1: \documentclass[12pt]{amsart}
2: \setcounter{tocdepth}{1}
3: \usepackage{amssymb}
4: \usepackage{epsfig}
5: \usepackage{graphicx}
6: \numberwithin{equation}{section}
7:
8:
9: \input xy
10: \xyoption{all}
11:
12:
13: \textheight 20cm \textwidth 13cm
14: \advance\headheight 2pt
15: \calclayout
16: \allowdisplaybreaks[3]
17:
18:
19:
20: \theoremstyle{plain}
21: \newtheorem{prop}{Proposition}
22: \newtheorem{Theo}[prop]{Theorem}
23: \newtheorem{theo}[prop]{Theorem}
24: \newtheorem{coro}[prop]{Corollary}
25: \newtheorem{cond}[prop]{Condition}
26: \newtheorem{lemm}[prop]{Lemma}
27: \newtheorem{assu}[prop]{Assumption}
28: \newtheorem{sublemm}[prop]{Sublemma}
29:
30: \theoremstyle{definition}
31: \newtheorem{defi}[prop]{Definition}
32: \newtheorem{prob}[prop]{Problem}
33: \newtheorem{ques}[prop]{Question}
34: \newtheorem{conj}[prop]{Conjecture}
35:
36: %\theoremstyle{remark}
37: \newtheorem{rema}[prop]{Remark}
38: \newtheorem{rems}[prop]{Remarks}
39: \newtheorem{exam}[prop]{Example}
40: \newtheorem{exer}[prop]{Exercise}
41: \newtheorem{nota}[prop]{Notation}
42:
43:
44:
45: %\usepackage[notref,notcite]{showkeys}
46:
47:
48: %\numberwithin{equation}{section}
49:
50: \newcommand{\bA}{\mathbb A}
51: \newcommand{\bC}{\mathbb C}
52: \newcommand{\bN}{\mathbb N}
53: \newcommand{\bP}{\mathbb P}
54: \newcommand{\bQ}{\mathbb Q}
55: \newcommand{\bZ}{\mathbb Z}
56:
57:
58: \newcommand{\cC}{\mathcal C}
59: \newcommand{\cD}{\mathcal D}
60: \newcommand{\cI}{\mathcal I}
61: \newcommand{\cK}{\mathcal K}
62: \newcommand{\cL}{\mathcal L}
63: \newcommand{\cM}{\mathcal M}
64: \newcommand{\cN}{\mathcal N}
65: \newcommand{\cO}{\mathcal O}
66: \newcommand{\cS}{\mathcal S}
67: \newcommand{\cT}{\mathcal T}
68: \newcommand{\cU}{\mathcal U}
69: \newcommand{\cV}{\mathcal V}
70: \newcommand{\cW}{\mathcal W}
71: \newcommand{\cX}{\mathcal X}
72:
73:
74: \newcommand{\ra}{\rightarrow}
75: \newcommand{\lra}{\longrightarrow}
76: \newcommand{\Spec}{{\rm Spec}}
77: \newcommand{\Gr}{{\rm Gr}}
78: \newcommand{\NS}{{\rm NS}}
79: \newcommand{\Hom}{{\rm Hom}}
80: \newcommand{\iHom}{{\mathcal H}\!{\mathit om}}
81: \newcommand{\Hilb}{{\mathcal H}\!{\mathit ilb}}
82: \newcommand{\Ext}{{\rm Ext}}
83: \newcommand{\iExt}{{\mathcal E}\!{\mathit xt}}
84: \newcommand{\mfm}{\mathfrak m}
85: \newcommand{\mfq}{\mathfrak q}
86: \newcommand{\mfr}{\mathfrak r}
87: \newcommand{\mfw}{\mathfrak w}
88: \newcommand{\mfo}{\mathfrak o}
89:
90: \makeatother
91: \makeatletter
92:
93:
94:
95:
96: \author{Brendan Hassett}
97: \address{Department of Mathematics \\
98: Rice University, MS 136 \\
99: Houston, TX 77251-1892}
100: \email{hassett@rice.edu}
101:
102: \author{Yuri Tschinkel}
103: \address{
104: Mathematisches Institut \\
105: Bunsenstr. 3-5 \\
106: 37073 G\"ottingen \\
107: Germany }
108: \email{
109: yuri@uni-math.gwdg.de}
110:
111:
112:
113:
114:
115:
116: \title[Potential density of rational points]{Potential
117: density of rational points for K3 surfaces
118: over function fields}
119:
120:
121:
122:
123:
124:
125: \begin{document}
126: \date{\today}
127:
128:
129: \begin{abstract}
130: We give examples of non-isotrivial
131: K3 surfaces over complex function fields with Zariski-dense rational
132: points and N\'eron-Severi rank one.
133: \end{abstract}
134:
135: \maketitle
136:
137: \section{Introduction}
138: \label{sect:introduction}
139:
140:
141: Let $F$ be a number field or a function field of a curve over the complex numbers.
142: For each variety $X$ defined over $F$, we can ask whether the rational points
143: $X(F)$ are dense in $X$. We say that rational points of $X$ are {\em potentially dense}
144: if there exists a finite extension $E/F$ with $X(E)$ dense in $X$. Potential
145: density of rational points is expected to be a geometric property, depending (for
146: smooth projective $X$) on the positivity of the canonical class $K_X$. For example,
147: if $X$ is a curve or a surface with $K_X$ negative, rational points are potentially dense;
148: indeed, such varieties are known to be rational over
149: some finite extension of $F$.
150: On the other hand, potential density fails for curves and for certain classes
151: of higher-dimensional varieties with $K_X$ positive, by a theorem of Faltings.
152: We refer the reader to \cite{Ca} for a general discussion
153: of the connections between classification theory and potential
154: density questions.
155:
156:
157: The intermediate case where $K_X=0$ remains mysterious. Already the case of K3 surfaces
158: is open, in general. Potential density has been proved for abelian varieties, Enriques
159: surfaces, and special classes of K3 surfaces, e.g., those with an elliptic fibration
160: or an infinite group of automorphisms \cite{HarTs} \cite{bt} \cite{bt1} \cite{h}. However, all the
161: existing examples of K3 surfaces with dense rational points have geometric
162: N\'eron-Severi rank
163: at least two.
164:
165: Our main result is:
166: \begin{theo}
167: \label{theo:one}
168: Let $B$ be a complex curve and $F=\bC(B)$ its function field.
169: There exist non-isotrivial K3 surfaces over $F$, with
170: geometric N\'eron-Severi rank one,
171: and a Zariski dense set of $F$-rational points.
172: \end{theo}
173: This partially answers a question of Campana \cite[Sect. 11]{Ca}.
174:
175: Over a number field, no examples of K3 surfaces with geometric
176: N\'eron-Severi rank one and a dense set of
177: rational points are known. Our method uses the fact that the ground field is uncountable (see
178: Remark~\ref{rema:uncountable}); it does not yield
179: examples over $\bar{\bQ}(B)$.
180:
181: \
182:
183: We summarize the contents of this paper. Sections~\ref{sect:deform}
184: through \ref{sect:fe} contain the proof of Theorem~\ref{theo:one}.
185: The reader may want to consult the heuristic discussion in
186: Section~\ref{sect:proj} before tackling the main argument.
187: Our general strategy is to deform reducible curves--unions of
188: sections with smooth
189: rational curves in the fibers--to new sections of larger degree
190: (see Proposition~\ref{prop:oneps}).
191: Sections~\ref{sect:deform} and \ref{sect:k3} review general results
192: on deformations of curves and moduli spaces of K3 surfaces.
193: Our examples are pencils of K3 surfaces of low degree;
194: Section~\ref{sect:k3pencil} describes these pencils
195: and Proposition~\ref{prop:makepencil} indicates the situations where
196: our construction applies. It is an open problem to find examples
197: of large degree with the properties described in Theorem~\ref{theo:one}.
198: Indeed, recent results of Gritsenko, Hulek, and Sankaran
199: imply that the moduli space of K3 surfaces of large degree is
200: of general type; in particular, there are no pencils through
201: the generic point. Section~\ref{sect:limit} partially explains
202: why pencils play a special r\^ole in our analysis. Section~\ref{sect:CY}
203: extends our results to higher-dimensional Calabi-Yau hypersurfaces.
204:
205:
206:
207:
208: \
209:
210:
211: \noindent {\bf Acknowledgments:} The first author was supported by the Sloan Foundation and
212: NSF grant 0134259. We benefited from conversations with
213: Izzet Coskun, Aise Johan de Jong, Joe Harris, and Jason Starr.
214: We are grateful to Coskun and Starr for allowing us to include
215: their results in Section \ref{sect:CY}.
216:
217:
218:
219: \section{Deformation theory}
220: \label{sect:deform}
221:
222: In this section we recall basic facts from deformation theory used in subsequent proofs.
223: Throughout, we work over an algebraically closed field $k$.
224:
225: Let $X$ be a smooth projective variety over $k$, $C\subset X$ a closed
226: subscheme. The tangent space
227: to the Hilbert scheme $\Hilb(X)$ at $[C]$ is
228: $$
229: \cT_{\Hilb(X)}|_{[C]}=\Hom_X(\cI_C,\cO_C)=\Hom_C(\cI_C/\cI_C^2,\cO_C).
230: $$
231: We have an exact sequence
232: $$
233: \cI_C/\cI^2_C\ra \Omega^1_{X}\otimes_{\cO_X} \cO_C \ra \Omega^1_C\ra 0;
234: $$
235: when $C$ is reduced, the left arrow is injective at generic points of $C$
236: \cite{Mat} pp. 188, \cite{Hart} II.8 Ex.1.
237:
238: Assume now that $C$ is a reduced local-complete-intersection
239: subscheme of $X$, so that $\cI_C/\cI_C^2$
240: is a locally free $\cO_C$-module. Since $C$ is reduced,
241: an $\cO_C$-linear homomorphism of locally-free modules is injective provided
242: it is injective at each generic point, and we obtain
243: $$
244: 0 \ra \cI_C/\cI^2_C\ra \Omega^1_{X}\otimes_{\cO_X} \cO_C \ra \Omega^1_C\ra 0.
245: $$
246: Dualizing our exact sequence yields
247: \begin{eqnarray*}
248: & & 0 \ra \iHom_C(\Omega^1_C,\cO_C) \ra \iHom_C(\Omega^1_X,\cO_C) \ra
249: \iHom_C(\cI_C/\cI_C^2,\cO_C) \quad \quad (\dagger) \\
250: & & \quad \ra \iExt^1_C(\Omega^1_C,\cO_C)\ra 0
251: \end{eqnarray*}
252: on the level of coherent sheaves and
253: \begin{eqnarray*}
254: & & 0 \ra \Hom_C(\Omega^1_C,\cO_C) \ra \Hom_C(\Omega^1_X,\cO_C) \ra \Hom_C(\cI_C/\cI^2_C,\cO_C) \\
255: & & \quad \ra \Ext^1_C(\Omega^1_C,\cO_C)
256: \end{eqnarray*}
257: on cohomology. We write
258: $$\cT_C=\iHom_C(\Omega^1_C,\cO_C) \
259: \cT_X|_C=\iHom_C(\Omega^1_X,\cO_C) \
260: \cN_{C/X}=\iHom_C(\cI_C/\cI_C^2,\cO_C);
261: $$
262: the last two of these are locally free. We can rewrite $(\dagger)$ as
263: \begin{equation}
264: \label{ES2}
265: 0 \ra \cT_C \ra \cT_X|_C \ra \cN_{C/X} \ra \iExt^1_C(\Omega^1_C,\cO_C) \ra 0.
266: \end{equation}
267: Moreover, we have
268: $$\cT_{\Hilb(X)}|_{[C]}=\Gamma(C,\cN_{C/X}).$$
269:
270: Assume that $C$ is a nodal curve and fix $p\in C$ a node. We can choose
271: local coordinates $x_1,\ldots,x_n$ for $X$ centered about $p$ so that
272: $$C=\{x_1x_2=x_3=\ldots=x_n=0 \}.$$
273: We give local presentations for the terms in exact sequence (\ref{ES2}).
274: The tangent bundle of $X$ restricted to $C$ is freely generated by
275: $$\partial/\partial x_1, \partial/\partial x_2,\ldots, \partial/\partial x_n.$$
276: The subsheaf $\cT_C$ is generated by
277: $ x_1\partial/\partial x_1-x_2\partial/\partial x_2 $
278: which induces the relation
279: $$ x_1\partial/\partial x_1=x_2\partial/\partial x_2 $$
280: in the quotient sheaf $\cT_X|_C/T_C$. The sheaf $\iExt^1_C(\Omega^1_C,\cO_C)$
281: has fiber at $p$ equal to the versal deformation space of the isolated singularity
282: $(C,p)\subset X$, which in this case is one-dimensional. The normal bundle $\cN_{C/X}$
283: is an extension of this $\Ext$-group by the quotient $\cT_X|_C/T_C$, with free
284: generators
285: \begin{equation} \label{GEN}
286: x_2^{-1}\partial/\partial x_1=x_1^{-1}\partial/\partial x_2,
287: \partial/\partial x_3,\ldots, \partial/\partial x_n. \end{equation}
288:
289: Let $\nu:C^{\nu}\ra C$ denote the normalization of $C$ at the point $p$,
290: with conductor $\nu^{-1}(p)=\{p',p''\}$.
291: Consider the exact sequence
292: $$0 \ra \cT_{C^{\nu}} \ra \cT_X|_{C^{\nu}} \ra \cN_{C^{\nu}/X} \ra 0.$$
293: There is an induced extension (see \cite{GHS}\S 2, \cite{AK}\S 25-27):
294: \begin{equation}
295: \label{EX1}
296: 0 \ra \cN_{C^{\nu}/X} \ra \nu^*\cN_{C/X} \ra Q \ra 0,
297: \end{equation}
298: where $Q$ is torsion, supported at $\{p',p''\}$ with length one
299: at each point. We interpret $\nu^*\cN_{C/X}$ in (\ref{EX1})
300: as the sheaf of sections of $\cN_{C^{\nu}/X}$ with simple poles at $p'$
301: (resp. $p''$) in the direction $\cT_{C^{\nu}}|_{p''}$ (resp. $\cT_{C^{\nu}}|_{p'}$.)
302: The elements listed in (\ref{GEN}) may be regarded as free generators for
303: both $\nu^*\cN_{C^{\nu}/X}|_{p'}$ and $\nu^*\cN_{C^{\nu}/X}|_{p''}$;
304: the resulting identification expresses the compatibility condition satisfied
305: by sections of $\nu^*\cN_{C/X}$ which descend to $\cN_{C/X}$.
306:
307:
308:
309:
310:
311:
312:
313:
314:
315:
316:
317:
318:
319:
320:
321:
322:
323:
324:
325:
326: \section{K3 surfaces}
327: \label{sect:k3}
328:
329: In this section we work over $\bC$.
330:
331: Let $S_0$ be a smooth proper complex variety and $D\subset S_0$
332: a smooth divisor. There is a natural exact sequence
333: $$
334: 0\ra \cT_{S_0}\langle -D\rangle \ra \cT_{S_0}\ra \cN_{D/S_0}\ra 0,
335: $$
336: where the first term is derivations with logarithmic zeroes along $D$.
337: This induces a long exact sequence
338: $$
339: \ra H^1(S_0,\cT_{S_0}\langle -D\rangle)\ra H^1(S_0,\cT_{S_0})\stackrel{\phi}{\ra} H^1(D,\cN_{D/S_0}) \ra
340: $$
341: where the second term parametrizes the first order deformations of $S_0$
342: and the first term parametrizes first order deformations of the pair $(S_0,D)$ (see \cite{kawamata}).
343:
344: Let $\pi: \cS\ra \Delta$ be a deformation of $S_0$
345: over $\Delta=\Spec(\bC[[z]])$, with $S_0=\pi^{-1}(0)$.
346: Consider the extension
347: \begin{equation} \label{ext1}
348: 0\ra \cN_{D/S_0}\ra \cN_{D/\cS}\ra \cN_{S_0/\cS}|_D\ra 0
349: \end{equation}
350: and the corresponding extension class
351: $$
352: \eta\in \Ext^1_{D}(\cN_{S_0/\cS}|_{D}, \cN_{D/S_0})\simeq \Ext^1_{D}(\cO_{D}, \cN_{D/S_0})=H^1(D,\cN_{D/S_0}).
353: $$
354: Let
355: $$
356: \kappa\in H^1(S_0,\cT_{S_0})
357: $$
358: be the Kodaira-Spencer class of $d/dz$; using the isomorphisms
359: $$H^1(S_0,\cT_{S_0})\simeq \Ext_{S_0}^1(\cO_{S_0},\cT_{S_0}) \quad
360: \cN_{S_0/\cS}\simeq \cO_{S_0}$$
361: we see that $\kappa$ classifies the extension
362: $$0 \ra \cT_{S_0} \ra \cT_{\cS}|_{S_0} \ra \cN_{S_0/\cS} \ra 0.$$
363: Restricting this to $D$ and taking quotients by $\cT_D$, we
364: obtain extension (\ref{ext1}). Thus we obtain:
365:
366:
367: \begin{lemm}
368: \label{lemm:gen}
369: The classes $\phi(\kappa)$ and $\eta$ are proportional by a non-zero constant.
370: \end{lemm}
371:
372: \begin{prop}\label{prop:minusone}
373: Let $\pi: \cS\ra \Delta$ be a deformation of a K3 surface $S_0$,
374: containing a smooth rational curve $\ell$.
375: Assume that the Kodaira-Spencer class $\kappa$ is not
376: in the tangent space to the locus of K3 surfaces containing
377: some deformation of $\ell$. Then
378: $$
379: \cN_{\ell/\cS} \simeq \cO_{\bP^1}(-1)\oplus \cO_{\bP^1}(-1).
380: $$
381: \end{prop}
382: \begin{proof}
383: The moduli space of K3 surfaces is smooth, with tangent space $H^1(S_0,\cT_{S_0})$
384: at $S_0$; the subspace parametrizing surfaces containing a deformation of $\ell$
385: is also smooth with tangent space equal to the image of
386: $$H^1(\cT_{S_0}\langle -\ell \rangle) \ra H^1(\cT_{S_0}).$$
387: Moreover, $\kappa$ is not in this subspace precisely when
388: $\phi(\kappa)\neq 0$.
389:
390: We have the extension
391: $$
392: 0\ra \cN_{\ell/S_0}\ra \cN_{\ell/\cS}\ra \cN_{S_0/\cS}|_{\ell}\ra 0
393: $$
394: with $\cN_{\ell/S_0}\simeq \cO_{\bP^1}(-2)$ and $\cN_{S_0/\cS}|_{\ell}\simeq \cO_{\bP^1}$.
395: The extension class $\eta$ is in the group
396: $$\Ext_{\ell}^1(\cN_{S_0/\cS}|_{\ell},\cN_{\ell/S_0})\simeq
397: H^1(\bP^1,\cO_{\bP^1}(-2))\simeq \bC$$
398: hence
399: $$
400: \cN_{\ell/\cS}=
401: \begin{cases}
402: \cO_{\bP^1}(-1) \oplus \cO_{\bP^1}(-1) & \text{ if } \eta \neq 0\\
403: \cO_{\bP^1} \oplus \cO_{\bP^1}(-2) & \text{ if } \eta = 0
404: \end{cases}.
405: $$
406: Lemma~\ref{lemm:gen} and the hypothesis guarantee
407: that $\eta \neq 0$.
408: \end{proof}
409:
410: \
411:
412: For each $g\ge 2$,
413: let $\cK_{2g-2}$ denote the moduli space of
414: K3 surfaces $S$ with a primitive polarization $f$ of degree
415: $2g-2$. For each $d>0$,
416: let $\cD_{g,d}$ denote the moduli space of triples $(S,f,\ell)$,
417: where $(S,f)\in \cK_{2g-2}$ and
418: $$\Lambda_{g,d}:=\left<f,\ell\right>=\bZ f+\bZ \ell \subset \NS(S)$$
419: is a saturated
420: sublattice with intersection form
421: $$
422: \begin{array}{c|cc}
423: & f & \ell \\
424: \hline
425: f & 2g-2 & d \\
426: \ell & d & -2
427: \end{array}.
428: $$
429: Let $\iota:\cD_{g,d}\ra \cK_{2g-2}$ be the induced projection, which is finite
430: onto its image.
431:
432: Each $\cD_{g,d}$ is an open subspace of the moduli spaces
433: of lattice-polarized K3 surfaces of
434: Nikulin \cite{Nik1} (see \cite{Dol} for detailed discussion).
435: Now $\Lambda_{g,d}$ admits a primitive imbedding into the
436: cohomology lattice of a K3
437: surface, which is unique up to conjugation by automorphisms
438: of the cohomology lattice \cite{Nik2} 1.14.4. Surjectivity
439: of the period map and the global Torelli theorem \cite{K3book}
440: Expos\'es X,IX guarantee $\cD_{g,d}$
441: is nonempty and irreducible.
442:
443: \begin{prop} \label{prop:existcurve}
444: Given $(S,f,\ell)\in \cD_{g,d}$ such that $\NS(S)=\left<f,\ell\right>$,
445: which is the case for a very general such triple. Then
446: there is a smooth
447: rational curve in $S$ with divisor class $\ell$. Moreover,
448: $f$ is very ample if $g>2$ and induces a branched double cover
449: $S\ra \bP^2$ if $g=2$.
450: \end{prop}
451: \begin{proof}
452: The assertions of the last sentence are applications of results
453: on linear series of K3 surfaces by Mayer and Saint-Donat:
454: \begin{enumerate}
455: \item{complete linear series have no isolated basepoints
456: \cite{SD} 3.1;}
457: \item{an ample divisor $f$ has fixed components only when there is an
458: smooth curve $E\subset S$ of genus one
459: and a smooth rational curve $\Gamma\subset S$ so that $f=gE+\Gamma$
460: \cite{SD} 8.1;}
461: \item{$f$ is hyperelliptic only when $f^2=2$ (in which
462: case $S$ is a double cover of $\bP^2$ branched over a sextic)
463: or there exists
464: a smooth curve $E\subset S$ of genus one
465: with $f\cdot E=2$ \cite{SD} 5.2;}
466: \item{except in the special cases listed above, $f$
467: is very ample \cite{SD} 6.1.}
468: \end{enumerate}
469: Our hypothesis implies that $\NS(S)$ has discriminant
470: $$\mathrm{Disc}\left<f,\ell\right>=-(f\cdot f)(\ell \cdot \ell)+
471: (f\cdot \ell)^2=4(g-1)+d^2>4.$$
472: In case (2), we have
473: $$\mathrm{Disc} \left<E,\Gamma\right>=-(E\cdot E)(\Gamma \cdot \Gamma)+
474: (E\cdot \Gamma)^2=1;$$
475: in the latter case of (3), the discriminant is $4$.
476:
477: Recall the description of the monoid
478: of effective classes of a polarized K3 surface $(S,f)$
479: \cite{LP} 1.6: It is generated by
480: \begin{enumerate}
481: \item{$D\in \NS(S)$ with $D^2\ge 0$ and $D\cdot f >0$;}
482: \item{$\Gamma \in \NS(S)$ with $\Gamma^2=-2$ and $\Gamma \cdot f>0$.}
483: \end{enumerate}
484: Moreover, indecomposable effective classes $\Gamma$ with $\Gamma^2=-2$
485: represent smooth rational curves. Thus $\ell$ is effective, and
486: represents a smooth rational curve if it is indecomposable.
487:
488: Assume the contrary, so
489: there exists an indecomposable effective class $m$ with
490: $\ell-m$ effective and nonzero.
491: As $f$ is ample, we have $f\cdot \ell>f\cdot m$.
492: Since $m$ is indecomposable, $m\cdot m \ge -2$.
493: The discriminants satisfy
494: $$\mathrm{Disc}\left<f,m\right>=-(f\cdot f)(m\cdot m)+(f\cdot m)^2<
495: -(f\cdot f)(\ell\cdot \ell)+(f\cdot \ell)^2=\mathrm{Disc}\left<f,\ell\right>.$$
496: However, since $\left<f,m\right> \subset \left<f,\ell\right>$ we have
497: $$\mathrm{Disc}\left<f,m\right>\ge
498: \mathrm{Disc}\left<f,\ell\right>,$$
499: a contradiction.
500: \end{proof}
501:
502:
503:
504: \section{Pencils of K3 surfaces}
505: \label{sect:k3pencil}
506: \begin{defi} \label{defn:pencilK3}
507: A {\em pencil of K3 surfaces} consists of
508: \begin{enumerate}
509: \item{a projective irreducible threefold $Y \subset \bP^N$;}
510: \item{a codimension-two linear subspace $\Lambda \subset \bP^N$
511: meeting $Y$ transversally along a smooth curve $Z$,
512: so that a generic hyperplane $\Lambda \subset H \subset \bP^N$
513: intersects $Y$ in a K3 surface.}
514: \end{enumerate}
515: Projection from $\Lambda$
516: $$p_{\Lambda}:\mathrm{Bl}_{\Lambda} \bP^N \ra \bP^1$$
517: induces a flat morphism
518: $$\pi:\cS:=\mathrm{Bl}_Z Y \ra \bP^1$$
519: with generic fiber a K3 surface.
520: \end{defi}
521:
522: Let $E$ denote the exceptional divisor of the blow-up
523: $\beta:\mathrm{Bl}_Z Y \ra Y$.
524: \begin{enumerate}
525: \item{Since $Z$ is a complete intersection in $Y$ we have
526: $$E\simeq \bP(\cN_{Z/Y})\simeq \bP(\cO_Z(1) \oplus \cO_Z(1))\simeq \bP^1 \times Z;$$
527: projection onto the first factor coincides with $\pi|_E$. Thus
528: for each $p\in \bP^1$, the fiber $\cS_p=\pi^{-1}(p)$ intersects $E$
529: along $\{p\}\times Z$.}
530: \item{For each $z\in Z$, the fiber $E_z:=\beta^{-1}(z)$ meets each
531: member of the pencil in one point and thus yields a section
532: $\mathsf s_z:\bP^1 \ra \cS$
533: of $\pi$.}
534: \item{The normal bundle of this section is computed by the exact sequence
535: $$
536: \begin{array}{ccccccccc}
537: 0 & \ra & \cN_{E_z/E} &\ra & \cN_{E_z/\cS} &\ra & \cN_{E/\cS}|_{E_z} &\ra & 0 \\
538: 0 & \ra & \cO_{\bP^1} & \ra & \cO_{\bP^1} \oplus \cO_{\bP^1}(-1) & \ra & \cO_{\bP^1}(-1) & \ra & 0.
539: \end{array}
540: $$}
541: \item{For each smooth fiber $\cS_p$ and each curve $\ell \subset \cS_p$,
542: the intersection $Z\cap \ell\neq \emptyset$. Given $z\in Z\cap \ell$,
543: $E_z$ meets $\ell$ transversally at $z$, i.e.,
544: the tangent spaces of $\ell$ and $E_z$ at $z$ are transverse.}
545: \end{enumerate}
546:
547: \
548:
549: \begin{exam}
550: Let $Y\subset \bP^N$ be a threefold with isolated singularities and $\omega_Y=\cO_Y(-1)$;
551: if $Y$ is smooth then it is a Fano threefold. We can choose $\Lambda$ so that it meets
552: $Y$ along a smooth curve, which yields a pencil of K3 surfaces as in Definition~\ref{defn:pencilK3}.
553: \end{exam}
554:
555: \begin{prop} \label{prop:makepencil}
556: Fix $2\leq g \leq 10$. There exists a projective variety
557: $X\subset \bP^M$ so that
558: the generic K3 surface of degree $2g-2$ can be
559: realized as an intersection of $X$ with a linear
560: subspace of codimension $\dim(X)-2$, and thus as a member
561: of a pencil of K3 surfaces.
562: \end{prop}
563: \begin{proof}
564: The generic K3 surface of degree $2$ is a hypersurface
565: of degree $6$ in weighted projective space
566: $\bP(1,1,1,2)$, which has isolated singularities. Imbed
567: $$X:=\bP(1,1,1,2) \hookrightarrow \bP^M$$
568: using weighted-homogeneous forms of degree six.
569: Smooth hyperplane sections are degree $2$ K3 surfaces.
570:
571: For $g=3,4,5,6,7,8,9,10$,
572: the generic K3 surface of degree $2g-2$ is a complete intersection
573: in a generalized flag variety \cite{Mukai}. We extract the
574: variety $X$ in each case:
575: \begin{itemize}
576: \item{$g=3$ quartic hypersurface in $\bP^3$: take $X$ to be
577: the $4$-fold Veronese
578: reimbedding of $\bP^3$;}
579: \item{$g=4$ complete intersection of a quadric and a cubic
580: in $\bP^4$: take $X$ to be the $3$-fold Veronese
581: reimbedding of a smooth quadric hypersurface $Q\subset \bP^4$;}
582: \item{$g=5$ a generic complete intersection of three quadrics: take
583: $X$ to be the $2$-fold Veronese reimbedding of $\bP^5$;}
584: \item{$g=6$ a complete intersection of $\Gr(2,5)$ with
585: a quadric and a codimension-three linear space: take $X$
586: to be the $2$-fold Veronese reimbedding of the generic codimension-three
587: linear section of $\Gr(2,5)$;}
588: \item{$g=7$ a codimension-eight linear section
589: of the isotropic Grassmannian $\mathrm{IGr}(5,10)\subset \bP^{15}$;}
590: \item{$g=8$ a codimension-six linear section of the Grassmainnian
591: $\Gr(2,6)\subset \bP^{14}$;}
592: \item{$g=9$ a codimension-four linear section of the Lagrangian
593: Grassmannian $\mathrm{LGr}(3,6)\subset \bP^{13}$;}
594: \item{$g=10$ a codimension-three linear section of the
595: flag variety of dimension five, associated with the adjoint representation
596: of $G_2$, imbedded in $\bP^{13}$.}
597: \end{itemize}
598:
599: We obtain pencils of K3 surfaces of degree $2g-2$ as follows:
600: Taking generic subspaces
601: $$\Lambda_{N-2} \subset \Lambda_{N} \subset \bP^M$$
602: of codimensions $\dim(X)-1$ and $\dim(X)-3$ respectively, there is a pencil
603: $$\Lambda_{N-2} \subset \Lambda_{N-1}(p) \subset \Lambda_N, \quad p\in \bP^1$$
604: of $(N-1)$-dimensional linear spaces. Taking $Y=\Lambda_N\cap X$
605: and $Z=\Lambda_{N-2}\cap X$, we obtain a pencil as in
606: Definition~\ref{defn:pencilK3}.
607: \end{proof}
608:
609:
610: \section{Deforming sections in pencils}
611: \label{sect:DSP}
612: \begin{prop}\label{prop:oneps}
613: Let $\pi:\cS \ra \bP^1$ be a pencil of K3 surfaces.
614: Suppose there exists a point $p\in \bP^1$ so that $\cS_p$ is smooth
615: and contains a smooth rational curve $\ell$ with normal bundle
616: $\cN_{\ell/\cS}\simeq \cO_{\bP^1}(-1) \oplus \cO_{\bP^1}(-1)$.
617: Then there exists a one-parameter family of sections of $\pi$, the closure
618: of which contains $\ell$.
619: \end{prop}
620: \begin{proof}
621: Let $Z$ be the base locus of the pencil.
622: Choose a point $z\in \ell \cap Z \subset \cS_p$ and consider the nodal
623: curve
624: $$C=\ell \cup_z E_z$$
625: with normalization $\nu:C^{\nu} \ra C$. Consider the exact sequence
626: $$0 \ra \cN_{C^{\nu}/\cS} \ra \nu^*\cN_{C/\cS} \ra Q \ra 0$$
627: with $Q$ a torsion sheaf, of length one at each of the points
628: $\{z',z''\}$ over $z$.
629:
630: It suffices to show that $\cN_{C/\cS}$ has no higher cohomology and
631: admits a global section not mapping to zero in $Q$ (see Section~\ref{sect:deform}):
632: Then the Hilbert scheme is smooth at $C$ and contains deformations of $C$
633: smoothing the node $z$ (cf. \cite{HT}, proof of Proposition 24
634: and \cite{GHS} \S 2).
635: Moreover, these deformations meet the generic fiber of $\pi$ at one
636: point, and thus are sections of $\pi$ not contained in $Z$.
637: Finally, since $\ell$ is a component of a degeneration of these
638: sections, it lies in the closure of the surface traced out
639: by these sections.
640:
641: We evaluate $\cN_{C/\cS}$ on each component of $C$: On $E_z$ we
642: have
643: \begin{equation} \label{EX2}
644: 0 \ra \cN_{E_z/\cS} \ra \cN_{C/\cS}|_{E_z} \ra Q(E_z) \ra 0
645: \end{equation}
646: with $Q(E_z)$ of length one at $z$; furthermore,
647: $\cN_{E_z/\cS}\simeq \cO_{\bP^1}(-1) \oplus \cO_{\bP^1}$
648: with the non-negative summand corresponding to the normal direction
649: to $E_z$ in $E$. Since the tangent vector to $\ell$ is not contained
650: in $E$, the $\cO_{\bP^1}$ component is saturated in the extension
651: (\ref{EX2}), and the $\cO_{\bP^1}(-1)$ component is {\em not} saturated
652: (\cite{HT} Proposition 23, cf. Sublemma 27). In particular, we conclude
653: $$\cN_{C/\cS}|_{E_z} \simeq \cO_{\bP^1} \oplus \cO_{\bP^1}.$$
654: As for $\ell$, we have
655: $\cN_{\ell/\cS}\simeq \cO_{\bP^1}(-1)^{\oplus 2}$ and an extension
656: $$0 \ra \cN_{\ell/\cS} \ra \cN_{C/\cS}|_{\ell} \ra Q(\ell) \ra 0$$
657: with $Q(\ell)$ of length one. In this case, the only possibility is
658: $$\cN_{C/\cS}|_{\ell} \simeq \cO_{\bP^1} \oplus \cO_{\bP^1}(-1).$$
659:
660: We observe that
661: \begin{enumerate}
662: \item{$\cN_{C/\cS}|_{E_z}$ is globally generated and has no higher
663: cohomology;}
664: \item{$\cN_{C/\cS}|_{\ell}$ has no higher cohomology and each nonzero
665: global section is nonzero in $Q(\ell)$.}
666: \end{enumerate}
667: Since $C$ is obtained by gluing $E_z$ and $\ell$ at a single point, $\cN_{C/\cS}$
668: also has no higher cohomology \cite{HT}, Lemma 21. Fixing a nonzero section
669: $t$ of $\cN_{C/\cS}|_{\ell}$, we can find a section of $\cN_{C/\cS}|_{E_z}$
670: agreeing with $t$ at $z$.
671: \end{proof}
672:
673:
674:
675: \section{Construction of the examples}
676: \label{sect:fe}
677:
678:
679: \begin{theo}
680: \label{theo:dense2}
681: Let $\pi:\cS\ra \bP^1$ be a pencil of K3 surfaces
682: of degree $2g-2$ with base locus
683: $Z$. Assume that there exists an infinite sequence of positive integers
684: $$
685: d_1 < d_2 < \ldots
686: $$
687: such that $\bP^1$ intersects
688: $\cD_{g,d_j}$ transversally at some point $p_j$ and the
689: class $\ell_j \in \Lambda_{g,d_j}$ is represented by a smooth
690: rational curve $\ell_j \subset \cS_{p_j}$.
691: Then $\cS$ has a Zariski dense set of sections of $\pi$.
692: \end{theo}
693: \begin{proof}
694: We have
695: $$\cN_{\ell_j/\cS}\simeq \cO_{\bP^1}(-1) \oplus \cO_{\bP^1}(-1)$$
696: by Proposition~\ref{prop:minusone}. Proposition~\ref{prop:oneps}
697: gives a one-parameter family of sections
698: $$\begin{array}{ccc}
699: \Sigma_j & \stackrel{\sigma_j}{\ra} & \cS \\
700: {\scriptstyle \psi_j} \downarrow \quad & & \\
701: R_j
702: \end{array},
703: $$
704: with $\sigma_j:\psi_j^{-1}(r)\ra \cS$ a section for $r\in R_j$ generic,
705: and $\sigma_j(\psi_j^{-1}(r_0))\supset \ell_j$ for the distinguished
706: point $r_0\in R_j$ around which the deformation was produced.
707: It follows that the irreducible surface
708: $\widetilde{\Sigma}_j:=\sigma_j(\Sigma_j)$ contains $\ell_j$.
709:
710: We claim that $\cup_j \widetilde{\Sigma}_j$ is dense in $\cS$. Let
711: $\Xi$ denote the closure of all these surfaces. If $\Xi \subsetneq \cS$
712: then $\Xi$ is a union of irreducible surfaces, each dominating $\bP^1$.
713: Thus the degree $\Xi\cap \cS_p \subset \cS_p$ is bounded;
714: In particular, $\Xi$ cannot contain $\ell_j$ when $d_j\gg 0$,
715: a contradiction.
716: \end{proof}
717:
718: \begin{coro}\label{coro:main}
719: Let $\cS \ra \bP^1$ be a very general pencil of degree $2g-2$,
720: for $2\le g \le 10$.
721: Then the hypothesis of
722: Theorem~\ref{theo:dense2} are satisfied, and sections are Zariski dense.
723: \end{coro}
724: \begin{proof}
725: By Proposition~\ref{prop:makepencil}, the moduli space
726: $\cK_{2g-2}$ is dominated by
727: an open subset of the Grassmannian $\Gr(N,M+1)$ parametrizing the $(N-1)$-dimensional
728: linear subspsaces in $\bP^M$.
729: For $d\gg 0$, $\cD_{g,d}$
730: determines a nonempty divisor in $\Gr(N,M+1)$, which is necessarily ample.
731: Each pencil determines a $\bP^1 \subset \Gr(N,M+1)$, so the pencils
732: are parametrized by a suitable Hilbert scheme. By
733: Proposition~\ref{prop:existcurve}, for a generic pencil
734: $\bP^1$ is transversal to $\cD_{g,d}$ at some
735: point and the corresponding fiber contains a smooth rational
736: curve of degree $d$. Take the countable intersection of these open subsets
737: in the space of all pencils; applying the Baire
738: category theorem, we are left with a nonempty dense subset of the space of all
739: pencils.
740: \end{proof}
741: \begin{rema}\label{rema:uncountable}
742: Our argument does not preclude the (unlikely) possibility that every pencil defined over
743: $\overline{\bQ}$ is tangent to $\cD_{g,d}$ for each $d \gg 0$.
744: \end{rema}
745:
746: We complete the argument for Theorem~\ref{theo:one}:
747: \begin{proof}
748: Corollary~\ref{coro:main}
749: gives pencils of K3 surfaces with dense rational points.
750: Proposition~\ref{prop:makepencil} implies that the generic K3 surface
751: of degree $2g-2,2\le g \le 10$ arises in such a pencil; in particular,
752: K3 surfaces with N\'eron-Severi rank one occur.
753: The N\'eron-Severi group of the geometric generic fiber
754: injects in the N\'eron-Severi group of each smooth member
755: of the pencil; indeed, the map on homology is an isomorphism.
756: Thus the N\'eron-Severi group of the geometric generic
757: fiber is also of rank one.
758:
759: For simplicity, we will only discuss non-isotriviality in the
760: degree four case: Any pair of quartic surfaces
761: $$S_i=\{f_i(w,x,y,z)=0 \}\subset \bP^3, \quad i=1,2$$
762: are contained in a pencil, e.g.,
763: $$\cS=\{s_1f_1+s_2f_2=0\} \subset \bP^3 \times \bP^1.$$
764: This is non-isotrivial whenever $S_1 \not \simeq S_2$, and
765: completes the proof when $B=\bP^1$.
766:
767:
768: For an arbitrary smooth complex curve $B$, we can express
769: $\bC(B)$ as a finite extension of $\bC(\bP^1)$. Any
770: K3 surface with dense $\bC(\bP^1)$-rational points
771: {\em a fortiori} has dense $\bC(B)$-rational points.
772: \end{proof}
773:
774:
775:
776:
777:
778: \section{Limitations}
779: \label{sect:limit}
780: In this section, we discuss constraints to deforming sections in
781: K3 fibrations. We hope to explain why pencils are a natural source
782: of examples of K3 surfaces
783: with potentially-dense rational points.
784:
785: Let $\pi:\cS \ra B$ be a fibration satisfying the following
786: \begin{enumerate}
787: \item{$\cS$ is smooth and the generic fiber is a K3 surface;}
788: \item{the singular fibers have at worst rational double points;}
789: \item{$\pi$ is non-isotrivial and projective.}
790: \end{enumerate}
791: Let $\mu:B \ra {\overline \cK_{2g-2}}$
792: be the classifying map into the Baily-Borel compactification of the
793: corresponding moduli space of polarized K3 surfaces. Recall that the
794: pull-back of the natural polarization on the Baily-Borel compactification is
795: $\cL:=\pi_*\omega_{\pi}$.
796: Since the fibers of $\pi$ have trivial dualizing sheaf,
797: $\omega_{\pi}$ is the pull-back of a line bundle from $B$, which
798: is necessarily $\cL$ (by the projection formula).
799:
800: Suppose we have
801: a section $\sigma:B\ra \cS$. Consider the exact sequence
802: $$0 \ra \cT_{\sigma(B)} \ra \cT_{\cS}|_{\sigma(B)} \ra \cN_{\sigma(B)/\cS} \ra 0.$$
803: It follows that
804: $$c_1(\cN_{\sigma(B)/\cS})=c_1(\cT_{\cS})-c_1(\cT_{\sigma(B)})=-c_1(\omega_{\pi}),$$
805: whence
806: $$\deg(\cN_{\sigma(B)/\cS})=-\deg(\mu(B))<0.$$
807: (For pencils of
808: quartic surfaces
809: the degree is $-1$, the smallest possible value.)
810: Hence $\cN_{\sigma(B)/\cS}$ cannot be globally generated at the generic point: If it were
811: then $\det(\cN_{\sigma(B)/\cS})$ would have a nonvanishing section,
812: contradicting our degree computation.
813: In particular, we cannot deform a section in
814: a two-parameter family sweeping out a dense subset of $\cS$.
815:
816:
817:
818: \section{Projective geometry of pencils of quartic surfaces}
819: \label{sect:proj}
820: Corollary~\ref{coro:main} says that a very general pencil of quartic
821: surfaces has a Zariski-dense set of sections. Here we interpret
822: these in terms of classical projective geometry.
823:
824: Recall the set-up of Theorem~\ref{theo:dense2}:
825: $$
826: \begin{array}{ccc}
827: \cS & \stackrel{\pi}{\ra} & \bP^1 \\
828: {\scriptstyle \beta} \downarrow \quad & & \\
829: \bP^3 & &
830: \end{array}
831: $$
832: where $\beta$ is the blowup along the base locus $Z$ of the pencil.
833: We analyze the proper transforms in $\bP^3$
834: of the sections of $\pi$
835: produced by
836: Proposition~\ref{prop:oneps}.
837:
838: Let $\ell_1$ be a line in some fiber of $\pi$; $\beta(\ell_1)$
839: is a four-secant line to $Z$. If $z\in Z\cap \beta(\ell_1)$ and
840: $E_z:=\beta^{-1}(z)$ then $\beta(\ell_1)\cup_z E_z$
841: deforms to a 3-secant line to $Z$.
842: These 3-secants move in a 1-parameter family.
843: Similarly, if $\ell_d$ is a smooth rational curve of degree $d$ in some
844: fiber of $\pi$ then $\beta(\ell_d)$ is a rational curve of degree $d$
845: meeting $Z$ in $4d$ points. The nodal curve $\beta(\ell_d)\cup_z E_z$
846: deforms to a 1-parameter family of rational curves in $\bP^3$ of degree $d$ meeting $Z$
847: in $4d-1$ points.
848:
849: Essentially, the proof of Proposition~\ref{prop:oneps}
850: shows that the space of rational curves of degree $d$ in $\bP^3$
851: meeting $Z$ in at least $4d-1$ points (counted with multiplicities)
852: has at least one irreducible component of the expected dimension
853: and the generic such curve meets $Z$ in $4d-1$ points.
854:
855:
856: \section{Higher-dimensional Calabi-Yau hypersurfaces}
857: \label{sect:CY}
858: We have seen in Corollary~\ref{coro:main} that a very general
859: pencil of quartic surfaces has a Zariski-dense set of sections.
860: The following generalization is due to Coskun
861: and Starr:
862: \begin{theo} \label{theo:dense3}
863: Let $\pi:\cX\ra \bP^1$ be a very general pencil of hypersurfaces
864: of degree $n+1$ in $\bP^n$, with $n\ge 4$. Then sections of
865: $\pi$ are Zariski dense.
866: \end{theo}
867: The first step is the following extension of Proposition~\ref{prop:oneps}
868: to higher-dimensional pencils (which has an analogous proof):
869: \begin{prop}\label{prop:onepsnew}
870: Let $\pi:\cX \ra \bP^1$ be a pencil of hypersurfaces of degree $n+1$
871: in $\bP^n$.
872: Suppose there exists a point $p\in \bP^1$ so that $\cX_p$ is smooth
873: and contains a smooth rational curve $\ell$ with normal bundle
874: $$\cN_{\ell/\cX}\simeq
875: \cO_{\bP^1}^{n-3} \oplus \cO_{\bP^1}(-1) \oplus \cO_{\bP^1}(-1).$$
876: Then there exists an $(n-2)$-parameter family of sections of $\pi$;
877: its closure contains the deformations of $\ell$ in $\cX_p$.
878: \end{prop}
879:
880: Just as in Theorem~\ref{theo:dense2}, Proposition~\ref{prop:onepsnew}
881: reduces Theorem~\ref{theo:dense3} to
882: the following result, which is a natural extension of
883: Theorem 1.27 of \cite{Cl} for quintic threefolds:
884: \begin{theo}
885: Let $X$ be a very general hypersurface of degree $n+1$ in $\bP^n$,
886: with $n\ge 4$.
887: Then there exists an infinite sequence of integers
888: $$d_1<d_2<\ldots $$
889: such that $X$ contains a rational curve $\ell$ of degree $d_j$
890: with
891: $$\cN_{\ell/X}\simeq
892: \cO_{\bP^1}^{n-3} \oplus \cO_{\bP^1}(-1) \oplus \cO_{\bP^1}(-1).$$
893: Furthermore, deformations of these curves are Zariski dense in $X$.
894: \end{theo}
895: It suffices
896: to produce one hypersurface $X$ with the desired properties;
897: even a singular $X$ will work, as long as it is smooth along each $\ell$.
898:
899: The proof proceeds by induction, following \cite{Cl}, which
900: establishes the base case $n=4$. Consider a hypersurface
901: $X\subset \bP^n$ of degree $n+1$
902: containing the codimension-two linear subspace
903: $\{x_0=x_n=0\}\subset \bP^n$. The defining equation for $X$
904: can be written uniquely
905: $$x_0 f_n(x_0,\ldots,x_{n-1})+x_n g_n(x_0,\ldots,x_n)=0 $$
906: where $f_n$ and $g_n$ have degree $n$.
907: Let $H$ denote the hyperplane $\{x_0=0\}$ and
908: $$W=\{f_n(x_0,\ldots,x_{n-1})=0\}\subset \{x_n=0\}\simeq \bP^{n-1},$$
909: a hypersurface of degree $n$.
910: The singular locus $\mathrm{Sing}(X)$ is defined by the equations
911: $$f_n+x_0\frac{\partial f_n}{\partial x_0} +
912: x_n\frac{\partial g_n}{\partial x_0}=
913: x_0\frac{\partial f_n}{\partial x_i}+x_n\frac{\partial g_n}{\partial x_i} =
914: x_n\frac{\partial g_n}{\partial x_n}+g_n=0,$$
915: where $i=1,\ldots,n-1.$ This contains the locus
916: $$\Sigma:=\{x_0=x_n=f_n=g_n=0\};$$
917: if this is a smooth complete intersection
918: then $\mathrm{Sing}(X)=\Sigma$.
919:
920: We make the following generality assumptions on $f_n$:
921: \begin{itemize}
922: \item{$W$ satisfies the inductive hypothesis, so
923: there is an infinite
924: sequence of integers $\{d_j\}$
925: so that $W$ contains a rational curve $\ell_j$ of
926: degree $d_j$ with normal bundle
927: $$\cN_{\ell_j/W}\simeq
928: \cO_{\bP^1}^{n-4} \oplus \cO_{\bP^1}(-1) \oplus \cO_{\bP^1}(-1).$$}
929: \item{$H$ intersects $W$ and each $\ell_j$ transversely.}
930: \end{itemize}
931: The Bertini theorem allows us to achieve the last condition
932: for a very general choice of $f_n$.
933: We also make assumptions on $g_n$:
934: \begin{itemize}
935: \item{$\Sigma$ is a smooth complete intersection, so that
936: $\mathrm{Sing}(X)=\Sigma$.}
937: \item{$\Sigma\cap \ell_j=\emptyset$ for each $j$.}
938: \end{itemize}
939: In particular, the $\ell_j$ avoid the singularities of $X$.
940:
941:
942: We can compute the normal bundle $\cN_{\ell_j/X}$ using
943: the extension
944: $$0 \ra \cN_{\ell_j/W} \ra \cN_{\ell_j/X} \ra \cN_{W/X}|_{\ell_j} \ra 0.$$
945: We have
946: $$\cN_{W\cup H/X}=\cO_{W\cap H}(+1)$$
947: and since $H$ is Cartier away from $\Sigma$
948: $$\cN_{W/X}|_{\ell_j}=\cN_{W\cup H/X}|_{\ell_j}(-H)=\cO_{\ell_j}.$$
949: The extension
950: $$0\ra \cO_{\bP^1}^{n-4} \oplus \cO_{\bP^1}(-1)^2 \ra \cN_{\ell_j/X} \ra
951: \cO_{\bP^1} \ra 0$$
952: is necessarily trivial and
953: $$\cN_{\ell_j/X}\simeq \cO_{\bP^1}^{n-3} \oplus \cO_{\bP^1}(-1)^2,$$
954: as desired.
955:
956: The deformations of
957: each $\ell_j$ sweep out a divisor in $W$; the union
958: of these divisors is dense in $W$.
959: It remains to check that the deformations of the $\ell_j$ are dense in
960: $X$. Each $\ell_j$ has deformations in $X$ which are not
961: contained in $W$. Indeed, we have shown that $\ell_j$
962: admits infinitesimal deformations
963: that are nonzero in $\cN_{W/X}|_{\ell_j}$. In particular, deformations of
964: $\ell_j$ sweep out an irreducible divisor $D_j \subset X$ not equal to $W$.
965: If the union $\cup_{j=1}^{\infty}D_j$ were a proper subvariety
966: $\Xi\subsetneq X$, then $\Xi\cap W$ would not be dense in $W$, contradicting the
967: inductive hypothesis.
968:
969:
970:
971:
972:
973:
974:
975:
976: \bibliographystyle{plain}
977: \bibliography{pd}
978:
979:
980: \end{document}
981:
982:
983:
984:
985:
986:
987:
988:
989:
990:
991:
992:
993:
994:
995:
996:
997:
998:
999: