a function $f:\mathbb N\to [0,\infty)$ 

555: is {\em sub-additive} if $f(n+m)\le f(n) + f(m)$ 

556: for all $n,m\in\mathbb N$. For example, given $r\ge 1,\, k>0$, 

557: the function $n\mapsto kn^r$ is sub-additive. 

558:  

559: \begin{lemma}  \label{disc} 

560: Let $f:\mathbb N\to [0,\infty)$ be a sub-additive function and let $\P$ 

561: be a finite presentation of a group. 

562: If  $\text{\rm{Area }} \Delta\le f(|\partial\Delta|)$ for every 

563: least-area disc diagram $\Delta$ over $\P$, then the Dehn 

564: function of $\P$ is $\le f(n)$. 

565: \end{lemma} 

566:  

567:  

568: \subsection{Presenting $F\rtimes\mathbb Z$} 

569:  

570: We shall establish the quadratic bound required for 

571: the Main Theorem by examing the nature of van Kampen 

572: diagrams over the following natural (aspherical) presentations of

573: free-by-cyclic groups. 

574:  

575: Given a finitely generated free group $F$ and 

576: an automorphism $\phi$ of $F$, we fix a basis 

577: $a_1,\dots,a_m$ for $F$, write $u_i$ to denote 

578: the reduced word equal to $\phi(a_i)$ in $F$, and 

579: present $ 

580: F\rtimes_\phi\mathbb Z$ by 

581: \begin{equation}\label{presentation} 

582: \P\iso \langle a_1,\dots,a_m,t\mid 

583: t^{-1}a_1tu_1^{-1},\dots, t^{-1}a_mtu_m^{-1}\rangle. 

584: \end{equation} 

585: We shall work exclusively with this presentation. 

586: \medskip 

587:  

588:  

589: \begin{figure}[htbp] 

590: \begin{center} 

591:   

592: \input{2cell.eps_t} 

593:   

594: \caption{A $2$-cell in a van Kampen diagram for $F \rtimes_{\phi} \mathbb Z$.} 

595: \label{figure:2cell} 

596: \end{center} 

597: \end{figure} 

598:  

599:  

600:  

601:  

602: \subsection{Time and $t$-Corridors with naive tops} 

603:  

604: The use of $t$-corridors as a tool for investigating 

605: van Kampen diagrams has 

606: become well-established in recent years. In the 

607: setting of van Kampen diagrams over the above presentation,  

608: $t$-corridors are easily described. 

609:  

610: Consider a van Kampen diagram $\Delta$ over 

611: the above presentation $\P$ and focus on an edge in the boundary 

612: $\partial \Delta$ that is labelled 

613: $t^{\pm 1}$ (read with positive orientation from the basepoint).  

614: If this edge lies in the boundary 

615: of a 2-cell, then the boundary  cycle of this 2-cell has 

616: the form $t^{-1}a_itu_i^{-1}$ (read with suitable orientation from 

617: a suitable point, see Figure \ref{figure:2cell}). In particular, there 

618: is  exactly one other edge  

619: in the boundary of the 2-cell that is labelled $t$; crossing 

620: this edge we enter another 2-cell with a similar boundary 

621: label, and iterating the argument we get a chain of 2-cells 

622: running across the diagram; this chain terminates at an edge of 

623: $\partial \Delta$ which (following the orientation of $\partial \Delta$ 

624: in the direction of our original edge labelled $t^{\pm 1}$) is labelled 

625: $t^{\mp 1}$. This chain of 2-cells is called a {\it{$t$-corridor}}. 

626: The edges labelled $t$ that we crossed in the above description 

627: are called the {\em vertical} edges of the corridor.  

628: The vertical edge on $\partial \Delta$ labelled $t^{-1}$ is 

629: called the {\em initial} end of the corridor, and at the other end one 

630: has the {\em terminal} edge. 

631:  

632: Formally, one should define a $t$-corridor to be a combinatorial map 

633: to $\Delta$ from a suitable subdivision of $[0,1]\times [0,1]$: the 

634: initial edge is the restriction of this map to $\{0\}\times [0,1]$; the 

635: vertical edges are the images of the 1-cells of the form $\{s\}\times [0,1]$, 

636: oriented so that the edge joining $(s,0)$ to $(s,1)$ is labelled $t$.

637: The {\em naive top} of the corridor is the edge-path obtained by restricting 

638: the above map to $[0,1]\times\{1\}$, and the {\em bottom} is the restriction 

639: to  $[0,1]\times\{0\}$. 

640: \smallskip 

641:  

642: \noindent{\bf Left/Right Terminology:} The orientation of a disc 

643: diagram induces an orientation on its corridors. Whenever 

644: we focus on an individual corridor, we shall regard its 

645:  initial as being {\em left}most  and its terminal 

646: edge as being {\em right}most. (This is just a suggestive way of saying 

647:  that the 

648: corridor map from $[0,1]\times (0,1)\subset \mathbb R^2$ to $\Delta\subset 

649: \mathbb R^2$ is 

650: orientation-preserving.) 

651:  

652: \smallskip 

653:  

654: 

655: \medskip 

656:  

657: \begin{figure}[htbp] 

658: \begin{center} 

659:   

660: \input{corridor.eps_t} 

661:   

662: \caption{A $t$-corridor} 

663: \label{figure:corridor} 

664: \end{center} 

665: \end{figure} 

666:  

667:  

668:  

669:  

670: See \cite{BG} for a  detailed account of $t$-corridors. 

671: Here we shall need only the following easy facts: 

672: \begin{enumerate} 

673: \item 

674: distinct $t$-corridors 

675: have disjoint interiors;  

676: \item 

677: if $\sigma$ is the edge-path in $\Delta$ running along 

678: the (naive) top or bottom  of a $t$-corridor, then $\sigma$ is 

679:  labelled 

680: by a word in the letters $\mathcal A^{\pm 1}$ 

681: that is equal in $F\rtimes\mathbb Z$   to the 

682: words labelling the subarcs of $\partial \Delta$ 

683: which share the  endpoints of $\sigma$ (given appropriate 

684: orientations); 

685: \item if we are in a least-area diagram  

686: then the word on the bottom of the corridor is freely reduced; 

687: \item the number of 2-cells in the 

688: $t$-corridor is the length of the word labelling the 

689: bottom side. 

690: \item In subsection 1.2 we described the map $h_\Delta$ 

691: associated to a van Kampen diagram. This map 

692: sends vertices of $\Delta$ to vertices of the Cayley graph $\mathcal C_\A$, 

693: i.e. elements of $F\rtimes \langle t \rangle$.  If the initial vertex of a  

694: directed edge in $\D$ is sent to an element of the form $wt^j$, with 

695: $w\in F$, then the edge is defined to occur at {\bf time} $j$. Note that the 

696: vertical edges of a fixed corridor all occur at the same time. 

697: \end{enumerate}     

698:  

699: We will

700: consider the {\em dynamics} 

701: of the automorphism $\phi$.  

702: 

703: 

704: \begin{definition} 

705: [Time and Length]  Item (5) above 

706: implies that the time of each 

707: $t$-corridor $S$ is well-defined;

708: we denote it $\height(S)$.  

709:  

710:  We define the {\em length} of a  

711: corridor $S$ to be the number of 2-cells that it 

712: contains, which is equal to the number of 1-cells along its bottom. 

713: We write $|S|$ to denote the length of $S$. 

714: \end{definition} 

715:  

716:  

717: \subsection{Conditioning the Diagram} 

718:  

719: We are working with the following presentation of $F\rtimes_\phi\mathbb Z$ 

720: $$ 

721: \mathcal P = \langle a_1,\dots,a_m,t\mid 

722: t^{-1}a_1tu_1^{-1},\dots, t^{-1}a_mtu_m^{-1}\rangle. 

723: $$ 

724:  

725: In the light of Lemma \ref{disc}, in order to prove the main 

726: theorem it suffices to consider only {\em disc diagrams}. Therefore, 

727: henceforth we shall assume that all diagrams are topological discs. 

728: We shall also assume that all of the discs considered are 

729: {\em least-area} diagrams for freely reduced words.  

730:  

731: \begin{lemma} 

732: Every least-area disc diagram  over $\P$ is  

733: the union of its $t$-corridors. 

734: \end{lemma} 

735:  

736: \begin{proof} Since the diagram is a disc, every 1-cell lies in 

737: the boundary of some 2-cell. The boundary of each 2-cell 

738: contains two edges labelled $t$. Consider the equivalence relation 

739: on 2-cells generated by $e\sim e'$ if the boundaries of $e$ and $e'$ share an 

740: edge labelled $t$. Each equivalence class forms either a $t$-corridor 

741: or else a $t$-ring, i.e. the closure of an annular sub-diagram 

742: whose internal and external cycles are labelled by a word in the 

743: generators of $F$. If the latter case arose, then since 

744: $F$ is a free group, the word $u$ on the external 

745: cycle would be freely equal to the empty word (since it contains no edges 

746:  labelled $t$). This would contradict the hypothesis that the diagram 

747:   is least-area, because one could reduce its area 

748: by excising the simply-connected sub-diagram bounded by this cycle, 

749: replacing it  with the zero-area diagram for $u$ over the free 

750: presentation of $F$. 

751: \end{proof} 

752:  

753:  

754: \subsection{Folded Corridors} 

755:  

756:  

757: In the light of the above lemma, we see that the diagrams $\Delta$ that we 

758: need to consider are essentially determined once one knows which 

759: pairs of boundary edges are connected by $t$-corridors. However, there 

760: remains a slight ambiguity arising from the fact that free-reduction in 

761: the free group is not a canonical process (e.g.  $x = (xx^{-1})x = x (x^{-1}x)$). 

762:  

763:  

764: To avoid this ambiguity, we  fix a least area disc diagram $\Delta$  

765: and assume that its corridors are {\em folded} in the sense of \cite{B-plms}. 

766: The topological closure  $T\subset\Delta$ of each corridor is a combinatorial 

767: disc. 

768: The hypothesis ``least area" alone 

769: forces the label on the {\em bottom} of the corridor 

770: to be a {\em freely reduced} word in the letters $a_i^{\pm 1}$. 

771: We define the 

772:  {\em top} of the (folded) corridor to be the  

773:  injective edge-path that remains when one deletes from the 

774: frontier of $T$ 

775:  the bottom and ends of the corridor. The word labelling 

776:  this path is the freely reduced word in $F$ that equals the 

777:  label on the naive top of the corridor. Note that, unlike the 

778: bottom of the corridor, the top may fail to intersect  the closure of some 

779: 2-cells --- see Figures \ref{figure:Fold1} and \ref{figure:Fold2} (where the automorphism is $a \mapsto a, 

780: b \mapsto ba^2, c \mapsto ca$). 

781:  

782: \begin{notation} 

783: We write $\top (S)$ and $\bot (S)$, respectively, to denote the top and bottom 

784: of a folded corridor $S$. 

785: \end{notation} 

786:  

787: \smallskip 

788:  

789: {\centerline{ 

790: {\em Henceforth we shall refer to folded $t$-corridors simply as ``corridors".}}} 

791:  

792:  

793: 

794: \bigskip 

795:  

796:  

797: \begin{figure}[htbp] 

798: \begin{center} 

799:   

800: \input{FoldPic1.eps_t} 

801:   

802: \caption{An unfolded corridor} 

803: \label{figure:Fold1} 

804: \end{center} 

805: \end{figure} 

806: 

807: \medskip 

808:  

809: \begin{figure}[htbp] 

810: \begin{center} 

811:   

812: \input{FoldPic2.eps_t} 

813:  

814: \caption{The  corresponding  unfolded corridor.} 

815: \label{figure:Fold2} 

816: \end{center} 

817: \end{figure} 

818:  

819:  

820: \subsection{Naive Expansion and Death} 

821:  

822: For each generator $a_i\in F$ we have the reduced word 

823: $u_i=\phi(a_i)$. Given a reduced  word $v=a_{i(1)}\dots a_{i(m)}$ 

824: we define the {\em naive expansion} of $\phi(v)$ to be  

825: the (unreduced) concatenation $u_{i(1)}\dots u_{i(m)}$. 

826:  

827: Note that if $v$ is the label on an interval of the bottom of a corridor, 

828: then the naive expansion of $\phi(v)$ is the label on the 

829: corresponding arc of the naive top of the corridor. 

830: 

831: 

832:  

833: An edge $\e$ on the bottom of a  corridor $S$ is said to {\em die} in $S$ 

834: if the 2-cell containing that edge  does not contain any edge of  

835: $\top(S)$.  (Equivalently, if $w$ is the label on  $\bot(S)$ and $a_i$ is 

836: the label on $\e$, then the subword  $u_i=\phi(a_i)$ in 

837: the naive expansion of $\phi(w)$ is cancelled 

838: completely during the free reduction encoded in $\Delta$.) In Figure \ref{figure:Fold2}

839: the edge labelled $a$ on the bottom of the corridor dies. 

840:  

841:  

842:  

843: \section{Singularities and Bounded Cancellation} \label{BCSection} 

844:  

845: We have noted that the structure  of a (folded, least-area disc) diagram 

846: over the natural presentation of a free-by-cyclic group 

847: is  the union of its 

848: corridors. 

849: In this section we pursue an 

850: understanding of how these corridors meet. 

851: \smallskip 

852:  

853: 

854: \medskip 

855:  

856:  

857: \begin{figure}[htbp] 

858: \begin{center} 

859:   

860: \input{Wrongway.eps_t} 

861:   

862: \caption{Corridors cannot meet this way in a least-area diagram} 

863: \label{figure:wrongway} 

864: \end{center} 

865: \end{figure} 

866:  

867:  

868: The first observation to make is that corridors cannot meet as in 

869: Figure \ref{figure:wrongway}. 

870:  

871: \begin{lemma} If $S\neq S'$, then $\bot(S)\cap\bot(S')$ 

872: consists of at most one point. 

873: \end{lemma} 

874:  

875: \begin{proof} For each letter $a$, there is only one type 

876: of 2-cell which has the label $a$ on its bottom side.  Thus, if two corridors 

877: were to meet in the manner of  Figure \ref{figure:wrongway}, then we would have a pair 

878: of 2-cells whose union was bounded by a loop labelled  

879: $u_it^{-1}tu_i^{-1}t^{-1}t$, which is 

880: freely equal to the identity. By excising this pair of 2-cells and 

881: filling the loop with a diagram of zero area, we would 

882: reduce the area of $\Delta$ without altering its boundary label --- 

883: but  $\Delta$  is assumed to be a least-area diagram. 

884:  

885: Thus $\bot(S)\cap\bot(S')$ contains no edges. To see that it cannot 

886: contain more than one vertex, follow the proof of Proposition 

887: \ref{SingularityProp}(1). 

888: \end{proof} 

889:  

890:  

891:  

892:  

893: \begin{definition}  

894: A {\em singularity} in $\Delta$ is a non-empty connected  component of the intersection 

895: of the tops of two 

896: distinct folded corridors. A 2-cell  is said to {\em hit} the 

897: singularity if 

898: it contains an edge of the singularity.  

899: 

900: The singularity   is said to be degenerate if it consists of a single point, and

901: otherwise it is {\em non-degenerate}.

902: \end{definition}  

903:  

904: \medskip 

905:  

906: \begin{figure}[htbp] 

907: \begin{center} 

908:   

909: \input{SingPic.eps_t} 

910:   

911: \caption{A `singularity'} 

912: \label{figure:Singularity} 

913: \end{center} 

914: \end{figure} 

915:  

916:  

917: Let $M$ be the maximum of the lengths of the words $u_i$ in 

918: our fixed presentation $\mathcal P$ of $F\rtimes_\phi \mathbb Z$. 

919:  

920: \begin{proposition}[Bounded singularities] \label{SingularityProp}$\ $ 

921:  

922:  

923: \begin{enumerate} 

924: \item[1.] If the tops of two corridors in a  least-area 

925: diagram meet, then their intersection is a singularity. 

926: \item[2.] 

927: There exists a constant $B$ depending only 

928: on $\phi$  such that less than $B$ 2-cells 

929: hit each singularity in a  least-area diagram over $\P$. 

930: \item[3.]  

931: If $\Delta$ is a least-area diagram over $\P$, 

932: then there are less than $2|\partial \Delta|$ non-degenerate singularities 

933: in $\Delta$, and each has length at most $MB$.

934: \end{enumerate} 

935: \end{proposition} 

936:  

937:  

938:  

939: \begin{proof} Suppose that the intersection of the tops of two corridors $S$ and $S'$ 

940: contains two distinct vertices, $p$ and $q$ say. Consider the unique subarcs 

941: of $\top(S)$ and $\top(S')$ connecting $p$ to $q$. 

942:  Each of these arcs is labelled by a reduced word in 

943: the generators of $F$; since the arcs have the same endpoints in $\Delta$, 

944: these words must be identical.  If the arcs did not coincide, then 

945: we could excise the subdiagram that they bounded and replace it with 

946: a zero-area diagram, contradicting our least-area hypothesis. This proves 

947: (1). 

948:  

949:  

950: 

951:  

952: \begin{figure}[htbp] 

953: \begin{center} 

954:   

955: \input{Sing.eps_t} 

956:   

957: \caption{The proof of Proposition \ref{SingularityProp}} 

958: \label{figure:BoundedSing} 

959: \end{center} 

960: \end{figure} 

961:  

962:  

963:  

964:  

965: Figure \ref{figure:BoundedSing} portrays the argument we use to prove (2). In $S$ 

966: (respectively $S'$), we choose  

967: an outermost pair of oriented edges $\e_1, \e_2$ (resp. 

968: $\e_1',\e_2'$) labelled $t$ whose termini lie on the  

969: singularity. We then connect their endpoints by shortest 

970: arcs in the singularity as shown. Note that 

971: each of the arcs labelled $x_1$ and $x_2$ is contained in the top 

972: of a single 2-cell, and hence has length at most $M$. 

973: We write $\alpha_i$ to denote the concatenation of $\e_i$, the arc labelled $x_i$ 

974: and the inverse of $\e_i'$.  

975:  

976:  

977: Let $U^{-1}_i\in F$ be the reduced word representing 

978: $\phi^{-1}(x_i)$. In $F\rtimes_\phi\mathbb Z$ we have $tx_it^{-1}U_i=1$; 

979: let  $\Delta_i$ be a least-area van Kampen diagram portraying  

980: this equality.   

981:  

982: Let $w$ (resp. $w'$) be the label on the edge-path 

983: in $\bot (S)$ (resp. $\bot(S')$) that connects 

984: the initial point of $\e_1$ (resp. $\e_1'$) to 

985: the initial point of $\e_2$ (resp. $\e_2'$).  

986:  

987:  

988:  

989:  

990: If we excise from $\Delta$ the subdiagram bounded by the loop whose label 

991: is 

992: $t^{-1}wtx_2t^{-1}{w'}^{-1}tx_1^{-1}$, then we reduce the area of  

993: $\Delta$ by $|w| + |w'|$. (Recall that the edges on the bottom 

994: of a corridor are in 1-1 correspondence with the 2-cells of the 

995: corridor.) We may then attach a copy of $\Delta_i$ along $\alpha_i$ 

996: and fill the resulting loop labelled $U_1wU_2^{-1}{w'}^{-1}$ with 

997: a diagram of zero area, because this word is equal to $1$ 

998: in the free group $F$.  

999: Thus we obtain a new van Kampen diagram whose boundary label 

1000: is the same as that of $\Delta$ and which has area 

1001: $$ 

1002: \area(\Delta) + \area(\Delta_1) + \area(\Delta_2) - |w| - |w'|. 

1003: $$ 

1004: Since $\Delta$ 

1005: is assumed to be least-area, this implies that  

1006: $ \area(\Delta_1) + \area(\Delta_2) \ge  |w| + |w'|.$ 

1007:  

1008:  

1009:  Let $B_0$ be an upper bound on the area of 

1010: all least-area van Kampen diagrams portraying equalities of the  

1011: form  $txt^{-1}\phi^{-1}(x)^{-1}=1$ with $|x|\le M$. 

1012: (It suffices to take $B_0=MM_{inv}$, where $M_{inv}$ is the maximum 

1013: of the lengths of the reduced words $\phi^{-1}(a_i)$.) By definition,  

1014: $ \area(\Delta_1) + \area(\Delta_2)\le 2B_0$, and hence 

1015: $|w| + |w'|\le 2B_0$. Thus for (2) it suffices to let $B=2B_0 + 1$. 

1016:  

1017: The length of the singularity in the above argument 

1018: is less than the sum of the lengths of  the naive 

1019: expansions of $\phi(w)$ and $\phi(w')$. Since  $|w|+|w'|\le B$,  

1020: the singularity has length less than $MB$. 

1021: 

1022: 

1023: It remains to bound the number of non-degenerate

1024: singularities in $\Delta$. To this

1025: end, we consider the subcomplex  $\Gamma\subset\Delta$ formed by the union of

1026: the tops of all folded corridors. Arguing as in (1), we see that the

1027: graph $\Gamma$ contains no non-trivial loops, i.e. it is a forest. Let 

1028: $V$ denote the set of vertices in $\Gamma$ that have valence at least

1029: 3 or else lie on $\partial \Delta$. (Thus $V$ is the set of 

1030: degenerate singularities, endpoints of non-degenerate singularities,

1031: and endpoints of the tops of corridors.)

1032: Let $E$ be the set of  connected components of

1033: $\Gamma\smallsetminus V$.

1034: 

1035: $|V|-|E|$ is the number $\pi_0$  of connected components of the forest $\Gamma$.

1036: The valence 1 vertices  $V^1\subset\Gamma$ are a subset of the endpoints of

1037: the tops of corridors, so there are less than $|\partial\Delta|$

1038: of them. One can calculate $|E|$ as half the sum of the valences of

1039: the vertices $v\in V$, so $3(|V|-|V^1|) +|V^1| \le 2|E|$. 

1040: Hence

1041: $$

1042: |E| = |V| - \pi_0 \le \frac 2 3 \big{(}|E| + |V^1|\big{)} -\pi_0

1043: < \frac 2 3 \big{(}|E| + |\partial\Delta|\big{)}.

1044: $$

1045: Therefore $|E| < 2|\partial\Delta|$. 

1046: 

1047: Each non-degenerate singularity determines an element of $E$, so

1048: the (crude) estimate in (3) is established. 

1049: \end{proof} 

1050:  

1051:  

1052:  

1053: \begin{lemma}[Bounded Cancellation Lemma] \label{BCL} There is a constant $B$, 

1054: depending only on $\phi$, such that if 

1055: $I$ is an interval consisting of $|I|$ edges 

1056: on the bottom of a (folded) corridor $S$ in a least-area diagram over $\P$, 

1057: and every edge of $I$ dies in $S$, then $|I| < B$. 

1058: \end{lemma} 

1059: \smallskip 

1060:  

1061: 

1062: \medskip 

1063:  

1064:  

1065: \begin{figure}[htbp] 

1066: \begin{center} 

1067:   

1068: \input{BCPic.eps_t} 

1069:   

1070: \caption{Bounded cancellation lemma} 

1071: \label{figure:BCL} 

1072: \end{center} 

1073: \end{figure} 

1074:  

1075: \begin{proof} The argument is entirely similar to that given for part (2) 

1076: of the previous proposition. 

1077: \end{proof} 

1078:  

1079:  

1080:  

1081: The above lemma is a reformulation of the 

1082:  Bounded Cancellation Lemma from \cite{Cooper}, 

1083: which Cooper attributes to Thurston.  

1084:  

1085:  

1086:  

1087:  

1088:  

1089:  

1090: \begin{remark} {\em `Singularities are only 1 pixel large.'} 

1091: The reader may find it useful to keep 

1092: in mind the following picture: think of   

1093: a least-area van Kampen diagram rendered on a computer 

1094: screen and assume that the length of the boundary of  

1095: the diagram is  large, so large that the constant $B$ 

1096: in Proposition \ref{SingularityProp} has to be scaled to something less 

1097: than 1 pixel in order to fit the picture on to the  computer's 

1098: screen. 

1099: In the resulting 

1100: image one sees blocks of $t$-corridors as shown in Figure \ref{figure:singflow} 

1101: below, and the singularities take on the appearance of classical $k$-prong

1102:  singularities in the time-flow of $t$-corridors. 

1103: \end{remark} 

1104:  

1105: 

1106: \medskip 

1107:  

1108:  

1109: \begin{figure}[htbp] 

1110: \begin{center} 

1111:   

1112: \input{SingFlowPic.eps_t} 

1113:   

1114: \caption{Schematic depiction of a singularity} 

1115: \label{figure:singflow} 

1116: \end{center} 

1117: \end{figure} 

1118:  

1119: \medskip 

1120:  

1121:  

1122: \section{Past, Future and Colour} \label{time} 

1123:  

1124: Our investigations thus far have led us to regard van Kampen diagrams 

1125: over $\P$ as flows of corridors  

1126: (at least schematically). We require some more vocabulary to pursue 

1127: this approach. 

1128:  

1129: We continue to work with a fixed disc diagram $\Delta$ over $\P$. 

1130:  

1131:  

1132: \begin{definition}[Ancestors and Colour]\label{defMu} 

1133:  Each edge $\varepsilon_1$ on the bottom of a corridor either 

1134: lies in the boundary of $\Delta$, or else lies in the top of 

1135: a unique 2-cell, the bottom of which we denote $\e_0$. We consider the 

1136: partial ordering on the set $\mathcal E$ of edges from the bottom of all corridors 

1137: generated by setting $\varepsilon_0 < \varepsilon_1$ whenever edges are related 

1138: in this way.  

1139:  

1140:  

1141:  

1142: If $\e'<\e$ then we 

1143: call $\e'$ an {\em ancestor} of $\e$. The {\em past} of $\e$ 

1144: is the set of its ancestors, and the {\em future} of $\e$ is  

1145: the set of edges $\e''$ such that $\e<\e''$. 

1146:  

1147: Two edges are defined to be of the same {\em colour} if  

1148: they have a common ancestor. Since every edge has 

1149: a unique ancestor on the boundary, colours are in 

1150: bijection with a subset\footnote{namely,  

1151: those edges of $\partial\Delta$ that lie on the bottom of 

1152: some 2-cell} of the edges in $\partial\Delta$ whose 

1153: label is not $t$; in particular there are less than 

1154: $|\partial\Delta|$ colours. 

1155:  

1156:  

1157: Each 2-cell in $\Delta$ has a unique edge 

1158: in the bottom of a corridor. Thus 

1159: we may also regard $\le$ as a partial 

1160: ordering on the 2-cells 

1161: of $\Delta$ and define the past, future and colour 

1162: of a 2-cell. 

1163:  

1164: We define the past (resp. future) of a {\em corridor} 

1165:  to be the union 

1166: of the pasts (resp. futures) of its closed 2-cells. 

1167: \end{definition} 

1168:  

1169: \begin{remark}\label{tree} 

1170: Each $e\in\E$ and each 2-cell has at most one immediate 

1171: ancestor (i.e. one that is maximal among its ancestors). 

1172: Consider the graph $\mathcal F$ with vertex set $\E$ that has an edge 

1173: connecting a pair of vertices if and only if 

1174: one is the immediate ancestor of the other. Note 

1175: that $\mathcal F$ is a forest 

1176: (union of trees). 

1177:  

1178: The {\em colours} in the diagram correspond to the 

1179:  connected components 

1180: (trees) of this forest.  

1181:  

1182:  There is a natural embedding 

1183: of $\mathcal F\hookrightarrow\Delta$: choose a point (`centre') 

1184:  in the interior of each 2-cell 

1185: and connect it to the centre of its immediate ancestor by an 

1186: arc that passes through their common edge. 

1187:  

1188: \end{remark} 

1189:  

1190: If the future of a corridor $S'$ intersects a corridor $S$ then 

1191: the intersection is connected: 

1192:  

1193: \begin{lemma}[Connected Pasts] \label{Connected}  

1194: If a pair of 2-cells $\alpha$ and $\beta$ in a 

1195: corridor $S$  have ancestors $\alpha'$ and $\beta'$ in a corridor $S'$, then every 

1196: $2$-cell $\gamma$ that lies between $\alpha$ and $\beta$ in $S$  has 

1197: an ancestor $\gamma'$ that lies between 

1198:  $\alpha'$ and $\beta'$  in $S'$. 

1199:  \end{lemma} 

1200:  

1201: 

1202:  

1203: \begin{figure}[htbp] 

1204: \begin{center} 

1205:   

1206: \input{Loop.eps_t} 

1207:   

1208: \caption{The `loop' picture} 

1209: \label{figure:loop} 

1210: \end{center} 

1211: \end{figure} 

1212:  

1213: \medskip 

1214:  

1215: \begin{proof} 

1216: Connect the centres of  $\alpha$ and $\beta$ by an arc in the 

1217: interior of $S$ that intersects only those 2-cells lying 

1218: between $\alpha$ and $\beta$, and connect the centres of $\alpha'$ 

1219: and $\beta'$ by 

1220: a similar arc in the interior of $S'$. Along with these two 

1221: arcs, we consider the embedded arcs 

1222: connecting $\alpha$ to $\alpha'$ and $\beta$ to $\beta'$ in the forest 

1223: $\mathcal F$ 

1224: described in Remark \ref{tree}. 

1225: These four arcs together form a  

1226: loop, and the disc that this loop encloses does not intersect 

1227: the boundary of $\Delta$. (Recall that $\Delta$ is a disc.) 

1228:  

1229: Consider the tree from $\mathcal F$ that contains $\gamma$. 

1230:  We may assume that 

1231: the arc in this tree 

1232: that connects $\gamma$ to its ancestor on the boundary does 

1233: not intersect the arc we chose in $S$. It must therefore intersect  

1234: our loop either 

1235: in $S'$, yielding the desired ancestor  $\gamma'$ in $S'$, or 

1236: else in one of the arcs connecting $\alpha$ to $\alpha'$, or 

1237: $\beta$ to $\beta'$. If the latter alternative pertains, $\alpha'$ or 

1238: $\beta'$ is an ancestor of $\gamma$, and we are done. 

1239: \end{proof} 

1240:  

1241: We highlight the degenerate case where the 2-cells $\alpha'$ and $\beta'$ 

1242: are equal and have their bottom   on $\partial\Delta$: 

1243:  

1244: \begin{corollary}\label{muConn} 

1245: Within a corridor, the 2-cells of each colour form a connected region.  

1246: \end{corollary} 

1247:  

1248:  

1249:  

1250: \section{Strategy, Strata and Conditioning}  

1251: \label{StrategySection} 

1252:  

1253:  

1254: Everything that has been said up to  this point has 

1255:  been true for  mapping tori of arbitrary automorphisms of 

1256:  finitely generated free groups. {\em Henceforth, 

1257:   we assume that the automorphism $\phi$ is positive.} 

1258:  

1259: A van Kampen diagram whose boundary cycle  has length $n$ contains at 

1260: most $n/2$ corridors. Thus our Main Theorem is an immediate consequence of: 

1261:  

1262: \begin{theorem}\label{BoundS} 

1263:  There is a constant $K$ depending only on $\phi$ 

1264: such that each  corridor in a least-area diagram $\Delta$ over $\P$ 

1265: has length at most $K\,|\partial\Delta|$. 

1266: \end{theorem} 

1267:  

1268:  

1269: In order to establish the desired bound on the 

1270: length of corridors, we must analyse how 

1271: corridors grow as they flow into the future, and 

1272: assess what cancellation can take place to inhibit this 

1273: growth. In the remainder of this section we shall 

1274: condition the automorphism to simplify 

1275: the discussion of growth. 

1276:  

1277: \begin{remark}  

1278:  The mapping torus $F\rtimes_{\phi^k}\mathbb Z$ is isomorphic to a  

1279:  subgroup of finite index in  $F\rtimes_{\phi}\mathbb Z$, namely 

1280:   $F\rtimes_{\phi}k\mathbb Z$. Thus, since the Dehn functions of 

1281:   commensurable 

1282:   groups are  

1283:   $\simeq$ equivalent, we are free to replace $\phi$ by a convenient 

1284:   positive power in our proof of the Main Theorem. 

1285:   \end{remark} 

1286:    

1287:    

1288:    

1289: \subsection{Strata} 

1290:  

1291:  In the following discussion we shall write $x$ to denote an 

1292: arbitrary choice of letter from our basis $\{a_1,\dots,a_m\}$ 

1293: for $F$. 

1294:  

1295:  

1296: Naturally associated to any positive automorphism one has 

1297: {\em supports} and 

1298:  {\em strata}.  The support 

1299:   $\supp(x)$ associated to $x$ is  

1300:    the set of all letters which appear in the freely 

1301:    reduced word $\phi^j(x)$ for some $j \geq 0$.  

1302:    The stratum $\S(x)\subset\supp(x)$ associated to $x$ consists 

1303:    of those $y\in\supp(x)$ such that $\supp(x)=\supp(y)$. 

1304:    

1305:    Note that $y\in \supp(x)$ implies $\supp(y)\subseteq\supp(x)$, 

1306:    and $y\in\S(x)$ implies $\S(y)=\S(x)$. 

1307:     

1308:     

1309:    

1310:     

1311:    There are two kinds of strata. 

1312:      The first are {\em parabolic\footnote{Bestvina 

1313:      {\em et al.} \cite{BFH} 

1314:       use the terminology {\em non-exponentially-growing} strata} strata},  

1315:       which are those of the 

1316:      form $\S(x)$ with $x\notin \supp(y)$ 

1317:       for all $y\in\supp(x)\ssm\{x\}$. 

1318:      The second kind  

1319:  are {\em exponential strata}, where one has $\S(x)=\S(y)$ for 

1320:  some distinct 

1321:  $x$ and $y$. The letter $x$ is defined to be {\em parabolic} 

1322:  or {\em  exponential} according to the type of $\S(x)$. 

1323:          

1324:          

1325:         If $x$ is exponential then $|\phi^j(x)|$ grows exponentially with $j$.  If 

1326:  all the edges of $\supp(x)$ are 

1327:   parabolic then $|\phi^j(x)|$ grows polynomially 

1328:   with $j$.  However, it may also happen that $x$ is a parabolic letter 

1329:   but $|\phi^j(x)|$ 

1330:   grows exponentially; this  

1331:  will be the case if  $\supp(x)$ contains 

1332:  letters $y$ such that $\S(y)$ is exponential. 

1333:                       

1334:                       

1335: \begin{example}  

1336:  Define $\phi: F_3\to F_3$ by $a_1\mapsto a_1^2a_2,\ a_2\mapsto 

1337:   a_1a_2,\ 

1338:  a_3\mapsto a_1a_2a_3$. Then $\S(a_1)=\S(a_2) =  

1339:  \{a_1, a_2\}$ is an 

1340:  exponential stratum, while $\S(a_3)=\{a_3\}$ 

1341:  is a parabolic stratum with $\supp(a_3)=\{a_1,a_2,a_3\}$. 

1342:  \end{example} 

1343:   

1344:  \begin{remark}\label{induct} 

1345:   The relation {\rm $[y< x$ if  

1346:  $\S(y)\subset\supp(x)\ssm \S(x)]$} generates a partial 

1347:  ordering on the letters $\{a_1,\dots,a_m\}$. For each 

1348:  $x$, the subgroup of $F$ generated by $\pre(x)=\{y\mid y<x\}$ 

1349:  is $\phi$-invariant. Let  $F\lfloor x \rfloor $ denote 

1350:   the quotient of $\langle\supp(x)\rangle$ 

1351:  by the normal closure of $\pre(x)\subset\supp(x)$, and 

1352:  let $F\lceil x\rceil $  denote 

1353:   the quotient of $F$ 

1354:  by the normal closure of $\pre(x)\subset F$. Note 

1355: that  $F\lfloor x \rfloor $  

1356: is a free group with basis (the images of) the letters in $\S(x)$, and $F\lceil x\rceil $ is the free 

1357:  group with basis $\{a_1,\dots,a_m\}\ssm\pre(x)$. 

1358:  

1359: The automorphisms of $\pre(x),\  F\lfloor x \rfloor$ 

1360: and $F\lceil x\rceil $ induced by $\phi$ are positive 

1361: with respect to the obvious bases, and their strata 

1362: are images of the strata of $\phi$. 

1363:  \end{remark} 

1364:   

1365:  \subsection{Conditioning the automorphism} 

1366:   

1367:  In the following proposition, the strata considered are those 

1368:   of $\phi^k$. 

1369:  (These may be smaller than the strata of $\phi$; consider 

1370:  the periodic case for example.) 

1371:   

1372:   \begin{proposition}\label{power} There exists a positive 

1373:    integer $k$ 

1374:   such that $\phi_0:=\phi^k$ has the following properties: 

1375: \begin{enumerate} 

1376: \item[1.]       Each letter $x$ appears in its own image under $\phi_0$. 

1377: \item[2.]       Each exponential letter $x$ appears 

1378:  at least $3$ times in its own image under $\phi_0$. 

1379: \item[3.]       For all $x$, each letter $y\in\supp(x)$ appears 

1380:  in $\phi_0(x)$. 

1381: \item[4.]       For all $x$ and all $j \geq 1$, the 

1382: leftmost  and rightmost letters of $\phi_0^j(x)$ 

1383: are the same as those of $\phi_0(x)$. 

1384: \item[5.]  For all $x$, all $j\geq 1$ 

1385:  and all strata $\S\subseteq\supp(x)$, 

1386:   the leftmost  (respectively, 

1387:   rightmost) letter 

1388:  from $\S$ in the reduced word $\phi_0^j(x)$ is the same as 

1389:   the leftmost (resp. 

1390:  rightmost) 

1391:  letter from $\S$ in $\phi_0(x)$. 

1392: \end{enumerate} 

1393: \end{proposition} 

1394:  

1395: \begin{proof} Items (1) to (3) can be seen as simple facts about 

1396: positive integer matrices, read-off from the action of $\phi$ 

1397: on the abelianization of $F$. 

1398: (By definition $a_j\in \S(a_i)$ if and only if the $(i,j)$ 

1399:  entry of some power 

1400: of the matrix describing this action is non-zero.) 

1401:  

1402:  

1403:  

1404: Assume that $\phi_1$ is a power of $\phi$ 

1405: that satisfies (1) to (3). Note that (3) implies 

1406: that the strata of $\phi_1$ coincide with those of any proper 

1407: power of it.  

1408:  

1409: Replacing $\phi_1$ by a positive power if necessary, we may 

1410: assume that if $\phi_1^j(x)$ begins with the letter $x$, for 

1411: any $j\ge 1$, then $\phi_1(x)$ begins with $x$. This ensures  

1412: that {\em{$[y\preceq_L x$ if some $\phi^j(x)$ begins with $y]$}} 

1413:  is a partial 

1414: ordering, for if $\phi_1^{j_k}(x_k)$ begins with $x_{k+1}$ for 

1415: $k=1,\dots,r$ and if $x_{r+1}=x_1$,  

1416: then $\phi_1^{\Sigma j_k}(x_1)=x_1$ 

1417: and hence $x_1=x_2=\dots = x_r$. 

1418: 

1419: If $\phi_1(x)$ begins with $z$ then $z\preceq_L x$, so 

1420: by raising $\phi_1$ to a suitable power 

1421: we can ensure for all $x$ 

1422: that $\phi_1(x)$ begins with a 

1423: letter that is $\preceq_L$-minimal. The $\preceq_L$-minimal 

1424: letters $y$ are precisely those such that $\phi_1(y)$ begins with $y$. 

1425: An entirely similar argument applies to the relation 

1426:  {\em{$[y\preceq_R x$ if some $\phi^j(x)$ ends with $y]$}}. 

1427:  This proves (4). 

1428:  

1429: Now assume that $\phi_0$ satisfies (1) to (4). The assertion  

1430: in (5) concerning leftmost letters from $\S$ is clear 

1431: for those $x$ where  $\phi_0(x)$ begins with $x$. If $\phi_0(x)$ begins

1432: with $y\neq x$, then either $\S\subset\supp(y)$ 

1433: or else the occurrences 

1434: of letters from $\S$ in $\phi_0^j(x)$ are 

1435: in 1-1 correspondence with the occurrences in the  

1436: image of $\phi_0^j(x)$ in $F\lceil y\rceil $.  (Notation of

1437: Remark \ref{induct}.) In the latter case,

1438: arguing by induction on the size of $\pre(y)$ we 

1439: may assume that  the induced automorphism   $\lceil \phi_0\rceil_y 

1440: :F\lceil y\rceil\to F\lceil y\rceil $ has 

1441: the property asserted in (5); the desired conclusion  for $\phi_0^j(x)$

1442: is then  tautologous. In the former case, if 

1443: we replace $\phi_0$ by $\phi_0^2$ then the conclusion 

1444: becomes as immediate as it was when $\phi_0(x)$ began with $x$.  

1445:  

1446: An entirely similar argument applies to rightmost letters. 

1447: \end{proof} 

1448:   

1449: \begin{remark} 

1450: Although we shall have no need of it here, it seems worth 

1451: recording that item (5) of the above proposition 

1452: remains true if one replaces strata $\Sigma \subset \supp(x)$ 

1453: by supports $\supp(y)\subset\supp(x)$. 

1454: \end{remark} 

1455:  

1456: \smallskip

1457: 

1458:  \begin{quote}{\em We now fix an automorphism $\phi=\phi_0$ and assume that 

1459:  is satisfies conditions (1)-(5) above. 

1460:  All of the constants discussed in the sequel 

1461:  will be calculated with respect to this $\phi$.} 

1462: \end{quote} 

1463:  

1464: \section{Preferred Futures, Fast Letters and Cancellation}  

1465: \label{PrefFutSec} 

1466:  

1467: Having conditioned our automorphism appropriately, we 

1468: are now in a position to analyse the fates of (blocks of) edges 

1469:  as they evolve in time. 

1470:   

1471:  \begin{definition}[Preferred futures]\label{pref-fut} 

1472: For each basis element $x\in\{a_1,\dots,a_n\}$,  

1473:  we choose an occurrence of $x$ in the reduced word 

1474:  $\phi(x)$ to be the (immediate) {\em preferred future of $x$}: 

1475:    if $x$ is a 

1476:   parabolic letter, there is only one possible choice; 

1477:   if $x$ is an  

1478:   exponential letter, 

1479:    we choose an occurrence of $x$ that is neither  

1480:   leftmost nor rightmost (recall that we have 

1481:   arranged for $x$ to appear 

1482:  at least three times in $\phi(x)$). More generally, we 

1483:  make a recursive definition of the {\em preferred future 

1484:  of $x$ in $\phi^n(x)$}: 

1485:  this is the occurrence of $x$ in $\phi^n(x)$ that 

1486:  is the preferred future of the 

1487:  preferred future of $x$ in $\phi^{n-1}(x)$.  

1488:   

1489:  The above definition distinguishes an edge $\e_1$  on the top of 

1490:  each 2-cell in our diagram $\Delta$, namely the edge 

1491:  labelled by the preferred future of the label at 

1492:  the bottom $\e_0$ of the 2-cell. We define $\e_1$ to 

1493:  be the (immediate) 

1494:   {\em preferred future} of $\e_0$. As with letters, 

1495:  an obvious recursion then defines a preferred future of $\e_0$ 

1496:  at each step in its future (for as long as it continues 

1497:  to exist). 

1498:   

1499:  Note that $\e_0$ has at most  one preferred 

1500:  future at each time. (It has exactly one until a preferred 

1501:  future dies in a corridor, 

1502:  lies on the boundary, or hits a singularity.) 

1503:   

1504:  If the bottom edge of a 2-cell is $\e_0$, then we define 

1505:  the preferred future of that 2-cell at time $t$ to be the unique 2-cell 

1506:  at time $t$ whose 

1507:  bottom edge is the preferred future of $\e_0$. 

1508:  \end{definition} 

1509:   

1510:   

1511:  

1512:  

1513: \subsection{Left-fast, constant letters, etc.} 

1514: We divide the letters $x\in\{a_1^{\pm 1},\dots,a_m^{\pm 1}\}$ into 

1515: classes according to the growth of the words 

1516: $\phi^j(x), j=1,2,\dots$, and divide the edges of $\Delta$ into 

1517: classes correspondingly. 

1518: \begin{enumerate} 

1519: \item[$\bullet$] 

1520: If $\phi(x) = x$ then $x$ is called a {\em constant 

1521:  letter}.  

1522:   

1523: \item[$\bullet$] If $x$ is a {\em non}-constant letter, then 

1524:   the function $n\mapsto |\phi^n(y)|$ grows 

1525:  like a polynomial of degree $d\in\{1,\dots,m-1\}$ or else as an exponential 

1526:    function of $n$. 

1527:  

1528: \item[$\bullet$] Let $x$ be a non-constant letter. 

1529:  If the distance between the preferred future of $x$ and the   

1530:  beginning of the word $\phi^n(x)$ grows at least quadratically as 

1531:  a function of $n$, we say that 

1532:  $x$ is {\em left-fast}; if this is not the case, 

1533:  we say that $x$ is 

1534:  {\em left-slow}.  {\em Right-fast} and  

1535:  {\em right-slow} are defined similarly. Note that $x$ is 

1536: left-fast (resp. slow) if and only if $x^{-1}$ is right-fast (resp. slow). 

1537:  

1538: \item[$\bullet$] Let $x$ be a non-constant letter. If $\phi(x) = uxv$ (the shown occurrence of $x$ need not be the preferred future), where $u$ consists only of constant letters,   

1539:  then we say that $x$ is {\em \lpl}. (We place no restriction on $v$; in particular 

1540:  it may contain occurrences of $x$.) {\em Right para-linear} is defined 

1541:  similarly. 

1542: \end{enumerate} 

1543:  

1544:  

1545: \bd 

1546: For \lpl letters, we define the {\em (left) para-preferred future} 

1547:  (pp-future) to be the left-most occurrence of $x$ in $\phi(x)$. 

1548:   The (right) pp-future of a \rpl letter is defined similarly, and 

1549:   edges in $\Delta$ inherit these designations from their labels. 

1550:   

1551:   (It is possible that a letter  

1552:    might be both \lpl  and right para-linear, and in such cases the 

1553:    left and right 

1554:     pp-futures need not agree. But when we discuss pp-futures, 

1555:     it will always be clear  

1556:      from the context whether we are favouring the left or the  right.) 

1557: \ed 

1558:  

1559:  

1560:  

1561: The following lemma indicates the origin of the 

1562: terminology `left-fast' (cf.~\cite[Lemma 4.2.2]{BFH}).  

1563: (A slight irritation arises from the fact that 

1564: there may exist letters $x$  such that $x$ is not left-fast but  

1565:  $\phi(x)$ contains left-fast letters; this difficulty accounts 

1566:  for a certain clumsiness in the statement of the lemma.) 

1567:  

1568:  

1569: \begin{lemma}\label{C_0} There exists a constant $C_0$ with the 

1570: following property: if $x\in\{a_1,\dots,a_n\}$ is such that 

1571:  $\phi(x)$ contains a left-fast letter $x'$ 

1572: and if $UVx\in F$ is a reduced word with $V$  positive\footnote{i.e. no 

1573:  inverses $a_j^{-1}$ appear 

1574: in $V$} and $|V|\ge C_0$, 

1575: then for all $j\ge 1$, the preferred 

1576: future of $x'$ is not cancelled 

1577: when one freely reduces $\phi^j(UVx)$. Moreover, 

1578: $|\phi^j(UVx)|\to\infty$  as $j\to\infty$. 

1579: \end{lemma} 

1580:  

1581: \begin{proof} We factorize the 

1582: reduced word $\phi^j(x)$ as $Y_{x,j}x'Z_{x,j}$ to emphasise the 

1583: placement of the preferred future of a fixed left-fast letter 

1584: $x'$ from $\phi(x)$. The fact that $x'$ is left-fast implies that $j\mapsto |Y_{x,j}|$ grows at least quadratically.  

1585:  

1586: Fix $C_0$ sufficiently large to 

1587: ensure that for each of the finitely many possible 

1588: $x\in\{a_1,\dots,a_n\}$, the integer 

1589: $|Y_{x,j}|$ is greater than $Bj$ whenever $j\ge C_0/B$, 

1590: where $B$ is the bounded cancellation constant. 

1591:  

1592: The Bounded Cancellation Lemma assures us that during the 

1593: free reduction of the naive expansion of $\phi(UVx)$, 

1594: at most $B$ letters of the positive word $\phi(Vx)$ will 

1595: be cancelled. At most $B$ further letters will be cancelled 

1596: when the naive expansion of $\phi^2(UVx)$ 

1597: is freely reduced, and so on. Since $V$ and $\phi$ are positive and 

1598: $ |V| \ge C_0$, it follows that $\phi^j(V)$ will 

1599: not be completely cancelled during the free reduction  

1600: of $\phi^j(UVx)$ if $j\le C_0/B$. When $j$ reaches $j_0:=\lceil C_0/B\rceil$ the 

1601: distance 

1602: from the preferred future of $x'$ to the left end of 

1603: the uncancelled segment of $\phi^j(Vx)$ is 

1604: at least  $|Y_{x,j_0}|$, which is greater than $Bj_0$ and hence $C_0$. 

1605: Repeating the argument with $Y_{x,j_0}$ in place of $V$, we conclude that 

1606: the length of the uncancelled segment of $\phi^j(Vx)$ in $\phi^j(UVx)$ 

1607: remains positive and goes to infinity  with $j$. 

1608: \end{proof} 

1609:  

1610: 

1611:  

1612: Significant elaborations of the previous argument will be developed in 

1613: Section \ref{ConstantSection}.  

1614:  

1615: \begin{definition}[New edges, cancellation and consumption]\label{new} 

1616: Fix a 2-cell in $\Delta$. One edge in the top of 

1617: the cell is the preferred future of the bottom 

1618: edge; this will be called  {\em old} and the 

1619: remaining edges will be called {\em new}. (These 

1620: concepts are unambiguous relative to a fixed 2-cell or (folded) corridor, but `old edge' would be 

1621: ambiguous if applied simply to a 1-cell of $\Delta$.) 

1622:  

1623: Two (undirected) edges $\e_1, \e_2$ 

1624: in the naive top of a  

1625: corridor are said to {\em cancel} each other if their images in the 

1626: folded corridor coincide. If $\e_1$ lies to the 

1627: left\footnote{Recall 

1628: that corridors have a left-right orientation.} of 

1629: $\e_2$, we say that $\e_2$ has been cancelled {\em 

1630: from the left} and $\e_1$ has been cancelled {\em 

1631: from the right}. 

1632: If $\e_1$ is the preferred future of an edge $\e$ 

1633: in the bottom of the corridor and $\e_2$ is a new 

1634: edge in the 2-cell whose bottom is $\e'$, then we 

1635: say that $\e'$ has {\em (immediately) consumed} $\e$ 

1636: {\em from the right}. `Consumed 

1637: from the left' is defined similarly. 

1638:  

1639: Let $e$ and $e'$ be edges in $\bot(S)$ for some 

1640: corridor $S$, with $e$ to the left (resp. right) of $e'$. 

1641: If an edge in the future of $e$ 

1642: cancels a preferred future of $e'$, then we say 

1643: that $e$ {\em eventually consumes} $e'$ {\em from 

1644: the left (resp. right).}  

1645: \end{definition} 

1646: \begin{lemma} \label{NoOldCanc} 

1647: A pair of old edges cannot cancel each other. 

1648: \end{lemma} 

1649:  

1650: \begin{proof} 

1651: Suppose that 

1652: two old edges in the naive top of a corridor $S$ 

1653: are labelled $x$ and cancel each other.  These 

1654: edges are the  preferred futures of edges on $\bot(S)$ 

1655: that bound an arc $\alpha$ labelled by a reduced word 

1656:  $x^{-1}wx$. 

1657: Consider the freely-reduced factorisation $\phi(x) = uxv$ where 

1658: the visible $x$ is the preferred future. 

1659: The arc in the naive top of $S$ corresponding 

1660: to $\alpha$ is labelled  

1661:  $v^{-1}x^{-1}u^{-1}Wuxv$, where $W$ is the naive 

1662:  expansion of $\phi(w)$. The old edges that we are considering 

1663:  are labelled by the visible occurrences of $x$ in this word and 

1664:  our assumption that these edges cancel means that the subarc 

1665:  labelled $x^{-1}u^{-1}Wux$ becomes a loop (enclosing a 

1666:  zero-area sub-diagram) in the diagram $\Delta$. 

1667:   

1668:  But this is impossible, because $x^{-1}wx$ is freely reduced, 

1669: which means that $W$ is not freely equal to the empty 

1670:  word, and hence neither is $x^{-1}u^{-1}Wux$. 

1671:  \end{proof} 

1672:   

1673:   

1674:   

1675:  \begin{corollary}\label{parabolicC} 

1676:   An edge labelled by a 

1677:  parabolic letter $x$ 

1678:  can only be consumed by an edge labelled $y$ with   

1679:  $\supp(x)$ strictly contained in $\supp(y)$. 

1680:  \end{corollary} 

1681:   

1682: \begin{remark} 

1683: A non-constant letter can only be (eventually) consumed from the left (resp. right) by a right-fast  

1684: (resp. left-fast) letter. 

1685: \end{remark} 

1686:  

1687:  

1688:  

1689: \begin{remark} The number of old letters in 

1690:  the naive top of a corridor $S$ is $|S|$, so 

1691:  the length of corridors in the future of $S$ 

1692:  will grow relentlessly unless old letters are 

1693:  cancelled by new letters or the corridor hits a 

1694:  boundary or a singularity. 

1695:  \end{remark} 

1696:  

1697:   

1698: An obvious separation argument provides 

1699: us with another useful observation concerning cancellation: 

1700:   

1701:   

1702: \begin{lemma}\label{perfect} Let $\e_1,\ \e_2$ and $\e_3$ be three 

1703: (not necessarily adjacent) edges that appear in 

1704: order of increasing subscripts as one reads from 

1705: left to right along the bottom of a corridor. If 

1706: the future of $\e_2$ contains an edge of $\partial\Delta$ 

1707: or of a singularity, then no edge in the future of 

1708: $\e_1$ can cancel with any edge in the future of $\e_3$. 

1709: \end{lemma} 

1710:  

1711:  

1712:  

1713: \section{Counting Non-constant Letters} \label{NonConstantSection} 

1714:  

1715: In this section we fix a corridor $S_0$ in $\Delta$ and 

1716: bound the contribution of non-constant letters to the 

1717: length of $\bot(S_0)$. 

1718:  

1719:  

1720:  

1721: \subsection{The first decomposition of $S_0$}\label{decomp} 

1722:  

1723: Choose an edge $\e$ on the bottom of $S_0$.  As we follow the 

1724: preferred future of $\e$ forward one of the following (disjoint) events must 

1725: occur: 

1726:  

1727: \begin{enumerate} 

1728: \item[1.] The last 

1729:   preferred future of $\e$ lies on the boundary of  

1730:  $\Delta$. 

1731:   

1732: \item[2.]  The last 

1733:   preferred future of $\e$ lies in a singularity. 

1734:  

1735: \item[3.]   The last 

1736:   preferred future of $\e$ dies in a corridor $S$ (i.e. 

1737:   cancels with another edge from the naive top of $S$). 

1738: \end{enumerate} 

1739:  

1740: We shall bound the length of $S_0$ 

1741: by finding a bound on the number of edges in each of these three 

1742: cases.  

1743:  

1744: We divide Case (3)  into two sub-cases: 

1745:  

1746:  

1747: \begin{enumerate} 

1748: \item[3a.] 

1749:  The preferred future of $\e$ dies when it is cancelled by an edge 

1750: that is not in the future of $S_0$.  

1751:  

1752: \item[3b.] 

1753: The preferred future of $\e$ 

1754: dies when it is cancelled by an edge 

1755: that is in the future of $S_0$.  

1756: \end{enumerate} 

1757:  

1758: \subsection{Bounding the easy bits} \label{EasyBounding} 

1759:  

1760: Label the sets of edges in $S_0$ which fall into the above classes 

1761: $S_0(1), S_0(2), S_0(3a)$ and $S_0(3b)$ respectively.   

1762:  We shall see that $S_0(3b)$ is by far the most troublesome 

1763:  of these sets. 

1764:     

1765: The first of the bounds in the following lemma is obvious, and the 

1766: second follows immediately from  Proposition \ref{SingularityProp}. 

1767: 

1768:  

1769: \begin{lemma}\label{bound1and2} 

1770:  $|S_0(1)| \leq \n \text{   \rm{and}   } 

1771:   |S_0(2)| \leq 2B\n$.    

1772:   \end{lemma} 

1773:  

1774: 

1775:  

1776: \begin{lemma}\label{bound3a} $|S_0(3a)|\le B\n$. 

1777: \end{lemma} 

1778:  

1779: \begin{proof} The preferred future of each $\e\in S_0(3a)$ 

1780: dies in some corridor in the future of $S_0$. Since 

1781: there are less than $\n /2$ corridors, we will be done 

1782: if we can argue that the preferred future of at most $2B$  

1783: such edges can die in each corridor $S$. 

1784:  

1785: Lemma \ref{Connected} tells us that the future of $S_0$ 

1786: intersects $S$ in a connected region, the 

1787: bottom of which is an interval $I$. The Bounded Cancellation Lemma 

1788: assures us that only the edges within a distance $B$ of the 

1789: ends of $I$ can be consumed in $S$ by an edge from 

1790: outside the interval. And by definition, if a preferred future 

1791: of an edge from  $S_0(3a)$ is to die in $S$, then it must 

1792: be consumed by an edge from outside $I$. 

1793: \end{proof} 

1794:  

1795: We have now reduced Theorem \ref{BoundS} to 

1796: the problem of bounding $S_0(3b)$, 

1797: i.e. of understanding cancellation {\em within} the future of $S_0$. 

1798: This will require a great deal of work. As a first step, 

1799: we further decompose 

1800: $S_0$, mingling the above decomposition based on the fates 

1801: of preferred futures of  

1802: edges with the natural decomposition of $S_0$ into 

1803: colours, as defined in Definition \ref{defMu}. 

1804:  

1805:  

1806:  

1807: \subsection{The chromatic decomposition of $S_0$} \label{chromatic} 

1808:  

1809: We fix a colour $\mu$ and 

1810: write $\mu(S_0)$ to denote the interval of $\bot(S_0)$ 

1811: consisting of edges coloured $\mu$.  

1812: We shall abuse terminology to the 

1813: extent of referring to $\mu(S_0)$ as {\em a colour}, evoking 

1814: the mental picture of the 2-cells in $S_0$ being painted 

1815: with their respective colours. (Recall 

1816: that the 2-cells of $S_0$ are in 1-1 correspondence with 

1817: the edges of $\bot(S_0)$.) 

1818:  

1819: We shall subdivide $\mu(S_0)$ into five subintervals 

1820: according to the fates  

1821: of the preferred futures of edges. To this end,  

1822: we define  $l_{\mu}(S_0)$ to be the rightmost edge in $\mu(S_0)$   

1823: whose immediate future contains  a left-fast edge that is 

1824: ultimately consumed 

1825: from the left by an edge of  $S_0$, and we define 

1826:  $A_1(S_0,\mu)$ to be the set of edges in $\bot(S_0)$ 

1827:  from the left end of $\mu(S_0)$ to $l_{\mu}(S_0)$, inclusive. 

1828: We define $A_2(\mu,S_0)\subset\mu(S_0)$ 

1829:  to consist of the remaining 

1830:  edges in $\mu(S_0)$ whose 

1831:  preferred futures  are  ultimately consumed 

1832: from the left by an edge of  $S_0$. 

1833:  

1834: Similarly, we define $r_{\mu}(S_0)$ to be 

1835:  the leftmost edge $\mu(S_0)$ that has a  right-fast edge in its immediate future 

1836:  that is ultimately consumed 

1837: from the right by an edge of  $S_0$, and we define 

1838:  $A_5(S_0,\mu)$ to be the set of edges in $\bot(S_0)$ 

1839:  from the right end of $\mu(S_0)$ to $r_{\mu}(S_0)$, inclusive. 

1840: We define $A_4(\mu,S_0)\subset\mu(S_0)$ 

1841:  to consist of the remaining 

1842:  edges in $\mu(S_0)$ whose 

1843:  preferred futures  are  ultimately consumed 

1844: from the right by an edge of  $S_0$. 

1845:  

1846: Finally, we define $A_3(S_0,\mu)$ to be the 

1847:  remainder of the edges in $\mu(S_0)$.  

1848: 

1849:  

1850: \begin{figure}[htbp] 

1851: \begin{center} 

1852:   

1853: \input{A1A5.eps_t} 

1854:  

1855: \caption{The second decomposition of $S_0$} 

1856: \label{figure:A1A5} 

1857: \end{center} 

1858: \end{figure} 

1859:  

1860: \medskip 

1861:  

1862:  

1863:  

1864: Modulo the fact that any of the $A_i(S_0,\mu)$ 

1865: might be empty, Figure 

1866: 10 is an accurate portrayal of  $\mu$: 

1867: the $A_i(S_0,\mu)$ are  connected 

1868: and they occur in 

1869: ascending order of suffix from left to right. 

1870:  

1871: The chromatic decomposition of $S_0$ is connected to the 

1872: decomposition of Subsection \ref{decomp} by the equality 

1873: in the following lemma, which is a tautology. The  

1874: inequality in this lemma is a restatement 

1875: of Lemmas \ref{bound1and2} and \ref{bound3a}. 

1876:  

1877: \begin{lemma} \label{A3Lemma} 

1878: $$ 

1879:  \bigcup_{\mu} A_3(S_0,\mu) = S_0 \ssm S_0(3b)\ \ \  

1880:  \text{  {\rm{and}}  }\ \ \  

1881:  \sum_{\mu}|A_3(S_0,\mu)| < \left({3B} + 1\right)\n . 

1882:  $$ 

1883:  \end{lemma}    

1884:  

1885: Thus the following lemma is a step towards bounding the 

1886: size of $S_0(3b)$. 

1887:  

1888: \begin{lemma} \label{A1A5Lemma} 

1889:  

1890: $$ 

1891: |A_1(S_0,\mu)|  \leq  C_0 \ \ \  

1892:  \text{  {\rm{and}}  }\ \ \  

1893: |A_5(S_0,\mu)|  \leq  C_0. 

1894: $$ 

1895: 

1896: \end{lemma} 

1897:  

1898: \begin{proof} 

1899: We prove the result only for $A_1(S_0,\mu)$; 

1900: the proof for $A_5(S_0,\mu)$ is entirely similar.  

1901: 

1902:  

1903: As in 

1904: Lemma \ref{perfect}, we know that the entire  future of the edges of  

1905: $A_1(S_0,\mu)$ to the left of $l_\mu(S_0)$ must 

1906: eventually be consumed from the left  by edges of $S_0$. This means 

1907: that we are essentially in the setting of Lemma \ref{C_0}, with  

1908: $l_{\mu}(S_0)$ in the role 

1909: of $x$ and $A_1(S_0,\mu)$ in the role of $Vx$. 

1910:  

1911: 

1912: Thus if the length of $A_1(S_0,\mu)$ were greater than $C_0$,  

1913: then we would conclude that  no  left-fast edge in the  

1914: immediate future of  $l_{\mu}(S_0)$ would be cancelled from the left by 

1915: an edge  of $\bot(S_0)$, contradicting the definition of $l_{\mu}(S_0)$.   

1916: \end{proof} 

1917:  

1918: \begin{corollary} 

1919: $$ 

1920: \sum_{\mu}|A_1(S_0,\mu)| \, \leq \, C_0\n 

1921:  \ \ \  

1922:  \text{  {\rm{and}}  }\ \ \  

1923: \sum_{\mu}|A_5(S_0,\mu)| \, \leq \, C_0\n . 

1924: $$ 

1925: \end{corollary} 

1926:  

1927: \subsection{A further decomposition of $A_2(S_0,\mu)$ 

1928:  and $A_4(S_0,\mu)$} 

1929:  

1930:  

1931:  

1932: It remains to bound $A_2(S_0,\mu)$ and $A_4(S_0,\mu)$.  

1933: We deal only with $A_4(S_0,\mu)$, the argument for $A_2(S_0,\mu)$ 

1934: being entirely similar. 

1935:  

1936: First partition $A_4(S_0,\mu)$ into subintervals $C_{(\mu,\mu')}$ 

1937: that consist of edges  that are eventually consumed by edges of a specified 

1938: colour $\mu'$. Then partition $C_{(\mu,\mu')}$ into two subintervals:  

1939: $C_{(\mu,\mu')}(1)$ begins at the   

1940: right of  $C_{(\mu,\mu')}$ 

1941: and ends with the last non-constant edge;  

1942:  $C_{(\mu,\mu')}(2)$ consists of the remaining (constant) edges. 

1943: See Figure \ref{figure:Cmumu}. 

1944: \medskip 

1945:  

1946:  

1947: \begin{figure}[htbp] 

1948: \begin{center} 

1949:   

1950: \input{Pic10.eps_t} 

1951:   

1952: \caption{$C_{(\mu,\mu')}(1)$ and $C_{(\mu,\mu')}(2)$.} 

1953: \label{figure:Cmumu} 

1954: \end{center} 

1955: \end{figure} 

1956:  

1957: In the course of this section we will bound the size of the intervals $C_{(\mu,\mu')}(1)$ and during the following four sections we bound the sum over all pairs $(\mu,\mu')$

1958: of the  sizes of the intervals $C_{(\mu,\mu')}(2)$ to get the desired bound on $|S_0(3b)|$.  In order to control this sum, we have 

1959: to address the question of which colours can be adjacent. 

1960:  

1961: \subsection{Adjacent Colours} 

1962:  

1963: 

1964: In Corollary \ref{muConn} we saw that in any corridor 

1965: $S$, the edges in $\bot(S)$ of a fixed colour form an interval. 

1966: We say that two distinct colours $\mu$ and $\mu'$ are 

1967: {\em adjacent} in $S$ if the closed intervals 

1968: $\mu(S)$ and $\mu(S')$ 

1969: have a common endpoint in $\bot(S)$. (Equivalently, 

1970: there is a pair of 2-cells in $S$, one  coloured $\mu$ and 

1971: the other $\mu'$, that share an edge labelled $t$.) 

1972: We write 

1973:  ${\vecZ}$ to denote the set of ordered pairs  

1974: $(\mu,\mu')$ such that  

1975:  $\mu$ and $\mu'$ are adjacent in some corridor $S$ 

1976: with $\mu(S)$ to the left of  $\mu'(S)$  in $\bot(S)$, 

1977: and we write $\vecZ$ to denote the set of unordered 

1978: pairs.  

1979:  

1980: \begin{lemma} \label{NoOfAdjacencies} 

1981: $$ 

1982: |{\vecZ}| < 2\n -3 .     

1983: $$ 

1984: \end{lemma} 

1985:  

1986: \begin{proof}  

1987:  We shall express this proof in the language 

1988: of the forest $\mathcal F$ introduced in Remark \ref{tree}. Suppose 

1989: that $\mu$ and 

1990: $\mu'$ are adjacent in $S$. 

1991: In $S$ we can connect the centre 

1992: of some 2-cell coloured $\mu$ to the centre of some 2-cell 

1993: coloured $\mu'$ by an arc contained in the union of 

1994: the pair of 2-cells. The union of this arc and the trees in $\mathcal F$ 

1995: corresponding to the colours $\mu$ and $\mu'$ disconnects 

1996: the disc $\Delta$; each of the other trees in $\mathcal F$ 

1997: is entirely contained in a 

1998: component of the complement, and the 

1999: colours with trees in different components can 

2000: never be adjacent in any corridor. 

2001:  

2002: We can encode adjacencies of colours by a chord diagram: draw 

2003: a round circle with marked points representing the colours of 

2004: $\Delta$ in the cyclic order that they appear in $\partial\Delta$, 

2005: then connect two points by a straight line if the corresponding 

2006: colours are adjacent in some corridor.  

2007:  The final phrase of 

2008: the preceding paragraph tells us that the lines in this 

2009: chord diagram do not intersect in the interior of the disc. 

2010: A simple count shows that since there are less than 

2011: $\n$ colours, there are less than $2\n -3$ lines in this diagram. 

2012: \end{proof} 

2013:   

2014:  

2015:  

2016: \subsection{Non-constant letters in 

2017:  $C_{(\mu,\mu')}$ that are not left-fast} 

2018:  \label{NonConstantSubsect} 

2019:  

2020: We stated in the introduction that a careful analysis of 

2021: van Kampen diagrams would allow us to reduce the Main Theorem to the 

2022: study of blocks of constant letters.  

2023: In this section we achieve the last step of this reduction. 

2024:  

2025: \begin{lemma} \label{C1Lemma}  

2026: There is a constant $C_1$ depending only on $\phi$ 

2027: with the following property: 

2028:  

2029: Let $S$ be a corridor and let $\mu_1$ and $\mu_2$ be 

2030:  colours that occur in $S$ with $\mu_1$ to the left of $\mu_2$ (but do not 

2031: assume that $\mu_1(S)$ is adjacent to $\mu_2(S)$).  Let 

2032:  $I\subset A_4(S,\mu_1)$ 

2033:   be a sub-interval that satisfies the 

2034:  following conditions 

2035: \begin{enumerate} 

2036: \item[1.] the left-most edge of $I$ is 

2037:  non-constant \mbox{and } 

2038:    

2039: \item[2.]  the preferred future of each edge in $I$ 

2040:  is eventually consumed by an edge of $\mu_2(S)$.\\ 

2041: \end{enumerate} 

2042: \noindent Then $|I| \leq C_1$. 

2043: In particular, $|C_{(\mu,\mu')}(1)| \leq C_1$ 

2044: for all $(\mu,\mu') \in {\vecZ}$. 

2045:  

2046: It suffices to take $C_1 = 2mB^2$, where $m$ is the rank of $F$, and 

2047: $B$ is the constant from the Bounded Cancellation Lemma. 

2048: \end{lemma}  

2049:  

2050:  

2051: \begin{proof} 

2052:   The region $I$ being considered contains no edge with a right-fast  

2053: letter in the $\phi$-image of its label. 

2054:  Since all exponential letters 

2055:  are both left-fast and right-fast, all non-constant edges in the future of $I$ are  parabolic.   

2056:  

2057: We begin the argument at the stage in time where $\mu_2$ 

2058: starts cancelling $I$.  For 

2059: notational convenience we assume that this time 

2060: is in fact $\height(S)$. (If it is not, then the  

2061: fact that the length of $I$ may 

2062: have increased in passing from $\time(S)$ to 

2063: this time adds greater strength to the 

2064: bound we obtain.) 

2065:   

2066:  We focus on the leftmost 

2067:  edge $\e_0$ of $I$ that is labelled 

2068:  by a non-constant letter $x$ for which $\supp(x)$  is maximal  

2069:  among the supports of all edge-labels 

2070:  from $I$ (with respect to inclusion). 

2071:  Let $y$ be the label on the edge 

2072:  $\e_0'$ of $\mu_2(S)$ that 

2073:  eventually consumes $\e_0$ (oriented as shown in Figure \ref{C0Pic}). 

2074:  Note that $\supp(x)$ is strictly contained in $\supp(y)$, 

2075:  by Corollary \ref{parabolicC}. If $\e_0'$ consumes $\e_0$ immediately, 

2076:  then the bounded cancellation lemma tells us that 

2077: $\e_0$ is a distance less than $B$ from the righthand end 

2078: of  $I$. If not, then we 

2079:  proceed one step into the future\footnote{proceeding one 

2080: step into the future also allows us to assume that there 

2081: are no letters coloured $\mu_1$ to the right of $I$} 

2082:  and  appeal to the 

2083:  conditioning done in Proposition \ref{power}(5) to assume 

2084:  that for all $j\ge 1$, the rightmost letter in $\phi^j(y)$ 

2085:  whose support includes $x$ is $y$. 

2086:  We shall call the edge in the future of $\e_0'$ 

2087: carrying the rightmost $y$  

2088:  the {\em highlighted} future of $\e_0'$ (perhaps it is not 

2089:  the preferred future).  

2090: 

2091: \begin{figure}[htbp] 

2092: \begin{center} 

2093:   

2094: \input{C0-Pic.eps_t} 

2095:   

2096: \caption{The edge labelled $\epsilon_0'$ will eventually consume

2097: $\epsilon_0$.}  

2098: \label{C0Pic} 

2099: \end{center} 

2100: \end{figure} 

2101:   

2102:   The first important point to observe is that  

2103:   the maximality of $\supp(x)$ ensures that 

2104:   there will never be any new edges labelled $x$ 

2105:   in the future of $I$ 

2106:   (`new' in the sense of \ref{new}). 

2107:    

2108:  The second important point to note is that  

2109:  the edges labelled $x$ in the future of 

2110:  $\e_0'$ that are to cancel with 

2111:  the futures of the 

2112:  edges labelled $x$ in $I$ must all lie to the 

2113:  left of the highlighted future of $\e_0'$. The point here is 

2114:  that the highlighted future of $\e_0'$ 

2115:  cannot be cancelled by an edge of $I$ (by the maximality 

2116:  of $x$), and in order for it to be cancelled from the 

2117:  other side, all the edges to 

2118:  its right labelled $x$ 

2119:  would have to be cancelled first, which would mean that they too 

2120:  were cancelling with something not in the future of $I$. 

2121:   

2122:  We now come to the key observation of the proof: at 

2123:  each stage $j$ steps into  the  

2124:   future of $S$, the leftmost\footnote{we 

2125:   have already noted that this is to 

2126:   the left of the highlighted future of $\e_0'$} 

2127:    edge $\e_j'$ in the future of $\e_0'$ 

2128:   that is labelled  

2129:   $x$ must be cancelled by an edge from 

2130:   the future of $I$ {\em immediately}, i.e. in the corridor 

2131:   where it appears at  $\height(S)+j$. 

2132:    Indeed if this 

2133:   were not the case, then  $\e_j'$ would develop a preferred 

2134:   future which, being an old  

2135:    edge (in the sense of Definition \ref{new}), could only  

2136:   cancel with a  new edge (Lemma \ref{NoOldCanc}) 

2137:   in the future of $I$. And since 

2138:   we have arranged that there be no new edges labelled $x$, 

2139:   the preferred future of $\e_j'$ would never cancel with 

2140:   an edge in the future of $I$. But this cannot be, because 

2141:   the continuing existence of a preferred future for $\e_j'$ would prevent 

2142:   anything to its {\em right} consuming  an 

2143:   edge in the future of $I$, and the penultimate sentence in the 

2144:   third paragraph of this proof implies that no new 

2145: edges labelled $x$ will ever appear to its {\em left} in the future of $\e_0'$. 

2146: Thus if $\e_j'$ is not 

2147:   cancelled immediately then we have a  contradiction 

2148:   to the fact that $\e_0'$ must 

2149:  eventually  consume $\e_0$. 

2150:   

2151:    

2152:   We have just proved that at $\height(S)+j$ the 

2153:   edge $\e_j'$ must cancel with the preferred 

2154:   future of an edge $\e_j$ in $I$ that is labelled $x$.  

2155:   According to 

2156:   the Bounded Cancellation Lemma, the preferred 

2157:   future of $\e_j$ at $(\height(S)+j-1)$ must lie within 

2158:   a distance $B$ of the right end of the future of $I$.  

2159:   Since there is  no cancellation within the 

2160:   future $I$, an iteration of this argument shows that 

2161:    for as 

2162:   long as there exist edges labelled $x$ in the future 

2163:   of $I$, each successive pair of these edges is separated 

2164:   by less than $B+|\phi(y)|\le 2B$ edges at each moment in time, 

2165:   and the rightmost must be within a distance $B$ of the 

2166:   right end of the future of $I$. 

2167:    

2168:   But since $\phi(x)$ contains at least 

2169:   one letter other than  the preferred future of $x$, 

2170:   it follows that there cannot be a pair 

2171:   of  edges of $I$ labelled $x$ that remain unconsumed 

2172:   at  $\height(S)+2B$, for otherwise 

2173:   they would have grown a distance more than 

2174:   $2B$  apart, contradicting the 

2175:   conclusion of the previous paragraph. And proceeding 

2176:   one more step into the future, the last edge labelled $x$ 

2177:   must be consumed.  

2178:    

2179:   Since at most 

2180:   $B$ letters of $I$ are cancelled at the right  

2181:    at each stage in its future, all of the edges of $I$ labelled $x$ 

2182:   are within a distance less than $2B^2$ of the right end of $I$, 

2183:   and they are all consumed when $I$ has flowed $2B$ steps 

2184:   into the future. 

2185:   If no non-constant edges remain in the future of $I$ 

2186:   at this stage, then we know 

2187:   that $|I|\le 4B^2$. 

2188:    

2189:   If there do remain non-constant edges, we take the maximal interval of the 

2190:    future of $I$ at   $\height(S)+2B$ whose leftmost 

2191:    edge is non-constant, and we repeat the argument. (This 

2192:    interval is obtained from the complete future of $I$ by 

2193:    removing a possibly-empty collection of constant edges 

2194:    at its left extremity.) 

2195:     

2196:   We proceed in this manner. The interval that 

2197:   we begin with at each iteration has strictly fewer 

2198:   strata than the previous one 

2199:   and therefore the procedure 

2200:   stops before $m=\text{\rm{rank}}(F)$ iterations. At 

2201:   the time when it stops (at most $\height(S)+2mB$), the future 

2202:   of $I$ has been cancelled entirely, except possibly for 

2203:   a block of 

2204:    constant edges at its left extremity.  

2205:    With one final appeal to the bounded cancellation 

2206:    lemma, we deduce that  $|I|\le 2mB^2$. 

2207:   \end{proof} 

2208:    

2209:  

2210:   \begin{corollary}\label{C1Corollary} 

2211: \[ \sum_{(\mu,\mu') \in {\vecZ}}|C_{(\mu,\mu')}(1)| < 2C_1\n . \] 

2212: \end{corollary} 

2213:  

2214: \begin{proof} This follows immediately 

2215:  from Lemmas \ref{NoOfAdjacencies} and \ref{C1Lemma}. 

2216: \end{proof} 

2217:  

2218:  

2219:  

2220:  

2221: \section{The Bound on $\sum\limits_{\mu \in S_0}|A_4(S_0,\mu)|$ and $\sum\limits_{\mu \in S_0}|A_2(S_0,\mu)|$}\label{A4sec} 

2222: 

2223: The sum of our previous arguments has reduced us to the nub of the

2224: difficulties that one faces in trying to prove the Main Theorem,

2225: namely the possible existence of large blocks of constant letters in

2226: the words labelling the bottoms of corridors.

2227: Now we must obtain a bound on

2228:  $$\sum\limits_{(\mu,\mu') \in {\vecZ}} |C_{(\mu,\mu')}(2)| 

2229: $$ 

2230: that will enable us to bound 

2231: $\sum\limits_{\mu \in S_0}|A_4(S_0,\mu)|$ and\footnote{In 

2232: practice 

2233:  we  only need concern ourselves with $A_4$, the arguments for $A_2$ 

2234: being entirely similar}   $\sum\limits_{\mu \in 

2235: S_0}|A_2(S_0,\mu)|$ by a linear function of $\n$.  These are the final 

2236: estimates required to complete the proof of the Main Theorem --- see 

2237: Section \ref{summary} for a r\'esum\'e of the proof.  

2238: 

2239:  The regions $C_{(\mu,\mu')}(2)$ are static, in the sense 

2240: that they do not change under iteration by $\phi$, so  

2241: the considerations of future growth 

2242: that helped us so much in previous 

2243: sections cannot be brought to bear directly. Rather, we must 

2244: analyse the complete history of blocks of constant letters, 

2245: understand how large blocks come into existence, and 

2246:  use global considerations to limit the sum of 

2247:  the sizes of all such blocks. 

2248:   

2249: Because of the global 

2250: nature of the arguments, 

2251:  we shall not obtain bounds on the sizes of the individual sets $C_{(\mu,\mu')}(2)$.

2252:  Instead, 

2253: we shall identify an associated block of 

2254: constant letters

2255: elsewhere in 

2256: the diagram  (a ``team") that is amenable to a delicate string of 

2257: balancing arguments that facilitates a bound on a union of 

2258: associated regions $C_{(\mu,\mu')}(2)$. 

2259: 

2260: Our strategy is motivated by the following considerations. 

2261: Believing Theorem \ref{BoundS} to be true, we seek 

2262: payment from the global geometry of $\Delta$ to compensate 

2263: us for having to handle the troublesome blocks of constant 

2264: edges $C_{(\mu,\mu')}(2)$;  the currencies of payment are 

2265: {\em consumed colours} 

2266: and dedicated subsets of 

2267: edges on $\partial\Delta$ --- since $\Delta$ can have at 

2268: most $\n$ of each, if we prove that adequate payment is available

2269: then our troubles will be bounded and the Main Theorem  

2270: will follow.  

2271: The chosen currencies are apposite

2272: because, as we shall  see in Section \ref{ConstantSection}, 

2273: a large block of edges labelled by constant letters can only 

2274: come into existence if  a colour (or colours) associated to a 

2275:  component of this block in the past was consumed completely, 

2276:   or else the boundary of $\Delta$ intruded into the past of 

2277:    the block (or else something nearby) causing smaller regions of constant edges to elide.  

2278:  

2279: 

2280:  

2281:  

2282: In the remainder of this section we shall explain how various estimates on the behaviour of 

2283: blocks of constant letters in $\Delta$ can be combined to obtain 

2284: the bounds that we require on  

2285: $\sum\limits_{\mu \in S_0}|A_4(S_0,\mu)|$ and  $\sum\limits_{\mu \in 

2286: S_0}|A_2(S_0,\mu)|$. We hope that this explanation will provide the 

2287: diligent reader with a useful road map and sufficient motivation to 

2288: sustain them through the many technicalities needed to establish the 

2289: estimates in subsequent sections. 

2290:  

2291: 

2292: In the following proposition, $M$ is the maximum length of the 

2293: images $\phi(x)$ of the basis elements of $F$, while

2294: $\ttt$ is the constant from the Pincer Lemma \ref{PincerLemma}, and

2295: $C_1$ is the upper bound on the lengths of the intervals

2296: $C_{(\mu,\mu')}(1)$ from Lemma \ref{C1Lemma}, $T_0$ comes from the Two

2297: Colour Lemma \ref{TwoColourLemma} and $C_4$ comes from Lemma

2298: \ref{G34pics}. The constant $\ll$ is

2299: defined above Definition \ref{NestingDef}, and $B$ is the bounded cancellation

2300: constant from Lemma \ref{BCL}.

2301: \smallskip

2302: 

2303: \noindent{\bf{The Constant $K_1$ is defined to be}}

2304: $$

2305:  \AFourC.  

2306: $$ 

2307: 

2308: \begin{proposition}\label{SummaryLemma} 

2309: \[      \sum_{\mu \in S_0}|A_4(S_0,\mu)| \leq K_1  \n . 

2310:  \]  

2311: \end{proposition} 

2312:  

2313:  

2314: \subsection{Dramatis Personae} 

2315: The ``proof" that we are about to present is essentially a scheme for

2316: reducing 

2317: the proposition to a series of technical lemmas that will be proved

2318: in Sections \ref{teamSec} and \ref{BonusScheme}. These lemmas are

2319: phrased in the language associated to {\em teams}, the precise

2320: definition of which will

2321: also be given  in  Section \ref{teamSec}. 

2322: Many of the proofs involve global cancellation arguments

2323: based on the {\em Pincer Lemma}, which will be proved in the

2324: next section. 

2325: Intuitively speaking, a {\em team}  

2326: (typically denoted $\T$) is a  contiguous 

2327: region of $\|\T\|$ constant letters all of which 

2328: are to be consumed by a  fixed \lpl edge (the {\em reaper}). Notwithstanding 

2329: this intuition, it is preferable for 

2330: technical reasons  to define a team to be a set of pairs of colours $(\mu,\mu')\in\vecZ$, 

2331: where $\mu'$ is fixed and the different {\em members} of the team correspond to  

2332: different values of $\mu$. We write $(\mu,\mu')\in\T$ to denote 

2333: membership. Teams also have {\em virtual members}, denoted $(\mu,\mu') \vin \T$ (see 

2334: Definition \ref{Virtual}). There are less than $2\n$ teams (Lemma \ref{allIn}).

2335:  

2336: Each pair $(\mu,\mu')$ with $C_{(\mu,\mu')}(2)$ non-empty 

2337: is either a member or a  virtual member  of a team (Lemma \ref{allIn}).  

2338: There are {\em short} teams (Definition \ref{newTeams}) and long teams, 

2339:  of which some are {\em distinguished} (Lemma \ref{Aget2Lemma}). 

2340: There are four types of {\em genesis} of a team, (G1), (G2), (G3) and

2341: (G4) (see Subsection \ref{genesis}).  Teams of genesis (G3) have

2342: associated to them a pincer $\Pin_{\T}$ (Definition \ref{pl}) yielding

2343: an auxiliary set of colours

2344: $\subT$. There is also a set of colours $\chi_P(\T)$ associated to the

2345: time before the pincer $\Pin_{\T}$ comes into play.  For long,

2346: undistinguished teams, we also need to consider certain sets

2347: $\CT$ and $\chi_{\delta}(\T)$ of colours consumed in the past of $\T$ (see

2348: the proof of Lemma \ref{Aget2Lemma}). Such teams may

2349: also have  three sets of edges in $\partial\Delta$ associated to  

2350:  them: $\partial^\T$, $\down_1(\T)$ and  

2351: $\down_2(\T)$. An important feature of the definitions of 

2352:  $\partial^\T$ and $\down_1(\T)$ is that the sets associated 

2353:   to different teams are disjoint.   This disjointness is crucial 

2354:    in  the following proof, where we use the fact that the sum 

2355:     of their cardinalities is at most $\n$. Similarly, the disjointness of

2356: the sets $\chi_c(\T)$ is used to estimate the sum of their cardinalities by

2357: $\n$ and likewise for $\chi_{\delta}(\T)$ and $\chi_P(\T)$.

2358: 

2359: It is not necessarily true that the sets $\down_2(\T)$ are disjoint

2360:  for different teams, but we shall explain how to account for the

2361:  amount of `double-counting' that can occur (see Lemma

2362:  \ref{Aget2Lemma}).

2363:  

2364: Associated to every team one has  the time $t_1(\T)$ at 

2365: which the reaper starts consuming the team (see Subsection

2366: \ref{t1}). Teams genesis (G3) also have two

2367: earlier times $t_2(\T)$ and $t_3(\T)$ associated to them as well as an

2368: auxiliary set of edges $\QT$, the definitions of which

2369: are somewhat  technical (see Definition \ref{PincerDef} {\em et seq.}). 

2370: 

2371: In Section \ref{BonusScheme} we describe a {\em bonus scheme} that

2372: assigns a set  of extra edges, $\bonus(\T)$ to each team.  These

2373: bonuses are assigned so as  to ensure that $|\bonus (\T)|+\|\T\|$

2374:  dominates  the sum of 

2375:  the cardinalities of the sets $\cmm$ 

2376: associated to the  members and virtual members of $\T$. 

2377:  

2378: \smallskip 

2379: \noindent{\bf Proof of Proposition \ref{SummaryLemma}.} 

2380:  

2381:  

2382: Recall that   $A_4(S_0,\mu)$ is partitioned  into disjoint regions $C_{(\mu,\mu')}$ 

2383: which in turn are partitioned into $C_{(\mu,\mu')}(1)$ and 

2384: $C_{(\mu,\mu')}(2)$.  

2385:  

2386: Given any $\mu_1$ and $\mu_2$, at most one  ordering of $\{\mu_1,\mu_2\}$ can 

2387: arise in  $S_0$. Thus Lemma \ref{NoOfAdjacencies} 

2388: implies that there are less than $2\n$ pairs $(\mu,\mu')\in\vecZ$ with 

2389: $C_{(\mu,\mu')}\subset\bot(S_0)$  non-empty. 

2390: It follows immediately from this observation and Lemma \ref{C1Lemma} that 

2391: \[      \sum_{(\mu,\mu') \in {\mathcal Z}}|C_{(\mu,\mu')}(1)| \leq 2C_1\n . 

2392: \] 

2393:  

2394: 

2395: Lemma  \ref{Aget2Lemma} accounts for the set of  distinguished 

2396:  long 

2397: teams $\dlong$: 

2398: \[      \sum_{{\mathcal T} \in \dlong}\sum_{(\mu,\mu') \in  

2399: {\mathcal T}}|C_{(\mu,\mu')}(2)| \leq 6B\n(T_1+T_0).   \]  

2400: For all other teams $\T$ we rely on Lemma \ref{C1toTeamLength} which

2401: states

2402: \begin{equation}\label{goodEq} 

2403: \sum_{(\mu,\mu') \in \T \mbox{ \tiny or } (\mu,\mu') \vin \T} |C_{(\mu,\mu')}(2)| \le

2404: \|\T\|  + |\bonus(\T)| + B. 

2405: \end{equation} 

2406: We next consider the {\em genesis} of teams. All teams of genesis (G4)

2407: are short (Lemma \ref{G4lemma}). And by Definition \ref{newTeams} for

2408: the short teams $\T\in\Sigma$ we have

2409: \[      \sum_{{\T} \in \Sigma}\sum_{(\mu,\mu') \in  

2410: {\mathcal T}}|C_{(\mu,\mu')}(2)| \leq 2\ll\n + \sum_{\T\in\Sigma}

2411: \big(|\bonus(\T)| + B\big).   \]

2412: 

2413: Lemma \ref{TeamAgeLemma} tells us that for teams of genesis (G1) and

2414: (G2) we have 

2415: \[      \|\T\| \leq 2MC_4|\down_1(\T)| + |\partial^{\T}|,  \]

2416: whilst for teams of genesis (G3) we have

2417: \[      \|\T\| \leq 2MC_4\big{(}|\down_1(\T)| +|\QT|\big{)}+

2418: T_0\big{(}|\chi_P(\T)| + 1\big{)} +|\partial^{\T}| + \ll.       \]

2419: 

2420: Let $\Gthree$ denote the set of teams of genesis (G3) with $\QT$

2421: non-empty. In Definition \ref{down2} we break $\QT$ into pieces so

2422: that

2423: $$

2424: |\QT|\ =  t_3(\T) - t_2(\T)  

2425: + |\down_2(\T)|.

2426: $$  

2427: Making crucial use of the Pincer Lemma, in Corollary \ref{t1-t2Corr} we prove that

2428: $$

2429: \sum\limits_{\T \in \Gthree} 

2430:  t_3(\T) - t_2(\T)  \ \le \ 3\ttt \, \n, 

2431: $$ 

2432: and in   Corollary \ref{downbound} we prove that

2433: $$ 

2434: \sum\limits_{\T \in \Gthree}|\down_2(\T)|

2435: \leq (2 + 3\ttt + 5T_0)\n.

2436: $$

2437: This completes the estimate on $|\QT|$ and hence $\|\T\|$. 

2438: 

2439: Section 10 is dedicated to the proof of Proposition  \ref{BonusBound}, which states

2440: \[      \sum_{\text{\small{teams}}}|\bonus(\T)| \leq \big( \Bb \big)\n. \]

2441: 

2442: Adding all of these estimates and recalling that there are less than $2\n$ teams, we deduce:

2443: \[      \sum_{\mu \in S_0}|A_4(S_0,\mu)| \leq K_1 \n ,    \]

2444: where $K_1$ is

2445: $$ 

2446:  \AFourC.

2447: $$

2448: Thus the proposition is proved.

2449: \hfill$\square$

2450: \smallskip 

2451: \begin{remark} The stated value of the constant $K_1$ 

2452:   is an artifact of our proof: we

2453: have simplified the estimates at each stage for the sake

2454:  of clarity rather than trying to optimise the

2455: constants involved. Nevertheless, we have made some effort

2456:  to make the arguments constructive

2457: so as to prove that there exists an algorithm to calculate  

2458: the Dehn function of $F\rtimes_\phi\mathbb Z$ directly from $\phi$.  

2459: This is explained in some detail in  \cite{BGconstants}.

2460: \end{remark} 

2461: By a precisely analogous argument, we also have

2462: \bp \label{A2Prop}

2463: \[      \sum_{\mu \in S_0}|A_2(S_0,\mu)| \leq K_1 \n ,

2464:  \] 

2465: where $K_1$ is the constant defined prior to Proposition \ref{SummaryLemma}.

2466: \end{proposition} 

2467:  

2468:  

2469: \section{The pleasingly rapid consumption of colours} \label{ConstantSection} 

2470:  

2471:  

2472: This section contains the  cancellation lemmas 

2473: that we need to control the manner in which colours are consumed. 

2474: The key result in this direction is the {\em Pincer Lemma} (Theorem \ref{PincerLemma}). 

2475:    

2476: \subsection{The Buffer Lemma} 

2477:   

2478:  

2479: \begin{lemma}\label{BufferLemma}  

2480: Let $I\subset\bot(S)$ be an interval of edges labelled by constant letters, and 

2481: suppose that the colours $\mu_1(S)$ and $\mu_2(S)$ lie either side 

2482: of $I$, adjacent to it.  

2483:  Provided that the whole of $I$ does not die in $S$,  no 

2484: non-constant edge  coloured  $\mu_1$ will ever cancel with  

2485: a non-constant edge coloured $\mu_2$. 

2486: \end{lemma}  

2487:  

2488: \begin{proof} Suppose that the future of $I$ in $\top(S)$ 

2489: is a non-empty interval labelled $w_0$. If $\mu_1(S)$ is to the left of $I$, 

2490: then reading from the left beginning with the last non-constant 

2491: edge coloured $\mu_1$, on the naive top of $S$ we have an interval labelled 

2492: $x w_1 y$, 

2493: where $y$ is a non-constant letter coloured $\mu_2$ and  

2494: $w_1$ contains $w_0$ and perhaps some constant letters 

2495: from $\mu_1$ and $\mu_2$.  

2496:  

2497: Our conditioning of $\phi$ (Proposition \ref{power}) ensures that, for all non-constant letters $z$, the rightmost non-constant letter in $\phi^j(z)$ is the same for all $j \geq 1$.  Therefore, in order for there to ever be cancellation between non-constant letters coloured $\mu_1$ and $\mu_2$, we must have $x = y^{-1}$.

2498: Thus on $\top(S)$ there is an interval labelled  $xwx^{-1}$, where $w$ is the  

2499: (non-empty) free-reduction 

2500: of $w_1$. 

2501:  

2502: At times greater than $\height(S)$, the future of the interval that 

2503: we are considering will continue to have a core subarc labelled $xw_jx^{-1}$,  

2504: where $w_j$ is a conjugate of $w$ by a (possibly-empty) 

2505:  word in constant letters (unless the interval hits a singularity or 

2506: the boundary). In particular, no non-constant letters from $\mu_1$ 

2507: and $\mu_2$ can ever cancel each other.  

2508: \end{proof} 

2509:  

2510:  

2511: In the light of the Bounded Cancellation Lemma we deduce: 

2512:  

2513:  

2514: \begin{corollary} \label{BufferCorollary} 

2515: Let $I\subset\bot(S)$ be an interval of edges labelled by constant letters, and 

2516: suppose that the colours $\mu_1(S)$ and $\mu_2(S)$ lie either side 

2517: of $I$, adjacent to it. 

2518: If $|I|\ge B$ 

2519: then there is never any cancellation between non-constant letters in $\mu_1$ and $\mu_2$. 

2520: \end{corollary}  

2521:  

2522:  

2523: \subsection{The Two Colour Lemma} 

2524:   

2525: \bd \label{Neuters} 

2526: Suppose that $U$ and $V$ are positive words\footnote{i.e. none of their letters are 

2527:  inverses $a_j^{-1}$} 

2528:  and that for some $k>0$ the only  negative exponents occurring   in $\phi^k(UV^{-1})$  

2529: are on constant letters.  Then we say that $U$   

2530: {\em $\phi$-neuters   $V^{-1}$ in at most $k$ steps}.  

2531: \ed 

2532:  

2533: We shall also apply the term $\phi$-neuters to describe  the 

2534: cancellation between  colours $\mu(S), \mu'(S) \subseteq \bot(S)$ that are adjacent in  corridors of van Kampen diagrams,  

2535: and the following lemma remains valid in that context. 

2536:  

2537:  

2538: \begin{prop}[Two Colour Lemma] \label{TwoColourLemma} 

2539: There exists a constant $\tz$ depending only on $\phi$ so that 

2540:  for all positive words $U$ and $V$, if  

2541: $U$ $\phi$-neuters  $V^{-1}$ then it does so in at most $\tz$ steps.  

2542: \end{prop} 

2543:  

2544: \begin{proof} We express 

2545:  $V^{-1}$ as a product of  three subwords:  reading from the left of

2546: $V^{-1}$, the first subword ends with the last letter $y$ such that

2547: $\phi(y)$ contains a left-fast letter; the second subword follows the

2548: first and ends with the last non-constant letter in $V^{-1}$; the

2549: remainder of $V^{-1}$ consists entirely of constant letters.

2550:  

2551: Lemma \ref{C_0} tells us that the length of the first subword is less

2552: than $C_0$, and the proof of Lemma \ref{C1Lemma} provides a  bound of

2553: $C_1$ on the length of the second subword.

2554:  

2555: Now consider the freely reduced form of $\phi^k(UV^{-1})$, and let $v_k$ denote its subword 

2556: that begins with the first letter of negative exponent and ends with the final non-constant 

2557: letter. The argument just applied to $V^{-1}$ shows that $v_k$ has length less than $C_0+C_1$ 

2558: for all $k\ge 0$. 

2559:  

2560: Suppose that $U$ $\phi$-neuters $V^{-1}$ in exactly $N$ steps, let $\alpha_{N-1}$ 

2561: be the letter of $\phi^{N-1}(UV^{-1})$ that consumes the last  letter of $v_{N-1}$, and 

2562: let $\alpha_k$ be the ancestor of $\alpha_{N-1}$ in $\phi^k(UV^{-1})$. Write 

2563: $\phi^k(UV^{-1}) = w_k\alpha_k u_k v_k w_k'$. 

2564:  

2565: Lemma \ref{C_0} shows that $|u_k| < C_0$ for all $k< N$, and we have just argued  

2566: that $|v_k| < C_0+C_1$. Thus we obtain a bound  (independent of $U$ and $V$) on 

2567: the number of words $\alpha_k u_k v_k$ that arise as $k$ varies --- call this 

2568: number $\tz$. If $N$ were greater 

2569: than $\tz$, then some configuration $\alpha_k u_k v_k$ with $v_k$ non-empty would recur. But 

2570: this is nonsense, because once there is this repetition, the words $v_k$ will continue to repeat, and thus $V^{-1}$ will never be $\phi$-neutered, contrary to assumption.  

2571: \end{proof} 

2572: 

2573: \begin{corollary} There exists a constant $\tz'$, depending only on $\phi$, with the following 

2574: property: if $U$ and $V$ are positive words, $V$ begins with a non-constant letter 

2575: and $\phi^k(UV^{-1})$ is positive for some $k>0$, then the least such $k$ 

2576: is less than $\tz'$. 

2577: \end{corollary}  

2578:  

2579: \begin{proof} The preceding lemma provides an upper bound on the least integer $N$ such that 

2580: $\phi^N(UV^{-1})$ contains no non-constant  letters with negative exponent. Up to 

2581: this point, the rightmost non-constant letter in $\phi^k(UV^{-1})$ may have been spawning 

2582: constant letters to its right, and thus  $\phi^k(UV^{-1})$ may have a terminal segment consisting 

2583: of constant letters. Since the rightmost non-constant letter of $\phi^k(V^{-1})$ does 

2584: not vary with $k$ when $k<N$ (by Proposition \ref{power}),  the length of this segment  

2585: grows at  a constant rate ($<M$) during each application of $\phi$. Similarly, its length 

2586: changes at a constant rate after time $N$, decreasing until it is eventually cancelled. 

2587:  

2588: Since $N\le \tz$, this segment of constant letters has length less than $M\tz$ 

2589: at time $N$, and hence is cancelled entirely before time $T_0(M+1)$. 

2590: \end{proof}  

2591:  

2592: 

2593: 

2594: \subsection{The disappearance of colours: Pincers and implosions} 

2595:  

2596: In this subsection we turn our attention to the detailed study of how non-adjacent colours 

2597: along a corridor in $\Delta$ can come together solely as a result of the mutual 

2598: annihilation of the intervening colours. Such an event determines a {\em pincer} (Figure \ref{PincerPic}), which is defined as follows. 

2599: 

2600: 

2601: \begin{figure}[htbp] 

2602: \begin{center} 

2603: 

2604: \input{Pincerpic.eps_t} 

2605: 

2606: \caption{A pincer.} 

2607: \label{PincerPic} 

2608: \end{center} 

2609: \end{figure} 

2610: 

2611:  

2612: \begin{definition}\label{pincerDef} 

2613: Consider a pair of paths $p_1, p_2$ in $\mathcal F \subseteq \Delta$  

2614: tracing the histories of $2$ non-constant edges $e_1, e_2$ that cancel in a corridor $S_t$.   

2615: Let $\mu_i$ denote the colour of the 2-cells along  $p_i$. 

2616: Suppose that at time $\tau_0$ these paths lie in a common corridor $S_b$.

2617: Under these circumstances, we define 

2618: the {\em pincer}  

2619: $\Pin=\Pin( p_1, p_2, \tau_0 )$ to be the subdiagram of $\Delta$ enclosed by the chains of  

2620: $2$-cells along $p_1$ and $p_2$, and the chain of $2$-cells connecting them in $S_b$. 

2621:   

2622: When it creates a desirable emphasis, we shall write $S_b(\Pin)$ and $S_t(\Pin)$ 

2623: in place of $S_b$ and $S_t$. 

2624:   

2625: We define $S_\Pin$ to be the earliest corridor of the pincer in which $\mu_1(S_\Pin)$ and 

2626: $\mu_2(S_\Pin)$ are adjacent. 

2627: We define  $\tilde{\chi}(\Pin)$ to be the set of colours $\mu\notin\{\mu_1,\mu_2\}$ such 

2628: that there is a 2-cell in $\Pin$ coloured $\mu$. And we define%\footnote{This

2629: %definition has attracted significant social commentary \cite{pi}.}  

2630: $$\life(\Pin) = \time(S_\Pi)-\time(S_b). 

2631: $$ 

2632: \end{definition} 

2633:  

2634: \begin{proposition}[Unnested Pincer Lemma]\label{prePincerLemma} 

2635: There  exists a constant $\hat{T_1}$, depending only on $\phi$, such that for any pincer $\Pin$ 

2636: \[      \life(\Pin) \leq \hat{T_1}(1+ |\tilde{\chi}(\Pin)|).        \] 

2637: \end{proposition} 

2638:  

2639: Fix a pincer $\Pin$ and assume $\life(\Pin)\neq 0$.

2640:  The idea of the proof of Proposition \ref{prePincerLemma} 

2641:  is as follows:  we shall identify a constant $\hat{T_1}$ and argue 

2642:  that if none of the colours $\mu\in\tilde{\chi}(\Pin)$ were consumed entirely  

2643:   by  $\time(S_b) + \hat{T_1}$, 

2644: the situation reached would be so stable  that no colours could be consumed in $\Pin$ at 

2645: subsequent times,  

2646:  contradicting the fact that all but $\mu_1$ and $\mu_2$ must be consumed by $\time(S_\Pi)$. 

2647:  

2648: With this approach in mind, we make the following definition: 

2649:  

2650: \begin{definition} Let $p$ be a positive integer.  

2651:  A {\em $p$-implosive array} of colours  in a corridor $S$ 

2652: is an ordered tuple $A(S)=[\nu_0(S),\dots,\nu_r(S)]$, with $r>1$, 

2653:  such that: 

2654: \begin{enumerate}

2655: \item each  pair of colours $\{\nu_j, 

2656: \nu_{j+1}\}$ is {\em essentially adjacent} in  $S$, meaning that there are no

2657: non-constant edges of any other colour separating $\nu_j(S)$ from $\nu_{j+1}(S)$;

2658: \item in each of the corridors $S=S^1,S^2,\dots, S^{p}$ in the future of $S$, 

2659:  every $\nu_j(S^i)$ contains a non-constant edge;

2660: \item in $S^p$,  {\em either} a non-constant edge coloured $\nu_0$ cancels a non-constant edge coloured $\nu_r$ 

2661: (and hence the colours $\nu_j$ with $j=1,\dots,r-1$ are consumed 

2662: entirely), {\em or else} all of the non-constant letters in $\nu_j(S^p)$, for $j=1,\dots, r-1$,

2663: are cancelled in $S^p$ by edges from one of the colours of the array, while

2664: $\nu_0(S^p)$ and $\nu_r(S^p)$ contain non-constant letters that survive in

2665: the free-reduction of the naive future of the interval $\nu_0(S^p)\dots\nu_r(S^p)\subset\bot(S^p)$

2666: (but  may nevertheless be cancelled in $S^p$ by edges from colours external to the array). 

2667: \end{enumerate}

2668: Arrays satisfying the first of the conditions in (3) are said to be of Type I, and those

2669: satisfying the second condition are said to be of Type II. (These types are not

2670: mutually exclusive.)

2671: 

2672: The {\em residual block} of an array of Type II is the interval of constant edges  between 

2673: the rightmost non-constant letter of $\nu_0$ and 

2674: the leftmost non-constant letter of $\nu_r$ in 

2675: the free reduction of the naive future of  $\nu_0(S^p)\dots\nu_r(S^p)$.

2676: The {\em enduring block} of the array  is the set of constant edges in 

2677: $ \bot(S)$ that have a future  in the residual block.

2678: 

2679:  Note that there may exist {\em unnamed colours} between

2680: $\nu_j(S)$ and $\nu_{j+1}(S)$ consisting entirely of constant edges.

2681: \end{definition} 

2682:  

2683: \begin{remarks}\label{subarray} Let $[\nu_0(S),\dots,\nu_r(S)]$ be a $p$-implosive array.

2684: 

2685: \smallskip

2686: (1) Any implosive subarray of $[\nu_0(S),\dots,\nu_r(S)]$ is $p$-implosive (same $p$).

2687: 

2688: (2) If an edge of $\nu_i$ cancels 

2689: with an edge of $\nu_j$ and $j-i>1$, then this cancellation 

2690: can only take place in $S^p$. If the edges cancelling are non-constant, 

2691:  then the subarray $[\nu_i(S),\dots, \nu_j(S)]$ is   $p$-implosive of Type I. 

2692: 

2693: 

2694: (3) Given $x,y,w\in F$,

2695: if the freely reduced words representing $x, y$ and $\phi(xwy)$ consist only 

2696: of constant letters, then

2697: so does the reduced form of $w$, since the subgroup generated by the constant

2698: letters is invariant under $\phi^{\pm 1}$. It follows that the residual block

2699: of any array of Type II contains edges from at most two of the colours $\nu_j$, and if

2700: there are two colours they must be essentially adjacent, i.e. $\nu_j(S^p), \nu_{j+1}(S^p)$.

2701: 

2702: (4) For the same reason, the enduring block of an implosive

2703: array of Type II is an interval involving at most two of the $\nu_j$, and if

2704: there are two such colours then they must be essentially adjacent.

2705: \iffalse

2706: (5) The concept ``essentially adjacent" admits the possibility that there may be

2707: several {\em unnamed colours} between $\mu_j(S)$ and $\mu_{j+1}(S)$, provided

2708: that all of the edges in these unnamed colours are constant. Note that the

2709: count defining the length of an array includes the edges in these unnamed colours.

2710: \fi

2711: \end{remarks}  

2712:    

2713: 

2714: \begin{lemma}\label{haveImp}  The ordered list of colours along each corridor before 

2715: $\time(S_\Pi)$ in a pincer $\Pi$ must contain an implosive array.

2716: \end{lemma}

2717: 

2718: \begin{proof} At the top of the pincer there is cancellation between non-constant

2719: edges. Lemma \ref{BufferLemma} tells us that before $\time(S_\Pi)$ the colours

2720: of these edges must have been separated by a non-constant letter of a different

2721: colour, hence the list of non-constant colours along the bottom of $S_\Pi$ is a

2722: 1-implosive array. This same list of colours defines an implosive array at

2723: each earlier time in the pincer until, going backwards in time, further non-constant

2724: colours appear. Suppose $\mu$ has non-constant letters in $\Pi$ at

2725: time $t$ but not time $t+1$. Let $\nu_0$ be  the first colour to the  left of $\mu$ that

2726: contains non-constant letters at time $t+1$, and let $\nu_r$ be the first such colour

2727: to the right. If $S_t$ is the corridor at time $t$, then the list of essentially-adjacent

2728: non-constant colours $[\nu_0(S_t),\dots,\mu(S_t),\dots,\nu_r(S_t)]$ is a

2729: 1-implosive array. And $[\nu_0(S_{t'}),\dots,\mu(S_{t'}),\dots,\nu_r(S_{t'})]$

2730: is a $(t'-t+1)$-implosive array for each earlier time $t'$ until (going backwards in

2731: time) either further non-constant colours appear or else we reach the bottom of the

2732: pincer.

2733: \end{proof}

2734: 

2735:  If, further to the above lemma, we can argue that there is a constant $\hat{T_1}$ such

2736: that each corridor before $\time(S_\Pi)$ contains a

2737: $p$-implosive array with $p\le\hat{T_1}$, then we will know that at least one of the colours 

2738: from $\tilde\chi(\P)$ is {\em essentially consumed} (i.e. comes to consist of constant

2739: edges only)

2740:  during each interval of $\hat{T_1}$ units in time during 

2741: the lifetime of the pincer. Thus Proposition \ref{prePincerLemma} 

2742:  is an immediate consequence of the following result, which will be proved

2743: in (\ref{RIP}).

2744: 

2745: 

2746: %\theoremstyle{remark} 

2747: %\newtheorem{RIproof}[theorem]{Proof of Regular Implosions}

2748:  

2749: \begin{proposition}[Regular Implosions]\label{implosion} 

2750: There is a constant $\hat{T_1}$ depending only on $\phi$ 

2751: such that every implosive array in any minimal area diagram $\Delta$ is $p$-implosive 

2752: for some $p\le\hat{T_1}$. 

2753: \end{proposition} 

2754:  

2755: The first restriction to note concerning implosive arrays is this: 

2756:  

2757: \begin{lemma} \label{OnlyBColours} 

2758: If $[\nu_0(S),\dots,\nu_r(S)]$ is implosive   %with $u$ unnamed colours.

2759: of Type I, then $r\le B$. If

2760: it is implosive of Type II, then $r< 2B$.

2761: \end{lemma} 

2762:  

2763: \begin{proof} In Type I arrays,  the interval 

2764:  $\nu_1(S^{p})\dots\nu_{r-1}(S^p)\subset\bot(S^p)$

2765:   is to die in $S^p$, so $r-1<B$ by

2766:   the Bounded Cancellation Lemma.

2767:   For Type II arrays, one applies

2768: the same argument to the intervals 

2769:  joining $\nu_0(S^p)$ and $\nu_r(S^p)$ to the residual block of constant letters. 

2770: \end{proof} 

2771:  

2772: \begin{remark} \label{shortisenough} 

2773: In the light of Lemma \ref{OnlyBColours}, an 

2774: obvious finiteness argument would provide the bound required for 

2775: Lemma \ref{implosion} if we were willing 

2776: to restrict ourselves to implosive arrays  with

2777:  a uniform bound on their

2778: length.

2779: %the  length of the $\nu_j(S)$ and the length of the unnamed colours present in the array.

2780: \iffalse

2781:  Moreover, in the case of Type II arrays, we can exclude

2782: the enduring block from the calculation of length, since it plays no role in any

2783: cancellation.

2784: \fi

2785: Motivated by this observation, we seek to prove  that every implosive array contains an

2786: implosive sub-array that is uniformly {\em short}.

2787: \end{remark} 

2788:  

2789: \smallskip 

2790:  

2791: In order to identify a suitable notion of {\em short}, 

2792:  we need to consider a further decomposition 

2793: of the colours $\nu_j(S_b)$ in a $p$-implosive array $[\nu_0(S_b),\dots,\nu_r(S_b)]$. 

2794:  

2795: Previously (Subsection \ref{chromatic}) we partitioned each colour 

2796: $\nu_j(S_b)$ into five intervals $A_1(S_b,\nu_j),\dots, 

2797: A_5(S_b,\nu_j)$ and then further decomposed $A_4$ into subintervals 

2798: $C_{(\nu_j,\nu')}(1)$ and $C_{(\nu_j,\nu')}(2)$ according to the 

2799: colours of the edges that were going to consume these subintervals in 

2800: the future. There is a corresponding decomposition of $A_2$ into 

2801: intervals which we denote $C^2_{(\nu_j,\nu')}(1)$ and 

2802: $C^2_{(\nu_j,\nu')}(2)$ (where $\nu'$ is now to the left of $\nu_j$ in 

2803: $S_b$).  

2804:  

2805: Adapting to our new focus, we now define $R_j(S_b)=A_5(\nu_1,S_b)\cup 

2806: C_{(\nu_j,\nu_{j+1})}(1)$, and $L_j(S_b)=A_1(\nu_1,S_b)\cup 

2807: C^2_{(\nu_j,\nu_{j+1})}(1)$. We also define  $C_j^R(S_b)$ to be 

2808: $C_{(\nu_j,\nu_{j-1})}(2)$ minus any edges from the excluded block, and $C_j^L(S_b)$ to 

2809: be $C^2_{(\nu_j,\nu_{j-1})}(2)$ minus any edges from the excluded block.

2810: Thus we obtain a decomposition of 

2811: $\nu_j(S_b)$ into five intervals (see Figure \ref{PincerDecomp})  

2812: $$ 

2813:  L_j(S_b),\ C_j^L(S_b), \ \mess(S_b,\nu_j),\    C_j^R(S_b),\ R_j(S_b) 

2814: $$ 

2815: where $\mess(S_b,\nu_j)$ contains the edges 

2816: whose preferred future dies at the time of  implosion together

2817: with edges from the excluded block\footnote{At this point the reader may 

2818: find it helpful to recall that only arrays of Type II have excluded

2819: blocks, and such a block is either contained in a single colour,

2820: or in adjacent colours $\nu_j(S_b)\cup\nu_{j+1}(S_b)$ with

2821: the intervening intervals $R_j(S_b) \dots L_{j+1}(S_b)$ empty.}.

2822: 

2823: 

2824: \begin{figure}[htbp] 

2825: \begin{center} 

2826: 

2827: \input{PincDecomp.eps_t} 

2828: 

2829: \caption{The decomposition of the colour $\nu_j$} 

2830: \label{PincerDecomp} 

2831: \end{center} 

2832: \end{figure} 

2833: 

2834:  

2835: The terminal colours in our array, $\nu_0$ and $\nu_r$, play a special 

2836: role. This is reflected in the fact that we shall only need to consider 

2837: the segment of $\nu_0$ from its right end 

2838:  up to and including the edge one to the left of $\mess(S_b,\nu_0)$. And 

2839:  in $\nu_r$ we shall only need to consider the segment from its left end 

2840: up to and including the edge one to the right of $\mess(S_b,\nu_r)$. 

2841:  We write $\mathcal L(\nu_j,S_b)$ and $\mathcal R(\nu_j,S_b)$, 

2842:   respectively, to denote these sub-intervals of $\nu_j(S_b)$. 

2843:   

2844:  \begin{definition} The length of $A(S)=[\nu_0(S),\dots,\nu_r(S)]$, written $\|A(S)\|$,

2845:  is the number of edges in the interval

2846: $\mathcal L(\nu_0,S)\dots\mathcal R(\nu_r,S)\subset \bot(S)$. (Note that $\|A(S)\|$ takes account of the

2847: unnamed colours.)

2848: \end{definition}

2849: 

2850: In keeping with the notation in the definition of $p$-implosive, we shall 

2851: write $S^t$ for the corridor $t$ steps into the future of $S_b$; in particular  $S^0=S_b$ and

2852: each $\nu_j$ with $j=1,\dots,r-1$ essentially vanishes in $S^p$. 

2853:  

2854: By definition, no preferred future of any 

2855:  edge in  $\mess(\nu_j,S_b)$ is cancelled 

2856: before $S^p$. Hence these intervals do not shrink in length before 

2857: that time, and as in the proof of Lemma \ref{OnlyBColours} we can use

2858: the Bounded Cancellation Lemma to bound the sum of their

2859: lengths:

2860: 

2861: \begin{lemma} \label{2Bcols}

2862: After excluding the edges of the enduring block, the sum of the lengths of the 

2863: intervals  $\mess(\nu_j,S_b)$ is   at most $2B$. 

2864: \end{lemma}

2865: 

2866: Combining this estimate with the bounds from Lemmas \ref{C_0} 

2867: and \ref{C1Lemma}, we deduce that for $j=1,\dots,r-1$  

2868:  

2869: \[   |\nu_j(S_b)| \le    |C^L_j(S_b)| + |C^R_j(S_b)|  + 2C_0 + 2C_1 + 2B +\mathcal E_j,  \] 

2870: where $\mathcal E_j$ is the number of edges from the excluded block coloured $\nu_j$.

2871: 

2872: Similarly, 

2873: $$ 

2874: |\L(\nu_0,S_b)|\le   

2875: |C^R_0(S_b)| + C_0 + C_1 + B +\mathcal E_0  

2876: $$ 

2877: and  

2878: $$ 

2879: |\R(\nu_{r},S_b)|\le  

2880: |C^L_{r}(S_b)| +C_0 + C_1 + B + \mathcal E_r. 

2881: $$ 

2882: This motivates us to define an array of colours $[\nu_0(S),\dots,\nu_r(S)]$ 

2883: to be {\em very short} if for $j=1,\dots,r-1$ we have 

2884: \[      |\nu_j(S)|  \leq 2C_0 + 2C_1 + 5B  + 1,    \]  

2885: and 

2886: \[      |\L(\nu_0,S)|  \leq C_0 + C_1 + 5B  + 1,    \] 

2887:  and  \[      |\R(\nu_r,S)|  \leq C_0 + C_1 + 5B + 1,        \] 

2888: and for $j=0,\dots,r-1$ the interval formed by the unnamed colours between $\nu_j(S)$

2889: and $\nu_{j+1}(S)$ has total length at most $B$.

2890: 

2891: An implosive array is said to be {\em short} if it satisfies the weaker 

2892: inequalities obtained by increasing  each of these bounds by $2B{\tz}$.   

2893:  

2894: \begin{lemma} \label{vshort} Let  $A=[\nu_0(S^0),\dots,\nu_r(S^0)]$ be  a 

2895: $p$-implosive array 

2896: with $p\ge \tz$. 

2897: \begin{enumerate}

2898: \item If $[\nu_0(S^\tz),\dots,\nu_r(S^\tz)]$ is very short, 

2899: then $A$ is short. 

2900: \item If $A$ is short, then $\|A\|\le 2B(2C_0+2C_1 + 5B +1+2BT_0) +2B^2(1+2T_0).$

2901: \end{enumerate}

2902: \end{lemma} 

2903:  

2904: \begin{proof} Item (1) is an immediate consequence of the Bounded Cancellation 

2905: Lemma \ref{BCL}. The (crude) bound in (2) is an immediate consequence of Lemma \ref{2Bcols}

2906: and the inequalities in the definition of {\em short}; the first summand is an estimate

2907: on the sum of the lengths of the named colours, and the second summand accounts for

2908: the unnamed colours.

2909: \end{proof} 

2910:    

2911: The following 

2912: lemma is the key step in  the proof of Proposition \ref{prePincerLemma}. 

2913:  

2914: \begin{lemma} \label{shorty} 

2915: If $A(S^0)=[\nu_0(S^0),\dots,\nu_r(S^0)]$

2916:  is a $p$-implosive array,

2917: then at least one of the following statements is true:

2918: \begin{enumerate}

2919: \item $p\le 2T_0$; 

2920: \item $A(S^0)$ is short;

2921: \item  $p > 2T_0$ and $A(S^{T_0})$ contains an 

2922: implosive sub-array $[\nu_k(S^\tz),\dots,\nu_l(S^\tz)]$ that is very short.

2923: \end{enumerate}

2924: \end{lemma} 

2925:  

2926: \begin{proof} Assume $p > 2T_0$ and that

2927:  $[\nu_0(S^0),\dots,\nu_r(S^0)]$ is not short. We claim that there is a 

2928:  block of at least $B+1$ constant letters in the interval determined by the

2929: array

2930:  $\mathcal L(\nu_0,S^{T_0})\dots\mathcal L(\nu_r,S^{T_0})$.

2931: Indeed, by definition, if an array

2932:  is not short then either one of the $\mathcal E_j$ has

2933: length at least $B+1$, or one of the

2934:  blocks of unnamed colours has length at

2935: least $B(2T_0+1)+1$, or

2936: else at least one of the intervals of 

2937: constant letters $C^L_j(S^{0})$ 

2938: or $C^R_j(S^{0})$ has length at least $B(T_0+1)+1$. 

2939: In the first case, since  $\mathcal E_j$  is in

2940: the excluded block, none of its edges are cancelled

2941:  before the moment of implosion, and

2942: hence it contributes a block of at least $ B+1$ constant

2943:  letters to $A(S^{T_0})$; in the second

2944: case, the Bounded Cancellation 

2945: Lemma assures us that the length of the appropriate block of unnamed colours  can decrease by 

2946: at most $2B$ at each step before the implosion of the array, 

2947: and hence it still contributes

2948: a block of  at least $ B+1$ constant edges to $A(S^{T_0})$;

2949: and similarly, in the third case,

2950:  $C^{\ast}_j(S^{0})$  can decrease by 

2951: at most $B$ at each step before the implosion of the array.

2952: 

2953: Let $\beta$ be a block of at least $ B+1$ constant edges in $A(S^{T_0})$ with non-constant

2954: edges $e_l$ and $e_{\rho}$ immediately to its left and right, respectively. 

2955: \iffalse

2956: The Two Colour Lemma 

2957: \ref{TwocolourLemma} assures us that the letters labelling the non-constant edges adjacent

2958: to the future of $\beta$ do not change between  $\time(S^{T_0})$ and $\time(S^p)$,

2959: the moment of implosion. 

2960: \fi

2961: The Buffer Lemma \ref{BufferLemma} assures us that the

2962: non-constant edges in the future of $e_l$ will never interact with the non-constant edges in the

2963: future of $e_{\rho}$. Thus at least one of $e_l$ or $e_{\rho}$ must be {\em stabbed in the back}, i.e. 

2964: its entire non-constant future must be consumed by edges on its own side of $\beta$. Suppose,

2965: for ease of notation, that it is $e_l$ and let $\nu_i$ be the colour of $e_l$. We claim that if $\nu_k$

2966: is the colour of the letter that ultimately consumes $e_l$, then $k\le i-2$.

2967: 

2968: We shall derive a contradiction

2969: from the assumption that the edge which ultimately

2970:  consumes  $e_l$ is  coloured $\nu_{i-1}$.

2971: There are two cases to

2972: consider according to whether $e_{\rho}$ is also coloured $\nu_i$. If it is, then we consider the

2973: word $V$ labelling the arc of $\bot(S^0)$

2974:  from the left end of $\nu_i(S^0)$ to the past of

2975: $e_l$; the consumption

2976: of the non-constant future of 

2977: $e_l$ completes the $\phi$-neutering  of $V$

2978:  by the word labelling $\nu_{i-1}(S^0)$, 

2979: in particular this neutering will have taken more than $T_0$ steps in time, contradicting

2980: the Two Colour Lemma \ref{TwoColourLemma}. If $e_{\rho}$ is not coloured $\nu_i$, then the

2981: consumption of the non-constant future of 

2982: $e_l$ results in a new essential adjacency of colours and hence can only be complete

2983:  at the moment of

2984: implosion, i.e. $\time(S^p)$. But this consumption constitutes the neutering of $\nu_i(S^{T_0})$

2985: by $\nu_{i-1}(S^{T_0})$, and according to the Two Colour Lemma this neutering 

2986: must be accomplished in at most $T_0$ units of time. Thus $p\le 2T_0$,

2987: contrary to our hypothesis. 

2988: 

2989: \def\kill{\!\!\!\searrow\!}

2990: 

2991: Thus we have proved that the edge which ultimately consumes $e_l$ is coloured

2992: $\nu_k$ where $k\le i-2$. Under these circumstances (or the 

2993: symmetric situation with $e_{\rho}$ in place of $e_l$) we say that {\em $\nu_k$

2994: neuters $\nu_i$ from behind} and write  $\nu_k \kill \nu_i$.

2995:  

2996: %***  We draw a `cancellation diagram' FIGURE to SHOW WHAT HAS HAPPENED... *** 

2997:  

2998: \begin{figure}[htbp] 

2999: \begin{center} 

3000:   

3001: \input{ArcDiag.eps_t} 

3002:   

3003: \caption{The nesting associated to $\kill$} 

3004: \label{figure:Arc} 

3005: \end{center} 

3006: \end{figure} 

3007:  

3008: \medskip 

3009:  

3010: There is a natural {\em nesting} among the $\kill$-related pairs of colours from the

3011: array: 

3012: % if $\{\nu_{k_1},\nu_{i_1}\}$ and $\{\nu_{k_2},\nu_{i_2}\}$ are two 

3013: % such pairs, 

3014:  $(\nu_{k_1},\nu_{j_1}) < (\nu_{k_2},\nu_{j_2})$ if $\nu_{k_1}$ and 

3015: $\nu_{j_1}$   

3016: both lie between $\nu_{k_2}$ and $\nu_{j_2}$ in $S^0$. See Figure 

3017: \ref{figure:Arc}. 

3018:  

3019: We focus our attention on an innermost (i.e. minimal) 

3020: pair with  $\nu_k\kill\nu_i$.  By definition $|k-i|\ge 2$. If there were

3021: a block of at least $ B+1$ constant letters between the closest non-constant

3022: letters of $\nu_k(S^{T_0})$ and $\nu_i(S^{T_0})$, 

3023: then the preceding argument 

3024: would yield a neutering from behind that contradicted the innermost nature

3025: of $\nu_k\kill\nu_i$. Thus  

3026: $[\nu_j(S^\tz),\dots,\nu_k(S^\tz)]$ is a very short array, and we are done.  

3027: \end{proof}

3028: 

3029: 

3030: 

3031: \begin{RIproof}\label{RIP}{\em Proof of Regular Implosions (Prop.\ref{implosion}):} Given the bound in Lemma \ref{vshort}(2), an obvious finiteness

3032: argument provides a constant $\tau$ such that every short implosive array is $p$-implosive

3033: with $p\le \tau$. And the same bound applies to implosive arrays that contain a short

3034: sub-array (Remark \ref{subarray}(1)). So in the light of Lemmas  \ref{shorty} and   \ref{vshort}(1),

3035: it suffices to let $\hat{T_1} = \max\{2T_0, \tau\}$.\hfill $\square$

3036: \end{RIproof}

3037:  

3038: \subsection{Super-Buffers}

3039: 

3040: 

3041: In this subsection we prove an important cancellation lemma based on

3042: Proposition \ref{prePincerLemma}, this lemma involves the following

3043: constant.

3044: 

3045: \begin{definition} \label{T1'Lemma}

3046: We fix an integer $T_1'$ such  that one gets repetitions  in all $T_1'$-long subsequences of $5$-tuples of reduced words

3047: \[	U_k:=\Big( u_{k,1}, u_{k,2}, u_{k,3}, u_{k,4}, u_{k,5}	 \Big) \ \ \ \ k=1,2,\dots \]

3048: with $|u_{k,1}| $ and $|u_{k,1}| $ at most $ C_0+ C_1 + 2B +1$, while

3049:  $|u_2^k|$ and $ |u_4^k|$ are at most $ C_0 + C_1$, and $|u_3^k| \leq 4B+1$. That is,

3050: for some $t_1\le t_2\leq T_1'$ and 

3051: \[	\Big( u_{t_1,1}, u_{t_1,2}, u_{t_1,3}, u_{t_1,4}, u_{t_1,5}	 \Big) = \Big( u_{t_2,1},  u_{t_2,2}, u_{t_2,3}, u_{t_2,4}, u_{t_2,5}	 \Big) .	\]

3052: \end{definition}

3053: 

3054: 

3055: \newtheorem{stipulation}[theorem]{Stipulation} 

3056: 

3057: 

3058: \begin{stipulation} Assume $T_1' \ge \hat{T_1}$.

3059: \end{stipulation}

3060:  

3061: The cancellation lemma we need is most easily phrased in terms of 

3062: colours of subwords, which we define as follows, keeping firmly in mind

3063: the example of a stack of partial corridors excised from the interior of a van Kampen

3064: diagram, retaining their memory of the colours to which the edges belong.

3065: 

3066: 

3067: We have a word $W$ with a  decomposition into preferred subwords

3068: $V = V_1 V_2

3069: \cdots V_k$, where each $V_i$ is either positive or negative;

3070: we think of these subwords as having colours $\mu_1, \ldots \mu_k$.

3071: Take the freely reduced words $\phi(V_i)$, concatenate them, then

3072: cancel to form a freely reduced word. There is some freedom in the

3073: choice of cancellation scheme, as in the folding of corridors, but we fix

3074: a choice, thus assigning to each letter of the freely reduced form of

3075: $\phi(V)$ the colour  $\mu_i$ of its ancestor. We repeat this process,

3076: thus assigning colours to the letters in the reduced form of $\phi^k(V)$ for

3077: each integer $k>0$.

3078: 

3079: The process that we have just described is an algebraic description of

3080: a choice of  minimal area van Kampen diagram for $t^{-k}Vt^k\phi^k(V)^{-1}$.

3081: Thus the following lemma is a comment on the form of  such

3082: diagrams.  

3083: 

3084: 

3085: \begin{proposition} \label{NoDoubleNeuter}  

3086: Let $V=V_1V_2V_3$ be a concatentation of words (coloured $\nu_1, \nu_2, \nu_3$)

3087: each of which is either positive or negative.

3088: If $W$ is a subword of the reduced

3089: form of  $\phi^{T_1'}(V)$ and   $W$ has a non-constant

3090: letter coloured $\nu_i$ for each $i\in\{1,2,3\}$, 

3091: then for all $k \geq 0$ there are

3092: non-constant letters  in $\phi^k(W)$ coloured $\nu_2$.

3093: \end{proposition}

3094: 

3095: \begin{proof} Let $\nu_i(W)$ denote the subword of $W$

3096: coloured $\nu_i$, and let $\nu_i^j$ denote  the maximal subword coloured $\nu_i$ in

3097: (the reduced word representing) $\phi^i(V_1V_2V_3)$ .

3098: Note that $\nu_2(W)=\nu_2^{T_1'}$, and more generally $\nu_2^{T_1'+j}$

3099: is the maximal word  in $\phi^j(W)$ coloured $\nu_2$.

3100: 

3101: Fix $k>T_1'$ and consider the diagram formed by the stack of corridors

3102: described prior to the proposition. The bottom

3103: of the first corridor is labelled $V$, and we regard it as being divided into

3104: three coloured intervals according to the decomposition $V_1V_2V_3$.

3105: Since $\nu_2(W)$ contains non-constant letters and $T_1'>\hat{T_1}$, 

3106: the array formed by these colours is not implosive (Proposition \ref{prePincerLemma}),

3107: and hence  

3108: $\nu_1(W)$ and $\nu_3(W)$ will never essentially consume $\nu_2(W)$.

3109: However, the proposition is not yet proved because there remains

3110: the possibility

3111: that  $\nu_2$ may essentially vanish because it  neuters $\nu_1(W)$, say, and is then

3112: neutered by $\nu_3(W)$. We proceed under this assumption, seeking

3113: a contradiction. (The case where the roles of 

3114: $\nu_1$ and $\nu_3$ are reversed is entirely similar.)

3115: 

3116: For each $1 \leq i \leq T_1'$, we have

3117: $\phi^i(V_1V_2V_3)=\nu_1^i, \nu_2^i$ and $\nu_3^i$.

3118: Write $\nu_2^i\equiv V^i(1)  V^i(2)  

3119: V^i(3)$, where $V^i(1)$ ends with last  non-constant letter in

3120: $\nu_2^i$ whose entire non-constant future is eventually consumed by

3121: letters coloured $\nu_1$, and $V^i(3)$ begins with the leftmost

3122: non-constant letter whose entire non-constant future is  

3123: eventually consumed by letters coloured $\nu_3$.

3124: Lemmas \ref{C_0} and \ref{C1Lemma} tell us that $V^i(1)$ and $V^i(3)$ have

3125: length at most

3126: $C_0 + C_1$.  

3127: 

3128: \noindent{\em Claim:} $V^i(2)$  contains exactly one non-constant edge

3129: and has length no more than $4B+1$.

3130: 

3131: We are assuming that $\nu_2(W)$ neuters $\nu_1(W)$. Consider the

3132: (non-constant) edge $\e_i$ in

3133: $\nu_2^i$ that will eventually consume the final non-constant edge in

3134: $\nu_1(W)$. Note that $\e_i$ is  the

3135: leftmost non-constant edge in $V^i(2)$. Moreover, we are assuming 

3136: that  $\nu_3(W)$ ultimately neuters

3137: $\nu_2(W)$, so in particular  it consumes the entire future of

3138: any edge to the right of $\e_i$, which

3139: forces $\e_i$ to be the rightmost non-constant edge in $V^i(2)$. The Buffer

3140: Lemma tells us that $\e_i$ must lie within $2B$ of both ends of $V^i(2)$,

3141: and hence the claim is proved.

3142: 

3143: Looking to the left

3144: of $V^i(1)$, we now consider the  subword    $L^i$ of $\nu_1^i$  

3145: that begins with the leftmost  non-constant edge in the future of

3146: which there is a non-constant letter that cancels with a letter

3147: coloured $\nu_2$. And looking to the right of  $V^i(3)$, we consider

3148: the subword that ends with the rightmost non-constant letter  in the future of

3149: which there is a non-constant letter that cancels with a letter

3150: coloured $\nu_2$. 

3151: any of whose non-constant future cancels

3152: with an edge painted $\nu_2$.  As in previous arguments, The Buffer Lemma and

3153: Lemmas \ref{C_0}, \ref{C1Lemma} tell is that  $|R^i|, |L^i| \le C_0 +

3154: C_1 + 2B +1$, for all $i$.

3155: 

3156: We have already bounded the lengths of $V^i(1), V^i(2)$ and $V^i(3)$

3157: by $C_0+C_1, 4B+1$ and $C_0+C_1$, respectively. Thus we are

3158: now in a position to invoke the repetitive behaviour described in Definition

3159: \ref{T1'Lemma}:

3160: for some positive integers $i $ and $t$ with $i+t\le T_1'$, we get a repetition 

3161: \[	\Big( R^i, V^i(1), V^i(2), V^i(3), L^i \Big) =  

3162: \Big( R^{i+t}, V^{i+t}(1), V^{i+t}(2), V^{i+t}(3), L^{i+t} \Big).

3163: \]

3164: For as long as we are assured of the continuing presence

3165: of $\nu_1^{i+s}$ and $\nu_3^{i+s}$,

3166:  the fate of $\nu_2^i=V^i(1)V^i(2)V^i(3)$ under $s$ iterations of $\phi$ depends

3167: only on $(R^i, V^i(1), V^i(2), V^i(3), L^i)$. Thus

3168: $$

3169: \Big( V^j(1), V^j(2), V^j(3) \Big) = \Big( V^{j+t}(1),

3170: V^{j+t}(2), V^{j+t}(3) \Big)

3171: $$

3172: for all $j\ge i$ within the time scale of this assurance. However this leads us

3173: to an absurd conclusion, because once $ \nu_1$ has become constant, 

3174: at all subsequent time,

3175: the surviving word coloured  $\nu_2$ contains as a proper subword, the

3176: $\nu_2$ word that existed at the corresponding times in the  cycles (of

3177: period $t$) before $T_1'$, and in particular they can never essentially

3178: vanish, contrary to our assumption that $\nu_3$ eventually neuters

3179: $\nu_2$. 

3180: \end{proof}

3181:  

3182: 

3183: 

3184: 

3185: 

3186: \subsection{Nesting and the Pincer Lemma}

3187: 

3188: In subsequent sections we would like to

3189:  bound the life of pincers by arguing

3190: that during the lifetime of a pincer,

3191:  colours must be consumed at a predictable rate (appealing

3192: to Proposition \ref{prePincerLemma}),

3193: noting that there  are only a limited number of colours. However, the bounds

3194: we need will require us to ascribe each consumed colour to a {\em unique}

3195: pincer. Thus we encounter problems whenever one pincer is contained in another.

3196: For reasons that will become apparent in subsequent sections,

3197: %(and  were foreshadowed in the proof of Lemma \ref{shorty})

3198: in situations where

3199: we must confront this problem, the inner of the two pincers will have a long block

3200: of constant edges along the corridor immediately above its peak. More precisely,

3201: we will find ourselves in the situation described in the following definition. The

3202: appearance of the constant $\ll := 2B(T_0+1)+1$ in the following definition is

3203: explained by the role that this constant played in the course of Lemma \ref{shorty}.

3204:  

3205: 

3206: \begin{definition} \label{NestingDef}

3207: Consider one pincer $\Pin_1$ contained in another $\Pin_0$.  Suppose

3208: that in the corridor $S \subseteq \Pin_0$ at the top of $\Pin_1$

3209: (where its boundary paths $p_1(\Pin_1)$ and $p_2(\Pin_1)$ come

3210: together) the future in $\top(S)$ of at least one of the edges

3211: containing $p_1(\Pin_1) \cap \bot(S)$ or $p_2(\Pin_1) \cap \bot(S)$

3212: contains no non-constant edges, and this future\footnote{We allow this

3213: future to be empty, in which case ``contained in" means that the immediate past

3214: of the long block of constant edges is not separated from $\Pin_1$ by any

3215: edge that has a future in $\top(S)$.}

3216: lies in an interval of at least $\ll$ constant edges contained in

3217: $\Pin_0$.  Then we say that $\Pin_1$ is {\em nested in} $\Pin_0$. (in

3218: Figure \ref{figure:Nest}, the $\ll$-long block of constant edges are shown in

3219: black.) We say that $\Pin_1$ is {\em left-loaded} or {\em right-loaded}

3220: according to the direction in which  the $\ll$-long block of constant edges

3221: extends from the peak of $\Pin_1$.

3222: \end{definition} 

3223: 

3224: \begin{figure}[htbp] 

3225: \begin{center} 

3226:   

3227: \input{Nest.eps_t} 

3228:   

3229: \caption{A depiction of nesting} 

3230: \label{figure:Nest} 

3231: \end{center} 

3232: \end{figure}

3233: 

3234: 

3235: \begin{remark}  A nested pincer cannot be both left-loaded and right-loaded (cf.

3236: Remark \ref{subarray}(3)).

3237: 

3238: If $\Pin_1$ is left-loaded, then  the future of  

3239: $p_1(\Pin_1) \cap \bot(S)$ contains no non-constant edges.

3240: It may happen that the future of $p_2(\Pin_1)$ also contains no 

3241: non-constant edges; in this case the colour $\mu$ of $p_2(\Pin_1)$

3242: essentially vanishes in $S$ due to   cancellation between non-constant edges 

3243: of $\mu$ and some colour to its right. Symmetric considerations apply

3244: to right-loaded pincers.

3245: \end{remark}

3246: 

3247: \begin{definition} \label{chiP}

3248: For a pincer $\Pin_0$, let $\{ \Pin_i \}_{i \in I}$ be the set of all

3249: pincers nested in $\Pin_0$.  Then define

3250: \[	\chi(\Pin_0) = \tilde{\chi}(\Pin_0) \smallsetminus \bigcup_{i \in

3251: I} \tilde{\chi}(\Pin_i).	\]

3252: \ed

3253: 

3254: 

3255: 

3256: \bl \label{nestLife}

3257: If the pincer $\Pin_1$ is nested in $\Pin_0$ then 

3258: $\time(S_t(\Pin_1)) < \time(S_{\Pin_0}).	$

3259: \end{lemma}

3260: \begin{proof}

3261: The presence of the hypothesised block of constant letters in

3262: $\top(S_t(\Pin_1))$ makes this an immediate consequence of the Buffer

3263: Lemma \ref{BufferLemma}.

3264: \end{proof}

3265: 

3266: Define $T_1 :=  T_1' + 2T_0$.

3267: The following theorem is the main result of this section.

3268: 

3269: 

3270: \begin{theorem}[Pincer Lemma] \label{PincerLemma} 

3271: For any pincer $\Pin$

3272: \[	\life(\Pin) \leq T_1(1 + |\chi(\Pin)|).	\]

3273: \end{theorem}

3274: 

3275: 

3276: 

3277: \begin{proof}

3278: The heart of our proof of Proposition \ref{prePincerLemma} was that

3279: in each block of $\hat T_1$  steps in time between  $\time(S_b)$ and

3280: $\time(S_\Pin)$ at least one colour essentially disappears. Our proof

3281: of the present theorem is an elaboration of that argument: we must

3282: argue for the essential disappearance of a  colour that is not

3283: contained in any of pincers nested in $\Pin$.  Thus we concentrate

3284: on that region of the pincer $\Pin$ that is exterior to the set of 

3285: {\em co-level\footnote{i.e. those that are maximal

3286: with respect to inclusion among the pincers nested in $\Pin$} 1} pincers nested in it;

3287: let $\{ \Pin_j\}, \ j=1,\dots,{J}$ be the set of such, indexed in order of appearance from

3288: left to right.

3289: 

3290: For $j=1,\dots,J-1$, let $\Sigma_j$ denote the set of  colours  along the bottom of $\Pin$

3291: that have a non-constant edge strictly

3292: between $\Pin_j$ and $\Pin_{j+1}$; if $\Pin_j$ is left-loaded, then we include

3293: the colour of $p_2(\Pin_j)$ in $\Sigma_j$, and if $\Pin_j$ is right-loaded, then we include

3294: the colour of $p_1(\Pin_j)$ in $\Sigma_{j-1}$. Likewise, we define $\Sigma_0$ to be

3295: the set of non-constant colours that lie to the left of $\Pin_1$ together with the

3296: colour of $p_1(\Pin)$, and we define

3297: $\Sigma_{J}$ to be

3298: the set of non-constant colours that lie to the right of $\Pin_J$ together with the

3299: colour of $p_2(\Pin)$. 

3300: 

3301: In order to prove the theorem, we derive a contradiction from the assumption that

3302: in the first $T_1$ units of time in the life of $\Pin$ no colours in the union of the

3303: $\Sigma_j$ essentially vanish. (There is no loss of generality in starting at the

3304: bottom of the pincer, since given any other starting time, one can discard the

3305: pincer below that level.) We label the corridors, beginning at the bottom of $\Pin$

3306: and proceeding in time as $S^0,S^1,\dots$

3307: 

3308: We focus on a single $\Sigma_j$, and write its colours in order as    $\nu_1, \ldots , \nu_r$.

3309: We analyse how the colours in $\Sigma_j$ come to vanish.  

3310: The first important observation is that  $2 \le i \le r-1$,

3311:  it is not possible for the colour $\nu_i$ to essentially vanish (at any time)

3312: due to cancellation merely between the colours in $\Sigma_j$.  

3313: For if this happened,  there would be an implosive array in $S^0$

3314: containing $\nu_i(S^0)$  and so, by Proposition \ref{prePincerLemma}, $\nu_i$ would vanish before $S^{T_1}$, contrary to our assumption.

3315: 

3316: There remains the possibility that $\nu_2$ may neuter  $\nu_1$ (after

3317: $S^{T_1}$).  This can happen in two ways.  The first is that $\Pin_{j-1}$ is left-loaded: in

3318: this case

3319: the neutering happens within time $T_0$ of the top of $\Pin_{j-1}$ (by Two Colour Lemma),

3320: and we are then in a stable situation in the sense that $\nu_3$ cannot subsequently neuter $\nu_2$,

3321: by Proposition \ref{NoDoubleNeuter}.  Now suppose that $\Pin_{j-1}$ is right-loaded.

3322: Consider the earliest time $t_0$ at which there is a block of  at least $B+1$ constant edges in the

3323: past of the $\lambda_0$-long block associated to $\Pin_{j-1}$. If $\nu_2$ is to neuter

3324: $\nu_1$, then it must do so within   $T_0$ steps of this time. Indeed, within $T_0$ steps,

3325: if the non-constant edges of $\nu_1$ to the right of the block have not been consumed

3326: by $\nu_2$, 

3327: then they will never be consumed by a colour from $\Sigma_j$. 

3328: 

3329: There is a further event that we must account for, which is closely related to

3330: neutering: it may

3331: happen that $\nu_1$ is the colour of $p_2(\Pin_{j-1})$ and that $\nu_2$ consumes

3332: all of the non-constant edges to the right of the block of constant edges discussed above;

3333: this is not a neutering but nevertheless the Two Colour Lemma applies. We would like

3334: to apply Proposition \ref{NoDoubleNeuter} in this situation to conclude that

3335: $\nu_3$ cannot subsequently neuter $\nu_2$.

3336: This is legitimate provided

3337: $t_0\ge\time (S^{T_1'})$.  If $t_0< \time (S^{T_1'})$, then we still know that $\nu_3$

3338: cannot  neuter $\nu_2$ before $S^{T_1}$, because by hypothesis no colour from

3339: $\Sigma_j$ essentially vanishes before this time. On the other hand, the Two Colour Lemma

3340: tells us that if $\nu_3$ is to neuter $\nu_2$, then it must do so within $T_0$ steps

3341: from $t_0$, and $t_0+T_0\le  \time (S^{T_1})$. Thus, once again, we conclude that

3342: $\nu_3$ can never neuter $\nu_2$.

3343: 

3344: Entirely similar arguments show that it cannot happen that $\nu_r$ is neutered

3345: by $\nu_{r-1}$ and that subsequently $\nu_{r-2}$ neuters $\nu_{r-1}$.

3346: 

3347: We have established the existence of a stable situation: proceeding past the point where

3348: the restricted amount of possible neutering within $\Sigma_j$

3349: has occurred, we may assume that the next

3350: essential disappearance of a colour from $\Sigma_j$ can only occur as a result of

3351: cancellation with a colour from some $\Sigma_i$ with $i\neq j$. Such further cancellation

3352: must occur, of course, because all but two\footnote{Degenerate cases with few

3353: colours are covered by the Two Colour Lemma and the Buffer Lemma.} of the

3354:  colours in $\bigcup_j\Sigma_j$ must be consumed within $\Pin$.

3355: 

3356: Passing to innermost pair of interacting $\Sigma_k$

3357: we may assume $i=j-1$ (cf. proof of Lemma \ref{shorty}). Thus our proof will be

3358: complete if we can argue that cancellation between non-constant edges

3359: from $\Sigma_{j-1}$ and $\Sigma_j$ is impossible. We have

3360: argued that the colours which are to cancel will be essentially adjacent within

3361: time $T_0$ of 

3362: the top of $\Pin_{j-1}$. On the other hand, there is a block of $\ll$ constant

3363: edges separating $\Sigma_{j-1}$-nonconstant edges and $\Sigma_{j}$-nonconstant edges

3364: at the top of $\Pin_{j-1}$. Since $\ll > 2B(T_0+1)$ at least $B+1$ of these constant edges

3365: remain $T_0$ steps later. The Buffer Lemma now obstructs the supposed

3366: cancellation between non-constant edges in $\Sigma_{j-1}$ and $\Sigma_j$. 

3367: \end{proof}

3368: 

3369:  

3370: \section{Teams and their Associates}\label{teamSec} 

3371:  

3372: We begin the process of grouping pairs of colours $(\mu,\mu')$ into 

3373: teams.  

3374:  

3375: \subsection{Pre-teams} \label{t1} 

3376: 

3377: The whole of $C_{(\mu,\mu')}(2)$ will ultimately be  consumed by  

3378: a single edge $\e_0\in\mu'(S_0)$. 

3379: We consider the time $t_0$ at which the future of $\e_0$ 

3380: starts consuming the future of  $C_{(\mu,\mu')}(2)$.  

3381: If $|C_{(\mu,\mu')}(2)|> 2B$, then this consumption will not be completed in 

3382: three steps of time  (Lemma \ref{BCL}). We claim that in  this circumstance, the 

3383: leftmost $\mu'$-coloured edge after the first two steps of the

3384: cancellation must be left para-linear. Indeed it is not left-constant since it must consume edges in 

3385: the future of $C_{(\mu,\mu')}(2)$, and since no non-constant $\mu'$-edges 

3386: are cancelled by $\mu$ in passing from the first to the second stage of  

3387: cancellation, the leftmost non-constant $\mu'$-label must remain the same (Proposition 

3388: \ref{power}). We denote this \lpl edge at time $t_0+2$ by $\e^\mu$. 

3389:  

3390: Let $\e_\mu$ be the rightmost edge in the future of $C_{(\mu,\mu')}(2)$  at time $t_0$. 

3391: We trace the ancestry of $\e_\mu$ and $\e^\mu$ in the trees of $\F\subset\Delta$ corresponding 

3392: to the colours $\mu$ and $\mu'$ (as defined in \ref{tree}). 

3393:  We go back to the last point in time 

3394: $\ptmm$ at which  

3395: both ancestors  lay in a common corridor 

3396:  {\em and} the interval on the bottom of this corridor between the pasts of 

3397:   $\e_{\mu}$ and $\e^\mu$  

3398: is comprised entirely of constant edges whose future is eventually  

3399:  consumed by the ancestor of 

3400: $\e^\mu$ at this time. We denote this corridor  $S_{\uparrow}$. 

3401:  

3402: \begin{definition}\label{preteam}  The ancestor of $\e^\mu$ at time $\ptmm$ is called the 

3403: {\em reaper} and is denoted $\prmm$.  The set of edges in $\bot(S_{\uparrow})$ 

3404:  which are eventually consumed by $\prmm$ is denoted $\pEmm$.  

3405:   This is a contiguous set of edges. 

3406: The {\em pre-team} $\pTmm$ is defined to 

3407: be the set of pairs $(\mu_1,\mu')$ such that $\pEmm$ contains 

3408: edges coloured $\mu_1$. The number of edges in $\pEmm$ is denoted 

3409: $\|\hat \T\|$. 

3410: \end{definition} 

3411:  

3412: In a little while we shall define {\em teams} to be   

3413: pre-teams satisfying a certain maximality condition (see Definition \ref{newTeams}). 

3414:   

3415:  

3416: \begin{remark} If $\ptmm<\time(S_0)$ then 

3417: near the right-hand end of $\pEmm$ one may have  an interval of 

3418: colours  $\nu$ such that $\nu(S_0)$ is empty.  

3419: \end{remark} 

3420:  

3421: \smallskip 

3422:  

3423: In the proof of Proposition \ref{SummaryLemma} we saw that it would be

3424: desirable if (whatever our final 

3425: definition of {\em team} and $\bonus$ may be) the following inequality

3426: (\ref{goodEq}) should hold for all teams: 

3427: \begin{equation} \label{preTeamInequality} 

3428: \sum_{(\mu,\mu') \in \T \mbox{ \tiny or } (\mu,\mu') \vin \T} |C_{(\mu,\mu')}(2)| \le 

3429: \|\T\|  + |\bonus(\T)| + B. 

3430: \end{equation} 

3431:  

3432: 

3433:  

3434: The following lemma shows that, even without introducing a  bonus scheme 

3435: or virtual members, the 

3436: desired inequality is straightforward for pre-teams with $\ptmm \geq \time(S_0)$. 

3437:  

3438: \begin{lemma} \label{t1high} 

3439:  If $\ptmm \geq \height(S_0)$ then $\pTmm$ satisfies 

3440: \[ \sum_{(\mu,\mu') \in \pTmm}|\cmm| \leq \|\pTmm\| + B.      \] 

3441: \end{lemma} 

3442:  

3443: \begin{proof} By definition   

3444: $\mu'(S_0)$ does not start consuming any  of the

3445:  $C_{(\mu_1,\mu')}(2)$ with $(\mu_1,\mu')\in\pT$ before $\ptmm$  

3446: (apart from a possible nibbling of  length $< B$ from the rightmost team 

3447: member  at time $\ptmm -1$).  

3448: Since each $C_{(\mu_1,\mu')}(2)$ consists only of edges consumed 

3449: by $\mu'(S_0)$, the future  of each $C_{(\mu_1,\mu')}(2)$ at time $\ptmm$ 

3450: will have the same length as $C_{(\mu,\mu')}(2)$ 

3451: (except that the rightmost may have lost these $< B$ edges).  

3452: And these futures are contained in $\pEmm$. 

3453: \end{proof} 

3454:  

3455:  

3456: The case where $\ptmm <\height(S_0)$ is more troublesome. As $\pEmm$ flows forwards in 

3457: time, the number of constant letters in the future of $\pEmm$ that are  

3458: consumed by $\prmm$ between $\ptmm$ and $\time (S_0)$ may be outweighed by the number of 

3459: constant letters generated to the left of the future of $\pEmm$ that will 

3460: ultimately be consumed by $\prmm$. 

3461:  

3462: It is to circumvent the failure of inequality 

3463: (\ref{preTeamInequality}) in this setting that we are 

3464: obliged to instigate the bonus scheme described in Section \ref{BonusScheme}. 

3465:  

3466: 

3467: 

3468: \subsection{The Genesis of pre-teams} \label{genesis} 

3469:  

3470: We fix $\pTmm$ with $\ptmm < \time(S_0)$ and consider 

3471: the various events that occur  at  $\ptmm$ to prevent 

3472: us pushing the pre-team back one step in time.  We write $S_\omega$ 

3473: to denote the corridor at time $\ptmm$ containing $\pEmm$. 

3474:  

3475: \begin{figure}[htbp] 

3476: \begin{center} 

3477:   

3478: \input{G1Pic.eps_t} 

3479:   

3480: \caption{A team of genesis (G1)}

3481: \label{G1Pic} 

3482: \end{center} 

3483: \end{figure} 

3484: 

3485: 

3486: \begin{figure}[htbp] 

3487: \begin{center} 

3488:   

3489: \input{G2Pic.eps_t} 

3490:   

3491: \caption{A team of genesis (G2)}

3492: \label{G2Pic} 

3493: \end{center} 

3494: \end{figure} 

3495: 

3496: \begin{figure}[htbp] 

3497: \begin{center} 

3498:   

3499: \input{G3Pic.eps_t} 

3500:   

3501: \caption{A team of genesis (G3)}

3502: \label{G3Pic} 

3503: \end{center} 

3504: \end{figure} 

3505: 

3506: \begin{figure}[htbp] 

3507: \begin{center} 

3508:   

3509: \input{G4Pic.eps_t} 

3510:   

3511: \caption{A team of genesis (G4)}

3512: \label{G4Pic} 

3513: \end{center} 

3514: \end{figure} 

3515: 

3516: 

3517: 

3518: \def\CC{C}

3519:  

3520: There are four types of events: 

3521:  

3522: \begin{enumerate} 

3523: \item[(G1)] The immediate past  of $\CC_{(\mu,\mu')}(S_\omega) 

3524: $ is separated from the 

3525: past of $\prmm$ by an intrusion of $\partial\Delta$ (Figure \ref{G1Pic}). 

3526: \item[(G2)] We are not in case (G1), but the immediate past  of

3527: $\CC_{(\mu,\mu')}(S_\omega)$ is separated from the 

3528: past of $\prmm$ because of a singularity (Figure \ref{G2Pic}). 

3529: \item[(G3)] The immediate past of $C_{(\mu,\mu')}(S_\omega)$ is  still

3530: in the same corridor as the past of $\prmm$, but it is  separated from

3531: it by a non-constant letter (Figure \ref{G3Pic}). 

3532: \item[(G4)] We are not in any of the above cases,  

3533: but the immediate past of the rightmost letter in  

3534:  $C_{(\mu,\mu')}(S_\omega)$ is not constant (Figure \ref{G4Pic}). 

3535: \end{enumerate} 

3536:  

3537: \smallskip  

3538:  

3539: \def\ST{S_\T} 

3540: \def\STmm{\S_{\T(\mu,\mu')}} 

3541:  

3542:  

3543:  

3544: The following lemma explains why Figures \ref{G3Pic} and \ref{G4Pic} are an

3545: accurate portrayal  of cases (G3) and (G4). 

3546:  

3547: Let $M_{inv}$ be the maximum length of $\phi^{-1}(x)$ over generators

3548: $x$ of $F$, and $C_4 = M_{inv}.M$.  

3549:  

3550: \begin{lemma}\label{G34pics} 

3551: If $I$ is an interval on $\top (S)$ labelled by a word $w$ in constant

3552: letters

3553: then the reduced word labelling the past of $I$ in $\bot(S)$ is of the 

3554: form $u\alpha v$, where $\alpha$ is a word in constant letters and 

3555: $|u|$ and $|v|$ are less than $C_4$. Moreover, if the past of the leftmost  

3556: (resp. rightmost) letter 

3557: in $w$ is constant, then $u$ (resp. $v$)  is empty. 

3558: 

3559: In particular, $|I| \leq |\alpha| + 2MC_4$.

3560: \end{lemma}  

3561:  

3562: 

3563: \begin{figure}[htbp] 

3564: \begin{center} 

3565:   

3566: \input{C4Pic.eps_t} 

3567:   

3568: \caption{The proof of Lemma \ref{G34pics}}

3569: \label{bottom-constant} 

3570: \end{center} 

3571: \end{figure} 

3572: 

3573: \begin{proof} See Figure \ref{bottom-constant}.  Follow the path from

3574: the left end of $I$ to $\bot(S)$.  This passes through a (possibly

3575: empty) path $a^{-1}$, followed by an edge labelled $t^{-1}$, where the

3576: length of $a$ is less than $M$ (since it can be chosen to be on the

3577: top of a $2$-cell which has an edge in $I$).

3578: Similarly, at the right end of $I$ we have a path labelled

3579: $bt^{-1}$, where the length of $b$ is less than $M$. The path along

3580: $\bot(S)$ joining the two endpoints of these paths is labelled by the

3581: reduced word freely equal in $F$ to $\phi^{-1}(awb) =

3582: \phi^{-1}(a)w\phi^{-1}(b)$.  The only non-constant edges in this word

3583: come from $\phi^{-1}(a)$ and $\phi^{-1}(b)$, which have lengths at

3584: most $M.M_{inv}$.  This proves the assertion in the first sentence.

3585:  

3586: The assertion in the second sentence follows from the observation that

3587: if $x,\, y$ and $\phi(x\beta y)$ consist only of constant letters,

3588: then so does the reduced form of $\beta$, and the assertion in the

3589: final sentence follows immediately from the first.

3590: \end{proof} 

3591:  

3592: \begin{remark}\label{decreell} 

3593:  It is convenient to assume that $MC_4 < \ll$. (In the unlikely 

3594: event that this is not the case, we simply increase $\ll$.) 

3595: \end{remark} 

3596:  

3597:  

3598: 

3599: We are finally in a position to make an appropriate definition of a team. 

3600:  

3601:  

3602: \begin{definition}\label{newTeams} \label{shortDef}

3603: All pre-teams $\pTmm$ with $\hat 

3604: t_1(\mu_1,\mu')\ge\time(S_0)$ 

3605: are defined to be teams, but the qualification criteria for pre-teams with 

3606:  $\hat t_1(\mu_1,\mu')<\time(S_0)$ are 

3607: more selective. 

3608:  

3609: If the genesis of  $\pTmm$ is of type  (G1) or (G2), then 

3610:  the rightmost component of the pre-team may  form a pre-team 

3611: at times before $\ptmm$.  In particular,  it may happen  

3612: that $(\mu_1,\mu')\in\pTmm$ but $\ptmm > \hat t_1(\mu_1,\mu')$ and hence 

3613: $(\mu,\mu')\not\in\pT(\mu_1,\mu')$. To avoid double 

3614: counting in our estimates on $\|\T\|$ we disqualify the  

3615: (intuitively smaller) pre-team $\pT(\mu_1,\mu')$ in these settings.  

3616:  

3617: If the genesis of $\pTmm$ is of type (G4), then again it may happen 

3618: that what remains to the right of $\pTmm$ at some time  before $\ptmm$ is a pre-team. 

3619: In this case, we disqualify the (intuitively larger) pre-team  $\pTmm$.   

3620:   

3621: The pre-teams that remain after these disqualifications 

3622: are now defined to be {\em teams}.  

3623:  

3624: A typical team will be denoted $\T$ 

3625: and all hats will be dropped from the notation for their associated objects 

3626: (e.g. we write $\Emm$ instead of $\pEmm$).  

3627:  

3628: A team is said to be {\em short} if $\|\T\|\le \ll$

3629: or $\sum\limits_{(\mu,\mu')\in\T} |C_{(\mu,\mu')}(2)| \le \ll$. Let

3630: $\S$ denote the set of short teams.

3631: \end{definition} 

3632:  

3633: \begin{lemma} \label{G4lemma} Teams of genesis (G4) are short. 

3634: \end{lemma} 

3635:  

3636: \begin{proof} Lemma \ref{G34pics} implies that $\ET$ is in the immediate 

3637: future of an interval of length at most $C_4$. And we have decreed (Remark 

3638: \ref{decreell}) that $MC_4< \ll$. 

3639: \end{proof} 

3640:  

3641: We wish our ultimate definition of a team to be such that every  pair  $(\mu,\mu')$ 

3642: with $C_{(\mu,\mu')}(2)$ non-empty is assigned to a team. The above definition 

3643: fails to achieve this because of two phenomena: first, a pre-team 

3644:  $\pTmm$ with genesis of type (G4) may

3645: have been disqualified, leaving $(\mu,\mu')$ teamless; second, in our initial discussion of  

3646: pre-teams (the first paragraph of Section \ref{t1}) we excluded pairs $(\mu,\mu')$ 

3647: with $|C_{(\mu,\mu')}(2)|\le 2B$. The following definitions remove these difficulties. 

3648:  

3649: \bd[Virtual team members] \label{Virtual} 

3650: If a pre-team $\pTmm$ of type (G4) is disqualified under the terms of Definition \ref{newTeams} 

3651: and the smaller team necessitating disqualification is $\pT(\mu_1,\mu')$,  

3652: then we define $(\mu,\mu')\vin\pT(\mu_1,\mu')$ and $\pTmm\subset_v\pT(\mu_1,\mu')$. 

3653: We extend the relation $\subset_v$ to be transitive and extend $\vin$ correspondingly. 

3654: If $(\mu,\mu')\vin\T$ then $(\mu_2,\mu')$ is said to be a {\em virtual member} of  

3655: the team $\T$. 

3656: \ed 

3657:  

3658: \bd If  $(\mu,\mu')$ is such that  $1\le |C_{(\mu,\mu')}(2)|\leq 2B$ and 

3659: $(\mu,\mu')$ is neither a member nor a virtual member of any previously 

3660: defined team, then we define $\T_{(\mu,\mu')}:=\{(\mu,\mu')\}$ to be a

3661: (short) team with $\|\T_{(\mu,\mu')}\|=|C_{(\mu,\mu')}(2)|$.  

3662: \ed 

3663:  

3664: \begin{lemma}\label{allIn}  

3665: Every   $(\mu,\mu')\in\vecZ$ with $C_{(\mu,\mu')}(2)$ non-empty is a member 

3666: or a virtual member of exactly one team, and there are less than $2\n$ teams. 

3667: \end{lemma} 

3668:  

3669: \begin{proof} The first assertion is an immediate consequence of the preceding 

3670: three definitions, and the second  follows 

3671:  from the fact that $|\vecZ| < 2\n$. 

3672: \end{proof} 

3673:  

3674: 

3675: \subsection{Pincers associated to teams of Genesis (G3)}

3676:  

3677: In this subsection we describe the pincer $\Pin_\T$ canonically

3678: associated to each team of genesis $(G3)$. 

3679: The definition of $\Pin_\T$

3680: involves the following concept which will prove important also for teams

3681: of other genesis.

3682: 

3683: 

3684: \begin{definition}\label{narrowPast}

3685: We define the {\em narrow past} of a team 

3686: $\T$ to be the set of constant edges that have a future in $\ET$. The narrow 

3687: past may have several components at each time, the set of which  

3688: are ordered left to right according to the ordering in $\ET$ of their futures.  We call these components {\em sections}. 

3689: \end{definition}

3690: 

3691: 

3692: {\center{\em{For the remainder of this subsection we 

3693: consider only long teams of genesis (G3). }}}

3694: 

3695: \begin{definition}[The Pincer $\tilde\Pin_\T$] \label{pl}\label{t2}

3696: The paths labelled $\hat p_l$ and $\hat p_r$ in Figure \ref{G3Pic}

3697:  determine a pincer and are defined as follows. Let $\xT$ be the

3698:  leftmost non-constant edge to the right of $\mu$ in the immediate

3699:  past  of  $\T$, and let  $x_1(\T)$  be the edge that consumes it.   

3700: Define $\tplT$ to be the path in $\F$ that traces the history of $\xT$

3701: to the boundary,  and let $\tprT$ be the path that traces the history

3702: of  $x_1(\T)$.

3703: (Note that  $x_1(\T)$ is left-fast.) 

3704: 

3705:  

3706: Define $\tilde t_2(\T)$ to be the earliest time at which the  

3707: paths $\tplT$ and $\tprT$ lie in the same corridor.

3708: The segments of the paths $\tplT$ and $\tprT$ after this time, together

3709: with the path joining them along the bottom of the

3710: corridor at  time $\tilde t_2(\T)$ form a pincer. We denote this pincer

3711: $\tilde\Pin_{\T}$. 

3712: \end{definition} 

3713: 

3714: 

3715: 

3716: The Pincer Lemma argues for the regular disappearance of colours

3717: within a pincer during those times when more than two colours continue

3718: to survive along the corridors of $\tilde\Pin_\T$. 

3719: However, when there are only two colours the situation  is

3720: more complicated.  

3721: 

3722: We claim that the following situation cannot arise:

3723: $\time(S_{\hat\Pin_\T}) \leq \tone - T_0$, the path

3724: $\tplT$ and the entire narrow past of $\T$ are in the same corridor at

3725: time $\tone - T_0$, and at this time they 

3726: are separated only by constant edges. For if this were the case,

3727: then the colour of $\tprT$ would $\phi$-neuter the colour of $\tplT$

3728: but would take more than $T_0$ steps to do so, contradicting

3729: the Two Colour Lemma.  Thus at least one of the three hypotheses in

3730: the first sentence

3731: of this paragraph is false; we consider the three possibilities. The

3732: troublesome case (3) leads to a cascade of pincers as depicted in

3733: Figure \ref{cascade}.

3734: 

3735: 

3736: \begin{definition}[The Pincer $\Pin_{\T}$ and times $t_2(\T)$ and

3737: $t_3(\T)$] \label{PincerDef} 

3738: 

3739: \ 

3740: 

3741: \begin{enumerate}

3742: \item {\em Some section of the narrow past of $\T$ is not in the same corridor as

3743:  $\tplT$ at time $\tone - T_0$:} In this case\footnote{this includes the

3744: possibility that $\tplT$ does not exist at time  $\tone - T_0$}

3745: we define $t_2(\T)=t_3(\T)$ to be the earliest time at which the entire

3746: narrow past of $\T$ lies in the same corridor as $\tplT$ and has length at least

3747: $\ll$.

3748: \item {\em Not case (1), there are no non-constant edges between $\tplT$

3749: and the narrow past of $\T$ at time $\tone - T_0$:} In this case

3750: $\time(S_{\tilde\Pin_{\T}}) > \tone - T_0$.  We define

3751: $\Pin_{\T} = \tilde\Pin_{\T}$ and  $t_3(\T) =

3752: \time(S_{\Pin_{\T}})$. If the narrow past of $\T$ at time $\tone - T_0$ 

3753: has length less than $\ll$, we define

3754: $t_2(\T) =  t_3(\T)$, and otherwise $t_2(\T) =\tilde t_2(\T)$. 

3755: \item 

3756: {\em Not in case (1) or case (2):}

3757: In this case there is at least one non-constant edge between the

3758: narrow past of $\T$ and $\tplT$ at 

3759: $\tone - T_0$. We  pass to the latest time at which there is such an

3760: intervening

3761: non-constant edge and consider  the path $\tilde p_l^\prime(\T)$

3762: that traces the history of the

3763: leftmost intervening non-constant edge $x'(\T)$ and the path $\tilde

3764: p_r^\prime(\T)$

3765: that traces the history of the edge $x_1^\prime(\T)$ that cancels

3766: with $x'(\T)$.

3767: We define  $\tilde t_2'(\T)$ to be the earliest time at which the  

3768: paths $\tilde p_l^\prime(\T)$ and $\tilde p_r^\prime(\T)$  lie in the

3769: same corridor

3770: and consider  the pincer formed by the  segments of the paths $\tilde

3771: p_l^\prime(\T)$ and $\tilde p_r^\prime(\T)$

3772: after  time $\tilde t_2'(\T)$ together

3773: with the path joining them along the bottom of the

3774: corridor at  time $\tilde t_2^\prime(\T)$.

3775: 

3776: We now repeat our previous analysis with the primed objects $\tilde p_l^\prime(\T), \tilde t_2^\prime(\T)$ {\em etc.} in

3777: place of $ \tilde p_l(\T), \tilde t_2(\T)$ {\em etc.}, checking whether we now fall into case (1) or (2);

3778: if we do not then we pass to $ \tilde p_l''(\T), \tilde t_2''(\T)$ {\em etc.}, and iterate the analysis until

3779: we do indeed fall into case (1) or (2), at which point we acquire the desired definitions of 

3780: $\Pin_\T,\, t_2(\T),\,  t_3(\T)$.

3781: \end{enumerate}

3782: 

3783: Define $p_l(\T)$ (resp. $\prT$) to be the left (resp. right)

3784: boundary path of the pincer $\Pin_\T$ extended backwards

3785: in time through $\F$ to $\partial\Delta$. Define $p_l^+(\T)$ to be the

3786: sequence of non-constant edges (one at each time) lying immediately to

3787: the right of the narrow past of $\T$ from the top of $\Pin_{\T}$ to

3788: time $\tone$. (These are edges of the leftmost of the primed $\tplT$

3789: considered in case (3).)

3790: \ed 

3791: 

3792: 

3793: \begin{definition}

3794: Let $\T$ be a long team of genesis (G3).  Let $\chi_P(\T)$ be the set

3795: of colours containing the paths $\tilde p_l(\T), \tilde p_l'(\T),\tilde p_l''(\T),\dots$ that

3796: arise in (iterated applications of) case (3) of Definition \ref{PincerDef} but

3797: do not become $p_l(\T)$.

3798: \ed 

3799: 

3800: \begin{figure}[htbp] 

3801: \begin{center} 

3802:   

3803: \input{FakePincers.eps_t} 

3804:   

3805: \caption{The cascade of pincers.}

3806: \label{cascade} 

3807: \end{center} 

3808: \end{figure}

3809: 

3810: 

3811: The preceding definitions are framed so as to make the following important

3812: facts self-evident.

3813: 

3814: \begin{lemma} 

3815:  \label{t1t3forTeam}\label{disj1}

3816: 

3817: \ 

3818: 

3819: \begin{enumerate}

3820: \item

3821: If $\T$ is a long team of genesis (G3),  

3822: \[      t_1(\T) - t_3(\T) \leq T_0(|\chi_P(\T)| + 1).   \] 

3823: \item

3824: If $\T_1$ and $\T_2$ are disjoint then  $\chi_P(\T_1)\cap \chi_P(\T_2)=\emptyset$.

3825: \end{enumerate}

3826:  \end{lemma}

3827: 

3828: \subsection{The length of teams} \label{TeamLemmas} 

3829: 

3830: 

3831:  

3832: \begin{definition}\label{down1}  

3833: Define $\down_1({\mathcal T})\subset\partial\Delta$ to consist of those 

3834: edges $e$ that are labelled $t$ and satisfy one of the following conditions: 

3835: \begin{enumerate} 

3836: \item[1.] $e$ is at the left end of a corridor containing a section of the narrow 

3837: past of $\T$ that is not leftmost at that time; 

3838: \item[2.]  $e$ is at the right end of a corridor containing a section of the narrow 

3839: past of $\T$ that is not rightmost at that time;  

3840: \item[3.]  $e$ is at the right end of a corridor which 

3841: contains the rightmost section of the narrow past of $\T$ at that time but which does 

3842: not intersect $\plT$.\\ 

3843: \end{enumerate} 

3844: \end{definition} 

3845:  

3846: All of the edges shown on the boundary in

3847: Figure \ref{figure:TeamAge} are contained in $\down_1(\T)$. 

3848: 

3849: \begin{definition} Define $\partial^\T\subset\partial\Delta$ to be the

3850: set of (necessarily constant) edges that have a preferred future in

3851: $\ET$.

3852: \end{definition} 

3853:  

3854: 

3855: We record an obvious disjointness property of the sets defined above.

3856: 

3857: \begin{lemma}\label{disj2}

3858: 

3859: \ 

3860: 

3861: \begin{enumerate}

3862: \item For distinct teams $\T_1$ and $\T_2$, $\partial^{\T_1}$ and

3863: $\partial_{\T_2}$ are disjoint.

3864: \item For distinct teams $\T_1$ and $\T_2$, $\down_1(\T_1)$ and

3865: $\down_1(\T_2)$ are disjoint. 

3866: \end{enumerate}

3867: \end{lemma}

3868: 

3869: \bd\label{QT}

3870: Suppose that $\T$ is a team of genesis (G3).  We define  $\QT$ be the

3871: set of edges $\e$ with the following properties:

3872: $\plT$ passes through $\e$  before time $t_3(\T)$, and the corridor $S$

3873: with $\e\in\bot(S)$ contains the entire narrow past of $\T$ and

3874: this narrow past has length at least $\ll$.

3875: \ed 

3876:   

3877: 

3878: The following lemma gives us a bound on $|\ET|$, which will reduce 

3879: our task to that of bounding $|\QT|$ for teams of genesis (G3). 

3880:  

3881: \begin{lemma}  \label{TeamAgeLemma}  

3882: 

3883: \ 

3884:  

3885: \begin{enumerate} 

3886: \item[1.] If the genesis 

3887: of $\T$ is of type (G1) or (G2), then 

3888: $$

3889: \|\T\| \leq 2MC_4\,|\down_1({\mathcal T})| +  |\partial^\T|  .

3890: $$ 

3891: \item[2.] If the genesis of $\T$ is of type (G3), then 

3892: $$

3893: \|\T\| \leq 2MC_4\,|\down_1({\mathcal T})| +|\partial^\T| + 2MC_4\,|\QT| +

3894: 2MC_4T_0\big(|\chi_P(\T)| +1\big) + \ll .

3895: $$  

3896: \end{enumerate} 

3897: \end{lemma} 

3898:  

3899: \medskip 

3900:  

3901: \begin{figure}[htbp] 

3902: \begin{center} 

3903:   

3904: \input{TeamAge.eps_t} 

3905:  

3906: \caption{Bounding the size of a team in terms of $|\down_1|$ and $|p_l|$} 

3907: \label{figure:TeamAge} 

3908: \end{center} 

3909: \end{figure} 

3910:  

3911: \begin{proof}     The first thing to observe is that at any 

3912: stage in the past of $\ET$ the set of letters lying in a 

3913: single corridor form a connected region. As in Lemma \ref{G34pics}, 

3914: this is simply a matter of noting that if $\phi(aub)=w$ where $w, a$ and $b$ 

3915: consist only of constant letters, then $u$ must equal a word in constant letters. 

3916:  

3917: Consider the   past of $\ET$ at a time $t$. Write $k_t$ for the number of 

3918: corridors that contain a non-trivial component of this past. 

3919: The total 

3920: increase in length of these components  when one goes forward to time $t+1$ 

3921: is bounded by $2MC_4k_t$, since the connectedness of the past 

3922: implies that  the only growth that can happen for existing components occurs 

3923: at their extremities, where a block of at most $MC_4$ constant letters

3924: may be added. This follows from Lemma \ref{G34pics}.  Also at time $t+1$,

3925: constant letters from $\partial\Delta$ may join the past of $\ET$, and

3926: there may be new components of constant letters (each of length less

3927: than $2MC_4$) whose ancestors at time $t$ were non-constant

3928: letters. Thus we have three possible causes of increase. The first and

3929: third account for growth of at most $2MC_4k_{t+1}$ and the second

3930: (boundary) contribution is the number of elements of $\partial^\T$

3931: that occur at time $t+1$.  

3932: If the genesis of $\T$ is of type (G1) or (G2), then at least $k_{t+1}$ 

3933: edges of $\down_1(\T)$ occur at time $t$, compensating us for the growth summand 

3934: $2MC_4k_{t+1}$. If the genesis of $\T$ is of type (G3) then we still have the 

3935: above compensation {\em except} at those times where no edges of $\down_1(\T)$ occur. 

3936: At these latter times the whole of the narrow past of $\T$ 

3937: lies in a single corridor through which $\plT$ passes. Since the

3938: narrow past lies

3939: in a single corridor, it is connected and  grows at most $2MC_4$ when moving 

3940: forward one unit 

3941: of time (unless added to by $\partial^\T$).  

3942: 

3943: The summands $2MC_4\,|\QT|$ and  $2MC_4T_0\big(|\chi_P(\T)| +1\big) $ in item (2) of the lemma

3944: account for the growth of the narrow past in the intervals of time

3945: below  $t_3(\T)$, and from $t_3(\T)$ to $t_1(\T)$, respectively. The additional summand $\ll$

3946: allows us to desist from our estimating if the narrow past of $\T$ ever shrinks to have

3947: length less than $\ll$. 

3948: \end{proof} 

3949: 

3950: 

3951: 

3952: \subsection{Bounding the size of $\QT$}\label{Proofs} 

3953: 

3954: For the remainder of this section we concentrate exclusively on long teams of genesis (G3)

3955: with $\QT$ non-empty. We denote the set of such teams by $\Gthree$.  

3956: Our goal is to bound $|\QT|$. (In the light of our

3957: previous results, this will complete the required analysis of the

3958: length of teams.)  

3959: 

3960: Recall from  Definition \ref{PincerDef} that for teams of genesis 

3961: (G3), the paths $\plT$ and $\prT$ and the chain 

3962: of 2-cells joining them in the corridor at time $\ttwo$ form a pincer denoted $\Pin_{\T}$. 

3963: The set $\subT$ was defined in Definition \ref{chiP}. 

3964: 

3965: 

3966: An important

3967: feature of  teams in $\Gthree$ is:

3968: 

3969: \begin{lemma}\label{GotBlock}

3970: If $\T \in \Gthree$ then there exists a block of  at least $\ll$

3971: constant edges   immediately adjacent to $\Pin_{\T}$ at each time from

3972: $t_3(\T)$ to the top of $\Pin_{\T}$, and adjacent to $p_l^+(\T)$ from

3973: then until $\tone$. (At time $\tone$ this block contains $\ET$.)

3974: \end{lemma}

3975: 

3976: \begin{proof} The hypothesis

3977: that $\QT$ is non-empty means that the narrow past

3978: of $\T$ at  some time before $t_3(\T)$ has length at

3979: least $\ll$ and is contained in the same corridor as 

3980: $\plT$ (see Definition \ref{QT}).  The definition of $t_3(\T)$ implies that

3981: the  narrow past of $\T$ is contained in a block of constant letters 

3982: immediately adjacent to $\plT$ or $p_l^+(\T)$ from time $t_3(\T)$

3983: until $\tone$. Since the  length of the narrow past of $\T$ does not

3984: decrease before $\tone$, these blocks of constant letters must have

3985: length at least $\ll$.

3986: \end{proof}

3987: 

3988: The following is an immediate consequence of the Pincer Lemma. 

3989: 

3990: \begin{lemma} \label{t1-t2Lemma}  

3991: For all $\T \in \Gthree$, 

3992: $$ 

3993:   t_3(\T) - \ttwo = \life(\Pin_{\T}) \leq \ttt (|\subT|+1) .   

3994: $$

3995: \end{lemma} 

3996:  

3997: 

3998: \begin{lemma} \label{nesters}

3999: If $\T_1, \T_2 \in \Gthree$ are distinct teams then $\chi(\Pin_{T_1})

4000: \cap \chi(\Pin_{T_2})  = \emptyset$.

4001: \end{lemma}  

4002:  

4003: \begin{proof} The pincers $\Pin_{\T_i}$ are either disjoint or else

4004: one is contained in the

4005: other. In the latter case, say  $\Pin_{\T_1}\subset\Pin_{\T_2}$,  

4006: the existence of the  block of $\ll$ constant edges established in

4007: Lemma \ref{GotBlock} means that $\Pin_{\T_1}$ is  

4008: actually nested in $\T_2$ in the sense of Definition \ref{chiP}. Thus 

4009: $\chi(\Pin_{\T_1}) \cap \chi(\Pin_{\T_2})  = \emptyset$ (by 

4010: Definition \ref{chiP}). 

4011: \end{proof}

4012:  

4013: \begin{corollary} \label{t1-t2Corr}

4014: $\sum\limits_{\T \in \Gthree} t_3(\T) - t_2(\T) \leq 3\ttt \n$.

4015: \end{corollary}

4016: 

4017:  

4018: It remains to bound the number of edges in $\QT$ which occur before $\ttwo$;

4019: this is  cardinality of the following set.

4020:  

4021: \begin{definition}\label{down2} 

4022: For $\T\in\Gthree$ we define $\down_2(\T)$ to be the set of edges in $\partial\Delta$ that 

4023: lie at the righthand end of a corridor containing an edge in $\QT$

4024: before time $\ttwo$.

4025: \end{definition} 

4026: 

4027: 

4028: The remainder of this section is dedicated to obtaining a bound on

4029: $$

4030: \sum\limits_{\T\in\Gthree}|\down_2(\T)|,

4031: $$ 

4032: (see Corollary \ref{downbound}).

4033:  

4034: At this stage our task of bounding $\|\T\|$ would be complete if

4035: the the sets $\down_2(\T)$ associated to distinct teams 

4036: were disjoint --- unfortunately they need not be, because of the possible 

4037: nesting of teams as shown in Figures \ref{figure:Nest}

4038: and \ref{figure:doublecount}. Thus we shall 

4039: be obliged to seek further pay-off for our troubles. To this end we 

4040: shall identify two sets of consumed colours $\chi_c(\T)$ and 

4041: $\chi_{\delta}(\T)$ that arise from 

4042: the nesting of teams. 

4043:  

4044: In order to analyse  the 

4045: effect of nesting we need the following vocabulary. 

4046:  

4047:  

4048:  

4049: There is an obvious left-to-right ordering of those paths in the forest 

4050: $\F$ which begin on the arc of $\partial\Delta\ssm\partial S_0$ that commences 

4051: at the initial vertex of the left end of $S_0$. (First one orders the trees, then 

4052: the relative order between paths in a tree is determined by the manner in 

4053: which they diverge; the only paths which are not ordered relative to each 

4054: other are those where one is an initial segment of the other, and this 

4055: ambiguity will not concern us.) 

4056: 

4057: \smallskip

4058: 

4059: \noindent{\bf Notation:} We write $\Gthree'$ for the set of teams $\T

4060: \in \Gthree$ such that $\down_2(\T) \neq \emptyset$.

4061: 

4062: \smallskip

4063: 

4064: We shall need the following obvious separation property. 

4065:  

4066:  

4067: \begin{lemma}\label{separate} 

4068: Consider $\T \in \Gthree'$.  If a

4069: path $p$ in $\F$ is to the left of $\plT$ and a path $q$ is the right

4070: of $\prT$,

4071: then there is no corridor connecting $p$ to $q$ at any time $t<\ttwo$. 

4072: \end{lemma} 

4073: 

4074: \begin{proof}  The hypothesis $\down_2 (\T) \neq \emptyset$ implies

4075: that before $\ttwo$ the paths $\plT$ and $\prT$ are not in the same

4076: corridor.

4077: \end{proof}

4078: 

4079: \begin{definition} \label{depthDef} 

4080: $\T_1\in\Gthree'$ is said to be {\em below} $\T_2\in\Gthree'$ if

4081: $p_l(\T_2)$ and $p_r(\T_2)$ both lie  between $p_l(\T_1)$ and

4082: $p_r(\T_1)$ in the left-right ordering described above.

4083: 

4084: $\T_1$ is said to be {\em to the left} of $\T_2$ if both  $p_l(\T_2)$ and $p_r(\T_2)$ 

4085: lie to the right of $p_r(\T_1)$. 

4086: 

4087:  

4088: We say that $\T$ is at {\em depth} $0$ if there are no teams above it. 

4089: Then, inductively, we say that a team is at depth $d+1$ if $d$ is the maximum 

4090: depth of those teams above $\T$.   

4091:  

4092: A {\em final depth} team is one with no teams below it. 

4093:  

4094: Note that there is a complete left-to-right ordering of teams $\T\in\Gthree$ at any given depth. 

4095: \end{definition} 

4096:  

4097:  

4098: \begin{lemma}\label{trapS_0} If there is a team from $\Gthree'$ below

4099: $\T\in\Gthree'$, then $\tone \ge \time (S_0)\ge \ttwo$. 

4100: \end{lemma}

4101: 

4102: \begin{proof} The first thing to note is that  if

4103:  $\time (S_0)$ were less than $\ttwo$, then  the narrow

4104: past of $\T$  at time $t_2(\T)$ must contain at least

4105: $\ll$ edges. This is because the length of the narrow past of $\T$ cannot

4106: decrease before $\tone$, and

4107: at $\time(S_0)$ the narrow past

4108: is the union of the intervals $C_{(\mu,\mu')}(2)$ with $(\mu,\mu')\in\T$,

4109: which has length at least $\ll$ since $\T$ is assumed not to be short.

4110: 

4111: Thus if $\time(S_0)<\ttwo$ then  we are in the non-degenerate

4112: situation of Definition \ref{PincerDef} and the defining property of $\ttwo$

4113: means that  before time $\ttwo$  no edge to the right of $\prT$ lies

4114: in the same

4115: corridor as all the colours of $\T$ (cf. Lemma \ref{separate}).  In

4116: particular this is true of

4117: the past of the reaper of $\T$ (assuming that it has a past at time

4118: $\ttwo$). On the

4119: other hand,   the reaper of $\T$ has a past in $S_0$ (by the very

4120: definition of a team), as do all of the colours of $\T$. And since they

4121: lie in a common corridor at $\time(S_0)$, they must also do so

4122: at all times up to $\tone$. This contradiction implies

4123: that in fact $\time(S_0)\ge\ttwo$. 

4124:  

4125: Consider Figure \ref{figure:Nest}.  

4126: Suppose that $\T'\in\Gthree'$ is below $\T$. The proof of Lemma \ref{GotBlock}

4127: tells us that there is a block of constant edges extending from the

4128: top of $\Pin_{\T'}$ containing the narrow past of $\T'$, and there is

4129: a similarly long block

4130: extending from the path $p_l^+(\T)$ at each subsequent time until

4131: $t_1(\T')$. Thereafter the future of the block is contained in the

4132: block of constant edges that evolves into the union of the

4133: $C_{(\mu,\mu')}(2) \subseteq \bot(S_0)$ with $(\mu,\mu') \in \T'$,

4134: which is long by hypothesis.

4135:  

4136:  At no time can this evolving block extend across $\plT$ 

4137: because by  definition the edges along $\plT$ are labelled by non-constant

4138: letters.  Thus the evolving

4139: block is trapped to the right of $\plT$ and to the left of 

4140: $\prT$. In particular, it must vanish entirely before the time

4141: at the top of the pincer $\Pin_\T$, which is no later than

4142: $\tone$ and therefore $\tone \ge \time (S_0)$. 

4143: \end{proof} 

4144:  

4145: The following is the main result of this section. 

4146: 

4147: 

4148:  

4149: \begin{lemma}   \label{Aget2Lemma} \label{chiT} There exist sets of

4150: colours $\chi_c(\T)$ and $\chi_{\delta}(\T)$ associated to each team

4151: $\T\in\Gthree'$ such that the sets associated to

4152: distinct teams are disjoint and the following inequalities hold. 

4153: 

4154: For each fixed team $\T_0 \in \Gthree'$ (of depth $d$ say),  

4155: the teams  of depth $d+1$ that lie below   $\T_0$ 

4156: may be described as follows: 

4157: \begin{enumerate} 

4158: \item[$\bullet$] There is at most one {\em distinguished team} $\T_1$,

4159: and

4160: $$ 

4161: \|\T_1\|\le 2B\Big(\ttt (1+ |\chi(\Pin_{\T_0})|) + T_0(|\chi_P(\T_0)| +

4162: 1)\Big).

4163: $$

4164: \item[$\bullet$] There are some number of final-depth teams. 

4165: \item[$\bullet$] For each of the remaining teams $\T$ we have 

4166:   $$|\down_2(\T_0) \cap\down_2(\T)| \le \ttt  \Big( 1 

4167: + |\chi_c(\T)| \Big) + T_0 \Big( |\chi_{\delta}(\T)| + 2 \Big).

4168: $$ 

4169: \end{enumerate}

4170: \end{lemma}  

4171: 

4172: \begin{proof}  The first thing to note is that if two teams $\T, \T'

4173: \in \Gthree'$ are at the same depth, then $\down_2(\T)$ and

4174: $\down_2(\T')$ are disjoint. Indeed if $\T$ is to the left of $\T'$,

4175: then at times before $\ttwo$ the paths $\plT$ and $p_l(\T')$ never lie in

4176: the same corridor.  Let $\T \in \Gthree'$ be a team of level $d+1$

4177: that is below $\T_0$ and consider the edge $e$  at the right end of a

4178: corridor earlier than $\ttwo$ that contains an edge in $\QT$. We are

4179: concerned with the fact that this edge may be in $\down_2(\T_0)$.  In

4180: this situation we say that $\T_0$ and $\T$ {\em double count} $e$.

4181: 

4182: 

4183: \begin{figure}[htbp] 

4184: \begin{center} 

4185:   

4186: \input{DoubleCount.eps_t} 

4187:   

4188: \caption{A depiction of double-counting} 

4189: \label{figure:doublecount} 

4190: \end{center} 

4191: \end{figure} 

4192:  

4193: Let $\T_1, \dots ,\T_r$ be the teams in $\Gthree'$ of depth $d+1$

4194: which double-count with $\T_0$, ordered from left to right, with the

4195: final-depth teams deleted.  We define $\chi_c(\T)$ to be empty for

4196: teams not on this list. $\T_1$ will be the distinguished team.

4197: 

4198: Since there is no double-counting between teams of the same level, 

4199:  the sets of times  at which 

4200: $\T_1, \dots,\T_r$ double-count with $\T_0$ must be disjoint. Indeed if 

4201:  $i < j$ then the set of times at which $\T_i$ double-counts 

4202:   with $\T_0$ is earlier than the 

4203: set of times at which $\T_j$ 

4204: double-counts with ${\T_0}$ (Lemma \ref{separate}). Moreover, 

4205:  the times for each $\T_i$ form an interval, which we denote $\I_i$.  

4206:  

4207: We assume $r\ge 2$ and describe the construction of 

4208: the sets $\chi_c(\T_i)$ and $\chi_{\delta}(\T_i)$ that account for

4209: double-counting.

4210: 

4211:  

4212: The first thing to note is that each $\I_i$ must be later than $t_2(\T_1)$, 

4213:  by Lemma \ref{separate}. 

4214: The second thing to note is that   the entire interval of time $\I_i$ 

4215: must also be earlier than $t_1(\T_1)$. Indeed if  some double-counting by 

4216: $\T_i$  and $\T_0$  were to occur  after $t_1(\T_1)$, then we would

4217: have $t_2(\T_k) > t_1(\T_1)$. But then

4218: $\time (S_0) > t_1(\T_1)$, so Lemma \ref{trapS_0} would imply  that

4219: there was no team below $\T_1$, contrary to hypothesis.  

4220: 

4221: We separately consider the intervals $\I_i \cap [ t_2(\T_1),t_3(\T_1)

4222: ]$ and  $\I_i \cap [ t_3(\T_1),t_1(\T_1) ]$, whose union is all of

4223: $\I_i$.

4224: 

4225: For that part of $\I_i$ before $t_3(\T_1)$, the proofs of

4226: the Pincer Lemma (Theorem \ref{PincerLemma}) and Proposition 

4227: \ref{prePincerLemma} tell us that colours in $\chi(\Pin_{\T_1})$

4228: will be consumed at the rate of at

4229: least one per $\ttt$ units of time. Define

4230: $\chi_c(\T_i)$ to be this set of consumed colours. We have

4231: $$

4232: \Big|\, \I_i \cap [ t_2(\T_1),t_3(\T_1) ]\, \Big| \leq \ttt (1 +

4233: |\chi_c(\T_i)|) .

4234: $$

4235: 

4236: Now consider  $\I_i \cap [ t_3(\T_1),t_1(\T_1)]$. Define

4237: $\chi_{\delta}(\T_i)$ as follows. The discussion in Definition

4238: \ref{PincerDef} shows that in any period of time of length $T_0$ in

4239: the interval $[t_3(\T_1),t_1(\T_1) ]$ at least one colour in

4240: $\chi_P(\T_1)$ disappears.  Let $\chi_{\delta}(\T_i)$ be

4241: the set of colours in $\chi_P(\T_1)$ which disappear during $\I_i

4242: \cap [t_3(\T_1), t_1(\T_1)]$ (these disappearances correspond to the

4243: discontinuities in the `path' $p_l^+(\T_1)$).  By construction, we

4244: then have\footnote{There is a 2 rather than the familiar 1 on the

4245: right to account for the colour containing

4246: $p_l(\T_1)$, which is not included in

4247: $\chi_P(\T_1)$; there might be up to $T_0$ corridors between

4248: $t_3(\T_1)$ and the top of $\Pin_{\T_1}$.}

4249: 

4250: \[	\Big| \, \I_i \cap  [ t_3(\T_1),t_1(\T_1)]\, \Big| \le

4251: T_0(|\chi_{\delta}(\T_i)| + 2),	\]

4252: and combining these estimates we have

4253: $$

4254: |\I_i| \le \ttt  \Big( 1 + |\chi_c(\T_i)| \Big) + T_0

4255: \Big(  |\chi_{\delta}(\T_i)| +  2\Big) , 

4256: $$

4257: as required. 

4258: Since the intervals $\I_i$ are disjoint, 

4259: the sets $\chi_c(\T_i),\, i=2,\dots,r$ are mutually disjoint. 

4260: And by construction, these sets are also disjoint from the sets 

4261: associated to teams other than 

4262: the $\T_i$ under consideration (i.e. those under other depth $d$  

4263: teams, or those of different 

4264: depths). The same considerations hold for the sets

4265: $\chi_{\delta}(\T_i),\, i=2,\ldots, r$.

4266: 

4267: In Figure \ref{figure:double}, the shaded region is where we recorded

4268: the regular disappearance of the colours forming $\chi_c(\T_i)$,

4269: whilst in Figure \ref{figure:doubletwo}, the shaded region is where we

4270: recorded the regular disappearance of the colours forming

4271: $\chi_{\delta}(\T_i)$.

4272: 

4273: \begin{figure}[htbp] 

4274: \begin{center} 

4275:   

4276: \input{Double.eps_t} 

4277:   

4278: \caption{Finding the colours $\chi_c(\T_i)$} 

4279: \label{figure:double} 

4280: \end{center} 

4281: \end{figure} 

4282: 

4283: 

4284: \begin{figure}[htbp]

4285: \begin{center}

4286: 

4287: \input{Doubletwo.eps_t}

4288: 

4289: \caption{Finding the colours $\chi_{\delta}(\T_i)$}

4290: \label{figure:doubletwo}

4291: \end{center}

4292: \end{figure}

4293:  

4294: It remains to establish the inequality 

4295: $$ 

4296: \|\T_1\|\le 2B\Big( \ttt (|\chi(\Pin_{\T_0})| + 1) +

4297: (|\chi_P(\T_0)|+1) \Big). 

4298: $$ 

4299: We first note (as in the proof of Lemma \ref{trapS_0})  that 

4300: $\ET_1$ is trapped between $\plT$ and $\prT$, so it must be consumed

4301: entirely between  the times $t_1(\T_1)$ and  $t_1(\T_0)$. But by the

4302: Bounded Cancellation Lemma, the length of  the future of $\ET_1$ can

4303: decrease by at most $2B$ at each step in time. Therefore  $\|\T_1\|\le

4304: 2B (t_1(\T_0) - t_1(\T_1))$.

4305:  

4306: $\T_1$ is assumed not be final-depth, so from Lemma \ref{trapS_0} we have 

4307: $t_2(\T_0) \le \time (S_0) \le t_1(\T_1)$. By combining these  

4308:  inequalities with Lemmas \ref{t1-t2Lemma} and \ref{t1t3forTeam}

4309: we obtain: 

4310: \begin{eqnarray*} 

4311: \|\T_1\| &\le& 2B\, \Big(t_1(\T_0) - t_1(\T_1)\Big)\\  

4312: & \leq & 2B\, \Big(t_1(\T_0) - \time (S_0)\Big)\\  

4313: & \leq & 2B\, \Big(t_1(\T_0) - t_2(\T_0)\Big)\\  

4314: & \leq & 2B \Big[ \ttt \Big(1+|\chi(\Pin_{\T_0})|\Big) + T_0

4315: \Big( |\chi_P(\T_0)| + 1 \Big) \Big]. 

4316: \end{eqnarray*} 

4317: \end{proof} 

4318:  

4319: \begin{corollary} \label{Firstdown2sum} Summing over the set of teams $\T \in \Gthree'$ that

4320: are not distinguished, we get 

4321: $$ 

4322: \sum_\T \Big|\down_2 (\T)\Big| \le2\,\Big| \bigcup_\T \down_2(\T)\Big|

4323: +\sum_{\T}  \ttt  \Big( 1 

4324: + |\chi_c(\T)| \Big) + \sum_{\T} T_0 \Big( |\chi_{\delta}(\T)| + 2

4325: \Big) .

4326: $$ 

4327: \end{corollary} 

4328:  

4329: \begin{proof} Suppose $\T \in \Gthree'$ of depth $d+1$ is not

4330: final-depth and not distinguished, and that $\T$ double-counts with

4331: some $\T_0$ of depth $d$ above it.  Then, by Lemma \ref{Aget2Lemma},

4332: we have

4333: \begin{eqnarray*}

4334: |\down_2(\T)| & = & |\down_2(\T) \smallsetminus \down_2(\T_0)| +

4335: |\down_2(\T) \cap \down_2(\T_0)| \\

4336: & \leq & |\down_2(\T) \smallsetminus \down_2(\T_0)| + \ttt (1 +

4337: |\chi_c(\T)|) + T_0 (2 + |\chi_{\delta}(\T)|).

4338: \end{eqnarray*}

4339: Suppose that $\T' \in \Gthree'$ is a team of depth $k < d$ and that

4340: $\T'$ is above $\T$. If $\T$ double-counts with $\T'$ at time $t$,

4341: then $\T$ double-counts with $\T_0$ at time $t$, by Lemma

4342: \ref{separate}.  Therefore, the set of edges that $\T$ double-counts

4343: with any team of lesser depth is exactly $\down_2(\T) \cap

4344: \down_2(\T_0)$.

4345: 

4346: Thus we have accounted for all double-counting other than than

4347: involving final depth teams.  The factor $2$ in the statement of the

4348: corollary accounts for this.

4349: \end{proof} 

4350: 

4351: And summing over the same set of teams again, we obtain: 

4352: \begin{corollary} \label{downbound} 

4353: $$ 

4354: \sum_\T |\down_2(\T)| \ \le \ \n (2 + 3\ttt + 5T_0). 

4355: $$

4356: \end{corollary} 

4357: \begin{proof} The sets of colours $\chi_c(\T)$ and $\chi_{\delta}(\T)$

4358: are disjoint. And the

4359: union of the sets $\down_2(\T)$  is a subset of $\partial\Delta$. The

4360: set of all colours and the  set of edges in $\partial\Delta$ each have

4361: cardinality at most $\n$.   And the number of teams is less than $2\n$

4362: (Lemma \ref{allIn}).

4363: \end{proof} 

4364:  

4365: 

4366:   

4367: \section{The Bonus Scheme} \label{BonusScheme} 

4368:  

4369:  

4370: We have defined teams and obtained a global bound on 

4371: $\sum\|\T\|$. 

4372: If $\cmm$ is non-empty then $(\mu,\mu')$ is a member or  

4373:  virtual member of a  unique team. 

4374: If this team is such that $\tone \ge \time (S_0)$, then no member of 

4375: the team is virtual  and we have the inequality

4376: $$\|\T\|>\sum\limits_{(\mu,\mu') \in \T}|\cmm| - B$$

4377:  established in Lemma \ref{t1high}.

4378:  We indicated following this lemma how this inequality

4379:  might fail in the case where

4380:   $\tone < \time (S_0)$. In this section we take up this

4381:   matter in detail

4382:    and introduce a {\em bonus scheme} that

4383:  assigns additional edges to teams in order to compensate for the possible failure

4384:  of the above inequality when   $\tone < \time (S_0)$.

4385:  

4386: By definition, at time $\tone$ the reaper $\r=\rT$ lies

4387:  immediately to the right of $\ET$.  The edges of 

4388: $\ET$ not consumed from the right by $\r$ by  $\time (S_0)$

4389:  have a preferred future in $S_0$ 

4390: that lies in $\cmm$ for some member $(\mu,\mu')\in\T$. 

4391: However, not all of the edges of 

4392: $\cmm$ need arise in this way:

4393:  some may not have  a constant ancestor at time $\tone$.

4394: And  if $(\mu,\mu')$ is only a virtual member of $\T$,

4395:  then no edge of $\cmm$ lies in the 

4396: future of $\ET$. The  {\em bonus} edges in $\cmm$

4397: are a certain subset of those  that do not have a constant 

4398: ancestor at time $\tone$. They are defined

4399: as follows.

4400: 

4401: \begin{definition} Let $\T$ be a team with $\tone < \time(S_0)$

4402: and consider a time $t$ with $\tone < t < \time(S_0)$.

4403: 

4404: The {\em swollen future} of $\T$ at time $t$ is the interval 

4405: of constant edges beginning immediately to the left of the pp-future of $\rT$. 

4406:  

4407: Let $e$ be a non-constant edge that lies immediately to the left of the 

4408: swollen future of $\T$ but whose ancestor is not a 

4409: right para-linear edge in this position. If $e$ is a right para-linear and

4410:  the (constant) rate 

4411: at which $e$ adds letters to the swollen future of $\ET$ is greater 

4412: than the (constant) rate at  which the future of the reaper cancels letters 

4413: in the future of $\ET$, then we define $e$ to be 

4414: a {\em rascal}; if $e$ is right-fast then we define it to be a {\em terror}.

4415: In both cases,

4416:  we define the

4417: {\em bonus provided by $e$} to be the set of edges in the swollen future

4418: of $\T$ in $S_0$ that have $e$ as their most recent non-constant

4419: ancestor, and are eventually consumed by $\rT$. 

4420: 

4421: The set $\bonusT$ is the union of the bonuses provided to $\T$ by all

4422: rascals and terrors.

4423: \end{definition} 

4424:  

4425: \begin{lemma}  \label{C1toTeamLength} 

4426: For any team $\T$, 

4427: \[      \sum_{(\mu,\mu') \in \T \mbox{ \tiny{or} } (\mu,\mu')

4428:  \vin T}|\cmm| \leq \|\T\| + |\bonus(\T)| + B.    \] 

4429: \end{lemma}

4430: 

4431: \begin{proof}

4432: If $\tone \geq \time(S_0)$, this follows immediately from Lemma

4433: \ref{t1high}.  If $\tone < \time(S_0)$ then at each step in time

4434: between $\tone$ and $\time(S_0)$ the only possible cause of growth in

4435: the length of the swollen future of the team is the possible action of

4436: a rascal or terror if such is present at that time.  (There is no

4437: interaction of the swollen future with the boundary or singularities,

4438: because of the exclusions in the second paragraph of Definition

4439: \ref{newTeams}.)

4440: 

4441: The swollen future has length $\|\T\|$ at time $\tone$ and length at

4442: least  $\sum |C_{(\mu,\mu')}(2)|$ at $\time(S_0)$. By definition,

4443: $|\bonus(\T)|$ is a bound on the growth in length between these

4444: times. (The summand $B$ is thus unnecessary in the case  $\tone <

4445: \time(S_0)$.)

4446: \end{proof}

4447:  

4448:  The following lemma shows that our main task in this

4449:  section will be to analyse the behaviour of rascals.

4450: 

4451: \begin{lemma} The sum of the lengths of the bonuses 

4452: provided to all teams by terrors is less than $2M\n$.

4453: \end{lemma}  

4454: 

4455: \begin{proof} Since it is right-fast, a terror will be separated

4456: from the team to which it is associated after one unit of

4457: time, and hence the bonus that it provides is less than $M$.

4458: There is at most one terror for each possible adjacency of colours

4459: and hence the total contributions of all terrors is less than

4460: $2M\n$.

4461: \end{proof}

4462:  

4463: The typical pattern of influence of rascals on a team 

4464: is shown in Figure \ref{belowS0};

4465: there may be several times at which rascals

4466: appear at the left of $\T$ and provide a  

4467: bonus for the team before being consumed from the left (or otherwise detached 

4468: from the team).  

4469:  

4470: \begin{figure}[htbp] 

4471: \begin{center} 

4472:   

4473: \input{belowS0.eps_t} 

4474:   

4475: \caption{The generic situation below $\time(S_0)$.} 

4476: \label{belowS0} 

4477: \end{center} 

4478: \end{figure}

4479: 

4480: \begin{definition}[Rascals' Pincers] \label{RascalPin}

4481: We fix a team $\T$ with $\tone < \time(S_0)$

4482:  and consider the interval of time $[\tau_0(e),\tau_1(e)]$,

4483: where $\tau_0(e)$ 

4484: is the time at which a rascal $e$ appears at the left end of the swollen 

4485: future of $\T$, and $\tau_1(e)$ is the time at which its future is no 

4486: longer to the immediate left of the future of the swollen future of $\T$.  

4487:  

4488: In the case where the pp-future $\hat e$ of $e$ at time  $\tau_1(e)$

4489: is cancelled from the left by an edge $e'$, we define

4490: $\tau_2(e)$ to be the earliest time when the pasts of $\hat e$ and $e'$ are

4491: in the same corridor. The path in $\F$ that traces the pp-future of

4492: $e$ up to  $\tau_1(e)$ is denoted $p_e$ and the path following through the

4493: ancestors of  $e'$ from

4494:  $\tau_2(e)$ to $\tau_1(e)$ is denoted  $p_e'$.

4495:  The pincer\footnote{to lighten the terminology, here we allow the

4496:  degenerate case where the ``pincer" has

4497:   no colours  other than those of $e$

4498:  and $e'$}

4499:   formed by $p_e$ and $p_{e}'$  with base at time  $\tau_2(e)$

4500: is denoted  $\Pin_e$. 

4501: \end{definition}

4502:  

4503: \begin{lemma} \label{lowRascals} 

4504: The total of all bonuses provided to all teams by  rascals $e$ with 

4505:  $\tau_1(e)\le\time(S_0)$ is less than $(3\ttt + 2T_0 +1)M\n$.  

4506: \end{lemma} 

4507:  

4508: \begin{proof} Consider a rascal $e$. We defer the case where $e$ hits

4509: a singularity or the  boundary. If this does not happen, the

4510: pp-future $\hat e$ of $e$ at time $\tau_1(e)$ is cancelled from the left

4511: by an edge $e'$ (which  is right-fast since $e$ is not

4512: constant). We consider the pincer $\Pin_e$ defined above.

4513: The presence of the swollen future of $\T$ at the top of the

4514: pincer allows us to apply the  Two Colour Lemma to conclude that

4515: $\tau_1(e) - T_0 \geq \time(S_{\Pi_e})$ (in the degenerate case

4516: discussed in the footnote, $ \time(S_{\Pi_e})$ is replaced by

4517: $\tau_2(e)$). And the

4518: Pincer Lemma tells us that 

4519: \[      \tau_1(e) - \tau_2(e) \leq \ttt \Big( 1 + |\chi(\Pin_e)| \Big)

4520: + T_0.       \]

4521: In fact, we could use $\tilde \chi(\Pin_e)$ instead of

4522: $\chi(\Pin_e)$ in this estimate because there cannot be any nesting

4523: amongst the pincers $\Pin_e$ with $\tau_1(e)\le \time(S_0)$,

4524: because nesting would imply that the swollen future of $\T$, which is

4525: immediately to the right of the lower rascal, would be trapped beneath

4526:  the upper pincer,

4527:   contradicting the fact that the team has a non-empty future in $S_0$. 

4528:  

4529: In the case where $e$ hits the boundary or is separated from the team

4530:  by a singularity (at time $\tau_1(e)$) we 

4531: define  $\tau_2(e)=\tau_1(e)$. No matter what the

4532: fate of $e$, we define

4533:  $\partial^e$ to be the set of edges in

4534:  $\partial\Delta$ at the left ends of corridors  

4535: containing the future of $e$ between 

4536: $\tau_0(e)$ and $\tau_2(e)$.

4537: The sets $\partial^e$ assigned to different rascals are disjoint,

4538: so summing over all rascals with $\tau_1(e)\le \time(S_0)$

4539: we have

4540: \begin{eqnarray*}

4541: \sum_e \Big(\tau_1(e) - \tau_0(e)\Big) & = & \sum_e (\tau_1(e) - \tau_2(e)) +

4542:  (\tau_2(e)-\tau_0(e))\\

4543: & \le & \sum_e \ttt\Big(1 + |\chi (\Pin_e)|\Big) + T_0 + |\partial^e|.

4544: \end{eqnarray*}

4545: Since the sets $\chi(\Pin_e)$ and $\partial^e$ are disjoint,

4546: the terms $\ttt |\chi (\Pin_e)|$ and $ |\partial^e|$ 

4547: contribute less than $(\ttt + 1)\n$ to this sum. And since the number of

4548:  rascals is bounded by the number of possible adjacencies of colours, the remaining

4549: terms contribute at most $(\ttt + T_0)2 \n$. Thus

4550: $$

4551: \sum_e  \Big(\tau_1(e) - \tau_0(e)\Big)\ \le\ (3\ttt + 2T_0 + 1) \n .

4552: $$

4553: The bonus produced by each rascal in each unit of time is less than $M$, so  

4554: the lemma  is proved.

4555: \end{proof} 

4556: 

4557: It remains to consider the size of the bonuses provided by rascals $e$ with 

4558:  $\tau_1(e)>\time (S_0)$. 

4559:  

4560: 

4561: The bonuses that are not accounted for in Lemma \ref{lowRascals}

4562: reside in blocks  of constant edges along $\bot(S_0)$ each of which is the swollen 

4563: future of some team, with

4564:  a \rpl letter at its left-hand end (the pp-future of

4565:   a rascal) and a \lpl letter at its left-hand end (the pp-future of

4566:   the team's reaper).  

4567:  

4568:  \begin{definition}

4569: A {\em left-biased} rascal $e$ is one with

4570:  $\tau_1(e) > \time(S_0)$ that satisfies the following properties: 

4571: \begin{enumerate} 

4572: \item[1.] the pp-future of the rascal  is (ultimately) 

4573: consumed from the left by an edge of $S_0$,  

4574: \item[2.] the swollen future of  $\T$ at 

4575: time $\tau_1(e)$ has length at least $\ll$ and  

4576: the pp-future of the reaper $\rT$ is still immediately to its right.

4577: \end{enumerate} 

4578:  \end{definition}

4579:  

4580: \def\life{\text{\rm{life}}}  

4581: \def\B{\text{\euf{B}}} 

4582:  

4583:  

4584: \begin{definition} Let $\B\subset\bot(S_0)$ be an interval of constant edges with a  

4585: right para-linear letter at its left-hand end and a left-linear letter $\r$ at its right-hand end. We 

4586: say that $\B$ is {\em right biased} if $\r$ is ultimately consumed by an edge (to its right) 

4587: in $S_0$. We define $\life(\B)$ to be the difference  between $\time(S_0)$ and  the time at which 

4588: the \lpl letter $\rho$ is consumed. 

4589: And we define the {\em effective volume} of $\B$ to be the number of edges in $\B$ 

4590: that are ultimately consumed by $\rho$. 

4591: \end{definition} 

4592:  

4593: We have the following tautologous tetrad of possibilities covering the swollen teams whose bonuses are 

4594: not entirely accounted for by Lemma \ref{lowRascals}. 

4595:  

4596: \begin{lemma} \label{tetrad} Let $\B\subset\bot(S_0)$ be an interval of constant edges that is 

4597: the swollen future of a team with a rascal 

4598: at its left-hand end and a \lpl letter $\r$ at its right-hand end.  

4599: Then at least one of the following holds: 

4600: \begin{enumerate}  

4601: \item[\rm{(i)}] the length of $\B$ is at most $\ll$;

4602: \item[\rm{(ii)}] $\B$ is the swollen future of a team with a left-biased rascal; 

4603: \item[\rm{(iii)}] $\B$ is right-biased;

4604: \item[\rm{(iv)}] neither of the non-constant letters at the ends of $\B$

4605:  is ultimately consumed by an edge of $S_0$. 

4606: \end{enumerate} 

4607: \end{lemma} 

4608: 

4609: We note here that when the length of $\B$ is at most $\ll$ then we

4610: have a short team, and we have already accounted for short teams.

4611: The following three lemmas correspond to eventualities (ii) to (iv).

4612:   

4613: \begin{lemma} \label{highRascals} 

4614: The sum of the bonuses provided to all teams by left-biased rascals is

4615: less than $(2M + 6M\ttt + 4MT_0 + 2\ll + 6B\ttt + 4BT_0) \n$.

4616: \end{lemma} 

4617: 

4618:  

4619: \begin{proof} The proof of this result is similar to the work done in

4620: the previous section.  We have a pincer $\Pin_e$ associated to the

4621: rascal $e$.  Since we are only concerned with the times when the

4622: rascal is immediately adjacent to a block of constant letters, it must

4623: be that at time $\tau_1(e) - T_0$ either we are below $\tau_0(e)$ or

4624: $\time(S_{\Pin_e})$ (cf. Definition \ref{PincerDef}). Therefore the

4625: following is an immediate consequence of the Pincer Lemma.

4626: \[	\tau_1(e) - \tau_2(e) \leq \ttt (1 + |\chi(\Pin_e)|) + T_0 .

4627: \]

4628: It now suffices to bound the amount of time for which $e$ is adjacent

4629: to the narrow past of $\B$ before $\tau_2(e)$.  We define $\tau_0'(e)$

4630: to be the latest time when the rascal $e$ has contributed less than

4631: $\ll$ edges to $\bonus(\T)$.  Then the bonus provided by $e$ is at

4632: most $M(\tau_1(e) - \tau_0'(e)) + \ll$. As in the previous section, we

4633: define $\down_2(e)$ to be those edges on the left end of corridors

4634: containing $e$ at times before $\tau_2(e)$ but after $\tau_0'(e)$.

4635: Just as in Lemma \ref{Aget2Lemma} and the corollaries immediately

4636: following it, we then have a notion of {\em depth} of

4637: rascals describing the nesting of the pincers $\Pin_e$\footnote{One

4638: extends the paths $p_e$ and $p_e'$ of Definition 

4639: \ref{RascalPin} back in time to $\partial\Delta$ so as to define the

4640: order definining depth}. We also have

4641: {\em distinguished} rascals (corresponding to the distinguished teams

4642: in Lemma \ref{Aget2Lemma}), and proceeding as in the proof of Lemma

4643: \ref{Aget2Lemma} we get the following estimates:

4644: 

4645: if $e_1$ is a distinguished rascal of depth $d+1$ and $e_0$ is the

4646: rascal of depth $d$ above it, then the bonus provided by $e_1$ is at

4647: most $2B\Big( T_1 (1 + |\chi(\Pin_{e_0}))|) + T_0 \Big)$, since all of

4648: the bonus provided by $e_1$ must disappear before $\tau_1(e_0)$;

4649: 

4650: for other rascals $e$ of depth $d+1$ which are below $e_0$ we have a

4651: set of colours $\chi_c(e)$, disjoint for distinct teams such that

4652: \[	|\down_2(e) \cap \down_2(e_0)| \leq T_1(1 + |\chi_c(e)|) +

4653: T_0.	\]

4654: Therefore, summing over the set of rascals which are not distinguished

4655: we get (cf Corollary \ref{Firstdown2sum})

4656: \[ \sum_e |\down_2(e)| \leq 2\Big| \bigcup_e \down_2(e)\Big| + \sum_e \Big

4657: ( \ttt (1 + |\chi_c(e)|) + T_0 \Big).	\]

4658: And summing over the same set of rascals, we get 

4659: \[ \sum_e |\down_2(e)| \leq (2 + 3\ttt + 2T_0)\n .	\]

4660: Therefore, for undistinguished rascals, we have

4661: \begin{eqnarray*}

4662: \sum_e \tau_1(e) - \tau_0'(e) & = & \sum(\tau_1(e) - \tau_2(e)) +

4663: \sum(\tau_2(e) - \tau_0'(e))\\

4664: & \leq & (3\ttt + 2T_0)\n + (2 + 3\ttt + 2T_0)\n,

4665: \end{eqnarray*}

4666: and so the contribution of all left-biased rascals is at most

4667: \[	\Big( (2 + 6\ttt + 4T_0)M + 2\ll + 6B\ttt + 4BT_0 \Big) \n,	\]

4668: as required.

4669: \end{proof}

4670:  

4671:  

4672: \begin{lemma} \label{RightBiased}  The sum $\sum \life(\B)$ over those

4673: $\B$ that are right-biased but  do not satisfy conditions (i) or (ii)

4674: of Lemma \ref{tetrad} is at most $(3\ttt B + 2T_0B)\n$.

4675: \end{lemma} 

4676: 

4677: \begin{figure}[htbp] 

4678: \begin{center} 

4679:   

4680: \input{Rteam.eps_t} 

4681: 

4682: \caption{A depiction of a right-biased team.} 

4683: \label{r-teams} 

4684: \end{center} 

4685: \end{figure}

4686: 

4687: \begin{proof}

4688: Once again, as in Lemmas \ref{lowRascals} and \ref{highRascals}, we

4689: obtain compensation for the continuing existence of a non-constant

4690: letter by using the Pincer Lemma to see that colours must be consumed

4691: at a constant rate in order to facilitate the life of $\r$. Thus we

4692: consider the left-fast edge that consumes the pp (i.e. left-most

4693: non-constant) future of $\r$; this edge is denoted $e(\r)$ in Figure

4694: \ref{r-teams}. The Pincer Lemma and the 2 Colour Lemma tell us that if

4695: $\Pin_{e(\r)}$ is the

4696: pincer associated to these paths (with $S_0$ at the bottom) then 

4697: \[      \life(\B) \leq \ttt (1 + |\chi(\Pin_{e(\r)})|) + T_0.   \]

4698: Suppose that $\B$ and $\B'$ are two right-biased blocks with

4699: associated edges $e(\r)$ and $e(\r')$ consuming their reapers. We claim

4700: that the sets $\chi(\Pin_{e(\r)})$ and $\chi(\Pin_{e(\r')})$ are

4701: disjoint.  The key point to observe is that since we are not in case

4702: (ii) of Lemma \ref{tetrad} the length of the swollen future of $\B$

4703: increases from $\time(S_0)$ to the top of $\Pin_{e(\r)}$; since $\B$

4704: had length at least $\ll$, we therefore have a block of more than

4705: $\ll$ of more than $\ll$ constant edges at the top of

4706: $\Pin_{e(\r)}$. Thus the pincers associated to $\B$ and $\B'$ are

4707: either disjoint or nested. Hence $\chi(\Pin_{e(\r)})$ and

4708: $\chi(\Pin_{e(\r')})$ are disjoint. Thus summing over all right-biased

4709: blocks $\B$ we obtain

4710: \[      \sum_{\B \mbox{ right-biased}} \life(\B) \leq (3\ttt B + 2T_0B)

4711: \n, \]

4712: as required.

4713: \end{proof} 

4714: Since any letter consumes less than $M$ constant letters in any unit of time, we conclude: 

4715: \begin{corollary} \label{rightCor} The sum  of the effective volumes of  

4716: all blocks that are right-biased but  do not satisfy conditions (i) and (ii) 

4717: of Lemma \ref{tetrad} is at most $(3M\ttt B + 2MT_0B)\n$.  

4718: \end{corollary} 

4719:  

4720: \begin{lemma} The sum of all blocks that satisfy condition (iv) of  

4721: Lemma \ref{tetrad} is at most $(2B +1)\n$. 

4722: \end{lemma} 

4723:  

4724: \begin{proof} Possibility (iv) involves several subcases: the key

4725: event which halts the growth of the swollen future of $\B$ may be a

4726: collision with $\partial\Delta$ or a singularity;  it may also be that

4727: the key event is that the future of the rascal or reaper adjacent to

4728: $\B$ is cancelled by an edge that is not in the future of $S_0$.  

4729:  

4730: But no matter what these key events may be, since we are in not in

4731: cases (ii) or (iii), associated to the blocks in case (iv) we have the

4732: following set of paths partitioning that part of the diagram $\Delta$

4733: bounded by $S_0$ and the arc of $\partial\Delta$ connecting the

4734: termini of the edges at the ends of $S_0$: 

4735: 

4736: 

4737: The path $\pi_l$ begins at $\time (S_0)$ and follows the pp-future of

4738: the rascal at the right-end of the future of $\B$ until it hits the

4739: boundary, a singularity, or else is cancelled by an edge $\e_l$ not in

4740: the future of $S_0$; if it hits the boundary, it ends; if it hits a

4741: singularity, $\pi_l$ crosses to the bottom of the corridor $S$ on the

4742: other side of the singularity, and turns left to follow $\bot(S)$ to

4743: the boundary (see Figure \ref{Pi_lOne}); if $\e_l$ cancels with the

4744: pp-future of the rascal, then $\pi_l$ follows the past of $\e_l$

4745: backwards in time to the boundary (see Figure \ref{Pi_lTwo}). 

4746:  

4747: 

4748: \begin{figure}[htbp] 

4749: \begin{center} 

4750:   

4751: \input{PiLone.eps_t} 

4752:   

4753: \caption{The path $\pi_l$ hits a singularity.} 

4754: \label{Pi_lOne} 

4755: \end{center} 

4756: \end{figure} 

4757: 

4758: \begin{figure}[htbp]

4759: \begin{center}

4760: 

4761: \input{PiLtwo.eps_t}

4762: 

4763: \caption{The path $\pi_l$ in cancelled from outside of the future of

4764: $S_0$}

4765: \label{Pi_lTwo}

4766: \end{center}

4767: \end{figure}

4768:  

4769: 

4770: The path $\pi_r$ describing the fate of $\r$ is defined similarly

4771: (except that it turns right if it hits a singularity). 

4772:  

4773: It is clear from the construction that no two of these paths can cross, thus we have the  partition  

4774: represented schematically in Figure \ref{partition}. 

4775:  

4776: \begin{figure}[htbp]

4777: \begin{center}

4778: 

4779: \input{partition.eps_t}

4780: 

4781: \caption{The schematic partition of $\Delta$ by the paths $\pi_l$ and

4782: $\pi_r$.}

4783: \label{partition}

4784: \end{center}

4785: \end{figure}

4786: 

4787: \def\bdy{\text{\rm{bdy}}} 

4788: \def\up{\text{\rm{up}}} 

4789: Given a swollen team $\B$ of type (iv), we follow the swollen future of $\B$ until its flow is 

4790: interrupted (at time $\iota(\B)$, say) 

4791: by meeting a singularity, the boundary of $\Delta$, or else its rascal or reaper is cancelled. 

4792: Consider the set of corridors that contain some component of the swollen future of $\B$ 

4793: after $\iota(\B)$. Consider also the set of edges $\bdy(\B) \subseteq

4794: \partial\Delta$ 

4795: that lie in the swollen future of $\B$. We keep account of the set of corridors by recording 

4796: the set of their ends on $\partial\Delta$, except that we ignore an end if we have to cross 

4797: a path $\pi_l$ or $\pi_r$ to reach it. Note that at least one end of each corridor is recorded. 

4798: Let $\up(\B)\subset\partial\Delta$ denote the  set  of ends recorded.  

4799:   

4800: Since the sets $\bdy(\B)$ and $\up(\B)$ are contained in the portion of $\partial\Delta$ accorded 

4801: to $\B$ by the partition formed by the paths $\pi_l$ and $\pi_r$, the sets associated to different 

4802: $\B$ are disjoint. In each unit of time beyond $\iota(\B)$ each

4803: component of the swollen future of $\B$ can shrink by at most $2B$ (by

4804: Lemma \ref{BCL}).  The set $\up(\B)$ measures the sum of the number of

4805: components over all such times, and $|\bdy(\B)|$ is the number of

4806: uncancelled edges. Thus we see that the length of

4807: the swollen future of $\B$ at time $\iota(\B)$ is at most $2B|\up(\B)|

4808: + |\bdy(\B)|$. Finally, the continued presence of the rascal ensures

4809: that the swollen future of $\B$ grows in each interval of time from

4810: $\time(S_0)$ to $\iota(\B)$.  Thus it follows that the

4811: length of $\B$ is also bounded by this number. So summing over all

4812: $\B$ of type (iv) we have: 

4813: $$ 

4814: \sum |\B| \,\le\, \sum \Big(2B|\up(\B)| + |\bdy(\B)|\Big) \le (2B +1)\n , 

4815: $$ 

4816: as required. 

4817: \end{proof} 

4818:  Summarising the results of this section we have 

4819:  

4820: \bl \label{BonusBound} Summing over all teams that are not short, we

4821: have 

4822: \[      \sum_{\T}|\bonus(\T)| \leq \Big( \Bb \Big) \n .  \]

4823: \end{lemma} 

4824:  

4825:  

4826: \section{The proof of the Main Theorem} \label{summary} 

4827:  

4828: Pulling all of the previous results together, define  

4829: $$

4830: K_1 = \AFourC,

4831: $$

4832: and

4833: $$ 

4834: K = \K. 

4835: $$ 

4836:  

4837: \begin{theorem} \label{S0<=Kn}

4838: $|S_0| \leq K\n$. 

4839:  

4840: \Prf: 

4841: The corridor $S_0$ can be subdivided into distinct colours which form connected regions. 

4842:   Each colour $\mu$ can be partitioned into connected (possibly empty) regions $A_1(S_0,\mu),  

4843:   A_2(S_0,\mu), A_3(S_0,\mu), A_4(S_0,\mu)$ and $A_5(S_0,\mu)$.  By Lemma \ref{A1A5Lemma}, 

4844:   Proposition \ref{SummaryLemma}, Lemma \ref{A3Lemma}, Proposition \ref{A2Prop} and Lemma \ref{A1A5Lemma}, respectively, 

4845: \begin{eqnarray*} 

4846: \sum_{\mu \in S_0}|A_1(S_0,\mu)|  & \leq & C_0\n ,\\ 

4847: \sum_{\mu \in S_0}|A_2(S_0,\mu)|  & \leq &  K_1  \n ,\\ 

4848: \sum_{\mu \in S_0}|A_3(S_0,\mu)|  & \leq & (2B+1)\n ,\\ 

4849: \sum_{\mu \in S_0}|A_4(S_0,\mu)|  & \leq &  K_1 \n,\mbox{ and} \\ 

4850: \sum_{\mu \in S_0}|A_5(S_0,\mu)|  & \leq & C_0\n . 

4851: \end{eqnarray*} 

4852: Summing completes the proof of Theorem \ref{S0<=Kn}.

4853: \et

4854: Since there are at most $\frac{\n}{2}$ corridors in $\Delta$,  

4855: \[      \mbox{Area}(\Delta) \leq \frac{K}{2}\n^2 ,       \] 

4856: which proves the Main Theorem. 

4857:  

4858:  

4859: \section{Glossary of Constants} 

4860:  

4861:  

4862: $B$ -- the Bounded Cancellation constant (Lemmas \ref{BCL} 

4863: and \ref{SingularityProp}). 

4864:  

4865: $C_0$ -- maximum distance a left-fast (right-fast)  

4866: letter can be from the left (right) edge of its colour if it is to be

4867: cancelled from the

4868: left (right) within the future of the corridor.  See Lemma \ref{A1A5Lemma}. 

4869:  

4870: $C_1$ -- an upper bound on the   

4871: lengths of the subintervals 

4872:  $C_{(\mu,\mu')}(1)$ of $A_4(S_0,\mu)$. By definition, $C_{(\mu,\mu')}(1)$ 

4873:  is consumed by $\mu'(S_0)$; it begins at the right end of

4874: $A_4(S_0,\mu)$ and ends   at the last non-constant letter. See Lemma

4875: \ref{C1Lemma}. Note that one can take $C_1=2mB^2$. 

4876:  

4877:  

4878: $M$ -- the maximum of the lengths of the images  $\phi(a_i)$ of the basis 

4879: elements $a_i$, i.e. the maximum length of $u_1, \ldots, u_m$ in

4880: the presentation $\P$ (see equation \ref{presentation}).

4881: 

4882: $M_{inv}$ -- the maximum of the lengths of $\phi^{-1}(a_i)$. 

4883:  

4884: $T_0$ -- the constant from the 2-Colour Lemma (Lemma

4885: \ref{TwoColourLemma}). For all positive words $U$ and $V$, if

4886: $U$ neuters  $V^{-1}$ then it does so in at most $T_0$ steps. 

4887: 

4888: $\hat{T_1}$ -- the constant from the Unnested Pincer Lemma, Theorem \ref{prePincerLemma}.

4889: 

4890: $T_1'$ -- the constant from Definition \ref{T1'Lemma}.  Recall that we stipulate that $T_1' \ge \hat{T_1}$.

4891: 

4892: $T_1 := T_1' + 2T_0$ -- $T_1$ is the constant from the Pincer Lemma, Theorem \ref{PincerLemma}. 

4893: 

4894: $C_4 := MM_{inv}$

4895: 

4896: $\ll := {\rm max} \{ 2B(T_0 + 1)+1, MC_4 \}$

4897: 

4898: Finally, $K_1$ is defined to be

4899: \[	\AFourC	,	\]

4900: and $K = 2C_0 + 2K_1 + 2B + 1$.

4901: 

4902:  

4903:  

4904: \begin{thebibliography}{99} 

4905: \bibitem{BestICM} M. Bestvina, The topology of $\text{\rm{Out}}(F_n)$,

4906: in {\em Proceedings of ICM, Bejing 2002, Vol.II},

4907: Higher Education Press, Bejing, 2002. pp. 373--384.

4908: \bibitem{BF}  M. Bestvina and M. Feighn, A combination theorem 

4909:  for negatively curved groups, \textit{J. Diff. Geom.}, {\bf 35}

4910: (1992), 85--101.

4911: \bibitem{BFH} M. Bestvina, M. Feighn and M. Handel, The Tits 

4912: alternative for $\text{\rm Out}(F_n)$ I: Dynamics of exponentially growing 

4913: automorphisms, \textit{Ann. of Math. (2)}, {\bf 151} (2000), 517--623. 

4914: \bibitem{BFH2} M. Betvina, M. Feighn and M. Handel, The Tits

4915: alternative for $\text{\rm Out}(F_n)$ II: A Kolchin type theorem,

4916: preprint. 

4917: \bibitem{BH2} M. Bestvina and M. Handel, Train tracks and 

4918: automorphisms of free groups, \textit{Ann. of Math. (2)}, {\bf 135}, 

4919: 1--51. 

4920: \bibitem{BB} N. Brady and M.R. Bridson, On the absence of 

4921: biautomaticity for graphs of abelian groups, preprint. 

4922: \bibitem{B-plms} M.R. Bridson, Polynomial Dehn functions and the 

4923: length of asynchronously automatic structures,  

4924: \textit{Proc. London Math. Soc.(3)}, {\bf 85} (2002), 441--466. 

4925: \bibitem{mb-shs} M.R. Bridson, On the subgroups of semihyperbolic

4926: groups,   

4927: \textit{Monog. L'Enseign. Math.}, {\bf 38} (2001), 85--111. 

4928: \bibitem{steer} M.R. Bridson, The geometry of the word problem, in

4929: \textit{``Invitations to geometry and topology"} (M.R. Bridson and

4930: S.M. Salamon, eds.), Oxford University Press, 2002.

4931: \bibitem{BG} M.R. Bridson and S.M. Gersten, The optimal isoperimetric 

4932: inequality for torus bundles over the circle, 

4933: \textit{Quart. J. Math. Oxford Ser. (2)}, {\bf 47} (1996), 1--23. 

4934: \bibitem{BGrovesII} M. R. Bridson and D. P. Groves, The quadratic 

4935:  isoperimetric inequality for mapping tori of free group automorphisms II: 

4936:   The general case, preprint. 

4937: \bibitem{BGconstants} M. R. Bridson and D.P. Groves, Alogrithms and 

4938: constants for isoperimetric functions of free-by-cyclic groups, in preparation. 

4939: \bibitem{BH} M.R. Bridson and A. Haefliger, \textit{Metric spaces of 

4940: non-positive curvature}, Springer-Verlag, Berlin, 1999. 

4941: \bibitem{BR} M.R. Bridson and L. Reeves, On the absence of automaticity 

4942: in certain free-by-cyclic groups, in preparation. 

4943: \bibitem{Brink} P. Brinkmann, Hyperbolic automorphisms of free groups, 

4944: \textit{GAFA}, {\bf 10} (2000), 1071--1089. 

4945: \bibitem{Cooper} D. Cooper, Automorphisms of free groups have finitely 

4946:  generated fixed point sets, \textit{J. Algebra}, {\bf 111} (1987), 453--456. 

4947: \bibitem{E+} D.P.A. Epstein, J.W. Cannon, D.F. Holt, S.V.F. Levy, 

4948: M.S. Paterson and W.P. Thurston, \textit{Word processing in groups}, 

4949: Jones and Bartlett, Boston, 1992. 

4950: \bibitem{FH} M. Feighn and M. Handel, Mapping tori of free group 

4951: automorphisms are coherent, \textit{Ann. of Math. (2)}, {\bf 149} 

4952: (1999), 1061--1077. 

4953: \bibitem{Ge} S.M. Gersten, The automorphism group of a free group is 

4954: not a CAT$(0)$ group, \textit{Proc. Amer. Math. Soc.}, {\bf 121} 

4955: (1994), 999--1002. 

4956: \bibitem{Gromov} M. Gromov, Hyperbolic groups, in \textit{Essays in group theory} 

4957:  (S.M. Gersten, ed.), Springer Verlag, MSRI Publ. {\bf 8} (1987), 75-263. 

4958: \bibitem{vK} E.R. van Kampen, On some lemmas in the theory of groups, 

4959:  \textit{Amer. J. Math.}, {\bf 55} (1933), 268--273. 

4960: \bibitem{leeb} B. Leeb, 3-Manifolds with(out) metrics of nonpositive 

4961: curvature, \textit{Invent. Math.} {\bf 122} (1995), 277--289. 

4962: \bibitem{Lu} M. Lustig, Structure and conjugacy for automorphisms of free groups, preprint. 

4963: \bibitem{LS} R.C. Lyndon and P.E. Schupp, Combinatorial group theory, 

4964: Springer-Verlag, Berlin, 1977. 

4965: \bibitem{Mac} N. Macura, Quadratic isoperimetric inequality for 

4966: mapping tori of polynomially growing automorphisms of free groups, 

4967: \textit{GAFA}, {\bf 10} (2000), 874--901. 

4968: \bibitem{Sela} Z. Sela, The Nielsen-Thurston classification and 

4969: automorphisms of a free group I, \textit{Duke Math. J.}, {\bf 84} 

4970: (1996), 379--397. 

4971: \end{thebibliography} 

4972:  

4973:  

4974: \end{document} 

4975:  

4976:  

4977: