1: %%%%%%%%

2: 

3: %%%%%

4: 

5: %%%%%%%%%%%%%

6: %\documentclass[10pt,oneside,amscd,amssymb,epsf]{amsart}

7: % epsf not used

8: \documentclass[10pt,oneside,amscd,amssymb,epsf]{amsart}

9: 

10: % delete 'oneside' for final version

11: 

12: % \usepackage[notcite]{showkeys}

13: \usepackage{amsthm}

14: \usepackage{graphicx}

15: \usepackage{pb-diagram, Bdiagx}

16: \usepackage{psfrag}

17: 

18: %\usepackage{showkeys}

19: 

20: % Delete the next two lines for single spaced copy.

21: 

22: %\parskip 1.75\parskip plus 3pt minus 1pt

23: %\renewcommand{\baselinestretch}{1.5}  (this is from submitted copy)

24: 

25: \newcommand{\TryPackage}[3]{\IfFileExists{#1.sty}{\usepackage{#1}

26: #2}{#3}}

27: \TryPackage{mathrsfs}{\renewcommand{\mathcal}{\mathscr }}{%

28:         % else try euler fonts

29:         \TryPackage{eucal}{}{}}

30: 

31: %\pagestyle{plain}

32: \pagenumbering{arabic}

33: %\footskip .5in

34: %\textheight 8.30in

35: %\headheight .4in

36: 

37: % switch to 6 in for final copy

38: \textwidth 6in

39: %\textwidth 4.5in

40: 

41: \textheight 8.5in

42: \evensidemargin .2in

43: \oddsidemargin .2in

44: \topmargin .25in

45: \headsep .2in

46: \headheight 0.2in

47: \footskip .5in

48: 

49: \begin{document}

50: \title[Calderon Projector for the Hessian ...]{Calderon Projector for

51: the

52: Hessian of the perturbed Chern-Simons function on a 3-manifold with

53: boundary}

54: 

55: \author{Benjamin Himpel}

56: \address{Department of Mathematics, Indiana University, Bloomington, IN

57: 47405}

58: \email{bhimpel@indiana.edu}

59: \urladdr{http://mypage.iu.edu/$\sim$bhimpel}

60: 

61: \author{Paul Kirk}

62: \address{Department of Mathematics, Indiana University, Bloomington, IN

63: 47405}

64: \email{pkirk@indiana.edu}

65: \urladdr{http://mypage.iu.edu/$\sim$pkirk}

66: 

67: \author{Matthias Lesch}

68: \address{Universit\"at zu K\"oln, Mathematisches Institut, Weyertal

69: 86--90,

70: D--50931 K\"oln}

71: \email{lesch@mi.uni-koeln.de}

72: \urladdr{http://www.mi.uni-koeln.de/$\sim$lesch}

73: 

74: \newcommand{\cormark}{\marginpar{$\Leftarrow$}}

75: \date{Feb. 18, 2003}

76: \begin{abstract} {The existence and continuity for the Calder\'on

77: projector of the perturbed odd signature  operator on a 3-manifold is

78: established. As an application we give a new proof of a result of Taubes

79: relating the mod 2 spectral flow of a family of operators on a  homology

80: 3-sphere  with the difference in local intersection numbers of the

81: character varieties  coming from a Heegard decomposition. }

82: \end{abstract}

83: 

84: \thanks{The second named author gratefully acknowledges the support of

85: the

86: National Science Foundation under   grant no. DMS-9971020.}

87: 

88: \maketitle

89: 

90: % MACROS

91: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

92: % Macros included by Matthias

93: \renewcommand{\tilde}{\widetilde}

94: \newcommand{\Cald}{Calder\'on projector}

95: \newcommand{\cV}{{\mathcal V}}

96: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

97: 

98: \newcommand{\const}{\mbox{const}}

99: \newcommand{\lto}{\longrightarrow}

100: % Greek

101: \newcommand{\al}{\alpha}

102: \newcommand{\be}{\beta}

103: \newcommand{\ga}{\gamma}

104: \newcommand{\de}{\delta}

105: \newcommand{\ep}{\epsilon}

106:        \renewcommand{\th}{\theta}

107: \newcommand{\la}{\lambda}

108: \newcommand{\La}{\Lambda}

109: \newcommand{\om}{\omega}

110: \newcommand{\si}{\sigma}

111: \newcommand{\Ga}{\Gamma}

112: \newcommand{\Om}{\Omega}

113: \newcommand{\Si}{\Sigma}

114: \newcommand{\etab}{\tilde{\eta}}

115: \newcommand{\tsig}{\tilde{\sigma}}

116: \newcommand{\hal}{\hat{\alpha}}

117: 

118: \newcommand{\grad}{\mbox{grad}}

119: \newcommand{\gra}{\mbox{grad}^\sim}

120: 

121: \newcommand\tr{\mathop {\rm tr}}

122: 

123: \newcommand{\tlam}{\widetilde \lambda}

124: 

125: % Black-board bold

126: \newcommand{\zz}{{\mathbb Z}}

127: \newcommand{\rr}{{\mathbb R}}

128: \newcommand{\cc}{{\mathbb C}}

129: \newcommand{\qq}{{\mathbb Q}}

130: \newcommand{\hh}{{\mathbb H}}

131: 

132: % Caligraphic

133: \newcommand{\cA}{{\mathcal A}}

134: \newcommand{\cU}{{\mathcal U}}

135: \newcommand{\cE}{{\mathcal E}}

136: \newcommand{\cD}{{\mathcal D}}

137: \newcommand{\cC}{{\mathcal C}}

138: \newcommand{\cL}{{\mathcal L}}

139: \newcommand{\cB}{{\mathcal B}}

140: \newcommand{\cF}{{\mathcal F}}

141: \newcommand{\cG}{{\mathcal G}}

142: \newcommand{\cM}{{\mathcal M}}

143: \newcommand{\cH}{{\mathcal H}}

144: \newcommand{\cT}{{\mathcal T}}

145: \newcommand{\cO}{{\mathcal O}}

146: \newcommand{\cP}{{\mathcal P}}

147: % Hats

148: \newcommand{\hA}{\widehat{A}}

149: \newcommand{\hB}{\widehat{B}}

150: 

151: % Tildes

152: \newcommand{\tB}{{\widetilde{\mathcal B}}}

153: \newcommand{\tM}{{\widetilde{\mathcal M}}}

154: \newcommand{\tC}{{\widetilde{C}}}

155: 

156: % Gothic

157: %\newcommand{\hh}{{\mathfrak h}}

158: %\newcommand{\hhp}{{{\mathfrak h}^\perp}}

159: % Operators

160: \newcommand{\im}{\operatorname{im}}

161: \newcommand{\Map}{\operatorname{Map}}

162: \newcommand{\hol}{\operatorname{{\it hol}}}

163: \newcommand{\proj}{\operatorname{Proj}}

164: \newcommand{\Mas}{\operatorname{Mas}}

165: \newcommand{\Spec}{\operatorname{Spec}}

166: \newcommand{\Hom}{\operatorname{Hom}}

167: \newcommand{\Hol}{\operatorname{Hol}}

168: \newcommand{\crit}{\operatorname{Crit}}

169: \newcommand{\hess}{\operatorname{Hess}}

170: \newcommand{\ind}{\operatorname{ind}}

171: \newcommand{\image}{\operatorname{image}}

172: \newcommand{\ad}{\operatorname{ad}}

173: \newcommand{\End}{\operatorname{End}}

174: \newcommand{\Sign}{\operatorname{Sign}}

175: \newcommand{\SF}{\operatorname{SF}}

176: \newcommand{\Id}{\operatorname{Id}}

177: \def\eqref#1{(\ref{#1})}

178: 

179: \newcommand{\del}{\partial}

180: 

181: %\def\image{\text{\rm image}}

182: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

183: % Makros by Matthias 2002-Jan

184: \newcommand{\chern}{\operatorname{ch}}

185: \renewcommand{\eqref}[1]{\textnormal{(\ref{#1})}}

186: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

187: 

188: %\newcommand{\bbl}{{[\![}}

189: %\newcommand{\bbr}{{]\!]}}

190: %\newcommand{\bbA}{{\bbl A \bbr}}

191: %\newcommand{\minA}{{\widehat{A}^-}}

192: %\newcommand{\maxA}{{\widehat{A}^+}}

193: %\newcommand{\minB}{{\widehat{B}^-}}

194: %\newcommand{\maxB}{{\widehat{B}^+}}

195:        \newcommand{\opX}{{\bar{X}}}

196: 

197:    \theoremstyle{definition}

198: 

199: \newtheorem{defn}{Definition}[section]

200: \theoremstyle{plain}

201: \newtheorem{lemma}[defn]{Lemma}

202: \newtheorem{thm}[defn]{Theorem}

203: \newtheorem{prop}[defn]{Proposition}

204:        \newtheorem{cor}[defn]{Corollary}

205:      \theoremstyle{definition}

206: \newtheorem{remark}[defn]{Remark}

207: 

208: \numberwithin{equation}{section}

209: 

210: \section{Introduction}

211: 

212: The purpose of this article is to establish the existence and continuity

213: of the Calder\'on projectors for families of perturbed signature

214: operators

215: on a 3-manifold with boundary.  We explain briefly the motivation.

216: 

217: The Morse theory  of the Chern-Simons function on the space of $SU(n)$

218: connections over a 3-manifold leads to the construction of topological

219: invariants, notably Taubes' construction of Casson's invariant

220: \cite{taubes}, Floer's instanton homology \cite{floer}, and more

221: recently

222: $SU(3)$ and $SU(n)$ extensions of Casson's invariant \cite{herald-boden,

223: BHK-int, CLM, herald2}.  These invariants are defined using the Morse

224: index of the Hessian of the Chern-Simons function at a critical point.

225: To

226: make this rigorous Taubes showed (in the $SU(2)$ setting) how to

227: understand this ill-defined Morse index in terms of the spectral flow of

228: the odd signature operator coupled to a path of connections. Most

229: importantly for the present article, he also showed how to construct

230: perturbations of the Chern-Simons function in order to make it suitably

231: non-degenerate.

232: 

233: Taubes' ideas form the blueprint for   generalizations of Casson's

234: invariant. In particular the definition of the generalized Casson

235: invariants involve the spectral flow of 1-parameter families of

236: perturbed

237: odd signature operators.  These perturbed odd signature operators (which

238: we describe in detail below) are compact perturbations of Dirac

239: operators,

240: but they are not differential operators.

241: 

242: With the construction of a Morse-theoretic topological invariant

243: established, the problem arises of computing the invariant to discover

244: what information it contains.  Casson's surgery formula and Floer's

245: exact

246: triangle are examples of computational formulas which make it possible

247: to

248: compute their invariants. In general one wishes to have surgery formulas

249: for the generalized Casson invariants.  This leads to the problem of

250: computing the spectral flow of the Hessian of the Chern-Simons function between

251:   critical points in terms of cut--and--paste  data for the underlying

252: 3-manifold.

253: 

254: There is a good cut--and--paste theory for the spectral flow of Dirac

255: operators which make it possible to carry out  calculations.

256: Nicolaescu's article \cite{nico} gives a splitting formula for the

257: spectral flow of a family of Dirac operators. This formula equates the

258: spectral flow to an infinite--dimensional Maslov index corresponding to

259: the

260: associated paths of Calder\'on projectors coming from two parts in a

261: decomposition of the manifold. (Roughly speaking, the \Cald\  for a

262: Dirac

263: operator $D$ is the projection in

264: $L^2(\partial X)$ onto the subspace of boundary values of solutions to

265: the equation $D\phi=0$.)

266:     Nicolaescu's result, combined with some techniques introduced in

267: \cite{daniel-kirk}, sufficed to compute $SU(3)$ Casson invariants for

268: certain Seifert fibered homology 3-spheres in

269: \cite{BHKK,CLM}, where one could avoid using perturbations (at least

270: near

271: the reducible critical points). However, the presence of singularities in the flat moduli space 

272: complicates calculations for more general

273: homology spheres, requiring the use of perturbations. To apply the powerful

274: methods of \cite{nico} one needs an extension of the results concerning

275: the existence and continuity of the Calder\'on projectors for families

276: of

277: operators more general than Dirac operators; in fact general enough to

278: include the perturbed odd signature operator. This is the task

279: accomplished in this article. Our main result is the following, which

280: combines Theorems

281: \ref{continuityCald} and \ref{Poissonop-a}.

282: 

283: \bigskip

284: 

285: \noindent{\bf Theorem.} {\em Consider the class $\cD$ of Dirac operators

286: on $E\to X$ for a manifold with boundary $X$ which are in cylindrical

287: form

288: near

289: $\partial X$,  and the class $\mathcal{V}$ of auxiliary operators

290: satisfying the pseudolocality property

291: \textnormal{\eqref{pseudolocality}}.  Then for $D\in \cD$ and $V\in

292: \mathcal{V}$ the Calder\'on projector of $D+V$ is well defined.

293: 

294: We say that a family $D_t\in\cD$ varies continuously if all coefficients

295: of $D_t$ (in any local chart) vary continuously. Moreover, we equip

296: $\mathcal{V}$ with the usual norm topology of $\cB(L^2(X,E))$. Then for

297: each

298: $n\in\mathbb{Z}_+$ the map

299: \[ \bigl\{(D,V)\in\cD\times\mathcal{V}\,\big|\, \dim Z_0(  D+

300: V)=n\bigr\}

301:       \longrightarrow \cB(L^2(\Sigma,E)), \quad (D,V)\mapsto C(  D+

302: V)

303: \] taking a pair $(D,V)$ to the  Calder\'on projector $C$  for the operator

304: $D+V$ is continuous. }

305: 

306: \smallskip

307: 

308:   In the statement, $Z_0(D+V)$   is a vector space isomorphic to $K\oplus

309: K$, where

310: $K$ denotes the space of sections $\phi$ which vanish on a collar of the

311: boundary and which satisfy  $(D+V)\phi=0$.  For Dirac operators the

312: unique continuation property implies that

313: $Z_0=0$. A novel feature of our approach is that we do not require

314: the unique continuation property to hold.

315: 

316: \bigskip

317:  

318: We are chiefly concerned with the application of this result to the

319: perturbed odd signature operator arising from the Morse theory of the

320: Chern-Simons function, as this will be  used crucially in a forthcoming

321: article \cite{BHK-pqr}. (In fact the completion of that article was held

322: up by  the need for the main result of the present article.)

323: Unfortunately

324: there is no suitably precise exposition of the various constructions

325: which

326: arise in this context and so we include  the details necessary  to

327: set up

328: the problem  precisely and also for the convenience of the reader.  One

329: of our

330: purposes  is to explicitly provide justification for comments like ``in

331: the non-generic

332: case one must perturb, but the arguments carry through,''    in the

333: context

334: of cut--and--paste techniques for spectral flow.

335: 

336: It is our hope that in addition to providing the proof of the theorem

337: stated

338: above this article complements  the  exposition \cite{Herald1, herald2}.

339: Thus in Section

340: \ref{set-up} we  explain the approach used in the Morse-theoretic study

341: of

342: the Chern-Simons function, with an emphasis on   how the perturbations

343: are constructed and on the effect  of perturbing on the Hessian of the

344: Chern-Simons function. We also include proofs of   some  (known) results

345: concerning the pseudolocality of the perturbation (Proposition

346: \ref{lem2.1}) and an estimate concerning the  size of the perturbation

347: (Lemma \ref{estimate}) that we require  to apply the result stated above

348: to the context of the  Hessian of the perturbed Chern-Simons function.

349: 

350: Theorem \ref{thm5.3} is our desired application. In this statement,

351: $D_{A,f}$ denotes the

352: perturbed twisted odd signature operator coupled to a connection $A$

353: and a perturbation $f$. It is

354: essentially the Hessian of the perturbed Chern-Simons function

355: $cs+f:\cA\to \mathbb{R}$.

356: 

357: \bigskip

358: 

359: \noindent{\bf Theorem \ref{thm5.3}.} {\em Let $X$ be a compact

360: 3-manifold

361: with boundary and $\cA$ the space of   connections in a principal

362: $SU(n)$

363: bundle over $X$. Then there exists a neighborhood

364: $U\subset \cA$ of the space of flat connections, and an

365: $\epsilon>0$  so that if

366: $B_{\cP}(0,\epsilon)$ denotes the

367: $\epsilon$ ball around $0$ in the space of perturbations $\cP$,  the map

368: $U\times B_{\cP}(0,\epsilon)\to \cB(L^2(\Sigma,E))$ sending a pair

369: $(A,f)$ to the Calder\'on projector $P_{A,f}$ of $D_{A,f}$   is

370: continuous.}

371: 

372: \bigskip

373: 

374: We finish the article   by giving a new proof in Section

375: \ref{application}, using our methods, of the main step in Taubes'

376: argument

377: that his invariant

378: equals Casson's invariant.

379: 

380: \bigskip

381: 

382: \noindent{\bf Theorem} (Taubes){\bf .} {\em The mod 2 Spectral flow of

383: the

384: perturbed odd signature operator between two generic perturbed flat

385: connections on a homology 3-sphere equals zero if and only if the

386: corresponding  local intersection numbers of the $SU(2)$ character

387: varieties of the Heegard handlebodies in the $SU(2)$ character variety

388: of

389: the Heegard surface are equal. }

390: 

391: \bigskip

392: 

393:   The strategy of our proof is to use an adiabatic stretching of a collar

394: of the Heegard surface  and symplectic reduction to reduce the

395: infinite--dimensional Maslov index (which equals the sign Taubes assigns

396: to

397: a critical point) to a finite--dimensional Maslov index in  the character

398: variety of a Heegard surface  (which determines Casson's sign). Having

399: established in  Theorem

400: \ref{thm5.3} that the Calder\'on projectors exist and vary  nicely, the

401: first step in the argument uses this and Nicolaescu's theorem

402: \cite{nico} to identify spectral flow with  the infinite--dimensional

403: Maslov index of the  Calder\'on projectors. The rest of the argument is

404: a study of the effect of stretching and a corresponding symplectic

405: reduction,  as analyzed in

406: \cite{daniel-kirk}.

407: 

408: The authors thank  H. Boden, G. Daskalopoulos, C. Herald, T. Mrowka,

409: and K.

410: Wojciechowski for helpful discussions.

411: 

412: \section{basic set-up}\label{set-up}

413: \subsection{Connections on a 3-manifold}

414: 

415: We begin by recalling the  spaces and operators of interest.

416: 

417: Let $X$ be a compact 3-manifold.  Fix a principal $SU(n)$ bundle  $P$

418: over

419: $X$. We associate two vector bundles to $P$: the $\cc^n$ vector bundle

420: $E\to X$ associated   to the standard representation

421: $SU(n)\to GL(\cc^n)$, and the Lie algebra bundle

422: $\ad P\to X$ associated to the adjoint representation $SU(n)\to

423: GL(su(n))$

424: of

425: $SU(n)$ on its Lie algebra $su(n)$.  Notice that the bundle $P$ is

426: trivializable since

427: $\pi_1(SU(n))$ and $\pi_2(SU(n))$  are trivial.

428: 

429: We let $\cA$ denote the space of smooth connections on $P$. A connection

430: $A\in \cA$ defines  a covariant derivative

431: $$d_A:C^\infty(X;E)\to \Om^1(X;E).$$ It is  convenient to view $\ad

432: P$

433: as a subbundle of  $\End(E)$. Then the covariant derivative on $\ad P$

434: associated to $A$  (which we also denote by $d_A:C^\infty(X,\ad P)\to

435: \Om^1(X;\ad P)$)  satisfies

436: $$(d_A(\phi))(e)=d_A(\phi(e))-\phi(d_A(e))\  \text{ for } \

437: \phi\in C^\infty(X;\ad P), \ e\in C^\infty(X;E).$$ The covariant

438: derivatives    extend  to $p$-forms $d_A:\Om^p(X; E)\to

439: \Om^{p+1}(X; E)$  and  $d_A:\Om^p(X; \ad P)\to

440: \Om^{p+1}(X; \ad P)$ by requiring the graded Leibnitz rule to hold. The

441: operator $d^2_A:\Omega^p(X;E)\to\Omega^{p+2}(X;E)$ has order zero and

442: is

443: given by multiplication by the   curvature form  $F(A)\in

444: \Omega^2(X;ad P)$.

445: 

446:      The space

447: $\cA$ is an affine space modeled  on  the space of Lie algebra valued

448: 1-forms

449: $\Om^1(X;\ad P)$: Fixing a connection $A_0$ yields an identification

450: \begin{equation}\label{eq2.1}\cA\xrightarrow\cong \Om^1(X;\ad P),

451: \ A\mapsto A-A_0.\end{equation} It is sometimes convenient to  fix a

452: trivialization $P\cong X\times SU(n)$; then we can take $A_0$ to be the

453: trivial (product) connection which we denote by $\Theta$. The space

454: $\cA$

455: is acted on by the   group of (smooth) gauge transformations

456: $\cG=C^\infty(X;P\times_{Ad} SU(n))$ in the usual way: $d_{gA}(e)=

457: g(d_A(g^{-1}(e)))$.

458: 

459: We fix a Riemannian metric on $X$. If the boundary of $X$ is non-empty

460: we

461: will always assume that a collar of the boundary $\del X$ in $X$ is

462: isometric to $[0,1)\times \del X$.

463: 

464: The $L^2$ inner product on

465: $\Om^i(X;\ad P)$ is then defined in terms of the Hodge $*$ operator on

466: $X$ by the formula

467: \begin{equation}\label{eq2.2}\langle b,c\rangle_{L^2}=-\int_X tr(b\wedge

468: *c).\end{equation}

469: 

470: For analytical reasons one must at times complete the spaces $\cA$ and

471: $\cG$ in appropriate Sobolev norms.

472:      Thus we fix some $s\ge 1$ and take

473: $\cA_s$ to be the

474: $L^2_s$ completion of $\cA\cong \Om^1(X;\ad P)$, and $\cG_{s+1}$ to be

475: the

476: closure of

477: $\cG$ in the

478: $L^2_{s+1}$ completion of $C^\infty(\End(E))$.  Here $L^2_s$ denotes

479:      the completion in the Sobolev norm corresponding to sections with

480: $s$

481: derivatives in $L^2$, extended in the standard way to all

482: $s\in \rr$. Then

483: $\cG_{s+1}$ acts smoothly on $\cA_{s}$ (see \cite{FU}). The restriction

484: $s\ge 1$ guarantees that the gauge transformations are continuous.

485: 

486: \subsection{The Chern-Simons function} The Chern-Simons function

487: $cs:\cA\to

488: \rr$ is defined in terms of a trivialization $P\cong X\times SU(n)$ as

489: follows. To a connection

490: $A\in \cA$ one can assign the $su(n)$-valued 1-form

491: $$a:=A-\Theta\in

492: \Om^1(X;\ad P)\cong\Om^1(X)\otimes su(n).$$ One then defines

493: $$cs(A)= \frac{1}{8\pi^2}\int_X \tr(da \wedge a +\tfrac{2}{3} a\wedge

494: a\wedge a).$$ In this expression (as in Equation \eqref{eq2.2}) the

495: wedge

496: product means the wedging of the differential forms and matrix

497: multiplication of the

498: $su(n)$  coefficients. Note that the definition (and value) of the

499: Chern-Simons function depends on the choice of trivialization of

500: $P$. If $X$ is {\it closed}, then the reduction of $cs(A)$ modulo

501: $\zz$ is independent of the choice of trivialization, and is gauge

502: invariant. In fact, $cs(g\cdot A)= cs(A) + \text{degree}(g) $, where the

503: degree is defined using the fact that

504: $H^3(SU(n);\zz)=\zz$.  Thus when $X$ is closed, one considers the

505: Chern-Simons function as a circle valued function on the quotient

506: $cs:\cA/\cG\to \rr/\zz=S^1$.

507: 

508: In \cite{taubes}, Taubes gave an interpretation of Casson's invariant of

509: homology 3-spheres by studying the Morse theory of the Chern-Simons

510: function. This was generalized in much subsequent work, including the work of

511: Floer

512: \cite{floer} and Herald-Boden \cite{herald-boden}.

513: Thus we look next   at the differential topology of

514: $\cA$,

515: $\cG$,  and $cs$.

516: 

517: The tangent space of $\cA$ at a connection $A$ is canonically identified

518: with $\Om^1(X;\ad P)$.  It is not hard to identify the tangent space of

519: $\cG$  at the identity with $\Om^0(X;\ad P)$. Fixing a connection $A\in

520: \cA$ yields  the {\em action map}

521: $$g_A:\cG\to \cA, \  g\mapsto g\cdot A$$ The differential of this map is

522: just the map $d_A:\Om^0(X;\ad P)\to \Om^1(X;\ad P)$. Precisely, the

523: diagram

524: \begin{equation}\label{eq2.3}

525: \begin{diagram}\dgmag{900}\dgsquash

526: \node{T_1\cG}\arrow{e,t}{dg_A}\arrow{s,l}{\cong}

527:           \node{T_A\cA}\arrow{s,l}{\cong}\\

528: \node{\Om^0(X;\ad P)}\arrow{e,t}{d_A}\node{\Om^1(X;\ad P)}

529: \end{diagram}\end{equation} commutes. Thus the tangent space to the

530: quotient $\cA/\cG$ at $[A]$,

531: $T_{[A]}\cA/\cG$ can be identified with the cokernel of $d_A$ (at least

532: at

533: smooth points of $\cA/\cG$: making this precise requires looking at

534: Sobolev completions and dealing with quotient singularities but we will

535: not emphasize this point here).

536: 

537: \subsection{The Hessian of the Chern-Simons function}\label{sec2.3} {\em

538: In this section

539: we assume that $X$ is a closed 3-manifold}.  This is because the

540: differential operators which arise from the Morse theory of the

541: Chern-Simons function  have a simple  description on a closed manifold

542: which is adequate for our purposes, even though we are ultimately

543: interested in cutting and pasting questions. This section could be

544: written

545: in the context of manifolds with boundary, but in that case the proper

546: way

547: to study the Chern-Simons function is as a section of a determinant

548: bundle

549: as in \cite{RSW} and \cite{Herald1}. (The material of this subsection is

550: well known. Standard references include \cite{taubes,

551: donaldson-kronheimer, FU}.   We include it to set up notation and to

552: prepare the reader for

553: the section on perturbations.)

554: 

555: The differential

556: $dcs_A:T_A\cA\to

557: \rr$ of the Chern-Simons function is computed (see \cite{taubes}) {\em

558: for

559: $X$ closed} as:

560: \[ dcs_A(b)=\frac{1}{4\pi^2}\int_X \tr(F(A)\wedge b),\] with $F(A)$

561: the curvature 2-form of $A$. (Recall that

562: in

563: a  trivialization,

564:      $F(A) =da+a\wedge a$.)

565:      From this one concludes that the critical points of $cs:\cA\to\rr$

566: are

567: those

568: $A$ such that $F(A)=0$, i.e. the {\em flat connections}.

569: 

570: The Hessian of $cs:\cA\to\rr$, $\hess_A(cs):T_A\cA\times T_A\cA\to

571: \rr$  is defined for any connection  $A\in\cA$ using the affine

572: structure

573: of $\cA$. It can be computed explicitly when

574: $X$ is closed (\cite{taubes}):

575: \[

576: \hess(cs)_A(b,c)=-\frac{1}{4\pi^2}\langle *d_Ab,c\rangle.

577: \]

578: 

579: The $L^2$ inner product can be viewed as a Riemannian metric on the

580: infinite--dimensional manifold $\cA$ (or $\cA_s$). Identifying

581: vectors and covectors via 

582: this metric  allows us to define the gradient vector field of

583: $cs$

584:      by the rule $\langle\grad(cs)_A,b\rangle_{L^2}=dcs_A(b)$. Similarly

585: the

586: linearization $H_A:T_A\cA\to T_A\cA$ of $\grad(cs)$ is obtained

587: from the Hessian by the rule

588: $\langle H_A(b),c\rangle_{L^2}=\hess(cs)_A(b,c)$.  From the above

589: formulas

590: one concludes that $$\grad(cs)_A= -\tfrac{1}{4\pi^2}*F(A)$$ and

591: $$H_A(b)= -\tfrac{1}{4\pi^2}*d_A(b).$$

592: 

593: The self--adjoint operator $H_A$ is not elliptic because of the gauge

594: invariance of

595: $cs$. For example, at a flat connection $A$, $d_Ad_A=F(A)=0$ , and so

596: $H_A(d_A(b))=0$ for all $b\in \Om^0(X;\ad P)$. In other words the kernel

597: of $H_A$ contains the infinite--dimensional subspace of tangent vectors

598: to

599: the

600: $\cG$ orbit

601:      through $A$. This causes problems if one wishes to consider the

602: Morse

603: theory of $cs$.

604: 

605: As explained in \cite{taubes} (and is typical in gauge theory), the

606: correct

607: way to resolve this lack of ellipticity is to consider $\grad(cs)$

608: as a

609: vector field   on the quotient

610: $\cA/\cG$, and thus $H_A$ as a family of elliptic self--adjoint

611: operators on

612: $T_{[A]}\cA/\cG$.    Using the slice theorem  and Diagram

613: \eqref{eq2.3} one identifies

614: $T_{[A]}\cA/\cG$ (one has to take care at the singular points) as the

615: orthogonal complement of the image of

616: $d_A:\Om^0(X;\ad P)\to \Om^1(X;\ad P)$, in other words as the kernel of

617: the adjoint $d_A^*:\Om^1(X;\ad P)\to \Om^0(X;\ad P)$.  The Bianchi

618: identity implies that

619: $d_A^*(\grad(cs)(A))=-\tfrac{1}{4\pi^2}d_A^*(*F(A))=0$. If $A$ is a

620: critical point of $cs$ (i.e.~a flat connection), then

621: $$d_A^*(H_A(b))=\tfrac{1}{4\pi^2}*d_A**d_A(b)=

622: \tfrac{1}{4\pi^2}*d_Ad_A(b)=\tfrac{1}{4\pi^2}*F(A)(b)=0.$$

623: But $H_A(b)$

624: is

625: not necessarily in the kernel of $d_A^*$ when $A$ is not a critical

626: point.

627: This motivates setting

628: $$\tilde{H}_A :\ker d_A^*\to \ker d_A^*, \ \tilde{H}_A(b)=

629: -\frac{1}{4\pi^2}\proj_{\ker d_A^*} (*d_A(b))$$ where $\proj$ denotes the $L^2$

630: orthogonal

631: projection. The operator $\tilde{H}_A$ agrees with the restriction of

632: $H_A$ to $\ker d_A^*$ if the connection $A$ is flat.

633: 

634: Then $\tilde{H}_A$ is  Fredholm  (i.e. its extension to

635: $L^2_s(\Omega^1(X;\ad P))\cap \ker d_A^*\to L^2_{s-1}(\Omega^1(X;\ad

636: P))\cap \ker d_A^*$ is Fredholm). Moreover, it is self--adjoint with

637: compact resolvent with respect to the $L^2$ inner product restricted to

638: $\ker d_A^*$. The spaces $\ker d_A^*$ form a smooth vector bundle over

639: each orbit-type stratum of

640: $\cA/\cG$. In particular, Taubes' idea was to use  the spectral flow

641:    of

642: a

643: family of operators $\tilde{H}_{A_t}$ as a substitute for the difference

644: in Morse index of critical points $A_0$ and $A_1$ of the Chern-Simons

645: function, and thereby to define an Euler characteristic for

646: $\cA/\cG$.

647: 

648: (The {\em spectral flow} $\SF(H_t)$ of a 1-parameter family of

649: self--adjoint operators $H_t$ is

650: the integer defined to be the   number of eigenvalues  that cross zero, counted with sign.

651: Some conventions need to be set if the

652: operators in question have kernel at the endpoints of the path, and one

653: needs to

654: establish that the concept is well defined  for appropriately

655: continuous  paths of

656: unbounded operators. A careful construction appropriate for our

657: purposes can can be

658: found in

659: \cite{booss-lesch-phillips}. The basic properties and conventions are

660: explained in

661: \cite{kirk-lesch}.)

662: 

663: Taubes observed that the spectral flow of a family $\tilde{H}_{A_t}$

664: equals the spectral flow of a slightly better behaved family of

665: operators.

666: Consider, for any connection $A$, the twisted de Rham sequence

667: \begin{equation}\label{eq2.4} 0\to \Om^0(X;\ad

668: P)\xrightarrow{d_A}\Om^1(X;\ad P)\xrightarrow{d_A}\Om^2(X;\ad

669: P)\xrightarrow{d_A}\Om^3(X;\ad P)\to 0.

670: \end{equation} The sequence \eqref{eq2.4} is an  elliptic  complex when

671: $A$ is flat.   It is not a complex at a non-flat connection, but can be

672: made into a complex if one substitutes

673: $d'_A:=-4\pi^2 *\tilde{H}_A\circ\proj_{\ker

674: d_A^*}$ for

675: $d_A:\Omega^1(X;\ad P)\to \Omega^2(X;\ad P)$.   When $A$ is flat,

676: $d'_A=d_A$.

677: 

678: Whether or not

679: $A$ is flat, the   {\it odd signature operator}  (obtained essentially

680: by folding up the sequence \eqref{eq2.4})

681:      \begin{equation}

682: \label{foldedupop}\begin{split} &D_A:\Om^0(X;\ad P)\oplus \Om^1(X;\ad

683: P)\to \Om^0(X;\ad P)\oplus

684: \Om^1(X;\ad P)\\ &D_A(b,c)=(d_A^* c, *d_A c+d_A b)

685: \end{split}\end{equation} is a self--adjoint Dirac operator in the

686: sense of   e.g. \cite{booss-w}

687:   for any

688: connection $A$.

689: 

690: One also has a  self--adjoint operator $D'_A(b,c)=(d_A^* c, *d'_A c+

691: d_Ab)$. For $A$ flat, $D_A=D'_A$. For a general irreducible connection

692: $A$,

693: $D_A-D'_A$ is a   relatively compact operator. 

694: In fact, letting $\Delta_A^{-1}:\Om^0(X;\ad P)\to \Om^0(X;\ad P)$ denote the inverse of $d_A^*d_A$  (which is invertible for $A$ irreducible) and observing   that

695: $\proj_{\ker

696: d_A^*}x= x-d_A\Delta_A^{-1}d_A^*x$, we see that 

697: $$(D_A-D'_A)(b,c)= \big( *F(A)\Delta_A^{-1} d_A^*- d_A\Delta_A^{-1}*F(A)+d_A\Delta_A^{-1}*F(A)d_A\Delta_A^{-1}d_A^*\big)(c).$$

698:  (see \cite{taubes} for details).     

699: 

700: 

701:   By decomposing

702: $\Om^0(X;\ad P)\oplus

703: \Om^1(X;\ad P)$ as $\Om^0(X;\ad P)\oplus  \text{Image }d_A\oplus\ker

704: d_A^*

705: $, one sees that   the spectrum of

706: $D'_A$ is the disjoint union of the spectrum of $\tilde{H}_A$ and a

707: symmetric (with respect to $\lambda\to-\lambda$) spectrum.

708: A simple

709: argument then shows that given a path $A_t$ of (say irreducible)

710: connections joining two critical points of $cs$,   the spectral flow of

711: the family

712: $\tilde{H}_{A_t}$ (which plays the role of the difference in  Morse

713: indices of $cs$ at $A_0$ and

714: $A_1$) equals the spectral flow of the family $D_{A_t}$. What is gained

715: is that the spectral flow of $D_{A_t}$ makes sense whether or not $A_t$

716: is a path of irreducible connections, i.e.~even when the subspaces

717: $\ker d_A^*\subset \Omega^1(X; \ad P)$ do not vary continuously.

718: 

719: \subsection{Perturbations} We come now to the construction of

720: perturbations and their effect on

721: the Hessian of the Chern-Simons  function.

722:    The critical point set of

723: $cs$ may be complicated and it is often necessary to perturb the

724: Chern-Simons function to make it as nice as possible.

725: 

726:      Making precise what ``nice" means can be quite difficult in the case

727: of

728: $SU(n)$ for $n>2$, or when the homology of $X$ is complicated. In the

729: case

730: of $SU(2)$ and $X$ a homology sphere  Taubes replaced $cs$ by a function

731: $cs+h$ whose restriction to the top stratum of

732: $\cA/\cG$ (i.e. the irreducible connections; see Section

733: \ref{application})   is Morse in a suitable sense, and then used this to

734: define a topological invariant in the spirit of the Poincar\'e-Hopf

735: theorem,  which he showed equals Casson's invariant. In

736: \cite{herald-boden} a similar analysis is carried out for $SU(3)$ and

737: $X$

738: a homology sphere leading to a definition of an $SU(3)$ Casson

739: invariant;

740: in that case in addition to being Morse on the top stratum of

741: $\cA/\cG$,

742: $cs+h$ must be suitably non-degenerate in higher codimensional strata.

743: In

744: more recent work \cite{herald2} Herald shows how with a suitable choice

745: of

746: perturbation

747: $h$ one can define versions of Casson's invariant for rational homology

748: spheres and $SU(n)$, $n=2,3,4,5$.

749: 

750: There are slightly different approaches to constructing a suitable

751: family

752: of perturbations $h$, but they are similar and our method will apply to

753: each approach.  We mention that the basic requirements that $h$ should

754: satisfy are:

755: \begin{enumerate}

756: \item The perturbation $h$ should be a smooth function on the Hilbert

757: manifold $\cA_1/\cG_2$. 

758: \item The operator 

759: defined by the Hessian of $cs +h$  at $A$ should be a

760: compact perturbation of

761: $\tilde{H}_A$.

762: \item The family of admissible $h$ should be large enough so that one

763: can

764: prove various general position results about $cs+h$ on the strata and

765: their normal bundles of $\cA/\cG$.

766: \end{enumerate}

767: 

768: The following types of perturbations,   due to Taubes and Floer,

769: accomplish these goals, as explained in

770: \cite{herald2}.

771: 

772: First, fix a collection of smooth embeddings $\gamma_i:D^2\times S^1\to

773: X,\ i=1,\cdots,N$. Taubes \cite{taubes} requires that the embeddings

774: have

775: disjoint images, whereas   Floer \cite{floer} and Herald

776: \cite{herald2}  assume that   $\gamma_i(x,1)=\gamma_1(x,1)$ for all

777: $i$ and $x\in D^2$ and that the derivatives of the $\gamma_i$ in the

778: $S^1$

779: direction  at points in $D^1\times \{1\}$ are independent of $i$. (In

780: fact, nothing is lost by assuming that there is an interval $I$ around

781: $1\in S^1$ so that $\gamma_i(x,u)$ is independent of $i$ for $u\in I$.)

782: The results of this article work equally well in   each setting. The

783: difference between them is that it is easier to show that the third

784: condition above holds for   perturbations of the type Floer considers

785: since  the base point issues are easier to control. We will assume, to

786: keep the notation under control, that we are using the approach of

787: Floer,

788: i.e. the common basepoint approach.

789:     We do not address the third condition in this article.  In the

790: language

791: of

792: \cite{Herald1,herald2,herald-boden} the third  condition holds if the

793: set

794: of perturbations is {\em abundant}.

795: 

796: Next,  let $\cP$ denote the set of $C^r$ functions (for some fixed large

797: $r$),

798: $f:SU(n)^N\to \rr$, invariant under the conjugation action of

799: $SU(n)$, i.e. $f(rg_1r^{-1},\cdots,rg_Nr^{-1})=f(g_1,\cdots,g_N)$.

800: 

801: Now let $P|_{D^2}$ denote the restriction of the principal bundle $P$ to

802: the disc $D^2=\gamma_i(D^2\times\{1\})$. Given a smooth connection $A\in

803: \cA$, consider the map

804: $\Phi_A:P|_{D^2}\to SU(n)^N$ defined  by the rule

805: $\Phi_A(p)=(g_1,\cdots ,g_N)$ where $pg_i\in P_x$ is the endpoint of

806: the

807: parallel lift of  the loop $\gamma_i(x,u), \ u\in [0,2\pi]$  with

808: respect

809: to

810: $A$ starting at

811: $p$ (where $x\in D^2$ is the image of $p$ via the bundle projection).

812: Then

813: $\Phi_A(pr)=(r^{-1}g_1 r,\cdots,r^{-1}g_N r)$ for $r\in SU(n)$. It

814: follows

815: that if $f\in \cP$, $f\circ \Phi_A$ descends to a well-defined function

816: on

817: $D^2$.

818: 

819: If the principal bundle is trivialized, $P\cong X\times SU(n)$, then one

820: can view parallel lifting as the holonomy, a function from loops to

821: $SU(n)$, and in our context this allows us to define

822: $$\Hol:\cA\times D^2\to SU(n)^N, \ \ (A,x)\mapsto

823: (\Hol_{\gamma_1(x,-)}(A),\cdots \Hol_{\gamma_N(x,-)}(A)).$$

824:      Thus we denote the function $f\circ \Phi_A$  by

825: $x\mapsto f(\Hol(A,x))$. Notice  that $f(\Hol(A,x))$ is independent of

826: the

827: trivialization, and is unchanged if we change $A$ by a gauge

828: transformation.

829: 

830: Fix a smooth positive cut--off function $\eta$ on $D^2$ which vanishes

831: near

832: the boundary. The space of {\em admissible perturbations} is defined to

833: be

834: the space of functions of the form

835: \begin{equation}\label{eq2.5}h_f:\cA\to \rr, \ A\mapsto

836: \int_{D^2}f(\Hol(A,x))\eta(x)\ d^2x.\end{equation}

837:      Thus we view  the admissible perturbation functions as being

838: parameterized by the (Fr{\'e}chet) space $\cP$.

839: 

840: Taubes and Floer show that $h_f$ extends to a smooth function on

841: $\cA_2$.  Note that $d(cs +h_f)_A= dcs_A + d({h_f})_A $ and

842: $\hess(cs+h_f)_A=\hess(cs)_A+\hess(h_f)_A$.  Thus one can define

843: $$Q_{A,f}:T_A\cA\cong\Om^1(X;\ad P)\to T_A\cA $$ by  the rule

844: $$\langle

845: Q_{A,f}(b),c\rangle=\hess(h_f)_A(b,c)=\tfrac{\partial^2}{\partial t 

846: \partial s}|_{s=t=0}\ h_f(A+sb +tc).$$  Then setting $d_{A,f}:=

847: d_A-4\pi^2

848: *Q_{A,f}$ we see that $-\tfrac{1}{4\pi^2} *d_{A,f}$

849:      is the operator defined by the Hessian of $cs+h_f$ at $A$.

850: 

851: A connection $A$ is called {\em $f$-perturbed flat} if it is a critical

852: point of $cs +h_f$  or, equivalently,  if $*F(A)=4\pi^2

853: \grad(h_f)_A$. As in Section \ref{sec2.3}, we form the sequence

854: \begin{equation}\label{eq2.6}0\to \Om^0(X;\ad P)\xrightarrow{d_A}

855: \Om^1(X;\ad P)\xrightarrow{d_{A,f}}

856:      \Om^2(X;\ad P)\xrightarrow{d_A}

857:       \Om^3(X;\ad P)\to 0.\end{equation} The sequence \eqref{eq2.6} is a

858: complex if $A$ is

859: $f$-perturbed flat, and for any connection $A$ and perturbation $f$ the

860: operator obtained from  \eqref{eq2.6} (compare to \eqref{foldedupop}),

861: \begin{equation}

862: \begin{split} &D_{A,f}:\Om^0(X;\ad P)\oplus \Om^1(X;\ad P) \to

863: \Om^0(X;\ad

864: P)\oplus

865: \Om^1(X;\ad P),\\

866: &D_{A,f}(b,c) =(d_A^* c ,*d_{A,f} c  + d_A b )

867: \end{split}

868: \end{equation}

869: is a self--adjoint operator called the {\em perturbed

870: twisted odd signature operator}.

871: 

872: The operator $*Q_{A,f}$ is independent of the choice of

873: Riemannian metric on $X$ since the definition of

874: $\hess(h_f)_A(b,c)$ does not use the metric, and since

875: $$\hess(h_f)_A(b,c)=\langle Q_{A,f}(b),c\rangle=-\int_X\tr(c\wedge

876: *Q_{A,f}(b)).

877: $$ It follows that $d_{A,f}$ is metric independent. In particular, if

878: $A$

879: is $f$-perturbed flat, then the cohomology of \eqref{eq2.6}, and hence

880: the

881: dimension of the kernel of $D_{A,f}$, is independent of the metric on

882: $X$.

883: 

884: We will give a proof  in Lemma \ref{estimate}  of Taubes' observation

885: that for any

886: $A$ and $f$, the operator

887: $D_{A,f}$ is Fredholm and in fact a compact perturbation of

888: $D_A$. Assuming this,  we conclude the following.

889: 

890: \begin{lemma}\label{independence} Suppose that $f_0$ and $f_1$ are

891: perturbations,

892: $A_0$ and

893: $A_1$ are connections so that $A_0$ is $f_0$ perturbed flat and

894: $A_1$ is

895: $f_1$ perturbed flat.  If $f_t$ is a path of perturbations from $f_0$ to

896: $f_1$ and

897: $A_t$ is a path of connections from $A_0$ to $A_1$ then

898: the spectral flow $\SF(D_{A_t,f_t})$ is independent of the choice of

899: paths

900: $A_t$ and

901: $f_t$ and independent of the  choice of Riemannian metric.

902: \end{lemma}

903: \begin{proof}

904: Since the space of connections and the space of perturbations are

905: contractible and the spectral flow is a homotopy invariant,

906: $\SF(D_{A_t,f_t})$ is independent of the choice of paths $A_t$ from

907: $A_0$ to $A_1$ and path $f_t$ from $f_0$ to $f_1$. Since the dimensions

908: of

909: the kernels of $D_{A_0,f_0}$ and $D_{A_1,f_1}$ are independent of the

910: Riemannian metric, it similarly follows that   $\SF(D_{A_t,f_t})$ is

911: independent of the choice of Riemannian metric.

912: \end{proof}

913: 

914: \subsection{Locality of $Q_{A,f}$}\label{secloc}

915:      The operator $D_{A,f}$ is not a differential operator, but it has

916: the

917: property that it differs from a differential operator by an operator

918: which

919: is  ``localized'' in the neighborhood of the solid tori used to

920: construct the perturbations.

921: 

922: Write

923: $$D_{A,f}(b,c)= D_A(b,c) + (0, -4\pi^2 Q_{A,f}(c))$$ so that $D_{A,f}$

924: differs from the Dirac operator $D_A$ by

925: $-4\pi^2 Q_{A,f}$.

926: 

927: \begin{prop}\label{lem2.1} Let $S\subset X$ denote the union of the

928: images

929: of the embeddings $\gamma_i$, $S=\cup_i\gamma_i(D^2\times S^1)$.

930: \begin{enumerate}

931: \item If $b,b'\in \Om^1(X;\ad P)$ and $b|_{S}=b'|_{S}$, then

932: $Q_{A,f}(b)=Q_{A,f}(b')$.

933: \item For any $b\in\Om^1(X;\ad P)$, $Q_{A,f}(b)$ vanishes outside of

934: $S$.

935: \item If $A,A'$ are connections and $A|_S=A'|_S$, then

936: $Q_{A,f}(b)=Q_{A',f}(b)$ for all $b\in\Om^1(X;\ad P)$.

937: \end{enumerate}

938: 

939: \end{prop}

940: \begin{proof} Suppose that  $b\in\Om^1(X;\ad P)$ vanishes on $S$.  Then

941: for each

942: $c\in\Om^1(X;\ad P)$,

943: $$\langle Q_{A,f}(b),c\rangle=\hess(h_f)_A(b,c)=

944: \tfrac{\partial^2}{\partial s  \partial

945: t}|_{s=t=0}\int_{D^2}f(\Hol(A+sb+tc,x))\eta(x)d^2x=0,$$

946: since the parallel lifts of the loops $\gamma_i(x,u), \ u\in[0,2\pi]$

947: with

948: respect to the connection $A+sb+tc$ are independent of $s$.  Hence

949: $Q_{A,f}(b)=0$. Part (1) now follows since $Q_{A,f}$ is linear and

950: $b-b'$

951: vanishes on $S$.

952: 

953: For part (2), Let $b\in\Om^1(X;\ad P)$ be arbitrary. Let $U\subset X\setminus S$

954:      and choose $c\in\Om^1(X;\ad P)$ supported in $U$.  Then

955:      $$\langle Q_{A,f}(b),c\rangle=\langle b,Q_{A,f}(c)\rangle=0$$

956:      by the same argument given for part (1) since $c$ vanishes on

957:      $S$.  Since the $L^2$ inner product of $Q_{A,f}(b)$ with $c$

958:      vanishes for every $c$ supported on $U$, $Q_{A,f}(b)$ must vanish on

959:      $U$.

960: 

961: Part (3) is clear.

962: 

963: \end{proof}

964: 

965:      Proposition \ref{lem2.1} implies that  the

966: restriction of

967: $D_{A,f}$  to

968:      a codimension 0 submanifold whose boundary misses

969: $S=\cup_i$Image$(\gamma_i)$ is well--defined. To be precise, suppose

970: that  $Y\subset X$ is a compact

971:      codimension 0 submanifold. We consider two cases:

972: $S\subset$interior$(Y)$

973:      and $S\subset X\setminus Y$.

974: 

975:      If $S\subset$interior$(Y)$, $A$ is any connection on $Y$ and

976:        $c$ is a 1-form  on $Y$ with values in $\ad

977:      P|_Y$, extend $A$ and $c$ arbitrarily to $\tilde{A}, \tilde{c}$ on

978: $X$.

979: Then

980:      $Q_{\tilde{A},f}(\tilde{c})$ vanishes outside of $S$ and is

981: independent

982: of the choice of extensions. Thus $Q_{A,f}$ is well-defined as an

983: operator

984: on $Y$.

985:      Since $D_A$ is a differential operator it is locally defined and

986: hence

987: has a well defined restriction to $Y$. It follows that

988: $D_{A,f}=D_A-4\pi^2

989: Q_{A,f}$ is well defined on $Y$.

990: 

991: If $Y\subset X\setminus S$, then a similar argument shows that $D_{A,f}$

992: restricts

993: to the differential operator $D_A$ on $Y$.

994: 

995: \subsection{An estimate}  We next give an  argument to show that

996: $D_{A,f}$ differs from $D_A$ by an operator which is bounded on $L^2$.

997: This result is essentially contained in Taubes' article \cite{taubes}.

998: See

999: also

1000: \cite[Proposition 2.8]{herald-boden}.

1001: 

1002: \begin{lemma} \label{estimate} There exists a constant

1003: $C$   so that for all $A\in

1004: \cA, f\in\cP$, and $b\in \Om^1(X;\ad P)$

1005: $$\|Q_{A,f}(b)\|_{L^2}\leq C \|f\|_{C^2}\|b\|_{L^2}.$$ In particular,

1006: $D_{A,f}$ is a compact perturbation of

1007: $D_A$, viewed as operators $L^2_1\to L^2$.

1008: \end{lemma}

1009: \begin{proof}

1010: 

1011: Denote $\Hol(A+sb+tc,x)$ by $H_x(s,t)$,  so

1012: $H_x:\rr^2\to  SU(n)^N$, and let

1013: $f:SU(n)^N\to \rr$ be invariant under the conjugation action of $SU(n)$.

1014: Then

1015: $$

1016: \hess(h_f)_A (b,c) = \left.\tfrac{\partial^2}{\partial

1017:       s\partial t}\right|_{s=t=0} \int_{D^2} f(H_x(s,t))\eta(x) d^2x =

1018: \int_{D^2}

1019: \left.\tfrac{\partial^2}{\partial

1020:       s\partial t}\right|_{s=t=0}  f(H_x(s,t)) \eta(x)d^2x.

1021: $$

1022: 

1023: We may think of $H_x:\rr^2\to SU(n)^N\subset

1024: \cc^{n^2N}$ and  extend  $f$  to a smooth function

1025: $\cc^{n^2N}\to \rr$.    By the chain and the product rule  for vector

1026: valued functions  we have

1027: \begin{eqnarray*}

1028: \lefteqn{\left.\tfrac{\partial^2}{\partial

1029:       s\partial t}\right|_{s=t=0}  f(H_x(s,t))}\\ & = & D_{H_x(0,0)} f

1030: \circ

1031: \left.\tfrac{\partial^2}{\partial

1032:       s\partial t}\right|_{s=t=0} H_x(s,t) +

1033:       \left(\left.\tfrac{\partial}{\partial s}\right|_{s=0}

1034:       H_x(s,0)\right)^T \circ D_{H_x(0,0)}^2 f \circ

1035: \left.\tfrac{\partial}{\partial t}\right|_{t=0}

1036:       H_x(0,t)

1037: \end{eqnarray*}

1038: 

1039: Then

1040: \begin{eqnarray*}

1041: \lefteqn{\left|\left.\tfrac{\partial^2}{\partial

1042:       s\partial t}\right|_{s=t=0}  f(H_x(s,t))\right|}\\

1043: & \leq & \|D_{H_x(0,0)} f \| \cdot \|

1044: \left.\tfrac{\partial^2}{\partial

1045:       s\partial t}\right|_{s=t=0} H_x(s,t) \|+

1046:       \|\left.\tfrac{\partial}{\partial s}\right|_{s=0}

1047:       H_x(s,0)\| \cdot \|D_{H_x(0,0)}^2 f\| \cdot \|

1048: \left.\tfrac{\partial}{\partial t}\right|_{t=0}

1049:       H_x(0,t)\|

1050: \end{eqnarray*}

1051: 

1052: Given  $c\in \Om^1(X;\ad P)$, $i\in \{1,\cdots ,N\}$,  and

1053: $x\in D^2$,  let

1054: $\tilde c_{i,x}:[0,1]\to su(n)$ be defined as follows.  Trivialize $\ad

1055: P$

1056: over the disc $D^2\subset X$, so $\ad P|_{D^2}\cong D^2\times su(n)$.

1057: Parallel  translation  by $A$ along the path $g_{i,x}:[0,1]\to X,

1058: t\mapsto\gamma_i(x,e^{2\pi i t})$ trivializes $g_{i,x}^*(\ad P)$ over

1059: $[0,1]$, and this way the pullback $g_{i,x}^*(c)$ is viewed as a 1-form

1060: with

1061: $su(n)$ coefficients. Contracting with the unit speed vector field

1062: $\tfrac{\del}{\del t}$ yields the function

1063: $\tilde c_{i,x}:[0,1]\to su(n)$, i.e.~

1064: $\tilde c_{i,x}=g_{i,x}^*(c)(\tfrac{\del}{\del t})$.

1065: 

1066: We have the estimate

1067: $$\int_{D^2} \int_0^1 |\tilde c_{i,x}(\nu)|\eta(x)\ d\nu\ d^2x

1068: \leq \int_{D^2} \int_0^1 |c(\nu)|\eta(x) \ d\nu\ d^2x\leq C_1\|

1069: c\|_{L^1(S^1\times D^2)}\leq C\| c\|_{L^1(X)}$$ for some constants $C_1,

1070: C$ independent of $c$.

1071: 

1072: Given a  connection $B$ let

1073: $h_{i,x}(B)$ denote the holonomy of $B$ around the loop

1074: $\gamma_i(x,-)$. Thus

1075: $$H_x(s,t)=(h_{1,x}(A+s b+t c),

1076: \cdots,h_{1,x}(A+s b+t c)).$$   Lemma 2.6    of

1077: \cite{herald-boden} establishes the formulas:

1078: \begin{eqnarray*}

1079: \left.\tfrac{d}{dt}\right|_{t=0} h_{i,x}(A+t b) & = & h_{i,x}(A)

1080: \int_0^1

1081: \tilde b_{i,x}(\nu) d\nu\\

1082: \left.\tfrac{\partial^2}{\partial s\ \partial t}\right|_{s=t=0}

1083: h_{i,x}(A+s b+tc) & = &

1084: h_{i,x}(A)

1085: \int_0^1\int_0^\nu (\tilde b_{i,x}(\nu)\tilde c_{i,x}(\mu) + \tilde

1086: c_{i,x}(\nu)\tilde b_{i,x}(\mu)) d\mu d\nu.

1087: \end{eqnarray*}

1088: 

1089: Notice that $h_{i,x}(A)\in SU(n)$, and so is bounded independently of

1090: $A$.  Thus  (letting $C$ denote possibly different constants)

1091: $$

1092: \int_{D^2}  |  \left.\tfrac{d}{dt}\right|_{t=0} H_x(s,0) | \ \eta(x)\

1093: d^2x

1094: \leq\sum_{i=1}^N C\|b\|_{L^1(X)} = NC\|b\|_{L^1(X)}\leq

1095: C\|b\|_{L^2(X)}.$$

1096: The last inequality follows from Cauchy-Schwarz.

1097: 

1098: Similarly

1099: \begin{eqnarray*}

1100:      \lefteqn{\int_{D^2} | \left.\tfrac{\partial^2}{\partial

1101:       s\partial t}\right|_{s=t=0} H_x(s,t)  | \ \eta(x) \ d^2x }\\

1102: & \leq  & C \int_{D^2} \int_0^1 |b(\nu)|d\nu \int_0^1|c(\mu)|d\mu d^2x

1103:    \leq C \left\|\int_0^1 |b(\nu)|d\nu\right\|_{L^2(D^2)} \cdot

1104: \left\|\int_0^1|c(\mu)|d\mu\right\|_{L^2(D^2)}\\

1105: & \leq &  C \| b\|_{L^2(S^1 \times D^2)} \cdot \| c \|_{L^2(S^1 \times

1106: D^2)}

1107: \leq C

1108: \|b\|_{L^2(X)}

1109: \cdot \|c\|_{L^2(X) }.

1110: \end{eqnarray*}

1111: 

1112: Also

1113: $$ \|D_{H_x(0,0)} f\| + \|D^2_{H_x(0,0)} f\| \leq C \|f\|_{C^2}.

1114: $$

1115: Thus for any $b,c\in \Om^1(X;\ad P)$,

1116: $$  \bigl|\hess(h_f)_A (b,c)\bigr|  \leq C \|f\|_{C^2}

1117: \|b\|_{L^2} \|c\|_{L^2},

1118: $$

1119: and hence $\|Q_{A,f}\|\le C \|f\|_{C^2}$.

1120: 

1121: For the  last statement of Lemma \ref{estimate}, the difference

1122: $D_{A,f}-D_A:L^2_1\to L^2$ equals the restriction  of

1123: $-4\pi^2Q_{A,f}$ to $L^2_1$. This restriction  can  be expressed as the

1124: composite of the compact inclusion

1125: $L^2_1\subset L^2$ and the bounded operator $-4\pi^2Q_{A,f}$, and hence

1126: is

1127: compact since the compact operators form a 2-sided ideal.

1128: \end{proof}

1129: 

1130: \section{The Poisson operator and the \Cald\ for perturbed Dirac

1131: operators

1132: on manifolds with boundary} A compact manifold with non-empty boundary

1133: always embeds  as a codimension 0 submanifold of a closed manifold

1134: (e.g.~

1135: its double) in such a way that a given principal bundle and $SU(n)$

1136: connection extend. It follows from Proposition \ref{lem2.1} that given:

1137: \begin{enumerate}

1138: \item a 3-manifold with boundary $X$,

1139: \item a principal bundle $P\to X$,

1140: \item a connection $A$ on $P$,

1141: \item a collection of embeddings $\gamma_i:D^2\times S^1\to

1142: $interior$(X)$, and

1143: \item a function $f\in \cP$

1144: \end{enumerate} one can unambiguously define operators $Q_{A,f}$ and

1145: $D_{A,f}$ on

1146: $X$.  In this context we will assume that the connection $A$ is in

1147: cylindrical form near the boundary. This means that in a collar

1148: neighborhood

1149: $[0,\epsilon)\times \partial X$  of $\partial X$, $A$ is the pullback

1150: of a

1151: connection $a$ on $\partial X$ with respect to the projection

1152: $[0,\epsilon)\times \partial X\to \partial X$ to the second factor. Then

1153: the operator $D_{A,f}$ equals $D_A$ on the collar, and this implies that

1154: $D_{A,f}$ takes the Atiyah--Patodi--Singer form (for details see

1155: \cite[Definitions 2.2 and 2.3]{BHKK})

1156: \begin{equation} D_{A,f}|_{[0,\epsilon)\times \partial

1157: X}=D_{A}|_{[0,\epsilon)\times

1158: \partial X}=

1159: \gamma(\tfrac{\partial}{\partial u} + S_a). \label{G4.1}

1160: \end{equation}

1161: 

1162: Here $S_a:\Om^*(\partial X;\ad P)\to \Om^*(\partial X;\ad P)$ is the

1163: tangential operator, given by the formula

1164: \begin{equation}\label{eq4.45}S_a(\alpha_0,\alpha_1,\alpha_2)=(*d_a\al_1

1165: ,

1166: -*d_a\al_0-d_a*\al_2, d_a*\al_1)

1167: \end{equation}

1168:     and $\gamma$ is the bundle isomorphism given by the formula

1169: \begin{equation}\label{eq4.4}

1170: \gamma(\alpha_0,\alpha_1,\alpha_2)

1171: =(-*\al_2,*\al_1,*\al_0)\end{equation}

1172: (in the  formulas \eqref{eq4.4} and \eqref{eq4.45}, $*$ refers to the

1173: Hodge

1174: $*$ operator for the metric on $\partial X$). The bundle isomorphism

1175: $\gamma$ satisfies $\gamma^2=-\Id$.

1176: 

1177: %%MARKE

1178: \subsection{The invertible double}

1179: 

1180: So far we have shown that $D_{A,f}$ is the sum of a Dirac operator $D_A$

1181: and  an  $L^2$--bounded operator $-4\pi^2 Q_{A,f}$. The operator

1182: $Q_{A,f}$

1183: is not a differential operator but it has certain locality properties as

1184: explained in Section \ref{secloc}. On a manifold with boundary we

1185: arranged

1186: so that $D_{A,f}$ has a product form in a collar of the  boundary.

1187: 

1188: In the study of Dirac operators on manifolds with boundary, their

1189: \Cald\ and their Poisson operator play a central role (cf.

1190: \cite{booss-w}).

1191: However, since $Q_{A,f}$ is far from being a differential operator,

1192: existence of the \Cald\ and of the Poisson operator is a priori not

1193: clear.

1194: We therefore treat the existence question and related issues. For the

1195: convenience of the reader our presentation will be more or less

1196: self--contained.

1197: 

1198: Let $X$ be a compact connected Riemannian manifold with boundary

1199: $\partial

1200: X=\Sigma$ and $D$ a self--adjoint Dirac type operator on the Hermitian

1201: vector bundle

1202: $E\to X$. We assume that we are in a product situation, i.e. that there

1203: is

1204: a collar $U=[0,\epsilon)\times \Sigma$ of $\Sigma$ in which $D$ takes

1205: the

1206: form

1207: \begin{equation}

1208:       D=\gamma\bigl(\frac{d}{dx}+B\bigr),

1209: \label{G4.10}

1210: \end{equation} where $\gamma$ is a bundle endomorphism and $B$ is a

1211: first

1212: order self--adjoint elliptic differential operator on $E|\Sigma$. One

1213: has

1214: the fundamental relations

1215: \begin{equation}

1216:      \gamma^2=-\Id,\quad \gamma^*=-\gamma,\quad\text{and}\quad

1217:      B\gamma=-\gamma B.\label{G4.11}

1218: \end{equation} Note that $D_A$ is of this form \eqref{G4.1}.

1219: 

1220: To model $Q_{A,f}$ we assume abstractly that we are given an auxiliary

1221: self--adjoint bounded linear operator $V$ in $L^2(X,E)$. In most cases

1222: $V$

1223: will be a pseudo--differential operator of order $0$, but this is not

1224: assumed here. More importantly, we assume the following locality

1225: property

1226: (cf. Sec. \ref{secloc}).

1227: \begin{equation}

1228: \text{There is a compact domain

1229: $S\subset X\setminus [0,\epsilon)\times \Sigma$ such that $M_\varphi

1230: V=0$

1231: for all $\varphi\in C^\infty_0(X\setminus S)$.}

1232: \label{pseudolocality}

1233: \end{equation}

1234: Here $M_\varphi$ denotes the operator of multiplication by $\varphi$.

1235: Note that by self--adjointness we automatically also have

1236: \begin{equation}

1237:       VM_\varphi=0\quad\text{for}\quad \varphi\in C^\infty_0(X\setminus

1238: S).

1239:       \label{G4.13}

1240: \end{equation} Moreover, if $f,g\in L^2(E)$ such that $f|W=g|W$ in a

1241: neighborhood $W$ of $S$ then choosing $\psi\in C^\infty_0(W)$ with

1242: $\psi\equiv 1$ in a neighborhood of $S$ we find

1243: \begin{equation} V(f-g)=V((1-\psi)(f-g))=(VM_{1-\psi})(f-g)=0.

1244: \label{G4.14}

1245: \end{equation}

1246: 

1247: The property \eqref{pseudolocality} implies that the restriction of

1248: $D+V$

1249: to the collar $U$ equals $D$ and hence takes the  product form

1250: \eqref{G4.10}. This and self--adjointness is all that is required to

1251: prove

1252: Green's formula (see \cite{booss-w})

1253: \begin{equation}

1254: \langle (D+V) f,g\rangle_{L^2(X,E)}-

1255: \langle f, (D+V) g\rangle_{L^2(X,E)}=\langle f|_\Sigma,

1256: \gamma(g|_\Sigma)\rangle_{L^2(\Sigma,E)}.

1257: \end{equation}

1258: 

1259: An important tool for studying elliptic boundary problems is the {\em

1260: invertible double construction} and the associated Poisson operator.

1261: Since

1262: it will be important in the sequel we recall the main facts (cf.

1263: \cite{booss-w}).

1264: 

1265: Let $\tilde X:=X_+\cup_{\Sigma} X_-$ where $X_\pm=X$. Let

1266: $\tilde E\to \tilde X$ be the vector bundle obtained by gluing

1267: $E$ with gluing function $\gamma$ over $\Sigma$. More precisely,

1268: consider

1269: the bundle $E\times\{\pm\} \to X\times\{\pm\}$. Then

1270: $$\tilde{X} =X\times\{\pm\}/\sim,  \text{ where } (x,+)\sim (y,-) \text{

1271: if and only if }

1272:     x=y\text{ and }

1273:     x\in\Sigma.$$  We write $X_\pm:=X\times\{\pm\}$.  Also,

1274: $$\tilde{E}=E\times\{\pm\}/\sim\text{ where } (x,v,+)\sim (y,w,-)

1275: \text{ if

1276: and only if }

1277:     x=y\in\Sigma \text{ and }  \gamma(v)=w.

1278:      $$  Denote by $m:\tilde X\to\tilde X$ the reflection interchanging

1279: $X_+$

1280: and $X_-$, i.e.  $m(x,\pm)=(x,\mp)$. Then $m$ is covered by a

1281: bundle

1282: map

1283: $\tilde{m}:\tilde{E}\to \tilde{E}$ defined by

1284: $\tilde{m}:(x,v,\pm)=(x,\mp

1285: v,\mp)$.  Notice that the restriction of $\tilde{m}$ to $

1286: \tilde{E}|\Sigma$

1287: equals $\gamma$ since if $x\in \Sigma,$

1288: $\tilde{m}:(x,v,\pm)=(x,\mp v, \mp)\sim (x, \gamma(v),\pm).$ The map

1289: $\tilde{m}$

1290: preserves the Hermitian metric on $E$  and

1291: $\tilde{m}^2=-\Id$.

1292: 

1293: The maps $m$ and  $\tilde{m}$  induce a unitary map

1294: \begin{equation}

1295: \Phi:L^2(\tilde X,\tilde E)\to L^2(\tilde X,\tilde E),\quad

1296: \Phi f(x):=\tilde{m} ( f(m(x)))

1297: \end{equation}

1298: satisfying

1299: $\Phi^2=-\Id$.  Thus $\Phi$ is defined by the diagram

1300: \[\begin{diagram}

1301: \node{\tilde{E}}\node{\tilde{E}}\arrow{w,t}{\tilde{m}}\\

1302: \node{\tilde{X}}\arrow{n,r}{\Phi

1303: f}\arrow{e,t}{m}\node{\tilde{X}}\arrow{n,r}{f}

1304: \end{diagram}\]

1305: 

1306: Then $D$ extends naturally to a Dirac operator $\tilde D$

1307: on

1308: $\tilde E$ satisfying

1309: \begin{equation}

1310:        \Phi^* \tilde D\Phi=-\tilde D.\label{G4.15}

1311: \end{equation} Similarly, in view of \eqref{pseudolocality}, $V$ extends

1312: to a bounded operator $\tilde V$ satisfying $\Phi^*\tilde V\Phi=-\tilde

1313: V$

1314: which has the same locality properties as $V$ with

1315: $\tilde S=S\cup m(S)$ instead of $S$.

1316: 

1317: Alternatively, we may view $\tilde D$ as a self--adjoint realization of

1318: a

1319: well--posed boundary value problem of $D\oplus (-D)$ acting on the

1320: bundle

1321: $E\oplus E$ over $X$. Namely, putting

1322: \begin{equation}

1323:           \begin{split}

1324:               \Psi:&L^2(\tilde X,\tilde E)\longrightarrow L^2(X,E\oplus

1325: E)\\

1326:                    &\Psi f:= f|X_+\oplus (\Phi f)|X_+

1327:           \end{split}\label{G4.16}

1328: \end{equation} we find

1329: \begin{equation}

1330:         \Psi\tilde D\Psi^{-1}=D\oplus (-D)

1331: \end{equation} and

1332: \begin{equation}

1333:       \Psi(\operatorname{dom}\tilde D)=\Psi(L^2_1(\tilde E))=

1334:         \bigl\{f\in L^2_1(X,E\oplus E)\,\big|\, (f|\Sigma)_2=\gamma

1335: (f|\Sigma)_1\bigr\}.\label{G4.17}

1336: \end{equation} It is indeed not difficult to show that

1337: $(f|\Sigma)_2=\gamma (f|\Sigma)_1$ is a well--posed boundary condition

1338: for

1339: $D\oplus (-D)$. The boundary operator of this boundary condition,

1340: an orthogonal projection in the pseudodifferential

1341: Grassmannian (see \cite{booss-w}),

1342: is given by

1343: \begin{equation}

1344:        \frac 12\begin{pmatrix} \Id & \gamma\\ -\gamma &

1345: \Id\end{pmatrix}.\label{G4.18}

1346: \end{equation}

1347: 

1348: We introduce the usual notations for restriction, extension, and trace

1349: operators:

1350: 

1351: \begin{defn}  Let $r^\pm$ be restriction of sections on $\tilde X$ to

1352: $X_\pm$ and

1353: $e^\pm$ extension by $0$ from $X_\pm$. $\varrho^\pm,\varrho$ denote the

1354: trace maps

1355: \begin{equation}\begin{split}

1356:        &L^2_s(X_\pm,E)\longrightarrow L^2_{s-1/2}(\Sigma,E),\\

1357:        &L^2_s(\tilde X,\tilde E)\longrightarrow L^2_{s-1/2}(\Sigma,E),

1358:                    \end{split}\quad s>1/2.\label{G4.19}

1359: \end{equation} Similarly, for $0<|t|<\epsilon$ we denote by $\varrho_t$

1360: the trace map which restricts sections to $\{t\}\times\Sigma$. For

1361: $\varrho_t$

1362: \eqref{G4.19} holds accordingly.

1363: 

1364: Note that, by the locality property \eqref{pseudolocality} of $V$, we

1365: have

1366: $r^+\tilde V=Vr^+.$

1367: \end{defn}

1368: 

1369: For convenience we abbreviate $T:=D+V$ resp. $\tilde T:=\tilde D+\tilde

1370: V$. The pseudolocality property of $\tilde V$ implies  that $\tilde T$

1371: has

1372: well--defined restrictions to $X_{\pm}$.  The following fact is

1373: well--known:

1374: 

1375: \begin{lemma}\label{ML-S4.3} For arbitrary $s\in [0,\infty)$ the trace

1376: map

1377: $\varrho^+$ extends to a bounded linear map

1378: \[   \ker T\cap L^2_s(X,E)\to L^2_{s-1/2}(\Sigma,E).\] More precisely,

1379: if

1380: $f\in \ker T\cap L^2_s(X,E)$ then

1381: $\varrho^+(f)=\lim\limits_{t\to 0+}\varrho_t(f)$ exists in

1382: $L^2_{s-1/2}(\Sigma,E)$ and the so defined $\varrho^+$ is bounded.

1383: \end{lemma}

1384: With some care the lemma could even be stated for all $s\in\mathbb{R}$.

1385: However,

1386: Sobolev spaces of negative order on manifolds with boundary are a

1387: nuisance.

1388: Since we will not need them we content ourselves to the case $s\ge 0$.

1389: \begin{proof} For $V=0$ the Lemma is well--known

1390: (\cite[Thm. 13.1]{booss-w}, cf. also \cite[Chap. I]{lions-magenes}).

1391: If $Tf=0$ then $Df=0$ in a

1392: collar of $\Sigma$. Since the result is local in a collar of

1393: $\Sigma$, we reach the conclusion.\end{proof}

1394: 

1395: In light of Lemma \ref{ML-S4.3} we can safely introduce  the following

1396: notation.

1397: 

1398: \begin{defn} Set

1399: \begin{align*}

1400:           Z_\pm^s&:=\bigl\{ f\in L^2_s(X_\pm,E)\,\big|\, \tilde

1401: T|_{X_\pm}

1402: f=0\bigr\},\quad s\ge 0,\\

1403:           \La_\pm^s&:=\varrho^\pm Z_\pm^{s+1/2},\quad s\ge -1/2,\\

1404:           Z_0&:=\bigl\{f\in L^2_1(\tilde X,\tilde E)\,\big|\, \tilde T

1405: f=0\bigr\}.

1406: \end{align*}

1407: \end{defn} Note that from elliptic regularity we immediately conclude

1408: that

1409: if

1410: $f\in L^2(\tilde X,\tilde E)$ and $\tilde Tf=0$ then

1411: $f\in L^2_1(\tilde X,\tilde E)$. Since $V$ is only assumed to be

1412: $L^2$--bounded, this cannot be improved. Also by elliptic theory,

1413: $\tilde T$ is a Fredholm operator (it is a relatively compact

1414: perturbation

1415: of $\tilde D$) and hence $\dim Z_0<\infty$.

1416: 

1417: \begin{prop}\label{ML-S4.5} If $f\in Z_0$ then $f$ vanishes in a collar

1418: of

1419: $\Sigma$ and hence $Z_0=r^+ Z_0\oplus r^-Z_0$.

1420: 

1421: In particular, $\tilde D$ is an invertible operator,   called the

1422: \emph{invertible double of $D$}.

1423: \end{prop}

1424: \begin{proof} We apply Green's formula and find

1425: \begin{equation}\begin{split}

1426:         \| \varrho f\|^2&=-\langle\varrho^+r^+f,\gamma \gamma

1427: \varrho^+r^+f\rangle= -\langle\varrho^+r^+f,\gamma \varrho^+r^+(\Phi

1428: f)\rangle\\

1429:              &= -\langle Tf,\Phi f\rangle_{L^2(X_+,E)}+\langle f,T\Phi

1430: f\rangle_{L^2(X_+,E)}=0

1431: \end{split}

1432: \end{equation} since $f\in Z_0$.

1433: 

1434: Thus we have $\varrho f=0$. Since the Dirac operator has the weak unique

1435: continuation property (cf. \cite{booss-ucp}) this implies that $f$

1436: vanishes

1437: in a collar neighborhood of $\Sigma$. If $V=0$ then the weak unique

1438: continuation property and the connectedness of $X$ imply together with

1439: the

1440: proven vanishing statement for $f\in Z_0$ that $\tilde D$ is

1441: invertible.\end{proof}

1442: 

1443: \begin{defn} We say that $D+V$ has the {\em weak unique continuation

1444: property with respect to $\partial X$} if $f=0$ is the only element of

1445: $Z_+^{1/2}$ with $\varrho^+f=0$.

1446: \end{defn}

1447: 

1448: In general we cannot expect $D+V$ to have the weak unique continuation

1449: property. However, for small $V$ it is true:

1450: 

1451: \begin{prop}\label{WUCP} The operator $D+V$ has the weak unique

1452: continuation property with respect to $\partial X$ if and only if

1453: $\tilde

1454: D+\tilde V$ is invertible. This is in particular the case if  $\|\tilde

1455: D^{-1}\tilde V\|<1$.

1456: \end{prop}

1457: \begin{remark} Booss--Bavnbek, Marcolli, and Wang \cite{BMW} prove

1458: the weak unique continuation property for a class of perturbed Dirac

1459: operators

1460: (perturbations may be nonlinear)

1461: arising in Seiberg--Witten theory. However, our perturbations

1462: are in general not admissible in their sense (cf. \cite[Def. 2.5]{BMW}).

1463: An operator $\mathfrak{P}$ is admissible in the sense of \cite{BMW} if

1464: there is a locally bounded function $P$ such that

1465: $|\mathfrak{P}u(x)| \le P(u,x)|u(x)|$. This implies in particular that

1466: $\mathfrak{P}u(x)=0$ if $u(x)=0$, i.e.

1467: $\mathfrak{P}$ is a local operator. If $\mathfrak{P}$ is a linear

1468: operator

1469: on smooth sections then this locality automatically implies that

1470: $\mathfrak{P}$

1471: is a differential operator of order $0$ (i.e. a bundle endomorphism). So

1472: the point in \cite{BMW} is that $\mathfrak{P}$ is allowed to be

1473: nonlinear.

1474: \end{remark}

1475: 

1476: \noindent{\it Proof of Proposition \ref{WUCP}.} In view of \eqref{G4.15}

1477: and Prop.

1478: \ref{ML-S4.5} the weak unique continuation

1479: property with respect to $\partial X$ implies $Z_0=0$ and hence the

1480: invertibility of $\tilde D+\tilde V$.

1481: 

1482: Conversely, assume that $\tilde D+\tilde V$ is invertible and consider

1483: $f\in Z_+^{1/2}$ with $\varrho^+f=0$. Then $e^+f\in Z_0=\{0\}$ and hence

1484: $f=0$.

1485: 

1486:     $\|\tilde D^{-1}\tilde V\|<1$ implies the invertibility of

1487: $\tilde D+\tilde V$ by means of the Neumann series.

1488: \qed

1489: 

1490: \subsection{Existence of Poisson operator and \Cald}\label{sec41}

1491: 

1492: Let $P_{Z_0}$ be the orthogonal projection onto $Z_0$ and denote by

1493: $\varrho^*$ the $L^2$--dual of $\varrho$. In other words, for a

1494: distributional section $f$ of $\tilde E|\Sigma$ and a test function

1495: $\varphi\in C^\infty(\tilde X,\tilde E)$ one has $(\varrho^*

1496: f,\varphi):=(f,\varrho

1497: \varphi)$.

1498: In view of Proposition \ref{ML-S4.5} one

1499: infers from this formula immediately that

1500: $P_{Z_0}\varrho^*=0$.

1501: 

1502: Furthermore, denote by $\tilde G$ be the pseudoinverse of $\tilde T$,

1503: i.e.

1504: \begin{equation}

1505:       \tilde Gf:=(\Id-P_{Z_0})(\tilde T+P_{Z_0})^{-1}f=

1506:       \begin{cases} \tilde T^{-1}f,& f\in Z_0^\perp,\\

1507:                             0,& f\in Z_0.

1508:       \end{cases}\label{G4.20}

1509: \end{equation} Note that since $\tilde T$ maps $L^2_s\to L^2_{s-1}, 0\le

1510: s\le 1$ the pseudoinverse maps $L^2_s\to L^2_{s+1}, -1\le s\le 0$.

1511: Together with the mapping properties of the trace map we infer that

1512: $\tilde G\varrho^*$ maps $L^2_s(\Sigma,E)$ continuously into

1513: $L^2_{s+1/2}(\tilde X,\tilde E)$ for each $-1/2\le s<0$. It is a

1514: fundamental fact that for $r^\pm \tilde D^{-1}\varrho^*$ this can be

1515: improved. Namely, one has:

1516: 

1517: \begin{thm}\label{Poissonop-a} $r^\pm\tilde D^{-1}\varrho^*$ maps

1518: $L^2_s(\Sigma,E)$ continuously into $L^2_{s+1/2}(X_\pm,E)$ for all

1519: $s\ge -1/2$.

1520: \end{thm}

1521: 

1522: For a proof see \cite{booss-w}. It will also follow from our discussion

1523: of

1524: the continuous dependence of the \Cald\ below. First we want to show

1525: that

1526: the previous theorem easily extends to hold for $\tilde T$ and hence a

1527: \Cald\ can be constructed for $\tilde T$.

1528: 

1529: \begin{thm}\label{Poissonop-b}

1530: $r^\pm\tilde G\varrho^*$ maps $L^2_s(\Sigma,E)$ continuously into

1531: $Z_\pm^{s+1/2}$ for $-1/2\le s\le 1/2$.

1532: 

1533: Furthermore, we have the following resolvent identity relating $\tilde

1534: G$

1535: and $\tilde D^{-1}$:

1536: \[\tilde G=\tilde D^{-1}-\tilde G \tilde V\tilde D^{-1}-P_{Z_0}\tilde

1537: D^{-1}.

1538: \]

1539: \end{thm}

1540: \begin{proof} The resolvent identity follows from

1541: \begin{equation}\begin{split}

1542:        \tilde G\tilde V\tilde D^{-1}= \tilde G(\tilde T-\tilde D)\tilde

1543: D^{-1}\\

1544:                  = (\Id-P_{Z_0})\tilde D^{-1}-\tilde G.

1545: \end{split}\end{equation} The operator $P_{Z_0}\tilde D^{-1}$ is

1546: bounded

1547: from $L^2_s(\tilde X,\tilde E)$ to  $L^2_1(\tilde X,\tilde E)$ for $s\ge

1548: -1$  (see    Prop. \ref{ML-S4.5} and the comment preceding it).

1549: Moreover,

1550: $\tilde G\tilde V\tilde D^{-1}$ maps

1551: $L^2_{-1}(\tilde X,\tilde E)$ continuously into $L^2_1(\tilde X,\tilde

1552: E)$.

1553:    From Theorem \ref{Poissonop-a} and the resolvent identity

1554: we thus conclude that

1555: $r^\pm\tilde G\varrho^*$ maps $L^2_s(\Sigma,E)$ continuously into

1556: $L^2_{s+1/2}(X_\pm,E)$.

1557: 

1558: Moreover, since

1559: $\tilde T \tilde G\varrho^*=(\Id-P_{Z_0})\varrho^*=\varrho^*$ and

1560: $\varrho^*$ is supported on $\Sigma$, we find that the operator actually

1561: maps into $Z_\pm^{s+1/2}$.

1562: \end{proof}

1563: 

1564: The restriction  $-1/2\le s\le 1/2$ in Theorem \ref{Poissonop-b}

1565: is due to the low regularity

1566: assumptions on $V$. If $V$ is a pseudodifferential operator, then

1567: Theorem \ref{Poissonop-a} holds verbatim for $\tilde T$ instead of

1568: $\tilde D$.

1569: The previous proof can easily be modified to this case.

1570: 

1571: With Theorem \ref{Poissonop-b} and Lemma  \ref{ML-S4.3}  in place we can

1572: make the following definitions.

1573: 

1574: \begin{defn}\begin{enumerate}

1575: \item Define $$K_\pm:=K(T)_\pm:=\pm r^\pm\tilde

1576: G\varrho^*\gamma:L^2_s(\Sigma,E)\to Z^{s+1/2}_{\pm}, \ -1/2\leq s\leq

1577: 1/2.$$

1578: Then $K:=K_+$ is called the {\em Poisson operator} for

1579: $T=D+V$.

1580: \item Define the  projection $$C_\pm:=C_\pm(T):=\varrho^\pm

1581: K(T)_\pm:L^2_s(\Sigma,E)\to

1582: L^2_s(\Sigma,E), \ -1/2\leq s\leq 1/2.$$   Then $C:=C_+$ is called the {\em

1583: \Cald\ }of

1584: $T$.

1585: \item Taking $s=0$, the subspace $$\La :=\La(T):=

1586: \La^0_+(T) =\operatorname{range}C_+(T)\subset

1587: L^2(\Sigma,E)

1588: $$ is called the {\em Cauchy data space} of $T$.

1589: \end{enumerate}

1590: \end{defn}

1591: 

1592: \begin{prop}\label{ML-S4.9} $C_\pm(D+V)$ are complementary orthogonal

1593: projections in $L^2(\Sigma,E)$ with

1594: $\operatorname{range}C_\pm=\La_\pm^0$.

1595: 

1596: Furthermore, $\gamma C_\pm\gamma^*=C_\mp$. In other words the subspaces

1597: $\La_\pm^0$ are Lagrangian subspaces with respect to the symplectic form

1598: $\omega(x,y)=\langle x,\gamma y\rangle_{L^2}.$

1599: \end{prop}

1600: 

1601: Recall that a Lagrangian subspace is a subspace $V\subset L^2(\Sigma,E)$

1602: satisfying $\gamma V=V^\perp$.

1603: 

1604: \begin{proof}

1605: $K_\pm$ maps $L^2(\Sigma,E)$ into $Z_\pm^{1/2}$ by Theorem

1606: \ref{Poissonop-b}. Hence in view of Lemma \ref{ML-S4.3} we find that

1607: $C_\pm$ is

1608: $L^2$--bounded with

1609: $\operatorname{range} C_\pm\subset \La_\pm^0$.

1610: 

1611: We show  $\La_+^0\perp \La_-^0$: pick $\xi\in \La_+^0,

1612: \eta\in \La_-^0$ and choose $f\in Z_+^{1/2}, g\in Z_-^{1/2}$ with

1613: $\xi=\varrho^+f, \eta=\varrho^-g$. Then Green's formula yields

1614: \begin{equation}

1615:        \begin{split}

1616: (\xi,\eta)&=(\varrho^+f,\varrho^-g)=-(\varrho^+f,\gamma\varrho^+(\Phi

1617: g))\\

1618:          &=-(T f,\Phi g)_{L^2(X_+,E)}+(f,T\Phi g)_{L^2(X_+,E)}=0.

1619:        \end{split}

1620: \end{equation} Hence $\La_+^0\perp \La_-^0$.

1621: 

1622: Next we show that $C_++C_-=\Id$: pick $\xi\in L^2(\Sigma,E)$ and $f\in

1623: C^\infty(\tilde X,\tilde E)$.   Green's formula gives

1624: \begin{equation}\begin{split}

1625:             ((C_++C_-)\xi,\gamma\varrho f)_\Sigma

1626: &= (\varrho^+r^+\tilde

1627: G\varrho^*\gamma \xi,\gamma\varrho f)_\Sigma-(\varrho^-r^-\tilde

1628: G\varrho^*\gamma\xi,\gamma\varrho f)_\Sigma\\

1629:       &=(TK_+\xi,f)_{X_+}-(\tilde G\varrho^*\gamma\xi,Tf)_{X_+}+

1630:         (TK_-\xi,f)_{X_-}-(\tilde G\varrho^*\gamma\xi,\tilde Tf)_{X_-}\\

1631:       &=-(\tilde G\varrho^*\gamma\xi,\tilde

1632: Tf)_X=-((\Id-P_{Z_0})\varrho^*\gamma

1633: \xi,f)_X\\

1634:       &=(-\gamma\xi,\varrho f)_\Sigma=(\xi,\gamma\varrho f)_\Sigma.

1635:            \end{split}

1636: \end{equation}

1637: Since $\gamma\varrho: C^\infty(\tilde X,\tilde E)\to C^\infty (\Sigma;

1638: E)$ is surjective, $(C_++C_-)\xi=\xi$.

1639: 

1640: In sum, $C_\pm$ are $L^2$--bounded with orthogonal range $\subset

1641: \La_\pm^0$

1642: and $C_++C_-=\Id$. Hence $C_\pm$ must be the orthogonal projections onto

1643: $\La_\pm^0$.

1644: 

1645: Finally, consider $\xi=\varrho^+f\in \La_+^0, f\in Z_+^{1/2}$. Then, by

1646: \eqref{G4.15} we have $\Phi f\in Z_-^{1/2}$ and thus

1647: $\gamma \xi=\varrho^-(\Phi f)\in \La_-^0$. Hence $\gamma$ interchanges

1648: $\La_+^0$ and $\La_-^0$. Since $\La_+^0\perp \La_-^0$ the spaces

1649: $\La_\pm^0$ must

1650: be

1651: Lagrangian and we are done.

1652: \end{proof}

1653: 

1654: \subsection{Continuous dependence of the \Cald}\label{sec42} Let us

1655: first

1656: describe the set--up in which we want to study the dependence of

1657: the \Cald\ on its input data:  We consider the class $\cD$ of Dirac

1658: operators on $E$ which are in cylindrical form near $\Sigma$ and the

1659: class

1660: $\mathcal{V}$ of auxiliary operators satisfying

1661: \textnormal{\eqref{pseudolocality}}. We say that a family $D_t\in\cD$

1662: varies continuously if all coefficients of $D_t$ (in any local chart)

1663: vary continuously. Moreover, we equip

1664: $\cV$ with the usual norm topology of $\cB(L^2(X,E))$.

1665: Finally, we put $(\cD\times \cV)(n)

1666: :=\bigl\{(D,V)\in\cD\times\mathcal{V}\,\big|\, \dim Z_0( D+V)=n\bigr\}$.

1667: 

1668: We start with some elementary statements on the continuous dependence of

1669: $P_{Z_0}$ and $\tilde G$:

1670: 

1671: \begin{prop}\label{contGtilde} The maps

1672: \begin{equation}

1673: \begin{split}P_{Z_0}, \tilde G&:(\cD\times\cV)(n)\to \cB(L^2(\tilde

1674: X,\tilde E),

1675: L^2_1(\tilde X,\tilde E)),\\

1676:      &P_{Z_0}(D,V):=\proj(\ker(\tilde D+\tilde V)),\\

1677:      &\tilde G(D,V):=(\Id-P_{Z_0}(D,V))(\tilde D+\tilde

1678: V+P_{Z_0}(D,V))^{-1},

1679:    \end{split}\label{ML-G3.21}

1680: \end{equation}

1681: are continuous.

1682: \end{prop}

1683: Note that this statement is slightly stronger than the continuity

1684: of $P_{Z_0}, \tilde G$ as maps into $\cB(L^2(\tilde X,\tilde E))$.

1685: 

1686: \begin{proof} Operators in $\cD$ are in cylindrical form near $\Sigma$,

1687: hence

1688: $\cD\times \cV\to \cB(L^2_1(\tilde X,\tilde E),L^2(\tilde X,\tilde E)),

1689: (D,V)\mapsto \tilde

1690: D+\tilde V$ is continuous. Thus also

1691: $\mathbb{C}\times\cD\times \cV\to \cB(L^2_1(\tilde X,\tilde

1692: E),L^2(\tilde

1693: X,\tilde E)), (\lambda,D,V)\mapsto \tilde

1694: D+\tilde V-\lambda$ is continuous. For a fixed pair $(D_0,V_0)\in

1695: (\cD\times\cV)$ there is an $\varepsilon>0$ such that $0$ is the only

1696: spectral point of $\tilde D_0+\tilde V_0$ in the closed disc

1697: $\{z\in\mathbb{C}\,|\,

1698: |z|\le \varepsilon\}$. By continuity this remains true for $(D,V)$ in

1699: a sufficiently small neighborhood of $(D_0,V_0)$ in

1700: $(\cD\times\cV)(n)$. For these $(D,V)$ we have

1701: \begin{equation}

1702:       P_{Z_0}(D,V)=\frac{1}{2\pi i}\oint_{|z|=\varepsilon}(\lambda-\tilde

1703:       D-\tilde V)^{-1}d\lambda.

1704: \end{equation}

1705: Since the inversion is continuous this proves the continuity statement

1706: on

1707: $P_{Z_0}$.

1708: 

1709: The continuity statement on $\tilde G(D,V)$ now follows immediately from

1710: the defining formula \eqref{ML-G3.21}.

1711: \end{proof}

1712: 

1713: The resolvent identity in Theorem

1714: \ref{Poissonop-b} shows that for $(D,V)\in(\cD\times \cV)(n)$

1715: the Poisson operator and hence $C_+(D+V)$

1716: depend continuously on $V$: namely $K_+(D+V)=K_+(D)-r^+(\tilde

1717: G(D,V)\tilde

1718: V\tilde D^{-1}+P_{Z_0}(D,V)\tilde D^{-1})\varrho^*\gamma$ and by the

1719: previous Proposition

1720: $$(D,V)\mapsto r^+(\tilde G(D,V)\tilde V\tilde D^{-1}

1721: +P_{Z_0}(D,V)\tilde D^{-1})\varrho^*\gamma$$

1722: is continuous

1723: $(\cD\times \cV)(n)\to \cB(L^2_{-1/2}(\Sigma,E),L^2_1(X_+,E))$.

1724: However, the dependence on

1725: $D$ itself is more subtle. In order to explore

1726: this we have to look a bit closer at the construction of the \Cald.

1727: 

1728: Fix a $c\in \mathbb{R}$ with $c\not\in\operatorname{spec} B$.

1729: We abbreviate the spectral projections

1730: $P_{>c}:=1_{(c,\infty)}(B),P_{\le c}:=1_{(-\infty,c]}(B)$ etc. Then we

1731: set,

1732: for $\xi\in L^2_s(\Sigma,E)$,

1733: \begin{equation}

1734:     Q(B)\xi(x):=e^{-xB}P_{\ge c}\xi, \quad x\ge 0.

1735: \end{equation}

1736: 

1737: We first need to study the mapping properties of $Q(B)$ with respect

1738: to Sobolev norms. The results we are going to present are fairly

1739: standard

1740: applications of the spectral theorem. However, to be as self--contained

1741: as possible we   present the details.

1742: 

1743: Let  $\epsilon >0$, $\epsilon <1$ and fix a smooth cut--off

1744: function $\varphi:\rr\to [0,1]$ with

1745: $\varphi(x)=1$ for $ |x| \le\epsilon/2$  and $\varphi(x)=0$ for $|x|\ge

1746: \epsilon$.

1747: Let $\psi:\rr\to \rr$ be any function in $

1748: C^\infty_0((-\epsilon,\epsilon)\setminus\{0\})$,  for example

1749: $\psi=\varphi'$.

1750: In the following proposition we only  use the restrictions of $\varphi$

1751: and $\psi$ to

1752: $\{x\ge 0\}$.

1753: 

1754: \begin{prop}\label{ML-S3.11}\hfill

1755: 

1756: \textnormal{1. } For all

1757: $s,s'\in\mathbb{R}$ we have $e^+\psi

1758: Q(B)\in\cB(L^2_s(\Sigma,E),L^2_{s'}(\tilde X,\tilde E))$.

1759: 

1760: \textnormal{2.}

1761: For $-1/2\le s\le 1/2$ the operator

1762: $\varphi Q(B)$ maps

1763: $L^2_s(\Sigma,E)$ continuously into $L^2_{s+1/2}(X_+,\tilde E)$.

1764: \end{prop}

1765: \begin{proof} First we need to recall the definition of Sobolev spaces

1766: on a  manifold with boundary.  Since $B$ is elliptic

1767: the Sobolev norms on $L^2_s(\Sigma,E)$ can be defined using

1768: $\Id+|B|$:

1769: \begin{equation}

1770:        \langle f,g\rangle_{L^2_s}:=\langle (\Id+|B|)^s f,(\Id+|B|)^s

1771: g\rangle_{L^2}

1772: \end{equation}

1773: for $f,g\in L^2_s(\Sigma,E)$. Furthermore, for a nonnegative integer

1774: $k\in \mathbb{Z}_+$ a Sobolev norm for $L^2_k(\mathbb{R}_+\times

1775: \Sigma,E)$

1776: is given by

1777: \begin{equation}

1778:       \|f\|_{L^2_k}^2=\sum_{j=0}^k \int_0^\infty

1779: \|\partial_x^j(\Id+|B|)^{k-j}f(x)\|_{L^2(\Sigma,E)}^2dx.\label{Sobnorm}

1780: \end{equation}

1781: The Sobolev norms on $X_+$ are obtained by  patching together  the

1782: usual interior Sobolev norms with the norms \eqref{Sobnorm}.

1783: 

1784: 1. By complex interpolation it suffices to prove the claim for $s'\in

1785: \mathbb{Z}_+$. Applying the Leibniz rule to

1786: $\partial^j(\Id+|B|)^{k-j}\psi

1787: Q(B)$ we see that in view of \eqref{Sobnorm} it is in fact sufficient

1788: to show that for any $k\in\mathbb{Z}_+$ the operator $(\Id+|B|)^k\psi

1789: Q(B)$ is bounded $L^2_s(\Sigma,E)\to L^2(X_+,E)$.

1790: 

1791: Furthermore, for $\xi\in L^2_s(\Sigma,E)$

1792: we have $(\Id+|B|)^{s}\xi\in L^2(\Sigma,E)$

1793: and 

1794: \[(\Id+|B|)^k\psi Q(B)\xi=(\Id+|B|)^{k-s}\psi

1795: Q(B)((\Id+|B|)^{s}\xi)\]

1796: so that  it suffices to prove the $L^2$--boundedness of

1797: $(\Id+|B|)^k\psi Q(B)$ for all $k$.

1798: 

1799: Next, since $\psi\in C^\infty_0(0,\epsilon)$ there is an $x_0>0$

1800: such that $\psi(x)=0$ for $x\le x_0$. For $x_0\le x\le\epsilon$ the

1801: spectral

1802: theorem gives

1803: \begin{equation}

1804:      \|(\Id+|B|)^ke^{-xB}P_{\ge c}\|_{L^2(\Sigma,E)}

1805: =\sup_{\lambda\ge c}|(1+\lambda)^ke^{-x\lambda}|\le C(k,x_0,c,\epsilon)

1806: \end{equation}

1807: and thus for $\xi\in L^2(\Sigma,E)$

1808: \begin{equation}\begin{split}

1809:        \|\psi (\Id+|B|)^kQ(B)\xi\|_{L^2(X_+,E)}^2&\le \|\psi\|_\infty^2

1810: \int_{x_0}^\epsilon

1811:               \|(\Id+|B|)^ke^{-xB}P_{\ge c}\xi\|_{L^2(\Sigma,E)}^2dx\\

1812:        &\le \epsilon \|\psi\|^2 C(k,x_0,c,\epsilon)^2 \|\xi\|^2.

1813:                 \end{split}

1814: \end{equation}

1815:   The proof of 1. now follows since $\psi$ has compact support in $(0,\infty)$

1816: and hence in this case extension by $0$ 

1817: is bounded $L^2_{s'}(X_+)\to L^2_{s'}(\tilde{X})$ for any $s'$.

1818: 

1819: \smallskip

1820: 

1821: 2. Let $(E_\lambda)_{\lambda\in\mathbb{R}}$ be the spectral resolution

1822: of

1823: $B$. Then we estimate for $\xi\in L^2(\Sigma,E)$

1824: \begin{equation}\label{ML-basicestimate}

1825:     \begin{split}

1826:         \|\varphi&(\Id+|B|)^{1/2}Q(B)\xi\|

1827: \le \int_0^\infty \| (\Id+|B|)^{1/2}\varphi(x)e^{-xB} P_{\ge

1828: c}(B)\xi\|^2dx\\

1829:      &\le \int_c^\infty \int_0^\epsilon (1+|\lambda|) e^{-2x\lambda}dx

1830: d\langle

1831:         E_\lambda \xi,\xi\rangle\\

1832:       &\le C(c,\epsilon) \|\xi\|^2.

1833:     \end{split}

1834: \end{equation}

1835: 

1836: For $\xi\in L^2_{-1/2}(\Sigma,E)$ we can now estimate using

1837: \eqref{ML-basicestimate}

1838: \begin{equation}\begin{split}

1839:       \|\varphi Q(B)\xi\|_{L^2(X_+,E)}&=\|\varphi

1840:       (\Id+|B|)^{1/2}Q(B)(\Id+|B|)^{-1/2}\xi\|_{L^2(X_+,E)}\\

1841:         &\le \|\varphi (\Id+|B|)^{1/2}Q(B)\|_{L^2(\Sigma,E)\to

1842:       L^2(X_+,E)}\|(\Id+|B|)^{-1/2}\xi\|_{L^2(\Sigma,E)}

1843:                 \end{split}

1844: \end{equation}

1845: proving the continuity of $\varphi Q(B)$ from $L^2_{-1/2}(\Sigma,E)$

1846: to $L^2(X_+,E)$.

1847: 

1848: Next consider $\xi\in L^2_{1/2}(\Sigma,E)$. Then analogously we find

1849: (taking

1850: $\psi=\varphi'$ and applying part 1)

1851: \begin{equation}\begin{split}

1852:       \|\partial_x\varphi Q(B)\xi\|_{L^2(X_+,E)}&\le \|\varphi'

1853:       Q(B)\xi\|_{L^2(X_+,E)}+ \|\varphi

1854:       B(\Id+|B|)^{-1/2}Q(B)(\Id+|B|)^{1/2}\xi\|_{L^2}\\

1855:      &\le C_1 \|\xi\|_{L^2}+C_2\|(\Id+|B|)^{1/2}\xi\|_{L^2}

1856:                 \end{split}

1857: \end{equation}

1858: and

1859: \begin{equation}\begin{split}

1860:       \|\varphi (\Id+|B|)Q(B)\xi\|_{L^2(X_+,E)}

1861:          &= \|\varphi (\Id+|B|)^{1/2}Q(B)(\Id+|B|)^{1/2}\xi\|_{L^2}\\

1862:      &\le C(c,\epsilon)\|(\Id+|B|)^{1/2}\xi\|_{L^2}.

1863:                 \end{split}

1864: \end{equation}

1865: In view of \eqref{Sobnorm} this proves the continuity of $\varphi Q(B)$

1866: from $L^2_{1/2}(\Sigma,E)$ to $L^2_1(X_+,E)$.

1867: 

1868: The complex interpolation method \cite[Sec. 4.2]{taylor} now yields the

1869: assertion for

1870: all $s\in [-1/2,1/2]$.

1871: \end{proof}

1872: 

1873: \newcommand{\Ell}{\operatorname{Ell}}

1874: \begin{prop}\label{ML-continuity} Let $\Ell^1(\Sigma,E)$ be the set

1875: of first order self--adjoint elliptic differential operators on

1876: $E|\Sigma$.

1877: Then for $\psi\in C^\infty_0((-\epsilon,\epsilon)\setminus{0})$ the map

1878: \[

1879:     \Ell^1(\Sigma,E)_c:=\{B\in\Ell^1(\Sigma,E)\,|\,

1880: c\not\in\operatorname{spec} B\}

1881:    \longrightarrow \cB(L^2(\Sigma,E),L^2(X_+,E)),\quad B\mapsto \psi Q(B)

1882: \]

1883: is continuous.

1884: \end{prop}

1885: \begin{proof} Again let $x_0>0$ be such that $\psi(x)=0$ for $0\leq x

1886: \le x_0$.

1887: By the spectral theorem the map

1888: \begin{equation}

1889:      [x_0,\epsilon]\times \Ell^1(\Sigma,E)_c\longrightarrow

1890: \cB(L^2(\Sigma,E)),

1891:     \quad (x,B)\mapsto e^{-xB} P_{\ge c}(B)

1892: \end{equation}

1893: is continuous.

1894: 

1895: Now fix $B_0\in \Ell^1(\Sigma,E)_c$ and a $\delta>0$. By compactness

1896: there is a neighborhood $W$ of $B_0$ such that for $

1897: B\in W$ and all $x_0\le x\le \epsilon$

1898: \begin{equation}

1899:      \|e^{-xB}P_{\ge c}(B)-e ^{-xB_0}P_{\ge c}(B_0)\|_{L^2(\Sigma,E)}\le

1900:      \delta/\sqrt{\epsilon}.

1901: \end{equation}

1902: Thus for $\xi\in L^2(\Sigma,E)$

1903: \begin{equation}

1904:     \|\psi Q(B)\xi-\psi Q(B_0)\xi\|^2\le C^2\int_{x_0}^\epsilon

1905:      \|e^{-xB}P_{\ge c}(B)-e ^{-xB_0}P_{\ge c}(B_0)\|_{L^2(\Sigma,E)}^2dx

1906:       \|\xi\|^2

1907:     \le C^2\delta^2\|\xi\|^2.

1908: \end{equation} 

1909: Thus $  \|\psi Q(B) -\psi Q(B_0) \| \le C\delta$,    completing the

1910: proof.

1911: \end{proof}

1912: 

1913: Let

1914: \begin{equation}\label{defapproxCald}

1915:        R(B)\xi(x):=\begin{cases}

1916:                   -e^{-xB}P_{\ge c}\gamma \xi  & \text{for }x\ge 0, \\

1917:                    e^{-xB}P_{\le c}\gamma \xi  & \text{for } x<0,

1918:      \end{cases}

1919: \end{equation}

1920: and define the {\em approximate    Poisson operator

1921: }   $K^0_\pm:=\pm r^\pm \varphi R(B)\gamma$.

1922: 

1923: Applying Proposition \ref{ML-S3.11} and \ref{ML-continuity}

1924: (and its obvious counterparts for $x\le 0$) to $R(B)$ gives:

1925: \begin{prop}\label{approxCald}\textnormal{1.} If $\psi\in

1926: C^\infty_0((-\epsilon,\epsilon)\setminus\{0\})$ then for all

1927: $s,s'\in\mathbb{R}$,

1928: $$\psi

1929: R(B)\in\cB(L^2_s(\Sigma,E),L^2_{s'}(\tilde X,\tilde E)).$$  Moreover,

1930: as an element of $\cB(L^2(\Sigma,E),L^2(\tilde X,\tilde E))$,

1931: $\psi R(B)$ depends continuously on the coefficients of $B$ as long

1932: as $c\not\in\operatorname{spec} B$.

1933: 

1934: \textnormal{2.} The map  $\varphi R(B)$ maps

1935: $L^2(\Sigma,E)$ continuously into

1936: $L^2_{1/2}(X_-\amalg X_+,\tilde E)$. 

1937: %Moreover, $\varphi R(B)\in

1938: %\cB(L^2(\Sigma,E),L^2_{1/2}(X_-\dot\cup X_+,\tilde E))$ depends

1939: %continuously on

1940: %the coefficients of $B$.

1941: \end{prop}

1942: 

1943: If $f\in C^\infty(\tilde X,\tilde E)$ then Green's formula gives

1944: \begin{equation}\begin{split}

1945:        (\varphi R\xi,\tilde T f)_{\tilde X}&=(r^+\varphi R\xi,r^+\tilde

1946: Tf)_{X^+}+(r^-\varphi R\xi,r^-\tilde

1947:        Tf)_{X^-}\\

1948:        &= (-\gamma P_{\ge c}\gamma\xi,\varrho f)_\Sigma-(\gamma P_{\le

1949:        c}\gamma\xi,\varrho f)_\Sigma+(S(B)\xi,f)_{\tilde X}\\

1950:        &=((\varrho^*+S(B))\xi,f)_{\tilde X}

1951:        \end{split}

1952: \end{equation}

1953: with $S(B)\xi=\varphi'\gamma R\xi$. In view of Prop.

1954: \ref{approxCald} we have in particular that $S(B)\in

1955: \cB(L^2(\Sigma,E),L^2(\tilde X,\tilde E))$ and that it depends

1956: continuously on the coefficients of $B$. Note furthermore, that

1957: $S(B)$ maps in fact into

1958: $L^2_{s,\textrm{comp}}(\tilde X\setminus\Sigma,\tilde E)$ for all $s\ge

1959: 0$. Hence we

1960: have

1961: shown that

1962: \begin{equation}

1963:       \tilde T\varphi R=\varrho^*+S(B).\label{ML-G4.21}

1964: \end{equation}

1965: 

1966: The definition of $K_\pm^0$ immediately gives

1967: \begin{equation}

1968:          \begin{split}

1969:            \varrho^+K_+^0\xi&:=\lim_{t\to 0+}\varrho_tK_+^0\xi=P_{\ge

1970: c}\xi,\\

1971:            \varrho^-K_-^0\xi&:=\lim_{t\to 0-}\varrho_tK_-^0\xi=P_{\le

1972: c}\xi.

1973:          \end{split}\label{ML-G4.22}

1974: \end{equation} Note that we assumed $c\not\in\operatorname{spec} B$.

1975: 

1976: \eqref{ML-G4.21} and \eqref{ML-G4.22} allow to express

1977: $K_\pm(\tilde D+\tilde V)$ and $C_\pm(\tilde D+\tilde V)$ in terms of

1978: the

1979: operator $R(B)$. Namely, we have in view of \eqref{ML-G4.21}

1980: \begin{equation}

1981:         (\Id-P_{Z_0})\varphi R=\tilde G\tilde T\varphi R=

1982:            \tilde G\varrho^*+\tilde G S(B)

1983: \end{equation} and hence

1984: \begin{equation}\begin{split}

1985:        K_+ &=r^+\tilde G\varrho^*\gamma\\

1986:            &=r^+(\Id-P_{Z_0})\varphi R\gamma-r^+\tilde GS(B)\gamma\\

1987:            &=K_+^0-r^+P_{Z_0}\varphi R\gamma-r^+\tilde GS(B)\gamma.

1988:           \end{split}

1989: \end{equation} Similarly,

1990: \begin{equation}

1991:        K_-=K_-^0+r^-P_{Z_0}\varphi R\gamma+r^-\tilde GS(B)\gamma.

1992: \end{equation} Finally, this gives     the formulas:

1993: \begin{equation}\begin{split}

1994:        C_+&=P_{\ge c}(B)-\varrho^+\tilde G S(B)\gamma,\\

1995:        C_-&=P_{\le c}(B)+\varrho^-\tilde G S(B)\gamma.

1996: \end{split}\label{ML-G4.26}

1997: \end{equation}

1998: In view of Proposition \ref{contGtilde} and Proposition \ref{approxCald}

1999: the operator

2000: $\varrho^\pm\tilde G S(B)\gamma$ maps $L^2(\Sigma,E)$ continuously into

2001: $L^2_{1/2}(\Sigma,E)$ and, as long as $\dim Z_0$ is constant, it

2002: depends continuously on $D$ and $V$.

2003: 

2004: Since $c\not\in\operatorname{spec}(B)$ the operators $P_{\ge c}(B),

2005: P_{\le

2006: c}(B)$ depend continuously on $B$. Note that the choice of $c$ is

2007: irrelevant. To show the continuity of

2008: $C_+$ at $D$ and $V$ we choose a $c$ not in $\operatorname{spec}(B)$.

2009: Then

2010: we will have $c\not\in\operatorname{spec}(B)$ in a full neighborhood of

2011: $D$ and

2012: $V$. Of course, $S(B)$ also depends on $c$.

2013: 

2014: Summing up we have proved:

2015: 

2016: \begin{thm}\label{continuityCald} We consider the class $\cD$ of Dirac

2017: operators on $E$ which are in cylindrical form near $\Sigma$ and the

2018: class

2019: $\mathcal{V}$ of auxiliary operators satisfying

2020: \textnormal{\eqref{pseudolocality}}. We say that a family $D_t\in\cD$

2021: varies continuously if all coefficients of $D_t$ (in any local chart)

2022: vary

2023: continuously. Moreover, we equip

2024: $\mathcal{V}$ with the usual norm topology of $\cB(L^2(X,E))$. Then for

2025: each

2026: $n\in\mathbb{Z}_+$ the map

2027: \[ \bigl\{(D,V)\in\cD\times\mathcal{V}\,\big|\, \dim Z_0( D+V)=n\bigr\}

2028:       \longrightarrow \cB(L^2(\Sigma,E)), \quad (D,V)\mapsto C( D+ V)

2029: \] is continuous.

2030: \end{thm}

2031: 

2032: Note also that in view of \eqref{ML-G4.26} the operator $C-P_{\ge

2033: 0}(B)$

2034: maps $L^2$ continuously into $L^2_{1/2}$, i.e. `is  an operator of order

2035: $-1/2$'. If $V$ is a pseudodifferential operator then one easily shows

2036: that $\varrho^\pm \tilde G S(B)$ is a smoothing operator and we recover,

2037: though with a different argument, the well--known fact that $C-P_{\ge

2038: 0}(B)$ is smoothing

2039: \cite[Prop. 2.2]{Scott}, \cite[Prop. 4.1]{Grubb}.

2040: 

2041: %%MARKE

2042: \subsection{Application to the perturbed Dirac operator}

2043: 

2044: We turn to our specific operator $D_{A,f}=D_A-4\pi^2 Q_{A,f}$.   The

2045: discussion in Section \ref{secloc} shows that

2046: $Q_{A,f}$ satisfies \eqref{pseudolocality}. Hence the general results of

2047: Sections \ref{sec41} and \ref{sec42} apply. In particular there exist

2048: the

2049: Poisson operator $K_+(D_{A,f})$ and the Calder{\'o}n projector

2050: $P_{A,f}:=C_+(D_{A,f})$. Next we show that in a neighborhood of the flat

2051: connections and for ``small'' $f$  the weak unique continuation property

2052: with respect to $\partial X$ holds for $D_{A,f}$.

2053: 

2054: \begin{lemma}\label{nbdU}  There is an open neighborhood

2055: $U\subset

2056: \cA$ of the set of flat connections and a constant $C$ (which depends

2057: only

2058: on $X$ as a Riemannian manifold) so that given any

2059: $A\in U$,

2060: $$\|\tilde D_{A}^{-1}\|\leq C.$$

2061:      \end{lemma}

2062:      \begin{proof} Given a flat connection

2063: $A$ and a gauge transformation $g$ over $X$ the operators $D_A$ and

2064: $D_{g\cdot A}$ are unitarily equivalent (i.e. $D_{gA}(a)=gD_A

2065: g^{-1}(a)$),

2066: and this equivalence extends to $\tilde D_A$ since $\cG$ acts on the Lie

2067: algebra coefficients and $\gamma=\pm *$ acts on the differential forms,

2068: thus $g_x\gamma_x(v)=\gamma_x(g_x(v))$ for $x\in \Sigma$. Hence the

2069: spectra of

2070: $\tilde{D}_A$ and $\tilde{D}_{gA}$ coincide.

2071: 

2072: The space $\cM$ of flat connections modulo gauge  transformations on $X$

2073: is compact, being homeomorphic to the  compact  real algebraic variety

2074: $\Hom(\pi_1X,SU(n))/\text{conjugation}$. Thus   there is a lower bound,

2075: say $k$, on the absolute value of the smallest eigenvalue of

2076: $\tilde{D}_A$ as $A$ varies in the set of flat connections on $X$.  It

2077: follows that the largest eigenvalue of $\tilde{D}_A^{-1}$ is bounded by

2078: $1/k$, and hence $\|\tilde{D}_A^{-1}\|\leq 1/k$. Set

2079: $C=2/k$, then by continuity the set $U=\{A\in\cA\ |\

2080:      \|\tilde{D}_A^{-1}\|\leq C\}$ is a neighborhood of the set of flat

2081: connections.

2082: \end{proof}

2083: 

2084: \begin{prop}\label{lem4.2} There is an $\epsilon>0$ so that given any

2085: connection $A$ in the neighborhood $U$ of Lemma \textnormal{\ref{nbdU}},

2086: and any

2087:      $f\in\cP$ satisfying $\|f\|_{C^r}<\epsilon$, then

2088: $\tilde D_A-4\pi^2\tilde Q_{A,f}$ is invertible and hence has the weak

2089: unique continuation property with respect to

2090: $\partial X$.

2091: \end{prop}

2092: \begin{proof}  The Neumann series shows that $\tilde D_A-4\pi^2\tilde

2093: Q_{A,f}$ is invertible  if $\|\tilde D_A^{-1} 4\pi^2 \tilde

2094: Q_{A,f}\|<1$.

2095: 

2096: By Lemmas \ref{nbdU} and

2097: \ref{estimate} there exists a constant $C$ independent of $A$ and $f$

2098: such that

2099: $$\|\tilde D_A^{-1}4\pi^2 \tilde Q_{A,f}\|\le 4\pi^2 \|\tilde D_A^{-1}\|

2100: \|\tilde Q_{A,f}\|\le C\|f\|_{C^2}.$$ Hence we may choose

2101: $\epsilon:=\tfrac{1}{C}$. Proposition \ref{WUCP} then sows that

2102: $D_{A,f}$

2103: has the weak unique continuation property.

2104: \end{proof}

2105: 

2106: \begin{thm}\label{thm5.3} The map

2107: $U\times B_{\cP}(0,\epsilon)\to \cB(L^2(\Sigma,E))$ sending a pair

2108: $(A,f)$ to the Calder\'on projector $C_{A,f}$ of $D_{A,f}$ is

2109: continuous,

2110: and hence the Cauchy data spaces $\Lambda_{A,f}:=

2111: \operatorname{range}(C_{A,f})$ vary continuously (in the gap metric on

2112: the

2113: Grassmannian of Lagrangian subspaces of $L^2(\Sigma,E)$) with respect to

2114: $A,f$.

2115: \end{thm}

2116: \begin{proof} By Prop. \ref{lem4.2} we have $Z_0(\tilde D_A-4\pi^2\tilde

2117: Q_{A,f})=0$ for all $(A,f)\in \cA\times B_{\cP}(0,\epsilon)$. Thus the

2118: claim follows from Theorem \ref{continuityCald}.

2119: \end{proof}

2120: 

2121: \section{An application}\label{application} Theorem \ref{thm5.3} allows

2122: us

2123: to use the technology developed in \cite{nico} and \cite{daniel-kirk}

2124: to

2125: study how the spectral flow of a path of operators

2126: $D_{A_t,f_t}$ on a closed 3-manifold behaves with respect to a

2127: decomposition along a separating surface. In particular, the  proof of

2128: Nicolaesu's adiabatic limit theorem and Lemma 3.2 of \cite{daniel-kirk},

2129: which consider the effect of stretching the collar of $X$ on the Cauchy

2130: data spaces,  do not depend on the operators being Dirac operators, but

2131: only on the fact that they have the Atiyah--Patodi--Singer form on a

2132: collar.  Thus     Theorem 5.1 of

2133: \cite{daniel-kirk}, and all its consequences (Section 6 of

2134: \cite{daniel-kirk}) are true for the operators

2135: $D_{A,f}$.

2136: 

2137: This will be used crucially in forthcoming   calculations of the

2138: $SU(3)$ Casson invariant for Brieskorn spheres

2139: \cite{BHK-pqr}.  We finish this article with a useful  application of

2140: our

2141: main result. We outline a different proof of a result of Taubes which is

2142: the main ingredient in his proof that his invariant equals Casson's

2143: $SU(2)$

2144: invariant for homology 3-spheres, and is the starting point for proofs

2145: of

2146: various approaches to problems known as the  Atiyah--Floer conjecture.

2147: It is also  one of the simplest

2148: non-trivial examples to which this machinery can be applied, since

2149: this situation avoids the delicate problem of singularities in the

2150: moduli space.

2151: 

2152: \bigskip

2153: 

2154:    First, we set up  some notation. Let $X$ be

2155: a homology 3-sphere.  Let

2156: $P\to X$ be  the trivial principal

2157: $SU(2)$ bundle  and let $\cA(X)$ denote the ($L^2_1$ completion of the)

2158: space of

2159: $SU(2)$ connections on

2160: $P$. Denote by $\cG(X)$ the ($L^2_2$ completion of the) gauge group

2161: $\Map(X,SU(2))$.  Then let $\cA^*(X)$ be the subspace of $\cA(X)$

2162: consisting of irreducible connections, that is, those connections whose

2163: stabilizer in  $\cG(X)$ is equal to  the center $\{\pm \Id \}

2164: \subset\cG(X)$. Thus $\cA^*(X)/\cG(X)$ is a smooth Hilbert manifold.

2165: 

2166:     Suppose that

2167: $X$ is given a Heegard decomposition

2168: $X=X_1\cup_\Sigma X_2$ of genus $g>2$. Choose    embeddings $\gamma_i$

2169: of

2170: solid tori so that $S=\cup_i\gamma_i(D^2\times S^1)$   is contained in

2171: the

2172: interior of

2173: $X_2$, and  choose   $f\in \cP$ so   that  the restriction of $cs+h_f$

2174: to

2175: the space $\cA^*(X)$ of irreducible connections  has only

2176: non-degenerate

2177: critical points (i.e.~the kernel of

2178: $D_{A,f}$ is zero at every irreducible perturbed flat connection $A$).

2179: In

2180: this context the fact  that such $f$ are dense in

2181:    any  neighborhood of $0\in\cP$   is proven in

2182: \cite{taubes} and

2183: \cite{floer}. We will  restrict $f$ further below.

2184: 

2185: It follows that the critical points of

2186: $cs+h_f:\cA^*(X)/\cG(X)\to \rr/\zz$ are finite and isolated.

2187:     Let $\cM_f^*(X)$ denote the critical set, i.e.~ the moduli space of

2188: irreducible perturbed flat connections on $X$:

2189: $$ \cM_f^*(X)=\{A\in\cA^*(X)\ | \ d(cs+h_f)(A)=0\}/\cG(X).$$

2190: 

2191:    To a pair $[A_0], [A_1]\in

2192: \cM^*_f(X)$, Taubes assigns the sign

2193: $$\Sign_T([A_0],[A_1])= (-1)^{\SF(D_{A_t,f})} $$ where $A_t$ is any path

2194: in $\cA^*(X)$ joining lifts of $[A_0]$ and $[A_1]$. In fact an

2195: application

2196: of the  index theorem shows that the mod $8$ reduction of

2197: $\SF(D_{A_t,f})$

2198: depends only on the gauge equivalence classes $[A_0]$ and $[A_1]$.

2199: 

2200: Restricting irreducible connections to the pieces of the  Heegard

2201: decomposition leads to a diagram (with the obvious notation)

2202: \[

2203: \begin{diagram}\dgsquash[3/4]

2204: \node[2]{\cM^*(X_1)}\arrow{se}\\

2205: \node{\cM^*_f(X)}\arrow{ne}\arrow{se}\node[2]{\cM^*(\Sigma)}\\

2206: \node[2]{\cM^*_f(X_2)}\arrow{ne}

2207: \end{diagram}

2208: \] All arrows are embeddings. The space $\cM^*(\Sigma)$ is a smooth

2209:      $6g-6$ dimensional  symplectic manifold. (We will describe the

2210: symplectic structure below.)    The subspaces

2211: $\cM^*(X_1)$ and

2212: $\cM_f^*(X_2)$ are smooth  and

2213: orientable $3g-3$ dimensional Lagrangian  submanifolds

2214: (\cite{taubes,Herald1}). They intersect transversely in the compact

2215: $0$--dimensional manifold $\cM_f^*(X)$ for   appropriate $f$,

2216: namely those perturbations $f$ so that $cs+h_f$ has only

2217: non-degenerate

2218: critical points in $\cA^*(X)$.   Notice that since

2219: $S$ lies in the interior of

2220: $X_2$,  perturbed flat connections restrict to flat connections on $X_1$

2221: and $\Sigma$.

2222: 

2223: We also assume that $\|f\|_{C^2}$ is small enough so  the following

2224: requirements  hold:

2225: \begin{enumerate}

2226: \item The space $\cM^*_f(X_2)$ is diffeomorphic to the flat (i.e.

2227: unperturbed) moduli space $

2228: \cM^*(X_2)$. This is possible since, as $f$ varies, $\cM^*_f(X_2)$ varies

2229: by

2230: a (Legendrian) cobordism (\cite{Herald1}) and so for small enough $f$

2231: the

2232: cobordism is a product. (Some care must be taken since $\cM^*_f(X_2)$

2233: is non-compact,

2234: but the results of \cite{Herald1}  can be used to reach this conclusion.

2235: Alternatively, Taubes shows that the perturbations can be chosen to fix

2236: the reducible

2237: stratum.)

2238: \item The perturbation should be small enough so that $\cM^*_f(X_2)$

2239: lies

2240: in the neighborhood $U$ specified by Lemma \ref{nbdU}. That this  is

2241: possible follows again from  \cite{Herald1}.

2242: \item The perturbation should be small enough so that  the requirement

2243: $\|f\|<\epsilon$ of  Proposition \ref{lem4.2} holds on $X_2$.

2244: \end{enumerate}

2245: 

2246: Then Theorem

2247: \ref{thm5.3} implies that the map taking $A\in \cM^*_f(X_2)$ to the

2248: Cauchy data space $\Lambda_{A,f}$ for $D_{A,f}$ is continuous.

2249: 

2250:     In this language Casson assigns the sign (after  orienting

2251:     the  moduli spaces $\cM^*(X_1),\ \cM_f^*(X_2)$ and $\cM^*(\Sigma)$)

2252: \[\Sign_C([A_0],[A_1])=\begin{cases} 1 & \text{if } 

2253:  \cM^*(X_1)\cdot_{[A_1]}\cM_f^*(X_2)=\cM^*(X_1)\cdot_{[A_0]}\cM_f^*(X_2)

2254:  \text{ and,}\\

2255:  -1  &  \text{otherwise}, 

2256: \end{cases}\]

2257: where

2258: $\cM^*(X_1)\cdot_{[A_1]}\cM_f^*(X_2)\in \{\pm 1\} $ denotes the local

2259: transverse

2260: intersection  number of $\cM^*(X_1)$ and $\cM_f^*(X_2)$ in

2261: $\cM^*(\Sigma)$

2262: at

2263: $[A_1]$.   

2264: 

2265: Taubes' proof that his invariant equals Casson's invariant (up to an

2266: overall sign which depends on a method for fixing particular

2267: orientations

2268: of all the moduli spaces)  reduces to the assertion that

2269: $\Sign_T=\Sign_C$. Notice that we are looking at relative signs, so that

2270: the  validity of $\Sign_T=\Sign_C$ is independent of the choice of

2271: orientations of the moduli  spaces.

2272: 

2273: We begin with a lemma whose proof appears  to be well  known to those

2274: who

2275: know it well  (e.g. \cite{dostoglou-salomon}), although  we were  unable

2276: to  find  an explicit statement in the literature.  The

2277:      statement is implicit in \cite{daska},  but is described in the

2278: language of stable holomorphic bundles.  We are indebted to G.

2279: Daskalopoulos for help with  the argument.

2280: 

2281: \begin{lemma}\label{lemma6.1} The  space $\cF^*(\Sigma)$ of irreducible

2282: flat

2283: $SU(2)$ connections over a closed 2-manifold $\Sigma$ of genus greater

2284: than 2 is   simply connected. The moduli space

2285: $\cM^*(\Sigma)$ of such connections is also  simply connected.

2286: 

2287: \end{lemma}

2288: 

2289: \begin{proof}   Before we start the proof, we remark that in the

2290: statement

2291: of  Lemma \ref{lemma6.1}, the space $\cF^*(\Sigma)$ can mean either the

2292: smooth flat irreducible connections, or the $L^2_r$ completion of this

2293: space for any $r$ large enough so that the

2294: $L^2_{r+1}$ completion of the gauge group consists of continuous gauge

2295: transformations,   e.g.~$r>0$.

2296: 

2297: We first set up notation.  We consider the trivial rank 2 smooth

2298: vector bundle

2299: $E=\cc^2\times\Sigma\to \Sigma$  endowed with a Hermitian

2300: metric.  We fix a holomorphic structure on

2301: $\Sigma$.  The space $\cA$ of

2302: $SU(2)$ connections on $E$  is  

2303: identified with the space  of holomorphic structures with

2304: holomorphically trivial determinant on

2305: $E$ via the standard construction (see

2306: \cite{Atiyah-Bott}) since $\Sigma$ is 1 complex--dimensional and so every

2307: structure is integrable. We denote by   $\cA^s$  the subspace of stable

2308: holomorphic structures on $E$ (see \cite{Atiyah-Bott}).  We take the

2309: $L^2_r$ completions of these spaces.

2310: 

2311: Denote by $\cF \subset \cA$ the subspace of flat connections, and

2312: by $\cF^* $ the subspace of irreducible flat connections.   We

2313:   take $\cG$ to be the ($L^2_{r+1}$

2314: completion of the) $SU(2)$ gauge group

2315: $\text{Map}(\Sigma,SU(2))$.

2316: 

2317: The slice theorem (see e.g.~\cite{donaldson-kronheimer}) shows that

2318: the

2319: quotient map $\cA\to

2320: \cA /\cG $, when restricted to the irreducible connections

2321: $\cA^*$    gives a locally trivial map, and hence a fibration

2322: $\cA^*\to \cA^*/\cG$ with fiber

2323: $\cG/\pm\Id$. Restricting to the flat irreducible connections

2324: gives the  fibration

2325: \begin{equation}\label{eq6.8}\cG /{\pm\Id}\to

2326: \cF^*\to \cM^*.

2327: \end{equation}

2328: 

2329: That $\cM^*$ is simply connected follows immediately from  \cite[Theorem

2330: 7.1]{daska} and the fact

2331:  that  $\cM^*$ is

2332: homeomorphic to

2333: the space of stable rank 2 holomorphic bundles over $\Sigma$ with

2334: trivial

2335: determinant (see \cite{narasimhan-sheshadri, Atiyah-Bott}).

2336: 

2337: The calculations in \cite{daska} show that

2338: $\pi_2(\cM^*)=\zz\oplus\zz/2$ and it is not  hard to show  that

2339: $\pi_1(\cG/{\pm\Id})=\zz\oplus\zz/2$.  If we knew that   the

2340: map induced by the connecting homomorphism  in the long exact sequence

2341: of

2342: homotopy groups for the fibration \eqref{eq6.8} were an isomorphism (or

2343: merely surjective), then the  result would follow from  this long exact

2344: sequence

2345: since

2346: $\pi_1(\cM^*)=0$.  This is true, in fact the results of \cite{daska},

2347: though stated for $U(2)$ instead of $SU(2)$, are equally valid for

2348: $SU(2)$.

2349: 

2350: Alternatively, we can show that  $ \pi_1(\cF^*)$ is zero using the

2351: result

2352: of \cite{rade}. Fix a flat irreducible connection $A\in \cF^* $. Let

2353: $\gamma:S^1\to \cF$ be a loop of flat connections. Corollary 2.7 of

2354: \cite{daskalopoulos-uhlenbeck}  shows that

2355: $\cA^s $ is simply connected, and since

2356: $\cF^*\subset \cA^s$ (see \cite{Atiyah-Bott})  

2357: $\gamma$ bounds a disc

2358: $\bar{\gamma}:D^2\to \cA^s $. The main result of \cite{rade}

2359: asserts that

2360: the Yang-Mills flow on $\cA$, restricted to

2361: $\cA^s$, defines a strong deformation retraction of $\cA^s$ to

2362: $\cF^*$. Thus $\gamma$ bounds a disc in $\cF^*$, completing the proof.

2363: 

2364: \end{proof}

2365: 

2366: The space of  irreducible $SU(2)$ representations of a free group is

2367: easily proven to be path connected. Thus the space

2368: $\cM^*(X_1)$, homeomorphic to the space of conjugacy classes of

2369: irreducible representations of $\pi_1(X_1)$,   is path connected. There

2370: is

2371: a fibration

2372: $$\cG(X_1)/\pm \Id\to\cF^*(X_1)\to \cM^*(X_1)$$ as in

2373: the

2374: proof of Lemma \ref{lemma6.1}. The gauge group

2375: $\cG(X_1)=\text{Map}(X_1,SU(2))$ is path connected since $X_1$ has the

2376: homotopy type of a 1-complex. Thus $\cF^*(X_1)$ is path connected.

2377: The same argument shows that

2378:     the space of irreducible perturbed flat connections on

2379: $X_2$ is

2380:     also path connected, since   the  perturbation was chosen small

2381: enough

2382: so that $\cM^*_f(X_2)$ is diffeomorphic to the unperturbed

2383: moduli space $\cM^*(X_2)$.

2384: 

2385: Combining the facts of the preceding paragraph, Lemma \ref{lemma6.1},

2386: and the observation that

2387: since the space of connections on a manifold is contractible, any family

2388: of connections on a submanifold of

2389: $X$ extends to a family of connections on all of $X$,  we obtain the

2390: following useful proposition.

2391: 

2392: \begin{prop} Given perturbed flat irreducible $SU(2)$ connections $A_0,

2393: A_1$ on the closed 3-manifold $X$, there exists a 2-parameter family

2394: $A_{s,t},\ s,t\in[0,1]$    of irreducible connections on $X$ satisfying:

2395: \begin{enumerate}

2396: \item $A_{0,t}=A_0$  and $A_{1,t}=A_1$ for all

2397: $t\in[0,1]$.

2398: \item $ A_{s,0} $ restricts to a path  of irreducible flat connections

2399: on

2400: $X_1$.

2401: \item $A_{s,1}$ restricts to a path of irreducible perturbed flat

2402: connections on

2403: $X_2$.

2404: \item $A_{s,t}$ restricts to an irreducible  flat connection

2405: $a_{s,t}$ on

2406: $\Sigma$ for all  $s,t$.

2407: \item The restriction of $A_{s,t}$ to a collar of $\Sigma$ is in product

2408: form:

2409: $A_{s,t}|_{\Sigma\times[-1,1]}=\tfrac{\del}{\del u} + a_{s,t}$.

2410: \qed\end{enumerate}

2411: \end{prop}

2412: 

2413: We now give an argument that $\Sign_T=\Sign_C$.  The strategy is to

2414: identify

2415: $\Sign_T$ with an infinite--dimensional Maslov index of the Cauchy data

2416: spaces, then

2417: to use  homotopy and stretching arguments to symplectically reduce to a

2418: finite--dimensional Maslov index which we identify with $\Sign_C$. We

2419: recommend that

2420: the reader who is unfamiliar with the type of argument we use look at

2421: the ``user's

2422: guide'' section of

2423: \cite{daniel-kirk}.

2424: 

2425: Let $\La^1_{s,t}$ denote the  Cauchy data space for the restriction of

2426: the

2427: operator

2428: $D_{A_{s,t},f}$ to $X_1$. Similarly define $\La^2_{s,t}$. These vary

2429: continuously by Theorem \ref{thm5.3}.  Note that in contrast to Section 3, the superscripts 

2430: $1$ and $2$ refer to the Cauchy data spaces for the two parts of the Heegard decomposition, 

2431: not to the Sobolev indices. Thus $\La^1_{s,t}$ and $\La^2_{s,t}$ are continuous families of Lagrangian subspaces of $L^2(\Om^*_\Sigma(\ad P))$. Moreover, for each $(s,t)$, $(\La^1_{s,t},\La^2_{s,t})$ is  a Fredholm pair.  Hence the Maslov index $\Mas(\La^1_\alpha,\La^2_\al)\in\zz$ is defined

2432: for any path $\al:[0,1]\to [0,1]\times[0,1]$ (for details see \cite{kirk-lesch}).

2433: 

2434:     Nicolaescu's theorem  says that the spectral flow of the family

2435: $D_{A_{s,0},f},

2436: \ s\in[0,1]$ equals the (infinite--dimensional) Maslov index

2437: $\Mas(\La^1_{s,0},\La^2_{s,0})$ (see \cite{nico,kirk-lesch}).

2438: Hence

2439: $$\Sign_T([A_0],[A_1])=(-1)^{\SF(D_{A_{s,0}})}

2440: =(-1)^{\Mas(\Lambda^1_{s,0},\Lambda^2_{s,0})}.$$

2441: We will

2442: use

2443: the homotopy properties of the Maslov index to find a simpler,

2444: finite--dimensional expression  for $\Mas(\La^1_{s,0},\La^2_{s,0})$ which

2445: can  be compared to the difference in intersection numbers.

2446: 

2447: The reduction  will be based on the following observations. Since the

2448: connection $a_{s,t}$ on $\Sigma$ is flat for all $s,t$,  the kernel of

2449: $S_{a_{s,t}}$ is identified via the Hodge theorem with  the cohomology

2450: $H^*(\Sigma;\ad P_{a_{s,t}})$ of the twisted de Rham complex

2451: \begin{equation}\label{eq6.10}  0\to \Omega^0(\Sigma,\ad

2452: P_{a_{s,t}})\xrightarrow{d_{a_{s,t}}}

2453: \Omega^1(\Sigma,\ad P_{a_{s,t}})\xrightarrow{d_{a_{s,t}}}

2454: \Omega^2(\Sigma,\ad P_{a_{s,t}})\to 0.

2455: \end{equation}

2456: 

2457: Since $a_{s,t}$ is irreducible,

2458: $H^0(\Sigma;\ad P_{a_{s,t}})=0$ (i.e.~using \eqref{eq2.3} for $\Sigma$)

2459: and so

2460: $H^2(\Sigma;\ad P_{a_{s,t}})=0$ by Poincar\'e duality.  Thus the kernel

2461: of

2462: $S_{a_{s,t}}$ is just

2463: $H^1(\Sigma;\ad P_{a_{s,t}})$. But this   is canonically identified

2464: with the tangent space of

2465: $\cM^*(\Sigma)$ at $[a_{s,t}]$, since the image of

2466: $d_{a_{s,t}}:\Omega^0(\Sigma,\ad P_{a_{s,t}})\to

2467: \Omega^1(\Sigma,\ad P_{a_{s,t}})$ is the tangent space of the

2468: $\cG(\Sigma)$

2469: orbit and   $d_{a_{s,t}}:\Omega^1(\Sigma,\ad P_{a_{s,t}})\to

2470: \Omega^2(\Sigma,\ad P_{a_{s,t}})$ is the linearization at $a_{s,t}$ of

2471: the

2472: curvature map

2473: $a\mapsto F(a)$.

2474: 

2475:     Via this identification, the cup product (or wedge product if one

2476: identifies cohomology with  harmonic forms)

2477: $$H^1(\Sigma;\ad P_{a_{s,t}})\times H^1(\Sigma;\ad P_{a_{s,t}})

2478: \xrightarrow{\cup}H^2(\Sigma;\rr)=\rr$$ defines a symplectic structure

2479: on

2480: $\cM^*(\Sigma)$. In particular, since

2481: $\cM^*(\Sigma)$ is a smooth $6g-6$ dimensional manifold,

2482: $\ker S_{a_{s,t}}=H^1(\Sigma;\ad P_{a_{s,t}})$ is a smoothly varying

2483: family of symplectic subspaces of $L^2(\Omega^*_\Sigma(\ad P))$.  For

2484: convenience we use the notation

2485: $$H^1_{s,t}(\Sigma)=H^1(\Sigma;\ad P_{a_{s,t}}).$$

2486: 

2487: Let $P^+_{s,t}$ denote the ($L^2$ closure of the) positive eigenspan of

2488: the

2489: tangential operator

2490: $S_{a_{s,t}}$, and let  $P^-_{s,t}$ denote its negative eigenspan.

2491: Thus

2492: we have a smoothly varying (in $s,t$) decomposition

2493: $$L^2(\Omega^*_\Sigma(\ad P))=P^-_{s,t}\oplus H^1_{s,t}(\Sigma)\oplus

2494: P^+_{s,t}.$$

2495: 

2496: \begin{lemma} \label{lem6.1} For each $s\in [0,1]$, let

2497: $T^1_{s,0}\subset

2498: H^1_{s,0}(\Sigma)$  be the   subspace defined  by

2499: $$T^1_{s,0}=\text{\rm image }H^1(X_1;\ad P_{a_{s,0}})\to H^1(\Sigma;\ad

2500: P_{a_{s,0}}).$$ Then the spaces $T^1_{s,0}$ are Lagrangian  and vary

2501: continuously. Via the identification  $T_{a_{s,0}}\cM^*(\Sigma)\cong

2502: H^1_{s,0}(\Sigma)$ the subspace $T^1_{s,0}$ corresponds to the tangent

2503: space at $a_{s,0}$ of the submanifold $\cM^*(X_1)$ of $\cM^*(\Sigma)$.

2504: 

2505: Moreover

2506: $$\SF(D_{A_{s,0},f})=\Mas(\La^1_{s,0},\La^2_{s,0})=

2507: \Mas(P^+_{s,0}\oplus T^1_{s,0},\La^2_{s,0}).$$

2508: \end{lemma}

2509: 

2510: \begin{proof}The statement of Lemma \ref{lem6.1}  refers only to the

2511: parameters $(s,0), s\in [0,1]$; in other words, throughout the proof the

2512: ``$t$'' parameter is zero.

2513: 

2514: The fact  that $T^1_{s,0}$ is Lagrangian is a  standard fact in

2515: geometric

2516: topology whose easy proof (based on Poincar\'e duality) we leave to the

2517: reader.  The identification of

2518: $T^1_{s,0}$ with the tangent space of $\cM^*(X_1)$ is also standard; one

2519: of many proofs is to consider the map of  de Rham complexes

2520: $\Omega^*(X_1;\ad P)\to \Omega^*(\Sigma;\ad P)$ in light of the diagram

2521: \eqref{eq2.3}.

2522:    Since $\cM^*(X_1)$ is a

2523: smooth submanifold, it follows that $T^1_{s,0}$ varies smoothly. Thus we

2524: focus on the statements about the Maslov index.

2525: 

2526: It follows from the adiabatic limit theorem of \cite{nico} and its

2527: refinement in

2528: \cite{daniel-kirk} that   replacing the collar of $X_1$ by an

2529: increasingly

2530: longer collar $\Sigma\times [-R,0]$ gives a homotopy of the path

2531: $s\mapsto

2532: \La^1_{s,0}$ to

2533: $s\mapsto P^+_{s,0}\oplus T^1_{s,0}$.  This is a consequence of the

2534: fact that

2535: there are no

2536: $L^2$ solutions to $D_{A_{s,0},f}\phi=0$ on  $(\Sigma\times

2537: (-\infty,0])\cup

2538: X_1$, which follows from \cite{APS} and two facts:

2539: \begin{enumerate}

2540: \item  The  restriction map

2541: $H^1(X_1;\ad P_{a_{s,0}})\to H^1(\Sigma;\ad P_{a_{s,0}}) $  is

2542: injective.

2543: \item The space $T^1_{s,0}$  equals  the {\em limiting values of

2544: extended

2545: $L^2$ solutions} in the sense of \cite{APS}. This was first observed by

2546: Yoshida \cite{yoshida} and in the present notation a proof can be found

2547: in

2548: \cite{daniel-kirk} and \cite{kirk-lesch}.

2549: \end{enumerate}

2550: 

2551: Let $r\mapsto \La^1_{s,0}(r)$ denote this homotopy. To be explicit,

2552: \[

2553: \La^1_{s,0}(r)=\begin{cases}\text{the Cauchy data space for   }

2554: D_{A_{s,0},f}

2555: \text{ on }(\Sigma\times [-\tfrac{1}{1-r},0])\cup X_1&\text{if } r<1,\\

2556: P^+_{s,0}\oplus T^1_{s,0}&\text{if } r=1.

2557: \end{cases}

2558: \]

2559: 

2560: This is not a homotopy rel endpoints, and so to see why the Maslov

2561: indices

2562: nevertheless agree, it suffices to show that  the dimensions of the

2563: intersections  $\La^1_{0,0}(r)\cap

2564: \La^2_{0,0}$ and

2565: $\La^1_{1,0}(r)\cap

2566: \La^2_{1,0}$ are independent of $r$, in fact these intersections are

2567: zero

2568: for

2569: all $r$.

2570: 

2571: For $r<1$ these intersections are isomorphic to the kernel of

2572:     the extension of $D_{A_{i,0},f}$ $(i=0,1)$ to $X(r)=(\Sigma\times

2573: [-\tfrac{1}{1-r},0])\cup X_1\cup X_2$, i.e. $X$ with a long collar

2574: inserted near $\Sigma$.  But the dimension of the kernel of

2575: $D_{A_{i,0},f}$ is independent of the choice of  Riemannian metric on

2576: $X$

2577: since $A_{i,0}$ is perturbed flat: it is isomorphic to the cohomology of the

2578: complex

2579: \eqref{eq2.6}, which is zero by the assumption that each perturbed flat

2580: connection is non-degenerate.

2581: 

2582: For $r=1$, we need to see that $(P^+_{0,0}\oplus T^1_{0,0})\cap

2583: \La^2_{0,0}$ and

2584: $(P^+_{1,0}\oplus T^1_{1,0})\cap \La^2_{1,0}$ are both zero. The

2585: quotient

2586: $$\frac{(P^+_{0,0}\oplus \ker S_{a_{0,0}}) \cap \La^2_{0,0}}{{

2587: {P^+_{0,0} \cap

2588: \La^2_{0,0}}}}\subset \ker S_{a_{0,0}} $$ is just $T^2_{0,0}$

2589: (\cite{APS,kirk-lesch})

2590: and the fact that each perturbed flat connection is non-degenerate

2591: implies

2592: that $T^1_{0,0}\cap T^2_{0,0}=0$. Moreover,  $ P^+_{0,0} \cap \La^2_{0,0}=0$ since 

2593: any section in this intersection is the restriction to the boundary of a solution to $D_{A_{0,0},f}\phi=0$ which extends to an $L^2$ solution on 

2594: $(\Sigma\times

2595: (-\infty,0])\cup

2596: X_1$.   The only such $\phi$ is zero, as we argued above. 

2597: Hence

2598: $(P^+_{0,0}\oplus T^1_{0,0})\cap \La^2_{0,0}=0$. The other endpoint is

2599: handled

2600: identically.

2601: 

2602: Since  the Maslov index is unchanged when the paths of Lagrangians are

2603: homotoped by a

2604: homotopy that preserves the dimension of the intersections at the

2605: endpoints

2606: (\cite{kirk-lesch}),

2607: $$\Mas( \La^1_{s,0},\La^2_{s,0})=\Mas(P^+_{s,0}\oplus

2608: T^1_{s,0},\La^2_{s,0}),$$ as

2609: desired.

2610: \end{proof}

2611: 

2612: The spaces $H^1_{s,t}=\ker S_{s,t}$ form a symplectic vector bundle over

2613: the square

2614: $[0,1]\times[0,1]$. In fact, this is just the pullback of the tangent

2615: bundle  $T_*\cM^*(\Sigma)$ via the map $[0,1]\times [0,1]\to

2616: \cM^*(\Sigma)$

2617: given by $(s,t)\mapsto [a_{s,t}]$.

2618: 

2619:     We have a Lagrangian subbundle

2620: $T^1_{s,0}$ defined over the top edge of this square, and also the 2

2621: sides

2622: since

2623: $A_{0,t}$ is independent of $t$ and $A_{1,t}$ is independent of $t$.

2624: Extend

2625: this arbitrarily (e.g.~ pull back via a retraction  of the square to

2626: its

2627: three sides) to a Lagrangian subbundle

2628: $T^1_{s,t}$ over the entire square. Then the 2-parameter family

2629: $P^+_{s,t}\oplus T^1_{s,t}$  provides a homotopy which shows that

2630: \begin{equation}\label{eq6.1}

2631: \Mas(P^+_{s,0}\oplus T^1_{s,0},\La^2_{s,0})=\Mas(P^+_{s,1}\oplus

2632: T^1_{s,1},\La^2_{s,1}).

2633: \end{equation} Together with Lemma \ref{lem6.1}, \eqref{eq6.1} implies

2634: that

2635: \begin{equation}\label{eq6.2} SF(D_{A_{s,0},f})= \Mas(P^+_{s,1}\oplus

2636: T^1_{s,1},\La^2_{s,1}).

2637: \end{equation} An argument very similar to the proof of Lemma

2638: \ref{lem6.1}, this time taking the adiabatic limit on $X_2$,  shows that

2639: \begin{equation}\label{eq6.3}

2640:      \Mas(P^+_{s,1}\oplus T^1_{s,1},\La^2_{s,1})=\Mas(P^+_{s,1}\oplus

2641: T^1_{s,1},P^-_{s,1}\oplus T^2_{s,1}),

2642: \end{equation} using the fact that the path of connections $A_{s,1}$ is

2643: perturbed flat on $X_2$.  Moreover, $T^2_{s,1}\subset H^1_{s,1}(\Si)$ is

2644: isomorphic to image of the differential of the inclusion

2645: $\cM_f^*(X_2)\to \cM^*(\Sigma)$ at $A_{s,1}$, i.e.~

2646: $T^2_{s,1}=\text{image}\big( T_{A_{s,1}}(\cM_f^*(X_2))\to

2647: T_{a_{s,1}}(\cM^*(\Sigma))\big)$.

2648:        Combining \eqref{eq6.2} and \eqref{eq6.3} and the facts that  the

2649: Maslov index is additive with respect to direct sum of symplectic spaces

2650: and that $P^+_{s,1}$ and $P^-_{s,1}$ are always orthogonal, one arrives

2651: at

2652: the conclusion

2653: \begin{equation}

2654: \label{eq6.5}

2655: \SF(D_{A_{s,0},f})=\Mas( T^1_{s,1},  T^2_{s,1}).

2656: \end{equation}

2657: 

2658: The spectral flow $\SF(D_{A_{s,0},f})$ depends only on $f$ and the

2659: endpoints

2660: $A_0$ and

2661: $A_1$ of $A_t$; see  Lemma

2662: \ref{independence}. Moreover its reduction modulo 2 (in fact modulo 8)

2663: only

2664: depends on the gauge equivalence classes $[A_0]$ and $[A_1]$.

2665: Equation \eqref{eq6.5} has identified $\SF(D_{A_{s,0},f})$

2666:   with a finite--dimensional Maslov index  corresponding to

2667: the tangent spaces of Lagrangian submanifolds of  a symplectic

2668: manifold. The

2669: following   lemma  identifies the mod 2 reduction of this Maslov index

2670: with

2671: the difference in intersection numbers of  $\cM^*(X_1)$ and

2672: $\cM^*(X_2)$ at

2673: $[A_0]$ and $[A_1]$.

2674: 

2675:     \begin{lemma}\label{lem6.2} Let $A$ and $B$ be  smooth  oriented

2676: Lagrangian submanifolds of  a (finite--dimensional) symplectic manifold

2677: $W$.   Let $p$  and $q$ be two transverse intersection points of

2678: $A$ and $B$. Let $\gamma_A$ be a path in $A$ from $p$ to $q$ and

2679: $\gamma_B$ a path in

2680: $B$ from $p$ to  $q$.  Suppose that  $F(s,t)$ is a homotopy rel

2681: endpoints

2682: in $W$ from

2683: $\gamma_A$ to  $\gamma_B$.  Let $T^A_{s,0}$ be the Lagrangian subbundle

2684: of

2685: the restriction of

2686: $T_*W$ to $\gamma_A$ consisting of tangent vectors to $A$, and similarly

2687: let

2688: $T^B_{s,1}$ be the Lagrangian subbundle of the restriction of

2689: $T_*W$ to $\gamma_B$ consisting of tangent vectors to $B$.

2690: 

2691: Then $T^A$ extends over the homotopy, and if $T^A_{s,t}$ is any

2692: extension

2693: (with

2694: $T^A_{0,s}$ and $T^A_{1,t}$ independent of $t$),  then

2695: $\Mas(T^A_{s,1},T^B_{s,1})$  is even if and  only if  intersection

2696: number

2697: of $A$  and $B$ at $p$ and $q$ coincide.

2698: \end{lemma}\qed

2699: 

2700: The straightforward proof can easily be constructed by musing on the

2701: following picture, and is left to the reader.

2702: \medskip

2703: \begin{center}

2704: \psfrag{TAs,t}{$T^A_{s,t}$}

2705: \psfrag{(s,t)}{$(s,t)$}

2706: \psfrag{A}{$A$}

2707: \psfrag{B}{$B$}

2708: \psfrag{p}{$p$}

2709: \psfrag{q}{$q$}

2710: \includegraphics{proof44.eps}

2711: \end{center}

2712: \medskip

2713: Lemma \ref{lem6.2} and Equation \eqref{eq6.5}  then imply Taubes'

2714: theorem.

2715: \begin{thm}[Taubes]$\Sign_C=\Sign_T$, and hence Casson's   and

2716: Taubes' invariants are equal up to an overall sign (which may depend

2717: on the

2718: homology 3-sphere).\end{thm}\qed

2719: 

2720:    Taubes proves more, namely he gives a construction  for specifying an

2721: overall sign

2722: so that  in fact  his invariant equals Casson's invariant for all

2723: homology 3-spheres.

2724: 

2725: \vfill\eject

2726: \begin{thebibliography}{99}

2727: 

2728: \bibitem{Atiyah-Bott} \textsc{Atiyah, M. F.} and\textsc{ Bott, R.} {\em

2729: The Yang-Mills equations over Riemann surfaces.}  Philos. Trans. Roy.

2730: Soc.

2731: London Ser. A  308  (1983),  no. 1505, 523--615.

2732: 

2733: \bibitem{APS}  \textsc{Atiyah, M. F.},  \textsc{Patodi, V. K.} and

2734: \textsc{

2735: Singer, I. M.} {\em Spectral asymmetry and Riemannian geometry. I.}

2736: Math.

2737: Proc. Cambridge Philos. Soc.  77  (1975), 43--69.

2738: 

2739: \bibitem{herald-boden}\textsc{Boden, Hans U.} and \textsc{Herald,

2740: Christopher M.} {\em The

2741: ${\rm SU}(3)$ Casson invariant for integral homology $3$-spheres.} J.

2742: Differential Geom.  50  (1998),  no. 1, 147--206.

2743: 

2744: \bibitem{BHK-int}\textsc{Boden, Hans U.},

2745: \textsc{Herald, Christopher M.},  and \textsc{ Kirk, Paul} {\em An

2746: integer

2747: valued ${\rm SU}(3)$ Casson invariant.} Math. Res. Lett. 8 (2001), no.

2748: 5-6, 589--603.

2749: 

2750: \bibitem{BHKK}\textsc{Boden, Hans U.},

2751: \textsc{Herald, Christopher M.},  \textsc{ Kirk, Paul}, and

2752: \textsc{Klassen, Eric P.} {\em   Gauge theoretic invariants of Dehn

2753: surgeries on knots.}  Geom. Topol.  5  (2001), 143--226

2754: 

2755: \bibitem{BHK-pqr}\textsc{Boden, Hans U.},

2756: \textsc{Herald, Christopher M.},  and \textsc{ Kirk, Paul} {\em The

2757: SU(3)

2758: Casson Invariant for Brieskorn spheres.} In preparation.

2759: 

2760: \bibitem{booss-ucp} \textsc{Booss--Bavnbek, B.}

2761: {\em The unique continuation property for Dirac operators revisited.}

2762: Geometry and topology: Aarhus (1998), Contemp. Math. 258 (2000), 21--32.

2763: 

2764: \bibitem{booss-lesch-phillips} \textsc{Booss--Bavnbek, B.}, \textsc{

2765: Lesch, Matthias},  and \textsc{Phillips, John}

2766:    {\em Unbounded Fredholm operators and

2767: spectral flow.} Preprint, math.FA/0108014

2768: 

2769: \bibitem{BMW} \textsc{Booss--Bavnbek, B.}, \textsc{Marcolli, M.},

2770: and \textsc{Wang, B.L.}

2771: {\em Weak UCP and perturbed monopole equations.}

2772: Preprint, math.DG/0203171

2773: 

2774: \bibitem{booss-w} \textsc{Booss-Bavnbek, Bernhelm}, and

2775: \textsc{Wojciechowski, Krzysztof P.} Elliptic boundary problems for

2776: Dirac

2777: operators. Mathematics: Theory \& Applications. Birkh\"{a}user Boston,

2778: Inc., Boston, MA, 1993. xviii+307 pp.

2779: 

2780: \bibitem{CLM} \textsc{Cappell, S.}, \textsc{Lee, R.}, and

2781: \textsc{Miller,

2782: E. Y.} {\em  A perturbative SU(3) Casson invariant.} Comment. Math.

2783: Helv. 77 (2002), no. 3, 491-523

2784: 

2785: \bibitem{daniel-kirk}\textsc{Daniel, A. M.} and \textsc{ Kirk, Paul}

2786: {\em

2787: A general splitting formula for the spectral flow.} With an appendix by

2788: \textsc{K. P. Wojciechowski}.  Michigan Math. J.  46  (1999),  no. 3,

2789: 589--617.

2790: 

2791: \bibitem{daska} \textsc{Daskalopoulos, Georgios D.}   {\em The topology

2792: of

2793: the space of stable bundles on a compact Riemann surface.}  J.

2794: Differential

2795: Geom.  36  (1992),  no. 3, 699--746.

2796: 

2797: \bibitem{daskalopoulos-uhlenbeck} \textsc{Daskalopoulos, Georgios D.}

2798: and

2799: \textsc{ Uhlenbeck, Karen K.} {\em  An application of transversality to

2800: the

2801: topology of the moduli space of stable bundles.}  Topology  34  (1995),

2802: no. 1, 203--215.

2803: 

2804: \bibitem{donaldson-kronheimer} \textsc{Donaldson, S. K.} and

2805: \textsc{Kronheimer, P. B.} The geometry of four-manifolds. Oxford

2806: Mathematical Monographs. Oxford Science Publications. The Clarendon

2807: Press,

2808: Oxford University Press, New York, 1990. x+440 pp.

2809: 

2810: \bibitem{dostoglou-salomon}\textsc{Dostoglou, Stamatis} and

2811: \textsc{Salamon, Dietmar A.}  {\em  Self-dual instantons and holomorphic

2812: curves.}  Ann. of Math. (2)  139  (1994),  no. 3, 581--640.

2813: 

2814: \bibitem{floer}\textsc{Floer, Andreas} {\em An instanton-invariant for

2815: $3$-manifolds.}  Comm. Math. Phys.  118  (1988),  no. 2, 215--240.

2816: 

2817: \bibitem{Herald1}\textsc{Herald, Christopher M.} {\em Legendrian

2818: cobordism

2819: and Chern-Simons theory on $3$-manifolds with boundary. } Comm. Anal.

2820: Geom.  2  (1994),  no. 3, 337--413

2821: 

2822: \bibitem{herald2}\textsc{Herald, Christopher M.} {\em  } in preparation.

2823: 

2824:      \bibitem{FU} \textsc{Freed, Daniel S.} and \textsc{Uhlenbeck, Karen

2825: K.}

2826: Instantons and four-manifolds. Second edition. Mathematical Sciences

2827: Research Institute Publications, 1. Springer-Verlag, New York, 1991.

2828: xxii+194 pp.

2829: 

2830: \bibitem{Grubb} \textsc{Grubb, Gerd} {\em Trace expansions for

2831: pseudodifferential boundary problems for Dirac--type operators and more

2832: general systems.} Ark. Math. 37 (1999), 45--86.

2833: 

2834: \bibitem{kirk-lesch}  \textsc{Kirk, Paul} and \textsc{Lesch, Matthias}

2835: {\em

2836: The eta-invariant, Maslov index, and spectral flow for Dirac-type

2837: operators

2838: on manifolds with boundary.} Preprint,  math.DG/0012123

2839: 

2840: \bibitem{lions-magenes} \textsc{Lions, J. } and \textsc{Magenes, E.}

2841: {\em Non--homogeneous boundary problems and applications.}

2842: Springer--Verlag, New York, 1972.

2843: 

2844: \bibitem{narasimhan-sheshadri} \textsc{Narasimhan, M. S.} and \textsc{

2845: Seshadri, C. S.} {\em  Stable and unitary vector bundles on a compact

2846: Riemann surface.}  Ann. of Math. (2)  82  1965 540--567.

2847: 

2848: \bibitem{nico} \textsc{Nicolaescu, Liviu I.} {\em  The Maslov index, the

2849: spectral flow, and decompositions of manifolds.}  Duke Math. J.  80

2850: (1995),  no. 2, 485--533.

2851: 

2852: \bibitem{rade} \textsc{Rade, Johan} {\em On the Yang-Mills heat

2853: equation in two and three dimensions.} J. Reine Angew. Math. 431

2854: (1992), 123--163.

2855: 

2856: \bibitem{RSW}  \textsc{Ramadas, T. R.},  \textsc{Singer, I. M.}, and

2857: \textsc{Weitsman, J.} {\em Some comments on Chern-Simons gauge theory.}

2858: Comm. Math. Phys.  126  (1989),  no. 2, 409--420.

2859: 

2860: \bibitem{Scott} \textsc{Scott, S. G.} {\em Determinants of Dirac

2861: boundary

2862: value problems over odd--dimensional manifolds.} Commun. Math. Phys. 173

2863: (1995), 43--76

2864: 

2865: \bibitem{taubes} \textsc{Taubes, Clifford Henry} {\em Casson's invariant

2866: and gauge theory.} J. Differential Geom. 31 (1990), no. 2, 547--599.

2867: 

2868: \bibitem{taylor} \textsc{Taylor, Michael E.} {\em Partial differential

2869: equations.} Applied Mathematical Sciences Vol 115, Springer--Verlag,

2870: New York, 1996.

2871: 

2872: \bibitem{yoshida} \textsc{Yoshida, Tomoyoshi} {\em Floer homology and

2873: splittings of manifolds.}  Ann. of Math. (2)  134  (1991),  no. 2,

2874: 277--323.

2875: 

2876: \end{thebibliography}

2877: 

2878: \end{document}

2879: 

2880: