1: \documentclass[12pt]{article}

2: 

3: %%%%%%%%%%

4: %

5: % VERSAO ATUAL (23 January 2007)

6: %

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

8: 

9: \usepackage{graphicx}

10: \usepackage{amscd}

11: \usepackage{theorem}

12: 

13: \hyphenation{mo-du-lo}

14: \hyphenation{mo-du-li}

15: \hyphenation{co-ni-cs}

16: \hyphenation{de-ve-lo-ped}

17: 

18: \newfont{\eulerfraktur}{eufm10 scaled\magstep1}

19: \newcommand{\TS}{\mbox{\eulerfraktur {X}}}

20: 

21: \newfont{\Bbb}{msbm10 scaled\magstep1}

22: \def\AA{\mbox{\Bbb A}}

23: \newcommand{\CC}{{\mbox{\Bbb C}}}

24: \newcommand{\EE}{{\mbox{\Bbb E}}}

25: \newcommand{\PP}{{\mbox{\Bbb P}}}

26: \newcommand{\GG}{{\mbox{\Bbb G}}}

27: \newcommand{\NN}{{\mbox{\Bbb N}}}

28: \newcommand{\QQ}{{\mbox{\Bbb Q}}}

29: \newcommand{\LL}{{\mbox{\Bbb L}}}

30: \newcommand{\RR}{{\mbox{\Bbb R}}}

31: \newcommand{\WW}{{\mbox{\Bbb W}}}

32: \newcommand{\DD}{{\mbox{\Bbb D}}}

33: \newcommand{\ZZ}{{\mbox{\Bbb Z}}}

34: 

35: \def\bdm{\begin{displaymath}}

36: \def\be{\begin{equation}}

37: \def\bfe{{\bf e}}

38: \def\bfu{{\bf u}}

39: \def\ka{\kappa_}

40: \def\bfv{{\bf v}}

41: \def\bfx{{\bf x}}

42: \def\bfy{{\bf y}}

43: \def\cw{\overline{wt(2)}}

44: 

45: \def\CMF{\overline{{\cal M}}_g}

46: \def\CMS{\overline{M}_g}

47: \def\CMSP{\overline{M}_{g,1}}

48: \def\dex{\frac{\d}{\d x}}

49: \def\dey{\frac{\d}{\d y}}

50: \def\d{{\partial}}

51: \def\edm{\end{displaymath}}

52: \def\ee{\end{equation}}

53: 

54: \def\fl{f{\hskip 1pt}l}

55: \def\f{~(\ref}

56: 

57: \def\hb{\lambda}

58: \def\KK{{\bf K}}

59: \def\kk{{\bf k}}

60: \def\lra{{\longrightarrow}}

61: 

62: \def\norma{\|.\|}

63: \def\ora{\overrightarrow}

64: 

65: \def\cocoa{{\hbox{\rm C\kern-.13em o\kern-.07em C\kern-.13em o\kern-.15em A}}}

66: \def\wh{\hat}

67: \def\wgt{{\rm wt}}

68: \renewcommand{\:}{\colon\,}

69: \newcommand{\x}{\times}

70: \newcommand{\ol}{\overline}

71: \newcommand{\wt}{\widetilde}

72: \newcommand{\w}{\omega}

73: 

74: \def\proof{{\bf Proof.}\,\,\,}

75: 

76: \def\qed{\hfill\vrule height4pt width4pt depth0pt}

77: \def\rds{{\omega_\pi}}

78: 

79: 

80: \def\gf{\pi:\TS\lra S}

81: \def\ggp{{\overline{\eta}}}

82: \def\uu{(u_1,\ldots,u_n)}

83: \def\vv{(v_1,\ldots,v_n)}

84: 

85: \def\xx{(x_1,\ldots,x_n)}

86: 

87: \def\dmyb{(\pi^*\bigwedge^g\pi_*\rds)^\vee}

88: \def\myb{\pi^*\bigwedge^g\pi_*\rds}

89: \def\wesp{{\otimes\frac{g(g+1)}{2}}}

90: \def\N{\frac{g(g+1)}{2}}

91: \def\umega{{\underline{\omega}}}

92: \def\hy{hyperelliptic\,\,}

93: 

94: \theoremstyle{change} 

95: \newtheorem{claim}{\hskip-0.2cm}[section]

96: 

97: \begin{document}

98: 

99: \title{Special ramification loci 

100: on the double product of a general curve

101: \thanks{2000 Mathematics Subject Classification: 

102: 14H10, 14H15, 14N10.}}

103: 

104: 

105: \author{{\small{{ C. CUMINO $^{(\natural)}$}, 

106: E. ESTEVES $^{(\flat)}$  AND  

107: {L. GATTO} $^{(\natural)}$ }}

108: \thanks{Work partially sponsored by MURST 

109: (Progetto Nazionale ``Geometria delle Variet\`a Proiettive'' 

110: Coordinatore Sandro Verra), and supported by 

111: GNSAGA-INDAM. The second author was also supported

112: by CNPq, Proc. 478625/03-0 and 301117/04-7, 

113: and CNPq/FAPERJ, Proc. E-26/171.174/2003.}\\

114: { }\\

115: {\normalsize $^{(\natural)}$ Dipartimento di 

116: Matematica, Politecnico di Torino,}\\

117: {\normalsize Corso Duca degli Abruzzi 24, 

118: 10129 Torino -- (ITALY)}\\{æ}

119: \\{\normalsize $^{(\flat)}$ Instituto Nacional de 

120: Matem\'atica Pura e Aplicada,}\\

121: {\normalsize Estrada Dona Castorina 110} \\ 

122: {\normalsize 22460-320 Rio de Janeiro RJ -- (BRAZIL)}}

123: 

124: \date{ }

125: 

126: \maketitle

127: 

128: \begin{abstract}

129: \noindent

130: Let $C$ be a general connected, smooth, projective 

131: curve of 

132: positive genus $g$. For each integer $i\geq 0$ we 

133: give formulas for the number of pairs 

134: $(P,Q)\in C\times C$ off the diagonal such that 

135: $(g+i-1)Q-(i+1)P$ is linearly equivalent to an 

136: effective divisor, and the number of pairs 

137: $(P,Q)\in C\times C$ off the diagonal such that 

138: $(g+i+1)Q-(i+1)P$ is linearly equivalent to a moving 

139: effective divisor.

140: \end{abstract}

141: 

142: \section{Introduction}\label{sec1}

143: 

144: Let $C$ be a general connected, smooth, projective 

145: curve of genus $g>0$. Put $C^2:=C\times C$, and let 

146: $\Delta\subset C^2$ be the diagonal. For each 

147: integer $i\geq 0$ consider the 

148: following loci on $C^2$:

149:      \begin{eqnarray*}

150:        D_i&:=&\{(P,Q)\in C^2-\Delta\,|\,

151:        h^0({\cal O}_C((g+i-1)Q-(i+1)P))>0\},\\

152:        E_i&:=&\{(P,Q)\in C^2-\Delta\,|\,

153:        h^0({\cal O}_C((g+i+1)Q-(i+1)P))>1\}.       

154:      \end{eqnarray*}

155: Our Proposition \ref{A2CC} 

156: claims that $D_i$ and $E_i$ are 

157: finite, and our main result, 

158: Theorem \ref{thm6DiazCuk}, gives 

159: formulas for the number of points in $D_i$ and $E_i$.

160: 

161: A formula for the number of points in $D_i$ appeared 

162: already as Lemma~6.3 on page 24 

163: of the seminal work by 

164: Diaz \cite{Diaz/Exc}, where the unnecessary extra 

165: hypotheses that $g\geq 2$ and $i\geq 2$ are made. 

166: Diaz used this formula to compute the class in 

167: the moduli space of genus-$g$ stable curves 

168: $\CMF$ of the closure $\ol{\cal D}_g$ 

169: of the locus of smooth 

170: curves $C$ having a Weierstrass point $P$ 

171: of type $g-1$, i.e. such that 

172: $h^0({\cal O}_C((g-1)P))\geq 2$.

173: 

174: Later on, Cukierman \cite{Cukie} gave a formula for 

175: the class in $\CMF$ of the closure $\ol{\cal E}_g$ 

176: of the locus of smooth 

177: curves $C$ containing a Weierstrass point $P$ of 

178: type $g+1$, 

179: i.e. such that $h^0({\cal O}_C((g+1)P))\geq 3$. 

180: He did not follow in Diaz's 

181: footsteps for this formula, but 

182: rather observed that the union 

183: $\ol{\cal D}_g\cup\ol{\cal E}_g$ is the branch 

184: locus of the Weierstrass divisor on the 

185: ``universal'' curve over $\CMF$, and used a Hurwitz 

186: formula with singularities to compute the class of 

187: this branch divisor.

188:  

189: Had Cukierman followed in Diaz's footsteps, he 

190: would probably have found he needed a formula for 

191: the number of points in $E_i$. We give this formula 

192: here.

193: 

194: In fact, in a sense to be explained below, it is 

195: slightly easier to obtain the number of points in 

196: $E_i$ than in $D_i$, though we obtain both in a 

197: quite integrated form here. 

198: To obtain these numbers, the 

199: natural procedure is to use Porteous formula 

200: to compute the virtual classes of certain 

201: natural ramification schemes $D_i^+$ and $E_i^+$ of 

202: maps of vector bundles on $C^2$; see 

203: Subsection \ref{srs}. Set-theoretically, $D_i^+$ and 

204: $E_i^+$ are given exactly as $D_i$ and $E_i$, but 

205: without the restriction that the pair $(P,Q)$ lies 

206: off $\Delta$. 

207: 

208: The problem is that $D_i^+$ and $E_i^+$ are both 

209: larger than $D_i$ and $E_i$. Indeed, $E_i^+$ is the 

210: union of $E_i$ with the set of points $(P,P)$ such 

211: that $P$ is a Weierstrass point of $C$ and, worse, 

212: $D_i^+$ is the union of $D_i$ and the whole diagonal 

213: $\Delta$. Since $E_i^+$ is finite, 

214: Porteous formula does give an 

215: expression for the number of points in $E_i^+$, with 

216: weights, and thus at least an upper 

217: bound for the number of points in $E_i$. But it 

218: does not a priori give any information on $D_i$.

219: 

220: To compute the number of points in $D_i$ and $E_i$, 

221: we use the fact that, by the Riemann-Roch Theorem, 

222: the union of $D_i$ and $E_i$ is the locus $SW_i$ of 

223: pairs $(P,Q)\in C^2-\Delta$ such that $Q$ is a 

224: special ramification point of the complete linear 

225: system $H^0(\omega_C((i+1)P))$, where $\omega_C$ 

226: is the canonical bundle of $C$. 

227: 

228: We give $SW_i$ a scheme structure as follows. First, 

229: we consider the ramification divisor 

230: $Z_i\subset C^2$ of the 

231: family of linear systems $H^0(\omega_C((i+1)P))$ 

232: parameterized by $P\in C$. Our Proposition \ref{prop5} 

233: implies that 

234: $Z_i$ contains $\Delta$ with multiplicity exactly 

235: $g$. Set $W_i:=Z_i-g\Delta$. Our 

236: Proposition \ref{prop7} 

237: gives an expression for the cycle $[W_i]$, and 

238: our Proposition \ref{lem} claims that $W_i$ is 

239: nonsingular. Furthermore, in Subsection \ref{srs} we 

240: observe that the branch divisor of $W_i$ with 

241: respect to the projection $p_1\: C^2\to C$ over the 

242: first factor has support $SW_i$. We give 

243: $SW_i$ the structure of this branch divisor. 

244: 

245: The advantage of considering $SW_i$ is 

246: that it is quite easy to compute its degree. 

247: Indeed, $[SW_i]$ is the second Chern class 

248: of the bundle of first-order relative jets of $p_1$ 

249: with coefficients in ${\cal O}_{C^2}(W_i)$. 

250: Having an expression for $[W_i]$ we derive very 

251: quickly an expression for $\int_{C^2}[SW_i]$ in 

252: Proposition \ref{prop07}

253: 

254: Now, giving $D_i$ and $E_i$ the subscheme 

255: structures induced from $D_i^+$ and $E_i^+$, our 

256: Proposition \ref{A2CC} shows that, as 0-cycles, 

257:         $$[D_i]+[E_i]=[SW_i].$$

258: Actually, $D_i$ and $E_i$ are reduced. Indeed, in 

259: the proof of Theorem \ref{thm6DiazCuk} we show 

260: that the weight of $(P,Q)$ in $[SW_i]$ is at most 2, 

261: and the maximum weight is achieved if and only if 

262: $(P,Q)\in D_i\cap E_i$. 

263: 

264: Now, as we already know $\int_{C^2}[SW_i]$, 

265: it is enough to compute either $\int_{C^2}[D_i]$ or 

266: $\int_{C^2}[E_i]$. As mentioned above, we compute 

267: the latter. In fact, we can get $\int_{C^2}[E_i^+]$ 

268: using Porteous formula, and a local analysis, done 

269: in Proposition \ref{A2CC}, 

270: shows that the weight of $(P,P)$ 

271: in $[E_i^+]$ is equal to $g+1$ for each Weierstrass 

272: point of $C$. Thus $\int_{C^2}[E_i]$ follows.

273: 

274: In a second article \cite{CEG}, we 

275: show how the knowledge of the number of points in 

276: $E_i$ can be used to compute the class of 

277: $\ol{\cal E}_g$ in $\ol{\cal M}_g$. 

278: This computation is not 

279: straightforward as, following in Diaz's footsteps, 

280: we have to determine 

281: the limits of 

282: special Weierstrass points of type $g+1$ on stable 

283: curves with just one node. This is the main result 

284: of \cite{CEG}. 

285: 

286: The limits of special 

287: Weierstrass points of type $g-1$ were computed by 

288: Diaz, using admissible covers. However, the same 

289: method does not apply to points of type $g+1$. For 

290: those we apply in \cite{CEG} 

291: the theory of limit linear series in 

292: a rather new way, using 2-parameter families. 

293: Actually, as in the present article, we use an 

294: integrated approach in \cite{CEG} that 

295: yields simultaneously 

296: the limits of special Weierstrass points 

297: of both types, and also formulas for 

298: the classes of

299: both $\ol{\cal D}_g$ and $\ol{\cal E}_g$.

300: 

301: Here is a layout of the article. In 

302: Section \ref{sec2} we 

303: review the theory of linear systems and ramification 

304: on 

305: a

306: smooth curve $C$, introduce the linear 

307: systems 

308: we will consider in the remaining of the article, 

309: $H^0(\omega_C((i+1)P))$ for $P\in C$, and prove a 

310: preliminary result about them. In 

311: Section \ref{sec3}, 

312: assuming $C$ is general, we obtain through 

313: degeneration methods results that bound the order 

314: sequence of $H^0(\omega_C((i+1)P))$ at any point of 

315: $C$. In Section \ref{sec4}, 

316: we describe the structure 

317: of the ramification divisor $Z_i\subset C^2$ of the 

318: family of linear systems $H^0(\omega_C((i+1)P))$ 

319: parameterized by $P\in C$. 

320: Finally, in Section \ref{sec5} we define the loci 

321: $D_i$ and $E_i$ and compute their number of points, 

322: through the study of the locus $SW_i$ of special 

323: ramification points of the family 

324: $H^0(\omega_C((i+1)P))$ for $P\in C$.

325: 

326: We thank Nivaldo Medeiros for discussions on related 

327: topics. Also, we acknowledge the use of \cocoa 

328: \cite{cocoa} for some of the computations.

329: 

330: \section{Setup}\label{sec2}

331: 

332: \begin{claim} 

333: {\rm ({\it Linear systems and ramification})

334: Let $C$ be a {\it smooth curve}, that is, 

335: a projective, connected,

336: smooth scheme of dimension 1 over $\CC$. Denote by $\w_C$

337: its canonical sheaf. Let $g:=h^0(C,\w_C)$, 

338: the genus of $C$.

339: 

340: Let $V$ be a $\CC$-vector space of sections

341: of a line bundle $\cal L$ on $C$. We call $V$ a

342: {\it linear system}. The linear system is

343: called {\it complete} if $V=H^0(C,{\cal L})$.

344: Let $r:=\dim V-1$ and $d:=\deg{\cal L}$. We call

345: $r$ the {\it rank} of $V$ and $d$ its {\it degree}. We say

346: as well that $\dim V$ is the {\it dimension} of $V$.

347: 

348: For each point $P$ of $C$, and each integer $j\geq 0$,

349: let $V(-jP)$ denote the vector subspace of $V$ of sections

350: of $\cal L$ that vanish with order at least

351: $j$ at $P$. We say that $j$ is an {\it order} of

352: $V$ at $P$ if $V(-jP)\neq V(-(j+1)P)$. There are

353: $r+1$ orders, which, in an increasing sequence, will be

354: denoted by

355:        $$\epsilon_0(V,P),\,\epsilon_1(V,P),\,\dots,\,

356:        \epsilon_r(V,P).$$

357: 

358: For each integer $\ell\geq 0$ and 

359: each line bundle ${\cal M}$ on

360: $C$, let ${\cal J}^\ell_C({\cal M})$ 

361: be the bundle of {\it jets},

362: or {\it principal parts}, of $\cal M$ 

363: truncated in order $\ell$.

364: Consider the map of rank-$r$ bundles,

365:       $$V\otimes{\cal O}_C\longrightarrow

366:       {\cal J}^r_C({\cal L}),$$

367: locally obtained by differentiating up to 

368: order $r$ the sections

369: of $\cal L$ in $V$. The {\it wronskian} $w_V$ of 

370: $V$ is the (nonzero) section of

371:       $${\cal L}^{\otimes r+1}\otimes

372:       \w_C^{\otimes r(r+1)/2}$$

373: induced by taking determinants in the

374: above map of bundles.

375: 

376: For each point $P$ of $C$, the {\it weight} $\wgt_V(P)$ of

377: $P$ in $V$ is the order of vanishing of $w_V$ at $P$.

378: We call $P$ a {\it ramification point} of $V$ if

379: $\wgt_V(P)>0$; otherwise we call $P$ {\it ordinary}.

380: A local analysis yields the formula

381:        $$\wgt_V(P)=\sum_{j=0}^r

382:        (\epsilon_j(V,P)-j).$$

383: We call $P$ a {\it simple} ramification point if

384: $\wgt_V(P)=1$; otherwise we call $P$ {\it special}.

385: The point $P$ is special if and only if the section

386:        $$Dw_V\in H^0\Big(C,{\cal J}^1_C

387:        ({\cal L}^{\otimes r+1}\otimes

388:        \w_C^{\otimes r(r+1)/2})\Big),$$

389: locally obtained from $w_V$ by differentiating, 

390: vanishes at $P$.

391: 

392: The {\it total weight} of the ramification points of

393: $V$ is the (finite) sum

394:       $$\wgt_V:=\sum_{P\in C}\wgt_V(P).$$

395: It is equal to the degree of the line bundle of which

396: $w_V$ is a section, that is,

397:       $$\wgt_V=(r+1)(d+(g-1)r),$$

398: a formula usually referred to as the

399: {\it Brill--Segre} or {\it Pl\"ucker formula}.

400: 

401: The {\it canonical system} is the complete linear

402: system of sections of $\w_C$. Its rank is $g-1$, and its

403: degree is $2(g-1)$. For each point $P$ of $C$, its

404: {\it Weierstrass weight} $\wgt(P)$

405: is its weight in the canonical system, and the

406: {\it Weierstrass order sequence} at

407: $P$ is the increasing sequence of orders at $P$ of

408: the canonical system.

409: 

410: For each integer $i\geq -1$ and each $P\in C$,

411: let $V_C(i,P)$ denote the complete linear system of

412: sections of $\w_C((i+1)P)$.}

413: \end{claim}

414: 

415: \begin{claim}\label{prop5}{\bf Proposition.} 

416: Let $C$ be a smooth

417: curve of genus $g$. For each integer

418: $i\geq 0$ and each $P\in C$, the following two

419: statements hold for $V:=V_C(i,P)$:

420: \begin{enumerate}

421: \item\label{it5}

422: The weight $\wgt_V(P)$ of $P$ as a ramification point of

423: $V$ satisfies

424:       \be

425:       \wgt_V(P)=g+\wgt(P),\label{eq:15}

426:       \ee

427: where $\wgt(P)$ is the Weierstrass weight of $P$.

428: \item \label{it4} The total weight $\wgt_V$ of

429: the ramification points of $V$ satisfies

430:       \be

431:       \wgt_V=g(g+i)^2.\label{eq:14}

432:       \ee

433: \end{enumerate}

434: \end{claim}

435: 

436: \proof From the Riemann--Roch theorem, for

437: each $j=0,\dots,i$,

438:       $$\dim V(-jP)=g+i-j.$$

439: In particular, comparing dimensions, we get

440:       $$V(-iP)=V(-(i+1)P)=H^0(C,\w_C).$$

441: Hence, the order sequence of $V$ at $P$ is

442:       $$0,\, 1,\, \dots,\, i-1,\,

443:       (i+1)+\epsilon_0,\, (i+1)+\epsilon_1,\,

444:       \dots,\, (i+1)+\epsilon_{g-1},$$

445: where $\epsilon_0,\,\epsilon_1,\,\dots,\,\epsilon_{g-1}$ 

446: is the Weierstrass order sequence at $P$. Thus

447:       $$\wgt_V(P)=\sum_{k=0}^{g-1}(i+1+\epsilon_k-i-k)

448:       =g+\wgt(P).$$

449: 

450: The second statement is a direct application of

451: the Brill--Segre formula, using that the rank of $V$

452: is $g+i-1$ and its degree is $2g-2+(i+1)$.

453: \qed

454: 

455: \section{The general curve}\label{sec3}

456: 

457: \begin{claim}\label{WiPgen}{\bf Proposition.}

458: Fix an integer $i_0\geq 0$.

459: Let $C$ be a general smooth curve of genus

460: $g\geq 1$. Then the following two statements hold for

461: each nonnegative integer $i\leq i_0$:

462: \begin{enumerate}

463: \item For a general point $P$ of $C$, the linear system

464: $V_C(i,P)$ ramifies at $P$ with weight $g$, and has

465: otherwise at most simple ramification points.

466: \item For any two points $P$ and $R$ of $C$,

467:     \begin{equation}\label{WP-R1}

468:       h^0(C,\w_C((i+1)P-(g+i)R))\leq 1

469:     \end{equation}

470: \vskip-0.8cm

471:     \begin{equation}\label{WP-R2}

472:       h^0(C,\w_C((i+1)P-(g+i+2)R)=0.

473:     \end{equation}

474: \end{enumerate}

475: \end{claim}

476: 

477: \proof Let us first observe that the property required

478: of $C$ is open. Indeed, let $f\: X\to S$ be any family of

479: smooth curves, that is, a projective, smooth map with

480: connected fibers of dimension 1. Consider the fibered

481: product $X^{(2)}:=X\x_S X$ of two copies of $f$, and

482: denote by $p_1$ and $p_2$ the projection maps. Denote by

483: $\Delta$ the diagonal subscheme of $X^{(2)}$. Let

484: $\w_f$ denote the relative canonical bundle of $f$.

485: Then $\w:=p_2^*\w_f$ is the relative canonical bundle

486: of $p_1$. Let

487:         $${\cal V}:=p_{1*}(\w((i+1)\Delta)).$$

488: A fiberwise analysis shows that 

489: $\cal V$ is a bundle of rank $g+i$

490: with formation commuting with base change. 

491: For each integer

492: $\ell\geq 0$, denote by $\cal J^\ell$ the 

493: bundle 

494: of rank $\ell+1$

495: of $p_1$-relative

496: jets of $\w((i+1)\Delta)$ truncated in 

497: order $\ell$, and denote by

498: $\psi_\ell\: p_1^*{\cal V}\to{\cal J}^\ell$ 

499: the map locally

500: obtained by differentiating the sections 

501: of $\w((i+1)\Delta)$ up

502: to order $\ell$ along the fibers of $p_1$. 

503: Let $W_{i,1}$ (resp.

504: $W_{i,2}$) be the closed subset of $X^{(2)}$ 

505: where $\psi_{g+i-1}$

506: (resp. $\psi_{g+i+1}$) has rank at most 

507: $g+i-2$ (resp. $g+i-1$).

508: Also, let $W_i$ be the closed subset 

509: of $X^{(2)}$ where

510: $\psi_{g+i-1}$ has rank at most 

511: $g+i-1$.

512: By

513: Proposition~\ref{prop5}, $W_i$ 

514: contains $\Delta$ with multiplicity

515: $g$. Let $W'_i:=W_i-g\Delta$ and 

516: $Z_i:=\Delta\cap W'_i$. Let

517: $W''_i\subset W'_i$ be the ramification scheme of the map

518: $p_1|_{W'_i}$. Let $U_i\subseteq S$ be the intersection of

519: $S-f(p_1(W_{i,1}\cup W_{i,2}))$ with $f(X-p_1(W''_i\cup Z_i))$.

520: Since $p_1$ is proper, and $f$ is both proper and open, $U_i$ is

521: an open subscheme of $S$. Let $U:=U_0\cap\cdots\cap U_{i_0}$. The

522: formation of $U$ commutes with base change. Thus a fiberwise

523: analysis reveals that $U$ consists of the set of points $s\in S$

524: such that the proposition holds for $C:=X(s)$.

525: 

526: Now, keeping in mind the existence of a versal family of

527: smooth curves, it is enough to exhibit a single curve $C$

528: for which the statement holds.

529: We will actually show a somewhat 

530: stronger existence result:

531: 

532: \begin{claim}\label{WCQ}{\bf Lemma.}

533: Fix nonnegative integers $i_0$ and $j_0$.

534: Let $g$ be a positive integer.

535: Then there is a smooth

536: pointed curve $(C,Q)$ of genus $g$ for which

537: the following three statements hold for

538: each nonnegative integers $i\leq i_0$ and $j\leq j_0$:

539: \begin{enumerate}

540: \item The linear system $V_C(j,Q)$ ramifies at

541: $Q$ with weight g, and has otherwise at most simple

542: ramification points.

543: \item For each $P\in C$ distinct from $Q$, either $Q$

544: is an ordinary point or a simple ramification point of

545: $V_C(i,P)$.

546: \item For each $P\in C$ distinct from $Q$, the linear

547: system $V$ of sections of $\w_C((i+1)P+(j+1)Q)$ given by

548:         $$V:=H^0(\w_C((i+1)P))+H^0(\w_C((j+1)Q))$$

549: satisfies

550:         $$\dim V(-(g+i+j)R)\leq 1\quad{\rm and}\quad

551:         V(-(g+i+j+2)R)=0$$

552: for each $R\in C$ distinct from $P$ and $Q$.

553: \end{enumerate}

554: \end{claim}

555: 

556: We will first see how the lemma implies the

557: proposition. Set $j_0=i_0$, and consider the pointed curve

558: $(C,Q)$ given by the lemma. Then the two statements of

559: Proposition~\ref{WiPgen} hold for $C$. Indeed,

560: the first statement holds for $P=Q$, whence

561: for $P$ in a neighborhood of $Q$, that is,

562: for a general $P$.

563: 

564: As for the second statement, first notice that 

565: (\ref{WP-R1}) and (\ref{WP-R2}) 

566: hold for $P=Q$ and every $R\in C$, a consequence of

567: the first statement of the lemma for $j:=i$. They hold

568: as well for $R=Q$ and every $P\in C$ distinct from $Q$, a

569: consequence of the second statement of the lemma. 

570: Furthermore, they hold for $R=P$ and any $P\in C$. 

571: Indeed, the first statement of the lemma for $j:=0$ 

572: implies that the canonical linear system

573: has at most simple ramification points. Thus

574: $h^0(\w_C((1-g)P))\leq 1$ and $h^0(\w_C(-(g+1)P))=0$.

575: 

576: Finally, fix a point $P\in C$ distinct from $Q$ and a 

577: point $R\in C$ distinct from $P$ and $Q$. For $j:=0$, 

578: the linear system $V$ defined in the lemma is the 

579: system of sections of $\w_C((i+1)P+Q)$ that are zero 

580: on $Q$. Since $R\neq Q$, the third statement of the

581: lemma yields (\ref{WP-R1}) and (\ref{WP-R2}).

582: 

583: It is thus enough to prove the lemma, 

584: what we do below.\qed

585: 

586: \vskip0.2cm

587: 

588: \proof (Lemma \ref{WCQ})

589: We will do induction on $g$.

590: The initial step is taken care of below.

591: 

592: Let $C$ be any elliptic curve and 

593: $Q\in C$ any point. Then the ramification points

594: of the complete linear system of sections of 

595: $\w_C((j+1)Q)$

596: are simple. (These are the $(j+1)^2$ points $R$

597: for which $Q-R$ is $(j+1)$-torsion, what includes $Q$.)

598: In fact, it follows from the Riemann--Roch

599: theorem that every complete linear system has only simple

600: ramification points. Thus Statements~1 and 2 of the lemma

601: hold. Now, given $P\in C$ distinct from $Q$, since the

602: vector subspace $V$ of $H^0(\w_C((i+1)P+(j+1)Q))$ defined

603: in  Statement~3 has codimension 1, the order sequence of

604: $V$ at a point $R$ is obtained either from

605:         $$0,\,1,\,\dots,\,g+i+j-1,\,g+i+j\quad{\rm or}\quad

606:         0,\,1,\,\dots,\,g+i+j-1,\,g+i+j+1$$

607: by removing an order. In any case, there is at most one order of

608: $V$ at $R$ above $g+i+j-1$, that is $\dim V(-(g+i+j)R)\leq 1$, and

609: all orders are at most $g+i+j+1$, that is $V(-(g+i+j+2)R)=0$.

610: 

611: Assume from now on that $g>1$, and that the claim holds for

612: smaller genera and any integers $i_0$ and $j_0$. We will

613: employ a degeneration technique in order to apply the

614: induction hypothesis.

615: 

616: Let $(Y,A)$ and $(Z,B)$ be nonrational smooth pointed curves

617: of genera $g_Y$ and $g_Z$, with $g_Y+g_Z=g$. From the

618: induction hypothesis, we may assume that the statements

619: of the lemma hold for $(C,Q)$ replaced by $(Y,A)$ and

620: all nonnegative integers $i\leq i_0$ and

621: $j\leq g_Z+i_0+j_0+1$, and for

622: $(C,Q)$ replaced by $(Z,B)$ and all

623: nonnegative integers $i\leq i_0$ and $j\leq g_Y+i_0+j_0+1$.

624: 

625: Let $C_0$ be the curve of compact type 

626: that is the union of

627: $Y$, of $Z$, and of a chain of rational curves

628: $E_1,\dots,E_{n-1}$ connecting $A$ to 

629: $B$, where $n\geq 2$.

630: Our convention is that 

631: $E_1$ contains $A$ and $E_{n-1}$ contains $B$.

632: Let $v$ be any integer such that $0<v<n$, and let

633: $Q_0$ be any point of $E_v$ that is not a node of $C_0$.

634: 

635: Let $S:={\rm Spec}(\CC[[t]])$, and denote its special point

636: by 0 and generic point by $\eta$. Since there are no

637: obstructions to deforming pointed nodal curves, there

638: are a projective, flat map $f\: X\to S$ and a section

639: $\lambda\: S\to X$ of $f$ such that

640: $(X(0),\lambda(0))=(C_0,Q_0)$ and

641: $(X(\eta),\lambda(\eta))$ is a smooth pointed curve over

642: the field of formal Laurent series $\CC[[t]][1/t]$.

643: 

644: Let $C$ be the base extension of $X(\eta)$ to the algebraic

645: closure of $\CC[[t]][1/t]$. Set $Q:=\lambda(\eta)$. It is

646: enough to see that the statements of the lemma hold for

647: $(C,Q)$. Indeed, the argument is quite standard, and is

648: summarized below. Though the pointed curve $(C,Q)$ is not

649: defined over $\CC$, it is defined over a finitely generated

650: extension $L$ of $\QQ$. If the statements of the lemma hold

651: for $(C,Q)$, they also hold for the base extension of

652: $(C,Q)$ over any algebraically closed field containing $L$.

653: But, since $\CC$ has many transcendentals

654: over $\QQ$, there is an algebraically closed field

655: containing $L$ which is isomorphic to $\CC$. 

656: So, if the statements of the lemma hold for $(C,Q)$, 

657: they hold as well for some pointed curve over $\CC$.

658: 

659: Now, any finite set of points of $C$ is 

660: defined over a finite field extension of $\CC[[t]][1/t]$.

661: Replacing $S$ by its normalization in this field extension,

662: we may assume that these are rational points of $X(\eta)$,

663: and thus that there are sections of $f$ intersecting

664: $X(\eta)$ at them. By making a further base extension, if

665: necessary, and a sequence of blowups at the singular

666: points of the special fiber, we may assume that the total

667: space $X$ is regular, and that these sections factor through

668: the smooth locus of $f$. The compensation for this

669: is a change of the special fiber. However, the special fiber

670: will have the same specification as the $C_0$ we described

671: above. Thus, no confusion will ensue if we keep calling

672: by $C_0$ this new fiber. Also, the section $\lambda$ can be

673: extended to a section of this new family.

674: 

675: Now, let $P$ and $R$ be points of $C$ with $P$ distinct from

676: $Q$ and $R$ distinct from $P$ and $Q$. As we mentioned

677: above, we may assume there are sections $\gamma\:S\to X$ and

678: $\rho\:S\to X$ through the smooth locus of $f$ such that

679: $\gamma(\eta)=P$ and $\rho(\eta)=R$. Set $P_0:=\gamma(0)$

680: and $R_0:=\rho(0)$. Let $\Gamma$ and $\Lambda$ be the images

681: of $\gamma$ and $\lambda$, respectively.

682: 

683: Fix nonnegative integers $i\leq i_0$ and $j\leq j_0$.

684: Let $\w$ be the relative dualizing bundle of $f\: X\to S$.

685: Let $V_\eta$ be the linear system of sections of the line

686: bundle $\w(\eta)((i+1)P+(j+1)Q)$ given by

687:        $$V_\eta:=H^0(\w(\eta)((i+1)P))+

688:        H^0(\w(\eta)((j+1)Q)).$$

689: Assume that $R$ is a ramification point of $V_\eta$. To

690: prove the statements of the lemma hold for $(C,Q)$, it is

691: enough to prove the following three statements:

692: \begin{enumerate}

693:     \item For $i=0$, the system $V_\eta$ ramifies at $Q$ with weight

694:     $g$, and $R$ is a simple ramification point of $V_\eta$.

695:     \item For $j=0$, the point $Q$ is a ramification point of

696:     $V_\eta$ of weight $g+i$ or $g+i+1$.

697:     \item $\dim V_\eta(-(g+i+j)R)\leq 1$ and

698:     $V_\eta(-(g+i+j+2)R)=0$.

699: \end{enumerate}

700: 

701: We will employ techniques of limit linear series, 

702: from

703: \cite{MathCont}, to show the

704: above three statements. 

705: There are two cases to consider:

706: 

707: \vskip0.2cm

708: 

709: {\noindent\emph{Case 1:} Assume that $P_0\in E_u$

710: for some $u$.}

711: 

712: \vskip0.1cm

713: 

714: Since $C_0$ is of compact type, there is an effective

715: divisor $D$ of $X$ supported on $C_0$ such that, letting

716:        $${\cal L}:=\w((i+1)\Gamma+(j+1)\Lambda+D),$$

717: we have ${\cal L}|_{E_m}\cong{\cal O}_{E_m}$ for each

718: $m=1,\dots,n-1$,

719:        $${\cal L}|_Z\cong\w_Z((g_Y+i+j+3)B)\quad

720:        {\rm and}\quad

721:        {\cal L}|_Y\cong\w_Y((1-g_Y)A).$$

722: Since, from the induction hypothesis, $A$ is a ramification

723: point of weight $g_Y$ of the complete linear system of

724: sections of $\w_Y(A)$, the point $A$ is not a Weierstrass

725: point of $Y$. Then $V:=H^0(X,{\cal L})\cap V_\eta$ restricts

726: to a linear system $V_Z$ of dimension $g+i+j$ of sections of

727: $\w_Z((g_Y+i+j+3)B)$. Also from the induction hypothesis,

728: $B$ is not a Weierstrass point of $Z$. So the order

729: sequence of $B$ in the complete linear system of sections of

730: $\w_Z((g_Y+i+j+3)B)$ is

731:        $$0,\,1,\,\dots,\,g_Y+i+j+1,\,g_Y+i+j+3,\,\dots,\,

732:        g+i+j+2.$$

733: As a consequence, the weight $w_B$ of $B$ as a ramification

734: point of the linear system $V_Z$ satisfies

735:        \be\label{wB1}

736:        w_B\leq 2(g_Y+i+j)+3g_Z,

737:        \ee

738: with equality if and only if $V_Z=H^0(\w_Z((g_Y+i+j+1)B))$.

739: 

740: Analogously, choosing an appropriate $D$, we obtain a linear

741: system $V_Y$ of dimension $g+i+j$ of sections of

742: $\w_Y((g_Z+i+j+3)A)$, and the weight $w_A$ of $A$

743: as a ramification point of $V_Y$ satisfies

744:        \be\label{wA1}

745:        w_A\leq 2(g_Z+i+j)+3g_Y,

746:        \ee

747: with equality if and only if $V_Y=H^0(\w_Y((g_Z+i+j+1)A))$.

748: 

749: Let $r:=g+i+j-1$. Using the Pl\"ucker formula, the number

750: $N$ of ramification points of $V_Y$ and $V_Z$ on

751: $(Y-A)\cup(Z-B)$, counted with their respective weights,

752: satisfies

753:        \begin{eqnarray*}

754:        N&=&(r+1)\Big((2g_Z+g_Y+i+j+1)+r(g_Z-1)\Big)-w_B\\

755:        &+&(r+1)\Big((2g_Y+g_Z+i+j+1)+r(g_Y-1)\Big)-w_A\\

756:        &=&N'+5g+4(i+j)-w_A-w_B,

757:        \end{eqnarray*}

758: where

759:        $$N':=(r+1)\Big((2g+i+j)+r(g-1)\Big)-2g-i-j.$$

760: 

761: Now, from the theory of limit linear series, each one of the

762: ramification points of $V_Y$ or $V_Z$ on $(Y-A)\cup(Z-B)$

763: is a limit of ramification points of $V_\eta$, and its

764: weight as a ramification point is the sum of the weights of

765: the ramification points of $V_\eta$ converging to it.

766: Besides those, since $P$ and $Q$ are ramification points

767: of $V_\eta$ with weights at least $g+j$ and $g+i$,

768: respectively, the points $P_0$ and $Q_0$ appear

769: as limits of ramification points of $V_\eta$ with weights

770: summing up to at least $2g+i+j$. Thus, from the Pl\"ucker

771: formula, at most $N'$ ramification points of $V_\eta$,

772: counted with their weights, converge to $(Y-A)\cup(Z-B)$. So

773:        $$5g+4(i+j)-w_A-w_B\leq 0.$$

774: 

775: However, Inequalities (\ref{wB1}) and (\ref{wA1}) for $w_B$

776: and $w_A$ yield the opposite inequality:

777:        $$5g+4(i+j)-w_A-w_B\geq 0.$$

778: Thus, equalities hold, and hence

779:        $$V_Y=H^0(\w_Y((g_Z+i+j+1)A))\quad{\rm and}\quad

780:        V_Z=H^0(\w_Z((g_Y+i+j+1)B)).$$

781: In addition, $P$ and $Q$ are ramification points of

782: $V_\eta$ of weights $g+j$ and $g+i$, respectively, and

783: all the other ramification points of $V_\eta$ converge

784: to $(Y-A)\cup(Z-B)$. In particular, Statement 2 

785: and the first part of Statement~1 are shown.

786: 

787: Now, since $R$ is a ramification point of $V_\eta$, and

788: $R$ is distinct from $P$ and $Q$, we have

789: $R_0\in(Y-A)\cup(Z-B)$. So $R_0$ is a

790: ramification point of either $V_Y$ or $V_Z$. From the

791: induction hypothesis, the complete linear systems of

792: sections of $\w_Y((g_Z+i+j+1)A)$ and $\w_Z((g_Y+i+j+1)B)$

793: have at most simple ramification points, other than $A$ or

794: $B$. Thus $R$ is the unique ramification point of $V_\eta$

795: converging to $R_0$ and its weight is 1.

796: So the remainder of Statement 1 is shown.

797: 

798: As for Statement 3, assume, 

799: without loss of generality, that

800: $R_0\in Z$. Set

801: $n:=\dim V_\eta(-(g+i+j)R)$, and let

802: $\sigma_1,\dots,\sigma_n$ form a $\CC[[t]]$-basis of

803: $V\cap V_\eta(-(g+i+j)R)$. Their restrictions to $Z$ are

804: sections of $V_Z$ vanishing with multiplicity at least

805: $g+i+j$ on $R_0$. Assume, by contradiction, that

806: $n\geq 2$. Since $R_0$ is a simple ramification point of

807: $V_Z$, the sections $\sigma_1|_Z,\dots,\sigma_n|_Z$

808: are linearly dependent. Thus, there is

809: a nonzero $n$-tuple $(c_1,\dots,c_n)\in\CC^n$ such that

810: $c_1\sigma_1+\cdots+c_n\sigma_n$ vanishes on $Z$, and hence

811: on the whole $C_0$. Thus

812:        \be\label{sigma}

813:        c_1\sigma_1+\cdots+c_n\sigma_n=t\sigma

814:        \ee

815: for some $\sigma\in H^0(X,{\cal L})$. Also $\sigma\in

816: V_\eta(-(g+i+j)R)$, and hence $\sigma$ is a $\CC[[t]]$-linear

817: combination of $\sigma_1,\dots,\sigma_n$. Plugging this linear

818: combination in (\ref{sigma}) we obtain a nontrivial

819: $\CC[[t]]$-linear relation among the sections $\sigma_i$, a

820: contradiction. Thus $n\leq 1$. A similar analysis, using that

821: $V_Z(-(g+i+j+2)R_0)=0$, shows that $V_\eta(-(g+i+2)R)=0$,

822: finishing the proof of Statement 3.

823: 

824: \vskip0.2cm

825: 

826: {\noindent\emph{Case 2:} Assume $P_0$ belongs to either $Y$

827: or $Z$.}

828: 

829: \vskip0.1cm

830: 

831: Without loss of generality, we may assume that $P_0\in Z$.

832: Again, since $C_0$ is of compact type, there is an effective

833: divisor $D$ of $X$ supported on $C_0$ such that, letting

834:        $${\cal L}:=\w((i+1)\Gamma+(j+1)\Lambda+D),$$

835: we have ${\cal L}|_{E_m}\cong{\cal O}_{E_m}$ for each

836: $m=1,\dots,n-1$,

837:        $${\cal L}|_Z\cong\w_Z((i+1)P_0+(g_Y+j+2)B)

838:        \quad{\rm and}\quad

839:        {\cal L}|_Y\cong\w_Y((1-g_Y)A).$$

840: As before, $V:=H^0(X,{\cal L})\cap V_\eta$ restricts to a

841: linear system $V_Z$ of dimension $g+i+j$ of sections of

842: $\w_Z((i+1)P_0+(g_Y+j+2)B)$.

843: 

844: Now,

845:        $$V\supseteq H^0(X,\w((i+1)\Gamma))+

846:        H^0(X,\w((j+1)\Lambda+D)).$$

847: Reasoning as in Case 1, we can show that

848: $H^0(X,\w((j+1)\Lambda+D))$ restricts to

849: $H^0(\w_Z((g_Y+j+1)B))$.

850: On the other hand, the exact sequence

851:        $$0\to H^0(\w_Z((i+1)P_0))\to

852:        H^0(\w((i+1)\Gamma)|_{C_0})

853:        \to H^0(\w_Y(A))$$

854: shows that $h^0(\w((i+1)\Gamma)|_{C_0})=g+i$, and hence

855: that $H^0(X,\w((i+1)\Gamma))$ restricts to a vector

856: subspace of $H^0(\w_Z((i+1)P_0+B))$ containing

857: the subspace $H^0(\w_Z((i+1)P_0))$. Thus

858:        $$V_Z\supseteq H^0(\w_Z((i+1)P_0))+

859:        H^0(\w_Z((g_Y+j+1)B)),$$

860: and a dimension count shows that equality holds.

861: 

862: The weight $w_B$ of $B$ as a ramification point of $V_Z$

863: depends on its weight as a ramification point of

864: $V_Z(i,P_0)$. Now, from the induction hypothesis,

865: $B$ is either an ordinary point or a simple ramification

866: point of $V_Z(i,P_0)$.

867: Hence, the order sequence at $B$ of the linear system

868: $V_Z$ is either

869:        $$1,\,2,\,\dots,\,g_Y+j,\,g_Y+j+2,\,g_Y+j+3,\,

870:        \dots,\,g+i+j,\,g+i+j+1$$

871: or

872:        $$1,\,2,\,\dots,\,g_Y+j,\,g_Y+j+2,\,g_Y+j+3,\,

873:        \dots,\,g+i+j,\,g+i+j+2.$$

874: At any rate,

875:        \begin{equation}\label{wB2}

876:        w_B\leq g_Y+j+2(g_Z+i)+1.

877:        \end{equation}

878: Notice that, if $i=0$, then $B$ is an ordinary point of

879: $V_Z(0,P_0)$, as it is an ordinary point of $Z$,

880: and thus Inequality (\ref{wB2}) is strict.

881: 

882: On the other hand, let $D'$ be an effective divisor of $X$

883: supported on $C_0$ such that, letting

884:        $${\cal M}:=\w((i+1)\Gamma+(j+1)\Lambda+D'),$$

885: we have ${\cal M}|_{E_m}\cong{\cal O}_{E_m}$ for each

886: $m=1,\dots,n-1$,

887:        $${\cal M}|_Y\cong\w_Y((g_Z+i+j+3)A)

888:        \quad{\rm and}\quad

889:        {\cal M}|_Z\cong\w_Z((i+1)P_0-(g_Z+i)B).$$

890: Since, as mentioned above, $B$ is either an ordinary point

891: or a simple ramification point of $V_Z(i,P_0)$, we have that

892: $H^0(X,{\cal M})\cap V_\eta$ restricts to a linear

893: system $V_Y$ of dimension $g+i+j$ of sections of

894: $\w_Y((g_Z+i+j+3)A)$.

895: 

896: Since $A$ is not a Weierstrass point of $Y$, the sequence of

897: orders at $A$ of the complete linear system

898: of sections of $\w_Y((g_Z+i+j+3)A)$ is

899:        $$0,\,1,\,\dots,\,g_Z+i+j+1,\,g_Z+i+j+3,\,

900:        g_Z+i+j+4,\,\dots,\,g+i+j+2.$$

901: Since $V_Y$ has codimension 2 in $H^0(\w_Y((g_Z+i+j+3)A))$,

902: the weight $w_A$ of $V_Y$ at $A$ satisfies

903:        \begin{equation}\label{wA2}

904:        w_A\leq 2(g_Z+i+j)+3g_Y,

905:        \end{equation}

906: with equality if and only if $V_Y=H^0(\w_Y((g_Z+i+j+1)A))$.

907: 

908: As in Case 1, using the Pl\"ucker formula, the number $N$ of

909: ramification points of $V_Y$ and $V_Z$ on $(Y-A)\cup(Z-B)$,

910: counted with their respective weights, satisfies

911:        $$N=N'+4g+4i+3j-w_A-w_B,$$

912: where

913:        $$N':=(g+i+j)(2g+i+j)+(g+i+j)(g+i+j-1)(g-1)-g-i.$$

914: As in Case 1, since $Q$ is a ramification point of $V_\eta$

915: with weight at least $g+i$, there are at most $N'$

916: ramification points of $V_\eta$, counted with their

917: respective weights, converging to $(Y-A)\cup(Z-B)$. So

918:        $$4g+4i+3j-w_A-w_B\leq 0.$$

919: On the other hand, Inequalities (\ref{wB2}) and (\ref{wA2})

920: yield

921:        $$w_A+w_B\leq 4g+4i+3j+1.$$

922: In particular, $w_A\geq 2(g_Z+i+j)+3g_Y-1$, whence

923:        $$V_Y\subset H^0(\w_Y((g_Z+i+j+2)A)).$$

924: Also, $Q$ has weight $g+i$ or $g+i+1$ in $V_\eta$.

925: Thus Statement 2 is shown.

926: Furthermore, 

927: if $i=0$ we have $w_A+w_B=4g+4i+3j$. In this case,

928: $V_Y=H^0(\w_Y((g_Z+i+j+1)A))$ and 

929: $Q$ has weight $g+i$ in $V_\eta$, showing the first 

930: part of Statement 1.

931: 

932: If $Q$ has weight $g+i+1$ in $V_\eta$, all other ramification

933: points converge to $(Y-A)\cup(Z-B)$. If $Q$ has weight $g+i$,

934: there is at most one ramification point of 

935: $V_\eta$, other than $Q$,

936: converging outside $(Y-A)\cup(Z-B)$, and that point is 

937: simple. If $R$ is that point, then

938: $\dim V_\eta(-(g+i+j)R)\leq 1$ and 

939: $V_\eta(-(g+i+j+2)R)=0$ because

940: of the simplicity of $R$.

941: 

942: Assume now that $R_0\in (Y-A)\cup(Z-B)$. Let us first

943: consider the case $R_0\in Y-A$. In this case, since,

944: from the induction hypothesis,

945: the complete linear system of sections of

946: $\w_Y((g_Z+i+j+2)A)$ has at most simple

947: ramification points, other than $A$, we have

948:        \begin{eqnarray*}

949:        h^0(\w_Y((g_Z+i+j+2)A-(g+i+j)R_0)&\leq& 1,\\

950:        h^0(\w_Y((g_Z+i+j+2)A-(g+i+j+2)R_0)&=& 0.

951:        \end{eqnarray*}

952: Thus $\dim V_Y(-(g+i+j)R_0)\leq 1$ and $V_Y(-(g+i+j+2)R_0)=0$ as

953: well. It follows, as in Case 1, that 

954: $$

955: \dim V_\eta(-(g+i+j)R)\leq 1\quad {\rm and}\quad 

956: V_\eta(-(g+i+j+2)R)=0.

957: $$ 

958: Furthermore,

959: if $i=0$, since in this case $V_Y=H^0(\w_Y((g_Z+i+j+1)A))$, all

960: the ramification points of $V_Y$ distinct from $A$ are simple.

961: Thus $R_0$ is simple in $V_Y$, and hence $R$ is simple in

962: $V_\eta$.

963: 

964: Assume now that $R_0\in Z-B$. There are two cases to consider.

965: First, assume $R_0=P_0$. Since, by induction hypothesis, the

966: complete linear system of sections of $\w_Z((g_Y+j+1)B)$ has at

967: most simple ramification points other than $B$, the weight of

968: $P_0$ as a ramification point of $V_Z$ is either $g+j$ or $g+j+1$.

969: Since $P$ has at least weight $g+j$ in $V_\eta$, and $R\neq P$,

970: the latter must hold, and $R$ must be a simple ramification point

971: of $V_\eta$. In particular, $\dim V_\eta(-(g+i+j)R)\leq 1$ and

972: $V_\eta(-(g+i+j+2)R)=0$.

973: 

974: Finally, assume $R_0\neq P_0$. Then 

975: $$

976: \dim V_Z(-(g+i+j)R_0)\leq 1\quad {\rm and} \quad 

977: V_Z(-(g+i+j+2)R_0)=0

978: $$

979: from the induction hypothesis, and

980: hence $\dim V_\eta(-(g+i+j)R)\leq 1$ and $V_\eta(-(g+i+j+2)R)=0$.

981: Thus Statement~3 is shown. Also, if $i=0$, then

982: $V_Z=H^0(\w_Z((g_Y+j+1)B))$, and, since $R_0\neq P_0$, the weight

983: of $R_0$ in $V_Z$ is equal to its weight in the complete linear

984: system of sections of $\w_Z((g_Y+j+1)B)$. By induction hypothesis,

985: this weight is one, and thus $R$ is a simple ramification point of

986: $V_\eta$. So Statement~1 is shown.\qed

987: 

988: \begin{claim}\label{ws}{\bf Corollary.} 

989: If $C$ is a general

990: smooth curve of genus $g\geq 1$, then all its

991: Weierstrass points are simple.

992: \end{claim}

993: 

994: \proof 

995: Apply Statement 1 of

996: Proposition~\ref{WiPgen} for $i_0:=0$ and $i:=0$. 

997: \qed

998: 

999: \begin{claim}\label{arbD}{\bf Proposition.} 

1000: Fix an integer $i_0\geq 0$. 

1001: Let $C$ be a general smooth curve of genus $g\geq 1$. 

1002: Then for any two distinct points $P$ and $R$ of $C$, 

1003: and any nonnegative integer $i\leq i_0$,

1004:        $$h^0(C,\omega_C((i+1)P-(g+i-2)R))=2.$$

1005: \end{claim}

1006: 

1007: \proof A line bundle of degree 2 on an elliptic curve has 

1008: (at most) 2 linearly independent sections. Thus we 

1009: may assume $g\geq 2$. Also, for $i=0$, 

1010:        $$h^0(C,\omega_C((i+1)P-(g+i-2)R)=

1011:          h^0(C,\omega_C(-(g-2)R))=2,$$

1012: since $R$ is at most a simple Weierstrass point of 

1013: $C$, a consequence of 

1014: Corollary~\ref{ws}. 

1015: So we need 

1016: only show the stated equality for integers $i>0$.

1017: 

1018: For each integer $j\geq 2$ (resp. $j\geq 1$), let 

1019: $M_j$ be the moduli space of smooth curves (resp. 

1020: let $M_{j,1}$ be the moduli space of smooth pointed 

1021: curves) of genus $j$. 

1022: Let $\ol M_j$ and $\ol M_{j,1}$ denote their 

1023: respective compactifications by stable (resp. stable, pointed) curves. 

1024: For each 

1025: positive integer $i\leq i_0$, let 

1026: $D^{(i)}\subseteq M_{g+i}$ be the subset 

1027: parameterizing curves admitting a covering 

1028: of degree at most $g+i-2$ of the projective line 

1029: totally ramified at a point. 

1030: By \cite{A}, Thm. 3.11, p. 333, the subvariety 

1031: $D^{(i)}$ is irreducible of codimension 2. Let 

1032: $\ol D^{(i)}\subset\ol M_{g+i}$ be the closure of 

1033: $D^{(i)}$. 

1034: 

1035: Let $\mu_i\:M_{g,1}\times M_{i,1}\to 

1036: \ol M_{g+i}$ be the natural map, associating to a 

1037: pair of smooth pointed curves the stable uninodal curve 

1038: which is the union of these curves identified at the 

1039: marked points. Let $E^{(i)}:=\mu_i^{-1}(\ol D^{(i)})$. 

1040: Let 

1041: $\rho_i\:E^{(i)}\to M_g$ be the natural map, 

1042: forgetting the second 

1043: pointed curve and the marked point on the first curve. 

1044: Since $C$ is general, we may assume that, 

1045: for each $i=1,\dots,i_0$, the 

1046: curve $C$ is parameterized 

1047: by a point of $M_g$ over which the fiber of 

1048: $\rho_i$ has minimum dimension. We claim this 

1049: dimension is at most $3i-3$, whence less than 

1050: $\dim M_{i,1}$. Indeed, if the dimension were 

1051: larger, then $E^{(i)}$ would have codimension at 

1052: most 1 

1053: in $M_{g,1}\times M_{i,1}$, and hence would dominate 

1054: $\ol D^{(i)}$ under $\mu_i$. So $\ol D^{(i)}$ would 

1055: be contained in the boundary $\ol M_{g+i}-M_{g+i}$, 

1056: an absurd. From the claim, for each $i=1,\dots,i_0$, 

1057: the general smooth pointed curve $(Y_i,B_i)$ of genus 

1058: $i$ is such that,

1059: for any $P\in C$, the 

1060: pair of pointed curves $((C,P),(Y_i,B_i))$ is 

1061: not parameterized by $E^{(i)}$. Consequently, the 

1062: stable uninodal curve $X_i$, union of $C$ and $Y_i$ 

1063: with $P$ and $B_i$ identified, is 

1064: parameterized by a point of 

1065: $\ol M_{g+i}$

1066: off $\ol D^{(i)}$, 

1067: for each $i=1,\dots,i_0$. 

1068: 

1069: Suppose, by contradiction, that for certain 

1070: distinct points $P$ and $Q$ of $C$, and a certain 

1071: positive integer $i\leq i_0$, we have

1072:        $$h^0(C,\w_C((i+1)P-(g+i-2)Q)\geq 3.$$

1073: Put $g':=g+i$. Since, by Riemann--Roch, 

1074: $h^0(C,\w_C((i+1)P-iQ))=g$, 

1075: there is an integer $j$ with $2\leq j<g$ 

1076: such that

1077:        $$h^0(C,\w_C((i+1)P-(g'-j)Q)=

1078:          h^0(C,\w_C((i+1)P-(g'-j-1)Q)=j+1.$$

1079: Again by Riemann--Roch, 

1080:        \begin{equation}

1081: 	 h^0(C,{\cal O}_C((g'-j)Q-(i+1)P))>

1082: 	 h^0(C,{\cal O}_C((g'-j-1)Q-(i+1)P)).

1083:        \end{equation}

1084: Thus, there is a map 

1085: $\phi\:C\lra \PP^1$ of degree $g'-j$ such that 

1086: $\phi^*(0)=(g'-j)Q$ and $\phi^*(\infty)\geq(i+1)P$. 

1087: Let $i'$ be the integer such that $i'+1$ is the 

1088: multiplicity of $P$ in 

1089: $\phi^*(\infty)$. 

1090: Then $i'\geq i$.

1091: 

1092: Set $Y:=Y_i$ and $B:=B_i$. 

1093: Since $B$ is general, $B$ is not a Weierstrass 

1094: point of $Y$. Thus, since $i'\geq i$, we have 

1095: $h^0(Y,{\cal O}_Y(i'B))<h^0(Y,{\cal O}_Y((i'+1)B))$.  

1096: So, there is a map 

1097: $\psi\:Y\lra\PP^1$ of degree $i'+1$ such that 

1098: $\psi^*(\infty)=(i'+1)B$. 

1099: 

1100: Putting together the maps 

1101: $\phi$ and $\psi$, we may construct the 

1102: covering with source $X_i$ depicted  

1103: %<below,

1104: %>

1105: in Figure~1 below,

1106: %

1107: \begin{figure}[ht]

1108: \begin{center}

1109: \includegraphics[height=5cm, angle=0]{admm.eps}

1110: \caption{The covering.}

1111: \end{center}

1112: \end{figure}

1113: %<\begin{center}

1114: %<\includegraphics[height=5cm]{admm.pdf}

1115: %<\end{center}

1116: which can be represented by a point $[X_i]$ of the 

1117: (compactification of the) Hurwitz scheme 

1118: parameterizing (pseudo)admissible 

1119: coverings of the projective line of 

1120: degree $(g'-j)$ totally ramified at a 

1121: point; see Remark \ref{rmk1}.

1122: Since 

1123: coverings of $\PP^1$ form a dense open subscheme of 

1124: this compactification, the curve 

1125: $X_i$ is limit of smooth curves equipped with a 

1126: degree-$(g'-j)$ map to the projective line totally 

1127: ramified at a 

1128: point. Since $j\geq 2$, it follows that $[X_i]$ 

1129: lies on the boundary of $D^{(i)}$,

1130: a contradiction.\qed

1131: 

1132: \begin{claim}\label{rmk1}{\bf Remark.} {\rm The 

1133: Hurwitz scheme we used in the proof of 

1134: Proposition \ref{arbD} is 

1135: mentioned in \cite{Diaz/Exc}, Section 5. It can be 

1136: constructed following the same reasoning used in the 

1137: construction of the Hurwitz scheme of (simple) 

1138: admissible coverings, given in the proof 

1139: of \cite{HaMu}, Thm. 4, p. 58. Also, the local 

1140: descriptions of both schemes are the same, given on 

1141: \cite{HaMu}, p. 62. From this description we see that 

1142: the Hurwitz scheme is equidimensional. Now, there 

1143: is a natural forgetful map from the Hurwitz scheme 

1144: to a corresponding moduli space of pointed genus-0 

1145: curves, taking a covering to its target. 

1146: This map is finite and surjective, also by 

1147: \cite{HaMu}, Thm. 4, p.~58. Since

1148: the moduli spaces of pointed genus-$0$ curves 

1149: are irreducible (see \cite{Kn} or \cite{Ke}),

1150: it follows that each irreducible 

1151: component of the Hurwitz scheme covers the target. So 

1152: coverings of $\PP^1$ form a dense open subscheme of 

1153: the Hurwitz scheme, a fact used in the proof 

1154: of Proposition \ref{arbD}.}

1155: \end{claim}

1156: 

1157: \begin{claim}\label{rmk2}{\bf Remark.} {\rm We tried 

1158: to prove Proposition \ref{arbD} using the same induction 

1159: argument used in the proof of Lemma \ref{WCQ}. 

1160: However, we could not prove the 

1161: initial step, that is, the following statement:}

1162: {\it Let $C$ be a general elliptic 

1163: curve, $Q\in C$ a general point, and $P\in C-\{Q\}$ any 

1164: point. Let $i$ and $j$ be nonnegative integers. Then the 

1165: linear system $V$ of sections of the line bundle 

1166: $\w_C((i+1)P+(j+1)Q)$ 

1167: generated by $H^0(\w_C((i+1)P))$ and $H^0(\w_C((j+1)Q))$ 

1168: satisfies $\dim V(-(i+j-1)R)=2$ for 

1169: each $R\in C-\{P,Q\}$.}

1170: \end{claim}

1171: 

1172: \section{Weierstrass divisors}\label{sec4}

1173: 

1174: \begin{claim}\label{wronski} {\rm ({\it Wronski maps})

1175: Let $C$ be a smooth curve of genus $g$. 

1176: For each integer $j\geq 0$, consider the family of

1177: linear systems $V_C(j,P)$ for $P$ varying on $C$.

1178: More precisely, let

1179: $p_1$ and $p_2$ denote the projections of $C\times C$ onto

1180: the first and second factors, and 

1181: $\Delta\subset C\times C$ the diagonal.

1182: The relative canonical bundle of $p_1$

1183: is simply the pullback $p_2^*\w_C$ of the canonical bundle

1184: $\w_C$ of $C$. For each integer $j\geq 0$, let

1185:        $${\cal L}_j:=p_2^*\w_C((j+1)\Delta),\quad

1186:        {\cal E}_j:=p_{1*}{\cal L}_j.$$

1187: Notice that, for each point $P$ of $C$,

1188: identifying $\{P\}\times C$ with $C$

1189: in the natural way,

1190: ${\cal L}_j|_{\{P\}\times C}=\w_C((j+1)P)$.

1191: Also, as $h^0(\w_C((j+1)P))=g+j$ for

1192: every $P\in C$, the sheaf ${\cal E}_j$ is a

1193: bundle of rank

1194: $g+j$ and ${\cal E}_j|_P=H^0(\w_C((j+1)P))$.

1195: 

1196: For each integer $\ell\geq 0$ and each line bundle

1197: ${\cal M}$ on $C\times C$, let

1198: ${\cal J}^\ell_{p_1}({\cal M})$

1199: be the bundle 

1200: of rank $\ell+1$ 

1201: of $p_1$-relative jets of ${\cal M}$

1202: truncated in order $\ell$. Let

1203:        $$\rho_{j,\ell}\:p_1^*{\cal E}_j\to

1204:        {\cal J}^\ell_{p_1}({\cal L}_j)$$

1205: be the map of bundles locally obtained by 

1206: differentiating 

1207: up to order $\ell$ along the fibers of $p_1$

1208: the sections of ${\cal L}_j$. We call $\rho_{j,\ell}$ 

1209: a {\it Wronski map}.

1210: 

1211: The map $\rho_{j,g+j-1}$ is a map of bundles of the

1212: same rank. Taking determinants, we get a section $z_j$

1213: of the line bundle

1214:         $$\bigwedge^{g+j}{\cal J}^{g+j-1}_{p_1}({\cal L}_j)

1215:         \otimes\bigwedge^{g+j}p_1^*{\cal E}_j^{\vee},$$

1216: which is naturally isomorphic, using the truncation

1217: sequence of the bundles of jets, to

1218:         $$p_2^*\w_C((j+1)\Delta)^{\otimes g+j}\otimes

1219:         p_2^*\w_C^{\otimes (g+j)(g+j-1)/2}\otimes

1220:         \bigwedge^{g+j}p_1^*{\cal E}_j^{\vee},$$

1221: or more simply to

1222:         $$p_2^*\w_C^{\otimes (g+j)(g+j+1)/2}

1223:         \Big((g+j)(j+1)\Delta\Big)\otimes\bigwedge^{g+j}

1224:         p_1^*{\cal E}_j^{\vee}.$$}

1225: \end{claim}

1226: 

1227: \begin{claim}\label{jWd} 

1228: {\rm ({\it Weierstrass divisors.}) 

1229: Keep the notation used in Subsection \ref{wronski}.

1230: Let $Z_j\subseteq C\times C$ denote the zero scheme 

1231: of $z_j$. The section $z_j$ is a relative wronskian.

1232: More precisely, for each $P\in C$, on $\{P\}\times C$,

1233: identified with $C$ in the natural way,

1234: the section $z_j$ restricts to the

1235: wronskian of the linear system $V_C(j,P)$. Hence, $Z_j$

1236: consists of the pairs $(P,Q)\in C\times C$ such that

1237: $V_C(j,P)$ ramifies at $Q$. Now, since $z_j$ is nonzero,

1238: being so on each fiber, $Z_j$ is a Cartier divisor.

1239: By Proposition~\ref{prop5}, the divisor $Z_j$ intersects

1240: each fiber $\{P\}\times C$ at $(P,P)$ with multiplicity

1241: $g+\wgt(P)$, where $\wgt(P)$ is the weight of $P$ in the

1242: canonical system of $C$. Thus $Z_j$ contains $\Delta$ with

1243: multiplicity exactly $g$. Let

1244:         $$W_j:=Z_j-g\Delta.$$

1245: Then $W_j$ is, set-theoretically, the locus of pairs

1246: $(P,Q)\in C\times C$ such that either $P=Q$ and $P$ is a

1247: Weierstrass point of $C$, or $P\neq Q$ and $Q$ is a

1248: ramification point of $V_C(j,P)$. We call $W_j$ the 

1249: {\it $j$-th Weierstrass divisor of $C$.}}

1250: \end{claim}

1251: 

1252: \begin{claim}\label{prop7}{\bf Proposition.} 

1253: Let $C$ be a smooth curve of

1254: genus $g\geq 1$ and $j$ a nonnegative integer. Let 

1255: $\Delta$ be the diagonal of $C\times C$, and 

1256: $p_1$ and $p_2$ the projections of $C\times C$ 

1257: onto the indicated factors. Let 

1258: $\w_C$ be the canonical bundle of $C$, and set 

1259: $K_\ell:=c_1(p_\ell^*\w_C)$ for $\ell=1,2$. 

1260: Let $W_j\subseteq C\times C$ be 

1261: the $j$-th Weierstrass divisor of $C$. 

1262: Then its class $[W_j]$ in the Chow group of

1263: $C\times C$ satisfies

1264:         \be

1265:         [W_j]=\frac{1}{2}(g+j)(g+j+1)K_2+j(g+j+1)[\Delta]

1266:         +\frac{1}{2}j(j+1)K_1.\label{eq:17}

1267:         \ee

1268: \end{claim}

1269: 

1270: \proof Use 

1271: the notation in Subsections \ref{wronski} 

1272: and \ref{jWd}. Since $W_j=Z_j-g\Delta$, and $Z_j$ is the

1273: zero scheme of a section of the line bundle

1274:         $$p_2^*\w_C^{\otimes (g+j)(g+j+1)/2}

1275:         \Big((g+j)(j+1)\Delta\Big)\otimes\bigwedge^{g+j}

1276:         p_1^*{\cal E}_j^{\vee},$$

1277: we get

1278:         \be

1279:     [W_j]=\frac{1}{2}(g+j)(g+j+1)K_2+j(g+j+1)[\Delta]-

1280:         p_1^*c_1({\cal E}_j).\label{eq:67}

1281:         \ee

1282: To finish, we need only show that

1283:         \be

1284:         c_1({\cal E}_j)=-\frac{1}{2}j(j+1)c_1(\w_C).

1285:     \label{eq:c1Ei}

1286:     \ee

1287: 

1288: We show (\ref{eq:c1Ei}) by induction on $j$.

1289: First of all,

1290:         $${\cal E}_0=p_{1*}p_2^*\w_C=

1291:         H^0(\w_C)\otimes{\cal O}_C.$$

1292: Since ${\cal E}_0$ is free, $c_1({\cal E}_0)=0$.

1293: 

1294: Assume now that $j>0$ and

1295: $c_1({\cal E}_{j-1})=-(j(j-1)/2)c_1(\w_C)$.

1296: Consider the natural short exact sequence

1297:         $$0\to p_2^*\w_C(j\Delta)\to p_2^*\w_C((j+1)\Delta)

1298:         \to p_2^*\w_C((j+1)\Delta)|_{\Delta}\to 0.$$

1299: Since $H^1(\w_C(jP))=0$ for each $P\in C$,

1300: applying $p_{1*}$ to the sequence above,

1301: we get the exact sequence

1302:         $$0\to{\cal E}_{j-1}\to{\cal E}_j\to

1303:         p_{1*}p_2^*\w_C((j+1)\Delta)|_{\Delta}\to 0.$$

1304: Now, $p_\ell|_\Delta$ is an isomorphism for $\ell=1,2$. So

1305: $p_{1*}p_2^*\w_C|_\Delta=\w_C$. In addition,

1306: $p_{1*}{\cal O}_{C\times C}(-\Delta)|_\Delta=\w_C$. Thus

1307:         \begin{eqnarray*}

1308:     c_1({\cal E}_j)&=&c_1({\cal E}_{j-1})+

1309:         c_1(p_{1*}p_2^*\w_C((j+1)\Delta)|_{\Delta})\\

1310:     &=&-(j(j-1)/2)c_1(\w_C)+(1-(j+1))c_1(\w_C)\\

1311:     &=&-(j(j+1)/2)c_1(\w_C),

1312:     \end{eqnarray*}

1313: as claimed.\qed

1314: 

1315: \begin{claim}\label{lem}{\bf Proposition.} 

1316: Let $C$ be a general

1317: smooth curve of genus $g\geq 1$ and $j$ a nonnegative

1318: integer. Let $W_j\subseteq C\times C$ be the 

1319: $j$-th Weierstrass divisor of $C$. Then $W_j$ 

1320: is nonsingular and intersects 

1321: the diagonal $\Delta$ transversally, at the pairs 

1322: $(P,P)$ such that $P$ is a

1323: Weierstrass point of $C$.

1324: \end{claim}

1325: 

1326: \proof Let us show first that $W_j$ intersects 

1327: $\Delta$ transversally. 

1328: As pointed out in Subsection \ref{jWd}, the intersection 

1329: $W_j\cap\Delta$ is,

1330: set-theoretically, the set of pairs $(P,P)$ such that

1331: $P$ is a Weierstrass point of $C$. As $C$ is general,

1332: by Corollary~\ref{ws},

1333: all its Weierstrass points are simple, and number

1334: $g^3-g$ by Pl\"ucker Formula.

1335: Now, since the intersection $W_j\cap\Delta$ is

1336: finite, the number of points of intersection, weighted

1337: by their intersection multiplicities, is equal to 

1338: the degree

1339: of the product $[W_j][\Delta]$. Using the notation 

1340: and Formula (\ref{eq:17}) of Proposition \ref{prop7}, 

1341: and using the Formulas

1342:        \be

1343:        \int_{C\times C}K_2[\Delta]=

1344:        \int_{C\times C}K_1[\Delta]=

1345:        -\int_{C\times C}[\Delta]^2=2g-2

1346:        \label{eq:kd}

1347:        \ee

1348: and

1349:        \be

1350:        \int_{C\times C}K_1K_2=4(g-1)^2,

1351:        \label{eq:kk}

1352:        \ee

1353: we get

1354:        $$\int_{C\times C}[W_j][\Delta]=

1355:        (g+j)(g+j+1)(g-1)-2j(g+j+1)(g-1)

1356:         +j(j+1)(g-1),$$

1357: which is exactly $g^3-g$. Thus the intersection

1358: multiplicities are all one.

1359: 

1360: As a corollary of the transversal intersection, 

1361: $W_j$ is nonsingular at its points on $\Delta$. So, 

1362: let now $(P,Q)\in W_j$ for $P$ and $Q$ distinct, and 

1363: let us show that $W_j$ is nonsingular at 

1364: $(P,Q)$ as well. 

1365: 

1366: Let $J:={\rm Pic}^{g-1}(C)$, the component of the 

1367: Picard scheme of $C$ parameterizing line bundles of 

1368: degree $g-1$. Let $\Theta\subset J$ be the theta divisor, 

1369: parameterizing line bundles with nontrivial global 

1370: sections. Let

1371:          $$\mu\: C\times C \to J$$

1372: be the map taking a pair $(R,S)$ to the point of 

1373: $J$ representing the bundle $\w_C((j+1)R-(g+j)S)$. 

1374: 

1375: We claim that $\mu(W_j)\subseteq\Theta$. Indeed, let 

1376: $C^{(3)}:=C\times C\times C$, and denote by $p_{1,2}$ 

1377: and $p_3$ the projection maps of 

1378: $C^{(3)}$ onto the indicated factors. Let 

1379: $\Delta_{1,3}$ and $\Delta_{2,3}$ be the indicated 

1380: diagonals of $C^{(3)}$. Set

1381:        $${\cal F}:=p_3^*\w_C((j+1)\Delta_{1,3}

1382:          -(g+j)\Delta_{2,3}).$$

1383: Recall the notation of Subsection \ref{wronski}. From 

1384: the construction of $\Theta$, to show that 

1385: $\mu(W_j)\subseteq\Theta$, it is enough to show that 

1386: the Wronski map $\rho_{j,g+j-1}$ represents 

1387: universally the cohomology of ${\cal F}$ or, put more 

1388: simply, that $\rho_{j,g+j-1}$ can be viewed as a 

1389: presentation of the right derived image 

1390: $R^1(p_{1,2})_*{\cal F}$. 

1391: 

1392: Let ${\cal G}:=p_3^*\w_C((j+1)\Delta_{1,3})$. Then 

1393: ${\cal F}\subseteq{\cal G}$. From the definition of 

1394: the Wronski map $\rho_{j,g+j-1}$, we get that 

1395: $\rho_{j,g+j-1}$ is the image under $(p_{1,2})_*$ of 

1396: the quotient map ${\cal G}\to{\cal G}/{\cal F}$. Thus, 

1397: the map $\rho_{j,g+j-1}$ 

1398: is the first map

1399: in the following 

1400: piece of the long derived sequence of 

1401: $0\to{\cal F}\to{\cal G}\to{\cal G}/{\cal F}\to 0$ 

1402: under $(p_{1,2})_*$:

1403:         $$(p_{1,2})_*{\cal G}\to

1404:           (p_{1,2})_*({\cal G}/{\cal F})\to

1405:           R^1(p_{1,2})_*{\cal F}\to 

1406: 	  R^1(p_{1,2})_*{\cal G}.$$

1407: Now, a fiberwise analysis shows that 

1408: $R^1(p_{1,2})_*{\cal G}=0$. Thus $\rho_{j,g+j-1}$ 

1409: is a presentation for $R^1(p_{1,2})_*{\cal F}$, finishing 

1410: the proof that $\mu(W_j)\subseteq\Theta$. 

1411: 

1412: Let ${\cal L}:=\w_C((j+1)P-(g+j)Q)$, and denote 

1413: by $[{\cal L}]$ the point of $J$ representing ${\cal L}$. 

1414: Since $(P,Q)\in W_j$, we have $[{\cal L}]\in\Theta$. 

1415: By Proposition \ref{WiPgen}, $h^0(C,{\cal L})=1$. Thus, 

1416: it follows from \cite{acgh}, Prop. (4.2), p. 189, that 

1417: $[{\cal L}]$ is a nonsingular point of $\Theta$. 

1418: Furthermore, identifying the cotangent space of 

1419: $J$ at $[{\cal L}]$ with $H^0(C,\w_C)$, the cotangent 

1420: space of $\Theta$ at $[{\cal L}]$ is the quotient by 

1421: the subspace $H^0(C,\w_C(-F))$, where $F$ is the 

1422: unique effective divisor of $C$ such that 

1423: ${\cal L}={\cal O}_C(F)$. 

1424: 

1425: Identifying the cotangent space of $C\times C$ at 

1426: $(P,Q)$ with $\w_C|_P\oplus\w_C|_Q$, the induced map 

1427: of cotangent spaces 

1428: $d\mu^*\: T^*_{J,[{\cal L}]}\to T^*_{C\times C,(P,Q)}$ 

1429: is equivalent 

1430: to the evaluation map,

1431:        $$\epsilon\: H^0(C,\w_C)\to \w_C|_P\oplus\w_C|_Q.$$

1432: We claim that $\epsilon(H^0(C,\w_C(-F)))\neq 0$. 

1433: Indeed, if that were not the case, we would have 

1434: $H^0(C,\w_C(-F))=H^0(C,\w_C(-F-P-Q))$, that is,

1435:        $$h^0(C,{\cal O}_C((g+j-1)Q-(j+2)P))

1436:          =h^0(C,{\cal O}_C((g+j)Q-(j+1)P)).$$

1437: By the Riemann--Roch theorem,

1438:        $$h^0(C,{\cal O}_C((g+j)Q-(j+1)P))=

1439:          h^0(C,{\cal L})=1,$$

1440: and thus, also by the Riemann--Roch theorem,

1441:        $$h^0(C,\w_C((j+2)P-(g+j-1)Q))=3.$$

1442: However, this contradicts Proposition \ref{arbD}. 

1443: 

1444: Since $\mu(W_j)\subseteq\Theta$, the image of 

1445: $\epsilon(H^0(C,\w_C(-F)))$ in the cotangent space of 

1446: $W_j$ at $(P,Q)$ is zero. Since 

1447: $\epsilon(H^0(C,\w_C(-F)))\neq 0$, that cotangent space 

1448: is a proper quotient of the cotangent space of 

1449: $C\times C$ at $(P,Q)$, and thus has dimension at most 1. 

1450: Since $W_j$ is a divisor, it follows that $W_j$ is 

1451: nonsingular at $(P,Q)$.\qed

1452: 

1453: \section{Special ramification classes}\label{sec5}

1454: 

1455: \begin{claim}\label{srl} 

1456: {\rm ({\it Special ramification loci.})

1457: Let $C$ be a smooth curve of genus $g\geq 1$. 

1458: For each nonnegative integer $i$, consider the 

1459: following loci in $C\times C$:

1460: \begin{enumerate}

1461: \item The locus $D^+_i$ of pairs $(P,Q)\in C\times C$

1462: such that

1463:        $$(g+i-1)Q-(i+1)P$$

1464: is linearly equivalent to an effective divisor.

1465: \item The locus $E^+_i$ of pairs $(P,Q)\in C\times C$ such

1466: that

1467:        $$(g+i+1)Q-(i+1)P$$

1468: is linearly equivalent to a moving effective divisor.

1469: \item The locus $SW^+_i$ of pairs

1470: $(P,Q)\in C\times C$ such 

1471: that

1472: $Q$ is a special ramification

1473: point of $V_C(i,P)$.

1474: \end{enumerate}

1475: We claim that, set-theoretically,

1476:        \be

1477:        SW^+_i=D^+_i\cup E^+_i.\label{eq:swde}

1478:        \ee

1479: 

1480: Indeed, by the Riemann--Roch Theorem, for a pair

1481: $(P,Q)\in C\times C$, the divisor

1482: $(g+i-1)Q-(i+1)P$ is linearly equivalent to an

1483: effective one if and only if

1484:         \be

1485:     h^0(\w_C((i+1)P-(g+i-1)Q))\geq 2,\label{eq:6b3}

1486:     \ee

1487: while $(g+i+1)Q-(i+1)P$ is linearly equivalent to a

1488: moving effective divisor if and only if

1489:         \be

1490:     h^0(\w_C((i+1)P-(g+i+1)Q))\geq 1.\label{eq:6b4}

1491:     \ee

1492: At any rate, if $(P,Q)\in D^+_i\cup E^+_i$, then $Q$ is a

1493: special ramification point of $V_C(i,P)$, that is,

1494: $(P,Q)\in SW^+_i$

1495: 

1496: On the other hand, let

1497: $(P,Q)\in C\times C-(D^+_i\cup E^+_i)$. Then

1498:         \begin{eqnarray*}

1499:     &&h^0(\w_C((i+1)P-(g+i-1)Q))=1,\\

1500:     &&h^0(\w_C((i+1)P-(g+i+1)Q))=0.

1501:     \end{eqnarray*}

1502: So, either $Q$ is an ordinary or a simple ramification

1503: point of $V_C(i,P)$, that is, $(P,Q)\not\in SW^+_i$.

1504: 

1505: Let $\Delta$ be the diagonal subscheme of $C\times C$.

1506: Notice that $E_i^+\cap\Delta$ consists of the pairs

1507: $(P,P)$ such that $P$ is a Weierstrass point of $C$.

1508: However, if $g>1$, both $D_i^+$ and $SW_i^+$ contain

1509: $\Delta$. (If $g=1$, then 

1510: $D_i^+=E_i^+=SW_i^+=\emptyset$.)

1511: 

1512: Let $D_i$, $E_i$ and $SW_i$ be the loci of points in

1513: $D_i^+$, $E_i^+$ and $SW_i^+$ that lie off $\Delta$. Of

1514: course, Expression (\ref{eq:swde}) implies

1515: $SW_i=D_i\cup E_i$. Our Proposition~\ref{A2CC} claims

1516: that, if $C$ is general, then

1517: $SW_i=D_i\cup E_i$ holds in a more

1518: refined way, in the cycle group of $C\times C$. Before

1519: stating it, we need to endow $D_i$, $E_i$ and $SW_i$

1520: with natural subscheme structures.}

1521: \end{claim}

1522: 

1523: \begin{claim}\label{srs} 

1524: {\rm ({\it Special ramification schemes}) 

1525: Keep the notation of Subsection \ref{srl}, and recall 

1526: that of Subsections \ref{wronski} and \ref{jWd}. 

1527: Notice that the subsets 

1528: $D_i^+$ and $E_i^+$ are the supports

1529: of the degeneracy schemes of $\rho_{i,g+i-2}$ and

1530: $\rho_{i,g+i}$, respectively. So we may give 

1531: $D_i^+$ and $E_i^+$ the

1532: corresponding scheme structures.

1533: Give $D_i$ and $E_i$ the

1534: corresponding open subscheme structures. We say that 

1535: $D_i$ and $E_i$ are the {\it $i$-th special ramification 

1536: schemes of type Diaz and Cukierman}, respectively. Call 

1537: $E_i^+$ the {\it $i$-th expanded special ramification 

1538: scheme of type Cukierman}.

1539: 

1540: In addition, differentiating along the fibers of $p_1$ a 

1541: section of ${\cal O}_{C\times C}(Z_i)$ defining $Z_i$, 

1542: we obtain a section of 

1543: ${\cal J}^1_{p_1}({\cal O}_{C\times C}(Z_i))$, 

1544: well-defined modulo $\CC^*$. By functoriality, its zero 

1545: scheme contains a pair $(P,Q)$ if and only if $Q$ is a 

1546: special Weierstrass point of $V_C(i,P)$. Thus the zero 

1547: scheme gives a scheme structure for

1548: $SW_i^+$. Give $SW_i$ the induced open subscheme 

1549: structure. We say that $SW_i$ is the {\it $i$-th special 

1550: ramification scheme of $C$}.

1551: 

1552: Now, $Z_i=W_i+g\Delta$. As done 

1553: for $Z_i$, we can differentiate along

1554: the fibers of $p_1$ a section of ${\cal O}_{C\times C}(W_i)$

1555: defining $W_i$ to obtain a section of ${\cal J}^1_{p_1}({\cal

1556: O}_{C\times C}(W_i))$. Its zero scheme $S$ coincides with the

1557: scheme $SW_i$ off $\Delta$, because $Z_i$ coincides with $W_i$

1558: there. Moreover, if $C$ is general, then $S$ does not intersect

1559: $\Delta$, and hence $S=SW_i$ scheme-theoretically. Indeed, let $P$

1560: be a point of $C$. If $(P,P)\in W_i$, then $P$ is a Weierstrass

1561: point of $C$. Moreover, as $C$ is general, by Corollary~\ref{ws},

1562: the point $P$ is a simple Weierstrass point. So, it follows from

1563: Proposition \ref{prop5} that $W_i$ intersects the fiber

1564: $\{P\}\times C$ transversally at $(P,P)$. Thus the derivative

1565: along $\{P\}\times C$ of a section defining $W_i$ does not vanish

1566: at $(P,P)$. So $S\cap\Delta=\emptyset$.}

1567: \end{claim}

1568: 

1569: \begin{claim}\label{matrices}{\bf Lemma.}

1570: Let ${\cal O}$ be a local 

1571: ring, and $r$ a nonnegative integer. Let $M$ be a matrix 

1572: with $r+2$ rows and $r+1$ columns and entries in 

1573: ${\cal O}$. Let $M_1$ and $M_2$ be the submatrices 

1574: obtained from $M$ by removing the last row, and

1575: the last two rows, respectively. Assume that the 

1576: matrix obtained from $M_1$ by taking residues 

1577: has rank at least $r$. Let $z$

1578: denote the determinant of $M_1$. Then there are 

1579: $u,v\in{\cal O}$ such that

1580: \begin{enumerate}

1581:     \item $(z,u)$ is the ideal of all maximal minors 

1582:       of $M_2$,

1583:     \item $(z,v)$ is the ideal of all maximal minors 

1584:       of $M$,

1585:     \item $(z,uv)$ is the ideal generated by the two 

1586:       maximal minors of $M$ obtained by removing each 

1587:       of the last two rows.

1588: \end{enumerate}

1589: \end{claim}

1590: 

1591: \proof We may write $M$ in the form

1592:         $$M=\left[\matrix{A&a&b\cr c&f_1&f_2

1593:         \cr d&g_1&g_2\cr e&h_1&h_2}\right],$$

1594: where $A$ is a square matrix of size $r-1$, where $a$ 

1595: and $b$ are column vectors of 

1596: size

1597: $r-1$, where 

1598: $c$, $d$ and $e$ are row vectors of 

1599: size

1600: $r-1$, 

1601: and where $f_1$, $f_2$, $g_1$, $g_2$,

1602: $h_1$ and $h_2$ are elements of ${\cal O}$.

1603: 

1604: Let $I$ and $J$ be the ideals of ${\cal O}$ generated,

1605: respectively, by all maximal minors of the submatrices

1606:         $$M_2=\left[\matrix{A&a&b\cr c&f_1&f_2}\right]

1607:         \quad{\rm and }\quad M=\left[

1608:         \matrix{A&a&b\cr c&f_1&f_2\cr d&g_1&g_2\cr

1609:     e&h_1&h_2}\right].$$

1610: Also, let $K\subseteq{\cal O}$ be the ideal generated by 

1611: the determinants of the square submatrices

1612:         $$M_1=\left[

1613:         \matrix{A&a&b\cr c&f_1&f_2\cr d&g_1&g_2}

1614:         \right]\quad{\rm and }\quad

1615:         M'_1:=\left[\matrix

1616: 	{A&a&b\cr c&f_1&f_2\cr e&h_1&h_2}

1617:         \right].$$

1618: Notice that the determinant of the first matrix is $z$.

1619: 

1620: From the hypothesis, the matrix obtained from $M_2$ 

1621: by taking residues has rank at least $r-1$. Thus, 

1622: performing row and column operations on $M$, including 

1623: column and row exchanges, we may, without changing the 

1624: ideals $I$, $J$ and $K$, assume that $A$ is the 

1625: identity matrix, $a=b=0$ and $c=d=e=0$. Then

1626: $z=f_1g_2-f_2g_1$ and

1627:     \begin{eqnarray*}

1628:     I&=&(f_1,\, f_2),\\

1629:     J&=&(f_1g_2-f_2g_1,\,f_1h_2-f_2h_1,\,g_1h_2-g_2h_1),\\

1630:     K&=&(f_1g_2-f_2g_1,\, f_1h_2-f_2h_1).

1631:     \end{eqnarray*}

1632: 

1633: Now, since the matrix obtained from $M_1$ 

1634: by taking residues has rank at least $r$, at least one 

1635: among $f_1,\, f_2,\, g_1,\, g_2$ is invertible.

1636: 

1637: If $f_1$ is invertible, then

1638:         $$g_1h_2-g_2h_1=(g_1/f_1)(f_1h_2-f_2h_1)-

1639:         (h_1/f_1)(f_1g_2-f_2g_1).$$

1640: Thus, the lemma holds for $u=1$ and $v=f_1h_2-f_2h_1$. 

1641: The case where $f_2$ is invertible is similar.

1642: 

1643: If $g_1$ is invertible, then

1644:         \begin{eqnarray*}

1645:         (f_1h_2-f_2h_1)&=&(f_1/g_1)(g_1h_2-g_2h_1)+

1646:         (h_1/g_1)(f_1g_2-f_2g_1),\\

1647:     f_2&=&(g_2/g_1)f_1-(1/g_1)(f_1g_2-f_2g_1).

1648:         \end{eqnarray*}

1649: Thus the lemma holds for $u=f_1$ and $v=g_1h_2-g_2h_1$. 

1650: A similar analysis holds if $g_2$ is invertible.\qed

1651: 

1652: \begin{claim}\label{A2CC}{\bf Proposition.}

1653: Let $C$ be a general

1654: smooth curve of genus $g\geq 1$ and $i$ a nonnegative

1655: integer. Let $\Delta$ be the diagonal of $C\times C$ and 

1656: $W_i$ the $i$-th Weierstrass divisor. 

1657: Let $SW_i$ be the $i$-th special ramification 

1658: scheme, and $D_i$ and $E_i$ the $i$-th special 

1659: ramification schemes of type Diaz and Cukierman, 

1660: respectively. Let $E_i^+$ be the $i$-th expanded 

1661: special ramification scheme of type Cukierman. 

1662: Then these ramification schemes are finite and satisfy, 

1663: in the cycle group of $C\times C$:

1664:         $$[SW_i]=[D_i]+[E_i]\quad{\rm and }\quad

1665:         [E_i^+]=[E_i]+(g+1)[W_i\cap\Delta].$$

1666: \end{claim}

1667: 

1668: \proof Since $C$ is general, by Statement 1 of

1669: Proposition~\ref{WiPgen}, the set $SW_i$ is finite

1670: for each $i\geq 0$. Thus, so are $D_i$ and $E_i$

1671: by Expression~(\ref{eq:swde}). It follows that

1672: $E_i^+$ is finite, because $E_i^+\cap\Delta$ is the

1673: set of points $(P,P)$ such that $P$ is Weierstrass, whence

1674: is finite.

1675: 

1676: Recall the notation of Subsections \ref{wronski}, 

1677: \ref{jWd}, \ref{srl} and \ref{srs}. 

1678: Set $r:=g+i-1$. Both equalities can be proved locally. 

1679: Thus, let $(P,Q)\in C\times C$ and ${\cal O}$ be the 

1680: local ring of $C\times C$ at $(P,Q)$. 

1681: As a map of ${\cal O}$-modules, 

1682: $\rho_{i,r+1}$ is given by a matrix $M$ of the form 

1683: described in the proof of Lemma \ref{matrices}. Let 

1684: us use the notation described in the statement of 

1685: that lemma.

1686: 

1687: Let $K\subseteq{\cal O}$ define $SW_i^+$. Then 

1688: $K=(z,z')$, where $z$ (resp. $z'$) 

1689: is the maximal minor obtained 

1690: from $M$ by removing the last (resp. last but one) row. 

1691: Notice that, 

1692: from the nature of $M$ as a ``wronskian matrix'', $z'$ is 

1693: also the derivative of $z$ along $p_1$. Let 

1694: $I$ and $J$ be the ideals of ${\cal O}$ defining $D_i^+$ 

1695: and $E_i^+$, respectively. Then $I$ and $J$ are the 

1696: ideals of all the

1697: maximal minors of $M_2$ and $M$, respectively.

1698: 

1699: Now, since $C$ is a general curve, by Statement 2 of 

1700: Proposition \ref{WiPgen},

1701:         $$h^0(\w_C((i+1)P-(g+i)Q))\leq 1.$$

1702: This translates in the matrix obtained from $M_1$ by 

1703: evaluating at $(P,Q)$ having rank at least $r$. 

1704: Applying Lemma \ref{matrices},

1705: there are $u,v\in{\cal O}$ such

1706:     $$I=(z,u),\quad J=(z,v),\quad K=(z,uv).$$

1707: Now, since $E_i^+$ is finite-dimensional and 

1708: $C\times C$ is smooth, the sequence $z,v$ is regular. 

1709: The same holds for the sequence $z,u$ 

1710: if $P\neq Q$. It follows 

1711: that $[SW_i]=[D_i]+[E_i]$.

1712: 

1713: The second equality in the statement of the

1714: proposition is obvious off $\Delta$. Thus, assume 

1715: $Q=P$. Since $E_i^+\cap\Delta=W_i\cap\Delta$, we may 

1716: also assume that $P$ is a Weierstrass point of $C$. 

1717: 

1718: Let $t$ be a local parameter of $C$ at $P$, and 

1719: $t_1,t_2\in{\cal O}$ be its pullbacks with respect to 

1720: the projections $p_1$ and $p_2$. Then $t:=t_2-t_1$ is a 

1721: local equation for $\Delta$. As we saw in Subsection 

1722: \ref{srs}, we have $z=t^gw$, where

1723: $w\in{\cal O}$ defines $W_i$, and is not divisible by 

1724: $t$. Letting $\partial$ denote the 

1725: derivative with respect to $t_2$, we have

1726:        $$z'=\partial z=\partial(t^gw)=

1727:        gt^{g-1}w+t^g\partial w.$$

1728: Thus $t^{g-1}$ divides $z$ and $z'$, and hence each 

1729: element of $K$, in particular $uv$. Since $E_i^+$ is 

1730: finite, $t$ does not divide $v$, and hence $t^{g-1}|u$. 

1731: Let $L:=t^{1-g}K$. Then there are two expressions for 

1732: $L$:

1733:        \be

1734:        L=(tw,uv/t^{g-1})\quad{\rm and }\quad

1735:        L=(tw,gw+t\partial w).

1736:        \label{I}

1737:        \ee

1738: Since $W_i\cap\Delta$ is finite, the 

1739: sequences $gw+t\partial w, t$ and $w, t$ are 

1740: regular. Thus, from the second expression for $L$ above, 

1741: we get

1742:        $$\ell({\cal O}/L)=2\ell({\cal O}/(t,w))+

1743:        \ell({\cal O}/(w,\partial w)).$$

1744: Now, $\ell({\cal O}/(w,\partial w))=0$ because $w$ and 

1745: $\partial w$ cut out $SW_i$, and $SW_i$ does not meet 

1746: $\Delta$. Also, by Lemma \ref{lem}, $W_i$ intersects 

1747: $\Delta$ transversally. Thus

1748: $\ell({\cal O}/(t,w))=1$, and hence $\ell({\cal O}/L)=2$.

1749: 

1750: Now, since the sequence $z, v$ is regular, and 

1751: $z=t^gw$, also the sequence $tw,v$ is regular. Thus, 

1752: from the first expression for $L$ in (\ref{I}), 

1753: we get

1754:        $$\ell({\cal O}/L)=\ell({\cal O}/(tw,u/t^{g-1}))+

1755:          \ell({\cal O}/(tw,v)),$$

1756: and whence $\ell({\cal O}/(tw,v))\leq 2$. Since 

1757: ${\cal O}$ is regular, and the sequence $tw,v$ is 

1758: regular, so is the sequence $v,w$. Thus

1759:        $$\ell({\cal O}/(tw,v))=\ell({\cal O}/(t,v))+

1760:          \ell({\cal O}/(w,v)).$$

1761: Since $E_i^+$ contains $(P,P)$, the function $v$ is zero 

1762: on $(P,P)$. Thus, since also $t$ and $w$ vanish on 

1763: $(P,P)$, we get 

1764: $\ell({\cal O}/(t,v))=\ell({\cal O}/(w,v))=1$. So, the 

1765: multiplicity of $E_i^+$ at $(P,P)$ is

1766:        $$\ell({\cal O}/(z,v))=

1767:        g\ell({\cal O}/(t,v))+\ell({\cal O}/(w,v))

1768:        =(g+1).$$

1769: Since, by Lemma \ref{lem}, the multiplicity of 

1770: $W_i\cap\Delta$ at $(P,P)$ is 1, we are done.\qed

1771: 

1772: \begin{claim}\label{prop07}{\bf Proposition.}

1773: Let $C$ be a general

1774: smooth curve of genus $g\geq 1$ and $i$ a nonnegative

1775: integer. Let $SW_i$ be the $i$-th special ramification 

1776: scheme of $C$. Then

1777:         \be

1778:     \int_{C\times C}[SW_i]=2ig(g-1)\Big((i+2)(g+i)^2

1779:     +2(g+i)+2\Big).

1780:     \label{eq:me1}

1781:     \ee

1782: \end{claim}

1783: 

1784: \proof Recall the notation of Subsections 

1785: \ref{wronski}, \ref{jWd}, \ref{srl} and \ref{srs}. 

1786: Since $C$ is general, $SW_i$ is finite. Also,

1787: $SW_i$ is the zero scheme of a section of the

1788: rank-2 bundle

1789: ${\cal J}^1_{p_1}({\cal O}_{C\times C}(W_i))$. Thus

1790: its class in

1791: the Chow group of $C\times C$ satisfies

1792:         $$[SW_i]=c_2({\cal J}^1_{p_1}

1793:         ({\cal O}_{C\times C}(W_i))).$$

1794: Using the truncation sequence for bundles of jets, we get

1795:         $$[SW_i]=[W_i](c_1(p_2^*\w_C)+[W_i]).$$

1796: Now, $c_1(p_2^*\w_C)=K_2$. Using Expression

1797: (\ref{eq:17}) for $j=i$, and taking into account

1798: that $K_\ell^2=0$ for $\ell=1,2$, we get

1799:         \begin{eqnarray*}

1800:     [SW_i]&=&i(g+i+1)\Big((g+i)^2+g+i+1\Big)K_2[\Delta]\\

1801:     &+&\frac{1}{2}i(i+1)\Big((g+i)^2+g+i+1\Big)K_1K_2\\

1802:     &+&i^2(g+i+1)^2[\Delta]^2+i^2(g+i+1)(i+1)K_1[\Delta].

1803:     \end{eqnarray*}

1804: Using Formulas (\ref{eq:kd}) and (\ref{eq:kk}), we get

1805:         \begin{eqnarray*}

1806:     \int_{C\times C}[SW_i]&=&i(g+i+1)

1807:     \Big((g+i)^2+g+i+1\Big)(2g-2)\\

1808:     &+&\frac{1}{2}i(i+1)\Big((g+i)^2+g+i+1\Big)4(g-1)^2\\

1809:     &-&i^2(g+i+1)^2(2g-2)+i^2(g+i+1)(i+1)(2g-2).

1810:     \end{eqnarray*}

1811: Simplifying, we get the claimed formula.\qed

1812: 

1813: \begin{claim}\label{thm6DiazCuk}{\bf Theorem.}

1814: Let $C$ be a general smooth curve of genus $g\geq 1$,

1815: and $i$ a nonnegative integer. Let $D_i$ and $E_i$ be the 

1816: $i$-th special ramification schemes of type Diaz and 

1817: Cukierman, respectively. Then $D_i$ and $E_i$ are 

1818: reduced, and

1819:       \be

1820:       \int_{C\times C}[D_i]=g(g-1)

1821:       \Big((g+i-1)^2(i+1)^2-(g-1)^2\Big)

1822:       \label{eq:ex0}

1823:       \ee

1824: and

1825:       \be

1826:       \int_{C\times C}[E_i]=g(g-1)

1827:       \Big((g+i+1)^2(i+1)^2-(g+1)^2\Big).

1828:       \label{eq:ex6b0}

1829:       \ee

1830: \end{claim}

1831: 

1832: \proof Recall the notation of 

1833: Subsections \ref{wronski}, \ref{jWd}, \ref{srl} and 

1834: \ref{srs}. We will 

1835: first compute the degrees of $D_i$ and $E_i$. 

1836: First of all, since $E_i^+$ is finite, and

1837: is the degeneracy scheme of $\rho_{i,g+i}$, applying

1838: Porteous formula (\cite{Fu}, Thm. 14.4, p. 254), 

1839: we get the

1840: following expression for the class $[E_i^+]$ in the

1841: Chow group of $C\times C$:

1842:       $$[E_i^+]=c_2({\cal J}^{g+i}_{p_1}({\cal L}_i)-

1843:       p_1^*{\cal E}_i).$$

1844: Now, $c_2({\cal E}_i)=c_1({\cal E}_i)^2=0$,

1845: since $C$ is one-dimensional. Thus

1846:       $$[E_i^+]=c_2({\cal J}^{g+i}_{p_1}({\cal L}_i))

1847:       -c_1({\cal J}^{g+i}_{p_1}({\cal L}_i))

1848:       c_1(p_1^*{\cal E}_i).$$

1849: Using the truncation sequence of the bundles of jets,

1850: we get

1851:       \begin{eqnarray*}

1852:       c_1({\cal J}^{g+i}_{p_1}({\cal L}_i))&=&

1853:       \sum_{\ell=1}^{g+i+1}(\ell K_2+(i+1)[\Delta]);\\

1854:       c_2({\cal J}^{g+i}_{p_1}({\cal L}_i))&=&

1855:       \sum_{m=2}^{g+i+1}\sum_{\ell=1}^{m-1}

1856:       (\ell K_2+(i+1)[\Delta])(m K_2+(i+1)[\Delta]).

1857:       \end{eqnarray*}

1858: Expanding, and using that $K_2^2=0$, we get

1859:       \begin{eqnarray*}

1860:       c_1({\cal J}^{g+i}_{p_1}({\cal L}_i))&=&

1861:       \frac{1}{2}(g+i+1)(g+i+2)K_2+(i+1)(g+i+1)[\Delta];\\

1862:       c_2({\cal J}^{g+i}_{p_1}({\cal L}_i))&=&

1863:       \frac{1}{2}(i+1)(g+i)(g+i+1)(g+i+2)K_2[\Delta]\\

1864:       &&+\frac{1}{2}(i+1)^2(g+i)(g+i+1)[\Delta]^2.

1865:       \end{eqnarray*}

1866: Finally, using Formula (\ref{eq:c1Ei}) for $j=i$, and

1867: Formulas (\ref{eq:kd}) and (\ref{eq:kk}), we get

1868:       $$\int_{C\times C}[E_i^+]=(i+1)^2g(g-1)(g+i+1)^2.$$

1869: 

1870: Now, it follows from Proposition \ref{lem} that

1871: $W_i$ meets $\Delta$

1872: transversally at $g^3-g$ points. Thus, using

1873: Proposition \ref{A2CC}, we get

1874:       \begin{eqnarray*}

1875:       \int_{C\times C}[E_i]&=&\int_{C\times C}[E_i^+]-

1876:       (g+1)(g^3-g)\\

1877:       &=&g(g-1)\Big((g+i+1)^2(i+1)^2-(g+1)^2\Big),

1878:       \end{eqnarray*}

1879: the stated formula for the degree of $[E_i]$.

1880: 

1881: Now, the expression for the degree of $[D_i]$ follows 

1882: now from the equality $[SW_i]=[D_i]+[E_i]$ proved in

1883: Proposition \ref{A2CC} and Formula (\ref{eq:me1})

1884: for the degree of $[SW_i]$ proved in Proposition

1885: \ref{prop07}.

1886: 

1887: Let us now show that $D_i$ and $E_i$ are reduced. 

1888: Let $(P,Q)\in SW_i$. Let $\wh{\cal O}$ be the completion 

1889: of the local ring of $C\times C$ at $(P,Q)$. 

1890: Let $t_1$ and $t_2$ be local equations in $\wh{\cal O}$ 

1891: for $\{P\}\times C$ and $C\times\{Q\}$, 

1892: respectively. Then $\wh{\cal O}=\CC[[t_1,t_2]]$. 

1893: Let $w\in{\cal O}$ be a local equation for 

1894: $W_i$. Since $(P,Q)\in SW_i$, and since $W_i$ is 

1895: nonsingular by Proposition \ref{lem}, we may assume that 

1896: $w=t_1+u$, where $u\in\CC[[t_2]]$. Now, 

1897: let $w'$ and $u'$ be the derivatives of $w$ and $u$ 

1898: with respect to $t_2$. Then the ideal defining 

1899: $SW_i$ at $(P,Q)$ is $(w,w')$, and the multiplicity of 

1900: the cycle $[SW_i]$ at $(P,Q)$ is 

1901: $\ell(\wh{\cal O}/(w,w'))$. Notice that $w'=u'$, and 

1902:        $$\frac{\wh{\cal O}}{(w,w')}\cong

1903:          \frac{\CC[[t_2]]}{(u')}=

1904:          \frac{\CC[[t_2]]}{(u,u')}\cong

1905: 	 \frac{\CC[[t_1,t_2]]}{(t_1,w,w')}.$$

1906: Thus the multiplicity of the cycle $[SW_i]$ at $(P,Q)$

1907: is the multiplicity $m$ of $SW_i\cap(\{P\}\times C)$ at 

1908: $(P,Q)$.

1909: 

1910: Since the formation of $SW_i$ commutes with base 

1911: change, this multiplicity $m$ satisfies

1912:        $$m=\wgt_V(Q)-1,$$

1913: where $V$ is the complete linear system of sections 

1914: of $\w_C((i+1)P)$. Now, by Propositions \ref{WiPgen} and 

1915: \ref{arbD}, the order sequence of $V$ at $Q$ satisfies

1916:        \begin{eqnarray*}

1917: 	 \epsilon_j(V,Q)&=&j\quad(j=0,1,\dots,g+i-3),\\

1918: 	 \epsilon_{g+i-2}(V,Q)&\leq&g+i-1,\\

1919: 	 \epsilon_{g+i-1}(V,Q)&\leq&g+i+1.

1920:        \end{eqnarray*}

1921: Thus $m\leq 2$, with equality if and only if 

1922:        $$h^0(\w_C((i+1)P-(g+i-1)Q))=2\quad{\rm and}

1923:          \quad h^0(\w_C((i+1)P-(g+i+1)Q))=1,$$

1924: that is, if and only if $(P,Q)\in D_i\cap E_i$. 

1925: Since $[SW_i]=[D_i]+[E_i]$ by Proposition~\ref{A2CC}, 

1926: it follows that $D_i$ and $E_i$ are reduced at 

1927: $(P,Q)$.\qed

1928: 

1929: \begin{thebibliography}{99}

1930: 

1931: \bibitem{A} E.~Arbarello, 

1932: {\it Weierstrass points and moduli of curves}. 

1933: Compo. Math. {\bf 29}, no. 3, (1974), 325--342.

1934: 

1935: \bibitem{acgh} E.~Arbarello, M. Cornalba, 

1936: P. A. Griffiths and J. Harris, 

1937: {\it Geometry of algebraic curves}.

1938: Grundlehren der mathematischen Wissenschaften, 

1939: vol.~267, Springer, New York, 1985.

1940: 

1941: \bibitem{cocoa} CoCoATeam, 

1942: {\it \cocoa: a system for doing Computations in 

1943: Commutative Algebra}.

1944: Available at http://cocoa.dima.unige.it.

1945: 

1946: \bibitem{Cukie} F.~Cukierman,

1947: {\it Families of Weierstrass points}.

1948: Duke Math. J. {\bf 58}, no. 2, (1989), 317--346.

1949: 

1950: \bibitem{CEG} C.~Cumino, E.~Esteves, and L.~Gatto,

1951: {\it Limits of special Weierstrass points}.

1952: To appear.

1953: 

1954: \bibitem{Diaz/Exc} S.~Diaz,

1955: {\it Exceptional Weierstrass points and the divisor

1956: on moduli space that they define}.

1957: Mem. Amer. Math. Soc., vol. 56, no. 327, Amer. Math. 

1958: Soc., Providence, 1985.

1959: 

1960: \bibitem{MathCont} E.~Esteves,

1961: {\it Linear systems and ramification points on

1962: reducible nodal curves}.

1963: In ``Algebra Meeting''  (A. Garcia, E. Esteves and

1964: A. Pacheco, Eds.), 21--35, Mat. Contemp., vol. 14,

1965: Soc. Bras. Mat., Rio de Janeiro, 1998.

1966: 

1967: \bibitem{Fu} W.~Fulton,

1968: {\it Intersection theory}.

1969: Grundlehren der mathematischen Wissenschaften, vol. 2,

1970: Springer, Berlin Heidelberg, 1984.

1971: 

1972: \bibitem{GP} L.~Gatto and F.~Ponza,

1973: {\it Derivatives of wronskians with applications to

1974: families of Weierstrass points}.

1975: Trans. Amer. Math. Soc. {\bf 351}, no. 6, (1999), 2233--2255.

1976: 

1977: \bibitem{HaMu} J.~Harris and D.~Mumford, 

1978: {\it On the Kodaira dimension of the moduli space of 

1979: curves}.

1980: Invent. Math. {\bf 67} (1982), 23--86.

1981: 

1982: \bibitem{Ke} S.~Keel,

1983: {\it Intersection theory of moduli space of stable 

1984: $n$-pointed curves of genus zero}. 

1985: Trans. Amer. Math. Soc. {\bf 330} (1992), 545--574.

1986: 

1987: \bibitem{Kn} F.~Knudsen,

1988: {\it The projectivity of the moduli space of 

1989: stable curves, II}. Math. Scand. {\bf 52} (1983), 

1990: 161--199.

1991: 

1992: \end{thebibliography}

1993: 

1994: \end{document}

1995: