1: \NeedsTeXFormat{LaTeX2e}

2: \documentclass[12pt,a4paper]{article}

3: 

4: 

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

6: %   Definitions for the article comin                   %

7: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

8: 

9: %%%%%%%%%%%%%%

10: %  Packages  %

11: %%%%%%%%%%%%%%

12: 

13: \usepackage{a4,enumerate}

14: \usepackage{amsthm}

15: \usepackage{amssymb}

16: \usepackage{amsmath}

17: \usepackage{latexsym}

18: \usepackage{pstricks,pst-coil,pst-plot,pst-node}

19: \usepackage{graphicx}

20: 

21: %%%%%%%%%%%%%%%%%%%%%%%%

22: % Arbeitseinstellungen %

23: %%%%%%%%%%%%%%%%%%%%%%%%

24: %\usepackage{showkeys}

25: %\long\def\regie#1{------{\bf #1}------}

26: %\long\def\regie#1{}

27: %\newcommand{\mnote}[1]{\marginpar{\tiny\em #1}}

28: %\def\mnote#1{}

29: \long\def\bild#1{#1}

30: %\long\def\bild#1{\ \newline{\bf !!!!!!Hier erscheint ein Bild !!!!}\newline}

31: \def\url#1{{\tt #1}}

32: 

33: %%%%%%%%%%%%%%%%%%%%%%%%%%

34: % Globale Formatierungen %

35: %%%%%%%%%%%%%%%%%%%%%%%%%%

36: 

37: \parindent0cm

38: \parskip=.5\baselineskip

39: 

40: 

41: %%%%%%%%%%%%%%%%%

42: %Seiten-Format  %

43: %%%%%%%%%%%%%%%%%

44: 

45: \oddsidemargin0.15cm

46: \evensidemargin0.73cm

47: \topmargin0.6cm

48: \headsep.9cm

49: \textwidth15cm

50: \textheight42\baselineskip

51: 

52: 

53: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

54: % Temporaer verwendete Groessen %

55: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

56: 

57: \newdimen\templaenge

58: 

59: %%%%%%%%%%%%%%%%

60: %    Fonts     %

61: %%%%%%%%%%%%%%%%

62: 

63: %%% Doppelbalkenfonts

64: \DeclareMathAlphabet{\doba}{U}{msb}{m}{n}

65: \gdef\mC{\doba{C}}

66: \gdef\mH{\doba{H}}

67: \gdef\mN{\doba{N}}

68: \gdef\mP{\doba{P}}

69: \gdef\mQ{\doba{Q}}

70: \gdef\mR{\doba{R}}

71: \gdef\mS{\doba{S}}

72: \gdef\mT{\doba{T}}

73: \gdef\mZ{\doba{Z}}

74: 

75: %%% Latex2e-Variante von \qed

76: \def\qed{{\leavevmode\unskip\nobreak\hfil\penalty 50\hskip 1em%

77:   \hbox{}\nobreak\hfil\lower 1pt\hbox{$\Box$\kern-.5pt}\parfillskip 0pt

78:   \finalhyphendemerits 0\par\bigbreak}}

79: \def\qedmath#1{\setbox0\hbox{$\displaystyle #1$}\templaenge=\textwidth\advance\templaenge by -\wd0%

80: \setbox1\hbox{$\Box$}\advance\templaenge by -2\wd1%

81: $$#1\hbox to0pt{\kern.5\templaenge$\Box$\kern-.5pt\hss}$$\par\bigbreak}

82: 

83: 

84: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

85: %    Buchstabenk"urzel                   %

86: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

87: 

88: %%% griechische Buchstaben

89: \def\al{{\alpha}}

90: \def\be{{\beta}}

91: \def\de{{\delta}}

92: \def\De{{\Delta}}

93: \def\om{{\omega}}

94: \def\Om{{\Omega}}

95: \def\la{{\lambda}}

96: \def\ka{{\kappa}}

97: \def\io{{\iota}}

98: \def\si{{\sigma}}

99: \def\Si{{\Sigma}}

100: \def\ga{{\gamma}}

101: \def\ep{{\varepsilon}}

102: \def\Ga{{\Gamma}}

103: \def\Tau{\mathcal{T}}

104: \def\th{{\vartheta}}

105: \def\Th{{\Theta}}

106: \def\ph{{\varphi}}

107: \def\phi{{\varphi}}

108: \def\Ph{{\Phi}}

109: \def\rh{{\rho}}

110: \def\ze{{\zeta}}

111: 

112: %%% beinahe griechische Buchstaben

113: \def\na{{\nabla}}

114: \def\pa{{\partial}}

115: \def\el{{\ell}}

116: 

117: %%% Kalligraphische Buchstaben

118: \def\cB{\mathcal{B}}

119: \def\cC{\mathcal{C}}

120: \def\cH{\mathcal{H}}

121: \def\cK{\mathcal{K}}

122: \def\cL{\mathcal{L}}

123: \def\cM{\mathcal{M}}

124: \def\cN{\mathcal{N}}

125: \def\cQ{\mathcal{Q}}

126: \def\cR{\mathcal{R}}

127: \def\cS{\mathcal{S}}

128: % \def\cT{\mathcal{T}} %Nicht benutzen: \Tau

129: \def\cW{\mathcal{W}}

130: \def\cV{\mathcal{V}}

131: \def\cZ{\mathcal{Z}}

132: 

133: %%% Zeichen und Symbole

134: \def\ohne{\smallsetminus}

135: \def\ti{\tilde}

136: \def\witi{\widetilde}

137: \def\wihat{\widehat}

138: \def\ol{\overline}

139: \def\gdw{\Longleftrightarrow}

140: \def\embed{\hookrightarrow}

141: \def\bs{\backslash}

142: \def\lan{\langle}

143: \def\ran{\rangle}

144: \def\LA{\Leftarrow}

145: \def\RA{\Rightarrow}

146: \def\LRA{\Leftrightarrow}

147: \def\trl{$\Leftrightarrow$}

148: \def\entsp{{\widehat =}}

149: \def\schlange{{\widetilde{\hspace{0.5cm}}}}

150: \def\Dflat{D^{\rm eucl}}

151: 

152: \def\torus{T^2}

153: \def\gflat{g}

154: %\def\Re{{\mathop{\rm Re\;}}}

155: %\def\Im{{\mathop{\rm Im\;}}}

156: \def\Ree{{\mathop{\rm Re\;}}}

157: \def\Imm{{\mathop{\rm Im\;}}}

158: 

159: 

160: %%%% \mR-Varianten

161: \def\Rplus{\mR^+}

162: \def\Roplus{\mR^+_0}

163: \def\Rminus{\mR^-}

164: \def\Rominus{\mR^-_0}

165: 

166: %Text-Makros

167: \def\ie{i.\thinspace e.\ \ignorespaces}

168: \def\Wlog{W.\thinspace l.\ \ignorespaces o.\thinspace g.\ \ignorespaces}

169: \def\nummerarray#1#2{\par\noindent\setbox0\hbox{\rm (#1)}\setbox1\hbox{$#2$}\unhcopy0%

170: \dimen0=.5\textwidth \advance\dimen0 by -\wd0 \advance\dimen0 by -.5\wd1 \kern\dimen0 \unhcopy1}

171: 

172: 

173: %%%%%%%%%%%%%%%%%%%%%%%%%%%%

174: %   Grosse Ausdr"ucke      %

175: %%%%%%%%%%%%%%%%%%%%%%%%%%%%

176: 

177: %%% Mathematical operators

178: \def\ker{\mathop{{\rm ker}}}

179: \def\image{\mathop{{\rm image}}}

180: \def\grad{{\mathop{{\rm grad}}}}

181: \def\Span{\mathop{{\rm span}}}

182: \def\spec{\mathop{{\rm spec}}}

183: \def\supp{\mathop{{\rm supp}}}

184: \def\dim{\mathop{{\rm dim}}}

185: \def\dist{\mathop{{\rm dist}}}

186: \def\vol{{\mathop{{\rm vol}}}}

187: \def\dvol{{\mathop{{\rm dvol}}}}

188: \def\area{{\mathop{{\rm area}}}}

189: \def\diam{{\mathop{{\rm diam}}}}

190: \def\injrad{{\mathop{{\rm injrad}}}}

191: \def\min{\mathop{{\rm min}}}

192: \def\max{\mathop{{\rm max}}}

193: \def\osc{\mathop{{\rm osc}}}

194: \def\mod{\mathop{{\rm mod}}}

195: \def\can{\mathop{{\rm can}}}

196: \def\Cl{\mathop{{\mC \rm l}}}

197: \def\ClR{\mathop{{\rm Cl}_{\mR}}}

198: \def\id{\mathop{{\rm id}}}

199: \def\Id{\mathop{{\rm Id}}}

200: \def\End{{\mathop{{\rm End}}}}

201: \def\GL{{\mathop{{\rm GL}}}}

202: \def\SO{\mathop{{\rm SO}}}

203: \def\U{{\rm U}}

204: \def\SU{{\rm SU}}

205: \def\Sp{{\rm Sp}}

206: \def\Spin{\mathop{{\rm Spin}}}

207: \def\CSO{{\mathop{{\rm CSO}}}}

208: \def\CSpin{{\mathop{{\rm CSpin}}}}

209: \def\PCSO{{P_{\rm CSO}}}

210: \def\PCSpin{{P_{\rm CSpin}}}

211: \def\PSpin{P_{\mbox{\rm \scriptsize Spin}}}

212: \let\psp\PSpin

213: \def\PSO{{P_{\mbox{\rm \scriptsize SO}}}}

214: \let\pso\PSO

215: \def\pson{{P_{\mbox{\rm \scriptsize SO(n)}}}}

216: \def\Hom{{\mathop{{\rm Hom}}}}

217: \def\scal{{\mathop{{\rm scal}}}}

218: \def\Ric{{\mathop{{\rm Ric}}}}

219: \def\im{{\mathop{{\rm im}}}}

220: \def\imD{{\im_{C^\infty} D_{g_0}}}

221: \def\imDg{{\im_{C^\infty} D_g}}

222: \def\Isom{{\mathop{{\rm Isom}}}}

223: 

224: %%% Mathematische Ausdr"ucke

225: \def\gnull{g}

226: \def\gf{g_f}

227: \def\Df{D_f}

228: \def\Dnull{D_1}

229: \def\Laf{\De_f}

230: \def\dvolf{\dvol_f}

231: \def\const{{{\mathop{\it const}}}}

232: 

233: \def\geucl{{g_{\mathop{\rm eucl}}}}

234: \def\gcan{{g_{\mathop{\rm can}}}}

235: \def\gstand{g_0}

236: \def\surfnc{{M_1}}

237: 

238: \def\platz{{\,\mathord{.}\,}}

239: \def\leersp{\lan\mathord{.}\mathord{,}\mathord{.}\ran}

240: \def\leerSP{(\mathord{.}\mathord{,}\mathord{.})}

241: %\def\CP{\mC{\mathrm P}}

242: %\def\RP{\mR{\mathrm P}}

243: \def\res#1#2{{#1}\lower .11ex\hbox{$|$}\lower .644ex\hbox{$\scriptstyle #2$}}

244: \def\stelle#1#2{\left. {#1}\right|_{#2}}

245: %\def\length#1#2{{\mbox{\rm length}_{#2}({#1})}}

246: %\def\Length#1#2{{\mbox{\rm length}_{#2}\Bigl({#1}\Bigr)}}

247: \let\mo=\mathopen

248: \let\mc=\mathclose

249: \def\GawtM#1#2{{{\ti\Ga}_{#1}^{#2}}}

250: \def\GaN#1#2{{{\Ga}_{#1}^{#2}}}

251: \def\nov{\mathbf{n}}

252: \def\zweiff{{\rm I\!I}}

253: \def\nob{{\mbox{\rmfamily\bfseries N}}}

254: \def\mkv{\mathbf{H}}

255: \def\tr{{\mathop{\rm tr}}}

256: \def\image{{\mathop{\rm image}}}

257: \def\lammin{{\lambda_{\rm min}^+}}

258: 

259: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

260: %  Environments und Strukturierelemente     %

261: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

262: 

263: \long\def\komment#1{}

264: %\long\def\longver#1{#1}

265: \long\def\longver#1{}

266: 

267: \def\proof#1{{\par\medbreak\noindent {\bf Proof\setbox0\hbox{#1}%

268: \ifdim\wd0=0pt .\else\ \ignorespaces #1.\fi}\enspace}}

269: \def\iop#1{{\par\medbreak\noindent {\bf Idea of proof\setbox0\hbox{#1}%

270: \ifdim\wd0=0pt .\else\ \ignorespaces #1.\fi}\enspace}}

271: %\def\example{{\noindent {\bf Example. }}}

272: \def\examples{{\noindent {\bf Examples. }\par\kern-\baselineskip}}

273: %\def\exampleno#1{{\noindent {\bf Example {#1}. }}}

274: 

275: 

276: \newtheoremstyle{remarks}{3pt}{3pt}{}{}{\bfseries}{}{ }{}

277: \newtheoremstyle{appendix}{3pt}{3pt}{}{}{\bfseries}{}{ }{}

278: 

279: \makeatletter

280: \@addtoreset{equation}{section}

281: \renewcommand{\theequation}{\the\c@section.\the\c@equation}

282: \makeatother

283: %\newtheorem{theorem}{\sc Theorem}[section]

284: \newtheorem{theorem}[equation]{\bf T{\footnotesize HEOREM}}

285: \newtheorem{proposition}[equation]{\bf P{\footnotesize ROPOSITION}}

286: \newtheorem{lemma}[equation]{\bf L{\footnotesize EMMA}}

287: \newtheorem{corollary}[equation]{\bf C{\footnotesize OROLLARY}}

288: \newtheorem{conjecture}[equation]{\bf C{\footnotesize ONJECTURE}}

289: \newtheorem*{cmcprinciple}{\bf Principle for construction of cmc-surfaces}

290: \newtheorem*{maintheorem}{\bf M{\footnotesize AIN} T{\footnotesize HEOREM}}

291: \newtheorem{openproblem}{\bf O{\footnotesize PEN} P{\footnotesize ROBLEM}}

292: \newtheorem{theorema}{\bf T{\footnotesize HEOREM}}[section]

293: \newtheorem{lemmaa}[theorema]{\bf L{\footnotesize EMMA}}

294: \newtheorem{propositiona}[theorema]{\bf P{\footnotesize ROPOSITION}}

295: \theoremstyle{definition}

296: \newtheorem*{remark}{Remark}

297: \newtheorem{definitionnr}[equation]{Definition}

298: \newtheorem*{definition}{Definition}

299: \newtheorem*{example}{Example}

300: 

301: \theoremstyle{remarks}

302: %\newtheorem*{remarks}{Remarks}

303: 

304: \hyphenation{know-ledge Ha-bi-li-ta-tions-schrift}

305: 

306: \def\eref#1{{\rm (\ref{#1})}}

307: 

308: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

309: % Extra-Definitionen fuer comin.tex  %

310: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

311: 

312: \def\pD{{p_D}}

313: \def\pY{{p_Y}}

314: \def\qD{{q_D}}

315: \def\Dir{D}

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

317: \def\cF{{\cal F}}

318: \let\<\langle

319: \let\>\rangle

320: \let\La\Lambda

321: \def\inj{\mathop{\rm inj}}

322: \def\Conf{{\rm Conf}}

323: \def\extconf{\ol{[g_0]}}

324: \def\extco#1{\ol{{\cM}(#1)}}

325: \def\RHS{{\rm RHS}}

326: 

327: 

328: 

329: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

330: %                                                                      %

331: % T h e   D o c u m e n t    b e g i n s    h e r e !                  %

332: %                                                                      %

333: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

334: 

335: 

336: \begin{document}

337: 

338: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

339: \title{The smallest Dirac eigenvalue in a spin-conformal class

340: and cmc-immersions}

341: \author{Bernd Ammann%

342: %\thanks{Partially supported

343: %by The European Contract Human Potential Programme,

344: %Research Training Networks HPRN-CT-2000-00101 and HPRN-CT-1999-00118}

345: }

346: \date{September 2005}

347: %\date{\today}

348: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

349: \maketitle

350: 

351: %Kommentar

352: \begin{abstract}

353: Let us fix a conformal class $[g_0]$ and a spin structure $\si$ on

354: a compact manifold $M$. For any $g\in [g_0]$, let $\la^+_1(g)$

355: be the smallest positive eigenvalue of the Dirac operator $D$ on $(M,g,\si)$.

356: In a previous article we have shown that

357:   $$\lammin(M,g_0,\si):=\inf_{g\in [g_0]} \la_1^+(g)\vol(M,g)^{1/n}>0.$$

358: In the present article, we enlarge the conformal class by adding 

359: certain singular metrics. We will show that

360: if $\lammin(M,g_0,\si)<\lammin(S^n)$,

361: then the infimum is attained on the enlarged conformal class.

362: For proving this, we solve a system of

363: semi-linear partial differential equations involving a nonlinearity

364: with critical exponent:

365:   $$D\phi= \la |\phi|^{2/(n-1)}\phi.$$

366: The solution of this problem has many analogies

367: to the solution of the Yamabe problem.

368: However, our reasoning is more involved than in the Yamabe problem

369: as the eigenvalues of the Dirac operator tend to $+\infty$ and $-\infty$.

370: 

371: Using the spinorial Weierstra\ss{} representation, 

372: the solution of this equation in dimension 2 shows the existence 

373: of many periodic constant mean curvature surfaces.

374: 

375: %{\bf Preliminary version, please do not distribute!}

376: \end{abstract}

377: {\bf Keywords:} Dirac operator, eigenvalues, conformal geometry,

378: critical Sobolev exponents

379: 

380: {\bf Mathematics Classification:} 58J50, 53C27 (Primary),

381: 58C40, 35P15, 35P30, 35B33 (Secondary)

382: 

383: 

384: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

385: \section{Introduction}

386: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

387: 

388: %The problem that we want to study in this article is originally motivated

389: %from the spectral theory of Dirac operators. Because of the conformally

390: %invariant character of the functional and the associated Euler-Lagrange 

391: %equations, blowup phenomena similar to those in the Yamabe problem arise.

392: %If the underlying manifold has dimension $2$, 

393: %the solutions to the Euler-Lagrange equations 

394: %can locally be translated into conformal constant mean curvature immersions

395: %into $\mR^3$. This translation can be used to construct periodic

396: %constant mean curvature surfaces.

397: 

398: %In this introduction, we will give a brief  

399: %overview over the history, explain the relations to the 

400: %Yamabe problem, and sketch the implied techniques; then we  

401: %will describe the application to constant mean curvature surfaces.

402: 

403: Let $M$ be a compact $n$-dimensional 

404: manifold, $n\geq 2$, with a fixed conformal class $[g_0]$ and a fixed 

405: spin structure $\si$.

406: Let $g$ be conformal to $g_0$, i.e.\ $ g\in [g_0]$. 

407: The classical Dirac operator $D_g$

408: on $(M,g,\si)$ has discrete real spectrum

409: with finite multiplicities. The eigenvalues tend to $+\infty$ and $-\infty$.

410: The dimension of the kernel of $D_g$ is a spin-conformal invariant, i.e.

411: $\dim \ker D_g=\dim \ker D_{g_0}$ for all $g\in [g_0]$.

412: We denote the first (=smallest) positive eigenvalue of $D_g$ 

413: by $\la_1^+(g)$,

414: and the first (=largest) negative eigenvalue by $\la_1^-(g)$.

415: 

416: Finding bounds for this eigenvalue has attracted much interest

417: during the last decades. Among several estimates 

418: \cite{lichnerowicz:63,friedrich:80,kirchberg:86,kirchberg:88,

419: kramer.semmelmann.weingart:98,kramer.semmelmann.weingart:99} 

420: in terms of a positive scalar curvature bound, 

421: let us mention the following estimate due to Friedrich~\cite{friedrich:80}.

422: If the minimum of the scalar curvature

423: of $(M,g)$ is at least $s>0$, then 

424: $\left(\la_1^\pm(g)\right)^2\geq {n\over 4(n-1)}s$.

425:   

426: An improvement of this inequality that is important in conformal 

427: geometry was derived by Hijazi \cite{hijazi:86} 

428: in terms of the first eigenvalue

429: $\la_1(L_g)$ 

430: of the conformal Laplacian $L_g=4\,{n-1\over n-2}+\scal_g$, 

431: namely

432: \begin{equation}\label{ineq.hij}

433:   \left(\la_1^\pm(g)\right)\geq {n\over 4(n-1)} \la_1(L_g)

434: \end{equation}

435: if $n\geq 3$. On the other hand, if the 

436: \emph{Yamabe invariant}

437:  $$Y(M,[g_0])=\inf_{g\in [g_0]}{\int_M \scal_g\dvol_g\over

438: \vol(M,g)^{(n-2)/n}}\in (-\infty, n(n-1)]$$ 

439: is non-negative, one can use the  conformal

440: transformation formula

441: for the scalar curvature \cite[Theorem~1.159]{besse:87}, and one obtains

442: that 

443:   $$ \la_1(L_g)\vol(M,g)^{2/n}\geq Y(M,[g_0]).$$

444: Together with \eref{ineq.hij}, we get

445: \begin{equation}\label{ineq.hij.der}

446: \left|\la_1^\pm(g)\right|\vol(M,g)^{1/n} \geq 

447:     \sqrt{{n\over 4(n-1)}\, Y(M,[g_0])}.

448: \end{equation}

449: That this inequality even holds in 

450: dimension $n=2$, was proved by C.~B\"ar \cite{baer:92b}. In this special case,

451: the Gauss-Bonnet theorem tells us that $Y(M)= 4\pi \chi(M)$, hence we obtain

452: \begin{equation}\label{ineq.baer}

453:  \left|\la_1^\pm(g)\right|\area(M,g)^{1/2}\geq 2\sqrt{\pi}

454: \end{equation}

455: if $M$ is diffeomorphic to $S^2$, but we do not get a bound for other surfaces.

456: Equality in \eref{ineq.hij.der} and \eref{ineq.baer} hold for the round 

457: spheres.

458: 

459: 

460: Hence, we have obtained an explicit positive 

461: lower bound $\left|\la_1^\pm(g)\right|\vol(M,g)^{1/n}$ that is uniform

462: on the conformal class $[g_0]$.

463: From now on, we will restrict to the first \emph{positive} eigenvalue, 

464: in order to simplify the presentation (see the remark below).

465: 

466: 

467: The \emph{existence} of a positive lower 

468: bound for $\la_1^+(g)\vol(M,g)^{1/n}$

469: had already been derived before by J. Lott. His estimate

470: does not require that $Y(M,[g_0])$ is positive, but needs the weaker assumption

471: that the Dirac operator is invertible, i.e.\ has $0$-dimensional kernel.

472: Unfortunately, his bound is not explicit, and for most spin-conformal

473: manifold the determination of the 

474: value of the associated infimum 

475: \begin{equation}\label{def.lammin}

476: \lammin(M,[g_0],\si)

477:     :=\inf_{g\in [g_0]} \la_1^+(g)\vol(M,g)^{1/n}

478: \end{equation}

479: is still a challenging open problem. 

480: 

481: A key idea of J. Lott's article is to derive a lower 

482: bound for the first Dirac eigenvalue in terms of the  

483: supremum of a conformally invariant functional, a version of this 

484: functional will be explained in Section~\ref{sec.varprin}.  

485: 

486: In our article \cite{ammann:03}, we started to study this

487: functional in more detail. In particular, we showed that Lott's result

488: extends to the case that the Dirac operator has non-trivial kernel,

489: namely 

490:   $$\lammin(M,[g_0],\si)>0$$

491: for any compact Riemannian spin manifold $(M,[g_0],\si)$.

492: 

493: Furthermore, it was  shown that

494:   $$\lammin(M,[g_0],\si) \leq \lammin(\mS^n),$$

495: where $\mS^n=(S^n,g_{\rm can})$ denotes the sphere with its standard 

496: Riemannian metric of constant sectional curvature $1$.

497: This bound has been proven in \cite{ammann:03} unless 

498: $\ker D\not=\{0\}$ and $n=2$. The remaining case $\ker D\not=\{0\}$ and $n=2$ 

499: was given in \cite{grosjean.humbert:p05}.

500: 

501: In the present article, we discuss whether the associated infimum in 

502: \eref{def.lammin} is attained. For having a well-behaved

503: minimization problem, it is reasonable to replace the conformal class

504: $[g_0]$ in 

505: \eref{def.lammin} by its $L^\infty$-completion $\overline{[g_0]}$.

506: Here, by definition a metric $f^2g_0$ is in $\overline{[g_0]}$ if 

507: $f$ is a real-valued $L^\infty$ function. The first positive 

508: eigenvalue of the Dirac operator extends naturally to this completion

509: (see Section~\ref{sec.spd}).

510:  We show that if we have the strict inequality

511:  \begin{equation}\label{eq.strict}

512:   \lammin(M,g_0,\si)< \lammin(\mS^n)= {n\over 2}\,\om_n^{1/n},

513: \end{equation}

514: then the infimum is obtained by a generalized metric $g\in \overline{[g_0]}$.

515: This minimizer has the form $g:=|\phi|^{4/(n-1)}g_0$ where 

516: $\phi$ is a spinor of regularity $C^2$. The set 

517: $\phi^{-1}(0):=\{x\in M\,|\,\phi(x)=0\}$ is  

518: called the \emph{nodal set of $\phi$} or the 

519: \emph{set of degeneration of $g$}.

520: 

521: \begin{theorem}\label{theo.main}

522: Let $M$ be a compact manifold of dimension $n\geq 2$

523: with a fixed conformal

524: class $[g_0]$ and a spin structure $\si$.

525: Assume that \eref{eq.strict}

526: holds. Let $\al:= 2/(n-1)$ if $n\geq 4$, and let $\al\in (0,1)$ if $n\in\{2,3\}$.  

527: \begin{enumerate}[\rm (A)]

528: \item Then there is a spinor field

529:    $\phi\in C^{2,\al}(\Si M)\cap C^\infty(\Si(M\ohne \phi^{-1}(0)))$ 

530:    on $(M,g_0)$ such that

531: \begin{equation}\label{eq.dirac.nonlin}

532:   D_{g_0}\phi = \lammin\, |\phi|^{2/(n-1)}\phi,\qquad \|\phi\|_{2n/(n-1)}=1.

533: \end{equation}

534: \item There is a $g\in \extconf$ such that

535:   $$\la^+_1(g)\vol(M,g)^{1/n}=\lammin.$$

536: The metric has the form $g=|\phi|^{4/(n-1)}g_0$ where 

537: $\phi$ is a spinor as in (A).

538: \item If $\dim M=2$, then the metric $g$ is smooth 

539: and the set of degeneration of $g$, denoted $\cS_g$,

540: is finite.

541: Furthermore

542:   $$\#\cS_g< {\rm genus}(M).$$

543: In particular,  if $M$ is diffeomorphic to a

544: $2$-torus, then the set of degeneration $\cS_g$ is empty.

545: \end{enumerate}

546: \end{theorem}

547: 

548: Inequality \eref{eq.strict} has been proven for several 

549: classes of manifolds.

550: It is known that it holds for non-conformally-flat manifolds of dimension

551: $\geq 7$ \cite{ammann.humbert.morel:p03av2}. 

552: 

553: Inequality~\eref{eq.strict} has also been shown if 

554: $M$ is conformally flat, if $D$ is invertible, 

555: and if the mass endomorphism is  not identically zero 

556: (after a possible change of orientation if $\dim M \equiv 3 \mod 4$) 

557: \cite{ammann.humbert.morel:p03b}. 

558: The mass endomorphism is a section of $\End(\Si M)$

559: defined as the zero order term of the development of the Green function

560: for the Dirac operator at the diagonal with respect to a 

561: conformal coordinate map.

562: 

563: Furthermore, \eref{eq.strict} is known for many Riemann surfaces (i.e. $n=2$), 

564: e.g.\ all rectangular tori have a spin structure such that 

565: \eref{eq.strict} holds.

566: 

567: The Euler-Lagrange equation~\eref{eq.dirac.nonlin} of the above minimization

568: problem has a particularly nice  interpretation in dimension~$n=2$. 

569: Locally the equation~\eref{eq.dirac.nonlin} 

570: can be translated into a conformal constant mean curvature

571: immersion into $\mR^3$. 

572: This translation

573: is a spinorial extension of the Weierstrass representation (see

574: Section~\ref{sec.weierrep}). By pasting together these local surfaces

575: one obtains a ``periodic branched conformal cmc immersion''. 

576: 

577: More exactly, let $(M,g)$ be a compact Riemann surface together 

578: with its universal covering $\pi:\witi M\to M$. A 

579: \emph{periodic branched conformal cmc immersion based on $(M,g)$}

580: is by definition a smooth map $F:\witi M\to \mR^3$ together with 

581: finitely many points $p_1,\ldots,p_k\in M$, the so-called 

582: \emph{branching points}, such 

583: that the following properties hold:

584: \begin{enumerate}[{\rm (1)}]

585: \item {\it Periodicity:}  There is a homomorphism 

586: $h:\pi(M)\to \mR^3$, the \emph{periodicity map}, such that 

587: for any $\gamma\in \pi_1(M)$, and $x\in\witi M$ one has

588:   $$F(x\cdot \gamma)=F(x)+h(\gamma).$$

589: Here $\cdot$ denotes the action of $\pi_1$ on $\witi M$ via Deck transformation.

590: \item {\it Conformality:} The restriction of $F$ to $\witi M\setminus \pi^{-1}(\{p_1,\ldots,p_k\})$

591:       is a conformal immersion.

592: \item {\it Branching points:} We have 

593: $dF_q=0$ for any $q\in \pi^{-1}(\{p_1,\ldots,p_k\})$. 

594: The order of the first non-vanishing term in the 

595: Taylor development of $dF$ in $q$ is called the 

596: \emph{branching index} of $F$ at $q$. 

597: \item {\it CMC:} The image $F\left(\witi M\setminus \pi^{-1}(\{p_1,\ldots,p_k\}\right)$ 

598: is an immersed surface 

599: with constant mean curvature.

600: \end{enumerate}

601: 

602: The principle yields the existence of many periodic branched conformal

603: cmc immersions. The set of periodic branched conformal

604: cmc immersions, $H\neq 0$, is 

605: essentially (see Section~\ref{sec.app.cmc} for a precise 

606: statement) 

607: in bijection with the stationary points of the variational problem

608: associated to our minimization problem. 

609: In particular, all minimizers of \eref{def.lammin} give rise

610: to periodic branched conformal

611: cmc immersions. We obtain

612: 

613: \begin{cmcprinciple}

614: Assume that the Riemann spin surface $(M,g,\si)$ carries a metric

615: $g$ such that the first positive eigenvalue of the Dirac operator is smaller

616: than  $2\sqrt{\pi/\area(M,g)}$. Then there is a periodic branched

617: conformal cmc immersion $F$ based on $(M,g)$. 

618: The regular homotopy class of $F$ is determined by the spin-structure $\si$.

619: The indices of all branching points are even, 

620: and the sum of these indices

621: is smaller than $2 {\rm genus}(M)$. In particular, 

622: if $M$ is a torus, there are no branching points.  

623: \end{cmcprinciple}

624: 

625: Some examples of branched

626: conformal cmc immersions that may arise by this principle

627: are given in Section~\ref{sec.app.cmc}. 

628: 

629: The problem that we are discussing in this article has many relations to the 

630: Yamabe problem. For a given compact conformal manifold $(M,[g_0])$ 

631: the Yamabe problem is the problem to find a metric of constant scalar curvature

632: $g$ in $[g_0]$. The problem has been affirmatively solved by Trudinger, 

633: Aubin, Schoen and Yau, see \cite{lee.parker:87} for a good overview. 

634: At first, due to the 

635: conformally invariant character of our problem and the Yamabe problem,

636: certain bounded, but non-compact Sobolev embeddings play an important role.

637: In the solution of both problems it is useful to break the conformal 

638: invariance by perturbing a parameter. For this perturbed parameter

639: the embeddings are compact, and standard methods yield the existence 

640: of a minimizer. It then has to be checked whether the perturbed minimizers

641: converge to a minimizer of the unperturbed problem, or whether they concentrate

642: in some points. In the Yamabe problem

643: the minimizers converge if 

644: \begin{equation}\label{eq.yamab}

645: Y(M,[g_0])<Y(\mS^n),

646: \end{equation}

647: holds whereas in our problem, the minimizers converge if 

648: \eref{eq.strict} holds.

649: Secondly, \eref{ineq.hij.der} implies that any manifold satisfying

650: \eref{eq.strict} satisfies \eref{eq.yamab} as well. Hence, proving 

651: \eref{eq.strict} for a given spin-conformal manifold $(M,[g_0],\si)$ 

652: solves the Yamabe problem on this manifold.

653: The third relationship is that some proofs of inequality \eref{eq.strict}

654: in special cases resemble to proofs of inequality, 

655: e.g.\ \cite{ammann.humbert.morel:p03av2} resembles to \cite{aubin:76}.

656: 

657: The structure of the article is as follows.

658: In Section~\ref{sec.varprin} we reformulate our problem 

659: as a variational problem. We will see that it is natural to admit 

660: in the infimum \eref{def.lammin} certain singular metrics namely ``metrics''

661: conformal to $g_0$ whose conformal factor might vanish 

662: somewhere. Such metrics --- called \emph{generalized metrics} ---

663: are the subject of Section~\ref{sec.spd}.  In the following section, 

664: we discuss the round sphere. This example is helpful to obtain a deeper 

665: understanding for the analytical difficulties. However, it can be skipped

666: if the reader is only interested in the main results of the article.

667: Section~\ref{sec.reg.theo} is devoted to certain regularity issues that will 

668: become important in Section~\ref{sec.solution}, where 

669: the variational principle is finally solved. We then show in 

670: Section~\ref{sec.proofmain} how this implies the main theorem.

671: The singularities of the minimizers are discussed in Section~\ref{sec.degen}.

672: In Section~\ref{sec.weierrep} we recall the spinorial 

673: Weierstrass representation. Section~\ref{sec.app.cmc} uses the spinorial 

674: Weierstrass representation to derive the application to constant mean 

675: curvature surfaces. Several examples are included.

676: In Appendix~\ref{sec.regularity} we summarize (without proofs)

677: some analytical tools. 

678: This appendix shall also serve as a reference for fixing the notations for 

679: Sobolev spaces and H\"older spaces. In Appendix~\ref{sec.scha.power} 

680: we proof a proposition about H\"older spaces, as we could not find a 

681: proof in the literature.

682: 

683: 

684: \begin{remark}[The first negative eigenvalue]

685: The infimum is also attained in \eref{def.lammin}

686: if we replace $\la_1^+$ by $|\la_1^-|$.  

687: In the case $n\not\equiv 3\mod 4$ this is obvious: 

688: there exists an automorphism

689: of the spinor bundle anticommuting with $D$, 

690: hence $\la_1^-(g)=-\la_1^+(g)$. 

691: In the case $n\equiv 3\mod 4$, the proof for $\la_1^-$

692: is up to the obvious sign changes completely identical.

693: In almost all references cited in the introduction, 

694: all statements for   $\la_1^+$ also hold for $|\la_1^-|$ and vice versa.

695: The only exception is \cite{ammann.humbert.morel:p03b}.

696: \end{remark}

697: 

698: {\bf Acknowledgements.}

699: I am very much indebted to Christian B\"ar for various support.

700: Thank you also to Emmanuel Humbert for his continuing interest in conformal 

701: spin geometry and his deep analytical ideas. Many electronic and personal 

702: discussions with Robert Kusner and Karsten Grosse-Brauckmann were also very helpful. I also want to thank B.~Booss-Bavnbek, Oussama Hijazi,

703: Sergiu Moroianu,Victor Nistor, Reiner Sch\"atzle and Guofang Wang

704: for several stimulating discussions related to this article.

705: %, and thanks to

706: %B.~Booss-Bavnbek for pointing us to \cite{booss.marcolli.wang:p02}.

707: 

708: 

709: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

710: \section{The associated variational principle}\label{sec.varprin}

711: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

712: 

713: The goal of this section is to reformulate the problem of minimizing

714: the first Dirac eigenvalue in a conformal class as a variational problem.

715: The choice of a good functional is not very easy, as 

716: we would like to find a functional which is both bounded and 

717: conformally invariant. 

718: %This is more involved than in the solution of the Yamabe problem.

719: The Dirac operator $D$ has a simple behavior under

720: conformal change of the metric, its square $D^2$ transforms in a more 

721: complicated way. Hence, it is desirable to use a functional that 

722: contains only terms in $D\phi$ and $\phi$ and no term depending

723: on $D^2\phi$. The $\cF_q$-functional defined in \eref{def.F}, $q=2n/(n+1)$,

724: is such a conformally invariant, bounded functional, and it will 

725: be turn out that by working with this functional we obtain

726: the desired results.

727: 

728: \begin{remark} In case, that the reader of our article

729: is already familiar with the analytical problems of the Yamabe problem,

730: he might find it enlightening to 

731: compare this problem to our problem.

732: Many techniques from the resolution of the Yamabe problem

733: can be carried over to our setting. However, several arguments from

734: the resolution of the Yamabe problem fail. Two main problems arise:

735: on the one hand the spectrum of $D$ is neither bounded from 

736: below nor from above, on the other hand, we are working with sections

737: of a vector bundle instead of functions, hence the standard maximum principle

738: is not available.

739: Such arguments will have to be replaced by other approaches.

740: \end{remark}

741: 

742: %In contrast to the Yamabe operator which is bounded from below, 

743: %the Dirac operator is neither bounded from above nor from below. 

744: %This is why spectral geometers typically work 

745: %with the square $\Dir^2$ of the Dirac operator instead of the Dirac operator 

746: %itself. 

747: %In our problem, working with $\Dir^2$ 

748: %is not a good approach, as the conformal 

749: %transformation formula for $\Dir^2$ is more complicated than for $\Dir$. 

750: %It is difficult to construct a conformally invariant 

751: %functional in the style of the Yamabe functional

752: %which contains $\Dir^2$, but not $\Dir$. 

753: %A functional that suits to our needs is $\cF_q$ defined in \eref{def.F}.

754: 

755: %We will see that the definition of the first positive Dirac eigenvalue

756: %extends naturally to a certain completion of the conformal class. This 

757: %completion will provide the natural set in which one should search 

758: %for minimizers of $\la^+_1(g)$.

759: 

760: At first, we recall some basic definitions and facts from spin geometry.

761: For details we refer to textbooks as for example 

762: \cite{lawson.michelsohn:89, roe:88,friedrich:00} 

763: or to the beautifully written 

764: self-contained introduction \cite{hijazi:01}.

765: 

766: Let $M$ be a compact manifold equipped with a Riemannian metric

767: $g_0$ and a spin structure $\si$. Assume that $g$

768: is a metric conformal to $g_0$. One associates to $(M,g,\si)$ 

769: a natural complex vector bundle over $M$ called the spinor bundle

770: $\Si(M,g,\si)\to M$. 

771: Sections of this bundle are called \emph{spinor fields}

772: or simply \emph{spinors}. The bundle carries a hermitian metric, a metric connection,

773: and a Clifford multiplication. These additional structures are used to define

774: the Dirac operator $D_g:\Gamma(\Si(M,g,\si)\to\Gamma(\Si(M,g,\si)$, which

775: is a first order elliptic differential operator. 

776: The Dirac operator $D_g$ 

777: is essentially self-adjoint, and hence it has a self-adjoint

778: extension. As a consequence of standard elliptic theory,

779: the spectrum is real and discrete, and all multiplicities

780: are finite.

781: 

782: In the following, the spin structure $\si$ will be fixed (and often suppressed

783: in the notation), whereas the metric $g$ varies inside the conformal class

784: $[g_0]$. Some objects 

785: will be noted with an index $g$, which means that they are

786: defined with respect to~$g$ whereas the same object without the index

787: $g$ indicates that it is defined with respect to the fixed background metric 

788: $g_0$ (with some exceptions that are apparent from the context). 

789: For example, $\dvol_g$ is the volume element associated to $g$ 

790: and $\dvol=\dvol_{g_0}$ is the one associated to $g_0$.

791: We will frequently use several norms, as for example

792: $\|\phi\|_{L^p}$ and $\|\phi\|_{H_1^q}$ 

793: that are defined with respect to $g_0$ 

794: unless otherwise stated. We summarize their definition and some important 

795: analytical tools in Appendix~\ref{sec.regularity}.

796: 

797: The spectrum of the Dirac operator $D_g$ will be denoted as 

798:     $$\ldots \leq \la_2^-(g)\leq \la_1^-(g)<0=\ldots=0

799:       <\la_1^+(g)\leq\la_2^+(g)\leq \ldots,$$ 

800: where each eigenvalue appears with its multiplicity. Note that $0$ 

801: may be an eigenvalue or not. In the former case $D_g$ is not invertible,

802: otherwise it is. Elliptic theory shows

803: $\lim_{k\to\infty}\la_k^+(g) =-\lim_{k\to\infty}\la_{-k}^+(g)=\infty$. 

804:  

805: The following transformation formula  will be of central importance.

806: To our knowledge, the formula was found by Hitchin \cite{hitchin:74}. 

807: Another reference written up in a more self-contained manner

808: is \cite{hijazi:01}.

809: %are \cite{hijazi:86,hijazi:01}.

810: 

811: \begin{proposition}[Conformal transformation formula for $D$]\label{prop.conf.change}

812: Let $g=f^2g_0$, $f:M\to \mR$ smooth and positive.

813: There is an isomorphism of vector bundles

814: ${F:\Si(M,g_0,\si)\to \Si(M,g,\si)}$

815: which is a fiberwise isometry such that

816:   $$\Dir_g(F(\phi))=F\left(f^{-{n+1\over 2}}\Dir_{g_0} f^{{n-1\over 2}}\phi\right).$$

817: \end{proposition}

818: 

819: It is convenient to define 

820:   $$\witi{F}(\phi)=F(f^{-{n-1\over 2}}\phi).$$

821: We will use this isomorphism $\witi{F}$

822: to identify spinors associated 

823: to conformal metrics. With this identification the conformal transformation 

824: formula reads as

825:   $$\Dir_g(\phi)= f^{-1}\Dir_{g_0}(\phi),$$

826: and $|\phi|_g=f^{-{n-1\over 2}}|\phi|_{g_0}$.

827: In particular, with this identification the kernels of $D_g$ and 

828: $\Dir_{g_0}$ coincide.

829: It is easy to verify that with this identification,

830: the expression $\int \<\Dir_g \phi,\phi\>_g\dvol_g$ is conformally invariant. 

831: The $L^p$-norm $\|\phi\|_{L^p(g)}:=\left(\int |\phi|_g^p\dvol_g\right)^{1/p}$ 

832: is conformally invariant 

833: if and only if $p$ has the value $\pD:=2n/(n-1)$. 

834: Similarly, $\|\Dir_g\phi\|_{L^q(g)}$ is conformally invariant if and only if

835: $q$ has the value $\qD:=2n/(n+1)$. Note that $\qD^{-1}+\pD^{-1}=1$. 

836: 

837: For any $q\in [\qD,2]$ and for any $H_1^q$-spinor $\phi$ 

838: that is not in the kernel of the 

839: Dirac operator we define

840: \begin{equation}\label{def.F}

841:   \cF_q^g(\phi)=  {\int\<\Dir_g\phi ,\phi\>_g\dvol_g\over \|\Dir_g\phi\|_{L^q(g)}^2}, \qquad

842:      \mu_q^g :=\mu_q(M,g,\si):= \sup \cF_q^g(\psi),

843: \end{equation}

844: and $\cF_q:= \cF_q^{g_0}$, $\mu_q:=\mu_q^{g_0}$.

845: The well-definedness of $\cF_q^g$ and some basic properties are given by the 

846: following lemma.

847: 

848: \begin{lemma}\label{lem.f.elem}

849: Let $q \in [\qD,\infty)$. Let $\phi$ be a spinor field

850: of regularity $H_1^q$, $D\phi\neq 0$.

851: Then,

852: $\cF_q(\phi)$ is well-defined and real. Furthermore

853: $\cF_q:H_1^q\setminus \ker D \to\mR$ 

854: is Fr\'echet differentiable with derivation given by 

855: \begin{equation}\label{eq.abl}

856: d\cF_q(\phi)(\psi) = {2\over \|\Dir\phi\|_{L^q}^2}

857:        \,\int\< \phi  -\rho_{q,\phi} |\Dir\phi|^{q-2}\Dir\phi,\Dir\psi\>,

858: \end{equation}

859: where $\rho_{q,\phi}=\cF_q(\phi) \|\Dir\phi\|_{L^q}^{2-q}$.

860: The supremum $\mu_q$ is positive and finite.

861: \end{lemma}

862: 

863: From the above considerations it is evident that the 

864: functional $\cF_q$ is conformally invariant if and only if $q=\qD$.

865: %This functional will take the role which is played by the Yamabe functional

866: %in the Yamabe problem.

867: 

868: \proof{}

869: Let $\phi$ be a spinor field of regularity $H_1^q$, $D\phi\neq 0$. 

870: Take $p$ with $p^{-1}+q^{-1}=1$, $q\ge {2n\over n+1}$.

871: Because of $\phi\in H_1^q\embed L^p$, we see with H\"older's inequality

872: that $\<D\phi,\phi\>$ is integrable. Thus, the numerator of $\cF_q$ is 

873: well-defined, and hence $\cF_q$ is well-defined. The self-adjointness

874: of $\Dir$ implies that $\cF_q(\phi)$ is real.

875: Moreover, because $H_1^q\embed L^p$ is bounded we see that 

876: $|\int_M \<D\phi,\phi\>|\leq \|D\phi\|_{L^q}\,\|\phi\|_{L^p}$

877: is bounded from above by a multiple of $\|\phi\|_{H_1^q}^2$. 

878: Using Theorem~\ref{theo.global.lp} we obtain

879: $\|\phi\|_{H_1^q}\leq \|D\phi\|_{L^q}+\|\phi\|_{L^q}$, 

880: and we see that $\cF_q$ is bounded on 

881: $H_1^q\cap (\ker D)^\perp\setminus \{0\}$. An arbitrary spinor field 

882: $\phi\in H_1^q$ is written as the sum of $\phi_1\in \ker D$ and a non-zero

883: $\phi_2\perp\ker D$. As $\cF_q(\phi)=\cF_q(\phi_2)$, we see that

884: $\mu_q$ is finite. 

885: 

886: Because of $\phi\in H_1^q$, we have 

887: $|\Dir \phi|^{q-2}\Dir\phi \in L^p$.

888: Hence the right hand side of \eref{eq.abl} defines a continuous

889: functional on $L^p$ and thus on $H_1^q$. We denote the functional

890: by $\psi\mapsto \RHS_\phi(\psi)$.

891: Similarly, one sees that

892:   $$\cF_q(\phi+ \psi)-\cF_q(\phi)- \RHS_\phi(\psi)\leq o(\|\psi\|_{H_1^q}),$$

893: hence $\cF_q$ is Fr\'echet differential with derivative 

894: $\psi\mapsto \RHS_\phi(\psi)$.

895: 

896: If $\phi$ is an eigenspinor to a positive eigenvalue, then 

897: $\mu_q\geq \cF_q(\phi)>0$.

898: \qed

899: 

900: 

901: \begin{proposition}[Properties of $\mu_q^g$]\label{prop.func}

902: The function $\mu_q^g:[\qD,\infty)\to (0,\infty)$ is continuous from the right,

903: and 

904:   $$\mu_2(M,g,\si)=(\la^+_1(g))^{-1}$$

905: Furthermore, if $\vol(M,g)=1$, then $\mu_q^g$ is non-increasing in $q$.

906: \end{proposition}

907: 

908: 

909: \proof{}

910: We assume $\vol(M,g)=1$, the statements for  $\vol(M,g)\not=1$ 

911: then follow by rescaling.

912: 

913: That $\mu_q^g$ is non-increasing follows easily from the H\"older inequality. 

914: 

915: In order to show

916: the continuity from the right, let $q\geq \qD$ be given.

917: We take a smooth spinor field $\phi$ such that

918: $\cF_q^g(\phi)\geq \mu_q^g-\ep$.

919: Observe that  $$\cF_{q'}^g(\phi)= { \|\Dir_g\phi\|_{L^q(g)}^2 \over \|D_g\phi\|_{L^{q'}(g)}^2}\,\cF_{q}^g(\phi).$$

920: The function $q'\mapsto \|\Dir_g\phi\|_{L^{q'}(g)}$ is continuous, hence

921: if $q'\geq q$ is sufficiently close to~$q$, then

922:   $$\mu_{q'}^g\geq \cF_{q'}^g(\psi)\geq \cF_q^g(\psi)-\ep \geq \mu_q^g -2 \ep.$$

923: Because $q\mapsto \mu_q^g$ is non-increasing, the continuity from the right follows.

924: 

925: The formula 

926: $\mu_2^g=(\la^+_1(g))^{-1}$ follows directly if one writes~$\phi$ as a sum of eigenspinors and evaluates

927: $\cF_2^g(\phi)$.

928: \qed

929: 

930: \begin{figure}

931: %

932: % Dies ist die Datei graph.pic.

933: % Dieses Bild zeigt

934: % die Funktion q\mapsto \mu_q

935: %

936: %

937: %

938: \newdimen\axdim

939: \axdim=1pt

940: \newdimen\cudim

941: \cudim=2pt

942: \def\ticklen{.2}

943: \def\abst{.5}

944: 

945: \begin{center}

946: \psset{unit=1cm}

947: 

948: \begin{pspicture}(-1,-\ticklen)(10.5,6)

949: \psset{linewidth=\axdim}

950: \psaxes[linewidth=\axdim,labels=none,ticks=none]{->}(0,0)(9.5,5)

951: \rput(9.9,0){$q$}

952: \rput(0,5.4){$\mu_q$}

953: \psline(2.4,-\ticklen)(2.4,\ticklen)

954: \psline(3.5,-\ticklen)(3.5,\ticklen)

955: \rput[t](2.4,-\abst){$\qD$}

956: \rput[t](3.5,-\abst){$2$}

957: \psline(-\ticklen,2.8)(\ticklen,2.8)

958: \psline(-\ticklen,3.7)(\ticklen,3.7)

959: \rput[r](-\abst,2.8){$(\la^+_1)^{-1}$}

960: \rput[r](-\abst,3.7){$\mu_\qD$}

961: \psline[linestyle=dotted]{-}(2.4,0)(2.4,3.7)

962: \psline[linestyle=dotted]{-}(3.5,0)(3.5,2.8)

963: \psline[linestyle=dotted]{-}(0,3.7)(2.4,3.7)

964: \psline[linestyle=dotted]{-}(0,2.8)(3.5,2.8)

965: \psset{linewidth=\cudim}

966: \psecurve[showpoints=false](2.2,3.74)(2.4,3.7)(3.5,2.8)(4,2.0)(5.1,1.6)(7.2,1.4)(7.5,1.2)

967: \end{pspicture}

968: \end{center}

969: \caption{$\mu_q$ as a function of $q$}

970: \end{figure}

971: 

972: \begin{remark}

973: In Proposition~\ref{prop.sol.subcrit} we will see that the supremum

974: defining $\mu_q^g$

975: is attained for $q>\qD$ by a $C^{2,\al}$-spinor. This implies

976: that the function

977:   $$[\qD,\infty)\to (0,\infty],\quad q\mapsto \mu_q^g$$

978: is also continuous from the left.

979: \end{remark}

980: 

981: 

982: %\begin{corollary} The smallest positive eigenvalue

983: %of the Dirac operator is bounded

984: %from below by $\mu_{\qD}^{-1}$:

985: %  $$\la^+_1\geq \mu_{\qD}^{-1}\qquad \lammin\geq \mu_{\qD}^{-1}.$$

986: %\end{corollary}

987: 

988: \begin{proposition}\label{prop.lamu}

989:   $$\mu_{\qD}(M,g_0,\si)={1\over\lammin(M,g_0,\si)}$$

990: \end{proposition}

991: \proof{}

992: We have already seen that $\mu_{\qD}$ is conformally invariant, \ie for $g_1\in [g_0]$

993:   $$\mu_{\qD}(M,g_0,\si)=\mu_{\qD}(M,g_1,\si).$$

994: The previous proposition states that 

995:   $$\mu_{\qD}(M,g_1,\si)\geq\mu_2(M,g_1,\si)=\left(\la_1^+(g_1)\right)^{-1}$$ 

996: if $\vol(M,g_1)=1$, and it follows

997:   $$\mu_{\qD}(M,g_0,\si)\geq {1\over \lammin(M,g_0,\si)}.$$

998: 

999: It remains to show the inverse inequality which amounts to showing 

1000:   $$\sup_{g\in [g_0]} \la_1^+(g)^{-1}\vol(M,g)^{-1/n}\geq \sup \cF_\qD.$$

1001: For any $\ep>0$ we take a smooth spinor $\phi_\ep$ with 

1002: $\cF_\qD(\phi_\ep)\geq \sup \cF_\qD-\ep$ and $\|\Dir\phi_\ep\|_{L^\qD}=1$.

1003: After a small perturbation of $\phi_\ep$ we can assume that 

1004: $\Dir\phi_\ep$ has no zeros.

1005: We set

1006:   $$g_\ep:=|\Dir\phi_\ep|^{4/(n+1)}g_0.$$

1007: Then we have $\vol(M,g_\ep)=1$, and $|\Dir\phi_\ep|_{g_\ep}$ has constant length $1$. 

1008: Hence,

1009:   $$\cF_\qD^{(M,g_0,\si)}(\phi_\ep)=\cF_\qD^{(M,g_\ep,\si)}(\phi_\ep)

1010:   = \cF_2^{(M,g_\ep,\si)}(\phi_\ep)\leq \mu_2(M,g_\ep,\si)=\la_1^+(g_\ep)^{-1}.$$ 

1011: This implies the proposition.\qed

1012: 

1013: Now, as we have understood the relationship between the supremum of $\cF_\qD$ and the infimum of $\la_1^+\vol^{1/n}$, 

1014: we want to establish a relationship of the maximizers of $\cF_\qD$ and the minimizers of $\la_1^+\vol^{1/n}$.

1015: This will require some knowledge about the Euler-Lagrange-equation.

1016: 

1017: One easily sees that 

1018:   $$\cF_q(\alpha \phi+\psi)= \cF_q(\phi)$$

1019: for any $\psi\in \ker\Dir$ and $\alpha\in \mR^*$. 

1020: Hence maximizers of $\cF_q$ appear in families and it will be 

1021: convenient to choose a good representative for each family of maximizers.

1022: 

1023: \begin{lemma}[Euler-Lagrange equations of $\cF_q$.]\label{lem.euler.lagrange}

1024: Let $q\in [\qD,2]$, and choose $p$ with ${q^{-1}+p^{-1}=1}$. 

1025: Suppose that $\cF_q$ has a maximizing spinor 

1026: $\phi_1\in H_1^q\setminus \ker \Dir$.

1027: Then there is a maximizing spinor $\phi\in H_1^q\setminus \ker \Dir$, $\phi\in \mR^*\phi_1 + \ker D$, 

1028: such that

1029: \begin{equation}\label{eq.nonlin.p}

1030: \Dir\phi= \mu_q^{-1}\, |\phi|^{p-2} \phi,\qquad \phi\in H_1^q, \qquad 

1031: \|\ph\|_{L^p}=1

1032: \end{equation}

1033: \end{lemma}

1034: 

1035: \proof{} We normalize $\phi_1$ such that $\|\Dir \phi_1\|_{L^q}=1$, thus $\rho_{q,\phi_1}=\mu_q$.

1036: As $\phi_1$ is a maximizer, $d\cF_q(\phi_1):H_1^q\setminus \ker \Dir\to \mR$ is identically zero.

1037: Because of \eref{eq.abl} we see that $\tau:=\phi_1-\mu_q|\Dir\phi_1|^{q-2}\Dir\phi_1\in L^p$ is a weak solution

1038: of $D\tau=0$, hence it is smooth and in the kernel of $\Dir$. Then, $\phi_2:=\phi_1-\tau=\mu_q|\Dir\phi_1|^{q-2}\Dir\phi_1$ satisfies $D\phi_1=D\phi_2$ and hence

1039: \begin{equation}\label{eq.nonlin.q}

1040:   \phi_2 = \mu_q |D\phi_2|^{q-2}D\phi_2, \qquad

1041:   \phi_2\in H_1^q, \qquad\|D\phi_2\|_{L^q}=1.

1042: \end{equation}

1043: Taking norms we obtain 

1044:   $$|\phi_2|=\mu_q|\Dir\phi_2|^{q-1}=\mu_q|\Dir\phi_2|^{q/p}.$$

1045:   $$|D\phi_2|^{q-2}=|D\phi_2|^{-(q-1)(p-2)}=\left(\mu_q^{-1} |\phi_2|\right)^{-(p-2)}$$

1046: Hence $\phi:=\mu_q^{-1}\phi_2$ satisfies \eref{eq.nonlin.p}.

1047: \qed

1048: 

1049: \begin{theorem}\label{theo.var.corresp}\ \\[-8mm]

1050: \begin{enumerate}[{\rm (a)}]

1051: \item Let $\psi$ be a maximizing spinor of $\cF_\qD$, 

1052: and suppose that $\psi$ is smooth and that $\Dir \psi$ vanishes nowhere. 

1053: Then $g:=|\Dir \psi|^{4/(n+1)}g_0$ is a smooth metric minimizing

1054: $\la_1^+\vol^{1/n}$ in the conformal class $[g_0]$.

1055: \item Let $g\in [g_0]$ be a (smooth) metric minimizing $\la_1^+\vol^{1/n}$,

1056: and let $\psi$ be an eigenspinor of $D_g$ to the eigenvalue $\la_1^+(g)$,

1057: then the length of $|\psi|_g$ is constant and $\psi$ maximizes 

1058: $\cF_\qD$.

1059: \end{enumerate}

1060: \end{theorem}

1061: 

1062: \proof{}

1063: \begin{enumerate}[(a)]

1064: \item Let $\psi$ be a maximizing spinor. According to the previous lemma, 

1065: there is an $\al\in \mR^*$ and a $\tau\in\ker \Dir$ such that

1066: $\phi:=\al\psi +\tau$ satisfies~\eref{eq.nonlin.p} qith $q=\qD$ and $p=\pD$. 

1067: In particular,

1068: $|\phi|^{\pD-1}=\mu_\qD |\Dir \phi|=\al\mu_\qD|\Dir\psi|$ vanishes nowhere.

1069: The metric $g_\phi:=|\phi|^{4\over n-1}g_0$ satisfies 

1070: $\vol(M,g_\phi)=1$ and $|\phi|_{g_\phi}=1$.  

1071: Then 

1072:   $$\left(\la_1^+(g_\phi)\right)^{-1}\geq \cF_2^{g_\phi}(\phi)=

1073:     \cF_\qD^{g_\phi}(\phi)=

1074:     \cF_\qD^{g_0}(\phi)=

1075:     {\mu_\qD^{-1}\int|\phi|^{\pD}\over \mu_\qD^{-2}

1076:     \left(\int|\phi|^{(\pD-1)\qD}\right)^{2/\qD}}

1077:     =\mu_\qD.$$

1078: As $\mu_\qD^{-1}=\lammin$ we see that $\la_1^+(g_\phi)\leq\lammin$, 

1079: hence $g_\phi$ minimizes $\la_1^+\vol^{1/n}$. By a simple rescaling argument 

1080: one sees that 

1081: $$g:=|\Dir \psi|^{4/(n+1)}g_0=\left({1\over \al}\right)^{4/(n+1)}|\Dir \phi|^{4/(n+1)}g_0=\left({\lammin\over \al}\right)^{4/(n+1)}g_\phi$$

1082: minimizes $\la_1^+\vol^{1/n}$ as well. 

1083: \item By rescaling we can assume that $\vol(M,g)=1$. Unless otherwise

1084: indicated all volume measures, norms, scalar products and 

1085: Dirac operators in this proof are with respect to $g$.

1086: In order to show that $|\psi|_g$ is constant, we define

1087:   $$f_t:={1+t|\psi|_g^2\over \left(\int \left(1+t|\psi|_g^2\right)^n\right)^{1/n}}.$$

1088: One calculates ${d\over dt}|_{t=0}f_t= |\psi|_g^2- \int|\psi|_g^2$.

1089: All metrics $f_t^2 g$ have volume $1$.

1090: Hence, as the infimum of $\la_1^+\vol^{1/n}$ is attained in $g$, we 

1091: have $\la_1^+(g)\leq\la_1^+(g_t)$, and hence

1092:   $$\mu_2(M,g,\si)=\cF_2^g(\psi)\geq \cF_2^{f_t^2g}(\psi)=

1093:     {\int \<D\psi,\psi\>\over \int f_t^{-1}|D\psi|^2}.$$

1094: For $t=0$ equality is attained. Hence 

1095:   $$\int |D\psi|^2\leq \int f_t^{-1} |D\psi|^2.$$

1096: Using $D\psi=\la_1^+(g) \psi$ and deriving with respect to $t$ yields 

1097:   $$0=- \left(\la_1^+(g)\right)^2\int \left(|\psi|_g^2- \int|\psi|_g^2\right)|\psi|_g^2.$$

1098: The right hand side is equal to 

1099: $-\left(\la_1^+(g)\right)^2\left\||\psi|_g^2- \int|\psi|_g^2\right\|_{L^2}=0$,

1100: hence $|\psi|_g$ is constant. 

1101: 

1102: This implies that $\cF_q^g(\psi)$ is independent of $q$, thus

1103:   $$ \cF_\qD(\psi)=\cF_2^g(\psi)=\left(\la_1^+(g)\right)^{-1}

1104:     =\left(\lammin\right)^{-1}.$$

1105: And hence $\cF_\qD$ attains it supremum in $\psi$. 

1106: \end{enumerate}

1107: \qed

1108: 

1109: \begin{remark}

1110: Later on, we will see that maximizers of $\cF_\qD$ that vanish nowhere

1111: are always smooth.

1112: \end{remark}

1113: 

1114: 

1115: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1116: \section{Generalized  metrics}\label{sec.spd}

1117: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1118: 

1119: Unfortunately, we cannot exclude that maximizers of $\cF_q$ vanish 

1120: somewhere. Maximizers with zeros correspond 

1121: to metrics with certain singularities, more precisely to metrics 

1122: whose conformal factor has zeros. 

1123: These metrics are the main object of this section.

1124: We will summarize some facts about the size of the zero set of maximizers 

1125: in Section~\ref{sec.degen}. 

1126: 

1127: Roughly speaking, generalized metrics are metrics of the form $f^2g_0$ 

1128: where $f\in L^\infty, f\geq 0$. However, for technical and formal 

1129: reasons it is better to use the following definition.

1130: 

1131: \begin{definition}

1132: Let $g_0$ be a smooth metric on a compact manifold $M$. 

1133: A \emph{generalized metric} is a tuple $(f,g)$ where $g\in [g_0]$ and

1134: $f\in L^\infty$, $f\geq 0$. If $h>0$ is smooth, we identify

1135: $(fh,g)$ with $(f,h^2g)$. Furthermore we identify $g\in[g_0]$ with $(1,g)$.

1136: Generalized metrics having a representative of the form $(f,g_0)$ are called

1137: \emph{conformal} to $g_0$, and those having a representative $(1,g)$, 

1138: $g\in [g_0]$ are called \emph{regular metrics} --- they correspond to metrics 

1139: in the ordinary sense.

1140: The set of all generalized metrics conformal to $g_0$ is called the 

1141: \emph{$L^\infty$-completion} $\ol{[g_0]}$ of the conformal class $[g_0]$.

1142: The \emph{volume of $g\in \ol{[g_0]}$} is defined as $\int f^n \dvol_{g_0}$.

1143: For a generalized metric we say that $f^{-1}(0)$ is the \emph{set of 

1144: degeneration}. 

1145: \end{definition}

1146: 

1147: \begin{remark}

1148: The reader should pay some attention to the following technical difficulty: 

1149: If $f$ vanishes on an open set, then the 

1150: $L^\infty-(2,0)$-tensor $f^2g_0$ does not determine (the equivalence class) 

1151: $(f,g_0)$.

1152: \end{remark}

1153: 

1154: However, despite of this remark and slightly abusing the notation, 

1155: we will write $f^2g_0$ instead of $(f,g_0)$. 

1156: Formally $f^2g_0$ is a generalized metric in the above sense, 

1157: not an $L^\infty-(2,0)$-tensor.

1158: 

1159: Let us now assume that $M$ carries a fixed spin structure $\si$. 

1160: For any generalized metric $g=f^2g_0$ 

1161: we want to define a spinor bundle and a Dirac operator 

1162: on $(M,g,\si)$ in such a way that the results of the previous section 

1163: carry over to this generalization. In particular, the functional $\cF_q$

1164: has to be defined and has to be conformally invariant for $q=\qD$.

1165: 

1166: As a vector bundle the spinor bundle $\Si(M,g,\si)$ 

1167: is defined to be $\Si(M,g_0,\si)$, and due

1168: to our identification of spinors for different metrics in a fixed conformal

1169: class this construction does only depend on the conformal class $[g_0]$ 

1170: and not on the metric $g_0$ itself. 

1171: For any $x\in M$ and any spinor $\phi$ in the fiber of $\Si(M,g,\si)$ over $x$

1172: we define the pointwise norm

1173: $$|\phi|_g:= \left\{

1174:     \begin{matrix}

1175:      f(x)^{-{n-1\over 2}}|\phi|_{g_0}\hfill & \mbox{if $f(x)\neq 0$}\hfill\cr

1176:      \infty\hfill& \mbox{if $f(x)=0$ and $\phi(x)\neq 0$}\hfill\cr

1177:      0\hfill& \mbox{if $f(x)=0$ and $\phi(x)= 0$}\hfill

1178:     \end{matrix}

1179: \right.

1180: $$

1181: Again, this norm depends only on the conformal class. For smooth sections

1182: $\phi$ of $\Si(M,g,\si)$ such that $|\phi|_g<\infty$ almost everywhere 

1183: we define the $H_1^2(M,g,\si)$-norm as 

1184:   $$\left(\int f^{-1}|\Dir_{g_0}\phi|_{g_0}^2\,\dvol_{g_0}\right)^{1/2}

1185:     + \left(\int |\phi|_{g_0}^\pD\,\dvol_{g_0}\right)^{1/\pD}$$

1186: where we used the conventions $0^{-1}r=\infty$ for $r>0$ and $0^{-1}0=0$. 

1187: The Sobolev space

1188: $H_1^2(M,g,\si)$ is the associated completion. 

1189: \begin{lemma}

1190: There is a natural inclusion 

1191:   $$H_1^2(M,g,\si)\embed H_1^2(M,g_0,\si)$$

1192: \end{lemma}

1193: \proof{}

1194: Cauchy sequences with  respect

1195: to the norm $H_1^2(M,g,\si)$ are also Cauchy sequences with respect to 

1196: $H_1^2(M,g_0,\si)$. Thus we obtain a bounded map  

1197: $H_1^2(M,g,\si)\to H_1^2(M,g_0,\si)$. In order to prove injectivity 

1198: of this map one shows that if $\phi_i$ is a Cauchy sequence in $H_1^2(M,g,\si)$

1199: converging to $0$ with respect to $H_1^2(M,g_0,\si)$, then it converges to $0$

1200: with respect to the $H_1^2(M,g,\si)$ as well.

1201: \qed

1202: 

1203: We define the Dirac operator 

1204: $\Dir_g:H_1^2(M,g,\si)\to L^2(M,g,\si)$, 

1205: $\Dir_g(\phi):=f^{-1}\Dir_{g_0}(\phi)$.

1206: The spinor $D_g (\phi)$ is well-defined almost everywhere, as

1207: $\{x\in M\,|\,D_{g_0}(\phi)(x)\neq 0 \mbox{ and }f(x)=0\}$ has measure zero.

1208: It is easy to verify that all these definitions only depend 

1209: on the conformal class of $g_0$ and not on $g_0$ itself.

1210: Furthermore, we can reformulate the definition of the above Sobolev space as

1211: \begin{eqnarray*}

1212:  H_1^2(M,g,\si) &=& \Bigl\{\phi \in \Gamma(\Si(M,g,\si))\;

1213:     \Big|\;{\textstyle\int} |\Dir_{g}\phi|_{g}^2\,\dvol_{g}<\infty\Bigr\}.

1214: \end{eqnarray*}

1215: 

1216: We now extend the definitions $\cF_2^g$ and $\la_1^+(g)$ 

1217: to the $L^\infty$-completion of the conformal class. 

1218: \begin{definition}

1219: For $g=f^2g_0\in \ol{[g_0]}$ and any $\phi\in H_1^2(M,g,\si)\setminus\ker\Dir$ 

1220: we define

1221:   $$\cF_2^g(\phi):={\int \<\Dir_{g_0}\phi,\phi\>_{g_0}\,\dvol_{g_0}

1222:     \over \int f^{-1} |\Dir_{g_0}\phi|_{g_0}^2\,\dvol_{g_0}}.$$

1223: \end{definition}

1224: Because of  

1225: \begin{eqnarray*}

1226:   \int f^{-1} |\Dir_{g_0}\phi|_{g_0}^2\,\dvol_{g_0}

1227:     &\geq& {1\over \|f\|_{L^\infty}}\, \int|\Dir_{g_0}\phi|_{g_0}^2\,\dvol_{g_0}\\

1228:   \cF_2^g(\phi)&\leq & \|f\|_{L^\infty}\,\cF_2^{g_0}(\phi)

1229: \end{eqnarray*}

1230: the functional is well-defined and bounded on $H_1^2(M,g,\si)\setminus \ker D$.

1231: 

1232: It is also not hard to see that the supremum is attained. In fact, 

1233: let $(\phi_i)$

1234: be a sequence of spinors in  $H_1^2(M,g,\si)\setminus \ker D$ with 

1235: $\cF_2^g(\phi_i)\to \sup \cF_2^g$, 

1236: normalized such that  $\int f^{-1} |\Dir\phi_i|^2=1$. 

1237: Then a subsequence $(\phi_{i_k})$ converges weakly in $H_1^2(M,g,\si)$,

1238: weakly in $H_1^2(M,g_0,\si)$ and strongly in $L^2(M,g_0,\si)$ towards

1239: a $\phi_\infty\in H_1^2(M,g,\si)\setminus \ker D$.

1240: Hence, $\lim \cF_2^g(\phi_{i_k})\leq \cF_2^g(\phi_\infty)$. Thus, 

1241: the supremum is attained in $\phi_\infty$. In analogy to

1242: Lemma~\ref{lem.f.elem} one gets for any smooth 

1243: test spinor $\psi$ 

1244:   $$\int \<f^{-1}\Dir \phi - (\cF_2^g(\phi))^{-1}\,\phi,\Dir \psi\>,$$

1245: which implies 

1246: % in analogy to Lemma~\ref{lem.euler.lagrange}

1247: $\tau:=\Dir_g\phi- (\cF_2^g(\phi))^{-1}\,\phi\in \ker D$ and

1248: finally for $\witi\phi_1:=\phi_\infty+\cF_2^g(\phi)\tau$ 

1249:   $$\Dir_g\witi\phi_1=(\cF_2^g(\witi\phi_1))^{-1}\,\witi\phi_1.$$ 

1250: 

1251: %Note that we have the embeddings  

1252: %  $$H_1^2(M,g,\si)\embed H_1^2(M,g_0,\si)\embed H_{1/2}^2(M,g_0,\si).$$

1253: %Hence, $\mu_2(M,g,\si):=\sup \cF_2^g<\infty$.

1254: \begin{definition}

1255: The \emph{first positive Dirac eigenvalue} 

1256: of $(M,g,\si)$, $g=f^2g_0\in \ol{[g_0]}$

1257: is 

1258:   $$\la_1^+(g):=\left(\sup \left\{\cF_2^g(\phi)\,|\,\phi \in H_1^2(M,g,\si)\setminus\ker\Dir\right\}\right)^{-1}.$$

1259: A non-trivial spinor with 

1260:   $$\Dir_g\phi=\la_1^+(g)\phi$$

1261: is an \emph{eigenspinor to the eigenvalue $\la_1^+(g)$}.

1262: \end{definition}

1263: 

1264: As we have already seen, this definition coincides with the 

1265: definition of the first positive Dirac eigenvalue and a corresponding 

1266: eigenspinor if $g$ is regular. 

1267:   

1268: Most of the statements of the previous section still hold in a 

1269: modified version for generalized metrics, the proofs are nearly identical.

1270: For example one can extend Proposition~\ref{prop.lamu} to

1271: \begin{proposition}\label{prop.bar.nonbar.gen}

1272:   $$\inf_{g\in [g_0])}\la^+_1(g)\vol(M,g)^{1/n}

1273:     =\inf_{g\in \ol{[g_0]}}\la^+_1(g)\vol(M,g)^{1/n} = \mu_{\qD}(M,g,\si)^{-1}.$$

1274: \end{proposition}

1275: The equation $\inf_{g\in [g_0]}\la^+_1(g)\vol(M,g)^{1/n}=\mu_{\qD}(M,g,\si)^{-1}$ 

1276: is exactly the statement of Proposition~\ref{prop.lamu} and the equation 

1277: $\inf_{g\in \ol{[g_0]}}\la^+_1(g)\vol(M,g)^{1/n} = \mu_{\qD}(M,g,\si)^{-1}$ 

1278: can be proven with exactly the same proof.

1279: 

1280: Similarly, we obtain an analogue of Theorem~\ref{theo.var.corresp}.

1281: 

1282: \begin{theorem}\label{theo.var.corresp.gen}

1283: Let $(M,g_0,\si)$ be a compact Riemannian spin manifold.\\[-7mm]

1284: \begin{enumerate}[{\rm (a)}]

1285: \item Let $\psi$ be a maximizing spinor of $\cF_\qD$ 

1286: (with regularity $H_1^\infty$).

1287: Then $g:=|\Dir \psi|^{4/(n+1)}g_0$ is a generalized metric minimizing

1288: $\la_1^+\vol^{1/n}$ in $\ol{[g_0]}$, the $L^\infty$-completion of 

1289: the conformal class $[g_0]$.

1290: \item Let $g=f^2g_0\in \ol{[g_0]}$ be a generalized 

1291: metric minimizing $\la_1^+\vol^{1/n}$,

1292: and let $\psi$ be an eigenspinor of $D_g$ to the eigenvalue $\la_1^+(g)$,

1293: then the length of $|\psi|_g$ is constant on $M\setminus f^{-1}(0)$ and 

1294: $\psi$ maximizes $\cF_\qD$.

1295: \end{enumerate}

1296: \end{theorem}

1297: 

1298: The proof of this theorem is essentially the same as the proof 

1299: of Theorem~\ref{theo.var.corresp}.

1300: Later, we will see that any maximizing spinor has regularity

1301: $C^{2,\al}$, hence $g=|D\psi|^{4/(n+1)}g_0$ is always a generalized metric.

1302: 

1303: 

1304: 

1305: 

1306: 

1307: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1308: \section{The case of the sphere}\label{sec.sphere}

1309: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1310: 

1311: 

1312: The sphere $\mS^n=(S^n,\can)$ with the round sphere 

1313: is an important example. On the one hand the invariant $\lammin$

1314: of the round sphere is contained 

1315: in many equations and inequalities. On the other hand the 

1316: round sphere has a large conformal group,

1317: and hence studying the minimizers on the sphere helps to understand 

1318: the analytical difficulties. In particular, we will see

1319: why the conclusion in Theorem~\ref{theo.anydim} does not hold if

1320: $\mu_\qD=\mu_\qD^{\mS^n}$. (Recall $\qD=2n/(n+1)$.)

1321: 

1322: The invariant $\lammin(\mS^n)$ is not hard to calculate. Recall

1323: that the Yamabe invariant $Y(\mS^n)$ is given by

1324:   $$Y(\mS^n):=\inf_{g\in [\can]}{\int \scal_g\,\dvol_g\over \vol(M,g)^{(n-2)/n}}.$$

1325: It is attained for $g=\can$, and hence $Y(\mS^n)= n (n-1) \, \om_n^{2/n}$,

1326: where $\om_n:=\vol(\mS^n)$.

1327: The Hijazi inequality (\ref{ineq.hij},\ref{ineq.hij.der}) tells us, that

1328: \begin{equation}\label{ineq.sphere.one}

1329:   \lammin(\mS^n)^2\geq {n\over 4 (n-1)}\,Y(\mS^n)= {n^2\over 4}\, \om_n^{2/n}.

1330: \end{equation}

1331: Recall that the sphere of constant sectional curvature $1$ carries

1332: a Killing spinor $\psi$ to the constant $-1/2$, i.e. it satisfies

1333:   $$\na_X \psi= -(1/2) X\cdot \psi.$$

1334: Note that this condition implies 

1335: that the length of $\psi$ is constant. Because of

1336: $D\psi= (n/2) \psi$ we obtain

1337: \begin{equation}\label{ineq.sphere.two}

1338:   \lammin(\mS^n)\leq \la_1^+(\mS^n)\vol(\mS^n)^{1/n}\leq  {n\over 2}\, \om_n^{1/n}.

1339: \end{equation}

1340: %Together with Proposition~\ref{prop.lamu} we obtain

1341: And then

1342:   $$\lammin(\mS^n)

1343:     = {n\over 2}\, \om_n^{1/n}.$$

1344: The above Killing spinor satisfies 

1345:   $$\left(\cF_{\qD}(\psi)\right)^{-1}

1346:     = {n\over 2}\, \om_n^{1/n} \left(\sup \cF_{\qD}\right)^{-1},$$

1347: hence $\cF_{\qD}$ attains its supremum in such Killing spinors.

1348:  

1349: Let $A:\mS^n\to \mS^n$ be an orientation preserving 

1350: conformal diffeomorphism. Then the pullback of any spinor 

1351: $\phi$ with respect to $A$ is a section 

1352: of $A^*(\Si(S^n,\can))=\Si(S^n,A^*\can)$, 

1353: and as before we identify $\Si(S^n,A^*\can)\cong \Si(S^n,\can)$ by using 

1354: the map 

1355: $\witi F$ described after Proposition~\ref{prop.conf.change}.

1356: If $\phi$ is a solution of \eref{eq.nonlin.p}, then this pullback, 

1357: denoted by $A^*\phi$, is a solution of this equation as well. 

1358: Furthermore if $\phi$ maximizes $\cF_{\qD}$, then $A^*\phi$ is also a maximizer.

1359: As a consequence, all conformal images

1360: of Killing spinors to the constant $-1/2$ are maximizers of $\cF_{\qD}$.

1361: The following proposition shows that there

1362: are no other maximizers on $\mS^n$.

1363: 

1364: %We will not use in the proof of our main results. However, it will enter into 

1365: %Remark~\ref{rem.sphereconc}.

1366: 

1367: \begin{proposition}

1368: %On the standard sphere we have

1369: %  $$\sup \cF_{\qD}={2\over n}\,\om_n^{-1/n},$$

1370: %and if the supremum is attained in $\psi$, 

1371: If $\psi$ is a spinor of regularity $C^2$ that 

1372: attains the supremum of $\cF_{\qD}$, 

1373: then there is a Killing spinor $\phi$ to the Killing constant $-1/2$

1374: and an orientation preserving conformal diffeomorphism $A:\mS^n\to \mS^n$

1375: with $A^*\phi=\psi$.

1376: \end{proposition}

1377: 

1378: Later on we will see that any maximizer of regularity $H_1^q$ 

1379: with $q>\qD$ is even $C^2$. Hence the statement also holds under this 

1380: weaker assumption.

1381: 

1382: \begin{lemma}\label{lem.scal.bound}

1383: Let $(M,g,\si)$ be an arbitrary Riemannian spin manifold 

1384: (not necessarily complete or compact). Assume that there is a spinor $\psi$

1385: of constant length $1$ and with $D\psi=\lambda \psi$. Then

1386:   $$\scal = 4  {n-1\over n}\,\lambda^2\,-4 |\witi \nabla \psi|^2,$$

1387: where $\witi\nabla_X \psi:=\na_X\psi + {\la\over n} X\cdot \psi$ denotes

1388: the \emph{Friedrich connection} on spinors.

1389: \end{lemma}

1390: 

1391: \proof{of the lemma}

1392: The Friedrich connection is a metric connection, hence

1393:   $$0={1\over 2}\,d^*d\<\psi,\psi\> = {\rm Re} \<\witi\na^* \witi\na \psi,\psi\> 

1394:   - \<\witi\na\psi,\witi\na\psi\>.$$

1395: The twisted version of the Schr\"odinger-Lichnerowicz formula 

1396: yields

1397:   $$\left(D-{\lambda\over n}\right)^2=\witi\na^*\witi\na + {\scal\over 4}- {(n-1) \lambda^2 \over n^2}.$$

1398: Hence 

1399:   $$\left(\la-{\la\over n}\right)^2= \<\witi\na^*\witi\na \psi,\psi\>+ {\scal\over 4}- {(n-1) \lambda^2 \over n^2}.$$

1400: We obtain

1401:   $$ {n-1\over n} \lambda^2

1402:    = \<\witi\na\psi,\witi\na\psi\> + {\scal\over 4}.$$

1403: \qed

1404: \proof{of the proposition}

1405: %As we have already seen above, Killing spinors to 

1406: %the Killing constant $-1/2$ are maximizers of $\cF_\qD$ and its images 

1407: %under conformal maps.

1408:  

1409: %As $\cF_\qD$ is conformally 

1410: %invariant, the supremum is also attained by the image of such a Killing spinor

1411: %under an orientation preserving

1412: %conformal map $\mS^n\to \mS^n$.

1413: 

1414: We have to show that the supremum is not attained by any other spinor. 

1415: For proving this, assume that $\psi$ is a maximizer, 

1416: $\|\psi\|_{L^\qD}=1$. 

1417: As the Dirac operator on $\mS^n$ has kernel $\{0\}$, the spinor 

1418: satisfies the Euler-Lagrange 

1419: equation \eref{eq.nonlin.p}.

1420: 

1421: On the open subset $S^n\setminus \psi^{-1}(0)$ we define the metric 

1422:   $$g_1:= |\psi|^{4\over n-1}\can.$$

1423: In this metric \eref{eq.nonlin.p} transforms into a solution of 

1424:   $$D_{g_1}\psi=\lammin \psi\qquad |\psi|_{g_1}\equiv 1.$$

1425: 

1426: On the one hand, one calculates 

1427: \begin{eqnarray*}

1428:    \scal_{g_1}&=& 

1429:     4 {n-1\over n-2}\,|\psi|^{-{n+2\over n-1}}\Delta_{\can} |\psi|^{n-2\over n-1}

1430:     + \scal_{\can}|\psi|^{-{4\over n-1}}\\

1431: \end{eqnarray*}

1432: and integration yields

1433: \begin{eqnarray}%\label{ineq.preyama}

1434: \int_{S^n\setminus \psi^{-1}(0)}\scal_{g_1}\,\dvol_{g_1}

1435:     &=& \int_{M\setminus \psi^{-1}(0)} 4 {n-1\over n-2}\,|\psi|^{{n-2\over n-1}}\Delta_{\can} |\psi|^{n-2\over n-1}

1436:     + n(n-1)|\psi|^{2{n-2\over n-1}}\,\dvol_{\can}\nonumber\\

1437: &=& \int_{S^n}

1438:     4 {n-1\over n-2}\,\left|d|\psi|^{n-2\over n-1}\right|_{\can}^2

1439:     + n(n-1)|\psi|^{2{n-2\over n-1}}\,\dvol_{\can},\label{ineq.preyamb}

1440: \end{eqnarray}

1441: where the last term arises by partial integration. In order to make this step 

1442: precise one has to exhaust $M\setminus \psi^{-1}(0)$ by smooth manifolds

1443: with boundary and partially integrate over these exhausting manifolds. 

1444: The boundary terms vanish in the limit, as $\psi\to 0$ on the boundaries and $d|\psi|$ is bounded.

1445: 

1446: It is a standard fact from the resolution of the Yamabe problem (see e.g.\

1447: \cite{lee.parker:87})

1448: that any $H_1^2$ function $f$ satisfies

1449: \begin{eqnarray}\label{ineq.yamafunc}

1450: {\int_{S^n}

1451:     4 {n-1\over n-2}\,\left|df\right|^2

1452:     + n(n-1)f^2\,\dvol_{\can}\over \left(\int_{S^n} f^{2n\over n-2}\,

1453:     \dvol_{\can}\right)^{n-2\over n}} &\geq& Y(\mS^n)=n(n-1)\om_n^{2/n}.

1454: \end{eqnarray}

1455: Setting $f:=|\psi|^{n-2\over n-1}$, we obtain $\int_{S^n} f^{2n\over n-2}=1$, 

1456: and hence the right side of \eref{ineq.preyamb} is bounded from below 

1457: by $n(n-1)\om_n^{2/n}$.

1458: 

1459: One the other hand, the previous lemma provides

1460:   $$\scal_{g_1}\leq 4 {n-1\over n}\left(\lammin\right)^2=n(n-1)\om_n^{2/n},$$

1461: and as  $\vol(S^n\setminus \psi^{-1}(0),g_1)=1$, we see that we must have

1462: equality in all inequalities involved, in particular in~\eref{ineq.yamafunc}. 

1463: An application of the maximum principle 

1464: \cite{lee.parker:87} yields that $f$ does not vanish, and hence $\psi$ 

1465: has no zeros. Furthermore, $g_1$ is a metric of constant scalar curvature

1466: conformal to $\mS^n$, and such a metric is necessarily of the form

1467: $g_1:=A^*\can$ for an orientation preserving conformal diffeomorphism

1468: $A:\mS^n\to \mS^n$. With respect to $g_1$ we obtain 

1469: $\witi \na \psi=0$, hence $\psi$ is a Killing spinor on $(S^n,g_1)$. 

1470: This implies that $\phi:=(A^{-1})^*\psi$ is a Killing spinor on $\mS^n$.

1471: \qed

1472: 

1473: 

1474: %According to standard results in the Yamabe problem, any metric in the form

1475: %$g=f^2 g_0$, $f$ continuous on $M$ and smooth on $U':=M\setminus f^{-1}(0)$

1476: %satisfies

1477: %  $$Y(M,[g_0])\leq{\int_{U'} \scal_g\dvol_g\over \vol(U',g)^{n-2\over n}}.$$

1478: %Equality in this inequality 

1479: %implies that $f^{-1}(0)=\emptyset$, and that $g$ is a Yamabe metric.

1480: %Applied to $M=S^n$, $f:= |\psi|^{2\over n+1}$ we obtain

1481: %  $$n(n-1)\om_m^{2/n}=Y(S^n,\gcan)\leq 4 {n-1\over n} 

1482: %    \left(\lammin\right)^2\,\int_U\dvol_{g_1} = (n-1)n\, \om_n^{2/n}.$$

1483: %Hence, we have equality, which implies that $g_1$ is a Yamabe metric of 

1484: %volume $1$ on $S^n$. This implies that there is an orientation preserving

1485: %conformal map $h:S^n\to S^n$ such that

1486: %$g_1:=\om_n^{-2/n} h^* \gcan$ and that $\psi$ is the pullback under $h$

1487: %of a Killing spinor on $(S^n,\gcan)$. 

1488: %\qed

1489: 

1490: 

1491: \begin{remark}

1492: There are solutions to \eref{eq.nonlin.p} that do not maximize the functional.

1493: An easy construction of such a solution is as follows. 

1494: The map $A:\mS^2=\mC\cup\{\infty\}\to \mS^2=\mC\cup\{\infty\}$, 

1495: $z\mapsto z^k$, $k\in\mN\setminus\{0\}$ 

1496: is conformal with branching points $0$ and $\infty$. If $\phi$ is a solution

1497: of $D\phi=c|\phi|^2\phi$, then 

1498: %with the help of the identification

1499: %of spinors for conformally related metrics following 

1500: %Proposition~\ref{prop.conf.change}, one can pullback 

1501: %the solution to a solution

1502: the pullback $\psi:=A^*\phi$  is a solution  of 

1503: $D\psi=c|\psi|^2\psi$ on $\mS^2\setminus\{0,\infty\}$.

1504: Here $\psi$ is a section of the pull-backed spinor bundle, which is defined 

1505: using the pull-backed spin structure. The pull-backed spin structure

1506:  on $\mS^2\setminus\{0,\infty\}$ coincides with the standard spin structure

1507: iff $k$ is odd. If $k$ is odd, one can show that the extension of $\psi$ by 

1508: setting $\psi(0)=0$ and $\psi(\infty)=0$ is a solution of 

1509: $D\psi=c|\psi|^2\psi$ on $\mS^2$.

1510: However, $\int|\psi|^4= k \int|\phi|^4$. This implies 

1511: $\cF(\psi)=k^{-1/3}\cF(\phi)$. Hence, if $\phi$ is a Killing spinor, 

1512: and $k\geq 3$, $k$ odd, then $\psi$ is a non-maximizing solution.

1513: \end{remark}

1514: 

1515: 

1516: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1517: \section{Regularity theorems for the Euler-Lagrange equations}\label{sec.reg.theo}

1518: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1519: 

1520: In this section we want to study solutions of the Euler-Lagrange equations.

1521: The first subsection ``Removal of singularities'' will be used in 

1522: the following section to 

1523: extend a solution of $\mR^n=\mS^n\setminus\{\mbox{South Pole}\}$ 

1524: to a solution on $\mS^n$. The second subsection states that solutions

1525: of the Euler-Lagrange equations are $C^{2,p-2}$.

1526: 

1527: \subsection{Removal of singularities}

1528: 

1529: \begin{theorem}[Removal of singularities theorem]\label{theo.sing.rem}

1530: Let $p\in[{n\over n-1},\infty)$.

1531: Let $(U,g)$ be a (not necessarily complete) Riemannian manifold 

1532: equipped with a spin structure, let $x\in U$.

1533: Assume that $\phi\in L^p(\Si(U\setminus\{x\}),g)$ 

1534: satisfies weakly on $U\setminus\{x\}$ the equation

1535: \begin{equation}\label{eq.singsol}

1536:     D\phi = \la |\phi|^{p-2}\phi.

1537: \end{equation}

1538: Then this equation even holds weakly on $U$. In particular, the distribution

1539: $D\phi$ does not have singular support in $x$ and is contained in 

1540: $L^q$. 

1541: \end{theorem}

1542: 

1543: \proof{}

1544: Let $\psi$ be a smooth spinor compactly supported in $U$.

1545: We have to show 

1546: \begin{eqnarray}\label{eq.singsol.show}

1547:   \int_U\<\phi,\Dir \psi\>= \la \int_U \< |\phi|^{p-2}\phi,\psi\>.

1548: \end{eqnarray}

1549: 

1550: For any small $\ep>0$ we choose a smooth 

1551: cut-off function $\eta_\ep:U\to[0,1]$

1552: with $\eta_\ep\equiv 1$ on $B_\ep(x)$, with $|\na \eta_\ep|\leq 2/\ep$

1553: and with support in $B_{2\ep}(x)$.

1554: We rewrite the left hand side as

1555: \begin{eqnarray}

1556:          \int_U\<\phi,\Dir \psi\>

1557:          &=&\int_U\Bigl\<\phi,\Dir \Bigl( (1-\eta_\ep) \psi + \eta_\ep \psi\Bigr)\Bigr\>\\

1558:          &=&\int_U\Bigl\<\phi,\Dir \Bigl( (1-\eta_\ep)\Bigr) \psi\Bigr\> 

1559:             + \int_U\<\phi,\eta_\ep \Dir \psi\> + \int_U\<\phi,\na \eta_\ep \cdot \psi\>\nonumber

1560: \end{eqnarray} 

1561: As $\phi$ is a weak solution of \eref{eq.singsol} 

1562: on $U\setminus\{x\}$, the first term

1563: equals to 

1564:   $$\la \int \< |\phi|^{p-2}\phi,(1-\eta_\ep)\psi\>,$$ 

1565: and for $\ep\to 0$ it tends to the right hand side of~\eref{eq.singsol.show}.

1566: 

1567: Let $q$ be related to $p$ via $1/q+1/p=1$.

1568: The absolute value of the second term is bounded by 

1569:   $$\|\phi\|_{L^p(B_{2\ep})}\,\|D\psi\|_{L^q(B_{2\ep})}$$

1570: which tends to $0$ for $\ep\to 0$.

1571: 

1572: Finally, the absolute value of the third term is bounded by

1573:   $${2\over \ep}\,\|\phi\|_{L^p(B_{2\ep})}\,\|\psi\|_{L^q(B_{2\ep})}\leq C\,\|\phi\|_{L^p(B_{2\ep})}\,

1574:     \ep^{{n\over q}-1}.$$

1575: Our condition $p\geq {n\over n-1}$ yields $q\leq n$, and hence the third term 

1576: also tends to $0$ for $\ep\to 0$.

1577: \qed

1578: 

1579: 

1580: \subsection{Regularity}

1581: 

1582: \begin{theorem}[$C^{1,\alpha}$-regularity theorem]\label{theo.reg}

1583: Suppose that $\phi\in H_1^q$, $q\in [\qD,2]$

1584: is a solution of equations~\eref{eq.nonlin.p}.

1585: Suppose that there is an $r>\pD$ such that $\|\phi\|_{L^r}<\infty$.

1586: Then $\phi$ is $C^{1,\alpha}$ for any $\alpha\in (0,1)$.\\

1587: Furthermore, we obtain a uniform bound of the $C^{1,\alpha}$-norm in the 

1588: following sense.

1589: Let us choose $k,K>0$ such that $\|\phi\|_{L^r}<k$ and $\mu_q\geq K$.

1590: Then for any $\al\in (0,1)$ there is a constant

1591:  $C$ depending only on $(M,g,\si)$, $p$, $r$, $K$, $k$ and $\al$ with

1592:   $$\|\phi\|_{C^{1,\al}}\leq C.$$

1593: \end{theorem}

1594: 

1595: \begin{remark}The following

1596: example shows that the theorem cannot hold without the $L^r$-bound. 

1597: Let $M=\mS^n$ and $p=\pD$. 

1598: Let $\psi$ be a Killing spinor to the Killing constant $-1/2$.

1599: Suppose $\|\psi\|_{L^\pD}=1$. 

1600: Let $A:\mS^n\to \mS^n$ be a M\"obius transformation, 

1601: such that the differential in $x\in \mS^n$ 

1602: satisfies $(dA)_x=2 \Id$.

1603: Then $\psi_i=d(A^i)\psi$ is a family of solutions of \eref{eq.nonlin.p}, 

1604: maximizing $\cF_\qD$.

1605: However, for any $r>\pD$ one can show that $\|\psi_i\|_{L^r}\to \infty$ 

1606: for $i\to \infty$.

1607: Hence, the $L^r$-bound is necessary for the theorem to hold. 

1608: \end{remark}

1609: 

1610: \begin{remark}

1611: The theorem will be applied in several versions. At first, we will apply it when

1612: $p<\pD$. In this case the Sobolev embedding already provides the required 

1613: $L^r$-bound on $\phi$. However, the $r$ given by the Sobolev embedding depends on $p$. It will

1614: be of central importance to obtain a bound that is uniform for $p\to \pD$. 

1615: After having proved Theorem~\ref{theo.anydim}  

1616: the uniformity statement in the above theorem will be used to 

1617: obtain a bound that is uniform for $p\to \pD$.

1618: Finally, the regularity theorem will be applied in the case $p=\pD$. In this case an additional $L^r$-bound 

1619: is required as well. 

1620: \end{remark}

1621: 

1622: \proof{}

1623: The proof uses the following ``bootstrap argument''.

1624: At first, we assume $r<n\,{n+1\over n-1}$.

1625: As $\phi$ is $L^r$, the right hand side of  \eref{eq.nonlin.p}, i.e.\ $|\phi|^{p-2}\phi$, 

1626: is $L^{r/(p-1)}\embed L^s$ with $s:={r/(\pD-1)}= r {n-1\over n+1}< n$.

1627: We apply the Global $L^p$-estimates~\ref{theo.global.lp} and get $\phi\in H_1^s$.

1628: Using the Sobolev embedding I, Theorem~\ref{theo.sobo} (a)

1629: one obtains $\phi\in L^{r'}$ with $r'={ns\over n-s}={rn {n-1\over n+1}\over n-r{n-1\over n+1}}$.

1630: Using $r>\pD=2n/n-1$ one sees that $r'>r$, hence we have obtained stronger regularity for $\phi$.

1631: We iterate this argument and get $L^{\ti r}$-bounds for arbitrarily 

1632: large~${\ti r}$. For any

1633: $\ti r>n\,{n+1\over n-1}$, we obtain $\phi\in H_1^{\ti s}$ with $\ti s:={\ti r/(\pD-1)}>n$. We apply 

1634: the Sobolev embedding theorem~II, Theorem \ref{theo.sobo} (c) and obtain $\phi\in C^{0,\al}$ for any $\al\in (0,1)$.

1635: Hence $|\phi|^{p-2}\phi$ is $C^{0,\al}$ as well, and applying  Schauder estimates \ref{theo.schauder}

1636: we get $\phi\in C^{1,\al}$ for arbitrary $\al$.

1637: 

1638: The uniformity of the upper bound is clear from the construction.

1639: \qed

1640: 

1641: The bootstrap can be continued and we obtain better regularity.

1642: 

1643: \begin{proposition}[Improved regularity]\label{prop.imp}

1644: We assume the assumptions of the previous theorem.

1645: \begin{enumerate}[{\rm (1)}]

1646: \item Let $U:=M\setminus\phi^{-1}(0)$. Then $\phi|_U\in C^{\infty}(U)$.

1647: \item If $p > 2$, 

1648: then $\phi\in C^{2,\alpha}$  for any $\alpha\in (0,1)\cap(0, p-2]$.

1649: Furthermore,

1650:   $$\|\phi\|_{C^{2,\al}}\leq C,$$

1651: where $C$ depends only on $(M,g,\si)$, $p$, $r$, $K$, $k$ and $\al$.

1652: \item If $n=2$ and $p=\pD=4$, then $\phi\in C^\infty$.  

1653: Furthermore,

1654:   $$\|\phi\|_{C^m}\leq C,$$

1655: where $C$ depends only on $(M,g,\si)$, $p$, $r$, $K$, $k$ and $m$.

1656: \end{enumerate}

1657: \end{proposition}

1658: 

1659: \proof{} (1) On $U$  we can continue the

1660: bootstrap argument and apply inductively the Schauder estimates

1661: Theorem~\ref{theo.schauder}.

1662: We conclude that $\phi$ is smooth on $U$. We obtain~(1). 

1663: However, uniform bounds will be difficult to obtain.

1664: 

1665: (2) The case $p=2$ is trivial. 

1666: Let $p>2$. We know that $\phi$ is $C^{1,\al}$ for any $\alpha$.

1667: Hence, using Appendix~\ref{sec.scha.power} one sees that 

1668: $|\phi|^{p-2}\phi$ is $C^{1,\al}$. The Schauder estimates imply that $\phi$

1669: is $C^{2,\al}$.

1670: 

1671: (3) Similarly, if $n=2$ and $p=4$, 

1672: then $\phi\mapsto |\phi|^{p-2}\phi$ is also smooth in $0$. Hence the bootstrap 

1673: can go on with higher order Schauder

1674: estimates, and we inductively get $\phi\in C^m(M)$ for any $m$. 

1675: The construction obviously provides uniform bounds. We obtain~(3).

1676: \qed

1677: 

1678: 

1679: 

1680: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1681: \section{Solution of the variational principle}\label{sec.solution}

1682: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1683: 

1684: \begin{proposition}\label{prop.sol.subcrit}

1685: Let $q$ and $p$ be related via $1/p+1/q=1$. For $q\in(\qD,2)$ 

1686: the supremum $\mu_q$ of $\cF_q$ is attained by a spinor field

1687: $\phi\in C^{2,\al}$, $\alpha\in (0,1)\cap(0, p-2]$, 

1688: $1/p+1/q=1$. The spinor $\phi$ can be chosen such that $\phi$

1689: is a solution of \eref{eq.nonlin.p}.

1690: \end{proposition}

1691: 

1692: \proof{}

1693: Let $\phi_i$ be a maximizing sequence for $\cF_q$, \ie

1694: $\cF_q(\phi_i)\to \mu_q$. We may assume $\|\Dir \phi_i\|_{L^q}=1$, and that 

1695: $\phi_i$ is orthogonal to $\ker \Dir$.

1696: After taking a subsequence there is a $\phi_\infty\in H_1^q$ 

1697: such that $\phi_i$ converges weakly to $\phi_\infty$ in $H_1^q$.

1698: Thus $\|\Dir\phi_\infty\|_{L^q}\leq \liminf \|\Dir\phi_i\|_{L^q}=1$.

1699: The compactness of the embedding $H_1^q\hookrightarrow L^p$ provides

1700: a subsequence, that converges strongly to

1701: $\phi_\infty$ in $L^p$. 

1702: This implies

1703:   $$\int\<D\phi_i,\phi_i\>=

1704:     \underbrace{\int\<D\phi_i,\phi_i-\phi_\infty\>}_{\leq\|D\phi_i\|_{L^q}\|\phi_i-\phi_\infty\|_{L^p}} 

1705:     +\underbrace{\int\<D\phi_i,\phi_\infty\>}_{\to\int\<D\phi_\infty,\phi_\infty\>}\to \int\<D\phi_\infty,\phi_\infty\>.

1706: $$

1707: %This implies 

1708: %$\int\<\Dir \phi_i,\phi_i\>\to\int\<\Dir \phi,\phi\>$. 

1709: Hence,

1710:   $$\mu_q= \lim \cF_q(\phi_i)= \lim

1711:     {\int\<\Dir \phi_i,\phi_i\>\over \|\Dir\phi_i\|_{L^q}^2}\leq 

1712:     {\int\<\Dir \phi_\infty,\phi_\infty\>\over \|\Dir\phi_\infty\|_{L^q}^2}=

1713:     \cF_q(\phi_\infty)\leq \mu_q.$$

1714: As a consequence, we have equality in all inequalities, in particular

1715: $\|\Dir\phi_\infty\|_{L^q}=1$.

1716: According to Lemma~\ref{lem.euler.lagrange} one can 

1717: find $\al\in \mR^*$ and $\tau\in \ker \Dir$ such that 

1718: $\phi=\al\phi_\infty+\tau$

1719: solves \eref{eq.nonlin.p}.

1720: Proposition~\ref{prop.imp} (Improved Regularity) 

1721: tells us that $\phi$ is $C^{2,\al}$.

1722: \qed

1723: 

1724: 

1725: \begin{theorem}[Uniform $C^0$-estimate]\label{theo.anydim}

1726: Let $\phi$ be a solution of \eref{eq.nonlin.p} with $p\in [2,\pD)$

1727: and $\mu_q \geq \mu_{\qD}^{\mS^n}+\ep$, $\ep>0$.

1728: Then there is a constant $C=C(M,g,\si,\ep)$ such that

1729:   $$\|\phi\|_{C^0} < C.$$

1730: \end{theorem}

1731: 

1732: \begin{remark}\label{rem.sphereconc}

1733: The conclusion of this theorem does not hold any longer, if we drop 

1734: the condition $\mu_q>\mu_\qD^{\mS^n}$. In fact, if $q=\qD$ and 

1735: $(M,g,\si)=\mS^n$, then there is the following counterexample: 

1736: Let $A$ be a non-isometric, but conformal map $\mS^n\to\mS^n$ 

1737: fixing North and South pole. Let $\phi$ be a Killing spinor. As seen in the 

1738: previous section $\phi$ is a maximizer of $\cF_\qD$ and a solution

1739: of \eref{eq.nonlin.p}. The images $\phi_k:=(A^k)_*\phi$ under the conformal 

1740: maps $A^k$ are also  maximizers of $\cF_\qD$ and solutions

1741: of \eref{eq.nonlin.p}. However, as easily seen, $\|\phi_k\|_{C^0}$

1742: tends to $\infty$ for $k\to \infty$.

1743: \end{remark}

1744: 

1745: 

1746: \proof{of the theorem}

1747: Assume that such a constant does not exist. Then we find a sequence

1748: of solutions $\phi_k$ of

1749: \begin{equation}\label{eq.nonlin.pk}

1750:     \Dir\phi_k= \mu_{q_k}^{-1}\, |\phi_k|^{p_k-2} \phi_k,\qquad \phi_k\in H_1^{q_k}, \qquad 

1751: \|\phi_k\|_{L^{p_k}}=1

1752: \end{equation}

1753: $1/q_k = 1 - 1/p_k$, $\mu_{q_k} \geq \mu_{\qD}^{\mS^n}+\ep$ and

1754: \begin{eqnarray}\label{form.div}

1755: \|\phi_k\|_{C^0}\to \infty.

1756: \end{eqnarray}

1757: 

1758: Let us assume for a moment that $p_\infty:=\liminf p_k < \pD$.

1759: In this case, we can choose a subsequence with $p_k\to p_\infty$.

1760: We have 

1761: $1=\|\phi_k\|_{L^{p_k}}^{p_k}= \Bigl\|\,|\phi_k|^{p_k-2}\phi_k\Bigr\|_{L^{q_k}}^{q_k}={\mu_{q_k}}^{q_k}\|D\phi_k\|_{L^{q_k}}^{q_k}$. 

1762: We conclude that 

1763: $\phi_k$ is bounded in $H_1^{\ti q}$ for a $\ti q > \qD$,

1764: and hence in $L^r$ for an $r>\pD$.

1765: Then the regularity theorem (Theorem~\ref{theo.reg}) says that

1766: $\|\phi_k\|_{C^0}$ is bounded, in contradiction to \eref{form.div}.

1767: Hence, the case $p_\infty>\pD$ can not occur, i.e.\ $\lim p_k = \pD$.

1768: 

1769: There is a sequence of points $s_k\in M$ with

1770:   $$m_k:=|\phi_k(s_k)|=\max\left\{|\phi_k(x)|\,\big|\,x\in M \right\}\to \infty.$$

1771: %Since $M$ is compact, we can assume that $s_k$ converges to $p\in M$.

1772: The idea is to blow up suitably the metric such that we obtain

1773: in the limit a solution on Euclidean $\mR^n$. 

1774: 

1775: We define

1776: \begin{eqnarray*} 

1777:   \ti g_k&:=&(m_k)^{2(p_k-2)}g\\

1778:   \ti \phi_k&:=& (m_k)^{(p_k-2)\frac{n-1}{2}-1} \phi_k.

1779: \end{eqnarray*}

1780: One easily verifies

1781: \begin{equation}\label{eq.normed}

1782:    |\ti \phi_k(s_k)|_{\ti g_k}=1,

1783: \end{equation}

1784: and obviously \eref{eq.nonlin.p} transforms into

1785: \begin{equation}\label{eq.preser}

1786:    D_{\ti g_k}\ti \phi_k= \frac{1}{\mu_{q_k}}|\ti\phi_k|_{\ti g_k}^{p_k-2}\ti\phi_k. 

1787: \end{equation}

1788: We calculate

1789: \begin{eqnarray*} 

1790: \left\|\ti\phi_k\right\|_{L^{p_k}(\ti g_k)}^{p_k} 

1791:    &= &\mu_{q_k}\int\<D_{\ti g_k}\ti\phi_k,\ti\phi_k\>_{\ti g_k}\,\dvol_{\ti g_k}\\

1792:    &=& \mu_{q_k}\int\<D_g\ti\phi_k,\ti\phi_k\>\,\dvol_g\\

1793:    &=&\left(m_k\right)^{2 \left((p_k-2)\frac{n-1}{2}-1\right)}

1794:    \underbrace{\left\|\phi_k\right\|_{L^{p_k}(g)}^{p_k}}_{=1}.  

1795: \end{eqnarray*}

1796: We can assume that $m_k\geq 1$. 

1797: As $p_k\leq\pD$ implies $(p_k-2)\frac{n-1}{2}-1\leq 0$, we obtain

1798: \begin{equation}\label{ineq.unif.lp}

1799:   \left\|\ti\phi_k\right\|_{L^{p_k}(\ti g_k)}\leq 1.

1800: \end{equation}

1801: The injectivity radius of $(M,\ti g_k)$ tends to infinity, i.e.\ 

1802: for any $R>0$ there is a $k_0=k_0(R)\in\mN$ 

1803: such that for all $k\geq k_0$ the exponential map 

1804: $\exp^{\ti g_k}_{s_k}:T_{s_k}M\to M$ 

1805: with respect to $\ti g_k$ and based in $s_k$ is a diffeomorphism 

1806: on the ball of radius $R$ around $0$.

1807: Now, we identify $(T_{s_k}M,\ti g_k)$ with $(\mR^n,\geucl)$.

1808: Then $\overline g_k:=(\exp^{\ti g_k}_{s_k})^*{\ti g_k}$ 

1809: is Riemannian metric on $B_R(0)$ that

1810: coincides with $\geucl$ in $0$.

1811: In the limit $k\to \infty$ the metrics $\overline g_k$ converge 

1812: to $\geucl$ in the $C^\infty$-topology.

1813: 

1814: As already said in previous sections, 

1815: the construction of the spinor bundle, its scalar product 

1816: and its connection depends on the Riemannian metric (and the spin structure).

1817: In order to work out the blowup construction one has to define

1818: a pull-back of the spinors 

1819: $\witi\phi_k\in \Gamma(\Si(M,g_k,\si))$ via normal coordinates and then obtain

1820: a spinor in $\Gamma(\Si(B_R(0),\geucl))$. 

1821: In the literature two such pull-backs are used, a pullback

1822: construction carried out in \cite{ammann.humbert.morel:p03av2} and inspired by

1823: \cite{bourguignon.gauduchon:92}, or via radial parallel transport 

1824: (see e.g.\ the solution of the index problem in \cite{roe:88} 

1825: using Getzler rescaling). Both pullbacks can be used here.

1826: The pullback in \cite{ammann.humbert.morel:p03av2} has better 

1827: approximation properties, and is an important tool if one wants to study

1828: fine asymptotics of the blowup. However, the pullback via 

1829: radial parallel transport is technically simpler to introduce, 

1830: hence it will be used here. 

1831: 

1832: For $R>0$ let $\Si_0(B_R(0),\geucl)$ (resp. $\Si_{s_k}(M,\overline g_k,\sigma)$)

1833: be the fiber of $\Si(B_R(0),\geucl)$ over $0$ 

1834: (resp.  $\Si(M,\overline g_k,\sigma)$ over $s_k$).

1835: Let  us define the radial vector

1836: field $X=r{\partial r}=\sum x^i\partial{x^i}$ on $B_R(0)\subset \mR^n$. Its 

1837: length is the distance from $0$.

1838: For sufficiently large $k$, the exponential map of $(M,\overline g_k)$ based 

1839: in $s_k$, denoted by $\exp^{\ti g_k}_{s_k}$ is a diffeomorphism from

1840: $B_R(0)$ onto its image.

1841: One chooses a (complex) linear isometry 

1842: $\Si_0(B_R(0),\geucl)\to\Si_{s_k}(M,\overline g_k,\sigma)$.

1843: This map extends uniquely to a fiber preserving map 

1844: $A:\Si(B_R(0),\geucl)\to\Si(M,\overline g_k,\sigma)$, such that

1845:   $$\begin{matrix}

1846:    %\matrix{

1847:     \Si(B_R(0),\geucl)& \stackrel{A}{\longrightarrow} & \Si(M,\overline g_k,\sigma)\cr

1848:    \downarrow & & \downarrow\cr

1849:    B_R(0) &  \stackrel{\exp^{\ti g_k}_{s_k}}{\longrightarrow} & M

1850:    %}

1851:  \end{matrix}

1852:  $$

1853: commutes and such that

1854:   $$A(\nabla_X \phi)= \nabla_{(\exp^{\ti g_k}_{s_k})_*(X)}A\phi.$$

1855: In a neighborhood of $s_k$, this condition can be equivalently 

1856: characterized by saying that 

1857: $\phi\mapsto A\circ \phi \circ (\exp^{\ti g_k}_{s_k})^{-1}$ maps

1858: parallel sections of $\Si(B_R(0),\geucl)$ to radially parallel 

1859: sections of $\Si(B_R(s_k),\overline g_k,\si)$.

1860: One easily sees that $A$ is a fiberwise isometry, but the connection 

1861: on the spinor bundle is not preserved. However, this will not matter, as 

1862: in the limit $k\to \infty$, $R$ fixed, the connections converge in the 

1863: $C^\infty$-topology.

1864: 

1865: Let $k\geq k_0(R)$. On $(B_R(0),\geucl)$ we define the spinor  

1866:   $$\overline \phi_k:=A^{-1}\circ \witi\phi_k\circ \exp^{\ti g_k}_{s_k}$$

1867: and the operator $D_k:\Gamma(\Si(B_R(0),\geucl))\to\Gamma(\Si(B_R(0),\geucl))$,

1868:   $$D_k:=A^{-1}\circ D_{\overline g_k}\circ A,$$

1869: where 

1870: $D_{\ol g_k}$ is the Dirac operator on $(B_R(0),\overline g_k)$.

1871: We obtain

1872:   $$D_k\overline{\phi}_k= {1\over \mu_{q_k}}\,|\overline{\phi}_k|^{p_k-2}\overline{\phi}_k$$

1873: Note that %with respect to $\overline g_k$

1874:   $$\|\overline\phi_k\|_{C^0(B_R(0))}\leq |\overline\phi_k(0)|=1.$$

1875: Hence, we may apply the interior $L^p$- and the Schauder-estimates 

1876: \ref{theo.int.lp} and \ref{theo.schauder}

1877: to conclude that

1878:   $$\|\overline{\phi}_k\|_{C^{1,\al}(B_{R/2}(0))}\leq C(R),$$

1879: with constants $C(R)$ and $k(R)$.

1880: The constant $C(R)$ does not depend on $k$ if $\overline g_k$ is sufficiently

1881: $C^\infty$-close to $\geucl$. In particular, the $C^\infty$-convergence

1882: $\overline g_k\to \geucl$ says that $C(R)$ does not depend on 

1883: $k$ for $k\geq k_1(R)$.

1884: %The constant $C(R)$ can be chosen such that it 

1885: %depends only on $R$, $n$, $\al$ and the

1886: %$L^\infty$-norms of the Riemann tensor and its first derivative 

1887: %in a sufficiently small neighborhood of $p$. However, the choice of 

1888: %$k(R)$ also depends on the sequence $m_k$. 

1889: % Note that the interior Schauder estimates do depend on the diameter

1890: % of the domain of definition.

1891: 

1892: Compare $D_k$ with the Dirac operator $\Dflat$ on Euclidean $\mR^n$.

1893: (For similar and more explicit calculations the reader might consider 

1894: e.g.\ \cite{pfaefflediss} for pullback via radial parallel transport,

1895: and  \cite{ammann.humbert.morel:p03av2} for the other pullback method). 

1896: We have

1897:   $$\|(\Dflat -D_k)\overline\phi_k\|_{C^{0,\al}(B_{R/2}(0),\geucl)}\leq \tau_{k,R} \|\overline\phi_k\|_{C^{1,\al}(B_{R/2}(0),\overline g_k)},$$

1898: with $\lim_{k\to \infty}\tau_{k,R}=0$. 

1899: The convergence is not uniform in $R$,

1900: but this will not matter in the following.

1901: 

1902: We choose a sequence of radii $R_m\to \infty$. 

1903: For each $R_m$, the Arcela-Ascoli theorem (Theorem~\ref{theo.arc.asc}) 

1904: allows us to choose a 

1905: subsequence of $(\overline\phi_k)$ converging in 

1906: $C^1(B_{R_m}(0),\geucl)$.

1907: After passing to a diagonal sequence, we see that there is a spinor

1908: $\overline{\phi}_\infty$ on $\mR^n$, such that 

1909: $\res{\overline{\phi}_k}{B_{R}(0)}$

1910: converges to 

1911: $\res{\overline{\phi}_\infty}{B_{R}(0)}\in C^1(B_R(0),\geucl)$ 

1912: for all $R>0$.

1913: 

1914: Then $\overline\phi_\infty$ is a solution of

1915:   $$\Dflat \overline\phi_\infty = {1\over\mu_\qD}\, |\overline\phi_\infty|^{\pD-2}

1916:      \overline\phi_\infty$$

1917: on $\mR^n$.

1918: 

1919: The estimate~\eref{ineq.unif.lp} says that

1920: $\|\witi\phi_k\|_{L^{p_k}(B_R(s_k),\witi g_k)}\leq 1$. Hence,

1921: for any $\ep>0$ and $R>0$ there is $k_2=k_2(R,\ep)$ such that

1922:   $$\|\overline\phi_k\|_{L^{p_k}(B_R(0))}\leq 1 + \ep$$

1923: for all $k\geq k_2$.

1924: Because of the $C^1$-convergence $\overline\phi_k\to \overline\phi_\infty$,

1925: Fatou's lemma yields

1926:   $$\|\overline\phi_\infty\|_{L^{\pD}(B_R(0))}\leq 1$$

1927: for any $R\in (0,\infty)$, and finally for $R=\infty$.

1928: 

1929: 

1930: We identify $\overline\phi_\infty$ via stereographic projection with an

1931: $L^\pD$-spinor $\widehat\phi_\infty$ on $\mS^n\setminus\{\mbox{South pole}\}$

1932: with the identification provided by the application $\witi F$ directly after 

1933: Proposition~\ref{prop.conf.change}. We obtain

1934: \begin{eqnarray}\label{eq.Sn}

1935:   D^{\mS^n} \widehat\phi_\infty = {1\over \mu_\qD}\, |\widehat\phi_\infty|^{\pD-2}

1936:      \widehat\phi_\infty

1937: \end{eqnarray}

1938: and $\|\widehat \phi_\infty\|_{L^\pD}\leq 1$. 

1939: The removal of singularities theorem, i.e.\ Theorem~\ref{theo.sing.rem}, 

1940: says that \eref{eq.Sn} holds on the whole sphere $\mS^n$. 

1941:  

1942: 

1943: \begin{eqnarray*}

1944:    \int_{\mS^n}\<D\widehat\phi_\infty,\widehat\phi_\infty\> &=&

1945:          \mu_\qD^{-1}\,\|\widehat\phi_\infty\|_{L^\pD(\mS^n)}^\pD,\\

1946:   \|D\widehat\phi_\infty\|_{L^\qD(\mS^n)}&=& 

1947:          \mu_\qD^{-1}\,\Bigl\|\,|\widehat\phi_\infty|^{\pD-1}\widehat\phi_\infty\Bigr\|_{L^\qD(\mS^n)}

1948:         =   \mu_\qD^{-1}\,\|\widehat\phi_\infty\|_{L^\pD(\mS^n)}^{\pD-1},\\

1949:    \mu_\qD^{\mS^n}\geq \cF^{\mS^n}_\qD(\widehat\phi_\infty)&=& 

1950:          \mu_\qD\|\widehat\phi_\infty\|_{L^\pD(\mS^n)}^{2-\pD}\geq \mu_\qD

1951: \end{eqnarray*}

1952: which is apparently a contradiction to our assumption

1953: $\mu_\qD\geq\mu_\qD^{\mS^n}+\ep$.

1954: \qed

1955: 

1956: 

1957: \begin{proposition}

1958: If there is a $p_0<\pD$ and an $r>\pD$ such that for all

1959: $t\in (p_0,\pD)$ there is a solution $\phi_t$ of equation~\eref{eq.nonlin.p}

1960: with $p=t$, $1/q+1/p=1$ and such that $\|\phi_t\|_{L^{r}}$ is bounded

1961: by a constant $C$ independent from $t$,

1962: then there is a sequence $t_i\to \pD$ such that $\phi_{t_i}$ converges

1963: in the $C^1$-topology

1964: to a solution of equation~\eref{eq.nonlin.p} with $p=\pD$.

1965: \end{proposition}

1966: 

1967: \proof{}

1968: For $p$ sufficiently close to $\pD$, we know because of

1969: Proposition~\ref{prop.func}

1970: that $\mu_t$ is bounded from below by a positive constant. Thus,

1971: we can apply the regularity theorem (Theorem~\ref{theo.reg}) which tells us

1972: that $(\phi_t)$ is uniformly bounded

1973: in $C^{1,\al}$. 

1974: Hence, for a  sequence $(t_i)$ with $t_i<\pD$,

1975: converging to $\pD$,

1976: the spinor fields $\phi_{t_i}$ converge in the $C^1$-topology to a

1977: $C^1$-spinor field $\phi_{\pD}$ which is a solution of 

1978: equation~\eref{eq.nonlin.p} with $p=\pD$. \qed

1979: 

1980: 

1981: 

1982: 

1983: 

1984: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1985: \section{Proof of the main theorem}\label{sec.proofmain}

1986: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1987: 

1988: In this section we want to prove the main result of this publication, namely

1989: Theorem~\ref{theo.main}.

1990: 

1991: \proof{of Theorem~\ref{theo.main}}

1992: Proposition~\ref{prop.sol.subcrit} tells us that

1993: for any $q\in (\qD,2)$, $\qD:=2n/(n+1)$, the functional $\cF_q$ is attained by a maximizer

1994: denoted $\phi_q$ satisfying 

1995: %\begin{equation}%\label{eq.nonlin.p}

1996: $$

1997: \Dir\phi_q= \mu_q^{-1}\, |\phi_q|^{p-2} \phi_q,\qquad \phi_q\in H_1^q, \qquad 

1998: \|\phi_q\|_{L^p}=1

1999: $$

2000: %\end{equation}

2001: where $p$ and $q$ are related via $p^{-1}+q^{-1}=1$.

2002: We have assumed that $\lammin(M,g,\si)< \lammin(\mS^n)$ 

2003: which is equivalent to $\mu_\qD>\mu_\qD^{\mS^n}$.

2004: As the function $q\mapsto \mu_q$ is continuous from the right

2005: (Proposition~\ref{prop.func}), we see that there is an $\ep>0$ such that

2006:  $\mu_q>\mu_\qD^{\mS^n}+\ep$ for $q$ close to $\qD$.

2007: For such $q$, Theorem~\ref{theo.anydim} implies that $\phi_q$ are uniformly bounded in the $C^0$-norm, 

2008: and then we can use Theorem~\ref{theo.reg} to conclude that these $\phi_q$ are even uniformly bounded 

2009: in $C^{1,\al}$. The Theorem of Arcela-Ascoli 

2010: (Theorem~\ref{theo.arc.asc}) 

2011: implies that there is there is a sequence $q_i\to \qD$ such 

2012: that $\phi_{q_i}$ converges in the $C^1$-norm 

2013: to a solution $\phi$ of 

2014: %\begin{equation}%\label{eq.nonlin.p}

2015: $$

2016: \Dir\phi= \lammin\, |\phi|^{\pD-2} \phi,\qquad \phi\in C^1, \qquad 

2017: \|\phi\|_{L^\pD}=1

2018: $$

2019: %\end{equation}

2020: Theorem~\ref{theo.reg} and Proposition~\ref{prop.imp} then show that $\phi$ has the desired regularity,

2021: and statement (A) is proven.

2022: 

2023: We will now show that statement (B) follows from

2024: statement (A).

2025: If we have a solution as in (A), then we set $g_1:=f^{2/(n-1)}g_0$ with

2026: $f=\<\phi,\phi\>$. Note that $\vol(M,g_1)=\int |\phi|^{2n/(n-1)}=1$.

2027: 

2028: The transformation formula for the Dirac operator under conformal changes

2029: (Proposition~\ref{prop.conf.change}) implies that there is a

2030: spinor $\phi_1$ on $(M,g_1,\si)$ such that

2031: %Then $\phi_1:=\phi/|\phi|$ is a bounded solution of

2032:   $$D_{g_1}\phi_1=\la \phi_1,\qquad |\phi_1|_{g_1}\equiv 1.$$

2033: Then obviously, $\la^+_1(g_1) = \lammin$ and (B) follows.

2034: 

2035: Statement (C) of Theorem~\ref{theo.main} will follow from 

2036: Proposition~\ref{prop.zero.est}, 

2037: proven in the next section.

2038: 

2039: %With Proposition~\ref{prop.lamu} we see that Theorem~\ref{theo.first} 

2040: %is equivalent to Theorem~\ref{theo.main} (B). 

2041: 

2042: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2043: \section{The nodal set}\label{sec.degen}

2044: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2045: 

2046: In this section we want to study that the zero set of solutions 

2047: of the Euler-Lagrange equation

2048: \begin{equation}\label{eq.nonlin.p.nn}

2049:   D\phi=c|\phi|^{p-2}\phi.

2050: \end{equation}

2051: Note that very often the zero set of a function $\phi$

2052: solving an equation of the above type is denoted as the 

2053: \emph{nodal set of $\phi$}.

2054: 

2055: The following theorem is due to C. B\"ar \cite{baer:97} for smooth

2056: $P$.

2057: 

2058: \begin{theorem}[Nodal sets for Dirac Operators]

2059: Let $(U,g)$ be a Riemannian manifold and let $\phi$ be a solution of

2060:   $$D\phi= P \cdot\phi$$

2061: where $P$ is a smooth function.

2062: Then the nodal set of $\phi$ has Hausdorff dimension at most $n-2$.

2063: \end{theorem}

2064: 

2065: Unfortunately, \eref{eq.nonlin.p.nn} has not the desired form

2066: as $P=c|\phi|^{p-2}$ is not smooth for non-integer $p$, and

2067: B\"ar's proof does not extend to such a $P$.

2068: Nevertheless, we conjecture that the theorem also holds for 

2069: $P=c|\phi|^{p-2}$, $p>2$. 

2070: \begin{conjecture}

2071: The nodal set of any solution of 

2072: \eref{eq.nonlin.p.nn} has Hausdorff dimension

2073: at most $n-2$.

2074: \end{conjecture}

2075: 

2076: %Using another method, we prove the following theorem for arbitrary

2077: %locally bounded $p$ in \cite{ammann:nodal}.

2078: % 

2079: %

2080: %The theorem tells us that the nodal set of

2081: %solutions of equation~\eref{eq.nonlin.p.nn} has Hausdorff dimension 

2082: %$\leq n-2$.

2083: 

2084: If $n=2$ and $p=\pD=4$, then 

2085: we have better regularity. 

2086: In this case solutions of equation~\eref{eq.nonlin.p.nn}

2087: and the corresponding $P=c|\phi|^2$ are smooth.

2088: Hence, using \cite[Main Theorem]{baer:97}, 

2089: one sees that in this case 

2090: the nodal set of a solution is a discrete subset. The following proposition 

2091: controls its cardinality.

2092: 

2093: 

2094: \begin{proposition}\label{prop.zero.est}

2095: On a compact spin surface $(M,g,\si)$ of genus $\gamma$ 

2096: let $\phi$ be a solution of equation

2097:   $$D\phi= \lambda |\phi|^2\phi,\quad \|\phi\|_{L^4}=1.$$

2098: Then the number of zeros of $\phi$ is at most

2099: $\gamma-1 + {\lambda^2\over 4\pi}.$

2100: \end{proposition}

2101: 

2102: In particular, this implies part (C) of Theorem~\ref{theo.main}.

2103: \proof{}

2104: We set $g_1:= |\phi|^4 g$. Then outside the zero set we know by 

2105: Lemma~\ref{lem.scal.bound} that

2106: the Gauss curvature of $g_1$ is at most $\lambda^2$. Furthermore 

2107: $\vol(M,g_1)=1$.

2108: Let $\phi(p)=0$. The integral 

2109: of the geodesic curvature with respect to $g_1$

2110: over small simply closed loop around $p$ is close to $-2 (2j_p+1) \pi$, 

2111: where $j_p$ is the order of the first non-vanishing term in the Taylor 

2112: expansion of $\phi$ in $p$.

2113: We remove small open disks

2114: around the zeros of $\phi$ from $M$, and we obtain a surface with boundary

2115: $M'$. 

2116: With the Gauss-Bonnet theorem we obtain

2117:   $$2\pi\, \chi(M')=\int_{M'} K_{g_1} + \int_{\pa M'} k_{g_1}

2118:            \leq \lambda^2-\sum (2j_p+1) 2\pi.$$

2119: And hence $2\pi\, (2-2\gamma)=2\pi\,\chi(M)\leq\lambda^2- 4\pi \sum j_p$, 

2120: which implies the proposition.

2121: \qed

2122: 

2123: 

2124: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2125: \section{The spinorial Weierstrass representation}\label{sec.weierrep}

2126: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2127: 

2128: The aim of this section is to recall the spinorial

2129: Weierstrass representation.

2130: 

2131: Weierstrass published a representation of minimal surfaces

2132: in $\mR^3$ in terms of holomorphic functions 

2133: \cite{weierstrass:66}.

2134: His article deals

2135: only with local questions, everything is described in a fixed

2136: conformal chart of the surface. From a modern (chart free)

2137: point of view, it is clear that

2138: these holomorphic sections should be interpreted as a section of

2139: the spinor bundle, and ``holomorphy'' translates into a ``harmonicity'',

2140: i.e.\ the surface is minimal iff the corresponding spinor $\phi$ satisfies

2141: $D\phi=0$.

2142: %However, this modern language was not available in the 19th century, 

2143: %and the local statement already admitted strong applications.

2144: 

2145: During the 20th century several attempts were undertaken to 

2146: globalize the Weierstrass representation and to adapt it to arbitrary surfaces.

2147: Unfortunately, most approaches replaced Weierstrass' original 

2148: approach by a formulation in terms of 

2149: a holomorphic $1$-form and a holomorphic function.

2150: The corresponding formula were quite involved, 

2151: and hence not very suitable for applications. 

2152: 

2153: An amazing breakthrough was achieved by work of

2154: D.~Sullivan,  R.~Kusner, and N.~Schmitt around 1990, 

2155: and independently by U.~Abresch. 

2156: In early 1989, Dennis Sullivan put together some unpublished notes explaining 

2157: the spinorial character of the Weierstrass representation for minimal 

2158: surfaces. In spring 1989, Robert Kusner realized that the spinor

2159: formalism is not limited to minimal surfaces, but extends to conformal

2160: immersions of arbitrary surfaces, as described below. These techniques were

2161: presented in Sullivan's CUNY seminar in 1992, and 

2162: spread around among the experts rapidly. 

2163: Kusner's results found their continuation

2164: in the PhD thesis of Nick Schmitt \cite{schmitt:diss}.

2165: %, a PhD student of Kusner, the thesis was 

2166: %published in 1993.

2167: Schmitt found many interesting applications of the spinorial Weierstrass.

2168: The results of Kusner and Schmitt led to the publications \cite{schmitt:diss}

2169: and \cite{kusner.schmitt:p96}. 

2170: 

2171: Independently, Abresch developed a spinorial Weierstrass for constant mean

2172: curvature surfaces.

2173: %, that can be understood as the special case $H=\const$

2174: %in Kusner's representation. 

2175: Unfortunately, 

2176: the resulting document of Abresch, some handwritten lecture notes from a 

2177: conference in Luminy, were never published.

2178: 

2179: We also want to mention an earlier result of Pinkall \cite{pinkall:85a}. 

2180: In the special class of oriented surfaces, Pinkall's result 

2181: establishes a bijection

2182: between regular homotopy classes of immersions of 

2183: oriented surfaces $M$ and $\mZ_2$-valued quadratic forms on $H^1(M,\mZ_2)$.

2184: These quadratic forms are

2185: in bijection with the spin structures obtained in 

2186: the spinorial Weierstrass representation.

2187: 

2188: %In the following years many generalizations of the spinorial Weierstrass 

2189: %representation were developed. We are aware of \cite{.....}, but we certainly

2190: %missed some important papers.

2191: 

2192: \komment{

2193: The idea to use spinors for this representation seems natural 

2194: when one reads Weierstrass' original article 

2195: \cite{weierstrass:66}, but it is very amazing that

2196: the representation takes such a simple formula in terms of spinors. 

2197: Unfortunately, the center of 

2198: mathematicians' interest moved away from Weierstrass' original 

2199: approach to a formulation in terms of 

2200: a holomorphic $1$-form and a holomorphic function, which 

2201: dominated the literature until the late 1980s.

2202: 

2203: Then, R. Kusner and D. Sullivan found out 

2204: that the spinor formulation gives a simple equation for

2205: non-minimal surfaces, and these ideas led to the preprint 

2206: \cite{kusner.schmitt:p96}.

2207: Although this preprint was not 

2208: published until now, this manuscript is the first document, 

2209: where the spinorial Weierstrass representation was explained.

2210: 

2211: It also has to be mentioned here that 

2212: independently form Kusner, Sullivan 

2213: and Schmitt, Abresch \cite{abresch:89} had already 

2214: worked with another, less simple version of the spinorial Weierstrass 

2215: representation before, but unfortunately, 

2216: the resulting document, some handwritten lecture notes from a conference in

2217: Luminy, seem to be unavailable to the public.

2218: }

2219: 

2220: More recent literature concerning this 

2221: representation can be found for example in 

2222: \cite{baer:98,friedrich:98,ammann:diss} and in articles by Pinkall, Taimanov, 

2223: M.U. Schmidt, Morel, Voss and their collaborators. However, this list 

2224: is far from being exhaustive.

2225:  

2226: In our exposition we roughly follow \cite{kusner.schmitt:p96,friedrich:98}. 

2227: The setting for the spinorial Weierstrass representation is as follows.

2228: Let $M$ be a compact Riemann surface of genus $\gamma$. 

2229: The vector bundles 

2230: $\Lambda^{1,0}T^* M$ and $\Lambda^{0,1}T^* M$ are defined 

2231: as the complex linear part and the complex anti-linear part of 

2232: $T^*M\otimes_\mR \mC$. The compositions 

2233:   $$I^{1,0}:T^*M\to T^*M\otimes_\mR \mC\to \Lambda^{1,0}M$$

2234:   $$I^{0,1}:T^*M\to T^*M\otimes_\mR \mC\to \Lambda^{0,1}M$$

2235: of the complexification and the projection on $\Lambda^{1,0} M$

2236: resp.\ $\Lambda^{0,1} M$

2237: define vector space homomorphisms $T^*M\cong \Lambda^{1,0}M$ and  

2238: $T^*M\cong \Lambda^{0,1}M$. These map $I^{1,0}$and $I^{0,1}$ 

2239: preserve the natural 

2240: connections. However, one should pay attention 

2241: to the fact, that these maps do not preserve lengths, but

2242: $2\Ree g(I^{1,0}(\al),I^{1,0}(\be))=2\Ree g(I^{0,1}(\al),I^{0,1}(\be))=g(\al,\be)$. 

2243: The maps $2(I^{1,0})^{-1}$ resp.\ $2(I^{0,1})^{-1}$ is denoted 

2244: as the \emph{real part}, namely $\Ree (\al)=2(I^{1,0})^{-1}(\al)$ for 

2245: $\al\in\Ga(\Lambda^{1,0}M)$ and the same notation is used in the $(0,1)$ case.

2246: \emph{Complex conjugation} maps $\Lambda^{0,1}M$ to $\Lambda^{1,0}M$ 

2247: and vice versa.

2248: 

2249: As the second Stiefel-Whitney-class of $M$ is the $\mod 2$ reduction 

2250: of the Euler-class of $TM\to M$, one sees that

2251: $M$ is spin. However, the space of spin structures on $M$ is not unique:

2252: it is an affine space 

2253: for the group $H^1(M,\mZ_2)=(\mZ_2)^{2\gamma}$, where $\gamma$ denotes 

2254: the genus of $M$. Hence, there are $4^\gamma$ spin structures on $M$.

2255: 

2256: If a spin structure is fixed, then the associated vector bundle with respect

2257: to the standard representation of $S^1$ on $\mC$ is a complex 

2258: line bundle $\Si^+M$ satisfying $\Si^+M \otimes \Si^+M\cong \Lambda^{0,1}M$.

2259: If we equip $\Si^+M$ with the natural hermitian metric, we can choose this map such that

2260: the hermitian metric (res.\ the connection) on $\Lambda^{0,1}M$ is 

2261: the tensor product metric (resp.\ tensor product connection).

2262: 

2263: We define $\Si^-M$ to be $\Si^+M$ with the conjugated complex structure. 

2264: In particular,

2265: there is a natural anti-linear conjugation map $\Si^+M\to \Si^-M$, the 

2266: hermitian product defines a complex bilinear metric contraction 

2267: $\Si^-M\otimes \Si^+M\to \mC$, and $\Si^-M\otimes \Si^-M\cong \La^{1,0}M$. 

2268: In particular,

2269: $\Si^-M\otimes \Lambda^{0,1}M=\Si^-M\otimes \Si^+M \otimes \Si^+M= \Si^+M$, and hence

2270: the Dolbeault operator is a map $\overline\pa:\Gamma(\Si^-M)\to \Gamma(\Si^+M)$, 

2271: and similarly $\overline\pa^*=-\partial: \Gamma(\Si^+M)\to \Gamma(\Si^-M)$.

2272: 

2273: We define $c^{1,0}$ and $c^{0,1}$ as the compositions

2274:   $$TM\stackrel{b}{\to}T^*M\stackrel{I^{1,0}}\to \Lambda^{1,0}M=\Si^-M\otimes \Si^-M=\Hom_\mC(\Si^+M,\Si^-M),$$

2275:   $$TM\stackrel{b}{\to}T^*M\stackrel{I^{0,1}}\to \Lambda^{0,1}M=\Si^+M\otimes \Si^+M=\Hom_\mC(\Si^-M,\Si^+M).$$

2276: Composing $c^{1,0}$ with complex conjugation yields $c^{1,0}$ and vice versa.

2277: %and by inserting the complex conjugation $\Lambda^{0,1}M\to \Lambda^{1,0}M$, one obtains $c^{1,0}$ 

2278: %as the composition 

2279: %  $$TM\stackrel{b}{\to}T^*M\stackrel{I}\to\Lambda^{0,1}M\to \Lambda^{1,0}M=

2280: %    \Si^-M\otimes \Si^-M=\Hom_\mC(\Si^+M,\Si^-M).$$

2281: One calculates

2282:   $$c^{1,0}(X) c^{0,1}(Y)+c^{1,0}(Y) c^{0,1}(X)=g(\overline{I(X)},I(Y))+ g(\overline{I(Y)},I(X))=g(X,Y).$$

2283: As a consequence, the map

2284:   $$TM\to\End\left(\Si^+M\oplus \Si^-M\right), \qquad X\mapsto 

2285: \sqrt{2} \begin{pmatrix}

2286: 0 &c^{0,1}(X)\\-c^{1,0}(X) & 0 

2287: \end{pmatrix}$$

2288: satisfies the Clifford relations.

2289: \longver{

2290: $$c^{0,1}(X)={1\over 2}\,(X^b- i J(X)^b)$$

2291: $$c^{1,0}(X)={1\over 2}\,(X^b+ i J(X)^b)$$

2292: Assume $e_1=\pa_x$, $e_2=J(e_1)=\pa_y$.

2293: We calculate on $\Si^-M$

2294:   $$e_1\pa_{e_1}= {1\over \sqrt{2}}(dx-i\,dy){\pa\over \pa x}$$

2295:   $$e_2\pa_{e_2}= {1\over \sqrt{2}}(dy+i\,dx){\pa\over \pa y}= {1\over \sqrt{2}}i(dx-i\,dy){\pa\over \pa y}$$

2296:   $$D = \sqrt{2} \,{dx- i\, dy\over 2}\left({\pa\over \pa x}+ i\,{\pa\over \pa y}\right)= 

2297:     \sqrt{2} \,\pa_{\ol z} \,d{\ol z}=\sqrt{2}\,\ol{\pa}$$

2298: }

2299: 

2300: One sees, that 

2301: the sum $\Si M:=\Sigma^+M\oplus \Sigma^-M$ can be identified with the standard spinor bundle on $M$ in 

2302: such a way that the above map is the Clifford multiplication, and such that

2303: $\Si^+M$ resp.\ $\Si^-M$ are the positive resp.\ negative half-spinors. 

2304: 

2305: The Dirac operator can be written in this notation as

2306: $$D= \sqrt{2}

2307: \begin{pmatrix}

2308: 0 &\overline\partial \\-\partial & 0 

2309: \end{pmatrix}: 

2310:     \Gamma\Big(\Sigma^+\oplus \Sigma^-\Big)\to 

2311:     \Gamma\Big(\Sigma^+\oplus \Sigma^-\Big).$$

2312: 

2313: Now let us assume that 

2314: $\phi=(\phi_+,\phi_-)$ is a solution of~$D\phi=H|\phi|^2\phi$, where

2315: $H$ is a real-valued function on $M$. 

2316: This means

2317:   $$-\sqrt{2}\pa \phi_+ = H |\phi|^2\phi_-$$

2318:   $$\sqrt{2}\pa \ol{\phi_-} = H |\phi|^2\ol{\phi_+}$$

2319: We define

2320: 

2321: 

2322: $$ \alpha := %\mathrm{Re}

2323: {\sqrt{2}}\,

2324: \begin{pmatrix}

2325:   \phi_+\otimes\phi_+ \,+\,\overline{\phi_-}\otimes \overline{\phi_-}\cr

2326:   i \phi_+\otimes \phi_+\,-i\,\overline{\phi_-}\otimes \overline{\phi_-}\cr

2327:   2 i\phi_+\otimes \overline{\phi_-}

2328: \end{pmatrix}\in\Gamma(\Lambda^{0,1}\otimes_\mR \mR^3).

2329: $$

2330: \longver{

2331: $$\pa \al= 2 i H |\phi|^2 

2332: \begin{pmatrix}

2333: -2\Imm (\phi_+\otimes \phi_-) \cr

2334: -2\Ree (\phi_+\otimes \phi_-) \cr

2335: |\phi|^2

2336: \end{pmatrix}\in\Gamma(i\mR^3).$$

2337: }

2338: 

2339: Let $\witi M$ denote the universal covering of $M$, and $\pi_1(M)$ the group

2340: of Deck transformations.

2341: 

2342: As $\pa\al=d\al$ is imaginary, we can find a function $F:\witi M\to \mR^3$, 

2343: such that $dF=\Ree \alpha$, and there is a homomorphism

2344: ${V:\pi_1(M)\to \mR^3}$, such that 

2345:   $$F(p\cdot\gamma)= F(p) + V(\gamma)\quad \forall p\in \witi M,

2346:     \;\gamma \in \pi_1(M)$$

2347: 

2348: One calculates that

2349: $F$ is a conformal map with possible branching points,

2350:   $$|dF|=|\Ree \alpha|={1\over \sqrt{2}}\,|\alpha|=|\phi|^2,$$

2351: and that $F(M)$ has mean curvature $H$. 

2352: 

2353: Hence, the map $F$ satisfies Properties (1) to (3) from the introduction,

2354: i.e. it is a \emph{periodic branched

2355: conformal immersion $F$ based on $(M,g)$} with mean curvature $H$.

2356: 

2357: In any zero of of the spinor $\phi$, the map $F$ has a branching point.

2358: If $F$ vanishes of order $k$, then $\al$ vanishes of order $2k$.

2359: Hence, all branching points of $F$ are necessarily of even order.

2360: 

2361: Summarizing the above statement, we obtain for any solution of 

2362: $D\phi=H\,|\phi|^2\phi$  

2363: a periodic branched conformal immersion of $\witi M$ into $\mR^3$ 

2364: which is uniquely 

2365: determined up to translation. If $\phi$ solves $D\phi=H\,|\phi|^2\phi$, 

2366: then  $-\phi$ as well, and the corresponding $F$ is the same. Hence, we obtain

2367: a well-defined map 

2368: $$\left\{\vbox{\hbox{solutions of}\vtop{\hbox{$D\phi=H\;|\phi|^2\phi$}%

2369:     \hbox{on $M$}}}\right\}/{\pm 1} 

2370: \quad\longrightarrow\quad

2371:  %$$

2372:  %$$\stackrel{1:1}{\longleftrightarrow}$$

2373:  %$$

2374: \left\{\vbox{\hbox{conformal periodic $H$-immersions}\vtop{\hbox{

2375:    $\witi M\to\mR^3$ %($S^3$, $H^3$) 

2376:    with branching} 

2377:    \hbox{points of even order}}}\right\}/\vtop{\mbox{trans-}\\\mbox{lations}}$$

2378: and one can show that this map is even a bijection.

2379: 

2380: The inverse of this map is given by restricting a parallel spinor

2381: on $\mR^3$ to $F(M)$ and

2382: by performing a conformal change \cite{baer:98}.

2383: 

2384: If $H$ is constant, then 

2385: there is also another version of the spinorial Weierstrass

2386: representation, where the target

2387: space is $S^3$ instead of $\mR^3$. We view $S^3$ as $\SU(2)$ 

2388: with a bi-invariant metric of constant curvature $1$, 

2389: the multiplication in $\SU(2)$ is denoted with 

2390: $\bullet$. The periodicity condition (1) has to be replaced

2391: by

2392: \begin{enumerate}[{\rm (1')}]

2393: \item {\it Left periodicity:}

2394: There is a homomorphism 

2395: $h:\pi(M)\to \SU(2)$, the \emph{periodicity map}, such that 

2396: for any $\gamma\in \pi_1(M)$, and $x\in\witi M$ one has

2397:   $$F(x\cdot \gamma)=h(\gamma)\bullet F(x).$$

2398: Here $\cdot$ denotes the action of $\pi_1$ on $\witi M$ via Deck transformation.\end{enumerate}

2399: 

2400: One obtains a bijection 

2401: $$\left\{\vbox{\hbox{solutions of}\vtop{\hbox{$D\phi=c\;|\phi|^2\phi$}%

2402:     \hbox{on $M$}}}\right\}/{\pm 1} 

2403: \quad\longrightarrow\quad

2404:  %$$

2405:  %$$\stackrel{1:1}{\longleftrightarrow}$$

2406:  %$$

2407: \left\{\vbox{\hbox{conformal left periodic $H$-im-}\vtop{\hbox{

2408: mersions $\witi M\to\SU(2)$ %($S^3$, $H^3$) 

2409: with} 

2410:    \hbox{branching points of even order}}}\right\}/\vtop{\mbox{Left mul-}\\\mbox{tiplication}}$$

2411: where $c=\sqrt{H^2+1}$.

2412: For details see 

2413: \cite{voss:dipl,morel:02,ammann:habil}.

2414: 

2415: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2416: \section{Applications to constant mean curvature surfaces}\label{sec.app.cmc}

2417: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2418: 

2419: If the dimension of $M$ is $2$, then \eref{eq.dirac.nonlin} reads as 

2420: \begin{equation}\label{eq.nonlin.2}

2421: \Dir\phi=\lammin\, |\phi|^{2} \phi,\qquad \phi\in H_1^{4/3}, \qquad 

2422: \|\phi\|_{L^4}=1

2423: \end{equation}

2424: and according to the regularity theory $\phi$ is even smooth.

2425: 

2426: The spinorial Weierstrass representation explained in the previous section

2427: tells us, that such a solution can be used to construct 

2428: certain immersions with constant mean curvature.

2429: 

2430: Combining the previous results we obtain the following application 

2431: that is a stronger version of the 

2432: ``Principle for construction of cmc-surfaces''

2433: mentioned in the introduction. 

2434: 

2435: \begin{proposition}

2436: Assume that the Riemann spin surface $(M,g,\si)$ satisfies 

2437: \begin{equation}\label{ineq.surface}

2438:   \lammin(M,[g],\si)<2\pi.

2439: \end{equation}

2440: Then there is a periodic branched

2441: conformal cmc immersion $F:\witi M\to \mR^3$ based on $(M,g)$. 

2442: The mean curvature is equal to $\lammin(M,[g],\si)$ and the area of a 

2443: fundamental domain is $1$.

2444: The regular homotopy class of $F$ is determined by the spin-structure $\si$.

2445: The indices of all branching points are even, 

2446: and the sum of these indices

2447: is smaller than $2 {\rm genus}(M)$. In particular, 

2448: if $M$ is a torus, there are no branching points.  

2449: \end{proposition}

2450: 

2451: 

2452: The proof is a direct consequence of 

2453: Theorem~\ref{theo.main} and the previous section.

2454: 

2455: There are many examples of stationary points of the functional.

2456: However, it is still open whether they are the maximizers or not.

2457: Note, that by changing the orientation of a surfaces, 

2458: a maximizers $\psi$ of $\cF_\qD$

2459: on $M$ turns into a minimizer $\psi'$ of $\cF_\qD$ on the surface with 

2460: reversed orientation $M'$, with $\cF^{M'}_\qD(\psi')=-\cF^{M}_\qD(\psi)$. 

2461: Let us study some examples.

2462: 

2463: \begin{examples}

2464: \begin{enumerate}[{\rm (a)}]

2465: \item 

2466: Let $(M,g)$ be a $2$-dimensional torus. Via a conformal change 

2467: we can achieve that $g$ is flat, i.e.\ $M=\mR^2/\Gamma$, equipped 

2468: with the Euclidean metric. 

2469: We assume that the lattice $\Gamma$ is generated by 

2470: $\begin{pmatrix}1\cr 0\end{pmatrix}$ and 

2471: $\begin{pmatrix}x\cr y\end{pmatrix}$, with $y>0$.

2472: The spinor bundle of a flat manifold is flat as well, hence the holonomy

2473: is a map $\Gamma\to \SU(\Si_p M)$. Indeed, the image of this map is contained

2474: in $\{\pm \Id\}$.

2475: We obtain a homomorphism $\chi:\Gamma\to \{\pm \Id\}$.

2476: This homomorphisms

2477: characterizes the spin structure $\si$ in the sense that two spin

2478: structures on $(M,g)$ are isomorphic iff the homomorphisms $\chi$

2479: coincide, and to each such  homomorphism there is a spin structure.

2480: The case $\chi\equiv+\Id$ corresponds to the so-called \emph{trivial} 

2481: spin structure $\si_{\mathrm{tr}}$, the other cases correspond to 

2482: \emph{non-trivial} 

2483: spin structures. 

2484: \footnote{Note that this notation is a bit misleading, as it is only the 

2485: \emph{trivial} spin structure on $M$ that defines a \emph{non-trivial}

2486: element in the bordism class.}

2487: 

2488: At first, we deal with the case $\si=\si_{\mathrm{tr}}$. 

2489: In this case, after a possible rotation and a possible homothety, 

2490: we can achieve

2491:   $$|x|\leq {1\over 2},\quad 

2492: y^2+x^2\geq 1,\quad y> 0.$$

2493: On $(M,g,\si_{\mathrm{tr}})$ the kernel of $\Dir$ has complex dimension $2$,

2494: and consists of parallel spinors. If one carries out the constructions

2495: from the last sections for a parallel spinor, then one obtains 

2496: an affine conformal map $F$. Such an $F$ is trivially

2497: a periodic conformal immersion $F$ based on $(M,g)$ 

2498: with \emph{vanishing} mean curvature. 

2499: However, if $y>\pi$ then one easily sees that 

2500: $\lammin(M,g,\si_{\mathrm{tr}})<2\pi$. The proposition

2501: yields the existence of a  periodic conformal immersion $F$ 

2502: based on $(M,g)$ with constant mean curvature $\lammin(M,g,\si_{\mathrm{tr}})$,

2503: and the area of a fundamental domain is $1$.

2504: However, the proposition only provides the existence of the solution, but 

2505: we cannot characterize the maximizer.

2506: In our example, a family of solutions can be explicitly written, namely

2507: it is the family of parameterized cylinders 

2508: \begin{eqnarray*}

2509:   F:\mR^2&\to &\mR^3\\

2510: \begin{pmatrix} a\cr b\end{pmatrix}& \mapsto &

2511: P\begin{pmatrix}{\sqrt{y}\over 4\pi}\cos{4\pi b\over y}\cr{\sqrt{y}\over 4\pi}\sin{4\pi b\over y}\cr {a\over \sqrt{y}} \end{pmatrix}+ X_0 

2512: \end{eqnarray*}

2513: for any $P\in O(3)$, $X_0\in \mR^3$. We conjecture that these solutions are

2514: exactly those that correspond to the maximizers and the minimizers of 

2515: $\cF_\qD$, when we normalize such that all spinors have $L^4$-norm $1$.

2516: 

2517: In the case  $\si\not=\si_{\mathrm{tr}}$

2518: we can achieve

2519: that 

2520: $$\chi\begin{pmatrix}1\cr 0\end{pmatrix}=\Id\qquad \chi\begin{pmatrix}1\cr 0\end{pmatrix}=-\Id,$$

2521:   $$|x|\leq {1\over 2},\quad 

2522: y^2+\left(|x|-{1\over 2}\right)^2\geq {1\over 4},\quad y> 0.$$

2523: The Dirac operator is always invertible.

2524: 

2525: One easily sees $\lammin(M,g,\si)\leq {\pi\over \sqrt{y}}$.

2526: Hence, the proposition yields solutions for $y>{4\over \pi}$. 

2527: Once again, solutions can be explicitly written, namely

2528: the parameterized cylinder 

2529: \begin{eqnarray*}

2530:   F:\mR^2&\to &\mR^3\\

2531: \begin{pmatrix} a\cr b\end{pmatrix}& \mapsto &

2532: P\begin{pmatrix}{\sqrt{y}\over 2\pi}\cos{2\pi b\over y}\cr{\sqrt{y}\over 2\pi}\sin{2\pi b\over y}\cr {a\over \sqrt{y}} \end{pmatrix}+ X_0 

2533: \end{eqnarray*}

2534: for any $P\in O(3)$, $X_0\in \mR^3$. 

2535: However, in some cases, e.g.\ if $x=0$ and $4/\pi<y<1$, 

2536: these solutions no longer correspond to maximizers and minimizers,

2537: but to saddle points of the functional. We conjecture

2538: that in the case $x=0$, $y<1$ the maximizers and minimizers 

2539: correspond to the unduloid immersions (see Figure~\ref{fig.undu}). 

2540: An unduloid is a surface of revolution of constant mean curvature. 

2541: \begin{figure}

2542: \begin{center}

2543: \bild{

2544: \includegraphics[scale=1]{unduloid1.eps}

2545: %\includegraphics[scale=.5]{unduloid2.eps}

2546: }

2547: \end{center}

2548: \caption{An unduloid in $\mR^3$, visualized by Nick Schmitt.}\label{fig.undu}

2549: \end{figure}

2550: \item If $M$ has genus $2$, then as in the case of the torus,

2551: the dimension of the kernel is independent of the metric, however it depends

2552: on the spin structure.

2553: If $\si$ is a spin structure such that

2554: $(M,\si)$ is spin-cobordant $0$, then the Dirac operator

2555: is invertible for any metric. Again, as in the torus case, one can find 

2556: for any $\ep>0$ a conformal classes

2557: ${g}$ on $M$ with 

2558: $\lammin(M,[g],\si)<\ep$. \cite{ammann.humbert:gluing}

2559: \item If the genus is larger than $2$, then the kernel of the Dirac operator

2560: on a Riemannian spin manifold~$(M,\si)$ 

2561: depends on the metric. For example if $M$ is 

2562: a surface of genus $3$ equipped with the spin structure $\si$ and the conformal

2563: structure ${g_0}$ associated to the periodic conformal immersion with 

2564: vanishing mean curvature indicated in Figure~\ref{fig.kgb.min}.

2565: \footnote{N. Schmitt's illustrations in this paragraph are available 

2566: on his website \url{http://www.gang.umass.edu/gallery/cmc/cmcgallery0302.html},

2567: K. Grosse-Brauckmann's illustrations are available on 

2568: \url{http://www.math.uni-bonn.de/people/kgb/Research/folie\_{}iwp.gif}}

2569: This immersion induces a harmonic spinor on $(M,g,\si)$. However, as 

2570: $(M,\si)$ is spin-cobordant $0$, there is a perturbation $[g_t]$ 

2571: of the conformal structure such that the Dirac operator on $(M,g_t,\si)$

2572: has a trivial kernel for small $t\neq 0$ \cite{maier:97}. In this case 

2573:   $$\lim_{t\to 0\atop t\neq 0}\lammin(M,[g_t],\si)=0,$$ 

2574: hence there exist 

2575: solutions of~\eref{eq.nonlin.2}. Such a solution is visualized in 

2576: Figure~\ref{fig.kgb.cmc}.

2577: \begin{figure}

2578: \begin{center}

2579: \bild{\includegraphics[scale=.5]{kgb_min.ps}}

2580: \end{center}

2581: \caption{A periodic branched conformal minimal surface%

2582: % based on a Riemann surface of genus $4$

2583: , visualized by K.~Grosse-Brauckmann}\label{fig.kgb.min}

2584: \end{figure}

2585: 

2586: \begin{figure}

2587: \begin{center}

2588: \bild{\includegraphics[scale=.5]{kgb_cmc.ps}}

2589: \end{center}

2590: \caption{A periodic branched conformal cmc surface%

2591: % based on a Riemann surface of genus $4$

2592: , visualized by K.~Grosse-Brauckmann}\label{fig.kgb.cmc}

2593: \end{figure}

2594: \item Constant  mean curvature immersions of $T^2$ into $\mR^3$, 

2595: in particular Wente tori and twisty tori (Figure~\ref{fig.twisty}) 

2596: also correspond to 

2597: stationary points of $\cF_\qD$. However, as they are not embedded  

2598: \cite{li.yau:82} tells us that $\int H^2 \geq 8\pi$.

2599: On the other hand maximizers and minimizers of $\cF_\qD$ satisfy 

2600: $\int H^2= \lammin^2\leq 4\pi$, hence these tori do \emph{not} 

2601: correspond to maximizers or minimizers.   

2602: \begin{figure}

2603: \begin{center}

2604: \bild{

2605: \includegraphics[scale=1]{twisty-torus4.eps}

2606: }\end{center}

2607: \caption{A twisty torus, a cmc immersed torus in $\mR^3$,

2608: visualized by Nick Schmitt}\label{fig.twisty}

2609: \end{figure}

2610: \end{enumerate}

2611: \end{examples}

2612: 

2613: Similar propositions also hold for immersions into $S^3$ and 

2614: into hyperbolic space $H^3$, see \cite{ammann:habil}. We will only 

2615: specify the case of $S^3=\SU(2)$.

2616: 

2617: \begin{proposition}

2618: Assume that the Riemann spin surface $(M,g,\si)$ satisfies 

2619: \begin{equation}

2620:   \lammin(M,[g],\si)<2\pi.

2621: \end{equation}

2622: Let $a\in(0,\lammin(M,[g],\si)$ be given.

2623: Then there is a left periodic branched

2624: conformal cmc immersion $F:\witi M\to \SU(2)$ based on $(M,g)$. 

2625: The mean curvature is equal to $H=\sqrt{(\lammin(M,[g],\si)/a)^2-1}$, 

2626: and the area of a 

2627: fundamental domain is $a^2$.

2628: The regular homotopy class of $F$ is determined by the spin-structure $\si$.

2629: The indices of all branching points are even, 

2630: and the sum of these indices

2631: is smaller than $2 {\rm genus}(M)$. In particular, 

2632: if $M$ is a torus, there are no branching points.  

2633: \end{proposition}

2634: 

2635: An example where the image of the periodicity map has a finite image in 

2636: $\SU(2)$ is given in Figure~\ref{fig.undu.s}.

2637: 

2638: \begin{figure}

2639: \begin{center}

2640: \bild{

2641: \includegraphics[scale=.3]{torus_undu_sphere.eps}

2642: }\end{center}

2643: \caption{An unduloid in $S^3$, visualized by Nick Schmitt.}\label{fig.undu.s}

2644: \end{figure}

2645: 

2646: 

2647: \begin{remark}

2648: If $(M,g)$ is an analytic Riemannian manifold of dimension $3$, then there

2649: is an analogue of the spinorial Weierstrass representation. This implies 

2650: that solutions $\phi$ of \eref{eq.nonlin.p} 

2651: can be geometrically interpreted as 

2652: a conformal cmc embedding of $M\setminus \phi^{-1}(0)$ 

2653: into a (non-complete) $4$-dimensional Riemannian manifold $(N,h)$.

2654: This manifold $(N,h)$ carries a parallel spinor whose restriction

2655: to $(M,g)$ is again the solution of~\eref{eq.nonlin.p}. The manifold 

2656: $(N,h)$ depends on $\phi$ and is unique up to restriction to subsets 

2657: and coverings. A more detailed exposition of this $3$-dimensional version

2658: is work in progress.

2659: \end{remark}

2660: 

2661: 

2662: \begin{appendix}

2663: 

2664: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2665: \section{Elliptic regularity}\label{sec.regularity}

2666: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2667: 

2668: 

2669: In this section we want to collect some facts about elliptic regularity

2670: for Dirac operators. The proofs of these statements are

2671: analogous to proofs of the corresponding statements for the Laplace operators

2672: as done e.g.\ in \cite{gilbarg.trudinger:77} and \cite{adams:75}.

2673: Details on how to prove most statements of this section

2674: are given also in \cite{ammann:habil}. Either a proof is provided there or 

2675: it is sketched how the statements are reduced to standard theorems.

2676: 

2677: Let $(N,g)$ be a Riemannian manifold, possibly with boundary,

2678: with a spin structure~$\si$. The interior of $N$ is denoted as $N_0$.

2679: The spinor bundle $\Si(N,g,\si)$ is a complex vector bundle

2680: carrying a natural connection and a natural hermitian metric.

2681: 

2682: We now define the Sobolev norms of spinors fields.

2683: 

2684: \begin{definition}[Sobolev spaces]

2685: For any spinor $\psi$ smooth on $N$, 

2686: and $q\in (1,\infty)$, $k\in \mN\cup\{0\}$,

2687: we define the $H^q_k$-norm of $\psi$ as

2688:   $$\|\psi\|_{H^q_k(N,g,\si)}

2689:    = \sum_{l=0}^k\,\Biggl(\int_N\Bigl|\underbrace{\na \ldots \na}_{l-\mbox{times}}\psi\Bigr|^q\,\dvol_g\Biggr)^{1/q}.$$

2690: When the domain of integration is clear, we simply write 

2691: $\|\psi\|_{H^q_k}$ instead of $\|\psi\|_{H^q_k(N,g,\si)}$.

2692: The closure of the space of 

2693: smooth spinors with respect to this norms is denoted as $H^q_k(\Si(N,g,\si))$.

2694: If the underlying structure of Riemannian spin manifold

2695: is clear from the context, we often write shortly 

2696: $H^q_k$,  or in other situations where we want to emphasize the metric

2697: we write  $H^q_k(g)$. We will also write 

2698: $L^q$ for $H^q_0$. 

2699: \end{definition}

2700: 

2701: \begin{theorema}[Interior $L^p$ estimates]\label{theo.int.lp}

2702: Let $(N,g,\si)$ be a compact Riemannian spin 

2703: manifold, possibly \emph{with boundary}. Let $K$ be a compact

2704: subset of $N_0$.

2705: Let $\psi$ be in $H_k^q(\Si(N,g,\si))$ and let $\phi\in H_1^1(\Si(N,g,\si))$ 

2706: be a weak solution (i.e.\ in the sense of distributions) of

2707:   $$D\phi = \psi$$

2708: on $N_0$.

2709: Then $\phi|_K\in H_{k+1}^q(\Si(K,g,\si))$ and

2710:   $$\|\phi\|_{H_{k+1}^q(\Si(K,g,\si))}\leq C\cdot

2711:     \left(\|\psi\|_{H_k^q(\Si(N,g,\si))} + \|\phi\|_{L^q(\Si(N,g,\si))}\right),$$

2712: where $C=C(N,g,\si,K,k,q)$, i.e.\ $C$ depends only on $N,g,\si,K,k$ and $q$.

2713: 

2714: Furthermore, 

2715: if a sequence of metrics $(g_i)_{i\in \mN}$ 

2716: converges in the $C^\infty$-topology

2717: to a Riemannian metric $g$, then the constants $C$ can be chosen 

2718: such that 

2719:   $$\sup_{i\in \mN}  C(N,g_i,\si,K,k)<\infty.$$   

2720: \end{theorema}

2721: %\cite[Theorem~7.63]{adams:75}

2722: 

2723: In order to prove the theorem, it is sufficient to prove 

2724: it in the case that $N$ and $K$ are small concentric geodesic balls, the radius

2725: being controlled in terms of curvature bounds, and bounds on derivatives

2726: of the curvature. As explained in \cite{ammann:habil} 

2727: this can be done in a way analogous to the corresponding statement 

2728: for the Laplacian in \cite{gilbarg.trudinger:77}.

2729: The general statement then can be reduced to the special 

2730: case by covering  (the general) $K$ by finitely many small open balls and an 

2731: associated partition of unity.

2732: 

2733: In the special case that the boundary of $N$ is the empty set,

2734: and that $K=N=N_0$, 

2735: then there is a stronger version, that will be used as well.

2736: 

2737: \begin{theorema}[Global $L^p$ estimates]\label{theo.global.lp}

2738: Let $(N,g,\si)$ be a compact Riemannian spin manifold 

2739: \emph{without boundary} and $\psi\in H_k^q(\Si(N,g,\si))$.

2740: Then any weak solution  $\phi\in H_1^1(\Si(N,g,\si))$ of

2741:   $$D\phi = \psi$$

2742: satisfies $\phi\in H_{k+1}^q(\Si(N,g,\si))$, and there is a constant $C=C(N,g,\si)$

2743: such that

2744:   $$\|\phi\|_{H_{k+1}^q}\leq C 

2745:     \left(\|\psi\|_{H_{k}^q}+\|\pi_{\mathop{\rm ker}(D)}(\phi)\|_{L^q}\right)$$

2746: where $\pi_{\mathop{\rm ker}(D)}$ is the $L^2$-orthogonal projection 

2747: to the kernel of $D$.

2748: \end{theorema}

2749: 

2750: The proof of this theorem is not difficult if one uses some facts 

2751: about pseudodifferential operators, explained for example in 

2752: \cite{taylor:81}. We abbreviate  $\pi:=\pi_{\mathop{\rm ker}(D)}$.

2753: The spectrum of the elliptic operator  

2754: $D+\pi$ is bounded away from $0$,

2755: hence it is invertible and its inverse is a pseudodifferential operator

2756: of degree $-1$. Pseudodifferential operators of degree $-1$

2757: are continuous from $L^q$ to $H_1^q$. Hence, 

2758:   $$\|\phi\|_{H_1^q}=\|(D+\pi)^{-1}(D+\pi)\phi\|_{H_1^q}\leq C\|(D+\pi)\phi\|_{L^q}\leq C\left(\|D\phi\|_{L^q}+\|\pi(\phi)\|_{L^q}\right).$$

2759: This is the statement of the theorem in the case $k=0$, and the statement

2760: for $k>0$ follows for example by using Theorem~\ref{theo.int.lp}.

2761: 

2762: However, a uniformity statement for converging $g_i\to g$ as in 

2763: Theorem~\ref{theo.int.lp} does not hold. If 

2764: $\dim \ker D_g >\limsup_i\dim \ker D_{g_i}$, then one easily shows that

2765: there are eigenspinors $\phi_i$ for $D_{g_i}$ to eigenvalues 

2766: $\la_i\to 0$, $\la_i\neq 0$ with  

2767:   $${\|D\phi_i\|_{L^q(g_i)}+\|\pi_{g_i}(\phi_i)\|_{L^q(g_i)}\over \|\phi_i\|_{H_1^q(g_i)}}=\la_i,$$

2768: which would contradict a uniform version of Theorem~\ref{theo.global.lp}.

2769: 

2770: We also needs some facts about H\"older norms.

2771: 

2772: \begin{definition}[H\"older spaces]

2773: Let $\al\in (0,1]$.

2774: On $C^\infty(\Si(N,g,\si))$ we define the

2775: \emph{H\"older norms}

2776: \begin{eqnarray*}

2777:   \|\phi\|_{C^{0,\al}} & := & \mbox{h\"ol}_\al(\phi)\\

2778:   \|\phi\|_{C^{1,\al}} & := & \|\phi\|_{C^0} + \mbox{h\"ol}_\al(\nabla \phi)\\

2779:   \mbox{h\"ol}_\al(Q)& := & \sup \Bigl\{{|Q(x)-P_\ga Q(y)|\over

2780:    d(x,y)^\al}\;|\; x,y\in M, x\neq y, P_\ga \mbox{ is the }\\

2781:  && \mbox{parallel transport along a shortest geodesic $\ga$ from $x$ to $y$}.\Bigr\}

2782: \end{eqnarray*}

2783: If the shortest geodesic is not unique, then we use the convention that

2784: the supremum runs over all possible choices of shortest geodesics.

2785: 

2786: The completions of $C^\infty(\Si(N,g,\si))$ with respect to the 

2787: H\"older norms $C^{0,\al}$ and $C^{1,\al}$ define the 

2788: \emph{H\"older spaces}

2789: $C^{0,\al}=C^{0,\al}(\Si(N,g,\si))$ and

2790: $C^{1,\al}=C^{1,\al}(\Si(N,g,\si))$.

2791: \end{definition}

2792: 

2793: H\"older norms are important as they admit Schauder estimates. 

2794: 

2795: \begin{theorema}[Schauder estimates]\label{theo.schauder}

2796: Let $(N,g,\si)$ be a compact Riemannian spin manifold, possibly

2797: \emph{with boundary}, and let $K$ be a compact subset of $N_0$. 

2798: Suppose $\psi\in C^{k,\al}(\Si(N,g,\si))$, $k\in \mN\cup\{0\}$.

2799: Then for any weak solution $\phi\in L^1 (\Si(N,g,\si))$ of

2800:   $$D\phi = \psi$$

2801: we have $\phi|_K\in C^{k+1,\al}(\Si(K,g,\si))$ and

2802:   $$\|\phi\|_{C^{k+1,\al}(\Si(K,g,\si))}\leq 

2803:     C \cdot (\|\psi\|_{C^{k,\al}(\Si(N,g,\si))}

2804:     +\|\phi\|_{C^0(\Si(N,g,\si)})$$

2805: where $C=C(N,g,\si,K,\al)$.

2806: 

2807: Furthermore, 

2808: if a sequence of metrics $(g_i)_{i\in \mN}$ 

2809: converges in the $C^\infty$-topology

2810: to a Riemannian metric $g$, then the constants $C$ can be chosen 

2811: such that 

2812:   $$\sup_{i\in \mN}  C(N,g_i,\si,K,k)<\infty.$$   

2813: \end{theorema}

2814: 

2815: As before, one important special case is that $N$ has empty boundary

2816: and $K=N=N_0$.

2817: 

2818: The proof can be done in a way analogous

2819: to the proof of the corresponding statements for the Laplacian in 

2820: \cite{gilbarg.trudinger:77}. Again it is sufficient to prove it for 

2821: small concentric geodesic balls, and to glue them together.

2822: We omit the details. For a proof for concentric geodesic 

2823: balls is provided for example in \cite{ammann:habil}.

2824: % where a version

2825: %with a more general uniformity statement is proven.

2826: 

2827: In local charts one easily reduces the following theorem to the standard

2828: Arcela-Ascoli theorem.

2829: 

2830: \begin{theorema}[Arcela-Ascoli] \label{theo.arc.asc}

2831: Let $(N,g,\si)$ be a Riemannian spin

2832: manifold, possibly \emph{with boundary}. 

2833: For $m\in \mN\cup \{0\}$, the inclusion

2834: $C^{m,\al}(\Si(N,g,\si))\to C^m(\Si(N,g,\si))$ is compact, i.e.\ 

2835: a bounded sequence in $C^{m,\al}(\Si(N,g,\si))$ has a subsequence 

2836: convergent in $C^m(\Si(N,g,\si))$.

2837: \end{theorema}

2838: 

2839: 

2840: Sobolev and H\"older spaces are related by a several embedding theorems,

2841: some of them are compact, others are only bounded. We summarize the embeddings

2842: that are needed in the article.

2843: %We will only need the Sobolev embeddings and the Rellich-Kondrakov theorem 

2844: %in the global version, i.e.\ for compact manifolds \emph{without boundary}.

2845: 

2846: \begin{theorema}[Embedding theorems]\label{theo.sobo}\

2847: Let $k,s\in \mN\cup\{0\}$, $k\geq s$ and $q,r\in (1,\infty)$.

2848: Let $(M,g,\si)$ be a compact Riemannian spin manifold \emph{without boundary}.

2849: All spaces of functions are defined on sections of $\Si(M,g,\si)$.

2850: \begin{enumerate}[{\rm (a)}]

2851: \item {\rm (Sobolev embedding theorem I).} If

2852: %\begin{equation}\label{eq.sob.a}

2853:  $$\phantom{\mathrm{(A.1)}}\qquad\qquad\qquad\qquad {1\over r} - {s\over n}\geq {1\over q} - {k\over n},\qquad\qquad\qquad\qquad \mathrm{(A.1)}$$

2854: %\end{equation}

2855: then $H_k^q$ is continuously embedded into $H_s^r$.

2856: \item {\rm (Rellich-Kondrakov theorem).}

2857: If strict inequality holds in {\rm (A.1)} %\eref{eq.sob.a}

2858: and if $k>s$, then the inclusion

2859: $H_k^q\embed H_s^r$ is a

2860: compact map.

2861: \item  {\rm (Sobolev embedding theorem II).}

2862: Suppose $0<\al <1$, $m\in\{0,1\}$ and

2863:   $${1\over q}\leq {k-m-\al \over n}.$$

2864: Then $H_k^q$ is continuously embedded into $C^{m,\al}$.

2865: \end{enumerate}

2866: \end{theorema}

2867: 

2868: 

2869: \komment{

2870: +++++++++++++++++++++

2871: 

2872: Let $(M,g)$ be a Riemannian manifold with a spin structure $\si$.

2873: 

2874: \begin{definition}[Sobolev spaces]

2875: For any smooth spinor $\psi$, and $q\in (1,\infty)$, $k\in \mN\cup\{0\}$,

2876: we define the $H^q_k$-norm of $\psi$ as

2877:   $$\|\psi\|_{H^q_k}= \|\underbrace{\na \ldots \na}_{k-\mbox{times}}\psi\|_{L^q}.$$

2878: \end{definition}

2879: 

2880: Obviously, the $H_q^k$-norms for different connections are equivalent.

2881: 

2882: If $M$ is compact, an alternative way to introduce

2883: Sobolev norms on spinors is by setting

2884:   $$\big\|\psi\big\|_{\witi H^q_k}=\Big\|\,|D|^k\psi\Big\|_{L^q} + \|\pi\psi\|_1,$$ where $\pi$ is the $L^2$-orthogonal projection to the kernel of $D$,

2885: where $\|\,\cdot\,\|_1$ is an arbitrary norm on the kernel, and

2886: where $|D|^k$ should be understood in the spectral sense, i.e.\

2887: if $\phi$ is an eigenspinor of $D$ to the eigenvalue $\la$,

2888: then $|D|^k\phi= |\la|^k$. Such powers of differential operators

2889: are well-understood because $M$ is compact (see \cite{seeley:67,taylor:81}).

2890: One consequence of the properties of such operators is that the norms

2891: $\ti H^q_k$ are equivalent to the $H^q_k$-norms.

2892: The definition of the $\witi H^q_k$-norms

2893: has the advantage that it extends to arbitrary $k\in \mR$.

2894: \komment{

2895: The definition of Sobolev-norms with $k\not\in \mZ$

2896: on non-compact manifolds can be easily extended, but we omit their definition

2897: as they will not be needed.

2898: }

2899: 

2900: We define the Sobolev space $H^q_k(\Si M)=H^q_k$ as

2901: the completion of the smooth spinors with respect to this norm.

2902: 

2903: 

2904: 

2905: \begin{definition}[H\"older spaces]

2906: For $\al\in (0,1]$, the \emph{H\"older-spaces}

2907: $C^{0,\al}(\Si M)$ and $C^{1,\al}(\Si M)$ are defined to be the completions of

2908: $C^\infty(\Si M)$ with respect to the

2909: \emph{H\"older norms}

2910: \begin{eqnarray*}

2911:   \|\phi\|_{C^{0,\al}} & := & \mbox{h\"ol}_\al(\phi)\\

2912:   \|\phi\|_{C^{1,\al}} & := & \|\phi\|_{\rm sup} + \mbox{h\"ol}_\al(\nabla \phi)\\

2913:   \mbox{h\"ol}_\al(Q)& := & \sup \Bigl\{{|Q(x)-P_\ga Q(y)|\over

2914:    d(x,y)^\al}\;|\; x,y\in M, x\neq y, P_\ga \mbox{ is the }\\

2915:  && \mbox{parallel transport along a shortest geodesic $\ga$ from $x$ to $y$}.\Bigr\}

2916: \end{eqnarray*}

2917: \end{definition}

2918: 

2919: %

2920: %In the following we assume for simplicity that either $s\in \mZ$ or $r=2$.

2921: %

2922: From elliptic theory, we know the following statements \cite{ammann:habil}.

2923: 

2924: \begin{theorema}[Sobolev embedding theorem]\label{theo.sob}\

2925: Let $k,s\in \mR$, $k\geq s$ and $q,r\in (1,\infty)$.

2926: \begin{enumerate}[{\rm (a)}]

2927: \item If

2928: %\begin{equation}\label{eq.sob.a}

2929:  $$\phantom{\mathrm{(A.1)}}\qquad\qquad\qquad\qquad {1\over r} - {s\over n}\geq {1\over q} - {k\over n},\qquad\qquad\qquad\qquad \mathrm{(A.1)}$$

2930: %\end{equation}

2931: then $H_k^q(\Si M)$ is continuously embedded into $H_s^r(\Si M)$.

2932: \item {\rm (Rellich-Kondrakov theorem).}

2933: If strict inequality holds in (A.1)%\eref{eq.sob.a}

2934: , then the inclusion

2935: $H_k^q(\Si M)\embed H_s^r(\Si M)$ is a

2936: compact map.

2937: \item Suppose $0<\al <1$, $m\in\{0,1\}$ and

2938:   $${1\over q}\leq {k-m-\al \over n}.$$

2939: Then $H_k^q(\Si M)$ is continuously embedded into $C^{m,\al}(\Si M)$.

2940: \item (Arcela-Ascoli) for $m\in \mN\cup \{0\}$, the inclusion

2941: $C^{m,\al}\to C^m$ is compact, i.e.\ 

2942: a bounded sequence in $C^{m,\al}$ has a subsequence convergent in $C^m$.

2943: \end{enumerate}

2944: \end{theorema}

2945: 

2946: \begin{theorema}[Interior $L^p$ estimates]\label{theo.int.lp}

2947: Let $\Omega$ be open in $\mR^n$, equipped with a Riemannian metric and

2948: the spin structure $\si$ induced from $\mR^n$, and

2949: $K\subset \Omega$ compact.

2950: We assume

2951: $g(v,v)\geq \zeta g_{\rm eucl}(v,v)$ for all $v\in T\Omega$ and

2952: $g\in C^{k+1}$ with $\|g\|_{C^{k+1}}\leq Z$. Let $D$ be the Dirac operator on

2953: $(\Omega,g,\si)$.

2954: Let $\psi$ be in $H_k^q(\Si \Omega)$ and let $\phi$ be a weak solution of

2955:   $$D\phi = \psi$$

2956: on $\Omega$.

2957: Then $\phi\in H_{k+1}^q(\Si K)$ and

2958:   $$\|\phi\|_{H_{k+1}^q(\Si K)}\leq C\cdot

2959:     \left(\|\psi\|_{H_k^q(\Si \Omega)} + \|\phi\|_{L^q(\Si \Omega)}\right),$$

2960: where $C=C(\zeta,Z,\Omega,K)$.

2961: \end{theorema}

2962: %\cite[Theorem~7.63]{adams:75}

2963: 

2964: 

2965: \begin{theorema}[Interior Schauder estimates]\label{theo.int.schauder}

2966: Let $\Omega$ be open in $\mR^n$, equipped with a Riemannian metric and

2967: the spin structure induced from $\mR^n$, and $K\subset \Omega$ compact.

2968: Let $\psi$ be a $C^{0,\al}$-spinor on $\Omega$.

2969: We assume

2970: $g(v,v)\geq \zeta g_{\rm eucl}(v,v)$ for all $v\in T\Omega$ and

2971: $\|g\|_{C^{k+1,\al}(\Omega)}\leq Z$.

2972: Then for any $C^1$-solution $\phi$ of

2973:   $$D\phi = \psi$$

2974: we have $\phi\in C^{k+1,\al}(K)$ and

2975:   $$\|\phi\|_{C^{k+1,\al}(K)}\leq C \cdot (\|\psi\|_{C^{k,\al}(\Omega)}+\|\phi\|_{C^0(\Omega)})$$

2976: where $C$ only depends on $n$, $\al$, $\diam (\Omega)$,

2977: $\dist(K,\partial \Omega)$, $\zeta$ and $Z$.

2978: \end{theorema}

2979: 

2980: By gluing together the local versions via charts, we obtain

2981: global $L^p$ and Schauder estimates.

2982: 

2983: \begin{theorema}[Global $L^p$ estimates]\label{theo.glob.lp}

2984: Let $(M,g,\si)$ be a compact Riemannian spin manifold

2985: and $\psi\in H_k^q(\Si M)$.

2986: Then any  weak solution of

2987:   $$D\phi = \psi$$

2988: satisfies $\phi\in H_{k+1}^q(\Si M)$, and there is a constant $C=C(M,g,\si)$

2989: such that

2990:   $$\|\phi\|_{H_{k+1}^q}\leq C \left(\|\psi\|_{H_{k}^q}+\|\phi\|_{L^q}\right).$$

2991: \end{theorema}

2992: \begin{theorema}

2993: Let $(M,g,\si)$ be a compact Riemannian spin manifold, and 

2994: let $\phi\in L^1(\Si M)$ with $D\phi\in H_k^q(\Si M)$.

2995: Then $\phi\in H_{k+1}^q(\Si M)$, 

2996: and there is a constant $C=C(M,g,\si)$ with 

2997:   $$\|\phi\|_{H_{k+1}^q}\leq C 

2998:     \left(\|D\phi\|_{H_{k}^q}+\|\pi_{\mathop{\rm ker}(D)}(\phi)\|_{L^q}\right)$$

2999: where $\pi_{\mathop{\rm ker}(D)}$ is the $L^2$-orthogonal projection 

3000: to the kernel of $D$.

3001: \end{theorema}

3002: 

3003: \begin{theorema}[Global Schauder estimates]\label{theo.glob.schauder}

3004: Let $(M,g,\si)$ be a compact Riemannian spin

3005: manifold and $\psi\in C^{k,\al}(\Si M)$.

3006: Then any solution of

3007:   $$D\phi = \psi,$$

3008: satisfies $\phi\in C^{k+1,\al}(\Si M)$,

3009: and there is a constant $C=C(M,g,\si)$ such that

3010:   $$\|\phi\|_{C^{k+1,\al}}\leq C \left(\|\psi\|_{C^{k,\al}}+\|\phi\|_{C^{0}}\right).$$

3011: \end{theorema}

3012: }

3013: 

3014: \komment{

3015: Similar norms and completions

3016: can be defined on $C^\infty(M)$ using the Laplacian

3017: on functions. The optimal constant in the embedding

3018: $H_1^2\embed L^{p_Y}$, $p_Y= 2n/(n+2)$ has been determined by Hebey.

3019: 

3020: 

3021: We recall an untechnical version of the theorem, which gives

3022: sufficient control for our application:

3023: %

3024: %We recall the following theorem which is a part of Hebey's research on the

3025: %optimal constant problem for Sobolev embeddings.

3026: %We set $p_Y=2n/(n-2)$, the critical exponent for the conformal Laplacian.

3027: 

3028: \begin{theorema}[{\cite[Theorem 4.12]{hebey:96}}]\label{theo.hebey}

3029: Let $(M,g)$ be a complete Riemannian $n$-manifold such that the Riemann

3030: curvature tensor fulfills $|R|\leq \La_1$ and $|\na R|\leq \La_2$ for some

3031: $\La_1, \La_2>0$ and $\inj(M,g)\geq i$ for some $i>0$. Let $\pY=2n/(n-2)$

3032: Then there exists a positive constant $B=B(n,\La_1,\La_2,i)$

3033: depending only on $n$, $\La_1$, $\La_2$ and $i$, such that for any $u\in H_1^2(M)$,

3034:   $$\|u\|^2_{L^{\pY}}\leq {4\over n (n-2)\, \om_n^{2/n}}\, \|du\|_{L^2}^2 +

3035:     B \,\|u\|^2_{L^2}.$$

3036: \end{theorema}

3037: }

3038: 

3039: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

3040: \section{Some facts about H\"older spaces}\label{sec.scha.power}

3041: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

3042: 

3043: In this section we want to include some proofs of probably 

3044: well-known statements about H\"older spaces.

3045: 

3046: \begin{lemmaa}

3047: Let $V$ be a Euclidean vector space, $\al\in (0,1)$, $m\in \mN$. 

3048: Then the map $V\to \bigotimes_m V$, 

3049:   $$T:x\mapsto |x|^{\al-m}\underbrace{x\otimes\cdots\otimes x}_{\mbox{$m$ times}}$$

3050: is $C^\al$.

3051: \end{lemmaa}

3052: 

3053: \proof{}

3054: Let $x,y\in V$. At first, we suppose that

3055: $\|x-y\|<\de$ and $\|x\|\geq \de$.

3056: Then we calculate

3057: %\begin{equation}\label{est.hoeeins}

3058: $$

3059: \|x\|\,\Big\|{x\over \|x\|}-{y\over \|y\|}\Big\|=\Big\|x-{\|x\|\over \|y\|}y\Big\|\leq \|x-y\|+\Big|1- {\|x\|\over \|y\|}\Big|\,\|y\|\leq 2\de

3060: $$

3061: %\end{equation}

3062: Hence 

3063: %\begin{equation}\label{est.hoezwei}

3064: $$

3065:    \Big\|{x\over \|x\|}\otimes \cdots \otimes {x\over \|x\|} 

3066:     - {y\over \|y\|}\otimes \cdots \otimes {y\over \|y\|}\Big\|\leq {2\de m\over \|x\|}

3067: $$

3068: %\end{equation}

3069: This implies 

3070:   $$|T(x)-T(y)|\leq 

3071:     \Big|\|x\|^\al -\|y\|^\al\Big|+ \|x\|^\al \,{2\de m\over \|x\|}\leq \|x-y\|^\al+ \|x\|^{\al-1}2\de m\leq (2m+1)\de^\al.$$

3072: 

3073: Now suppose that

3074: $\|x-y\|<\de$ and $\|x\|< \de$. Then 

3075:   $$|T(x)-T(y)|\leq \|x\|^\al + \|y\|^\al\leq 3\de^\al.$$

3076: Hence, we obtain $|T(x)-T(y)|\leq (2m+1)\|x-y\|$ for all $x,y\in V$, and hence $T$ is $C^\al$.

3077: \qed

3078: 

3079: 

3080: \begin{propositiona}

3081: If $V$ is a vector bundle over a compact Riemannian manifold, 

3082: and if $\phi$ is a

3083: $C^{1,\al}$-section of $V$, $\al\in (0,1)$ then 

3084: $\psi=|\phi|^\beta\phi$, $\be>0$ is a $C^{1,\gamma}$-section

3085: for $\gamma:=\min\{\al,\be\}$.

3086: \end{propositiona}

3087: 

3088: \proof{}

3089: The section $\phi$ is obviously Lipschitz, hence $|\phi|^\be$ is $C^\be$.

3090: We have to show that 

3091: \begin{eqnarray*} 

3092:   \na\psi&=&|\phi|^\be\na\phi + \beta\<\na \phi,\phi\>|\phi|^{\be-2}\phi

3093: \end{eqnarray*}

3094: is $C^\ga$. The first summand is a product of $C^\be$ and $C^\al$, 

3095: hence $C^\gamma$. According to the previous lemma, 

3096: $|\phi|^{\be-2}\phi\otimes \phi$ is $C^\be$, hence the second summand

3097: is a product of $C^\al$ and $C^\be$, hence also $C^\ga$.

3098: \qed  

3099: 

3100: 

3101: \end{appendix}

3102: 

3103: 

3104: 

3105: 

3106: 

3107: %\bibliographystyle{amsalpha}

3108: %\bibliographystyle{amsbernd}

3109: %\bibliography{literatur}

3110: 

3111: \providecommand{\bysame}{\leavevmode\hbox to3em{\hrulefill}\thinspace}

3112: \providecommand{\MR}{\relax\ifhmode\unskip\space\fi MR }

3113: % \MRhref is called by the amsart/book/proc definition of \MR.

3114: \providecommand{\MRhref}[2]{%

3115:   \href{http://www.ams.org/mathscinet-getitem?mr=#1}{#2}

3116: }

3117: \providecommand{\href}[2]{#2}

3118: \begin{thebibliography}{AHM03b}

3119: 

3120: \bibitem[Ada75]{adams:75}

3121: R.~Adams, \emph{{S}obolev {S}paces}, Academic Press, 1975.

3122: 

3123: \bibitem[AH05]{ammann.humbert:gluing}

3124: B.~Ammann and E.~Humbert, \emph{A gluing formula for the spinorial

3125:   $\tau$-invariant (preliminary title)}, Work in progress, 2005.

3126: 

3127: \bibitem[AHM03a]{ammann.humbert.morel:p03b}

3128: B.~Ammann, E.~Humbert, and B.~Morel, \emph{Mass endomorphism and spinorial

3129:   {Y}amabe type problems}, to appear in {C}omm. {A}nal. {G}eom., 2003.

3130: 

3131: \bibitem[AHM03b]{ammann.humbert.morel:p03av2}

3132: \bysame, \emph{A spinorial analogue of {A}ubin's inequality}, {A}r{X}iv

3133:   math.{DG}/0308107, 2003.

3134: 

3135: \bibitem[Amm98]{ammann:diss}

3136: B.~Ammann, \emph{{S}pin-{S}trukturen und das {S}pektrum des

3137:   {D}irac-{O}perators}, Ph.D. thesis, University of {F}reiburg, {G}ermany,

3138:   1998, Shaker-Verlag Aachen 1998, ISBN 3-8265-4282-7.

3139: 

3140: \bibitem[Amm03a]{ammann:03}

3141: \bysame, \emph{A spin-conformal lower bound of the first positive {D}irac

3142:   eigenvalue}, {D}iff. {G}eom. {A}ppl. \textbf{18} (2003), 21--32.

3143: 

3144: \bibitem[Amm03b]{ammann:habil}

3145: \bysame, \emph{A variational problem in conformal spin geometry},

3146:   Habilitationsschrift, Universit\"at Hamburg, 2003.

3147: 

3148: \bibitem[Aub76]{aubin:76}

3149: T.~Aubin, \emph{{{\'E}quations diff{\'e}rentielles non lin{\'e}aires et

3150:   probl{\`e}me de Yamabe concernant la courbure scalaire.}}, J. Math. Pur.

3151:   Appl., IX. Ser. \textbf{55} (1976), 269--296.

3152: 

3153: \bibitem[B{\"a}r92]{baer:92b}

3154: C.~B{\"a}r, \emph{Lower eigenvalue estimates for {D}irac operators}, Math. Ann.

3155:   \textbf{293} (1992), 39--46.

3156: 

3157: \bibitem[B{\"a}r97]{baer:97}

3158: \bysame, \emph{On nodal sets for {D}irac and {L}aplace operators}, Comm. Math.

3159:   Phys. \textbf{188} (1997), 709--721.

3160: 

3161: \bibitem[B{\"a}r98]{baer:98}

3162: \bysame, \emph{Extrinsic bounds for eigenvalues of the {D}irac operator}, Ann.

3163:   Global Anal. Geom. \textbf{16} (1998), 573--596.

3164: 

3165: \bibitem[Bes87]{besse:87}

3166: A.~L. Besse, \emph{Einstein manifolds}, Ergebnisse der Mathematik und ihrer

3167:   Grenzgebiete, 3.~Folge, no.~10, Springer-Verlag, 1987.

3168: 

3169: \bibitem[BG92]{bourguignon.gauduchon:92}

3170: J.-P. Bourguignon and P.~Gauduchon, \emph{Spineurs, op{\'e}rateurs de {D}irac

3171:   et variations de m{\'e}triques}, Comm. Math. Phys. \textbf{144} (1992),

3172:   581--599.

3173: 

3174: \bibitem[Fri80]{friedrich:80}

3175: T.~Friedrich, \emph{Der erste {E}igenwert des {D}irac-{O}perators einer

3176:   kompakten {R}iemannschen {M}annigfaltigkeit nicht-negativer {K}r{\"u}mmung},

3177:   Math. Nach. \textbf{97} (1980), 117--146.

3178: 

3179: \bibitem[Fri98]{friedrich:98}

3180: \bysame, \emph{{On the spinor representation of surfaces in Euclidean

3181:   $3$-space.}}, J. Geom. Phys. \textbf{28} (1998), 143--157.

3182: 

3183: \bibitem[Fri00]{friedrich:00}

3184: \bysame, \emph{{D}irac {O}perators in {R}iemannian {G}eometry}, {G}raduate

3185:   {S}tudies in {M}athematics 25, AMS, Providence, Rhode Island, 2000.

3186: 

3187: \bibitem[GH05]{grosjean.humbert:p05}

3188: F.~Grosjean and E.~Humbert, \emph{The first eigenvalue of {D}irac and {L}aplace

3189:   operators on surfaces (preliminary title)}, preprint in preparation, 2005.

3190: 

3191: \bibitem[GT77]{gilbarg.trudinger:77}

3192: D.~Gilbarg and N.~Trudinger, \emph{Elliptic partial differential equations of

3193:   second order}, Grundlehren der mathematischen Wissenschaften, no. 224,

3194:   Springer-Verlag, 1977.

3195: 

3196: \bibitem[Hij86]{hijazi:86}

3197: O.~Hijazi, \emph{A conformal lower bound for the smallest eigenvalue of the

3198:   {D}irac operator and {K}illing spinors}, Comm. Math. Phys. \textbf{104}

3199:   (1986), 151--162.

3200: 

3201: \bibitem[Hij01]{hijazi:01}

3202: \bysame, \emph{{Spectral properties of the {D}irac operator and geometrical

3203:   structures.}}, {Ocampo, Hernan (ed.) et al., Geometric methods for quantum

3204:   field theory. Proceedings of the summer school, Villa de Leyva, Colombia,

3205:   July 12-30, 1999. Singapore: World Scientific. 116-169 }, 2001.

3206: 

3207: \bibitem[Hit74]{hitchin:74}

3208: N.~Hitchin, \emph{Harmonic spinors}, Adv. Math. \textbf{14} (1974), 1--55.

3209: 

3210: \bibitem[Kir86]{kirchberg:86}

3211: K.-D. Kirchberg, \emph{An estimation for the first eigenvalue of the {D}irac

3212:   operator on closed {K}\"ahler manifolds of positive scalar curvature}, Ann.

3213:   Global Analysis and Geometry \textbf{4} (1986), 291--325.

3214: 

3215: \bibitem[Kir88]{kirchberg:88}

3216: \bysame, \emph{Compact six-dimensional {K}{\"a}hler spin manifolds of positive

3217:   scalar curvature with the smallest possible first eigenvalue of the {D}irac

3218:   operator}, Math.\ Ann. \textbf{282} (1988), 157--176.

3219: 

3220: \bibitem[KS96]{kusner.schmitt:p96}

3221: R.~Kusner and N.~Schmitt, \emph{The spinor representation of surfaces in

3222:   space}, preprint, http://www.arxiv.org/abs/dg-ga/9610005, 1996.

3223: 

3224: \bibitem[KSW98]{kramer.semmelmann.weingart:98}

3225: W.~Kramer, U.~Semmelmann, and G.~Weingart, \emph{The first eigenvalue of the

3226:   {D}irac operator on quaternionic {K}\"ahler manifolds}, Comm. Math. Phys.

3227:   \textbf{199} (1998), no.~2, 327--349.

3228: 

3229: \bibitem[KSW99]{kramer.semmelmann.weingart:99}

3230: \bysame, \emph{Eigenvalue estimates for the {D}irac operator on quaternionic

3231:   {K}\"ahler manifolds}, Math. Z. \textbf{230} (1999), no.~4, 727--751.

3232: 

3233: \bibitem[Lic63]{lichnerowicz:63}

3234: A.~Lichnerowicz, \emph{Spineurs harmoniques}, C. R. Acad. Sci. Paris

3235:   \textbf{257} (1963), 7--9.

3236: 

3237: \bibitem[LM89]{lawson.michelsohn:89}

3238: H.-B. Lawson and M.-L. Michelsohn, \emph{Spin geometry}, Princeton University

3239:   Press, Princeton, 1989.

3240: 

3241: \bibitem[LP87]{lee.parker:87}

3242: J.~M. Lee and T.~H. Parker, \emph{{The Yamabe problem.}}, Bull. Am. Math. Soc.,

3243:   New Ser. \textbf{17} (1987), 37--91.

3244: 

3245: \bibitem[LY82]{li.yau:82}

3246: P.~Li and S.-T. Yau, \emph{A new conformal invariant and its applications to

3247:   the {W}illmore {C}onjecture and the first eigenvalue of compact surfaces},

3248:   Invent. Math. \textbf{69} (1982), 269--291.

3249: 

3250: \bibitem[Mai97]{maier:97}

3251: S.~Maier, \emph{Generic metrics and connections on spin- and

3252:   spin-$\,^c$-manifolds}, Comm. Math. Phys. \textbf{188} (1997), 407--437.

3253: 

3254: \bibitem[Mor02]{morel:02}

3255: B.~Morel, \emph{Surfaces in {$S^3$} and {$H^3$} via spinors}, preprint, 2002,

3256:   math.{DG}/0204090.

3257: 

3258: \bibitem[Pf{\"a}02]{pfaefflediss}

3259: F.~Pf{\"a}ffle, \emph{{E}igenwertkonvergenz f{\"u}r {D}irac-{O}peratoren},

3260:   Ph.D. thesis, University of {H}amburg, {G}ermany, 2002, Shaker Verlag Aachen

3261:   2003, ISBN 3-8322-1294-9.

3262: 

3263: \bibitem[Pin85]{pinkall:85a}

3264: U.~Pinkall, \emph{Regular homotopy classes of immersed surfaces}, Topology

3265:   \textbf{24} (1985), 421--434.

3266: 

3267: \bibitem[Roe88]{roe:88}

3268: J.~Roe, \emph{Elliptic operators, topology and asymptotic methods}, Pitman

3269:   Research Notes in Mathematics Series, no. 179, Longman, 1988.

3270: 

3271: \bibitem[Sch93]{schmitt:diss}

3272: N.~Schmitt, \emph{Minimal surface with planar embedded ends}, {P}h.{D}.

3273:   dissertation, University of Amherst, 1993.

3274: 

3275: \bibitem[Tay81]{taylor:81}

3276: M.~E. Taylor, \emph{Pseudodifferential operators}, Princeton University Press,

3277:   Princeton, N.J., 1981.

3278: 

3279: \bibitem[Vos99]{voss:dipl}

3280: L.~Voss, \emph{Eigenwerte des {D}irac-{O}perators auf {H}yperfl{\"a}chen},

3281:   {D}iplomarbeit, {H}umboldt {U}niversit\"at zu {B}erlin, 1999.

3282: 

3283: \bibitem[Wei66]{weierstrass:66}

3284: K.~Weierstrass, \emph{{U}ntersuchungen {\"u}ber die {F}l{\"a}chen, deren

3285:   mittlere {K}r{\"u}mmung {\"u}berall gleich {N}ull ist}, available in

3286:   {W}eierstrass, {M}athematische {W}erke. {III}, 39--52, 1866.

3287: 

3288: \end{thebibliography}

3289: 

3290: \pagebreak[2]

3291: 

3292: \vspace{1cm}

3293: Author's address:

3294: \nopagebreak

3295: \vspace{5mm}\\

3296: \parskip0ex

3297: \vtop{

3298: \hsize=7cm\noindent

3299: \obeylines

3300: Bernd Ammann

3301: Institut \'Elie Cartan BP 239              

3302: Universit\'e Henri Poincar\'e, Nancy 1               

3303: 54506 Vandoeuvre-l\`es-Nancy Cedex               

3304: France                           

3305: }

3306: 

3307: 

3308: \vspace{0.5cm}

3309: 

3310: E-Mail:

3311: {\tt bernd.ammann@gmx.de}

3312: 

3313: 

3314: WWW:

3315: {\tt http://www.berndammann.de/uni}

3316: 

3317: \end{document}

3318: