1:  

2: %Version of Jan. 9, 2001

3: 

4: 

5: 

6: 

7: \documentclass[10pt]{article} 

8:  

9: \usepackage{amsmath} 

10: \usepackage{amsthm} 

11: \usepackage{amssymb} 

12: \usepackage{amsfonts} 

13: \usepackage{epsfig}

14:  

15: \input xypic 

16:  

17:  

18: \relpenalty=10000 

19: \binoppenalty=10000 

20: \tolerance=50000 

21:  

22:  

23: \newtheorem{The}{Theorem}[section] 

24: \newtheorem{Cor}[The]{Corollary} 

25: \newtheorem{Pro}[The]{Proposition} 

26: \newtheorem{Exa}[The]{Example} 

27: \newtheorem{Rem}[The]{Remark} 

28: \newtheorem{Lem}[The]{Lemma} 

29: \newtheorem{Def}[The]{Definition} 

30: \newtheorem{Con}[The]{Conjecture} 

31:  

32: \def\proof{\vspace{2ex}\noindent{\bf Proof.} } 

33: 

34: \def\endproof{\relax\ifmmode\expandafter\endproofmath\else 

35: \unskip\nobreak\hfil\penalty50\hskip.75em\hbox{}\nobreak\hfil\bull 

36: {\parfillskip=0pt \finalhyphendemerits=0 \bigbreak}\fi} 

37: \def\endproofmath$${\eqno\bull$$\bigbreak} 

38: \def\bull{\vbox{\hrule\hbox{\vrule\kern3pt\vbox{\kern6pt}\kern3pt\vrule}\hrule}} 

39:  

40: \def\cancel#1#2{\ooalign{$\hfil#1\mkern1mu/\hfil$\crcr$#1#2$}}

41: 

42: \def\dirac{\mathpalette\cancel\partial} 

43: \def\Dirac{\mathpalette\cancel D}

44:  

45:  

46: \newcommand{\C}{{\mathbb C}} 

47: \newcommand{\R}{{\mathbb R}} 

48: \newcommand{\Z}{{\mathbb Z}} 

49: \newcommand{\Q}{{\mathbb Q}} 

50: \renewcommand{\P}{{\mathbb P}}

51: \newcommand{\AAA}{{\mathbb A}} 

52: \newcommand{\A}{{\cal A}} 

53: \newcommand{\E}{{\cal E}} 

54: \newcommand{\G}{{\cal G}} 

55: \newcommand{\M}{{\cal M}} 

56: \newcommand{\B}{{\cal B}} 

57:  

58: \newcommand{\ind}{\mathfrak{i}} 

59: \renewcommand{\j}{\mathfrak{j}} 

60:  

61: \newcommand{\la}{\langle} 

62: \newcommand{\ra}{\rangle} 

63: \newcommand{\ba}{\begin{eqnarray}} 

64: \newcommand{\na}{\end{eqnarray}} 

65: \newcommand{\beq}{\begin{equation}} 

66: \newcommand{\eeq}{\end{equation}}  

67: \newcommand{\s}{\mathfrak{s}} 

68: \renewcommand{\t}{\mathfrak{t}} 

69: \newcommand{\spinc}{\mathrm{Spin}^c} 

70:  

71:  

72: \title{Exact triangles in monopole homology and the Casson-Walker invariant}  

73: \author{Matilde Marcolli and Bai-Ling Wang} 

74: \date{}

75:  

76: \begin{document} 

77: \maketitle 

78:  

79: \tableofcontents

80:  

81: \section{Introduction} 

82: 

83: The purpose of this paper is to give a general outline of the problem

84: of the exact triangles in Seiberg--Witten--Floer theory. We present

85: here the most general case, where the problem consists of producing a

86: surgery formula relating the monopole homology of a compact oriented

87: 3--manifold $Y$ with an embedded knot $K$, and the monopole homologies

88: of some 3--manifolds obtained by Dehn surgery on $K$. 

89: 

90: In the series of papers \cite{CMW} \cite{MW2} \cite{MW3} \cite{MW4},

91: we studied the problem in the case of an integral homology 

92: 3-sphere $Y$, and the 3--manifolds $Y_1$  

93: and $Y_0$ obtained by Dehn surgery on $K$ with framing $1$ and $0$,

94: respectively. 

95: 

96: The results of \cite{CMW} \cite{MW2} \cite{MW3} \cite{MW4} are, at

97: this stage, still to be considered as ``work in progress'', where some

98: of the proofs need more rigorous presentations. The main

99: result of that series of papers is that the

100: Seiberg-Witten-Floer homologies of $Y$ and of the manifolds $Y_1$   

101: and $Y_0$ are related by an exact triangle

102: {\small

103: \[

104: \diagram

105: HF^{SW}_{*} (Y_1, g_1) \rto^{w^1_{*}}

106:  & HF^{SW}_{*}(Y, g_0, \mu)

107: \dlto_{w^0_{*}} \\

108:  \bigoplus_{k\in \Z} HF^{SW}_{(*)}(Y_0, \s_k) \uto_{w_{(*)}}

109: & \\

110: \enddiagram

111: \] }

112: In this triangle the maps $w_*^1, w_*^0$ and $w_{(*)}$ are induced 

113: by the surgery cobordisms connecting $Y_1$ and $Y$,

114: $Y$ and $Y_0$, $Y_0$ and $Y_1$, respectively, and $\mu$ is the surgery

115: perturbation that simulates the effect of Dehn surgery. 

116: 

117: In this paper, we explain how, following the same strategy for the

118: proof of the surgery formula which we have introduced in the previous

119: work, we may be able to extend this exact triangle to the more

120: general case of any closed oriented 3-manifold with a smoothly

121: embedded knot. 

122: 

123: In the last section of this paper we give a topological application of

124: the kind of arguments that lead to the proof of the ``geometric

125: triangles'', namely the surgery formula for monopole homology viewed

126: at the level of generators. We show that a modified

127: version of the Seiberg-Witten invariant agrees with the Casson-Walker

128: invariant, for any closed and oriented rational homology 3-sphere.

129: 

130: Let Y be a  closed oriented 3-manifold, 

131: with a smooth embedded knot $K$, let $\nu(K)$

132: be the  tubular neighbourhood of $K$ in $Y$.

133: Choose an identification of $\nu(K)$ with 

134: $D^2\times S^1$:

135: \ba

136: \nu(K) \cong D^2\times S^1,

137: \label{frame}

138: \na 

139: where $K$ is mapped to the core of the

140: solid torus $D^2\times S^1$. Under the identification (\ref{frame}), 

141: on the boundary $T^2$, we fix a basis $m, l$ of $H_1(T^2, \Z)$ such that

142: $l$ is the longitude (parallel to $K$ under the identification (\ref{frame})) 

143: and $m$ is the right-handed meridian (intersecting  $l$ once), the orientation

144: determined by $m\wedge l$ coincides with

145: the orientation induced from $Y$. The corresponding

146: longitude and meridian in the knot complement

147: $ V = Y\backslash \nu (K) $

148: are denoted by $m', l'$ respectively. 

149:  Similarly, let $m''$ and $l''$ be the meridian

150: and longitude in the tubular neighbourhood of the knot $\nu(K)$. The

151: meridian $m''$ bounds a disk $D^2$ in $\nu(K)$, and $l''$ generates

152: $H_1(\nu(K),\Z)$ and parallels to $K$. Let $p$ and $q$ be two relatively prime

153: integers, the Dehn surgery with coefficient $p/q\in \Q\cup \{\infty\}$

154: on $K$ is the operation of removing $\nu(K)$ and gluing in $D^2\times S^1$ 

155: by an orientation reversing diffeomorphism $f_{p/q}$

156: of $T^2$ that satisfies

157: \[

158: f_{p/q} (m'') = pm'-ql'.

159: \]

160: The resulting manifold is denoted by $Y_{p/q}$. Note that in general 

161: $Y_{p/q}$ depends on the choice of the identification (\ref{frame}).

162: 

163: Let $\s$ be  a $\spinc$ structure  on $Y$. We shall see that, with a

164: suitable choice of metrics and perturbation, $(Y, \s)$ has

165: non-empty monopole moduli space only if $\s|_{\nu(K)}$ 

166: has trivial determinant, hence we shall always assume that

167: the   $\spinc$ structure $\s$ is trivial around $K$.

168: If $K$ represents a trivial homology class 

169: in $H_1(Y, \Z)$, then there is only one $\spinc$ structure  on 

170: $Y$ which agrees with $\s$ over $Y-\nu(K)$ and $\nu(K)$. 

171: Suppose that $K$ represents a torsion element of order $n$ in 

172: $H_1(Y, \Z)$, which means,

173: \ba

174: \label{order:n}

175: \displaystyle{

176: \frac {|\hbox{Torsion}( H_1(Y, \Z))|}{|\hbox{Torsion}( H_1(Y-\nu(K),  \Z))|}}

177: =n.

178: \na

179: In other words, $n$ is the minimal positive integer such 

180: that $n[K]$ is homologous to zero in $H_1(Y, \Z)$. 

181: Then there exists a $\Z_n$-family of $\spinc$ structures 

182: $

183: \s\otimes L_k (k=1, \cdots, n)

184: $

185: which agree with $\s$ over $Y-\nu(K)$ and $\nu(K)$, where 

186: $L_k$ is a complex line bundle whose Euler

187: class is given by $kPD([K])$. If $[K] \neq 0$ in $H_1(Y, \Q)$,

188: then  there exists a $\Z$-family of $\spinc$ structures 

189: $

190: \s\otimes L_k (k\in \Z)

191: $

192: which agree with $\s$ over $Y-\nu(K)$ and $\nu(K)$, where 

193: $L_k$ is a complex line bundle whose Euler

194: class is given by $kPD([K])$.

195: 

196: Let $Y_1$ be the $(+1)$-surgery on $K$, 

197: and $Y_0$ be the 0-surgery on $K$, 

198: we can consider separately the following cases.

199: 

200: 

201: 

202: \noindent $\bullet$ 

203: First we assume that $Y$ is a rational homology 3-sphere with

204: a smoothly embedded knot $K$ representing

205: a torsion element of order $n$ in $H_1(Y, \Z)$ in the sense of

206: (\ref{order:n}). Then $Y_1$ is a rational homology

207: 3-sphere, and $Y_0$ is a rational homology $S^1\times S^2$.

208: Let $\s$ be a $\spinc$ structure on $Y$, which is

209: trivial on $\nu(K)$. Gluing

210: the $\spinc$ structures $s|_{Y-\nu(K)}$ and $\s|_{\nu(K)}$

211: along $T^2$ via different gauge transformations

212: on $T^2$ results in

213: a $\Z_n$-family of $\spinc$ structures on $Y$ and $Y_1$,

214: and a $\Z$-family of $\spinc$ structures on $Y_0$. 

215: Without confusion, thinking $K$ as the core of 

216: the attaching solid torus $\nu(K)$,  we denote these structures

217: on $Y, Y_1$ and $Y_0$ all by

218: $ \s\otimes L_k (k\in Z),$

219: where $L_k$ is a complex line bundle of Euler

220: class $kPD([K])$.\, With this notation, it is understood

221: that for $Y$ and $Y_1$, and with $n[K]=0$, there is only a

222: $\Z_n$-family of $\spinc$ structures. 

223: Then, with a careful choice of metrics and perturbations

224: on $Y, Y_1$ and $Y_0$ as in Part I \cite{CMW}, we will

225: obtain the following exact triangles for the Seiberg-Witten-Floer

226: homologies: 

227: {\small

228: \[

229: \diagram 

230:  HF^{SW}_{*} (Y_1, \s\otimes L_m, g_1) \rto^{w^1_{*}}  

231:  & HF^{SW}_{*}(Y,\s\otimes L_m, g, \mu) 

232: \dlto_{w^0_{*}} \\  

233:  \bigoplus_{k\in \Z} HF^{SW}_{(*)}(Y_0, \s\otimes L_{nk+p}) \uto_{w_*}  

234: & \\ 

235: \enddiagram 

236: \] }

237: which holds for any fixed choice of $m$ and $p$ in $\{ 0,\ldots, n-1

238: \}$, and for a corresponding choice of perturbation, we have 

239: {\small

240: \[

241: \diagram 

242: \bigoplus_{k=1}^n HF^{SW}_{*} (Y_1, \s\otimes L_k, g_1) \rto^{w^1_{*}}  

243:  & \bigoplus_{k=1}^n HF^{SW}_{*}(Y,\s\otimes L_k, g, \mu) 

244: \dlto_{w^0_{*}} \\  

245:  \bigoplus_{k\in \Z} HF^{SW}_{(*)}(Y_0, \s\otimes L_k) \uto_{w_*}  

246: & \\ 

247: \enddiagram 

248: \] }

249: In both cases the homomorphisms $w^1_{*}, w^0_{*}$ and 

250: $w_*$ are induced by the surgery cobordisms.

251: These exact triangles generalize the results of 

252: Part I-IV \cite{CMW} \cite{MW2}\cite{MW3} \cite{MW4},

253: in the sense that the above exact triangle  reduces to the

254: exact triangle for an integral homology 3-sphere

255: when $n=1$.

256: 

257: \noindent $\bullet$ 

258: Now we assume that $(Y, \s)$ is a closed oriented 

259: 3-manifold of $b_1(Y)>0$ and with a $\spinc$ structure $\s$.

260:  Let $K$ be a knot representing

261: a torsion homology class in $H_1(Y, \Z)$ of order $n$

262: in the sense of (\ref{order:n}).

263: If $\s$ has  non-trivial determinant in the sense

264: that $c_1(\s) \neq 0$ in $H^2(Y, \Q)$, then

265: the Seiberg-Witten-Floer homology

266: for $(Y, \s)$ is $\Z_{2\ell}$-graded, where

267: $2\ell$ is the multiplicity of $c_1(\s)$ in

268: $H^2(Y, \Z)/\hbox{Torsion}$, i.e,

269: $c_1(\s) (H_2(Y, \Z)/\hbox{Torsion}) = 2\ell$. The gluing of the

270: $\spinc$ structures corresponding to these three different

271: surgeries gives rise to a $\Z_n$-family of $\spinc$

272: structures on $Y, Y_1$ and a $\Z$-family of

273: $\spinc$ structures on $Y_0$, which

274: we denote by 

275: $ \s\otimes L_k \qquad (k\in \Z),$

276: with the convention as before. Then, for any

277: $\spinc$ structure  $\s\otimes L_k$ ($k=1, \cdots, n)$ on $Y$

278: and $Y_1$, the corresponding Seiberg-Witten-Floer

279: homologies are all $\Z_{2\ell}$-graded, while 

280: for any $\spinc$ structure  $\s\otimes L_k$ ($k\in \Z$) on $Y_0$,

281: the Seiberg-Witten-Floer

282: homology $HF^{SW}_* (Y_0, \s\otimes L_k)$ is 

283: $\Z_{\ell_{[k]}}$-graded, where $\ell_{[k]}$ is the greatest common

284: factor in $2\ell$ and $2k$. Similar to Part IV \cite{MW4},

285: the $\Z_{\ell_{[k]}}$-graded homology $HF^{SW}_* (Y_0, \s\otimes L_k)$

286: can be lifted to a $\Z_{\ell}$-graded homology 

287: $HF^{SW}_{(*)} (Y_0, \s\otimes L_k)$. For this case, we derive in this 

288: paper the following exact triangles:

289: {\small

290: \ba

291: \diagram 

292:  HF^{SW}_{*} (Y_1, \s\otimes L_m, g_1) \rto^{w^1_{*}}  

293: & HF^{SW}_{*}(Y, \s\otimes L_m, g, \mu) 

294: \dlto_{w^0_{*}} \\  

295:  \bigoplus_{k\in \Z} HF^{SW}_{(*)}(Y_0, \s\otimes L_{nk+p}) \uto_{w_{(*)}}.  

296: & \\ 

297: \enddiagram 

298: \label{b_1>0:n:non-torsion:mp}

299: \na }

300: which holds for any choice of $m$ and $p$ in $\{ 0,\ldots, n-1 \}$,

301: and for a corresponding choice of perturbation, we have 

302: {\small

303: \ba

304: \diagram 

305: \bigoplus_{k=1}^n HF^{SW}_{*} (Y_1, \s\otimes L_k, g_1) \rto^{w^1_{*}}  

306: & \bigoplus_{k=1}^n HF^{SW}_{*}(Y, \s\otimes L_k, g, \mu) 

307: \dlto_{w^0_{*}} \\  

308:  \bigoplus_{k\in \Z} HF^{SW}_{(*)}(Y_0, \s\otimes L_k) \uto_{w_{(*)}}.  

309: & \\ 

310: \enddiagram 

311: \label{b_1>0:n:non-torsion}

312: \na }

313: 

314: \noindent $\bullet$ 

315: If $\s$ is a torsion $\spinc$ structure on $Y$, then for any

316: $k=1, \cdots, n$, 

317: $HF^{SW}_{*}(Y, \s\otimes L_k, \Z[[t]])$

318: and $HF^{SW}_{*}(Y_1, \s\otimes L_k, \Z[[t]])$ are $\Z$-graded

319: with $\Z[[t]]$-coefficient. The reduced

320: versions $HF^{SW}_{*}(Y, \s\otimes L_k)$

321: and $HF^{SW}_{*}(Y_1, \s\otimes L_k)$ are obtained by setting

322: $t=0$. The Seiberg-Witten-Floer homology 

323: $HF^{SW}_*(Y_0, \s\otimes L_k)$ is $\Z_{2k}$-graded, and

324: can be lifted to a $\Z$-graded version, denoted by

325: $HF^{SW}_{(*)}(Y_0, \s\otimes L_k)$. Then the

326: exact triangles take the same form as (\ref{b_1>0:n:non-torsion:mp})

327: and (\ref{b_1>0:n:non-torsion}).

328: 

329: \noindent $\bullet$ 

330: The remaining case is when a smoothly embedded

331: knot $K$ represents a non-trivial 

332: homology class in $H_1(Y, \Q)$. Let $n$ be the 

333: minimal positive integer such that there is

334: a  2-cycle in $H_2(Y, \Z)$  intersecting

335: $K$ at $n$ points.  

336: Then the $0$-surgery on $Y$ along $K$ yields $Y_0$

337: satisfying $b_1(Y_0) = b_1(Y) -1$. More precisely,

338: $H_1(Y_0, \Z)$ is obtained

339: by replacing the $\Z$-component $\Z \la [K]\ra$ of $H_1(Y, \Z)$

340: by $\Z_n \la [m'']= [l']\ra$. Notice that 

341: $Y$ can be thought of as the  manifold obtained by

342: $0$-surgery on $m'\subset Y-\nu(K) \subset Y_0$,

343: and $Y_1$ can be thought as the result

344: of $(+1)$-surgery on

345: $m'\subset Y-\nu(K) \subset Y_0$. Since $m'$ represents

346: a torsion element of order $n$ in $H_1(Y_0, \Z)$ in the

347: sense of (\ref{order:n}), we have 

348: the exact triangles for $(Y, \s, K)$ obtained from

349: the corresponding exact triangles for $(Y_0, \s, m')$

350: in the form of (\ref{b_1>0:n:non-torsion:mp})

351: and (\ref{b_1>0:n:non-torsion}). Thus,

352: it is enough to establish the exact triangle 

353: for a general closed oriented 3-manifold 

354: $Y$ with a smoothly embedded knot representing 

355: a torsion element of order $n$ in $H_1(Y, \Z)$. 

356: 

357: 

358: 

359: We now summarize the main theorem of this paper.

360: 

361: \begin{The}\label{main:theorem}

362: Let $(Y, \s)$ be a closed oriented 3-manifold  with

363: a $\spinc$ structure which is trivial around a

364: smoothly embedded knot $K$. Assume that $K$ represents

365: a torsion element of order $n$ in $H_1(Y, \Z)$ in the sense

366: of (\ref{order:n}). Assume that the

367: canonical framing of $K$ is given by

368: an identification of $D^2\times S^1$ with the 

369: tubular neighbourhood $\nu(K)$ such that $K$ is given

370: by $\{0\} \times S^1$. Here the parallel simple curve

371: on $T^2$ provides  the longitude of $K$ and the right handed meridian

372: is given by $\partial (D^2) \times \{ pt\}$. The orientation

373: determined by $m\wedge l$ coincides with the orientation induced  from $Y$.

374: Let $Y_1$ and $Y_0$ be the manifolds obtained, respectively,

375: by $(+1)$ and $0$ surgery along $K$ in $Y$.

376: With a careful choice of metrics and perturbations, we obtain the 

377: following exact triangle induced

378: by the surgery cobordisms after possible grading shifts:

379: {\small

380: \[

381: \diagram 

382: \bigoplus_{k=1}^n HF^{SW}_{*} (Y_1, \s\otimes L_k) \rto^{w^1_{*}}  

383:  & \bigoplus_{k=1}^nHF^{SW}_{*}(Y, \s\otimes L_k) 

384: \dlto_{w^0_{*}} \\  

385:  \bigoplus_{k\in \Z} HF^{SW}_{(*)}(Y_0, \s\otimes L_k) \uto_{w_{(*)}}  

386: & \\ 

387: \enddiagram   

388: \] }

389: Here $\s\otimes L_k$ is the $\spinc$ structure obtained

390: by tensoring a $\spinc$ structure $\s$ with a complex line

391: bundle $L_k$ of Euler class $kPD([K])$. Moreover, for any

392: fixed $m, p\in \{0, \cdots, n-1\}$, we have the

393: following more refined version of the exact triangle:

394: {\small

395: \[

396: \diagram

397: HF^{SW}_{*} (Y_1, \s\otimes L_m) \rto^{w^1_{*}}

398:  & HF^{SW}_{*}(Y, \s\otimes L_m)

399: \dlto_{w^0_{*}} \\

400:  \bigoplus_{k\in \Z} HF^{SW}_{(*)}(Y_0, \s\otimes L_{nk+p}) \uto_{w_{(*)}}

401: & \\

402: \enddiagram

403: \] }

404: Here again the maps are induced by the surgery cobordisms, possibly

405: after a shift in the grading.

406: \end{The}

407: 

408: The major technical steps required in the proof are described in

409: \cite{CMW}\cite{MW2}\cite{MW3}\cite{MW4}.  Thus, in this paper,

410: we shall address only those issues that are relevant to 

411: this general exact triangle, while we refer to the previous papers for the

412: general setting and results. 

413: 

414: In the last section of the paper, we show that a suitably modified

415: version of the Seiberg--Witten invariant of a rational homology

416: 3-sphere agrees with the Casson--Walker invariant.

417: For any rational homology 3-sphere $(Y, \s, g)$ with

418: a $\spinc$ structure $\s$ and a Riemannian metric, 

419: the counting of the irreducible Seiberg-Witten

420: monopoles defines the Seiberg-Witten invariant

421: \[

422: SW_Y(\s, g) = \# \bigl(\M_Y^*(\s, g)\bigr),

423: \]

424: where each irreducible  monopole in $\M_Y^*(\s, g)$ has

425: a natural orientation from the linearization of the Seiberg-Witten

426: equations. As studied in \cite{MW1}, $SW_Y(\s, g)$ depends on the metric and

427: perturbation used in the definition, in order to obtain a topological

428: invariant, we can modify $SW_Y(\s, g)$ by a metric and perturbation

429: dependent correction term as follows. Choose any four manifold $X$

430: with boundary $Y$, such  that $X$ is endowed with a 

431: cylindrical-end metric modeled on $(Y, g_Y)$. Choose

432: a $\spinc$ structure $\s_X$ on $X$ which agrees with

433: $\s$ on $Y$ over the end, and choose a connection $A$

434: on $(X, \s_X)$ which extends the unique reducible

435: $\theta_{\s}$ on $(Y, \s)$. Then we set 

436: \ba

437: \xi_Y(\s, g) = Ind_\C (\Dirac^X_A) -\displaystyle{\frac{1}{8} }

438: \bigl( c_1(\s_X)^2 -\sigma (X)\bigr),

439: \label{correction:term}

440: \na

441: where $Ind_\C (\Dirac^X_A)$ is the complex index of the

442: Dirac operator on $(X, \s_X)$ twisted with

443: the extending $\spinc$ connection $A$ and $\sigma (X)$ is

444: the signature of $X$. By the Atiyah-Patodi-Singer index theorem,

445: $\xi_Y(\s, g)$ is independent of the choice of

446: $(X, \s_X)$ and $A$, actually, $\xi_Y(\s, g)$ can be expressed as a

447: combination of the Atiyah-Patodi-Singer eta invariants for the

448: Dirac operator and signature operator on $(Y, \s)$:

449: \[

450: \xi_Y(\s, g) = \displaystyle{-

451: \frac 14 \eta^{\dirac_{\theta_{\s}}}_Y (0) -  

452: \frac 18 \eta^{sign}_Y(0).}

453: \]

454: 

455: The modified version of the  Seiberg-Witten invariant is defined

456: as

457: \ba

458: \hat{SW}_Y(\s) = SW_Y (\s, g) - \xi_Y(\s, g).

459: \label{hat:SW}

460: \na

461: Then we prove the following equivalence between 

462: $\hat{SW}_Y$ and the Casson-Walker invariant.

463: 

464: \begin{The}

465: Let $Y$ be a rational homology 3-sphere. Then,

466: \[

467: \sum_{\s\in \spinc (Y)} \hat{SW}_Y(\s) = \displaystyle{\frac{1}{2}}

468: |H_1(Y, \Z)| \lambda (Y),

469: \]

470: where $\lambda (Y)$ is the Casson-Walker invariant of $Y$ (cf. \cite{Walker}).

471: \end{The}

472: 

473: 

474: 

475: 

476: 

477: 

478: 

479: \vskip .2in

480: 

481: {\bf Acknowledgements} BLW likes to acknowledge the paper

482: of Ozsv\'ath and Szab\'o \cite{OS} on the theta divisor and the Casson-Walker 

483: invariant which leads to his proof of the equivalence 

484: of $\hat{SW}_Y$ and the Casson-Walker invariant, hence 

485: proving the conjecture formulated in \cite{OS} on the  equivalent

486: between $\hat{SW}_Y$ 

487: and their $\hat\theta$ invariant for all rational homology 3-sphere. 

488: BLW is partially supported by Australia Research Council Fellowship.

489: 

490: 

491: 

492: \section{The geometric triangle}

493: 

494: In this section, we identify the

495: monopoles on $Y$ with monopoles on $Y_1$ and $Y_0$.  Suppose given a

496: smoothly embedded knot   

497: $K$ in $(Y, \s)$, which represents a torsion element of order $n$ in

498:  $H_1(Y, \Z)$. We can split $Y$ along a torus as in 

499: \cite{CMW},

500: \[

501: Y= V\cup_{T^2} \nu(K).

502: \]

503: We choose a metric on $Y$ with a long cylinder

504: $[-r, r] \times T^2$, and denote the resulting manifold

505: as

506: \[

507: Y(r) = V\cup_{T^2} ([-r, r] \times T^2) \cup_{T^2} \nu(K).

508: \]

509: We additionally require that the chosen metric on $Y$ satisfies the

510: condition of Lemma 

511: 3.18 in \cite{CMW} in the neighbourhood $\nu(K)$ of the knot:

512: it has non-negative scalar curvature, strictly

513: positive away from the boundary. This induces a natural

514: metric on $Y_0(r)$. On $Y_1(r)$, we need to choose a metric which

515: agrees with the original metric on $Y (r)$ when restricted

516: to the knot complement $V$. The induced

517: metric from $Y_1(r)$ in the torus neighbourhood of $\nu(K)$

518: is the metric described in Lemma 3.21 \cite{CMW}. 

519: 

520: 

521: With this choice of the metric, the moduli space of monopoles

522: on $(Y, \s)$ is non-empty only if $\s|_{\nu(K)} $ is trivial.

523: Gluing the two $\spinc$ structures $\s|_V$ and $\s|_{\nu(K)}$

524: along $T^2$ by a gauge transformation on $T^2$ gives rise to a

525: $\Z_n$-family of $\spinc$ structures on $Y$ and $Y_1$, and to

526: a $\Z$-family of $\spinc$ structures on $Y_0$. The resulting

527: $\spinc$ structures  can be classified as the result of tensoring the

528: original $\spinc$ structure $\s$ with complex

529: line bundles $L_k (k\in \Z)$ whose

530: Euler class is given by $kPD([K])$.  The gluing

531: theorem for 3-dimensional monopoles and the 

532: perturbation $\mu$ on

533: $\nu(K)$, ``simulating the effect of surgery'',

534: provide the decomposition

535: of the moduli space for

536: \[

537: \cup_{k\in \{ 1, \cdots, n\}} \M_Y (\s\otimes L_k).

538: \]

539: 

540: \begin{The}

541: \label{decomposition}

542: With the  choice of perturbations and metrics on $Y, Y_1$ and

543: $Y_0$ described above, we have the following relation between the

544: critical sets of the Chern-Simons-Dirac functional

545: on the manifolds $Y, Y_1$ and $Y_0$:

546: \[\begin{array}{lll}

547: &&\bigcup_{k\in \{ 1, \cdots, n\}} \M_Y (\s\otimes L_k)\\[2mm]

548: &=& \bigcup_{k\in \{ 1, \cdots, n\}} \M_{Y_1}  (\s\otimes L_k)

549: \cup \bigcup_{k\in \Z} \M_{Y_0}  (\s\otimes L_k).

550: \end{array}

551: \]

552: \end{The}

553: \begin{proof}

554: First we assume that $Y$ is a rational homology 3-sphere. When

555: stretching the neck in $Y(r)$, as $r\to \infty$, we get two manifolds,

556: each endowed with an infinite cylindrical end,

557: \[

558: V(\infty) = V\cup_{T^2} ([0, \infty) \times T^2)\]\[

559: \nu(K) (\infty) = \nu(K) \cup_{T^2} ((-\infty, 0] \times T^2).

560: \]

561: The Seiberg-Witten monopole moduli space of $Y(r)$, for sufficiently large

562: $r$, can be described in terms of the

563: moduli spaces of $V(\infty)$ and $\nu(K) (\infty)$ as

564: analyzed in \cite{CMW}. 

565: 

566: Notice that the moduli spaces of flat connections on $T^2$, modulo

567: the subgroups of the gauge transformations on $T^2$ which

568: can be extended to the whole manifolds $V$ or $\nu(K)$, define

569: the following character varieties:

570: \[

571: \chi_0(T^2, V)

572: = H^1(T^2, \R)/2n\Z \la PD([l])\ra,

573: \]\[

574: \chi_0(T^2, \nu(K)) 

575: = H^1(T^2, \R)/2\Z \la PD([m])\ra.

576: \]

577: Thus the character variety $\chi_0(T^2, Y)$ is

578: $ \chi_0(T^2, Y) = H^1(T^2, \R).$

579: The covering maps between these character varieties are illustrated as 

580: follows

581: {\small \[

582: \diagram

583: 2\Z \langle PD([m]) \rangle \drto &  & 2n\Z \langle PD([l]) \rangle\dlto \\

584: & \chi_0(T^2,Y) \dlto^{\pi_1}\drto^{\pi_2}  & \\

585: \chi_0(T^2,V)\drto &  & \chi_0(T^2,\nu(K))\dlto \\

586: &  \chi(T^2)  & \\

587: \enddiagram

588: \]}

589: where the maps $\pi_i$ are the covering maps with  fibers as indicated.

590: 

591: Based on the analysis of the space $\M_V^* (\s|_V)$ of irreducible

592: monopoles on a 3-manifold with 

593: a cylindrical end modelled on $T^2$, as in \cite{CMW},

594: we see that the asymptotic limit map defines a continuous

595: map:

596: \[

597: \M_V^* (\s|_V) \stackrel{\partial_{\infty}}{\to}

598: \chi_0(T^2,V). 

599: \]

600: The reducibles on $V$ embed in the character variety $\chi_0(T^2,V)$.

601: Then, for a sufficiently large $r$, the gluing theorem gives a 

602: diffeomorphism:

603: \[

604: \#_Y : {\cal M}^*_V\backslash \partial_\infty^{-1}(U_\theta)

605: \times_{\chi_0(T^2,Y)}\chi (\nu(K)) 

606: \to\bigcup_{k=1}^n {\cal M}^*_{Y(r)}(\s\otimes L_k),

607: \]

608: here $U_\Theta$ is a small neighbourhood of the ``bad points'' $\Theta$ 

609: in $\chi_0(T^2, Y)$ where the twisted Dirac operator

610: on $T^2$ has non-trivial kernel, and $ \chi (\nu(K))$  

611: are the reducible lines in $\chi_0(T^2,Y)$. The above

612: fibred product is obtained (cf.\cite{CMW}) by taking the 

613: pullbacks of the images of the boundary value maps  under the

614: projections 

615: {\small

616: \[

617: \diagram

618: & \chi_0(T^2,Y) \dlto^{\pi_1}\drto^{\pi_2}  & \\

619: \chi_0(T^2,V) &  & \chi_0(T^2,\nu(K)). \\

620: \enddiagram

621: \]}

622: 

623: 

624: Let $(u, v)$ be the coordinates on

625: $\chi_0(T^2,Y) \cong H^1(T^2, \R)$ corresponding

626: to the holonomy around the longitude $l$ and

627: the meridian $m$ respectively. In the gluing map $\#_Y$ above, 

628: $\chi(\nu(K))$ corresponds to the lines

629: $\{v=2k, k=1, \cdots, n\}$. For each line $\{v=2k\}$,

630: the image of the gluing map gives

631: a diffeomorphism onto $\M^*_{Y(r)} (\s\otimes L_k)$.

632: 

633: For each $\spinc$ structure $\s\otimes L_k$, there is a

634: unique reducible monopole 

635: on $Y(r)$, which is given by the 

636: intersection of $\chi(V, \s)$, the flat connections

637: on $(V, \s)$, with the line $\{ v= 2k\}$ in $\chi_0(T^2, Y)$:

638: \[

639: \theta_Y(k) = \{ u=u(\s)\} \times_{\chi_0(T^2, Y)} \{ v= 2k\}.

640: \]

641: Here $u(\s)$ is the holonomy of the flat connections

642: in $\chi(V, \s)$ around the longitude $l'$.

643: 

644: Similarly, we have gluing maps for monopoles on $Y_1(r)$ and

645: $Y_0(r)$, respectively.

646: In the gluing map $ \#_{Y_1}$

647: for $ {\cal M}^*_{Y_1(r)}(\s\otimes L_k)$,

648: \[

649:  \#_{Y_1} : {\cal M}^*_V \backslash

650: \partial_\infty^{-1}(U_\theta)\times_{\chi_0(T^2,Y_1)}\chi (\nu(K)) 

651: \to\bigcup_{k=1}^n {\cal M}^*_{Y_1(r)}(\s\otimes L_k), 

652: \]

653: $\chi (\nu(K))$ is identified with $v-u = 2k+1$.

654: Similarly, we have the gluing map $ \#_{Y_0}$

655: for $ {\cal M}^*_{Y_0(r)}(\s\otimes L_k)$,

656: \[

657:  \#_{Y_0} : {\cal M}^*_V \backslash

658: \partial_\infty^{-1}(U_\theta)\times_{\chi_0(T^2,Y_0)}\chi (\nu(K)) 

659: \to \bigcup_{k\in\Z}{\cal M}^*_{Y_0(r)}(\s\otimes L_k), 

660: \]

661: where $\chi (\nu(K))$ is given by $ u = 2k$. For each $k\in \{1,

662: \cdots, n\}$, the reducible monopole for $(Y_1, \s\otimes L_k)$,

663: consists of the unique point 

664: \[

665: \theta_{Y_1}(k) = \{ u=u(\s)\} \times_{\chi_0(T^2, Y)} \{ v=u+ 2k+1\}.

666: \]

667: 

668: For $Y_0$ with a non-trivial $\spinc$ structure 

669: $\s\otimes L_k$ ($k\in \Z, k\neq 0$), the set of reducibles is empty

670: for any generic perturbation, and for 

671: $\s\otimes L_0= \s$, it consists of one circle of reducibles

672: $u=0$ in the cylinder $\chi_0(T^2,Y_0) = \chi_0(T^2,V)$, which

673: can be perturbed away by introducing a small perturbation as in

674: Theorem 6.13 \cite{CMW}.

675: 

676: Now we apply the perturbation to simulate the effect of Dehn

677: surgery. This amounts to a careful choice of perturbation as

678: in Section 6 \cite{CMW}, which we now briefly describe.

679: 

680: Choose a compactly supported 2-form $\mu$  representing the generator

681: of $ H^2_{cpt} (D^2 \times S^1)$, defined as in Lemma  3.18 \cite{CMW},

682: such that we have

683: \ba

684: \int_{D^2 \times \{pt \}}\mu  = 1 \label{normalization}

685: \na

686:  for any point on $S^1$. Under the isomorphism $H^2_{cpt} (\nu(K))

687: \cong H_1(\nu(K))$, given by Poincar\'e duality, this form corresponds

688: to the generator $[\mu]= PD_{\nu(K)}(l)$.

689: The class of $\mu$ in $H^2( D^2 \times S^1)$

690: is trivial, and we can write $\mu= d\nu$, where $\nu$ is a 1-form

691: satisfying $\int_{S^1 \times \{pt \}}\nu  = 1$, i.e.

692: $\nu= PD_{T^2}(l)$.

693: Choose on $\nu(K)$ a metric as in Lemma 3.18 \cite{CMW}.

694: 

695: Fix a $U(1)$-connection $A_0$ representing the trivial connection on $T^2$,

696: For any $U(1)$-connection $A$, define  

697: $T_A$ to be

698: \[

699: T_A(z)= -i\int_{\{z\in D^2\} \times S^1} (A-A_0).

700: \]

701: 

702: For any  given $\epsilon >0$, we can choose a function $\hat f:\R\to\R$ with

703: the following properties.

704: 

705: (a) $\hat f$ is continuously differentiable on $(-1,1)$ and satisfies the

706: periodicity $\hat f(t+2)=\hat f(t)$

707: 

708: (b) the derivative $\hat f'$ has range $\hat f'(t)\in [-1,1]$ for all $t\in

709: [-1,1]$, and satisfies $\hat f'(1-t) = \hat f'(1+t)$ for $t\in\R$.

710: 

711: (c) the following estimate holds:

712: $\sup_{t\in [-1+\epsilon, 1-\epsilon]} | \hat f'(t)-t | <

713: \epsilon.$

714: 

715: Now, for the $\spinc$ structure $\s\otimes L_k$ ($k=1, \cdots, n$),

716: consider the function $f_k'(t)= \hat f'(t+1)+2k$ and define a

717: perturbation of the Seiberg-Witten equations

718: on $(Y, \s\otimes L_k)$ in the following

719: way:

720: \ba

721: \left\{\begin{array}{l}

722:  F_A = *\sigma(\psi,\psi) + f'_k(T_A) \mu \\[2mm]

723: \dirac_A (\psi ) = 0

724: \end{array}

725: \right. .

726: \label{p SW}

727: \na

728: 

729: With respect to the chosen metric on $\nu(K)$, with sufficiently large

730: positive scalar curvature on the support of $\mu$ as specified in

731: Lemma 3.18 \cite{CMW}, 

732: the only solutions

733: of the perturbed monopole equations are reducibles $(A, 0)$, that satisfy

734: \ba

735: F_A = f_k'(T_A)\mu.

736: \label{deformflat}

737: \na

738: 

739: In addition to this surgery perturbation, we consider another

740: perturbation of the Seiberg--Witten equations on the

741: tubular neighbourhood $\nu(K)$ in $Y$, $Y_1$, and $Y_0$. This

742: perturbation has the effect of producing a global shift in the

743: character variety to avoid the bad points on $H^1(T^2, \R)$ when

744: we deform the unperturbed geometric triangles in 

745: $\chi (T^2)$ to the perturbed geometric triangles in 

746: $\chi (T^2)$.

747: 

748: Let $\mu$ be a compactly supported 2-form in $D^2\times S^1$

749: satisfying (\ref{normalization}). Let $\eta >0$ be some small real

750: parameter. Consider an additional perturbation

751: \ba F_A= *\sigma(\psi,\psi) \pm\eta \mu \label{eta:pert} \na

752: of the curvature equation on $\nu(K)$ inside $Y$ and  inside $Y_0$.

753:  

754: This perturbation has the effect of shifting the asymptotic values by

755: an amount $\eta$. We choose the sign so that the line of reducibles on

756: $\nu(K)\subset Y$ for $\s\otimes L_k$ 

757: becomes $\{ (u,v) | v=2k +\eta \}$, the line of reducibles on

758: $\nu(K)\subset Y_1$ for $\s\otimes L_k$

759:  remains the same $\{ (u,v) | v-u =2k+ 1 \}$, and the line

760: of reducibles on $\nu(K)\subset Y_0$ for $\s\otimes L_k$

761:  becomes $\{ (u,v) | u=2k+ \eta \}$.

762: 

763: On $\nu(K)$ inside $Y$ for $\s\otimes L_k$

764:  we shall consider the perturbed curvature

765: equation 

766: \ba  F_A= *\sigma(\psi,\psi) +

767: (f'_k(T_A-\eta)+\eta)\mu. \label{eta:mu:pert} \na 

768: 

769: Therefore, we can partition

770: the moduli spaces for $(Y, \s\otimes L_k)$ ($k=1,\cdots, n$) into the union

771: of the moduli spaces for $(Y_1, \s\otimes L_k)$ ($k=1,\cdots, n$)

772: and $(Y, \s\otimes L_k)$ ($k\in \Z)$,  as in Theorem 6.3

773: \cite{CMW}. This completes the proof of the 

774: theorem for the case of rational homology 3-sphere $Y$ with

775: a knot $K$ representing a torsion element of order n in $H_1(Y, \Z)$.

776: 

777: For a general 3-manifold $Y$ with a smoothly embedded

778: knot $K$ representing a torsion element of order n

779: in  $H_1(Y, \Z)$, the proof is essentially the same as the case

780: of rational homology spheres discussed above, and we omit the details

781: here. 

782: 

783: \end{proof}

784: 

785: The perturbation can be illustrated as in Figure

786: \ref{deform:triangle1} where $n=4$. From now on, we will

787: use the following notations to denote the reducibles lines

788: for $Y$, $Y_1$ and $Y_0$ repectively:

789: \ba\begin{array}{c}

790:  L_Y(\s_k)= \{ (u,v) | \, v=f'_k(u-\eta) +\eta \}\\[2mm]

791:  L_{Y_1}(\s_k)=\{ (u,v) | \, v=u+2k+1 \}\\[2mm] 

792:  L_{Y_0}(\s_k)= \{ (u,v) | \, v=2k+\eta \}. 

793: \end{array}

794: \label{reducible:lines}

795: \na

796: 

797: \begin{figure}[ht]

798: \epsfig{file= ex5fig0.eps,angle= 0}

799: \caption{The perturbed geometric triangle as in Theorem \ref{decomposition}}

800: \label{deform:triangle1}

801: \end{figure}

802: 

803: 

804: 

805: With a more careful study of the perturbed geometric triangles, we have

806: the following decomposition of 3-dimensional monopoles

807: on $Y$ under the Dehn surgery.

808: 

809: \begin{The}

810: With the perturbations and metrics on $Y, Y_1$ and

811: $Y_0$ as in Theorem \ref{decomposition}, and for any

812: fixed $m, p \in \{0, \cdots, n-1\}$, there exists

813:  a further perturbation on $Y$,

814: such that we have the following relation between the

815: critical sets of the Chern-Simons-Dirac functional

816: on the manifolds $Y, Y_1$ and $Y_0$:

817: \ba \label{split-mp}

818:  \M_Y (\s\otimes L_m)

819: = \M_{Y_1}  (\s\otimes L_m)

820: \cup \bigcup_{k\in \Z} \M_{Y_0}  (\s\otimes L_{nk+p}).

821: \na

822: \label{refine:decomposition}

823: \end{The}

824: \begin{proof}

825: In the proof of Theorem \ref{decomposition}, we know

826: that the perturbed Seiberg-Witten monopoles on $Y, Y_1$

827: and $Y_0$ are given by the following gluing models (here 

828: we assume that $Y$ is a rational homology 3-sphere):

829: \[

830: {\cal M}^*_{Y(r)}(\s\otimes L_k) 

831: \cong {\cal M}^*_V\backslash \partial_\infty^{-1}(U_\theta)

832: \times_{\chi_0(T^2,Y)} \{ v=2k \},

833: \]

834: \[

835: {\cal M}^*_{Y_1(r)}(\s\otimes L_k) 

836: \cong {\cal M}^*_V\backslash \partial_\infty^{-1}(U_\theta)

837: \times_{\chi_0(T^2,Y_1)} \{ v= u+2k+1\},

838: \]

839: \[

840: {\cal M}^*_{Y_0(r)}(\s\otimes L_k) 

841: \cong {\cal M}^*_V\backslash \partial_\infty^{-1}(U_\theta)

842: \times_{\chi_0(T^2,Y_0)} \{ u=2k\}.

843: \]

844: Note that the additional perturbation (\ref{eta:pert}) of the 

845:  curvature equation on $\nu(K)$ inside $Y$ and  inside $Y_0$

846: introduce a shift of coordinates $(u, v)$

847: to $(u+\eta, v+\eta)$. We can introduce these new coordinates,

848: still denoted by $(u, v)$. Then the reducible line for $V$

849: is given by $u = -\eta$ in $H^1(T^2, \R)$. 

850: 

851: We will show that there exists a further surgery

852: perturbation on $Y$ that suits the purpose of

853: identifying monopoles on $Y, Y_1$ and

854: $Y_0$ as stated in the Theorem.  

855: Without loss of the generality,

856: after possible coordinates change, we can assume that

857: $\s\otimes L_m = \s\otimes L_0$. 

858: Fix $p\in \{0, \cdots, n-1\}$. We will construct a function

859: $f_0': \R \to \R$, 

860: which depends on a small $\epsilon >0$, such that,

861: as $\epsilon \to 0$, the curve $v=f_0'(u)$ approaches

862: the union of lines

863: \[ L_{Y_1}(\s\otimes L_0)\cup \bigcup_{k\in \Z} L_{Y_0}(\s\otimes

864: L_{nk+p}) \] 

865: where

866: $ L_{Y_1}(\s\otimes L_0)=\{ v=u+1\}, 

867:  L_{Y_0}(\s\otimes L_{nk+p})= \{ u= 2nk+2p\}.$

868: 

869: We identify $\chi_0(T^2, V)$ with the fundamental domain

870: \[

871: \{ u\in \R\} \times \{ 0\le v < 2n \},

872: \]

873: in $H^1(T^2, \R)$. The asymptotic values

874: $\partial_\infty (\M^*_V) \subset \chi_0(T^2, V)$ can

875: be lifted to $H^1(T^2, \R)$ periodically. Using this

876: $2n$--periodicity, we only need to construct

877: a function $f_0: [-1, 2n-1] \to [0, 2n]$,

878: which depends on $\epsilon$, such that, for any given

879: $\epsilon\leq \epsilon_0$, we have

880: \[

881: \sup_{t\in [-1, 2n-1]\backslash [-\epsilon , \epsilon]}

882: |f'_0 (t) - t | < \epsilon.

883: \]

884: Such function can be easily constructed as in the 

885: proof of Theorem \ref{decomposition}. Then 

886: over $[2nk-1, 2nk+2n-1]$, $f_0'(t)$ is defined

887: to be $f'_0 (t)+2nk$.  See Figure 

888:   \ref{deform:triangle2} where the perturbation is

889: illustrated in the cases with $n=4, m=p=0$ and

890: with $n=4, m=0, p=2$, respectively.  The general case

891: can be proved by a similar method.

892: \end{proof}

893: 

894: 

895: \begin{figure}[ht]

896: \epsfig{file= ex5fig1.eps,angle= 0}

897: \caption{The perturbed geometric triangle as in Theorem

898: \ref{refine:decomposition} 

899: \label{deform:triangle2}} 

900: \end{figure}

901: 

902: 

903: \section{The relative gradings}

904: 

905: In the previous section, we have the following decomposition:

906: \[

907:  \M_{Y, \mu} (\s\otimes L_m)

908: = \M_{Y_1}  (\s\otimes L_m)

909: \cup \bigcup_{k\in \Z} \M_{Y_0}  (\s\otimes L_{nk+p})  \]

910: for any fixed $m, p\in  \{0, \cdots,  n-1\}$. 

911: Assume that $Y$ is a rational homology 3-sphere.

912: First we fix a grading on  $\M_{Y, \mu} (\s\otimes L_0)$ defined

913: in terms of the spectral flow of the linearization operator

914: for the 3-dimensional Seiberg-Witten equations along a path

915: connecting an irreducible monopole in $\M_{Y,\mu} (\s\otimes L_0)$

916: to the unique reducible $\theta_Y(0)$ in the configuration

917: space for $(Y, \s\otimes L_0)$.  Then

918: the analysis of the relative grading in Part I section 7 \cite{CMW}

919: can be applied to induce a compatible grading 

920: on $\M_{Y_1}  (\s\otimes L_0) \cup 

921: \bigcup_{k\in \Z} \M_{Y_0}  (\s\otimes L_{nk+p})$ as follows, cf.

922: Proposition 7.3  -- Corollary 7.7 in \cite{CMW}.

923: 

924: \begin{Pro}\label{relative:grading0} Let $Y$ be a rational homology 3-sphere.

925: For any fixed $p \in \{0, \cdots, n-1\}$, the Floer complexes

926: \[

927: C_*(Y, \s\otimes L_0, \mu) = \oplus _{a\in \M_{Y,\mu} (\s\otimes L_0)}

928: \Z \la a \ra,

929: \]

930: \[

931: C_*(Y_1, \s\otimes L_0) = \oplus _{a\in \M_{Y_1} (\s\otimes L_0)} \Z

932: \la a \ra,

933: \]

934: \[

935: C_*(Y_0, \s\otimes L_{nk+p}) = \oplus _{a\in \M_{Y_0} 

936: (\s\otimes L_{nk+p})} \Z \la a\ra,

937: \]

938: have a compatible relative grading of generators in the

939: following sense.

940: \begin{enumerate}

941: \item Suppose given two irreducible critical points $a, b$

942: in $\M^*_{Y_1}(\s\otimes L_0)$, and the corresponding elements 

943: $a^\epsilon, b^\epsilon$ in $\M^*_{Y,\mu}(\s\otimes L_0)$

944: under the above decomposition (\ref{split-mp}). Then

945: \[

946: \deg_{Y,\mu}(a^\epsilon)-\deg_{Y,\mu}(b^\epsilon)=

947: \deg_{Y_1}(a)-\deg_{Y_1}(b).

948: \]

949: \item Suppose given two monopoles $a, b$

950: in $\M^*_{Y_0}(\s\otimes L_{nk+p})$, and the corresponding elements 

951: $a^\epsilon, b^\epsilon$ in $\M^*_{Y, \mu}(\s\otimes L_0)$

952: under the above decomposition (\ref{split-mp}). Then 

953: \[

954: \deg_{Y_0,\s_k}(a)-

955: \deg_{Y_0,\s_k}(b) = \deg_{Y,\mu}(a^\epsilon)-

956: \deg_{Y,\mu}(b^\epsilon) \ \ \hbox{ mod } (2nk+2p).

957: \]

958: Therefore, the grading $\deg_{Y,\mu}$ defines a $\Z$-valued lift

959: of the $\Z_{2nk+2p}$-valued relative index on 

960: $\M^*_{Y_0}(\s\otimes L_{nk+p})$ under the decomposition

961: of $\M^*_{Y, \mu}(\s\otimes L_0)$.

962: \end{enumerate}

963: 

964: \end{Pro}

965: 

966: For a general 3-manifold $(Y, \s)$ with $b_1(Y)> 0$

967: and  a knot $K$ representing

968: a torsion element of order $n$, if $\s$ is a torsion

969: $\spinc$ structure, we know that,

970: for any $k \in \{0, \cdots, n-1\}$, the $\spinc$ structure 

971: $\s\otimes L_k$ also has a torsion class $c_1(\s\otimes L_k)$. Thus,

972: after a small perturbation to get rid of the $(S^1)^{b_1(Y)}$-family

973: of reducibles, we have a $\Z$-graded

974: $$\M^*_{Y, \mu}(\s\otimes L_k) \cong \M_{Y,\mu}(\s\otimes L_k).$$

975: Then it is easy to see that the results

976: in Part I section 7 \cite{CMW} hold in this case as well 

977: without any substantial change.

978: 

979: Now assume that $c_1(\s)$ is a non-torsion element, with multiplicity $2\ell$

980: in $H^2(Y, \Z)/\hbox{Torsion}$. Then, for

981: any $k \in \{0, \cdots, n-1\}$, the set of generators

982: $$\M^*_{Y, \mu}(\s\otimes L_k) \cong \M_{Y, \mu}(\s\otimes L_k)$$

983: is $2\ell$-graded, and so is 

984: $$\M^*_{Y_1}(\s\otimes L_k) \cong \M_{Y_1}(\s\otimes L_k).$$

985: Then for any $k\in \Z$, any non-empty moduli space

986: $$\M^*_{Y,\mu}(\s\otimes L_k) \cong \M_{Y, \mu}(\s\otimes L_k)$$

987: is $\Z_{2\ell_{[k]}}$-graded, where $2\ell_{[k]}$

988: is the maximum common factor of $2\ell$ and $2k$.

989: We can choose a relative grading on $\M^*_{Y, \mu}(\s\otimes L_0)$

990: by the spectral flow of the linearization operator along

991: a path connecting any irreducible monopole

992: in $\M^*_{Y}(\s\otimes L_0)$ to a fixed monopole

993: $a_0$ in $\M^*_{Y, \mu}(\s\otimes L_0)$.  

994: Note that this grading is $2\ell$-graded. Then the analysis 

995: in Part I section 7 can also be applied to obtain the

996: following proposition.

997: 

998: \begin{Pro}\label{relative:grading1}

999: For $p \in \{0, \cdots, n-1\}$ and $k\in \Z$, the Floer complexes

1000: $C_*(Y, \s\otimes L_0), C_*(Y_1, \s\otimes L_0)$, and 

1001: $C_*(Y_0, \s\otimes L_{nk+p})$ have a compatible 

1002: relative grading of generators in the following sense.

1003: \begin{enumerate}

1004: \item Suppose given two irreducible critical points $a, b$

1005: in $\M^*_{Y_1}(\s\otimes L_0)$, and the corresponding elements

1006: $a^\epsilon, b^\epsilon$ in $\M^*_{Y, \mu}(\s\otimes L_0)$, 

1007: then

1008: \[

1009: \deg_{Y,\mu}(a^\epsilon)-\deg_{Y,\mu}(b^\epsilon)=

1010: \deg_{Y_1}(a)-\deg_{Y_1}(b),

1011: \]

1012: as a $\Z_{\ell}$-valued function.

1013: \item Suppose given two monopoles $a, b$

1014: in $\M^*_{Y_0}(\s\otimes L_{nk+p})$, and the corresponding elements

1015: $a^\epsilon, b^\epsilon$ in $\M^*_{Y, \mu}(\s\otimes L_0)$,

1016: then

1017: \[

1018: \deg_{Y_0,\s_k}(a)-

1019: \deg_{Y_0,\s_k}(b) = \deg_{Y,\mu}(a^\epsilon)-

1020: \deg_{Y,\mu}(b^\epsilon) \ \ \hbox{ mod } (2\ell_{[nk+p]}).

1021:  \]

1022: Thus, the induced grading $ \deg_{Y,\mu}$ on 

1023: $\M^*_{Y_0}(\s\otimes L_{nk+p})$, under the

1024: decomposition (\ref{split-mp}), defines a choice of a

1025: $\Z_{2\ell}$-valued lifting of the $\Z_{2\ell_{[nk+p]}}$-valued

1026: relative index. 

1027: \end{enumerate}

1028: \end{Pro} 

1029: 

1030: 

1031: 

1032: In the rest of this section we discuss the induced

1033: gradings on the various $\M_Y^*(\s\otimes L_m)$, for $m \in \{1,

1034: \cdots, n-1\}$. 

1035: Notice that, in the case of a rational homology 3-sphere $Y$,

1036: the perturbed line $L_{Y}(\s\otimes L_m)$ is a

1037: deformation of the line $L_{Y}(\s\otimes L_0)$. Along this

1038: deformation, the corresponding parameterized

1039: reducibles form a path

1040: connecting $\theta_Y(0)$ to $\theta_Y(m)$. As the following Lemma

1041: shows, this deformation

1042: can be realized as a perturbation of the monopole equations

1043: for $(Y, \s\otimes L_0)$. 

1044: 

1045: \begin{Lem} 

1046: For any $m \in \{1, \cdots, n-1\}$, there exists a perturbation

1047: $\mu_m$ for the Seiberg-Witten monopole equations

1048: on $(Y, \s\otimes L_0)$,

1049:  such that there is a diffeomorphism:

1050: \[

1051: \M_{Y, \mu_m} (\s\otimes L_0) \cong \M_Y(\s\otimes L_m).

1052: \]

1053: Moreover, if $Y$ is a rational homology 3-sphere, 

1054: the grading on $\M_Y^*(\s\otimes L_m)$

1055: induces a grading on $\M_{Y, \mu_m}^* (\s\otimes L_0)$. Under the

1056: above identification,

1057: the resulting grading differs from the original grading

1058: on  $\M_{Y}^* (\s\otimes L_0)$ by

1059: the wall-crossing formulae studied in

1060: \cite{MW1} for $(Y, \s\otimes L_0)$. If $b_1(Y)> 0$,

1061: then the induced grading on $\M_Y(\s\otimes L_m)$

1062: from $\M_{Y, \mu_m} (\s\otimes L_0)$ agrees with the

1063: relative grading on $\M_Y(\s\otimes L_m)$. 

1064: \label{pert-0m}

1065: \end{Lem}

1066: 

1067: \begin{proof}

1068: The first claim follows from the gluing models

1069: of the monopoles in $\M_{Y} (\s\otimes L_0)$

1070: and $ \M_Y (\s\otimes L_m)$. Then, using the results in \cite{MW1}, we

1071: know that, in the 

1072: case of the rational homology 3-sphere,

1073: the spectral flow of the twisted Dirac operator

1074: along the path of reducibles gives the 

1075: index shift on

1076: $\M_{Y}^* (\s\otimes L_0)$ and $\M_{Y, \mu_m}^* (\s\otimes L_0)$

1077: according to the wall-crossing formulae derived in \cite{MW1}. Again

1078: by the results of \cite{MW1},

1079: for $Y$ with $b_1(Y)>0$, the induced

1080: grading from the parametrized spectral flow is same as the

1081: original relative index on $ \M_Y (\s\otimes L_m)$. 

1082: \end{proof}

1083: 

1084: Thus, Lemma \ref{pert-0m} provides a consistent way of assigning a

1085: choice of absolute grading on the various $\M_Y (\s\otimes L_m)$.

1086: The degree shift of Lemma \ref{pert-0m} can be described as follows. 

1087: Let $\theta_Y(t)$ be the path of reducibles 

1088: on $(Y, \s\otimes L_0) $ for the family of perturbations

1089: connecting $\M_{Y}^* (\s\otimes L_0)$

1090: to $\M_{Y, \mu_m}^* (\s\otimes L_0)$,

1091: then the wall crossing formulae in \cite{MW1}

1092: tell us that

1093: the index shift is given by the complex spectral flow

1094: \ba

1095: SF_\C(\dirac_{\theta_Y(t)}). 

1096: \label{index:shift}

1097: \na

1098: 

1099: 

1100: \section{Geometric limits and the holomorphic triangles}

1101: 

1102: 

1103: \subsection{Surgery cobordisms}

1104: 

1105: We first briefly describe the surgery cobordisms from

1106: $Y_1$ to $Y$, from $Y$ to $Y_0$, and from $Y_0$ to $Y_1$,

1107: respectively, as in \cite{MW3}.  The cobordism $W_1$, from

1108: $Y_1$ to $Y$, is obtained by removing from the trivial cobordism 

1109: $Y_1\times [0, 1]$ an $S^1\times D\cong \nu(K)\times \{ 1\}$, where $D$ is

1110: a disk, and $\nu(K)$ is the tubular neighbourhood of the knot in

1111: $Y_1$, and then attaching a 2-handle with framing $-1$.

1112: We denote by $D_1$ the core disk of the 2-handle in $W_1$. 

1113: Similarly, the cobordism $W_0$, form $Y$ to $Y_0$, is obtained by removing

1114: from the trivial cobordism $Y_0\times [0, 1]$ an $S^1\times D\cong

1115: \nu(K)\times \{ 0\}$ and attaching a 2-handle with framing zero.

1116: We denote by $D_0$ the core disk of the 2-handle in $W_0$.

1117: Attaching the two-handle has the effect of modifying the

1118: boundary component $Y_1\times \{ 1\}$ in the trivial cobordism to the

1119: boundary component $Y\times \{ 1 \}$ in the non-trivial cobordism

1120: $W_1$, or, respectively,  the

1121: boundary component $Y_0\times \{ 0 \}$ in the trivial cobordism 

1122: to the boundary component $Y\times \{ 0 \}$ in $W_0$. 

1123: The cobordism $\bar W_2$ connecting $Y_0$ and $Y_1$,

1124: satisfies the relation

1125: $$ \bar W = \bar W_2 \# \C \P^2, $$

1126: where $\bar W$ is the composite cobordism $\bar W=\bar W_0 \cup_Y \bar

1127: W_1$.

1128: 

1129: We assume that the 3-manifolds $Y_1$, $Y$, and $Y_0$ are endowed with metrics

1130: with a long cylinder $T^2\times [-r,r]$, as specified in

1131: section 2 (see also \cite{CMW}). We consider the manifolds $W_1$ and

1132: $W_0$ endowed 

1133: with infinite cylindrical ends $Y_1 \times (-\infty,

1134: -T_0]$ and $Y\times [T_0,\infty)$, and $Y_0 \times

1135: [T_0,\infty)$ and $Y\times (-\infty, -T_0]$, respectively. 

1136: As in \cite{MW3}, we can decompose the cobordisms $W_i$ as

1137: \ba 

1138: W_i= V\times \R \cup_{T^2\times \R} T^2\times [-r,r]\times \R

1139: \cup_{T^2\times \R}  W_i(\nu(K)). 

1140: \label{split:cobord} 

1141: \na

1142: The  region $W_i(\nu(K))$ has the following property. There 

1143: is a compact set ${\cal K}$ in $W_i$ such that the intersection

1144: ${\cal K} \cap W_i(\nu(K))$ is obtained by attaching a 2-handle 

1145: $D\times D$ to the product $\nu(K)\times [-T_0, T_0]$, and, outside

1146: of ${\cal K}$, the region ${\cal K}^c \cap W_i(\nu(K))$ consists of product

1147: regions $\nu(K)\times [ T_0 ,\infty)$ and $\nu(K)\times (-\infty, -T_0]$,

1148: and $T^2\times [r_0,r] \times [-T_0,T_0]$.

1149: 

1150: As in \cite{MW3},  consider an interior point $x_i$

1151: contained in the core disk of the 2-handle, $x_i \in D_i$,

1152: and we denote by $\hat W_i$ the 

1153: punctured cobordism $\hat W_i = W_i \backslash \{ x_i \}$. Similarly,

1154: we can consider the punctured manifold

1155: $$ \hat W_i(\nu(K)) = W_i(\nu(K))\backslash \{ x_i \}. $$

1156: In  the manifolds $\hat W_i(\nu(K))$, endowed with an extra

1157: asymptotic end of the form $S^3 \times [0,\infty)$ at

1158: the puncture, we can identify a product region 

1159: \ba

1160:  {\cal V}=\nu(K)_{r_0}\times \R \cong D\times (D_i\backslash \{ x_i

1161: \}). 

1162: \label{prodW} 

1163: \na 

1164: Thus, we identify the manifold $W_i$ with a

1165: connected sum

1166: $$ W_i = \hat W_i \# Q_i, $$

1167: with a long cylindrical neck $S^3\times [-T(r),T(r)]$, and with $Q_i$ a

1168: 4-ball, where $S^3$ is  decomposed as the union of

1169: two solid tori in the standard way,

1170: $S^3=\nu \cup \tilde \nu$, with $\nu\cong \tilde\nu \cong D\times

1171: S^1$. Then the product region ${\cal V}$ of (\ref{prodW}) 

1172: in $W_i$ identifies the standard solid torus $\nu$ in $S^3$ with the

1173: neighbourhood 

1174: $\nu(K)$ of the knot $K$ in $Y$. Similarly, there is a product

1175: region $\tilde {\cal V}$ which identifies the other solid torus

1176: $\tilde\nu$ in $S^3$ with the 

1177: tubular neighbourhood $\nu(K)$ in $Y_i$, after the surgery.

1178: The resulting punctured cobordism can be written as

1179: \ba

1180:  \hat W_i(r)=( V_r\times \R) \cup {\cal V}(r) \cup \tilde {\cal V}(r).

1181: \label{hat:W(r)}

1182: \na

1183: 

1184: We now impose a choice of metrics and perturbations for

1185: the Seiberg-Witten equations

1186: on the cobordisms as in subsection 2.2 of \cite{MW3}. Then we can adopt the

1187: results of \cite{MW2}\cite{MW3} to understand

1188: the asymptotic limits of finite energy monopoles, under the

1189: splitting of the punctured cobordisms as $r\to\infty$, as in

1190: (\ref{hat:W(r)}). 

1191: 

1192: 

1193: \subsection{Geometric limits and holomorphic triangles}

1194: 

1195: We assume that $Y$ and $Y_1$ are endowed with the

1196: $\spinc$ structure $\s\otimes L_0$. The other $\spinc$ structures can

1197: be studied analogously. 

1198: 

1199: On the surgery cobordism $W_1(r)$, there

1200: is a $\Z$-family of $\spinc$ structures whose

1201: restrictions to the two ends agree with

1202: $\s\otimes L_0$ on $Y$ and $Y_1$ respectively. 

1203: We will study the moduli spaces of monopoles on

1204: $W_1(r)$ with asymptotic values in

1205: $\M_{Y_1}(\s\otimes L_0)$ and 

1206: $\M_{Y, \mu}(\s\otimes L_0)$ at the two ends,

1207: where $\mu$ is the surgery perturbation defined by $f_0'$ as

1208: in the proof of Theorem \ref{refine:decomposition}.

1209: In the case $b_1(Y)>0$, we only consider

1210: the components of minimal energy, as defined in \cite{MW3} \cite{MW4}, 

1211: among all the possible moduli spaces with different $\spinc$

1212: structures and with the given asymptotic values.

1213: In particular, for a rational homology 3-sphere $Y$, 

1214: we only consider the moduli spaces of minimal

1215: dimension among the $\Z$-family of $\spinc$ structures.

1216: 

1217: With this convention understood, we denote by

1218: $\M^{W_1}(a_1, a)$ 

1219: the moduli space with asymptotic values

1220:  $a_1\in\M_{Y_1}(\s\otimes L_0)$ and $a\in\M_{Y,\mu}(\s\otimes L_0)$.

1221: Similarly, we denote by $\M^{W_0}(a, a_0)$ 

1222: the moduli space with asymptotic values

1223:  $a\in\M_{Y,\mu}(\s\otimes L_0)$ and 

1224: $a_0 \in \M_{Y_0}(\otimes L_{nk+p})$ for $k\in \Z$, for a

1225: fixed $p\in\{0, \ldots, n-1\}$. Under  a generic choice of the perturbation,

1226: all these moduli spaces $\M^{W_1}(a_1,a)$ and

1227: $\M^{W_0}(a,a_0)$ are cut out transversely and of the expected

1228: dimension.

1229: 

1230: The convergence and gluing arguments developed in \cite{MW1} can be

1231: applied to this case as well, to give the following 

1232: compactifications of $\M^{W_1}(a_1, a)$ and $\M^{W_0}(a, a_0)$. 

1233: 

1234: \begin{Pro}\label{compactification}

1235: Suppose that $\M^{W_1}(a_1, a)$ is non-empty, then $\M^{W_1}(a_1, a)$

1236: admits a compactification to a manifold

1237: with corners, where the  

1238: codimension $1$ boundary strata consist of

1239: \ba 

1240: \label{compact:W1}

1241: \begin{array}{c}

1242: \bigcup_{c\in \M_{Y, \mu}^*(\s\otimes L_0)}

1243: \M^{W_1}(a_1,c)\times \hat \M_{Y,\mu}(c,a)

1244: \\[2mm]

1245: \cup \bigcup_{c_1 \in \M_{Y_1}^*(\s\otimes L_0)} \hat\M_{Y_1}(a_1,c_1)\times

1246: \M^{W_1}(c_1,a), 

1247: \end{array} 

1248: \na

1249: and with extra components

1250: \ba 

1251: \begin{array}{c} 

1252: \hat \M_{Y_1}(a_1,\theta_1)\times U(1) \times

1253: \M^{W_1}(\theta_1,a) \\[2mm]

1254: \M^{W_1}(a_1,\theta)\times U(1) \times\hat

1255: \M_{Y,\mu}(\theta,a), 

1256: \end{array} 

1257: \label{red:compact:W1} 

1258: \na

1259: when splitting through the reducibles $\theta_1$ and

1260: $\theta$ in $ \M_{Y_1}^*(\s\otimes L_0)$ and

1261: $\M_{Y, \mu}^*(\s\otimes L_0)$ respectively . 

1262: We also have the similar compactification for 

1263: $\M^{W_0} (a, a_0)$.

1264: \end{Pro}

1265: 

1266: 

1267: Now we can describe the geometric limits of monopoles

1268: in  $\M^{W_1}(a_1, a)$ and $\M^{W_0} (a, a_0)$

1269: when stretching $r\to \infty$ inside

1270: \[

1271:  \hat W_i(r)= V_r\times \R \cup {\cal V}(r) \cup \tilde {\cal V}(r).

1272:  \]

1273: We only describe the case of $\hat W_1(r)$. With 

1274: similar arguments we have the corresponding 

1275: geometric limits for $\hat W_0(r)$.

1276: The proof of the results stated below on these geometric limits

1277: follows from the same arguments of \cite{MW2}\cite{MW3}.

1278: 

1279: Consider elements $a_i^{(1)}\in \M_{Y_1}^*(\s\otimes L_0)$

1280: and $a_j(\epsilon) \in \M_{Y,\mu}^*(\s\otimes L_0)$,

1281: which we can write as

1282: $$ a_i^{(1)}=[(A_i^-,\psi_i^-)\#(a_{\infty,i}^-,0)] $$

1283: $$ a_j(\epsilon)= [(A_j^+(\epsilon),\psi_j^+(\epsilon))\#

1284: (a_{\infty,j}^+(\epsilon),0)], $$

1285: with 

1286: $$[A_i^-,\psi_i^-], \, \, [A_j^+(\epsilon),\psi_j^+(\epsilon)]\in \,

1287: \, \M_V^*(\s\otimes L_0|_V)$$

1288: $$ a_{\infty,i}^-\in \, \, L_{Y_1}(\s\otimes L_0), 

1289:  a_{\infty,j}^+(\epsilon)\in \, \, L_{Y,\mu}(\s\otimes L_0).$$

1290: 

1291: \begin{The} (Proposition 3.1 and Remark 6.2 in \cite{MW3})

1292: Assume that $\M^{W_1(r)}\bigl(a_i^{(1)}, a_j(\epsilon)\bigr)$

1293: is non-empty for all

1294: sufficiently large $r$. Then a

1295:  family of solutions $[\A_1(r), \Psi_1(r)]$ in 

1296: $\M^{W_1(r)}\bigl(a_i^{(1)}, a_j(\epsilon)\bigr)$

1297: defines the following geometric limits on 

1298: $ V_r\times \R \cup {\cal V}(r) \cup \tilde {\cal V}(r)$

1299: as $r\to \infty$.

1300: 

1301: \noindent 

1302:  (a). A finite energy solution $[\A',\Psi']^\epsilon$ of the perturbed

1303: Seiberg-Witten equations  on $V \times \R$, with a radial limit

1304: $a_\infty(\epsilon)$ in $\partial_\infty (\M_V^*)\subset

1305: \chi_0(T^2,V)$, and with temporal limits $[A,\psi]_1^\epsilon$ and

1306: $[\tilde A, \tilde \psi]_1^\epsilon$ in

1307: $ \partial_\infty^{-1}(a_\infty(\epsilon)) \subset \M_V^*. $

1308: 

1309: \noindent 

1310: (b). Two paths $[A(t),\psi(t)]_1^\epsilon$ in $\M_V^*$, for

1311: $t\in [-1,0)$ and $t\in (0,1]$, with

1312: \[

1313: \begin{array}{lll}

1314:  [A(-1),\psi(-1)]_1^\epsilon = [A_i^-, \psi_i^-], &\quad&

1315:  \lim_{t\to 0-} [A(t),\psi(t)]_1^\epsilon =

1316: [A,\psi]_1^\epsilon \\[2mm]

1317:  [A(1),\psi(1)]_1^\epsilon = [ A_j^+(\epsilon) ,

1318: \psi_j^+(\epsilon)], &\quad&

1319:  \lim_{t\to 0+} [A(t),\psi(t)]_1^\epsilon = [\tilde A,

1320: \tilde \psi]_1^\epsilon 

1321: \end{array}\]

1322: These paths induce a continuous, piecewise smooth path 

1323: $a_1^\epsilon(t)$ on $\partial_\infty (\M_V^*)$

1324: satisfying $ a_1^\epsilon(t)=\partial_\infty

1325: \bigl([A(t),\psi(t)]_1^\epsilon\bigr)$, with

1326: $$ a_1^\epsilon (-1)=a_{\infty,i}^-  \ \  a_1^\epsilon

1327: (0)=a_\infty(\epsilon) \ \  a_1^\epsilon (1)=

1328: a_{\infty,j}^+(\epsilon). $$ 

1329: As $\epsilon\to 0$, these geometric limits define paths

1330: $[A(t),\psi(t)]$ and $a(t)$  with

1331: $$ a(-1)=a_{\infty,i}^-, \ \  a(0)=a_\infty, \ \ 

1332: a(1)=a_{\infty,j}^+, $$

1333: in $\partial_\infty (\M_V^*)\subset

1334: \chi_0(T^2,V)$,  and $a_\infty=\lim_\epsilon a_\infty(\epsilon)$.

1335: 

1336: (c) There is  a holomorphic triangle in $H^1(T^2,\R)$ with

1337: vertices at

1338: $$ a^-_{\infty,i}, \vartheta_1, a^+_{\infty,j}(\epsilon) $$

1339: and sides given by parameterized arcs along the lines 

1340: $L_{Y_1}(\s\otimes L_0)$, $L_{Y,\mu}(\s\otimes L_0)$ and 

1341: $ \{ a_1^\epsilon(t) \}\subset \partial_\infty(\M_V^*).$

1342: Here we denote 

1343: $ \vartheta_1 = L_{Y_1}(\s\otimes L_0)\cap L_{Y,\mu}(\s\otimes

1344: L_0).$ 

1345: \label{geom:lim1} 

1346: \end{The}

1347: 

1348: This Theorem shows that the 

1349: moduli space $\M^{W_1(r)} \bigl(a_i^{(1)}, a_j(\epsilon)\bigr)$

1350: is characterized by the geometric limits on $V\times \R$ from (a) and

1351: the holomorphic triangles in (c). Two typical

1352:  holomorphic triangles

1353: for $\M^{W_1(r)} \bigl(a_i^{(1)}, a_j(\epsilon)\bigr)$

1354: or  $\M^{W_0(r)} \bigl(a_i(\epsilon), a_j^{(0)}\bigr)$

1355: are illustrated in 

1356: Figure \ref{holomorphic:triangles} where $n=4, m=k=0$ and $p=2$. 

1357: Here the points $\vartheta_i$ are the intersection points

1358: $$ \vartheta_1 = L_{Y_1}(\s\otimes L_0) \cap L_{Y,\mu}(\s\otimes

1359: L_0)$$

1360: for $W_1$ and

1361: $$ \vartheta_0 = L_{Y,\mu}(\s\otimes L_0) \cap L_{Y_0}(\s\otimes

1362: L_{nk+p}) $$

1363: in the case of $W_0$. 

1364: In other words, $\vartheta_i$ is the restriction to

1365: $T^2=\partial 

1366: \nu=\partial \tilde \nu$ of the unique reducible point $\theta_{S^3}$

1367: at the puncture in the cobordism.

1368: 

1369: \begin{figure}[ht]

1370: \epsfig{file=ex5fig2.eps,angle= 0}

1371: \caption{Holomorphic triangles for $W_1(r)$ and $W_0(r)$}

1372: \label{holomorphic:triangles}

1373: \end{figure}

1374: 

1375: 

1376: 

1377: In order to glue back the

1378: geometric limits and the holomorphic triangles, there are some

1379: admissible conditions for the geometric limits on $V\times \R$,

1380: which are studied in section 6.1,  section 6.2 and 

1381: section 6.3 \cite{MW3}. 

1382: As in section 6 \cite{MW3}, assuming that the element 

1383: $a_\infty$ is away from the bad

1384: points in the character variety of $T^2$, we denote by

1385: $\hat\M_{V\times \R}(a_\infty)$ the {\em balanced energy} moduli space

1386:  of the Seiberg-Witten equations on $V\times \R$ with 

1387: asymptotic value $a_\infty$ in the radial direction. Then we have

1388: $$ \hat \M_{V\times \R}(a_\infty) = \bigcup_{[A,\psi],[\tilde A,\tilde

1389: \psi]}\hat \M_{V\times \R}( [A,\psi],[\tilde A,\tilde

1390: \psi], a_\infty), $$

1391: with 

1392: $$ [A,\psi],[\tilde A,\tilde\psi] \in \partial_\infty^{-1}(a_\infty)

1393: \subset \M^*_V. $$  

1394: Each moduli space 

1395: $$ \M_{V\times \R}( [A,\psi],[\tilde A,\tilde

1396: \psi], a_\infty), $$

1397: for a fixed choice of $[A,\psi]$ and $[\tilde A,\tilde

1398: \psi]$ in $\partial_\infty^{-1}(a_\infty)$ in $\M_V^*$, is a smooth

1399: finite dimensional oriented manifold of the expected dimension, where

1400: the orientation is given by the corresponding determinant line bundle

1401: of the linearization operator for the monopole equations on 

1402: $V\times \R$.

1403: 

1404: Remember that we are only considering the components with minimal

1405: energy or minimal dimension among all the moduli spaces

1406: of finite energy monopoles with fixed asymptotic values.

1407: Recall that there is a notion of admissible triples

1408: (cf. Definition 6.4 \cite{MW3}),  which

1409: singles out those elements $([A,\psi],[\tilde A,\tilde \psi], a_\infty)$

1410: which arise as part of the geometric limits of solutions in

1411: $\M^{W_1(r)}(a_1,a)$  

1412: (or $\M^{W_0(r)}(a_1,a_0)$, or $\M_{Y(r)\times \R}(a,b)$ etc).

1413: For example, in the case of $\M^{W_1(r)}(a_1,a)$, with

1414: $$ a_1=[(A^-,\psi^-)\#(a^-,0)] $$

1415: $$ a=[(A^+,\psi^+)\#(a^+,0)], $$ 

1416:  an admissible triple  $([A,\psi],[\tilde A,\tilde \psi], a_\infty)$  

1417: must satisfy the following conditions:

1418: there exists a smooth regular

1419: parameterization $a(t)$, for $t\in [-1,1]$ of the path in

1420: $\partial_\infty(\M_V^*)$ connecting $a^-$ and $a^+$, such that

1421: $a(0)=a_\infty$, and corresponding smooth paths $[A(t),\psi(t)]$ in

1422: $\M_V^*$, for $t\in [-1,0)$ and $t\in (0,1]$, satisfying  

1423: $\partial_\infty[A(t),\psi(t)]=a(t)$, and with

1424: $$ [A(-1),\psi(-1)]=[A^-,\psi^-] \ \ \hbox{ and

1425: } \ \ \lim_{t\to

1426: 0_-}[A(t),\psi(t)]=[A,\psi] $$

1427: $$ \lim_{t\to 0_+}[A(t),\psi(t)]=[\tilde A,\tilde \psi] \ \ \hbox{ and

1428: } \ \ [A(1),\psi(1)]=[A^+,\psi^+]. $$

1429: By the results of \cite{MW3}, we know that the possible choices of

1430: $a_\infty$ and of the admissible data in $\hat\M_{V\times

1431: \R}(a_\infty)$ are uniquely determined by the inequivalent holomorphic 

1432: triangles $\Delta$ with vertices

1433: $\{ a^-,\vartheta_1, a^+ \}$ and sides along the union of

1434: Lagrangians $\ell \cup \ell_1 \cup \ell_\mu$, with $\ell$ defined by

1435: the asymptotic values $\partial_\infty(\M_V^*)$. We denote by

1436: $\Xi_{W_1} (a_1, a)$ the set of such inequivalent holomorphic

1437: triangles. 

1438: Moreover, under the identification of the choice of admissible triples 

1439: $([A,\psi],[\tilde A,\tilde \psi], a_\infty)$

1440: with the choice of inequivalent oriented holomorphic triangles in 

1441: $\Xi_{W_1} (a_1, a)$,  the gluing map

1442: gives an orientation preserving diffeomorphism 

1443: \ba \label{glue:W_1}

1444: \#_{W_1}: 

1445: \bigcup_{([A,\psi],[\tilde A,\tilde \psi], a_\infty)\in \Xi_{W_1} (a_1, a)}

1446: \M_{V\times\R}( [A,\psi],[\tilde A,\tilde \psi], a_\infty)

1447:  \to

1448: \M^{W_1}(a_1,a). 

1449: \na 

1450: 

1451: Similarly, there are orientation preserving diffeomorphisms

1452: given by the gluing maps for $W_0(r)$, $(Y(r)\times \R, \s\otimes L_0)$,

1453: $(Y_1(r)\times \R, \s\otimes L_0)$ and $(Y_0(r)\times \R, \s\otimes

1454: L_{nk+p})$ defined over the set of admissible triples, which are in turn 

1455: determined by the corresponding  inequivalent oriented holomorphic triangles

1456: or holomorphic discs. The corresponding sets

1457: are denoted by 

1458: $\Xi_{W_0}(a, a_0)$, $\Xi_{Y}(a, b)$,  $\Xi_{Y_1}(a_1, b_1)$

1459: and  $\Xi_{Y_0}(a_0, b_0)$, respectively. 

1460: We summarize all the gluing theorems for the

1461: geometric limits and holomorphic triangles (discs) as follows:

1462: 

1463: \begin{The} \label{final:glue}

1464: Suppose given a pair of monopoles $a^{(1)}, b^{(1)}$ in

1465: $\M_{Y_1}(\s\otimes L_0)$, and a pair of monopoles

1466: $a(\epsilon), b(\epsilon)$ in $\M_{Y_1, \mu}(\s\otimes L_0)$,

1467: where the surgery perturbation $\mu$ determined by $f_0'$ depends on a 

1468: small parameter $\epsilon$. Suppose given a pair of monopoles 

1469: $a^{(0)}, b^{(0)}$ in $\M_{Y_0}(\s\otimes L_{nk+p})$

1470: for some $k\in \Z$ and $p\in \{0, \cdots, n-1\}$. Then, for

1471: sufficiently large $r$,

1472: the gluing maps give the following

1473: orientation preserving diffeomorphisms

1474: \[\begin{array}{l}

1475: \bigcup_{\Delta\in \Xi_{Y} (a(\epsilon), 

1476: b(\epsilon))}

1477: \M_{V\times\R}( \Delta)

1478: \stackrel{\#_{Y(r)\times \R}}{\longrightarrow}  

1479: \M_{Y\times\R} (a(\epsilon), b(\epsilon)),

1480: \\

1481: \bigcup_{\Delta\in \Xi_{Y_1}(a^{(1)}, 

1482: b^{(1)})}

1483: \M_{V\times\R}( \Delta)

1484:  \stackrel{\#_{Y_1(r)\times \R}}{\longrightarrow} \M_{Y_1\times\R} (a^{(1)}, 

1485: b^{(1)}),

1486: \\

1487: \bigcup_{\Delta\in \Xi_{Y_0

1488: }(a^{(0)}, b^{(0)})}\M_{V\times\R}( \Delta)

1489:  \stackrel{\#_{Y_0(r)\times \R}}{\longrightarrow} 

1490: \M_{Y_0\times\R} (a^{(0)}, b^{(0)}),

1491: \\

1492: \bigcup_{\Delta\in \Xi_{W_1} (a^{(1)}, 

1493: a(\epsilon))}

1494: \M_{V\times\R}( \Delta)

1495: \stackrel{\#_{W_1}}{\longrightarrow} \M^{W_1}(a^{(1)}, 

1496: a(\epsilon)),

1497: \\

1498: \bigcup_{\Delta\in \Xi_{W_0} 

1499: (a(\epsilon), a^{(0)})}

1500: \M_{V\times\R}( \Delta)

1501: \stackrel{\#_{W_0}}{\longrightarrow} \M^{W_0} (a(\epsilon), a^{(0)}),

1502: \end{array}\]

1503: where, for simplicity, we denoted a triple $([A,\psi],[\tilde A,\tilde

1504: \psi], a_\infty)$ by $\Delta$. 

1505: \end{The}

1506: 

1507: 

1508: 

1509: 

1510: 

1511: \section{Proof of exactness and the surgery triangle}

1512: 

1513: In this section, we will prove the main theorem of this paper. Notice

1514: that we only prove the refined exact triangle in Theorem

1515: \ref{main:theorem} for $m=0$. The arguments for this case can be

1516: adapted to give an analogous proof for $m\neq 0$.

1517: 

1518: \subsection{The chain homomorphisms}

1519: 

1520: Recall that the moduli spaces for $W_1$ and $W_0$ which we consider

1521: are only the components of minimal energy (or dimension). They are

1522: smooth and oriented manifolds of the expected dimension, and they can

1523: be compactified according to Proposition \ref{compactification}. In

1524: particular, whenever one such moduli space is 0-dimensional, we have a 

1525: counting of points with the orientation. Thus,

1526: as in \cite{MW3}, we can define the chain homomorphisms

1527: between the Floer chain groups of $(Y, \mu, \s\otimes L_0)$,

1528: $(Y_1,  \s\otimes L_0)$ and $(Y_0, \s\otimes L_{nk+p})$ as follows.

1529: 

1530: \begin{Def}

1531: We define the map $w_*^1: C_*(Y_1)\to C_*(Y,\mu)$ by assigning the

1532: matrix elements 

1533: $$ \la a , w_*^1(a_1) \ra =\# \M^{W_1} (a, a_1),$$

1534: where the right hand side is a counting of points with the orientation

1535: if $\M^{W_1} (a, a_1)$ is 0-dimensional and non-empty, and  it is 

1536: $\la a , w_*^1(a_1) \ra =0$ otherwise. 

1537: Similarly, we define the map 

1538: $w_*^0: C_*(Y,\mu)\to \oplus_k C_{(*)}(Y_0, \s\otimes L_{nk+p})$,

1539: with the the matrix coefficients 

1540: $$ \la a_0 , w_*^0(a) \ra =\# \M^{W_0}(a, a_0) $$

1541: if $\M^{W_0}(a, a_0))$ is 0-dimensional and non-empty, and  

1542: $ \la a_0 , w_*^0(a) \ra =0$ otherwise.

1543: Here the chain groups $C_{(*)}(Y_0, \s\otimes L_{nk+p})$ are 

1544: equipped with the lifting of the relative

1545: grading on $\M_{Y_0}(\s\otimes L_{nk+p})$ as described in 

1546: Proposition \ref{relative:grading1}.

1547: \label{w10}

1548: \end{Def} 

1549: 

1550: Notice that on $W_1$ there is a $\Z$-family of 

1551: $\spinc$ structures which agree with 

1552: $\s\otimes L_0$ when restricted to the two

1553: ends. From Lemma 4.1 in \cite{MW3}, the choice of components with 

1554: minimal energy and dimension essentially 

1555: excludes other $\spinc$ structures when

1556: we consider the moduli space $\M^{W_1}(a_1, a)$ with fixed

1557: asymptotic values $a_1 \in \M_{Y_1}^*(\s\otimes L_0)$ and

1558: $a\in \M^*_{Y, \mu}(\s\otimes L_0)$.

1559: 

1560: In the case of $b_1(Y)>0$ and $c_1(\s\otimes L_0)$ is non-torsion,

1561: when both  $\M_{Y, \mu}(\s\otimes L_0)$ and 

1562: $\M_{Y_1}(\s\otimes L_0)$ are $\Z_{2\ell}$-graded, with

1563: $2\ell$ the multiplicity of $c_1(\s\otimes L_0)$ in

1564: $H^2(Y, \Z)/\hbox{Torsion}$, then Proposition \ref{relative:grading1}

1565: ensures that there exists

1566: a compatible grading on $\M_{Y_1}(\s\otimes L_0)$

1567: and a  $\Z_{2\ell}$--lifting of the relative grading 

1568: on any $\M_{Y_0}(\s\otimes L_{nk+p})$. For those

1569: $\spinc$ structures $\s\otimes L_0$  with torsion $c_1(\s\otimes L_0)$,

1570: the corresponding Floer homology is defined

1571: to be

1572: \[

1573: HF_*^{SW}(Y, \s\otimes L_0, \Z[[t]])|{t=0},

1574: \]

1575: as described in \cite{MW4}.  Then it is easy to see that

1576: the choice of components with minimal energy and dimension makes 

1577: $w_*^1$ and $w_*^0$ well-defined.  

1578: 

1579: Then, with the help of the compactifications in

1580: Proposition \ref{compactification},

1581:  the proofs of Lemma 4.3 and Lemma 4.4 in \cite{MW3}

1582: go through without any substantial change to give the following

1583:  Lemma.

1584: 

1585: \begin{Lem} 

1586: \begin{enumerate}

1587: \item The maps $w_*^i$ are chain homomorphisms.

1588: \item Suppose given $a_1$ in $\M_{Y_1}^*(\s\otimes L_0)$

1589: and $a_0$ in $\M_{Y_0}(\s\otimes L_{nk+p})$,  such that the relative

1590: index induced from $\M_{Y, \mu}^*(\s\otimes L_0)$ as in

1591: Proposition \ref{relative:grading1} is zero, then the 

1592: composite map $w^0_* \circ w^1_*$ is given by

1593: \[

1594:  \la w^0_* \circ w^1_*(a_1), a_0 \ra =

1595: \# \M^{W} (a_1, a_0).

1596: \]

1597: Here we use the same convention on the choice of the

1598: moduli spaces for $W =W_1\#_Y W_0$.

1599: \end{enumerate}

1600: \label{chain:homo}

1601: \end{Lem}

1602: 

1603: Thus, we have obtained a sequence of

1604: chain complexes induced by the surgery

1605: cobordisms:

1606: \[

1607:  0\to C_*(Y_1, \s\otimes L_0)\stackrel{w^1_*}{\longrightarrow}

1608: C_*(Y,\mu, \s\otimes L_0)\stackrel{w^0_*}{\longrightarrow} 

1609: \oplus_{k\in \Z} C_{(*)}(Y_0,\s\otimes L_{nk+p}) \to 0,

1610: \]

1611: for any fixed $p\in \{0, \cdots, n-1\}$.

1612: We shall prove that this sequence is exact and that the corresponding

1613: exact triangle is the surgery triangle, namely the connecting

1614: homomorphism is also defined via a surgery cobordism. The gluing Theorem

1615: \ref{final:glue}

1616: for the admissible geometric limits on $V\times \R$ 

1617: and the corresponding holomorphic triangles (or discs)

1618: will play a crucial role in the proof of these statements. 

1619: 

1620: 

1621: 

1622: \subsection{Proof of exactness}

1623: We first prove  that $w^1_*$ is injective and

1624: $w^0_*$ is surjective, then we show that as in \cite{MW3},

1625: $w^0_* \circ w^1_* =0$. This, together with the specific properties of 

1626: the maps $w^1_*$ and $w^0_*$ will be sufficient to prove the exactness

1627: in the middle term of the sequence:

1628: \[ 0\to C_*(Y_1, \s\otimes L_0)\stackrel{w^1_*}{\longrightarrow}

1629: C_*(Y,\mu, \s\otimes L_0)\stackrel{w^0_*}{\longrightarrow}

1630: \oplus_{k\in \Z} C_{(*)}(Y_0,\s\otimes L_{nk+p}) \to 0. \]

1631: 

1632: We can partition the moduli space

1633: $\M_{Y, \mu} (\s\otimes L_0)$ according to the

1634: decomposition (\ref{split-mp}) of Theorem \ref{refine:decomposition}, 

1635: \[

1636: \M^*_{Y, \mu} (\s\otimes L_0) \cong 

1637:  \M^*_{Y_1}  (\s\otimes L_0) \cup \bigcup_{k\in \Z} 

1638: \M_{Y_0}  (\s\otimes L_{nk+p}),

1639: \]

1640: where $\mu$ represents the surgery perturbation determined by $f_0'$

1641: as in the proof of Theorem \ref{refine:decomposition}. Here 

1642: $f_0'$ depends on a small parameter $\epsilon >0$. That is,

1643: we identify $\M^*_{Y, \mu} (\s\otimes L_0) $ with 

1644:  a collection of points

1645: \ba

1646:  {\cal M}_{Y,\mu}^* (\s\otimes L_0)

1647: = \{ a^{(1)}_i(\epsilon) \}_{i=1,\ldots r} \cup \{

1648: a^{(0)}_j(\epsilon) \}_{j=r+1,\ldots, s }, 

1649: \label{partition}

1650: \na

1651: so that, as we let $\epsilon \to 0$, the points

1652: $\{ a^{(1)}_i(\epsilon) \}_{i=1,\ldots r} $

1653: get identified with the corresponding elements

1654: $\{ a^{(1)}_i\}_{i=1,\ldots r}$

1655: in $ \M^*_{Y_1}  (\s\otimes L_0)$ and, similarly, the points

1656: $\{ a^{(0)}_j(\epsilon) \}_{j=r+1,\ldots, s }$

1657: get identified with the corresponding elements 

1658: $\{ a^{(0)}_j \}_{j=r+1,\ldots, s }$

1659: in $ \cup_{k\in \Z} \M_{Y_0}  (\s\otimes L_{nk+p})$.

1660: 

1661: \begin{Lem}\label{inj:surj} The coefficients of the maps $w_*^1$ and

1662: $w_*^0$ satisfy

1663: \begin{enumerate}

1664: \item $ \la a^{(1)}_i(\epsilon), w_*^1(a^{(1)}_j \ra =\delta_{ij}$;

1665: \item $ \la a^{(0)}_i, w_*^0(a^{(0)}_j(\epsilon) \ra

1666: =\delta_{ij}$. \end{enumerate}\noindent

1667: Thus, $w_*^1$ is injective and $w_*^0$

1668: is surjective.

1669: \end{Lem}

1670: \begin{proof} Using the gluing Theorem \ref{final:glue}, we can 

1671: describe the moduli space  

1672: $\M^{W_1}\bigl(a_i^{(1)}, a_j^{(1)}(\epsilon)\bigr)$ for

1673: two monopoles $a_i^{(1)}, a_j^{(1)}(\epsilon)$ of

1674: relative index $0$. With our convention

1675: on the choice of components for these moduli spaces, we know

1676: that $\M^{W_1}\bigl(a_i^{(1)}, a_j^{(1)}(\epsilon)\bigr)$

1677: is zero-dimensional, if non-empty, and obtained

1678: as the gluing of the admissible geometric limits

1679: on $V\times \R$ and the corresponding holomorphic

1680: triangles. 

1681: 

1682: As we let $\epsilon \to 0$, it is easy to see that the holomorphic 

1683: triangles degenerate to certain holomorphic discs, and

1684: the admissible geometric limits

1685: on $V\times \R$ for $\M^{W_1}\bigl(a_i^{(1)}, a_j^{(1)}(\epsilon)\bigr)$

1686: are identified with the admissible geometric limits

1687: on $V\times \R$  for $\M_{Y_1\times \R}(a_i^{(1)}, a_j^{(1)})$.

1688: Since the relative index of $a_i^{(1)}, a_j^{(1)}$ is

1689: zero, $\M_{Y_1\times \R}(a_i^{(1)}, a_j^{(1)})$,

1690: being a zero-dimensional moduli space of minimal energy,

1691: is empty unless $a_i^{(1)}=a_j^{(1)}$, in which case

1692: $\M_{Y_1\times \R}(a_i^{(1)}, a_i^{(1)})$ consists of a unique

1693: solution. This proves that

1694: \[

1695: \la a^{(1)}_i (\epsilon), w_*^1(a^{(1)}_j\ra =\delta_{ij}.

1696:  \]

1697: Similarly, we obtain

1698: $\la a^{(0)}_i, w_*^0(a^{(1)}_j(\epsilon) \ra =\delta_{ij}.$\end{proof}

1699: 

1700: Now we prove the exactness in the middle term. We proceed

1701: as in \cite{MW3} to show that $w_*^0\circ w_*^1=0$, which, together

1702: with Lemma \ref{inj:surj} is

1703: sufficient to establish the exact triangle.  Again, we

1704: will use heavily the gluing theorem \ref{final:glue}

1705: to analyze the moduli spaces on the 

1706: cobordisms. The following Lemma is the direct

1707: consequence of the results of Lemma \ref{chain:homo} and Lemma

1708: \ref{inj:surj}.

1709: 

1710: \begin{Lem}

1711: Suppose given  $a^{(1)}_i\in \M_{Y_1}(\s\otimes L_0)$ and

1712: $a^{(0)}_j(\epsilon) \in \M_{Y, \mu}(\s\otimes L_0)$ which

1713: corresponds to $a^{(0)}_j \in \cup_{k\in \Z} \M_{Y_0}(\s\otimes L_{nk+p})$. 

1714: The coefficients of 

1715: the composition  map $w_*^0\circ w_*^1$ satisfy

1716: \[

1717: \la a^{(0)}_j, w_*^0\circ w_*^1 (a^{(1)}_i) \ra

1718: = \la a^{(0)}_j(\epsilon), w_*^1 (a^{(1)}_i) \ra

1719: + \la a^{(0)}_j, w_*^0 (a^{(1)}_i(\epsilon)) \ra.\]

1720: \end{Lem}

1721: 

1722: Since the coefficient $\la a^{(0)}_j(\epsilon), w_*^1 (a^{(1)}_i) \ra$

1723: is given by the counting of monopoles

1724: in $\M^{W_1} (a^{(1)}_i, a^{(0)}_j(\epsilon) )$ with the orientation,

1725: and $\la a^{(0)}_j, w_*^0 (a^{(1)}_i(\epsilon)) \ra$ is given by the counting

1726: of monopoles

1727: in $\M^{W_0} (a^{(1)}_i(\epsilon), a^{(0)}_j )$ with the orientation,

1728: the gluing Theorem \ref{final:glue}, for the admissible geometric limits

1729: on $V\times \R$ and the corresponding holomorphic

1730: triangles, yields the following Lemma (cf. Theorem 6.9 \cite{MW3}).

1731: 

1732: \begin{Lem}

1733: \label{W_1:W_0} 

1734: For small enough $\epsilon$ and large $r\ge r_0$, there 

1735: is an orientation reversing  diffeomorphism  

1736: \[   

1737: \M^{W_1(r)}\bigl(a_i^{(1)}, a_j^{(0)}(\epsilon) \bigr) 

1738: \cong \M^{W_0(r)} \bigl(a_i^{(1)}(\epsilon), a_j^{(0)}\bigr).

1739:  \] 

1740: Hence, we have $ w_*^0\circ w_*^1 =0$.

1741: \end{Lem}

1742: 

1743: The Lemma corresponds to the fact that, in the two cases, the same

1744: triangles are counted with the reverse orientation.

1745: 

1746: With these Lemmata at hand, the arguments in Section 6.5 \cite{MW3}

1747: yield the following exact triangle for any $p \in \{0, \cdots, n-1\}$:

1748: {\small

1749: \ba

1750: \diagram

1751: HF^{SW}_{*} (Y_1, \s\otimes L_0) \rto^{w^1_{*}}

1752:  & HF^{SW}_{*}(Y, \s\otimes L_0)

1753: \dlto_{w^0_{*}} \\

1754:  \bigoplus_{k\in \Z} HF^{SW}_{(*)}(Y_0, \s\otimes L_{nk+p})

1755: \uto_{\Delta_{(*)}} 

1756: & \\

1757: \enddiagram

1758: \label{triangle}

1759: \na }

1760: 

1761: After a possible shift of relative index, we obtain the general

1762: exact triangle for any $m, p \in \{0, \cdots, n-1\}$:

1763: {\small

1764: \[

1765: \diagram

1766: HF^{SW}_{*} (Y_1, \s\otimes L_m) \rto^{w^1_{*}}

1767:  & HF^{SW}_{*}(Y, \s\otimes L_m)

1768: \dlto_{w^0_{*}} \\

1769:  \bigoplus_{k\in \Z} HF^{SW}_{(*)}(Y_0, \s\otimes L_{nk+p})

1770: \uto_{\Delta_{(*)}} 

1771: & \\

1772: \enddiagram

1773: \] }

1774: This completes the proof of our main result, Theorem \ref{main:theorem},

1775: except for the claim that  the connecting homomorphisms

1776:  $\Delta_{(*)}$ are induced by the surgery cobordism

1777: connecting $Y_0$ to $Y_1$. We shall discuss this statement in the next

1778: subsection. 

1779: 

1780: 

1781: \subsection{The surgery triangle}

1782: 

1783: To give a precise description of the

1784: connecting homomorphism $\Delta_{(*)}$, 

1785: we need to study the discrepancy between 

1786: the boundary operator

1787: $\partial_Y$ of the Floer complex $C_*(Y,\s\otimes L_0, \mu)$

1788: and the operator

1789: $\partial_{Y_1}\oplus \bigoplus_{k\in \Z}

1790: \partial_{Y_0,k}$ on 

1791: $$ C_*(Y_1, \s\otimes L_0) \oplus 

1792: \bigoplus_{k\in \Z} C_{(*)}(Y_0,\s\otimes L_{nk+p}). $$

1793: Let us identify again the points of $\M_{Y,\mu}^*(\s\otimes L_0)$

1794: as in (\ref{partition}) with 

1795: $$ {\cal M}_{Y,\mu}^* (\s\otimes L_0)

1796: = \{ a^{(1)}_i(\epsilon) \}_{i=1,\ldots r} \cup \{

1797: a^{(0)}_j(\epsilon) \}_{j=r+1,\ldots, s }.$$

1798: Then we have the following Lemma which

1799: gives the connecting homomorphism $\Delta_{(*)}$.

1800: 

1801: \begin{Lem}

1802: Suppose given a cycle in $\sum_i x_i a^{(0)}_i$ in

1803: $C_{(*)}(Y_0,\s\otimes L_{nk+p})$. 

1804: The image of $\sum_i x_i a^{(0)}_i$ under the

1805: connecting homomorphism $\Delta$ is given by

1806: $$ \Delta(\sum_i x_i a^{(0)}_i)= \sum_{i, j} x_i

1807: \#\bigl(\M_{Y\times \R} (a_i^{(0)}(\epsilon), a_j^{(1)}(\epsilon))\bigr) 

1808: a_j^{(1)}.$$

1809: \end{Lem}

1810: \begin{proof} (Lemma 7.1, Lemma 7.2 in \cite{MW3}) 

1811: Using the diagram chasing,  Lemma \ref{inj:surj} and Lemma

1812: \ref{W_1:W_0}, we see that we have

1813: \[

1814:  \begin{array}{lll}

1815: && \partial_{Y,\mu}(\sum_i x_i

1816: a^{(0)}_i(\epsilon)) \\[2mm]

1817: &=& \sum_{i,j} x_i 

1818: \#\bigl(\M_{Y\times\R}(a_i^{(0)}(\epsilon),a_j^{(1)}(\epsilon))\bigr)

1819: a_j^{(1)}(\epsilon) \\[2mm] 

1820: &&- \sum_{i,j,k} x_i 

1821: \#\bigl(\M_{Y\times\R}(a_i^{(0)}(\epsilon),a_j^{(1)}(\epsilon))\bigr)

1822: \#\bigl(\M^{W_0}(a_j^{(1)}(\epsilon), a_k^{(0)}))\bigr)

1823: a_k^{(0)}(\epsilon)\\[2mm]

1824: &=& \sum_{i,j} x_i 

1825: \#\bigl(\M_{Y\times\R}(a_i^{(0)}(\epsilon),a_j^{(1)}(\epsilon))\bigr)

1826: a_j^{(1)}(\epsilon) \\[2mm] 

1827: &&+ \sum_{i,j,k} x_i

1828: \#\bigl(\M_{Y\times\R}(a_i^{(0)}(\epsilon),a_j^{(1)}(\epsilon))\bigr)

1829: \#\bigl(\M^{W_1}(a_j^{(1)}, a_k^{(0)})(\epsilon))\bigr)

1830: a_k^{(0)}(\epsilon).

1831: \end{array} 

1832: \]

1833: By comparing this expression with

1834: $$ 

1835: \begin{array}{lll} 

1836: &&w_*^1( \sum_i x_i \#\bigl( \M_{Y\times \R} (a_i^{(0)}(\epsilon),

1837: a_j^{(1)}(\epsilon))\bigr) a_j^{(1)})  \\[2mm]

1838: &=&\sum_j x_j \#\bigl( \M_{Y\times \R} (a_j^{(0)}(\epsilon),

1839: a_i^{(1)}(\epsilon))\bigr) a_i^{(1)}(\epsilon)  \\[2mm]

1840: && +\sum_{i, j, k} x_i \#\bigl(\M_{Y\times \R}(a_i^{(0)}(\epsilon),

1841: a_j^{(1)}(\epsilon))\bigr) \#\bigl(\M^{W_1}(a_j^{(1)}, a_k^{(0)}(\epsilon))

1842: a_k^{(0)}(\epsilon)\bigr), 

1843: \end{array} $$

1844: we obtain 

1845: $$ \Delta(\sum_i x_i a^{(0)}_i)= \sum_{i, j} x_i

1846: \#\bigl(\M_{Y\times \R} (a_i^{(0)}(\epsilon), a_j^{(1)}(\epsilon)\bigr)

1847: a_j^{(1)}.$$

1848: This completes the proof. 

1849: \end{proof}

1850: 

1851: In next proposition, we show that  the connecting homomorphism 

1852: in the exact sequence can also be described as a map $w_{(*)}$,

1853: induced by a surgery cobordism $\bar W_2$ connecting

1854: $Y_0$ to $Y_1$, which satisfies $W_1 \#_Y W_0 = W_2 \# \bar{ \C\P^2}$.

1855: The resulting diagram 

1856: $$ C_*(Y_1, \s\otimes L_0) \stackrel{w_*^1}{\to} C_*(Y,\s\otimes L_0,\mu) 

1857: \stackrel{w^0_*}{\to} \oplus_k 

1858: C_{(*)}(Y_0,\s\otimes L_{nk+p}) \stackrel{w_{(*)}}{\to} C_*(Y_1,

1859: \s\otimes L_0)[-1] $$

1860: is therefore a distinguished triangle, the surgery triangle.  

1861: 

1862: \begin{Pro} (Proposition 7.3 \cite{MW3}) 

1863: The connecting homomorphism $\Delta_{(*)}$

1864:  in the exact triangle is given by 

1865: the following expression,

1866: $$ \Delta_{(*)}(\sum_i x_i a^{(0)}_i)= 

1867: w_{(*)}(\sum_i x_i a^{(0)}_i),

1868: $$

1869: for any cycle $\sum_i x_i a^{(0)}_i$ in

1870: $\oplus_k  C_{(*)}(Y_0,\s\otimes L_{nk+p})$ for any

1871: fixed $p \in \{0, \cdots, n-1\},$ where

1872: $$ w_{(*)} : \oplus_k C_{(*)}(Y_0,\s\otimes L_{nk+p})

1873: \to C_*(Y_1, \s\otimes L_0)[-1]$$

1874: is  the homomorphism defined by counting solutions in the

1875: zero-dimensional components of the moduli spaces

1876: $$ \M^{\bar W_2}(a_j^{(0)},a_i^{(1)}), $$

1877: over the cobordism $\bar W_2$.

1878: \end{Pro}

1879: \begin{proof}

1880: The argument is exactly the same as in the proof of Proposition

1881:  7.3 in  \cite{MW3}, therefore we shall omit the proof here.

1882: \end{proof}

1883: 

1884: 

1885: 

1886: 

1887: 

1888: 

1889: \section{Seiberg--Witten and Casson--Walker invariant}

1890: 

1891: In this section, we derive the relation between the

1892: topologically invariant version of the Seiberg-Witten invariant and

1893: the Casson-Walker invariant for rational homology 3-spheres. Together

1894: with the equivalence between the Casson-Walker invariant

1895: and the theta invariant introduced by Ozsv\'ath and Szab\'o in

1896: \cite{OS}, our result proves their  

1897: conjecture relating the Seiberg-Witten invariant and their

1898: theta invariant.

1899: 

1900: Let Y be a rational homology 3-sphere with a smoothly embedded knot $K$

1901: representing a torsion element of order $n$ in $H_1(Y, \Z)$,

1902: \[

1903: \displaystyle{

1904: \frac {|H_1(Y, \Z)|}{|\hbox{Torsion}( H_1(Y-\nu(K),  \Z))|}}

1905: =n,

1906: \]

1907: and endowed with the canonical framing  $(m, l)$ in

1908: a fixed identification:

1909: $\nu (K)\cong D^2 \times S^1$. Let $p$ and $q$  be relatively prime

1910: integers. The Dehn surgery with coefficient $p/q \in \Q \cup\{\infty\}$

1911: on $K$ gives rise to another closed manifold $Y_{p/q}$.   

1912: 

1913: Denote by $\spinc (V)$ the set of equivalence classes of $\spinc$

1914: structures on $V= Y\backslash \nu (K)$ with trivial restriction

1915: to the boundary $T^2$. Then, for any $Y_{p/q}$, there

1916: is a surjective map:

1917: \[

1918: \iota_{Y_{p/q}}: \qquad \spinc (Y_{p/q}) \to \spinc (V),

1919: \]

1920: where, for any $\s\in \spinc (Y_{p/q})$, $\iota_{Y_{p/q}} (\s)$

1921: is given by the restriction to $V\subset Y_{p/q}$. The fiber of

1922: $\iota_{Y_{p/q}}$ is given by a cyclic group

1923: generated by the Poincar\'e dual of the core

1924: of $Y_{p/q}\backslash V$.

1925: Formally, for $\iota_{Y}(\s)$ with $\s \in \spinc (Y)$, we identify the fiber

1926: of $\iota_{Y}$  with

1927: the following set of $\spinc$ structures

1928: \[

1929: \iota_{Y}^{-1}(\iota_{Y}(\s)=\bigcup_{m=0, \cdots, n-1}

1930: \{\s\otimes L_m| c_1(L_m) = mPD([K]) \in H^2(Y, \Z) \}.

1931: \]

1932: Similarly the fiber of $\iota_{Y_{p/q}}$ is given  by

1933: \[

1934: \bigcup_{m=0, \cdots, np-1} \{\s\otimes L_m|  c_1(L_m) 

1935: = mPD([K]) \in H^2(Y_{p/q}, \Z) \},

1936: \]

1937: and the fiber of $\iota_{Y_{0}}$ is given  by

1938: \[

1939: \bigcup_{ m\in \Z}\{\s\otimes L_m| c_1(L_m) 

1940: = mPD([K]) \in H^2(Y_0, \Z) \}.

1941: \]

1942: Here we use the same notation $\s$ on $Y_{p/q}$ ($Y$, or $Y_0$)

1943: as the corresponding $\spinc$ structure obtained by gluing

1944: $\s \in \spinc (V)$ with the trivial $\spinc$ structure on $\nu(K)$

1945: by the trivial gauge transformation on $T^2$. We hope this

1946: notation will not cause any confusion. 

1947: 

1948: Assume that $V$ and $\nu(K)$

1949: are equipped with a metric with a cylindrical end modeled on $T^2$.

1950: Let $\s$ be a $\spinc$ structure on $Y$. By the

1951: result of \cite{CMW} on the moduli space of finite energy monopoles on

1952: $(V, \iota_Y (\s))$, 

1953: we know that

1954: the irreducible part, denoted $\M^*_V(\s)$, is a smooth,

1955: oriented 1-dimensional manifold. The 

1956: asymptotic values along the cylindrical end define a

1957: continuous map:

1958: \[

1959: \partial_\infty: \qquad \M^*_V(\s) \rightarrow  \chi_0 (T^2, V),

1960: \]

1961: where $\chi_0 (T^2, V)$ is a $\Z\times \Z_n$-covering of

1962: the character torus $\chi (T^2)$. Sometime it is convenient

1963: to compose the above asymptotic value map with this

1964: covering map and define a boundary value map:

1965: \ba

1966: \partial_\infty: \qquad \M^*_V(\s) \rightarrow \chi (T^2).

1967: \label{partial:infty}

1968: \na

1969: 

1970: Notice that the reducible part $\chi (V)$ of the moduli space on

1971: $(V, \iota_Y (\s))$

1972: is an embedded circle $\chi (V) \subset \chi_0 (T^2, V)$ under the

1973: asymptotic value map. This becomes a circle of 

1974: multiplicity $n$ in $\chi (T^2)$. There is a ``bad point'' 

1975: in $\chi (T^2)$, given by the flat connections such that the

1976: corresponding twisted Dirac operator has a non-trivial kernel.

1977: We can endow $\chi (T^2)$ with a coordinate system $(u, v)$ defined

1978: by the holonomy around the longitude $l$ and the meridian $m$,

1979: respectively, so that the bad point corresponds to $(u, v) =(1,

1980: 1)$. The reducible circle $\chi (V)$, with the holonomy around the

1981: longitude $l$ of order $n$, is given by $u = u(\s)$, with

1982: $u(\s) \in \{0, 2/n, \cdots, 2(n-1)/n\}$. After a suitable perturbation,

1983: and a corresponding shift of coordinates, as discussed in \cite{CMW},

1984: we can assume that the bad point does not lie on any of these $n$

1985: possible circles $u = u(\s)$ of reducibles $\chi (V)$. 

1986: 

1987:  From the result in \cite{CMW}, we know

1988: that, under the map $\partial_\infty$ in (\ref{partial:infty}), the

1989: boundary points $\partial ( \M^*_V(\s))$ are

1990: either mapped to the bad point in $\chi(T^2)$ or mapped

1991: to the reducible circle $u= u(\s)$ on $\chi(T^2)$. 

1992: 

1993: Let $\chi (\nu(K) \subset Y_{p/q})$ be the reducible circle

1994: on $\nu(K) \subset Y_{p/q}$, which maps to a closed curve on $\chi(T^2)$

1995: with slope $p/q$ in the $(u, v)$-coordinates: parallel to $pv =qu$. 

1996: Looking at the induced Spin structure on $T^2 \subset Y_{p/q}$,

1997: we know that the curve $\chi (\nu(K) \subset Y_{p/q})$ goes through

1998: $(0, 1)$ if $q$ is odd or goes through

1999: $(0, 0)$ if $q$ is even, cf.\cite{CMW}. Again, after a suitable

2000: perturbation as in (\ref{eta:pert}), and the corresponding shift of 

2001: coordinates, we can assume that this $p/q$-curve is away from the bad 

2002: point on $\chi(T^2)$ and does not meet $u= u(\s)$ 

2003: along the coordinate line $v=0$. Then we know that

2004: $u= u(\s)$ intersects $\chi (\nu(K) \subset Y_{p/q})$ 

2005: inside $\chi(T^2)$ at $p$ points, which

2006: are denoted by $\theta_1, \cdots, \theta_p$, ordered according the

2007: orientation of $ u= u(\s) \subset \chi (T^2)$. They can be

2008: lifted to $pn$ points in $\chi_0(T^2, V)$. We denote these points by

2009: $\theta_1^{(k)},  \cdots,  

2010: \theta_p^{(k)}$, ($k=0, 1, \cdots, n-1$)

2011: according to the order. Denote by $\theta_0$ the intersection point of 

2012: $u= u(\s)$ with $v=0$ in $\chi (T^2)$. This

2013: can be lifted to $n$-points $\theta_0^{(0)}, 

2014: \theta_0^{(1)}, \cdots, \theta_0^{(n-1)}$

2015: on $\chi_0(T^2, V)$.  

2016: Moreover, we can assume that the map $\partial_\infty$ in

2017: (\ref{partial:infty}) 

2018: is transverse to the curves $u= u(\s)$, $v=0$ and

2019: $\chi (\nu(K) \subset Y_{p/q})$, by a suitable perturbation of the

2020: Seiberg-Witten equations on $V$ as in \cite{CMW}. We can also assume

2021: that the image   

2022: $\partial_\infty\M^*_V(\s)$ does not meet the points $\theta_0,

2023: \theta_1, \cdots \theta_p$ in $\chi (T^2)$, again

2024: by suitable perturbation, as discussed in \cite{CMW}.

2025: 

2026: Let $I$ be any open interval in

2027: \[

2028:  \chi (V) = \{ u = u(\s)\} \subset \chi_0(T^2, V).

2029: \]

2030: We denote by $SF_\C (\dirac^V_I)$

2031: the complex spectral flow of Dirac operator

2032: on $V$ twisted with the path of reducible connections $I$ on $V$. From  

2033: the analysis in \cite{CMW} and \cite{MW1},

2034: we know that 

2035: \ba

2036: \# \bigl( \partial_\infty|_{\partial \M^*_V(\s)} \bigr)^{-1}

2037: (I) = SF_\C (\dirac^V_I).

2038: \label{boundary:on:I}

2039: \na

2040:    For convenience, we define

2041: \ba

2042: SF_\C (\dirac^V_{[\theta_i, \theta_j]}) = \sum_{k=0}^{n-1}

2043: SF_\C (\dirac^V_{[\theta_i^{(k)}, \theta_j^{(k)}]}).

2044: \label{spectralflow:sum}

2045: \na

2046: 

2047: With this notation understood, we can state the following

2048: proposition relating the Seiberg-Witten invariants on $Y_{p/q}$,

2049: $Y$ and $Y_0$. 

2050: 

2051: \begin{Pro}\label{p/q-formulae:SW} 

2052: Consider generic compatible small perturbations of the Seiberg-Witten

2053: equations 

2054: on $Y_{p/q}$, $Y$ and $Y_0$, such that the map $\partial_\infty$ as in

2055: (\ref{partial:infty}) 

2056: is transverse to the curves $u= u(\s)$, $v=0$ and

2057: $\chi (\nu(K) \subset Y_{p/q})$ and misses the points

2058: $\theta_0, \theta_1, \cdots, \theta_p$ in $\chi (T^2)$. Then

2059: we have the following relation:

2060: \[\begin{array}{lll}

2061: &&\sum_{k=0}^{pn-1} SW_{Y_{p/q}} (\s\otimes L_k, g_{Y_{p/q}})\\[2mm]

2062: &=& p\sum_{k=0}^{n-1} SW_{Y} (\s\otimes L_k, g_{Y}) +

2063: q \sum_{k\in \Z} SW_{Y_0} (\s\otimes L_k)\\[2mm]

2064: && + \sum_{i=1}^p SF_\C (\dirac^V_{[\theta_0, \theta_i]}).

2065: \end{array}

2066: \]

2067: \end{Pro}

2068: \begin{proof} By the gluing theorem for 3-dimensional monopoles as in

2069: \cite{CMW}, we have 

2070: \[

2071: \bigcup_{k=0}^{pn-1} \M^*_{Y_{p/q}} (\s\otimes L_k) 

2072: = \M^*_V (\s) \times_{\chi (T^2)} \chi (\nu(K) \subset Y_{p/q}),

2073: \]

2074: where $\chi (\nu(K) \subset Y_{p/q})$ is the $p/q$-curve on 

2075: $\chi (T^2)$. Thus, we obtain 

2076: \ba

2077: \sum_{k=0}^{pn-1} SW_{Y_{p/q}} (\s\otimes L_k, g_{Y_{p/q}})

2078: = \# \bigl( \M^*_V (\s) \times_{\chi (T^2)} \chi (\nu(K) \subset Y_{p/q})

2079: \bigr).

2080: \label{p:q}

2081: \na

2082: Notice that the set 

2083: $

2084: \{\theta_1^{(k)}, \cdots, \theta_p^{(k)}: k = 0, 1, \cdots, n-1 \}

2085: $

2086: consists of  the unique reducible monopole for each 

2087: $(Y_{p/q}, \s\otimes L_k)$. 

2088: 

2089: 

2090: Similarly, we have

2091: \ba

2092: \sum_{k=0}^{n-1} SW_{Y} (\s\otimes L_k, g_{Y})

2093: =  \# \bigl( \M^*_V (\s) \times_{\chi (T^2)} \{ v=0\}\bigr).

2094: \label{Y}

2095: \na

2096: Here the reducible set consists of

2097: $

2098: \{ \theta_0^{(0)}, \theta_0^{(1)}, \cdots, \theta_0^{(n-1)} \}.

2099: $

2100: 

2101: In order to avoid the circle of reducibles on $(Y_0, \s \otimes L_0)$,

2102: we need to introduce a small perturbation such that

2103: $\chi (\nu(K)\subset Y_0)$ on $\chi (T^2)$ is a small parallel

2104: shifting of $u = u(\s)$ such that the bad point is not

2105: contained in the narrow strip bounded by these

2106: two parallel curves. We denote this small shift

2107: of $u = u(\s)$ by $u = u(\s) +\eta$, where $\eta$ is a sufficiently small

2108: positive number. This can be achieved by a perturbation of the

2109: equations as in \cite{CMW}. Then we have 

2110: \ba

2111: \sum_{k\in \Z}  SW_{Y_0} (\s\otimes L_k) 

2112: = \# \bigl( \M^*_V (\s) \times_{\chi (T^2)} \{ u=  u(\s) +\eta\}\bigr).

2113: \label{Y_0}

2114: \na

2115: 

2116: In order to compare the three countings in (\ref{p:q}) -- (\ref{Y_0}), we

2117: need to choose an oriented 2-chain $ C$ in $\chi (T^2)$ whose

2118: boundary 1-chain is given by 

2119: \[\begin{array}{lll}

2120: &&\chi (\nu(K)\subset Y_{p/q}) - p \chi (\nu(K)\subset Y) -q

2121: \chi (\nu(K)\subset Y_0) \\[2mm]

2122: &=& \chi (\nu(K)\subset Y_{p/q}) - p \{ v= 0\} 

2123: - q\{ u= u(\s) +\eta\},

2124: \end{array}

2125: \]

2126: and such that $C$ does not contain the bad point in $\chi (T^2)$. Then, 

2127: counting the boundary points of $\partial_\infty^{-1}(C)$, as a

2128: 0-chain, we obtain

2129: \ba

2130: \begin{array}{lll}

2131: &&\# \bigl( \partial_\infty^{-1} (\chi (\nu(K)\subset Y_{p/q}) \bigr)

2132: \\[2mm]

2133: &=& p \# \bigl( \partial_\infty^{-1} (\{ v= 0\}) \bigr)

2134:  + q  \# \bigl( \partial_\infty^{-1}(\{ u= u(\s) +\eta\}) \bigr)\\[2mm]

2135: && + \# \bigl( \partial_\infty\mid_{\partial (\M^*_V(\s)) }\bigr)^{-1} (C).

2136: \label{count:C}

2137: \end{array}

2138: \na

2139: 

2140: As $C$ does not contain the bad points, we know

2141: that the possible points of $\partial_\infty (\partial (\M^*_V(\s)) )

2142: \cap C$ all lie on the curve $u = u(\s)$, away from

2143: the points $\theta_0, \theta_1, \cdots, \theta_p$. It is easy to see

2144: that $C$ covers the intervals of $u = u(\s)$ between two consecutive

2145: points $\theta_i$ with different multiplicities:

2146: the multiplicities are 

2147: $ p, p-1, \cdots, 1, 0$,

2148: for the intervals 

2149: \[

2150: [\theta_0, \theta_1], [\theta_1, \theta_2],

2151: \cdots, [\theta_{p-1}, \theta_p], [\theta_p, \theta_0],

2152: \]

2153: respectively. 

2154: By the identity (\ref{boundary:on:I}) and the definition 

2155: (\ref{spectralflow:sum}), we know that 

2156: \ba

2157: \# \bigl( \partial_\infty|_{\partial (\M^*_V(\s))} \bigr)^{-1}

2158:   (C) = \sum_{i=1}^p SF_\C (\dirac ^V_{[\theta_0, \theta_i]}).

2159: \label{count:u(s)}

2160: \na

2161: Combining all the identities in (\ref{p:q}), (\ref{Y}),

2162: (\ref{Y_0}),  (\ref{count:C}) and (\ref{count:u(s)}),

2163: we obtain the proof of the proposition.

2164: \end{proof}

2165: 

2166: The Seiberg-Witten invariant for any rational homology 3-sphere

2167: depends on metric and perturbation (cf.\cite{MW1}). We now consider 

2168: the correction term (\ref{correction:term}) as defined in the

2169: introduction. We have the following proposition relating the

2170: correction terms for $Y_{p/q}$ and $Y$.

2171: 

2172: \begin{Pro}\label{p/q-formulae:correction}

2173: \begin{enumerate}

2174: \item For any rational homology 3-sphere $Y$ with a $\spinc$ structure

2175: $\s$ and a Riemannian metric $g_Y$, 

2176: \[

2177: \hat{SW}_Y (\s) = SW_Y (\s, g_Y) - \xi (\s, g_Y)

2178: \]

2179: is a well-defined topological invariant.

2180: \item For any relatively prime integers $p$ and $q$, a positive integer $n$,

2181: and $u\in \{0, 2/n,\ldots, 2(n-1)/n \}$, we have that

2182:  \[

2183: \sum_{k=0}^{pn-1} \xi_{Y_{p/q}} (\s\otimes L_k, g_{Y_{p/q}})

2184: - p \sum_{k=0}^{n-1} \xi_{Y} (\s\otimes L_k, g_{Y}) -

2185: \sum_{i=1}^p SF_\C (\dirac ^V_{[\theta_0, \theta_i]}) \]

2186:  is independent of the 

2187: manifold $Y$ and depends only on $p, q, n$, and

2188: $u(\s) \in \{0, 2/n, \ldots, 2(n-1)/n\}.$

2189: \end{enumerate}

2190: \end{Pro} 

2191: \begin{proof} Claim (1) follows from the wall-crossing formulae

2192: in \cite{MW1} and the Atiyah-Patodi-Singer index theorem. 

2193: The proof of claim (2) is analogous to the proof of Proposition

2194: 7.9 in \cite{OS}. We adapt their arguments to our situation.

2195: We write the standard surgery cobordism between

2196: $S^3$ and the Lens space $L(p, q)$ as

2197: \[

2198: W(S^3, L(p, q)) = \bigl( [0, 1] \times S^1\times D^2 \bigr) 

2199: \cup_{ [0, 1] \times S^1 \times S^1} X_{p/q},

2200: \]

2201: Then the surgery cobordism  between $Y$ and $Y_{p/q}$

2202: can be identified as 

2203: \[

2204: W_{p/q} =  \bigl( [0, 1] \times V\bigr) 

2205: \cup_{ [0, 1] \times S^1 \times S^1} X_{p/q}.

2206: \]

2207: We fix a metric on $W_{p/q}$ which respects the product structure

2208: $[0, 1] \times V$ and $[0, 1] \times S^1 \times S^1$,

2209: and agrees with $g_Y$ and $g_{Y_{p/q}}$ on the boundaries

2210: $Y$ and $Y_{p/q}$, respectively.

2211: 

2212: For a $\spinc$ structure  $\s\otimes L_i^{(m)}$

2213: in $\{\s\otimes L_k: k=0, \cdots, pn-1\}$ on

2214: $Y_{p/q}$, whose reducible monopole corresponds

2215: to $\theta_i^{(m)}$ (with $i\in \{1, \cdots, p\}$ and

2216: $m\in \{ 0, \cdots n-1\}$), we consider the $\spinc$ structure

2217: $\s\otimes L_m$ on $Y$ whose  reducible monopole is

2218: $\theta_0^{(m)}$. Then we claim that 

2219: \ba\label{xi:m:i}

2220: \xi_{Y_{p/q}} (\s\otimes L_i^{(m)}, g_{Y_{p/q}}) 

2221: - \xi_{Y} (\s\otimes L_m, g_Y) -SF_\C

2222: (\dirac^V_{[\theta_0^{(m)}, \theta_i^{(m)}]})\na

2223: is independent of $Y$ and depends only on $p, q, n$ and on

2224: $u(\s) \in \{0, 2/n, \cdots, 2(n-1)/n\}.$

2225: 

2226: To prove this claim, we choose a $\spinc$ structure $\tilde \s$

2227: on $W_{p/q}$ whose restriction to $Y$ and $Y_{p/q}$ is given by

2228: $\s\otimes L_m$ and $\s\otimes L_i^{(m)}$, respectively, and such that

2229: $c_1(\tilde\s)^2 =1$. On $(W_{p/q}, \tilde \s)$, we choose

2230: a connection $A$, whose restriction to

2231: $V\times [0, 1]$ is the path of reducibles connecting

2232: $\theta_0^{(m)}$ to $\theta_i^{(m)}$ along the curve

2233: $\chi (V) \subset \chi_0(T^2, V)$. Then we have

2234: \ba\begin{array}{lll}

2235: && \xi_{Y_{p/q}} (\s\otimes L_i^{(m)}, g_{Y_{p/q}}) 

2236: - \xi_{Y} (\s\otimes L_m, g_Y) \\[2mm]

2237: &=& Ind_\C (\Dirac_A^{W_{p/q}}) - \displaystyle {

2238: \bigl( \frac {c_1(\tilde \s)^2 -\sigma (W_{p/q}) }{8}\bigr)}\\[2mm]

2239: &=& Ind_\C (\Dirac_A^{W_{p/q}})\\[2mm]

2240: &=& Ind_\C (\Dirac_A^{[0,1]\times V}) + Ind_\C (\Dirac_A^{X_{p/q}})

2241: \end{array}\label{splitting}

2242: \na

2243: where the third equality follows from the

2244: splitting principle for the index, as the

2245: Dirac operator has no kernel on the various boundaries and

2246: corners \cite{CLM}. Notice that we have

2247: \[

2248: Ind_\C (\Dirac_A^{[0,1]\times V}) =

2249: SF_\C (\dirac^V_{[\theta_0^{(m)}, \theta_i^{(m)}]}),

2250:  \]

2251: and the connection $A|_{X_{p/q}}$ extends to connection $A_0$

2252: on $W(S^3, L(p, q))$ by a flat connection, whose index on

2253: $[0,1] \times S^1 \times D^2$ satisfies

2254: \[

2255: Ind_\C (\Dirac_{A_0}^{[0,1] \times S^1 \times D^2} )=0.

2256: \]

2257: In fact, we can choose the metric on $W(S^3, L(p, q))$ with

2258: a positive scalar curvature metric on $[0,1] \times S^1 \times D^2$.

2259: Therefore, we have

2260: \[

2261: Ind_\C (\Dirac_A^{X_{p/q}}) = 

2262: Ind_\C (\Dirac_{A_0}^{W(S^3, L(p, q))}),

2263: \]

2264: which depends only on $p, q, n$ and $u(\s)$, and so does the quantity

2265: \ba

2266: \label{splitting2}

2267:  \xi_{Y_{p/q}} (\s\otimes L_i^{(m)}, g_{Y_{p/q}})

2268: - \xi_{Y} (\s\otimes L_m, g_Y)-

2269: SF_\C (\dirac^V_{[\theta_0^{(m)}, \theta_i^{(m)}]}).

2270: \na 

2271: When summing the identity (\ref{splitting2})

2272: over $i\in \{1, \cdots, p\}$ and $m\in \{0, \cdots, n-1\}$, 

2273: notice that the term $\xi_{Y} (\s\otimes L_m, g_Y)$ is independent of 

2274: $i\in \{1, \cdots, p\}$, 

2275: hence we obtain the proof of the claim (2) by using the definition 

2276: (\ref{spectralflow:sum}).  

2277: \end{proof}

2278: 

2279: With these two propositions in place, we now have the following surgery

2280: formula for the modified version of the Seiberg-Witten invariant.

2281: 

2282: \begin{The}\label{p/q-formulae:hatSW}

2283: Given any two relatively prime integers $p$ and $q$, a positive

2284: integer $n$ and $u\in \{0, 2/n, 2(n-1)/n\}$, there is a rational valued

2285: function $s(p, q, n, u)$, depending only on $p, q, n$ and $u$,

2286: satisfying the

2287: following property. Let $Y$ be a rational homology 3-sphere with a smoothly

2288: embedded knot and a canonical framing $(m, l)$

2289: such that $\nu (K)\cong D^2\times S^1$. Assume that

2290: $K$ represents a torsion element of order $n$ in $H_1(Y, \Z)$. Let

2291: $\s$ be a $\spinc$ structure on $Y$. Then we have

2292: \[

2293: \begin{array}{lll}

2294: &&\sum_{k=0}^{pn-1} \hat{SW}_{Y_{p/q}} (\s\otimes L_k)\\[2mm]

2295: &=& p\sum_{k=0}^{n-1} \hat{SW}_{Y} (\s\otimes L_k) +

2296: q \sum_{k\in \Z} SW_{Y_0} (\s\otimes L_k)\\[2mm]

2297: && + s(p, q, n, u). \end{array}

2298: \]

2299: \end{The}

2300: \begin{proof} Following from Proposition \ref{p/q-formulae:correction}, 

2301: we know that

2302: \ba

2303: \sum_{k=0}^{pn-1} \xi_{Y_{p/q}} (\s\otimes L_k, g_{Y_{p/q}})

2304: -p \sum_{k=0}^{n-1} \xi_{Y} (\s\otimes L_k, g_{Y}) -

2305: \sum_{i=1}^p SF_\C (\dirac ^V_{[\theta_0, \theta_i]})

2306: \label{s(p,q,n,u)}

2307: \na

2308:  depends only on $p, q, n$ and

2309: $u=u(\s) \in \{0, 2/n, \cdots, 2(n-1)/n\}$. We denote this term by

2310: $s(p, q, n, u)$. By subtracting (\ref{s(p,q,n,u)}) from the surgery

2311: formula for the Seiberg-Witten invariants in Proposition  

2312: \ref{p/q-formulae:SW}, we obtain the proof of this theorem.

2313: \end{proof}

2314: 

2315: Now we can establish the equivalence between

2316: the modified version of the Seiberg-Witten invariant $\hat{SW}$

2317: and the Casson-Walker invariant for rational homology 3-spheres.

2318: 

2319: \begin{The}

2320: For any rational homology 3-sphere $Y$, we have

2321: \[

2322: \sum_{\s\in \spinc (Y)} \hat {SW}_Y (\s) 

2323: = \displaystyle{\frac 12} |H_1(Y, \Z)| \lambda (Y)

2324: \]

2325: where $\lambda (Y)$ is the Casson-Walker invariant.

2326: \end{The}

2327: \begin{proof}

2328: We first derive the surgery formula for the invariant

2329: $\sum_{\s\in \spinc (Y)} \hat {SW}_Y (\s)$ from 

2330: Theorem \ref{p/q-formulae:hatSW} and the Seiberg-Witten invariant for

2331: $Y_0$ (a rational homology $S^1\times S^2$, i.e., $b_1(Y_0)=1$) (see

2332: \cite{MT} \cite{Don}):

2333: \ba\label{surgery:formuale:SW}

2334: \begin{array}{lll}

2335: &&\sum_{\s\in \spinc (Y_{p/q})} 

2336: \hat{SW}_{Y_{p/q}} (\s)\\[2mm]

2337: &=& p\sum_{\s\in \spinc (Y)} \hat{SW}_{Y} (\s) +

2338: q \sum_{j=0}^\infty a_j j^2 + |H_1(Y, \Z)| s(p, q, n) \end{array}

2339: \na

2340: where $s(p, q, n)= \sum_u s(p, q, n, u)/n$ and $a_j$ is the coefficient

2341: of the symmetrized Alexander polynomial of $Y_0$,

2342: \[

2343: A(t) = |\hbox{Torsion} (H_1(Y_0, \Z))| + \sum_{j=1}^\infty a_j (t^j + t^{-j})

2344: \]

2345: normalized such that 

2346: \[

2347: A(1) =  |\hbox{Torsion} (H_1(Y_0, \Z))|.

2348: \]

2349: Set $\bar \lambda (Y) = \displaystyle{\frac 12}|H_1(Y, \Z)| \lambda (Y)$

2350: as the normalized Casson-Walker invariant. Then

2351: the surgery formula in \cite{Walker}

2352: for $\bar \lambda (Y)$ can be expressed as (cf. \cite{OS}):

2353: \ba\label{surgery:formuale:CW}

2354: \begin{array}{lll}

2355: \bar \lambda (Y_{p/q}) &=& p \bar \lambda (Y) + q  \sum_{j=0}^\infty a_j j^2 

2356: \\[2mm]

2357: && + |H_1(Y, \Z)| \bigl( \displaystyle{

2358: \frac{q(n^2-1)}{12n^2} -\frac{p s(p, q)}{2}}\bigr).

2359: \end{array}\na

2360: Here $s(p, q)$ is the Dedekind sum of relatively prime integers $p$

2361: and $q$ (cf. \cite{Walker}). 

2362: Comparing (\ref{surgery:formuale:SW}) and (\ref{surgery:formuale:CW}),

2363: we only need to show that 

2364: \ba

2365: s(p, q, n) =  \displaystyle{

2366: \frac{q(n^2-1)}{12n^2} -\frac{p s(p, q)}{2}}.

2367: \label{s(p,q,n)}

2368: \na

2369: 

2370: Since $s(p,q,n)$ is independent of the manifold $Y$, we can choose

2371: some examples that can be computed explicitly, and use them to

2372: identify the coefficient $s(p,q,n)$.  

2373: The Lens space $L(p, q)$ can be obtained by a $p/q$-surgery

2374: on an unknot in $S^3$. The calculation of Nicolaescu \cite{Nic}

2375: for $L(p, q)$ gives us that 

2376: \[

2377: \sum_{\s\in \spinc (L(p, q))} \hat{SW}_{L(p, q)}(\s) = 

2378: -\displaystyle{

2379: \frac{ps(p, q)}{2}}.

2380: \]

2381: This implies that (\ref{s(p,q,n)}) holds for $n=1$. Now

2382: we can prove (\ref{s(p,q,n)}) by induction on $n$. 

2383: This is exactly the same argument as in the proof of Theorem 7.5 in

2384: \cite{OS} on the equivalence of their theta invariant and the

2385: Casson-Walker invariant. The example is the Seifert manifold

2386: $M(n, 1; -n, 1; q, -p)$, obtained by $p/q$ surgery on a knot of order $n$ in 

2387: $L(n, 1) \# \overline{L(n, 1)}$. By Kirby calculus it is possible to show

2388: that $M(n, 1; -n, 1; q, -p)$ can be obtained as

2389: $(-n)$-surgery on a knot in the Lens space $L(pn-q, q)$, and

2390: can be obtained as a sequence of surgeries on knots of order less than $n$,

2391: see the proof of Theorem 7.5 in \cite{OS} for details. 

2392: \end{proof}

2393: 

2394: 

2395: \begin{thebibliography}{999999}

2396: 

2397: \bibitem{CLM} S. Cappell, R. Lee, E. Miller,  {\em

2398:       Self-adjoint elliptic operators and manifold decompositions,

2399:      part II: Spectral flow and Maslov index.} Comm. Pure Appl. Math.,

2400:       49:869-909, 1996.

2401:  

2402: \bibitem{CMW} A. Carey, M. Marcolli, B.L. Wang, {\em Exact triangles

2403:        in Seiberg-Witten Floer theory. Part I: the geometric triangle},

2404:        preprint, math.DG/9907065.

2405: 

2406: \bibitem{Don} S. Donaldson, {\em  Topological field theories and formulae 

2407:       of Casson and Meng-Taubes, } Proceedings of the Kirbyfest 

2408:     (Berkeley, CA, 1998), 87--102 (electronic), Geom.

2409:      Topol. Monogr., 2, Geom. Topol., Coventry, 1999

2410: 

2411:  

2412: \bibitem{MW1} M. Marcolli, B.L. Wang, {\em Equivariant

2413:       Seiberg-Witten Floer theory}, preprint, dg-ga/9606003.

2414: 

2415: \bibitem{MW2} M. Marcolli, B.L. Wang, {\em Exact triangles

2416:        in Seiberg-Witten Floer theory. Part II: geometric limits of flow

2417:        lines}, preprint, math.DG/9907080.

2418: 

2419: \bibitem{MW3} M. Marcolli, B.L. Wang, {\em Exact triangles

2420:        in Seiberg-Witten Floer theory. Part III: proof of exactness},

2421:        preprint, math.DG/0009157.

2422: 

2423: \bibitem{MW4} M. Marcolli, B.L. Wang, {\em Exact triangles

2424:        in Seiberg-Witten Floer theory. Part IV: $\Z$-graded monopole

2425:        homology}, preprint, math.DG/0009159.

2426: 

2427: \bibitem{MT} G. Meng, C. Taubes, {\em $\underline{\rm SW}=$ Milnor torsion},

2428:           Math. Res. Lett. 3 (1996), no. 5, 661--674.

2429: 

2430: \bibitem{Nic} L. Nicolaescu, {\em Seiberg-Witten theoretic invariants 

2431:               of lens spaces}, preprint, math.DG/9901071.

2432: 

2433: \bibitem{OS} P. Ozsv\'ath, Z. Szab\'o, {\em The theta divisor and

2434:              the Casson-Walker invariant}, preprint,  math.DG/0006194.

2435: 

2436: \bibitem{Walker} K. Walker, {\em An extension of Casson's invariant},

2437:          Ann. Math. Studies, Vol. 126, Princeton Univ. Press, 1992.

2438: 

2439: 

2440: \end{thebibliography}

2441: 

2442: 

2443: \noindent {\bf Matilde Marcolli}\par 

2444: \noindent Max--Planck--Institut f\"ur Mathematik, Vivatsgasse 7,

2445: D-53111 Bonn, Germany. \par 

2446: \noindent marcolli\@@mpim-bonn.mpg.de

2447: 

2448: \vskip .2in

2449: 

2450: \noindent {\bf Bai-Ling Wang}\par

2451: \noindent  Department of Pure

2452: Mathematics, University of Adelaide, Adelaide SA 5005, Australia. \par

2453: \noindent bwang\@@maths.adelaide.edu.au

2454: 

2455: \end{document}

2456:    

2457: 

2458: 

2459: