1: \documentclass[11pt]{amsart}

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

3: %\usepackage{epstopdf}

4: 

5: \textwidth=14.5cm

6: \textheight=21cm

7: 

8: % for 100% printing

9: %\hoffset-1.2cm

10: %\voffset+0.5cm

11: 

12: %for 110% printing

13: \hoffset-1.0cm

14: \voffset-0.6cm

15: 

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

17: 

18: \theoremstyle{plain}

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

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

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

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

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

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

25: 

26: \theoremstyle{definition}

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

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

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

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

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

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

33: 

34: \numberwithin{equation}{section}

35: 

36: % bibliography

37: 

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

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

40: 

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

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

43: 

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

45: 

46: \catcode`\¨=\active\def¨{~} 

47: 

48: \renewcommand{\aa}[3]{a_{#1,#2,#3}}%_Entry_of_matrix

49: \newcommand{\aab}[2]{\overline{a}_{#1,#2}}%_Entry_of_matrix

50: \newcommand{\aas}[3]{\widehat a_{#1, #2,#3}}

51: 

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

53: height5.5pt #1}}

54: \newcommand{\bb}[2]{b_{#1,#2}}

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

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

57: \newcommand{\bs}{x}%_A_simple

58: \newcommand{\bt}{y}%_Another_simple

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

60: \newcommand{\by}{y}

61: 

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

63: \newcommand{\charpol}[1]{P_{#1}(x)}

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

65: \newcommand{\comp}[2]{[#2]_{#1}}

66: \newcommand{\concat}{\mathbin{{}^{\scriptscriptstyle\frown}}}

67: 

68: \newcommand{\D}{\Delta}

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

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

71: \renewcommand{\div}{\prec}

72: \newcommand{\Div}{\mathrm{Div}}

73: \newcommand{\dive}{\preccurlyeq}

74: \newcommand{\DivL}{D_{\!L}}

75: \newcommand{\DivR}{D_{\!R}}

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

77: \newcommand{\pz}{z}

78: 

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

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

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

82: 

83: \newcommand{\ff}{f}

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

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

86: 

87: \newcommand{\g}{\gamma}

88: \let\ge=\geqslant

89: \renewcommand{\gg}{g}

90: 

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

92: \newcommand{\II}{I}

93: \let\ince=\subseteq

94: \newcommand{\ind}{i}

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

96: \newcommand{\indd}{j}

97: \newcommand{\Int}[1]{[\![1, #1]\!]}

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

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

100: \newcommand{\ik}{i}

101: \newcommand{\ip}{p}

102: \newcommand{\iq}{q}

103: \newcommand{\ir}{k}

104: \newcommand{\is}{\ell}

105: 

106: \newcommand{\JJ}{J}

107: 

108: \newcommand{\kk}{k}

109: \newcommand{\KK}{K}

110: 

111: \renewcommand{\l}{\lambda}

112: \let\le=\leqslant

113: \renewcommand{\ll}{\ell}

114: 

115: \newcommand{\m}{\mu}

116: \newcommand{\Mat}[1]{M_{#1}}%_Transition_matrix

117: \newcommand{\Matt}[1]{M'_{#1}}

118: \newcommand{\Mattt}[1]{\overline M_{#1}}

119: \newcommand{\mm}{m}

120: 

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

122: \newcommand{\nn}{n}

123: \newcommand{\NN}{N}

124: \newcommand{\nno}{{\nn-1}}

125: \newcommand{\NNN}{N}

126: 

127: \newcommand{\op}{\cdot}

128: 

129: \renewcommand{\part}[2]{\{#2\}_{#1}}

130: \newcommand{\perm}[1]{\pi(#1)}

131: \newcommand{\pp}{p}

132: \newcommand{\ppp}{p}

133: \newcommand{\pppp}{p}

134: \newcommand{\Pw}{\mathfrak{P}}

135: 

136: \newcommand{\qq}{q}

137: \newcommand{\QQ}{\mathbb{Q}}

138: 

139: \newcommand{\resp}{{\it resp.{¨}}}

140: \newcommand{\rr}{r}

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

142: \newcommand{\rrr}{r}

143: 

144: \let\s=\sigma

145: \newcommand{\set}[1]{\widetilde{#1}}

146: \newcommand{\sh}{\mathrm{sh}}

147: \renewcommand{\sp}[1]{\tau_{#1}}

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

149: \renewcommand{\SS}{S}

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

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

152: 

153: \newcommand{\uu}{u}

154: 

155: \newcommand{\var}{u}

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

157: \newcommand{\vv}{v}

158: 

159: \newcommand{\ww}{w}

160: 

161: \newcommand{\xx}{x}

162: \newcommand{\XX}{X}

163: 

164: \newcommand{\yy}{y}

165: 

166: \newcommand{\zz}{z}

167: 

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

169: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

170: \begin{document}

171: 

172: \author{Patrick DEHORNOY}

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

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

175: 14032~Caen, France}

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

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

178: 

179: \title{Combinatorics of normal sequences of

180: braids}

181: 

182: \keywords{braid group; normal form; fundamental

183: braid; counting; generating function}

184: 

185: \subjclass{20F36, 05A05}

186: 

187: \begin{abstract}

188: Many natural counting problems arise in

189: connection with the normal form of braids---and

190: seem to have not been much considered so far. Here

191: we solve some of them by analysing the normality

192: condition in terms of the associated

193: permutations, their descents and the

194: corresponding partitions. A number of different

195: induction schemes appear in that framework.

196: \end{abstract}

197: 

198: \maketitle

199: 

200: Ubiquitous and connected with a number of domains,

201: Artin's braid groups have received much attention

202: in the recent years. However, not so many works

203: are devoted to a purely combinatorial study of

204: braids, presumably because counting arguments did

205: not prove so far to be much helpful for

206: investigating braids. Nevertheless, although

207: braid groups are infinite, they admit several

208: filtrations leading to finite sets and,

209: therefore, to natural enumeration problems.

210: 

211: For each presentation of braid groups, (at least) two

212: natural counting poblems arise, namely, on the one hand,

213: counting how many braids admit an expression of

214:  a given length, in particular evaluating the associated

215: growth rate---we shall refer to this as Question¨1 in the

216: sequel---and, on the other hand, counting, for a given

217: braid, how many words represent that specific braid, a

218: relevant question when the number is finite,

219: typically when we discard the inverses of the

220: generators and only consider positive expressions,

221: \ie, when we restrict to some submonoid of~$B_\nn$---we

222: shall refer to that as Question¨2. 

223: 

224: In the case of the Artin generators¨$\ss i$, both types of

225: questions have been addressed, and at least partially

226: solved: Question¨1, actually not for¨$B_\nn$ but for the

227: submonoid¨$B_\nn^+$ of¨$B_\nn$ generated by the¨$\ss i$'s, was

228: investigated in~\cite{Xua}, and completely solved

229: in~\cite{Bro}. As for Question~2, it is natural in this context to

230: address it for the particular elements¨$\D_n^d$, where

231: $\D_n$ is Garside's fundamental braid \cite{Gar}. It was

232: investigated and solved for~$n = 3$ in~\cite{CrH}.

233: 

234: In this paper, we address Question~1 for another natural 

235: generating set, namely the so-called simple braids, also called

236: the Garside generators below~\cite{Gar}. 

237: These generators, which are the divisors  of~$\D_n$ in 

238: the monoid~$B_\nn^+$, are in one-to-one

239: correspondence with permutations of $n$~objects,

240: and they give rise to a remarkable unique

241: decomposition for each braid, usually called its normal

242: form \cite{Dlg, Adj, Eps, ElM}. Because of its

243: uniqueness and of its many nice properties,

244: expressed in particular in the existence of a bi-automatic 

245: structure, the normal form

246: of braids is the preferred way of specifying

247: braids in many recent developments, in particular

248: those of algorithmic or cryptographical nature \cite{Geb, KoL}.

249: 

250: In the case of the Garside generators and the associated

251: normal form(s), what we called Question¨2 above is not

252: relevant, as each braid has one unique distinguished

253: decomposition in terms of the generators, and counting the non-normal expressions would appear artificial. But Question¨1, namely counting the

254: number of braids with a normal form of a given length, is

255: quite natural. The question was briefly considered

256: by R.\,Charney in¨\cite{Chb}: she observed that, because

257: normal words can be recognized by a finite state automaton,

258: the number of braids with length¨$\dd$ obeys a linear

259: induction rule, and the associated generating function is

260:  rational. Explicit values are given in the case of

261: $3$¨and $4$¨strand braids.

262: 

263: The aim of this paper is to go further in the investigation

264: of counting problems connected with the normal form of

265: braids. We consider the case of the monoid¨$B_\nn^+$, and

266: mainly study the number of positive $n$¨strand braids

267: with a normal form of length (at most)¨$\dd$, \ie, we

268: consider for Garside generators the problem solved

269: in¨\cite{Bro} in the case of Artin generators. The main

270: difference is that, in the case of

271: Garside generators, the length is no longer an additive

272: parameter: for instance, multiplying two simple braids may

273: result in a simple braid, \ie, multiplying two braids of

274: length¨$1$ may result in a braid of length¨$1$. This makes

275: the current study much more uneasy.

276: 

277: Let $\bb\nn\dd$ denote the number of positive $n$¨strand

278: braids with a normal form of length at most¨$\dd$, \ie, the

279: number of divisors of¨$\DD\nn\dd$ in¨$B_\nn^+$. We establish

280: various results about the numbers¨$\bb\nn\dd$,

281: and about the connected numbers¨$\bbb\nn\dd\bs$ that count,

282: for $\bs$ a simple braid, the positive $n$¨strand braid

283: with a normal form of length at most¨$\dd$ whose

284: $\dd$th factor is precisely¨$\bs$. Two types of results are established, 

285: namely results for fixed braid index¨$n$, and results for fixed degree¨$\dd$.

286: When $n$ is fixed and $\dd$ varies, as was recalled above, 

287: the numbers¨$\bb\nn\dd$ and $\bbb\nn\dd\bs$ obey a linear

288: induction rule associated with a certain $n! \times n!$

289: incidence matrix¨$\Mat\nn$. Here we show that $\Mat\nn$

290: can be replaced with a smaller matrix of size $p(\nn)

291: \times p(\nn)$, where $p(\nn)$ is the number of partitions

292: of¨$\nn$. The result relies on analysing the descents

293: of the permutations associated with simple braids, and it

294: is connected with a classical result by Solomon¨\cite{Sol}.

295: It is then easy to deduce the numerical value of¨$\bb\nn\dd$

296: for small¨$\nn, \dd$, as well as explicit formulas, at least

297: for $n \le 4$. We are also led to several

298: conjectures about the eigenvalues of the matrix¨$\Mat\nn$

299: that seem to have never been considered so far. The most

300: puzzling one claims that the characteristic polynomial

301: of¨$\Mat{\nno}$ divides that of¨$\Mat\nn$. It holds at

302: least for $\nn \le 10$.

303: 

304: When $\dd$ is fixed and $\nn$ varies, quite different

305: induction rules appear. Everything is trivial for $\dd =

306: 1$, and an explicit formula for¨$\bb\nn2$ can be deduced from

307: the results of¨\cite{CSV, CSV2}. It seems difficult to

308: go further in general, but new results (and new

309: induction schemes) appear when we consider the numbers

310: $\bbb\nn\dd{\D_{\nn-\rrr}}$ with $1 \le \rr \le \nn$,

311: typically in the (non-trivial) case $\rr = 1$, and, more

312: generally, when $\rr$ is fixed. In particular, we

313: obtain explicit values for $\bbb\nn3{\D_{\nno}}$,

314: $\bbb\nn3{\D_{\nn-2}}$, and $\bbb\nn4{\D_{\nno}}$.

315: 

316: The specific questions investigated in this paper, in

317: particular that of the value

318: of¨$\bbb\nn\dd{\D_{\nn-\rrr}}$, arose in¨\cite{Dhh}. There

319: exists a distinguished linear ordering of braids¨$<$ relying

320: on the notion of a

321: $\s$-positive braid word¨\cite{Dgr}, and the aim

322: of¨\cite{Dhh} is to develop a new approach to that ordering

323: based on the study of its connection with the Garside

324: structure. It turns out that certain parameters describing the

325: restriction of¨$<$ to positive $n$-braids of degree at

326: most¨$\dd$ can be expressed in terms of the

327: numbers¨$\bbb\nn\dd{\D_\rr}$, an initial

328: motivation for our current study of these numbers. However,

329: we think that the formulas and methods developed in the

330: current paper go beyond the above specific applications. In

331: particular, the great diversity of the induction

332: schemes appearing in connection with various

333: specializations of the general problem is remarkable. At the

334: least, the current study should demonstrate the richness of

335: the combinatorics underlying the normal form of braids.

336: 

337: Still other presentations of the braid groups are known, in

338: particular the one involving the so-called dual

339: monoid¨\cite{Bes, BKL}, which gives rise to an alternative

340: Garside structure, and, therefore, to an alternative

341: normal form analogous to that considered here, where the

342: role of simple braids is played by elements that are in

343: one-to-one correspondence with non-crossing partitions.

344: All questions considered in the current paper could be

345: similarly addressed for the dual structure, and, more

346: generally, for the many presentations of¨$B_\nn$ known to

347: date. Similarly, Artin's braid groups~$B_\nn$ belong to larger

348: families of groups, typically Artin-Tits groups of spherical

349: type and, more generally, Garside groups~\cite{Dfx, Dgk, McC}. 

350: Once again, all

351: questions considered here extend to such frameworks

352: naturally. However, mainly because of the specific

353: applications mentioned above, we find it interesting to

354: consider here the specific framework of braids and

355: permutations, and we leave the extensions for further

356: investigation.

357: 

358: The paper is organized as follows. Section¨\ref{S:Back}

359: sets the framework and the basic definitions. In

360: Section¨\ref{S:Adj} we introduce the incidence

361: matrix¨$\Mat\nn$ that controls the sequences¨$\bb\nn\dd$ for

362: fixed¨$\nn$ and show how to reduce their size

363: from¨$\nn!$ to¨$2^\nno$. In Section¨\ref{S:Part}, we

364: show how to further reduce the size to¨$p(\nn)$, and solve

365: the induction for small values of¨$\nn$. Finally, in

366: Section¨\ref{S:Deg}, we turn to the cases when the

367: degree is fixed and the braid index varies. 

368: 

369: \section*{Acknowledgment}

370: 

371: The author thanks C.\,Hohlweg, F.\,Hivert and

372: J.C.\,Novelli for interesting discussions about the topics

373: investigated in this paper, in particular Remark¨\ref{R:Hohlweg}.

374: 

375:   \section{Background and preliminary results}

376: \label{S:Back}

377: 

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

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

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

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

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

383: B_\nn = \left<\ss1, \dots, \ss{\nno} \,;\,

384: \begin{array}{cl}

385: \ss i \ss j = \ss j \ss i 

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

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

388: \   \ss i \ss j \ss i  = \ss j \ss i \ss j

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

390: \end{array} \right>.

391: \end{equation}

392: So, $B_1$ is a trivial group¨$\{1\}$, while $B_2$

393: is the free group generated by¨$\ss1$. The elements of¨$B_\nn$

394: are called $\nn$¨strand braids, or simply {\it $\nn$-braids}.

395: We use $B_\infty$ for the group generated by an infinite

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

397: of¨\eqref{E:Present}, \ie, the direct limit of all¨$B_\nn$'s

398: under the inclusion of¨$B_\nn$ into¨$B_{n+1}$.

399: 

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

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

402: \le i < n$. Such a expression is called an $\nn$¨strand {\it

403: braid word}. Two braid words¨$\ww, \ww'$ representing the

404: same braid are said to be {\it equivalent}; the braid represented 

405: by a braid word¨$\ww$ is denoted¨$\cl\ww$. 

406: 

407: It is standard to associate with every $\nn$¨strand braid

408: word¨$\ww$ an $\nn$¨strand braid {\it diagram} by stacking 

409: elementary diagrams associated with the successive letters 

410: according to the rules

411: \begin{center}

412: \begin{picture}(83,13)(0,0)

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

414: \put(-4,3){$\ss i\mapsto$}

415: \put(9.5,9){$1$}

416: \put(19.5,9){$2$}

417: \put(50,9){$i$}

418: \put(58,9){$i{+}1$}

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

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

421: \end{picture}

422: \end{center}

423: \begin{center}

424: \begin{picture}(83,10)(0,0)

425: \put(10,0){\includegraphics{SigmaInv.eps}}

426: \put(-4,3){$\ss i\inv \mapsto$}

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

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

429: \end{picture}

430: \end{center}

431: Then two braid words are

432: equivalent if and only if the diagrams they encode are the

433: projections of ambient isotopic figures in¨$\RR^3$, \ie, one

434: can deform one diagram into the other without allowing the

435: strands to cross or moving the endpoints. 

436: 

437: \subsection{The monoid¨$B_\nn^+$ and the braids¨$\D_\nn$}

438: 

439: Let $B_\nn^+$ be the monoid admitting

440: the presentation¨\eqref{E:Present}. The elements of¨$B_\nn^+$

441: are called {\it positive} $\nn$-braids.

442: 

443: \begin{defi}

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

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

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

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

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

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

450: \end{defi}

451: 

452: As $B_\nn^+$ is not commutative for $\nn \ge 3$, there are

453: the symmetric notions of a right divisor and a left

454: multiple---but we shall mostly use left divisors here.

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

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

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

458: always specify the index¨$\nn$.

459: 

460: Wih respect to left divisibility, $B_\nn^+$ has the structure

461: of a lattice \cite{Gar}: any two positive $\nn$-braids¨$\bx,

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

463: $\gcd(\bx, \by)$, and a least common right multiple. A

464: special role is played by the lcm of the elements¨$\ss1$,

465: \dots, $\ss{\nno}$, traditionally denoted¨$\D_\nn$, which we

466: recall is inductively defined by

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

468: \D_1 = 1, \qquad

469: \D_\nn = \ss1 \ss2 \dots \ss{\nno} \, \D_{\nno}.

470: \end{equation}

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

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

473: that the inner automorphism¨$\flip n$ of¨$B_\nn$ corresponding

474: to conjugation by¨$\D_\nn$ exchanges¨$\ss i$ and¨$\ss{n-i}$

475: for $1 \le i \le \nno$. 

476: 

477: \subsection{The normal form}

478: 

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

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

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

482: {\it simple} in the sequel. Geometrically, simple braids

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

484: braid diagram in which any two strands cross at most once.

485: 

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

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

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

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

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

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

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

493: 

494: \begin{defi}

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

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

497: each¨$\kk$, one has $\sx_\kk =

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

499: \end{defi}

500: 

501: Clearly, each positive braid admits a unique normal

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

503: normal expression as unbounded on the right by

504: completing it with as many trivial factors¨$1$ we need. In

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

506: normal form) of¨$\bx$ for each positive braid¨$\bx$. We

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

508: the largest integer such that the $\dd$th factor of¨$\bx$ is

509: not¨$1$. It is well known that the positive $\nn$-braids of

510: degree at most¨$\dd$ coincide with the (left or oright)

511: divisors of¨$\DD\nn\dd$. 

512: 

513: The only properties of the normal form we shall use here

514: are as follows:

515: 

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

517: Assume that $(\sx_1, \dots, \sx_\dd)$ is a sequence

518: of simple

519: $\nn$-braids. Then the following are equivalent:

520: 

521: $(i)$ The sequence $(\sx_1, \dots, \sx_r)$ is normal;

522: 

523: $(ii)$ For $1 \le \kk < \dd$, the sequence

524: $(\sx_\kk, \sx_{\kk+1})$ is normal;

525: 

526: $(iii)$ For $1 \le \kk < \dd$, every¨$\ss i$

527: dividing¨$\sx_{\kk+1}$ on the left divides¨$\sx_\kk$

528: on the right.

529: \end{lemm}

530: 

531: \begin{defi}

532: For¨$\sx$ a simple $\nn$-braid, we define

533: $\DivL(\sx)$ (\resp $\DivR(\sx)$) to be the set of

534: all¨$i$'s such that $\ss i$ is a left (\resp

535: right) divisor of¨$\sx$.

536: \end{defi}

537: 

538: The example of¨$\ss2$ and $\ss2\ss1\ss3\ss2$,

539: for which both¨$\DivL$ and¨$\DivR$ is¨$\{2\}$,

540: shows that these sets do not determine a

541: simple braid. However, as far as normal sequences are

542: concerned, they contain all needed

543: information, as Lemma¨\ref{L:Normal}

544: can be restated as:

545: 

546: \begin{lemm} \label{L:NormalBis}

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

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

549: \dd$, we have $\DivR(\sx_\kk)

550: \supseteq \DivL(\sx_{\kk+1})$.

551: \end{lemm}

552: 

553: \subsection{Connection with permutations}

554: 

555: Everywhere in the sequel, we write¨$\Int\nn$ for $\{1,

556: \dots, \nn\}$. By mapping¨$\ss i$ to the transposition¨$(i,

557: i+1)$, one defines a surjective homomorphism¨$\pi$

558: of¨$B_\nn$ onto the symmetric group¨$\Sym\nn$. The

559: restriction of¨$\pi$ to simple braids is a bijection:

560: for every permutation¨$f$ of¨$\Int\nn$, there exists exactly

561: one simple braid¨$\bs$ satisfying $\perm\bs = f$.

562: 

563: The Exchange Lemma for Coxeter groups connects

564: the sets¨$\DivL(\sx)$ and¨$\DivR(\sx)$ with the

565: permutation associated with¨$\sx$ and their descents. 

566: For¨$\ff$ a permutation, use $\ell(\ff)$ for the minimal 

567: number of factors occurring in a decomposition of¨$\ff$ 

568: as a product of transpositions. The precise statement is

569: 

570: \begin{lemm} \label{L:Exchange}

571: Let $\bs$ be a simple $\nn$-braid¨$\bs$. For $1 \le i

572: < \nn$, the following are equivalent:

573: 

574: $(i)$ The braid¨$\ss i$ is a left divisor of¨$\bs$

575: in¨$B_\nn^+$, \ie, $i$ belongs to¨$\DivL(\bs)$; 

576: 

577: $(ii)$ The strands starting at positions¨$i$ and¨$i+1$

578: cross in any (positive) diagram for¨$\bs$;

579: 

580: $(iii)$ We have $\pi(\bs)\inv(i) > \pi(\bs)\inv(i+1)$;

581: 

582: $(iv)$ We have $\ell(\pi(\ss i\bs)) < \ell(\perm\bs)$, \ie, $i$ is

583: a descent of¨$\perm\bs\inv$.\\

584: Symmetrically, the following are equivalent:

585: 

586: $(i')$ The braid¨$\ss i$ is a right divisor of¨$\bs$

587: in¨$B_\nn^+$, \ie, $i$ belongs to¨$\DivR(\bs)$;  

588: 

589: $(ii')$ The strands finishing at positions¨$i$ and¨$i+1$

590: cross in any (positive) diagram for¨$\bs$;

591: 

592: $(iii')$ We have $\pi(\bs)(i) > \pi(\bs)(i+1)$;

593: 

594: $(iv')$ We have $\ell(\pi(\bs\ss i)) < \ell(\perm\bs)$, \ie, $i$ is

595: a descent of¨$\perm\bs$.

596: \end{lemm}

597: 

598: So, for¨$\bs$ a simple braid, the indices¨$i$ such that $\ss i$ is a right

599: divisor of¨$\bs$ are the descents of the associated permutation¨$\perm\bs$,

600: while those such that $\ss i$ is a left divisor of¨$\bs$ are the

601: descents of¨$\perm\bs\inv$.

602: 

603: \subsection{The numbers $\bb\nn\dd$ and $\bbb\nn\dd\bs$}

604: 

605: Our aim in this paper is to solve various counting problems

606: involving the normal form of positive braids. The main

607: numbers we investigate are as follows:

608: 

609: \begin{defi}

610: For $\nn, \dd \ge 1$, we denote by¨$\bb\nn\dd$ the number

611: of positive $\nn$¨strand braids of degree at most¨$\dd$,

612: \ie, the number of divisors of¨$\DD\nn\dd$ in the braid

613: monoid.

614: \end{defi}

615: 

616: By Lemma¨\ref{L:NormalBis}, $\bb\nn\dd$ is the number of

617: normal sequences of length¨$\dd$, \ie, the number of

618: sequences $(\sx_1, \dots, \sx_\dd)$ where all¨$\sx_\kk$ are

619: simple braids and $\DivL(\sx_\kk) \supseteq

620: \DivR(\sx_{\kk+1})$ holds for $\kk < \dd$. By

621: Lemma¨\ref{L:Exchange}, it is also the number of sequences

622: of permutations $(\ff_1, \dots, \ff_\dd)$ such that, for

623: each¨$\kk < \dd$, the descents of¨$\ff_{\kk+1}\inv$ are

624: included in those of¨$\ff_\kk$.

625: 

626: For¨$\dd = 1$, the bijection between simple $\nn$¨strand

627: braids and permutations of¨$\Int\nn$ immediately gives

628: \begin{equation}

629: \bb\nn1 = n!,

630: \end{equation}

631: which implies for all¨$\nn, \dd$

632: \begin{equation} \label{E:Upper}

633: \bb\nn\dd \le (\nn!)^\dd.

634: \end{equation}

635: 

636: In the sequel, we shall have to count normal sequences

637: satisfying some constraints. So we introduce one more

638: notation.

639: 

640: \begin{defi}

641: For $\nn, \dd \ge 1$ and $\bs$ a simple $\nn$-braid, we

642: denote by¨$\bbb\nn\dd\bs$ the number of positive

643: $\nn$¨strand braids of degree at most¨$\dd$ with $\dd$th

644: factor equal to¨$\bs$.

645: \end{defi}

646: 

647: In other words, $\bbb\nn\dd\bs$ is the number of normal

648: sequences of the form $(\bs_1, \dots, \bs_{\dd-1}, \bs)$.

649: Some connections are obvious:

650: 

651: \begin{prop} \label{P:Total}

652: For all¨$\nn, \dd$, we have

653: \begin{equation} \label{E:Last}

654: \bb\nn\dd = \sum_{\mbox{\Small\rm $\bs$ simple}} \bbb\nn\dd\bs =

655: \bbb\nn{\dd+1}1.

656: \end{equation}

657: \end{prop}

658: 

659: \begin{proof}

660: The first equality is obvious. The second one follows from

661: the fact that $(\bs_1, \dots, \bs_\dd)$ is normal if and

662: only if $(\bs_1, \dots, \bs_\dd, 1)$ is: indeed, $1$ has no

663: left divisor but itself, so, by Lemma¨\ref{L:Normal}, every

664: sequence $(\bs, 1)$ is normal.

665: \end{proof}

666: 

667: \section{Adjacence matrices}

668: \label{S:Adj}

669: 

670: In this section, we study the numbers $\bb\nn\dd$ and

671: $\bbb\nn\dd\bs$ when $\nn$ is fixed and $\dd$ varies. By

672: Lemma¨\ref{L:Normal}, normal sequence of simple braids are

673: characterized by a purely local criterion that only involves

674: adjacent entries. It follows that the set of all normal

675: sequences can be recognized by a finite state automaton

676: \cite{Eps}, and, as a consequence, the associated counting

677: numbers obey a linear induction rule specified by a certain

678: incidence matrix¨\cite{Eps2}. In this section, we define

679: the matrix involved in the current situation, and show how

680: its size, which is originally¨$\nn!$, can be lowered

681: to¨$2^\nno$.

682: 

683: \subsection{Enumeration of simple braids}

684: 

685: Below we consider matrices whose entries are indexed

686: by simple braids (or, equivalently, permutations). Fixing an

687: enumeration of simple braids is not important at a

688: conceptual level, but this is necessary when the objects

689: are to be specified explicitly. We shall use the

690: restriction of the canonical linear ordering of braids

691: denoted¨$<^{\scriptscriptstyle\phi}$

692: in¨\cite{Dgr}---which gives for each¨$\nn$ a well-ordering of

693: ordinal type¨$\omega^{\omega^{\nn-2}}$ on¨$B_\nn^+$. 

694: The corresponding increasing enumeration of simple

695: $\nn$-braids can be constructed directly using induction

696: on¨$\nn$. We start from the following easy remark:

697: 

698: \begin{lemm} \label{L:Factor}

699: For $1 \le i \le \nn$, write $\ss{i, \nn}$ for $\ss i

700: \ss{i+1} \dots \ss{\nno}$ (so that $\ss{\nn, \nn}$

701: is¨$1$). Then every simple $\nn$-braid¨$\bs$ admits a unique

702: decomposition $\bs = \ss{i, \nn} \, \bt$ with $1

703: \le i \le \nn$ and $\bt$ a simple $(\nno)$-braid.

704: \end{lemm}

705:  

706: \begin{proof}

707: (Figure¨\ref{F:Factor})

708: Let $i = \perm\bs(\nn)$. Then we can realize¨$\bs$ by a

709: diagram in which the $i$th strand is first sent to the

710: rightmost position, and it remains a simple $(\nno)$-braid.

711: Conversely, we have $i = \perm{\ss{i, \nn}\bt}(\nn)$, so the

712: decomposition is unique.

713: \end{proof}

714: 

715: \begin{figure} [htb]

716: \begin{picture}(35,20)(0, 0)

717: \put(0,0){\includegraphics{Factor.eps}}

718: \put(34,9){$\bs$}

719: \put(11.5,5){$\bt$}

720: \put(12,19.5){$i$}

721: \end{picture}

722: \caption{\smaller Proof of Lemma¨\ref{L:Factor}}

723: \label{F:Factor}

724: \end{figure}

725: 

726: \begin{defi}

727: We inductively define an enumeration¨$S_\nn$ of simple

728: $\nn$-braids by

729: \begin{equation}

730: S_1:= (1), \quad

731: S_\nn:= S_{\nno} \concat \ss{\nno, \nn} S_{\nno} \concat

732: \dots \concat \ss{1, \nn} S_{\nno},

733: \end{equation}

734: where $\concat$ stands for list concatenation, and $\bs S$ is

735: the list obtained from¨$S$ by multipliying all entries

736: by¨$\bs$ on the left. The $\kk$th element in¨$\bigcup_\nn

737: S_\nn$ is denoted¨$\sp\kk$.

738: \end{defi}

739: 

740: The first¨$\sp\kk$'s are, in increasing order,

741: $$\sp1 = 1\ , 

742: \sp2 = \ss1\ , 

743: \sp3 = \ss2\ , 

744: \sp4 = \ss2\ss1\ ,

745: \sp5 =  \ss1\ss2\ , 

746: \sp6 = \ss1\ss2\ss1\ , 

747: \sp7 = \ss3, \dots$$

748: Lemma¨\ref{L:Factor} guarantees that all simple braids

749: occur in the above enumeration. Note that, for every¨$\nn$,

750: we have $\D_\nn = \sp{\nn!}$.

751: 

752: It is easy to check that

753: the ordering of simple braids we use corresponds to a

754: reversed antilexicographic ordering of the inverses of the

755: associated permutations: $\bs$ occurs before¨$\bt$ if and

756: only if we have $\perm\bs\inv < \perm\bt\inv$, where 

757: $\ff < \gg$ is said to hold if we have $\ff(i) > \gg(i)$ for

758: the largest¨$i$ for which $\ff$ and $\gg$ do not agree.

759: 

760: \subsection{The matrix¨$\Mat\nn$}

761: 

762: Everywhere in the sequel, we write $(M)_{\xx, \yy}$ for the

763: $(\xx, \yy)$-entry of a matrix¨$M$.

764: 

765: \begin{defi} 

766: For $\nn \ge 1$, we define $\Mat\nn$ to be the $\nn! \times

767: \nn!$ matrix satisfying

768: $$

769: (\Mat\nn)_{\kk, \ll} = 

770: \begin{cases}

771: 1 & \mbox{if $(\sp\kk, \sp\ll)$ is normal,}\\

772: 0 & \mbox{otherwise.}

773: \end{cases}$$

774: \end{defi}

775: 

776: Instead of referring to integer entries, it will be often

777: convenient to think of the entries of¨$\Mat\nn$ as directly

778: indexed by simple braids; for $\bs, \bt$ simple braids, we

779: simply write $(\Mat\nn)_{\bs, \bt}$ for the

780: corresponding entry.

781: 

782: \begin{exam}

783: The first 3 matrices¨$\Mat\nn$ are

784: $$

785: \Mat1 = 1, \quad

786: \Mat2 = 

787: \left(\begin{matrix}1&0\\1&1\end{matrix}\right), \quad

788: \Mat3 = \left(

789: \begin{matrix}

790: 1&0&0&0&0&0\\

791: 1&1&0&0&1&0\\

792: 1&0&1&1&0&0\\

793: 1&1&0&0&1&0\\

794: 1&0&1&1&0&0\\

795: 1&1&1&1&1&1

796: \end{matrix}

797: \right).$$

798: \end{exam}

799: 

800: The construction of the matrix¨$\Mat\nn$ immediately

801: implies the following results:

802: 

803: \begin{lemm} \label{L:Stack}

804: $(i)$ The first column and the last row of¨$\Mat\nn$

805: contain only¨$1$'s; the first row, its first entry

806: excepted, and the last column, its last entry excepted, 

807: contain only¨$0$'s.

808: 

809: $(ii)$ The first $(\nno)!$ columns of¨$\Mat\nn$ consist of

810: $\nn$¨stacked copies of¨$\Mat{\nno}$. 

811: 

812: $(iii)$ If $\DivR(\sp\kk) = \DivR(\sp{\kk'})$ holds, then

813: the $\kk$th and the $\kk'$th rows in¨$\Mat\nn$ coincide.

814: Similarly, if $\DivL(\sp\ll) = \DivL(\sp{\ll'})$ holds, then

815: the $\ll$th and the $\ll'$th columns in¨$\Mat\nn$ coincide.

816: \end{lemm}

817: 

818: \begin{proof}

819: $(i)$ By  construction, we have $\sp1 = 1$ and

820: $\sp{\nn!} = \DD\nn{}$. Now $(\bs, 1)$ is always normal,

821: and so is $(\DD\nn{}, \bs)$. On the other hand, $(1 , \bs)$

822: is normal only for $\bs = 1$, and $(\bs, \DD\nn{})$ is

823: normal only for $\bs = \DD\nn{}$.

824: 

825: $(ii)$ Assume $\kk \le (\nno)!$ and $\kk' = \kk + (\nn-i)

826: \cdot (\nno)!$. Our enumeration of simple braids implies

827: $\sp{\kk'} = \ss{i, \nn} \sp\kk$. Then

828: Figure¨\ref{F:Factor} makes the equality

829: $\DivR(\sp{\kk'}) \cap \Int{\nno} = \DivR(\sp\kk)$ clear.

830: For every simple $(\nno)$-braid¨$\bt$, the

831: set¨$\DivL(\bt)$ is included in¨$\Int{\nno}$, and it

832: follows that $(\sp{\kk'}, \bt)$ is normal if and only

833: if¨$(\sp\kk, \bt)$ is. In other words, we have 

834: $(\Mat\nn)_{\kk', \ll} = (\Mat\nn)_{\kk, \ll}$ for $\ll \le

835: (\nno)!$.

836: 

837: $(iii)$ By Lemma¨\ref{L:Normal}, the value of

838: $(\Mat\nn)_{\kk, \ll}$ only depends on¨$\DivR(\sp\kk)$ and

839: on¨$\DivL(\sp\ll)$.

840: \end{proof}

841: 

842: The connection between the numbers¨$\bbb\nn\dd\bs$ and the

843: matrix¨$\Mat\nn$ is straightforward:

844: 

845: \begin{lemm} \label{L:Comput}

846: For every simple¨$\bt$ and every

847: $\dd \ge 1$, we have

848: \begin{equation} \label{E:Number}

849: \bbb\nn\dd{\bt} = ((1, 1, \dots, 1) \,

850: \Mat\nn\!\!\!{}^{\dd-1})_\bt.

851: \end{equation}

852: \end{lemm}

853: 

854: \begin{proof}

855: Induction on¨$\dd$. For $\dd=1$, and for each simple

856: $\nn$-braid¨$\bs$, there is exactly one braid of

857: degree at most¨$1$ whose first factor is¨$\bt$,

858: namely $\bt$¨itself, and we have $\bbb\nn1\bt = 1$.

859: Assume $\dd \ge 2$. By Lemma¨\ref{L:Normal}, $(\bs_1,

860: \dots, \bs_{\dd-1}, \bt)$ is normal if and only if $(\bs_1,

861: \dots, \bs_{\dd-1})$ and $(\bs_{\dd-1}, \bt)$ are normal,

862: so we get 

863: \begin{equation*} \label{E:Comput}

864: \bbb\nn\dd{\bt} 

865: = \sum_{(\bs, \bt)\mbox{ \Small normal}}

866: \bbb\nn{\dd-1}{\bs} = \sum_\bs

867: \bbb\nn{\dd-1}{\bs}

868: \, (\Mat\nn)_{\bs, \bt},

869: \end{equation*}

870: and \eqref{E:Number} follows inductively. 

871: \end{proof}

872: 

873: \begin{rema}

874: As last row of¨$\Mat\nn$ is $(1, \dots, 1)$, we have

875: $(1, \dots, 1) = (0, \dots, 0, 1) \Mat\nn$, and we can

876: replace¨\eqref{E:Number} with

877: \begin{equation} \label{E:NumberBis}

878: \bbb\nn\dd{\bt} = ((0, \dots, 0, 1) \,

879: \Mat\nn\!\!\!{}^{\dd})_\bt.

880: \end{equation}

881: \end{rema}

882: 

883: \begin{exam}

884: Using the value of¨$\Mat2$, we immediately

885: find $\bbb2\dd1 = \dd$, $\bbb2\dd{\ss1} = 1$, as

886: could be expected: there are $\dd+1$ braids of degree

887: at most¨$\dd$, namely the braids¨$\ss1^\kk$ with

888: $\kk < \dd$, whose $\dd$th factor is¨$1$, and

889: $\ss1^\dd$, whose $\dd$th factor is¨$\D_2$, \ie,¨$\ss1$. 

890: \end{exam}

891: 

892: The computation for $\nn \ge 3$ is more complicated, and we

893: postpone it. For the moment, we just point that, as the

894: numbers¨$\bbb\nn\dd\bs$ obey the linear

895: recurrence¨\eqref{E:Number}, standard arguments imply  that

896: they can be expressed in terms of the eigenvalues

897: of¨$\Mat\nn$:

898: 

899: \begin{prop} \label{P:Expansion}

900: Let $\ev_1$, \dots, $\ev_\ir$ be the non-zero

901: eigenvalues of¨$\Mat\nn$. Then,  for

902: each simple $\nn$-braid¨$\bs$, there

903: exist polynomials $P_1, \dots, P_\ir$ with $\deg(P_\ind)$

904: at most the multiplicity of¨$\ev_\ind$ for¨$\Mat\nn$ such

905: that, for each¨$\dd \ge 0$, we have

906: \begin{equation} \label{E:General}

907: \bbb\nn\dd\bs = P_1(\dd) \ev_1^\dd + \cdots +

908: P_\ir(\dd) \ev_\ir^\dd.

909: \end{equation} 

910: \end{prop}

911: 

912: \begin{coro}

913: For all¨$\nn, \bs$, the generating function of the

914: numbers¨$\bbb\nn\dd\bs$'s with respect to¨$\dd$ is rational.

915: \end{coro}

916: 

917: \subsection{Reducing the size}

918: 

919: The size¨$\nn!$ of the incidence¨$\Mat\nn$ is uselessly

920: large, and we shall see now how to lower it. This will be

921: done in two steps. The first one relies on the fact,

922: pointed out in Lemma¨\ref{L:Stack}$(iii)$, that many columns

923: in¨$\Mat\nn$ are equal. For subsequent use, it will be

924: useful to introduce a new sequence of numbers:

925: 

926: \begin{defi}

927: For $\II, \JJ \ince \Int{\nno}$, we denote

928: by $\aa\nn\II\JJ$ (\resp $\aas\nn\II\JJ$) the

929: number of simple $\nn$-braids satisfying $\DivL(\bs) =

930: \II$ (\resp $\DivL(\bs) \supseteq \II$) and

931: $\DivR(\bs) \supseteq \JJ$.

932: \end{defi}

933: 

934: \begin{lemm} \label{L:Reduction}

935: For $\nn \ge 1$, let $\Matt\nn$ be the $2^\nno \times

936: 2^\nno$ matrix with entries indexed by subsets

937: of¨$\Int{\nno}$ defined by $(\Matt\nn)_{\II, \JJ} = 

938: \aa\nn\II\JJ$. Then the characteristic polynomials

939: of¨$\Matt\nn$ and $\Mat\nn$ coincide up to a power of¨$\xx$,

940: and, for every simple¨$\bt$ with $\DivL(\bt) = \JJ$ and

941: every $\dd \ge 1$, we have

942: \begin{equation} \label{E:Numberr}

943: \bbb\nn\dd{\bt} = ((1, 1, \dots, 1) \,

944: \Matt\nn\!{}^{\dd-1})_\JJ.

945: \end{equation}

946: \end{lemm}

947: 

948: \begin{proof}

949: Gathering the columns corresponding to simples with

950: the same¨$\DivL$¨set and summing the corresponding

951: lines amounts to replacing¨$\Mat\nn$ with a similar

952: matrix of the form

953: $\left(\begin{matrix}

954: \Matt\nn&0\\\dots&0 \end{matrix}\right)$, so the

955: result about the characteristic polynomial is clear.

956: 

957: As for the value of¨$\bbb\nn\dd\bt$, the argument

958: is similar to that for Lemma¨\ref{L:Comput}. The

959: induction starts as $\aa\nn\II{\Int{\nno}} =

960: 1$ holds for each¨$\II$. For the general step, we

961: find

962: \begin{align*} \label{E:Comput}

963: \bbb\nn\dd{\bt} 

964: &= \sum_{(\bs, \bt)\mbox{ \Small normal}}

965: \bbb\nn{\dd-1}{\bs} 

966: = \sum_{\DivR(\bs) \supseteq\JJ}

967: \bbb\nn{\dd-1}{\bs} 

968: = \sum_\II \sum_{

969: \underset{\scriptstyle \DivL(\bs) = \II}

970: {\scriptstyle \DivR(\bs) \supseteq\JJ}}

971: \bbb\nn{\dd-1}{\bs} \\

972: &= \sum_\II  \bbbb\nn{\dd-1}{\II} \aa\nn\II\JJ

973: = \sum_\II  ((1, \dots, 1)

974: \Matt\nn{}^{\dd-2})_{\II} (\Matt\nn)_{\II, \JJ}

975: = ((1, \dots, 1)

976: \Matt\nn{}^{\dd-1})_{\JJ},

977: \end{align*}

978: where $\bbbb\nn\dd\II$ denotes the common

979: value of $\bbb\nn\dd\bs$ for $\bs$ with¨$\DivL(\bs) =

980: \II$. 

981: \end{proof}

982: 

983: For $\nn = 3$, and using the enumeration $\emptyset$,

984: $\{1\}$, $\{2\}$, $\{1,2\}$ that is induced by our

985: enumeration of simple braids, we obtain 

986: $\Matt3 = \left(\begin{matrix}

987: 1&0&0&0\\2&1&1&0\\

988: 2&1&1&0\\

989: 1&1&1&1

990: \end{matrix}\right)$. Observe that the second

991: and third columns in¨$\Matt3$ coincide, which suggests a

992: further reduction step.

993: 

994: \section{Partitions associated with a simple braid}

995: \label{S:Part}

996: 

997: We can indeed reduce the size of the matrices once more: 

998: we can replace the

999: incidence matrix¨$\Matt\nn$ with a new matrix¨$\Mattt\nn$,

1000: whose size is¨$p(\nn)$, the number of partitions of¨$\nn$.

1001: Here the result is deduced from elementary remarks

1002: about simple braids (or, equivalently, about

1003: permutations); it can also be deduced from classical

1004: results about Solomon's algebra of descents---and 

1005: therefore extends to all Artin--Tits groups of

1006: spherical type. 

1007: 

1008: \subsection{Computation of¨$\aas\nn\II\JJ$}

1009: 

1010: We shall start from an explicit determination of the value

1011: of the numbers¨$\aas\nn\II\JJ$ in terms of the block

1012: compositions of¨$\II$ and¨$\JJ$. We first recall the notions

1013: of composition and partition. 

1014: 

1015: \begin{defi}

1016: For $\II \ince \Int{\nno}$, the {\it

1017: $\nn$-composition}¨$\comp\nn\II$ of~$\II$ is defined to be

1018: the sequence $(\ip_1, \ip_2-\ip_1,

1019: \dots,

1020: \ip_\ir-\ip_{\ir-1})$, where

1021: $\ip_1, \dots, \ip_\ir$ is the increasing

1022: enumeration of $\Int\nn \setminus \II$. The {\it

1023: $\nn$-partition}¨$\part\nn\II$ of~$\II$ is the

1024: non-increasing rearrangement of¨$\comp\nn\II$.

1025: \end{defi}

1026: 

1027: \begin{exam}

1028: The composition of~$\II$ consists of the sizes of the blocks

1029: of adjacent elements in~$\II$, augmented by¨$1$: for

1030: instance, the $10$-composition of

1031: $\{1,2,4,5,6,9\}$ is¨$(3, 4, 1, 2)$, as the blocks are  of

1032: size¨$2, 3, 0, 1$. The $10$-partition of~$\II$ is therefore

1033: $(4, 3, 2, 1)$. Note that the $\nn$-composition of~$\II$

1034: determines~$\II$, but its $\nn$-partition does not.

1035: \end{exam}

1036: 

1037: The geometric observation is the following one:

1038: 

1039: \begin{lemm} \label{L:Sequence}

1040: Assume $\JJ \ince \Int{\nno}$, and let

1041: $(\iq_1, \dots, \iq_\is)$ be the $\nn$-composition

1042: of¨$\JJ$. For $\bs$ a simple $\nn$-braid, define

1043: $\ff_\bs : \Int\nn \to \Int\is$ by $\ff_\bs(i) = j$

1044: if the $i$th strand of¨$\bs$ finishes in the

1045: $j$th block of¨$\JJ$. Then $\bs \mapsto \ff_\bs$

1046: establishes a bijection between the simple

1047: $\nn$-braids¨$\bs$ that satisfy

1048: $\DivR(\bs) \supseteq \JJ$ and the functions

1049: from¨$\Int\nn$ to¨$\Int\is$. Moreover, for

1050: $\DivR(\bs) \supseteq \JJ$, we have, for every¨$i$, 

1051: \begin{equation} \label{E:Equiv}

1052: i \in \DivL(\bs) 

1053: \quad  \Longleftrightarrow \quad

1054: \ff_\bs(i) \ge \ff_\bs(i+1).

1055: \end{equation}

1056: \end{lemm} 

1057: 

1058: \begin{proof}

1059: Assume that $\bs$ is a simple $\nn$-braid

1060: satisfying $\DivR(\bs) \supseteq \JJ$, \ie, that is

1061: right divisible by¨$\ss j$ for each¨$j$ in¨$\JJ$.

1062: By hypothesis, the first block of consecutive

1063: elements of¨$\JJ$ is $1, \dots \iq_1-1$. Then $\bs$

1064: being right divisible by $\ss1$, \dots,

1065: $\ss{\iq_1-1}$ is equivalent to its being right

1066: divisible by the left lcm of these elements, which

1067: is¨$\D_{\iq_1}$. Similarly, the second block

1068: in¨$\JJ$ is $\ss{\iq_1+1}$, \dots, $\ss{\iq_1 +

1069: \iq_2-1}$, and being right divisible by these

1070: elements amounts to being right divisible by their

1071: left lcm, which is $\sh^{\iq_1}(\D_{\iq_2})$, where

1072: $\sh$ denotes the shift endomorphism of¨$B_\infty$

1073: that maps each¨$\ss i$ to¨$\ss{i+1}$. Now, by

1074: construction, the blocks $\{1, \dots, \iq_1-1\}$

1075: and $\{\iq_1+1 \dots, \iq_1 + \iq_2-1\}$ are

1076: separated by¨$\iq_1$, and, therefore, the

1077: corresponding¨$\s$'s commute. In particular,

1078: $\D_{\iq_1}$ and $\sh^{iq_1}(\D_{\iq_2})$ commute,

1079: and their left lcm is their product. Finally, a

1080: simple braid¨$\bs$ satisfies

1081: $\DivR(\bs) \supseteq \JJ$ if and only if it is a

1082: right multiple of the element

1083: $$\D_\JJ = \D_{\iq_1} \, \sh^{\iq_1}(\D_{\iq_2}) \,

1084: \dots \, \sh^{\iq_1 + \cdots +

1085: \iq_{\is-1}}(\D_{\iq_\is}),$$ 

1086: \ie, we have $\bs = \bs' \D_\JJ$ for some¨$\bs'$.

1087: 

1088: We claim that $\ff_\bs$ determines¨$\bs'$,

1089: hence¨$\bs$. Indeed, in a simple braid, any two

1090: strands cross at most once. Now, in a $\D$-diagram,

1091: any two strands cross. So, if the $i$th and the

1092: $i'$th strands go to the same block of¨$\JJ$, \ie,

1093: if we have $\ff_\bs(i) = \ff_\bs(i')$, then these

1094: strands cross in the final¨$\D$-part, and therefore

1095: they cannot cross in (any diagram

1096: representing)¨$\bs'$. So, when

1097: $\ff_\bs$ is given, there is only one way to

1098: construct¨$\bs'$, namely taking the strands to the

1099: entrance of the specified $\D$-block in increasing

1100: order (Figure¨\ref{F:Blocks}). 

1101: 

1102: Consider now¨$i$, $1 \le i < \nn$. We wonder

1103: whether $\ss i$ is a left divisor of¨$\bs$, \ie, if

1104: the $i$th and the $i + 1$st strands cross in the

1105: diagram of¨$\bs$. If we have $\ff_\bs(i) =

1106: \ff_\bs(i+1)$, the $i$th and $i+1$st strand go to

1107: the same $\JJ$-block, where they certainly

1108: cross. If we have $\ff_\bs(i) >

1109: \ff_\bs(i+1)$, then the $i$th strand goes to a

1110: $\JJ$-block on the right of the $\JJ$-block to

1111: which the $i+1$st strand goes, so they must cross

1112: in the $\bs'$ part. On the contrary, for

1113: $\ff_\bs(i) < \ff_\bs(i+1)$, the strands cannot

1114: cross in the $\bs'$ part---if they crossed once,

1115: they would have to cross a second time before

1116: exiting, and this is forbidden---and they do not

1117: cross in the $\D$ part either. So \eqref{E:Equiv}

1118: holds.

1119: \end{proof}

1120: 

1121: \begin{figure} [htb]

1122: \begin{picture}(92,44)(0, 0)

1123: \put(0,4){\includegraphics{Blocks.eps}}

1124: \put(75,32){the $\bs'$ part}

1125: \put(75,14){the $\D$ part}

1126: \put(1,15){$\D_{\iq_1}$}

1127: \put(46,15){$\D_{\iq_\is}$}

1128: \put(0,0){first block of¨$\JJ$}

1129: \put(30,0){$\dots$}

1130: \put(55,35){$\dots$}

1131: \put(45,0){last block of¨$\JJ$}

1132: \end{picture}

1133: \caption{\smaller A simple braid divisible by all

1134: $\ss j$'s with $j$ in¨$\JJ$ can be represented by a

1135: diagram finishing with $\D$'s corresponding to the

1136: blocks of¨$\JJ$; the strands going to the same block

1137: cannot cross in the $\bs'$ part, as they cross

1138: inside the block; so the strands starting from¨$i$

1139: and¨$i'$ with $i < i'$ cross if and only if  they

1140: go to different blocks and $i$ goes to a block on

1141: the right of the block $i'$ goes to.}

1142: \label{F:Blocks}

1143: \end{figure}

1144: 

1145: We deduce the following characterization

1146: of¨$\aas\nn\II\JJ$:

1147: 

1148: \begin{prop} \label{P:EES}

1149: Assume that $\II, \JJ$ are subsets of~$\Int{\nno}$

1150: with respective $\nn$-composi\-tions $(\ip_1, \dots,

1151: \ip_\ir)$ and $(\iq_1, \dots, \iq_\is)$. Then $\aas\nn\II\JJ$

1152: is the number of $\ir \times \is$ matrices

1153: with nonnegative integer entries such that, for

1154: all~$i, j$, the $i$th row has sum~$\ip_i$ and the

1155: $j$th~column has sum~$\iq_j$. In particular, we have

1156: \begin{equation} \label{E:Value}

1157: \aas\nn\II\emptyset = \frac{\nn!}{\ip_1! \dots

1158: \ip_\ir!}

1159: \mbox{\quad and \quad}

1160: \aas\nn\emptyset\JJ = \frac{\nn!}{\iq_1! \dots

1161: \iq_\is!}.

1162: \end{equation}

1163: \end{prop}

1164: 

1165: \begin{proof}

1166: Lemma¨\ref{L:Sequence} immediately implies

1167: \begin{gather} 

1168: \label{E:Sequence}

1169: \aa\nn\II\JJ = \card

1170: \{\ff \!:\! \Int\nn \!\to\! \Int\is \,;\, 

1171: (\forall j)(\card\ff\inv(j) = \iq_j) 

1172: \,\&\,

1173: (i \in \II \Leftrightarrow \ff(i) \ge \ff(i+1))\},\\

1174: \label{E:SequenceBis}

1175: \aas\nn\II\JJ = \card\{\ff \!:\! \Int\nn \!\to\!

1176: \Int\is \,;\,  (\forall j)(\card\ff\inv(j) = \iq_j) 

1177: \,\&\,

1178: (i \in \II \Rightarrow \ff(i) \ge \ff(i+1))\}.

1179: \end{gather}

1180: 

1181: 

1182: 

1183: Assume that $\ff$ is a function of~$\Int\nn$

1184: to~$\Int\is$ satisfying the constraints

1185: of~\eqref{E:SequenceBis}. Let $A_\ff$ be the $\ir

1186: \times \is$-matrix whose $(i, j)$-entry is the

1187: number of~$k$'s in the $i$th block of~$\II$

1188: satisfying~$\ff(k) = \iq_j$. By construction, the

1189: the sum of the $i$th row of~$A_\ff$ is the

1190: size~$\ip_i$ of the $i$th block of~$\II$, while

1191: the sum of the $j$th column is the number of~$k$'s

1192: satisfying $\ff(k) = j$, \ie, it is~$\iq_j$. We

1193: claim that $A_\ff$ determines~$\ff$. Indeed,

1194: \eqref{E:SequenceBis} requires that $\ff$ be

1195: non-increasing on each block of~$\II$, so there is

1196: only one possibility once the number of~$k$'s

1197: going to the various~$j$ is fixed.

1198: 

1199: The first equality in¨\eqref{E:Value} follows:

1200: for¨$\is = \nn$, there is exactly one nonzero entry

1201: in each column, so choosing a convenient matrix

1202: amounts to choosing among $\nn$~elements the

1203: $\iq_1$~columns with a¨$1$ in the first row, the

1204: $\iq_2$~columns with a¨$1$ in the second row, \etc\

1205: The second equality is similar with rows and

1206: columns exchanged. 

1207: \end{proof}

1208: 

1209: \begin{coro} \label{C:Depends}

1210: $(i)$ The number¨$\aas\nn\II\JJ$ only depends on the

1211: partitions¨$\part\nn\II$ and¨$\part\nn\JJ$.

1212: 

1213: $(ii)$ For each~$\nn$ and¨$\II$, the number¨$\aa\nn\II\JJ$

1214: only depends on the partition¨$\part\nn\JJ$.

1215: \end{coro}

1216: 

1217: \begin{proof}

1218: Point¨$(i)$ directly follows from the characterization of

1219: Proposition¨\ref{P:EES}, as the latter clearly involves the

1220: sizes of the blocks of¨$\II$ and¨$\JJ$ only. As for¨$(ii)$, 

1221: the usual inclusion-exclusion formula gives

1222: $$\aa\nn\II\JJ = \sum_{\KK \cap\II = \emptyset}

1223: (-1)^{\card\KK} \aas\nn{\II \cup \KK}\JJ.$$

1224: By¨$(i)$, each term in the sum only depends

1225: on¨$\part\nn\JJ$, and so does the sum.

1226: \end{proof}

1227: 

1228: It is easy to check that the value of¨$\aa\nn\II\JJ$ does

1229: not only depend on¨$\part\nn\II$ in

1230: general: when we apply the inclusion-exclusion formula, 

1231: the sizes of the blocks in¨$\II \cup \KK$ do not only depend

1232: on the sizes of the block in¨$\II$.

1233: 

1234: \begin{rema} \label{R:Hohlweg}

1235: Corollary¨\ref{C:Depends} can also be deduced from the classical results by

1236: Solomon about the descent algebra---and, therefore, it extends to all

1237: Artin--Tits groups of spherical type. The argument is as follows. For¨$f$ a

1238: permutation, let

1239: $D(f)$ denote the sets of descents of¨$f$. In the group

1240: algebra¨$\QQ[\Sym\nn]$, let $d_I:= \sum\{f; D(f) = I\}$ and

1241: $e_\JJ= \sum\{f; D(f) \cap \JJ = \emptyset\}$. Using $w_0$

1242: for the flip permutation, we have $w_0 e_\JJ = \sum\{f; D(f) \supseteq

1243: \JJ\}$, and therefore $\aa\nn\II\JJ = \langle d_\II, w_0e_\JJ\rangle$,

1244: where $\langle.,.\rangle$ is the inner product defined by $\langle f,

1245: g\rangle = 1$ for $g = f\inv$, and¨$=0$ otherwise. Using the isometry result

1246: of¨\cite{Hoh}, we deduce $\aa\nn\II\JJ = \langle \theta(\dd_\II),

1247: \theta(w_0e_\JJ)\rangle$, where $\theta(e_\KK)$ denotes the character

1248: of¨$\Sym\nn$ induced by the trivial character of the standard parabolic

1249: subgroup generated by (the transpositions¨$s_i$ with¨$i$ in)¨$\KK$.

1250: By¨\cite{Sol}, the subspace of¨$\QQ[\Sym\nn]$ generated by the¨$e_\KK$ is

1251: a subalgebra, and the kernel of¨$\theta$ is generated by the elements

1252: $\dd_\II - \dd_{\JJ}$ with

1253: $\II, \JJ$ associated with the same partition, and it follows from the above

1254: expression that $\aa\nn\II\JJ$ only depends on the partition

1255: associated with¨$\JJ$.

1256: \end{rema} 

1257: 

1258: 

1259: \subsection{The matrix¨$\Mattt\nn$}

1260: 

1261: We can now come back to the matrix¨$\Matt\nn$, and

1262: replace it for the computation of the numbers¨$\bb\nn\dd$

1263: with a new matrix of smaller size. Indeed, a direct

1264: application of Corollary¨\ref{C:Depends} is

1265: 

1266: \begin{lemm}

1267: Assume that $\JJ$, $\JJ'$ are subsets of¨$\Int\nno$ with

1268: the same $\nn$-partition. Then the $\JJ$th and $\JJ'$th

1269: columns of¨$\Matt\nn$ are equal.

1270: \end{lemm}

1271: 

1272: Thus the process used to replace¨$\Mat\nn$ with¨$\Matt\nn$

1273: can be applied again, \ie, the form a new matrix by

1274: gathering the equal columns and summing the corresponding

1275: rows.

1276: 

1277: \begin{defi}

1278: $(i)$ For¨$\l$ a partition (or a composition) of¨$\nn$, we

1279: denote by¨$\set\l$ the unique subset¨$\II$ of¨$\Int\nno$

1280: satisfying $\comp\nn\II = \l$.

1281: 

1282: $(ii)$ For $\l, \m \vdash \nn$ (\ie, partitions of¨$\nn$),

1283: we put

1284: $$\aab\l\m 

1285: = \sum_{\part\nn\II = \l} \aa\nn\II{\set\m}

1286: = \card\{\bs \,;\, \part\nn{\DivL(\bs)} = \l

1287: \,\&\, \DivR(\bs) \supseteq \set\m\},$$

1288: and we let $\Mattt\nn$ be the matrix with rows and columns

1289: indexed by partitions of¨$\nn$ and whose $(\l, \m)$-entry

1290: is¨$\aab\l\m$.

1291: \end{defi}

1292: 

1293: In this way the size of the matrix has been reduced

1294: from¨$\nn!$ to¨$\nbpart\nn$, the number of partitions

1295: of¨$\nn$. For instance, enumerating partitions in the order

1296: induced by the previous order on $\Pw(\Int\nn)$, we obtain

1297: $$\Mattt3 = 

1298: \left(\begin{matrix}

1299: 1&0&0\\4&2&0\\1&1&1

1300: \end{matrix}\right), \quad

1301: \Mattt4 =

1302: \left(\begin{matrix}1&0&0&0&0\\11&4&1&0&0\\5&3&2&1&0\\

1303: 6&4&2&2&0\\1&1&1&1&1

1304: \end{matrix}\right), \quad

1305: \Mattt5 =

1306: \left(\begin{matrix}

1307: 1&0&0&0&0&0&0\\

1308: 26&8&0&2&0&0&0\\

1309: 23&12&4&5&0&1&0\\

1310: 43&21&5&10&0&2&0\\

1311: 8&6&4&4&2&2&0\\

1312: 18&12&6&8&2&4&0\\

1313: 1&1&1&1&1&1&1

1314: \end{matrix}\right).$$

1315: 

1316: Applying the same argument as for

1317: Lemma¨\ref{L:Reduction}, we obtain:

1318: 

1319: \begin{prop} \label{P:NNNumber}

1320: For $\nn \ge 1$, the characteristic polynomials

1321: of¨$\Mattt\nn$ and $\Mat\nn$ coincide up to a power of¨$\xx$,

1322: and, for every simple $\nn$-braid¨$\bs$, we have for

1323: each¨$\dd$

1324: \begin{equation}

1325: \bbb\nn\dd{\bs} = ((1,1, \dots, 1) \,

1326: \Mattt\nn^{\dd-1})_\l,

1327: \end{equation}

1328: where $\l$ is the $\nn$-partition of¨$\DivL(\bs)$.

1329: In particular we have 

1330: \begin{equation}

1331: \bb\nn\dd = ((1,1, \dots, 1) \,

1332: \Mattt\nn^{\dd})_{(1, 1, \dots, 1)}.

1333: \end{equation}

1334: \end{prop}

1335: 

1336: Table¨\ref{T:Values} gives the first few values deduced

1337: from the above formulas.

1338: 

1339: \begin{table}[htb]

1340: $$\begin{tabular}{c|r|r|r|r|r|r}

1341: \quad$\dd$

1342: &1&2&3&4&5&6\\

1343: \hline

1344: \vrule width0pt height12pt depth6pt

1345: $\bbb2\dd1$

1346: & 1

1347: & 2 & 3 & 4 & 5 & 6 \\

1348: \hline

1349: \vrule width0pt height12pt 

1350: $\bbb3\dd1$

1351: & 1

1352: & 6 & 19 & 48 & 109 & 234 \\

1353: \vrule width0pt depth6pt

1354: $\bbb3\dd{\D_2}$

1355: & 1

1356: & 3 & 7 & 15 & 31 & 63 \\

1357: \hline

1358: \vrule width0pt height12pt 

1359: $\bbb4\dd1$

1360: & 1

1361: &24&211&1\,380&8\,077&45\,252

1362: \\

1363: $\bbb4\dd{\D_2}$

1364: & 1

1365: &12&83&492&2\,765&15\,240\\

1366: \vrule width0pt depth6pt

1367: $\bbb4\dd{\D_3}$

1368: & 1

1369: &4&15&64&309&1\,600 \\

1370: \hline

1371: \vrule width0pt height12pt 

1372: $\bbb5\dd1$

1373: & 1

1374:  & 120 & 3\,651 & 79\,140 

1375: & 1\,548\,701 

1376: & 29\,375\,460 

1377: \\

1378: $\bbb5\dd{\D_2}$

1379: & 1

1380:  & 60 & 1\,501 & 30\,540 

1381: & 585\,811

1382: & 11\,044\,080 

1383: \\

1384: $\bbb5\dd{\D_3}$

1385: & 1

1386:  & 20 & 311 & 5\,260 

1387: & 94\,881

1388: & 1\,755\,360 

1389: \\

1390: \vrule width0pt depth6pt

1391: $\bbb4\dd{\D_4}$

1392: & 1

1393:  & 5 & 31 & 325 

1394: & 4\,931 

1395: & 86\,565 

1396: \\

1397: \hline

1398: \vrule width0pt height12pt 

1399: $\bbb6\dd1$

1400: & 1

1401: & 720 

1402: & 90\,921 

1403: & 7\,952\,040 

1404: & 634\,472\,921

1405: & 49\,477\,263\,360

1406: \\

1407: $\bbb6\dd{\D_2}$

1408: & 1

1409: & 360 

1410: & 38\,559 

1411: & 3\,228\,300 

1412: & 254\,718\,389

1413: & 19\,808\,530\,620\\

1414: $\bbb6\dd{\D_3}$

1415: & 1

1416:  & 120 & 8\,727 & 649\,260 

1417: & 49\,654\,757

1418: & 3\,831\,626\,580 \\

1419: $\bbb6\dd{\D_4}$

1420: & 1

1421:  & 30 & 1\,075 & 61\,620 

1422: & 4\,387\,195

1423: & 332\,578\,230 \\

1424: $\bbb6\dd{\D_5}$

1425: & 1

1426:  & 6 & 63 & 1\,955 

1427: & 116\,423

1428: & 8\,448\,606 

1429: \end{tabular}$$

1430: \medskip

1431: \caption{\smaller First values of $\bbb\nn\dd{\D_\rr}$

1432: for $1 \le \rr < \nn$;  the values of¨$\bb\nn\dd$ can be also

1433: read, as Proposition¨\ref{P:Total} gives $\bb\nn\dd =

1434: \bbb\nn{\dd+1}1$, so, for instance, we find $\bb33= 48$, and

1435: $\bb45 = 45\,252$.}

1436: \label{T:Values}

1437: \end{table}

1438: 

1439: \subsection{Small values of¨$\nn$}

1440: 

1441: For small values of¨$\nn$, it is easy to complete the

1442: computations and to obtain an explicit form for the

1443: expansion of¨$\bbb\nn\dd\bs$ announced in

1444: Proposition¨\ref{P:Expansion}.

1445: 

1446: \begin{exam}

1447: Assume $\nn = 3$. The matrix¨$\Mattt3$ is invertible with

1448: eigenvalues¨$1$ (double) and¨$2$. By solving the

1449: recurrences, we find

1450: \begin{equation*} %\label{E:Number3}

1451: \bbb3\dd\bs =

1452: \begin{cases}

1453: 4 \cdot 2^\dd - 3\dd -4

1454: &\mbox{¨for $\bs$ with partition $(1, 1, 1)$, \ie,

1455: $\bs = 1$}\\ 

1456: 2^\dd - 1, 

1457: &\mbox{¨for $\bs$ with partition $(2, 1)$,

1458: \ie, $\bs = \ss1, \ss2, \ss2\ss1$, or $\ss1\ss2$,}\\

1459: 1

1460: &\mbox{¨for $\bs$ with partition $(3)$, \ie, $\bs =

1461: \D_3$,}

1462: \end{cases}

1463: \end{equation*}

1464: and we deduce $\bb3\dd = 8 \cdot 2^\dd - 3\dd - 7$.

1465: \end{exam}

1466: 

1467: \begin{exam}

1468: Assume now $\nn = 4$. The matrix¨$\Mattt4$ admits 4

1469: eigenvalues, namely those of¨$\Mattt3$, plus $\ev_1 = 3 +

1470: \sqrt6$ and $\ev_2 = 3 - \sqrt6$.  Solving the recurrences

1471: yields for¨$\bbb\nn\dd\bs$ with associated partition as

1472: indicated

1473: $$\begin{array}{lccrccrccrc}

1474: (1,1,1,1):

1475: &\frac1{20}(18 + 7\sqrt6)

1476: &\hspace{-3mm}\ev_1^\dd

1477: &+

1478: &\frac1{20}(18-7\sqrt6)

1479: &\hspace{-3mm}\ev_2^\dd 

1480: &-  

1481: &\frac{256}5 

1482: &\hspace{-3mm}\cdot\, 2^\dd 

1483: &+ 

1484: &6k + 11,\\

1485: (2,1,1):

1486: &\frac1{60}(18 + 7\sqrt6)

1487: &\hspace{-3mm}\ev_1^\dd 

1488: &+

1489: &\frac1{60}(18-7\sqrt6)

1490: &\hspace{-3mm}\ev_2^\dd 

1491: &- 

1492: &\frac85

1493: &\hspace{-3mm}\cdot\, 2^\dd 

1494: &+ 

1495: &1,\\

1496: (2,2):

1497: &\frac1{12}(\sqrt6)

1498: &\hspace{-3mm}\ev_1^\dd  

1499: &- 

1500: &\frac1{12}(\sqrt6)

1501: &\hspace{-3mm}\ev_2^\dd,\\

1502: (3,1):

1503: &\frac1{60}(6 -\sqrt6)

1504: &\hspace{-3mm}\ev_1^\dd

1505: &+

1506: &\frac1{60}(6+\sqrt6)

1507: &\hspace{-3mm}\ev_2^\dd 

1508: &+ 

1509: &\frac45

1510: &\hspace{-3mm}\cdot\, 2^\dd 

1511: &-

1512: &1,\end{array}

1513: $$

1514: and¨$1$ for associated partition $(4)$, \ie, for $\bs =

1515: \D_4$. As the characteristic polynomial of¨$\Mattt4$ is

1516: $(x^2 - 6x + 3)(x - 2)(x-1)^2$,

1517: we can equivalently determine $\bbb4\dd\bs$ and $\bb4\dd$ by

1518: inductions on¨$\dd$ of the form

1519: \begin{equation} \label{E:Recur}

1520: \var_\dd = 6\var_{\dd-1} - 3\var_{\dd-2}

1521: + \alpha 2^\dd + \beta \dd + \gamma

1522: \end{equation}

1523: where $\alpha, \beta, \gamma$ are determined using special

1524: values of¨$\var_\dd$. For instance, $\bb4\dd$ is determined

1525: by¨\eqref{E:Recur} with $\alpha = 32$, $\beta = -12$,

1526: $\gamma = -34$ and the values $\var_{-1} = 0$, $\var_0= 1$. 

1527: Generating functions can be deduced easily.

1528: \end{exam}

1529: 

1530: \subsection{Eigenvalues of $\Mat\nn$}

1531: 

1532: By Proposition¨\ref{P:Expansion}, the value of¨$\bb\nn\dd$

1533: and¨$\bbb\nn\dd\bs$, and in particular its asymptotic

1534: behaviour when $\dd$ grows to infinity, are connected with

1535: the non-zero eigenvalues of¨$\Mat\nn$, which, by

1536: Proposition¨\ref{P:NNNumber}, coincide with those

1537: of¨$\Mattt\nn$. The characteristic polynomial

1538: of¨$\Mattt\nn$---hence of¨$\Mat\nn$ up to an

1539: $x^\dd$¨factor---for small values of¨$\nn$ is displayed in

1540: Table¨\ref{T:Charpol}.

1541: 

1542: \begin{table}[t]

1543: \begin{tabbing}

1544: \hspace{10mm}\=\hspace{85mm}\=\hspace{2cm}\kill

1545: \> $\charpol{\Mattt1} 

1546: = x - 1$\\

1547: \> $\charpol{\Mattt2} 

1548: = \charpol{\Mattt1}  \cdot  (x - 1)$\\

1549: \> $\charpol{\Mattt3} 

1550: = \charpol{\Mattt2}  \cdot  (x - 2)$\\

1551: \> $\charpol{\Mattt4} 

1552: = \charpol{\Mattt3}  \cdot  (x^2 - 6x +3)$\\

1553: \> $\charpol{\Mattt5} 

1554: =\charpol{\Mattt4}  \cdot  (x^2 - 20 x +24)$\\

1555: \> $\charpol{\Mattt6} 

1556: = \charpol{\Mattt5}

1557:  \cdot  (x^4 - 82 x^3 +359x^2 - 260x + 60)$\\

1558: \> $\charpol{\Mattt7} 

1559: = \charpol{\Mattt6} \cdot

1560: (x^4  - 390 x^3  + 6024 x^2 - 13680 x + 8640)$\\

1561: \> $\charpol{\Mattt8} 

1562: = \charpol{\Mattt7}

1563: \cdot (x^7  - 2134 x^6  + 139976 x^5 - 1321214 x^4

1564: + 3780975 x^3$\\

1565: \> \hspace{6cm}$- 3305160 x^2  + 1341900 x - 226800)$

1566: \end{tabbing}

1567: \begin{tabular}{c|c|c|c|c|c|c|c|c}

1568: $\nn$&1&2&3&4&5&6&7&8\\

1569: \hline

1570: \vline width0pt height10pt depth6pt

1571: $\ev_{max}(\Mat\nn)$&1&1&2&5.449 &18.717 

1572: &77.405 &373.990 &2066.575 \\

1573: $\displaystyle\frac{\ev_{max}(\Mat\nn)}{\nn\cdot\ev_{max}(\Mat\nno)}$

1574: &-&0.5& 0.667 &0.681 &0.687 &0.689 &0.690 &0.691\\ 

1575: \end{tabular}

1576: \bigskip

1577: \caption{\smaller Characteristic polynomial

1578: of¨$\Mattt\nn$ for $\nn \le 8$, and the

1579: corresponding largest eigenvalue---which is to be

1580: compared with¨$\nn!$, the growth rate

1581: for the number of $\nn$-braids of degree at

1582: most¨$\dd$ if all sequences were normal.}

1583: \label{T:Charpol}

1584: \end{table} 

1585: 

1586: These values support the following

1587: 

1588: \begin{conj} \label{C:Inclusion}

1589: For each¨$\nn$, the characteristic polynomial

1590: of¨$\Mat\nno$ divides that of¨$\Mat\nn$. More precisely,

1591: the sprectrum of¨$\Mattt\nn$ is the spectrum

1592: of¨$\Mattt\nno$, plus $p(\nn) - p(\nno)$ simple non-zero

1593: eigenvalues.

1594: \end{conj}

1595: 

1596: It is not hard to check the above statement for $\nn \le

1597: 10$. Specifically, let¨$\widehat M_\nn$ be the size

1598: $p(\nn)-2$ matrix obtained from¨$\Mattt\nn$ by deleting the

1599: first and the last rows and columns. For each small value

1600: of¨$\nn$, one can directly check that $\widehat M_\nn$ is

1601: similar to a matrix of the form $\left(\begin{matrix}

1602: \widehat{M}_\nno&0\\\dots&\dots\end{matrix}\right)$, and

1603: deduce the properties asserted in

1604: Conjecture¨\ref{C:Inclusion}. But ne generic argument is

1605: known so far.

1606: 

1607: The growth rate of the numbers¨$\bbb\nn\dd\bs$

1608: is connected with the largest eigenvalue

1609: $\ev_{max}(\Mat\nn)$ of¨$\Mat\nn$. For¨$\nn \le 6$,

1610: all¨$\bbb\nn\dd\bs$ except¨$\bbb\nn\dd{\D_\nn}$,

1611: which is¨$1$, and therefore all $\bb\nn\dd$ as

1612: well, grow like $\ev_{max}(\Mat\nn)^\dd$. 

1613: 

1614: \begin{ques}

1615: Do all¨$\bbb\nn\dd\bs$ except¨$\bbb\nn\dd{\D_\nn}$

1616: grow like $\ev_{max}(\Mat\nn)^\dd$?

1617: \end{ques}

1618: 

1619: \begin{ques}

1620: What is the asymptotic behaviour of $\ev_{max}(\Mat\nn)$

1621: with¨$\nn$?

1622: \end{ques}

1623: 

1624: The trivial upper bound of¨\eqref{E:Upper} suggests to

1625: compare $\ev_{max}(\Mat\nn)$ with¨$\nn!$, or, rather,

1626: $\ev_{max}(\Mat\nn)$ with¨$\nn \cdot \ev_{max}(\Mat\nno)$.

1627: 

1628: \section{Letting the degree vary}

1629: \label{S:Deg}

1630: 

1631: So far, we kept the braid index¨$\nn$ fixed, and studied

1632: how the numbers¨$\bb\nn\dd$ or¨$\bbb\nn\dd\bs$ vary

1633: with¨$\dd$, thus letting linear inductions appear. Quite

1634: different induction schemes appear when we fix the degree

1635: and let the braid index vary. No systematic method is

1636: known so far, and we only mention a few partial results

1637: motivated by the approach of¨\cite{Dhh}.

1638: 

1639: \subsection{The numbers¨$\bb\nn2$}

1640: 

1641: Very little is known about¨$\bb\nn\dd$ in general. The case

1642: $\dd = 1$ is trivial, as we already observed the equality 

1643: $$\bb\nn1 = \nn!.$$

1644: For $\dd = 2$, the value can be deduced from earlier results

1645: of¨\cite{CSV, CSV2} about permutations. We shall use the

1646: following very general observation about duality in Garside

1647: groups:

1648: 

1649: \begin{lemm} \label{L:Dual}

1650: For¨$\xx$ in¨$\DD\nn{}$, let $\dual\xx$

1651: and $\xx^*$ be defined by $\dual\xx\,\xx = \xx \,

1652: \xx^* = \D_\nn$. Then $\xx \mapsto \dual\xx$ and

1653: $\xx \mapsto \xx^*$ are permutations of¨$\DD\nn{}$,

1654: and, for each simple¨$\xx$, we have

1655: \begin{equation} \label{E:Dual}

1656: \DivR(\dual\xx) = \Int\nn \setminus \DivL(\xx)

1657: \mbox{\quad and \quad}

1658: \DivL(\xx^*) = \Int\nn \setminus \DivR(\xx).

1659: \end{equation}

1660: \end{lemm}

1661: 

1662: \begin{proof}

1663: Assume $\xx \in \DD\nn{}$. Then, by hypothesis,

1664: $\xx$ is a left and a right divisor of¨$\D_\nn$,

1665: hence $\dual\xx$ and $\xx^*$ are positive

1666: braids, and they are divisors of¨$\D_\nn$

1667: in¨$B_\nn^+$, so they are simple. That the

1668: mappings $\xx \mapsto \dual\xx$ and

1669: $\xx \mapsto \xx^*$ are injective is clear,

1670: and the surjectivity follows from the

1671: finiteness of¨$\DD\nn{}$.

1672: 

1673: Now, $\ss i$ being a right divisor of¨$\dual\xx$ is

1674: equivalent to $\dual\xx \ss i$ not being simple,

1675: hence to the non-existence of¨$\yy$ satisfying

1676: $\dual\xx \ss i \yy = \D_\nn$, and finally to the

1677: non-existence of¨$\yy$ satisfying $\xx = \ss i

1678: \yy$. This implies the first equality

1679: in¨\eqref{E:Dual}. The second equality follows from

1680: a symmetric argument.

1681: \end{proof}

1682: 

1683: \begin{prop}

1684: The numbers¨$\bb\nn2$ are determined by the

1685: induction

1686: \begin{equation}

1687: \bb02=1, \qquad 

1688: \bb\nn2 = \sum_{\ind=0}^\nno (-)^{\nn + \ind

1689: +1} {\nn \choose \ind}^2 

1690: \bb\ind2.

1691: \end{equation}

1692: Their double exponential generating function is 

1693: \begin{equation}

1694: \sum_{\nn = 0}^\infty \bb\nn2

1695: \frac{x^\nn}{\nn!^2} = 

1696: \bigg( \sum_{\nn = 0}^\infty (-1)^\nn

1697: \frac{x^\nn}{\nn!^2}

1698: \bigg)\inv = \frac1{J_0(\sqrt\xx)},

1699: \end{equation}

1700: where $J_0(\xx)$ is the Bessel function.

1701: \end{prop}

1702: 

1703: \begin{proof}

1704: By definition, $\bb\nn2$ is the number of pairs

1705: of simple¨$\nn$-braids $(\sx_1, \sx_2)$ satisfying

1706: $\DivR(\sx_1) \supseteq \DivL(\sx_2)$, \ie, by

1707: Lemma¨\ref{L:Dual},

1708: $\DivR(\sx_1) \cap \DivR(\dual\sx_2) = \emptyset$. By

1709: Lemma¨\ref{L:Exchange}, this number is also

1710: the number of pairs of permutations¨$(\ff, \gg)$

1711: in¨$\Sym\nn$ with no descent in common, \ie, such

1712: that there exists no¨$i$ satisfying both $\ff(i) >

1713: \ff(i+1)$ and $\gg(i) > \gg(i+1)$. Such pairs of

1714: permutations have been counted

1715: in¨\cite{CSV, CSV2} (see also¨\cite{Rio}),

1716: with the result indicated above.

1717: \end{proof}

1718: 

1719: \subsection{The numbers $\bbb\nn2{\D_{\nn - \rrr}}$}

1720: 

1721: Specific results appear when we consider the

1722: numbers¨$\bbb\nn\dd\bs$ with¨$\bs$ of the form¨$\D_{\nn

1723: - \rrr}$ with $1 \le \rrr \le \nn$. In particular, we can

1724: complete the computation when $\rrr$ is fixed and $\dd$ is

1725: small. We obviously have

1726: $\bbb\nn1{\D_{\nn - \rrr}} = 1$ for $1 \le \rrr \le \nn$, so

1727: the first case to consider is $\dd = 2$. The general

1728: principle that makes the computation of $\bbb\nn\dd{\D_{\nn

1729: - \rrr}}$ relatively easy is the following observation:

1730: 

1731: \begin{lemm} \label{L:Prelim}

1732: For all¨$\nn, \dd, \rrr$, we have

1733: \begin{equation} \label{E:Prelim}

1734: \bbb\nn\dd{\D_{\nn - \rrr}} = \sum_{\mbox{\Small\rm $\bs$ right divisible

1735: by¨$\D_{\nn - \rrr}$}} \bbb\nn{\dd-1}\bs.

1736: \end{equation}

1737: \end{lemm}

1738: 

1739: \begin{proof}

1740: The argument is similar to that for

1741: Proposition¨\ref{P:Total}. A sequence $(\bs_1,

1742: \dots, \bs_{\dd-1}, \D_{\nn-\rr})$ is normal if and only if

1743: both $(\bs_1, \dots, \bs_{\dd-1})$ and $(\bs_{\dd-1},

1744: \D_{\nn-\rr})$ are normal. Now $(\bs_{\dd-1}, \D_{\nn-\rr})$

1745: is normal if and only if every¨$\ss i$

1746: dividing¨$\D_{\nn-\rr}$ on the left divides¨$\bs_{\dd-1}$

1747: on the right. The $\ss i$'s dividing $\D_{\nn-\rr}$ on the

1748: left are $\ss1$, \dots, $\ss{\nn - \rr - 1}$. The simple

1749: braids that are right divisible by $\ss1$, \dots, $\ss{\nn -

1750: \rr - 1}$ are those right divisible by¨$\D_{\nn-\rr}$. Then

1751: \eqref{E:Prelim} follows.

1752: \end{proof}

1753: 

1754: \begin{prop}

1755: For $1 \le \rrr \le \nn$, we have

1756: \begin{equation} \label{E:K1}

1757: \bbb\nn2{\D_{\nn - \rrr}} = \frac{\nn!}{(\nn - \rrr)!}.

1758: \end{equation}

1759: \end{prop}

1760: 

1761: \begin{proof}

1762: By Lemma¨\ref{L:Prelim}, $\bbb\nn2{\D_{\nn - \rrr}}$ is

1763: the number of simple $\nn$-braids¨$\bs$ that are

1764: right divisible by¨$\D_{\nn - \rrr}$, \ie., that satisfy

1765: $\DivR(\bs) \supseteq \Int{{\nn - \rrr}}$. The block

1766: composition of¨$\Int{\nn - \rrr}$ in¨$\nn$ is $({\nn - \rrr}, 1,

1767: \dots, 1)$, so \eqref{E:Value} directly

1768: gives¨\eqref{E:K1}.

1769: \end{proof}

1770: 

1771: \subsection{The numbers $\bbb\nn3{\D_{\nn -\rrr}}$}

1772: 

1773: Things become more interesting for¨$\dd = 3$. 

1774: 

1775: \begin{prop} \label{P:K2}

1776: For $1 \le \rrr \le \nn$, there exist polynomials¨$P_1$, \dots, $P_{n-r}$

1777: with integer coefficients and $P_i$ of degree at most¨${\nn - \rrr} - i +

1778: 1$ such that, for every¨$\nn$, we have

1779: \begin{align} 

1780: \label{E:K2}

1781: \bbb\nn3{\D_{\nn-\rrr}} 

1782: &= (\nn - \rrr)!\,({\nn - \rrr}+1)^\nn + \sum_{i = 1}^{\nn -

1783: \rrr} P_i(\nn) \, i^{\rrr +i-1}.\\

1784: \intertext{The explicit values for $\rrr

1785: = 1, 2$ are}

1786: \label{E:K21}

1787: \bbb\nn3{\D_{\nno}} &= 2^\nno, \\

1788: \label{E:K22}

1789: \bbb\nn3{\D_{\nn-2}} &= 2 \cdot 3^\nn - (\nn + 6) \cdot

1790: 2^\nno + 1.

1791: \end{align}

1792: \end{prop}

1793: 

1794: \begin{proof}

1795: We begin with¨\eqref{E:K21}. By

1796: Lemma¨\ref{L:Prelim}, $\bbb\nn3{\D_{\nno}}$ is the sum of

1797: all $\bbb\nn2\bs$ with $\bs$ right divisible by¨$\D_{\nno}$, \ie, it is the number of normal sequences

1798: $(\sx_1, \sx_2)$ such that $\sx_2$ is right divisible

1799: by¨$\D_{\nno}$. Let $\SS$ be the set of all such

1800: normal sequences. We partition¨$\SS$ according to the

1801: value of¨$\DivL(\sx_2)$, \ie, for each subset¨$\II$

1802: of¨$\Int{\nno}$, we count how many pairs¨$(\sx_1,

1803: \sx_2)$ satisfy¨$\DivL(\sx_2) = \II$. So assume

1804: that $\sx_2$ is right divisible by¨$\D_{\nno}$. Two

1805: cases are possible. Either $\sx_2$ is right

1806: divisible by (hence equal to)¨$\D_\nn$, and then we

1807: have

1808: $\DivL(\sx_2) = \Int{\nno}$. Or

1809: $\sx_2$ is not divisible by¨$\D_{\nno}$, and then

1810: Lemma¨\ref{L:Sequence} shows that

1811: $\sx_2$ must be $\ss{i, \nn}\D_{\nno}$ for some¨$i$

1812: with $2 \le i \le \nn$, so that the

1813: block composition of¨$\DivL(\sx_2)$ is $(i-1,

1814: \nn - i+1)$. So the possible compositions

1815: for the set¨$\DivL(\sx_2)$ are¨$(\nn)$, and $(\ip,

1816: \nn - \ip)$ with $1 \le \ip \le \nno$. Conversely,

1817: the previous analysis shows that, for each¨$\II$ of

1818: the previous form, there exists exactly one

1819: possible¨$\sx_2$. Now, Proposition¨\ref{P:EES} says

1820: that there is one choice for¨$\sx_1$ in the case

1821: of¨$(\nn)$---namely $\sx_1 = \D_\nn$---and $\nn

1822: \choose \ip$ choices for¨$\sx_1$ in the case

1823: of¨$(\ip, \nn - \ip)$. We deduce

1824: $$\bbb\nn3{\D_{\nno}} = 1 + \sum_{\ip= 1}^\nno {\nn

1825: \choose{\ip}} = 2^\nno.$$

1826: 

1827: The method is similar for

1828: computing¨$\bbb\nn3{\D_{\nn-2}}$ in¨\eqref{E:K22}. Assume

1829: that

1830: $(\sx_1, \sx_2)$ is a normal sequence with¨$\sx_2$ right

1831: divisible by¨$\D_{\nn-2}$. The hypothesis is $\DivR(\sx_2) \supseteq

1832: \Int{\nn-3}$, so three cases may occur, namely $\DivR(\sx_2)

1833: \supseteq \Int{\nn-2}$, $\DivR(\sx_2) = \Int{\nn-3}$, and

1834: $\DivR(\sx_2) =  \Int{\nn-3} \cup \{\nno\}$. The first case was analysed

1835: above. In the second case, $\DivL(\sx_2)$ has three blocks, and,

1836: conversely, each set¨$\II$ with three blocks gives

1837: exactly one eligible¨$\sx_2$. In the third case,

1838: $\DivL(\sx_2)$ has either two blocks, or it has

1839: three blocks with the middle one at least¨$2$;

1840: conversely, each set¨$\II$ of the previous form

1841: gives one eligible¨$\sx_2$. Using as above

1842: Proposition¨\ref{P:EES} to count the

1843: eligible¨$\sx_1$'s for each possible¨$\II$, we

1844: obtain that $\bbb\nn3{\D_{\nn-2}}$ is

1845: \begin{equation} \label{E:Sum}

1846: \bbb\nn3{\D_{\nno}} 

1847: + \sum_{\overset{\scriptstyle\ip_1 + \ip_2 + \ip_3 =

1848: \nn} {\ip_1, \ip_2, \ip_3 \ge 1}}

1849: \frac{\nn!}{\ip_1! \ip_2! \ip_3!}

1850: + \sum_{\overset{\scriptstyle\ip_1 + \ip_2 =

1851: \nn} {\ip_1, \ip_2 \ge 1}}

1852: \frac{\nn!}{\ip_1! \ip_2!}

1853: + \sum_{\overset{\scriptstyle\ip_1 + \ip_2 + \ip_3 =

1854: \nn} {\ip_1, \ip_3 \ge 1,\ip_2 \ge 2}}

1855: \frac{\nn!}{\ip_1! \ip_2! \ip_3!}.

1856: \end{equation}

1857: Using the fact that $3^\nn$ is the sum of all

1858: $\frac{\nn!}{\ip_1! \ip_2! \ip_3!}$ with $\ip_1

1859: + \ip_2 + \ip_3 = \nn$, one deduces¨\eqref{E:K22}

1860: by bookkeeping.

1861: 

1862: Applying the same method in the general case leads

1863: to¨\eqref{E:K2}. Indeed, always by

1864: Lemma¨\ref{L:Sequence}, specifying a simple

1865: $\nn$-braid¨$\sx_2$ satisfying¨$\DivR(\sx_2)

1866: \supseteq \Int{\rrr -1}$ amounts to choosing a

1867: permutation of the ${\nn - \rrr}$¨last strands and the

1868: ${\nn - \rrr}$¨positions $(i_1, \dots, i_{\nn - \rrr})$ where these

1869: strands start from. In the generic case, the

1870: resulting set¨$\DivL(\sx_2)$ is $\{i_1 - 1, \dots,

1871: i_{\nn - \rrr} - 1\}$, whose composition consists of

1872: ${\nn - \rrr}$¨blocks. The special cases are when at least

1873: two adjacent strands among the last ${\nn - \rrr}$¨ones

1874: start from adjacent positions; according to

1875: whether these strands cross or not in the final

1876: part, one then obtains either a composition with a

1877: block of size¨2 at least, or a composition with

1878: less than ${\nn - \rrr}$¨blocks. Conversely, for every

1879: subset¨$\II$ of¨$\Int{\nno}$ with ${\nn - \rrr}$¨blocks,

1880: there exists in general $(\nn - \rrr)!$¨eligible¨$\sx_2$'s,

1881: one for each choice of the final permutation of the

1882: last $\nn - \rrr$¨strands. There may be less than

1883: $(\nn - \rrr)!$¨choices for¨$\sx_2$ when $1$ occurs in the

1884: composition of¨$\II$. Also, subsets

1885: of¨$\Int{\nno}$ with fewer than¨${\nn - \rrr}$ blocks

1886: may lead to eligible¨$\sx_2$'s. Multiplying by the

1887: number of eligibles¨$\sx_1$'s for each¨$\II$ and

1888: summing up yields an expression similar

1889: to¨\eqref{E:Sum}, involving $(\nn - \rrr)!$¨sums of the form

1890: $\sum_{\ip_1 + \dots + \ip_{{\nn - \rrr}+1} = \nn}

1891: \frac{\nn!}{\ip_1! \dots \ip_{{\nn - \rrr}+1}!}$ with

1892: possible order constraints on¨$\ip_1$, \dots,

1893: $\ip_{{\nn - \rrr}+1}$. Each of them leads to a

1894: factor¨$({\nn - \rrr}+1)^\nn$, plus additional factors

1895: corresponding to specializing arguments to¨$0$

1896: or¨$1$ or to grouping them.

1897: \end{proof}

1898: 

1899: \subsection{The numbers $\bbb\nn4{\D_{\nno}}$}

1900: 

1901: For $\dd = 4$, it seems hopeless to complete the computation

1902: of¨$\bbb\nn\dd{\D_{\nn-\rr}}$. However, this can be done

1903: for¨$\rrr = 1$. The remarkable point is that still another

1904: induction scheme appears.

1905: 

1906: \begin{prop}

1907: For¨$\nn \ge 1$, we have

1908: \begin{equation} \label{E:KKK3}

1909: \bbb\nn4{\DD{\nno}{}} = \sum_{\ind=0}^\nno

1910: \frac{\nn!}{\ind!}.

1911: \end{equation}

1912: \end{prop}

1913: 

1914: \begin{proof}

1915: According to Lemma¨\ref{L:Prelim} again, we have

1916: now to count the normal sequences

1917: $(\sx_1, \sx_2, \sx_3)$ with¨$\sx_3$ of the

1918: form¨$\ss{i, \nn}\D_{\nno}$, $2 \le i \le \nn$.

1919: We partition the family according to the value¨$\II$

1920: of¨$\DivL(\sx_2)$, and count how many sequences may

1921: correspond to a given¨$\II$. Let $(\ip_1, \dots,

1922: \ip_\ir)$ denote the block composition of¨$\II$.

1923: 

1924: Let us first consider the case $\II =

1925: \Int{\nno}$. Then we must have $\sx_2 = \D_\nn$,

1926: hence $\sx_1 = \D_\nn$ as well. There are

1927: $\nn$¨possible choices for¨$\sx_3$, and the total

1928: number of corresponding sequences¨$(\sx_1, \sx_2,

1929: \sx_3)$ is¨$\nn$.

1930: 

1931: We assume now $\II \not= \Int{\nno}$, \ie, $\ir

1932: \ge 2$. As for¨$\sx_1$, Proposition¨\ref{P:EES} directly

1933: gives the number of choices, namely

1934: $\frac{\nn!}{\ip_1! \dots \ip_\ir!}$. So we are left

1935: with counting how many pairs¨$(\sx_2, \sx_3)$ are

1936: eligible. The case $\sx_3 = \D_n$ is excluded

1937: since it implies $\sx_2 = \D_\nn$ hence $\II =

1938: \Int{\nno}$. As in the case of¨$\bbb\nn3{\D_{\nno}}$,

1939: the hypothesis that $\sx_3$ is $\ss{i, \nn}

1940: \D_{\nno}$ for some¨$i$ with $2 \le i \le \nn$

1941: implies that the block composition

1942: of¨$\DivL(\sx_3)$ consists of two nonempty blocks,

1943: and, conversely, each partition of¨$\Int\nn$ into

1944: two nonempty blocks gives a unique¨$\sx_3$ of the

1945: convenient form. So the number of pairs¨$(\sx_2,

1946: \sx_3)$ associated with¨$\II$ is the number

1947: of¨$\sx_2$'s satisfying

1948: $\DivL(\sx_2) = \II$ and such that $\DivR(\sx_2)$

1949: has two blocks.

1950: 

1951: By¨\eqref{E:Sequence}, this number is the

1952: number of functions¨$\ff$ of¨$\Int\nn$ to¨$\{1,

1953: 2\}$ such that $\ff(i) < \ff(i+1)$ holds exactly for

1954: $i \notin \II$. As only two values are possible,

1955: this condition means that we have $\ff(i) = 1$ and

1956: $\ff(i+1) = 2$ for¨$i \notin \II$, and

1957: $\ff(i+1) \le \ff(i)$ for¨$i \in \II$. Consider

1958: the blocks of¨$\II$. In each block,

1959: except possibly the first and the last ones, the

1960: value of¨$\ff$ has to be¨$2$ on the first element,

1961: and to be¨$1$ on the last element. Inbetween, $\ff$

1962: is non-increasing. So the values consist of a series

1963: of¨$2$'s, followed by a series of¨$1$'s. The

1964: only parameter to specify is the position

1965: where¨$\ff$ switches from¨$2$ to¨$1$, so, for a

1966: block of size¨$\ip$, there are $\ip-1$ possible

1967: choices (see Figure¨\ref{F:Function}). The cases of

1968: the first and the last blocks are special, because

1969: there is no constraint on the left for the first

1970: block, and on the right for the last block. So, in

1971: these special cases, there are

1972: $\ip$¨choices instead of¨$\ip-1$. The conclusion is

1973: that, for¨$\II$ of block composition $(\ip_1, \dots,

1974: \ip_\ir)$, there are

1975: $\ip_1(\ip_2-1) \dots (\ip_{\ir-1}-1) \ip_\ir$

1976: choices for the pairs¨$(\sx_2, \sx_3)$ associated

1977: with¨$\II$. Merging the result for¨$\sx_1$ and

1978: for¨$(\sx_2, \sx_3)$ and summing up over¨$\II$

1979: gives

1980: \begin{equation} \label{E:K3}

1981: %\sum_{\underset{\scriptstyle p_1, \dots, p_r \ge 1}{p_1 +

1982: %\cdots + p_r = n}} 

1983: \bbb\nn4{\DD{\nno}{}} = \sum

1984: \frac{\nn!}{\ip_1! \dots \ip_\ir!} \,  

1985: \ip_1(\ip_2-1)

1986: \dots (\ip_{\ir-1}-1)\ip_\ir.

1987: \end{equation}

1988: the sum being taken over all finite compositions

1989: $(\ip_1, \dots, \ip_\ir)$ of¨$\nn$: indeed, the

1990: value for¨$\II = \Int{\nno}$, namely¨$\nn$,

1991: corresponds to the missing term $\frac{\nn!}{\nn!}

1992: \nn$ of the sum.

1993: 

1994: \begin{figure} [htb]

1995: \begin{picture}(82,20)(0, 0)

1996: \put(0,3){\includegraphics{Function.eps}}

1997: \put(15,0){$\ip_1$}

1998: \put(38,0){$\ip_2$}

1999: \put(57,0){$\ip_3$}

2000: \put(72,0){$\dots$}

2001: \end{picture}

2002: \caption{\smaller Proof of¨\eqref{E:K3}: the

2003: rises are fixed, so it just remains to choose the

2004: position of the fall in each block of¨$\II$,

2005: whence $\ip-1$ choices for a size¨$\ip$ block

2006: except the first and the last ones.}

2007: \label{F:Function}

2008: \end{figure}

2009: 

2010: We can now simplify the right hand term

2011: in¨\eqref{E:K3}. To this end, we observe that

2012: \begin{equation} \label{E:KK3}

2013: \sum_{\underset{\scriptstyle \ip_1, \dots, \ip_\ir \ge 1}{\ip_1 +

2014: \cdots + \ip_\ir = \ind + 1}}

2015: \frac{\ip_1-1}{\ip_1!} \dots \frac{p_{r-1}-1}{p_{r-1}!} \,

2016: \frac{\ip_\ir}{\ip_\ir!} = 1

2017: \end{equation} 

2018: holds for¨$\ind \ge 0$. Indeed, let $F(\ind)$ be the

2019: left hand side of¨\eqref{E:KK3}. We

2020: prove¨\eqref{E:KK3} using induction on¨$\ind$. For

2021: $\ind=0$, we get $1=1$. Assume $\ind \ge 1$ and

2022: consider the sequences

2023: $(\ip_1, \dots, \ip_\ir)$ satisfying $\ip_1 + \cdots + 

2024: \ip_\ir = \ind + 1$.  On the one hand, we have

2025: $(\ind + 1)$, whose contribution to¨$F(\ind)$ is

2026: $\frac{1}{\ind!}$. On the other hand, we have

2027: the sequences of length at least¨$2$. Now, for

2028: each¨$\ip$ with $1 \le \ip \le \ind$, the

2029: contribution of $(\ip, \ip_2, \dots, \ip_\ir)$

2030: to¨$F(\ind)$ is

2031: $\frac{\ip-1}{\ip!}$ times the contribution

2032: of $(\ip_2, \dots, \ip_\ir)$ to¨$F(\ind - \ip)$.

2033: Hence the total contribution of the sequences

2034: beginning with¨$\ip$ to¨$F(\ind)$ is

2035: $\frac{\ip-1}{\ip!}

2036: \, F(\ind - \ip)$, so, by induction hypothesis, it

2037: is $\frac{(\ip-1)}{\ip!}$. We deduce

2038: $F(\ind) = \frac{0}{1!} + \frac{1}{2!} + \cdots +

2039: \frac{\ind -1}{\ind!} + \frac1{\ind!}$,  

2040: which is clearly¨$1$. 

2041: 

2042: Consider now the right hand side in¨\eqref{E:K3}.

2043: For

2044: $0 \le \ind < \nno$, the contribution of $(\ind+1,

2045: \ip_2, \dots, \ip_\ir)$ to the sum is

2046: $\frac{\nn!}{\ind!}$ times the  quantity

2047: $\frac{\ip_1-1}{\ip_1!} \,

2048: \frac{\ip_{\ir_1}-1}{\ip_{\ir-1}!}

2049: \, \frac{\ip_\ir}{\ip_\ir!}$ involved

2050: in¨\eqref{E:KK3}. Using the latter equality, we

2051: deduce that the total contribution of the sequences

2052: beginning with¨$\ind + 1$ is

2053: $\frac{\nn!}{\ind!}$. As for¨$\ind = \nno$, the

2054: contribution of¨$(\nn)$ to the right

2055: hand side in¨\eqref{E:K3} is¨$\nn$, which is

2056: $\frac{\nn!}{(\nno)!}$, so the general formula

2057: remains valid. By summing over¨$\ind$, we

2058: obtain¨\eqref{E:KKK3}.

2059: \end{proof}

2060: 

2061: \begin{coro}

2062: The numbers $\bbb\nn4{\DD{\nno}{}}$ are

2063: determined by the induction

2064: $$\var_1 = 1, \qquad

2065: \var_\nn = \nn \var_{\nno} +

2066: 2\nno.$$

2067: \end{coro}

2068: 

2069: Another consequence of¨\eqref{E:KKK3} is the equality

2070: $$\bbb\nn4{\DD{\nno}{}} = \lfloor \nn! e \rfloor - 1,$$

2071: with $e = \exp(1)$.

2072: 

2073: \begin{thebibliography}{99}

2074: 

2075: \Reff Adj; S.I.\,Adyan; Fragments of the word Delta in a

2076: braid group; Mat. Zam. Acad. Sci. SSSR; 36-1; 1984;

2077: 25--34; translated Math. Notes of the Acad. Sci. USSR;

2078: 36-1 (1984) 505--510.

2079: 

2080: \Ref Bes; D. Bessis; The dual braid monoid; An. Sci. Ec.

2081: Norm. Sup.; 36; 2003; 647--683.

2082: 

2083: \Ref Bir; J.\,Birman; Braids, links, and mapping 

2084: class groups; Annals of Math. Studies 82, Princeton

2085: Univ.  Press  (1975). 

2086: 

2087: \Reff BKL; J.~Birman, K.H. Ko \& S.J. Lee; A new approach

2088: to the word problem in the braid groups; Advances in

2089: Math.; 139-2; 1998; 322-353.

2090:    

2091: \Ref Bro; A.\,Bronfman; Growth function of a class of

2092: monoids; Preprint (2001).

2093: 

2094: \Reff CSV; L.\,Carlitz, R.\,Scoville \&

2095: T.\,Vaughan; Enumeration of pairs of permutations

2096: and sequences; Bull. Amer. Math. Soc.; 80;

2097: 1974;  881--884.

2098: 

2099: \Reff CSV2; L.\,Carlitz, R.\,Scoville \&

2100: T.\,Vaughan; Enumeration of

2101: pairs of permutations; Discrete Math.; 14;

2102: 1976; 215--239.

2103: 

2104: \Reff  Chb; R.\,Charney; Geodesic automation and

2105: growth functions for Artin groups of finite type; Math.

2106: Ann.; 301-2; 1995; 307--324.

2107: 

2108: \Ref CrH; C.R.\,Cromwell \& S.\,Humphries; Counting

2109: fundamental paths in $2$-generator Artin semigroups;

2110: Preprint (2004).

2111: 

2112: \Reff Dgk; P.\,Dehornoy; Groupes de Garside; 

2113: Ann. Scient. Ec. Norm. Sup.; 35; 2002; 267--306.

2114: 

2115: \Ref Dgr; P.\,Dehornoy, I. Dynnikov, D. Rolfsen,

2116: B. Wiest; Why are braids orderable?; Panoramas \&

2117: Synth\`eses vol. 14, Soc. Math. France (2002).

2118: 

2119: \Ref Dhh; P.\,Dehornoy; Still another approach to the

2120: braid ordering; arXiv: math.GR/0506495.

2121: 

2122: \Reff Dfx; P.\,Dehornoy \& L.\,Paris; Gaussian

2123: groups and Garside groups, two generalizations of Artin

2124: groups; Proc. London Math. Soc.; 79-3; 1999; 569--604.

2125: 

2126: \Reff Dlg; P. Deligne; Les immeubles des groupes de

2127: tresses g\'en\'eralis\'es; Invent. Math.; 17; 1972;

2128: 273--302.

2129: 

2130: \Reff ElM; E.A.\,Elrifai \& H.R.\,Morton; Algorithms for

2131: positive braids; Quart. J. Math. Oxford; 45-2; 1994;

2132: 479--497.

2133: 

2134: \Ref Eps; D.\,Epstein, J.\,Cannon, D.\,Holt, S.\,Levy,

2135: M.\,Paterson \& W.\,Thurston; Word Processing in Groups;

2136: Jones \& Bartlett Publ. (1992).

2137: 

2138: \Reff Eps2; D.\,Epstein, A.\,Iano-Fletscher, \& U.\,Zwick;

2139: Growth functions and automatic groups; Experiment. Math; 5;

2140: 1996; 297--315.

2141: 

2142: \Reff Gar; F.A.\,Garside; The braid group and

2143: other groups; Quart. J. Math. Oxford; 20-78; 1969;

2144: 235--254.

2145: 

2146: \Ref Geb; V.\,Gebhardt; A new approach to the conjugacy

2147: problem in Garside groups; J. of Algebra, to appear;

2148: math.GT/0306199.

2149: 

2150: \Ref Hoh; C.\,Hohlweg; Properties of the Solomon algebra

2151: homomorphism; arXiv: math.RT/0302309.

2152: 

2153: \Ref KoL; K.H.\,Ko, S.\,Lee, J.H.\,Cheon, J.W.\,Han,

2154: J.\,Kang, C.\,Park; New public-key cryptosystem using

2155: braid groups; Crypto 2000, 166--184.

2156:   

2157: \Ref McC; J.\,Mc Cammond; An introduction to

2158: Garside structures; Preprint (2005).

2159: 

2160: \Reff Rio; J.\,Riordan; Inverse relations and

2161: combinatorial identities; Amer. Math. Monthly; 

2162: 71; 1964; 485--498.

2163: 

2164: \Reff Sol; L.\,Solomon; A Mackey formula in the

2165: group ring of a Coxeter group; J. Algebra;

2166: 41; 1976; 255--268.

2167: 

2168: \Reff Xua; P.\,Xu; Growth of the positive braid

2169: groups; J. Pure Appl. Algebra; 80; 1992;

2170: 197--215.

2171: 

2172: \end{thebibliography}

2173: 

2174: \end{document}

2175: 

2176: