1: %LATEX%
2: \NeedsTeXFormat{LaTeX2e}
3: \newif\ifsmfart
4: \IfFileExists{smfart.cls}
5: {\documentclass[12pt,english]{smfart}
6: \IfFileExists{smfenum.sty}{\usepackage{smfenum}}{}
7: \usepackage{bull}
8: \smfarttrue}
9: {\message{^^J*** It would be better to typeset this file with smfart.cls ***^^J^^J}
10:
11:
12: \documentclass[12pt]{amsart}\let\Subsection\subsection}
13: \setcounter{tocdepth}{1}
14: \usepackage{amssymb}
15: \numberwithin{equation}{section}
16:
17:
18:
19: \ifx\pdftexversion\undefined
20: \usepackage[dvips]{graphicx}
21: \else
22: \usepackage[pdftex]{graphicx}
23: \fi
24:
25: \input xy
26: \xyoption{all}
27:
28:
29: \textheight 20cm \textwidth 13cm
30: \advance\headheight 2pt
31: \calclayout
32: \allowdisplaybreaks[3]
33:
34: \theoremstyle{plain}
35: \newtheorem{prop}{Proposition}
36: \newtheorem{Theo}[prop]{Theorem}
37: \newtheorem{theo}[prop]{Theorem}
38: \newtheorem{coro}[prop]{Corollary}
39: \newtheorem{cond}[prop]{Condition}
40: \newtheorem{lemm}[prop]{Lemma}
41: \newtheorem{assu}[prop]{Assumption}
42: \newtheorem{sublemm}[prop]{Sublemma}
43:
44: \theoremstyle{definition}
45: \newtheorem{defi}[prop]{Definition}
46: \newtheorem{prob}[prop]{Problem}
47: \newtheorem{ques}[prop]{Question}
48: \newtheorem{conj}[prop]{Conjecture}
49:
50: \theoremstyle{remark}
51: \newtheorem{rema}[prop]{Remark}
52: \newtheorem{rems}[prop]{Remarks}
53: \newtheorem{exam}[prop]{Example}
54: \newtheorem{exer}[prop]{Exercise}
55: \newtheorem{nota}[prop]{Notations}
56: \newcommand{\bA}{\mathbb A}
57: \newcommand{\bC}{\mathbb C}
58: \newcommand{\bN}{\mathbb N}
59: \newcommand{\bP}{\mathbb P}
60: \newcommand{\cC}{\mathcal C}
61: \newcommand{\cI}{\mathcal I}
62: \newcommand{\cL}{\mathcal L}
63: \newcommand{\cN}{\mathcal N}
64: \newcommand{\cO}{\mathcal O}
65: \newcommand{\cT}{\mathcal T}
66: \newcommand{\cU}{\mathcal U}
67: \newcommand{\cV}{\mathcal V}
68: \newcommand{\cW}{\mathcal W}
69: \newcommand{\cX}{\mathcal X}
70: \newcommand{\cY}{\mathcal Y}
71: \newcommand{\ra}{\rightarrow}
72: \newcommand{\lra}{\longrightarrow}
73: \newcommand{\Spec}{{\rm Spec}}
74: \newcommand{\Hom}{{\rm Hom}}
75: \newcommand{\iHom}{{\mathcal H}\!{\mathit om}}
76: \newcommand{\Ext}{{\rm Ext}}
77: \newcommand{\iExt}{{\mathcal E}\!{\mathit xt}}
78: \newcommand{\mfm}{\mathfrak m}
79: \newcommand{\mfp}{\mathfrak p}
80: \newcommand{\mfq}{\mathfrak q}
81: \newcommand{\mfr}{\mathfrak r}
82:
83:
84: \makeatother
85: \makeatletter
86:
87: \def\Title {Weak approximation over function fields}
88: \def\Author {Brendan Hassett and Yuri Tschinkel}
89: \def\Subject {Algebraic geometry}
90: \def\Keywords {rationally connected varieties, weak approximation}
91:
92: \newif\ifpdf
93: \ifx\pdfoutput\undefined
94: \pdffalse
95: \else
96: \pdfoutput=1
97: \pdftrue
98: \fi
99: \ifpdf
100: \usepackage[pdftex,
101: hyperindex=true,
102: pdftitle={\Title},pdfauthor={\Author},pdfsubject={\Subject},
103: pdfkeywords={\Keywords},pdfpagemode={UseOutlines},
104: bookmarks=true,bookmarksopen=true,
105: bookmarksopenlevel=0,bookmarksnumbered=true,
106: pdfstartview={FitV},
107: colorlinks=true
108: ]{hyperref}
109: \pdfcompresslevel=9
110: \else
111: \usepackage{hyperref}
112: \fi
113:
114:
115: \begin{document}
116:
117:
118:
119: \title[Weak approximation over function fields]
120: {Weak approximation over function fields}
121: \author{Brendan Hassett}
122: \address{Department of Mathematics \\
123: Rice University, MS 136 \\
124: Houston, TX 77251-1892}
125: \email{hassett@rice.edu}
126: \author{Yuri Tschinkel}
127: \address{Mathematisches Institut\\
128: Bunsenstr. 3-5 \\
129: 37073 G\"ottingen, Germany}
130: \email{yuri@uni-math.gwdg.de}
131:
132:
133:
134: %\begin{document}
135: \date{\today}
136:
137:
138: \begin{abstract}
139: We prove that
140: rationally connected varieties over the function field of a complex curve
141: satisfy weak approximation for places of good reduction.
142: \end{abstract}
143:
144: \maketitle
145: \tableofcontents
146:
147: %\input epsf
148:
149: \section{Introduction}
150: \label{sect:introduction}
151:
152:
153: Let $F$ be a number field and $X$ an algebraic variety over $F$.
154: Does there exist an $F$-rational point on $X$? If so, are they
155: ubiquitous on $X$? For many classes of varieties, such problems
156: are analyzed using local-to-global principles.
157: The Hasse principle says that $X$ has an $F$-rational point
158: provided it has a rational point over each completion of $F$.
159: The principle of weak approximation says that, given a finite collection of
160: places of $F$ together with a point of $X$ over each of the corresponding
161: completions, there exists an $F$-rational point approximating these
162: arbitrarily closely.
163:
164:
165: \
166:
167:
168: The impetus for this paper was
169: the following result by Graber, Harris and Starr:
170:
171: \begin{theo}[\cite{GHS},Theorem 1.2]
172: \label{theo:great}
173: Let $F$ be the function field of a smooth curve over $\bC$.
174: Every proper rationally connected variety $X$ over
175: $F$ has an $F$-rational point.
176: \end{theo}
177:
178: An algebraic variety is rationally connected if any two points can be
179: joined by a rational curve (see Section~\ref{sect:rat} for more details).
180: Rational and unirational varieties are rationally connected.
181: We refer the reader to \cite{dJS} for related results
182: in positive characteristic.
183:
184: \
185:
186:
187: From an arithmetic viewpoint, proving such a theorem
188: entails surmounting two obstacles: First
189: one needs to show that there are no obstructions to the existence
190: of a local point, i.e., $X(F_{\nu})\neq \emptyset$ for all completions $F_{\nu}$
191: of $F$. If $B$ is the smooth projective curve with $F=\bC(B)$
192: and $\pi:\cX\ra B$ is a model for $X$ over $B$,
193: one has to show there exist local analytic sections of $\pi$
194: at each point of $B$.
195: Secondly, one has to prove the Hasse principle for $X$ over $F$,
196: which entails constructing a global section of $\pi$.
197: Note that over a number field, the Hasse principle may fail
198: even for cubic surfaces.
199:
200:
201: \
202:
203: Theorem~\ref{theo:great} naturally leads one to ask
204: whether rationally connected varieties over $\bC(B)$ satisfy weak
205: approximation as well. In this paper we prove this away from singular
206: fibers of $\pi$, i.e., away from the places of bad reduction:
207: \begin{theo}
208: \label{theo:main}
209: Let $X$ be a smooth, proper,
210: rationally connected variety over the
211: function field of a curve over $\bC$. Then $X$ satisfies
212: weak approximation for places of good reduction.
213: \end{theo}
214: Theorem~\ref{theo:great} and Theorem 2.13 of \cite{KMM} give
215: the zeroth-order case:
216: There exists a section of $\pi:\cX \ra B$ passing through arbitrary points
217: of smooth fibers.
218:
219:
220: For varieties over function fields of curves, weak approximation
221: is satisfied in the following cases \cite{CT}:
222: \begin{itemize}
223: \item stably rational varieties;
224: \item connected linear algebraic groups and homogeneous spaces
225: for these groups;
226: \item homogeneous space fibrations over varieties that satisfy weak approximation,
227: for example, conic bundles over rational varieties;
228: \item Del Pezzo surfaces of degree at least four.
229: \end{itemize}
230: Weak approximation is not known for general cubic surfaces.
231: Madore has a manuscript addressing weak approximation for
232: cubic surfaces away from places of bad reduction.
233:
234: \
235:
236: {\bf Acknowledgments:}
237: The first author was partially supported by the
238: Sloan Foundation and NSF Grants 0134259 and 0196187.
239: The second author was partially supported by NSF Grant 0100277.
240: Part of this work was done while both authors were visiting
241: the American Institute of Mathematics in Palo Alto.
242: We thank J. de Jong, T. Graber, J. Harris, J. Koll\'ar and J. Starr
243: for conversations about this topic.
244:
245:
246:
247:
248:
249: \section{Basic properties of weak approximation}
250: \label{sect:not}
251:
252: \subsection{Definition}
253:
254:
255: Let $F$ be a number field or a function field
256: of a smooth projective curve $B$ over
257: an algebraically closed ground field $k$ of
258: characteristic zero. For each place $\nu$ of $F$,
259: let $F_{\nu}$ denote the $\nu$-adic completion of $F$.
260: Let $X$ be an algebraic variety
261: of dimension $d$ over $F$; in this paper,
262: all varieties are assumed to be geometrically integral.
263: Let $X(F)$ denote the set of $F$-rational points of $X$.
264: One says that rational points on $X$ {\em satisfy weak approximation}
265: if, for any finite set of places
266: $\{\nu_i\}_{i\in I}$ of $F$ and
267: $\nu_i$-adic open subsets $U_i\subset X(F_{\nu_i})$,
268: there is a rational point $x\in X(F)$
269: such that its image in each $X(F_{\nu_i})$ is contained in $U_i$.
270: In particular, for any collection of $x_i \in X(F_{\nu_i}), i\in I,$
271: there exists an $x\in X(F)$ arbitrarily close to each $x_i$.
272:
273: It is well known that weak approximation is a birational property:
274: If $X_1$ and $X_2$ are smooth varieties birational over $F$
275: then $X_1$ satisfies weak approximation if and only if $X_2$
276: satisfies weak approximation. Given a smooth proper variety $X_1$,
277: after applying
278: Chow's lemma and resolution of singularities we obtain a smooth
279: projective variety $X_2$ birational to $X_1$. Thus in proving
280: weak approximation, it usually suffices to consider
281: projective varieties. In particular, Theorem~\ref{theo:main}
282: reduces to this case.
283:
284:
285:
286: \
287:
288: For the rest of this paper we restrict our attention to the function
289: field case. Places $\nu$ of $F$ correspond to points $b$ on $B$.
290: We also assume that $X$ is projective, so it admits a projective model
291: $$\pi:\cX \ra B,$$
292: i.e., a flat projective morphism
293: with generic fiber $X$;
294: for each $b\in B$,
295: the fiber over $b$ is denoted
296: $$\cX_b=\cX\times_B \Spec(\cO_{B,b}/\mfm_{B,b}).$$
297: Sections of $\pi$ yield $F$-valued points of $X$ and
298: conversely, each $F$-valued point of $X$ extends to a section of
299: $\pi$.
300: Let $\widehat{B}_b$ denote the completion of $B$ at $b$ and
301: $(\widehat{\cO}_{B,b},\widehat{\mfm}_{B,b})$ the
302: associated complete local ring, which has fraction field $F_{\nu}$.
303: Restricting to this formal neighborhood of $b$ gives
304:
305: \centerline{
306: \xymatrix{
307: \cX\times_B \widehat{B}_b \ar[d]_{\hat{\pi}_b} \ar[r] & \cX\ar[d]_{\pi} \\
308: \widehat{B}_b \ar[r] & B
309: }}
310:
311: \noindent
312: Sections of $\hat{\pi}_b$ restrict to $F_{\nu}$-valued points of $X$ and
313: conversely, each $F_{\nu}$-valued point of $X$ extends to a section of
314: $\hat{\pi}_b$. Basic $\nu$-adic open subsets of $X(F_{\nu})$
315: consist of those sections of $\hat{\pi}_b$
316: which agree with a given section over
317: $\Spec(\widehat{\cO}_{B,b}/\widehat{\mfm}^{N+1}_{B,b})
318: \subset\widehat{B}_b$, for
319: some $N\in \bN$.
320: Weak approximation means that for any finite set of points $\{b_i\}_{i\in I}$
321: in $B$, sections $\hat{s}_i$ of $\hat{\pi}_{b_i}$,
322: and $N\in \bN$, there exists a section $s$ of $\pi$
323: agreeing with $\hat{s}_i$ modulo $\widehat{\mfm}^{N+1}_{B,b_i}$ for each $i$.
324:
325:
326:
327: \subsection{Fibers of good reduction}
328: \label{sect:good}
329: We continue to assume that $X$ is a smooth projective variety over $F=k(B)$,
330: $\nu$ a place of $F$, and $b\in B$ the corresponding point.
331: A place $\nu$ is {\em of good reduction} for $X$ if there exists
332: a scheme
333: $$\widehat{\cX}_b \ra \widehat{B}_b,$$
334: proper and smooth over $\widehat{B}_b$,
335: with generic fiber isomorphic to $X$ over $F_{\nu}$.
336: Let $S$ denote the finite set of places of bad reduction.
337:
338:
339: \begin{defi}
340: A variety
341: $X$ satisfies {\em weak approximation for places of good reduction} if,
342: for any finite set of places of good reduction
343: $\{\nu_i\}_{i\in I}$ and
344: $\nu_i$-adic open subsets $U_i\subset X(F_{\nu_i})$,
345: there is a rational point $x\in X(F)$
346: such that its image in each $X(F_{\nu_i})$ is contained in $U_i$.
347: \end{defi}
348: Suppose we have a model $\pi: \cX\ra B$ smooth over $B\setminus S$.
349: Then we can express this in more geometric terms:
350: For each finite set of points $\{b_i\}_{i\in I}$
351: in $B\setminus S$, sections $\hat{s}_i$ of $\hat{\pi}_{b_i}$
352: and $N\in \bN$, there exists a section $s$ of $\pi$
353: agreeing with $\hat{s}_i$ modulo $\widehat{\mfm}^{N+1}_{B,b_i}$ for each $i$.
354:
355:
356: \begin{prop}\label{prop:goodmodel}
357: Retain the notation introduced above.
358: There exists an algebraic space
359: $$\pi:\cX \ra B,$$
360: proper and flat over $B$, smooth over the places of good reduction,
361: and with generic fiber $X$.
362: Such a space is called a good model of $X$ over $B$.
363: \end{prop}
364: \begin{proof}
365: Choose a projective model $\pi':\cX'\ra B$ for $X$ over $B$.
366: If $\pi'$ is smooth over $B\setminus S$ there is nothing to prove. Otherwise,
367: let $\{b_j\}\subset B\setminus S$ denote the points over which
368: $\cX'_{b_j}$ is singular; let $\widehat{\cX}'_{b_j}$ denote the
369: completion of $\cX'$ along the central fiber $\cX'_{b_j}$.
370: By assumption,
371: there exists a proper smooth scheme
372: $$\widehat{\cX}_{b_j}\ra \widehat{B}_{b_j},$$
373: which is isomorphic to
374: our original model over the generic point.
375: Resolving the indeterminacy of the rational map
376: $$\widehat{\cX}_{b_j}\dashrightarrow \widehat{\cX}'_{b_j},$$
377: we find that these are
378: related by a sequence of modifications in the central fiber.
379: This gives a sequence of formal modifications
380: to $\cX'$ along the singular fibers,
381: in the sense of Artin \cite{Art70}.
382: Theorems 3.1 and 3.2
383: of \cite{Art70} give a unique proper algebraic
384: space $\pi:\cX \ra B$ realizing these formal modifications to $\cX'$.
385: \end{proof}
386: \begin{exam}
387: There are simple examples justifying the introduction of algebraic spaces.
388: Let $\pi:\cX'\ra B$ be a flat projective morphism such that each
389: fiber is a cubic surface with rational double points and the
390: generic fiber $X$ is smooth. Suppose that near each point $b\in B$ the
391: local monodromy representation
392: $$
393: \mathrm{Gal}(\bar{F}_{\nu}/F_{\nu})\ra
394: \mathrm{Aut}(\mathrm{Pic}(\bar{X}))
395: $$
396: is trivial.
397: By a theorem of Brieskorn \cite{Br1} \cite{Br2},
398: there exists a simultaneous resolution
399: $$
400: \begin{array}{rcccl}
401: \cX & & \stackrel{\varrho}{\lra} & &\cX' \\
402: & \searrow & & \swarrow& \\
403: & & B & &
404: \end{array},
405: $$
406: where $\cX$ is a smooth proper algebraic space over
407: $B$ and $\varrho_b:\cX_b \ra \cX'_b$
408: is the minimal resolution of $\cX'_b$ for each $b\in B$.
409: However, $\cX$ is constructed by making modifications
410: of $\cX'$ in formal neighborhoods of the singular fibers,
411: and hence is not necessarily a scheme.
412: Note that blowing up the singularities of $\cX'$
413: will usually introduce exceptional divisors in the fibers.
414: \end{exam}
415:
416:
417:
418:
419:
420: \begin{defi}
421: Let $b\in B\setminus S$ be a point of good reduction and
422: $
423: \pi: \cX\ra B
424: $
425: a good model. An $N$-{\em jet} of
426: $\pi$ at $b$ is a section of
427: $$
428: \cX\times_B \Spec(\cO_{B,b}/\mfm^{N+1}_{B,b})\ra
429: \Spec(\cO_{B,b}/\mfm^{N+1}_{B,b}).
430: $$
431: \end{defi}
432:
433:
434: Hensel's lemma guarantees that every $N$-jet is
435: a restriction of a section of $\hat{\pi}_b$.
436: Let $\{b_i\}_{i\in I}$ be a finite set of points of
437: good reduction and $j_i$ an $N$-jet of $\pi$ at $b_i$.
438: We write $J=\{ j_i\}_{i\in I}$ for the
439: corresponding collection of $N$-jets.
440:
441:
442: \subsection{Iterated blowups}
443: \label{sect:iter}
444: Let $\pi:\cX\ra B$ be good model of $X$
445: and $J=\{ j_i\}_{i\in I}$ a finite collection of $N$-jets
446: at points of good reduction $\{b_i\}$.
447: The {\em iterated blowup} associated with $J$
448: $$\beta(J):\cX(J)\ra \cX$$
449: is obtained by performing
450: the following sequence of blowups:
451: For each $i\in I$ choose a section $\hat{s}_i$ of $\hat{\pi}_{b_i}$
452: with jet $j_i$.
453: Now blow up $\cX$ successively $N$ times, where at each stage
454: the center is the point at which
455: the proper transform of
456: $\hat{s}_i$ meets
457: the fiber over $b_i$.
458: Observe that a blowup of $\cX$ centered in the
459: fibers of $\pi$ is uniquely determined by the corresponding
460: blowup of the completions along those fibers.
461: Note that at each stage we blow up a smooth point of
462: the fiber of the corresponding model and that the
463: result does not depend on the order of the $b_i$ or
464: on the choice of $\hat{s}_i$.
465:
466:
467:
468: The fiber $\cX(J)_{b_i}$ decomposes into irreducible components
469: $$
470: \cX(J)_{b_i}=E_{i,0} \cup \ldots \cup E_{i,N}
471: $$
472: where
473: \begin{itemize}
474: \item $E_{i,0}$ is the proper transform of $\cX_{b_i}$,
475: isomorphic to the blowup of $\cX_{b_i}$ at $r_{i,0}:=\hat{s}_i(b_i)$;
476: \item $E_{i,n}$, $n=1,\ldots, N-1$, is the blowup of $\bP^d$
477: at $r_{i,n}$, the point where the proper transform
478: of $\hat{s}_i$ meets the fiber over $b_i$ of the $n$th blowup;
479: \item $E_{i,N}\simeq \bP^d$.
480: \end{itemize}
481: The intersection $E_{i,n}\cap E_{i,n+1}$
482: is the exceptional divisor $\bP^{d-1}\subset E_{i,n}$
483: and a proper transform of a hyperplane
484: in $E_{i,n+1}$, for $n=0,\ldots, N-1$.
485: Let $r_i \in E_{i,N}\setminus E_{i,N-1}$ denote
486: the intersection of $\hat{s}_i$ with $E_{i,N}$.
487:
488:
489:
490: \begin{figure}[htb]
491: \centerline{\hskip 7cm\includegraphics[width=\textwidth]{weakfig1a}}
492: \caption{One fiber of the iterated blowup}
493: \label{fig1}
494: \end{figure}
495:
496:
497: For each section $s':\cX(J)\ra B$ the composition $\beta(J)\circ s'$
498: is a section of $\pi:\cX \ra B$. Conversely, given a section $s$ of
499: $\pi:\cX \ra B$, its proper transform $s'$ is the unique section
500: of $\cX(J)\ra B$ lifting $s$.
501: Sections $s'$ of $\cX(J)\ra B$ with $s'(b_i)=r_i$
502: yield sections of $\cX\ra B$ with
503: $N$-jet $j_i$ at $b_i$.
504: We therefore have the following criterion for weak
505: approximation in fibers of good reduction:
506:
507: \begin{prop}
508: \label{prop:jets}
509: $X$ satisfies weak approximation in fibers of good reduction
510: if and only if
511: there exists a good model $\pi:\cX \ra B$ with the following
512: property:
513: For each $N\in \bN$ and collection of $N$-jets $J$
514: with corresponding iterated blowup $\cX(J)$, and for
515: any choice of points $r_i\in E_{i,N}\setminus E_{i,N-1}, i\in I$,
516: there exists a section $s$ of $\cX(J)\ra B$ with $s(b_i)=r_i$
517: for each $i\in I$.
518: \end{prop}
519:
520:
521:
522:
523:
524:
525:
526: \section{Rationally connected varieties}
527: \label{sect:rcv}
528: We retain the notation introduced in Section \ref{sect:not}.
529: In particular, the ground field $k$ is algebraically closed
530: of characteristic zero.
531:
532: \subsection{Terminology and fundamental results}
533: \label{sect:rat}
534: Rational connectedness
535: was introduced in the classification of Fano
536: varieties \cite{Ca} \cite{KMM}. However, rationally connected
537: varieties are now of independent interest:
538: \begin{defi}[\cite{kollar} IV.3.2]
539: \label{defi:ko}
540: A variety $Y$ is {\em rationally chain connected}
541: (resp. {\em rationally connected}) if there is a
542: family of proper and connected curves $g:U\ra Z$
543: whose geometric fibers have only rational components
544: (resp. are irreducible rational curves) and a
545: cycle morphism $g:U\ra Y$ such that
546: $$
547: u^{(2)}:U\times_Z U \ra Y\times Y.
548: $$
549: is dominant.
550: \end{defi}
551: Our definition of `rationally chain connected' makes sense
552: for reducible schemes $Y$.
553:
554: \begin{exam}The class of rationally connected varieties includes
555: unirational varieties and smooth Fano varieties, \cite{kollar} IV.3.2.6,
556: V.2.13. In particular, smooth hypersurfaces of degree $\le m$
557: in $\bP^m$ are rationally connected.
558: \end{exam}
559:
560:
561: \begin{defi}
562: Let $Y$ be a smooth algebraic space of dimension $d$ and
563: $f:\bP^1 \ra Y$ a morphism, so we have an isomorphism
564: $$
565: f^*T_Y\simeq \cO_{\bP^1}(a_1) \oplus \ldots \oplus \cO_{\bP^1}(a_d)
566: $$
567: for suitable integers $a_1,\ldots,a_d$.
568: Then $f$ is {\em free} (resp. {\em very free}) if each $a_i\ge 0$
569: (resp. $a_i\ge 1$).
570: \end{defi}
571:
572: We recall some key properties:
573: \begin{itemize}
574: \item[(1)]Let $Y$ be a proper rationally chain connected variety.
575: Then any two closed points are contained in a connected curve
576: with rational irreducible components, \cite{kollar} IV.3.5.1.
577: \item[(2)]
578: If the ground field $k$ is uncountable
579: then $Y$ is rationally connected (resp. rationally
580: chain connected) if any two very general closed points $y_1$
581: and $y_2$ are contained in an irreducible rational curve
582: (resp. connected curve with rational irreducible components),
583: \cite{kollar} IV.3.6.
584: \item[(3)] Let $Y$ be a smooth proper rationally connected variety
585: and $y_1,\ldots,y_m$ points in $Y$. Then
586: there exists a very free morphism $f:\bP^1 \ra Y$
587: such that $y_1,\ldots,y_m \in f(\bP^1)$. We may take $f$
588: to be an immersion if $\dim(Y)=2$ and an embedding if $\dim(Y)\ge 3$,
589: \cite{kollar} IV.3.9.
590: \item[(4)]
591: A smooth variety $Y$ is rationally connected if
592: it is rationally chain connected, \cite{kollar} IV.3.10.3.
593: \item[(5)]
594: Let $\pi:\cY \ra B$ be a proper equidimensional morphism over
595: an irreducible base. If the generic fiber of $\pi$ is
596: rationally chain
597: connected then every fiber is rationally {\em chain} connected,
598: \cite{kollar} IV.3.5.2.
599: \item[(6)]
600: If $\pi:\cY \ra B$ is a smooth morphism then the locus
601: $$\{ b\in B: \cY_b \text{ is rationally connected } \}$$
602: is open, \cite{kollar} IV.3.11.
603: \end{itemize}
604: Since Properties (5) and (6) are local on the base,
605: they also hold for good models (which are only assumed to
606: be algebraic spaces over $B$).
607:
608:
609: \begin{exam}
610: \label{exam:cubic}
611: Property (5) does not guarantee that every
612: fiber is rationally connected: Consider the
613: family of cubic surfaces
614: $$
615: \cX:=\{(w,x,y,z;t): x^3+y^3+z^3=tw^3 \}\ra \bA^1_t.
616: $$
617: The generic fiber is rationally connected but the fiber
618: $\cX_0$ is a cone over an elliptic curve, which
619: is not rationally connected.
620: \end{exam}
621:
622:
623:
624: \subsection{Producing sections through prescribed points}
625: Theorem~\ref{theo:great}, when combined with the machinery of
626: Section~\ref{sect:rat}, has the following
627: important consequence:
628:
629: \begin{theo}[\cite{kollar} IV.6.10, \cite{KMM} 2.13]
630: \label{theo:ghsk}
631: Let $X$ be a smooth projective
632: rationally connected variety over the
633: function field of a curve. Given a projective model $\pi:\cX \ra B$,
634: a finite collection of points $\{b_i \}_{i\in I}$ such that each
635: $\cX_{b_i}$ is smooth,
636: and points $x_i \in \cX_{b_i}$, there exists a section
637: $s:B\ra \cX$ such that $s(b_i)=x_i$, for each $i\in I$.
638: \end{theo}
639:
640:
641:
642:
643: It is natural to wonder whether we can relax the hypothesis that
644: the fibers $\cX_{b_i}$ be smooth. For simplicity, assume
645: that the total space $\cX$ of our model is regular; this
646: can always be achieved by resolving singularities. Then
647: for each section $s$, $s(b)\in \cX_b$ is necessarily a smooth point;
648: otherwise, the intersection multiplicity of the section with
649: $\cX_b$ would be $>1$. In light of this, the most optimistic
650: generalization of Theorem~\ref{theo:ghsk} would be:
651: \begin{conj}
652: Let $X$ be a smooth projective
653: rationally connected variety over the
654: function field of a curve. Given a regular model $\pi:\cX \ra B$,
655: a finite collection of points $\{b_i \}_{i\in I}\subset B$
656: and smooth points $r_i \in \cX_{b_i}$, there exists a section
657: $s\,:\,B\ra \cX$ such that $s(b_i)=r_i$ for each $i\in I$.
658: \end{conj}
659:
660: Applying this to the iterated blowups as described
661: in Proposition~\ref{prop:jets},
662: we obtain:
663: \begin{conj}
664: \label{conj:main}
665: A smooth rationally connected variety
666: over the function field of a curve
667: satisfies weak approximation.
668: \end{conj}
669:
670: \
671:
672: We outline the main issues in the proof of Theorem~\ref{theo:main};
673: details are given in Section~\ref{sect:main-proof}.
674: By Proposition~\ref{prop:jets}, we are reduced to proving the
675: existence of a section passing through specific smooth points $r_i$
676: of {\em singular} fibers of the iterated blow-up.
677: Theorem~\ref{theo:ghsk} does not immediately imply this, but it does
678: guarantee a section $\sigma$ passing through some point
679: $x_i$ of each of these fibers.
680: Property (5) of rationally connected varieties from
681: Section~\ref{sect:rat} guarantees the existence
682: of some chain $T_i$ of rational curves in the corresponding fiber
683: joining $x_i$ and $r_i$.
684: The difficulty is to choose these so that
685: $C:=\sigma(B) \cup_{i\in I} T_i$ deforms to a
686: section containing $r_i$, for each $i\in I$.
687: In particular, it is necessary that $C$ intersect the
688: fibral exceptional divisor $E_{i,N}$ containing $r_i$ in one point
689: and not intersect the other components of the corresponding fiber;
690: this constrains the homology class of $T_i$.
691: Furthermore, we must describe each $T_i$ explicitly so the deformation
692: space of $C$ can be analyzed.
693:
694:
695:
696:
697:
698: \section{Deformation theory}
699: We continue to work over an algebraically closed ground field
700: of characteristic zero.
701: In this section, a {\em curve} is a connected reduced scheme
702: of dimension one.
703: \subsection{Combs}
704: Recall the {\em dual graph} associated with a nodal curve $C$:
705: Its vertices are indexed by the irreducible components of $C$
706: and its edges are indexed by
707: the intersections of these components.
708: \begin{defi}
709: A projective nodal curve $C$ is {\em tree-like} if
710: \begin{itemize}
711: \item{each irreducible component of $C$ is smooth;}
712: \item{the dual graph of $C$ is a tree.}
713: \end{itemize}
714: \end{defi}
715:
716: We shall require a slight generalization of the standard notion
717: of a comb (cf. \cite{kollar}):
718: \begin{defi}
719: A {\em comb with $m$ broken teeth} is a projective nodal curve $C$ with
720: $m+1$ subcurves
721: $D, T_1, \ldots, T_m$ such that
722: \begin{itemize}
723: \item{$D$ is smooth and irreducible;}
724: \item{$T_{\ell}\cap T_{\ell'}=\emptyset$, for all $\ell\neq \ell'$;}
725: \item{each $T_{\ell}$
726: meets $D$ transversally
727: in a single point; and }
728: \item{each $T_{\ell}$ is a chain of $\bP^1$'s.}
729: \end{itemize}
730: Here $D$ is called the {\em handle} and the
731: $T_{\ell}$ the {\em broken teeth}.
732: \end{defi}
733:
734:
735:
736: \begin{figure}[h]
737: %\centerline{\psfig{figure=weakfig3a}}
738: \centerline{\includegraphics{weakfig3a}}
739: \caption{A comb with five broken teeth}\label{fig3}
740: \end{figure}
741:
742:
743:
744: \subsection{Vector bundle lemmas}
745: \label{sect:vblem}
746: Let $C$ be a smooth curve and $\cV$ a vector
747: bundle on $C$. Given a collection of distinct
748: points $\mfq=\{q_1,\ldots,q_m\}\subset C$ and one-dimensional
749: subspaces of the fibers $\xi_{q_{\ell}} \subset \cV_{q_{\ell}},\ell=1,\ldots,m$,
750: there exists a rank-one subbundle $\cL \subset \cV$ with fiber at
751: $q_{\ell}$ equal to $\xi_{q_{\ell}}$, $\ell=1,\ldots,m$.
752: The extension
753: $$
754: 0 \longrightarrow \cL \longrightarrow \cL\otimes \cO_C(\mfq)
755: \longrightarrow \cL \otimes (\cO_C(\mfq))_{\mfq} \longrightarrow 0
756: $$
757: induces
758: $$
759: 0 \longrightarrow \cV \longrightarrow \cV' \longrightarrow Q
760: \longrightarrow 0,
761: $$
762: where $Q:=(\cL\otimes \cO_C(\mfq))_{\mfq}$
763: is supported on $\mfq$ and has
764: length one at each $q_{\ell}$.
765: This extension depends on the $q_{\ell}$ and $\xi_{q_{\ell}}$ but not
766: on $\cL$. The saturation of $\cL$ in $\cV'$ is isomorphic
767: to $\cL\otimes \cO_C(\mfq)$.
768: \begin{lemm} \label{lemm:sat}
769: Retain the notation introduced above.
770: A subbundle $\cU \subset \cV$ is also a subbundle of $\cV'$ if
771: $\xi_{q_{\ell}}\cap \cU_{q_{\ell}}=0$ for each $\ell$.
772: \end{lemm}
773: \begin{proof}
774: We have exact sequences
775: \begin{eqnarray*}
776: & & 0 \ra \Hom(\cL\otimes \cO_C(\mfq),\cV) \ra \Hom(\cL,\cV) \ra
777: \Ext^1(Q,\cV) \\
778: & & 0 \ra \iHom(\cL\otimes \cO_C(\mfq),\cV) \ra \iHom(\cL,\cV) \ra
779: \iExt^1(Q,\cV) \ra 0
780: \end{eqnarray*}
781: and the extension class $\eta_{\cV'} \in \Ext^1(Q,\cV)$
782: is the image of the inclusion $\cL \hookrightarrow \cV$ under
783: the connecting homomorphism. Since $\cL$ is saturated in
784: $\cV$ at $q_{\ell}$, $\eta_{V'}$ localizes to a nonzero
785: element of $\iExt^1(Q,\cV)_{q_{\ell}}$
786: for each $\ell$.
787:
788: Since $\cU\subset \cV$ is a subbundle, $\cV/\cU$ has no torsion and
789: thus is locally free.
790: The class $\eta_{\cV'}$ naturally induces an
791: extension
792: $$0 \longrightarrow \cV/\cU \longrightarrow \cV'/\cU \longrightarrow Q \ra 0,$$
793: classified by $\eta_{\cV'/\cU}\in \Ext^1(Q,\cV/\cU)$,
794: the image of the
795: composition
796: $$
797: \cL \hookrightarrow \cV \twoheadrightarrow \cV/\cU
798: $$
799: under the connecting homomorphism. Our hypothesis guarantees
800: that $\cL$ is a subbundle of $\cV/\cU$ near $q_{\ell}$, hence
801: $\eta_{\cV'/\cU}$ localizes to a nonzero element of
802: $\iExt^1(Q,\cV/\cU)_{q_{\ell}}$ for each $\ell$.
803: It follows that $\cV'/\cU$ is torsion-free, so $\cU$ is a subbundle.
804: \end{proof}
805:
806:
807: \begin{lemm}[\cite{GHS}, Lemma 2.5]
808: \label{lemm:vb}
809: Retain the notation introduced above. Fix an integer $N$ and
810: a vector bundle $\cV$. Then there exist points
811: $q_1, \ldots, q_m$ and one-dimensional subspaces $\xi_{q_\ell}\subset \cV_{q_{\ell}}$, $\ell=1,\ldots,m$, such that
812: $$H^1(\cV'\otimes \cO_C(-w_1-\ldots-w_N))=0$$
813: for any points $w_1,\ldots,w_N \in C$.
814: \end{lemm}
815:
816: Our next lemma is well known (cf. \cite{NS}, Section 2) but we
817: provide a proof for the convenience of the reader:
818: \begin{lemm}
819: \label{lemm:tr-like}
820: Let $C$ be a tree-like curve and $\cV$ a vector bundle on $C$.
821: If for each irreducible component $C_{\ell}$ of $C$
822: the restriction $\cV\otimes \cO_{C_{\ell}}$ is globally generated
823: then $\cV$ is globally generated.
824: Furthermore,
825: $$H^1(C,\cV)\ra \oplus_{\ell} H^1(\cV\otimes \cO_{C_{\ell}})$$
826: is an isomorphism.
827: \end{lemm}
828:
829: \begin{proof}We do induction on the number of
830: irreducible components; the case of one component
831: is trivial. Otherwise, express $C$ as a union $D\cup D^c$,
832: where $D$ is irreducible with connected complement in $C$
833: and $D^c=\overline{C\setminus D}$ is tree-like. Let $q$
834: be the node of $C$ joining $D$ and $D^c$,
835: $$
836: g:C':=D\amalg D^c\ra C
837: $$
838: the partial normalization of $C$ at $q$, and
839: $r,r^c$ the points of $C'$
840: with $g(r)=g(r^c)=q$.
841: The descent data for $\cV$ consist of the pullback $g^*\cV$
842: and an isomorphism
843: $$\phi:(g^*\cV)_r \ra (g^*\cV)_{r^c}$$
844: induced by identifications of fibers
845: $$
846: (g^*\cV)_r\simeq \cV_q \simeq (g^*\cV)_{r^c}.
847: $$
848: Recall the exact sequences relating the cohomology
849: of $\cV$ and $g^*\cV$:
850: $$
851: 0 \ra H^0(C,\cV) \ra
852: H^0(C',g^*\cV) \ra
853: (g^*\cV)_r\oplus (g^*\cV)_{r^c}
854: \stackrel{(-\phi,\mathrm{Id})}{\ra}
855: (g^*\cV)_{r^c} \ra 0
856: $$
857: $$
858: H^0(C',g^*\cV) \ra
859: (g^*\cV)_r\oplus (g^*\cV)_{r^c} \ra
860: H^1(C,\cV) \ra H^1(C',g^*\cV) \longrightarrow 0.
861: $$
862:
863: By the inductive hypothesis, $g^*\cV$ is globally generated
864: on $C'$. Since $r$ and $r^c$ are on different connected
865: components of $C'$, the second exact sequence guarantees
866: that $H^1(C,\cV)\ra H^1(C',g^*\cV)$ is injective;
867: the cohomology statement follows.
868: Since $g^*\cV$ is globally generated, for each section over
869: $D$ there exists a section over $D^c$ compatible under
870: the isomorphism $\phi$, and vice versa. These compatible
871: pairs of sections descend to elements of $H^0(C,\cV)$.
872: Thus given $p\in D$ and $v\in \cV_p$, a section
873: $t\in H^0(D,\cV\otimes \cO_D)$ with $t(p)=v$
874: extends to a section over $C$.
875: \end{proof}
876:
877:
878:
879:
880: \subsection{Analysis of normal bundles}
881: \label{sect:nb}
882: We describe
883: the normal bundle of a nodal curve immersed in
884: a smooth algebraic space. Our main references are
885: Section 2 of \cite{GHS} and Section 6 of \cite{AK}.
886: See \cite{Art69} and \cite{kollar} I.5
887: for foundational results on Hilbert `schemes' of algebraic spaces,
888: \cite{kollar} II.1
889: for applications to morphisms of curves into spaces,
890: and \cite{dJS} for an extension of
891: Theorem~\ref{theo:great} to
892: positive characteristic using this machinery.
893:
894: If $C$ is a nodal curve imbedded into a smooth space $Y$
895: then $\cN_{C/Y}$ is defined as the dual to the kernel of the restriction
896: homomorphism of K\"ahler differentials
897: $\Omega^1_Y\otimes \cO_C \twoheadrightarrow \Omega^1_C;$
898: a local computation shows this is locally free.
899: First order deformations of $C\subset Y$ are given by
900: $H^0(C,\cN_{C/Y})$; obstructions are given by $H^1(C,\cN_{C/Y})$.
901:
902: Let $D$ be the union of irreducible components of
903: $C$ and $\mfq=\{q_1,\ldots,q_m\}$ the locus
904: where $D^c:=\overline{C\setminus D}$ meets $D$.
905: At a node of $C$, the tangent cone is a union of two one-dimensional
906: subspaces, the tangents to the transverse branches.
907: The tangent to $D^c$ at $q_{\ell}$ yields a one-dimensional subspace
908: $$\xi_{q_{\ell}}\subset (\cN_{D/Y})_{q_{\ell}}.$$
909: As in Section \ref{sect:vblem}, these induce a natural extension
910: $\cN'_{D/Y}$ of $\cN_{D/Y}$, which coincides with
911: the restriction to $D$ of the normal bundle to $C$ in $Y$
912: $$
913: 0 \lra \cN_{D/Y} \lra \cN_{C/Y}\otimes \cO_D \lra Q \lra 0.
914: $$
915: Here $Q$ is a torsion sheaf supported on $\mfq$, with length one
916: at each point $q_{\ell},\ell=1,\ldots,m$. Sections of
917: $\cN_{C/Y}\otimes \cO_D$ can be interpreted as sections of $\cN_{D/Y}$
918: with poles at the $q_{\ell}$ in the directions $T_{q_{\ell}}D^c$.
919:
920: We shall need a slight generalization: We continue to assume that
921: $C$ is a nodal curve and $Y$ is nonsingular. Let $f:C \ra Y$ denote
922: a closed immersion whose image is a nodal curve. The
923: restriction homomorphism
924: $f^*\Omega^1_Y \rightarrow \Omega^1_C$
925: is surjective and the dual to its kernel is still locally free.
926: This is denoted $\cN_f$ and coincides with $\cN_{C/Y}$ when $f$
927: is an embedding.
928: First order deformations of $f:C\ra Y$ are given by
929: $H^0(C,\cN_f)$; obstructions are given by $H^1(C,\cN_f)$.
930: This set-up differs from the standard deformation theory of morphisms
931: in that we ignore reparametrizations of $C$.
932: When $D$ is a union of irreducible
933: components of $C$ as above, then the analogous extension
934: takes the form (cf. Lemma 2.6 of \cite{GHS}):
935: $$
936: 0 \lra \cN_{f|_D} \lra \cN_f \otimes \cO_D \lra Q \lra 0.
937: $$
938:
939: This analysis gives the following infinitesimal smoothing criterion:
940: (cf. Lemma 2.6 of \cite{GHS}):
941: \begin{lemm} \label{lemm:smooth}
942: Retain the above notation. A first
943: order deformation $t\in H^0(C,\cN_f)$ smooths the node
944: $q_{\ell}$ if the restriction
945: $t|_D \in \cN_f\otimes \cO_D$ is nonzero in $Q_{q_{\ell}}$.
946: \end{lemm}
947:
948:
949: \begin{prop}
950: \label{prop:diag}
951: Let $Y$ be a smooth algebraic space, $E\subset Y$ a smooth subspace of
952: codimension one, $C$ a nodal curve, $D\subset C$ a union of irreducible
953: components of $C$, $D^c=\overline{D\setminus C}$, and
954: $\mfq=\{q_{\ell}\}=D\cap D^c$.
955: Let $f:C \rightarrow Y$ be an immersion with image a nodal
956: curve so that $f(D)\subset E$, $f(D\setminus C)\subset Y\setminus E$,
957: and $f(D^c)$ is transverse to
958: $E$ at each point of $f(\mfq)$.
959: If $g:D\ra E$ is the resulting immersion of $D$
960: into $E$ then
961: $\cN_g$ is saturated in $\cN_f\otimes \cO_D$
962: and we have the following diagram:
963:
964: \centerline{
965: \xymatrix{
966: & & 0 \ar[d] & 0\ar[d] & \\
967: 0\ar[r] &\cN_g\ar[r]\ar@{=}[d]& \cN_{f|_D}\ar[r]\ar[d] & g^*\cN_{E/Y} \ar[d]\ar[r] & 0 \\
968: 0\ar[r] &\cN_g\ar[r] & \cN_f\otimes \cO_D\ar[r]\ar[d]& g^*\cN_{E/Y}\otimes \cO_D(\mfq)\ar[d]\ar[r] & 0 \\
969: & 0\ar[r] & Q\ar[r]\ar[d] & g^*\cN_{E/Y}\otimes (\cO_D(\mfq))_{\mfq}\ar[d] \ar[r]& 0\\
970: & & 0 & 0 &
971: }}
972:
973: \end{prop}
974:
975: \begin{proof}
976: For each $\ell$ the composition
977: $$T_{q_{\ell}}D^c \ra T_{q_{\ell}}Y \ra
978: T_{q_{\ell}}Y/ T_{q_{\ell}}D$$
979: determines a one-dimensional subspace
980: $\xi_{\ell}\subset (\cN_{f|_D})_{q_{\ell}}$.
981: The transversality hypothesis implies
982: $(\cN_g)_{q_{\ell}}\cap \xi_{q_{\ell}}=0$ in $(\cN_{f|_D})_{q_{\ell}}$.
983: Using Lemma \ref{lemm:sat}, we conclude that $\cN_{f|_D}$ is
984: a subbundle in $\cN_f\otimes \cO_D$, so the quotient
985: $$R=\left(\cN_f\otimes \cO_D\right) / \cN_{f|_D}$$
986: is locally free. This sheaf arises as an extension
987: \begin{equation}
988: \label{eq:extend}
989: 0 \longrightarrow g^*\cN_{E/Y} \longrightarrow R \longrightarrow
990: Q \longrightarrow 0,
991: \end{equation}
992: where $Q$ is supported on $\mfq$ with length
993: one at each $q_{\ell}$. We may therefore identify extension
994: (\ref{eq:extend}) with the tensor product of $g^*\cN_{E/Y}$ with
995: $$
996: 0 \longrightarrow \cO_D \longrightarrow \cO_D(\mfq)
997: \longrightarrow \left(\cO_D(\mfq)\right)_{\mfq} \longrightarrow 0.
998: $$
999: \end{proof}
1000:
1001:
1002:
1003:
1004:
1005: \begin{prop}
1006: \label{prop:tree}
1007: Let $C$ be a tree-like curve, $Y$ a smooth algebraic space, and
1008: $f:C \ra Y$ an immersion with nodal image.
1009: Suppose that for each irreducible component $C_\ell$ of $C$,
1010: $H^1(C_{\ell},\cN_f\otimes \cO_{C_{\ell}})=0$ and
1011: $\cN_f\otimes \cO_{C_{\ell}}$
1012: is globally generated.
1013: Then $f:C \ra Y$ deforms to an immersion of a smooth curve
1014: into $Y$.
1015:
1016: Suppose furthermore that
1017: $\mfp=\{p_1,\ldots,p_w \}\subset C$
1018: is a collection of smooth points
1019: such that for each component $C_{\ell}$
1020: $H^1(\cN_f\otimes \cO_{C_{\ell}}(-\mfp))=0$
1021: and the sheaf $\cN_f\otimes \cO_{C_{\ell}}(-\mfp)$
1022: is globally generated.
1023: Then $f:C \ra Y$ deforms to an immersion of a smooth curve
1024: into $Y$ containing $f(\mfp)$.
1025: \end{prop}
1026:
1027: \begin{proof}
1028: Our argument is similar to the constructions of Section 2 of
1029: \cite{GHS} and Lemma 65 of \cite{AK}.
1030: Lemma~\ref{lemm:tr-like} implies that $H^1(C,\cN_f)=0$ and $\cN_f$
1031: is globally generated. Hence
1032: the space of maps is unobstructed and every first-order
1033: deformation of $f$ lifts to an actual deformation.
1034: Global generation implies the existence of
1035: $t\in H^0(C,\cN_f)$ so that,
1036: for each component $C_\ell$, the image of $t$ in
1037: $$
1038: H^0(C_{\ell},Q(C_{\ell})), \quad
1039: Q(C_{\ell}):=\left(\cN_f\otimes \cO_{C_{\ell}}\right)/\cN_{f|_{C_{\ell}}}
1040: $$
1041: is nonzero at each point of the support of $Q(C_{\ell})$.
1042: The first-order deformation $t$ smooths each node of $C$ by
1043: Lemma~\ref{lemm:smooth}.
1044:
1045: For the second part, consider those maps with image containing
1046: $f(\mfp)$. Our cohomology assumption
1047: guarantees that this space is unobstructed;
1048: in addition, $\cN_f\otimes \cO_C(-\mfp)$ is globally generated.
1049: Hence there exists a $u\in H^0(\cN_f\otimes \cO_C(-\mfp))$ so that,
1050: for every component $C_\ell$, the image of $u$ in $Q(C_{\ell})$
1051: is nonzero at each point of its support.
1052: Note that
1053: $$
1054: (\cN_f\otimes \cO_{C_{\ell}}(-\mfp))/
1055: (\cN_{f|_{C_{\ell}}}\otimes \cO_{C_{\ell}}(-\mfp))=
1056: (\cN_f\otimes \cO_{C_{\ell}})/\cN_{f|_{C_{\ell}}},
1057: $$
1058: since the quotient is a torsion sheaf with support disjoint from $\mfp$.
1059: Hence the first-order deformation $u$ smooths each node of $C$ and
1060: contains $f(\mfp)$.
1061: \end{proof}
1062:
1063:
1064: \section{Proof of the main theorem}
1065: \label{sect:main-proof}
1066:
1067: The theorem is well known when $d=\dim(X)=1$. The
1068: only smooth proper rationally connected curve is $\bP^1$,
1069: which satisfies weak approximation. We may therefore assume
1070: $d\ge 2$.
1071:
1072:
1073: Recall the set-up of Proposition~\ref{prop:jets}:
1074: It suffices to show that for
1075: each integer $N$, finite set $\{b_i\}_{i\in I}\subset B\setminus S$,
1076: and collection of $N$-jet data $J$ supported in the
1077: fibers over $\{b_i\}_{i\in I}$, there exists
1078: a section in the iterated blowup $\cX(J)$ passing through
1079: prescribed points $r_i \in E_{i,N}\setminus E_{i,N-1}$.
1080:
1081: We proceed by induction on $N$; the base case $N=0$ is essentially
1082: Theorem~\ref{theo:ghsk}. However, our assumptions are slightly weaker:
1083: We are not assuming $\cX$ is a scheme.
1084:
1085: The total space of $\cX$ is smooth along the fibers $\cX_{b_i}, i\in I$,
1086: so we may resolve the singularities of $\cX$ without altering these fibers.
1087: Theorem~\ref{theo:great} gives a section $\sigma$ of $\pi$.
1088: Let $q_i=\sigma(b_i)$ for each $i\in I$; let $I'\subset I$
1089: (resp. $I''\subset I$) denote those indices with $q_i\neq r_i$
1090: (resp. $q_i =r_i$).
1091:
1092: We shall construct a comb $C$ with handle $\sigma(B)$ and smooth
1093: teeth $T_1,\ldots,T_m$ and an immersion $f:C \ra \cX$
1094: with nodal image so that:
1095: \begin{itemize}
1096: \item{the $T_{\ell}$ are free rational curves
1097: in distinct smooth fibers of $\pi$;}
1098: \item{for each $i\in I'$, there is a tooth $T_i$
1099: containing $r_i$ as a smooth point;}
1100: \item{let $\mfr$ denote the sum of the points of $C$
1101: mapping to the $r_i,i\in I$; then the restriction of
1102: $\cN_f\otimes \cO_C(-\mfr)$
1103: to each irreducible component of $C$
1104: is globally generated and has no higher cohomology.}
1105: \end{itemize}
1106: We emphasize that $f$ can be taken to be an
1107: embedding if $d>2$.
1108:
1109: Proposition~\ref{prop:tree} implies $f:C\ra \cX$
1110: admits a deformation $\tilde{f}:\tilde{C}\ra \cX$,
1111: where $\tilde{C}$ is smooth and $\tilde{f}(\tilde{C})$ contains
1112: each of the $r_i$. Since all the teeth $T_1,\ldots,T_m$
1113: are contained in fibers of $\cX\ra B$,
1114: $C$ intersects the generic fiber
1115: in one point. Thus the deformed curve $\tilde{C}$ also
1116: meets the generic fiber in one point and hence is a section
1117: of $\cX \ra B$.
1118:
1119: Here are the details of the construction. For each $\ell$
1120: $$f_{\ell}:T_{\ell} \ra \cX_{b_{\ell}}$$
1121: is a free rational curve with nodal image,
1122: so that $\sigma(b_{\ell})\in f_{\ell}(T_{\ell})$ as a smooth
1123: point. For $\ell=1,\ldots,|I'|$ we choose these so that
1124: $r_i$ is contained in the image as a smooth point.
1125: For $\ell=|I'|+1,\ldots,m$, we choose these in generic fibers
1126: of good reduction with generic tangent directions
1127: $\xi_{\ell} \subset T_{\sigma(b_{\ell})}\cX_{b_{\ell}}$
1128: satisying the hypotheses of Lemma~\ref{lemm:vb},
1129: so that the extension
1130: $$
1131: 0 \lra \cN_{\sigma(B)} \lra \cN_f\otimes\cO_{\sigma(B)} \lra Q(\sigma(B)) \lra 0
1132: $$
1133: is globally-generated and has no higher cohomology, even
1134: after twisting by $\cO_{\sigma(B)}(-\sum_{i\in I''}r_i)$.
1135:
1136: \
1137:
1138:
1139: We next address the inductive step.
1140: Let $J'$ denote order-$(N-1)$ truncation
1141: of $J$, i.e., if
1142: $$
1143: j_i: \Spec(\cO_{B,b_i}/\mfm_{B,b_i}^{N+1}) \ra
1144: \cX\times_B\Spec(\cO_{B,b_i}/\mfm_{B,b_i}^{N+1})
1145: $$
1146: then
1147: $$j'_i=j_i|\Spec(\cO_{B,b_i}/\mfm_{B,b_i}^N).$$
1148: The inductive hypothesis applied to $J'$ guarantees the
1149: existence of a section $s':B\ra \cX(J')$ passing through arbitrary
1150: points
1151: $$
1152: r'_i\in (E'_{i,N-1}\setminus E'_{i,N-2})\subset \cX(J').
1153: $$
1154: Specifically,
1155: we choose $s'$ so that it has jet data $J'$ over the
1156: points $\{b_i\}_{i\in I}$. Let $\sigma:B \ra \cX(J)$ denote
1157: the proper transform of $s'$ in $\cX(J)$. By construction,
1158: $\sigma$ meets $\cX(J)_{b_i}$ in a point
1159: $q_i\in E_{i,N}\setminus E_{i,N-1}$ for
1160: each $i$. Our goal is to find a section $s:B \ra \cX(J)$
1161: such that for each $i\in I$ and $r_i\in E_{i,N}\setminus E_{i,N-1}$,
1162: $s(b_i)=r_i$.
1163: Again, $I'\subset I$
1164: (resp. $I''\subset I$) denotes those indices with $q_i\neq r_i$
1165: (resp. $q_i =r_i$).
1166:
1167: Next, we construct a comb $C$ with handle $\sigma(B)$ and
1168: broken teeth $T_1,\ldots,T_m$ and an immersion $f:C \ra \cX(J)$
1169: with nodal image so that:
1170: \begin{itemize}
1171: \item{for each $i\in I'$,
1172: there is a broken tooth $T_i$
1173: mapped to $\cX(J)_{b_i}$ and containing $r_i$;}
1174: \item{$C$ is smoothly embedded at $r_i$ for each $i\in I'$, so there is a unique
1175: component $T_{i,N}\subset C$ containing $r_i$;}
1176: \item{the remaining broken teeth $T_{|I'|+1},\ldots,T_m$
1177: are free rational curves contained
1178: in generic fibers of $\cX(J)\ra B$ of good reduction;}
1179: \item{the restriction of $\cN_f\otimes \cO_C(-\mfr)$
1180: to each irreducible component
1181: is globally generated and has no higher cohomology.}
1182: \end{itemize}
1183: Again, $f$ can be taken to be an
1184: embedding if $d>2$.
1185:
1186: Proposition~\ref{prop:tree} implies that $f:C\ra \cX(J)$
1187: admits a deformation $\tilde{f}:\tilde{C}\ra \cX(J)$,
1188: where $\tilde{C}$ is smooth and $\tilde{f}(\tilde{C})$ contains
1189: each of the $r_i$. The image $\tilde{f}(\tilde{C})$
1190: is the desired section of $\cX(J) \ra B$.
1191:
1192: We start by describing the teeth $T_i$ with $i\in I'$.
1193: Recall from Section \ref{sect:iter}
1194: that $E_{i,N}\simeq \bP^d$ and $E_{i,N}\cap E_{i,N-1}$ is a hyperplane
1195: section in this $\bP^d$. Let $T_{i,N}$ denote the unique line joining
1196: $r_i$ to $q_i=\sigma(b_i)$; let $q_{i,N-1}$ denote the intersection of
1197: this line with $E_{i,N-1}$. We have
1198: $$E_{i,N-1}\simeq \mathrm{Bl}_{r_{i,N-1}}\bP^d$$
1199: with exceptional divisor $E_{i,N-1} \cap E_{i,N}\simeq \bP^{d-1}$;
1200: there is a unique line in $\bP^d$ containing $r_{i,N-1}$ whose
1201: proper transform $T_{i,N-1}\subset E_{i,N-1}$ meets $q_{i,N-1}$.
1202: Let $q_{i,N-2}$ denote the intersection of this line with $E_{i,N-2}$.
1203: Continuing in this way, we obtain a sequence of embedded
1204: smooth rational curves
1205: $$T_{i,n}\subset E_{i,n}\simeq \mathrm{Bl}_{r_{i,n}}\bP^d, \quad n>0,$$
1206: each the proper transform of a line meeting $r_{i,n}$.
1207: Let $q_{i,0}$ denote the intersection of $T_{i,1}$ with
1208: $E_{i,0}$, which is a point in the exceptional divisor of
1209: $$E_{i,0}=\mathrm{Bl}_{r_{i,0}}\ra \cX_{b_i}.$$
1210: Let $g_{i,0}:T_{i,0} \rightarrow E_{i,0}$ be a free
1211: rational curve, immersed so that the image is a nodal curve,
1212: with $q_{i,0}\in g_{i,0}(T_{i,0})$
1213: as a smooth point. Property (3) of rationally connected varieties
1214: gives such a curve; $g_{i,0}$ can be taken to be an embedding when $d>2$.
1215: Let $f_{i,0}$ denote the composition of $g_{i,0}$ with the
1216: inclusion $E_{i,0}\subset \cX(J)$, and
1217: $$f_i:T_i=T_{i,0}\cup \ldots \cup T_{i,N}\lra \cX(J)$$
1218: the resulting map of the broken tooth into $\cX(J)$.
1219:
1220:
1221: \begin{figure}[h]
1222: \centerline{\hskip7cm \includegraphics{weakfig2a}}
1223: \caption{Attaching broken teeth and moving the section}\label{fig2}
1224: \end{figure}
1225:
1226:
1227: We verify the normal bundle conditions for the components
1228: of these teeth. First, observe that
1229: \begin{eqnarray*}
1230: \cN_{T_{i,n}/E_{i,n}}&\simeq&\begin{cases}
1231: \cO_{\bP^1}(+1)^{d-1} & \text{ for } n=N \\
1232: \cO_{\bP^1}^{d-1} & \text{ for } n=1,\ldots, N-1\\
1233: \end{cases} \\
1234: \cN_{g_{i,0}} &\simeq &
1235: \oplus_{w=1}^{d-1} \cO_{\bP^1}(a_w), \quad a_w \ge 0;
1236: \end{eqnarray*}
1237: the $a_w$ are nonnegative because $\cN_{g_{i,0}}$ is a quotient
1238: of $g_{i,0}^*T_{E_{i,0}}$, which is nonnegative.
1239: Fibers of $\cX(J)\ra B$ restrict to the zero divisor on each $T_{i,n}$
1240: and $\sum_{n=0}^N E_{i,n}$ is equivalent to the class of a fiber,
1241: hence
1242: $$
1243: E_{i,n}|T_{i,n}=(-\sum_{n'\neq n} E_{i,n'})|T_{i,n}.
1244: $$
1245: It follows that
1246: $$
1247: \cN_{E_{i,n}/\cX(J)}\otimes \cO_{T_{i,n}} =
1248: \cO_{\cX(J)}(E_{i,n})\otimes \cO_{T_{i,n}} \simeq
1249: \begin{cases}
1250: \cO_{\bP^1}(-1) & \text{ for } n=N \\
1251: \cO_{\bP^1}(-2) & \text{ for } n=1, \ldots, N-1\\
1252: \cO_{\bP^1}(-1) & \text{ for } n=0.
1253: \end{cases}
1254: $$
1255: For $n>0$ we have the exact sequence
1256: $$
1257: 0 \lra \cN_{T_{i,n}/E_{i,n}} \lra \cN_{T_{i,n}/\cX(J)} \lra
1258: \cN_{E_{i,n}/\cX(J)}\otimes \cO_{T_{i,n}} \lra 0,
1259: $$
1260: which splits in our situation. Therefore, we find
1261: $$
1262: \cN_{T_{i,n}/\cX(J)}\simeq
1263: \begin{cases}
1264: \cO_{\bP^1}(+1)^{d-1}\oplus \cO_{\bP^1}(-1) & \text{ for } n=N \\
1265: \cO_{\bP^1}^{d-1}\oplus \cO_{\bP^1}(-2) & \text{ for } n=1,\ldots, N-1
1266: \end{cases}.
1267: $$
1268: For $n=0$ we have
1269: $$
1270: 0 \lra \cN_{g_{i,0}} \lra \cN_{f_{i,0}} \lra
1271: g_{i,0}^*\cN_{E_{i,0}/\cX(J)} \lra 0,
1272: $$
1273: which implies
1274: $$\cN_{f_{i,0}}\simeq
1275: \oplus_{w=1}^{d-1} \cO_{\bP^1}(a_w)\oplus \cO_{\bP^1}(-1).$$
1276:
1277:
1278: On first examination, the negative summands would make it hard to satisfy
1279: the hypotheses of Proposition~\ref{prop:tree}. However, the nodes
1280: in the broken teeth give enough positivity to overcome the negative
1281: factors. We use Proposition~\ref{prop:diag} to analyze
1282: the relationship between the normal bundles to the $T_{i,n}$ and
1283: the restriction to the normal bundle of the broken comb to these
1284: components. When $n>0$, we have an exact sequence
1285: $$
1286: 0 \lra \cN_{T_{i,n}/\cX(J)} \lra \cN_f\otimes \cO_{T_{i,n}}
1287: \lra Q(T_{i,n}) \lra 0,
1288: $$
1289: where $Q(T_{i,n})$ is a torsion sheaf, supported at the nodes of
1290: $C$ on $T_{i,n}$.
1291: However, the positive summands of $\cN_{T_{i,n}/\cX(J)}$ are saturated
1292: in $\cN_f$; only the negative summand fails to be saturated.
1293: When $n=1,\ldots,N-1$, the negative summand is isomorphic
1294: to $\cO_{\bP^1}(-2)$ and
1295: $Q(T_{i,n})$ has length two, so the saturation
1296: is $\cO_{\bP^1}$. When $n=N$, the negative summand is isomorphic to
1297: $\cO_{\bP^1}(-1)$ and $Q(T_{i,N})$ has length two and
1298: support $\{r_i,q_{i,N-1}\}$, so the saturation
1299: is $\cO_{\bP^1}$.
1300: When $n=0$ we have
1301: $$
1302: 0 \lra \cN_{f_{i,0}} \lra \cN_{f_i}\otimes \cO_{T_{i,0}}
1303: \lra Q(T_{i,0}) \lra 0,
1304: $$
1305: with $Q(T_{i,0})$ of length one and supported at $q_{i,0}$.
1306: The negative summand of $\cN_{f_{i,0}}$ is isomorphic to
1307: $\cO_{\bP^1}(-1)$, so the extension above induces
1308: $$
1309: 0 \ra \cO_{\bP^1}(-1) \ra \cO_{\bP^1} \ra Q(T_{i,0}) \ra 0,
1310: $$
1311: i.e., the saturation of the negative factor is $\cO_{\bP^1}$.
1312:
1313: To summarize, we have shown
1314: $$
1315: \cN_f\otimes \cO_{T_{i,n}} =\begin{cases}
1316: \cO_{\bP^1}(+1)^d & \text{ for } n=N \\
1317: \cO_{\bP^1}^d & \text{ for } n=1,\ldots, N-1 \\
1318: \oplus_{w=1}^{d-1} \cO_{\bP^1}(a_w)\oplus \cO_{\bP^1}, \quad a_w\ge 0&
1319: \text{ for } n=0
1320: \end{cases},
1321: $$
1322: so the hypotheses of Proposition~\ref{prop:tree}
1323: hold for the broken teeth $T_i$.
1324:
1325:
1326: For $\ell=|I'|+1,\ldots,m$ we take
1327: $$f_{\ell}:T_{\ell} \ra \cX_{b_{\ell}}$$
1328: to be free rational curves, immersed in generic fibers of
1329: good reduction so that the images are nodal,
1330: with $\sigma(b_{\ell})\in f_{\ell}(T_{\ell})$ as a smooth
1331: point. We choose these with generic tangent directions
1332: $\xi_{\ell} \subset T_{\sigma(b_{\ell})}\cX_{b_{\ell}}$
1333: so that Lemma~\ref{lemm:vb} guarantees
1334: $\cN_f\otimes \cO_{\sigma(B)}(-\sum_{i\in I''}r_i)$
1335: is globally-generated and has
1336: no higher cohomology.
1337:
1338:
1339:
1340:
1341: \bibliographystyle{smfplain}
1342: \bibliography{wa3}
1343:
1344:
1345: \end{document}
1346:
1347:
1348: