1: 

2: % this version has Jones polynomial finiteness moved in

3: % and simplified

4: 

5: \documentstyle[txmac,a4,%leqno,%

6: amssymb,%

7: bibenc,%

8: case,%

9: twoside,%

10: nocaphead2,%

11: epsf,%

12: leqno,%

13: %varthm,%

14: %rotate,%

15: myrot,%

16: %multicol,%

17: mypic,%

18: times,mathptm%

19: ]{article}

20: 

21: \advance\oddsidemargin by -1.9cm

22: \advance\evensidemargin by -1.9cm

23: \advance\textwidth by 3.8cm

24: % \advance\oddsidemargin by -0.3cm

25: % \advance\evensidemargin by -0.3cm

26: % \advance\textwidth by 0.6cm

27: 

28: \def\mynewtheo#1#2{%

29: \newtheorem{@#1}{#2}[section]%

30: \newenvironment{#1}{\begin{@#1}\rm}{\end{@#1}}}

31: 

32: \mynewtheo{lemma}{Lemma}

33: \mynewtheo{exer}{Exercise}

34: \mynewtheo{theo}{Theorem}

35: \mynewtheo{rem}{Remark}

36: \mynewtheo{defi}{Definition}

37: \mynewtheo{conj}{Conjecture}

38: \mynewtheo{corr}{Corollary}

39: \mynewtheo{prop}{Proposition}

40: \mynewtheo{question}{Question}

41: \mynewtheo{exam}{Example}

42: 

43: \makeatletter

44: 

45: \newenvironment{theorem}{\begin{theo}}{\end{theo}}

46: \newenvironment{conjecture}{\begin{conj}}{\end{conj}}

47: \newenvironment{myitems}{%

48: \def\@listi{\leftmargin \leftmargini \parsep 4\p@ \@plus

49: 2\p@ \@minus \p@\topsep1\p@ \@plus 2\p@ \@minus 1\p@

50: \itemsep 4\p@ \@plus 2\p@ \@minus \p@} \begin{itemize}

51: \itemsep\z@\relax \partopsep\z@\relax}{\end{itemize}}

52:  

53: 

54: 

55: \parskip 5pt plus 3pt minus 2pt\relax

56: 

57: \parindent0pt\relax

58: 

59: \pagestyle{headings}

60: 

61: \begin{document}

62: 

63: \input{myeqn.tex}

64: % \input{young.tex}

65: \renewenvironment{eqn}{\begin{equation}}{\end{equation}\ignorespaces}

66: 

67: \def\rottab#1#2#3{

68: \expandafter\advance\csname c@#3\endcsname by -1\relax

69: \centerline{%

70: %\hfill

71: %\fboxsep0pt\relax

72: %\fbox{%

73: \rbox{\centerline{\vbox{\setbox1=\hbox{#1}%

74: \hbox to \wd1{\hfill\vbox{{%

75: %\tracingmacros1

76: \caption{#2}}}\hfill}%

77: \vskip6mm

78: \box1}}%

79: }

80: %}

81: }%

82: }

83: 

84: 

85: % \let\old@diag\diag \def\diag{\special{ps: pstack (====) ==}\old@diag}

86: 

87: \newcommand{\mybin}[2]{\text{$\Bigl(\begin{array}{@{}c@{}}#1\\#2%

88: \end{array}\Bigr)$}}

89: \newcommand{\mybinn}[2]{\text{$\biggl(\begin{array}{@{}c@{}}%

90: #1\\#2\end{array}\biggr)$}}

91: 

92: \def\overtwo#1{\mbox{\small$\mybin{#1}{2}$}}

93: \newcommand{\mybr}[2]{\text{$\Bigl\lfloor\mbox{%

94: \small$\displaystyle\frac{#1}{#2}$}\Bigr\rfloor$}}

95: \def\mybrtwo#1{\mbox{\mybr{#1}{2}}}

96: 

97: \def\myfrac#1#2{\raisebox{0.2em}{\small$#1$}\!/%

98: \!\raisebox{-0.2em}{\small$#2$}}

99: \def\ffrac#1#2{\mbox{\small$\ds\frac{#1}{#2}$}}

100: 

101: \def\noloop{{\diag{0.5cm}{0.5}{1}{\picline{0.25 0}{0.25 1}}}}

102: 

103: \def\vrt#1{{\picfillgraycol{0}\picfilledcircle{#1}{0.09}{}}}

104: 

105: \def\ReidI#1#2{

106:   \diag{0.5cm}{0.9}{1}{

107:     \pictranslate{0.4 0.5}{

108:       \picscale{#11 #21}{

109:         \picmultigraphics[S]{2}{1 -1}{

110: 	  \picmulticurve{-6 1 -1 0}{0.5 -0.5}{0.5 0}{0.1 0.3}{-0.2 0.3}

111: 	} %picmultigraphics

112: 	\piccirclearc{-0.2 0}{0.3}{90 270}

113:       }%picscale

114:     }%pictranslate

115:   }%diag

116: }

117: \def\Pos#1#2{{\diag{#1}{1}{1}{#2

118: \picmultiline{-5 1 -1 0}{0 1}{1 0}

119: \picmultiline{-5 1 -1 0}{0 0}{1 1}

120: }}}

121: \def\Neg#1#2{{\diag{#1}{1}{1}{#2

122: \picmultiline{-5 1 -1 0}{0 0}{1 1}

123: \picmultiline{-5 1 -1 0}{0 1}{1 0}

124: }}}

125: \def\Nul#1#2{{\diag{#1}{1}{1}{#2

126: \piccirclearc{0.5 1.4}{0.7}{-135 -45}

127: \piccirclearc{0.5 -0.4}{0.7}{45 135}

128: }}}

129: \def\Inf#1#2{{\diag{#1}{1}{1}{#2

130: \piccirclearc{0.5 1.4 x}{0.7}{135 -135}

131: \piccirclearc{0.5 -0.4 x}{0.7}{-45 45}

132: }}}

133: \def\pos{\Pos{0.5em}{\piclinewidth{10}}}

134: \def\neg{\Neg{0.5em}{\piclinewidth{10}}}

135: \def\nul{\Nul{0.5em}{\piclinewidth{10}}}

136: \def\inf{\Inf{0.5em}{\piclinewidth{10}}}

137: 

138: \def\noloop{{\diag{0.5cm}{0.5}{1}{\picline{0.25 0}{0.25 1}}}}

139: 

140: \def\ReidI#1#2{

141:   \diag{0.5cm}{0.9}{1}{

142:     \pictranslate{0.4 0.5}{

143:       \picscale{#11 #21}{

144:         \picmultigraphics[S]{2}{1 -1}{

145: 	  \picmulticurve{-6 1 -1 0}{0.5 -0.5}{0.5 0}{0.1 0.3}{-0.2 0.3}

146: 	} %picmultigraphics

147: 	\piccirclearc{-0.2 0}{0.3}{90 270}

148:       }%picscale

149:     }%pictranslate

150:   }%diag

151: }

152: 

153: \def\GD{\szCD{6mm}}

154: \def\szCD#1#2{{\let\@nomath\@gobble\small\diag{#1}{2.4}{2.4}{

155:   \picveclength{0.27}\picvecwidth{0.1}

156:   \pictranslate{1.2 1.2}{

157:     \piccircle{0 0}{1}{}

158:     #2

159: }}}}

160: \def\CD{\szCD{4mm}}

161: 

162: \def\point#1{{\picfillgraycol{0}\picfilledcircle{#1}{0.08}{}}}

163: \def\labpt#1#2#3{\pictranslate{#1}{\point{0 0}\picputtext{#2}{$#3$}}}

164: \def\vrt#1{{\picfillgraycol{0}\picfilledcircle{#1}{0.09}{}}}

165: 

166: \def\chrd#1#2{\picline{1 #1 polar}{1 #2 polar}}

167: \def\arrow#1#2{\picvecline{1 #1 polar}{1 #2 polar}}

168: 

169: \def\labch#1#2#3{\chrd{#1}{#2}\picputtext{1.3 #2 polar}{$#3$}}

170: \def\labar#1#2#3{\arrow{#1}{#2}\picputtext{1.3 #2 polar}{$#3$}}

171: \def\labbr#1#2#3{\arrow{#1}{#2}\picputtext{1.3 #1 polar}{$#3$}}

172: 

173: \def\@dcont{}

174: \def\svCD#1{\ea\glet\csname #1\endcsname\@dcont}

175: \def\rsCD#1{\ea\glet\ea\@dcont\csname #1\endcsname\ea\glet

176: \csname #1\endcsname\relax}

177: 

178: \def\addCD#1{\ea\gdef\ea\@dcont\ea{\@dcont #1}}

179: \def\drawCD#1{\szCD{#1}{\@dcont}}

180: 

181: 

182: \def\epsfs#1#2{{\catcode`\_=11\relax\ifautoepsf\unitxsize#1\relax\else

183: \epsfxsize#1\relax\fi\epsffile{#2.eps}}}

184: \def\epsfsv#1#2{{\vcbox{\epsfs{#1}{#2}}}}

185: \def\vcbox#1{\setbox\@tempboxa=\hbox{#1}\parbox{\wd\@tempboxa}{\box

186:   \@tempboxa}}

187: \newbox\@tempboxb

188: \def\vtbox#1{\setbox\@tempboxa=\hbox{\shortstack[t]{#1}}%

189:   \setbox\@tempboxb=\hbox{2rut}\@tempdima\ht\@tempboxa\relax

190:   \advance\@tempdima-\ht\@tempboxb\relax\raise-\@tempdima\box \@tempboxa

191:   %\dp\@tempboxa=\@tempdima

192:   %\box \@tempboxa

193:   }

194: \def\p{\epsfsv{2cm}}

195: 

196: \def\@test#1#2#3#4{%

197:   \let\@tempa\go@

198:   \@tempdima#1\relax\@tempdimb#3\@tempdima\relax\@tempdima#4\unitxsize\relax

199:   \ifdim \@tempdimb>\z@\relax

200:     \ifdim \@tempdimb<#2%

201:       \def\@tempa{\@test{#1}{#2}}%

202:     \fi

203:   \fi

204:   \@tempa

205: }

206: 

207: \def\go@#1\@end{}

208: \newdimen\unitxsize

209: \newif\ifautoepsf\autoepsftrue

210: 

211: \unitxsize4cm\relax

212: \def\epsfsize#1#2{\epsfxsize\relax\ifautoepsf

213:   {\@test{#1}{#2}{0.1 }{4   }

214: 		{0.2 }{3   }

215: 		{0.3 }{2   }

216: 		{0.4 }{1.7 }

217: 		{0.5 }{1.5 }

218: 		{0.6 }{1.4 }

219: 		{0.7 }{1.3 }

220: 		{0.8 }{1.2 }

221: 		{0.9 }{1.1 }

222: 		{1.1 }{1.  }

223: 		{1.2 }{0.9 }

224: 		{1.4 }{0.8 }

225: 		{1.6 }{0.75}

226: 		{2.  }{0.7 }

227: 		{2.25}{0.6 }

228: 		{3   }{0.55}

229: 		{5   }{0.5 }

230: 		{10  }{0.33}

231: 		{-1  }{0.25}\@end

232: 		\ea}\ea\epsfxsize\the\@tempdima\relax

233: 		\fi

234: 		}

235: 

236: \def\vrt#1{{\picfillgraycol{0}\picfilledcircle{#1}{0.09}{}}}

237: \def\fdot#1{{\picfillgraycol{0}\picfilledcircle{#1}{0.02}{}}}

238: \def\cycl#1#2#3#4{\vrt{#1}\vrt{#2}\vrt{#3}\vrt{#4}

239: \picline{#1}{#2}\picline{#2}{#3}\picline{#3}{#4}\picline{#4}{#1}}

240: \def\xcycl#1#2#3#4#5#6{\vrt{#1}\vrt{#2}\vrt{#3}\vrt{#4}\vrt{#5}

241: \vrt{#6}\picline{#1}{#2}\picline{#2}{#3}\picline{#3}{#4}

242: \picline{#4}{#5}\picline{#5}{#6}\picline{#6}{#1}}

243: 

244: %usage ODD # of  coords, then {}

245: \def\@curvepath#1#2#3{%

246:   \@ifempty{#2}{\piccurveto{#1 }{@stc}{@std}#3}%

247:     {\piccurveto{#1 }{#2 }{#2  #3  0.5 conv}

248:     \@curvepath{#3}}%

249: }

250: \def\curvepath#1#2#3{%

251:   \piccurve{#1 }{#2 }{#2 }{#2  #3  0.5 conv}%

252:   \picPSgraphics{/@stc [ #1  #2  -1 conv ] $ D /@std [ #1  ] $ D }%

253:   \@curvepath{#3}%

254: }

255: 

256: %usage EVEN # of  coords, then {}

257: \def\@opencurvepath#1#2#3{%

258:   \@ifempty{#3}{\piccurveto{#1 }{#1 }{#2 }}%

259:     {\piccurveto{#1 }{#2 }{#2  #3  0.5 conv}\@opencurvepath{#3}}%

260: }

261: \def\opencurvepath#1#2#3{%

262:   \piccurve{#1 }{#2 }{#2 }{#2  #3  0.5 conv}%

263:   \@opencurvepath{#3}%

264: }

265: 

266: 

267: \author{A. Stoimenow\footnotemark[1]\\[2mm]

268: \small Research Institute for Mathematical Sciences, \\

269: \small Kyoto University, Kyoto 606-8502, Japan\\

270: \small e-mail: {\tt stoimeno@kurims.kyoto-u.ac.jp}\\

271: \small WWW: {\hbox{\web|http://www.kurims.kyoto-u.ac.jp/~stoimeno/|}}

272: }

273: 

274: {\def\thefootnote{\fnsymbol{footnote}}

275: \footnotetext[1]{Supported by 21st Century COE Program.}

276: }

277: 

278: \title{\large\bf \uppercase{Properties of closed 3-braids}\\

279: [1mm]\uppercase{and other link braid representations}\\

280: [4mm]

281: \phantom{\small\it This is a preprint. I would be grateful

282: for any comments and corrections %but I prefer you not circulate it

283: !}}

284: 

285: \date{\large Current version: \curv\ \ \ First version:

286: \makedate{22}{9}{2004}}

287: 

288: \maketitle

289: 

290: \makeatletter

291: 

292: \let\vn\varnothing

293: \let\point\pt

294: \let\ay\asymp

295: \let\pa\partial

296: \let\ap\alpha

297: \let\bt\beta

298: \let\be\beta

299: \let\Dl\Delta

300: \let\Gm\Gamma

301: \let\gm\gamma

302: \let\de\delta

303: \let\dl\delta

304: \let\eps\epsilon

305: \let\lm\lambda

306: \let\Lm\Lambda

307: \let\sg\sigma

308: \let\vp\varphi

309: \let\zt\zeta

310: \let\om\omega

311: \let\diagram\diag

312: \let\nb\nabla

313: \let\prt\partial

314: \let\wh\widehat

315: \let\wt\widetilde

316: \def\tW{\wt W}

317: \let\sm\setminus

318: \let\tl\tilde

319: % \def\ncap{\not\mathrel{\cap}}

320: % \def\sgn{\text{\rm sgn}\,}

321: \def\inx{\mathop {\operator@font ind}}

322: \def\spn{\mathop {\operator@font span}\nolimits}

323: \def\dig{\mathop {\operator@font diag}}

324: \def\Mc{\max\cf}

325: \def\Md{\max\deg}

326: \def\md{\min\deg}

327: \def\mc{\min\cf}

328: \def\vol{\text{\rm vol}\,}

329: \def\Ra{\Rightarrow}

330: \def\Lra{\Longrightarrow}

331: \def\lra{\longrightarrow}

332: \def\so{\Rightarrow}

333: \def\So{\Longrightarrow}

334: \def\nin{\not\in}

335: \let\ds\displaystyle

336: \let\llra\longleftrightarrow

337: \let\reference\ref

338: % \def\ul#1{\fbox{$#1$}}

339: \let\ol\overline

340: \let\ul\underline

341: \let\u\underline

342: \let\h\hat

343: % \def\ul#1{\diag{1em}{1}{1}{\picbox{0.5 0.5}{1.0 1.0}{$#1$}}}

344: \let\es\enspace

345: \def\lfra{\leftrightarrow}

346: \def\cf{\text{\rm cf}\,}

347: 

348: \def\TM{$^\text{\raisebox{-0.2em}{${}^\text{TM}$}}$}

349: % \def\lb{\linebreak[0]}

350: % \def\lz{\lb\verb}

351: \def\ssim{\stackrel{\ds \sim}{\vbox{\vskip-0.2em\hbox{$\scriptstyle

352: *$}}}}

353: 

354: 

355: \long\def\@makecaption#1#2{%

356:    % \tm

357:    \vskip \abovecaptionskip 

358:    {\let\label\@gobble

359:    \let\ignorespaces\@empty

360:    \xdef\@tempt{#2}%

361:    }%

362:    \ea\@ifempty\ea{\@tempt}{%

363:    \sbox\@tempboxa{%

364:       \fignr#1#2}%

365:       }{%

366:    \sbox\@tempboxa{%

367:       {\fignr#1:}\capt\ #2}%

368:       }%

369:    \ifdim \wd\@tempboxa >\captionwidth {%

370:       %\centerline{\parbox{\captionwidth}{\unhbox \@tempboxa}}%

371:       \rightskip=\@captionmargin\leftskip=\@captionmargin

372:       \unhbox\@tempboxa\par

373:      }%

374:    \else

375:       \centerline{\box \@tempboxa}%

376:       % \hbox to\captionwidth{\hfil\box\@tempboxa\hfil}%

377:    \fi

378:    \vskip \belowcaptionskip

379:    }%

380: %

381: \def\fignr{\small\sffamily\bfseries}%

382: \def\capt{\small\sffamily}%

383: 

384: % \long\def\@makecaption#1#2{%

385: %    % \tm

386: %    \vskip 10pt

387: %    {\let\label\@gobble

388: %    \let\ignorespaces\@empty

389: %    \xdef\@tempt{#2}%

390: %    %\typeout{`#2'}%

391: %    }%

392: %    \ea\@ifempty\ea{\@tempt}{%

393: %    \setbox\@tempboxa\hbox{%

394: %       \fignr#1#2}%

395: %       }{%

396: %    \setbox\@tempboxa\hbox{%

397: %       {\fignr#1:}\capt\ #2}%

398: %       }%

399: %    \ifdim \wd\@tempboxa >\captionwidth {%

400: %       \rightskip=\@captionmargin\leftskip=\@captionmargin

401: %       \unhbox\@tempboxa\par}%

402: %    \else

403: %       \hbox to\captionwidth{\hfil\box\@tempboxa\hfil}%

404: %    \fi}%

405: % %

406: % \def\fignr{\small\sffamily\bfseries}%

407: % \def\capt{\small\sffamily}%

408: 

409: 

410: \newdimen\@captionmargin\@captionmargin2cm\relax

411: \newdimen\captionwidth\captionwidth0.8\hsize\relax

412: 

413: 

414: \def\eqref#1{(\protect\ref{#1})}

415: 

416: \def\proof{\@ifnextchar[{\@proof}{\@proof[\unskip]}}

417: \def\@proof[#1]{\noindent{\bf Proof #1.}\enspace}

418: 

419: \def\hint{\noindent Hint: }

420: \def\problem{\noindent{\bf Problem.} }

421: % \def\question{\noindent{\bf Question.} }

422: 

423: \def\@mt#1{\ifmmode#1\else$#1$\fi}

424: \def\qed{\hfill\@mt{\Box}}

425: \def\qqed{\hfill\@mt{\Box\enspace\Box}}

426: 

427: % \let\Bbb\bf

428: 

429: \def\cU{{\cal U}}

430: \def\cC{{\cal C}}

431: \def\cP{{\cal P}}

432: \def\fg{{\frak g}}

433: \def\tr{\text{tr\,}}

434: \def\cZ{{\cal Z}}

435: \def\cD{{\cal D}}

436: \def\bR{{\Bbb R}}

437: \def\cE{{\cal E}}

438: \def\bZ{{\Bbb Z}}

439: \def\bN{{\Bbb N}}

440: 

441: 

442: \def\bysame{\same[\kern2cm]\,}

443: 

444: \def\br#1{\left\lfloor#1\right\rfloor}

445: \def\BR#1{\left\lceil#1\right\rceil}

446: 

447: \def\abstractname{}

448: 

449: 

450: \@addtoreset {footnote}{page}

451: 

452: \renewcommand{\section}{%

453:    \@startsection

454:          {section}{1}{\z@}{-2.5ex \@plus -1ex \@minus -.2ex}%

455:                {3ex \@plus.2ex}{\Large\bf}%

456: }

457: 

458: \renewcommand{\subsubsection}{%

459:    \@startsection

460:          {subsubsection}{1}{\z@}{-1.5ex \@plus -1ex \@minus -.2ex}%

461:                {1ex \@plus.2ex}{\large\bf}%

462: }

463: \renewcommand{\@seccntformat}[1]{\csname the#1\endcsname .

464: \quad}

465: 

466: \def\bC{{\Bbb C}}

467: \def\bP{{\Bbb P}}

468: 

469: 

470: \makeatletter

471: 

472: 

473: \let\old@tl\~\def\~{\raisebox{-0.8ex}{\tt\old@tl{}}}

474: \let\lra\longrightarrow

475: \let\sm\setminus

476: \let\eps\varepsilon

477: \let\ex\exists

478: \let\fa\forall

479: \let\ps\supset

480: 

481: \def\rs#1{\raisebox{-0.4em}{$\big|_{#1}$}}

482: 

483: {\let\@noitemerr\relax

484: \vskip-2.7em\kern0pt\begin{abstract}

485: \noindent{\bf Abstract.}\enspace

486: We show that 3-braid links with given (non-zero) Alexander

487: or Jones polynomial are finitely many, and can be effectively

488: determined. We classify among closed 3-braids strongly

489: quasi-positive and fibered ones, and show that 3-braid links have

490: a unique incompressible Seifert surface. We also classify the

491: positive braid words with Morton-Williams-Franks

492: bound 3 and show that closed positive braids of braid index 3 are

493: closed positive 3-braids. For closed braids on more

494: strings, we study the alternating links occurring. In particular

495: we classify those of braid index 4, and show that their Morton-%

496: Williams-Franks inequality is exact. Finally, we use the Burau

497: representation to obtain new braid index criteria, including an

498: efficient 4-braid test.\\[2mm]

499: %

500: {\it Keywords:} link polynomial, positive braid, strongly

501: quasi-positive link, 3-braid link, alternating link, Seifert surface,

502: fiber, incompressible surface, braid index, Burau representation

503: \\[2mm]

504: {\it AMS subject classification:} 57M25 (primary), 20F36, 32S55

505: (secondary)

506: \end{abstract}

507: }\vspace{7mm}

508: 

509: {\parskip0.2mm\tableofcontents}

510: \vspace{7mm}

511: 

512: \section{Introduction}

513: 

514: Originating from the pioneering work of Alexander

515: \cite{Alexander2} and Artin \cite{Artin,Artin2}, braid

516: theory has become intrinsically interwoven with knot theory,

517: and over the years, braid representations of different types

518: have been studied, many of them with motivation coming

519: from fields outside of knot theory, for example dynamical

520: systems \cite{Williams} or 4-dimensional QFTs \cite{Kreimer}.

521: 

522: The systematic study of closed 3-braids was begun by Murasugi

523: \cite{Murasugi2}. Later 3-braid links have been classified

524: in \cite{BirMen}, in the sense that 3-braid representations

525: of the same link are exactly described. This is, up to

526: a few exceptions, mainly just conjugacy. The conjugacy

527: problem of 3-braids has a series of solutions, starting with

528: Schreier's algorithm \cite{Sc}, going over Garside

529: \cite{Garside} (for arbitrary braid groups), Xu \cite{Xu}

530: (and later more generally Birman-Ko-Lee \cite{BKL}), until,

531: for example, a recent algorithm of linear complexity due to

532: Fiedler and Kurlin \cite{FK}. So Birman-Menasco's work allows

533: to decide if two 3-braids have the same closure link.

534: 

535: However, many properties of links (except achirality and

536: invertibility) are not evident from (3-strand) braid

537: representations, and thus to classify 3-braid links with

538: special properties remains a non-trivial task. A first result

539: was given in \cite{Murasugi} for rational links, and then

540: improved in \cite{3br}, where we completed this project for

541: alternating links.

542: 

543: {}From the opposite point of view, a natural question concerning

544: braid representations of links is: 

545: 

546: \begin{question}\label{q1}

547: If a braid representation of a particular type exists, does

548: also one exist with the minimal number of strands (among all

549: braid representations of the link)?

550: \end{question}

551: 

552: The minimal number of

553: strands for a braid representation of a link $L$ is called the

554: \em{braid index} of the link, and will be denoted by $b(L)$.

555: 

556: In \cite{Bennequin}, Bennequin studied in relation to contact

557: structures braid representations by bands, independently considered

558: by Rudolph \cite{Rudolph} in the context of algebraic curves, and

559: more recently in \cite{BKL} from a group theoretic point of view.

560: A band representation naturally spans a Seifert surface of the

561: link. See figure \reference{figbd}.

562:  

563: \begin{figure}[h]

564: \[

565: % \begin{array}{c}

566: \epsfs{5cm}{fig1}

567: % \end{array}

568: \]\\[-14mm]

569: \caption{A braided Seifert surface obtained from a band representation

570: \label{figbd}}

571: \end{figure}

572:  

573: % In figure \reference{fig2}, for the move (a) $\to$ (b), we slide

574: % $B$ and $C$, respectively, along $D$ and $A$, then delete $D$

575: % (by deplumbing a Hopf band), and then slide $C$ back along

576: % $A$. For the move (b) $\to$ (c), we slide $C$ along $B$,

577: % then slide $B$ along $C$. The $k$ bands $A$ can be subsequently

578: % removed by Murasugi desumming a $(2,k)$-torus link fiber surface.

579: % Thus the surface $F$, spanned by $\hat\bt$, turns after

580: % (de)summing Hopf bands into a surface $F'$ consisting of $k$

581: % copies of $F'_0$ and one negative band $N$. See figure \ref{fig3} (a).

582: %  

583: % \begin{figure}[h]

584: % \[

585: % \begin{array}{cc}

586: %  (a) & (b) \\

587: % \end{array}

588: % \]

589: % \caption{\label{fig3}}

590: % \end{figure}

591: %  

592: % In figure \reference{fig3}, we perform the move (a) $\to$ (b), sliding

593: % $B$ along $N$. The last surface $F'$, in figure \reference{fig3} (b),

594: % is Murasugi sum of a fibre surface spanning the $(2,2,-2)$-pretzel

595: % link and $\tl F$, where $\tl F$ consists 

596: % of $k - 1$ copies of $F'_0$ and the band $N$. By induction on

597: % $k$, we see that $F'$ is a fibre surface, and hence $\hat\bt$

598: % is a fibred link. \qed

599: %  

600: % \end{appendix}

601: Bennequin proved in \cite{Bennequin} that a minimal genus band

602: representation always exists on 3-braids. Such surface was called

603: by Birman-Menasco \cite{BirMen2} a Bennequin surface. Later, by

604: examples found by Ko and Lee, and Hirasawa and myself, the

605: existence of Bennequin surfaces on the minimal number of strands

606: was found not to extend to 4-braids. (We will see in this paper,

607: though, among many other things, that it does extend for

608: alternating links.) Since by Rudolph's work \cite{Rudolph},

609: a Bennequin surface exists for any link, on a possibly large

610: number of strands, the answer to the question \reference{q1}

611: is negative for minimal genus band representations. 

612: 

613: A special case of band representations are the positive ones,

614: called strongly quasi-positive in \cite{Rudolph}. For such

615: representations the above examples do not apply. The existence

616: of minimal such representations is a question of Rudolph, whose

617: answer is not known. Our first result is a positive answer to

618: Rudolph's question for 3-braids. 

619: 

620: \begin{theorem}\label{thqp}

621: If $L$ is a strongly quasi-positive link, and has braid index 3,

622: then $L$ is has a strongly quasi-positive 3-braid representation.

623: \end{theorem}

624: 

625: This result is an application of the work in \cite{3br} on

626: Xu's normal form \cite{Xu}. It legitimizes the term ``strongly

627: quasi-positive 3-braid link'', which otherwise would be

628: linguistically sloppy, for it may mean a 3-braid link that is strongly

629: quasi-positive, or the link (which is the closure) of a strongly

630: quasi-positive 3-braid.

631: 

632: 

633: As an (\em{a-priori}-so-seeming) improvement of Bennequin's

634: theorem, Birman-Menasco proved \cite{BirMen2} that

635: for 3-braid links not only some but in fact \em{any} minimal genus

636: surface is a Bennequin surface. Xu used this result and the

637: classification in \cite{BirMen} to conclude that most 3-braid links

638: have a unique minimal genus surface. Unfortunately, she failed to

639: deal (in the oriented sense) with some exceptions, which we complete

640: in section \reference{SA} (see theorem \reference{wu}).

641: 

642: The following work involves a further study of Xu's normal form.

643: % A proof of this result is given by the authors in an appendix.

644: %

645: A natural relation of this form to the skein polynomial was

646: exhibited in \cite{3br}. Here we extend this study to the Alexander

647: polynomial $\Dl$. As a result, we can describe 3-braid links with

648: given $\Dl$, and in particular show no non-trivial 3-braid

649: knot has trivial polynomial. In \cite{Birman}, Birman constructed

650: some pairs of different 3-braid links with the same polynomials

651: and proposed the problem to understand the values of the various

652: link polynomials on 3-braid links. Our work solves (to a large

653: extent at least) Birman's problem for the Alexander polynomial.

654: We also have a solution for the Jones and Brandt-Lickorish-%

655: Millett-Ho $Q$ polynomial. 

656: 

657: Parallel to the Alexander polynomial description, and using

658: a recent result of Hirasawa-Murasugi, we classify among 3-braid

659: links the fibered ones. (This result was proved independently

660: by Yi Ni in \cite{Ni}, apparently unaware of this work, which

661: I did to make widely public immediately.) We will see then

662: that a 3-braid knot is fibered if (and only if) its Alexander

663: polynomial is monic, but that this is \em{not} true for 2- or

664: 3- component links.

665: 

666: The fact that most 3-braid links were found fibered,

667: and this could be proved independently from Bennequin and

668: Birman-Menasco, opened the hope for a more natural approach to

669: some of their results. Finally we were indeed able to complete this

670: project, with the assistance of M. Hirasawa and M. Ishiwata, obtaining

671: 

672: \begin{theorem}\label{thuq}

673: Any 3-braid link has a unique (non-closed) incompressible surface.

674: \end{theorem}

675: 

676: This result subsumes all previous uniqueness theorems. Besides,

677: its proof is entirely different, in that it fully avoids the contact

678: geometry approach of Bennequin and the considerable complexity

679: in Birman-Menasco's subsequent braid foliation work. Instead,

680: we use the sutured manifold theory of Gabai \cite{Gabai,Gabai2}

681: (to deal with the fibered links) and Kobayashi-Kakimizu \cite{%

682: Kobayashi,Kakimizu} (for the remaining cases), as well as some

683: of our preceding work on the Alexander polynomial (to rule out

684: disconnected Seifert surfaces). Since Kobayashi-Kakimizu use

685: a slightly stronger notion of uniqueness (see the beginning of

686: \S\reference{Uq}), theorem \reference{thuq} is an improvement even

687: for minimal genus surfaces. Moreover, its proof underscores the

688: geometric meaning of Xu's form, that remained unclear in \cite{BirMen}.

689: While we still cannot provide an ``easy'' proof of Birman-Menasco's

690: classification theorem (see remark on p.~34 of \cite{BirMen}), our

691: work will likely lead to simplifications in their very lengthy

692: treatment, which makes heavy use of incompressible surfaces.

693: 

694: The study and applications of Xu's form occupy the first main

695: part of the paper. Then some related results are included,

696: whose treatment requires different methods. These

697: results are dealt with in the later sections of the paper.

698: 

699: As a first such topic, more substantial effort will be made to

700: prove an analogous result to theorem \reference{thqp} for

701: positive (in the usual Artin generators) braid representations.

702: 

703: \begin{theorem}\label{thg}

704: If $L$ is a positive braid link, and has braid index 3, then 

705: $L$ is a positive 3-braid link.

706: \end{theorem}

707: 

708: So our work here can be understood to answer question

709: \reference{q1} for braid index 3 and strongly quasi-positive

710: or positive braid links. (The state of quasi-positive links,

711: and of quasi-positive braid representations with regard to 

712: question \reference{q1}, in contrast remains open. There is,

713: however, an algorithm, found by orevkov \cite{Orevkov}

714: to decide if a 3-braid is quasi-positive.)

715: 

716: The proof of theorem \reference{thg} requires a study of positive

717: braids where the braid index bound in the Morton-Williams-Franks

718: inequality \cite{Morton,WilFr} is 3. This study builds on and extends

719: (but also considerably simplifies) the work of Nakamura

720: \cite{Nakamura} for Morton-Williams-Franks bound 2. Then some

721: detailed case distinction and calculation are necessary.

722: In \cite{posex_bcr} it was shown, again, that theorem

723: \reference{thg} is not true for 4-braids: there are two 16 crossing

724: knots that have positive 16 crossing 5-braid representations,

725: but braid index 4 (and consequently only non-positive 4-braid

726: representations). It is in fact for these examples that I was

727: led to investigate about the theorem.

728: 

729: In section \ref{Spos},

730: %we come back to theorem \reference{thg}

731: % and show how to obtain its alternative proof

732: %. As a partial step, which is equally

733: % noteworthy,

734: using the arguments in the proof of theorem \reference{thg}

735: and a criterion of Yokota \cite{Yokota}, we consider 3-braid

736: representations of links that are (diagrammatically) positive.

737: We obtain strong restrictions on such representations,

738: and in particular determine which of the links are not fibered.

739: (See theorem \ref{thposq}.)

740: 

741: % We will show two completely different approaches to

742: % theorem \reference{thg}. The first uses the

743: % Morton-Williams-Franks inequality and tries to identify

744: % the positive braids with Morton-Williams-Franks bound 3,

745: % extending the work of Nakamura \cite{Nakamura} for

746: % bound 2. The proof is slightly technical, and requires

747: % the verification of certain special cases.

748: % Another proof, following a longer, but much

749: % more insightful road, is then given by continuing

750: % the arguments that establish theorem \reference{thqp}.

751: 

752: Another problem proposed by Birman, in \cite{MortonPb},

753: was to understand the relation of alternation of links

754: (as a pre-eminent diagram defined property) and braid

755: representations. Early substantial results on the braid

756: index of alternating links were due to Murasugi \cite{Murasugi},

757: who determined the braid index for rational and fibered

758: alternating links. In \cite{3br} we used Xu's form to classify

759: the braid index 3 alternating links. Here, in \S\reference{Sfg},

760: we will easily recover this result and push it forward to braid

761: index 4. For this we use an argument based on the Jones polynomial,

762: and connect the celebrated Kauffman-Murasugi-Thistlethwaite work

763: \cite{Kauffman2,Murasugi3,Thistle2} to braid representations.

764: The existence of the Bennequin surface of alternating links on a

765: 4 string braid (corollary \reference{cfT}) is among several easy

766: consequences we obtain.

767: 

768: The last section \S\reference{SBurau} discusses some applications

769: of the Squier unitarization \cite{Squier} of the Burau

770: representation, that concern also braids on more strings. A

771: first series of conditions are estimates on the norm of special

772: values of the Alexander and Jones polynomial. They relate to and

773: partly extend estimates of Jones in \cite{Jones}. In particular, we

774: will see that the Alexander polynomial provides conditions for every

775: given braid index. Some properties of the skein polynomial, and a

776: relation to Mahler measure are also discussed.

777: 

778: In the case of a 4-braid, one can go further and (almost)

779: identify the Eigenvalues of the Burau matrix

780: from the Alexander and Jones polynomial (of its closure).

781: Using this, a criterion for braid index 4 is derived.

782: We show examples exhibiting the efficiency of this test, including

783: such where not only the Morton-Williams-Franks inequality itself,

784: but also its 2-cable version fails (and so our test seems the only

785: practicable option).

786: 

787: % and is included here to complete the picture.

788: 

789: In the appendix we collect some work of Hirasawa, Ishiwata and

790: Murasugi, which completes the proof of several of our results.

791: 

792: \section{Preliminaries, basic definitions and conventions}

793: % \noindent{\bf Convention.} 

794: 

795: Basic concepts that appear throughout the paper are summarized.

796: `W.l.o.g.' will abbreviate `without loss of generality';

797: `r.h.s.' will stand for `right hand-side'.

798: 

799: \subsection{Links and link diagrams\label{Sdg}}

800: 

801: Links are represented by diagrams; we assume diagrams are

802: oriented (though sometimes orientation is not relevant).

803: 

804: A crossing $p$ in a link diagram $D$ is called \em{reducible} (or

805: nugatory) if it looks like on the left of figure \reference{figred}.

806: $D$ is called reducible if it has a reducible crossing, else it is

807: called \em{reduced}. The reducing of the reducible crossing $p$

808: is the move depicted on figure \reference{figred}. Each diagram $D$

809: can be (made) reduced by a finite number of these moves.

810: 

811: \begin{figure}[htb]

812: \begin{eqn}\label{eqred}

813: \diag{6mm}{4}{2}{

814:   \picrotate{-90}{\rbraid{-1 2}{1 1.4}}

815:   % \piccirclearc{2 0.5}{1.3}{45 135}

816:   % \piccirclearc{2 1.5}{1.3}{-135 -45}

817:   \picputtext[u]{2 0.7}{$p$}

818:   \picscale{1 -1}{

819:     \picfilledcircle{0.7 -1}{0.8}{$P$}

820:   }

821:   \picfilledcircle{3.3 1}{0.8}{$Q$}

822: } \qquad\lra\qquad

823: \diag{6mm}{4}{2}{

824:   \piccirclearc{2 0.5}{1.3}{45 135}

825:   \piccirclearc{2 1.5}{1.3}{-135 -45}

826:   \picfilledcircle{1 1}{0.8}{$P$}

827:   \picfilledcircle{3 1}{0.8}{$Q$}

828: } 

829: \end{eqn}

830: \caption{\label{figred}}

831: \end{figure}

832: 

833: We assume in the following all diagrams reduced, unless otherwise

834: stated.

835: 

836: The diagram on the right of figure \reference{figtan}

837: is called \em{connected sum} $A\# B$ of the diagrams $A$ and $B$.

838: If a diagram $D$ can be represented as the connected sum of 

839: diagrams $A$ and $B$, such that both $A$ and $B$ have at least one

840: crossing, then $D$ is called \em{disconnected} (or composite), else

841: it is called \em{connected} (or prime). $K$ is \em{prime} if whenever

842: $D=A\# B$ is a composite diagram of $K$, one of $A$ and $B$

843: represent an unknotted arc (but not both; the unknot is

844: not considered to be prime per convention).

845: 

846: By $c(D)$ we denote the

847: \em{number of crossings} of $D$, $n(D)$ the number of components

848: of $D$ (or $K$, $1$ if $K$ is a knot), and $s(D)$ the \em{number

849: of Seifert circles} of $D$. The \em{crossing number} $c(K)$ of a

850: knot or link $K$ is the minimal crossing number of all diagrams

851: $D$ of $K$.

852: $!D$ is the \em{mirror image} of $D$, and $!K$ is the mirror image

853: of $K$. Clearly $g(!D)=g(D)$ and $g(!K)=g(K)$.

854: 

855: \begin{figure}[htb]

856: \[

857: \diag{6mm}{3}{2}{

858:   \piccirclearc{1.8 1}{0.5}{-120 120}

859:   \picfilledcircle{1 1}{0.8}{$A$}

860: }\, \#\,

861: \diag{6mm}{3}{2}{

862:   \piccirclearc{1.2 1}{0.5}{60 300}

863:   \picfilledcircle{2 1}{0.8}{$B$}

864: }\quad =\quad

865: \diag{6mm}{4}{2}{

866:   \piccirclearc{2 0.5}{1.3}{45 135}

867:   \piccirclearc{2 1.5}{1.3}{-135 -45}

868:   \picfilledcircle{1 1}{0.8}{$A$}

869:   \picfilledcircle{3 1}{0.8}{$B$}

870: } 

871: \]

872: \caption{\label{figtan}}

873: \end{figure}

874: 

875: The \em{Seifert graph} $\Gm(D)$ of a diagram $D$ is defined to

876: be the graph whose vertices are the Seifert circles of $D$ and

877: whose edges are the crossings. The \em{reduced Seifert graph}

878: $\Gm'(D)$ is defined by removing multiple copies of an edge

879: between two vertices in $\Gm(D)$, so that a simple edge

880: remains. (See \cite{MurPrz,MurPrz2} for example.)

881: 

882: The \em{(Seifert) genus} $g(K)$ resp.\ \em{Euler characteristic}

883: $\chi(K)$ of a knot or link $K$ is said to be the minimal genus

884: resp.\ maximal Euler characteristic of Seifert surface of $K$.

885: For a diagram $D$ of $K$, $g(D)$ is defined to be the genus

886: of the Seifert surface obtained by Seifert's algorithm on $D$,

887: and $\chi(D)$ its Euler characteristic. We have $\chi(D)=s(D)-c(D)$

888: and $2g(D)=2-n(D)-\chi(D)$.

889: 

890: The numbering of knots we use is as in the tables of \cite

891: [appendix]{Rolfsen} for prime knots of crossing number $\le 10$,

892: and as in \cite{KnotScape} for those of crossing number $11$

893: to $16$. KnotScape's numbering is reorganized so that for

894: given crossing number non-alternating knots are appended after

895: alternating ones, instead of using `a' and `n' superscripts.

896: 

897: \subsection{Polynomial link invariants\label{brD}}

898: 

899: Let $X\in\bZ[t,t^{-1}]$. The \em{minimal} or \em{maximal degree} 

900: $\md X$ or $\Md X$ is the minimal resp.\ maximal exponent of $t$ 

901: with non-zero coefficient in $X$. Let $\spn_tX=\Md_tX-\md_tX$.

902: The coefficient in degree $d$ of $t$ in $X$ is denoted $[X]_{t^d}$

903: or $[X]_{d}$. The \em{leading coefficient} $\Mc\,X$ of $X$ is its

904: coefficient in degree $\Md X$. If $X\in\bZ[x_1^{\pm 1},x_2^{\pm 1}]$,

905: then $\Md_{x_1}X$ denotes the maximal degree in $x_1$. Minimal

906: degree and coefficients are defined similarly, and of course

907: $[X]_{x_1^k}$ is regarded as a polynomial in $x_2^{\pm 1}$.

908: 

909: Let $P(v,z)$ be the \em{skein polynomial} \cite{HOMFLY,LickMil}.

910: It is a Laurent polynomial in two variables of oriented knots and

911: links. We use here the convention of \cite{Morton}, i.e. with the

912: polynomial taking the value $1$ on the unknot, having the variables

913: $v$ and $z$ and satisfying the skein relation

914: \begin{eqn}\label{1}

915: v^{-1}\,P\bigl(

916: \diag{5mm}{1}{1}{

917: \picmultivecline{-5 1 -1.0 0}{1 0}{0 1}

918: \picmultivecline{-5 1 -1.0 0}{0 0}{1 1}

919: }

920: \bigr)\,-\,

921: v \,P\bigl(

922: \diag{5mm}{1}{1}{

923: \picmultivecline{-5 1 -1 0}{0 0}{1 1}

924: \picmultivecline{-5 1 -1 0}{1 0}{0 1}

925: }

926: \bigr)\,=\,

927: z\,P\bigl(

928: \diag{5mm}{1}{1}{

929: \piccirclevecarc{1.35 0.5}{0.7}{-230 -130}

930: \piccirclevecarc{-0.35 0.5}{0.7}{310 50}

931: }

932: \bigr)\,.

933: \end{eqn}

934: 

935: We will denote in each triple as in \eqref{1} 

936: the diagrams (from left to right) by $D_+$, $D_-$ and $D_0$.

937: For a diagram $D$ of a link $L$, we will use all of the notations

938: $P(D)=P_D=P_D(l,m)=P(L)$ etc.\ for its skein polynomial, with

939: the self-suggestive meaning of indices and arguments. So we

940: can rewrite \eqref{1} as

941: \begin{eqn}\label{srel}

942: v^{-1}P_+(v,z)-vP_-(v,z) = zP_0(v,z)\,.

943: \end{eqn}

944: 

945: The \em{writhe} is a number ($\pm1$), assigned to any crossing in a

946: link diagram. A crossing as on the left in \eqref{1} has writhe 1 and

947: is called \em{positive}. A crossing as in the middle of \eqref{1} has

948: writhe $-1$ and is called \em{negative}. The writhe of a link diagram

949: is the sum of writhes of all its crossings.

950: 

951: Let $c_{\pm}(D)$ be the number of positive, respectively negative

952: crossings of a diagram $D$, so that $c(D)=c_+(D)+c_-(D)$ and

953: $w(D)=c_+(D)-c_-(D)$. 

954: 

955: The \em{Jones polynomial} \cite{Jones2} $V$, and (one variable)

956: \em{Alexander polynomial} \cite{Alexander} $\Dl$ are obtained

957: from $P$ by variable substitutions

958: \begin{eqn}\label{DP}

959: \Dl(t)=P(1,t^{1/2}-t^{-1/2})\,,

960: \end{eqn}

961: and

962: \begin{eqn}\label{VPsub}

963: V(t)=P(t,t^{1/2}-t^{-1/2})\,.

964: \end{eqn}

965: 

966: Hence these polynomials also satisfy corresponding skein relations.

967: (In algebraic topology, the Alexander polynomial is usually

968: defined only up to units in $\bZ[t,t^{-1}]$; the present

969: normalization is so that $\Dl(t)=\Dl(1/t)$ and $\Dl(1)=1$.)

970: 

971: In very contrast to its relatives, the \em{range} of the 

972: Alexander polynomial (i.e., set of values it takes) is known.

973: Let us call a polynomial $\Dl\in\bZ[t^{1/2},t^{-1/2}]$

974: \em{admissible} if it satisfies for some natural number

975: $n\ge 1$ the three properties\\

976: \mbox{}\ \ (i)\quad $t^{(n-1)/2}\Dl\in \bZ[t^{\pm 1}]$,\\

977: \mbox{}\ (ii)\quad  $\Dl(t)=(-1)^{n-1}\Dl(1/t)$ and \\

978: \mbox{}(iii)\quad   $(t^{1/2}-t^{-1/2})^{n-1}\mid

979:                     \Dl$ for $n>1$, or $\Dl(1)=1$ for $n=1$. \\

980: It is well-known that these are exactly the polynomials that occur

981: as (1-variable) Alexander polynomials of some $n$-component link. 

982: 

983: The \em{Kauffman polynomial} \cite{Kauffman} $F$ is usually defined

984: via a regular isotopy invariant $\Lm(a,z)$ of unoriented links.

985: We use here a slightly different convention for the variables

986: in $F$, differing from \cite{Kauffman,Thistle2} by the interchange

987: of $a$ and $a^{-1}$. Thus in particular we have for a link diagram $D$

988: the relation $F(D)(a,z)=a^{w(D)}\Lm(D)(a,z)$, where $\Lm(D)$ is the

989: writhe-unnormalized

990: version of the polynomial, given in our convention by the properties

991: \begin{eqnarray}

992: & \Lambda\bigl(\Pos{0.5cm}{}\bigr)\ +\ \Lambda\bigl(\Neg{0.5cm}{}\bigr)\ =\ z\ \bigl(\ 

993: \Lambda\bigl(\Nul{0.5cm}{}\bigr)\ +\ \Lambda\bigl(\Inf{0.5cm}{}\bigr)\ \bigr)\,,\label{wseven}\\[2mm]

994: & \label{wseven.5} \Lambda\bigl(\ \ReidI{-}{-}\bigr) = a^{-1}\ \Lambda\bigl(\noloop\bigr);\quad

995: \Lambda\bigl(\ \ReidI{-}{ }\bigr) = a\ \Lambda\bigl(\noloop\bigr)\,,\\[2mm]

996: & \label{wseven.7} \Lambda\bigl(\,\mbox{\Large $\bigcirc$}\,\bigr) = 1\,.

997: \end{eqnarray}

998: 

999: The \em{Brandt-Lickorish-Millett-Ho polynomial} \cite{BLM} $Q$

1000: is given by $Q(z)=F(1,z)$.

1001: 

1002: \subsection{Semiadequacy\label{Ssaq}}

1003: 

1004: An alternative description of $V$ is given by the Kauffman bracket in

1005: \cite{Kauffman2}. We do not need this description directly, but

1006: for self-containedness is it useful to recall the related concept of

1007: semiadequacy that was popularized in \cite{LickThis}.

1008: 

1009: Let $D$ be an unoriented link diagram.

1010: A \em{state} is a choice of splittings of type $A$ or 

1011: $B$ for any single crossing (see figure \ref{figsplit}), 

1012: Let the \em{$A$-state} of $D$ be the state where all

1013: crossings are $A$-spliced; similarly define the $B$-state.

1014: 

1015: We call a diagram \em{$A$-(semi)adequate} if in the $A$-state

1016: no crossing trace (one of the dotted lines in figure

1017: \reference{figsplit}) connects a loop with itself.

1018: Similarly we define $B$-(semi)adequate. A diagram is

1019: \em{semiadequate} if it is $A$- or $B$-semiadequate, and

1020: \em{adequate} if it is simultaneously $A$- and $B$-semiadequate.

1021: A link is adequate/semiadequate if it has an adequate/semiadequate

1022: diagram. (Here $A$-adequate and $B$-adequate is what is called

1023: $+$adequate resp.\ $-$adequate in \cite{Thistle}.)

1024: 

1025: \begin{figure}[htb]

1026: \[

1027: \diag{8mm}{4}{2}{

1028:    \picline{0 0}{4 2}

1029:    \picmultiline{-5.0 1 -1.0 0}{0 2}{4 0}

1030:    \picputtext{2.7 1}{A}

1031:    \picputtext{1.3 1}{A}

1032:    \picputtext{2 1.6}{B}

1033:    \picputtext{2 0.4}{B}

1034: } \quad

1035: \diag{8mm}{4}{2}{

1036:    \pictranslate{2 1}{

1037:        \picmultigraphics[S]{2}{1 -1}{

1038:            \piccurve{-2 1}{-0.3 0.2}{0.3 0.2}{2 1}

1039:        }

1040:        {\piclinedash{0.05 0.05}{0.01}

1041:         \picline{0 -0.4}{0 0.4}

1042:        }

1043:    }

1044:    \picputtext{2.7 1}{A}

1045:    \picputtext{1.3 1}{A}

1046: } \quad

1047: \diag{8mm}{4}{2}{

1048:    \pictranslate{2 1}{

1049:        \picmultigraphics[S]{2}{-1 1}{

1050:            \piccurve{2 -1}{0.1 -0.5}{0.1 0.5}{2 1}

1051:        }

1052:        {\piclinedash{0.05 0.05}{0.01}

1053:         \picline{0 -0.6 x}{0 0.6 x}

1054:        }

1055:    }

1056:    \picputtext{2 1.6}{B}

1057:    \picputtext{2 0.4}{B}

1058: }

1059: \]

1060: \caption{\label{figsplit}The A- and B-corners of a

1061: crossing, and its both splittings. The corner A (resp. B)

1062: is the one passed by the overcrossing strand when rotated 

1063: counterclockwise (resp. clockwise) towards the undercrossing 

1064: strand. A type A (resp.\ B) splitting is obtained by connecting 

1065: the A (resp.\ B) corners of the crossing. It is useful to

1066: put a ``trace'' of each splitted crossing as an arc connecting

1067: the loops at the splitted spot.}

1068: \end{figure}

1069: 

1070: We also need a part of the semiadequacy formulas for the Jones

1071: polynomial. As in \cite{ntriv}, for an non-negative integer $i$, we

1072: write $V_i$ for $[V]_{\md V+i}$ and $\bar V_i=[V]_{\Md V-i}$. (These 

1073: are the $i+1$-st and $i+1$-st last coefficient of $V$ in its

1074: coefficient list.) For an A- (resp. B-) semiadequate diagram, 

1075: in \cite{ntriv}, and independently in \cite{DasLin}, formulas were 

1076: obtained for $V_{1,2}$ (resp. $\bar V_{1,2}$).

1077: 

1078: \parbox[b]{12.5cm}{

1079: \parskip 5pt plus 3pt minus 2pt\relax

1080: {}From the $A$-state $A(D)$ of a diagram $D$ we define two

1081: graphs.  The first graph, we call it the \em{$A$-graph}

1082: $G(A)=G(A(D))$ has vertices for each loop in $A(D)$, and

1083: an edge between each pair of loops that are connected

1084: by a trace of at least one crossing in $D$. If there are

1085: at least two such traces, we call the edge \em{multiple}.  

1086: 

1087: Let $\bigtriangleup(D)=\bigtriangleup(A(D))$ be the number of

1088: cycles of length 3 (\em{triangles}) in $G(A(D))$.  

1089: 

1090: The second graph, we call it the \em{intertwining graph}

1091: $IG(A)=IG(A(D))$, has vertices for each multiple edge

1092: in $A(D)$, and an edge in $IG(A)$ is drawn between each

1093: pair of vertices in $IG(A)$, whose corresponding multiple

1094: traces in $A(D)$ contain traces of the form shown on the right.

1095: }

1096: \enspace\parbox[t]{4.0cm}{

1097: \vbox to 4.0cm{\vss

1098: \[

1099: \diag{10mm}{4}{4}{

1100:   \picline{2 0}{2 4}

1101:   {\piclinedash{0.1 0.1}{0.01}

1102:    \picline{1 0.8}{2 0.8}

1103:    \picline{3 1.6}{2 1.6}

1104:    \picline{1 2.4}{2 2.4}

1105:    \picline{3 3.2}{2 3.2}

1106:   }

1107:   \picfilledellipse{1.6 1 x}{0.3 1.2}{}

1108:   \picfilledellipse{2.4 3 x}{0.3 1.2}{}

1109: }

1110: \]\vss}

1111: }

1112: 

1113: \begin{theorem}(\cite{ntriv,DasLin})\label{T02}

1114: Assume that $D$ is a connected $A$-adequate diagram. Then

1115: $V_0V_1=\chi(G(A))-1$ and

1116: \[

1117: V_0V_2=\mybin{2-\chi(G(A))}{2}+\chi(IG(A))-\bigtriangleup(A(D))\,.

1118: \]

1119: \end{theorem}

1120: 

1121: Here $\chi$ is the Euler characteristic (number of vertices minus

1122: number of edges); $V_0=\pm 1$ by \cite{LickThis}.

1123: Analogous formulas hold for $B$-adequate diagrams.

1124: 

1125: \subsection{Braids and braid words\label{brs}}

1126: 

1127: The $n$-string \em{braid group} $B_n$ is considered generated by

1128: the Artin \em{standard generators} $\sg_i$ for $i=1,\dots,n-1$. These

1129: are subject to relations of the type $[\sg_i,\sg_j]=1$ for $|i-j|>1$,

1130: which we call \em{commutativity relations} (the bracket denotes the

1131: commutator) and $\sg_{i+1}\sg_i\sg_{i+1}= \sg_i\sg_{i+1}\sg_i$,

1132: which we call \em{Yang-Baxter} (or shortly YB) \rm{relations}.

1133: 

1134: We will make one noteworthy modification of this notation.

1135: In the following $\sg_3$ will stand for the usual Artin generator

1136: for braids of $4$ or more strands (such braids are considered

1137: explicitly only in sections \reference{S2.2} and \reference{SBurau}),

1138: while on $3$-braids (considered in all other sections) it will denote

1139: the ``band'' generator $\sg_2\sg_1\sg_2^{-1}=\sg_1^{-1}\sg_2\sg_1$.

1140: 

1141: It will be often convenient in braid words to write $\pm i$ for

1142: $\sg_i^{\pm 1}$. For example, $[21(33-4)^232]$ is an alternative

1143: writing for $\sg_2\sg_1(\sg_3^2\sg_4^{-1})^2\sg_3\sg_2$.

1144: 

1145: The following definition summarizes basic terminology of

1146: braid words used throughout the paper.

1147: 

1148: \begin{defi}\label{dfext}

1149: Choose a word 

1150: \begin{eqn}\label{wd}

1151: \bt=\prod_{i=1}^n\sg_{k_i}^{l_i}

1152: \end{eqn}

1153: with $k_i\ne k_{i+1}$ and $l_i\ne 0$. We understand such words

1154: \em{in cyclic order}.

1155: 

1156: Call $\sg_{k_i}^{l_i}$ the \em{syllables}

1157: of $\bt$. For such a syllable, let $k_i$ be called the \em{index}

1158: of the syllable, and $l_i$ its \em{exponent} or \em{length}.

1159: We call a syllable $\sg_{k_i}^{l_i}$ \em{non-trivial} if $|l_i|>1$

1160: and \em{trivial} if $|l_i|=1$.

1161: 

1162: We say $n$ is the \em{syllable length} of $\bt$ in \eqref{wd}.

1163: The \em{(word) length} $c(\bt)$ of $\bt$ is $\sum_{i=1}^n |l_i|$, and

1164: $c_{\pm}(\bt)=\sum_{\pm l_i>0}\pm l_i$ are the \em{positive/negative

1165: length} of $\bt$. A word is \em{positive} if $c_-(\bt)=0$, or

1166: equivalently, if all $l_i>0$. The \em{exponent sum} of $\bt$ is

1167: defined to be $[\bt]=\sum_{i=1}^n l_i$. The \em{index sum} of

1168: $\bt$ is $\sum_{i=1}^n k_i\cdot |l_i|$ (i.e.

1169: each letter, not just syllable in $\bt$, contributes to that sum).

1170: 

1171: For $3$-braid words, the $k_i$ interchange between $1$ and $2$.

1172: Thus the vector of the $l_i$, considered up to cyclic permutations,

1173: determines the conjugacy class unambiguously. We call it

1174: the \em{Schreier vector}.

1175: 

1176: Let $\bt$ be a positive word.

1177: We call $\tl\bt$ an \em{extension} of $\bt$ if $\tl\bt$ is obtained by

1178: replacing some (possibly no) trivial syllables in $\bt$ by non-trivial

1179: ones of the same index. Contrarily, we call $\bt$ a \em{syllable

1180: reduction} of $\tl\bt$. We call $\bt$ \em{non-singular} if $[\bt]_k

1181: >1$ for all $k=1,\dots,n-1$, where $[\bt]_k=\sum_{k_i=k}l_i$ is the

1182: exponent sum of $\sg_k$ in $\bt$. If $[\bt]_k=1$, we say that the

1183: syllable of index $k$ in $\bt$ is \em{isolated} or \em{reducible}.

1184: \end{defi}

1185: 

1186: To avoid confusion, it seems useful to clarify \em{a priori}

1187: the following use of symbols (even though we recall it at

1188: appropriate places later).

1189: 

1190: A comma separated list of integers

1191: will stand for a \em{sequence of syllable indices} of a braid word. 

1192: The `at' sign `@', written after such an index means that the

1193: corresponding syllable is trivial, while by an exclamation mark

1194: `$!$' we indicate that the syllable is non-trivial. (As the

1195: exponent for non-trivial syllables will be immaterial, it is

1196: enough to distinguish only whether the syllable is trivial or not.)

1197: If none of $!$ and @ is specified, we do not exclude explicitly

1198: any of either types. 

1199: 

1200: A bracketed but non-comma separated list of integers will stand

1201: for a braid word. An asterisk `$*$' put after a letter (number)

1202: in such a word means that this letter may be repeated

1203: (it need not be repeated, but it must not be omitted).

1204: So a (possibly trivial) index-2 syllable can be written

1205: as $22*$. The expression `$[23]$' should mean a letter which

1206: is either `2' or `3', and `$[23]+$' means a possibly empty

1207: sequence of letters `$2$' and `$3$'.

1208: 

1209: \subsection{Braid representations of links\label{brt}}

1210: 

1211: By a theorem of Alexander \cite{Alexander}, any link is the \em{closure}

1212: $\hat\bt$ of a braid $\bt$. The \em{braid index} $b(L)$ of a link $L$

1213: is the smallest number of strands of a braid $\be$ whose closure

1214: $\hat\be$ is $L$. See \cite{Morton,WilFr,Murasugi}. Such $\bt$ are

1215: also called \em{braid representations} of $L$. The closure operation

1216: gives for a particular braid word $\bt$ a link diagram $D=\hat\bt$.

1217: Then we have for example $w(\hat\bt)=[\bt]$, $c(\bt)=c(\hat\bt)$,

1218: $c_\pm(\bt)=c_\pm(\hat\bt)$, and $s(\hat\bt)$

1219: is the number of strings of $\bt$ (i.e. $n$ for $\bt\in B_n$).

1220: 

1221: Many properties of braid words we will deal with

1222: relate to the corresponding properties of their link diagrams.

1223: For example, a braid word is called \em{positive} if it contains no

1224: $\sg_i^{-1}$, or in other words, its closure diagram is positive.

1225: A braid is positive if it has a positive braid word.

1226: In a similar fashion, we say that braid is ($A$/$B$-)adequate if

1227: it has an ($A$/$B$-)adequate word representation, and a word is

1228: adequate if the link diagram obtained by its closure is adequate. 

1229: 

1230: If a braid word $\bt$ is written as $\sg_1^{\pm 1}\ap\sg_1^p\ap'$,

1231: where $p\in\bZ$ and none of $\ap$ and $\ap'$ contains a syllable of

1232: index $1$, then the diagram admits a \em{flype}, which exchanges

1233: the syllables of index $1$ in $\bt$, so that we obtain $\sg_1^p

1234: \ap\sg_1^{\pm 1}\ap'$. This operation preserves the isotopy

1235: type of the closure link $\hat\bt$, but in general changes

1236: the braid conjugacy class. The phenomenon is explained in

1237: \cite{BirMen}. In the context of general link diagrams,

1238: the flype has been studied also extensively, most prominently

1239: in \cite{MenThis}.

1240: 

1241: Alternatively to the standard Artin generators, one

1242: considers also a representation of the braid groups by

1243: means of an extended set of generators (and their inverses)

1244: \[

1245: \sg_{i,j}^{\pm 1}=\sg_i\dots\sg_{j-2}\sg_{j-1}^{\pm 1}

1246: \sg_{j-2}^{-1}\dots\sg_i^{-1}

1247: \]

1248: for $1\le i<j\le n$. Note that 

1249: \begin{eqn}\label{op}

1250: \sg_i=\sg_{i,i+1}\,.

1251: \end{eqn}

1252: A representation of a braid $\bt$, and its closure link $L=\hat\bt$,

1253: as word in $\sg_{i,j}^{\pm 1}$ is called a \em{band representation}

1254: \cite{BKL}. A band representation of $\bt$ spans naturally a

1255: Seifert surface of the link $L$ as in figure \ref{figbd}: one

1256: glues disks into the strands, and connects them by half-twisted

1257: bands along the $\sg_{i,j}$. The resulting surface is called

1258: \em{braided Seifert surface} of $L$. A minimal genus Seifert

1259: surface of $L$, which is a braided Seifert surface, is also

1260: called a \em{Bennequin surface} \cite{BirMen2}.

1261: 

1262: In this paper we will deal exclusively with band representations

1263: in $B_3$. Then we have three band generators $\sg_{i,i\bmod 3+1}$

1264: (where $i=1,2,3$, and `$\bmod$' is taken with values between $0$ and

1265: $2$). With \eqref{op}, we have $\sg_1=\sg_{1,2}$ and $\sg_2=\sg_{2,3}$,

1266: and with the special meaning of $\sg_3\in B_3$ introduced above,

1267: $\sg_3=\bar\sg_{1,3}$, where bar denotes the mirror image.

1268: (This mirroring is used here for technical reasons related to

1269: Xu's normal form, as explained below.)

1270: 

1271: If a band representation contains only positively half-twisted bands

1272: (i.e. no $\sg_{i,j}^{-1}$ occur), it is called \em{band-positive}

1273: or \em{strongly quasi-positive}. A link with a strongly quasi-positive

1274: band (braid) representation is called strongly quasi-positive.

1275: Such links have an importance in connection to algebraic curves;

1276: see \cite{Rudolph2}.

1277: 

1278: Using the skein polynomial, define a quantity by

1279: \begin{eqn}\label{mwfb}

1280: MWF(L)\,=\,\frac 12\,\big(\Md_vP-\md_vP\big)+1\,.

1281: \end{eqn}

1282: The \em{Morton-Williams-Franks braid index inequality}

1283: \cite{Morton,WilFr} (abbreviated as MWF) states that

1284: \begin{eqn}

1285: b(L)\ge MWF(L)

1286: \end{eqn}

1287: for every link $L$. This inequality is often exact (i.e. an

1288: equality). The study of links where it is exact or not

1289: has occupied a significant part of previous literature.

1290: Most noteworthy is the work of Murasugi \cite{Murasugi}

1291: and Murasugi-Przytycki \cite{MurPrz2}.

1292: 

1293: The Morton-Williams-Franks 

1294: results from two other inequalities, due to Morton,

1295: namely that for a diagram $D$, we have

1296: \begin{eqn}\label{mq}

1297: 1-s(D)+w(D)\,\le\,\md_lP(D)\,\le\,\Md_lP(D)\,\le\,s(D)-1+w(D)\,.

1298: \end{eqn}

1299: Williams-Franks showed these inequalities for the case of braid

1300: representations (i.e. when $D=\hat\bt$ for some braid $\bt$).

1301: Later it was observed from the

1302: algorithm of Yamada \cite{Yamada} and Vogel \cite{Vogel}

1303: that the braid version is actually equivalent to (and not just

1304: a special case of) the diagram version.

1305: 

1306: These inequalities were later improved in \cite{MurPrz2} in a way 

1307: that allows to settle the braid index problem for many links

1308: (see \S\ref{Sfg} or also \cite{Ohyama}). 

1309: 

1310: In the special case of $3$-braids (and with the special meaning

1311: of $\sg_3$ as described above), Xu \cite{Xu} gives a normal form

1312: of a conjugacy class in $\sg_{1,2,3}$. By Xu's algorithm, each

1313: $\bt\in B_3$ can be written in one of the two forms

1314: {\def\theenumi{\Alph{enumi}}

1315: \def\labelenumi{(\theenumi)}

1316: \begin{enumerate}

1317: \item\label{itA} $[21]^kR$ or $L^{-1}[21]^{-k}$ ($k\ge 0$), or

1318: \item\label{itB} $L^{-1}R$,

1319: \end{enumerate}

1320: }where $L$ and $R$ are positive words in $\sg_{1,2,3}$ with (cyclically)

1321: non-decreasing indices (i.e. each $\sg_i$ is followed by $\sg_i$

1322: or $\sg_{i\bmod 3+1}$, with `$\bmod$' taken with values between $0$

1323: and $2$). Since the form \reference{itB} must be cyclically

1324: reduced, we may assume that $L$ and $R$ do not start or end with

1325: the same letter. This form is the shortest word in $\sg_{1,2,3}$

1326: of a conjugacy class. By Bennequin's aforementioned result, the

1327: braided surface is then a minimal genus (or Bennequin) surface.

1328: 

1329: \subsection{Gau\ss{} sum invariants}

1330: 

1331: We recall briefly the definition of Gau\ss{} sum invariants.

1332: They were introduced first in \cite{Fied} for braids, and later

1333: \cite{Fied2,VirPol} for knots. It is known that all

1334: they give formulas for Vassiliev invariants.

1335: 

1336: \begin{defi}(\protect\cite{Fied2})\ 

1337: A Gau\ss{} diagram of a knot diagram is an oriented circle with

1338: arrows connecting points on it mapped to a crossing and

1339: oriented from the preimage of the undercrossing (underpass)

1340: to the preimage of the overcrossing (overpass).

1341: \end{defi}

1342: 

1343: We will call a pair of crossings whose arrows intersect in

1344: the Gau\ss{} diagram a \em{linked} pair.

1345: 

1346: \begin{exam}

1347: As an example, figure \ref{fig6_2} shows the knot $6_2$ in its

1348: commonly known projection and the corresponding Gau\ss{} diagram.

1349: \end{exam}

1350: 

1351: \begin{figure}[htb]

1352: {%\small

1353: \[

1354: \begin{array}{c@{\qquad}c}

1355:      \diag{2cm}{2.0}{2.0}{

1356:   \picputtext[dl]{0 0}{\autoepsffalse\epsfs{4cm}{6_2}}

1357:   \picputtext[u]{1.25 1.65}{$1$} 

1358:   \picputtext[u]{0.8 1.65}{$2$} 

1359:   \picputtext[u]{0.6 0.6}{$3$} 

1360:   \picputtext[u]{1.0 0.35}{$4$} 

1361:   \picputtext[u]{1.3 0.9}{$5$} 

1362:   \picputtext[u]{0.9 1.3}{$6$} 

1363: 	}

1364:  & 

1365: \diag{1.6cm}{2.4}{2.4}{\pictranslate{1.2 1.2}{

1366: \piclinewidth{60}

1367: \piccircle{0 0}{1}{}

1368: \labar{240}{ 30}{1}

1369: \labar{ 60}{210}{2}

1370: \labar{300}{ 90}{3}

1371: \labar{120}{330}{4}

1372: \labar{  0}{150}{5}

1373: \labar{180}{270}{6}

1374: } % \pictranslate

1375: } % \diag

1376: \end{array}

1377: \]}

1378: \caption{The standard diagram of the knot $6_2$ and its Gau\ss{}

1379: diagram.\label{fig6_2}}

1380: \end{figure}

1381: 

1382: The simplest (non-trivial) Vassiliev knot invariant is the

1383: \em{Casson invariant} $v_2$, with $v_2=\Dl''(1)/2=-V''(1)/6$,

1384: for which Polyak-Viro

1385: \cite{VirPol,VirPol2} gave the simple Gau\ss{} sum formula

1386: \begin{eqn}\label{pv2_}

1387: v_2\,=\,\GD{\arrow{225}{45}

1388: \arrow{315}{135}

1389: \picfillgraycol{0}\picfilledcircle{1 90 polar}{0.09}{}

1390: }\,.

1391: \end{eqn}

1392: Here the point on the circle corresponds to a point on the

1393: knot diagram, to be placed arbitrarily except on a crossing.

1394: (The expression does not alter with the position of the

1395: basepoint; we will hence have, and need, the freedom to place

1396: it conveniently.)

1397: 

1398: We will use the symmetrized version of \eqref{pv2_}

1399: w.r.t. taking the mirror image of the knot diagram:

1400: \begin{eqn}\label{pv2}

1401: v_2\,=\,\frac{1}{2}\es\left(

1402: \GD{\arrow{225}{45}

1403: \arrow{315}{135}

1404: \picfillgraycol{0}\picfilledcircle{1 90 polar}{0.09}{}

1405: }+

1406: \GD{\arrow{45}{225}

1407: \arrow{135}{315}

1408: \picfillgraycol{0}\picfilledcircle{1 90 polar}{0.09}{}

1409: }\right)\,.

1410: \end{eqn}

1411: 

1412: \section{Xu's form and Seifert surfaces\label{SA}}

1413: 

1414: We studied the relation of Xu's algorithm and the skein

1415: polynomial in \cite{3br}, and here we will go further to

1416: connect fiberedness, the and Alexander and Jones polynomial

1417: to Xu's form.

1418: 

1419: \subsection{Strongly quasi-positive links among links of braid index 3}

1420: 

1421: Theorem \reference{thqp} follows relatively easily from

1422: the work in \cite{3br}, but it is a good starting point for

1423: the later more substantial arguments basing on Xu's form.

1424: 

1425: \proof[of theorem \reference{thqp}]

1426: The inequalities of \eqref{mq} for the $v$-degree

1427: of $P$ applied on a positive band representation show that,

1428: with $P=P(L)$ and $\chi=\chi(L)$,

1429: \begin{eqn}\label{x}

1430: \md_vP\ge 1-\chi\,.

1431: \end{eqn}

1432: Because of \cite[proposition 21]{LickMil}, which now says

1433: \begin{eqn}\label{p21}

1434: P(v,v^{-1}-v)=1\,,

1435: \end{eqn}

1436: we have $\md_lP\le

1437: \Md_zP$, and by \cite{3br} $\Md_zP=1-\chi$. So from \eqref{x}

1438: we obtain 

1439: \begin{eqn}\label{y}

1440: \md_lP=1-\chi\,. 

1441: \end{eqn}

1442: 

1443: It is known that the minimal degree term in $z$ of the skein polynomial

1444: of a $n$-component link is divisible by $(v-v^{-1})^{n-1}$. (In

1445: \cite{Kneissler} in fact all occurring terms are classified.) So

1446: if $MWF (K)=1$, then $K$ is a knot. Now, the identity \eqref{p21}

1447: implies that if $MWF=1$ for some knot $K$, then $P(K)=1$. For

1448: any non-trivial knot $1-\chi=2g>0$, so for any non-trivial strongly

1449: quasi-positive knot from \eqref{x} we have $\md_vP>0$, and so $P\ne 1$.

1450: 

1451: So a strongly quasi-positive link cannot have $P=1$, and always

1452: $MWF\ge 2$.  % (The split links are easy to deal with.)

1453: Therefore, when the braid index is 3, we have

1454: $MWF\in \{2, 3\}$. Then \eqref{y} and the inequalities \eqref{mq}

1455: for the $v$-degree of $P$ show that a 3-braid representation $\bt$

1456: has exponent sum $[\bt]=3-\chi$, unless $MWF=2$ and $[\bt]=1-\chi$.

1457: 

1458: In former case, we can find a minimal genus band representation

1459: from $\bt$ by Xu's algorithm \cite{Xu}, and this representation

1460: must be positive. In latter case, we will have one negative band,

1461: and have Xu's form $L^{-1}R$. Here $L$ and $R$ are positive words in

1462: the letters $\sg_1,\sg_2,\sg_3=\sg_2\sg_1\sg_2^{-1}$ with $\sg_i$

1463: followed by $\sg_i$ or $\sg_{i\bmod 3+1}$ (as described

1464: in \S\reference{brt}), and $L$ is a single letter.

1465: 

1466: W.l.o.g. assume $L=\sg_3$. If the first or last letter in $R$

1467: is a $\sg_3$ then we can cancel two bands, and have a positive

1468: representation. If some $\sg_3$ occurs in $R$, then we can write

1469: $L^{-1}R=\sg_3^{-1}\ap\sg_3\ap'$ where $\ap,\ap'$ are positive

1470: band words. Such a representation is quasi-positive with a Seifert

1471: ribbon \cite{Rudolph} of smaller genus. However, it is known that

1472: for strongly quasi-positive links the genus and 4-genus coincide

1473: (see for example \cite{posqp}). If $R$ has no $\sg_3$, then

1474: we have up to cyclic letter permutation a braid of the form

1475: $\sg_1^k\sg_2^l\sg_1^{-1}\sg_2^{-1}$ with $k\ge 0$, $l>0$. This

1476: braid is easily seen to reduce to (a positive one on) two strands.

1477: \qed

1478: 

1479: \begin{rem}\label{rqw}

1480: It is not true that all 3-braids that have the skein polynomial of

1481: $(2,n)$-torus links have positive 3-band representations. Birman's

1482: \cite{Birman} construction (see definition \reference{DEFB} below)

1483: yields examples like $12_{2037}$ in \cite{KnotScape} with one

1484: negative band. This construction gives in fact all non-obvious

1485: 3-braid $MWF=2$ examples, a circumstance shown in \cite{ntriv} by

1486: applying theorem \ref{thq} (and reportedly in previous unpublished

1487: work of El-Rifai). The result is explained and used below in the

1488: proof of theorem \reference{thJC}.

1489: \end{rem}

1490: 

1491: \subsection{Uniqueness of minimal genus Seifert surfaces\label{Uq}}

1492: 

1493: The discussion here came about from the desire to complete Xu's

1494: uniqueness theorem for Seifert surfaces of 3-braid links. While

1495: our result will be further improved later using work to follow,

1496: we need to introduce some notation and basic tools. We will use

1497: the work of Kobayashi \cite{Kobayashi}, implying that the property

1498: a surface to be a unique minimal genus surface is invariant under

1499: Hopf (de)plumbing.

1500: 

1501: Subsequently Mikami Hirasawa advised me about a subtlety

1502: concerning the notion of `uniqueness' which must be explained.

1503: Xu's work considers Seifert surfaces unique in the sense isotopic

1504: to each other, if we \em{may move} the link by the isotopy.

1505: In particular, such isotopy may interchange link components.

1506: However, we demand the isotopy to preserve component orientation.

1507: (For unoriented isotopy Xu's result is complete.) Contrarily,

1508: Kobayashi's setting assumes uniqueness in the sense that

1509: Seifert surfaces are isotopic to each other \em{fixing} the link.

1510: Hirasawa explained that the definitions are not equivalent, and

1511: that a unique (if we may move the link) minimal genus surface

1512: may get not unique when one (de)plumbs a single Hopf band. To

1513: account for this discrepancy, we should establish proper language.

1514: 

1515: \begin{defi}

1516: Let us use the term \em{unique} for

1517: Kobayashi's notion of uniqueness (up to isotopy fixing the

1518: link), and let us call Xu's notion of uniqueness (up to

1519: isotopy which may move the link or permute components,

1520: but preserves orientation) \em{weakly unique}.

1521: \end{defi}

1522: 

1523: Here we state the following extension of Xu's uniqueness theorem.

1524: 

1525: \begin{theorem}\label{wu}

1526: Every 3-braid link has a weakly unique minimal genus Seifert surface.

1527: \end{theorem}

1528: 

1529: 

1530: \proof Birman-Menasco showed that any minimal genus Seifert surface

1531: of a 3-braid link is isotopic to a Bennequin surface. Then, Xu

1532: showed that every conjugacy class of 3-braids carries a canonical

1533: (up to oriented isotopy) Bennequin surface. The conjugacy classes

1534: of 3-braids with given closure link were classified in \cite{BirMen}.

1535: Most links admit a single conjugacy class, and we are done, as in

1536: \cite{Xu}. The exceptional cases are easy to deal with, except the

1537: ``flype admitting'' braids 

1538: \begin{eqn}\label{fl3}

1539: \sg_1^{\pm p}\sg_2^{\pm q}\sg_1^{\pm r}\sg_2^{\pm 1}\,.

1540: \end{eqn}

1541: The flype interchanges $\pm q$ and $\pm 1$, and (in general) gives

1542: a different conjugacy class, which differs from the original one

1543: by orientation. So we have (as in \cite{Xu}), two Bennequin surfaces

1544: isotopic only up to orientation. We will settle this case now also

1545: for oriented isotopy, that is, show that these surfaces are

1546: isotopic to themselves with the opposite orientation.

1547: 

1548: Recall that for a fibered link, a minimal genus surface is the

1549: same as a fiber surface, and such a surface is unique (Neuwirth-%

1550: Stallings theorem). Moreover, by work of Gabai \cite{Gabai,Gabai2,%

1551: Gabai3} and Kobayashi \cite{Kobayashi} the properties of a surface

1552: to ba a minimal genus surface, a unique minimal genus surface or a

1553: fiber surface are invariant under Murasugi (de)sum with a fiber

1554: surface. In particular, this invariance holds for (de)plumbing

1555: a Hopf band (which is understood to be an unknotted annulus with 

1556: one full, positive or negative,

1557: twist). Now we note that in the cases in \eqref{fl3} where $p=r$,

1558: the flype is trivial (i.e. realized by a conjugacy), so that the

1559: surface is weakly unique. However, since Kobayashi's theorem may

1560: fail for weakly unique surfaces, we cannot reduce our surfaces to

1561: this case. We will use Hopf (de)plumbings to recur all cases to fiber

1562: surfaces or a 2-full twisted annulus. Then we understand that our

1563: surfaces are unique, and in particular weakly unique. The type of

1564: Hopf (de)plumbings we will apply is to interconvert all powers

1565: of a given band generator of given sign (for example $\sg_3^k$ is

1566: equivalent to $\sg_3^{-1}$ for each $k<0$).

1567: 

1568: W.l.o.g. assume in \eqref{fl3} that we have $-$ in $\pm 1$ and that none

1569: of $p,q,r$ is $0$ (the other cases are easy). We assume $p,q,r>0$

1570: and vary the signs before $p,q,r$ properly. Also, since the flype

1571: interchanges $\pm p$ and $\pm r$, we may assume $\pm p\ge \pm r$.

1572: So we exclude the sign choice $(-p,+r)$.

1573: 

1574: \begin{caselist}

1575: \case $p,q,r>0$. We can write \eqref{fl3} up to cyclic permutation

1576: as $(\sg_2^{-1}\sg_1\sg_2)^p\sg_2^{q-1}\sg_1^r$, and conjugating with

1577: $\sg_2\sg_1\sg_2$ we have $\sg_3^p\sg_1^{q-1}\sg_2^r$. Since $p,r>0$,

1578: this is Hopf plumbing equivalent to $\sg_3\sg_1\sg_2$ or $\sg_3\sg_2$.

1579: These cases are a disk and a 2-full twisted annulus for

1580: the reverse $(2,4)$-torus link, and we are done.\label{cs1}

1581: 

1582: \case $p,r>0$, $-q<0$. We have an alternating braid, which is

1583: fibered.

1584: 

1585: \case $p,q,r<0$. We have a negative braid, which is fibered.

1586: 

1587: \case $-p,-r<0$, $q>0$. We have a word of the form 

1588: $\sg_2^{q}\sg_1^{-r}\sg_2^{-1}\sg_1^{-p}$. If one of

1589: $p$ and $r$ is $1$, then we have a $(2,\,.\,)$-torus link.

1590: If $q=1$, then we can go over to the mirror image, and land in

1591: case \reference{cs1}. Now, when $p,q,r\ge 2$, the

1592: minimal word in Xu's form is $2^{q-2}(-3)(-2)^{r-1}(-1)^{p-1}$.

1593: Since $p-1$ and $r-1$ are non-zero, its surface is plumbing

1594: equivalent to the one of $[2-3-2-1]$ (if $q>2$) or $[-3-2-1]$

1595: (if $q=2$). These are the annulus for the reverse $(2,-4)$-torus link,

1596: and the fiber of the $(-2,-2,2)$-pretzel link, and we are done. 

1597: 

1598: \case $p>0$, $-r<0$, $q>0$.

1599: We have 

1600: \[

1601: \sg_1^p\sg_2^q\sg_1^{-r}\sg_2^{-1}=\sg_1^p\sg_2^{q-1}

1602: (\sg_2\sg_1^{-1}\sg_2^{-1})^r=\sg_1^p\sg_2^{q-1}\sg_3^{-r}\,.

1603: \]

1604: Such a surface is plumbing equivalent to the one for $\sg_1\sg_3^{-1}$

1605: (if $q=1$) resp. $\sg_1\sg_2\sg_3^{-1}$ (if $q>1$), which are the

1606: fibers of the unknot and Hopf link resp.

1607: 

1608: \case $p>0$, $-r<0$, $-q<0$. We have 

1609: \[

1610: \sg_1^{-r}\sg_2^{-1}\sg_1^p\sg_2^{-q}=\sg_1^{1-r}(\sg_1^{-1}\sg_2^

1611: {-1}\sg_1)\sg_1^{p-1}\sg_2^{-q}=\sg_1^{1-r}\sg_3^{-1}\sg_1^{p-1}

1612: \sg_2^{-q}\,.

1613: \]

1614: If $-r=-1$ then we have a $(2,p-q-1)$-torus link. If $-r<-1$, then

1615: the band surface from the right word above is plumbing equivalent to the

1616: one for $\sg_1^{-1}\sg_3^{-1}\sg_2^{-1}$ (if $p=1$) or 

1617: $\sg_1^{-1}\sg_3^{-1}\sg_1\sg_2^{-1}$ (if $p>1$), which are the

1618: annulus for the reverse $(2,4)$-torus link, and the fiber of the

1619: $(2,2,-2)$-pretzel link. \qed

1620: \end{caselist}

1621: 

1622: \subsection{Fiberedness\label{ppp}}

1623: 

1624: For the rest of the paper we normalize

1625: $\Dl$ so that $\Dl(1)=1$ and $\Dl(t)=\Dl(1/t)$.

1626: 

1627: \begin{theorem}\label{Thq}

1628: Let $L$ be a strongly quasi-positive 3-braid link. Then the following

1629: are equivalent:

1630: \def\labelenumi{\theenumi)}

1631: \begin{enumerate}

1632: \item $L$'s minimal genus surface is a Hopf plumbing,

1633: \item $L$'s minimal genus surface is a fiber surface,

1634: \item $\Md\Dl(L)=1-\chi(L)$ and $\Mc\Dl(L)=\pm 1$,

1635: \item $L$'s Xu normal form is \em{not} $R$, with syllable

1636: length of $R$ divisible by $3$. In other words, the Xu normal

1637: form is not an extension (in the sense of definition

1638: \ref{dfext}) of $[(123)^k]$ for $k>0$.

1639: \item Some minimal band form of $L$ contains

1640: $\sg_1^k\sg_2^l\sg_1^m$ or $\sg_2^k\sg_1^l\sg_2^m$

1641: as subword for $k,l,m>0$, or is $\sg_1^k\sg_2^l$.

1642: \end{enumerate}

1643: \end{theorem}

1644: 

1645: \proof  $(1)\So (2)\So (3)$ are clear.

1646: 

1647: $(5)\So (1)$. Assume after deplumbing, all letters occur in single

1648: power, and up to conjugacy the word starts with $\sg_1\sg_2\sg_1$

1649: or $\sg_2\sg_1\sg_2$. By adjusting one of the two, we can have a

1650: $\sg_1^2$ or $\sg_2^2$ if the next letter is $\sg_1$ or $\sg_2$. In

1651: that case we deplumb a Hopf band. If the next letter is $\sg_3$, then

1652: we have $\sg_1\sg_2\sg_1\sg_3=\sg_1\sg_2^2\sg_1$, can can also deplumb

1653: a Hopf band. Then we reduce the surface for that of $\sg_1\sg_2\sg_1$

1654: which is the Hopf band.

1655: 

1656: $(3)\So (4)$. We prove the contrary. Assume $(4)$ does not hold. 

1657: Band-positive surfaces are always of minimal genus, so that the

1658: properties we investigate are invariant under Hopf (de)plumbings.

1659: Under applying skein relations at non-trivial syllables we are

1660: left with powers of $\sg_1\sg_2\sg_3$. Apply the skein relation

1661: for $\Dl$ at the last band. Then $\bt_{-}$ and $\bt_0$ are both

1662: of minimal length. We already proved, in $(5)\so (1)$, that

1663: $\bt_0$ is fibered, so $\Md\Dl=1-\chi(\bt_0)$ and $\Mc\Dl=\pm 1$.

1664: The same holds for $\bt_{-}$ by \cite[proposition 2]{3br}, since 

1665: $\bt_{-}$ is of Xu's minimal form $L^{-1}R$ and is not positive.

1666: So the terms in degree $1-\chi(\bt)$ of $\Dl(\bt)$ either cancel, or

1667: give $\pm 2$. 

1668: 

1669: $(4)\So (5)$. We prove the contrary. Assume $(5)$ does not hold.

1670: If we do not have a word $\bt=R$ in Xu's form of length divisible by

1671: $3$, another option would be to have a word with cyclically decreasing

1672: indices. But note that $[321]=[121]$ is the Hopf band, while

1673: $[321321]=[211211]$ contains a $121$, too, and so we

1674: are done. Otherwise the index array of

1675: the syllables of $\bt$ must contain the same entry with distance $2$,

1676: and such a word is conjugate to the ones excluded.

1677: \qed

1678: 

1679: \begin{theorem}\label{xLR}

1680: Any closed braid of Xu's form $L^{-1}R$ is fibered.

1681: \end{theorem}

1682: 

1683: The proof uses some work of Hirasawa-Murasugi.

1684: A consequence of their result is the following lemma,

1685: which we require. It is proved in appendix \reference{HM}.

1686: 

1687: \begin{lemma}\label{lki}

1688: The links $L_k$, given by the closed 3-braids $[(123)^k\,-2]$ for $k>0$,

1689: are fibered.

1690: \end{lemma}

1691: 

1692: % \proof Consider the knots $L_k'$ given by the closed 3-braid

1693: % words $[(12)^{3k+1}(-2)^{3k+l}]$, where $l=2$ for

1694: % $k$ even, and $l=3$ for $k$ odd. The Xu normal form is found using

1695: % the identity $2^3(123)=(12)^3=(123)2^3$ to be

1696: % \[

1697: % (12)^{3k+1}(-2)^{3k+l}\to 12(123)^k(-2)^l\to 21223(123)^{k-1}(-2)^l\to

1698: % 1223(123)^{k-1}(-2)^{l-1}\,,

1699: % \]

1700: % and this surface is Hopf plumbing equivalent to that of $L_k$.

1701: % 

1702: % We know from \cite{3br} (as explained further in \cite{ntriv})

1703: % that if $\bt\in B_3$ is not strongly quasi-positive, then

1704: % $2\Md\Dl=1-\chi$ and $\Mc\Dl=\pm 1$.

1705: % 

1706: % Now, since $L_k'$ has Xu normal form of the type $L^{-1}R$, we

1707: % find that $\Mc\Dl(L_k')=\pm 1$. Then, by the result of

1708: % Hirasawa-Murasugi $L_k'$ is fibered, and hence so is $L_k$. \qed

1709: 

1710: 

1711: \proof[of theorem \reference{xLR}]

1712: We use induction on the length of $L^{-1}R$ and for fixed

1713: length on the exponent sum. Under Hopf deplumbings assume all

1714: syllables in $L$ and $R$ are trivial.

1715: Assume up to mirroring that $L$ is not shorter than $R$.

1716: Permute by conjugacy $R$ to

1717: the left, and permute the indices so that $L$ starts with $-3$. 

1718: 

1719: The following transformations also offer a Hopf deplumbing

1720: \[

1721: 1-3-2-1\to -2 1 -2 -1\to -2 -2 -1 2\to -2 -1 2

1722: \]

1723: and

1724: \[

1725: 1 2 -3\to -2 1 1

1726: \]

1727: These reductions fail both if either $L$ and $R$ have length at most

1728: $2$, or $R$ has length $1$. (Remember $L$ is not shorter than $R$.)

1729: In former case one checks directly that one has a disk, Hopf band or

1730: connected sum of two Hopf bands. We consider latter case.

1731: 

1732: By conjugacy permute the indices so that $R=1$; also assume $L$ has

1733: length at least $3$. If $L$ starts and ends with $-2$ then we do

1734: \[

1735: 1 -2 -1 \dots -2 \to -2 3 -1 \dots -2\to 3 -1 \dots -2 -2

1736: \]

1737: (where the right transformation is a conjugacy) and deplumb a

1738: Hopf band. If $L$ starts with $-3$ then we transform as before

1739: \[

1740: 1 -3 -2 -1\to -2 1 -2 -1\to -2 -2 -1 -2

1741: \]

1742: and deplumb a Hopf band. So $L$ starts with $-2$ and ends on $-3$.

1743: Then the mirror image of $\bt$ is up to conjugacy of the form

1744: $[(123)^k\,-2]=[(1 2 2 1)^k (-2)^{k+1}]$, which we dealt with in

1745: the lemma before.  \qed

1746: 

1747: Combinedly, we obtain

1748: 

1749: \begin{corr}

1750: Let $W$ be a 3-braid link. Then the following are equivalent:

1751: \def\labelenumi{\theenumi)}

1752: \begin{enumerate}

1753: \item $W$ is fibered,

1754: \item $\Md\Dl(W)=1-\chi(W)$ and $\Mc\Dl(W)=\pm 1$,

1755: \item $W$'s Xu form is not of the type $L^{-1}$ or $R$,

1756: with syllable length of $L$ or $R$ divisible by $3$. \qed

1757: \end{enumerate}

1758: \end{corr}

1759: 

1760: \begin{exam}

1761: The routine verification, with Mikami Hirasawa, of the tables in

1762: \cite{KnotScape} for fibered knots, has shown non-fibered knots 

1763: with monic Alexander polynomial of degree matching the genus

1764: start at 12 crossings. One of these knots, $12_{1752}$, has

1765: braid index $4$, so that (as expected) the corollary does not

1766: hold for 4-braids.

1767: \end{exam}

1768: 

1769: \begin{corr}

1770: Any non-split closed 3-braid $\hat\bt$ with $|[\bt]|\le 2$ is fibered,

1771: in particular so is any slice 3-braid knot. \qed

1772: \end{corr}

1773: 

1774: Again $6_1$ and $9_{46}$ show the second part is not true for 4-braids.

1775: The first statement in the corollary extends simultaneously

1776: the property of amphicheiral knots, which follows also

1777: from \cite{BirMen}, since amphicheiral 3-braid

1778: knots are closed alternating 3-braids.

1779: 

1780: Originally the insight about a unique minimal genus surface motivated

1781: the fibered 3-braid link classification. Still this insight lacks

1782: asset as to the somewhat improper way it emerges. Note for example,

1783: that the stronger version of \cite{BirMen2} of Bennequin's theorem

1784: enters decisively into \cite{Xu} and the proof of theorem \ref{wu},

1785: but then paradoxically latter imply that Birman-Menasco's formulation

1786: is actually equivalent to, and not really an improvement of,

1787: Bennequin's theorem. Also, Birman-Menasco's classification

1788: \cite{BirMen} priorizes Schreier's conjugacy algorithm and lacks

1789: any geometric interpretation of Xu's normal form, while such an

1790: interpretation becomes evident in our setting. This provided strong

1791: motivation for theorem \reference{thuq}. Its proof is completed

1792: by dealing with the non-fibered cases in appendix \reference{HI}.

1793: It requires also a part of the further detailed consideration of the

1794: Alexander polynomial that follows next.

1795: 

1796: \section{Polynomial invariants}

1797: 

1798: \subsection{Alexander polynomial}

1799: 

1800: In \cite{Birman}, Birman proposed (but considered as very difficult)

1801: the problem to classify 3-braid

1802: links with given polynomials. In \cite{3br} we dealt with the skein

1803: polynomial. Now we can extend our results to the Alexander polynomial

1804: (with the convention in the beginning of \S\reference{ppp}).

1805: The following discussion gives a fairly exact description how to

1806: find the 3-braid links, if such exist, for any possible admissible

1807: (as specified in \S\reference{brD}) polynomial.

1808: 

1809: A solution for the Jones polynomial is presented afterwards.

1810: % (though not explained in detail)

1811: 

1812: \begin{lemma}

1813: If $\bt$ is strongly quasi-positive and fibered, then $\Mc\Dl=+1$.

1814: \end{lemma}

1815: 

1816: (Here it is essential to work with the leading, not trailing

1817: coefficient of $\Dl$ and with strongly quasi-positive links and

1818: not their mirror images.)

1819: 

1820: \proof We know $\md_vP=\Md_zP=1-\chi$. So $[P]_{z^{1-\chi}}$ has

1821: a term in degree $v^{1-\chi}$. The coefficient must be $+1$

1822: because of \eqref{p21} and because it is the only coefficient that

1823: contributes to the absolute term in $P(v,v^{-1}-v)$. Now from

1824: the classification of leading $z$-terms of $P$ in \cite{3br}

1825: it follows that $[P]_{z^{1-\chi}}$ can have at most one further term,

1826: with coefficient $\pm 1$. If such term exists, the substitution

1827: \eqref{DP} would either cancel the terms

1828: in degree $t^{(1-\chi)/2}$ in $\Dl(t)$, or give coefficient $\pm 2$,

1829: so our link cannot be fibered. Thus a second term does not exist,

1830: and the claim follows from \eqref{DP}. \qed

1831: 

1832: \begin{lemma}\label{ty}

1833: Let $\bt'=[(123)^k]$ be an even power $k$ of $[123]$.

1834: Assume $\bt$ has Xu normal form $R$, and after syllable

1835: reduction becomes $\bt'$, but $\bt\ne\bt'$ itself (i.e. some

1836: syllable in $\bt$ is non-trivial). Then $\hat\bt$ satisfies

1837: $\Md\Dl=-1-\chi$; moreover $\Mc\Dl>0$ and is equal to the 

1838: number of non-trivial syllables in $\bt$.

1839: \end{lemma}

1840: 

1841: \proof We use for fixed $k$ induction on the exponent sum.

1842: If exactly one syllable is non-trivial with exponent $2$,

1843: then applying the skein relation at the exponent-2

1844: syllable shows that $\bt=\bt_+$ inherits the Alexander

1845: polynomial of $\bt_-$ with positive sign (since $\bt_0=\bt'$,

1846: whose closure has zero polynomial). 

1847: Now $\bt_-$ is positive and has an index array with a subsequence

1848: of the form $xyx$, and so is fibered by theorem \reference{Thq}.

1849: Then by lemma \reference{ty}, $\Mc\Dl=+1$ and

1850: $\Md\Dl=1-\chi(\hat\bt_-)=-1-\chi(\hat\bt)$.

1851: 

1852: If $[\bt]-[(123)^k]>1$, then applying the skein relation

1853: at any non-trivial exponent $2$ syllable gives (with positive

1854: sign) the Alexander polynomials of two closed braids

1855: $\bt_0$ and $\bt_-$, former of which is fibered and

1856: latter of which has the requested property by induction.

1857: Then the maximal terms in degree $1-\chi(\hat\bt_-)=

1858: -\chi(\hat\bt_0)$ are positive and do not cancel.

1859: The one of $\bt_-$ is $+1$, while the one of $\bt_0$

1860: by induction one less than the number of non-trivial 

1861: syllables in $\bt$ (since in $\bt_0$ one more syllable

1862: becomes trivial).

1863: 

1864: If some syllable in $\bt$ has exponent $>2$ then $\bt_-$ is

1865: not minimal, and the degree and leading coefficient of $\Dl$

1866: are inherited (with positive sign) from $\bt_0$. \qed

1867: 

1868: \begin{lemma}\label{tz}

1869: Assume $\bt$ has Xu normal form $R$, which after syllable

1870: reduction becomes an odd power $k$ of $[123]$.

1871: Then $\Md\Dl=1-\chi$, and $\Mc\Dl=+2$.

1872: \end{lemma}

1873: 

1874: \proof First we prove the claim if $\bt=[(123)^k]$.

1875: We know that $\Md\Dl=1-\chi$ and $\Mc\Dl=\pm 2$,

1876: so we must exclude $\Mc\Dl=-2$. Applying the skein relation gives

1877: the polynomials of $\bt_0$ and $\bt_-$ with positive sign.

1878: $\bt_0$ is positive and fibered as before, so $\Mc\Dl=+1$.

1879: Then clearly $\bt_-$ (which is of the form $L^{-1}R$ and also

1880: fibered) cannot have $\Mc\Dl=-3$. 

1881: 

1882: If $\bt\ne [(123)^k]$, then it has a non-trivial syllable.

1883: Applying the skein relation at a letter in that syllable

1884: we find that $\bt_-$ reduces. So the leading term comes

1885: from $\bt_0$, and with positive sign. \qed

1886: 

1887: 

1888: 

1889: \begin{theorem}\label{THA}

1890: Fix some admissible Alexander polynomial $\Dl$. Then

1891: \def\labelenumi{\theenumi)}

1892: \begin{enumerate}

1893: \item If $\Dl=0$, then the 3-braid links with such polynomial

1894: are the split links and the closures of (incl.\ negative)

1895: even powers of $[123]$.

1896: \item If $\Dl\ne 0$, there are only finitely many 3-braid links

1897: with this $\Dl$. They all have $1-\chi=2\Md\Dl$ or $-1-\chi=2\Md\Dl$.

1898: In latter case they are up to mirroring strongly quasi-positive.

1899: \item If $\Mc\Dl\le -2$, then no 3-braid knot or 3-component

1900: link has such Alexander polynomial, and any 2-component link

1901: is strongly quasi-negative.

1902: \item If $\Mc\Dl\ge +2$, any 3-braid knot or 3-component link

1903: with such Alexander polynomial is strongly quasi-positive or a

1904: mirror image of it. Any 2-component link is strongly quasi-positive.

1905: \item If $|\Mc\Dl|>2$, then any 3-braid link 

1906: with such Alexander polynomial has $2\Md\Dl=-1-\chi$.

1907: \end{enumerate}

1908: \end{theorem}

1909: 

1910: \begin{defi}\label{DEFB}

1911: Let for $\bt\in B_3$ with $6\mid [\bt]$, the \em{Birman dual}

1912: $\bt^*$ be defined by $\bt^{-1}\dl^{2[\bt]/3}$, where

1913: $\dl=[121]$ and $\dl^2$ generates the center of $B_3$.

1914: \end{defi}

1915: 

1916: Birman \cite{Birman} shows that $\hat\bt$ and $\hat\bt^*$

1917: have the same skein polynomial. This observation relates to

1918: our explanation at a couple of places, for example, in remark

1919: \reference{rqw}, and also in the below arguments.

1920: 

1921: \proof[of theorem \reference{THA}]

1922: Let us exclude \em{a priori} trivial and split links.  The claims

1923: follow from the discussion of $2\Md\Dl$ and $\Mc\Dl$ in cases.

1924: 

1925: We know by theorem \reference{xLR} (or by \cite{3br,ntriv}, as noted

1926: before) that if $\bt\in B_3$ is not (up to mirroring) strongly

1927: quasi-positive, then $2\Md\Dl=1-\chi$ and $\Mc\Dl=\pm 1$.

1928: 

1929: It remains to deal with the Xu form $R$. The form $L$ is just the

1930: mirror image, and mirroring preserves the Alexander polynomial for

1931: knots and 3-component links and alters the sign for 2-component

1932: links.

1933: 

1934: If making trivial all syllables in $\bt$, the new word

1935: $\bt'$ is not a power of $[123]$, then we proved that $\hat\bt$

1936: is fibered, so again $2\Md\Dl=1-\chi$ and $\Mc\Dl=\pm 1$.

1937: 

1938: If $\bt'$ is an even power of $[123]$, then by Birman duality we

1939: conclude that $\Dl(\hat\bt')=0$. If $\bt\ne \bt'$, then one uses

1940: lemma \reference{ty}.

1941: % induction on the syllables exponents. If $\bt$ and $\bt'$

1942: % differ only by a letter, then applying the skein relation at

1943: % the exponent-2 syllable shows that $\bt=\bt_+$ inherits the Alexander

1944: % polynomial of $\bt_-$ (since $\bt_0=\bt'$). This has form $L^{-1}R$

1945: % and so $2\Md\Dl=1-\chi(\hat\bt_-)=-1-\chi(\hat\bt)$ and $\Mc\Dl=\pm 1$.

1946: % Then the degree and leading term are successively propagated.

1947: Thus $2\Md\Dl=-1-\chi$.

1948: % and also here $\Mc\Dl=\pm 1$.

1949: 

1950: If $\bt'$ is an odd power of $[123]$, then use the observation in

1951: lemma \reference{tz}  (or make one syllable to exponent $4$ and

1952: use Birman duality) to conclude that $2\Md\Dl=1-\chi$ and

1953: $\Mc\Dl=+2$. \qed

1954: 

1955: \begin{exam}

1956: In certain situations this theorem gives the most rapid test to

1957: exclude closed 3-braids. For example, the knot $13_{6149}$ has

1958: $MWF=3$, but seeing that $\Dl$ has $\Mc\Dl=-2$ we immediately

1959: conclude that it cannot be a closed 3-braid.

1960: \end{exam}

1961: 

1962: \begin{corr}

1963: There are only finitely many 3-braid links with given $\Md\Dl$

1964: (provided $\Dl\ne 0$). \qed

1965: \end{corr}

1966: 

1967: It is actually true (as we will prove below) that for links with

1968: \em{any} bounded braid index there are only finitely many different

1969: Alexander polynomials of given degree admitted. However, such

1970: Alexander polynomials may be admitted by infinitely many different

1971: links (of that braid index). See \S\reference{s4.2} for some remarks.

1972: % Using the non-faithfulness of

1973: % the Burau representation on $\ge 5$ strands, it should not

1974: % be too hard to find such families. The 4-strand case however

1975: % seems open.

1976: % 

1977: % \begin{question}

1978: % Are there only finitely many 4-braid links with given $\Dl\ne 0$

1979: % (or equivalently $\Md\Dl$)?

1980: % \end{question}

1981: 

1982: \begin{corr}\label{cbd}

1983: Non-split 3-braid links bound no disconnected Seifert surfaces (with

1984: no closed components). In particular there are no non-trivial 3-braid

1985: boundary links.

1986: \end{corr}

1987: 

1988: \proof For such links the Alexander polynomial is zero. The only

1989: non-split 3-braid links $L$ of zero polynomial are closures of

1990: $[(123)^k]$ for even $k$. These cases are easily ruled out by linking

1991: numbers. Any pair of components of $L$ has non-zero linking number.

1992: So a connected component of any Seifert surface $S$ of $L$ must have

1993: at least two boundary components of $L$, and if $S$ is disconnected,

1994: $L$ has at least 4 components, which is clearly not the case. \qed

1995: 

1996: The examples that falsify this claim for 4-braids are again

1997: easy: consider the (closure links of) words in $[1\pm 2-1]$ and

1998: $[2\pm 3-2]$.

1999: 

2000: \begin{corr}

2001: If $\Dl$ is the Alexander polynomial of a 3-braid link, then

2002: $|\Mc\Dl|\le \Md\Dl+2$. For a knot $|\Mc\Dl|\le\Md\Dl+1$.

2003: \end{corr}

2004: 

2005: \proof If $|\Mc\Dl|\le 2$ then we are easily done. 

2006: ($\Dl=\pm 2$ cannot occur for a knot.) Otherwise we have

2007: up to mirroring a strongly quasi-positive word $\bt$ reducing to

2008: an even power of $[123]$. Now $|\Mc\Dl|$ counts non-trivial syllables,

2009: so $[\bt]=3-\chi\ge 2|\Mc\Dl|$, and $-1-\chi=2\Md\Dl$. For a knot

2010: the number of syllables with exponent $\ne 2$ is at least $2$,

2011: so $[\bt]-2\ge 2|\Mc\Dl|$. \qed

2012: 

2013: Note in particular that the proof shows that $|\Mc\Dl|$

2014: can be any given natural number, and how to find the link

2015: that realizes this number. We emphasize this here, because

2016: later we will prove a contrary statement in the case of

2017: \em{alternating} links for \em{every} arbitrary braid index

2018: (see corollary \reference{cSG}).

2019: 

2020: \begin{corr}

2021: A 3-braid \em{knot} is fibered if and only if $\Mc\Dl=\pm 1$.

2022: \end{corr}

2023: 

2024: \proof It remains to explain why no 3-braid knot has $\Mc\Dl=\pm 1$

2025: but $\Md\Dl<1-\chi$. Latter condition would imply that we have

2026: up to mirroring a strongly quasi-positive word $\bt$ reducing to

2027: an even power of $[123]$, and former condition that $\bt$ has only one

2028: non-trivial syllable. But $[(123)^{2k}]$ has 3-component closure,

2029: and making one syllable non-trivial cannot give a knot. \qed

2030: 

2031: In particular, it is worth noting

2032: 

2033: \begin{corr}

2034: No non-trivial 3-braid knot has trivial Alexander polynomial. \qed

2035: \end{corr}

2036: 

2037: Again, the two 11 crossing knots immediately show that this is

2038: not true for 4-braids.

2039: 

2040: \begin{exam}\label{x29}

2041: We can also easily determine the 3-braid links for some small degree

2042: Alexander polynomials. For example, we see that no other 3-braid

2043: knot has the polynomial of $3_1$, $4_1$ or $5_2$. Similarly we can

2044: check that no 3-braid knot has the polynomial of $9_{42}$ and

2045: $9_{49}$ (which shows that these knots have braid index 4), a

2046: fact we will derive in the last section using entirely different

2047: representation theory arguments. 

2048: \end{exam}

2049: 

2050: \subsection{\label{Js}Jones polynomial}

2051: 

2052: The control of the Jones polynomial on 3-braid links was the

2053: object of main attention in \cite{Birman}. We can accomplish

2054: this with a similar argument to $\Dl$. The result we obtain

2055: can be conveniently described in our setting and is as follows:

2056: 

2057: \begin{theorem}\label{Hh}

2058: Let $L$ be a non-split 3-braid link, and $L=\hat\bt$ with $\bt\in

2059: B_3$. Then

2060: \begin{eqn}\label{g}

2061: \spn V(L)\,\le\,4-\chi(L)\,.

2062: \end{eqn}

2063: Equality holds if and only if $L$ is strongly quasi-signed (i.e.

2064: -positive or -negative, or equivalently $|[\bt]|=3-\chi(L)$), and

2065: not fibered. More specifically, the following holds:

2066: \def\labelenumi{\theenumi)}

2067: \begin{enumerate}

2068: \item If $L$ is strongly quasi-positive, then 

2069: $\ds\quad

2070: \md V\,=\,\frac{1-\chi}{2}\quad\mbox{and}\quad\mc V\,=\,\pm 1\,.

2071: $\\

2072: Analogously, if $L$ is strongly quasi-negative, then

2073: $\ds\quad

2074: \Md V\,=\,\frac{\chi-1}{2}\quad\mbox{and}\quad\Mc V\,=\,\pm 1$\,.

2075: \item If $L$ is strongly quasi-positive and fibered, then

2076: $\spn V(L)\,\le\,3-\chi(L)$\,.

2077: If $L$ is strongly quasi-positive and not fibered, then \eqref{g}

2078: is an equality and $\Mc V\,=\,\pm 1$. (The properties for strongly

2079: quasi-negative are analogous.)

2080: \item If $L$ is not strongly quasi-signed and $|[\bt]|<1-\chi(L)$,

2081: then 

2082: \[

2083: \mc V(L)\,=\,\pm 1\,, \quad\Mc V(L)\,=\,\pm 1\quad\mbox{and}\quad

2084: \spn V(L)\,=\,3-\chi(L)\,.

2085: \]

2086: \item If $L$ is not strongly quasi-signed and $[\bt]=1-\chi(L)$, then

2087: \[

2088: \Mc V(L)\,=\,\pm 1\quad\mbox{and}\quad

2089: \Md V(L)\,=\,\frac{5}{2}-\frac{3}{2}\chi(L)\,.

2090: \]

2091: Moreover, $\md V\ge \ds\frac{-1-\chi}{2}$, and if equality holds,

2092: then $\mc V=\pm 2$. (The case $[\bt]=\chi(L)-1$ is analogous.)

2093: \end{enumerate}

2094: \end{theorem}

2095: 

2096: Apart from solving Birman's problem how to determine 3-braids

2097: with given Jones polynomial, theorem \ref{Hh} easily implies that

2098: no non-trivial 3-braid link has trivial (i.e. unlink) polynomial. We

2099: defer the discussion of the non-triviality of the Jones polynomial

2100: to \cite{ntriv}, where we work in the much more general context of

2101: semiadequate links. In that paper we will show that semiadequate

2102: links have non-trivial Jones polynomial. This result in fact motivated

2103: theorem \reference{thq}, which then provides a different conclusion

2104: about the non-triviality of the polynomial. A further application

2105: will be the classification of the 3-braid links with unsharp

2106: Morton-Williams-Franks inequality (mentioned in remark \ref{rqw}).

2107: 

2108: \begin{corr}\label{fintg}

2109: For a given Jones polynomial $V$ (actually the pair

2110: $(\md V,\Md V)$ is enough), there are at most three

2111: values of $\chi(L)$ of a 3-braid link $L$ with $V(L)=V$.

2112: If $\md V\cdot \Md V\le 0$, then $\chi(L)$ is unique.

2113: \end{corr}

2114: 

2115: \proof The theorem shows that the value of $\chi$ is determined

2116: by one of $\Md V$, $\md V$ or $\spn V=\Md V-\md V$.

2117: In particular for a pair $(\md V,\Md V)$ there exist at most three

2118: values of $\chi$ of 3-braid links with a Jones polynomial realizing

2119: this pair. If $\md V\cdot \Md V\le 0$, then the options that a 3-braid

2120: $\bt$ with $V(\hat\bt)=V$ is strongly quasi-signed or almost

2121: quasi-signed are excluded (up to a few simple cases that can be

2122: checked directly), so $\chi(L)$ is determined (unambiguously) by

2123: $\spn V$. \qed

2124: 

2125: In particular, since 3-braid links of given $\chi$

2126: are only finitely many, we have

2127: 

2128: \begin{corr}\label{finth}

2129: There are only finitely many closed 3-braids with the same

2130: Jones polynomial, actually with the same pair $(\md V,\Md V)$. \qed

2131: \end{corr}

2132: 

2133: For example, one

2134: easily sees that no 3-braid knot has the polynomial of $9_{42}$.

2135: Similarly, no other 3-braid knot has the polynomial of the figure-8-knot

2136: (there is, however, a 4-braid knot with such polynomial, $11_{386}$).

2137: 

2138: That there are only finitely many closed 3-braids with the same

2139: skein polynomial was known from \cite{3br}. For the Jones polynomial

2140: one should note that infinite families were constructed

2141: by Traczyk \cite{Traczyk} if one allows polynomials up to units.

2142: Traczyk's examples show that for fibered

2143: (strongly quasi-)positive links $\spn V$ may remain the same

2144: while $\chi\to-\infty$, so one cannot expect a full (lower) 

2145: control on $\spn V$ from $\chi$. From these links, one obtains

2146: by connected sum infinite families with the same

2147: polynomial for 5-braids. (The status of 4-braids remains unclear.)

2148: Also Kanenobu \cite{Kanenobu2} constructed finite families of 3-braids

2149: of any arbitrary size, so that our result is the maximal possible.

2150: 

2151: A question that surfaces naturally with these remarks in mind is 

2152: 

2153: \begin{question}

2154: Does for 3-braid links the Jones polynomial (or Alexander polynomial)

2155: determine the skein polynomial? In other words, do any two

2156: 3-braid links with the same $V$ (or $\Dl$) have also equal $P$?

2157: \end{question}

2158: 

2159: We have at least the following partial result, whose proof we

2160: postpone after the proof of theorem \reference{Hh}.

2161: 

2162: \begin{corr}\label{CVP}

2163: A given Jones polynomial $V=V(L)$ is realized for 3-braid links

2164: $L$ by at most three different skein polynomials $P(L)$. If

2165: $\md V\cdot \Md V\le 0$, then $V(L)$ determines $P(L)$.

2166: \end{corr}

2167: 

2168: In the general case one cannot expect a positive answer to the

2169: above question. At least for links there is now the method

2170: of \cite{EKT} available, which should yield large families of

2171: links with the same Jones but different skein (or Alexander)

2172: polynomial. For constructing families with the same Alexander

2173: (but different Jones or skein) polynomial, further techniques

2174: are available, applicable also for knots, like the non-faithfulness

2175: of the Burau representation (see \cite{Bigelow}) or tangle surgeries

2176: (see \cite{Bleiler}).

2177: 

2178: % \begin{corr}\label{fintg}

2179: % For a given Jones polynomial $V$ (actually the pair

2180: % $(\md V,\Md V)$ is enough), there are at most three

2181: % values of $\chi(L)$ of a 3-braid link $L$ with $V(L)=V$.

2182: % If $\md V\cdot \Md V\le 0$, then $\chi(L)$ is unique.

2183: % \end{corr}

2184: % 

2185: % \proof The theorem shows that the value of $\chi$ is determined

2186: % by one of $\Md V$, $\md V$ or $\spn V=\Md V-\md V$.

2187: % 

2188: % In particular for a pair $(\md V,\Md V)$ there exist at most three

2189: % % values of $\chi$ of 3-braid links with a Jones polynomial realizing

2190: % this pair; such links are only finitely many. \qed

2191: 

2192: For the proof of theorem \ref{Hh} we use the previous work in

2193: \cite{3br} (that in

2194: particular answered Birman's question) on the skein polynomial. We

2195: apply again the result in \cite{3br} that $\Md_zP=1-\chi$ for closed

2196: 3-braids. As in that paper, we distinguish the cases of band-%

2197: positive, band-negative 3-braids, and such of Xu's form $L^{-1}R$. 

2198: 

2199: By the Morton-Williams-Franks inequalities \eqref{mq},

2200: a 3-braid $\bt$ of exponent sum (writhe) $[\bt]$ has $v$-degrees of

2201: $P$ in $[\bt]-2$ we call \em{left degree}, $[\bt]$ we call \em{middle

2202: degree} and $[\bt]+2$ we call \em{right degree}. The terms of

2203: $[P]_{z^k}$ for some $k$ in these degrees will be called left,

2204: middle and right terms\footnote{Note that here the brackets for

2205: polynomials and for braids have a completely different meaning.}. 

2206: 

2207: \begin{lemma}\label{lemg}\mbox{}\\

2208: (a) If $\bt$ is band-positive, then $\Mc_zP$ has $v$-terms in the left

2209: degree and possibly in the middle degree. If $\bt$ is band-negative,

2210: $\Mc_zP$ has $v$-terms in the right degree and possibly in the

2211: middle degree. In either situation a term in the middle degree occurs

2212: if and only if $\hat\bt$ is not fibered.\\

2213: (b) If $\bt$ is $L^{-1}R$

2214: then $\Mc_zP$ has $v$-terms in the middle degree only.\\

2215: In both cases all non-zero occurring coefficients are $\pm 1$.

2216: \end{lemma}

2217: 

2218: \proof In (a) we prove only the first claim (the second

2219: claim is analogous). Let $\chi=\chi(\hat\bt)$. Since

2220: $\md_lP\le \Md_zP=1-\chi$ and $1-\chi=[\bt]-2\ge 

2221: \md_lP$ by MWF, we have $\md_lP=\Md_zP$, and then \eqref{p21}

2222: implies that $\Mc_zP$ has a $v$-term in the left

2223: degree, with coefficient $\pm 1$. From \cite[Theorem 3]{3br} we

2224: have then that it has no right-degree term, and that if it has

2225: a middle-degree term, the coefficient is $\pm 1$. Now the

2226: previous work and the substitution $v=1$, $z=t^{1/2}-t^{-1/2}$

2227: for $\Dl$ in \eqref{DP} easily show that the middle term

2228: occurs if and only if $\hat\bt$ is not fibered.

2229: 

2230: Now consider (b). If $\Mc_zP$ has a left-degree term, then using

2231: the substitution \eqref{p21}, we saw that $\bt$ is band-positive.

2232: Otherwise, $[\bt]<3-\chi$, so the contribution of the coefficient

2233: of $z^{1-\chi}v^{[\bt]-2}$ in \eqref{p21}, which is not cancelled,

2234: is not in degree $0$. Analogously one argues if $\Mc_zP$ has a

2235: right-degree term. \qed

2236: 

2237: Note that keeping track of $[\bt]$ and distinguishing between

2238: left and middle term is important here. By remark \ref{rqw},

2239: we have links with equal polynomials such that the left

2240: term of the one is the middle term of the other.

2241: 

2242: \begin{lemma}\label{lemldg}

2243: (a) If $\bt$ is $L^{-1}R$ and $L$ has exponent sum $[L]>1$, then

2244: $[P]_{z^{-1-\chi}}$ has a left-degree term, and the coefficient

2245: is $\pm 1$. \\

2246: (b) If $[L]=1$ and

2247: $[P]_{z^{-1-\chi}}$ has a left-degree term, then the coefficient

2248: is $\pm 2$. \\

2249: The two analogous statements hold replacing $L$ by $R$

2250: and left-degree term by right-degree term.

2251: \end{lemma}

2252: 

2253: \proof We proved already that $\Mc_zP$ has only a middle-degree term.

2254: So it must remain under the substitution of \eqref{p21} if and only

2255: if $[\bt]=1-\chi$. The cases (a) and (b) occur when this term must

2256: be cancelled, and complemented to $1$, respectively. \qed

2257: 

2258: \proof[of theorem \reference{Hh}]

2259: The statements follow mainly by putting together the last two

2260: lemmas and looking in which degrees the

2261: non-cancelling contributions of the coefficients of $P$

2262: occur under the substitution \eqref{VPsub}.

2263: 

2264: If part (a) of lemma \reference{lemg} applies, we established already

2265: (in theorem \reference{thqp})

2266: that the positive band form of $\bt$ is equivalent to the strong

2267: quasi-positivity of $\hat\bt$. In this case $\md V$ comes from

2268: the left-degree term in $\Md_zP$. If $\hat\bt$ is not

2269: fibered, then $[P]_{z^{-1-\chi}}$ must have a right-degree

2270: term (with coefficient $\pm 1$) to cancel the middle-degree

2271: term of $\Md_zP$ under \eqref{p21}.

2272: 

2273: If part (b) of lemma \reference{lemg} applies, we use the

2274: further information of lemma \ref{lemldg}.

2275: In case (a) of lemma \ref{lemldg}, the left and right

2276: terms in $[P]_{z^{-1-\chi}}$ determine the degrees

2277: and edge coefficients of $V$.

2278: 

2279: In case (b) of lemma \ref{lemldg},

2280: the maximal term in $V$ comes from the right-degree term in

2281: $[P]_{z^{-1-\chi}}$. A coefficient in $t^{(-1-\chi)/2}$

2282: may come only from a left-degree term in $[P]_{z^{-1-\chi}}$,

2283: which, if occurring, is with coefficient $\pm 2$. \qed

2284: 

2285: This proof underscores the significance of \eqref{VPsub}

2286: as a tool for studying the Jones polynomial. So far it

2287: seems to have been useful just for calculating specific

2288: Jones polynomials from $P$. In \S\reference{Sfg} we will

2289: see further results that come out of considering this

2290: substitution.

2291: 

2292: We note the following equalities that follow from the proof of

2293: theorem \ref{Hh}. These will be needed in the study of the Q

2294: polynomial, and are also helpful for corollary \reference{CVP}.

2295: 

2296: \begin{lemma}\label{lemma_}

2297: If $\bt$ is a not strongly quasi-signed 3-braid of exponent sum $e$

2298: and $V$, $\chi$ the Jones polynomial resp. Euler characteristic

2299: of its closure, then

2300: \begin{eqnarray}

2301: \label{one_}\Md V-e & = & \frac{1-\chi}{2}+1 \\

2302: \label{two_}\md V-e & = & -\frac{1-\chi}{2}-1 

2303: \end{eqnarray}

2304: \em{Exceptions} are \eqref{two_} if $\bt$ is strongly

2305: almost quasi-positive, and \eqref{one_} if $\bt$ is strongly

2306: almost quasi-negative. \qed

2307: \end{lemma}

2308: 

2309: \proof[of corollary \reference{CVP}]

2310: Depending on whether a 3-braid representation $\bt$ is strongly

2311: quasi-signed, almost quasi-signed, or none of both, $V$ determines

2312: $e=[\bt]$ via $\chi(\hat\bt)$ and/or \eqref{one_} or \eqref{two_}.

2313: With $V$ and $e$, one can recover the trace of the Burau matrix

2314: $\psi_2(\bt)$, and from that also $P(\hat\bt)$ (see \cite{Birman} or

2315: \eqref{V_3} below). Again, if $\md V\cdot \Md V\le 0$, then the

2316: options that $\bt$ is strongly quasi-signed or almost quasi-signed

2317: are ruled out easily. \qed

2318: 

2319: \subsection{Q polynomial}

2320: 

2321: We extend the scope of the previous results to the

2322: Brandt-Lickorish-Millett-Ho polynomial. The main aim here is to prove

2323: 

2324: \begin{theo}\label{thQ}

2325: Only finitely many non-trivial 3-braid links have given $Q$

2326: polynomial, and none has trivial (i.e., unlink) polynomial.

2327: \end{theo}

2328: 

2329: Note that by Kanenobu's work \cite{Kanenobu3}, again

2330: there are finite families of arbitrary large size,

2331: so we claim again a sort of contrary result.

2332: 

2333: While this result may be considered less relevant

2334: that its analoga for $\Dl$ and $V$, its proof displays

2335: the largest variety of tools necessary to apply, and 

2336: shows how the various approaches to link invariants

2337: (skein relations, state models and representation theory), 

2338: which are often considered in isolation, can usefully 

2339: complement each other. Indeed, $Q$ seems in general 

2340: more difficult to treat than $V$. Apart from Kidwell's 

2341: results \cite{Kidwell} for alternating links, and those

2342: in \cite{pos} for positive \em{knots}, neither the 

2343: non-triviality nor the finiteness property seem to have 

2344: been known previously for any other class of links.

2345: 

2346: The proof makes use of the full extent of the study of 

2347: Xu's form. Various additional ingredients will be necessary.

2348: One such is J. Murakami's $Q$-$V$ formula, found in

2349: \cite{Murakami} (by representation theory, and proved also

2350: later by Kanenobu \cite{Kanenobu} using skein relations).

2351: It allows to recur in many cases the problem from $Q$ to $V$.

2352: When dealing with $V$, we use beside the previous discussion

2353: the (Kauffman bracket based) formulas of \S\ref{Ssaq}.

2354: 

2355: In cases Murakami's formula is not helpful, we apply the

2356: Polyak-Viro formula \eqref{pv2} for Casson's knot invariant

2357: $v_2=\myfrac{1}{2}\Dl''(1)$,

2358: and the following formula, (likewise skein theoretic and) due

2359: to Kanenobu \cite{Kanenobu4}, that relates $v_2$ to $Q$.

2360: 

2361: \begin{theo}(\cite{Kanenobu4})

2362: For a link $L$ with $n$ components $K_i$, $i=1,\dots,n$, we have

2363: \begin{eqn}\label{Kanf}

2364: Q_L'(-2)=3(-2)^n\sum_iv_2(K_i)+3(-2)^{n-2}\sum_{i<j}lk(K_i,K_j)^2

2365: +(n-1)(-2)^{n-3}\,,

2366: \end{eqn}

2367: where $lk$ is the linking number.

2368: \end{theo}

2369: 

2370: Using this formula, we prove first a special case of theorem

2371: \ref{thQ}.

2372: 

2373: \begin{prop}\label{pgw}

2374: Only finitely many (non-trivial) strongly quasi-positive 3-braid

2375: links have given $Q$ polynomial, and none has trivial polynomial.

2376: \end{prop}

2377: 

2378: For the case of knots, we have the following more specific statement,

2379: that has also some independent meaning.

2380: 

2381: \begin{prop}\label{pgq}

2382: If $K$ is a strongly quasi-positive 3-braid knot, then $v_2(K)\ge

2383: g(K)$.

2384: \end{prop}

2385: 

2386: Note that in \cite{pos} we proved the same inequality for a

2387: general \em{positive} knot. Since the pretzel knots with

2388: trivial Alexander polynomial are strongly quasi-positive, nothing

2389: like this holds for a general strongly quasi-positive knot, though.

2390: 

2391: We need some preparations. Let us first fix a convention. We number

2392: braid strands by 1,2,3 from left to right in the bottom of the

2393: braid by and compose words from bottom to top. We propagate strand

2394: number through the crossings. In particular strand( number)s may 

2395: appear permuted in the middle or on top of the braid. To refer to the 

2396: ordering of the strands with regard ot the local position in the braid

2397: diagram, we speak of left, middle and right strand. So, for example, 

2398: if strands 1 and 2 enter below a $\sg_1$ (strand 1 as left strand and

2399: strand 2 as middle strand), then they exit above in the order 2\ 1 

2400: (strand 2 as left strand and strand 1 as middle strand). 

2401: 

2402: Let for a (not necessarily pure, and not closed) braid word the

2403: linking numbers of strand $i$, $j$ numbered as explained above be

2404: the sum of the writhes of all crossings these two strands pass.

2405: (Note that for links this differs by the additional factor 1/2,

2406: which is not relevant, though.)

2407: 

2408: \begin{lemma}\label{lmlk}

2409: Let $\bt$ be a strongly quasi-positive 3-braid word

2410: (not necessarily pure, and not closed). Then the

2411: linking numbers $l_{ij}$ between pairs $(i,j)$ of 

2412: strands satisfy $l_{13}+l_{23}>0$, unless $\bt$ is split.

2413: \end{lemma}

2414: 

2415: \proof The sum of writhes of crossings a strand passes

2416: through each of $\sg_{1,2,3}$ is non-negative. So

2417: $l_{13}+l_{23}\ge 0$. If $l_{13}+l_{23}=0$, then

2418: strand 3 passes only from below as left or middle

2419: strand into a $\sg_3=\sg_1^{-1}\sg_2\sg_1$. But since it

2420: starts (at the bottom of the braid) as a rightmost 

2421: strand, this means that is passes no crossing, so $\bt$ is split. 

2422: \qed

2423: 

2424: We prove proposition \ref{pgq}

2425: by induction on the length of the positive word

2426: in $\sg_{1,2,3}$. Assume $\bt$ is written as such a word.

2427: We can w.l.o.g. cyclically permute the indices. 

2428: 

2429: \begin{lemma}

2430: If $\hat\bt$ is a strongly quasi-positive 3-braid, then a positive 

2431: word $\bt$ can be chosen so that it contains a non-trivial syllable,

2432: \em{unless} $\bt$ is a power of $[123]$. In particular, it is 

2433: always possible if $\hat\bt$ is a knot, or a fibered link.

2434: \end{lemma}

2435: 

2436: \proof Choose Xu's form. It is $[12]^k\bt'$, where $k\ge 0$ and $\bt'$ 

2437: has cyclically non-decreasing indices. This word $\bt$ has a non-trivial

2438: syllable up to cyclic permutations of the letters, unless $\bt'$

2439: is a power of $[123]$. In this case, if $k>0$, we apply a YB relation 

2440: at the initial $121$ in $\bt$, and are done. \qed

2441: 

2442: \proof[of proposition \ref{pgq}] Let $\hat\bt$ be a knot. 

2443: We can w.l.o.g. now assume $\bt$ is written as a word with a 

2444: non-trivial syllable $s$. Now we apply the skein relation

2445: of $\Dl$ at a letter of $s$. Let $K=K,K_-,K_0$ be the skein

2446: triple, and $\bt_{\pm,0}$ the corresponding braids. Also we can 

2447: cyclically permute the indices in $\bt_{\pm,0}$. We choose them so 

2448: that the strand fixed by the permutation of $\bt_0$ is number 3. We 

2449: know from the skein relation of $\Dl$ 

2450: that $v_2(K)-v_2(K_-)=lk(L,M)$ where $L,M$ are the components 

2451: of $K_0$. Now that linking number is positive by lemma \ref{lmlk} 

2452: (unless $\bt_0$ is a split braid, but this is impossible because $s$

2453: is non-trivial). By induction on $g(K)$, the claim of proposition

2454: \ref{pgq} follows. \qed

2455: 

2456: 

2457: 

2458: \proof[of proposition \reference{pgw}] We distinguish three

2459: cases depending on the number of components of $\hat\bt$.

2460: 

2461: \begin{caselist}

2462: \case 1 component. In this case $Q'(-2)$, by Kanenobu's formula

2463: \eqref{Kanf},

2464: is a multiple of Casson's invariant $v_2$. The claim we wish to 

2465: show then is implied by the estimate in proposition \reference{pgq}.

2466: 

2467: \case 2 components. In this case Kanenobu's formula 

2468: involves only the square of the linking number of both components 

2469: $K,L$ of the closure link of $\bt$ and $v_2$ of the 2-string subbraid 

2470: component $K$. Looking at Xu's normal form, we see that if $\bt$ is 

2471: a strongly quasi-positive 3-braid, then writing $\bt$ as a word

2472: in $\sg_{1,2,3}$, and then expanding $\sg_3=\sg_1^{-1}\sg_2\sg_1$,

2473: we obtain a word in $\sg_{1,2}$ of some length $c$ with at most 

2474: $c/5$ negative letters. Since thus $\bt$ has at most $c/5$ negative

2475: crossings, it is easy to see that not both $v_2(K)$ and $lk(K,L)$

2476: can be zero, and one grows unboundedly when $c$ grows. So we are done.

2477: 

2478: \case 3 components. In this case Kanenobu's formula \eqref{Kanf} 

2479: reduces to the square sum of the linking number of pairs components.

2480: Let $l_{12},l_{13},l_{23}$ be these linking numbers. Again it is 

2481: clear that not all of $l_{12},l_{13},l_{23}=0$, and that when 

2482: $l_{12}+l_{13}+l_{23}\ge 2c/5$ grows unboundedly, then so will

2483: $l_{12}^2+l_{13}^2+l_{23}^2$, so again we are done.

2484: \end{caselist}

2485: \qed

2486: 

2487: Now we move on to settle the remaining cases in theorem \ref{thQ}.

2488: Here the tools differ considerably. Beside the study of Xu's form,

2489: we need the help of the following theorem, proved in \cite{ntriv}.

2490: 

2491: \begin{theorem}\label{thq}

2492: Any $3$-braid is up to conjugacy $A$-adequate or $B$-adequate.

2493: \end{theorem}

2494: 

2495: We proved also the following

2496: 

2497: \begin{theorem}

2498: A $3$-braid word is $B$-adequate if and only if it is (a) a

2499: negative word, or (b) it contains no $[121]$ as subword, and negative

2500: entries in the Schreier vector appear isolated. (That is, in

2501: cyclic order they are preceded and followed by a positive entry.)

2502: \end{theorem}

2503: 

2504: Using this criterion we prove now

2505: 

2506: \begin{lemma}\label{oia}

2507: Assume $\bt$ is strongly $\le 2$-almost positive and of exponent 

2508: sum $e\ge 0$. Then $\bt$ is $B$-adequate.

2509: \end{lemma}

2510: 

2511: \proof

2512: If $\bt$ is strongly quasi-positive or almost strongly

2513: quasi-positive, then it is of Xu's form $L^{-1}R$

2514: where $L$ is a single letter. By direct observation

2515: we can verify that when permuting indices cyclicly

2516: properly and writing out $\sg_3$ in $\sg_{1,2}$, then

2517: the resulting word has no $[121]$ subword, and negative

2518: entries in the Schreier vector appear isolated. (In case

2519: of $[(312)^k31-2]$, we must cancel 2 crossings first.)

2520: 

2521: If $\bt$ is 2-almost strongly quasi-positive, then then

2522: it is of Xu's form $L^{-1}R$ where $L$ has 2 letters.

2523: We can assume that $L=[12]$ or $L=[1^2]$, then we write down

2524: the subword of $L^{-1}R$ consisting of the first and last

2525: letters of $R$ and $L$ (herein $k\ge 1$):

2526: \[

2527: \begin{array}{ll}

2528: [[23] \dots 33-2-1] & [[23] \dots 22-1-1] \\{}

2529: [[23] \dots 23-2-1] & [[23] \dots 31^k2-1-1] \\{}

2530: [[23] \dots 31^k-2-1] & [[23] \dots 33-1-1] \\{}

2531: & [[23] \dots 23-1-1]

2532: \end{array}

2533: \]

2534: (Here `[23]' means as in \S\ref{brs} that the word is to begin with

2535: `2' or `3'.) The reductions and $B$-adequacy test are done

2536: case-by-case. \qed

2537: 

2538: Using the formulas in \S\reference{Ssaq}, we prove

2539: 

2540: \begin{lemma}\label{yu}

2541: If $L=\hat\bt$ is a link, which is the closure of an $A$-semiadequate 

2542: $3$-braid $\bt$, and $V_0V_1=-1$, then $V_0V_2=1$ or $2$.

2543: \end{lemma}

2544: 

2545: \proof By \cite{ntriv} a A-semiadequate $3$ braid is either

2546: positive, in which case $V_1=0$, or has no $[-1-2-1]$ and

2547: positive entries are isolated in the Schreier vector.

2548: In \cite{ntriv} we described the words  for latter

2549: type which have $V_1=0$. They are (up to braid relations

2550: and cyclic letter permutations) of the form%

2551: \footnote{Note that these are exactly the 3-braids which are

2552: reducible in the terminology of dynamic properties;

2553: this reducibility has nothing to do, though, with the

2554: reducibility in Markov's theorem.}

2555: \[

2556: [1^{-2}2^{-2}1^{-2}\dots 1^{-2}2^{-1}1^p2^{-1}]\,.

2557: \]

2558: 

2559: Analogously to that study, we can see that if $V_0V_1=-1$,

2560: then $\bt$ has a word which is obtained from the above type

2561: by replacing some $1^{-2}$ by $1^{-k}$ 

2562: for $k\ge 3$. The intertwining graph of

2563: the $A$-state is a path (all vertices have valence 2, except two

2564: of valence 1), so $\chi(IG)=1$. We have $\bigtriangleup

2565: =1$ if $k=3$, and otherwise $\bigtriangleup=0$. So by

2566: theorem \reference{T02}, we have $V_0V_2=1$ for

2567: $k=3$, and $V_0V_2=2$ otherwise. \qed

2568: 

2569: To explain what such a property has to do with $Q$, 

2570: now we must introduce J. Murakami's formula \cite{Murakami}.

2571: (See also Kanenobu \cite[Theorem 2]{Kanenobu}.)

2572: 

2573: Let $i=\sqrt{-1}$, $u=\sqrt{-t}$ and $x=u+u^{-1}$. Let further%

2574: \footnote{Note that $V$ is here what is written as $J$, and not $V$, 

2575: in \cite{Kanenobu}.} for a braid $\bt$ of exponent sum $e$,

2576: \begin{eqn}\label{chi}

2577: \chi(\bt,t)=i^eu^{-2e}V_{\hat\bt}(t)+u^{-e}(x^2-2)\,.

2578: \end{eqn}

2579: 

2580: Then Murakami's formula is

2581: 

2582: \begin{theo}(J. Murakami)\label{TMF}

2583: If $L$ is a closure of a 3-braid $\bt$ of exponent sum $e$,

2584: then

2585: \begin{eqn}\label{sonjo}

2586: Q(L,x)\,=\,\chi(\bt,\sqrt{t})^2-1\,+\,\frac{2(x^2+x-1)}{x^2(x^2-3)}

2587: (u^e+e^{-e})+\frac{-x^4-2x^3+3x^2+4x-4}{x^2(x^2-3)}\chi(\bt,t)\,.

2588: \end{eqn}

2589: \end{theo}

2590: 

2591: \proof[of theorem \ref{thQ}] We can w.l.o.g. (taking the mirror image)

2592: assume that $e\ge 0$, and (excluding trivial to check special cases)

2593: that $\chi\le 0$. Clearing denominators and absolute terms in

2594: \eqref{sonjo}, we find

2595: \begin{eqnarray}

2596: \label{Q1} Q_1(L,x):=x^2(x^2-3)(Q(L,x)+1) & = &

2597: \bigl[ i^e(-t)^{-e}V(t)+(-t)^{-e/2}(-t-\frac{1}{t}) \bigr]

2598: \cdot (-x^4-2x^3+3x^2+4x-4) \\

2599: \nonumber & &  + \bigl[ i^e(-t)^{-e/2}V(-\sqrt{-t})+(-t)^{-e/4}x \bigr]^2

2600: \cdot x^2(x^2-3) \\

2601: \nonumber & & +2(x^2+x-1)(\sqrt{-t}^e+\sqrt{-t}^{-e})\,.

2602: \end{eqnarray}

2603: 

2604: Rearranging, we need to show that the sum of the following 5

2605: expressions,

2606: regarded as a polynomial in $t^{\pm 1/2}$, has either arbitrarily

2607: small minimal degree or arbitrarily large maximal degree.

2608: (Note that by \eqref{sonjo} this sum must be self-conjugate in 

2609: $t$ up to coefficient signs, which is not at all evident directly.)

2610: 

2611: \begin{eqnarray*}

2612: T'_1 & = & i^e(-t)^{-e}V(t)\cdot (-x^4-2x^3+3x^2+4x-4) \\

2613: T'_2 & = & t^{-e}\bigl[V(-\sqrt{-t})\bigr]^2\cdot x^2(x^2-3) \\

2614: T'_3 & = & 2i^e(-t)^{-3e/4}V(-\sqrt{-t})x^3(x^2-3) \\

2615: T'_4 & = & (-t)^{-e/2}\bigl[(x^2-2)(-x^4-2x^3+3x^2+4x-4)+x^4(x^2-3)+2(x^2+x-1)

2616:  \bigr] \\

2617: T'_5 & = & (-t)^{e/2}\bigl[2(x^2+x-1) \bigr] 

2618: \end{eqnarray*}

2619: 

2620: Switching $-t\to t$ to simplify the expressions, and regrouping terms, 

2621: we have

2622: \begin{eqnarray*}

2623: T_1 & = & i^et^{-e}V(-t)\cdot (-1\ -2\ -1\ -2\ [-4]\ -2\ -1\ -2\ -1) \\

2624: T_2 & = & (-t)^{-e}\bigl[V(-\sqrt{t})\bigr]^2\cdot (1\ 0\ 1\ 0\ [0]\ 0\ 1\ 0\ 1) \\

2625: T_3 & = & 2i^et^{-3e/4}V(-\sqrt{t})\cdot (1\ 0\ 2\ 0\ 1\ [0]\ 1\ 0\ 2\ 0\ 1) \\

2626: T_4 & = & t^{-e/2}\bigl[-2(x-1)(x^2-3)(x^2-1) \bigr] \\

2627: T_5 & = & t^{e/2}\bigl[2(x^2+x-1) \bigr] 

2628: \end{eqnarray*}

2629: Now $u=t^{1/2}$ and $x=t^{1/2}+t^{-1/2}$. The last factors for $T_

2630: {1,2,3}$ are given as a list of coefficients, with the absolute term

2631: put in brackets. Note that these are polynomials in $\sqrt{t}$, so for

2632: example, the first two polynomials have degree $2$ in $t$.

2633: 

2634: Since we are concerned with cancellations of the leading and trailing

2635: coefficients of $T_k$, let us compile their minimal $m_k=\md_tT_k$ and

2636: maximal degrees $M_k=\Md_tT_k$ (note that by taking degrees w.r.t.

2637: $t$, and not $\sqrt{t}$, the $m_k$, $M_k$ are only half-integers).

2638: \[

2639: \begin{array}{c|c|c}

2640: \raisebox{-1em}{\ry{2.3em}} k & m_k & M_k \\

2641: \hline

2642: \raisebox{-1em}{\ry{2.5em}}1 & \md V-e-2 & \Md V-e+2 \\[2mm]

2643: 2 & \md V-e-2 & \Md V-e+2 \\[2mm]

2644: 3 & -\ffrac{3e}{4}+\ffrac{\md V}{2}-\ffrac{5}{2} & -\ffrac{3e}{4}+

2645:     \ffrac{\Md V}{2}+\ffrac{5}{2}\\[2mm]

2646: 4 & -\frac{e}{2}-\frac{5}{2} & -\frac{e}{2}+\frac{5}{2} \\[2mm]

2647: 5 & \frac{e}{2}-1 & \frac{e}{2}+1\raisebox{-1em}{\ry{1.7em}}

2648: \end{array}

2649: \]

2650: (Here again $\md V=\md_t V(t)$ is a half-integer, and similarly

2651: $\Md V$.)

2652: 

2653: Our aim will be to determine what of the $M_k$ is the largest, and 

2654: to show that the coefficients of the contributing $T_k$ do not cancel.

2655: An important case where we exclude (problematic) cancellation is

2656: 

2657: \begin{lemma}\label{lmcan}

2658: Assume that 

2659: \begin{eqnarray}

2660: \label{GG} M_1=M_2 & \ge & M_k+3/2\quad\mbox{ and } \\

2661: \label{gg} m_1=m_2 & \le & m_k-3/2

2662: \end{eqnarray}

2663: for $k=3,4,5$.

2664: Then $\Md_t Q_1(L,x)\ge M_1-1$. (Here $Q_1(L,x)$ refers to \eqref{Q1}.)

2665: If $L$ the closure of a 3-braid which is $A$-adequate resp. $B$-adequate

2666: then assuming \eqref{gg} resp.\ \eqref{GG} alone is sufficient.

2667: \end{lemma}

2668: 

2669: \proof By theorem \ref{thq}, $L$ is the closure of a 3-braid $\bt$ which 

2670: is either $A$-adequate or $B$-adequate. We consider the $B$-adequate case 

2671: and show that $M_1=M_2\ge M_k+3/2$ implies $\Md_t Q_1(L,x)\ge M_1-1$.

2672: If $\bt$ is $A$-adequate, we obtain similarly $\md_t Q_1(L,x)\le 

2673: m_1+1$, and the result follows by the (anti)symmetry of $Q_1$.

2674: 

2675: Now if $M_1=M_2\ge M_k+3/2$, the last three coefficients of $M_{1,2}$

2676: are not cancelled from other $M_k$ (remember these are polynomials

2677: in $\sqrt{t}$, and coefficients are meant for such polynomials), 

2678: so it is enough to show that they do not (completely) cancel 

2679: among each other. Now a look at the formulas for $T_{1,2}$ shows

2680: that if these three coefficients are all to cancel, then (apart from

2681: proper sign coincidences) we must have $\bar V_1\bar V_0=-1$ and 

2682: $\bar V_2=0$. (Take also into account the switch $-t\to t$.)

2683: However, this situation was ruled out by lemma \ref{yu}. \qed

2684: 

2685: We assumed $e\ge 0$; we also excluded the case $e=3-\chi$ of

2686: strongly quasi-positive braids. To apply the lemma, we need to

2687: establish \eqref{gg} and/or \eqref{GG}. For this we use now

2688: lemma \reference{lemma_}.

2689: 

2690: Clearly, a strongly almost quasi-negative braid should not 

2691: be considered for $e\ge 0$; the almost quasi-positive braid

2692: require a small extra argument, which is given below. Now

2693: \begin{eqn}\label{hj}

2694: \Md V-e\,=\,\frac{1-\chi}{2}+1\,\ge\,\frac{e}{2}+1\,,

2695: \end{eqn}

2696: so $M_3\ge M_4$, and we can discard $M_4$. Also from \eqref{hj}

2697: we obtain

2698: \[

2699: M_1-M_5-1=\Md V-\frac{3e}{2}\ge 1\,,

2700: \]

2701: so $M_1>M_5+1$, and can neglect $M_5$. Similarly, if

2702: $\bt$ is not almost strongly quasi-positive, we have with 

2703: \eqref{two_} and $\md V<e/2$ that

2704: \begin{eqn}\label{mmm}

2705: m_3< m_4<m_5\,,

2706: \end{eqn}

2707: so $m_{4,5}$ are also irrelevant.

2708: 

2709: So we need to deal only with $m_k,M_k$ for $k\le 3$. Moreover, we 

2710: see from \eqref{one_} that $M_1\ge (1-\chi)/2+3$. So whenever we can 

2711: apply lemma \ref{lmcan}, we have for $\chi\le 0$ that $\Md Q_1\ge 

2712: M_1-1>2$, so $\Md_z Q>0$, and $Q$ is not an unlink 

2713: polynomial. Moreover, we have for a sequence of links

2714: from $\chi\to -\infty$ also $\Md M_1\to \infty$, so

2715: $\Md Q\to \infty$, and a given (even just degree of a)

2716: $Q$ polynomial occurs only finitely many times, as desired.

2717: A similar argument applies if we use \eqref{two_}.

2718: 

2719: To apply lemma \ref{lmcan}, we need to check $M_3<M_1-1$ and $m_3>m_1+

2720: 1$. Note that if $\bt$ is almost strongly quasi-positive, then \eqref

2721: {two_} may not hold. So may not be able to apply lemma \ref{lmcan}

2722: directly. However, we remedied this by showing in lemma \ref{oia}

2723: that $\bt$ is $B$-adequate. Then \eqref{one_} holds, and it is enough

2724: to use $M_3<M_1-1$ (the condition \eqref{mmm} becomes also irrelevant).

2725: 

2726: Now using \eqref{one_} and \eqref{two_}

2727: \begin{eqnarray}

2728: \nonumber

2729: m_1-m_3+1 & = & \md V -e -2 +\frac{3e}{4}-\frac{\md V}{2}+\frac{5}{2}+1

2730: \,=\,\frac{\md V}{2}-\frac{e}{4}+\frac{3}{2} \\

2731: \nonumber

2732: & = & \frac{e}{2}-\frac{1-\chi}{4}-\frac{1}{2}-\frac{e}{4}+\frac{3}{2}\,=\,

2733: \frac{e}{4}-\frac{1-\chi}{4}+1\,, \\

2734: \label{iua}

2735: M_1-M_3-1 & = & \frac{\Md V}{2}-\frac{e}{4}-\frac{3}{2} \\

2736: \nonumber

2737: & = & \frac{e}{2}+\frac{1-\chi}{4}+\frac{1}{2}-\frac{e}{4}-\frac{3}{2}\,=\,

2738:       \frac{1-\chi}{4}+\frac{e}{4}-1\,.

2739: \end{eqnarray}

2740: 

2741: Then \eqref{GG} becomes equivalent to

2742: \[

2743: \frac{1-\chi}{4}+\frac{e}{4}>1\,.

2744: \]

2745: So, assuming $\chi\le 0$ and $e\le 1-\chi$, for \eqref{GG} 

2746: we are left to consider only $(e,\chi)=(0,-1)$, $(0,-3)$,

2747: $(1,0)$, $(1,-2)$, or $(2,-1)$, which are trivial cases to check. 

2748: 

2749: For \eqref{gg} we must deal with $A$-adequate but not $B$-adequate

2750: braids with $e\ge -3-\chi$, i.e. $\le 3$-almost strongly

2751: quasi-positive braids. For $\le 2$-almost strongly

2752: quasi-positive braids we can use lemma \reference{oia}.

2753: While one may extend the argument there to $3$-almost strongly

2754: quasi-positive braids, it would require a longer

2755: case-by-case analysis. We provide instead a different argument.

2756: 

2757: We try to modify the argument for proposition \reference{pgw}.

2758: In case $\hat\bt$ is a link of 2 or 3 components, we can argue again

2759: as there, and are left with the cases $e=0,1$ of $6$ and $7$ bands,

2760: which are easy to deal with.

2761: 

2762: So assume $\hat\bt$ is a knot.

2763: Now from \eqref{iua}, we have $m_3=m_1+1$, so if still some of the

2764: first two coefficients of $T_1$, $T_2$ do not cancel, we have

2765: $\md_t Q_1\le m_1+\myfrac{1}{2}$, and are done.

2766: Otherwise again $V_1V_0=-1$. So by the proof of lemma \reference{yu},

2767: \[

2768: \bt=[1^{-k}2^{-2}1^{-2}\dots 1^{-2}2^{-1}1^l2^{-1}]\,,

2769: \]

2770: with $k\ge 3$ and $l\ge 1$. Since $\hat\bt$ is a knot, $k$ and $l$

2771: are easily observed to be odd. Now a $\le 3$-almost positive

2772: word in Xu's form can be chosen, when written out as a word in

2773: $\sg_{1,2}$ of $c$ letters, to have $c_-\le c/5+3$ negative

2774: letters/crossings. If we reduce the word by cancelling

2775: crossings, this inequality remains true.

2776: 

2777: We estimate the Casson invariant $v_2$ of $\hat\bt$ using the Gau\ss{}

2778: diagram formula of Polyak-Viro \eqref{pv2}. Let us put the

2779: basepoint right after the group of $l$ positive crossings (which

2780: of either strands is immaterial). 

2781: 

2782: Now $l$ is odd and the $l$ positive crossings are pairwise linked,

2783: so their contribution to the Gau\ss{} diagram sum \eqref{pv2} is,

2784: independently on the location of the basepoint, equal to $v_2(T_{2,l})

2785: =\ffrac{l^2-1}{8}$, where $T_{2,l}$ is the $(2,l)$-torus knot. The

2786: analogous claim is true for the group of $k$ negative crossings.

2787: 

2788: If some of the positive crossings is linked with a negative crossing,

2789: it must be a negative crossing of a $\sg_2^{-1}$. Now by our choice

2790: of basepoint, in

2791: each syllable $\sg_2^{-2}$ only one of the two crossings gives a pair

2792: that contributes to \eqref{pv2}, and the contribution is always

2793: $-1/2$. Putting this together, we have

2794: \begin{eqn}\label{oiz}

2795: v_2\,\ge\,\frac{l^2-1}{8}+\frac{k^2-1}{8}-\frac{c_--k+2}{4}\cdot l

2796: \cdot \frac{1}{2}\,.

2797: \end{eqn}

2798: Assuming $c\ge 10$ (the other cases are checked directly), we have

2799: \[

2800: l\,\ge\frac{4c}{5}-3\,\ge\,\frac{c}{5}+3\,\ge\,c_-=c-l\,.

2801: \]

2802: So the r.h.s. in \eqref{oiz} is minimized for $l\,\ge\ffrac{4c}{5}-3$

2803: when putting $l=\ffrac{4c}{5}-3$. Then we have, with $k\ge 3$ and

2804: $c_-=c-l$, the estimate

2805: \[

2806: v_2\,\ge\,\frac{3c^2}{50}-\frac{29c}{40}+\frac{11}{4}\,,

2807: \]

2808: which is positive for $c\ge 10$, and grows when $c\to \infty$. 

2809: 

2810: With this theorem \ref{thQ} is proved. \qed

2811: 

2812: \begin{rem}

2813: It is interesting whether $\Md Q\to\infty$ also for the strongly 

2814: quasi-positive braids, but in Murakami's formula massive cancellations 

2815: become possible and $Q_1$ cannot be easily controlled.

2816: At least one can prove using theorem \ref{thq} and

2817: some results of Thistlethwaite in \cite{Thistle} that

2818: there are only finitely many 3-braid links with given

2819: $\Md_zF$, where $F$ is the Kauffman polynomial.

2820: \end{rem}

2821: 

2822: \section{Positivity of 3-braid links}

2823: 

2824: \subsection{Positive braid links\label{S2.2}}

2825: 

2826: \subsubsection{The Morton-Williams-Franks bound}

2827: 

2828: For the proof of theorem \reference{thg} we will need to study

2829: the behaviour of the bound \eqref{mwfb} in the Morton-Williams-Franks

2830: inequality (we abbreviate as MWF) on positive

2831: braids. This was begun by Nakamura \cite{Nakamura},

2832: who settled the case $MWF=2$ in the suggestive way: such braids

2833: represent only the $(2,n)$-torus links. (The case $MWF=1$ is

2834: trivial.) We will introduce a method that considerably simplifies

2835: his proof (but still makes use of some of his ideas), and

2836: then go on to deal with $MWF=3$. The example of non-sharp

2837: MWF inequality, $13_{9365}$ in \cite{KnotScape} (the connected

2838: 2-cable of the trefoil), given in \cite{MorSho}, is in fact

2839: only among a small family of exceptional cases.

2840: 

2841: \begin{theorem}\label{thmwf}

2842: If $b(\hat\bt)>MWF(\hat\bt)=3$, then $\bt$ reduces to a 4-braid,

2843: and is given by one of the following forms (assuming that $3*$

2844: denotes a sequence of at least one letter $3$, and $11*$

2845: resp.\ $22*$ sequences of at least two letters $1$ or $2$):\\

2846: \[

2847: [22*3*122*11*23211*], \quad\mbox{and}\quad [22*3122*11*23*211*].

2848: \]

2849: \end{theorem}

2850: 

2851: 

2852: We will reduce the proof to a finite number of words to

2853: check, which is done by calculation using the program of

2854: \cite{MorSho}. Since a direct computation is more reliable

2855: than an increasingly difficult mathematical argument, we have

2856: not tried to minimize the calculation by all means. However,

2857: we point out that for the sake of theorem \reference{thg}

2858: alone (rather than its refinement, theorem \reference{thmwf}),

2859: the following weaker statement is sufficient, for which

2860: a considerable part of the case-by-case calculations can

2861: be dropped. This corollary requires the notion of

2862: semiadequacy \cite{LickThis}, and can be deduced from theorem

2863: \reference{thmwf} by direct check of the exceptional words.

2864: (We will sometimes write $MWF(\bt)$ for $MWF(\hat\bt)$.)

2865: 

2866: \begin{corr}\label{cort}

2867: If $\bt$ is a positive braid word, and $MWF(\bt)=3$, then $\bt$

2868: reduces (up to Markov equivalence) to a positive 4-braid word $\bt'$,

2869: and the diagram $\hat\bt'$ is not $B$-adequate.

2870: \end{corr}

2871: 

2872: Here the notions of $A$-adequate and $B$-adequate for diagrams

2873: and braids are as explained in \S\ref{brD} and \S\ref{brt}. We

2874: note, as a consequence of \cite{Thistle}, that a braid is ($A$/%

2875: $B$-)adequate if and only if some, or equivalently any,

2876: minimal length word of a braid in its conjugacy class it is so.

2877: 

2878: We will thus prove theorem \reference{thg} only using corollary

2879: \reference{cort}, and indicate in the proof of theorem \reference{thmwf}

2880: the point where the corollary follows (and the rest of the argument

2881: is not needed). The argument that elegantly replaces the remaining

2882: case-by-case checks requires theorem \reference{thq}. Since (by

2883: taking again the full extent of our proof) theorem \reference{thq}

2884: is nonetheless not indispensable, we permit ourselves to defer its

2885: proof to a separate paper \cite{ntriv}.

2886: 

2887: 

2888: % \proof It is easy to see that each positive irreducible $\bt\in B_4$

2889: % with a $\sg_2$ can be written as $\sg_2^{k_1}\sg_1^{l_1}\sg_3^{m_1}\dots

2890: % \sg_2^{k_n}\sg_1^{l_n}\sg_3^{m_n}$, where $m_i>0$ and $l_i+m_i>0$.

2891: % 

2892: % Now if $\bt$ contains no YB relation, then all if $l_i+m_i>1$,

2893: % and $n>1$ or $l_1,m_1>1$.

2894: % 

2895: % Now if all syllables of $\sg_2$ are non-trivial, we can remove

2896: % in the positive resuloution tree and have the split union of

2897: % two non-trivial (because $\bt$ irreducible) $(2,k)$-torus links,

2898: % so $MWF=4$$. If ....

2899: % \proof Check directly. ($B$-adequacy is invariant under regular isotopy,

2900: % so any other 4-braid word giving the same knot is not $B$-adequate

2901: % either.) \qed

2902: 

2903: \proof[of theorem \reference{thg}]

2904: If $MWF(\bt)\le 2$, then we are done. So assume $MWF(\bt)=3$.

2905: Assume first $L$ has an $A$-adequate 3-braid. Since $A$-adequate

2906: diagrams minimize the number of negative crossings, and $L$ has

2907: a positive (braid) diagram, the $A$-adequate 3-braid diagram is

2908: positive, and we are done. So let $L$ have a $B$-adequate 3-braid

2909: diagram $\hat\bt'$. Now by corollary \reference{cort}, we find

2910: that $L$ reduces to a positive 4-braid $\bt$

2911: with $c_+(\bt)$ positive crossings, and $\bt$ is not $B$-adequate.

2912: Since $\bt'$ is $B$-adequate and $B$-adequate diagrams

2913: minimize the number of positive crossings, $c_+(\bt')<c_+(\bt)$.

2914: On the other hand, $MWF(L)=3$ and the inequalities of MWF for the

2915: $v$-degree of $P$ show that $\bt'$ must have exponent sum

2916: $[\bt']=c_+(\bt)-1$. Since $[\bt']\le c_+(\bt')$, we must have

2917: equality, so $\bt'$ is positive. \qed

2918: 

2919: Note the following easy and useful consequence of theorem 

2920: \reference{thg}:

2921: 

2922: \begin{corr}

2923: A link which is a closure of a positive braid of at most 4 strings,

2924: has a minimal crossing diagram as a closed positive braid, and a

2925: minimal string positive braid representation.

2926: \end{corr}

2927: 

2928: \proof The case of the braid representation is straightforward,

2929: and it implies the minimal crossing diagram statement by

2930: looking at $\Md_zP$ and using Morton's inequalities. \qed

2931: 

2932: The examples in \cite{posex_bcr}, mentioned after theorem \ref{thg},

2933: show that the corollary is not true in case of positive 5-braids at

2934: least for the positive minimal braid representation. So far no examples

2935: are known where no minimal crossing positive braid diagram exists

2936: (it was known to exist from \cite{WilFr,Murasugi} for closed

2937: positive braids with a full twist, which include the torus links,

2938: and from \cite{posex_bcr} for positive braid knots of at most

2939: 16 crossings), but the pathologies for minimal strings hint to caution.

2940: It was shown in \cite{posex_bcr} that one, and in \cite{ntriv} that

2941: infinitely many fibered positive knots

2942: have no minimal crossing positive diagram.

2943: 

2944: \subsubsection{Maximal subwords}

2945: 

2946: Here we start the technical considerations needed to prove 

2947: theorem \reference{thmwf}. We consider the form \eqref{wd},

2948: now with all $l_i>0$.

2949: 

2950: \begin{defi}

2951: We define \em{summit syllables} in \eqref{wd}:

2952: \def\theenumi{\alph{enumi}}

2953: \def\labelenumi{\theenumi)}

2954: \begin{enumerate}

2955: \item All $\sg_{n-1}^{l_i}$ are summit syllables, and

2956: \item if $\ap=\sg_k^{l_i}$ and $\ap'=\sg_k^{l_j}$ are summit

2957: syllables, with no $\sg_{k'}^{l'}$ for $k'\ge k$ occurring between

2958: $\ap$ and $\ap'$, then all $\sg_{k-1}^{l'}$ occurring between $\ap$

2959: and $\ap'$ are summit syllables.

2960: \end{enumerate}

2961: \end{defi}

2962: 

2963: Note that, according to definition \reference{dfext}, we

2964: consider syllables in cyclic order. The relation ``between'' in

2965: the above definition should also be understood in that sense:

2966: a syllable occurring after the last index $i$ syllable

2967: $\ap$ and/or before the first index $i$ syllable $\ap'$ is

2968: considered to be between $\ap$ and $\ap'$.

2969: 

2970: For the following considerations it is (not necessary but) helpful

2971: to visualize $\bt$ by the braid scheme explained in \cite{bind}.

2972: 

2973: Separate $\bt$ in

2974: \eqref{wd} into subwords $\ap_1\dots \ap_n$, such that $\ap_i$

2975: contains only syllables of odd or even index, and this parity

2976: changes between $\ap_i$ and $\ap_{i+1}$. Then for a syllable

2977: $\sg_k^l$ occurring in $\ap_i$, put the integer $l$ at the point

2978: $(k,i)\in \bN\times \bN\subset \bR^2$ in the plane. Here $(k,i)$

2979: is the point in the $i$-th row and $k$-th column, with rows

2980: numbered (as in Cartesian coordinates) from bottom to top and

2981: columns from left to right.

2982: 

2983: One obtains a certain checkerboard pattern of integers we call

2984: \em{braid scheme} of $\bt$. (If we do not put any integer on

2985: a point $(k,i)$, we assume its ``content'' is zero, or it is

2986: ``empty''. So for all non-empty points $(k,i)$ in the scheme,

2987: $i+k$ is always even or always odd.)

2988: 

2989: One can \em{reduce} the

2990: scheme by moving an integer $l$ at $(k,i)$ to $(k,i-2)$ if $i>2$

2991: and the points $(k\pm 1,i-1)$ are empty. We call the scheme

2992: reduced if it does not admit any such move. Then in a reduced

2993: scheme, summit syllables of $\bt$ are those, whose entries in

2994: the scheme are ``on top'' when viewing the scheme from the left.

2995: {}From this viewangle the following ``geographic'' choice of

2996: terminology becomes more plausible.

2997: 

2998: 

2999: \begin{defi}

3000: Summit syllables still have a cyclic order from \eqref{wd}. We call

3001: the subword $\bt'$ of $\bt$ in \eqref{wd} made of summit syllables

3002: the \em{maximal subword}. The subword made of non-summit syllables

3003: (i.e. the subword obtained by deleting in $\bt$ all syllables in

3004: $\bt'$) is called \em{non-maximal subword}. 

3005: 

3006: Note that neighbored summit syllables have indices $k_i$ differing

3007: by $\pm 1$. We say that a summit syllable is \em{minimal} resp.

3008: \em{maximal} if its both neighbors have higher resp.\ lower index.

3009: 

3010: We call $\bt$ \em{summit reduced} if all its minimal summit syllables

3011: are non-trivial.  We call $\bt$ \em{index reduced} if it is

3012: non-singular and its index sum $\sum_{i=1}^n k_i\cdot l_i$ cannot be

3013: reduced by a Yang-Baxter relation, i.e. $\bt$ contains no

3014: $\sg_{i+1}\sg_i\sg_{i+1}$ as subword.

3015: \end{defi}

3016: 

3017: \begin{lemma}

3018: Index reduced $\So$ summit reduced. In particular a summit reduced

3019: form always exists.\qed

3020: \end{lemma}

3021: 

3022: Recall that a positive resolution tree is a rooted tree with directed 

3023: edges,

3024: whose vertices (nodes) contain positive braid words, the root labelled

3025: by $\bt$. Every vertex has exactly one incoming edge, except the

3026: root that has none, and zero or two outoutgoing edges. In former

3027: case it is labelled by an unlink (terminal node).

3028: In latter case it is labelled by

3029: a word of the form $\ap\sg_i^2\ap'$, with $\ap,\ap'$ positive

3030: words, and the two vertices connected by the outgoing edges 

3031: are labelled by $\ap\sg_i\ap'$ and $\ap\ap'$, or positive words

3032: obtained therefrom by Markov equivalence (isotopy of the closure link).

3033: 

3034: In \cite{Nakamura} the following fact was observed, and used

3035: decisively, and we shall do the same here.

3036: 

3037: \begin{theorem}(Nakamura \cite{Nakamura})

3038: $MWF(\bt)$ is the maximal number of components of a (link in a) node

3039: in a positive resolution tree for $\bt$.

3040: \end{theorem}

3041: 

3042: In particular, $MWF$ is monotonous (does not decrease) under word

3043: extension, and does not depend on the exponent of non-trivial

3044: syllables.

3045: 

3046: \begin{lemma}\label{lemh}

3047: If $\bt$ is \em{summit reduced}, and $\bt'$ is obtained from $\bt$

3048: by removing all summit syllables, then there is a positive resolution

3049: tree for $\bt$ that contains $\bt'$ as a node.

3050: \end{lemma}

3051: 

3052: \proof Since minimal syllables are non-trivial, one can delete them

3053: in the resolution tree. The two neighbors in the maximal subword

3054: join to a non-trivial new minimal syllable, and so one iterates the

3055: procedure. \qed

3056: 

3057: Since all $\sg_{n-1}^{l'}$, $\sg_{n-2}^{l'}$ in \eqref{wd} occur as

3058: summit syllables, $\bt'$ has split last two strands, and so we have

3059: a quick proof of Nakamura's main result.

3060: 

3061: \begin{corr}(Nakamura \cite{Nakamura})

3062: Any summit reduced positive word on $n\ge 3$ strands has $MWF\ge 3$.

3063: \qed

3064: \end{corr}

3065: 

3066: In particular any positive braid representation of a $(2,n)$-torus

3067: link can be reduced to the standard one by index-decreasing YB

3068: relations and removals of nugatory crossings.  

3069: 

3070: \subsubsection{The proof of Theorem \reference{thmwf}:

3071: Initial simplifications\label{4.0}}

3072: 

3073: The following fact is well-known:

3074: 

3075: \begin{theorem}\label{th4.1}

3076: $MWF(\bt)=1$ if an only if $[\bt]_i=1$ for $i=1,\dots,n-1$.

3077: \end{theorem}

3078: 

3079: Now for $MWF=3$ it suffices to ensure that (either we can reduce the

3080: braids) or can find words, whose non-maximal subwords do not

3081: give an unknot. \em{For the rest of the section we assume that

3082: $\bt'$ gives the unknot.}

3083: 

3084: We will work by induction on the number of strands, and for

3085: fixed number of strands on the index sum. So we consider a

3086: positive braid word $\bt$, and assume w.l.o.g. it has the

3087: smallest index sum among positive braid representatives of

3088: its closure link for the same number of strands. For such

3089: $\bt$, we will either reduce it (by at least one strand or

3090: crossing), or show $MWF\ge 4$.

3091: 

3092: Most braids $\bt$ will be easily dealt with, but there remain certain

3093: families of words that require a case-by-case study. We decided not

3094: to omit too many of the (tedious) details of this part, in order

3095: to keep the proof followable, even if it may not contribute to

3096: its (esthetic) appearance.

3097: 

3098: Note that in order to prove $MWF(\bt)\ge 4$, it suffices to

3099: go over to a (link in a) suitably chosen node in a positive

3100: resolution tree for $\bt$ and show $MWF\ge 4$ for this node. In

3101: particular, we can remove from $\bt$ all syllables of index $\le

3102: k$ and the $k$ resulting left isolated strands.

3103: 

3104: The case of reductions is more delicate. In some situations

3105: we can describe them directly, but this is not always the case.

3106: Then we proceeded as follows. First we took generic examples, in

3107: making all syllables non-trivial whose triviality we have not argued

3108: about. We adjust parities so that the closure is a knot, and checked

3109: using KnotScape \cite{KnotScape} that the braid reduces (by at least one

3110: strand/crossing).

3111: 

3112: Later we wrote a computer program that seeks reductions by keeping

3113: given crossings rigid. Such reductions would commute with replacing

3114: rigid crossings by any tangle, in particular by any non-trivial

3115: braid word syllable. (Non-trivial syllables behave

3116: similarly to  rigid vertices, and suggest that the reduction

3117: is likely to work in general.) This way we can find reductions for

3118: infinite families of braids on a given number of strands. By turning

3119: all summit index-1 syllables into rigid crossings, one can also

3120: handle the braids that occur for an increasing number of strands.

3121: 

3122: The technical details of the application of this program are,

3123: however, tedious and little insightful. Instead we content

3124: ourselves with giving the examples we processed with KnotScape.

3125: % The alternative proof of theorem \reference{thg} given later

3126: % is another reason why we felt not motivated to make this

3127: % way the approach here even more technical.

3128: 

3129: 

3130: \begin{defi}

3131: A \em{valley} resp. \em{mountain} is a subword of the maximal word

3132: that starts with same index syllables and contains only one minimal

3133: resp. maximal syllable. This syllable is called the \em{bottom} of

3134: the valley resp.\ \em{summit} or \em{top} of the mountain.

3135: The index of the bottom/top is the \em{depth} resp.\ \em{height}.

3136: \end{defi}

3137: 

3138: We assume there are at least two mountains of maximal height (i.e.

3139: $n-1$). Otherwise we have a split component or a $(2,n)$-torus

3140: connected component, or a reducible braid and can work by induction

3141: on the number of strands. Similarly at least one valley has

3142: depth $1$, otherwise $\sg_1$ in $\bt'$ remains reducible in $\bt$.

3143: 

3144: The following operation will be somewhat important, and we will

3145: call it ``filling the valley''.

3146: 

3147: \begin{lemma}(``filling the valley'')

3148: Any valley can be removed from the maximal subword in the positive

3149: resolution tree.

3150: \end{lemma}

3151: 

3152: \proof Same as for lemma \reference{lemh}. \qed

3153: 

3154: \begin{lemma}

3155: If a mountain $M$ is not of maximal height

3156: (i.e. $n-1$), then $MWF(\bt)\ge 4$.

3157: \end{lemma}

3158: 

3159: \proof Let $k<n-1$ be the height of $M$.

3160: We assumed there are at least two mountains of maximal height. So

3161: now w.l.o.g. assume some, say

3162: the left, of the neighbored mountains of $M$ has height $k'>k$.

3163: Fill the two valleys around $M$ starting with $\sg_{k-1}$.

3164: Then the maximal subword has a syllable index sequence

3165: $k+1,k,k-1,k,k-1,k$. Make the second and third syllable trivial

3166: (if not already), and apply YB relations, moving the fourth

3167: syllable to the left: $k+1,\ul{k-1},k,k-1(,k-1),k$. The result is a

3168: summit reduced word, in which a new syllable of index $k-1$

3169: (the underlined one) was removed and it became non-maximal. Hence the

3170: non-maximal subword has exponent sum $[\bt']_{k-1}>1$, and so

3171: $MWF\ge 4$. \qed

3172: 

3173: \begin{lemma}\label{lem2val}

3174: If $\bt$ has $>2$ valleys of depth at most $n-3$, then $MWF(\bt)\ge 4$.

3175: \end{lemma}

3176: 

3177: \proof % It suffices to check it for 4-braids.

3178: It suffices to check for 3 mountains (as one can fill

3179: separate valleys) and 4-braids (as one can fill valleys by

3180: levels as in the proof of lemma \reference{lemh} and the remark

3181: after theorem \reference{th4.1}). This is just

3182: the word [1232112321121321], which is easily checked (to

3183: have $MWF=4$). \qed

3184: 

3185: \subsubsection{Two mountains\label{S4.1}}

3186: 

3187: We assume in \S\reference{S4.1} and \S\reference{S4.2} that $n\ge 5$.

3188: The case of $4$-braids is considered later in \S\reference{S4.3}.

3189: We refer to \S\reference{brs} for the use of notation we will employ.

3190: 

3191: We assume first $\bt$ has two mountains. By the previous remarks they

3192: are both of height $n-1$.

3193: 

3194: So now consider words with syllable index sequence

3195: \begin{eqnarray}\nonumber

3196: & 1,2,\dots,n-2,n-1,\quad

3197: p_1@,\dots,p_k@,\quad

3198: n-2,n-3,\dots,g+1,g!,\quad \\

3199: \label{xx} &

3200: g+1,\dots,n-2,n-1,\quad

3201: q_1@,\dots,q_l@,\quad

3202: n-2,n-3,\dots,2,1

3203: \end{eqnarray}

3204: such that $l+k=n-3$ and

3205: $\{p_1,\dots,p_k,q_1,\dots,q_l\}=\{1,\dots,n-3\}$.

3206: We will distinguish only between non-trivial and trivial syllables

3207: (in former case exponent is immaterial). For non-trivial syllables

3208: we write an exclamation mark

3209: after the index, for trivial ones an `at' (@) sign.

3210: If none of $!$ and @ is specified, we do not exclude explicitly any of

3211: either types. We write $\bt_1,\dots,\bt_6$ for the subwords separated by

3212: space in \eqref{xx}.

3213: 

3214: Assume w.l.o.g. (up to reversing the braid's orientation) that

3215: some $p_i$ is $1$, and let up to commutativity the $p_1,\dots,p_k$

3216: subword be written as $h,h-1,\dots,1,p_1',\dots,p_{k-h}'$ (the 

3217: $p_i'$ contain the indices above $h$ occurring as $p_i$).

3218: 

3219: Assume the maximal syllable of $\sg_{h+1}$ in $\bt_1$ is non-trivial.

3220: Then one can write $\bt$ as word with a subword of index sequence 

3221: \begin{eqn}\label{siks}

3222: 1,\dots,h+1!,h@,\dots,1@\,. 

3223: \end{eqn}

3224: If now $h<n-3$, we can make in \eqref{siks} all syllables

3225: trivial except $h+1$, which we make of exponent $2$, then

3226: split a loop (component of a link in a node of a positive

3227: resolution tree) by removing all the syllables in \eqref{siks}

3228: together with the terminating `$1$' in \eqref{xx}. We obtain

3229: \begin{eqnarray*}

3230: & h+2,\dots,n-2,n-1,\quad

3231: p_1'@,\dots,p_{k-h}'@,\quad

3232: n-2,n-3,\dots,g+1,g!,\quad \\

3233: & g+1,\dots,n-2,n-1,\quad

3234: q_1@,\dots,q_l@,\quad

3235: n-2,n-3,\dots,2.

3236: \end{eqnarray*}

3237: (Here no syllables of index 1 occur, and the split loop is the

3238: isolated leftmost strand.) Then we fill the valley starting with the

3239: index-$n-1$-summits, splitting another loop (the rightmost strand),

3240: \[

3241: h+2,\dots,n-3,n-2,\quad p_1'@,\dots,p_{k-h}'@,\quad q_1@,\dots,q_l@,

3242: \quad n-2,n-3,\dots,2,

3243: \]

3244: and are left with a

3245: word that has at least two $\sg_{n-2}$. So $MWF\ge 4$. 

3246: 

3247: \begin{caselist}

3248: \case Assume the smaller valley has depth $g>1$.

3249: The $\bt$ has two index-1 syllables, a trivial (non-summit)

3250: and a non-trivial (summit) one. By a flype one can exchange

3251: them, and so have a non-summit reduced word. Then one

3252: can change $\bt$ to a word of smaller index sum, and so

3253: we are done by induction.

3254: 

3255: % So we have $h=n-3$ (in particular $\bt_5$ is trivial).

3256: % Now if $g>2$, then the $p_l=1$ occurring at the

3257: % end in $\bt_2$, can be moved to the end of $\bt$ right

3258: % before the $\dots,2,1$ of $\bt_6$,

3259: % and by reversal of the word we have a situation with $h=1$.

3260: % We can exclude $g>2$ by assuming the present $h$ is the minimal

3261: % possible.

3262: % 

3263: % Thus $g=2$. Then we have a braid of the form

3264: % \begin{eqn}\label{seven}

3265: % 1*,2,\dots,n-2*,\quad

3266: % n-3@,\dots,1@,\quad

3267: % n-1,\dots,3,2*,\quad

3268: % 3,\dots,n-1,\quad

3269: % n-2,\dots,2\,.

3270: % \end{eqn}

3271: % According to the spacing group the syllables into subwords

3272: % $\bt'_{1}$ to $\bt'_5$.

3273: % 

3274: % By making all syllables trivial except the index-$n-2$ syllable in

3275: % $\bt'_1$ and the index-2 syllable in $\bt'_3$, we have a 3-component

3276: % link. Two components have no self-crossings (the first and last strand).

3277: % The third component has self-crossings, namely the syllables in

3278: % $\bt'_5$. If in $\bt$ some of the syllables in $\bt'_5$ were

3279: % non-trivial, we could now delete it, and have a 4-component link,

3280: % so that $MWF(\bt)\ge 4$.

3281: % So in \eqref{seven} the syllables in $\bt'_5$ are trivial.

3282: % 

3283: % % Now, by drawing a closed braid picture, we see that when the

3284: % % $n-2$-index syllable in $\bt'_5$ is trivial, and so is the

3285: % % $n-1$-index syllable in $\bt_3$, then the braid is Markov reducible.

3286: % % Thus the $n-1$-index syllable in $\bt_3$ is non-trivial.

3287: % % 

3288: % % Now remove the $n-1$-index syllable and also the other in $\bt_5$.

3289: % % Call the bew braid $\tl\bt$. It has 4 non-trivial syllables:

3290: % % in cycl order of inex $1$, $n-2$, $2$, $n-2$. 

3291: % % 

3292: % % When we make trivial the last two non-trivial syllables, the closure 

3293: % % becomes again a 3-component link, whose one component's self crossings

3294: % % are in $\bt_3$, except the $2*$. By the previous argument all syllables

3295: % % in $\bt_3$ must be trivial, except the $2*$.

3296: % 

3297: % Now, we argue that all syllables in $\bt'_3$ must be trivial, 

3298: % except the $n-1$ and $2*$. Namely, assume otherwise the $b$-index

3299: % syllable is non-trivial for $2<b<n-1$. Then remove all index $\le

3300: % b-2$ syllables in $\bt$. First let $b<n-2$. Let $\bt_i''$

3301: % be the result of $\bt_i'$ (for $1\le i\le 5$) under this

3302: % deletion. The $\sg_{b-1}$ in $\bt_1''$ now forms with $\bt_2''$

3303: % the letter-order-reversed subword \eqref{siks} (with $h=1$).

3304: % By the same loop split argument following \eqref{siks} we have

3305: % $MWF\ge 4$, since a $\sg_{n-2}$ remains in $\bt_2''$ and

3306: % $\bt_4''$. (Here $1=h<n-3$, or $n\ge 5$, is needed.)

3307: % If $b=n-2$, then we obtain after deleting index $\le b-2$

3308: % syllables a 5-braid. Making all syllables we can trivial,

3309: % we have the word 112332143323432. By direct check it has $MWF=

3310: % 4$, and so $MWF(\bt)\ge 4$. 

3311: % 

3312: % Similarly we have, as assumed, that all

3313: % syllables in $\bt'_2$ are trivial. So we have

3314: % \[

3315: % 1*,2,\dots,n-3,n-2*,\quad n-3@,\dots,1@,\quad

3316: % n-1,n-2@,\dots,3@,2*,\quad 3,\dots,n-1,\quad n-2@,\dots,2@

3317: % \]

3318: % 

3319: % By testing the 5-braids, we see that $MWF\ge 4$ if both

3320: % index $n-1$-syllables are non-trivial. If one is trivial,

3321: % one has to verify now that such braids reduce to positive

3322: % braids on a smaller number of strands (and then work by

3323: % induction). We checked this up to $n=9$, and it suggests

3324: % a reduction is possible for higher $n$.

3325: 

3326: \case Now $g=1$. We write $\bt_1,\dots, \bt_6$ for the

3327: 6 subwords separated by spacing in \eqref{xx}.

3328: By a similar argument as after \eqref{siks} we can argue that

3329: if one can reorder the syllables in $\bt_{2,5}$ so that

3330: $\bt$ has a subword with an index sequence 

3331: \begin{eqn}\label{(eight)}

3332: k,k+1,\dots,h-1,h!,h-1,\dots,k\,,

3333: \end{eqn}

3334: with the first or last

3335: $h-k$ syllables belonging to $\bt_{2,5}$ and the others

3336: to $\bt_{1,3,4,6}$, and $h<n-2$, then $MWF\ge 4$.

3337: 

3338: W.l.o.g. assume $1\in \bt_5$ (which is meant to abbreviate that

3339: $\bt_5$ contains an index-$1$ syllable).

3340: 

3341: \begin{caselist}

3342: \case

3343: Now if $2\not\in\bt_5$ (so $2\in\bt_2$) then the

3344: $2$-index syllables in $\bt_{4,6}$ are trivial

3345: (because we have otherwise \eqref{(eight)} with $k=1$, $h=2$).

3346: 

3347: We distinguish several cases by the subwords of

3348: (non-summit) syllables of index $1,2$ and $3$ in $\bt_{2,5}$.

3349: We separate the subwords between $\bt_{2}$ and $\bt_5$

3350: by a vertical line `$|$'. (We assume here that $n\ge 6$.

3351: The case $n=5$ must be handled by a separate, but

3352: simplified, argument.) Note that by symmetries we can

3353: exchange the words of $1,2$ and $3$ left and right from `$|$'

3354: and also (simultaneously) reverse both, and can also use the

3355: commutativity of $1$ and $3$. Then we are left with

3356: the following cases.

3357: 

3358: \begin{caselist}

3359: \case

3360: $2|31$. If $3\in \bt_5$, then both $3$ in $\bt_{1,3}$ are trivial.

3361: (Otherwise, we would have \eqref{(eight)} for $k=2$ and $h=3$.)

3362: % \begin{verbatim}

3363: Below we give pairs of words, the first obtained by extending all

3364: admissible syllables to be non-trivial, and the second one by

3365: extending the first word to one with knot closure, which was then

3366: checked to reduce (by at least one strand, not necessarily to a

3367: 3 braid).

3368: \\

3369: $[1122344  2 55665544322111   2334455 3441 66554433211]$, \\

3370: $[11222344 2 5556655444322111 2334455 3441 666554433211]$ reduces. \\

3371: % \end{verbatim}

3372: \case

3373: If $3\in \bt_2$, then one of both $3$-index syllables in

3374: $\bt_{1,3}$ must be trivial. (It is the syllable in $\bt_3$ if `$3$'

3375: occurs before `$2$' in $\bt_2$, or the syllable in $\bt_1$

3376: otherwise.)

3377: 

3378: \begin{caselist}

3379: \case

3380: $ 23|1$.

3381: % \begin{verbatim}

3382: \\

3383: $[1122344 23 556655443322111    2334455 441 66554433211]$, \\

3384: $[1122344 23 556665554443322111 2334455 441 66554433211]$ reduces. \\

3385: % \end{verbatim}

3386: \case $ 32|1$

3387: % \begin{verbatim}

3388: \\

3389: $[11223344  32 55665544322111   2334455 441 66554433211]$, \\

3390: $[112233444 32 5556665544322111 2334455 441 66554433211]$ reduces.  \\

3391: % \end{verbatim}

3392: \end{caselist}

3393: \end{caselist}

3394: 

3395: \case

3396: Now assume $2\in\bt_5$. Since $1\in\bt_5$, now the exclusion

3397: of \eqref{(eight)} shows that only one of the $2$-index syllables

3398: in $\bt_{4,6}$ is trivial, but we will show that also one in

3399: $\bt_{1,3}$ is. Namely, by making the proper index-$1$ (summit)

3400: syllable to exponent 2, and one of the $2$-index syllables of $\bt_

3401: {1,3}$ trivial, one can slide by braid relations the $1$-index

3402: syllable $\sg_1=X$ from $\bt_5$ to $\bt_2$.

3403: By applying the previous argument, both

3404: $2$-index syllables in (the now modified) $\bt_{1,3}$ are trivial.

3405: One of them was previously made trivial to slide $X$ in, but the

3406: condition on the other one persists for the original braid.

3407: 

3408: 

3409: In all situations,

3410: make all the other syllables in $\bt_{1,3,4,6}$ non-trivial and

3411: check using KnotScape that the braid reduces. 

3412: 

3413: % MAKE THIS EXACT!!!

3414: 

3415: \begin{caselist}

3416: \case $ 3| 12$. In this case the above argument shows that the

3417: syllable $2\in\bt_1$ is trivial.

3418: % \begin{verbatim}

3419: \\

3420: $[11233445566  344 55443322111 234455   12 66554432211]$, \\

3421: $[112334455666 344 55443322111 23444555 12 665544322111]$ reduces. \\

3422: % \end{verbatim}

3423: \case $ 3| 21$. Here $2\in\bt_3$ is trivial.

3424: % \begin{verbatim}

3425: \\

3426: $[112233445566 344 5544332111 2234455 21 6655443211]$, \\

3427: $[112233445566 344 5544332111 2234455 21 66655544432111]$ reduces. \\

3428: % \end{verbatim}

3429: \case $| 123$. As before $2\in\bt_1$ is trivial.

3430: % \begin{verbatim}

3431: \\

3432: $[11233445566 55443322111 234455 123 665544332211]$, \\

3433: $[11233445566 55443322111 234455 123 666555444332211]$ reduces. \\

3434: % \end{verbatim}

3435: \case $| 132$.

3436: % \begin{verbatim}

3437: \\

3438: $[11233445566 55443322111 2334455 132 66554432211]$, \\

3439: $[11233445566 55443322111 2334455 132 66655544432211]$ reduces. \\

3440: % \end{verbatim}

3441: \end{caselist}

3442: \end{caselist}

3443: 

3444: \end{caselist}

3445: 

3446: \subsubsection{More than two mountains\label{S4.2}}

3447: 

3448: % VERY VAGUE AND MUCH NEEDS TO BE DONE USING COMPUTATIONS !!!

3449: To deal with the general case, 

3450: now we make the following modifications. We call a

3451: summit syllable sequence with indices $n-2$ and $n-1$

3452: terminated on both sides by $n-1$'s a modified

3453: mountain or \em{plateau}. We have again by lemma

3454: \reference{lem2val} only two valleys of depth $<n-2$,

3455: or alternatively only two plateaus (now instead of mountains).

3456: The case of more than two mountains is thus mainly a

3457: adaptation of the case of two mountains, replacing

3458: mountains by plateaus.

3459: 

3460: % Also we allow $\bt$ and the maximal summit subword

3461: % to throw in occurrance of non-trivial $\sg_{n-1}$ syllables.

3462: 

3463: % The words that admit reductions apply as in the previous

3464: % two-mountains-case,

3465: % 

3466: % 2 deep, 1 shallow valley

3467: % 123211232213211 reduces

3468: % 123121123223211 reduces 

3469: % 

3470: % CHECK CAREFULLY !!!

3471: % 

3472: % , but for proving $MWF\ge 4$ the

3473: % following modifications of the argument are needed.

3474: 

3475: Again we may assume non-maximal subwords have exactly one (and trivial)

3476: syllable per index. The elimination of the maximal subwords can be done

3477: similarly. % only that when arriving at $\dots,n,n-1@,n,n-1@,n$

3478: 

3479: We distinguish two cases as in the above study of the 2-mountain

3480: words, depending on the depth $g$ of the second valley (the

3481: other valley has depth $1$ by the same argument as above).

3482: 

3483: \begin{caselist}

3484: \case $g>1$. We use the previous flyping argument.

3485: % In the first case above the same arguments show $g=2$ and (after

3486: % deleting the index-$n-2$ syllables in the plateaus) that

3487: % the syllables in $\bt_{3,5}$ in the form \eqref{seven} are trivial,

3488: % except the index-$n-2$ syllables in both and the index-$2$ and

3489: % $n-1$ syllable in $\bt_{3}$.

3490: % (Otherwise, fill one valley, remove a loop from \eqref{hh}, and

3491: % something non-trivial remains.) Also $n-2\in\bt_1$ may now be trivial.

3492: % 

3493: % {$112233\hat 4554455\ \ 321\ 5544\ul 32\ 2233445544\ 5544

3494: % \ul 3\ul 2$}

3495: % 

3496: % % CHECK ANYTH REDUCES !!!

3497: % One has to check that such braids reduce. For example, \\

3498: % % \begin{verbatim}

3499: % 11223345544455 321 55544322 22333445544 5544432 reduces, and \\

3500: % 1122334445544455 321 55544322 22333445544 5544432 reduces. 

3501: % \end{verbatim}

3502: 

3503: \case $g=1$.

3504: In the second case we had restrictions on exponents of syllables

3505: with index $2$ and $3$ occurring in the maximal subwords from

3506: the position of syllables with index $1$ and $2$ occurring in

3507: the non-maximal subwords. The restrictions on $2$-(index) syllables

3508: from $1$-(index) syllables remain. So do the restrictions on

3509: $3$-syllables from $2$-syllables unless we have $\le 5$

3510: strands. The argument is the same: one can still pull

3511: out two loops and has at least two letters of $\sg_{n-2}$.

3512: % (additionally discarding all additional non-trivial $\sg_{n-1}$

3513: % syllables)

3514: 

3515: Now we check restrictions on $3$-index syllables for $5$ strands

3516: and reducibility. We have up to extensions a finite number of

3517: special braids to verify. Clearly, extensions are never

3518: admissible for non-summit syllables, and always admissible for

3519: summit index-1-syllables (since they are all non-trivial). 

3520: Also extensions of subwords $(n-2\ n-2\ n-1)^k$ occurring

3521: repeatedly are redundant, since doubling the letter $n-1$ is the

3522: same as deleting the $n-2\ n-2$ for the next $k$. We

3523: will verify that the property $MWF\ge 4$ resp. reducibility

3524: does not depend on the value of $k$ as soon as $k>0$.

3525: We will thus deal only with the other extensions.

3526: 

3527: \begin{caselist}

3528: \case The non-summit $2$-index syllable is between two valleys

3529: of depth $n-2=3$ - this is handled as before: \\

3530: $[1234334123343211234321]$ has $MWF=4$, \\

3531: $[1234334213343211234321]$ has $MWF=4$, \\

3532: $[1234334233432112314321]$ has $MWF=4$.

3533: 

3534: \case The non-summit $2$-index syllable is between

3535: one depth $n-2=3$ valley and one depth $1$ valley. These are words

3536: of the form $[123412(334)^k3211234(334)^l321]$, \\

3537: $[123421(334)^k3211234(334)^l321]$,\\

3538: $[12342(334)^k32112341(334)^l321]$ \\

3539: for $k>0$, $l\ge 0$, and their extensions.

3540: 

3541: \begin{caselist}

3542: \case $[123412(334)^k3211234(334)^l321]$ and extensions.

3543: 

3544: \begin{caselist}

3545: \case $l=0$. To display the extendability of syllables,

3546: in the following notation the necessarily trivial syllables are

3547: hatted, while a possibly trivial syllable is underlined.

3548: 

3549: $[1\hat 2\hat 3\ul 412(334)^k\ul 3\ul 211\hat 2\hat 3\hat 4\ul 3\ul 21]$. 

3550: 

3551: When $l=0$, then making non-trivial any single of the hatted syllables

3552: makes $MWF=4$ already for $k=1$, while for any $k>0$ making any

3553: combination (possibly all) of the underlined syllables non-trivial

3554: gives $MWF=3$.

3555: 

3556: Now again check that braids reduce, for example: \\ 

3557: $[1234123343343211234321]$ has $MWF=3$, \\

3558: $[12341233433433432111234321]$ reduces to 3 strands, \\

3559: $[1234441233433433221123433221]$ reduces to 3 strands.

3560: 

3561: 

3562: \case 

3563: When $l=1$, the already for $k=1$ we have without extensions

3564: $[1234123343211234334321]$ and $MWF=4$.

3565: \end{caselist}

3566: 

3567: \case $[123421(334)^k3211234(334)^l321]$ and extensions.

3568: 

3569: \begin{caselist}

3570: \case When $l=0$, we have for all $k$ the extendability

3571: 

3572: $[1\ul 2\hat 3\ul 421(334)^k\ul 3\hat 211\ul 2\hat 3\hat 4\ul 3\hat 21]$,

3573: 

3574: with the same explanation as before.

3575: 

3576: For example, $[122344213343343343321122343321]$ reduces to 3 strands.

3577: 

3578: \case When $l\ge 1$, then already for $k=1$ and no extensions

3579: $MWF([1234213343211234334321])=4$.

3580: 

3581: \end{caselist}

3582: 

3583: \case $[12342(334)^k32112341(334)^l321]$ and extensions.

3584: 

3585: \begin{caselist}

3586: \case When $l=0$, we have

3587: 

3588: $[1\ul 2\hat 3\ul 42(334)^k\ul 3\ul 211\hat 2\hat 3\hat 41\ul 3\hat 21]$.

3589: 

3590: For example, $[122344233433433433221123413321]$ reduces to 3 strands.

3591: 

3592: \case When $l>0$, already for $k=1$, $l=1$ without extension

3593: we have $[1234233432112341334321]$ and $MWF=4$.

3594: \end{caselist}

3595: \end{caselist}

3596: 

3597: \case The non-summit $2$-index syllable is between

3598: two depth $1$ valleys. These are (up to symmetry) words of the form\\

3599: $[1234(334)^k321123421321]$ and \\

3600: $[12341(334)^k32112342321]$, \\

3601: for $k>0$, and their extensions (here necessarily $l=0$).

3602: 

3603: \begin{caselist}

3604: \case $[1234(334)^k32112342132]$ and extensions:

3605: 

3606: $[1\ul 2\ul 3\ul 4(334)^k\ul 3\hat 211\ul 2\hat 3\hat 421\hat 3\hat 21]$.

3607: 

3608: Again for $k=1$ making non-trivial all underlined syllables 

3609: gives $MWF=3$, while making non-trivial any of the hatted

3610: syllables gives $MWF=4$. 

3611: 

3612: For example,

3613: $[12 3  4 3343       2112 3 214321]$ has $MWF=3$, and its extensions\\

3614: $[122333443343       211223 214321]$ and \\

3615: $[12233 4433433433433211223421 321]$ \\

3616: were checked to reduce to 3 strands.

3617: 

3618: The same is the outcome for $k>1$.

3619: 

3620: \case $[12341(334)^k32112342321]$ and extensions:

3621: 

3622: $[1\h 2\u{34}1(334)^k\u 3\h 211\u 2\h 3\h 42\h 3\u 21]$

3623: 

3624: The reducibility cases follow analogously. % CHECK !!

3625: For example, $[123344133433433433211223423221]$ reduces to 3 strands.

3626: 

3627: % For $l>0$ we have always $MWF=4$.

3628: 

3629: \end{caselist}

3630: 

3631: \end{caselist}

3632: \end{caselist}

3633: 

3634: \subsubsection{4-braids\label{S4.3}}

3635: 

3636: If some mountain is not of height 3, or $>2$ valleys of

3637: depth $1$ exist, then we are done as before (see \S\reference{4.0}).

3638: 

3639: So the maximal subword is of the form

3640: $1,2,3,(2!,3)^p,2,1,1,2,3,(2!,3)^n,2,1$, and the non-maximal

3641: subword is a single $\sg_1^1$. We separate the summit syllables

3642: and their letters by the summit syllables of index $1$ into

3643: a \em{left} and \em{right plateau}. Assume w.l.o.g. the (non-summit)

3644: $\sg_1^1$ is in the left plateau. The word `in' is to mean that

3645: in cyclic order of the syllables of $\bt$ the syllable $\sg_1^1$

3646: can be written to occur just before or after a syllable with

3647: index $n-1$ that belongs to the left plateau. This means that

3648: we can write $\bt$ as

3649: \begin{eqn}\label{blmn}

3650: \bt_{l,m,n}=[1\vtbox{2\\.}3(223)^n1(223)^m\vtbox{2\\..} 1123(223)^l21]

3651: \end{eqn}

3652: with $n,m,l\ge 0$, or some of its extensions. Note that MWF will

3653: be monotonous in $m,n,l$, i.e. $MWF(\bt_{l,m,n+1})\ge

3654: MWF(\bt_{l,m,n})$ etc. Using symmetry assume $n\ge m$.

3655: 

3656: One can check already at this stage that such words are not

3657: $B$-adequate. So we obtain corollary \reference{cort}, and

3658: for the proof of theorem \reference{thg} the rest of the

3659: argument here can be replaced by the application of theorem

3660: \reference{thq}. Note that $B$-adequacy is invariant

3661: under isotopy preserving writhe and crossing number, so

3662: any other positive 4-braid word giving the same link is

3663: not $B$-adequate either.

3664: 

3665: If $n+l>0$, then the $2$-index syllable $\vtbox{2\\..}$

3666: in \eqref{blmn} must be trivial.

3667: % there are $2*$ of the left plateau on the left (right) of

3668: % $\sg_1^1$ or some in the right plateau, then the right

3669: % (left) $2$-index syllable of that mountain must be trivial.

3670: Otherwise remove all $2!$ in $(2!,3)^m$ (if any), and split two loops

3671: as explained after \eqref{siks}. The `$2!$' in $(2!,3)^l$ or

3672: $(2!,3)^n$ remain, and so $MWF=4$.

3673: With a similar argument we see that if $m+l>0$, then the $2$-index

3674: syllable $\vtbox{2\\.}$ is trivial.

3675: 

3676: We distinguish 3 cases depending on whether these arguments

3677: apply or not.

3678: 

3679: \begin{caselist}

3680: \case Both non-triviality arguments apply. So we have a family of

3681: words $\bt_{l,m,n}=[123(223)^n1(223)^m21123(223)^l21]$ with $n,m>0$,

3682: or $l>0$ and their extensions, and both $\vtbox{2\\..}$ and

3683: $\vtbox{2\\.}$ are trivial.

3684: 

3685: \begin{caselist}

3686: \case $l=0$. We assumed $n,m>0$, and already for $n=m=1$,

3687: the word $[12322312232112321]$, we have $MWF=4$.

3688: 

3689: \case $l>0$.

3690: 

3691: \begin{caselist}

3692: \case $m=n=0$. These are extensions of $[1231211123(223)^l21]$,

3693: and the admissibility is found to be:

3694: 

3695: $[1\h 2\h 31\h 211\ul{23}(223)^l\ul 21]$.

3696: 

3697: The two letters `$2$' in the left plateau cannot be doubled

3698: ($[123122112322321]$, $[122312112322321]$), neither the `$3$' \\

3699: ($[123132112322321]$), since $MWF=4$ already for $l=1$.

3700: 

3701: Without extension, $l=1$ ($[123121112322321]$) and $l=3$

3702: ($[123121112322322322321]$) reduce.

3703: % 

3704: % $[1231211223223223221]$ reduces.

3705: % 

3706: % % So so do (sloppy) $123121123[23]+21$ in regexp notation.

3707: % 

3708: % $[123121112223223223223221]$ has $MWF=3$. \\

3709: % $[1231211122232232232232221]$ reduces. \\

3710: So we find that $[12312112*3[23]+2*1]$ reduce.

3711: (Recall that, while `$2*$' in

3712: a braid word should mean at least one letter `$2$', the term `$[23]+$'

3713: should mean a possibly empty sequence of letters `$2$' and `$3$'.

3714: We distinguish braid words from index sequences by not putting

3715: commas between the numbers.)

3716: 

3717: \case $m+n>0$; this reduces to the case of $m=n=0$ with some of the

3718: twos or the three in the left plateau doubled, where we found $MWF=4$.

3719: 

3720: \end{caselist}

3721: 

3722: \end{caselist}

3723: 

3724: 

3725: \case In the case one of the non-triviality conditions on

3726: $\vtbox{2\\..}$ and $\vtbox{2\\.}$ does not apply,

3727: we have $123(223)^n12112321$ with $n>0$ and its extensions.

3728: (Now the right plateau is a mountain.)

3729: 

3730: % \begin{caselist}

3731: % \case

3732: The case $n=3$ ($[123223223223121112321]$) simplifies. \\

3733: % So so do (sloppy) $123[23]+312112321$ in regexp notation.

3734: % Also 12223223223223121112321, so so do (sloppy) $12*[23]+312112321$.

3735: $[122332232232232231211222321]$ simplifies.

3736: 

3737: 

3738: However, $MWF([123223122112321])=4$, so the right $2$-index syllable

3739: in the left plateau ($\vtbox{2\\..}$ in \eqref{blmn}) must be

3740: trivial.

3741: 

3742: The right 2 and the 3 in the right plateau must be trivial: \\

3743: $[1232231211123221]$, \\ $[1232231211123321]$ have $MWF=4$.

3744: 

3745: But the left `$2$' of the left and right plateau may not be trivial:

3746: $[12223223223223121112222321]$ has $MWF=3$. \\

3747: $[122232232232231211122222321]$ simplifies. \\

3748: So $[12*[23]+312112*321]$ simplifies. % ?

3749: 

3750: We arrive at the form $[1\u{23}(223)^n1\h 211\u 2\h 3\h 21]$.

3751: 

3752: % \dots

3753: % \end{caselist}

3754: 

3755: \case

3756: In case both non-triviality conditions do not apply, we have

3757: $[12312112321]$ and its extensions. (So both plateaus are

3758: mountains.)

3759: 

3760: Since non-triviality is nonetheless possible, we may have non-trivial

3761: $2$-index syllables in the left plateau.

3762: We distinguish three cases again according to whether

3763: $\vtbox{2\\..}$ and $\vtbox{2\\.}$ are trivial or not.

3764: 

3765: 

3766: \begin{caselist}

3767: \case Both $2$-index syllables are trivial: $[12312112321]$.

3768: 

3769: % 12312112321

3770: 

3771: By direct check: \\

3772: % 122312112321 $MWF=3$

3773: $[123312112321]$ has $MWF=3$, \\

3774: % 123122112321 $MWF=3$

3775: $[123121122321]$ has $MWF=3$, \\

3776: $[123121123321]$ has $MWF=3$, \\

3777: $[123121123221]$ has $MWF=3$. \\

3778: 

3779: % 12233122112321 $MWF=3$

3780: % 122331221122321 $MWF=4$

3781: % 122331221123321 $MWF=4$

3782: % 122331221123221 $MWF=4$

3783: 

3784: $[12312112233221]$ has $MWF=3$, \\

3785: % 122312112233221 $MWF=4$

3786: $[123312112233221]$ has $MWF=4$. \\

3787: % 123122112233221 $MWF=4$

3788: 

3789: So one can extend the right mountain's `$2$'s and one of the

3790: left or right mountain's `$3$'s, but not both `$3$'s.

3791: 

3792: $[123121122233221]$ reduces, \\

3793: $[123312112223221]$ reduces, \\

3794: $[1233121122233221]$ has $MWF=4$.

3795: 

3796: So a reducing check is to be made on $[11*231211*22*3*22*]$.

3797: % and 11*22*3*122*11*232

3798: 

3799: $[123121112233221]$ reduces, \\

3800: $[123121112222333322221]$ reduces, \\

3801: $[1233312111222322221]$ reduces.

3802: 

3803: % \begin{caselist}

3804: % \case

3805: % 11*231211*22*3*22*

3806: % 

3807: % \case 11*22*3*122*11*232

3808: % 

3809: % 122331222112321 DOES NOT REDUCE !!!

3810: % 122331222112321 connected-2-cabled MWF =7 !!!!!

3811: % 

3812: % So we have exceptions 11*22*3*122*11*232 !

3813: % (Or 22*3*122*11*23211*, which is the

3814: % first family in theorem \reference{thmwf}.)

3815: % 

3816: % 

3817: % 12331222112321 2-cabled MWF =6 !!!!!

3818: % 112331222112321 reduces

3819: % 11233331222112321 reduces

3820: % 

3821: % 12231222112321 2-cabled MWF =7 !!!!!

3822: % 1223122112321 2-cabled MWF =7 !!!!!

3823: % 

3824: % So we have exceptions 11*22*3122*11*232 !

3825: % (Or 22*3122*11*23211*, which is a special case of the

3826: % first and second family in theorem \reference{thmwf}.)

3827: % 

3828: % 1223312112321 2-cabled MWF =6 !!!!!

3829: % 12222333312112321 reduces

3830: 

3831: \case

3832: One $2$-index syllable is non-trivial. This is the word

3833: $[123122112321]$, with $MWF=3$ (the case

3834: $[122312112321]$ is symmetric).

3835: 

3836: We have the following extensions:

3837: 

3838: $[1233122112321]$ has $MWF=3$ (and reduces), \\

3839: $[1231221122321]$ has $MWF=4$, \\

3840: $[1231221123221]$ has $MWF=3$ (and reduces), \\

3841: $[1231221123321]$ has $MWF=3$ (and reduces).

3842: 

3843: The following combined extensions are to check:

3844: $[12331221123321]$ has $MWF=4$, \\

3845: $[12331221122321]$ has $MWF=4$, \\

3846: $[12312211223321]$ has $MWF=4$. \\

3847: 

3848: Thus we are left to deal with

3849: $[1233*122*112321]$,\\

3850: $[123122*112322*1]$,\\

3851: $[123122*11233*21]$,\\

3852: and check that they all reduce:

3853: 

3854: $[12333312222112321]$ reduces,\\

3855: $[12312222112322221]$ reduces,\\

3856: $[12312222112333321]$ reduces.

3857: 

3858: \case Both $2$-index syllables are non-trivial:

3859: $[1223122112321]$. This is a braid word $\bt_0$ for $13_{9465}$.

3860: We have the following extensions:

3861: 

3862: $[12233122112321]$ has $MWF=3$, \\

3863: $[12231221122321]$ has $MWF=4$, \\

3864: $[12231221123221]$ has $MWF=4$, \\

3865: $[12231221123321]$ has $MWF=3$.

3866: 

3867: The only common extension of the two $MWF=3$ extensions is \\

3868: $[122331221123321]$, which has $MWF=4$.

3869: 

3870: So it remains to verify that (the closures of)

3871: $[22*3*122*11*23211*]$ and $[22*3122*11*23*211*]$ have braid index 4.

3872: 

3873: % 1222233312211232111 no reduce

3874: % 12223122211123321 no reduce

3875: 

3876: For this we use the two-cabled MWF inequality. Let for a braid

3877: $\bt\in B_n$, the ``two-cabled'' braid $(\bt)_2\in B_{2n}$ be

3878: obtained from $\bt$ by replacing in \eqref{wd} each $\sg_i$

3879: by $\sg_{2i}\sg_{2i-1}\sg_{2i+1}\sg_{2i}$. Then $(\bt)_2$

3880: is a braid representation of the (blackboard framed) two-cable

3881: link $L_2$ of the closure $L=\hat\bt$ of $\bt$. We consider now

3882: for $\bt$ the above braid $\bt_0$.

3883: 

3884: We know, from the computations described in \cite{MorSho,WilFr},

3885: that $MWF((13_{9465})_2)=7$. This is in fact true also for the

3886: connected cable (the one with braid representation $(\bt_0)_2

3887: \cdot\sg_1$). Now we claim that reducing or resolving a clasp

3888: (changing a $\sg_i^2$ into a $\sg_i$ or deleting it) does not

3889: reduce the two-cabled MWF bound. The 2-cable of a clasp can be

3890: resolved by resolving 4 clasps. The 2-cable of a crossing in a

3891: 2-cabled clasp can be resolved by resolving one clasp and changing

3892: twice $\sg_i^2\to \sg_i$. Finally, the 2-cable of an isolated 

3893: $\sg_i$ can be reduced into two internal twists of the

3894: doubled original strand. Such twists can be collected for

3895: every doubled component, resolved for each doubled component to

3896: one, and joined if doubled components are joined by reducing

3897: a doubled crossing in a doubled clasp. So the two-cabled MWF

3898: reduces to the one of the connected cable of $13_{9465}$ and

3899: we are done.

3900: \end{caselist}

3901: \end{caselist}

3902: 

3903: The proof of theorem \reference{thmwf} is now completed.

3904: 

3905: \subsection{Positive links\label{Spos}}

3906: 

3907: In this section, we will refine the arguments proving

3908: theorem \reference{thqp} to restrict the possible 

3909: 3-braid representations of positive links. 

3910: % and then, % in an alternative way, classify

3911: % positive braid links of braid index 3.

3912: % 

3913: Our positivity considerations will make use of the criterion

3914: of Yokota \cite{Yokota}, and the Kauffman polynomial $F$.

3915: We recall the properties \eqref{wseven}~--~\eqref{wseven.7}\,

3916: that determine $F$ and its writhe-unnormalized version $\Lm$.

3917: 

3918: Note that for $P$ one can similarly define a regular isotopy

3919: invariant

3920: \begin{eqn}\label{dPtl}

3921: \tl P(D)(a,z)=(ia)^{-w(D)}P(D)(ia,iz)\,,

3922: \end{eqn}

3923: with $i=\sqrt{-1}$. Then $\tl P$ satisfies similar relations to

3924: \eqref{wseven}~-- \eqref{wseven.7}. The difference to $\Lm$ is that

3925: $\tl P$ is defined on oriented link diagrams, and that the term making

3926: orientation incompatible on the right of \eqref{wseven} is missing.

3927: 

3928: \begin{theorem}\label{thF}(Yokota \cite{Yokota})

3929: If $L$ is a positive link, then 

3930: \[

3931: \md_a F(L)\,=\,\md_v P(L)\,=\,1-\chi(L)\,,

3932: \]

3933: and

3934: \[

3935: [F(L)]_{a^{1-\chi(L)}}=[P(L)(ia,iz)]_{a^{1-\chi(L)}}\,.

3936: \]

3937: \end{theorem}

3938: 

3939: We also require an extension of braids to the context of $F$.

3940: This was described in \cite{BW} and \cite{Murakami}, but we use

3941: only the generators of the algebra defined there. Strings will be

3942: assumed numbered from left to right and words will be composed

3943: from bottom to top. We write $\sg_i$ for a braid generator, where

3944: strand $i$ from the lower left corner, passing over strand $i+1$,

3945: goes to the upper right corner.

3946: 

3947: We add elements $\dl_i$ of the following form:

3948: \[

3949: \diag{4mm}{6}{2.5}{

3950:   \pictranslate{0 0.5}{

3951:   \picline{0 0}{0 2}

3952:   \picmultigraphics{3}{0.2 0}{

3953:     \picfilledcircle{0.55 1}{0.03}{}

3954:   }

3955:   \picline{6 0}{6 2}

3956:   \picmultigraphics{3}{0.2 0}{

3957:     \picfilledcircle{5.05 1}{0.03}{}

3958:   }

3959:   \picline{1.5 0}{1.5 2}

3960:   \picline{4.5 0}{4.5 2}

3961:   \piccirclearc{3 0}{0.5}{0 180}

3962:   \piccirclearc{3 2}{0.5}{180 0}

3963:   \picputtext{2.5 -.3}{\footnotesize $i$}

3964:   \picputtext{3.5 -.3}{\footnotesize $i+1$}

3965:   }

3966: }\es.

3967: \]

3968: By hat we denote the usual closure operation.

3969: 

3970: A word in the described generators gives rise to an unoriented tangle

3971: diagram that turns into an unoriented link diagram under closure. 

3972: If the word has no $\dl_i$ then this diagram can be oriented to

3973: give a(n oriented) closed braid diagram. We will assume this

3974: orientation is chosen. Otherwise, a coherent orientation is not

3975: generally possible. In particular, the sign of exponents of $\sg_i$

3976: in this context may not coincide with the sign of the corresponding

3977: crossings after some (or even any) orientation choice of the diagram.

3978: For kinks (the diagram fragments occurring in \eqref{wseven.5} on the

3979: left hand-sides),

3980: however, a sign is definable since any possible component

3981: orientation chosen gives rise to the same (skein) sign. So we will be

3982: able (and we will need) to distinguish between positive (in the

3983: left equation of \eqref{wseven.5}) and negative

3984: kinks (in the right one).

3985: 

3986: \begin{lemma}\label{lemma1}

3987: Let 

3988: \begin{eqn}\label{Df}

3989: D=\wh{\es}\,\left(\,

3990: \sg_1^{k_{1,1}}\sg_2^{k_{1,2}}\dots \sg_1^{k_{1,n_1}}\sg_2^{-1}\quad

3991: \sg_1^{k_{2,1}}\dots \sg_1^{k_{2,n_2}}\sg_2^{-1}\quad\dots

3992: \quad \sg_l^{k_{l,1}}\dots \sg_1^{k_{l,n_l}}\dl_2\,\right)

3993: \end{eqn}

3994: where $n_i\ge 1$ odd and $k_{i,j}\ge 2$ when $1<j<n_i$ and

3995: $k_{i,j}\ge 1$ when $j=1$ or $n_i$.

3996: Then $\md_a\Lm(D)=-l$.

3997: \end{lemma}

3998: 

3999: \proof For $l=1$ see lemma \reference{lemma4} below. Then use induction

4000: on $l$. Change a crossing of a $\sg_2^{-1}$ in $D=D_-$. Then:

4001: 

4002: $D_+$ can be reduced by the same argument as in the proof of

4003: lemma 5.2 of \cite{ntriv} until we have a form \eqref{Df} with

4004: smaller $l$, and a non-zero number of negative kinks added.

4005: Then since negative kinks shift the $a$-degree of $\Lm$ up,

4006: we have $\md_a\Lm(D_+)>-l$.

4007: 

4008: $D_0$ has $\md_a\Lm(D_0)=1-l>-l$ by induction.

4009: 

4010: $D_\infty=D_{l_1}\# D_{l_2}$ with $l_1+l_2=l$, so $\md_a\Lm(D_

4011: \infty)=-l_1-l_2=-l$, and $\md_a\Lm(D)$ is inherited from

4012: $\Lm(D_\infty)$.

4013: \qed

4014: 

4015: \begin{lemma}\label{lemma2}

4016: Let 

4017: \begin{eqn}\label{Df2}

4018: D=\wh{\es}\,\left(\,\sg_1^{k_{1,1}}\sg_2^{k_{1,2}}\dots

4019: \sg_1^{k_{1,n_1}}\sg_2^{-1}\quad

4020: \sg_1^{k_{2,1}}\dots \sg_1^{k_{2,n_2}}\sg_2^{-1}\quad\dots

4021: \quad \sg_l^{k_{l,1}}\dots \sg_1^{k_{l,n_l}}\sg_2^{-1}\,\right)

4022: \end{eqn}

4023: where $n_i\ge 1$ odd and $k_{i,j}\ge 2$ when $1<j<n_i$ and

4024: $k_{i,j}\ge 1$ when $j=1$ or $n_i$.

4025: Then $\md_a\Lm(D)\ge -2$ when $l\le 2$ and $\md_a\Lm(D)=-l$

4026: when $\l\ge 3$.

4027: \end{lemma}

4028: 

4029: \proof Assume first we proved the result for $l\le 2$, and that

4030: $l>2$. We argue by induction on $l$.

4031: 

4032: Apply the $\Lm$-relation at a $\sg_2^{-1}$ crossing in $D=D_-$.

4033: Then $D_0$ has $\md_a\Lm=1-l$ by induction. $D_+$ simplifies

4034: as in the proof of lemma 5.2 of \cite{ntriv}. This simplification

4035: only removes negative letters. It can be iterated until one

4036: of two situations occurs. It can (a) happen that all negative

4037: letters disappear. Then we have a positive braid and $\md_a\Lm=-2>-l$

4038: by Yokota's result. Or it can (b) occur that no $\sg_1\sg_2\sg_1$

4039: or $\sg_2\sg_1\sg_2$ occur as subwords. Then $k_{i,j}\ge 2$

4040: for $1<j<n_i$, and the number $l$ of negative crossings has

4041: decreased strictly. So we have by induction $\md_a\Lm(D_+)>-l$.

4042: Finally we must deal with the $D_\infty$ term. This follows

4043: from lemma \reference{lemma1}.

4044: 

4045: It remains to justify the claim $\md_a\Lm(D)\ge -2$ when $l\le 2$.

4046: This is done exactly with the same argument, only that now we

4047: observe that all of $\Lm(D_{0,+,\infty})$ have $a$-degree $\ge -2$. \qed

4048: 

4049: \begin{lemma}\label{lemma3}

4050: The closed braids in \eqref{Df2} are not positive for $l\ge 2$.

4051: \end{lemma}

4052: 

4053: \proof The representation \eqref{Df2} clearly gives rise to

4054: a positive band representation by replacing $\sg_2\sg_1^{k_{i,n_i}}

4055: \sg_2^{-1}$ by $\sg_3^{k_{i,n_i}}$. So $[\bt]=3-\chi(\hat\bt)$.

4056: If $l>2$, then we have

4057: \[

4058: \md_a F(D)\,=\,w(D)\,+\,\md_a \Lm(D)\,=\,3-\chi(\hat\bt)+\md_a \Lm(D)\,

4059: <\,1-\chi(\hat\bt)\,

4060: \]

4061: but $\md_v P(D)=1-\chi(\hat\bt)$ as before, so $\md_a F\ne \md_v P$

4062: and we are done by theorem \reference{thF}.

4063: 

4064: If $l=1$, then if $n_1=1$, we have a $(2,k_{1,1})$-torus

4065: link, and for $n_1=3$ we have the $(1,k_{1,1},k_{1,2},k_{1,3})

4066: $-pretzel link. Otherwise we apply the relations for $\tl P$

4067: and $\Lm$ at the negative crossing. Then $\md_a\Lm(D_{+,0})=

4068: \md_a\tl P(D_{+,0})=-2$, and

4069: $[\Lm(D_{+,0})]_{a^{-2}}=[\tl P(D_{+,0})]_{a^{-2}}$

4070: by theorem \reference{thF} since $D_{+,0}$ are positive braids.

4071: That the extra term $\Lm(D_\infty)$ has no contribution to

4072: $a^{-2}$ follows from lemma \reference{lemma4} below. So

4073: $D$ satisfies Yokota's conditions in theorem \reference{thF}.

4074: (We do not always know if $D$ depicts a positive link; see

4075: remark \reference{rty}.)

4076: 

4077: If $l=2$ then resolve a negative crossing in $D=D_-$ via the

4078: relation $\tl P_-=z \tl P_0-\tl P_+$ and via the $\Lm$-relation

4079: $\Lm_-=z\Lm_0+z\Lm_\infty-\Lm_+$. Now $[\tl P_{+,0}]_{a^{-2}}=

4080: [\Lm_{+,0}]_{a^{-2}}$. This follows from the above argument

4081: for $l=1$. The additional term $\Lm_\infty$ has $a$-degree

4082: $-2$ by lemma \reference{lemma1}, so $[\Lm_-]_{a^{-2}}\ne

4083: [\tl P_-]_{a^{-2}}$. Again by theorem \reference{thF},

4084: $D=D_-$ can therefore not belong to a positive link.  \qed

4085: 

4086: \begin{rem}

4087: Ishikawa asks in \cite{Ishikawa} whether strongly

4088: quasi-positive knots satisfy the equality $TB=2g_s-1$,

4089: where $TB$ is the maximal Thurston-Bennequin invariant.

4090: The proof of the lemma shows that there are infinitely many

4091: strongly quasi-positive 3-braid knots with $TB<2g_s-1$:

4092: take any of the knots with $l>2$. (There are also many other

4093: such knots, like the counterexamples to Morton's conjecture

4094: in \cite{posex_bcr}.)

4095: \end{rem}

4096: 

4097: \begin{lemma}\label{lemma4}

4098: If 

4099: \[

4100: D=D_{[n]}=\wh{\es} (\sg_1^{k_1}\sg_2^{k_2}\dots\sg_1^{k_n}\dl_2)

4101: \]

4102: with $n\ge 1$ odd, $k_1,k_n\ge 1$ and $k_i\ge 2$ when $i=2,\dots,n-1$,

4103: then $\md_a \Lm(D)=-1$. Writing $\ol{k}=(k_1,\dots,k_n)$ and

4104: \[

4105: \tl w(\ol{k})\,=\,(k_1-1)+\sum_{l=2}^{n}(k_l-2)\,, 

4106: \]

4107: we have $\Md_z [\Lm(D)]_{a^{-1}}\,=\,\tl w(\ol{k})$.

4108: \end{lemma}

4109: 

4110: \proof If $n=1$ we check directly (we have a reduced diagram

4111: of the $(2,k_1)$-torus link), so let $n\ge 3$.

4112: 

4113: Consider first three special forms of $\ol{k}$.

4114: 

4115: If $\ol{k}=(12^*1)$ (with $2^*$ being a sequence of `$2$'),

4116: then $D$ is regularly isotopic to a trivial

4117: 2-component link diagram. If $\ol{k}=(12^*)$ or $(2^*1)$, then $D$

4118: is regularly isotopic to an unknot diagram with one positive kink.

4119: In both situations the claims follow directly.

4120: 

4121: Now let $\ol{k}=(2^*)$. We orient $D$ so as to become negative.

4122: Then $D$ depicts the $(2,-n-1)$-torus

4123: link. We can evaluate $\Lm$ on $L$ from $F(L)$ by

4124: normalization. For its mirror image $!L$,

4125: we can use theorem \reference{thF} to conclude that

4126: $\md_a F(!L)=n$. Now it is also

4127: known that $\spn_a F(!L)=c(!L)$, and $c(!L)=n+1$, so

4128: \[

4129: \Md_a F(!L)=\md_a F(!L)+\spn_a F(!L)=n+(n+1)=2n+1\,.

4130: \]

4131: Thus $\md_a F(L)=-1-2n$, and since $w(D)=-2n$, we have

4132: \[

4133: \md_a \Lm(D)=-w(D)+\md_a F(D)=-(-2n)-1-2n=-1\,.

4134: \]

4135: That $\Md_z[\Lm(D)]_{a^{-1}}=1$ can also be obtained by

4136: direct calculation.

4137: 

4138: If $\ol{k}$ is not of these special types, then $k_l\ge 3$ for some

4139: $1\le l\le n$. We resolve a positive crossing in $\sg_j^{k_l}$.

4140: We have (with $D=D_+$)

4141: \begin{eqn}\label{*}

4142: \Lm (D_+)\,=\,z\Lm (D_0)\,+z\Lm(D_\infty)-\Lm(D_-)\,.

4143: \end{eqn}

4144: Here $D_0$ has $\tl w$ by one less and comes with a $z$-factor,

4145: so it is enough to show that $\Lm(D_-)$ and $\Lm(D_\infty)$

4146: do not contribute.

4147: 

4148: If $k_l>3$, then $D_-$ is of the required form and has $\tl w$

4149: by two less, so by induction $\Lm(D_-)$ has too small $z$-degree

4150: in $[\Lm]_{a^{-1}}$.

4151: 

4152: Now consider $k_l=3$. As in the proof of lemma 5.2 of \cite{ntriv} we

4153: can move

4154: by braid relations (regular isotopy) a $\sg_1$ in $\sg_1\sg_2\sg_1$

4155: until we get a $\sg_i\sg_i^{-1}$ (and then cancel). Here we

4156: replaced $\sg_i^{-1}$ by $\dl_i$, so that such $\sg_i$ right before

4157: a $\dl_i$ becomes a kink, which is negative. By repeating this

4158: transformation, we obtain a form that has no $\sg_i\sg_{i\pm 1}\sg_i$

4159: (i.e. $k_i>1$ for $1<i<n$) and a certain non-zero number of

4160: negative kinks collected at both ends. The negative kinks

4161: shift the degree in $a$ of $\Lm$ up, so by induction $D_-$

4162: has no contribution to $[\Lm]_{a^{-1}}$.

4163: 

4164: It remains to deal with the term of $D_\infty$ in \eqref{*}.

4165: It is the connected sum $D_{[n_1]}\# D_{[n_2]}$ with $n_1+n_2=n-1$,

4166: and with $k_l-1>1$ negative kinks. So by induction $\md_a\Lm(D_\infty)

4167: =k_l-1+(-1)+(-1)>-1$, and $\Lm(D_\infty)$ gives no

4168: contribution to $[\Lm(D_+)]_{a^{-1}}$. \qed

4169: 

4170: \begin{theorem}\label{thposq}

4171: If a 3-braid link is positive, then it is the closure

4172: of a positive or almost positive 3-braid. Along these links

4173: the non-fibered ones are exactly the $(1,p,q,r)$-pretzel

4174: links.

4175: \end{theorem}

4176: 

4177: \proof Consider the first claim.

4178: % If a 3-braid is almost positive, and does not

4179: % reduce to a positive one, then it is of the form

4180: % \eqref{Df2} with $l=1$. % We observed in the proof

4181: % of lemma \reference{lemma3}, that such links are positive.

4182: % 

4183: Since $L$ is positive, it is strongly

4184: quasi-positive, and so has a positive 3-braid band representation

4185: $\bt$ by theorem \reference{thqp}. So $\bt$ is of Xu's form $R$

4186: or $(21)^kR$ (with $k>0$) up to extensions. If $\bt$ contains

4187: $\sg_1\sg_2\sg_1$ or $\sg_2\sg_1\sg_2$ we can reduce it

4188: as before, until (a) it becomes positive, or (b)

4189: it still has a positive band representation, but it

4190: does not contain $\sg_1\sg_2\sg_1$ or $\sg_2\sg_1\sg_2$.

4191: In case (a) we are done. In case (b) we observe that

4192: $\bt$ is of the form \eqref{Df2}, apply lemma

4193: \reference{lemma3}, and conclude that $l\le 1$.

4194: 

4195: It remains to argue which links are not fibered.

4196: Positive braids are always fibered, and by direct

4197: observation for an almost positive braid we have

4198: Xu's form $R$ or $(21)^kR$ depending on whether

4199: in \eqref{Df2} (with $l=1$) we have $n_1=3$ or

4200: $n_1>3$. We proved in theorem \reference{Thq}

4201: that the forms $(21)^kR$ (for $k>0$) give fibered

4202: closure links. For $n_1=3$ we have as before the

4203: $(1,p,q,r)$-pretzel links. \qed

4204: 

4205: \begin{rem}\label{rty}

4206: Unfortunately, we cannot completely determine which of the almost

4207: positive braid representations with $l=1$ and $n_1>3$ in \eqref{Df2}

4208: give positive links. For example the Perko knot $10_{161}$ has

4209: such a representation, and it is positive. The Perko move (see

4210: \cite{HTW}), turning the closed braid diagram of $10_{161}$

4211: into a positive diagram, applies for more general examples.

4212: However, some knots, like $14_{46862}$, do not seem subjectable

4213: to this or similar moves, and their positivity status remains

4214: unclear at this point.

4215: \end{rem}

4216: 

4217: Similarly, one would hope to prove that

4218: among these links none is a positive braid link (and this way

4219: to obtain a different, but much more insightful proof of theorem

4220: \reference{thg}). Using the polynomials, one can exclude certain

4221: families, for example all links of $n_1=5$, but a complete argument

4222: again does not seem possible.

4223: 

4224: % we have a link, which belongs to a family

4225: % of the Perko knot $10_{161}$; latter arises when $n_1=5$, $k_{1,*}=

4226: % (1,2,3,2,1)$. The Perko move (see \cite{HTW}) turns this diagram

4227: % into a positive diagram, and the same move applies for the more

4228: % general examples.

4229: 

4230: % At least we can derive theorem \reference{thg} (again),

4231: % now possibly in a more favorable way.

4232: % 

4233: % \proof[of theorem \reference{thg}]

4234: % The one direction is trivial. Assume thus $L$ is braid positive and

4235: % of braid index $3$. In particular, $L$ is positive, so by theorem

4236: % \reference{thposq}, it is a closed positive or almost positive 

4237: % 3-braid $\bt$. Assume (trivially) we have latter case. $L$ is

4238: % also fibered, so $n_1>3$ in \eqref{Df2} (with $l=1$).

4239: % 

4240: % Now we use the skein polynomial $P$ and claim

4241: %  

4242: % \begin{lemma}\label{lemma6}

4243: % We have for such closed braids $L=\hat\bt$

4244: % that $[P]_{z^{-1-\chi}v^{3-\chi}}=0$. %and $V_1=V_2=0$, where

4245: % % $n(L)$ is the number of components of $L$.

4246: % \end{lemma}

4247: %  

4248: % It is easy to see that no positive braid has such

4249: % polynomial. The term in $[P]_{(zv)^{1-\chi}}$

4250: % comes from a terminal node in the positive resolution

4251: % tree in which all crossings are smoothed out until

4252: % a braid as in theorem \reference{th4.1} remains.

4253: % By considering a node in the tree in which all crossings

4254: % are smoothed out except exactly one, we have a Hopf

4255: % link, and a Hopf link has $[P]_{z^{-1}v^{3}}=1$.

4256: % Since its contribution is not cancelled, a non-trivial

4257: % positive braid must have $[P]_{z^{-1-\chi}v^{3-\chi}}\ne 0$.

4258: % \qed

4259: % 

4260: % % By \cite{posqp}, we have for a (non-trivial, non-split)

4261: % % positive braid link that $V_2\ne 0$, so $L$ is not the

4262: % % closure of apos braid, and we are done. \qed

4263: % % 

4264: % \proof[of lemma \reference{lemma6}]

4265: % The minimal possible word {\tt 1 2 2 1 1 2 2 1 -2} is found to have

4266: % $P$ of the stated type by direct calculation. Now we use induction

4267: % on the word length and apply the skein relation for $P$ at

4268: % some positive letter in a non-trivial syllable. We work

4269: % again with the form \eqref{Df2} (with $l=1$). If $n_1=5$ we

4270: % choose $i$ so that the syllable has exponent $k_{1,i}\ge 3$

4271: % if $1<i<n_1$ or $k_{1,i}\ge 2$ if $i=1$ or $n_1$. We have 

4272: % \begin{eqn}\label{prel}

4273: % P(D_+)=v^2P(D_-)+vzP(D_0)\,.

4274: % \end{eqn}

4275: % Now all three occurring links are strongly quasi-positive.

4276: % So 

4277: % \[

4278: % \md_v P(D_-)+2=\md_v P(D_0)+1=\md_v P(D_+)\, .

4279: % \]

4280: % % (This can be concluded from the explanations in

4281: % % \cite{posqp}.) Also the terms $t^2V_(D_-)$ and $t^{3/2}

4282: % % V_(D_0)$ have $V_0=\pm 1$ (of oppsoite sign!) and $V_1=0$

4283: % % Because either braids either reduce to a positive braid,

4284: % 

4285: % The choice of $k_i$ for $n_1=5$ prevents

4286: % us from descending into the undesriable situation that

4287: % $D_-$ or $D_0$ has $n_1=3$. 

4288: % $D_0$ is always of the required form.

4289: % So in particular 

4290: % \begin{eqn}\label{pd0}

4291: % [P(D_0)]_{v^{-1-\chi(D_0)}z^{3-\chi(D_0)}}=0

4292: % \end{eqn}

4293: % by induction.

4294: % 

4295: % $D_-$ may reduce to a positive braid if $k_i=3$ and $1<i<n_1$ or $k_i

4296: % =2$ for $i=1$ or $n_1$. Otherwise it

4297: % is also of the previous form with $n_1>3$.

4298: % In either case $D_-$ is fibered. Now it follows from the

4299: % description of $[P]_{1-\chi}$ in \cite{3br} (more explicitly

4300: % given in \S\reference{SA}) that fibered strongly quasi-positive

4301: % 3-braids have

4302: % \begin{eqn}\label{pd-}

4303: % [P(D_-)]_{z^{1-\chi(D_-)}v^{3-\chi(D_-)}}=0\,.

4304: % \end{eqn}

4305: % Putting now \eqref{pd0} and \eqref{pd-} into

4306: % \eqref{prel} gives the result. \qed

4307: 

4308: \section{Studying alternating links by braid index\label{Sfg}}

4309: 

4310: The combination of the identity \eqref{p21} and the skein-Jones

4311: substitution \eqref{VPsub} was already used in \S\reference{Js}

4312: to translate the determination of the 3-braid link genus from

4313: $P$ to $V$. A similar line of thought will now enable us to

4314: extend the other main result in \cite{3br}, the description

4315: of alternating links of braid index 3. This result was

4316: motivated by the work of Murasugi \cite{Murasugi},

4317: and Birman's problem in \cite{MortonPb} how to

4318: relate braid representations and diagrammatic properties

4319: of links. We will see how via \eqref{VPsub} and 

4320: the famous Kauffman-Murasugi-Thistlethwaite theorem

4321: \cite{Kauffman2,Murasugi3,Thistle2} the Jones polynomial

4322: enters in a new way into the braid representation picture.

4323: The argument will lead to the braid index 3

4324: result surprisingly easily, and then also to the

4325: classification for braid index 4 (which seems out of scope

4326: with the methods in \cite{3br} alone). We also obtain a

4327: good description of the general (braid index) case.

4328: 

4329: Our starting point is the following general result concerning the

4330: MWF-bound \eqref{mwfb}. A diagram is called special if it

4331: has no separating Seifert circles; see \cite{Cromwell}.

4332: The number of Seifert circles of $D$ is denoted by $s(D)$.

4333: 

4334: \begin{theorem}\label{_ty}

4335: Assume $L$ is a non-trivial non-split alternating link, and $MWF(L)=

4336: k$. Then an alternating (reduced) diagram $D$ of $L$ has $s(D)\le 2k-

4337: 2$ Seifert circles, and equality holds only if $D$ is special.

4338: \end{theorem}

4339: 

4340: \proof As $L$ is non-trivial and non-split, we have $1-\chi(L)> 0$.

4341: It is also well known, that $\Md_zP(L)=1-\chi(L)$ (see

4342: \cite{Cromwell}). Now $\spn_v P(L)\le 2k-2$ by assumption. Under

4343: the substitution in \eqref{mwfb} this translates to 

4344: \begin{eqn}\label{sV}

4345: \spn V(L)\le 1-\chi(L)+2k-2\,.

4346: \end{eqn}

4347: On the other hand, by \cite{Kauffman2,Murasugi3,Thistle2},

4348: $\spn V(L)=c(D)$, and also $1-\chi(L)=c(D)-s(D)+1$.

4349: So

4350: \begin{eqn}\label{ff}

4351: c(D)\le c(D)-s(D)+1+2k-2\,,

4352: \end{eqn}

4353: and then $s(D)\le 2k-1$. Now if this is an equality, then

4354: so is \eqref{sV}. Then one easily sees that $P(L)$ must have

4355: non-zero coefficients in both monomials $z^{\Md_zP}v^{\Md_vP}$

4356: and $z^{\Md_zP}v^{\md_vP}$. Now under the substitution in \eqref{p21},

4357: both these monomials give a non-cancelling contribution,

4358: and one of them is not in ($v$-)degree 0, so the identity

4359: \eqref{p21} cannot hold. Moreover, if \eqref{sV} fails

4360: just by one, then still one of the two coefficients must

4361: be non-zero. In order its (non-cancelling) contribution on

4362: the left of \eqref{p21} to be in degree $0$, we see that

4363: either $\Md_zP=\md_vP$, or $\Md_zP=-\Md_vP$. Using \cite

4364: {Cromwell}, one concludes then that these are precisely the cases of a

4365: (positively or negatively) special alternating link. \qed

4366: 

4367: Using \cite{Yamada} and \cite{Vogel} we have some simple

4368: estimate on the unsharpness of MWF for alternating links.

4369: 

4370: \begin{corr}

4371: For an alternating non-trivial non-split link $L$, we have

4372: $b(L)\le 2MWF(L)-2$. \qed

4373: \end{corr}

4374: 

4375: Of course, for many (in particular alternating) links

4376: $b(L)=MWF(L)$ (that is, MWF is exact), or at least $b-MWF$

4377: is small. So the above estimate should be considered as

4378: a worst-case-analysis. Even if not strikingly sharp,

4379: it is still far from trivial, in view of what we

4380: already know can occur for non-alternating knots. Namely,

4381: using the construction in \cite{Kanenobu2} and the work

4382: in \cite{BirMen3} (see remark \reference{rfn} below),

4383: one can find sequences of knots $(K_i)$ for which $MWF$

4384: (in fact the full $P$ polynomial) is constant, but

4385: $b(K_i)\to\infty$.

4386: 

4387: Another observation is that one can now extend the outcome

4388: of the work in \cite{SV} by replacing crossing number of

4389: an alternating knot by its braid index.

4390: 

4391: \begin{corr}

4392: Let $n_{g,b}$ denote the number of alternating knots of

4393: braid index $b$ and genus $g$. Then $n_{g,b}$ is finite.

4394: Moreover, for $g$ fixed, we have that

4395: $\lim_{b\to \infty} n_{g,b}/b^{6g-4}=C_g$ is a constant.

4396: \end{corr}

4397: 

4398: \proof The finiteness of $n_{g,b}$ follows by \eqref{ff}.

4399: When $\chi(L)$ is fixed, and $c(L)\to\infty$, one easily

4400: sees from \eqref{ff} that $MWF(L)$ behaves asymptotically

4401: (up to an $O(1)$, i.e. bounded, term)

4402: at least like $c(D)/2$. Now Ohyama's result \cite{Ohyama}

4403: implies that $b(L)$ (and $MWF(L)$ as well) behave asymptotically

4404: exactly as $c(L)/2$. So from \cite{SV} we have the result.

4405: \qed

4406: 

4407: \begin{rem}\label{rfn}

4408: Of course we could also gain, as in \cite{SV}, an estimate on the

4409: $C_g$ and its asymptotics for $g\to \infty$, it would just multiply

4410: by $2^{6g-4}$. One should also note that the finiteness of

4411: $n_{g,b}$, which one sees from \eqref{ff}, is not necessarily

4412: clear \em{a priori}. In fact, however, Birman-Menasco proved

4413: \cite{BirMen3} that $n_{g,b}$ is a finite number even for \em{general}

4414: (i.e. without restriction to alternating) knots. Their methods seem,

4415: though, quite unhelpful to estimate these numbers properly.

4416: \end{rem}

4417: 

4418: Theorem \ref{_ty} immediately leads to the first slight sharpening

4419: of the description of alternating links of braid index 3 in

4420: \cite{3br}. (The case of $MWF=2$ is even more obvious, and omitted.)

4421: 

4422: \begin{corr}\label{opi}

4423: An alternating link has $MWF=3$ if and only if it has braid index 3.

4424: \end{corr}

4425: 

4426: \proof If $MWF(L)\le 3$, then the alternating diagram has at

4427: most 4 Seifert circles, and exactly 4 only if it is special.

4428: Apart from connected sums (which are easily handled), we obtain

4429: the diagrams of closed alternating 3-braids and the $(p,q,r,s)$-

4430: pretzel diagrams. By direct calculation of $P$ we saw in

4431: \cite{3br} that if $\min(p,q,r,s)\ge 2$, then $MWF\ge 4$,

4432: and that otherwise the pretzel link has braid index 3. \qed

4433: 

4434: The case of 4-braids is now not too much more difficult.

4435: % (even though

4436: % it was completely out of scope with the method in \cite{3br}).

4437: 

4438: \begin{theorem}\label{tty}

4439: Let $L$ be a prime non-split alternating link. The following 3

4440: conditions are equivalent:

4441: \begin{enumerate}

4442: \item\label{Xa} $MWF(L)=4$

4443: \item\label{Xb} $b(L)=4$

4444: \item\label{Xc} $L$ is one of the links, whose reduced 

4445: alternating diagrams are described (up to mirror images) as follows

4446: \begin{enumerate}

4447: \item\label{Ya} The Murasugi (or connected) sum of three

4448:   $(2,n_i)$-torus links (with $|n_i|>1$),

4449: \item\label{Yb} The Murasugi (or connected) sum of a $(2,n)$-torus

4450:   link with a $(p,q,r,s)$-pretzel link, with one of $p,q,r,s$ equal

4451:   to 1, or its mirror image, or

4452: \item\label{Yc} a special diagram whose Seifert graph

4453: (see \S\ref{Sdg}) is as shown in figure \reference{figSG}.

4454: \end{enumerate}

4455: \end{enumerate}

4456: \end{theorem}

4457: 

4458: \proof $\reference{Xb}\So \reference{Xa}$. This follows from

4459: corollary \ref{opi}.

4460: 

4461: $\reference{Xc}\So \reference{Xb}$. That the links in

4462: \reference{Xc} have braid index at least 4 follows from the

4463: description of the links with $b(L)\le 3$ in \cite{3br}

4464: (which also comes out of the proof of corollary \eqref{opi}).

4465: It is also not too hard to check that for all these links

4466: $b=4$ by exhibiting a diagram $D$ with $s(D)=4$. The most

4467: systematic way seems to apply the graph index inequality of

4468: Murasugi-Przytycki \cite{MurPrz2} (see also \cite{Ohyama}).

4469: 

4470: $\reference{Xa}\So \reference{Xc}$. By applying theorem

4471: \reference{_ty}, we need to deal with non-special

4472: diagrams of at most 5 and special diagrams of at most 6

4473: Seifert circles.

4474: 

4475: First consider the non-special diagrams.

4476: 

4477: For the fibered links (the reduced Seifert graph is a tree),

4478: Murasugi's result \cite{Murasugi} leads directly to case

4479: \reference{Ya}. For non-fibered links, the Seifert graph

4480: must have a cycle, which must be of length at least 4 (the

4481: Seifert graph is bipartite). Then the only option

4482: that remains is case \reference{Yb}. We must still

4483: argue why one of $p,q,r,s$ must be $\pm 1$. This can

4484: be done using Murasugi-Przytycki's work, but one easily

4485: sees it also by a direct skein theoretic argument, which

4486: we explain.

4487: 

4488: Let $L(p,q,r,s)$ be the corresponding link and $P(p,q,r,s)$

4489: its skein polynomial. Look first at $p=0$, $q,r,s\ge 2$. Then

4490: $L(p,q,r,s)$ a connected sum of two $(2,n)$-torus

4491: links, and a closed alternating 3-braid. So

4492: $MWF(L(p,q,r,s))=5$. Now since the case $p=1$, $q,r,s\ge 2$ has

4493: $MWF(L(1,q,r,s))\le b(L(1,q,r,s))=4$, the skein relation \eqref{srel}

4494: easily shows that the maximal degree coefficient $\Mc_vP(p,q,r,s)$

4495: of $v$ in $P$ (which is a polynomial in $z$) for $p=2$ is inherited

4496: from $p=0$. Then further applications of \eqref{srel} show that

4497: the $z$-degree of $\Mc_vP(p,q,r,s)$ increases with $p$, so 

4498: in particular this term never vanishes, and so $MWF(L(p,q,r,s))=5$.

4499: 

4500: \begin{figure}[htb]

4501: {\small

4502: \[

4503: \begin{array}{c}

4504:   \diag{6mm}{3}{2}{

4505:   \pictranslate{0 0.5 x}{

4506:     \diag{6mm}{3}{2}{

4507:       \cycl{0.5 2}{0.5 0.5}{2 0.5}{2 2}

4508:     }

4509:   }

4510:   \picputtext{1.75 0}{\scriptsize (all multiplicities $\ge 2$)}

4511:   }

4512:   \qquad

4513:   \diag{6mm}{3}{2}{

4514:     \cycl{1.3 2}{0 1}{1.1 0}{1.5 1}

4515:     \cycl{1.3 2}{2.8 1}{1.1 0}{1.5 1}

4516:     \picputtext{2.5 0.4}{$1$}

4517:   }

4518:   \qquad

4519:   \diag{6mm}{3}{2}{

4520:     \pictranslate{0 0.5}{

4521:     \xcycl{0 0}{1.5 0}{3 0}{3 1.5}{1.5 1.5}{0 1.5}

4522:     \picputtext{0.75 -0.3 x}{$1$}

4523:     \picputtext{0.75 -0.3}{$1$}

4524:     }

4525:   }

4526:   \qquad

4527:   \diag{6mm}{3}{2}{

4528:     \pictranslate{0 0.5}{

4529:     \xcycl{0 0}{1.5 0}{3 0}{3 1.5}{1.5 1.5}{0 1.5}

4530:     \picline{1.5 1.5}{1.5 0}

4531:     \picputtext{2.25 -0.3}{$1$}

4532:     \picputtext{0.75 -0.3}{$1$}

4533:     }

4534:   }

4535:   \qquad

4536:   \diag{6mm}{4}{2.5}{

4537:     \pictranslate{0 0.5}{

4538:     \xcycl{0 0}{1.5 0}{3 0}{3 1.5}{1.5 1.5}{0 1.5}

4539:     \picline{1.5 1.5}{1.5 0}

4540:     \picputtext{2.25 1.8}{$1$}

4541:     \picputtext{0.75 -0.3}{$1$}

4542:     \opencurvepath{0 1.5}{0 2}{0.5 2.5}{3 2.5}{4 2}{4 0.5}{3.5 0}{3 0}{}

4543:     }

4544:   }

4545:   \\[5mm]

4546: \end{array}

4547: \]

4548: }

4549: \caption{The reduced Seifert graphs of the alternating diagrams

4550: of special alternating links of braid index 4. Simple edges

4551: have their multiplicity (1) attached, and the other edges (of

4552: multiplicity one or more) are unlabelled. In the first graph,

4553: exceptionally, all edges have multiplicity at least 2.\label{figSG}}

4554: \end{figure}

4555: 

4556: 

4557: Now consider the special diagrams. For them one considers the

4558: Seifert graph, and needs to write

4559: down all bipartite planar graphs on at most 6 vertices, which

4560: have no cut vertex. Since the diagram $D$ is special, the placement

4561: of multiple copies of an edge give diagrams equivalent up to

4562: flypes, so it is enough to consider simple graphs (the reduced

4563: Seifert graph), and have the multiplicities of an edge written

4564: as its label. The graphs can be easily compiled using the observation

4565: that they must contain a cycle of length 4 or 6; see figure

4566: \reference{figSG}. By direct inspection we see that the edge

4567: multiplicities must be as specified in the figure. (In fact if an

4568: edge is multiple it turns out irrelevant what its multiplicity

4569: is, so in this case we just omit the label.) We rule out the

4570: remaining multiplicities by a skein theoretic calculation,

4571: similar to the one explained for case \reference{Yb}.

4572: 

4573: Let w.l.o.g. (up to mirroring) $D$ be positive. For each edge

4574: $e$ in $\Gm(D)$ of variable multiplicity $\ge i$ (where $i=1,2$)

4575: we calculate the skein polynomial of the diagram that

4576: corresponds to $\Gm$ for multiplicities $i$ and $i+1$ of $e$.

4577: That is, if $\Gm(D)$ has $l$ edges of variable multiplicity,

4578: we have $2^l$ polynomials to calculate. Then we check for

4579: each such set of $2^l$ polynomials that $Q=P_{v^{9-\chi(D)}}$

4580: is non-zero, and $\Md_zP-\Md_zQ$ as well as $\Mc_zQ$ is

4581: constant within this set of $2^l$ polynomials. Then

4582: by \eqref{srel} this property is inherited to diagrams

4583: $D$ whose $\Gm(D)$ have edges of higher multiplicity,

4584: and in particular $MWF\ge 5$.

4585: 

4586: There is one more graph,

4587: \[

4588: \diag{6mm}{3}{2}{

4589:   \cycl{2.0 2}{0 1}{1.8 0}{1.2 1}

4590:   \cycl{2.0 2}{2.4 1}{1.8 0}{3.5 1}

4591: }

4592: \]

4593: not included in figure \reference{figSG}. In that

4594: case (by the method we just explained we verify that)

4595: $MWF\ge 5$ for all non-zero edge multiplicities. \qed

4596: 

4597: In \cite{MurPrz2}, Murasugi-Przytycki define a certain

4598: quantity $\inx(D)$, assigned to a link diagram $D$,

4599: called \em{index}. (We omit here the detailed discussion;

4600: one can consult also \cite{Ohyama} or \cite{gener}.)

4601: Their motivation was to give an upper estimate 

4602: \begin{eqn}\label{mp}

4603: b(L)\le s(D)-\inx(D)

4604: \end{eqn}

4605: for the braid index of the underlying link $L$. Their origin

4606: of \eqref{mp} consists in an appropriate move (see

4607: figure 8.2 in \cite{MurPrz}) which reduces

4608: the number of Seifert circles of the diagram.

4609: Murasugi-Przytycki conjectured that for an alternating

4610: diagram $D$, the inequality \eqref{mp} is exact. This

4611: conjecture is also confirmed for alternating links up

4612: to braid index 4. 

4613: 

4614: Based on the Murasugi-Przytycki procedure, we can re-enter

4615: the Bennequin surface topic.

4616: 

4617: \begin{corr}\label{cfT}

4618: Alternating links of braid index at most 4 carry a Bennequin

4619: surface on a minimal string braid.

4620: \end{corr}

4621: 

4622: \proof In \cite{gener} it was explained how to apply

4623: restrictedly the Seifert circle reduction move of

4624: Murasugi-Przytycki (and then those of Yamada \cite{Yamada})

4625: so as to obtain a braided surface. This modified reduction

4626: is easily checked to lead to the minimal number of strings for the

4627: links in question. \qed

4628: 

4629: % (It was verified in \cite{gener}

4630: % for alternating knots of genus up to 4.)

4631: 

4632: \begin{corr}

4633: Let $L$ be an alternating link of braid index $4$. Then

4634: $|\Mc\Dl(L)|\le 5$, and if $|\Mc\Dl(L)|>2$, then $L$ is

4635: special alternating.

4636: \end{corr}

4637: 

4638: \proof It is well known, that $|\Mc\Dl(L)|$ is multiplicative

4639: under Murasugi sum and for a special diagram depends only

4640: on the reduced Seifert graph. The result follows by calculation

4641: for the specific types. \qed

4642: 

4643: The proof of this corollary, and the extension of the multiplicativity

4644: of $\Mc\Dl$ to $\Mc_zP$ for diagrammatic Murasugi sum \cite{MurPrz}

4645: demonstrates also the following more general principle:

4646: 

4647: \begin{corr}\label{cSG}

4648: For any given braid index, there are only finitely many values of

4649: $\Mc\Dl$ and $\Mc_zP$ among Alexander and skein polynomials of

4650: alternating links of that braid index. \qed

4651: \end{corr}

4652: 

4653: We saw that for $\Dl$ this statement is wrong for non-alternating

4654: links even among 3-braids. On the other hand, by \cite{3br},

4655: it is true for $P$, and we do not know if it remains true for

4656: (closed) braids on more strings. (One could also ask if

4657: infinitely many leading coefficients of $\Dl$ occur if $\Md\Dl

4658: =1-\chi$, but we se no deeper meaning in this question, so will

4659: not dwell further upon it here.)

4660: 

4661: The following is also worth observing. Call a subclass

4662: $\cC'$ of a class $\cC$ of links \em{generic} in $\cC$ if

4663: \[

4664: \lim_{n\to\infty}\,\frac{\#\,\{L\in \cC'\,:\,c(L)=n\,\}}

4665: {\#\,\{L\in \cC\,:\,c(L)=n\,\}}\,=\,1\,.

4666: \]

4667: 

4668: \begin{corr}

4669: The number of special alternating links of given braid index

4670: grows polynomially in the crossing number. In particular,

4671: a generic alternating link of given braid index is not

4672: special alternating.

4673: \end{corr}

4674: 

4675: \proof A special alternating link is determined by the Seifert

4676: ($=$ checkerboard) graph of its alternating diagram. The number

4677: of such graphs with a fixed number of vertices grows polynomially

4678: in the number of edges. The second claim in the corollary

4679: follows because it is easy to see that in contrast the number

4680: of non-special alternating links grows exponentially (due to

4681: exponentially many, in the crossing number, ways to perform the

4682: Murasugi sum at a separating Seifert circle of the alternating

4683: diagram). \qed

4684: 

4685: \begin{rem}

4686: Note that in contrast we showed in \cite{SV} that a generic

4687: alternating knot (and the case of links is analogous) of given

4688: \em{genus is} special alternating. This shows from yet another

4689: point of view the opposition between genus and braid index.

4690: \end{rem}

4691: 

4692: Another immediate and useful consequence of theorem \ref{tty}

4693: and the preceding remarks is

4694: 

4695: \begin{corr}\label{cft}

4696: If an alternating link $L$ has $MWF(L)\le 4$ (in particular

4697: if $b(L)\le 4$), then the MWF inequality is exact

4698: (i.e. an equality) for $L$. \qed

4699: \end{corr}

4700: 

4701: This gives a nice complement to the MWF exactness results in

4702: \cite{Murasugi,MurPrz2}. (As another such amplification, we 

4703: proved the case of \em{knots} and genus $\le 4$ in \cite{gener}.)

4704: 

4705: Note that at MWF bound 5 we hit already at the Murasugi-Przytycki

4706: examples \cite{MurPrz2} of non-exact MWF (with $b=6$). So corollary

4707: \reference{cft} is not true for $MWF\ge 5$ or $b\ge 6$. We do

4708: not know about the case $b=5$. However, ruling out braid index

4709: $5$ for the Murasugi-Przytycki family is a serious computational

4710: problem (only two specific members were dealt with, a 4-component

4711: 15 crossing link and an 18 crossing knot; see \cite[\S 19]

4712: {MurPrz2}). Already with this circumstance in mind,

4713: one cannot expect to easily extend the corollary (or theorem

4714: \reference{tty}) for $b=5$ either, even if it may be true.

4715: 

4716: On the other hand, leaving these troublesome exceptions aside,

4717: the above discussion should fairly clearly explain how

4718: the general picture continues for alternating links with MWF

4719: bound 5 and more.

4720: 

4721: \section{Applications of the representation theory\label{SBurau}}

4722: 

4723: So far the representation theory behind the skein, Jones, and

4724: Alexander polynomial was not used. We will give some applications

4725: of it now. The theory is well explained in \cite{Jones}. We will

4726: use Jones's conventions, unless otherwise specified.

4727: 

4728: \subsection{The Jones conjecture\label{SJS}}

4729: 

4730: There is a conjecture, often attributed to Jones (who speculated

4731: on it at least for knots and in it its weaker form, as given below;

4732: see 357 l.-6 of his paper), stating that a minimal

4733: string braid representation has a unique exponent sum.

4734: 

4735: \begin{conj}(Jones's conjecture)\label{jcc}

4736: \def\labelenumi{\theenumi)}

4737: \begin{enumerate}

4738: \item (weaker version)\label{weV}

4739: If $\bt,\bt'\in B_n$ satisfy $\hat\bt=\hat\bt'=L$ and $n=b(L)$,

4740: then $[\bt]=[\bt']=:w_{\min}(L)$.

4741: \item (stronger version) Part \reference{weV} holds, and if

4742: $\bt''\in B_n'$ for $n'>n$ has $\hat\bt''=L$, then 

4743: \begin{eqn}\label{stz}

4744: \big|[\bt'']-w_{\min}(L)\big|\le n'-n\,.

4745: \end{eqn}

4746: \end{enumerate}

4747: \end{conj}

4748: 

4749: It was observed in \cite{posex_bcr} that counterexamples

4750: to the conjecture would make MWF and \em{all} of its cabled

4751: versions unsharp. (In \cite{posex_bcr} the weaker version was

4752: focussed on, but the same arguments address the stronger version

4753: too.) Thus, for example, corollary \reference{cft} can also

4754: be regarded as a partial solution to Jones's conjecture.

4755: Similarly, the work in \S\reference{S2.2} shows:

4756: 

4757: \begin{corr}

4758: If a link $L$ is the closure of a positive braid, and $b(L)\le 4$,

4759: then the strong Jones conjecture is true for $L$. 

4760: \end{corr}

4761: 

4762: \proof We use theorem \reference{thmwf}, with the remark that

4763: the exceptional braid words therein were shown to have braid

4764: index 4 using the 2-cabled MWF. \qed

4765: 

4766: For 3-braids the weaker version was known to be true again

4767: from Birman-Menasco's classification result. We will show now

4768: the stronger version, as a consequence of the description in

4769: \cite{3br} of 3-braid links with unsharp MWF inequality (and thus

4770: with a much simpler proof than appealing to Birman-Menasco). See

4771: also remark \reference{rJC}.

4772: 

4773: \begin{theorem}\label{thJC}

4774: The stronger version of Jones's conjecture holds for 3-braid links.

4775: \end{theorem}

4776: 

4777: \noindent{\bf Convention.}

4778: The letter $\Dl$, with an integer subscript $n$, is used in

4779: \S\ref{SJS},

4780: in deviation from other sections, exclusively for the half-twist

4781: braid on $n$ strings, and \em{not} for the Alexander polynomial.

4782: (The Alexander polynomial will not appear in \S\reference{SJS}.)

4783: 

4784: \proof[of theorem \reference{thJC}]

4785: As explained, it suffices to deal only with the 3-braid links

4786: of unsharp MWF inequality. In \cite{3br} these links were

4787: described fully. If $\mu=\sg_1^{6a\pm 1}\sg_2^{\mp 1}$,

4788: then for the Birman dual $\bt=\mu^*$ of $\mu$ (see definition

4789: \ref{DEFB}) we have $\hat\mu^*\ne \hat\mu$ when $6a\pm 1>6$, but

4790: $K=\hat\bt$ has the skein polynomial of the $(2,6a\pm 1)$-torus

4791: knot $\hat \mu$. In \cite{3br} is was proved that these knots $K$

4792: form the full list of 3-braid links with unsharp MWF inequality.

4793:  

4794: Consider first %(with a slight change of notation)

4795: the knots $K=\hat\bt$ with $\bt=\mu^*$ for

4796: $\mu=\sg_1^{6a+1}\sg_2^{-1}$ and $a>0$. It is easy

4797: to bring $\bt$ into Xu's normal form, and to observe

4798: that it has exactly one negative band. This exhibits

4799: a Seifert ribbon (see \cite{Rudolph}) of smaller genus

4800: than $g(K)$, i.e. $g_s(K)<g(K)$ (see also the proof of

4801: theorem \reference{thqp}). Now from Xu's form one sees $w_{\min}

4802: (K)=[\bt]=2g(K)$. Take this for a moment as a definition of

4803: $w_{\min}(K)$; this writing will be justified when we

4804: show that the writhe is unique. (One could also quote

4805: Birman-Menasco here, but the complexity of their argument

4806: is unnecessary.) So 

4807: \begin{eqn}\label{iopa}

4808: 1-\chi_s(L)<w_{\min}(L)\,.

4809: \end{eqn}

4810: Using \eqref{mq},

4811: we see that if \eqref{stz} is violated for some $\bt''\in B_{n'}$,

4812: then $[\bt'']=w_{\min}(L)+2+(n'-3)$. However, by the Rudolph-%

4813: Bennequin inequality \cite{Rudolph2} (see also \cite{pos}),

4814: we would have then 

4815: \[

4816: 1-\chi_s(L)\ge [\bt'']-n'+1=w_{\min}(L)\,,

4817: \]

4818: in contradiction to \eqref{iopa}.

4819: 

4820: The effort focusses on the other braids $\bt=\mu^*$

4821: for $\mu=\sg_1^{6a-1}\sg_2$.

4822: In this case we consider $!K$, whose Xu form has only negative

4823: bands. The inequalities \eqref{mq} show that we must rule out a

4824: braid representation $\bt''\in B_{n'}$ of $!K$ with $[\bt'']=w_{\min}

4825: (!K)+2+(n'-3)$, where (we set in the same \em{a posteriori} to

4826: be justified manner) $w_{\min}(!K)=-2g(K)-2$.

4827: 

4828: We prohibit $\bt''$ by evaluating the 2-cable skein polynomial

4829: $P(\hat\gm_a)$ of $!K$ for

4830: \begin{eqn}\label{gma}

4831: \gm_a\,=\,[4354]\cdot [2132]^{6a-1}\,(\Dl_6^2)^{-2a}\,\in\,B_6,

4832: \end{eqn}

4833: with $a>1$ and $\Dl_6^2=[12345]^6$ generating the center of $B_6$,

4834: and showing that the polynomial has non-zero terms in $v$-degree

4835: $[\gm_a]-5$ or $[\gm_a]-3$. This may appear a banality, but in

4836: fact requires a substantial use of Jones' work in \cite{Jones}.

4837: 

4838: Consider the more general 6-braids

4839: \[

4840: \bt_{k,l}\,:=\,[4354]\cdot [2132]^k\,\Dl_6^{2l}\,.

4841: \]

4842: (We are of course only interested in the special case $k=6a-1$,

4843: $l=-2a$ and $a>1$, but it is useful to treat the 2-parameter family

4844: first.)

4845: 

4846: The skein polynomial of $\hat\bt_{k,l}$ can be evaluated using

4847: the representation theory in \cite{Jones}. We adopt the convention

4848: that \em{all references} to lemmas, page and paragraph numbers,

4849: equations of the form ($x.y$), etc. for the rest of this proof

4850: are understood to be \em{to Jones's paper}, unless noted otherwise.

4851: 

4852: Jones's % paper gives a clear and detailed account on the

4853: % calculation of the skein polynomial. His

4854: version of the skein polynomial, 

4855: \[

4856: X(q,\lm)=P(v,z)\mbox{\quad with $v=\sqrt{\lm q}$\ \ and

4857: \ $z=\sqrt{q}-1/\sqrt{q}$\,},

4858: \]

4859: can be evaluated on a closed

4860: $n$-braid $\bt$ by a weighted sum, with weights $W_Y$ in $\lm$ and

4861: $q$, of traces (in $q$) of irreducible representations (irreps) $\pi_Y$

4862: of $B_n$, indexed by Young diagrams (or tableaux\footnote{Here

4863: consistently `tableau' is used as a synonym for (Young) diagram,

4864: i.e. with no additional information attached to it.}) $Y$, or

4865: equivalently by partitions of $n$. (A partition of $n$ is a

4866: tuple $(n_1,\dots,n_k)$ with $n_k>0$, $n_i\ge n_{i+1}$ and

4867: $\sum n_i=n$.) We may identify the Young diagram with its

4868: partition, counting partitions in horizontal rows. For

4869: example the partition $(4)=\Young{4}$ means one row (trivial

4870: representation), while $(1111)=\Young{1,1,1,1}$ means one column

4871: (parity representation); thus $\pi_{3,1}=\pi_{\Young[1.5mm]

4872: {3,1}}$. Since the calculation of $W_Y$ was given in

4873: \cite{Jones}, we will not repeat it in detail. We will

4874: also deal (mostly) with $Y$ where the calculation of

4875: $\pi_Y$ is explained in \cite{Jones}.

4876: 

4877: Using Definition 6.1 (of \cite{Jones}), (5.5) and the formula

4878: for $W_Y(q,\lm)$ from p. 347 top, one has for a 6-braid $\bt$

4879: of exponent sum $e=[\bt]$,

4880: \begin{eqn}\label{STR}

4881: X_{\hat\bt}(q,\lm)\,=\,-\sqrt{\lm}^{e-5}\,\sum_{Y\vdash 6}

4882: \wt W_Y(q,\lm)\,\tr \pi_Y(\bt)\,.

4883: \end{eqn}

4884: Here $\wt W_Y$ denote the slightly rescaled weights

4885: \[

4886: \wt W_Y(q,\lm)\,=\,\frac{R_{Y}(q,\lm)}{Q_Y(q)}\,\cdot\,

4887: \frac{1-q}{1-\lm q}\,,

4888: \]

4889: with $R_Y$ being specified on p.\ 347 after Figure 5.6, and

4890: $Q_Y(q)$ being the hook length product term from p. 346 middle.

4891: The symbol `$\mathord{\vdash}$' is taken from partition theorists

4892: and is used to mean here that $Y$ is a Young tableau of $6$ boxes

4893: (or equivalently a partition of $6$).

4894: In order to avoid denominators it is useful to multiply

4895: \eqref{STR} by $(1-q^2)\cdot\dots\cdot(1-q^6)$, so

4896: \begin{eqn}\label{STR2}

4897: -(1-q^2)\cdot\dots\cdot(1-q^6)X_{\hat\bt}\,=\,

4898: \sqrt{\lm}^{e-5}\,\sum_{Y\vdash 6}

4899: \wh W_Y(q,\lm)\,\tr \pi_Y(\bt)\,,

4900: \end{eqn}

4901: with 

4902: \[

4903: \wh W_Y(q,\lm)\,=\,\frac{R_Y(q,\lm)}{1-\lm q}\,\cdot\,

4904: \frac{(1-q)(1-q^2)\cdot\dots\cdot(1-q^6)}{Q_Y(q)}\,

4905: \]

4906: becoming a Laurent polynomial in $\lm$ and $q$\,.

4907: 

4908: Note that $\pi_Y$ are representations that involve only the

4909: variable $q$, not $\lm$. The obvious question is how to

4910: evaluate their traces.

4911: 

4912: Consider first 

4913: \[

4914: \bt_k=\bt_{k,0}\,=\,[4354]\,[2312]^k\,.

4915: \]

4916: 

4917: There are 11 Young diagrams $Y$ of 6 boxes. Now $\bt_k\in B_4\subset

4918: B_6$, and according to p. 340 top, the restriction of $\pi_Y$ to $B_4$

4919: splits into a direct sum of representations $\pi_{Y'}$ indexed by

4920: 4-box Young diagrams $Y'$, which we name for shorthand

4921: \begin{eqn}\label{shh}

4922: A=\Young{1,1,1,1}\quad

4923: B=\Young{2,1,1}\quad

4924: C=\Young{2,2}\quad

4925: D=\Young{3,1}\quad

4926: E=\Young{4}\,.

4927: \end{eqn}

4928: 

4929: The representations $\pi_{Y'}$ are clarified completely in \S 8

4930: (see also \S\reference{SUn} in \em{this} paper below). Denote by

4931: $\psi_{n-1}$ the (reduced) $(n-1)$-dimensional representation

4932: of $B_n$; see \S 2. Let us also write $-1$ for the parity

4933: representation $(-1)^{[\,.\,]}$, and $-\rho=-1\otimes \rho$ for the

4934: direct (tensor, or Kronecker) product of a representation $\rho$ with

4935: the parity. Then $\pi_A=-1$, $\pi_E$ is given by $q^{[\,.\,]}$,

4936: $\pi_B=-\psi_3$ (Note 5.7; the sign disappears here, though,

4937: because for us always $e=[\bt]$ is even), $\pi_D=-\bigwedge^2\psi_3$

4938: and $\pi_E=\pi_{\Young[1.5mm]{2,1}}\circ\ol{\ry{0.6em}\es}$,

4939: with $\ol{\es\ry{0.6em}}$ being the homomorphism $B_4\to B_3$

4940: given by $\ol{\sg_{1,2,3}\ry{0.6em}}=\sg_{1,2,1}$ (p. 355).

4941: Also $\pi_{\Young[1.5mm]{2,1}}=-\psi_2$.

4942: 

4943: Let us write below $\ol{Y}$ for the \em{transposed} (or \em{dual})

4944: Young diagram to $Y$, given by exchanging rows and columns.

4945: For example, for $Y=\Young{3,1}$, we have $\ol{Y}=\Young{2,1,1}$.

4946: The relation between $\pi_Y$ and $\pi_{\ol{Y}}$ is given in Note

4947: 4.6.

4948: 

4949: Now $\psi_3(\ap)$ and $\psi_2(\bar\ap)$ for 

4950: \[

4951: \ap=[2132]

4952: \]

4953: are easy to

4954: calculate (see \S 2). We find former's eigenvalues to be $\pm t$ and

4955: $-t^2$, and latter's $t$ and $t^3$. In particular, both matrices are

4956: diagonalizable (because the eigenvalues are distinct for generic $t$).

4957: Setting $t=q$ as in Note 5.7, then we

4958: have the following table of eigenvalues of the $\pi_{Y'}(\ap)$

4959: \[

4960: \begin{array}{|c|c|c|c|c|}

4961: \hline

4962: \ry{1.3em}A & B & C & D & E \\[1mm]

4963: \hline

4964: \ry{1.5em}1& -q   & q^3 & -q^3 & q^4 \\

4965: 	 &  q   & q   &  q^3 &     \\

4966: 	 & -q^2 &     & -q^2 &     \\[2mm]

4967: \hline

4968: \end{array}

4969: \kern2cm

4970: \begin{array}{c|*{7}{|c}}

4971: \ry{1.2em}i & 1 & 2 & 3 & 4 & 5 & 6 & 7 \\[1.5mm]

4972: \hline  

4973: \ry{1.2em}\dl_i & 1 & q^4 & q & -q & -q^2 & q^3 & -q^3\\[1mm]

4974: \end{array}

4975: \]

4976: Let us number the 7 possible eigenvalues of $\pi_Y'(\ap)$ by

4977: $\dl_i$ as shown on the right. 

4978: 

4979: Thus $\pi_Y(\ap)$ are all diagonalizable, with eigenvalues $\dl_i$.

4980: The multiplicities of $\pi_{Y'}$ in $\pi_Y$ are also easy to

4981: calculate. They are the number of descending paths from $Y'$ to

4982: $Y$ in the Figure 3.3 (continued one more row to the bottom).

4983: Table \reference{tab1} shows the multiplicities of $\pi_{Y'}$

4984: (in the shorthand of \eqref{shh}) in $\pi_Y$ and the resulting ones,

4985: which we write $M(i,Y)$, of the eigenvalues $\dl_i$ in $\pi_Y(\ap)$.

4986: 

4987: \begin{table}[ptb]

4988: \captionwidth\vsize\relax

4989: \newpage

4990: \vbox to \textheight{\vfil

4991: \rottab{ 

4992: \hbox to \vsize{%\hss%\footnotesize

4993: \hss

4994: % \setbox\@tempboxa=\hbox{%

4995: % \textwidth\vsize\relax

4996: $\displaystyle

4997: \begin{array}{|c||*{11}{c|}}

4998: \hline\ry{1.0cm}%

4999: Y & \Young{1,1,1,1,1,1} & % 1

5000: \Young{2,1,1,1,1} &       % 2

5001: \Young{2,2,1,1} &         % 3

5002: \Young{2,2,2} &           % 4

5003: \Young{3,1,1,1} &         % 5

5004: \Young{3,2,1} &           % 6

5005: \Young{3,3} &             % 7

5006: \Young{4,1,1} &           % 8

5007: \Young{4,2} &             % 9

5008: \Young{5,1} &             % 10

5009: \Young{6} \\[6mm]         % 11

5010: \hline\ry{1.6em}%

5011: Y' & A & 2A+B & A+2B+C & B+C & A+2B+D & 2B+2C+2D & C+D & B+2D+E & C+2D+E

5012: & D+2E & E \\[2mm]

5013: \hline\ry{1.6em}%

5014: %  PARTITION #7 & #8 ARE SWOPPED FROM MY NOTES

5015: %      #1 #2  #3  #4  #5  #6  #8  #7  #9  #10 #11

5016: 1    & 1 & 2 & 1 &   & 1 &   &   &   &   &   &   \\

5017: q^4  &   &   &   &   &   &   &   & 1 & 1 & 2 & 1 \\

5018: q    &   & 1 & 3 & 2 & 2 & 4 & 1 & 1 & 1 &   &   \\

5019: -q   &   & 1 & 2 & 1 & 2 & 2 &   & 1 &   &   &   \\

5020: -q^2 &   & 1 & 2 & 1 & 3 & 4 & 1 & 3 & 2 & 1 &   \\

5021: q^3  &   &   & 1 & 1 & 1 & 4 & 2 & 2 & 3 & 1 &   \\

5022: -q^3 &   &   &   &   & 1 & 2 & 1 & 2 & 2 & 1 &   \\[2mm]

5023: \hline\ry{1.6em}%

5024: d    & 1 & 5 & 9 & 5 & 10& 16& 5 &10 & 9 & 5 & 1 \\

5025: r    & 0 & 1 & 3 & 2 & 4 & 8 & 3 & 6 & 6 & 4 & 1 \\

5026: 30r/d& 0 & 6 &10 &12 & 12& 15& 18& 18& 20& 24& 30\\[2mm]

5027: \hline

5028: \end{array}%

5029: $%

5030: % }%

5031: % \@tempdima0.3\vsize\relax\advance\@tempdima by -0.3\wd\@tempboxa\relax

5032: % \kern\@tempdima

5033: % \hss

5034: % \copy\@tempboxa

5035: \hss}%

5036: }{This table displays the Young tableaux $Y$ of 6 boxes, the

5037: decomposition into $\pi_{Y'}$ for 4-box Young tableaux $Y'$ of the

5038: sub-representation of $B_4$ in $\pi_Y$ (writing $Y'$ for $\pi_{Y'}$),

5039: the multiplicities $M(i,Y)$

5040: of the 7 possible eigenvalues $\dl_i$ of $\pi_Y(\ap)$, and the

5041: quantities $d$ ($=\dim \pi_Y$) and $r$ (rank of the idempotent

5042: $e_i$) occurring in Jones's lemma 9.3.\label{tab1}}{table}

5043: \vss}

5044: \newpage

5045: \end{table}

5046: 

5047: Since with $\pi_{Y'}(\ap)$ also all $\pi_Y(\ap)$ are

5048: diagonalizable, we have

5049: \begin{eqn}\label{XX}

5050: \tr \pi_Y(\bt_k)\,=\,\tr \pi_Y(\ap^k\cdot [4354])\,=\,

5051: \sum_{i=1}^7\,c(i,Y)\,\dl_i^k\,,

5052: \end{eqn}

5053: with $c(i,Y)$ given as follows. Consider $\pi_Y([4354])$ in

5054: the basis of $\pi_Y$ that diagonalizes $\pi_Y(\ap)$. Then to

5055: obtain $c(i,Y)$, sum the $(j,j)$-entries of the matrix

5056: of $\pi_Y([4354])$ over rows/columns $j$, for which the

5057: $(j,j)$-entry of $\pi_Y(\ap)$ is $\dl_i$.

5058: 

5059: So the problem to evaluate $\tr \pi_Y(\bt_k)$ transforms into

5060: the one to determine $c(i,Y)$. There are \em{a priori} 77

5061: of those, given by combining $7$ eigenvalues $\dl_i$ with

5062: 11 Young diagrams $Y$. However, one immediately notes that

5063: clearly $c(i,Y)=0$ when $M(i,Y)=0$. This leaves 45 of the 77 values.

5064: 

5065: If we can calculate $\tr \pi_Y$ for general 6-braids, then we

5066: can use \eqref{XX} as a linear equation for $c(i,Y)$.

5067: If $\pi_Y(\ap)$ has $l_Y$ different eigenvalues, we can

5068: determine $c(i,Y)$ for that $Y$ by calculating the l.h.s. of

5069: \eqref{XX} for $l_Y$ different values of $k$ and solving for $c(i,Y)$.

5070: (Always $l_Y\le 6$, as evident from table \reference{tab1}.)

5071: 

5072: In case $Y$ is one of the one-hook diagrams like $\Young{4,1,1}$,

5073: then $\pi_Y$ is by p. 354 bottom (a tensor product of a parity,

5074: which disappears at even exponent sum, with) an exterior power

5075: of $\psi_5$. Thus, as in \S 7 (of Jones's paper), one can evaluate

5076: $\tr \pi_Y(\bt_k)$ from the characteristic polynomial of the Burau

5077: matrix $\psi_5(\bt_k)$. We determined this way the corresponding

5078: $c(i,Y)$, and 

5079: verified them with a few extra values of $k$ in \eqref{XX}. While

5080: it is clear that all $c(i,Y)$ should be rational expressions in $q$,

5081: we expected them to be in fact Laurent polynomials. So we were a bit

5082: startled by the denominators $1+q^2$. However, according to p. 343

5083: top, the Hecke algebra may degenerate at roots of unity $q$, which

5084: justifies at least cyclotomic polynomials as denominators.

5085: 

5086: \begin{table}[ptb]

5087: \captionwidth\vsize\relax

5088: \newpage

5089: \vbox to \textheight{\vfil

5090: \rottab{ 

5091: \hbox to \vsize{%\hss%\footnotesize

5092: \hss

5093: % \setbox\@tempboxa=\hbox{%

5094: % \textwidth\vsize\relax

5095: $\displaystyle

5096: \begin{array}{|c||*{11}{c|}}

5097: \hline\ry{1.0cm}%

5098: Y & \Young{1,1,1,1,1,1} & % 1

5099: \Young{2,1,1,1,1} &       % 2

5100: \Young{2,2,1,1} &         % 3

5101: \Young{2,2,2} &           % 4

5102: \Young{3,1,1,1} &         % 5

5103: \Young{3,2,1} &           % 6

5104: \Young{3,3} &             % 7

5105: \Young{4,1,1} &           % 8

5106: \Young{4,2} &             % 9

5107: \Young{5,1} &             % 10

5108: \Young{6} \\[6mm]         % 11

5109: \hline\ry{1.6em}%

5110: %  PARTITION #7 & #8 ARE SWOPPED FROM MY NOTES

5111: %      #1 #2  #3  #4  #5  #6  #8  #7  #9  #10 #11

5112: 1 & 1 & -\left( \frac{q^4}{1 + q^2} \right)  & \frac{q^7}

5113:    {\left( 1 + q^2 \right) \,\left( 1 + q + q^2 \right) } & 0 & 0 & 0 & 0 & \

5114: 0 & 0 & 0 & 0 \\

5115: %

5116: q^4  & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & \frac{q}

5117:    {\left( 1 + q^2 \right) \,\left( 1 + q + q^2 \right) } & -\left( \frac{q^2}{1 + q^2} \right)  & q^4 \\

5118: %

5119: q    & 0 & \frac{1}{2} & \frac{-\left( 

5120:        \left( -1 + q \right) \,q \right) }{2} & q + \frac{q^3}{2} & \frac{q^2 + 2\,q^4}{-2 - 2\,q^2} & \frac{q\,\left( 1 - 4\,q + q^2 \right) }

5121:     {2} - c(6,\Young[1.5mm]{ 3,2,1} ) & 0 & \frac{q^6}{2 + 2\,q^2} & \frac{q^6}

5122:    {1 + q + q^2} & 0 & 0 \\[2mm]

5123: %

5124: -q &   0 & \frac{1}{2} & \frac{-\left( 

5125:        \left( -1 + q \right) \,q \right) }{2} & \frac{q^3}{2} & \frac{q^2 + 

5126:      2\,q^4}{-2 - 2\,q^2} & \frac{{\left( 1 - q \right) }^2\,q}{2} - 

5127:    \frac{q^2}{1 + q} - c(6,\Young[1.5mm]{ 3,2,1} ) & 0 & \frac{q^6}

5128:    {2 + 2\,q^2} & 0 & 0 & 0 \\

5129: %

5130: -q^2 &   0 & \frac{1}{1 + q^2} & \frac{q\,

5131:      \left( 1 + \left( -1 + q \right) \,q \right) }{1 + q^2} & -q^2 & 

5132: -q^2 & q\,\left( 1 + \left( -1 + q \right) \,q \right)  & -q^2 & -q^2 & \frac{q^3\,\left( 1 - q + q^2 \right) }{1 + q^2} & \frac{q^6}

5133:    {1 + q^2} & 0 \\[2mm]

5134:    %

5135: q^3 &   0 & 0 & \frac{1}{1 + q + q^2} & 0 & \frac{1}

5136:    {2 + 2\,q^2} & c(6,\Young[1.5mm]{ 3,2,1} ) & \frac{q}{2} + q^3 & \frac{-\left( q^2\,

5137:        \left( 2 + q^2 \right)  \right) }{2\,\left( 1 + q^2 \right) } & \frac{\left( -1 + q \right) \,q^2}{2} & \frac{q^4}

5138:    {2} & 0 \\[3mm]

5139: -q^3 &   0 & 0 & 0 & 0 & \frac{1}{2 + 2\,q^2} & \frac{q^2}{1 + q} + 

5140:    c(6,\Young[1.5mm]{ 3,2,1} ) & \frac{q}{2} & \frac{-\left( q^2\,

5141:        \left( 2 + q^2 \right)  \right) }{2\,\left( 1 + q^2 \right) } & \frac{\left( -1 + q \right) \,q^2}{2} & \frac{q^4}{2} & 0 \\[2mm]

5142: \hline

5143: \end{array}%

5144: $%

5145: % }%

5146: % \@tempdima0.3\vsize\relax\advance\@tempdima by -0.3\wd\@tempboxa\relax

5147: % \kern\@tempdima

5148: % \hss

5149: % \copy\@tempboxa

5150: \hss}%

5151: }{This table shows the values $c(i,Y)$ of \eqref{XX} for Young tableaux

5152: $Y$ of 6 boxes and the 7 possible eigenvalues $\dl_i$ of $\pi_Y(\ap)$.

5153: \label{tab2}}{table}

5154: \vss}

5155: \newpage

5156: \end{table}

5157: 

5158: For $Y=\Young{2,2,2}$, $\pi_Y$ was written down directly on p. 362,

5159: and for its transposed (dual) Young diagram $\Young{3,3}$, one

5160: uses Note 4.6. 

5161: 

5162: There remain 3 representations, for $Y=\Young{3,2,1}$,

5163: $\Young{4,2}$, and its dual $\Young{2,2,1,1}$, with a total of

5164: 15 unknown $c(i,Y)$. These were more complicated to find, since

5165: we knew of no way to evaluate $\tr \pi_Y$ directly. To help

5166: ourselves, first observe that we have, for fixed $Y$, the trace

5167: identities

5168: \begin{eqn}\label{tri}

5169: \sum_{i=1}^7\,c(i,Y)\,=\,\tr \pi_Y([4354])\,=\,

5170: \tr \pi_Y([2132])\,=\,\sum_{i=1}^7 M(i,Y)\dl_i\,,

5171: \end{eqn}

5172: which again give a linear condition on the $c(i,Y)$. To find

5173: further identities, we used \eqref{STR2} in a ``backward'' manner.

5174: We calculated for small\footnote{In the parametrization $P(v,z)$,

5175: used by Morton-Short, unlike for $X$, the coefficients of

5176: $P(\hat\bt_{k,l})$ become quickly large and produce machine

5177: size integer overflows. In particular, we could not calculate

5178: correctly polynomials for $|l|>2$.} $k,l$ ($k$ odd, $|k|\le 5$,

5179: $|l|\le 1$) the polynomial on the left using Morton-Short's

5180: program \cite{MorSho2}. We substituted the known $c(i,Y)$ on

5181: the right of \eqref{STR2}, obtaining thus linear conditions

5182: for the yet unknown $c(i,Y)$. To determine the coefficients,

5183: it remains to understand the effect on $\tr\pi_Y$ of multiplying

5184: with the full twist $\Dl_6^2$. This was, however, done also by

5185: Jones in \S 9, lemma 9.3:

5186: \begin{eqn}\label{l93}

5187: \pi_Y(\Dl_n^2)\,=\,q^{rn(n-1)/d}\,Id_{\pi_Y}\,.

5188: \end{eqn}

5189: For given $Y$, the number $r=r_Y$ is calculated as in lemma 9.1

5190: from figure 3.3, and $d=d_Y=\dim \pi_Y$ more easily by the hook

5191: length formula on p. 341. Call $e_Y=n(n-1)r/d$ (with $n=6$) the

5192: exponent\footnote{This is not to be confused with the variable

5193: $e$, which we use for exponent sum of a braid, or with the

5194: idempotent $e_i$ from Jones's lemma 9.1.} of $q$ on the right

5195: hand-side of \eqref{l93}. These values are given in table

5196: \ref{tab1}. Two simple checks are $\sum\limits_{Y\vdash n}d_Y^2=

5197: n!$ (because the multiplicity of each irrep in the Hecke algebra

5198: equals its dimension), and that $e_Y$ are integers (Remark on p. 358

5199: bottom) and satisfy $e_{Y}+e_{\ol{Y}}=n(n-1)$ for all $Y$

5200: (because of Note 4.6). So \eqref{STR2} gets

5201: \begin{eqn}\label{STR3}

5202: -

5203: (1-q)\cdot\dots\cdot(1-q^6)\,X(\hat\bt_{k,l})(q,\lm)\,=\,

5204: \sqrt{\lm}^{e-5}\,\sum_{Y\vdash 6}\,\wh W_Y(q,\lm)\,

5205: q^{l\cdot e_Y}\,\sum_{i=1}^7\,c(i,Y)\,\dl_i^k\,,

5206: \end{eqn}

5207: with $e=[\bt_{k,l}]=4+4k+30l$.

5208: 

5209: Actually, each polynomial

5210: $X(\hat\bt_{k,l})$ gives 6 equations for $c(i,Y)$, because

5211: there are 6 relevant $\lm$-coefficients on both hand-sides

5212: of \eqref{STR2} (in degrees $\ffrac{e-5}{2}$, $\ffrac{e-3}{2}$,

5213: \dots, $\ffrac{e+5}{2}$\,; it is helpful to multiply again by

5214: $(1+q^2)$ to get disposed of the denominators of the known

5215: $c(i,Y)$). We have with 18 polynomials 111 equations (6 equations per

5216: polynomial plus the three relevant trace equations \eqref{tri}).

5217: Still the resulting system was too hard to solve by computer, using 

5218: MATHEMATI\-CA\TM{} \cite{Wolfram}, since its coefficients are

5219: (Laurent) polynomials in $q$, with dozens of terms each.

5220: 

5221: However, substituting some (rational) values of $q$, the system

5222: can be solved immediately. We used this to check first the rank of

5223: the matrix (i.e. which equations are linearly redundant). Again we

5224: were surprised that for 15 variables $c(i,Y)$ the rank was only

5225: $14$. This, however, can be explained from our restraint to odd

5226: $k$ (which we chose for some, purely technical,

5227: component number concerns in the calculation

5228: with Morton-Short's program). Whenever two opposite eigenvalues

5229: $\dl_i$ and $\dl_{i'}=-\dl_i$ occur in $\pi_Y(\ap)$, the equations

5230: \eqref{STR2} for odd $k$ can detect only $c(i,Y)-c(i',Y)$. The

5231: trace equations \eqref{tri}, which involve $c(i,Y)$ and $c(i',Y)$

5232: with the same sign, remedy the shortcoming for $\Young{4,2}$ and

5233: its transposed diagram, but for $\Young{3,2,1}$ we have two pairs

5234: $(i,i')$ of opposite eigenvalues, so we still lose one dimension.

5235: This is, however, not really a problem, because in the

5236: braids $\gm_a=\bt_{6a-1,-2a}$ of \eqref{gma} we

5237: need for our proof, $k=6a-1$ is always odd, so we need only

5238: $c(i,Y)-c(i',Y)$ to evaluate the polynomial in \eqref{STR2}.

5239: 

5240: We used the special evaluations to select equations that give

5241: the full matrix rank, and to guess the formula for general $q$

5242: for some of the $c(i,Y)$. (The ones we already found suggest that

5243: these formulas should not be so complicated.) Substituting these

5244: (yet potential) solutions too, gives an even simpler linear system

5245: for the still unknown $c(i,Y)$, which then could be solved in $q$.

5246: 

5247: Since we know that our matrix has rank 14, it is enough to check

5248: our solution (up to the 1-dimentional ambiguity, which will

5249: disappear in \eqref{STR2}) with the 111 equations we have. The

5250: result was confirmed, and is shown in table \ref{tab2}.

5251: 

5252: With all $c(i,Y)$ determined, the main work is done.

5253: So far we can evaluate $X(\hat\bt_{k,l})$ for odd $k$.

5254: Multiplying \eqref{STR3} by $Z=(1+q)(1+q^2)(1+q+q^2)$ to clear

5255: all denominators in the $c(i,Y)$, normalizing and taking

5256: coefficients in $\lm^m$ for $m=0,\dots,5$, we have

5257: (with $e_Y$ being the exponents of $q$ in \eqref{l93} and

5258: $e=[\bt_{k,l}]$)

5259: \begin{eqn}\label{OPI}

5260: \left[-\frac{1}{\sqrt{\lm}^{e-5}}\,\cdot\,Z\,\cdot\,

5261: (1-q)\cdot\dots\cdot(1-q^6)\,

5262: X_{\hat\bt_{k,l}}(q,\lm)\,\right]_{\lm^m}\,=\,

5263: \sum_{Y\vdash 6}\,\sum_{i=1}^7\,

5264: (Z\cdot c(i,Y))\,\dl_i^k\,

5265: q^{l\cdot e_Y}\,\bigl[\wh W_Y(q,\lm)\,\bigr]_{\lm^m}\,.

5266: \end{eqn}

5267: 

5268: Recall that for our proof it is enough to show that this term

5269: becomes non-zero for $m=0$ or $m=1$, when $k=6a-1$ and $l=-2a$

5270: for an integer $a>1$ (and $\bt_{k,l}=\gm_a$ in \eqref{gma}).

5271: Now, for odd $k$ (and fixed $Y$), we can group

5272: the sum over $7$ terms $Z\cdot c(i,Y)$ into 5 terms $\tl c(i,Y)$

5273: accounting for $\dl_3=-\dl_4$ and $\dl_6=-\dl_7$, and thus

5274: excluding $i=4,7$. So the above sum in \eqref{OPI} becomes

5275: \begin{eqn}\label{yy}

5276: \sum_{Y\vdash 6}\,\sum_{\scbox{\shortstack{$i=1$\\$i\ne 4$}}}^6

5277: \tl c(i,Y)\dl_i^k\,q^{l\cdot e_Y}\,\bigl[\wh W_Y(q,\lm)\,\bigr]

5278: _{\lm^m}\,.

5279: \end{eqn}

5280: (Now in the $\tl c(i,Y)$, the 1-degree ambiguity of $c(i,Y)$ for

5281: $Y=\Young{3,2,1}$, as explained above, cancels out.)

5282: 

5283: Among the 55 possible $(i,Y)$ (with $i\ne 4,7$), only 21

5284: of the $\tl c(i,Y)$ are non-zero. It turns out that for $m=0$,

5285: when $k=6a-1$, $l=-2a$ and $a>1$, there is a unique term among

5286: the 21 summands in \eqref{yy} whose minimal degree in $q$ is the

5287: smallest (it is $14-36a$). Thus $[P(\hat\bt_{k,l})]_{v^{e-5}}\ne 0$.

5288: We calculated the polynomial with Morton-Short's program for $a=2$

5289: (where the calculation was still feasible), and it confirmed that

5290: all 6 $v$-terms appear. We also calulated that for $m=0$ and

5291: $a=0,1$ there are two terms of smallest minimal degree. This

5292: better ought to be so, because in that case $\hat\gm_{a}$ are just

5293: a 2-cable of the unknot and $!5_1$, resp., and the coefficients

5294: for $m=0,1$ in \eqref{yy} must be $0$, which we checked once

5295: more separately. (Also we found that for $m=1$ there are two

5296: terms of smallest minimal degree in \eqref{yy} for all $a\ge 0$.)

5297: With this the proof of theorem \reference{thJC} is complete.

5298: 

5299: Let us finally say that the computer part of the calculation owed

5300: a lot to the use of MATHEMATI\-CA. While, if done properly, it could

5301: be carried out in a few minutes, it required a week of work to

5302: find the way of skillfully programming MATHEMATI\-CA

5303: to do all the separate steps in an efficient way. \qed

5304: 

5305: \begin{rem}

5306: Note that one could handle the cases $\mu=\sg_1^{6a+1}\sg_2^{-1}$

5307: from the beginning of the proof also using the representation

5308: theoretic argument, by looking at $m=4,5$ for $a<0$. We waived on

5309: this investigation, though, since the proof for $a>0$ was laborious

5310: enough.

5311: \end{rem}

5312: 

5313: \subsection{Unitarity of the Burau representation\label{SUn}}

5314: 

5315: In the following $q$ and $t$ are unit norm complex numbers.

5316: We define $\arg (e^{is}):=s\bmod 2\pi$ for $s\in\bR$.

5317: We continue using the formalism of Young tableaux,

5318: the representations $\pi_Y$ and the notations $X(q,\lm)$

5319: and $W_Y$ of the proof of theorem \reference{thJC}.

5320: 

5321: For the Alexander polynomial $\Dl$ (we resume the

5322: notational convention from before \S\ref{SJS}), as well as

5323: for $3$- and $4$-braids, the

5324: representations $\pi_Y$ are given by Burau representations. We note

5325: (again, and more explicitly) the following descriptions of $\pi_Y$

5326: in terms of the Burau representation $\psi_n$ given in \cite{Jones}.

5327: Again the indexing is chosen so that $\psi_{n-1}$ is the reduced

5328: $(n-1)$-dimensional representation of $B_n$, and by $-\rho$ we denote

5329: the direct product of $\rho$ with the parity representation.

5330: 

5331: \Youngunitlength1.5mm

5332: As before, $e$ stands for the exponent sum of a braid $\bt$.

5333: For $3$-braids we have the following properties (with

5334: reference to the explanation in \cite{Jones}):

5335: \begin{enumerate}

5336: \item $\pi_{\Young{2,1}}(\bt)(q)=(-1)^e\psi_2(\bt)(q)=

5337: q^e\psi_2(\bt^{-1})(q)$ (because of

5338: row-column symmetry; see Note 4.6).

5339: \item $\pi_{\Young{1,1,1}}(\bt)=(-1)^e$ and

5340: $\pi_{\Young{3}}(\bt)=q^e$ (Note 4.7)

5341: \end{enumerate}

5342: 

5343: For $4$-braids we have:

5344: \begin{enumerate}

5345: \item $\pi_{\Young{2,1,1}}=-\psi_3$. So

5346: $\pi_{\Young{2,1,1}}(\bt)=(-1)^e\psi_3(\bt)$. (Note 5.7)

5347: \item $\pi_{\Young{3,1}}(\bt)(q)=q^e\psi_3(\bt^{-1})(q)$ (because of

5348: row-column symmetry; see Note 4.6). Also 

5349: \[

5350: \pi_{\Young{3,1}}=-\pi_{\Young{2,1,1}}\wedge

5351: \pi_{\Young{2,1,1}}=-\psi_3\wedge \psi_3\,,

5352: \]

5353: where wedge denotes antisymmetric product (see p.354 bottom).

5354: \item $\pi_{\Young{2,2}}(\bt)=(-1)^e\psi_2(\bar\bt)$, where bar denotes

5355: the homomorphism from $B_4$ to $B_3$ given by $\bar\sg_{1,2,3}=\sg_{1,

5356: 2,1}$. (p. 355)

5357: \item $\pi_{\Young{1,1,1,1}}(\bt)=(-1)^e$ and

5358: $\pi_{\Young{4}}(\bt)=q^e$ (Note 4.7)

5359: \end{enumerate}

5360: 

5361: Now Squier observes in \cite{Squier}, that $\psi_i(\bt^{-1})(q)$

5362: and $\psi_i(\bt)(q^{-1})$ are conjugate and so have the same

5363: trace. So by the self-symmetry of $\pi_{\Young{2,1}}$ we have

5364: $\tr \pi_{\Young{2,1}}(t), \tr \pi_{\Young{2,2}}(t)\in (-t)^{e/2}\bR$.

5365: Similarly $(-t)^{-e/2}\tr \pi_{\Young{2,1,1}}(t)$ and $(-t)^{-e/2}\tr

5366: \pi_{\Young{3,1}}(t)$ are conjugate complex numbers. 

5367: These properties will be important below.

5368: 

5369: \begin{rem}

5370: Squier uses a different convention for

5371: $\psi_i$ from Jones. He transposes and changes sign in

5372: matrix entries with odd row-column sum (i.e., conjugates by

5373: $\dig(1,-1,1,-1,\dots)$). This, however, does not affect

5374: our arguments. 

5375: \end{rem}

5376: 

5377: The key point in arguments below is Squier's result.

5378: We write $M^*$ for the conjugate transposed of a matrix $M$.

5379: (That is, $M^*_{i,j}=\ol{M}_{j,i}$.)

5380: 

5381: \begin{theorem}(Squier \cite{Squier})

5382: For any $n\ge 1$ there exists a Hermitian matrix $J=J^{[n+1]}$ and a

5383: regular matrix $M$, such that with $J_0=M^*JM$ we have $\psi_n^*J_0

5384: \psi_n=J_0$.

5385: \end{theorem}

5386: 

5387: In particular, $J$ is degenerate or definite iff $J_0$ is so.

5388: Moreover,

5389: \[

5390: J_{i,j}\,=\,\left\{\,\begin{array}{c@{\quad}l}

5391: -1 & \mbox {if\ $|i-j|=1$} \\

5392: \sqrt{t}+1/\sqrt{t} & \mbox {if\ $i=j$} \\

5393: 0 & \mbox {otherwise}

5394: \end{array}\right.\,.

5395: \]

5396: 

5397: It is easy to see that if $t=1$, then $J$ is positive definite.

5398: Now definiteness is an open condition, so for $t$ close to $1$,

5399: it is still valid. 

5400: % By using the criterion of positivity of the principle minors,

5401: One can determine when $J$ loses this

5402: property. % in simple cases. It is easy to verify that for $3$-braids

5403: % this occurs in $e^{\pm 2\pi i/3}$, while for $4$-braids in $\pm i$.

5404: % In general we have

5405: 

5406: \begin{prop}\label{p4.1}

5407: The Squier form $J^{[n]}$ on $B_n$ degenerates exactly in the $n$-th

5408: roots of unity. In particular, it is positive definite exactly

5409: when $|\arg t|<2\pi /n$.

5410: \end{prop}

5411: 

5412: \proof Denote by $J^{[n]}$ the form corresponding to $n$-braids,

5413: i.e. the one given by restricting $J$ to the first $n-1$

5414: rows and columns. It is not too hard to calculate the determinant

5415: of $J^{[n]}$. By development in the last row,

5416: \[

5417: \det J^{[n]}\,=\,(\sqrt{t}+1/\sqrt{t})\det J^{[n-1]}-\det J^{[n-2]}\,,

5418: \]

5419: whence

5420: \[

5421: \det J^{[n]}\,=\,\frac{t^n-1}{\sqrt{t}-1/\sqrt{t}}\,\cdot\,

5422: \frac{1}{(\sqrt{t})^n}\,.

5423: \]

5424: Then the claim follows easily. To see definiteness use the

5425: (positivity of) the principal minor criterion. Since $e^{\pm 2\pi i/n}$

5426: is a simple zero of $\det J^{[n]}$, the determinant must become

5427: negative for $|\arg t|\in(2\pi/n,4\pi/n)$. Then applying this

5428: argument to all $n'<n$ shows that $J^{[n]}$ is not positive definite

5429: for $|\arg t|\ge 2\pi/n$. \qed

5430: 

5431: 

5432: So on the arcs of $S^1$ that connect the primitive $n$-th root of

5433: unity to $1$, we have that $J$ is positive definite. Now

5434: if $J$ is such, it can be written as $Q^*Q$, and then

5435: conjugating $\psi_i$ by $QM$ we obtain a $U(n-1)$-representation.

5436: This means in particular that all eigenvalues of $\psi_i$

5437: have unit norm. We will below derive implications of this

5438: circumstance for the link polynomials.

5439: 

5440: \subsection{Norm estimates\label{s4.2}}

5441: 

5442: The Jones polynomial $V$ can be specified, for our

5443: purposes, by $V(t)=X(t,t)$. In the following, which

5444: root of complex numbers is taken is irrelevant,

5445: important is though that it be kept fixed in subsequent

5446: calculations. By $\Re$ we denote the real part of

5447: a complex number.

5448: 

5449: \begin{theorem}\label{th1}

5450: If $|t|=1$, $\Re t>0$ and $\bt$ is a $4$-braid, then 

5451: $\big |V_{\hat\bt}(t)\big|\,\le\,(2\Re \sqrt{t})^3$.

5452: If $\bt$ is a $3$-braid and $\Re t>-1/2$, then

5453: $\big |V_{\hat\bt}(t)\big|\,\le\,(2\Re \sqrt{t})^2$.

5454: \end{theorem}

5455: 

5456: \proof We have from \cite{Jones} that if $\bt\in B_4$ with

5457: $[\bt]=e$, then

5458: \begin{eqn}\label{Vf}

5459: V_{\hat\bt}(t)\,=\,\left(-\sqrt{t}\right)^{e-3}

5460: \left[\,\frac{t(1-t^3)}{1-t^2}\,\tr \psi_3+\frac{t^2}{1+t}

5461: \tr \bar\psi_2\,+\,\frac{1-t^5}{1-t^2}\,\right]\,,

5462: \end{eqn}

5463: where $\bar\psi_2$ is the composition of $\psi_2$ with

5464: $\bar{ }\,:\,B_4\to B_3$. Taking norms and using that

5465: $\bar\psi_2$ and $\psi_3$ are unitary, we find

5466: \begin{eqn}\label{X}

5467: \big |V(t)\big|\,\le 3\left|\frac{1-t^3}{1-t^2}\right|+

5468: \frac{2}{|1+t|}+\left|\frac{1-t^5}{1-t^2}\right|\,.

5469: \end{eqn}

5470: It is now a routine (but somewhat tedious) calculation to

5471: verify that the r.h.s. is equal to $(2\Re \sqrt{t})^3$

5472: for $|t|=1$, $\Re t>0$. 

5473: 

5474: For $\bt\in B_3$ we have similarly

5475: \begin{eqn}\label{V_3}

5476: V_{\hat\bt}(t)\,=\,\left(-\sqrt{t}\right)^{e-2}

5477: \left[\,t\cdot \tr\psi_2\,+\,(1+t^2)\,\right]\,,

5478: \end{eqn}

5479: and the result follows using $|\tr\psi_2|\le 2$. \qed

5480: 

5481: This theorem generalizes Jones's result \cite[proposition 15.3]{Jones}

5482: for $n\le 4$, where he considers $t=e^{2\pi i/k}$, $k\ge 5$. In fact

5483: the comparison to (and established coincidence with) Jones's

5484: estimate led to the simplification of the r.h.s. of \eqref{X}.

5485: In \cite{posex_bcr} we noted that Jones's estimate can be

5486: better than MWF when $MWF=3$, but by connected sum one

5487: can give an example for $MWF=4$. Again it appears that for

5488: $\bt\in B_4$ and $t\ne e^{\pm \pi i/3}$ the set $\{\,|V_{\hat

5489: \bt}(t)|\,\}$ is dense in $[0,(2\Re \sqrt{t})^3]$, and similarly it

5490: is in $[0,(2\Re \sqrt{t})^2]$ for $\bt\in B_3$ and $t\ne e^{\pm \pi

5491: i/3},e^{\pm \pi i/5}$. (See also the remarks at the end of \S 12

5492: in \cite{Jones}.)

5493: 

5494: \begin{conjecture}

5495: If $\bt\in B_n$ and $|t|=1$ with $|\arg t|<2\pi /n$ then

5496: $|V(t)|\,\le\,(2\Re \sqrt{t})^{n-1}$.

5497: \end{conjecture}

5498: 

5499: In the case of the Alexander polynomial $\Dl(t)=X(t,1/t)$, we

5500: can say something on general braids.

5501: 

5502: \begin{theorem}

5503: For each $n\ge 2$, %there is an $\eps_n>0$ such that

5504: if $|t|=1$ and $|\arg t|\le 2\pi/n$ and $\bt\in B_n$ then

5505: \[

5506: \big |\Dl_{\hat\bt}(t)\big|\,\le\,\frac{2^{n-1}|1-t|}{|1-t^n|}\,.

5507: \]

5508: \end{theorem}

5509: 

5510: \proof

5511: % Let $\eps_n>0$ be so that $\ds\frac{|1-t^n|}{|1-t|}>n-0.1$ and

5512: $J_0$ is positive definite when $|t|=1$ and $|\arg t|<2\pi/n$.

5513: Then

5514: \[

5515: \big |\Dl_{\hat\bt}(t)\big|\,\frac{|1-t|}{|1-t^n|}\,=\,

5516: \big |\det (1-\psi_{n-1}(\bt))\big|\,,

5517: \]

5518: and all eigenvalues of $1-\psi_{n-1}$ have norm $\le 2$.

5519: The case $|\arg t|=2\pi/n$ follows by continuity. \qed

5520: 

5521: \begin{corr}

5522: For all $n,k$ the set

5523: \[

5524: \big\{\,\Dl(\hat\bt)\,:\,\bt\in B_n\,,\ \ \deg\Dl\le k\,\big\}

5525: \]

5526: is finite. That is, among closed braids of given number of strands

5527: only finitely many Alexander polynomials of given degree occur.

5528: \end{corr}

5529: 

5530: \proof $\Dl$ is determined by $\Dl(t_i)$ for $k$ different

5531: $t_i$ with $|t_i|=1$, $0<\arg t_i<2\pi/n$, by means of a 

5532: linear transformation using the (regular Vandermonde)

5533: matrix $M=(t_i^j)_{i,j=1}^n$.

5534: So $||[\Dl(t)]_{t^j}||\,\le\,||M^{-1}||\cdot ||\Dl(t_i)||$\,.

5535: \qed

5536: 

5537: This result should be put in contrast to the various constructions

5538: of knots with any given Alexander polynomial. For example

5539: a recent construction of Nakamura \cite{Nakamura2} allows to

5540: realize the degree of the polynomial by the (actually braidzel)

5541: genus of the knot. (We were subsequently independently

5542: able to further specialize this result to canonical genus.) 

5543: % do not know the following 

5544: % 

5545: % \begin{question}

5546: % Is any Alexander polynomial admitted by a knot with canonical

5547: % genus equal to the degree of the polynomial?

5548: % \end{question}

5549: % 

5550: A different construction of Fujii 

5551: \cite{Fujii} shows that knots with 3 bridges admit all

5552: Alexander polynomials. So the situation between braid and

5553: plat closures is completely different.

5554: 

5555: Compare also Birman-Menasco's result in \cite{BirMen3}, mentioned

5556: (for knots) in remark \ref{rfn}, that there are only finitely

5557: many closed braids of given number of strands with given genus. 

5558: Note that we do not claim that only finitely many 

5559: closed braids of given number of strands with given Alexander

5560: polynomial (degree) occur. For $3$-braids it is true, but

5561: from $5$-braids on the non-faithfulness of the Burau

5562: representation should (in principle, modulo the evaluability

5563: of another invariant) make it possible to construct

5564: infinite families of links with the same (for example, trivial)

5565: polynomial. It makes some sense to ask about the status of 4-braids.

5566: 

5567: \begin{question}

5568: Are there only finitely many 

5569: closed 4-braids of given Alexander polynomial (degree)?

5570: \end{question}

5571: 

5572: When working with $\Dl$, for $3$- and $4$-braids we can be

5573: more explicit.

5574: 

5575: \begin{corr}\label{pp1}

5576: If $\bt\in B_4$, then

5577: \begin{eqn}\label{Xx}

5578: \big|\,\Dl_{\hat\bt}(t)\,\big|\,\le\,\frac{8\,|1-t|}{|1-t^4|}\,,

5579: \end{eqn}

5580: when $|t|=1$ and $\Re t>0$. If $\bt\in B_3$ and $\Re t>-1/2$, then

5581: \begin{myeqn}{\qed}

5582: \big|\,\Dl_{\hat\bt}(t)\,\big|\,\le\,\frac{4\,|1-t|}{|1-t^3|}\,.

5583: \end{myeqn}

5584: \end{corr}

5585: 

5586: Putting $t=e^{2\pi i/5}$ in \eqref{Xx} we have $\big|\,\Dl(e^{2\pi

5587: i/5})\,\big|\,\le\,8\,.$ This improves the bound $10.47\dots$ in

5588: \cite{posex_bcr} suggested to replace Jones's (incorrect) value 6.5

5589: in \cite{Jones2}. For 3-braid \em{knots} Jones gives in

5590: \cite[proposition 15.2]{Jones} the better bound 3 when $t=i$, using

5591: the property $V(i)=\pm 1$.

5592: 

5593: \begin{exam}\label{4249}

5594: The simplest knots with $MWF=4$ which can be excluded from being a

5595: 4-braid using corollary \reference{pp1} are $13_{8385}$ (where one

5596: can use $t=e^{2\pi i/9}$) and $14_{37492}$ (with $t=e^{2\pi i/8}$).

5597: For $3$-braids we can deal with the known examples $9_{42}$

5598: and $9_{49}$. (This gives now an alternative proof that

5599: there are no 3-braid knots with such Alexander polynomial, as

5600: we explained in example \reference{x29}.)

5601: \end{exam}

5602: 

5603: If we know the exponent sum of $\bt$ we can do better.

5604: 

5605: \begin{prop}\label{pp2}

5606: If $\bt\in B_4$, $e=[\bt]$ and $t$ as before, then

5607: \[

5608: \left|\left(

5609: -\sqrt{t}\right)^{e-3}\Dl_{\hat\bt}(t)\frac{1-t^4}{1-t}-1

5610: +(-t)^e\,\right|\,\le 6.

5611: \]

5612: If $\bt\in B_3$ and $\Re t>-1/2$, then

5613: \begin{myeqn}{\qed}

5614: \left|\left(

5615: -\sqrt{t}\right)^{e-2}\Dl_{\hat\bt}(t)\frac{1-t^3}{1-t}-1

5616: -(-t)^e\,\right|\,\le 2\,.

5617: \end{myeqn}

5618: \end{prop}

5619: 

5620: \begin{exam}\label{lxd}

5621: Since one can determine the possible $e$ via MWF from $P$,

5622: one can apply proposition \reference{pp2} for given $P$. In

5623: the $3$-braid case we can exclude $10_{150}$ this way. The

5624: remaining two 10 crossing knots with unsharp MWF, $10_{132}$

5625: and $10_{156}$, fail~-- understandably, since they share the

5626: skein polynomials of (the closed 3-braids) $5_1$ and $8_{16}$

5627: resp. (See the table\footnote{The second duplication was

5628: noted in the remarks after Jones's table, but not referred

5629: to correctly in its last column.} in \cite{Jones}.) For

5630: $4$-braids several new 14 crossing knots can be ruled

5631: out, for example $14_{21199}$.

5632: \end{exam}

5633: 

5634: 

5635: \begin{rem}

5636: The use of $P$ to restrict the possible values of $e$ is

5637: usually most effective, but not indispensable. There are

5638: other conditions on $e$, originating from Bennequin's

5639: work \cite{Bennequin}, that can be more applicable in

5640: certain cases where the calculation of $P$ is tedious.

5641: Also, when $t$ is a root of unity of order $n$, the tests

5642: depend only on $e\bmod 2n$.

5643: \end{rem}

5644: 

5645: \begin{rem}

5646: Note that the quantity $\chi$ in \eqref{chi} is equal to $u\,

5647: \tr \psi_2$, where $u=\sqrt{-t}+\ds\frac{1}{\sqrt{-t}}$ and

5648: $\psi_2$ is the Burau matrix of $B_3$ (see the explanation in

5649: \cite{Kanenobu}). Thus one can obtain

5650: similar estimates for values of $Q$ on 3-braids.

5651: \end{rem}

5652: 

5653: 

5654: \subsection{Skein polynomial}

5655: 

5656: Now it is natural to look at the full 2-variable skein polynomial

5657: $X$. We have, as in \eqref{STR}, for $\bt\in B_4$ of exponent

5658: sum $e$,

5659: \[

5660: X_{\hat\bt}(q,\lm)\,=\,-\sqrt{\lm}^{e-3}\,

5661: \sum_{Y\vdash 4}\,\tr\pi_Y(q)\,\tW_Y(q,\lm)\,,

5662: \]

5663: where the weights $\tW_Y$ are given in $\lm$-coefficients by

5664: table \reference{table1}. (They are all polynomials in $\lm$ of

5665: degree $3$, with coefficients being rational expressions in

5666: $q$.) Now, with given $e$ and $P$, we have $4$ equations

5667: (the coefficients in $\lm$) in $3$ unknowns (the traces of

5668: $\pi_{211}$, $\pi_{31}$ and $\pi_{22}$; we use here the

5669: partition notation for the subscripts). However, the

5670: restriction of the matrix in table \reference{table1} to

5671: the columns of $\pi_{211}$, $\pi_{31}$ and $\pi_{22}$

5672: has rank $2$. This means that two of the $X_i=[X]_{\lm^

5673: {i+(e-3)/2}}$ for $0\le i\le 3$ determine the other two.

5674: One could believe now to use

5675: this as a 4-braid test. However, these two relations

5676: result from the general substitutions $\lm=1$ and $\lm=1/q^2$

5677: that turn $X$ into the component parity count or $1$.

5678: These substitutions kill all trace weights except of the trivial

5679: or parity representation, and for these representations the weights

5680: become also independent on the braid group. Thus the

5681: relations between the $X_i$ will hold whenever $MWF\le 4$,

5682: and are useless as a 4-braid test. In a similar vein, one has 

5683: 

5684: \newbox\@tempboxb

5685: \setbox\@tempboxb=\hbox{

5686: \capt The weights of the traces contributing to the

5687: $\lm$-coefficients of $X$ for a $4$-braid. Each table entry

5688: must be multiplied with the factor on the right in the

5689: first row to obtain the contribution of the corresponding trace

5690: to $-\sqrt{\lm}^{e-3}$ times the power of $-\lm$ in the

5691: first column. In the first row the symbol $[i]$ denotes $1-q^i$.}

5692: 

5693: \begin{table}

5694: \[

5695: \begin{array}{c*5{|c}}

5696: & 

5697: \ds \tW_{1111}=\frac{1}{[2][3][4]}\times &

5698: \ds \tW_{211}=\frac{q}{[2][1][4]}\times &

5699: \ds \tW_{22}=\frac{q^2}{[2][2][3]}\times &

5700: \ds \tW_{31}=\frac{q^3}{[2][1][4]}\times &

5701: \ds \tW_{4}=\frac{q^6}{[2][3][4]}\times \\[5mm]

5702: \hline

5703: \ry{6mm}

5704: \ds 1 & 1 & 1 & 1 & 1 & 1 \\[1mm]

5705: -\lm  & q^2+q^3+q^4 & 1+q^2+q^3   & 1+q+q^2   & 1+1/q+q^2 & 1+1/q+1/q^2 \\[2mm]

5706: \lm^2 & q^5+q^6+q^7 & q^3+q^2+q^5 & q+q^2+q^3 & 1/q+q+q^2 & 1/q+1/q^2+1/q^3 \\[2mm]

5707: -\lm^3& q^9         & q^5         & q^3       & q         & 1/q^3 \\[2mm]

5708: \end{array}

5709: \]

5710: \caption{\unhbox\@tempboxb

5711: \label{table1}}

5712: \end{table}

5713: 

5714: \begin{prop}\label{pp7.6}

5715: If a braid $\bt$ has $MWF(\bt)\le 4$, then $V(\hat\bt)$ and

5716: $\Dl(\hat\bt)$ together with the exponent sum $[\bt]=e$,

5717: determine $P(\hat\bt)$.

5718: \end{prop}

5719: 

5720: \proof From $e$, $V$ and $\Dl$ we know

5721: \[

5722: \tl X_a\,=\,\sum_{i=0}^3\,X_i(q)\,q^{a(i+(e-3)/2)}\,,

5723: \]

5724: for $a\in\{-2,-1,0,1\}$. (For $a=-2$ and $a=0$ we have the

5725: trivializing substitutions, $a=-1$ corresponds to $\Dl$ and

5726: $a=1$ to $V$.) So one can recover $X_i$ from $\tl X_a$

5727: and $e$. (One can write down an explicit formula easily.) \qed

5728: 

5729: This condition is thus equally unhelpful as a 4-braid

5730: test. For a similar reason, I expect (though I have not

5731: rigorously derived) an explanation of the (experimentally

5732: observed) failure of Jones's conditions \cite[\S 8]{Jones}

5733: (see formula (8.10)) to obstruct to a 4-braid.

5734: 

5735: \begin{exam}

5736: The knot $11_{386}$, known from \cite{LickMil}, has the

5737: Jones polynomial of the figure-8-knot. So $11_{386}$ and its mirror

5738: image show that the dependence of $P$ on $e$ in proposition

5739: \reference{pp7.6} is essential.

5740: \end{exam}

5741: 

5742: \begin{question}

5743: Are there 5-braids $\bt_{1,2}$ (at least one of which has 

5744: $MWF=5$), with the same exponent sum,

5745: $V(\hat\bt)$ and $\Dl(\hat\bt)$, but different $P(\hat\bt)$?

5746: \end{question}

5747: 

5748: The lack of such examples, after some check in the knot tables,

5749: is at least not fully explainable. Only 6-braids could be found.

5750: 

5751: \begin{exam}

5752: The knots $16_{443392}$ and $!15_{223693}$ have the same $V$

5753: and $\Dl$ but different $P$ polynomial. They have $MWF=5$ resp. $4$,

5754: with $5$- resp. $4$-braid representations of exponent sum $2$ resp.

5755: $-1$, so one obtains 6-braids of exponent sum $1$ by stabilization.

5756: \end{exam}

5757: 

5758: One can now, as before, go over to the norms in each

5759: row in table \reference{table1}, or of arbitrary linear

5760: combinations of such rows. Then for $q$ where $J_0$ is

5761: definite we obtain again estimates on $|X_i(q)|$, or of

5762: $|X(q,\lm)|$ for any non-zero complex number $\lm$. (In

5763: particular, for $\lm=q$ we obtain theorem \reference{th1} and

5764: for $\lm=1/q$ corollary \reference{pp1}.) Although they still

5765: contain the Jones and Alexander polynomial conditions (which

5766: were observed both non-trivial in comparison to MWF), such

5767: skein polynomial norm estimates have not proved in practice,

5768: as a 4-braid test, an efficient improvement over their special

5769: cases and MWF. Below we will explain how to do much better.

5770: 

5771: Knots like $10_{132}$ and $10_{156}$ in example \ref{lxd} show

5772: a disadvantage of our test, resulting from not taking into account

5773: information of other invariants. On the opposite hand, the

5774: mere use of $\Dl$ or $P$ reduces calculation complexity, and

5775: excludes any potential further knot with such polynomial.

5776: In the case of $P$ this gives yet a different way to

5777: answer negatively Birman's question if one can realize any

5778: skein polynomial by a link making MWF sharp (see \cite

5779: {posex_bcr,3br}).

5780: 

5781: Using the Brandt-Lickorish-Millett-Ho polynomial, Murakami

5782: \cite{Murakami} and later Kanenobu \cite{Kanenobu} gave with

5783: theorem \reference{TMF} a more

5784: efficient (in excluding examples, though less in calculation

5785: complexity) test for a 3-braid. But the work in \cite{3br} and

5786: in the previous sections of this paper makes the study of 

5787: polynomials of 3-braids anyway less relevant. $4$-braids

5788: become much more of interest, and in this case, after MWF,

5789: the problem to find applicable conditions has been largely

5790: unsettled for quite a while.

5791: 

5792: \subsection{Recovering the Burau trace}

5793: 

5794: \subsubsection{Conditions on the eigenvalues}

5795: 

5796: Using just norms clearly weakens the conditions considerably,

5797: and so one would like to identify the Burau eigenvalues directly.

5798: However, the relations between the $X_i$ do not allow to recover 

5799: by simple algebraic means the individual traces from $P$. 

5800: 

5801: Using Squier's unitarity, there is an analytic way to recover

5802: the Burau trace of 4-braids (at least for generic $t$ and

5803: up to finite indeterminacy). Since by the previous remarks

5804: the use of $P$ is not essential, we will describe the

5805: procedure given $\Dl$, $V$ and $e$. This gives the most

5806: significant practical enhancement to the above 4-braid tests.

5807: 

5808: In the following we fix a 4-braid $\bt$ of exponent sum $e$, whose

5809: closure link has Jones polynomial $V$ and Alexander polynomial $\Dl$.

5810: We let $t$ be a unit norm complex number with non-negative real part.

5811: 

5812: We use that

5813: \[

5814: \tr \psi_3\wedge \psi_3 = (-t)^{e/2}\ol{\tr \psi_3(-t)^{-e/2}}\,.

5815: \]

5816: Then we have

5817: \[

5818: \Dl(t)\,=\,\frac{1-t}{1-t^4}\,\left(-\frac{1}{\sqrt{t}}\right)^{e-3}

5819: \cdot\,\left[ 1-\tr \psi_3 + (-t)^{e/2}\ol{\tr \psi_3(-t)^{-e/2}}-

5820: (-t)^e\,\right]\,.

5821: \]

5822: So, with $i=\sqrt{-1}$,

5823: \[

5824: \dl\,:=-\frac{1}{2i}(-t)^{-e/2}

5825: \left[ \Dl(t)\frac{1-t^4}{1-t}\left(-\frac{1}{\sqrt{t}}\right)^{3-e}

5826: -1 + (-t)^e\,\right]

5827: \]

5828: is a real number. Now when $\lm_{1,2,3}$ are the eigenvalues of

5829: $\psi_3(\bt)$, we have

5830: \begin{eqnarray*}

5831: A & = & \lm_1+\lm_2+\lm_3\,=\,(-t)^{e/2}(y+i\dl) \\

5832: B & = & \lm_1\lm_2+\lm_1\lm_3+\lm_2\lm_3\,=\,(-t)^{e/2}(y-i\dl) \\

5833: C & = & \lm_1\lm_2\lm_3\,=\,(-t)^e\,.

5834: \end{eqnarray*}

5835: Here $y$ is a real number we do not know, and we try to determine.

5836: Since $|\lm_k|=1$, we have

5837: as before $|A|\le 3$ and $|B|\le 3$. So the range

5838: for $y$ is $[-y_0,y_0]$ with $y_0=\sqrt{9-\dl^2}$.

5839: (If $|\dl|>3$ we are done as before.) Then

5840: \[

5841: \tr\psi_3\,\in\,[(-t)^{e/2}(-y_0+i\dl),\, (-t)^{e/2}(y_0+i\dl)]\,,

5842: \]

5843: with the interval understood lying in $\bC$.

5844: Let $\psi_-$ and $\psi_+$ be the endpoints of this interval.

5845: 

5846: Now one can restrict the interval $[-y_0,y_0]$ for $y$ using the

5847: Jones polynomial as follows.

5848: 

5849: Let $\rho\,:=(-t)^{-e/2}\tr\bar\psi_2\in [-2,2]$. The restriction

5850: to the given range follows because $J^{[3]}$ (see the proof of

5851: proposition \reference{p4.1}) is also definite when

5852: $J^{[4]}$ is. From \eqref{Vf} we have

5853: \[

5854: \tr \psi_3\,=\,\frac{1-t^2}{t(1-t^3)}\,\left[\,

5855: V(t)\,\left(-\frac{1}{\sqrt{t}}\right)^{e-3}-\frac{1-t^5}{1-t^2}

5856: -\frac{t^2}{1+t}(-t)^{e/2}\cdot \rho\,\right]\,=:\,\tl\psi(\rho)\,,

5857: \]

5858: so

5859: \[

5860: \tl\psi(\rho)=\frac{1-t^2}{t(1-t^3)}\left [

5861: V(t)\,\left(\frac{-1}{\sqrt{t}}\right)^{e-3}-

5862: \frac{1-t^5}{1-t^2}\right] -\frac{\rho}{1+2\Re t}(-t)^{e/2}\,.

5863: \]

5864: (Just for the purpose of defining $\tl\psi$, we should regard here

5865: $\rho$ as a formal parameter, rather than as a concrete value.)

5866: 

5867: Let $\tl\psi(\pm 2)=:\tl\psi_\pm$. Since $1+2\Re t>0$, we have

5868: $\Re((-t)^{-e/2}\tl\psi_+)>\Re((-t)^{-e/2}\tl\psi_-)$.

5869: 

5870: Then $[\tl\psi_-,\tl\psi_+]\subset \bC$ is an interval

5871: of the same slope as $[\psi_-,\psi_+]$, so we check if they overlap.

5872: 

5873: Let

5874: \[

5875: \tl y_{\pm}= (-t)^{-e/2}\tl\psi_\pm -i\delta.

5876: \]

5877: 

5878: Then for a consistent restriction on $\tr \psi_3$ the following holds:

5879: \begin{enumerate}

5880: \item $\tl y_{\pm}$ are real

5881: \item $ \tl y_+\tl y_-\le 0$ or $\min(\tl y_\pm^2+\delta^2)\le 9$.

5882: \end{enumerate}

5883: 

5884: Potentially these conditions may be violated, but in

5885: practice they seem always to hold. (We have not elaborated on why

5886: this is so, though it may be worth understanding.) Then at least,

5887: we consider 

5888: \begin{eqn}\label{qw0}

5889: y\in [-y_0,y_0]\cap [\tl y_-,\tl y_+]\,.

5890: \end{eqn}

5891: 

5892: We have now the cubic

5893: \[

5894: x^3 +ax^2 +bx +c \,:=\,x^3-A x^2+B x-C \,=\,0\,.

5895: \]

5896: One solution is obtained by Cardano's formula

5897: \begin{eqn}\label{qw}

5898: \lm_1\,=\,-\frac{a}{3}

5899: -\frac{\sqrt[3]{2}(-a^2+3b)}{3\Gm} + \frac{\Gm}{3\sqrt[3]{2}}\,,

5900: \end{eqn}

5901: where 

5902: \begin{eqn}\label{Gm}

5903: \Gm=\sqrt[3]{

5904:   -2a^3 + 9ab - 27c + 

5905:   \sqrt{

5906:     27\,\left( -a^2b^2 + 4b^3 + 4a^3c -18abc + 27c^2

5907:     \right)

5908:   }

5909: }\,.

5910: \end{eqn}

5911: We must have that $|\lm_1|=1$. Then we must check $|\lm_{2,3}|=1$.

5912: For this their exact determination is not necessary.

5913: We have $\lm_2+\lm_3=A-\lm_1$ and $\lm_2\lm_3=C/\lm_1$.

5914: In order $|\lm_{2,3}|=1$, we must have

5915: \[

5916: \lm_2+\lm_3=c\cdot \sqrt{\lm_2\lm_3}\,,

5917: \]

5918: with $c\in [-2,2]$, so $A-\lm_1=c\sqrt{C/\lm_1}$, which is

5919: equivalent to

5920: \begin{eqn}\label{qw2}

5921: \frac{(A-\lm_1)^2\lm_1}{C}\,\in\,[0,4]\,.

5922: \end{eqn}

5923: 

5924: \subsubsection{Applications and examples}

5925: 

5926: The image of the r.h.s. of \eqref{qw0} under \eqref{qw}

5927: will be some curve in $\bC$ that generically intersects

5928: $S^1$ in a finite number of points. This parametric

5929: equality can be examined numerically and allows to recover

5930: $A=\tr \psi_3$ up to finite indeterminacy. In particular,

5931: we have

5932: 

5933: \begin{prop}\label{4brt}

5934: Assume for some $t\in S^1$ with $|\arg t|\le \pi/2$ we

5935: cannot find $y$ as in \eqref{qw0}, such that $\lm_1$

5936: given by \eqref{qw} has norm 1 and \eqref{qw2} holds. Then

5937: there is no $\bt\in B_4$ with the given $e$ whose closure has

5938: the given Alexander and Jones polynomial. \qed

5939: \end{prop}

5940: 

5941: \begin{rem}

5942: To apply the test in practice, one chooses a small stepwidth $s$

5943: for $y$ in the interval \eqref{qw0}, and calculates the derivative

5944: of the r.h.s. of \eqref{qw} in $y$ to have an error bound on

5945: $||\lm_1|-1|$ in terms of $s$. When some of the radicands in \eqref{Gm}

5946: becomes close to $0$, some care is needed. One situation where

5947: such degeneracy occurs are the knots $15_{144634}$, $15_{144635}$,

5948: and $15_{145731}$. They have representations $\bt\in B_4$, whose

5949: Burau matrix is trivial for $t=e^{\pi i/3}$. This may be noteworthy

5950: on its own in relation to the problem whether $\psi_3$ is faithful. 

5951: (Clearly $\psi_3$ is unfaithful at any root of unity on the

5952: center of $B_4$, but here $\bt$ is not even a pure braid.)

5953: \end{rem}

5954: 

5955: \begin{exam}

5956: Applying proposition \reference{4brt} we can exclude $11_{387}$, one

5957: of the 7 prime 11 crossing knots with $b=5$ but $MWF\le 4$. Eleven

5958: prime knots with 12 crossings, and 63 of 13 crossings where braid

5959: index 4 is not prohibited by MWF can be ruled out. The correctness

5960: of these examples was later verified by the 2-cabled MWF. Up to 16

5961: crossings more than 4000 examples were obtained. (Let us note that

5962: from them only about 100 can be identified using the norm estimates.)

5963: \end{exam}

5964: 

5965: \begin{exam}

5966: The check of prime 14 crossing knots to which our criterion applied,

5967: revealed 6 knots with 2-cable MWF bound 8. One, $14_{22691}$,

5968: can be excluded from braid index 4 (as done with $14_{45759}$ 

5969: in \cite{posex_bcr}), by making the $v$-degrees of the

5970: polynomial of the 2-cable contradict the exponent sum of its

5971: possible 8-braid representation. However, for the other 5

5972: knots, $14_{28220}$, $14_{30960}$, $14_{41334}$, $14_{41703}$,

5973: and $14_{44371}$, the argument fails, and so our condition

5974: seems the only applicable one. (Clearly a 3-cable polynomial

5975: is not a computationally reasonable option, and even the 2-cable

5976: requires up to several hours, while our test lasts a few seconds.)

5977: \end{exam}

5978: 

5979: Still our criterion leaves open several interesting examples

5980: of (apparent) failure of the 2-cable MWF inequality, among

5981: them the knot $13_{9684}$ encountered with M.~Hirasawa. A more

5982: general, and important, possible application is as follows.

5983: 

5984: \begin{rem}\label{rJC}

5985: In relation to Jones's conjecture \reference{jcc}, we

5986: already quoted (in the proof of theorem \ref{thJC})

5987: the observation in \cite{posex_bcr} that counterexamples

5988: to the conjecture would make MWF and \em{all} of its cabled

5989: versions unsharp. In that sense, our 4-braid test may be

5990: the first possible approach toward identifying such a

5991: counterexample. Birman-Menasco claimed indeed a family

5992: of 6-string potential counterexamples, and K.~Kawamuro

5993: gave later a simpler family on 5 strings. Extensive

5994: checks with our test of Kawamuro's knots failed to turn up

5995: successful cases. This puzzled us a while, until Kawamuro

5996: reported recently that in fact H.~Matsuda falsified all

5997: Birman-Menasco (and hence also Kawamuro's) candidates.

5998: \end{rem}

5999: 

6000: \subsection{Mahler measures}

6001: 

6002: \subsubsection{3-braids}

6003: 

6004: The material in section \reference{SBurau} originated from

6005: a question of S.~Kamada whether the Alexander polynomial

6006: Mahler measure $M(\Dl)$ is bounded on closed 3-braids. 

6007: 

6008: \begin{defi}

6009: For a Laurent polynomial $p\in\bZ[t^{\pm 1}]$, and $n\in\bN_+$,

6010: we define the $n$-norm of $p$ by

6011: \[

6012: ||p||_n\,:=\,\sqrt[n]{\sum_{k=-\infty}^{\infty}\bigl([p]_k\bigr)^n}\,,

6013: \]

6014: and its Mahler measure by

6015: \[

6016: M(p)\,:=\,\prod_{|t|\ge 1,\,p(t)=0}\,|t|\,.

6017: \]

6018: This is extended to $p\in\bZ[t^{\pm 1/2}]$ in the obvious way.

6019: See \cite{SSW,CK}.

6020: \end{defi}

6021: 

6022: Kamada's question is related to controlling $|V(t)|$ and $|\Dl(t)|$

6023: for $t\in S^1$, since the 2-norm $||\,.\,||_2$ and the Mahler

6024: measure $M$ of polynomials have circle integral formulas (see

6025: \eqref{55'} and \eqref{Ji}). One

6026: can thus ask whether $||\Dl\cdot W_n||_2$ is bounded for proper

6027: $W_n\in \bZ[t,1/t]$, or (weaker) whether $M(\Dl)$ is bounded

6028: (and similarly $M(V)$ and $||V\cdot W_n||_2$) for braid

6029: index $\le n$.

6030: 

6031: For values of $t$ where $J_0$ is not definite, however, there

6032: seems little one can say on the range (on closed 3-braids)

6033: of $|V(t)|$ or $|\Dl(t)|$. Likely they are dense in $\bR_+$.

6034: % So bounded 2-norms seem unlikely, but even if $|\Dl(t)|$

6035: % is unbounded for any $t$ with indefinite $J_0$, that is

6036: % not sufficient.

6037: We can conclude boundedness properties in

6038: special cases where the indefinite $J_0$ values of $t$ are

6039: controllable. For example the following can be proved easily.

6040: (Note that for polynomials with integer coefficients the properties

6041: a set of polynomials to have bounded $2$-norms, or finitely many

6042: distinct $2$-norms, are equivalent, and also equivalent to the same

6043: two properties for the $1$-norm.)

6044: 

6045: \begin{prop}

6046: The set

6047: \[

6048: \{\,||(1-t^{2n})\Dl_{\hat\bt}(t)||_2\,:\,

6049: \bt\in B_{2n},\,\Dl_{\hat\bt}\in\bZ[t^{\pm n}]\,\}

6050: \]

6051: is finite for any $n\ge 2$.

6052: \end{prop}

6053: 

6054: \proof We have 

6055: \begin{eqn}\label{55'}

6056: ||X||_2^2\,=\,\int_{0}^{1}\,\big|\,X(e^{2\pi is})\,\big|^2\,ds\,.

6057: \end{eqn}

6058: If $\Dl\in\bZ[t^{\pm n}]$, then this integral for 

6059: \[

6060: X=\frac{(1-t^{2n})\Dl(t)}{1-t}=\det(1-\psi_{2n-1})

6061: \]

6062: is controlled from its part over $0<s<\ffrac{1}{2n}$, and

6063: this in turn is controlled by the unitarity of $\psi_{2n-1}$.

6064: To eliminate the denominator, observe that the norm in

6065: \eqref{55'} is just the one in $L^2(S^1)$, and so

6066: $||P\cdot Q||_2\le ||P||_2\cdot ||Q||_2$. \qed

6067: 

6068: However, in general a bound on the Mahler measure of

6069: arbitrary polynomials of closed $n$-braids does not exist

6070: even for $n=3$.

6071: 

6072: \begin{prop}

6073: The Mahler measure of $\Dl$ and $V$ is unbounded on closed

6074: 3-braids.

6075: \end{prop}

6076: 

6077: {\def\gm{g_{\min}'}

6078: \proof In the following $x\le O(y)$ means $\limsup |x/y|<\infty$

6079: (the limit following from the context), $x\ge O(y)$ means $\liminf

6080: |x/y|>0$ and $x=O(y)$ means both $x\le O(y)$ and $x\ge O(y)$.

6081: The quantities $C$ and $C'$ will stand for real constants, with

6082: $C$ being positive. They are understood to be independent on the

6083: indices of the terms in the formula they occur in, but

6084: may vary between the various inequalities.

6085: 

6086: We consider the links $L_n$ given by the closed 3-braids

6087: $(\sg_1\sg_2^{-1})^n$. Using the Burau matrix eigenvalues, we find

6088: \[

6089: V_n\,:=V(L_n)\,=\,t+\frac{1}{t}+e_-^n(t)+e_+^n(t)\,,

6090: \]

6091: where 

6092: \[

6093: e_{\pm}(t)\,=\,-\frac{t+1/t-1}{2}\,\pm

6094: \sqrt{\left( \frac{t+1/t-1}{2} \right)^2-1 }\,.

6095: \]

6096: Similarly

6097: \[

6098: \Dl_n\,:=\,\Dl(L_n)\,=\,\frac{1}{t+1/t+1}\left(2-e_-^n(t)-e_+^n(t)

6099: \right)\,.

6100: \]

6101: Now the Jensen integral gives

6102: \begin{eqn}\label{Ji}

6103: \log\,M(V_n)\,=\,\int_0^1\,\log\,|V_n(e^{2\pi is})|\,ds\,.

6104: \end{eqn}

6105: Since $V_n$ is reciprocal, it suffices to consider the integral

6106: over $s\in[0,1/2]$. Herein, with $t = e^{2\pi is}$, for

6107: $s\in (1/3,1/2]$, we have real $e_{\pm}(t)$, one with norm $>1$. So

6108: \[

6109: \int_{1/3}^{1/2}\,\log\,|V_n(e^{2\pi is})|\,ds\,\ge\,O(n)\,.

6110: \]

6111: We must argue about $s\in[0,1/3]$. In that case, $e_{\pm}(t)$

6112: are conjugate complex numbers of unit norm. This implies that

6113: the integrand is bounded above when $n\to\infty$, but our problem

6114: is to show that it does not decrease too quickly with $n$.

6115: The assignment (with $i=\sqrt{-1}$)

6116: \[

6117: t\,\longmapsto\,f(t)\,=\,-\frac{t+1/t-1}{2}\,+

6118: i\sqrt{ 1-\left( \frac{t+1/t-1}{2} \right)^2 }\,

6119: \]

6120: maps $e^{2\pi i[0,1/3]}\to e^{2\pi i[1/6,1/2]}$, so

6121: \[

6122: \frac{1}{2\pi}\,\arg t\,\stackrel{g}{\longmapsto}\,

6123: \frac{1}{2\pi}\,\arg f(t)

6124: \]

6125: is a map $g\,:\,[0,1/3]\to[1/6,1/2]$, which is

6126: bijective and monotonous, and $C^1$ except in $t=1/3$

6127: where $g'\to\infty$. In particular,

6128: \[

6129: \gm\,=\,\min\,\{\,g'(s)\,:\,s\in[0,1/3]\,\}\,>0\,.

6130: \]

6131: Now consider the functions 

6132: \[

6133: h(s)\,:=\,2\cos 2\pi ng(s)\quad\mbox{and}\quad

6134: m(s)\,:=\,-2\cos 2\pi s

6135: \]

6136: for $s\in[0,1/3]$. The singularities of the Jensen integral \eqref{Ji}

6137: correspond to the intersections of the graphs of $h$ and $m$.

6138: 

6139: If $n\gg \gm$, then all intersection points\footnote{For the rest

6140: of the proof $i$ is used as an index rather than the complex unit.}

6141: $s_i$ of the graphs of $h$ and $m$ have $|m'(s_i)|>|h'(s_i)|$. (Here

6142: we use also that $m\ne h$ when $s=0$.) In particular, between two

6143: critical points $x_i$ of $h$, it intersects $m$ at most once. So the

6144: number $n'$ of intersections $s_i$ satisfies

6145: \[

6146: \left|\,n'-\frac{2}{3}n\,\right|\,\le\,O(1)\,,\mbox{\quad

6147: and therefore\quad}\,n'=O(n)\,.

6148: \]

6149: 

6150: Moreover, 

6151: \begin{eqn}\label{20.5}

6152: |m'(s_i)|-|h'(s_i)|\,\ge\,(2-|m(s_i)|)\,\cdot C\,,

6153: \end{eqn}

6154: for a constant $C$ independent on $i$ and $n$. Now when $n$

6155: increases, the $s_i$ will concentrate around $s=\frac{1}{3}$,

6156: but since $\gm>0$, for $s$ small, the $x_i$ will be at distance

6157: $\ge O(\frac{1}{n})$. So

6158: \begin{eqn}\label{20.6}

6159: 2-|m(x_i)|\ge C\cdot \frac{i^2}{n^2}

6160: \end{eqn}

6161: for a constant $C$ independent on $i$ and $n$. Then

6162: if $s_1,\dots,s_{n'}$ are the $n'$ intersection points

6163: of $h$ and $m$, we can estimate for values of $s$

6164: between the critical points $x_i$ and $x_{i+1}$ of $h$

6165: \def\tsi{\bigl(2-|m(s_i)|\bigr)}

6166: \begin{eqn}\label{x5}

6167: |m(s)-h(s)|\,\ge\,C\,\cdot \tsi\,\cdot |s-s_i|^2

6168: \end{eqn}

6169: for a number $C$ independent on $i$ and $s$. Inequality

6170: \eqref{x5} follows because we can control the behaviour

6171: of $|m(s)-h(s)|$ from \eqref{20.5} for $s$ around $s_i$,

6172: and though $|m-h|$ decreases slightly around the endpoints

6173: of the interval $[x_i,x_{i+1}]$, this decrease can be

6174: controlled by \eqref{20.6}. So

6175: \begin{eqn}\label{x4}

6176: \int_{x_i}^{x_{i+1}}\,\log\,|\,m(s)-h(s)\,|\,ds\,\ge\,

6177: (x_{i+1}-x_i)\Bigl(\,\log\,\tsi\,+\log C\,\Bigr)\,+\,

6178: \,2\,\int_{x_i}^{x_{i+1}}\,\log |s-s_i|\,ds\,.

6179: \end{eqn}

6180: Since for small $i$ (when $2-|m(s_i)|$ is

6181: small), the $x_i$ will be at distance $\ge O(\frac{1}{n})$, using

6182: \eqref{20.6} and $m(s)\le 1$ when $0\le s\le \myfrac{1}{3}$ we have

6183: \begin{eqn}\label{x2}

6184: \sum_{i=1}^{n'}\,\Bigl|\,\log\,\tsi\,\Bigr|\,\le\,n'\cdot

6185: \left(\,C\cdot\left|\int_{0}^{1}\log |x|\,dx\right|+C'\,\right)\,=

6186: \,O(n)\,.

6187: \end{eqn}

6188: 

6189: Let $d_i:=x_{i+1}-x_i$. Because of $\gm>0$, we have 

6190: \begin{eqn}\label{63'}

6191: \max_{i}\,d_i\,\le\,O(1/n)\,.

6192: \end{eqn}

6193: So, similarly as in \eqref{x2}

6194: \begin{eqn}\label{x3}

6195: \sum_{i=1}^{n'}\,\Bigl|\,d_i\,\log\,\tsi\,\Bigr|\,

6196: \le O(1)\,.

6197: \end{eqn}

6198: Also, because of \eqref{63'} and $\int 1+\log x\ dx=x\cdot\log x+C$,

6199: for $n$ large

6200: \[

6201: \left|\,\int_{x_i}^{x_{i+1}}\,\bigl(\,1+\log |s-s_i|\,\bigr)\,ds\,

6202: \right|\,\le\,

6203: -C\cdot d_i\cdot \left[\,\log d_i\,+\,C'\,\right]\,\le\,

6204: \sqrt{d_i}\,\le\,O\left(\frac{1}{\sqrt{n}}\right)\,

6205: \]

6206: for some real number $C'$ independent on $i$ and $n$, and so

6207: \[

6208: \sum_{i=1}^{n'}\,\left|\,\int_{x_i}^{x_{i+1}}\,2\cdot

6209: \log |s-s_i|\,ds\,\right|\,\le\,O(\sqrt{n})\,.

6210: \]

6211: This, \eqref{x3} and \eqref{x4} imply that for $s\in[0,1/3]$ the

6212: (part of the) Jensen integral

6213: \eqref{Ji} has a lower bound that behaves like $-O(\sqrt{n})$, so the

6214: dominating part is for $s\in[1/3,1/2]$, and we are done for $V$.

6215: 

6216: For $\Dl$ the argument is similar. There $m(s)\equiv 2$, so

6217: instead of \eqref{x5} we have for $s\in[x_i,x_{i+1}]$ the estimate

6218: $|m(s)-h(s)|\ge |s-s_i|^2\cdot C\cdot \gm$, with the same further

6219: reasoning.  \qed

6220: }

6221: 

6222: \begin{rem}

6223: Dan Silver pointed out that for the Alexander polynomial a proof

6224: can be obtained (possibly more elegantly, but not immediately)

6225: from the ideas in \cite{SW}.

6226: \end{rem}

6227: 

6228: \subsubsection{Skein polynomial and generalized twisting}

6229: 

6230: By using the representation theory, one can extend the scope of

6231: the results in \cite{SSW,CK} on bounded Mahler measure

6232: to (parallel) multi-strand twisting.

6233: We give a version for the skein polynomial, since its special case,

6234: the Jones polynomial, was discussed in \cite{CK}. One can obtain

6235: this special case by setting $\lm=q$ in the below theorem. (The

6236: reversal of component orientation is not a serious problem for

6237: $V$, unlike for $P$.) We write as before $P_L(v,z)$ for the skein

6238: polynomial and use the skein rule $v^{-1}P_+-vP_-=zP_0$.

6239: 

6240: A \em{template} $T$ consists, as in \cite{SunThis}, of a number of

6241: strands, that create crossings and \em{slots}, only that for us

6242: strands are oriented and slots have the more general form

6243: (with an arbitrary number of in- and outputs)

6244: \[

6245: \diag{7mm}{3}{2}{

6246:   \picmultigraphics{5}{0.5 0}{\picvecline{0.5 0}{0.5 2}}

6247:   \picfilledbox{1.5 1}{2.8 1}{}

6248: }\,.

6249: \]

6250: A(n oriented) diagram $D$ is \em{associated} to $T$, if $D$ is

6251: obtained from $T$ by inserting into each slot of $T$ a braid of a

6252: certain number of full twists (on the proper number of strands).

6253: 

6254: The $1$-norm $||P||_1$ of a (Laurent) polynomial $P$ (even in

6255: several variables) is understood to be the sum of absolute values

6256: of all its (non-zero) coefficients (taken in all variables).

6257: We write $M(P)$ for its Mahler measure. It is known that for

6258: polynomials $P$ (of real coefficients), $M(P)\le ||P||_1$

6259: (see \S 2 of \cite{SSW}).

6260: 

6261: \begin{theorem}

6262: For each $n$ there is a polynomial $D_n(q)$ such that for

6263: each template of $n_1,\dots,n_k$-strand parallel twist

6264: slots and each diagram $D$ associated to $T$ we have

6265: \[

6266: \left|\left|\es\prod_{i=1}^k D_{n_i}(q)\es\cdot\es

6267: P_D(\lm,\sqrt{q}-1/\sqrt{q})\es\right|\right|_1\le C_T\,,

6268: \]

6269: where $C_T$ is a constant that depends only on $T$.

6270: Furthermore $D_n$ are made of a product of terms $1-q^{l}$

6271: for some $1\le l\le n$. In particular, 

6272: \[

6273: \{\,M(\,P_D(\lm,\sqrt{q}-1/\sqrt{q})\,)\,:\,\mbox{\ $D$ is

6274: associated to $T$\,}\,\}

6275: \]

6276: is bounded.

6277: \end{theorem}

6278: 

6279: \proof Write $D(t_1,\dots,t_k)=D_{t_1,\dots,t_k}$ for the diagram

6280: associated to $T$ by putting into the $i$-th slot $t_i$ full

6281: twists (on $n_i$ strands).

6282: 

6283: Let first $k=1$. Then the complement of the slot is a room with

6284: $n=n_1$ parallel in- and outputs. The skein module of such

6285: a room is generated by (positive permutation) braids, and

6286: thus it suffices to work with the $n$-strand braid

6287: group. Again, by \cite[Proposition 6.2]{Jones}

6288: \[

6289: P_L(\lm,\sqrt{q}-1/\sqrt{q})=X_L(q,\lm^2/q)\,,

6290: \]

6291: where

6292: % $X_L(q,\lm)$ is the invariant treated in \cite{Jones}.

6293: % By the work of \cite{Jones} we know that

6294: $X_L$ is a writhe-strand normalized (Definition 6.1 in \cite{Jones})

6295: weighted trace sum (equation 5.5), and the full twist has the

6296: effect of multiplying the traces by a power of $q$ (Lemma 9.3).

6297: 

6298: All terms in the demoninator of the (rational) generating series 

6299: \begin{eqn}\label{GS}

6300: \sum_{t=0}^{\infty}\,X(D_t)(q,\lm^2/q)\,x^t\,

6301: \end{eqn}

6302: are obtained from demoninators in the definition

6303: of $X_L(q,\lm)$ and quantities that differ by the

6304: number $t=t_1$ of twists. To identify these terms we 

6305: will work up to units in $\bZ[q^{\pm 1/2},\lm^{\pm 1/2}]$

6306: that do not depend on the number $t$ of twists but just 

6307: on $T$. The quantities depending on $t$ are only

6308: the power of $q$ in the traces and the term $\sqrt

6309: {\lm}^e$ in Definition 6.1, where $e$ is the exponent sum.

6310: Since the full twist has $e=n(n-1)$, we find the terms

6311: $1-\lm^{n(n-1)/2} q^jx$ for some values of $j\ge 0$. The

6312: exponent $n(n-1)/2$ becomes $n(n-1)$ and $j\ge -n(n-1)/2$

6313: when replacing $\lm^2/q$ for $\lm$. It suffices now

6314: to collect the demoninator terms occurring in the

6315: definition of $X_L$ (and replace $\lm^2/q$ for $\lm$).

6316: {}From Definition 6.1 we have $\lm^{(n-1)/2}$ (which is

6317: a unit, fixed for given $T$, and remains so when

6318: replacing $\lm^2/q$ for $\lm$) and some power of $1-q$.

6319: {}From the text below Figure 5.6 and (5.5) we have

6320: some power of $1-\lm q$ (that becomes $1-\lm^2$ under

6321: substitution) and (in $Q(q)$) products of terms $1-q^l$

6322: for some $1\le l\le n$ (with $l$ being a hook length of a

6323: box in a Young diagram of $n$ boxes). Thus the demoninator

6324: of \eqref{GS} is something we can take to be $D_n(q)$ times 

6325: a product of terms of the form $1-\lm^2$ and $1-\lm^{n

6326: (n-1)}q^{j}x$. 

6327: 

6328: Now if $k>1$, we first observe that $X(D_{t_1,\dots,t_k})(q,\lm)$

6329: satisfy a linear recurrence (with coefficients in $\bZ[

6330: q^{\pm 1/2},\lm^{\pm 1/2}]$) in $t_i$ for any fixed value

6331: of $t_j, j\ne i$. Moreover that recurrence itself does not

6332: depend on $t_j, j\ne i$ for fixed $i$, only its initial values

6333: do depend. Then one inductively argues over $k$ that the

6334: generating series 

6335: \[

6336: \sum_{t_1,\dots,t_k\ge 0} X(D_{t_1,\dots,t_k})(q,\lm^2/q)\es

6337: x_1^{t_1} \dots x_k^{t_k}

6338: \]

6339: has a demoninator which is the product of the $D_{n_i}$ and terms

6340: $1-\lm^2$ and $1-\lm^{n_i(n_i-1)}q^{j}x_i$ for $1\le i\le k$ and $j\ge 

6341: -n_i(n_i-1)/2$. The cases when $t_i<0$ are analogous. From this

6342: one easily concludes the claim up to the factors $1-\lm^2$.

6343: Since for fixed $T$ the power $p_T$ of $1-\lm^2$ is fixed,

6344: and the $\lm$-span of $P_D$ is bounded by the Morton-Williams-%

6345: Franks inequality (see proposition 15.1 of \cite{Jones}),

6346: one can get disposed of the $1-\lm^2$

6347: factors by linear combinations of the coefficients of

6348: $(1-\lm^2)^{p_T}P_D\cdot \prod D_{n_i}$ in the powers

6349: of $\lm$. \qed

6350: 

6351: 

6352: \subsection{Knots with unsharp MWF inequality}

6353: 

6354: This brief part was motivated by the paper of Kawamuro

6355: \cite{Kawamuro}, which I came across during the continuous work

6356: on my paper. Kawamuro was interested in finding infinitely many

6357: knots for which the MWF inequality is strict, in particular such

6358: satisfying Jones' conjecture in \S\ref{SJS}. The point to make

6359: is that we have at least two ways allowing us to obtain her, and

6360: further such knots much more easily. (In particular it is not

6361: necessary to appeal to the heavy geometric machinery of

6362: Neumann-Rudolph-Giroux.)

6363: 

6364: We already saw an infinite family of examples with unsharp MWF

6365: (and satisfying Jones' conjecture) in the proof of theorem

6366: \ref{thJC}. This family is, of course, little insightful,

6367: since it is settled already by Birman-Menasco's 3-braid

6368: work. The 4-braid examples they proposed, but could not

6369: decide about, deserve more attention. As given in figure 4

6370: of \cite{Kawamuro}, consider for $(x,y,z,w)\in\bZ^4$ the

6371: 4-braids $\bt_{x,y,z,w}=[212^213^x2^y-12^z3^w]$, and let

6372: $K_{x,y,z,w}=\hat\bt_{x,y,z,w}$. Birman-Menasco observe that

6373: $MWF(K_{x,y,z,w})\le 3$ (see lemma 2.9 in \cite{Kawamuro}),

6374: but suspect that many of the $K_{x,y,z,w}$ have braid index 4.

6375: One can exhibit an infinite family of $K_{x,y,z,w}$ of $b=4$,

6376: and thus recover theorem 2.8 of \cite{Kawamuro}, like this.

6377: 

6378: \begin{prop}

6379: There is a natural number $N$ and 6 tuples $(\ap_1,\dots,

6380: \ap_4)\in \bZ^4$ of different mod-2-reductions in $(\bZ/2)

6381: ^4$, such that if $K_\gm:=K_{\gm_1,\dots,\gm_4}$ is a knot,

6382: then $b(K_\gm)=4$ whenever $(\ap_i)\equiv (\gm_i)\bmod

6383: 2$ and $\gcd\Bigl(\ffrac{\gm_i-\ap_i}{2}\Bigr)\ge N$.

6384: \end{prop}

6385: 

6386: \proof Since (complex) roots of unity are dense on the complex

6387: unit circle, choose a root of unity $t$, for which the 3-braid

6388: test in corollary \ref{pp1} applies for $9_{42}$ or $9_{49}$ (see

6389: example \ref{4249}). Then, using the formula for $\Dl$ in lemma 3.1

6390: of \cite{SSW}, we see that for each such $t$ there is an $n\in2\bZ

6391: \sm\{0\}$, such that $\Dl_{\hat\bt}(t)$ is preserved when $\bt$

6392: is replaced by $\sg_i^n\bt$.

6393: 

6394: Moreover, it is easy to check that for all 6 of the 16 possible

6395: vectors of parities of $(x,y,z,w)$, for which $K_{x,y,z,w}$ is a

6396: knot, there is a representation of (at least one of) $9_{42}$ or

6397: $9_{49}$ of that parity combination. (The representations given

6398: after definition 2.7 of \cite{Kawamuro} show 3 of the parity

6399: types.) Now, take such a representation $\bt_{\ap_1,\dots,\ap_4}$

6400: and vary the parameters $\ap_i$ by multiples of $n$. Then, since

6401: we can choose (for proper $t$) $n$ to be any sufficiently

6402: large even natural number, we obtain in the claim. \qed

6403: 

6404: Of course, this elementary argument could be further concretified

6405: and strengthened. Presumably, one can do much better using

6406: J.~Murakami's test theorem \reference{TMF} (since it is an equality).

6407: Note that our proof does not confirm Jones' conjecture on any

6408: $K_{x,y,z,w}$, so it may trigger the question: would it help

6409: to find a counterexample? This seems very optimistic, though,

6410: as we will soon see from an alternative approach to Kawamuro's

6411: theorem, using the work in \S\ref{SJS}. First we can settle

6412: (also with regard to Jones' conjecture) the concrete examples

6413: $K_{-1,-2,m,2}$ for $m\ge 2$ even, obtained in her proof.

6414: 

6415: \begin{prop}\label{pZ}

6416: We have $b(K_{-1,-2,m,2})=4$ for $m\ne -1,0$.

6417: \end{prop}

6418: 

6419: \proof We give just a brief explanation. We try to calculate the

6420: skein polynomial of the 2-cable $\bt^{[2]}_{-1,-2,m,2}$ of $\bt_

6421: {-1,-2,m,2}$, obtained by replacing $\sg_i$ by $\sg_{2i}\sg_{2i+1}

6422: \sg_{2i-1}\sg_{2i}$, and we may consider just the coefficient

6423: of $v^{4[\bt_{-1,-2,m,2}]+7}$. Again $\pi_Y([2312])$ has at

6424: most 7 different eigenvalues $\dl_i$, for all $Y\vdash 8$ taken

6425: together. But now we have no full twist, and so do not need to

6426: deal with the various $Y$ (and their weights $W_Y$ etc.) one by

6427: one. We can sum $W_Y\cdot c(i,Y)$ over $Y\vdash 8$, and are left

6428: with 7 different $q$-rational expressions $c_i$ to determine. 

6429: Again this can be done from 7 explicit polynomials, which can be

6430: obtained using Morton's program. (If we focus on one parity of $m$

6431: only, we can again reduce the unknowns $c_i$ and test polynomials 

6432: to 5, but need to calculate polynomials from slightly more

6433: complicated braids.)

6434: 

6435: We calculated in fact 11 polynomials, for $|m|\le 5$, and

6436: used those 9 for $|m|\le 4$ in the determination of $c_i$

6437: to have extra safety. The resulting linear

6438: equation system for $c_i$ (with matrix $\{\dl_i^m\}$) is now

6439: not a serious problem to MATHEMATICA, which gives the solution

6440: (for generic $q$) in just a few seconds. Clearing denominators,

6441: and looking already at the extremal $q$-degrees, we see that for

6442: $|m|\ge 6$ among the 7 terms $c_i\dl_i^m$ the one for $i=2$ has

6443: the lowest minimal or highest maximal degree (as a unique term,

6444: except for $m=6$, where we must use also that the leading

6445: coefficient of $c_6\dl_6^m$ has the same sign). With an

6446: explicit check for the other $m$, we conclude the claim. \qed

6447: 

6448: This method of proof has also the advantage of easily leading

6449: to a qualitative improvement of Kawamuro's theorem, which is

6450: closer to what one should expect.

6451: 

6452: \begin{prop}

6453: The tuples $(x,y,z,w)$ for which $K_{x,y,z,w}$ has braid index 4 (and

6454: satisfies Jones' conjecture) are generic in $\bZ^4$, in the sense that

6455: \begin{eqn}\label{bla}

6456: \lim_{n\to\infty}\es

6457:   \frac{

6458:    \big|\,

6459:     \{\,(x,y,z,w)\in \bZ^4\,:\,b(K_{x,y,z,w})=4\,\}\,\cap\, {[-n,n]^4} 

6460:    \,\big|

6461:   \rule[-0.6em]{\z@}{1.3em}}

6462:   {\big|\,[-n,n]^4\,\big|\rule[0.1em]{\z@}{1.0em}}

6463: \es=\,1\,.

6464: \end{eqn}

6465: \end{prop}

6466: 

6467: \proof Extending the argument for proposition \ref{pZ}, we see that 

6468: \[

6469: \tl P(x,y,z,w):=

6470:   [P(\hat\bt^{[2]}_{x,y,z,w})]_{v^{4[\bt_{x,y,z,w}]+7}}(

6471:   \sqrt{q}-1/\sqrt{q})\cdot \sqrt{q}^{\,4[\bt_{x,y,z,w}]+7}

6472:   \,=\,\sum_{i_x,i_y,i_z,i_w=1}^7\,c_{i_x,i_y,i_z,i_w}\dl_{i_x}^x

6473:   \dl_{i_y}^y\dl_{i_z}^z\dl_{i_w}^w\,,

6474: \]

6475: where $c_{i_x,i_y,i_z,i_w}$ are again some rational expressions

6476: in $q$. They are obviously not all zero, and the polynomials $\tl

6477: P(x,y,z,w)$ have the following property: if one fixes three of the

6478: parameters (say $x,y,z$), and $\tl P(x,y,z,w)=0$ for 7 different

6479: values of the remaining parameter (here $w$), then it vanishes

6480: for all values of that parameter (at the given fixed 3 others).

6481: It is then not hard to deduce \eqref{bla} from this (see e.g.

6482: the proof of lemma 11.1 of \cite{gen2}). \qed

6483: 

6484: 

6485: 

6486: \noindent{\bf Acknowledgement.} I would like to thank to M. Hirasawa,

6487: M. Ishiwata and K.~Murasugi for their contribution, and S.~Kamada,

6488: K.~Kawamuro, T.~Nakamura and D.~Silver for some helpful remarks.

6489: The calculations were performed largely using the programs of

6490: \cite{MorSho2} and \cite{KnotScape}, and MATHEMATI\-CA\TM{}

6491: \cite{Wolfram}. I also wish to thank to my JSPS host Prof. T.~Kohno

6492: at University of Tokyo for his support.

6493: 

6494: % \vspace{1cm}\mbox{}

6495: % \newpage

6496: 

6497: \begin{appendix}

6498: 

6499: \renewcommand{\thesection}{\Alph{section}}

6500: {\def\@seccntformat#1{Appendix \csname the#1\endcsname . \quad}

6501: 

6502: \section{Postliminaries}

6503: }

6504: % \def\@seccntformat#1{\csname the#1\endcsname . \quad}

6505: 

6506: \subsection{Fibered Dean knots (Hirasawa-Murasugi)\label{HM}}

6507: 

6508: Here we present some material due to Hirasawa and Murasugi,

6509: who studied fibering of generalized Dean knots. An overview is given

6510: in \cite{Hirasawa}. As this is work in progress, a longer exposition

6511: may appear subsequently elsewhere.

6512: 

6513: \begin{defi}

6514: The Dean knot $K(p,q|r,rs)$ is given by the closed $p$-braid

6515: \[

6516: (\sg_{p-1}\sg_{p-2}\cdots \sg_1)^{q}(\sg_1\sg_2\cdots \sg_{r-1})^{rs}\,,

6517: \]

6518: with $p>r>1$ and $q,s$ non-zero integers such that $(q,p)=1$.

6519: \end{defi}

6520: 

6521: Hirasawa and Murasugi proposed a conjecture on these knots

6522: and obtained so far the following partial results.

6523:  

6524: \begin{conjecture}(Hirasawa-Murasugi)

6525: A Dean's knot $K(p,q|r,rs)$ is a fibred knot if and

6526: only if its Alexander polynomial is monic, that is,

6527: $\Mc\Dl=\pm 1$.

6528: \end{conjecture}

6529: 

6530: \begin{figure}[th]

6531: \[

6532: \begin{array}{ccc}

6533: \epsfs{6cm}{fig2a} & 

6534: \epsfs{5cm}{fig2b} & 

6535: \epsfs{5cm}{fig2c}\\

6536:  (a) & (b) & (c) \\

6537: \end{array}

6538: \]

6539: \caption{\label{fig2}}

6540: \end{figure}

6541: 

6542: \begin{prop}(Hirasawa-Murasugi)

6543: This conjecture has been proven for the following cases.

6544: \def\theenumi{\alph{enumi}}

6545: \def\labelenumi{(\theenumi)}

6546: \begin{enumerate}

6547: \item     $q = kp + 1$, and $r$ and $s$ are arbitrary,

6548: \item     $q =kp - 1$, and  $r$ and $s$ are arbitrary,

6549: \item     $r =  p - 1$, and  $q$ and $s$ are arbitrary.

6550: \end{enumerate}

6551: \end{prop}

6552: 

6553: The last case implies in particular that the conjecture is

6554: true for $p=3$. Below follows a part of the argument in this

6555: case that settles lemma \reference{lki}. (Other parts of their

6556: proof are very similar to some of our previous arguments.

6557: It seems, for example, that Hirasawa and Murasugi were to

6558: some extent aware of Theorem \reference{Thq}.)

6559: 

6560: \proof [of lemma \reference{lki}] We consider the braids $[(123)^k-2]$,

6561: with $k>0$ fixed. By isotopies and Hopf plumbings, we modify our

6562: surface 

6563: \begin{eqnarray*}

6564: % & (123)^k-2\to 1223(123)^{k-1}-2\to -21223(123)^{k-1}= & \\

6565: % & 12 -123(123)^{k-1}\to 23(123)^{k-1}12 -1= 2(312)^k-1\to

6566: % (2312)^k-1\,.

6567: &\rx{-1mm} [(123)^k -2] \to [(1223)^k - 2]=[12(2312)^{k-1}23 - 2] \to

6568: [(2312)^{k-1}23 - 212] = [(2312)^{k-1}2312 -1]=[(2312)^k - 1].

6569: \end{eqnarray*}

6570: Let $F$ be the surface of the band representation $\bt=[(2312)^k-1]$.

6571: We will show that $F$ is a fiber surface.

6572: 

6573: Consider the subsurface $F_0$ of $F$ that spans $[2 3 1 2]$

6574: in the natural manner.

6575: We deform the $k$ copies of $F_0$ to the (isotopic sub)surfaces

6576: $F'_0$ by a series of diagrams, see figure \reference{fig2}.

6577: (Here strands are numbered from right to left and words composed

6578: downward.)

6579: 

6580: In figure \reference{fig2}, for the move (a) $\to$ (b), we slide

6581: $B$ and $C$, respectively, along $D$ and $A$, then delete $D$

6582: (by deplumbing a Hopf band), and then slide $C$ back along

6583: $A$. For the move (b) $\to$ (c), we slide $C$ along $B$,

6584: then slide $B$ along $C$. The $k$ bands $A$ can be subsequently

6585: removed by Murasugi desumming a $(2,k)$-torus link fiber surface.

6586: Thus the surface $F$, spanned by $\hat\bt$, turns after

6587: (de)summing Hopf bands into a surface $F'$ consisting of $k$

6588: copies of $F'_0$ and one negative band $N$. See figure \ref{fig3} (a).

6589:  

6590: 

6591: {

6592: % \tm

6593: \def\@captype{figure}

6594: \long\def\@makecaption#1#2{%

6595:    % \tm

6596:    \vskip 10pt

6597:    {\let\label\@gobble

6598:    \let\ignorespaces\@empty

6599:    \xdef\@tempt{#2}%

6600:    %\typeout{`#2'}%

6601:    }%

6602:    \ea\@ifempty\ea{\@tempt}{%

6603:    \setbox\@tempboxa\hbox{%

6604:       \fignr#1#2}%

6605:       }{%

6606:    \setbox\@tempboxa\hbox{%

6607:       {\fignr#1:}\capt\ #2}%

6608:       }%

6609:    \ifdim \wd\@tempboxa >\captionwidth {%

6610:       \rightskip=\@captionmargin\leftskip=\@captionmargin

6611:       \unhbox\@tempboxa\par}%

6612:    \else

6613:       \hbox to\captionwidth{\hfil\box\@tempboxa\hfil}%

6614:    \fi}%

6615: % %

6616: % \def\fignr{\small\sffamily\bfseries}%

6617: % \def\capt{\small\sffamily}%

6618: % 

6619: % 

6620: % \newdimen\@captionmargin\@captionmargin2cm\relax

6621: % \newdimen\captionwidth\captionwidth\hsize\relax

6622: 

6623: \

6624: \captionwidth0.3\textwidth\relax

6625: \[

6626: \begin{array}{cc@{\qquad}c}

6627: \epsfs{5cm}{fig3a} & \epsfs{5cm}{fig3b} & \epsfs{5cm}{surf} \\

6628: (a) & (b) \\[-2mm]

6629: \multicolumn{2}{c}{ \vbox{\caption{\label{fig3}}} } &

6630: \raisebox{1em}{\vbox{\caption{\label{fig4}}}}

6631: \end{array}

6632: \]

6633: 

6634: }

6635: 

6636: In figure \reference{fig3}, we perform the move (a) $\to$ (b), sliding

6637: $B$ along $N$. The last surface $F'$, in figure \reference{fig3} (b),

6638: is Murasugi sum of a fibre surface spanning the $(2,2,-2)$-pretzel

6639: link and $\tl F$, where $\tl F$ consists 

6640: of $k - 1$ copies of $F'_0$ and the band $N$. By induction on

6641: $k$, we see that $F'$ is a fibre surface, and hence $\hat\bt$

6642: is a fibred link. \qed

6643:  

6644: \subsection{$A$-decomposition (joint with Hirasawa-Ishiwata)\label{HI}}

6645: 

6646: For the proof of theorem \reference{thuq} we introduce the

6647: $A$-decomposition due to T.~Kobayashi \cite{Kobayashi}.

6648: 

6649: A \em{sutured manifold} in the sense of Gabai \cite{Gabai3} can be

6650: understood as a pair $(L,H)$ consisting of a closed 3-dimensional

6651: submanifold $H$ of $\bR^3$ with boundary $S=\prt H$ a connected

6652: surface, and a set of oriented loops $L\subset H$. We require that

6653: one can orient the connected components of $S\sm L$ so that the

6654: induced orientations on $L$ coincide from both sides of $L$ (in

6655: particular a connected component of $S\sm L$ never bounds to itself

6656: along a loop of $L$), and are given by the orientation of $L$.

6657: 

6658: Let $F$ be a connected Seifert surface of a(n oriented) link $L=\prt

6659: F$. We embed $F$ as $F\times\{0.5\}$ into the \em{bicolar} $H=

6660: F\times I$ (with $I=[0,1]$). Then $(L,H)$ becomes a sutured

6661: manifold. We call it \em{canonical sutured manifold} $C(F)$ of $F$.

6662: 

6663: We describe some basic operations on sutured manifolds $(L,H)$.

6664: 

6665: A \em{decomposition disk} $D$ is a disk with $P=\prt D\subset \prt

6666: H$, properly embedded in the complement of $H$ (i.e. $D\cap H=P$).

6667: We require that $D$ is not parallel to $S=\prt H$, and

6668: satisfies $P\cap F\ne\vn$. We assume also that the intersection of

6669: $P$ and $F$ is transversal, so that it is a collection of points.

6670: 

6671: Since $L=\prt F$ is separating on $S$, the intersection $D\cap L= P

6672: \cap F$ is an even number of points, and the orientation of $L$ at

6673: the intersection points is alternating (with respect to the

6674: orientation of the loop $P$). See figure \reference{fig5} (a).

6675: 

6676: \begin{figure}[th]

6677: \[

6678: \begin{array}{c@{\kern1.4cm}c}

6679: \diag{1cm}{3}{3}{

6680:   \piccircle{1.5 d}{0.6}{$D$}

6681:   \pictranslate{1.5 1.5}{

6682:     { \piclinedash{0.05 0.05}{0.02}

6683:       \picmultigraphics[rt]{4}{90}{

6684:         \picline{0.0 0.6}{-0.3 1.4}

6685:       }

6686:     }

6687:     \picputtext{0.4 0.9}{$L$}

6688:     \picmultigraphics[rt]{2}{180}{

6689:       \picvecline{0.3 1.4}{0.0 0.6 0.3 1.4 0.5 conv}

6690:       \picline{0.0 0.6}{0.0 0.6 0.3 1.4 0.5 conv}

6691:       \picrotate{90}{

6692:         \picvecline{0.0 0.6}{0.0 0.6 0.3 1.4 0.5 conv}

6693:         \picline{0.3 1.4}{0.0 0.6 0.3 1.4 0.5 conv}

6694:       }

6695:     }

6696:   }

6697: }

6698: &

6699: \diag{1cm}{2.5}{2.0}{

6700:   \picline{1 0}{1 2}

6701:   \picline{0 0.5}{1 0.5}

6702:   \picvecline{2 0.5}{1 0.5}

6703:   \picvecline{0 1.5}{2 1.5}

6704:   \picputtext{1 2.3}{$D$}

6705:   \picputtext{2.3 0.5}{$L$}

6706:   \picputtext{2.3 1.5}{$L$}

6707:   \picputtext{0.8 1.0}{$a$}

6708: }\quad\lra\quad

6709: \diag{1cm}{2}{2.0}{

6710:   \picline{1 0}{1 2}

6711:   \picputtext{1 2.3}{$D$}

6712:   \pictranslate{1 1}{

6713:     \picmultigraphics[rt]{2}{180}{

6714:       \picline{1 -0.5}{0.2 -0.5}

6715:       \piclineto{0.2 0.5}

6716:       \picveclineto{1 0.5}

6717:     }

6718:   }

6719: }   

6720: \\

6721: \ry{1.7em}(a) & \raisebox{0.8em}{$(b)$} \\

6722: \end{array}

6723: \]

6724: \caption{\label{fig5}}

6725: \end{figure}

6726: 

6727: Then $L\cap P$ separates $P$ into a collection of intervals or

6728: \em{arcs}. Let $a$ be such an arc. An \em{$A$-operation} on $D$ along

6729: $a$ is a transformation of $(L,H)$ into a sutured manifold $(L',H)$,

6730: where $L'$ is obtained by splicing $L$ along $a$. See figure

6731: \ref{fig5} (b).

6732: 

6733: A \em{product decomposition} along $D$ is a similar operation, due

6734: to Gabai \cite{Gabai3}, and can be described as an $A$-operation

6735: if $|L\cap D|=2$ (in which case which of the two arcs is chosen

6736: is irrelevant), followed by a subsequent gluing of a $D^2\times I$ into

6737: $H$ along a neighborhood $N(P)\simeq S^1\times I$ of $P$ on $S$.

6738: 

6739: \begin{defi}

6740: We define a sutured manifold $(L,H)$ to be \em{$A$-decomposable} as

6741: follows:

6742: \def\theenumi{\arabic{enumi}}

6743: \def\labelenumi{\theenumi)}

6744: \begin{enumerate}

6745: \item Assume $H$ is a standardly embedded handlebody (i.e. so that

6746: $\ol{S^3\sm H}$ is also one). If $L$ is a collection of trivial loops

6747: on $\prt H$, and all loops bound \em{disjoint} disks in $\prt H$,

6748: then $(L,H)$ is $A$-decomposable.

6749: %

6750: \item If $(L',H')$ is obtained from $(L,H)$ by a product decomposition

6751: (along some decomposition disk $D$),

6752: and $(L',H')$ is $A$-decomposable, then so is $(L,H)$.

6753: %

6754: \item Let $D$ be a decomposition disk of $H$ with $|L\cap D|=2n$ and

6755: choose among the $2n$ arcs on $P=\prt D$ a collection of $n$ cyclically

6756: consecutive arcs $a_1,\dots,a_n$. (Consecutive is to mean that, taken

6757: with their boundary in $L\cap P$, their union is a single interval

6758: in $P$, and not several such intervals.) Let $(L_i,H)$ be obtained

6759: from $(L,H)$ by $A$-decomposition on $D$ along $a_i$ for $i=1,\dots,n$.

6760: Then if all $(L_i,H)$ are $A$-decomposable, so is $(L,H)$.

6761: \end{enumerate}

6762: \end{defi}

6763: 

6764: \begin{theorem}(T.~Kobayashi \cite{Kobayashi}, O.~Kakimizu

6765: \cite{Kakimizu}) 

6766: \def\theenumi{\arabic{enumi}}

6767: \def\labelenumi{\theenumi)}

6768: \begin{enumerate}

6769: \item A fiber surface is a unique incompressible surface.

6770: \item The property a surface to be a unique incompressible surface 

6771:   is invariant under Hopf (de)plumbing.

6772: \item If $C(F)$ is $A$-decomposable, then $F$ is a

6773:   unique incompressible surface for $L=\prt F$.

6774: \end{enumerate}

6775: \end{theorem}

6776: 

6777: \begin{figure}%[th]

6778: \[

6779: \epsfsv{4.5cm}{dec1} \stackrel{a_1}{\lra} \epsfsv{4.7cm}{dec2}

6780: \]

6781: \[

6782: \epsfsv{4.5cm}{dec3} \qquad \epsfsv{4.5cm}{dec4} 

6783: \]

6784: \[

6785: \epsfsv{4.5cm}{dec5} \qquad \epsfsv{4.5cm}{dec6}

6786: \]

6787: \[

6788: \epsfsv{4.2cm}{dec7} \qquad \epsfsv{4.2cm}{dec8}

6789: \]

6790: \[

6791: \epsfsv{3.5cm}{dec9} \qquad \epsfsv{3.7cm}{dec10}

6792: \]

6793: \[

6794: \epsfsv{3.5cm}{dec11} \qquad \epsfsv{3.2cm}{dec12}

6795: \]

6796: \caption{\label{fig6} This sequence of diagrams describes the

6797: $A$-decomposition along arc $a_1$ of the canonical sutured

6798: manifold corresponding to the surface in figure \reference{fig4}.

6799: The $A$-operation is the change between the first and the second

6800: diagram. The following diagrams (left to right in each row)

6801: display isotopies and product decompositions, leading finally to

6802: a trivial curve on a torus. The $A$-operation along $a_2$

6803: must be performed too, but the decomposition is similar.}

6804: \end{figure}

6805: 

6806: Now we complete the proof of theorem \reference{thuq}.

6807: 

6808: \proof[of theorem \reference{thuq}] 

6809: Let $L$ be a 3-braid link. If $L$ is split, the splitting sphere

6810: also splits any incompressible surface for $L$. Since

6811: $L$ is a split union of a 2-braid link and an unknot

6812: (incl.~2 and 3-component unlinks), the claim is easy.

6813: Excluding split links, corollary \reference{cbd} shows we

6814: need to consider only connected Seifert surfaces. It also

6815: suffices to deal with the non-fibered links only. These are

6816: equivalent under Hopf (de)plumbing to some of $(123)^k$ ($k\ne

6817: 0$). For such links, we can modify the transformation of

6818: Hirasawa and Murasugi from \S\reference{HM}. Then we can turn

6819: by Hopf (de)plumbing the surfaces into those like in figure

6820: \reference{fig3} (a), consisting of $k$ copies of $F_0'$, but

6821: now \em{without} the lower band $N$. (Figure \reference{fig4}

6822: shows the case $k=4$.) Then we use $A$-decomposition, as

6823: shown in figure \reference{fig6}. One applies $A$-operations

6824: along the arcs $a_{1,2}$. We show only the result for $a_1$,

6825: the case of $a_2$ and other $k$ is analogous. \qed

6826: 

6827: \end{appendix}

6828: 

6829: % \vspace{1cm}\mbox{}

6830: 

6831: % \tm

6832: {\small

6833: \begin{thebibliography}{BLM}

6834: \bibitem[Al]{Alexander} J. W.~Alexander, \em{A lemma on systems of

6835:         knotted curves}, Proc. Nat. Acad. Sci. U.S.A.

6836:         {\bf 9} (1923), 93--95.

6837: \bibitem[Al2]{Alexander2} \bysame, \em{Topological invariants

6838:         of knots and links}, Trans.\ Amer.\ Math.\ Soc. {\bf 30} (1928),

6839:         275--306.

6840: \bibitem[Ar]{Artin} E.~Artin, \em{Theorie der Z\"opfe}, 

6841:         Abh. Math. Sem. Hamburgischen Univ. {\bf 4} (1926), 47--72.

6842: \bibitem[Ar2]{Artin2} \bysame, \em{Theory of braids}, 

6843:         Ann. of Math. {\bf 48(2)} (1947), 101--126.

6844: \bibitem[Be]{Bennequin} D.~Bennequin, \em{Entrelacements et \'equations

6845: 	de Pfaff}, Soc.\ Math.\ de France, Ast\'erisque {\bf 107-108}

6846: 	(1983), 87--161.

6847: \bibitem[Bi]{Bigelow} S.~Bigelow, \em{Representations of braid

6848: 	groups}, Proceedings of the International Congress of

6849: 	Mathematicians, Vol. {\bf II} (Beijing, 2002),  37--45.

6850: \bibitem[B]{Birman} J.\ S.\ Birman, \em{On the Jones polynomial of

6851: 	closed $3$-braids}, Invent. Math. {\bf 81(2)} (1985), 287--294.

6852: \bibitem[BKL]{BKL} \bysame, K.\ Ko and S.\ J.\ Lee, \em{A new

6853: 	approach to the word and conjugacy problems in the braid

6854: 	groups}, Adv. Math. {\bf 139(2)} (1998), 322--353.

6855: \bibitem[BM]{BirMen} \bysame\ and W.~W.~Menasco, \em{Studying knots

6856: 	via braids III: Classifying knots which are closed 3 braids},

6857: 	Pacific J.~Math. {\bf 161}(1993), 25--113.

6858: \bibitem[BM2]{BirMen2} \bysame\ and \bysame\,, \em{Studying links

6859: 	via closed braids II: On a theorem of Bennequin},

6860: 	Topology Appl. {\bf 40(1)} (1991), 71--82.

6861: \bibitem[BM3]{BirMen3} \bysame\, and \bysame\,, \em{Studying knots

6862: 	via braids VI: A non-finiteness theorem},

6863: 	Pacific J.~Math. {\bf 156} (1992), 265--285.

6864: \bibitem[BW]{BW} \bysame\ and H.\ Wenzl, \em{Braids, link polynomials

6865: 	and a new algebra}, Trans. Amer. Math. Soc. {\bf 313(1)}

6866: 	(1989), 249--273.

6867: \bibitem[Bl]{Bleiler} S.~A.~Bleiler, \em{Realizing concordant

6868: 	polynomials with prime knots}, Pacific J. Math. {\bf 100(2)}

6869: 	(1982), 249--257.

6870: \bibitem[BLM]{BLM} R.~D.~Brandt, W. B. R. Lickorish and K. Millett,

6871: 	\em{A polynomial invariant for unoriented knots and links},

6872: 	Inv. Math. {\bf 74} (1986), 563--573.

6873: \bibitem[CK]{CK} A.~Champanerkar and I.~Kofman, \em{On the Mahler

6874: 	measure of Jones polynomials under twisting},

6875: 	Algebr. Geom. Topol. {\bf 5} (2005), 1--22.

6876: \bibitem[Cr]{Cromwell} P.~R.~Cromwell, {\em Homogeneous links},

6877: 	J. London Math. Soc. (series 2) {\bf 39} (1989), 535--552.

6878: \bibitem[DL]{DasLin} O.~Dasbach and X.-S.~Lin, {\em On the Head and

6879: 	the Tail of the Colored Jones Polynomial}, Compositio Math.

6880: 	{\bf 142(5)} (2006), 1332--1342.

6881: \bibitem[EKT]{EKT} S.\ Eliahou, L.\ H. Kauffman and M.\ Thistlethwaite,

6882: 	{\em Infinite families of links with trivial Jones polynomial},

6883: 	Topology {\bf 42(1)} (2003), 155--169.

6884: \bibitem[Fi]{Fied} T.~Fiedler, {\em A small state sum for knots},

6885: 	Topology {\bf 32~(2)} (1993), 281--294.

6886: \bibitem[Fi2]{Fied2} \bysame, {\em Gauss sum invariants for knots

6887: 	and links}, Kluwer Academic Publishers, Mathematics and Its

6888: 	Applications Vol {\bf 532} (2001).

6889: \bibitem[FK]{FK} \bysame\, and V.\ Kurlin, \em{A one-parameter

6890: 	approach to knot theory}, preprint \web|math.GT/0606381|.

6891: \bibitem[FW]{WilFr} J.\ Franks and R.\ F.\ Williams, \em{Braids and the

6892: 	Jones-Conway polynomial}, Trans.\ Amer.\ Math.\ Soc. {\bf 303}

6893: 	(1987), 97--108.

6894: \bibitem[F\&]{HOMFLY} P. Freyd, J. Hoste, W. B. R. Lickorish,

6895: 	K. Millett, A. Ocneanu and D. Yetter, {\it A new polynomial

6896: 	invariant of knots and links}, Bull. Amer. Math. Soc.

6897: 	{\bf 12} (1985), 239--246.

6898: \bibitem[Fu]{Fujii} H.~Fujii, \em{Geometric indices and the Alexander

6899: 	polynomial of a knot},

6900: 	Proc. Amer. Math. Soc. {\bf 124(9)} (1996), 2923--2933.

6901: \bibitem[Ga]{Gabai} D.~Gabai, \em{The Murasugi sum is a natural

6902: 	geometric operation}, Low-dimensional topology (San Francisco,

6903: 	Calif., 1981), Contemp. Math. {\bf 20}, 131--143, Amer. Math.

6904: 	Soc., Providence, RI, 1983.

6905: \bibitem[Ga2]{Gabai2} \bysame, \em{The Murasugi sum is a natural

6906: 	geometric operation II}, Combinatorial methods in

6907: 	topology and algebraic geometry (Rochester, N.Y., 1982),

6908: 	93--100, Contemp. Math. {\bf 44}, Amer. Math. Soc.,

6909: 	Providence, RI, 1985.

6910: \bibitem[Ga3]{Gabai3} \bysame, \em{Detecting fibred links in $S^3$},

6911:  	Comment. Math. Helv.  {\bf 61(4)} (1986), 519--555.

6912: \bibitem[Gr]{Garside} F.~Garside, \em{The braid group and other groups},

6913: 	Q.~J.~Math.~Oxford {\bf 20} (1969), 235--264.

6914: \bibitem[HM]{Hirasawa} M.~Hirasawa and K.~Murasugi, \em{Double-torus

6915: 	fibered knots and pre-fiber surfaces}, Musubime to Teijigen

6916: 	Topology (Dec. 1999), 43--49.

6917: \bibitem[HT]{KnotScape} J.~Hoste and M.~Thistlethwaite, {\em

6918: 	KnotScape}, a knot polynomial calculation and table access

6919: 	program, available at \webb:|http://www.math.utk.edu/\~morwen|.

6920: \bibitem[HTW]{HTW} \bysame, \bysame\, and J.~Weeks, \em{The first

6921: 	1,701,936 knots}, Math. Intell. {\bf 20 (4)} (1998), 33--48.

6922: \bibitem[I]{Ishikawa} M.~Ishikawa, {\em On the Thurston-Bennequin

6923: 	invariant of graph divide links}, Math. Proc. Cambridge

6924: 	Philos. Soc. {\bf 139(3)} (2005), 487--495.

6925: \bibitem[J]{Jones} V.~F.~R.~Jones, {\em Hecke algebra representations 

6926: 	of braid groups and link polynomials}, Ann. of Math.

6927: 	{\bf 126} (1987), 335--388.

6928: \bibitem[J2]{Jones2} \bysame, {\em A polynomial

6929: 	invariant of knots and links via von Neumann algebras},

6930: 	Bull. Amer. Math. Soc. {\bf 12} (1985), 103--111.

6931: \bibitem[Kk]{Kakimizu} O.~Kakimizu, \em{Classification of the

6932: 	incompressible spanning surfaces for prime knots of $\leq 10$

6933: 	crossings}, Hiroshima Math. J. {\bf 35} (2005), 47--92.

6934: \bibitem[K]{Kanenobu} T.~Kanenobu, \em{Relations between the

6935: 	Jones and Q polynomials of 2-bridge and 3-braid links},

6936: 	Math. Ann. {\bf 285} (1989), 115--124.

6937: \bibitem[K2]{Kanenobu2} \bysame, \em{Examples on polynomial invariants

6938: 	of knots and links II}, Osaka J. Math. {\bf 26(3)} (1989),

6939: 	465--482.

6940: \bibitem[K3]{Kanenobu3} \bysame, \em{Examples on polynomial invariants

6941: 	of knots and links}, Math. Ann. {\bf 275} (1986), 555--572.

6942: \bibitem[K4]{Kanenobu4} \bysame, \em{An evaluation of the first

6943: 	derivative of the $Q$ polynomial of a link},

6944: 	Kobe J. Math. {\bf 5(2)} (1988), 179--184.

6945: \bibitem[Ka]{Kauffman} L.\ H.\ Kauffman, {\em An invariant of regular

6946: 	isotopy}, Trans. Amer. Math. Soc. {\bf 318} (1990), 417--471.

6947: \bibitem[Ka2]{Kauffman2} \bysame, {\em State models and

6948: 	the Jones polynomial}, Topology {\bf 26} (1987), 395--407.

6949: \bibitem[Kw]{Kawamuro} K.~Kawamuro, \em{The algebraic crossing number

6950: 	and the braid index of knots and links},

6951: 	Algebr. Geom. Topol. {\bf 6} (2006), 2313--2350. 

6952: \bibitem[Ki]{Kidwell} M.~E.~Kidwell, \em{On the degree of the

6953: 	Brandt-Lickorish-Millett-Ho polynomial of a link},

6954: 	Proc. Amer. Math. Soc. {\bf 100(4)} (1987), 755--762.

6955: \bibitem[Kn]{Kneissler} J. A. Kneissler, \em{Woven braids and their

6956: 	closures}, J. Knot Theory Ramifications {\bf 8(2)} (1999),

6957: 	201--214.

6958: \bibitem[Ko]{Kobayashi} T.~Kobayashi, \em{Uniqueness of minimal

6959: 	genus Seifert surfaces for links},

6960: 	Topology Appl. {\bf 33(3)} (1989), 265--279.

6961: \bibitem[Kr]{Kreimer} D.\ Kreimer, \em{Knots and Feynman diagrams},

6962: 	Cambridge Lecture Notes in Physics {\bf 13},

6963: 	Cambridge University Press, Cambridge, 2000.

6964: \bibitem[LM]{LickMil} W. B. R. Lickorish and K.~C.~Millett, {\em A

6965: 	polynomial invariant for oriented links}, Topology

6966: 	{\bf 26 (1)} (1987), 107--141.

6967: \bibitem[LT]{LickThis} \bysame\, and M.~B.~Thistlethwaite,

6968: 	\em{Some links with non-trivial polynomials and their crossing

6969: 	numbers}, Comment. Math. Helv. {\bf 63} (1988), 527--539.

6970: \bibitem[MT]{MenThis} W.~W.~Menasco and M.~B.~Thistlethwaite, \em{%

6971:   	The Tait flyping conjecture}, Bull. Amer. Math. Soc. {\bf

6972:     	25 (2)} (1991), 403--412.

6973: \bibitem[Mo]{Morton} H.~R.~Morton, \em{Seifert circles and knot

6974: 	polynomials}, Proc. Camb. Phil. Soc. {\bf 99} (1986), 107--109.

6975: \bibitem[Mo2]{MortonPb} \bysame\ (ed.), {\em Problems},  in

6976: 	``Braids'', Santa Cruz, 1986 (J.~S.~Birman and A.~L.~Libgober,

6977: 	eds.), Contemp. Math. {\bf 78}, 557--574.

6978: \bibitem[MS]{MorSho}  \bysame\ and H. B. Short, \em{The

6979: 	$2$-variable polynomial of cable knots}, Math. Proc.

6980: 	Cambridge Philos. Soc. {\bf 101(2)} (1987), 267--278.

6981: \bibitem[MS2]{MorSho2} \bysame\ and \bysame, {\tt br9z.p}, a Pascal

6982: 	program for calculation of the skein polynomial from braids,

6983: 	\web|http://www.liv.ac.uk/~su14/knotprogs.html|.

6984: \bibitem[Mr]{Murakami} J.\ Murakami, \em{The Kauffman polynomial of

6985: 	links and representation theory},

6986: 	Osaka J. Math. {\bf 24(4)} (1987), 745--758.

6987: \bibitem[Mu]{Murasugi} K.~Murasugi, \em{On the braid index of

6988: 	alternating links}, Trans. Amer. Math. Soc. {\bf 326 (1)}

6989: 	(1991), 237--260.

6990: \bibitem[Mu2]{Murasugi2} \bysame, \em{On closed 3-braids},

6991:   	Memoirs AMS {\bf 151} (1974), AMS, Providence.

6992: \bibitem[Mu3]{Murasugi3} \bysame\,, \em{Jones polynomial and classical

6993: 	conjectures in knot theory}, Topology {\bf 26} (1987), 187--194.

6994: \bibitem[MP]{MurPrz} \bysame\ and J.~Przytycki, \em{The skein

6995: 	polynomial of a planar star product of two

6996: 	links}, Math. Proc. Cambridge Philos. Soc. {\bf 106(2)}

6997: 	(1989), 273--276.

6998: \bibitem[MP2]{MurPrz2} \bysame\ and \bysame, \em{An index of a graph

6999: 	with applications to knot theory}, Mem. Amer. Math. Soc.

7000: 	{\bf 106 (508)} (1993).

7001: \bibitem[Na]{Nakamura} T.~Nakamura, \em{Notes on the braid index of

7002: 	closed positive braids}, Topology Appl. {\bf 135(1-3)}

7003: 	(2004), 13--31.

7004: \bibitem[Na2]{Nakamura2} \bysame, \em{Braidzel surfaces and the

7005: 	Alexander polynomial}, Proceedings of the Workshop

7006: 	``Intelligence of Low Dimensional Topology'', Osaka

7007: 	City University (2004), 25--34.

7008: \bibitem[Ni]{Ni} Y.~Ni, \em{Closed 3-braids are nearly fibred},

7009: 	preprint \web|math.GT/0510243|.

7010: \bibitem[Oh]{Ohyama} Y.~Ohyama, \em{On the minimal crossing number

7011: 	and the braid index of links}, Canad. J. Math. {\bf 45(1)}

7012: 	(1993), 117--131.

7013: \bibitem[Or]{Orevkov} S.~Orevkov, \em{Quasipositivity problem for

7014: 	3-braids}, Turkish Journal of Math. {\bf 28} (2004),

7015: 	89--93; also available at

7016: 	\web|http://picard.ups-tlse.fr/homepage/orevkov.html|.

7017: \bibitem[PV]{VirPol} M.~Polyak and O.~Viro,

7018: 	{\em Gauss diagram formulas for Vassiliev invariants},

7019: 	Int.\ Math.\ Res.\  Notes {\bf 11} (1994), 445--454.

7020: \bibitem[PV2]{VirPol2} \bysame\, and \bysame, {\em On the Casson

7021: 	knot invariant}, Knots in Hellas '98, Vol. {\bf 3} (Delphi),

7022: 	J. Knot Theory Ramifications {\bf 10(5)} (2001), 711--738.

7023: \bibitem[Ro]{Rolfsen} D.~Rolfsen, {\em Knots and links}, Publish

7024: 	or Perish, 1976.

7025: \bibitem[Ru]{Rudolph} L.\ Rudolph, \em{Braided surfaces and

7026: 	Seifert ribbons for closed braids}, Comment. Math. Helv.

7027: 	{\bf 58} (1983), 1--37.

7028: \bibitem[Ru2]{Rudolph2} \bysame, \em{Quasipositivity as an

7029: 	obstruction to sliceness}, Bull. Amer. Math. Soc. (N.S.)

7030: 	{\bf 29(1)} (1993), 51--59.

7031: \bibitem[Sc]{Sc} O.~Schreier, \em{\"Uber die Gruppen $A^aB^b=1$},

7032: 	Abh. Math. Sem. Univ. Hamburg {\bf 3} (1924), 167--169.

7033: \bibitem[SSW]{SSW} D.~Silver, A.~Stoimenow and S.~G.~Williams, \em{%

7034: 	Euclidean Mahler measure and Twisted Links},

7035: 	\web|math.GT/0412513|,

7036: 	Algebr. Geom. Topol. {\bf 6} (2006), 581--602.

7037: \bibitem[SW]{SW} \bysame\, and S.~Williams, \em{Coloring link diagrams

7038: 	with a continuous palette}, Topology {\bf 39} (2000),

7039: 	1225--1237.

7040: \bibitem[Sq]{Squier} C.\ Squier, \em{The Burau representation is

7041:   	unitary}, Proc.\ Amer.\ Math.~Soc. {\bf 90} (1984), 199--202.

7042: \bibitem[St]{posqp} A.~Stoimenow, {\em On polynomials and surfaces of

7043: 	variously positive links}, \web|math.GT/0202226|,

7044: 	Jour. Europ. Math. Soc. {\bf 7(4)} (2005), 477--509.

7045: \bibitem[St2]{3br} \bysame, {\em The skein polynomial of closed

7046: 	3-braids}, J. Reine Angew. Math. {\bf 564} (2003), 167--180.

7047: \bibitem[St3]{posex_bcr} \bysame, {\em On the crossing number of

7048:    	positive knots and braids and braid index criteria of Jones

7049: 	and Morton-Williams-Franks},

7050:         Trans.\ Amer.\ Math.\ Soc. {\bf 354(10)} (2002), 3927--3954.

7051: \bibitem[St4]{ntriv} \bysame, {\em Coefficients and non-triviality

7052: 	of the Jones polynomial}, preprint {\tt math.GT/0606255}.

7053: \bibitem[St5]{gener} \bysame, {\em Diagram genus, generators and

7054: 	applications}, preprint.

7055: \bibitem[St6]{bind} \bysame, {\em The braid index and the growth of

7056: 	Vassiliev invariants}, J.~Of Knot Theory and Its Ram.

7057: 	{\bf 8(6)} (1999), 799--813.

7058: \bibitem[St7]{pos} \bysame, {\em Positive knots, closed braids,

7059: 	and the Jones polynomial}, {\tt math.GT/9805078}, Ann. Scuola

7060: 	Norm. Sup. Pisa Cl. Sci. {\bf 2(2)} (2003), 237--285.

7061: \bibitem[St8]{gen2}\bysame, {\em Knots of genus two}, preprint

7062: 	\web|math.GT/0303012|.

7063: \bibitem[SV]{SV} \bysame\ and A.\ Vdovina, {\em Counting alternating

7064:   	knots by genus}, Math.\ Ann. {\bf 333} (2005), 1--27.

7065: \bibitem[ST]{SunThis} C.~Sundberg and M.~B.~Thistlethwaite, {\em The

7066: 	rate of growth of the number of prime alternating links and

7067: 	tangles}, Pacific Journal of Math. {\bf 182 (2)} (1998),

7068: 	329--358.

7069: \bibitem[Th]{Thistle} M.~B.~Thistlethwaite, {\em On the Kauffman

7070: 	polynomial of an adequate link}, Invent.\ Math. {\bf 93(2)}

7071: 	(1988), 285--296.

7072: \bibitem[Th2]{Thistle2} \bysame, {\em A spanning tree expansion for the

7073: 	Jones polynomial}, Topology {\bf 26} (1987), 297--309.

7074: \bibitem[Tr]{Traczyk} P.~Traczyk, \em{$3$-braids with proportional

7075:   	Jones polynomials}, Kobe J. Math. {\bf 15(2)} (1998), 187--190.

7076: \bibitem[Vo]{Vogel} P.~Vogel, \em{Representation of links by braids:

7077: 	A new algorithm}, Comment. Math. Helv. {\bf 65} (1990),

7078: 	104--113.

7079: \bibitem[Wi]{Williams} R. F. Williams, \em{Lorenz knots are prime},

7080: 	Ergodic Theory Dynam. Systems  {\bf 4(1)} (1984), 147--163. 

7081: \bibitem[Wo]{Wolfram} S. Wolfram, {\em Mathematica --- a system

7082: 	for doing mathematics by computer}, Addison-Wesley, 1989.

7083: \bibitem[Xu]{Xu} Peijun Xu, {\em The genus of closed $3$-braids},

7084: 	J. Knot Theory Ramifications {\bf 1(3)} (1992), 303--326.

7085: \bibitem[Y]{Yamada} S.~Yamada, \em{The minimal number of Seifert

7086: 	circles equals the braid index}, Invent. Math. {\bf 88}

7087: 	(1987), 347--356.

7088: \bibitem[Yo]{Yokota} Y.\ Yokota, {\em Polynomial invariants of

7089: 	positive links}, Topology {\bf 31(4)} (1992), 805--811.

7090: \end{thebibliography}

7091: }

7092: 

7093: \end{document}

7094:  

7095: