1: \documentclass[10pt]{amsart}

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

3: %\usepackage{epstopdf}

4: 

5: \hoffset-1.2cm

6: \voffset-0.5cm

7: 

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

9: 

10: \theoremstyle{plain}

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

12: \newtheorem*{thrmA}{Theorem A}

13: \newtheorem*{thrmB}{Theorem B}

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

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

16: \newtheorem{ques}[prop]{Question}

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

18: 

19: \theoremstyle{definition}

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

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

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

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

24: \newtheorem*{rema*}{Remark}

25: 

26: \numberwithin{equation}{section}

27: 

28: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

29: \renewcommand{\a}{\alpha}

30: \renewcommand{\AA}{A}

31: \renewcommand{\aa}{a}%

32: \newcommand\AlgoD{\mathcal{A}}

33: 

34: \renewcommand{\b}{\beta}

35: \newcommand{\bb}{b}%

36: \newcommand{\BB}[1]{B_{#1}^{\scriptscriptstyle+}}

37: \newcommand{\BBB}[1]{\boldsymbol{B}_{\!\,#1}^{\scriptscriptstyle+}}

38: \newcommand\BSucc[2]{#2^{[#1]}}%successor

39: 

40: \newcommand\card{\mathrm{card}}

41: \newcommand{\cc}{c}

42: \newcommand{\CC}{C}

43: \newcommand{\cl}[1]{[#1]}%class mod 2

44: \newcommand{\CL}[1]{\overline{\vrule height5pt width0pt#1}}%class in monoid

45: \newcommand{\conc}{\mathbin{\hbox to 2mm{$^{\scriptscriptstyle\frown}$}}}

46: 

47: \renewcommand{\d}{\delta}%

48: \let\D=\Delta

49: \newcommand{\dd}{d}

50: \newcommand{\DD}[2]{D\!_{{#1}}(#2)}%iterdec

51: \newcommand{\ddd}[1]{\delta_{#1}}

52: \newcommand{\DDhat}[2]{\widehat{\Delta}_{#1, #2}}

53: \newcommand{\DDfe}[2]{\exp{D}_{#1}(#2)}%expseq

54: \newcommand{\DDf}[2]{D_{#1}(#2)}%iterdec

55: \renewcommand{\div}{\prec}

56: \newcommand{\dive}{\preccurlyeq}

57: \newcommand{\divL}{\prec_{_L}}

58: \newcommand{\DR}{\mathrm{Div}_{\scriptscriptstyle\! R}}

59: 

60: \newcommand{\e}{\varepsilon}

61: \newcommand{\ea}{{\scriptstyle\varnothing}}%

62: \newcommand\ee{e}

63: \newcommand\emin{e^{\scriptscriptstyle\mathtt{min}}}

64: \newcommand{\etc}{\emph{etc.}}

65: \renewcommand{\exp}[1]{{#1}^{*}}

66: 

67: \newcommand\ff[1]{\Phi_{#1}}

68: 

69: \newcommand{\g}{\gamma}

70: \let\ge=\geqslant

71: \newcommand{\gen}{g}

72: \newcommand{\gf}{>^{\scriptscriptstyle+}}

73: \newcommand{\gfe}{\ge^{\scriptscriptstyle+}}

74: \newcommand\Gen[1]{\gen_{#1}}

75: \newcommand\GGen{\boldsymbol{\gen}}

76: \newcommand{\gs}{>}

77: \newcommand{\gse}{\ge}

78: 

79: \newcommand{\h}{\theta}

80: 

81: \newcommand\ie{\emph{i.e.}}

82: \newcommand\ii{i}

83: \newcommand\II{I}

84: \newcommand\ince{\subseteq}

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

86: 

87: \newcommand{\jj}{j}

88: \newcommand{\JJ}{J}

89: 

90: \newcommand{\kk}{k}

91: 

92: \renewcommand{\l}{\lambda}

93: \let\le=\leqslant

94: \newcommand\Ldots{...\,}

95: \newcommand\lf{<^{\scriptscriptstyle+}}

96: \newcommand\lfe{\le^{\scriptscriptstyle+}}

97: \renewcommand{\lg}[1]{\vert#1\vert}

98: \newcommand{\lmin}{\mathrel{\widetilde{<}}}

99: \newcommand\ls{<}

100: \newcommand\lse{\le}

101: \newcommand\lSL{<^{\mathtt{\scriptscriptstyle ShortLex}}}

102: \newcommand\lSLe{\le^{\mathtt{\scriptscriptstyle ShortLex}}}

103: 

104: \newcommand{\mm}{m}

105: \newcommand{\MM}{M}

106: \newcommand{\MMM}{\boldsymbol{M}}

107: \newcommand{\Mon}[1]{\langle#1\rangle^{\scriptscriptstyle\!+}}

108: \newcommand{\mult}{\succ}

109: \newcommand{\multe}{\succcurlyeq}

110: 

111: \newcommand{\Nat}{\mathbb{N}}

112: \newcommand{\nn}{n}

113: \newcommand{\nno}{{n-1}}

114: \newcommand{\nnt}{{n-2}}

115: \newcommand{\norm}[1]{\vert\!\vert#1\vert\!\vert}

116: 

117: \newcommand\om{\omega}

118: \newcommand\Ord{\mathbf{Ord}}

119: \newcommand\OVER[1]{\,#1\inv}

120: 

121: \newcommand{\perm}{\pi}

122: \newcommand{\pp}{p}

123: \newcommand{\ppo}{{p-1}}

124: \newcommand{\prece}{\preccurlyeq}

125: \newcommand{\pref}{\mathbin{\scriptstyle\sqsubset}}

126: 

127: \newcommand{\qq}{q}

128: 

129: \newcommand{\rr}{r}

130: \newcommand{\rro}{{r-1}}

131: \newcommand{\rrt}{{r-2}}

132: \def\resp{\emph{resp.}~}

133: 

134: \newcommand{\s}{\sigma}

135: \newcommand{\ShortLex}{\mathtt{ShortLex}}

136: \renewcommand{\ss}[1]{\sigma_{\!#1}^{\vrule height5pt

137: width0pt}}%braidSigma 

138: \renewcommand{\SS}{S}

139: \newcommand{\sss}[1]{\sigma_{#1}^{-1}}%braidSigma 

140: \newcommand\Succ[2]{#2^{(#1)}}%successor

141: 

142: \newcommand{\Tail}[2]{\mathrm{tail}(#1,#2)}

143: \newcommand{\TTT}[2]{\underline{D}_{#1}(#2)}%tree

144: \newcommand{\TTTe}[2]{\exp{\underline{D}}_{#1}(#2)}%tree

145: 

146: \newcommand{\uu}{u}

147: \newcommand{\uuu}{\boldsymbol{s}}

148: 

149: \newcommand{\vv}{v}

150: \newcommand{\vvv}{\boldsymbol{t}}

151: 

152: \newcommand{\ww}{w}

153: \newcommand{\WW}[1]{\underline{B}_{#1}^{\scriptscriptstyle+}}

154: \newcommand{\www}{\boldsymbol{s}}

155: \newcommand{\WWW}[1]{\underline{\boldsymbol{B}}_{\!\,#1}^{\scriptscriptstyle+}}

156: 

157: \newcommand{\xs}{s}

158: \newcommand{\xt}{t}

159: \newcommand{\xx}{x}

160: \newcommand{\XX}{X}

161: 

162: \newcommand{\yy}{y}

163: 

164: \newcommand{\zz}{z}

165: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

166: \begin{document}

167: 

168: \hfill{\tiny 2008-02}

169: 

170: \author{Patrick DEHORNOY}

171: \address{Laboratoire de Math\'ematiques Nicolas Oresme,

172: Universit\'e de Caen, 14032 Caen, France}

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

174: \urladdr{//www.math.unicaen.fr/\!\hbox{$\sim$}dehornoy}

175: 

176: \title{Alternating normal forms for braids and locally

177: Garside monoids}

178: 

179: \keywords{braid group, braid ordering, normal form,

180: Garside monoid, Artin--Tits monoid, locally Garside monoid}

181: 

182: \subjclass{20F36, 20M05, 06F05}

183: 

184: \begin{abstract}

185: We describe new types of normal forms for braid

186: monoids, Artin--Tits monoids, and, more generally, for all

187: monoids in which divisibility has some convenient lattice

188: properties (``locally Garside monoids''). We show that, in

189: the case of braids, one of these normal forms coincides with the

190: normal form introduced by Burckel and  deduce that the latter

191: can be computed easily. This approach leads to a new, simple

192: description for the standard ordering (``Dehornoy order'')

193: of~$B_\nn$ in terms of that of~$B_\nno$, and to a quadratic

194: upper bound for the complexity of this ordering.

195: \end{abstract}

196: \maketitle

197: 

198: The first aim of this paper is to improve our understanding

199: of the well-ordering of positive braids and of the Burckel normal form 

200: of~\cite{BuT, Bur}, which after more than ten years remain

201: mysterious objects. This aim is achieved, at least

202: partially, by giving  a new, alternative definition for the

203: Burckel normal form that makes it natural and easily

204: computable. This new description is direct, involving right

205: divisors only, while Burckel's original approach resorts to

206: iterating some  tricky reduction procedure. It turns out that

207: the construction we describe below relies on a very general

208: scheme for which many monoids are eligible, and we hope for

209: further applications beyond the case of braids.

210: 

211: After the seminal work of F.A.\,Garside~\cite{Gar},

212: we know that braid monoids and, more generally, spherical Artin--Tits

213: monoids and Garside monoids that generalize them, can be 

214: equipped with a normal form, namely the so-called

215: greedy normal form of~\cite{BrS, Adj, ElM, Thu}, which gives

216: for each element of the monoid a distinguished representative

217: word. This normal form is excellent both in theory and in

218: practice as it provides a bi-automatic structure and it is

219: easily computable~\cite{Eps, Cha, Dgk}.

220: 

221: What we do in this paper is to construct a new type of

222: normal form for braid monoids and their generalizations.

223: Our construction keeps one of the ingredients of the (right)

224: greedy normal form, namely considering the maximal right

225: divisor that lies in some subset~$\AA$, but, instead of taking for~$\AA$ the

226: set of so-called simple elements, \ie, the divisors of the

227: Garside element~$\D$, we choose $\AA$ to be some

228: standard parabolic submonoid~$\MM_\II$ of~$\MM$,

229: \ie, the monoid generated by some subset~$\II$ of the

230: standard generating set~$\SS$. When $\II$ is a proper

231: subset of~$\SS$, the submonoid~$\MM_\II$ is a proper subset

232: of~$\MM$, and the construction stops after one step.

233: However, by considering two parabolic submonoids~$\MM_\II$, $\MM_\JJ$ which

234: together generate~$\MM$, we can obtain a well-defined, unique

235: decomposition consisting of alternating factors in~$\MM_\II$

236: and~$\MM_\JJ$, as in the case of an amalgamated product. By considering

237: convenient families of  submonoids, we can iterate the process and

238: obtain a unique normal form for each element of~$\MM$.

239: When it exists, typically in all Artin--Tits monoids, such a

240: normal form is exactly as easy to compute as the greedy

241: normal form, and, as the greedy form, it solves the word problem in quadratic time. 

242: 

243: The above construction is quite general, as it only

244: requires the ground monoid~$\MM$ to be what is

245: now called locally right Garside---or locally left

246: Gaussian in the obsolete terminology of~\cite{Dgl}.

247: However, our main interest in the current paper lies in the case of braids and, more specifically, their ordering. For a

248: convenient choice of the parameters, the alternating normal form  turns out to

249: coincide with the Burckel normal form of~\cite{Bur}. As a

250: consequence, we at last obtain both an easy algebraic

251: description of the latter, and an efficient algorithm for

252: computing it. Mainly, because of the connection between the

253: Burckel normal form and the standard ordering of braids

254: (``Dehornoy order''), we obtain a new characterization of the

255: latter. The result can be summarized as follows. As usual,

256: $\BB\nn$ denotes the monoid of positive $\nn$-strand

257: braids. We use $\ff\nn$ for the involutive flip

258: automorphism of~$\BB\nn$ that maps~$\ss\ii$

259: to~$\ss{\nn-\ii}$ for each~$\ii$, \ie, for conjugation by the Garside element~$\D_\nn$, and $<$ for the upper

260: version of the standard braid ordering.

261: 

262: \begin{thrmA}

263: $(i)$ Every positive $\nn$-strand braid~$\xx$ admits a unique decomposition

264: $$\xx= \ff\nn^\ppo(\xx_\pp) \cdot ...  \cdot \ff\nn^2(\xx_3) \cdot \ff\nn(\xx_2) \cdot \xx_1$$ 

265: with $\xx_\pp, \Ldots, \xx_1$ in~$\BB\nno$ such that, for

266: each $\rr \ge2$, the only generator~$\ss\ii$ that divides

267: $\ff\nn^{\pp-\rr}(\xx_\pp) \cdot ... \cdot \ff\nn(\xx_{\rr+1}) \cdot

268: \xx_{\rr}$ on the right is~$\ss1$. Starting from~$\xx^{(0)} = \xx$, the

269: element~$\xx_\rr$  is determined  by the condition that $\xx_\rr$ is the

270: maximal right divisor of~$\xx^{(\rr)}$ that lies in~$\BB\nno$, and

271: $\xx^{(\rr)}$ is $\ff\nn(\xx^{(\rr-1)}\xx_\rr\inv)$.

272: 

273: $(ii)$ Let $\xx, \yy$ be positive $\nn$-strand braids. Let $(\xx_\pp, \Ldots, \xx_1)$ and $(\yy_\qq,\Ldots,\yy_1)$

274: be the sequences associated with~$\xx$ and~$\yy$ as in~$(i)$. Then $\xx<\yy$ holds in~$\BB\nn$ if and only if we

275: have either $\pp<\qq$, or $\pp=\qq$ and, for some~$\rr \le \pp$, we have $\xx_{\rr'}=\yy_{\rr'}$ for

276: $\rr < \rr' \le \pp$ and $\xx_\rr<\yy_\rr$ in~$\BB\nno$.

277: \end{thrmA}

278: 

279: In other words, via the above decomposition, the ordering of~$\BB\nn$ is a $\ShortLex$-extension of that

280: of~$\BB\nno$, this meaning the variant of lexicographical extension in which the length is given priority.

281: In the above  statement, Point~$(i)$---Proposition~\ref{P:FSplitting} below---is easy, but

282: Point~$(ii)$---Corollary~\ref{C:RecOrder}---is not. Another outcome of the

283: current approach is the following complexity upper

284: bound for the braid ordering---Corollary~\ref{C:Complexity}:

285: 

286: \begin{thrmB}

287: For each~$\nn$, the standard ordering of~$B_\nn$ has at most a quadratic

288: complexity: given two $\nn$-strand braid words~$\uu, \vv$ of

289: length~$\ell$, we can decide whether the braid represented by~$\uu$ is

290: smaller than the braid represented by~$\vv$ in time~$O(\ell^2)$.

291: \end{thrmB}

292: 

293: We think that the tools developed in this paper might be useful for addressing other types of questions, typically

294: those involving conjugacy in~$B_\nn$.

295: 

296: The paper is organized as follows. In Section~\ref{S:Alt}, we describe the alternating decompositions obtained when

297: considering two submonoids in a locally Garside monoid. In Section~\ref{S:Iter}, we show how to iterate the

298: construction using a binary tree of nested submonoids. In Section~\ref{S:NormalForm}, we deduce a

299: normal form result in the case when the base submonoids are generated by atoms. From Section~\ref{S:Flip}, we

300: concentrate on the specific case of braids and investigate what we call the $\Phi$-splitting and the $\Phi$-normal

301: form of a braid. Finally,  in Section~\ref{S:Burckel}, we investigate the connection between the $\Phi$-normal form

302: and the Burckel normal form, and deduce the above mentioned applications to the braid ordering.

303: 

304: \begin{rema*}

305: All constructions developed in this paper involve right

306: divisibility and the derived notions. This choice is dictated

307: by the braid applications of Section~\ref{S:Burckel}. Of

308: course, we could use left divisibility instead and obtain

309: symmetric versions, in the framework of  monoids that are

310: locally Garside on the left.

311: \end{rema*}

312: 

313: We use~$\Nat$ for the set of all nonnegative integers.

314: 

315: \section{Alternating decompositions}

316: \label{S:Alt}

317: 

318: We construct unique decompositions for the elements of

319: monoids in which enough least common left multiples (left

320: lcm's) exist. If $\MM$ is such a monoid

321: and $\AA$ is a subset of~$\MM$ that is closed under the left lcm

322: operation, then, under weak additional assumptions, every

323: element~$\xx$ admits a distinguished decomposition $\xx= \xx'

324: \xx_1$, where $\xx_1$ is a maximal right divisor of~$\xx$ that

325: lies in~$\AA$. The element~$\xx_1$ will be called the

326: $\AA$-tail of~$\xx$. If we assume that every non-trivial (\ie,

327: $\not=1$) element of~$\MM$ has a non-trivial $\AA$-tail, we can

328: consider the $\AA$-tail of~$\xx'$, and, iterating the process,

329: obtain a distinguished decomposition of~$\xx$ as a product of

330: elements of~$\AA$, as done for the standard greedy normal

331: form of Garside monoids. Here, we drop the assumption

332: that every non-trivial element has a non-trivial $\AA$-tail, but

333: instead consider two subsets~$\AA_1, \AA_2$ of~$\MM$ with

334: the property that, for every non-trivial~$\xx$, at least one of

335: the $\AA_1$- or $\AA_2$-tails of~$\xx$ is non-trivial. Then, we

336: obtain a distinguished decomposition of~$\xx$ as an alternating

337: product of elements of~$\AA_1$ and of~$\AA_2$.

338: 

339: \subsection{Locally Garside monoids}

340: 

341: Divisibility features play a key r\^ole throughout 

342: the paper, and we first fix some notation. 

343: 

344: \begin{nota}

345: For $\MM$ a monoid and $\xx,\yy\in\MM$, we say that

346: $\yy$ is a \emph{right divisor} of~$\xx$, or, equivalently,

347: that $\xx$ is a \emph{left multiple} of~$\yy$, denoted

348: $\xx\multe\yy$, if $\xx=\zz\yy$ holds

349: for some~$\zz$; we write $\xx\mult\yy$ if $\xx=\zz\yy$

350: holds for some~$\zz\not=1$. The set of all right divisors

351: of~$\xx$ is denoted by~$\DR(\xx)$. 

352: \end{nota}

353: 

354: The approach considered below turns out to be relevant

355: for the following monoids.

356: 

357: \begin{defi}

358: \label{D:LocGarside}

359: We say that a monoid~$\MM$ is a \emph{locally right Garside} if:

360: 

361: $(C_1)$ The monoid $\MM$ is right cancellative, \ie,

362: $\xx\zz=\yy\zz$ implies $\xx=\yy$;

363: 

364: $(C_2)$ Any two elements of~$\MM$ that admit a

365: common left multiple admit a left~lcm; 

366: 

367: $(C_3)$ For every~$\xx$ in~$\MM$, there is no infinite

368: chain $\xx_1 \div \xx_2 \div ... $ in~$\DR(\xx)$.

369: \end{defi}

370: 

371: If $\MM$ is a locally right Garside monoid, and

372: $\xx,\yy$ are elements of~$\MM$ satisfying

373: $\xx\multe\yy$, the element~$\zz$ satisfying $\xx=\zz\yy$

374: is unique by right cancellativity, and we denote it

375: by~$\xx\OVER{\yy}$.

376: 

377: \begin{exam}

378: \label{X:Artin}

379: According to~\cite{BrS} and~\cite{Mic}, all Artin--Tits

380: monoids are locally right (and left) Garside. We recall that

381: an Artin--Tits monoid is a monoid generated by  a

382: set~$\SS$ and relations of the form

383: $\xs\xt\xs...=\xt\xs\xt...$ with $\xs,\xt \in \SS$, both sides of the same length,

384: and at most one such relation for each pair~$\xs,\xt$. An important

385: example is Artin's braid monoid~$\BB\nn$ \cite{KaT}, which corresponds to

386: $\SS=\{\ss1, \Ldots, \ss\nno\}$ with

387: \begin{equation}

388: \label{Pres}

389: \ss i \ss j = \ss j \ss i

390: \quad\mbox{for $\vert i - j\vert \ge 2$,}

391: \qquad

392:  \ss i \ss j \ss i = \ss j \ss i \ss j 

393: \quad\mbox{for $\vert i - j\vert = 1$}.

394: \end{equation}

395: As the name suggests, more general examples of locally

396: Garside monoids are the Garside monoids

397: of~\cite{Dfx, Dgk, CMW, ChM, Pid}, which include torus

398: knot monoids~\cite{Pik}, dual braid monoids~\cite{BKL},

399: and many more.

400: \end{exam}

401: 

402: If $\MM$ is locally right Garside, then no non-trivial

403:  element of~$\MM$ is invertible: if we

404: had $\xx\yy=1$ with $\xx\not=1$, hence $\yy\not=1$, the

405: sequence $\xx, 1, \xx, 1,...$ would

406: contradict~$(C_3)$. So right divisibility

407: is antisymmetric, and, therefore, it is a

408: partial ordering on~$\MM$. As a consequence, the left lcm, when it exists, is

409: unique.

410: 

411: Definition~\ref{D:LocGarside}---which also appears

412: in~\cite{DiM}---is satisfactory in that it exclusively involves

413: the right divisibility relation, and it directly leads to

414: Lemma~\ref{L:Lcm} below. Actually, it does not coincide with

415: the definitions of~\cite{Dgk} and~\cite{Dfx}, where

416: $(C_3)$ is replaced with some condition involving left

417: divisibility. However, both definitions are equivalent. For a 

418: while, we use $\divL$ for the proper left divisibility relation,

419: \ie, $\xx\divL\yy$ means $\yy=\xx\zz$ with $\zz\not=1$. We

420: denote by~$\Ord$ the class of ordinals.

421: 

422: \begin{lemm}

423: \label{L:LocGarside}

424: $(i)$ If $\MM$ is right cancellative,

425: Condition~$(C_3)$ is equivalent~to

426: 

427: $(C'_3)$ There is no infinite descending

428: chain in~$(\MM,\divL)$.

429: 

430: $(C''_3)$ There exists $\l: \MM\to\Ord$ such that $\yy\not=1$

431: implies $\l(\xx\yy) > \l(\xx)$.

432: 

433: \noindent $(ii)$ In any case, Conditions~$(C_3)$--$(C''_3)$

434: follow from

435: 

436: $(C_3^+)$ There exists $\l\!:\!\MM\to\Nat$

437: such that $\yy\not=1$ implies

438: $\l(\xx\yy)\ge\l(\xx)+\l(\yy) >\nobreak \l(\xx)$.

439: \end{lemm}

440: 

441: \begin{proof}

442: $(i)$ Assume that $\MM$ is right cancellative and

443: $(C_3)$ fails in~$\MM$. There exists~$\xx$

444: in~$\MM$ and a sequence $\xx_1, \xx_2,...$ in~$\DR(\xx)$

445: such that

446: $\xx_{\nn+1}\mult \xx_\nn$ holds for every~$\nn$. So,

447: for each~$\nn$, there exists~$\yy_\nn\not=1$ satisfying

448: $\xx_{\nn+1}=\yy_\nn\xx_\nn$. On the other hand, as

449: $\xx_\nn$ belongs to~$\DR(\xx)$, there exist~$\zz_\nn$

450: satisfying $\xx=\zz_\nn\xx_\nn$. We find

451: $$\xx=\zz_\nn\xx_\nn=\zz_{\nn+1}\xx_{\nn+1}=

452: \zz_{\nn+1}\yy_\nn\xx_\nn.$$

453: By cancelling~$\xx_\nn$

454: on the right, we deduce $\zz_\nn=\zz_{\nn+1}\yy_\nn$,

455: hence $\zz_{\nn+1}\divL\zz_\nn$ for each~$\nn$, and the

456: sequence $\zz_0, \zz_1, ...$ witnesses that $(C'_3)$ fails.

457: 

458: Conversely, assume that $(C'_3)$ fails in~$\MM$.

459: Let $\zz_0, \zz_1, ...$ be a descending chain for~$\divL$. For

460: each~$\nn$, choose~$\yy_\nn\not=1$ satisfying

461: $\zz_\nn=\zz_{\nn+1}\yy_\nn$. Let $\xx=\zz_0$,

462: $\xx_1=1$, and, inductively,

463: $\xx_{\nn+1}=\yy_\nn\xx_\nn$. By construction, we

464: have $\xx_{\nn+1} \mult\xx_\nn$ for each~$\nn$.

465: Now, we also have $\xx=\zz_\nn\xx_\nn$ for each~$\nn$,

466: so all elements~$\xx_\nn$ belong to~$\DR(\xx)$, and the sequence

467: $\xx_1, \xx_2, ...$ witnesses that $(C_3)$ fails.

468: 

469: The equivalence of~$(C'_3)$ and~$(C''_3)$ is standard, and

470: $(C_3^+)$ strengthens~$(C''_3)$.

471: \end{proof}

472: 

473: Condition~$(C_3^+)$ holds in particular in every monoid

474: that is presented by homogeneous relations, \ie, relations of

475: the form~$\uu=\vv$ where $\uu$ and~$\vv$ are words of

476: the same length: then we can define~$\l(\xx)$ to

477: be the length of any word representing~$\xx$. This is the

478: case for the Artin--Tits monoids of Example~\ref{X:Artin}.

479: 

480: Lemma~\ref{L:LocGarside} implies that locally right Garside

481: monoids coincide with the monoids called locally left Gaussian

482: in~\cite{Dgk}, in connection with the left Gaussian monoids

483: of~\cite{Dfx}. The reason for changing terminology is that

484: the current definition is coherent with~\cite{DiM} and it is

485: more natural: locally right Garside monoids involve right

486: divisibility, and the normal forms we discuss below are

487: connected with what is usually called the right normal form.

488: 

489: Assume that $\MM$ is a locally right Garside monoid.

490: Condition~$(C_2)$ is equivalent to saying that, for

491: every~$\xx$ in~$\MM$, any two elements of~$\DR(\xx)$

492: admit a left lcm, and it follows that any finite subset

493: of~$\DR(\xx)$ admits a global left lcm. By the Noetherianity

494: condition~$(C_3)$, the result extends to arbitrary subsets.

495: We say that a set~$\XX$ is \emph{closed under left lcm} if the

496: left lcm of any two elements of~$\XX$ exists and lies

497: in~$\XX$ whenever it exists in~$\MM$, \ie, by~$(C_2)$,

498: whenever these elements admit a common left multiple

499: in~$\MM$.

500: 

501: \begin{lemm}

502: \label{L:Lcm}

503: Assume that $\MM$ is a locally right Garside monoid, and

504: $\xx\in\MM$. Then every nonempty subset~$\XX$ of~$\DR(\xx)$ admits

505: a global left lcm~$\xx_1$; if moreover $\XX$ is closed

506: under left lcm, then $\xx_1$ belongs to~$\XX$.

507: \end{lemm}

508: 

509: \begin{proof}

510: Assume first that $\XX$ is closed under left lcm. By the

511: axiom of dependent choices, Condition~$(C_3)$ implies

512: that $(\DR(\xx),\mult)$ is a well-founded poset, so

513: $\XX$ admits some $\mult$-minimal, \ie, some

514: $\div$-maximal, element~$\xx_1$: so $\xx'\xx_1\in\XX$

515: implies~$\xx'=1$.  Then $\xx_1$ is a global left lcm for~$\XX$.

516: Indeed, assume $\yy_1 \in \XX$. By hypothesis,

517: $\xx_1$ and~$\yy_1$ lie in~$\DR(\xx)$, so, by~$(C_2)$, they

518: admit a left lcm~$\zz$, which can be expressed

519: as~$\zz = \yy'\yy_1=\xx'\xx_1$. The hypothesis that $\XX$ is

520: closed under left lcm implies $\zz\in\XX$. The choice

521: of~$\xx_1$ implies $\xx'=1$, hence $\xx_1 \multe \yy_1$.

522: 

523: If the assumption that $\XX$ is closed under left

524: lcm is dropped, we can apply the above result to the

525: closure~$\widehat\XX$ of~$\XX$ under left lcm. Then the global left

526: lcm~$\xx_1$ of~$\widehat\XX$ is a global left lcm for~$\XX$, but we

527: cannot be sure that

528: $\xx_1$ lies in~$\XX$---yet it is certainly the left lcm of some

529: finite subset of~$\XX$.

530: \end{proof}

531: 

532: Although standard, the previous result is crucial. By applying

533: Lemma~\ref{L:Lcm} to the subset $\DR(\xx)\cap\DR(\yy)$

534: of~$\DR(\xx)$, we deduce that any two elements~$\xx,\yy$

535: of a locally right Garside monoid~$\MM$ admit a right gcd

536: (greatest common divisor), and, therefore, for every~$\xx$

537: in~$\MM$, the structure $(\DR(\xx),\multe)$ is a lattice,

538: with minimum~$1$ and maximum~$\xx$. 

539: 

540: \subsection{The $\AA$-tail of an element}

541: 

542: If $\MM$ is a  monoid and $\xx, \xx_\pp, \Ldots, \xx_1$ belong to~$\MM$, we say

543: that $(\xx_\pp, \Ldots, \xx_1)$ is a \emph{decomposition} of~$\xx$ if $\xx = \xx_\pp

544: ... \xx_1$ holds. The basic observation is that,  for each subset~$\AA$

545: of the monoid~$\MM$ that contains~$1$ and is closed under left

546: lcm, and every~$\xx$ in~$\MM$, Lemma~\ref{L:Lcm} leads to a distinguished

547: decomposition~$(\xx', \xx_1)$ of~$\xx$ with~$\xx_1 \in \AA$. 

548: 

549: \begin{lemm}

550: \label{L:Tail}

551: Assume that $\MM$ is a locally right Garside monoid and

552: $\AA$ is a  subset of~$\MM$ that contains~$1$ and is closed under left lcm.

553: Then, for each element~$\xx$ of~$\MM$, there exists a

554: unique right divisor~$\xx_1$ of~$\xx$ that lies in~$\AA$

555: and is maximal with respect to right divisibility, namely the

556: left lcm of $\DR(\xx)\cap\AA$.

557: \end{lemm}

558: 

559: \begin{proof}

560: Apply Lemma~\ref{L:Lcm} with $\XX=\DR(\xx)\cap\AA$.

561: The latter set  is nonempty as it contains at least~$1$, and it is closed under left lcm as it is the

562: intersection of two sets that are closed under left lcm.

563: \end{proof}

564: 

565: \begin{defi}

566: Under the hypotheses of Lemma~\ref{L:Tail}, the

567: element~$\xx_1$ is called the \emph{$\AA$-tail} of~$\xx$,

568: and denoted~$\Tail\xx\AA$.

569: \end{defi}

570: 

571: \begin{exam}

572: \label{X:Closed}

573: Let $\MM$ be an Artin--Tits monoid with standard set of

574: generators~$\SS$. We assume in addition that $\MM$ is of

575: spherical type, which means that the Coxeter group

576: obtained by adding the relation~$\xs^2=1$ for each~$\xs$ in~$\SS$ is finite.

577: Then, Garside's theory shows that any two elements of~$\MM$ 

578: admit a common left multiple, hence a left lcm. We shall

579: consider two types of closed subsets of~$\MM$. The first,

580: standard choice consists in considering the set~$\Sigma$ of

581: so-called simple elements in~$\MM$, namely the divisors

582: of the lcm~$\D$ of~$\SS$. By construction,

583: $\Sigma$ contains~$1$ and is closed under left (and right) divisor, and under

584: left (and right) lcm. For each~$\xx$

585: in~$\MM$, the $\Sigma$-tail of~$\xx$ is the right gcd

586: of~$\xx$ and~$\D$.

587: 

588: A second choice consists in considering $\II \subseteq \SS$, and taking

589: for~$\AA$ the standard parabolic submonoid~$\MM_\II$ of~$\MM$ generated

590: by~$\II$. The specific form of the Artin--Tits relations

591: implies that $\MM_\II$ is closed under left (and right)

592: divisor, and under left (and right) lcm, hence it is

593: eligible for our approach. Denote by~$\D_\II$

594: the lcm of~$\II$. Then, for every

595: element~$\xx$ of~$\MM$, the

596: $\MM_\II$-tail~$\xx_1$ of~$\xx$ is the right gcd of~$\xx$

597: and~$\D_\II^{\lg\xx}$, where $\lg\xx$ denotes the common

598: length of all words representing~$\xx$. Indeed, let~$\xx'_1$

599: be the latter gcd, and let $\ell=\lg\xx$. By definition,

600: $\xx_1$ is a right divisor of~$\xx$, so we have

601: $\lg{\xx_1}\le\ell$, and, as for each~$\zz$ in~$\MM_\II$

602: satisfying

603: $\lg\zz\le\ell$ is, we have $\D_\II^\ell \multe \xx'_1$,

604: hence $\xx'_1 \multe \xx_1$. Conversely, $\xx'_1$

605: is an element of~$\DR(\xx)\cap\MM_\II$, hence we have

606: $\xx_1 \multe \xx'_1$, and, finally, $\xx_1=\xx'_1$.

607: Note that the previous approach does not require

608: that $\MM$ be of spherical type, but only that

609: $\MM_\II$ is. Actually, $\MM_\II$ is a closed submonoid

610: even if it is not of spherical type---but, then, the

611: characterization of the $\MM_\II$-tail in terms of

612: powers of~$\D_\II$ vanishes.

613: \end{exam}

614: 

615: \subsection{Alternating decompositions}

616: 

617: In the second case of Example~\ref{X:Closed}, the

618: involved subset is a submonoid of~$\MM$, \ie, in

619: addition to being closed under left lcm, it is closed

620: under multiplication. From now on, we shall

621: concentrate on this situation. Then, the decomposition of

622: Lemma~\ref{L:Tail} takes a specific form.

623: 

624: \begin{defi}

625: \label{D:Closed}

626: Assume that $\MM$ is a locally right Garside monoid. We say that a submonoid~$\MM_1$ of~$\MM$ is \emph{closed} if it is closed under both left lcm and left divisor, \ie,

627: every left lcm of elements of~$\MM_1$ belongs to~$\MM_1$

628: and every left divisor of an element of~$\MM_1$ belongs

629: to~$\MM_1$.

630: \end{defi}

631: 

632: \begin{exam}

633: If $\MM$ is an Artin--Tits monoid with standard set of

634: generators~$\SS$, then every standard parabolic submonoid

635: of~$\MM$ is closed. This need not be the case in every locally

636: right Garside monoid, or even in every Garside monoid. For

637: instance, the monoid $\Mon{\aa,\bb \mid

638: \aa\bb\aa=\bb^2}$ is Garside, hence locally right

639: Garside---the associated Garside group is the braid

640: group~$B_3$. However, the submonoid generated by~$\bb$

641: is not closed, as it contains~$\bb^2$, which is

642: $\aa\bb\aa$, but it contains neither~$\aa$ nor~$\aa\bb$,

643: which are left divisors of~$\bb^2$.

644: \end{exam}

645: 

646: \begin{nota}

647: For~$\MM$ a monoid, $\xx\in\MM$ and $\AA\ince\MM$,

648: we write $\xx\perp\AA$ if no non-trivial element of~$\AA$ is a

649: right divisor of~$\xx$, \ie, if $\DR(\xx)\cap\AA$ is

650: either~$\emptyset$ or~$\{1\}$.

651: \end{nota}

652: 

653: \begin{lemm}

654: \label{L:Decomp}

655: Assume that $\MM$ is a locally right Garside monoid, and

656: $\MM_1$ is a closed submonoid of~$\MM$. Then, for each~$\xx$ in~$\MM$,

657: there exists a unique decomposition

658: $(\xx', \xx_1)$ of~$\xx$ satisfying

659: \begin{equation}

660: \label{E:Decomp}

661: \xx' \perp\MM_1\text{\quad and\quad}

662: \xx_1\in\MM_1,

663: \end{equation}

664: namely the one given by

665: $\xx_1=\Tail\xx{\MM_1}$ and $\xx'=\xx\OVER{\xx_1}$.

666: \end{lemm}

667: 

668: \begin{proof}

669: Let $\xx_1=\Tail\xx{\MM_1}$ and $\xx'=\xx\OVER{\xx_1}$. We

670: claim that, for each decomposition $(\yy', \yy_1)$ of~$\xx$ with

671: $\yy_1\in\MM_1$, we have

672: \begin{equation}

673: \label{E:CharTail}

674: \yy' \perp \MM_1

675: \quad\Longleftrightarrow\quad

676: \yy_1=\xx_1.

677: \end{equation}

678: First, assume $\zz \in \DR(\xx')\cap\MM_1$.

679: Then we have $\xx'=\xx''\zz$ for some $\xx''$, hence

680: $\xx=\xx''\zz\xx_1$, and $\zz\xx_1\in\DR(\xx)$.

681: As $\zz$ and $\xx_1$ belong to~$\MM_1$ and the

682: latter is a submonoid of~$\MM$, we deduce

683: $\zz\xx_1\in\MM_1$, hence $\zz=1$ by definition

684: of~$\xx_1$. So $\xx' \perp \MM_1$ holds, and the $\Longleftarrow$

685: implication in~\eqref{E:CharTail} is true.

686: 

687: Conversely, assume $\xx=\yy'\yy_1$ with

688: $\yy_1\in\MM_1$. By definition of the $\MM_1$-tail, we have

689: $\xx_1=\zz\yy_1$ for some~$\zz$. The assumption that $\MM_1$ is closed

690: under left divisor implies $\zz\in\MM_1$. Then we find

691: $\yy'\yy_1=\xx= \xx'\xx_1 = \xx'\zz\yy_1$, hence

692: $\yy'=\xx'\zz$ by cancelling~$\yy_1$,

693: and finally $\zz\in\DR(\yy')\cap\MM_1$. Then

694: $\DR(\yy')\cap\MM_1=\{1\}$ implies $\zz=1$, \ie,

695: $\yy_1=\xx_1$, and, from there, $\yy'=\xx'$. So the

696: $\Longrightarrow$ implication in~\eqref{E:CharTail} is true.

697: \end{proof}

698: 

699: By definition, the relation $\xx' \perp \MM_1$

700: of~\eqref{E:CharTail} is equivalent to $\Tail{\xx'}{\MM_1} =

701: 1$. This shows that iterating the decomposition of

702: Lemma~\ref{L:Decomp} makes no sense: we extracted the

703: maximal right divisor of~$\xx$ that lies in~$\MM_1$, so,

704: after that, there remains nothing to extract any longer. But

705: assume that $\MM$ is locally right Garside, and that

706: $\MM_2,\MM_1$ are \emph{two} closed submonoids of~$\MM$.

707: For each~$\xx$ in~$\MM$, Lemma~\ref{L:Decomp} gives

708: a distinguished decomposition $(\xx', \xx_1)$ of~$\xx$

709: with~$\xx_1$ in~$\MM_1$. If $\xx'$ is not~$1$, and if

710: $\MM_2 \cup \MM_1$ generates~$\MM$, the $\MM_2$-tail

711: of~$\xx'$ is not~$1$, and we obtain a new decomposition

712: $(\xx'' , \xx_2 , \xx_1)$ of~$\xx$ with $\xx_2\in\MM_2$

713: and $\xx_1\in\MM_1$. If $\xx''$ is not~$1$, we repeat the

714: process with~$\MM_1$, \etc\  finally obtaining a

715: decomposition of~$\xx$ as an alternating sequence of

716: elements of~$\MM_2$ and~$\MM_1$.

717: 

718: \begin{defi}

719: \label{D:Covering}

720: If $\MM$ is a locally right Garside monoid, we say that $(\MM_2,\MM_1)$

721: is a \emph{covering} of~$\MM$ if $\MM_2$ and $\MM_1$ are

722: closed submonoids of~$\MM$ and  $\MM_2\cup\MM_1$

723: generates~$\MM$ (as a monoid).

724: \end{defi}

725: 

726: \begin{exam}

727: Let $\MM$ be an Artin--Tits monoid with standard

728: set of generators~$\SS$, and let $\SS_2,

729: \SS_1$ be two subsets of~$\SS$ satisfying

730: $\SS_2\cup\SS_1=\SS$. For $\kk =2,1$, let

731: $\MM_\kk$ be the standard parabolic submonoid

732: of~$\MM$ generated by~$\SS_\kk$. Then $(\MM_2, \MM_1)$

733: is a covering of~$\MM$. Indeed, we already observed that

734: $\MM_1$ and $\MM_2$ are closed submonoids of~$\MM$.

735: Moreover, $\SS$ is included in~$\MM_2\cup\MM_1$, so

736: the latter generates~$\MM$. 

737: \end{exam}

738: 

739: Similar results hold for every locally right Garside monoid

740: that is generated by the union of two sets~$\SS_2, \SS_1$

741: provided we define $\MM_\kk$ to be the smallest \emph{closed}

742: submonoid of~$\MM$ generated by~$\SS_\kk$. 

743: 

744: \begin{nota}

745: \label{N:Parity}

746: For each (nonnegative) integer~$\rr$, we define $\cl\rr$ to

747: be~$1$ if $\rr$ is odd, and $2$ if $\rr$ is even.

748: \end{nota}

749: 

750: \begin{prop}

751: \label{P:AltDec}

752: Assume that $\MM$ is a locally right Garside monoid and

753: $(\MM_2,\MM_1)$ is a covering of~$\MM$. Then, for 

754: every non-trivial element~$\xx$ of~$\MM$, there exists a

755: unique decomposition $(\xx_\pp, \Ldots, \xx_1)$ of~$\xx$ satisfying

756: $\xx_\pp\not=1$ and,  for each~$\rr \ge 1$,

757: \begin{equation}

758: \label{E:AltDec1}

759: \xx_p...\xx_{\rr+1} \perp \MM_{\cl\rr}

760: \text{\quad and\quad}

761: \xx_\rr\in\MM_{\cl\rr}.

762: \end{equation}

763: The elements~$\xx_\rr$ are determined from

764: $\xx^{(0)}= \xx$ by

765: \begin{equation}

766: \label{E:AltDec2}

767: \xx_\rr= 

768: \Tail{\xx^{(\rr-1)}}{\MM_{\cl\rr}}

769: \text{\quad and\quad}

770: \xx^{(\rr)}=\xx^{(\rr-1)}\OVER{\xx_\rr}. 

771: \end{equation}

772: Moreover, we have $\xx_\rr\not=1$ for $\rr\ge2$.

773: \end{prop}

774: 

775: \begin{proof}

776: Let $\xx$ belong to~$\MM$, and let $\xx_\rr$,

777: $\xx^{(\rr)}$ be as specified by~\eqref{E:AltDec2}.

778: Using induction on $\rr\ge1$, we first prove the

779: relations

780: \begin{gather}

781: \label{E:Alt3}

782: \xx=\xx^{(\rr)}\xx_\rr\cdots\xx_1,\\

783: \label{E:Alt4}

784: \xx^{(\rr)} \perp\MM_{\cl\rr}.

785: \end{gather}

786: For $\rr=1$, Lemma~\ref{L:Decomp} for~$\xx$

787: and~$\MM_1$ gives $\xx=\xx^{(1)}\xx_1$, which

788: is~\eqref{E:Alt3}, and

789: $\xx^{(1)} \perp \MM_1$, which is~\eqref{E:Alt4}.

790: Assume $\rr\ge2$. Then \eqref{E:AltDec2} implies

791: $\xx^{(\rr-1)}=\nobreak\xx^{(\rr)}\xx_\rr$, and, 

792: susbtituting in $\xx=\xx^{(\rr-1)}\xx_{\rr-1}...\xx_1$, which

793: holds by induction hypothesis, we obtain~\eqref{E:Alt3}.

794: Moreover, Lemma~\ref{L:Decomp} for~$\xx^{(\rr)}$ and

795: $\MM_{\cl\rr}$ gives~\eqref{E:Alt4}.

796: 

797: By construction, the sequence $\xx_1$, $\xx_2\xx_1$,

798: $\xx_3\xx_2\xx_1$, ... is increasing

799: in~$(\DR(\xx), \div)$. By Condition~$(C_3)$, it is

800: eventually constant. By right cancellability, this implies

801: that there exists~$\pp$ such that $\xx_\rr=\xx^{(\rr)}=1$

802: holds for all~$\rr\ge\pp$. Then \eqref{E:Alt3} implies

803: $\xx=\xx_\pp...\xx_1$, with $\xx_\pp\not=1$ provided

804: $\pp$ is chosen to be minimal and $\xx$ is not~$1$.

805: 

806: So the expected sequence $(\xx_\pp, \Ldots, \xx_1)$ exists and

807: satisfies~\eqref{E:AltDec1} and~\eqref{E:AltDec2}. We show now

808: $\xx_\rr\not=1$ for $\rr\ge2$. Indeed, assume

809: $\xx^{(\rr-1)}\not=1$. By hypothesis,

810: $\MM_2\cup\MM_1$ generates~$\MM$, implying

811: $\xx^{(\rr-1)} \not\perp (\MM_2\cup\MM_1)$.

812: By~\eqref{E:Alt4}, we have $\xx^{(\rr-1)} \perp

813: \MM_{\cl{\rr-1}}$, hence $\xx^{(\rr-1)} \not\perp

814: \MM_{\cl\rr}$. Therefore the $\MM_{\cl\rr}$-tail

815: of~$\xx^{(\rr-1)}$, which by definition is~$\xx_\rr$, is

816: not~$1$---the argument fails for $\rr=1$ because $\xx^{(0)}

817: \perp \MM_{\cl0}$ need not hold.

818: 

819: We turn to uniqueness. Consider any decomposition

820: $(\yy_\qq,  \Ldots, \yy_1)$ of~$\xx$ satisfying $\yy_\qq\not=1$ with 

821: $\yy_\rr\in\MM_{\cl\rr}$ and

822: $\yy_\qq...\yy_{\rr+1} \perp\MM_{\cl\rr}$ for

823: each~$\rr$. We inductively prove $\yy_\rr=\xx_\rr$

824: and $\yy_\qq...\yy_{\rr+1}=\xx^{(\rr)}$ for

825: $\rr\ge1$. For $\rr=1$, the hypotheses $\xx=

826: (\yy_\qq...\yy_2)\yy_1$ with $\yy_1\in\MM_1$ and

827: $\yy_\qq...\yy_2 \perp \MM_1$ imply

828: $\yy_1=\xx_1$ and $\yy_\qq...\yy_2= \xx^{(1)}$ by

829: Lemma~\ref{L:Decomp}. Assume $\rr\ge2$. By induction

830: hypothesis, we have

831: $\yy_\qq...\yy_{\rr}= \xx^{(\rr-1)}$, and the hypotheses

832: about the elements~$\yy_\jj$ give

833: $\xx^{(\rr-1)} = (\yy_\qq...\yy_{\rr+1}) \yy_\rr$ with

834: $\yy_\rr\in\MM_{\cl\rr}$ and

835: $\yy_\qq...\yy_{\rr+1} \perp \MM_{\cl\rr}$. Then

836: Lemma~\ref{L:Decomp} implies

837: $\yy_\rr=\Tail{\xx^{(\rr-1)}}{\MM_{\cl\rr}} = \xx_\rr$ and

838: $\yy_\qq...\yy_{\rr+1}= \xx^{(\rr-1)}\OVER{\xx_\rr} =

839: \xx^{(\rr)}$. Finally,  $\qq>\pp$ would imply

840: $\xx_\qq=\yy_\qq\not=1$, contradicting the choice of~$\pp$.

841: \end{proof}

842: 

843: \begin{defi}

844: \label{D:Decomp}

845: In the framework of Proposition~\ref{P:AltDec}, the sequence

846: $(\xx_\pp, \Ldots, \xx_1)$ is called the  \emph{$(\MM_2,\MM_1)$-decomposition} of~$\xx$.

847: \end{defi}

848: 

849: \begin{exam}

850: \label{X:Delta}

851: Consider the $4$-strand braid monoid~$\BB4$. Let

852: $\MM_1$ be the submonoid generated by~$\ss1$

853: and~$\ss2$, \ie, $\BB3$, and $\MM_2$ be the submonoid

854: generated by~$\ss2$ and~$\ss3$. Choose

855: $\xx=\D_4^2=(\ss1\ss2\ss1\ss3\ss2\ss1)^2$. The

856: computation of the $(\MM_2,\MM_1)$-decomposition

857: of~$\xx$ is as follows:

858: \begin{align*}

859: \smash{\xx^{(0)}} 

860: &= \xx = \smash{\D_4^2}

861: &\xx_1

862: &=\smash{\Tail{\xx^{(0)}}{\MM_1}

863: =\D_3^2},\\

864: \smash{\xx^{(1)}}

865: &=\smash{\xx^{(0)}}\OVER{\xx_1}

866: =\ss3\ss2\s_1^2\ss2\ss3

867: &\xx_2 

868: &=\smash{\Tail{\xx^{(1)}}{\MM_2}}

869: =\ss2\ss3,\\

870: \xx^{(2)}

871: &=\xx^{(1)}\OVER{\xx_2}

872: =\smash{\ss3\ss2\s_1^2}

873: &\xx_3 

874: &=\Tail{\xx^{(2)}}{\MM_1}

875: =\smash{\ss2\s_1^2},\\

876: \xx^{(3)}

877: &=\xx^{(2)}\OVER{\xx_3}

878: =\ss3

879: &\xx_4 

880: &=\Tail{\xx^{(3)}}{\MM_2}

881: =\ss3,\\

882: \xx^{(4)}

883: &=\smash{\xx^{(3)}\OVER{\xx_4}}

884: =1.

885: \end{align*}

886: Thus the $(\MM_2,\MM_1)$-decomposition of~$\D_4^2$ is the

887: sequence $(\ss3, \ss2\s_1^2, \ss2\ss3, \D_3^2)$---see Figure~\ref{F:Alt} for an illustration in terms of

888: standard braid diagrams. Note that the decomposition depends on the

889: order of the submonoids: the

890: $(\MM_1,\MM_2)$-decomposition of~$\D_4^2$ is

891: $(\ss1, \ss2\s_3^2, \ss2\ss1, (\ss2\ss3\ss2)^2)$.

892: \end{exam}

893: 

894: \begin{figure}[htb]

895: \begin{picture}(72,23)

896: \put(0,0){\includegraphics{Decomp.eps}}

897: \put(60.8,10){$\xx_1$}

898: \put(44,16){$\xx_2$}

899: \put(26.8,10){$\xx_3$}

900: \put(10,16){$\xx_4$}

901: \put(-4,13){$\dots$}

902: \end{picture}

903: \caption{\sf Diagram associated with the $(\MM_2,

904: \MM_1)$-decomposition of a $4$-braid: starting from the right,

905: we alternatively select the maximal right divisor that does not

906: involve the $\nn$th strand and the first strand.}

907: \label{F:Alt}

908: \end{figure}

909: 

910: \begin{rema}

911: In the framework of Proposition~\ref{P:AltDec}, $\xx_1$ is the

912: left lcm of all right divisors of~$\xx$ that lie in~$\MM_1$. Comparing with the case of the greedy normal form, we

913: might expect that, similarly, $\xx_2 \xx_1$ is the left lcm of

914: all right divisors of~$\xx$ of the form~$\yy_2 \yy_1$ with

915: $\yy_\kk \in \MM_\kk$, \ie, lying in $\MM_2\MM_1$. This is

916: not the case. Consider Example~\ref{X:Delta} again, and let

917: $\xx = \s_1^\ee\ss2\ss1$ with~$\ee\ge1$. Then the

918: $(\MM_2, \MM_1)$-decomposition of $\xx$ is $(\s_1^\ee,

919: \ss2, \ss1)$, so $\xx_2 \xx_1$ is $\ss2\ss1$ here. Now, we also

920: have $\xx = \ss2 \ss1 \s_2^\ee$, so $\s_2^\ee$, \ie,

921: $\s_2^\ee \cdot 1$, is a right divisor of~$\xx$ that belongs

922: to~$\MM_2\MM_1$ and does not divide $\ss2 \ss1$.

923: More generally, we see that the braids~$\ss\ii$ that are

924: right divisors of~$\xx$ cannot be retrieved from the last two

925: elements of the $(\MM_2, \MM_1)$-decomposition of~$\xx$.

926: \end{rema}

927: 

928: \begin{rema}

929: \label{R:Shortest}

930: Assume that $\MM$ is a locally right Garside monoid, and $(\MM_2, \MM_1)$ is a covering of~$\MM$.

931: Define an \emph{$(\MM_2, \MM_1)$-sequence} to be any finite

932: sequence $(\xx_\pp, \Ldots, \xx_1)$ such that $\xx_\rr$ belongs

933: to~$\MM_{\cl\rr}$ for each~$\rr$.

934: Then the $(\MM_2, \MM_1)$-decomposition of~$\xx$ is a

935: certain decomposition of~$\xx$ that is a $(\MM_2,

936: \MM_1)$-sequence. As we take the maximal right divisor at

937: each step, we might expect to obtain a short

938: $(\MM_2, \MM_1)$-sequence, possibly the shortest possible one.

939: We shall see in Section~\ref{S:Burckel} below that this is

940: indeed the case for the covering of Example~\ref{X:Delta}.

941: However, this is not the case in general.

942: Indeed, keep the braid monoid~$\BB4$, but consider

943: the covering $(\MM'_2, \MM'_1)$, where $\MM'_1$

944: (\resp $\MM'_2$) is the submonoid generated by~$\ss1$

945: and~$\ss3$ (\resp by~$\ss2$ and~$\ss3$). Let $\xx$ be $\ss3\ss2\ss1\ss2\s_3^2\ss2$. The

946: $(\MM'_2,\MM'_1)$-decomposition of~$\xx$ turns out to be

947: $(\ss1 , \s_2^2 , \ss3\ss1 , \ss2 , \s_1)$, a sequence of

948: length~$5$, but another decomposition of~$\xx$ is the

949: $(\MM'_2, \MM'_1)$-sequence 

950: $(\ss3\ss2, \ss1, \ss2\s_3^2\ss2, 1)$, which has length~$4$: choosing the

951: maximal right divisor at each step does \emph{not} guarantee that we obtain the

952: shortest sequence.

953: \end{rema}

954: 

955: Finally, it should be clear that, instead of

956: considering two closed submonoids $\MM_2,\MM_1$

957: of~$\MM$, we could consider any finite family of such

958: submonoids $\MM_{\mm}, \Ldots, \MM_1$. Provided the union

959: of all~$\MM_\jj$'s generates~$\MM$, we can extend

960: Proposition~\ref{P:AltDec} and obtain for every

961: element~$\xx$ of~$\MM$ a distinguished decomposition

962: $(\xx_\pp, \Ldots, \xx_1)$ such that $\xx_\rr$ belongs to

963: $\MM_{\cl\rr}$ and

964: $\xx_p...\xx_{\rr+1}\perp\MM_{\cl\rr}$ holds for

965: every~$\rr$, where $\cl\rr$ now denotes the unique

966: element of~$\{1, \Ldots, \mm\}$ that equals~$\rr$

967: mod~$\mm$. The only difference is that

968: the condition $\xx_\rr\not=1$ for $\rr\ge2$ has to be relaxed to

969: $\xx_{\rr+\mm-2}...\xx_\rr\not=1$ for $\rr \ge \mm$, since

970: the conjunction of $\xx\not=1$ and

971: $\xx\perp\MM_{\cl\rr}$ need not guarantee 

972: $\xx\not\perp\MM_{\cl{\rr+1}}$, but only

973: $\xx\not\perp

974: (\MM_{\cl{\rr+\mm-1}}\cup...\cup\MM_{\cl{\rr+1}})$. 

975: Adapting is easy---see~\cite{Fro} for an example.

976: 

977: \subsection{Algorithmic aspects}

978: 

979: Computing the alternating decomposition is easy provided one can efficiently

980: perform right division in the ground monoid. To give a precise statement, we

981: recall from~\cite{Dfx} the notion of word norm (or pseudolength)

982: that generalizes the standard notion of word length. In the

983: sequel, for $\SS$ included in~$\MM$ and~$\ww$ a word on~$S$, we denote by~$\CL\ww$ the element of~$\MM$

984: represented by~$\ww$.

985: 

986: \begin{defi}

987: \label{D:Norm}

988: Assume that $\MM$ is a locally right Garside monoid that satisfies Condition~$(C_3^+)$, and $\SS$

989: generates~$\MM$. For~$\ww$ a word on~$\SS$, we denote

990: by~$\norm\ww$ the maximal length of a word~$\ww'$

991: satisfying~$\CL{\ww'}=\CL\ww$. 

992: \end{defi}

993: 

994: Condition~$(C_3^+)$ is precisely what is needed to guarantee that $\norm\ww$ exists for every word~$\ww$.

995: Indeed, if $\lambda: \MM \to \Nat$ witnesses that $(C_3^+)$ is satisfied, then every word~$\ww'$

996: satisfying~$\CL{\ww'}=\CL\ww$ must satisfy $\lg\ww \le \l(\CL\ww)$. Conversely, if $\norm\ww$ exists for each

997: word~$\ww$, then the map $\ww \mapsto \norm\ww$ induces a well-defined map of~$\MM$ to~$\Nat$ that

998: witnesses~$(C_3^+)$. In the case of Artin--Tits monoids and, more generally, of monoids presented by homogeneous

999: relations,  $\norm\ww$ coincides with the length~$\lg\ww$.

1000: 

1001: \begin{prop}

1002: \label{P:Complexity1}

1003: Assume that $\MM$ is a locally right Garside monoid,

1004: generated by some finite set~$\SS$, and satisfying

1005: Condition~$(C_3^+)$ plus:

1006: \begin{quote}

1007: $(*)$ There exists an algorithm~$\AlgoD$ that, for~$\ww$ a word on~$\SS$ and $\xs$ in~$\SS$, runs in

1008: time~$O(\norm\ww)$, recognizes whether

1009: $\CL\ww\multe\xs$ holds and, if so, returns a word

1010: representing~$\CL\ww\OVER{\xs}$. 

1011: \end{quote} 

1012: Let $\SS_2,\SS_1\ince\SS$ satisfying $\SS_2\cup\SS_1=\SS$. Let

1013: $\MM_\kk$ be the submonoid of~$\MM$ generated

1014: by~$\SS_\kk$, and suppose that $\MM_2, \MM_1$ are closed. 

1015: Then there exists a algorithm that, for~$\ww$

1016: a word on~$\SS$, runs in time~$O(\norm\ww^2)$ and

1017: computes the $(\MM_2,\MM_1)$-decomposition

1018: of~$\CL\ww$.

1019: \end{prop}

1020: 

1021: \begin{proof}

1022: Having listed the elements of~$\SS_1$ and~$\SS_2$, and

1023: starting with~$\ww$, we use~$\AlgoD$ to divide by

1024: elements of~$\SS_1$ until division fails, then we divide by

1025: elements of~$\SS_2$ until division fails, \etc\ We stop when

1026: the remainder is~$1$. If we start with a word~$\ww$

1027: satisfying $\norm\ww = \ell$, then the words~$\ww_\rr$

1028: subsequently occurring represent the elements~$\xx^{(\rr)}$

1029: of~\eqref{E:AltDec2}, which are left divisors of~$\xx$, and,

1030: hence, we have $\lg{\ww_\rr} \le \norm{\ww_\rr} \le

1031: \ell$. Moreover, at each step, $\norm{\ww_\rr}$ decreases by at least~$1$,

1032: so termination occurs after at most

1033: $\card(\SS) \times \ell$~division steps. By hypothesis, the

1034: cost of each division step is bounded above by~$O(\ell)$,

1035: whence a quadratic global upper bound.

1036: \end{proof}

1037: 

1038: \begin{exam}

1039: Let $\MM$ be an Artin--Tits of spherical type, or, more

1040: generally, a Garside monoid, and let $\SS$ be the set of

1041: atoms in~$\MM$. Then there exist division

1042: algorithms running in linear time, \emph{e.g.}, those

1043: involving a rational transducer based on the (right)

1044: automatic structure~\cite{Eps}. Alternatively, for the specific

1045: question of dividing by an atom, the reversing method

1046: of~\cite{Dgp} is specially convenient.

1047: \end{exam}

1048: 

1049: \section{Iterated alternating decompositions}

1050: \label{S:Iter}

1051: 

1052: If the submonoids involved in a covering are monogenerated, it makes no sense to iterated the alternating

1053: decomposition. But, in general, for instance in the case of Example~\ref{X:Delta}, the covering submonoids

1054: need not be monogenerated, and they can in turn be covered by smaller submonoids. In such cases, it is natural to

1055: iterate the alternating decomposition using a sequence of nested coverings. This is the idea we develop in this

1056: section. The main observation is that the result of the iterated decomposition can be obtained directly, without any

1057: iteration.

1058: 

1059: \subsection{Iterated coverings}

1060: \label{S:IterCov}

1061: 

1062: The possibility of iterating the alternating decomposition

1063: relies on the following trivial observation:

1064: 

1065: \begin{lemm}

1066: \label{L:Sub}

1067: Every closed submonoid of a locally right Garside monoid is

1068: locally right Garside.

1069: \end{lemm}

1070: 

1071: \begin{proof}

1072: Assume that $\MM_1$ is a closed submonoid of

1073: a locally right Garside monoid~$\MM$. First,

1074: $\MM_1$ admits right cancellation as every submonoid of a

1075: right cancellative monoid does. Then, if $\xx,

1076: \yy$ belong to~$\MM_1$ and admit a common left

1077: multiple~$\zz$ in~$\MM_1$, then $\zz$ is a common left

1078: multiple of~$\xx$ and~$\yy$ in~$\MM$, so, in~$\MM$,

1079: the left lcm~$\zz'$ of~$\xx$ and~$\yy$ exists. The

1080: hypothesis that $\MM_1$ is closed under left lcm implies

1081: $\zz' \in \MM_1$, and, then,

1082: $\zz'$ must be a left lcm for~$\xx$ and~$\yy$ in the sense

1083: of~$\MM_1$. Finally, the right divisibility

1084: relation of~$\MM_1$ is included in the right divisibility

1085: relation of~$\MM$, so a sequence contradicting

1086: Condition~$(C_3)$ in~$\MM_1$ would also

1087: contradict~$(C_3)$ in~$\MM$. 

1088: \end{proof}

1089: 

1090: Assume that $\MM$ is a locally right Garside monoid and $(\MM_2,\MM_1)$ is a covering of~$\MM$.

1091: By Lemma~\ref{L:Sub}, $\MM_2$ and $\MM_1$ are locally

1092: right Garside, and we can repeat the process: assuming

1093: that $(\MM_{\kk,2},\MM_{\kk,1})$ is a covering of~$\MM_\kk$

1094: for~$\kk=2,1$, every element of~$\MM_\kk$ admits a

1095: $(\MM_{\kk,2},\MM_{\kk,1})$-decomposition, and,

1096: therefore, every element of~$\MM$ admits a distinguished

1097: decomposition in terms of the four

1098: monoids~$\MM_{22}$, $\MM_{21}$,

1099: $\MM_{12}$, and $\MM_{22}$---we drop commas in

1100: indices.

1101: 

1102: \begin{exam}

1103: \label{X:IterCov}

1104: As in Example~\ref{X:Delta}, consider the $4$-strand

1105: braid monoid~$\BB4$, and let $\MM_2, \MM_1$ be the

1106: parabolic submonoids respectively generated

1107: by~$\ss3, \ss2$, and by $\ss2, \ss1$. Then let $\MM_{22}$,

1108: $\MM_{21}$, $\MM_{12}$, and $\MM_{11}$ be the

1109: submonoids respectively generated by~$\ss2$, $\ss3$,

1110: $\ss2$, and~$\ss1$. Then  $(\MM_{\kk2},\MM_{\kk1})$ is

1111: a covering of~$\MM_\kk$ for~$\kk=2,1$.

1112: \end{exam}

1113: 

1114: To make the construction formal, we introduce the notion of an iterated covering.

1115: 

1116: \begin{defi}

1117: Assume that $\MM$ is a locally right Garside monoid. We say that $\MM$ is a \emph{$0$-covering} of itself, and

1118: then, for $\nn \ge 1$, we define an \emph{$\nn$-covering} of~$\MM$ to be a pair $(\MMM_2, \MMM_1)$

1119: such that there exists a covering $(\MM_2, \MM_1)$ of~$\MM$ such that $\MMM_\kk$ is an $(\nno)$-covering

1120: of~$\MM_\kk$ for $\kk = 1,2$.

1121: \end{defi}

1122: 

1123: So a $1$-covering of~$\MM$ is just an ordinary covering, and, for instance, a $2$-covering of~$\MM$ consists of a

1124: covering $(\MM_2, \MM_1)$ of~$\MM$, plus coverings of~$\MM_2$ and~$\MM_1$, as in Example~\ref{X:IterCov}.

1125: 

1126: An iterated covering of a monoid~$\MM$ has the structure of a binary tree, and we can specify the various

1127: submonoids by using finite sequences of twos and ones---or of ones and zeroes, or of letters `L' and `R'---to

1128: indicate at each forking which direction is to be taken. In the sequel, such a finite sequence of length~$\nn$ is called

1129: a \emph{binary $\nn$-address}. In this way, an $\nn$-covering of a monoid~$\MM$ is a sequence of submonoids

1130: $\MM_\a$ indexed by binary addresses of length at most~$\nn$, such that, for each~$\a$ of length smaller

1131: than~$\nn$, the pair~$(\MM_{\a2}, \MM_{\a1})$ is a covering of~$\MM_\a$, and $\MM_\ea$ is~$\MM$---using

1132: $\ea$ for the empty address. In the sequel, if $\MMM$ is an iterated covering, we shall always use~$\MM_\a$ for

1133: the $\a$-entry in~$\MMM$.

1134: 

1135: If the ground monoid~$\MM$ has some distinguished generating set~$\SS$, we can specify an

1136: $\nn$-covering by choosing a subset~$\SS_\a$ of~$\SS$ for

1137: each~$\a$ in~$\{2,1\}^\nn$, and, for~$\b$ in~$\{2,1\}^\nn$ with $\mm\le\nn$, defining~$\MM_\b$ to be

1138: the submonoid generated by all~$\SS_{\a}$'s such that $\b$ is a

1139: prefix of~$\a$. We obtain an $\nn$-covering provided each

1140: submonoid~$\MM_\b$ is closed. For such coverings, we can 

1141: display the inclusions in a binary tree---see Figure~\ref{F:Skeleton}.

1142: 

1143: \begin{figure}[htb]

1144: \begin{picture}(30,22)

1145: %\Grid

1146: \put(0,-1.5){\includegraphics{Skeleton.eps}}

1147: \put(0,0){$\ss2$}

1148: \put(10,0){$\ss3$}

1149: \put(20,0){$\ss2$}

1150: \put(30,0){$\ss1$}

1151: \put(2,10){$\ss2,\!\ss3$}

1152: \put(22,10){$\ss1,\!\ss2$}

1153: \put(10,20){$\ss1,\!\ss2,\!\ss3$}

1154: \end{picture}

1155: \caption{\sf Skeleton of the $2$-covering of~$\BB4$  of Example~\ref{X:Iter}: a depth~$2$ binary

1156: tree displaying the inclusions between the generating sets of the successive submonoids; this example corresponds

1157: to $\SS_{22}=\SS_{12}=\{\ss2\}$, $\SS_{21}=\{\ss3\}$, and $\SS_{11}=\{\ss1\}$;  we find for instance $\MM_2 =

1158: \Mon{\ss2,\ss3}$, and $\MM_{12} = \Mon{\ss2}$.}

1159: \label{F:Skeleton}

1160: \end{figure}

1161: 

1162: \subsection{Iterated $\MMM$-decomposition}

1163: \label{S:IterDec}

1164: 

1165: As was shown in Section~\ref{S:Alt}, each covering $(\MM_2, \MM_1)$ of a monoid~$\MM$ leads to a

1166: distinguished decomposition for the elements of~$\MM$ in terms of elements of~$\MM_2$ and~$\MM_1$. An

1167: iterated covering similarly leads to what can be called an iterated decomposition.

1168: 

1169: \begin{defi}

1170: \label{D:IterDec}

1171: Assume that $\MM$ is a locally right Garside monoid, and $\MMM$ is an $\nn$-covering of~$\MM$. For

1172: $\xx$ in~$\MM$, we define the \emph{$\MMM$-decomposition}~$\DD\MMM\xx$ of~$\xx$ by $\DD\MMM\xx=\xx$ for

1173: $\nn=0$, and, for $\nn \ge 1$ and $\MMM = (\MMM_2, \MMM_1)$, by

1174: \begin{equation}

1175: \label{E:IterDec}

1176: \DD\MMM\xx=

1177: (\DD{\MMM_{\cl\pp}}{\xx_\pp} \,,\, ... \,,\,

1178: \DD{\MMM_1}{\xx_1}),

1179: \end{equation}

1180: where $(\xx_\pp, \Ldots, \xx_1)$ is the $(\MM_2,

1181: \MM_1)$-decomposition of~$\xx$. 

1182: \end{defi}

1183: 

1184: \begin{exam}

1185: \label{X:Iter}

1186: Consider the braid~$\D_4^2$ of $\BB4$ and the covering~$\MMM$ of Example~\ref{X:IterCov}. We saw in

1187: Example~\ref{X:Delta} that the $(\MM_2, \MM_1)$-decomposition of~$\D_4^2$ is

1188: $$(\ss3, \ss2\s_1^2,  \ss2\ss3,  \D_3^2).$$

1189: Now, the $(\MM_{12}, \MM_{11})$-decomposition of~$\D_3^2$ turns out to be

1190: $(\ss2, \s_1^2,\ss2,\s_1^2)$. Similarly, the $(\MM_{22}, \MM_{21})$-decomposition of~$\ss2\ss3$

1191: is $(\ss2, \ss3)$. Continuing in this way, we obtain 

1192: \begin{equation}

1193: \label{E:XIterDec1}

1194: \DD\MMM{\D_4^2} = ((\ss3), (\ss2,\s_1^2),(\ss2,\ss3),

1195: (\ss2, \s_1^2,\ss2,\s_1^2)), 

1196: \end{equation}

1197: corresponding to the factorization

1198: $\D_4^2 = (\ss3)\,\cdot\, (\ss2 \cdot \s_1^2)\,\cdot\, (\ss2

1199: \cdot \ss3)\,\cdot\, (\ss2\cdot\s_1^2\cdot \ss2 \cdot

1200: \s_1^2)$.

1201: \end{exam}

1202: 

1203: For $\nn \ge 2$, the $\MMM$-decomposition of an element is a sequence of sequences. More precisely, it is an

1204: \emph{$\nn$-sequence}, defined to be a single element for $\nn = 0$, and to be a sequence of $(\nno)$-sequences

1205: for $\nn \ge 1$. Such iterated sequences can naturally be viewed as trees, on the model of Figure~\ref{F:Tree} (left).

1206: 

1207: Entries in an ordinary sequence of length~$\pp$ are usually specified using numbers from~$1$

1208: to~$\pp$---or rather $\pp$ to~$1$ in the context of this paper where we start from the right. Entries in an iterated

1209: sequence are then specified using finite sequence of numbers, as done in Section~\ref{S:IterCov} with binary

1210: addresses. In the sequel, a length~$\nn$ sequence of positive numbers is called an \emph{$\nn$-address}: for

1211: instance, $32$ is a typical

1212: $2$-address---in examples, we drop brackets and separating commas. If $\www$ is an $\nn$-sequence, and $\h$ is

1213: an $\mm$-address with $\mm \le \nn$, we denote by~$\www_\h$ the $\h$-subsequence of~$\www$, \ie, the

1214: $(\nn-\mm)$-sequence made by those entries in~$\www$ whose address begins with~$\h$---when it exists, \ie,

1215: when the considered sequences are long enough---see Figure~\ref{F:Tree} (right). 

1216: 

1217: \begin{figure}[htb]

1218: \begin{picture}(95,27)

1219: \put(-1,3.3){\includegraphics{Tree.eps}}

1220: \put(49.2,3){\includegraphics{Tree.eps}}

1221: \put(0,0){$\ss3$}

1222: \put(5,0){$\ss2$}

1223: \put(10,0){$\s_1^2$}

1224: \put(15,0){$\ss2$}

1225: \put(20,0){$\ss3$}

1226: \put(25,0){$\ss2$}

1227: \put(30,0){$\s_1^2$}

1228: \put(35,0){$\ss2$}

1229: \put(40,0){$\s_1^2$}

1230: \put(0,12){$\ss3$}

1231: \put(7,12){$\ss2\s_1^2$}

1232: \put(17,12){$\ss2\ss3$}

1233: \put(33,12){$\D_3^2$}

1234: \put(19,23){$\D_4^2$}

1235: \put(49,0){$41$}

1236: \put(54,0){$32$}

1237: \put(59,0){$31$}

1238: \put(64.5,0){$22$}

1239: \put(69.5,0){$21$}

1240: \put(74.5,0){$14$}

1241: \put(79.5,0){$13$}

1242: \put(84.5,0){$12$}

1243: \put(90,0){$11$}

1244: \put(50,12){$4$}

1245: \put(60,12){$3$}

1246: \put(70,12){$2$}

1247: \put(85,12){$1$}

1248: \put(70,23){$\ea$}

1249: \end{picture}

1250: \caption{\sf The tree associated with the

1251: $2$-sequence of~\eqref{E:XIterDec1}: on the

1252: left,  the braid entries, on the right, the addresses; the entry list specifies

1253: the name of the leaves, while the address

1254: list specifies the shape of the tree; for each address~$\h$, the $\h$-subsequence~$\www_\h$ corresponds to what

1255: lies below~$\h$ in~$\www$; here, the

1256: $31$-subsequence is~$\s_1^2$, while the $2$-subsequence

1257: is~$(\ss2, \ss3)$. The $23$-subsequence does not exist.}

1258: \label{F:Tree}

1259: \end{figure}

1260: 

1261: Note that addresses are just a way of specifying brackets in an iterated sequence: an

1262: $\nn$-sequence is determined by its unbracketing---that is, the (ordinary) sequence obtained by removing all

1263: inner brackets---and its address list. For instance, in the $2$-sequence of~\eqref{E:XIterDec1}, the unbracketing and

1264: the address list are

1265: \begin{equation}

1266: \label{E:XIterDec3}

1267: (\ss3, \ss2,\s_1^2, \ss2,\ss3, \ss2, \s_1^2,\ss2,\s_1^2)

1268: \text{\quad and \quad}

1269: (41,32,31,22,21,14,13, 12,11).

1270: \end{equation}

1271: 

1272: Assume that $\www$ is the $\MMM$-decomposition of an element~$\xx$. For each~$\h$ that is the address of a

1273: node of~$\www$ (viewed as a tree), write $\xx_\h$ for the product of the subsequence~$\www_\h$. Then, by

1274: definition, if

1275: $\h$ is the address of an inner node and $\h\pp, \Ldots,

1276: \h1$ are the addresses of the nodes that lie immediately below~$\h$ in~$\www$, the sequence

1277: $(\xx_{\h\pp}, \Ldots, \xx_{\h1})$ is the $(\MM_{\cl\h2}, \MM_{\cl\h1})$-decomposition of~$\xx_\h$, where 

1278: $\cl\h$ denotes the binary address obtained by replacing each~$\rr$ ocurring in~$\h$ with~$\cl{\rr}$---which is

1279: coherent with Notation~\ref{N:Parity}. Applying Proposition~\ref{P:AltDec} immediately gives the following

1280: characterization.

1281: 

1282: \begin{prop}

1283: \label{P:IterDec}

1284: Assume that $\MM$ is a locally right Garside monoid, $\MMM$ is an $\nn$-covering of~$\MM$, and $\www =

1285: \DD\MMM\xx$. For each address~$\h$ in~$\www$, let

1286: $\xx_\h$ denote the product of~$\www_\h$. Assume that $\h$ is the address of an inner node and

1287: $\h\pp, \Ldots, \h1$ are the addresses of the nodes that lie immediately below~$\h$ in~$\www$. Then, the

1288: elements~$\xx_{\h\rr}$ are determined from

1289: $\xx_\h^{(0)}= \xx_\h$ by

1290: \begin{equation}

1291: \label{E:IterDec1}

1292: \xx_{\h\rr}=\Tail{\xx_\h^{(\rr-1)}}{\MM_{\cl{\h\rr}}}

1293: \text{\quad and\quad}

1294: \xx_\h^{(\rr)} = \xx_\h^{(\rr-1)}\OVER{\xx_{\h\rr}}.

1295: \end{equation}

1296: \end{prop}

1297: 

1298: \begin{exam}

1299: \label{X:Complicated}

1300: In the context of Example~\ref{X:Iter} and Figure~\ref{F:Tree}, \eqref{E:IterDec1} gives

1301: $$\xx_1=

1302: \ss2\s_1^2\ss2\s_1^2 = 

1303: \Tail\xx{\BB3}, \quad

1304: \xx_2= \ss2\ss3 = 

1305: \Tail{\ss3\ss2\s_1^2\ss2\ss3}{\Mon{\ss2,\ss3}}, \etc$$

1306: which involve the whole of~$\xx$, but also, at the next level,

1307: we have

1308: $$\xx_{11}= \s_1^2 = 

1309: \Tail{\ss2\s_1^2\ss2\s_1^2}{\BB2}, \quad

1310: \xx_{12}= \ss2 = 

1311: \Tail{\ss2\s_1^2\ss2}{\Mon{\ss2}}, \etc

1312: $$

1313: which only involve the element

1314: $\xx_1$, namely~$\ss2\s_1^2\ss2\s_1^2$, and not the

1315: whole of~$\xx$.

1316: \end{exam}

1317: 

1318: \subsection{A transitivity lemma}

1319: \label{S:Iter2}

1320: 

1321: Proposition~\ref{P:IterDec} looks intricate, and it is not

1322: satisfactory in that it does not give a global characterization

1323: of the $\MMM$-decomposition and a way to obtain it directly. This is

1324: what we shall do now. The point is that, according to the following result, there is no

1325: need to consider local remainders when computing

1326: iterated tails.

1327: 

1328: \begin{lemm}

1329: \label{L:BiTail}

1330: Assume that $\MM$ is a locally right Garside monoid, that $\MM_1$ is a closed submonoid of~$\MM$, and

1331: that $\MM_{11}$ is a closed submonoid of~$\MM_1$. Then,

1332: for every~$\zz$ in~$\MM$ and every left divisor~$\yy$

1333: of~$\Tail\zz{\MM_1}$, we have

1334: \begin{equation}

1335: \label{E:Bitail}

1336: \Tail{(\zz\OVER{\Tail\zz{\MM_1}})\yy}{\MM_{11}}.

1337: = \Tail{\yy}{\MM_{11}}.

1338: \end{equation}

1339: \end{lemm}

1340: 

1341: \begin{proof}

1342: Put $\zz_1= \Tail\zz{\MM_1}$ and $\zz'=\zz\OVER{\zz_1}$. By

1343: definition,

1344: $\Tail{\yy}{\MM_{11}}$ is a right divisor of

1345: $\Tail{\zz'\yy}{\MM_{11}}$, hence the point is to prove

1346: that every right divisor of~$\zz'\yy$ lying

1347: in~$\MM_{11}$ is a right divisor of~$\yy$. So assume

1348: $\zz'\yy=\xx'\xx$ with

1349: $\xx\in\MM_{11}$. By hypothesis, we have

1350: $\zz_1=\yy\zz'_1$ for some~$\zz'_1$, necessarily lying

1351: in~$\MM_1$. Then, we have $\zz= \zz'\zz_1 =

1352: \zz'\yy\zz'_1 = \xx'\xx\zz'_1$. Now $\xx\in\MM_{11}$

1353: implies $\xx\in\MM_1$, hence $\xx\zz'_1\in\MM_1$, and

1354: $\xx\zz'_1$ has to be a right divisor of~$\Tail\zz{\MM_1}$,

1355: \ie, of~$\zz_1$, which is also~$\yy\zz'_1$. It follows that

1356: $\xx$ is a right divisor of~$\yy$, as expected.

1357: \end{proof}

1358: 

1359: In particular, when we choose~$\yy$ to be~$\zz_1$ itself,

1360: \eqref{E:Bitail} gives

1361: \begin{equation}

1362: \label{E:BiTail2}

1363: \Tail\zz{\MM_{11}}=\Tail{\Tail\zz{\MM_1}}{\MM_{11}},

1364: \end{equation}

1365: which is vaguely reminiscent of the equality

1366: $\Tail{\zz\yy}\Sigma = 

1367: \Tail{\Tail\zz\Sigma\yy}\Sigma$ that is crucial

1368: in the construction of the right greedy normal form in a Garside

1369: monoid.

1370: 

1371: \subsection{Global characterization of the iterated decomposition}

1372: \label{S:Global}

1373: 

1374: We shall now give a direct description of the

1375: $\MMM$-decomposition not involving the

1376: intermediate values~$\xx_\h$. Consider Examples~\ref{X:Iter} and~\ref{X:Complicated} again. The

1377: problem is as follows: in the case of the $1$-covering

1378: of~$\BB3$, only two submonoids are involved, and the final

1379: decomposition consists of alternating blocks belonging to

1380: each of them; in the case of the $2$-covering

1381: of~$\BB4$, the decomposition consists of blocks

1382: of~$\ss1$'s, $\ss2$'s, and~$\ss3$'s, but the order in which

1383: these blocks appear is not so simple. Indeed, on the left of a

1384: block of~$\ss2$'s, there may be either a block of~$\ss1$'s or

1385: a block of~$\ss3$'s, depending on the

1386: current address, \ie, on the position in

1387: (the skeleton of) the covering, typically on which of the

1388: two occurrences of~$\ss2$ in the tree of

1389: Figure~\ref{F:Skeleton} the considered block of~$\ss2$'s is

1390: to be associated: on the left of a block of~$\ss2$'s associated

1391: with the rightmost~$\ss2$ in Figure~\ref{F:Skeleton}, $\ss1$ is expected, while $\ss3$

1392: is expected in the other case. This is what

1393: Proposition~\ref{P:GlobalIter} below says, namely that the

1394: $\MMM$-decomposition can be obtained directly provided

1395: we keep track of some position specified by a binary address. 

1396: 

1397: To make the description precise, we introduce the notion of

1398: successors of an address. It comes in two versions, one for

1399: general addresses, one for binary addresses. 

1400: 

1401: \begin{defi}

1402: \label{D:Successor}

1403: For $\h$ an $\nn$-address and $0\le\mm\le\nn$, the

1404: \emph{$\mm$-successor}~$\Succ\mm\h$ of~$\h$ is 

1405: the $\nn$-address obtained by keeping the first

1406: $\mm$ digits of~$\h$, adding~$1$ to the next one, and

1407: completing with~$1$'s, \ie, for $\h=\dd_1...\dd_\nn$,

1408: the $\mm$-successor is $\dd'_1...\dd'_\nn$ with

1409: $\dd'_\rr=\dd_\rr$ for $\rr\le\mm$, and, if $\mm<\nn$

1410: holds,  $\dd'_{\mm+1}=\dd_{\mm+1}+1$ and $\dd'_\rr=1$

1411: for $\rr>\mm+1$. For $\a$ a binary $\nn$-address, the

1412: \emph{binary $\mm$-successor}~$\BSucc\mm\a$ of~$\a$ is

1413: defined to be~$\cl{\Succ\mm\a}$. 

1414: \end{defi}

1415: 

1416: \begin{exam}

1417: Let $\h=3612$. The successors of~$\h$ are

1418: $$\Succ0\h=4111, \quad

1419: \Succ1\h=3711, \quad

1420: \Succ2\h=3621, \quad

1421: \Succ3\h=3613, \quad

1422: \Succ4\h=3612.$$

1423: Similarly, the binary successors of~$\a=1212$ are

1424: $$\BSucc0\a=2111, \quad

1425: \BSucc1\a=1111, \quad

1426: \BSucc2\a=1221, \quad

1427: \BSucc3\a=1211, \quad

1428: \BSucc4\a=1212.$$

1429: \end{exam}

1430: 

1431: Note that $\Succ\nn\h=\h$ holds for every $\nn$-address~$\h$. We recall that specifying an iterated sequence

1432: amounts to specifying both its unbracketing and its address list.

1433: 

1434: \begin{prop}

1435: \label{P:GlobalIter}

1436: Assume that $\MM$ is a locally right Garside monoid, and $\MMM$ is an $\nn$-covering of~$\MM$.

1437: Then, for~$\xx$ in~$\MM$, the unbracketing

1438: $(\xx_\pp, \Ldots, \xx_1)$ and the address list $(\h_\pp, \Ldots,

1439: \h_1)$ of~$\DD\MMM\xx$ are inductively determined

1440: from

1441: $\xx^{(0)}=\xx$ and $\h_1=1^\nn$ by

1442: \begin{equation}

1443: \label{E:GlobalIter}

1444: \xx_\rr= 

1445: \Tail{\xx^{(\rr-1)}}{\MM_{\cl{\h_\rr}}}

1446: \text{\ , \quad}

1447: \xx^{(\rr)}=\xx^{(\rr-1)}\OVER{\xx_\rr}

1448: \text{\ , \quad and\quad}

1449: \h_{\rr+1}=\Succ\mm{\h_\rr},

1450: \end{equation}

1451: where $\mm$ is the length of the longest prefix~$\h$

1452: of~$\h_\rr$ that satisfies $\xx^{(\rr)} \not\perp\MM_{\cl\h}$.

1453: \end{prop}

1454: 

1455: \begin{proof}

1456: As can be expected, we use an induction on~$\nn$. The argument relies on the

1457: transivity relation of Lemma~\ref{L:BiTail}.

1458: 

1459: For $\nn=0$, everything is

1460: trivial, and, for $\nn=1$, the result is a restatement of

1461: Proposition~\ref{P:AltDec}: in this case, the

1462: $1$-address~$\h_\rr$ is~$\rr$, the longest

1463: prefix of~$\h_\rr$ satisfying $\xx^{(\rr)} \not\perp

1464: \MM_{\cl\h}$ is~$\ea$, and the induction  rule

1465: reduces to $\h_{\rr+1}=\rr+1$.

1466: 

1467: Assume $\nn\ge2$. Let $(\yy_\qq, \Ldots, \yy_1)$ be the

1468: $(\MM_2, \MM_1)$-decomposition of~$\xx$. By

1469: definition, we have

1470: \begin{equation}

1471: \label{E:GlobalDec1}

1472: \DD\MMM\xx=

1473: (\DD{\MMM_{\cl\qq}}{\yy_\qq} \Ldots, 

1474: \DD{\MMM_1}{\yy_1}).

1475: \end{equation}

1476: For $\qq\ge\jj\ge1$, let $(\yy_{\jj,\pp_\jj}, \Ldots,

1477: \yy_{\jj,1})$ and

1478: $(\h_{\jj,\pp_\jj}, \Ldots, \h_{\jj,1})$ be the unbracketing and

1479: the address list in~$\DD{\MMM_{\cl\jj}}{\yy_\jj}$.

1480: Then, by~\eqref{E:GlobalDec1}, we have

1481: \begin{equation}

1482: \label{E:Concat1}

1483: (\xx_\pp, \Ldots, \xx_1) = (\yy_{\qq,\pp_\qq}, \Ldots,

1484: \yy_{\qq,1}) \conc ... \conc (\yy_{1,\pp_1}, \Ldots,

1485: \yy_{1,1}),

1486: \end{equation}

1487: where $\conc$ denotes concatenation, and, similarly,

1488: \begin{equation}

1489: \label{E:Concat2}

1490: (\h_\pp, \Ldots, \h_1) = (\qq\h_{\qq,\pp_\qq}, \Ldots,

1491: \qq\h_{\qq,1}) \conc ... \conc (1\h_{1,\pp_1}, \Ldots,

1492: 1\h_{1,1}).

1493: \end{equation}

1494: By induction hypothesis, the sequences of~$\yy_\jj$'s

1495: and~$\h_{\jj,\kk}$'s satisfy the counterpart

1496: of~\eqref{E:GlobalIter}, and we wish to

1497: deduce~\eqref{E:GlobalIter}, \ie, dropping the

1498: elements~$\xx^{(\rr)}$, to prove

1499: $$\xx_\rr = \Tail{\xx_\pp ... \xx_\rr}{\MM_{\h_\rr}}

1500: \text{\quad and\quad}

1501: \h_{\rr+1}=\Succ\mm{\h_\rr}$$ where $\mm$ is the

1502: length of the maximal prefix~$\h$ of~$\h_\rr$ satisfying

1503: $\xx_\pp ... \xx_{\rr+1} \not\perp \MM_{\cl\h}$.

1504: We use induction on $\rr\ge1$.

1505: 

1506: Assume that $\xx_\rr$ corresponds

1507: to some entry~$\yy_{\jj,\kk}$ in~\eqref{E:Concat1}.  By

1508: construction, we have $\h_\rr=\jj\h_{\jj,\kk}$.

1509: Let $\yy =\yy_{\jj,\pp_\jj}...\yy_{\jj,\kk}$. The induction

1510: hypothesis gives

1511: \begin{equation}

1512: \label{E:Tail1}

1513: \xx_\rr = \yy_{\jj,\kk} =

1514: \Tail{\yy}{\MM_{\cl{\jj\h_{\jj,\kk}}}} =

1515: \Tail{\yy}{\MM_{\cl{\h_\rr}}}.

1516: \end{equation}

1517: On the other hand, by construction, $\yy$ is a left divisor of

1518: $\yy_{\jj,\pp_\jj}...\yy_{\jj,1}$, \ie, of~$\yy_\jj$, and

1519: $\yy_\jj$ is the $\MM_{\cl\jj}$-tail of

1520: $\yy_\qq...\yy_\jj$, \ie, putting $\zz=\yy_\qq ...

1521: \yy_\jj$, we have

1522: \begin{equation}

1523: \label{E:Tail2}

1524: \yy_\jj = \Tail{\zz}{\MM_{\cl\jj}}.

1525: \end{equation}

1526: Applying Lemma~\ref{L:BiTail} to the

1527: monoids $\MM_{\cl{\h_\rr}} \subseteq \MM_{\cl\jj}

1528: \subseteq \MM$, we deduce from~\eqref{E:Tail1}

1529: and~\eqref{E:Tail2} the relation

1530: $\xx_\rr = \Tail{(\zz\OVER{\yy_\jj})\yy}{\MM_{\cl{\h_\rr}}}$,

1531: which is $\xx_\rr = \Tail{\xx_\pp ...

1532: \xx_\rr}{\MM_{\cl{\h_\rr}}}$, as, by construction,

1533: we have $(\zz\OVER{\yy_\jj})\yy = \xx_\pp...\xx_\rr$.

1534: 

1535: Consider now~$\h_{\rr+1}$. Two

1536: cases are possible, according to whether $\xx_\rr$

1537: corresponds to an initial or a non-initial entry in

1538: some sequence of~$\yy$'s, \ie, with the above notation,

1539: according to whether $\kk = \pp_\jj$ holds or not. 

1540: Assume first $\kk < \pp_\jj$. Then $\h_{\jj,\kk+1}$ exists, and

1541: the induction hypothesis implies that

1542: $\h_{\jj,\kk+1}$ is the $\mm$-successor of~$\h_{\jj,\kk}$,

1543: where $\mm$ is the length of the maximal prefix~$\h$

1544: of~$\h_{\jj,\kk}$ for which $\yy_{\jj,\pp_\jj}...

1545: \yy_{\jj,\kk+1} \not\perp \MM_{\cl{\jj\h}}$ holds. The latter

1546: relation is equivalent to $\xx_\pp ... \xx_{\rr+1} \not\perp

1547: \MM_{\cl{\jj\h}}$: indeed,

1548: $\xx\not\perp\AA$ is equivalent to $\Tail\xx\AA\not=1$,

1549: and, as above, Lemma~\ref{L:BiTail} implies

1550: $\Tail{\xx_\pp ... \xx_{\rr+1}}{\MM_{\cl{\jj\h}}}

1551: = \Tail{\yy_{\jj,\pp_\jj}...

1552: \yy_{\jj,\kk+1}}{\MM_{\cl{\jj\h}}}$. Therefore,

1553: $\h_{\rr+1}$, which is $\jj\h_{\jj,\kk+1}$, is the

1554: $\mm+1$-successor of~$\jj\h_{\jj,\kk}$, \ie, of~$\h_\rr$,

1555: where $\mm$ is the length of the maximal prefix~$\h$

1556: of~$\h_{\jj,\kk}$ for which $\xx_\pp ... \xx_{\rr+1} \not\perp

1557: \MM_{\cl{\jj\h}}$ holds, hence $\mm+1$ is the length of the

1558: maximal prefix~$\h'$ of~$\h_\rr$ (namely~$\jj\h$) for which 

1559: $\xx_\pp ... \xx_{\rr+1} \not\perp \MM_{\cl{\h'}}$ holds.

1560: 

1561: Finally, assume $\kk = \pp_\jj$, \ie, $\h_{\jj,\kk}$ is the

1562: leftmost address in the $\MMM_{\cl\jj}$-decomposition

1563: of~$\yy_\jj$. In this case, by hypothesis, we have

1564: $\h_{\rr+1}=(\jj+1)1^\nno$. Now, the hypothesis implies

1565: $\yy_\qq...\yy_{\jj+1} \perp \MM_{\cl\jj}$, \ie,

1566: $\xx_\pp...\xx_{\rr+1} \perp \MM_{\cl\jj}$. So, in

1567: this case, the only prefix~$\h$ of~$\h_\rr$, \ie,

1568: of~$\jj\h_{\jj,\pp_\jj}$, for which $\xx_\pp...\xx_{\rr+1}

1569: \not\perp \MM_{\cl\h}$ may hold is the empty

1570: address~$\ea$, which is the expected relation with

1571: $\mm=0$. 

1572: \end{proof}

1573: 

1574: \begin{exam}

1575: \label{X:IterBis}

1576: Consider the case of§ $\BB4$ and $\D_4^2$ again.

1577: Proposition~\ref{P:GlobalIter} directly gives

1578: the $\MMM$-decomposition of~$\D_4^2$ as follows. We

1579: start with $\xx = \D_4^2$ and $\h_1=11$. Then we compute

1580: $\MM_{11}$-tail, \ie, here the $\Mon{\ss1}$-tail,

1581: of~$\xx^{(0)}$, which turns out to be~$\s_1^2$, and call

1582: the quotient~$\xx^{(1)}$. Then the address~$\h_2$ is

1583: obtained by looking at the maximal prefix~$\h$ of~$\h_1$,

1584: \ie, of~$11$, for which $\MM_{\cl\h}\not\perp\xx^{(1)}$ holds.

1585: In the current case, we have

1586: $\xx^{(1)}\perp\MM_{11}$ and

1587: $\xx^{(1)}\not\perp\MM_1$, hence $\h=1$, so $\h_2$ is

1588: obtained from~$11$ by incrementing the second digit,

1589: leading to~$\h_2=12$, which corresponds to

1590: $\MM_{\cl{\h_2}}=\Mon{\ss2}$. We take the

1591: $\Mon{\ss2}$-tail of~$\xx^{(1)}$, call the

1592: remainder~$\xx^{(2)}$, and iterate. The successive values are

1593: displayed in Table~\ref{T:Iter}.

1594: \end{exam}

1595: 

1596: \begin{table}[htb]

1597: \begin{tabular}{c|lcccc}

1598: $\rr$\quad

1599: &$\xx^{(\rr)}$

1600: &$\h_\rr$

1601: &$\cl{\h_\rr}$

1602: &$\MM_{\cl{\h_\rr}}$

1603: &$\xx_\rr$

1604: \\

1605: \hline

1606: $0$

1607: &$\ss1\ss2\ss1\ss3\ss2\ss1\ss1\ss2\ss1\ss3\ss2\ss1$

1608: \rule{0pt}{12pt}

1609: \\

1610: $1$

1611: &$\ss2\ss1\ss3\ss2\ss1\ss1\ss2\ss3\ss1\ss2$

1612: &$11$&$11$

1613: &$\Mon{\ss1}$

1614: &$\s_1^2$

1615: \\

1616: $2$

1617: &$\ss2\ss1\ss3\ss2\ss1\ss1\ss2\ss3\ss1$

1618: &$12$&$12$

1619: &$\Mon{\ss2}$

1620: &$\ss2$

1621: \\

1622: $3$

1623: &$\ss2\ss3\ss2\ss1\ss1\ss2\ss3$

1624: &$13$&$11$

1625: &$\Mon{\ss1}$

1626: &$\s_1^2$

1627: \\

1628: $4$

1629: &$\ss3\ss2\ss1\ss1\ss2\ss3$

1630: &$14$&$12$

1631: &$\Mon{\ss2}$

1632: &$\ss2$

1633: \\

1634: $5$

1635: &$\ss3\ss2\ss1\ss1\ss2$

1636: &$21$&$21$

1637: &$\Mon{\ss3}$

1638: &$\ss3$

1639: \\

1640: $6$

1641: &$\ss3\ss2\ss1\ss1$

1642: &$22$&$22$

1643: &$\Mon{\ss2}$

1644: &$\ss2$

1645: \\

1646: $7$

1647: &$\ss3\ss2$

1648: &$31$&$11$

1649: &$\Mon{\ss1}$

1650: &$\s_1^2$

1651: \\

1652: $8$

1653: &$\ss3$

1654: &$32$&$12$

1655: &$\Mon{\ss2}$

1656: &$\ss2$

1657: \\

1658: $9$

1659: &$1$

1660: &$41$&$21$

1661: &$\Mon{\ss3}$

1662: &$\ss3$

1663: \end{tabular}

1664: \bigskip

1665: \caption{\sf Direct determination of the iterated

1666: decomposition of~$\D_4^2$: at step~$\rr$, we extract the

1667: maximal right divisor~$\xx_\rr$ of the current

1668: remainder~$\xx^{(\rr-1)}$ that lies in the

1669: monoid~$\MM_{\cl{\h_\rr}}$, we update the remainder

1670: into~$\xx^{(\rr)}$, and we define the next address~$\h_{\rr+1}$

1671: to be the maximal successor~$\h$ of~$\h_\rr$ for which

1672: $\xx^{(\rr)}$ is not orthogonal to~$\MM_{\cl\h}$; we stop

1673: when only $1$ is left.}

1674: \label{T:Iter}

1675: \end{table}

1676: 

1677: \section{The alternating normal form}

1678: \label{S:NormalForm}

1679: 

1680: We shall now deduce normal form results in (good) locally Garside monoids. The initial

1681: remark is that, if $\MM$ is a locally Garside monoid generated by an element~$\gen$, then $\MM$

1682: must be torsion-free by Condition~$(C_3)$, hence it is a free monoid, and every element of~$\MM$

1683: admits a unique expression as~$\gen^\ee$ with~$\ee \in \Nat$. Now, if $\MM$ is an arbitrary locally right Garside

1684: monoid and if $\MMM$ is an (iterated) covering of~$\MM$, then each element of~$\xx$ has been given a

1685: distinguished decomposition in terms of the factor monoids~$\MM_\a$ of~$\MMM$. If, moreover, each of the

1686: monoids~$\MM_\a$ happens to be generated by a single element~$\gen_\a$, the $\MMM$-decomposition gives a

1687: unique distinguished expression in terms of the elements~$\gen_\a$. This situation

1688: occurs for instance in the case of the $2$-covering of Example~\ref{X:IterCov}.

1689: 

1690: \subsection{Atomic coverings}

1691: 

1692: From now on, we consider locally right Garside monoids that satisfy Condition~$(C_3^+)$. It is easily seen that such

1693: monoids are generated by atoms, \ie, elements~$\gen$ such that $\gen = \xx\yy$ implies $\xx = 1$ or $\yy =

1694: 1$---see for instance~\cite{Dfx}. In view of the above remarks, it is natural to concentrate on

1695: coverings that involve submonoids generated by atoms.

1696: 

1697: \begin{defi}

1698: \label{D:Atomic}

1699: Assume that $\MM$ is a locally right Garside monoid, and $\GGen$ is an $\nn$-sequence of atoms of~$\MM$. We

1700: say that an $\nn$-covering of~$\MM$ is \emph{atomic}  based on the sequence~$\GGen$ if, for each

1701: $\nn$-address~$\a$, the monoid~$\MM_\a$ is the submonoid of~$\MM$ generated by the atom~$\Gen\a$.

1702: \end{defi}

1703: 

1704: For instance, the $2$-covering of Example~\ref{X:IterCov} is atomic, based on $((\ss2,\ss3),(\ss2,\ss1))$. Note that a

1705: base sequence must contain all atoms of~$\MM$, as, by definition, it generates~$\MM$. An arbitrary sequence of

1706: atoms need not always define a covering, as a submonoid generated by a family of atoms is not necessarily closed in

1707: the sense of Definition~\ref{D:Closed}. This however is true in

1708: braid monoids---and in all Artin--Tits monoids. 

1709: 

1710: Before going on and defining the $\MMM$-normal form, we discuss one more general point, namely whether

1711: $\MMM$-decompositions may have gap, this meaning that a trivial factor~$1$ may appear between two non-trivial

1712: factors. 

1713: 

1714: \begin{exam}

1715: \label{X:Gap}

1716: Let $\MM$ be the $5$-strand braid monoid~$\BB5$, and

1717: $\MMM$ be the $2$-covering based on~$((\ss4,\ss3),(\ss2,\ss1))$. One easily checks that the

1718: $\MMM$-decomposition of~$\xx$ is $((\ss4, 1), (\ss1))$,

1719: which has a trivial entry lying between two non-trivial entries.

1720: \end{exam}

1721: 

1722: It is easy to state conditions that exclude such gaps.

1723: 

1724: \begin{lemm}

1725: \label{L:Dense}

1726: Say that an $\nn$-covering~$\MMM$ is  \emph{dense} if, for each binary

1727: address~$\b$ of length~$\mm$ with $0\le\mm<\nn$,

1728: \begin{equation}

1729: \label{E:Dense}

1730: \text{$\MM_\b$ is generated by 

1731: $\MM_{\b1}$ and

1732: $\MM_{\b21^{\nn-\mm-1}}$, and by $\MM_{\b2}$ and

1733: $\MM_{\b1^{\nn-\mm}}$}. 

1734: \end{equation}

1735: Then, decomposi\-tions associated with a dense covering have no gap.

1736: \end{lemm}

1737: 

1738: \begin{proof}

1739: Owing to Proposition~\ref{P:GlobalIter}, the

1740: point is to prove that, if, for some

1741: binary $\nn$-address~$\a$ and some~$\mm$, 

1742: writing $\b$ (\resp $\b'$) for the length~$\mm$ (\resp

1743: $\mm+1$) prefix of~$\a$, we have both

1744: $\xx \not\perp \MM_\b$ and

1745: $\xx \perp \MM_{\b'}$, then necessarily the

1746: $\MM_{\BSucc\mm\a}$-tail of~$\xx$ is not trivial. Write

1747: $\b' = \b\rr$. For $\rr=1$, a sufficient condition for the

1748: previous implication is that $\MM_{\b}$ is generated

1749: by~$\MM_{\b1}$ and~$\MM_{\b21^{\nn-\mm-1}}$: then, a

1750: non-trivial right divisor of~$\xx$ lying in~$\MM_\b$ cannot

1751: be right divisible by any factor in~$\MM_{\b1}$ and,

1752: therefore, it must be right divisible by some factor

1753: in~$\MM_{\b21^{\nn-\mm-1}}$, and, by definition, we have

1754: $\b21^{\nn-\mm-1}=\BSucc\mm\a$. For $\rr=2$, the

1755: argument is similar, replacing~$\b1$ with~$\bŽ$, and

1756: $\b21^{\nn-\mm-1}$ with $\b1^{\nn-\mm}$. So, the

1757: conditions in~\eqref{E:Dense} are sufficient.

1758: \end{proof}

1759: 

1760: In the case of an atomic covering, the density condition of Lemma~\ref{L:Dense} requires

1761: that the base sequence be highly redundant. Such conditions are important in practice because they strongly limit

1762: the patterns that can be used in the construction of dense atomic coverings.

1763: 

1764: \begin{prop}

1765: \label{P:Successors}

1766: Assume that $\MMM$ is a dense atomic

1767: $\nn$-covering of~$\MM$ based on~$\GGen$. Then, for each

1768: $\nn$-address~$\a$, the set

1769: $\{\Gen{\BSucc\mm\a} \mid 0\le\mm\le\nobreak \nn\}$ is the atom

1770: set of~$\MM$, and the latter contains at most

1771: $\nn+1$ elements.

1772: \end{prop}

1773: 

1774: \begin{proof}

1775: Use induction on~$\nn\ge0$. The case $\nn=0$ is

1776: obvious. Assume $\nn\ge1$. Write $\a=\dd\b$ with

1777: $\dd=1$ or~$2$. Assume first $\dd=1$. By~\eqref{E:Dense},

1778: $\MM$ is generated by $\Gen{21^\nno}$, which is the

1779: $0$-successor of~$\a$, and~$\MM_1$. By induction

1780: hypothesis, the latter is generated by the family of all

1781: $\Gen{1\BSucc\mm\b}$'s, so $\MM$ is generated by the

1782: successors of~$\a$. The argument is symmetric for $\dd=2$,

1783: using the second part of~\eqref{E:Dense}. By

1784: construction, every $\nn$-address admits $\nn+1$

1785: successors, hence there are at most

1786: $\nn+1$ atoms in~$\MM$.

1787: \end{proof}

1788: 

1789: We shall see in Section~\ref{S:Flip} that dense atomic

1790: $\nn$-coverings involving $\nn\nobreak+\nobreak1$ atoms

1791: exist for each~$\nn$. For $\nn=2$, the

1792: only possible pattern is (up to renaming) that of Figure~\ref{F:Skeleton}. For

1793: $\nn\ge3$, several non-isomorphic patterns exist---see Figure~\ref{F:Solutions}.

1794: 

1795: \begin{figure}[htb]

1796: \begin{picture}(85,34)

1797: \put(0,-0.5){\includegraphics{Solutions.eps}}

1798: \put(0,0){$3$}

1799: \put(5,0){$2$}

1800: \put(10,0){$3$}

1801: \put(15,0){$4$}

1802: \put(20,0){$2$}

1803: \put(25,0){$3$}

1804: \put(30,0){$2$}

1805: \put(35,0){$1$}

1806: \put(2,10){$23$}

1807: \put(12,10){$34$}

1808: \put(22,10){$23$}

1809: \put(32,10){$12$}

1810: \put(6,20){$234$}

1811: \put(26,20){$123$}

1812: \put(15,30){$1234$}

1813: \put(50,0){$2$}

1814: \put(55,0){$3$}

1815: \put(60,0){$2$}

1816: \put(65,0){$4$}

1817: \put(70,0){$2$}

1818: \put(75,0){$3$}

1819: \put(80,0){$2$}

1820: \put(85,0){$1$}

1821: \put(52,10){$23$}

1822: \put(62,10){$24$}

1823: \put(72,10){$23$}

1824: \put(82,10){$12$}

1825: \put(56,20){$234$}

1826: \put(76,20){$123$}

1827: \put(65,30){$1234$}

1828: \end{picture}

1829: \caption{\sf The two possible patterns for a dense

1830: $3$-covering involving four atoms.}

1831: \label{F:Solutions}

1832: \end{figure}

1833: 

1834: \subsection{The  $\MMM$-normal form}

1835: \label{S:Normal}

1836: 

1837: We are now ready to convert the results of

1838: Sections~\ref{S:Iter} into the

1839: construction of a normal form. We recall that,

1840: for~$\SS$ generating~$\MM$ and~$\ww$ a word on~$\SS$,

1841: we denote by~$\CL\ww$ the element of~$\MM$ represented

1842: by~$\ww$. We write~$\ww(\kk)$ for the $\kk$th letter

1843: in~$\ww$ \emph{from the right}.

1844: 

1845: \begin{defi}

1846: \label{D:NF}

1847: Assume that $\MM$ is a locally right Garside monoid with atom set~$\SS$, and that $\MMM$ is a dense atomic

1848: $\nn$-covering of~$\MM$ based on~$\GGen$. A

1849: length~$\ell$ word~$\ww$ on~$\SS$ is said

1850: to be \emph{$\MMM$-normal} if

1851: \begin{quote}

1852: There exist $\nn$-addresses $\a_\ell, \Ldots, \a_0$ with

1853: $\a_0 = 1^\nn$ such that, for each~$\kk$, 

1854: $\ww(\kk)=\Gen{\a_\kk}$ holds, where $\a_\kk$ is the

1855: maximal successor of~$\a_{\kk-1}$---\ie, is

1856: $\BSucc\mm{\a_{\kk-1}}$ with maximal~$\mm$---for which

1857: $\Gen{\a_\kk}$ is a right divisor of

1858: $\CL{\ww(\ell) ... \ww(\kk)}$.

1859: \end{quote}

1860: \end{defi}

1861: 

1862: The above definition may look convoluted, but handling a few

1863: examples should make it easily understandable. 

1864: Table~\ref{T:XNF} shows that our favourite example, namely 

1865: $\ss3\ss2\ss1\ss1\ss2\ss3\ss2\ss1\ss1\ss2\ss1\ss1$,

1866: is $\MMM$-normal with respect to the $2$-covering

1867: of~Example~\ref{X:IterCov}.

1868: 

1869: The expected existence and uniqueness of the $\MMM$-normal form is the following easy result.

1870: 

1871: \begin{prop}

1872: \label{P:NF}

1873: Assume that $\MM$ is a locally right Garside monoid

1874: with atom set~$\SS$, and $\MMM$ is a dense atomic

1875: $\nn$-covering of~$\MM$ based on~$\GGen$.  Then each

1876: element~$\xx$ of~$\MM$ admits a unique

1877: $\MMM$-normal representative, namely

1878: $\Gen{{\a_\ell}}...\Gen{{\a_1}}$, where

1879: $\a_\ell, \Ldots, \a_1$ are inductively determined

1880: from $\xx^{(0)}=\xx$ and $\a_0=1^\nn$ by

1881: \begin{equation}

1882: \label{E:NF2}

1883: \a_\kk=\BSucc\mm{\a_{\kk-1}}

1884: \text{\quad and \quad}

1885: \xx^{(\kk)}= \xx^{(\kk-1)}\OVER{\Gen{{\a_\kk}}},

1886: \end{equation}

1887: where $\mm$ is maximal such that $\Gen{\BSucc\mm{\a_{\kk-1}}}$ is a right divisor of~$\xx^{(\kk)}$.

1888: Moreover, $\Gen{{\a_\ell}}...\Gen{{\a_1}}$ is the word

1889: obtained from the $\MMM$-decomposition of~$\xx$ by

1890: concatenating the entries and possibly deleting the final~$1$. 

1891: \end{prop}

1892: 

1893: \begin{proof}

1894: The existence follows from the assumption that

1895: $\MMM$ is dense, which guarantees that, as long as the

1896: remainder~$\xx^{(\kk)}$ is not trivial, there must exist a

1897: successor~$\BSucc\mm{\a_{\kk-1}}$ of the address~$\a_{\kk-1}$ such that

1898: $\Gen{\BSucc\mm{\a_{\kk-1}}}$ is a right divisor of~$\xx^{(\kk)}$.

1899: Uniqueness follows from the choice of that successor.

1900: 

1901: The inductive construction of~\eqref{E:NF2} is

1902: essentially the construction of the $\MMM$-decomposition

1903: as given in Proposition~\ref{P:GlobalIter}. The only

1904: difference is that, here, we do not extract the whole tail of

1905: the current remainder, but only one letter at each step. For

1906: instance, if, at some point, the generator to be looked for

1907: is~$\gen$ and the current remainder~$\xx^{(\kk-1)}$ is

1908: divisible by~$\gen^2$, then

1909: $\xx^{(\kk)}$ is~$\xx^{(\kk-1)}\OVER{\gen}$, and, at the next step,

1910: $\a_\kk$ is the $\nn$-successor of~$\a_{\kk-1}$, \ie, it

1911: is~$\a_{\kk-1}$ again, and the next letter of the normal form

1912: is~$\gen$ again. In such a case, we have $\mm=\nn$.

1913: By contrast, in Proposition~\ref{P:GlobalIter}, the

1914: parameter~$\mm$ is never~$\nn$.

1915: \end{proof}

1916: 

1917: Under the hypotheses of Proposition~\ref{P:NF}, the

1918: word~$\ww$ is called the \emph{$\MMM$-normal form} of~$\xx$. The construction described in

1919: Proposition~\ref{P:NF} is an algorithm, displayed in Table~\ref{T:Algo}.  A

1920: typical example  is given in Table~\ref{T:XNF}.

1921: 

1922: \begin{table}[htb]

1923: \begin{tabular}{l}

1924: \hline

1925: \rule{0pt}{12pt}{\bf Input:} A word~$\ww$ on~$\SS$;\\

1926: %\hline

1927: \rule{0pt}{12pt}{\bf Procedure:}\\

1928: \hspace{7mm}$\ww':=\mathtt{emptyword}$;\\

1929: \hspace{7mm}$\a:= 1^\nn$;\\

1930: \hspace{7mm}$\mathtt{while}\

1931: \ww\not=\mathtt{emptyword} 

1932: \ \mathtt{do}$\\

1933: \hspace{14mm}$\mm:=\nn$;\\

1934: \hspace{14mm}$\mathtt{while}\ \mathtt{quotient}

1935: (\ww,\Gen{{\BSucc\mm{\a}}})\ =\mathtt{error}\ 

1936: \mathtt{do}$\\

1937: \hspace{21mm}$\mm:= \mm-1$;\\

1938: \hspace{14mm}$\mathtt{od}$;\\

1939: \hspace{14mm}$\a:=\BSucc\mm{\a}$;\\

1940: \hspace{14mm}$\ww:= \mathtt{quotient}(\ww,\Gen{\a})$;\\

1941: \hspace{14mm}$\ww':=\mathtt{concat}(\Gen{\a}, \ww')$;\\

1942: \hspace{7mm}$\mathtt{od}$.\\

1943: \rule[-6pt]{0pt}{12pt}{\bf Output:} The unique

1944: $\MMM$-normal word~$\ww'$ that is equivalent

1945: to~$\ww$.\\

1946: \hline

1947: \end{tabular}

1948: \bigskip

1949: \caption{\sf Algorithm for the $\MMM$-normal

1950: form; we assume that $\SS$ is the atom set of~$\MM$, and

1951: $\MMM$ is a dense atomic $\nn$-covering of~$\MM$

1952: based on~$\GGen$; moreover, we assume

1953: that $\mathtt{quotient}(\ww,\gen)$ is a subroutine that, for

1954: $\ww$ a word on~$\SS$ and $\gen$ in~$\SS$,

1955: returns $\mathtt{error}$ if $\gen$ is not a right divisor

1956: of~$\CL\ww$, and returns a word

1957: representing~$\CL\ww\OVER{\gen}$ otherwise.}

1958: \label{T:Algo}

1959: \end{table}

1960: 

1961: \begin{table}[htb]

1962: \begin{tabular}{c|ccccccc}

1963: $\kk$\quad

1964: &$\ww_\kk$\hfill$\ww'_\kk$

1965: \rule[-10pt]{0pt}{12pt}%

1966: &\hbox{\quad$\a_{\kk-1}$\quad}

1967: &$\mm$

1968: &$\BSucc\mm{\a_{\kk-1}}$

1969: &$\Gen{\BSucc\mm{\a_{\kk-1}}}$

1970: &$\CL{\ww_\kk} \multe

1971: \Gen{\BSucc\mm{\a_{\kk-1}}}?$

1972: \\

1973: \hline

1974: \rule{0pt}{12pt}%

1975: $0$

1976: &$\ss1\ss2\ss1\ss3\ss2\ss1\ss1\ss2\ss1\ss3\ss2\ss1$

1977: \qquad\hfill

1978: -

1979: &$11$

1980: &$2$

1981: &$11$

1982: &$\ss1$

1983: &yes\\

1984: $1$

1985: &$\ss1\ss2\ss1\ss3\ss2\ss1\ss1\ss2\ss1\ss3\ss2$

1986: \qquad\hfill

1987: $\ss1$

1988: &$11$

1989: &$2$

1990: &$11$

1991: &$\ss1$

1992: &yes\\

1993: $2$

1994: &$\ss2\ss1\ss3\ss2\ss1\ss1\ss2\ss3\ss1\ss2$

1995: \qquad\hfill

1996: $\ss1\ss1$

1997: &$11$

1998: &$2$

1999: &$11$

2000: &$\ss1$

2001: &no\\

2002: &&

2003: &$1$

2004: &$12$

2005: &$\ss2$

2006: &yes\\

2007: $3$

2008: &$\ss2\ss1\ss3\ss2\ss1\ss1\ss2\ss3\ss1$

2009: \qquad\hfill

2010: $\ss2\ss1\ss1$

2011: &$12$

2012: &$2$

2013: &$12$

2014: &$\ss2$

2015: &no\\

2016: &&

2017: &$1$

2018: &$11$

2019: &$\ss1$

2020: &yes\\

2021: $4$

2022: &$\ss2\ss1\ss3\ss2\ss1\ss1\ss2\ss3$

2023: \qquad\hfill

2024: $\ss1\ss2\ss1\ss1$

2025: &$11$

2026: &$2$

2027: &$11$

2028: &$\ss1$

2029: &yes\\

2030: $5$

2031: &$\ss2\ss3\ss2\ss1\ss1\ss2\ss3$

2032: \qquad\hfill

2033: $\ss1\ss1\ss2\ss1\ss1$

2034: &$11$

2035: &$2$

2036: &$11$

2037: &$\ss1$

2038: &no\\

2039: &&

2040: &$1$

2041: &$12$

2042: &$\ss2$

2043: &yes\\

2044: $6$

2045: &$\ss3\ss2\ss1\ss1\ss2\ss3$

2046: \qquad\hfill

2047: $\ss2\ss1\ss1\ss2\ss1\ss1$

2048: &$12$

2049: &$2$

2050: &$12$

2051: &$\ss2$

2052: &no\\

2053: &&

2054: &$1$

2055: &$11$

2056: &$\ss1$

2057: &no\\

2058: &&

2059: &$0$

2060: &$21$

2061: &$\ss3$

2062: &yes\\

2063: $7$

2064: &$\ss3\ss2\ss1\ss1\ss2$

2065: \qquad\hfill

2066: $\ss3\ss2\ss1\ss1\ss2\ss1\ss1$

2067: &$21$

2068: &$2$

2069: &$21$

2070: &$\ss3$

2071: &no\\

2072: &&

2073: &$1$

2074: &$22$

2075: &$\ss2$

2076: &yes\\

2077: $8$

2078: &$\ss3\ss2\ss1\ss1$

2079: \qquad\hfill

2080: $\ss2\ss3\ss2\ss1\ss1\ss2\ss1\ss1$

2081: &$22$

2082: &$2$

2083: &$22$

2084: &$\ss2$

2085: &no\\

2086: &&

2087: &$1$

2088: &$21$

2089: &$\ss3$

2090: &no\\

2091: &&

2092: &$0$

2093: &$11$

2094: &$\ss1$

2095: &yes\\

2096: $9$

2097: &$\ss3\ss2\ss1$

2098: \qquad\hfill

2099: $\ss1\ss2\ss3\ss2\ss1\ss1\ss2\ss1\ss1$

2100: &$11$

2101: &$2$

2102: &$11$

2103: &$\ss1$

2104: &yes\\

2105: $10$

2106: &$\ss3\ss2$

2107: \qquad\hfill

2108: $\ss1\ss1\ss2\ss3\ss2\ss1\ss1\ss2\ss1\ss1$

2109: &$11$

2110: &$2$

2111: &$11$

2112: &$\ss1$

2113: &no\\

2114: &&

2115: &$1$

2116: &$12$

2117: &$\ss2$

2118: &yes\\

2119: $11$

2120: &$\ss3$

2121: \qquad\hfill

2122: $\ss2\ss1\ss1\ss2\ss3\ss2\ss1\ss1\ss2\ss1\ss1$

2123: &$12$

2124: &$2$

2125: &$12$

2126: &$\ss2$

2127: &no\\

2128: &&

2129: &$1$

2130: &$11$

2131: &$\ss1$

2132: &no\\

2133: &&

2134: &$0$

2135: &$21$

2136: &$\ss3$

2137: &yes\\

2138: $12$

2139: &-

2140: \qquad\hfill

2141: $\ss3\ss2\ss1\ss1\ss2\ss3\ss2\ss1\ss1\ss2\ss1\ss1$

2142: &21

2143: &-

2144: &-

2145: &-\\

2146: \end{tabular}

2147: \bigskip

2148: \caption{\sf Computation of the

2149: $\MMM$-normal form of~$\D_4^2$, for

2150: $\MMM$ the

2151: $2$-covering of Example~\ref{X:Iter},

2152: starting from the word

2153: $(\ss1\ss2\ss1\ss3\ss2\ss1)^2$: at

2154: each step, we try to divide the current word~$\ww_\kk$ by

2155: some generator~$\ss\rr$ and, when succesful, we add

2156: this~$\ss\rr$ on the left of~$\ww'_\kk$, until no letter is left

2157: in~$\ww_\kk$; the point is to know in which order the

2158: generators are tried, and this is specified by the

2159: address~$\a_\kk$: we try the successors

2160: of~$\a_{\kk-1}$ starting with the last one, \ie,

2161: with~$\a_{\kk-1}$, and then consider shorter and

2162: shorter prefixes of~$\a_{\kk-1}$; density guarantees that we

2163: cannot get stuck until $\ww_\kk$ is empty.}

2164: \label{T:XNF}

2165: \end{table}

2166: 

2167: As for complexity, computing the $\MMM$-normal

2168: form is as easy as computing the $\MMM$-decomposition.

2169: In our current atomic context, the existence of the norm

2170: (Definition~\ref{D:Norm}) is guaranteed~\cite{Dfx}.

2171: 

2172: \begin{prop}

2173: \label{P:Complexity2}

2174: Assume that $\MM$ is a locally right Garside monoid with atom set~$\SS$, that $\MMM$ is a dense atomic

2175: $\nn$-covering of~$\MM$ based on~$\GGen$, and that Condition~$(*)$ of

2176: Proposition~\ref{P:Complexity1} is satisfied. Then, for  each

2177: word~$\ww$ on~$\SS$, the algorithm of

2178: Table~\ref{T:Algo} runs in time~$O(\norm\ww^2)$.

2179: \end{prop}

2180: 

2181: \begin{proof}

2182: The only change with respect to

2183: Proposition~\ref{P:Complexity1} is that we have to keep

2184: track of binary addresses of fixed length~$\nn$ so as to

2185: know in which order the divisions have to be tried. Getting a

2186: new letter of the normal word under construction requires at

2187: most $\nn+1$~divisions, but the rest is similar.

2188: \end{proof}

2189: 

2190: \subsection{The exponent sequence}

2191: 

2192: We conclude this section with an easy remark about  $\MMM$-decompositions in the context of atomic coverings,

2193: namely that an element of the monoid is non-ambiguously determined by the iterated sequence of exponents in its

2194: $\MMM$-decomposition, \ie, we can forget about names of atoms and only keep track of exponents

2195: without losing information.

2196: 

2197: \begin{defi}

2198: For $\MM, \MMM$ as in Definition~\ref{D:NF}, and for $\www$ an iterated sequence whose entries are of the

2199: form~$\Gen\a^{\ee_\a}$, we define the \emph{exponent sequence}~$\exp\www$ of~$\www$ to be the iterated

2200: sequence obtained by replacing~$\Gen\a^{\ee_\a}$ with~$\ee_\a$ everywhere in~$\www$. 

2201: \end{defi}

2202: 

2203: For instance, in the context of Example~\ref{X:Iter}, the $\MMM$-decomposition of~$\D_4^2$ is the $2$-sequence 

2204: $((\ss3), (\ss2,\s_1^2), (\ss2,\ss3), (\ss2, \s_1^2,\ss2,\s_1^2))$, so the exponent

2205: sequence is the $2$-sequence of natural numbers

2206: $$((1), (1,2), (1,1), (1,2,1,2)).$$

2207: As in the case of every iterated sequence,  specifying the exponent sequence of~$\DD\MMM\xx$ amounts to

2208: giving two ordinary sequences, namely its unbracketing---in the above

2209: example $(1, 1,2, 1,1, 1,2,1,2)$---and its address list---$(41,32,31,22,21$, $14,13, 12,11)$ above. Easy examples show

2210: that, taken separately, neither of the above sequences is sufficient to recover~$\xx$. But, when we take them

2211: simultaneously, we can recover~$\xx$.

2212: 

2213: \begin{prop}

2214: \label{P:Exponent}

2215: If $\MMM$ is an atomic $\nn$-covering of~$\MM$, then, for

2216: every~$\xx$ in~$\MM$, the exponent sequence of~$\DD\MMM\xx$

2217: determines~$\xx$.

2218: \end{prop}

2219: 

2220: \begin{proof}

2221: Let~$\GGen$ be the base sequence of~$\MMM$, and let

2222: $(\ee_\pp, \Ldots, \ee_1)$ and $(\h_\pp, \Ldots, \h_1)$ be

2223: the unbracketing and the address list in the exponent sequence of~$\DD\MMM\xx$. Then

2224: we recover~$\DD\MMM\xx$ itself, and

2225: therefore~$\xx$, by replacing for each~$\rr$ the

2226: entry~$\ee_\rr$ corresponding to an address~$\h_\rr$

2227: with~$\Gen{\cl{\h_\rr}}^{\ee_\rr}$. The formal proof is an easy induction on the degree of the

2228: covering~$\MMM$---see Figure~\ref{F:TreeNat} for an

2229: example.

2230: \end{proof}

2231: 

2232: \begin{figure}[htb]

2233: \begin{picture}(42,23)

2234: %\Grid

2235: \put(0.5,3){\includegraphics{TreeNat.eps}}

2236: \put(0,0){$1$}

2237: \put(5,0){$1$}

2238: \put(10,0){$2$}

2239: \put(15,0){$1$}

2240: \put(20,0){$1$}

2241: \put(25,0){$1$}

2242: \put(30,0){$2$}

2243: \put(35,0){$1$}

2244: \put(40,0){$2$}

2245: \end{picture}

2246: \caption{\sf Tree representation of  the exponent

2247: sequence of~$\DD\MMM{\D_4^2}$, \ie, of  $((1), (1,2), (1,1),

2248: (1,2,1,2))$; Proposition~\ref{P:Exponent} states that the

2249: geometry of the tree determines the missing names: for

2250: instance, the leftmost~$2$ has address~$31$ in the tree, so

2251: it corresponds to the generator~$\Gen{\cl{31}}$,

2252: which is~$\ss1$; hence, this entry~$2$ must correspond to a

2253: factor~$\s_1^2$ in $\DD\MMM{\D_4^2}$.}

2254: \label{F:TreeNat}

2255: \end{figure}

2256: 

2257: \section{The $\Phi$-normal form of braids}

2258: \label{S:Flip}

2259: 

2260: From now on, we concentrate on the specific case of braids. In order to apply the previous results, we fix for

2261: each~$\nn$ a covering of~$\BB\nn$ by two copies of~$\BB\nno$, namely

2262: $\BB\nno$ and its image under the flip automorphism~$\ff\nn$. We study the decomposition associated with this

2263: covering, as well as an iterated version and the derived normal form, called the

2264: $\Phi$-normal form. This naturally leads to introducing a certain linear

2265: ordering of~$\BB\nn$, which will be subsequently proved to be connected with the standard braid ordering.

2266: 

2267: \subsection{The $\Phi$-splitting of a braid}

2268: 

2269: In the sequel, we always  consider $\BB\nno$ as a submonoid of~$\BB\nn$: an

2270: $(\nn-1)$-strand braid is a particular $\nn$-strand braid. 

2271: We denote by~$\ff\nn$ the flip automorphism of~$\BB\nn$ that exchanges~$\ss\ii$ and~$\ss{\nn-\ii}$ for

2272: each~$\ii$. It is well-known---see for instance~\cite[Chapter~1]{Dhr}---that $\ff\nn$ is the conjugation by the

2273: Garside element~$\D_\nn$. We also use~$\ff\nn$ for $\nn$-strand braid words, thus denoting by~$\ff\nn(\ww)$

2274: the image of a braid word~$\ww$ under~$\ff\nn$ letter by letter.  

2275: 

2276: The initial, obvious observation is that, for each~$\nn\ge 3$,

2277: the monoids~$\BB\nno$ and~$\ff\nn(\BB\nno)$ are closed submonoids

2278: of~$\BB\nn$, and that the pair $(\ff\nn(\BB\nno), \BB\nno)$ is a

2279: covering of~$\BB\nn$ in the sense of

2280: Definition~\ref{D:Covering}. Thus Proposition~\ref{P:AltDec}

2281: gives for every $\nn$-strand braid a distinguished

2282: decomposition as an alternating product of elements

2283: of~$\BB\nno$ and~$\ff\nn(\BB\nno)$, according to the scheme of Figure~\ref{F:Alt}. We now

2284: restate the general result so as to emphasize the r\^ole of the flip

2285: automorphism.

2286: 

2287: \begin{prop}

2288: \label{P:FSplitting}

2289: Every braid~$\xx$ in~$\BB\nn$ admits a unique decomposition

2290: \begin{equation}

2291: \label{E:Flip}

2292: \xx=\ff\nn^\ppo(\xx_\pp) \cdot ... \cdot \ff\nn(\xx_2) \cdot \xx_1

2293: \end{equation}

2294: with  $\xx_1, \Ldots, \xx_\pp$ in~$\BB\nno$

2295: such that, for each~$\rr\ge2$, the only $\ss\ii$

2296: that is a right divisor of

2297: $\ff\nn^{\pp-\rr}(\xx_\pp) \cdot ... \cdot \ff\nn(\xx_{\rr+1})

2298: \cdot \xx_\rr$ is~$\ss1$. The braids~$\xx_\rr$

2299: are determined from

2300: $\xx^{(0)}= \xx$ by

2301: \begin{equation}

2302: \label{E:FlipDec2}

2303: \xx_\rr= 

2304: \Tail{\xx^{(\rr-1)}}{\BB\nno}, 

2305: \quad

2306: \xx^{(\rr)}=\ff\nn(\xx^{(\rr-1)}\OVER{\xx_\rr}). 

2307: \end{equation}

2308: \end{prop}

2309: 

2310: \begin{proof}

2311: As $\ff\nn$ is an automorphism of~$\BB\nn$, the relation

2312: $\yy_1=\Tail\yy{\ff\nn(\BB\nno)}$ is equivalent to

2313: $\ff\nn(\yy_1) =\Tail{\ff\nn(\yy)}{\BB\nno}$. Moreover

2314: $\ff\nn$ is an automorphism for the quotient operation

2315: as well. Then \eqref{E:Flip} and the divisibility constraints

2316: just express that the sequence 

2317: $(\ff\nn^\ppo(\xx_\pp), \Ldots, \ff\nn(\xx_2), \xx_1)$ is the

2318: $(\ff\nn(\BB\nno), \BB\nno)$-decomposition of~$\xx$.

2319: \end{proof}

2320: 

2321: \begin{defi}

2322: \label{D:Splitting}

2323: The sequence $(\xx_\pp, \Ldots, \xx_1)$ involved

2324: in~\eqref{E:Flip} is called the \emph{$\nn$-splitting}

2325: of~$\xx$; the parameter~$\pp$ is called the \emph{$\nn$-breadth of~$\xx$}.

2326: \end{defi}

2327: 

2328: The only difference between the $(\ff\nn(\BB\nno),

2329: \BB\nno)$-decomposition and the $\nn$-splitting is that the

2330: flip~$\ff\nn$ is applied to each other entry. The benefit is that

2331: all entries in the $\nn$-splitting of a braid of~$\BB\nn$ are braids of~$\BB\nno$, and not elements of~$\BB\nno$

2332: and~$\ff\nn(\BB\nno)$, alternately.  Note that the

2333: $\nn$-splitting of~$\xx$ is obtained by repeating a

2334: single operation, namely finding the $\BB\nno$-tail of~$\xx$---hence the right gcd of~$\xx$

2335: and~$\D_\nno^\infty$ as was seen in

2336: Example~\ref{X:Closed}---and flipping the quotient.

2337: 

2338: \begin{exam}

2339: \label{X:Splitting}

2340: Let $\xx$ be the $4$-strand braid $\D_4^2$. The $\BB3$-tail of~$\xx$ is $\D_3^2$, with associated

2341: quotient $\ss3\ss2\s_1^2\ss2\ss3$, hence, after a flip,

2342: $\xx^{(1)} = \ss1\ss2\s_3^2\ss2\ss1$. The $\BB3$-tail of~$\xx^{(1)}$ is $\ss2\ss1$,

2343: with quotient $\ss1\ss2\s_3^2$, hence, after a flip,

2344: $\xx^{(2)} = \ss3\ss2\s_1^2$. The $\BB3$-tail of~$\xx^{(2)}$ is $\ss2\s_1^2$, with

2345: quotient~$\ss3$, hence, after a flip, $\xx^{(3)} = \ss1$, which belongs to~$\BB3$. Thus $\D_4^2$ has

2346: $4$-breadth~$4$, and its

2347: $4$-splitting is $(\ss1, \ss2\s_1^2, \ss2\ss1,

2348: \D_3^2)$---compare with the

2349: $(\ff4(\BB3), \BB3)$-decomposition of~$\D_4^2$ as computed

2350: in Example~\ref{X:Delta}.

2351: \end{exam}

2352: 

2353: Note that, as in the case of the $(\ff\nn(\BB\nno), \BB\nno)$-decomposition, the non-final entries in an

2354: $\nn$-splitting are never~$1$, but the final (rightmost) entry may:

2355: the $3$-splitting of~$\ss2$ is $(\ss1, 1)$, as $\ss2$ is not divisible by~$\ss1$.

2356: 

2357: \subsection{The flip covering of~$\BB\nn$}

2358: 

2359: The $\nn$-splitting operation associates with every braid of~$\BB\nn$ a sequence of braids of~$\BB\nno$. We

2360: can now iterate the construction, so as to associate with every braid of~$\BB\nn$ an iterated sequence of braids

2361: of~$\BB2$. According to the general framework of Section~\ref{S:Iter}, this entails introducing an

2362: iterated $(\nnt)$-covering of the monoid~$\BB\nn$.

2363: 

2364: \begin{defi}

2365: For $\nn\ge2$, we denote~$\BBB\nn$ the $(\nn\!-\!2)$-covering of~$\BB\nn$ defined by

2366: \begin{equation}

2367: \label{E:DefFlipCov}

2368: \BBB2=\BB2, \quad

2369: \BBB\nn=(\ff\nn(\BBB\nno), \BBB\nno).

2370: \end{equation}

2371: \end{defi}

2372: 

2373: Applying the recursive definition, we find

2374: \begin{gather*}

2375: \BBB3 = (\ff3(\BB2), \BB2) = (\Mon{\ss2}, \Mon{\ss1}),\\

2376: \BBB4 = (\ff4(\BBB3), \BBB3) = ((\Mon{\ss2}, \Mon{\ss3}), (\Mon{\ss2}, \Mon{\ss1})), 

2377: \end{gather*}

2378: which is the $2$-covering of Example~\ref{X:IterCov}. More generally, writing $\BB{\nn,\a}$ for the $\a$-entry

2379: in~$\BBB\nn$, we deduce from~\eqref{E:DefFlipCov} the rules

2380: \begin{equation}

2381: \label{E:FlipCov4}

2382: \BB{2,\ea}=\BB2, \quad

2383: \BB{\nn,1\a}=\BB{\nn-1,\a},\quad \text{and} \quad

2384: \BB{\nn,2\a}=\ff{\nn}(\BB{\nn-1,\a}).

2385: \end{equation}

2386: 

2387: The above values show that $\BBB3$ and~$\BBB4$ are dense atomic coverings. This result extends to all values

2388: of~$\nn$, with the following description of the base sequence.

2389: 

2390: \begin{prop}

2391: \label{P:BraidCovering}

2392: For $\nn \ge 2$, define the $(\nnt)$-sequence~$\GGen_\nn$  by

2393: \begin{equation}

2394: \label{E:FlipCov}

2395: \GGen_2=\ss1, \quad

2396: \GGen_\nn= (\ff\nn(\GGen_\nno), \GGen_\nno).

2397: \end{equation}

2398: Then, for each  binary address~$\a$ of length $\nn\!-\!2$, we have

2399: $\Gen{\a} =\ss\ii$ with

2400: \begin{equation}

2401: \label{E:Name}

2402: \ii = -\mm_1 + \mm_2 - ... +

2403: (-1)^\rr \mm_\rr +

2404: \begin{cases}

2405: 1&\text{if $\rr$ is even},\\

2406: \nn&\text{if $\rr$ is odd},

2407: \end{cases}

2408: \end{equation}

2409: if $\a = \dd_1...\dd_{\nn-2}$ and $\mm_1 < ... <

2410: \mm_\rr$ are the~$\mm$'s for which $\dd_\mm$ is even. 

2411: Moreover, $\BBB\nn$ is a dense atomic covering based

2412: on~$\GGen_\nn$. 

2413: \end{prop}

2414: 

2415: \begin{proof}

2416: Firstly, we prove~\eqref{E:Name} using induction on~$\nn \ge 2$. For $n= 2$, \eqref{E:Name}

2417: reduces to $\Gen\ea=\ss1$, which is true. Assume $\nn\ge3$, and let

2418: $\a'=\dd_2...\dd_{\nn-2}$. Putting $\Gen{\a'}=\ss{\ii'}$,

2419: we aim at proving $\ii=\ii'$ if $\dd_1$ is odd, and

2420: $\ii=\nn-\ii'$ if $\dd_1$ is even.  Write $\SS$ for $-\mm_1 +

2421: \mm_2 - ... + (-1)^\rr

2422: \mm_\rr$, and $\rr'$, $\mm'_1$, $\mm'_2$, \Ldots, $\SS'$,

2423: $\nn'$ for the similar parameters associated with~$\a'$.

2424: Assume first that $\dd_1$ is odd. Then we have $\rr=\rr'$,

2425: and $\mm_\jj=\mm'_\jj+1$ for each~$\jj$, hence

2426: $\SS=\SS'$ if $\rr$ is odd, and

2427: $\SS=\SS'-1$ if $\rr$ is even. The induction

2428: hypothesis gives $\ii'=\SS'+1$ if $\rr$ is odd,

2429: $\SS'+\nn'$ if $\rr$ is even. We deduce

2430: $\ii= \SS+1 = \SS'+1=\ii'$ if $\rr$ is even, and 

2431: $\ii= \SS+\nn = \SS'-1+\nn'+1=\ii' $ if $\rr$ is odd.

2432: 

2433: Assume now that $\dd_1$ is even. Then we  have

2434: $\rr=\rr'+1$, $\mm_1=1$, and $\mm_{\jj+1} =\mm'_\jj+1$

2435: for each~$\jj\ge1$, hence $\SS=-\SS'$ if $\rr$ is

2436: odd, and $\SS=-\SS'-1$ if $\rr$ is even. The induction

2437: hypothesis gives $\ii'=\SS'+\nn'$ if $\rr$ is

2438: odd, $\SS'+1$ if $\rr$ is even. We deduce

2439: $\ii = \SS+1 = - \SS'+1= \nn - \ii'$ if $\rr$ is odd, and $\ii = 

2440: \SS+\nn = - \SS'-1+\nn=\nn-\ii'$ if $\rr$ is

2441: even.

2442: 

2443: Nex, the braids~$\ss\ii$ are the atoms of~$\BB\nn$, and every parabolic submonoid of~$\BB\nn$ is closed, so

2444: every surjective sequence of atoms defines a covering. An obvious induction on~$\nn$ shows that, for~$\nn \ge 2$,

2445: each of $\ss1, \Ldots, \ss\nno$ occurs in the sequence~$\GGen_\nn$. Moreover, comparing~\eqref{E:DefFlipCov}

2446: and~\eqref{E:FlipCov} makes it straightforward that $\BBB\nn$ is precisely the covering based on~$\GGen_\nn$.

2447: 

2448: As for density, the point is

2449: to show that $\BB\nn$ is generated by~$\BB\nno$

2450: and~$\BB{\nn,21^{\nn-3}}$. Now \eqref{E:Name} gives

2451: $\Gen{21^{\nn-3}} = \ss\nno$, precisely

2452: the atom of~$\BB\nn$ missing in~$\BB\nno$.

2453: \end{proof}

2454: 

2455: It is easy to see that, for each~$\nn$, the unbracketing of ~$\GGen_\nn$ is the length~$2^{\nn-2}$ suffix

2456: of some left infinite sequence $\GGen_\infty$ where indices are

2457: $$\quad...,6,3,4,3,2,4,3,4,5,3,2,3,4,2,3,2,1.$$

2458: An example of application for the rule of~\eqref{E:Name} is as follows: in the length~$7$ address~$1221212$, there

2459: are even digits at positions $2,3,5,7$ (from the left), so \eqref{E:Name}

2460: gives $\ii=(-2+3-5+7)+1=4$, hence $\Gen{1221212}=\ss4$. 

2461: 

2462: As $\BBB\nn$ is a dense atomic covering of~$\BB\nn$, it is eligible for the results of Section~\ref{S:Iter}. We fix

2463: some specific, simplified notation.

2464: 

2465: \begin{nota}

2466: For~$\xx$ in~$\BB\nn$, the $\BBB\nn$-decomposition of~$\xx$ is denoted by~$\DDf\nn\xx$, and its exponent

2467: sequence is denoted by~$\DDfe\nn\xx$.

2468: \end{nota}

2469: 

2470: The recursive definition of~$\BBB\nn$  implies the following connection between the splitting and the

2471: $\BBB\nn$-decomposition.

2472: 

2473: \begin{lemm}

2474: \label{L:DecompBraid}

2475: For $\nn \ge 3$ and  $\xx$ in~$\BB\nn$, we have

2476: \begin{equation}

2477: \label{E:BiFlip21}

2478: \DDf\nn\xx=(\ff\nn^\ppo(\DDf\nno{\xx_\pp}), \Ldots,

2479: \ff\nn(\DDf\nno{\xx_2}), \DDf\nno{\xx_1}).

2480: \end{equation}

2481: where $(\xx_\pp, \Ldots, \xx_1)$ is the $\nn$-splitting of~$\xx$.

2482: \end{lemm}

2483: 

2484: \begin{proof}

2485: By definition, the $(\ff\nn(\BB\nno), \BB\nno)$-decomposition of~$\xx$ is the sequence

2486: $$(\ff\nn^\ppo(\xx_\pp), \Ldots, \ff\nn(\xx_2), \xx_1),$$

2487: and, therefore, by definition again, we have

2488: \begin{equation*}

2489: \DDf\nn\xx = 

2490: (\DD{\ff\nn^\ppo(\BBB\nno)}{\ff\nn^\ppo(\xx_\pp)},

2491: \Ldots, 

2492: \DD{\ff\nn(\BBB\nno)}{\ff\nn(\xx_2)}, 

2493: \DD{\BBB\nno}{\xx_1}).

2494: \end{equation*}

2495: Now, as $\ff\nn$ is an automorphism of~$\BB\nn$, we have

2496: $\DD{\ff\nn(\BBB\nno)}{\ff\nn(\yy)} = \ff\nn(\DD{\BBB\nno}{\yy})$

2497: for each~$\yy$ in~$\BB\nno$, \ie, $\DD{\ff\nn(\BBB\nno)}{\ff\nn(\yy)} = \ff\nn(\DDf\nno\yy)$, and 

2498: \eqref{E:BiFlip21} follows.

2499: \end{proof}

2500: 

2501: \begin{exam}

2502: \label{X:FlipNormal}

2503: (See Figure~\ref{F:Splitting})

2504: We saw in Example~\ref{X:Splitting} that the $4$-splitting

2505: of~$\D_4^2$ is $(\ss1, \ss2\s_1^2, \ss2\ss1, \D_3^2)$.

2506: Now, the $3$-splitting of~$\D_3^2$ turns out to be

2507: $(\ss1, \s_1^2, \ss1, \s_1^2)$, that of $\ss2\ss1$ is $(\ss1,

2508: \ss1)$, etc. Gathering the results, and applying the needed

2509: flips, we find

2510: \begin{equation}

2511: \label{E:XFlip}

2512: \DDf4{\D_4^2}=((\ss3), (\ss2,\s_1^2), (\ss2,\ss3), 

2513: (\ss2, \s_1^2,\ss2,\s_1^2)),

2514: \end{equation}

2515: as already seen in Example~\ref{X:Iter}. The associated exponent sequence is

2516: \begin{equation}

2517: \DDfe4{\D_4^2}=((1), (1,2), (1,1), (1,2,1,2)),

2518: \end{equation}

2519: \end{exam}

2520: 

2521: \begin{figure}[htb]

2522: \begin{picture}(43,35)

2523: \put(-1,12){\includegraphics{Tree.eps}}

2524: \put(0,9){$\ss1$}

2525: \put(5,9){$\ss1$}

2526: \put(10,9){$\s_1^2$}

2527: \put(15,9){$\ss1$}

2528: \put(20,9){$\ss1$}

2529: \put(25,9){$\ss1$}

2530: \put(30,9){$\s_1^2$}

2531: \put(35,9){$\ss1$}

2532: \put(40,9){$\s_1^2$}

2533: \put(0,21){$\ss2$}

2534: \put(7,21){$\ss2\s_1^2$}

2535: \put(17,21){$\ss1\ss2$}

2536: \put(33,21){$\D_3^2$}

2537: \put(19,31.5){$\D_4^2$}

2538: \put(6,27){$\scriptstyle\ff{\!4}$}

2539: \put(18.2,27){$\scriptstyle\ff{\!4}$}

2540: \put(4.5,15){$\scriptstyle\ff{\!3}$}

2541: \put(14.2,15){$\scriptstyle\ff{\!3}$}

2542: \put(25.5,15){$\scriptstyle\ff{\!3}$}

2543: \put(33.2,15){$\scriptstyle\ff{\!3}$}

2544: \put(-40,4){\ie, after reintroducing the flips,}

2545: \put(0,0){$\ss3$}

2546: \put(5,0){$\ss2$}

2547: \put(10,0){$\s_1^2$}

2548: \put(15,0){$\ss2$}

2549: \put(20,0){$\ss3$}

2550: \put(25,0){$\ss2$}

2551: \put(30,0){$\s_1^2$}

2552: \put(35,0){$\ss2$}

2553: \put(40,0){$\s_1^2$}

2554: \put(0,0){$\ss3$}

2555: \put(5,0){$\ss2$}

2556: \end{picture}

2557: \caption{\sf The $\BBB4$-decomposition of~$\D_4^2$ viewed as an iterated

2558: splitting: we split the initial braid of~$\BB4$ into a sequence

2559: of braids in~$\BB3$, then we split each of them into a sequence of

2560: braids in~$\BB2$, \ie, of powers of~$\ss1$; the sequence $\DDf4{\D_4^2}$ is obtained by

2561: iteratively flipping each other entry.}

2562: \label{F:Splitting}

2563: \end{figure}

2564: 

2565: \subsection{The $\Phi$-normal form}

2566: 

2567: The iterated covering~$\BBB\nn$ is atomic and, therefore, it gives raise to a unique normal form on~$\BB\nn$. 

2568: According to Proposition~\ref{P:NF}, the $\BBB\nn$-normal form of a braid~$\xx$ of~$\BB\nn$ is the word

2569: obtained by concatenating the (unique) expressions of the successive entries in its $\BBB\nn$-decomposition as

2570: powers of atom. For instance, from the $\BBB4$-decomposition of~$\D_4^2$ given in~\eqref{E:XFlip}, we deduce

2571: the $\BBB4$-normal form $\ss3 \ss2 \s_1^2 \ss2 \ss3 \ss2 \s_1^2 \ss2 \s_1^2$.

2572: 

2573: If $\xx$ belongs to~$\BB\nno$, then the $\nn$-splitting of~$\xx$ is

2574: the length one sequence~$(\xx)$. Therefore, we have $\DD\nn\xx = (\DD\nno\xx)$, and the normal

2575: form of~$\xx$ as an element of~$\BB\nno$ coincides with its normal form as an element of~$\BB\nn$. Owing to

2576: this remark, we shall forget about subscripts, and put the following without ambiguity.

2577: 

2578: \begin{defi}

2579: For $\xx$ in~$\BB\nn$, the $\BBB\nn$-normal form of~$\xx$ is called the \emph{$\Phi$-normal form} of~$\xx$.

2580: \end{defi}

2581: 

2582: Lemma~\ref{L:DecompBraid} implies that the $\Phi$-normal form has the following simple

2583: connection with the splitting operation---which could be taken as an alternative definition:

2584: 

2585: \begin{prop}

2586: For $\nn \ge 3$ and  $\xx$ in~$\BB\nn$, the $\Phi$-normal form of~$\xx$ is the word

2587: \begin{equation}

2588: \label{E:NFBraid}

2589: \ff\nn^\ppo(\ww_\pp) \cdot ...\cdot \ff\nn(\ww_2) \cdot \ww_1,

2590: \end{equation}

2591: where $(\xx_\pp, \Ldots, \xx_1)$ is the $\nn$-splitting of~$\xx$, and, for each~$\rr$, the word~$\ww_\rr$ is the

2592: $\Phi$-normal form of~$\xx_\rr$.

2593: \end{prop}

2594: 

2595: The results of Section~\ref{S:Normal} imply that, in addition to the above recursive definitions, the $\Phi$-normal form

2596: also admits direct characterizations. We shall now state such characterizations. Several

2597: equivalent statements are possible---and can be used in practical implementations. The principle is always:

2598: \begin{quote}

2599: An $\nn$-strand braid word~$\ww$ is $\Phi$-normal if, for

2600: each~$\kk$, the $\kk$th letter of~$\ww$ starting from the

2601: right is the smallest~$\ss\ii$ that is a right divisor of the

2602: braid represented by the prefix of~$\ww$ finishing at that

2603: letter, smallest referring to some local ordering of

2604: the~$\ss\ii$'s that is updated at each step and corresponds to

2605: a position in the skeleton of the covering~$\BBB\nn$.

2606: \end{quote}

2607: The formal definition includes a description of the

2608: local ordering of the~$\ss\ii$'s. The latter can be encoded in

2609: several equivalent ways, involving addresses, or numbers, or

2610: permutations. If the local ordering were the

2611: fixed order $\ss1\!< \!...\!< \!\ss\nno$, then being normal

2612: would simply mean being lexicographically minimal. 

2613: 

2614: We recall that, for $\a$ a binary address,

2615: $\BSucc\mm\aa$ denotes the binary $\mm$-successor

2616: of~$\a$ (Definition~\ref{D:Successor}), and that, for $\ww$ a braid word, $\CL\ww$

2617: denotes the braid represented by~$\ww$.

2618: 

2619: \begin{prop}

2620: \label{P:NormalBraid}

2621: A length~$\ell$ positive $\nn$-strand braid

2622: word~$\ww$ is $\Phi$-normal if and only if any

2623: one of the following equivalent conditions holds:

2624: 

2625: $(i)$ There exist binary addresses  $\a_\ell, \Ldots,

2626: \a_0$ with $\a_0 = 1^{\nn-2}$ such that,

2627: for each~$\kk$, $\ww(\kk) =

2628: \Gen{\a_\kk}$ holds, and $\a_\kk$ is

2629: the maximal binary successor of~$\a_{\kk-1}$ such that

2630: $\Gen{\a_\kk}$ is a right divisor of~$\CL{\ww(\ell)  ... \ww(\kk)}$.

2631: 

2632: $(ii)$ There exist numbers $\mm_\ell, \Ldots, \mm_1$ in

2633: $\{0, \Ldots, \nn\}$ such that, putting $\a_0= 1^{\nn-2}$

2634: and inductively defining  $\a_\kk=

2635: \BSucc{\mm_\kk}{\a_{\kk-1}}$, then, for each~$\kk$, we

2636: have $\ww(\kk)=\Gen{\a_\kk}$ and $\CL{\ww(\ell)  ... \ww(\kk)}

2637: \not\multe\Gen{\a}$ for every $\mm$-successor~$\a$ of~$\a_{\kk-1}$

2638: with $\mm>\mm_\kk$.

2639: 

2640: $(iii)$ There exist permutations $\perm_\ell, \Ldots,

2641: \perm_0$ of~$\{1, \Ldots, \nn-1\}$ such that $\perm_0$ is

2642: the identity, and, for each~$\kk$, we have

2643: $\ww(\kk) = \ss{\perm_\kk(1)}$ and $\perm_\kk$ is obtained

2644: from~$\perm_{\kk-1}$ as follows: let $\pp$ be minimal

2645: satisfying~$\CL{\ww(\ell)  ... \ww(\kk)}

2646: \multe \ss{\perm_{\kk-1}(\pp)}$; then we have

2647: $\perm_\kk(1)= \perm_{\kk-1}(\pp)$, 

2648: $\perm_\kk(\qq)=\perm_{\kk-1}(\qq)$ for $\qq>\pp$,

2649: and $(\perm_\kk(2), \Ldots, \perm_\kk(\pp))$ is

2650: the increasing (\resp decreasing) enumeration of

2651: $\{\perm_{\kk-1}(1), \Ldots, \perm_{\kk-1}(\pp-1)\}$ if the

2652: latter are larger (\resp smaller) than~$\perm_{\kk}(1)$ in the usual ordering of integers.

2653: \end{prop}

2654: 

2655: \begin{proof}

2656: Point~$(i)$ is Definition~\ref{D:NF} and~$(ii)$ is a direct

2657: reformulation. As for~$(iii)$, $\perm_\kk$ is the

2658: enumeration of the names of the successors of~$\a_\kk$,

2659: starting from the bottom, \ie, for each~$\mm$, we have

2660: $\Gen{\BSucc\mm{\a_\kk}} =

2661: \ss\ii$ with $\ii = \perm_\kk(\nn-\mm-1)$. At each step, we

2662: select the maximal successor satisfying the divisibility

2663: requirement, hence, here, the first entry in the

2664: permutation~$\perm_{\kk-1}$; the updating rules come

2665: from the specific definition of the covering~$\BBB\nn$. 

2666: \end{proof}

2667: 

2668: As for complexity, a direct application of

2669: Proposition~\ref{P:Complexity2} gives:

2670: 

2671: \begin{prop}

2672: \label{P:BraidNF}

2673: Running on a positive $\nn$-strand braid word of

2674: length~$\ell$, the algorithm of Table~\ref{T:Algo} returns the

2675: $\Phi$-normal word that is equivalent to~$\ww$

2676: in~$O(\ell^2 \nn\log\nn)$ steps; in the meanwhile, it also

2677: determines the address list of~$\DDf\nn{\CL\ww}$.

2678: \end{prop}

2679: 

2680: \begin{proof}

2681: As for~$(ii)$, we recall from~\cite[Chapter 9]{Eps} that

2682: there exists a division algorithm running in

2683: time~$O(\ell\nn\log\nn)$.

2684: \end{proof}

2685: 

2686: We refer to Table~\ref{T:Algo} for the algorithm 

2687: determining the $\Phi$-normal form, and to

2688: Table~\ref{T:XNF} for the details of the computation

2689: for~$\D_4^2$. Note that, apart from the fact

2690: that letters come gathered in blocks in the former, the only

2691: difference between the unbracketing of the $\BBB\nn$-decomposition and the

2692: $\Phi$-normal form viewed as a sequence of letters is that the

2693: $\BBB\nn$-decomposition always finishes with a power of~$\ss1$,

2694: possibly~$\s_1^0$, \ie,~$1$: for instance, the $\Phi$-normal

2695: form of~$\ss2$ is~$\ss2$, \ie, the length one

2696: sequence~$(\ss2)$, while its $\BBB3$-decomposition is the

2697: length two sequence~$(\ss2, 1)$.

2698: 

2699: \subsection{A linear ordering on~$\BB\nn$}

2700: \label{S:FlipOrder}

2701: 

2702: As the monoid~$\BB2$ is isomorphic to~$\Nat$, it is

2703: equipped with a natural linear ordering. Now, as the

2704: $\nn$-splitting associates with every braid of~$\BB\nn$ a

2705: distinguished finite sequence of braids, of~$\BB\nno$, we can recursively define a linear

2706: ordering of~$\BB\nn$.

2707: 

2708: \begin{defi}

2709: \label{D:FlipOrder}

2710: For $\nn\ge2$, we define the relation~$\lf_\nn$

2711: on~$\BB\nn$ as follows:

2712: 

2713: $(i)$ For $\xx,\yy$ in~$\BB2$, we say that $\xx\lf_2\yy$ holds

2714: for $\xx=\s_1^\pp$ and $\yy=\s_1^\qq$ with $\pp<\qq$;

2715: 

2716: $(ii)$ For $\xx,\yy$ in~$\BB\nn$ with $\nn\ge3$, we say that

2717: $\xx\lf_\nn\yy$ holds if, letting $(\xx_\pp, \Ldots, \xx_1)$ and $(\yy_\qq, \Ldots,

2718: \yy_1)$ be the $\nn$-splittings of~$\xx$

2719: and~$\yy$, we have either $\pp<\qq$, or 

2720: $\pp=\qq$ and for some~$\rr \le \pp$ we have

2721: $\xx_{\rr'}=\yy_{\rr'}$ for $\pp \ge \rr' > \rr$ and

2722: $\xx_\rr \lf_\nno \yy_\rr$.

2723: \end{defi}

2724: 

2725: Thus, $\lf_\nn$ is a sort of lexicographic extension of the

2726: natural order on~$\BB2$, \ie, on~$\Nat$, via  splittings.

2727: The extension is not exactly lexicographic: before comparing

2728: componentwise, we first compare the lengths of the

2729: sequences, \ie, the $\nn$-breadths of the considered braids,

2730: a comparison method called $\ShortLex$ in~\cite{Eps}.

2731: 

2732: \begin{prop}

2733: \label{P:ShortLex}

2734: $(i)$ For~$\nn\ge2$, the relation~$\lf_\nn$ is a linear

2735: ordering of~$\BB\nn$, which is a well-ordering. For

2736: each braid~$\xx$, the immediate $\lf_\nn$-successor of~$\xx$

2737: is~$\xx\ss1$.

2738: 

2739: $(ii)$ For $\nn\ge3$, the order~$\lf_\nn$ extends the

2740: order~$\lf_\nno$, and $\BB\nno$ is the initial segment

2741: of~$\BB\nn$ determined by~$\ss\nno$, \ie, we

2742: have $\BB\nno = \{\xx\in\BB\nn \mid \xx\lf_\nn \ss\nno\}$. 

2743: \end{prop}

2744: 

2745: \begin{proof}

2746: $(i)$ The relation~$\lf_2$ is a linear ordering of~$\BB2$. Then, $\lf_\nn$ being a

2747: linear ordering of~$\BB\nn$ follows from $\lf_\nno$

2748: being a linear ordering of~$\BB\nno$ and

2749: the $\nn$-splitting being unique. That  $\lf_\nn$ is a

2750: well-order results from a similar induction, owing to the

2751: standard result that the

2752: $\ShortLex$-extension of a well-ordering is a well-ordering. Finally,

2753: if the $\nn$-splitting of~$\xx$ is $(\xx_\pp, \Ldots, \xx_1)$,

2754: the $\nn$-splitting of~$\xx\ss1$ is $(\xx_\pp, \Ldots,

2755: \xx_1\ss1)$, making it clear that

2756: $\xx\ss1$ is the immediate successor of~$\xx$.

2757: 

2758: $(ii)$ For $\xx, \yy$ in~$\BB\nno$, the

2759: $\nn$-splittings of~$\xx$ and~$\yy$ are

2760: the length one sequences~$(\xx)$ and~$(\yy)$, so, by

2761: definition,

2762: $\xx\lf_\nn\yy$ is equivalent to $\xx\lf_\nno\yy$. On the

2763: other hand, the $\nn$-splitting of~$\ss\nno$ is $(\ss1,

2764: 1)$, so $\xx\lf_\nn\ss\nno$ holds for each~$\xx$

2765: in~$\BB\nno$. Conversely, assume

2766: $\xx\in\BB\nn$ and $\xx\lf_\nn\ss\nno$. By construction,

2767: if $(\xx_2,\xx_1)$ is a $\nn$-splitting,

2768: $\xx_2$ is not~$1$, hence, by~$(i)$, we have

2769: $\xx_2\gfe_\nn\ss1$. So, if

2770: $\xx\lf_\nn\ss\nno$ holds, the only possibility is that the

2771: $\nn$-breadth of~$\xx$ is~$1$, \ie, that $\xx$ belongs

2772: to~$\BB\nno$.

2773: \end{proof}

2774: 

2775: Owing to Proposition~\ref{P:ShortLex}$(ii)$, we shall skip the

2776: index~$\nn$ and write~$\lf$ for~$\lf_\nn$.

2777: 

2778: \begin{exam}

2779: \label{X:Compar}

2780: The $3$-splittings of $\ss1$ and $\ss2$

2781: respectively are $(\ss1)$ and $(\ss1, 1)$, \ie, their

2782: respective $3$-breadths are~$1$ and~$2$. Hence we have

2783: $\ss1 \lf \ss2$.

2784: 

2785: Similarly, the $3$-splittings of $\D_3$ and

2786: $\s_1^2\s_2^2$ are $(\ss1, \ss1, \ss1)$ and $(\s_1^2, \s_1^2, 1)$. The $3$-breadth is~$3$ in both

2787: cases, and we compare lexicographically. The first entries

2788: are $\ss1$ and $\s_1^2$. The former is smaller, hence $\D_3 \lf \s_1^2\s_2^2$ holds.

2789: \end{exam}

2790: 

2791: The order~$\lf$ has been introduced above by means of the splitting. It can be introduced equivalently by appealing

2792: to the exponent sequence of the $\BBB\nn$-decomposition and to the following ordering of iterated sequences of

2793: integers.

2794: 

2795: \begin{defi}

2796: If $\uuu, \vvv$ are $\nn$-sequences of natural numbers, we say that $\uuu$ is $\ShortLex$-smaller than~$\vvv$,

2797: denoted $\uuu \lSL \vvv$, if we have $\nn = 0$ and $\uuu$ is smaller than~$\vvv$ with respect to the standard

2798: order on~$\Nat$, or $\nn \ge 1$ and either $\uuu$---viewed as a sequence of $(\nno)$-sequences---is shorter

2799: than~$\vvv$, or they have equal length and $\uuu$ is lexicographically smaller than~$\vvv$, \ie, writing $\uuu

2800: = (\uuu_\pp, \Ldots, \uuu_1)$ and $\vvv = (\vvv_\pp, \Ldots, \vvv_1)$, there exists $\rr \le \pp$ such that we

2801: have $\uuu_{\rr'} = \vvv_{\rr'}$ for $\pp \ge \rr' > \rr$ and $\uuu_\rr \lSL \vvv_\rr$.

2802: \end{defi}

2803: 

2804: \begin{lemm}

2805: \label{L:OrderCo1}

2806: For~$\xx,\yy$ in~$\BB\nn$, we have

2807: \begin{equation}

2808: \label{E:FlipOrder}

2809: \xx\lf\yy

2810: \quad\Longleftrightarrow\quad

2811: \DDfe\nn\xx \lSL \DDfe\nn\yy.

2812: \end{equation}

2813: \end{lemm}

2814: 

2815: \begin{proof}

2816: As the relations involved in both sides of ~\eqref{E:FlipOrder} are linear orderings, it is enough to prove one

2817: implication. We shall prove using  induction on~$\nn\ge2$ that $\xx\lf\yy$ implies $\DDfe\nn\xx \lSL

2818: \DDfe\nn\yy$. The result is obvious for $\nn=2$. Assume $\nn\ge3$ and $\xx \lf \yy$ in~$\BB\nn$. Let

2819: $(\xx_\pp, \Ldots, \xx_1)$ and $(\yy_\qq, \Ldots, \yy_1)$ be the

2820: $\nn$-splittings  of~$\xx$ and~$\yy$.

2821: By~\eqref{E:BiFlip21}, we have

2822: \begin{gather}

2823: \label{E:Expon1}

2824: \DDfe\nn\xx=(\DDfe\nno{\xx_\pp}, \Ldots,

2825: \DDfe\nno{\xx_1}), \quad

2826: \DDfe\nn\yy=(\DDfe\nno{\yy_\qq}, \Ldots,

2827: \DDfe\nno{\yy_1})

2828: \end{gather}

2829: ---as the names of the generators are forgotten, the flips do not appear in exponent

2830: sequences. According to the definition of~$\lf$, two cases are possible. If $\pp < \qq$ holds, then the left sequence

2831: in~\eqref{E:Expon1} is shorter than the right sequence, so $\DDfe\nn\xx \lSL\nobreak

2832: \DDfe\nn\yy$ holds. Otherwise, for some~$\rr \le \pp$, we must have $\xx_{\rr'} = \yy_{\rr'}$ for $\pp

2833: \ge \rr' > \rr$ and $\xx_\rr \lf \yy_\rr$. We deduce $\DDfe\nno{\xx_{\rr'}} = \DDfe\nno{\yy_{\rr'}}$ for $\pp

2834: \ge \rr' > \rr$ and, using the induction hypothesis, $\DDfe\nno{\xx_\rr} \lSL \DDfe\nno{\yy_\rr}$. Here again,

2835: we find $\DDfe\nn\xx \lSL \DDfe\nn\yy$.

2836: \end{proof}

2837: 

2838: For instance, we saw in Example~\ref{X:Compar} that

2839: $\D_3 \lf \s_1^2\s_2^2$ holds. Another way to

2840: see it is to compare $\DDfe3{D_3}$ and $\DDfe3{\s_1^2\s_2^2}$ with respect to~$\lSL$. The

2841: respective values are $(1,1,1)$ and $(2,2,0)$: the former is $\lSL$-smaller.

2842: 

2843: \subsection{The braids~$\DDhat\nn\pp$}

2844: 

2845: Few properties of the order~$\lf$ are visible

2846: directly. Typically, whether $\xx\lf\yy$ implies

2847: $\zz\xx\lf\zz\yy$ is unclear because we do not know much

2848: about the $\nn$-splittings of~$\zz\xx$ and~$\zz\yy$ as

2849: compared with those of~$\xx$ and~$\yy$. We shall come

2850: back on the question in Section~\ref{S:Burckel}.

2851: For the moment, we conclude this section with a technical result about~$\lf$, namely we determine the least upper

2852: bound of the braids of~$\BB\nn$ whose $\nn$-breadth is at most~$\pp$. 

2853: 

2854: \begin{nota}

2855: (See Figure~\ref{F:DeltaHat}) 

2856: For $\nn\ge2$ and $\dd\ge 1$, we set

2857: \begin{equation}

2858: \label{E:dddd}

2859: \ddd\nn=\ss\nno...\ss1

2860: \text{\  and \ }

2861: \DDhat\nn\dd=\ff\nn^{\dd+1}(\ddd\nn) \cdot  ...

2862: \cdot \ff\nn^2(\ddd\nn) \cdot \ff\nn(\ddd\nn).

2863: \end{equation}

2864: \end{nota}

2865: 

2866: In other words, $\DDhat\nn\dd$ is the

2867: length~$\dd(\nn-1)$ zigzag $...\ss\nno ... \ss1\ss1 ...

2868: \ss\nno$ with $\dd-1$ alternations, finishing

2869: with~$\ss\nno$. For instance, 

2870: $\DDhat42$ is the braid~$\ss3\ss2\s_1^2\ss2\ss3$.

2871: 

2872: \begin{figure}[htb]

2873: \begin{picture}(97,16)

2874: \put(0,-2){\includegraphics{DeltaHat.eps}}

2875: \end{picture}

2876: \caption{\sf The braids~$\DDhat34$ (left) and $\DDhat43$ (right):

2877: starting from the right, the upper strand of~$\DDhat\nn\dd$ forms $\dd$~half-twists around all other strands.}

2878: \label{F:DeltaHat}

2879: \end{figure}

2880: 

2881: \begin{lemm}

2882: \label{L:dddd}

2883: $(i)$ For $\nn\ge2$ and $\dd\ge 1$, we have

2884: \begin{equation}

2885: \label{E:Delta}

2886: \D_\nn^\dd = \DDhat\nn\dd\,\D_\nno^\dd.

2887: \end{equation}

2888: 

2889: $(ii)$ For~$\nn\ge2$,  $\dd\ge1$, and $\xx \in \BB\nno$,

2890: the $\nn$-splitting  of~$\DDhat\nn\dd \, \xx$ is

2891: \begin{equation}

2892: \label{E:dddd3}

2893: (\ss1 \ ,\ 

2894: \underbrace{\mbox{$\d_\nno\ss1 \ , \ ... \ , \

2895: \d_\nno\ss1$}}_{\dd-1\text{\ times}}\ ,\  \d_\nno \ ,\

2896: \xx).

2897: \end{equation}

2898: This holds in particular for~$\DDhat\nn\dd$

2899: with~$\xx=1$, and for~$\D_\nn^\dd$ with

2900: $\xx=\D_\nno^\dd$.

2901: \end{lemm}

2902: 

2903: \begin{proof}

2904: $(i)$ Among the many equivalent inductive definitions

2905: of~$\D_\nn$, we choose the recursive definition $\D_1= 1$ and $\D_\nn= \ss1...\ss\nno\D_\nno$, 

2906: \ie, $\D_\nn=\DDhat\nn1\D_\nno$, for $\nn \ge 2$. Then \eqref{E:Delta} holds

2907: for~$\dd=1$. For~$\dd\ge2$, we use induction:

2908: \begin{multline*}

2909: \D_\nn^\dd

2910: =\D_\nn\D_\nn^{\dd-1}

2911: =\D_\nn\DDhat\nn{\dd-1}\D_\nno^{\dd-1}

2912: =\ff\nn(\DDhat\nn{\dd-1})\D_\nn\D_\nno^{\dd-1}\\

2913: =\ff\nn(\DDhat\nn{\dd-1})\DDhat\nn1\D_\nno\D_\nno^{\dd-1}

2914: =\DDhat\nn\dd\D_\nno^\dd.

2915: \end{multline*}

2916: 

2917: $(ii)$ When we evaluate the sequence of~\eqref{E:dddd3}

2918: by flipping each other entry, we obtain~$\DDhat\nn\dd\,\xx$. On the other

2919: hand, each entry in~\eqref{E:dddd3} except possibly the last one is

2920: right divisible by~$\ss1$, and by no  other~$\ss\ii$. Hence, by

2921: Proposition~\ref{P:FSplitting}, the considered sequence is the

2922: $\nn$-splitting of the braid it represents.

2923: \end{proof}

2924: 

2925: In particular, the $3$-splitting of~$\D_3^\dd$ is 

2926: $(\ss1 ,\s_1^2 , \ ... \ , \s_1^2 ,\ss1 ,\s_1^\dd)$, $\dd-1$~times~$\s_1^2$, 

2927: which is $(\ss1, \ss1, \ss1)$ for $\dd=1$, corresponding to

2928: $\D_3=\ss1\ss2\ss1$, and $(\ss1, \s_1^2, \ss1, \s_1^2)$

2929: for $\dd=2$, corresponding to

2930: $\D_3^2=\ss2\s_1^2\ss2\s_1^2$. 

2931: 

2932: We shall see that $\DDhat\nn\ppo$ is the least upper bound for the

2933: braids of~$\BB\nn$ whose $\nn$-breadth at most~$\pp$. To prove this, we shall show that the $\nn$-splitting

2934: of~$\DDhat\nn\ppo$ is minimal among all $\nn$-splittings of length~$\pp+1$. Therefore, we first investigate the

2935: constraints satisfied by

2936: $\nn$-splittings.

2937: 

2938: \begin{lemm}

2939: \label{L:Small}

2940: For $\nn\ge2$, the braids in~$\BB\nn$ that satisfy

2941: $\xx\lf\ddd\nn$ are of those of the form

2942: $\ss\nno...\ss\mm\yy$ with $\nn\ge\mm\ge2$ and

2943: $\yy\in\BB{\mm-1}$.

2944: \end{lemm}

2945: 

2946: \begin{proof}

2947: We use induction on $\nn\ge2$. For $\nn=2$, we

2948: have $\ddd\nn=\ss1$, and the result is true, as $\xx\lf\ss1$

2949: implies $\xx=1$, and $1$ is the only element of~$\BB1$.

2950: Assume $\nn\ge3$, and $\xx\lf\ddd\nn$. The $\nn$-splitting 

2951: of~$\ddd\nn$ is $(\ss1,\ddd\nno)$. By definition, two

2952: cases are possible: either the $\nn$-breadth of~$\xx$

2953: is~$1$, which means that $\xx$ lies in~$\BB\nno$, or the

2954: $\nn$-breadth of~$\xx$ is~$2$ and, letting $(\xx_2,

2955: \xx_1)$ be its $\nn$-splitting, we have

2956: either $\xx_2\lf\ss1$, which is impossible, or

2957: $\xx_2=\ss1$ and $\xx_1\lf\ddd\nno$. In the latter

2958: case, by induction hypothesis, there exist~$\mm$ with

2959: $\nn-1\ge\mm\ge2$ and $\yy$ in~$\BB{\mm-1}$ such that

2960: $\xx_1= \ss{\nn-2}...\ss\mm\yy$ holds, and, then, we

2961: find $\xx=\ss\nno\ss{\nn-2}...\ss\mm\yy$.

2962: \end{proof}

2963: 

2964: \begin{prop}

2965: \label{P:Constraints}

2966: Assume that $(\xx_\pp, \Ldots, \xx_1)$ is the $\nn$-splitting of some braid in~$\BB\nn$. Then the following

2967: constraints are satisfied:

2968: \begin{equation}

2969: \label{E:Constraints}

2970: \xx_\pp \gfe \ss1, \quad

2971: \xx_\rr \gfe \ddd\nno \ss1

2972: \text{\ for $\pp > \rr \ge 3$}, \quad

2973: \xx_2\gfe \ddd\nno

2974: \text{\ if $\pp \ge 3$ holds}.

2975: \end{equation}

2976: \end{prop}

2977: 

2978: \begin{proof}

2979: First, we have $\xx_\pp \not= 1$ by hypothesis, hence $\xx_\pp \gfe \ss1$ by

2980: Proposition~\ref{P:ShortLex}$(i)$.

2981: 

2982: Then, $\xx_\rr$ is right divisible by~$\ss1$ for $\rr\ge2$.

2983: Indeed, by Proposition~\ref{P:AltDec}, we have $\xx_\rr \not=

2984: 1$, hence $\xx_\rr \multe \ss\ii$ for some~$\ii$. Now

2985: $\xx_\rr \multe \ss\ii$ implies

2986: $\ff\nn^{\pp-\rr}(\xx_\pp) \cdot  ... \cdot \xx_\rr \multe \ss\ii$,

2987: and $\ii\ge2$ would contradict the $\nn$-splitting

2988: condition of Proposition~\ref{P:FSplitting} at position~$\rr$. 

2989: 

2990: Assume $\pp > \rr\ge3$, and $\xx_\rr \lf \ddd\nno

2991: \ss1$. Write $\xx_\rr = \yy_\rr \ss1$. By

2992: Proposition~\ref{P:ShortLex}$(i)$,

2993: $\xx_\rr$ is the immediate successor of~$\yy_\rr$, so  $\xx_\rr

2994: \lf \ddd\nno \ss1$ implies $\yy_\rr \lf \ddd\nno$. By

2995: Lemma~\ref{L:Small}, we have $\yy_\rr = \ss{\nn-2} ...

2996: \ss\mm\yy$ with~$\yy$ in~$\BB{\mm-1}$ and

2997: $\nn-1\ge\mm\ge\nobreak2$. The condition $\xx_{\rr+1}

2998: \not= 1$ implies $\ff\nn(\xx_{\rr+1}) \multe

2999: \ss\nno$, hence $\ff\nn(\xx_{\rr+1}) \cdot \xx_\rr \multe

3000: \ss\nno \, ...\, \ss\mm \yy \ss1$. Assume first $\nn > \mm \ge

3001: 3$. Then $\ss{\mm}$ commutes with~$\yy_\rr$ and

3002: with~$\ss1$, and we obtain $\ff\nn(\xx_{\rr+1}) \cdot \xx_\rr

3003: \multe \ss\mm$, which contradicts the $\nn$-splitting

3004: condition at position~$\rr$. Assume now $\mm = 2$, hence

3005: $\yy=1$. Then we have

3006: $\ff\nn(\xx_{\rr+1}) \cdot \xx_\rr \multe \ss\nno \, ... \, \ss1$,

3007: hence 

3008: $$\xx_{\rr+1} \cdot \ff\nn(\xx_\rr) \cdot \xx_{\rr-1} \multe

3009: \ss1\, ... \, \ss\nno \xx_{\rr-1}.$$

3010: Now, for~$\ii

3011: \le \nn-2$, we have $\ss1\, ... \, \ss\nno \ss\ii = \ss{\ii+1}

3012: \ss1\, ... \, \ss\nno$, so there exists~$\xx'$ for which

3013: $\ss1\, ... \, \ss\nno \xx_{\rr-1} = \xx' \ss1\, ... \, \ss\nno$  holds. We deduce $\xx_{\rr+1} \cdot

3014: \ff\nn(\xx_\rr)

3015: \cdot \xx_{\rr-1} \multe \ss\nno$, contradicting the

3016: $\nn$-splitting condition at position~$\rr-1$.

3017: 

3018: Assume finally $\xx_2 \lf \ddd\nno$. By

3019: Lemma~\ref{L:Small}, we can write $\xx_2 = \ss{\nn-2}\, ... \, \ss\mm\xx$ with~$\xx$ in~$\BB{\mm-1}$ and

3020: $\nn-1\ge\mm\ge2$. As above, we deduce $\ff\nn(\xx_3)

3021: \cdot \xx_2 \multe \ss\nno\, ... \, \ss\mm \xx$. If $\nn > \mm

3022: \ge 3$ holds, $\ss\mm$ commutes with~$\xx$, and we obtain 

3023: $\ff\nn(\xx_3) \cdot \xx_2 \multe\nobreak \ss\mm$, which contradicts

3024: the $\nn$-splitting condition at position~$2$. For $\mm

3025: = 2$, hence $\xx=1$, we obtain $\ff\nn(\xx_3) \cdot \xx_2

3026: \multe \ss2$ directly, and the same contradiction.

3027: \end{proof}

3028: 

3029: \begin{prop}

3030: \label{P:UpperBound}

3031: For $\pp\ge1$, the braid~$\DDhat\nn\ppo$ is the $\lf$-least upper bound of the elements of~$\BB\nn$ whose

3032: $\nn$-breadth is at most~$\pp$.

3033: \end{prop}

3034: 

3035: \begin{proof}

3036: By Lemma~\ref{L:dddd}$(ii)$, $\DDhat\nn\ppo$ has

3037: $\nn$-breadth~$\pp+1$, hence $\xx\lf\DDhat\nn\ppo$ holds

3038: for~$\xx$ with $\nn$-breadth at most~$\pp$.

3039: Conversely, assume that the $\nn$-breadth of~$\xx$ is at

3040: least~$\pp+1$. If it is~$\pp+2$ or more, then

3041: $\xx\gf\DDhat\nn\ppo$ holds by definition of~$\lf$.

3042: Otherwise, let $(\xx_{\pp+1}, \Ldots, \xx_1)$ be the

3043: $\nn$-splitting of~$\xx$. Proposition~\ref{P:Constraints} says

3044: that the sequence $(\xx_{\pp+1}, \Ldots, \xx_1)$ is at least

3045: $(\ss1, \ddd\nno\ss1, \Ldots, \ddd\nno\ss1, \ddd\nno,

3046: 1)$, which is the $\nn$-splitting of~$\DDhat\nn\ppo$.

3047: Hence we have $\xx\gfe\DDhat\nn\ppo$.

3048: \end{proof}

3049: 

3050: \section{Connection with the braid order}

3051: \label{S:Burckel}

3052: 

3053: Defining a unique normal representative is of little interest, unless the normal form has some specific

3054: additional properties that make it useful. At the moment, the most interesting

3055: property of the $\Phi$-normal form of braids

3056: seems to be its connection with the so-called Dehornoy order.

3057: 

3058: \subsection{The braid order}

3059: \label{S:Order}

3060: 

3061: We shall establish a simple connection between the

3062: $\lf$-ordering of~$\BB\nn$, \ie, the ordering deduced from

3063: the $\nn$-splitting, and the standard linear ordering of braids of~\cite{Dhr}. We recall the definition of the latter.

3064: Considering~$\BB\nno$ as a submonoid of~$\BB\nn$, we

3065: denote by~$\BB\infty$ the union of all~$\BB\nn$'s, and by~$B_\infty$

3066: the group of fractions of~$\BB\infty$, \ie, the braid group on

3067: unboundedly many strands.

3068: 

3069: \begin{defi}

3070: \label{D:Order}

3071: For $\xx,\yy$ in~$B_\infty$, we say that $\xx\ls\yy$

3072: holds if the braid~$\xx\inv\yy$ admits at least one

3073: word representative in which the generator~$\ss\ii$ with

3074: maximal index occurs positively only, \ie, $\ss\ii$ occurs

3075: but $\sss\ii$ does not.

3076: \end{defi}

3077: 

3078: \begin{thrm}

3079: \label{T:Order}

3080: $(i)$ \cite{Dfb}

3081: The relation~$\ls$ is a linear ordering of~$B_\infty$ that is

3082: compatible with multiplication on the left.

3083: 

3084: $(ii)$ \cite{Lve}

3085: The restriction of~$\ls$ to~$\BB\infty$ is a well-ordering.

3086: 

3087: $(iii)$ \cite{Bur}

3088: For each~$\nn\ge2$, the restriction of~$\ls$ to~$\BB\nn$,

3089: which is the interval $(1, \ss\nn)$ of~$(\BB\infty, \ls)$, is a

3090: well-ordering of type~$\om^{\om^{\nn-2}}$.

3091: \end{thrm}

3092: 

3093: In the framework of~\cite{Dhr}, the ordering of Definition~\ref{D:Order} is called the upper version of the braid

3094: order. In some sources, in particular the early ones, the lower variant is considered, namely the relation~$\lmin$

3095: referring to the letter~$\ss\ii$ with \emph{minimal} index,

3096: instead of maximal as above. Both relations are similar as $\xx\ls\yy$ is equivalent

3097: to $\ff\nn(\xx)\lmin\ff\nn(\yy)$ for all~$\xx, \yy$

3098: in~$B_\nn$. However, as first noted by S.\,Burckel

3099: in~\cite{BuT}, the statements involving

3100: the well-order property are more natural with~$\ls$.

3101: 

3102: \subsection{Adding brackets in a braid word}

3103: 

3104: In order to connect the braid orders~$\lf$ and~$\ls$,

3105: we shall compare the $\Phi$-normal form of Section~\ref{S:Flip}

3106: with some other normal form introduced by S.\,Burckel in his

3107: remarkable work, and we first need to introduce some notions

3108: from~\cite{Bur}. The original description of~\cite{Bur} is

3109: formulated in terms of trees. However, the latter are equivalent to the

3110: iterated sequences of Section~\ref{S:Iter}, and we can easily

3111: describe the fragment of Burckel's construction needed here

3112: in terms of iterated sequences. Here , we give a new

3113: description that is more directly connected to our

3114: approach. In terms of trees, this amounts to starting from the

3115: top and the right, while Burckel's approach starts from the

3116: bottom and the left. The equivalence of both descriptions is

3117: established in Proposition~\ref{P:CoincidenceNF} below.

3118: 

3119: Our basic observation here is that a free monoid is locally right Garside: this is a trivial result, as the

3120: right divisibility relation of a free monoid is simply the relation of being a suffix. Then, applying the decomposition

3121: process of Sections~\ref{S:Alt} and~\ref{S:Iter} to a word~$\ww$ in a free monoid amounts to grouping the letters

3122: of~$\ww$ into blocks, \ie, in adding brackets in~$\ww$. We shall consider the iterated covering of the free

3123: braid word monoids that mimicks the covering~$\BBB\nn$ of Section~\ref{S:Flip}.

3124: 

3125: \begin{nota}

3126: We denote by~$\WW\nn$ the free monoid consisting of all positive $\nn$-strand braid words, and by~$\WWW\nn$

3127: the atomic iterated covering of~$\WW\nn$ based on the sequence~$\GGen_\nn$---the same as in the case

3128: of~$\BBB\nn$.

3129: \end{nota}

3130: 

3131: We shall now use the $\WWW\nn$-decomposition of a word in~$\WW\nn$.

3132: As in Section~\ref{S:Flip}, it is convenient to take advantage of the recursive definition of

3133: the covering~$\WWW\nn$, and to introduce the counterpart of the $\nn$-splitting. 

3134: 

3135: \begin{defi}

3136: For $\nn \ge 3$ and $\ww$ in~$\WW\nn$, the \emph{$\nn$-splitting} of~$\ww$ is defined to be the unique

3137: sequence $(\ww_\pp, \Ldots, \ww_1)$ of words in~$\WW\nno$ such that $(\ff\nn^\ppo(\ww_\pp), \Ldots,

3138: \ff\nn(\ww_2), \ww_1)$ is the $(\ff\nn(\WW\nno), \WW\nno)$-decomposition of~$\ww$.

3139: \end{defi}

3140: 

3141: As being a right divisor in a free monoid is equivalent to being a suffix, Proposition~\ref{P:AltDec} implies that

3142: $(\ww_\pp, \Ldots, \ww_1)$ is the $\nn$-splitting of~$\ww$ if and only if, for each~$\rr$, the word~$\ww_\rr$ is

3143: the longest suffix of~$\ff\nn^{\pp-\rr}(\ww_\pp) \cdot ... \cdot \ww_\pp$ that lies in~$\WW\nno$. 

3144: 

3145: \begin{exam}

3146: \label{X:WordSplitting}

3147: Let $\ww$ be the $4$-strand braid word $\ss3\ss2\s_1^2\ss2\ss3

3148: \ss2\s_1^2\ss2 \s_1^2$. The longest suffix of~$\ww$ that lies in~$\WW3$ is $\ss2\s_1^2\ss2 \s_1^2$, and the 

3149: remaining prefix is $\ss3\ss2\s_1^2\ss2\ss3$, \ie,

3150: $\ff4(\ww^{(1)})$ with $\ww^{(1)} =

3151: \ss1\ss2\s_3^2\ss2\ss1$. The longest suffix of~$\ww^{(1)}$

3152: that does not contain~$\ss3$ is $\ss2\ss1$, with remaining

3153: prefix~$\ss1\ss2\s_3^2$, \ie, $\ff4(\ww^{(2)})$ with

3154: $\ww^{(2)} = \ss3\ss2\s_1^2$. The longest suffix

3155: of~$\ww^{(2)}$ that does not contain~$\ss3$ is

3156: $\ss2\s_1^2$, with remaining prefix~$\ss3$. So, by definition, the $4$-splitting of the word~$\ww$ is the sequence

3157: of $3$-strand braid words $(\ss1,\, \ss2\s_1^2,\, \ss2\ss1,\, \ss2\s_1^2\ss2 \s_1^2)$. 

3158: \end{exam}

3159: 

3160: Imitating for braid words the notation used for braids in Section~\ref{S:Flip}, we put:

3161: 

3162: \begin{nota}

3163: For $\ww$ in~$\WW\nn$, we denote the $\WWW\nn$-decomposition of~$\ww$ by~$\TTT\nn\ww$, and its

3164: exponent sequence by~$\TTTe\nn\ww$.

3165: \end{nota}

3166: 

3167: By construction, the iterated sequence~$\TTT\nn\ww$ is a certain bracketing of~$\ww$. Before giving an example,

3168: we note the following connection between the $\WWW\nn$-decomposition and the splitting.

3169: 

3170: \begin{lemm}

3171: \label{L:DecompWord}

3172: For $\nn \ge 3$ and  $\ww$ in~$\WW\nn$, we have

3173: \begin{equation}

3174: \label{E:DecompWord}

3175: \TTT\nn\ww=(\ff\nn^\ppo(\TTT\nno{\ww_\pp}), \Ldots,

3176: \ff\nn(\TTT\nno{\ww_2}), \TTT\nno{\ww_1}).

3177: \end{equation}

3178: where $(\ww_\pp, \Ldots, \ww_1)$ is the $\nn$-splitting of~$\ww$.

3179: \end{lemm}

3180: 

3181: The proof is exactly similar to that of Lemma~\ref{L:DecompBraid}.

3182: 

3183: \begin{exam}

3184: \label{X:Bracketing}

3185: Let again $\ww$ be the $4$-strand braid word $\ss3\ss2\s_1^2\ss2\ss3

3186: \ss2\s_1^2\ss2 \s_1^2$. We saw in Example~\ref{X:WordSplitting} that the $4$-splitting of~$\ww$ is

3187: $(\ss1, \ss2\s_1^2, \ss2\ss1, \ss2\s_1^2\ss2 \s_1^2)$. Then, we can easily see that the

3188: $3$-splitting of the word $\ss2\s_1^2\ss2 \s_1^2$ is $(\ss2, \s_1^2, \ss2, \s_1^2)$, etc.

3189: Using~\eqref{E:DecompWord}, we conclude that the $\WWW4$-decomposition of~$\ww$ is the

3190: $2$-sequence

3191: \begin{equation}

3192: \label{E:ExampleWord}

3193: \TTT4\ww = ( (\ss3),  (\ss2,\s_1^2),  (\ss2, \ss3),  (\ss2, \s_1^2, \ss2, \s_1^2)  ).

3194: \end{equation}

3195: \end{exam}

3196: 

3197: The braid word~$\ww$ considered in Example~\ref{X:Bracketing} is the $\Phi$-normal form of~$\D_4^2$.   By

3198: comparing~\eqref{E:XFlip} and~\eqref{E:ExampleWord}, we see that, up to identifying the word~$\s_\ii^\ee$ with

3199: the braid it represents, the $\WWW4$-decomposition of the word~$\ww$ is the

3200: $\BBB4$-decomposition of~$\D_4^2$. This phenomenon is general. 

3201: 

3202: \begin{lemm}

3203: \label{L:BracketNormal}

3204: If $\ww$ is a $\Phi$-normal $\nn$-strand braid word, we have

3205: $\TTT\nn\ww =

3206: \DDf\nn{\CL\ww}$.

3207: \end{lemm}

3208: 

3209: \begin{proof}

3210: We use induction on~$\nn$. For $\nn=2$, the result

3211: is obvious. Otherwise, let $(\xx_\pp, \Ldots,

3212: \xx_1)$ be the $\nn$-splitting of~$\CL\ww$, and, for

3213: each~$\rr$, let $\ww_\rr$ be the $\Phi$-normal form

3214: of~$\xx_\rr$. By construction, each word~$\ww_\rr$ with

3215: $\rr \ge 2$ finishes with~$\ss1$, so $(\ww_\pp, \Ldots, \ww_1)$

3216: is the $\nn$-splitting of~$\ww$. The induction hypothesis

3217: implies 

3218: $\TTT\nno{{\ww_\rr}} = \DDf\nno{\xx_\rr}$ for

3219: each~$\rr$. Applying~\eqref{E:DecompWord}, we deduce

3220: \begin{align*}

3221: \TTT\nn\ww 

3222: &=  (\ff\nn^\ppo(\TTT\nno{{\ww_\pp}}), \Ldots, 

3223: \ff\nn(\TTT\nno{{\ww_2}}), 

3224: \TTT\nno{{\ww_1}})\\

3225: &=(\ff\nn^\ppo(\DDf\nno{\xx_\pp}), \Ldots, 

3226: \ff\nn(\DDf\nno{\xx_2}),  

3227: \DDf\nno{\xx_1}).

3228: \end{align*}

3229: By~\eqref{E:BiFlip21}, the latter sequence

3230: is~$\DDf\nn{\CL\ww}$.

3231: \end{proof}

3232: 

3233: At this point, we can easily establish the connection between our current notion of $\WWW\nn$-decomposition and

3234: Burckel's notion of ``the tree of a braid word''.

3235: 

3236: \begin{lemm}

3237: \label{L:Bracketing}

3238: Assume $\nn\ge3$ and  $\ww \in \WW\nn$ with $\TTT\nn\ww = (\www_\pp, \Ldots,

3239: \www_1)$. Then, for $1 \le \ii \le \nno$, and assuming $\www_\pp =\nobreak \TTT\nno{\ww_\pp}$, we have 

3240: $$\TTT\nn{{\ss\ii}\ww} = 

3241: \begin{cases}

3242: ((...(\ss\ii )...),  \www_\pp, \Ldots, \www_1)

3243: &\text{for $\pp$ even and $\ii = 1$},\\

3244: &\text{\quad and for $\pp$ odd and $\ii=\nno$},\\

3245: {(\TTT\nno{\ff\nn^\ppo(\ss\ii) \ww_\pp}, 

3246: \www_\ppo, \Ldots, \www_1)}

3247: &\text{otherwise}.

3248: \end{cases}

3249: $$

3250: \end{lemm}

3251: 

3252: \begin{proof}

3253: Let $(\ww_\pp, \Ldots, \ww_1)$ be the $\nn$-splitting of~$\ww$. Then the $\nn$-splitting of~$\ss\ii \ww$ is

3254: $(\ss1, \ww_\pp, \Ldots, \ww_1)$ for $\pp$ even and $\ii = 1$, and for $\pp$ odd and $\ii = \nno$,

3255: and it is $(\ff\nn^\ppo(\ss\ii) \, \ww_\pp,  \ww_\ppo, \Ldots,

3256: \ww_1)$ otherwise. Indeed, the point is whether the additional letter~$\ss\ii$ can be

3257: incorporated in the same entry as~$\ww_\pp$. Taking the

3258: flips into account, this depends on whether $\ff\nn^\ppo(\ss\ii)$

3259: is~$\ss\nno$ or not. The value of~$\TTT\nn{{\ss\ii}\ww}$ directly follows. 

3260: \end{proof}

3261: 

3262: As the rule of Lemma~\ref{L:Bracketing} directly mimicks the inductive construction of the tree associated

3263: with~$\ww$ in the sense of~\cite{Bur}, we deduce:

3264: 

3265: \begin{prop}

3266: \label{P:CoincidenceNF}

3267: For each positive $\nn$-strand braid word~$\ww$, the tree

3268: associated with~$\TTT\nn\ww$ coincides with the tree of~$\ww$ as defined in~\cite{Bur}. 

3269: \end{prop}

3270: 

3271: Before going to Burckel's results, let us observe that the braid ordering~$\lf$ of  of Definition~\ref{D:FlipOrder} 

3272: admits a simple characterization in terms of $\Phi$-normal words. 

3273: 

3274: \begin{prop}

3275: \label{P:Connection1}

3276: For all~$\xx,\yy$ in~$\BB\nn$, we have

3277: \begin{equation}

3278: \label{E:Connection1}

3279: \xx\lf\yy \quad\Longleftrightarrow\quad

3280: \TTTe\nn\uu \lSL\TTTe\nn\vv,

3281: \end{equation}

3282: where $\uu$ and~$\vv$ are the $\Phi$-normal representatives

3283: of~$\xx$ and~$\yy$.

3284: \end{prop}

3285: 

3286: \begin{proof}

3287: By Lemma~\ref{L:BracketNormal}, we have

3288: $\DDf\nn\xx=\TTT\nn\uu$ and 

3289: $\DDf\nn\yy=\TTT\nn\vv$, so $\xx\lf\yy$, which is equivalent to

3290: $\DDfe\nn\xx\lSL\DDfe\nn\yy$ by Lemma~\ref{L:OrderCo1}, is also equivalent to

3291: $\TTTe\nn\uu \lSL\TTTe\nn\vv$.

3292: \end{proof}

3293: 

3294: \begin{rema}

3295: For $\ww$ in~$\WW\nn$, define a $\WWW\nn$-bracketing of~$\ww$ to be any $(\nnt)$-sequence~$\www$ such

3296: that the unbracketing of~$\www$ is~$\ww$ and, for each address~$\h$ of length~$\nn\!-\!2$, the

3297: entry~$\www_\h$ belongs to~$\WW{\cl\h}$ when it exists. So a $\WWW\nn$-bracketing of~$\ww$ is any way of

3298: adding brackets in~$\ww$ so that the resulting iterated sequence has its entries correctly dispatched with respect to

3299: the skeleton of the iterated covering~$\WWW\nn$. By construction, $\TTT\nn\ww$ is always a

3300: $\WWW\nn$-bracketing of~$\ww$, but it is not the only one. For instance, both $(\s_2^2, \ss1)$ and $(\ss2, \e,

3301: \ss2, \e, \e, \ss1)$ are

3302: $\WWW3$-bracketings of the word~$\s_2^2\ss1$. Then it is easy to check that $\TTT\nn\ww$ is, among all

3303: $\WWW\nn$-bracketings of~$\ww$, the one that has the $\lSL$-smallest exponent sequence. 

3304: \end{rema}

3305: 

3306: \subsection{The Burckel normal form}

3307: 

3308: We now appeal to Burckel's results in~\cite{Bur} to state a connection between the braid ordering~$\ls$ and the

3309: $\lSL$-ordering of the exponent sequences.  

3310: 

3311: \begin{defi}

3312: A positive $\nn$-strand braid word~$\ww$ is said to

3313: be \emph{Burckel normal} if the exponent

3314: sequence~$\TTTe\nn\ww$ is

3315: $\lSL$-minimal among all~$\TTTe\nn{\ww'}$

3316: with~$\ww'\equiv\ww$.

3317: \end{defi}

3318: 

3319: \begin{exam}

3320: Let us consider the two positive $3$-strand braid words that represent~$\D_3$, namely $\ss1\ss2\ss1$ and

3321: $\ss2\ss1\ss2$. Then we find $\TTT3{\ss1\ss2\ss1} = (\ss1, \ss2, \ss1)$, and $\TTT3{\ss2\ss1\ss1} = (\ss2, \ss1,

3322: \ss2, \e)$---here we use the empty word $\e$ to emphasize that we consider words. So we have $\TTTe3{\ss1 \ss2

3323: \ss1} = (1, 1, 1)$, and $\TTTe3{\ss2 \ss1 \ss2} = (1, 1, 1, 0)$. As $(1, 1, 1)$ is shorter, hence $\lSL$-smaller, than

3324: $(1, 1, 1, 0)$, we conclude that $\ss1 \ss2 \ss1$ is Burckel normal, while $\ss2\ss1\ss2$ is not.

3325: \end{exam}

3326: 

3327: Burckel normal words are called \emph{irreducible}

3328: in~\cite{Bur}. As the $\ShortLex$-ordering of $\nn$-sequences on~$\Nat$ is a

3329: well-ordering, each nonempty set of $\nn$-sequences in~$\Nat$

3330: contains a $\lSL$-least element. Therefore, each

3331: positive braid admits a unique Burckel normal representative. 

3332: 

3333: \begin{thrm} [Burckel, \cite{Bur}]

3334: \label{T:Burckel}

3335: For~$\xx,\yy$ in~$\BB\nn$, we have

3336: \begin{equation}

3337: \label{E:Connection2}

3338: \xx\ls\yy \quad\Longleftrightarrow\quad

3339:  \TTTe\nn\uu \lSL \TTTe\nn\vv,

3340: \end{equation}

3341: where $\uu$ and~$\vv$ are the Burckel normal

3342: representatives of~$\xx$ and~$\yy$.

3343: \end{thrm}

3344: 

3345: Burckel's proof of Theorem~\ref{T:Burckel} is quite subtle for $\nn\ge4$ and requires

3346: a transfinite induction. The point is to define a combinatorial operation called reduction so

3347: that, if a braid word~$\ww$ is not Burckel normal, then its

3348: reduct~$\ww'$ is equivalent to~$\ww$ and satisfies

3349: $\TTTe\nn{\ww'} \lSL \TTTe\nn\ww$. 

3350: 

3351: In the sequel, we shall only use the following consequence of

3352: Theorem~\ref{T:Burckel}.

3353: 

3354: \begin{coro}

3355: \label{C:Burckel}

3356: If $\uu$ and $\vv$ are the Burckel normal representatives

3357: of~$\xx$ and~$\xx\ss\ii$, then $\TTTe\nn\uu \lSL

3358: \TTTe\nn\vv$ holds.

3359: \end{coro}

3360: 

3361: \begin{proof}

3362: By definition, we have $\xx \ls \xx\ss\ii$, as the quotient $\xx\inv \xx\ss\ii$ has an expression,

3363: namely~$\ss\ii$, in which the generator with highest index appears positively only.

3364: \end{proof}

3365: 

3366: \subsection{Connecting the normal forms}

3367: \label{S:Connection}

3368: 

3369: At this point, two distinguished word representatives have

3370: been introduced for each positive braid, namely its

3371: $\Phi$-normal form, and its Burckel normal form.  We shall now

3372: prove that these \emph{a priori} unrelated normal representatives actually

3373: coincide.

3374: 

3375: \begin{prop}

3376: \label{P:Connection}

3377: The Burckel normal form coincides with the $\Phi$-normal

3378: form.

3379: \end{prop}

3380: 

3381: \begin{proof}

3382: As each braid admits a unique Burckel normal

3383: representative and a unique $\Phi$-normal representative,

3384: proving one implication is sufficient. Here we prove using

3385: induction on~$\nn\ge2$ that an $\nn$-strand braid word that is not

3386: $\Phi$-normal is not either Burckel normal. For $\nn=2$, every

3387: word, namely every power of~$\ss1$, is normal in both

3388: senses. Assume $\nn\ge3$, and assume that $\ww$ is a word in~$\WW\nn$ that is not $\Phi$-normal. We aim at

3389: proving that $\ww$ is not Burckel normal. Owing to the definition of a Burckel normal word, it is enough to exhibit

3390: a word~$\ww'$ that represents the same braid as~$\ww$ and is such that $\TTTe\nn{\ww'}$ is

3391: $\ShortLex$-smaller than~$\TTTe\nn\ww$.

3392: 

3393: Let $(\ww_\pp, \Ldots, \ww_1)$ be the $\nn$-splitting of~$\ww$. By

3394: Lemma~\ref{L:DecompWord}, the value of~$\TTTe\nn\ww$ is

3395: \begin{equation}

3396: \label{E:FirstSeq}

3397: (\TTTe\nno{{\ww_\pp}}, \Ldots,

3398: \TTTe\nno{{\ww_2}},

3399: \TTTe\nno{{\ww_1}})

3400: \end{equation} 

3401: ---as we consider exponent sequences, we can forget about flips. The hypothesis that $\ww$ is not $\Phi$-normal may

3402: have two causes, namely that one of the words~$\ww_\rr$ is not $\Phi$-normal, or that all words~$\ww_\rr$ are $\Phi$-normal but $(\CL{\ww_\pp}, \Ldots, \CL{\ww_1})$ is not the $\nn$-splitting of the braid~$\CL\ww$.

3403: 

3404: Assume first that some word~$\ww_\rr$ is not $\Phi$-normal. By induction hypothesis, $\ww_\rr$ is not Burckel

3405: normal either. Hence there exists a word~$\ww'_\rr$ equivalent to~$\ww_\rr$ satisfying 

3406: $$\TTTe\nno{\ww'_\rr} \lSL \TTTe\nno{\ww_\rr}.$$

3407: Let $\ww'$ be the word obtained from~$\ww$ by replacing the subword~$\ff\nn^\rro(\ww_\rr)$

3408: with $\ff\nn^\rro(\ww'_\rr)$. Then

3409: $\ww'$ is equivalent to~$\ww$, and, by construction, one has

3410: $$\TTTe\nn{\ww'} \lSL \TTTe\nn{\ww},$$

3411: hence $\ww$ cannot be Burckel normal.

3412: 

3413: Assume now that each word~$\ww_\rr$ is $\Phi$-normal and $(\CL{\ww_\pp}, \Ldots, \CL{\ww_1})$ is not the

3414: $\nn$-splitting of~$\CL\ww$. Then there exists~$\rr$ such that the

3415: braid represented by 

3416: $$\vv = \ff\nn^{\pp-\rr}(\ww_\pp) \cdot ... \cdot \ff\nn(\ww_{\rr+1}) \cdot \ww_\rr$$

3417:  is right divisible by some~$\ss\ii$ with $\ii \ge 2$. We shall show that the factor~$\ss\ii$ can be

3418: removed from~$\ww_\rr$ and incorporated in the next factor~$\ww_\rro$,

3419: so as to give rise to a new word~$\ww'$ equivalent to~$\ww$ and satisfying $\TTTe\nn{\ww'} \lSL

3420: \TTTe\nn\ww$---see Figure~\ref{F:Proof}.

3421: 

3422: Indeed, let $\vv'$ be the Burckel normal form of~$\CL{\vv} \OVER{\s_\ii}$, and let $\ww'$ be the word

3423: $\ff\nn^\rro(\vv') \cdot \ff\nn^\rrt(\ss{\nn-\ii} \ww_\rro) \cdot ... \cdot \ff\nn(\ww_2) \cdot \ww_1$. By

3424: construction, $\ww'$ is equivalent to~$\ww$. The $\nn$-splitting of~$\vv$ is $(\ww_\pp, \Ldots, \ww_\rr)$.

3425: Let $(\ww'_{\pp'}, \Ldots, \ww'_\rr)$ be that of~$\vv'$. Then the $\nn$-splitting of~$\ww'$ is

3426: $(\ww'_{\pp'}, \Ldots, \ww'_\rr, \ss{\nn-\ii}\ww_\rro, \ww_\rrt, \Ldots, \ww_1)$, 

3427: and so, by Lemma~\ref{L:DecompWord}, the value of~$\TTTe\nn{\ww'}$ is

3428: \begin{multline}

3429: \label{E:SecondSeq}

3430: (\TTTe\nno{\ww'_{\pp'}}, \Ldots, \TTTe\nno{\ww'_\rr}, 

3431: \TTTe\nno{\ss{\nn-\ii}\ww_\rro},  \TTTe\nno{\ww_\rrt}, \Ldots, \TTTe\nno{\ww_1}).

3432: \end{multline}

3433: Now---this is the point---Corollary~\ref{C:Burckel} implies $\TTTe\nn{\vv'}

3434: \lSL \TTTe\nn\vv$, \ie, always by  Lemma~\ref{L:DecompWord}, 

3435: $$(\TTTe\nno{\ww'_{\pp'}}, \Ldots, \TTTe\nno{\ww'_\rr}) 

3436: \lSL (\TTTe\nno{\ww_\pp}, \Ldots, \TTTe\nno{\ww_\rr})$$

3437: ---hence in particular $\pp' \le \pp$. Adding~$\rr-1$ entries on the right of the above sequences does not change

3438: their order, and we deduce that the sequence of~\eqref{E:SecondSeq} is $\ShortLex$-smaller than that

3439: of~\eqref{E:FirstSeq}, \ie, $\TTTe\nn{\ww'}$ is $\ShortLex$-smaller than~$\TTTe\nn\ww$. This shows that $\ww$

3440: is not Burckel normal.

3441: \end{proof}

3442: 

3443: \begin{figure}[htb]

3444: \begin{picture}(106,25)

3445: \put(0,3){\includegraphics{Proof.eps}}

3446: \put(20,23){$\TTT\nn\ww$}

3447: \put(75,23){$\TTT\nn{\ww'}$}

3448: \put(2,0){$\ww_\pp$}

3449: \put(17,0){$\ww_\rr$}

3450: \put(27,0){$\ww_\rro$}

3451: \put(44,0){$\ww_1$}

3452: \put(44,0){$\ww_1$}

3453: \put(62,0){$\ww'_{\pp'}$}

3454: \put(71,0){$\ww'_\rr$}

3455: \put(80,0){$\ss\ii$}

3456: \put(84,0){$\ww_\rro$}

3457: \put(101,0){$\ww_1$}

3458: \end{picture}

3459: \caption{\sf Proof of Proposition~\ref{P:Connection}: if

3460: $\ww$ is not $\Phi$-normal because some $\ss\ii$ with $\ii \ge

3461: 2$ right divides the braid associated with the $\pp\!-\nobreak\!\rr\!+\nobreak\!1$

3462: left factors, then that $\ss\ii$ can be removed from

3463: the left part and incorporated in the next factor; Corollary~\ref{C:Burckel} guarantees that the new left part is

3464: smaller than the old one, so the new word~$\ww'$ is

3465: equivalent to~$\ww$ but its exponent sequence is

3466: smaller than that of~$\ww$.}

3467: \label{F:Proof}

3468: \end{figure}

3469: 

3470: \begin{rema*}

3471: It is natural to wonder whether  Proposition~\ref{P:Connection} extends to every dense atomic

3472: covering~$\MMM$ of a locally right monoid~$\MM$, \ie, whether the $\MMM$-normal form of an element~$\xx$

3473: of~$\MM$ is always the representative whose $\underline{\MMM}$-decomposition---defined in the

3474: obvious way from the considered atoms---has the $\lSL$-minimal exponent sequence. This is

3475: \emph{not} the case. Indeed, as in Remark~\ref{R:Shortest}, consider the $2$-covering~$\MMM$ of~$\BB4$

3476: based on $((\ss3,\ss2), (\ss3,\ss1))$. Then, the $\MMM$-normal form of

3477: $\ss3\ss2\ss1\ss2\s_3^2\ss2$ turns out to be the word $\ss1\s_2^2\ss1\ss3\ss2\ss1$. Now the

3478: $\underline{\MMM}$-decompositions of the words

3479: $\ss3\ss2\ss1\ss2\s_3^2\ss2$ and $\ss1\s_2^2\ss1\ss3\ss2\ss1$ respectively are

3480: $$(  (\ss3,\ss2), (\ss1), (\ss2, \s_3^2, \ss2), (\e)

3481:  )

3482: \text{\quad and \quad}

3483: ( (\ss1), (\s_2^2), (\ss1), (\ss3, \ss2),g (\ss1)  ),$$ with lengths $4$ and $5$---this is the same

3484: example as in Remark~\ref{R:Shortest}. The latter has a

3485: $\lSL$-larger exponent sequence, so the $\MMM$-normal

3486: form does not correspond to the smallest exponent sequence.  Technically, the point is that the counterpart to

3487: Corollary~\ref{C:Burckel} fails: the breadth may decrease under right

3488: multiplication. For instance, for $\yy =

3489: \ss1\s_2^2\ss1\ss3\ss2\s_1^2$, the

3490: $\MMM$-decomposition of~$\yy$ is $(\ss1 , \s_2^2 ,

3491: \ss1\ss3 , \ss2 , \s_1^2)$, which has length~$5$, while that

3492: of~$\yy\ss2$ is $(\ss3\s_2^2 , \ss1 , \ss2\s_3^2\ss2 , \ss1)$,

3493: which has length~$4$. This shows that the covering~$\BBB\nn$ is quite specific.

3494: \end{rema*}

3495: 

3496: \subsection{Applications}

3497: 

3498: Once we know that the $\Phi$-normal form and the Burckel

3499: normal form coincide, each one inherits the properties

3500: of the other, and we easily deduce several consequences, in

3501: particular in terms of braid orderings.

3502: 

3503: \begin{prop}

3504: \label{P:CompOrder}

3505: For~$\xx,\yy\in\BB\infty$, the relations $\xx\ls\yy$

3506: and $\xx\lf\yy$ are equivalent.

3507: \end{prop}

3508: 

3509: \begin{proof}

3510: Let $\uu$ and $\vv$ be the $\Phi$-normal representatives

3511: of~$\xx$ and~$\yy$. By Proposition~\ref{P:Connection},

3512: $\uu$ and~$\vv$ also are the Burckel normal

3513: representatives of~$\xx$ and~$\yy$. The equivalences

3514: $$\xx \ls \yy 

3515: \quad\Longleftrightarrow\quad 

3516: \TTTe\nn\uu \lSL \TTTe\nn\vv

3517: \quad\Longleftrightarrow\quad 

3518: \xx \lf \yy$$

3519: then follow from Proposition~\ref{P:Connection1}

3520: and Theorem~\ref{T:Burckel}.

3521: \end{proof}

3522: 

3523: We deduce that the standard braid ordering~$\ls$ inherits the recursive definition of the

3524: ordering~$\lf$, which is Theorem~A$(ii)$ in the introduction:

3525: 

3526: \begin{coro}

3527: \label{C:RecOrder}

3528: Let $\xx, \yy$ be positive $\nn$-strand braids. Let $(\xx_\pp, \Ldots, \xx_1)$ and $(\yy_\qq,\Ldots,\yy_1)$

3529: be the $\nn$-splittings of~$\xx$ and~$\yy$. Then $\xx<\yy$ holds in~$\BB\nn$ if and only if we

3530: have either $\pp<\qq$, or $\pp=\qq$ and there exists~$\rr \le \pp$ such that we have $\xx_{\rr'}=\yy_{\rr'}$ for

3531: $\pp\ge\rr'>\rr$ and $\xx_\rr<\yy_\rr$ in~$\BB\nno$.

3532: \end{coro}

3533: 

3534: In the other direction, we deduce that the ordering~$\lf$ satisfies the known properties of the ordering~$\ls$:

3535: 

3536: \begin{coro}

3537: \label{C:Compat2}

3538: The order~$\lf$ is compatible with multiplication on the left, and $\xx\lf\xx\ss\ii$ always holds.

3539: \end{coro}

3540: 

3541: Futher consequences involve the algorithmic complexity. The following result deals with the braid order~$\ls$,

3542: and it is Theorem~B in the introduction.

3543: 

3544: \begin{coro}

3545: \label{C:Complexity} 

3546: For each~$\nn$, the braid order~$\ls$ on~$B_\nn$ can be

3547: decided in quadratic time: if $\ww$ is a (non necessarily

3548: positive) $\nn$-strand braid word of length~$\ell$, then

3549: whether $\CL\ww\gs1$ holds can be decided in

3550: time~$O(\ell^2\nn^3\log\nn)$.

3551: \end{coro}

3552: 

3553: \begin{proof}

3554: We first observe

3555: that, if $\uu,\vv$ are positive $\nn$-strand braid words of

3556: length at most~$\ell$, then $\CL\uu\ls\CL\vv$ can be

3557: decided in time~$O(\ell^2\nn\log\nn)$. Indeed, by

3558: Proposition~\ref{P:BraidNF}$(ii)$, we can compute the

3559: decompositions~$\DDf\nn{\CL\uu}$

3560: and~$\DDf\nn{\CL\vv}$ within the indicated amount of time;

3561: the extra cost of subsequently comparing the corresponding

3562: exponent sequences with respect to the

3563: $\ShortLex$-ordering is linear in~$\ell\nn$. 

3564: 

3565: If $\ww$ is an arbitrary $\nn$ strand braid word of

3566: length~$\ell$, according to~\cite[Chapter~9]{Eps}, we can

3567: find two positive braid words~$\uu, \vv$ of length

3568: in~$O(\ell\nn^2)$ such that $\ww$ is equivalent

3569: to~$\uu\inv\vv$ in time~$O(\ell^2\nn\log\nn)$. Then

3570: $\CL\ww\gs1$ is equivalent to $\CL\uu\ls\CL\vv$, which,

3571: by the above observation, can be decided in

3572: time~$O(\ell^2\nn^5\log\nn)$. Actually, we can lower the

3573: exponent of~$\nn$ to~$3$ because an upper bound for the

3574: computation of the $\Phi$-normal form is~$O(\ell\ell_c\nn\log\nn)$, where

3575: $\ell_c$ is the canonical length, \ie, the

3576: number of divisors of~$\D_\nn$ involved in the right

3577: greedy normal form. When we go from~$\ww$

3578: to~$\uu\inv\vv$, the canonical lengths of~$\uu$

3579: and~$\vv$ are bounded above by that of~$\ww$, leading

3580: to~$O(\ell\ell_c\nn^3\log\nn)$ for the whole comparison.

3581: \end{proof}

3582: 

3583: Finally, another application is that, for each~$\nn$, the Burckel normal form of a positive

3584: $\nn$-strand braid word can be computed in quadratic time w.r.t.\,the

3585: length of the initial word, which is clear from Proposition~\ref{P:BraidNF} and the fact that the Burckel normal form

3586: coincides the $\Phi$-normal form. In the approach of~\cite{Bur}, the Burckel normal form comes as the final result of an

3587: iterated reduction process whose convergence is guaranteed by the fact that an ordinal decreases, and no complexity

3588: analysis has been published so far.

3589: 

3590: \section{Open questions and further work}

3591: 

3592: \subsection{The $\Phi$-normal form}

3593: 

3594: We have seen in Proposition~\ref{P:Constraints} that an arbitrary sequence of braids in~$\BB\nno$ need not be the

3595: $\nn$-splitting of a braid in~$\BB\nn$. An obvious question is whether the constraints of 

3596: Proposition~\ref{P:Constraints} are sufficient conditions.

3597: 

3598: \begin{ques}

3599: Assume that $\xx_\pp, \Ldots, \xx_1$ are braids of~$\BB\nno$ that satisfy

3600: \begin{equation}

3601: \xx_\pp \gse \ss1, \quad

3602: \xx_\rr \gse \ddd\nno \ss1

3603: \text{\  for $\pp > \rr \ge 3$}, \quad

3604: \xx_2\gse \ddd\nno

3605: \text{\  if $\pp \ge 3$ holds}.

3606: \end{equation}

3607: Does there exist a braid in~$\BB\nn$ whose $\nn$-splitting is $(\xx_\pp, \Ldots, \xx_1)$?

3608: \end{ques}

3609: 

3610: The only case where a (positive) answer is known is $\nn = 3$.

3611: 

3612: \begin{prop}

3613: A sequence $(\s_1^{\ee_\pp}, \Ldots, \s_1^{\ee_1})$ is the $3$-splitting of a braid of~$\BB3$ if and only if the

3614: numbers~$\ee_\rr$ satisfy the inequalities: 

3615: \begin{equation}

3616: \ee_\pp \ge 1, \quad

3617: \ee_\rr \ge 2 \text{\ for $\pp > \rr \ge 3$}, \quad

3618: \ee_2\ge 1 \text{\  if $\pp \ge 3$ holds}.

3619: \end{equation}

3620: \end{prop}

3621: 

3622: \begin{proof}

3623: What remains to be shown is that, if at least one of the above conditions fails, then $(\s_1^{\ee_\pp}, \Ldots,

3624: \s_1^{\ee_1})$ is not a $3$-splitting. Now, by Lemma~\ref{L:Dense}, no gap may exist in a $3$-splitting, so

3625: $\ee_\rr = 0$ is impossible for $\pp > \rr \ge 2$.

3626: 

3627: On the other hand, assume $\ee_\rr = 1$ with $\pp > \rr \ge 3$. As we have $\s_1^{\ee_{\pp+1}} \ss2

3628: \s_1^{\ee_\ppo} = \s_1^{\ee_{\pp+1}-1} \s_2^{\ee_\ppo} \ss1 \ss2$, the braid $\ff3^{\ppo-\rr}(\s_1^{\ee_\rr})

3629: \cdot ... \cdot \s_1^{\ee_\ppo}$ is right divisible by~$\ss2$, contradicting the characteristic property of a

3630: $3$-splitting.

3631: \end{proof}

3632: 

3633: The result can be restated as

3634: 

3635: \begin{coro}

3636: \label{C:Normal3}

3637: Set $\emin_1 = 0$, $\emin_2 = 1$, and $\emin_\rr = 2$ for $\rr \ge 3$. Then a positive $3$-strand braid word

3638: $\s_{\cl\pp}^{\ee_\pp} \, ... \, \s_2^{\ee_2} \, \s_1^{\ee_1}$ with $\ee_\pp \ge 1$ is $\Phi$-normal if and only if the

3639: inequality $\ee_\rr \ge \emin_\rr$ is satisfied for all indices~$\rr$ except possibly~$\pp$.

3640: \end{coro}

3641: 

3642: \begin{rema}

3643: A priori, the $\Phi$-normal form of a positive braid is completely different from its right greedy normal form

3644: of~\cite[Chapter~9]{Eps}. However, it was observed by J.~Mairesse (private communication) that, in the case of

3645: $3$-strands, there is a rather simple connection: starting from the right greedy normal form of a positive $3$-strand

3646: braid, we can obtain its $\Phi$-normal form by replacing the final factor~$\D_3^\ee$ with its $\Phi$-normal

3647: form, and, depending on the parity of~$\ee$ and on the final letter in the next factor, possibly push some

3648: factors~$\s_\ii^\dd$ through~$\D_3^\ee$---see~\cite{Dhq} for details.

3649: \end{rema}

3650: 

3651: \subsection{The braid ordering}

3652: 

3653: The proof of Proposition~\ref{P:Connection} heavily

3654: relies on Burckel's Theorem~\ref{T:Burckel}, a highly

3655: non trivial combinatorial result in the general case.

3656: 

3657: \begin{ques}

3658: \label{Q:Final}

3659: Is there a direct proof for the following results?

3660: 

3661: $(i)$ The orders~$\lf$ and~$\ls$ coincide. 

3662: 

3663: $(ii)$ The order~$\lf$ is compatible with multiplication on

3664: the left.

3665: 

3666: $(iii)$ The relation $\xx\lf\xx\ss\ii$ always holds.

3667: \end{ques}

3668: 

3669: So far we have no general answer. We mention

3670: below partial results toward a positive answer to

3671: Question~\ref{Q:Final}$(i)$, namely proving that, for

3672: all braids~$\xx,\yy$, the relation $\xx\lf\yy$ implies

3673: $\xx\ls\yy$---as we are dealing with linear orders, one

3674: implication is enough. Here we consider special values

3675: for~$\yy$. By Proposition~\ref{P:ShortLex}$(ii)$, we already know that

3676: $\xx\lf\ss\nno$ is equivalent to~$\xx\ls\ss\nno$, as

3677: both are equivalent to

3678: $\xx\in\BB\nno$. Here is another result of this kind.

3679: 

3680: \begin{prop}

3681: \label{P:Compar}

3682: For every~$\xx$ in~$\BB\nn$, the relation

3683: $\xx\lf\DDhat\nn\dd$ implies $\xx<\DDhat\nn\dd$.

3684: \end{prop}

3685: 

3686: \begin{proof}

3687: Assume $\xx\lf\DDhat\nn\dd$. By

3688: Proposition~\ref{P:UpperBound}, the $\nn$-breadth of~$\xx$ is

3689: at most~$\dd+1$, and we can write $\xx =

3690: \ff\nn^\dd(\xx_{\dd+1}) \cdot ... \cdot

3691: \ff\nn(\xx_2) \cdot \xx_1$ for some $\xx_{\dd+1}, \Ldots, \xx_1$

3692: in~$\BB\nno$. An easy computation using~\eqref{E:Delta} and the equalities $\ff\nn(\xx_\rr\inv)=\D_\nn

3693: \xx_\rr\inv \D_\nn\inv$ gives

3694: \begin{equation}

3695: \label{E:Quotient}

3696: \xx\inv\,\DDhat\nn\dd = 

3697: \xx_1\inv \cdot \D_\nn\xx_2\inv \cdot \D_\nn\xx_3\inv ...

3698: \cdot \D_\nn \xx_{\dd+1}\inv \cdot \D_\nno^{-\dd}.

3699: \end{equation}

3700: This leads to an expression of the quotient

3701: $\xx\inv \DDhat\nn\dd$ in which the letter~$\ss\nno$

3702: occurs $\dd$~times, while neither

3703: $\sss\nno$ nor any letter~$\s_\jj^{\pm1}$ with

3704: $\jj\ge\nn$ does. Indeed, each factor~$\D_\nn$ admits a

3705: positive expression in which $\ss\nno$ occurs once, namely

3706: the one arising from the decomposition $\D_\nn=\DDhat\nn1

3707: \D_\nno$, while the negative factors~$\xx_\rr\inv$ and

3708: $\D_\nno^{-\dd}$ belong to~$B_\nno$ and therefore can

3709: be expressed using neither~$\ss{n-1}$ nor~$\sss\nno$.

3710: Therefore $\xx\ls\DDhat\nn\dd$ holds.

3711: \end{proof}

3712: 

3713: It is not hard to deduce that, for every~$\xx$ in~$\BB\nn$, the relation

3714: $\xx\lf\D_\nn^\dd$ implies $\xx<\D_\nn^\dd$, as well as various similar compatibility results between~$\lf$

3715: and~$\ls$. But, so far, we have no complete answer to Question~\ref{Q:Final}$(i)$ in the

3716: general case. 

3717: 

3718: It is however easy to provide such an answer in the case $\nn = 3$. Indeed, in this special case, the exact form of $\Phi$-normal words is known, and a direct computation similar to that of Proposition~\ref{P:Compar} shows that, for $\xx,

3719: \yy$ in~$\BB3$, the relation~$\xx \lf \yy$ implies that the braid

3720: $\xx\inv \yy$ are an expression where $\ss2$ occurs but $\sss2$ does not, or an expression where $\ss1$ occurs

3721: but none of~$\sss1, \ss2, \sss2$ does, \ie, that $\xx \ls \yy$ holds.

3722: 

3723: By~\cite{Lve} and~\cite{Bur}, we know that $(\BB3, \ls)$ is a well-ordering of order type~$\om^\om$. Hence

3724: the position of every braid of~$\BB3$ is unambiguously specified by an ordinal number, called the \emph{rank}

3725: of~$\xx$, namely the order type of the initial segment of~$(\BB3, \ls)$ determined by~$\xx$. Using the

3726: formula for the $\Phi$-normal form given in Corollary~\ref{C:Normal3}, we deduce the following explicit

3727: value for the rank of a $3$-strand braid.

3728: 

3729: \begin{prop}

3730: \label{P:Rank}

3731: The rank of the braid with $\Phi$-normal form $\s_{\cl\pp}^{\ee_\pp} \, ... \, \s_2^{\ee_2} \, \s_1^{\ee_1}$ in the

3732: well-ordering $(\BB3, \ls)$ is the ordinal number

3733: \begin{equation}

3734: \label{E:Rank}

3735: \om^\ppo \cdot {\ee_\pp} + \sum_{\rr = \ppo}^{\rr = 1} \om^\rro \cdot (\ee_\rr - \emin_\rr),

3736: \end{equation}

3737: where the (absolute) numbers~$\emin_\rr$ are those of Corollary~\ref{C:Normal3}.

3738: \end{prop}

3739: 

3740: \begin{proof}

3741: The point is to determine which $\Phi$-normal words correspond to braids smaller than the considered one. By

3742: Corollary~\ref{C:Normal3}, $\Phi$-normal words are characterized by the inequalities $\ee_\rr \ge \emin_\rr$ for

3743: $\rr < \pp$, and \eqref{E:Rank} follows.

3744: \end{proof}

3745: 

3746: For instance, we saw in Lemma~\ref{L:dddd} that the $3$-splitting of~$\D_3^\dd$ is the

3747: length~$\dd+\nobreak 2$ sequence $(\ss1, \s_1^2, \Ldots, \s_1^2, \ss1, \s_1^\dd)$. Proposition~\ref{P:Rank}

3748: shows that,  for each~$\dd$, the rank of~$\D_3^\dd$ in~$(\BB3, \ls\nobreak)$ is the ordinal~$\om^{\dd+1} +

3749: \dd$: only the initial~$1$ and the final~$\dd$ contribute here, as all intermediate exponents have the minimal legal

3750: value~$\emin_\rr$.

3751: 

3752: \begin{ques}

3753: \label{Q:Rank}

3754: Does there exist a similar explicit formula for the rank of an arbitrary positive braid in the

3755: well-ordering~$(\BB\infty, \ls)$?

3756: \end{ques}

3757: 

3758: We refer to~\cite{But} for partial results about Question~\ref{Q:Rank}, and to~\cite{Dhq} for further applications,

3759: consisting of unprovability statements involving braids.

3760: 

3761: \subsection{Artin--Tits monoids and other Garside monoids}

3762: 

3763: We proved in Section~\ref{S:Iter} that

3764: $\MMM$-decompositions exist in every locally right Garside

3765: monoid~$\MM$ in which enough closed submonoids exist.

3766: This is in particular the case for every Artin--Tits monoid

3767: with respect to the standard set of generators~$\SS$, as every

3768: subset of~$\SS$ generates a closed submonoid that is closed.

3769: Thus, dense atomic coverings exist for every Artin--Tits monoid~$\MM$, and each of them leads to

3770: $\MMM$-decompositions similar to those of Section~\ref{S:Flip}. Then, we can

3771: adapt Section~\ref{S:FlipOrder} and define a linear

3772: ordering~$<_{\MMM}$ of~$\MM$ using the

3773: $\ShortLex$-ordering on $\MMM$-decompositions.

3774: 

3775: \begin{ques}

3776: \label{Q:Artin}

3777: Let $\MM$ be an Artin--Tits monoid. Is any of the 

3778: linear orders~$<_{\MMM}$ invariant under left multiplication?

3779: \end{ques}

3780: 

3781: In type~$A_\nn$, \ie, if $\MM$ is a braid monoid,

3782: Corollary~\ref{C:Compat2} provides a positive  answer. But the

3783: proof depends on the connection between the

3784: orders~$\lf$ and~$\ls$ and it is quite specific. More general positive results would presumably entail a direct

3785: proof in the case of braids,

3786: \ie, an answer to Question~\ref{Q:Final}$(ii)$.

3787: 

3788: Another possible extension of the current approach consists in

3789: addressing braids again, but in connection with

3790: other monoids. Laver's proof of Theorem~\ref{T:Order}$(ii)$

3791: implies that the restriction of~$\ls$ to any finitely generated

3792: submonoid of~$B_\infty$ generated by conjugates of

3793: the~$\ss\ii$'s is a well-ordering. In particular, the

3794: restriction of~$\ls$ to the dual braid monoids

3795: of~\cite{BKL} is a well-ordering. The latter are Garside monoids,

3796: and they are directly relevant for our approach. Natural analogs to the $\Phi$-normal forms exist, and

3797: investigating their connection with the braid ordering is an

3798: obvious task, recently achieved by J.\,Fromentin in~\cite{Fro}. It turns out that the dual framework is more

3799: suitable than the standard one, in that a positive answer to the counterpart of Question~\ref{Q:Final} can be given,

3800: with a direct proof that requires no transfinite induction.

3801: 

3802: \subsection{Geometric and dynamic properties}

3803: 

3804: Not much is known about the $\Phi$-normal form

3805: of braids. As every braid admits a canonical

3806: decomposition as a fraction~$\xx\yy\inv$ with $\xx,\yy$ in~$\BB\infty$ with no common right divisor, we can

3807: extend the $\Phi$-normal form of~$\BB\infty$ into a unique normal form

3808: on~$B_\infty$. Experiments suggest that the behaviour of this

3809: normal form is rather different from that of the greedy

3810: normal form, and many questions arise about the geometry

3811: it induces on the Cayley graph of~$B_\nn$. In particular, it is

3812: natural to ask for a possible associated automatic structure.

3813: The answer seems to be negative.

3814: 

3815: \begin{prop}

3816: $(i)$ For each~$\nn$, the set of all (positive) $\Phi$-normal

3817: $\nn$-strand braid words is rational, \ie, recognized by a finite state

3818: automaton.

3819: 

3820: $(ii)$ For $\nn\ge3$, $\Phi$-normal words do not satisfy the Fellow Traveler Property \cite{Eps}

3821: with respect to multiplication on the right.

3822: \end{prop}

3823: 

3824: \begin{proof}[Proof (sketch)]

3825: $(i)$ By Proposition~\ref{P:NormalBraid}, a positive $\nn$-strand braid word~$\ww$ is $\Phi$-normal if and only if

3826: each letter occurring in~$\ww$ is the smallest~$\ss\ii$ that right divides the braid represented by the

3827: prefix finishing at that letter, with respect to an ordering of~$\{\ss1, \Ldots, \ss\nno\}$ that depends on the suffix

3828: starting at that letter (actually at the next one). It is easy to construct an automaton that, when reading a braid word,

3829: returns the set of all~$\ss\ii$ that right divide the braid represented by that word. Similarly, it is easy to

3830: construct a reversed automaton that, reading a braid word from the right, returns the local ordering 

3831: of~$\{\ss1, \Ldots,

3832: \ss\nno\}$ that is involved in the above construction. Standard techniques from the theory of automata enable one

3833: to mix both constructions, and to build an automaton that recognizes the family of all $\Phi$-normal $\nn$-strand

3834: braid words.

3835: 

3836: $(ii)$ For odd (\resp even) $\dd\ge0$, the $\Phi$-normal

3837: form of~$\D_3^\dd$ is~$\uu_\dd = \DDhat3\dd

3838: \s_1^\dd$, while that of~$\D_3^\dd \ss2$ is~$\vv_\dd =

3839: \ss1 \uu_\dd$ (\resp $\ss2 \uu_\dd$)---as $\DDhat\nn\dd$ is a braid that admits a unique positive word

3840: representative, there is no danger here in using the same notation for the word and the braid. For

3841: $\ell = 1,

3842: \Ldots, 3\dd+1$, the successive distances between the length~$\ell$

3843: prefixes of~$\uu_\dd$ and~$\vv_\dd$ turn out to be

3844: 

3845: $0, 2, 4, 4, 6, 6, \Ldots, 2(\dd-1), 2(\dd-1), 2\dd, 2\dd, 2\dd,

3846: 2(\dd-1), \Ldots, 6, 4, 2, 1$.\\ There is no uniform upper bound for

3847: the above distances, hence the $\kk$-Fellow Traveler Property

3848: fails for every~$\kk$.

3849: \end{proof}

3850: 

3851: Investigating the dynamical properties of the $\Phi$-normal

3852: form along the lines addressed in~\cite{BNV, NeV, MaM, Mai,

3853: MaM2} is also a natural task. The generic problem is to study

3854: growth and stabilization in random walks through~$B_\nn$

3855: or~$\BB\nn$:  one compares the successive normal forms,

3856: typically looking at whether the first factors become

3857: eventually constant. Each new normal form induces a new

3858: problem. Let

3859: $\bb(\xx)$ denote the $\nn$-breadth of~$\xx$, and

3860: $\cc_\rr(\xx)$ denote the $\rr$th entry, starting from the

3861: right, in the $\nn$-splitting of~$\xx$. 

3862: 

3863: \begin{ques}

3864: Let $\XX$ be the random walk

3865: through~$\BB\nn$ defined by $\XX_{\kk+1}=\nobreak \ss\ii\, \XX_\kk$

3866: with $\ii$ equidistributed in $\{1,\Ldots, \nn-1\}$. What are the

3867: distributions of $\frac1\kk\bb(\XX_\kk)$ and

3868: $\frac1\kk\lg{\cc_\rr(\XX_\kk)}$ for each fixed~$\rr$?

3869: \end{ques}

3870: 

3871: Experiments suggest that the length of~$\cc_0(\XX_\kk)$

3872: might grow like~$\kk/(\nn+2)$, while

3873: $\cc_\rr(\XX_\kk)$ with $\rr\ge1$ tends to stabilize

3874: to~$\ddd\nno\ss1$, of constant length,

3875: and $\bb(\XX_\kk)$ might be connected with~$\sqrt{\kk}$.

3876: 

3877: \begin{thebibliography}{9}

3878: 

3879: \def\Reff#1; #2; #3; #4; #5; #6; #7\par{%

3880: \bibitem{#1} #2, \emph{#3}, #4 {\bf #5} (#6) #7}

3881: 

3882: \def\Ref#1; #2; #3; #4\par{%

3883: \bibitem{#1} #2, \emph{#3}, #4}

3884: 

3885: \Reff Adj; S.I. Adyan; Fragments of the word Delta in a

3886: braid group; Mat. Zam. Acad. Sci. SSSR; 36-1; 1984;

3887: 25--34; translated Math. Notes of the Acad. Sci. USSR;

3888: 36-1 (1984) 505--510.

3889: 

3890: \Reff Art; E.~Artin; Theory of Braids; Ann. of Math.; 48;

3891: 1947; 101--126.

3892: 

3893: \Reff BNV; R.\,Bikbov, S.\,Nechaev \& A.\,Vershik; Statistical

3894: properties of locally free groups with applications to braid

3895: groups and growth of random heaps; Comm. Math.

3896: Phys.; 212-2; 2000; 469--501. 

3897: 

3898: \Reff BKL; J.~Birman, K.H. Ko \& S.J. Lee; A new approach

3899: to the word problem in the braid groups; Advances in

3900: Math.; 139-2; 1998; 322-353.

3901:    

3902: \Reff BrS; E.\,Brieskorn \& K.\,Saito; Artin-Gruppen und

3903: Coxeter-Gruppen; Invent. Math.; 17; 1972; 245--271.

3904: 

3905: \Ref BuT; S.\,Burckel; Le bon ordre sur les tresses

3906: positives; PhD Thesis, Universit\'e de Caen (1994).

3907: 

3908: \Reff Bur; S.\,Burckel; The wellordering on

3909: positive braids; J. Pure Appl. Algebra; 120-1; 1997; 1--17.

3910: 

3911: \Reff But; S.\,Burckel; Computation of the ordinal of

3912: braids; Order; 16; 1999; 291--304.

3913: 

3914: \Ref Dhq; L.\,Carlucci, P.\,Dehornoy, A.\,Weiermann; Unprovability statements 

3915: involving braids; Preprint; math.LO/0711.3785.

3916: 

3917: \Reff Cha; R.\,Charney; Artin groups of finite type are

3918: biautomatic; Math. Ann.; 292-4; 1992; 671--683.

3919: 

3920: \Reff CMW; R.\,Charney, J.\,Meier \& K.\,Whittlesey;

3921: Bestvina's normal form complex and the homology of

3922: Garside groups; Geom. Dedicata; 105; 2004; 171-188.

3923: 

3924: \Reff ChM; R.\,Charney \& J.\,Meier;

3925: The language of geodesics for Garside groups; Math. Zeitschr; 248; 2004; 495--509.

3926: 

3927: \Reff Dfb; P.\,Dehornoy; Braid groups and left

3928: distributive  operations; Trans. Amer. Math. Soc.; 345-1;

3929: 1994; 115--151. 

3930: 

3931: \Reff Dgk; P.\,Dehornoy; Groupes de Garside; 

3932: Ann. Scient. Ec. Norm. Sup.; 35; 2002; 267--306.

3933: 

3934: \Reff Dgp; P.\,Dehornoy; Complete positive group 

3935: presentations; J. of Algebra; 268; 2003; 156--197.

3936: 

3937: % \Ref Dgr; P. Dehornoy, I. Dynnikov, D. Rolfsen, B. Wiest;

3938: % Why are braids orderable?; Panoramas \& Synth\`eses

3939: % vol. 14, Soc. Math. France (2002).

3940: % 

3941: \Ref Dhr; P. Dehornoy, I. Dynnikov, D. Rolfsen, B. Wiest;

3942: Ordering Braids; Mathematical Surveys and Monographs, Amer. Math. Soc., to appear.

3943: 

3944: \Reff Dgl; P.\,Dehornoy \& Y. Lafont; Homology of

3945: Gaussian groups;  Ann. Inst. Fourier; 53-2; 2003;

3946: 1001--1052.

3947: 

3948: \Reff Dfx; P.\,Dehornoy \& L. Paris; Gaussian

3949: groups and Garside groups, two generalizations of Artin

3950: groups; Proc. London Math. Soc.; 79-3; 1999; 569--604.

3951: 

3952: \Ref DiM; F.\,Digne \& J.\,Michel; Garside and locally

3953: Garside categories; Preprint; math.GR/0612652.

3954: 

3955: \Reff ElM; E. A.\,Elrifai \& H.R.\,Morton; Algorithms for

3956: positive braids; Quart. J. Math. Oxford; 45-2; 1994;

3957: 479--497.

3958: 

3959: \Ref Eps; D.\,Epstein, J.\,Cannon, D.\,Holt, S.\,Levy,

3960: M.\,Paterson \& W.\,Thurston; Word Processing in Groups;

3961: Jones \& Bartlett Publ. (1992).

3962: 

3963: \Ref Fro; J.\,Fromentin; The cycling normal form on dual braid monoids;  Preprint; math.GR/0712.3836.

3964: 

3965: \Reff Gar; F. A.\,Garside; The braid group and

3966: other groups; Quart. J. Math. Oxford; 20-78; 1969;

3967: 235--254.

3968: 

3969: \Ref KaT; C.\,Kassel \& V.\,Turaev; Braid groups; Springer (2007).

3970: 

3971: \Reff Lve; R.\,Laver; Braid group actions on left

3972: distributive structures and well-orderings in the braid

3973: group; J. Pure Appl. Algebra; 108-1; 1996; 81--98.

3974: 

3975: \Reff Mai; J.\,Mairesse; Random walks on groups and 

3976: monoids with a Markovian harmonic measure; Electron. J.

3977: Probab.; 10; 2005; 1417--1441.

3978: 

3979: \Reff MaM; J.\,Mairesse \& F.\,Math\'eus; Random walks on 

3980: free products of cyclic groups; J. London Math. Soc.; 75-1;

3981: 2007; 47--66.

3982: 

3983: \Reff MaM2; J.\,Mairesse \& F.\,Math\'eus; Randomly 

3984: growing braid on three strands and the manta ray; Ann.

3985: Appl. Probab.; 17-2; 2007; 502--536.

3986: 

3987: \Reff Mic; J.\,Michel; A note on words in braid monoids; J.

3988: Algebra; 215; 1999; 366--377.

3989: 

3990: \Reff NeV; S.\,Nechaev \& R.\,Voituriez; Random walks on

3991: 3-strand braids and on related hyperbolic groups; J. Phys.

3992: A; 36-1; 2003; 43--66.

3993: 

3994: \Reff Pid; M.\,Picantin; The center of thin

3995: Gaussian groups; J. of Algebra; 245-1; 2001; 92--122.

3996: 

3997: \Reff Pik; M.\,Picantin; Automatic structures of torus

3998: knot groups; J. Knot Th. and its Ramif.; 12-6; 2003; 833-866.

3999: 

4000: \Ref Thu; W.\,Thurston; Finite state algorithms for the

4001: braid group; Circulated notes (1988).

4002: 

4003: \end{thebibliography}

4004: 

4005: \end{document}

4006: