1: 

2: 

3: 

4: \documentclass[12pt]{amsart}

5: 			 

6: \usepackage{amsfonts}

7: \usepackage{amssymb}

8: \usepackage{amscd, graphics}

9: \usepackage{epsfig}

10: \input xy

11: \xyoption{all}

12: 

13: \setlength{\parindent}{.4 in}

14: \setlength{\textwidth}{5.8 in}

15: \setlength{\topmargin} {-.3 in}

16: \setlength{\evensidemargin}{0 in}

17: \setlength{\oddsidemargin}{0 in}

18: \setlength{\footskip}{.3 in}

19: \setlength{\headheight}{.3 in}

20: \setlength{\textheight}{8.5 in}

21: \setlength{\parskip}{.1 in}

22: 

23: \newcommand{\marginlabel}[1]%

24:   {\mbox{}\marginpar{\raggedleft\hspace{0pt}\bfseries\sf#1}}

25: 

26: 

27: %%----------------------------------------------------------

28: %%  SHORTCUTS

29: %%----------------------------------------------------------

30: \def\ZZ{{\mathbb Z}}

31: \def\NN{{\mathbb N}}

32: \def\FF{{\mathbb F}}

33: \def\CC{{\mathbb C}}

34: \def\AA{{\mathbb A}}

35: \def\RR{{\mathbb R}}

36: \def\QQ{{\mathbb Q}}

37: \def\PP{{\textbf P}}

38: \def\OO{{\mathcal O}}

39: 

40: 

41: \def\cD{\mathcal{D}}

42: \def\cA{\mathcal{A}}

43: \def\cF{\mathcal{F}}

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

45: \def\cO{\mathcal{O}}

46: \def\cI{\mathcal{I}}

47: \def\I{\mathcal{I}}

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

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

50: \def\cU{\mathcal{U}}

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

52: \def\cE{\mathcal{E}}

53: 

54: \def\Mg{\cM_{g}}

55: \def\Mgn{\cM_{g,n}}

56: \def\Mg1{\cM_{g,g+1}}

57: \def\Pici{{\rm Pic}^{g-2i-1}(C)}

58: \def\hsigma{\widehat{\sigma}}

59: \def\iso{\simeq}

60: \def\reg{\rm reg}

61: \def\tensor{\otimes}

62: 

63: %%----------------------------------------------------------

64: %%  NEW OPERATORS

65: %%----------------------------------------------------------

66: \DeclareMathOperator{\Coker}{Coker}

67: \DeclareMathOperator{\Ker}{Ker}

68: \DeclareMathOperator{\depth}{depth}

69: \DeclareMathOperator{\Pic}{Pic}

70: \DeclareMathOperator{\Hom}{Hom}

71: \DeclareMathOperator{\HOM}{\mathcal{Hom}}

72: \DeclareMathOperator{\rank}{rank}

73: \DeclareMathOperator{\Spec}{Spec}

74: \DeclareMathOperator{\Cl}{Cl}

75: 

76: 

77: %%----------------------------------------------------------

78: %%  OTHER ENVIRONMENTS    

79: %%----------------------------------------------------------

80: 

81: %%\swapnumbers

82: 

83: \theoremstyle{plain}

84: \newtheorem*{introtheorem}{Theorem}   

85: \newtheorem*{introcorollary}{Corollary}

86: \newtheorem{theorem}{Theorem}[section]

87: \newtheorem{proposition}[theorem]{Proposition}

88: \newtheorem{corollary}[theorem]{Corollary}

89: \newtheorem{lemma}[theorem]{Lemma}

90: 

91: \theoremstyle{definition}

92: \newtheorem{definition}[theorem]{Definition}

93: \newtheorem{remark}[theorem]{Remark}

94: \newtheorem{example}[theorem]{Example}

95: \newtheorem{sketch}[theorem]{Sketch}

96: \newtheorem{conjecture}[theorem]{Conjecture}

97: \newtheorem{conjecture/question}[theorem]{Conjecture/Question}

98: \newtheorem{question}[theorem]{Question}  

99: 

100: \pagestyle{myheadings}

101: \theoremstyle{remark}

102: \newtheorem*{nctheorem}{Theorem}

103: 

104: 

105: %%----------------------------------------------------------

106: %%  TITLE PAGE INFO  

107: %%----------------------------------------------------------

108: \begin{document}

109: 

110: \title{Divisors on $\Mg1$ and the Minimal Resolution Conjecture

111: for points on canonical curves}

112: 

113: \author[G. Farkas]{Gavril Farkas}

114: \address{Department of Mathematics, University of Michigan, 

115: 525 East University, Ann Arbor, MI, 48109-1109}

116: \email{{\tt gfarkas@umich.edu}}

117: 

118: \author[M. Musta\c{t}\v{a}]{Mircea~Musta\c{t}\v{a}}

119: \address{Department of Mathematics, University of California,

120: Berkeley, CA, 94720 and Institute of Mathematics of

121: the Romanian Academy}

122: \email{{\tt mustata@math.berkeley.edu}}

123: 

124: \author[M. Popa]{Mihnea Popa} 

125: \address{Department of Mathematics, Harvard University, 

126: One Oxford Street, Cambridge, MA 02138}

127: \email{{\tt mpopa@math.harvard.edu}}

128: 

129: \subjclass{Primary 14H10; Secondary 13D02, 14F05.}

130: \keywords{Minimal free resolutions, difference varieties, moduli spaces of curves with 

131: marked points}

132: \maketitle

133: \markboth{G. FARKAS, M. MUSTA\c T\v A and M. POPA}

134: {\bf THE MINIMAL RESOLUTION CONJECTURE}

135: 

136: %%----------------------------------------------------------

137: %%  INTRODUCTION    

138: %%----------------------------------------------------------

139: \section*{\bf Introduction}

140: 

141: The Minimal Resolution Conjecture for points in projective space has 

142: attracted considerable attention in recent years, starting with the 

143: original \cite{lorenzini1}, \cite{lorenzini2} and continuing most notably 

144: with \cite{gaeta}, \cite{BG}, \cite{walter},

145: \cite{simpson}, \cite{popescu}, \cite{epsw}. The purpose

146: of this paper 

147: is to explain how a completely analogous problem can be formulated for 

148: sets of points on arbitrary varieties embedded in projective space, and 

149: then study in detail the case of curves. Similarly to 

150: the well-known analysis of syzygies of curves carried out by Green and 

151: Lazarsfeld (\cite{gl1}, \cite{gl2}, \cite{gl3}), we divide our work into

152: a study of resolutions of points on canonical curves and on curves of 

153: large degree. The central result of the paper states that the Minimal 

154: Resolution Conjecture is true on any canonical curve. In contrast, it 

155: always fails for curves embedded with large degree, although a weaker 

156: result, called the Ideal Generation Conjecture, holds also in this case.

157: These results turn out to have surprisingly deep connections with the 

158: geometry of difference varieties in Jacobians, special divisors on 

159: moduli spaces of curves with marked points, and moduli spaces of stable

160: bundles.  

161: 

162: Let $X$ be a projective variety over an algebraically closed field,

163: embedded by a (not necessarily complete) linear series. 

164: We begin by formulating a general version of the 

165: Minimal Resolution Conjecture (MRC), in analogy with the case of $\PP^n$,

166: predicting how the Betti numbers of 

167: a general subset of points of $X$ in the given embedding are related to the 

168: Betti numbers of $X$ itself. More precisely (cf. Theorem 

169: \ref{general_results} below),  

170: for a large enough general set of points $\Gamma$ on $X$, the Betti diagram 

171: consisting of the graded Betti numbers $b_{i,j}(\Gamma)$ is obtained from 

172: the Betti diagram of $X$ by adding two more nontrivial rows, at places 

173: well determined by the length of $\Gamma$. Recalling that the Betti diagram 

174: has the Betti number $b_{i,j}$ in the $(j,i)$-th position, and assuming 

175: that the two extra rows are indexed by $i=r-1$ and $i=r$, for some integer 

176: $r$, the MRC predicts that 

177: $$b_{i+1,r-1}(\Gamma)\cdot b_{i,r}(\Gamma)= 0,$$

178: i.e. at least one of the two Betti numbers on any "diagonal" is zero. 

179: As the difference $b_{i+1, r-1} - b_{i,r}$ can be computed exactly,

180: this implies a precise knowledge of the Betti numbers in these two rows.

181: Summing up, knowing the Betti diagram of $\Gamma$ would be the same as knowing 

182: the Betti diagram of $X$. A subtle question is however to understand

183: how the shape of the Betti diagram of $X$ influences whether MRC

184: is satisfied for points on $X$. An example illustrating this is given 

185: at the end of Section 1.

186: 

187: The Minimal Resolution Conjecture

188:  has been extensively studied in the case $X=\PP^n$. The

189: conjecture holds for $n\leq 4$ by results of

190: Gaeta, Ballico and Geramita, and Walter (see \cite{gaeta},

191: \cite{BG} and

192: \cite{walter}, respectively).

193: Moreover, Hirschowitz and Simpson 

194: proved in \cite{simpson}

195:  that it holds if the number of points is large enough with respect

196: to $n$. However, the conjecture does not hold in general:

197: it fails for every $n\geq 6$, $n\neq 9$ for almost $\sqrt{n}/2$

198: values of the number of points, by a result of Eisenbud, Popescu, Schreyer

199: and Walter (see \cite{epsw}). We refer to \cite{popescu} and

200: \cite{epsw} for a nice introduction and an account of the present status

201: of the problem in this case.

202: 

203: 

204: The main body of the paper is dedicated to a detailed study of MRC

205: in the case of curves. 

206: We will simply say that a curve \emph{satisfies MRC} in a given embedding 

207: if MRC is satisfied by a general set of points $\Gamma$ of any 

208: sufficiently large degree

209: (for the precise numerical statements see Section 1). We will also 

210: sometimes say that MRC holds for a line bundle $L$ if it 

211: holds for $C$ in the embedding given by $L$.  

212: Our main result says that MRC holds in the most significant case,

213: namely the case of canonically embedded nonhyperelliptic curves.

214: 

215: \begin{introtheorem}

216: If $C$ is a canonical curve, then $C$ satisfies MRC.

217: \end{introtheorem}

218: 

219: In contrast, under very mild assumptions on the genus, 

220: the MRC always fails in the case of curves of large degree, at well-determined 

221: spots in the Betti diagram (cf. Section 2 for precise details). The statement

222: $b_{2,r-1}\cdot b_{1,r}=0$, i.e. the case $i=1$, does hold though; this is 

223: precisely the Ideal Generation Conjecture, saying that the 

224: minimal number of generators of $I_\Gamma / I_X$ is as small as possible.

225: 

226: \begin{introtheorem}

227: (a) If $L$ is a very ample line bundle of degree $d\geq 2g$, then $IGC$ 

228: holds for $L$.

229: \newline

230: (b) If $g\geq 4$ and $L$ is a line bundle of degree $d\geq 2g+16$,

231: then there exists 

232: a value of $\gamma$ such that $C\subseteq\PP H^{0}(L)$ does not satisfy MRC 

233: for $i=\lfloor\frac{g+1}{2}\rfloor$. The same holds if $g\geq 15$ and

234:  $d\geq 2g+5$.

235: \end{introtheorem}

236: 

237: It is interesting to note that by the "periodicity" property of Betti

238: diagrams of 

239: general points on curves (see \cite{mustata} \S2), the theorem above

240: implies that 

241: on curves of high degree, MRC fails for sets of points of

242: arbitrarily large length.

243: This provides a very different picture from the case of projective space 

244: (cf. \cite{simpson}),

245: where asymptotically the situation is as nice as possible.

246: 

247: We explain the strategy involved in the proof of these results

248:  in some detail, as 

249: it appeals to some new geometric techniques in the study of

250:  syzygy related questions.

251: For simplicity we assume here that $C$ is a smooth curve embedded

252:  in projective space 

253: by means of a complete linear series corresponding to a very ample

254:  line bundle $L$

255: (but see \S2 for more general statements).

256:  A well-known geometric approach, developed  

257: by Green and Lazarsfeld in the study of syzygies of curves

258:  (see \cite{lazarsfeld} for a survey),

259: is to find vector bundle statements equivalent to the algebraic ones,

260:  via Koszul cohomology.

261: This program can be carried out completely in the case of MRC,

262:  and for curves we get a 

263: particularly clean statement. Assume that $M_L$ is the kernel of

264:  the evaluation map

265: $H^0(L)\otimes \OO_C\rightarrow L$ and $Q_L:=M_L^{*}$.

266: Then (cf. Corollary \ref{criterion} below) MRC holds 

267: for a collection of $\gamma\geq g$ general points on $C$

268:  if and only if the following 

269: is true:

270: $$h^0(\wedge^i M_L \otimes \xi)=0, ~{\rm for~all}~ i  ~{\rm and}~

271: \xi\in

272:  {\rm Pic}^{g-1+\lfloor{\frac{di}{n}}\rfloor}(C)~{\rm general}.\,\,\,\,(*)$$

273: Condition $(*)$ above is essentially the condition studied by Raynaud  

274: \cite{raynaud}, related to the existence of theta divisors for semistable 

275: vector bundles. In the particular situation of $\wedge^i M_L$, 

276: with $L$ a line bundle of large degree, it has been 

277: considered in \cite{popa} in order to produce base points for the determinant 

278: linear series on the moduli spaces $SU_C(r)$ of semistable bundles of rank $r$ 

279: and trivial determinant. A similar approach shows here the failure of

280: condition $(*)$ (and so of MRC) for $i=[\frac{g+1}{2}]$. On the

281: other hand, the fact that 

282: IGC holds is a

283:  rather elementary application of the Base Point Free Pencil Trick

284: \cite{ACGH} III \S3.

285: 

286: The case of canonical curves is substantially more involved, but in the end 

287: one is rewarded with a positive answer. As above, it turns out that MRC is 

288: equivalent to the vanishing:

289: $$h^0(\wedge^i Q\otimes \xi)=0, ~{\rm for~all}~i ~{\rm and~}~\xi\in

290:  {\rm Pic}^{g-2i-1}(C) 

291: ~{\rm general},$$

292: where $Q$ is the dual of the bundle $M$ defined by the evaluation sequence: 

293: $$0\longrightarrow M\longrightarrow H^0(\omega_C)\otimes \OO_C\longrightarrow 

294: \omega_C\longrightarrow 0.$$

295: As the slope of $\wedge^i Q$ is $2i\in \ZZ$, this is in turn equivalent to the 

296: fact that $\wedge^i Q$ has

297:  a theta divisor $\Theta_{\wedge^i Q}\in {\rm Pic}^{g-2i-1}(C)$. 

298: On a fixed curve, if indeed a divisor,

299: $\Theta_{\wedge^i Q}$ will be identified as being precisely 

300: the difference variety $C_{g-i-1}-C_i \subseteq\Pici$ (cf. \cite{ACGH} Ch.V.D), where 

301: $C_n$ is the $n$-th symmetric product of $C$.

302: This is achieved via a 

303: filtration argument and a cohomology class calculation similar to the 

304: classical Poincar\'e theorem

305:  (cf. Proposition \ref{1}). A priori though, on an arbitrary curve 

306: the nonvanishing locus

307:  $\{\xi~|~h^0(\wedge^i Q\otimes \xi)\neq 0\}$ may be the whole $\Pici$, 

308: in which case this identification 

309: is meaningless. We overcome this problem by working with all curves at once, that 

310: is by setting up a similar universal construction on the moduli space 

311: of curves with marked points $\Mg1$.

312:  Here we slightly oversimplify the exposition in order 

313: to present the main idea,

314:  but for the precise technical details see Section 3. 

315: We essentially consider the "universal nonvanishing locus" in $\Mg1$: 

316: $$\cZ = \{(C,x_1,\ldots ,x_{g-i},y_1,\ldots ,y_{i+1})|h^0(\wedge^i

317: Q_C\otimes\OO(x_1+\ldots 

318: +x_{g-i}-y_1-\ldots -y_{i+1}))\neq 0\}.$$

319: The underlying idea is that the difference line bundles $\OO_C(x_1+\ldots 

320: +x_{g-i}-y_1-\ldots -y_{i+1})$ in fact cover the whole $\Pici$ (i.e. 

321: $C_{g-i}-C_{i+1}= \Pici$), and so 

322: for any given curve $C$,

323:  $\cZ|_C$ is precisely the nonvanishing locus described above. 

324: The advantage of writing

325: it in this form is that we are led to performing a computation on $\Mg1$ rather

326: than on a universal Picard,

327:  where for example one does not have a canonical choice of 

328: generators for the Picard group.  

329: A ``deformation to hyperelliptic

330: curves'' argument easily implies that MRC holds for general canonical 

331: curves, so $\cZ$ is certainly a divisor.

332: We then show that $\cZ$ is the degeneracy locus of a morphism of vector

333: bundles of the 

334: same rank and compute its class using a Grothendieck-Riemann-Roch 

335: argument (cf. Proposition \ref{degeneracy_class}). 

336: 

337: On the other hand, one can define an (a priori different) divisor $D$ in

338: $\Mg1$ which 

339: is a global analogue of the preimage of $C_{g-i-1}-C_i$ in 

340: $C^{g-i}\times C^{i+1}$ via

341: the difference map. It is convenient to see $D$ as the locus of curves 

342: with marked points $(C,x_1,\ldots ,x_{g-i},y_1,\ldots ,y_{i+1})$ having a

343: $\mathfrak g^1_g$ 

344: which contains $x_1,\ldots , x_{g-i}$ in a fiber and $y_1, \ldots ,

345: y_{i+1}$ in a different fiber.

346: An equivalent formulation of the discussion above is that 

347: $D\subseteq\cZ$, and in order for MRC to hold

348:  for all canonical curves one should 

349: have precisely $D = \cZ$. As we show that $D$

350:  is reduced (cf. Proposition \ref{reduced}), it suffices 

351: then to prove that the class of $D$ coincides with that of $\cZ$.

352: To this end we consider the closure of $D$ in the 

353: compactification $\overline{\cM}_{g,g+1}$,

354:  where the corresponding boundary condition is defined

355: by means of limit linear series. The computation of the class of $D$ via 

356: this closure is essentially independent of the rest of 

357: the paper. 

358: It relies on degeneration and enumerative techniques 

359: in the spirit of \cite{HM} and \cite{EH1}.

360: 

361: The results of both this and the computation

362:  of the class of $\mathcal{Z}$ are summarized 

363: in the following theorem. For the statement,

364:  we recall that ${\rm Pic}(\mathcal{M}_{g,n})_{\QQ}$

365: is generated by the class $\lambda$ of

366:  the Hodge bundle and the classes $\psi_j$, $1\leq j\leq n$,

367: where $\psi_j:= c_1(p_j^* \omega)$,

368:  with $\omega$ the relative dualizing sheaf on the universal 

369: curve $\mathcal{C}_g \rightarrow

370:  \mathcal{M}_g$ and $p_j : \Mgn \rightarrow \mathcal{C}_g$

371: the projection onto the $j$-th factor.

372: 

373: \begin{introtheorem}

374: The divisors $\cZ$ and $D$ defined above have the 

375: same class in $\Pic(\cM_{g,g+1})_{\QQ}$, namely

376: $$-\Bigl({g-1\choose i}-10{g-3\choose {i-1}}\Bigr)

377: \lambda+{{g-2}\choose i}\Psi_x

378: +{{g-2}\choose{i-1}}\Psi_y,$$

379: where $\Psi_x=\sum_{j=1}^{g-i}\psi_j$ and $\Psi_y=\sum_{j=g-i+1}^{g+1}\psi_j$. 

380: In particular $D = \cZ$.

381: \end{introtheorem}

382: 

383: As mentioned above, this implies that $\wedge^i Q$ always has a 

384: theta divisor, for all $i$, so equivalently that MRC holds

385: for an arbitrary canonical curve. We record the more 

386: precise identification of this theta divisor, which now follows

387: in general.

388: 

389: \begin{introcorollary}

390: For any nonhyperelliptic curve $C$, $\Theta_{\wedge^i Q} = C_{g-i-1}

391: - C_i$. 

392: \end{introcorollary}

393: 

394: In this particular form, our result answers positively a conjecture 

395: of R. Lazarsfeld.

396: It is worth mentioning that it also answers negatively a question that

397: was raised in connection with \cite{popa}, namely if $\wedge^i Q$ provide

398: base points for determinant linear series on appropriate moduli

399: spaces of vector bundles.

400: 

401: 

402: The paper is structured as follows. In Section 1 we give some 

403: equivalent formulations of the Minimal Resolution Conjecture and 

404: we describe the vector bundle setup used in the rest of the paper.

405: In Section 2 we treat the case of curves embedded with large degree,

406: proving IGC and showing that MRC fails. Section 3 is devoted to the 

407: main result, namely the proof of MRC for canonical curves, and here 

408: is where we look at the relationship with difference varieties and 

409: moduli spaces of curves with marked points. The divisor 

410: class computation in $\overline{\mathcal{M}}_{g,g+1}$, on which part of 

411: the proof relies, is carried out in Section 4 by means of limit linear

412: series.

413: 

414: \medskip

415: \noindent 

416: {\bf Acknowledgments.} We would like to thank D. Eisenbud and 

417: R. Lazarsfeld for sharing with us numerous useful ideas on the subject.

418: 

419: 

420: \section{\bf Several formulations of the Minimal Resolution Conjecture}

421: 

422: \subsection*{\bf Notations and conventions}

423: 

424: We work over an algebraically closed field $k$ which, unless

425: explicitly mentioned otherwise, has

426: arbitrary characteristic. Let $V$ be a vector space over $k$

427: with $\dim_kV=n+1$ and $S=Sym(V)\iso k[X_0,\ldots,X_n]$

428: the homogeneous coordinate ring of the corresponding

429: projective space $\PP V\iso\PP^n$. 

430: 

431: For a finitely generated graded $S$-module $N$,

432: the \emph{Betti~ numbers} $b_{i,j}(N)$ of $N$ are defined from the

433: minimal free resolution $F_{\bullet}$ of $N$ by

434: $$F_i=\oplus_{j\in\ZZ}S(-i-j)^{b_{i,j}(N)}.$$

435: The \emph{Betti ~diagram} of $N$ has in the $(j,i)$-th  position

436: the Betti number $b_{i,j}(N)$. The \emph{regularity}

437: ${\rm reg}(N)$ of $N\neq 0$

438: can be defined as the index of the last nontrivial row

439: in the Betti diagram of $N$ (see \cite{eisenbud},

440: 20.5 for

441: the connection with the cohomological definition).

442: 

443: We will use the computation of Betti numbers via

444: Koszul cohomology: $b_{i,j}(N)$ is the dimension over $k$

445: of the cohomology of the following piece of the

446: Koszul complex:

447: $$\wedge^{i+1}V\tensor N_{j-1}\longrightarrow

448: \wedge^iV\tensor N_j\longrightarrow\wedge^{i-1}V\tensor N_{j+1}$$

449: (see \cite{green} for details).

450: 

451: For an arbitrary subscheme $Z\subseteq \PP^n$, we denote by

452: $I_Z\subseteq S$ its saturated ideal and let $S_Z=S/I_Z$.

453: We denote by $P_Z$ and $H_Z$ the Hilbert polynomial and

454: Hilbert function of $Z$, respectively. The regularity 

455: ${\rm reg}(Z)$ of $Z$ is defined to be the regularity of

456: $I_Z$, if $Z\neq \PP^n$, and $1$ otherwise. Notice that

457: with this convention, in the Betti diagram of $Z$,

458: which by definition is the Betti diagram of $S_Z$,

459: the last nontrivial row is always indexed by ${\rm reg}(Z)-1$. 

460: 

461: For a projective variety $X$, a line bundle $L$ on $X$, and

462: a linear series $V\subseteq H^0(L)$ which generates $L$, 

463: we denote by $M_V$ the vector bundle which is the kernel of the

464: evaluation map 

465: $$0\longrightarrow M_V \longrightarrow V\tensor\cO_X

466:  \overset{ev}{\longrightarrow} 

467: L\longrightarrow 0.$$

468: When $V=H^0(L)$ we use the

469: notation $M_L:=M_V$. If $C$ is a smooth curve of genus $g\geq 1$,

470: and $\omega_C$ is the canonical line bundle, then 

471: $M_C$ denotes the vector bundle $M_{\omega_C}$. The dual vector bundles

472: will be denoted by $Q_V$, $Q_L$ and $Q_C$, respectively. 

473: Whenever there is no risk of confusion, we will simply write

474: $M$ and $Q$, instead of $M_C$ and $Q_C$.

475: 

476: 

477: \subsection*{\bf The Minimal Resolution Conjecture for points on

478:  embedded varieties.} 

479: In this section $X\subseteq\PP\,V\iso\PP^n$ is a fixed irreducible projective 

480: variety of positive dimension.

481:  We study the Betti numbers of a general set of $\gamma$ points $\Gamma

482: \subseteq X$.

483:  Since the Betti numbers are upper semicontinuous functions,

484: for every positive integer $\gamma$, there is an open subset $U_{\gamma}$ of

485: $X^{\gamma}\setminus\cup_{p\neq q}\{x\colon x_p=x_q\}$ such that

486: for all $i$ and $j$, $b_{i,j}(\Gamma)$ takes its minimum value

487: for $\Gamma\in U_{\gamma}$. Notice that as the regularity is bounded in terms

488: of $\gamma$, we are concerned with finitely many Betti numbers.

489: From now on, $\Gamma$ \emph{general} means $\Gamma\in U_{\gamma}$.

490: 

491: It is easy to determine the Hilbert function of a general set of points 

492: $\Gamma$ 

493: in terms of the Hilbert function of $X$ 

494: (see \cite{mustata}). We have the following:

495: 

496: \begin{proposition}\label{hilbert_function}

497: If $\Gamma\subseteq X$ is a general set of $\gamma$ points, then

498: $$H_{\Gamma}(t)={\rm min}\,\{H_X(t),\gamma\}.$$

499: \end{proposition}

500: 

501: 

502: 

503: To determine the Betti numbers of a general set of points $\Gamma$ is a 

504: much more subtle problem. If $\gamma$ is large enough, then 

505: the Betti diagram of $\Gamma$ looks as follows: in the upper part we

506: have the Betti diagram of $X$ and there are two extra nontrivial rows at

507: the bottom. Moreover, the formula in Proposition~\ref{hilbert_function}

508: gives an expression for the differences of the Betti numbers in these

509: last two rows. We record the formal statement in the following theorem

510: and for the proof we refer to \cite{mustata}.

511: 

512: \begin{theorem}\label{general_results}

513: Assume that $\Gamma\subseteq X$ is a general set of $\gamma$ points,

514: with $P_X(r-1)\leq\gamma<P_X(r)$ for some $r\geq m+1$, where $m={\rm reg}\,X$.

515: 

516: \item{\rm (i)} For every $i$ and $j\leq r-2$, we have $b_{i,j}(\Gamma)

517: =b_{i,j}(X)$.

518: \item{\rm (ii)} $b_{i,j}(\Gamma)=0$, for $j\geq r+1$ and there is an  

519: $i$ such that $b_{i,r-1}(\Gamma)\neq 0$.

520: \item{\rm (iii)} For every $j\geq m$, we have

521: $$b_{i,j}(\Gamma)=b_{i-1,j+1}(I_{\Gamma}/I_X)=b_{i-1,j+1}(\oplus_{l\geq 0}

522: H^0(\cI_{\Gamma/X}(l))).$$

523: \item{\rm (iv)} If $d=\dim\,X$, then for every $i\geq 0$, we have

524: $b_{i+1,r-1}(\Gamma)-b_{i,r}(\Gamma)=Q_{i,r}(\gamma)$, where

525: $$Q_{i,r}(\gamma)=\sum_{l=0}^{d-1}

526: (-1)^l{{n-l-1}\choose{i-l}}\Delta^{l+1}P_X(r+l)-

527: {n\choose i}(\gamma-P_X(r-1)).$$

528: \end{theorem}

529: 

530: 

531: We will focus our attention on the Betti numbers in the bottom two rows

532: in the Betti diagram of $\Gamma$. The equation in Theorem~\ref

533: {general_results} (iv) gives lower bounds for these numbers, namely

534: $b_{i+1,r-1}(\Gamma)\geq {\rm max}\,\{Q_{i,r}(\gamma),0\}$

535: and $b_{i,r}(\Gamma)\geq {\rm max}\,\{-Q_{i,r}(\gamma),0\}$.

536: 

537: \medskip

538: \begin{definition}

539: In analogy with the case $X=\PP^n$

540: (see \cite{lorenzini1} and \cite{lorenzini2}), we say that the 

541: \emph{Minimal Resolution Conjecture} (to which we refer from now on as MRC)

542: holds for a fixed value of $\gamma$ as above

543: if for every $i$ and every  general set $\Gamma$,

544:  $b_{i+1,r-1}(\Gamma)={\rm max}\,\{Q_{i,r}(\gamma),0\}$

545:  and $b_{i,r}(\Gamma)={\rm max}\,\{-Q_{i,r}(\gamma),0\}$.

546:  Equivalently, it says that 

547: $$b_{i+1,r-1}(\Gamma)\cdot b_{i,r}(\Gamma)

548: =0 {\rm ~for~ all~} i.$$

549: \end{definition}

550: 

551: 

552: This conjecture has been extensively studied in the case $X=\PP^n$,

553:  $L=\OO_{\PP^n}(1)$. 

554: It is known to hold for small values of $n$ ($n=2$, $3$ or $4$) and for

555: large values of $\gamma$, depending on $n$, but not in

556: general. In fact, it has been shown that for every $n\geq 6$, $n\neq 9$,

557: MRC fails for almost $\sqrt{n}/2$ values of $\gamma$ (see

558: \cite{epsw}, where one can find also a detailed account of the problem).

559: 

560: Note that the assertion in MRC holds obviously for $i=0$. The first

561: nontrivial case $i=1$ is equivalent by Theorem~\ref{general_results}

562: to saying that the minimal number of generators of $I_{\Gamma}/I_X$

563: is as small as possible. This suggests the following:

564: 

565: \medskip

566: \begin{definition}

567: We say that the \emph{Ideal Generation Conjecture}

568: (IGC, for short) holds for

569: $\gamma$ as above if for a general set of points $\Gamma\subseteq X$ of

570:  cardinality $\gamma$,

571: we have $b_{2,r-1}(\Gamma)\cdot b_{1,r}(\Gamma)=0$.

572: \end{definition}

573: 

574: \smallskip

575: 

576: 

577: \begin{example}(\cite{mustata})\label{case1}

578:  MRC holds for every $X$ when $\gamma=P_X(r-1)$,

579: since in this case $b_{i,r}(\Gamma)=0$ for every $i$.

580: Similarly, MRC holds for every $X$ when $\gamma= P_X(r)-1$,

581: since in this case $b_{1,r-1}=1$ and $b_{i,r-1}(\gamma)=0$ for $i\geq 2$.

582: \end{example}

583: 

584: 

585: 

586: We derive now a cohomological interpretation of MRC.

587: From now on we assume that $X$ is nondegenerate, so that

588: we have $V\subseteq H^0(\cO_X(1))$.

589: Using a standard Koszul cohomology argument, we can express the

590: Betti numbers in the last two rows of the Betti diagram of $\Gamma$

591: as follows.

592: 

593: 

594: \begin{proposition}\label{cohom}

595: With the above notation, we have in general for every $i\geq 0$ 

596: $$b_{i+1,r-1}(\Gamma)=h^0(\wedge^i M_V\tensor\cI_{\Gamma/X}(r)),$$

597: $$b_{i,r}(\Gamma)=h^1(\wedge^i M_V\tensor\cI_{\Gamma/X}(r)).$$

598: \end{proposition}

599: 

600: \begin{proof}

601: We compute the Betti numbers via Koszul cohomology,

602: using the formula in Theorem~\ref{general_results} (iii).

603: 

604: Consider the complex:

605: $$ \wedge^iV\tensor H^0(\I_{\Gamma/X}(r))\overset{f}{\longrightarrow}

606: \wedge^{i-1}V\tensor H^0(\I_{\Gamma/X}(r+1))

607: \overset{h}{\longrightarrow}\wedge^{i-2}V\tensor H^0(\I_{\Gamma/X}(r+2))$$

608: 

609: 

610: Since $H^0(\cI_{\Gamma/X}(r-1))=0$, it follows

611: that $\dim_k({\rm Ker}f)

612: =b_{i+1,r-1}(\Gamma)$ and $\dim_k({\rm Ker}\,h/{\rm Im}\,f)

613: =b_{i,r}(\Gamma)$.

614:  The exact sequence

615: $$0\longrightarrow M_V\longrightarrow V\tensor\OO_X\longrightarrow\OO_X(1)

616: \longrightarrow 0$$ induces long exact sequences

617: $$0\longrightarrow\wedge^i M_V\longrightarrow\wedge^i V\tensor\OO_X

618: \longrightarrow\wedge^{i-1} M_V\tensor\OO_X(1)\longrightarrow 0.\,\,\,\,

619: (*)$$

620: By tensoring with $\cI_{\Gamma/X}(r)$ and taking global sections,

621: we get the exact sequence

622: $$H^0(\wedge^i M_V\tensor\I_{\Gamma/X}(r))

623: \hookrightarrow\wedge^iV\tensor H^0(\cI_{\Gamma/X}(r))

624: \overset{f}{\longrightarrow}\wedge^{i-1}V\tensor H^0(\cI_{\Gamma/X}(r+1)).$$

625: This proves the first assertion in the proposition.

626: 

627: \smallskip

628: 

629: 

630: We have a similar exact sequence:

631: $$H^0(\wedge^{i-1} M_V\tensor\cI_{\Gamma/X}(r+1))

632: \hookrightarrow\wedge^{i-1} V\tensor H^0(\cI_{\Gamma/X}(r+1))

633: \overset{h}{\longrightarrow}

634: \wedge^{i-2}V\tensor H^0(\cI_{\Gamma/X}(r+2)).$$

635: Therefore $b_{i,r}(\Gamma)$ is the dimension

636: over $k$ of the cokernel of

637: $$g\,:\,\wedge^iV\tensor H^0(\cI_{\Gamma/X}(r))\longrightarrow

638: H^0(\wedge^{i-1}M_V\otimes\cI_{\Gamma/X}(r+1)).$$

639: 

640: Using again the exact sequence $(*)$, by tensoring with $\cI_{\Gamma/X}(r)$

641: and taking a suitable part of the long exact sequence, we get:

642: $$\wedge^iV\tensor H^0(\cI_{\Gamma/X}(r))\longrightarrow

643: H^0(\wedge^{i-1} M_V\tensor\cI_{\Gamma/X}(r+1))\longrightarrow$$

644: $$H^1(\wedge^i M_V\tensor\cI_{\Gamma/X}(r))\longrightarrow\wedge^iV

645: \tensor H^1(\cI_{\Gamma/X}(r)).$$

646: 

647: Since ${\rm reg}\,\Gamma\leq r+1$, we have ${\rm reg}\,\cI_{\Gamma/X}\leq r+1$

648: and therefore

649: $H^1(\cI_{\Gamma/X}(r))=0$. From the above exact sequence we see

650: that ${\rm Coker}\,g\simeq

651: H^1(\wedge^i M_V\tensor\cI_{\Gamma/X}(r))$, which proves

652: the second assertion of the proposition.

653: \end{proof}

654: 

655: \smallskip

656: 

657: 

658: 

659: \begin{remark}\label{higher}

660:  The higher cohomology groups 

661: $H^p(\wedge^i M_V\tensor\I_{\Gamma/X}(r))$, $p\geq 2$, always vanish. 

662: Indeed, using the exact sequences in the proof of the proposition,

663: we get $$h^p(\wedge^i M_V\tensor\I_{\Gamma/X}(r))=

664: h^1(\wedge^{i-p+1} M_V\tensor\I_{\Gamma/X}(r+p-1))=

665: b_{i-p+1,r+p-1}(\Gamma)=0.$$

666: Therefore we have $Q_{i,r}(\gamma)=\chi(\wedge^i M_V\tensor\I_{\Gamma/X}(r))$

667: and MRC can be interpreted as saying that for general $\Gamma$,

668: the cohomology of $\wedge^i M_V\tensor\I_{\Gamma/X}(r)$ is supported

669: in cohomological degree either zero or one.

670: \end{remark}

671: 

672: \smallskip

673: 

674: In the case of a curve $C$, 

675: MRC can be reformulated using 

676: Proposition~\ref{cohom}

677: in terms of general line bundles on $C$.

678: We will denote by $\lfloor x\rfloor$ and $\lceil x\rceil$

679: the integers defined by $\lfloor x\rfloor\leq x <\lfloor x\rfloor +1$

680: and $\lceil x \rceil -1<x\leq\lceil x\rceil$. 

681: 

682: \begin{corollary}\label{criterion}

683: Suppose that $C\subseteq\PP\,V$

684: is a nondegenerate, integral curve of arithmetic genus  $g$ 

685: and degree $d$.

686: We consider the following two statements:

687: 

688: \item{\rm (i)} For every  $i$ and for a general line bundle

689: $\xi\in \Pic^j(C)$, where $j=g-1+\lceil{\frac{di}{n}}\rceil$, we have

690: $H^1(\wedge^i M_V\tensor \xi)=0$. 

691: \item{\rm (ii)} For every $i$ and for a general line bundle

692: $\xi\in\Pic^j(C)$, where $j=g-1+\lfloor{\frac{di}{n}}\rfloor$, we have

693: $H^0(\wedge^i M_V\tensor \xi)=0$.

694: 

695: Then MRC holds for $C$ for every $\gamma\geq \max\{g, P_C({\rm reg}\,X)\}$

696:  if and only if

697: both {\rm (i)} and {\rm (ii)} are true. Moreover, if $C$ is locally Gorenstein,

698: then {\rm (i)} and {\rm (ii)} are equivalent.

699: \end{corollary}

700: 

701: 

702: \begin{proof} 

703: If $\gamma\geq g$, then for a general set $\Gamma$ of $\gamma$ points,

704: $\I_{\Gamma/C}$ is a general line bundle on $C$ of degree $-\gamma$.

705: Since in this case $\I_{\Gamma/C}(r)$ is a general line bundle of

706: degree $j=dr-\gamma$ and $d(r-1)+1-g\leq \gamma\leq dr+1-g$,

707: Proposition~\ref{cohom}

708: says that MRC holds for every $\gamma\geq \max\{g, P_C({\rm reg}\,C)\}$

709:  if and only if

710: for every $j$ such that $g-1\leq j\leq d+g-1$ and for

711: a general line bundle $\xi'\in\Pic^j(C)$,

712: either $H^0(\wedge^i M_V\tensor \xi')=0$ or

713: $H^1(\wedge^i M_V\tensor \xi')=0$. 

714: 

715: Since $\dim\,C=1$, we have $b_{i+1,r-1}(\Gamma)-b_{i,r}(\Gamma)=

716: d{n-1\choose i} -(\gamma-P_C(r-1)){n\choose i}$. It follows immediately

717: that $b_{i+1,r-1}(\Gamma)-b_{i,r}(\Gamma)\geq 0$ if and only if

718: $j\geq g-1+di/n$.

719: 

720: The first statement of the corollary follows now from the fact that

721: if $E$ is a vector bundle on a curve and $P$ is a point, then

722: $H^0(E)=0$ implies $H^0(E\tensor\OO(-P))=0$ and

723: $H^1(E)=0$ implies $H^1(E\tensor\OO(P))=0$. The last statement

724: follows from Serre duality and the isomorphism 

725: $\wedge^iQ_V\simeq\wedge^{n-i}M_V\otimes\cO_C(1)$.

726: \end{proof}

727: 

728: \begin{remark}\label{caseIGC}

729:  The corresponding assertion for IGC says that

730: $X$ satisfies IGC for every $\gamma\geq \max\{g, P_C({\rm reg}\,C)\}$

731:  if and only if both 

732: (i) and (ii) are true for $i=1$. Note that if $X$ is locally Gorenstein,

733: then by Serre duality condition (ii) for $i=1$ is equivalent to

734: condition (i) for $i=n-1$.

735: \end{remark}

736: 

737: \begin{remark}\label{half}

738: If $C$ is a locally  Gorenstein integral curve

739: such that $d/n\in\ZZ$, then in order to check MRC for all

740: $\gamma\geq \max\{g, P_C({\rm reg}\,C)\}$,

741:  it is enough to check condition (i) in Corollary~\ref

742: {criterion} only for $i\leq n/2$. Indeed, using Serre duality

743: and Riemann-Roch, we see that the conditions for $i$

744: and $n-i$ are equivalent.

745: \end{remark}

746: 

747: In light of Corollary~\ref{criterion}, we make the following:

748: 

749: \begin{definition}

750: If $C\subseteq \PP\,V$ is a nondegenerate integral curve of arithmetic genus

751: $g$ and regularity $m$, we say that $C$ satisfies MRC if 

752: a general set of $\gamma$ points on $C$ satisfies MRC

753: for every $\gamma\geq{\rm max}\{g, P_C(m)\}$. If $L$ is a very ample line

754: bundle on a curve $C$ as before, we say that $L$ satisfies MRC if

755: $C\subset\PP\,H^0(L)$ satisfies MRC. Analogous definitions are made for IGC.

756: \end{definition}

757: 

758: \begin{example}({\bf Rational quintics in $\PP^3$.})\label{quintics}

759: We illustrate the above discussion in the case of smooth rational quintic

760: curves in $\PP^3$. We consider two explicit examples, the first when the curve

761: lies on a (smooth) quadric and the second when it does not.

762: Let $X$ be given parametrically by

763: $(u,v)\in\PP^1\longrightarrow (u^5,u^4v,uv^4,v^5)\in\PP^3$, so that

764: it lies on the quadric $X_0X_3=X_1X_2$. The Betti diagram of $X$ is

765: \begin{center}

766: \renewcommand{\arraystretch}{1.25}

767: \begin{tabular}[l]{c|c}

768: \hline

769: 0 & 1 -- -- -- \\

770: 1 & -- 1 -- --  \\

771: 2 & -- -- -- -- \\

772: 3 & -- 4 6 2  \\

773: \end{tabular}

774: \end{center}

775: and if $\Gamma\subset X$ is a set of $28$ points, then the Betti diagram of

776: $\Gamma$ is

777: \begin{center}

778: \renewcommand{\arraystretch}{1.25}

779: \begin{tabular}[l]{c|c}

780: \hline

781: 0 & 1 -- -- -- \\

782: 1 & -- 1 -- -- \\

783: 2 & -- -- -- -- \\

784: 3 & -- 4 6 2 \\

785: 4 & -- -- -- -- \\

786: 5 & -- 3 4 1 \\

787: 6 & -- -- 2 2 \\

788: \end{tabular}

789: \end{center}

790: As $b_{3,5}(\Gamma)=1$ and $b_{2,6}(\Gamma)=2$, we see that MRC

791: is not satisfied by $X$ for this number of points.

792: 

793: Let now $Y$  be the curve given parametrically by

794: $(u,v)\in\PP^1\longrightarrow (u^5+u^3v^2, u^4v-u^2v^3,uv^4,v^5)\in\PP^3$.

795: In this case $Y$ does not lie on a quadric, and in fact, its

796: Betti diagram is given by

797: \begin{center}

798: \renewcommand{\arraystretch}{1.25}

799: \begin{tabular}[l]{c|c}

800: \hline

801: 0 & 1 -- -- -- \\

802: 1 & -- -- -- -- \\

803: 2 & -- 4 3 -- \\

804: 3 & -- 1 2 1 \\

805: \end{tabular}

806: \end{center}

807: If $\Gamma'\subset Y$ is a set of $28$ points, then the Betti diagram of

808: $\Gamma'$ is

809: \begin{center}

810: \renewcommand{\arraystretch}{1.25}

811: \begin{tabular}[l]{c|c}

812: \hline

813: 0 & 1 -- -- -- \\

814: 1 & -- -- -- -- \\

815: 2 & -- 4 3 -- \\

816: 3 & -- 1 2 1 \\

817: 4 & -- -- -- -- \\

818: 5 & -- 3 4 -- \\

819: 6 & -- -- 1 2 \\

820: \end{tabular}

821: \end{center}

822: which shows that MRC is satisfied for $Y$ and this number of points.

823: 

824: These two examples show the possible behavior with respect to the

825: MRC for smooth rational quintics in $\PP^3$. The geometric condition

826: of lying on a quadric translates into a condition on the splitting type

827: of $M_V=\Omega_{\PP^3}(1)\vert_X$. More precisely, it is proved in

828: \cite{ev} that if $X\subset\PP^3$ is a smooth rational quintic curve, then

829: $X$ lies on a quadric if and only if we have

830: $\Omega_{\PP^3}(1)\vert_X\simeq\cO_{\PP^1}(-3)\oplus\cO_{\PP^1}(-1)^{\oplus 2}$

831: (the other possibility, which is satisfied by

832: a general such quintic, is that $\Omega_{\PP^3}(1)\vert_X

833: \simeq\cO_{\PP^1}(-1)\oplus\cO_{\PP^1}(-2)^{\oplus 2}$).

834: Corollary~\ref{criterion} explains therefore the behaviour with respect

835: to MRC in the above examples.

836: \end{example}

837: 

838: 

839: 

840: \section{\bf Curves of large degree and a counterexample to MRC}

841: 

842: In this section we assume that $C$ is a smooth projective curve

843: of genus $g$ and $L$ is a very ample line bundle on $C$. Our aim is to 

844: investigate whether $C$ satisfies MRC, or at least IGC,  for every

845: $\gamma\geq g$, in the embedding given by the complete linear series $|L|$. 

846: As before, $m$ will denote the regularity of $C$.

847: 

848: \begin{example}\label{g_small}

849: If $g=0$ or $1$, then $C$ satisfies MRC for all $\gamma\geq P_C(m)$ 

850: in every embedding given by a complete linear series

851: (see \cite{mustata}, Proposition 3.1).

852: \end{example}

853: 

854: In higher genus we will concentrate on the study

855:  of MRC for canonical curves and 

856: curves embedded with high degree, in direct analogy with the syzygy questions 

857: of Green-Lazarsfeld (cf. \cite{gl1}, \cite{gl2}, \cite{gl3}).

858: The main conclusion of this section will be that, while 

859: IGC is satisfied in both situations, the high-degree embeddings

860: always fail to satisfy MRC at a well-specified spot in the Betti diagram. 

861: This is in contrast with our main result, proved in \S3, that MRC 

862: always holds for canonical curves, and the arguments involved here  

863: provide an introduction to that section. The common theme of the proofs is the

864: vector bundle interpretation of MRC described in \S1.

865: 

866: \medskip

867: \noindent 

868: {\bf Review of filtrations for $Q_L$ and $Q$ \cite{lazarsfeld}.}

869:  Here we recall a basic property of 

870: the vector bundles $Q_L$ which will be essential for our arguments. 

871: Let $L$ be a very ample line bundle on $C$

872: of degree $d$, and recall from \S1 that $Q_L$ is given by the defining sequence

873: $$0\longrightarrow L^{-1}\longrightarrow H^0(L)^{*}\otimes

874:  \OO_{C}\longrightarrow

875: Q_{L}\longrightarrow 0.$$

876: Assume first that $L$ is non-special and $x_{1},\ldots,x_{d}$ are the points

877: of a general hyperplane section of $C\subseteq\PP H^{0}(L)$.

878:  One shows (see e.g. \cite{lazarsfeld} \S1.4) 

879: that there exists an exact sequence 

880: \begin{equation}\label{large_degree_filtration}

881: 0\longrightarrow \underset{i\in \{1,\ldots, d-g-1\}}

882: {\bigoplus}\OO_{C}(x_{i})\longrightarrow Q_{L}

883: \longrightarrow \OO_{C}(x_{d-g}+ \ldots +x_{d})\longrightarrow 0.

884: \end{equation}

885: On the other hand, assuming that $C$ is nonhyperelliptic and $L=\omega_C$, if 

886: $x_{1},\ldots,x_{2g-2}$ are the points of a general hyperplane section,

887: the analogous sequence reads:

888: \begin{equation}\label{canonical_filtration}

889: 0\longrightarrow \underset{i\in \{1,\ldots, g-2\}}

890: {\bigoplus}\OO_{C}(x_{i})\longrightarrow Q

891: \longrightarrow \OO_{C}(x_{g-1}+ \ldots +x_{2g-2})\longrightarrow 0.

892: \end{equation}

893: 

894: 

895: \medskip

896: We start by looking at the case of curves embedded with large degree.

897:  The main results 

898: are summarized in the following:

899: 

900: \begin{theorem}\label{curves_of_large_degree}

901: {\rm (a)} If $L$ is a very ample line bundle of degree $d\geq 2g$, then $IGC$ 

902: holds for $L$.

903: \newline

904: {\rm (b)} If $g\geq 4$ and $L$ is a line bundle of degree $d\geq 2g+10$,

905:  then there exists 

906: a value of $\gamma$ such that $C\subseteq\PP H^{0}(L)$ does not satisfy MRC 

907: for $i=\lfloor\frac{g+1}{2}\rfloor$. The same holds if $g\geq 14$ and

908:  $d\geq 2g+5$.

909: \end{theorem}

910: 

911: \begin{proof}

912: (a)

913: Let $L$ be a very ample line bundle of degree $d\geq 2g$.

914: By Corollary \ref{criterion} and Serre duality, it is easy to see that IGC 

915: holds for $L$ if:

916: \medskip

917: \newline

918: (i) $h^{1}(Q_{L}\otimes \eta)=0$ for $\eta\in {\rm Pic}^{g-2}(C)$ general

919: \newline

920: and

921: \newline

922: (ii) $h^{0}(Q_{L}\otimes \eta)=0$ for $\eta\in {\rm Pic}^{g-3}(C)$ general.

923: 

924: Condition (i) is a simple consequence of the filtration

925:  (\ref{large_degree_filtration}). 

926: More precisely, if $x_{1},\ldots,x_{d}$ are the points

927: of a general hyperplane section of $C\subseteq\PP H^{0}(L)$,

928:  from the exact sequence

929: $$0\longrightarrow \underset{i\in \{1,\ldots, d-g-1\}}

930: {\bigoplus}\OO_{C}(x_{i})\longrightarrow Q_{L}

931: \longrightarrow \OO_{C}(x_{d-g}+ \ldots +x_{d})\longrightarrow 0.$$

932: we conclude that it would be enough to prove:

933: $$h^{1}(\eta(x_{i}))=0 {\rm ~and~} h^{1}(\eta(x_{d-g}+ \ldots +x_{d}))=0$$

934: for $\eta\in {\rm Pic}^{g-2}(C)$ general. Now for every

935:  $i\in \{1,\ldots ,d-g-1\}$, $\eta(x_{i})$

936: is a general line bundle of degree $g-1$, so $h^{1}(\eta(x_{i}))=0$.

937:  On the other hand

938: ${\rm deg}~\eta(x_{d-g}+ \ldots +x_{d})\geq2g-1$, so clearly

939:  $h^{1}(\eta(x_{d-g}+ \ldots +x_{d}))=0$.

940: 

941: For condition (ii) one needs a different argument.

942:  By twisting the defining sequence of $Q_{L}$: 

943: $$0\longrightarrow L^{-1}\longrightarrow

944: H^{0}(L)^{*}\otimes \OO_{C}\longrightarrow

945: Q_{L}\longrightarrow 0$$

946: by $\eta\in {\rm Pic}^{g-3}(C)$ general and taking cohomology,

947:  we see that (ii) holds if and only 

948: if the map

949: $$\alpha^{*}:H^{1}(L^{-1}\otimes \eta)\rightarrow

950: H^{0}(L)^{*}\otimes H^{1}(\eta)$$

951: is injective, or dually if and only if the cup-product map

952: $$\alpha: H^{0}(L)\otimes H^{0}(\omega_{C}

953: \otimes \eta^{-1})\rightarrow H^{0}(L\otimes \omega_{C}

954: \otimes \eta^{-1})$$

955: is surjective. We make the following:

956: \medskip

957: \newline

958: \emph{Claim.} $|\omega_{C}\otimes \eta^{-1}|$

959:  \emph{is a base point free pencil}.

960: 

961: \medskip

962: Assuming this for the time being, one can apply the

963:  Base Point Free Pencil Trick

964: (see \cite{ACGH} III \S3) to conclude that

965: $${\rm Ker~}\alpha= H^{0}(L\otimes \omega_{C}^{-1}\otimes \eta).$$

966: But $L\otimes \omega_{C}^{-1}\otimes \eta$ is a general line bundle of

967:  degree $d-g-1\geq g-1$  

968: and so $h^{1}(L\otimes \omega_{C}^{-1}\otimes \eta)=0$.

969:  By Riemann-Roch this means

970: $h^{0}(L\otimes \omega_{C}^{-1}\otimes \eta)=d-2g$.

971:  On the other hand $h^{0}(L)=d-g+1$,

972: $h^{0}(\omega_{C}\otimes \eta^{-1})=2$ and

973:  $h^{0}(L\otimes \omega_{C}\otimes \eta^{-1})=d+2$, so

974: $\alpha$ must be surjective.

975: 

976: We are only left with proving the claim. Since $\eta\in {\rm Pic}^{g-3}(C)$

977:  is general,

978: $h^{0}(\eta)=0$, and so we easily get:

979: $$ h^{0}(\omega_{C}\otimes \eta^{-1})=h^{1}(\eta)=g-1-(g-3)=2.$$

980: Also, for every $p\in C$, $\eta(p)\in {\rm Pic}^{g-2}(C)$ is general,

981:  hence still noneffective. Thus:

982: $$h^{0}(\omega_{C}\otimes \eta^{-1}(-p))=h^{1}(\eta(p))=g-1-(g-2)=1.$$

983: This implies that $|\omega_{C}\otimes \eta^{-1}|$ is base point free.

984: 

985: \medskip

986: (b) Here we follow an argument in \cite{popa} leading to the required

987: nonvanishing statement.

988: First note that it is clear from (\ref{large_degree_filtration})

989:  that for every $i$ with $1\leq i\leq d-g-1$

990: there is an inclusion

991: $$\OO_{C}(x_{1}+\ldots +x_{i})\hookrightarrow \wedge^{i}Q_{L},$$

992: where $x_{1},\ldots , x_{i}$ are general points on $C$.

993: This immediately implies that 

994: $$h^{0}(\wedge^{i}Q_{L}\otimes \OO_{C}(E_{i}-D_{i}))\neq 0,$$

995: where $E_{i}$ and $D_{i}$ are general effective divisors on $C$ of degree $i$.

996: On the other hand 

997: we use the fact (see e.g. \cite{ACGH} Ex. V. D) 

998: that every line bundle $\xi \in {\rm Pic}^{0}(C)$ can be written as a 

999: difference

1000: $$\xi=\OO_{C}(E_{\lfloor\frac{g+1}{2}\rfloor}-

1001: D_{\lfloor\frac{g+1}{2}\rfloor}),$$

1002: which means that 

1003: $$h^{0}(\wedge^{\lfloor\frac{g+1}{2}\rfloor}

1004: Q_{L}\otimes \xi)\neq 0,~\forall \xi 

1005: \in {\rm Pic}^{0}(C) {\rm ~general~}.$$

1006: Now by Serre duality:

1007: $$H^{0}(\wedge^{i}Q_{L}\otimes \xi)\cong H^{1}(\wedge^{i}M_{L}\otimes 

1008: \omega_{C}\otimes \xi^{-1})^{*},$$

1009: so that Corollary \ref{criterion} easily implies that $C$

1010:  does not satisfy MRC for 

1011: $i=\lfloor\frac{g+1}{2}\rfloor$ as long as $2g-2\geq g-1+ \frac{di}{d-g}$.

1012: A simple computation gives then the stated conclusion.

1013: \end{proof}

1014: 

1015: \begin{remark}

1016: Motivation for the argument in (b) above was quite surprisingly

1017: provided by the study \cite{popa} of 

1018: the base locus of the determinant linear series on the moduli space $SU_C(r)$ 

1019: of semistable bundles of rank $r$ and trivial determinant on a curve $C$. 

1020: In fact this argument produces 

1021: explicit base points for the determinant linear series under appropriate

1022:  numerical conditions.

1023: \end{remark}

1024: 

1025: \begin{remark}

1026: The technique in Theorem \ref{curves_of_large_degree} (b) can be extended

1027:  to produce 

1028: examples of higher dimensional varieties for which appropriate

1029:  choices of $\gamma$ 

1030: force the failure of MRC for general sets of $\gamma$ points.

1031:  More precisely, the 

1032: varieties in question are projective bundles $\PP E \rightarrow C$ over a

1033:  curve $C$, 

1034: associated to very ample 

1035: vector bundles $E$ on $C$ of arbitrary rank and large degree,

1036:  containing sub-line bundles of 

1037: large degree. Using the interpretation given in Proposition \ref{cohom},

1038:  the problem 

1039: is reduced to a cohomological question about the exterior powers

1040:  $\wedge^i M_E$, where 

1041: $M_E$ is defined analogously as the kernel of the evaluation map

1042: $$0\longrightarrow M_E \longrightarrow H^0(E)\otimes

1043:  \OO_X\longrightarrow E\longrightarrow 0.$$

1044: This question is then treated essentially as above,

1045:  and we do not enter into details. 

1046: Unfortunately once a bundle $E$ of higher rank is fixed,

1047:  this technique does not seem 

1048: to produce couterexamples for arbitrarily large values of $\gamma$,

1049:  as in the case of 

1050: line bundles. Such examples would be very interesting,

1051:  in light of the asymptotically nice

1052: behavior of general points in $\PP^n$ (cf. \cite{simpson}).

1053: \end{remark}

1054: 

1055: Finally we turn to the case of canonical curves with the goal of

1056:  providing an introduction 

1057: to the main result in Section 3. Let $C$ be 

1058: a nonhyperelliptic curve of genus $g$, $V=H^0(\omega_C)$

1059: and $C\hookrightarrow\PP V\iso\PP^{g-1}$ the canonical embedding.

1060: We note here that an argument similar to

1061:  Theorem \ref{curves_of_large_degree} (a) 

1062: immediately implies IGC for $C$. This will be later subsumed in the general 

1063: Theorem \ref{canonical_curves}.  

1064: 

1065: \begin{proposition}\label{igc_canonical}

1066: IGC holds for the canonical curve $C$.

1067: \end{proposition}

1068: \begin{proof}

1069: The argument is similar (and in fact simpler) to the proof of (ii) in 

1070: Theorem \ref{curves_of_large_degree} (a). 

1071: In this case, again by interpreting Proposition~\ref{cohom}

1072: (se Remark~\ref{caseIGC}

1073:  IGC holds if and only if

1074: $$H^{0}(Q\otimes \xi)=0~{\rm  for}~\xi\in {\rm Pic}^{g-3}(C) ~{\rm general}.$$

1075: This is in turn equivalent to the surjectivity of the multiplication map:

1076: $$H^{0}(\omega_{C})\otimes H^{0}(\omega_{C}\otimes \xi^{-1})\rightarrow

1077: H^{0}(\omega_{C}^{\otimes 2}\otimes \xi^{-1}),$$

1078: which is again a quick application  of the Base Point Free Pencil Trick.

1079: \end{proof}

1080: 

1081: The geometric picture in the present case of canonical curves

1082:  can be described a

1083: little more precisely. In fact, for $\xi\in {\rm Pic}^{g-3}(C)$, we have

1084: $$\mu(Q\otimes \xi)=g-1,$$

1085: where  $\mu(E):={\rm deg}\,(E)/{\rm rk}\,(E)$

1086: denotes in general the \emph{slope} of the vector bundle $E$.

1087: By standard determinantal results, the subset 

1088: $$\Theta_{Q}:=\{\xi~|~ h^{0}(Q\otimes \xi)\neq 0\}\subseteq{\rm Pic}^{g-3}(C)$$

1089: is either a divisor or the whole variety.

1090: The statement of IGC is then equivalent to saying that  

1091: $\Theta_{Q}$ is indeed a divisor

1092: in ${\rm Pic}^{g-3}(C)$ (one says that $Q$ \emph{has a theta divisor}). 

1093: A simple filtration argument based on the sequence (3) above shows that in fact

1094: $$\Theta_{Q}=C_{g-2}-C:=\{\OO_{C}(p_{1}+\ldots+p_{g-2}-q)~|~p_{1},\ldots ,

1095: p_{g-2}, q \in C\},$$

1096: which has already been observed by Paranjape and Ramanan in

1097: \cite{paranjape}

1098: A generalization of this observation to the higher exterior powers

1099:  $\wedge^iQ$ will be 

1100: the starting point for our approach to proving MRC for canonical curves in

1101:  what follows.

1102: 

1103: 

1104: 

1105: \section{\bf MRC for canonical curves}

1106: 

1107: In this section $C$ will be a canonical curve, i.e. a smooth curve

1108: of genus $g$ embedded in $\PP^{g-1}$ by the canonical linear series

1109: $|\omega_C|$ (in particular $C$ is not hyperelliptic). Our goal is to prove

1110: the following:

1111: 

1112: \begin{theorem}\label{canonical_curves}

1113: If $C$ is a canonical curve, then $C$ satisfies MRC.

1114: \end{theorem}

1115: 

1116: \begin{remark}

1117: In fact, since $C$ is canonically embedded, its regularity is

1118: $m=4$, and as $g\geq 3$ we always have $P_C(m)=7(g-1)\geq g$.

1119: Thus the statement means that MRC holds for every $\gamma \geq P_C(m)$.

1120: \end{remark}

1121: 

1122: The general condition required for a curve to 

1123: satisfy MRC which was stated in Corollary \ref{criterion}

1124: (see also Remark~\ref{half}) 

1125: takes a particularly clean form in the case of canonical embeddings.

1126: We restate it for further use.

1127: 

1128: \begin{lemma}\label{restatement}

1129: Let $C$ be a canonical curve. Then $C$ satisfies MRC if and only if, for all 

1130: $1 \leq i\leq \frac{g-1}{2}$ we have

1131: $$h^0(\wedge^i M\otimes \eta) =

1132: h^1(\wedge^i M\otimes \eta)=0, ~{\rm for}~ \eta\in 

1133: {\rm Pic}^{g+2i-1}(C)~{\rm general},$$ 

1134: or equivalently

1135: $$(*) ~~h^0(\wedge^i Q\otimes \xi) =h^1(\wedge^i Q\otimes \xi)=0, 

1136: ~{\rm for}~ \xi\in \Pici~

1137: {\rm general}.$$ 

1138: \end{lemma} 

1139: 

1140: \begin{remark}

1141: Note that $\mu(Q) = 2$, so $\mu(\wedge^i Q)=2i\in \ZZ$.

1142:  This means that the condition 

1143: $(*)$ in Lemma~\ref{restatement} is equivalent to saying that $\wedge^i Q$

1144:  has a theta divisor (in 

1145: $\Pici$), which we denote $\Theta_{\wedge^i Q}$.

1146:  In other words, the set defined by 

1147: $$\Theta_{\wedge^i Q}:= \{ \xi\in \Pici~|~h^0(\wedge^i Q\otimes \xi)\neq 0\}$$

1148: with the scheme structure

1149:  of a degeneracy locus of a map of vector bundles of the same 

1150: rank is an actual divisor as expected (cf. \cite{ACGH} II \S4).

1151: \end{remark}

1152: 

1153: 

1154: \subsection*{\bf Hyperelliptic curves.}  Note that the statement $(*)$

1155: in Lemma~\ref{restatement} makes sense even for hyperelliptic curves.

1156:  Again $Q$ is the dual of $M$, where $M$ is the 

1157: kernel of the evaluation map for the canonical line

1158: bundle.

1159: Therefore we will say slightly abusively 

1160: that MRC is satisfied for some smooth curve of genus

1161: $g\geq 2$ if $(*)$ is satisfied for all $i$, $1\leq i\leq (g-1)/2$.

1162: In fact, the hyperelliptic case is the only one for which we can give a direct

1163: argument.

1164: 

1165: \begin{proposition}\label{hyperelliptic}

1166: MRC holds for hyperelliptic curves.

1167: \end{proposition}

1168: 

1169: \begin{proof}

1170: We show that for every $i$,

1171: $h^{0}(\wedge^{i}Q\otimes \xi)=0$, if $\xi\in {\rm Pic}^{g-2i-1}(C)$

1172: is general.

1173: Since $C$ is hyperelliptic, we have a degree two morphism

1174: $f:C\rightarrow\PP^1$ and if $L=f^*(\OO_{\PP^1}(1))$,

1175: then $\omega_C=L^{g-1}$. Therefore the morphism

1176: $\widetilde{f}\,:\,C\longrightarrow\PP^{g-1}$ defined by $\omega_C$

1177: is the composition of the Veronese embedding $\PP^1\hookrightarrow

1178: \PP^{g-1}$ with $f$. Note that we have $M=\widetilde{f}^*(\Omega_{\PP^{g-1}}

1179: (1))$.

1180: 

1181: Since on $\PP^1$ we have the exact sequence:

1182: $$0\longrightarrow

1183: \OO_{\PP^1}(-1)^{\oplus(g-1)}\longrightarrow

1184:  H^0(\OO_{\PP^1}(g-1))\tensor\OO_{\PP^1}

1185: \longrightarrow\OO_{\PP^1}(g-1)\longrightarrow 0,$$

1186: we get $M\iso (L^{-1})^{\oplus(g-1)}$.

1187: Therefore for every $i$, we have 

1188: $$\wedge^{i}Q\iso(L^i)^{\oplus{{g-1}\choose i}}.$$

1189: Now if $\xi\in\Pic^{g-2i-1}(C)$ is general,

1190: then $\xi\tensor L^{i}$ is a general line bundle

1191: of degree $g-1$ and so $h^0(\wedge^{i}Q\tensor \xi)=0$.

1192: \end{proof}

1193: 

1194: 

1195: 

1196: \subsection*{\bf Theta divisors and difference varieties for a fixed curve.}

1197: We noted above that MRC is satisfied for $C$ if and only if

1198:  $\Theta_{\wedge^i Q}$ 

1199: is a divisor. We now identify precisely what the divisor should be,

1200:  assuming that 

1201: this happens. (At the end of the day this will hold for all canonical curves.)

1202: Recall that by general theory, whenever a divisor,

1203: $\Theta_{\wedge^i Q}$ belongs 

1204: to the linear series

1205: $|{{g-1} \choose {i}}\Theta|$, where we slightly abusively  

1206: denote by $\Theta$ a certain theta divisor on ${\rm Pic}^{g-2i-1}(C)$ (more

1207: precisely $\Theta_N$, where $N$ is a ${{g-1} \choose {i}}$-th root of 

1208: ${\rm det}(\wedge^i Q)$).

1209: 

1210: From now on we always assume that we are in this situation.

1211: The Picard variety

1212:  ${\rm Pic}^{g-2i-1}(C)$ contains a \emph{difference subvariety}

1213: $C_{g-i-1}-C_{i}$ defined as the image of the difference map

1214: $$\phi:C_{g-i-1}\times C_{i}\longrightarrow {\rm Pic}^{g-2i-1}(C)$$

1215: $$(x_1+\ldots +x_{g-i-1},y_1+\ldots +y_i)\rightarrow \mathcal{O}_C

1216: (x_1+\ldots +x_{g-i-1} -y_1 -\ldots - y_i).$$

1217: The geometry of the difference varieties has interesting links with the 

1218: geometry of the curve \cite{Rob} and \cite{ACGH} (see below). 

1219: The key observation is that our theta divisor 

1220: is nothing else but the difference variety above.

1221: 

1222: \begin{proposition}\label{1}

1223: For every smooth curve $C$ of genus $g$, we have

1224: $$C_{g-i-1} - C_{i}\subseteq\Theta_{\wedge^i Q}.$$

1225: Moreover, if $C$ is nonhyperelliptic and $\Theta_{\wedge^i Q}$

1226: is a divisor, then the above inclusion is an equality.

1227: \end{proposition}

1228: 

1229: We start with a few properties of the difference varieties, which for 

1230: instance easily imply that $C_{g-i-1} - C_{i}$ is a divisor. More 

1231: generally, we study the difference variety $C_a - C_b$, $a\geq b$, defined 

1232: analogously. Note that this study is suggested in a series of exercises 

1233: in [ACGH] Ch.V.D and Ch.VI.A in the case $a=b$, but the formula in V.D-3 

1234: there giving the cohomology class of $C_a - C_a$ is unfortunately incorrect,

1235:  as we first learned from R. Lazarsfeld. 

1236: The results we need are collected in the following:

1237: 

1238: \begin{proposition}\label{2}

1239: (a) Assume that $1\leq b\leq a\leq \frac{g-1}{2}$. Then the difference map:

1240: $$\phi: C_a \times C_b \longrightarrow C_a - C_b\subseteq{\rm

1241: Pic}^{a-b}(C)$$

1242: is birational onto its image if $C$ is

1243: nonhyperelliptic. When $C$ is hyperelliptic, $\phi$ has

1244:  degree ${{a} \choose {b}}\,2^b$ onto its image.

1245: \newline

1246: \noindent

1247: (b) If $C$ is nonhyperelliptic, the cohomology 

1248: class $c_{a,b}$ of $C_a - C_b$ in ${\rm Pic}^{a-b}(C)$ is given by

1249: $$c_{a,b} = {{a+b} \choose {a}}\,\theta^{g-a-b},$$

1250: where $\theta$ is the class of a theta divisor.

1251: \end{proposition}

1252: 

1253: Assuming this, the particular case $a = g-i-1$ and $b = i$ quickly

1254: implies

1255: the main result.

1256: 

1257: \begin{proof}(of Proposition \ref{1})

1258: From Proposition \ref{2} (b) we see that if $C$ is nonhyperelliptic, then

1259: the class of $C_{g-i-1} - C_{i}$ is given by:

1260: $$c_{g-i-1,i} = {{g-1} \choose {i}} \theta.$$

1261: On the other hand, as $\Theta_{\wedge^i Q}$ is associated to the

1262: vector bundle $\wedge^i Q$, if it is a divisor, then its cohomology class is 

1263: ${{g-1} \choose {i}}\theta$ (recall that $\Theta_{\wedge^i Q}$ has the same 

1264: class as ${{g-1}

1265: \choose {i}}\Theta$). As in this case both $\Theta_{\wedge^i Q}$ and

1266:  $C_{g-i-1} - C_{i}$ are

1267: divisors, in order to finish the proof of the proposition it is enough

1268: to prove the first statement.

1269:  

1270: To this end, we follow almost verbatim the argument in

1271:  Theorem \ref{curves_of_large_degree}

1272: (b). Namely, the filtration (2) in \S2 implies that for every $i\geq 1$

1273:  there is an inclusion:

1274: $$\OO_C(x_1+\ldots + x_i)\hookrightarrow \wedge^i Q,$$

1275: where $x_1,\ldots , x_i$ are general points on $C$. This means that 

1276: $$h^0(\wedge^i Q\otimes \OO_C(E_{g-i-1}-D_i))\neq 0$$

1277: for all general effective divisors $E_{g-i-1}$

1278:  of degree $g-i-1$ and $D_i$ of degree $i$,

1279: which gives the desired inclusion.

1280: \end{proof}

1281: 

1282: We are left with proving Proposition \ref{2}. This follows by more or less

1283: standard arguments in the study of Abel maps and Poincar\' e formulas for

1284: cohomology classes of images of symmetric products.

1285: 

1286: \begin{proof}(of Proposition \ref{2})

1287: (a) This is certainly well known (cf. \cite{ACGH} Ch.V.D), and we do not reproduce 

1288: the proof here.

1289: \newline

1290: \noindent

1291: (b) Assume now that $C$ is nonhyperelliptic, so that 

1292: $$\phi: C_a \times C_b \longrightarrow C_a - C_b$$

1293: is birational onto its image. For simplicity we will map 

1294: everything to the Jacobian of $C$, so fix a point $p_0 \in C$ and consider 

1295: the commutative diagram:

1296: $$\xymatrix{

1297: & C^{a+b} \ar[dl]_{\psi} \ar[dr]^{\alpha} \\

1298: C_a\times C_b \ar[r]^{\phi} & C_a - C_b \ar[r]^{-(a-b)p_0} & J(C)}$$

1299: where $C^{a+b}$ is the $(a+b)$-th cartesian product of the curve and the

1300: maps 

1301: are either previously defined or obvious.

1302: We will in general denote by $[X]$ the fundamental class of the compact

1303: variety $X$. Since $\psi$ clearly has degree $a!\cdot b!$, and since

1304: $\phi$ is birational by (a), we have:

1305: $$\alpha_{*}[C^{a+b}] = a!\cdot b!\cdot c_{a,b}.$$

1306: This means that it is in fact enough to prove that $\alpha_{*}[C^{a+b}]

1307: = (a+b)!\cdot \theta^{g-a-b}$, and note that 

1308: $(a+b)!\cdot\theta^{g-a-b}$ is the same as the class

1309: $u_{*}[C^{a+b}]$, where $u$ is the usual Abel map:

1310: $$u: C^{a+b}\longrightarrow J(C)$$

1311: $$(x_1,\ldots , x_{a+b})\rightarrow \OO_C(x_1 + \ldots + x_{a+b} - (a+b)p_0).$$

1312: The last statement is known as Poincar\'e's formula (see e.g. [ACGH] I

1313: \S5).

1314: We are now done by the following lemma,  which essentially says that 

1315: adding or subtracting points is the same when computing cohomology

1316: classes. 

1317: \end{proof}

1318: 

1319: \begin{lemma}

1320: Let $u,\alpha : C^{a+b} \longrightarrow J(C)$ defined by:

1321: $$u(x_1 ,\ldots ,x_{a+b}) = \OO_C(x_1 + \ldots + x_{a+b} - (a+b)p_0)$$

1322: and 

1323: $$\alpha(x_1 ,\ldots , x_{a+b}) = \OO_C(x_1 + \ldots + x_a - x_{a+1} - \ldots -

1324: x_{a+b} - (a-b)p_0).$$

1325: Then $u_{*}[C^{a+b}] = \alpha_{*}[C^{a+b}]\in H^{2(g-a-b)}(J(C),

1326: \mathbb{Z})$.

1327: \end{lemma}

1328: \begin{proof}

1329: For simplicity, in this proof only, we will use additive divisor notation,

1330: although we actually mean the associated line bundles.

1331: Consider the auxiliary maps:

1332: $$u_0: C^{a+b}\longrightarrow J(C)^{a+b}$$

1333: $$(x_1, \ldots , x_{a+b})\rightarrow (x_1 - p_0 ,\ldots , x_{a+b} -

1334: p_0),$$

1335: $$\alpha_0: C^{a+b}\longrightarrow J(C)^{a+b}$$

1336: $$(x_1,\ldots , x_{a+b})\rightarrow (x_1  - p_0, \ldots , x_a - p_0, 

1337: p_0 - x_{a+1}, \ldots , p_0 - x_{a+b})$$

1338: and the addition map:

1339: $$a: J(C)^{a+b}\longrightarrow J(C)$$

1340: $$(\xi_1, \ldots , \xi_{a+b})\rightarrow \xi_1 + \ldots + \xi_{a+b}.$$

1341: Then one has:

1342: $$u = au_0 ~{\rm and}~\alpha = a\alpha_0 = a\mu u_0,$$

1343: where $\mu$ is the isomorphism

1344: $$\mu: J(C)^{a+b}\longrightarrow J(C)^{a+b}$$

1345: $$(\xi_1, \ldots ,\xi_{a+b})\rightarrow (\xi_1,\ldots,\xi_a , -\xi_{a+1},

1346: \ldots , -\xi_{a+b}).$$

1347: Now the statement follows from the more general fact that $\mu$ induces 

1348: an isomorphism on cohomology. This is in turn a simple consequence

1349: of the fact that the involution $x \rightarrow -x$ induces the identity on

1350: $H^{1}(J(C), \mathbb{Z})$.

1351: \end{proof}

1352: 

1353: \begin{remark}\label{hyperelliptic_case}

1354: The equality in Proposition~\ref{1} holds set-theoretically also for

1355: hyperelliptic curves. Indeed, we have  

1356: the inclusion $C_{g-i-1}-C_i\subseteq\Theta_{\wedge^iQ}$,

1357: and we have seen in the proof of

1358: Proposition~\ref{hyperelliptic} that $\Theta_{\wedge^iQ}$ is irreducible

1359: (with the reduced structure, it is just a translate of the usual theta

1360:  divisor).

1361: \end{remark}

1362: 

1363: \subsection*{\bf General canonical curves.}

1364: Since we have seen that MRC holds for hyperelliptic curves, a standard

1365: argument shows that it holds for general canonical curves. In fact, our

1366: previous result about the expected

1367: form of the theta divisors $\Theta_{\wedge^iQ}$ allows us

1368: to say something more precise about the set of curves in ${\mathcal M}_g$

1369: which might not satisfy MRC.

1370: 

1371: \begin{proposition}\label{generic}

1372: For every $i$, the set of curves

1373: $\{[C]\in {\mathcal M}_g\mid\Theta_{\wedge^iQ_C}=\Pic^{g-2i-1}(C)\}$

1374: is either empty or has pure codimension one. In particular,

1375: the same assertion is true for the set of curves

1376:  in ${\mathcal M}_g$

1377: which do not satisfy MRC.

1378: \end{proposition}

1379: 

1380: \begin{proof}

1381: As the arguments involved are standard we will just sketch the proof.

1382: 

1383: Start by considering, for a given $d\geq 2g+1$,   

1384: the Hilbert scheme $\mathcal H$ of curves in $\PP^{d-g}$ with  

1385: Hilbert polynomial $P(T)=d\,T+1-g$ and ${\mathcal U}\subseteq\mathcal{H}$

1386: the open subset corresponding to smooth connected nondegenerate curves.

1387:  

1388: Let $f\,:\,{\mathcal Z}\longrightarrow {\mathcal U}$

1389:  be the universal family over $\mathcal U$,

1390: which is smooth of relative dimension $1$,

1391:  and $\omega_{{\mathcal Z}/{\mathcal U}}

1392: \in\Pic({\mathcal Z})$ the relative cotangent bundle.

1393: By base change there is an exact sequence

1394: $$0\longrightarrow {\mathcal Q}^{\vee}

1395: \longrightarrow f^*f_*\omega_{{\mathcal Z}/

1396: {\mathcal U}}

1397: \longrightarrow 

1398: \omega_{{\mathcal Z}/{\mathcal U}}\longrightarrow 0,$$

1399: where ${\mathcal Q}$ is a vector bundle on $\cZ$ such that 

1400:  if $u\in {\mathcal U}$ corresponds to a curve

1401:  $C=\cZ_u$ (in a suitable embedding), then ${\mathcal Q}\vert_{{\mathcal Z}_u}

1402: \iso Q_C$.

1403: 

1404: The usual deformation theory arguments show that ${\mathcal H}$ is smooth

1405: and has dimension $(d-g+1)^2+4(g-1)$. Moreover, the universal family

1406: ${\mathcal Z}$ defines a surjective

1407: morphism $\pi\,:\,{\mathcal U}\longrightarrow

1408: \cM_g$ whose fibers are irreducible and have dimension $(d-g+1)^2+g-1$.

1409: It is immediate to see from this that $\cU$ is connected.

1410: 

1411: Fix now $l$ such that $d=(g-2i-1)+l(2g-2)\geq 2g+1$. Consider

1412: $\cU$ and $\cZ$ as above and let $\cF:=\wedge^i\cQ\otimes\omega_{\cZ/\cU}^{-l}

1413: \otimes p^*\cO_{\PP^{d-g}}(1)$, where $p$ is the composition of the

1414: inclusion $\cZ\hookrightarrow\cU\times\PP^{d-g}$ and the projection onto the

1415: second factor.

1416: 

1417: We consider also the closed subset of $\cU$:

1418: $${\mathcal D}_1=\{u\in\cU\mid h^0(\cF\vert_{\cZ_u})\geq 1\}.$$

1419: It is clear by definition that $\pi^{-1}([C])\subseteq{\mathcal D}_1$

1420: if and only if $\Theta_{\wedge^iQ_C}=\Pic^{g-2i-1}(C)$.

1421:  In particular,

1422:  Proposition~\ref{hyperelliptic} implies that ${\mathcal D}_1\neq \cU$.

1423: 

1424: ${\mathcal D}_1$ is the degeneracy locus of a morphism between two vector

1425:  bundles of the same rank. Indeed, if $H\subset\PP^{d-g}$ is a hyperplane,

1426: $\widetilde{H}=p^{-1}H$,

1427: and $r\gg 0$, then ${\mathcal D}_1$ is the degeneracy locus of 

1428: $$f_*(\cF\otimes\cO_{\mathcal Z}(r\widetilde{H}))

1429: \longrightarrow f_*(\cF\otimes \cO_{r\widetilde{H}}(r\widetilde{H})).$$

1430: Note that these are both vector bundles of rank $rd{{g-1}\choose i}$

1431: (we use base change and the fact that by Corollary~3.5 in 

1432: \cite{paranjape}, for every

1433: smooth curve $C$, the bundle $Q_C$ is semistable). We therefore conclude that

1434: ${\mathcal D}_1$ is a divisor on $\cU$.

1435: 

1436: 

1437: On the other hand, it is easy to see that the set

1438: $${\mathcal D}_2=\{u\in\cU\mid \cO_{\cZ_u}(1)\otimes\omega_{\cZ_u}^{-l}

1439: \in (\cZ_u)_{g-i-1}-(\cZ_u)_i\}$$ is closed.

1440: Moreover, Proposition~\ref{1}

1441: (see also Remark~\ref{hyperelliptic_case}) shows that

1442: ${\mathcal D}_2\subseteq {\mathcal D}_1$, 

1443: and if $\pi^{-1}([C])\not\subseteq{\mathcal D}_1$,

1444: then $\pi^{-1}([C])\cap {\mathcal D}_1=\pi^{-1}([C])\cap {\mathcal D}_2$.

1445: 

1446: Let $\mathcal S$ be the set of irreducible components of ${\mathcal D}_1$

1447: which are not included in ${\mathcal D}_2$. Using the fact that $\pi$

1448: has irreducible fibers, all of the same dimension, it is easy to see that

1449: if $Y\in{\mathcal S}$, then $\pi(Y)$ is closed in $\cM_g$,

1450: that it is in fact a divisor, and $Y=\pi^{-1}(\pi(Y))$. Moreover, the locus of

1451: curves in $\cM_g$ for which $\Theta_{\wedge^iQ}$ is not a divisor is

1452: $\cup_{Y\in{\mathcal S}}

1453: \pi(Y)$, which proves the proposition.

1454: \end{proof}

1455: 

1456: 

1457: 

1458: 

1459: \subsection*{\bf The class of the degeneracy locus on $\cM_{g,g+1}$}

1460: 

1461: We first fix the notation. We will denote by ${\mathcal M}_g^{0}$ the open

1462: subset of ${\mathcal M}_g$ which corresponds to curves with no

1463: nontrivial automorphisms.

1464: From now on we assume that $g\geq 4$, since for $g=3$ MRC is equivalent

1465: to IGC, which is the content of Proposition \ref{igc_canonical}.

1466: Thus $\cM_g^{0}$ is 

1467: nonempty and its complement has codimension 

1468: $g-2\geq 2$ (see \cite{HaMo} pg.37), so working with this subset 

1469: will not affect the answers we get for divisor class computations

1470: on $\cM_g$ or $\cM_{g,n}$.

1471: 

1472: In this case we have a universal family over $\cM_g^{0}$ denoted by

1473: ${\mathcal C}_g^{0}$

1474:  and for every $n\geq 1$, the open subset of

1475: $\cM_{g,n}$ lying over $\cM_g^{0}$

1476: (which we denote by $\cM_{g,n}^{0}$) is equal to the fiber product

1477: $(\times_{\cM_g^{0}}{\mathcal C}_g^{0})^n$

1478:  minus all the diagonals. We assume that $n\geq g+1$.

1479: 

1480: Consider the following cartezian diagram:

1481: \[

1482: \begin{CD}

1483: {\mathcal X}@>{q}>> \cM_{g,n}^{0}\\

1484: @VV{f}V@VV{h}V\\

1485: {\mathcal C}_g^{0}@>{p}>>\cM_g^{0}\\

1486: \end{CD}

1487: \]

1488: in which all the morphisms 

1489: and varieties are smooth and $p$ (hence also $q$) is proper.

1490: 

1491: Let $\omega\in\Pic({\mathcal C}_g^{0})$ be the relative canonical line

1492:  bundle for $p$, $E=p_*(\omega)$ the Hodge vector bundle and

1493:  $\cQ$ the rank $g-1$

1494: vector bundle on ${\mathcal C}^{0}_g$ such that $\cQ^{\vee}$ is the

1495: kernel of the evaluation map $p^*E\longrightarrow\omega$. For every

1496: $[C]\in\cM_g^{0}$, we have $\cQ\vert_{p^{-1}([C])}\simeq Q_C$.

1497: 

1498: The projection on the $j$-th factor $p_j\,:\,\cM_{g,n}^{0}

1499: \longrightarrow{\mathcal C}_g^{0}$ induces a section 

1500: $q_j\,:\,\cM_{g,n}^{0}\longrightarrow {\mathcal X}$ of $q$.

1501: If $E_j={\rm Im}(q_j)$, then $E_j$ is a relative divisor over $\cM_{g,n}

1502: ^{0}$. Consider the following vector bundle on ${\mathcal X}$:

1503: $$\cE=\wedge^if^*\cQ\left(\sum_{j=1}^{g-i}E_j-\sum_{j=g-i+1}^{g+1}E_j\right)$$

1504: and let

1505:  $\cZ=\{u\in\cM_{g,n}^{0}\vert ~h^0(\cE\vert_{{\mathcal X}_u})\geq 1\}$.

1506: 

1507: The algebraic set

1508: $\cZ$ comes equipped with a natural stucture of degeneracy locus. 

1509: Suppose for example that $Y$ is a sum of $m$ divisors $E_j$

1510: (possibly with repetitions), where $m\gg 0$. In this case, $\cZ$

1511: is the degeneracy locus of the morphism

1512:  $$\cF:=q_*(\cE\otimes\cO_{\mathcal X}(Y))

1513: \longrightarrow \cF':=q_*(\cE\otimes\cO_{\mathcal X}(Y)\vert_Y).$$

1514:  This scheme 

1515: structure does not depend on the divisor $Y$ we have chosen.

1516: In fact, it is the universal subscheme over which the $0$-th Fitting ideal

1517: of the first higher direct image of $\cE$ becomes trivial (see e.g.

1518: \cite{ACGH} Ch.IV \S3

1519:  for the proof of an analogous property). Note that $\cZ$ is a divisor

1520: and not the whole space, since by Proposition~\ref{generic} we know that

1521: for a general curve $C$, there is $\xi\in\Pic^{g-2i-1}(C)$

1522: such that $H^0(\wedge^iQ_C\otimes\xi)=0$ (note also 

1523:  that the difference map

1524: $C^{g-i}\times C^{i+1}\longrightarrow \Pic^{g-2i-1}(C)$ is surjective,

1525: cf. \cite{ACGH} Ch.V.D).

1526: 

1527: 

1528: We will use the notation $\lambda=c_1(h^*(E))$, $\psi_j=c_1(p_j^{*}(\omega))$,

1529: and $\Psi_x=\sum_{j=1}^{g-i}\psi_j$ and $\Psi_y=\sum_{j=g-i+1}^{g+1}\psi_j$.

1530: It is well known that $\lambda$ together with $\psi_j$, $1\leq j\leq n$,

1531: form a basis for $\Pic(\cM_{g,n}^{0})_{\QQ}$.

1532: 

1533: \begin{proposition}\label{degeneracy_class}

1534: With the above notation, for every $n\geq g+1$, the class of $\cZ$

1535: in $\Pic(\cM_{g,n}^{0})_{\QQ}$ is given by

1536: \begin{equation}

1537: [\cZ]=-\left({{g-1}\choose i}-10{{g-3}\choose{i-1}}\right)

1538: \lambda+{{g-2}\choose i}\Psi_x

1539: +{{g-2}\choose{i-1}}\Psi_y.

1540: \end{equation}

1541: \end{proposition}

1542: 

1543: 

1544: \begin{proof}

1545: Note that the pull-back of divisors induced by the projection to

1546:  the first $(g+1)$ components induces injective homomorphisms

1547: $\Pic(\cM_{g,g+1}^{0})_{\QQ}\hookrightarrow

1548: \Pic(\cM_{g,n}^{0})_{\QQ}$. From the universality of the scheme

1549: structure on $\cZ$ it follows that

1550:  the computation of $c_1(\cZ)$ is

1551: independent of $n$. Therefore we may assume

1552: that $n$ is large enough, so that in defining the

1553: scheme structure of $\cZ$ as above, we may take $Y=\sum_{j=g+2}^nE_j$.

1554: We introduce also the notation $\Psi_z=\sum_{j=g+2}^n\psi_j$.

1555: 

1556: As a degeneracy locus, the class of $\cZ$ is given by $c_1(\cF')-c_1(\cF)$.

1557: It is clear that we have $E_j\cap E_l=\emptyset$ if $j\neq l$

1558: and via $E_j\simeq\cM_{g,n}^{0}$, we have $\cO_{E_j}(-E_j)\simeq

1559: p_j^*(\omega)$. Since $\cQ^{\vee}$ is the kernel of the evaluation map for

1560: $\omega$, we get 

1561: \begin{equation}\label{first_formula}

1562: c_1(f^*Q)=f^*(c_1(\omega))-q^*(\lambda).

1563: \end{equation}

1564: 

1565: 

1566: Before starting the computation of $c_1(\cF)$ and $c_1(\cF')$, we record the

1567: following well-known formulas for Chern classes.

1568: 

1569:  \begin{lemma}\label{chern_classes}

1570: Let $R$ be a vector bundle of rank $n$ on a variety $X$

1571: and $L\in {\rm Pic}(X)$. 

1572: \item{\rm (i)} $c_1(R\otimes L)=c_1(R)+nc_1(L)$.

1573: \item{\rm (ii)} $c_2(R\otimes L)=

1574: c_2(R)+(n-1)c_1(R)c_1(L)+{n\choose 2}c_1(L)^2$.

1575: \item{\rm (iii)} $c_1(\wedge^iR)=

1576: {{n-1}\choose{i-1}}c_1(R)$, if $n\geq 2$ and $1\leq i

1577: \leq n$.

1578: \item{\rm (iv)} $c_2(\wedge^iR)=\frac12

1579: {{n-1}\choose{i-1}}\left({{n-1}\choose {i-1}}-1\right)

1580: c_1(R)^2+{{n-2}\choose{i-1}}c_2(R)$, if $n\geq 3$ and $1\leq i\leq n$.

1581: \end{lemma}

1582: 

1583: {}From the previous discussion and Lemma~\ref{chern_classes} (i)

1584:  and (iii), we get

1585: $$c_1(q_*(\cE\otimes\cO_{E_j}(Y)))=-{{g-2}\choose{i-1}}\lambda-

1586: {{g-2}\choose i}\psi_j,$$

1587: {}from which we deduce

1588: $$c_1(\cF')=-(n-g-1){{g-2}\choose{i-1}}\lambda-{{g-2}\choose i}\Psi_z.$$

1589: 

1590: In order to compute $c_1(\cF)$, we apply the Grothendieck-Riemann-Roch

1591: formula for $q$ and $\cE(Y)$

1592: (see \cite{fulton} 15.2).

1593:  Note that the varieties are smooth and $q$ is smooth and proper.

1594: Moreover, since we assume $n\gg 0$, we have $R^jq_*(\cE(Y))=0$, for $j\geq 1$.

1595: Therefore we get

1596: $$ch(q_*(\cE(Y)))=q_*(ch(\cE(Y))\cdot td(f^*\omega^{-1})).$$

1597: From this we deduce

1598: $$(*)\,

1599: c_1(\cF)=q_*\left(\frac12 c_1(\cE(Y))^2-c_2(\cE(Y))-

1600: \frac12 f^*c_1(\omega)\cdot c_1(\cE(Y))

1601: +\frac1{12}{{g-1}\choose i}f^*c_1(\omega)^2\right). 

1602: $$

1603: 

1604: We compute now each of the classes involved in the above equation.

1605: In order to do this we need to know how to make the push-forward of the

1606: elementary classes on $\mathcal X$. We list these rules in the following: 

1607: 

1608: \begin{lemma}\label{rules}

1609: With the above notation, we have

1610: \item {\rm (i)} $q_*(f^*c_1(\omega)^2)=12\lambda$.

1611: \item {\rm (ii)} $q_*(q^*\lambda\cdot f^*c_1(\omega))=(2g-2)\lambda$.

1612: \item {\rm (iii)} $q_*q^*(\lambda^2)=0$.

1613: \item {\rm (iv)} $q_*(c_1(E_j)\cdot q^*\lambda)=\lambda$.

1614: \item {\rm (v)} $q_*(c_1(E_j)\cdot f^*c_1(\lambda))=\psi_j$.

1615: \item {\rm (vi)} $q_*q^*c_2(h^*E)=0$.

1616: \item {\rm (vii)} $q_*(c_1(E_j)^2)=-\psi_j$.

1617: \end{lemma}

1618: 

1619: \begin{proof}[Proof of Lemma~\ref{rules}]

1620: The proof of (i) is analogous to that of the relation $p_*(c_1(\omega)^2)=

1621: 12\lambda$ (see \cite{HaMo} \S 3E). The other formulas are straightforward.

1622: \end{proof}

1623: 

1624: Using Lemma~\ref{chern_classes} (i) and (iii) and the formula 

1625: (\ref{first_formula}) for

1626: $c_1(f^*Q)$, we deduce that 

1627: $$c_1(\cE(Y))={{g-2}\choose{i-1}}(f^*c_1(\omega)-q^*\lambda)

1628: +{{g-1}\choose i}\left(\sum_{j=1}^{g-i}c_1(E_j)-\sum_{j=g-i+1}^{g+1}c_1(E_j)

1629: +\sum_{j=g+2}^nc_1(E_j)\right).$$

1630: Applying Lemma~\ref{rules}, we get

1631: $$q_*(c_1(\cE(Y))^2/2)=\left((8-2g){{g-2}\choose{i-1}}^2

1632: -(n-2i-2){{g-2}\choose{i-1}}

1633: {{g-1}\choose i}\right)\lambda+$$

1634: $${{g-2}\choose{i-1}}{{g-1}\choose i}(\Psi_x-\Psi_y+\Psi_z)

1635: -\frac12{{g-1}\choose i}^2(\Psi_x+\Psi_y+\Psi_z).$$

1636: 

1637: 

1638: {}From the above formula for $c_1(\cE(Y))$ and Lemma~\ref{rules}, we get

1639: $$q_*\left(-\frac12 f^*c_1(\omega)\cdot c_1(\cE(Y))\right)

1640: =(g-7){{g-2}\choose{i-1}}\lambda

1641: -\frac12{{g-1}\choose i}(\Psi_x-\Psi_y+\Psi_z).$$

1642: 

1643: 

1644: Lemma~\ref{rules} (i) gives

1645: $$q_*\left(\frac1{12}{{g-1}\choose i}f^*c_1(\omega)^2\right)

1646: ={{g-1}\choose i}\lambda.$$

1647: 

1648: {}From the defining exact sequence of $\cQ^{\vee}$ we compute

1649: $$c_2(f^*\cQ)=q^*c_2(h^*E)+f^*c_1(\omega)\cdot(f^*c_1(\omega)-q^*\lambda).$$

1650: Using now Lemma~\ref{chern_classes} (ii) and (iv) and Lemma~\ref{rules},

1651:  we deduce

1652: $$q_*c_2(\cE(Y))= \left((8-2g){{g-2}\choose{i-1}}\left(

1653: {{g-2}\choose{i-1}}-1\right)+(14-2g){{g-3}\choose{i-1}}\right)\lambda-$$

1654: $$(n-2i-2){{g-2}\choose{i-1}}\left({{g-1}\choose i}-1\right)\lambda+

1655: {{g-2}\choose{i-1}}\left({{g-1}\choose i}-1\right)(\Psi_x-\Psi_y+\Psi_z)-$$

1656: $$\frac12{{g-1}\choose i}\left({{g-1}\choose i}-1\right)(\Psi_x

1657: +\Psi_y+\Psi_z).$$

1658: 

1659: 

1660: 

1661: Using these formulas and equation $(*)$, we finally obtain

1662: $$c_1(\cF)=\left((2g-14){{g-3}\choose{i-1}}-(n+g-2i-3){{g-2}\choose{i-1}}

1663: +{{g-1}\choose i}\right)\lambda-$$

1664: $${{g-2}\choose i}(\Psi_x+\Psi_z)

1665: -{{g-2}\choose{i-1}}\Psi_y.$$

1666: 

1667: Since the class of $\cZ$ is equal with $c_1(\cF')-c_1(\cF)$, we deduce

1668: the statement of the proposition.

1669: \end{proof}

1670: 

1671: \subsection*{\bf Proof of the main result.}

1672: We introduce next a divisor

1673:  $D$ on $\mathcal{M}_{g,g+1}$ which is a global analogue of the

1674:  preimage of $C_{g-i-1}-C_{i}$ under the difference map

1675:  $C^{g-i}\times C^{i+1}\rightarrow \mbox{Pic}^{g-2i-1}(C)$.

1676:  This is motivated by Proposition \ref{1}, and our goal is roughly 

1677:  speaking to prove a global version of that result. 

1678: 

1679: \begin{definition} For $g\geq 3$ and

1680:  $1\leq i\leq \frac{g-1}{2}$ we define the divisor $D$

1681: on $\mathcal{M}_{g,g+1}$ to be the locus of smooth pointed curves

1682:  $(C, x_1,\ldots ,x_{g-i}, y_1,\ldots y_{i+1})$

1683:  having a linear series $\mathfrak g^1_{g}$ containing 

1684: $x_1+\cdots +x_{g-i}$ in a fiber and $y_1+\cdots +y_{i+1}$ in another fiber.

1685: Note that this means that we can in fact write the line bundle

1686: $\OO_C(x_1+\ldots +x_{g-i}-y_1-\ldots-y_{i+1})$ 

1687: as an element in $C_{g-i-1}-C_i$.

1688: \end{definition}

1689: 

1690: We consider in what follows the divisor $\overline{\mathcal{Z}}$,

1691: the closure in ${\mathcal M}_{g,g+1}$ of the divisor 

1692: ${\mathcal Z}$ studied above (we take now $n=g+1$,  but

1693: as we mentioned, this does not affect the formula for its class).

1694: In Section 4 we prove that $D$ is reduced and that

1695:  $D\equiv_{\QQ-{lin}} \overline{\mathcal{Z}}$

1696: (cf. Theorem \ref{class_D}).

1697: This being granted we are in a position to complete the proof of Theorem 3.1:

1698: 

1699: \begin{proof}[Proof of Theorem~\ref{canonical_curves}] 

1700: Note that for $g=3$ our assertion is just the statement of Proposition 

1701: \ref{igc_canonical}. Thus we can assume $g\geq 4$.

1702: As mentioned above $D$ is reduced, and from Proposition \ref{1}

1703: we see that

1704:  $\mbox{supp}(D)\subseteq \mbox{supp}(\overline{\mathcal{Z}})$.

1705:  We get that $\overline{\mathcal{Z}}-D$ is effective,

1706:  and in fact $\overline{\mathcal{Z}}-D=h^*(E)$, where $E$ is an effective

1707:  divisor on $\mathcal{M}_g$ and $h:\mathcal{M}_{g,g+1}\rightarrow

1708:  \mathcal{M}_g$ is the projection. 

1709: Moreover, the map $h^*:\mbox{Pic}(\mathcal{M}_g)_{\QQ}\rightarrow

1710:  \mbox{Pic}(\mathcal{M}_{g,g+1})_{\QQ}$ is injective (cf. \cite{AC2}),

1711:  hence $E\equiv_{\QQ-lin}0$. Since 

1712: the Satake compactification of $\mathcal{M}_g$ has boundary

1713:  of codimension $2$ (see e.g \cite{HaMo}, pg.45) this implies $E=0$,

1714:  that is, $\overline{\mathcal{Z}}=D$.

1715:  Therefore $\Theta_{\wedge ^i Q_C}$ is a divisor in

1716:  $\mbox{Pic}^{g-2i-1}(C)$ and the identification

1717:  $\Theta_{\wedge ^i Q_C}=C_{g-i-1}-C_{i}$ holds for

1718:  \emph{every} nonhyperelliptic curve $C$.

1719: \end{proof}

1720: 

1721: 

1722: 

1723: \section{\bf A divisor class computation on $\Mg1$}

1724: 

1725: 

1726: In this section we compute the class of the divisor

1727:  $D$ on $\mathcal{M}_{g,g+1}$ defined in the previous section.

1728: We start by recalling a few facts about line bundles on

1729:  $\overline{\mathcal{M}}_{g,n}$. Let us fix $g\geq 3$, $n\geq 0$

1730:  and a set $N$ of $n$ elements.

1731:  Following \cite{AC2}, we identify $\overline{\mathcal{M}}_{g,n}$

1732:  with the moduli space

1733:  $\overline{\mathcal{M}}_{g,N}$

1734:  of stable curves of genus $g$ with marked points indexed by $N$.

1735:  We denote by $\pi_{q}:\overline{\mathcal{M}}_{g,N\cup\{q\}}\rightarrow

1736:  \overline{\mathcal{M}}_{g,N}$ the map forgetting the marked

1737:  point indexed by $q$.

1738:  For each $z\in N$ we define the tautological class $\psi_z=c_1(\mathbb L_z)\in \mbox{Pic}(\overline{\mathcal{M}}_{g,N})_{\mathbb Q}$,

1739: where $\mathbb L_z$ is the line bundle over $\overline{\mathcal{M}}_{g,N}$ whose fiber over the moduli point $[C,\{x_i\}_{i\in N}]$ is the cotangent space $T_{x_z}^*(C)$. Note that although we are

1740:  using an apparently different definition, these $\psi$ classes are the same 

1741:  as those which appear in the previous section.

1742: 

1743:  For $0\leq i\leq g$ and $S\subseteq N$, the boundary divisor

1744:  $\Delta_{i:S}$  corresponds to the closure in $\overline{\mathcal{M}}_{g,N}$

1745: of the locus of nodal curves

1746:  $C_1\cup C_2$, with $C_1$ smooth of genus $i$, $C_2$ smooth

1747:  of genus $g-i$, and such that the marked points sitting on

1748:  $C_1$ are precisely those labelled by $S$.

1749:  Of course $\Delta_{i:S}=\Delta_{g-i:S^{c}}$ and we set

1750:  $\Delta_{0:S}:=0$ when $|S|\leq 1$. We also consider the divisor $\Delta_{irr}$ consisting of irreducible pointed curves

1751: with one node. We denote by $\delta_{i:S}\in \mbox{Pic}(\overline{\mathcal{M}}_{g,n})_{\QQ}$ the class of $\Delta_{i:S}$ and by $\delta_{irr}$ that of $\Delta_{irr}$. 

1752:  It is well known that the Hodge class $\lambda, \delta_{irr}$, the $\psi_z$'s and the

1753:  $\delta_{i:S}$'s freely generate

1754:  $\mbox{Pic}(\overline{\mathcal{M}}_{g,n})_{\QQ}$ (cf. \cite{AC2}). 

1755: 

1756:  For a smooth curve $C$ and for a pencil $\mathfrak g^1_d$ on $C$, we say that an effective divisor $E$ on $C$ is in a fiber of the pencil if there exists 

1757: $E'\in \mathfrak g^1_d$ such that $E'-E$ is an effective divisor.

1758: 

1759: Recall that for $g\geq 3$ and

1760:  $0\leq i\leq \frac{g-1}{2}$ we have defined the divisor $D$

1761: on $\mathcal{M}_{g,g+1}$ to be the locus of curves

1762:  $(C, x_1,\ldots ,x_{g-i}, y_1,\ldots ,y_{i+1})$

1763:  having a linear series $\mathfrak g^1_{g}$ containing 

1764: $x_1+\cdots +x_{g-i}$ in a fiber and $y_1+\cdots +y_{i+1}$ in another fiber.

1765: We denote by $\overline{D}$ the closure of $D$ in

1766:  $\overline{\mathcal{M}}_{g,g+1}$.

1767: 

1768: The divisor $D$ comes equipped with a scheme structure induced by the forgetful map $\mathcal{G}\rightarrow \mathcal{M}_{g,g+1}$. Here $\mathcal{G}$ is the variety parametrizing objects $[C,\vec{x},\vec{y},l]$, where $\vec{x}=(x_1,\ldots,x_{g-i})$ and $\vec{y}=(y_1,\ldots, y_{i+1})$ are such that $[C,\vec{x},\vec{y}]\in \mathcal{M}_{g,g+1}$ and $l$ is a linear series $\mathfrak g^1_g$ on $C$ such that $l(-\sum_{j=1}^{g-i} x_j)\neq \emptyset$ and

1769: $l(-\sum_{j=1}^{i+1} y_j)\neq \emptyset$.

1770: It is well-known that $\mathcal{G}$ is smooth of pure dimension $4g-3$ (see e.g. \cite{AC1}, pg. 346). Note also that there is a natural action of $S_{g-i}\times S_{i+1}$  on $\mathcal{M}_{g,g+1}$ (and hence on $\mathcal{G}$) by permuting the components of $\vec{x}$ and $\vec{y}$ separately.

1771: 

1772: The main result of the section is the following:

1773: 

1774: \begin{theorem}\label{class_D}

1775: The divisor $D$ is reduced and its class in $\Pic(\overline{\mathcal{M}}_

1776: {g,g+1})_{\QQ}$ is

1777: $$[D]=-\Bigl({g-1\choose  i}

1778: -10{g-3\choose i-1}\Bigr)\ \lambda+{g-2\choose i}\Psi

1779: _x+{g-2\choose i-1}\Psi_y, $$

1780: where $\Psi_x=\sum_{j=0}^{g-i} \psi_{x_j}$

1781:  and $\Psi_y=\sum_{j=0}^{i+1} \psi_{y_j}$.

1782: \end{theorem}

1783: 

1784: \noindent

1785: We begin by proving the first part of Theorem~\ref{class_D}:

1786: 

1787: \begin{proposition}\label{reduced}

1788: The divisor $D$ is reduced.

1789: \end{proposition}

1790: 

1791:  \begin{proof}

1792:  Since the variety $\mathcal{G}$ introduced above is smooth, it suffices to show that the projection $\pi:\mathcal{G}\rightarrow

1793:  \mathcal{M}_{g,g+1}$ given by $\pi([C,\vec{x},\vec{y},l])=[C,\vec{x},\vec{y}]$,

1794: is generically injective. 

1795: 

1796: We pick a component $\mathcal{X}$ of $\mathcal{G}\times _{\mathcal{M}_{g,g+1}}\mathcal{G}$ whose general point corresponds to a marked curve $[C,\vec{x},\vec{y}]\in \mathcal{M}_{g,g+1}$ together with two {\emph{different}} base-point-free $\mathfrak g^1_g$'s on $C$, both containing $\vec{x}$ and $\vec{y}$ in different fibers. Clearly $\mbox{dim}(\mathcal{X})\geq 4g-4$ and if we show that $\mbox{dim}(\mathcal{X})\leq 4g-4$, then we are done.

1797: For a general point in $\mathcal{X}$ we denote by $f_1,f_2:C\rightarrow \PP^1$ the induced $g$-sheeted maps. We may assume that $f_1(\vec{x})=f_2(\vec{x})=0$ and $f_1(\vec{y})=f_2(\vec{y})=\infty.$

1798: The product map $f=(f_1,f_2):C

1799: \rightarrow \PP^1\times \PP^1$ is birational onto its image

1800:  and $\Gamma=f(C)$ will have points of multiplicity

1801:  at least $g-i$ and $i+1$ at $a=(0,0)$ and $b=(\infty,\infty)$ respectively.

1802: 

1803:  If $S=\mbox{Bl}_{\{a,b\}}(\PP^1\times \PP^1)$ we set $\gamma= gl+gm-(g-i)E_a-(i+1)E_

1804: b\in \mbox{Pic(S)}$, where $l$ and $m$ are pullbacks of

1805:  the rulings on $\PP^1\times \PP^1$,

1806:  and $E_a$, $E_b$ are the exceptional divisors. We denote by $V(S,\gamma)$ 

1807: the Severi variety of curves $Y\subset S$ homologous to $\gamma$. 

1808: The discussion above shows that $\mathcal{X}$ lies in the closure of the image of the rational map $V(S,\gamma)- ->(\mathcal{G}\times_{\mathcal{M}_{g,g+1}}\mathcal{G})/S_{g-i}\times S_{i+1}$

1809: obtained by projecting $S$ onto the two factors. Thus $\mbox{dim}(\mathcal{X})\leq \mbox{dim }V(S,\gamma)-\mbox{dim} \mbox{ Aut}(S).$

1810: 

1811: 

1812: On the other hand an argument identical to that in \cite{AC1}, Proposition 2.4, 

1813: shows that since $S$ is a regular surface, every irreducible component $M$ of $V(S,\gamma)$ 

1814: having dimension $\geq g+1$ is of the expected dimension provided by deformation theory, 

1815: that is, 

1816: $\mbox{dim}(M)=g-1-\gamma \cdot K_S$. 

1817: Therefore $\mbox{dim}(\mathcal{X})\leq g-1-\gamma \cdot K_S-

1818: \mbox{dim }\mbox{Aut}(S)=4g-4$.

1819: \end{proof}

1820: 

1821: 

1822: We will prove the second part of Theorem~\ref{class_D} using degeneration

1823: techniques and enumerative geometry.

1824:  

1825: 

1826: \subsection*{Recap on limit linear series (cf. \cite{EH1})}

1827: We recall that for a smooth curve $C$,

1828:  a point $p\in C$ and a linear series $l=(L,V)$

1829:  with $L\in \mbox{Pic}^d(C)$ and $V\in G(r+1, H^0(L))$,

1830:  the \emph {vanishing sequence} of $l$ at $p$

1831:  is obtained by ordering the set $\{\mbox{ord}_p(\sigma)\}_{\sigma \in

1832:  V}$, and it is denoted by 

1833: $$a^l(p):0\leq a_0^l(p)<\ldots <a_r^l(p)\leq d.$$

1834: The \emph{weight} of $l$ at $p$ is defined as

1835:  $w^l(p):=\sum_{i=0}^r (a_i^l(p)-i)$. 

1836: 

1837:  Given a curve $C$ of compact type, a \emph{limit}

1838:  $\mathfrak g^r_d$ on $C$ is a collection of honest

1839:  linear series $l_Y=(L_Y,V_Y)\in G^r_d(Y)$ for each

1840:  component $Y$ of $C$, satisfying the compatibility

1841:  condition that if $Y$ and $Z$ are components of $C$ 

1842: meeting at $p$ then

1843: $$a_i^{l_Y}(p)+a_{r-i}^{l_Z}(p)\geq d\mbox{ }\mbox{ for }i=0,\ldots ,r.$$

1844: We note that limit linear series appear as limits of

1845:  ordinary linear series in $1$-dimensional families of

1846:  curves and there is a useful sufficient criterion

1847:  for a limit $\mathfrak g^r_d$ to be

1848:  \emph{smoothable} (cf. \cite{EH1}, Theorem 3.4).

1849: 

1850: \smallskip

1851: \noindent

1852: We will need the following enumerative result (cf. \cite{H}, Theorem 2.1):

1853: 

1854: \begin{proposition}\label{enumerative}

1855: Let $C$ be a general curve of genus $g$,

1856:  $d\geq \frac{g+2}{2}$ and $p\in C$ a general point.

1857: \begin{itemize}

1858: \item The number of $\mathfrak g^1_d$'s on $C$ containing

1859:  $(2d-g)q$ in a fiber, where $q\in C$ is an unspecified point, is

1860: $$b(d,g)=(2d-g-1)(2d-g)(2d-g+1)\frac{g!}{d!(g-d)!}\mbox{ }.$$

1861: \item If $\beta \geq 1$, $\gamma \geq 1$ are integers

1862:  such that $\beta+\gamma=2d-g$, the number of

1863:  $\mathfrak g^1_d$'s on $C$ containing $\beta p+\gamma q$

1864:  in a fiber for some point $q \in C$ is

1865: $$c(d,g,\gamma)=(\gamma^2(2d-g)-\gamma)\frac{g!}{d!\ (g-d)!}\ .$$

1866: \end{itemize}

1867: \end{proposition}

1868: 

1869: \noindent

1870: The following simple observation will be used repeatedly:

1871: 

1872: \begin{proposition}\label{glue}

1873: Fix $y,z\in N$ and denote by

1874:  $\pi_z:\overline{\mathcal{M}}_{g,N}\rightarrow

1875:  \overline{\mathcal{M}}_{g,N-\{z\}}$ the map forgetting

1876:  the marked point labelled by $z$. If $E$ is any divisor class

1877:  on $\overline{\mathcal{M}}_{g,N}$, then the $\lambda$ and the $\psi_x

1878: $ coefficients of $E$ are the same as those of

1879:  $(\pi_z)_*(E\cdot \delta_{0:yz})$ for all $x\in \{y,z\}^c.$

1880: \end{proposition}

1881: 

1882: \begin{proof}

1883:  We write $E$ uniquely as a combination of $\lambda$, tautological classes 

1884: $\psi_y,\psi_z$ and $\psi_x$ with $x\in \{y,z\}^c$ and boundary divisors. To express $(\pi_z)_*(E\cdot \delta_{0:yz})$ in $\mbox{Pic}(\overline{\mathcal{M}}_{g,N-\{z\}})_{\mathbb Q}$ we use that

1885: $(\pi_z)_*(\lambda\cdot \delta_{0:yz})=

1886: \lambda,\mbox{ }(\pi_z)_*(\psi_x\cdot \delta_{0:yz})

1887: =\psi_x \mbox{ for }x\in \{y,z\}^c$ and that $(\pi_z)_*(\psi_x\cdot \delta_{0:yz})=0$ for $x\in \{y,z\}.$ Moreover we have that $(\pi_z)_*(\delta_{i:S}\cdot \delta_{0:yz})$ is boundary in all cases

1888: except that $(\pi_z)_*(\delta_{0:yz}^2)=-\psi_y$ (cf. \cite{AC2}, Lemma 1.2

1889:  and \cite{logan}, Theorem 2.3). The conclusion follows immediately. 

1890: \end{proof}

1891: 

1892: 

1893:  By a succession of push-forwards, using Proposition \ref{glue}

1894:  we will reduce the problem of computing the class of $D$

1895:  to two divisor class computations in $\overline{\mathcal{M}}_{g,3}$. The main idea is to let all the points $x_j$ and then all the points $y_j$ come together and understand how the geometric condition defining $D$ changes under degeneration. Recall that by $\overline{D}$ we denote the closure of $D$ in $\overline{\mathcal{M}}_{g,g+1}$.

1896: 

1897:  We define the following sequence of divisors:

1898:  starting with $\overline{D}=D_{y_{i+1}}$, for $1\leq j\leq i$

1899:  we define inductively the divisors $D_{y_j}$ on

1900:  $\overline{\mathcal{M}}_{g,g-i+j}$ by

1901: $$D_{y_j}:=(\pi_{y_{j+1}})_*(\Delta_{0:y_jy_{j+1}}\cdot D_{y_{j+1}}).$$

1902: Loosely speaking, $D_{y_j}$ is obtained from $D_{y_{j+1}}$

1903:  by letting the marked points $y_j$ and $y_{j+1}$ come together.

1904: Then we define $D_{x_{g-i}}:=D_{y_1}$ and we let the

1905:  marked points $x_2,\ldots,x_{g-i}$ come together

1906: : for $2\leq j\leq g-i-1$

1907:  we define inductively the divisors $D_{x_j}$ on

1908:  $\overline{\mathcal{M}}_{g,j+1}$ by

1909: $$D_{x_j}:=(\pi_{x_{j+1}})_*(\Delta_{0:x_j x_{j+1}} \cdot D_{x_{j+1}}).$$

1910: Proposition~\ref{glue}

1911:  ensures that the $\psi_{x_1}$ and the $\lambda$ coefficients

1912: of $[\overline{D}]$ are the same as those of $[D_{x_2}]$.

1913: 

1914: \begin{proposition}\label{reduce}

1915:  The divisor $D_{x_2}$ is reduced and

1916:  it is the closure in $\overline{\mathcal{M}}_{g,3}$

1917:  of the locus of those smooth pointed curves $(C,x_1,x_2,y)$

1918:  for which there exists a $\mathfrak g^1_g$ with

1919:  $(i+1)y$ in a fiber and $x_1+(g-i-1)x_2$ in another fiber.

1920: \end{proposition}

1921: 

1922: \begin{proof}

1923:  For simplicity we will only prove that

1924:  $D_{y_i}$ is reduced and that it is the closure of the locus

1925:  of those smooth pointed curves $(C,x_1,\ldots,x_{g-i},y_1,\ldots,y_i)$

1926:  for which $x

1927: _1+\cdots +x_{g-i}$ and $y_1+\cdots +y_{i-1}+2y_i$ are

1928:  in different fibers of the same $\mathfrak g^1_g$.

1929:  Then by iteration we will get a similar statement for $D_{x_2}$.

1930: 

1931:  Let $(X=C\cup_q \PP^1,x_1,\ldots ,x_{g-i}, y_1,\ldots ,y_{i+1})$

1932:  with $y_i, y_{i+1}\in \PP^1$ be a general point in a

1933:  component of $D_{y_{i+1}}\cap \Delta_{0:y_i y_{i+1}}$. A standard dimension count shows that $C$ must be smooth. There exists a limit $\mathfrak g^1_g$ on $X$,

1934:  say $l=(l_C,l_{\PP^1})$, together with sections

1935:  $\sigma_{\PP^1} \in V_{\PP^1}$ and

1936:  $\sigma_C, \tau_C \in V_C$, such that

1937:  $\mbox{div}(\tau_C)\geq x_1+\cdots +x_{g-i}$,

1938:  $\mbox{div}(\sigma_C)\geq y_1+\cdots +y_{i-1}$,

1939:  $\mbox{div}(\sigma_{\PP^1})\geq y_i+y_{i+1}$ and moreover

1940:  $\mbox{ord}_q(\sigma_{\PP^1})+\mbox{ord}_{q}(\sigma_C)\geq g$

1941:  (apply \cite{EH1}, Proposition 2.2). 

1942: 

1943: Clearly $\mbox{ord}_q(\sigma_{\PP^1})\leq g-2$, hence

1944:  $\mbox{div}(\sigma_C)\geq 2q+y_1+\cdots +y_{i-1}$.

1945:  The contraction map $\pi_{y_{i+1}}$ collapses $\PP^1$

1946:  and identifies $q$ and $y_i$, so the second part of the claim follows.

1947: 

1948:  To conclude that $D_{y_i}$ is also reduced we use

1949:  that both $D_{y_{i+1}}$ and $\Delta_{0:y_i y_{i+1}}$

1950:  are reduced and that they meet transversally.

1951:  This is because the limit $\mathfrak g^1_g$ we found on $X$

1952:  is smoothable in such a way that all ramification is kept

1953:  away from the nodes (cf. \cite{EH1}, Proposition 3.1), hence the tangent spaces to $D_{y_i}$ and $\Delta_{0:y_iy_{i+1}}$ at the intersection point $(X,\vec{x},\vec{y})$ cannot be equal.

1954: \end{proof}

1955: 

1956: \smallskip

1957: 

1958:  In a similar way, by letting first all $x_j$ with

1959:  $1\leq j\leq g-i$ and then all $y_j$ with $2\leq j\leq i+1$

1960:  coalesce, we obtain a reduced divisor $D_{y_2}$ on

1961:  $\overline{\mathcal{M}}_{g,3}$ which is the closure of the locus

1962:  of smooth curves $(C,x,y_1,y_2)$ having a $\mathfrak g^1_g$

1963:  with $(g-i)x$ and $y_1+iy_2$ in different fibers.

1964:  Moreover, the $\lambda$ and the $\psi_{y_1}$ coefficients of

1965:  $[\overline{D}]$ coincide with those of $[D_{y_2}]$.

1966:  Once more applying Proposition~\ref{glue} it follows that the $\lambda$

1967:  and the $\psi_{y_1}$ coefficients of $[D_{y_2}]$ are the same

1968:  as those of 

1969:  $(\pi_x)_*([D_{y_2}]\cdot \delta_{0:xy_2})$. Similarly, 

1970: the $\psi_{x_1}$ coefficient of $[D_{x_2}]$ is the same as that

1971:  of $(\pi_y)_*([D_{x_2}]\cdot \delta_{0:x_2y})$.

1972:  

1973: \begin{proposition}\label{sum_Y}

1974:  We have that 

1975: $$(\pi_x)_*(D_{y_2}\cdot \Delta_{0:xy_2})=\sum_{j=0}^i

1976: Y_j,$$

1977: where for $j<i$ the reduced divisor $Y_j$ is the closure in

1978:  $\overline{\mathcal{M}}_{g,2}$ of the locus of curves

1979:  $(C,y_1,y_2)$ having a $\mathfrak g^1_{g-j}$ with

1980:  $(g-2j-1)y_2+y_1$ in a fiber, while the reduced divisor $Y_i$ consists of curves

1981:  $(C,y_1,y_2)$ with

1982: a $\mathfrak g^1_{g-i}$ having $(g-2i)y_2$ in a fiber

1983:  (and no condition on $y_1$).

1984: \end{proposition}

1985: 

1986: \begin{proof}

1987:  Once again, let $(X=C\cup_q \PP^1,x,y_1,y_2)$

1988:  be a point in $D_{y_2}\cap \Delta_{0:xy_2}$, with $y_1\in C$

1989:  and $x,y_2\in \PP^1$. Then there exists a limit

1990:  $\mathfrak g^1_g$, say $l=(l_C,l_{\PP^1})$ on $X$ together

1991:  with sections $\sigma_{\PP^1},\tau_{\PP^1}\in

1992:  V_{\PP^1}$ and $\sigma_C\in V_C$ such that 

1993: $\mbox{div}(\sigma_{\PP^1})\geq iy_2, \mbox{div}

1994: (\tau_{\PP^1})\geq (g-i)x, \mbox{div}(\sigma_C)\geq y_1$

1995:  and moreover $\mbox{ord}_q(\sigma_{\PP^1})

1996: +\mbox{ord}_q(\sigma_C)\geq g$.

1997: 

1998:  The Hurwitz formula on $\PP^1$ and the condition defining 

1999:  a limit linear series give that

2000:  $w^{l_C}(q)\geq w^{l_{\PP^1}}(x)+w^{l_{\PP^1}}(y_2)\geq g-2$.

2001:  On the other hand, since $(X,x,y_1,y_2)$ moves in a family

2002:  of dimension $\geq 3g-2$ it follows that $(C,q)$ also moves in a

2003:  family of dimension $\geq 3g-3$ in $\mathcal{M}_{g,1}$

2004:  (i.e. codimension $\leq 1$). Since according to \cite{EH2}, Theorem 1.2,

2005:  the locus of pointed curves $[C,q]\in \mathcal{M}_{g,1}$ carrying a $\mathfrak g^1_g$ having $w(q)\geq g$ has codimension $\geq 2$, we get $w^{l_C}(q)\leq g-1$. There are two possibilities:

2006: 

2007:  \textbf{i)} $w^{l_C}(q)=g-2$. Let us denote $j=a_0^{l_C}(q)$,

2008:  hence $a_1^{l_C}(q)=g-1-j$ and $a_k^{l_C}(q)+a_{1-k}^{l_{\PP^1}}(q)=g$

2009:  for $k=0,1$. Therefore

2010:  $j+1=a_0^{l_{\PP^1}}(q)\leq \mbox{ord}_q(\tau_{\PP^1})\leq i.$

2011:  Moreover, since $\mbox{ord}_{q}(\sigma_{\PP^1})

2012: \leq g-i\leq g-j-1$, we obtain that $\mbox{ord}_q(\sigma_C)\geq j+1$,

2013:  hence $\mbox{div}(\sigma_C)\geq y_1+(g-1-j)q$,

2014:  that is, $l_C(-jq)$ is a $\mathfrak g^1_{g-j}$

2015:  on $C$ with $(g-2j-1)q+y_1$ in a fiber, or equivalently $

2016: [C,y_1,q]\in Y_j$, where $0\leq j\leq i-1$.

2017: 

2018: 

2019: To see that conversely $\bigcup_{j=0}^{i-1}Y_j\subseteq

2020:  (\pi_{x})_*(D_{y_2}\cdot \Delta_{0:xy_2})$ we pick a general

2021:  pointed curve $(C,y_1,q)$ having a $\mathfrak g^1_{g-j}$ with

2022:  $(g-2j-1)q+y_1$ in a fiber and we construct a Harris-Mumford admissible

2023:  covering $f:X'\rightarrow B$ of degree $g$,

2024:  where $X'$ is a curve semistably equivalent to $X$ defined as above, and

2025:  $B=(\PP^1)_1\cup_{t}(\PP^1)_2$ is the transversal

2026:  union of two lines (see Fig. 1): we take $f_{|C}:C\rightarrow (\PP^1)_1$

2027:  to be the degree $g-j$ covering 

2028: such that $(g-2j-1)q+y_1\subseteq f_{|C}^*(t)$, while

2029:  $f_{|\PP^1}:\PP^1\rightarrow (\PP^1)_2$

2030:  is the degree $g-j-1$ map containing $(g-i)x$ and $iy_2$

2031:  in different fibers and with $(g-2j-1)q$ in the fiber over $t$.

2032:  It is clear that there

2033:  is a unique such $\mathfrak g^1_{g-j-1}$ on $\PP^1$.

2034:  Furthermore, at $y_1$ we insert a rational curve $R$ mapping

2035:  isomorphically onto $(\PP^1)_2$ and at the remaining $j$ points

2036:  in $f_{|C}^{-1}(t)-\{y_1,q\}$ we insert rational curves mapping 

2037: with degree $1$ onto $(\PP^1)_2$ while at the $g-j$ points

2038:  in $f_{|\PP^1}^{-1}(t)-\{q\}$ we insert

2039:  copies of $\PP^1$ mapping isomorphically onto $(\PP^1)_1$. 

2040: We denote the resulting curve by $X'$. If $y_1'=f_{|R}^{-1}(f(y_2))$, then

2041: $(X',x,y_1',y_2)$ is stably equivalent to

2042:  $(X,x,y_1,y_2)$ and $iy_2+y_1'$ and $(g-i)x$ appear in distinct fibers of the $g$-sheeted map $f:X'\rightarrow B$. Thus we get that $[X,x,y_1,y_2]\in D_{y_2}\cap \Delta_{0:xy_2}$.

2043: 

2044: 

2045: \begin{figure}[ht]

2046: \begin{center}

2047: \mbox{\epsfig{file=cov.eps,width=8cm,height=6cm,angle=0}}

2048: \end{center}

2049: \label{}

2050: \caption{}

2051: \end{figure}

2052: 

2053: 

2054: 

2055: 

2056: 

2057: 

2058: %\begin{figure}[!ht]\label{Boundary Map}

2059:  %       \scalebox{.4}{\includegraphics{covering.eps}}

2060: %\end{figure}        

2061: 

2062: \smallskip

2063: 

2064: \textbf{ii)} $w^{l_C}(q)=g-1$.

2065:  We denote $a^{l_C}(q)=a^{l_{\PP^1}}(q)=(j,g-j)$.

2066:  Since $\mbox{ord}_q(\tau_{\PP^1})\leq i$ we get that

2067:  $j\leq i$. Now $w^{l_C}(q)=g-1$ is already a

2068:  codimension $1$ condition on $\mathcal{M}_{g,1}$,

2069:  so it follows that $\mbox{ord}_q(\sigma_C)=j$,

2070: hence $\mbox{div}(\sigma_{\PP^1})\geq (g-j)q+iy_2$. This yields $i=j$ and

2071: $\mbox{div}(\sigma_{\PP^1})=(g-i)q+iy_2$.

2072:  We thus get that $[C,y_1,q]\in Y_i$. 

2073: 

2074:  Conversely, given $(C,y_1,q)\in \mathcal{M}_{g,2}$ together

2075:  with a $\mathfrak g^1_{g-i}$ on $C$ with $(g-2i)q$ in a fiber,

2076:  we construct a degree $g$ admissible covering

2077:  $f:X'\rightarrow (\PP^1)_1\cup_t(\PP^1)_2$,

2078: which will prove that $[C,q,y_2]\in (\pi_x)_*(D_{y_2}\cdot \Delta_{0:xy_2})$:

2079: we first take $f_{|C}:C\rightarrow (\PP^1)_1$ of degree

2080:  $g-i$ with $(g-2i)q\subseteq f_{|C}^*(t)$.

2081:  Then $f_{|\PP^1}:\PP^1\rightarrow (\PP^1)_2$

2082:  is of degree $g-i$, completely ramified at $x$ and with

2083:  $f_{|\PP^1}^{-1}(t)=(g-2i)q+iy_2$. At $y_2\in \PP^1$ we insert a 

2084: rational curve $R$ which we map $i:1$ to $(\PP^1)_1$

2085:  such that we have total ramification both at $y_2$

2086:  and at the point $y_2'\in R$ characterized by

2087: $f_{|C}(y_1)=f_{|R}(y_2')$. Finally, at each of the points in

2088:  $f_{|C}^{-1}(t)-\{q\}$ we insert a $\PP^1$

2089:  which we map isomorphically onto $(\PP^1)_2$. 

2090: 

2091: Thus we have proved that $\mbox{supp}(\pi_x)_*(D_{y_2}\cdot \Delta_{0:xy_2})=\cup_{j=0}^i \mbox{supp}(Y_j)$. The conclusion now follows if we notice that $D_{y_2}$ is reduced and all admissible coverings we constructed are smoothable, hence $D_{y_2}\cdot \Delta_{0:xy_2}$ is reduced too.

2092: \end{proof}

2093: 

2094: We have thus reduced the problem of computing $[D]$ to that of computing the class of all divisors $Y_j$ on $\mathcal{M}_{g,2}$ for $0\leq j\leq i$.

2095: 

2096: \begin{proposition}\label{class_Y}

2097: For $0\leq j\leq i$ we have the following relations

2098:  in $\rm{Pic}$$(\mathcal{M}_{g,2})_{\mathbb Q}$:

2099: $$Y_j\equiv_{lin}a_j\lambda+b_{1j}\psi_{y_1}+b_{2j}

2100: \psi_{y_2},\mbox{ }\mbox{   where}$$

2101: 

2102: $$a_j=-\frac{g-2j}{g}{g \choose j}+\frac{10(g-2j)}{g-2}{g-2 \choose j-1}

2103:  \mbox{ }\mbox{ }\mbox{ for all }0\leq j\leq i,$$

2104: $$b_{1j}=\frac{g-2j-1}{g-1}{g-1 \choose j}\mbox{ when }

2105: j\leq i-1, b_{1i}=0, b_{2i}=\frac{(g-2i)^3-(g-2i)}{2g-2}{g \choose i}$$

2106:  $$b_{2j}=\frac{(g-2j-1)(g^3-g^2-4g^2j+4j^2g+2jg-2j)(g-2)!}{2j!(g-1)!}

2107:  \mbox{ for }j\leq i-1.$$

2108: \end{proposition}

2109: 

2110: \begin{proof}

2111:  We will compute the class of $Y_j$

2112:  when $j\leq i-1$. The class of $Y_i$ is computed similarly.

2113:  Let us write the following relation in $\mbox{Pic}(\overline{\mathcal{M}}_{g,2})_{\mathbb Q}$:

2114: $$Y_j\equiv_{lin} a_j\lambda+b_{1j}\psi_{y_1}+

2115: b_{2j}\psi_{y_2}-c_j\delta_{0:y_1y_2}+(\mbox{ other boundary terms }).$$

2116: We use the method of test curves to determine the coefficients $a_j,b_{1j}$ and $b_{2j}$, that is, we intersect the classes appearing on both sides of the previous relation with curves inside $\overline{\mathcal{M}}_{g,2}.$ By computing intersection numbers we obtain linear relations between the coefficients $a_j,b_{1j},b_{2j}$.

2117: 

2118: 

2119: By Proposition 4.5 we have that 

2120: \begin{equation}

2121: Z_j:=(\pi_{y_2})_*(Y_j\cdot \Delta_{0:y_1 y_2})

2122: \equiv_{lin} a_j\lambda+c_j\psi_{y_1}+(\mbox{ boundary }).

2123: \end{equation}

2124: Using  the same reasoning as in Proposition~\ref{reduce},

2125:  we obtain that $Z_j$ is the closure in $\overline{\mathcal{M}}_{g,1}$

2126:  of the locus of curves $(C,y_1)$ carrying a

2127:  $\mathfrak g^1_{g-j}$ with $(g-2j)y_1$ in a fiber.

2128: 

2129: In order to determine the coefficient $c_j$ we intersect both sides 

2130: of $(5)$ with a general fiber $F$ of the map

2131:  $\overline{\mathcal{M}}_{g,1}\rightarrow

2132:  \overline{\mathcal{M}}_{g}$: we get that

2133:  $c_j=Z_j\cdot F/\psi_{y_1}\cdot F=b(g-j,g)/(2g-2)$ (cf. Proposition \ref{enumerative}). 

2134: 

2135:   To determine $b_{1j}$ and $b_{2j}$ we use two

2136:  test curves in $\overline{\mathcal{M}}_{g,2}$:

2137:  first, we fix a general curve $C$ of genus $g$ and we obtain

2138:  a family $C_{[1]}=\{(C,y_1,y_2)\}_{y_1\in C}$,

2139:  by fixing a general point $y_2\in C$ and letting $y_1$ vary on $C$.

2140:  From $(5)$, clearly

2141:  $C_{[1]}\cdot Z_j=(2g-1)b_{1j}+b_{2j}-c_j$.

2142:  On the other hand, according to Proposition~\ref{enumerative}

2143: $C_{[1]}\cdot Z_j=c(g-j,g, 1)$.

2144: 

2145:  For a new relation between $b_{1j}$ and $b_{2j}$

2146:  we use the test curve $C_{[2]}=\{(C,y_1,y_2)\}_{y_2\in C}$

2147:  in $\overline{\mathcal{M}}_{g,2}$,

2148:  where this time $y_1$ is a fixed general point

2149:  while $y_2$ varies on $C$. We have the equation $(2g-1)b_

2150: {2j}+b_{1j}-c_j=C_{[2]}\cdot Z_j=c(g-j,g,g-2j-1)$, and since

2151:  $c_j$ is already known we get in this way both $b_{1j}$ and $b_{2j}$.

2152: 

2153:  We are only left with the computation of $a_j$.

2154:  From \cite{EH2}, Theorem 4.1 we know that the class of

2155:  $Z_j$ is a linear combination  of the Brill-Noether

2156:  class and of the class of the divisor of Weierstrass points,

2157:  that is, $Z_j\equiv_{lin} \mu

2158: BN+\nu \mathcal{W}$, where

2159: $$BN:=(g+3)\lambda-\frac{g+1}{6}\delta_{irr}-\sum_{i=1}^{g-1}i(g-i)

2160: \delta_{i:y_1}\mbox{ and }$$

2161: $$\mathcal{W}:=-\lambda+\frac{g(g+1)}{2}\psi_{y_1}-

2162: \sum_{i=1}^{g-1}{g-i+1\choose 2}\delta_{i:y_1}.$$

2163: We already know that $\nu=2c_j/(g(g+1))$.

2164:  To determine $\mu$ we use the following test curve in

2165:  $\overline{\mathcal{M}}_{g,1}$: we take a general curve $B$

2166:  of genus $g-1$ and a general $2$-pointed elliptic curve $(E,0,y_1)$.

2167:  We consider the family $\overline{B}=

2168: \{X_q=B\cup_{q\sim 0}E,y_1\}_{q\in B}$ obtained by identifying

2169:  the variable point $q\in B$ with the fixed point $0\in E$.

2170:  We easily get $\overline{B}\cdot \psi_{y_1}=

2171: \overline{B}\cdot \lambda=0$,

2172:  $\overline{B}\cdot \delta_{1:y_1}=-\mbox{deg}K_B=4-2g$, 

2173: while $\overline{B}$ vanishes on all the other boundaries.

2174:  On the other hand $\overline{B}\cdot Z_j$ is the number of

2175:  limit $\mathfrak g^1_{g-j}$'s on the curves $X_q$ having vanishing

2176:  $g-2j$ at the fixed point $y_1\in E$. If $l=(l_B,l_E)$ is such a linear 

2177: series, then using again the additivity of the Brill-Noether numbers

2178:  (cf. \cite{EH1}, Proposition 4.6) and the assumption that $y_1-0\in

2179:  \mbox{Pic}^0(E)$ is not torsion, we obtain that $w^{l_B}(q)=g-2j$,

2180:  so either $a^{l_B}(q)=(1,g-2j)$ or $a^{l_B}(q)=(0,g-2j+1)$.

2181:  Thus $\overline{B}\cdot Z_j=b(g-j-1,g-1)+b(g-j,g-1)$ and we

2182:  can write a new relation enabling us to compute $a_j$.

2183: \end{proof}

2184: 

2185: 

2186: We can now complete the proof of Theorem~\ref{class_D}:

2187: \begin{proof}[Proof of Theorem~\ref{class_D}]

2188:  Let us write

2189:  $D\equiv_{lin}A\lambda+B_1\Psi_x+B_2\Psi_y$, where

2190:  $\Psi_x:=\sum_{j=1}^{g-i}\psi_{x_j}$ and $\Psi_{y}:=

2191: \sum_{j=1}^{i+1}\psi_{y_j}$. As noticed before, the

2192:  $\{\lambda, \Psi_y\}$-part of $[D]$ and the $\{\lambda, \psi_{y_1}\}$-part

2193:  of $\sum_{j=0}^i [Y_j]$ coincide, hence using Proposition \ref{class_Y}

2194: $$A=\sum_{j=0}^i a_j=-{g-1 \choose i}+10{g-3 \choose i-1}\mbox{ and }

2195:  B_2=\sum_{j=0}^i b_{1j}={g-2\choose i-1}.$$ 

2196: 

2197:  Finally, to determine $B_1$ one has to compute

2198:  the $\psi_{x_1}$ coefficient of the divisor $D_{x_2}$

2199:  on $\overline{\mathcal{M}}_{g,3}$. Arguing in a way that

2200:  is entirely similar to Proposition \ref{sum_Y} we obtain that

2201:  $B_1={g-2 \choose i}$.

2202: \end{proof}

2203:  

2204:  

2205: 

2206: 

2207: 

2208: %----------------------------------------

2209: %%  REFERENCES

2210: %%----------------------------------------------------------

2211: %%\bibliographystyle{amsalpha}

2212: %%\bibliography{MRC}

2213: 

2214: \providecommand{\bysame}{\leavevmode\hbox to3em{\hrulefill}\thinspace}

2215: 

2216: \begin{thebibliography}{EMS}

2217: 

2218: \bibitem[AC1]{AC1} 

2219: E. Arbarello and M. Cornalba, \emph{Footnotes to a paper of Beniamino 

2220: Segre}, Math. Ann. \textbf{256} (1981), 341--362.

2221: 

2222: \bibitem[AC2]{AC2} 

2223: E. Arbarello and M. Cornalba, \emph{Calculating

2224:  cohomology groups of moduli spaces of curves via

2225:  algebraic geometry}, Inst. Hautes Etudes Sci. Publ. Math.

2226:  No. 88 (1998), 97--127.

2227: 

2228: \bibitem[ACGH]{ACGH}

2229: E. Arbarello, M. Cornalba, P.A. Griffiths and J. Harris,

2230: \emph{Geometry of algebraic curves}, Grundlehren 267, Springer, 1985.

2231: 

2232: \bibitem[BG]{BG}

2233: E. Ballico and A.V. Geramita, \emph{The minimal free resolution of the

2234: ideal of $s$ general points in $\PP^3$}, Proceedings of the 1984

2235: Vancouver conference in algebraic geometry, 1--10, CMS Conf. Proc. \textbf{6},

2236: Amer. Math. Soc., Providence, RI, 1986.

2237: 

2238: \bibitem[Ei]{eisenbud}

2239: D. Eisenbud, \emph{Commutative algebra with a view

2240: toward algebraic geometry}, Springer, 1995.

2241: 

2242: \bibitem[EH1]{EH1}

2243: D. Eisenbud and J. Harris,

2244:  \emph{Limit linear series: basic theory}, Invent. Math. \textbf{85} (1986),

2245: no. 2, 337--371.

2246: 

2247: \bibitem[EH2]{EH2}

2248: D. Eisenbud and J. Harris, \emph{Irreducibility of some families

2249:  of linear series with Brill-Noether number $-1$},

2250:  Ann. Scient. Ec. Norm. Sup. (4), \textbf{22} (1989), no. 1, 33--53.

2251: 

2252: \bibitem[EP]{popescu}

2253: D. Eisenbud and S. Popescu, \emph{Gale duality

2254: and free resolutions of ideals of points}, Invent. Math. \textbf{136}

2255: (1999), no. 2, 419--449.

2256: 

2257: \bibitem[EPSW]{epsw}

2258: D. Eisenbud, S. Popescu, F.--O. Schreyer, Ch. Walter,

2259: \emph{Exterior algebra methods for the Minimal Resolution Conjecture}, 

2260: preprint 2000, math.AG/0011236.

2261: 

2262: \bibitem[EV]{ev}

2263: D. Eisenbud and A. Van de Ven, \emph{On the normal bundles of smooth

2264: rational space curves}, Math. Ann. \textbf{256} (1981), 453--463.

2265: 

2266: \bibitem[Fu]{fulton}

2267: W. Fulton, \emph{Intersection theory}. Second edition.

2268: Springer-Verlag, Berlin, 1998.

2269: 

2270: \bibitem[Ga]{gaeta}

2271: F. Gaeta, \emph{A fully explicit resolution of the ideal

2272: defining $N$ generic points in the plane}, preprint 1995.

2273: 

2274: \bibitem[Gr]{green}

2275: M. Green, \emph{Koszul cohomology and geometry}, in 

2276: \emph{Lectures on Riemann surfaces}, World Scientific Press, Singapore, 

2277: 177--200.

2278: 

2279: \bibitem[GL1]{gl1}

2280: M. Green and R. Lazarsfeld, \emph{On the projective normality of complete 

2281: linear series on an algebraic curve}, Invent. Math. \textbf{83} (1986), 73--90.

2282: 

2283: \bibitem[GL2]{gl2}

2284: M. Green and R. Lazarsfeld,

2285:  \emph{A simple proof of Petri's theorem on canonical 

2286: curves}, in \emph {Geometry Today}, Progress in Math. Birkhauser, 1986.

2287: 

2288: \bibitem[GL3]{gl3}

2289: M. Green and R. Lazarsfeld, \emph{Some results on the syzygies of finite sets

2290: and algebraic curves}, Compositio Math. \textbf{67} (1988), 301--314.

2291: 

2292: \bibitem[Ha]{H}

2293: J. Harris, \emph{On the Kodaira dimension of the moduli

2294:  space of curves. The even genus case},

2295: Invent. Math. \textbf{75} (1984), no.3, 437--466.

2296: 

2297: \bibitem[HaMo]{HaMo}

2298: J. Harris and I. Morrison, \emph{Moduli of curves}, Springer-Verlag,

2299: New York 1998.

2300: 

2301: \bibitem[HaMu]{HM}

2302: J. Harris and D. Mumford, \emph{On

2303:  the Kodaira dimension of the moduli space of curves},

2304: Invent. Math. \textbf{67} (1982), no.1, 23--88.

2305: 

2306: \bibitem[HS]{simpson}

2307: A. Hirschowitz and C. Simpson,

2308:  \emph{La r\'esolution minimale de l'arrangement d'un 

2309: grand nombre de points dans $\PP^n$}, Invent. Math. \textbf{126} (1996), no.3, 467--503. 

2310: 

2311: \bibitem[Lz1]{lazarsfeld}

2312: R. Lazarsfeld, \emph{A sampling of vector bundle techniques in

2313: the study of linear series}, in Lectures on Riemann Surfaces, \emph{World

2314: Scientific Press}, Singapore 1989, pp. 500--559.

2315: 

2316: \bibitem[Lz2]{Rob}

2317: R. Lazarsfeld, private communication.

2318: 

2319: \bibitem[Log]{logan} 

2320: A. Logan, \emph{Moduli spaces of curves with marked points}, Ph.D. thesis, Harvard University, 1999.

2321: 

2322: \bibitem[Lo1]{lorenzini1}

2323: A. M. Lorenzini, \emph{On the Betti numbers of points in projective

2324: space}, Ph.D. thesis, Queen's University, Kingston, Ontario, 1987.

2325: 

2326: \bibitem[Lo2]{lorenzini2}

2327: A. M. Lorenzini, \emph{The minimal resolution conjecture},

2328: J. Algebra \textbf{156} (1993), no. 1, 5--35. 

2329: 

2330: \bibitem[Mu]{mustata}

2331: M. Musta\c{t}\v{a}, 

2332: \emph{Graded Betti numbers of general finite

2333: subsets of points on projective varieties}, Le Matematiche, vol. LIII, 1998,

2334: 53--81.

2335: 

2336: \bibitem[PR]{paranjape}

2337: K. Paranjape and S. Ramanan,

2338:  \emph{On the canonical ring of a curve}, Algebraic geometry

2339: and commutative algebra, vol. II, 503--516, Kinokuriya, 1988.

2340: 

2341: \bibitem[Po]{popa}

2342: M. Popa, \emph{On the base locus of the generalized theta divisor},  

2343: C. R. Acad. Sci. Paris \textbf{329} (1999), S\'erie I, 507--512.

2344: 

2345: \bibitem[Ra]{raynaud}

2346: M. Raynaud, \emph{Sections des fibr\'es vectoriels sur une courbe}, Bull.

2347: Soc. Math. France \textbf{110} (1982), 103--125.

2348: 

2349: \bibitem[Wa]{walter}

2350: Ch. Walter, \emph{The minimal free resolution of the homogeneous ideal

2351: of $s$ general points in $\PP^4$}, Math. Zeitschrift \textbf{219} (1995),

2352: no. 2, 231--234.

2353: 

2354: 

2355: \end{thebibliography}

2356: 

2357: \end{document}

2358: 

2359: