1: \documentclass{amsart}

2: \usepackage{amsmath, amssymb, amsthm, amscd, graphicx}

3: 

4: \textwidth=13.5cm

5: \textheight=20cm

6: 

7: % for 100% printing

8: \hoffset-1.2cm

9: \voffset+0.5cm

10: 

11: %for 110% printing

12: \hoffset-1.8cm

13: \voffset-0.6cm

14: 

15: \setlength{\unitlength}{1mm}

16: 

17: \theoremstyle{plain}

18: \newtheorem{prop}{Proposition}[section]

19: \newtheorem{coro}[prop]{Corollary}

20: \newtheorem{conj}[prop]{Conjecture}

21: \newtheorem{lemm}[prop]{Lemma}

22: \newtheorem{thrm}[prop]{Theorem}

23: 

24: \theoremstyle{definition}

25: \newtheorem{defi}[prop]{Definition}

26: \newtheorem*{propA}{Property A}

27: \newtheorem*{propC}{Property C}

28: \newtheorem{nota}[prop]{Notation}

29: \newtheorem{exam}[prop]{Example}

30: \newtheorem{rema}[prop]{Remark}

31: 

32: \numberwithin{equation}{section}

33: 

34: % bibliography

35: 

36: \def\Reff#1; #2; #3; #4; #5; #6; #7\par{%

37: \bibitem{#1} #2, {\it #3}, #4 {\bf #5} (#6) #7}

38: 

39: \def\Ref#1; #2; #3; #4\par{%

40: \bibitem{#1} #2, {\it #3}, #4}

41: 

42: % Specific macros of this text $$$$

43: 

44: \catcode`\¨=\active\def¨{~} 

45: 

46: \renewcommand{\a}{\mathtt{a}}

47: \newcommand{\A}{\mathtt{A}}

48: 

49: \renewcommand{\Bar}[1]{\overline{\vline width0pt

50: height5.5pt #1}}

51: %\renewcommand{\Bar}{\bar}

52: \renewcommand{\b}{\mathtt{b}}

53: \newcommand{\B}{\mathtt{B}}

54: \newcommand{\bbb}[3]{b_{#1,#2}(#3)}

55: \newcommand{\bbbb}[3]{b'_{#1,#2}(#3)}

56: \newcommand{\bx}{x}%_A_braid

57: \newcommand{\by}{y}

58: 

59: \renewcommand{\c}{\mathtt{c}}

60: \newcommand{\card}{\mathtt{\#}}

61: \newcommand{\cc}{c}

62: \newcommand{\ccc}[2]{\cc(\DD{#1}{#2})}

63: \newcommand{\CC}{C}

64: \newcommand{\CCC}{\mathbb{C}}

65: \newcommand{\CG}{\Gamma}

66: \newcommand{\cl}[1]{[#1]}

67: \newcommand{\Class}{C}

68: 

69: \newcommand{\D}{\Delta}

70: \newcommand{\Dd}[2]{\delta_{#1,#2}}

71: \newcommand{\Ddt}[2]{\widetilde\delta_{#1,#2}}

72: \newcommand{\DD}[2]{\D_{#1}^{#2}}

73: \newcommand{\dd}{d}%_Degree

74: \newcommand{\DDD}[2]{\Div(\D_{#1}^{#2})}

75: \newcommand{\DDDD}[2]{(\Div(\D_{#1}^{#2}),\nobreak\sm\nobreak)}

76: \renewcommand{\div}{\prec}

77: \newcommand{\Div}{\mathrm{Div}}

78: \newcommand{\DDiv}[1]{(\Div(#1),\nobreak\sm\nobreak)}

79: \newcommand{\dive}{\preccurlyeq}

80: \newcommand{\DivL}{D_L}

81: \newcommand{\DivR}{D_R}

82: \newcommand{\pz}{z}

83: 

84: \newcommand{\ea}{a}

85: \newcommand{\eb}{b}

86: \newcommand{\eg}{{\it e.g.}}

87: \newcommand{\etc}{{\it etc.}}

88: \newcommand{\ev}{\rho}%_Eigenvalue

89: \newcommand{\expo}{e}%_Number_of_occurrences

90: 

91: \newcommand{\ff}{f}

92: \newcommand{\flip}[1]{\phi_{#1}}

93: \newcommand{\first}[1]{(#1)_{\!_1}}

94: 

95: \newcommand{\g}{\gamma}

96: \let\ge=\geqslant

97: \newcommand{\G}{\Gamma}

98: \renewcommand{\gcd}{\mathrm{gcd}}

99: \renewcommand{\gg}{g}

100: \newcommand{\GG}[2]{\CG(\DD{#1}{#2})}

101: \newcommand{\gr}{>}

102: \newcommand{\gre}{\ge}

103: 

104: \newcommand{\hh}[1]{h_{#1}}

105: \newcommand{\hhh}[3]{\hh{#1}(\DD{#2}{#3})}

106: 

107: \newcommand{\ie}{{\it i.e.}}

108: \let\ince=\subseteq

109: \newcommand{\ind}{i}

110: \newcommand{\Ind}{\mathrm{ind}}

111: \newcommand{\indd}{j}

112: \newcommand{\Init}[1]{R_{#1}}

113: \newcommand{\Initt}[1]{R'_{#1}}

114: \newcommand{\Inittt}[1]{R''_{#1}}

115: \newcommand{\Int}[1]{[\![#1]\!]}

116: \newcommand{\inv}{^{-1}}

117: \newcommand{\Inv}{{}^{-1}}

118: \newcommand{\ik}{i}

119: \newcommand{\ip}{p}

120: \newcommand{\iq}{q}

121: \newcommand{\ir}{k}

122: \newcommand{\is}{\ell}

123: 

124: \newcommand{\kk}{k}

125: 

126: \renewcommand{\l}{\lambda}

127: \newcommand{\last}[1]{(#1)_{\!_\infty}}

128: \let\le=\leqslant

129: \newcommand{\lcm}{\mathrm{lcm}}

130: 

131: \newcommand{\m}{\mu}

132: \newcommand{\mm}{m}

133: 

134: \newcommand{\Na}[1]{\mathtt{\#}_{\!\mathstrut\ss1}#1}

135: \newcommand{\Nb}[1]{\mathtt{\#}_{\!\mathstrut\ss2}#1}

136: \newcommand{\nbpart}[1]{p(#1)}% number of partitions

137: \newcommand{\nn}{n}

138: \newcommand{\NN}{N}

139: \newcommand{\NNN}{N}

140: 

141: \renewcommand{\o}{\omega}

142: \newcommand{\op}{\cdot}

143: \newcommand{\opp}{\mathord{\cdot}}

144: 

145: \newcommand{\pp}{p}

146: \newcommand{\ppp}{p}

147: \newcommand{\pppp}{p}

148: 

149: \newcommand{\qq}{q}

150: 

151: \newcommand{\resp}{{\it resp.{~}}}

152: \newcommand{\rr}{r}

153: \newcommand{\RR}{\mathbb{R}}

154: 

155: \let\s=\sigma

156: \newcommand{\sh}{\mathrm{sh}}

157: \newcommand{\sm}{<}

158: \newcommand{\sme}{\le}

159: \renewcommand{\sp}{s}%_A_simple

160: \newcommand{\spp}{t}%_Another_simple

161: \renewcommand{\ss}[1]{\sigma_{#1}}

162: \newcommand{\sss}[2]{\sigma_#1^{#2}}

163: \renewcommand{\SS}[2]{S_{#1}^{#2}}

164: \newcommand{\SSS}[2]{\underline{S}_{#1}^{#2}}

165: \newcommand{\sx}{x}%_A_simple_factor_of_x

166: \newcommand{\Sym}[1]{\mathfrak{S}_{#1}}

167: 

168: \newcommand{\uu}{u}

169: 

170: \newcommand{\var}{u}

171: \newcommand{\vs}{\emph{vs.} }

172: \newcommand{\vv}{v}

173: 

174: \newcommand{\ww}{w}

175: \newcommand{\WW}[2]{\vec w_{#1}^{#2}}

176: 

177: \newcommand{\xx}{x}

178: \newcommand{\XX}{X}

179: 

180: \newcommand{\yy}{y}

181: 

182: \newcommand{\zz}{z}

183: 

184: \hyphenation{mon-oid mon-oids Lem-ma}

185: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

186: \begin{document}

187: 

188: \author{Patrick DEHORNOY}

189: \address{Laboratoire de Math\'ematiques Nicolas

190: Oresme UMR 6139\\ Universit\'e de Caen,

191: 14032~Caen, France}

192: \email{dehornoy@math.unicaen.fr}

193: \urladdr{//www.math.unicaen.fr/\textasciitilde dehornoy}

194: 

195: \title{Still another approach to the braid ordering}

196: 

197: \keywords{braid group; orderable group;

198: well-ordering; normal form; fundamental braid}

199: 

200: \subjclass{20F36, 05A05, 20F60}

201: 

202: \begin{abstract}

203: We develop a new approach to the linear ordering of the

204: braid group~$B_\nn$, based on investigating its

205: restriction to the set¨$\DDD\nn\dd$ of all divisors

206: of~$\DD\nn\dd$ in the monoid¨$B_\infty^+$, \ie, to

207: positive $n$-braids whose normal form has length at

208: most¨$\dd$. In the general case, we compute several

209: numerical parameters attached with the finite orders

210: $\DDDD\nn\dd$. In the case of $3$~strands, we moreover

211: give a complete description of the increasing

212: enumeration of~$\DDDD3\dd$. We deduce a new and

213: specially direct construction of the ordering

214: on~$B_3$, and a new proof of the result that its

215: restriction to~$B_3^+$ is a well-ordering of ordinal

216: type~$\omega^\omega$.

217: \end{abstract}

218: 

219: \maketitle

220: 

221: The general aim of this paper is to investigate the

222: connection between the Garside structure of Artin's

223: braid groups and their distinguished linear ordering

224: (sometimes called the Dehornoy ordering). This leads

225: to a new, alternative construction of the ordering.

226: 

227: Artin's braid groups¨$B_\nn$ are endowed with several

228: interesting combinatorial structures. One of them

229: stems from Garside's analysis

230: \cite{Gar} and is nowadays known as a Garside

231: structure¨\cite{Dgk, McC}. It describes $B_\nn$ as the group

232: of fractions of a monoid¨$B_\nn^+$ with a rich divisibility

233: theory. One of the outcomes of this theory is a unique

234: normal decomposition for every braid in¨$B_\nn$ in

235: terms of simple braids, which are the divisors of

236: Garside's fundamental braid~$\D_\nn$, a finite

237: family of¨$B_\nn^+$ in one-to-one correspondence

238: with the permutations of $\nn$~objects. One obtains a

239: natural graduation of the monoid¨$B_\nn^+$ by

240: considering the family~$\DDD\nn\dd$ of all divisors

241: of~$\DD\nn\dd$, which also are the elements

242: of~$B_n^+$ whose normal form has length at

243: most~$\dd$.

244: 

245: On the other hand, the braid groups are equipped with a

246: distinguished linear ordering, which is compatible with

247: multiplication on the left, and admits a simple

248: combinatorial characterization¨\cite{Dfb}: a

249: braid¨$\bx$ is smaller than another braid¨$\by$ if,

250: among all expressions of the quotient¨$\bx\inv\by$ in

251: terms of the standard generators¨$\ss i$, there exists

252: at least one expression in which the

253: generator¨$\ss\mm$ with maximal (or minimal)

254: index¨$\mm$ appears only positively, \ie,

255: $\ss\mm$ occurs, but $\ss\mm\inv$ does not. Several

256: deep results about that ordering are known, in

257: particular the fact that its restriction

258: to¨$B_\infty^+$ is a well-ordering, and a number of

259: equivalent constructions are known¨\cite{Dgr}.

260: 

261: Although both combinatorial in nature, the previous

262: structures remain mostly uncon\-nected---and connecting

263: them may appear as one of the most natural questions of

264: braid combinatorics. For degree¨$1$, \ie, for simple

265: braids, the linear ordering corresponds to a

266: lexicographical ordering of the associated

267: permutations¨\cite{Dgb}. But this connection does not

268: extend to higher degrees, and almost nothing is known

269: about the restriction of the linear ordering to positive

270: braids of a given degree. In particular, no connection is

271: known between the above mentioned Garside normal form and

272: the alternative normal form constructed by S.\,Burckel

273: in¨\cite{Bus, But, Buu}, one that makes comparison with

274: respect to the linear ordering easy: to give an example,

275: the Garside normal form of¨$\DD3{2\dd}$

276: is¨$(\ss1\ss2\ss1)^{2\dd}$, while its Burckel normal form

277: is $(\ss2\ss1^2\ss2)^\dd\ss1^{2\dd}$.

278: 

279: Our aim in this paper is to investigate the finite

280: linearly ordered sets $\DDDD\nn\dd$. A

281: nice way of thinking of this structure is to consider the

282: increasing enumeration of¨$\DDD\nn\dd$, and to view it as

283: a distinguished path from¨$1$ to¨$\DD\nn\dd$ in the Cayley

284: graph of¨$B_\nn$. A complete description of this path

285: would arguably be an optimal solution to the rather vague

286: question of connecting the Garside and the ordered

287: structures of braid groups. Such a description seems to

288: be extremely intricate from a combinatorial point of

289: view, and it remains out of reach for the moment, but we

290: prove partial results in this direction, namely 

291: 

292: - $(i)$ in the general case, a determination of some

293: numerical parameters attached with $\DDDD\nn\dd$ that in

294: some sense measure its size, with explicit values for

295: small values of¨$\nn$ and¨$\dd$, and

296: 

297: - $(ii)$ in the special case $\nn = 3$, a complete

298: description of the increasing enumeration

299: of¨$\DDDD\nn\dd$.

300: 

301: More specifically, the parameters we investigate are

302: the complexity and the heights.

303: The complexity¨$\ccc\nn\dd$ is defined as the

304: maximal number of occurrences of¨$\ss{\nn-1}$ in an

305: expresion of¨$\DD\nn\dd$ containing no¨$\ss{\nn-1}$.

306: It is connected with the termination of the handle

307: reduction algorithm of¨\cite{Dfo}, and its

308: determination was left as an open question in the

309: latter paper. The

310: $\rr$-height¨$\hhh\rr\nn\dd$ is defined to

311: be the number of¨$\rr$-jumps in the increasing

312: enumeration of $\DDDD\nn\dd$ (augmented by¨$1$), where

313: the term $\rr$-jump refers to some natural filtration of

314: the linear ordering¨$\sm$ by a sequence of partial

315: orderings¨$\sm_\rr$. When $\rr$ increases, $\rr$-jumps

316: are higher and higher, so $\hhh\rr\nn\dd$ counts how

317: many big jumps exist in $\DDDD\nn\dd$.  We prove that the

318: complexity $\ccc\nn\dd$ equals the height

319: $\hhh{\nn-1}\nn\dd$ (Proposition¨\ref{P:MainHeight}), and

320: that, for each¨$\rr$, the $\rr$-height $\hhh\rr\nn\dd$ is

321: the number of divisors of¨$\DD\nn\dd$ whose $\dd$th

322: factor of the normal form is right divisible by¨$\D_\rr$

323: (Proposition¨\ref{P:Main}). Together with the

324: combinatorial results of¨\cite{Dhi}, this allows for

325: computing the explicit values listed in

326: Table¨\ref{T:Values}, and for establishing

327: various inductive formulas (Propositions¨\ref{P:Values}

328: and¨\ref{P:Values34}, among others).

329: 

330: Besides the enumerative results, we also prove a

331: general structural result that connects the ordered set

332: $\DDDD\nn\dd$ with (subsets of) $\DDDD{\nn-1}\dd$

333: (Corollary¨\ref{C:Structure}). This result

334: suggests an inductive method for directly constructing

335: the increasing enumeration of $\DDDD\nn\dd$ starting

336: from those of $\DDDD{\nn-1}\dd$ and $\DDDD\nn{\dd-1}$.

337: This approach is completed here for $\nn = 3$

338: (Proposition¨\ref{P:Enum3}). In some sense,

339: $3$~strand braids are simple objects, and the result

340: may appear as of modest interest; however, the order

341: on~$B_3^+$ is a well-ordering of ordinal

342: type~$\omega^\omega$, hence not a so simple object. The

343: interesting point is that this approach leads to a

344: new, alternative construction of the braid ordering,

345: with in particular a new and simple proof for the

346: so-called Comparison Property which is the hard core

347: in the construction, namely the part that guarantees

348: the linearity of the ordering. In this way, one

349: obtains not only one more construction of an ordering

350: that already has many constructions

351: \cite{Dgr}, but arguably the optimal one, as it makes

352: all proofs simple once the initial inductive

353: definition is correctly stated, and as the connection

354: with the Garside structure is then explicit.

355: 

356: The paper is organized as follows. After a first

357: introductory section recalling basic properties and setting

358: the notation, we introduce the parameters $\ccc\nn\dd$

359: and $\hhh\rr\nn\dd$ in Section¨\ref{S:Meas}, and we

360: establish their connection. In Section¨\ref{S:Deter}, we

361: connect in turn $\hhh\rr\nn\dd$ with the number of

362: $\nn$-braids whose $\dd$th factor in the normal form

363: satisfy certain constraints, and deduce explicit

364: values. Finally, in Section¨\ref{S:N3}, we study the

365: specific case of¨$\DDDD3\dd$ and describe its

366: increasing enumeration, resulting in the new

367: construction of the braid ordering in this case.

368: 

369: \section{Background and preliminary results}

370: 

371: Our notation is standard, and we refer to textbooks

372: like¨\cite{Bir} or¨\cite{Eps} for basic results about

373: braid groups. We recall that the $\nn$¨strand braid group¨$B_\nn$

374: is defined for $\nn \ge 1$ by the presentation

375: \begin{equation} \label{E:Present}

376: B_\nn = \left<\ss1, \dots, \ss{\nn-1} \,;\,

377: \begin{array}{cl}

378: \ss i \ss j = \ss j \ss i 

379: &\text{\quad for $\vert

380: i - j\vert \ge 2$}\\

381: \   \ss i \ss j \ss i  = \ss j \ss i \ss j

382: &\text{\quad for $\vert i - j\vert = 1$}

383: \end{array} \right>.

384: \end{equation}

385: So, $B_1$ is the trivial one-element group, while

386: $B_2$ is the free group generated by¨$\ss1$. The

387: elements of¨$B_\nn$ are called $\nn$¨strand braids,

388: or simply {\it $\nn$-braids}. We use $B_\infty$

389: for the group generated by an infinite sequence

390: of¨$\ss i$'s subject to the relations

391: of¨\eqref{E:Present}, \ie, the direct limit of

392: all¨$B_\nn$'s with respect to the inclusion of¨$B_\nn$

393: into¨$B_{n+1}$.

394: 

395: By definition, every $\nn$-braid¨$\bx$ admits (infinitely

396: many) expressions in terms of the generators¨$\ss i$, $1

397: \le i < n$. Such an expression is called an

398: $\nn$¨strand {\it braid word}. Two braid words¨$\ww,

399: \ww'$ representing the same braid are said to be {\it

400: equivalent}; the braid represented by a braid

401: word¨$\ww$ is denoted¨$\cl\ww$. 

402: 

403: \subsection{Positive braids and the element $\D_\nn$}

404: 

405: We denote by¨$B_\nn^+$ the monoid

406: admitting the presentation¨\eqref{E:Present},

407: and by¨$B_\infty^+$ the union (direct limit) of

408: all¨$B_\nn^+$'s. The elements of¨$B_\nn^+$ are called {\it

409: positive} $\nn$-braids. In¨$B_\infty^+$, no element

410: except¨$1$ is invertible, and we have a natural notion of

411: divisibility: 

412: 

413: \begin{defi}

414: For $\bx, \by$ in¨$B_\nn^+$, we say that $\bx$ is a {\it

415: left divisor} of¨$\by$, denoted $\bx \dive \by$, or,

416: equivalently, that $\by$ is a {\it right multiple}

417: of¨$\bx$, if $\by = \bx \pz$ holds for some¨$\pz$

418: in¨$B_\nn^+$. We denote by¨$\Div(\by)$ the (finite) set of

419: all left divisors of¨$\by$ in¨$B_\nn^+$.

420: \end{defi}

421: 

422: The monoid~$B_\nn^+$ is not commutative for $\nn \ge

423: 3$, and therefore there are distinct symmetric

424: notions of a right divisor and a left multiple---but

425: we shall mostly use left divisors here. Note that

426: $\bx$ is a (left) divisor of¨$\by$ in the sense

427: of¨$B_\nn^+$ if and only if it is a (left) divisor in

428: the sense of¨$B_\infty^+$, so there is no need to

429: specify the index¨$\nn$.

430: 

431: According to Garside's theory \cite{Gar},

432: $B_\nn^+$ equipped with the left divisibility

433: relation is a lattice: any two

434: positive $\nn$-braids¨$\bx,

435: \by$ admit a greatest common left divisor, denoted

436: $\gcd(\bx, \by)$, and a least common right multiple,

437: denoted¨$\lcm(\bx, \by)$. A special role is played by the

438: lcm¨$\D_\nn$ of¨$\ss1$, \dots, $\ss{\nn-1}$, which can

439: be inductively defined by

440: \begin{equation} \label{E:Delta}

441: \D_1 = 1, \qquad

442: \D_\nn = \ss1 \ss2 \dots \ss{n-1} \, \D_{\nn-1}.

443: \end{equation}

444: It is well known that $\D_\nn^2$ belongs to the centre

445: of¨$B_\nn$ (and even generates it for $\nn \ge 3$),

446: and that the flip automorphism¨$\flip n$ of¨$B_\nn$

447: corresponding to conjugation by¨$\D_\nn$

448: exchanges¨$\ss i$ and¨$\ss{n-i}$ for

449: $1 \le i \le \nn-1$. It follows that $\flip n$

450: also exchanges¨$\ss{n, 1}$ and¨$\ss{1, n}$.

451: 

452: In¨$B_\nn^+$, the left and the right divisors of¨$\D_\nn$

453: coincide, and they make a finite sublattice of¨$(B_\nn^+,

454: \dive)$ with $\nn!$¨elements. These braids will be called

455: {\it simple} in the sequel. When braid

456: words are represented by diagrams as

457: mentioned in Figure~\ref{F:Diagram}, simple braids

458: are those positive braids that can be represented by a

459: diagram in which any two strands cross at most

460: once.

461: 

462: \begin{figure}[htb]

463: \begin{center}

464: \begin{picture}(83,21)(0,0)

465: \put(10,0){\includegraphics{Sigma.eps}}

466: \put(0,13){$\ss i$~:}

467: \put(0,3){$\ss i\inv$~:}

468: \put(9.5,19){$1$}

469: \put(19.5,19){$2$}

470: \put(50,19){$i$}

471: \put(58,19){$i{+}1$}

472: \put(27,3){\dots}

473: \put(77,3){\dots}

474: \put(27,13){\dots}

475: \put(77,13){\dots}

476: \end{picture}

477: \end{center}

478: \caption{\smaller One associates with

479: every $\nn$¨strand braid word¨$\ww$ an $\nn$¨strand

480: braid diagram by stacking elementary diagrams as

481: above; then two braid words are equivalent if and only

482: if the associated diagrams are the projections of

483: ambient isotopic figures in¨$\RR^3$, \ie, one can

484: deform one diagram into the other without allowing the

485: strands to cross or moving the endpoints.}

486: \label{F:Diagram}

487: \end{figure}

488: 

489: By mapping¨$\ss i$ to the transposition¨$(i, i+1)$, one

490: defines a surjective homomorphism¨$\pi$ of¨$B_\nn$ onto the

491: symmetric group¨$\Sym\nn$. The restriction of¨$\pi$ to

492: simple braids is a bijection: for every permutation¨$f$

493: of¨$\{1, \dots, \nn\}$, there exists exactly one simple

494: braid¨$\bx$ satisfying $\pi(\bx) = f$. It follows that

495: the number of simple $\nn$-braids is¨$\nn!$. 

496: 

497: \begin{exam} \label{X:Degree}

498: The set¨$\DDD3{}$ consists of six elements, namely $1$,

499: $\ss1$, $\ss2$, $\ss2\ss1$, $\ss1\ss2$, and¨$\D_3$. In

500: examples, we shall often use the shorter notation¨$\tt a$

501: for¨$\ss1$, $\tt b$ for¨$\ss2$, \etc\ Thus, the six

502: simple $3$-braids are $1$, $\tt a$, $\tt b$, $\tt ba$,

503: $\tt ab$, $\tt aba$.  

504: \end{exam}

505: 

506: \subsection{The normal form}

507: 

508: For each positive $\nn$-braid¨$\bx$ distinct of¨$1$,

509: the simple braid $\gcd(\bx, \D_\nn)$ is the maximal

510: simple left divisor of¨$\bx$, and we obtain a

511: distinguished expression $\bx = \sx_1 \bx'$

512: with¨$\sx_1$ simple. By decomposing¨$\bx'$ in the

513: same way and iterating,  we obtains the so-called

514: normal expression¨\cite{ElM, Eps}. 

515: 

516: \begin{defi}

517: A sequence $(\sx_1, \dots, \sx_\dd)$ of simple

518: $\nn$-braids is said to be {\it normal} if, for

519: each¨$\kk$, one has $\sx_\kk =

520: \gcd(\D_\nn, \sx_\kk \dots \sx_\dd)$.

521: \end{defi}

522: 

523: Clearly, each positive braid admits a unique normal

524: expression. It will be convenient here to consider the

525: normal expression as unbounded on the right by

526: completing it with as many trivial factors¨$1$ as needed.

527: In this way, we can speak of the {\it $\dd$th factor} (in

528: the normal form) of¨$\bx$ for each positive braid¨$\bx$.

529: We say that a positive braid has {\it degree¨$\dd$} if

530: $\dd$ is the largest integer such that the $\dd$th factor

531: of¨$\bx$ is not¨$1$. We shall use the following two

532: properties of the normal form: 

533: 

534: \begin{lemm} \cite{ElM} \label{L:Normal}

535: A sequence of simple $\nn$-braids $(\sx_1, \dots,

536: \sx_\dd)$ is normal if and only if, for each¨$\kk <

537: \dd$, each¨$\ss i$ that divides¨$\sx_{\kk+1}$

538: on the left divides¨$\sx_\kk$ on the right.

539: \end{lemm}

540: 

541: \begin{lemm} \cite{ElM} \label{L:Degree}

542: For $\bx$ a positive braid in¨$B_\nn^+$, the following

543: are equivalent:

544: 

545: $(i)$ The braid¨$\bx$ belongs to¨$\DDD\nn\dd$, \ie, is a

546: (left or right) divisor of¨$\DD\nn\dd$; 

547: 

548: $(ii)$ The degree of¨$\bx$ is at most¨$\dd$.

549: \end{lemm}

550: 

551: \begin{exam} \label{X:Enum}

552: There are 19 divisors of~$\DD32$, which also are the

553: $3$-braids of degree at most~$2$. Their enumeration in

554: normal form---in an ordering that may seem strange now,

555: but should become familiar soon---is: $1$,

556: $\a$, $\a\opp\a$, $\b$, $\b\a$,

557: $\b\a\opp\a$, $\b\opp\b$, $\b\opp\b\a$, $\a\b$,

558: $\a\b\a$, $\a\b\a\opp\a$, $\a\b\opp\b$,

559: $\a\b\opp\b\a$, $\a\opp\a\b$, $\a\b\a\opp\b$,

560: $\a\b\a\opp\b\a$, $\b\a\opp\a\b$, $\a\b\a\opp\a\b$,

561: $\a\b\a\opp\a\b\a$. 

562: \end{exam}

563: 

564: By Lemma~\ref{L:Degree}, every divisor

565: of¨$\DD\nn\dd$ can be expressed as the product of at

566: most¨$\dd$ divisors of¨$\D_\nn$, so we certainly have

567: $\card\DDD\nn\dd \le (\nn!)^\dd$ for all¨$\nn,

568: \dd$.

569: 

570: \subsection{The braid ordering}

571: 

572: The basic notion is the following one:

573: 

574: \begin{defi}

575: Let $\ww$ be a nonempty braid word. We say that

576: $\ss\mm$ is the {\it main} generator in¨$\ww$

577: if $\ss\mm$ or $\ss\mm\inv$ occurs in~$\ww$,

578: but no $\ss i^{\pm1}$ with $i > \mm$ does. We

579: say that $\ww$ is {\it

580: $\s$-positive} (\resp {\it $\s$-negative}) if the main

581: generator occurs only positively (\resp negatively)

582: in¨$\ww$.

583: \end{defi}

584: 

585: A positive nonempty braid word, \ie, one that contains

586: no¨$\ss i\inv$ at all, is $\s$-positive, but the inclusion

587: is strict: for instance, $\ss1\inv \ss2$ is not positive,

588: but it is $\s$-positive, as its main generator,

589: namely¨$\ss2$, occurs positively (one¨$\ss2$) but not

590: negatively (no¨$\ss2\inv$).

591: 

592: Then we have the following two properties, which have

593: received a number of independent proofs¨\cite{Dgr}:

594: 

595: \begin{propA}

596: {\it A $\s$-positive braid word does not represent¨$1$.}

597: \end{propA}

598: 

599: \begin{propC}

600: {\it Every braid except¨$1$ can be represented by a

601: $\s$-positive word or by a $\s$-negative word.}

602: \end{propC}

603: 

604: It is then straightforward to order the braids:

605: 

606: \begin{defi}

607: If $\bx, \by$ are braids, we say that $\bx \sm \by$ holds

608: if the braid¨$\bx\inv\by$ admits at least one $\s$-positive

609: representative.

610: \end{defi} 

611: 

612: It is clear that the relation¨$\sm$ is transitive and

613: compatible with multiplication on the left;  Property¨A

614: implies that $\sm$ has no cycle, hence is a

615: strict partial order, and Property¨C then implies that it is

616: actually a linear order.

617: 

618: As every nonempty positive braid word is

619: $\s$-positive, $\bx \dive \by$ implies $\bx \sme \by$

620: for all positive braids¨$\bx, \by$, but the converse is

621: not true: $\ss1$ is not a left divisor of¨$\ss2$, but $\ss1

622: \sm \ss2$ holds, since $\ss1\inv\ss2$ is a $\s$-positive

623: word. 

624: 

625: \begin{exam} \label{X:List}

626: The increasing enumeration of the set¨$\DDD3{}$ is

627: $$\tt 1 \sm a \sm b \sm ba \sm ab \sm aba.$$

628: For

629: instance, we have $\tt ba \sm ab$, \ie, $\ss2\ss1

630: \sm \ss1\ss2$, as the quotient, namely $\ss1\inv

631: \ss2\inv \ss1 \ss2$ (or $\tt ABab$), also admits the

632: expression $\ss2 \ss1\inv$, a $\s$-positive word.

633: Similarly, the reader can check that the increasing

634: enumeration of¨$\DDD32$ is the one given in

635: Example~\ref{X:Enum}.

636: \end{exam}

637: 

638: \begin{lemm} \label{L:Extend}

639: The linear ordering¨$\sm$ extends the left divisibility

640: ordering¨$\div$.

641: \end{lemm}

642: 

643: \begin{proof}

644: By definition, $1 \sm \ss i$ holds for every¨$i$. As the

645: ordering¨$\sm$ is compatible with multiplication on the

646: left, it follows that $\xx \sm \xx \ss i$ holds for

647: all¨$i, \xx$, and, therefore, $\xx \sm \xx \yy$ holds

648: whenever $\yy$ is a non-trivial positive braid.

649: \end{proof}

650: 

651: Lemma¨\ref{L:Extend} implies that $1$ is always the

652: first element of¨$\DDDD\nn\dd$, and

653: $\DD\nn\dd$ is always its last element.

654: It may be noted that a deep result by

655: Laver¨\cite{Lve} shows that, although $\sm$ is not

656: compatible with right multiplication in general,

657: nevertheless $\xx \sm \ss i

658: \xx$ always holds,

659: \ie, $\sm$ also extends the right divisibility ordering. 

660: 

661: By Property~C, every nontrivial braid admits

662: at least one $\s$-positive or $\s$-negative

663: expression. In general, such a $\s$-positive or

664: $\s$-negative expression is not unique, but the main

665: generator in such expressions is uniquely defined:

666: 

667: \begin{lemm} \label{L:Index}

668: If a braid¨$\bx$ admits a $\s$-positive 

669: expression, then the main generators in any

670: two $\s$-positive expressions of¨$\bx$

671: coincide.

672: \end{lemm}

673: 

674: \begin{proof}

675: Assume that $\ww$, $\ww'$ are

676: $\s$-positive  expressions of¨$\bx$, and let

677: $\ss\mm$, $\ss{\mm'}$ be their main

678: generators. Assume for instance $\mm <

679: \mm'$. Then $\ww\inv \ww'$ is a

680: $\s$-positive word, and it represents the

681: trivial braid¨$1$: this contradicts Property¨A.

682: \end{proof}

683: 

684: So, there will be no ambiguity in referring to

685: {\it the} main generator of some non-trivial braid¨$\bx$: 

686: this means the main generator in any $\s$-positive (or

687: $\s$-negative) expression of¨$\bx$.

688: 

689: \begin{rema}

690: Our current definition corresponds to the order¨$<^\phi$

691: of¨\cite{Dgr}. It differs from the one most usually

692: considered in literature in that we refer to the maximal

693: index rather than to the minimal one in the

694: definition of a $\s$-positive word. Switching from one

695: definition to the other amounts to conjugating

696: by¨$\D_\nn$, \ie, to applying the flip automorphism. Results

697: are entirely similar for both versions. However, it is much

698: more convenient to consider the ``max'' choice here, because

699: it guarantees that $B_\nn^+$ is an initial segment

700: of¨$B_{n+1}^+$. Using the ``min'' convention would make the

701: statements of the forthcoming sections less natural.

702: \end{rema}

703: 

704: \section{Measuring the ordered sets¨$\DDDD\nn\dd$}

705: \label{S:Meas}

706: 

707: As was said above, our aim is

708: to investigate the finite ordered sets¨$\DDDD\nn\dd$,

709: and, more generally, $(\Div(\pz), \sm)$ for¨$\pz$ a

710: positive braid. We shall do it by defining numerical

711: parameters that somehow measure their size. The first

712: parameter involves the length of the

713: $\s$-positive words that are, in some natural sense

714: defined below, drawn in the Cayley graph

715: of¨$\DD\nn\dd$. It will be called the {\it complexity}

716: of¨$\DD\nn\dd$, because it is directly connected with

717: the complexity analysis of the handle reduction

718: algorithm of¨\cite{Dfo}. The other parameters involve

719: a filtration of the linear ordering by

720: the¨$\ss i$'s, and they will be called the {\it

721: heights} of¨$\DD\nn\dd$ because they count the jumps

722: of a given height in¨$\DDDD\nn\dd$. 

723: 

724: \subsection{Sigma-positive paths in the Cayley graph}

725: 

726: The first parameter we attach to $(\Div(\pz),

727: \sm)$ involves the $\s$-positive paths in the Cayley

728: graph of¨$\pz$.

729: 

730: We recall that the Cayley graph of

731: the group¨$B_\nn$ with respect to the standard

732: generators $\ss i$ is the labeled graph  with vertex

733: set¨$B_\nn$ and such that there exists a $\ss

734: i$-labeled edge from¨$\bx$ to¨$\by$ if and only if

735: $\by = \bx \ss i$ holds. The Cayley graph of the

736: monoid¨$B_\nn^+$ is obtained by restricting the vertices

737: to¨$B_\nn^+$. Note that the Cayley graph of¨$B_\nn$

738: (and a fortiori of¨$B_\nn^+$) can be seen as a subgraph

739: of the Cayley graph of¨$B_\infty$.

740: 

741: \begin{defi}

742: (Figure¨\ref{F:Cayley})

743: For $\pz$ a positive braid, we denoted by¨$\CG(\pz)$

744: the subgraph of the Cayley graph of¨$B_\infty$

745: obtained by restricting the vertices to¨$\Div(\pz)$,

746: and only keeping those edges that connect two vertices

747: in¨$\Div(\pz)$.

748: \end{defi}

749: 

750: As every element of¨$B_\nn^+$ is a left divisor of¨$\DD\nn\dd$

751: for¨$\dd$ large enough, the Cayley graph of¨$B_\nn^+$ is 

752: the union of all graphs¨$\GG n k$ when $\dd$¨varies.

753: 

754: \begin{figure} [htb]

755: \begin{picture}(110,51)(0, -3)

756: \put(5,8){\includegraphics{Path31.eps}}

757: \put(50,0){\includegraphics{Path32.eps}}

758: \put(2,20){$1$}

759: \put(10,34){$\ss2$}

760: \put(27,34){$\ss2\ss1$}

761: \put(10,7){$\ss1$}

762: \put(27,7){$\ss1\ss2$}

763: \put(36,20){$\D_3$}

764: \put(47,20){$1$}

765: \put(55,33){$\ss2$}

766: \put(60,44){$\ss2^2$}

767: \put(70,44){$\ss2^2\ss1$}

768: \put(80,44){$\ss2\ss1^2$}

769: \put(90,44){$\ss2\ss1^2\ss2$}

770: \put(65,33){$\ss2\ss1$}

771: \put(65,33){$\ss2\ss1$}

772: \put(86,33){$\D_3\ss1$}

773: \put(99,33){$\D_3\ss1\ss2$}

774: \put(55,8){$\ss1$}

775: \put(65,8){$\ss1\ss2$}

776: \put(86,8){$\D_3\ss2$}

777: \put(99,8){$\D_3\ss2\ss1$}

778: \put(82,20){$\D_3$}

779: \put(108,20){$\D_3^2$}

780: \put(60,-3){$\ss1^2$}

781: \put(70,-3){$\ss1^2\ss2$}

782: \put(80,-3){$\ss1\ss2^2$}

783: \put(90,-3){$\ss1\ss2^2\ss1$}

784: \end{picture}

785: \caption{\smaller The graphs

786: of¨$\GG3{}$ and $\GG32$;

787: the dotted edges represent¨$\ss1$, the plain

788: ones¨$\ss2$; observe that the graph

789: of¨$\DD32$ is not planar; in grey: two

790: $\s$-positive words traced in the graphs, namely

791: $\tt aAbab$ and $\tt bbabAbab$ ({\it cf.}

792: Lemma¨\ref{L:Path})}

793: \label{F:Cayley}

794: \end{figure}

795: 

796: A path in the Cayley graph can be specified by its

797: initial vertex and the list of the labels of its

798: successive edges, \ie, by a braid word. For each¨$i <

799: n$ and each¨$\bx$ in¨$B_\nn$, there is exactly one $\ss

800: i$-labeled edge with target¨$\bx$, and one

801: $\ss i$-labeled edge with source¨$\bx$ in the Cayley

802: graph of¨$B_\nn$. Hence, in the complete

803: Cayley graph of¨$B_\nn$, for each initial vertex¨$\bx$ and

804: each $\nn$-braid word¨$\ww$, there is always one path

805: labeled¨$\ww$ starting from¨$\bx$. When we restrict to some

806: fragment¨$\G$, this need not be the case, but we have an

807: unambiguous notion of¨$\ww$ being drawn in¨$\G$

808: from¨$\bx$. Formally, this corresponds to

809: 

810: \begin{defi}

811: For $\G$ a subgraph of the Cayley graph

812: of¨$B_\infty$, and $\bx$ a vertex in¨$\G$, we say

813: that a braid word¨$\ww$ is {\it drawn} from¨$\bx$

814: in¨$\G$ if, for every prefix¨$\vv\ss i$ (\resp $\vv\ss

815: i\inv$) of¨$\ww$, there exists a $\ss i$-labeled edge

816: starting from (\resp finishing at)¨$\bx

817: \, \cl{\vv}$ in¨$\G$.

818: \end{defi}

819: 

820: For instance, we can check on Figure¨\ref{F:Cayley} that

821: the word¨$\ss1^2$ is drawn from¨$\ss2$

822: in¨$\GG32$, but not in¨$\GG3{}$. In algebraic terms,

823: we have the following characterization:

824: 

825: \begin{lemm} \label{L:Path}  

826: Assume that $\pz$ is a positive braid, and $\ww$ is a

827: braid word. Then $\ww$ is drawn from¨$\bx$ in¨$\CG(\pz)$

828: if and only if $\bx \cl{\vv} \dive \pz$ holds for each

829: prefix¨$\vv$ of¨$\ww$.

830: \end{lemm}               

831: 

832: \begin{proof}

833: The condition is sufficient. Indeed, assume it is

834: satisfied by¨$\ww$, and $\vv \ss i$ is a prefix of¨$\ww$.

835: Then, by hypothesis, $\bx \cl{\vv}$ and

836: $\bx \cl{\vv}\ss i$ are left divisors of¨$\pz$, hence

837: are vertices in¨$\CG(\pz)$, and, therefore,

838: there is a $\ss i$-labeled edge between¨$\bx \cl{\vv}$ and

839: $\bx \cl{\vv} \ss i$ in¨$\CG(\pz)$. The argument is

840: similar for a prefix of the form $\vv \ss i\inv$. Using

841: induction on the length of¨$\ww$, we deduce that $\ww$ is

842: drawn from¨$\bx$ in¨$\CG(\pz)$. 

843: 

844: Conversely, if there exists a $\ww$-labeled path

845: from¨$\bx$ in¨$\CG(\pz)$, then, for each prefix¨$\vv$

846: of¨$\ww$, the braid¨$\bx \cl{\vv}$ has to represent some

847: vertex in¨$\CG(\pz)$, hence it is a left divisor

848: of¨$\pz$. 

849: \end{proof}

850: 

851: For¨$\pz$ a positive braid, we shall investigate the

852: $\s$-positive words drawn in the graph¨$\CG(\pz)$. It is

853: clear that, even if $\Div(\pz)$ is a finite set,

854: arbitrary long words are drawn in¨$\CG(\pz)$ whenever the

855: latter contains at least 2¨vertices, \ie, $\pz$ is

856: not¨$1$. The example of Figure¨\ref{F:Cayley} shows that

857: restricting to $\s$-positive words does not change the

858: result: for instance, for each¨$\kk$, the

859: word $(\ss1\ss1\inv)^\kk\ss2\ss1\ss2$ is a

860: $\s$-positive expression of¨$\D_3$, and it is drawn

861: in¨$\GG3{}$. So we cannot hope for any finite upper

862: bound on the length of the $\s$-positive words drawn

863: in¨$\CG(\pz)$ in general. Now, the situation changes

864: if we concentrate on the main generators, \ie, we

865: forget about the generators with non-maximal index.

866: 

867: \begin{lemm} \label{L:Finite}

868: Assume that $\G$ is subgraph of the Cayley graph

869: of¨$B_\infty$, and $\ww$ is a $\s$-positive word drawn

870: in¨$\CG(\pz)$. Then the number of occurrences of the main

871: generator in¨$\ww$ is at most the number of

872: non-terminal vertices in¨$\G$.

873: \end{lemm}

874: 

875: \begin{proof}

876: Assume that $\ww$ is drawn from¨$\bx$ in¨$\G$. Let

877: $\ss\mm$ be the main generator in¨$\ww$. As there is at

878: most one $\ss\mm$-labeled edge starting from each vertex

879: of¨$\G$, it suffices to show that the number of¨$\ss\mm$'s

880: in¨$\ww$ is bounded above by the number of

881: $\ss\mm$-edges in¨$\G$. Hence, it suffices to show that

882: the path¨$\gamma$ associated with¨$\ww$ cannot cross the

883: same $\ss\mm$-edge twice. Now assume that some

884: $\ss\mm$-edge starts from the vertex¨$\by$, and that $\g$

885: crosses this edge twice. This means that $\g$ contains a

886: loop from¨$\by$ to¨$\by$. Let

887: $\vv$ be the subword of¨$\ww$ labeling that loop. By

888: construction, $\vv$ begins with¨$\ss\mm$, it contains

889: no¨$\ss\mm\inv$ and no $\ss i^{\pm1}$ with $i > m$ as

890: it is a subword of¨$\ww$, and it represents the

891: braid¨$1$ as it labels a loop in the Cayley graph

892: of¨$B_\infty$: this means that $\vv$ is a

893: $\s$-positive word representing¨$1$, which contradicts

894: Property¨A.

895: \end{proof}

896: 

897: Lemma¨\ref{L:Finite} applies in particular to every

898: graph¨$\CG(\pz)$ with¨$\pz$ a positive braid. So we can

899: introduce our first parameter measuring

900: the size of the ordered set¨$\DDiv\pz$:

901: 

902: \begin{defi}

903: (Figure~\ref{F:Cayley})

904: For¨$\pz$ a positive braid with main generator¨$\ss\mm$,

905: the {\it complexity}¨$\cc(\pz)$ of¨$\pz$ is defined to

906: be the maximal number of¨$\ss\mm$'s in a $\s$-positive

907: word drawn in¨$\CG(\pz)$. 

908: \end{defi}

909: 

910: \begin{exam}

911: The word¨$\ss2 \ss1 \ss2$ is a

912: $\s$-positive word drawn from¨$1$ in¨$\GG3{}$, and

913: it contains two¨$\ss2$'s, hence we have $\ccc3{}

914: \ge 2$. Actually, it is not hard to obtain the

915: exact value $\ccc3{} = 2$. Indeed, if a

916: $\s$-positive path¨$\g$ contains the two

917: $\ss2$-edges starting from¨$1$ and¨$\ss1\ss2$, it

918: cannot come back to¨$\ss2$ for possibly crossing the

919: third $\ss2$-edge; and if $\g$ contains the

920: $\ss2$-edge that starts from¨$\ss1$, it can never

921: come back to¨$1$ or to¨$\ss2\ss1$ and therefore

922: contains at most one $\ss2$-edge. 

923: As we have $\DD3\dd = (\ss2 \ss1 \ss2)^\dd$,

924: we deduce $\ccc3\dd \ge 2\dd$ for every¨$\dd$;

925: this value is certainly not optimal, since the

926: example displayed in Figure¨\ref{F:Cayley} contains

927: five¨$\ss2$'s, proving $\ccc32 \ge 5$---the exact

928: value is¨$6$, and, more generally, we have $\ccc\nn\dd

929: = 2^{\dd+1} - 2$, as will be seen in

930: Section¨\ref{S:Deter}.

931: \end{exam}

932: 

933: \begin{rema}

934: Restricting to $\s$-positive words drawn in¨$\CG(\pz)$

935: is essential: for instance, for each¨$\kk$, we have

936: \begin{equation} \label{E:Unbounded}

937: \D_3 = \ss2^{\kk+1} \ss1\ss2\ss1^{-\kk},

938: \end{equation}

939: a $\s$-positive word containing $\kk + 2$ letters¨$\ss2$.

940: Now, for¨$\kk \ge 1$, the word involved

941: in¨\eqref{E:Unbounded} is not drawn in¨$\GG3{1}$, as

942: its prefix¨$\ss2^2$ is not. Thus the

943: parameter¨$\cc(\pz)$ really involves the left

944: divisors of¨$\pz$.

945: \end{rema}

946: 

947: A direct application of Lemma¨\ref{L:Finite} gives:

948: 

949: \begin{prop}

950: Every positive braid has a finite complexity; more

951: precisely, for¨$\pz$ of length¨$\ell$ in¨$B_\nn^+$  with

952: $\nn \ge 3$, we have $\cc(\pz) \le (\nn-1)^\ell$.

953: \end{prop}

954: 

955: \begin{proof}

956: The number of non-terminal vertices in¨$\CG(\pz)$, \ie,

957: the number of proper left divisors of¨$\pz$, is at most

958: $1 + (\nn-1) + (\nn-1)^2 + \cdots + (\nn-1)^{\ell-1}$.

959: \end{proof}

960: 

961: As the length of any positive expression of¨$\D_\nn$ is

962: $\nn(\nn-1)/2$, we obtain in particular for all¨$\nn, \dd$

963: \begin{equation} \label{E:Coarse}

964: \ccc\nn\dd \le

965: (\nn-1)^{\dd\nn(\nn-1)/2}.

966: \end{equation}

967: 

968: Before going further, let us observe that, in the

969: definition of the complexity of¨$\pz$, we can restrict

970: to decompositions of¨$\pz$, \ie, instead of considering

971: paths starting from and finishing at arbitrary vertices,

972: we can restrict to paths starting from¨$1$ and finish

973: at¨$\pz$:

974: 

975: \begin{lemm} \label{L:Decomp}

976: Assume that $\pz$ is a positive braid with main

977: generator¨$\ss\mm$. Then $\cc(\pz)$ is the

978: maximal number of¨$\ss\mm$'s in any $\s$-positive

979: decomposition of¨$\pz$ drawn in¨$\CG(\pz)$.

980: \end{lemm}

981: 

982: \begin{proof}

983: Let $\cc'(\pz)$ be the number involved in the above

984: statement. Clearly we have $\cc'(\pz) \le \cc(\pz)$.

985: Conversely, assume that $\ww$ is drawn in¨$\CG(\pz)$

986: from¨$\bx$, and that the $\ww$-labeled path starting

987: from¨$\bx$ finishes at¨$\by$. Let $\uu$ be a positive

988: expression of¨$\bx$, and $\vv$ be a positive expression

989: of¨$\by\inv \pz$. The latter exists as, by hypothesis,

990: $\by$ is a left divisor of¨$\pz$. Then $\uu\ww\vv$ is a

991: $\s$-positive decomposition of¨$\pz$ drawn in¨$\CG(\pz)$.

992: Hence we have $\cc'(\pz) \ge \cc(\pz)$.

993: \end{proof}

994: 

995: 

996: \begin{rema}

997: Let us call Property~A$^*$ the fact that all

998: numbers¨$\ccc\nn\dd$ are finite. Above we derived 

999: Property~A$^*$ from Property¨A. Actually, the

1000: implication is an equivalence, \ie, we can also

1001: deduce Property¨A from Property~A$^*$. Indeed,

1002: assume that some $\s$-positive braid word¨$\ww$

1003: represents¨$1$. The word¨$\ww$ may involve negative

1004: letters and the problem is to find a vertex¨$\bx$

1005: such that there exists a path labeled¨$\ww$

1006: from¨$\bx$ in some¨$\GG\nn\dd$. Let $\ss m$ be the

1007: main generator in¨$\ww$. The

1008: word¨$\ww$ has finitely many prefixes, say

1009: $\ww_0$, \dots, $\ww_\ell$. By Garside's theory, each

1010: word¨$\ww_i$ is equivalent to a word of the form¨$u_i\inv

1011: \vv_i$ with $\uu_i, \vv_i$ positive. Let $\bx$

1012: be the least common left multiple of the

1013: positive braids¨$\cl{\uu_0}$,

1014: \dots, $\cl{\uu_\ell}$. Then, for each¨$i$, the

1015: braid¨$\bx\cl{\ww_i}$ is positive. Moreover, there

1016: exist¨$\nn$ and¨$\dd$ such that $\bx\cl{\ww_0}$, \dots,

1017: $\bx\cl{\ww_\ell}$ all are divisors of¨$\DD\nn\dd$. This

1018: means that the word¨$\ww$ is drawn from¨$\bx$ 

1019: in¨$\GG\nn\dd$, and the associated path is a loop

1020: around¨$\bx$. It follows that $\ww^\kk$ is drawn

1021: in¨$\GG\nn\dd$ from¨$\bx$ for each¨$\kk$. By

1022: construction, $\ww^\kk$ contains at least

1023: $\kk$¨generators¨$\ss m$, hence $\ccc\nn\dd$ cannot

1024: be finite.

1025: \end{rema}

1026: 

1027: \subsection{Connection with handle reduction}

1028: 

1029: Handle reduction¨\cite{Dfo} is an algorithmic solution to

1030: the word problem of braids that relies on the braid

1031: ordering---actually the most efficient method

1032: available to-date in practice. It is proved to be

1033: convergent, but the complexity upper bound resulting

1034: from the argument of¨\cite{Dfo} is exponential with

1035: respect to the length of the input word, seemingly

1036: very far from sharp. 

1037: 

1038: Each step of handle reduction involves a specific

1039: generator¨$\ss i$, and, for an induction, the point is to

1040: obtain an upper bound on the number of reduction steps

1041: involving the main generator. The latter will naturally

1042: be called the {\it main} reduction steps. The

1043: connection between handle reduction and the complexity

1044: as defined above relies on the following

1045: technical result: 

1046: 

1047: \begin{lemm} \cite{Dfo}

1048: Assume that $\pz$ is a positive braid with main

1049: generator¨$\ss\mm$, and $\ww$ is drawn in¨$\CG(\pz)$.

1050: Then, for each sequence of handle reductions from¨$\ww$,

1051: \ie, each sequence¨$\vec\ww$ with $\ww_0= \ww$ such that

1052: $\ww_\kk$ is obtained by reducing one handle

1053: from¨$\ww_{\kk-1}$ for each¨$\kk$, there exists a

1054: witness-word¨$\uu$ that is  $\s$-positive, drawn

1055: in¨$\Div(\pz)$, and such that the number of¨$\ss\mm$'s

1056: in¨$\uu$ is the number of main reductions in¨$\vec\ww$.

1057: \end{lemm}

1058: 

1059: It follows that the number of main reduction steps in any

1060: sequence of handle reductions starting with a word drawn

1061: in¨$\CG(\pz)$ is bounded above by¨$\cc(\pz)$. In

1062: particular, if we start with an $n$¨strand braid

1063: word¨$\ww$ of length¨$\ell$, then it is easy to show

1064: that $\ww$ is drawn in¨$\GG\nn\ell$, and, applying the

1065: upper bound of¨\eqref{E:Coarse}, we deduce an

1066: upper bound for the number of possible main reductions

1067: from¨$\ww$, one that is exponential with respect

1068: to¨$\ell$.

1069: 

1070: A natural way of improving this coarse upper bound would

1071: be to determine the value of¨$\ccc\nn\dd$ more precisely.

1072: This will be done in Section¨\ref{S:Deter} below.

1073: However, the explicit formulas show that, for $\nn \ge

1074: 3$, the growth rate with respect to¨$\dd$ is exponential,

1075: thus discarding any hope of proving the expected

1076: polynomial upper bound for the number of reduction steps

1077: by this approach.

1078: 

1079: \subsection{A filtration of the braid ordering}

1080: 

1081: We now introduce new numerical parameters for the ordered

1082: sets¨$\DDiv\pz$. These numbers appear in connection with a

1083: natural filtration of the ordering¨$\sm$, using an

1084: increasing sequence of partial orderings that we

1085: introduce now.

1086: 

1087: By Lemma¨\ref{L:Index}, the index of the main generator

1088: of a non-trivial braid is well defined. We can use this

1089: index to  measure the height of the jump between two

1090: braids¨$\bx, \by$ satisfying $\bx \sm \by$:

1091: 

1092: \begin{defi}

1093: For $\bx, \by$ in~$B_\infty$ and $\rr

1094: \ge 1$, we say that $\bx \sm_\rr \by$ holds

1095: or, equivalently, that $(\bx, \by)$ is a

1096: {\it $\rr$-jump}, if $\bx\inv\by$ admits a

1097: $\s$-positive expression in which the main

1098: generator is~$\ss\mm$ with $\mm \ge \rr$.

1099: \end{defi}

1100: 

1101: \begin{lemm}

1102: For each $\rr \ge 1$, the

1103: relation~$\sm_\rr$ is a strict partial

1104: order that refines~$\sm$; the

1105: relation~$\sm_1$ coincides with~$\sm$,

1106: and $\rr \le \qq$ implies that $\sm_\qq$

1107: refines~$\sm_\rr$. 

1108: \end{lemm}

1109: 

1110: \begin{proof}

1111: That $\sm_\rr$ is transitive follows from

1112: the fact that the concatenation of a

1113: $\s$-positive word with main

1114: generator~$\ss{\mm}$ and a $\s$-positive word 

1115: with main generator~$\ss{\mm'}$ is a

1116: $\s$-positive word with main

1117: generator~$\ss{\max(\mm, \mm')}$. 

1118: \end{proof}

1119: 

1120: In the sequel, we consider the $\sm_\rr$-chains

1121: included in~$\Div(\pz)$, and their length:

1122: 

1123: \begin{defi}

1124: For $\pz$ a positive braid and $\rr \ge 1$,

1125: we define the {\it $\rr$-height}¨$\hh\rr(\pz)$ of¨$\pz$

1126: to be the maximal length of a $\sm_\rr$-chain included

1127: in~$\Div(\pz)$. 

1128: \end{defi}

1129: 

1130: Before giving examples, we observe the

1131: connection between¨$\hh\rr(\pz)$ and

1132: the increasing enumeration of the

1133: set~$\Div(\pz)$:

1134: 

1135: \begin{lemm} \label{L:Enum}

1136: Let $\pz$ be a positive braid and $\rr \ge

1137: 1$. Then $\hh\rr(\pz)-1$ is the

1138: number of $\rr$-jumps in the increasing

1139: enumeration of~$\DDiv\pz$.

1140: \end{lemm}

1141: 

1142: \begin{proof}

1143: If the number of $\rr$-jumps in the

1144: increasing enumeration of~$\Div(\pz)$

1145: is~$\NN_\rr-1$, we can extract from~$\Div(\pz)$ a

1146: $\sm_\rr$-chain of length~$\NN_\rr$. Conversely,

1147: assume that $(\by_0, \dots \by_{\NN_\rr})$ is a

1148: $\sm_\rr$-chain in~$\Div(\pz)$. 

1149: Let $\pz_0 \sm \ldots \sm \pz_\NNN$ be the

1150: increasing enumeration of~$\Div(\pz)$. As

1151: $\sm_\rr$ refines~$\sm$, there exists an

1152: increasing function~$f$ of $\{0, \ldots,

1153: \NN_\rr\}$ into $\{0, \ldots, \NNN\}$ such that

1154: $\by_i = \pz_{f(i)}$ holds for every~$i$. Now

1155: the hypothesis $\pz_{f(i)} \sm_\rr

1156: \pz_{f(i+1)}$ implies that there exists at

1157: least one $\rr$-jump between $\pz_{f(i)}$

1158: and~$\pz_{f(i+1)}$. Indeed, by

1159: Lemma~\ref{L:Index}, it is impossible that a

1160: concatenation of $\mm$-jumps with $\mm < \rr$

1161: results in a $\rr$-jump. So the number

1162: of $\rr$-jumps in $(\pz_0, \dots, \pz_\NNN)$ is

1163: at least~$\NN_\rr$.

1164: \end{proof}

1165: 

1166: In other words, in order to determine¨$\hh\rr(\pz)$,

1167: there is no need to consider arbitrary chains: it

1168: is enough to consider the maximal chain

1169: obtained by enumerating $\Div(\pz)$ exhaustively. 

1170: 

1171: \begin{exam} \label{X:Height}

1172: Refining the increasing enumeration of¨$\DDD3{}$

1173: given in Example¨\ref{X:List} by indicating for

1174: each step the height of the corresponding jump, we

1175: obtain:

1176: \begin{equation} \label{E:List31}

1177: \tt 1 \sm_1 a \sm_2 b

1178: \sm_1 ba

1179: \sm_2 ab \sm_1 \D_3,

1180: \end{equation}

1181: where we recall $\tt a, b, \dots$ stand for

1182: $\ss1, \ss2, \dots$. For instance,

1183: $\tt (ba, ab)$ is a $2$-jump, as we have $\tt

1184: (ba)\inv(ab) = ABab = AabA = bA$, a

1185: $\s$-positive decomposition with main

1186: generator~$\ss2$. The number of $1$-jumps

1187: in~\eqref{E:List31},

1188: \ie, the number of symbols~$\sm_\rr$ with

1189: $\rr \ge 1$, is~$5$, while

1190: the number of~$2$-jumps is~$2$, so, by

1191: Lemma¨\ref{L:Enum}, we deduce

1192: $\hhh13{} = 6$ and $\hhh23{} = 3$.

1193: Similarly, we obtain for¨$\DD32$ 

1194: \begin{multline*}

1195: \tt 1 \sm_1 a \sm_1 aa \sm_2 b \sm_1 ba 

1196: \sm_1 baa \sm_2 bb \sm_1 bba 

1197: \sm_2 ab \sm_1

1198: aba  \sm_1 abaa \sm_2 abb 

1199: \\ \tt 

1200: \sm_1 abba \sm_2

1201: aab

1202: \sm_1  aaba \sm_1 aabaa \sm_2 baab \sm_1 baaba

1203: \sm_1 baabaa,

1204: \end{multline*}

1205: leading to $\hhh132 = 19$ and $\hhh232 = 7$.

1206: \end{exam}

1207: 

1208: \begin{prop} \label{P:Decreasing}

1209: $(i)$ For every braid~$\pz$ in~$B_\nn^+$, we

1210: have

1211: \begin{equation}

1212: \hh1(\pz) = \card\Div(\pz)  \ge

1213: \hh2(\pz)

1214: \ge \dots \ge \hh\nn(\pz) = 1.

1215: \end{equation}

1216: 

1217: $(ii)$ For all positive braids~$\pz, \pz'$

1218: and $\rr \ge 1$, we have

1219: \begin{equation} \label{E:Addition}

1220: \hh\rr(\pz\pz') \ge \hh\rr(\pz) + \hh\rr(\pz').

1221: \end{equation}

1222: \end{prop}

1223: 

1224: \begin{proof}

1225: $(i)$ A $\sm_1$-chain is simply a $\sm$-chain,

1226: hence every subset of~$\Div(\pz)$ gives such

1227: a chain. So the maximal $\sm_1$-chain in

1228: $\Div(\pz)$ is $\Div(\pz)$ itself, and

1229: $\hh1(\pz)$ is the cardinality of~$\Div(\pz)$.

1230: 

1231: On the other hand, no $\sm_\nn$-chain

1232: in~$B_\nn^+$ has length more than~$1$, as the

1233: main generator of a $\s$-positive $\nn$~strand

1234: braid word cannot be $\ss\nn$ or above. So

1235: $\hh\nn(\pz)$ is~$1$.

1236: 

1237: Then, for $\qq \le \rr$, every

1238: $\sm_\rr$-chain is a $\sm_\qq$-chain, which

1239: implies $\hh\rr(\pz) \ge \hh\qq(\pz)$.

1240: 

1241: Point~$(ii)$ is obvious, as the

1242: concatenation of two $\sm_\rr$-chains is a

1243: $\sm_\rr$-chain.

1244: \end{proof}

1245: 

1246: From~\eqref{E:Addition} we deduce

1247: $\hh\rr(\pz^\dd) \ge \dd \cdot \hh\rr(\pz)$

1248: for all~$\rr, \pz$. By Lemma¨\ref{L:Degree}, every

1249: divisor of~$\DD\nn\dd$ can be decomposed as the

1250: product of at most~$\dd$ divisors

1251: of~$\D_\nn$, and the latter are

1252: $\nn!$~in number, so we obtain the (coarse) bounds

1253: \begin{equation}

1254: \dd \cdot \hhh\rr\nn{} \le \hhh\rr\nn\dd \le

1255: (\nn!)^\dd

1256: \end{equation}

1257: for all~$\rr, \nn, \dd$. Better estimates

1258: will be given below.

1259: 

1260: \begin{rema}

1261: Instead of restricting to subsets of~$B_\infty$ of the

1262: form~$\Div(\pz)$, we can define the complexity and the

1263: $\rr$-height for every (finite) set of braids~$\XX$. Most

1264: general results extend, but, when $\XX$ is not closed

1265: under left division, nothing can be said about the

1266: number of~$\ss\rr$'s involved in a

1267: $\rr$-jump. Considering such an extension is not useful

1268: here. 

1269: \end{rema}

1270: 

1271: \subsection{Connection with the complexity}

1272: 

1273: We shall now connect the complexity¨$\cc(\pz)$ with

1274: the numbers¨$\hh\rr(\pz)$ just defined. The result

1275: is simple:

1276: 

1277: \begin{prop} \label{P:MainHeight}

1278: For $\pz$ a positive braid with main

1279: generator¨$\ss\mm$, we have

1280: \begin{equation}

1281: \cc(\pz) = \hh\mm(\pz) - 1.

1282: \end{equation}

1283: In particular, for $\nn \ge 2$ and $\dd

1284: \ge 0$, we have 

1285: \begin{equation} \label{E:Complexity}

1286: \ccc\nn\dd = \hhh{\nn-1}\nn\dd - 1.

1287: \end{equation}

1288: \end{prop}

1289: 

1290: One inequality is easy:

1291: 

1292: \begin{lemm} \label{L:MainHeight}

1293: For $\pz$ a positive braid with main

1294: generator¨$\ss\mm$, we have $\cc(\pz) \le

1295: \hh\mm(\pz) - 1$.

1296: \end{lemm}

1297: 

1298: \begin{proof} 

1299: The argument is reminiscent of that used for

1300: Lemma~\ref{L:Enum}, but requires a little more care.

1301: Assume that $\ww$ is a $\s$-positive word drawn

1302: in¨$\CG(\pz)$ from¨$\bx$ containing

1303: $\NN_\mm$¨occurrences of¨$\ss\mm$. By

1304: Lemma¨\ref{L:Decomp}, we can assume $\bx = 1$ without

1305: loss of generality. Let $\pz_0

1306: \sm \pz_1 \sm \dots \sm \pz_\NNN$ be the increasing

1307: enumeration of¨$\Div(\pz)$. By definition, all prefixes

1308: of¨$\ww$ represent divisors of¨$\pz$, so, letting¨$\ell$

1309: be the length of¨$\ww$, there exists a mapping $f : \{0,

1310: \dots, \ell\} \to  \{0,

1311: \dots, \NNN\}$ such that, for each¨$\kk$, the

1312: length¨$\kk$ prefix of¨$\ww$

1313: represents¨$\pz_{f(\kk)}$. By construction, we have

1314: $f(0) = 0$ and $f(\ell) = \NNN$.

1315: 

1316: The difference with Lemma~\ref{L:Enum} is

1317: that $f$ need not be increasing. Now, let

1318: $p_1$, \dots, $p_{\NN_\mm}$ be the

1319: $\NN_\mm$~positions in¨$\ww$ where the

1320: generator~$\ss m$ occurs, completed with

1321: $p_0 = 0$. Then, in the prefix of¨$\ww$ of

1322: length¨$p_1$, \ie, in the subword of¨$\ww$

1323: corresponding to positions from¨$p_0+1$

1324: to¨$p_1$, there is one¨$\ss m$, plus

1325: letters¨$\ss i^{\pm1}$ with $i <

1326: m$ (Figure¨\ref{F:Jump}). This subword is

1327: therefore $\s$-positive, hence we

1328: must have $\pz_{f(p_0)} \sm \pz_{f(p_1)}$,

1329: which requires

1330: $f(p_0) < f(p_1)$. Moreover, the quotient

1331: $\pz_{f(p_0)}\inv

1332: \pz_{f(p_1)}$ is a braid that admits at least

1333: one $\s$-positive expression containing¨$\ss m$,

1334: hence $\pz_{f(p_0)}

1335: \sm_\mm \pz_{f(p_1)}$ holds. Now the same

1336: is true between¨$f(p_1)$ and¨$f(p_2)$,

1337: \etc\ Hence the number of $\mm$-jumps in the

1338: increasing enumeration of~$\Div(\pz)$ is at

1339: least~$\NN_\mm$, \ie, we have $\hh\mm(\pz) \ge

1340: \NN_\mm + 1$.

1341: \end{proof}

1342: 

1343: \begin{figure} [htb]

1344: \begin{picture}(120,27)(0, -0)

1345: \put(0,2){\includegraphics{Jump.eps}}

1346: \put(87,23){almost-positive}

1347: \put(87,19){expression}

1348: \put(102,15){of¨$\pz$}

1349: \put(87,5){increasing }

1350: \put(87,1){enumeration of $\Div(\pz)$}

1351: \put(0,0){$\pz_0$}

1352: \put(10,0){$\pz_1$}

1353: \put(20,0){$\pz_2$}

1354: \put(24,0){$\sm_\mm$}

1355: \put(30,0){$\pz_3$}

1356: \put(40,0){$\pz_4$}

1357: \put(50,0){$\pz_5$}

1358: \put(54,0){$\sm_\mm$}

1359: \put(60,0){$\pz_6$}

1360: \put(70,0){$\pz_7$}

1361: \put(80,0){$\pz_8$}

1362: \put(2,9){$\scriptstyle f(0)$}

1363: \put(12,9){$\scriptstyle f(2)$}

1364: \put(22,9){$\scriptstyle f(1)$}

1365: \put(32,9){$\scriptstyle f(4)$}

1366: \put(42,9){$\scriptstyle f(3)$}

1367: \put(62,9){$\scriptstyle f(5)$}

1368: \put(62,12){$\scriptstyle f(7)$}

1369: \put(72,9){$\scriptstyle f(6)$}

1370: \end{picture}

1371: \caption{\smaller Proof of

1372: Lemma~\ref{L:MainHeight}: the main

1373: generator~$\ss\mm$ corresponds to the bold

1374: arrow: the function¨$f$ need not be

1375: increasing, but the projection of a bold

1376: arrow upstairs must include at least one bold

1377: arrow downstairs, \ie, at least one

1378: $\mm$-jump.}

1379: \label{F:Jump}

1380: \end{figure}

1381: 

1382: It remains to prove the second inequality in

1383: Proposition~\ref{P:MainHeight}, \ie, to prove

1384: that, if $\pz$ is a positive $\nn$-braid

1385: satisfying $\hh{\mm}(\pz) = \NN+1$, then

1386: $\pz$ admits an almost-positive expression

1387: containing $\NN$~generators~$\ss\mm$. 

1388: The problem is as follows: if $\pz$ is a

1389: positive braid and $\bx, \by$ are left

1390: divisors of~$\pz$ satisfying $\bx \sm \by$,

1391: then, by definition, the quotient $\bx\inv

1392: \by$ admits some $\s$-positive

1393: expression~$\ww$, but nothing {\it

1394: a priori } guarantees that $\ww$ be drawn

1395: in~$\CG(\pz)$. In other words, we might have

1396: $\bx \sm \by$ but no $\s$-positive witness

1397: for this inequality inside~$\Div(\pz)$.

1398: This however cannot happen, but the proof

1399: requires a rather delicate argument. 

1400: 

1401: \begin{prop} \label{P:Handle}

1402: Let $\pz$ be a positive braid. Then, 

1403: for all $\bx, \by$ in¨$\Div(\pz)$,

1404: the following are equivalent:

1405: 

1406: $(i)$ The relation $\bx \sm \by$ holds,

1407: \ie, there exists a $\s$-positive path

1408: from~$\bx$ to~$\by$ in the Cayley graph

1409: of~$B_\infty$;

1410: 

1411: $(ii)$ There exists a $\s$-positive path 

1412: from¨$\bx$ to¨$\by$ in the Cayley graph

1413: of~$B_\nn$;

1414: 

1415: $(iii)$ There exists a $\s$-positive path 

1416: from¨$\bx$ to¨$\by$ in¨$\CG(\pz)$.

1417: \end{prop}

1418: 

1419: \begin{proof}

1420: Clearly $(iii)$ implies¨$(ii)$, which 

1421: implies¨$(i)$. We shall prove that $(i)$

1422: implies~$(iii)$---and thus reprove that

1423: $(i)$ implies~$(ii)$, which was first

1424: proved in¨\cite{Lrb}---by

1425: using the handle reduction method of¨\cite{Dfo, Dgr}. The

1426: problem is to prove that, among all $\s$-positive paths

1427: connecting¨$\bx$ to¨$\by$ in the Cayley graph

1428: of¨$B_\infty$, at least one is drawn in~$\CG(\pz)$.

1429: 

1430: Now, let $\uu, \vv$ be positive words representing

1431: $\bx$ and¨$\by$. Then the word¨$\uu\inv\vv$

1432: represents¨$\bx\inv\by$, and, by hypothesis, it is drawn

1433:  in~$\CG(\pz)$ from¨$\bx$. Handle

1434: reduction is an operation that transforms a braid word into

1435: equivalent words and eventually produces a $\s$-positive

1436: word if it exists. It is proved in¨\cite{Dfo} that, for

1437: every $\nn$¨strand braid word¨$\ww$, there exists a

1438: finite fragment¨$\G_\ww$ of the Cayley

1439: graph of~$B_\nn^+$ and a vertex¨$\bx_\ww$

1440: of¨$\G_\ww$ such that

1441: $\ww$ and all words obtained from¨$\ww$ by handle reduction

1442: are drawn from¨$\bx_\ww$ in¨$\G_\ww$. Moreover,

1443: when $\ww$ has the form¨$\uu\inv\vv$ with $\uu, \vv$

1444: positive, then all vertices in¨$\G_\ww$ are the

1445: left divisors of the least common right multiple of

1446: the braids represented by¨$\uu$ and¨$\vv$, here

1447: ¨$\bx$ and¨$\by$,  while $\bx_\ww$ is the braid

1448: represented by¨$\uu$, \ie,¨$\bx$. As

1449: $\bx$ and¨$\by$ are divisors of¨$\pz$, so is their

1450: least common right multiple, and the

1451: graph¨$\G_\ww$ is included in~¨$\CG(\pz)$.

1452: It follows that every word obtained

1453: from¨$\uu\inv\vv$ using handle reduction is

1454: drawn from¨$\bx$ in¨$\CG(\pz)$. The

1455: termination of handle reduction guarantees

1456: that, among these words, at least one is

1457: $\s$-positive, so $(iii)$ follows.

1458: \end{proof}

1459: 

1460: A direct application of Proposition~\ref{P:Handle} is the

1461: existence of $\s$-positive quotient-sequences drawn in

1462: the Cayley graph. The definition is as follows:

1463: 

1464: \begin{defi} \label{D:Witness}

1465: Assume that $\pz$ is a positive braid and

1466: $\XX$ is a subset of~$\Div(\pz)$. Let

1467: $\bx_0 \sm \dots \sm \bx_\NN$ be the

1468: increasing enumeration of~$\XX$. We say that

1469: a sequence of words $\vec\ww = (\ww_1, \dots,

1470: \ww_\NNN)$ is a {\it quotient-sequence} for~$\XX$ if,

1471: for each~$\kk$, the word~$\ww_\kk$ is an  expression

1472: of~$\bx_{\kk-1}\inv \bx_\kk$ for each¨$\kk$. We

1473: say that $\vec\ww$ is {\it $\s$-positive} if every

1474: entry in¨$\vec\ww$ is $\s$-positive, and that

1475: $\vec\ww$ is {\it drawn in¨$\CG(\pz)$} (from¨$\bx_0$)

1476: if $\ww_\kk$ is drawn from~$\bx_{\kk-1}$ in~$\CG(\pz)$

1477: for each¨$\kk$.

1478: \end{defi}

1479: 

1480: \begin{coro} \label{C:Witness}

1481: Assume that $\pz$ is a positive braid. Then

1482: every subset of~$\Div(\pz)$ admits a

1483: $\s$-positive quotient-sequence drawn in¨$\CG(\pz)$.

1484: \end{coro}

1485: 

1486: \begin{exam} \label{X:Jump}

1487:  (Figure¨\ref{F:MaxPath})

1488: By computing the successive quotients in the

1489: increasing enumeration of¨$\DDD32$ given in

1490: Example¨\ref{X:List}, we easily find that

1491: $$\tt (a, a, AAb, a, a, AAb, a, AAb, a, a, bAA, a,

1492: bAA, a, a, bAA, a, a)$$

1493: is a $\s$-positive quotient-sequence for $\Div(\DD32)$

1494: drawn in¨$\GG32$. This sequence turns out to be the

1495: unique sequence with the above properties, but this

1496: uniqueness is specific to the case of $3$-braids ({\it

1497: cf.} Figure~\ref{F:Decomp41} below). 

1498: \end{exam}

1499: 

1500: \begin{figure} [htb]

1501: \begin{picture}(60,45)(0, -0)

1502: \put(0,0){\includegraphics{MaxPath32.eps}}

1503: \put(1,26){$1$}

1504: \put(57,18){$\D_3^2$}

1505: \end{picture}

1506: \caption{\smaller The increasing enumeration of the

1507: divisors of¨$\DD32$, and a $\s$-positive

1508: quotient-sequence drawn in¨$\GG32$: the associated path

1509: visits every vertex, and is labeled

1510: $\tt aaAAbaaAAbabAAaaAAbabAAaabAAaa$; it crosses

1511: $6$ $\ss2$-edges (and no $\ss2\inv$)}

1512: \label{F:MaxPath}

1513: \end{figure}

1514: 

1515: We can now easily complete the proof of

1516: Proposition~\ref{P:MainHeight}:

1517: 

1518: \begin{proof}[Proof of

1519: Proposition~\ref{P:MainHeight}]

1520: Let $(\pz_0, \dots, \pz_\NN)$

1521: be the $\sm$-increasing enumeration\break

1522: of $\Div(\pz)$. By Corollary¨\ref{C:Witness},

1523: there exists a $\s$-positive quotient-sequence¨$\vec\ww$

1524: for¨$\Div(\pz)$ that is drawn in¨$\CG(\pz)$. Let $\ww =

1525: \ww_1 \ldots \ww_\NN$. By construction, $\ww$ is a

1526: $\s$-positive word drawn in¨$\CG(\pz)$, and the number of

1527: occurrences of the main generator¨$\ss\mm$ in¨$\ww$ is

1528: (at least) the number of $\mm$-jumps in¨$(\pz_0, \dots,

1529: \pz_\NN)$. So we have $\cc(\pz) \ge \hh\mm(\pz)

1530: -1$. Owing to Lemma¨\ref{L:MainHeight}, this

1531: completes the proof.

1532: \end{proof}

1533: 

1534: \begin{rema}

1535: Assume that $\vec\ww$ is a $\s$-positive

1536: quotient-sequence  for~$\Div(\pz)$, and

1537: $\ss\mm$ is the main generator

1538: occurring in~$\vec\ww$. Then

1539: each word~$\ww_i$ contains zero

1540: or one letter~$\ss\mm$. Indeed, if $\ww_i$

1541: contained two~$\ss\mm$'s or more, then the

1542: vertex reached after the first~$\ss\mm$ ought

1543: to lie in the open $\sm$-interval determined by two

1544: successive entries of~$\vec\pz$, and the

1545: latter is empty by construction since all

1546: elements of~$\Div(\pz)$ occur in~$\vec\pz$.

1547: \end{rema}

1548: 

1549: \section{A decomposition result for~$\DDiv\pz$}

1550: \label{S:Deter}

1551: 

1552: In this section, we establish a structural result

1553: describing $(\DDD\nn\dd, <)$ as the concatenation of

1554: $\ccc\nn\dd+1$ intervals isomorphic to subsets

1555: of¨$(\DDD{\nn-1}\dd, <)$. We deduce an explicit formula

1556: connecting¨$\hhh\rr\nn\dd$ with the number of braids

1557: in¨$\DDD\nn\dd$ whose $\dd$th factor is right divisible

1558: by¨$\D_\rr$, which in turn enables us to complete the

1559: computation of¨$\ccc\nn\dd$ and¨$\hhh\rr\nn\dd$ for

1560: small values of¨$\rr$, $\nn$ and¨$\dd$.

1561: 

1562: \subsection{$B_\rr$-classes}

1563: 

1564: In order to analyse the linearly ordered

1565: sets¨$\DDDD\nn\dd$, and, more generally, 

1566: $(\Div(\pz), \sm\nobreak)$ for¨$\pz$ a positive braid,

1567: we introduce convenient partitions. As $B_\rr$ is a

1568: group for each~$\rr$, it is clear that the relation

1569: $\bx\inv \by \in B_\rr$ defines an equivalence

1570: relation on (positive) braids, so we may put:

1571: 

1572: \begin{defi}

1573: For $\rr \ge 1$ and $\bx, \by$

1574: in¨$B_\infty^+$, we say that $\bx$ and

1575: $\by$ are {\it $B_\rr$-equivalent} if

1576: $\bx\inv \by$ belongs to¨$B_\rr$.

1577: \end{defi}

1578: 

1579: By construction, $B_\rr$-equivalence is

1580: compatible with multiplication on the left.

1581: In the sequel, we consider the restriction

1582: of $B_\rr$-equivalence to finite subsets

1583: of~$B_\infty^+$ of the form~$\Div(\pz)$,

1584: \ie, we use $B_\rr$-equivalence to partition

1585: $\Div(\pz)$ into subsets, naturally called

1586: $B_\rr$-classes.

1587: 

1588: \begin{exam}

1589: As $B_1$ is trivial, $B_1$-equivalence is

1590: equality, and, therefore, the $B_1$-classes are

1591: singletons. On the other hand, any two elements

1592: of¨$B_\nn$ are

1593: $B_\rr$-equivalent for each¨$\rr \ge n$, so, for $\pz$

1594: in¨$B_\nn^+$, there is only one $B_\rr$-class for $\rr \ge

1595: n$, and the only interesting relations arise for

1596: $1 < \rr < n$. For instance, $\DDD3{}$

1597: contains three $B_2$-classes, while $\DDD32$

1598: contains seven of them

1599: (Figure¨\ref{F:Classes}).

1600: \end{exam}

1601: 

1602: \begin{figure} [htb]

1603: \begin{picture}(100,43)(0, -0)

1604: \put(0,0){\includegraphics{Classes.eps}}

1605: \end{picture}

1606: \caption{\smaller The $B_2$-classes

1607: in¨$\DDD3{}$ and¨$\DDD32$}

1608: \label{F:Classes}

1609: \end{figure}

1610: 

1611: Saying that there is an $\rr$-jump

1612: between two braids¨$\bx$ and¨$\by$ means that

1613: $\bx\inv \by$ is $\s$-positive and does not

1614: belong to¨$B_\rr$, so, for $\bx \sm \by$, we

1615: have the equivalence 

1616: \begin{equation} \label{E:Jump}

1617: \big(\mbox{$\bx, \by$ are not

1618: $B_\rr$-equivalent}\big)

1619: \Longleftrightarrow

1620: \left(

1621: \begin{matrix}

1622: \mbox{there is a $\rr$-jump between} \\

1623: \mbox{between¨$\bx$ and¨$\by$}

1624: \end{matrix}

1625: \right).

1626: \end{equation} 

1627: 

1628: \begin{lemm} \label{L:Interval}

1629: Assume that $\pz$ is a positive braid. Then,

1630: each $B_\rr$-class in¨$\Div(\pz)$ is an

1631: interval for¨$\sm$ and there is an $\rr$-jump

1632: between each $B_\rr$-class and the next one.

1633: \end{lemm}

1634: 

1635: \begin{proof}

1636: Assume $\bx \sm \by \in \Div(\pz)$. By

1637: \eqref{E:Jump}, if $\bx$ and $\by$

1638: are not $B_\rr$-equivalent, there is an

1639: $\rr$-jump between¨$\bx$ and¨$\by$, hence

1640: between¨$\bx$ and any element of¨$\Div(\pz)$

1641: above¨$\by$, so no such element may be

1642: $B_\rr$-equivalent to¨$\bx$. This implies

1643: that each $B_\rr$-class is an

1644: $\sm$-interval.

1645: \end{proof}

1646: 

1647: \begin{coro} \label{C:NbClasses}

1648: For each $\rr \ge 1$, 

1649: the number of $B_\rr$-classes in¨$\Div(\pz)$

1650: is¨$\hh\rr(\pz)$.

1651: \end{coro}

1652: 

1653: \begin{proof}

1654: By¨\eqref{E:Jump}, there is no

1655: $\rr$-jump between two elements of the same

1656: $B_\rr$-class, and there is one between two

1657: elements not in the same $B_\rr$-class. Thus

1658: the number of $B_\rr$-classes is the number

1659: of $\rr$-jumps in the $\sm$-increasing

1660: enumeration of¨$\Div(\pz)$ augmented by¨$1$,

1661: hence, by Lemma~\ref{L:Enum}, it is¨$\hh\rr(\pz)$.

1662: \end{proof}

1663: 

1664: $B_\rr$-equivalence provides a partition

1665: of¨$(\Div(\pz), \sm)$ into finitely many

1666: subintervals. The interest of this partition

1667: is that we can describe $B_\rr$-classes rather

1668: precisely and, typically, connect them with

1669: subsets of¨$B_\rr$. In particular, this will

1670: allow for connecting the ordered

1671: sets¨$\DDDD\nn\dd$ with the

1672: sets¨$(\DDD{\nn-1}\dd, \sm)$.

1673: 

1674: \begin{prop} \label{P:Lattice}

1675: (Figure¨\ref{F:Decomp})

1676: Assume $\pz \in B_\infty^+$ and $\rr \ge 1$. Let

1677: $\Class$ be a

1678: $B_\rr$-class in¨$\Div(\pz)$, and let

1679: $\ea, \eb$ be its $\sm$-extremal elements. Then $\ea$

1680: divides every element of¨$\Class$ on the left, and the left

1681: translation by¨$\ea$ defines an isomorphism of

1682: $(\Div(\ea\inv\eb), \dive, \sm)$ onto¨$(\Class, \dive,

1683: \sm)$. In particular, $(\Class, \dive)$ is a

1684: lattice.

1685: \end{prop}

1686: 

1687: \begin{proof}

1688: By Lemma¨\ref{L:Interval}, $\Class$ is the

1689: $\sm$-interval determined by¨$\ea$ and¨$\eb$,

1690: \ie, we have $$\Class = \{\bx

1691: \in \Div(\pz) ; \ea \sm \bx \sm \eb\}.$$

1692: 

1693: We know that $\Div(\pz)$ is a

1694: lattice with respect to left divisibility:

1695: any two elements¨$\bx, \by$ of¨$\Div(\pz)$

1696: admit a greatest left common divisor, here

1697: denoted $\gcd(\bx, \by)$, and a least common

1698: right multiple, denoted¨$\lcm(\bx, \by)$.

1699: Firstly, we claim that $\Class$ is a lattice with

1700: respect to left divisibility, \ie, the left gcd 

1701: and the right lcm of two elements of¨$\Class$

1702: lie in¨$\Class$. So assume $\bx, \by \in \Class$. Let $\bx_0,

1703: \by_0$ be defined by $\bx = \gcd(\bx, \by)

1704: \bx_0$ and $\by = \gcd(\bx,

1705: \by) \by_0$.  The hypothesis that $\bx\inv\by$ belongs

1706: to¨$B_\rr$ implies that there exist $\bx_1, \by_1$ in¨$B_\rr^+$

1707: satisfying $\bx\inv \by = \bx_1\inv \by_1$. By definition of

1708: the gcd, there must exist a positive

1709: braid¨$\pz_1$ satisfying $\bx_1 = \pz_1 \bx_0$

1710: and $\by_1 = \pz_1 \by_0$. Because

1711: $\pz_1$ is positive, $\bx_1 \in B_\rr^+$

1712: implies $\bx_0 \in B_\rr^+$, hence

1713: $\gcd(\bx, \by) \in \Class$. As for the lcm,

1714: the conjunction of $\bx = \gcd(\bx,

1715: \by)

1716: \bx_0$ and $\by = \gcd(\bx, \by) \by_0$

1717: implies $$\lcm(\bx, \by) = \gcd(\bx,

1718: \by)\lcm(\bx_0, \by_0).$$ As $\bx_0, \by_0

1719: \in B_\rr^+$ implies $\lcm(\bx_0, \by_0) \in

1720: B_\rr^+$, we deduce $\lcm(\bx, \by) \in \Class$.

1721: 

1722: As $\Class$ is finite, it follows that $\Class$

1723: admits a global gcd. Because the linear

1724: ordering¨$\sme$ extends the partial

1725: divisibility ordering¨$\dive$, this global

1726: gcd must be the $\sm$-minimum¨$\ea$ of¨$\Class$.

1727: Symmetrically, $\Class$ admits a global lcm,

1728: which must be the $\sm$-maximum¨$\eb$. So, at

1729: this point, we know that $\ea$ is a left

1730: divisor of every element in¨$\Class$, and $\eb$

1731: is a right multiple of each such element,

1732: \ie, we have 

1733: \begin{equation} \label{E:Inclusion}

1734: \Class \ince \{\bx \in B_\infty^+ ; \ea \dive

1735: \bx \dive \eb\}.

1736: \end{equation}

1737: Moreover, $\ea \dive \bx \dive \eb$

1738: implies  $\ea \sme \bx \sme \eb$, hence $\bx

1739: \in \Class$, so the inclusion

1740: in¨\eqref{E:Inclusion} is an

1741: equality.

1742: 

1743: Now, put $F(\bx) = \ea\bx$ for¨$\bx$

1744: in¨$\Div(\ea\inv \eb)$. As $B_\infty^+$ is

1745: left cancellative, $F$ is injective.

1746: Moreover, for $\bx$ a positive braid, $\bx

1747: \dive

1748: \ea\inv

1749: \eb$ is equivalent to $\ea\bx \dive \eb$, so

1750: the image of¨$F$ is $\{\bx \in B_\infty^+ ;

1751: \ea \dive

1752: \bx \dive \eb\}$, hence is¨$\Class$. Finally, by

1753: construction, $F$ preserves both¨$\dive$

1754: and¨$\sm$.

1755: \end{proof}

1756: 

1757: \begin{figure} [htb]

1758: \begin{picture}(86,44)(0, 0)

1759: \put(2,0){\includegraphics{Decomp.eps}}

1760: \put(25,32){$C$}

1761: \put(25,7){$a$}

1762: \put(39,29){$b$}

1763: \put(0,0){$1$}

1764: \put(21,19){$\ss\rr$}

1765: \put(41,23){$\ss\rr$}

1766: \put(62,27){$\ss\rr$}

1767: \put(86,40){$\pz$}

1768: \end{picture}

1769: \caption{\smaller Decomposition of $(\Div(\pz),

1770: \sm)$ into $B_\rr$-classes: each class¨$C$ is a

1771: lattice with respect to divisibility; the increasing

1772: enumeration of¨$\Div(\pz)$ exhausts the first class,

1773: then jumps to the next one by an $\rr$-jump, \etc\ The

1774: number of classes is¨$\hh\rr(\pz)$.}

1775: \label{F:Decomp}

1776: \end{figure}

1777: 

1778: For¨$\rr = 1$, each $B_\rr$-class is a singleton, and

1779: Proposition¨\ref{P:Lattice} says nothing; similarly, if

1780: the main generator of¨$\pz$ is¨$\ss\mm$,

1781: there is only one $B_\rr$-class for $\rr >

1782: \mm$, and we gain no information. But, for

1783: $1 < \rr \le \mm$, and specially for $\rr =

1784: \mm$, Proposition¨\ref{P:Lattice} states that

1785: the chain¨$\Div(\pz)$ is obtained by

1786: concatenating $\hh\rr(\pz)$ copies of sets of

1787: the form¨$\Div(\pz')$ with¨$\pz'$ of index at

1788: most¨$\rr$. In particular, for $\pz = \DD

1789: n\dd$, we have

1790: 

1791: \begin{coro} \label{C:Structure}

1792: For each¨$\nn$ and each¨$\rr$ with $\rr < \nn$, the

1793: chain $\DDDD\nn\dd$ is obtained by  concatenating

1794: $\hhh\rr\nn\dd$ in\-tervals, each of which,

1795: when equipped with¨$\dive$, is a translated

1796: copy of some initial sublattice

1797: of¨$(\DDD\rr\dd, \dive)$.

1798: \end{coro}

1799: 

1800: The case of¨$\DD32$ and¨$\DD4{}$ are illustrated in

1801: Figures¨\ref{F:Decomp32} and¨\ref{F:Decomp41}.

1802: 

1803: \begin{figure} [htb]

1804: \begin{picture}(64,43)(0, 0)

1805: \put(2,0){\includegraphics{Decomp32.eps}}

1806: \put(1,25){$1$}

1807: \put(63,20){$\DD32$}

1808: \end{picture}

1809: \caption{\smaller Decomposition of

1810: $\DDDD32$ into $B_2$-classes: the increasing

1811: enumeration of¨$\DDDD32$ is the concatenation

1812: of the increasing enumeration of the successive

1813: classes, separated by $2$-jumps (compare with

1814: Figure¨\ref{F:MaxPath}); in this case, $B_2$-classes

1815: are simply chains with respect to divisibility}

1816: \label{F:Decomp32}

1817: \end{figure}

1818: 

1819: \begin{figure} [htb]

1820: \begin{picture}(104,48)(0, 0)

1821: \put(2,0){\includegraphics{Decomp41.eps}}

1822: \put(7,0){$1$}

1823: \put(97,46){$\DD4{}$}

1824: \end{picture}

1825: \caption{\smaller Decomposition of

1826: $\DDDD4{}$ into $B_3$-classes; note that the

1827: $\ss3$-arrows (thick) corresponding to 3-jumps

1828: are not unique; in this case, all $B_3$-classes are

1829: isomorphic to the lattice $(\DDD3{}, \sm, \dive)$, \ie,

1830: to the Cayley graph of¨$\D_3$}

1831: \label{F:Decomp41}

1832: \end{figure}

1833: 

1834: 

1835: \subsection{Extremal elements}

1836: 

1837: The next step is to observe that extremal

1838: points in $B_\rr$-classes admit a simple

1839: characterization in terms of divisibility.

1840: 

1841: \begin{prop} \label{P:Extremal}

1842: Assume that $\pz$ is a positive braid. 

1843: 

1844: $(i)$ An element¨$\bx$ of¨$\Div(\pz)$ is the maximum of its

1845: $B_\rr$-class if and only if the relation $\bx\ss i

1846: \dive\pz$ fails for $1 \le i < \rr$.

1847: 

1848: $(ii)$ An element¨$\bx$ of¨$\Div(\pz)$ is the minimum of its

1849: $B_\rr$-class if and only if no¨$\ss i$ with $1 \le i <

1850: \rr$ divides¨$\bx$ on the right.

1851: \end{prop}

1852: 

1853: \begin{proof}

1854: $(i)$ The condition is necessary: if $\bx \ss i$ lies

1855: in¨$\Div(\pz)$ for some¨$i$ with $i < \rr$, then $\bx \ss

1856: i$ lies in the same $B_\rr$-class as¨$\bx$, and it is larger

1857: both for¨$\dive$ and¨$\sm$, so $\bx$ cannot be maximal in

1858: its $B_\rr$-class. Conversely, assume that $\bx$ is

1859: not maximal in its $B_\rr$-class. Then there

1860: exists¨$\by$ satisfying $\bx \sm \by$ and $\by$ is

1861: $B_\rr$-equivalent to¨$\bx$. Now, by 

1862: Proposition¨\ref{P:Lattice},

1863: the lcm of¨$\bx$ and¨$\by$ is also $B_\rr$-equivalent

1864: to¨$\bx$, which means that there exists¨$\by_1$ in¨$B_\rr^+$

1865: satisfying $\lcm(\bx, \by) = \bx\by_1$. Now $\bx \sm \by$

1866: implies $\by_1 \not= 1$, so there must exist¨$i < m$ such

1867: that $\ss i$ is a left divisor of¨$\by_1$. Then we

1868: have $\bx\ss i \dive \bx\by_1 \dive \pz$, hence $\bx

1869: \ss i \dive \pz$.

1870: 

1871: $(ii)$ The argument is symmetric. If we have $\bx = 

1872: \by \ss i$ for some positive braid¨$\by$ and $i < \rr$, then

1873: $\by$ belongs to the $B_\rr$-class of¨$\bx$, and $\bx$

1874: cannot be minimal in its $B_\rr$-class. Conversely, assume

1875: that $\bx$ is not minimal in its $B_\rr$-class. Then there

1876: exists¨$\by$ satisfying $\by \sm \bx$ and $\by$ is

1877: $B_\rr$-equivalent to¨$\bx$. By

1878: Proposition¨\ref{P:Lattice} again, the gcd

1879: of¨$\bx$ and¨$\by$ is also $B_\rr$-equivalent

1880: to¨$\bx$, which means that there

1881: exists¨$\by_0$ in¨$B_\rr^+$ satisfying

1882: $\gcd(\bx, \by) \by_0 = \bx$. As $\by \sm

1883: \bx$ implies $\by_0 \not= 1$, there must exist¨$i < m$

1884: such that $\ss i$ is a right divisor of¨$\by_0$, hence

1885: of¨$\bx$.

1886: \end{proof}

1887: 

1888: When we apply the previous criterion to the

1889: braids¨$\DD\nn\dd$, we obtain:

1890: 

1891: \begin{prop} \label{P:MaxClass}

1892: For $\bx$ in¨$\DDD\nn\dd$ and $1 \le

1893: \rr \le \nn$, the

1894: following are equivalent:

1895: 

1896: $(i)$ The element¨$\bx$ is $\sm$-maximal in its

1897: $B_\rr$-class;

1898: 

1899: $(ii)$ The element¨$\bx \ss i$ belongs to¨$\Div(\DD\nn\dd)$

1900: for no¨$i < \rr$;

1901: 

1902: $(iii)$ The $\dd$th factor of¨$\bx$ is

1903: right divisible by¨$\D_\rr$.

1904: 

1905: $(iv)$ The $\dd+1$st factor of¨$\bx \D_\rr$

1906: is¨$\D_\rr$.

1907: \end{prop}

1908: 

1909: \begin{proof}

1910: The equivalence of¨$(i)$ and¨$(ii)$ is given

1911: by Proposition¨\ref{P:Extremal}$(i)$. It

1912: remains to establish the equivalence

1913: of¨$(ii)$--$(iv)$. For $\rr = 1$, $(ii)$ is

1914: vacuously true, while $(iii)$ and

1915: $(iv)$ always hold. So the expected equivalences are true.

1916: We henceforth assume $\rr \ge 2$.

1917: 

1918: Let $\bx$ belong to¨$\DDD\nn\dd$, and let $\sx_\dd$ be the $\dd$th

1919: factor in the normal form of¨$\bx$. For $i < n$, saying that

1920: $\bx \ss i$ does not belong to¨$\Div(\DD\nn\dd)$ means that

1921: the normal form of¨$\bx \ss i$ has length $\dd+1$, hence,

1922: equivalently, that the normal form of¨$\sx_\dd \ss i$ has

1923: length¨$2$. This occurs if and only if

1924: $\ss i$ is a right divisor of¨$\sx_\dd$. So, for

1925: $\rr \le n$,

1926: $(ii)$ is equivalent to $\sx_\dd$ being right divisible by

1927: all¨$\ss i$'s with $1 \le i < \rr$, hence to

1928: $\sx_\dd$ being right divisible by the (left)

1929: lcm of these elements, which is¨$\D_\rr$.

1930: 

1931: Finally, $(iii)$ and $(iv)$ are equivalent. Indeed, if

1932: the $\dd$th factor¨$\sx_\dd$ in the normal form of¨$\bx$ is

1933: divisible by¨$\D_\rr$ on the right, then $(\sx_\dd, \D_\rr)$ is a

1934: normal sequence as no $\ss i$ with $i < \rr$

1935: from¨$\D_\rr$ may pass to¨$\sx_\dd$. Hence

1936: $(\sx_1, \ldots, \sx_\dd, \D_\rr)$ is a normal

1937: sequence, necessarily the normal form

1938: of¨$\bx\D_\rr$.  Conversely, assume that the

1939: normal form of¨$\bx\D_\rr$ is

1940: $(\sx_1, \ldots, \sx_\dd, \D_\rr)$. The hypothesis that

1941: $(\sx_\dd, \D_\rr)$ is normal implies that $\sx_\dd$ is divisible

1942: on the right by each¨$\ss i$ with $i < \rr$,

1943: hence is divisible on the right by¨$\D_\rr$.

1944: Now $(\sx_1, \ldots,

1945: \sx_\dd)$ is the normal form of¨$\bx$.

1946: \end{proof}

1947: 

1948: Observe that, for $\rr \ge 2$, an element

1949: of¨$\DDD\nn\dd$ that is $\sm$-maximal in its

1950: $B_\rr$-class cannot

1951: belong to¨$\DDD\nn{\dd-1}$, \ie, cannot have

1952: degree $\dd-1$ or less, since the

1953: $\dd$th factor of its normal form cannot be¨$1$.

1954: 

1955: Similar conditions characterize the minimal elements of the

1956: $B_\rr$-classes. Because the normal form has a priviledged

1957: orientation, the results are not entirely symmetric of

1958: those of Proposition¨\ref{P:MaxClass}

1959: 

1960: \begin{prop} \label{P:MinClass}

1961: For $\bx$ in¨$\DDD\nn\dd$ and $1 \le

1962: \rr \le \nn$, the following are equivalent:

1963: 

1964: $(i)$ The element¨$\bx$ is $\sm$-minimal of its

1965: $B_\rr$-class;

1966: 

1967: $(ii)$ No¨$\ss i$ with $i < \rr$ is a right divisor

1968: of¨$\bx$;

1969: 

1970: $(iii)$ The degrees of¨$\bx$ and¨$\bx \D_\rr$ are equal.

1971: \end{prop}

1972: 

1973: \begin{proof}

1974: The equivalence of¨$(i)$ and¨$(ii)$ is given by

1975: Proposition¨\ref{P:Extremal}$(ii)$. On the other hand,

1976: everything is obvious for¨$\rr = 1$. So it remains to

1977: establish the equivalence of¨$(ii)$ and¨$(iii)$ in the case

1978: $\rr \ge 2$. Now, assume that¨$(ii)$ holds an $\bx$ has

1979: degree¨$\dd$. The hypothesis that $\ss i$ is not a right

1980: divisor of¨$\bx$ implies that $\bx \ss i$ is a divisor

1981: of¨$\DD\nn\dd$. As this holds for each $i <

1982: \rr$, the lcm of $\bx \ss1$, \dots, $\bx \ss{\rr-1}$, which

1983: is $\bx \D_\rr$, also divides¨$\DD\nn\dd$, which means that

1984: $\bx \D_\rr$ has degree (at most)¨$\dd$. So $(ii)$

1985: implies¨$¨(iii)$.

1986: 

1987: Conversely, assume that $\ss i$ divides¨$\bx$ on the right.

1988: Then the degree of¨$\bx \ss i$ is strictly larger than that

1989: of¨$\bx$, and, {\it a fortiori}, the same is true for¨$\bx

1990: \D_\rr$.

1991: \end{proof}

1992: 

1993: \subsection{Determination of¨$\hhh\rr\nn\dd$}

1994: 

1995: A direct application of the previous results is a formula

1996: connecting the number of $B_\rr$-classes

1997: in¨$\Div(\DD\nn\dd)$, \ie, the

1998: numbers¨$\hhh\rr\nn\dd$, with the number of

1999: braids whose normal form ends with some

2000: specific factor.

2001: 

2002: \begin{defi}

2003: For $\nn, \dd \ge 1$ and for¨$\sp$ a simple

2004: $\nn$-braid, we denote by¨$\bbb\nn\dd\sp$ the number

2005: of positive braids of degree at most¨$\dd$, \ie, of

2006: divisors of¨$\DD\nn\dd$, whose $\dd$th factor is¨$\sp$.

2007: \end{defi}

2008: 

2009: \begin{prop} \label{P:Main}

2010: For $1 \le \rr \le \nn$, we have

2011: \begin{equation} \label{E:Main}

2012: \hhh\rr\nn\dd = \sum_{\mbox{\Small $s$ right divisible by

2013: $\D_\rr$}} \bbb\nn\dd s  = \bbb \nn{\dd+1}{\D_\rr} .

2014: \end{equation} 

2015: \end{prop}

2016: 

2017: In words: The number of $\rr$-jumps in $\DDDD\nn\dd$ is

2018: the number of

2019: $\nn$-braids of degree at most¨$\dd$ whose $\dd$th factor

2020: is right divisible by¨$\D_\rr$.

2021: 

2022: \begin{proof}

2023: By Corollary¨\ref{C:NbClasses},

2024: $\hhh\rr\nn\dd$ is the number of

2025: $B_\rr$-classes in¨$\Div(\DD\nn\dd)$. Each

2026: class contains exactly one maximum element, and, by

2027: Proposition¨\ref{P:MaxClass}, the latter are

2028: characterized by the property that their

2029: $\dd$th factor is right

2030: divisible by¨$\D_\rr$. The first equality

2031: in¨\eqref{E:Main} follows. The second one

2032: follows from the equivalence of¨$(iii)$

2033: and¨$(iv)$ in Proposition¨\ref{P:MaxClass}.

2034: \end{proof}

2035: 

2036: For $\rr = 1$, as every simple braid is divisible by¨$1$ on

2037: the right, Relation¨\eqref{E:Main} reduces to

2038: \begin{equation} \label{E:MainOne}

2039: \hhh1\nn\dd = \sum_{\sp} \bbb\nn\dd\sp = \bbb

2040: \nn{\dd+1}{1},

2041: \end{equation}

2042: a special case of the relation $\hh1(\pz) =

2043: \card\Div(\pz)$ of

2044: Proposition¨\ref{P:Decreasing}. 

2045: For $\rr = \nn$, as the only normal sequence of

2046: length¨$\dd$  that finishes with¨$\D_\nn$ is  $(\D_\nn,

2047: \dots, \D_\nn)$, Relation¨\eqref{E:Main}

2048: reduces to

2049: \begin{equation} \label{E:Mainn}

2050: \hhh\nn\nn\dd = 1,

2051: \end{equation}

2052: also noted in Proposition¨\ref{P:Decreasing}.

2053: Finally, for $\rr = \nn-1$, we obtain using

2054: Proposition¨\ref{P:MainHeight}:

2055: 

2056: \begin{coro} \label{C:Main}

2057: For $\nn \ge 2$, we have

2058: \begin{equation} \label{E:MainNO}

2059: \ccc\nn\dd = \hhh{\nn - 1}\nn\dd -1 = \sum_{i =

2060: 2}^{\nn}

2061: \bbb\nn\dd{\ss i \ss{i+1} \dots \ss{\nn-1}\D_{\nn - 1}} =

2062: \bbb\nn{\dd+1}{\D_{\nn - 1}} - 1.

2063: \end{equation}

2064: \end{coro}

2065: 

2066: \begin{proof}

2067: The simple $\nn$-braids that are right divisible

2068: by¨$\D_{\nn-1}$ are the braids of the form

2069: $\ss i \ss{i+1} \dots \ss{\nn-1}$ with $1 \le i\le

2070: \nn$. Indeed,

2071: it is clear that every such braid is simple

2072: and right divisible by¨$\D_{\nn-1}$.

2073: Conversely, the only possibility for $\pz

2074: \D_{\nn-1}$ to be simple is that $\pz$ moves the $\nn$th

2075: strand to some position between¨$1$ and¨$\nn$, but

2076: introduces no crossing between the remaining strands.

2077: Finally, $\ss1 \ss2 \dots \ss{\nn-1}\D_{\nn-1}$

2078: is¨$\D_\nn$, and we already observed that

2079: $\bbb\nn\dd{\D_\nn}$ is¨$1$, so we obtain the first

2080: equality in¨\eqref{E:MainNO}.

2081: \end{proof}

2082: 

2083: \subsection{Computation of¨$\bbb\nn\dd\sp$}

2084: 

2085: By Lemma¨\ref{L:Normal}, normal sequences are

2086: characterized by a local condition involving only pairs

2087: of consecutive elements. It follows that the set of all

2088: normal sequences is a rational set, \ie, it can be

2089: recognized by a finite state automaton. Standard

2090: arguments then show that the numbers¨$\bbb\nn\dd\sp$ obey

2091: a linear recurrence. Building on this observation,

2092: seemingly first used in the case of braids in¨\cite{Chb},

2093: we can obtain explicit formulas for the

2094: parameters¨$\ccc\nn\dd$ and¨$\hhh\rr\nn\dd$ for small

2095: values of¨$\rr$, $\nn$, and/or¨$\dd$. We shall not go

2096: into details in the current paper, and refer

2097: to¨\cite{Dhi} where all formulas are established---and

2098: where, more generally, the rich combinatorics underlying

2099: the normal form of braids is investigated.

2100: 

2101: In the sequel, we write $(M)_{\xx, \yy}$ for the $(\xx,

2102: \yy)$-entry of a matrix¨$M$. The general principle for

2103: computing the numbers¨$\bbb\nn\dd\sp$ for some

2104: fixed¨$\nn$ is as follows:

2105: 

2106: \begin{lemm} \label{L:Comput}

2107: For $\nn \ge 1$, let $M_\nn$ be the square matrix with

2108: entries indexed by simple $\nn$-braids defined by

2109: $$

2110: (M_\nn)_{\sp, \spp} = 

2111: \begin{cases}

2112: 1 & \mbox{if $(\sp, \spp)$ is normal,}\\

2113: 0 & \mbox{otherwise.}

2114: \end{cases}$$

2115: Then, for every simple¨$\spp$ and every

2116: $\dd \ge 1$, we have $\bbb\nn\dd{\spp} = ((1, 1, \dots, 1)

2117: \, M_\nn\!\!\!{}^{\dd-1})_\spp$.

2118: \end{lemm}

2119: 

2120: The proof is an easy induction on¨$\dd$.

2121: 

2122: \begin{exam}

2123: The matrix¨$M_1$ is

2124: $(1)$, corresponding to

2125: $\bbb1\dd1 = 1$. For $\nn=2$, using the enumeration $(1,

2126: \ss1)$ of simple $2$-braids, we find $M_2 = 

2127: \left(\begin{matrix}1&0\\1&1\end{matrix}\right)$,

2128: leading to

2129: $\bbb2\dd1 = \dd$, $\bbb2\dd{\ss1} = 1$, as could

2130: be expected: there are $\dd+1$ braids of degree

2131: at most¨$\dd$, namely the braids¨$\ss1^\expo$ with

2132: $\expo < \dd$, whose $\dd$th factor is¨$1$, and

2133: $\ss1^\dd$, whose $\dd$th factor is¨$\D_2$, \ie,

2134: ¨$\ss1$. For $\nn = 3$, using the enumeration $(1, \ss1,

2135: \ss2,\ss2\ss1, \ss1\ss2, \D_3)$ of simple $3$-braids,  we

2136: obtain

2137: $$M_3 = \left(

2138: \begin{matrix}

2139: 1&0&0&0&0&0\\

2140: 1&1&0&0&1&0\\

2141: 1&0&1&1&0&0\\

2142: 1&1&0&0&1&0\\

2143: 1&0&1&1&0&0\\

2144: 1&1&1&1&1&1

2145: \end{matrix}

2146: \right),$$

2147: from which we can deduce for instance $\bbb331= 19$ or

2148: $\bbb34{\ss1} = 15$ using Lemma¨\ref{L:Comput}.

2149: \end{exam}

2150: 

2151: Using Proposition¨\ref{P:Main}, we deduce

2152: 

2153: \begin{prop} \label{P:Values}

2154: With $M_\nn$ as in Lemma¨\ref{L:Comput}, we have

2155: for $\nn \ge \rr \ge 1$ and $\dd \ge 1$

2156: \begin{gather*}

2157: \ccc\nn\dd = ((1, 1, \dots, 1) \, M_\nn^\dd)_{\D_{\nn-1}}

2158: - 1,\\

2159: \hhh\rr\nn\dd = ((1, 1, \dots, 1) \,

2160: M_\nn^\dd)_{\D_{\rr}}.

2161: \end{gather*}

2162: \end{prop}

2163: 

2164: \begin{coro}

2165: $(i)$ For fixed¨$\nn, \rr$, the generating functions for

2166: the sequences $\ccc\nn\dd$ and $\hhh\rr\nn\dd$ are

2167: rational.

2168: 

2169: $(ii)$ For fixed¨$\nn, \rr$, the numbers $\ccc\nn\dd$ and

2170: $\hhh\rr\nn\dd$ admit expressions of the form

2171: \begin{equation} \label{E:General}

2172: P_1(\dd) \ev_1^\dd + \cdots +

2173: P_\ir(\dd) \ev_\ir^\dd.

2174: \end{equation} 

2175: where $\ev_1$, \dots, $\ev_\ir$ are the non-zero

2176: eigenvalues of¨$M_\nn$ and $P_1, \dots,

2177: P_\ir$ are polynomials with $\deg(P_\ind)$ at

2178: most the multiplicity of¨$\ev_\ind$ for¨$M_\nn$.

2179: \end{coro}

2180: 

2181: As the matrix¨$M_\nn$ is an $\nn! \times \nn!$ matrix,

2182: completing the computation is not so easy, even for small

2183: values of¨$\nn$. Actually, it is shown in¨\cite{Dhi} how

2184: to replace¨$M_\nn$ with a smaller matrix¨$\overline

2185: M_\nn$ of size $p(\nn) \times p(\nn)$, where $p(\nn)$ is

2186: the number of partitions of¨$\nn$. The property is

2187: connected with classical results by Solomon about the

2188: descents of permutations¨\cite{Sol}. With such methods,

2189: one easily obtains the values listed in

2190: Table¨\ref{T:Values}.

2191: 

2192: \begin{table}[htb]

2193: $$\begin{tabular}{c|r|r|r|r|r|r|r}

2194: \quad$\dd$

2195: & 0 

2196: &1&2&3&4&5&6\\

2197: \hline

2198: \vrule width0pt height12pt depth6pt

2199: $\hhh12\dd$

2200: & 1

2201: & 2 & 3 & 4 & 5 & 6 & 7 \\

2202: \hline

2203: \vrule width0pt height12pt 

2204: $\hhh13\dd$

2205: & 1

2206: & 6 & 19 & 48 & 109 & 234 & 487\\

2207: \vrule width0pt depth6pt

2208: $\hhh23\dd$

2209: & 1

2210: & 3 & 7 & 15 & 31 & 63 & 127\\

2211: \hline

2212: \vrule width0pt height12pt 

2213: $\hhh14\dd$

2214: & 1

2215: &24&211&1\,380&8\,077&45\,252&249\,223

2216: \\

2217: $\hhh24\dd$

2218: & 1

2219: &12&83&492&2\,765&15\,240

2220: & 83\,399 \\

2221: \vrule width0pt depth6pt

2222: $\hhh34\dd$

2223: & 1

2224: &4&15&64&309&1\,600 

2225: & 8\,547

2226: \\

2227: \hline

2228: \vrule width0pt height12pt 

2229: $\hhh15\dd$

2230: & 1

2231:  & 120 & 3\,651 & 79\,140 

2232: & 1\,548\,701 

2233: & 29\,375\,460 

2234: & 551\,997\,751 

2235: \\

2236: $\hhh25\dd$

2237: & 1

2238:  & 60 & 1\,501 & 30\,540 

2239: & 585\,811

2240: & 11\,044\,080 

2241: & 207\,154\,921 

2242: \\

2243: $\hhh35\dd$

2244: & 1

2245:  & 20 & 311 & 5\,260 

2246: & 94\,881

2247: & 1\,755\,360 

2248: & 32\,741\,851 

2249: \\

2250: \vrule width0pt depth6pt

2251: $\hhh45\dd$

2252: & 1

2253:  & 5 & 31 & 325 

2254: & 4\,931 

2255: & 86\,565 

2256: & 1\,590\,231 

2257: \\

2258: \hline

2259: \vrule width0pt height12pt 

2260: $\hhh16\dd$

2261: & 1

2262: & 720 

2263: &\!90\,921 

2264: &\!7\,952\,040 

2265: &\!634\,472\,921

2266: &\!49\,477\,263\,360

2267: &\!3\,836\,712\,177\,121

2268: \\

2269: $\hhh26\dd$

2270: & 1

2271: & 360 

2272: &\!38\,559 

2273: &\!3\,228\,300 

2274: &\!254\,718\,389

2275: &\!19\,808\,530\,620

2276: &\!1\,535\,016\,069\,499 \\

2277: $\hhh36\dd$

2278: & 1

2279:  & 120 & 8\,727 & 649\,260 

2280: & 49\,654\,757

2281: & 3\,831\,626\,580 

2282: & 296\,361\,570\,667\\

2283: $\hhh46\dd$

2284: & 1

2285:  & 30 & 1\,075 & 61\,620 

2286: & 4\,387\,195

2287: & 332\,578\,230 

2288: & 25\,612\,893\,355\\

2289: $\hhh56\dd$

2290: & 1

2291:  & 6 & 63 & 1\,956

2292: & 116\,423

2293: & 8\,448\,606 

2294: & 643\,888\,543

2295: \end{tabular}$$

2296: \bigskip

2297: \caption{\smaller First values of $\hhh\rr\nn\dd$

2298: for $1 \le \rr < \nn$---the value is¨$1$ for $\rr

2299: \ge \nn$. For instance, we read that the number of

2300: $3$-strand braids of degree at most¨$2$, \ie,

2301: $\hhh132$, is¨$19$---as was seen in

2302: Example¨\ref{X:Height}---while the maximal number

2303: of¨$\ss3$'s in a $\s$-positive word drawn in¨$\GG44$,

2304: \ie, $\ccc44$, which is $\hhh344-1$

2305: according to Proposition¨\ref{P:MainHeight}, is¨$308$.}

2306: \label{T:Values}

2307: \end{table}

2308: 

2309: Using the reduced matrices 

2310: $\overline M_3 = \left(\begin{matrix}1&0&0\\4&2&0\\1&1&1

2311: \end{matrix}\right)$ and 

2312: $\overline M_4 =

2313: \left(\begin{matrix}1&0&0&0&0\\11&4&1&0&0\\5&3&2&1&0\\

2314: 6&4&2&2&0\\1&1&1&1&1

2315: \end{matrix}\right)$, we obtain the

2316: following explicit form for¨\eqref{E:General} involving

2317: the non-zero eigenvalues of¨$M_3$, namely $1$ (double),

2318: $2$ and of¨$M_4$, namely $1$ (double), $2$, and $3 \pm

2319: \sqrt6$:

2320: 

2321: \begin{prop} \label{P:Values34}

2322: Let $\ev_1 = 3 + \sqrt{6}$ and $\ev_2 = 3 - \sqrt{6}$.

2323: Then, for $\dd \ge 1$, we have

2324: \begin{align*}

2325: \hhh13\dd &= 8\cdot 2^\dd - 3 \dd - 7,\\

2326: \hhh23\dd &= \ccc3\dd+1 = 2 \cdot 2^\dd - 1,\\

2327: \hhh14\dd &= 

2328: \frac3{20}(32 + 13\sqrt6) \ev_1^\dd  

2329: + \frac3{20}(32 - 13\sqrt6) \ev_2^\dd 

2330: - \frac{128}5 \cdot\, 2^\dd

2331: +  6\dd + 17,\\

2332: \hhh24\dd &= 

2333: \frac1{20}(32 + 13\sqrt6) \ev_1^\dd 

2334: + \frac1{20}(32 - 13\sqrt6) \ev_2^\dd

2335: - \frac{16}5 \cdot\, 2^\dd 

2336: + 1,\\

2337: \hhh34\dd &= \ccc34 +1 = 

2338: \frac1{20}(4 + \sqrt6) \ev_1^\dd

2339: + \frac1{20}(4 - \sqrt6) \ev_2^\dd 

2340: + \frac85 \cdot\, 2^\dd - 1.

2341: \end{align*}

2342: \end{prop}

2343: 

2344: The main interest of the above formulas are to show that

2345: each of the involved parameters has an exponential growth

2346: with respect to¨$\dd$, in $O(2^\dd)$ for $\nn = 3$, and in

2347: $O((3+\sqrt{6})^\dd)$ for $\nn = 4$. For practical

2348: purposes, it may be more convenient to resort to inductive

2349: formulas, for instance

2350: \begin{gather}

2351: \label{E:Recur3}

2352: \hhh13\dd = 2 \hhh13{\dd-1} + 3 \dd + 1,\\

2353: \label{E:Recur4}

2354: \hhh14\dd = 6\hhh14{\dd-1} - 3\hhh14{\dd-2}

2355: + 32 \cdot 2^\dd - 12 \dd - 34,

2356: \end{gather}

2357: together with initial values $\hhh130 = \hhh140 = 1$,

2358: $\hhh141 = 24$ (or $\hhh14{-1}= 0$).

2359: 

2360: \subsection{Small values of¨$\dd$}

2361: 

2362: Another approach is to keep¨$\dd$ fixed and let¨$\nn$ vary.

2363: Once again, we only mention a few results, and refer the

2364: reader to¨\cite{Dhi} for the proofs and additional

2365: comments. For $\dd = 1$, it is easy to determine all

2366: values:

2367: 

2368: \begin{prop} [\cite{Dhi}]

2369: For $\nn \ge \rr \ge 1$, we have

2370: \begin{equation*} \label{E:K1}

2371: \hhh\rr\nn{} = \frac{\nn!}{\rr!}.

2372: \end{equation*}

2373: \end{prop}

2374: 

2375: For $\dd = 2$, it is easier to complete the

2376: computation for $\nn - \rr$ rather than¨$\rr$ fixed. 

2377: 

2378: \begin{prop} [\cite{Dhi}] \label{P:K2}

2379: For $\nn \ge \rr \ge 1$, we have

2380: \begin{align*} 

2381: \hhh{\nn-\rr}\nn2 

2382: &= \rr!\,(\rr+1)^\nn + \sum_{i = 1}^\rr P_i(\nn) \,

2383: i^{\nn-\rr+i-1},\\

2384: \intertext{for some polynomial $P_i$ of degree at

2385: most¨$\rr - i + 1$. The values for $\rr =

2386: 1, 2$ are}

2387: \hhh{\nn-1}\nn2 &= 2^\nn - 1, \\

2388: \hhh{\nn-2}\nn2 &= 2 \cdot 3^\nn - (\nn + 6) \cdot

2389: 2^{\nn-1} + 1.

2390: \end{align*}

2391: \end{prop}

2392: 

2393: For $\rr$ fixed, no general formula is known. Let us

2394: mention the case of¨$\hhh1\nn2$, which follows from

2395: results of¨\cite{CSV}:

2396: 

2397: \begin{prop} [\cite{Dhi}]

2398: The numbers¨$\hhh1\nn2$ are determined by the

2399: induction

2400: \begin{equation*}

2401: \hhh102=1, \qquad 

2402: \hhh1\nn2 = \sum_{\ind=0}^{\nn-1} (-1)^{\nn + \ind

2403: +1} {\nn \choose \ind}^2 

2404: \hhh1\ind2.

2405: \end{equation*}

2406: Their double exponential generating function is, with 

2407: $J_0(\xx)$ is the Bessel function,

2408: \begin{equation*}

2409: \sum_{\nn = 0}^\infty \hhh1\nn2

2410: \frac{z^\nn}{\nn!^2} = 

2411: \bigg( \sum_{\nn = 0}^\infty (-1)^\nn

2412: \frac{z^\nn}{\nn!^2}

2413: \bigg)\inv = \frac1{J_0(\sqrt{z})}.

2414: \end{equation*}

2415: \end{prop}

2416: 

2417: Finally, for $\dd = 3$, the computation can be completed

2418: at least in the case $\nn - \rr = 1$:  

2419: 

2420: \begin{prop} [\cite{Dhi}]

2421: For¨$\nn \ge 1$, we have, with $e = \exp(1)$,

2422: \begin{equation*} \label{E:KKK3}

2423: \hhh{\nn-1}\nn3 = \sum_{\ind=0}^{\nn-1}

2424: \frac{\nn!}{\ind!} = \lfloor\nn! e\rfloor - 1.

2425: \end{equation*}

2426: \end{prop}

2427: 

2428: Using Proposition¨\ref{P:MainHeight}, we deduce the

2429: following explicit values for $\ccc\nn\dd$, \ie, for the

2430: maximal number of occurrences of¨$\ss{\nn-1}$ is a

2431: $\s$-positive word drawn in the Cayley graph

2432: of¨$\DD\nn\dd$:

2433: $$\ccc\nn{} = \nn - 1, \qquad

2434: \ccc\nn2 = 2^\nn - 2, \qquad

2435: \ccc\nn3 = \sum_{\ind=0}^{\nn-1}

2436: \frac{\nn!}{\ind!} - 1 = \lfloor\nn! e\rfloor - 2.

2437: $$

2438: The formulas listed above show that a number

2439: of different induction schemes appear, suggesting that the

2440: combinatorics of normal sequences of braids is very rich. 

2441: 

2442: \section{A complete description of $\DDDD3\dd$}

2443: \label{S:N3}

2444: 

2445: Our ultimate goal would be a complete description

2446: of each chain $\DDDD\nn\dd$, this typically meaning

2447: that we are able to explicitly specify the increasing

2448: enumeration of its elements. This goal remains out of

2449: reach in the general case, but we shall show now how the

2450: process can be completed in the case $\nn = 3$. The

2451: counting formulas of Section¨\ref{S:Deter} play a key

2452: role in the construction, and, in particular, the Pascal

2453: triangle of Table¨\ref{T:Triangle} below is directly

2454: connected with the $2^\dd$¨factor in the inductive

2455: formulas of Proposition¨\ref{P:Values34}.  As an

2456: application, we deduce a new proof of Property¨C and of

2457: the well-ordering property, hence a complete

2458: re-construction of the braid ordering in the case $\nn

2459: = 3$.

2460: 

2461: The general principle is to make the decomposition of

2462: Corollary¨\ref{C:Structure} explicit. The latter shows

2463: that, for all¨$\nn$ and¨$\dd$, the chain $\DDDD\nn\dd$

2464: can be decomposed into $\ccc\nn\dd$¨subintervals

2465: each of which is a copy of some fragment

2466: of¨$\DDDD{\nn-1}\dd$. Moreover, the approach of

2467: Section¨\ref{S:Deter} suggests an induction on¨$\dd$

2468: as well, so, finally, we are led to looking for a

2469: description of¨$\DDDD\nn\dd$ in terms

2470: of¨$\DDDD{\nn-1}\dd$ and¨$\DDDD\nn{\dd-1}$---\ie, in

2471: the current case, a description of $\DDDD3\dd$ in terms

2472: of

2473: $\DDDD2\dd$ and $\DDDD3{\dd-1}$.

2474: 

2475: \subsection{The braids¨$\Dd\nn\pp$}

2476: 

2477: The subsequent construction will appeal to a double

2478: series of braid called¨$\Dd\nn\pp$, and we begin with a

2479: few preliminary properties of these braids.

2480: 

2481: \begin{defi}

2482: For $\nn \ge 2$, let $\ss{\nn, 1}$ and $\ss{1, \nn}$

2483: respectively denote the braid words

2484: $\ss{\nn-1} \ss{\nn-2} \dots \ss1$ and $\ss1 \ss2

2485: \dots \ss{\nn-1}$. For $\pp \ge 0$, we define 

2486: $\Ddt\nn\pp$ to  be (the braid represented by) the

2487: length¨$\pp$ prefix of the right infinite

2488: word¨$(\ss{\nn, 1}\ss{1, \nn})^\infty$, and $\Dd\nn\pp$ to

2489: be (the braid represented by) the length¨$\pp$ suffix of

2490: the left infinite word¨$^\infty(\ss{\nn, 1}\ss{1, \nn})$. 

2491: \end{defi}

2492: 

2493: For instance, we find

2494: $\Dd30 = 1$, $\Dd31 = \b$, $\Dd32 = \a\b$, \dots,

2495: $\Dd34 = \b\a\a\b$, \dots, $\Dd37 = \a\a\b\b\a\a\b$, \etc\

2496: Similarly, we have $\Dd46 = \c\b\a\a\b\c$, and, more

2497: generally, $\Dd\nn{2\nn-2} = \Ddt\nn{2\nn-2}= \ss{\nn,

2498: 1}\ss{1, \nn}$. Note that, as a word,

2499: $\Dd\nn\pp$ is obtained by reversing the order of the

2500: letters in¨$\Ddt\nn\pp$.

2501: 

2502: \begin{lemm}

2503: For $\nn \ge 2$ and $\pp, \qq \ge 0$ satisfying $\pp

2504: + \qq = \dd(\nn-1)$, we have

2505: \begin{equation} \label{E:Deltadelta}

2506: \Dd\nn\pp \, \DD{\nn-1}\dd \, \Ddt\nn\qq = \DD\nn\dd.

2507: \end{equation}

2508: \end{lemm}

2509: 

2510: \begin{proof}

2511: We first prove using induction on¨$\dd$ the relation

2512: \begin{equation} \label{E:Deltak0}

2513: \Dd\nn{\dd(\nn-1)} \, \DD{\nn-1}\dd = \DD\nn\dd,

2514: \end{equation}

2515: \ie, \eqref{E:Deltadelta} with $\qq =

2516: 0$. For $\dd = 0$, \eqref{E:Deltak0} reduces to $1 =

2517: 1$. Assume $\dd \ge 1$.  By definition,

2518: $\Dd\nn{\dd(\nn-1)}$ is $\ss{\nn,1} \, 

2519: \Dd\nn{(\dd-1)(\nn-1)}$ for

2520: $\dd$¨odd, and is $\ss{1, \nn} \,

2521: \Dd\nn{(\dd-1)(\nn-1)}$ for

2522: $\dd$¨even, so, in any case, we can write

2523: $$\Dd\nn{\dd(\nn-1)} = \flip\nn^{\dd-1}(\ss{1, \nn}) \,

2524: \Dd\nn{(\dd-1)(\nn-1)},$$ 

2525: where we recall $\flip\nn$ denotes the flip

2526: automorphism of¨$B_\nn$ that exchanges¨$\ss i$ and

2527: $\ss{\nn - i}$. Using the induction hypothesis

2528: and¨\eqref{E:Delta}, we find

2529: \begin{align*}

2530: \Dd\nn{\dd(\nn-1)} \, \DD{\nn-1}\dd 

2531: &= \flip\nn^{\dd-1}(\ss{1, \nn}) \, \Dd\nn{(\dd-1)(\nn-1)} \,

2532: \DD{\nn-1}{\dd-1} \, \D_{\nn-1}\\

2533: &= \flip\nn^{\dd-1}(\ss{1, \nn}) \, \DD n{\dd-1} \, \D_{\nn-1}

2534: = \DD n{\dd-1} \, \ss{1, \nn} \, \D_{\nn-1}

2535: = \DD n{\dd-1} \, \D_\nn 

2536: = \DD\nn\dd.

2537: \end{align*} 

2538: 

2539: We return to the general case

2540: of¨\eqref{E:Deltadelta}. For $\dd$ even, we have

2541: $\Dd\nn{\dd(\nn-1)} = \Ddt n{\dd(\nn-1)}$, hence

2542: $\Ddt\nn\qq \, \Dd\nn\pp = \Dd\nn{\dd(\nn-1)}$. If

2543: $\dd$ is odd, we have $\Dd\nn{\dd(\nn-1)} =

2544: \flip\nn(\Ddt n{\dd(\nn-1)})$, which implies

2545: $\flip\nn(\Ddt\nn\qq) \, \Dd\nn\pp = \Dd\nn{\dd(\nn-1)}$. So $\flip\nn^\dd(\Ddt\nn\qq) \, \Dd\nn\pp = \Dd\nn{\dd(\nn-1)}$ holds in both

2546: cases. Now, using¨\eqref{E:Deltak0}, we find

2547: \begin{align*}

2548: \flip\nn(\Ddt\nn\qq) \, \Dd\nn\pp \, \DD{\nn-1}\dd \, \Ddt\nn\qq

2549: &= \Dd\nn{\dd(\nn-1)} \, \DD{\nn-1}\dd \, \Ddt\nn\qq

2550: = \DD\nn\dd \, \Ddt\nn\qq

2551: = \flip\nn(\Ddt\nn\qq) \, \DD\nn\dd,

2552: \end{align*} 

2553: from which we deduce¨\eqref{E:Deltadelta} by cancelling 

2554: $\flip\nn(\Ddt\nn\qq)$ on the left.

2555: \end{proof}

2556: 

2557: \begin{lemm}

2558: For $1 \le i \le \nn-2$ we have

2559: \begin{equation}

2560: \label{E:Commut}

2561: \Dd\nn{\dd(\nn-1)} \, \ss i = \ss{i+e} \,

2562: \Dd\nn{\dd(\nn-1)}

2563: \end{equation}

2564: with $e = 0$ if $\dd$ is even, and $e = 1$ if $\dd$ is odd.

2565: \end{lemm}

2566: 

2567: \begin{proof}

2568: For $1 \le i \le \nn-2$, we have

2569: \begin{equation}

2570: \ss{1,n} \, \ss i = \ss{i+1} \, \ss{1,n},

2571: \mbox{\quad and \quad}

2572: \ss{\nn, 1} \, \ss{i+1} = \ss i \, \ss{\nn, 1}, 

2573: \end{equation}

2574: as an easy induction shows. This implies

2575: $\ss{\nn,1} \, \ss{1,n} \, \ss i = \ss i \,

2576: \ss{\nn,1} \, \ss{1,n}$, and therefore $(\ss{\nn,1}

2577: \, \ss{1,n})^\dd \, \ss i = \ss i \,

2578: (\ss{\nn,1} \, \ss{1,n})^\dd$, \ie, 

2579: $\Dd\nn{2\dd(\nn-1)} \, \ss i = \ss i \,

2580: \Dd\nn{2\dd(\nn-1)}$, for every¨$\dd$. On the other

2581: hand, we have 

2582: $\Dd\nn{(2\dd+1)(\nn-1)} = \ss{1,n} \, \Dd\nn{2\dd(\nn-1)}$, hence

2583: $$\Dd\nn{(2\dd+1)(\nn-1)} \, \ss i 

2584: = \ss{1,n} \, \ss i \, \Dd\nn{2\dd(\nn-1)} 

2585: = \ss{i+1} \, \ss{1,n} \, \Dd\nn{2\dd(\nn-1)} =

2586: \ss{i+1} \, \Dd\nn{(2\dd+1)(\nn-1)},$$

2587: as was expected.

2588: \end{proof}

2589: 

2590: \subsection{A Pascal triangle}

2591: 

2592: We shall now construct for every¨$\dd$ a sequence of

2593: positive braids¨$\SS3\dd$ that will turn out to be

2594: the increasing enumeration of¨$\DDDD3\dd$.

2595: The construction relies on an induction similar to

2596: a Pascal triangle. In order to make it easily

2597: understandable, it is convenient to start with a

2598: construction in the (trivial) cases¨$\nn =

2599: 1$ and¨$\nn = 2$.

2600: 

2601: As $B_1$ is the trivial group, then for every¨$\dd$

2602: there is exactly one element of degree at

2603: most¨$\dd$, namely¨$1$, and we can state:

2604: 

2605: \begin{prop}

2606: Let $\SS1\dd$ be defined for

2607: $\dd \ge 0$ by

2608: \begin{equation} \label{E:Def1}

2609: \SS1\dd = (1).

2610: \end{equation}

2611: Then $\SS1\dd$ is the increasing enumeration

2612: of¨$\DDD1\dd$.

2613: \end{prop}

2614: 

2615: The group¨$B_2$ is the rank¨$1$ free group

2616: generated by¨$\ss1$. The fundamental braid¨$\D_2$

2617: is just¨$\ss1$, and the braids of degree at

2618: most¨$\dd$, \ie, the divisors of¨$\DD2\dd$, consist

2619: of the $\dd+1$ braids¨$1, \ss1, \ldots, \sss1\dd$.

2620: On the other hand, we have $\ss{1, 2} = \ss{2, 1} =

2621: \ss1$, and $\Dd1i = \sss1i$ for every¨$i$. 

2622: 

2623: \begin{nota}

2624: If $S_1, S_2$ are sequences (of braids), we denote

2625: by $S_1 + S_2$ the concatenation of¨$S_1$ and¨$S_2$,

2626: \ie, the sequence obtained by appending¨$S_2$ after¨$S_1$.

2627: If $S$ is a sequence of braids, and $\xx$ is a

2628: braid, we denote by¨$\xx S$ the translated sequence

2629: obtained by multiplying each entry in¨$S$ by¨$\xx$

2630: on the left. 

2631: \end{nota}

2632: 

2633: With these conventions, the sequence $(1, \ss1,

2634: \dots, \ss1^\dd)$ can be expressed as $\Dd20(1)

2635: + \Dd21(1) + \cdots + \Dd2\dd(1)$, and we can

2636: state:

2637: 

2638: \begin{prop} \label{P:Enum3}

2639: Let $\SS2\dd$ be defined for $\dd \ge 0$ by

2640: \begin{equation} \label{E:Def2}

2641: \SS2\dd = \Dd20\SS1\dd + \Dd21\SS1\dd + \cdots + \Dd2\dd\SS1\dd.

2642: \end{equation}

2643: Then $\SS2\dd$ is the increasing enumeration

2644: of¨$\DDD2\dd$.

2645: \end{prop}

2646: 

2647: We repeat the process for¨$\nn = 3$, introducing

2648: a sequence¨$\SS3\dd$ by a definition similar

2649: to¨\eqref{E:Def2} that involves $\SS2\dd$

2650: and¨$\SS3{\dd-1}$. The result we shall prove is:

2651: 

2652: \begin{prop} \label{P:Main3}

2653: Let $\SS3\dd$ be defined for $\dd \ge 0$ by

2654: \begin{align} \label{E:Def3}

2655: \SS3\dd =  \Dd30\SS2 \dd + \SS3{\dd, 1}

2656: + \Dd31\SS2 \dd + \cdots 

2657: + \Dd3{2\dd-1}\SS2 \dd + \SS3{\dd, 2\dd} + \Dd3{2\dd} \SS2 \dd,

2658: \end{align}

2659: where $\SS3{\dd, 1}, \cdots, \SS3{\dd, 2\dd}$ are 

2660: defined by $\SS3{\dd, 1} =

2661: \SS3{\dd, 2\dd} =

2662: \emptyset$ and, for $2 \le \ppp \le 2\dd-1$, 

2663: \begin{equation*}

2664: \SS3{\dd, \ppp} = 

2665: \begin{cases}

2666: \phantom{\ss2}\ss1(\SS3{\dd-1, \ppp-1} +

2667: \Dd3{\ppp-1}\SS2{\dd-1} +

2668: \SS3{\dd-1, \ppp}) & \mbox{for $\ppp = 0 \pmod 4$,}\\

2669: \ss2\ss1(\SS3{\dd-1, \ppp-2} + \Dd3{\ppp-1}\SS2{\dd-1} +

2670: \SS3{\dd-1, \ppp-1}) & \mbox{for $\ppp = 1 \pmod 4$,}\\

2671: \phantom{\ss2}\ss2(\SS3{\dd-1, \ppp-1} +

2672: \Dd3{\ppp-1}\SS2{\dd-1} +

2673: \SS3{\dd-1, \ppp}) & \mbox{for $\ppp = 2 \pmod 4$,}\\

2674: \ss1\ss2(\SS3{\dd-1, \ppp-2} + \Dd3{\ppp-1}\SS2{\dd-1} +

2675: \SS3{\dd-1, \ppp-1}) & \mbox{for $\ppp = 3 \pmod 4$.}

2676: \end{cases}

2677: \end{equation*}

2678: Then $\SS3\dd$ is the increasing enumeration

2679: of¨$\DDD3\dd$.

2680: \end{prop}

2681: 

2682: The general scheme is illustrated in 

2683: Table¨\ref{T:Triangle}: the sequence¨$\SS3\dd$ is

2684: constructed by starting with $2\dd+1$¨copies

2685: of¨$\SS2\dd$ translated by¨$\Dd30$, \dots,

2686: $\Dd3{2\dd}$ and inserting (translated copies

2687: of) fragments of the previous sequence¨$\SS3{\dd-1}$.  

2688: 

2689: \begin{table}[t]

2690: \begin{picture}(130,47)(1, 0)

2691: \put(0,10){$\Dd30 \SS23$}

2692: \put(10,10){$(\SS3{3,1})$}

2693: \put(20,10){$\Dd31 \SS23$}

2694: \put(32,10){$\SS3{3,2}$}

2695: \put(40,10){$\Dd32 \SS23$}

2696: \put(52,10){$\SS3{3,3}$}

2697: \put(60,10){$\Dd33 \SS23$}

2698: \put(72,10){$\SS3{3,4}$}

2699: \put(80,10){$\Dd34 \SS23$}

2700: \put(92,10){$\SS3{3,5}$}

2701: \put(100,10){$\Dd34 \SS23$}

2702: \put(110,10){$(\SS3{3,6})$}

2703: \put(120,10){$\Dd36 \SS23$}

2704: \put(10,9){$\underbrace{\hbox to 28mm{\hfill}}$}

2705: \put(14,4){$\ss2\cdot$}

2706: \put(29,4){$\ss1\ss2\cdot$}

2707: \put(17.5,3){$\swarrow$}

2708: \put(26.5,3){$\searrow$}

2709: \put(52,9){$\underbrace{\hbox to 26mm{\hfill}}$}

2710: \put(56,4){$\ss1\cdot$}

2711: \put(70,4){$\ss2\ss1\cdot$}

2712: \put(59,3){$\swarrow$}

2713: \put(68,3){$\searrow$}

2714: \put(92,9){$\underbrace{\hbox to 27mm{\hfill}}$}

2715: \put(96,4){$\ss2\cdot$}

2716: \put(111,4){$\ss1\ss2\cdot$}

2717: \put(99,3){$\swarrow$}

2718: \put(108,3){$\searrow$}

2719: \put(14,0){$\cdots$}

2720: \put(31,0){$\cdots$}

2721: \put(54,0){$\cdots$}

2722: \put(71,0){$\cdots$}

2723: \put(95,0){$\cdots$}

2724: \put(112,0){$\cdots$}

2725: 

2726: \put(20,22){$\Dd30 \SS22$}

2727: \put(30,22){$(\SS3{2,1})$}

2728: \put(40,22){$\Dd31 \SS22$}

2729: \put(52,22){$\SS3{2,2}$}

2730: \put(60,22){$\Dd32 \SS22$}

2731: \put(72,22){$\SS3{2,3}$}

2732: \put(80,22){$\Dd33 \SS22$}

2733: \put(90,22){$(\SS3{2,4})$}

2734: \put(100,22){$\Dd34 \SS22$}

2735: \put(30,21){$\underbrace{\hbox to

2736: 28mm{\hfill}}$}

2737: \put(36,16){$\ss2\cdot$}

2738: \put(50,16){$\ss1\ss2\cdot$}

2739: \put(38,14.5){$\swarrow$}

2740: \put(47,14.5){$\searrow$}

2741: \put(72,21){$\underbrace{\hbox to

2742: 27mm{\hfill}}$}

2743: \put(77,16){$\ss1\cdot$}

2744: \put(91,16){$\ss2\ss1\cdot$}

2745: \put(79,14.5){$\swarrow$}

2746: \put(88,14.5){$\searrow$}

2747: 

2748: \put(40,34){$\Dd30 \SS21$}

2749: \put(50,34){$(\SS3{1,1})$}

2750: \put(60,34){$\Dd31 \SS21$}

2751: \put(70,34){$(\SS3{1,2})$}

2752: \put(80,34){$\Dd32 \SS21$}

2753: \put(50,33){$\underbrace{\hbox to

2754: 28mm{\hfill}}$}

2755: \put(56,28){$\ss2\cdot$}

2756: \put(70,28){$\ss1\ss2\cdot$}

2757: \put(58,26.5){$\swarrow$}

2758: \put(67,26.5){$\searrow$}

2759: 

2760: \put(60,44){$\Dd30 \SS20$}

2761: \end{picture}

2762: \smallskip

2763: \caption{\smaller The inductive construction

2764: of¨$\SS3\dd$ as a Pascal triangle: the subsequence¨$

2765: \SS3{\dd,\ppp}$ is obtained by (translating and)

2766: concatenating the previous subsequences¨$

2767: \SS3{\dd-1,\ppp-1}$ and¨$

2768: \SS3{\dd-1,\ppp}$, or¨$\SS3{\dd-1,\ppp-2}$ and¨$

2769: \SS3{\dd-1,\ppp-1}$, depending on the parity of¨$\ppp$;

2770: the parenthesized sequences are empty; if we forget about

2771: the subsequences¨$\Dd3\qq\SS2\dd$, we have the

2772: Pascal triangle.}

2773: \label{T:Triangle}

2774: \end{table}

2775: 

2776: \begin{exam} \label{X:SS3}

2777: The difference between the definition of¨$\SS3\dd$

2778: in¨\eqref{E:Def3} and that of¨$\SS2\dd$ in¨\eqref{E:Def2} is

2779: the insertion of the additional factors $ \SS3{\dd,\ppp}$

2780: between the consecutive terms¨$\Dd3\qq\SS2\dd$. Because

2781: $\SS3{\dd, 1}$ and $\SS3{\dd, 2\dd}$ are empty, the difference

2782: occurs for $\dd \ge 2$ only. The first values are:

2783: 

2784: $\SS30 = \Dd30 \SS20 = (1)$,

2785: 

2786: $\SS31 = \Dd30 \SS21 + \SS3{1,1} + \Dd31 \SS21 +

2787: \SS3{1,2} + \Dd32 \SS21$

2788: 

2789: \hspace{1cm} $= (1, \a) + \emptyset

2790: + \b(1, \a) + \emptyset 

2791: + \a\b(1, \a)$

2792: 

2793: \hspace{1cm} $= (1, \a, \b, \b\a, \a\b, \a\b\a)$,

2794: 

2795: $\SS32  = \Dd30 \SS22 + \SS3{2,1} +

2796: \Dd31 \SS22 + \SS3{2,2} +

2797: \Dd32 \SS22 + \SS3{2,3} +

2798: \Dd33 \SS22 + \SS3{2,4} +

2799: \Dd34 \SS22$

2800: 

2801: \hspace{1cm} $ = (1, \a, \a\a) + \emptyset

2802: + \b(1, \a, \a\a) + \b(\b, \b\a)

2803: + \a\b(1, \a, \a\a)$

2804: 

2805: \hspace{4cm} $ + \a\b(\b, \b\a)

2806: + \a\a\b(1, \a, \a\a) + \emptyset 

2807: + \b\a\a\b(1, \a, \a\a)$

2808: 

2809: \hspace{1cm} $ = (1, \a, \a\a, \b, \b\a, \b\a\a,

2810: \b\b, \b\b\a,

2811: \a\b, \a\b\a, \a\b\a\a,

2812: \a\b\b, \a\b\b\a, \a\a\b$, 

2813: 

2814:  \hspace{6cm} $\a\a\b\a, \a\a\b\a\a, \b\a\a\b,

2815: \b\a\a\b\a, \b\a\a\b\a\a)$.\\

2816: It is easy to check directly that the sequence

2817: $\SS3\dd$ provides the increasing enumeration

2818: of¨$\DDD3\dd$ for $\dd = 0, 1, 2$. 

2819: \end{exam}

2820: 

2821: The proof of Proposition¨\ref{P:Main3} will be

2822: split into several pieces, each of which is

2823: established using an induction on the degree¨$\dd$.

2824: 

2825: \begin{lemm} \label{L:Inclusion3}

2826: All entries in¨$\SS3\dd$ are divisors of¨$\DD3\dd$.

2827: \end{lemm}

2828: 

2829: \begin{proof}

2830: The result is true for $\dd = 0$. Assume $\dd \ge 1$. By

2831: construction, each entry in¨$\SS3\dd$ either is of the form

2832: $\Dd3\qq \ss1^\expo$ with $0 \le \qq \le 2\dd$ and $0 \le

2833: \expo \le \dd$, or belongs to some

2834: subsequence¨$\SS3{\dd,\ppp}$ with $2 \le \ppp \le 2\dd -

2835: 1$. In the first case,

2836: $\Dd3\qq\ss1^\expo$ is a right divisor of $\Dd3{2\dd}

2837: \ss1^\expo$, which itself is a left divisor of

2838: $\Dd3{2\dd} \ss1^\dd$. By¨\eqref{E:Deltadelta}, the latter

2839: is¨$\DD3\dd$. Hence each $\Dd3\qq

2840: \ss1^\expo$ is a divisor of¨$\DD3\dd$. As for the

2841: entries coming from some subsequence¨$\SS3{\dd,\ppp}$,

2842: by definition they are of the form¨$\xx\yy$ with $\xx$

2843: one of $\ss2, \ss1\ss2, \ss1,

2844: \ss2\ss1$ and $\yy$ an entry in¨$\SS3{\dd-1}$. Then

2845: $\xx$ is a divisor of¨$\D_3$, while, by induction

2846: hypothesis, $\yy$ is a divisor of¨$\DD3{\dd-1}$, so

2847: $\xx\yy$ is a divisor of¨$\DD3\dd$.

2848: \end{proof}

2849: 

2850: \begin{lemm} \label{L:Card3}

2851: The length of the sequence¨$\SS3\dd$ equals the

2852: cardinality of¨$\DDD3\dd$.

2853: \end{lemm}

2854: 

2855: \begin{proof}

2856: Let $\ell_\dd$ denote the length of¨$\SS3\dd$.

2857: Computing¨$\ell_\dd$ is not very difficult. However,

2858: there is need to do it, which amounts to solve a

2859: recursive formula unnecessarily. Indeed, we saw in

2860: Section¨\ref{S:Deter} that the cardinality¨$\hhh13\dd$

2861: of¨$\DDD3\dd$ obeys the inductive rule¨\eqref{E:Recur3}.

2862: So it will be enough to check that

2863: $\ell_\dd$ satisfies the relation

2864: \begin{equation} \label{E:Double}

2865: \ell_\dd = 2 \ell_{\dd-1} + 3\dd + 1,

2866: \end{equation}

2867: and starts from the initial¨$\ell_1 = 6$ (or $\ell_0 =

2868: 1$). The latter point was checked in Example¨\ref{X:SS3}.

2869: 

2870: Now Table¨\ref{T:Triangle} shows that most entries

2871: in¨$\SS3{\dd-1}$ give rise to two entries in¨$\SS3\dd$.

2872: More precisely, each entry of¨$\SS3{\dd-1}$ not

2873: belonging to a factor of the form¨$\Dd3{2\qq}

2874: \SS2{\dd-1}$ gives rise to two entries in¨$\SS3\dd$,

2875: and, conversely, each entry in¨$\SS3\dd$ not belonging

2876: to a factor¨$\Dd3{\qq}

2877: \SS2{\dd}$ comes from such an entry in¨$\SS3{\dd-1}$.

2878: There are $\dd$¨factors¨$\Dd3{2\qq} \SS2{\dd-1}$

2879: in¨$\SS3{\dd-1}$, each of length¨$\dd$, and

2880: $2\dd+1$¨factors¨$\Dd3{2\qq} \SS2{\dd}$ in¨$\SS3\dd$,

2881: each of length¨$\dd+1$. So we obtain

2882: $$\ell_\dd - (2\dd+1)(\dd+1) = 2(\ell_{\dd-1} - \dd^2),$$

2883: which gives¨\eqref{E:Double}.

2884: \end{proof}

2885: 

2886: At this point, we cannot (yet) conclude that each

2887: divisor of¨$\DD3\dd$ occurs exactly once in¨$\SS3\dd$,

2888: as there could be some repetitions.

2889: 

2890: \subsection{A quotient-sequence for¨$\SS3\dd$}

2891: 

2892: Our next aim is to show that $\SS3\dd$ is

2893: $\sm$-increasing. To this end, we shall explicitly

2894: determine the quotient of adjacent entries

2895: in¨$\SS3\dd$, \ie, we shall specify a quotient-sequence

2896: for¨$\SS3\dd$ in the sense of Definition¨\ref{D:Witness}.

2897: 

2898: A preliminary step consists in determining the first and

2899: the last entries of the sequence¨$ \SS3{\dd,\ppp}$.

2900: For¨$S$ a nonempty sequence, we denote by $\first S$

2901: (\resp $\last S$) the first (\resp last) entry in¨$S$.

2902: 

2903: \begin{lemm} \label{L:FirstLast}

2904: For $1 < i < 2\dd$, we have

2905: \begin{equation}

2906: \first{ \SS3{\dd,\ppp}} = \Dd3{\ppp-1} \, \ss2, 

2907: \mbox{\quad and \quad}

2908: \last{ \SS3{\dd,\ppp}} \, \ss2 = \Dd3\ppp \, \sss1\dd. 

2909: \end{equation}

2910: \end{lemm}

2911: 

2912: \begin{proof}

2913: The result is vacuously true for $\dd = 0, 1$. Assume $\dd \ge

2914: 2$ with $\ppp  = 0 \pmod 4$. Using the definition, the

2915: induction hypothesis, and¨\eqref{E:Commut}, we find

2916: \begin{gather*}

2917: \first{ \SS3{\dd,\ppp}} 

2918: = \ss1 \, \first{ \SS3{\dd-1,\ppp-1}} = 

2919: \ss1 \, \Dd3{\ppp-2} \, \ss2

2920: = \Dd3{\ppp-1} \, \ss2, \\

2921: \last{ \SS3{\dd,\ppp}} \, \ss2 

2922: = \ss1 \, \last{ \SS3{\dd-1,\ppp}} \, \ss2 

2923: =  \ss1 \, \Dd3\ppp \, \sss1{\dd-1} 

2924: = \Dd3\ppp \, \sss1\dd.

2925: \end{gather*}

2926: Similarly, for $\ppp  = 1 \pmod 4$, we have

2927: \begin{gather*}

2928: \first{ \SS3{\dd,\ppp}} 

2929: = \ss2\ss1 \, \first{ \SS3{\dd-1,\ppp-2}} = 

2930: \ss2\ss1 \, \Dd3{\ppp-3} \, \ss2

2931: = \Dd3{\ppp-1} \, \ss2, \\

2932: \last{ \SS3{\dd,\ppp}} \, \ss2 

2933: = \ss2\ss1 \, \last{ \SS3{\dd-1,\ppp-1}} \, \ss2 

2934: =  \ss2\ss1 \, \Dd3{\ppp-1} \, \sss1{\dd-1} 

2935: = \ss2\, \Dd3{\ppp-1} \, \sss1\dd

2936: = \Dd3\ppp \, \sss1\dd.

2937: \end{gather*}

2938: Then, for $\ppp  = 2 \pmod 4$, we have

2939: \begin{gather*}

2940: \first{ \SS3{\dd,\ppp}} 

2941: = \ss2 \, \first{ \SS3{\dd-1,\ppp-1}} = 

2942: \ss2 \, \Dd3{\ppp-2} \, \ss2

2943: = \Dd3{\ppp-1} \, \ss2, \\

2944: \last{ \SS3{\dd,\ppp}} \, \ss2 

2945: = \ss2 \, \last{ \SS3{\dd-1,\ppp}} \, \ss2 

2946: =  \ss2 \, \Dd3\ppp \, \sss1{\dd-1} 

2947: = \Dd3\ppp \, \sss1\dd.

2948: \end{gather*}

2949: Finally, for $\ppp  = 3 \pmod 4$, we find

2950: \begin{gather*}

2951: \first{ \SS3{\dd,\ppp}} 

2952: = \ss1\ss2 \, \first{ \SS3{\dd-1,\ppp-2}} = 

2953: \ss1\ss2 \, \Dd3{\ppp-3} \, \ss2

2954: = \Dd3{\ppp-1} \, \ss2, \\

2955: \last{ \SS3{\dd,\ppp}} \, \ss2 

2956: = \ss1\ss2 \, \last{ \SS3{\dd-1,\ppp-1}} \, \ss2 

2957: =  \ss1\ss2 \, \Dd3{\ppp-1} \, \sss1{\dd-1} 

2958: =  \ss1\ss2\ss1\ss2 \, \Dd3{\ppp-3} \, \sss1{\dd-1}\\

2959: \hspace{4cm}

2960: =  \ss1\ss1\ss2\ss1 \, \Dd3{\ppp-3} \, \sss1{\dd-1} 

2961: =  \ss1\ss1\ss2 \, \Dd3{\ppp-3} \, \sss1\dd

2962: = \Dd3\ppp \, \sss1\dd.

2963: \end{gather*}

2964: This completes the argument.

2965: \end{proof}

2966: 

2967: We shall now construct an explicit quotient-sequence

2968: for¨$\SS3\dd$, \ie, a sequence of braid words

2969: representing the quotients of the consecutive entries

2970: of¨$\SS3\dd$. Before doing it for¨$\SS3\dd$, let us

2971: consider the (trivial) cases of¨$\SS1\dd$ and¨$\SS2\dd$.

2972: As $\SS1\dd$ consists of one single entry, it vacuously

2973: admits the empty sequence as a quotient-sequence. As

2974: for¨$\SS2\dd$, we can state:

2975: 

2976: \begin{lemm}

2977: For $\dd \ge 0$, let $\WW1\dd$ be the empty sequence, and

2978: let¨$\WW2\dd$ be defined by

2979: \begin{equation} \label{E:DDef2}

2980: \WW2\dd = \WW1\dd + (\ss1) + \WW1\dd + \cdots + \WW1\dd

2981: + (\ss1) + \WW1\dd,

2982: \end{equation}

2983: $\dd$¨times¨$(\ss1)$. Then $\WW2\dd$ is a

2984: quotient-sequence for¨$\SS2\dd$.

2985: \end{lemm}

2986: 

2987: On a similar way, we shall prove:

2988: 

2989: \begin{prop} \label{P:Quotient3}

2990: Let $\WW3\dd$ be the sequence defined by

2991: $\WW30 = \emptyset$ and 

2992: \begin{align} \label{E:DDef3}

2993: \WW3\dd =  \WW2\dd 

2994: \!+\! (\ss1^{-\dd}\ss2) 

2995: &+ \WW2\dd 

2996: \!+\! (\ss1^{-\dd}\ss2) 

2997: \!+\! \WW3{\dd, 2} \!+\!

2998: (\ss2\ss1^{-\dd}) \\ 

2999: \notag &+ \WW2\dd 

3000: \!+\! (\ss1^{-\dd}\ss2) 

3001: \!+\! \WW3{\dd, 3} 

3002: \!+\!

3003: (\ss2\ss1^{-\dd}) 

3004: \!+\! \cdots \\  

3005: \notag &+ \WW2\dd 

3006: \!+\! (\ss1^{-\dd}\ss2) 

3007: \!+\! \WW3{\dd, 2\dd-1} \!+\!

3008: (\ss2\ss1^{-\dd}) 

3009: \!+\! \WW2\dd 

3010: \!+\! (\ss2\ss1^{-\dd})

3011: \!+\! \WW2\dd,

3012: \end{align}

3013: \begin{tabbing} 

3014: with \= $\WW3{\dd, 2} = \WW3{\dd, 3} 

3015: = \WW2{\dd-1} 

3016: \!+\! (\ss2\sss1{-\dd+1}) 

3017: \!+\! \WW3{\dd-1, 2},$\\ 

3018: \> \vline width0pt height15pt

3019: $\WW3{\dd, 2\dd-2} =\WW3{\dd, 2\dd-1}

3020: = \WW3{\dd-1, 2\dd-3} 

3021: \!+\! (\sss1{-\dd+1}\ss2) 

3022: \!+\! \WW2{\dd-1}$,\\

3023: and

3024: \> \vline width0pt height15pt

3025: $\WW3{\dd, 2\pppp}  = \WW3{\dd, 2\pppp+1} 

3026: = \WW3{\dd-1,2\pppp-1} 

3027: \!+\! (\sss1{-\dd+1}\ss2) 

3028: \!+\! \WW2{\dd-1} 

3029: \!+\! (\ss2\sss1{-\dd+1}) 

3030: \!+\! \WW3{\dd-1, 2\pppp}$

3031: \end{tabbing}

3032: for $4 \le 2\pppp \le 2\dd-4$. Then

3033: $\WW3\dd$ is a quotient-sequence for¨$\SS3\dd$.

3034: \end{prop}

3035: 

3036: \begin{exam} \label{X:Quotient}

3037: We find

3038: $\WW31 

3039:  = \WW21 + (\A\b) + \WW21 + (\b\A) + \WW21

3040:  = (\a, \A\b, \a, \b\A, \a)$, and

3041: \begin{align*}

3042: \WW32 

3043: = \WW22 + (\A\A\b) 

3044: &+ \WW22 + (\A\A\b) 

3045: + \WW3{2, 2} + (\b\A\A)\\

3046: &+ \WW22 +

3047: (\A\A\b) + \WW3{2, 3} + (\b\A\A) + \WW22 +

3048: (\b\A\A) + \WW22

3049: \end{align*}

3050:  with $\WW3{2, 2} = \WW3{2, 3} = \WW21 = (\a)$,

3051: whence

3052: $$\WW32 

3053:  = (\a, \a, \A\A\b, \a, \a, \A\A\b, \a, \b\A\A, \a, \a,

3054: \A\A\b, \a, \b\A\A, \a, \a, \b\A\A, \a, \a).$$

3055: \end{exam}

3056: 

3057: \begin{proof} [Proof of Proposition¨\ref{P:Quotient3}]

3058: We prove using induction on¨$\dd$ that $\WW3\dd$ is a

3059: quotient-sequence for¨$\SS3\dd$ with the $4\dd-2$¨terms

3060: in¨\eqref{E:DDef3} corresponding to the $4\dd-1$¨nonempty

3061: terms in¨\eqref{E:Def3}---so, in particular, for $2

3062: \le \ppp \le 2\dd-1$, the subsequence¨$\WW3{\dd,\ppp}$ is

3063: a quotient-sequence for ¨$ \SS3{\dd,\ppp}$. The result is

3064: vacuously true for

3065: $\dd=0$.  

3066: Assume $\dd \ge 1$. By definition, the sequence¨$\SS3\dd$

3067: consists of the concatenation of the $2\dd+1$ sequences

3068: $\Dd30 \SS2 \dd, \cdots, \Dd3{2\dd} \SS2 \dd$, in which the $2\dd-2$

3069: sequences $\SS3{\dd, 2}, \dots, \SS3{\dd, 2\dd-1}$ are inserted.

3070: We shall consider these subsequences separately, and then

3071: consider the transitions between consecutive subsequences.

3072: 

3073: First, $\WW2\dd$ is a quotient-sequence for¨$\SS2\dd$,

3074: hence it is a quotient-sequence for every sequence

3075: $\Dd3\qq

3076: \SS2\dd$ as well, since, by definition, the quotients we

3077: consider are invariant under left translation. Then, by

3078: construction, each subsequence

3079: $\SS3{\dd, 2\pppp}$ or $\SS3{\dd, 2\pppp+1}$ appearing in¨$\SS3\dd$ is

3080: obtained by translating some subsequence¨$S$

3081: of¨$\SS3{\dd-1}$, namely

3082: $$S =  \SS3{\dd-1, 2\pppp-1} + \Dd3{\qq-1} \SS2{\dd-1} +

3083: \SS3{\dd-1, 2\pppp}.$$

3084: By induction hypothesis, the sequence

3085: $$\WW3{\dd-1, 2\pppp-1} +

3086: (\sss1{-\dd+1}\ss2) + \WW2{\dd-1} +

3087: (\ss2\sss1{-\dd+1}) +

3088: \WW3{\dd-1, 2\pppp},$$

3089: which is precisely $\WW3{\dd, 2\pppp}$ and $\WW3{\dd, 2\pppp+1}$

3090: by definition, is a quotient-sequence for¨$S$. The

3091: property remains true in the special cases $\pppp = 1$

3092: and $\pppp = \dd$, which correspond to removing the

3093: initial term¨$\SS3{\dd-1,2\pppp-1}$, and/or the final

3094: term¨$\SS3{\dd-1, 2\pppp}$, respectively. Then

3095: $\WW3{\dd, 2\pppp}$ and $\WW3{\dd, 2\pppp+1}$ are also

3096: quotient-sequences for any sequence obtained from¨$S$ by

3097: a left translation, in particular for¨$\SS3{\dd,

3098: 2\pppp}$ and¨$\SS3{\dd, 2\pppp+1}$.

3099: 

3100: So it only remains to study the transitions between the

3101: consecutive terms in the expression¨\eqref{E:Def3}

3102: of¨$\SS3\dd$, \ie, to compare the last entry in each term

3103: with the first entry in the next term. Four cases are to be

3104: considered, namely the special case of the first two terms

3105: and of the final two terms, and the generic cases of the

3106: transitions from¨$\Dd3\qq \SS2\dd$ to¨$\SS3{\dd,\ppp+1}$

3107: and from¨$\SS3{\dd,\ppp}$ to¨$\Dd3\qq \SS2\dd$.

3108: 

3109: As for the first two terms, namely $\Dd30 \SS2\dd$ and

3110: $\Dd31 \SS2\dd$, \ie, $\SS2\dd$ and $\ss2\SS2\dd$, the last

3111: entry in¨$\SS2\dd$ is¨$\sss1\dd$, while the first entry

3112: in¨$\ss2\SS2\dd$ is¨$\ss2$, so $\sss1{-\dd}\ss2$ is a quotient.

3113: As for the last two terms, namely $\Dd3{2\dd-1} \SS2\dd$ and

3114: $\Dd3{2\dd} \SS2\dd$, the last entry in¨$\Dd3{2\dd-1}\SS2\dd$

3115: is¨$\Dd3{2\dd-1}\,\sss1\dd$, while the first entry

3116: in¨$\Dd3{2\dd}\SS2\dd$ is¨$\Dd3{2\dd}$. Now,

3117: by¨\eqref{E:Deltadelta}, we have 

3118: $\Dd3{2\dd-1}\,\sss1\dd\ss2 =

3119: \Dd3{2\dd}

3120: \,\sss1\dd$, so $\ss2\sss1{-\dd}$ is an expression of the

3121: quotient.

3122: 

3123: Consider now the transition from¨$\Dd3\qq \SS2\dd$

3124: to¨$\SS3{\dd,\qq+1}$. The last entry in¨$\Dd3\qq \SS2\dd$

3125: is¨$\Dd3\qq\, \sss1\dd$, while, by Lemma¨\ref{L:FirstLast}, the

3126: first entry in¨$\SS3{\dd,\qq+1}$ is¨$\Dd3\qq \,\ss2$. Hence

3127: $\sss1{-\dd}\ss2$ represents the quotient. Finally,

3128: consider the transition from¨$\SS3{\dd,\ppp}$ to¨$\Dd3\qq

3129: \SS2\dd$. By Lemma¨\ref{L:FirstLast} again, the last

3130: entry¨$\xx$ in¨$\Dd3\qq \SS2\dd$ satisfies¨$\xx\, \ss2

3131: = \Dd3\qq \, \sss1\dd$, while the first entry in¨$\Dd3\qq

3132: \SS2\dd$ is¨$\Dd3\qq$. Hence $\ss2\sss1{-\dd}$

3133: represents the quotient.

3134: \end{proof}

3135: 

3136: \begin{coro} \label{C:Distinct3}

3137: For each¨$\dd$ the sequence¨$\SS3\dd$

3138: is $\sm$-increasing; so, in particular, it consists of

3139: pairwise distinct braids.

3140: \end{coro}

3141: 

3142: \begin{proof}

3143: By definition, every word in¨$\WW3\dd$ is

3144: $\s$-positive, hence, by Property¨A, it does not

3145: represent¨$1$.

3146: \end{proof}

3147: 

3148: As $\SS3\dd$ consists of pairwise distinct divisors

3149: of¨$\DD3\dd$, Lemma¨\ref{L:Card3} implies that every

3150: divisor of¨$\DD3\dd$ occurs exactly once in¨$\SS3\dd$.

3151: Then, as $\SS3\dd$ is $\sm$-increasing, it must be the

3152: increasing enumeration of¨$\DDD3\dd$, and the proof of

3153: Proposition¨\ref{P:Main3} is complete.

3154: 

3155: \begin{rema}

3156: Once we know that $\SS3\dd$ is the increasing

3157: enumeration of¨$\DDD3\dd$ and that $\WW3\dd$ is a

3158: $\s$-positive quotient-sequence for¨$\SS3\dd$, we can

3159: count the $2$-jumps in¨$\SS3\dd$ and obtain the value

3160: of¨$\hhh23\dd$ directly: this amounts to forgetting

3161: about all¨$\ss1^{\pm1}$ in the construction

3162: of¨$\WW3\dd$, and it is then fairly obvious that there

3163: only remains $2^\dd-2$ times¨$\ss2$.

3164: \end{rema}

3165: 

3166: \subsection{Larger values of¨$\nn$}

3167: 

3168: The same construction can be developped for¨$\nn = 4$

3169: and further. The general scheme is clear, namely to

3170: define¨$\SS4\dd$ using an inductive rule

3171: \begin{align} \label{E:Def4}

3172: \SS4\dd =  \Dd40 \SS3\dd + \SS4{\dd, 1}

3173: + \Dd41 \SS3\dd + \cdots 

3174: + \Dd4{3\dd-1} \SS3\dd + \SS4{\dd, 3\dd} + \Dd4{3\dd}

3175: \SS3\dd,

3176: \end{align}

3177: where the intermediate factor¨$\SS4{\dd,\ppp}$ is

3178: constructed by concatenating and translating convenient

3179: fragments of¨$\SS4{\dd-1}$. Owing to the inductive

3180: rule¨\eqref{E:Recur4} satisfied by the number of

3181: elements¨$\hhh14\dd$ of¨$\DDD4\dd$, we can expect that the

3182: generic entry of¨$\SS4{\dd-1}$ has to be repeated 6¨times

3183: in¨$\SS4\dd$, but with some entries from¨$\SS4{\dd-2}$

3184: repeated 3¨times only. Once the inductive definition

3185: of¨$\SS4\dd$ is made complete, showing that the sequence

3186: is $\sm$-increasing and counting its entries should be

3187: easy. As we have no complete description so far, we

3188: leave the question open here.

3189: 

3190: \subsection{A new construction for the linear ordering

3191: of¨$B_3$}

3192: 

3193: The main interest of the approach described above is not

3194: only to connect the Garside structure of¨$B_\nn$ with

3195: its linear ordering, but also to provide a new

3196: independent construction of the braid ordering, at least

3197: in the case of¨$B_3$ as the latter is the only one for

3198: which the construction was completed so far.

3199: 

3200: As was recalled in the introduction, the existence of

3201: the linear ordering of braids relies on two properties

3202: of braids, namely Property¨A and Property¨C. These

3203: properties have received a number of independent

3204: proofs¨\cite{Dgr}. In particular, Property¨A has now

3205: a very short proof based on Dynnikov's coordinization

3206: for triangulations of a punctured disk (\cite{Dgr},

3207: Chapter¨9). As for Property¨C, no really simple proof

3208: exists so far: not to mention the initial argument

3209: involving self-distributive algebra, the combinatorial

3210: proofs based on handle reduction or on Burckel's

3211: uniform tree approach, as well as the geometric proofs

3212: based on standardization of curve diagrams require some

3213: care. So, at the moment, one can estimate that the

3214: optimal proof of Property¨C is still missing.

3215: 

3216: A direct application of our construction of the

3217: sequence¨$\SS3\dd$ is

3218: 

3219: \begin{prop}

3220: Property¨C holds for¨$B_3$, \ie, every non-trivial

3221: $3$-braid admits a $\s$-positive or a $\s$-negative

3222: expression.

3223: \end{prop}

3224: 

3225: \begin{proof}[New proof]

3226: We take as an hypothesis that Property¨A is true, so

3227: that the relation¨$\sm$ is a partial ordering, but we do

3228: not assume that $\sm$ is linear. As every braid in¨$B_3$

3229: is the quotient of two positive braids in¨$B_3^+$,

3230: proving Property¨C for¨$B_3$ amounts to proving that, if

3231: $\bx, \by$ are arbitrary elements of¨$B_3^+$, then the

3232: quotient $\bx\inv \by$ admits a $\s$-positive or a

3233: $\s$-negative expression.

3234: 

3235: Now the construction of¨$\SS3\dd$ is self-contained, and

3236: so is that of¨$\WW3\dd$. Then, by construction, every

3237: word in¨$\WW3\dd$ is $\s$-positive. As any concatenation

3238: of $\s$-positive words is $\s$-positive, it follows

3239: that, if $\bx, \by$ are any braids occurring

3240: in¨$\bigcup_\dd\SS3\dd$, then the quotient¨$\bx\inv \by$

3241: admits a $\s$-positive or a $\s$-negative expression,

3242: according to whether $\bx$ occurs before or after¨$\by$

3243: in¨$\SS3\dd$. So, in order to conclude that Property¨C is

3244: true, it just remains to check that each positive

3245: $3$-braid occurs in¨$\bigcup_\dd\SS3\dd$. As

3246: every entry of¨$\SS3\dd$ belongs to¨$\DDD3\dd$, this

3247: is equivalent to proving that each divisor of¨$\DD3\dd$

3248: occurs in¨$\SS3\dd$. Property¨A guarantees that the

3249: entries of¨$\SS3\dd$ are pairwise distinct

3250: (Corollary¨\ref{C:Distinct3}), so it suffices to compare

3251: the length of¨$\SS3\dd$ with the cardinality

3252: of¨$\DDD3\dd$, and this is what we

3253: made in Lemma¨\ref{L:Card3}.

3254: \end{proof}

3255: 

3256: Actually the construction of¨$\SS3\dd$ gives more.

3257: The approach developped by S.\,Burc\-kel in¨\cite{Bus}

3258: consists in introducing a convenient notion of normal

3259: braid words such that every positive braid admits

3260: exactly one normal expression. In the case of $3$¨

3261: strand braids, the definition is as follows. Every

3262: positive $3$¨strand braid word¨$\ww$ can be written as

3263: an alternated product of blocks¨$\ss1^{\expo}$

3264: and¨$\ss2^{\expo}$. Then we define the {\it code}

3265: of¨$\ww$ to be the sequence made by the sizes of these

3266: blocks. To avoid ambiguity, we decide that the last

3267: block is considered to be a block of¨$\ss1$'s, \ie, we

3268: decide that the code of¨$\ss1$ is¨$(1)$, while the code

3269: of¨$\ss2$ is¨$(1, 0)$. For instance, the code

3270: of¨$\ss2^2

3271: \ss1^3\ss2^5$ is $(2, 3, 5, 0)$.

3272: 

3273: \begin{defi}

3274: A positive $3$¨strand braid word¨$\ww$ is said to be

3275: {\it normal in the sense of Burckel} if its code has the

3276: form $(e_1, \dots, e_\ell)$ with $e_\kk \ge 2$ for $2

3277: \le \kk\le \ell-2$.

3278: \end{defi}

3279: 

3280: Burckel shows in¨\cite{Bus} that every positive

3281: $3$-braid admits a unique normal expression and,

3282: moreover, that $\bx \sm \by$ holds if and only if the

3283: normal form of¨$\bx$ is $\mathtt{ShortLex}$-smaller than

3284: the normal form of¨$\by$, where $\mathtt{ShortLex}$

3285: refers to the variant of the lexicographic ordering of

3286: sequences in which the length is given priority: $(e_1,

3287: \dots, e_\ell) <_{\mathtt{ShortLex}} (e'_1, \dots,

3288: e'_{\ell'})$ holds for $\ell < \ell'$, and for $\ell =

3289: \ell'$ and $(e_1, \dots, e_\ell)$ lexicographically

3290: smaller than¨$(e'_1, \dots, e'_{\ell'})$. Burckel's

3291: method consists in defining an iterative reduction

3292: process on non-normal braid words. Our current

3293: approach allows for an alternative, simpler method.

3294: First, a direct inspection shows:

3295: 

3296: \begin{lemm}

3297: Let $\SSS3\dd$ be the sequence of braid words defined by

3298: the inductive rule¨\eqref{E:Def3}. Then $\SSS3\dd$

3299: consists of words that are normal in the sense of

3300: Burckel.

3301: \end{lemm}

3302: 

3303: Then, by construction, every braid in¨$\SS3\dd$ is

3304: represented by a word of¨$\SSS3\dd$, so, as every

3305: positive $3$-braid occurs in¨$\bigcup\SS3\dd$, we

3306: immediately deduce:

3307: 

3308: \begin{prop}

3309: Every positive $3$-braid admits an expression that is

3310: normal in the sense of Burckel.

3311: \end{prop}

3312: 

3313: This in turn enables us to obtain a simple proof for

3314: the following deep, and so far not very well understood

3315: result originally due to Laver¨\cite{Lve} (and

3316: to Burckel \cite{Bus} for the ordinal type):

3317: 

3318: \begin{coro}

3319: The restriction of¨$\sm$ to¨$B_3^+$ is a well-ordering

3320: of ordinal type¨$\omega^\omega$.

3321: \end{coro}

3322: 

3323: \begin{proof}

3324: The $\mathtt{ShortLex}$ ordering of sequences of

3325: nonnegative integers is a well-ordering of ordinal

3326: type¨$\omega^\omega$, so its restriction to codes of

3327: normal words in the sense of Burckel is a well-ordering

3328: as well. The type of the latter cannot be less

3329: than¨$\omega^\omega$ as one can easily exhibit an

3330: increasing sequence of length¨$\omega^\omega$.

3331: \end{proof}

3332: 

3333: Burckel's approach extends to all braid

3334: monoids¨$B_\nn^+$, at the expense of introducing a

3335: convenient notion of normal word and defining an

3336: associated reduction process which is very intricate.

3337: Completing the construction of the sequences¨$\SS4\dd$

3338: and, more generally, $\SS\nn\dd$ along the lines

3339: described above would hopefully allow for an alternative

3340: much simpler approach. In particular, once the correct

3341: definition is given, all subsequent proofs should reduce

3342: to easy inductive verifications.

3343: 

3344: \begin{thebibliography}{99}

3345: 

3346: \Ref Bir; J.\,Birman; Braids, links, and mapping 

3347: class groups; Annals of Math. Studies 82, Princeton

3348: Univ.  Press  (1975). 

3349: 

3350: \Reff Bus; S.\,Burckel; The wellordering on

3351: positive braids; J. Pure Appl. Algebra; 120-1; 1997; 1--17.

3352: 

3353: \Reff But; S.\,Burckel; Computation of the

3354: ordinal of braids; Order; 16; 1999; 291--304.

3355: 

3356: \Reff Buu; S.\,Burckel; Syntactical methods for

3357: braids of three strands; J. Symb. Comp.; 31;

3358: 2001; 557--564.

3359: 

3360: \Reff CSV; L.\,Carlitz, R.\,Scoville \&

3361: T.\,Vaughan; Enumeration of

3362: pairs of permutations; Discrete Math.; 14;

3363: 1976; 215--239.

3364: 

3365: \Reff  Chb; R. Charney; Geodesic automation and

3366: growth functions for Artin groups of finite type; Math.

3367: Ann.; 301-2; 1995; 307--324.

3368: 

3369: \Reff Dfb; P. Dehornoy; Braid groups and left

3370: distributive  operations; Trans. Amer. Math. Soc.; 345-1;

3371: 1994; 115--151. 

3372: 

3373: \Reff Dfo; P.\,Dehornoy; A fast method for comparing

3374: braids; Advances in Math.; 125; 1997; 200--235.

3375: 

3376: \Reff Dgb; P.\,Dehornoy; Strange questions about braids;

3377: J. Knot Th. and its Ramifications; 8-5; 1999; 589--620.

3378: 

3379: \Reff Dgk; P.\,Dehornoy; Groupes de Garside; 

3380: Ann. Scient. Ec. Norm. Sup.; 35; 2002; 267--306.

3381: 

3382: \Ref Dgr; P.\,Dehornoy, I. Dynnikov, D. Rolfsen, B.

3383: Wiest; Why are braids orderable?; Panoramas \&

3384: Synth\`eses vol. 14, Soc. Math. France (2002).

3385: 

3386: \Ref Dhi; P.\,Dehornoy; Combinatorics of normal sequences

3387: of braids; Preprint.

3388: 

3389: \Reff ElM; E.A.\,Elrifai \& H.R.\,Morton; Algorithms for

3390: positive braids; Quart. J. Math. Oxford; 45-2; 1994;

3391: 479--497.

3392: 

3393: \Ref Eps; D.\,Epstein, J.\,Cannon, D.\,Holt, S.\,Levy,

3394: M.\,Paterson \& W.\,Thurston; Word Processing in Groups;

3395: Jones \& Bartlett Publ. (1992).

3396: 

3397: \Reff Gar; F.A.\,Garside; The braid group and

3398: other groups; Quart. J. Math. Oxford; 20-78; 1969;

3399: 235--254.

3400: 

3401: \Ref Lrb; D.\,Larue; Left-distributive and

3402: left-distributive idempotent algebras; Ph D Thesis,

3403: University of Colorado, Boulder (1994).

3404: 

3405: \Reff Lve; R.\,Laver; Braid group actions on left

3406: distributive structures and well-orderings in the braid

3407: group; J. Pure Appl. Algebra; 108-1; 1996; 81--98.

3408: 

3409: \Ref McC; J.\,Mc Cammond; An introduction to Garside

3410: structures; Preprint (2005).

3411: 

3412: \Reff Sol; L.\,Solomon; A Mackey formula in the group ring

3413: of a Coxeter group; J. Algebra; 41; 1976; 255-268.

3414: 

3415: \end{thebibliography}

3416: 

3417: \end{document}

3418: