1: \documentstyle[12pt,psfig]{article}

2: 

3: 

4: \def\figura#1#2{#2}

5: 

6: 

7: 

8: \def\hatM{{\widehat{M}}}

9: \def\hatv{{\widehat{v}}}

10: \def\hattt{{\widehat{\calt}}}

11: \def\caltbar{\overline{\calt}}

12: \def\hatpp{{\widehat{\pp}}}

13: \def\interior{{\rm Int}}

14: \def\index{{\rm ind}}

15: \def\vbar{{\overline v}}

16: \def\Wbar{{\overline W}}

17: \def\Kbar{{\overline K}}

18: \def\Ebar{{\overline E}}

19: \def\fbar{{\overline f}}

20: \def\hbar{{\overline h}}

21: \def\GL{{\rm GL}}

22: \def\defin#1#2{\smallskip\noindent{\bf Definition (#1).} #2}

23: \def\Ccell{{\rm C}^{\rm cell}}

24: \def\Cphi{{\rm C}^\varphi}

25: \def\Xw{X_{\rm w}}

26: \def\taubar{\overline{\tau}}

27: \def\tauw{\tau_{\rm w}}

28: \def\taubarw{\taubar_{\rm w}}

29: \def\basis{{\hbox{\Got b}}}

30: \def\hbasis{{\hbox{\Got h}}}

31: \def\gbasis{{\hbox{\Got g}}}

32: \def\orient{{\hbox{\Got o}}}

33: \def\ristr#1{\big|_{#1}}

34: \def\move#1{\mathop{

35: \begin{picture}(10,5)

36: \put(0,-3.8){$\tilde{ }$}

37: \put(1,1){$\to$}

38: \end{picture}}

39: \limits^{#1}}

40:  

41: 

42: 

43: \def\stwotriv{S^2_{\rm triv}}

44: \def\bthtr{B^3_{\rm triv}}

45: 

46: 

47: \textwidth=15.5truecm

48: \textheight=21truecm

49: \topmargin=0truecm

50: \oddsidemargin=.2truecm

51: \evensidemargin=-.8truecm

52: 

53: 

54: \marginparsep=.15truecm

55: \marginparwidth=2.0truecm

56: \reversemarginpar

57: \def\nota#1{\marginpar{\tiny #1}}

58: 

59: 

60: 

61: \makeatletter

62: \def\@begintheorem#1#2{\it \trivlist \item[\hskip \labelsep{\bf #1\ #2.}]}

63: \newtheorem{teo}{Theorem}[section]

64: \newtheorem{rem}[teo]{Remark}

65: \newtheorem{lem}[teo]{Lemma}

66: \newtheorem{sublem}[teo]{Sublemma}

67: \newtheorem{cor}[teo]{Corollary}

68: \newtheorem{exa}[teo]{Example}

69: \newtheorem{prop}[teo]{Proposition} 

70: \newtheorem{que}[teo]{Question} 

71: 

72: \def\finedim#1{{\hfill\hbox{\enspace\fbox{\ref{#1}}}}\vspace{5pt}}

73: \def\dim#1{\vspace{1pt}\noindent{\it Proof of} {\hspace{2pt}}\ref{#1}.}

74: \def\compo{\,{\scriptstyle\circ}\,}

75: \def\cont{{\rm C}}

76: \def\combd{{\rm Comb}^\partial}

77: \def\objd{{\rm Obj}^\partial}

78: \def\recd{r^\partial}

79: \def\leg{{\rm Leg}}

80: 

81: 

82: \font\capfont=cmr10

83: \font\scpicc=cmcsc10

84: \font\sc=cmcsc10 scaled 1200

85: \def\mycaption#1{\caption{{\small #1}}} 

86: \def\mylabel#1{\label{#1}}

87: 

88: \newfont{\Bbb}{msbm10 scaled 1200}

89: \def\mr{{\hbox{\Bbb R}}}

90: \def\mc{{\hbox{\Bbb C}}}

91: \def\mz{{\hbox{\Bbb Z}}}

92: \def\mq{{\hbox{\Bbb Q}}}

93: \def\mh{{\hbox{\Bbb H}}}

94: \def\mk{{\hbox{\Bbb K}}}

95: \def\mn{{\hbox{\Bbb N}}}

96: \def\matp{{\hbox{\Bbb P}}}

97: 

98: \newfont{\mycal}{eusm10 scaled 1200}

99: \def\ttt{{\hbox{\mycal T}}}

100: 

101: \newfont{\Got}{eufm10 scaled 1200}

102: \def\permu{{\hbox{\Got S}}}

103: 

104: \def\ff{{\cal F}}

105: \def\cc{{\cal C}}

106: \def\bb{{\cal B}}

107: \def\vv{{\cal V}}

108: \def\uu{{\cal U}}

109: \def\ss{{\cal S}}

110: \def\mm{{\cal M}}

111: \def\nn{{\cal N}}

112: \def\calt{{\cal T}}

113: \def\oo{{\cal O}}

114: \def\gg{{\cal G}}

115: \def\ee{{\cal E}}

116: \def\rr{{\cal R}}

117: \def\kk{{\cal K}}

118: \def\hh{{\cal H}}

119: \def\pp{{\cal P}}

120: \def\qq{{\cal Q}}

121: \def\dd{{\cal D}}

122: 

123: \def\trasvint{\cap\hspace{-8.5pt}|\hspace{5pt}}

124:  

125: \def\eul{{\rm Eul}}

126: \def\euls{{\rm Eul}^{\rm s}}

127: \def\eulc{{\rm Eul}^{\rm c}}

128: \def\Thetas{\Theta^{\rm s}}

129: \def\Thetac{\Theta^{\rm c}}

130: \def\alphas{\alpha^{\rm s}}

131: \def\alphac{\alpha^{\rm c}}

132: 

133: 

134: 

135: \title{Reidemeister Torsion of 3-Dimensional\\ Euler Structures

136: with Simple Boundary Tangency\\ and Pseudo-Legendrian Knots}

137: 

138: 

139: \author{Riccardo Benedetti\qquad Carlo Petronio\thanks{The second named

140: author gratefully acknowledges financial support by GNSAGA-CNR}}

141: 

142: 

143: 

144: 

145: 

146: \begin{document}

147: 

148: \maketitle

149: 

150: \noindent{\small{\scpicc Abstract}. We generalize Turaev's definition of torsion invariants of pairs $(M,\xi)$, where $M$ is a 3-dimensional manifold and $\xi$ is an Euler structure on $M$ 

151: (a non-singular

152: vector field up to homotopy relative to $\partial M$  and local modifications in $\interior(M)$). Namely, we

153: allow $M$ to have arbitrary boundary and $\xi$ to have simple (convex and/or concave) tangency circles to

154: the boundary.  We prove that Turaev's $H_1(M)$-equivariance formula holds also in  our generalized context.

155: Our torsions apply in particular to (the exterior of) pseudo-Legendrian knots 

156: ({\em i.e.}~knots transversal to a given vector field),

157: and hence to Legendrian knots in contact

158: 3-manifolds. We show that torsion, as an

159: absolute invariant, contains a lifting to pseudo-Legendrian knots of the classical

160: Alexander invariant. We also precisely analyze the information carried by torsion

161: as a relative invariant of pseudo-Legendrian knots which are framed-isotopic.

162: Using branched standard spines to describe

163: vector fields we

164: show how to explicitly invert Turaev's reconstruction map  from combinatorial to smooth Euler structures,

165: thus making the computation of torsions a more effective one. As an example we work out a specific

166: calculation.}

167: 

168: \vspace{.5cm}

169: 

170: \noindent{\small{\scpicc Mathematics Subject Classification (1991)}: 57N10

171: (primary), 57Q10, 57R25 (secondary).}

172: 

173: 

174: 	

175: \vspace{.5cm}

176: 

177: 

178: \section*{Introduction}

179: Reidemeister torsion is a classical yet very vital topic in

180: 3-dimensional topology, and it was recently used in a variety of important

181: developments. To mention a few, torsion is a fundamental ingredient of the

182: Casson-Walker-Lescop invariants (see {\em e.g.}~\cite{lescop}), and more

183: generally of the perturbative approach to quantum invariants (see {\em

184: e.g.}~\cite{lemuoh}). Relations have been pointed out between torsion and

185: hyperbolic geometry~\cite{porti}. Turaev's torsion of Euler 

186: structures~\cite{turaev:Euler} has recently been 

187: recognized by Turaev himself (\cite{turaev:spinc},~\cite{turaev:nuovo}) to have deep

188: connections with the Seiberg-Witten

189: invariants of ${\rm Spin}^{\rm c}$-structures on 3-manifolds, after 

190: the proof of Meng and Taubes~\cite{meng:taub} that a suitable combination of these

191: invariants can be identified with the classical Milnor torsion.

192: 

193: 

194: 

195: Turaev's theory~\cite{turaev:Euler} actually exists in all dimensions. We

196: quickly review it before proceeding. A {\em smooth Euler structure} $\xi$ on a

197: compact oriented manifold $M$, possibly with $\partial

198: M=\emptyset$, is a non-singular vector field on $M$ viewed up to local

199: modifications  in $\interior(M)$ and  homotopy relative to $\partial M$. 

200: Orientability of $M$ is not strictly necessary, but we find

201: it convenient to assume it. Turaev

202: allows only ``monochromatic'' boundary components, {\it i.e.}~black ones (on

203: which the field points outwards) and white ones (on which it points inwards).

204: This implies the constraint that $\chi(M,W)=0$, where $W$ is the white portion

205: of  $\partial M$, but in~\cite{turaev:spinc} and~\cite{turaev:nuovo} Turaev

206: only focuses on the more specialized case where $M$ is 3-dimensional and closed

207: or bounded by tori. In all dimensions, the set $\euls(M,W)$ of smooth Euler

208: structures compatible with $(M,W)$ is an affine space over  $H_1(M;\mz)$. The

209: two main ingredients of Turaev's theory are as follows. First, he defines a

210: certain set of $1$-chains, called the space $\eulc(M,W)$ of {\em combinatorial}

211: Euler structures compatible with $(M,W)$, he shows that this is again affine

212: over $H_1(M;\mz)$,  and he describes an $H_1(M;\mz)$-equivariant bijection

213: $\Psi:\eulc(M,W)\to\euls(M,W)$ called the {\em reconstruction map}. Second, for

214: $\xi\in\eulc(M,W)$ and for  any representation $\varphi$ of $\pi_1(M)$ into the

215: units of a suitable ring $\Lambda$ he defines a torsion  invariant

216: $\tau^\varphi(M,\xi)$, or more generally $\tau^\varphi(M,\xi,\hbasis)$, with

217: values in $K_1(\Lambda)/(\pm1)$. This invariant is by definition a lifting of

218: the classical Reidemeister torsion (see~\cite{milnor}) $\tau^\varphi(M)\in

219: K_1(\Lambda)/(\pm\varphi(\pi_1(M)))$, and it satisfies the

220: $H_1(M;\mz)$-equivariance formula \begin{equation} \label{tau:equivariance}

221: \tau^\varphi(M,\xi',\hbasis)=\tau^\varphi(M,\xi,\hbasis)\cdot \varphi(\xi'-\xi)

222: \end{equation} where $\xi'-\xi\in H_1(M;\mz)$. For $\xi\in\euls(M,W)$ one

223: defines $\tau^\varphi(M,\xi)$ as $\tau^\varphi(M,\Psi^{-1}(\xi))$, and the 

224: $H_1(M;\mz)$-equivariance of the reconstruction map $\Psi$ implies that 

225: formula~(\ref{tau:equivariance}) holds also for smooth structures. We emphasize

226: that the definition of $\Psi$ is based on an explicit geometric construction,

227: but its bijectivity is only established through $H_1(M;\mz)$-equivariance. This

228: makes the definition of torsion for smooth structures somewhat implicit.

229: 

230: In the present paper, and in other papers in preparation, we are concerned with

231: generalizations and improvements of Turaev's theory. Here we consider

232: 3-manifolds. This work had two main initial aims. Our first aim was to find a

233: geometric description of the map $\Psi^{-1}$, and hence to turn the computation

234: of Turaev's torsion into a more effective procedure, using our 

235: encoding~\cite{lnm} of non-singular vector fields up to homotopy (also

236: called ``combings'') in terms of branched standard spines. Our second aim was

237: to define torsion invariants of pseudo-Legendrian pairs $(v,L)$, 

238: consisting of a link $L$ transversal to a non-singular vector field $v$, viewed up to 

239: {\em pseudo-Legendrian isotopy}, namely transversality-preserving

240: simultaneous isotopy of $L$ and homotopy of $v$. For a given $v$ we will just say

241: that $L$ is pseudo-Legendrian in $(M,v)$, or just in $v$. Note that

242: an ordinary Legendrian link in a given oriented contact structure $\xi$

243: is pseudo-Legendrian in $\xi^\perp$, and Legendrian isotopy implies pseudo-Legendrian

244: isotopy.

245: A specific motivation to look for invariants of pseudo-Legendrian

246: links comes  from the remarkable relation recently discovered by Fintushel and

247: Stern~\cite{fist} between  the Alexander polynomial ({\em i.e.}~Milnor torsion)

248: of a knot $K\subset S^3$ and  (a suitable combination of) the Seiberg-Witten

249: invariants of the  ``surgered'' $4$-manifold $X_K$ obtained using $K$ (and a

250: suitable base $4$-manifold $X$). Both our initial aims lead us to consider

251: Euler structures on $3$-manifolds $M$ (without restrictions on $\partial M$)

252: allowing simple tangency circles to $\partial M$ of  {\em concave} type (see

253: Fig.~\ref{conc:conv:tang} below). On the other  hand it turns out that, to

254: define torsion, the natural objects to deal with  are Euler structures with

255: {\em convex} tangency  circles. It is a fortunate fact, peculiar of dimension

256: 3, that there is a canonical way to associate  a convex field to any {\em

257: simple} ({\em i.e.}~mixed concave and convex) one. This allows to define

258: torsion for all smooth simple Euler structures, and eventually to achieve both

259: the objectives we had in mind.

260: 

261: Let us now summarize the contents of this paper. The foundational part of

262: our work consists in extending to the

263: context of Euler structures with simple tangency the notions of combinatorial

264: structure $\eulc$ and reconstruction map $\Psi$. 

265: This part follows the same scheme as~\cite{turaev:Euler}

266: and relies on technical results of Turaev. Our main contribution here

267: is the proof that the natural transformations of a concave structure

268: into a convex one, viewed at the smooth level and at the combinatorial level,

269: actually correspond to each other under the reconstruction map

270: (Theorem~\ref{diagram:commutes}).

271: After setting the foundations, we prove the following main result

272: (stated informally here:

273: see Sections~\ref{Eul:def:section} 

274: and~\ref{spines:section} for precise definitions and statements).

275: 

276: 

277: 

278: \begin{teo}\label{informal:algorithmic}

279: Let $\xi$ be an Euler structure with concave

280: tangency circles. If $P$ is a branched standard spine 

281: which represents $\xi$, then $P$ allows to explicitly 

282: find a representative of $\Psi^{-1}(\xi)\in\eulc$, and hence to compute

283: the torsion of $\xi$ in terms of the finite combinatorial data which encode $P$.

284: \end{teo}

285: 

286: 

287: 

288: 

289: 

290: After the foundations, we concentrate on pseudo-Legendrian knots $(v,K)$, assuming

291: for simplicity the ambient manifold to be closed. The connection

292: comes from the fact that the restriction of $v$ defines

293: a concave Euler structure on the exterior $E(K)$ of $K$, with two parallel tangency lines

294: on $\partial E(K)$ determined by the framing defined by $v$ on $K$. We 

295: show that torsion, as an {\em absolute} invariant, contains (in a suitable sense)

296: a lifting to pseudo-Legendrian knots of the classical Alexander invariant.

297: Then we carefully analyze the {\em relative} information carried by torsion

298: for pseudo-Legendrian pairs $(v_0,K_0)$ and $(v_1,K_1)$ such that $(v_0,v_1)$ 

299: are homotopic to each other and $(K_0,K_1)$ are framed-isotopic to each other.

300: A relevant point which emerges from this analysis is that in general torsion does not

301: provide a single-valued relative invariant, because the

302: action of a certain mapping class group (which depends on the framed isotopy class only)

303: must be taken into account. This leads us to the notion of `good' framed knots,

304: for which the action is trivial, and the study of torsion is simpler. 

305: We show that many knots are good (for instance, all knots in a homology sphere are good,

306: and most knots with hyperbolic complement are good).

307: Concentrating on good knots we then prove that in a homology sphere the relative torsion

308: of two knots essentially coincides with the difference of their rotation numbers

309: (Maslov indices), so torsion basically detects whether the knots are isotopic through

310: pseudo-Legendrian immersions.

311: 

312: 

313: Moreover we analyze the effect on torsion

314: of the framed first Reidemeister move (which does not change the framing

315: but locally changes the winding number by $\pm2$), and we show that for

316: homology spheres the winding number is just the difference of Maslov indices,

317: thus getting an alternative proof of the relation between torsion and rotation number.

318: Using the fact (proved in~\cite{second:paper}) that framed isotopy is

319: generated by pseudo-Legendrian isotopy and the framed first Reidemeister move,

320: we then obtain several interesting consequences, among which we state the following:

321: 

322: \begin{teo}\label{informal:good:for:knots}

323: Consider pseudo-Legendrian knots

324: $(v_0,K_0)$ and $(v_1,K_1)$. Assume that $K_0$ is good and that the meridian of $K_0$ has

325: infinite order in $H_1(E(K_0);\mz)$. Then the knots are pseudo-Legendrian

326: isotopic if and only if they have trivial relative torsion invariants.

327: \end{teo}

328: 

329: This paper is organized

330: as follows. In Section~\ref{Eul:def:section} we provide the

331: formal definitions of smooth and combinatorial Euler structure.

332: In Section~\ref{torsion:def:section} we introduce torsion and

333: state the equivariance property. In Section~\ref{spines:section} we

334: show how branched standard spines can be used for computing torsion.

335: In Section~\ref{knots:section} we specialize to pseudo-Legendrian 

336: knot exteriors and analyze torsion both as an absolute and as a relative invariant.

337: In Section~\ref{exa:section} we carry out a specific computation using

338: the technology of Section~\ref{spines:section}. 

339: In Sections~\ref{Eul:def:section}

340: to~\ref{knots:section} proofs which are long and require the introduction of 

341: ideas and techniques not used elsewhere are omitted. Section~\ref{proofs} contains all

342: these proofs.

343: 

344: We conclude this introduction by announcing related results which we have

345: recently obtained and partially written down. In~\cite{second:paper} we extend

346: to the case with boundary our combinatorial presentation~\cite{lnm} of combed

347: manifolds in terms of branched spines, and we provide similar presentations

348: of framed and pseudo-Legendrian links, using $\cont^1$ diagrams on branched 

349: spines. Some results from~\cite{second:paper} are actually used also in the

350: present paper (see Sections~\ref{spines:section} and~\ref{knots:section}). 

351: In~\cite{third:paper} we use the results of~\cite{second:paper} to develop

352: an approach to torsion entirely based on combinatorial techniques, getting

353: slightly different generalizations of Turaev's theory. In~\cite{fourth:paper} we

354: generalize the theory of Euler structures and (with some restrictions) of

355: torsion to all dimensions and allowing any generic (Whitney-Morin-type)

356: tangency to the boundary.  

357: 

358: 

359: \section{Euler structures}\label{Eul:def:section}

360: In this section we define smooth and combinatorial

361: Euler structures and explain their correspondence.

362: Fix once and for ever a compact oriented 3-manifold $M$, possibly with 

363: $\partial M=\emptyset$.

364: Using the {\it Hauptvermutung}, we will always freely intermingle the

365: differentiable, piecewise linear and topological viewpoints. Homeomorphisms

366: will always respect orientations. All vector fields mentioned in this paper will be 

367: non-singular, and they will be termed just {\em fields} for the sake of brevity.

368: 

369: \paragraph{Smooth and combinatorial Euler structures}

370: We will call {\em boundary pattern} on $M$ a partition $\pp=(W,B,V,C)$ of

371: $\partial M$ where $V$ and $C$ are finite unions of 

372: disjoint circles, and $\partial W=\partial B=V\cup C$. In particular, $W$ and $B$ are

373: interiors of compact surfaces embedded in $\partial M$. Even if $\pp$ can actually 

374: be determined by less data, {e.g.} the pair $(W,V)$,

375: we will find it convenient to refer to $\pp$ as a quadruple.

376: Points of $W$, $B$, $V$ and $C$ will be called {\em white}, {\em black}, 

377: {\em convex} and {\em concave} respectively.

378: We define the set

379: of {\em smooth Euler structures} on $M$ compatible with $\pp$, denoted by

380: $\euls(M,\pp)$, as the set of equivalence classes of

381: fields on $M$ which point inside on $W$, point

382: outside on $B$ and have simple tangency to $\partial M$ 

383: of {\em convex} type along $V$ and {\em concave} type 

384: along $C$,

385: as shown in a cross-section in Fig.~\ref{conc:conv:tang}.

386: \begin{figure}

387: \centerline{\psfig{file=cvtang.eps,width=6cm}}

388: \caption{\label{conc:conv:tang} Convex (left) and concave (right) tangency to the boundary.}

389: \end{figure}

390: Two such fields are equivalent if they are obtained from each other by 

391: homotopy through fields of the same type and modifications supported into

392: interior balls. The following

393: variation on the Poincar\'e-Hopf formula is established in Section~\ref{proofs}:

394: 

395: \begin{prop}\label{p:h:formula}

396: $\euls(M,\pp)$ is non-empty if and only if $\chi(\Wbar)=\chi(M)$. 

397: \end{prop}

398: 

399: \noindent We remark here that $\chi(\Wbar)=\chi(W)$, $\chi(\overline{B})=\chi(B)$, $\chi(V)=\chi(C)=0$ and

400: $\chi(W)+\chi(B)=\chi(\partial M)=2\chi(M)$, so there are various ways to

401: rewrite the relation $\chi(\Wbar)=\chi(M)$, the most intrinsic of which is actually

402: $\chi(M)-(\chi(\Wbar)-\chi(C))=0$ (see below for the reason).

403: 

404: Now, given $\xi,\xi'\in\euls(M,\pp)$ we can choose generic representatives $v,v'$, so

405: that the set of points of $M$ where $v'=-v$ is a union of loops contained in the

406: interior of $M$. A standard procedure allows to give these loops a canonical orientation,

407: thus getting an element $\alphas(\xi,\xi')\in H_1(M;\mz)$. The following result is easily

408: obtained along the lines of the well-known analogue for closed manifolds.

409: 

410: \begin{lem}

411: $\alphas$ is well-defined and turns $\euls(M,\pp)$ into an affine space over $H_1(M;\mz)$.

412: \end{lem}

413: 

414: A (finite) cellularization $\cc$ of $M$ 

415: is called {\em suited} to $\pp$ if $V\cup C$ is

416: a subcomplex, so $W$ and $B$ are unions of cells. 

417: Here and in the sequel by ``cell'' we will always mean an {\em open} one.

418: Let such a $\cc$ be given.

419: For $\sigma\in\cc$ define $\index(\sigma)=(-1)^{{\rm dim}(\sigma)}$. We define

420: $\eulc(M,\pp)_\cc$ as the set of equivalence classes of integer singular 1-chains $z$ in $M$

421: such that 

422: $$\partial z=\sum_{\sigma\subset M\setminus (W\cup V)}\index(\sigma)\cdot p_\sigma$$

423: where $p_\sigma\in\sigma$ for all $\sigma$. Two chains $z$ and $z'$ 

424: with $\partial z=\sum\index(\sigma)\cdot p_\sigma$ and 

425: $\partial z'=\sum\index(\sigma)\cdot p'_\sigma$

426: are defined to be equivalent if there exist $\delta_\sigma:([0,1],0,1)\to

427: (\sigma,p_\sigma,p'_\sigma)$ such that

428: $$z-z'+\sum_{\sigma\subset M\setminus (W\cup V)}\index(\sigma)\cdot \delta_\sigma$$

429: represents $0$ in $H_1(M;\mz)$. Elements of

430: $\eulc(M,\pp)_\cc$ are called {\em combinatorial Euler structures} relative to $\pp$ and $\cc$,

431: and their representatives are called {\em Euler chains}.

432: The definition implies that, for $\xi,\xi'\in\eulc(M,\pp)_\cc$, their difference $\xi-\xi'$

433: can be defined as an element $\alphac(\xi,\xi')$ of $H_1(M;\mz)$. The following is easy:

434: 

435: \begin{lem}

436: $\eulc(M,\pp)_\cc$ is non-empty if and only if $\chi(\Wbar)=\chi(M)$, and in this case

437: $\alphac$ turns it into an affine space over $H_1(M;\mz)$.

438: \end{lem}

439: 

440: \noindent Since $\Wbar=W\cup V\cup C$, the alternating sum of dimensions of cells in $W\cup V$ is 

441: intrinsically interpreted as $\chi(\Wbar)-\chi(C)$, which explains why the most meaningful

442: way to write the relation $\chi(\Wbar)=\chi(M)$ is $\chi(M)-(\chi(\Wbar)-\chi(C))=0$.

443: From now on we will always assume that this relation holds.

444: Turaev~\cite{turaev:Euler} only considers the case where $V=C=\emptyset$, so 

445: $W=\overline{W}$ and $B=\overline{B}$, and our relation takes the usual form

446: $\chi(M,W)=0$.

447: The following result was established by Turaev in~\cite{turaev:Euler}

448: in his setting, but the proof

449: extends {\em verbatim} to our context, so we omit it.

450: Only the first assertion is hard.

451: We state the other two because we will use them.

452: 

453: \begin{prop}\label{combin:prop:statement}

454: \begin{enumerate}

455: \item If $\cc'$ is a subdivision of $\cc$ then there exists a canonical

456: $H_1(M;\mz)$-isomorphism $\eulc(M,\pp)_\cc\to\eulc(M,\pp)_{\cc'}$. In particular

457: $\eulc(M;\mz)$ is canonically defined up to $H_1(M;\mz)$-isomorphism independently 

458: of the cellularization.

459: \item\label{connected:spider:point} If 

460: $\cc$ is a cellularization of $M$ suited to $\pp$

461: and $x_0\in M$ is an assigned point, any element of $\eulc(M,\pp)$ can be represented,

462: with respect to $\cc$, as a sum $\sum_{\sigma\subset M\setminus(W\cup V)}

463: \index(\sigma)\cdot \beta_\sigma$

464: with $\beta_\sigma:([0,1],0,1)\to(M,x_0,\sigma)$.

465: \item\label{bary:sub:point} If $\calt$ is a triangulation of $M$ suited to $\pp$,

466: any element of $\eulc(M,\pp)$ can be represented, with respect to $\calt$, as 

467: a simplicial $1$-chain in the first barycentric subdivision of $\calt$.

468: \end{enumerate}

469: \end{prop}

470: 

471: 

472: 

473: \noindent Our first main result, proved in Section~\ref{proofs},

474: is the extension to the case under consideration of

475: Turaev's correspondence between $\eulc$ and $\euls$.

476: 

477: 

478: 

479: \begin{teo}\label{reconstruction:statement}

480: There exists a canonical $H_1(M;\mz)$-equivariant isomorphism 

481: $$\Psi:\eulc(M,\pp)\to\euls(M,\pp).$$

482: \end{teo}

483: 

484: \noindent The definition of $\Psi$ is based on an explicit geometric construction,

485: but its bijectivity is only established through $H_1(M;\mz)$-equivariance.

486: As already mentioned in the introduction, this makes in general a very difficult task to

487: determine the inverse of $\Psi$. One of the features of this paper is the description

488: of $\Psi^{-1}$ in terms of the combinatorial encoding of fields by 

489: means of branched spines: Theorem~\ref{spider:structure:teo} 

490: describes $\Psi^{-1}$ when $\pp$ is concave,

491: and Theorem~\ref{diagram:commutes} shows that from a general $\pp$ we can

492: effectively pass to a unique convex $\pp$, and hence to a unique concave $\pp$,

493: and conversely.

494: 

495: 

496: In view of Theorem~\ref{reconstruction:statement}, when no confusion risks to arise,

497: we shortly write $\eul(M,\pp)$ for either $\euls(M,\pp)$ or $\eulc(M,\pp)$,

498: and $\alpha$ for the map giving the affine $H_1(M;\mz)$-structure on this space.

499: 

500: 

501: 

502: \paragraph{Convex Euler structure associated to an arbitrary one}

503: Let $M$ and $\pp=(W,B,V,C)$ be as in the definition of $\eul(M,\pp)$.

504: The pattern $\theta(\pp)=(W,B,V\cup C,\emptyset)$ is a convex one canonically

505: associated to $\pp$. We define a map 

506: $$\Thetas:\euls(M,\pp)\to\euls(M,\theta(\pp))$$

507: as geometrically described in Fig.~\ref{conc:to:conv}. Concerning

508: this figure, note that the loops in $C$ can be oriented as 

509: components of the boundary of $B$, which is oriented as a subset of the boundary of $M$.

510: \begin{figure}

511: \centerline{\psfig{file=conc_to_conv.eps,width=10cm}}

512: \caption{\label{conc:to:conv} Turning a concave tangency circle $\gamma$ into a convex one: the 

513: apparent singularity in the cross-section is removed by adding a small bell-shaped field 

514: directed parallel to $\gamma$, {\em i.e.}~orthogonal to the cross-section.}

515: \end{figure}

516: 

517: \begin{lem}\label{concave:convex:smooth}

518: $\Thetas$ is a well-defined $H_1(M;\mz)$-equivariant bijection.

519: \end{lem}

520: 

521: \dim{concave:convex:smooth}

522: The first two properties are easy and imply the third property. The inverse of $\Thetas$

523: may actually be described geometrically by a figure similar to Fig.~\ref{conc:to:conv}, but we leave this to the reader.

524: \finedim{concave:convex:smooth}

525:  

526: We define now a combinatorial version of $\Thetas$. Consider a cellularization $\cc$ suited to 

527: $\pp$, and denote by $\gamma_1,\dots,\gamma_n$ the 1-cells contained in $C$. We choose the 

528: parameterizations $\gamma_j:(0,1)\to C$ so that they respect the natural orientation of $C$ 

529: already discussed above, and we extend the $\gamma_j$ to $[0,1]$, without changing notation.

530: Now let $z$ be an Euler chain relative to $\pp$. It 

531: easily seen that $z-\sum_{j=1}^n\gamma_j\ristr{[1/2,1]}$ is an Euler chain

532: relative to $\theta(\pp)$. Setting 

533: $$\Thetac([z])=\left[z-\sum_{j=1}^n\gamma_j\ristr{[1/2,1]}\right]$$ 

534: we get a map $\Thetac:\eulc(M,\pp)\to\eulc(M,\theta(\pp))$.

535: 

536: \begin{lem}\label{concave:convex:comb}

537: $\Thetac$ is a well-defined $H_1(M;\mz)$-equivariant bijection.

538: \end{lem}

539: 

540: \dim{concave:convex:comb}

541: Again, the first two properties are easy and imply the third one. 

542: \finedim{concave:convex:comb}

543: 

544: In Section~\ref{proofs} we will see the following:

545: 

546: 

547: 

548: 

549: \begin{teo}\label{diagram:commutes}

550: If $\Psi$ is the reconstruction map of Theorem~\ref{reconstruction:statement} then the

551: following diagram is commutative:

552: $$\matrix{

553: \eulc(M,\pp) & \stackrel{\Thetac}{\longrightarrow} & \eulc(M,\theta(\pp))\phantom{.} \cr

554: \Psi\downarrow\phantom{\Psi} & & \phantom{\Psi}\downarrow\Psi \cr

555: \euls(M,\pp) & \stackrel{\Thetas}{\longrightarrow} & \euls(M,\theta(\pp)). \cr}$$

556: \end{teo}

557: 

558: Using this result we will sometimes just write 

559: $\Theta:\eul(M,\pp)\to\eul(M,\theta(\pp))$.

560: 

561: 

562: \section{Torsion of an Euler structure}\label{torsion:def:section}

563: In this section we define torsion. We set up the usual

564: algebraic environment~\cite{milnor} in which 

565: torsion can be defined, fixing a ring $\Lambda$ with unit, with the property

566: that if $n$ and $m$ are distinct positive integers then $\Lambda^n$ and 

567: $\Lambda^m$ are not isomorphic as $\Lambda$-modules. The 

568: Whitehead group $K_1(\Lambda)$ is defined as the

569: Abelianization of $\GL_\infty(\Lambda)$, and $\Kbar_1(\Lambda)$ is the

570: quotient of $K_1(\Lambda)$ under the action of $-1\in\GL_1(\Lambda)=

571: \Lambda_*$. (Later in this paper the symbol $K_1$ will also be used for a knot,

572: but the meaning will always be clear from the context.)

573: 

574: 

575: We will directly define torsion only for a {\em convex} Euler structure,

576: but the definition easily extends to any Euler structure $\xi$ with simple

577: boundary tangency, taking the torsion of the convexified structure $\Theta(\xi)$.

578: So, we fix a manifold $M$, a {\em convex} boundary pattern $\pp=(W,B,V,\emptyset)$ on $M$,

579: a cellularization $\cc$ suited to $\pp$ and

580: a representation $\varphi:\pi_1(M)\to\Lambda_*$. 

581: We will denote by $\varphi$ again the extension

582: $\mz[\pi_1(M)]\to\Lambda$ (a ring homomorphism). 

583: 

584: We consider now the universal cover $q:\tilde{M}\to M$ and

585: the twisted chain complex $\Cphi_*(M,W\cup V)$, where

586: $\Cphi_i(M,W\cup V)$ is defined as 

587: $\Lambda\otimes_{\varphi}\Ccell_i(\tilde{M},q^{-1}(W\cup V);\mz)$,

588: and the boundary operator is induced from the ordinary boundary.

589: The homology of this complex is denoted by $H^\varphi_*(M,W\cup V)$ and

590: called the $\varphi$-twisted homology. We assume that each $H^\varphi_i(M,W\cup V)$

591: is a free $\Lambda$-module and fix a basis $\hbasis_i$.

592: 

593: \begin{rem}\label{twisted:remarks}{\em

594: \begin{enumerate}

595: \item\label{basepoint:remark} To have a formal completely intrinsic definition of  

596: $H^\varphi_*(M,W\cup V)$, one should fix from the

597: beginning a basepoint $x_0\in M$ for $\pi_1(M)$, and consider pointed

598: universal covers $q:(\tilde M,\tilde x_0)\to(M,x_0)$, because any two such covers

599: are {\em canonically} isomorphic, and the action of $\pi_1(M)$ on $\tilde{M}$ is

600: {\em canonically} defined on them. 

601: \item To define $H^\varphi_*(M,W\cup V)$

602: we have used in an essential way the fact that $W\cup V=\overline{W}$ is 

603: closed, because otherwise $\Cphi_*(M,W\cup V)$ cannot be defined.

604: \item\label{free:basis:remark} $\Cphi_i(M,W\cup V)$ is a free $\Lambda$-module, and

605: each $\mz[\pi_1(M)]$-basis of $\Ccell_i(\tilde{M},q^{-1}(W\cup V);\mz)$ determines a

606: $\Lambda$-basis of $\Cphi_i(M,W\cup V)$.

607: \item\label{varphibar:defin} If we compose $\varphi$ with the projection 

608: $\Lambda_*\to\Kbar_1(\Lambda)$ we get a homomorphism of $\pi_1(M)$ into an

609: {\em Abelian} group, so we get a homomorphism

610: $\overline{\varphi}:H_1(M;\mz)\to\Kbar_1(\Lambda).$

611: \end{enumerate}}

612: \end{rem}

613: 

614: \noindent Now let $\xi\in\eulc(M,\pp)$ and choose a representative of $\xi$ 

615: as in point~\ref{connected:spider:point} of 

616: Proposition~\ref{combin:prop:statement}, namely

617: $$\sum_{\sigma\in\cc,\ \sigma\subset M\setminus(W\cup V)}

618: \index(\sigma)\cdot \beta_\sigma$$

619: with $\beta_\sigma(0)=x_0$ for all $\sigma$, $x_0$ being a fixed point of $M$.

620: We choose $\tilde x_0\in q^{-1}(x_0)$ and consider the liftings

621: $\tilde\beta_\sigma$ which start at $\tilde x_0$. For $\sigma\subset M\setminus(W\cup V)$

622: we select its preimage $\tilde\sigma$ which contains $\tilde\beta_\sigma(1)$, and

623: define $\gbasis(\xi)$ as the collection of all these $\tilde\sigma$.

624: Arranging the $i$-dimensional elements of $\gbasis(\xi)$ in any order, by 

625: Remark~\ref{twisted:remarks}(\ref{free:basis:remark}) we

626: get a $\Lambda$-basis $\gbasis_i(\xi)$ of $\Cphi_i(M,W\cup V)$. 

627: We consider a set ${\tilde\hbasis}_i$ of elements of $\Cphi_i(M,W\cup V)$

628: which project to the fixed basis $\hbasis_i$ of $H^\varphi_i(M,W\cup V)$.

629: 

630: Now note that, given a free $\Lambda$-module $L$ 

631: and two finite bases $\basis=(b_k)$,

632: $\basis'=(b'_k)$ of $M$, the assumption made on $\Lambda$ guarantees that

633: $\basis $ and $\basis'$ have the same number of elements, so there

634: exists an invertible square matrix $(\lambda^h_k)$ such that

635: $b'_k=\sum_h\lambda^h_k b_h$. We will denote by

636: $[\basis'/\basis ]$ the image of $(\lambda^h_k)$ in $K_1(\Lambda)$.

637:  

638: 

639: 

640: 

641: \begin{prop}\label{defining:proposition}

642: If $\basis_i\subset\Cphi_i(M,W\cup V)$ is such that $\partial\basis_i$ is a

643: $\Lambda$-basis of $\partial(\Cphi_i(M,W\cup V))$, then

644: $(\partial\basis_{i+1})\cdot\tilde\hbasis_i\cdot\basis_i$ is a $\Lambda$-basis of

645: $\Cphi_i(M,W\cup V)$, and 

646: $$\tau^\varphi(M,\pp,\xi,\hbasis)=\pm

647: \prod_{i=0}^3\Big[ \Big( (\partial\basis_{i+1})\cdot\tilde\hbasis_i\cdot

648: \basis_i\Big) \;\Big/\;\gbasis_i(\xi)\Big]^{(-1)^i}\in

649: \Kbar_1(\Lambda)$$

650: is independent of all choices made. Moreover 

651: \begin{equation}

652: \tau^\varphi(M,\pp,\xi',\hbasis)=\tau^\varphi(M,\pp,\xi,\hbasis)\cdot 

653: \overline{\varphi}(\alphac(\xi',\xi)).\label{combinatorial:equivariance:formula}

654: \end{equation}

655: \end{prop}

656: 

657: \dim{defining:proposition}

658: The first assertion and independence of the $\basis_i$'s is purely algebraic and classical,

659: see~\cite{milnor}. Now note that $\xi\in\eulc(M,\pp)$ was used to select the bases 

660: $\gbasis_i(\xi)$. The $\gbasis_i(\xi)$ are of course not uniquely determined themselves,

661: but we can show that different choices lead to the same value of $\tau^\varphi$.

662: 

663: First of all, the arbitrary ordering in the $\gbasis_i(\xi)$ is inessential because

664: torsion is only regarded up to sign. Second, consider the effect of choosing a

665: different representative of $\xi$. This leads to a new family $\tilde\sigma'$ of

666: cells. If $\tilde\sigma'=a(\sigma)\cdot \tilde\sigma$, with $a(\sigma)\in\pi_1(M)$,

667: and $\overline{a}(\sigma)$ is the image in $H_1(M;\mz)$, we automatically have

668: $$\sum_{\sigma\subset M\setminus{W\cup V}}\index(\sigma)\cdot \overline{a}(\sigma)=0\in

669: H_1(M;\mz),$$

670: which allows to conclude that also the representative chosen is inessential.

671: The choice of the lifting $\tilde x_0$ can be shown to be inessential either

672: in the spirit of Remark~\ref{twisted:remarks}(\ref{basepoint:remark}),

673: or by showing that a simultaneous $a$-translation of all $\tilde\sigma$, for

674: $a\in\pi_1(M)$, multiplies the torsion by $\overline{\varphi}(a)^{\chi(M)-\chi(W\cup V)}=1$.

675: 

676: Formula~(\ref{combinatorial:equivariance:formula}) is readily established by choosing

677: representatives $\sum\index(\sigma)\cdot \beta_\sigma$ and 

678: $\sum\index(\sigma)\cdot \beta'_\sigma$ of $\xi$ and $\xi'$ 

679: such that $\beta'_\sigma=\beta_\sigma$ for all 

680: $\sigma$ but one.\finedim{defining:proposition}

681: 

682: Since the above construction uses the cellularization $\cc$ in a

683: way which may appear to be essential,

684: we add a subscript $\cc$ to the torsion we have defined.

685: The next result, which can be established following Turaev~\cite{turaev:Euler},

686: shows that dependence on $\cc$ is actually inessential.

687: 

688: \begin{prop}

689: Let $\cc$ and $\cc'$ be cellularizations suited to $\pp$. Assume that

690: $\cc'$ subdivides $\cc$, and consider the bijection

691: $\ss_{(\cc',\cc)}:\eulc(M,\pp)_\cc\to\eulc(M,\pp)_{\cc'}$ of

692: Proposition~\ref{combin:prop:statement}, and the canonical isomorphism

693: $j_{(\cc',\cc)}:H^\varphi_*(M,W\cup V)_\cc\to H^\varphi_*(M,W\cup V)_{\cc'}$. Then,

694: with obvious meaning of symbols we have:

695: $$\tau^\varphi_\cc(M,\pp,\xi,\hbasis)=

696: \tau^\varphi_{\cc'}(M,\pp,\ss_{(\cc',\cc)}(\xi),j_{(\cc',\cc)}(\hbasis)).$$

697: \end{prop}

698: 

699: 

700: 

701: It is maybe appropriate here to remark that the choice of a basis $\hbasis$  of

702: $H^\varphi_*(M,W\cup V)$ and the definition of $\tau^\varphi(M,\pp,\xi,\hbasis)$ implicitly 

703: assume a description of the universal cover of $M$, which is typically undoable in

704: practical cases. However,

705: if one starts from a representation of $\pi_1(M)$ into the units of a {\em commutative}

706: ring  $\Lambda$, {\em i.e.} a representation which factors through

707:  one of $H_1(M;\mz)$,

708: one can use from the very beginning the maximal Abelian rather

709: than the universal cover, which makes computations more feasible.

710: 

711: \begin{rem}{\em Turaev~\cite{turaev:Reidemeister} has shown that a homological

712: orientation yields a sign-refinement of torsion, {\em i.e.}~a lifting from 

713: $\Kbar_1(\Lambda)$ to $K_1(\Lambda)$. This refinement extends with minor

714: modifications to our setting of boundary tangency. This sign-refinement,

715: in the closed and monochromatic case, is often an essential component of the 

716: theory (for instance, it is crucial for the relation with

717: the 3-dimensional Seiberg-Witten invariants~\cite{turaev:spinc},~\cite{turaev:nuovo}

718: and for the definition of the Casson invariant~\cite{lescop}), but we will

719: not address it in the present paper.}

720: \end{rem}

721: 

722: 

723: 

724: \paragraph{Computation of torsion via disconnected spiders} 

725: In this paragraph we show that 

726: to determine the family of lifted cells necessary to define torsion one can use representatives

727: of Euler structures more general than those used above. This is a technical point which we

728: will use below to compute torsions using branched spines (Section~\ref{spines:section}).

729: 

730: 

731: We fix $M$, $\pp$, $\cc$ and $\varphi$ as above, and $\xi\in\eulc(M,\pp)$. Let

732: $\gbasis(\xi)=\{\tilde\sigma\}$ be the family of liftings of the cells lying in

733: $M\setminus(W\cup V)$ determined by a connected spider as explained above. Note

734: that if $\gbasis'=\{\tilde\sigma'\}$ is any other family of liftings we have 

735: $\tilde\sigma'=a(\sigma)\cdot\tilde\sigma$ for some $a\in\pi_1(M)$, and we can

736: define $$h(\gbasis',\gbasis(\xi))=\sum_{\sigma\subset M\setminus (W\cup

737: V)}\index(\sigma)\cdot \overline{a}(\sigma)\in H_1(M;\mz).$$

738: 

739: 

740: 

741: 

742: \begin{prop}\label{no:need:to:lift}

743: Assume there exists a partition $\cc_1\sqcup\dots\sqcup\cc_k$ of the set of cells 

744: lying in $M\setminus(W\cup V)$, and let $\xi\in\eulc(M,\pp)$ have a representative of the form

745: $$z=\sum_{j=1}^k\left(\sum_{\sigma\in\cc_j\setminus\{\sigma_j\}}

746: \index(\sigma)\cdot \gamma^{(j)}_\sigma\right)$$

747: where $\sigma_j\in\cc_j$ and $\gamma^{(j)}_\sigma:([0,1],0,1)\to(M,p_{\sigma_j},p_\sigma)$.

748: Choose any lifting $\tilde p_{\sigma_j}$ of $p_{\sigma_j}$, lift 

749: $\gamma^{(j)}_\sigma$ to $\tilde\gamma^{(j)}_\sigma$ starting from $\tilde p_{\sigma_j}$,

750: let $\tilde\sigma'$ be the lifting of $\sigma$ containing $\tilde\gamma^{(j)}_\sigma(1)$,

751: and let $\gbasis'$ be the family of all these liftings.

752: Then $h(\gbasis',\gbasis(\xi))=0\in H_1(M;\mz)$. In particular $\gbasis'$ can be used to compute 

753: $\tau^\varphi(M,\pp,\xi,\hbasis)$.

754: \end{prop}

755: 

756: \dim{no:need:to:lift}

757: Note first that the coefficient of $p_{\sigma_j}$ in $\partial z$ is exactly

758: $$-\sum_{\sigma\in\cc_j\setminus\{\sigma_j\}}\index(\sigma).$$

759: On the other hand this coefficient must be equal to $\index(\sigma_j)$. 

760: Summing up we deduce that $\sum_{\sigma\in\cc_j}\index(\sigma)=0$.

761: 

762: Now choose $x_0\in M$ and $\delta^{(j)}:([0,1],0,1)\to(M,x_0,p_{\sigma_j})$. For 

763: $\sigma\in\cc_j$ define

764: $$\beta_\sigma=\cases{

765: \delta^{(j)} & if $\sigma=\sigma_j$ \cr

766: \delta^{(j)}\cdot\gamma^{(j)}_\sigma & otherwise,}$$

767: so that $\beta_\sigma:([0,1],0,1)\to(M,x_0,p_\sigma)$, whence 

768: $w=\sum_{\sigma\subset M\setminus(W\cup V)}\beta_\sigma$ is an Euler chain.

769: Moreover:

770: $$w-z=\sum_{j=1}^k\left(\sum_{\sigma\in\cc_j}\index(\sigma)\right)

771: \cdot\delta^{(j)}=0\in H_1(M;\mz),$$

772: so $[w]=\xi$. Now choose $\tilde x_0$ over $x_0$, lift the $\delta^{(j)}$ and 

773: $\beta_\sigma$ starting from $\tilde x_0$, and let

774: $a^{(j)}\in\pi_1(M)$ be such that $\tilde p_{\sigma_j}=a^{(j)}\cdot\tilde\delta^{(j)}(1)$.

775: Then

776: $$h(\gbasis',\gbasis(\xi))=\sum_{j=1}^k\left(\sum_{\sigma\in\cc_j}\index(\sigma)\right)

777: \cdot\overline{a}^{(j)}=0\in H_1(M;\mz),$$

778: and the proof is complete.\finedim{no:need:to:lift}

779: 

780: The next result follows directly from the definition, but it is worth stating

781:  because it shows how torsions may be used to distinguish triples $(M,\pp,\xi)$ from each other.

782: 

783: \begin{prop}\label{homeo:action}

784: Let $f: M\to M'$ be a homeomorphism, consider $\xi\in\eul(M,\pp)$, 

785: $\varphi:\pi_1(M)\to\Lambda_*$ and a $\Lambda$-basis $\hbasis$ of 

786: $H^\varphi_*(M,\overline{W})$. Then

787: $$\tau^{\varphi\circ f_*^{-1}}(M',f_*(\pp),f_*(\xi),f_*(\hbasis))=

788: \tau^\varphi(M,\pp,\xi,\hbasis).$$

789: \end{prop} 

790: 

791: 

792: 

793: \section{Spines and computation of torsion}\label{spines:section}

794: In this section we show how to geometrically invert the reconstruction map $\Psi$, and

795: how to compute torsions starting from a combinatorial

796: encoding of vector fields. We first review

797: the theory developed in~\cite{lnm}. See the beginning of Section~\ref{Eul:def:section}

798: for our conventions on manifolds, maps, and fields. In addition to the terminology

799: introduced there, we will need the notion of {\em traversing} field on a manifold

800: $M$, defined as a field whose orbits eventually intersect $\partial M$ 

801: transversely in both directions (in other words, orbits are compact intervals).

802: 

803: \paragraph{Branched spines}

804: A {\em simple} polyhedron $P$ is a  finite connected 2-dimensional polyhedron

805: with singularity of stable nature (triple lines and points where six non-singular

806: components meet). Such a $P$ is called {\it standard} if all the components of

807: the natural stratification given by singularity are open cells. Depending on

808: dimension, we will call the components {\it vertices, edges} and {\it regions}.

809: 

810: A {\em standard spine} of a $3$-manifold $M$ with $\partial M\neq\emptyset$

811: is a standard polyhedron $P$ embedded in $\interior(M)$ so that $M$ collapses onto $P$.

812: Standard spines of oriented $3$-manifolds are characterized among standard polyhedra

813: by the property of carrying an {\em orientation}, defined 

814: (see Definition~2.1.1 in~\cite{lnm}) as a ``screw-orientation''

815: along the edges (as in the left-hand-side of Fig.~\ref{screw:branch}),

816: with an obvious compatibility at vertices

817: (as in the centre of Fig.~\ref{screw:branch}).

818: \begin{figure}

819: \centerline{\psfig{file=screwbra.eps,width=12cm}}

820: \caption{\label{screw:branch} Convention on screw-orientations, compatibility 

821: at vertices, and geometric interpretation of branching.}

822: \end{figure}

823: It is the starting point of the theory of standard spines

824: that every oriented $3$-manifold $M$ with $\partial M\neq\emptyset$ has an oriented

825: standard spine, and can be reconstructed (uniquely up to homeomorphism) 

826: from any of its oriented standard spines. See~\cite{casler} for the non-oriented version 

827: of this result and~\cite{manuscripta} or Proposition~2.1.2 in~\cite{lnm} for the (slight) oriented refinement.

828: 

829: A {\it branching} on a standard polyhedron $P$ is an

830: orientation for each region of $P$, such that no edge is induced the same

831: orientation three times. See the right-hand side of Fig.~\ref{screw:branch}

832: and Definition 3.1.1 in~\cite{lnm} for the geometric meaning of this notion.

833: An oriented standard spine $P$ endowed with a branching is shortly named 

834: {\em branched spine}. We will never use specific notations for the extra structures: 

835: they will be considered to be part of $P$.

836: The following result, proved as Theorem~4.1.9 in~\cite{lnm}, is the starting point of our constructions.

837: 

838: \begin{prop}\label{from:spine:to:field}

839: To every branched spine $P$ there corresponds a manifold $M(P)$ 

840: with non-empty boundary and a concave traversing field $v(P)$ on $M(P)$.

841: The pair $(M(P),v(P))$ is well-defined up to diffeomorphism.

842: Moreover an embedding $i:P\to\interior(M(P))$ is defined, 

843: and has the property that $v(P)$ is positively transversal to $i(P)$.

844: \end{prop}

845: 

846: 

847: The topological construction which underlies this proposition is actually quite

848: simple, and it is illustrated in Fig.~\ref{constr:M}. Concerning the last

849: \begin{figure}

850: \centerline{\psfig{file=flowspin.eps,width=4cm}}

851: \caption{\label{constr:M} Manifold and field associated to a branched spine.}

852: \end{figure}

853: assertion of the proposition, note that the branching allows to define an

854: oriented tangent plane at each point of $P$.

855: 

856: \paragraph{Combinatorial encoding of combings}

857: Let $P$ be a branched spine, and define $v(P)$ on $M(P)$ as just explained.

858: Assume that in $\partial M(P)$ there is only one component which is homeomorphic to $S^2$

859: and is split by the tangency line of $v(P)$ to $\partial M(P)$ into two

860: discs. (Such a component will be denoted by $\stwotriv$.) Now, notice that $\stwotriv$ is

861: also the boundary of the closed $3$-ball with constant vertical field, 

862: denoted by $B^3_{\rm triv}$. This shows that we can cap off

863: $\stwotriv$ by attaching a copy of $B^3_{\rm triv}$, 

864: getting a compact manifold $\hatM(P)$ and a field $\hatv(P)$ on $\hatM(P)$. If we denote by $\hatpp(P)$ the boundary pattern of $\hatv(P)$ on

865: $\hatM(P)$, we easily see that the pair $(\hatM(P),\hatv(P))$ is only well-defined

866: up to homeomorphism of $\hatM(P)$ and homotopy of $\hatv(P)$ through fields compatible

867: with $\hatpp(P)$. Note also that $\hatpp(P)$ is automatically concave.

868: 

869: If $\pp$ is a boundary pattern on $M$, we define ${\rm Comb}(M,\pp)$ as

870: the set of fields compatible with $\pp$ under homotopy through fields also

871: compatible with $\pp$. An element of ${\rm Comb}(M,\pp)$

872: is called a {\em combing} on $(M,\pp)$. Note that we have a projection

873: ${\rm Comb}(M,\pp)\to\eul(M,\pp)$.

874: 

875: The above construction shows that

876: a branched spine $P$ with only one $\stwotriv$ on $\partial M(P)$ defines

877: an element $\Phi(P)$ of ${\rm Comb}(\hatM(P),\hatpp(P))$.

878: In~\cite{second:paper} we will establish the following:

879: 

880: \begin{teo}\label{bounded:surg} 

881: If $M$ is any compact oriented 

882: $3$-manifold and $\pp$ is a concave boundary pattern

883: on $M$ not containing $\stwotriv$ components, then $\Phi$ maps surjectively

884: $\{P:\ \hatM(P)\cong M,\ \hatpp(P)\cong\pp\}$ onto ${\rm Comb}(M,\pp)$.

885: \end{teo}

886: 

887: This theorem generalizes

888: the main achievement of~\cite{lnm} (Theorems~1.4.1 and~5.2.1),

889: where it is proved in the special case of closed $M$. The complete statement

890: includes also the description of a finite set of local moves on branched spines

891: generating the equivalence relation induced by $\Phi$. We will not need the

892: moves in this paper. The following

893: geometric interpretation the theorem may however be of some interest.

894: 

895: \begin{rem}\label{trivial:pieces}

896: {\em In general, the dynamics of a field, even a concave one, can be very complicated,

897: whereas the dynamics of a traversing field (in particular, $B^3_{\rm triv}$) is simple.

898: Theorem~\ref{bounded:surg} means that for any 

899: (complicated) concave field

900: there exists a sphere $S^2$ which splits the field into two (simple) pieces:

901: a standard $B^3_{\rm triv}$ and a concave traversing field.}

902: \end{rem}

903: 

904: 

905: We can give here an easy special proof of Theorem~\ref{bounded:surg}

906: for the case we are most interested in, namely link exteriors.

907: Note that our argument relies on the results of~\cite{lnm}.

908: 

909: \vspace{1pt}

910: \noindent{\em Proof of~\ref{bounded:surg} 

911: for link exteriors}. We have to show that if $M$ is closed, $v$ is a field on $M$

912: and $L$ is transversal to $v$, then the exterior $E(L)$ of $L$ with the restricted field

913: is represented by some branched spine in the sense explained above.

914: 

915: The construction explained in Section~5.1  of~\cite{lnm} shows that 

916: there exists a branched standard spine $P$ such that $v$ is positively 

917: transversal to $P$, and the complement of $P$, with the 

918: restriction of $v$, is isomorphic to

919: the open 3-ball with the constant vertical field. 

920: The last condition easily implies that $L$ can be 

921: isotoped through links transversal to $v$ to a link lying in an arbitrarily small 

922: neighbourhood of $P$, with the further property that its natural projection

923: on $P$ is $\cont^1$, possibly with crossings. 

924: 

925: Once $L$ has been isotoped to a $\cont^1$ link on $P$,

926: a branched spine of $(E(L),v\ristr{E(L)})$ is obtained

927: by digging a tunnel in $P$ along the projection of $L$, as shown in Fig.~\ref{dig:tunnel}.

928: \begin{figure}

929: \centerline{\psfig{file=digtunne.eps,width=10cm}}

930: \caption{\label{dig:tunnel} How to dig a tunnel in a spine.}

931: \end{figure}

932: A crossing in the projection will of course give rise to 4 vertices in the spine.

933: Note that the spine which results from the digging may occasionally be 

934: non-standard, but it is standard as soon as the projection is complicated enough 

935: ({\em e.g.}~if on each component there are both a crossing and an intersection with $S(P)$).

936: {{\hfill\hbox{\enspace\fbox{\ref{bounded:surg}$E(L)$}}}\vspace{5pt}}

937: 

938: 

939: \paragraph{Spines and ideal triangulations} 

940: We remind the reader that an {\it

941: ideal triangulation} of a manifold $M$ with non-empty boundary is  a partition

942: $\calt$ of $\interior(M)$ into open cells of dimensions 1, 2 and 3, induced by a

943: triangulation $\calt'$ of the space $Q(M)$, where:

944: \begin{enumerate}

945: \item $Q(M)$ is obtained from $M$ by collapsing each component of

946: $\partial M$ to a point;

947: \item $\calt'$ is a triangulation only in a loose sense, namely

948: self-adjacencies and multiple adjacencies of tetrahedra are allowed;

949: \item The vertices of $\calt'$ are precisely the points of $Q(M)$

950: which correspond to components of $\partial M$.

951: \end{enumerate}

952: 

953: \noindent It turns out (see for instance~\cite{mafo},~\cite{tesi},~\cite{matv:new}) that there exists a natural

954: bijection between standard spines and ideal triangulations

955: of a 3-manifold. Given an ideal triangulation, the

956: corresponding standard spine is just the 2-skeleton of the dual

957: cellularization, as illustrated in Figure~\ref{duality}.

958: \begin{figure}

959: \centerline{\psfig{file=verttetr.eps,width=3cm}}

960: \caption{\label{duality} Duality between standard spines and ideal triangulations.}

961: \end{figure}

962: The inverse of this correspondence will be denoted by $P\mapsto\calt(P)$. 

963: 

964: Now let $P$ be a branched spine. First of all, we can realize $\calt(P)$ in such

965: a way that its edges are orbits of the restriction of $v(P)$ to

966: $\interior(M(P))$, and the 2-faces are unions of such orbits.  Being orbits,

967: the edges of $\calt(P)$ have a natural orientation, and the branching condition,

968: as remarked in~\cite{gr:1},

969: is equivalent to the fact that on each tetrahedron of $\calt(P)$ 

970: exactly one of the vertices is a sink and one is a source. 

971: 

972: \begin{rem}\label{all:oriented}

973: {\em It turns out that if $P$ is a branched spine, 

974: not only the edges, but also the faces

975: and the tetrahedra of $\calt(P)$ have natural orientations. For tetrahedra, we

976: just restrict the orientation of $M(P)$. For faces, we first note that the edges

977: of $P$ have a natural orientation (the prevailing orientation induced by the

978: incident regions). Now, we orient a face of $\calt(P)$ so that the algebraic

979: intersection in $M(P)$ with the dual edge is positive. }

980: \end{rem}

981: 

982: \paragraph{Euler chain defined by a branched spine}

983: We fix in this paragraph a standard spine $P$ and consider its manifold $M=M(P)$.

984: We start by noting that the ideal triangulation $\calt=\calt(P)$ defined by $P$ 

985: can be interpreted as a realization of $\interior(M)$ by face-pairings on a finite

986: set of tetrahedra with vertices removed. If, instead of 

987: removing vertices, we remove open conic neighbourhoods of the vertices, thus getting {\em truncated} tetrahedra, after the 

988: face-pairings we obtain $M$ itself. This shows that $P$ determines a cellularization

989: $\caltbar=\caltbar(P)$ 

990: of $M$ with vertices only on $\partial M$ and 2-faces which are either triangles

991: contained in $\partial M$ or hexagons contained in $\interior(M)$, with edges contained alternatingly in $\partial M$ and in $\interior(M)$.

992: 

993: 

994: Now assume that $P$ is branched and 

995: that $\partial M$ contains only one $\stwotriv$ component, so 

996: $\hatM=\hatM(P)$ is defined. Note that $\hatM$ can be thought of as the space obtained from

997: $M$ by contracting $\stwotriv$ to a point, so a projection $\pi:M\to\hatM$ is defined,

998: and $\pi(\caltbar)$ is a cellularization of $\hatM$. Next, we modify 

999: $\pi(\caltbar)$ by subdividing the triangles on

1000: $\partial\hatM$ as shown in Fig.~\ref{trunc:tetra}.

1001: \begin{figure}

1002: \centerline{\psfig{file=truncated.eps,width=8.5cm}}

1003: \caption{\label{trunc:tetra}Truncated tetrahedra and subdivision of the triangles on the

1004: boundary}

1005: \end{figure}

1006: The result is a cellularization $\hattt=\hattt(P)$ of $\hatM$. 

1007: Note that $\hattt$ on $\partial\hatM$ 

1008: consists of ``kites'', with long edges coming from tetrahedra and short edges coming from 

1009: subdivision. Note also that $\hattt$ has exactly one vertex $x_0$ in $\interior(\hatM)$, and

1010: that the cells contained in $\interior(\hatM)$, except $x_0$, are the duals to the 

1011: cells of the natural cellularization $\uu=\uu(P)$ of $P$. For $u\in\uu$ we denote by $\hat{u}$ 

1012: its dual and by $p_u=p_{\hat{u}}$ the point where $u$ and $\hat{u}$ intersect, called the {\em centre} of both.

1013: 

1014: We will now use the field $\hatv=\hatv(P)$ to construct a combinatorial Euler chain

1015: on $\hatM$ with respect to

1016: $\hattt$. It is actually convenient to consider, instead of $\hatv$, the field 

1017: $\vbar=\pi(v)$, which 

1018: coincides with $\hatv$ except near $x_0$, where it has a 

1019: (removable) singularity. For $u\in\uu$ we denote by $\beta_u$ the arc obtained by integrating $\vbar(P)$ in the positive direction, starting from $p_u$, until the 

1020: boundary or the singularity is reached. We define:

1021: $$s(P)=\sum_{u\in\uu}\index(u)\cdot \beta_u.$$

1022: 

1023: Let us consider now the pattern $\hatpp=\hatpp(P)=(W,B,\emptyset,C)$ defined by $P$.

1024: If $p$ is a vertex of $\pi(\caltbar)$ contained in $B$, we define its star ${\rm St}(p)$ as

1025: the sum of the straight segments going from $p$ to the centres of all the kites containing $p$,

1026: minus the sum of the straight

1027: segments going from $p$ to the centres of all the long edges 

1028: containing $p$. If $\sigma$ is an edge of $\pi(\caltbar)$ contained in $B$ we define its

1029: bi-arrow ${\rm Ba}(\sigma)$ as the sum of the two straight segments going from the centre

1030: $p_\sigma$ of $\sigma$ to the centres of the two short kite-edges containing $p_\sigma$.

1031: A star and a bi-arrow are shown in Fig.~\ref{starfig}.

1032: \begin{figure}

1033: \centerline{\psfig{file=starfig.eps,width=7cm}}

1034: \caption{\label{starfig}The star ${\rm St}(p)$ centred at a vertex $p$ contained in $B$

1035: and the bi-arrow ${\rm Ba}(\sigma)$ based at the midpoint of an edge $\sigma$ contained in $B$}

1036: \end{figure}

1037: We define:

1038: $$s'(P)=s(P)+\sum_{p\in B\cap\caltbar(P)^{(0)}}{\rm St}(p)+

1039: \sum_{\sigma\in\hattt(P)^{(1)},\sigma\subset B}{\rm Ba}(\sigma).$$

1040: 

1041: \begin{lem}\label{s:prime:Euler}

1042: $s'(P)$ defines an element of $\eulc(\hatM,\theta(P))$.

1043: \end{lem}

1044: 

1045: \dim{s:prime:Euler}

1046: Recall that $\theta(\hatpp)=(W,B,C,\emptyset)$, {\em i.e.}~the concave line $C$ is turned into

1047: a convex one. So by definition we have to show that $\partial s'(P)$ contains, with the right 

1048: sign, the centres of all cells of $\hattt$ except those of $W\cup C$. 

1049: 

1050: It will be convenient to analyze first the natural lifting of $s(P)$ to $M$, denoted by

1051: $\tilde s(P)=\sum_{u\in\uu}\index(u)\cdot \tilde\beta_u$ with obvious meaning of symbols. So 

1052: \begin{equation}

1053: \partial\tilde s(P)=\sum_{u\in\uu}-\index(u)\cdot \tilde\beta_u(0)+

1054: \sum_{u\in\uu}\index(u)\cdot \tilde\beta_u(1).

1055: \label{s:tilde:boundary}

1056: \end{equation}

1057: 

1058: Since the cellularization $\caltbar$ of $M$ is dual to $\uu$, the first half of 

1059: (\ref{s:tilde:boundary}) gives the centres of the cells contained in $\interior(M)$, with right 

1060: sign. One easily sees that the second half gives exactly the centres of the cells 

1061: (of $\caltbar$) contained in $B$, also with right sign. 

1062: 

1063: When we project to $\hatM$ and consider $\partial s(P)$, the first half of 

1064: (\ref{s:tilde:boundary}) again provides (with right sign)

1065: the centres of the all cells contained in $\interior(\hatM)$, except the special

1066: vertex $x_0$ obtained by collapsing $\stwotriv$. We can further split

1067: the points of the second half of (\ref{s:tilde:boundary}) into those which lie on $\stwotriv$

1068: and those which do not. The points of the first type project to $x_0$, and the resulting coefficient of $x_0$ is $\chi(B\cap\stwotriv)$, but $B\cap\stwotriv$ is an open 2-disc,

1069: so the coefficient is 1. (We are here using the very special property of dimension 2

1070: that $\chi$ can be computed using a finite cellularization of an open manifold, because the

1071: boundary of the closure has $\chi=0$.) The points of the second type faithfully project to 

1072: $\hatM$, giving the centres of the simplices contained in $B$ of the triangulation

1073: $\pi(\caltbar)\ristr{\partial\hatM}$.

1074: However $\hattt$ on $\partial\hatM$ is a subdivision of $\pi(\caltbar)$,

1075: and this is the reason why we have added the stars and the bi-arrows to $s(P)$ getting $s'(P)$.

1076: The following computation of the coefficients in $\partial s'(P)$ of the centres of the 

1077: cells of $\hattt$ contained in $B$ concludes the proof.

1078: \begin{enumerate}

1079: \addtocounter{enumi}{-1}

1080: \item Cells of dimension 0 are listed as follows:

1081: \begin{enumerate}

1082: \item Centres of triangles of $\pi(\caltbar)$, 

1083: which receive coefficient $+1$ from $\partial s(P)$;

1084: \item Midpoints of edges of $\pi(\caltbar)$, 

1085: which receive coefficient $-1$ from $\partial s(P)$

1086: and $+2$ from the bi-arrows they determine;

1087: \item Vertices of $\pi(\caltbar)$ receive $+1$ from $\partial s(P)$ and

1088: (algebraically) $0$  from the star they determine;

1089: \end{enumerate}

1090: \item Cells of dimension 1 are:

1091: \begin{enumerate}

1092: \item Short edges of kites, whose midpoints receive $-1$ from the bi-arrows;

1093: \item Long edges of kites, whose midpoints receive $-1$ from the stars;

1094: \end{enumerate}

1095: \item Cells of dimension 2 are kites, and their centres receive $+1$ from the stars.

1096: \end{enumerate}

1097: \finedim{s:prime:Euler}

1098: 

1099: \noindent Now we denote by $\gamma_j:(0,1)\to C$, for $j=1,\dots,n$, 

1100: orientation-preserving parameterizations of the 1-cells of $\hattt$ contained in $C$, and we extend the $\gamma_j$ to $[0,1]$, without changing notation. We define

1101: $$s''(P)=s'(P)+\sum_{j=1}^n\gamma_j\ristr{[1/2,1]}.$$ 

1102: 

1103: \begin{lem}\label{s:second:Euler}

1104: $s''(P)$ defines an element of $\eulc(\hatM,\hatpp)$, and 

1105: $$[s'(P)]=\Thetac([s''(P)])\in \eulc(\hatM,\theta(\pp)).$$

1106: \end{lem}

1107: 

1108: \dim{s:second:Euler} At the level of representatives, the second assertion is obvious,

1109: and it implies the first assertion.

1110: \finedim{s:second:Euler} 

1111: 

1112: \noindent We defer to Section~\ref{proofs} the proof of the next result, which shows that

1113: the map $P\mapsto[s''(P)]\in\eulc(\hatM,\hatpp)$ allows,

1114: using branched spines, to explicitly find the inverse

1115: of the reconstruction map $\Psi$ of 

1116: Theorem~\ref{reconstruction:statement}. This result was informally announced as

1117: Theorem~\ref{informal:algorithmic} in the Introduction.

1118: 

1119: \begin{teo}\label{spider:structure:teo}

1120: $\Psi([s''(P)])=[\hatv(P)]\in\euls(\hatM,\hatpp)$.

1121: \end{teo}

1122: 

1123: Recall now that we have defined torsions directly only for convex patterns,

1124: and we have extended the definition to concave patterns via the map $\Theta$.

1125: As a consequence of Lemma~\ref{s:second:Euler} and Theorem~\ref{spider:structure:teo},

1126: and by direct inspection of $s'(P)$, we have the following result which summarizes

1127: our investigations on the relation between spines, Euler structures, and torsion:

1128: 

1129: \begin{teo}\label{spines:compute:teo}

1130: If $P$ is a branched spine which represents a manifold $\hatM$ with 

1131: concave boundary pattern $\hatpp=(W,B,\emptyset,C)$

1132: in the sense of Theorem~\ref{bounded:surg}, then for any 

1133: representation $\varphi:\pi_1(M)\to\Lambda_*$ and any $\Lambda$-basis

1134: $\hbasis$ of $H^\varphi_*(\hatM,W\cup C)$, the torsion

1135: $\tau^\varphi(\hatM,\hatpp,[\hatv(M)],\hbasis)$ can be computed using

1136: (in the sense of Proposition~\ref{no:need:to:lift}) the lifting

1137: to the universal cover of $\hatM$ of the chain $s'(P)$ defined above. In particular, 

1138: $s'(P)$ can be used directly, without replacing it by a connected spider.

1139: \end{teo}

1140: 

1141: \paragraph{Computational hints} 

1142: To actually compute torsion starting from a branched spine $P$, besides describing

1143: the universal (or maximal Abelian) cover of $\hatM=\hatM(P)$ and determining the preferred

1144: liftings of the cells in $\hatM\setminus(W\cup C)$, one needs to compute

1145: the boundary operators in the twisted chain complex 

1146: $\Cphi_*(M,W\cup C)$. These operators are of course twisted liftings of the

1147: corresponding operators in the cellular chain complex of $(\hatM,W\cup C)$,

1148: with respect to $\hattt$. We briefly describe here the form of the latter operators.

1149: Recall first that $\hattt$ consists of a special vertex $x_0$, the kites

1150: (with their vertices and edges) on $\hatM$, and the duals of the

1151: cells of $P$. On $\partial\hatM$ the situation is easily described, so we consider the

1152: internal cells. 

1153: \begin{enumerate}

1154: \item If $R$ is a region of $P$, the ends of its dual edge 

1155: $\hat{R}$ are either $x_0$ or vertices of $\partial\hatM$ 

1156: contained only in long edges of kites. 

1157: \item If $e$ is an edge of $P$ then

1158: $\partial\hat{e}$ is given by

1159: $\hat{R}_1+\hat{R}_2-\hat{R}_0$ plus 3 long edges of kites, where $R_0,R_1,R_2$ are the

1160: regions incident to $e$, numbered so that $R_1$ and $R_2$ induce on $e$ the same

1161: orientation. Here $R_0,R_1,R_2$ need not be different from each other, so

1162: the formula may actually have some cancellation. The 3 long edges

1163: of kites must be given an appropriate sign, and some of them 

1164: may actually be collapsed to the point $x_0$. Note that we have only 3 kite-edges, out

1165: of the 6 which geometrically appear on $\partial\hat{e}$, because the other 3 are white.

1166: \item If $v$ is a vertex of $P$ then

1167: $\partial\hat{v}$ is given by 

1168: $\hat{e}_1+\hat{e}_2-\hat{e}_3-\hat{e}_4$ plus 6 kites, where $e_1,e_2$ are

1169: the edges which (with respect to the natural orientation) are leaving $v$, and

1170: $e_3,e_4$ are those which are reaching it. Again, there could be

1171: repetitions in the $e_i$'s. The kites all have coefficient $+1$, and again some of them may

1172: actually be collapsed to $x_0$. As above, we have only

1173: 6 kites because the other 6 are white.

1174: \end{enumerate}

1175: 

1176: \begin{rem}\label{do:not:cut}{\em To define the cellularization $\hattt(P)$ associated to 

1177: a spine we have decided to subdivide all the triangles on $\partial\hatM$

1178: into 3 kites, but when doing actual computations this is not necessary and 

1179: impractical. The only triangles 

1180: which we really need to subdivide are those intersected by $C$, because we need the 

1181: cellularization to be suited to the pattern. Let us consider the 4

1182: triangles corresponding to the ends of a certain tetrahedron. If in each of them

1183: we count the number of black kites and the number of white kites, we get respectively

1184: $(3,0)$, $(2,1)$, $(1,2)$, $(0,3)$. So, the first and last triangles do not have to

1185: be subdivided, and the other two can be subdivided using a segment only. Summing up,

1186: for each vertex of $P$ we only need to add two segments on the boundary. Before

1187: projecting $M(P)$ to $\hatM(P)$ one sees that the number of cells, with respect to

1188: $\caltbar(P)$, is increased in all dimensions 0, 1 and 2 by twice the number of vertices of $P$.

1189: When projecting to $\hatM(P)$ the cells lying in $\stwotriv$ get collapsed to points.}

1190: \end{rem}

1191: 

1192: 

1193: 

1194: 

1195: \section{Pseudo-Legendrian knots}\label{knots:section}

1196: 

1197: We fix in this section a compact oriented manifold $M$ and a boundary pattern $\pp$ on $M$.

1198: The boundary of $M$ may be empty or not. Recall that if $v$ is a vector field on $M$ and 

1199: $K$ is a knot in $\interior(M)$, we have defined $K$ to be pseudo-Legendrian in $(M,v)$ 

1200: if $v$ is transversal to $K$. We will also call $(v,K)$ a pseudo-Legendrian pair.

1201: Having fixed $\pp$, we will only consider fields $v$ 

1202: compatible with $\pp$. Some of the results we will establish hold

1203: also for links, but we will stick to knots for the sake of 

1204: simplicity. First, we need to spell out the equivalence relation

1205: which we consider.

1206: 

1207: Let $v_0,v_1$ be compatible with $\pp$ and let $K_0,K_1$ be pseudo-Legendrian in

1208: $(M,v_0)$ and $(M,v_1)$ respectively. We define $(v_0,K_0)$ to be {\em weakly equivalent}

1209: to $(v_1,K_1)$ if there exist a homotopy $(v_t)_{t\in[0,1]}$ through fields compatible

1210: with $\pp$ and an isotopy $(K_t)_{t\in[0,1]}$ such that $K_t$ is transversal

1211: to $v_t$ for all $t$. If $v_0=v_1$ then $K_0$ and $K_1$ are

1212: called {\em strongly equivalent} if the homotopy $(v_t)$ can be chosen to be constant.

1213: 

1214: \begin{rem}\label{weak:strong:rem}

1215: {\em

1216: \begin{enumerate}

1217: \item Of course strong equivalence implies weak equivalence. Weak equivalence is the natural

1218: relation to consider on pseudo-Legendrian pairs $(v,K)$, while strong equivalence

1219: is natural for pseudo-Legendrian knots in a fixed $(M,v)$. See~\cite{second:paper}

1220: for a further discussion on this point.

1221: \item The term `pseudo-Legendrian isotopy', used in the introduction and 

1222: in~\cite{second:paper}, corresponds to `weak equivalence.' For the sake of brevity,

1223: and to emphasize the difference with strong equivalence, we will only use 

1224: the term `weak equivalence' in the rest of this paper.

1225: \end{enumerate}}

1226: \end{rem}

1227: 

1228: Before proceeding note that if $K$ is pseudo-Legendrian in $(M,v)$ then

1229: $v$ turns $K$ into a framed knot, which we will denote by $K^{(v)}$.

1230: The framed-isotopy class of $K^{(v)}$ is of course invariant under weak

1231: equivalence.

1232: 

1233: \paragraph{Euler structures on knot exteriors}

1234: For a knot $K$ in $M$ we consider a (closed) tubular neighbourhood $U(K)$ of $K$

1235: in $M$ and we define $E(K)$ as the closure of the complement of $U(K)$.

1236: If $F$ is a framing on $K$ we extend the boundary pattern $\pp$ previously fixed

1237: on $M$ to a boundary pattern $\pp(K^F)$ on $E(K)$, by splitting $\partial U(K)$

1238: into a white and a black longitudinal annuli, the longitude being the one defined by

1239: the framing $F$. As a direct application of Proposition~\ref{p:h:formula} one sees

1240: that $\eul(E(K),\pp(K^F))$ is non-empty (assuming $\eul(M,\pp)$ to be non-empty).

1241: 

1242: A convenient way to think of $\pp(K^F)$ is as follows. The framing $F$ 

1243: determines a transversal vector field along $K$. If we extend this field near

1244: $K$ and choose  $U(K)$ small enough then the pattern we see on $\partial U(K)$

1245: is exactly as required. With this picture in mind, it is clear that if $K$ is

1246: pseudo-Legendrian in $(M,v)$, where $v$ is compatible with $\pp$, then the

1247: restriction of $v$ to $E(K)$ defines an element

1248: $$\xi(v,K)\in\eul(E(K),\pp(K^{(v)})$$ 

1249: (this structure was already considered in the partial proof of Theorem~\ref{bounded:surg}).

1250: So the theory of torsion applies. We will discuss in this section torsion both

1251: as an {\em absolute} and as a {\em relative} invariant, splitting the section into

1252: two subsections.

1253: 

1254: 

1255: \subsection{Absolute torsion and the Alexander invariant}

1256: We will show in this section that torsion as an absolute invariant of

1257: pseudo-Legendrian knots lifts the classical Alexander invariant, but the

1258: relation between the two objects is more complicated than in Turaev's

1259: situation (\cite{turaev:Reidemeister} and~\cite{turaev:Euler}),

1260: because here two different algebraic complexes are involved at the same time.

1261: 

1262: \paragraph{Turaev's lifting of Milnor torsion} 

1263: Let us first recall again in what sense Turaev's torsion lifts the classical one. Let

1264: $M$ be a manifold which is closed or bounded by tori, and take

1265: a representation $\varphi:H_1(M;\mz)\to\Lambda_*$, where $\Lambda$

1266: is as usual. The classical theory~\cite{milnor} allows to

1267: define an invariant

1268: $$\tau^\varphi(M)\in K_1(\Lambda)

1269: \Big/ (\pm\varphi(H_1(M;\mz))),$$

1270: usually stipulated to be $1$ if

1271: the $\varphi$-twisted homology of $M$ does not vanish, {\em i.e.},

1272: using the above notation, if the complex 

1273: $C^\varphi_*(M)=\Lambda\otimes_\varphi

1274: C_*^{\rm cell}(\tilde M;\mz)$ is not acyclic, where

1275: $q:\tilde M\to M$ is the maximal Abelian cover.

1276: When $\xi$ is an Euler structure on $M$ with monochromatic boundary components,

1277: Turaev~\cite{turaev:Euler} 

1278: shows that his torsion $\tau^\varphi(M,\xi)\in\Kbar_1(\Lambda)$ is a lifting

1279: of $\tau^\varphi(M)$ with respect to the obvious projection of

1280: $K_1(\Lambda)/\pm\varphi(H_1(M;\mz)$ onto $\Kbar_1(\Lambda)$.

1281: 

1282: In the special case where 

1283: $\Lambda$ is the field of fractions obtained from the 

1284: group ring of $H_1(M;\mz)$ modulo torsion,

1285: and $\varphi:H_1(M;\mz)\to\Lambda$ is the natural projection, the invariant

1286: $\tau^\varphi(M)$ is called Milnor torsion, and its

1287: sign-refinement provided by Turaev 

1288: in~\cite{turaev:Reidemeister} has been shown to be equivalent

1289: to the classical Alexander invariant. So Turaev's torsion for Euler structures contains

1290: a lifting of the Alexander invariant. We will discuss in the rest of this

1291: subsection the extent to what the same holds when the Euler structure 

1292: arises from a pseudo-Legendrian knot. What we will say applies to any allowed

1293: representation $\varphi:H_1(M;\mz)\to\Lambda$, but we keep in mind that the relation

1294: with the Alexander invariant emerges for a special choice of $\varphi$ and $\Lambda$.

1295: Since we will also drop the condition that the involved complexes be acyclic, we note

1296: that torsion is only defined when the resulting homology is free. This is not true in

1297: general, but it is for instance when $\Lambda$ is a field.

1298: 

1299: 

1300: 

1301: 

1302: \paragraph{Torsion of a knot complement} Let us restrict to the case of a closed

1303: manifold $M$, and let us consider a pseudo-Legendrian pair $(v,K)$ in $M$

1304: and a representation $\varphi:H_1(E(K);\mz)\to\Lambda$ as

1305: usual. We would like to interpret the torsion of the Euler structure $\xi(v,K)$ 

1306: on $E(K)$ with respect to $\varphi$ 

1307: as a lifting of $\tau^\varphi(E(K))$, but a difficulty immediately emerges, because

1308: the algebraic complexes used to compute these torsions do not coincide.

1309: 

1310: To be more specific, let us first spell out how the torsion of $\xi(v,K)$ is defined.

1311: Let $\pp(K^{(v)})=(B,W,\emptyset,C)$ be the boundary pattern

1312: defined on $E(K)$. Then we define

1313: $\tau^\varphi(M,v,K,\hbasis)$ as $\tau^\varphi(M,\pp(K^{(v)}),\Theta(\xi(v,K)),\hbasis)$.

1314: More specifically, 

1315: $\tau^\varphi(M,v,K,\hbasis)$ is the torsion of the complex

1316: $C^\varphi_*(E(K),\overline{W})$,

1317: where $W$ is the (open) white annulus on $\partial E(K)$,

1318: as above the maximal Abelian cover of $E(K)$ is used to define the complex, 

1319: the preferred cell lifting is obtained using

1320: an Euler chain for the convexified structure $\Theta(\xi(v,K))$, and $\hbasis$ is

1321: a basis of the twisted homology of $E(K)$ relative to $\overline{W}$.

1322: 

1323: Now, $\tau^\varphi(E(K))$ is the torsion of

1324: $C_*^\varphi(E(K))$, and

1325: this complex can be radically different from the previous one. For instance,

1326: when $M$ is a homology sphere, the absolute complex is always acyclic, while

1327: the complex relative to $\overline{W}$, which depends only on

1328: the framed knot $K^{(v)}$, very often is not. 

1329: We will see how to overcome this difficulty

1330: using the fundamental multiplicativity properties of torsion.

1331: 

1332: \paragraph{How to turn a torus into black}

1333: We will describe in this paragraph two explicit methods for

1334: modifying $\xi(v,K)$ to an Euler structure $\beta(v,K)$

1335: such that $\partial E(K)$ becomes monochromatic black.

1336: These methods are respectively a geometric and an algebraic one. 

1337: The fact that they actually lead to the same result is

1338: true but not very important, so we will omit the proof.

1339: Both methods involve the choice of an orientation of $K$.

1340: The first method is explained in a cross-section in Fig.~\ref{geomblak}

1341: \begin{figure}

1342: \centerline{\psfig{file=geomblak.eps,width=8cm}}

1343: \caption{\label{geomblak} Black field on a knot complement.}

1344: \end{figure}

1345: The cross-section is transversal to $K$, and the apparent singularity

1346: of the modified field is removed by summing a field parallel to $K$ and

1347: supported near the singularity ({\em cf.}~Fig.~\ref{conc:to:conv}, where a similar

1348: method was used).

1349: 

1350: To describe the algebraic construction of $\beta$, recall that if $z$ is a 1-chain

1351: representing $\xi(v,K)$ then $\partial z$ contains,

1352: with the appropriate sign, the centres of all cells in $E(K)\setminus W$. 

1353: Knowing the subdivision rule for Euler chains (Proposition~\ref{combin:prop:statement})

1354: we can also assume that the cellularization on $W$ has a particularly simple shape.

1355: We assume it consists of rectangles as in Fig.~\ref{zw_chain} (left), where we also

1356: show a 1-chain $z_W$ having the property that $\partial z_W$ contains the centres of all

1357: cells in $W$. We can now define $\beta(v,K)$ as the Euler structure carried by $z+z_W$.

1358: The boundary of $E(K)$ is completely black with respect to this structure,

1359: because $\partial(z+z_W)$ contains the centres of all cells of $E(K)$.

1360: \begin{figure}

1361: \centerline{\psfig{file=zw_chain.eps,width=10cm}}

1362: \caption{\label{zw_chain} A 1-chain on the annulus $W$.}

1363: \end{figure}

1364: 

1365: One easily sees from both our descriptions of $\beta$ that it is canonically

1366: defined and $H_1$-equivariant.

1367: Since we will need these properties, we spell out their meaning, starting from 

1368: an oriented framed knot $K^F$ rather than a pseudo-Legendrian knot.

1369: Let $(W,B,\emptyset,C)$ be the concave boundary pattern determined

1370: by $F$ on $E(K)$: then $\beta:\eul(E(K),(W,B,\emptyset,C))\to\eul(E(K);

1371: (\emptyset,\partial E(K),\emptyset,\emptyset)$ 

1372: is well-defined (depending on $K^F$ only) and $H_1(E(K);\mz)$-equivariant.

1373: 

1374: \begin{rem}{\em If $-K$ denotes the same knot $K$ with reversed orientation then

1375: $$\alpha(\beta(v,K),\beta(v,-K))=[\lambda]\in H_1(E(K);\mz)$$

1376: where $\lambda$ is the longitude on $\partial E(K)$ determined by the framing $K^{(v)}$.}

1377: \end{rem}

1378: 

1379: A geometric interpretation of the second description of $\beta$ is possible

1380: and used below. We have mentioned that a theory of Euler structures exists in all dimensions.

1381: While the case $n\geq 4$ requires some technicalities~\cite{fourth:paper},

1382: the reader can easily work out the case $n=2$ using the case

1383: $n=3$ treated in the present paper. 

1384: And one easily sees that $z_W$ is just an Euler chain of the inward-pointing

1385: Reeb field $r_W$ on $W$ shown in Fig.~\ref{zw_chain} (right). Moreover $r_W$ can

1386: be canonically turned into an outward-pointing field $\Theta(r_W)$, which of course

1387: is the outward-pointing Reeb field (but the core spins in the opposite direction).

1388: So a torsion $\tau^\psi(W,\Theta(r_W))$ can be computed

1389: (possibly with a basis of the twisted homology added to the data).

1390: 

1391: 

1392: \paragraph{Knot torsion as a lifting of Milnor's torsion}

1393: Let as above $(v,K)$ be pseudo-Legendrian and let $\varphi:H_1(E(K);\mz)\to\Lambda$

1394: be a representation. If $i:W\to E(K)$ is the inclusion, we set 

1395: $\varphi_W=\varphi\compo i_*$. Considering the twisted homology of the pair

1396: $(E(K),W)$ we get an exact sequence

1397: $$\hh=\Big(\cdots\longrightarrow H^{\varphi_W}_i(W)\longrightarrow

1398: H^\varphi_i(E(K))\longrightarrow H^\varphi(E(K),W)\longrightarrow

1399: H^{\varphi_W}_{i-1}(W)\longrightarrow\cdots\Big).$$

1400: We choose bases $\hbasis$, $\hbasis'$, and $\hbasis''$ respectively for 

1401: $H^\varphi_*(E(K);\overline{W})$, $H^\varphi_*(E(K))$ and $H^{\varphi_W}_*(W)$,

1402: so we can compute  

1403: $\tau^\varphi(M,v,K,\hbasis)$, $\tau^\varphi(E(K),\beta(v,K),\hbasis')$ and 

1404: $\tau^{\varphi_W}(W,\Theta(r_W),\hbasis'')$. Moreover we can compute 

1405: $\tau(\hh,\hbasis,\hbasis',\hbasis'')$. The following

1406: result is a refinement of Theorem~3.2 in~\cite{milnor}, and

1407: a proof can be given imitating the argument given in~\cite{turaev:spinc}

1408: (where a special case of the result is established).

1409: 

1410: \begin{prop}\label{Alex:lift}

1411: The following equality holds:

1412: \begin{equation}\label{Alex:lift:eq}

1413: \tau^\varphi(E(K),\beta(v,K),\hbasis')=

1414: \tau^\varphi(M,v,K,\hbasis)\cdot

1415: \tau^{\varphi_W}(W,\Theta(r_W),\hbasis'')\cdot

1416: \tau(\hh,\hbasis,\hbasis',\hbasis'').

1417: \end{equation}

1418: \end{prop}

1419: 

1420: The following comments on the previous proposition eventually explain

1421: in what sense our torsion can be viewed as a lifting of the classical torsion

1422: (in particular, Milnor torsion and the Alexander invariant).

1423: 

1424: \begin{rem}{\em 

1425: In equation~(\ref{Alex:lift:eq}) the term $\tau^\varphi(E(K),\beta(v,K),\hbasis')$

1426: is one of Turaev's torsion, so it is indeed a lifting of the classical torsion.

1427: The term $\tau^\varphi(M,v,K,\hbasis)$ is the torsion for pseudo-Legendrian knots

1428: introduced in this paper, while 

1429: $\tau^{\varphi_W}(W,\Theta(r_W),\hbasis'')$ and 

1430: $\tau(\hh,\hbasis,\hbasis',\hbasis'')$ can be viewed as normalizing terms.

1431: One can for instance choose homology bases so that $\tau(\hh,\hbasis,\hbasis',\hbasis'')=1$,

1432: and note that $\tau^{\varphi_W}(W,\Theta(r_W),\hbasis'')$ depends only

1433: on the framed knot $K^{(v)}$, not on the Euler structure.}

1434: \end{rem}

1435: 

1436: \begin{rem}{\em 

1437: The factor $\tau^{\varphi_W}(W,\Theta(r_W),\hbasis'')$ 

1438: may be understood quite easily. Denoting by $1$ the generator of $H_1(W;\mz)$,

1439: the result depends on $\varphi_W(1)$. If

1440: $\varphi_W(1)=1$ then the $\varphi_W$-twisted homology of $W$ is not

1441: twisted at all, so it is non-zero and free, and we can

1442: choose $\hbasis''$ so that $\tau^{\varphi_W}(W,\Theta(r_W),\hbasis'')=1$.

1443: On the contrary, if $\varphi_W(1)-1$ is invertible, then the $\varphi_W$-twisted homology

1444: is zero, and $\tau^{\varphi_W}(W,\Theta(r_W))$  is computed to be

1445: $\varphi_W(1)-1$. In the intermediate cases where $\varphi_W(1)-1$ is neither zero nor a unit,

1446: which can only occur when $\Lambda$ is not a field,

1447: $\tau^{\varphi_W}(W,\Theta(r_W))$ is likely not to be defined.}

1448: \end{rem}

1449: 

1450: \begin{rem}{\em

1451: We can further specialize the understanding of $\tau^{\varphi_W}(W,\Theta(r_W),\hbasis'')$ 

1452: when $M$ is a homology sphere and $\varphi:H_1(E(K);\mz)\to\Lambda$ is the representation

1453: which gives the Milnor torsion. In this case $\varphi_W(1)=\varphi(\mu)^n$, where $\mu$ is the

1454: meridian of $K$ and $n$ is the framing. So $\tau^{\varphi_W}(W,\Theta(r_W),\hbasis'')$

1455: can be normalized to be $1$ for $\varphi(\mu)^n=1$, and it equals $\varphi(\mu)^n-1$ when

1456: $\varphi(\mu)^n\neq1$}\end{rem}

1457: 

1458: 

1459: 

1460: 

1461: 

1462: \subsection{Torsion as a relative invariant}

1463: We study in this section how torsion can be employed to distinguish

1464: pseudo-Legendrian knots from each other. We first show that as a relative invariant

1465: torsion is only well-defined as a multi-valued function, the ambiguity being

1466: given by the action of a group. Then we concentrate on the knots (called `good' below)

1467: for which this action is trivial, and we interpret the relative information carried

1468: by torsion as a difference of winding numbers (or Maslov indices).

1469: 

1470: 

1471: 

1472: 

1473: 

1474: \paragraph{Group action on Euler structures}

1475: Consider a knot $K$ and a

1476: self-diffeomorphism $f$ of $E(K)$ which is the identity near $\partial E(K)$.

1477: Then $f$ extends to a self-diffeomorphism $\widehat f$ of $M$, where 

1478: $\widehat f\ristr{U(K)}={\rm id}_{U(K)}$. We define $G(K)$ as the group of all such $f$'s

1479: with the property that $\widehat f$ is isotopic to the identity on $M$.

1480: Elements of $G(K)$ are regarded up to isotopy relative to $\partial E(K)$.

1481: If $F$ is a framing on $K$ then the pull-forward of vector fields induces

1482: an action of $G(K)$ on $\eul(E(K),\pp(K^{(v)})$. We will now see 

1483: that an obstruction to weak equivalence can be expressed in terms

1484: this group action.

1485: 

1486: Let $(v_0,K_0)$ and $(v_1,K_1)$ be pseudo-Legendrian pairs in $M$, and assume

1487: that  $K_0^{(v_0)}$ is framed-isotopic to $K_1^{(v_1)}$ under a diffeomorphism

1488: $f$ relative to $\partial M$. Using the restriction of $f$ and the pull-back of

1489: vector fields we get a bijection

1490: $$f^*:\eul(E(K_1),\pp(K_1^{(v_1)}))\to\eul(E(K_0),\pp(K_0^{(v_0)})).$$

1491: 

1492: \begin{prop}\label{group:obstr}

1493: Under the current assumptions, if $(v_0,K_0)$ and $(v_1,K_1)$ are weakly equivalent to each

1494: other then $f^*(\xi(v_1,K_1))$ belongs to the $G(K_0)$-orbit of $\xi(v_0,K_0)$ in

1495: $\eul(E(K_0),\pp(K_0^{(v_0)})$.

1496: \end{prop}

1497: 

1498: \dim{group:obstr}

1499: By assumption $K_0,K_1$ and $v_0,v_1$ embed in continuous families $(K_t)_{t\in[0,1]}$

1500: and $(v_t)_{t\in[0,1]}$, where $v_t$ is transversal to $K_t$ for all $t$. Now

1501: $(K_t^{(v_t)})_{t\in[0,1]}$ is a framed-isotopy, so 

1502: there exists a continuous family $(g_t)_{t\in[0,1]}$ of diffeomorphisms of $M$

1503: fixed on $\partial M$ and such that $g_0={\rm id}_M$ and $g_t(K_0^{(v_0)})=K_t^{(v_t)}$. 

1504: So we get a map

1505: $$[0,1]\ni t\mapsto

1506: \alpha(\xi(v_0,K_0),g_t^*(\xi(v_t,K_t)))\in H_1(E(K_0);\mz).$$

1507: Since $H_1(E(K_0);\mz)$ is discrete and the map is continuous,

1508: we deduce that the map is identically 0. So $g_1^*(\xi(v_1,K_1))=\xi(v_0,K_0)$.

1509: Now

1510: $$f^*(\xi(v_1,K_1))=(f^*\compo (g_1)_*\compo g_1^*)(\xi(v_1,K_1))=

1511: (f^{-1}\compo g_1)_*(\xi(v_0,K_0))$$

1512: and the conclusion follows because $f^{-1}\compo g_1$ defines an element of $G(K_0)$.

1513: \finedim{group:obstr}

1514: 

1515: The group $G(K)$ is in general rather difficult to understand (see~\cite{hatcher}), 

1516: so we introduce a special terminology for the case where its action

1517: can be neglected. We will say that a framed knot $K^F$ is {\em good} if

1518: $G(K)$ acts trivially on $\eul(E(K),\pp(K^F))$. If $K^F$ is good for all framings $F$,

1519: we will say that $K$ itself is good.

1520: The following are easy examples of good knots:

1521: \begin{itemize} 

1522: \item $M$ is $S^3$ and $K$ is the trivial knot;

1523: \item $M$ is a lens space $L(p,q)$ and $K$ is the core of one of the handlebodies

1524: of a genus-one Heegaard splitting of $M$.

1525: \end{itemize} 

1526: \noindent The reason is that in both cases $E(K)$ is a solid torus, and we know that an automorphism of

1527: the solid torus which is the identity on the boundary is isotopic 

1528: to the identity relatively to the boundary, so $G(K)$ is trivial. The next three

1529: results show that on one hand $G(K)$ is very seldom trivial, but on the other hand 

1530: many knots are good. We will give proofs in the sequel, after introducing

1531: some extra notation. In the statements, by `$E(K)$ is hyperbolic' 

1532: we mean `$\interior(E(K))$ is complete, finite-volume hyperbolic.'

1533: 

1534: \begin{prop}\label{seldom:triv}

1535: If $M$ is closed and $E(K)$ is hyperbolic then $G(K)$ is non-trivial.

1536: \end{prop}

1537: 

1538: \begin{teo}\label{many:good}

1539: If $M$ is closed, $E(K)$ is hyperbolic and 

1540: either ${\rm Out}(\pi_1(E(K)))$ is trivial or $H_1(E(K);\mz)$ is torsion-free

1541: then $K$ is good.

1542: \end{teo}

1543: 

1544: \begin{teo}\label{homology:good}

1545: If $M$ is a homology sphere then every knot in $M$ is good.

1546: \end{teo}

1547: 

1548: The next result, which follows directly from Proposition~\ref{group:obstr},

1549: the definition of goodness, and Proposition~\ref{homeo:action},

1550: shows that for good knots torsion can be used as an

1551: obstruction to weak (and hence strong) equivalence.

1552: 

1553: \begin{prop}\label{torsion:obstr}

1554: Let $(v_0,K_0)$ and $(v_1,K_1)$ be pseudo-Legendrian pairs in $M$, and assume that 

1555: $K_0^{(v_0)}$ is framed-isotopic to $K_1^{(v_1)}$ under a diffeomorphism $f$ relative

1556: to $\partial M$. Suppose that $K_0^{(v_0)}$ is good, and that for some representation

1557: $\varphi:\pi_1(E(K_0))\to\Lambda$ and some $\Lambda$-basis $\hbasis$ of

1558: $H^\varphi_*(E(K_0),\overline{W(\pp(K_0^{(v_0)}))})$ we have

1559: \begin{equation}\label{tors:obs:eq}

1560: \tau^\varphi(E(K_0),\pp(K_0^{(v_0)}),\xi(v_0,K_0),\hbasis)\neq

1561: \tau^{\varphi\compo f_*^{-1}}(E(K_1),\pp(K_1^{(v_1)}),\xi(v_1,K_1),f_*(\hbasis)).

1562: \end{equation}

1563: Then $(v_0,K_0)$ and $(v_1,K_1)$ are not weakly equivalent.

1564: \end{prop}

1565: 

1566: \begin{rem}{\em

1567: \begin{enumerate}

1568: \item The right-hand side of equation~(\ref{tors:obs:eq}) actually equals

1569: \begin{eqnarray*}

1570: & & \tau^\varphi(E(K_0),\pp(K_0^{(v_0)}),f^*(\xi(v_1,K_1)),\hbasis) \\

1571: &=& \overline{\varphi}(\alpha(v_0,f^*(v_1))\cdot

1572: \tau^\varphi(E(K_0),\pp(K_0^{(v_0)}),\xi(v_0,K_0),\hbasis).

1573: \end{eqnarray*}

1574: This shows that the most torsion can capture as a relative invariant

1575: of $(v_0,K_0)$ and $(v_1,K_1)$ is $\alpha(v_0,f^*(v_1))$. 

1576: We will show below that in some cases torsion indeed allows

1577: to determine $\alpha(v_0,f^*(v_1)$ completely.

1578: \item By definition of goodness the homology class

1579: $\alpha(v_0,f^*(v_1))$ just considered is actually

1580: independent of $f$. We will denote it by $\alpha((K_0,v_0),(K_1,v_1))$.

1581: \item For non-good knots the relative invariant is an orbit of the action of

1582: $G(K_0)$. So an obstruction in terms of torsion could be given also for non-good knots,

1583: but the statement would become awkward, and we have 

1584: refrained from giving it.

1585: \item If equation~(\ref{tors:obs:eq}) holds for some basis $\hbasis$ then it holds

1586: for any basis.

1587: \end{enumerate}}

1588: \end{rem}

1589: 

1590: \noindent To conclude this paragraph we note that using the technology of

1591: Turaev~\cite{turaev:Euler},  one can actually see that the action on Euler

1592: structures of an automorphism  is invariant under {\em homotopy} (not only

1593: isotopy) relative to the boundary. We will not use this fact.

1594: 

1595: \paragraph{Good knots} 

1596: We introduce now some notation needed for the

1597: proofs of Proposition~\ref{seldom:triv} and Theorem~\ref{many:good} (for 

1598: Theorem~\ref{homology:good} we will use a different approach, see below).

1599: Recall that $(M,\pp)$ is fixed for the whole section. We temporarily fix also

1600: a framed knot $K^F$ in $M$, a regular neighbourhood $U$ of $K$, and we denote by $T$ the

1601: boundary torus of $U$. On $T$ we consider 1-periodic coordinates $(x,y)$

1602: such that $x\mapsto (x,0)$ is a meridian of $U$ and $y\mapsto (0,y)$ is 

1603: a longitude compatible with $F$. We denote a collar of $T$ in $E(K)$ by 

1604: $V$ and parameterize $V$ as $T\times[0,1]$, where $T=T\times\{0\}$. We consider on

1605: $[0,1]$ a coordinate $s$. For $p,q\in\mz$ we define automorphisms

1606: $\dd_{(p,q)}$ of $E(K)$ as follows. 

1607: Each $\dd_{(p,q)}$ is supported in $V$, and on $V$, using the

1608: coordinates just described, it is given by 

1609: $$\dd_{(p,q)}(x,y,s)=(x+p\cdot s, y +q\cdot s, s).$$

1610: We will call such a map a {\em Dehn twist}. It is easy to verify that

1611: the extension of $\dd_{(p,q)}$ to $M$ is isotopic to the identity of $M$.

1612: Note that $\dd_{(p,q)}$ is actually not smooth on $T\times\{1\}$, but 

1613: we can consider some smoothing and identify $\dd_{(p,q)}$

1614: to an element of $G(K)$, because the equivalence class is

1615: independent of the smoothing.

1616: 

1617: \dim{seldom:triv}

1618: We show that $\dd_{(p,q)}$ is non-trivial in $G(K)$ for all $(p,q)\neq(0,0)$. 

1619: Fix the basepoint $a_0=(0,0)\in T$ for the fundamental groups of $T$ and $E(K)$. Then 

1620: $\dd_{(p,q)}$ acts on $\pi_1(E(K),a_0)$ as the conjugation by $i_*(p,q)$, where 

1621: $i:T\to E(K)$ is the inclusion and $(p,q)\in\mz\times\mz=\pi_1(T,a_0)$.

1622: If $\dd_{(p,q)}$ is trivial in $G(K)$, {\em i.e.} it is isotopic to the identity

1623: relatively to $\partial E(K)$, in particular it acts trivially on 

1624: $\pi_1(E(K),a_0)$. This implies that $i_*(p,q)$ is in the centre of $\pi_1(E(K),a_0)$.

1625: Now it follows from hyperbolicity that this centre is trivial and $i_*$ is injective,

1626: whence the conclusion.

1627: \finedim{seldom:triv}

1628: 

1629: \noindent The proof of Theorem~\ref{many:good} will rely on properties of hyperbolic

1630: manifolds and on the following fact, which we consider to be quite remarkable

1631: (note that the 2-dimensional analogue,

1632: which may be stated quite easily, is false).

1633: Remark that the result applies in particular to Dehn twists.

1634: 

1635: \begin{prop}\label{collar:trivial}

1636: If $[f]\in G(K)$ and $f$ is supported in the collar $V$ of $\partial U$ then $[f]$

1637: acts trivially on $\eul(E(K),\pp(K^F))$.

1638: \end{prop}

1639: 

1640: \dim{collar:trivial}

1641: Consider a vector field $v$ on $E(K)$ compatible with $\pp(K^F)$. Since $v$ and $f_*(v)$

1642: differ only on $V$, their difference belongs to the image of $H_1(V;\mz)$ in

1643: $H_1(E(K);\mz)$. So we may as well assume that $E(K)=V$, {\em i.e.} $M$ is

1644: the solid torus $U\cup V$.

1645: 

1646: By contradiction, let $\xi\in\eul(V,\pp(K^F))$ be such that 

1647: $\alpha(\xi,(\dd_{(p,q)})_*(\xi))$ is non-zero in $H_1(V;\mz)$, so it is given

1648: by $k\cdot[\gamma]$ for some $k\in\mz\setminus\{0\}$ and some simple closed

1649: curve $\gamma$ on $T\times\{1\}\subset\partial V$. Let us now take another

1650: simple closed curve $\delta$ on $T\times\{1\}$ which intersects $\gamma$

1651: transversely at one point. Let us define $N$ as the manifold obtained by

1652: attaching the solid torus to $V$ along $T\times\{1\}$, in such a way that the

1653: meridian of the solid torus gets identified  with $\delta$. Note that $N$ is

1654: again a solid torus and that the homology class of $\gamma$ in

1655: $H_1(N;\mz)\cong\mz$ is a generator. Now we can apply

1656: Proposition~\ref{p:h:formula} to extend $\xi$ to an Euler structure $\xi_N$ on

1657: $N$. Moreover we can extend $f$ to an automorphism $g$ of $N$ which is the

1658: identity on $\partial N=T\times\{0\}$. Now by construction

1659: $\alpha(\xi_N,g_*(\xi_N))$ equals $k\cdot[\gamma]$ in $H_1(N;\mz)\cong\mz$, so

1660: it is non-zero. But $g$ is isotopic to the identity of $N$ relatively to the

1661: boundary of $N$, so we have a contradiction. \finedim{collar:trivial}

1662: 

1663: For the proof of Theorem~\ref{many:good} we will also need the

1664: following easy fact. 

1665: 

1666: \begin{lem}\label{power:action}

1667: Let $f$ be an automorphism of $M$ relative to $\partial M$, and consider the

1668: induced automorphisms of $H_1(M;\mz)$ and $\eul(M,\pp)$, both denoted by $f_*$. Then:

1669: $$\alpha(f_*(\xi_0),f_*(\xi_1))=f_*(\alpha(\xi_0,\xi_1)),\qquad

1670: \forall\xi_0,\xi_1\in\eul(M,\pp).$$

1671: \end{lem}

1672: 

1673: \dim{power:action}

1674: Take representatives of $\xi_0$ and $\xi_1$ such that $\alpha(\xi_0,\xi_1)$ can be

1675: viewed as the anti-parallelism locus. The formula is then obvious.

1676: \finedim{power:action}

1677: 

1678: \dim{many:good} Consider $[f]\in G(K)$. It follows from the work of Johansson

1679: (see~\cite{hatcher}) that, under the assumption that $E(K)$ is hyperbolic, the

1680: group generated by Dehn twists has finite index in the mapping class group of

1681: $E(K)$ relative to the boundary. More precisely, the quotient group can be

1682: identified to a subgroup of ${\rm Out}(\pi_1(E(K))$, which is finite as a

1683: consequence of Mostow's rigidity. If ${\rm Out}(\pi_1(E(K))$ is trivial then

1684: $[f]$ is equivalent to a Dehn twist, so $f$ acts trivially on

1685: $\eul(E(K),\pp(K^F))$ by Proposition~\ref{collar:trivial}. 

1686: 

1687: We are left to deal with the case where $H_1(E(K);\mz)$ is torsion-free. By

1688: Johansson's result, there exists an integer $n$ such that $f^n$ acts trivially

1689: on $\eul(E(K),\pp(K^F))$. Consider now $\xi\in\eul(E(K),\pp(K^F))$, and set

1690: $\alpha=\alpha(\xi,f_*(\xi))$.  We must show that  $\alpha=0$. We denote by

1691: $\widehat\alpha$ the image of $\alpha$ in $H_1(M;\mz)$, and by  $\widehat f$

1692: the extension of $f$ to $M$. Since $\widehat f$ is isotopic to the identity,

1693: we  have $\widehat f_*(\widehat\alpha)=\widehat\alpha$. If we take an oriented

1694: 1-manifold $a$ representing $\alpha$ and disjoint from $\partial U(K)$, this

1695: means that there exists an  oriented surface $\Sigma$ in $M$ such that

1696: $\partial\Sigma=a\cup(-f(a))$. Up to isotopy we can  assume that $\Sigma$

1697: intersects $\partial U(K)$ transversely in a union of circles. This shows that

1698: $f_*(\alpha)=\alpha+k\cdot\mu$, where $\mu$ is the meridian of $T$. Note that

1699: $f_*(\mu)=\mu$, so for all integers $m$ we have $f_*^m(\alpha)=\alpha+m\cdot

1700: k\cdot\mu$. Now, using Lemma~\ref{power:action}, we have:

1701: 	\begin{eqnarray*}

1702: 0 & = & \alpha(\xi,f_*^n(\xi)=\sum_{m=0}^{n-1}\alpha(f_*^m(\xi),f_*^{m+1}(\xi))\\

1703: & = & \sum_{m=0}^{n-1} f_*^m(\alpha(\xi,f_*(\xi)))=\sum_{m=0}^{n-1}f_*^m(\alpha)=

1704: \sum_{m=0}^{n-1}(\alpha+m\cdot k\cdot\mu)\\

1705: & = & n\cdot \alpha +{n(n-1)\over 2}\cdot k\cdot \mu.

1706: \end{eqnarray*}

1707: This shows that $2\cdot \alpha+(n-1)\cdot k\cdot \mu$ is a torsion element of $H_1(E(K);\mz)$,

1708: so it is null by assumption. So $(1-n)\cdot k\cdot \mu=2\cdot \alpha$. If we apply $f_*$ to

1709: both sides of this equality we get

1710: $(1-n)\cdot k\cdot f_*(\mu)=2\cdot f_*(\alpha)$. Using the equality again and the relations

1711: $f_*(\mu)=\mu$ and $f_*(\alpha)=\alpha+k\cdot\mu$ we get

1712: $$(1-n)\cdot k\cdot \mu=2\cdot\alpha+2\cdot k\cdot\mu=(1-n)\cdot k\cdot

1713: \mu+2\cdot k\cdot\mu.$$

1714: Therefore $k\cdot\mu$ is a torsion element, and hence null. But

1715: $2\cdot\alpha=(1-n)\cdot k\cdot\mu$, so also $\alpha$ is null.\finedim{many:good}

1716: 

1717: 

1718: \paragraph{Rotation number, and more good knots} 

1719: We will show in this section that 

1720: in a homology sphere the rotation number of a pseudo-Legendrian knot can be 

1721: (defined and) expressed 

1722: in terms of Euler structures on the exterior. This will lead us to a 

1723: simple interpretation of torsion as a relative invariant of knots,

1724: and it will allow us to show

1725: that in a homology sphere all knots are good (Theorem~\ref{homology:good}).

1726: 

1727: To begin, we note that the definition of the rotation number, classically defined in the

1728: contact case, actually extends to the situation we are considering. Since we will need

1729: this definition, we recall it. Let $M$ be a homology sphere, let

1730: $v$ be a field on $M$ and let $K$ be an oriented pseudo-Legendrian knot 

1731: in $(M,v)$. Take a plane field $\eta$ transversal to $v$ and tangent to $K$, 

1732: and a Seifert surface

1733: $S$ for $K$. Up to isotopy of $S$ we can assume that $\eta$ is tangent to

1734: $S$ only at isolated points. Then ${\rm rot}_v(K)$ is the sum of a contribution

1735: for each of these tangency points $p$. Define ${\rm o}(p)$ to be $+1$ if

1736: $\eta_p=T_pS$ and $-1$ if $\eta_p=-T_pS$. If $p\in\partial S=K$ then $p$ contributes

1737: just with ${\rm o}(p)$. If $p\in\interior(S)$ we can consider near $p$ a section of 

1738: $\eta\cap TS$ which vanishes at $p$ only, and denote by ${\rm i}(p)$ its index. Then

1739: $p$ contributes to ${\rm rot}_v(K)$ with ${\rm o}(p)\cdot{\rm i}(p)$.

1740: 

1741: It is quite easy to see that the resulting number is indeed independent from $\eta$

1742: and $S$. Moreover ${\rm rot}_v(K)$ is unchanged under homotopies of $v$ relative to $K$,

1743: and local modifications away from $K$, so we can actually define

1744: ${\rm rot}_\xi(K)$ where $\xi=\xi(v,K)\in\eul(E(K),\pp(K^{(v)})$.

1745: 

1746: \begin{prop}\label{alpha:rot}

1747: Let $M$ be a homology sphere, let $v$ be a field on $M$ 

1748: and let $K_0$ and $K_1$ be oriented pseudo-Legendrian knots in $(M,v)$. 

1749: Assume that there exists a framed-isotopy $f$ which maps $K_1^{(v)}$ to 

1750: $K_0^{(v)}$. Identify $H_1(E(K_0);\mz)$ to $\mz$ using a meridian. Then:

1751: $${\rm rot}_v(K_1)={\rm rot}_v(K_0)+2\alpha(f_*(\xi(v,K_1)),\xi(v,K_0)).$$

1752: \end{prop}

1753: 

1754: \dim{alpha:rot}  Let $K:=K_0$, $v_0:=v$ and $v_1:=f_*(v)$. Note that $v_0$ and

1755: $v_1$ coincide along $K$.  Of course ${\rm rot}_v(K_1)={\rm rot}_{v_1}(K)$. We

1756: are left to show that 

1757: $${\rm rot}_{v_1}(K)={\rm rot}_{v_0}(K)+2\alpha(\xi(v_1,K)),\xi(v_0,K)).$$ 

1758: We can now homotope $v_0$ and $v_1$ away from $K$ until they differ only in

1759: the  neighbourhood $W(L)$ of an oriented link $L$, and within this

1760: neighbourhood they differ exactly by a ``Pontrjagin move'', as defined for

1761: instance in~\cite{lnm}. Namely, $v_0$  runs parallel to $L$ in $W(L)$, while

1762: $v_1$ runs opposite to $L$ on $L$ and has non-positive radial component on

1763: $W(L)$ (see below for a picture). Note that $L$ represents

1764: $\alpha(\xi(v_1,K)),\xi(v_0,K))$.

1765: 

1766: Let us choose now a Seifert surface $S$ for $K$ and a Riemannian metric on $M$,

1767: and define $\eta_i=v_i^\perp$, for $i=0,1$. Since

1768: $\eta_0\ristr{K}=\eta_1\ristr{K}$, the contributions along $K$ to ${\rm

1769: rot}_{v_0}(K)$ and ${\rm rot}_{v_1}(K)$ are the same. Up to isotoping $S$ we

1770: may assume that $L$ is transversal but never orthogonal to $S$. At the points

1771: where $\eta_0$ is tangent to $S$ also $\eta_1$ is tangent to $S$, and the

1772: contributions are the same. So ${\rm rot}_{v_1}(K)-{\rm rot}_{v_0}(K)$ is given

1773: by the sum of the contributions of the tangency points of $\eta_1$ to $S$

1774: within $W(L)$. We will show that each point of $L\cap S$ gives rise to exactly

1775: two tangency points, which both contribute with $+1$ or $-1$ according to the

1776: sign of the intersection of $L$ and $S$ at the point. This will show that ${\rm

1777: rot}_{v_1}(K)-{\rm rot}_{v_0}(K)$ is twice the algebraic intersection of $L$

1778: and $S$. This algebraic intersection is exactly the value of

1779: $[L]=\alpha(\xi(v_1,K)),\xi(v_0,K))$ as a multiple of $[m]$, so the local

1780: analysis at $L\cap S$ will imply the desired conclusion.

1781: 

1782: For the sake of simplicity we only examine a positive intersection point of $L$

1783: and $S$. This is done in a cross-section in Fig.~\ref{rot:alpha:fig}, which

1784: shows the local effect

1785: \begin{figure}

1786: \centerline{\psfig{file=rotalfa.eps,width=15.5cm}}

1787: \caption{\label{rot:alpha:fig}Effect of the Pontrjagin move.}

1788: \end{figure}

1789: of the move. The fields pictured both have a rotational symmetry, suggested in the

1790: figure. The two tangency points which arise are a positive focus (on the right) and

1791: a negative saddle (on the left), 

1792: so the local contribution is indeed $+2$, and the proof is complete.

1793: \finedim{alpha:rot}

1794: 

1795: \begin{rem}{\em The definition of rotation number and

1796: Proposition~\ref{alpha:rot}

1797: easily extend to the case of manifolds which are not homology spheres, 

1798: by restricting to homologically trivial knots and choosing a relative homology class in the

1799: exterior.}

1800: \end{rem}

1801: 

1802: \noindent We can now prove that in a homology sphere all knots are good.

1803: 

1804: \dim{homology:good}

1805: Consider $[f]\in G(K)$, a framing $F$ on $K$ and $\xi\in\eul(E(K),\pp(K^F))$. We must

1806: show that $f_*(\xi)=\xi$. Let $\xi=[v]$ and denote by $\widehat{v}$ the

1807: obvious extension of $v$ to $M$. As above, let $\widehat{f}$ be the extension of $f$ to $M$.

1808: During the proof of Proposition~\ref{alpha:rot} we have shown that

1809: $${\rm rot}_{\widehat{f}_*(\widehat{v})}(K)-{\rm rot}_{\widehat{v}}(K)=

1810: 2\alpha(f_*(v),v).$$

1811: 

1812: But ${\rm rot}_{\widehat{f}_*(\widehat{v})}(K)$ is actually equal to 

1813: ${\rm rot}_{\widehat{v}}(K)$, because 

1814: $\widehat{f}$ is the identity near $K$. Therefore

1815: $f_*(v)$ and $v$ differ by a torsion element of

1816: $H_1(E(K);\mz)\cong\mz$, so they are equal. By definition

1817: $f_*(\xi)=[f_*(v)]$ and $\xi=[v]$, and the proof

1818: is complete.

1819: \finedim{homology:good}

1820: 

1821: Theorems~\ref{many:good} and~\ref{homology:good} provide a partial answer

1822: to the problem of determining which knots are good. The general problem does not appear 

1823: to be straight-forward, and we leave it for further investigation.

1824: We will only show below an example of knot which is not good.

1825: 

1826: 

1827: \paragraph{Curls and the winding number}  

1828: We show in this paragraph the relation between the relative invariant 

1829: $\alpha((v_0,K_0),(v_1,K_1))$ of two pseudo-Legendrian knots (when

1830: this invariant is well-defined) and

1831: an analogue of the winding number (the invariant which allows

1832: to distinguish framed-isotopic planar link diagrams which are not equivalent under

1833: the second and third of Reidemeister's moves, see~\cite{trace}). 

1834: Moreover we will give an

1835: example of knot which is not good. The proof of the next result uses the

1836: example of Section~\ref{exa:section}, so it is deferred to

1837: Section~\ref{proofs}.

1838: 

1839: \begin{prop}\label{wind:sensitive}

1840: Consider a field $v$ on $M$ and a portion of $M$ on which $v$

1841: can be identified to the vertical field in $\mr^3$. Consider oriented knots $K_0$ and

1842: $K_{\pm1}$ which are transversal to $v$ and differ only within the chosen portion of $M$,

1843: as shown in Fig.~\ref{wind:fig}.

1844: \begin{figure}

1845: \centerline{\psfig{file=new_wind_fig.eps,width=12cm}}

1846: \caption{\label{wind:fig}Knots which differ for a positive or a negative double curl.}

1847: \end{figure}

1848: Choose the positive meridian $m$ of $K_0$, as also shown in the figure.

1849: Let $f$ be an isotopy which maps $K_{\pm1}^{(v)}$ to $K_0^{(v)}$ and is supported in

1850: a tubular neighbourhood of $K_0$. Then:

1851: $$\alpha(\xi(v,K_0),f_*(\xi(v,K_{\pm1})))=\pm[m]\in H_1(E(K_0);\mz).$$

1852: \end{prop}

1853: 

1854: \begin{prop}\label{hyper:curl} Let $(v,K_0)$ be a pseudo-Legendrian pair in

1855: $M$, and denote by $[m]\in H_1(E(K_0);\mz)$ the homology class of the meridian of

1856: $U(K_0)$. Assume either that $K_0^{(v)}$ is good and $[m]\neq 0$ or that

1857: $E(K_0)$ is hyperbolic and $[m]$ has infinite order.

1858: Let $K_1$ be a knot obtained

1859: from $K_0$ as in Fig.~\ref{wind:fig}. Then $(v,K_0)$ and $(v,K_1)$  are not

1860: weakly equivalent. \end{prop}

1861: 

1862: \dim{hyper:curl} By contradiction, using 

1863: Propositions~\ref{group:obstr} and~\ref{wind:sensitive}, we would get elements

1864: $\xi_0,\xi_1$ of $\eul(E(K_0),\pp(K_0^{(v)})$ such that $\alpha(\xi_0,\xi_1)=[m]$

1865: and $\xi_1=f_*(\xi_0)$ for some $[f]\in G(K_0)$. If $K_0^{(v)}$ is good and $[m]\neq 0$ 

1866: this is a contradiction. Assume now that $E(K_0)$ is hyperbolic and $[m]$ has infinite order.

1867: Since $f_*([m])=[m]$, using 

1868: Lemma~\ref{power:action} we easily see that $\alpha(\xi_0,f_*^k(\xi_0))=k\cdot[m]$ for

1869: all $k$. Proposition~\ref{collar:trivial} and the result of Johansson already used in the proof of

1870: Theorem~\ref{many:good}

1871: now imply that $f^k$ acts trivially on $\eul(E(K_0),\pp(K_0^{(v)})$ for some

1872: $k$, whence the contradiction.\finedim{hyper:curl}

1873: 

1874: 

1875: 

1876: \noindent As an application of Proposition~\ref{wind:sensitive}, 

1877: we can show that there exist knots which are not good. Consider $S^2\times[0,1]$

1878: with vector field parallel to the $[0,1]$ factor. Let $K_0$ be the equator

1879: of $S^2\times\{1/2\}$, and let $K_1$ be obtained from $K_0$ by the

1880: modification described in Fig.~\ref{wind:fig}.

1881: Using Proposition~\ref{wind:sensitive}, if we choose a framed-isotopy

1882: $g$ of $K_1^{(v)}$ onto $K_0^{(v)}$ supported in $U(K_0)$, we have

1883: $$\alpha(\xi(v,K_0),(g\ristr{E(K_1)})_*(\xi(v,K_1)))=[m],$$

1884: where $[m]$ is a generator of $H_1(E(K_0);\mz)\cong\mz$.

1885: On the other hand, $K_1$ is strongly equivalent to $K_0$ in $(M,v)$

1886: (the winding number only exists on $\mr^2$, not on $S^2$). 

1887: So there exists an isotopy $h$ of $K_1^{(v)}$ onto $K_0^{(v)}$ through links

1888: transversal to $v$, and we have

1889: $$\alpha(\xi(v,K_0),(h\ristr{E(K_1)})_*(\xi(v,K_1)))=0.$$

1890: This implies that $(h\compo g^{-1})\ristr{E(K_0)}$ acts non-trivially

1891: on $\xi(v,K_0)\in\eul(E(K_0),\pp(K_0^{(v)}))$.

1892: 

1893: 

1894: To conclude our discussion on the relative invariant $\alpha$ between two pseudo-Legendrian

1895: knots, we state now a result proved in~\cite{second:paper}. The consequences we deduce

1896: easily follow from Proposition~\ref{wind:sensitive}.

1897: 

1898: \begin{prop}\label{curl:diff:prop}

1899: Let $(v_0,K_0)$ and $(v_1,K_1)$ be pseudo-Legendrian in $M$, assume that 

1900: $v_0$ and $v_1$ are homotopic fields, and that $K_0^{(v_0)}$ and $K_1^{(v_1)}$

1901: are isotopic as framed knots. Then $(v_0,K_0)$ and $(v_1,K_1)$ become

1902: weakly equivalent up to a finite number of local moves

1903: $K_0\longrightarrow K_{\pm1}$ as in Fig.~\ref{wind:fig}.

1904: \end{prop}

1905: 

1906: \begin{cor}\label{rel:wind:cor}

1907: Under the same assumptions, assume also that $K_0^{(v_0)}$ is good, so

1908: $\alpha((K_0,v_0),(K_1,v_1))$ is defined. Then 

1909: $$\alpha((K_0,v_0),(K_1,v_1))=

1910: {\rm w}((K_0,v_0),(K_1,v_1))\cdot[m_0]\in H_1(E(K_0);\mz)$$

1911: where $m_0$ is the meridian of $K_0$ and 

1912: ${\rm w}((K_0,v_0),(K_1,v_1))$ is the 

1913: (non-well-defined) algebraic number of moves 

1914: $K_0\longrightarrow K_{\pm1}$ needed to make $K_0$ and $K_1$ weakly equivalent.

1915: \end{cor}

1916: 

1917: Concerning the statement of the previous corollary, note that 

1918: both ${\rm w}((K_0,v_0),(K_1,v_1))$ and $[m_0]$ depend on the

1919: choice of an orientation on $K_0$, but their product does not.

1920: 

1921: \begin{cor}

1922: Assume furthermore that $[m_0]$ has infinite order in $H_1(E(K_0);\mz)$. Then 

1923: ${\rm w}((K_0,v_0),(K_1,v_1))\in\mz$ is a well-defined integer relative

1924: invariant, which we call the {\em relative winding number}.

1925: \end{cor}

1926: 

1927: 

1928: \begin{rem}{\em 

1929: If $M$ is a homology sphere then the local moves of Fig.~\ref{wind:fig}

1930: which modify the winding number also change the rotation number, and 

1931: Corollary~\ref{rel:wind:cor} is consistent with Proposition~\ref{alpha:rot}.}\end{rem}

1932: 

1933: The next proposition contains in particular Theorem~\ref{informal:good:for:knots} stated

1934: in the introduction.

1935: 

1936: \begin{prop}\label{no:tors:same}

1937: Under the assumptions of Proposition~\ref{curl:diff:prop}, assume that 

1938: $K_0^{(v_0)}$ is good and that $[m_0]$ has infinite order in 

1939: $H_1(E(K_0);\mz)$. The following facts are pairwise equivalent:

1940: \begin{enumerate}

1941: \item\label{no:wind:point} the relative winding 

1942: number of $(K_0,v_0)$ and $(K_1,v_1)$ vanishes;

1943: \item\label{no:tors:point} all relative torsion 

1944: invariants of $(K_0,v_0)$ and $(K_1,v_1)$ are trivial;

1945: \item\label{equiv:point} $(K_0,v_0)$ and $(K_1,v_1)$ are weakly equivalent.

1946: \end{enumerate}

1947: \end{prop}

1948: 

1949: \dim{no:tors:same}

1950: Equivalence of (\ref{no:wind:point}) and (\ref{equiv:point})

1951: comes from the previous discussion and from the fact that 

1952: a positive double curl and a negative double curl cancel via

1953: weak equivalence. To show that (\ref{no:wind:point}) and

1954: (\ref{no:tors:point}) are equivalent we only need to consider torsion

1955: with respect to a representation $\varphi:H_1(E(K_0);\mz)\to\Lambda$ such that

1956: $\varphi([m_0])$ has infinite order.

1957: \finedim{no:tors:same}

1958: 

1959: 

1960: \begin{cor}\label{immers:cor}

1961: Under the assumptions of Proposition~\ref{curl:diff:prop},

1962: assume that $M$ is a homology sphere. Then the facts (1), (2), and (3) 

1963: of Proposition~\ref{no:tors:same} are also equivalent to the following:

1964: \begin{enumerate}

1965: \item[(4)]\label{no:rot:point} $(K_0,v_0)$ and $(K_1,v_1)$ 

1966: have the same rotation number.

1967: \end{enumerate}

1968: \end{cor}

1969: 

1970: 

1971: \dim{immers:cor}

1972: Equivalence of (\ref{no:wind:point}) 

1973: and (\ref{no:rot:point}) comes from the previous discussion and 

1974: Proposition~\ref{wind:sensitive}.\finedim{immers:cor}

1975: 

1976: 

1977: Since

1978: in a homology sphere two pseudo-Legendrian knots which are homotopic

1979: through pseudo-Legendrian immersions certainly have the same Maslov index,

1980: the previous corollary seems to suggest that all torsion can capture in a homology

1981: sphere is the homotopy class through immersions.

1982: We believe that it would be interesting to check if also 

1983: for a general manifold $M$, under the assumptions of Corollary~\ref{rel:wind:cor},

1984: homotopy through pseudo-Legendrian immersions implies

1985: ${\rm w}((K_0,v_0),(K_1,v_1))\cdot[m_0]=0$. We conclude by informing the reader that

1986: in~\cite{second:paper} we have discussed the extent to which the category of

1987: pseudo-Legendrian knots can be represented by the category of genuine Legendrian

1988: knots in overtwisted contact structures.

1989: 

1990: \section{An example}\label{exa:section}

1991: Figure~\ref{abalone} shows a neighbourhood of the singular set of the so-called

1992: \begin{figure}

1993: \centerline{\psfig{file=abalone.eps,width=15.5cm}}

1994: \caption{\label{abalone}The abalone, and a $\cont^1$ knot on it.}

1995: \end{figure}

1996: abalone, a branched standard spine of $S^3$, which we denote by $A$. Note that $A$ has one vertex,

1997: two edges and two regions. The figure on the left is easier to understand, but 

1998: it does not represent the genuine embedding of $A$ in $S^3$, which is instead

1999: shown in the centre (hint: compute linking numbers). On the right we show (using the

2000: easier picture) a $\cont^1$ knot $K$ on $A$. Using the genuine picture one sees that

2001: $K$ is actually trivial in $S^3$, and its framing is $-1$.

2002: So the knot exterior $E(K)$ is actually a solid torus,

2003: with an induced Euler structure $\xi$, and the white annulus $W\subset\partial E(K)$

2004: is a longitudinal one. Let us now take the representation $\varphi:\pi_1(E(K))\to\mz[t^{\pm1}]$

2005: which maps the generator to $t$. It is not hard to see that $H_*^\varphi(E(K),\overline{W})=0$, so we can

2006: compute $\tau^\varphi(E(K),\xi)$. We describe the method to be followed, skipping several details

2007: and all explicit formulae.

2008: 

2009: We can apply directly the method described in the (partial) proof of 

2010: Theorem~\ref{bounded:surg}, to get a branched standard spine $P$

2011: (in the sense of Theorem~\ref{bounded:surg}) of $E(K)$.

2012: This $P$ is easily recognized to have 5 vertices (denoted $v_1,\dots,v_5$),

2013: 10 edges (denoted $e_0,\dots,e_9$) and 6 regions (denoted $r_1,\dots r_6$).

2014: Figure~\ref{dualtria} shows the truncated ideal triangulation dual to $P$.

2015: \begin{figure}

2016: \centerline{\psfig{file=dualtria.eps,width=15.5cm}}

2017: \caption{\label{dualtria}Truncated ideal triangulation of the knot exterior.}

2018: \end{figure}

2019: In the figure the hat denotes duality as usual. We have written

2020: $-\hat e_i$ instead of $\hat e_i$ when $\hat e_i$ lies on 

2021: $\hat v_j$ but the natural

2022: orientation of $\hat e_i$ is not induced by the orientation of $\hat v_j$.

2023: The letters $S$ and $T$ refer to the boundary sphere and torus respectively 

2024: ($S$ should actually be collapsed to one point $x_0$, but the picture is

2025: easier to understand before collapse).

2026: 

2027: Recall that the algebraic complex of which we must compute the

2028: torsion has one generator for each cell in the cellularization 

2029: of $E(K)$ arising from $P$,

2030: excluding the white cells and the tangency circles on the boundary.

2031: From Fig.~\ref{dualtria} we can see how many such cells there will

2032: be in each dimension, namely 3 in dimension 0 ($x_0$ and two vertices

2033: on $T$), 14 in dimension 1 (the $\hat r_i$'s and 8 edges on $T$),

2034: 16 in dimension 2 (the $\hat e_i$'s and the 6 black kites on $T$)

2035: and 5 in dimension 3 (the $\hat v_i$'s). We can also easily describe

2036: the combinatorial Euler chain $s'(P)$ which will be used to find

2037: the preferred cell liftings: besides the orbits of the field there are

2038: only one star and one bi-arrow; the support of $s'(P)$ has 3 connected 

2039: components (one spider with 19 legs and head at $x_0$, the star union

2040: the second half of $\hat r_2$, and the bi-arrow union a segment 

2041: contained in $\hat e_3$).

2042: 

2043: To actually determine the preferred liftings we need an effective

2044: description of the lifting of the cellularization

2045: to the universal cover $\tilde E(K)\to E(K)$.

2046: Since $\pi_1(E(K))=\mz$, each cell $c$ will have liftings $\tilde c^{(n)}$

2047: for $n\in\mz$, where $\tilde c^{(n)}$ is the $n$-th translate of $\tilde c^{(0)}$.

2048: The choice of $\tilde c^{(0)}$ itself is of course arbitrary, but to

2049: understand the cover we must express the $\partial\tilde c^{(0)}$'s in terms of

2050: the other $\tilde d^{(n)}$'s. To do this we start with a lifting $\tilde x_0$

2051: of the basepoint $x_0$ and we lift the other cells one after each other,

2052: taking into account the relations in $\pi_1(E(K)$ and making sure that the union

2053: of cells already lifted is always connected. When a cell $c$ is reached for the first

2054: time, its lifting is chosen arbitrarily and declared to be $\tilde c^{(0)}$, but

2055: its boundary will involve in general $\tilde d^{(n)}$'s with $n\neq 0$.

2056: Once the lifted cellularization is known, it is a simple matter

2057: to determine preferred cell liftings: since the support of $s'(P)$ consists

2058: of 3 spiders, we only need to choose liftings of the 3 heads and then lift

2059: the legs.

2060: 

2061: Carrying out the computations we have explicitly found the algebraic complex

2062: with coefficients in $\mz[t^{\pm1}]$, and the preferred generators of the

2063: 4 moduli appearing. Then, using Maple, we have checked that indeed

2064: the complex is acyclic, and we have computed its torsion as follows:

2065: $$\tau^\varphi(E(K),\xi)=\pm t^{-1}.$$

2066: Note that as an application of Proposition~\ref{wind:sensitive},

2067:  by adding curls, we can 

2068: easily construct a family $\{K_n\}$ of pseudo-Legendrian knots such that

2069: $\tau^\varphi(E(K_n),\xi_n)=\pm t^{n}$.

2070: 

2071: 

2072: \section{Main proofs}\label{proofs}

2073: 

2074: In this section we provide all the proofs which we have

2075: omitted in the rest of the paper. We will always refer to the statements

2076: for the notation.

2077: 

2078: \dim{p:h:formula}

2079: Let us first recall the classical Poincar\'e-Hopf formula, according to which if $v$ is a

2080: vector field with isolated singularities on a manifold $M$, and $v$ points outwards on

2081: $\partial M$ ({\em i.e.}~$\partial M$ is black), then the sum of the indices of all 

2082: singularities is $\chi(M)$. Assume now that $v$ has isolated singularities and on 

2083: $\partial M$ it is compatible with a pattern $\pp=(W,B,V,C)$. We claim that 

2084: if $\cc$ is a cellularization of $M$ suited to $\pp$ we have:

2085: \begin{equation}\label{ph:formula}

2086: \sum_{x\in{\rm Sing}(v)}\index_x(v)=\chi(M)-\sum_{\sigma\in\cc,\ \sigma\subset W\cup V}

2087: \index(\sigma).

2088: \end{equation}

2089: This formula is enough to prove the statement: if a non-singular field $v$ compatible

2090: with $\pp$ exists then the left-hand side of~\ref{ph:formula} vanishes,

2091: and the right-hand side of~\ref{ph:formula} equals the obstruction of the statement.

2092: On the other hand, if the obstruction vanishes, then one can first consider a singular

2093: field compatible with $\pp$, then group up the singularities in a ball, and remove them.

2094: 

2095: To prove~\ref{ph:formula} we consider the manifold $M'$ obtained by attaching

2096: a collar $\partial M\times[0,1]$ to $M$ along $\partial M=\partial M\times\{0\}$.

2097: Of course $M'\cong M$. We will now extend $v$ to a field $v'$ on $M'$ with the property that

2098: $v'$ points outwards on $\partial M'$, and in $\partial M\times(0,1)$ the field

2099: $v'$ has exactly one singularity for each cell $\sigma\subset W\cup V$, with index

2100: $\index(\sigma)$. An application of the classical Poincar\'e-Hopf formula then

2101: implies the conclusion. The construction of $v'$ is done cell by cell. We first

2102: show how the construction goes in dimension 2, see Fig.~\ref{two:obstr}.

2103: \begin{figure}

2104: \centerline{\psfig{file=two_obstr.eps,width=12cm}}

2105: \caption{\label{two:obstr}Extension of a field to the collared manifold: dimension 2}

2106: \end{figure}

2107: 

2108: For the 3-dimensional case, it is actually convenient to choose a cellularization $\cc$

2109: of special type. Namely, we assume that $\cc\ristr{\partial M}$ 

2110: consists of rectangles and triangles, each rectangle having exactly one edge on $V\cup C$,

2111: and the union of rectangles covering a neighbourhood of $V\cup C$. We suggest 

2112: in Fig.~\ref{three:obstr:1} how to 

2113: \begin{figure}

2114: \centerline{\psfig{file=three_obstr_1.eps,width=9cm}}

2115: \caption{\label{three:obstr:1}Extension of a field to the collared manifold: white

2116: cells in dimension 3}

2117: \end{figure}

2118: define $v'$ on $\sigma\times[0,1]$ for $\sigma\subset W$ of dimension 0, 1 and 2 respectively.

2119: By the choices we have made the situation near $\partial W$ contains the 2-dimensional

2120: situation as a transversal cross-section, and it is not too difficult to extend $v'$ further

2121: and check that indices of the singularities are as required. As an example, we suggest in

2122: Fig.~\ref{three:obstr:2}

2123: \begin{figure}

2124: \centerline{\psfig{file=three_obstr_2.eps,width=7cm}}

2125: \caption{\label{three:obstr:2}Extension of a field to the collared manifold: convex edge in

2126: dimension 3}

2127: \end{figure}

2128: how to do this near a convex edge.\finedim{p:h:formula}

2129: 

2130: 

2131: \dim{reconstruction:statement}

2132: Our proof follows the scheme given by Turaev in~\cite{turaev:Euler}, with some 

2133: technical simplifications and some extra difficulties due to the tangency circles.

2134: We first recall that if $\ss$ is a (smooth) triangulation of a manifold $N$, a (singular)

2135: vector field $w_\ss$ on $N$ can be defined by the requirements 

2136: that: (1) each simplex is a union of orbits; (2)

2137: the singularities are exactly the barycentres of the

2138: simplices; (3) barycentres of higher dimensional simplices are

2139: more attractive that those of lower dimensional simplices.

2140: More precisely, each orbit (asymptotically) goes from a barycentre $p_\sigma$

2141: to a barycentre $p_{\sigma'}$, where $\sigma\subset\sigma'$.

2142: It is automatic that $\index_{p_\sigma}(w_\ss)=\index(\sigma)$.

2143: See Fig.~\ref{elem:field} for a description of $w_\ss$ on a 2-simplex of $\ss$.

2144: \begin{figure}

2145: \centerline{\psfig{file=elem_field.eps,width=4cm}}

2146: \caption{\label{elem:field}The singular field $w_\ss$ on a 2-simplex}

2147: \end{figure}

2148: 

2149: 

2150: 

2151: Let us consider now a triangulation $\calt$ of $M$, and let us choose a 

2152: representative $z$ of the given $\xi\in\eulc(M,\pp)$ as in

2153: Proposition~\ref{combin:prop:statement}(\ref{bary:sub:point}).

2154: We consider now the manifold $M$ obtained by attaching $\partial M\times[0,\infty)$

2155: to $M$ along $\partial M=\partial M\times\{0\}$. Note that $M'\cong\interior(M)$.

2156: Moreover $\calt$ extends to a ``triangulation'' $\calt'$ of $M'$, where on $M\times(0,\infty)$

2157: we have simplices with exactly one ideal vertex, obtained by taking cones

2158: over the simplices in $\partial M$ and then removing the vertex.

2159: Even if $\calt'$ is not strictly speaking a triangulation, the construction of

2160: $w_{\calt'}$ makes sense, because the missing vertex at infinity would be

2161: a repulsive singularity anyway. We arrange things in such a way that if 

2162: $\sigma\subset\partial M$ then the singularity in $\sigma\times(0,\infty)$

2163: is at height $1$, so it is $p_\sigma\times\{1\}$.

2164: 

2165: We will define now a smooth function $h:\partial M\to (0,\infty)$ and set $M_h=M\cup\{(x,t)\in\partial M\times[0,\infty):\ t\leq h(x)\}$, in such a way that

2166: $w_{\calt'}$ is non-singular on $\partial M_h$, and, modulo the natural homeomorphism

2167: $M\cong M_h$, it induces on $\partial M_h$ the desired boundary pattern $\pp$.

2168: Later we will show how to use $z$ to remove the singularities of $w_{\calt'}$ on $M_h$.

2169: 

2170: To define the function $h$ we consider a (very thin) left half-collar $L$ of $V$ on 

2171: $\partial M$ and a right half-collar $R$ of $C$. Here ``left'' and ``right'' refer

2172: to the natural orientations of $\partial M$ and of $V\cup C$.

2173: Note that $L\subset B$ and $R\subset W$.

2174: Now we set $h\ristr{B\setminus L}\equiv 1/2$, and 

2175: $h\ristr{W\setminus R}\equiv 2$. Figures~\ref{blacktria} and~\ref{whitetria}

2176: \begin{figure}

2177: \centerline{\psfig{file=blacktria.eps,width=10cm}}

2178: \caption{\label{blacktria}Where $h=1/2$ the field points outwards }

2179: \end{figure}

2180: respectively show that away from $V\cup C$ indeed the pattern of $w_{\calt'}$ on

2181: \begin{figure}

2182: \centerline{\psfig{file=whitetria.eps,width=10cm}}

2183: \caption{\label{whitetria}Where $h=2$ the field points inwards}

2184: \end{figure}

2185: $\partial M_h$ is as required. Now we identify $L$ to $V\times[-1,0]$ and $R$ to

2186: $C\times[0,1]$, and we define $h(x,s)=f(s)$ for $(x,s)\in V\times[-1,0]$

2187: and $h(x,s)=f(s-1)$ for $(x,s)\in C\times[0,1]$, where $f:[-1,0]\to[1/2,2]$ is

2188: a smooth monotonic function with all the derivatives vanishing at $-1$ and $0$.

2189: Instead of describing $f$ explicitly we picture it and show that also near

2190: $V\cup C$ the pattern is as required. This is done near $V$ and $C$ respectively 

2191: in Figg.~\ref{convline}

2192: \begin{figure}

2193: \centerline{\psfig{file=convline.eps,width=15.5cm}}

2194: \caption{\label{convline}On $V$ the field has convex tangency}

2195: \end{figure}

2196: and~\ref{concline}.

2197: In both pictures we have only considered a special configuration for the 

2198: triangulation on $\partial M$, and we have refrained from picturing the orbits

2199: of the field in the 3-dimensional figure. Instead, we have separately shown

2200: the orbits on the vertical simplices on which the value of $h$ changes.

2201: 

2202: The conclusion is now exactly as in Turaev's argument, so we only give a sketch.

2203: The chosen representative $z$ of $\xi\in\eulc(M,\pp)$ can be described as an integer linear

2204: combination of orbits of $w_{\calt'}$, which we can describe as segments

2205: $[p_\sigma,p_{\sigma'}]$ with $\sigma\subset\sigma'$. Now we consider the chain

2206: \begin{equation}\label{complete:chain:to:collar}

2207: z'=z-\sum_{\sigma\subset W\cup V}\index(\sigma)\cdot p_\sigma\times[0,1].

2208: \end{equation}

2209: By definition of $h$ we have that $z'$ is a 1-chain in $M_h$, and 

2210: $\partial z'$ consists exactly of the singularities of $w_{\calt'}$ contained in $M_h$, 

2211: each with its index. For each segment $s$ which appear in $z'$

2212: we first modify $w_{\calt'}$ to a field which is ``constant''

2213: on a tube $T$ around $s$, and then we modify the field again

2214: within $T$, in a way which depends on the coefficient of $s$ in $z'$.

2215: The resulting field has the same singularities as $w_{\calt'}$, but 

2216: one checks that these singularities can be removed by 

2217: a further modification supported within small balls centred at the singular points.

2218: We define $\Psi(\xi)$ to be the class in $\euls(M,\pp)$ of this final field.

2219: Turaev's proof that $\Psi$ is indeed well-defined and $H_1(M;\mz)$-equivariant

2220: \begin{figure}

2221: \centerline{\psfig{file=concline.eps,width=15.5cm}}

2222: \caption{\label{concline}On $C$ the field has concave tangency}

2223: \end{figure}

2224: applies without essential modifications.\finedim{reconstruction:statement}

2225: 

2226: 

2227: \begin{rem}\label{good:cell:ok}

2228: {\em In the previous proof we have defined $\Psi$ using triangulations, in order to

2229: apply directly Turaev's technical results (in particular, invariance under subdivision).

2230: However the geometric construction makes sense also for cellularizations $\cc$ more

2231: general than triangulations, the key point being the possibility of defining a field

2232: $w_{\cc}$ satisfying the same properties as the field defined for triangulations.

2233: This is certainly true, for instance, for cellularizations $\cc$ of $M$ induced by

2234: realizations of $M$ by face-pairings on a finite number of polyhedra, assuming

2235: that the projection of each polyhedron to $M$ is smooth.}

2236: \end{rem}

2237: 

2238: \dim{diagram:commutes}

2239: To help the reader follow the details, we first outline the scheme of the proof:

2240: \begin{enumerate}

2241: \item By identifying $M$ to a collared copy of itself, we choose a representative $z$

2242: of the given $\xi\in\eulc(M,\pp)$ such that the extra terms added to define $\Thetac(\xi)$

2243: cancel with terms already appearing in $z$. (We know {\em a priori} that this happens

2244: at the level of boundaries, but it may well not happen at the level of $1$-chains.)

2245: \item We apply Remark~\ref{good:cell:ok} and choose a cellularization of $M$ in which

2246: it is particularly easy to analyze $\Psi(\xi)$ and $\Psi(\Thetac(\xi))$, both constructed

2247: using the representative $z$ already obtained.

2248: \end{enumerate}

2249: 

2250: 

2251: \noindent We consider a cellularization $\cc$ of

2252: $M$ satisfying the same assumptions on $\partial M$ as those considered in the proof of

2253: Proposition~\ref{p:h:formula}, namely $C\cup V$ is surrounded on both sides

2254: by a row of rectangular tiles, and the other tiles are triangular. 

2255: We denote by $\gamma_1,\dots,\gamma_n$ the segments in $C$, oriented as $C$.

2256: 

2257: 

2258: Let us consider a representative $z$ relative to $\cc$ of the given $\xi\in\eulc(M,\pp)$.

2259: We construct a new copy $M_1$ of $M$ by attaching $\partial M\times[-1,0]$ to

2260: $M$ along $\partial M=\partial M\times\{-1\}$, and we extend $\cc$ to $\cc_1$ by 

2261: taking the product cellularization on $\partial M\times[-1,0]$. We define a new chain as

2262: \begin{eqnarray*}

2263: z_1 & = & z+\sum_{\sigma\subset B}\index(\sigma)\cdot p_\sigma\times[-1/2,0]

2264: -\sum_{\sigma\subset W\cup V}\index(\sigma)\cdot p_\sigma\times[-1,-1/2]\\

2265: & & + \sum_{j=1}^n\Big(\gamma_j\ristr{[1/2,1]}\times\{-1/2\}-

2266: \gamma_j\ristr{[1/2,1]}\times\{0\}\Big).

2267: \end{eqnarray*}

2268: Note that $z_1$ is an Euler chain in $M_1$ with respect to $\cc_1$. Consider the natural

2269: homeomorphism $f:M\to M_1$ and the class

2270: $$a=\alphac(f_*(\xi),[z_1])\in H_1(M_1;\mz)$$

2271: which may or not be zero. Since the inclusion of $M$ into $M_1$ is an isomorphism

2272: at the $H_1$-level, $a$ can be represented by a $1$-chain in $M$, so $z_1$ can

2273: be replaced by a new Euler chain $z_2$ such that $[z_2]=f_*(\xi)$ and $z_2$

2274: differs from $z_1$ only on $M$.

2275: 

2276: Renaming $M_1$ by $M$ and $z_2$ by $z$ we have found a representative $z$ of $\xi$

2277: such that $z=z_\theta+\sum_{j=1}^n\gamma_j\ristr{[1/2,1]}$, where $z_\theta$ is a sum

2278: of simplices contained in $B\cup\interior M$. Note that of course $\Thetac(\xi)=[z_\theta]$.

2279: To conclude the proof we need now to analyze $\Psi(\xi)$, constructed using $z$, and

2280: $\Psi(\Thetac(\xi))$, constructed using $[z_\theta]$, and show that

2281: $\Psi(\Thetac(\xi))=\Thetas(\Psi(\xi))$. By construction $\Psi(\xi)$ and 

2282: $\Psi(\Thetac(\xi))$ will only differ near $C$, and we concentrate on one component

2283: of $C$ to show that the difference is exactly (up to homotopy) as in the definition of

2284: $\Thetas$, {\em i.e.}~as in Fig.~\ref{conc:to:conv}.

2285: 

2286: The difference between $\Psi(\xi)$ and $\Psi(\Thetac(\xi))$ is best visualized on

2287: a cross-section of the form $C\times[0,\infty)$. We leave to the reader to analyze

2288: the complete 3-dimensional pictures. To understand the cross-section, we 

2289: follow the various steps in the proof of Theorem~\ref{reconstruction:statement}.

2290: 

2291: The first step in the definition of $\Psi(\xi)$ (respectively, $\Psi(\Thetac(\xi))$)

2292: consists in choosing the height function $h$ (respectively, $h_\theta$)

2293: and replacing the chains $z$ (respectively, $z_\theta$)

2294: by a chain $z'$ (respectively, $z'_\theta$)

2295: as in formula~(\ref{complete:chain:to:collar}).

2296: This is done in Fig.~\ref{comm:diag:1}

2297: \begin{figure}

2298: \centerline{\psfig{file=comm_diag_1.eps,height=4cm}}

2299: \caption{\label{comm:diag:1}Local difference between $z'$ (left) and $z'_\theta$ (right)}

2300: \end{figure}

2301: where only the difference between the chains is shown.

2302: 

2303: To conclude we must modify the field $w_\cc$ within a small neighbourhood of 

2304: the support of $z'$ and $z'_\theta$. This is done in Figg.~\ref{comm:diag:2}

2305: \begin{figure}

2306: \centerline{\psfig{file=comm_diag_2.eps,height=4cm}}

2307: \caption{\label{comm:diag:2}Construction of $\Psi(\xi)$ on $C\times[0,\infty)$.

2308: On the left we show $w_\cc$ and the zones where it must be modified.}

2309: \end{figure}

2310: and~\ref{comm:diag:3} respectively. The rightmost picture in Fig.~\ref{comm:diag:3} 

2311: \begin{figure}

2312: \centerline{\psfig{file=comm_diag_3.eps,height=4cm}}

2313: \caption{\label{comm:diag:3}Construction of $\Psi(\Thetac(\xi))$ on $C\times[0,\infty)$}

2314: \end{figure}

2315: is obtained by homotopy on the previous one. The representatives of $\Psi(\xi)$ and

2316: $\Psi(\Thetac(\xi))$ can be compared directly, and indeed they differ by a curve

2317: parallel to $C$ and directed consistently with $C$, so 

2318: $\Psi(\Thetac(\xi))=\Thetas(\Psi(\xi))$.\finedim{diagram:commutes}

2319: 

2320: 

2321: We give now the proof omitted in Section~\ref{knots:section}.

2322: 

2323: \dim{wind:sensitive}

2324: Let us first note that the comparison class which we must show to be $[m]$ is independent

2325: of $f$ by Proposition~\ref{collar:trivial}. We

2326: will give two completely independent (but somewhat sketchy) proofs that this

2327: class is indeed $[m]$.

2328: 

2329: For a first proof, instead of comparing a ``straight'' knot with one with two curls, 

2330: we compare two knots with one curl, chosen so that the framing is the same but the 

2331: winding number is different. This is of course equivalent. The two knots are shown in

2332: Fig.~\ref{curls} as thick tubes, together with one specific orbit of the

2333: field they are immersed in. The resulting bicoloration on the boundary

2334: of the tubes is also outlined.

2335: \begin{figure}

2336: \centerline{\psfig{file=curls.eps,height=4cm}}

2337: \caption{\label{curls}Differently curled tubes in the vertical field.}

2338: \end{figure}

2339: To compare the curls we isotope the bicolorated tubes to the same

2340: straight tube, and we show how the orbit of the field is transformed under

2341: this isotopy. This is done in Fig.~\ref{nocurls}.

2342: \begin{figure}

2343: \centerline{\psfig{file=nocurls.eps,height=3.2cm}}

2344: \caption{\label{nocurls}Straightened curls.}

2345: \end{figure}

2346: Also from this very partial picture it is quite evident that the resulting fields

2347: wind in opposite directions around the tube. A more accurate picture

2348: would show that the difference is actually a meridian of the tube.

2349: 

2350: Another (indirect) proof goes as follows. Note first that the comparison class which we

2351: must compute certainly is a multiple of $[m]$, say $k\cdot[m]$. Note also that

2352: $k$ is independent of the ambient manifold $(M,v)$. Moreover, by symmetry, we

2353: will have $\alpha(\xi(v,K_0),f_*(\xi(v,K_{-1})))=-k\cdot[m]$ if $K_{-1}$ is

2354: obtained by locally adding a double curl with opposite winding number.

2355: 

2356: We take now $M$ to be $S^3$, with the field $v$ carried by the abalone $P$

2357: as in Section~\ref{exa:section}, and $K$ to be a trivial knot contained in the 

2358: ``smaller'' disc of the $P$. Using either the classical machinery of obstruction

2359: theory or the techniques developed in~\cite{second:paper}, one can see that there

2360: exists another pseudo-Legendrian knot $K'$ in $(S^3,v)$ such that 

2361: $\alpha(\xi(v,K),\xi(v,K'))=[m]$, where by simplicity we are omitting the

2362: framed-isotopies necessary to compare these classes. As already remarked in

2363: Section~\ref{spines:section}, we can assume that $K'$ has a $\cont^1$-projection

2364: on $P$. If one examines $P$ carefully one easily sees that $K'$ can actually be slid

2365: over $P$ to lie again in the small disc of $P$. Now $K'$ is a planar projection of the

2366: trivial knot, so through Reidemeister moves of types II and III,

2367: which correspond to isotopies through knots transversal to $v$, it can be transformed

2368: into a projection which differs from the trivial one only for a finite (even) number

2369: of curls. Summing up, we have a knot $K'$ such that $\alpha(\xi(v,K),\xi(v,K'))=[m]$

2370: and $K'$ differs from $K$ only for a finite number of transformations of the form

2371: $K\mapsto K_1$ or $K\mapsto K_{-1}$. This shows that $[m]$ is a multiple of 

2372: $k\cdot[m]$, so $k=\pm1$.\finedim{wind:sensitive}

2373: 

2374: We conclude the paper by establishing

2375: the only statement given in Section~\ref{spines:section} and

2376: not proved there. As above, we do not recall all the notation.

2377: 

2378: \dim{spider:structure:teo}

2379: We fix $P$ and set $s''=s''(P)$, $\hatv=\hatv(P)$.

2380: Using Remark~\ref{good:cell:ok} we see that the construction of

2381: $\Psi([s''])$ explained in the proof of Theorem~\ref{reconstruction:statement}

2382: can be directly applied to the cellularization $\hattt=\hattt(P)$ of $\hatM$.

2383: Recall that this construction requires identifying $\hatM$ to a collared copy

2384: of itself, and extending $s''$ to a chain $s'''$ whose boundary consists precisely

2385: of the singularities of a field $w$. A representative of $\Psi([s''])$ is then

2386: obtained by applying to $w$ a certain desingularization procedure.

2387: This desingularization is supported in a

2388: neighborhood of $s'''$, and one can easily check that

2389: the connected components of the support of $s'''$ (denoted henceforth by $S$)

2390: are actually contractible.

2391: Therefore, {\em any} desingularization of $w$ supported in a

2392: neighbourhood of $s'''$ will give a representative of $\hatv$.

2393: We will prove the desired formula $\Psi([s''])=[\hatv]$

2394: by exhibiting one such desingularization which is nowhere antipodal to $\hatv$.

2395: In our argument we will always neglect the contraction of $\stwotriv$ which maps

2396: $M$ onto $\hatM$. (The desired formula actually holds at the level of $M$, and it

2397: easily implies the formula for $\hatM$.) 

2398: 

2399: By the above observations, the following claims easily imply the conclusion of the proof:

2400: \begin{enumerate}

2401: \item\label{claim:antip} The set of points where $w$ is antipodal to 

2402: $\hatv$ is contained in $S$.

2403: \item\label{claim:desing} If $S_0$ is a component of

2404: $S$ then $w$ can be desingularized within a neighbourhood of

2405: $S_0$ to a field which is not antipodal to $\hatv$ in the neighbourhood.

2406: \end{enumerate}

2407: \noindent We prove claim~\ref{claim:antip} by first noting that the cells dual

2408: to those of $P$ are unions of orbits of both $w$ and $\hatv$. So we can analyze cells

2409: separately. We do this explicitly only for 2-dimensional cells, leaving to 

2410: the reader the other cases. In Fig.~\ref{hexagon:1}

2411: \begin{figure}

2412: \centerline{\psfig{file=hexagon_1.eps,width=3.5cm}}

2413: \caption{\label{hexagon:1}The field $\hatv$ on a hexagon}

2414: \end{figure}

2415: we describe $\hatv$. In the left-hand side of Fig.~\ref{hexagon:2}

2416: \begin{figure}

2417: \centerline{\psfig{file=hexagon_2.eps,width=12cm}}

2418: \caption{\label{hexagon:2}The field $w$ and the trace of $S$ on a hexagon}

2419: \end{figure}

2420: we describe $w$ on the collared hexagon. In the right-hand side of the same figure

2421: we only show the singularities of $w$ on the renormalized hexagon,

2422: and the intersection of $S$ with the hexagon. In this figure the 7 short segments come

2423: from $s'''-s''$; the other bits of $S$ have been labeled by `Or', `St', `Ba' or `He'

2424: to indicate that they come from

2425: orbits of $\hatv$, stars, bi-arrows or half-edges.

2426: 

2427: This proves claim~\ref{claim:antip}. Comparing Fig.~\ref{hexagon:2}

2428: with Fig.~\ref{hexagon:1}, and carrying out the same analysis for

2429: 3-cells, one actually shows also claim~\ref{claim:desing} for 

2430: components $S_0$ coming from $s'''-s''$. Components of $S$ other than these can

2431: be described in one of the following ways:

2432: \begin{itemize}

2433: \item[(a)] an orbit of $\hatv$ emanating from a vertex of $P$;

2434: \item[(b)] a half-edge of $C$;

2435: \item[(c)] a bi-arrow together with an orbit of $\hatv$ emanating from

2436: the centre of an edge of $P$ and reaching the centre of the bi-arrow;

2437: \item[(d)] a star together with an orbit of $\hatv$ emanating from

2438: the centre of a disc of $P$ and reaching the centre of the star.

2439: \end{itemize}

2440: \noindent All cases can be treated with the same method, we only do case (c). 

2441: Figure~\ref{fix:biarrow}

2442: \begin{figure}

2443: \centerline{\psfig{file=fix_bia.eps,width=8.5cm}}

2444: \caption{\label{fix:biarrow}An enhanced bi-arrow and the field $w$ near it}

2445: \end{figure}

2446: shows the component placed so that $\hatv$ can be thought of as the 

2447: constant vertical field pointing upwards, and the field $w$ near the component.

2448: The conclusion easily follows.

2449: \finedim{spider:structure:teo}

2450: 

2451: 

2452: \begin{thebibliography}{99}

2453: 

2454: 

2455: 

2456: 

2457: \bibitem[1]{manuscripta} {\sc R.~Benedetti, C.~Petronio}, {\it A finite

2458: graphic calculus  for $3$-manifolds}, Manuscripta Math. {\bf 88} (1995),

2459: 291-310.

2460: 

2461: \bibitem[2]{lnm}  {\sc R.~Benedetti, C.~Petronio}, ``Branched Standard

2462: Spines of 3-Manifolds'', Lecture Notes in Math. n. 1653, Springer-Verlag,

2463: Berlin-Heidelberg-New York, 1997.

2464: 

2465: \bibitem[3]{contspin} {\sc R.~Benedetti, C.~Petronio},  {\it Branched spines

2466: and contact structures on $3$-manifolds}, Ann. Mat. Pura Appl. (to appear).

2467: 

2468: \bibitem[4]{second:paper} {\sc R.~Benedetti, C.~Petronio},  

2469: {\it Combed 3-manifolds with concave boundary,

2470: framed links, and pseudo-Legendrian links},

2471: preprint January 2000, {\tt math.GT/0001162 }, submitted.

2472: 

2473: \bibitem[5]{third:paper}  {\sc R.~Benedetti, C.~Petronio},  

2474: {\it Reidemeister torsion via branched spines}, (in preparation).

2475: 

2476: \bibitem[6]{fourth:paper} {\sc R.~Benedetti, C.~Petronio},  {\it

2477: Euler structures with generic boundary and torsion of

2478: Reidemeister type}, (in preparation).

2479: 

2480: \bibitem[8]{casler} {\sc B.~G.~Casler}, {\it An imbedding theorem for

2481: connected $3$-manifolds with boundary}, Proc. Amer. Math. Soc. {\bf 16} (1965),

2482: 559-566.

2483: 

2484: \bibitem[9]{christy} {\sc J.~Christy}, {\it Branched surfaces and attractors

2485: I}, Trans. Amer. Math. Soc. {\bf 336} (1993), 759-784.

2486: 

2487: \bibitem[11]{fist} {\sc R.~Fintushel, R. Stern}, {\it Knots, links,

2488: and $4$-manifolds}, Invent. Math. {\bf 134} (1998), 363-400.

2489: 

2490: \bibitem[12]{gr:1} {\sc D.~Gillman, D.~Rolfsen}, {\it The Zeeman conjecture

2491: is equivalent to the Poincar\'e conjecture}, Topology {\bf 22} (1983), 315-323.

2492: 

2493: \bibitem[13]{hatcher} {\sc A.~Hatcher, D.~McCullough}, {\it Finiteness

2494: of classifying spaces of relative diffeomorphism groups of $3$-manifolds}, 

2495: Geometry and Topology {\bf 1} (1997), 91-109.

2496: 

2497: \bibitem[14]{ishii} {\sc I.~Ishii}, {\it Moves for flow-spines and

2498: topological invariants of $3$-manifolds}, Tokyo J. Math. {\bf 15} (1992),

2499: 297-312.

2500: 

2501: \bibitem[15]{lemuoh}{\sc T.~T.~Q.~Le, J.~Murakami, T.~Ohtsuki},

2502: {\it On a universal perturbative invariant of $3$-manifolds},

2503: Topology {\bf 37} (1998), 359-374.

2504: 

2505: \bibitem[16]{lescop}{\sc C.~Lescop},``Global Surgery Formula for the Casson-Walker invariant'',

2506: Ann. of Math. Studies n. 140, Princeton University Press, Princeton, NJ, 1996.

2507: 

2508: \bibitem[18]{mafo} {\sc S.~V.~Matveev, A.~T.~Fomenko},

2509: {\it Constant energy surfaces of Hamiltonian systems, enumeration

2510: of three-dimensional manifolds in increasing order of complexity,

2511: and computation of volumes of closed hyperbolic manifolds},

2512: Russ. Math. Surv. {\bf 43} (1988), 3-25.

2513: 

2514: \bibitem[19]{matv:mossa} {\sc S.~V.~Matveev}, {\it Transformations of special

2515: spines and the Zeeman conjecture}, Math. USSR-Izv. {\bf 31} (1988), 423-434.

2516: 

2517: \bibitem[20]{matv:new}

2518: {\sc S.~V.~Matveev}, ``Algorithmic Methods in

2519: 3-Manifold Topology'', in preparation.

2520: 

2521: \bibitem[21]{meng:taub}{\sc G.~Meng, C.~H.~Taubes},

2522: {\it $\underline{SW}\,=\,$Milnor torsion}, Math. Res. Lett. {\bf 3} (1996), 661-674

2523: 

2524: \bibitem[22]{milnor}{\sc J.~Milnor}, {\it Whitehead torsion}, Bull. Amer. Math. Soc

2525: {\bf 72} (1966), 358-426.

2526: 

2527: \bibitem[23]{tesi} {\sc C.~Petronio}, ``Standard Spines and 3-Manifolds'', Scuola

2528: Normale Superiore, Pisa, 1995.

2529: 

2530: \bibitem[24]{piergallini} {\sc R.~Piergallini}, {\it Standard moves for  

2531: standard polyhedra and spines}, Rendiconti Circ. Mat. Palermo {\bf 37}, suppl.

2532: 18 (1988), 391-414.

2533: 

2534: \bibitem[25]{porti} {\sc J.~Porti},  ``Torsion de Reidemeister pour les

2535: Vari\'et\'es Hyperboliques'', Memoirs n. 612, Amer. Math. Soc., Providence, RI,

2536: 1997.

2537: 

2538: \bibitem[26]{trace} {\sc B.~Trace}, {\it On the Reidemeister

2539: moves of a classical knot}, Proc. Amer. Math. Soc. {\bf 89} (1983), 722-724.

2540: 

2541: \bibitem[27]{turaev:Reidemeister} {\sc V.~G.~Turaev}, {\it Reidemeister

2542: torsion in knot theory}, Russ. Math. Surv. {\bf 41} (1986), 119-182.

2543: 

2544: \bibitem[28]{turaev:Euler} {\sc V.~G.~Turaev}, {\it Euler structures, nonsingular

2545: vector fields, and torsion of Reidemeister type}, Math. USSR-Izv. {\bf 34}

2546: (1990), 627-662.

2547: 

2548: \bibitem[29]{turaev:spinc} {\sc V.~G.~Turaev}, {\it Torsion invariants of

2549: Spin$^{\it c}$-structures on $3$-manifolds}, Math. Res. Lett. {\bf 4} (1997), 679-695.

2550: 

2551: \bibitem[30]{turaev:nuovo} {\sc V.~G.~Turaev}, {\it A combinatorial

2552: formulation for Seiberg-Witten invariants of $3$-manifolds},

2553: Math. Res. Lett. {\bf 5} (1998), 583-598.

2554: 

2555: 

2556: 

2557: \end{thebibliography}

2558:  

2559: 

2560: \vspace{1cm}

2561: 

2562: 

2563: \hspace{8cm} benedett@dm.unipi.it

2564: 

2565: \hspace{8cm} petronio@dm.unipi.it

2566: 

2567: \hspace{8cm} Dipartimento di Matematica

2568: 

2569: \hspace{8cm} Via F.~Buonarroti, 2

2570: 

2571: \hspace{8cm} I-56127, PISA (Italy)

2572: 

2573: \end{document}

2574: 

2575: