1: \documentclass[10pt]{amsart}

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

3: \usepackage{epstopdf}

4: 

5: \renewcommand{\subjclassname}{%

6: \textup{2000} Mathematics Subject Classification}

7: 

8: \hoffset-1.2cm

9: \voffset+0.5cm

10: 

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

12: 

13: \theoremstyle{plain}

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

15: \newtheorem{theo}[prop]{Theorem}

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

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

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

19: 

20: \theoremstyle{definition}

21: \newtheorem{conv}[prop]{Convention}

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

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

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

25: \newtheorem*{rema}{Remark}

26: 

27: \numberwithin{equation}{section}

28: 

29: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

30: 

31: \renewcommand\aa{a}

32: \newcommand\act{\cdot}

33: \newcommand\bb{b}

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

35: \newcommand\Bi{B_\infty}

36: \newcommand\br[1]{\widehat{#1}}

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

38: \newcommand\DD[1]{\Delta_{#1}}

39: \newcommand\divel{\preccurlyeq_{\scriptscriptstyle\! L}}

40: \newcommand\dual[1]{{}^*\!#1}

41: \newcommand{\e}{\varepsilon}

42: \newcommand{\ee}{e}

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

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

45: \newcommand\FDyn{F}

46: \newcommand\ff{f}

47: \newcommand\flip{\phi}

48: \newcommand\gcdl{\mathrm{gcd}_{\scriptscriptstyle\! L}}

49: \newcommand\GDyn{\overline{F}}

50: \let\ge=\geqslant

51: \renewcommand\gg{g}

52: \newcommand\gives{\rightsquigarrow}

53: \newcommand\HH{\alpha}

54: \newcommand\id{\mathrm{id}}

55: \def\ie{{\it i.e.}}

56: \newcommand\ii{i}

57: \newcommand{\Int}{\mathbb{Z}}

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

59: \newcommand{\jj}{j}

60: \newcommand{\kk}{k}

61: \newcommand\lcml{\mathrm{lcm}_{\scriptscriptstyle\! L}}

62: \let\le=\leqslant

63: \newcommand{\mm}{m}

64: \newcommand\Neg[1]{#1^-}

65: \newcommand{\nn}{n}

66: \newcommand{\NN}{N}

67: \newcommand{\perm}{\pi}

68: \newcommand\Pos[1]{#1^+}

69: \newcommand\pp{p}

70: \newcommand{\qq}{q}

71: \newcommand{\redrr}{\curvearrowright}

72: \newcommand{\redrl}{\mathrel{\raisebox{5pt}{\rotatebox{180}{$\curvearrowleft$}}}}

73: \newcommand{\Ree}{\mathbb{R}}

74: \def\resp{{\it resp.}~}

75: \newcommand\rr{r}

76: \newcommand\s{\sigma}

77: \renewcommand\ss[1]{\sigma_{#1}^{\vrule height5pt width0pt}}%braidSigma 

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

79: \newcommand\ssss[1]{\sigma_{#1}^{\pm1}}%braidSigma 

80: \newcommand\tta{\mathtt{a}}

81: \newcommand\ttA{\mathtt{A}}

82: \newcommand\ttb{\mathtt{b}}

83: \newcommand\ttB{\mathtt{B}}

84: \newcommand\ttc{\mathtt{c}}

85: \newcommand\ttC{\mathtt{C}}

86: \newcommand\under{\mathord{\backslash}}

87: \newcommand{\uu}{\ww'}

88: \newcommand{\vs}{{\it vs.}~}

89: \newcommand{\vv}{\ww''}

90: \newcommand{\ww}{w}

91: \newcommand\xs{s}

92: \newcommand\xt{t}

93: \newcommand\xu{u}

94: \newcommand\xv{v}

95: \newcommand\xx{x}

96: \newcommand{\yy}{y}

97: \newcommand{\zz}{z}

98: 

99: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

100: \begin{document}

101: 

102: \hfill\raisebox{20pt}{$\scriptscriptstyle 2007-03$}

103: 

104: \author{Patrick DEHORNOY}

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

106: UMR 6139 CNRS,  Universit\'e de Caen BP 5186, 14032 Caen,

107: France}

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

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

110: 

111: \title{Efficient solutions to the braid isotopy problem}

112: 

113: \keywords{braid group, isotopy problem, word problem, greedy

114: normal form, word redressing, handle reduction, Dynnikov's coordinates}

115: 

116: \thanks{}

117: 

118: \subjclass{20F36, 94A60}

119: 

120: \begin{abstract}

121: We describe the most efficient solutions to the word problem

122: of Artin's braid group known so  far, \ie, in other

123: words, the most efficient solutions to the braid isotopy

124: problem, including the Dynnikov method, which could be especially

125: suitable for cryptographical applications. Most results appear in

126: literature; however, some results about the greedy normal form 

127: and the symmetric normal form and their connection with grid 

128: diagrams may have never been stated explicitly.  

129: \end{abstract}

130: 

131: \maketitle

132: 

133: Because they are both not too simple and not too complicated,

134: Artin's braid groups~$B_n$ have been and remain one of the most

135: natural and promising platform groups for non-commutative

136: group-based cryptography~\cite{AAG, KLC, Dgw}. More precisely,

137: braid groups are not too simple in that they lead to problems with

138: presumably difficult instances, typically the conjugacy problem and

139: the related conjugacy  and multiple conjugacy search problems, and

140: they are not too complicated in that there exist efficient solutions to

141: the word problem, a preliminary requirement when one aims at

142: practically computing in a group. It turns out that, since the

143: founding paper~\cite{Art} appeared in~1947, the word problem of

144: braid groups---which is also the braid isotopy problem---has

145: received a number of solutions: braid groups might even be the

146: groups for which the number of known solutions to the word

147: problem is currently the highest one. 

148: 

149: In this paper, we review some of these solutions, namely those that, at

150: the moment, appear as the most efficient ones for practical

151: implementation, and, therefore, the most promising ones for

152: cryptographical applications. What makes the subject specially

153: interesting is that these solutions relie on deep underlying

154: structures that explain their efficiency. Five solutions are

155: described, and they come in two families, namely those

156: based on a normal form, and those that use no normal form. In the

157: first family, we consider the so-called greedy normal form, both in its

158: non-symmetric and  symmetric versions. In the second family, we

159: consider the so-called word redressing method, which, like the greedy

160: normal forms, has a quadratic complexity, the handle

161: reduction method, whose complexity remains unknown

162: but which is very efficient in practice, and Dynnikov's coordinization

163: method, which relies on an entirely different, geometric approach,

164: and might turn to be very efficient. In this description, we

165: deliberately discard lots of alternative solutions which are

166: intrinsically exponential in complexity.

167: 

168: The paper is organized as follows. In Section~\ref{S:Normal}, after

169: setting the background, we describe the greedy normal form and the

170: symmetric normal form, without providing explicit rules to compute

171: them. All results in this section are standard. In Section~\ref{S:Grid},

172: we introduce grid diagrams and explain---and prove---how to use

173: such diagrams to compute the normal forms of

174: Section~\ref{S:Normal}. Though more or less equivalent to that

175: of~\cite[Chapter 9]{Eps}, this approach is less standard, and, to the

176: best of our knowledge, the results may have never been stated in the

177: form given here.  Finally, in Section~\ref{S:Direct}, we describe the

178: word redressing, handle reduction, and Dynnikov coordinates

179: methods. The results here already appeared in literature, but 

180: Dynnikov's approach, which appears in~\cite[Chapter~7]{Dgr},

181: has not yet become classical. Also, the formulae of Section~\ref{S:Dynnikov} 

182: have been optimized to make implementation easy.

183: 

184: \section{Solutions based on a normal form}

185: \label{S:Normal}

186: 

187: This section deals about solutions to the braid word problem that

188: consist in defining for each braid~$\xx$ a unique distinguished

189: representative called the normal form of~$\xx$. When this is done,

190: in order to compute with braids, it is in practice sufficient to work

191: with normal representatives. There exist excellent normal forms

192: for braids, namely those connected with the so-called {\it greedy 

193: normal form} based on Garside's theory~\cite{Gar}. Here we describe 

194: them, successively in a non-symmetric and a symmetric version.

195: 

196: \subsection{Braid groups}

197: 

198: We first recall a few basic definitions and general results about braid groups.

199: 

200: \subsubsection{Presentation}

201: 

202: Artin's braid groups are infinite non-commu\-tative

203: groups. They appear in several {\it a priori} unrelated

204: frameworks, and they admit many equivalent definitions. 

205: In our case, it will be convenient to introduce them by means of explicit

206: presentations. 

207: 

208: \begin{defi}

209: For $n \ge 2$, the braid group~$B_n$ is defined by the

210: presentation

211: \begin{equation}\label{E:Pres}

212:  \langle \ss1, ..., \ss{n-1} \, ; \,

213:  \ss i \ss j = \ss j \ss i 

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

215:  \ss i \ss j \ss i = \ss j \ss i \ss j 

216: \mbox{ for $\vert i - j\vert = 1$}

217:  \rangle.

218: \end{equation}

219: \end{defi}

220: 

221: An element of~$B_n$ will be called an {\it $n$-braid}. For each~$n$,

222: the identity mapping on~$\{\ss1, ..., \ss{n-1}\}$ induces an

223: embedding of~$B_n$ into~$B_{n+1}$, so that we can consider an

224: $\nn$-braid as a particular $(\nn+1)$-braid. Note that

225: $B_2$ is an infinite cyclic group, \ie, is isomorphic to the

226: group~$\Int$ of integers. For $\nn\ge3$, the group~$B_\nn$ is not

227: commutative: the center of~$B_n$ is the cyclic subgroup generated

228: by the element~$\DD\nn^2$, where $\DD\nn$ is introduced

229: in~\eqref{E:Delta} below.

230: 

231: When a group is specified using a presentation, each element of

232: the group is an equivalence class of words with respect to

233: the congruence generated by the relations of the presentation. In the sequel, a

234: word on the letters~$\ssss1$, ..., $\ssss{n-1}$ will be called an

235: {\it $n$-braid word}. So, every $n$-braid is an equivalence class of

236: $n$-braid words under the congruence~$\equiv$ generated

237: by the relations of~\eqref{E:Pres}. If the braid~$\xx$ is the

238: equivalence class of the word~$\ww$, we say that $\ww$ is a

239: representative of~$\xx$, and we write $\xx= \cl\ww$. 

240: 

241: \subsubsection{The word problem}

242: 

243: Using $\e$ for the empty word, the {\it word

244: problem} of~\eqref{E:Pres} is the algorithmic problem:

245: \begin{quote}

246: Given one braid word~$\ww$, does $\ww\equiv\e$ hold, \ie, does

247: $\ww$ represent the unit braid~$1$?

248: \end{quote}

249: This is the problem we investigate in the sequel.

250: Because $B_\nn$ is a group, the above one parameter problem is

251: equivalent to the two parameter problem:

252: \begin{quote}

253: Given two braid words~$\ww, \ww'$, does $\ww\equiv\ww'$ hold,

254: \ie, do $\ww$ and $\ww'$ represent the same braid?

255: \end{quote}

256: Indeed, $\ww\equiv\ww'$ is equivalent to $\ww\inv\ww'\equiv\e$,

257: where $\ww\inv$ is the word obtained from~$\ww$ by reversing

258: the order of the letters and exchanging~$\ss\ii$ and~$\sss\ii$

259: everywhere.

260: 

261: \subsubsection{Geometric interpretation}

262: 

263: The elements of~$B_n$ can be interpreted as geometric $n$~strand

264: braids~\cite{Bir, KaT, Dgr}. To this end, one

265: associates with every braid word the plane diagram obtained by

266: concatenating the elementary diagrams of Figure~\ref{F:Braid}

267: corresponding to the successive letters. 

268: 

269: \begin{figure}[htb]

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

271: \begin{picture}(110,31)(0,-1)

272: %\Grid

273: \put(5,2){\includegraphics{Braid.eps}}

274: \put(1,1){$1$}

275: \put(1,11){$\ii$}

276: \put(-1.3,17){$\ii\!+\!1$}

277: \put(1,27){$\nn$}

278: \put(7,-2){$\ss\ii$}

279: \put(16,-2){$\sss\ii$}

280: \put(8,6){$\vdots$}

281: \put(18,6){$\vdots$}

282: \put(8,21){$\vdots$}

283: \put(18,21){$\vdots$}

284: \put(31,3){$\ss1$}

285: \put(35,3){$\sss2$}

286: \put(41,3){$\ss1$}

287: \put(46,3){$\ss2$}

288: \put(51,3){$\ss1$}

289: \put(56,3){$\ss3$}

290: \put(59.5,3){$\sss1$}

291: \put(65,3){$\sss2$}

292: \put(70.5,3){$\sss1$}

293: \put(76,3){$\sss2$}

294: \put(81.5,3){$\sss1$}

295: \put(87.5,3){$\ss2$}

296: \put(91.5,3){$\ss2$}

297: \put(95.5,3){$\sss3$}

298: \put(101,3){$\sss2$}

299: \end{picture}

300: \caption{\smaller\sf The $\nn$~strand braid diagrams

301: associated with~$\ss i$ and~$\sss i$, and the

302: $4$~strand braid diagram associated with the $4$-\nobreak braid

303: word

304: $\ss1\sss2\ss1\ss2\ss1\ss3\sss1\sss2\sss1\sss2\sss1\ss2\ss2\sss3\sss2$---also

305: denoted $\mathtt{aBabacABABAbbCB}$ in the sequel: one

306: concatenates the successive $4$~strand diagrams

307: corresponding to the successive letters of the word.}

308: \label{F:Braid}

309: \end{figure}

310: 

311: A braid diagram can be seen as a plane projection of a

312: three-dimensional figure consisting on $n$~disjoint

313: curves connecting the points $(1, 0, 0),

314: ..., (n, 0, 0)$ to the points $(1, 0, 1), ..., (n, 0, 1)$ in~$\Ree^3$. Then

315: the relations of~\eqref{E:Pres} correspond to ambient isotopy, \ie,

316: to continuously moving the curves without moving their

317: ends and without allowing them to intersect. It is easy to check that

318: each relation in~\eqref{E:Pres} corresponds to such an isotopy; the

319: converse implication, \ie, the fact that the projections

320: of isotopic 3D figures can always be encoded in words connected

321: by~\eqref{E:Pres} was proved by E.\,Artin in~\cite{Art}. Thus the

322: braid word problem for the presentation~\eqref{E:Pres} is also 

323: the {\it braid isotopy problem}---thus similar to the (much more difficult) 

324: knot isotopy problem.

325: 

326: \subsubsection{Positive braids}

327: 

328: A braid word is said to be {\it positive} if it contains no

329: letter~$\sss\ii$. A braid is said to be {\it positive} if it can be

330: represented by at least one positive word. Positive $n$-braids form a

331: submonoid denoted~$\BB\nn$ of the group~$B_n$. Garside's theory~\cite{Gar} 

332: implies that $\BB\nn$ admits, as a monoid, the

333: presentation~\eqref{E:Pres} and that $B_n$ is a group of fractions

334: of~$\BB\nn$, \ie, every braid in~$B_n$ can be expressed

335: as~$\yy\inv\xx$ with $\xx,\yy$ in~$\BB\nn$.

336: 

337: For $\xx, \yy$ in~$\BB\nn$, we say that $\xx$ is a {\it

338: left divisor} of~$\yy$, or, equivalently,  that $\yy$ is a {\it

339: right multiple} of~$\xx$, denoted $\xx\divel\yy$, if $\yy = \xx

340: \zz$ holds for some~$\zz$ in~$\BB\nn$. Right divisors and left

341: multiples are defined symmetrically. With respect to left divisibility (and 

342: to right divisibility as well),

343: $\BB\nn$ has the structure of a lattice: any two positive

344: $\nn$-braids~$\xx, \yy$ admit a greatest common left

345: divisor~$\gcdl(\xx, \yy)$, and a least common right multiple. 

346: 

347: \subsubsection{Permutation of a braid}

348: 

349: The geometric interpretation makes it clear that mapping~$\ss i$

350: to the transposition that exchanges~$i$ and~$i+1$

351: induces a surjective homomorphism of the braid group~$B_n$ onto

352: the symmetric group~$S_n$. Under this homomorphism, here

353: denoted~$\perm$, a braid~$\xx$ is mapped to the permutation~$f$

354: of~$\{1, ..., n\}$ such that the strand that finishes at position~$i$ in

355: any braid diagram representing~$\xx$ begins at position~$f(i)$. 

356: 

357: \subsubsection{Simple braids}

358: 

359: A special r\^ole is played by the positive $\nn$-braid~$\DD\nn$

360: inductively defined by

361: \begin{equation} 

362: \label{E:Delta}

363: \DD1 = 1, \qquad

364: \DD\nn = \ss1 \ss2 \dots \ss{\nn-1} \, \DD{\nn-1}.

365: \end{equation}

366: In~$\BB\nn$, the left and the right divisors of~$\DD\nn$ coincide,

367: they include each of~$\ss1$, ..., $\ss{\nn-1}$, and they make

368: a finite sublattice of~$\BB\nn$ with $\nn!$~elements. These divisors

369: of~$\DD\nn$ are called {\it simple} braids. Geometrically, simple

370: braids are those positive braids that can be represented by a braid

371: diagram in which any two strands cross at most once. Moreover, the

372: restriction of the projection~$\perm$ to simple braids is a bijection:

373: for each permutation~$\ff$ in~$S_\nn$, there exists exactly

374: one simple braid~$\xs$ satisfying $\perm(\xs)=\ff$. This simple

375: braid will be denoted by~$\br\ff$.

376: 

377: \subsection{The greedy normal form}

378: \label{S:Greedy}

379: 

380: The seminal results of F.A.\,Garside~\cite{Gar} subsequently

381: developed in~\cite{Thu, Adj, ElM} imply that braid groups can be

382: equipped with a remarkable normal form, the so-called

383: greedy normal form. The latter  is excellent both in theory and in

384: practice as it provides a  bi-automatic structure, and it is easily

385: computable. 

386: 

387: \subsubsection{Description}

388: 

389: The greedy normal form exists in several variants, in particular left

390: and symmetric right versions. Here we shall consider the left versions

391: only. Again, there exist two different versions. Both consist in

392: expressing an arbitrary braid as a quotient of two positive braids,

393: \ie, as a fraction. In the version considered in this section, all

394: denominators have some special form, namely they are powers of

395: the element~$\DD\nn$. By contrast, in the version

396: considered in Section~\ref{S:Symmetric} below, the numerator and

397: the denominator of fractions play symmetric r\^oles.

398: 

399: \begin{defi}

400: \label{D:Normal}

401: $(i)$ A sequence of simple braids $(\xs_1, ..., \xs_\pp)$ is said to be

402: {\it normal} if, for each~$\kk<\pp$, every~$\ss\ii$ that is a left

403: divisor of~$\xs_{\kk+1}$ is a right divisor of~$\xs_\kk$.

404: 

405: $(ii)$ A sequence of permutations $(\ff_1, ..., \ff_\pp)$ is said to be

406: {\it normal} if, for each~$\kk<\pp$,  every recoil

407: of~$\ff_{\kk+1}$ is a descent of~$\ff_\kk$, \ie, if

408: $\ff_{\kk+1}\inv(i) > \ff_{\kk+1}\inv(i+1)$ implies $\ff_\kk(i) >

409: \ff_\kk(i+1)$.

410: \end{defi}

411: 

412: The connection between~$(i)$ and~$(ii)$ in

413: Definition~\ref{D:Normal} is that, if $\xs$ is a simple $\nn$-braid

414: and $\ff$ is the associated permutation, then $\ss\ii$ is a left (\resp

415: right) divisor of~$\xs$ if and only if we

416: have $\ff\inv(\ii)>\ff\inv(\ii+1)$ (\resp $\ff(\ii)> \ff(\ii+1)$).

417: So a sequence of simple braids $(\xs_1, ..., \xs_\pp)$ is normal if and

418: only if the associated sequence of permutations $(\perm(\xs_1),

419: ...,\perm(\xs_\pp))$ is normal.

420: 

421: We denote by~$\omega_\nn$ the flip permutation of~$\{1, ...,

422: \nn\}$ defined by $\omega_\nn(\ii) = \nn-\ii+1$.

423: 

424: \begin{theo}

425: \cite[Chapter~9]{Eps}

426: \label{T:Normal}

427: $(i)$ Every braid~$\zz$ in~$B_n$ admits a unique decomposition of

428: the form $\DD\nn^\mm \xs_1 ... \xs_\pp$ with $\mm$ in~$\Int$

429: and $(\xs_1, ..., \xs_\pp)$ a normal sequence of simple braids

430: satisfying $\xs_1 \not=\DD\nn$ and $\xs_\pp\not=1$.

431: 

432: $(ii)$ Every braid~$\zz$ in~$B_n$ admits a unique decomposition of

433: the form $\DD\nn^\mm \br{\ff_1} ... \br{\ff_\pp}$ with $\mm$

434: in~$\Int$ and $(\ff_1, ..., \ff_\pp)$ a normal sequence of

435: permutations satisfying $\ff_1 \not=\omega_\nn$ and

436: $\ff_\pp\not=\id$.

437: \end{theo}

438: 

439: In the situation of Theorem~\ref{T:Normal}$(i)$, the sequence

440: $(\mm; \xs_1, ..., \xs_\pp)$ is called the {\it greedy normal form}

441: of~$\zz$---or the $\nn$-greedy normal form of~$\zz$ if we wish to insist

442: on the braid index~$\nn$. As simple braids are in

443: one-to-one correspondence with permutations, and by the remark

444: above, the braid form and the permutation form of the greedy normal

445: form are equivalent. So there is no problem

446: in also calling the sequence

447: $(\mm; \ff_1, ..., \ff_\pp)$ of Theorem~\ref{T:Normal}$(ii)$ the

448: greedy normal form of~$\zz$. 

449: 

450: Clearly, $(0; \emptyset)$ is the greedy normal form of~$1$, and the

451: uniqueness of the greedy normal form implies the following solution to the

452: braid isotopy problem---but a solution that remains uneffective as long 

453: as we give no method for computing from an arbitrary braid word~$\ww$

454: the greedy normal form of~$\cl\ww$, \ie,

455: until Section~\ref{S:Grid} below:

456: 

457: \begin{coro}

458: \label{C:Greedy}

459: A braid word~$\ww$ represents~$1$ in the braid group if and

460: only if the greedy normal form of~$\cl\ww$ is~$(0; \emptyset)$.

461: Two braid words~$\ww, \ww'$ represent the same braid

462: in~$B_\nn$ if and only if the greedy normal forms of the

463: braids~$\cl\ww$ and~$\cl{\ww'}$ coincide.

464: \end{coro}

465: 

466: \begin{exam}

467: \label{X:NF}

468: In order to obtain shorter notation, we shall in the sequel use $\tta,

469: \ttb, \ttc...$ for~$\ss1, \ss2, \ss3...$, and, symmetrically,

470: $\ttA, \ttB...$ for $\sss1, \sss2$... (as in the caption of

471: Figure~\ref{F:Braid}). Then, a typical greedy normal form for a

472: $4$-braid is the sequence

473: 

474: $(-2; \mathtt{ac, abcb, bcba, a})$,

475: 

476: \noindent\ie, equivalently, using $(\ff(1), ..., \ff(\nn))$ to

477: specify a permutation~$\ff$ of~$\{1, ..., \nn\}$,

478: 

479: $(-2; (2,1,4,3), (2,4,3,1), (4,1,3,2), (2,1,3,4))$,

480: 

481: \noindent consisting of an integer and four simple $4$-braids, or,

482: equivalently, four permutations of~$\{1, ..., 4\}$: for instance, 

483: $(2,1,4,3)$ is the permutation associated

484: with~$\mathtt{ac}$, \ie, with~$\ss1\ss3$. To check that we have a greedy 

485: normal form, we observe for instance that the descents of  $(2,1,4,3)$ 

486: are $1$ and~$3$, while the recoils of $(2,4,3,1)$, \ie, the descents of~$(4,1,2,3)$, are

487: $1$ and~$3$ as well, so the normality condition is satisfied between

488: $(2,1,4,3)$ and~$(2,4,3,1)$. The other verifications are similar.

489: So, the above sequences are two versions of the greedy normal

490: form of the $4$-braid represented by $\Delta_4^{-2}.\mathtt{ac.abcb.bcba.a}$, \ie,

491: by

492: 

493: $\ww = \mathtt{ABACBA.ABACBA.ac.abcb.bcba.a}$.

494: 

495: \noindent As the above normal form is not $(0, \emptyset)$, we

496: deduce from Corollary~\ref{C:Greedy} that $\ww$ does not represent~$1$ in~$B_4$.

497: \end{exam}

498: 

499: \subsubsection{Explanation}

500: 

501: The existence and uniqueness of the greedy normal form follows

502: from two results: 

503: 

504: $(i)$ For every braid~$\zz$ in~$B_\nn$, there exist $\mm\in\Int$ 

505: and $\xx\in\BB\nn$ satisfying $\DD\nn^\mm\xx=\zz$, the

506: decomposition being unique provided $\DD\nn\not\divel\xx$ is

507: required;

508: 

509: $(ii)$ For every positive braid~$\xx$, there exists a unique normal

510: sequence $(\xs_1, ..., \xs_\pp)$ of simple $\nn$-braids with

511: $\xs_\pp\not=1$ satisfying $\xs_1 ... \xs_\pp = \xx$.

512: 

513: The proof of~$(i)$ is an easy induction on the length of a braid

514: word representing~$\zz$ once one knows, for each~$\ii$, the

515: relations $\ss\ii\divel\DD\nn$ and $\DD\nn

516: \ss\ii=\ss{\nn-\ii}\DD\nn$, which imply that, by multiplying an

517: $\nn$-braid by a sufficient large power of~$\DD\nn$, one can

518: always obtain a positive braid.

519: 

520: As for~$(ii)$, the existence of left gcd's in the monoid~$\BB\nn$

521: implies that each positive $\nn$-braid~$\xx$ can be expressed as

522: $\xx = \xs_1\xx'$ with $\xs_1 = \gcdl(\xx, \DD\nn)$, and

523: $\xx\not=1$ implies $\xs_1\not=1$. By iterating the process,

524: thus writing $\xx' = \xs_2\xx''$, {\it etc.}, one eventually obtains a

525: decomposition $\xx= \xs_1 ... \xs_\pp$. By construction, the

526: sequence $(\xs_1, ..., \xs_\pp)$ consists of divisors of~$\DD\nn$,

527: \ie, of simple $\nn$-braids, and, for each~$\kk<\pp$, one has

528: $\xs_\kk = \gcdl(\xs_\kk

529: \xs_{\kk+1} ... \xs_\pp, \DD\nn)$, hence, {\it a fortiori}, $\xs_\kk =

530: \gcdl(\xs_\kk \xs_{\kk+1}, \DD\nn)$. The point is that the

531: latter relations are equivalent to $(\xs_\kk, \xs_{\kk+1})$

532: being normal for each~$\kk$, and, therefore, to $(\xs_1, ...,

533: \xs_\pp)$ being normal. The uniqueness comes from the fact that, if

534: $(\xs_1, ..., \xs_\pp)$ is a normal sequence, then, necessarily, one

535: has $\xs_1 = \gcdl(\xs_1...\xs_\pp, \DD\nn)$.

536: 

537: \subsubsection{Discussion}

538: 

539: What is missing in the above description of the greedy

540: normal form is an algorithm for computing the (unique)

541: normal form of a braid~$\zz$ starting from an arbitrary

542: representative of~$\zz$. Clearly, the existence of such an algorithm is

543: a necessary condition for using the normal form in

544: practice. Such an algorithm will be provided in

545: Section~\ref{S:Grid} below, and discussing the practical

546: implementation of the greedy normal form will be possible only then.

547: 

548: Actually, the point is not necessarily to find the normal form

549: equivalent to an arbitrary word, but rather to find the normal form of

550: the product or the quotient of two normal forms. Indeed, whenever

551: one chooses to work with normal forms, one may forget about

552: non-normal words provided one is able to perform the basic

553: operations inside the family of normal forms. Of course, as the

554: generator~$\ss\ii$ is itself a braid, with normal form~$(0; \ss\ii)$,

555: an algorithm computing the product of two normal forms will in

556: particular determine the product of a normal form by~$\ss\ii$ and,

557: therefore, inductively determine the normal form of any product

558: of~$\ssss\ii$'s, but the general philosophy is not exactly that of a

559: normalizing algorithm.

560: 

561: Note that, while the permutation variant of the greedy normal form is 

562: non-ambiguously defined, the

563: braid word variant is not: for instance, the first braid factor in 

564: Example~\ref{X:NF}, namely~$\mathtt{ac}$, is uniquely

565: defined as a simple braid, but it can be represented by two different

566: positive braid words, namely $\mathtt{ac}$ and~$\mathtt{ca}$. So

567: the braid word form becomes unique only when a distinguished

568: word representative has been chosen for every simple braid. This

569: explains why the permutation form is often more convenient.

570: 

571: \subsection{The symmetric normal form}

572: \label{S:Symmetric}

573: 

574: In the greedy normal form where the denominator is always a power of~$\DD\nn$. 

575: The symmetric normal form is a variant in which the numerator and the 

576: denominator play symmetric r\^oles.

577: 

578: \subsubsection{Description}

579: 

580: The symmetric normal form appeals to the same notion of a normal

581: sequence of simple braids as the greedy normal form.

582: 

583: \begin{theo}

584: \cite[Chapter~9]{Eps}

585: \label{T:Symmetric}

586: $(i)$ Every braid~$\zz$ admits a unique decomposition as

587: $\xt_\qq\inv\, ... \,\xt_1\inv \,\xs_1\, ... \,\xs_\pp$ with $(\xs_1, ...,

588: \xs_\pp)$, $(\xt_1, ..., \xt_\qq)$ two normal sequences of simple braids

589: satisfying $\xs_\pp\not=1$, $\xt_\qq\not=1$, and

590: $\gcdl(\xs_1, \xt_1)=1$.

591: 

592: $(ii)$ Every braid~$\zz$ admits a unique decomposition

593: as $\br{\gg_\qq}{}\inv \,...\, \br{\gg_1}{}\inv\,\br{\ff_1}\, ...

594: \,\br{\ff_\pp}$ with $(\ff_1, ..., \ff_\pp)$, $(\gg_1, ..., \gg_\qq)$ two

595: normal sequences of permutations satisfying

596: $\ff_\pp, \gg_\qq \not= \id$ and

597: such that $\ff_1\inv(\ii)>\ff_1\inv(\ii+1)$ implies

598: $\gg_1\inv(\ii)\le\gg_1\inv(\ii+1)$.

599: \end{theo}

600: 

601: In the situation of Theorem~\ref{T:Symmetric}$(i)$, the

602: double sequence $(\xt_1, ..., \xt_\qq; \xs_1, ..., \xs_\pp)$ is called

603: the {\it symmetric normal form} of~$\zz$. As for the greedy normal

604: form, both versions of the symmetric normal forms are equivalent,

605: and the double sequence of permutations $(\gg_1, ..., \gg_\qq; \ff_1,

606: ..., \ff_\pp)$ of Theorem~\ref{T:Symmetric}$(ii)$ is also called the

607: symmetric normal form of~$\zz$. As $(\emptyset;

608: \emptyset)$ is the symmetric normal form of~$1$, the uniqueness 

609: of the symmetric normal form provides the following

610: solution to the braid isotopy problem:

611: 

612: \begin{coro}

613: \label{C:Symmetric}

614: A braid word~$\ww$ represents~$1$ in the braid group if and

615: only if the symmetric normal form of~$\cl\ww$ is~$(\emptyset;

616: \emptyset)$. Two braid words~$\ww, \ww'$ represent the same braid

617: in~$B_\nn$ if and only if the symmetric normal forms of~$\cl\ww$

618: and~$\cl{\ww'}$ coincide.

619: \end{coro}

620: 

621: \begin{exam}

622: \label{X:Symmetric}

623: A typical symmetric normal form is

624: 

625: $(\mathtt{ab, bacb; bcba, a})$,

626: 

627: \noindent\ie, equivalently, 

628: 

629: $((2,3,1,4), (3,4,1,2); (4,1,3,2), (2,1,3,4))$,

630: 

631: \noindent For instance, the normality condition between the first factors of the

632: two sequences holds as $1$ is the only recoil of~$(2,3,1,4)$, while

633: $2$ and~$3$ are those of~$(4,1,3,2)$. In other words, the simple

634: braids~$\mathtt{ab}$ and~$\mathtt{bcba}$ admit no nontrivial

635: common left divisor. Thus the above expressions specify the

636: symmetric normal form of the braid represented by

637: 

638: $\ww = \mathtt{BCAB.BA.bcba.a}$,

639: 

640: \noindent which we shall see below coincides with the braid of

641: Example~\ref{X:NF}---and of Figure~\ref{F:Braid}. As the above

642: normal form is not $(\emptyset, \emptyset)$, we deduce that $\ww$

643:  does not represent~$1$ in~$B_4$.

644: \end{exam}

645: 

646: \subsubsection{Explanation}

647: 

648: As $B_\nn$ is a group of fractions for the monoid~$\BB\nn$, every

649: element of~$B_\nn$ can be expressed as a fraction~$\yy\inv\xx$

650: with $\xx, \yy$ in~$\BB\nn$, and the decomposition is unique if, in

651: addition, $\gcdl(\xx,\yy)$ is required. The symmetric normal form

652: is obtained by taking the greedy normal forms of the positive

653: braids~$\xx, \yy$ so obtained, with the only difference that the 

654: $\DD\nn$ factors are not separated. The specific properties of 

655: normal sequences imply that $\gcdl(\xx,\yy)=1$ is

656: equivalent to $\gcdl(\xs_1, \xt_1)=1$, where $\xs_1$ and $\xt_1$

657: respectively are the first factors in the normal forms of~$\xx$

658: and~$\yy$.

659: 

660: \subsubsection{Discussion}

661: 

662: As in Section~\ref{S:Greedy}, our description will be complete only

663: when we give algorithms for computing the symmetric normal form

664: of a product or of a quotient. This will be done in

665: Section~\ref{S:Grid}.

666: 

667: \section{Grid properties of the normal form}

668: \label{S:Grid}

669: 

670: The interest of the greedy and symmetric normal forms of braids

671: lies in the existence of simple computing rules for determining the

672: normal form of a product or of a quotient. Here we shall solve the

673: following two problems, which then easily leads to

674: complete algorithms for all problems connected with the

675: normal forms:

676: \begin{quote}

677: Starting with the greedy normal forms of two positive

678: braids~$\xx, \yy$, find the greedy normal form of the

679: product~$\yy\xx$ and, assuming that $\yy$ is a right divisor

680: of~$\xx$, the greedy normal form of the quotient~$\xx\yy\inv$.

681: \end{quote}

682: The solutions we shall describe involve grid diagrams that visualize

683: the properties of normal sequences. To this end, it will be convenient

684: to associate with every braid~$\xx$ an arrow labelled~$\xx$, so

685: that a  relation of the form $\xx\yy=\zz$ corresponds to a

686: commutative diagram for the associated arrows. We

687: shall indicate that a sequence $(\xs_1, ...,

688: \xs_\pp)$ is normal by drawing an  arc connecting the final end of

689: each arrow with the initial end of the next one, on the shape of

690: \begin{center}

691: \begin{picture}(75,6)

692: \put(0,0){\includegraphics{Sequence.eps}}

693: \put(6,2){$\xs_1$}

694: \put(21,2){$\xs_2$}

695: \put(36,2){$\xs_3$}

696: \put(50,0.5){$\ldots$}

697: \put(66,2){$\xs_\pp$}

698: \end{picture}.

699: \end{center}

700: 

701: \subsection{Prerequisites}

702: 

703: The only properties of simple braids needed in the forthcoming

704: proofs are summarized in the following lemma, which is standard:

705: 

706: \begin{lemm}

707: \cite{Cha}

708: \label{L:Head}

709: For~$\xx$ in~$\BB\nn$, let $\HH(\xx):=\gcdl(\xx, \DD\nn)$. Then,

710: for all~$\xx, \yy$, we have:

711: \begin{gather}

712: \label{E:Head2}

713: \HH(\xx) \divel \xx,

714: \text{\quad and $\HH(\xx)=\xx$ if $\xx$ is simple},\\

715: \label{E:Head3}

716: \xx\divel\yy  \text{ implies } 

717: \HH(\xx) \divel \HH(\yy),\\

718: \label{E:Head4}

719: \HH(\xx\yy) =\HH(\xx\HH(\yy)).

720: \end{gather}

721: \end{lemm}

722: 

723: Then the definition of normal sequences immediately rewrites in

724: terms of the function~$\HH$ as follows---and this is the

725: technical form we shall use in the sequel:

726: 

727: \begin{lemm}

728: A sequence of simple braid~$(\xs_1, ..., \xs_\pp)$ is normal if and

729: only if, for each~$\kk<\pp$, we have $\xs_\kk = \HH(\xs_\kk

730: \xs_{\kk+1})$.

731: \end{lemm}

732: 

733: In the sequel, we mostly deal with positive braids. When we say that

734: $(\xs_1, ..., \xs_\pp)$ is the normal form of a positive $\nn$-braid~$\xx$, we

735: mean that $(\xs_1, ..., \xs_\pp)$ is normal and that $\xx = \xs_1 ... \xs_\pp$.

736: Equivalently, this means that the greedy normal form of~$\xx$ is

737: $(\mm; \xs_{\mm+1}, ..., \xs_\pp)$, where $\mm$ is the number of initial~$\kk$'s

738: such that $\xs_\kk$ equals~$\DD\nn$.

739: 

740: \subsection{Grid properties for the quotient}

741: \label{S:GridComplement}

742: 

743: We start with the problem of finding the normal form

744: of~$\xx\yy\inv$ from those of~$\xx$ and~$\yy$, where $\xx$

745: and~$\yy$ belong to~$\BB\nn$. In principle, the computation

746: makes sense only if $\yy$ happens to be a right divisor of~$\xx$.

747: Actually, in any case, the positive braids~$\xx, \yy$ admit a left lcm

748: in~$\BB\nn$, and what we shall do is to compute the normal form of

749: this left lcm, denoted~$\lcml(\xx,\yy)$, and of the associated left 

750: complements defined as follows:

751: 

752: \begin{defi}

753: For $\xx,\yy$ in~$\BB\nn$, the unique~$\xx'$ in~$\BB\nn$

754: satisfying $\lcml(\xx,\yy)=\xx'\yy$ is called the {\it left

755: complement} of~$\xx$ in~$\yy$, and denoted by~$\xx/\yy$.

756: \end{defi}

757: 

758: If $\yy$ happens to be a right divisor of~$\yy$, \ie, if

759: $\yy/\xx = 1$ holds, then we have $\lcml(\xx,\yy) = \yy$,

760: and $\xx/\yy = \xx\yy\inv$. Thus an algorithm

761: computing the left complement is in particular an algorithm

762: computing the right quotient when it exists.

763: 

764: A standard observation is that simple braids are closed under left

765: lcm and left complement. Indeed, assume that $\xs,

766: \xt$ are simple, and let $\xu = \lcml(\xs, \xt) = \xs' \xt = \xt' \xs$,

767: \ie, $\xs'= \xs/\xt$ and $\xt' = \xt/\xs$. Then, $\xs$ and $\xt$ are

768: right divisors of~$\DD\nn$, hence $\xu$ is also a right divisor

769: of~$\DD\nn$, and it is therefore a simple braid. Then, $\xs'$

770: and~$\xt'$ are simple as well, as every left divisor of a

771: simple braid is a simple braid. Moreover, it is straightforward to

772: check that $\xs'\xt$ being the left lcm of~$\xs$ and~$\xt$ is

773: equivalent to $\xs'$ and $\xt'$ admitting no common left divisor

774: except~$1$, \ie, to $\gcdl(\xs', \xt')=1$.

775: 

776: \begin{defi}

777: A commutative diagram consisting of four simple braids~$\xs, \xt,

778: \xs', \xt'$ satisfying $\xs'\xt=\xt'\xs$ and $\gcdl(\xs', \xt')=1$ is

779: called a {\it $C$-tile} (like ``complement tile''), and it is represented

780: as in Figure~\ref{F:CTile}.

781: \end{defi}

782: 

783: \begin{figure}[htb]

784: \begin{picture}(23,19)

785: %\Grid

786: \put(0,.5){\includegraphics{CTile.eps}}

787: \put(-2,8){$\xt'$}

788: \put(22,8){$\xt$}

789: \put(10,-2){$\xs$}

790: \put(10,17){$\xs'$}

791: \end{picture}

792: \caption{\smaller\sf A $C$-tile: a commutative diagram involving

793: four simple braids such that the two initial ones have no nontrivial

794: common left divisor---here indicated by a perpendicularity sign.}

795: \label{F:CTile}

796: \end{figure}

797: 

798: We establish normality conditions for diagrams involving

799: the above $C$-tiles.

800: 

801: \begin{lemm}

802: \label{L:Complement1}

803: (Figure~\ref{F:Complement1}) 

804: Assume that $\xs_1, \xs_2, \xs'_1, \xs'_2, \xt_0, \xt_1,

805: \xt_2$ are simple braids satisfying $\xt_2\xs_1=\xs'_1\xt_1$, 

806: $\xt_1\xs_2=\xs'_2\xt_0$, $\gcdl(\xs'_2,\xt_1)=1$, and that

807: $(\xs_1,\xs_2)$ is normal. Then $(\xs'_1, \xs'_2)$ is normal as well.

808: \end{lemm}

809: 

810: \begin{figure}[htb]

811: \begin{picture}(41,24)

812: %\Grid

813: \put(0,1){\includegraphics{Complement1.eps}}

814: \put(-3,11){$\xt_2$}

815: \put(22,11){$\xt_1$}

816: \put(42,11){$\xt_0$}

817: \put(9,1){$\xs_1$}

818: \put(29,1){$\xs_2$}

819: \put(9,21){$\xs'_1$}

820: \put(29,21){$\xs'_2$}

821: \end{picture}

822: \caption{\smaller\sf A grid property involving $C$-tiles:

823: assuming that the right rectangle is a $C$-tile and the bottom

824: sequence $(\xs_1, \xs_2)$ is normal, the top sequence $(\xs'_1,

825: \xs'_2)$ is normal as well.}

826: \label{F:Complement1}

827: \end{figure}

828: 

829: \begin{proof}

830: With the notation of Figure~\ref{F:Complement1}, we compute:

831: \begin{align*}

832: \HH(\xs'_1\xs'_2)

833: &\divel \HH(\xs'_1\xs'_2\xt_0)

834: &&\text{by~\eqref{E:Head3}}\\

835: &= \HH(\xt_2\xs_1\xs_2)

836: &&\text{by commutativity}\\

837: &= \HH(\xt_2\HH(\xs_1\xs_2))

838: &&\text{by~\eqref{E:Head4}}\\

839: &= \HH(\xt_2 \xs_1)

840: &&\text{by the hypothesis that $(\xs_1,\xs_2)$ is normal}\\

841: &= \HH(\xs'_1\xt_1)

842: &&\text{by commutativity}\\

843: &\divel \xs'_1\xt_1

844: &&\text{by~\eqref{E:Head2}}.

845: \end{align*}

846: On the other hand, by~\eqref{E:Head2} again, we have

847: $\HH(\xs'_1\xs'_2) \divel \xs'_1\xs'_2$, hence

848: $$\HH(\xs'_1\xs'_2) \divel \gcdl(\xs'_1\xs'_2, \xs'_1\xt_1) =

849: \xs'_1 \cdot \gcdl(\xs'_2, \xt_1) = \xs'_1,$$

850: using the hypothesis $\gcdl(\xs'_2, \xt_1)=1$. It follows that 

851: $(\xs'_1, \xs'_2)$ is normal.

852: \end{proof}

853: 

854: Similarly, we have the following result involving the diagonals of

855: $C$-tiles, \ie, the left lcm's of the corresponding simple braids.

856: 

857: \begin{lemm}

858: \label{L:Complement2}

859: (Figure~\ref{F:Complement2})

860: Assume that $(\xs_1, \xs_2)$ and $(\xt_1, \xt_2)$ are normal

861: sequences of simple braids. Let $\xu_1, \xu_2$ be as in

862: Figure~\ref{F:Complement2}. Then $(\xu_1, \xu_2)$, $(\xu_1,

863: \xs'_2)$, and $(\xu_1, \xt'_2)$ are normal as well. 

864: \end{lemm}

865: 

866: \begin{figure}[htb]

867: \begin{picture}(41,35)

868: %\Grid

869: \put(0,1){\includegraphics{Complement2.eps}}

870: \put(-3,11){$\xt''_2$}

871: \put(22,11){$\xt'_2$}

872: \put(42,11){$\xt_2$}

873: \put(-3,26){$\xt''_1$}

874: \put(22,26){$\xt'_1$}

875: \put(42,26){$\xt_1$}

876: \put(9,1){$\xs_1$}

877: \put(29,1){$\xs_2$}

878: \put(9,21){$\xs'_1$}

879: \put(29,21){$\xs'_2$}

880: \put(9,36){$\xs''_1$}

881: \put(29,36){$\xs''_2$}

882: \put(9,29){$\xu_1$}

883: \put(29,14){$\xu_2$}

884: \end{picture}

885: \caption{\smaller\sf Another grid property involving $C$-tiles:

886: assuming that the bottom right rectangle is a $C$-tile and the

887: bottom and right sequences $(\xs_1, \xs_2)$ and $(\xt_1, \xt_2)$

888: are normal, the diagonal sequence $(\xu_1, \xu_2)$ is normal

889: as well, and so are the diagonal-then-horizontal sequence

890: $(\xu_1,\xs'_2)$ and the diagonal-then-vertical sequence

891: $(\xu_1,\xt'_2)$.}

892: \label{F:Complement2}

893: \end{figure}

894: 

895: \begin{proof}

896: With the notation of Figure~\ref{F:Complement2}, we compute

897: \begin{align*}

898: \HH(\xu_1\xu_2)

899: &= \HH(\xt''_2\xt''_1\xs_1\xs_2)

900: &&\text{by commutativity}\\

901: &= \HH(\xt''_2\xt''_1\HH(\xs_1\xs_2))

902: &&\text{by~\eqref{E:Head4}}\\

903: &= \HH(\xt''_2\xt''_1\xs_1)

904: &&\text{by the hypothesis that $(\xs_1, \xs_2)$ is normal}\\

905: &= \HH(\xu_1\xt'_2)

906: &&\text{by commutativity}\\

907: &\divel \xu_1\xt'_2

908: &&\text{by~\eqref{E:Head2}}.

909: \end{align*}

910: Similarly, we obtain $\HH(\xu_1\xu_2)\divel \xu_1\xs'_2$,

911: and we deduce

912: $$\HH(\xu_1\xu_2)\divel \gcdl(\xu_1\xt'_2,

913: \xu_1\xs'_2) = \xu_1 \cdot \gcdl(\xs'_2, \xt'_2) = \xu_1$$ from

914: the hypothesis $\gcdl(\xs'_2, \xt'_2)=1$. So 

915: $(\xu_1,\xu_2)$ is normal. As $\xs'_2$ is a left divisor

916: of~$\xu_2$, the normality of~$(\xu_1, \xu_2)$ implies

917: that of~$(\xu_1, \xs'_2)$, and, similarly, that of~$(\xu_2,

918: \xt'_2)$.

919: \end{proof}

920: 

921: We are ready to establish:

922: 

923: \begin{prop}

924: \label{P:Complement}

925: $(i)$ Assume that $(\xs_1, ..., \xs_\pp)$ and $(\xt_1, ..., \xt_\qq)$

926: are normal sequences of simple braids. Let $\mathcal{D}$ be the grid

927: diagram obtained by starting from the right column $(\xt_1, ...,

928: \xt_\qq)$ and the bottom row $(\xs_1, ..., \xs_\pp)$ and filling the

929: diagram with $C$-tiles from right to left, and from bottom to top,

930:  as shown in Figure~\ref{F:Complement}.

931: Then every path in~$\mathcal{D}$ consisting of diagonal arrows

932: followed by horizontal arrows, as well as every path consisting of

933: diagonal arrows followed by horizontal arrows  corresponds to a

934: normal sequence. 

935: 

936: $(ii)$ Let $\xx = \xs_1...\xs_\pp$ and $\yy=\xt_1...\xt_\qq$. Let

937: $(\xs'_1, ..., \xs'_\pp)$ be the top row of~$\mathcal{D}$, 

938: let $(\xt'_1, ..., \xt'_\qq)$ be its left column, and, assuming

939: $\pp\ge\qq$, let $(\xu_1, ..., \xu_\qq)$ be the diagonal from the

940: top-left corner. Then $(\xs'_1, ..., \xs'_\pp)$ is the normal form

941: of~$\xx/\yy$, while $(\xt'_1, ..., \xt'_\qq)$ is the normal form

942: of~$\yy/\xx$, and $(\xu_1, ..., \xu_\qq, \xs_{\qq+1}, ..., \xs_\pp)$

943: is the normal form of the left lcm of~$\xx$ and~$\yy$.

944: \end{prop}

945: 

946: \begin{figure}

947: \begin{picture}(110,45)

948: %\Grid

949: \put(0,-0.5){\includegraphics{Complement.eps}}

950: \put(8,0){$\xs_1$}

951: \put(36,0){$\xs_{\qq-1}$}

952: \put(52,0){$\xs_{\qq}$}

953: \put(66,0){$\xs_{\qq+1}$}

954: \put(98,0){$\xs_{\pp}$}

955: \put(8,45){$\xs'_1$}

956: \put(36,45){$\xs'_{\qq-1}$}

957: \put(52,45){$\xs'_{\qq}$}

958: \put(66,45){$\xs'_{\qq+1}$}

959: \put(98,45){$\xs'_{\pp}$}

960: \put(1,7){\llap{$\xt'_\qq$}}

961: \put(1,17){\llap{$\xt'_{\qq\!-\!1}$}}

962: \put(1,37){\llap{$\xt'_1$}}

963: \put(109,7){$\xt_\qq$}

964: \put(109,17){$\xt_{\qq\!-\!1}$}

965: \put(109,37){$\xt_1$}

966: \put(2,26){$\vdots$}

967: \put(17,26){$\vdots$}

968: \put(32,26){$\vdots$}

969: \put(47,26){$\vdots$}

970: \put(62,26){$\vdots$}

971: \put(77,26){$\vdots$}

972: \put(92,26){$\vdots$}

973: \put(107,26){$\vdots$}

974: \put(23,2.5){$\ldots$}

975: \put(23,12.5){$\ldots$}

976: \put(23,22.5){$\ldots$}

977: \put(23,32.5){$\ldots$}

978: \put(23,42.5){$\ldots$}

979: \put(83,2.5){$\ldots$}

980: \put(83,12.5){$\ldots$}

981: \put(83,22.5){$\ldots$}

982: \put(83,32.5){$\ldots$}

983: \put(83,42.5){$\ldots$}

984: \put(10,38){$\xu_1$}

985: \put(40,18){$\xu_{\qq\!-\!1}$}

986: \put(55,8){$\xu_{\qq}$}

987: \end{picture}

988: \caption{\smaller\sf Construction of the left lcm and the

989: left complements by the grid method: we start from the

990: bottom and right sides, and fill the grid with $C$-tiles, from right

991: to left and from bottom to top; then every row, every column, and

992: every diagonal in the grid consist of normal sequences.}

993: \label{F:Complement}

994: \end{figure}

995: 

996: \begin{proof}

997: By construction, the diagram~$\mathcal{D}$ is commutative. Let

998: $\zz = \xs'_1 ... \xs'_\pp \xt_1 ... \xt_\qq$. Then $\zz$ is a

999: common left multiple of~$\xx$ and~$\yy$. Moreover, an easy

1000: induction shows that every common left multiple~$\zz'$ of~$\xx$

1001: and~$\yy$ has to be a left multiple of~$\zz$: indeed, $\zz'$, being

1002: a common left multiple of~$\xs_\pp$ and~$\xt_\qq$, has to be a

1003: left multiple of the braid represented by the diagonal of the

1004: bottom-right rectangle in~$\mathcal{D}$, and we argue similarly for

1005: each of the $\pp\qq$~rectangles in~$\mathcal{D}$. Hence $\zz$ is

1006: the left lcm of~$\xx$ and~$\yy$, and, therefore, we have $\xs'_1 ...

1007: \xs'_\pp = \xx/\yy$ and $\xt'_1 ... \xt'_\qq = \yy/\xx$.

1008: 

1009: Then, by repeatedly applying Lemma~\ref{L:Complement1}, we

1010: obtain that every row and every column in~$\mathcal{D}$ is a

1011: normal sequence. As for the diagonals, and the diagonals followed by

1012: rows or columns, we similarly apply Lemma~\ref{L:Complement2}.

1013: \end{proof}

1014: 

1015: As a straightforward application, we obtain the following computing

1016: rule for determining the normal form of~$\xx\xu\inv$ from that

1017: of~$\xx$.

1018: 

1019: \begin{coro}

1020: \label{C:Complement}

1021: Assume that $(\xs_1, ..., \xs_\pp)$ is the normal form of

1022: a positive braid~$\xx$, and that $\xu$ is a simple braid that

1023: divides~$\xx$ on the right. Then the normal form of~$\xx\xu\inv$

1024: is the sequence $(\xs'_1, ..., \xs'_\pp)$ determined by

1025: the grid diagram of Figure~\ref{F:ComplementOne}.

1026: \end{coro}

1027: 

1028: \begin{figure}[htb]

1029: \begin{picture}(100,22)

1030: %\Grid

1031: \put(0,0.7){\includegraphics{ComplementOne.eps}}

1032: \put(9,1){$\xs_1$}

1033: \put(29,1){$\xs_2$}

1034: \put(69,1){$\xs_{\pp-1}$}

1035: \put(89,1){$\xs_\pp$}

1036: \put(-3.5,11){$\xu_0$}

1037: \put(22,11){$\xu_1$}

1038: \put(82,11){$\xu_{\pp\!-\!1}$}

1039: \put(96.5,11){$\xu_\pp$}

1040: \put(102,11){$\xu$}

1041: \put(9,21){$\xs'_1$}

1042: \put(29,21){$\xs'_2$}

1043: \put(69,21){$\xs'_{\pp-1}$}

1044: \put(89,21){$\xs'_\pp$}

1045: \put(50,3.5){$\ldots$}

1046: \put(50,18.5){$\ldots$}

1047: \end{picture}

1048: \caption{\smaller\sf Normal form of~$\xx\xu\inv$ from

1049: that of~$\xx$: start from the plain arrows and fill the

1050: diagram with $C$-tiles, from right to left; the expected normal

1051: form is the sequence $(\xs'_1, ..., \xs'_\pp)$ read on the top line; the

1052: hypothesis that $\xu$ is a right divisor of~$\xx$ guarantees that

1053: $\xu_0$ is~$1$---without any hypothesis, $(\xs'_1, ..., \xs'_\pp)$ is

1054: the normal form of~$\xx/\xu$, and $\xu_0$ is then~$\xu/\xx$.}

1055: \label{F:ComplementOne}

1056: \end{figure}

1057: 

1058: \subsection{Grid properties for the product}

1059: \label{S:GridProduct}

1060: 

1061: We turn to the product, with the question of determining the normal

1062: form of~$\yy\xx$ from the normal forms of~$\xx$ and~$\yy$. The

1063: method is similar to that of the previous section, with another type

1064: of basic tile.

1065: 

1066: For all simple braids~$\xt_1, \xt_2$, there exist simple

1067: braids~$\xs_1, \xs_2$ satisfying $\xs_1 \xs_2 = \xt_1 \xt_2$ and

1068: $(\xs_1, \xs_2)$ is normal. Indeed, let $\xs_1=\HH(\xt_1\xt_2)$,

1069: and~$\xs_2$ satisfying $\xt_1\xt_2 = \xs_1\xs_2$.

1070: By~\eqref{E:Head2}, $\xt_1\divel\xt_1\xt_2$ implies $\xt_1 =

1071: \HH(\xt_1) \divel \HH(\xt_1\xt_2)=\xs_1$, so we have

1072: $\xt_2=\xu\xs_2$ for some~$\xu$,  which forces~$\xs_2$ to be

1073: simple. In other words, the normal form of the product of two

1074: simple braids must consist of at most two simple braids.

1075: 

1076: \begin{defi}

1077: A commutative diagram consisting of four simple braids~$\xs_1$,

1078: $\xs_2$, $\xt_1$, $\xt_2$ satisfying $\xs_1 \xs_2 = \xt_1 \xt_2$

1079: and such that $(\xs_1, \xs_2)$ is normal is called a {\it $P$-tile}

1080: (like ``product tile''), and it is represented as in

1081: Figure~\ref{F:PTile}.

1082: \end{defi}

1083: 

1084: \begin{figure}[htb]

1085: \begin{picture}(21,16)

1086: %\Grid

1087: \put(0,0.2){\includegraphics{PTile.eps}}

1088: \put(-3,8){$\xt_1$}

1089: \put(22,8){$\xs_2$}

1090: \put(10,-2){$\xt_2$}

1091: \put(10,17){$\xs_1$}

1092: \end{picture}

1093: \caption{\smaller\sf A $P$-tile: a commutative diagram involving

1094: four simple braids such that two of them make a normal sequence.}

1095: \label{F:PTile}

1096: \end{figure}

1097: 

1098: As was done above with $C$-tiles, we establish normality results

1099: in diagrams involving $P$-tiles.

1100: 

1101: \begin{lemm}

1102: \label{L:Product1}

1103: (Figure~\ref{F:Product1}) 

1104: Assume that $\xs_1, \xs_2, \xs'_1, \xs'_2, \xt_0,

1105: \xt_1, \xt_2$ be simple braids satisfying $\xt_0\xs_1=\xs'_1\xt_1$, 

1106: $\xt_1\xs_2=\xs'_2\xt_2$, and such that $(\xs_1,\xs_2)$ and

1107: $(\xs'_1, \xt_1)$ are normal. Then $(\xs'_1, \xs'_2)$ is normal as

1108: well.

1109: \end{lemm}

1110: 

1111: \begin{figure}[htb]

1112: \begin{picture}(41,21)

1113: %\Grid

1114: \put(0,0){\includegraphics{Product1.eps}}

1115: \put(-3,10){$\xt_0$}

1116: \put(22,10){$\xt_1$}

1117: \put(42,10){$\xt_2$}

1118: \put(9,0){$\xs_1$}

1119: \put(29,0){$\xs_2$}

1120: \put(9,20){$\xs'_1$}

1121: \put(29,20){$\xs'_2$}

1122: \end{picture}

1123: \caption{\smaller\sf A grid property involving $P$-tiles:

1124: assuming that the left rectangle is a $P$-tile and the bottom

1125: sequence $(\xs_1, \xs_2)$ is normal, the top sequence $(\xs'_1,

1126: \xs'_2)$ is normal as well.}

1127: \label{F:Product1}

1128: \end{figure}

1129: 

1130: \begin{proof}

1131: $(ii)$ With the notation of Figure~\ref{F:Product1}, we compute:

1132: \begin{align*}

1133: \HH(\xs'_1\xs'_2)

1134: &\divel \HH(\xs'_1\xs'_2\xt_2)

1135: &&\text{by~\eqref{E:Head3}}\\

1136: &= \HH(\xt_0\xs_1\xs_2)

1137: &&\text{by commutativity}\\

1138: &= \HH(\xt_0\HH(\xs_1\xs_2))

1139: &&\text{by~\eqref{E:Head4}}\\

1140: &= \HH(\xt_0\xs_1)

1141: &&\text{by the hypothesis that $(\xs_1, \xs_2)$ is normal}\\ 

1142: &= \HH(\xs'_1\xt_1)

1143: &&\text{by commutativity}\\

1144: &= \xs'_1

1145: &&\text{by the hypothesis that $(\xs'_1, \xt_1)$ is normal},

1146: \end{align*}

1147: so $(\xs'_1, \xs'_2)$ is normal.

1148: \end{proof}

1149: 

1150: To go further, we need to use the duality with respect to~$\DD\nn$.

1151: For each simple $n$-braid~$\xs$, there exists a unique simple

1152: $n$-braid~$\xs^*$ satisfying $\xs \xs^* = \DD\nn$. Then, 

1153: we have $\xs\DD\nn = \xs\xs^*\xs^{**} =

1154: \DD\nn\xs^{**}$, hence $\xs^{**}$ is the conjugate~$\DD\nn\inv

1155: \xs \DD\nn$, which is easily seen to be the image of~$\xs$ under

1156: the flip automorphism~$\flip_\nn$ that exchanges~$\ss\ii$

1157: and~$\ss{\nn-\ii}$ for $1 \le \ii<\nn$. 

1158: 

1159: \begin{lemm}

1160: \label{L:Dual}

1161: Assume that $\xs,\xt$ are simple braids. Then $(\xs,

1162: \xt)$ is normal if and only if $\gcdl(\xs^*, \xt)=1$ holds.

1163: \end{lemm}

1164: 

1165: \begin{proof}

1166: We have 

1167: $$\HH(\xs \xt) 

1168: = \gcdl(\xs \xt, \DD\nn)

1169: = \gcdl(\xs \xt, \xs \xs^*)

1170: = \xs \cdot \gcdl(\xt, \xs^*),$$

1171: hence $\HH(\xs \xt) = \xs$ is equivalent to $\gcdl(\xt, \xs^*)=1$.

1172: \end{proof}

1173: 

1174: \begin{lemm}

1175: \label{L:Product2}

1176: (Figure~\ref{F:Product2}) 

1177: Assume that $\xs_0, \xs_1, \xs_2, \xt_1, \xt_2, \xt'_1, \xt'_2$ are

1178: simple braids satisfying $\xt_1\xs_1=\xs_0\xt'_1$, 

1179: $\xt_2\xs_2=\xs_1\xt'_2$, and such that $(\xt_1,\xt_2)$ and

1180: $(\xs_1, \xt'_2)$ are normal. Then $(\xt'_1, \xt'_2)$ is normal as

1181: well.

1182: \end{lemm}

1183: 

1184: \begin{figure}[htb]

1185: \begin{picture}(27,33)

1186: %\Grid

1187: \put(0.2,0.5){\includegraphics{Product2.eps}}

1188: \put(0,23){$\xt_1$}

1189: \put(0,8){$\xt_2$}

1190: \put(24,23){$\xt'_1$}

1191: \put(24,8){$\xt'_2$}

1192: \put(12,33){$\xs_0$}

1193: \put(12,18){$\xs_1$}

1194: \put(12,-1){$\xs_2$}

1195: \end{picture}

1196: \caption{\smaller\sf Another grid property involving $P$-tiles:

1197: assuming that the bottom rectangle is a $P$-tile and the left

1198: sequence $(\xt_1, \xt_2)$ is normal, the right sequence $(\xt'_1,

1199: \xt'_2)$ is normal as well.}

1200: \label{F:Product2}

1201: \end{figure}

1202: 

1203: \begin{proof}

1204: Introduce $\xs_0^*$, $\xs_1^*$ and $\xs_2^*$. Then the diagram of

1205: Figure~\ref{F:Product2bis} is commutative. As $\flip_\nn$ is an

1206: automorphism of~$\BB\nn$---and of the group~$B_\nn$ too---the

1207: hypothesis that $(\xt_1, \xt_2)$ is normal implies that

1208: $(\flip_\nn(\xt_1), \flip_\nn(\xt_2))$ is normal as well. 

1209: 

1210: Now, by Lemma~\ref{L:Dual}, the hypothesis that $(\xs_1, \xt'_2)$

1211: is normal gives $\gcdl(\xs_1^*,\xt'_2)=\nobreak1$. By

1212: Lemma~\ref{L:Complement1}, the latter condition together with the

1213: normality  of~$(\flip_\nn(\xt_1),

1214: \flip_\nn(\xt_2))$ implies that $(\xt'_1,

1215: \xt'_2)$ is normal.

1216: \end{proof}

1217: 

1218: \begin{figure}[htb]

1219: \begin{picture}(47,33)

1220: %\Grid

1221: \put(0.2,0.5){\includegraphics{Product2bis.eps}}

1222: \put(0,23){$\xt_1$}

1223: \put(0,8){$\xt_2$}

1224: \put(24,23){$\xt'_1$}

1225: \put(24,8){$\xt'_2$}

1226: \put(12,33){$\xs_0$}

1227: \put(12,18){$\xs_1$}

1228: \put(12,-1.5){$\xs_2$}

1229: \put(32,33){$\xs_0^*$}

1230: \put(32,18){$\xs_1^*$}

1231: \put(32,-1.5){$\xs_2^*$}

1232: \put(44,23){$\flip_\nn(\xt_1)$}

1233: \put(44,8){$\flip_\nn(\xt_2)$}

1234: \end{picture}

1235: \caption{\smaller\sf Proof of Lemma~\ref{L:Product2}: one

1236: introduces the dual of the horizontal arrows; the

1237: hypothesis that the left column~$(\xt_1, \xt_2)$ is normal implies

1238: that the right column $(\flip_\nn(\xt_1),\flip_\nn(\xt_2))$ is 

1239: normal as well, and, then, we apply

1240: Lemma~\ref{L:Complement1} to come back to $(\xt'_1, \xt'_2)$.}

1241: \label{F:Product2bis}

1242: \end{figure}

1243: 

1244: We deduce

1245: 

1246: \begin{prop}

1247: \label{P:Product}

1248: $(i)$ Assume that $(\xs_1, ..., \xs_\pp)$ and $(\xt_1, ..., \xt_\qq)$

1249: are normal sequences of simple braids. Let $\mathcal{D}$ be the grid

1250: diagram obtained by starting from the left column $(\xt_1, ...,

1251: \xt_\qq)$ and the bottom row $(\xs_1, ..., \xs_\pp)$ and filling the

1252: diagram with $P$-tiles from left to right, and from bottom to top,

1253: as shown in Figure~\ref{F:Product}. Then

1254: every path in~$\mathcal{D}$ consisting of horizontal arrows

1255: followed by vertical arrows corresponds to a normal sequence. 

1256: 

1257: $(ii)$ Let $\xx = \xs_1...\xs_\pp$ and $\yy=\xt_1...\xt_\qq$. Let

1258: $(\xs'_1, ..., \xs'_\pp)$ be the top row of~$\mathcal{D}$ and

1259: $(\xt'_1, ..., \xt'_\qq)$ is its right column. Then $(\xs'_1, ...,

1260: \xs'_\pp, \xt'_1, ..., \xt'_\qq)$ is the normal form of~$\yy\xx$.

1261: \end{prop}

1262: 

1263: \begin{figure}[htb]

1264: \begin{picture}(72,47)

1265: %\Grid

1266: \put(0,-0.5){\includegraphics{Product.eps}}

1267: \put(0,37){$\xt_1$}

1268: \put(3,17){\llap{$\xt_{\qq\!-\!1}$}}

1269: \put(0,7){$\xt_\qq$}

1270: \put(9,0){$\xs_1$}

1271: \put(24,0){$\xs_2$}

1272: \put(59,0){$\xs_\pp$}

1273: \put(3,26){$\vdots$}

1274: \put(18,26){$\vdots$}

1275: \put(33,26){$\vdots$}

1276: \put(53,26){$\vdots$}

1277: \put(68,26){$\vdots$}

1278: \put(41,2.5){$\ldots$}

1279: \put(41,12.5){$\ldots$}

1280: \put(41,22.5){$\ldots$}

1281: \put(41,32.5){$\ldots$}

1282: \put(41,42.5){$\ldots$}

1283: \put(69,37){$\xt'_1$}

1284: \put(69,17){$\xt'_{\qq\!-\!1}$}

1285: \put(69,7){$\xt'_{\qq}$}

1286: \put(9,44.5){$\xs'_1$}

1287: \put(24,44.5){$\xs'_2$}

1288: \put(59,44.5){$\xs'_\pp$}

1289: \end{picture}

1290: \caption{\smaller\sf Computation of the normal form of a product

1291: by means of a grid: we start from the left and the bottom sides, and

1292: fill the diagram using $P$-tiles, from left to right and from bottom to

1293: top; then the top row and the right column are normal sequences as

1294: well, and so is the sequence obtained by concatenating them.}

1295: \label{F:Product}

1296: \end{figure}

1297: 

1298: \begin{proof}

1299: By hypothesis, the bottom row is normal, hence, by

1300: Lemma~\ref{L:Product1}, we inductively deduce that the $\kk$th

1301: row from the bottom is normal. Similarly, by hypothesis, the left

1302: column is normal, hence, by Lemma~\ref{L:Product2}, we inductively

1303: deduce that $\kk$th column from the left is normal. Finally, the

1304: sequence $(\xs'_1, ...,

1305: \xs'_\pp, \xt'_1, ..., \xt'_\qq)$ is normal, as it is the concatenation

1306: of two normal sequences and, moreover, $(\xs'_\pp, \xt'_1)$ is

1307: normal by construction. As, by construction, the

1308: diagram~$\mathcal{D}$ is commutative, the product of the latter

1309: sequence is also the product of the left column and the bottom row,

1310: \ie, it is the braid~$\yy\xx$.

1311: \end{proof}

1312: 

1313: As in the case of the quotient, we deduce in particular rules for

1314: computing the normal form of $\xu\xx$ and~$\xx\xu$ from that

1315: of~$\xx$ when $\xu$ is a simple braid.

1316: 

1317: \begin{coro}

1318: \label{C:Product}

1319: Assume that $(\xs_1, ..., \xs_\pp)$ is the normal form

1320: of the positive braid~$\xx$, and that $\xu$ is a simple braid. Then

1321: the normal forms of~$\xu\xx$ and~$\xx\xu$ are the sequences

1322: $(\xs'_1, ..., \xs'_\pp, \xu'_\pp)$ and $(\xu''_0, \xs''_1, ..., \xs''_\pp)$ 

1323: determined by the grid diagrams of Figure~\ref{F:ProductOne}.

1324: \end{coro}

1325: 

1326: \begin{figure}[htb]

1327: \begin{picture}(101,45)

1328: %\Grid

1329: \put(0,.3){\includegraphics{ProductOne.eps}}

1330: \put(9,0){$\xs''_1$}

1331: \put(29,0){$\xs''_2$}

1332: \put(68,0){$\xs''_{\pp-1}$}

1333: \put(89,0){$\xs''_\pp$}

1334: \put(9,20){$\xs_1$}

1335: \put(29,20){$\xs_2$}

1336: \put(68,20){$\xs_{\pp-1}$}

1337: \put(89,20){$\xs_\pp$}

1338: \put(9,26){$\xs_1$}

1339: \put(29,26){$\xs_2$}

1340: \put(68,26){$\xs_{\pp-1}$}

1341: \put(89,26){$\xs_\pp$}

1342: \put(9,45.5){$\xs'_1$}

1343: \put(29,45.5){$\xs'_2$}

1344: \put(68,45.5){$\xs'_{\pp-1}$}

1345: \put(89,45.5){$\xs'_\pp$}

1346: \put(49,3.5){$\ldots$}

1347: \put(50,18.5){$\ldots$}

1348: \put(49,28.5){$\ldots$}

1349: \put(50,43.5){$\ldots$}

1350: \put(-3.5,11){$\xu''_0$}

1351: \put(22,11){$\xu''_1$}

1352: \put(82,11){$\xu''_{\pp-1}$}

1353: \put(96.5,11){$\xu''_{\pp}$}

1354: \put(102,11){$\xu$}

1355: \put(-2,36){$\xu$}

1356: \put(2,36){$\xu'_0$}

1357: \put(22,36){$\xu'_1$}

1358: \put(82,36){$\xu'_{\pp-1}$}

1359: \put(102,36){$\xu'_\pp$}

1360: \end{picture}

1361: \caption{\smaller\sf Construction of the normal

1362: from of~$\xu\xx$ (above) and $\xx\xu$ (below) from the normal

1363: form $(\xs_1, ..., \xs_\pp)$ of~$\xx$: start with $\xu'_0:=\xu$ (\resp 

1364: with $\xu''_\pp:= \xu$)  and fill the diagram with $P$-tiles from left to right

1365: (\resp from right to left); then $(\xs'_1, ..., \xs'_\pp,

1366: \xu'_\pp)$ (\resp $(\xu''_0, \xs''_1, ..., \xs''_\pp)$) is the expected

1367: normal form---up to removing $\xu'_\pp$ or~$\xs''_\pp$ if the

1368: latter happen to be~$1$.}

1369: \label{F:ProductOne}

1370: \end{figure}

1371: 

1372: \subsection{Application to computing normal forms}

1373: 

1374: We can now easily provide algorithms that compute the greedy

1375: normal form and the symmetric normal form of a braid starting from

1376: an arbitrary word that represents it. Clearly, the point is, starting 

1377: from the greedy (\resp symmetric) normal form of a braid~$\zz$, to

1378: be able to determine the greedy (\resp symmetric) normal form

1379: of~$\zz\xu^{\pm1}$ when $\xu$ is a simple braid, hence in

1380: particular a generator~$\ss\ii$. The solutions come as easy

1381: applications of the results of the previous sections. 

1382: 

1383: \subsubsection{Computing the greedy normal form}

1384: 

1385: We recall that, for~$\xs$ a simple $\nn$-braid, $\xs^*$ denotes the

1386: unique (simple) braid satisfying $\xs \xs^* = \DD\nn$.

1387: Symmetrically, we denote by $\dual\xs$ the unique (simple) braid

1388: satisfying $\dual\xs \xs=\DD\nn$. Note that, by construction, 

1389: $(\dual\xs)^*=\xs$ holds for each simple braid~$\xs$.

1390: 

1391: \begin{prop}

1392: \label{P:AlgoGreedy}

1393: Assume that the $\nn$-greedy normal form of~$\zz$

1394: is $(\mm; \xs_1, ..., \xs_\pp)$ and $\xu$ is a simple $\nn$-braid. 

1395: 

1396: $(i)$ Let $\xs'_\pp$, $\xu_{\pp-1}$, ..., $\xs'_1$, $\xu_0$ be determined 

1397: by filling the top diagram of Figure~\ref{F:AlgoGreedy}. Then the greedy 

1398: normal form of~$\zz\xu$ is $(\mm+1; \xs'_1, ..., \xs'_\pp)$ if

1399: $\xu_0=\DD\nn$ holds, and $(\mm; \xu_0, \xs'_1, ..., \xs'_\pp)$ otherwise.

1400: 

1401: $(ii)$ Let $\xs'_\pp$, $\xu_{\pp-1}$, ..., $\xs'_1$, $\xu_0$ be determined by

1402: filling the bottom diagram of Figure~\ref{F:AlgoGreedy}. Then the greedy 

1403: normal form of~$\zz\xu\inv$ is $(\mm;  \xs'_1, ..., \xs'_\pp)$ if $\xu_0=1$ holds,

1404: and $(\mm-1; \dual{u_0}, \xs'_1, ..., \xs'_\pp)$ otherwise.

1405: \end{prop}

1406: 

1407: \begin{figure}[htb]

1408: \begin{picture}(120,50)

1409: %\Grid

1410: \put(0,0){\includegraphics{AlgoGreedy.eps}}

1411: \put(5,-0.3){$\DD\nn^{\mm-1}$}

1412: \put(28,0){$\DD\nn$}

1413: \put(48,0){$\xs_1$}

1414: \put(87,0){$\xs_{\pp\!-\!1}$}

1415: \put(108,0){$\xs_\pp$}

1416: \put(48,20){$\xs'_1$}

1417: \put(87,20){$\xs'_{\pp\!-\!1}$}

1418: \put(108,20){$\xs'_\pp$}

1419: \put(68,3.5){$\ldots$}

1420: \put(68,18.5){$\ldots$}

1421: \put(28,20){$\dual{\xu_0}$}

1422: \put(21,11){$1$}

1423: \put(41,11){$\xu_0$}

1424: \put(61,11){$\xu_1$}

1425: \put(81,11){$\xu_{\pp\!-\!2}$}

1426: \put(101,11){$\xu_{\pp\!-\!1}$}

1427: \put(116,11){$\xu_\pp$}

1428: \put(121,11){$\xu$}

1429: \put(17,47){$\DD\nn^\mm$}

1430: \put(38,47){$\xs_1$}

1431: \put(77,47){$\xs_{\pp\!-\!1}$}

1432: \put(98,47){$\xs_\pp$}

1433: \put(38,27){$\xs'_1$}

1434: \put(77,27){$\xs'_{\pp\!-\!1}$}

1435: \put(98,27){$\xs'_\pp$}

1436: \put(58,30){$\ldots$}

1437: \put(58,45){$\ldots$}

1438: \put(31,37){$\xu_0$}

1439: \put(51,37){$\xu_1$}

1440: \put(71,37){$\xu_{\pp\!-\!2}$}

1441: \put(91,37){$\xu_{\pp\!-\!1}$}

1442: \put(106,37){$\xu_\pp$}

1443: \put(111,37){$\xu$}

1444: \end{picture}

1445: \caption{\smaller\sf Greedy normal form of~$\zz\xu$

1446: (top) and~$\zz\xu\inv$ (bottom) from the greedy normal

1447: form $(\mm; \xs_1, ..., \xs_\pp)$ of~$\zz$: starting from

1448: $\xu_\pp:= \xu$, fill the diagram using $P$-tiles (\resp $C$-tiles)

1449: from right to left, and adapt at the the left end to guarantee the

1450: connection with the $\DD\nn^\mm$-factor, namely

1451: include $\xu_0$ in the latter if needed (top), or factorize

1452: $\DD\nn\xu_0\inv$ into~$\dual{\xu_0}$ (bottom).}

1453: \label{F:AlgoGreedy}

1454: \end{figure}

1455: 

1456: \begin{proof}

1457: $(i)$ By commutativity of the diagram, we have $\zz\xu =

1458: \DD\nn^\mm\xu_0\xs'_1...\xs'_\pp$, so the point is to check that

1459: the sequence $(\mm; \xu_0, \xs'_1, ..., \xs'_\pp)$, or $(\mm+1;

1460: \xs'_1, ..., \xs'_\pp)$, is normal. Now Corollary~\ref{C:Product}

1461: guarantees that $(\xu_0, \xs'_1, ..., \xs'_\pp)$ is normal. According

1462: to whether $\xu_0$ equals~$\DD\nn$ or not, we integrate the

1463: factor~$\xu_0$ in~$\DD\nn^\mm$, and we obtain a greedy normal

1464: form, hence the greedy normal form of~$\zz\xu$ by uniqueness.

1465: 

1466: $(ii)$ By commutativity, we have $\zz\xu\inv = \DD\nn ^\mm \xu_0\inv

1467: \xs'_1 ... \xs'_\pp = \DD\nn^{\mm-1} \dual{\xu_0} \xs'_1 ...

1468: \xs'_\pp$, and the point is to check that the expected sequences are

1469: greedy normal forms. Corollary~\ref{C:Complement} guarantees

1470: that $(\xs'_1, ..., \xs'_\pp)$ is normal. So, if $\xu_0=1$

1471: holds, $(\mm; \xs'_1, ..., \xs'_\pp)$ is a greedy normal form, hence it

1472: is the greedy normal form of~$\zz\xu\inv$; otherwise, we observe

1473: that $(\mm\nobreak-\nobreak1; \dual{\xu_0}, \xs'_1, ..., \xs'_\pp)$

1474: is a greedy normal form, as $\xu_0\not=1$ implies

1475: $\dual{\xu_0}\not=\DD\nn$, and, by Lemma~\ref{L:Dual}, 

1476: the hypothesis $\gcdl(\xu_0, \xs'_1)=1$ is equivalent to

1477: $(\dual{\xu_0}, \xs'_1)$ being normal since $\xu_0 =

1478: (\dual{\xu_0})^*$ holds.

1479: \end{proof}

1480: 

1481: \begin{exam}

1482: \label{X:AlgoGreedy}

1483: Let $\ww_0$ be the braid word $\mathtt{aBabacABABAbbCB}$---the

1484: randomly chosen $4$-braid word illustrated in

1485: Figure~\ref{F:Braid}, which will be repeatedly considered in

1486: the sequel. Starting with $(0; \emptyset)$, which is the greedy

1487: normal form of~$1$, and applying the algorithm of 

1488: Proposition~\ref{P:AlgoGreedy} to the successive letters of~$\ww$,

1489: we obtain the $4$-greedy normal form of the prefixes

1490: of~$\ww$, namely (in braid words form):

1491: 

1492: $0: \e \gives (0; \emptyset)$

1493: 

1494: $1: \mathtt{a} \gives (0; \mathtt{a})$

1495: 

1496: $2: \mathtt{aB} \gives (-1; \mathtt{abcb,ba})$

1497: 

1498: $3: \mathtt{aBa} \gives (-1; \mathtt{abcb,ba,a})$

1499: 

1500: $4: \mathtt{aBab} \gives (-1; \mathtt{abcb,ba,ab})$

1501: 

1502: $5: \mathtt{aBaba} \gives (0; \mathtt{a,ab})$

1503: 

1504: $6: \mathtt{aBabac} \gives (0; \mathtt{a,abc})$

1505: 

1506: \dots

1507: 

1508: $14:  \mathtt{aBabacABABAbbC} \gives (-2;

1509: \mathtt{ac, abcb, bcba, ab})$

1510: 

1511: $15:  \mathtt{aBabacABABAbbCB} \gives (-2;

1512: \mathtt{ac, abcb, bcba, a})$.

1513: 

1514: \noindent Thus the greedy normal form of the braid represented

1515: by~$\ww$ is the sequence

1516: 

1517: $(-2; \mathtt{ac, abcb, bcba, a})$, 

1518: 

1519: \noindent\ie,

1520: in permutation form, 

1521: 

1522: $(-2; (2,1,4,3), (2,4,3,1), (4,1,3,2), (2,1,3,4))$

1523: 

1524: \noindent---this is the greedy normal

1525: form of Example~\ref{X:NF}. As the initial word~$\ww_0$ contains

1526: $15$~letters, computing its greedy normal form entails

1527: $15$~applications of Proposition~\ref{P:AlgoGreedy}. However, one

1528: can speed up the process by gathering adjacent letters that

1529: together represent a simple braid: for instance, $\mathtt{abac}$

1530: represents a simple braid, so steps~$3$ to~$6$ above

1531: can be gathered into a single step corresponding to multiplying

1532: by the simple braid~$\mathtt{abac}$. By Corollary~\ref{C:Greedy},

1533: we deduce that the braid word~$\ww_0$ does not represent~$1$

1534: in~$B_4$.

1535: \end{exam}

1536: 

1537: As for complexity analysis, two cases are to be considered. If the

1538: braid index~$\nn$ is fixed, and, for practical implementations, has a

1539: small value, say $\nn\le6$, then one can precompute the

1540: tables of the binary operations $(\xs,\xt)\mapsto \HH(\xs\xt)$,

1541: $(\xs,\xt)\mapsto \HH(\xs\xt)\inv\xs\xt$, and $(\xs,\xt)\mapsto

1542: \gcdl(\xs,\xt)$, in which case the determination of the greedy

1543: normal form for a braid word of length~$\ell$ has

1544: complexity~$O(\ell^2)$, and is easy and quick in practice. Otherwise,

1545: there are too many simple $\nn$-braids to store all results, and one

1546: has to compute the values of~$\HH(\xs\xt)$ and~$\gcdl(\xs,\xt)$

1547: locally. As explained in~\cite[Chapter 9]{Eps},

1548: this is essentially a sorting process, and, therefore, each such

1549: computation can be done in time~$O(\nn\log\nn)$, resulting in a

1550: global time complexity~$O(\ell^2\nn\log\nn)$ for the computation

1551: of the normal form for an $\nn$-braid word of length~$\ell$.

1552: 

1553: \subsubsection{Computing the symmetric normal form}

1554: 

1555: We now consider the symmetric normal form of

1556: Section~\ref{S:Symmetric}, and show how to compute the symmetric

1557: normal form of~$\zz\xu$ and $\zz\xu\inv$ form that of~$\zz$

1558: by using convenient grid diagrams. The method is similar to that for

1559: the greedy normal, but a bit more care is needed for the transition

1560: between the numerator and the denominator in the case of a

1561: product.

1562: 

1563: \begin{prop}

1564: \label{P:AlgoSymmetric}

1565: Assume that the symmetric normal form of~$\zz$

1566: is the double sequence $(\xt_1, ..., \xt_\qq; \xs_1, ..., \xs_\pp)$, and

1567: $\xu$ is a simple braid. 

1568: 

1569: $(i)$ Let $\xs'_\pp$, $\xu_{\pp-1}$, ..., $\xs'_1$, $\xu_0$, $\xv_0$,

1570: $\xv_1$, $\xt'_2$, $\xv_2$,  ... , $\xt'_\qq$, $\xv_\qq$ be

1571: determined by filling the top diagram of

1572: Figure~\ref{F:AlgoSymmetric}. Then the symmetric normal 

1573: form of~$\zz\xu$ is $(\xt'_2, ..., \xt'_\qq, \xv_\qq; \xv_0^*,  \xs'_1, ..., \xs'_\pp)$, 

1574: 

1575: $(ii)$ Let $\xs'_\pp$, $\xu_{\pp-1}$, ..., $\xs'_1$, $\xu_0=\xv_0$,

1576: $\xv_1$, $\xt'_2$, ... , $\xt'_\qq$, $\xv_\qq$ are determined by

1577: filling the bottom diagram of Figure~\ref{F:AlgoSymmetric}. Then the 

1578: symmetric normal form of~$\zz\xu\inv$ is $(\xt'_1, ...,

1579: \xt'_\qq, \xv_\qq;  \xs'_1, ..., \xs'_\pp)$ , 

1580: \end{prop}

1581: 

1582: \begin{figure}[htb]

1583: \begin{picture}(140,51)

1584: %\Grid

1585: \put(-10,-0.5){\includegraphics{AlgoSymmetric.eps}}

1586: \put(9,0){$\xt_\qq$}

1587: \put(49,0){$\xt_1$}

1588: \put(69,0){$\xs_1$}

1589: \put(109,0){$\xs_\pp$}

1590: \put(9,20){$\xt'_\qq$}

1591: \put(49,20){$\xt'_1$}

1592: \put(69,20){$\xs'_1$}

1593: \put(109,20){$\xs'_\pp$}

1594: \put(29,2.7){$\ldots$}

1595: \put(29,17.5){$\ldots$}

1596: \put(89,2.7){$\ldots$}

1597: \put(89,17.5){$\ldots$}

1598: \put(-3,10){$\xv_\qq$}

1599: \put(21.5,10){$\xv_{\qq\!-\!1}$}

1600: \put(41.5,10){$\xv_1$}

1601: \put(57,10){$\xv_0$}

1602: \put(62,10){$\xu_0$}

1603: \put(82,10){$\xu_1$}

1604: \put(101.5,10){$\xu_{\pp\!-\!1}$}

1605: \put(116,10){$\xu_\pp$}

1606: \put(122,10){$\xu$}

1607: 

1608: \put(-1,48.5){$\xt_\qq$}

1609: \put(39,48.5){$\xt_2$}

1610: \put(59,48.5){$\xt_1$}

1611: \put(79,48.5){$\xs_1$}

1612: \put(119,48.5){$\xs_\pp$}

1613: \put(-1,28.5){$\xt'_\qq$}

1614: \put(39,28.5){$\xt'_2$}

1615: \put(78,25){$\xv_0^*$}

1616: \put(84,28.5){$\xs'_1$}

1617: \put(119,28.5){$\xs'_\pp$}

1618: \put(19,46.7){$\ldots$}

1619: \put(99,46.7){$\ldots$}

1620: \put(19,31.7){$\ldots$}

1621: \put(99,31.7){$\ldots$}

1622: \put(-8.5,39){$\xv_\qq$}

1623: \put(11,39){$\xv_{\qq\!-\!1}$}

1624: \put(31.5,39){$\xv_2$}

1625: \put(51.5,39){$\xv_1$}

1626: \put(76,39){$\xu_0$}

1627: \put(55,28.5){$\xv_0$}

1628: \put(66,30){$1$}

1629: \put(71,30){$\DD\nn$}

1630: \put(60,39){$\dual{\xu_0}$}

1631: \put(91.5,39){$\xu_1$}

1632: \put(111.5,39){$\xu_{\pp\!-\!1}$}

1633: \put(126.5,39){$\xu_\pp$}

1634: \put(131.5,39){$\xu$}

1635: \end{picture}

1636: \caption{\smaller\sf Symmetric normal form of~$\zz\xu$

1637: (top) and~$\zz\xu\inv$ (bottom) from the symmetric

1638: normal form $(\xt_1, ..., \xt_\qq; \xs_1, ..., \xs_\pp)$

1639: of~$\zz$: start from $\xu_\pp := \xu$, and fill the diagram 

1640: from right to left using $P$-tiles  (\resp

1641: $C$-tiles, then $P$-tiles); in the case of the product, the transition 

1642: is as follows: having got~$\xu_0$, we find $\dual{\xu_0}$ using a

1643: $C$-tile, then let $(\xv_0, \xv_1)$ be the normal form of $\dual{\xu_0}\xt_1$

1644: using a $P$-tile, and

1645: continue with $\xv_1, \xt_2$, \etc.; in the case of the

1646: quotient, we simply put $\xv_0:=\xu_0$, and continue

1647: with $\xv_0, \xt_1$, \etc}

1648: \label{F:AlgoSymmetric}

1649: \end{figure}

1650: 

1651: \begin{proof}

1652: $(i)$ By commutativity of the diagram, we have $\zz\xu =

1653: \xv_\qq\inv \xt'_\qq{}\inv ... \xt'_2{}\inv \xv_0^* \xs'_1 ... \xs'_\pp$, and

1654: the point is to show that the double sequence $(\xt'_2, ..., \xt'_\qq, \xv_\qq; 

1655: \xv_0^*, \xs'_1, .., \xs'_\pp)$ is a symmetric normal form. 

1656: Corollary~\ref{C:Product} guarantees that the sequences $(\xu_0, \xs'_1, ..., \xs'_\pp)$

1657: and $(\xv_0, \xt'_2, ..., \xt'_\qq, \xv_\qq)$ are normal. There

1658: remain two points to check, namely that $\gcdl(\xt'_2,

1659: \xv_0^*)$ is~$1$, and that $(\xv_0^*,\xs'_1)$ is normal.

1660: For the first relation, Lemma~\ref{L:Product1} implies that $(\xv_0,

1661: \xt'_2)$ is normal, which is equivalent to

1662: $\gcdl(\xt'_2,

1663: \xv_0^*)=\nobreak1$ by Lemma~\ref{L:Dual}.  

1664: 

1665: As for the second relation, \ie, for the normality of~$(\xv_0^*,\xs'_1)$,

1666:  by Lemma~\ref{L:Dual} again, it is equivalent to

1667: $\gcdl(\xv_0^{**},\xs'_1)=1$, hence to

1668: $\gcdl(\xv_0,\flip_\nn\inv(\xs'_1))=1$. Now  we have

1669: $\xv_0\xv_1= \dual{\xu_0}\xt_1$, and $\xs_1\xu_1 =

1670: \xu_0\xs'_1$, hence 

1671: $$\dual{\xu_0} \xs_1 \xu_1 

1672: = \DD\nn \xs'_1

1673: = \flip_\nn\inv(\xs'_1) \DD\nn

1674: = \flip_\nn\inv(\xs'_1) \dual{\xu_1} \xu_1,$$

1675: hence $\flip_\nn\inv(\xs'_1) \dual{\xu_1} = \dual{\xu_0}\xs_1$.

1676: Assume $\xs$ is a simple left divisor of~$\xv_0$

1677: and~$\flip_\nn\inv(\xs'_1)$. Then, {\it a fortiori}, we have

1678: $\xs\divel \xv_0 \xv_1$ and $\xs \divel

1679: \flip_\nn\inv(\xs'_1)\dual{\xu_1}$, hence, by the above

1680: computations, $\xs \divel \dual{\xu_0} \xt_1$ and $\xs \divel

1681: \dual{\xu_0} \xs_1$, and, therefore, $\xs \divel \dual{\xu_0}$ as,

1682: by hypothesis, $\gcdl(\xs_1, \xt_1)= 1$ holds. Now, by hypothesis,

1683: $(\xu_0, \xs'_1)$ is normal, hence, by Lemma~\ref{L:Dual}, we have

1684: $\gcdl(\xu_0^*, \xs'_1)=1$, and therefore

1685: $\gcdl(\flip_\nn\inv(\xu_0^*), \flip_\nn\inv(\xs'_1)) =1$, \ie,

1686: $\gcdl(\dual{\xu_0}, \flip_\nn\inv(\xs'_1))=1$. So we must have

1687: $\xs=1$, implying $\gcdl(\xv_0,\flip_\nn\inv(\xs'_1))=1$, which

1688: was seen above to be equivalent to $(\xv_0^*,\xs'_1)$ being normal.

1689: So the proof is complete.

1690: 

1691: $(ii)$ The argument for the quotient is similar. By commutativity of

1692: the diagram, we have $\zz\xu\inv = \xv_\qq\inv

1693: \xt'_\qq{}\inv ... \xt'_1{}\inv \xs'_1... \xs'_\pp$, and, once

1694: again, the point is to check that the double sequence

1695: $(\xt'_1, ..., \xt'_\qq, \xv_\qq; \xs'_1, ..., \xs'_\pp)$ is a

1696: symmetric normal form. Corollary~\ref{C:Complement} 

1697: implies that $(\xs'_1, ..., \xs'_\pp)$ is normal, and

1698: Corollary~\ref{C:Product} implies that $(\xt'_1, ..., \xt'_\qq,

1699: \xv_\qq)$ is normal, so the only remaining point to check is

1700: $\gcdl(\xs'_1, \xt'_1)=1$. Now assume that $\xs$ is a simple left

1701: divisor of~$\xs'_1$ and~$\xt'_1$. Then, {\it a fortiori}, we have

1702: $\xs \divel \xs'_1\xu_1 = \xu_0\xs_1$ and $\xs \divel \xt'_1 \xv_1

1703: = \xu_0 \xt_1$. As $\gcdl(\xs_1, \xt_1)=1$ holds by hypothesis, we

1704: deduce $\xs \divel \xu_0$, and, finally, $\xs=1$ as, by construction,

1705: we have $\gcdl(\xu_0, \xs'_1)=1$. Hence $\gcdl(\xs'_1, \xt'_1)=1$

1706: holds.

1707: \end{proof}

1708: 

1709: \begin{exam}

1710: \label{X:AlgoSymmetric}

1711: Let $\ww_0$ be the braid word $\mathtt{aBabacABABAbbCB}$ again.

1712: Applying the algorithm of  Proposition~\ref{P:AlgoSymmetric} to the

1713: successive letters of~$\ww$ leads to the symmetric normal forms

1714: (here in braid word form):

1715: 

1716: $0: \e \gives (\emptyset; \emptyset)$

1717: 

1718: $1: \mathtt{a} \gives (\emptyset; \mathtt{a})$

1719: 

1720: $2: \mathtt{aB} \gives (\mathtt{ab}; \mathtt{ba})$

1721: 

1722: $3: \mathtt{aBa} \gives (\mathtt{ab}; \mathtt{ba,a})$

1723: 

1724: $4: \mathtt{aBab} \gives (\mathtt{ab}; \mathtt{ba,ab})$

1725: 

1726: $5: \mathtt{aBaba} \gives (\emptyset; \mathtt{a,ab})$

1727: 

1728: $6: \mathtt{aBabac} \gives (\emptyset; \mathtt{a,abc})$

1729: 

1730: \dots

1731: 

1732: $14:  \mathtt{aBabacABABAbbC} \gives (\mathtt{ab, bacb};

1733: \mathtt{bcba, ab})$

1734: 

1735: $15:  \mathtt{aBabacABABAbbCB} \gives (\mathtt{ab, bacb};

1736: \mathtt{bcba, a})$.

1737: 

1738: \noindent Thus the symmetric normal form of the braid represented

1739: by~$\ww_0$ is the double sequence  $(\mathtt{ab, bacb};

1740: \mathtt{bcba, a})$, \ie, in permutation form, 

1741: 

1742: $((2,3,1,4), (3,4,1,2); (4,1,3,2), (2,1,3,4))$

1743: 

1744: \noindent---this is the symmetric

1745: normal form of Example~\ref{X:Symmetric}. As in the case of the

1746: greedy normal form, we observe that the steps corresponding to a

1747: single simple factor can be gathered. We deduce from

1748: Corollary~\ref{C:Symmetric} that $\ww_0$ does not represent~$1$

1749: in~$B_4$.

1750: \end{exam}

1751: 

1752: The complexity analysis is the same as in the case of the greedy

1753: normal form: for a fixed braid index~$\nn$, the algorithm of

1754: Proposition~\ref{P:AlgoSymmetric} is quadratic in the length of the

1755: initial word; when $\nn$ is not fixed, a multiplicative

1756: factor~$\nn\log\nn$ has to be inserted.

1757: 

1758: \begin{rema}

1759: The diagrams of Figures~\ref{F:AlgoGreedy}

1760: and~\ref{F:AlgoSymmetric} make it clear that the greedy and

1761: symmetric normal forms satisfy the fellow traveler property

1762: of~\cite{Eps} and therefore are connected with an automatic

1763: structure on the braid group~$B_\nn$.

1764: 

1765: Also, let us point here that the greedy and symmetric normal 

1766: forms exist for a class of structures that is much wider than braid

1767: groups, namely the so-called Garside groups of~\cite{Dfx, Dgk}, and

1768: even more: thin groups of fractions of~\cite{Dgo}, Garside

1769: categories of~\cite{Kra, DiM, Bes}---all eligible for the grid properties

1770: described above.

1771: 

1772: Finally, we mention the existence of alternative normal

1773: forms~\cite{Bur, Dho} whose computation has the same complexity as

1774: the greedy and symmetric normal forms, but which rely on different

1775: bases and are connected with the braid order alluded to in

1776: Section~\ref{S:Handle} below. So far, there seems to be no reason to

1777: expect these normal forms to be more suitable for practical

1778: applications than those described above.

1779: \end{rema}

1780: 

1781: \section{Direct solutions}

1782: \label{S:Direct}

1783: 

1784: Using a normal form is not the only way for solving the braid isotopy

1785: problem. Besides the solutions of Section~\ref{S:Normal}, there exist

1786: alternative solutions directly deciding whether a given braid

1787: word~$\ww$ represents the trivial braid~$1$ or not. 

1788: 

1789: Using such solutions entails working with arbitray

1790: braid words. As discussed in~\cite{Dgw}, this option makes the

1791: computation of product and inverse obvious (merely concatenating

1792: or reversing words), at the expense of making equivalence not

1793: obvious, whereas the option of using a normal form and restricting

1794: to normal words makes equivalence obvious (a mere equality), but

1795: makes the algebraic operations of product and inverse less obvious,

1796: as normalization processes such as those described in

1797: Propositions~\ref{P:Product} and~\ref{P:Complement} are needed.

1798: 

1799: Here we describe three direct solutions to the braid isotopy problem,

1800: namely two syntactic solutions based on some word rewrite systems,

1801: and one geometric method due to I.\,Dynnikov which consists in

1802: attributing integral coordinates to every braid.

1803: 

1804: \subsection{Word redressing}

1805: \label{S:Redressing}

1806: 

1807: Word redressing (also called reversing in literature) is a simple

1808: syntactic transformation that consists in pushing the positive  

1809: letters~$\ss\ii$ in one direction and the negative letters~$\sss\ii$ in

1810: the other direction, until a word of the form ``all positive, then all

1811: negative'' is obtained. This leads to solutions of the word problem

1812: that admit a quadratic complexity. The underlying theory is Garside

1813: theory again.

1814: 

1815: \subsubsection{Description}

1816: 

1817: Redressing a braid word consists in looking for the subwords of the

1818: form~$\sss\ii\ss\jj$, \ie, one negative letter followed by a positive

1819: letter, and transforming them into equivalent patterns consisting of

1820: positive letters followed by negative letters, \ie, replacing

1821: negative--positive subwords with equivalent positive--negative words.

1822: 

1823: \begin{defi}

1824: \label{D:Redr}

1825: \cite{Dff, Dgp}

1826: Assume that $\ww, \ww'$ are braid words. We say that $\ww$ is {\it

1827: right redressible} to~$\ww'$ in one step if $\ww'$ is obtained

1828: from~$\ww$ either by deleting some subword of the

1829: form~$\sss\ii\ss\ii$, or by replacing some subword of the

1830: form~$\sss\ii\ss\jj$ with $\vert\ii-\jj\vert\ge2$ by $\ss\jj\sss\ii$,

1831: or by replacing some subword of the

1832: form~$\sss\ii\ss\jj$ with $\vert\ii-\jj\vert=1$ by

1833: $\ss\jj\ss\ii\sss\jj\sss\ii$. We say that $\ww\redrr\ww'$ holds if

1834: there is a finite sequence $\ww_0=\nobreak\ww$, ..., $\ww_\NN=\ww'$

1835: such that $\ww_\kk$ is right redressible to~$\ww_{\kk+1}$ in one step

1836: for each~$\kk$.

1837: \end{defi}

1838: 

1839: By construction, the words that are terminal for redressing are the

1840: words that include no subword of the form~$\sss\ii\ss\jj$, \ie, the

1841: words of the form~$\uu\vv{}\inv$ with $\uu,\vv$ positive braid

1842: words. 

1843: 

1844: \begin{theo}

1845: \cite{Dff}

1846: \label{T:Redressing}

1847: For each braid word~$\ww$, there exist two unique positive

1848: braid words~$\uu,\vv$ such that $\ww \redrr \uu\vv{}\inv$ holds.

1849: \end{theo}

1850: 

1851: A new solution to the braid isotopy problem follows: 

1852: 

1853: \begin{coro}

1854: \label{C:Redressing}

1855: A braid word~$\ww$ represents~$1$ in the braid group if and

1856: only if, denoting by~$\uu$ and~$\vv$ the positive words for which

1857: $\ww \redrr \uu\vv{}\inv$ holds, we have $\vv{}\inv\uu \redrr

1858: \e$, where $\e$ denotes the empty word.

1859: \end{coro}

1860: 

1861: \begin{exam}

1862: \label{X:Redressing}

1863: Let us start again with the braid

1864: word~$\ww_0 = \mathtt{aBabacABABAbbCB}$ of

1865: Examples~\ref{X:AlgoGreedy} and~\ref{X:AlgoSymmetric}. Owing to

1866: Corollary~\ref{C:Redressing}, we decide whether $\ww$ is trivial

1867: or not by redressing~$\ww$ to a word of the form~$\uu\vv{}\inv$

1868: with $\uu, \vv$ positive words, then redressing $\vv{}\inv\uu$ again,

1869: and looking whether we finally obtain the empty word. In the

1870: current case, selecting at each step the leftmost pattern of the

1871: form~$\sss\ii\ss\jj$ (underlined in the words

1872: below), the successive words are as follows

1873: 

1874: $0: \mathtt{a\underline{Ba}bacABABAbbCB}$

1875: 

1876: $1: \mathtt{aabA\underline{Bb}acABABAbbCB}$

1877: 

1878: $2: \mathtt{aab\underline{Aa}cABABAbbCB}$

1879: 

1880: $3: \mathtt{aabcABAB\underline{Ab}bCB}$

1881: 

1882: $4: \mathtt{aabcABA\underline{Bb}aBAbCB}$

1883: 

1884: \ldots

1885: 

1886: $11: \mathtt{aabcbABBACB}$.

1887: 

1888: \noindent At this point, we switch the positive and negative

1889: subword, and redress again:

1890: 

1891: $12: \mathtt{ABBAC\underline{Ba}abcb}$

1892: 

1893: $13: \mathtt{ABBA\underline{Ca}bABabcb}$

1894: 

1895: $14: \mathtt{ABB\underline{Aa}CbABabcb}$

1896: 

1897: \ldots

1898: 

1899: $37: \mathtt{cbaa\underline{Bb}cBCABBA}$.

1900: 

1901: $38: \mathtt{cbaacBCABBA}$.

1902: 

1903: \noindent The latter word is not empty: we conclude that

1904: $\ww_0$ does not represent~$1$~in~$B_4$.

1905: \end{exam}

1906: 

1907: \subsubsection{Explanation}

1908: 

1909: Clearly, redressing replaces a braid word with an equivalent braid word,

1910: and it is easy to show that, if $\ww$ redresses to a word of the form

1911: $\uu\vv{}\inv$ with $\uu,\vv$ positive, then $\uu$ and $\vv$ are uniquely

1912: determined. Now the problem is that, as Example~\ref{X:Redressing} shows, 

1913: redressing may increase the length of the words, and it is not obvious that

1914: the process terminates. The point is that, if $\ww$ has the form $\ww_1\inv

1915: \ww_2$ where $\ww_1, \ww_2$ are positive words representing simple

1916: braids, then $\ww$ redresses to some word $\uu\vv{}\inv$ where $\uu$

1917: and~$\vv$ are positive and again represent simple braids. Thus, with respect

1918: to an enhanced alphabet containing all positive braid words representing

1919: simple braids, the length does not increase under redressing, and this

1920: leads to the termination result. Actually,

1921: what redressing does is to compute the right lcm in the braid monoid, and

1922: so the properties behing redressing are the Garside theory.

1923: 

1924: \subsubsection{Discussion}

1925: 

1926: The advantage of the redressing method is the simplicity of its

1927: implementation: there is a unique syntactic operation, 

1928: involving length~$2$ subwords of the considered word only.

1929: From that point of view, the method is more easily implemented than

1930: the greedy or symmetric normal form. At a

1931: theoretical level, both methods are essentially equivalent: there exist

1932: positive constants~$C, C'$ such that, for each $\nn$-braid

1933: word~$\ww$, if $\NN_1(\ww)$ (\resp $\NN_2(\ww)$) denotes the

1934: number of braid relations needed to put~$\ww$ into a greedy normal

1935: form (\resp to redress~$\ww$), then we have

1936: $C\NN_1(\ww) \le \NN_2(\ww) \le C'\NN_1(\ww)$. However,

1937: by representing simple braids by permutations and using 

1938: fast sorting algorithms to compute the normal

1939: form as explained in~\cite[Chapter 9]{Eps}, one presumably obtains

1940: a more efficient algorithm.

1941: 

1942: According to Theorem~\ref{T:Redressing}, the redressing method

1943: starting from a braid word~$\ww$ yields a final braid word~$\ww'$

1944: such that $\ww\equiv\e$ is equivalent to $\ww'=\e$. However, in

1945: general, $\ww'\equiv\ww$ fails: because

1946: of the exchange of the negative and positive factors between the two

1947: passes, $\ww'$ is only equivalent to a conjugate of~$\ww$. Now, it is easy to describe a variant of the

1948: redressing method that avoids the median conjugation. Indeed, let 

1949: {\it left redressing}, denoted $\ww\redrl\ww'$, be the symmetric 

1950: counterpart to (right)

1951: redressing consisting in replacing each pattern~$\ss\ii\sss\jj$ with

1952: an equivalent negative--positive word. Formally, we may define

1953: $\ww\redrl\ww'$ to mean

1954: $\widetilde\ww\redrr \widetilde{\ww'}$, where

1955: $\widetilde\ww$ is the word obtained from~$\ww$ by

1956: reversing the order of the letters.

1957: 

1958: \begin{coro}

1959: \cite{Dff}

1960: \label{C:RedressingBis}

1961: A braid word~$\ww$ represents~$1$ in the braid group if and only if, denoting

1962: by~$\uu$ and~$\vv$ the positive words for which $\ww

1963: \redrr \uu\vv{}\inv$ holds, we have $\uu\vv{}\inv \redrl \e$.

1964: \end{coro}

1965: 

1966: So the method is the same as in Corollary~\ref{C:Redressing},

1967: with the only difference that we do not change the word

1968: obtained at the end the first redressing pass, but instead continue

1969: with left redressing. The criterion remains the same, namely that the

1970: initial word~$\ww$ is trivial if and only if the final word~$\ww''$ is

1971: empty. The advantage of this variant is that $\ww''$ is equivalent

1972: to~$\ww$, and, moreover, it gives a fractionary decomposition

1973: of~$\cl\ww$ which is geodesic, \ie, has minimal length among all

1974: fractionary expressions of~$\cl\ww$. This implies that

1975: the negative and the positive parts of~$\ww''$ must be equivalent to

1976: the two components of the symmetric normal form

1977: of~$\cl\ww$, although they need not be in normal form in general.

1978: Assuming that an algorithm for computing the normal form

1979: of a positive braid is available, one obtains in this way an alternative way 

1980: for computing the symmetric normal form of an arbitrary braid that is

1981: more simply implemented than the method of Proposition~\ref{P:AlgoSymmetric}: 

1982: perform double redressing, then put the numerator and the denominator 

1983: in normal form. 

1984: 

1985: \begin{exam}

1986: Starting with~$\mathtt{aBabacABABAbbCB}$ once more, the first

1987: $11$~steps of the redressing algorithm are as in

1988: Example~\ref{X:Redressing}, but, then, we appeal to left redressing,

1989: and the sequel is different. We find:

1990: 

1991: $0: \mathtt{a\underline{Ba}bacABABAbbCB}$

1992: 

1993: \ldots

1994: 

1995: $11: \mathtt{aabc\underline{bA}BBACB}$

1996: 

1997: $12: \mathtt{aab\underline{cA}BabBBACB}$

1998: 

1999: $13: \mathtt{aa\underline{bA}cBabBBACB}$

2000: 

2001: $14: \mathtt{a\underline{aA}BabcBabBBACB}$

2002: 

2003: \ldots

2004: 

2005: $39: \mathtt{BACBBAcbaac\underline{bB}}$.

2006: 

2007: $40: \mathtt{BACBBAcbaac}$.

2008: 

2009: \noindent Once again, the latter word is not empty, and we

2010: conclude that $\ww_0$ does not represent~$1$ in~$B_4$. In

2011: addition, we obtain that $\mathtt{BACBBAcbaac}$ is a shortest

2012: expression of~$\cl{\ww_0}$ as a negative--positive fraction, which is

2013: coherent with Example~\ref{X:Symmetric}, where it was shows that

2014: the symmetric normal form of~$\cl{\ww_0}$ is $(\mathtt{ab, bacb},

2015: \mathtt{bcba, a})$: indeed,

2016: $(\mathtt{ab, bacb})$ is the normal form of~$\mathtt{abbcab}$,

2017: and $(\mathtt{bcba, a})$ is the normal form of~$\mathtt{cbaac}$.

2018: \end{exam}

2019: 

2020: As a final remark, let us observe that, because redressing is efficient

2021: at computing lcm's and complements in the braid monoid, it also

2022: provides an easy way to compute gcd's and, from there, normal forms.

2023: Though not as efficient as those based on quick sorting, the algorithms

2024: based on word redressing are more easily implemented than the 

2025: latter, and they are convenient for small and medium size braid words.

2026: 

2027: \subsection{Handle reduction}

2028: \label{S:Handle}

2029: 

2030: Handle reduction is another syntactic braid word transformation

2031: which, like redressing, consists in iterating some basic word

2032: transformation and concluding that the initial braid word

2033: represents~$1$ if and only if the final word is empty. 

2034: The basic transformation step is more complicated that the one

2035: involved in redressing but the number of steps is much lower, and

2036: the method turns out to be extremaly efficient in practice. The

2037: underlying structure behind handle reduction is a linear

2038: ordering of braids, which pilots the reduction process and

2039: heuristically explains its efficiency.

2040: 

2041: \subsubsection{Description}

2042: 

2043: Handle reduction is an extension of free reduction. The latter 

2044: consists in iteratively deleting patterns

2045: of the form $x x\inv$ or $x\inv x$. Handle reduction involves not

2046: only patterns of the form~$\ss i \sss i$ or $\sss i \ss i$, but also

2047: more general patterns of the form $\ss i \dots

2048: \sss i$ or $\sss i \dots \ss i$ with intermediate letters between the

2049: letters~$\ss i$ and~$\sss i$.

2050: 

2051: \begin{defi}

2052: $(i)$ A {\it $\ss i$-handle} is a braid word of the form

2053: \begin{equation}\label{E:Handle}

2054: w = \s_i^e \, w_0 \, \s_{i+1}^d \, w_1 \, \s_{i+1}^d

2055: \dots

2056: \s_{i+1}^d \, w_m \, \s_i^{-e},

2057: \end{equation}

2058: with $e, d = \pm 1$, $m \ge 0$, and $w_0, ..., w_m$

2059: containing no~$\ssss j$ with $j \le i+1$. 

2060: Then the {\it  reduct} of~$w$ is defined to be

2061: \begin{equation}

2062: w' = w_0 \, \s_{i+1}^{-e}\s_i^d\s_{i+1}^e \, w_1 \,

2063: \s_{i+1}^{-e}\s_i^d\s_{i+1}^e \dots

2064: \s_{i+1}^{-e}\s_i^d\s_{i+1}^e \, w_m,

2065: \end{equation}

2066: \ie, we delete the initial and final letters~$\ssss i$, and 

2067: we replace each letter~$\ssss{i+1}$ with

2068: $\s_{i+1}^{-e}\ssss i\s_{i+1}^e$. 

2069: 

2070: $(ii)$ We say that a braid word~$\ww$ is {\it reduced} if it

2071: contains no $\ss\ii$-handle, where $\ss\ii$ is the generator with

2072: minimal index occurring in~$\ww$.

2073: \end{defi}

2074: 

2075: \begin{figure}[htb]

2076: \setlength{\unitlength}{1mm}

2077: \begin{picture}(125,20)

2078: \put(0,0){\includegraphics[scale=0.9]{Handle.eps}}

2079: \end{picture}

2080: \caption{\smaller\sf Reduction of a handle, here a $\ss1$-handle: 

2081: the dashed strand has the shape of a handle, and reduction consists

2082: in pushing that strand so that it skirts above the next crossings

2083: instead of below.}

2084: \label{F:Handle}

2085: \end{figure}

2086: 

2087: A braid word of the form $\ss i \sss i$ or $\sss i \ss i$ is a handle,

2088: and reducing it means deleting it, so handle reduction generalizes

2089: free reduction.  As illustrated in Figure~\ref{F:Handle}, reducing

2090: a handle yields an equivalent braid word. So, as in the case of

2091: free reduction, if there is a reduction sequence from a braid

2092: word~$w$ to the empty word, \ie, a sequence $w = w_0, 

2093: ..., w_N = \e$ such that, for each~$k$, the

2094: word~$w_{k+1}$ is obtained from~$w_k$ by replacing

2095: some handle of~$w_k$ by its reduct, then $w$ represents~$1$

2096:  in the braid group. The

2097: point is that the converse is also true.

2098: 

2099: \begin{theo}

2100: \cite{Dfo}

2101: \label{P:HRed}

2102: For every braid word~$\ww$, every sequence of handle

2103: reductions from~$\ww$ leads in finitely many steps to a reduced

2104: word~$\ww'$. Moreover, a reduced word~$\ww'$ represents~$1$ if

2105: and only if it is empty.

2106: \end{theo}

2107: 

2108: We obtain a new solution to the braid isotopy problem:

2109: 

2110: \begin{coro}

2111: \label{C:Handle}

2112: A braid word~$\ww$ represents~$1$ in the braid group if and only

2113: if any sequence of handle reductions starting from~$\ww$

2114: terminates with the empty word.

2115: \end{coro}

2116: 

2117: \begin{exam}

2118: \label{X:Handle}

2119: Consider $\ww_0 = \mathtt{aBabacABABAbbCB}$ again. Choosing to

2120: reduce the leftmost handle at each step and underlying it, we

2121: succesively obtain:

2122: 

2123: $0: \mathtt{aBab\underline{acA}BABAbbCB}$

2124: 

2125: $1: \mathtt{aBa\underline{bcB}ABAbbCB}$

2126: 

2127: $2: \mathtt{aB\underline{aCbcA}BAbbCB}$

2128: 

2129: $3: \mathtt{aBCBa\underline{bcB}AbbCB}$

2130: 

2131: $4: \mathtt{aBCB\underline{aCbcA}bbCB}$

2132: 

2133: $5: \mathtt{aBCBCBabcbbCB}$.

2134: 

2135: \noindent The latter word contains no $\ss1$-handle, so it is

2136: reduced, and it is not empty, so we conclude that $\ww_0$ does not

2137: represent~$1$ in~$B_4$.

2138: \end{exam}

2139: 

2140: \subsubsection{Explanation}

2141: 

2142: Two different structures lie behind handle reduction, namely

2143: Garside's theory that was already involved on the previous solutions,

2144: and, in addition, some linear order that is compatible with left

2145: multiplication on~$B_\nn$. It can be seen that each handle reduction

2146: is essentially a composition of redressing steps, but the main

2147: difference with the algorithms of Sections~\ref{S:Normal}

2148: and~\ref{S:Redressing} is that, here, we do not perform all

2149: redressing steps systematically, but only some of them

2150: according to a general strategy provided by the underlying braid

2151: order. This should make it natural why the handle reduction method

2152: is, in practice, much more efficient than the redressing method and

2153: than the greedy and symmetric normal form methods ($5$

2154: \vs $40$ steps in our example).

2155: 

2156: \subsubsection{Discussion}

2157: 

2158: A braid word may contain many handles, so building an actual

2159: algorithm requires to fix a strategy prescribing in which

2160: order the handles will be reduced. In Example~\ref{X:Handle},

2161: we chose to reduce the leftmost handle, but more

2162: efficient strategies exist. As can be expected, the most

2163: efficient ones use a divide-and-conquer trick. Although the only

2164: upper bound for space and time complexity proved so far is exponential, 

2165: handle reduction is extremely efficient in practice, as show the statistics

2166: of~\cite{Dgw}. Also, reduction being a local

2167: procedure, the amount of memory needed to implement it is what is

2168: needed to just store the braid under reduction. So, using arbitrary 

2169: words together with handle reduction instead of normal words could

2170: be specially interesting when the computing resources are limited. 

2171: 

2172: It was mentioned above that the symmetric normal form, as well as the

2173: redressing method, yield fractionary expressions of minimal length.

2174: Such fractionary expressions need not be expressions of

2175: minimal length: there may be shorter expressions that are not

2176: fractions, \ie, in which all negative letters are not gathered in one

2177: block and all positive letters in another block. The general problem

2178: of finding the shortest expression of a braid is difficult: its $\Bi$

2179: version is known to be $NP$-complete~\cite{PaR}. Although no actual

2180: result is proved, it has been observed that handle reduction is good at

2181: providing short expressions. Actually, the definition of handle

2182: reduction is not symmetric and the left side plays a distinguished

2183: r\^ole. One can compensate this lack of symmetry by defining {\it

2184: iterated handle reduction} as follows: starting with~$\ww$, we

2185: reduce~$\ww$ to~$\ww'$, then flip~$\ww'$ using~$\flip_\nn$,

2186: reduce it, and flip the result. Equivalently, after one handle

2187: reduction, one performs the symmetric operation in which the right

2188: side is distinguished. By doing so, and possibly iterating the process,

2189: one empirically obtains short expressions. It is conjectured by the

2190: author, as well as by A.\,Miasnikov and A.\,Ushakov, that there

2191: might exist a constant~$C$ such that applying the previous process

2192: to any braid word~$\ww$ leads to a final word of length at

2193: most~$C\ell_{\mathrm{min}}(\cl\ww)$, where, for~$\xx$ a braid, 

2194: $\ell_{\mathrm{min}}(\xx)$ denotes the length of the shortest

2195: word~$\xx$.

2196: 

2197: \begin{exam}

2198: \label{X:HandleBis}

2199: Starting once more with $\ww_0 = \mathtt{aBabacABABAbbCB}$,

2200: iterated handle reduction from~$\ww_0$ leads in 3~steps to the

2201: word $\mathtt{acBCCBa}$, which happens to be a geodesic

2202: representative of the braid~$\cl{\ww_0}$.

2203: \end{exam}

2204: 

2205: \subsection{Dynnikov coordinates}

2206: \label{S:Dynnikov}

2207: 

2208: We conclude with still another solution to the braid isotopy

2209: problem, namely the one provided by the so-called Dynnikov

2210: coordinates. This solution relies on a completely different approach

2211: steming from geometry and deep results by L.\,Mosher about

2212: the existence of an automatic structure for all mapping class

2213: groups~\cite{Mos}.

2214: 

2215: \subsubsection{Description}

2216: 

2217: The principle consists in associating with every $\nn$-braid word a

2218: sequence of $2\nn$~integers, that can be thought of as

2219: coordinates for the braid. This coordinization is faithful in that

2220: two braid words receive the same coordinates if and only if they

2221: represent the same braid.

2222: 

2223: \begin{defi}

2224: $(i)$ For $\xx$ in~Ê$\Int$, write $\Pos\xx$ for~$\max(0,\xx)$,

2225: and $\Neg\xx$ for~$\min(\xx,0)$. Let~$\FDyn,

2226: \GDyn: \Int^4 \to \Int^4$ be defined by $\FDyn =

2227: (\FDyn_1, ..., \FDyn_4)$, $\GDyn=(\GDyn_1, ...,

2228: \GDyn_4)$ with

2229: \begin{equation}

2230: \label{E:Dynnikov}

2231: \begin{cases}

2232: \FDyn_1(\xx_1, \yy_1, \xx_2, \yy_2)

2233: &:=\quad \xx_1 + \Pos{\yy_1} + \Pos{(\Pos{\yy_2} - \zz_1)},\\

2234: \FDyn_2(\xx_1, \yy_1, \xx_2, \yy_2)

2235: &:=\quad \yy_2 - \Pos\zz_1,\\

2236: \FDyn_3(\xx_1, \yy_1, \xx_2, \yy_2)

2237: &:=\quad  \xx_2 + \Neg{\yy_2} + \Neg{(\Neg{\yy_1} + \zz_1)},\\

2238: \FDyn_4(\xx_1, \yy_1, \xx_2, \yy_2)

2239: &:=\quad  \yy_1 + \Pos\zz_1,\\

2240: \GDyn_1(\xx_1, \yy_1, \xx_2, \yy_2)

2241: &:=\quad \xx_1 - \Pos{\yy_1} -  \Pos{(\Pos{\yy_2} + \zz_2)},\\

2242: \GDyn_2(\xx_1, \yy_1, \xx_2, \yy_2)

2243: &:=\quad  \yy_2 + \Neg\zz_2,\\

2244: \GDyn_3(\xx_1, \yy_1, \xx_2, \yy_2)

2245: &:=\quad  \xx_2 - \Neg{\yy_2} - \Neg{(\Neg{\yy_1} -\zz_2)},\\

2246: \GDyn_4(\xx_1, \yy_1, \xx_2, \yy_2)

2247: &:=\quad  \yy_1 - \Neg\zz_2,

2248: \end{cases}

2249: \end{equation}

2250: where we put 

2251: $\zz_1 := \xx_1 - \Neg{\yy_1} - \xx_2 + \Pos{\yy_2}$

2252: and $\zz_2 := \xx_1 + \Neg{\yy_1} - \xx_2 - \Pos{\yy_2}$. 

2253: 

2254: $(ii)$ For $(\aa_1, \bb_1, ..., \aa_\nn, \bb_\nn)$

2255: in~$\Int^{2n}$, put $(\aa_1, \bb_1, ..., \aa_\nn, \bb_\nn) \act

2256: \s_\ii^\ee =  (\aa'_1, \bb'_1, ..., \aa'_\nn, \bb'_\nn)$ with

2257: $\aa'_\kk = \aa_\kk$ and $\bb'_\kk=\bb_\kk$

2258: for $\kk\not=\ii,\ii+1$, and

2259: \begin{equation*}

2260: (\aa'_\ii, \bb'_\ii, \aa'_{\ii+1}, \bb'_{\ii+1}) =

2261: \begin{cases}

2262: \FDyn(\aa_\ii, \bb_\ii, \aa_{\ii+1}, \bb_{\ii+1})

2263: &\text{for $\ee=+1$},\\

2264: \GDyn(\aa_\ii, \bb_\ii, \aa_{\ii+1}, \bb_{\ii+1})

2265: &\text{for $\ee=-1$}.\\

2266: \end{cases}

2267: \end{equation*}

2268: Then, for $\ww$ an $n$-braid word, we recursively define

2269: \begin{equation*}

2270: (\aa_1, \bb_1, ..., \aa_\nn, \bb_\nn) \act \ww =

2271: \begin{cases}

2272: (\aa_1, \bb_1, ..., \aa_\nn, \bb_\nn)

2273: &\text{for $\ww= \e$},\\

2274: ((\aa_1, \bb_1, ..., \aa_\nn, \bb_\nn) \act \ww') \act \s_\ii^\ee

2275: &\text{for $\ww = \ww'\s_\ii^\ee$}.

2276: \end{cases}

2277: \end{equation*}

2278: The {\it Dynnikov coordinates} of~$\ww$

2279: are defined to be the sequence $(0,1,0,1,..., 0,1) \act\ww$.

2280: \end{defi}

2281: 

2282: \begin{theo}

2283: \cite[Propositions~8.5.3 and~8.5.4]{Dgr}

2284: \label{T:Dynnikov}

2285: The Dynnikov coordinates of an $n$-braid word~$\ww$ characterize

2286: the braid~$\cl\ww$ represented by~$\ww$: the coordinates

2287: of~$\ww$ and~$\ww'$ are equal if and only if $\cl\ww=\cl{\ww'}$ holds.

2288: \end{theo}

2289: 

2290: We deduce still another solution to the braid isotopy problem:

2291: 

2292: \begin{coro}

2293: An $n$-strand braid word represents~$1$ in~$B_\nn$ if

2294: and only if its Dynnikov coordinates are~$(0,1,0,1,..., 0,1)$.

2295: \end{coro}

2296: 

2297: \begin{exam}

2298: \label{X:Dynnikov}

2299: Consider $\ww_0 = \mathtt{aBabacABABAbbCB}$ once more. By

2300: applying the formulae of~\eqref{E:Dynnikov} inductively, we

2301: compute the Dynnikov coordinates of the successive prefixes

2302: of~$\ww_0$:

2303: 

2304: $0: \e \gives (0, 1, 0, 1, 0, 1, 0, 1)$

2305: 

2306: $1: \mathtt{a} \gives (1, 0, 0, 2, 0, 1, 0, 1)$

2307: 

2308: $2: \mathtt{aB} \gives (1, 0, -2, 0, 0, 3, 0, 1)$

2309: 

2310: $3: \mathtt{aBa} \gives (1, -3, -2, 3, 0, 3, 0, 1)$

2311: 

2312: $4: \mathtt{aBab} \gives (1, -3, 3, 2, 0, 4, 0, 1)$

2313: 

2314: $5: \mathtt{aBaba} \gives (1, -1, 3, 0, 0, 4, 0, 1)$

2315: 

2316: \ldots

2317: 

2318: $14: \mathtt{aBabacABABAbbC} \gives (1, -7, 5, -1, -7, 4, 0, 8)$

2319: 

2320: $15: \mathtt{aBabacABABAbbCB} \gives (1, -7, -6, 4, 1, -1, 0, 8)$.

2321: 

2322: \noindent The latter coordinates are not $(0,1,0,1,0,1,0,1)$, so we conclude that $\ww_0$ does not

2323: represent~$1$ in~$B_4$.

2324: \end{exam}

2325: 

2326: \subsubsection{Explanation}

2327: 

2328: At first, the formulae~\eqref{E:Dynnikov} seem quite mysterious. Actually,

2329: there is no miracle here, but a very clever use of the simple formula

2330: \begin{equation}

2331: \label{E:Flip}

2332: \xx+\xx'=\max(\xx_1+\xx_3,\xx_2 + \xx_4)

2333: \end{equation}

2334: that compares the number of intersections of a family of curves with

2335: two triangulations obtained one from the other by switching one

2336: diagonal in a quadrilateral (flip transformation). 

2337: 

2338: The framework consists in considering an $n$-braid as the isotopy

2339: class of a homeomorphism of a disk with $n$~punctures: then the

2340: generator~$\ss\ii$ corresponds to the homeomorphism that

2341: exchanges the $\ii$th and $(\ii+1)$st punctures by a half-turn.

2342: Dynnikov's idea is to let the braid act on a family of curves in the

2343: punctured disk, and to count their intersections with a fixed

2344: triangulation, or, equivalently, to let the braid act on

2345: the triangulation and count its intersections with a fixed family of

2346: curves. The action of~$\ss\ii$ on the chosen triangulation can then

2347: be decomposed into the composition of four flips, and

2348: applying~\eqref{E:Flip} repeatedly leads to the mysterious

2349: formulae~\eqref{E:Dynnikov}. 

2350: 

2351: \subsubsection{Discussion}

2352: 

2353: The remarkable point about the Dynnikov coordinates is that they

2354: involve the semiring $(\Int, \max, +, 0)$, which explains their

2355: efficiency. Multiplying by one generator~$\ss\ii$ can only increase

2356: the size of the coordinates by one unit, whereas similar formulae in

2357: the ring $(\Int, +, \times, 1)$ would double

2358: the size in the worst case. It follows that the solution to the braid word

2359: problem given by Theorem~\ref{T:Dynnikov} has a linear space

2360: complexity, and a quadratic time complexity.

2361: 

2362: It is not so easy to compare the solution based on the Dynnikov

2363: coordinates with the other solutions to the braid isotopy problem,

2364: because its practical efficiency much depends on the way large

2365: integer arithmetic is implemented---large integers do appear

2366: when long braid words are considered, typically length~$O(\ell)$ binary

2367: integers for a length~$\ell$ braid word representing a pseudo-Anosov 

2368: braid. However, the only

2369: arithmetic operations involved are addition and maximum, and

2370: both can be implemented very efficiently and easily. No statistical

2371: study has been completed so far, and it would be desirable to

2372: compare Dynnikov's method with handle reduction. The only weak

2373: point of the former is that, at the moment, it is purely

2374: incremental: the only available formulae correspond to multiplying by

2375: one single letter~$\ss\ii$ or~$\sss\ii$, so, in particular, no

2376: divide-and-conquer variant exists.

2377: 

2378: \begin{thebibliography}{99}

2379: 

2380: \def\Reff#1; #2; #3; #4; #5; #6; #7\par{%

2381: \bibitem{#1} #2, {\it #3}, #4 {\bf #5} (#6) #7}

2382: 

2383: \def\Ref#1; #2; #3; #4\par{%

2384: \bibitem{#1} #2, {\it #3}, #4}

2385: 

2386: \Reff Adj; S.I.\,Adyan; Fragments of the word Delta in a

2387: braid group; Mat. Zam. Acad. Sci. SSSR; 36-1; 1984;

2388: 25--34; translated Math. Notes of the Acad. Sci. USSR;

2389: 36-1 (1984) 505--510.

2390: 

2391: \Reff AAG; I.~Anshel, M.~Anshel, \& D.~Goldfeld; An

2392: algebraic method for public-key cryptography; Math.

2393: Research Letters; 6; 1999; 287--291.

2394: 

2395: \Reff Art; E.~Artin; Theory of Braids; Ann. of Math.; 48;

2396: 1947; 101--126.

2397: 

2398: \Ref Bes; D.\,Bessis; Garside categories, periodic loops and  cyclic

2399: sets; Preprint; math.GR/0610778.

2400: 

2401: \Ref Bir; J.~Birman; Braids, Links, and Mapping  Class

2402: Groups; Annals of Math. Studies {\bf 82} Princeton Univ. 

2403: Press  (1975). 

2404:    

2405: \Reff Bur; S. Burckel; The wellordering on

2406: positive braids; J. Pure Appl. Algebra; 120-1; 1997; 1--17.

2407: 

2408: \Reff Cha; R.\,Charney; Artin groups of finite type are

2409: biautomatic; Math. Ann.; 292-4; 1992; 671--683.

2410: 

2411: \Reff Dff; P.\,Dehornoy; Groups with a complemented

2412: presentation; J. Pure Appl. Algebra; 116; 1997; 115--137.

2413: 

2414: \Reff Dfo; P.\,Dehornoy; A fast method for comparing

2415: braids; Advances in Math.; 125; 1997; 200--235.

2416: 

2417: \Reff Dgk; P.\,Dehornoy; Groupes de Garside; 

2418: Ann. Scient. Ec. Norm. Sup.; 35; 2002; 267--306.

2419: 

2420: \Reff Dgo; P.\,Dehornoy; Thin groups of fractions;

2421: Contemp. Math.; 296; 2002; 96--129.

2422: 

2423: \Reff Dgp; P.\,Dehornoy; Complete positive group 

2424: presentations; J. of Algebra; 268; 2003; 156--197.

2425: 

2426: \Reff Dgw; P.\,Dehornoy; Braid-based cryptography;

2427: Contemp. Math.; 360; 2004; 5--33.

2428: 

2429: \Ref Dho; P.\,Dehornoy; Alternating normal forms for braids

2430: and locally Garside monoids monoids; Preprint; math.GR/0702592.

2431: 

2432: \Ref Dgr; P. Dehornoy, I. Dynnikov, D. Rolfsen, B. Wiest;

2433: Why are braids orderable?; Panoramas \& Synth\`eses

2434: vol. 14, Soc. Math. France (2002).

2435: 

2436: \Reff Dfx; P.\,Dehornoy \& L. Paris; Gaussian

2437: groups and Garside groups, two generalizations of Artin

2438: groups; Proc. London Math. Soc.; 79-3; 1999; 569--604.

2439: 

2440: \Ref DiM; F.\,Digne \& J.\,Michel; Garside and locallly Garside 

2441: categories; Preprint; math.GR/0612652.

2442: 

2443: \Reff ElM; E. A.\,Elrifai \& H.R.\,Morton; Algorithms for

2444: positive braids; Quart. J. Math. Oxford; 45-2; 1994;

2445: 479--497.

2446: 

2447: \Ref Eps; D.\,Epstein, J.\,Cannon, D.\,Holt, S.\,Levy,

2448: M.\,Paterson \& W.\,Thurston; Word Processing in Groups;

2449: Jones \& Bartlett Publ. (1992).

2450: 

2451: \Reff Gar; F. A.\,Garside; The braid group and

2452: other groups; Quart. J. Math. Oxford; 20-78; 1969;

2453: 235--254.

2454: 

2455: \Ref KaT; C.\,Kassel \& V.\,Turaev; Braid groups; Springer (2007).

2456: 

2457: \Ref KLC; K.H. Ko, S.J. Lee, J.H. Cheon, J.W. Han, 

2458: J.S. Kang, \& C. Park; New public-key cryptosystem

2459: using braid groups; Crypto 2000; Springer Lect. Notes

2460: in Comput. Sci., 1880 (2000) 166--184.

2461: 

2462: \Ref Kra; D.\,Krammer; A class of Garside groupoid structures on 

2463: the pure braid group; Trans. Amer. Math. Soc., to appear.

2464: 

2465: \Reff Mos; L. Mosher; Mapping class groups are

2466: automatic; Ann. of Math.; 142; 1995; 303--384.

2467: 

2468: \Reff PaR; M.S.\,Paterson \& A.A.\,Razborov; The set of minimal 

2469: braids is co-NP-complete; J. Algorithms; 12--3; 1991; 393--408.

2470:    

2471: \Ref Thu; W.\,Thurston; Finite state algorithms for the

2472: braid group; Circulated notes (1988).

2473: 

2474: \end{thebibliography}

2475: 

2476: \end{document}

2477: 

2478: