1: % This is LATEX

2: % using the style file of G&T: http://www.maths.warwick.ac.uk/gt

3: %

4: \documentclass{gtart}

5: \usepackage{amssymb, enumerate, epsfig}

6: %\input epsf

7: %\usepackage{pictex}

8: %

9: \def\<{\langle}

10: \def\>{\rangle}

11: \def\implies{\Longrightarrow}

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

13: \def\basl{\backslash}

14: \def\Q{{\mathbb Q}}

15: \def\C{{\mathbb C}}

16: \def\N{{\mathbb N}}

17: \def\Z{{\mathbb Z}}

18: \def\R{{\mathbb R}}

19: \def\bH{{\mathbb H}}

20: \def\MCG{\mathcal{MCG}}

21: \def\marg#1{{\marginpar{\tiny #1}}}

22: \def\calG{\mathcal{G}}

23: \def\pA{pseudo-Anosov}

24: \def\BnD{B_n / \langle \Delta^2 \rangle}

25: %\def\Bp{B_\bigcirc}

26: %\def\Bpp{B_\odot}

27: %\def\Dp{D_\bigcirc}

28: %\def\Dpp{D_\odot}

29: \def\tld{\widetilde}

30: \renewcommand{\phi}{\varphi}

31: %

32: \newtheorem{lemma}{Lemma}[section]

33: \newtheorem{theorem}[lemma]{Theorem}

34: \newtheorem{prop}[lemma]{Proposition}

35: \newtheorem{cor}[lemma]{Corollary}

36: \newtheorem{definition}[lemma]{Definition}

37: \newtheorem{lem}[lemma]{Lemma}

38: \newtheorem{conj}[lemma]{Conjecture}

39: \newtheorem{remark}[lemma]{Remark}

40: \newtheorem{example}[lemma]{Example}

41: \newtheorem{question}[lemma]{Question}

42: %

43: 

44: \begin{document}

45: \title{On the structure of the centralizer of a braid}

46: \author{Juan Gonz\'alez-Meneses\footnote{Partially supported by MCYT,

47: BFM2001-3207 and FEDER.} and Bert Wiest}

48: \address{Dpto.~de Matem\'atica Aplicada I, E.T.S.~Arquitectura, Universidad

49: de Sevilla, Avda.~Reina Mercedes, 2. 41012 Sevilla, Spain;

50: {\tt meneses@us.es}\\

51: and\\

52: IRMAR (UMR 6625 du CNRS), Universit\'e de Rennes 1, Campus de Beaulieu,

53: \\ 35042 Rennes cedex, France; {\tt bertold.wiest@math.univ-rennes1.fr}}

54: 

55: \begin{abstract}

56: The mixed braid groups are the subgroups of Artin braid groups whose elements preserve a

57: given partition of the base points. We prove that the centralizer of any braid can be

58: expressed in terms of semidirect and direct products of mixed braid groups. Then we

59: construct a generating set of the centralizer of any braid on $n$ strands, which has at most

60: $\frac{k(k+1)}{2}$ elements if $n=2k$, and at most $\frac{k(k+3)}{2}$ elements if $n=2k+1$.

61: These bounds are shown to be sharp, due to work of N.V.Ivanov and of S.J.Lee. Finally, we

62: describe how one can explicitly compute this generating set.

63: \end{abstract}

64: \keywords{braid, centralizer, Nielsen-Thurston theory.}

65: \primaryclass{20F36}\secondaryclass{20E07, 20F65.}

66: %\maketitlepage

67: \makeshorttitle

68: 

69: % ------------------------------------------------------------------

70: 

71: \section{Introduction and statement of the results}\label{S:IN}

72: 

73: In 1971, Makanin \cite{Makanin} gave an algorithm for computing a

74: generating set of the centralizer $Z(\beta)$ of any given element $\beta$

75: of the $n$-string braid group $B_n$. His method, however, tends to yield

76: very large, and highly redundant generating sets.

77: %

78: One hint that much smaller generating sets could be found came from the experimental results

79: of Gonz\'alez-Meneses and Franco, which were obtained with a radically improved version of

80: Makanin's algorithm, based on new theoretical work \cite{GMFr}. Also, it has probably been

81: clear to specialists for a long time that Nielsen-Thurston theory could be used to improve

82: upon Makanin's results. However, there seems to be no such result in the literature, and the

83: aim of the present paper is to fill this gap.

84: 

85: Although our main interest was to compute, for any given $\beta\in B_n$,

86: a small generating set of $Z(\beta)$, we succeed in describing this

87: centralizer in terms of semidirect and direct products of

88: {\em mixed braid groups} (see~\cite{Manfredini,Orevkov}). These groups

89: are defined as follows: let $X=\{P_1,\ldots,P_n\}$ be the base points of

90: the braids in $B_n$. Given a partition ${\cal P}$ of $X$, the mixed braid

91: group $B_{\cal P}$ consists of those braids whose associated permutation

92: preserves each coset of ${\cal P}$.

93: 

94: The well known classification of mapping classes of a punctured surface

95: into periodic, reducible and pseudo-Anosov ones, yields an analogous

96: classification for braids. If $\beta$ is reducible, then one can decompose

97: it, in a certain sense, into a {\em tubular braid} $\widehat{\beta}$, and

98: some {\em interior braids} $\beta_{[1]},\ldots, \beta_{[t]}$, all of

99: them having less than $n$ stands. The main result of this paper is the

100: following:

101: 

102: \begin{theorem}\label{T:main1}

103: Let $\beta\in B_n$. One has:

104: \begin{enumerate}

105: \item If $\beta$ is pseudo-Anosov , then $Z(\beta)\simeq \mathbb{Z}^2$.

106: 

107: \item If $\beta$ is periodic, then $Z(\beta)$ is either $B_n$ or isomorphic

108: to a braid group on an annulus.

109: 

110: \item If $\beta$ is reducible, then there exists a split exact sequence:

111: $$

112: 1 \longrightarrow Z(\beta_{[1]})\times \cdots \times Z(\beta_{[t]})

113: \longrightarrow Z(\beta) \longrightarrow Z_0(\widehat{\beta})\longrightarrow 1,

114: $$

115: where $Z_0(\widehat{\beta})$ is a subgroup of $Z(\widehat{\beta})$,

116: isomorphic either to $\Z^2$ or to a mixed braid group.

117: \end{enumerate}

118: \end{theorem}

119: 

120: Notice that $\mathbb{Z}\simeq B_2=B_{\{\{1,2\}\}}$, also

121: $B_n=B_{\{\{1,\ldots,n\}\}}$, and finally the braid group over an annulus

122: on $k$ strands is isomorphic to $B_{\{\{1,\ldots,k\},\{k+1\}\}}\subset

123: B_{k+1}$. Hence all these groups can be seen as mixed braid groups. Then,

124: by recurrence on the number of strands we deduce the following:

125: 

126: \begin{cor}

127:  For every $\beta\in B_n$, the centralizer $Z(\beta)$ can be expressed in

128: terms of semidirect and direct products of mixed braid groups.

129: \end{cor}

130: 

131: Using the above structure we shall construct, for any braid $\beta\in B_n$,

132: a generating set of $Z(\beta)$ having very few elements. More precisely,

133: we obtain:

134: 

135: \begin{theorem}\label{T:main}

136: If $\beta\in B_n$, then the centralizer $Z(\beta)$ can

137: be generated by at most $\frac{k(k+1)}{2}$ elements if $n=2k$,

138: and at most $\frac{k(k+3)}{2}$ elements if $n=2k+1$.

139: \end{theorem}

140: 

141: We will present an example, communicated to us by S. J. Lee, showing that

142: the above bound is sharp.

143: That is, we will define, for every positive

144: integer $n$, a braid in $B_n$ whose centralizer cannot be generated by

145: less than $\frac{k(k+1)}{2}$ elements if $n=2k$, or less than

146: $\frac{k(k+3)}{2}$ elements if $n=2k+1$.

147: (The first to observe that the number of generators of the centralizer

148: may grow quadratically with the number of strands was N.V.Ivanov

149: \cite{IvanovTalk}.)

150: 

151: However, the above bound refers to the worst case, and one could be interested

152: in the minimal number of generators of a particular braid. We shall give

153: a generating set which is in some sense the smallest ``natural'' generating

154: set for the centralizer of a braid. However, we shall also give an example

155: that illustrates the difficulty of finding the absolutely minimum possible

156: number of generators.

157: 

158: Let us mention that, for the special case of reducible braids conjugated

159: to a generator $\sigma_i$, its centralizer has already been described

160: in~\cite{FRZ}.

161: 

162: The plan of the paper is as follows: in section \ref{S:NiTh} we set up

163: notation and some standard machinery, and give the mentioned example by

164: S.\ J.\ Lee. In section~\ref{S:per} we study $Z(\beta)$ in the case where

165: $\beta$ is periodic, section~\ref{S:pA} deals with the {\pA} case, and

166: section~\ref{S:red} the reducible one, which is the most involved. In

167: section~\ref{S:upperbound} we define a generating set which is no larger

168: than the stated upper bound. In section~\ref{S:minimal} we describe a

169: generating set which is as small as possible while still reflecting the

170: geometric structure of the Nielsen-Thurston decomposition. We also give an

171: example to show that by algebraic trickery, even smaller sets can be obtained.

172: Finally in section~\ref{S:alg} we discuss how the generating set that

173: we defined can be found algorithmically.

174: 

175: % ------------------------------------------------------------------

176: 

177: \section{Prerequisites from Nielsen-Thurston theory}\label{S:NiTh}

178: 

179: We denote by $D$ the closed disk of radius $2$ centered at $0$ in the

180: complex plane. For any $n\in \N$, the disk $D$, together with any choice

181: of $n$ distinct points in its interior, is denoted $D_n$, and the

182: distinguished points are called the {\sl punctures}. We shall use

183: different choices for the exact position of the punctures at different

184: times - they may be lined up on the real axis, or regularly distributed on

185: a circle of radius $1$, or again one of them may be in the centre while

186: the remaining $n-1$ are distributed over the circle of radius $1$. In most

187: instances, the position of the punctures is irrelevant, and we shall leave

188: it unspecified.

189: 

190: We recall that the braid group $B_n$ is the group of isotopy classes of

191: homeomorphisms fixing (pointwise) the boundary and permuting the punctures

192: of $D_n$. Here the isotopies must fix pointwise the boundary and the punctures.

193: Alternatively, $B_n$ could be defined as the group of isotopy

194: classes of disjoint movements of the punctures, starting and ending with

195: the configuration of $D_n$. Yet another definition of $B_n$ is as the set

196: of isotopy classes of braids with $n$ strings in the cylinder

197: $D\times [0,1]$, where the start and end points of the strings are exactly

198: the puncture points in $D_n\times \{0\}$ and $D_n\times \{1\}$. We shall

199: use all three points of view.

200: 

201: We shall often work with a certain quotient of the group $B_n$, rather

202: than with $B_n$ itself.

203: We recall that the center of $B_n$ is isomorphic to the integers,

204: and generated by the full twist $\Delta^2$ (where $\Delta$ is Garside's

205: half twist). Geometrically, the group projection

206: $B_n \to B_n/\langle \Delta^2 \rangle$ is given by smashing the boundary

207: curve of $D_n$ to a puncture, so that $\BnD$ is naturally a subgroup

208: of the mapping class group of the sphere with $n+1$ punctures.

209: In order to keep notation manageable, we shall use the same letters for

210: elements of the braid group $B_n$ and for their image in the quotient $\BnD$.

211: This abuse of notation should not cause confusion.

212: 

213: We say that an element $\beta\in B_n$

214: is {\sl periodic} if the element of $\BnD$ represented by

215: $\beta$ is of finite order. Equivalently, $\beta$ is periodic if there

216: exists a $k\in \N$  such that in $B_n$ we have that $\beta^k$ is equal

217: to some power of $\Delta^2$.

218: 

219: We say an element $\beta$ of $B_n$ is {\sl reducible} if there exists a

220: nonempty multicurve $C$ in $D_n$ (i.e.~a system of disjoint simple closed

221: curves in $D_n$, none of them isotopic to the boundary or enclosing a

222: single puncture)

223: which is stabilized by $\beta$, i.e.~such that $\beta(C)$ is isotopic to $C$.

224: Note that $\beta$ may permute different components of the multicurve $C$.

225: 

226: The following definition is taken from \cite{BLM} (see also \cite{Ivanov}).

227: To every reducible braid $\beta\in B_n$ one can associate a canonical

228: invariant multicurve: its {\sl canonical  reduction system}, which by

229: definition is the collection of all isotopy classes $c$ of simple closed

230: curves which have the following two properties:

231: firstly, $c$ must be stabilized by some power of $\beta$, and secondly

232: any simple closed curve which has non-zero geometric intersection number

233: with $c$ must {\it not} be stabilized by any power of $\beta$. For instance,

234: let us

235: consider the punctured disk $D_6$, where the 6 punctures are arranged

236: uniformly on the circle of radius $1$ around $0$. Then the

237: rotation of the punctures around the circle by an angle of $\frac{2\pi}{3}$

238: is a periodic element of $B_6$ (of period $3$), it is

239: also reducible (e.g.~the three simple closed curves encircling punctures

240: 1 and 2, 3 and 4, and 5 and 6 respectively form an invariant multicurve),

241: but its canonical  reduction system is empty. This example, however, is

242: somewhat untypical: if a {\it non-periodic} braid is reducible, then

243: its canonical reduction system is nonempty (see \cite{Ivanov}).

244: 

245: If $C$ is an invariant multicurve of a reducible braid $\beta$,

246: then we define the {\sl tubular braid} induced by $\beta$ and $C$ to be the

247: braid on fewer strings obtained from $\beta$ by removing from $D_n$ all the

248: disks bounded by outermost curves of $C$, and collapsing each outermost

249: curve of $C$ to a puncture point. It should be stressed that this braid

250: is only defined up to conjugacy.

251: 

252: An alternative way to look at the same defintion is the following:

253: let us consider again $\beta$ as an isotopy class of $n$ disjoint strings

254: in $D\times [0,1]$ with extremal points at the puncture points of

255: $D_n\times\{0\}$ and $D_n\times \{1\}$, such that each disk

256: $D\times \{t\}$ intersects each string exactly once.

257: Now our picture can be completed by embedded cylinders in

258: $D\times [0,1]$ which are disjoint from each other and from the strings

259: of the braid, each of which intersects each disk $D\times \{t\}$ in exactly

260: one circle, and whose boundary components are exactly the outermost curves

261: of $C$ in $D\times \{0\}$ and $D\times \{1\}$. We can interpret the solid

262: cylinders bounded by these cylinders as ``fat strings'', and the resulting

263: braid with some fat strings is exactly the tubular braid defined above.

264: 

265: The {\em interior} braids induced by $\beta$ and $C$ are the braids on

266: fewer strings induced by $\beta$ at the interior of the discs bounded by

267: the outermost curves of $C$. They can be thought of as the braids `inside'

268: the tubes of the tubular braid. Therefore, for every reducible braid

269: $\beta$, and every invariant multicurve $C$, we can decompose $\beta$ into

270: one tubular braid and some interior braids -- as many as the number of

271: outermost curves in $C$.

272: 

273: Finally, we have the notion of a {\sl \pA} element of $B_n$, for which we

274: refer to \cite{FLP} or \cite{Ivanov}. Roughly speaking, $\beta\in B_n$

275: is {\pA} if it is represented by a homeomorphism of $D_n$ which preserves

276: two transverse measured foliations on $D_n$ (called the ``stable'' and the

277: ``unstable'' foliation), while scaling the measure of the unstable one by

278: some factor $\lambda$ which is greater than 1, and the measure of the stable

279: one by $\frac{1}{\lambda}$.

280: 

281: Thurston's theorem \cite{Thclass,FLP} states that every irreducible

282: element of $B_n$ is either periodic or pseudo-Anosov.

283: 

284: We end this section with the promised example, due to S. J. Lee, that should

285: be helpful for understanding the relationship between the Nielsen-Thurston

286: decomposition and the centralizer subgroup of a braid $\beta\in B_n$.

287: This example was also found independently by N. V. Ivanov and

288: H. Hamidi-Tehrani~\cite{IvanovRecent}.

289: 

290: \begin{example}\label{E:purebraid} \rm

291:  Suppose that $n=2m$, and denote by $\sigma_i$ the standard generator of

292: $B_n$, in which the $i$th and the $(i+1)$st punctures permute their

293: positions in a clockwise sense. We define $\beta=\sigma_1 \sigma_3^2

294: \sigma_5^3 \cdots \sigma_{2m-1}^{m}$.

295: 

296: 

297: The canonical reduction system of $\beta$ consists of $m$ circles, the

298: $i$th one enclosing the punctures $2i-1$ and $2i$. The corresponding

299: tubular braid is the trivial braid of $B_m$, and the interior braids are,

300: respectively, $\sigma_1$, $\sigma_1^2$, \ldots, $\sigma_1^m$ (notice that

301: all of them are non-conjugate, since conjugate braids have the same

302: exponent sum).

303: 

304: Let $D_{(1)}, \ldots, D_{(m)}$ be the disks bounded by the above circles.

305: As we shall see, any braid that commutes with $\beta$ has to send each

306: disk $D_{(i)}$ to itself (since the interior braids are non-conjugate).

307: A generating set of the centralizer subgroup of $\beta$ is given by

308: \begin{enumerate}

309: \item[(i)] for each $i\in \{1,\ldots,m\}$, the braid $\sigma_{2i-1}$, whose

310:  support is contained in $D_{(i)}$,

311: \item[(ii)] any generating set for the pure braid group on $m$ strings $P_m$--

312: all the generators here act as the identity on

313: $D_{(1)}\cup\ldots\cup D_{(m)}$, and can be seen as a pure tubular braid on $m$

314: strings (tubes), where the $i$th tube starts and ends at $D_{(i)}$.

315: \end{enumerate}

316: It can be easily shown that, in this case, $Z(\beta)\simeq\Z^m\times P_m$.

317: The essential observation now is the following: it can be deduced by the

318: presentation given in~\cite{Bir}, that the abelianization of $P_m$ is

319: isomorphic to $\Z^{m(m-1)/2}$  (see also~\cite{Arnold}). Hence, the

320: abelianization of $Z(\beta)$ is isomorphic to $\Z^m\times \Z^{m(m-1)/2}$.

321: Therefore, at least $m+\frac{m(m-1)}{2}=\frac{m(m+1)}{2}$ generators are

322: needed for the centralizer of the braid $\beta$.

323: 

324:  The case when $n=2m+1$ is analogous. The braid proposed by S. J. Lee is:

325: $\beta=\sigma_2 \sigma_4^2 \sigma_6^3 \cdots \sigma_{2m}^m$. This time the

326: first strand is not enclosed by any curve of the canonical reduction

327: system of $\beta$, and one has: $Z(\beta)\simeq \Z^m \times P_{m+1}$.

328: Hence, in this case the minimal possible number of generators is

329: $m+\frac{m(m+1)}{2}=\frac{m(m+3)}{2}$.

330: \end{example}

331: 

332: By proving theorem~\ref{T:main}, we will show that the above examples are

333: the worst one can find.

334: 

335: % ------------------------------------------------------------------

336: 

337: \section{The periodic case}\label{S:per}

338: 

339: We have to start by describing the periodic elements of $B_n$.

340: In order to state this classification result, which is classical, we

341: need to define two braids.

342: 

343: If $D_n$ is the disk with $n$ punctures arranged regularly on the circle

344: of radius $1$, then the braid which we shall call $\delta_{(n)}$

345: is represented by a clockwise

346: movement of all punctures on this circle by an angle $\frac{2\pi}{n}$.

347: If no confusion is possible, we shall simply write $\delta$, without

348: indicating the number of strands (note that this braid is the Garside

349: element of the Birman-Ko-Lee structure of $B_n$~\cite{BKL}).

350: 

351: 

352: Similarly, if we think of $D_{n}$ as having one puncture in the centre, and

353: $n-1$ punctures arranged circularly around it, then we define

354: $\gamma_{(n)}\in B_{n}$ to

355: be the braid given by a circular movement of the $n-1$ punctures by an

356: angle of $\frac{2\pi}{n-1}$, while leaving the central puncture fixed.

357: Again, for simplicity we shall often only write $\gamma$ instead of

358: $\gamma_{(n)}$.

359: 

360: 

361: The result that classifies periodic braids, which is due to

362: Eilenberg \cite{Eilenberg} and de~Ker\'ekj\'art\'o \cite{deK}

363: (see~\cite{coko} for a modern exposition) is:

364: 

365: \begin{lemma}\label{L:per_gamma_delta}

366: Every periodic braid in $B_n$ is conjugate to a power of $\delta_{(n)}$ or

367: $\gamma_{(n)}$.

368: \end{lemma}

369: 

370: Thus we only need to consider the centralizer subgroups of $\delta_{(n)}^k$

371: and $\gamma_{(n)}^k$ for all $n, k\in \Z$, since the centralizers of

372: conjugate elements are isomorphic by an inner automorphism of $B_n$. This

373: problem has been solved by Bessis, Digne and Michel~\cite{BDM}, on the

374: wider context of complex reflexion groups. We shall explain their result in

375: the particular case of braid groups:

376: 

377: We suppose first that $\beta=\delta_{(n)}^k$ where, without loss of

378: generality, $k\geqslant 0$. Let $d=\gcd (n,k)$. For $u=1,\ldots,n$,

379: we will denote $P_u=e^{i 2\pi u/n}$ the punctures of $D_n$,

380: so $\beta=\delta_{(n)}^k$ sends $P_u$ to $P_{u+k}$ for every $u$

381: (the indices are taken modulo $n$). Hence the permutation induced

382: by $\beta$ has $d$ orbits (cycles) of length $r=\frac{n}{d}$, that we

383: denote by ${\cal C}_1,\ldots, {\cal C}_d$. See in figure~\ref{F:delta_orbits}

384: an example where $n=12$, $k=9$, $d=3$ and $r=4$: the braid $\delta_{(12)}$

385: and the three orbits of $\delta_{(12)}^9$.

386: %

387: \begin{figure}[htb]

388: \centerline{

389: \epsfbox{delta_orbits.eps}

390: }

391: \caption{The braid $\delta\in B_{12}$, and the three orbits of

392: $\delta^9$ (in black, white and grey).}

393: \label{F:delta_orbits}

394: \end{figure}

395: 

396: If $r>1$ (that is if $d<n$), consider the once punctured disc

397: $D^*=D\backslash \{0\}$, and the $r-$sheeted covering $\theta=\theta_r\co

398: D^* \rightarrow D^*$ defined by $\theta(ae^{it})=ae^{itr}=ae^{itn/d}$. The

399: orbits ${\cal C}_1,\ldots,{\cal C}_d$ are sent by $\theta$ to the points

400: $Q_1,\ldots,Q_d$, where $Q_u=e^{i2\pi u/d}$. If we consider the half-line

401: $L=\{ae^{i\pi/d},\; a\in ]0,2]\}$

402: (notice that $L$ passes between $Q_d$ and $Q_1$), then $D^*\backslash L$

403: is a fundamental region for $\theta$ (see figure~\ref{F:theta}).

404: %

405: \begin{figure}[htb]

406: \centerline{ \epsfbox{theta.eps} } \caption{The covering map $\theta=\theta_4$ associated to

407: $\delta_{(12)}^9$. } \label{F:theta}

408: \end{figure}

409: 

410: Now notice that every braid in $B_d(D^*)$ can be lifted, by $\theta^{-1}$,

411: to a braid in $B_n$ in a natural way. The resulting braid is a $\frac{2\pi

412: d}{n}$-symmetric braid, that is, it is invariant under a rotation by an

413: angle of $\frac{2\pi d}{n}$. But then it is also invariant under a

414: rotation of angle $\frac{2\pi k}{n}$; in other words, the resulting braid

415: commutes with $\beta$. Hence we have a natural homomorphism: $\theta^*\co

416: B_d(D^*) \rightarrow B_n$ whose image is contained in $Z(\beta)$. Then one

417: has

418: 

419: \begin{theorem}[\cite{BDM}]\label{T:centrdelta}

420: The natural homomorphism

421: $\theta^*\co B_d(D^*)\rightarrow Z(\delta_{(n)}^k)$ is an isomorphism.

422: \end{theorem}

423: 

424: In other words, every element in the centralizer of $\beta=\delta_{(n)}^k$

425: can be seen (via $\theta$) as a braid on a once punctured disc, that is, a

426: braid on an annulus. Notice that if $r=1$ (that is, if $k$ is a multiple

427: of $n$), then $\beta$ is a power of $\delta_{(n)}^n=\Delta_{(n)}^2$. In this

428: case $\theta$ is the identity map, and the fundamental region is the whole

429: $D_n$. Hence the centralizer of $\beta$ is the whole $B_n$, as one should

430: expect.

431: 

432: 

433: Since we are interested in minimising the set of generators, we observe

434: that if $d=n$ (thus $r=1$), then $Z(\beta)=B_n$ is generated by two

435: elements, namely Artin's $\sigma_1$ and Birman-Ko-Lee's $\delta$. In a

436: similar way, if $1<d<n$, then the braid group $B_d(D^*)$ is generated by

437: just two elements, namely $\delta_{(n)}=\theta^*(\delta_{(d)})$ and the

438: braid $\theta^*(\sigma_1)$ shown in figure \ref{F:gendelgam}(a). Notice

439: that this case contains the above one, where $\theta^*$ is the identity.

440: Finally, if $d=1$ then $B_1(D^*)$ is cyclic. Thus we have:

441: 

442: \begin{prop}\label{P:centrdelta'}

443: If $k$ and $n$ are coprime, then $Z(\delta_{(n)}^k)$ is generated by a single

444: element, namely $\delta_{(n)}$. If, by contrast, $gcd(k,n)\geqslant 2$, then

445: $Z(\delta_{(n)}^k)$ is generated by two elements: $\delta_{(n)}$ and the braid

446: $\theta^*(\sigma_1)$.

447: \end{prop}

448: 

449: \begin{figure}[htb]

450: \centerline{

451: \epsfbox{gendelgam.eps} % exported at 35pc

452: }

453: \caption{Generators $\theta_3^*(\sigma_1)$ and $\bar{\theta}_3^*(\sigma_1)$

454: of the centralisers of $\delta_{(12)}^4$ and $\gamma_{(13)}^4$.}

455: \label{F:gendelgam}

456: \end{figure}

457: 

458: It is clear that the generating set given by proposition~\ref{P:centrdelta'}

459: is indeed minimal.

460: %Following with the work in~\cite{BDM}, we will study the centralizer of

461: %$\beta=\gamma_{(n)}^k$.

462: Next we study the centralizer of $\beta=\gamma_{(n)}^k$, still following the

463: work in \cite{BDM}.  This time we call $d=\gcd (n-1,k)$, and $r=(n-1)/d$.

464: If $d<n-1$, the above map $\theta$ induces a natural homomorphism

465: $\bar{\theta}^*=\bar{\theta}_r^*\co B_d(D^*)\rightarrow B_n$, where this

466: time the central point of $D$ is considered as a puncture. Hence, the central

467: strand of every braid coming from $B_d(D^*)$ is trivial. We observe that the

468: image of this homomorphism is contained in $Z(\beta)$, and in fact one has:

469: 

470: \begin{theorem}[\cite{BDM}]\label{T:centrdelta2}

471: The natural homomorphism

472: $\bar{\theta}_r^*\co B_d(D^*)\rightarrow Z(\gamma_{(n)}^k)$ is an isomorphism.

473: \end{theorem}

474: 

475: By contrast, if $d=n-1$, then $\beta$ is a power of $\gamma^{n-1}=\Delta^2$,

476: so $\overline{\theta}_r^*=1$, $Z(\beta)=B_n$ and everything works as above.

477: Hence we have

478: 

479: \begin{prop}\label{P:centrgamma}

480: If $k$ and $n-1$ are coprime, then $Z(\gamma_{(n)}^k)$ is generated by a

481: single element, namely $\gamma_{(n)}$. If, by contrast, $gcd(k,n-1)=d

482: \geqslant 2$, then $Z(\gamma_{(n)}^k)$ is generated by two elements:

483: $\gamma_{(n)}=\bar{\theta}^*(\delta_{(d)})$ and the braid

484: $\bar{\theta}^*(\sigma_1)$.

485: \end{prop}

486: 

487: See figure \ref{F:gendelgam}(b) for an illustration of the braid

488: $\bar{\theta}^*(\sigma_1)$. We summarize all the results in this

489: section as follows:

490: 

491: \begin{cor}

492:   The centralizer of any periodic braid in $B_n$ either equals $B_n$ or is

493:   isomorphic to $B_d(D^*)$, for some $d<n$. In particular, it can be

494:   generated by at most two elements.

495: \end{cor}

496: 

497: We end with a result that will be helpful later:

498: 

499: \begin{cor}\label{C:addstring}

500: If $k$ is not a multiple of $n$, then

501: $Z(\delta_{(n)}^k)\cong Z(\gamma_{(n+1)}^k)$.

502: \end{cor}

503: 

504: \begin{proof}

505:  Both groups are isomorphic to $B_d(D^*)$, where $d=\gcd(n,k)$.

506: An actual isomorphism can be defined as follows: take any element

507: $\alpha\in Z(\delta_{(n)}^k)$, isotope it to make it

508: $\frac{2\pi k}{n}$-symmetric, and then add a

509: trivial strand based at the central point of $D_n$.

510: \end{proof}

511: 

512: 

513: % ------------------------------------------------------------------

514: 

515: \section{The pseudo-Anosov case}\label{S:pA}

516: 

517: \begin{prop}\label{P:pA}

518: If $\beta\in B_n$ is pseudo-Anosov, then the centralizer of $B_n$ is free

519: abelian and generated by two elements: some {\pA} $\alpha$ which has the

520: same stable and unstable projective measured foliation as $\beta$ (possibly

521: $\beta$ itself), and one periodic braid $\rho$ (a root of $\Delta^2$,

522: possibly $\Delta^2$ itself).

523: \end{prop}

524: 

525: We stress that the generating set promised by proposition \ref{P:pA} is

526: obviously minimal.

527: For proving this result, it is more convenient to think about the quotient

528: group $\BnD$. Since $\langle \Delta^2 \rangle$ is the center of $B_n$,

529: it is contained in the centralizer of any element. Hence the centralizer of

530: an element in $B_n$ is just the preimage of the centralizer of its

531: corresponding mapping class in $\BnD$. Thus, for the rest of this section,

532: we shall work in this quotient $\BnD$; we shall prove the following

533: result, from which proposition \ref{P:pA} will then be deduced:

534: 

535: \begin{prop}\label{P:pA'} If $\beta\in \BnD$ is \pA, then the centralizer

536: of $\beta$ is abelian, and is generated by some {\pA} $\alpha$ which has the

537: same stable and unstable projective measured foliation as $\beta$, and

538: possibly one element $\rho$ of finite order.

539: \end{prop}

540: 

541: \begin{proof}[Proof of proposition \ref{P:pA'}]

542: We start by observing that the {\pA} element $\beta$ cannot commute with

543: any reducible element $a\in \BnD$, except possibly with periodic ones --

544: thus all elements of $Z(\beta)\subset \BnD$ are either {\pA} or periodic.

545: To see this, let us assume that the canonical reduction system $C$ of $a$

546: is non-empty. Then the canonical reduction system of

547: $\beta^{-1} a\beta$ is $\beta(C)$. If it were true that $\beta^{-1} a\beta=a$,

548: then we would have $\beta(C)=C$, which is impossible since it is well known

549: that {\pA} homeomorphisms do not stabilise any curves or multicurves.

550: (This result is also a special case of corollary 7.13 of \cite{Ivanov}.)

551: 

552: Our next claim is that all pseudo-Anosov elements in $Z(\beta)$ have the

553: same stable and unstable projective measured foliations.

554: In order to prove this, we can apply Corollaries 7.15 and 8.4 of

555: \cite{Ivanov}: since the centralizer subgroup of $\beta$

556: is infinite and irreducible,

557: it follows that $Z(\beta)$ contains an infinte cyclic group as a subgroup

558: of finite index. It follows that if $a$ is any {\pA} element in the

559: centralizer of $\beta$, then there exist $k, k'\in \N$ such that

560: $a^k=\beta^{k'}$. Since all powers of a {\pA} element have the same stable

561: and unstable projective measured foliation, it follows that $a$ has

562: the same stable and unstable projective measured foliations as $\beta$, and

563: so do all {\pA} elements of $Z(\beta)\subseteq \BnD$.

564: 

565: Next we make an essential observation which only works for braid groups,

566: and does not generalize to mapping class groups of surfaces with no boundary,

567: or with more than two boundary components: all elements of

568: $\BnD$, regarded as a subgroup of the mapping class group of the $n+1$ times

569: punctured sphere, fix the puncture which came from collapsing the boundary

570: of $D_n$. Moreover, there are singular leaves of the stable and unstable

571: foliation of $\beta$ emanating from this puncture, at least one of each

572: (like for every other puncture). In the cyclic ordering around the puncture,

573: singular leaves of the stable and unstable foliation alternate. If an

574: element $a$ of $\BnD$ commutes with $\beta$, then the action of $a$ has to

575: preserve the projective stable

576: and unstable foliations. Thus in the cyclic ordering around our preferred

577: puncture, the action of $a$ can only induce a cyclic (possibly trivial)

578: permutation of the singular leaves (sending stable to stable, and unstable

579: to unstable leaves, nevertheless).

580: 

581: Now we see that an element $a$ of $Z(\beta)\subseteq \BnD$ is uniquely

582: determined by just two data: firstly the stretch factor $\lambda$ by which

583: its action on the unstable measured foliation of $\beta$ multiplies the

584: measure of that foliation. (This factor $\lambda$ equals $1$ if $a$ is

585: periodic, and belongs to the set $\R_+\setminus \{1\}$ if $a$ is \pA).

586: And secondly by the cyclic permutation of the leaves of the stable projective

587: foliation emanating from the distinguished puncture of the $n+1$ times

588: punctured sphere. Indeed, if $a$ and $b$ share both data, then $ab^{-1}$ has

589: stretch factor 1 (so it is periodic), and preserves the singular leaves.

590: Hence it is the identity in $\BnD$, so $a=b$.

591: 

592: This implies that the set of periodic elements of

593: $Z(\beta)$ forms a subgroup of $Z(\beta)$ which is either trivial or

594: isomorphic to $\Z / k\Z$, where $k$ is a divisor of the number of singular

595: leaves of the stable foliation emanating from the preferred puncture.

596: Any generator of this subgroup can play the r\^ole of our desired generator

597: $\rho$ of $Z(\beta)\subseteq \BnD$.

598: 

599: Now $\rho$ commutes with any other element in $Z(\beta)$, because their

600: commutator has stretch factor 1 and induces the trivial permutation of the

601: prongs around the preferred singularity.

602: 

603: Now notice that the stretch factor yields a multiplicative map

604: from $Z(\beta)$ to $\mathbb{R}^+$. But it is known that the set of

605: possible stretch factors for a given foliation is discrete (see~\cite{Ivanov}),

606: so the image of $Z(\beta)$ under this map must be a cyclic subgroup of $\R^+$.

607: Take an element $\alpha$ whose stretch factor $\lambda$ generates this group.

608: Then $\alpha$ is {\pA} and the stretch factor of any element in

609: $Z(\beta)$ must be a power of $\lambda$.

610: 

611: 

612: We now have that $\alpha$ and $\rho$ generate $Z(\beta)\in \BnD$, because any

613: element in $Z(\beta)$ can be multiplied by some power of $\alpha$ so as

614: to obtain an element with stretch factor $1$, i.e.~a power of $\rho$.

615: 

616: It follows that $Z(\beta)\subset \BnD$ is isomorphic to $\Z \times \Z/ k\Z$,

617: with generators $\alpha$ and $\rho$. This completes the proof of

618: proposition \ref{P:pA'}.

619: \end{proof}

620: 

621: 

622: \begin{proof}[Proof of proposition~\ref{P:pA}]

623:  By proposition~\ref{P:pA'}, $Z(\beta)\subset \BnD$ is isomorphic to $\Z

624: \times \Z/ k\Z$, with generators $\alpha$ and $\rho$. But then

625: $Z(\beta)\subset B_n$ is just the preimage of $Z(\beta)\subset \BnD$ under

626: the natural projection. Consider the subgroup

627: $\langle \rho \rangle \subset Z(\beta) \subset \BnD$. Its preimage is an

628: infinte cyclic group in $B_n$ that contains $\langle \Delta^2 \rangle$.

629: We can suppose (up to choosing an appropriate $\rho$), that the generator

630: of this cyclic group projects to $\rho$, so we call it $\rho$ as well.

631: Notice that $\rho$ is a root of $\Delta^2$, since $\Delta^2$ belongs to

632: $\langle \rho \rangle$. Then we choose an element in $B_n$ that projects to

633: $\alpha$, and we also call it $\alpha$. We must prove that in $B_n$ we still

634: have $Z(\beta) = \langle \alpha \rangle \times \langle \rho \rangle$.

635: 

636: But every element in $Z(\beta)\subset B_n$ can be written as $\alpha^k

637: \rho^l \Delta^{2m}$. Since $\Delta^2$ is a power of $\rho$, then

638: $\{\alpha,\rho\}$ is a set of generators of $Z(\beta)$. On the other hand,

639: the commutator of $\alpha$ and $\rho$ projects to the trivial mapping

640: class, hence it equals $\Delta^{2k}$ for some $k$. But the algebraic

641: number of crossings of the braid $\Delta^{2k}$ is $kn(n-1)$, while for the

642: commutator of any two elements this number is zero. Hence $k=0$, so

643: $\alpha$ and $\rho$ commute. Finally, it is well known that $B_n$ is

644: torsion-free, so $Z(\beta)$ is isomorphic to $\Z \times \Z$, as we wanted

645: to show.

646: \end{proof}

647: 

648: 

649: % ------------------------------------------------------------------

650: 

651: \section{The reducible case}\label{S:red}

652: 

653: 

654: It remains to study the centralizer of a non-periodic reducible braid

655: $\beta$. Recall that for every braid $\gamma$ one has

656: $Z(\gamma^{-1}\beta\gamma)= \gamma^{-1} Z(\beta) \gamma$.

657: Hence, in general we will not study $Z(\beta)$, but the centralizer of a

658: suitable conjugate of $\beta$, which will be easier to describe.

659: Throughout this section we shall think of the punctures of the disk $D_n$

660: as being lined up on the real axis.

661: 

662: % ............................................................

663: 

664: \subsection{Reducible braids in regular form}\label{S:reg_form}

665: 

666: As we saw in section~\ref{S:NiTh}, if $\beta$ is a non-periodic reducible

667: element, then its canonical reduction system is nonempty. We denote by

668: $R'(\beta)$ the set of outermost curves in the canonical reduction system

669: of $\beta$. It is determined by $\beta$ up to isotopy fixing the

670: punctures. Since we can study any conjugate of $\beta$, we can suppose

671: that $R'(\beta)$ is a family of disjoint circles centered at the real axis,

672: with disjoint interiors, each one enclosing more than one and less than

673: $n$ punctures.

674: 

675: Notice that there could be punctures in $D_n$ not enclosed by any circle

676: in $R'(\beta)$. In order to simplify the notations below, we define the

677: system of curves $R(\beta)$ to contain exactly the curves of $R'(\beta)$,

678: plus one circle around each such puncture of $D_n$. These new circles are

679: called the degenerate circles of $R(\beta)$. We now have that every puncture

680: in $D_n$ is enclosed by exactly one circle in $R(\beta)$.

681: 

682:  Notice that $\beta$ preserves $R(\beta)$, but it could permute the circles.

683: We will suppose that this permutation has $t$ orbits (or cycles)

684: ${\cal C}_1, \ldots, {\cal C}_t$. That is, ${\cal C}_i$ is a family

685: of circles $\{C_{i,1},\ldots,C_{i,r_i}\}\subset R(\beta)$ such that $\beta$

686: sends $C_{i,k}$ to $C_{i,k+1}$ (here the second index is taken modulo

687: $r_i$). Then one has $R(\beta)={\cal C}_1\cup \cdots \cup {\cal C}_t =$

688: $\{C_{1,1},\ldots,C_{1,r_1}\}\cup \cdots \cup \{C_{t,1},\ldots,C_{t,r_t}\}$.

689: If $m_i$ is the number of punctures inside $C_{i,k}$, for any $k$, then

690: $1\leqslant m_i <n$ and $m_1r_1+\cdots +m_tr_t=n$.

691: 

692: 

693: Let $\widehat{\beta}$ be the tubular braid induced by $\beta$ and

694: $R(\beta)$. Then $\widehat{\beta}\in B_m$, where

695: $m=r_1+\cdots+r_t$. For $i=1,\ldots, t$ and $k=1,\ldots, r_i$, let

696: $\beta_{i,k}$ be the braid induced by $\beta$ in the interior of

697: $C_{i,k}$. In other words, $\beta_{i,k}$ is the braid inside the

698: tube of $\widehat{\beta}$ which starts at $C_{i,k}$ and ends at

699: $C_{i,k+1}$. We will call the braids $\beta_{i,k}$ the {\em interior braids}

700: of $\beta$. Notice that the interior braids of each degenerate circle is

701: just a trivial braid on one string.

702: %Notice that the degenerate circles of $R(\beta)$ are

703: %irrelevant, since they do not modify the corresponding tubular braid, and

704: %their corresponding interior braids are always trivial.

705: 

706: In figure~\ref{F:tubular} we can see an example of a reducible braid

707: $\beta\in B_{13}$, and its corresponding tubular braid $\widehat{\beta}\in

708: B_6$. In this example we have three orbits, and the following data:

709: $r_1=3$, $r_2=2$, $r_3=1$; $m_1=2$, $m_2=3$, $m_3=1$,

710: $\beta_{1,1}=\sigma_1^2$, $\beta_{1,2}=\sigma_1^{-1}$, $\beta_{1,3}=1$,

711: $\beta_{2,1}=\sigma_1 \sigma_2$, $\beta_{2,2}=\sigma_1^{-1}\sigma_2$,

712: $\beta_{3,1}=1$ and $\widehat{\beta}=\sigma_3^2\sigma_2\sigma_1

713: \sigma_5^2\sigma_4$.

714: 

715: 

716: \begin{figure}[htb]

717: \centerline{

718:  \epsfbox{tubular.eps} %

719: }

720: \caption{Example of a reducible braid $\beta$, and its corresponding tubular

721: braid $\widehat{\beta}$.}\label{F:tubular}

722: \end{figure}

723: 

724: 

725: It would be desirable for $\beta$ to have its interior braids as simple

726: as possible, in order to study its centralizer. We propose the following:

727: 

728: \begin{definition}

729: Let $\beta\in B_n$ be a non-periodic reducible braid. Then $\beta$ will be

730: said to be in {\em regular form} if (using the notation introduced above) it

731: satisfies the following conditions:

732: \begin{enumerate}

733: 

734: % \item $R(\beta)$ is a family of circles, whose orbits under $\beta$ will

735: %be denoted by ${\cal C}_1=\{C_{1,1},\ldots,C_{1,r_1}\},\ldots, {\cal C}_t=

736: %\{C_{t,1},\ldots,C_{t,r_t}\}$. * Remarque de Bert: ce n'est pas une condition!

737: 

738:  \item The only non-trivial interior braids in $\beta$ are

739: $\beta_{1,r_1}, \beta_{2,r_2},\ldots, \beta_{t,r_t}$ -- we shall denote these

740: braids by $\beta_{[1]}, \beta_{[2]}, \ldots, \beta_{[t]}$.

741: 

742:  \item For $i,j\in \{1,\ldots,t\}$, if $\beta_{[i]}$ and $\beta_{[j]}$ are

743: conjugate, then $\beta_{[i]}=\beta_{[j]}$.

744: 

745: \end{enumerate}

746: \end{definition}

747: 

748: 

749:  Hence, if $\beta$ is in regular form, there is at most one non-trivial

750: interior braid for each orbit, and any two interior braids are either equal or

751: non-conjugate. Fortunately, one can conjugate every non-periodic reducible

752: braid $\beta$ to another one in regular form, as we are going to see.

753: 

754: First, consider the subgroup $B_{R(\beta)}\subset B_n$ consisting of those

755: braids preserving $R(\beta)$. For $\alpha\in B_{R(\beta)}$, we can consider

756: the tubular braid $\widehat{\alpha}$ induced by $\alpha$ and $R(\beta)$.

757: Every $\alpha\in B_{R(\beta)}$ is completely determined by $\widehat{\alpha}$

758: and its interior braids $\alpha_{i,k}$, for

759: $i=1,\ldots t$ and $k=1,\ldots, r_i$.

760: 

761:  Now consider, in $\beta$, an orbit ${\cal C}_i =

762: \{C_{i,1},\ldots,C_{i,r_i}\}$ and the interior braids

763: $\beta_{i,1},\ldots, \beta_{i,r_i}\in B_{m_i}$. We define $\alpha\in

764: B_{R(\beta)}$ as follows: $\widehat{\alpha}$ is trivial, $\alpha_{j,k}=1$

765: if $j\neq i$, and $\alpha_{i,k}=\beta_{i,k}\beta_{i,k+1}\cdots

766: \beta_{i,r_i}$. If we conjugate $\beta$ by $\alpha$, we obtain

767: $\beta'=\alpha^{-1}\beta \alpha$, which has the following properties:

768: \begin{itemize} \item $\widehat{\beta}'=\widehat{\beta}$.

769: 

770: \item $\beta'_{j,k}=\beta_{j,k}$, for $j\neq i$.

771: 

772: \item $\beta'_{i,k}= (\alpha_{i,k})^{-1} \beta_{i,k} \alpha_{i,k+1} =

773: (\beta_{i,r_i}^{-1}\cdots \beta_{i,k}^{-1})(\beta_{i,k} \cdots

774: \beta_{i,r_i})=1$, for $k\neq r_i$.

775: 

776: \item $\beta'_{i,r_i}=(\alpha_{i,r_i})^{-1} \beta_{i,r_i} \alpha_{i,1} =

777: \beta_{i,r_i}^{-1} \beta_{i,r_i}(\beta_{i,1} \cdots \beta_{i,r_i}) =

778: \beta_{i,1} \cdots \beta_{i,r_i}$.

779: 

780: \end{itemize}

781: 

782: In other words, if we conjugate $\beta$ by $\alpha$ we `transfuse' all the

783: interior braids in ${\cal C}_i$ to the last tube $C_{i,r_i}$, so

784: $\beta'_{i,r_i}$ becomes the only nontrivial interior braid in ${\cal

785: C}_i$. In figure~\ref{F:transfuse} we can see an example of such a

786: conjugation, where $\beta_{[i]}$ denotes the product $\beta_{i,1}\cdots

787: \beta_{i,r_i}$. We can now do the same for every $i=1,\ldots,t$.

788: Therefore, since we are interested in $\beta$ up to conjugacy, we can

789: suppose that $\beta_{i,k}=1$ if $k\neq r_i$ and denote

790: $\beta_{[i]}=\beta_{i,r_i}$, for every $i=1,\ldots,t$.

791: 

792: 

793: \begin{figure}[htb]

794: \centerline{

795:  \epsfbox{transfuse.eps} %

796: }

797: \caption{How to conjugate $\beta$ to simplify interior braids.}

798: \label{F:transfuse}

799: \end{figure}

800: 

801: 

802: Now suppose that some $\beta_{[i]}$ is conjugate to some $\beta_{[j]}$,

803: and let $h_{i,j}$ be a conjugating braid, that is, $h_{i,j}^{-1}

804: \beta_{[i]} h_{i,j}= \beta_{[j]}$. Consider the braid $\alpha\in

805: B_{R(\beta)}$ such that $\widehat{\alpha}=1$, $\;\alpha_{j,k}=1$ for

806: $j\neq i$ and $\alpha_{i,k}=h_{i,j}$ for every $k$. As we can see in

807: figure~\ref{F:betai_betaj}, if we conjugate $\beta$ by $\alpha$, then

808: $\beta_{[i]}$ is replaced by $\beta_{[j]}$. Therefore, we can assume that

809: for $i,j\in \{1,\ldots,t\}$, either $\beta_{[i]}=\beta_{[j]}$ or

810: $\beta_{[i]}$ and $\beta_{[j]}$ are not conjugate, and therefore we can

811: suppose that $\beta$ is in regular form.

812: 

813: 

814: \begin{figure}[htb]

815: \centerline{

816:  \epsfbox{betai_betaj.eps} %

817: }

818: \caption{How to replace $\beta_{[i]}$ by $\beta_{[j]}$ if they are conjugate.}

819: \label{F:betai_betaj}

820: \end{figure}

821: 

822: 

823:  Notice that we have chosen to put $\beta_{[i]}$ into the tube starting at

824: $C_{i,r_i}$. But we can move it to any other tube of ${\cal C}_i$ if we

825: wish, by a suitable conjugation, and later on we will need to use this.

826: Hence we define, for $i\in \{1,\ldots,t\}$ and $k\in\{1,\ldots, r_i-1\}$,

827: a braid $\mu=\mu(i,k)$ that will `move' the interior braid $\beta_{[i]}$

828: to the tube $C_{i,k}$. This braid is defined as follows: the tubular braid

829: $\widehat{\mu}$ is trivial, and the interior braids are all trivial except

830: $\mu_{i,k+1}=\mu_{i,k+2}=\cdots =\mu_{i,r_i}=\beta_{[i]}$. We can see in

831: figure~\ref{F:mu_ik} how this works.

832: 

833: \begin{figure}[htb]

834: \centerline{

835:  \epsfbox{mu_ik.eps} %

836: }

837: \caption{How to move $\beta_{[i]}$  from $C_{i,4}$ to $C_{i,2}$,

838: when $r_i=4$.}\label{F:mu_ik}

839: \end{figure}

840: 

841: % ............................................................

842: 

843: \subsection{Centralizer of a braid in regular form}

844: 

845:  We will now study the centralizer of $\beta$, assuming that $\beta$ is in

846: regular form. Recall that the only non-trivial interior braids of $\beta$

847: are denoted $\beta_{[1]},\ldots,\beta_{[t]}$, and that $\widehat{\beta}$

848: is the tubular braid associated to $\beta$ and $R(\beta)$. In this section

849: we will show that there is an exact sequence:

850: $$

851:   1 \rightarrow Z(\beta_{[1]})\times \cdots \times Z(\beta_{[t]})

852: \stackrel{g}{\longrightarrow}

853: Z(\beta) \stackrel{p}{\longrightarrow} Z_0(\widehat{\beta}) \rightarrow 1,

854: $$

855: where $Z_0(\widehat{\beta})$ is a subgroup of $Z(\widehat{\beta})$. Later

856: on we will see that this sequence splits.

857: 

858: 

859: For $i\in \{1,\ldots,t\}$, consider the centralizer $Z(\beta_{[i]})$ in

860: $B_{m_i}$. We define a map

861: $g_i\co Z(\beta_{[i]}) \rightarrow B_{R(\beta)}$ as follows:

862: given $\gamma\in Z(\beta_{[i]})$, $g_i(\gamma)$

863: is the braid $\alpha\in B_{R(\beta)}$ satisfying $\widehat{\alpha}=1$,

864: $\;\alpha_{j,k}=1$ for $j\neq i$, and $\alpha_{i,k}=\gamma$ for

865: $k=1,\ldots,r_i$. We need to show the following:

866: 

867: 

868: \begin{prop}\label{P:gi_injective}

869:  The map $g_i$ defined above is an injective homomorphism, and its image

870: is contained in $Z(\beta)$.

871: \end{prop}

872: 

873: \begin{proof}

874: % Let $\gamma,\delta\in Z(\beta_{[i]})$. Since the tubular braids

875: %associated to $g_i(\gamma)$ and $g_i(\delta)$ are trivial, then the

876: %interior braid

877: %$(g_i(\gamma)g_i(\delta))_{j,k}=(g_i(\gamma))_{j,k}(g_i(\delta))_{j,k}$,

878: %for every possible $j$ and $k$. Hence, one has

879: %$(g_i(\gamma)g_i(\delta))_{j,k}=1$ if $j\neq i$, and one also has

880: %$(g_i(\gamma)g_i(\delta))_{i,k}=\gamma\delta$. So the interior braids of

881: %$g_i(\gamma)g_i(\delta)$ coincide with the interior braids of

882: %$g_i(\gamma\delta)$. Since the tubular braids are both trivial, this

883: %implies that $g_i(\gamma)g_i(\delta)=g_i(\gamma\delta)$, so $g_i$ is a

884: %homomorphism.

885: %

886: %Moreover, since $g_i(\gamma)_{i,1}=\gamma$ for every

887: %$\gamma\in Z(\beta_{[i]})$, then $g_i(\gamma)=1$ implies $\gamma=1$,

888: %so $g_i$ is one to one.

889: The map $g_i$ is given by the diagonal homomorphism

890: $Z(\beta_{[i]})\to Z(\beta_{[i]}) \times \ldots \times Z(\beta_{[i]})$

891: ($r_i$ factors), followed by the homomorphism induced by an inclusion of

892: $r_i$ copies of an $m_i$-times punctured disk into $r_i$ disjoint subdisks

893: (each containing $m_i$ punctures) of $D_n$. By the results of

894: \cite{ParisRolfsen} we can deduce that $g_i$ is indeed an injective

895: homomorphism.

896: 

897: 

898: It remains to show that for every $\gamma\in Z(\beta_{[i]})$ one has

899: $\alpha=g_i(\gamma)\in Z(\beta)$. Since $\widehat{\alpha}$ is trivial,

900: $\widehat{\alpha^{-1}\beta \alpha}= \widehat{\alpha}^{-1}\widehat{\beta}

901: \widehat{\alpha}=\widehat{\beta}$. So we just need to show that the

902: interior braids of $\alpha^{-1} \beta \alpha$ and $\beta$ coincide.

903:   For $j\neq i$, the braids $\alpha_{j,k}$ are trivial for every $k$, so

904: $\left(\alpha^{-1} \beta \alpha\right)_{j,k}=\beta_{j,k}$.

905:  Now, for $k\neq r_i$, one has $\left(\alpha^{-1} \beta \alpha\right)_{i,k} =

906: \alpha_{i,k}^{-1} \; \beta_{i,k}\;  \alpha_{i,k+1} =

907: \gamma^{-1} 1 \gamma = 1= \beta_{i,k}$.

908:  Finally, since $\gamma$ commutes with $\beta_{[i]}$, one has

909: $\left(\alpha^{-1} \beta \alpha\right)_{i,r_i} =

910: \alpha_{i,r_i}^{-1} \; \beta_{i,r_i}\;  \alpha_{i,1} =

911: \gamma^{-1} \beta_{[i]} \gamma = \beta_{[i]}= \beta_{i,r_i}$.

912: Therefore $\alpha^{-1}\beta \alpha = \beta$, so the image of $g_i$ is

913: contained in $Z(\beta)$.

914: \end{proof}

915: 

916: 

917: \begin{prop}\label{P:g_injective}

918:  The map  $\; g\co Z(\beta_{[1]})\times \cdots \times Z(\beta_{[t]})

919: \longrightarrow Z(\beta)$ defined by

920: $g(\gamma_1,\ldots,\gamma_t)=g_1(\gamma_1)\cdots g_t(\gamma_t)$

921: is an injective homomorphism.

922: \end{prop}

923: 

924: \begin{proof} Given $\gamma\in Z(\beta_{[i]})$, the only nontrivial

925: strands in $g_i(\gamma)$ are those inside the tubes

926: $C_{i,1},\ldots, C_{i,r_i}$.

927: Hence if $i\neq j$, $\gamma\in Z(\beta_{[i]})$ and

928: $\delta\in Z(\beta_{[j]})$, then $g_i(\gamma)$ and $g_j(\delta)$ commute.

929: Since every $g_i$ is a homomorphism, this shows that $g$ is also a

930: homomorphism. But we know by the previous proposition that $g_i$ is

931: injective for $i=1,\ldots,t$. Using an argument similar to the proof of

932: proposition \ref{P:gi_injective}, one can deduce that $g$ is also injective.

933: \end{proof}

934: 

935: Now we will relate $Z(\beta)$ and $Z(\widehat{\beta})$.

936: Every braid in $Z(\beta)$ preserves the canonical reduction system of

937: $\beta$ (see~\cite{Ivanov}), so it must preserve $R(\beta)$. That is,

938: $Z(\beta)\subset B_{R(\beta)}$. Let $p\co B_{R(\beta)}\rightarrow B_m$ be

939: the homomorphism which sends $\alpha$ to $\widehat{\alpha}$, the tubular

940: braid induced by $\alpha$ and $R(\beta)$. If we take $\alpha\in Z(\beta)$

941: then $\beta=\alpha^{-1}\beta \alpha$, so

942: $p(\beta)=p(\alpha^{-1}\beta \alpha)=p(\alpha)^{-1}p(\beta) p(\alpha)$.

943: Hence $p(\alpha)$ commutes with $p(\beta)=\widehat{\beta}$. Therefore, if

944: we restrict $p$ to $Z(\beta)$ we get

945: $\;p\co Z(\beta)\rightarrow Z(\widehat{\beta})$.

946: 

947: 

948: Unfortunately, neither $p\co B_{R(\beta)}\rightarrow B_m$ nor its restriction

949: $\;p\co Z(\beta)\rightarrow Z(\widehat{\beta})$ are surjective, but we shall

950: see that the

951: elements in the image of $p$ in either case can be easily characterised by

952: the permutation they induce. Notice that $p$ induces a bijection

953: $\widetilde{p}$ from $R(\beta)$ to $\{P_1,\ldots,P_m\}$, the punctures of

954: $D_m$. We denote by $\tau$ the inverse of $\widetilde{p}$.

955: 

956: \begin{definition}

957:  Let $\eta\in B_m$, and let $\pi_{\eta}$ be the permutation induced by

958: $\eta$ on the punctures of $D_m$. We say that $\pi_{\eta}$ is

959: {\em consistent with} $R(\beta)$ if, for $i=1,\ldots,m$,

960: $\;\tau(P_i)$ and $\tau(\pi_{\eta}(P_i))$ enclose the same number of punctures.

961: \end{definition}

962: 

963: \begin{prop}\label{P:cons_Rbeta}

964: An element $\eta\in B_m$ is in the image of

965: $\;p\co B_{R(\beta)}\rightarrow B_m$ if and only if

966: $\pi_{\eta}$ is consistent with $R(\beta)$.

967: \end{prop}

968: 

969: \begin{proof}

970:  If $\eta$ is in the image of $p$, let $\alpha\in B_{R(\beta)}$ with

971: $p(\alpha)=\eta$. Then, for every $i=1,\ldots,m$, $\;\tau(P_i)$ and

972: $\tau(\pi_{\eta}(P_i))$ are the top and bottom circles of a tube

973: determined by $\alpha$. Hence they must enclose the same number of

974: punctures (the number of strands inside the tube).

975: 

976: Conversely, suppose that $\pi_{\eta}$ is consistent with $R(\beta)$. Take

977: $i\in \{1,\ldots,m\}$ and suppose that $\tau(P_i)=C_{j,k}$. Then take the

978: $i$th strand of $\eta$ and consider it as a tube, enclosing the trivial

979: braid on $m_j$ strands. Do this for every $i=1,\ldots,m$. The resulting

980: braid, $\psi(\eta)$, is well defined since $\pi_{\eta}$ is consistent with

981: $R(\beta)$, and it belongs to $B_{R(\beta)}$. Moreover,

982: $p(\psi(\eta))=\eta$ by construction.

983: \end{proof}

984: 

985: The homomorphism $\psi$ introduced in this proof will play a prominent

986: r\^ole in what follows: if $\eta\in B_m$, then $\psi(\eta)$ is the braid

987: in $B_{R(\beta)}$ whose tubular braid equals $\eta$, and whose interior

988: braids are all trivial.

989: 

990: All the elements in $B_m$ that shall be considered from now on will have

991: permutations consistent with $R(\beta)$. Hence, by abuse of notation, we

992: will identify $C_{i,k}=\widetilde{p}(C_{i,k})$ and

993: ${\cal C}_i=\widetilde{p}({\cal C}_i)$ if it does not lead to confusion.

994: 

995: 

996: We still need to characterise the elements in the image of

997: $\;p\co Z(\beta)\rightarrow Z(\widehat{\beta})$. We just know that their

998: permutations must be consistent with $R(\beta)$, but this is not

999: sufficient. Recall that the permutation induced by $\beta$ on the

1000: components of $R(\beta)$ has $t$ orbits, ${\cal C}_1, \ldots {\cal C}_t$.

1001: The key observation now is that every element $\alpha\in Z(\beta)$

1002: preserves these orbits setwise, though it could permute them. Therefore,

1003: for $i=1,\ldots,t$, one has $\alpha({\cal C}_i)={\cal C}_j$ for some $j$.

1004: In the same way, for any $\eta\in Z(\widehat{\beta})$ one has

1005: $\alpha({\cal C}_i)={\cal C}_j$ for some $j$.

1006: 

1007: \begin{lemma}\label{L:perm-conj}

1008: Let $\alpha\in Z(\beta)$. If $\alpha({\cal C}_i)={\cal C}_j$ for some

1009: $i,j\in \{1,\ldots, t\}$, then $\beta_{[i]}=\beta_{[j]}$.

1010: \end{lemma}

1011: 

1012: 

1013: \begin{proof}

1014: Since $\alpha({\cal C}_i)={\cal C}_j$, the two orbits have the same length,

1015: which we shall denote $r$; thus $r=r_i=r_j$.

1016: Now $\beta^r$ is a braid that preserves $C_{i,k}$ and $C_{j,k}$ for every $k$,

1017: and is such that $(\beta^r)_{i,k}=\beta_{[i]}$ and

1018: $(\beta^r)_{j,k}=\beta_{[j]}$.

1019: Now since $\alpha$ commutes with $\beta$, then it also commutes with

1020: $\beta^{r}$. Suppose that $\alpha$ sends $C_{i,1}$ to $C_{j,k}$. Then

1021: $\beta_{[j]}=(\beta^{r})_{j,k} = (\alpha^{-1}\beta^{r}\alpha)_{j,k} =

1022: (\alpha_{i,1})^{-1} (\beta^{r})_{i,1} \alpha_{i,1} =

1023: (\alpha_{i,1})^{-1} \beta_{[i]}\alpha_{i,1}.$

1024: Therefore $\beta_{[i]}$ and $\beta_{[j]}$ are conjugate, and since $\beta$

1025: is in regular form, $\beta_{[i]}=\beta_{[j]}$, as we wanted to prove.

1026: \end{proof}

1027: 

1028: 

1029: Lemma~\ref{L:perm-conj} imposes another condition for a braid in

1030: $Z(\widehat{\beta})$ to be in $p(Z(\beta))$:

1031: 

1032: 

1033: \begin{definition}

1034:  Let $\eta\in Z(\widehat{\beta})$. We say that $\pi_{\eta}$ is

1035: {\em consistent with} $\beta$ if it is consistent with $R(\beta)$ and,

1036: furthermore, for every $i,j\in\{1,\ldots, t\}$ such that

1037: $\eta({\cal C}_i)={\cal C}_j$, one has $\beta_{[i]}=\beta_{[j]}$.

1038: \end{definition}

1039: 

1040: 

1041: \begin{definition}

1042: $Z_0(\widehat{\beta})$ is the subgroup of $Z(\widehat{\beta})$ consisting of

1043: those elements whose permutation is consistent with $\beta$.

1044: \end{definition}

1045: 

1046: Then lemma~\ref{L:perm-conj} can be restated as follows: If

1047: $\alpha\in Z(\beta)$ then $p(\alpha)\in Z_0(\widehat{\beta})$. Moreover,

1048: we can prove the following:

1049: 

1050: 

1051: \begin{prop}\label{P:p_surjective}

1052:  The homomorphism $p\co Z(\beta)\longrightarrow Z_0(\widehat{\beta})$ is

1053: surjective.

1054: \end{prop}

1055: 

1056: 

1057: \begin{proof}

1058:   Let $\eta\in Z_0(\widehat{\beta})$. We shall construct a preimage of $\eta$

1059: under $p$ in two steps. Since $\pi_{\eta}$ is consistent with $\beta$

1060: (thus with $R(\beta)$), we can, as a first step, consider the braid

1061: $\psi(\eta)\in B_n$.

1062: We then have $p(\psi(\eta))=\eta$; but $\psi(\eta)$ does not necessarily

1063: commute with

1064: $\beta$, since the interior braids of $\psi(\eta)^{-1}\beta \psi(\eta)$

1065: could differ from those of $\beta$. Actually, since the interior braids of

1066: $\psi(\eta)$ are all trivial, conjugating $\beta$ by $\psi(\eta)$ just

1067: permutes the interior braids of $\beta$. More precisely, the braid

1068: $\psi(\eta)^{-1} \beta \psi(\eta)$ equals $\beta$, except that, for each

1069: $i\in\{1,\ldots,t\}$, it may not be the tube $C_{i,r_i}$ which contains the

1070: nontrivial interior braid $\beta_{[i]}$, but some other tube from the

1071: family $\mathcal{C}_i$. Our aim in the second step is thus to

1072: fill the tubes of $\psi(\eta)$ with more suitable interior braids, in order

1073: to obtain a braid that commutes with $\beta$.

1074: %so that it will commute with $\beta$.

1075: 

1076:  For every $i\in\{1,\ldots,t\}$, we know that $\psi(\eta)$ sends ${\cal

1077: C}_i$ to some ${\cal C}_j$. Let $k_i\in \{1,\ldots,r_i\}$ be such that

1078: $\psi(\eta)$ sends $C_{i,k_i}$ to $C_{j,r_j}$, and consider the braid

1079: $\mu(i,k_i)$ defined at the end of Subsection~\ref{S:reg_form}. If we

1080: conjugate $\beta$ by $\mu(i,k_i)$ we move $\beta_{[i]}$ from $C_{i,r_i}$

1081: to $C_{i,k_i}$. If we further conjugate by $\psi(\eta)$, then

1082: $\beta_{[i]}$ goes to $C_{j,r_j}$. But $\eta$ is consistent with $\beta$,

1083: so $\beta_{[i]}=\beta_{[j]}$. Hence, the interior braids in ${\cal C}_j$

1084: are preserved. We can do this for $i=1,\ldots,t$, so we obtain that the

1085: braid

1086: $$

1087: \left(\prod_{i=1}^t{\mu(i,k_i)}\right)\psi(\eta)

1088: $$

1089: commutes with $\beta$ and its tubular braid is $\eta$, so it is in

1090: $p^{-1}(\eta)\cap Z(\beta)$. This shows the result.

1091: \end{proof}

1092: 

1093: 

1094: We can finally bring together all the results in this section to state the

1095: following:

1096: 

1097: 

1098: \begin{theorem}\label{T:exactseq}

1099: Let $\beta\in B_n$ be a non-periodic reducible braid in regular form.

1100: Then the sequence

1101: $$

1102: 1 \rightarrow Z(\beta_{[1]})\times \cdots \times Z(\beta_{[t]})

1103: \stackrel{g}{\longrightarrow}

1104: Z(\beta) \stackrel{p}{\longrightarrow}  Z_0(\widehat{\beta}) \rightarrow 1

1105: $$

1106: is exact.

1107: \end{theorem}

1108: 

1109: 

1110: \begin{proof}

1111:   By proposition~\ref{P:g_injective} $g$ is injective, and by

1112: proposition~\ref{P:p_surjective} $p$ is surjective. It just remains to

1113: show that $\mbox{im}(g) =\ker(p)$.

1114: 

1115: By construction, every element in the image of $g$ induces a trivial

1116: tubular braid, so $\mbox{im}(g) \subset \ker(p)$. Let then

1117: $\alpha\in \ker(p)$, that is, $\widehat{\alpha}=1$. Since

1118: $\alpha\in Z(\beta)$, we have $\alpha^{-1}\beta\alpha=\beta$, and since

1119: $\beta_{i,k}=1$ for $k\neq r_i$, we must have

1120: $\alpha_{i,k}^{-1} 1 \alpha_{i,k+1}=1$, so $\alpha_{i,k}=\alpha_{i,k+1}$ for

1121: $k=1,\ldots,r_i-1$. Hence $\alpha_{i,1}=\alpha_{i,2}=\cdots = \alpha_{i,r_i}$

1122: for every $i$. Moreover,  we have

1123: $\beta_{[i]}=\beta_{i,r_i}=\alpha_{i,r_i}^{-1} \beta_{i,r_i} \alpha_{i,1}=

1124: \alpha_{i,1}^{-1} \beta_{[i]} \alpha_{i,1}$, so

1125: $\alpha_{i,1}\in Z(\beta_{[i]})$. Therefore,

1126: $\alpha=g_1(\alpha_{1,1})g_2(\alpha_{2,1})\cdots g_t(\alpha_{t,1})=

1127: g(\alpha_{1,1},\alpha_{2,1},\ldots,\alpha_{t,1})$. That is,

1128: $\ker(p)\subset \mbox{im}(g)$.

1129: \end{proof}

1130: 

1131: % ............................................................

1132: 

1133: \subsection{Finding a section for $\mathbf{p}$}

1134: 

1135:  In this subsection we will prove that the exact sequence of

1136: theorem~\ref{T:exactseq} splits. We recall that $\widehat{\beta}$ is

1137: obtained from $\beta$ by collapsing the disks bounded by outermost curves

1138: in the canonical reduction system of $\beta$ to single punctures. In

1139: particular, the canonical reduction system of $\widehat{\beta}$ must be

1140: empty. Hence, $\widehat{\beta}$ is either periodic or pseudo-Anosov. We will

1141: distinguish these two cases, to define a multiplicative section for

1142: $p$, but first we will show an easy particular case. Recall that a braid

1143: is pure if it induces the trivial permutation of its base points.

1144: 

1145: \begin{prop}\label{P:section_trivial}

1146:  If $\widehat{\beta}$ is pure, there is a homomorphism

1147: $\:h\co Z_0(\widehat{\beta}) \rightarrow Z(\beta)$ such that $p\circ h=1$.

1148: \end{prop}

1149: 

1150: \begin{proof}

1151:   We shall prove that in this case, the homomorphism $\psi$ constructed

1152: in the proof of proposition \ref{P:cons_Rbeta} is such a section.

1153:   Let $\eta\in Z_0(\widehat{\beta})$. Since $\widehat{\beta}$ is pure,

1154: ${\cal C}_i=\{C_{i,1}\}$ for all $i$. Hence, if $\eta$ sends ${\cal C}_i$ to

1155: ${\cal C}_j$ then it sends the tube $C_{i,1}$ (containing $\beta_{[i]}$)

1156: to the tube $C_{j,1}$ (containing $\beta_{[j]}=\beta_{[i]}$, since $\beta$

1157: is in regular form). Therefore, filling every tube in $\eta$ with the

1158: trivial braid, that is, defining $h(\eta)=\psi(\eta)$, yields indeed an

1159: element of $Z(\beta)$.

1160: \end{proof}

1161: 

1162: Next we study the general case, depending whether $\widehat{\beta}$ is

1163: periodic or pseudo-Anosov.

1164: 

1165: \begin{prop}\label{P:section_periodic}

1166: If $\widehat{\beta}$ is periodic, there is a homomorphism

1167: $\:h\co Z_0(\widehat{\beta}) \rightarrow Z(\beta)$ such that $p\circ h=1$.

1168: \end{prop}

1169: 

1170: \begin{proof}

1171: Recall that we are studying $\beta$ up to conjugacy. This implies that we

1172: can also study $\widehat{\beta}$ up to conjugacy since, for every

1173: $\xi\in B_m$, if we conjugate $\beta$ by $\psi(\xi)$ we are conjugating

1174: $\widehat{\beta}$ by $\xi$. Moreover, after conjugating by $\psi(\xi)$,

1175: $\beta$ continues to be in regular form (up to renaming the circles in

1176: $R(\beta)$). Therefore we can suppose, up to conjugacy, that

1177: $\widehat{\beta}$ is a rigid rotation of the disc, that is, a power of

1178: $\delta_{(m)}$ or $\gamma_{(m)}$.

1179: 

1180: Suppose first that $\widehat{\beta}=\delta^k_{(m)}$ for some $k$. We can

1181: suppose that $k$ is not a multiple of $m$, since in that case

1182: $\widehat\beta$ would be a power of $\Delta^2_{(m)}$, thus it would be pure,

1183: and this case has already been studied in proposition~\ref{P:section_trivial}.

1184: Recall the analysis of periodic braids in section~\ref{S:per}: the base

1185: points $Q_1,\ldots, Q_m$ of $\widehat{\beta}$ will be evenly distributed

1186: along a circle of radius 1 around 0. Let $d=\gcd(m,k)<m$ and $r=m/d$.

1187: Then $\widehat\beta $ sends $Q_i$ to $Q_{i+k}$, and there are $d$ orbits

1188: ${\cal C}_1,\ldots,{\cal C}_d$ of length $r$. The orbit ${\cal C}_i$ will

1189: contain the points $Q_u$ where $u\equiv i$ (mod $d$). Since we can choose

1190: which tubes of $\beta $ contain the interior braids, we will suppose that

1191: these are the tubes starting at $Q_{m-d+1}, Q_{m-d+2},\ldots, Q_{m}$, that

1192: is, the last $d$ points of $D_m$.

1193: 

1194: We will consider now some line segments in $D$ which separate the points

1195: $Q_1,\ldots,Q_m$ into $r$ sets of $d$ points. Let $L$ be the line segment

1196: joining the origin with the border of $D$, passing between the points

1197: $Q_{m-d}$ and $Q_{m-d+1}$, and let $L'$ be the segment passing between

1198: $Q_m$ and $Q_1$. Notice that $L$ and $L'$ determine a sector which contains

1199: the points $Q_{m-d+1},\ldots, Q_m$, corresponding to the tubes of $\beta$

1200: with nontrivial interior braids. Let $\phi: \C \rightarrow \C $ be the

1201: rotation around the origin by an angle of $2\pi k/m$ (the angle induced

1202: by $\widehat\beta$), and denote $L_i=\phi^i(L)$. Since $\gcd(m,k)=d$, the

1203: segments $L_0,\ldots,L_{r-1}$ divide $D$ into $m/d=r$ sectors, each

1204: one of angle $2\pi/r$ and containing the points $Q_{id+1},\ldots, Q_{id+d}$

1205: for some $i$. Take the smallest integer $e>0$ such that $\phi^e (L)=L'$.

1206: Then one has $L_0=L$ and $L_e=L'$. We are interested in the union of segments

1207: ${\cal L}=L_1 \cup L_2\cup \cdots \cup L_e$ (see figure~\ref{F:L0_Le} for

1208: an example).

1209: %

1210: \begin{figure}[htb]

1211: \centerline{ \epsfbox{L0_Le.eps} }

1212: \caption{The segments $L$, $L'$, and the union of

1213: segments ${\cal L}$, for $\widehat\beta=\delta^6\in B_{15}$.} \label{F:L0_Le}

1214: \end{figure}

1215: 

1216: 

1217: Let then  $\eta\in Z_0(\widehat{\beta})$.  In order to define $h(\eta)$, it

1218: suffices to define its interior braids. This is done as follows: recall that,

1219: since $\eta$ commutes with $\widehat{\beta}$, it can be isotoped to a

1220: symmetric braid (with respect to the rotation $\phi $), so we take a

1221: symmetric representative of $\eta$. For every base point $Q_i$ of

1222: $\widehat{\beta}$ (corresponding to a circle $C_{j,u}$), consider the strand

1223: of $\eta$ starting at $Q_i$ (the $i$th strand of $\eta$). Then we define the

1224: interior braid $h(\eta)_{j,u}=(\beta_{[j]})^{L(\eta,i)}$, where

1225: $L(\eta,i)\in \Z$ is the algebraic number of times that the $i$th strand of

1226: $\eta$ crosses ${\cal L}$. This is well defined by theorem~\ref{T:centrdelta}

1227: (if you take two distinct representatives of $\eta$ as a symmetric braid,

1228: they are isotopic through symmetric braids, so the strands never touch the

1229: origin and the intersection number $L(\eta,i)$ is preserved).

1230: 

1231: In other words, we define $h(\eta)$ as follows: we start with trivial

1232: interior braids, and we follow the movement of the strands of $\eta$.

1233: Each time a strand crosses a segment of ${\cal L}$ in the positive sense,

1234: we multiply its interior braid by $\beta_{[j]}$ (where $j$ is the index

1235: of the orbit ${\cal C}_j$ of that strand). And every time a strand crosses

1236: ${\cal L}$ in the negative sense, we multiply its interior braid by

1237: $\beta_{[j]}^{-1}$.

1238: 

1239: We have thus defined a map

1240: $h\co Z_0(\widehat{\beta})\rightarrow B_{R(\beta)}$. To show that

1241: $h$ is a homomorphism, it suffices to see that the interior braids of

1242: $\eta \xi$ are the product of those of $\eta$ and $\xi$, for

1243: $\eta,\xi\in Z_0(\widehat{\beta})$. Suppose that the $i$th strand of

1244: $\eta $ goes from $Q_i$ (corresponding to $C_{j,u}$) to $Q_{i'}$

1245: (corresponding to $C_{j',u'}$). Hence $\eta $ sends ${\cal C}_j$ to

1246: ${\cal C}_{j'}$, and since $\eta\in Z_0(\widehat{\beta})$, it follows

1247: that $\beta_{[j]}=\beta_{[j']}$. One also has, by definition,

1248: $L(\eta\xi ,i)=L(\eta ,i)+ L(\xi ,i')$. Therefore $(\eta \xi)_{j,u} =

1249: (\beta_{[j]})^{L(\eta\xi,i)} =

1250: (\beta_{[j]})^{L(\eta,i)} (\beta_{[j]})^{L(\xi,i')} =

1251: \eta_{j,u}\xi_{j',u'}$, so  $h$ is a homomorphism.

1252: 

1253: We must finally show that, with this definition, $h(\eta)\in Z(\beta )$,

1254: for every $\eta \in Z_0(\widehat\beta )$. We will define first some special

1255: braids. For every $i,j\in \{1,\ldots,d\}$ such that  $i<j$ and

1256: $\beta_{[i]}=\beta_{[j]}$, define the symmetric braid $S_{i,j} =

1257: S_{j,i}=\theta_r^*(\sigma_i \cdots \sigma_{j-2} \sigma_{j-1} \sigma _{j-2}

1258: \cdots \sigma_i)$  (see figure~\ref{F:gendelgam} in section~\ref{S:per} to

1259: recall the definition of $\theta_r^*$, and figure~\ref{F:S_13} here for an

1260: example). The braid $S_{i,j}$ commutes with $\widehat\beta $ (since it is

1261: symmetric), and it permutes the orbits ${\cal C}_i$ and ${\cal C}_j$,

1262: preserving the others. Hence $S_{i,j}\in Z_0(\widehat\beta)$. Moreover, its

1263: strands do not cross ${\cal L}$, so by definition of $h$ one has

1264: $h(S_{i,j})=\psi(S_{i,j})$ (the interior braids are trivial).

1265: %

1266: \begin{figure}[htb]

1267: \centerline{ \epsfbox{S_13.eps} }

1268: \caption{The braid $S_{1,3}$, for

1269: $\widehat\beta=\delta^6\in B_{15}$ (assuming that $\beta_{[1]}=\beta_{[3]}$).}

1270: \label{F:S_13}

1271: \end{figure}

1272: 

1273: But $h(S_{i,j})$ commutes with $\beta$, since the only tubes it permutes

1274: are those of the orbits ${\cal C}_i$ and ${\cal C}_j$; among these tubes,

1275: the only two with non-trivial interior braids are exchanged, and their

1276: corresponding interior braids are equal ($\beta_{[i]}=\beta_{[j]}$). Hence

1277: the interior braids of $\beta$ are preserved by $\psi(S_{i,j})=h(S_{i,j})$,

1278: so $h(S_{i,j})\in Z(\beta)$.

1279: 

1280: Take then an arbitrary $\eta\in Z_0(\widehat\beta )$. We must show that

1281: $h(\eta) \in Z(\beta)$. Suppose that $\eta$ sends ${\cal C}_i$ to

1282: ${\cal C}_j$ for some $i,j$. Then $\beta_{[i]}=\beta_{[j]}$, so $S_{i,j}$

1283: is defined, and the braid $\eta S_{i,j}$ preserves the orbit ${\cal C}_i$.

1284: We can continue this way, until we obtain a braid

1285: $\eta S_{i_1,j_1} \cdots S_{i_k,j_k}$ that commutes with $\widehat\beta $

1286: and preserves every orbit ${\cal C}_i$, for $i=1,\ldots,d$.  Since

1287: $h(S_{i,j})\in Z(\beta )$ for every $i,j$, and $h$ is a homomorphism, in

1288: order to show that $h(\eta)\in Z(\beta)$ it suffices to show that $h(\eta

1289: S_{i_1,j_1}\cdots S_{i_k,j_k})\in Z(\beta)$. Therefore, we can suppose that

1290: $\eta$ preserves every orbit ${\cal C}_i$.

1291: 

1292: Denote $\alpha =h(\eta)$. We need to show that the interior braids of $\alpha^{-1} \beta

1293: \alpha $ coincide with those of $\beta$. Since $\eta$ preserves all orbits, we will consider

1294: just the tubes of ${\cal C}_1$, the other ones being analogous. Suppose that $\alpha$ sends

1295: the circle $C_{1,u}$ to $C_{1,r}$. Then it must send $C_{1,v}$ to $C_{1,v-u}$ for every $v$

1296: (the indices are taken modulo $r$).

1297: 

1298: We will identify the points $Q_1,\ldots,Q_m$ with their corresponding circles $C_{i,v}$. For

1299: every $v=1,\ldots,r$, let $b_v$ be the strand of $\eta $ starting at $C_{1,v}$. Since $\eta$

1300: is symmetric, we have $\phi (b_v)=b_{v+1}$. Suppose that $b_v$ crosses $t$ times the segment

1301: $L_i$, where $i\in\{0,\ldots,r-1\}$. Then $b_{v+1}$ will cross $t$ times the segment

1302: $\phi(L_i)=L_{i+1}$. Therefore, if $b_v$ crosses $l$ times ${\cal L}$, and if it crosses

1303: $l_0$ times $L_0$ and $l_e$ times $L_e$, then $b_{v+1}$ crosses $l-l_e+l_0$ times ${\cal

1304: L}$.

1305: 

1306: If $v\neq r$ and $v\neq u$, then $b_v$ neither starts nor ends at $C_{1,r}$. Then it crosses

1307: $L_0$ and $L_e$ the same number of times. Hence, $b_v$ and $b_{v+1}$ cross ${\cal L}$ the

1308: same number of times, say $l$. Therefore, if $v\neq r,u$, one has $(\alpha^{-1} \beta

1309: \alpha)_{1,v-u}=(\alpha_{1,v})^{-1} \beta_{1,v} \alpha_{1,v+1} = \beta_{[1]}^{-l} \: 1 \:

1310: \beta_{[1]}^l= 1 = \beta_{1,v-u}$.

1311: 

1312: If $u=v=r$, then $b_v$ starts at ends at $C_{1,r}$. Hence, as above, it crosses $L_0$ and

1313: $L_e$ the same number of times, so $b_v=b_r$ and $b_{v+1}=b_1$ cross ${\cal L}$ the same

1314: number of times, say $l$. We then have $(\alpha^{-1} \beta \alpha)_{1,v-u}=(\alpha^{-1}

1315: \beta \alpha)_{1,r} = (\alpha_{1,r})^{-1} \beta_{1,r} \alpha_{1,1}= \beta_{[1]}^{-l}

1316: \beta_{[1]} \beta_{[1]}^l = \beta_{[1]} = \beta_{1,r} = \beta_{1,v-u}$. Hence, if $u=r$, we

1317: have already seen all the possible cases. We will then suppose that $u\neq r$.

1318: 

1319: If $v=r$, then $b_v$ starts (but does not end) at $C_{1,r}$. Hence, it crosses $L_e$ one

1320: more time (in the positive sense) than it crosses $L_0$. Therefore, if $b_v=b_r$ crosses $l$

1321: times ${\cal L}$, then $b_{v+1}=b_1$ crosses it $l-1$ times. One has: $(\alpha^{-1} \beta

1322: \alpha)_{1,v-u}=(\alpha^{-1} \beta \alpha)_{1,r-u} = (\alpha_{1,r})^{-1} \beta_{1,r}

1323: \alpha_{1,1}= \beta_{[1]}^{-l} \beta_{[1]} \beta_{[1]}^{l-1} = 1 =

1324: \beta_{1,r-u}=\beta_{1,v-u}$.

1325: 

1326: Finally, if $v=u$ then $b_v$ ends (but does not start) at $C_{1,r}$. In this case, it

1327: crosses $L_e$ one less time (in the positive sense) than it crosses $L_0$. Hence, if

1328: $b_v=b_u$ crosses $l$ times ${\cal L}$, then $b_{v+1}=b_{u+1}$ crosses it $l+1$ times. One

1329: then has: $(\alpha^{-1} \beta \alpha)_{1,v-u}=(\alpha^{-1} \beta \alpha)_{1,r} =

1330: (\alpha_{1,u})^{-1} \beta_{1,u} \alpha_{1,u+1}= \beta_{[1]}^{-l} \: 1 \: \beta_{[1]}^{l+1} =

1331: \beta_{[1]} = \beta_{1,r} = \beta_{1,v-u}$.

1332: 

1333: Therefore, in every possible case we have  $(\alpha^{-1} \beta \alpha)_{1,v-u}=

1334: \beta_{1,v-u}$, for every $v$. This means that the interior braids of $(\alpha^{-1} \beta

1335: \alpha)$ and of $\beta $ coincide, that is, $\alpha =h(\eta)$ commutes with $\beta $, as we

1336: wanted to show.

1337: 

1338: This completes the proof of proposition \ref{P:section_periodic} in the case

1339: $\widehat{\beta}=\delta^k_{(m)}$, and it only remains to deal with the case when

1340: $\widehat\beta=\gamma_{(m)}^k$. As above, we can suppose that $k$ is not a multiple of

1341: $m-1$, since in that case $\widehat\beta$ would be pure, and this case has already been

1342: treated in proposition~\ref{P:section_trivial}. Hence, the only fixed point in the

1343: permutation induced by $\widehat\beta $ is the origin. Therefore, every $\eta$ commuting

1344: with $\widehat\beta $ must fix the origin. This means that, for every $\eta\in

1345: Z_0(\widehat\beta )$, we can fill its central tube with the trivial braid, and the other

1346: tubes in the same way as above (defining ${\cal L}$, and counting the number of times each

1347: strand crosses ${\cal L}$). This defines a homomorphism $h: Z_0(\widehat\beta)\rightarrow

1348: Z(\beta)$ which is a section of $p$. The proof is the same as above.

1349: \end{proof}

1350: 

1351: 

1352: It remains to study the case when $\widehat{\beta}$ is \pA.

1353: 

1354: \begin{prop}\label{P:section_pA}

1355: If $\widehat{\beta}$ is pseudo-Anosov, then there is a homomorphism

1356: $h\co Z_0(\widehat{\beta})\rightarrow Z(\beta)$ such that $p\circ h=1$.

1357: \end{prop}

1358: 

1359: \begin{proof}

1360: In this case, we know that $Z(\widehat{\beta})$ is a free abelian group of rank 2, generated

1361: by a pseudo-Anosov and a periodic braid. Hence, $Z_0(\widehat{\beta})$ is an abelian group

1362: of rank one or two. Notice that $\Delta_{(m)}^2\in Z_0(\widehat{\beta})$, because this braid

1363: commutes with $\widehat{\beta}$ and because $\pi_{\Delta^2}$ is trivial, and thus consistent

1364: with $\beta$. Hence $Z_0(\widehat{\beta})$ contains at least one periodic element. On the

1365: other hand, $\widehat{\beta}$ belongs itself to $Z_0(\widehat{\beta})$, since

1366: $\pi_{\widehat{\beta}}$ is clearly consistent with $\beta$. Hence in $Z_0(\widehat{\beta})$

1367: there are also pseudo-Anosov braids. Since all powers of a periodic braid are periodic, and

1368: all powers of a pseudo-Anosov braid are pseudo-Anosov, it follows that

1369: $Z_0(\widehat{\beta})$ has in fact rank two.  More precisely, $Z_0(\widehat{\beta}) =

1370: \langle \eta\rangle \times \langle \rho \rangle$, where $\eta$ is pseudo-Anosov and $\rho$

1371: is periodic. In particular, we have

1372: $\widehat\beta \in \langle \eta\rangle \times \langle \rho \rangle$, and the three braids

1373: $\widehat{\beta}$, $\eta$ and $\rho$ are mutually commuting.

1374: 

1375: Our aim is to define two commuting braids $h(\rho)$ and $h(\eta)$ in

1376: $Z(\beta)$ which are preimages of $\rho$ respectively $\eta$ under $p$.

1377: The definition of $h(\rho)$ is very simple: we take an arbitrary preimage

1378: of $\rho$ under $p$ -- this is possible since $p$ is surjective by

1379: proposition \ref{P:p_surjective}. It remains to construct $h(\eta)$.

1380: 

1381: \begin{lemma}\label{L:pAlem1}

1382: Suppose $\alpha\in B_{R(\beta)}$, that is, the braid $\alpha$ preserves the set of outermost

1383: curves in the canonical reduction system of $\beta$. Suppose also that $\mu, \nu \in

1384: Z(\widehat{\alpha})$. Suppose that $\iota_\mu\in B_{R(\beta)}$ is a braid with trivial tubes

1385: (i.e.\ $\widehat{\iota}_\mu=1$) such that $\psi(\mu)\cdot\iota_\mu \in Z(\alpha)$. Finally,

1386: suppose that $\mu$ and $\nu$ induce the same permutation. Then we have as well that

1387: $\psi(\nu)\cdot\iota_\mu \in Z(\alpha)$.

1388: \end{lemma}

1389: 

1390: In other words, if two tubular braids commute with $\widehat{\alpha}$, if they induce the

1391: same permutation, and if some ``filling'' of one of them commutes even with $\alpha$, then

1392: the same filling of the other will also commute with $\alpha$.

1393: 

1394: \begin{proof}[Proof of lemma \ref{L:pAlem1}] Conjugating $\alpha$ by

1395: $\psi(\nu)\cdot\iota_\mu \in Z(\alpha)$ yields a certain braid $\alpha'$; we have to check

1396: that $\alpha'=\alpha$. Firstly, we have an equality of tubular braids

1397: $\widehat{\alpha'}=\widehat{\alpha}$, because $\nu$, the tubular braid of

1398: $\psi(\nu)\cdot\iota_\mu$, commutes with $\widehat{\alpha}$. Moreover, since $\mu$ and $\nu$

1399: induce the same permutations, we have for $i=1,\ldots, m$ that the $i$th tube of $\alpha'$

1400: contains the same braid as the $i$th tube of

1401: $(\psi(\mu)\cdot\iota_\mu)^{-1}\cdot\alpha\cdot(\psi(\mu)\cdot\iota_\mu)$. Since

1402: $\psi(\mu)\cdot\iota_\mu$ commutes with $\alpha$, this is in turn the same as the $i$th tube

1403: of $\alpha$. In summary, $\alpha$ and $\alpha'$ have the same tubular braids, and

1404: corresponding tubes contain the same interior braids, which implies that $\alpha=\alpha'$.

1405: \end{proof}

1406: 

1407: Next we have to think in detail about the orbit structure of

1408: $\widehat{\beta}$. Let us choose arbitrarily a puncture $P$ of the disk

1409: $D_m$ (on which $\widehat{\beta}$ acts), and let $O(\widehat{\beta},\rho)$ be

1410: the orbit of that puncture under the action of the subgroup

1411: $\langle \widehat{\beta}\rangle \times \langle \rho \rangle$ of

1412: $Z_0(\widehat{\beta})$. Let $O(\widehat{\beta},\rho,\eta)$ be the orbit

1413: of $P$ under the action of the group

1414: $\langle\rho\rangle \times \langle \eta \rangle$ (note that this

1415: group is also isomorphic to $\Z^2$, and contains $\widehat{\beta}$).

1416: 

1417: We are going to suppose without loss of generality that

1418: $O(\widehat{\beta},\rho,\eta)$ contains {\it all} punctures of $D_m$,

1419: and we shall specify how the tubes of $\eta$ corresponding to this orbit

1420: shall be filled -- indeed, if there are other orbits, then these can be

1421: treated in same way, independently.

1422: 

1423: {\bf Special case: } Let us start by considering the simpler special case that

1424: $O(\widehat{\beta},\rho,\eta) = O(\widehat{\beta},\rho)$, i.e.\ that the

1425: action of $\eta$ preserves the $(\widehat{\beta},\rho)$-orbit. In this case

1426: we have

1427: %

1428: \begin{lemma}\label{L:pAlem2}

1429: There exist integers $k$ and $l$ such that $\eta$ and

1430: $\widehat{\beta}^k\cdot \rho^l$ induce the same permutations on

1431: $O(\widehat{\beta},\rho)$.

1432: \end{lemma}

1433: %

1434: \begin{proof}[Proof of lemma \ref{L:pAlem2}] One can choose $k$ and $l$ such

1435: that $\widehat{\beta}^k \rho^l(P) = \eta(P)$, simply because $\eta(P)$ is in

1436: the orbit of $P$ under the action of $\widehat{\beta}$ and $\rho$. Now if

1437: $P'$ is another point in the orbit, then

1438: $P'=\widehat{\beta}^\kappa \rho^\lambda(P)$ for some $\kappa, \lambda \in \Z$.

1439: Since $\widehat{\beta}, \rho$, and $\eta$ are mutually commuting, we get

1440: $\widehat{\beta}^k \rho^l(P') =

1441: \widehat{\beta}^k \rho^l(\widehat{\beta}^\kappa \rho^\lambda(P)) =

1442: \widehat{\beta}^\kappa\rho^\lambda(\widehat{\beta}^k \rho^l(P))=

1443: \widehat{\beta}^\kappa\rho^\lambda(\eta(P))=

1444: \eta(\widehat{\beta}^\kappa \rho^\lambda(P)) = \eta(P')$.

1445: \end{proof}

1446: 

1447: We already know a nice preimage of $\widehat{\beta}^k\cdot \rho^l$ under $p$:

1448: the braid $\beta^k\cdot h(\rho)^l$ belongs to $Z(\beta)$, because both

1449: $\beta$ and $h(\rho)$ do. This braid can be reexpressed as

1450: $\psi(\widehat{\beta}^k \rho^l)\cdot\iota$, where $\iota$ is

1451: some braid in $B_{R(\beta)}$ with $\widehat{\iota}=1$. (That is, we define

1452: $\iota$ to consist of the interior braids of the tubes of

1453: $\beta^k\cdot h(\rho)^l$).

1454: 

1455: Now we define our filling of $\eta$ by $h(\eta):=\psi(\eta)\cdot\iota$. By

1456: lemma \ref{L:pAlem1} we have that indeed $\psi(\eta)\cdot\iota\in Z(\beta)$.

1457: In order to see that $\psi(\eta)\cdot\iota$ lies also in the centralizer of

1458: $h(\rho)$ one can use a very similar argument. Explicitly, both $\eta$

1459: and $\widehat{\beta}^k\rho^l$ lie in the centralizer of $\rho$, and they

1460: induce the same permutation of the punctures. Moreover,

1461: $\psi(\widehat{\beta}^k \rho^l) \cdot\iota = \beta^k h(\rho)^l\in Z(h(\rho))$.

1462: By lemma \ref{L:pAlem1} we conclude again that

1463: $\psi(\eta)\cdot\iota \in Z(h(\rho))$, also.

1464: 

1465: {\bf General case: } In the case where $\eta$ does not preserve

1466: $O(\widehat{\beta},\rho)$, the strategy is to work not with $\beta$ itself

1467: but with a certain conjugate of $\beta$. The details are as follows.

1468: We have a finite number of disjoint $(\widehat{\beta},\rho)$-orbits in

1469: $O(\widehat{\beta},\rho,\eta)$, and since $\eta$ commutes with

1470: $\widehat{\beta}$ and $\rho$, the action of $\eta$ permutes these

1471: orbits cyclically:

1472: $$

1473: O(\widehat{\beta},\rho)\stackrel{\eta-{\rm action}}{\longrightarrow}

1474: \eta(O(\widehat{\beta},\rho)) \stackrel{\eta-{\rm action}}{\longrightarrow}

1475: \ldots \stackrel{\eta-{\rm action}}{\longrightarrow}

1476: \eta^s(O(\widehat{\beta},\rho))=O(\widehat{\beta},\rho).

1477: $$

1478: Let us denote $\widehat{\beta}_*, \rho_*$ and $\eta^s_*$ the braids

1479: which are obtained from $\widehat{\beta}, \rho$ and $\eta^s$ by retaining

1480: only the strands corresponding to $O(\widehat{\beta},\rho)$, and forgetting

1481: the strands corresponding to all other $(\widehat{\beta},\rho)$-orbits.

1482: Similarly, let $\beta_*$ be the corresponding restriction of $\beta$.

1483: Our first aim is to fill the tubes of $\rho_*$ and $\eta^s_*$ so as to obtain

1484: commuting braids in $Z(\beta_*)$. This can be done as in the ``special case'':

1485: for $\rho_*$ we choose any filling in $Z(\beta_*)$, and for $\eta^s_*$ there

1486: exists a braid $\iota_*$ with trivial tubes such that

1487: $\psi(\eta_*^s) \cdot \iota_*$ commutes with $\beta_*$ and the filling of

1488: $\rho_*$.

1489: 

1490: We have succeeded in finding a filling of certain tubes of $\eta^s$, but

1491: not yet of $\eta$ itself. Also, we have so far only filled the tubes

1492: of $\rho$ which correspond to $O(\widehat{\beta},\rho)$, but not yet

1493: those in the $\eta$-translates of this orbit. We first notice that the

1494: $\eta$-action sends $\widehat{\beta}$-orbits to $\widehat{\beta}$-orbits,

1495: and that in each $\widehat{\beta}$-orbit there is exactly one tube whose

1496: preimage in $\beta$ contains a nontrivial braid (the same for all

1497: $\widehat{\beta}$-orbits), and all other tubes are filled with

1498: a trivial braid. Thus, up to cyclically changing the numbering of the

1499: orbits of each tube of $\widehat{\beta}$, we may assume that the

1500: $\eta$-action sends each tube of $\widehat{\beta}$ in

1501: $O(\widehat{\beta},\rho)$ to a tube of $\widehat{\beta}$ in

1502: $\eta(O(\widehat{\beta},\rho))$ which is filled with the nontrivial braid

1503: if and only if the tube of $O(\widehat{\beta},\rho)$ is. Similarly, for

1504: $i=1,\ldots,s-1$ we may assume that $\eta^i$ sends each $\widehat{\beta}$-tube

1505: in $O(\widehat{\beta},\rho)$ to a $\widehat{\beta}$-tube in

1506: $\eta^i(O(\widehat{\beta},\rho))$ which has the same filling in $\beta$.

1507: 

1508: Now we can use the same property as a construction recipe for $h(\rho)$:

1509: a tube of $\rho$ in $\eta^i(O(\widehat{\beta},\rho))$ (where $i=1,\ldots,s$)

1510: is filled in the same way as its preimage under $\eta^i$. With this

1511: definition, $h(\rho)$ commutes with $\beta$. Finally we are ready to

1512: define $h(\eta)$: we take the braid $\psi(\eta)$, but modify the braids

1513: in the tubes that terminate at positions corresponding to

1514: $O(\widehat{\beta},\rho)$ by multiplying them on the right by $\iota_*$. In

1515: other words, the braid $h(\eta)$ is obtained from $\eta$ as follows: we fill

1516: those tubes of $\eta$ which connect points in $\eta^i(O(\widehat{\beta},\rho))$

1517: to points in $\eta^{i+1}(O(\widehat{\beta},\rho))$ (with $i=0,\ldots,s-2$)

1518: with the trivial braid, and we fill the tubes that start in

1519: $\eta^{s-1}(O(\widehat{\beta},\rho))$ and terminate in

1520: $O(\widehat{\beta},\rho)$ with the interior braids of $\iota_*$.

1521: By construction, this braid $h(\eta)$ commutes with both $\beta$ and $h(\rho)$.

1522: This concludes the proof of proposition \ref{P:section_pA}.

1523: \end{proof}

1524: 

1525: 

1526: \begin{cor}

1527: Suppose that $\beta$ is a non-periodic reducible braid in regular form. Then

1528: the exact sequence

1529: $$

1530: 1\longrightarrow Z(\beta_{[1]})\times \cdots \times Z(\beta_{[t]})

1531: \stackrel{g}{\longrightarrow} Z(\beta)  \stackrel{p}{\longrightarrow}

1532: Z_0(\widehat{\beta}) \longrightarrow 1

1533: $$

1534: splits. That is,

1535: $Z(\beta)\cong\left(Z(\beta_{[1]})\times \cdots

1536: \times Z(\beta_{[t]})\right) \rtimes Z_0(\widehat{\beta})$.

1537: \end{cor}

1538: 

1539: \begin{proof}

1540:  Since $\widehat{\beta}$ cannot be reducible, the result is a direct

1541: consequence of propositions \ref{P:section_trivial}, \ref{P:section_periodic}

1542: and \ref{P:section_pA}.

1543: \end{proof}

1544: 

1545: % ............................................................

1546: 

1547: \subsection{Structure of $Z_0(\widehat{\beta})$}\label{S:C_0beta}

1548: 

1549: %The only part of theorem~\ref{T:main1} that remains to be shown

1550: %says that, if $\beta$ is a non-periodic

1551: %reducible braid, then $Z_0(\widehat{\beta})$ is isomorphic either to

1552: %$\Z^2$ or to a mixed braid group. This will be shown in this subsection.

1553: The proof of theorem~\ref{T:main1} is now completed by the following result.

1554: \begin{prop}\label{P:C_0beta}

1555: Suppose that $\beta$ is a non-periodic reducible braid, and that its tubular

1556: braid $\widehat{\beta}$ has $m$ strands. Then $Z_0(\widehat{\beta})$ is

1557: isomorphic either to $\Z^2$ or to a mixed braid group on $k$ strands,

1558: where $k\leqslant m$.

1559: \end{prop}

1560: 

1561: \begin{proof}

1562: As usual, there are three subcases, depending wether $\widehat{\beta}$ is

1563: trivial, periodic or \pA.

1564: Recall that we are assuming that $\beta$ is in regular form.

1565: 

1566: Suppose first that $\widehat{\beta}=1$. In this case, $Z(\widehat{\beta})=B_m$.

1567: Hence $Z_0(\widehat{\beta})$ contains any braid whose permutation is

1568: consistent with $\beta$. Denote by ${\cal P}$ the following partition of

1569: $\{P_1,\ldots,P_m\}=\{C_{1,1},\ldots,C_{m,1}\}$: we say that $P_i$ and $P_j$

1570: belong to the same coset of ${\cal P}$ if and only if

1571: $\beta_{[i]}=\beta_{[j]}$. By definition, a braid's permutation is

1572: consistent with $\beta$ if and only if it preserves ${\cal P}$. Therefore,

1573: $Z_0(\widehat{\beta})=B_{\cal P}$, and we are done. (In this case, we have

1574: $k=m$).

1575: 

1576: 

1577: If $\widehat{\beta}$ is {\pA}, it is shown in proposition~\ref{P:section_pA}

1578: that $Z_0(\widehat{\beta})\simeq \Z^2$, so this case is already known.

1579: 

1580: Finally, suppose that $\widehat{\beta}$ is periodic. If it is a power of

1581: $\Delta^2$, then its centralizer is the whole $B_m$, and its corresponding

1582: permutation is trivial, so this case is equivalent to the first one.

1583: 

1584: If $\widehat{\beta}$ is periodic but not a power of $\Delta^2$, then we know

1585: by theorems~\ref{T:centrdelta} and \ref{T:centrdelta2} that

1586: $Z(\widehat{\beta})\simeq B_d(D_*)$, for some $d\geqslant 1$, where $D^*$

1587: is the once punctured disk. But every base point $Q_i$ in $D^*$ corresponds

1588: to an orbit ${\cal C}_i$ of $\widehat{\beta}$ (see figure~\ref{F:theta} in

1589: section~\ref{S:per}), so we can define the following partition ${\cal P}'$

1590: of $\{Q_1,\ldots,Q_d\}$: $Q_i$ and $Q_j$ belong to the same coset if and

1591: only if $\beta_{[i]}=\beta_{[j]}$. This partition lifts by $\theta^{-1}$ to

1592: a partition of $\{P_1,\ldots,P_m\}$, in such a way that any braid in

1593: $B_d(D_*)$ preserves ${\cal P}'$ if and only if its corresponding permutation

1594: in $Z(\widehat{\beta})$ is consistent with $\beta$. Therefore,

1595: $Z_0(\widehat{\beta})\simeq B_{{\cal P}'}(D^*)$. Now it suffices to consider

1596: the central puncture of $D^*$ as another base point, $Q_{d+1}$, and to

1597: notice that $B_{{\cal P}'}(D^*)\cong B_{\cal P}$, where

1598: ${\cal P}={\cal P}'\cup \{\{Q_{d+1}\}\}$. To summarize, in this case we have

1599: $Z_0(\widehat{\beta})\cong B_{{\cal P}'}(D^*)\cong B_{\cal P}$, and the

1600: partition $\cal P$ has $k=d+1$ cosets. Since $d$ must be a proper divisor

1601: of $m$, we get that $k=d+1<m$, and the result follows.

1602: \end{proof}

1603: 

1604: In particular, $Z_0(\widehat{\beta})$ is isomorphic either to $\Z^2$ or to

1605: a mixed braid group. Theorem~\ref{T:main1} is thus proven.

1606: 

1607: % ------------------------------------------------------------------

1608: 

1609: \section{An upper bound for the number of generators}\label{S:upperbound}

1610: 

1611: Once decomposed $Z(\beta)$, if $\beta$ is reducible, as a semi-direct product

1612: of $(Z(\beta_{[1]})\times \cdots \times Z(\beta_{[t]}))$ and

1613: $Z_0(\widehat{\beta})\subset Z(\widehat{\beta})$, we will define a small set

1614: of generators for $Z(\beta)$. We will proceed by induction on the number of

1615: strings, but we need to define first a generating set for

1616: $Z_0(\widehat{\beta})$. We do it as follows:

1617: 

1618: \begin{prop}\label{P:genC_0}

1619:  Let $\beta\in B_n$ be a non-periodic reducible braid, and let

1620: $\widehat{\beta}\in B_m$ be its corresponding tubular braid. Then

1621: $Z_0(\widehat{\beta})$ can be generated by at most $\frac{m(m-1)}{2}$ elements.

1622: \end{prop}

1623: 

1624: \begin{proof}

1625:  If $m=2$ then $Z_0(\widehat{\beta})$ is cyclic, so let us assume that

1626: $m\geqslant 3$. We know by subsection~\ref{S:C_0beta} that

1627: $Z_0(\widehat{\beta})$ is either isomorphic to $\Z^2$ or to a mixed braid

1628: group. The case $\Z^2$ satisfies our result, so we will assume that

1629: $Z_0(\widehat{\beta})$ is isomorphic to a mixed braid group on $k$ strings.

1630: 

1631: Mixed braid groups have been studied in~\cite{Manfredini}, where a

1632: presentation in terms of generators and relations is given. Since we are

1633: mainly interested in the generators, we will extract from those

1634: in~\cite{Manfredini} a small generating set: Let ${\cal P}$ be a partition

1635: of the set $\{1,\ldots,k\}$, having $d$ cosets of length $m_i$

1636: (for $i=1,\ldots,d$). A generating set for $B_{\cal P}$ is given by the

1637: following:

1638: \begin{enumerate}

1639: 

1640:  \item For $i=1,\ldots, d$, a generating set for $B_{m_i}$ (if $m_i>1$).

1641: 

1642:  \item A generating set for the pure braid group $P_d$.

1643: 

1644: \end{enumerate}

1645: It is clear that the first kind of generators corresponds to the movements

1646: of the points inside a coset, while the second one corresponds to the

1647: movement of the points of a coset with respect to those of the others.

1648: For instance, if $k=6$ and ${\cal P}=\{\{1\},\{2,3\},\{4,5,6\}\}$, then one

1649: possible generating set would be:

1650: $$

1651:    \{ \sigma_2\}\cup \{\sigma_4,\;\sigma_5\} \cup \{\sigma_1^2, \;

1652:    \sigma_1\sigma_2\sigma_3^2\sigma_2^{-1} \sigma_1^{-1},\; \sigma_3^2\}.

1653: $$

1654: In order to minimise these generators we recall that $B_2$ is cyclic and,

1655: if $m>2$, then $B_m$ can be generated by two elements. Hence, if we denote

1656: $e_i=m_i-1$ if $m_i< 3$ and $e_i=2$ otherwise, then $e_i$ is a minimal number

1657: of generators for $B_{m_i}$. On the other hand, a minimal number of generators

1658: for $P_d$ is $\frac{d(d-1)}{2}$. Therefore, the minimal number of generators

1659: for $B_{\cal P}$ is:

1660: $$

1661:    g_{\cal P}=\left(\sum_{i=1}^d{e_i}\right)+

1662:    \frac{d(d-1)}{2}\leqslant  \left(\sum_{i=1}^d(m_i-1)\right)+\frac{d(d-1)}{2}

1663: $$

1664: $$

1665:    = k-d+ \frac{d(d-1)}{2}=k+\frac{d(d-3)}{2} \leqslant  k+\frac{k(k-3)}{2}

1666:    =\frac{k(k-1)}{2}.

1667: $$

1668: Notice that if ${\cal P}=\{\{1\},\{2\},\ldots,\{k\}\}$ (so $d=k$), then

1669: $g_{\cal P}=\frac{k(k-1)}{2}$, and this is the worst possibility by the above

1670: formula.

1671: 

1672: Finally we recall from proposition \ref{P:C_0beta} that $k\leqslant m$, so

1673: that $g_{\cal P}\leqslant \frac{m(m-1)}{2}$.

1674: %Now we turn to $Z_0(\widehat{\beta})$. Recall that we are assuming that it

1675: %is isomorphic to a mixed braid groups on $k$ strings, so the result will

1676: %follow if we prove that $k\leqslant m$. We have two cases, depending

1677: %whether $\widehat{\beta}$ is trivial or periodic.

1678: %If $\widehat{\beta}=1$, then we saw in section~\ref{S:C_0beta} that $k=m$

1679: %(this proves that the bound is sharp). If $\widehat{\beta}$ is periodic,

1680: %then either it is a power of $\Delta^2$ -- and in this case $k=m$ again --

1681: %or $Z(\widehat{\beta})\simeq B_d(D^*)$, so $k=d+1$. But in this case,

1682: %$d$ must be a proper divisor of $m$, hence $k=d+1<m$, and the result follows.

1683: \end{proof}

1684: 

1685: 

1686: The first generating set $G'$ of $Z(\beta)$ that we will present

1687: is the following: if $\beta$ is periodic or {\pA}, we have already defined

1688: in sections~\ref{S:per} and \ref{S:pA} a minimal generating set of $Z(\beta)$,

1689: having one or two elements. So suppose that $\beta$ is reducible. Then, by

1690: induction on the number of strings, and by proposition~\ref{P:genC_0}, we

1691: can suppose that we have defined $G_1,\ldots, G_t$ and $G_0$, generating sets

1692: for $Z(\beta_{[1]}), \ldots, Z(\beta_{[t]})$ and $Z_0(\widehat{\beta})$

1693: respectively (if some $\beta_{[i]}$ has one string, then $G_i=\emptyset$).

1694: Then we define $G'=g_1(G_1)\cup \cdots \cup g_t(G_t) \cup h(G_0)$, which is

1695: clearly a generating set for $Z(\beta)$.

1696: 

1697: 

1698: \begin{proof}[Proof of theorem~\ref{T:main}.]

1699: Denote $p(n)$ the upper bound proposed in theorem~\ref{T:main}, that is,

1700: $p(n)=\frac{k(k+1)}{2}$ if $n=2k$ or $p(n)=\frac{k(k+3)}{2}$ if $n=2k+1$.

1701: We will show that the generating set $G'$ defined above has at most

1702: $p(n)$ elements. The case $n=2$ is trivial, so we can suppose that $n>2$ and

1703: that the result is true for any smaller number of strings. We can also assume

1704: that $\beta$ is non-periodic and reducible.

1705: 

1706: The strategy now is to successively replace $\beta$ by different braids,

1707: in such a way that during each replacement step the number of generators

1708: of its centralizer, as given by the above construction, increases.

1709: 

1710: The first modification of $\beta$ will be to replace the tubular braid

1711: $\widehat{\beta}$ by the trivial braid. At the same time, we shall modify the

1712: interior braids, with the aim of rendering them pairwise non-conjugate.

1713: More precisely, we notice that, for any braid $\alpha$ with at least two

1714: strings, the number of generators of $Z(\alpha)$ and $Z(\Delta^{2p}\alpha)$

1715: is the same, while $\Delta^{2p}\alpha$ and $\Delta^{2q}\alpha$ are conjugate

1716: if and only if $p=q$. Thus after multyplying each interior braid $\beta_{[i]}$

1717: by a suitable power of twists $\Delta_{(m_i)}^2$, we can assume that all the

1718: interior braids with at least two strings are pairwise non-conjugate, so that

1719: $t=m$. As seen in the proof of proposition~\ref{P:genC_0}, this first

1720: replacement has increased (or left unchanged) the number of generators of

1721: $G_0$, according to our construction.

1722: 

1723: % Recall that $\widehat{\beta}$ has $m$ strands, while $m_i$ is the number of strands of $\beta_{[i]}$.

1724: % By proposition~\ref{P:genC_0}, the number of elements of $G_0$ is maximised when $\widehat{\beta}$

1725: %is trivial, and all the interior braids are pairwise non-conjugate. Notice also that, for any braid $\alpha$

1726: %with at least two strings, the number of generators of $Z(\alpha)$ and $Z(\Delta^{2p}\alpha)$ is the same, while

1727: %$\Delta^{2p}\alpha$ and $\Delta^{2q}\alpha$ are conjugate if and only if $p=q$. Therefore, we can assume that

1728: %$\widehat{\beta}=1$ and (up to multiplying by some powers of $\Delta^2_{(m_i)}$) that all the interior braids

1729: %with at least two strings are non-conjugate, since this would not decrease the number of elements in $G'$.

1730: %We are thus assuming  that $t=m$.

1731: 

1732: Suppose, without loss of generality, that $m_1=m_2=\cdots=m_d=1$,  that $m_i=2s_i$ for $i=d+1,\ldots,d+u$,

1733: and that $m_i=2s_i+1$, for $i=d+u+1,\ldots,d+u+v$, where $d+u+v=m$. Hence $u$ is the number of interior braids

1734: with an even number of strings, and $v$ is the number of interior braids with an odd (but greater than one) number

1735: of strings. If $d\geqslant 2$, then we shall make further modifications to the braid $\beta$, with the aim

1736: of lowering $d$. More precisely, if $d\leqslant 2$, then we can decrease $d$ by multiplying $\beta$ by

1737: $\sigma_1^p$ for some $p$, where $p$ is chosen in such a way that no other interior braid of $\beta$ equals

1738: $\sigma_1^p$. This replacement increases $u$ by one,

1739: and decreases $d$ by two. Thus the number of generators in $G_0$ decreases by one (if $d=2$) or increases

1740: (if $d>2$). But we would have a new interior braid, $\sigma_1^p$, yielding one new generator. Hence, the total

1741: number of elements in $|G'|$ will not decrease. In other words, without decreasing the number of elements

1742: of $|G'|$ we can replace $\beta$ by a braid with $d\leqslant 1$.

1743: 

1744: %we can decrease $d$ if we multiply $\beta$ by $\sigma_1^p$ for some $p$: if no other

1745: %interior braid of $\beta$ equals $\sigma_1^p$, then we would increase $u$ by one and decrease $d$ by two.

1746: %The number of generators in $G_0$ would decrease by one (if $d=2$) or would increase (if $d>2$). But we would have

1747: %a new interior braid, $\sigma_1^p$, yielding one new generator. Hence, the total number of elements in $|G'|$ would

1748: %not decrease. In other words, we can suppose that $d\leqslant 1$.

1749: 

1750: Denote $a=s_{d+1}+\cdots +s_{d+u}$, $\; b=s_{d+u+1}+\cdots+s_{m}$ and $S=a+b$.

1751: Then one has $n=d+2S+v$. By induction on the number of strings, we have the following bound on the number of

1752: elements in $G'$:

1753: \begin{eqnarray*}

1754: |G'| & \leqslant & \sum_{i=d+1}^{m}{p(m_i)}+ \frac{m(m-1)}{2}  \\

1755:    & = & \sum_{i=d+1}^{d+u}{\frac{s_i(s_i+1)}{2}}+\sum_{i=d+u+1}^m{\frac{s_i(s_i+3)}{2}}+

1756: \left( \begin{array}{c} m \\2 \end{array}\right) \\

1757:   & = &  \sum_{i=d+1}^m{\left( \begin{array}{c} s_i+1 \\2 \end{array}\right)} + \sum_{i=d+u+1}^m{s_i} +

1758: \left( \begin{array}{c} m \\2 \end{array}\right) \\

1759:   & = &  \sum_{i=d+1}^m{\left( \begin{array}{c} s_i+1 \\2 \end{array}\right)} + b +

1760: \left( \begin{array}{c} m \\2 \end{array}\right) \\

1761: \end{eqnarray*}

1762: where $s_i\geqslant 1$ for $i=d+1,\ldots,m$.

1763: 

1764: 

1765:  Given two positive integers $x$ and $y$, one has:

1766: $$

1767:   \left( \begin{array}{c} x+1 \\ 2 \end{array}\right) + \left( \begin{array}{c} y+1 \\ 2 \end{array}\right) =

1768:  \left( \begin{array}{c} x+y+1 \\ 2 \end{array}\right) - xy.

1769: $$

1770: This yields:

1771: \begin{eqnarray*}

1772:  |G'| & \leqslant & \left( \begin{array}{c} S+1 \\ 2 \end{array}\right)

1773:  -\left(\sum_{d+1\leqslant i<j \leqslant m}{s_is_j}\right) + b +

1774: \left( \begin{array}{c} m \\ 2 \end{array}\right).

1775: \end{eqnarray*}

1776: 

1777:  Now we distinguish two cases. If $d=0$, then $m=u+v$ and $n=2S+v$. Also,

1778: \begin{eqnarray*}

1779:  |G'| & \leqslant & \left( \begin{array}{c} S+1 \\ 2 \end{array}\right)

1780:  -\left(\sum_{1\leqslant i<j\leqslant m}{s_is_j}\right) + b +

1781: \left( \begin{array}{c} m \\ 2 \end{array}\right) \\   & \leqslant & \left(

1782: \begin{array}{c} S+1 \\ 2 \end{array}\right)

1783:  -  \left( \begin{array}{c} m \\ 2 \end{array}\right) + b +

1784: \left( \begin{array}{c} m \\ 2 \end{array}\right)

1785: \quad = \quad \frac{S(S+1)}{2} + b.

1786: \end{eqnarray*}

1787: 

1788: 

1789: If $v=0$ one has $b=0$, so $S=a$ and

1790: $|G'|\leqslant \frac{S(S+1)}{2}=\frac{a(a+1)}{2}$; but also $n=2k=2a$,

1791: so $p(n)=\frac{a(a+1)}{2}$ and we are done.

1792: 

1793: If $v=1$ then $n=2S+1$, hence $k=S$ and $p(n)=\frac{S(S+3)}{2}$. But in this case

1794: $|G'|\leqslant \frac{S(S+1)}{2}+b\leqslant \frac{S(S+1)}{2}+S = \frac{S(S+3)}{2}=p(n)$.

1795: 

1796: If $v\geqslant 2$, since $n=2S+v$ one has $k\geqslant S+1$. Then $|G'|\leqslant \frac{S(S+1)}{2} + b

1797: < \frac{S(S+1)}{2} + (S+1) =\frac{(S+2)(S+1)}{2} \leqslant \frac{k(k+1)}{2} \leqslant p(n)$.

1798: 

1799: Therefore, the result is true if $d=0$. Suppose now that $d=1$. In this case $m=u+v+1$ and

1800: $n=2S+v+1$. Then one has:

1801: \begin{eqnarray*}

1802:  |G'| & \leqslant & \left( \begin{array}{c} S+1 \\ 2 \end{array}\right)

1803:  -\left(\sum_{2\leqslant i<j\leqslant m}{s_is_j}\right) + b +

1804: \left( \begin{array}{c} m \\ 2 \end{array}\right) \\

1805: & \leqslant & \left( \begin{array}{c} S+1 \\ 2 \end{array}\right)

1806:  -  \left( \begin{array}{c} m-1 \\ 2 \end{array}\right) + b +

1807: \left( \begin{array}{c} m \\ 2 \end{array}\right) \\

1808:  & = & \frac{S(S+1)}{2} + b + m-1 \quad = \quad \frac{S(S+1)}{2} + b+u+v \\

1809:  & \leqslant & \frac{S(S+1)}{2} + S+ v  \quad = \quad  \frac{S(S+3)}{2} + v.

1810: \end{eqnarray*}

1811: 

1812: 

1813: 

1814: If $v=0$ then $b=0$ and $\;k=S$, so $|G'|\leqslant \frac{S(S+3)}{2} = p(n)$.

1815: 

1816: If $v=1$ then $n=2S+2$ and $k=S+1$. Then $|G'|\leqslant \frac{S(S+3)}{2}+1=

1817: \frac{(S+1)(S+2)}{2}=p(n)$.

1818: 

1819: If $v=2$ then $n=2S+3$ and $k=S+1$. Then $|G'|\leqslant \frac{S(S+3)}{2}+2=

1820: \frac{S^2+3S+4}{2} <\frac{(S+1)(S+4)}{2}=p(n)$.

1821: 

1822: 

1823: Finally, if $v\geqslant 3$ then $n=2S+v+1$ so $k\geqslant S+v/2$. Hence

1824: $$

1825:   p(n)\; \geqslant \; \frac{(S+v/2)(S+v/2+1)}{2} \;=

1826:   \;\frac{S^2+(v+1)S+v(v+2)/4}{2}

1827: $$

1828: $$

1829:   \geqslant \; \frac{S(S+3)}{2} + S/2 + v/2

1830:   \; > \; \frac{S(S+3)}{2}+ v \; \geqslant \; |G'|.

1831: $$

1832: Therefore, in every case $|G'|\leqslant p(n)$, and theorem~\ref{T:main} is

1833: proved.

1834: \end{proof}

1835: 

1836: 

1837: Recall that, in example~\ref{E:purebraid}, %(which is due to S. J. Lee),

1838: we defined braids of any number of strands whose centralizer could not be

1839: generated by less than $p(n)$ elements. Therefore, the bound given by

1840: theorem~\ref{T:main} is the best possible one.

1841: 

1842: % ------------------------------------------------------------------

1843: 

1844: \section{Small generating sets}\label{S:minimal}

1845: 

1846: We saw in the previous section an upper bound for the number of generators of the

1847: centralizer of a braid $\beta$, in terms of its number of strings. But one could obtain a

1848: better bound if more information about $\beta$ is given. In this section we will define a

1849: new generating set $G$ for $Z(\beta)$, which is in most cases smaller than the set $G'$

1850: defined before. It is also the smallest possible ``natural'' generating set, in the sense

1851: that each generator belongs to one of the $t+1$ factors in the semidirect product

1852: decomposition in theorem \ref{T:main1}(c). Thus in a philosophical sense, $G$ is the

1853: ``right'' generating set, even though it is not in general the smallest possible one, as we

1854: shall see at the end of this section.

1855: 

1856: If $\beta$ is periodic or {\pA}, we already know a minimal generating set, with at most two

1857: elements. We also know a minimal generating set for any mixed braid group (see the proof of

1858: proposition~\ref{P:genC_0}). Hence we can define $G$ by induction on the number of strands,

1859: when $\beta$ is a reducible, non-periodic braid. We can also suppose that $\beta$ is in

1860: regular form. We recall that the interior braids are denoted $\beta_{[1]},\ldots,

1861: \beta_{[t]}$, and the tubular braid $\widehat{\beta}$.

1862: 

1863: \begin{definition}

1864: We will say that $i,j\in \{1,\ldots, t\}$ are {\em permutable}

1865: if there exists some $\eta\in Z_0(\widehat{\beta})$ such that

1866: $\eta({\cal C}_i)={\cal C}_j$.

1867: \end{definition}

1868: 

1869: Remark that permutability is an equivalence relation,

1870: and the definition of $Z_0(\widehat\beta )$ says that if $i$ and $j$

1871: are permutable then $\beta_{[i]}=\beta_{[j]}$.

1872: 

1873: Let then $\{i_1,\ldots,i_r\}\subset\{1,\ldots,t\}$ be coset representatives

1874: for permutability. Let $G_{i_k}$ be a minimal set of generators for

1875: $Z(\beta_{[i_k]})$, and $G_0$ be a minimal set of generators for

1876: $Z_0(\widehat{\beta})$. Then we define

1877: $G=g_{i_1}(G_{i_1})\cup \cdots \cup g_{i_r}(G_{i_r}) \cup h(G_H)$.

1878: Notice that $G\subset G'$, and they coincide if and only if there is no

1879: pair of permutable indices.

1880: 

1881: \begin{prop}

1882: $G$ is a generating set of $Z(\beta)$.

1883: \end{prop}

1884: 

1885: \begin{proof}

1886: From the exact sequence of theorem~\ref{T:exactseq} it follows that,

1887: if $G_i$ is a set of generators for $Z(\beta_{[i]})$, then a set of

1888: generators for $Z(\beta)$ is

1889: $G'=g_{1}(G_{1})\cup \cdots \cup g_{t}(G_{t}) \cup h(G_0)$. Hence, we just

1890: need to show that if $j\in\{1,\ldots, t\}\backslash \{i_1,\ldots,i_r\}$,

1891: then every element in $g_j(G_j)$ can be written as a product of elements

1892: in $G$.

1893: 

1894: Take then $j$ as above. There must be some $i_k$ permutable with $j$, so $\beta_{[j]}=\beta_{[i_k]}$

1895: and there is some $\eta\in Z_0(\widehat{\beta})$ such that $\eta({\cal C}_{i_k})={\cal C}_{j}$.

1896: Notice that $G_j$ is a set of generators for $Z(\beta_{[j]})=Z(\beta_{[i_k]})$, so every $\gamma\in G_j$

1897: can be written as a product of elements in $G_{i_k}$. Hence the braid

1898: $\alpha=h(\eta)^{-1} g_{i_k}(\gamma) h(\eta)$ can be written as a product of elements in $G$.

1899: Moreover, one has $\widehat{\alpha}=\widehat{h(\eta)}^{-1} 1 \widehat{h(\eta)} = 1$, and the

1900: only nontrivial interior braids in $\alpha$ are those corresponding to ${\cal C}_j$. Since the

1901: interior braids $h(\eta)_{i_k,l}$ for every $l$  are just powers of $\beta_{[i_k]}=\beta_{[j]}$,

1902: and $\gamma$ commutes with $\beta_{[j]}$, it follows that for every $l$, $\alpha_{j,l}=\gamma$.

1903: Therefore $\alpha=g_j(\gamma)$, so every element in $g_j(G_j)$ can be written as a product of elements

1904: in $G$, thus $G$ is a generating set for $Z(\beta)$.

1905: \end{proof}

1906: 

1907: The generating set we have just defined is, unfortunately, not always the

1908: smallest possible one:

1909: 

1910: \begin{example}\rm

1911: Consider the five string braid

1912: $\beta = \sigma_3 \sigma_4 \sigma_2 \sigma_3 \sigma_1

1913: \sigma_2 \sigma_2 \sigma_3 \sigma_4 \sigma_1 \sigma_2 \sigma_3$ --

1914: the canonical reduction system of this braid has two round circles,

1915: one containing punctures number 1, 2 and 3, the other punctures number

1916: 4 and 5; the tubular braid is just a full twist of the two fat

1917: strings: $\widehat{\beta} = \sigma_1^2$. Moreover, the interior braids of

1918: each tube is trivial. According to theorem \ref{T:main1}, the centralizer

1919: of this braid is

1920: $$

1921: Z(\beta)\cong (B_3 \times B_2) \rtimes PB_2 \cong (B_3 \times \Z)\rtimes \Z

1922: $$

1923: and the generating set constructed in this section has four elements:

1924: two for $B_3$, and one for each factor $\Z$. We now claim that this

1925: generating set is not as small as possible.

1926: 

1927: Indeed, $B_3 \times \Z$ can be generated by only two elements (and thus

1928: $Z(\beta)$ can be generated by three elements). To see this, recall that

1929: the 3-string braid group is isomorphic to the group of the $(2,3)$-torus knot.

1930: Thus $B_3$ has a presentation $\langle y,z\ |\ y^3z^{-2} = 1\rangle$

1931: (with $y=\sigma_1\sigma_2$ and $z=\sigma_1 \sigma_2 \sigma_1$).

1932: Moreover, the factor $\Z$ is generated by $\sigma_4$. Now the two generators

1933: $(y,\sigma_4)$ and $(z,\sigma_4)$ generate $B_3\times \Z$, because

1934: $(1,\sigma_4)$ can be written as $(y,\sigma_4)^3 (z,\sigma_4)^{-2}$.

1935: \end{example}

1936: 

1937: % ------------------------------------------------------------------

1938: 

1939: \section{Some algorithmic aspects}\label{S:alg}

1940: 

1941: The aim of this section is to present the essential ingredients for an

1942: algorithm which, for any given braid, finds a generating set of its

1943: centralizer subgroup that matches the description of the previous sections.

1944: Since, for any braid $\beta$ and any $k\in \Z$, the centralizer subgroups of

1945: $\beta$ and $\beta \Delta^{2k}$ coincide, we can always assume that $\beta$

1946: is positive.

1947: 

1948: We start by mentioning that algorithms that perform the Nielsen-Thurston

1949: classification, and give the invariant folitations in the {\pA} case (in the

1950: form of train tracks), are available -- notably, there are Bestvina-Haendel's

1951: \cite{BestvinaHaendel} and of Los' \cite{Los} algorithms; and computer

1952: implementations are available on the web \cite{Brinkmann, Hall}.

1953: 

1954: We recall briefly the idea of the two automatic structures on braid groups

1955: that are relevant for us: for the first one, given by Garside \cite{Garside}

1956: and Thurston \cite{ThBnaut} (and refined by El-Rifai and Morton \cite{EM}),

1957: we think of $D_n$ has having the $n$ punctures lined up on the real line

1958: in the disk $D$. For the second one, given by Birman, Ko, and Lee \cite{BKL},

1959: we think of $B_n$ as having the $n$ punctures regularly spaced on the circle

1960: of radius $1$. Apart from that, the structures are exactly analogue.

1961: In the Garside-Thurston structure, there is a canonical way to write $\beta$

1962: as a product of divisors of $\Delta$, namely by pushing each crossing between

1963: two strings into a factor as far to the  left as possible. This normal form

1964: is called the {\sl left greedy normal form}. For instance, in this normal

1965: form all factors which are {\it equal} to $\Delta$ (not just divisors of it)

1966: are grouped together at the very left of the product decomposition.

1967: Analogously, Birman-Ko-Lee write

1968: each braid as a product of divisors of $\delta$ in a left-greedy way.

1969: If $\beta$ is a positive braid, then its {\sl super summit set} is the

1970: subset of all elements $\alpha$ of its conjugacy class which satisfy the

1971: following conditions:

1972: \begin{enumerate}

1973: \item[(i)] $\alpha$ is positive,

1974: \item[(ii)] the writing of $\alpha$ in left greedy normal form has as

1975: few factors as possible among all elements satisfying (i),

1976: \item[(iii)]  the writing of of $\alpha$ in left greedy normal form has

1977: as many factors on the left as possible equal to $\Delta$ (or $\delta$),

1978: among all elements satisfying (i) and (ii).

1979: \end{enumerate}

1980: Two positive elements of $B_n$ are conjugate if and only if their super

1981: summit sets coincide. Given $\beta\in B_n$ there is an algorithm, given

1982: in~\cite{FrGM_conj} (which is an improvement of the algorithm in~\cite{EM}),

1983: to compute its super summit set. It is as follows: first we repeatedly

1984: {\sl cycle} $\beta$ (i.e.~move the first factor different from $\Delta$,

1985: respectively $\delta$, to the end and calculate the left greedy form of the

1986: resulting braid), until this process runs into a loop. At this point we are

1987: guaranteed to have achieved condition (ii) above. Then we repeatedly

1988: {\sl decycle} (i.e.~move the last factor to the front

1989: and calculate the left greedy form of the resulting braid) until we run

1990: into a loop. Then all elements of this loop belong to the super summit set.

1991: Afterwards, all other elements of the super summit set can be found

1992: recursively by conjugating already known elements by (suitable) divisors of

1993: $\Delta$ (respectively $\delta$), and retaining the result if it belongs to

1994: the super summit set.

1995: 

1996: This algorithm for computing the super summit set is necessary for our

1997: purposes. Now suppose we are given a braid $\beta\in B_n$ and we want to

1998: compute its centralizer. First we need to determine if $\beta$ is periodic,

1999: reducible or {\pA}, and then we can use the results in this paper.

2000: 

2001: \begin{remark}

2002: Very recently, V. Gebhardt~\cite{Gebhardt} presented a better algorithm for

2003: the conjugacy problem in braid groups. He defined the {\em ultra summit set},

2004: which is in general much smaller than the super summit set described here.

2005: \end{remark}

2006: 

2007: % ..................................................................

2008: 

2009: \sh{Periodic elements}

2010: 

2011: Deciding whether a given element $\beta$ of $B_n$ is periodic is very

2012: easy: one calculates the $n-1$st and the $n$th power of $\beta$. Then $\beta$

2013: is periodic if and only if one of the two results is a power of $\Delta^2$.

2014: 

2015: If $\beta^{n-1}=\Delta^{2k}$ for some $k\in \N$, then $\beta$ is conjugate

2016: to $\gamma_{(n)}^k$ (as can be easily seen from lemma~\ref{L:per_gamma_delta}),

2017: and a conjugating element can be found explicitly using either of the

2018: two standard algorithms. Similarly, if $\beta^n=\Delta^{2k}$, then $\beta$

2019: is conjugate to $\delta_{(n)}^k$, and either algorithm yields an

2020: explicit conjugating element. In either case, one can find explicitly

2021: a generating set of the centralizer subgroup with only two elements, using

2022: propositions \ref{P:centrdelta'} or \ref{P:centrgamma}.

2023: 

2024: % ..................................................................

2025: 

2026: \sh{Finding reducing curves of reducible elements}

2027: 

2028: After establishing that an element $\beta$ of $B_n$ is not periodic, we need

2029: to check whether it is reducible, and if it is, we want to find explicitly

2030: an invariant multicurve. This is, in fact, a standard part of

2031: Bestvina-Haendel's \cite{BestvinaHaendel} and of Los' \cite{Los}

2032: algorithms.

2033: 

2034: We want to point out one particulary elegant alternative, which is due to

2035: Benardete, Gutierrez and Nitecki \cite{BGN2} (see also~\cite{BGN}). We think of

2036: $D_n$ as having the $n$ punctures lined up horizontally, and we look at

2037: Garside-Thurston's left greedy normal form. The key observation from

2038: \cite{BGN2} is the following: suppose that $C$ is an invariant multicurve of

2039: a braid $\beta$, and that the normal form of $\beta$ is

2040: $\beta=\beta_1\cdot\ldots\cdot \beta_k$, where $\beta_1,\ldots,\beta_k\in B_n$

2041: are divisors

2042: of $\Delta$. Moreover, suppose that all components of $C$ are {\sl round}

2043: (i.e.~actual geometric circles in $D_n$). Then we have not only that

2044: $\beta_1\cdot\ldots\cdot\beta_n(C)=C$, but also that all components of all the

2045: multicurves $\beta_1\cdot\ldots\cdot\beta_i(C)$ are round for $i=1,\ldots,k$.

2046: 

2047: As remarked in \cite{BGN2} this implies as a corollary that invariant

2048: multicurves are visible as round curves in the super summit set of $\beta$,

2049: and in particular the reducibilty of a braid is easily detectable from the

2050: super summit set. To prove the corollary we note that $\beta$ has a conjugate

2051: in which all components of the curve system $C$ are round; moreover,

2052: $\beta$ and its conjugate have the same super summit set. Now cycling and

2053: decycling this conjugate does not change the fact that there  is a {\it round}

2054: invariant curve system, by the key observation above. At the end of the

2055: cycling/decycling procedure we have found elements of the super summit set

2056: which contain the desired round invariant curves.

2057: 

2058: Now it is shown in~\cite{BGN2} how to determine if a

2059: given braid preserves a system of disjoint round curves. And there is a finite

2060: number of these systems. Moreover, since for each element of the super summit

2061: set we know how it can be conjugated to obtain $\beta$, we can find explicitly

2062: all curves that belong to a reduction system for $\beta$. We can then easily

2063: determine, by its definition, which of these curves belong to the canonical

2064: reduction system of $\beta$. That is, we can compute the canonical reduction

2065: system of $\beta$.

2066: 

2067: By the results in this paper, $Z(\beta)$ is then a semi-direct product of two

2068: groups that can be computed by induction on the number of strings. Hence, it

2069: only remains to study the case when $\beta$ is {\pA}.

2070: 

2071: 

2072: % ..................................................................

2073: 

2074: \sh{Pseudo-Anosov elements: commutation with $\delta_{(n)}^k$}

2075: 

2076: Suppose that our braid $\beta$ fails the tests of periodicity and reducibility,

2077: hence it is known to be {\pA}. We need to check if it commutes with a

2078: periodic braids other than powers of $\Delta^2$.

2079: 

2080: We shall think of $D_n$ as having its $n$ punctures uniformly distributed

2081: over the circle of radius $1$, and we consider Birman-Ko-Lee's

2082: left-greedy normal form. We want to decide

2083: algorithmically whether $\beta$ is conjugate to a braid $\alpha$ with

2084: the property that $\alpha$ commutes with $\delta_{(n)}^k$ for some

2085: positive integer $k<n$. If it is, we want to know the conjugating braid

2086: explicitly. The following result yields such an algorithm.

2087: %

2088: \begin{prop}\label{P:commdelta}

2089: Suppose that a {\pA} braid $\beta$ has a conjugate which commutes with

2090: $\delta_{(n)}^k$ for some integer $k$. Then there exists an element $\alpha$

2091: of the super summit set of $\beta$ which has the property that $\alpha$,

2092: and in fact every factor of the left greedy normal form of $\alpha$,

2093: commutes with $\delta_{(n)}^k$.

2094: \end{prop}

2095: %

2096: \begin{proof} Let $\beta'$ be a conjugate of $\beta$ which commutes with

2097: $\delta_{(n)}^k$. If $\beta'=\beta'_1\cdot\ldots\cdot\beta'_r$ is the

2098: left-greedy normal form of $\beta'$, then each factor $\beta'_i$ is a divisor

2099: of $\delta_{(n)}$ which is $\frac{2\pi k}{n}$-symmetric. This follows from

2100: the fact that the very definition of the left-greedy normal form is

2101: completely rotation symmetric. More precisely, the fact that two

2102: consecutive factors $\beta'_i \beta'_{i+1}$ determine a left-greedy normal

2103: form is not modified by rotating them. Hence, the product

2104: $(\delta_{(n)}^{-k} \beta'_1 \delta_{(n)}^k ) \cdots

2105: (\delta_{(n)}^{-k} \beta'_r\delta_{(n)}^k)$ is in left-greedy normal form.

2106: Since this product equals $\delta_{(n)}^{-k} \beta'\delta_{(n)}^k=\beta'$,

2107: whose left-greedy normal form is $\beta'_1\cdots \beta'_k$, we obtain that

2108: $\delta_{(n)}^{-1} \beta'_i \delta_{(n)}=\beta'_i$, for $i=1,\ldots,r$.

2109: 

2110: Using the same argument inductively, we see that the cycling and decycling

2111: procedure only ever creates braids in left greedy normal form in which all

2112: factors are $\frac{2\pi k}{n}$-symmetric.

2113: \end{proof}

2114: 

2115: Now we notice that it is very easy to decide if a given divisor of $\delta$ (in

2116: the Birman-Ko-Lee context) is invariant under a given rotation. Hence one

2117: can determine if a braid commutes with an (explicitly computable) conjugate of

2118: $\delta_{(n)}^k$ by looking at the elements of its super summit set.

2119: 

2120: 

2121: 

2122: % ..................................................................

2123: 

2124: \sh{Pseudo-Anosov elements: commutation with $\gamma_{(n)}^k$}

2125: 

2126:  Now we want to determine if a given {\pA} braid commutes with a conjugate

2127: of $\gamma_{(n)}^k$, for a given positive integer $k<n-1$. This is only

2128: possible if there is some index $i\in \{1,\ldots,n\}$ such that $\beta$

2129: preserves $P_i$, as can be easily seen by looking at the corresponding

2130: permutations.

2131: 

2132: Call ${\cal P}_i=\{\{P_i\},\{P_1,\ldots,P_{i-1},P_{i+1},\ldots,P_n\}\}$, a

2133: partition of $\{P_1,\ldots,P_n\}$. Then $\beta$ should belong to

2134: $B_{{\cal P}_i}$. There is a natural map

2135: $f_i\co B_{{\cal P}_i} \rightarrow B_{n-1}$ which consists of forgetting

2136: the $i$th string. Notice that, if a braid $\alpha$ commutes with

2137: $\gamma_{(n)}^k$ (where $P_1$ is considered to be the central point of

2138: $D_{(n)}$) then $f_1(\alpha)$ commutes with

2139: $f_1(\gamma_{(n)}^k)=\delta_{(n-1)}^k$.

2140: 

2141: Hence we have a necessary condition that must be satisfied. If $\beta$

2142: preserves a puncture $P_i$, then we conjugate it to some $\alpha$ that

2143: preserves $P_1$, and we test whether a conjugate of  $f_1(\alpha)$ commutes

2144: with $\delta_{(n-1)}^k$ for some $k<n-1$. If this does not happen, for

2145: $i=1,\ldots, n$, then no conjugate of $\beta$ commutes with $\gamma_{(n)}^k$.

2146: 

2147: This necessary condition is of course not sufficient. A sufficient

2148: and testable condition is now given by the following result. Recall that, by

2149: corollary~\ref{C:addstring}, there is an isomorphism

2150: $\chi=(\bar{\theta}^*)^{-1}\theta^*$ from

2151: $Z(\delta_{(n-1)}^k)$ to $Z(\gamma_{(n)}^k)$, given by adding a trivial

2152: string at the centre of $D_{n-1}$. Notice that, if

2153: $\zeta\in Z(\gamma_{(n)}^k)$, then $\chi(f_1(\zeta))=\zeta$. Then one has:

2154: 

2155: \begin{prop}

2156: Suppose that $\alpha\in B_n$ preserves $P_1$, and $\tld\alpha=f_1(\alpha)$

2157: commutes with $\delta_{(n-1)}^k$. Then the following two statements are

2158: equivalent.

2159: \begin{enumerate}

2160: \item[(i)] $\alpha$ is conjugate to an element $\zeta$ of $B_n$

2161: which commutes with $\gamma_{(n)}^k$, and the conjugating homeomorphism

2162: preserves $P_1$.

2163: \item[(ii)]  $\alpha$ is conjugate to $\chi(\tld\alpha)$.

2164: \end{enumerate}

2165: \end{prop}

2166: 

2167: 

2168: \begin{proof}

2169: The implication (ii)$\Rightarrow$(i) is immediate, by choosing

2170: $\zeta := \chi(\tld\alpha)$.

2171: 

2172: For the implication (i)$\Rightarrow$(ii), we suppose that (i) holds, that is,

2173: there is an element $\eta\in B_{{\cal P}_1}$ such that

2174: $\eta^{-1} \alpha \eta= \zeta$, where $\zeta\in Z(\gamma_{(n)}^k)$. We can

2175: apply $f_1$ to all these elements, denoting $\tld\eta=f_1(\eta)$ and

2176: $\tld\zeta=f_1(\zeta)$.

2177: This yields  $(\tld\eta)^{-1} \tld\alpha \:\tld\eta = \tld\zeta$,

2178: where $\tld\alpha, \tld\zeta \in Z(\delta_{(n-1)}^k)$.

2179: 

2180: If we show that $\tld\eta\in Z(\delta_{(n-1)}^k)$, then we can apply $\chi$ to

2181: all factors, obtaining

2182: $\chi(\tld\eta)^{-1} \chi(\tld\alpha) \:\chi(\tld\eta) =

2183: \chi(\tld\zeta)=\zeta$, hence $\chi(\tld\alpha)$ is conjugate to $\zeta$

2184: which is conjugate to $\alpha$, and the result follows.

2185: 

2186: Let us then show that $\tld\eta$ commutes with $\delta_{(n-1)}^k$.

2187: Notice that $\zeta$ is a {\pA} braid that commutes with $\gamma_{(n)}^k$.

2188: Hence it preserves a projective foliation ${\cal F}_{\zeta}$, which is

2189: invariant under a rotation by an angle of $\frac{2\pi k}{n-1}$. But in this

2190: case $\tld\zeta$ also preserves ${\cal F}_{\zeta}$, with the same

2191: stretch factor, hence it is also {\pA}. Since $\tld\alpha$ is conjugated

2192: to $\tld\zeta$, then it is {\pA} as well, and we call ${\cal F}_{\tld\alpha}$

2193: its corresponding projective foliation (which is also invariant under the

2194: same rotation, since $\tld\alpha$ commutes with $\delta_{(n-1)}^k$).

2195: Since $(\tld\eta)^{-1} \tld\alpha \:\tld\eta = \tld\zeta$, we have that

2196: $\tld\eta$ sends ${\cal F}_{\tld\alpha}$ to ${\cal F}_{\zeta}$.

2197: 

2198: Now consider the braid $d=\tld\eta^{-1} \delta_{(n-1)}^k \tld\eta$.

2199: It is conjugate to $\delta_{(n-1)}^k$, and hence periodic.

2200: Moreover, it preserves ${\cal F}_{\zeta}$, so

2201: it commutes with $\tld\zeta$. But the periodic elements in the centralizer

2202: of $\tld\zeta$ form a cyclic group containing $\delta_{(n-1)}^k$, and

2203: $\delta_{(n-1)}^k$ is the only element having exponent sum $(n-2)k$. Since

2204: $d$ has exactly the same exponent sum, it follows that $d=\delta_{(n-1)}^k$.

2205: Hence $\tld\eta$ commutes with $\delta_{(n-1)}$, and the result follows.

2206: \end{proof}

2207: 

2208: An algorithm for testing whether a braid $\beta$ is conjugate to a braid

2209: which commutes with $\gamma_{(n)}^k$ is now easy to construct: for each of

2210: the $n$ punctures test whether the puncture is fixed by $\beta$, and whether

2211: forgetting this puncture yields a braid which is conjugate to a braid

2212: $\tld{\alpha}$ that commutes with $\delta_{(n)}^k$. (We know how to do

2213: this, by the results of the previous subsection). For each puncture

2214: that does satisfy this property, test whether $\chi(\tld\alpha)$  (which is

2215: obtained from $\tld{\alpha}$ by adding a ``trivial'' string in the

2216: centre), is conjugate to $\beta$. If, for one of the punctures, this is the

2217: case, then the answer is ``yes'', otherwise ``no''.

2218: 

2219: % ..................................................................

2220: 

2221: \sh{Pseudo-Anosov elements: finding roots}

2222: 

2223: It remains to describe a last step for computing a generating set for

2224: $Z(\beta)$, when $\beta$ is {\pA}. We assume that we have already computed

2225: the subgroup $\langle \rho \rangle$ of periodic braids commuting with $\beta$.

2226: Then we can multiply $\beta$ by a suitable power of $\rho$, to obtain a braid

2227: $b$ that preserves the singular leaves of the projective foliations

2228: corresponding to $\beta$. Then we know that

2229: $Z(\beta)=\langle \alpha \rangle \times \langle \rho \rangle$, where $\alpha$

2230: is the smallest possible root of $b$.

2231: 

2232: The last problem, therefore, is to determine whether a given {\pA} braid $b$

2233: has a $k$th root, for given $k$, and to compute that root. This problem has

2234: been solved in~\cite{Sty} (generalised to all Garside groups in \cite{Sibert}).

2235: Moreover, since the number of possible values of $k$ is finite (we are

2236: assuming that $b$ is positive), we have an algorithm for computing $\alpha$,

2237: thus a generating set for $Z(\beta)$.

2238: 

2239: {\bf Acknowledgements } We are grateful to a number of people for discussions

2240: and valuable ideas that greatly contributed to this research. The examples of

2241: Nikolai V.\ Ivanov \cite{IvanovTalk,IvanovRecent}, which we learned about

2242: through discussions with Mustafa Korkmaz, were an important inspiration and

2243: greatly helped us clarify our ideas. It was thus from Ivanov (via Korkmaz)

2244: that we learned that the number of generators may have to grow quadratically

2245: with the number of strings, contradicting a conjecture in \cite{GMFr}. We are

2246: very grateful to Sang Jin Lee who later, but independently of Ivanov, came up

2247: with his examples, conjectured that they represent the worst case, and kindly

2248: communicated these ideas to us by email. (Hessam Hamidi-Teherani found the

2249: same examples as Lee immediately after listening to Ivanov's talk, but we

2250: didn't learn this until very recently.) We also thank David Bessis for useful

2251: discussions, and Joan Birman for telling us about the references~\cite{BGN}

2252: and \cite{BGN2}.

2253: 

2254: 

2255: 

2256: % ------------------------------------------------------------------

2257: 

2258: \begin{thebibliography}{999}

2259: 

2260: \bibitem{Arnold} V. I. Arnold, The cohomology ring of the group of dyed

2261: braids, Mat.Zametki, V. 5 (1969), 227-231.

2262: 

2263: \bibitem{BGN} D.~Benardete, M.~Gutierrez, Z.~Nitecki, A combinatorial

2264: approach to reducibility of mapping classes, Contemporary Math. 150

2265: (1993), 1--31.

2266: 

2267: \bibitem{BGN2} D.~Benardete, M.~Gutierrez, Z.~Nitecki, Braids and the

2268: Nielsen-Thurston classification, J. Knot Theory and its Ramifications

2269: 4 (1995), 549-618.

2270: 

2271: \bibitem{BDM} D.~Bessis, F.~Digne, J.~Michel, Springer theory in braid

2272: groups and the Birman-Ko-Lee monoid, Pacific J. Math. 205 (2002), no. 2,

2273: 287--309.

2274: 

2275: \bibitem{BestvinaHaendel} M.~Bestvina, M.~Haendel, Train-tracks for surface

2276: homeomorphisms, Topology 34 (1995), 109--140.

2277: 

2278: \bibitem{Bir} J.~Birman, Braids, links, and mapping class groups, Annals

2279: of Math.~Studies, 82, Princeton University Press 1975.

2280: 

2281: \bibitem{BKL} J.~Birman, K.~H.~Ko, S.~J.~Lee, A new approach to the word

2282: and conjugacy problems in the braid groups, Adv. Math. 139 (1998), 322--353.

2283: 

2284: \bibitem{BLM} J.~Birman, A.~Lubotzky and J.~McCarthy,

2285: Abelian and solvable subgroups of the mapping class groups, Duke Math.J.\ 50

2286: (1983), 1107-1120.

2287: 

2288: \bibitem{Brinkmann} P.~Brinkmann, An implementation of the Bestvina-Handel

2289: algorithm for surface homeomorphisms,  Experiment. Math. 9 (2000), 235--240.

2290: Computer program available at

2291: {\tt www.math.uiuc.edu/\~{ }brinkman/software/train/}

2292: 

2293: \bibitem{coko} A. Constantin, B. Kolev, The theorem of

2294: Ker\'{e}kj\'{a}rt\'{o} on periodic homeomorphisms of the disc and the

2295: sphere, Enseign. Math (2) 40 (1994) No 3-4, 193-204.

2296: %Preprint (2003) {\tt arXiv:math.GN/0303256}.

2297: 

2298: \bibitem{Eilenberg} S.~Eilenberg, Sur les transformations p\'eriodiques de

2299: la surface de sph\`ere, Fund.~Math.~22 (1934), 28--41.

2300: 

2301: \bibitem{EM} E.~A.~El-Rifai, H.~R.~Morton, Algorithms for positive braids,

2302: Quart.~J. Math. Oxford Ser.~(2) 45 (1994), no.~180, 479--497.

2303: 

2304: \bibitem{FLP} A.~Fathi, F.~Laudenbach, V.~Poenaru, Travaux de Thurston sur

2305: les surfaces - seminaire Orsay, Asterisque 66--67, Soci\'et\'e Math.~de

2306: France, 1991.

2307: 

2308: \bibitem{FRZ} R. Fenn, D. Rolfsen, J. Zhu, Centralisers in the braid group and

2309: singular braid monoid, l'Enseignment Math\'ematique 42 (1996), 75--96.

2310: 

2311: \bibitem{FrGM_conj} N.~Franco, J.~Gonz\'alez-Meneses, Conjugacy problem for braid

2312: groups and Garside groups. To appear in Journal of Algebra.

2313: 

2314: \bibitem{GMFr} N.~Franco, J.~Gonz\'alez-Meneses,  Computation of centralizers

2315: in braid groups and Garside groups. To appear in Revista Matem\'atica

2316: Iberoamericana.

2317: 

2318: \bibitem{Garside} F.~A.~Garside, The braid group and other groups,

2319: Quart.~J.~Math.~Oxford 20 (1969), 235--254.

2320: 

2321: \bibitem{Gebhardt} V. Gebhardt, A New Approach to the Conjugacy Problem in

2322: Garside Groups, Preprint (2003). {\tt arXiv:math.GT/0306199}.

2323: 

2324: \bibitem{Hall} T. Hall, Computer implementation of Bestvina-Handel algorithm

2325: available at

2326: {\tt www.liv.ac.uk/maths/PURE/MIN\_SET/CONTENT/members/T\_Hall.html}

2327: 

2328: \bibitem{Ivanov} N.~V.~Ivanov, Subgroups of Teichm\"uller modular groups,

2329: Translations of mathematical monographs vol.~115 (1992), AMS.

2330: 

2331: \bibitem{IvanovTalk} N.~V.~Ivanov, talk at the special session ``Mapping

2332: class groups and the geometric theory of Teichm\"uller spaces'' at the

2333: $974^{\rm th}$ meeting of the AMS, Ann Harbour, MI, March 1--3, 2002.

2334: 

2335: \bibitem{IvanovRecent} N.~V.~Ivanov, Examples of centralizers in the Artin

2336: braid groups, Preprint (2003), {\tt arXiv:math.GT/0306418}

2337: 

2338: \bibitem{deK} B.~de Ker\'ekj\'art\'o, \"Uber die periodischen Transformationen

2339: der Kreisscheibe und der Kugelfl\"ache, Math.~Annalen 80 (1919), 3--7.

2340: 

2341: \bibitem{Los} J.~Los, Pseudo-Anosov maps and invariant train tracks in the

2342: disc: a finite algorithm, Proc. London Math. Soc. (3) 66 (1993), 400--430.

2343: 

2344: \bibitem{Makanin} G.~S.~Makanin, On normalizers in the braid group,

2345: Mat.Sb.~86 (128), 1971, 171--179.

2346: 

2347: \bibitem{Manfredini} S.~Manfredini, Some subgroups of Artin's braid group.

2348: Special issue on braid groups and related topics (Jerusalem, 1995).

2349: Topology Appl. 78 (1997), no. 1-2, 123--142.

2350: 

2351: \bibitem{Orevkov} S.~Yu.~Orevkov, Quasipositivity test via unitary

2352: representations of braid groups and its applications to real algebraic curves,

2353: J. Knot Theory Ramifications~10 (2001), no. 7, 1005--1023.

2354: 

2355: \bibitem{ParisRolfsen} L.~Paris, D.~Rolfsen, Geometric subgroups of

2356: surface braid groups, Ann. Inst. Fourier 49 (1999), 101-156

2357: 

2358: \bibitem{PeHa} R.~C.~Penner, J.~L.~Harer,  Combinatorics of train tracks,

2359: Annals of Mathematics Studies, 125. Princeton University Press,

2360: Princeton, NJ, 1992.

2361: 

2362: \bibitem{Sibert} H.~Sibert, Extraction of roots in Garside groups,

2363: Comm. Algebra 30 (2002), no. 6, 2915--2927.

2364: 

2365: \bibitem{Sty}  V.~B.~Sty\v snev,  Izv. Akad. Nauk SSSR Ser. Mat. 42 (1978),

2366: no. 5, 1120--1131, 1183.

2367: 

2368: \bibitem{Thclass} W. P. Thurston,  On the geometry and dynamics of

2369: diffeomorphisms of surfaces, Bull. Amer. Math. Soc. (N.S.) 19 (1988),

2370: 417--431.

2371: 

2372: \bibitem{ThBnaut} W. P. Thurston, Braid Groups, Chapter 9 of ``Word

2373: processing in groups",

2374: D.~B.~A.~Epstein, J.~W.~Cannon, D.~F.~Holt, S.~V.~F.~Levy, M.~S.~Paterson and

2375: W.~P.~Thurston, Jones and Bartlett Publishers, Boston, MA, 1992.

2376: 

2377: \end{thebibliography}

2378: 

2379: \end{document}

2380: