1: %; whizzy document
2: \documentclass[10 pt]{article}
3: \usepackage{amsmath,amsthm,amssymb,amscd}
4: \setcounter{tocdepth}{1}
5:
6: %theorems and the like
7:
8: %\swapnumbers
9:
10: \theoremstyle{plain}
11:
12: \newtheorem{thm}{Theorem}[]
13: %\newtheorem{thmm}[thm]{Main Theorem}
14: \newtheorem{pro}[thm]{Proposition}
15: \newtheorem{lem}[thm]{Lemma}
16: %\newtheorem{alm}[thm]{Ampleness lemma}
17: \newtheorem{cor}[thm]{Corollary}
18: \newtheorem{con}[thm]{Conjecture}
19:
20: \theoremstyle{definition}
21:
22: \newtheorem{dfn}[thm]{Definition}
23: \newtheorem{conv}[thm]{Convention}
24: %\newtheorem{dfncor}[thm]{Definition-Corollary}
25: %\newtheorem{dfnthm}[thm]{Definition-Theorem}
26: \newtheorem{exa}[thm]{Example}
27: \theoremstyle{remark}
28:
29: \newtheorem{exe}[thm]{Exercise}
30: \newtheorem{prb}[thm]{Problem}
31: \newtheorem{rem}[thm]{Remark}
32: %\newtheorem{hop}[thm]{Hope}
33: \newtheorem{que}[thm]{Question}
34:
35:
36: %definitions and such
37:
38: %operators
39:
40: \DeclareMathOperator{\rk}{rk}
41: \DeclareMathOperator{\ch}{ch}
42: \DeclareMathOperator{\td}{td}
43: \DeclareMathOperator{\Tr}{Tr}
44: \DeclareMathOperator{\ad}{ad}
45: %\DeclareMathOperator{\codim}{codim}
46: %\DeclareMathOperator{\Bir}{Bir}
47: %\DeclareMathOperator{\Aut}{Aut}
48: \DeclareMathOperator{\Hom}{Hom}
49: \DeclareMathOperator{\End}{End}
50: \DeclareMathOperator{\Spin}{Spin}
51: \DeclareMathOperator{\mult}{mult}
52: %\DeclareMathOperator{\Ima}{Im}
53: \DeclareMathOperator{\Ker}{Ker}
54: %\DeclareMathOperator{\Mov}{Mov}
55: \DeclareMathOperator{\Br}{Br}
56: \DeclareMathOperator{\Bs}{Bs}
57: \DeclareMathOperator{\CH}{CH}
58: \DeclareMathOperator{\Pic}{Pic}
59: \DeclareMathOperator{\Gal}{Gal}
60: %\DeclareMathOperator{\Proj}{Proj}
61: %\DeclareMathOperator{\Diff}{Diff}
62: \DeclareMathOperator{\Ins}{Int}
63: %\DeclareMathOperator{\Exc}{Exc}
64: \DeclareMathOperator{\Div}{\overline{Div}}
65: \DeclareMathOperator{\NE}{\overline{NE}}
66: \DeclareMathOperator{\NA}{\overline{NA}}
67: \DeclareMathOperator{\ME}{\overline{Mob}}
68: \DeclareMathOperator{\NM}{\overline{NM}}
69: \DeclareMathOperator{\Bl}{Bl}
70: %\DeclareMathOperator{\Def}{Def}
71: %\DeclareMathOperator{\DEF}{\underline{Def}}
72: %\DeclareMathOperator{\ISOM}{\underline{Isom}}
73: %\DeclareMathOperator{\Isom}{Isom}
74: %\DeclareMathOperator{\Spec}{Spec}
75: %\DeclareMathOperator{\Ext}{Ext}
76: %\DeclareMathOperator{\Sing}{Sing}
77:
78: %other
79:
80: %\newcommand{\h}{\mathcal{H}}
81: \renewcommand{\o}{\mathcal{O}}
82: \renewcommand{\l}{\mathcal{L}}
83: %\newcommand{\d}{\mathcal{D}}
84: %\newcommand{\m}{\overline{\mathcal{M}}}
85: \newcommand{\q}{\mathbb{Q}}
86: \renewcommand{\r}{\mathbb{R}}
87: \renewcommand{\a}{\mathbb{A}}
88: \newcommand{\p}{\mathbb{P}}
89: \newcommand{\z}{\mathbb{Z}}
90: \newcommand{\F}{\mathbb{F}}
91: \renewcommand{\c}{\mathbb{C}}
92: \newcommand{\cstar}{\c^\times}
93:
94: \begin{document}
95: \bibliographystyle{alpha}
96:
97: \title{Birational Geometry of 3-fold Mori Fibre Spaces}
98:
99: \author{
100: Gavin Brown\thanks{gavinb@maths.warwick.ac.uk}\\
101: Mathematics Institute\\
102: University of Warwick\\
103: Coventry, CV4 7AL, UK
104: \and
105: Alessio Corti\thanks{a.corti@dpmms.cam.ac.uk}\\
106: Dep.~of Pure Math.~and Math.~Stat.\\
107: University of Cambridge\\
108: Cambridge, CB3 0WB, UK\\
109: \and
110: Francesco Zucconi\thanks{zucconi@dimi.uniud.it}\\
111: DIMI, Universit\`a di Udine\\
112: via delle Scienze 208\\
113: 33100 Udine, Italy}
114:
115: \date{15 July, 2003}
116:
117: \maketitle
118:
119: \begin{abstract}
120: In this paper we study standard 3-fold conic bundles $X/\p^2$ and stable
121: cubic del Pezzo fibrations $X/\p^1$. These are the key examples of
122: 3-fold Mori fibre spaces (Mfs).
123: We begin systematically to chart the \emph{geography} of these
124: varieties, namely the classification of their deformation families
125: and discrete invariants.
126: Given a 3-fold Mfs $X/S$, we aim to understand the set of all Mfs
127: $Y/T$ with $X$ birational to $Y$. For example, we say that $X/S$ is
128: \emph{birationally rigid} if this is a 1-element set.
129: We state conjectures on the rigidity of Mfs, with the goal of
130: approaching optimal criteria.
131: Our main results are summarised in Tables~\ref{tab:1} and~\ref{tab:2}
132: and Figure~\ref{fig:geog}.
133: We want this to become your standard guidebook to Mori fibre spaces
134: and their birational geometry.
135: \end{abstract}
136: \pagebreak
137:
138: \tableofcontents
139:
140: \section{Introduction}
141: \label{sec:introduction}
142:
143:
144: \subsection{Mori fibre spaces}
145:
146: A \mbox{3-fold} Mori fibre space---we often use the acronym
147: Mfs---is a \mbox{3-fold} $X$ together with an extremal
148: contraction $f \colon X \to S$ of \emph{fibering type}.
149: This means that $X$ has $\q$-factorial terminal
150: singularities, $-K_X$ is ample on fibres, the relative Picard rank
151: $\rho = \rk N^1(X) -\rk N^1 (S)$ equals $1$, and $\dim S < \dim X$. This
152: structure is the higher-dimensional generalisation of minimally ruled
153: surfaces and $\p^2$.
154:
155: This paper is devoted to \emph{strict} \mbox{3-fold} Mori fibre
156: spaces, that is, we always assume $\dim S\geq 1$. There are two cases:
157: \emph{conic bundles} when $\dim S =2$ and \emph{del
158: Pezzo fibrations}---we write $dP_k$ fibrations when fibres are del
159: Pezzo surfaces of degree $k$---when $\dim S =1$. We also always assume that
160: the base variety $S$ is rational.
161:
162: \begin{dfn} Let $f \colon X\to S$, $g \colon Y \to T$ be Mori fibre spaces.
163: A birational map
164: \[
165: \varphi \colon X \dasharrow Y
166: \]
167: is \emph{square} if it maps a general fibre of $f$
168: isomorphically to a general fibre of $g$. We say that $X\to S$
169: and $Y \to T$ are \emph{square birational}, or \emph{square
170: equivalent}, when there is a square birational map between them.
171: \end{dfn}
172:
173: \subsection{Geography}
174:
175: Our first goal in this paper is to begin a systematic study of the
176: \emph{geography} of strict Mori fibre spaces. We focus on the case of
177: \emph{standard} conic bundles $X\to \p^2$ and \emph{stable} $dP_3$
178: fibrations $X\to \p^1$.
179:
180: Recall that a conic bundle $X \to S$ is standard if $X$ is
181: nonsingular; this in turn implies that $S$ is also nonsingular, and
182: the discriminant locus $\Delta \subset S$ is a nodal curve. Every
183: conic bundle is square birational to a standard conic bundle (it may
184: be necessary to blow up $S$). With qualifications, the assumption is not really
185: restrictive. In Section~\ref{sec:conic-bundles} of this paper we begin
186: a systematic study of standard conic bundles $X\to \p^2$. Necessarily,
187: $X$ embeds in a projectivised rank 3 vector bundle on $\p^2$, and we
188: learn how to write down the bundle and equations of $X$ explicitly. As
189: an illustration, we construct standard conic bundles for all
190: ramification data with a discriminant curve of degree 7. (There are 4
191: deformation families.) Our methods apply to higher degrees and to
192: surfaces other than $\p^2$; what we do here is a starting point
193: for future work.
194:
195: The definition of stable del Pezzo fibration is technical; in this
196: paper we almost always work with the slightly stronger condition that
197: $X$ be nonsingular. Again, a $dP$ fibration is always square
198: equivalent to a stable one.
199: In Section~\ref{sec:dp_3-fibrations} of this paper we
200: begin a systematic study of the geography of stable $dP_3$ fibrations
201: $X\to \p^1$. Necessarily, $X$ is a member of a linear system $|3M+nL|$
202: on a rational scroll $\F(0,a,b,c)$. We determine the invariants
203: $n,a,b,c$ such that a general element of the linear system is a stable
204: $dP_3$ fibration. We study the degree 3 case because it is the most
205: interesting, but our methods apply to degree 1 and 2.
206: Grinenko has studied del Pezzo fibrations of degree 1 and 2
207: in a series of recent papers \cite{grinenko:01a, grinenko:01b,
208: grinenko:01c, grinenko:00, gri02}, but he always works under the
209: assumption that the total space $X$ is Gorenstein. In the case of $dP$
210: fibrations of degree 1 and 2, this assumption is too restrictive.
211:
212: \subsection{Sarkisov links}
213: \label{sec:sarkisov-links}
214:
215: Both in the case of conic bundles and del Pezzo fibrations $X/S$, we study
216: in detail nonsquare Sarkisov links originating from $X/S$.
217: Recall that Sarkisov links are
218: the ``elementary'' building blocks of the birational geometry of Mfs:
219: any birational map between Mfs can be factored as a chain of Sarkisov
220: links. We usually, but not always, assume that $X$ is nonsingular and
221: sufficiently general in moduli.
222: To understand Sarkisov links, we are led to compute explicitly the
223: Mori cone of $X$ and the cone of mobile divisors on $X$. In our
224: examples $X$ has rank 2, so these cones
225: are 2-dimensional, and we just need to determine
226: the two edges in terms of explicit loci on $X$. The mobile cone of $X$
227: is partitioned into chambers (in our case these are just wedges)
228: corresponding to moves of the 2-ray game \cite{corti:00} starting with $X$.
229: We study these moves explicitly to see if the 2-ray
230: game leads to a Sarkisov link from $X\to S$ to a new Mori fibre space.
231: When $X$ is embedded in a scroll $\F$ over $S$ the moves are
232: often, but not always, induced by moves of $\F$.
233:
234:
235: \subsection{Rigid Mori fibre spaces and known criteria}
236:
237: \begin{dfn} \cite[Foreword]{MR2001f:14004}
238: The \emph{pliability} of a Mori fibre space $X \to S$ is the set
239: \[
240: \mathcal{P}(X/S) = \{ \text{Mfs} \; Y/T \mid Y \; \text{is
241: birational to} \quad X\}/\sim
242: \]
243: where $\sim$ denotes \emph{square birational equivalence}. We say
244: that $X\to S$ is \emph{birationally rigid} when $\mathcal{P} (X/S)$
245: consists of one element.
246: \end{dfn}
247:
248: \noindent A few general criteria for birational rigidity have been known
249: for quite some time, see \cite{MR98e:14009}, \cite{corti:00}.
250: These criteria and their proofs are based on exploiting
251: properties of the 1-cycle $K^2_X$. The importance of the following
252: condition was first recognised in a series of brilliant papers by
253: A.\ Pukhlikov. His point of view is explained for example
254: in \cite{MR2001j:14010}.
255:
256: \begin{dfn} \cite{MR2000d:14017, MR99f:14016, MR98j:14014}
257: We say that a variety $X$ satisfies the \emph{$K^2$~condition} if
258: $K_X^2$ is not in the interior of the Mori cone $\NE X$ of effective
259: 1-dimensional cycles on $X$.
260: \end{dfn}
261:
262: \begin{thm} \cite{MR84h:14047, MR82g:14035, MR81d:14023},
263: \cite{MR2000d:14017, MR99f:14016, MR98j:14014}
264: Let $X$ be a nonsingular \mbox{3-fold} and $X\to S$ a conic bundle or a $dP_3$
265: fibration (satisfying an additional ``genericity'' condition on the
266: singularities---see Theorem~\ref{thm:dp3} below). If
267: $X$ satisfies the $K^2$~condition, then $X \to S$ is birationally
268: rigid.
269: \end{thm}
270:
271: More recent results explore examples that lie on the boundary between rigid and
272: nonrigid \cite{cm02, mella02}, \cite{grin03}, \cite{MR2003g:14017}.
273: These papers study Mfs $X/S$ which are either \emph{bi-rigid} in the
274: sense that they have exactly two models as Mori fibre spaces
275: (more precisely, $\mathcal{P}(X/S)$ is a set with two elements), or that are
276: rigid in a subtle way, that is, there exist Mfs $Y/T$ and nonsquare maps $X
277: \dasharrow Y$, but $Y/T$ is always square birational to $X/T$.
278: In this paper we construct many examples of varieties that are likely
279: to behave similarly.
280:
281: \subsection{Conjectures}
282: \label{sec:mfs_con}
283:
284: The known general criteria for rigidity of Mori fibre spaces are not
285: optimal. Our second goal in this paper is to state some conjectures, based on
286: the intuition built on looking at the examples we construct. Some of
287: these conjectures we expect to be able to prove in the near future,
288: others, we feel, are definitely harder, and some are probably wrong.
289: The $K^2$~condition is not natural. Our conjectures are stated in term
290: of a geometrically more natural condition, whose importance was also
291: recognised in work by Grinenko \cite[Conjecture 1.5]{grin99}
292: \cite[Conjecture 1.6]{grinenko:00} \cite[Conjecture 2.5]{gri02}.
293:
294: \begin{dfn} \label{dfn:con-star} A variety $X$ \emph{satisfies
295: condition~$(\ast)$} if the anticanonical class $-K$ is not in the interior of
296: the cone of mobile divisors of $X$.
297: \end{dfn}
298:
299: \noindent Note that condition~$(\ast)$ is stronger than the $K^2$~condition.
300: Indeed, if $-K$ is in the interior of the mobile cone, we can write
301: for some $n>0$
302: \[
303: -nK=M_1+M_2=H_1+H_2
304: \]
305: where $M_1\sim H_1$ have no component in common, and,
306: similarly, $M_2\sim H_2$ have no component in common.
307: Then writing
308: \[
309: n^2K^2=\sum M_iH_j
310: \]
311: shows that $K^2$ is in the interior of the Mori cone.
312:
313: In Sections~\ref{sec:rigid-conic-bundles} and~\ref{sec:dp-con}
314: we state some conjectures on the rigidity of standard conic bundles
315: and stable $dP_3$ fibrations in terms of condition~$(\ast)$.
316: For many of the examples that we construct which seem to lie on the
317: boundary between rigid and nonrigid, it would be natural to conjecture
318: that they are bi-rigid, or that they are still rigid, though they do not
319: satisfy condition~$(\ast)$. We have resisted the temptation to make
320: such conjectures, but we invite our readers to make their own and
321: possibly prove some theorems.
322:
323: It is by now well understood how to use Pukhlikov's $K^2$~condition
324: to prove rigidity of Mfs, but it seems difficult to go further.
325: On the other hand, we have as yet no experience or success
326: using condition~$(\ast)$ to prove rigidity of Mfs.
327: We hope that this paper will help to introduce a
328: new point of view on the birational geometry of Mori fibre spaces,
329: where the geometric properties of $-K$ play more prominent a role through
330: condition~$(\ast)$.
331:
332: \subsection{Tables and Figures}
333: \label{sec:tables-figures}
334:
335: Tables~\ref{tab:1} and~\ref{tab:2}, and Figure~\ref{fig:geog}
336: summarise and collect key information which is obtained in many
337: calculations throughout the paper.
338: Table~\ref{tab:1} shows the list of standard conic bundles over $\p^2$
339: with discriminant of degree 7 and contains information about their
340: nonsquare Sarkisov links to alternative models as Mori fibre spaces.
341: Figure~\ref{fig:geog} depicts the geography of $dP_3$ fibrations and
342: contains information about the general member of each
343: family, showing whether it satisfies condition~$(\ast)$ and whether we know an
344: alternative model as a Mori fibre space.
345: This information is supplemented in Table~\ref{tab:2}, which lists all
346: families of $dP_3$ fibration for which we know an element that has a
347: nonsquare Sarkisov link to some Mfs $Y/T$ and contains detailed
348: information about the link.
349:
350: \subsection{The Appendix}
351: \label{sec:appendix}
352:
353: The Appendix sets out our notation for rational scrolls and
354: information about their birational maps. A scroll $\F$ is a quotient of
355: affine space by an action of the group $G=\cstar \times \cstar$. Though
356: one is not normally aware of this, the quotient depends on the
357: choice of a $G$-linearisation. Different $G$-linearisations produce
358: different quotients. In this manner we can understand the 2-ray game
359: originating from the scroll. In the Appendix, we introduce notation for
360: this which is used throughout the paper when constructing Sarkisov
361: links of Mfs $X\subset \F$; indeed, it is often the case that
362: moves of $\F$ induce moves of $X$.
363:
364: \subsection{Various cones}
365: \label{sec:various-cones}
366:
367: Let $X$ be a projective variety. We denote $N^1(X)=N^1(X, \r)$ the
368: vector space of Cartier divisors on $X$ with real coefficients, modulo
369: \emph{numerical} equivalence, and $N_1(X)=N_1(X, \r)$ the dual space of
370: 1-dimensional cycles on $X$ with real coefficients, modulo
371: numerical equivalence. Throughout this paper, it is crucial
372: to be aware of various cones associated to $X$:
373: \begin{description}
374: \item[The Mori cone] $\NE_1(X)$, or simply $\NE (X)$, is the closure
375: of the cone in $N_1(X)$ generated by the effective 1-cycles.
376: \item[The ample cone] $\NA^1(X)$ is the cone in $N^1(X)$ generated by
377: ample divisors; Kleiman's criterion states that it is the dual cone
378: to the Mori cone.
379: \item[The mobile cone] $\NM^1(X)$ is the cone in $N^1(X)$ generated by
380: mobile divisors. We say that a divisor $D$ on $X$ is \emph{mobile}
381: if a positive multiple $nD$ moves in a linear system with no fixed
382: divisor.
383: \item[The quasieffective cone] $\NE^1 (X)$ is the closure of the
384: cone in $N^1(X)$ generated by effective divisors. Divisor in this
385: cone are called \emph{quasieffective}.
386: \end{description}
387:
388:
389: \subsection{Disclaimer}
390:
391: This paper was prepared against a tight (for us) deadline. We did not
392: test some of the conjectures as much as we would have liked,
393: and it is likely that some of the definitions, calculations,
394: etc.\ contain mistakes. The responsibility for these is of course
395: ours, but we still want to apologise for the inconvenience we
396: may be causing our readers.
397:
398: \section{The Sarkisov category}
399: \label{sec:sarkisov-category}
400:
401: The \emph{Sarkisov category} is the category of Mori fibre spaces and
402: birational maps between them. We begin with a simple result
403: stating that Mfs that have a model as a Fano \mbox{3-fold} belong to
404: finitely many deformation families.
405:
406: \begin{pro} \label{pro:finite}
407: There is a finite collection of algebraic families of Mori fibre
408: spaces with the following property: if $X\to S$ is birational to a
409: Fano \mbox{3-fold} $Y$, then $X/S$ is square birational to a member of one
410: of the families in the collection.
411: \end{pro}
412:
413: \begin{proof}
414: We briefly explain the idea of the proof. By the Sarkisov program
415: \cite{MR96c:14013}, nonrigid Mfs are parametrised by nonsquare
416: links of the Sarkisov program. These are in turn parametrised by
417: weak terminal Fano \mbox{3-folds} of rank 2, and these are bounded by
418: \cite{MR2001h:14053}.
419: \end{proof}
420:
421: We hope that we will soon be able to prove the following.
422:
423: \begin{con}
424: If $X\to S$ is a Mori fibre space, the pliability $\mathcal{P}
425: (X/S)$ is in a natural way an algebraic variety.
426: \end{con}
427:
428: \noindent The key difficulty is this. Let $\mathcal{X}\to T$ be a family
429: of Fano \mbox{3-folds} with a birational selfmap
430: $\sigma \colon \mathcal{X} \dasharrow \mathcal{X}$ that maps fibres
431: birationally into fibres and hence induces a birational selfmap of the
432: base $T$. If this induced birational selfmap of $T$ has
433: infinite order and $t_1\in T$ is a general point, then the fibre $X_{t_1}$
434: is birational to $X_{t_2}$, and then to $X_{t_3}$ and
435: so on. It seems possible that all the values $t_i\in T$, but not the
436: whole of $T$, may contribute to the pliability.
437:
438: Recent experience \cite{cpr} \cite{cm02, mella02} suggests the
439: following rather optimistic conjectures.
440:
441: \begin{con}
442: For the total space $X$ of a \mbox{3-fold} Mfs $X\to S$ to be rational is a
443: topological property. In other words, if $X/S$ and $Y/T$ are \mbox{3-fold}
444: Mori fibre spaces, $X$ is rational and $Y$ is diffeomorphic to $X$,
445: then $Y$ is also rational.
446: \end{con}
447:
448: \begin{con}
449: For a \mbox{3-fold} Mori fibre space to be rigid is a topological
450: property.
451: \end{con}
452: \noindent In particular, being rigid or rational is constant along algebraic
453: families where all fibres are diffeomorphic.
454:
455: \begin{con}
456: The pliability $\mathcal{P}(X/S)$ of a \mbox{3-fold} Mori fibre space is in
457: a natural way a topological invariant.
458: \end{con}
459:
460:
461: \section{Conic bundles}
462: \label{sec:conic-bundles}
463:
464: We do not remind you of the known sufficient conditions for rigidity of conic
465: bundles or Iskovskikh's conjectural characterisation of conic bundles
466: with rational total space
467: \cite{MR97j:14044, MR93i:14033, MR92k:14036, MR88i:14038}
468: \cite{corti:00}. We do illustrate a
469: method to write down equations of a conic bundle with assigned
470: discriminant. This method is based on Catanese's work on the Babbage
471: conjecture \cite{MR83c:14026} and is explicitly computable, unlike the
472: traditional
473: abstract approach via Brauer groups. We focus on conic bundles over
474: $\p^2$ with discriminant a nodal plane curve of degree~7. Motivated by
475: these examples, we state some conjectures on the rigidity of conic bundles.
476:
477: \subsection{Conic bundles and Brauer groups}
478: \label{sec:conic-brauer}
479:
480: For details on this section see \cite{MR48:299}, \cite{MR84h:14047}.
481: Let $X \to S$ be a \emph{standard} conic bundle over a rational surface
482: $S$. By definition, this means that $X$ is nonsingular
483: and the relative Picard rank is 1, a consequence of which is that
484: $S$ is itself nonsingular and the discriminant curve $\Delta \subset
485: S$ is nodal. Over
486: $\Delta$, the fibre is generically the sum of two lines, which
487: specifies a \mbox{2-to-1} \emph{admissible cover} $N \to \Delta$.
488: (When $C$ is nonsingular, admissible just
489: means \'etale; in general, an admissible \mbox{2-to-1} cover is required to
490: ramify over both branches of each node of $C$.)
491: Together, we refer to the double cover $N\to \Delta$ and the embedding $\Delta
492: \subset S$ as the \emph{ramification data} of the conic bundle.
493:
494: It is well known that there is a standard conic bundle $X\to S$ with
495: any preassigned ramification data $\Delta \subset S$ and \mbox{2-to-1}
496: admissible cover $N \to \Delta$. The total space $X$ is not unique, but any two
497: choices are square birational over the base.
498:
499: The traditional proof uses the exact sequence \cite{MR48:299} (in which $S$ is
500: a rational surface with field of rational functions $K$)
501: \[
502: 0\to \Br S \to \Br K\to\bigoplus_{\text{curves}\;C \subset S}
503: H^1(K(C), \q/\z) \to \bigoplus_{\text{points}\; P \in S} \mu^{-1} \to
504: \mu^{-1} \to 0
505: \]
506: and the fact, due to Platonov and reproduced in \cite{MR84h:14047},
507: that every element of order 2 in the Brauer group $\Br K$ can be
508: represented by a quaternion algebra. See the references for details
509: and explanations.
510:
511: In the following subsections, we illustrate an effective proof of this
512: statement in two steps: first we learn how to specify effectively a
513: \mbox{2-to-1} cover $N \to \Delta$, then we write down explicit equations for
514: a conic bundle $X$.
515:
516: \subsection{Catanese's theory}
517: \label{sec:cataneses-theory}
518:
519: Let $C$ be a nonsingular curve of degree $d$ in $\p^2$. (With a bit of
520: care, the theory works unchanged for nodal curves.) We fix coordinates
521: $u_0, u_1, u_2$ on $\p^2$ and denote $S=k[u_0,u_1,u_2]$ the
522: homogeneous coordinate ring.
523: We are interested in admissible \mbox{2-to-1} covers of $C$. We summarise the
524: part of Catanese's paper \cite{MR83c:14026} which is relevant to us;
525: see also \cite{Dixon}.
526: The theory works identically for covers which in addition ramify over
527: a specified hyperplane section of $C$, and we treat these as well.
528:
529: An admissible \mbox{2-to-1} cover corresponds to a line bundle
530: $\mathcal{L}$ on $C$ with a choice of isomorphism: either
531: $\mathcal{L}^2 =\o_C$, or $\mathcal{L}^2 =\o_C(-1)$.
532: Denote by $i\colon C \hookrightarrow \p^2$ the inclusion. By
533: \cite{MR83c:14026}, Theorems~2.16 and 2.19, $i_\ast \mathcal{L}$ is a
534: Cohen-Macaulay $\o_{\p^2}$-module, so it has a 2-step symmetric locally
535: free resolution:
536: \[
537: 0 \to \oplus \o_{\p^2} (-l_i) \stackrel{A}{\to} \oplus \o_{\p^2} (-r_i)
538: \to i_\ast \mathcal{L} \to 0.
539: \]
540: In other words, $A$ is a symmetric $n\times n$ matrix whose
541: entries are homogeneous forms on $\p^2$.
542:
543: Note that $\det A$ is a homogeneous equation of $C$.
544: On the other hand, writing $C$ as a symmetric determinantal specifies the line
545: bundle $\l$ and the double cover.
546:
547: Denote by $d_i$ the degree of the $i$-th diagonal entry $a_{ii}$ of
548: $A$; $d = \sum d_i$ is a partition
549: of $d$. The degree of $a_{ij}$ is $(d_i+d_j)/2$,
550: so that all the $d_i$s have the same parity.
551: We have the following numerology:
552: \begin{equation*}
553: \begin{aligned}
554: r_i &= (d+e-d_i)/2\\
555: l_j &= (d+e+d_j)/2
556: \end{aligned}
557: \quad
558: \text{where}
559: \quad
560: e=
561: \begin{cases}
562: 0 \quad \text{if}\quad \l^2 = \o_C\\
563: 1 \quad \text{if}\quad \l^2 = \o_C(1).
564: \end{cases}
565: \end{equation*}
566: Let
567: \[
568: M = \bigoplus_{n \geq 0} M_n \quad \text{where} \quad M_n =
569: H^0 \bigl(\p^2, i_\ast \l (n)\bigr)
570: \]
571: be the Serre module of the sheaf $i_\ast \l$.
572: The direct summand $\o_{\p^2} (-r_i)$ in the presentation
573: corresponds to a generator $m_i$ of degree $r_i$ of $M$.
574: The matrix $A$ encodes the structure of $M$ as a graded module over
575: the homogeneous coordinate ring $S=k[u_0,u_1,u_2]$ of $\p^2$, as well
576: as the identity $m_i m_j = b_{ij} \in S$
577: in which $(b_{ij})=B = \ad A$ is the adjugate matrix of $A$.
578:
579: We work out the well-known example
580: of theta characteristics of plane quartics in detail; this
581: corresponds to the case $d=4$, $e=1$ of the theory.
582: Here $\l$ is a line bundle with $\l^2 =\o(-1)$, and $\l(1)$ is a theta
583: characteristic since $K_C = \o(1)$. A plane quartic has genus $g=3$
584: and $\deg \l = -2$, so Riemann--Roch gives
585: \[
586: h^0(C,\l(1))-h^1(C,\l(1))=0.
587: \]
588: We have two cases: $ h^0(\l (1) )= 0$ (even theta characteristic), and
589: $ h^0(\l (1) )= 1$ (odd theta characteristic).
590: We work out the odd case.
591:
592: By assumption, we have exactly one generator
593: $m \in M_1 = H^0(C,\l(1))$ of degree 1.
594: Next, $\l(2)$ has degree 6 and Riemann--Roch calculates
595: \[
596: h^0(C,\l(2))=6+1-3=4,
597: \]
598: so $M_2$ is based by 4 elements: they are
599: $u_0m, u_1m, u_2m$
600: and a new generator $n$. Noting $h^0(C,\l(3)) = 8$, we see that there
601: must be a relation in degree 2 between the 9 elements $u_iu_jm$,
602: $u_kn$; this relation must take the form $n f_1 = m f_2$
603: where $f_1$ is a linear form in $u_0, u_1, u_2$ and $f_2$ a
604: quadratic form. The matrix $A$ has the form
605: \[
606: A=
607: \begin{pmatrix}
608: f_1& f_2 \\
609: f_2& f_3
610: \end{pmatrix}.
611: \]
612: The whole structure is specified by writing the curve
613: $C=(f_1f_3-f_2^2=0)$ as a symmetric determinantal.
614: This presentation reveals a preferred
615: bitangent $f_1=0$, which ``is" the odd section $m$.
616:
617: In the even case, you will find 4 generators $m_1,\dots,m_4$
618: of degree 2, and 4 linear relations
619: among the 12 elements $u_i m_j$: $A$ is a $4\times 4$
620: symmetric matrix of linear forms.
621:
622: \subsection{Plane curves of degree 7}
623: \label{sec:deg-7}
624: We study the case $d=7$, $e=0$, that is, line bundles $\l$
625: with $\l^2 = \o_C$ on plane curves $C$ of degree $\deg C = 7$.
626: Here $K_C = \o(4)$, so $\l(2)$ is a theta characteristic. A plane curve of
627: degree $7$ has genus $g=15$ and canonical degree $\deg K_C=28$.
628: Riemann--Roch and Serre duality give at once
629: \begin{eqnarray*}
630: \label{eq:rr7}
631: h^0\bigl(C,\l(n)\bigr) - h^1\bigl(C,\l(n)\bigr) & = & 7(n-2) \\
632: h^1\bigl(C,\l(n)\bigr) & = & h^0\bigl(C,K_C\otimes\l(n)^\ast\bigr) \\
633: & = & h^0\bigl(C,\l(4-n)\bigr).
634: \end{eqnarray*}
635: We use these equations without comment throughout.
636: It is clear that $p_1 = h^0(C,\l(1))$
637: and $p_2 = h^0(C,\l(2))$ determine all the $p_n=h^0(C, \l(n))$: indeed,
638: the Hilbert series of $i_\ast\l$ as an $S$-module is
639: \begin{multline*}
640: P(t) =\sum_{n\geq 0} h^0(C, \l(n)) t^n = \\
641: p_1t + p_2t^2 + (7 + p_1)t^3 + 14t^4 + \cdots + 7it^{i-2} + \cdots.
642: \end{multline*}
643: % & = & N(t)/(1-t)^3
644: % where $N(t) = \ell_1t + N_2t^2 + N_3t^3 - N_3t^4 - N_2t^5 - \ell_1t^6$
645: % and $N_2 = \ell_2 - 3\ell_1$, $N_3 = 7-3\ell_2+4\ell_1$.
646:
647: There are four cases, corresponding to the four possible values of
648: $p_2 = h^0(C, \l(2))$. We show later that the first two cases
649: occur on a generic curve, while the last two cases can only occur on
650: curves with special moduli. In the four cases,
651: the degrees of the diagonal entries of the matrix $A$ are the
652: summands of the four different partitions of the number $d=7$
653: into odd summands.
654: \begin{enumerate}
655: \item $h^0(C, \l(2))=0$ (generic even theta),
656: \item $h^0(C, \l(2))=1$ (generic odd theta),
657: \item $h^0(C, \l(2))=2$ (special even theta),
658: \item $h^0(C, \l(2))=3$ (special odd theta).
659: \end{enumerate}
660:
661: \subsubsection{$h^0(C, \l(2))=0$}
662: \label{sec:case-1}
663: We have $h^0(C, \l(3)) = 7$ generators of degree 3. Multiplying these
664: by $u_0$, $u_1$, $u_2$, we get 21 elements in the 14-dimensional space
665: $M_4$, so there are 7 equations between them.
666: In this case $A$ is a 7 by 7 symmetric matrix of linear forms.
667:
668: \subsubsection{$h^0(C, \l(2))=1$}
669: \label{sec:case-2}
670: Again $h^0(C, \l(3))=7$, but here we assume a generator $m$ of
671: degree 2, giving $u_i m$ in $M_3$, so we need 4 more generators $n_1,\dots,n_4$
672: in $M_3$. In this case $A$ is a $5 \times 5$ symmetric matrix of homogeneous
673: forms of degrees
674: \[
675: \begin{pmatrix}
676: 3 & 2 & 2 & 2 &2 \cr
677: & 1 & 1 & 1 &1 \cr
678: & & 1 & 1 &1 \cr
679: & & & 1 &1 \cr
680: & & & &1
681: \end{pmatrix}
682: \]
683: in which the degrees of the diagonal elements express $d=7=3+1+1+1+1$
684: as a partition into odd summands.
685:
686:
687: \subsubsection{$h^0(C, \l(2))=2$}
688: \label{sec:case-3}
689:
690: This case corresponds to the partition $7=3+3+1$. The assumption is
691: that there are two generators $m_1, m_2$ in $M_2$, so, as usual,
692: $h^2(C, \l(3))=7$ then shows that we need one more generator $n$
693: in degree 3. We conclude that $A$ is a $3 \times 3$ symmetric matrix
694: of homogeneous forms of degrees
695: \begin{equation}
696: \label{mx:case3}
697: \begin{pmatrix}
698: 3& 3& 2\cr
699: & 3& 2\cr
700: & & 1
701: \end{pmatrix}.
702: \end{equation}
703: We show below that not all plane septics have a theta-characteristic
704: of this type. More precisely, the locus of those who do is of
705: codimension 1 in moduli.
706:
707:
708: \begin{exa}[Counting moduli]
709: Let $C = (\det A = 0)\subset\p^2$ for $A$ a symmetric
710: $3\times 3$ matrix of forms of degrees given by matrix
711: format (\ref{mx:case3}) above. The matrix A depends on
712: $3\times 10 + 2\times 6 + 3 = 45$ parameters, and $A$ is uniquely
713: determined by the choice of the generators of the
714: Serre module. In this case, that choice amounts to the choice of
715: $m_1,m_2$ in $M_2$ (which has the 4 degrees of freedom of the basis of a
716: 2-dimensional vector space) and the choice of the additional generator
717: $n$ in $M_3$ (7 degrees of freedom) for a total of $4+7=11$ degrees of
718: freedom. The space of septics of the given form has dimension $45 - 11 = 34$,
719: whereas the space of all septics has dimension $35$.
720: \end{exa}
721:
722: \subsubsection{$h^0(C, \l(2))=3$}
723: \label{sec:case-4}
724:
725: This case corresponds to the partition $7=5+1+1$. Here
726: $p_1=1$ and the element $m$ in $M_1$ generates a basis
727: $u_i m$ of $M_2$. This time $h^0(C,\l(3)) = 8$,
728: so we need two additional generators $n_1,n_2$ in degree 3.
729: The matrix $A$ is a $3\times 3$ matrix of homogeneous forms of degrees
730: \[
731: \begin{pmatrix}
732: 5& 3& 3\cr
733: & 1& 1\cr
734: & & 1
735: \end{pmatrix}.
736: \]
737: Once again, the locus of septics who own a theta characteristic of
738: this type is of codimension 1 in moduli.
739:
740:
741: \subsection{Conic bundles over $\p^2$ with discriminant of degree 7}
742: \label{sec:delta-7}
743:
744: We construct standard conic bundles over $\p^2$ with ramification data
745: a \mbox{2-to-1} admissible cover $N \to \Delta \subset \p^2$ of a plane curve
746: of degree $7$. We make four deformation families of such \mbox{3-folds},
747: following the analysis in the previous section.
748: We show that general members $X$ of the first three families have an
749: alternative model as a Mori fibre space. (This may be true for all
750: their members, but we didn't check that). Conjecture~\ref{con:cb} below
751: implies that a general member of the fourth family is birationally rigid.
752: This is a striking prediction, because this example is far
753: from the numerical range where the known rigidity criteria apply.
754:
755: \subsubsection{$h^0(\Delta, \l(2))=0$}
756: \label{sec:cb7-1}
757:
758: Let $N \to \Delta$ be as in \ref{sec:case-1}. A conic bundle with this
759: ramification data was known classically, and it is birational to the
760: Fano \mbox{3-fold} $Y = Y_{2,2,2} \subset \p^6$, the codimension 3 complete
761: intersection of three quadrics in $\p^6$. Starting from $Y$, the
762: link $Y\dasharrow X$ begins with the projection from a line $\ell
763: \subset Y$. The details of this construction are well known, but see
764: also the following example.
765:
766: \subsubsection{$h^0(C, \l(2))=1$}
767: \label{sec:cb7-2}
768: Let $N \to \Delta$ be as in \ref{sec:case-2}.
769: We want explicit equations for a conic
770: bundle with ramification data $N \to \Delta$.
771: Consider the variety
772: \[
773: Z = \{ {}^t \mathbf{x} A \mathbf{x} = 0 \} \subset \F = \F (1,0,0,0,0).
774: \]
775: This is a bundle of 3-dimensional quadrics immersed in a scroll over
776: $\p^2$ with ramification data $N\to \Delta$. Indeed, a singular
777: 3-dimensional quadric is, generically, a cone over $\p^1 \times
778: \p^1$, and the two rulings give back the cover $N \to \Delta$.
779: It is natural to try to construct our conic bundle from $Z$ by a kind
780: of ``dimensional reduction''.
781:
782: \paragraph{Claim 1}
783: \label{sec:claim-1}
784:
785: We can choose coordinates in the scroll $\F$ such that the matrix $A$
786: has the form
787: \[
788: A=
789: \begin{pmatrix}
790: B & \mathbf{b} \cr
791: {}^t \mathbf{b} & 0
792: \end{pmatrix}
793: \]
794: where $\mathbf{b}$ is a $1\times 4$ column vector of linear
795: forms. To see that this is possible, we look for a coordinate change
796: of the form:
797: \[
798: \begin{pmatrix}
799: x_0 \\
800: \mathbf{\xi}
801: \end{pmatrix}
802: \mapsto
803: \begin{pmatrix}
804: 1 & 0 \\
805: \mathbf{\beta} & M
806: \end{pmatrix}
807: \begin{pmatrix}
808: x_0 \\
809: \mathbf{\xi}
810: \end{pmatrix}
811: \]
812: where ${}^t\mathbf{\xi} = (x_1,x_2,x_3,x_4)$, $M=(m_{ij}) \in
813: GL(4,\c)$, and $\mathbf{\beta}$ is a $4\times 1$ column vector of
814: linear forms in $u_0,u_1,u_2$. The coordinate change
815: brings $A$ in the
816: wanted form if and only if
817: the column vector with entries
818: $(0,m_{14},m_{24},m_{34},m_{44})$
819: is a nonzero null-vector of $A$. It is easy to see that such a vector
820: exists; indeed the $4\times 4$ symmetric submatrix $(a_{ij})$
821: of $A$ is a matrix of linear forms and corresponds to a net of
822: 3-dimensional quadrics. A base point of the net of quadrics gives the
823: null-vector we want.
824:
825: \paragraph{Equations of $X$}
826: \label{sec:equations-x}
827:
828: Define
829: \[
830: X=\{ {}^t \mathbf{x} B \mathbf{x} = \mathbf{b} \cdot \mathbf{x}= 0 \}
831: \subset \F (1,0,0,0)
832: \]
833: where $\F(1,0,0,0)$ is the subscroll $\{x_4=0\} \subset \F(1,0,0,0,0)$.
834: One can easily check that $X$ is a conic bundle with the required
835: ramification data $N \to \Delta$. Notice that, in terms of
836: projectivised vector bundles on $\p^2$, the linear form expresses $X$
837: as a conic in a projectivised nonsplit rank 3 vector bundle on $\p^2$.
838:
839: \paragraph{Claim 2}
840: \label{claim:claim-2}
841:
842: If $\mathbf{b}$ is sufficiently general, we may further change
843: coordinates so that
844: \[
845: \mathbf{b}=
846: \begin{pmatrix}
847: 0 \\
848: u_0\\
849: u_1\\
850: u_2
851: \end{pmatrix}.
852: \]
853:
854: \paragraph{Model as a Fano \mbox{3-fold}}
855: \label{sec:model-as-fano}
856:
857: We construct a Sarkisov link from $X$ to a
858: codimension 3 Fano \mbox{3-fold} $Y=Y_{2,3,3,3,3} \subset
859: \p(1^6,2)$. The ideal of $Y$ is generated by
860: the $4\times 4$ Pfaffians of a $5\times 5$ antisymmetric matrix of homogeneous
861: forms of degrees
862: \[
863: \begin{pmatrix}
864: 2&2&2&2\\
865: &1&1&1\\
866: & &1&1\\
867: & & &1
868: \end{pmatrix}
869: \]
870: on $\p(1^6,2)$. Assuming that $X$ is generic, we exhibit a birational
871: map, in fact a link of the Sarkisov program, from $X$ to a Fano \mbox{3-fold}
872: of the type just described.
873: This alternative model of $X$ as a Mori fibre space
874: demonstrates, in particular, that $X$ is not birationally rigid.
875:
876: If we choose coordinates as in Claim~\ref{claim:claim-2}, then $X \subset
877: \F(1,0,0,0)$ is given by equations
878: \[
879: \begin{cases}
880: {}^t \mathbf{x} B \mathbf{x} =0 \\
881: u_0x_1+u_1x_2+u_2x_3 = 0
882: \end{cases}
883: \]
884: where $B$ is a $4\times 4$ symmetric matrix of forms of degrees
885: \[
886: \begin{pmatrix}
887: 3 & 2 & 2& 2\\
888: & 1 & 1& 1\\
889: & & 1& 1\\
890: & & & 1
891: \end{pmatrix}
892: \]
893: in the coordinates $u_0,u_1,u_2$.
894:
895: The morphism $p\colon \F \to \p^5=\p H^0(\F, M)^\ast$ given by
896: \[
897: (x_1,x_2,x_3, x_4, x_5,x_6) =
898: (x_1,x_2,x_3,u_0x_0,u_1x_0,u_2x_0)
899: \]
900: identifies $\F$ with the blow up of $\p^5$ along the 2-plane $\Pi
901: =\{ x_4=x_5=x_6=0\}$. Under this morphism $X$ maps to
902: $p(X)=\overline{Y}_{2,3}$, a generic complete intersection of a
903: quadric and a cubic in $\p^5$ containing the plane $\Pi$.
904: To find the equations of $\overline{Y}$, note that we
905: can find unique homogeneous quadrics
906: $q_0$, $q_1$, $q_2$ in six variables, such that
907: \[
908: {}^t \mathbf{x} B \mathbf{x} =\sum_{i=0}^2
909: u_iq_i(x_1,x_2,x_3,u_0x_0,u_1x_0,u_2x_0).
910: \]
911: (If $B$ is generic, the quadrics $q_0,q_1,q_2$ are also
912: generic.) It is easy to see that $\overline{Y}$ is given by the
913: following equations:
914: \[
915: \begin{pmatrix}
916: x_1 & x_2 & x_3\\
917: q_0 & q_1 & q_2
918: \end{pmatrix}
919: \begin{pmatrix}
920: x_4\\
921: x_5\\
922: x_6
923: \end{pmatrix}
924: =0.
925: \]
926: Starting from $\overline{Y}_{2,3}$, we construct $Y$ as an
927: \emph{unprojection}, that is, the contraction of the Weil
928: divisor $\Pi \subset \overline{Y}$; see \cite{kinosaki},
929: \cite{MR1941620} for details.
930:
931:
932: \subsubsection{$h^0(\Delta, \l(2))=2$}
933: \label{sec:cb7-3}
934:
935: Let $N \to \Delta$ be as in \ref{sec:case-3}. In this case, we can
936: immediately write down equations for a conic bundle with this
937: ramification data. Indeed, $A$ is a
938: $3\times 3$ symmetric matrix of homogeneous forms of degrees
939: \[
940: \begin{pmatrix}
941: 3&3&2\\
942: &3&2\\
943: & &1
944: \end{pmatrix}.
945: \]
946: We can take
947: \[
948: X= \{{}^t \mathbf{x} A \mathbf{x}=0 \} \in |2M+L| \subset \F(1,1,0).
949: \]
950:
951: We now exhibit an alternative model of $X$ as a $dP_3$ fibration.
952: We begin with a preliminary discussion of the geometry of the scroll
953: $\F=\F(1,1,0)$, which we think of as a \emph{geometric} quotient
954: \[
955: \F= \a /\!\!/ G
956: \]
957: of $\a=\c^6$, with coordinates $u_0,u_1,u_2, x_0,x_1,x_2$, by the
958: group
959: $G=\c^\times \times \c^\times$, with coordinates $\lambda,
960: \mu$, acting in the usual way (see the Appendix). Indeed,
961: a $G$-linearisation of the trivial line bundle on $\a$ is the same
962: as a character $\chi\colon G \to \c^\times$. The group $G$ acts on
963: global sections $f\colon \a \to \c$ by $gf(x)=\chi(g)f(g^{-1}x)$.
964: We say that a linearisation is \emph{useful} if the (open) subset
965: \[
966: \a^{\textit{ss}}_\chi =
967: \{x \in \a \mid \exists \; f \in \o^\chi_\a,\;f(x) \not = 0\}
968: \]
969: of semistable points is nonempty, where
970: \[
971: \o_\a^\chi = \{f \colon \a \to
972: \c \mid f(gx)=\chi(g) f(x)\}
973: \]
974: is the set of $G$-invariant sections.
975: We have that
976: \[
977: \a^{\textit{ss}}_\chi=(\c^3 \setminus \{0\}) \times (\c^3 \setminus \{0\})
978: \quad \Longleftrightarrow \quad \chi \in \r_+[L]+\r_+[M].
979: \]
980: There are other useful linearisations. In fact, the cone of useful
981: linearisations is the cone $\r_+[L]+\r_+[M-L]$. This cone is
982: naturally partitioned in two chambers, and
983: \[
984: \a^{\textit{ss}}_\chi=(\c^4 \setminus \{0\}) \times (\c^2 \setminus \{0\})
985: \quad \Longleftrightarrow \quad \chi \in \r_+[M]+\r_+[M-L].
986: \]
987: gives the second chamber.
988: When choosing a linearisation in the second chamber, the geometric
989: quotient is a different scroll over $\p^1$:
990: \[
991: \a /\!\!/ G=\F^\prime = \F(0,1,1,1).
992: \]
993: Crossing the wall separating the two chambers corresponds to a
994: birational map $\F \dasharrow \F^\prime$. The equation of
995: $X$, viewed as the equation of a hypersurface $Y \subset
996: \F^\prime$, is a section of the line bundle $\o_{\F^\prime} (3M -L)$.
997: The most convincing way to see this is to make the substitutions
998: \begin{align*}
999: u_0, u_1, u_2, x_0 & \mapsto y_1,y_2,y_3,y_0 \cr
1000: x_1,x_2 & \mapsto t_0,t_1
1001: \end{align*}
1002: and to think of $t_0,t_1,y_0,y_1,y_2,y_3$ as natural coordinates on
1003: the scroll $\F^\prime$; the action of $\lambda, \mu$ on these
1004: coordinates is
1005: \begin{align*}
1006: \lambda \colon (t_0,t_1,y_0,y_1,y_2,y_3) \mapsto & (\lambda^{-1}
1007: t_0,\lambda^{-1} t_1, y_0,\lambda y_1,\lambda y_2, \lambda y_3)\cr
1008: \mu \colon (t_0,t_1,y_0,\dots,y_3) \mapsto & (\mu t_0,\mu
1009: t_1,\mu y_0,y_1,y_2,y_3).
1010: \end{align*}
1011: To recover the standard presentation of the scroll $\F^\prime =
1012: \F(0,1,1,1)$, we
1013: change coordinates in $G$: the 1-parameter subgroups
1014: $\lambda^\prime \to (\lambda^{\prime -1},1) $ and
1015: $\mu^\prime \to (\mu^\prime,\mu^\prime)$ act
1016: as
1017: \begin{align*}
1018: \lambda^\prime \colon (t_0,t_1, y_0,y_1,y_2,y_3) \mapsto & (\lambda^\prime
1019: t_0,\lambda^\prime t_1, y_0,\lambda^{\prime -1} y_1,\lambda^{\prime
1020: -1} y_2, \lambda^{\prime -1} y_3)\cr
1021: \mu^\prime \colon (t_0,t_1,y_0,y_1,y_2,y_3) \mapsto & (t_0, t_1,
1022: \mu^\prime y_0,\mu^\prime y_1,\mu^\prime y_2, \mu^\prime y_3).
1023: \end{align*}
1024: These calculations show that, with the stated substitutions,
1025: $H^0\bigl(\F^\prime ,3M-L\bigr)$ is canonically identified with
1026: \[
1027: H^0\bigl(\F, 3M-(-L+M)\bigr) = H^0(\F, 2M+L\bigr).\]
1028: It is easy to see that the map $X \dasharrow Y$ is a flop which, by
1029: what we just said, is a Sarkisov link (of type IV according to
1030: \cite{MR96c:14013}) from the conic bundle
1031: $X/\p^2$ to a $dP_3$ fibration $Y/\p^1$.
1032:
1033: \subsubsection{$h^0(\Delta, \l(2))=3$}
1034: \label{sec:cb7-4}
1035:
1036: Let $N \to \Delta$ be as in \ref{sec:case-4}. In this case, $A$ is a
1037: $3\times 3$ symmetric matrix of homogeneous forms of degrees
1038: \[
1039: \begin{pmatrix}
1040: 5&3&3\\
1041: &1&1\\
1042: & &1
1043: \end{pmatrix}
1044: \]
1045: and we can take
1046: \[
1047: X= \{{}^t \mathbf{x} A \mathbf{x}=0 \}\in |2M+L| \subset \F(2,0,0).
1048: \]
1049: The divisor $E=\{x_2=0\} \cap X$ is the exceptional divisor of
1050: a (2,1)-contraction $g\colon X \to Y$ with $K=0$, that is, $E$
1051: contracts to a curve
1052: of strictly canonical singularities on $Y$. This is a bad link and no
1053: alternative model as a Mfs is produced. We suspect that $X$ is
1054: birationally rigid. Observe that this follows from Conjecture~\ref{con:cb}.
1055:
1056: \subparagraph{}
1057:
1058: Table~\ref{tab:1} summarises the examples discussed so far.
1059: \begin{table}[ht]
1060: \begin{center}
1061: $
1062: \begin{array}{|c|c|c|c|}
1063: \hline
1064: h^0\bigl(\Delta,\l(2)\bigr) & \text{model over} \; \p^2 & \text{link} &
1065: \text{other model} \\
1066: \hline
1067: \hline
1068: 0 &X_{2M+L,M+L,M+L} & \text{flop and} & Y_{2,2,2} \subset \p^6\\
1069: & \subset \p^2 \times \p^4 & \text{(2,1)-contraction} & \\
1070: \hline
1071: 1 &X_{2M+L,M+L} &\text{flop and} & Y_{2,3,3,3,3}\subset \p(1^6,2) \\
1072: & \subset \F(1, 0^3) &\text{(2,0)-contraction} & \\
1073: \hline
1074: 2 &X_{2M+L} \subset \F(1,1,0)& \text{flop}
1075: & Y_{3M-L} \subset \F(1^3,0)\\
1076: & & & \\
1077: \hline
1078: 3 &X_{2M+L} \subset \F(2,0,0)& \text{bad $K$ trivial}
1079: &\text{$X$ rigid ?} \\
1080: & &\text{(2,1)-contraction} & \\
1081: \hline
1082: \end{array}
1083: $
1084: \end{center}
1085: \caption{Conic bundles over $\p^2$ with discriminant $\Delta$ of
1086: degree 7\label{tab:1}}
1087: \end{table}
1088:
1089: \subsection{Conjectures on rigid conic bundles}
1090: \label{sec:rigid-conic-bundles}
1091:
1092: We briefly sketch a few more examples of conic bundles, mainly
1093: nonstandard or over surfaces other than $\p^2$, then state some conjectures.
1094: %Topics: nodal curves, weighted projective spaces, hilbert series,
1095:
1096: \begin{exa}
1097: We show that a general codimension 3 Fano \mbox{3-fold}
1098: $Y_{3,3,4,4,4}\subset \p(1^5,2,3)$ is
1099: linked to a \emph{nonstandard} conic bundle $X$.
1100: The equations of $Y$ are the Pfaffians of a $5\times 5$
1101: antisymmetric matrix of homogeneous forms which, in suitable
1102: coordinates, can be written as follows:
1103: \[
1104: \begin{pmatrix}
1105: z &a_1 &a_2 &a_3 \cr
1106: &b_1 &b_2 &b_3 \cr
1107: & &x_5 &x_4 \cr
1108: & & &x_3
1109: \end{pmatrix}.
1110: \]
1111: (Here as usual $z$ is a coordinate of weight $3$, $y$ is a coordinate
1112: of degree $2$, and $x_1,\dots,x_5$ are coordinates of weight 1. The $a$s and
1113: the $b$s are homogeneous forms of degree 2 in the $x$s and $y$.)
1114: Projecting $Y$ to $\p(1^5, 2)$ we obtain a complete intersection
1115: $\overline{Y}_{3, 3}$ given by equations:
1116: \[
1117: \begin{pmatrix}
1118: a_1 &a_2 &a_3\cr
1119: b_1 &b_2 &b_3
1120: \end{pmatrix}
1121: \begin{pmatrix}
1122: x_3 \\
1123: x_4 \\
1124: x_5
1125: \end{pmatrix}
1126: =0,
1127: \]
1128: containing the weighted plane $\Pi =\{ x_3=x_4=x_5=0\}$. We now
1129: construct a model as a conic bundle. First we construct a suitable
1130: ambient space. Consider the quotient
1131: \[
1132: \F = \bigl((\c^3 \setminus \{0\}) \times (\c^4 \setminus \{0\})\bigr)/
1133: (\c^\times \times \c^\times)
1134: \]
1135: by the action
1136: \begin{align*}
1137: \lambda \colon (u_0,u_1,u_2,x_0,x_1,x_2,y) & \mapsto (\lambda u_0,
1138: \lambda u_1, \lambda u_2, \lambda^{-1} x_0, x_1, x_2, y) \cr
1139: \mu \colon (u_0,u_1,u_2,x_0,x_1,x_2,y) & \mapsto (u_0,u_1,u_2,\mu x_0, \mu
1140: x_1, \mu x_2, \mu^2 y).
1141: \end{align*}
1142: We can think of $\F$ as a scroll over $\p^2$ with fibre the weighted
1143: projective space $\p(1^3, 2)$. We define a morphism $f\colon \F \to
1144: \p(1^5,2)$ by setting:
1145: \[
1146: (u_0,u_1,u_2,x_0,x_1,x_2,y) \mapsto (x_1,x_2,u_0x_0,u_1x_0,u_2x_0,y).
1147: \]
1148: The proper preimage $X =f^{-1} \overline{Y}$
1149: is a general complete intersection of the form
1150: $X=X_{2M+L,2M+L}\subset \F$. The
1151: ramification data $N \to \Delta$ is a \mbox{2-to-1} covering of a
1152: special plane octic with a triple point $P\in \Delta$, so $X$
1153: is not a standard conic bundle. It is easy to see that $X$ has an
1154: index 2 orbifold point over $P$. Because of the singular point, the
1155: threshold invariant $\tau(X/\p^2)$ \cite[\S 4.1]{corti:00}
1156: is $5/2$, and not the expected $8/2$.
1157: Blowing up the point $P\in \p^2$ expresses $X$ as a conic bundle
1158: over $\F^1$ with discriminant of relative degree $5$, and this in
1159: turn has a model as a $dP_4$ fibration over $\p^1$.
1160: \end{exa}
1161:
1162: \begin{exa}
1163: A general codimension 3 Fano \mbox{3-fold} $Y_{4,4,4,4,4} \subset \p(1^3,
1164: 2^4)$ is linked to a nonstandard conic bundle
1165: $X/\p^2$. The discriminant $\Delta$ is a
1166: plane curve of degree $9$ with four ordinary triple points, so $X$ is
1167: not a standard conic bundle. It is easy to see that $X$ has
1168: index 2 orbifold singularities over the singular points of
1169: $\Delta$. Please do your own calculations.
1170: \end{exa}
1171:
1172: \begin{exa}
1173: Takagi and Reid \cite{MR1924722}, \cite{kinosaki} construct a
1174: codimension 4 Fano \mbox{3-fold} $Y \subset \p(1^6,2^2)$ with $h^0(-K)= g+2
1175: =6$ and $-K^3=7$ with 2 index 2 orbifold points. There are two
1176: deformation families with these invariants, corresponding to the
1177: ``Tom'' and ``Jerry'' formats of unprojection. If $Y$ is a general
1178: Fano in the Tom family, then $Y$ is linked to a general conic bundle
1179: $X\to \p^2$ with discriminant curve of degree 6. The cover $N \to
1180: \Delta$ corresponds to the partition $6=2+2+2$, but we did not
1181: yet carry out all the necessary calculations to understand the model
1182: $X$ explicitly.
1183: \end{exa}
1184:
1185: \begin{exa}
1186: There is no reason why we should only work with conic bundles over
1187: $\p^2$, since Catanese's theory works essentially unchanged for curves on
1188: a weighted projective plane. For example, there are two
1189: Fano \mbox{3-folds} in codimension 4 and 5, $Y \subset \p(1^5,2^3)$ and
1190: $Y\subset \p(1^5,2^4)$, with $-K^3=11/2$ and $6$, see \cite{MR1924722},
1191: which can be linked to conic bundles over $\p(1,1,2)$ with discriminants in
1192: $\o(8)$ and $\o(6)$.
1193: \end{exa}
1194:
1195: We have written down explicit equations of several families of conic bundles
1196: with discriminant of low degree. We believe that the \emph{explicit
1197: geometry} of these varieties will play an increasingly more prominent
1198: role in the study of their birational geometry. We state the following
1199: rather optimistic conjecture.
1200:
1201: \begin{dfn}
1202: We say that a standard conic bundle $\pi \colon X\to S$ satisfies
1203: condition~$(\ast)$ if $-K_X \not \in \Ins \NM^1 X$.
1204: \end{dfn}
1205:
1206: \begin{con} \label{con:cb}
1207: A standard conic bundle over $\p^2$ is
1208: birationally rigid if it satisfies condition~$(\ast)$.
1209: \end{con}
1210: We feel like making the following conjecture for which we have little
1211: evidence.
1212: \begin{con}
1213: A standard conic bundle over $\p^2$ is
1214: birationally rigid if the discriminant has degree \mbox{$\ge 9$}.
1215: \end{con}
1216:
1217: \section{$dP_3$ fibrations }
1218: \label{sec:dp_3-fibrations}
1219:
1220: \begin{conv} \label{conv:dp3}
1221: In this section a \emph{$dP_3$ fibration} is a \mbox{3-fold} $X$
1222: together with a
1223: morphism $f\colon X \to \p^1$ satisfying the following conditions:
1224: \begin{enumerate}
1225: \item The nonsingular fibres of $f$ are cubic surfaces.
1226: \item $X$ has Gorenstein terminal singularities (these are precisely
1227: the isolated hypersurface singularities with a DuVal section). In particular
1228: $X\subset \F= \p (E)$ is naturally embedded in a rational scroll
1229: over $\p^1$ (a natural choice is $E=f_\ast \o (-K_X)$).
1230: \item $X$ has Picard rank $\rho = 2$, that is, the morphism $f\colon X
1231: \to \p^1$ is extremal.
1232: \item The local rings of $X$ are unique factorisation domains,
1233: that is, Weil divisors on $X$ are Cartier.
1234: \end{enumerate}
1235: \end{conv}
1236: \noindent In addition, we often assume that $X$ is nonsingular, or at
1237: least that $f\colon X \to \p^1$ is \emph{semistable} in the sense of
1238: \cite{MR98g:11076}.
1239:
1240: \begin{rem}
1241: Corti and Koll\'{a}r \cite{MR98e:14037}
1242: \cite{MR98g:11076} show that if $f\colon X\to \p^1$ is a Mori fibre space
1243: $dP_3$ fibration, then $X/\p^1$ is square birational to a semistable
1244: $g\colon Y \to \p^1$; in particular $Y$ is Gorenstein and the
1245: conditions above are satisfied. It is therefore not restrictive to
1246: limit our attention to semistable fibrations.
1247: \end{rem}
1248:
1249: In this section, we aim to do two things. First, we want to determine
1250: the geography of $dP_3$ fibrations, that is, determine all integers
1251: $n,a,b,c$ such that a general member $X \in |3M+nL|$ on $\F(0,a,b,c)$
1252: is a $dP_3$ fibration in our sense.
1253: Second, we want to state some
1254: conjectures on the birational geometry of $dP_3$ fibrations.
1255: M.~Grinenko has been studying del Pezzo
1256: fibrations systematically in a series of recent papers
1257: \cite{grinenko:01a, grinenko:01b, grinenko:01c}. Because he is
1258: primarily concerned with fibres of degree 1 and 2, there is little
1259: overlap between his work and what we do here.
1260:
1261: \subsection{The $K^2$~condition}
1262: \label{sec:k2-cond}
1263:
1264: \begin{dfn} \label{dfn:k2-cond}
1265: We say that $X$ satisfies the \emph{$K^2$~condition} if
1266: \[
1267: K_X^2 \not \in \Ins \NE X.
1268: \]
1269: \end{dfn}
1270:
1271: \begin{thm} \cite{MR98j:14014} \label{thm:dp3}
1272: Let $f\colon X\to \p^1$ be a $dP_3$ fibration
1273: satisfying the following technical conditions:
1274: \begin{itemize}
1275: \item the total space $X$ is nonsingular, and $f$ has Lefschetz
1276: singularities, that is, it has only ordinary critical points,
1277: with distinct critical values.
1278: \item If $X_b$ is a singular fibre, then there are exactly six lines
1279: of $X_b$ passing through the unique singular point.
1280: \end{itemize}
1281: If in addition $X$ satisfies the
1282: $K^2$~condition, then $X$ is birationally rigid.
1283: \end{thm}
1284:
1285: \begin{rem}
1286: It follows from the above theorem that, under its assumptions,
1287: Iskovskikh's conjecture \cite{MR2000i:14019} holds.
1288: \end{rem}
1289:
1290: If a $dP_3$ fibration $X\to S$, with $X$ nonsingular,
1291: is nonrigid, then $X\to S$ belongs to one of finitely
1292: many algebraic families. (This can be proved using
1293: Proposition~\ref{pro:finite}.) On the other hand, as we see in
1294: Section~\ref{sec:geog}, infinitely many families do not
1295: satisfy the $K^2$~condition. This shows that Pukhlikov's theorem is
1296: not optimal.
1297:
1298: \subsection{Conjectures}
1299: \label{sec:dp-con}
1300:
1301: \begin{dfn}
1302: We say that $X$ satisfies the \emph{condition~$(\ast)$} if $-K \not \in \Ins
1303: \NM^1 X$.
1304: \end{dfn}
1305:
1306: \begin{rem}
1307: In Section~\ref{sec:nonrigid} we study (among other things)
1308: condition~$(\ast)$ for the general members of families of
1309: $dP_3$. Though we don't prove it completely, we believe that
1310: condition~$(\ast)$ is satisfied by a general member of all but a
1311: handful of families of $dP_3$ fibrations listed in Table~\ref{tab:2}.
1312: \end{rem}
1313:
1314: Grinenko has recently made the following striking conjecture.
1315:
1316: \begin{con} \cite[Conjecture 1.5]{grin99} \cite[Conjecture 1.6]{grinenko:00}
1317: \cite[Conjecture 2.5]{gri02} \label{con:dp-ast}
1318: A $dP_3$ fibration with nonsingular total space
1319: is birationally rigid if it satisfies condition~$(\ast)$.
1320: \end{con}
1321:
1322: In the remaining part of this subsection we make a few comments on the
1323: meaning of the conjecture.
1324:
1325: \begin{con} \label{con:dp-ast-2}
1326: Same as \ref{con:dp-ast}, only assuming that $X/\p^1$ is
1327: semistable in the sense of \cite{MR98g:11076}.
1328: \end{con}
1329:
1330: \begin{con} \label{con:dp-equiv}
1331: Let $X\to \p^1$ be a $dP_3$ fibration, with $X$ nonsingular. Let
1332: $X^\prime \to \p^1$ be a \mbox{3-fold} Mori fibre space, square birational
1333: to $X/\p^1$. If $X/\p^1$ satisfies condition~$(\ast)$, then so does
1334: $X^\prime/\p^1$.
1335: \end{con}
1336:
1337: \begin{con} \label{con:dp-equiv-2}
1338: Same as \ref{con:dp-equiv}, only assuming that $X/\p^1$ is
1339: semistable.
1340: \end{con}
1341: \noindent More experimentation is needed before we can have any confidence
1342: in these conjectures. Here we only briefly touch on these matters in
1343: the Example in Section~\ref{sec:unstable}.
1344:
1345: \begin{pro}
1346: Conjecture~\ref{con:dp-ast} follows from
1347: Conjecture~\ref{con:dp-equiv}, and Conjecture~\ref{con:dp-ast-2}
1348: from Conjecture~\ref{con:dp-equiv-2}.
1349: \end{pro}
1350:
1351: \noindent This indicates that further progress is likely to
1352: come from a systematic study of square birational maps involving a
1353: semistable $dP_3$ fibration and a Mori fibre space.
1354:
1355: \begin{proof}
1356: We sketch the proof. If $X\to \p^1$ is not rigid, there is
1357: a Mfs $Y \to T$ and a nonsquare birational map $X \dasharrow
1358: Y$. Applying the Sarkisov program as in \cite{MR96c:14013} gives a
1359: Mfs $f^\prime \colon X^\prime \to \p^1$ square birational to
1360: $X\to \p^1$, and a linear system
1361: \[
1362: \mathcal{H}^\prime \subset |-nK_{X^\prime} + f^{\prime \ast} A|
1363: \]
1364: where $A$ is a divisor on $\p^1$ of strictly negative degree. It follows that
1365: $X^\prime$ does not satisfy condition~$(\ast)$.
1366: \end{proof}
1367:
1368: \subsection{Geography for $dP_3$ fibrations}
1369: \label{sec:geog}
1370:
1371: \subsubsection{Notation and basic numerology}
1372: \label{sec:notat}
1373:
1374: \paragraph{Notation}
1375: A $dP_3$ fibration $X\to\p^1$, as defined formally in Convention~\ref{conv:dp3},
1376: always admits model as a hypersurface in a 4-fold scroll $\F$
1377: that is a $\p^3$ bundle over $\p^1$. We fix the notation in use
1378: throughout this section.
1379: \begin{enumerate}
1380: \item $X \in |3M + nL|\subset \F = \F(0,a,b,c)$, where $a,b,c,n$ are
1381: integers with $0\le a \le b \le c$. We write $d=a+b+c$.
1382: \item We write $u,v,x,y,z,t$ for the homogeneous coordinates of $\F$,
1383: where $u,v$ are the homogeneous coordinates on the base $\p^1$ and
1384: $x,y,z,t$ are the fibre coordinates.
1385: \item We denote $L$ and $M$ the natural basis of $\Pic (\F)$. These
1386: line bundles have sections $u\in H^0(\F, L)$ and $x\in H^0(\F, M)$,
1387: and we sometimes identify $L$, $M$ with the actual divisors $u=0$,
1388: $x=0$.
1389: \item We write $\Gamma = \{y=z=t=0\}$ and $B=\{z=t=0\}$. Note that
1390: $\Gamma$ generates an extremal ray of $\NE (\F)$.
1391: \item We denote by $F\in \o_\F (3M+nL)$ the equation of $X$. The
1392: polynomial $F$ is a sum $\sum\alpha m$ of \emph{terms} $\alpha m$
1393: where the sum ranges over the \emph{fibre monomials} $m$ that are
1394: cubics in $x,y,z,t$ with coefficients $\alpha=\alpha(u,v)$ to
1395: fix up homogeneity. We write both $\alpha m\in F$ and $m\in F$ to
1396: mean that the term (with implicit coefficient $\alpha$ if not
1397: mentioned) appears in $F$ with nonzero coefficient.
1398: \end{enumerate}
1399:
1400: \paragraph{Basic numerology}
1401: \begin{enumerate}
1402: \item $M^3L = 1$, $M^4 = d$.
1403: \item $-K_X = -K_{|X}$, where $-K = M + (2 - d - n)L$ is a divisor on $\F$.
1404: \item $X \cdot \Gamma = n$. Apart from the trivial case $X\in |3M|$ on
1405: $\p^1 \times \p^3$ (a constant family of cubic surfaces) there are
1406: two cases:
1407: \[
1408: \begin{cases}
1409: n\geq 0 &\quad \text{and $3M+nL$ is nef and big},\\
1410: n < 0. &
1411: \end{cases}
1412: \]
1413: When $3M+nL$ is nef and big, it is base point free and a general $X
1414: \in |3M+nL|$ is nonsingular. It then follows from the Lefschetz
1415: hyperplane theorem that $\rho (X)=2$ and $X$ is a $dP_3$ fibration in
1416: our sense. Almost everything we say below refers to the much more
1417: interesting case when $n<0$.
1418: \end{enumerate}
1419:
1420: \subsubsection{What the picture says} \label{sec:geog_pic}
1421:
1422: \begin{figure}
1423: \setlength{\unitlength}{1.2mm}
1424: \begin{center}
1425: \begin{picture}(100,155)
1426: % \put(0,0){\framebox(100,155){}}
1427: % labeled axes
1428: \put(0,5){\vector(1,0){95}}
1429: \put(65,0){\vector(0,1){150}}
1430: \put(95,0){\makebox(5,10){$n$}}
1431: \put(60,150){\makebox(10,5){$d$}}
1432: \put(82,0){\makebox(5,5){$2$}}
1433: \put(72,0){\makebox(5,5){$1$}}
1434: \put(64,0){\makebox(5,5){$0$}}
1435: \put(52,0){\makebox(5,5){$-1$}}
1436: \put(42,0){\makebox(5,5){$-2$}}
1437: \put(32,0){\makebox(5,5){$-3$}}
1438: \put(22,0){\makebox(5,5){$-4$}}
1439: \put(12,0){\makebox(5,5){$-5$}}
1440: \put(2,0){\makebox(5,5){$-6$}}
1441: \put(65,40){\makebox(5,10){$4$}}
1442: \put(65,60){\makebox(5,10){$6$}}
1443: \put(65,80){\makebox(5,10){$8$}}
1444: \put(65,100){\makebox(5,10){$10$}}
1445: \put(65,120){\makebox(5,10){$12$}}
1446: \put(65,140){\makebox(5,10){$14$}}
1447: % boundary and pukhlikov line
1448: \put(65,45){\line(3,-5){25}}
1449: \put(65,45){\line(-3,5){65}}
1450: \put(65,25){\line(1,-1){25}}
1451: \put(65,25){\line(-1,1){20}}
1452: \put(45,45){\line(-1,2){10}}
1453: \put(35,65){\line(0,1){10}}
1454: \put(35,75){\line(-1,1){20}}
1455: \put(15,95){\line(-1,2){10}}
1456: \put(5,115){\line(0,1){10}}
1457: \put(5,125){\line(-1,1){5}}
1458: % dots mark dP3s about which we know nothing
1459: \multiput(84.5,4.2)(0,10){15}{$\cdot$}
1460: \multiput(74.5,34.2)(0,10){12}{$\cdot$}
1461: \multiput(64.5,54.2)(0,10){10}{$\cdot$}
1462: \multiput(54.5,64.2)(0,10){9}{$\cdot$}
1463: \multiput(44.5,84.2)(0,10){7}{$\cdot$}
1464: \multiput(34.5,94.2)(0,10){6}{$\cdot$}
1465: \multiput(24.5,114.2)(0,10){4}{$\cdot$}
1466: \multiput(14.5,134.2)(0,10){2}{$\cdot$}
1467: % bullets mark known nonrigid dP3s
1468: \put(74.3,4){$\bullet$}
1469: \put(74.3,14){$\bullet$}
1470: \put(64.3,14){$\bullet$}
1471: \put(64.3,24){$\bullet$}
1472: \put(54.3,34){$\bullet$}
1473: \put(54.3,44){$\bullet$}
1474: \put(44.3,44){$\bullet$}
1475: \put(44.3,54){$\bullet$}
1476: \put(44.3,64){$\bullet$}
1477: \put(34.3,64){$\bullet$}
1478: \put(34.3,74){$\bullet$}
1479: % circles mark known dP3s with a bad (K=0) link
1480: \put(74.3,24){$\circ$}
1481: \put(64.3,34){$\circ$}
1482: \put(64.3,44){$\circ$}
1483: \put(54.3,54){$\circ$}
1484: \put(44.3,74){$\circ$}
1485: \put(34.3,84){$\circ$}
1486: \put(34.3,94){$\circ$}
1487: \put(24.3,84){$\circ$}
1488: \put(24.3,94){$\circ$}
1489: \put(24.3,104){$\circ$}
1490: \put(14.3,94){$\circ$}
1491: \put(14.3,104){$\circ$}
1492: \put(14.3,114){$\circ$}
1493: \put(14.3,124){$\circ$}
1494: \put(4.3,114){$\circ$}
1495: \put(4.3,124){$\circ$}
1496: \put(4.3,134){$\circ$}
1497: \put(4.3,144){$\circ$}
1498: % commentary: links
1499: \put(69,10){\makebox(10,5){{\footnotesize $001$}}}
1500: \put(57,21){\makebox(10,5){{\footnotesize $011$}}}
1501: \put(49,30){\makebox(10,5){{\footnotesize $111$}}}
1502: \put(50,40){\makebox(10,5){{\footnotesize $[112]$}}}
1503: \put(42,47){\makebox(10,5){{\footnotesize $[113]$}}}
1504: \put(39,40){\makebox(10,5){{\footnotesize $112$}}}
1505: \put(40,50){\makebox(10,5){{\footnotesize $122$}}}
1506: \put(28,60){\makebox(10,5){{\footnotesize $123$}}}
1507: \put(40,60){\makebox(10,5){{\footnotesize $[123]$}}}
1508: \put(27,71){\makebox(10,5){{\footnotesize $133$}}}
1509: % commentary: K = 0
1510: \put(49,54){\makebox(10,5){{\footnotesize $(122)$}}}
1511: \put(41,64){\makebox(10,5){{\footnotesize $(222)$}}}
1512: \put(44,75){\makebox(10,5){{\footnotesize $(133)$}}}
1513: \put(32,75){\makebox(10,5){{\footnotesize $(223)$}}}
1514: \put(31,84){\makebox(10,5){{\footnotesize $(233)$}}}
1515: \put(33,94){\makebox(10,5){{\footnotesize $(144)$}}}
1516: \put(23,84){\makebox(10,5){{\footnotesize $(224)$}}}
1517: \put(21,94){\makebox(10,5){{\footnotesize $(234)$}}}
1518: \put(19,104){\makebox(10,5){{\footnotesize $(244)$}}}
1519: \put(13,94){\makebox(10,5){{\footnotesize $(225)$}}}
1520: \put(11,104){\makebox(10,5){{\footnotesize $(235)$}}}
1521: \put(11,114){\makebox(10,5){{\footnotesize $(245)$}}}
1522: \put(15,124){\makebox(10,5){{\footnotesize $(255)$}}}
1523: \put(4,114){\makebox(10,5){{\footnotesize $(236)$}}}
1524: \put(3,124){\makebox(10,5){{\footnotesize $(246)$}}}
1525: \put(1,134){\makebox(10,5){{\footnotesize $(256)$}}}
1526: \put(4,143){\makebox(10,5){{\footnotesize $(266)$}}}
1527: % legend
1528: \put(3,150){\parbox[b]{70mm}{The Pukhlikov line $3d + 5n = 12$}}
1529: \put(0,55){\parbox[b]{30mm}{Symbols:}}
1530: \put(0,50){\parbox[b]{50mm}{$\cdot$ possible $dP_3$ coordinate}}
1531: \put(0,45){\parbox[b]{50mm}{$\circ$ for $X$ with a $K=0$ bad link}}
1532: \put(0,40){\parbox[b]{50mm}{$\bullet$ for known nonrigid $X$}}
1533: \put(0,30){\parbox[b]{30mm}{Labels:}}
1534: \put(0,25){\parbox[b]{70mm}{$abc$ when every $X\subset\F(a,b,c)$ admits the link}}
1535: \put(0,20){\parbox[b]{70mm}{$[abc]$ when only special $X$ admit the link}}
1536: \put(0,15){\parbox[b]{70mm}{$(abc)$ when every $X$ has $K=0$ bad link}}
1537: \end{picture}
1538: \end{center}
1539: \caption{Geography of $|3M+nL|\subset\F(a,b,c)$ with $d=a+b+c$
1540: \label{fig:geog}}
1541: \end{figure}
1542:
1543: We plot the families of $dP_3$ fibrations on
1544: a graph of $n$ against $d$ as in Figure~\ref{fig:geog} and we refer to this
1545: picture as \emph{geography}. We regard the figure as a pictorial
1546: statement of a theorem that we now spell out.
1547:
1548: The figure summarises information
1549: obtained in calculations carried out in the remaining part of the
1550: paper; in particular the figure displays the following.
1551: \begin{enumerate}
1552: \item Pairs $(n,d)$ for which there is a family of $dP_3$ fibrations $X\in
1553: |3M+nL|$ in $\F(0,a,b,c)$ with $d=a+b+c$.
1554: \item Triples $(a,b,c)$ such that a general or a special member of the
1555: corresponding family is known to be nonrigid.
1556: \item Triples $(a,b,c)$ such that a general member of the
1557: corresponding family does not satisfy condition~$(\ast)$. We
1558: do not prove it completely, but we believe that the figure displays
1559: all such triples.
1560: \end{enumerate}
1561:
1562: We plan to use the picture as a primary testing ground for
1563: Grinenko's Conjecture and as a starting point possibly to prove
1564: it. Here we explain what the picture and its
1565: various elements mean. Precise details are worked out in the following
1566: subsections.
1567:
1568: \subparagraph{Families of $dP_3$ fibrations}
1569:
1570: We mark points $(n,d)$ of Figure~\ref{fig:geog} by a dot $\cdot$
1571: if and only if there are values $0\leq a \leq b \leq c $ with
1572: $d=a+b+c$, such that a general $X \in |3M+dL|$ in $\F(0,a,b,c)$ is a
1573: $dP_3$ fibration in our sense. A dot can be a bullet~$\bullet$ or a
1574: circle~$\circ$; we explain what these mean below. The geography
1575: consists of the first quadrant minus the origin, and the region to the
1576: right of a curve with a periodic behaviour which we define by the
1577: picture itself. A point of the picture may house several deformation families:
1578: for example, $(n,d)$ is $(-2,5)$ for a family of $dP_3$ fibrations in
1579: $\F(1,1,3)$ and also for a different family that lies in $\F(1,2,2)$.
1580: We explain how this region is drawn in
1581: Section~\ref{sec:how-geogr-obta} below.
1582:
1583: \subparagraph{Nonrigid families}
1584:
1585: A bullet~$\bullet$ marks a point $(n,d)$ corresponding to
1586: some family of $dP_3$ fibrations for which we know that a member $X$ is
1587: nonrigid. When $X$ is general in its family, we specify the family
1588: by writing $abc$ under the bullet, indicating the
1589: invariants of $\F(0,a,b,c)$. In some cases, we only know that a
1590: \emph{special} $X$ in the family is nonrigid. We indicate this by
1591: writing $[abc]$ instead. In all cases when we know that $X$ is
1592: nonrigid, the 2-ray game from $X/\p^1$ is a nonsquare link of the
1593: Sarkisov program: we refer to this as \emph{the} link
1594: and we say that \emph{$X$ admits the link}. We describe the link
1595: explicitly in Section~\ref{sec:nonrigid} and summarise our calculations in
1596: Table~\ref{tab:2}.
1597:
1598: \subparagraph{Condition~$(\ast)$}
1599:
1600: We do not prove it completely, but we believe that
1601: condition~$(\ast)$ does not hold for a general member
1602: $X\in |3M+nL|$ in $\F(a,b,c)$
1603: if and only if $(n,d)$ has $\bullet$ with $abc$.
1604: After the argument of Section~\ref{sec:mfs_con}, we only study points
1605: to the left of the line $3d + 5n - 12 = 0$; the arguments of
1606: Section~\ref{sec:family5} show immediately that about two thirds
1607: of those with $n<3$ satisfy condition~$(\ast)$, and we believe
1608: the remaining cases to be only slightly more difficult.
1609: Grinenko's
1610: conjecture is then equivalent to the statement that Table~\ref{tab:2}
1611: is a complete list of families of $dP_3$
1612: fibrations with nonrigid general element.
1613:
1614: \subparagraph{$K$ trivial contractions}
1615:
1616: Unless there is already a bullet at this point, we mark $(n,d)$
1617: with a circle $\circ$ when we know that
1618: there is a corresponding family of $dP_3$
1619: fibrations such that the 2-ray game starting with a general member
1620: $X$ terminates with a $K$-trivial contraction and a variety with
1621: strictly canonical singularities. We specify the family by writing
1622: $(abc)$ above the circle. Note that, in each example
1623: of $[abc]$, a general $X$ in the family has a $K$-trivial bad link
1624: and we do not repeat $(abc)$ there. These examples are not our main
1625: interest here, but they are cases where the question of
1626: rigidity is particularly intriguing.
1627:
1628: \subparagraph{The Pukhlikov line}
1629:
1630: The figure also shows the line $3d+5n=12$. As we show in
1631: Lemma~\ref{lem:explicitk2} below, the
1632: general member of a family satisfies the $K^2$~condition
1633: if the point $(n,d)$ lies to the right of this line. (For $n$
1634: negative, this is if and only if, which leaves out a couple of
1635: cases with $n>0$ where we are not completely sure.) The graph shows
1636: that the line leaves out a thin strip of the geography, containing
1637: infinitely many families.
1638:
1639: \subsubsection{How the geography is obtained}
1640: \label{sec:how-geogr-obta}
1641:
1642: In this subsection we make the statements from
1643: which the geography of $dP_3$ fibrations is derived.
1644: \begin{pro} \label{pro:geog}
1645: Figure~\ref{fig:geog} marks all pairs $(n,d)$ for which there are integers
1646: $0\leq a \leq b \leq c$ such that:
1647: \begin{enumerate}
1648: \item The relative surface $B=\{z=t=0\}$ is not contained in the
1649: base locus of the linear system $|3M+nL|$. (When $a=b$, we require
1650: that a general member of the linear system does not contain a
1651: surface of the form $\{s=t=0\}$ where $s$ is some section of $\o_\F (M-bL)$.)
1652: \item The curve $\Gamma = \{y=z=t=0\}$ is contained in the base
1653: locus of $|3M+nL|$ with multiplicity at most one, that is, when
1654: $n<0$, a general member $X$ in the linear system is nonsingular
1655: generically along $\Gamma$.
1656: \end{enumerate}
1657: \end{pro}
1658: \noindent The conditions imply that a general member $X\in |3M+nL|$ is
1659: nonsingular outside $\Gamma$ and possibly has isolated singularities
1660: along $\Gamma$. In fact, $X$ is a $dP_3$ fibration.
1661: \begin{pro} \label{pro:dp3}
1662: Let $(n,d)$ and $X \in |3M+nL|$ in $\F(0,a,b,c)$ be as in
1663: Proposition~\ref{pro:geog}
1664: above. Then:
1665: \begin{enumerate}
1666: \item For any $X$, $\Pic (X)= H^2(X) = \z^2$, that is, $X$ has
1667: Picard rank 2.
1668: \item If all fibres of $f\colon X \to \p^1$ are reduced and
1669: irreducible, we have an exact sequence
1670: \[
1671: 0\to \z [L] \to A_2 (X) \to \Pic (X_\eta) \to 0
1672: \]
1673: where $A_2(X)$ is the group of 2-dimensional cycles, that is, Weil
1674: divisors on $X$ modulo rational equivalence, and $X_\eta$ is the
1675: generic fibre.
1676: \item If $X$ is general, then $\Pic (X_\eta) = \z[-K]$.
1677: \item If $X$ is general, then $X$ has isolated singularities of $cA$
1678: type.
1679: \end{enumerate}
1680: \end{pro}
1681: \begin{cor}
1682: With the same assumptions, if $X$ is general, then $X/\p^1$ is a
1683: $dP_3$ fibration.
1684: \end{cor}
1685:
1686: \paragraph{Proof of Proposition~\ref{pro:geog}}
1687:
1688: The proposition is a direct consequence of the following more precise lemma.
1689: Indeed, it is elementary to check that the inequalities in the lemma specify
1690: the region marked in Figure~\ref{fig:geog}.
1691:
1692: \begin{lem} \label{lem:ineqs}
1693: Consider the linear system $|3M+nL|$ on $\F(0,a,b,c)$ as above.
1694: \begin{enumerate}
1695: \item $B \not \subset \Bs |3M+nL|$ if and only if $n\geq -3a$.
1696: \item If $n<0$ and $a=b$, then all $X$ contain a surface
1697: $t=\ell(z,y)=0$, where $\ell$ is a linear form if and only if $n=-3a$.
1698: \item If $n<0$, so that $\Gamma \subset \Bs|3M+nL|$, then a
1699: general member $X\in |3M+nL|$ is nonsingular generically along
1700: $\Gamma$ if and only if $n\geq c$. In this case also $n+d \geq 2a$.
1701: \end{enumerate}
1702: \end{lem}
1703:
1704: \begin{proof}
1705: If the first inequality fails, then every monomial in $F$ is divisible
1706: by $z$ or $t$ so $X$ contains $B$.
1707:
1708: If $-n=3a>0$ and $a=b$, we can refine this analysis slightly.
1709: We can write $F=f_3(y,z)+tf_2(x,y,z,t)$, where $f_2$, $f_3$ are
1710: homogeneous forms of the indicated degrees. If $\ell(y,z)$ is a linear
1711: factor of $f_3$, then $X$ necessarily contains the surface $t=\ell(z,y)=0$.
1712:
1713: Suppose that $n<0$, so that $a>-(1/3)n>0$ by the first inequality.
1714: The fibre monomial $x^3$ cannot now occur in any
1715: term of $F$, so the inclusion $\Gamma\subset X$ is clear.
1716:
1717: The total space $X$ is nonsingular generically along $\Gamma$ if and
1718: only if, for general values of $u,v$, the polynomial $F$ contains
1719: at least one of the terms $x^2y$, $x^2z$, $x^2t$.
1720: If $x^2t\in F$, then $n \ge -c$,
1721: and that is the weakest inequality of the three. Finally
1722: \[
1723: d + n = (a+b) + (c+n) \ge 2a
1724: \]
1725: with equality only if $n=-c$ and $a=b$.
1726: \end{proof}
1727:
1728: \begin{rem}
1729: Some of our statements in Section~\ref{sec:geog_pic} follow
1730: from Lemma~\ref{lem:ineqs}.
1731: Let $\sigma$ be the closed subcone of $\NM^1(X)$ generated
1732: by the mobile rays $L$, $D_z=(z=0)\cap X$. Then writing
1733: $-K_X = (2-a-c-n)L + D_z$ shows that
1734: \begin{itemize}
1735: \item[(a)]
1736: $2-a-c-n > 0
1737: \quad
1738: \text{if and only if}
1739: \quad
1740: -K_X \in\Ins\sigma
1741: $,
1742: % \quad
1743: % \Longrightarrow
1744: % \quad
1745: % -K_X \in\Ins\NM^1(X)$
1746: \item[(b)]
1747: $2-a-c-n \ge 0
1748: \quad
1749: \text{if and only if}
1750: \quad
1751: -K_X \in\sigma
1752: $.
1753: % \quad
1754: % \Longrightarrow
1755: % \quad
1756: % -K_X \in\NM^1(X)$
1757: \item[(c)]
1758: If condition~$(\ast)$ holds for $(n;a,b,c)$, then $2-a < c+n$.
1759: \end{itemize}
1760: It is easy to solve these inequalities for $a,c,n$ with the results
1761: of Lemma~\ref{lem:ineqs}. In case (a), the solutions are exactly those
1762: 4-tuples $(n;a,b,c)$ for which $(n,d)$ is marked by $\bullet_{abc}$.
1763: In case (b), the solutions are exactly those $(n;a,b,c)$ for which
1764: $(n,d)$ is marked by $\circ^{(abc)}$. Condition~$(\ast)$ holds for
1765: each of the solutions that have $a=2$; it fails only at the special
1766: member of some solutions when $a=1$, which are the [abc] cases.
1767: Part (c), which is virtually no condition at all when $n\ge 0$.
1768: \end{rem}
1769:
1770: \paragraph{Proof of Proposition~\ref{pro:dp3}}
1771:
1772: We briefly sketch the proof, which is standard.
1773: We take on the first assertion first. If $n\geq 0$, then $X$ is nef
1774: and big. Denote $U=\F \setminus X$ the open complement.
1775: The Lefschetz hyperplane theorem implies that
1776: $H^{8-i}_c(U)=H_i(U)=(0)$ for $i>4$ hence $H^2_c(U) = H^3_c (U)=(0)$
1777: and from the standard exact sequence
1778: \[
1779: H^2_c(U) \to H^2 (\F) \to H^2 (X) \to H^3_c(U)
1780: \]
1781: we deduce that $H^2(X)=H^2(\F)=\z^2$. Since $H^1(X, \o_X)=(0)$, it
1782: follows that $\Pic(X)=\z^2$, based by $L$ and $M$. When $n<0$
1783: $X$ is not nef and we can proceed in various ways. For example, we can
1784: observe that the proper transform $X^\prime \subset \F^\prime$ is nef
1785: and big on the variety $\F^\prime$ obtained from $\F$ by flipping the
1786: curve $\Gamma$ (see the Appendix for information on $\F^\prime$). By
1787: Lefschetz then $H^2(X^\prime)= \z^2$. The birational map
1788: $X^\prime \dasharrow X$ is either an isomorphism in codimension 1, or
1789: else $X^\prime$ contains the whole of the flipped set $\p(a,b,c)$. In
1790: either case, we conclude that $H^2 (X)=\z^2$ and the statement follows.
1791:
1792: The exact sequence in 2.\ is standard.
1793:
1794: Consider now the generic fibre $S$ of $X\to\p^1$ which is a nonsingular
1795: cubic surface embedded in $\p^3$ over the function field $\c(s)$
1796: of $\p^1$ with coordinates $x,y,z,t$. See \cite{KSC}
1797: for a tutorial in the elementary methods of such cubics. We claim
1798: that $S$ has Picard rank~1. An example of such $S$ is the surface
1799: \[
1800: S_0 = py^3 + qz^3 + rt^3 + x^2t
1801: \]
1802: where $p,q,r\in \c(s)$ are chosen generally.
1803: It is easy to calculate the 27 lines on such a cubic surface,
1804: and then to figure out the action of the Galois group $\Gal(K/\c(s))$
1805: for any extension $K/\c(s)$ in which the lines are geometric:
1806: for general $p,q,r$, the Galois group is $\z/3\times \z/3 \times \z/2$ and
1807: the 27 lines split into orbits of size 3,6,9,9. One checks
1808: by explicit computation that the smaller orbits are made of
1809: unions of entire hyperplane sections over $K$ so there are no
1810: Galois orbits of disjoint lines. Segre's theorem, \cite{KSC}
1811: Theorem~2.16, then implies that the Picard rank of $S_0$ is 1.
1812: Since a special surface $S_0$ in the family has rank~1, so
1813: does the general surface $S$. This shows statement 3.
1814:
1815: The final claim follows from an elementary monomial argument. Indeed $x^2t\in F$
1816: with coefficient a homogeneous polynomial $\varphi (u, v)$. The
1817: singularities of $X$ along $\Gamma$ are at the zeros of $\varphi$
1818: and if $X$ is general we may assume that these all have
1819: multiplicity 1. If $u|\varphi$, for example, the singularity at
1820: $u=0$ is of the form $ut+\cdots =0$. This implies that the general
1821: surface section has a singularity of type $A_k$ there. \qed
1822:
1823:
1824: \subsubsection{The $K_X^2$ condition and Pukhlikov's Theorem}
1825:
1826: \begin{lem}
1827: \label{lem:explicitk2}
1828: If $X\to\p^1$ is a $dP_3$ fibration, then
1829: \[
1830: K_X^2\in\Ins\NE(X)
1831: \quad
1832: \text{implies that}
1833: \quad
1834: 3d + 5n < 12.
1835: \]
1836: If in addition $n<0$, then
1837: \begin{itemize}
1838: \item[(a)]
1839: the 1-cycles $M^2L$ and $\Gamma\subset X$ bound
1840: the Mori cone $\NE(X)$
1841: \item[(b)]
1842: $K_X^2\in\Ins\NE(X)$ if and only if $3d + 5n < 12$.
1843: \end{itemize}
1844: \end{lem}
1845: \begin{proof}
1846: Suppose first that $n<0$.
1847: The general fibre of $X\to\p^1$ is a cubic surface, so $M^2L$,
1848: a line in the fibre, lies on one extreme ray of $\NE(X)$.
1849: Since $a>0$, there is a projective morphism from the scroll $\F$
1850: that contracts $\Gamma$, so the curve $\Gamma\subset X$ generates
1851: the other ray of $\NE(X)$, proving (a).
1852:
1853: Part (b) follows from (a) by calculating the 1-cycle $K_X^2$ in the
1854: basis $\Gamma, M^2L$. Denoting $i\colon X \hookrightarrow\F$ for the
1855: inclusion and noting that
1856: \[
1857: \Gamma=(M-aL)(M-bL)(M-cL)=M^3 - dM^2L,
1858: \]
1859: we calculate $i_\ast K_X^2=(-K)^2X$ on $\F$ as
1860: \[
1861: \begin{array}{rcl}
1862: \bigl(M + (2 - d - n)L\bigr)^2(3M + nL) & = & 3M^3 + \bigl(6(2-d-n) +
1863: n\bigr)M^2L \\
1864: & = & 3\Gamma + (12 - 3d - 5n)M^2L.
1865: \end{array}
1866: \]
1867:
1868: Now let $n$ be any integer. Suppose that $K_X^2\in\Ins\NE(X)$. We
1869: calculate in ${\mathrm N}_1(\F)$ omitting $i_\ast$. Let $\sigma$ be the
1870: closed subcone of effective divisors spanned by $K_X^2$ and $M^2L$.
1871: If $\Gamma\notin\sigma$, then $K_X^2$ is a strictly convex combination
1872: of $\Gamma$ and $M^2L$ so the inequality holds. If $\Gamma\in\sigma$,
1873: then some multiple $\Sigma=k\Gamma$ of the class of $\Gamma$ (on $\F$)
1874: contains an effective curve that lies on $X$. But $\Gamma\subset\F$
1875: can be contracted, so $\Sigma$ must be supported on $\Gamma$.
1876: Therefore $\Gamma\subset X$ and $\Gamma$ generates an extremal ray
1877: of $\NE(X)$. The inequality follows as before.
1878: \end{proof}
1879:
1880: \subsection{Nonrigid $dP_3$ fibrations}
1881: \label{sec:nonrigid}
1882:
1883: We summarise all the examples of nonrigid $dP_3$ fibrations that
1884: we know in Table~\ref{tab:2}. We do not dwell on the well-known cases:
1885: \begin{itemize}
1886: \item
1887: $(n,d)=(0,1)$ links to the cubic \mbox{3-fold} $Y=Y_3\subset\p^4$,
1888: the $dP_3$ fibration being the pencil of $\p^3$s through a plane
1889: intersecting $Y$ in a cubic curve
1890: \item
1891: $(n,d)=(1,0)$ links to $Y=\p^3$, the $dP_3$ fibration being the
1892: pencil of any pair of transverse cubic surfaces in $Y$
1893: \item
1894: $(n,d) = (1,1)$ links to $Y_{3,3}\subset\p(1^5,2)$, the $dP_3$
1895: fibration being a pencil of divisors having maximal vanishing
1896: at the singular point; see \cite{cm02}, \cite{BZ03}.
1897: \end{itemize}
1898: We discuss the table below and then make three detailed studies
1899: of examples.
1900:
1901: \begin{table}[ht]
1902: \begin{center}
1903: \renewcommand{\arraycolsep}{3mm}
1904: $
1905: \begin{array}{|c|ccc|l|l|}
1906: \hline
1907: \text{No.} & n & a,b,c & \mu & \text{Link of $-\mu K_X-L$}
1908: & \text{other model} \\
1909: \hline
1910: \hline
1911: 1 & 1 & 0,0,1 & 3 & \text{9-flop then $(2,0)$ to}
1912: & Y^\prime_{3,3}\subset\p^5(1^5,2) \\
1913: & & & & \text{$\frac{1}{2}(1,1,1)$ singularity}
1914: & \text{general in its family} \\
1915: \hline
1916: 2 & 0 & 0,1,1 & 1 & \text{3-flop} & \text{$dP_3$ fibration, same} \\
1917: & & & & & \text{numerology as $X$} \\
1918: \hline
1919: 3 & -1 & 1,1,1 & 1 & \text{flop} & \text{conic bundle over }\p^2 \\
1920: & & & & & \text{with $\deg\Delta = 7$} \\
1921: \hline
1922: 4 & -2 & 1,1,2 & 1 & \text{flop then $(2,1)$ to} &
1923: \text{$Y^\prime_4\subset \p^4(1^4,2)$} \\
1924: & & & & \text{linear }\p^1\cong\ell\subset Y^\prime
1925: & \\
1926: \hline
1927: 5 & -2 & 1,2,2 & 1 & \text{Francia antiflip}
1928: & \text{$dP_2$ fibration with} \\
1929: & & & & \text{then flop} & \text{$\frac{1}{2}(1,1,1)$ on 1 fibre} \\
1930: \hline
1931: 6 & -3 & 1,2,3 & 1 & \text{Francia antiflip then} &
1932: Y^\prime_6\subset \p^4(1^3,2,3) \\
1933: & & & & \text{$(2,0)$ to $P\in Y^\prime$}
1934: & \text{$P$ a $c\mathrm{D}_4$ singularity} \\
1935: \hline
1936: 7 & -3 & 1,3,3 & 1 & \text{toric antiflip}
1937: & \text{$dP_1$ fibration with} \\
1938: & & & & (1,1,-1,-3) & \text{$\frac{1}{3}(1,1,2)$ on 1 fibre} \\
1939: \hline
1940: \hline
1941: 8a & -1 & 1,1,2 & 5
1942: & (1,1,-1,-1,-3),
1943: & Y^\prime\subset\p(1^4,2,3,4) \\
1944: & & &
1945: & \text{7-flop}, (2,0)
1946: & \text{general}, P=\frac{1}{4}(1,1,3) \\
1947: \hline
1948: 8b & -1 & 1,1,2 & 3
1949: & (1,1,-1,-1,-4),
1950: & \text{$Y^\prime\subset\p^5(1^4,2^2)$\,?} \\
1951: & & &
1952: & \text{3-flop, $(2,0)$}
1953: & \\
1954: \hline
1955: 9 & -2 & 1,1,3 & 3
1956: & (1,1,-1,-1,-4),
1957: & \text{$Y^\prime\subset\p^5(1^2,2^2,3,5)$\,?} \\
1958: & & &
1959: & \text{3-flop, $(2,0)$}
1960: & \\
1961: \hline
1962: 10 & -2 & 1,2,3 & 3
1963: & (1,1,-1,-2,-7),
1964: & \text{$Y^\prime\subset\p^5(1^2,2^3,3)$\,?} \\
1965: & & & & (2,0) & \\
1966: \hline
1967: \end{array}
1968: $
1969: \end{center}
1970: \caption{Nonrigid $dP_3$ fibrations $X\in |3M+nL|\subset\F(0,a,b,c)$
1971: \label{tab:2}}
1972: \end{table}
1973:
1974: \subsubsection{Overview of Table~\ref{tab:2}}
1975:
1976: Each entry of Table~\ref{tab:2} represents a family of $dP_3$ fibrations
1977: \[
1978: X\in |3M+nL| \subset \F(0,a,b,c)
1979: \]
1980: for which the 2-ray game on some member $X$ results in a
1981: Sarkisov link to another model of $X$ as a Mfs. The general member $X$
1982: of families~1--7 is nonsingular. In families~3--7, every $X$ (not just
1983: the general member) admits the link as described.
1984: Families~8--10 necessarily have a singularity on $\Gamma$, which
1985: is described (in new coordinates) for general $X$ as follows.
1986: \begin{center}
1987: \renewcommand{\arraycolsep}{3mm}
1988: $
1989: \begin{array}{|l|c|c|c|c|}
1990: \hline
1991: \text{No.} & 8a & 8b & 9 & 10 \\
1992: \hline
1993: \text{Equation} & xy=zt & xy = z^3 + t^3 & xy = z^3 + t^3 & xy = z^3 + t^6 \\
1994: \hline
1995: \end{array}
1996: $
1997: \end{center}
1998: In families~8--10, only special members $X$ admit the link as
1999: described.
2000:
2001:
2002: Our method for calculating the other model is to calculate the
2003: graded ring of $-\mu K_X - L$. In these four cases, though,
2004: this does not present the other model well, since, we believe,
2005: the contraction is to a non-Gorenstein singularity $P\in Y^\prime$.
2006: We have not yet made the required calculations, but it seems
2007: likely that the other model in families 8b, 9, 10 is better
2008: presented as a complete intersection in weighted projective
2009: spaces $\p^5(1^4,2^2)$, $\p^5(1^2,2^2,3,5)$ and $\p^5(1^2,2^3,3)$
2010: respectively, the index of $P\in Y^\prime$ being 2, 5, 2 respectively.
2011:
2012: We draw attention to family~2. This links pair of $dP_3$ fibrations
2013: $X\dasharrow X'$ that lie in the same family. We guess that, in general,
2014: they are not isomorphic and that they form a bi-rigid pair. Indeed,
2015: if they were isomorphic, this link would be an `untwisting' link
2016: (in the sense of \cite{cpr}), and we would be inclined to guess
2017: that $X$ is rigid. We know little about this case.
2018:
2019: The numbering of cases in the first column is arbitrary. Apart from
2020: this, the information for each entry is separated into three columns,
2021: and we describe the contents of each of these in turn:
2022: \begin{enumerate}
2023: \item
2024: This lists the integers $n$ and $a,b,c$ that determine the family
2025: in question, as well as the $\mu>0$ for which
2026: $-\mu K_X - L$ determines an edge of the mobile cone $\NM^1(X)$.
2027: Note that, as Conjecture~\ref{con:dp-ast} predicts, condition~$(\ast)$
2028: is not satisfied by these examples
2029: \item
2030: This describes the link. We express the antiflips as $\cstar$ quotients
2031: by listing the characters of a $\cstar$ action; we comment further below.
2032: The word `flop' means the flop of a single rational curve. We say
2033: `$n$-flop' when, for general $X$, an analytic neighbourhood of the
2034: flopping curve consists of the disjoint union of $n$ rational curve
2035: neighbourhoods.
2036: The notation $(2,p)$ indicates a divisorial contraction
2037: to a point (when $p=0$) or a line (when $p=1$).
2038: \item
2039: This final column gives a sketch of the other model.
2040: \end{enumerate}
2041: We say more about the antiflips. The Francia antiflip replaces a
2042: $\p^1$ in the nonsingular locus having normal bundle
2043: ${\cal O}(-1)\oplus{\cal O}(-2)$ by a $\p^1$ passing through an
2044: index~2 terminal quotient singularity. This is worked out in
2045: Section~\ref{sec:family5}. The notation $(1,1,-1,-3)$ denotes a
2046: toric \mbox{3-fold} antiflip similar to the Francia flip, but with an
2047: index~3 singularity. An example of the hypersurface antiflips
2048: $(1,1,-1,-1,-m)$ is described in Section~\ref{sec:family8}:
2049: typically, these replace a single rational curve passing through a
2050: terminal Gorenstein point with a bouquet of $m-1$ rational curves
2051: meeting in an index~$m$ singular point. See \cite{MR2000f:14018}
2052: for details and lists of these flips.
2053:
2054: \paragraph{Other models as strict Mfs}
2055: Four cases link to another strict Mfs. Family~2 is discussed above,
2056: and family~3 is linked to the conic bundle of Section~\ref{sec:cb7-2}.
2057:
2058: The other two, families~5 and 7, share two novelties.
2059: First, the use of weighted scrolls
2060: for describing $dP_1$ and $dP_2$ fibrations contrasts with Grinenko's
2061: use of finite morphisms to nonsingular scrolls.
2062: As calculated in Section~\ref{sec:family5}, the other model of family~5
2063: is the general element
2064: \[
2065: X^\prime\in |4M-L|
2066: \quad
2067: \text{in the weighted scroll}
2068: \quad
2069: \left(
2070: \begin{array}{rrrrrr}
2071: 0 & 0 & 1 & 2 & 1 & 1 \\
2072: 1 & 1 & 0 & -1 & -1 & -1
2073: \end{array}
2074: \right).
2075: \]
2076: Similarly, one sees the other model of family~7 as the general element
2077: \[
2078: X^\prime\in |6M-3L|
2079: \quad
2080: \text{in the weighted scroll}
2081: \quad
2082: \left(
2083: \begin{array}{rrrrrr}
2084: 0 & 0 & 2 & 3 & 1 & 1 \\
2085: 1 & 1 & -1 & -2 & -1 & -1
2086: \end{array}
2087: \right).
2088: \]
2089: Second, these examples present nonrigid $dP_1$ and $dP_2$ fibrations.
2090: Grinenko \cite{gri02} has complete classifications of such nonrigid $dP_k$
2091: in the Gorenstein case, and the nonrigid examples are rare. The specimens
2092: here have singularities of index~2 and 3, so they are not subject to
2093: Grinenko's classifications, but they do invite one to extend Grinenko's
2094: results to the higher index case.
2095:
2096: \subsubsection{Family~5: general members are nonrigid}
2097: \label{sec:family5}
2098:
2099: We work out the link arising from the 2-ray game for
2100: family~5 using the wall-crossing methods of Section~\ref{sec:cb7-3}
2101: and the Appendix. The calculations are similar for families~2--7.
2102: Consider any $dP_3$ fibration
2103: \[
2104: X\colon (F=0) \in |3M - 2L| \subset \F = \F(0,1,2,2).
2105: \]
2106: The polynomial $F$ is a combination of the monomials of the Newton polygon
2107: \[
2108: \begin{array}{cc}
2109: \text{$\deg_{u,v}$} & \text{fibre monomial} \\
2110: \hline
2111: 0 & xy^2, x^2z, x^2t \\
2112: 1 & y^3, xyz, xyt \\
2113: 2 & y^2z, y^2t, xz^2, xzt, xt^2 \\
2114: 3 & yz^2, yzt, yt^2 \\
2115: 4 & z^3, z^2t, zt^2, t^3
2116: \end{array}
2117: \]
2118: where $\deg_{u,v}$ denotes the degree in $u,v$ of the coefficient.
2119: We know that $F$ must involve both $xy^2+\varphi_1(u,v)y^3$
2120: and $x^2z + x^2t$ nontrivially.
2121: % , and we may assume $F\owns uy^3$ and $F\owns x^2t$.
2122: We continue to use the notation $\Gamma\colon (y=z=t=0)\subset X$.
2123:
2124: The chamber defining $\F$ is $\r_+[L]+\r_+[M]$. Crossing the wall
2125: into the next chamber $\r_+[M]+\r_+[M-L]$ corresponds to a birational
2126: map $\F\dasharrow\F_1$ that factors through the contraction of
2127: $\Gamma\subset\F$ and the extraction of a surface
2128: $E_1\colon (u=v=0)\subset\F_1$. In fact, $E_1\cong\p(1,2,2)$ with
2129: coordinates $y,z,t$.
2130:
2131: The birational image $X_1\subset\F_1$ of $X$ is still $F=0$ so
2132: $X_1\cap E_1$ is a $\p(1,2)$: when $u=v=0$ we can set $x=1$, as usual
2133: in projective geometry, so the intersection is $y^2+z+t=0$ in $\p(1,2,2)$.
2134: We see that $-K\Gamma = (M-L)\Gamma = -1$, so the map
2135: $X\dasharrow X_1$ is an antiflip of $\Gamma$ and the link proceeds
2136: with $X_1$ if and only if $X_1$ has terminal singularities.
2137: We check this condition in coordinates.
2138:
2139: We calculate one patch of $X_1$ in detail, and leave the others to
2140: the reader. The unstable locus of the quotient defining $\F_1$ is
2141: \[
2142: (u=v=x=0) \cup (y=z=t=0)
2143: \]
2144: so the open set $xz\not= 0$ is a well-defined affine patch $U$ on $\F_1$.
2145: There is a residual $\z/2$ action (the stabiliser of $z$-axis by the
2146: $\mu$-action) defining the chart $\a^4\to U$ that acts on coordinates
2147: $u,v,y,t$ of $\a^4$ by the character $(1,1,1,0)$. The equation of $X_1$
2148: in $U$ includes the monomial $t$, and the equation of the antiflipped
2149: curve $\Gamma_1\subset X_1$ is $u=v=t=0$. An analytic neighbourhood of
2150: $X_1$ near the origin in $U$ is isomorphic to $t=0$ in the quotient
2151: $\a^4/(\z/2)(1,1,1,0)$, and this is terminal.
2152: The flip $X_1\to X$ is the Francia flip.
2153:
2154: We cross to the next chamber giving $\F_1\dasharrow\F^\prime$
2155: which factors as the contraction of $\p^2\subset\F_1$ (with
2156: coordinates $u,v,x$) and the extraction of a $\p^1\subset\F^\prime$
2157: (with coordinates $z,t$).
2158: The birational image of $X$ is $X^\prime\subset\F^\prime$.
2159: Since $F$ involves $xy^2 + uy^3$, the intersection of $X$ with
2160: the exceptional $\p^2$ is a line $\Sigma_1$. Clearly $-K\Sigma_1=0$
2161: so the map $X_1\dasharrow X^\prime$ is a flop and, in particular, $X^\prime$ has
2162: terminal singularities.
2163:
2164: The variety $\F^\prime$ has a morphism $\F^\prime\to\p^1$ given by the ratio
2165: $z,t$; fibres are $\p(1,1,2,1)$ with coordinates $u,v,x,y$.
2166: To see this, make row operations on the character matrix
2167: (basis changes in $\c^\times\times\c^\times$ that do not alter the quotients):
2168: \[
2169: \left(
2170: \begin{array}{rrrrrr}
2171: 0 & 0 & 1 & 1 & 1 & 1 \\
2172: 1 & 1 & 0 & -1 & -2 & -2
2173: \end{array}
2174: \right)
2175: \sim
2176: \left(
2177: \begin{array}{rrrrrr}
2178: -1 & -1 & -1 & 0 & 1 & 1 \\
2179: 1 & 1 & 2 & 1 & 0 & 0
2180: \end{array}
2181: \right).
2182: \]
2183: The right-hand matrix is better adapted to the unstable locus
2184: $(u=v=x=y=0) \cup (z=t=0)$ for $\F^\prime$. The appropriate basis of
2185: $\Pic(\F^\prime)$ is $M^\prime$, $L^\prime$, the line bundles corresponding
2186: to the characters $0\choose 1$ and $1\choose 0$ in the basis of the
2187: right-hand matrix.
2188:
2189: The induced map $X^\prime\to\p^1$ describes $X^\prime$ as a $dP_2$ fibration:
2190: since $x^2t\in F$, it is clear that
2191: \[
2192: X^\prime\subset |4M^\prime - L^\prime| \subset \F^\prime.
2193: \]
2194: One can check that $X^\prime$ is a general in its family if $X$ was
2195: to start with.
2196:
2197: \paragraph{}
2198: In this example, every step of the 2-ray game on $X$ was inherited
2199: (by computing birational images) from the steps of the 2-ray game on
2200: $\F$, itself an easy toric calculation. Most examples we know
2201: have this feature.
2202: \begin{dfn}
2203: Let $X\subset\F$ be a $dP_3$ fibration in a scroll $\F\to\p^1$ (of
2204: any dimension).
2205: We say that \emph{the link follows the scroll} if the birational
2206: images of $X$ in the 2-ray game of $\F$, together with the birational maps
2207: induced between them, make up the 2-ray game of $X$.
2208: \end{dfn}
2209: This property is one of the characteristics of \emph{Mori dream spaces},
2210: as introduced by Keel--Hu \cite{MR2001i:14059}. Our less precise
2211: definition is a convenient shorthand for the purposes of this paper only.
2212:
2213: \subsubsection{Family~8: special members are nonrigid}
2214: \label{sec:family8}
2215:
2216: Consider $dP_3$ fibrations in the family
2217: \[
2218: X\colon (F=0) \in |3M-L|\subset\F = \F(0,1,1,2).
2219: \]
2220: The polynomial $F$ is a combination of the monomials of the Newton polygon
2221: \[
2222: \begin{array}{cc}
2223: \text{$\deg_{u,v}$} & \text{fibre monomial} \\
2224: \hline
2225: 0 & x^2y, x^2z \\
2226: 1 & xy^2, xyz, xz^2, x^2t \\
2227: 2 & y^3, y^2z, yz^2, z^3, xyt, xzt \\
2228: 3 & y^2t, yzt, z^2t, xt^2 \\
2229: 4 & yt^2, zt^2 \\
2230: 5 & t^3
2231: \end{array}.
2232: \]
2233: We define
2234: \[
2235: {\mathrm{val}}(F) =
2236: \min\{ \deg_{u,v} m\ |\ m\in F \text{ a monomial with nonzero coefficient} \}
2237: \]
2238: to separate cases of $F$ according to their leading term in $u,v$.
2239:
2240: \paragraph{Case 1: ${\mathrm{val}}(F) = 0$.}
2241:
2242: This is the general case: the coefficient of $x^2y$ or $x^2z$ in $F$ is
2243: not zero and so the birational link of such $X$ follows the scroll.
2244: Since $-K_X=M-L$, one calculates that the final divisorial contraction
2245: is trivial against $-K_X$, so this is a bad link on $X$.
2246:
2247: \paragraph{Case 2: ${\mathrm{val}}(F) = 1$.}
2248:
2249: Every term of $F$ is divisible by $u$ or $v$ so we write
2250: \[
2251: F=uf - vg
2252: \quad
2253: \text{with}
2254: \quad
2255: f,g\in |3M-2L|.
2256: \]
2257: Since $u$ and $v$ do not vanish simultaneously on $X$, the rational
2258: section
2259: \begin{equation}
2260: \label{eq:unproj}
2261: \xi = f/v = g/u
2262: \quad
2263: \text{of}
2264: \quad
2265: {\cal O}_X(3M-3L)
2266: \end{equation}
2267: is regular: $\xi\in H^0(X,3M-3L)$. We regard $\xi$ as a new variable
2268: and, using equations (\ref{eq:unproj}), recompute the link starting from
2269: \[
2270: X \colon
2271: (v\xi = f, u\xi = g)
2272: \in |3M-2L| \cap |3M-2L| \subset
2273: \F^5
2274: \]
2275: where the weighted scroll $\F^5$ is defined by the $\cstar\times\cstar$ action
2276: \[
2277: \left(
2278: \begin{array}{rrrrrrr}
2279: 0 & 0 & 1 & 1 & 1 & 3 & 1 \\
2280: 1 & 1 & 0 & -1 & -1 & -3 & -2
2281: \end{array}
2282: \right)
2283: \]
2284: on $\a^7$ with coordinates $u,v,x,y,z,\xi,t$. We have re-embedded $X$
2285: isomorphically in a larger scroll: projection from $\xi$ is the
2286: isomorphism to the original embedding. Necessarily $F\owns x^2t$, since
2287: otherwise $X$ would be singular along $\Gamma$ (the negative section
2288: $y=z=\xi=t=0$), and then either $f\owns x^2t$, or $g\owns
2289: x^2t$. Possibly after renaming $u,v$, we may assume that $f\owns x^2t$.
2290: This case now divides into two subcases.
2291:
2292: \paragraph{Case 2a: ${\mathrm{val}}(F) = 1$, ${\mathrm{val}}(g) = 0$.}
2293:
2294: We can write the equations of $X$ in the form
2295: \begin{equation}
2296: \label{eq:2a}
2297: \left(
2298: \begin{array}{ccc}
2299: f_2 & f_1 + \xi & f_3 \\
2300: g_1 + \xi & g_2 & g_3 \\
2301: \end{array}
2302: \right)
2303: \left(
2304: \begin{array}{c}
2305: u \\ v \\ x
2306: \end{array}
2307: \right)
2308: = 0.
2309: \end{equation}
2310: Following the link of $\F^5$, we see $X\dasharrow X_1$. We solve
2311: (\ref{eq:2a}) by
2312: \begin{equation}
2313: \label{eq:eta}
2314: \eta =
2315: \frac{f_2g_2 - (f_1+\xi)(g_1+\xi)}{x} =
2316: \frac{f_2g_3-f_3(g_1+\xi)}{v} =
2317: \frac{(f_1+\xi)g_3-f_3g_2}{u}
2318: \end{equation}
2319: and conclude that $\eta \in H^0(X_1,5M-6L)$ since the semistable locus of
2320: the action defining $X_1$ does not include $u=v=x=0$. Once more, we
2321: make a new scroll $\F^6$, by including $\eta$ among the coordinates,
2322: and re-embed $X\subset\F^6$:
2323: \[
2324: X \colon ({\mathrm{Pfaff}}_{4\times 4}M = 0) \subset
2325: \F^6
2326: \]
2327: where $\F^6$ is defined by the $\cstar\times\cstar$ action
2328: \[
2329: \left(
2330: \begin{array}{rrrrrrrr}
2331: 0 & 0 & 1 & 1 & 1 & 3 & 5 & 1 \\
2332: 1 & 1 & 0 & -1 & -1 & -3 & -6 & -2
2333: \end{array}
2334: \right)
2335: \]
2336: on $\a^8$ with coordinates $u,v,x,y,z,\xi,\eta,t$. Manipulating
2337: (\ref{eq:eta}), the equations of $X\subset\F^6$ are the $4\times 4$
2338: Pfaffians of a $5\times 5$ skew symmetric matrix
2339: \[
2340: M =
2341: \left(
2342: \begin{array}{cccc}
2343: \eta & f_2 & f_1 + \xi & f_3 \\
2344: & g_1 + \xi & g_2 & g_3 \\
2345: & & x & -v \\
2346: & & & u
2347: \end{array}
2348: \right).
2349: \]
2350: The 2-ray game of $X$ follows the link of the scroll $\F^6$.
2351: We sketch the steps:
2352: \begin{enumerate}
2353: \item
2354: Antiflip $X\dasharrow X_1$. This factors as $X\to Z_1\leftarrow X_1$,
2355: by the contraction of $\Gamma$ to $Z_1$, both maps being the restrictions
2356: of maps of the scroll. The exceptional locus $u=v=0$ of
2357: $Z_1\leftarrow X_1$ is defined by the equations of $X_1$ in
2358: $\p^4(1,1,3,5,1)$ with coordinates $y,z,\xi,\eta,t$, since that $\p^4$ is
2359: exceptional in the scroll. We set $x=1$ to see that $t=0$, since
2360: $f_3\owns xt$, and also that $\eta$ is eliminated.
2361: So antiflipped locus is
2362: \[
2363: \bigl\{\text{three $\p(1,3)$s meeting in a point}\bigr\} =
2364: \bigl(g_3(y,z) = 0\bigr)\subset\p(1,1,3).
2365: \]
2366: One can check in coordinates, as in Section~\ref{sec:family5}, that this
2367: point is an index~3 terminal singularity on $X_1$.
2368: In Table~\ref{tab:2}, this antiflip is denoted by $(1,1,-1,-1,-3)$.
2369: \item
2370: Flop $X_1\dasharrow X^\prime$. Since $-K_{X^\prime}=M-L$, curves
2371: that are contracted are trivial against $-K_{X^\prime}$.
2372: See \cite{BZ03} for a detailed calculation, that also counts the
2373: number of contracted curves.
2374: \item
2375: Divisorial contraction to a point $X^\prime\to Y^\prime$.
2376: The divisor $F=(t=0)$ on the scroll is contracted to a point, so
2377: $F\cap X^\prime$ is too. The linear system $|6M-5L|$ (and its multiples)
2378: define the morphism, so since
2379: \[
2380: -6K_X = 6(M-L) = (6M-5L) + L
2381: \]
2382: is negative on only the flipped curves, it must be relatively ample.
2383: Therefore $X^\prime\to Y^\prime$ is extremal, and $Y^\prime$ has
2384: terminal singularities.
2385: \end{enumerate}
2386: The result $Y^\prime$ is No.\ 6 of Alt{\i}nok's list of codimension~3
2387: Fano \mbox{3-folds} \cite{alt98}. This Sarkisov link is already known to us,
2388: calculated from the $Y^\prime$ end of the link in \cite{BZ03} following
2389: Example~9.16 of \cite{kinosaki}.
2390:
2391: \paragraph{Case 2b: ${\mathrm{val}}(F) = 1$, ${\mathrm{val}}(g) = 1$.}
2392:
2393: Now $g$ is further specialised:
2394: \[
2395: g=uf_1 + vg_1
2396: \quad
2397: \text{with}
2398: \quad
2399: f_1,g_1\in |3M-3L|.
2400: \]
2401: As usual, since $u,v$ do not vanish simultaneously on $X$ and $u\xi - g$
2402: is identically zero we write
2403: \begin{equation}
2404: \label{eq:unproj2}
2405: u\xi = uf_1 + vg_1
2406: \quad
2407: \text{and}
2408: \quad
2409: \eta = (\xi - f_1)/v = g_1/u
2410: \end{equation}
2411: and conclude that $\eta\in H^0(X,3M-4L)$. Again we include $\eta$ as
2412: a new variable and compute the link again starting from
2413: \begin{multline*}
2414: X \in |3M-2L| \cap |3M-3L| \subset \F^5 \\
2415: \text{determined by the action}
2416: \quad
2417: \left(
2418: \begin{array}{rrrrrrr}
2419: 0 & 0 & 1 & 1 & 1 & 3 & 1 \\
2420: 1 & 1 & 0 & -1 & -1 & -4 & -2
2421: \end{array}
2422: \right).
2423: \end{multline*}
2424: Unlike case 2a, here we have eliminated $\xi$ from the coordinates
2425: using the equation $\xi = v\eta + f_1$ of (\ref{eq:unproj2}).
2426: Substituting for $\xi$ calculates the equations of $X\subset\F^5$:
2427: \[
2428: v^2\eta + vf_1 = f,\quad
2429: u\eta = g_1.
2430: \]
2431: The 2-ray game of $X$ follows the link of the scroll $\F^5$.
2432: The final contraction in the link is given by the linear system
2433: $|3M-4L|$ and its multiples on $X^\prime$.
2434:
2435: \subsubsection{Unstable $dP_3$ fibrations and condition $(\ast)$}
2436: \label{sec:unstable}
2437:
2438: Consider the family of $dP_3$ fibrations
2439: \[
2440: X\colon (F=0) \in |3M-4L|\subset\F(0,2,2,4).
2441: \]
2442: A general $X$ has a 2-ray game that follows the scroll, but this
2443: gives a $K_X$-trivial bad link. Nevertheless, we find
2444: a special $X$ which is nonrigid.
2445:
2446: The polynomial $F$
2447: is a combination of the monomials of the Newton polygon
2448: \[
2449: \begin{array}{cc}
2450: \text{degree} & \text{fibre monomial} \\
2451: \hline
2452: 0 & xy^2, xyz, xz^2, x^2t \\
2453: 2 & y^3, y^2z, yz^2, z^3, xyt, xzt \\
2454: 4 & y^2t, yzt, z^2t, xt^2 \\
2455: 6 & yt^2, zt^2 \\
2456: 8 & t^3
2457: \end{array}
2458: \]
2459: and we impose conditions on the coefficient polynomials $\alpha(u,v)$:
2460: we require $u^i$ to divide the coefficient of fibre monomials according
2461: to the table
2462: \[
2463: \begin{array}{cc}
2464: i & \text{fibre monomial} \\
2465: \hline
2466: 1 & xyt, xzt \\
2467: 2 & y^2t, yzt, z^2t \\
2468: 3 & xt^2 \\
2469: 4 & yt^2, zt^2 \\
2470: 6 & t^3
2471: \end{array}
2472: \]
2473: with coefficients otherwise general. The reader can check that a
2474: general such $X$ is nonsingular away from a $c{\mathrm{D}}_4$
2475: singularity at the point $(0,1;0,0,0,1)$.
2476: Note that $X$ satisfies condition~$(\ast)$ because, even though it is
2477: special in the family, the defining equation of $X$ involves $x^2t$
2478: with nontrivial coefficient, hence it is still true that the 2-ray
2479: game from $X$ follows the scroll (and ends in a bad link).
2480:
2481: Following Corti--Koll\'ar \cite{MR98e:14037}, \cite{MR98g:11076},
2482: $X$ is unstable with respect to the
2483: weight system $w = (3,2,2,0)$. Indeed, $F$ is divisible by $u^6$
2484: after the substitution
2485: \[
2486: u^3x^\prime, u^2y^\prime, u^2z^\prime, t^\prime
2487: \quad
2488: \text{for}
2489: \quad
2490: x,y,z,t.
2491: \]
2492: Cancelling the $u^6$ factor gives a $dP_3$ fibration
2493: \[
2494: X_{\mathrm{st}}\in |3M-L|\subset\F(0,1,1,1)
2495: \]
2496: that is square birational to $X\to\p^1$. Note that the fibre at $u=0$
2497: has an Eckardt point:
2498: \[
2499: X_{\mathrm{st}}\cap (u=0) = \bigl(tf_2(x,y,z) = g_3(y,z)\bigr).
2500: \]
2501: Even though it is not general in its family, $X_{\mathrm{st}}$
2502: fails condition~$(\ast)$, and $X_{\mathrm{st}}$ does have a Sarkisov link,
2503: following the scroll, to a conic bundle over $\p^2$.
2504:
2505: \addcontentsline{toc}{section}{References}
2506:
2507: \bibliography{bibbcz}
2508:
2509: \appendix
2510:
2511: \section{Birational transformations of scrolls}
2512: \label{sec:scrolls}
2513:
2514: \subsection{Definition of scrolls}
2515: \label{sec:def-scrolls}
2516:
2517: We set our notation for rational scrolls and toric links between them. Our
2518: treatment follows closely \cite[Chapter 2]{MR98d:14049}.
2519:
2520: Throughout the Appendix, we consider actions of the group
2521: $G=\c^\times \times \c^\times$ on affine space. The elements of
2522: $G$ are ordered pairs $(\lambda, \mu)$ where $\lambda, \mu \in
2523: \c^\times$. We denote by $\mathbb{X}=\Hom (G, \c^\times)$
2524: the lattice of characters of $G$, with basis the coordinate functions
2525: (projections on the two factors) $\chi_1$, $\chi_2$ such that
2526: $\chi_1(\lambda, \mu) = \lambda$ and $\chi_2 (\lambda, \mu) = \mu$.
2527: The dual lattice $\mathbb{X}^\ast = \Hom (\mathbb{X}, \z)$ is based by
2528: the 1-parameter subgroups $e_1$, $e_2$ such that $e_1(\lambda) =
2529: (\lambda, 1)$ and $e_2(\mu) = (1, \mu)$. Sometimes we abuse notation
2530: and write $\lambda$ for the element $e_1(\lambda) \in
2531: G$ (and, similarly, $\mu$ for $e_2(\mu)$). Occasionally we abuse even
2532: further and identify $\lambda$ with the coordinate function
2533: $\chi_1\colon G \to \c^\times$ (and, similarly, we identify
2534: $\mu$ with $\chi_2$).
2535:
2536: We now define rational scrolls.
2537: Fix a base $\p=\p^k$, with homogeneous coordinates $u_0, \dots,
2538: u_k$; in this paper, we only work with $k=1$ or $k=2$. Consider
2539: now $\c^{n+1}$, with
2540: coordinates $x_0,\dots,x_n$. Fix integers $a_0,\dots,a_n$, usually
2541: nonnegative and in increasing order. Consider the action of $G$ on the
2542: affine space $\a = \c^{k+1} \times \c^{n+1}$, where the two factors
2543: of $G$ act by
2544: \begin{align*}
2545: \lambda \colon (u_0,\dots,u_k,x_0,\dots,x_n) \mapsto & (\lambda u_0,\dots,
2546: \lambda u_k,\lambda^{-a_0} x_0,\dots,\lambda^{-a_n} x_n)\\
2547: \mu \colon (u_0,\dots,u_k,x_0,\dots,x_n) \mapsto & (u_0,\dots,u_k,\mu
2548: x_0,\dots,\mu x_n).
2549: \end{align*}
2550: We summarise this action by writing down the matrix:
2551: \[
2552: \begin{pmatrix}
2553: 1 &\dots& 1 &-a_0&\dots&-a_n\\
2554: 0 &\dots& 0 &1 &\dots&1
2555: \end{pmatrix}.
2556: \]
2557: By definition, the scroll $\F=\F(a_0,\dots,a_n)$ is the following quotient:
2558: \[
2559: \F(a_0,a_1,\dots,a_n) = \bigl(\c^{k+1}\setminus \{0\}\bigr) \times
2560: \bigl(\c^{n+1}\setminus \{0\}\bigr)/G.
2561: \]
2562: It is clear that $\F$ is a $\p^n$-bundle over $\p^k$.
2563:
2564: \subsection{Line bundles on scrolls}
2565: \label{sec:lbs-scrolls}
2566:
2567: There is a 1-to-1 correspondence between line bundles on the scroll
2568: $\F$ and characters $\chi\colon G \to \cstar$ of the group
2569: $G$. To a character $\chi$ we associate a
2570: \emph{$G$-linearisation} of the trivial line bundle over $\a$, by
2571: acting with $\chi$ in the direction of fibres. Taking the quotient of
2572: the $G$-linearisation by $G$, we form a bundle on $\F$.
2573: Let us denote by $L_\chi$ the resulting bundle. It is easy to
2574: chase through the definition and see that the
2575: sections of $L_\chi$ are the global
2576: eigenfunctions with eigenvalue $\chi$:
2577: \[
2578: H^0(\F, L_\chi)=\{f\colon (\c^{k+1} \setminus \{0\}) \times
2579: (\c^{n+1} \setminus \{0\}) \to \c \mid f(gx)=\chi(g) f(x) \}.
2580: \]
2581: Sometimes, we abuse notation and confuse the line bundle and the
2582: corresponding character.
2583:
2584: We denote by $L$ and $M$ the line bundles corresponding to $\chi_1$ and
2585: $\chi_2$. By what we just said, the
2586: sections of $L$ are the functions $f\colon (\c^{k+1}\setminus\{0\})
2587: \times (\c^{n+1}\setminus \{0\}) \to \c$ such that
2588: \begin{eqnarray*}
2589: f(\lambda u_0,\dots, \lambda u_k,\lambda^{-a_0} x_0,\dots,\lambda^{-a_n}
2590: x_n)&=& \lambda f(u_0,\dots,u_k,x_0,\dots,x_n)\\
2591: f(u_0,\dots,u_k,\mu x_0,\dots,\mu x_n)&=& f(u_0,\dots,u_k,x_0,\dots,x_n).
2592: \end{eqnarray*}
2593: It follows that $H^0(\F, L)$ is based by the coordinate
2594: functions $u_0,\dots,u_n$ and $L$ is the pull-back $\pi^\ast \o(1)$ by
2595: the natural morphism $\pi \colon \F\to \p^k$.
2596: Similarly, the
2597: group $H^0(\F, M)$ of global sections of $M$ is based by the
2598: monomials
2599: \[
2600: u_0^{w_0}\cdots u_k^{w_k} x_i \quad \text{where} \quad w_0+\cdots +w_k = a_k.
2601: \]
2602:
2603: \subsection{$\F$ as a geometric quotient}
2604:
2605: Using the language of $G$-linearisations, we can view $\F$ as a
2606: \emph{geometric} quotient $\a^{k+n+2} /\!\!/ G$ in the sense of
2607: Geometric Invariant Theory. If $G$ acts on $\a=\a^{k+n+2}$ as before and
2608: $\chi\colon G \to \cstar$ is a character, then $G$ acts on functions
2609: $f\in \o_\a$ by $gf(x)=\chi(g)f(g^{-1}x)$, and we denote $\o_\a^\chi$ the
2610: invariants. The set of semistable points is by definition
2611: \[
2612: \a^{\text{ss}}_\chi=\{x\in \a \mid \exists f\in \o_\a^\chi, \;f(x) \not = 0\}.
2613: \]
2614: The group $G$ acts with finite stabilisers (in fact, freely) on the
2615: set of semistable points, and the geometric quotient is by definition
2616: \[
2617: \a /\!\!/G = \a^{\text{ss}}_\chi/G.
2618: \]
2619:
2620: \subsection{Linearisations and geometric quotients}
2621: \label{sec:geom-quot}
2622:
2623: Different linearisations lead to different quotients. We state what is
2624: going on and leave the elementary proofs to the reader.
2625: We say that a linearisation is \emph{useful} if the set of semistable points is
2626: nonempty. To fix ideas, let us assume that $0=a_0\leq a_1 \leq \dots
2627: \leq a_{n}$.
2628: The cone of useful linearisations is the cone
2629: \[
2630: (\r_+[L]+\r_+[M-a_{n-1} L])\cap \mathbb{X}
2631: \]
2632: This cone is partitioned into chambers corresponding to different
2633: geometric quotients. For example if $\chi \in \r_+[L] + \r_+[M]$, then
2634: $\a^{\text{ss}}_\chi=(\c^{k+1}\setminus \{0\}) \times (\c^{n+1} \setminus
2635: \{0\})$ and we get our $\F$ back. However there are the other chambers
2636: \[
2637: \sigma_i = (\r_+[M-a_{i-1}L] + \r_+[M-a_{i}L]) \cap \mathbb{X},
2638: \]
2639: whenever $a_{i-1}<a_{i}$, and if $\chi \in \sigma_i$
2640: \[
2641: \a^{\text{ss}}_\chi = (\c^{k+1+i}\setminus \{0\})\times (\c^{n+1-i}
2642: \setminus \{0\}).
2643: \]
2644: The corresponding quotient
2645: \[
2646: \F_i= \a^{\text{ss}}_\chi /G
2647: \]
2648: for $\chi \in \sigma_i$ is birational to $\F$. The sequence of
2649: birational maps $\F=\F_0 \dasharrow \F_1 \dasharrow \cdots$ is a 2-ray game
2650: in the sense of \cite[Section 2.2]{corti:00}. For example if $a_1>0$, the move
2651: $\F\dasharrow \F_1$ is the antiflip of the section $\Gamma=
2652: \{x_1=x_2=\dots =x_n=0\}$ which generates one of the two extremal
2653: rays of $\NE (\F)$.
2654:
2655: \subsection{Generalisations}
2656: \label{sec:generalisations}
2657:
2658: The above can be generalised slightly to the action of $G$ on
2659: $\a^{n+1}$ given by the matrix
2660: \[
2661: \begin{pmatrix}
2662: \alpha_0 & \dots & \alpha_n \\
2663: \beta_0 & \dots & \beta_n
2664: \end{pmatrix}.
2665: \]
2666: We use this notation in the text as a shorthand for the action
2667: \begin{align*}
2668: x_0, \dots, x_n \mapsto \lambda^{\alpha_0}x_0, \dots,
2669: \lambda^{\alpha_n} x_n \\
2670: x_0, \dots, x_n \mapsto \mu^{\beta_0}x_0, \dots,
2671: \mu^{\beta_n} x_n.
2672: \end{align*}
2673: Here we assume:
2674: \begin{enumerate}
2675: \item the rows are linearly independent (so we get a faithful action
2676: of $G$) and all columns are nonzero,
2677: \item all $\beta_i\geq 0$, and $\beta_n >0$,
2678: \item the ratios $\alpha_i/\beta_i$ are in decreasing order.
2679: \end{enumerate}
2680: As before we denote $L$, $M$ the $G$-linearisations corresponding to
2681: the characters $\chi_1$, $\chi_2$. The cone of useful linearisations
2682: \[
2683: (\r_+[\alpha_0M-\beta_0L]+\r_+[\alpha_{n-1}M-\beta_{n-1}L]) \cap \mathbb{X}
2684: \]
2685: is partitioned into chambers
2686: \[
2687: \sigma_i =
2688: (\r_+[\alpha_{i-1}M-\beta_{i-1}L]+\r_+[\alpha_{i}M-\beta_{i}L]).
2689: \]
2690: (when $\alpha_{i-1}/\beta_{i-1}> \alpha_i/\beta_i $).
2691: Choosing a character $\chi \in \sigma_i$ gives a semistable locus
2692: $\a^{\text{ss}}_\chi= (\c^{i+1}\setminus \{0\}) \times (\c^{n-i}\setminus
2693: \{0\})$ and a geometric quotient $\F_i=\a^{\text{ss}}_\chi/G$.
2694:
2695:
2696: \subsection{Alternative viewpoints}
2697: \label{sec:altern-viewp}
2698:
2699: There are at least two other points of view on birational maps of
2700: scrolls.
2701:
2702: We can identify the cone of useful $G$-linearisations with the mobile
2703: cone $\NM^1(\F)$ of the scroll. From this point of view, the main focus
2704: is the bigraded ring
2705: \[
2706: \bigoplus_{e,n} H^0(\F, eM+nL).
2707: \]
2708:
2709: We can also view scrolls, and their generalisations, as special cases
2710: of rank~2 toric varieties.
2711:
2712:
2713:
2714:
2715: \end{document}
2716:
2717: In this article, we give some examples and state some conjectures on the
2718: birational structure of Mori fibre spaces. We focus mainly on the
2719: case of \emph{strict} Mori fibre spaces, i.~e.\ those that are not
2720: Fano 3-folds. Fano 3-folds then appear as possible alternative
2721: birational models.
2722:
2723: