1: 

2:  \documentclass{amsart}

3: 

4: 

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

6: 

7: 

8: 

9: %\documentclass[12pt,leqno,amssymbols]{amsbook}

10: 

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

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

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

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

15: 

16: 

17: 

18: \usepackage{amssymb}

19: 

20: 

21: 

22: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

23: 

24: % Definitions for new environments

25: % theorem style plain --- default

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

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

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

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

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

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

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

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

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

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

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

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

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

39:    

40: \theoremstyle{definition}

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

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

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

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

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

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

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

48: \newtheorem{probs}[thm]{Problems}

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

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

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

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

53: %\renewcommand{\theremark}{} 

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

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

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

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

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

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

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

61: 

62: \theoremstyle{remark}

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

64: \newtheorem{case}{Case} 

65: \newtheorem{subcase}{Subcase}   

66: \newtheorem{step}{Step}

67: \newtheorem{approach}{Approach}

68: %\newtheorem{principle}{Principle}

69: \newtheorem{fact}{Fact}

70: \newtheorem{subsay}{}

71: 

72: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

73: \setcounter{section}{0}

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

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

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

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

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

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

80: %\pagestyle{empty}

81: 

82: 

83: \let \cedilla =\c

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

85: \let \crossedo =\o

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

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

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

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

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

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

92: 

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

94: 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

109: 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

131: \newcommand{\res}[0]{\operatorname{Res}} 

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

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

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

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

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

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

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

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

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

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

142: 

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

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

145: 

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

147: 

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

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

150: 

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

152: 

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

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

155: 

156: 

157: 

158: 

159: 

160: %\UseAMSsymbols

161: 

162: %\DeclareMathSymbol\subsetneq 2328

163: %\DeclareMathSymbol\onto 1310

164: 

165: %\DeclareMathSymbol\twoheadrightarrow 1310

166: %\DeclareMathSymbol\rtimes 226F

167: %\DeclareMathSymbol\ltimes 226E

168: 

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

170: 

171: 

172: \begin{document}

173: \bibliographystyle{amsalpha}

174: 

175: \title{Positive Sasakian structures on  5-manifolds}

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

177: %\today

178: \maketitle

179: 

180: A quasi--regular Sasakian structure on a manifold $L$

181: is equivalent to writing $L$ as the unit circle subbundle

182: of a holomorphic Seifert $\c^*$-bundle over a complex algebraic

183: orbifold $(X,\Delta=\sum (1-\frac1{m_i})D_i)$. The Sasakian structure is called

184: positive if the orbifold first Chern class 

185: $c_1(X)-\Delta=-(K_X+\Delta)$ is positive. We are especially interested

186: in the case when the Riemannian metric part of the 

187: Sasakian structure is Einstein. By a result of

188: \cite{kob, bg00}, this happens iff

189: \begin{enumerate}

190: \item $-(K_X+\Delta)$ is positive, 

191: \item the first Chern class of the Seifert bundle

192: $c_1(L/X)$ is a rational multiple of $-(K_X+\Delta)$, and

193: \item there is an orbifold K\"ahler--Einstein metric on $(X,\Delta)$.

194: \end{enumerate}

195: If the first two conditions  hold, we say that

196: $f:L\to (X,\Delta)$ is  a {\it pre-SE} 

197: (or pre--Sasakian--Einstein) Seifert bundle.

198: 

199: Note that if $H_2(L,\q)=0$ then $H_2(S,\q)\cong \q$

200: and so  the second condition is automatic.

201: 

202: 

203: 

204: While it is not true that all pre-SE Seifert bundles

205:  carry a 

206: Sasakian--Einstein  structure, all known topological obstructions

207: to the existence of a Sasakian--Einstein  structure

208: in dimension 5 are  consequences of the pre-SE condition.

209: 

210: 

211: A 2--dimensional orbifold $(S,\Delta)$

212: such that $-(K_S+\Delta)$ is positive is also called

213: a {\it log Del Pezzo surface}. For any such, $S$ is a rational surface

214: with quotient singularities. 

215: By \cite[2.4]{em5}, Seifert $\c^*$-bundles

216: over $(S, \sum (1-\frac1{m_i})D_i)$ are

217: uniquely classified by a homology class

218: $B\in H_2(S,\z)$ and integers

219: $0< b_i<m_i$ with $(b_i,m_i)=1$.

220: We denote the corresponding Seifert  $\c^*$-bundle 

221: (resp.\ $S^1$-bundle) by

222: $$

223: Y\bigl(S,B, \textstyle{\sum} \tfrac{b_i}{m_i}D_i\bigr)

224: \qtq{resp.} L\bigl(S, B, \textstyle{\sum} \tfrac{b_i}{m_i}D_i\bigr).

225: $$

226: 

227: \begin{say}[Classification problems] Following

228: \cite{em5}, we want to address 3 problems.

229: \begin{enumerate}

230: \item Given $(S,\Delta)$ and $B$, determine the 5--manifold

231:  $L(S,B,\sum \tfrac{b_i}{m_i}D_i)$.

232: \item Given a 5--manifold $L$, decide if it can be

233: written as $L(S,B,\sum \tfrac{b_i}{m_i}D_i)$ for some  $(S,\Delta)$ and $B$.

234: \item Given a 5--manifold $L$, describe all of its representations

235:  as $L(S,B,\sum \tfrac{b_i}{m_i}D_i)$ for some  $(S,\Delta)$ and $B$.

236: \end{enumerate}

237: 

238: The case when $L$ is simply connected was studied in

239: \cite{em5}. 

240: Problem (1) was solved in \cite[5.7]{em5}.

241: As to Problem (2), note that

242: by a result of \cite{smale}, such  manifolds are

243: determined by the second homology group $H_2(L,\z)$

244: if $w_2=0$. (See \cite{barden} for the $w_2\neq 0$ case.)

245: The torsion part of $H_2(L,\z)$ can be completely described 

246: as follows.

247: (There are partial results about the rank of the free part.)

248: 

249: 

250: \begin{thm}\label{h2.tors.thm} \cite[1.4]{em5}

251: Let $L$ be a compact 5--manifold  

252: such that $H_1(L,\z)=0$.

253: Assume that $L$ has a positive Sasakian structure

254: $f:L\to (S,\Delta)$. 

255: Then  the torsion subgroup of the second homology, $\tors H_2(L,\z)$,

256: is determined by $(S,\Delta)$ and it

257: is one of the following.

258: \begin{enumerate}

259: \item $(\z/m)^2$ for some $m\geq 1$,

260: \item $(\z/5)^4$ or $(\z/4)^4$,

261: \item $(\z/3)^4,(\z/3)^6$ or $(\z/3)^8$,

262: \item $(\z/2)^{2n}$ for some $n\geq 1$.

263: \end{enumerate}

264: \end{thm}

265: 

266: Note that the theorem gives restrictions on $L$ but does not say anything about

267: $(S,\Delta)$. 

268: While it  seems hopelessly complicated to describe all

269: Seifert bundle  structures on $S^5$, there  are 

270: few positive Sasakian structures on  complicated

271: 5--manifolds. In particular, Problem (3) was

272: settled for 

273:  $\tors H_2(L,\z)\cong (\z/5)^4$  \cite[1.8.2]{em5}.

274: \end{say}

275: 

276:  

277: 

278: The aim of this note is fourfold. First, 

279:  we extend several of the results to

280:  5--manifolds which are not simply connected. 

281: Second, I would like to correct

282: the mistake in the classification of the case 

283: (\ref{h2.tors.thm}.1) which was brought to my attention by

284: K.\ Galicki. Third, Problem (3) is solved 

285: in  the  $(\z/4)^4$ and    $(\z/3)^8$  cases.

286: Finally, we show that in in many cases, all positive Sasakian

287: structures are given by hypersurfaces in $\c^4$

288: and  also give explicit equations

289: for these. 

290: 

291: Toward the first goal, we show that the groups 

292: listed in (\ref{h2.tors.thm}.1--4) give large subgroups  of all possible

293: $\tors H_2(L,\z)$.

294: 

295: \begin{thm} \label{h2.tors.new.thm}

296: Let $L$ be a compact 5--manifold with a positive

297:  Sasakian structure. Then $\tors H_2(L,\z)$ has a subgroup

298: $G$ as in (\ref{h2.tors.thm}.1--4) such that the quotient group

299:  $\tors H_2(L,\z)/G$ is

300:   $\z/r$, $\z/r+\z/2$ or $\z/r+\z/3$  for some $r$.

301: \end{thm}

302: 

303: Our next aim is to prove that in most cases

304: the existence of the subgroup $G$ controls 

305: $L$ and the Sasakian structure very tightly.

306: 

307: \begin{thm}\label{h2.tors.gen.cases} 

308: Let $f:L\to (S,\Delta)$ be a compact 5--manifold with a positive

309:  Sasakian structure.

310:  Assume that  $H_2(L,\z)\supset (\z/m)^2$ 

311: where  $m\geq  28$ or $m\geq 12$ and $(m,6)=1$.

312: Then 

313: \begin{enumerate}

314: \item $S$ is a Del Pezzo surface with cyclic Du Val singularities only

315: and $\Delta=(1-\frac1{m'})C$ where $C\in |-K_S|$ is a smooth elliptic curve

316: and $m'$ is a multiple of $m$.

317: \item there are 132 families of such $(S,C)$ up to deformations that

318: preserve the singularities,

319: \item for every such $S$,

320: $\pi_1(S^0)$ is abelian of order $\leq 9$,

321: where $S^0:=S\setminus \sing S$ denotes the set of smooth points.

322: $\pi_1^{orb}(S,(1-\frac1{m'})C)$ is also abelian of order $\leq 9$,

323: save 3 cases which are nonabelian of order $2^3, 2^4$ or $3^3$.

324: The precise values are in Table 2.

325: \item The fundamental group of $L$ is given by  a central extension

326: $$

327: 0\to \z/d\to \pi_1(L)\to \pi_1^{orb}(S,(1-\tfrac1{m'})C) \to 1,

328: $$

329: for some $d$.

330: \item The torsion of the second homology group of $L$  

331: sits in an exact sequence

332: $$

333: 0\to (\z/m')^2\to \tors H_2(L,\z)\to  \pi_1(S^0).

334: $$

335: \end{enumerate}

336: \end{thm}

337: 

338: It is quite likely that the above theorem holds for all

339: $m\geq 12$. However, for $m<12$, there are many counter examples,

340: see \cite[6.6--7]{em5}.

341: 

342: \begin{say}[Sasakian--Einstein structures]  In each of the 132 cases the

343: existence of positive Sasakian and of pre--Sasakian--Einstein structures

344: over $(S,\Delta)$ is  effectively decidable, but some of the

345: computations may be lengthy.

346: 

347: There are 93 cases of $(S,C)$ when $\pi_1(S^0)=1$. 

348: In \cite[1.8.1]{em5} I claimed, incorrectly, that they all

349: admit a Seifert bundle with Sasakian--Einstein structure.

350: We see in (\ref{corr181.say})

351: that these all admit Seifert bundles

352: with a positive Sasakian  structure,  but

353: only 19 of them admit a Seifert bundle with pre--SE structure.

354: 

355: For the remaining 39 cases 

356: with $\pi_1(S^0)\neq 1$

357: my computations are not yet complete.

358: There are some cases that do not have any smooth Seifert bundles

359: over them (\ref{sm.sb.ex.say}), and in many other cases

360: Sasakian--Einstein structures exist.

361: \end{say}

362: 

363: \begin{say}[Hypersurface examples]

364: Concrete examples of Sasakian structures

365: can be obtained using $\c^*$-actions

366: on algebraic hypersurfaces.

367: A linear $\c^*$-action on $\c^n$ can be diagonalized

368: and so given by weights $(w_1,\dots,w_n)$.

369: We consider only actions which are effective  (thus the weights

370: are relatively prime)

371: and  the origin is an attracting fixed point

372: (thus $w_i> 0$ for every $i$).

373: Let $Y\subset \c^n$ be an algebraic variety, smooth away from the origin,

374: which is invariant under the $\c^*$-action.

375: Then its link

376: $L:=Y\cap S^{2n-1}(1)$ inherits a natural quasi--regular Sasakian

377: structure and every quasi--regular Sasakian

378: structure arises this way.

379: 

380: This description is especially simple when $Y$ is a hypersurface.

381: In this case everything is described

382: by the weights $(w_1,\dots,w_n)$ and a weighted homogeneous

383: polynomial $f(x_1,\dots,x_n)$ which defines $Y$.

384: 

385: Thus, up to deformations, we need to specify

386: the weights $(w_1,\dots,w_n)$ and the weighted degree

387: $d$ of $f$. 

388: I write

389: $L^*(w_1,\dots,w_n;d)$ for any such quasi--regular Sasakian

390: manifold. (The $^*$ is to remind one that using weights

391: is in some sense dual to the notation in \cite{bgk}.)

392: 

393: The simplest examples where $\pi_1(L)\neq 1$

394: are created by taking a quotient by a subgroup

395: of  $\c^*$. 

396: These are all cyclic. Thus the symbol

397: $$

398: L^*(w_1,\dots,w_n;d)/(\z/m)\qtq{or}

399: L^*(w_1,\dots,w_n;d)/\tfrac1{m}(w_1,\dots,w_n)

400: $$

401: stands for any Sasakian manifold obtained

402: as $L/(\z/m)\subset Y/(\z/m)$ where

403: $Y:=(f=0)$ is the zero set of a degree $d$ weighted homogeneous

404: polynomial such that $Y$ is smooth outside the origin

405: and we take the quotient by the $\z/m$-action generated by

406: $$

407: (x_1,\dots,x_n)\mapsto (\epsilon^{w_1}x_1,\dots,\epsilon^{w_n}x_n)

408: \qtq{where} \epsilon=e^{2\pi i/m}.

409: $$

410: We call these the {\it obvious quotients}.

411: 

412: One can also take quotients by groups that are

413: not contained in $\c^*$. For instance, one can take

414: quotients

415: $$

416: L^*(w_1,\dots,w_n;d)/\bigl(\tfrac1{r}(a_1,\dots,a_n)\times

417: \tfrac1{m}(w_1,\dots,w_n)\bigr)

418: $$

419: where the first factor of $\z/r\times \z/m$

420: acts via

421: $$

422: (x_1,\dots,x_n)\mapsto (\eta^{a_1}x_1,\dots,\eta^{a_n}x_n)

423: \qtq{where} \eta=e^{2\pi i/r},

424: $$

425: and the second factor acts as above.

426: One should keep in mind that not all hypersurfaces

427: $L^*(w_1,\dots,w_n;d)$ admit such a $\z/r$-action.

428: In using this notation, we assume that we consider a case

429: when the action exists.

430: \end{say}

431: 

432: We are now ready to state that classification theorem

433: for those cases when $H_2(L,\z)$ contains a large torsion subgroup.

434: 

435: \begin{thm} \label{main.thm} 

436: Let $L$ be a compact 5--manifold with a positive

437:  Sasakian

438: structure.  

439: \begin{enumerate}

440: \item  If $H_2(L,\z)\supset (\z/5)^4$ then 

441: $H_2(L,\z)= (\z/5)^4$ and 

442: $L=L^*(5,5,15,6;30)$ or an obvious quotient by $\z/m$ for  $(m,15)=1$.

443:  The simplest equation is

444: $$

445: (x^{6}+y^{6}+z^2+t^{5}=0)/\tfrac1{m}(5,5,15,6).

446: $$

447: The moduli space of Sasakian--Einstein structures is naturally

448: parametrized by the moduli space of  genus 2 curves.

449: 

450: \item  If $H_2(L,\z)\supset (\z/3)^8$ then 

451:  $H_2(L,\z)=(\z/3)^8$ and 

452: $L=L^*(3,3,15,10;30)$ or an obvious quotient by $\z/m$ for  $(m,15)=1$.

453:  The simplest equation is

454: $$

455: (x^{10}+y^{10}+z^2+t^{3}=0)/\tfrac1{m}(3,3,15,10).

456: $$

457:  The moduli space  of Sasakian--Einstein structures  is naturally

458: parametrized by the moduli space of  hyperelliptic curves of  genus 4.

459: 

460: 

461: \item If $H_2(L,\z)\supset (\z/4)^4$ then 

462: either $H_2(L,\z)= (\z/4)^4$ or  $H_2(L,\z)= \z+(\z/4)^4$.

463: In the first case, there are 5  families: 

464: \begin{enumerate}

465: \item $L=L^*(4,4,8,5;20)$ or an obvious quotient

466: by $\z/m$ for $(m,2)=1$,  

467:  with sample equation 

468: $$

469: (x^{5}+y^{5}+yz^2+t^{4}=0)/\tfrac1{m}(4,4,8,5),

470: $$

471: \item $Y=L^*(2,2,6,3;12)/\frac12(1,1,1,e)$

472: where $e\in \{0,1\}$ or an obvious quotient by $\z/m$ for  $(m,6)=1$.

473: The simplest equation is

474: $$

475: (x^6+y^6+z^2+t^4=0)/\bigl(\tfrac12(1,1,1,e)\times \tfrac1{m}(2,2,6,3)\bigr).

476: $$ 

477: 

478: \item 

479:  $L=L^*(4,8,20,9;40)$ or an obvious quotient by $\z/m$ for  $(m,6)=1$.

480: The simplest equation is

481: $$

482: (x^{10}+y^{5}+z^2+xt^{4}=0)/\tfrac1{m}(4,8,20,9).

483: $$

484: \item $L=L^*(2,4,10,5;20)/\frac12(1,0,1,1)$

485: or an obvious quotient by $\z/m$ for  $(m,10)=1$.

486: The simplest equation is

487: $$

488: (x^{10}+y^{5}+z^2+t^4=0)/

489: \bigl(\tfrac12(1,0,1,1)\times \tfrac1{m}(2,4,10,5)\bigr).

490: $$

491: \end{enumerate}

492: 

493: In the second case,

494: there are infinitely many

495: positive Sasakian structures  with the same $\pi_1$ but only one

496: which is pre--SE. These are:

497: \begin{enumerate}\setcounter{enumii}{4}

498: \item  

499: $L=L^*(4,4,12,5;24)$ 

500: or an obvious quotient by $\z/m$ for  $(m,10)=1$.

501: The simplest equation is

502: $$

503: (x^{6}+y^{6}+z^2+xt^4=0)/\tfrac1{m}(4,4,12,5).

504: $$

505: \end{enumerate}

506: The moduli space  of Sasakian--Einstein structures  is naturally

507: parametrized by the moduli space of  pairs

508: $(C,p)$ where $\pi:C\to \p^1$ is a genus 2 curve and $p\in C$ is not a

509: branch point in cases (a),(b),(e) and  $p\in C$ is  a

510: branch point in cases (c),(d).

511: \end{enumerate}

512: \end{thm}

513: 

514: The above parametrizations of the moduli spaces of 

515: Sasakian--Einstein structures are one--to--one.

516: However, if we  consider the moduli spaces of

517: the resulting Einstein metrics, we get a two--to--one

518: parametrization since a curve and its complex conjugate

519: give the same Einstein metric, see \cite[18--21]{bgk}.

520: 

521: I see no a priori reason why all these cases should be

522: realizable as hypersurface quotients. K.\ Galicki  told

523:  me that he found the

524: example $(x^{6}+y^{6}+z^2+t^5=0)$

525:  with  $H_2(L,\z)=(\z/5)^4$.

526: This lead me to realize that in fact all  examples with 

527: $H_2(L,\z)=(\z/5)^4$

528: are hypersurfaces. The equations are also easy to see

529: in the $(\z/3)^8$ case, but are rather mysterious

530: from the point of view of my proof for several

531: of the $(\z/4)^4$ cases.

532: 

533: 

534: 

535: \section{Reduction to algebraic geometry}

536: 

537: 

538: \begin{say}\label{algred.say} As in  \cite{bgk, em5},

539: to $L=Y\cap S^{2n-1}(1)$ we associate a projective

540: algebraic variety $X=Y\setminus\{0\}/\c^*$

541: with cyclic quotient singularities. There is also

542: a natural $\q$-divisor $\Delta=\sum (1-\tfrac1{m_i})D_i$

543: on $X$ where $D_i\subset X$ is a divisor such that

544: the stabilizer of the  $\c^*$-action has order

545: $m_i$ over the points of  $D_i$.

546: The positivity of the Sasakian structure is

547: equivalent to the log Fano condition:

548: $$

549: -\bigl(K_X+\sum (1-\tfrac1{m_i})D_i\bigr)\qtq{is ample.}

550: $$

551: Thus if $\dim L=5$, we are looking for

552: pairs $(S,\Delta)$

553: where 

554: \begin{enumerate}

555: \item $S$ is a projective surface with cyclic quotient singularities,

556: \item $\Delta=\sum (1-\tfrac1{m_i})D_i$ is a $\q$-divisor, and

557: \item $-(K_S+\Delta)$ is ample.

558: \end{enumerate}

559: There are a few more conditions which we do not need for now.

560: 

561: A rather easy result \cite[6.3]{em5} on log Del Pezzo surfaces 

562: gives the following restriction on the $D_i$:

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

564: \item there is at most one index $i$ such that $g(D_i)>0$.

565: \end{enumerate}

566: Because of the special role of $D_i$, we 

567: frequently set $D:=D_i$, $r:=m_i$ and 

568: use the notation $(S,(1-\tfrac1{r})D+\Delta)$.

569: 

570: It turns out that the ampleness of $-(K_S+(1-\tfrac1{r})D+\Delta)$ forces

571: $g(D)$ or $r$ to be quite small and this accounts for

572:  the restrictions in Theorem \ref{h2.tors.thm}.

573: 

574: This is relevant to our purposes since

575: \cite[5.7]{em5} 

576:  relates the torsion of $H_2(L,\z)$

577: to the divisor $\sum (1-\tfrac1{m_i})D_i$:

578: 

579: \begin{enumerate}\setcounter{enumi}{4}

580: \item If $H_1(L,\z)=0$ then $\tors H_2(L,\z)=\sum (\z/m_i)^{2g(D_i)}$.

581: Thus, in the log Del Pezzo case,  $\tors H_2(L,\z)\cong (\z/r)^{2g(D)}$.

582: \end{enumerate}

583: 

584: 

585: 

586: 

587: If $H_1(L,\z)\neq 0$, the spectral sequence  in \cite[5.10]{em5}

588: becomes quite messy, but in the  log Del Pezzo case only

589: one nonzero map involves $H^3(L,\z)$.

590: The same proof gives the following weaker result

591: when the  first ordinary homology of the base is trivial.

592: \end{say}

593: 

594: \begin{prop} \label{thm2.pf} 

595: Let $f:L^5\to (S,\Delta=\sum (1-\frac1{m_i})D_i)$

596: be a smooth Seifert bundle over  a projective surface with

597: quotient singularities. Assume that $H_1(S,\z)=0$ and

598: let $s=\rank H^2(S,\q)$. Then

599: there is an exact sequence

600: $$

601: H_1(S^0,\z)\cong H^3(S,\z)\to H^3(L,\z)\to \z^{s-1}+

602: \sum (\z/m_i)^{2g(D_i)}\to 0. 

603: \eqno{(\ref{thm2.pf}.1)}

604: $$

605: \end{prop}

606: 

607: 

608: Proof.

609:  The argument very closely follows

610: \cite[5.9--10]{em5}.  

611: As there, we have exact sequences

612: $$

613: 0\to R^1f_*\z_L\stackrel{\tau}{\to} \z_S \to Q\to 0

614: \eqno{(\ref{thm2.pf}.2)}

615: $$

616: and 

617: $$

618: 0\to \sum_i \z_{P_j}/n_j \to Q \to \sum_i \z_{D_i}/m_i\to 0,

619: \eqno{(\ref{thm2.pf}.3)}

620: $$

621: where $P_j\in S$ are the singular points.

622: This implies that

623: $H^i(S,Q)=\sum_i H^i(D_i,\z/m_i)$ for $i\geq 1$.

624: The key piece of the long cohomology sequence of (\ref{thm2.pf}.2)

625:  is

626: $$

627: H^1(S,\z)\to \sum H^1(D_i,\z/m_i)\to H^2(S,R^1f_*\z_L)

628: \to H^2(S,\z) \to \sum H^2(D_i,\z/m_i)

629: $$

630: Here $H^1(S,\z)=0$ and $H^2(S,\z)\cong \z^s$

631: by assumption.

632: The right hand group is torsion, hence

633: there is a noncanonical isomorphism

634: $$

635: H^2(S,R^1f_*\z_L)\cong \z^s+\sum_i H^1(D_i,\z/m_i).

636: $$

637: Therefore, in

638:   the Leray spectral sequence

639: $H^i(S,R^jf_*\z_L)\Rightarrow H^{i+j}(L,\z)$ the $E_2$ term is

640: $$

641: \begin{array}{lccll}

642: \z \quad & (\mbox{torsion}) & \quad\z^{s}+\sum_i (\z/m_i)^{2g(D_i)}\quad & 

643: (\mbox{torsion})\quad & \z\\

644: \z & 0 & \z^{s} & H^3(S,\z) & \z.

645: \end{array}

646: \eqno{(\ref{thm2.pf}.4)}

647: $$

648: Since $H^3(S,\z)\cong H_1(S^0,\z)$ by \cite[4.2]{em5},

649: we get the exact sequence

650: $$

651: H_1(S^0,\z)\to \tors H^3(L,\z)\to \sum_i H^1(D_i,\z/m_i)\to 0. \qed

652: $$

653: 

654: 

655: \begin{cor} Notation and assumptions as in (\ref{thm2.pf}).

656: Let $p_j\in S$ be the singular points and $n_j$ the order of the local

657: fundamental group.

658: \begin{enumerate}

659: \item If any two of the numbers $m_i, n_j$ are relatively prime

660: then the sequence (\ref{thm2.pf}.1) is left exact.

661: \item If, in addition  $|H_1(S^0,\z)|$ is relatively prime to $\prod m_i$

662: then the sequence (\ref{thm2.pf}.1)  splits.

663: \end{enumerate}

664: \end{cor}

665: 

666: Proof. If any two of the numbers $m_i, n_j$ are relatively prime,

667:  the sequence  in (\ref{thm2.pf}.2)

668: splits

669: and $H^0(S,Q)\cong \z/(\prod n_j\cdot \prod m_i)$.

670: This implies that $H^1(S,R^1f_*\z_L)=0$.

671: The $E_2$ term of the Leray spectral sequence

672: (\ref{thm2.pf}.4) is 

673: $$

674: \begin{array}{lccll}

675: \z \quad & 0 & \quad\z^{s}+\sum_i (\z/m_i)^{2g(D_i)}\quad & 

676: (\mbox{torsion})\quad & \z\\

677: \z & 0 & \z^{s} & H^3(S,\z) & \z.

678: \end{array}

679: $$

680: Thus $H^3(L,\z)$ sits in an exact sequence

681: $$

682: 0\to H^3(S,\z) \to H^3(L,\z)\to \z^{s}+\sum_i (\z/m_i)^{2g(D_i)}

683: \to \z

684: $$

685: As before,  $H^3(S,\z)\cong H_1(S^0,\z)$

686: and the extension of the torsion part splits

687: if $|H_1(S^0,\z)|$ is relatively prime to $\prod m_i$.\qed

688: \medskip

689: 

690: This result turns out to be quite useful, since

691: the groups $H_1(S^0,\z)$ are rather special 

692:  when

693:  $(S,\Delta)$ is a log Del Pezzo surface. Thus $S$ is

694: a rational surface and so $H_1(S,\z)=0$.

695:  A project of C.\ Xu aims  to give a complete determination of

696: the possible fundamental groups $\pi_1(S^0)$

697: where $S$ is a rational surface with quotient singularities.

698: While this is not easy, the first homology

699:   $H_1(S^0,\z)$ is not hard to control.

700: 

701: The proof of (\ref{h2.tors.new.thm}) is obtained by combining

702: (\ref{thm2.pf}) and (\ref{fundgr.lem}).

703: 

704: 

705: \begin{lem} \label{fundgr.lem}

706: Let  $S$ be a log Del Pezzo surface with quotient singularities.

707: Then  $H_1(S^0,\z)$

708: is either $\z/m$,  $\z/m+\z/2$ or $\z/m+\z/3$ for some $m\geq 1$.

709: \end{lem}

710: 

711: Proof. Let $S'\to S$ be the corresponding Galois cover

712: with  Galois group $G$. The stabilizers of points

713: are subgroups of the local fundamental groups at the

714: singularities, and hence cyclic.

715: Moreover, no $1\neq g\in G$ fixes a curve   pointwise.

716: Thus it is enough to prove the following.

717: 

718: \begin{lem} Let $G$ be 

719:   an  Abelian group  acting

720: on a rational surface with cyclic stabilizers.

721: Assume also that no element fixes a curve of genus $\geq 1$ pointwise.

722:  Then

723: $G$ is  either $\z/m$,  $\z/m+\z/2$ or $\z/m+\z/3$.

724: \end{lem}

725: 

726: Proof.

727: We can take a $G$-equivariant resolution 

728: and then pass to the $G$-minimal model $T$.

729: (Note that if a smooth point $p\in T$ is fixed by

730: an Abelian group $H$ then $H$ also has a fixed point on

731: the blow up $B_pT$, so a fixed point on one

732:  model gives a fixed point on any other  model.)

733: 

734: Our aim is to find a $G$-invariant subset $Z\subset T$

735: such that either $Z\cong \p^1$ or $Z$ is at most 3 points.

736: In the first case, $G$ acts on $\p^1$ hence it  either

737: has a fixed point, or it acts through $(\z/2)^2$ and an index 2

738: subgroup has a fixed point. In the second case

739: a subgroup of index $\leq 3$ has a fixed point. Such

740: a subgroup is cyclic and so $G$ is

741:  $\z/m$,  $\z/m+\z/2$ or $\z/m+\z/3$ as required.

742: 

743: Consider first the case when  there is a 

744: $G$-equivariant ruling $f:T\to \p^1$. 

745: We are done if $f_*:G\to \aut(\p^1)$ is injective.

746: Otherwise, any $g\in \ker f_*$ fixes  2 points in each fiber of $f$,

747: thus $g$ fixes either 1 or 2 irreducible curves pointwise

748: and their union is $G$-invariant.

749: We are done if there is a $G$-invariant curve.

750: 

751: Otherwise, every $g\in \ker f_*$

752:  fixes 2 disjoint curves $E_1, E_2$ pointwise and $G$ permutes these curves.

753: By the Hodge index theorem 

754: $E_i^2\leq 0$, 

755: thus they generate the ``other'' extremal ray of the cone of curves.

756: In particular, either $T=\p^1\times \p^1$ 

757: or these curves are unique and  every element of

758: $\ker f_*$ fixes the same curves. If $G$ has a fixed point

759:  $p\in \p^1$ then $Z:=f^{-1}(p)\cap (E_1\cup E_2)$ is a 2--element set fixed

760: by $G$. If there is no such $p$ then  $G$ acts

761: on $\p^1$ and also on $E_1\cup E_2$ through $(\z/2)^2$.

762: Furthermore, there is an index 2 subgroup $H\subset G$ which

763: fixes each $E_i$ and has a fixed point on each $E_i$.

764: Thus $H$ is cyclic

765:  and so   $G\cong \z/m$ or   $G\cong \z/m+\z/2$.

766: 

767: The $\p^1\times \p^1$ case is left to the end.

768: 

769: Otherwise $T$ is a Del Pezzo surface of $G$-Picard number 1.

770: 

771: Assume that some $g\in G$  pointwise fixes some curves $C_i$.

772:  The $C_i$ are smooth,

773:  disjoint, rational curves.

774: By the adjunction formula, $C_i^2\in\{0,-1\}$ or $T\cong \p^2$.

775: Their sum $\sum C_i$ is $G$-invariant and not ample, but

776: this contradicts $G$-minimality. Thus either

777: $T\cong \p^2$ or every  element of $G$

778: acts with isolated fixed points.

779: 

780: Assume that  there is $H<G$ with  $H\cong (\z/2)^2$.

781: Then $T/H$ is a Del Pezzo surface with $A_1$-singularities only.

782: From Table 2 in (\ref{nontriv.pi1.lisat})

783:   we see that $\deg T/H=2$ and

784: $T\cong \p^1\times \p^1$. Thus, aside from this case, 

785: the 2--part of $G$ is cyclic.

786: 

787: 

788: 

789:  If $T=\p^2$,

790: take any $g\in G$. It has either $3$ fixed points or a fixed point 

791: and a fixed line (and the second case must happen if

792: $g^2=1$). In the second case, the fixed point is $G$-fixed

793: (and so $G$ is cyclic) 

794: and in the former case either all 3  points are $G$-fixed or

795: $G$ permutes them cyclically.

796: 

797: If $\deg T\in \{1,2,3,4,5\}$ then $\aut(T)$ acts faithfully

798: on $H^2(T,\z)$. Indeed, if $g$ acts trivially on $H^2(T,\z)$

799: then it descends to an automorphism of $\p^2$ with

800: $\geq 4$ fixed points in general position, hence $g=1$.

801: Thus, $G$ is a subgroup of the Weyl group of

802: $E_8,E_7,E_6,D_5,A_4$  (cf.\ \cite[Sec.25]{manin}).

803: 

804: 

805: 

806: 

807: If $\deg T=1$ then $|K_T|$ has a unique base point which is fixed by

808: $\aut(T)$, so $G$ is cyclic.

809: 

810: 

811: If $\deg T=2$ then there is a unique degree two morphism

812: $T\to \p^2$. If $|G|$ is odd and $H<G$ then 

813:  any $H$-fixed point on $\p^2$ is

814: dominated by (one or two) $H$-fixed point(s) on $T$. 

815: If $G$ is even, then a $G$-fixed point on $\p^2$

816: is  dominated by  two points and each is fixed by an

817: index 2 subgroup of $G$.

818: 

819: If $\deg T\in \{3,4,5\}$ then by looking at the order 

820: of the Weyl  groups, we see that the odd part 

821: $G_{odd}\subset \z/15$, except possibly

822: when $\deg T=3$ where the 3--part could be bigger.

823: Any action of a group on a cubic surface $T$ induces an

824: action on $\p^3\cong |K_T|$. For odd order groups this action lifts to

825:  $\c^4$ since the kernel of

826: $SL_4\to PSL_4$ is $\z/4$. Thus we get an eigenvector on $\c^4$

827: and a fixed point on the cubic $T$.

828: 

829:  

830: If $\deg T=6$ then only $\z/3$ can act on the $(-1)$-curves

831: nontrivially, and even for the $\z/3$-action there is

832: a $\z/3$-invariant set of 3 disjoint $(-1)$-curves. Thus the

833: action descends to $\p^2$.

834: 

835: There are no $G$-minimal  Del Pezzo surfaces of degree 7

836: and in degree 8 we get $\p^1\times \p^1$.

837: 

838: 

839: Thus we are left with $G$ acting on $\p^1\times \p^1$.

840: If any $g\in G$ interchanges the two factors, then

841: $g$ has 2 fixed points or a fixed rational curve

842: from the Lefschetz fixed point formula. In both cases 

843: an index 2 subgroup of $G$ has a fixed point.

844: 

845: 

846: The last case is when $G$ preserves the coordinate projections.

847: The image of $G$ in each $\aut(\p^1)$ is either cyclic

848: or $(\z/2)^2$. If the first case happens at least once, then

849: an index 2 subgroup has a fixed point. Finally we have to

850: deal with subgroups of $G\subset (\z/2)^4$. If the order is 8 or 16

851: then there are $g_1, g_2\in G$ which act trivially

852: on the first (resp. second) factor. Thus $\langle g_1,g_2\rangle$

853: is a noncyclic subgroup with a fixed point, a contradiction.

854: \qed

855: \medskip

856: 

857: 

858: Putting these together, we obtain the following.

859: Note that the torsion in $H_2$ is dual to the torsion in $H^3$,

860: thus a quotient of $H^3$ becomes a subgroup of $H_2$.

861: 

862: \begin{cor}\label{h2.tors.gen.thm} Let $L$ be a compact 5--manifold  

863: with a positive Sasakian structure.

864: Then  the torsion subgroup of the second homology, $\tors H_2(L,\z)$

865: has a subgroup $G$ such that

866: \begin{enumerate}

867: \item  $G$ is  $(\z/m)^2$,

868:  $(\z/5)^4$, $(\z/4)^4$,

869:   $(\z/3)^{2n}$ for $n\in \{2,3,4\}$ or $(\z/2)^{2n}$ and

870: \item $\tors H_2(L,\z)/G$ is 

871:  $\z/r$, $\z/r+\z/2$ or

872:  $\z/r+\z/3$ for some $r$. \qed

873: \end{enumerate}

874: \end{cor}

875: 

876: 

877: 

878: \begin{rem} Most Abelian groups can not be written in the above form.

879: If $G$ exists, then it is almost always unique up to

880: isomorphism. The only ambiguity is with the 6-torsion part.

881: 

882: To see this, we can consider the $p$-parts separately.

883: The main cases are

884: \begin{enumerate}

885: \item  $A_p=(\z/p^a)^2+\z/p^b$ and $A_p/G\cong \z/p^b$, 

886: \item  $A_p=\z/p^a+\z/p^b$ and $A_p/G\cong \z/p^{b-a}$.

887: \end{enumerate}

888: These are the only possibilities for $p\geq 7$,

889: with a few more cases for $p=2,3,5$.

890: \end{rem}

891: 

892: 

893: 

894: 

895: \begin{say}[Plan of the proofs of

896: (\ref{h2.tors.gen.cases}) and (\ref{main.thm})]\label{plan.say}

897: We follow the approach in \cite[6.8]{em5}. 

898: Start with  $(S,(1-\tfrac1{r})D+\Delta)$ satisfying the conditions

899: (\ref{algred.say}.1--3). Let $g:S'\to S$ be the

900: minimal resolution of $S$ and $h:S'\to S^m$ a minimal model of $S'$.

901: In the sequence of blow ups leading from $S$ to $S'$

902: and the subsequent blow downs leading from $S'$ to $S^m$

903: every intermediate surface $T$ satisfies the following condition:

904: \begin{enumerate}

905: \item[($*$)]  We can write $-K_T\equiv (1-\tfrac1{r})D_T+\Delta_T+H_T$

906: where  $D_T$ denotes the birational transform of $D$ on $T$,

907:  $\Delta_T$ is an effective linear combination

908: of rational curves (coming from $\Delta$ and the exceptional curves of $g$)

909:  and $H_T$ is nef and big

910: (this is a general divisor numerically equivalent to

911:  the pull back/push forward of 

912: $-(K_S+(1-\tfrac1{r})D+\Delta)$).

913: \end{enumerate}

914: This turns out to be very restrictive in many cases.

915: 

916: It is easy to see that $S$ is a rational surface, so

917: $S^m$ is either $\p^2, \p^1\times \p^1$ or a minimal ruled

918: surface $\f_n$ for some $n\geq 2$. For the latter, let

919: $E\subset \f_n$ denote the negative section and $F$ a fiber.

920: By an easy case analysis (cf.\ \cite[6.8]{em5}) we get the following.

921: \begin{enumerate}

922: \item If $r\geq 6$ and $g\geq 1$ then $S^m=\p^2, \p^1\times \p^1$

923: or $\f_2$ and $D^m\in |-K_{S^m}|$ is  smooth and elliptic.

924: \item If $r\geq 5, g\geq 2$ then $r=5$, $S^m=\f_3$ and $D^m\in |2E+6F|$

925: has genus $2$.

926: \item If $r\geq 3, g\geq 4$ then $r=3$, $S^m=\f_5$ and $D^m\in |2E+10F|$

927: has genus $4$.

928: \item If $r=4, g\geq 2$ then either $S^m=\f_3$ and $D^m\in |2E+6F|$

929: or $S^m=\f_2$ and $D^m\in |2E+5F|$. In both case $D^m$ has genus $2$.

930: \end{enumerate}

931: The plan is, in each case, to start with 

932: $(S^m,(1-\tfrac1{r})D^m+\Delta^m)$ and to write

933: down all possible $(S',(1-\tfrac1{r})D'+\Delta')$.

934: It is then easy to get a complete list of all

935: $(S,(1-\tfrac1{r})D+\Delta)$. Finally we need to check the additional

936: conditions, especially \cite[3.6 and 4.8]{em5}.

937: \end{say}

938: 

939: 

940: 

941: 

942: 

943: \section{The exceptional cases}

944: 

945: \begin{say}[The $(\z/5)^4$ and $(\z/3)^8$ cases]\label{5and3.pf}

946: 

947: Assume that $H_2(L.\z)$ contains a subgroup

948: isomorphic to $(\z/5)^4$ or $(\z/3)^8$.

949: Let $G\subset H_2(L.\z)$  be the subgroup given in

950: (\ref{h2.tors.gen.thm}). Since the odd torsion

951: in $H_2(L,\z)/G$ is either cyclic or  $(\z/3)^2$,

952: we conclude that either

953: \begin{enumerate}

954: \item $G= (\z/5)^4$, 

955: \item $G= (\z/3)^8$, or

956: \item  $G=(\z/3)^6$ and 

957: $H_2(L,\z)/G=(\z/3)^2$.

958: \end{enumerate}

959: Thus  $g(D)=2$ in the first case and $g(D)\geq 3$ in the second and third  

960: cases.

961: 

962: This implies that $S^0$ is simply connected.

963: Indeed, any nontrivial cover $\pi:S'\to S$ would give

964: another log Del Pezzo surface where $D':=\pi^{-1}(D)$

965: has genus $\geq 3$ in  the first case and genus $\geq 5$ in the other cases.

966: By the list in (\ref{plan.say}) there are no such surfaces.

967: In particular, the case  $G=(\z/3)^6$ and 

968: $H_2(L,\z)/G=(\z/3)^2$ does not happen.

969: 

970: Next, as in \cite[9.9]{em5}, we show that $S'=S^m$ and

971: $S$ is obtained by contracting the negative section $E$.

972: Indeed, any one point blow up of $\f_m$ maps

973: either to $\f_{m-1}$ (if we blow up a point not on $E$)

974: or  $\f_{m+1}$ (if we blow up a point  on $E$).

975: Since $S^m$ is unique in each case

976: by the list in (\ref{plan.say}),

977: $S'\to S^m$ is an isomorphism. 

978: 

979: If $S=S^m$ then 

980: $$

981: \Delta^m+H^m\equiv \sum (1-\tfrac1{r_i})D_i+H

982: $$

983: where the $D_i$ are rational and $H$ is ample. 

984: In the $(\z/5)^4$ case, this would give

985: $$

986: \tfrac25 E+\tfrac15F\equiv \sum (1-\tfrac1{r_i})D_i+H.

987: $$

988: The intersection number of the left hand side with

989: $F$ is $\tfrac25<\frac12$, so 

990: any $D_i$ is a fiber. Now intersecting with $E$ shows that

991: there  can not be any $D_i$. This is impossible since

992:  the left hand side is not  nef and big.

993: 

994: Similarly, in the $(\z/3)^8$ case we would need to solve

995: $$

996: \tfrac23 E+\tfrac13F\equiv \sum (1-\tfrac1{r_i})D_i+H,

997: $$

998: which is also impossible.

999: 

1000: 

1001: 

1002: This proves  (\ref{main.thm}), except for the

1003: equations. 

1004: 

1005: There is a systematic way to obtain $Y$ and the $\c^*$-action from

1006: $(S,\Delta)$ (cf.\ \cite[2.5]{em5}), but I found it much easier to

1007: do this by guessing. Note that by contracting the negative section

1008: in $\f_n$, we get the weighted projective plane

1009: $\p(1,1,n)$ and $D\in |2E+2nF|$ is a curve given by a

1010: degree $2n$ equation. After completing the square,

1011: these are of the form $z^2+f_{2n}(x,y)$.

1012: The explicit relationship between $\Delta$ and the weights

1013: exhibited in \cite[6]{bgk} now leads to 

1014: $f=t^5+z^2+f_{6}(x,y)$ in case $r=5, S^m=\f_3$

1015: and to 

1016: $f=t^3+z^2+f_{10}(x,y)$ in case $r=3, S^m=\f_5$.\qed

1017: \end{say}

1018: 

1019: The case analysis is harder for $(\z/4)^4$ and

1020: we need some way to see how to get $S'$ from $S^m$.

1021: 

1022: \begin{say}[Blow up criterion]\label{bu.crit} Assume that

1023: $\pi:T_1\to T$ is the inverse of the blowing up of

1024: $p\in T$ with exceptional curve $E_1\subset T_1$ and that

1025: $(T_1,(1-\tfrac1{r})D_1+\Delta_1+H_1)$ satisfies  (\ref{plan.say}.$*$).

1026: That is, $-K_{T_1}\equiv (1-\tfrac1{m})D_1+\Delta_1+H_1$,

1027: $\Delta_1$ is  effective  and $H_1$ is nef and big.

1028: 

1029: Set $D:=\pi_*D_1, \Delta:=\pi_*\Delta_1$ and

1030: $H:=\pi_*H_1$. Then 

1031: $(T,(1-\tfrac1{r})D+\Delta+H)$ also satisfies  (\ref{plan.say}.$*$).

1032: Since $(K_{T_1}\cdot E_1)=-1$, we conclude that

1033: $$

1034: \left((1-\tfrac1{r})D_1+\Delta_1+H_1\right)\cdot E_1=1,

1035: $$

1036: hence 

1037: $$

1038: \mult_p \left((1-\tfrac1{r})D+\Delta+H\right)\geq 1.

1039: $$

1040: In practice we know $(T,(1-\tfrac1{r})D)$ 

1041:  and we would like to

1042: find $p$. This is possible if the numerical class of

1043: $\Delta+H$ is small, but we get many possibilities if

1044: $\Delta+H$ is bigger.

1045: \end{say}

1046: 

1047: \begin{say}[The $(\z/4)^4$ case]

1048: Here there are two possibilities for $(S^m, D^m)$

1049: and $S'\to S^m$ need not be an isomorphism.

1050: 

1051: Let us start with  $S^m=\f_2$ and $D^m\in |2E+5F|$.

1052: \medskip

1053: 

1054: {\it Case 1.}

1055: If $S'=S_m$ then, as in (\ref{5and3.pf}),

1056:  from $(E\cdot (\Delta^m+H^m))<0$ we see that

1057: $E$ must be contracted. Thus $S$ is the 

1058:  weighted projective plane

1059: $\p(1,1,2)$ 

1060:  and $D$ is given by a weighted degree 5 equation

1061: $(f_5(x,y,z)=0)$ where $w(z)=2$.

1062: (Alternatively,  $S$ is the quadric cone

1063: in $\p^3$ and $D$ is a curve of degree 5

1064: passing through its vertex.)

1065: Thus we get the equation

1066: $(t^4+f_5(x,y,z)=0)$ for $Y$.

1067: This gives us (\ref{main.thm}.3.a).

1068: \medskip

1069: 

1070: {\it Case 2.}

1071: If we perform at least one blow up in going from $S^m$ to $S'$,

1072: then as before, we could have ended up with

1073: $S^m=\f_1$ or $S^m=\f_3$ instead.

1074: The first case is impossible from the list of (\ref{plan.say})

1075:  and the second one

1076: we consider next.

1077: 

1078: \medskip

1079: {\it Case 3.}

1080: Thus assume that $S^m=\f_3$ and $C\in |2E+6F|$ is a

1081: smooth curve. Thus  $\Delta^m+H^m\equiv \tfrac12 E+\tfrac12 F$.

1082: 

1083: If $S'=S^m$  then, as before,  we see that

1084: $E$ must be contracted. 

1085: This leads to the surface

1086: $S=\p(1,1,3)$ and $D\in |\o_S(6)|$ a smooth curve.

1087: 

1088: This never leads to simply connected  5--manifolds  by \cite[4.8]{em5}.

1089: All Seifert bundles are of the form

1090: $Y(S, B, \frac{b}{4}D)$ where $B=aF$ for some $a\in\z$ and $b\in \{1,3\}$.

1091: The corresponding Chern class is

1092: $(a+\frac{3b}{2})F$, always a half integer.

1093: Thus we get the basic cases with $c_1(Y/S)=\frac12$ and their

1094: obvious quotients.

1095: 

1096: 

1097: Since $D\in |\o_S(6)|$ and $6$ is even,

1098: $H_1^{orb}(S,\frac34D)=\z/2$

1099: and there are two basic cases: $a=-1,b=1$ and $a=-4,b=3$.

1100: 

1101: The double cover of $(S,\frac34D)$

1102: is given by

1103: $$

1104: S_6=(f_6(x,y,z)+T^2=0)\subset \p(1,1,3,3),

1105: $$

1106: and the involution is $(x:y:z:T)\mapsto (x:y:z:-T)$.

1107: Keep in mind that these are projective coordinates,

1108: so the same  involution can be given as $(x:y:z:T)\mapsto (-x:-y:-z:T)$.

1109: The basic Seifert bundle pulls back to

1110: $Y=L^*(2,2,6,3;12)$ with typical equation

1111: $$

1112: x^6+y^6+z^2+t^4=0.

1113: $$ 

1114: There are 2 ways to lift the involution to a fixed point free

1115: involution on $Y$. These are

1116: $(x,y,z,t)\mapsto (-x,-y,-z,-t)$ and

1117: $(x,y,z,t)\mapsto (-x,-y,-z,t)$.

1118: These give the 2 families listed in  (\ref{main.thm}.3.b).

1119: 

1120: \medskip

1121: {\it Case 4.}

1122: Next we blow up at least 1 point on $S^m$.

1123: This point can not be on $E$ since the resulting surface

1124: would also dominate $\f_4$. 

1125: Write

1126: $$

1127: \tfrac12 E+\tfrac12 F\equiv \Delta^m+H^m=\sum a_iF_i+bE+R^m,

1128: $$

1129: where the $F_i$ are distinct fibers and $R^m$

1130:  has no irreducible component which is $E$ or a fiber.

1131: By intersecting with $F$ we see that $b\leq \frac12$.

1132: Intersecting with $E$ gives 

1133: $$

1134: 0\leq (E\cdot R^m)= 3b-1-\sum a_i,

1135: $$

1136: in particular $b\geq \tfrac13$ and

1137: $\tfrac12-b\leq \tfrac13(\tfrac12-\sum a_i)$,

1138: thus $\sum a_i\leq \frac12$.

1139: If $p$ lies on $F_1$ then

1140: $$

1141: \mult_p(\Delta^m+H^m)\leq a_1+(F\cdot R^m)

1142: =a_1+(\tfrac12-b)\leq \tfrac16+\tfrac23 a_1-\tfrac13\sum_{i\neq 1}a_i

1143: \leq \tfrac12.

1144: $$

1145: The condition (\ref{plan.say}.$*$) is 

1146: $$

1147: \mult_p(\tfrac34 D+\Delta^m+H^m)\geq 1,

1148: $$

1149: which is only possible if $p\in C$ and $a_1\geq 1/8$.

1150: Furthermore, if $a_1\leq a_2$ then we get

1151: $\tfrac 14\leq a_1\leq a_2$. But $\sum a_i\leq \tfrac12$,

1152: and this would lead to $R^m=(\tfrac12-b)E$

1153: which is impossible.

1154: Thus we conclude:

1155: \medskip

1156: 

1157: {\it Claim.} There is a unique fiber $F$ such that

1158: $S'$ is obtained from $S^m$ by blowing up points

1159: in or above $F\cap D$. 

1160: \medskip

1161: 

1162: The general case is when $F\cap D$ consists of 2 points

1163: and the special case is when $F\cap D$ is a single point

1164: where $F$ and $D$ are tangent.

1165: 

1166: \medskip

1167: {\it Case 5.}

1168: Let us deal first with the general case and blow up $p\in F\cap D$.

1169: We get $S_1\to S^m$ with exceptional curve $G_1$. Let

1170: $F_1$ and $ D_1$ be the birational transforms of $F$ and $D$.

1171: Write

1172: $$

1173: \Delta_1+H_1=aF_1+bE_1+cG_1+R_1.

1174: $$

1175: From (\ref{bu.crit}) we see that 

1176: $$

1177: c\leq \mult_p(\tfrac34 D+\Delta^m+H^m)-1

1178: \leq \tfrac34+a+\tfrac13(\tfrac12-a)-1=\tfrac23a-\tfrac1{12}.

1179: $$

1180: So at every point  $q\in G_1\setminus(D_1+F_1)$ the multiplicity

1181: of $\Delta_1+H_1$ is at most

1182: $$

1183: c+\mult_qR_1\leq c+\mult_pR^m\leq \tfrac23a-\tfrac1{12}

1184: + \tfrac16-\tfrac13 a=\tfrac1{12}+\tfrac13a<1,

1185: $$

1186: thus we can not blow these up. At $G_1\cap F_1$

1187: the multiplicity

1188:  is at most

1189: $$

1190: a+c+\mult_qR_1\leq a+\tfrac1{12}+\tfrac13a<1.

1191: $$

1192: Finally, at $G_1\cap D_1$

1193: the multiplicity

1194:  is at most

1195: $$

1196: \tfrac34+\tfrac1{12}+\tfrac13a=\tfrac56+\tfrac13 a<1,

1197: $$

1198: except when $a=\tfrac12$.

1199: But  this again would lead to $R^m=(\tfrac12-b)E$

1200: which is impossible.

1201: 

1202: 

1203: Thus we conclude that we can blow up one or both of the points

1204: $F\cap D$ and then we get $S'$. 

1205: 

1206: If we blow up only one point, we have to contract $F_1$

1207: and we are in the already considered case when $S'=\f_2$.

1208: 

1209: 

1210: \medskip

1211: {\it Case 6.}

1212: If we blow up both points of $F\cap D$, we have to contract $E_1\cup F_1$.

1213: We get a surface with Picard number 2 and a single

1214: cyclic quotient singularity of the form

1215: $\c^2/\tfrac15(1,2)$.

1216: 

1217: \medskip

1218: 

1219: {\it Claim.} The surface $S$ is isomorphic to a

1220: quasi-smooth hypersurface $S_6\subset \p(1,1,3,5)$

1221: and $D$ is the complete intersection of $S_6$ with 

1222: $(t=0)$.

1223: \medskip

1224: 

1225: I found this isomorphism by computing the quotient 

1226: by the hyperelliptic involution of $D$ which acts

1227: on $S^m,S'$ and also on $S$. Once the isomorphism is guessed,

1228: it is easier to verify  by working backwards from 

1229: the surface $S_6\subset \p(1,1,3,5)$.

1230: Its equation can be written, after coordinate changes, as

1231: $$

1232: S_6=(z^2+f_6(x,y)+\ell_1(x,y)t=0)\subset \p(1,1,3,5).

1233: $$

1234: Notice that its intersection with $(\ell_1=0)$

1235: is the reducible curve $(z^2+f_6(x,\alpha x)=0)\subset \p(1,3,5)$

1236: for some $\alpha$.

1237: The two irreducible components correspond to the two 

1238: exceptional curves of $S'\to S^m$.

1239: The rest is a straightforward computation. 

1240: The surface is specified by  choosing 

1241: 6 points in $\p^1$ (given by $f_6=0$) plus one more

1242: corresponding to the choice of $\ell_1$.

1243: 

1244: This leads to the case (\ref{main.thm}.3.e).

1245: 

1246: 

1247: 

1248: \medskip

1249: {\it Case 7.}

1250: Next we deal with the special case  when $F\cap D$ is a single point

1251: where $F$ and $D$ are tangent. Computations as above

1252: yield that in this case we can blow up the point $p$ on $D$ 

1253: at most 3--times.

1254: 

1255: 

1256: \medskip

1257: {\it Case 8.}

1258: If $S'$ is obtained by 1 blow up, we factor through $\f_2$ as before.

1259: If we do 2 blow ups, we get $S=\p(1,2,5)$ and

1260: $D\in |\o(10)|$.  

1261: As in Case 3, $H_1^{orb}(S,\frac34D)=\z/2$.

1262: Set $L:=(x=0)$.

1263: There are two basic cases, corresponding to

1264: $Y(S,-2L, \frac14D)$ and $Y(S,-7L, \frac34D)$.

1265: The index 2 point $(0:1:0)\in S$ adds a further complication

1266: since the first of these does not satisfy  the smoothness  condition

1267: \cite[4.8.1]{em5}.

1268:  

1269: The orbifold double cover of $(S,\frac34D)$ is

1270: $$

1271: S_{10}:=(f_{10}(x,y,z)+T^2=0)\subset \p(1,2,5,5)

1272: $$

1273: and the involution is

1274: $(x:y:z:T)\mapsto (x:y:z:-T)$.

1275: Since these are weighted projective coordinates, this is the same as

1276: $(x:y:z:T)\mapsto (-x:y:-z:T)$.

1277: (Note that $-1$ acts by sending a coordinate $u$ to $(-1)^{wt(u)}u$,

1278: hence the $+$ sign in front of $y$.)

1279: 

1280: This leads to the basic examples

1281: $L=L^*(2,4,10,5;20)$ with sample equation  $(x^{10}+y^{5}+z^2+t^4=0)$,

1282: The lifting of the involution is

1283: either $(x,y,z,t)\mapsto (-x,y,-z,-t)$ or

1284:  $(x,y,z,t)\mapsto (-x,y,-z,t)$. Here the second action has fixed point,

1285: this is consistent with our earlier considerations.

1286: Thus we get only one family, as in  (\ref{main.thm}.3.d).

1287: 

1288: \medskip

1289: {\it Case 9.}

1290: Finally, if we blow up 3--times, we can contract the

1291: birational transforms of $E,F$ and of the first 2

1292: exceptional curves. This gives a  surface with a single singular point

1293: of the form $\c^2/\tfrac19(2,5)$. 

1294: 

1295: 

1296: 

1297: \medskip

1298: 

1299: {\it Claim.} The surface $S$ is isomorphic to a

1300: quasi-smooth hypersurface $S_{10}\subset \p(1,2,5,9)$

1301: and $D$ is the complete intersection of $S_{10}$ with 

1302: $(t=0)$.

1303: \medskip

1304: 

1305:  Once again, 

1306: it is easier to verify this by working backwards.

1307: The equation of $S_{10}$ is

1308: $$

1309: (f_5(x^2,y)+z^2+xt=0)\subset \p(1,2,5,9).

1310: $$

1311: Its intersection 

1312: with $(x=0)$ is the  curve $(ay^5+bz^2=0)\subset \p(2,5,9)$.

1313: This is a smooth (but not quasi-sooth) rational curve,

1314: and it  corresponds to the last 

1315: exceptional curve of $S'\to S^m$.

1316: The surface is specified by  choosing 

1317: 6 points in $\p^1$ (given by $xf_5=0$), that is

1318: a genus 3 hyperelliptic curve plus a specified branch point.

1319: The rest is a straightforward computation. 

1320: 

1321: Thus we obtain (\ref{main.thm}.3.c).

1322: \qed

1323: \end{say}

1324: 

1325: 

1326: \begin{say}[Other examples] It is easy to write down

1327: infinitely many positive Sasakian structures

1328: on  certain simply connected 5--manifolds $L$

1329: either by hand or by consulting the

1330: (partially unpublished) lists of Boyer and Galicki.

1331: Below I write the orbifolds $(S,\Delta)$ and the simplest equation.

1332: (Here $C_k$ denotes a general curve of degree $k$ 

1333: in the weighted projective plane $\p(a,b,c)$ with coordinates

1334: $x,y,z$ and

1335: $\ell:=(x=0)$.)

1336: \begin{enumerate}

1337: \item $H_2(L,\z)=(\z/3)^2$. For any $(k,6)=1$, take 

1338: $$

1339: \bigl(\p(1,1,2), (1-\tfrac13)C_4+(1-\tfrac1{k})\ell\bigr)\qtq{and}

1340: x^{4k}+y^4+z^2+t^3.

1341: $$

1342: \item $H_2(L,\z)=(\z/5)^2$. For any $(k,30)=1$, take 

1343: $$

1344: \bigl(\p(1,2,3), (1-\tfrac15)C_6+(1-\tfrac1{k})\ell\bigr)\qtq{and}

1345: x^{6k}+y^3+z^2+t^5.

1346: $$

1347: \item $H_2(L,\z)=(\z/3)^4$. For any $(k,30)=1$, take 

1348: $$

1349: \bigl(\p(1,2,5), (1-\tfrac13)C_{10}+(1-\tfrac1{k})\ell\bigr)\qtq{and}

1350: x^{10k}+y^5+z^2+t^3.

1351: $$

1352: \item $H_2(L,\z)=(\z/2)^{2n}$. For $n\geq 0$ and $(k,2n(2n+1))=1$  take

1353: $$

1354: \bigl(\p(1,1,n), (1-\tfrac12)C_{2n+1}+(1-\tfrac1{k})\ell\bigr)\qtq{and}

1355: x^{k(2n+1)}+y^{2n+1}+yz^2+t^2.

1356: $$

1357: \end{enumerate}

1358: Sasakian--Einstein structures are known to exist

1359: in the first 3 cases, but for the last one the  criterion

1360: of \cite{dk} fails and existence is not known.

1361: 

1362: The situation is more complicated in the next two cases:

1363: \begin{enumerate}\setcounter{enumi}{4}

1364: \item $H_2(L,\z)=(\z/4)^2$ or $(\z/6)^2$. For any $k$, take 

1365: $$

1366: \bigl(\p(1,2,3), (1-\tfrac14)C_6+(1-\tfrac1{k})\ell\bigr)

1367: \qtq{resp.}

1368: \bigl(\p(1,2,3), (1-\tfrac16)C_6+(1-\tfrac1{k})\ell\bigr).

1369: $$

1370: These are the right examples as log Del Pezzo surfaces, but

1371: the condition \cite[4.8.2]{em5} fails, so the corresponding Seifert bundles

1372: are not simply connected. Since my approach is to rule out everything at the

1373: surface level, these cases could be very hard to settle using my methods.

1374: \end{enumerate}

1375: 

1376:  This leaves

1377: open the finiteness question for

1378: \begin{enumerate}\setcounter{enumi}{5}

1379: \item $H_2(L,\z)=(\z/m)^2$ with  $7\leq m\leq 11$, and

1380: \item $H_2(L,\z)=(\z/3)^6$.

1381: \end{enumerate}

1382: The  computations in Section \ref{klt.sect} suggest that in these cases

1383: there should be only finitely many families of 

1384: pre-SE  structures.

1385: 

1386: However, as the examples with $H_2(L,\z)=(\z/4)^4$  suggest, the

1387: case analysis can be rather tricky and unexpected special

1388: configurations may arise.

1389: \end{say}

1390: 

1391: \section{The main series}

1392: 

1393: 

1394: 

1395: 

1396: Let us start with fixing an error in \cite{em5}.

1397: 

1398: 

1399: 

1400: \begin{say}[Correction to {\cite[Thm.1.8.1(b)]{em5}} ]

1401: \label{corr181.say}

1402: The theorem asserts that for each $m\geq 12$, there are

1403: exactly 93  pre-SE Seifert bundles

1404: $f:L\to (S,\Delta)$ on 

1405:  compact 5--manifolds $L$  

1406: satisfying $H_1(L,\z)=0$ and $\tors H_2(L,\z)\cong (\z/m)^2$. 

1407: 

1408: The construction of the 93 families of pairs

1409: $(S,\Delta)$ is correct.

1410: Going from $(S,\Delta)$ to the Seifert bundle is, however,

1411: done incorrectly  since the conditions for  a

1412: Sasakian structure and for  a pre--Sasakian--Einstein

1413: structure have been thoroughly mixed up.

1414: 

1415: 

1416: The construction of the 93 families given in \cite[7.6]{em5}

1417: starts with a surface $T$ which is

1418: one of $\p^1\times \p^1, \p^2, Q,  S_5, \p(1,2,3)$.

1419: For each of these write $-K_T\sim d(T)H$ where $H\in \weil(T)$

1420: is a positive generator. We have

1421: $$

1422: d(\p^1\times \p^1)=2,  d(\p^2)=3, d(Q)=4,  d(S_5)=5,  d(\p(1,2,3))=6.

1423: $$

1424: Next we perform some weighted blow ups \cite[7.3]{em5}

1425: to get $S=B_{m_1,\dots,m_k}T$.

1426: There are $k$ exceptional curves $E_1,\dots, E_k$.

1427: Each $E_i$ passes through a unique singular point $p_i\in S$

1428: and $E_i$ generates the local class group which is $\z/m_i$.

1429: Set $d(S):=\gcd(m_1,\dots,m_k, d(T))$.

1430: 

1431: The divisor class group $\weil(S)$ is freely generated by

1432: $$

1433: \pi^*H, E_1,\dots, E_k\qtq{and}

1434: K_S\sim -d(T)\pi^*H+\sum m_iE_i.

1435: $$ 

1436: 

1437: 

1438: The main condition that was overlooked is the

1439: smoothness criterion for Seifert bundles \cite[3.6]{em5}.

1440: 

1441: In our case there is only one curve $D$ and $S$ is smooth along

1442: $D$. Thus, by \cite[3.6]{em5},

1443:  the corresponding Seifert bundle 

1444: $Y(S,B,\frac{b}{m}D)$ is smooth iff 

1445: $$

1446: \mbox{$B$ generates the local class group at every  point.}

1447: \eqno{(\ref{corr181.say}.1)}

1448: $$

1449: \medskip

1450: 

1451: {\it Claim \ref{corr181.say}.2.} Notation as above. Then

1452: $$

1453: Y(B_{m_1,\dots,m_k}T, aH+c_1E_1+\cdots+c_kE_k, \frac{b}{m}D)

1454: \qtq{is smooth}

1455: $$

1456: iff  $(m_i,c_i)=1$ for $i=1,\dots,k$ and

1457: $(a,d(T))=1$ if $T$ is singular.

1458: \medskip

1459: 

1460: As a consequence, we see that all 93 cases correspond 

1461: to positive Sasakian structures on smooth 5--manifolds.

1462: 

1463: A Seifert bundle is pre-SE iff

1464: its Chern class $c_1(Y/S)=B+\frac{b}{m}D$ is a rational multiple of 

1465: $-(K_S+(1-\frac1{m})D)$. In our case $D\sim -K_S$, hence

1466: $B$ itself is a rational multiple of $-K_S$.

1467: Thus

1468: $$

1469: B=r\bigl(\tfrac{d(T)}{d(S)}\pi^*H-\sum \tfrac{m_i}{d(S)}E_i\bigr)

1470: $$

1471: for some positive integer $r$.

1472: Thus if  (\ref{corr181.say}.1)

1473: holds then $d(S)=m_i$ for every $i$.

1474: 

1475: Furthermore, in the cases when $T$ is singular,

1476: $B$ generates the local class group at each singular point of $T$

1477: iff $d(T)=d(S)$. 

1478: Thus we obtain the following.

1479: \medskip

1480: 

1481: {\it Claim \ref{corr181.say}.3.} Notation as above. Then

1482: $$

1483: Y(B_{m_1,\dots,m_k}T, aH+c_1E_1+\cdots+c_kE_k, \tfrac{b}{m}D)

1484: \qtq{is smooth and pre--SE iff}

1485: $$

1486: \begin{enumerate} 

1487: \item $m_1=\cdots=m_k=d(S)$, 

1488: \item 

1489: $aH+c_1E_1+\cdots+c_kE_k= r(-\frac{d(T)}{d(S)}H+E_1+\cdots+E_k)$

1490: for some $r\in \z$, and

1491: \item  if $T$ is singular, then $(r,d(T))=1$ and $d(S)=d(T)$. \qed

1492: \end{enumerate}

1493: \medskip

1494: 

1495: 

1496: 

1497: These conditions cut down considerably

1498: the list given in \cite[7.6]{em5} and we

1499:  get the following 19  cases:

1500: $$

1501: \begin{array}{c}

1502: \begin{array}{lc}

1503: \qquad \mbox{surfaces} & \mbox{$d(S)$}\\*[1ex]

1504: B_{1}\p^2, B_{11}\p^2,\dots, B_{11111111}\p^2 & 1\\

1505: \p^1\times \p^1, 

1506: B_2\p^1\times \p^1, B_{22}\p^1\times \p^1, B_{222}\p^1\times \p^1 & 2\\

1507: \p^2, B_3\p^2, B_{33}\p^2 &  3\\

1508: Q, B_4Q & 4\\

1509: S_5 & 5\\

1510: \p(1,2,3) & 6

1511: \end{array}

1512: \ \\

1513: \mbox{Table 1}

1514: \end{array}

1515: $$

1516: 

1517: 

1518: 

1519: 

1520: The condition \cite[4.8.2]{em5} says that

1521: the resulting Seifert bundle $L$ satisfies

1522: $H_1(L,\z)=0$ iff 

1523: $H^2(S,\z)\to H^2(D,\z)\to \z/m $ is surjective.

1524: If $S=T$, that is, we do no blow ups at all, then 

1525: by \cite[9.8]{em5}

1526: $$

1527: \im\bigl[H^2(T,\z)\to H^2(D,\z)\bigr]=d(T)H^2(D,\z).

1528: $$

1529: After blow ups, the new curves that come in are the

1530:  exceptional curves $E_i$. Here $(E_i\cdot D)=1$ but the $E_i$ pass through the

1531: singular points and they are only homology classes.

1532: To get a cohomology class (or Cartier divisor), we need to

1533: take $m_iE_i$ since $m_i$ is also the index of the singular point.

1534: Thus we conclude:

1535: \medskip

1536: 

1537: {\it Claim \ref{corr181.say}.4.} For $S=B_{m_1,\dots,m_k}T$,

1538: $\im\bigl[H^2(S,\z)\to H^2(D,\z)\bigr]= d(S)\cdot

1539: H^2(D,\z)$.\qed

1540: \end{say}

1541: 

1542: \medskip

1543: 

1544: \begin{cor} Let $S$ be a projective surface with 

1545: Du Val singularities 

1546: such that $\pi_1(S^0)=1$. Let

1547:  $D\in |-K_S|$ be a smooth elliptic curve.

1548: There is a simply connected Seifert bundle

1549: $f:L\to (S,(1-\frac1{m})D)$ with a pre-SE structure iff

1550: \begin{enumerate}

1551: \item $S$ is one of the surfaces in Table 1, and

1552: \item  $m$ is relatively prime to $d(S)$.

1553: \end{enumerate}

1554: In these cases, $f:L\to (S,(1-\frac1{m})D)$ is

1555: uniquely determined by $(S,(1-\frac1{m})D)$

1556: (up to reversing the orientation of the fibers) and

1557: $L$ carries a Sasakian--Einstein metric for $m\geq 7$.

1558: \end{cor} 

1559: 

1560: Proof. we have proved eveything, except the claims

1561: about the existence of Sasakian--Einstein metrics.

1562: This will be established in (\ref{ke.conds}).\qed

1563: 

1564: \begin{say}[Equations] Quite surprisingly, all the singular surfaces on the

1565: list can be realized in weighted projective 3--spaces.

1566: All of these examples are on the Boyer--Galicki lists.

1567: Here  also claim the converse: every singular Del Pezzo surface

1568: in Table 1 is isomorphic to a corresponding surface below.

1569: \begin{enumerate}

1570: \item $B_2\p^1\times \p^1$:  $S_3\subset \p(1,1,1,2)$ with simplest equation

1571: $$

1572: x^3+y^3+z^3+xt^m=0 \qtq{for $(m,2)=1$.}

1573: $$

1574: \item $B_{22}\p^1\times \p^1$:  $S_4\subset \p(1,1,2,2)$ with simplest equation

1575: $$

1576: x^4+y^4+z^2+zt^m=0 \qtq{for $(m,2)=1$.}

1577: $$

1578: \item $B_{222}\p^1\times \p^1$:  $S_6\subset \p(1,2,3,2)$

1579:  with simplest equation

1580: $$

1581: x^6+y^3+z^2+t^{3m}=0 \qtq{for $(m,2)=1$.}

1582: $$

1583: \item $B_3\p^2$:   $S_4\subset \p(1,1,2,3)$ with simplest equation

1584: $$

1585: x^4+y^4+z^2+xt^m=0 \qtq{for $(m,3)=1$.}

1586: $$

1587: \item $B_{33}\p^2$:  $S_6\subset \p(1,2,3,3)$ with simplest equation

1588: $$

1589: x^6+y^3+z^2+zt^m=0 \qtq{for $(m,3)=1$.}

1590: $$

1591: \item $Q$:  $S_4\subset \p(1,1,2,4)$ with simplest equation

1592: $$

1593: x^4+y^4+z^2t^m=0 \qtq{for $(m,2)=1$.}

1594: $$

1595: \item $B_4Q$:  $S_6\subset \p(1,2,3,4)$ with simplest equation

1596: $$

1597: x^6+y^3+z^2+yt^m=0 \qtq{for $(m,2)=1$.}

1598: $$

1599: \item $S_5$:  (sorry for the notation) 

1600: $S_6\subset \p(1,2,3,5)$ with simplest equation

1601: $$

1602: x^6+y^3+z^2+zt^m=0 \qtq{for $(m,5)=1$.}

1603: $$

1604: \item $\p(1,2,3)$:  $S_6\subset \p(1,2,3,6)$ with simplest equation

1605: $$

1606: x^6+y^3+z^2+t^m=0 \qtq{for $(m,6)=1$.}

1607: $$

1608: \end{enumerate}

1609: \end{say}

1610: 

1611: 

1612: \begin{rem} There are no hypersurface links 

1613: on the Boyer--Galicki lists giving infinitely many

1614: $(S,\Delta)$ where $S\in \{B_1\p^2, \dots, B_{11111}\p^2\}$.

1615: 

1616: I claim that these  can not be realized

1617:  as the link $L$ of a hypersurface with a $\c^*$-action, 

1618: at least when $H_2(L,\z)\supset (\z/m)^2$ for

1619: $m\geq 12$ and $(S,\Delta)$ is log Del Pezzo.

1620: Assume the contrary. Then we get that

1621: $S$ is a hypersurface in a weighted projective space

1622: $\p(a,b,c,d)$ such that $K_S+(1-\frac1{m})D$ is proportional to

1623: $H|_S$ where $H$ is the hyperplane class 

1624: of the weighted projective space. Since $D\in |-K_S|$, we conclude that

1625: $K_S$ is proportional to

1626: $H$.  By the Grothendieck--Lefschetz theorem,

1627: this implies that $-K_S=dH$ for some $d\in \z$. 

1628: In our cases $-K_S$ is not divisible, so $d=1$.

1629: This implies that

1630: $$

1631: h^0(S,\o_S(-K_S))=h^0(\p(a,b,c,d), \o_{\p}(1))\leq 4,

1632: $$

1633: since this dimension is the number of times that 1 occurs among $a,b,c,d$.

1634: In the above cases, however, $h^0(S,\o_S(-K_S))\geq 5$.

1635: 

1636: Galicki told me that the link $L^*(2,3,4,7;14)$

1637: realizes  $S=B_{1111}\p^2$, but the corresponding $\Delta$

1638: is not a rational multiple of $-K_S$.

1639: \end{rem}

1640: 

1641: 

1642: \begin{say}[The cases with nontrivial fundamental group]

1643: \label{nontriv.pi1.lisat} The classification of minimal 

1644: Del Pezzo surfaces with Du Val singularities

1645: is completed in \cite{furushima, mi-zh1, mi-zh2, ye}. 

1646: We are interested only in those that have cyclic quotient singularities.

1647: The 5 cases where $\pi_1(S^0)=1$ were considered in \cite{em5}.

1648: The following table lists the remaining ones.

1649: $$

1650: \begin{array}{c}

1651: \begin{array}{cccccc}

1652: \mbox{degree} & \mbox{singularities}& \pi_1(S^0) & \weil/\pic & 

1653: \mbox{univ. cover}& \pi_1(S^0\setminus D)\\*[1ex]

1654: 1 & A_8 & \z/3 & \z/3 & B_{3111}\p^2 & \z/3\\

1655: 1 & A_7+A_1  & \z/4  &\z/4 & B_{2111}\p^2& \z/4\\

1656: 1 & A_5+A_2+A_1 & \z/6 & \z/6 &B_{111}\p^2& \z/6\\

1657: 1 & 4A_2 & (\z/3)^2 & (\z/3)^2 & \p^2& G_{27}\\

1658: 1 & 2A_3+2A_1 & \z/2+\z/4 & \z/2+\z/4 &\p^1\times \p^1 & G_{16}\\

1659: 1 & 2A_4 & \z/5 & \z/5 &B_{1111}\p^2 & \z/5\\

1660: 1 & A_3+4A_1 & (\z/2)^2 & (\z/2)^3+\z/4 & B_{22}\p^1\times \p^1  & \z/2+\z/4\\

1661: 2 & A_7  & \z/2 & \z/4 & B_{41}\p^2& \z/2\\

1662: 2 & A_5+A_2 & \z/3 & \z/6 &B_{11}Q & \z/3\\

1663: 2 & 2A_3+A_1 & \z/4 & \z/2+\z/4 &\p^1\times \p^1& \z/2+\z/4 \\

1664: 2 & 6A_1 & (\z/2)^2 & (\z/2)^4 &\p^1\times \p^1& G_8 \\

1665: 2 & 2A_3 & \z/2 & \z/2+\z/4 & B_{22}\p^1\times \p^1 & \z/4\\

1666: 3 & A_5+A_1 & \z/2 & \z/6 & B_3\p^2 & \z/6\\

1667: 3 & 3A_2 & \z/3 & (\z/3)^2 & \p^2& (\z/3)^2\\

1668: 4 & A_3+2A_1 & \z/2 & \z/2+\z/4 & Q & (\z/2)^2 \\

1669: 4 & 4A_1 & \z/2 & (\z/2)^3 & \p^1\times \p^1 & (\z/2)^2\\

1670: \end{array}\\

1671: \ \\

1672: \mbox{Table 2}

1673: \end{array}

1674: $$

1675: Here $G_n$ is a nonabelian group of order $n$.

1676: \begin{enumerate}

1677: \item $G_8$ is the quaternion group,

1678: \item $G_{16}\subset GL_4$ is generated by

1679: $(x,y,z,t)\mapsto (z,t,-x,y)$ and 

1680: $(x,y,z,t)\mapsto (y,-x,t,-z)$.

1681: \item $G_{27}\subset GL_3$ is generated by

1682: $(x,y,z)\mapsto (y,z,x)$ and 

1683: $(x,y,z)\mapsto (x,\epsilon y,\epsilon ^2 z)$ where $\epsilon^3=1$.

1684: \end{enumerate}

1685: 

1686: 

1687: The computation of the table:

1688: The papers \cite[p.13-15]{furushima}, \cite[p.71]{mi-zh1}, 

1689: \cite[p.193]{mi-zh2} and \cite{ye}

1690: contain tables for the first 4 columns, except

1691: $\weil/\pic$ in the 4 cases with Picard number 2.  (The fundamental group of

1692: $A_7+A_1$ is listed in \cite[p.71]{mi-zh1} as $(\z/2)^2$, but it

1693: is $\z/4$.) The Picard number and the singularities of the

1694: universal cover are listed in \cite[p.71]{mi-zh1}, from this it

1695: is easy to work out where the surface is on the list \cite[7.6]{em5}.

1696: 

1697: By \cite[1.2 and 1.6]{ye}, for each singularity type there is a unique surface,

1698: except for $2A_3$ for which there is a 1--parameter family.

1699: 

1700: Once we have $\bar S\to S$ as the universal cover

1701: and $\bar D\subset \bar S$ is the preimage of $D$ then we have

1702: an exact sequence

1703: $$

1704: \pi_1(\bar S^0\setminus \bar D)\to 

1705: \pi_1(S^0\setminus  D)\to \pi_1(S^0)\to 1.

1706: $$

1707: Each time $\bar S$ is obtained by a blow up of weight 1,

1708: the resulting $\p^1\subset \bar S$ intersects $\bar D$ transversally

1709: at 1 point, so $\pi_1(\bar S^0\setminus \bar D)=1$.

1710: In the remaining cases one needs to write down the

1711: action of the $\pi_1(S^0)$ and see how it lifts to

1712: the universal cover of $\bar S^0\setminus \bar D$.

1713: 

1714: Computing any entry of the table is an elementary task.

1715: Some computations are quick but a few are quite tedious.

1716: It is unfortunately easy to miss or misdraw a $-1$-curve

1717: after performing many blow ups, so anyone wishing to rely on

1718: a particular entry is advised to recheck it.

1719: \end{say}

1720: 

1721: 

1722: In  the simply connected case there are many isomorphisms between

1723: blow ups, but this does not happen for the general case.

1724: 

1725: \begin{lem}\label{no.two.isom}  Let $S$ be a Del Pezzo surface

1726:  with cyclic quotient singularities such that $|\pi_1(S^0)|> 1$.

1727: There is a unique line in Table 2 such that $S$ is a weighted blow up

1728: of a surface on that line.

1729: (We do not claim that the blow up itself is unique.)

1730: \end{lem}

1731: 

1732: Proof. The blow ups do not change the fundamental group

1733: \cite[7.3]{em5} and we create only $A_1$ and $A_2$ singularities

1734: since we can only have blow ups $B_{m_1,\dots,m_r}T$

1735: where $\sum m_i<\deg T-1\leq 3$.

1736: It turns out that the fundamental group and the collection

1737: of $A_i:i\geq 3$ singularities uniquely determine in which line

1738: of  Table 2 the surface is. The Picard number and the

1739: $A_i:i\leq 2$ singularities now determine the number and

1740: type of  blow ups performed.\qed

1741: \medskip

1742: 

1743: 

1744: If $\deg T=1$ (resp.\ $2,3,4$) then we get $1$ (resp.\ $2,4,7$) blown up 

1745: surfaces,

1746: including $T$ itself. 

1747: Thus we get

1748: 39 deformation types of  Del Pezzo surfaces

1749:  with cyclic quotient singularities such that $|\pi_1(S^0)|> 1$.

1750: The 93 cases where $\pi_1(S^0)=1$ were enumerated

1751: in \cite[7.6]{em5}, giving

1752:  a total of 132 deformation types.

1753: 

1754: \begin{say}[Existence of smooth Seifert bundles] \label{sm.sb.ex.say}

1755: Let $S$ be one of the surfaces in Table 2

1756: and $D\in |-K_S|$ a smooth elliptic curve.

1757: As in (\ref{corr181.say}),

1758: we are considering Seifert bundles

1759: $Y(S,B,\frac{b}{m}D)$ where $B$ is a Weil divisor class on $S$.

1760: 

1761: By (\ref{corr181.say}.1) $Y(S,B,\frac{b}{m}D)$ is smooth iff 

1762:  $B$ generates the local class group at every  point.

1763: 

1764: Finding all such $B$ 

1765:  requires a detailed computation of the divisor class group

1766: $\weil(S)$ and the restriction map

1767: $$

1768: \weil(S)/\pic(S)\to \sum_{p\in \sing S} \weil(p,S).

1769: $$

1770: On a Del Pezzo surface of degree $\leq 7$, the

1771: curves $C$ with $(C\cdot K_S)=-1$ generate $\weil(S)$.

1772: On the minimal desingularization $S'\to S$ these are the $-1$ curves.

1773: Thus if we have a description of $S'$ as a blow up

1774: of $\p^2$, we see all such curves by looking at

1775: lines through 2 blow up points, conics through 5 blow up points,

1776: etc. (See \cite[Sec.26]{manin} for the complete list in degrees 2 and 1.)

1777: The description given in \cite{furushima} gives exactly these

1778: blow ups. See also \cite[Figure 1]{mi-zh1}.

1779: 

1780: In some cases 

1781: $\weil(S)/\pic(S)$ is too small to get surjection onto

1782: some $\weil(p,S)$, and then  

1783:  there are no smooth Seifert bundles at all.

1784: This happens in 3 cases:

1785: $$

1786: A_8, A_7+A_1, A_7.

1787: $$

1788: More surprising is the mildly singular $A_3+2A_1$ case

1789: which again has no smooth Seifert bundle over it.

1790: On the minimal desingularization, the configuration of

1791: $-1$ and $-2$ curves is

1792: $$

1793: \stackrel{-2}{\circ} - \stackrel{-1}{\circ} - \stackrel{-2}{\circ} - 

1794: \stackrel{-2}{\circ} - \stackrel{-2}{\circ} - \stackrel{-1}{\circ} - 

1795: \stackrel{-2}{\circ}

1796: $$

1797: In order to generate both $\z/2$ on the ends, we need to take the

1798: two $-1$-curves with odd coefficients, but then

1799: we get only twice the generator of the $\z/4$

1800: of the middle singularity.

1801: 

1802: The more complicated $A_5+A_1$ case leads to the configuration

1803: $$

1804: \stackrel{-2}{\circ} - \stackrel{-1}{\circ} - \stackrel{-2}{\circ} - 

1805: \stackrel{-2}{\circ} - \stackrel{-2}{\circ} - 

1806: \stackrel{-2}{\circ} - 

1807: \stackrel{-2}{\circ}

1808: $$

1809: plus an extra $-1$-curve. The $-1$-curve shown

1810: generates both local class groups, so there are

1811: smooth Seifert bundles, even with SE structure. 

1812: 

1813: A glance at the diagrams for $A_5+A_2+A_1$ and for

1814: $2A_4$ in \cite[Figure 1]{mi-zh1} shows $-1$-curves

1815: which generate all local class groups.

1816: 

1817: 

1818: As another concrete example, 

1819: the group $G_{27}$ operates freely on 

1820: $\c^3$ outside the origin, thus

1821: $S^5/G_{27}\to \p^2/(\z/3)^2$ is a smooth Seifert bundle.

1822: \end{say}

1823: 

1824: Just for illustration, let us compute one simple case

1825: completely.

1826: 

1827: \begin{exmp}[$3A_2$ case]  We can write this as

1828: $S=\p^2/(\z/3)$ by the action 

1829: $$

1830: (x:y:z)\mapsto (x:\epsilon y: \epsilon^2 z)\qtq{where} \epsilon^3=1.

1831: $$

1832: We can take $D=(x^3+y^3+z^3=0)$. The universal cover of

1833: $\p^2\setminus D$ is the cubic

1834: $$

1835: (x^3+y^3+z^3=t^3)\subset \p^3.

1836: $$

1837: We get a $(\z/3)^2$-action generated by

1838: $$

1839: (x:y:z:t)\mapsto (x:\epsilon y: \epsilon^2 z:t)\qtq{and} 

1840: (x:y:z:t)\mapsto (x: y: z:\epsilon t).

1841: $$

1842: 

1843: The 3 coordinate lines give the curves $A,B,C\subset \p^2/(\z/3)$.

1844: These generate $\weil(S)$ subject to the relations

1845: $3A=3B=3C=A+B+C$. We can thus rewrite

1846: $$

1847: \weil(S)=\z[A]+\z/3[A-B].

1848: $$

1849: By explicit computations, the class

1850: $B_{uv}:=uA+v(A-B)$ corresponds to a smooth Seifert bundle

1851: iff

1852: none of the numbers $u, v, u+v$ is divisible by 3.

1853: Whenever this holds, there is a smooth pre-SE  Seifert bundle

1854: $L_{uv}\to (S, B_{uv}, (1-\frac1{m})D)$ for any $(m,3)=1$.

1855: It has a Sasakian--Einstein metric for every $m\geq 4$.

1856: \end{exmp}

1857: 

1858: \begin{rem} As a consequence of the classification,

1859: \cite[Thm, p.184]{mi-zh2} concludes that the fundamental group

1860: of a  Del Pezzo surface with Du Val singularities is abelian.

1861: 

1862: This is easy to see directly as follows. Let $g:S'\to S$ be

1863: the universal cover. Pick a smooth elliptic curve

1864: $C\in |-K_S|$. Then $C':=g^{-1}(C)\in |-K_{S'}|$, thus

1865: it is also a smooth elliptic curve. 

1866: Hence the fundamental group is the same as the kernel

1867: of the group homomorphism $C'\to C$, hence an abelian group

1868: with at most 2 generators.

1869: \end{rem}

1870: 

1871: 

1872: 

1873: 

1874: 

1875: 

1876: 

1877: 

1878: 

1879: 

1880: 

1881: 

1882: 

1883: 

1884: 

1885: 

1886: 

1887: 

1888: 

1889: \section{Klt conditions}

1890: \label{klt.sect}

1891: 

1892: 

1893: While we did not use it in the proof, it is instructive to see

1894: the relationship between finding $S'$ and the

1895: klt condition for $(S^m,(1-\tfrac1{r})D^m+\Delta^m+H^m)$.

1896: 

1897: Since $\Delta'$ is a nonnegative linear combination, all

1898: exceptional curves of $S'\to S^m$ have nonpositive

1899: discrepancy with respect to

1900: $(S^m,(1-\tfrac1{r})D^m+\Delta^m+H^m)$. If the latter

1901: pair is klt, then, by an observation of Shokurov,

1902: there are only finitely many such curves on

1903: any birational model of $S^m$ and they can be found

1904: explicitly (cf.\ \cite[2.36]{kmbook}). 

1905: 

1906: The problem with using this result is that

1907:  we do not know $\Delta^m$ and $H^m$, only

1908: the numerical class 

1909: $$

1910:  \Delta^m+H^m\equiv -\left(K_{S^m}+(1-\tfrac1{r})D^m\right).

1911: $$

1912: The usual proofs of the above finiteness result start

1913: by taking a log resolution of 

1914: $(S^m,(1-\tfrac1{r})D^m+\Delta^m+H^m)$, and in essence

1915: we would need to understand all smooth surfaces

1916: dominating $S^m$.

1917: 

1918: The key part of the proofs 

1919: in (\ref{5and3.pf}) and (\ref {corr181.say}) is to describe all

1920: exceptional curves over $S^m$ which have nonpositive

1921: discrepancy with respect to some

1922: $(S^m,(1-\tfrac1{r})D^m+\Delta^m+H^m)$.

1923: 

1924: To this end one can use 

1925: the following result which sharpens the klt conditions

1926: used in \cite[Sec.8]{em5}.

1927: 

1928: 

1929: \begin{prop} \label{klt.crit} Let $S$ be a smooth surface,

1930: $C$ a smooth curve on $S$ and $D$ an effective $\q$-divisor

1931: on $S$ such that $C\not\subset \supp D$.

1932:  Let $p\in S$ be  point and $n$ a natural number.

1933: Then $(S, (1-\frac1{n})C+cD)$ is klt at $p$ for

1934: $$

1935: c< \min\left\{

1936: \frac1{(C\cdot D)_p}+\frac1{n\cdot \mult_pD},\frac1{\mult_pD}\right\}.

1937: $$

1938: \end{prop}

1939: 

1940: Proof. Choose local coordinates $(x,y)$ such that

1941: $C=(y=0)$ and let $f(x,y)=0$ be an equation of $mD$ for some $m$

1942: such that $mD$ is an integral divisor.

1943: 

1944: Consider the local degree $n$ cover $\pi:T\to S$ given by $y=z^n$.

1945: By \cite[20.3.2]{k-etal},  $(S, (1-\frac1{n})C+cD)$ is klt at $p$

1946: iff $(T, \frac{c}{m}(f(x,z^n)=0))$ is klt at $p$.

1947: 

1948: We aim to apply a theorem of Var\v cenko \cite{varc}

1949: which gives a condition for  $(T, \gamma(g(x,z)=0))$ to be klt

1950: in terms of the Newton polygon of $g$ in a suitable coordinate system

1951: which is achieved after a series of coordinate changes

1952: of the form $(x,z)\mapsto (x-\alpha z^i, z)$ or

1953: $(x,z)\mapsto (x, z-\alpha x^i)$.

1954: The problem is that we can handle only those coordinate changes

1955: which are compatible with $\pi$. 

1956: That is, those of the form $(x,z)\mapsto (x-\alpha z^{ni}, z)$. 

1957: Thus we have to look

1958: carefully at the proof, not just the final result.

1959: 

1960: We state the result in the rather artificial form that we 

1961: need.

1962:  The reader should consult the proof given in

1963: \cite[6.40]{ksc}, especially pp.172--3.

1964: 

1965: \begin{lem} \label{varch.thm}

1966: Write $g(x,z)=\sum b(i,j)x^iz^j$. Assume that one of the

1967: following holds.

1968: \begin{enumerate}

1969: \item (Main case.) There are $(i,j)$ and $(i',j')$ such that $b(i,j)\neq 0$,

1970: $b(i',j')\neq 0$ and the line segment

1971: $[(i,j), (i',j')]$ contains a point $(\gamma,\gamma)$ with $\gamma< c^{-1}$.

1972: \item (Degenerate case.) There is $(i,j)$  such that $b(i,j)\neq 0$

1973: and $i,j<c^{-1}$.

1974: \end{enumerate}

1975: Then one of the following also holds:

1976: \begin{enumerate}\setcounter{enumi}{2}

1977: \item (Klt case.) $(T, c(g(x,z)=0))$ is klt.

1978: \item (Coordinate change.) There are natural numbers

1979: $u,v,w$ and $e$  such that $(u,v)=1$,  $e>w/(u+v)$ and 

1980: $$

1981: \sum_{ui+vj=w} b(i,j)x^iy^j\qtq{is divisible by}

1982: (\alpha x^v+\beta y^u)^e.

1983: $$

1984: In this case necessarily $u=1$ or $v=1$, there is a unique such factor

1985: and the  new coordinates are

1986: $(x,\alpha x^v+\beta y)$ if $u=1$ and

1987: $(\alpha x+\beta y^u,y)$ if $v=1$.

1988: \end{enumerate}

1989: \end{lem}

1990: 

1991: 

1992: Set $a=(L\cdot D)_p$ and $d=\mult_pD$.

1993: As before,  write

1994: $f=\sum b(i,j)x^iy^j$.

1995: Then $b(am,0)\neq 0$ and

1996: $b(i,j)\neq 0$ for some $i+j=md$.

1997: Thus

1998: $$

1999: f(x,z^n)= \sum b(i,j)x^iz^{nj}.

2000: $$

2001: If $b(i,j)\neq 0$ for some $i+j=md$ and $j\geq md/(n+1)$ then 

2002: the main case 

2003: (\ref{varch.thm}.1) applies using the  line segment $[(am,0), (i,nj)]$.

2004: The maximum value of $\gamma$ is achieved when $i=0$, giving the condition

2005: $c<a^{-1}+(nd)^{-1}$.

2006: 

2007: Otherwise $b(i,j)\neq 0$ for some $i+j=md$ and $j< md/(n+1)$

2008: Thus $i<md$. $nj<md$ and  (\ref{varch.thm}.2) applies, giving the condition

2009: $c<d^{-1}$.

2010: 

2011: Thus either  $(T, \frac{c}{m}(f(x,z^n)=0))$ is klt

2012: as required or we are in the 

2013:  coordinate change case (\ref{varch.thm}.4). 

2014: Thus we consider

2015: $\sum_{ui+vnj=w} b(i,j)x^iz^{nj}$.

2016: Set $U=u/(u,n), V=vn/(u,n), W=w/(u,n)$

2017: and look at the irreducible factorization

2018: $$

2019: \sum_{Ui+Vj=W} b(i,j)x^iy^j=

2020: \prod_{\ell} (\alpha_{\ell} x^V+\beta_{\ell} y^U)^{e_{\ell}}.

2021: $$

2022: Correspondingly, 

2023: $$

2024: \sum_{ui+vnj=w} b(i,j)x^iz^{nj}=

2025: \prod_{\ell} (\alpha_{\ell} x^V+\beta_{\ell} z^{Un})^{e_{\ell}}.

2026: $$

2027: Since (\ref{varch.thm}.4) applies, either $\min\{V,Un\}=1$, that is 

2028: $V=1$, or some $(\alpha_{\ell} x^V+\beta_{\ell} z^{Un})$

2029: factors further. However, in the latter case we have

2030: $(V,Un)>1$ factors of the same multiplicity

2031: $e_{\ell}>w/(v+un)$, which is impossible.

2032: Thus 

2033: $V=1$ and  the coordinate change is

2034: of the form $(x,z)\mapsto (x-\gamma z^{Un}, z)$

2035: which can be realized on $S$ as

2036: $(x,y)\mapsto (x-\gamma y^{U}, y)$.\qed

2037: \medskip

2038: 

2039: The following consequences show that the

2040: klt condition is satisfied for the cases

2041: leading to $H_2(L,\z)=(\z/m)^2$ for $m\geq 7$

2042: and for $H_2(L,\z)=(\z/3)^6$. This is why I expect 

2043: to have only finitely many families for them.

2044: 

2045: \begin{cor}\label{klt.cond.cor1} Let $C\subset S$ be a smooth elliptic curve

2046: where $S$ is  $\p^2, \p^1\times \p^1$ or $\f_2$.

2047: Let $D$ be an effective $\q$-divisor such that

2048:  $D\equiv \frac1{n}C$. Then

2049: $(S, (1-\frac1{n})C+D)$ is klt for $n\geq 7$.

2050: \end{cor}

2051: 

2052: Proof. Start with the $S=\p^2$ case.

2053: If $H\subset \p^2$ is a line then

2054: $C\equiv 3H$ and so $D\equiv \frac3{n}H$.

2055: Thus $(C\cdot D)_p\leq \frac9{n}$ and 

2056: $\mult_pD\leq (H\cdot D)=\frac3{n}$. Since

2057: $$

2058: \tfrac{n}9+\tfrac13>1\qtq{for $n\geq 7$,}

2059: $$

2060: the rest   follows from (\ref{klt.crit}).

2061: The other two cases are similarly easy.

2062: \qed

2063: 

2064: \begin{cor} Let $C\subset \f_4$ be a smooth curve

2065: of genus 3 such that $C\equiv 2E+8F$.

2066: Let $D$ be an effective $\q$-divisor such that

2067:  $D\equiv -(K+\frac23C)$. Then

2068: $(\f_3, \frac23C+D)$ is klt.

2069: \end{cor}

2070: 

2071: Proof. Here  $D\equiv \frac23(E+F)$, 

2072: thus 

2073: $$

2074: (C\cdot D)_p\leq \tfrac23\cdot (2E+8F)\cdot (E+F)=\tfrac43,

2075: $$ 

2076: and 

2077: $\mult_pD\leq \frac23$. Since

2078: $\tfrac34+\frac12>1$

2079: the rest   follows from (\ref{klt.crit}).

2080: \qed

2081: \medskip

2082: 

2083: One can also use (\ref{klt.crit}) to check

2084:  the existence condition \cite{dk} for orbifold

2085: K\"ahler--Einstein metrics.

2086: 

2087: \begin{cor}\label{ke.conds} Let $S$ be a Del Pezzo surface

2088: with quotent singularities and $C\in |-K_S|$ a smooth elliptic curve.

2089: Then $(S, (1-\frac1{n})C)$ has an orbifold

2090: K\"ahler--Einstein metric whenever $n>\frac23 K_S^2$.

2091: \end{cor}

2092: 

2093: Proof. Set $d=K_S^2$. Let $D$ be an effective $\q$-divisor such that

2094:  $D\equiv \frac1{n}C$.  We need to check that

2095: $(S, (1-\frac1{n})C+\frac23D)$ is klt.

2096: As in (\ref{klt.cond.cor1}), this holds if

2097: $$

2098: \tfrac23 < \min\left\{\tfrac{n}{d}+\tfrac{n}{nd}, \tfrac{n}{d}\right\}=

2099: \tfrac{n}{d}.\qed

2100: $$

2101: 

2102:  \begin{ack}  I thank Ch.\ Boyer and K.\ Galicki for many comments

2103: and corrections and the University of Utah where some of this work was done.

2104: Partial financial support  was provided by  the NSF under grant number 

2105: DMS-0500198. 

2106: \end{ack}

2107: 

2108: \bibliography{refs}

2109: 

2110: \vskip1cm

2111: 

2112: \noindent Princeton University, Princeton NJ 08544-1000

2113: 

2114: \begin{verbatim}kollar@math.princeton.edu\end{verbatim}

2115: 

2116: 

2117: 

2118: 

2119: 

2120:  

2121: \end{document}

2122: 

2123: 

2124: 

2125: \begin{say}[The $(\z/m)^2$ cases] By \cite[6.5]{em5}, there are

2126: 4 series where $H_1(L,\z)=0$, $H_2(L,\z)\cong (\z/m)^2$ and these

2127: exhaust all possibilities when $m\geq 12$.

2128: Here $S$ is one of the surfaces 

2129: $$

2130: \p^2, \p(1,1,2), S_5, \p(1,2,3),

2131: $$

2132: where $S_5\subset \p^5$ is a  certain degree 5 Del Pezzo surface

2133: which can also be given as a degree 6 hypersurface

2134: in $\p(1,2,3,5)$. Correspondingly, we get the sample equations

2135: \begin{enumerate}

2136: \item $x^3+y^3+z^3+t^m$ for $(3,m)=1$,

2137: \item $x^2+y^4+z^4+t^m$ for $(2,m)=1$,

2138: \item $x^2+y^3+z^6+zt^m$ for $(5,m)=1$.

2139: \item $x^2+y^3+z^6+t^m$ for $(6,m)=1$,

2140: \end{enumerate}

2141: \end{say}

2142: 

2143:  

2144: