1: \documentclass{amsart}

2: 

3: 

4: %\documentstyle[amsmath,amssymb,amscd,graphicx,psfrag,epsf,amsthm]{article}

5: 

6: 

7: 

8: %\documentclass[10pt,leqno,amssymbols]{amsbook}

9: 

10: %\documentstyle[12pt,twoside]{amsart}

11: %\documentstyle[amstex,12pt,leqno]{book}

12: %\documentstyle[amstex,12pt,leqno]{amsart}

13: %\documentstyle[12pt,leqno]{amsbook}

14: 

15: 

16: \usepackage{amssymb}

17: 

18: 

19: 

20: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

21: 

22: % Definitions for new environments

23: % theorem style plain --- default

24: %\newtheorem{thm}{Theorem}[chapter]

25: \newtheorem{thm}{Theorem}%[section]

26: \newtheorem{mainthm}[thm]{Main Theorem}

27: \newtheorem{question}[thm]{Question}

28: \newtheorem{lem}[thm]{Lemma}

29: \newtheorem{cor}[thm]{Corollary}

30: \newtheorem{cors}[thm]{Corollaries}

31: \newtheorem{prop}[thm]{Proposition}

32: \newtheorem{crit}[thm]{Criterion}

33: \newtheorem{conj}[thm]{Conjecture}

34: \newtheorem{principle}[thm]{Principle} %!!!!!!!!!!!!!!!!!!!!!!

35: \newtheorem{complem}[thm]{Complement}%!!!!!!!!!!!!!!!!!!!!!!

36: \newtheorem{pitfall}[thm]{Pitfall}%!!!!!!!!!!!!!!!!!!!!!!

37:    

38: \theoremstyle{definition}

39: \newtheorem{defn}[thm]{Definition}

40: \newtheorem{condition}[thm]{Condition}

41: \newtheorem{say}[thm]{}

42: \newtheorem{exmp}[thm]{Example}

43: \newtheorem{hint}[thm]{Hint}

44: \newtheorem{exrc}[thm]{Exercise}

45: \newtheorem{prob}[thm]{Problem}

46: \newtheorem{const}[thm]{Construction}   %!!!!!!!!!!!!!!!!

47: \newtheorem{ques}[thm]{Question}    %!!!!!!!!!!!!!!!!!!!!

48: \newtheorem{alg}[thm]{Algorithm}

49: \newtheorem{rem}[thm]{Remark}          

50: \newtheorem{aside}[thm]{Aside}          

51: %\renewcommand{\theremark}{} 

52: \newtheorem{note}[thm]{Note}            %\renewcommand{\thenote}{} 

53: \newtheorem{summ}[thm]{Summary}         %\renewcommand{\thesumm}{} 

54: \newtheorem{ack}{Acknowledgments}       \renewcommand{\theack}{} 

55: \newtheorem{notation}[thm]{Notation}   

56: \newtheorem{warning}[thm]{Warning}  

57: \newtheorem{defn-thm}[thm]{Definition--Theorem}  %!!!!!!!!!!!!!!!!!!!!!!!!

58: \newtheorem{defn-lem}[thm]{Definition--Lemma}  %!!!!!!!!!!!!!!!!!!!!!!!!

59: \newtheorem{convention}[thm]{Convention}  %!!!!!!!!!!!!!!!!!!!!!!!!!!!

60: \newtheorem{keyidea}[thm]{Key idea} 

61: \newtheorem{expl}[thm]{Explanation}   

62: \newtheorem{meth}[thm]{Method}   

63: 

64: 

65: \theoremstyle{remark}

66: \newtheorem{claim}[thm]{Claim}  

67: \newtheorem{case}{Case} 

68: \newtheorem{subcase}{Subcase}   

69: \newtheorem{step}{Step}

70: \newtheorem{approach}{Approach}

71: %\newtheorem{principle}{Principle}

72: \newtheorem{fact}{Fact}

73: \newtheorem{subsay}{}

74: 

75: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

76: \setcounter{section}{0}

77:  %\documentstyle[12pt]{book}

78: %\setlength{\topmargin}{0in}

79: %\setlength{\headsep}{0in}

80: %\setlength{\headheight}{0in}

81: %\setlength{\textheight}{8.25in}

82: %\setlength{\parindent}{0in}

83: %\pagestyle{empty}

84: 

85: 

86: \let \cedilla =\c

87: \renewcommand{\c}[0]{{\mathbb C}}  

88: \let \crossedo =\o

89: \renewcommand{\o}[0]{{\mathcal O}} 

90: %\renewcommand{\o}[0]{{\mathcal O}} 

91: \newcommand{\z}[0]{{\mathbb Z}}

92: \newcommand{\n}[0]{{\mathbb N}}

93: \let \ringaccent=\r  %%% \r  shorthand for 'ring accent'

94: \renewcommand{\r}[0]{{\mathbb R}} 

95: 

96: \renewcommand{\a}[0]{{\mathbb A}} 

97: 

98: \newcommand{\h}[0]{{\mathbb H}}

99: \newcommand{\p}[0]{{\mathbb P}}

100: \newcommand{\f}[0]{{\mathbb F}}

101: \newcommand{\q}[0]{{\mathbb Q}}

102: \newcommand{\map}[0]{\dasharrow}

103: \newcommand{\qtq}[1]{\quad\mbox{#1}\quad}

104: \newcommand{\spec}[0]{\operatorname{Spec}}

105: \newcommand{\pic}[0]{\operatorname{Pic}}

106: \newcommand{\gal}[0]{\operatorname{Gal}}

107: \newcommand{\cont}[0]{\operatorname{cont}}

108: \newcommand{\rank}[0]{\operatorname{rank}}

109: \newcommand{\mult}[0]{\operatorname{mult}}

110: \newcommand{\discrep}[0]{\operatorname{discrep}}

111: \newcommand{\totaldiscrep}[0]{\operatorname{totaldiscrep}}

112: 

113: \newcommand{\supp}[0]{\operatorname{Supp}}    

114: \newcommand{\red}[0]{\operatorname{red}}    

115: \newcommand{\codim}[0]{\operatorname{codim}}    

116: \newcommand{\im}[0]{\operatorname{im}}    

117: \newcommand{\proj}[0]{\operatorname{Proj}}    

118: \newcommand{\wt}[0]{\operatorname{wt}}    

119: \newcommand{\socle}[0]{\operatorname{socle}}    

120: \newcommand{\coker}[0]{\operatorname{coker}}    

121: \newcommand{\ext}[0]{\operatorname{Ext}}    

122: \newcommand{\Hom}[0]{\operatorname{Hom}}    

123: \newcommand{\trace}[0]{\operatorname{Trace}}  

124: \newcommand{\cent}[0]{\operatorname{center}}

125: \newcommand{\bs}[0]{\operatorname{Bs}}

126: \newcommand{\aut}[0]{\operatorname{Aut}}    

127: \newcommand{\specan}[0]{\operatorname{Specan}}    

128: \newcommand{\inter}[0]{\operatorname{Int}}    

129: \newcommand{\sing}[0]{\operatorname{Sing}}    

130: \newcommand{\ex}[0]{\operatorname{Ex}}    

131: \newcommand{\diff}[0]{\operatorname{Diff}}  

132: \newcommand{\tors}[0]{\operatorname{tors}}  

133: \newcommand{\tor}[0]{\operatorname{Tor}}  

134: \newcommand{\res}[0]{\operatorname{\mathcal R}} 

135: \newcommand{\disc}[0]{\operatorname{Discr}} 

136: \newcommand{\jac}[0]{\operatorname{Jac}} 

137: \newcommand{\chow}[0]{\operatorname{Chow}}

138: \newcommand{\chr}[0]{\operatorname{char}}

139: \newcommand{\comp}[0]{\operatorname{Comp}}

140: \newcommand{\br}[0]{\operatorname{Br}}

141: \newcommand{\var}[0]{\operatorname{Var}}

142: \newcommand{\hilb}[0]{\operatorname{Hilb}}

143: \newcommand{\weil}[0]{\operatorname{Weil}}

144: \newcommand{\weilf}[0]{\operatorname{Weil_{tf}}}

145: \newcommand{\cl}[0]{\operatorname{Cl}}

146: \newcommand{\gr}[0]{\operatorname{gr}}

147: \newcommand{\len}[0]{\operatorname{length}}

148: \newcommand{\stab}[0]{\operatorname{stab}}

149: 

150: %\renewcommand{\conj}[0]{\operatorname{conj}} 

151: \newcommand{\ric}[0]{\operatorname{Ricci}}

152: 

153: \newcommand{\nec}[1]{\overline{NE}({#1})}

154: 

155: \newcommand{\rup}[1]{\lceil{#1}\rceil}

156: \newcommand{\rdown}[1]{\lfloor{#1}\rfloor}

157: 

158: \newcommand{\onto}[0]{\twoheadrightarrow}

159: 

160: \newcommand{\lcm}[0]{\operatorname{lcm}}

161: \newcommand{\wdeg}[0]{\operatorname{wdeg}}

162: \newcommand{\obs}[0]{\operatorname{Obstr}}

163: \newcommand{\ann}[0]{\operatorname{Ann}}

164: \newcommand{\diag}[0]{\operatorname{diag}}

165: \newcommand{\ord}[0]{\operatorname{ord}}

166: \newcommand{\mord}[0]{\operatorname{max-ord}}

167: \newcommand{\der}[0]{\operatorname{Der}}

168: \newcommand{\cosupp}[0]{\operatorname{cosupp}}

169: 

170: 

171: 

172: 

173: 

174: 

175: %\UseAMSsymbols

176: 

177: %\DeclareMathSymbol\subsetneq 2328

178: %\DeclareMathSymbol\onto 1310

179: 

180: %\DeclareMathSymbol\twoheadrightarrow 1310

181: %\DeclareMathSymbol\rtimes 226F

182: %\DeclareMathSymbol\ltimes 226E

183: 

184: \def\into{\DOTSB\lhook\joinrel\to}

185: 

186: %\makeindex

187: 

188: 

189: \begin{document}

190: \bibliographystyle{amsalpha}

191: 

192: \title[Bogomolov--Miyaoka--Yau inequality]{Is there a topological\\

193: Bogomolov--Miyaoka--Yau inequality?}

194: \author{J\'anos Koll\'ar}

195: 

196: \today

197: 

198: \maketitle

199: 

200:     

201: %\tableofcontents

202: 

203: 

204: Let $S$ be a smooth, complex,  projective, minimal  surface

205: of general type. The Bogomolov--Miyaoka--Yau inequality states

206: that $c_1(S)^2\leq 3 c_2(S)$ \cite{bog-4, rei-bog, miy-3, yau}.

207: In this note I want to address the

208: following question:

209: $$

210: \mbox{\it Is there a  topological analog of the

211: Bogomolov--Miyaoka--Yau inequality?}

212: $$

213: \noindent The $11/8$-conjecture 

214: \cite{matsu, furuta} can be viewed as such, but 

215: in Section 1 I write down another

216: possible variant. 

217:  Section 2 explores its relationship with the

218: Montgomery--Yang problem on  differentiable circle actions on

219: $S^5$ and Section 3 examines its connection

220: with the  H--cobordism group of 3--manifolds.

221: Related examples  and questions on algebraic surfaces 

222: are discussed in Section 4.

223: The last section studies the remarkable series of

224: hypersurfaces

225: $$

226: (x_1^{a_1}x_2+x_2^{a_2}x_3+ \cdots + x_{n-1}^{a_{n-1}}x_n+ x_n^{a_n}x_1=0)

227: $$

228: in weighted projective spaces. 

229: 

230: The aim of this note is two fold.

231: On the one hand, I would like to call attention to

232: several questions about algebraic surfaces with quotient

233: singularities that have interesting connections with

234: the topology of 3-- or 5--manifolds.

235: The study of algebraic orbifolds led to many interesting

236: examples in  topology and differential geometry 

237: (see, for instance, \cite{briesk, or-wa} or the 

238:  recent papers \cite{bgk, e-surf}),

239:  but there should be many

240: more connections.

241: 

242: On the other hand, more speculatively, I hope that

243: methods developed around the  Bogomolov--Miyaoka--Yau inequality

244: can be adapted to the topological setting,  leading to

245: progress on the questions mentioned in Sections 2 and 3.

246: 

247: \section{The orbifold Bogomolov--Miyaoka--Yau inequality}

248: 

249:  

250: The Bogomolov--Miyaoka--Yau inequality can be

251: generalized to orbifolds. These are normal projective surfaces

252: whose singularities are locally analytically isomorphic to

253:  quotient singularities $\c^2/G$ where $G\subset GL(2,\c)$

254: is a finite group whose action is fixed point free outside the origin.

255: Such a surface has finitely many singular points,

256: and locally at each of them $S$ is topologically the cone

257: over a 3--manifold $S^3/G$ where $G\subset U(2,\c)$

258: is a subgroup acting without fixed points.

259: This 3--manifold is called the

260: {\it link} of $s\in S$ and it is denoted by $L_s$.

261: (The version where a surface is allowed to have orbifold

262: structure in codimension one is also interesting \cite{bgk, e-surf}, but

263: it will not be considered here.)

264: 

265: An easy way to get such an orbifold is to take the quotient of  a smooth 

266: projective surface $X$ by a finite group $G$ acting on $X$

267: with only isolated fixed points.

268: The most interesting examples are, however, those that do not

269: arise as a global quotient. 

270:   For a complex surface, $c_2(S)$ is

271: the same as the topological Euler characteristic.

272: Following Thurston,

273: we introduce the

274: {\it orbifold Euler characteristic}

275: $$

276: e_{orb}(S):=e(S)-\sum_{s\in \sing S} \Bigl( 1-\frac1{|\pi_1(L_s)|}\Bigr).

277: $$

278: The orbifold version of the

279: Bogomolov--Miyaoka--Yau inequality is the following,

280: developed in the series of papers

281: \cite{sak, miy, kns, megy}. 

282: 

283: \begin{thm}\label{obmy.thm}  Let $S$ be a normal projective surface with

284: quotient singularities such that $-c_1(S)$

285: is ample (or at least nef). Then

286: $$

287: c_1(S)^2\leq 3e_{orb}(S).

288: \eqno{(\ref{obmy.thm}.1)}

289: $$

290: \end{thm}

291: 

292: Since  $-c_1(S)$

293: is  nef,  $c_1(S)^2\geq 0$, thus we also get

294: the following weaker inequality,

295: which also holds when $c_1(S)$ is nef by \cite{ke-mc}

296: $$

297: 0\leq e_{orb}(S).

298: \eqno{(\ref{obmy.thm}.2)}

299: $$

300: While (\ref{obmy.thm}.2) gives nothing interesting 

301: for smooth surfaces, it has very interesting

302: consequences for singular surfaces.

303: 

304:  Let $S$ be a normal projective surface with

305: quotient singularities whose 

306:  second Betti number  is 1. Then either $\pm c_1(S)$

307: is ample or $c_1(S)=0$. Thus (\ref{obmy.thm}.2) applies

308: and so $0\leq e_{orb}(S)$. 

309: It is easy to see (by looking at the Albanese map)

310: that $b_1(S)=0$, thus $e(S)=3$ and so $S$ is a

311: rational homology $\c\p^2$. Thus we get to following 

312:  expanded version

313: of  (\ref{obmy.thm}.2):

314: $$

315: \sum_{s\in \sing S} \Bigl( 1-\frac1{|\pi_1(L_s)|}\Bigr)\leq  3.

316: \eqno{(\ref{obmy.thm}.3)}

317: $$

318: In particular, $S$ has at most 6 singular points.

319: \medskip

320: 

321: Let us now make the bold (or foolish)

322: guess that the inequality

323: (\ref{obmy.thm}.3) is topological in nature.

324: It may be more convenient to formulate the conjecture

325: for smooth, compact 4--manifolds $M$ whose

326: boundary components are spherical, that is, their universal cover is $S^3$.

327: One can then attach cones to each boundary component

328: to get a 4--dimensional orbifold $S$.

329: We are mainly interested in the cases when $S$ is a

330:  homology $\c\p^2$. Correspondingly,

331: $H_1(M,\z)=0$ and $H_2(M,\z)\cong \z$.

332: 

333: 

334: \begin{conj}[Smooth Bogomolov--Miyaoka--Yau inequality]

335: \label{tbmy.thm}

336: Let $M^4$ be a smooth, compact 4--manifold with spherical

337: boundary components $\partial M^4=\cup_i L_i$. Assume that

338: $H_1(M,\z)=0$ and $H_2(M,\z)\cong \z$. Then

339: $$

340: \sum_i \Bigl( 1-\frac1{|\pi_1(L_i)|}\Bigr)< 3.

341: \eqno{(\ref{tbmy.thm}.1)}

342: $$

343: In particular, $M$ has at most 5 boundary components.

344: \end{conj}

345: 

346: \begin{say}[The origins of the conjecture] While, to the best of my knowledge,

347: the above conjecture is new, various forms of it have been implicit

348: in the works of several authors.

349: 

350: Controlling the singularities of a variety by global invariants has

351: been one of the  aims of the papers

352: \cite{miy,  megy, ke-mc}. It is quite natural to go from

353: an algebraic formulation to a purely topological one as above.

354: 

355: On the topological side, the Montgomery--Yang problem on circle actions

356: on 5--manifold was explicitly studied as a problem  on 4--dimensional

357: orbifolds in \cite{fi-st}. As we discuss in Section 2, the

358: relationship with (\ref{tbmy.thm}) is close but the two problems are

359: not identical.

360: 

361: The history of the topological work on 

362: H--cobordisms of Seifert fibered 3--manifolds

363: is reviewed in \cite{sav}.

364: The same topic appears in singularity theory as the study of smoothings

365: of quasihomogeneous surface singularities. The link of such a singularity

366: is a Seifert fibered 3--manifold, and each smoothing exhibits the link as

367: the boundary of the Milnor fiber. 

368: In this setting, (\ref{tbmy.thm}) is closely related to 

369: \cite[3.10]{wahl-my}. His results imply (\ref{tbmy.thm})

370: for smoothings of negative weight.

371: \end{say}

372: 

373: 

374: 

375: 

376: \begin{say}[Comments and possible generalizations] 

377: 

378: 1. In the algebraic case,

379: if equality holds in (\ref{obmy.thm}.3) then $c_1(S)=0$ and

380: $S$ is the quotient of an Abelian or a K3 surface by a cyclic group.

381: In particular, $H_1(S\setminus \sing S,\z)\neq 0$. This is why

382: the $\leq$ in (\ref{obmy.thm}.3)

383:  was changed to a strict inequality in (\ref{tbmy.thm}.1).

384: An algebraic rational homology $\c\p^2$   can never have 6 singular points

385: and I don't know any examples with 5 singular points.

386: There are, however, many examples with 4 singular points

387: (\ref{4pt.icosa}). 

388: 

389: 

390: 

391: 

392: 2. Unlike in the algebraic case, the restriction

393: $H_1(M,\z)=0$ is not a consequence of $H_2(M,\z)=\z$.

394: Indeed, attaching a 1--handle to  $M$ increases

395: $H_1(M,\z)$ while leaving $H_2(M,\z)$ unchanged.

396: 

397: 3. The assumption $H_1(M,\z)=0$ is somewhat arbitrarily chosen.

398: One could require instead 

399: $\pi_1(M)=1$.

400: The variant with $H_1(M,\z)=0$ is the relevant

401: condition for integral H--cobordism questions

402: and $\pi_1(M)=1$ connects directly with the 

403: Montgomery--Yang problem on circle actions.

404: 

405: 4. In the algebraic case the inequality 

406: (\ref{tbmy.thm}.1) holds even if we only assume that

407: $H_1(M,\q)=0$, but in the topological case this is not enough.

408: There are lens spaces which bound a rational homology ball,

409: and taking connected sum of these with $\c\p^2$ gives examples

410: with an arbitrary number of boundary components.

411: 

412: The simplest algebraic example is the following.

413: Let $C\subset \c\p^2$ be a smooth conic and $C\subset N$

414: a regular neighborhood with boundary $L$. 

415: Since the normal bundle of $C$ in $\c\p^2$ has degree 4,

416: we see that $L$ is a $\z/4$-quotient of $S^3$.

417: Set

418: $M:=\c\p^2\setminus \inter N$. $M$ is a rational homology ball

419: with $\pi_1(M)=\z/2$ which bounds $L$.

420: 

421: There are many such examples, see, for instance, \cite{ca-ha}.

422: 

423: To get more algebraic ones, let  $u,v$ be  relatively prime

424: natural numbers.  Then

425:  the complement of a  regular neighborhood of the curve

426: $$

427: C:=(x^{u+v}-yz=0)\subset \p^2(1,u,v).

428: $$

429: is a rational homology ball $M$ with $\pi_1(M)=\z/(u+v)$

430: which bounds a lens space $L$ with $\pi_1(L)=\z/(u+v)^2$.

431: 

432: 5. By \cite[p.360]{fi-st-po}, the Poincar\'e sphere

433: is not H--cobordant to 0, thus one can not use 

434: connected sums to get counter examples to (\ref{tbmy.thm}).

435: 

436: 

437: 6. It is possible that Conjecture \ref{tbmy.thm} can be generalized

438: to arbitrary $H_2(M,\z)$. In this case, 

439:  (\ref{tbmy.thm}.1),  should be

440: replaced by 

441: $$

442: \sum_i \Bigl( 1-\frac1{|\pi_1(L_i)|}\Bigr)\leq 2+\dim H_2(M,\q).

443: \eqno{(\ref{tbmy.thm}.2)}

444: $$

445: This may hold if $\pi_1(M)=1$ 

446:  but --  as pointed out to me by P.~Hacking --

447: $H_1(M,\z)=0$ is not sufficient, not even for algebraic surfaces.

448: The surfaces in question arise as the minimal compactifications

449: of Seifert $\c^*$-bundles over $\p^1$.

450: These were  studied by \cite{dolgachev, pinkham, demazure, fl-za}. 

451: 

452: One of the simplest examples is the following.

453: 

454: \begin{exmp} Rational surfaces $S_m$ with 

455: $H_1(S_m,\z)=0, H_2(S_m,\z)\cong \z^2$ and with

456:  $2m$  quotient singularities.

457: \end{exmp}

458:   

459: Let us start with any minimal ruled surface $f:S\to \p^1$.

460: Pick points $p_i\in F_i\subset S$ where the $F_i$ are different fibers of

461: $f$ and natural numbers $r_i$. Blow up $(2+r_i)$-times each $p_i\in F_i$

462: and in each fiber contract all but the last $(-1)$-curve.

463: We get a singular ruled surface

464: $$

465: g=g(r_1,\dots,r_m): S(r_1,\dots,r_m)\to \p^1

466: \qtq{with} H_2(S(r_1,\dots,r_m),\q)\cong \q^2.

467: $$

468: The fiber over $f(p_i)$ is a smooth rational curve with

469: multiplicity $r_i$. Moreover, $S(r_1,\dots,r_m)$

470: has 2 singular points  along these fibers.

471: They are cyclic quotients of the form $\c^2/\frac1{r_i}(1,1)$

472: and $\c^2/\frac1{r_i}(1,-1)$.

473: 

474: It is easy to see that the fundamental group of the

475: smooth part is given by generators and relations as:

476: $$

477: \pi_1(S(r_1,\dots,r_m)^0)\cong \langle a_1,\dots, a_m: 

478: a_1^{r_1}=\cdots=a_m^{r_m}=

479: a_1\cdots a_m=1\rangle.

480: $$

481: If the $r_i$ are pairwise relatively prime

482: then $H_2(S(r_1,\dots,r_m),\z)\cong \z^2$

483: but the fundamental group is never trivial for $m\geq 3$.

484: \medskip

485: 

486:  It is worth noting that

487: (\ref{tbmy.thm}) completely fails 

488: if $M$ is only a topological manifold.

489: The number of boundary components can be arbitrary,

490: see (\ref{top-different}).

491: 

492: \end{say}

493: 

494: 

495: 

496: 

497: 

498: 

499: \section{The Montgomery--Yang problem}

500: 

501: Fixed point free differentiable circle actions on

502: $S^7$ with finitely many

503: nonfree orbits are classified in \cite{mo-ya}, see also \cite{pet}.

504: Such actions are frequently called {\it pseudo-free}.

505: The main idea of their classification is the 

506: following.

507: 

508: We work on $S^7$ and on the

509: orbifold quotient $X:=S^7/S^1$.

510: Let $O_i\subset S^7$ be a nonfree orbit

511: and $x_i\in X$ the corresponding orbifold point.

512: $O_i$ bounds a disk in $S^7$

513: whose image is a 2--sphere  $S_i\subset X$

514: containing $x_i$. 

515: Since $\dim X>4$, we can arrange these $S_i$ to be disjoint.

516: The classification now has 2 parts.

517: \begin{enumerate}

518: \item[i)] Describe  the

519: circle action  over a neighborhood of each $S_i$.

520: These turn out to be diffeomorphic to

521: linear circle actions on $S^3\times D^4$.

522: \item[ii)] Describe how these  local models glue together.

523: \end{enumerate}

524: 

525: By contrast, if we work on $S^5$, then

526: $\dim S^5/S^1=4$, thus the different $S_i$

527: do intersect, and we can not

528: separate the local models from each other. 

529: (Note that if  a manifold $M$ admits a 

530: fixed point free differentiable circle action, then

531: $e(M)=0$, thus this theory is

532: most interesting for odd dimensional manifolds.)

533: 

534: Montgomery and Yang found that it is

535: difficult to construct examples

536: of differentiable circle actions on $S^5$

537: with many nonfree orbits.

538: This led to the following:

539: 

540: \begin{conj}[Montgomery--Yang problem] 

541: \cite[p.41]{mo-ya}

542: Let $S^1\times S^5\to S^5$ be a differentiable circle action

543: with only finitely many nonfree  orbits.

544: Then there are at most 3 nonfree orbits.

545: \end{conj}

546: 

547: (Note that the proposed partial solution

548: in \cite{fi-st} is incorrect, see the review, MR0874031.)

549: 

550: 

551: There are many different actions

552: with 3 exceptional orbits.

553: The simplest ones are linear actions but  there are many

554: nonlinear examples too.

555: 

556: \begin{exmp} Let $S^5=(|x|^2+|y|^2+|z|^2=1)\subset \c^3$

557: be the unit sphere. Let $a,b,c\geq 2$ be pairwise relatively prime

558: natural numbers. Then

559: $S^1\times S^5\to S^5$, given by 

560: $$

561: (\lambda,x,y,z)\mapsto (\lambda^{bc}x,\lambda^{ca}y,\lambda^{ab}z)

562:  $$

563: is a differentiable circle action

564: with  3 nonfree orbits on the 3 coordinate axes.

565: 

566: Note that in this case the quotient

567: $S^5/S^1$ can be thought of 

568: as the (complex) weighted projective plane 

569: $\p^2(a,b,c)$.

570: \end{exmp}

571: 

572: 

573: In general, one can try to classify

574: circle actions on 5--manifolds $L$ in terms of the

575: 4--dimensional quotient orbifold $L/S^1$.

576: 

577: Let $L$ be a 5--manifold and

578: $S^1\times L\to L$ a differentiable circle action.

579: Assume that there are finitely many $S^1$-orbits

580: $O_1,\dots,O_m$ such that the action is 

581: free on $L\setminus \cup_iO_i$.

582: For each $O_i\subset L$ take an $S^1$-invariant

583: tubular neighborhood 

584: $O_i\subset T_i\subset L$. Then

585: $L^0:=L\setminus \cup_iT_i$ is a compact

586: 5--manifold with boundary and with a free circle action.

587: Thus $M:=L^0/S^1$ is a compact

588: 4--manifold with boundary, where the

589: boundary components are lens spaces.

590: 

591: The 4--manifold $M$ uniquely determines

592: $L$ and the $S^1$-action up to diffeomorphism in many cases.

593: See \cite{mo-ya, fi-st-po} for the pseudo-free case and \cite{e-surf}

594: in general. (While \cite{e-surf} considers only the algebraic case,

595: the result is valid in general using the methods

596: of \cite{circ-act}.) One of the simplest cases is

597: pseudo-free  circle actions on

598: 5--dimensional rational homology spheres.

599: 

600: \begin{thm}\label{S1.act.char}

601: There is a one--to--one correspondence between:

602: \begin{enumerate}

603: \item Pseudo-free differentiable circle actions on

604: 5--dimensional rational homology spheres $L$  with $H_1(L,\z)=0$.

605: \item Smooth, compact 4--manifolds $M$  with

606: boundary such that

607: \begin{enumerate}

608: \item  $\partial M=\cup_i L_i$ 

609: is a disjoint union of lens spaces

610: $L_i=S^3/\z_{m_i}$,

611: \item the $m_i$ are relatively prime to each other,

612: \item $H_1(M,\z)=0$ and $H_2(M,\z)\cong\z$.

613: \end{enumerate}

614: \end{enumerate}

615: Furthermore, $L$ is diffeomorphic to $S^5$ iff

616: $\pi_1(M)=1$.

617: \end{thm}

618: 

619: The smooth Bogomolov--Miyaoka--Yau--type conjecture

620: \ref{tbmy.thm} would give the following 

621: for circle actions:

622: 

623: \begin{conj} \label{gen.my.conj} Let $L$ be a 

624: 5--dimensional rational  homology sphere with $H_1(L,\z)=0$ admitting

625:  a pseudo-free differentiable circle action.

626: Let 

627: $O_1,\dots,O_k$ be the nonfree orbits with

628: stabilizers $\z/m_1,\dots,\z/m_k$. Then

629: $$

630: \sum_i \bigl(1-\tfrac1{m_i}\bigr)\leq 3.

631: $$

632: \end{conj}

633: 

634: In fact, using \cite{circ-act}, one can generalize

635: this to fixed point free actions which are not

636: pseudo-free.

637: Let us say that an orbit $O\subset L$ of an $S^1$-action is {\it exceptional}

638: if the order of its stabilizer $|\stab(O)|$

639: is bigger than the least common multiple of the

640:  orders of the stabilizers of nearby orbits,

641: that is   $\lcm\{|\stab(O_p)|:p\not\in O\}$.

642: It is easy to see that the quotient 

643: $$

644: m(O):=|\stab(O)|/\lcm\{|\stab(O_p)|:p\not\in O\}

645: $$

646: is an integer which is also the order of the local

647: fundamental group of $L/S^1$ at the image of $O$ \cite[Prop.15]{circ-act}.

648: Thus (\ref{tbmy.thm}) would imply the following

649: generalization of (\ref{gen.my.conj}):

650: 

651: \begin{conj} \label{gengen.my.conj} Let $L$ be a 

652: 5--dimensional rational homology sphere with $H_1(L,\z)=0$ admitting

653:  a fixed point free differentiable circle action.

654: Let 

655: $O_1,\dots,O_k$ be the exceptional orbits. Then

656: $$

657: \sum_i \bigl(1-\tfrac1{m(O_i)}\bigr)\leq 3.

658: $$

659: \end{conj}

660: 

661: 

662: \begin{say}

663: A positive answer to Conjecture \ref{tbmy.thm}

664: would come  close to settling the

665: Montgomery--Yang problem. Indeed, 

666: first we   enumerate

667: all sequences of pairwise  relatively prime

668: natural numbers $m_1,\dots,m_k$ such that

669: $$

670: \sum_{i=1}^k \bigl(1-\tfrac1{m_i}\bigr)\leq 3.

671: $$

672:  The list turns out to be very short and we get 

673: one of the following cases:

674: \begin{enumerate}

675: \item  $k\leq 3$, as required by the Montgomery--Yang problem,

676: \item $(2,3,5,n)$ for any $(n,30)=1$,

677: \item $(2,3,7,n)$ for $n\in\{11,13,17,19,23,25,29,31,37,41\}$, or

678: \item $(2,3,11,13)$.

679: \end{enumerate}

680: So, right away we get that there are at most 4

681: nonfree orbits (assuming Conjecture \ref{tbmy.thm}).

682: 

683: I do not know any 4--manifolds $M$ 

684: with $H_1(M,\z)=0$ corresponding

685: to the cases (2--4) above

686: but I can not exclude these even in the algebraic case.

687: (See, however, (\ref{4pt.icosa}) for some examples where one

688: of the singularities is not a cyclic quotient.)

689: 

690: Even if such a manifold exists,

691: it leads to a counterexample to the

692:  Montgomery--Yang problem only if it is  simply connected.

693: \end{say}

694: 

695: 

696: The methods to prove (\ref{S1.act.char}) lead to a complete

697: characterization of  compact, simply connected  5--manifolds

698: which admit a fixed point free differentiable circle action.

699: Before we state the result, we need

700: to define an invariant for   5--manifolds.

701: 

702: \begin{defn}\label{i(L).defn}

703: For any manifold $L$, write its  second homology as

704:  a direct sum

705: of cyclic groups of prime power order

706: $$

707: H_2(L,\z)=\z^k+\sum_{p,i} \bigl(\z/p^i\bigr)^{c(p^i)}

708: \qtq{for some $k=k(L), c(p^i)=c(p^i,L)$.}

709: \eqno{(\ref{i(L).defn}.1)}

710: $$

711: The numbers $k, c(p^i)$ are  uniquely determined by

712: $H_2(L,\z)$.

713: One can choose the decomposition

714:  such that 

715: the second Stiefel--Whitney class map

716: $$

717: w_2:H_2(L,\z)\to \z/2

718: $$ is zero 

719: on all but one summand  $\z/2^n$. This value $n$ is unique

720: and it is   denoted by $i(L)$ \cite{barden}.

721: Alternatively, 

722: $i(L)$ is the

723: smallest $n$ such that there is an $\alpha\in H_2(L,\z)$ such that

724: $w_2(\alpha)\neq 0$ and $\alpha$ has order $2^n$.

725: 

726: By  \cite{smale, barden}, a compact, simply connected  5--manifolds

727: is determined by the invariants $k(L), c(p^i,L)$ and $i(L)$.

728: \end{defn}

729: 

730: 

731: \begin{thm}\cite[Thm.3]{circ-act}\label{main.thm}

732: Let $L$ be a  compact, simply connected  5--manifold

733: with invariants  $k(L), c(p^i,L), i(L)$ as in (\ref{i(L).defn}.1).

734: Then $L$ 

735:  admits a fixed point free differentiable circle action

736: iff the following conditions hold:

737: \begin{enumerate}

738: \item  for every prime $p$, there are

739: at most $k+1$ nonzero $c(p^i)$,

740: \item  $i(L)\in \{0,1,\infty\}$, and

741: \item if $i(L)=\infty$ then there are

742: at most $k$ nonzero $c(2^i)$.

743: \end{enumerate}

744: \end{thm}

745: 

746: \begin{say}[Comments]

747: 

748: 1. The above result gives a complete

749: characterization of those manifolds that admit a 

750: fixed point free circle action,

751: but it does {\em not} describe all possible circle actions.

752: For any $L$ satisfying the above assumptions, infinitely

753: many topologically distinct circle actions are

754: constructed in \cite{circ-act} and 

755: a complete classification seems unlikely.

756: 

757: 2. The circle actions constructed in \cite{circ-act}

758: have 2--parameter families of nonfree orbits.

759: Pseudo actions exist only if $H_2(L,\z)$

760: is torsion free (cf.\ \cite[Prop.28]{circ-act}).

761: \end{say}

762: 

763: 

764: 

765: 

766: \section{H-cobordism of 3--manifolds}

767: 

768: 

769: \begin{defn} 

770: Let $\Sigma$ be an integral homology sphere,

771: that is $H_*(\Sigma,\z)=H_*(S^3,\z)$. 

772: $\Sigma$ is {\it H--cobordant} to zero if there is a

773: smooth 4--dimensional homology cell $W$ 

774:  with boundary $\partial W=\Sigma$.

775: Similarly, one can study rational homology spheres

776: which are rationally H--cobordant to zero.

777: 

778: It is an interesting and difficult question to decide

779: which integral/rational  homology spheres are 

780: H--cobordant to zero; see \cite[Chap.7]{sav}

781: for a  survey.

782: 

783: The case when  $\Sigma$ is Seifert fibered is

784: very much connected with the 

785: Bogomolov--Miyaoka--Yau inequality.

786: \end{defn}

787: 

788: 

789: \begin{defn}[Seifert fiber spaces]\label{seifert.def}{\ }

790: 

791:  A {\it Seifert fibered} 3--manifold is a

792: proper morphism of a 3--manifold to a surface $f:M\to S$

793: such that every point $s\in S$ has a neighborhood $s\in D_s\subset S$

794: such that the pair $f^{-1}( D_s)\to  D_s$ is fiber preserving

795: homeomorphic to one of the  normal forms $f_{c,d}$  defined 

796: as follows.

797: 

798: Let $S^1(z),D^2(z)\subset \c$ denote the unit circle (resp.\ open unit disk)

799: where $(z)$ indicates the name of the coordinate.

800: For a pair of integers $c,d$ satisfying  $0\leq c<d$ and

801: $(c,d)=1$, define

802: $$

803: f_{c,d}:S^1(u)\times D^2(z)\to D^2\qtq{by}

804:  f_{c,d}(u,z)=u^cz^d.

805: $$

806: $f_{c,d}$ restricts to a fiber bundle

807: $S^1\times (D^2\setminus \{0\})\to D^2\setminus \{0\}$.

808: The fiber of $f_{c,d}$ over the origin is still $S^1$, but 

809: $f_{c,d}^{-1}(0)$ has

810: multiplicity $d$. 

811: 

812: A Seifert fibered 3--manifold $M\to F$ is determined

813: by the location of the singular fibers $p_i\in F$, 

814: the above numbers $(c_i,d_i)$ at each point $p_i$

815: and by a global invariant,

816: see \cite{seif, seif-book, scott}.

817: For us the following consequence of the

818: classification is more important:

819: \end{defn}

820: 

821: \begin{prop} Given pairwise relatively prime natural numbers

822: $d_1,\dots,d_k$, there is a unique

823: integral  homology sphere $\Sigma(d_1,\dots,d_k)$

824: which has a Seifert fibered structure over $S^2$

825: with fiber multiplicities $d_1,\dots,d_k$.

826: \end{prop}

827: 

828: 

829: \begin{say}[Seifert disk bundles]

830: Let $\epsilon$ be a fixed $d$th root of unity

831: and $\mu_d\subset \c$ the group of $d$th roots of unity.

832: Consider the quotient

833: $$

834: D^2(x)\times D^2(y)/\mu_d(\epsilon, \epsilon^c),

835: $$

836: where we use this shorthand to denote the quotient

837: of $D^2(x)\times D^2(y)$ be the $\mu_d$-action

838: $(x,y)\mapsto (\epsilon x, \epsilon^cy)$.

839: The second projection map $y\to y^d$ descends to

840: $$

841: F_{c,d}:D^2(x)\times D^2(y)/\mu_d(\epsilon, \epsilon^c) {\to}

842:  D^2(y)/\mu_d(\epsilon^c)\cong D^2(y^d).

843: $$

844: Restricting to $S^1(x)\times D^2(y)$ we get isomorphisms

845: $$

846: S^1(x)\times D^2(y)/\mu_d(\epsilon, \epsilon^c)\cong

847: S^1(x)\times D^2(x^{-c}y)/\mu_d(\epsilon, 1)\cong

848: S^1(x^d)\times D^2(x^{-c}y),

849: $$

850: and setting $u=x^d, z=x^{-c}y$ the projection map $F_{c,d}$ becomes

851: $$

852: (u,v)\mapsto u^cz^d=(x^d)^c(x^{-c}y)^d=y^d.

853: $$

854: 

855: Thus we conclude  that $f_{c,d}$ can be extended 

856: to the projection map $F_{c,d}$. 

857: 

858: We can glue these together and obtain that

859: every Seifert fibered 3--manifold $M\to F$ can be obtained

860: as the boundary of a Seifert disk bundle $D(M)$.

861: The multiple fibers of multiplicity $m$

862: correspond to  cyclic quotient singularities

863: $\c^2/\mu_d(\epsilon, \epsilon^c)$ on $D(M)$.

864: 

865: Note that the zero section gives an embedding $F\into D(M)$

866: and $D(M)$ retracts to $F$.

867: The description of $M$ can also be given in terms of 

868: the self intersection of $F\subset D(M)$.

869: \end{say}

870: 

871: 

872: \begin{lem}\label{seif.h_1.lem}

873:  Let $M\to F=S^2$ be a Seifert fibered 3--manifold

874: with pairwise relatively prime  fiber multiplicities $d_1,\dots,d_k$

875: and corresponding Seifert disk bundle $D(M)$.

876: Then 

877: $$

878: (F\cdot F)=\pm \frac{|H_1(M,\z)|}{d_1\cdots d_k}.

879: $$

880: \end{lem}

881: 

882: 

883: There are  several series of 

884:  examples of Seifert fibered  homology spheres

885: which are H--cobordant to zero

886: \cite{ak-ki, ca-ha, st-br, neu}, but for all of them

887: the number of singular fibers is $\leq 3$. 

888: (See also \cite{orevkov}.)

889: Let us make the lack of

890: known examples into a formal conjecture:

891: 

892: \begin{conj}\label{seif.conj}  

893: Let $M \to S^2$ be a 

894:  Seifert fibered rational homology sphere

895: with fiber multiplicities $d_1,\dots,d_k$.

896: \begin{enumerate}

897: \item If $M$ is rationally H--cobordant to zero

898: then $\sum (1-\tfrac1{d_i})<3$.

899: \item If $M=\Sigma(d_1,\dots,d_k)$ and $M$ is   H--cobordant to zero

900: then $k\leq 3$.

901: \end{enumerate}

902: \end{conj}

903: 

904: 

905: It was observed in \cite{fi-st} that this problem

906: is closely  related to the existence of

907: 4--manifolds whose boundary components are lens spaces,

908: and hence to Conjecture \ref{tbmy.thm}.

909: 

910: 

911: Assume that $\Sigma(d_1,\dots,d_k)$ is the boundary of a

912: homology cell $W$. Above we wrote

913: $\Sigma(d_1,\dots,d_k)$ as the boundary of  a

914: Seifert disk bundle $D(\Sigma(d_1,\dots,d_k))$.

915: Gluing them together, we get

916: a 4--dimensional orbifold

917: $S$. The Mayer--Vietoris sequence shows that

918: $H_*(S,\z)=H_*(\c\p^2,\z)$.

919: Thus (\ref{tbmy.thm}) implies that we have at most 4 singular points.

920: 

921: Note, however, that (\ref{seif.conj}.2) is not equivalent

922: to (\ref{tbmy.thm}). If $S$ is a 4--dimensional orbifold

923: then usually one can not find an embedded copy of

924: $S^2$ passing through all singular points 

925: whose regular neighborhood is  a Seifert disk bundle.

926: 

927: \begin{exmp}[H--cobordisms of lens spaces]\label{lens.H.cob}

928: 

929: We give 2 algebraic constructions  showing that the

930: lens spaces

931: $$

932: \begin{array}{rcl}

933: L_{(n^2, nc-1)}&:=&S^3/(\z/n^2)(1,nc-1)\\

934: &:=&(|x|^2+|y|^2=1)/

935: (x,y)\mapsto (\epsilon x, \epsilon^{nc-1} y)

936: \qtq{where}\epsilon^{n^2}=1,

937: \end{array}

938: $$

939: are rationally H--cobordant to 0 for any $(c,n)=1$.

940: 

941: The first construction uses deformation of singularities as in

942:  \cite{wahl2} or \cite[Sec.3]{k-sb}.

943: Under the action

944: $$

945: (x,y)\mapsto (\epsilon x, \epsilon^{nc-1} y)

946: \qtq{where}\epsilon^{n^2}=1,

947: $$

948: the subgroup of $n$th roots of unity acts

949: as 

950: $(x,y)\mapsto (\epsilon^n x, \epsilon^{-n} y)$,

951:  and the quotient is

952: $$

953: (uv-w^n=0)\subset \c^3\qtq{where} u=x^n, v=y^n, w=xy.

954: $$

955: The induced quotient action of $\z/n\cong (Z/n^2)/(\z/n)$

956: is given by

957: $$

958: (u,v,w)\mapsto (\epsilon^n u, \epsilon^{-n} v, \epsilon^{nc} w)

959: $$

960: The Euler characteristic of the Milnor fiber

961: $(uv-w^n=1)$ is $n$, hence the 

962: Euler characteristic of the quotient Milnor fiber

963: $(uv-w^n=1)/(\z/n)$ is $1$. Therefore it is a

964: rational homology ball.

965: 

966: The second construction uses the  curve

967: $$

968: C:=(xy=z^{a+b})\subset \p^2(a,b,1).

969: $$

970: $C$ is smooth, rational and passes through 2 quotient

971: singularities. Thus the bounday $M(a,b)$ of its tubular neighborhood

972: is Seifert fibered over $C$ with 2 multiple fibers, hence

973: it is a lens space. Since $(C\cdot C)=(a+b)^2/ab$, we see from

974: (\ref{seif.h_1.lem}) that $|H_1(L,\z)|=(a+b)^2$.

975: 

976: 

977: One needs some explicit computations to decide which lens space.

978: The chart $z\neq 0$ is isomorphic to $\c^2$

979: with coordinates $X:=xz^{-a}, Y:=yz^{-b}$. The affine equation

980: of $C$ is $(XY=1)$.

981: 

982: $\p^2(a,b,1)\setminus (x=0)\cup C$ is isomorphic to

983: $\c^*\times \c$ and one can choose the isomorphism such that

984: in the $X,Y$-coordinates the corresponding coordinate functions

985: are

986: $$

987: (XY-1)^aX^{-a-b}\qtq{and} X^{b'-a'}(XY-1)^{-b'}

988: $$

989: where $aa'+bb'=1$. Correspondingly, 

990: $\p^2(a,b,1)\setminus (y=0)\cup C$ is also isomorphic to

991: $\c^*\times \c$ and  the corresponding coordinate functions

992: are

993: $$

994: (XY-1)^bY^{-a-b}\qtq{and} Y^{a'-b'}(XY-1)^{-a'}.

995: $$

996: These determine how the two charts are patched together

997: along 

998: $$

999: \c^*\times \c^*\cong \p^2(a,b,1)\setminus (xy=0)\cup C

1000: $$

1001: with coordinate functions

1002: $$

1003: XY(XY-1)^{-1}\qtq{and} X^bY^{-a}.

1004: $$

1005: A somewhat messy explicit computation gives that 

1006: one gets the lens space

1007: $$

1008: M(a,b)\cong L_{(a+b)^2, (a'-b')(a+b)-1}.

1009: $$

1010: Since $aa'+bb'=1$, we get that

1011: $(a+b)a'-(a'-b')b=1$, thus $a+b$ and $a'-b'$ are relatively prime.

1012: 

1013: It is easier to see this by putting together the

1014:  two descriptions  as follows.

1015: Fix $n$ and $a,c$ such that $ac\equiv 1\mod n$.

1016: Consider the family of weighted affine hypersurfaces

1017: $$

1018: X(\lambda):=(xy-z^n+\lambda t=0)\subset \p(a,n-a,1,n)\setminus (t=0).

1019: $$

1020: For $\lambda=0$ we get 

1021: $$

1022: X(0)=(xy-z^n=0)/\tfrac1{n}(a,n-a,1)

1023: \cong (xy-z^n=0)/\tfrac1{n}(1,-1,c).

1024: $$

1025: For $\lambda\neq 0$ we can eliminate $t$ to get that

1026: $$

1027: X(\lambda)\cong \p(a,n-a,1)\setminus (xy-z^n=0).

1028: $$

1029: Thus the quotient Milnor fiber of the first constructions

1030: is isomorphic to the complement of the curve $C$

1031: in the second construction.

1032: 

1033: 

1034: \end{exmp}

1035: 

1036: 

1037: The following series of 

1038: \cite[5.9.2]{wahl-sm}

1039: gives an algebraic realization of

1040: the rational H--cobordisms constructed in \cite[p.132]{neu}.

1041: See  (\ref{qs.ratcurves}) for its projective version.

1042: 

1043: \begin{exmp}\label{H.cob.exmps}

1044:  Let $a,b,c\geq 1$ be  integers. Set

1045: $S_{abc}:=(x^ay+y^bz+z^cx=0)\subset\c^3$.

1046: $S_{abc}$  is quasi homogeneous with weights

1047: $(bc-c+1, ca-a+1, ab-b+1)$,

1048: thus its link is a Seifert fibered 3--manifold $L_{abc}\to S^2$

1049: with 3 multiple fibers of multiplicities

1050: $ab-b+1, bc-c+1, ca-a+1$.

1051: The Milnor fiber is

1052: $$

1053: M_{abc}:=(x^ay+y^bz+z^cx=1)\subset\c^3,

1054: $$

1055: and its Euler characteristic  is

1056: $abc+1$. Let $\omega$ be a primitive $(abc+1)$st root of unity.

1057: The cyclic group $\z/(abc+1)$ acts freely on $M_{abc}$ by

1058: $$

1059: (x,y,z)\mapsto (\omega x, \omega^{-a}y,\omega^{ab}z).

1060: $$

1061: The quotient $M_{abc}/\bigl(\z/(abc+1)\bigr)$

1062: has Euler characteristic 1, thus it is a

1063: rational homology ball. 

1064: It bounds the Seifert fibered 3--manifold $L_{abc}/\bigl(\z/(abc+1)\bigr)$

1065: which also has  3 multiple fibers of multiplicities

1066: $ab-b+1, bc-c+1, ca-a+1$.

1067: \end{exmp}

1068: 

1069: 

1070: The following remarkable example of  \cite{w-let} shows that there are 

1071: even examples with 4 singular fibers. Further such examples are in 

1072: \cite{st-sz-w}.

1073: 

1074: \begin{exmp}[J.\ Wahl] A Seifert fibered rational homology sphere with

1075: 4 multiple fibers which is rationally H--cobordant to 0.

1076: 

1077: The classification of all finite subgroups of $O(n)$

1078: which act freely on the sphere $S^{n-1}$ is the

1079:  spherical space form problem.

1080:  See  \cite[Part III]{wolf} for a thorough treatment.

1081: One such action is obtained as follows

1082: (cf.\ \cite[5.5.6]{wolf}).

1083: 

1084: Fix an integer $u\geq 2$ and set

1085: $$

1086: m:= 3u^2-3u+1,\ n:=9u,\ r:=3u^2-6u+2,\ a:=3u-1.

1087: $$

1088: Let $\zeta$ (resp.\ $\omega$)  be a primitive $m$th 

1089: (resp.\ $3u$th) root of unity

1090: and consider the subgroup $G\subset GL(3,\c)$ 

1091: generated by

1092: $$

1093: A:= (x,y,z)\mapsto (\zeta x,\zeta^r y,\zeta^{r^2}z)

1094: \qtq{and}

1095: B:= (x,y,z)\mapsto (\omega^{-1}z, x,y).

1096: $$

1097: $G$ is nilpotent, 

1098: its order is $nm$ and it is given by

1099: the relations

1100: $$

1101: \langle A,B: A^m=B^n=1, BAB^{-1}=A^r\rangle.

1102: $$

1103: Consider now the function

1104: $$

1105: f(x,y,z):= x^ay+y^az+\omega z^ax.

1106: $$

1107: \medskip

1108: {\it Claim.} Notation as above.

1109: \begin{enumerate}

1110: \item  The link of the singularity

1111: $(f=0)/G$ is  Seifert fibered with 4 singular fibers

1112: of multiplicities  $(3,3,3,m)$.

1113: \item The Milnor fiber $(f=1)/G$ is a rational homology ball.

1114: \end{enumerate}

1115: \medskip

1116: 

1117: Proof. The singularity $(f=0)$ 

1118: can be resolved by 1 blow up. The exceptional curve

1119: $C$ is  smooth  of genus $\binom{a}{2}$.

1120: 

1121: Note that $B^3=\omega{\mathbf 1}$ acts trivially on $C$.

1122: The group $G/\langle B^3\rangle$ acts on $C$ with 4

1123: nonfree orbits. These are:

1124: 

1125: -- the orbits of $(1:\lambda \omega : \lambda^2 \omega)$

1126: where $\lambda^3 \omega=1$; these have order 3 stabilizers, and

1127: 

1128: -- the orbit of $(1:0:0)$; this has order $m$ stabilizer.

1129: 

1130: \noindent The Hurwitz formula now shows that

1131: $C/G\cong \p^1$. Moreover $B_0\c^3/G$ has 4 singular points,

1132: three have index 3 and one has index $m$. This shows the first part.

1133: 

1134: 

1135: To see the second part, note that

1136: by explicit computation,

1137: the  Milnor number of $f$ is 

1138: $$

1139: \dim \c[x,y,z]/(\partial f/\partial x,

1140: \partial f/\partial y, \partial f/\partial z)= a^3.

1141: $$

1142: Thus the Euler characteristic of the Milnor fiber $(f=1)$ is

1143: $1+a^3=mn=|G|$. Therefore

1144:  the Euler characteristic of the quotient Milnor fiber

1145:   $(f=1)/G$ 

1146: is 1, and hence it is a rational homology ball.\qed

1147: \end{exmp}

1148: 

1149: 

1150: \begin{say}[Smoothing surface singularities]

1151: Let $(0\in S)$ be a normal surface singularity

1152: with link $L$ and let $M$ be the Milnor fiber of a smoothing of

1153: $S$. With suitable care, $M$ is a 4--manifold whose boundary is

1154: $L$. We are especially interested in the case when

1155: $M$ is a rational homology ball.

1156: 

1157: This imposes a very strong restriction on $(0\in S)$.

1158: For instance,  it seems to have been known  for some time

1159:  that $(0\in S)$ is a rational singularity,

1160: cf.\ \cite[Sec.2.3]{st-sz-w}. 

1161: The latter also shows that (\ref{seif.conj}.1) holds

1162: for the link of a normal surface singularity

1163:  which has a smoothing 

1164: whose Milnor fiber is a rational homology ball.

1165: 

1166: 

1167: For other relationships between the algebraic geometry

1168: of a surface singularity and the Seiberg-Witten theory

1169: of its link see \cite{ne-wa, w-char, n-n1, n-n2, n-n3}.

1170: \end{say}

1171: 

1172: 

1173: 

1174: 

1175: \begin{rem}\label{top-different} By the results of 

1176: \cite{free}, every  integral homology sphere

1177: bounds a topological homology 4--cell.

1178: By the above construction this implies that

1179: there are topological 

1180: integral 

1181: homology $\c\p^2$-s with any number of singular points.

1182: \end{rem}

1183: 

1184: 

1185: 

1186: 

1187: \section{Open problems on algebraic surfaces}

1188: 

1189: On the algebraic geometry side, all of the

1190: questions can be gathered into one central problem:

1191: 

1192: \begin{prob}\label{hopeless} Classify all integral/rational

1193:  homology $\c\p^2$-s with

1194: quotient singularities.

1195: \end{prob}

1196: 

1197: I do not expect this to be  feasible.

1198: The case when the canonical class is

1199: anti ample, that is, we are looking at log--Del Pezzo

1200: surfaces, received a lot of attention,  but

1201: there seem to be too many cases for a complete structure theorem;

1202: see \cite{miyanishi, ke-mc, shokurov}.

1203: 

1204: The case where the canonical class is numerically trivial should be

1205: hard but there is a clear path to follow.

1206: If $mK_S\sim 0$ then $S$ has a degree $m$ cover

1207: which is either an Abelian surface or a K3 surface with

1208: Du~Val singularities.  Thus the problem reduces to the

1209: classification of cyclic group actions on

1210: Abelian and K3 surfaces.

1211: The 7 cases where the Picard number of

1212: the K3 surface is maximal  are classified in \cite{og-zh}.

1213: 

1214: 

1215: Very little is known about the case when the canonical class is

1216: ample. The recent classification in the smooth case

1217: \cite{prasad} is very significant, but it probably

1218: says very little about the singular case.

1219: 

1220: The following examples are worked out in (\ref{rhcp2.say}).

1221: 

1222: 

1223: \begin{exmp} \label{rat.surf.exmp} Let $a_1,a_2,a_3,a_4$ be

1224: natural numbers  such that

1225: $a_2a_3a_4-a_3a_4+a_4-1$ and  

1226: $a_1a_2a_3a_4-1$ are relatively prime.

1227: (This holds in at least 75\% of all cases.)

1228: Let $S$ be the surface

1229: $$

1230: S:=S(a_1,\dots,a_4):=

1231: (x_1^{a_1}x_2+x_2^{a_2}x_3 + x_3^{a_3}x_4+ x_4^{a_4}x_1=0)

1232: \subset \p(w_1,\dots,w_4)

1233: $$

1234: where, using subscripts modulo 4,

1235: $$

1236: w_i=a_{i+1}a_{i+2}a_{i+3}-a_{i+2}a_{i+3}+a_{i+3}-1.

1237: $$

1238: Note that  $S$ contains the curves

1239: $(x_1=x_3=0)$ and $(x_2=x_4=0)$ and they can be 

1240: contracted. Thus we obtain  a surface

1241:  $S^*=S^*(a_1,a_2,a_3,a_4)$. 

1242: We compute in Section 5 that:

1243: \begin{enumerate}

1244: \item $S^*$ is a rational surface with

1245: quotient singularities,

1246: \item  $H_*(S^*,\q)=H_*(\c\p^2,\q)$,

1247: \item if $a_i\geq 4$ then 

1248: the canonical class of $S^*$ is ample, and

1249: \item $(K_{S^*}^2)$ converges to 1 as $\min \{a_i\}\to \infty$.

1250: \end{enumerate}

1251: \end{exmp}

1252: 

1253: Let us also note that in positive characteristic

1254: one can get examples with many quotient singularities.

1255: 

1256: \begin{exmp}[Characteristic $p$]  Let $k$ be a  field 

1257: of characteristic $p$. For some $q=p^m$, blow up all the

1258: $\f_q$-points of $\p^2$. The birational transforms of the

1259: $\f_q$-lines become disjoint, smooth, rational curves with self intersection

1260: $-q$. They can be contracted to obtain a surface

1261: $X_q$ defined over $\f_q$ with $q^2+q+1$ singular points.

1262: These are quotient singularities of the type $\a^2/\mu_q$

1263: where $\mu_q$ is the subgroup scheme of $q$th roots of unity

1264: $\spec k[t,t^{-1}]/(t^q-1)$. 

1265: (Note that in characteristic $p$ it is the 

1266: $\z/q$ quotients that behave very badly (cf.\ \cite{artin})

1267: and the $\mu_q$-quotients are the correct characteristic $p$

1268: analogs of characteristic $0$ quotient singularities.)

1269: 

1270: 

1271: 

1272: $X_2$ is a Del Pezzo surface of degree 2, but 

1273: the canonical class is ample for $q\geq 3$. 

1274: \end{exmp}

1275: 

1276: 

1277: Below I list some special cases of Problem \ref{hopeless}

1278: which are of interest either as partial steps in the

1279: classification or as having immediate topological

1280: consequences.

1281: 

1282: \begin{conj}\cite[4.17]{e-surf}\label{sconnconj}

1283:  Let $S$ be a rational homology $\c\p^2$ with

1284: quotient singularities.

1285: If $S^0:=S\setminus \sing S$ is simply connected then $S$ is rational.

1286: \end{conj}

1287: 

1288: This is similar in spirit  to the result 

1289: that in dimension 2, all algebraic  $\q$--homology cells are rational

1290: \cite{gu-pr}. (\ref{sconnconj}) was verified by \cite{keum}

1291: when the singularities are not very complicated.

1292: 

1293: 

1294: 

1295: \begin{conj}[Algebraic Montgomery--Yang problem]

1296:  Let $S$ be a rational homology $\c\p^2$ with

1297: quotient singularities.

1298: If $S^0:=S\setminus \sing S$

1299:  is simply connected then $S$ has at most 3 singular points.

1300: \end{conj}

1301: 

1302: \begin{exmp}\label{4pt.icosa}

1303:  Let $G\subset SL(2,\c)$ be 

1304: subgroup which contains no quasi--reflections such that its

1305: image in $PSL(2,\c)$ is the  icosahedral group $I$

1306: (see \cite{briesk-s} for a complete list and the

1307: corresponding quotient singularities). Let $Z\subset G$ be its center,

1308: then $G/Z\cong I$.

1309: Extend the $G$--action on $\c^2$ to $\c\p^2$. 

1310: The center acts trivially on the line at infinity

1311: and $\c\p^2/Z$ is a cone over the rational normal  curve of degree $|Z|$.

1312: Then

1313: $S_G:=\c\p^2/G=(\c\p^2/Z)/I$ has 4 quotient  singularities,

1314: one of type $\c^2/G$ at the origin, 3 of types

1315: $\c^2/\z_2, \c^2/\z_3,\c^2/\z_5$ at infinity. The fundamental group

1316: of $S^0_G$ is $I$, thus $S_G$ is an integral homology $\c\p^2$.

1317: 

1318: These examples can also be obtained by starting with a minimal ruled surface

1319: and blowing up inside 3 of the fibers.

1320: \end{exmp}

1321: 

1322: \begin{prob} Classify all integral/rational homology $\c\p^2$-s with

1323: 4 or more quotient singularities.

1324: \end{prob}

1325: 

1326: 

1327: 

1328:  It is also of considerable interest to study algebraic surfaces

1329: that lead to H--cobordisms of Seifert fibered manifolds.

1330: If we want to stay completely algebraic, then 

1331: we are lead to the following

1332: 

1333: \begin{prob}\label{alg.seif.Hcob} Classify all  pairs $(S,C)$ such that

1334: \begin{enumerate}

1335: \item $S$ is a rational homology $\c\p^2$,

1336: \item $C$ is a rational curve, homeomorphic to $S^2$.

1337: \end{enumerate}

1338: \end{prob}

1339: 

1340: 

1341: In this case $S\setminus C$ is a {\it rational homology plane}

1342: or a {\it $\q$-acyclic surface}. That is a (nonproper) surface 

1343: $X$  such that $H_*(X,\q)=H_*(\c^2,\q)$.

1344: 

1345: 

1346: 

1347: 

1348: There is a huge body of literature devoted to classifying

1349: integral/rational homology planes. 

1350: See \cite{1, 5, 2, 3, 4, 6, gu-pr, dr1, dr2, dr3}

1351:  and the many references there.

1352: Nonetheless, most rational homology planes can not be compactified

1353: to get a rational homology $\c\p^2$, thus 

1354: (\ref{alg.seif.Hcob}) may be a much easier problem.

1355: 

1356: 

1357: \begin{prob} Topological smoothings of surfaces with

1358: quotient singularities.

1359: \end{prob}\label{top.smooth.prob}

1360: 

1361: Let $S$ be a proper surface with quotient singularities

1362: $p_i\in S$. Let $p_i\in M_i$ be small conical neighborhoods.

1363:  For each singularity $(p_i\in M_i)$ choose a 

1364: 4--manifold such that $\partial N_i=\partial M_i$.

1365: For instance, $N_i$ could be the general fiber of a smoothing

1366: of $M_i$.

1367:  In general, there need not be an

1368: algebraic deformation of $S$ which realizes these local smoothings.

1369: However, one can always get a differentiable manifold

1370: by replacing each singular $M_i$ with the smooth $N_i$.

1371: 

1372: It is interesting to find this way 

1373: differentiable manifolds which are not algebraic.

1374: This idea, though formulated purely topologically, 

1375: was used in \cite{st-sza, p-s-sz, fi-st-5} to construct 

1376: exotic differentiable structures on $\c\p^2$

1377: blown up at $\geq 5$ points.

1378: 

1379: It is especially interesting to work with the case

1380: when each $N_i$ is a rational homology ball.

1381: In the early seventies, Casson, Gordon and Conway

1382: (unpublished) 

1383: found 3  such classes:

1384: \begin{enumerate}

1385: \item $\c^2/\frac1{n^2}(1,na- 1)$ where $(n,a)=1$,

1386: \item $\c^2/\frac1{n^2}(1,d(n- 1))$ where $d|n- 1$ is odd, and

1387: \item $\c^2/\frac1{n^2}(1,d(n- 1))$ where $d|2n+ 1$.

1388: \end{enumerate}

1389: Recently \cite{lisca} proved that these are in fact all,

1390: if we also take into account two elementary observations:

1391: \begin{enumerate}\setcounter{enumi}{3}

1392: \item replacing the generator of $\z/p$ by its inverse

1393: shows that $\c^2/\frac1{p}(1,q)\cong \c^2/\frac1{p}(1,q')$

1394: for $qq'\equiv 1 \mod p$, and

1395: \item  conjugating one of the coordinates shows

1396: that $\c^2/\frac1{p}(1,q)$ is diffeomorphic to $\c^2/\frac1{p}(1,p-q)$.

1397: \end{enumerate}

1398: 

1399: Two algebraic realizations of the first series were given

1400: in (\ref{lens.H.cob}). I do not know algebraic descriptions

1401: of the other two.

1402: 

1403: Let $S$ be a proper rational homology $\c\p^2$

1404:  with quotient singularities

1405: $p_i\in S$ which are on the list

1406: (\ref{top.smooth.prob}.1--5).

1407: The resulting topological smoothing is then a smooth

1408: 4--manifold which is also a  rational homology $\c\p^2$.

1409: 

1410: 

1411: 

1412: 

1413: 

1414: In an earlier version of this note I

1415: raised the possibility that such surfaces could lead to

1416: a fake $\c\p^2$, that is, a smooth 4--manifold

1417: homeomorphic but not diffeomorphic to $\c\p^2$.

1418: Moreover,  I was hoping to do this with smoothings that

1419: are locally complex analytic.

1420: To this end, one needs to 

1421: find rational homology $\c\p^2$-s with

1422: singularities of the form  $\c^2/\frac1{n^2}(1,na-1)$.

1423: 

1424: If $S$ has such  singularities and $-K_S$ is ample,

1425: then the local deformations of the singular points can be globalized,

1426: and $S$ is the degeneration of smooth Del Pezzo surfaces.

1427: These were studied and partially classified in

1428: \cite{manetti, hac-pro}.

1429: The  general type examples, however,  have  a tendency to be

1430: rigid (cf.\ \cite[6.12]{6}), and they may lead to

1431: new differentiable 4--manifolds.

1432: Unfortunately, we get nothing interesting.

1433: 

1434: 

1435: \begin{thm}[P.\ Hacking] Let $S$ be a 

1436: rational homology $\c\p^2$ with

1437: singularities of the form  $\c^2/\frac1{n_i^2}(1,n_ia_i-1)$

1438: for some $(n_i,a_i)=1$.

1439: If $K_S$ is nef then $S$ is smooth.

1440: \end{thm}

1441: 

1442: Proof. Since $K_S$  is nef,

1443: we have the orbifold BMY inequality 

1444: $c_1(S)^2 \le 3e_{orb}(S)$. 

1445: 

1446: Noether's formula $\chi(\o_S)=\frac1{12}(c_1(S)^2+e(S))$

1447: usually needs a correction term for

1448: each quotient singularity, but for our singularities

1449: the

1450: correction term vanishes, cf.\ \cite[Prop.3.5]{hac-pro}.

1451: Thus  $c_1(S)^2=9$ and so $3\leq e_{orb}(S)$ thus $S$ is smooth.\qed

1452: \medskip

1453: 

1454: If we also allow the other singularities on the

1455: list (\ref{top.smooth.prob}.1--5), there are

1456: interesting examples. For instance

1457: $$

1458: \bigl(u^2=(x^2+y^2-z^2)(x^2+2y^2-z^2)\bigr)\subset \p^3(1,1,1,2)

1459: $$

1460: is a degree 2 Del Pezzo surface with Picard number 2

1461: and two singular points of type $A_3$, that is,

1462: $\c^2/\frac14(1,3)$. Topological smoothing

1463: creates out of it a 4--manifold with $b_2=2$

1464: (which is probably not simply connected).

1465: 

1466: It would be interesting to start a systematic study of such examples.

1467: 

1468: 

1469: \section{Examples of rational homology projective spaces}

1470: 

1471: 

1472: In this section we investigate hypersurfaces in weighted projective

1473: spaces given by an equation

1474: $$

1475: H(a_1,\dots,a_n):=

1476: (x_1^{a_1}x_2+x_2^{a_2}x_3+ \cdots + x_{n-1}^{a_{n-1}}x_n+ x_n^{a_n}x_1=0)

1477: \subset \p(w_1,\dots,w_n).

1478: $$

1479: These  hypersurfaces,

1480: or rather, the corresponding cones, play a fundamental role

1481: in the classification of quasi--homogeneous singularities

1482:  \cite{or-ra, kou}, but they  have

1483: many other remarkable properties as well.

1484: 

1485: The best known examples arise when

1486: $a_1=\cdots=a_n=a$, giving hypersurfaces in

1487: ordinary projective space

1488: $$

1489: H(a):=

1490: (x_1^{a}x_2+x_2^{a}x_3+ \cdots + x_{n-1}^{a}x_n+ x_n^{a}x_1=0)

1491: \subset \p^{n-1}.

1492: $$

1493: These have been studied for their  large group of automorphisms

1494: among others. (The case $a=n=3$ is Klein's curve of genus 3

1495: with  a simple group of order 168 as automorphisms.)

1496: 

1497: It turns out, however, that this case is  completely misleading

1498:  and for general $a_1,\dots,a_n$ we get

1499: very different behavior.

1500: 

1501: The closely related examples of 

1502: \cite{libg, bar-dim}

1503: $$

1504: (x_1^{a}x_2^{d-a}+x_2x_3^{d-1}+ \cdots + x_{n-1}x_n^{d-1}+ x_{n+1}^{d}=0)

1505: \subset \p^n

1506: $$

1507: give hypersurfaces with (non--quotient) isolated singularities

1508: which have the same integral homology  as $\c\p^{n-1}$

1509: if $\gcd(a,d(d-1))=1$.

1510: 

1511: 

1512: \begin{say}[Summary of the results]\label{summary.say}

1513: 

1514: Let $H=H(a_1,\dots,a_n)$ be as above.

1515: Then $H$

1516: has only cyclic quotient singularities for $n\geq 4$.

1517: Under a mild but not very explicit restriction on $ a_1,\dots,a_n$

1518: (\ref{num.conds}) we show that

1519: \begin{enumerate}

1520: \item $H$ is birational to $\c\p^{n-2}$, but

1521: \item if every $a_i\geq n$ and $n\geq 4$

1522: then the canonical class $K_{H}$ is ample and

1523: its self intersection $(K_H^{n-2})$ converges to 1 

1524:  as $\min\{ a_i\}\to \infty$.

1525: \end{enumerate}

1526: Moreover, if $n$ is odd then 

1527: \begin{enumerate}\setcounter{enumi}{2}

1528: \item  $H$ is a  rational homology  $\c\p^{n-2}$, and

1529: \item $\p(w_1,\dots,w_n)\setminus H$ is a rational homology $\c^{n-1}$.

1530: \end{enumerate}

1531: For $n=3$ this gives many examples of 

1532: Seifert fibered rational homology spheres which are

1533: rationally $H$-cobordant to 0, see (\ref{H.cob.exmps}).

1534: 

1535: If $n=2m$ is even then  $H_{n-2}(H,\q)$ has dimension 3. 

1536: However, if $n=4$,  then $H$ contains

1537: 2 disjoint contractible curves and after

1538: contracting them we get $H\to H^*$ and in (\ref{rat.surf.exmp})

1539: we show that

1540: \begin{enumerate}\setcounter{enumi}{5}

1541: \item  $H^*$ is a  rational homology  $\c\p^{n-2}$, and 

1542: \item if every $a_i\geq 4$ then  the canonical class $K_{H^*}$ is ample.

1543: \end{enumerate}

1544: \end{say}

1545: 

1546: \begin{rem} In singularity theory and in topology the

1547: Brieskorn--Pham singularities

1548: $$

1549: (x_1^{a_1}+x_2^{a_2}+ \cdots + x_n^{a_n}=0)\subset \c^n

1550: $$

1551: are much better known.

1552: When the link of a Brieskorn--Pham singularity

1553: is a homology sphere, then the corresponding projective hypersurface

1554: is isomorphic to a weighted projective space \cite{briesk}

1555: and all the intricate geometry is concentrated

1556: in the corresponding Seifert bundle structure 

1557: (see, for instance, \cite{or-wa}).

1558: 

1559: By contrast, for the singularities

1560: $$

1561: (x_1^{a_1}x_2+x_2^{a_2}x_3+ \cdots + x_{n-1}^{a_{n-1}}x_n+ x_n^{a_n}x_1=0)

1562: \subset \c^n

1563: $$

1564: the projective hypersurfaces are also very interesting

1565: and the Seifert bundle structure  is usually simple.

1566: \end{rem}

1567: 

1568: 

1569: \begin{say}[Numerical conditions]\label{num.conds}

1570: 

1571: The exponents $a_1,\dots,a_n$,

1572: the weights $w_1,\dots,w_n$ and the degree $d$ are related by the

1573: equations

1574: $$

1575: a_iw_i+w_{i+1}=d\qtq{for} i=1,\dots,n,

1576: $$

1577:  where we write all subscripts modulo $n$.

1578: Let us fix the exponents $a_i$ and assume that $\prod a_i\neq (-1)^n$.

1579: Using (\ref{linalg.lem}) the system can be solved

1580: explicitly. Set

1581: $$

1582: W_i:=\sum_{j=1}^n(-1)^{j-1}\prod_{\ell=i+j}^{i+n-1}a_{\ell}

1583: \qtq{and} D:=\prod_{\ell=1}^n a_{\ell}+(-1)^{n-1},

1584: $$

1585: where $D$ is the determinant of the system.

1586: It is easy to check that

1587: $$

1588: a_iW_i+W_{i+1}=D\qtq{for} i=1,\dots,n.\eqno{(\ref{num.conds}.1)}

1589: $$

1590: We would like to have a well formed weighted projective space,

1591: thus we have  to divide the weights by

1592: their greatest common divisor

1593: $$

1594: w^*:=\gcd(W_1,\dots,W_n).\eqno{(\ref{num.conds}.2)}

1595: $$

1596: Note that the equations (\ref{num.conds}.1) imply that

1597: $$

1598: w^*=\gcd(W_i,D)\ \forall i\qtq{and}

1599: w^*=\gcd(W_i,W_{i+1})\ \forall i.\eqno{(\ref{num.conds}.3)}

1600: $$

1601: It turns out that the cases with $w^*=1$

1602: have many special properties that are not shared by the

1603: examples with $w^*>1$. 

1604: 

1605:  It is not clear to me how to determine whether $w^*=1$,

1606: other than actually computing it. It is, however, easy to see

1607: that $w^*=1$ happens frequently.

1608: 

1609: Note that we can write $W_n=a_{n-1}A\pm 1$ and

1610: $D=a_na_{n-1}B\pm 1$ where $A,B$ depend on

1611: $a_1,\dots,a_{n-2}$ only. Fix a prime $p$

1612: and $a_1,\dots,a_{n-2}$.

1613: No matter what $A,B$ are, there is at most one choice

1614: for $a_{n-1}$ modulo $p$ such that 

1615:  $p$ divides $W_n$ and then at most one choice

1616: for $a_{n}$ modulo $p$ such that 

1617:  $p$ divides $D$.

1618:  By the Chinese remainder theorem,

1619: the conditions for different primes are independent.

1620: Thus the proportion of the $n$-tuples  with $w^*=1$ is

1621: asymptotically at least

1622: $$

1623: \prod_p\bigl( 1-\frac{1}{p^2}\bigr)=\frac{1}{\zeta(2)}=\frac {6}{\pi^2}

1624: =0.607...

1625: $$

1626: Note that $D=\prod a_i\pm 1$ is divisible by 2 only if all the

1627: $a_i$ are odd, and then $w_i\equiv n\mod 2$. This 

1628: allows us to conclude that the above $0.607$ can be

1629: improved to $0.8$ if $n$ is odd and to $0.75$ if $n$ is even.

1630: 

1631: Set

1632: $$

1633: w_i:=\tfrac1{w^*}W_i\qtq{and} 

1634:  d:= \tfrac1{w^*}D.\eqno{(\ref{num.conds}.4)}

1635: $$

1636: From (\ref{num.conds}.3) we conclude that

1637: $$

1638: \gcd(w_i,d)=1\ \forall i\qtq{and}

1639: \gcd(w_i,w_{i+1})=1\ \forall i.\eqno{(\ref{num.conds}.5)}

1640: $$

1641: \end{say}

1642: 

1643: Note  that

1644: $H$  can also be viewed as a general element of the

1645: linear system

1646: $$

1647: |H|=|x_1^{a_1}x_2,x_2^{a_2}x_3, \cdots, x_{n-1}^{a_{n-1}}x_n, x_n^{a_n}x_1|.

1648: $$

1649: Indeed, take any $\lambda_i\neq 0$ and consider 

1650: $$

1651: H(\lambda):=(\lambda_1x_1^{a_1}x_2+\lambda_2x_2^{a_2}x_3+ \cdots + 

1652: \lambda_{n-1}x_{n-1}^{a_{n-1}}x_n+ \lambda_nx_n^{a_n}x_1=0)

1653: $$

1654: Choose $\mu_i$ such that $\mu_i^{a_i}\mu_{i+1}=\lambda_i^{-1}$.

1655: Such a choice is possible since we can view these equations as a 

1656: linear system for the $\log \mu_i$ whose determinant is

1657: $D\neq 0$. Thus $H$ and $H(\lambda)$ differ only by a 

1658: coordinate change.

1659: 

1660: \begin{thm}\label{main.exmp.thm}

1661:  Assume that $\prod a_i\neq (-1)^n$

1662: and write all subscripts modulo $n$. 

1663: Define $w_i,d$ and $w^*$ as in (\ref{num.conds}.4). Then

1664: \begin{enumerate}

1665: \item $\p(w_0,\dots,w_n)$ is a 

1666: well formed weighted projective space

1667: whose singular set has dimension $\leq \rdown{n/2}-1$.

1668: \item The hypersurface

1669: $$

1670: H(a_1,\dots,a_n):=

1671: (x_1^{a_1}x_2+x_2^{a_2}x_3+ \cdots + x_{n-1}^{a_{n-1}}x_n+ x_n^{a_n}x_1=0)

1672: \subset \p(w_1,\dots,w_n)

1673: $$

1674: is quasi--smooth.

1675: \item $\p(w_1,\dots,w_n)\setminus H(a_1,\dots,a_n)$

1676: is smooth.

1677: \item If $w^*=1$ then $H$ is birational to $\c\p^{n-2}$.

1678: \item $K_H$ is ample if $\min\{a_i\}\geq n$ and 

1679:  $(K_H^{n-2})\to 1$  as $\min\{a_i\}\to \infty$.

1680: \item If $n$ is odd and $w^*=1$ then

1681: $$

1682: \begin{array}{l}

1683: H_*(H(a_1,\dots,a_n),\q)\cong H_*(\c\p^{n-2},\q)\qtq{and}\\

1684: H_*(\p(w_1,\dots,w_n)\setminus H(a_1,\dots,a_n),\q)\cong H_*(\c^{n-1},\q).

1685: \end{array}

1686: $$

1687: \item If $n=2m$ is even and $w^*=1$ then

1688: $$

1689: \begin{array}{l}

1690: H_j(H(a_1,\dots,a_n),\q)\cong H_j(\c\p^{n-2},\q)\qtq{for $j\neq n-2$, and}\\

1691: H_{n-2}(H(a_1,\dots,a_n),\q)\cong \q^3.

1692: \end{array}

1693: $$

1694: The middle homology is spanned by

1695: the the complete intersection class $c_1(\o_{\p}(1))^{m-1}\cap [H]$ and the 

1696: two disjoint weighted linear subspaces

1697: $$

1698: (x_1=x_3=\cdots=x_{2m-1}=0)\qtq{and}

1699: (x_2=x_4=\cdots=x_{2m}=0).

1700: $$

1701: \end{enumerate}

1702: \end{thm}

1703: 

1704: Proof. The singular locus of

1705: $\p(w_1,\dots,w_n)$ is a union of weighted linear subspaces

1706: $L_I$ where

1707:  $I\subset \{1,\dots,n\}$ is a  subset such that

1708: $\gcd\{w_i:i\in I\}\neq 1$ and

1709:  $$

1710: L_I:=\{(x_1,\dots,x_n):x_j=0\ \forall \ j\not\in I\}.

1711: $$

1712: As we noted in (\ref{num.conds}.5), $I$ does not contain 

1713: any pair of indices whose

1714: difference is 1. Thus $|I|\leq n/2$ and so

1715: $\dim L_I\leq \rdown{n/2}-1$.

1716: We also see that $L_I\subset H$ for every such $I$.

1717: This shows (\ref{main.exmp.thm}.1) and (\ref{main.exmp.thm}.3). 

1718: 

1719: Outside $(\prod x_i=0)$ the hypersurface

1720: $H$ is smooth by Bertini.

1721: Assume that $H$ is not quasi--smooth at the point

1722: $(p_1,\dots,p_n)$ and $p_i=0$. 

1723: Then $\partial h/\partial x_i=0$ shows that $p_{i-1}=0$

1724: and by repeating the argument we get that all the $p_j=0$. Thus

1725: $H$ is quasi--smooth, proving (\ref{main.exmp.thm}.2).

1726: 

1727: 

1728: 

1729: Assuming that $w^*=1$,  we show that the

1730: linear system

1731: $$

1732: |H|=|x_1^{a_1}x_2,x_2^{a_2}x_3, \cdots, x_{n-1}^{a_{n-1}}x_n, x_n^{a_n}x_1|

1733: $$

1734: maps $\p(w_1,\dots,w_n)$ birationally to $\p^{n-1}$ and so

1735:  $H$ is mapped birationally to

1736: a hyperplane in $\p^{n-1}$. Note that $|H|$ 

1737: restricts to a homomorphism between the tori

1738: $$

1739: \eta:(\c^*)_{\mathbf x}^n\to (\c^*)_{\mathbf y}^n

1740: \qtq{given by} \eta^*y_i=x_i^{a_i}x_{i+1},

1741: $$

1742: where $(y_1,\dots,y_n)$ are coordinates on the target $\p^{n-1}$.

1743: The degree of $\eta$  is the determinant of the

1744: matrix of exponents, which we already computed to be

1745: $D=\prod_i a_i+(-1)^{n-1}$.

1746: 

1747: Let us now restrict $\eta$ to  the 1--parameter subgroup

1748: $(\lambda^{w_1},\dots, \lambda^{w_n})$. We get

1749: a homomorphism of degree $d$:

1750: $$

1751: \eta:(\lambda^{w_1},\dots, \lambda^{w_n})\to 

1752: (\lambda^d,\dots,\lambda^d).

1753: $$

1754: Note that  $w^*=1$ iff $d=D$,

1755: thus if $w^*=1$ then $\eta$ descends to an isomorphism

1756: $$

1757: (\c^*)_{\mathbf x}^n/(\lambda^{w_1},\dots, \lambda^{w_n})\to

1758:  (\c^*)_{\mathbf y}^n/(\lambda,\dots,\lambda).

1759: $$

1760: This is exactly the map given by $|H|$,

1761: proving (\ref{main.exmp.thm}.4).

1762: 

1763: Although it is not needed for our purposes,

1764: one can also write down the inverse of the map

1765: given by $|H|$. First we get the formulas

1766: $$

1767: \begin{array}{lcl}

1768: x_i^D&=&\prod_{j=0}^{n-1}y_{i+j}^{b_{ij}}\qtq{where}

1769: b_{ij}:=(-1)^{j+1}\prod_{\ell=i+j-1}^{i+n-1}a_{\ell}\qtq{and}\\

1770: x_i^{w_{i+1}}x_{i+1}^{-w_i}&=& x_i^Dy_i^{-w_i}.

1771: \end{array}

1772: $$

1773: Then one can easily check that the monomials

1774: $x_i^D$ and $x_i^{w_{i+1}}x_{i+1}^{-w_i}$

1775: generate the subring $\c(x_1,\dots,x_n)^{(D)}$ of those

1776: elements whose degree is divisible by $D$.

1777: Thus we get an explicit isomorphism

1778: $$

1779: |H|^*:\c(y_1,\dots,y_n)\cong \c(x_1,\dots,x_n)^{(D)}.

1780: $$

1781: Nevertheless, I found it very difficult to compute

1782: anything based on these formulas.

1783: 

1784: For $n\geq 5$ the singular set of $\p(w_1,\dots,w_n)$

1785: has codimension $\geq 3$, thus the 

1786:  canonical class of $H$ is given by the adjunction formula,

1787: $$

1788: K_H=(K_{\p}+H)|_H=\o_{\p}(d-\sum w_i)|_H.

1789: $$

1790: The minor modifications needed in the

1791:  few cases when $n=4$ and $\p(w_1,\dots,w_n)$

1792: has 1--dimensional singular set are discussed in

1793: (\ref{rhcp2.say}).

1794: If $w^*=1$ then the two highest terms in the coefficient of $K_H$ are

1795: $$

1796: \bigl(\ \prod a_i\bigr) \bigl(1-\sum \tfrac1{a_i}\bigr),

1797: $$

1798: which is positive as soon as $\min\{a_i\}\geq n$. In fact it is easy to see

1799: that $d-\sum w_i>0$ if $\min\{a_i\}\geq n$.

1800: The self intersection of $K_H$ is computed asymptotically by 

1801: $$

1802: (K_H^{n-2})=(\o_{\p}(d-\sum w_i)^{n-2}\cdot \o_{\p}(d))

1803: \sim \frac{d^{n-1}}{\prod w_i}\sim w^*.

1804: $$

1805:  

1806: We compute the homology groups of $H$ using the

1807: Milnor--Orlik formula  \cite{mi-or}.

1808: 

1809: Let $f(x_1,\dots,x_n)$ be  a weighted homogeneous polynomial

1810: of weighted degree $d$ where the variable $x_i$ has weight $w_i$.

1811: Assume that $(f=0)$ has an isolated singularity at the origin

1812: and let $L=L(f):=(f=0)\cap S^{2n-1}(1)$ be its link.

1813: Then $L$ is $(n-3)$--connected and the rank of the

1814: middle homology groups is given by 

1815: $$

1816: \dim H_{n-2}(L,\q)=

1817: \sum_{I\subset \{1,\dots,n\}} (-1)^{n-|I|}

1818: \frac{\prod_{i\in I} (d/w_i)}{\lcm\{u_i:i\in I\}},

1819: $$

1820: where we write $\frac{d}{w_i}=\frac{u_i}{v_i}$ in lowest terms.

1821: 

1822: 

1823: In our case $(d,w_i)=1$ for every $i$  (\ref{num.conds}.5)

1824: an so $u_i=d$. Thus 

1825: $\lcm\{u_i:i\in I\}=d$ save for $I=\emptyset$. Thus

1826: the formula becomes

1827: $$

1828: \begin{array}{l}

1829: \sum_{I\subset \{1,\dots,n\}} (-1)^{n-|I|}

1830: \frac{\prod_{i\in I} (d/w_i)}{\lcm\{u_i:i\in I\}}=\\

1831: \qquad\qquad\qquad\qquad=(-1)^{n}+\frac{(-1)^{n-1}}{d}+

1832: \frac1{d}\sum_{I\subset \{1,\dots,n\}} (-1)^{n-|I|}

1833: \prod_{i\in I} \frac{d}{w_i}\\

1834: \qquad\qquad\qquad\qquad

1835: =(-1)^{n}+\frac{(-1)^{n-1}}{d}+\frac1{d}\prod_i\Bigl(\frac{d}{w_i}-1\Bigr)\\

1836: \qquad\qquad\qquad\qquad

1837: =(-1)^{n}+\frac{(-1)^{n-1}}{d}+

1838: \frac1{d}\prod_i\Bigl(\frac{a_{i-1}w_{i-1}}{w_i}\bigr)\\

1839: \qquad\qquad\qquad\qquad=(-1)^{n}+\frac{(-1)^{n-1}}{d}+\frac1{d}\prod_i a_i\\

1840: \qquad\qquad\qquad\qquad=(-1)^n+w^*,

1841: \end{array}

1842: $$

1843: where at the last step we took into account that 

1844: $dw^*=D=\prod_i a_i +(-1)^{n-1}$.

1845: 

1846: The link $L$ is a Seifert $S^1$-bundle

1847: over $X:=(f=0)\subset \p(w_1,\dots,w_n)$

1848: and the resulting Leray spectral sequence

1849: is  easy to compute (with rational coefficients),

1850: see, e.g. \cite{or-wa}. This gives

1851: (\ref{main.exmp.thm}.6--7) except for the

1852: precise identification of

1853: $H_{n-2}(H,\q)$ in the $n=2m$ case.

1854: 

1855: We use this only for $n=4$, where it is worked out in

1856: (\ref{rhcp2.say}).\qed

1857: 

1858: \begin{lem}\label{linalg.lem} Assume that $\prod a_i\neq (-1)^n$

1859: and write all subscripts modulo $n$. 

1860: Then  the system

1861: $$

1862: a_iv_i+v_{i+1}=1\quad i=1,\dots,n

1863: $$

1864: has determinant $\prod_{\ell=1}^n a_{\ell}+(-1)^{n-1}$

1865: and a unique solution  given by

1866: $$

1867: v_i=\frac{W_i}{D}:=

1868: \frac{\sum_{j=1}^n(-1)^{j-1}\prod_{\ell=i+j}^{i+n-1}a_{\ell}}

1869: {\prod_{\ell=1}^n a_{\ell}+(-1)^{n-1}}. \qed

1870: $$

1871: \end{lem}

1872: 

1873: 

1874: 

1875: Next we consider in greater detail the 

1876: two low dimensional cases.

1877: 

1878: 

1879: \begin{say}[Quasi--smooth rational curves]\label{qs.ratcurves} 

1880: 

1881: The case $n=3$ gives quasi--smooth rational curves

1882: in weighted projective planes.

1883: Here we have a system

1884: $$

1885: a_1w_1+w_2=a_2w_2+w_3=a_3w_3+w_1=d

1886: $$

1887: with solutions

1888: $$

1889: w_1=\frac{a_2a_3-a_3+1}{w^*},\

1890: w_2=\frac{a_3a_1-a_1+1}{w^*},\

1891: w_3=\frac{a_1a_2-a_2+1}{w^*},\

1892: d=\frac{a_1a_2a_3+1}{w^*}.

1893: $$

1894: We can also compute the genus of  

1895: the general member of the linear system

1896: $$

1897: C\in |x_1^{a_1}x_2, x_2^{a_2}x_3, x_3^{a_3}x_1|

1898: $$

1899: using the adjunction formula (\ref{all.qs.rcs}.4).

1900: By explicit computation

1901: $$

1902: \frac{d(d-w_1-w_2-w_3)}{w_1w_2w_3}=w^*-

1903: \frac1{w_1}-\frac1{w_2}-\frac1{w_3},

1904: $$

1905: Thus  by (\ref{all.qs.rcs}.4),

1906:  the genus of $C$ is  $(w^*-1)/2$.

1907: Thus $C$ is a smooth rational curve iff $w^*=1$,

1908: that is, when 

1909: $a_2a_3-a_3+1, a_3a_1-a_1+1$ and $ a_1a_2-a_2+1$

1910:  are relatively prime.

1911: 

1912: 

1913: The Kodaira dimension of the pair $(P:=\p(w_1,w_2,w_3),C)$

1914: is determined by the sign of 

1915: $$

1916: \begin{array}{rcl}

1917: \deg (C+K_{P})&=&a_1a_2a_3+1-(a_1a_2-a_2+1+ a_2a_3-a_3+1+a_3a_1-a_1+1)\\

1918: &=&

1919: (a_1-1)(a_2-1)(a_3-1)-1.

1920: \end{array}

1921: $$

1922: This is negative if one of $a_1,a_2,a_3$ is 1.

1923: If say $a_3=1$ then we get weighted projective planes

1924: $\p^2(a_2,1, a_1a_2-a_2+1)$ with only 2 singular points

1925: and the corresponding link is a lens space.

1926: 

1927: If $a_1=a_2=a_3=2$ then the relatively prime conditions is not satisfied.

1928: In all other cases $(a_1-1)(a_2-1)(a_3-1)-1>0$ so the Kodaira dimension is

1929: 2. \qed

1930: \medskip

1931: 

1932: As a side remark we note that 

1933: (\ref{H.cob.exmps}) lists all

1934: interesting quasi--smooth rational curves in

1935: weighted projective planes.

1936: 

1937: \begin{prop}\label{all.qs.rcs}

1938:  Let $P:=\p(u,v,w)$ be a well formed weighted

1939: projective plane and $C=C_d\subset P$ a quasi--smooth

1940: rational curve of degree $d$.

1941: Then, up to permuting the coordinates and

1942: isomorphism,  the pair $d,C$ is one of the following

1943: \begin{enumerate}

1944: \item $d=w$ and $C=(z=0)$.

1945: \item $d=u+v, w|u+v$ and $C\in |xy, z^{(u+v)/w}, \dots|$ where the

1946: existence of other degree $d$ monomials depends on further numerical

1947: coincidences.

1948: \item $(P,C)$ is as in (\ref{H.cob.exmps}).

1949: \end{enumerate}

1950: \end{prop}

1951: 

1952: Proof. Let $C\subset S$ be a quasi--smooth curve on a surface $S$

1953: which passes through the singular points $P_i$

1954: which are cyclic quotients by $\z/m_i$. The adjunction

1955: formula (cf.\ \cite{cort}) says that

1956: $$

1957: C(C+K_S)=2g(C)-2+\sum \bigl(1-\tfrac1{m_i}\bigr).

1958: \eqno{(\ref{all.qs.rcs}.4)}

1959: $$

1960: Assume now that $S=P$ is a weighted

1961: projective plane, $C$ is rational of degree $d$ and it

1962: passes through at most 2 singular points of indices $u,v$,

1963: where $u=1$ or $v=1$ are allowed. Then we get that

1964: $$

1965: \tfrac{d(d-u-v-w)}{uvw}=-\tfrac1{u}-\tfrac1{v}.

1966: $$

1967: Thus $d<u+v+w$ and so if $z$ is the coordinate with the

1968: biggest weight then it  appears in one of the monomials $z, z^2, zy,zx$.

1969: The rest follows by an easy case analysis, giving the

1970: first two possibilities.

1971: 

1972: It remains to consider the case when $C$ passes through

1973: all 3 singular points and $u,v,w\geq 2$. This gives the equation

1974: $$

1975: \tfrac{d(d-u-v-w)}{uvw}=1-\tfrac1{u}-\tfrac1{v}-\tfrac1{w}.

1976: $$

1977: Aside from the case $\{u,v,w\}=\{2,3,5\}$, the right hand side

1978: is positive and there is no easy upper bound for $d$.

1979: 

1980: The quasi--smoothness conditions 

1981: show that, up to permuting the coordinates,

1982: we have monomials

1983: $$

1984: x^ay,y^bz,z^cx\qtq{or}  x^ay, y^bx, z^cx.

1985: $$

1986: The first of these leads to (\ref{H.cob.exmps}) and to our

1987: last possibility. (In fact one can check that

1988: in the 3 singular point case, there are no other monomials

1989: with the same weighted degree.)

1990: 

1991: Finally we  exclude

1992: the case $x^ay, y^bx, z^cx$. All of these are divisible

1993: by $x$, thus we also must have another monomial

1994: $z^{c'}y$. Thus $(c-c')w=v-u$ and so $w|u-v$.

1995: Up to interchanging $x,y$ we can assume that $u\geq w+1$.

1996: 

1997: From $d=ua+v=bv+u$ we obtain that $(a-1)u=(b-1)v$. Since

1998: $(u,v)=1$ we get that $d=muv+u+v$ for some $m\geq 1$.

1999: Substituting into the genus formula and rearranging we get

2000: $(mu+1)(mv+1)=(m+1)w$.  But

2001: $mu+1\geq u+1>w$ and $mv+1\geq 2m+1$ give a contradiction.\qed

2002: \end{say}

2003: 

2004: 

2005: \begin{say}[Examples of rational homology $\c\p^2$-s]\label{rhcp2.say}

2006: 

2007: For $n=4$ we get

2008: a surface 

2009: $$

2010: S=S(a_1,a_2,a_3,a_4):=(x_1^{a_1}x_2+x_2^{a_2}x_3+x_3^{a_3}x_4+x_4^{a_4}x_1=0)

2011: \subset \p(w_1,w_2,w_3,w_4),

2012: $$

2013: where the $a_i$ and $w_i$ satisfy a system of equations

2014: $$

2015: a_1w_1+w_2=a_2w_2+w_3=a_3w_3+w_4=a_4w_4+w_1=d

2016: $$

2017: with solutions

2018: $$

2019: w_1=\frac{a_2a_3a_4-a_3a_4+a_4-1}{w^*}, \dots\qtq{and} 

2020: d=\frac{a_1a_2a_3a_4-1}{w^*}.

2021: $$

2022: If $w^*=1$ then

2023: $S(a_1,a_2,a_3,a_4)$ is a rational surface

2024: with 4 quotient  singularities at the coordinate vertices

2025: and with $H_2(S,\q)\cong \q^3$.

2026: 

2027: 

2028: 

2029: Note that $S$ contains the two rational curves

2030: $$

2031: C_1:=(x_1=x_3=0)\qtq{and} C_2:=(x_2=x_4=0).

2032: $$

2033: Both of these are quasi--smooth in $S$.

2034: Thus by the adjunction formula (\ref{all.qs.rcs}.4),

2035: $$

2036: (K_S+C_1)\cdot C_1=-\frac1{w_2}-\frac1{w_4}\qtq{and}

2037: (K_S+C_2)\cdot C_2=-\frac1{w_1}-\frac1{w_3}.

2038: $$

2039: This implies that both curves 

2040: have negative intersection with 

2041: $K_S+(1-\epsilon)(C_1+C_2)$ for $0<\epsilon\ll 1$, and so they are

2042: are extremal rays for

2043: the $K_S+(1-\epsilon)(C_1+C_2)$ minimal model program.

2044: (See \cite{km-book} for an introduction.)

2045: Thus 

2046: $C_1$ and $C_2$ are both contractible

2047: to quotient singularities and we get

2048: rational surfaces

2049: $$

2050: \pi:S\to S^*=S^*(a_1,a_2,a_3,a_4).

2051: $$

2052: 

2053: If the $\{w_i\}$ are pairwise relatively prime, then

2054: the canonical class of $S$ is

2055: $$

2056: K_S=\o_{\p}(\textstyle{\prod} a_i-1-\textstyle{\sum} w_i).

2057: $$

2058: If the pairwise relatively prime assumption fails

2059: then the general adjunction formula  \cite{cort} says that

2060: $$

2061: K_S+\Bigl(1-\frac1{\gcd(w_1,w_3)}\Bigr)C_1

2062: +\Bigl(1-\frac1{\gcd(w_2,w_4)}\Bigr)C_2

2063: =\o_{\p}(\textstyle{\prod} a_i-1-\textstyle{\sum} w_i).

2064: $$

2065: 

2066: 

2067: 

2068: Note that if $a_1,a_2,a_3,a_4\geq 4$ then

2069: $K_{S^*}$ is ample. 

2070: One can write down an explicit formula for the

2071: self intersection of $K_{S^*}$, but it

2072: is rather complicated. In any case, one sees that

2073: it also converges to 1 as $a_1,a_2,a_3,a_4\to\infty$.

2074: \end{say}

2075: 

2076: 

2077: 

2078: 

2079: \begin{ack}  I thank V.\ Alexeev, D.\ Auroux, J.\ Amoros, 

2080: W.\ Chen, I.\ Dolgachev,

2081: C.\ Gordon, 

2082: J.\ Keum, T.\ Mrowka, A.\ N\'emethi and D.-Q.\ Zhang

2083: for useful comments  and especially 

2084: M.\ Zaidenberg who called my attention to numerous

2085: related papers.

2086: I am grateful to P.\ Hacking and J.\ Wahl

2087: for allowing me to use their unpublished results.

2088: J.M.\ Johnson helped with computer experiments

2089: that lead to the discovery of various properties of the

2090: examples in Section 5.

2091: Partial financial support was provided by  the NSF under grant number 

2092: DMS-0500198. 

2093: \end{ack}

2094: 

2095: 

2096: \bibliography{refs}

2097: 

2098: \bigskip

2099: 

2100: \noindent Princeton University, Princeton NJ 08544-1000

2101: 

2102: \begin{verbatim}kollar@math.princeton.edu\end{verbatim}

2103: 

2104: 

2105: \end{document}

2106: 

2107: 

2108: 

2109: 

2110: 

2111: 

2112: \end{document}

2113: 

2114: 

2115: 

2116: 

2117: 

2118: 

2119: 

2120: 

2121: 

2122: \begin{say}\label{def.tor-stor}

2123: [Torus and solid tous]{\ }

2124: 

2125: As a general reference, see \cite[Chap.2]{Rolfsen76}.

2126: 

2127: Let $S^1$ be the unit circle $(|u|=1)\subset \c_u$ and $D^2$ the unit

2128: disc

2129: $(|z|\leq 1)\subset \c_z$. 

2130: $S^1\times D^2$ is called the {\it solid torus}. Its boundary

2131: $$

2132: \partial (S^1\times D^2)=(|u|=1)\times (|z|=1)\sim S^1\times S^1

2133: $$

2134: is a {\it torus}. Up to isotopy, the homeomorphisms of a torus are given

2135: by

2136: $$

2137: (u,z)\mapsto (u^az^b,u^cz^d),\qtq{where}

2138: \left(

2139: \begin{array}{cc}

2140: a& b\\

2141: c&d

2142: \end{array}

2143: \right)

2144: \in GL(2,\z).

2145: $$

2146: Up to isotopy, the homeomorphisms of a solid torus are given by

2147: $$

2148: (u,z)\mapsto (u^{\pm 1},u^mz)\qtq{or}

2149: (u,z)\mapsto (u^{\pm 1},u^m\bar z),\qtq{where} m\in\z.

2150: $$

2151: Let $g:S^1\times S^1\to S^1$ be an $S^1$-bundle. Up to isotopy,

2152: it can be written as

2153: $$

2154: g_{c,d}:(u,z)\mapsto u^cz^d, \qtq{where} (c,d)=1.

2155: $$

2156: Given a solid torus $S^1\times D^2$, let 

2157: $g_{c,d}:\partial (S^1\times D^2)\to S^1$ be an $S^1$-bundle.

2158: This can be extended to a  map

2159: $$

2160: G_{c,d}:S^1\times D^2\to D^2 \qtq{by}

2161: (u,z)\mapsto

2162: \left\{

2163: \begin{array}{lll}

2164: u^cz^d&\mbox{if}& d> 0,\\

2165: u^c\bar z^{(-d)}&\mbox{if}& d< 0,\\

2166: u^c|z| &\mbox{if}& d= 0.

2167: \end{array}

2168: \right.

2169: $$

2170: The fiber $G_{c,d}^{-1}(1)$ is parametrized by

2171: $$

2172: \phi:S^1\to S^1\times D^2\qtq{given as} t\mapsto (t^d,t^{-c}).

2173: $$

2174: ($\phi$ is  injective since $(d,c)=1$.) Composing $\phi$ with the

2175: first projection $S^1\times D^2\to S^1$ we obtain a map of degree $m$

2176: from $S^1$ to $S^1$. Thus the fiber of $G_{c,d}$ over the origin

2177: has multiplicity $|d|$ (\ref{mult.fibr.def}). 

2178: 

2179: 

2180: Applying a homeomorphism of the solid torus, we can get

2181: every $G_{c,d}$ to

2182: one of the following normal forms:

2183: \begin{enumerate}

2184: \item 

2185: For a pair of integers $c,d$ satisfying  $0\leq c<d$ and

2186: $(c,d)=1$, define

2187: $$

2188: f_{c,d}:S^1\times D^2\to D^2\qtq{by}

2189:  f_{c,d}(u,z)=u^cz^d.

2190: $$

2191: $f_{c,d}$ restricts to a fiber bundle

2192: $S^1\times (D^2\setminus \{0\})\to D^2\setminus \{0\}$.

2193: The fiber of $f_{c,d}$ over the origin is still $S^1$, but 

2194: $f_{c,d}^{-1}(0)$ has

2195: multiplicity $d$. 

2196: \item $f_{1,0}(u,z)=u\cdot |z|$.  In this case the fibers

2197: $f_{1,0}^{-1}(t)$ for $t\neq 0$ are null homotopic in 

2198: the solid torus. 

2199: 

2200: Instead of $f_{1,0}$, I prefer to use the map

2201: $$

2202: f_0:(u,z)\mapsto u\frac{1+|z|}{2}\in \{t\vert 1/2\leq |t|\leq 1\}\subset

2203: \c_t.

2204: $$

2205: The image is a closed annulus  $A_{1/2,1}$ with boundary circles

2206: $C_{1/2}$ and $C_1$. 

2207: $f_0$ is an $S^1$-bundle over the annulus $A_{1/2,1}\setminus C_{1/2}$

2208: and $f_0^{-1}(C_{1/2})\to C_{1/2}$ is a homeomorphism.

2209: 

2210: $f_{1,0}$ is obtained by composing $f_0$ with $t\mapsto

2211: 2t(1-\frac{1}{2|t|})$ which contracts the circle $C_{1/2}$ to a point. 

2212: \end{enumerate}

2213: \end{say}

2214: 

2215: 

2216: 

2217: