math0308182/e69.tex
1: \NeedsTeXFormat{LaTeX2e}
2: \newif\ifsmfart
3: \IfFileExists{smfart.cls}
4:    {\documentclass[12pt,english]{smfart}
5:     \IfFileExists{smfenum.sty}{\usepackage{smfenum}}{}
6:     \usepackage{bull}
7:     \smfarttrue}
8: {\message{^^J*** It would be better to typeset this file with smfart.cls ***^^J^^J}
9:     
10: \documentclass[12pt]{amsart}\let\Subsection\subsection}
11: 
12: \setcounter{tocdepth}{1}
13: \usepackage{amssymb}
14: \usepackage{epsfig}
15: \usepackage{graphicx}
16: \numberwithin{equation}{section}
17: 
18: \input xy
19: \xyoption{all}  
20: 
21: 
22: \textheight 20cm \textwidth 13cm
23: \advance\headheight 2pt
24: \calclayout
25: \allowdisplaybreaks[3]
26: 
27: 
28: 
29: \theoremstyle{plain}
30: \newtheorem{prop}[subsection]{Proposition}
31: \newtheorem{Theo}[section]{Theorem}
32: \newtheorem{theo}[subsection]{Theorem}
33: \newtheorem{coro}[subsection]{Corollary}
34: \newtheorem{cond}[subsection]{Condition}
35: \newtheorem{lem}[subsection]{Lemma}
36: \newtheorem{assu}[subsection]{Assumption}
37: \newtheorem{defn}[subsection]{Definition}
38: \newtheorem{sublemm}[subsection]{Sublemma}
39: 
40: \theoremstyle{definition}
41: \newtheorem{prob}[subsection]{Problem}
42: \newtheorem{ques}[subsection]{Question}
43: \newtheorem{conj}[subsection]{Conjecture}
44: 
45: \theoremstyle{remark}
46: \newtheorem{rem}[subsection]{Remark}
47: \newtheorem{rems}[subsection]{Remarks}
48: \newtheorem{exam}[subsection]{Example}
49: \newtheorem{exer}[subsection]{Exercise}
50: \newtheorem{nota}[subsection]{Notations}
51: 
52: %\usepackage{palatino}\usepackage[mathcal]{euler}
53: %\let\mathsec\mathrm
54: 
55: \newcommand{\bA}{{\mathbb A}}
56: \newcommand{\bC}{{\mathbb C}}
57: \newcommand{\bG}{{\mathbb G}}
58: \newcommand{\bN}{{\mathbb N}}
59: \newcommand{\bQ}{{\mathbb Q}}
60: \newcommand{\bR}{{\mathbb R}}
61: \newcommand{\bP}{{\mathbb P}}
62: \newcommand{\bZ}{{\mathbb Z}}
63: \newcommand{\cF}{{\mathcal F}}
64: \newcommand{\cO}{{\mathcal O}}
65: \def\cL{{\mathcal L}}
66: \def\cX{{\mathfrak X}}
67: \newcommand{\ra}{{\rightarrow}}
68: \newcommand{\hra}{{\hookrightarrow}}
69: \newcommand{\Spec}{{\rm Spec}}
70: \newcommand{\Move}{{\rm Mov}}
71: \newcommand{\Proj}{{\rm Proj}}
72: 
73: \newcommand{\tS}{\tilde S}
74: \newcommand{\NE}{\mathrm{NE}}
75: \newcommand{\NM}{\mathrm{NM}}
76: \newcommand{\Pic}{\mathrm{Pic}}
77: \newcommand{\Hom}{\mathrm{Hom}}
78: \newcommand{\Cone}{\mathrm{Cone}}
79: \newcommand{\Cox}{\mathrm{Cox}}
80: \newcommand{\tor}{\left<\mathrm{torsion}\right>}
81: \newcommand{\Dfour}{{\mathbf D}_4}
82: \newcommand{\Esix}{{\mathbf E}_6}
83: \def\la{{\lambda}}
84: \def\rA{{\mathrm A}}
85: \def\rN{{\mathrm N}}
86: \def\Tr{{\rm Tr}}
87: \def\Gal{{\rm Gal}}
88: 
89: 
90: %\usepackage[notref,notcite]{showkeys}
91: 
92: \makeatother
93: \makeatletter
94: 
95: 
96: 
97: \author{Brendan Hassett}
98: \address{Department of Mathematics \\
99: Rice University \\
100: Houston, TX 77005}
101: \email{hassett@rice.edu}
102: \author{Yuri Tschinkel}
103: \address{Department of Mathematics \\
104:          Princeton University\\ 
105:          Fine Hall, Washington Road\\
106:          Princeton, NJ 08544-1000,  U.S.A.}
107: \email{ytschink@math.princeton.edu}
108: \thanks{The first author was 
109: partially supported by the Sloan Foundation and by
110: NSF Grants 0196187 and 0134259. 
111: The second author was partially supported by NSF Grant 0100277.}
112: 
113: 
114: 
115: \def\Title    { }
116: \def\Author   {Brendan Hassett and Yuri Tschinkel}
117: \def\Subject  {Algebraic geometry}
118: \def\Keywords {Universal torsors, Cox rings}
119: 
120: 
121: 
122: 
123: 
124: \usepackage{graphicx}
125: 
126: 
127: 
128: 
129: 
130: 
131: \title[Universal torsors and Cox rings]{Universal torsors and Cox rings}
132: \date{\today}
133: \begin{document}
134: 
135: 
136: \begin{abstract}
137: We study the equations of universal torsors on rational surfaces. 
138: \end{abstract}
139: 
140: \maketitle
141: \tableofcontents
142: 
143: 
144: \setcounter{section}{0}
145: \section*{Introduction}
146: \label{sect:introduction}
147: 
148: The study of surfaces over nonclosed fields $k$ leads naturally
149: to certain auxiliary varieties, called {\em universal torsors}.
150: The proofs of the Hasse principle and weak approximation
151: for certain Del Pezzo surfaces
152: required a very detailed knowledge of the projective geometry, in fact,
153: explicit equations, for these torsors \cite{CTS}, \cite{CSS}, \cite{CSS2},
154: \cite{SS}, \cite{S}, \cite{Sb}.
155: More recently, Salberger proposed using universal torsors to count
156: rational points of bounded
157: height, obtaining the first sharp upper bounds on split Del Pezzo surfaces
158: of degree 5 and asymptotics on split toric varieties over $\mathbb Q$
159: \cite{S-ast}. This approach was further developed in the work of
160: Peyre, de la Bret\`eche, and Heath-Brown 
161: \cite{Peyre}, \cite{Peyre2}, \cite{Br}, \cite{HB-C}.
162: 
163: \
164: 
165: Colliot-Th\'el\`ene and Sansuc have given a general formalism for
166: writing down equations for these torsors.
167: We briefly sketch their method:
168: Let $X$ be a smooth projective variety and $\{ D_j\}_{j\in J} $ a finite set
169: of irreducible divisors on $X$ such that
170: $U:=X\setminus  \cup_{j\in J} D_j$ has trivial Picard group.
171: In practice, one usually chooses generators
172: of the effective cone of $X$, e.g., the lines on the Del Pezzo surface.
173: Consider the resulting exact sequence:
174: $$
175: 0\longrightarrow \bar{k}[U]^*/\bar{k}^*\longrightarrow \oplus_{j\in J}
176: \bZ D_j \longrightarrow \Pic(X_{\bar{k}})\longrightarrow 0.
177: $$
178: Applying $\Hom(-,\bG_m)$, one obtains
179: an exact sequence of tori
180: $$
181: 1\longrightarrow T(X)\longrightarrow T\longrightarrow R\longrightarrow 1,
182: $$
183: where the first term is the {\em N\'eron-Severi torus} of $X$.
184: Suppose we have a collection
185: of rational functions, invertible on $U$, which form a basis for
186: the relations among the $\{D_j\}_{j\in J}$.
187: These can be interpreted as a section
188: $U\ra R\times U$, and thus naturally
189: induce a $T(X)$-torsor over $U$, which  canonically extends
190: to the universal torsor over $X$.  In practice,
191: this extension can be made explicit,
192: yielding equations for the universal torsor.
193: 
194: 
195: However, when the cone generated by $\{D_j\}_{j\in J}$ is simplicial, there
196: are {\em no} relations and this method
197: gives little information.  In this paper, we outline an alternative
198: approach to the construction of universal torsors
199: and illustrate it in specific examples
200: where the effective cone of $X$ is simplicial.
201: 
202: 
203: \
204: 
205: We will work with varieties $X$ such that 
206: the Picard and the N\'eron-Severi groups
207: of $X$ coincide and such that the ring
208: $$
209: \Cox(X):=\bigoplus_{L\in \Pic(X)} \Gamma(X,L),
210: $$
211: is finitely generated.  This ring admits a natural action of 
212: the N\'eron-Severi torus and the corresponding affine variety
213: is a natural embedding of the universal torsor of $X$.
214: The challenge is to 
215: actually compute $\Cox(X)$ in specific examples;
216: Cox has shown that it is
217: a polynomial ring precisely when $X$ is toric \cite{Cox}.
218:   
219: \
220:  
221: Here is a roadmap of the paper: In Section~\ref{sect:general} 
222: we introduce Cox rings and discuss their general properties.
223: Finding generators for the Cox ring entails embedding
224: the universal torsor into affine space, which yields embeddings
225: of our original variety into toric quotients of this affine space.
226: We have collected several useful facts about toric varieties in 
227: Section~\ref{sect:toric}.  Section~\ref{sect:e6} is devoted to 
228: a detailed analysis of the unique cubic surface $S$ with an isolated 
229: singularity of type $\Esix$.  We compute the (simplicial) effective
230: cone of its minimal desingularization $\tilde{S}$, 
231: and produce 10 distinguished sections in $\Cox(\tilde{S})$.
232: These satisfy a unique equation and we show the 
233: universal torsor naturally embeds in 
234: the corresponding hypersurface in ${\mathbb A}^{10}$. 
235: More precisely, we get a homomorphism from the coordinate ring
236: of $\bA^{10}$ to $\Cox(\tilde{S})$ and the main point 
237: is to prove its surjectivity.   Here we use
238: an embedding of $\tilde{S}$ into a simplicial toric threefold $Y$, 
239: a quotient of ${\mathbb A}^{10}$ 
240: under the action of the N\'eron-Severi torus
241: so that $\Cox(Y)$ is the polynomial ring over the above 10 generators. 
242: The induced restriction map on the level of Picard groups
243: is an isomorphism respecting the moving cones.  We conclude
244: surjectivity for each degree by finding an appropriate 
245: birational projective model of $Y$ and using 
246: vanishing results on it. 
247: Finally, in Section~\ref{sect:d4} we write down equations 
248: for the universal torsors (the Cox rings) of
249: a split and a nonsplit cubic surface 
250: with an isolated singularity of 
251: type $\Dfour$. 
252: 
253: For an account of the Cox rings of smooth Del Pezzo surfaces,
254: we refer the reader to the paper of Batyrev and Popov 
255: \cite{BP}.
256: 
257: 
258: 
259: 
260: \
261: 
262: \noindent {\bf Acknowledgments:}
263: The results of this paper have been reported at the
264: American Institute of Mathematics conference ``Rational and integral
265: points on higher dimensional varieties''.  We benefited from the
266: comments of the other participants, in particular, V. Batyrev and
267: J.L. Colliot-Th\'el\`ene.  We also thank S. Keel for several helpful
268: discussions about Cox rings and M. Thaddeus for advice about the
269: geometric invariant theory of toric varieties.
270: 
271: 
272: \
273: 
274: \section{Generalities on the Cox ring}
275: \label{sect:general}
276: 
277: For any finite subset $\Xi$ of a real vector
278: space, let $\Cone(\Xi)$ denote the closed
279: cone generated by $\Xi$.  
280: 
281: Let $X$ be a normal projective variety of dimension $n$ over an
282: algebraically closed field $k$ of characteristic zero.  Let 
283: $\rA_{n-1}(X)$ and $\rN_{n-1}(X)$ denote Weil divisors on $X$ up to 
284: linear and numerical equivalence, respectively.  
285: Let $\rA_1(X)$ and $\rN_1(X)$ denote the classes of curves
286: up to equivalence.  
287: Let $\NE_{n-1}(X) \subset \rN_{n-1}(X)_{\bR}$ denote the
288: cone of (pseudo)effective divisors, i.e., the 
289: smallest real closed cone containing all the
290: effective divisors of $X$.  
291: Let $\NE_1(X)\subset \rN_1(X)_{\bR}$ denote
292: the cone of effective curves and   
293: $\NM^1(X) \subset \rN_{n-1}(X)_{\bR}$ 
294: the cone of nef Cartier divisors, which is
295: dual to the cone of effective Cartier divisors.  
296: By Kleiman's criterion,
297: this is the smallest real closed cone containing all
298: ample divisors of $X$.  
299: 
300: \
301: 
302: 
303: Let $L_1,\ldots,L_r$ be invertible sheaves on $X$.
304: For $\nu=(n_1,\ldots,n_r)\in \bN^r$ write
305: $$L^{\nu}:=L_1^{\otimes n_1} \otimes \ldots \otimes L_r^{\otimes n_r}.$$
306: Consider the ring
307: $$R(X,L_1,\ldots,L_r):=\bigoplus_{\nu\in \bN^r}\Gamma(X,L^{\nu}),$$
308: which need not be finitely generated in general.
309: 
310: By definition, an invertible sheaf $L$ on $X$ is {\em semiample}
311: if $L^N$ is globally generated for some $N>0$:
312: 
313: \begin{prop}{(\cite{HK}, Lemma 2.8)}
314: \label{prop:coxfg} 
315: If $L_1,\ldots,L_r$ are semiample then
316: $R(X,L_1,\ldots,L_r)$ is finitely generated.
317: \end{prop}
318: 
319: \begin{rem}
320: If the $L_i$ are ample then, after replacing each
321: $L_i$ by a large multiple,  $R(X,L_1,\ldots,L_r)$ is
322: generated by
323: $$
324: \Gamma(X,L_1) \otimes \ldots \otimes \Gamma(X,L_r).
325: $$ 
326: However, this is not generally the case if the $L_i$
327: are only semiample (despite the assertion in
328: the second part of Lemma 2.8 of \cite{HK}).  
329: Indeed, let $X\ra \bP^1\times \bP^1$ 
330: be a double cover and $L_1$ and $L_2$ be the pull-backs
331: of the polarizations on the $\bP^1$'s to $X$.  For suitably
332: large $n_1$ and $n_2$, $L_1^{n_1}\otimes L_2^{n_2}$ is very
333: ample and its sections embed $X$.  However, 
334: $$
335: \Gamma(X,L^{n_1}_1) \otimes \Gamma(X,L^{n_2}_2) \simeq
336: \Gamma(\bP^1,\cO_{\bP^1}(n_1)) \otimes \Gamma(\bP^1,\cO_{\bP^1}(n_2)),
337: $$
338: and any morphism induced by these sections factors through 
339: $\bP^1\times \bP^1$.   
340: \end{rem}  
341: 
342: 
343: \begin{prop}
344: \label{prop:hu} 
345: Let $L_1,\ldots, L_r$ be a set of invertible sheaves on $X$
346: such that $L_j$ is generated by sections $s_{j,0},...,s_{j,d_j}$. 
347: Assume that the induced morphism $X\ra \prod_{j} \bP^{d_j}$ 
348: is birational into its image. 
349: Then the ring generated by the $s_{j,k}$'s 
350: has the same fraction field as $R(X,L_1,\ldots, L_r)$. 
351: \end{prop}
352: 
353: \begin{proof}
354: Both rings have fraction field $k(X)(t_1,...,t_r)$,
355: where $t_j$ is a nonzero section of $L_j$.  
356: \end{proof}
357: 
358: 
359: \
360: 
361: \begin{defn}
362: \cite{HK} 
363: Let $X$ be a nonsingular projective
364: variety so that $\Pic(X)$ is a free abelian group of rank $r$.  
365: The {\em Cox ring} for $X$ is defined
366: $$
367: \Cox(X):=R(X,L_1,\ldots,L_r),
368: $$
369: where $L_1,\ldots,L_r$ are lines bundles so that
370: \begin{enumerate}
371: \item{the $L_i$ form a $\bZ$-basis of $\Pic(X)$;}
372: \item{the cone $\Cone(\{L_1,\ldots,L_r\})$ contains $\NE_{n-1}(X)$.}
373: \end{enumerate}
374: This ring is naturally graded by $\Pic(X)$:
375: for $\nu\in \Pic(X)$ the $\nu$-graded piece
376: is denoted $\Cox(X)_{\nu}$.  
377: \end{defn}
378: 
379: \begin{prop}
380: \label{prop:unique} \cite{HK}
381: The ring $\Cox(X)$ does not depend on the choice of generators
382: for $\Pic(X)$.  
383: \end{prop}
384: 
385: \begin{proof} 
386: Consider two sets of generators
387: $L_1,\ldots,L_r$ and $M_1,\ldots,M_r$.  
388: Since $\Cone(\{L_i\})$ and $\Cone(\{M_i\})$ contain
389: all the effective divisors, the nonzero graded pieces
390: of both $R(X,L_1,\ldots,L_r)$ and
391: $R(X,M_1,\ldots,M_r)$ are indexed by the effective divisor
392: classes in $\Pic(X)$.  Choose isomorphisms
393: $$M_j\simeq L^{(a_{1j},\ldots,a_{rj})}, 
394: \quad i=1,\ldots,r, A=(a_{ij}) $$
395: which naturally induce isomorphisms
396: $$
397: \Gamma(M^{\nu})\simeq \Gamma(L^{A\nu}), 
398: \quad A\nu=(a_{11}\nu_1+\ldots+a_{1r}\nu_r,
399: \ldots,a_{r1}\nu_1+\ldots+a_{rr}\nu_r).
400: $$
401: Thus we find $R(X,L_1,\ldots,L_r)\simeq R(X,M_1,\ldots,M_r)$.
402: \end{proof}
403: 
404: As $\Cox(X)$ is graded by $\Pic(X)$, a free abelian group 
405: of rank $r$, the torus
406: $$T(X):=\Hom(\Pic(X),\bG_m)$$
407: acts on $\Cox(X)$.  Indeed, each $\nu\in \Pic(X)$ naturally yields
408: a character $\chi_{\nu}$ of $T(X)$, and the action is given by
409: $$
410: t\cdot \xi=\chi_{\nu}(t)\xi, \quad 
411: \xi \in \Cox(X)_{\nu}, t\in T(X).
412: $$
413: Thus the isomorphism constructed in Proposition~\ref{prop:unique}
414: is {\em not} canonical:  Two such isomorphisms differ by
415: the action of an element of $T(X)$.  It is precisely
416: this ambiguity that makes descending the
417: universal torsor to nonclosed fields an interesting question.
418: 
419: 
420: \
421: 
422: The following conjecture is a special case of
423: 2.14 of \cite{HK}:
424: 
425: \begin{conj}[Finiteness of Cox ring]
426: Let $X$ be a log Fano variety. 
427: Then $\Cox(X)$ is finitely generated. 
428: \end{conj}
429: 
430: 
431: \begin{rem}
432: \label{rem:cox-fg}
433: Note that if $\Cox(X)$ is finitely generated 
434: it follows trivially that $\NE_{n-1}(X)$ is finitely generated. 
435: Moreover, the nef cone $\NM^1(X)$ is also finitely generated. 
436: 
437: Indeed, the nef cone corresponds to one of the chambers in the 
438: group of characters of $T(X)$ governed by the stability
439: conditions for points  $v\in \Spec(\Cox(X))$.  These
440: chambers are bounded by finitely many hyperplanes (see Theorem 0.2.3 in 
441: \cite{DH} for more details).  
442: \end{rem}
443: 
444: 
445: 
446: It has been conjectured by Batyrev \cite{Ba}
447: that the pseudo-effective cone of a Fano variety
448: is finitely generated. 
449: However, the finiteness
450: of the Cox ring is not a formal consequence of the
451: finiteness of the pseudo-effective cone.  
452: 
453: 
454: \begin{exam}
455: \label{exam:9}
456: Let $p_1,\ldots,p_9\in H\subset \bP^3$ be nine
457: distinct coplanar points given as a
458: complete intersection of two generic
459: cubic curves in the hyperplane $H$, and let $X$ be the blow-up of
460: $\bP^3$ at these points.  Then $\NE^1(X)$ is finitely generated
461: but $\Cox(X)$ is not.  Indeed, $X$ is an equivariant
462: compactification of the additive group $\bG_a^3$,
463: acting by translation on the
464: affine space $\bP^3-H$.  The group action can
465: be used to show that $\NE^1(X)$ is generated
466: by the boundary components (see \cite{HT1}). Similarly,
467: one can show that the cone
468: $\NE_1(X)$ is generated by classes of curves in the boundary
469: components, e.g., the proper transform $\tilde{H}\subset X$
470: of $H$.  It is well-known that $\NE_1(\tilde{H})$ is infinite
471: \cite{KM} \S 1.23(4):  The pencil of cubic plane curves
472: with base locus $p_1,\ldots, p_9$ induces an elliptic fibration,
473: $$
474: \tilde{H} \ra \bP^1,
475: $$
476: for which the nine exceptional curves of $\tilde{H}\ra H$ are sections.
477: Addition in the group law gives an infinite number of
478: sections, which are also $(-1)$-curves and generators of
479: $\NE_1(\tilde{H})$.  These are also generators of
480: $\NE_1(X)$, since the sections (other than the nine
481: exceptional curves) intersect $\tilde{H}$ negatively.
482: It follows that $\NE_1(X)$ and $\NM^1(X)$ are not finitely generated
483: and hence $\Cox(X)$ is not finitely generated
484: (see Remark~\ref{rem:cox-fg}).
485: \end{exam}
486: 
487: 
488: 
489: \
490: 
491: 
492: 
493: 
494: \begin{prop}
495: \label{prop:semiample}
496: Let $X$ be a nonsingular projective variety
497: whose anticanonical divisor $-K_X$ is
498: nef and big.  Suppose that $D$ is a nef divisor on $X$.
499: Then $H^i(X,\cO_X(D))=0$ for each $i>0$ and $D$
500: is semiample.
501: \end{prop} 
502: 
503: \begin{proof}
504: The first assertion is a consequence of
505: Kawamata-Viehweg vanishing \cite{KM} \S 2.5.
506: The second is a special case of the Kawamata
507: Basepoint-freeness Theorem \cite{KM} \S 3.2.
508: \end{proof}
509: 
510: Proposition~\ref{prop:semiample}
511: largely determines the Hilbert function of the Cox ring:
512: 
513: \begin{coro}
514: \label{coro:hilbert}
515: Retain the assumptions of Proposition~\ref{prop:semiample}.
516: Then for nef classes $\nu$ we have
517: $$
518: \dim \Cox(X)_{\nu}=\chi(\cO_X(\nu)).
519: $$
520: \end{coro}
521: 
522: \begin{rem}
523: In practice, this will help us to find generators of $\Cox(X)$. 
524: \end{rem}
525: 
526: 
527: 
528: \section{Generalities on toric varieties}
529: \label{sect:toric}
530: 
531: 
532: 
533: We recall quotient constructions
534: of toric varieties, following Brion-Procesi \cite{brionproc},
535: Cox \cite{Cox}, and Thaddeus \cite{thad}.
536: 
537: 
538: Let $T\simeq \bG_m^r$ be a torus with character
539: group $\cX^*(T)$.  Suppose that $T$ acts
540: faithfully on the polynomial 
541: ring $k[x_1,\ldots,x_{n+r}]$ by the formula
542: $$t(x_j)=\chi_j(t)x_j, \quad t\in T, $$
543: where $\{\chi_1,\ldots,\chi_{n+r} \}\subset \cX^*(T).$ 
544: Define $M$ as the kernel of the surjective morphism
545: $$\chi:=(\chi_1,\ldots,\chi_{n+r}): \bZ^{n+r} \ra
546: \cX^*(T).$$
547: We interpret $M$ as the character group of the quotient torus
548: $\bG_m^{n+r}/T$. 
549: Set $N=\Hom(M,\bZ)$ so that dualizing gives
550: $$(\bZ^{n+r})^* \ra N \ra 0.$$
551: Let $e_1,\ldots,e_{n+r}$ and $e_1^*,\ldots,e_{n+r}^*$ denote
552: the coordinate vectors in $\bZ^{n+r}$ and  $(\bZ^{n+r})^*$;
553: let $\bar{e}_1^*,\ldots,\bar{e}_{n+r}^*\in N$ denote 
554: the images of the $e_i^*$ in $N$.  Concretely, the $\bar{e}_i^*$
555: are the columns of the $n\times (n+r)$ matrix of
556: dependence relations among the $\chi_j$.  
557: 
558: Consider a toric $n$-fold $X$ associated with a fan 
559: having one-skeleton 
560: $\{\bar{e}_1^*,\ldots,\bar{e}_{n+r}^*\}$.  
561: In particular, we assume that none of $\bar{e}_i^*$ is
562: zero or a positive multiple of any of the others.  
563: The variety $X$ is a categorical quotient of an invariant
564: open subset $U\subset \bA^{n+r}$ under the action of
565: $T$ described above (see \cite{Cox} 2.1).  
566: Elements $\nu\in \cX^*(T)$ classify
567: $T$-linearied invertible sheaves $\cL_{\nu}$ on $\bA^{n+r}$ and 
568: $$\Gamma(\bA^{n+r},\cL_{\nu})\simeq k[x_1,\ldots,x_{n+r}]_{\nu}.$$
569: We have $\rA_{n-1}(X)\simeq \cX^*(T)$ and we can identify
570: $$
571: \Gamma(\cO_X(D))\simeq k[x_1,\ldots,x_n]_{\nu(D)},
572: $$
573: where $\nu(D) \in \cX^*(T)$ is associated with
574: the divisor class of $D$.  The variables $x_i$ are
575: associated with the irreducible torus-invariant 
576: divisors $D_i$ on $X$ (see \cite{fulton} \S 3.4), and
577: the cone of effective divisors $\NE_{n-1}(X)$ is generated by 
578: $\{D_1,\ldots,D_{n+r}\}$.
579: Geometrically, the effective cone in $\cX^*(T)$ is the
580: image of the standard simplicial cone generated
581: by $e_1,\ldots,e_{n+r}$ under the projection homomorphism
582: $\chi:\bZ^{n+r} \ra \cX(T)$.  
583: 
584: 
585: Recall that the {\em moving cone}
586: $$
587: \Move(X) \subset \NE_{n-1}(X)
588: $$
589: is defined as the smallest
590: closed subcone containing the effective divisors on $X$
591: without fixed components.  
592: \begin{prop}\label{prop:toricmove}
593: Retaining the notation and assumptions above,
594: $$
595: \Move(X) = \bigcap_{i=1,\ldots,n+r} 
596: \Cone(\chi_1,\ldots,\chi_{i-1},\chi_{i+1},\ldots,\chi_{n+r})
597: $$
598: and has nonempty interior.
599: \end{prop}
600: \begin{proof}
601: The fixed components of
602: $\Gamma(X,\cO_X(D))$ are necessarily invariant under the
603: torus action, hence are taken from 
604: $\{D_1,\ldots,D_{n+r}\}$.  Moreover, $D_i$ is fixed
605: in each $\Gamma(X,\cO_X(dD)),d>0$ if and only if 
606: $x_i$ divides each element of $k[x_1,\ldots,x_{n+r}]_{d\nu(D)}$.  
607: This is the case exactly when
608: $$\nu(D) \in \Cone(\chi_1,\ldots,\chi_{n+r}) - 
609: \Cone(\chi_1,\ldots,\chi_{i-1},\chi_{i+1},\ldots,\chi_{n+r}).$$
610: 
611: Suppose that the interior of the moving cone is empty.
612: After permuting indices there are two possibilities:  Either 
613: $\Cone(\chi_2,\ldots,\chi_{n+r})$ has
614: no interior, or the cones 
615: $\Cone(\chi_2,\ldots,\chi_{n+r})$ and
616: $\Cone(\chi_1,\chi_3,\ldots,\chi_{n+r})$ have nonempty interiors
617: but meet in a cone with positive codimension.  As the $T$-action is faithful,
618: the $\chi_i$ span $\cX^*(T)$.  In the first case,
619: $\chi_2,\ldots,\chi_{n+r}$ span a codimension-one
620: subspace of $\cX^*(T)$ that does not contain $\chi_1$, so 
621: that each dependence relation 
622: $$
623: c_1\chi_1 + \ldots + c_{n+r}\chi_{n+r}=0
624: $$
625: has $c_1=0$.  This translates into $\bar{e}_1^*=0$,
626: a contradiction.  In the second case, $\chi_3,\ldots,\chi_{n+r}$
627: span a hyperplane, and $\chi_1$ and $\chi_2$ are on opposite
628: sides of this hyperplane.  Putting the dependence relations 
629: among the $\chi_i$ in row echelon form, we obtain a unique
630: relation with nonzero first and second entries, and
631: these two entries are both positive.  This translates into
632: the proportionality of $\bar{e}_1^*$ and $\bar{e}_2^*$.
633: \end{proof}
634: 
635: We now seek to characterize the projective toric
636: $n$-folds $X$ with one-skeleton 
637: $\{\bar{e}_1^*,\ldots,\bar{e}_{n+r}^*\}$.
638: These are realized as Geometric Invariant Theory
639: quotients $\bA^{n+r}//T$ associated with the various 
640: linearizations of our $T$-action.  
641: We consider the graded ring
642: $$
643: R:=\sum_{d\ge 0}\Gamma(\bA^{n+r},\cL_{d\nu})=
644: \sum_{d\ge 0 }k[x_1,\ldots,x_{n+r}]_{d\nu}.
645: $$
646: 
647: \begin{prop}[see \cite{thad} \S 2,3]
648: \label{prop:whenprojective}
649: Retain the notation above
650: and set $X:=\Proj(R)$.
651: 
652: \begin{enumerate}
653: \item{
654: $X$ is projective over $k$ if and only if
655: $0$ is not contained in the convex hull of
656: $\{\chi_1,\ldots,\chi_{n+r} \}$.}
657: \item{
658: $X$ is toric of dimension $n$ if
659: $\nu$ is in the interior of 
660: $$
661: \Cone(\chi_1,\ldots,\chi_{n+r}).
662: $$}
663: \item{In this case, 
664: the one-skeleton of $X$ is contained in
665: $\{\bar{e}_1^*,\ldots,\bar{e}_{n+r}^*\}$.  
666: Equality holds if $\nu$ is in the interior of the moving cone 
667: $$\bigcap_{i=1,\ldots,n+r}
668: \Cone(\chi_1,\ldots,\chi_{i-1},\chi_{i+1},\ldots,\chi_{n+r}).$$}
669: \end{enumerate}
670: \end{prop}
671: \begin{rem}
672: Our proof will show that
673: $X$ may still be of dimension $n$ even when $\nu$
674: is contained in a facet of 
675: $$
676: \Cone(\chi_1,\ldots,\chi_{n+r}).
677: $$
678: 
679: Similary, the one-skeleton of $X$ may still be 
680: $\{\bar{e}_1^*,\ldots,\bar{e}_{n+r}^*\}$
681: even when $\nu$ is contained in a facet of
682: $$\bigcap_{i=1,\ldots,n+r}
683: \Cone(\chi_1,\ldots,\chi_{i-1},\chi_{i+1},\ldots,\chi_{n+r}).$$
684: \end{rem}
685: \begin{proof}
686: The monomials which appear in $R$ are in one-to-one correspondence
687: to solutions of
688: $$a_1\chi_1+\ldots+a_{n+r}\chi_{n+r}=d\nu, \quad
689: a_i \in \bZ_{\ge 0}.$$
690: In geometric terms, the monomials appearing
691: in $R$ coincide with the elements of $\bZ^{n+r}$
692: in the cone
693: $$\chi^{-1}(\Cone(\nu))\cap \Cone(e_1,\ldots,e_{n+r}).$$
694: By Gordan's Lemma in convex geometry, $R$ is generated
695: as a $k$-algebra by a finite set of monomials 
696: $x^{m_1},\ldots,x^{m_s}$.
697: The monomials appearing in the $d$th graded piece $R_d$ coincide
698: with elements of $\bZ^{n+r}$ in the polytope
699: $$P_{d\nu}:=\chi_{\bR}^{-1}(d\nu) \cap \Cone(e_1,\ldots,e_{n+r}).$$
700: Note that $\chi^{-1}(d\nu)$ is a translate of $M$.
701: 
702: 
703: For the first part, recall that 
704: $\Proj(R)$ is projective over $\Spec(R_0)$,
705: where $R_0$ is the degree-zero part.  
706: Now $0$ is in the convex hull of $\{\chi_1,\ldots,\chi_{n+r}\}$
707: if and only if there are nonconstant elements of $R$ of degree
708: zero.  Our hypothesis just says that $R_0=k$ and thus is equivalent
709: to the projectivity of $X$ over $k$.  
710: 
711: As for the second part,  $T$ acts 
712: on $R$ by homotheties and thus acts trivially on $\Proj(R)$,
713: so we have an induced action of $\bG_m^{n+r}/T$ on $\Proj(R)$.
714: We claim this action is faithful, so 
715: the quotient is toric of dimension $n$.  
716: Let $\mu_1,\ldots,\mu_n$ be generators for 
717: $M=\cX^*(\bG_m^{n+r}/T)$. 
718: Choose $v \in \bZ^{n+r}$ in the interior of 
719: $\Cone(e_1,\ldots,e_{n+r})$ so that $\chi_{\bR}(\Cone(v))=\Cone(\nu)$.
720: Replacing $v$ by a suitably large integral multiple, 
721: we may assume each $v+\mu_i,i=1,\ldots,n$, is in 
722: $\Cone(e_1,\ldots,e_{n+r})$.  If $\chi(v)=d\nu$ then $R_d$ contains
723: a set of generators for $M$, so the induced representation
724: of $\bG_m^{n+r}/T$ on $R_d$ is faithful.  
725: 
726: For the third part, we extract the fan classifying $X$
727: from $P_{d\nu}$, following \cite{fulton} \S 1.5 and 3.4:
728: For each face $Q$ of $P_{d\nu}$, consider the cone
729: $$\sigma_Q=\{v \in N_{\bR}: \left<u,v \right> \leq \left<u',v \right>
730: \text{ for all }u \in Q, u' \in P_{d\nu} \}.$$
731: This assignment is inclusion reversing, so the one-dimensional
732: cones of the fan correspond to facets of $P_{d\nu}$.  Moreover,
733: each facet of $P_{d\nu}$ is induced by one of the facets
734: of $\Cone(e_1,\ldots,e_{n+r})$.  The corresponding one-dimensional
735: cone in $N_{\bR}$ is spanned by $\bar{e}_i^*$.  
736: It remains to verify that each facet of $\Cone(e_1,\ldots,e_{n+r})$
737: actually induces a facet of $P_{d\nu}$.  
738: The hypothesis that $\nu$ is in the moving cone 
739: means that $P_{d\nu}$ intersects each of the 
740: $\Cone(e_1,\ldots,e_{i-1},e_{i+1},\ldots,e_{n+r})$.
741: If $\nu$ is in the interior of the moving cone then
742: the intersection of $P_{d\nu}$ with 
743: $\Cone(e_1,\ldots,e_{i-1},e_{i+1},\ldots,e_{n+r})$
744: meets the relative interior of this cone, hence
745: this cone induces a facet of $P_{d\nu}$.  
746: \end{proof}
747: 
748: 
749: Proposition \ref{prop:whenprojective} yields the following
750: nice consequence:
751: 
752: \begin{prop} 
753: \label{prop:projmodels}
754: Let $X$ be a complete toric variety and $\nu$
755: a divisor class in the interior of $\Move(X)$. 
756: Then there exists a projective toric variety $Y_{\nu}$,
757: with the same one-skeleton as $X$, and polarized by $\nu$.  
758: \end{prop}
759: 
760: \noindent
761: For generic $T$-linearized invertible sheaves on $\bA^{n+r}$,
762: all semistable points are actually stable;  hence 
763: $Y_{\nu}$ is a simplicial toric
764: variety for generic $\nu$ (see \cite{brionproc} 1.2 and \cite{Cox} 2.1).  
765: For the special values $\nu_0$, contained in the
766: walls of the chamber decomposition of \cite{thad}, this
767: fails to be the case.  However, for each special $\nu_0$,
768: there exists a generic $\nu$ so that 
769: $\Cone(\nu)$ is very close to $\Cone(\nu_0)$ and
770: there is a projective, torus-equivariant morphism
771: $Y_{\nu} \ra Y_{\nu_0}$ \cite{thad} 3.11.  
772: The polarization associated to $\nu_0$ pulls back to $Y_{\nu}$,
773: so we obtain the following:
774: 
775: \begin{prop} 
776: \label{prop:goodmodels}
777: Let $X$ be a complete toric variety and $\nu_0$
778: a divisor class in the moving cone of $X$. 
779: Then there exists a simplicial projective toric variety
780: $Y$, with the same one-skeleton as $X$, so that $\nu_0$ is
781: semiample on $Y$.
782: \end{prop}
783: Of course, $\nu_0$ is big when it is in the interior
784: of the effective cone.  
785: 
786: 
787: 
788: 
789: 
790: 
791: \section{The $\Esix$ cubic surface}
792: \label{sect:e6}
793: 
794: By definition, 
795: the {\em $\Esix$ cubic surface} is given by the homogeneous
796: equation
797: \begin{equation} 
798: \label{eqn:main}
799: S=\{(w,x,y,z):xy^2+yw^2+z^3=0\}\subset \bP^3.
800: \end{equation}
801: We recall some elementary properties (see \cite{bruce-wall}
802: for more details on singular cubic surfaces):
803: 
804: \begin{prop}
805: \label{prop:classical}
806: $
807: {}
808: $
809: \begin{enumerate}
810: \item The surface $S$ has a single singularity at the
811: point $p:=(0,1,0,0)$, of type $\Esix$.
812: \item $S$ is the unique cubic surface with this property,
813: up to projectivity.
814: \item $S$ contains a unique line, satisfying the equations
815: $y=z=0$.
816: \end{enumerate}
817: \end{prop}
818: 
819: Any smooth cubic surface may be
820: represented as the blow-up of $\bP^2$ at six points
821: in `general position'.  There is an analogous property
822: of the $\Esix$ cubic surface:
823: 
824: \begin{prop}
825: \label{prop:sixpoints}
826: The $\Esix$ cubic surface $S$ is 
827: the closure of the image of $\bP^2$ under the
828: linear series
829: $$w=a^2c \quad x=-(ac^2+b^3) \quad y=a^3 \quad z=a^2b,$$
830: where 
831: $$
832: \Gamma(\bP^2,\cO_{\bP^2}(1))=\langle a,b,c\rangle.
833: $$
834: This map is the inverse of the projection of $S$ 
835: from the double point  $p$.
836: The affine open subset
837: $$
838: \bA^2:=\{a\neq 0\} \subset \  \bP^2 
839: $$
840: is mapped isomorphically onto $S-\ell$.  In particular,
841: $S\setminus \ell \simeq \bA^2$, so the $\Esix$ cubic surface is a 
842: compactification of $\bA^2$.
843: \end{prop}
844: 
845: \begin{rem}
846: Note that $S$ is not an {\em equivariant} compactification of 
847: $\bG_a^2$, so the general theory of \cite{CLT} does
848: not apply. 
849: 
850: 
851: Indeed, if $S$ were an equivariant compactification of $\bG_a^2$
852: then the projection from $p$ would be $\bG_a^2$-equivariant (see \cite{HT1}).
853: Therefore, the map $\bP^2 \dashrightarrow S$ given above 
854: has to be $\bG_a^2$-equivariant. 
855: The only $\bG_a^2$-action on $\bP^2$ under which a line is invariant
856: is the standard translation action \cite{HT1}.  
857: However, the linear series above is not invariant under the 
858: standard translation action 
859: $$
860: b\mapsto b+\beta a \quad c \mapsto c+\gamma a.
861: $$
862: \end{rem}
863: 
864: \
865: 
866: We proceed to compute the effective cone of the  minimal 
867: resolution $\phi_{\ell}:\tS\ra S$. 
868: Let $\ell \subset \tS$ be the proper transform of the
869: line mentioned in Proposition~\ref{prop:classical}.  
870: 
871: \begin{prop}
872: \label{prop:PIC}
873: The Picard group $\Pic(\tS)$ is a free abelian group of rank seven, generated
874: by $\ell$ and the exceptional curves of $\phi_{\ell}$.  
875: For a suitable ordering $\{F_1,F_2,F_3,F_4,F_5,F_6\}$
876: of the exceptional curves, the intersection
877: pairing takes the form
878: \begin{equation}
879: \label{Fmatrix}
880: \begin{array}{c|ccccccc}
881:  & F_1 & F_2 & F_3 & \ell & F_4 & F_5 & F_6 \\
882: \hline
883: F_1 & -2 & 0 & 1 & 0 & 0 & 0 & 0 \\
884: F_2 & 0 & -2 & 0 & 0  & 0 & 0 & 1   \\
885: F_3 & 1 & 0 & -2 & 0 & 0 & 0 & 1  \\
886: \ell& 0 & 0 & 0 & -1  & 1 & 0  & 0   \\
887: F_4 & 0 & 0 & 0 & 1 & -2 & 1 & 0   \\
888: F_5 & 0 & 0 & 0 & 0 & 1 & -2 & 1    \\
889: F_6 & 0 & 1 & 1 & 0 & 0 & 1 & -2  
890: \end{array}.
891: \end{equation}
892: \end{prop}
893: 
894: \begin{figure}[t]
895: \label{E6figure}
896: \centerline{\includegraphics{E6.eps}}
897: \caption{Dynkin diagram of $\Esix$}
898: \end{figure}
899: 
900: 
901: \begin{prop}
902: \label{prop:effcone}
903: The effective cone $\NE(\tS)$ 
904: is simplicial and generated by 
905: $\Phi:=\{F_1,F_2,F_3,\ell,F_4,F_5,F_6\}$.  
906: Each nef divisor is contained in the monoid generated by 
907: the divisors:
908: \begin{eqnarray*}
909: A_1      &=&\quad \quad  F_2 + F_3 + 2\ell + 2F_4 + 2F_5 +  2F_6  \\
910: A_2      &=& F_1  + F_2 + 2F_3 + 3\ell + 3F_4 + 3F_5 +  3F_6   \\
911: A_3      &=& F_1 + 2F_2 + 2F_3 + 4\ell + 4F_4 + 4F_5 +  4F_6  \\
912: A_{\ell} &=& 2F_1 + 3F_2 + 4F_3 + 3\ell + 4F_4 + 5F_5 +  6F_6  \\
913: A_4      &=& 2F_1 + 3F_2 + 4F_3 + 4\ell + 4F_4 + 5F_5 +  6F_6   \\
914: A_5      &=& 2F_1 + 3F_2 + 4F_3 + 5\ell + 5F_4 + 5F_5 +  6F_6  \\
915: A_6      &=& 2F_1 + 3F_2 + 4F_3 + 6\ell + 6F_4 + 6F_5 +  6F_6  
916: \end{eqnarray*}
917: Moreover $A_{\ell}$ is the anticanonical class $-K_{\tS}$
918: and its sections induce the resolution morphism $\phi_{\ell}:\tS \ra S$.  
919: \end{prop}
920: 
921: \begin{proof}
922: The intersection form in terms of
923: $A:=\{A_1,\ldots,A_6  \}$ is:
924: \begin{equation}
925: \label{Amatrix}
926: \begin{array}{c|ccccccc}
927:  & A_1 & A_2 & A_3 & A_{\ell} & A_4 & A_5 &  A_6  \\
928: \hline
929: A_1      & 0 & 1 & 1 & 2 & 2 & 2 &  2  \\
930: A_2      & 1 & 1 & 2 & 3 & 3 & 3 &  3   \\
931: A_3      & 1 & 2 & 2 & 4 & 4 & 4 &  4  \\
932: A_{\ell} & 2 & 3 & 4 & 3 & 4 & 5 &  6  \\
933: A_4      & 2 & 3 & 4 & 4 & 4 & 5 &  6   \\
934: A_5      & 2 & 3 & 4 & 5 & 5 & 5 &  6  \\
935: A_6      & 2 & 3 & 4 & 6 & 6 & 6 &  6  
936: \end{array}.
937: \end{equation}
938: This is the inverse of the intersection
939: matrix \eqref{Fmatrix} written in terms of the basis $\Phi$,
940: so the $A_i$ generate the dual to 
941: $\Cone(\Phi)$.  Observe that all the 
942: entries of matrix \eqref{Amatrix} are nonnegative and
943: $$\Cone(A)\subset \Cone(\Phi).$$
944: 
945: Suppose that $D$ is an effective divisor on $\tS$.  
946: We write $D$ as a sum of the fixed components
947: contained in $\{F_1,\ldots,F_6,\ell\}$ and the
948: parts moving relative to $\Phi$:
949: $$
950: D=M_{\Phi}+a_1F_1+\ldots +a_6F_6+a_{\ell}\ell, \quad
951: a_1,\ldots,a_6,a_{\ell}\ge 0.
952: $$
953: {\em A priori}, $M_{\Phi}$ may have fixed components,
954: but they are not contained in $\Phi$ (however, see Lemma~\ref{lemm:fin}).  
955: It follows that 
956: $M_{\Phi}$ intersects each element of $\Phi$ nonnegatively,
957: i.e., it is contained in $\Cone(A)$ and thus
958: in $\Cone(\Phi)$.  We conclude that $D\in \Cone(\Phi)$.
959: Since $A_1, ..., A_6, A_{\ell}$ generate $\rN_1(\tS)$ over $\bZ$
960: each nef divisor can be written as a nonnegative linear combination
961: of these divisors. 
962: 
963: To see that $A_{\ell}$ is the anticanonical divisor, we apply
964: adjunction
965: $$
966: K_{\tS}F_i=0, i=1,\ldots, 6 \quad K_{\tS}\ell=-1.
967: $$
968: Nondegeneracy of the intersection form implies
969: $A_{\ell}=-K_{\tS}$.  Since $S$ has rational double points,
970: the resolution map $\phi_{\ell}$ is crepant, i.e., 
971: $\phi_{\ell}^*K_S=K_{\tS}$.
972: Thus 
973: $$
974: \Gamma(A_{\ell})=\Gamma(-K_{\tS})=\Gamma(-\phi_{\ell}^*K_S)=
975: \Gamma(\phi_{\ell}^*\cO_S(+1))
976: $$
977: so the sections of $A_{\ell}$ induce $\phi_{\ell}$.  
978: \end{proof}
979: 
980: 
981: Choose nonzero sections
982: $\xi_1,\ldots,\xi_{\ell}$
983: generating $\Gamma(F_1),\ldots,\Gamma(\ell)$:
984: $$\Gamma(F_1)=\left<\xi_1\right>, \ldots, 
985: \Gamma(F_6)=\left<\xi_6\right>,  \Gamma(\ell)=\left<\xi_{\ell}\right>.$$
986: These are canonical up to scalar multiplication.
987: Each effective divisor
988: $$D=b_1F_1+b_2F_2+b_3F_3+b_{\ell}\ell + b_4F_4+ b_5F_5+b_6F_6$$
989: has a distinguished nonzero section
990: $$\xi^{(b_1,b_2,b_3,b_{\ell},b_4,b_5,b_6)}:=
991: \xi_1^{b_1}\ldots \xi_6^{b_6}\xi_{\ell}^{b_{\ell}}.$$ 
992: The distinguished section of $A_j$
993: is denoted $\xi^{\alpha(j)}$.  
994: Note that we have an injective ring homomorphism 
995: \begin{equation}
996: \label{eqn:ring-h}
997: k[\xi_1,\ldots, \xi_6,\xi_{\ell}]\ra \Cox(\tilde{S}). 
998: \end{equation}
999: 
1000: 
1001: There is a partial order on the monoid of
1002: effective divisors of $\tS$:  $D_1\prec D_2$
1003: if $D_2-D_1$ is effective.  The restriction of this
1004: order to the generators of the nef cone
1005: is illustrated in the diagram below:
1006: $$
1007: \begin{array}{ccccc}
1008:  & & A_6 & & \\
1009:  & & | & & \\
1010:  & & A_5 & & \\
1011:  & & | & & \\
1012:  & & A_4 & & \\
1013:  & \diagup &  & \diagdown & \\
1014: A_{\ell} &  &  &  & A_3 \\
1015:  & \diagdown &  & \diagup & \\
1016:  & & A_2 & & \\
1017:  & & | & & \\
1018:  & & A_1 & & 
1019: \end{array}
1020: $$
1021: Whenever $D_1\prec D_2$
1022: we have an inclusion 
1023: $$\Gamma(D_1)\hra \Gamma(D_2)$$
1024: which is natural up to scalar multiplication.
1025: Indeed, express
1026: $$D_1-D_2=b_1F_1+b_2F_2+\ldots+b_6F_6+b_{\ell}\ell,
1027: \quad b_j \ge 0$$
1028: so we have
1029: \begin{eqnarray*}
1030: s_1 &\mapsto&
1031: \xi^{(b_1,b_2,b_3,b_{\ell},b_4,b_5,b_6)}s_1 \\
1032: \Gamma(D_1) & \hra & \Gamma(D_2).
1033: \end{eqnarray*}
1034: 
1035: 
1036: The homomorphism \eqref{eqn:ring-h} is not surjective, and we
1037: now look for generators of 
1038: $\Cox(\tS)$ beyond the $\xi_j$.  
1039: Consider the subring
1040: $$
1041: \Cox_a(\tS)=\bigoplus_{\nu\in \NM(\tS)}\Cox(\tS)_{\nu}
1042: $$ 
1043: obtained by restricting to degrees corresponding 
1044: to nef classes on $\tS$.  
1045: The following lemma implies that 
1046: any homogeneous element $\mathsf s_D\in \Cox(\tS)$ 
1047: can be written in the form
1048: $$
1049: \mathsf s_D=m_D\xi_1^{b_1}\cdots \xi_6^{b_6}\xi_{\ell}^{b_\ell}
1050: $$
1051: with nonnegative exponents and $m_D\in \Cox_a(\tS)$.  
1052: \begin{lem}
1053: \label{lemm:fin}
1054: Let $D$ be an effective divisor on $\tS$ with fixed part $F_D$ and moving
1055: part $M_D$. Then $F_D$ is supported in 
1056: $\{ F_1, \ldots , F_6, \ell\}$, and $M_D$ is     
1057: a linear combination of 
1058: $A_1, \ldots, A_6,A_{\ell}$ with nonnegative
1059: coefficients.    
1060: \end{lem}
1061: \begin{proof} 
1062: Clearly $M_D$ is nef, so the description of the nef divisors in
1063: Proposition~\ref{prop:effcone} gives the expression
1064: in terms of the $A_i$.  
1065: Proposition~\ref{prop:semiample} shows $M_D$ is
1066: semiample with vanishing higher cohomology;  the last part of
1067: Proposition~\ref{prop:effcone} gives the requisite
1068: positivity of the anticanonical class.  
1069: 
1070: Let $F$ be a fixed component of $D$ not supported in 
1071: $\{ F_1, \ldots , F_6, \ell\}$. 
1072: To arrive at a contradiction, we need to show that
1073: $h^0(\cO_{\tS}(M_D+F))> h^0(\cO_{\tS}(M_D))$. 
1074: Since $M_D$ has vanishing higher cohomology and 
1075: $$
1076: h^2(\cO_{\tS}(F+M_D))=h^0(\cO_{\tS}(K_{\tS}-F-M_D)) =0 
1077: $$
1078: it suffices to show that 
1079: $$
1080: \chi(\cO_{\tS}(F+M_D))> \chi(\cO_{\tS}(M_D)).
1081: $$
1082: By Riemann-Roch, it suffices to show that 
1083: $$
1084: F^2+2M_DF-K_{\tS}F >0
1085: $$
1086: or, equivalently, 
1087: $$
1088: F^2+K_{\tS}F +2M_DF -2K_{\tS}F = 2g(F)-2 + 2M_DF -2K_{\tS}F >0.
1089: $$
1090: Since $F$ is irreducible, $g(F)\ge 0$ 
1091: and $M_DF\ge 0$ and $-K_{\tS}F > 1$, 
1092: as  $M_D$ is nef and $-K_{\tS}$ is nonpositive only along the
1093: exceptional curves and has degree 1 only on the line $\ell$ 
1094: (see Proposition~\ref{prop:classical}). 
1095: \end{proof}
1096: 
1097: 
1098: Corollary \ref{coro:hilbert} gives
1099: the dimensions of the graded pieces of $\Cox_a(\tS)$.
1100: We focus first on the generators of the nef cone,
1101: introducing sections $\tau_j\in \Gamma(A_j)$ as needed to 
1102: achieve the prescribed dimensions:
1103: \begin{eqnarray*}
1104: \Gamma(A_1)&=&\left<\xi^{\alpha(1)},\tau_1\right>\\
1105: \Gamma(A_2)&=&\left<\xi^{\alpha(2)},\xi^{\alpha(2)-\alpha(1)}\tau_1,
1106: \tau_2 \right> \\
1107: \Gamma(A_{\ell})&=&\left<\xi^{\alpha(\ell)},\xi^{\alpha(\ell)-\alpha(1)}\tau_1,
1108: \xi^{\alpha(\ell)-\alpha(2)}\tau_2,\tau_{\ell} \right> 
1109: \end{eqnarray*}
1110: The sections of $A_{\ell}$ induce $\phi_{\ell}:\tS\ra S\subset \bP^3$
1111: by Proposition \ref{prop:effcone}, and can be identified with
1112: the coordinates $w,x,y,z$ of Equation~\eqref{eqn:main}.  
1113: Since $A_1\prec A_2 \prec A_{\ell}$,
1114: we have 
1115: $$\Gamma(A_1)\hookrightarrow \Gamma(A_2) \hookrightarrow\Gamma(A_{\ell}).$$
1116: We can identify $\Gamma(A_1)=\left<y,z\right>$;  these
1117: correspond to projecting $S$ from the line $\ell=\{y=z=0\}$
1118: and induce a conic bundle structure
1119: $$
1120: \phi_1:\tS \ra \bP^1.
1121: $$
1122: We have $\Gamma(A_2)=\left<x,y,z\right>$;  these correspond
1123: to projecting $S$ from the singularity $p=\{w=y=z=0\}$ and induce the
1124: blow-up realization
1125: $$
1126: \phi_2:\tS \ra \bP^2.
1127: $$
1128: Therefore, we may choose $\tau_1,\tau_2,$ and $\tau_{\ell}$ so that 
1129: $$
1130: y=\xi^{\alpha(\ell)} \quad w=\xi^{\alpha(\ell)-\alpha(2)}\tau_2 \quad
1131: z=\xi^{\alpha(\ell)-\alpha(1)}\tau_1 \quad x=\tau_{\ell}.
1132: $$
1133: 
1134: We obtain the following induced sections for $A_3,A_4,A_5,$ and $A_6$:
1135: \begin{eqnarray*}
1136: \Gamma(A_3)&=&\left<\xi^{\alpha(3)},\xi^{\alpha(3)-\alpha(1)}\tau_1,
1137: \xi^{\alpha(3)-\alpha(2)}\tau_2,\xi^{\alpha(3)-2\alpha(1)}\tau_1^2\right>\\
1138: \Gamma(A_4)&=&\left<\xi^{\alpha(4)},\xi^{\alpha(4)-\alpha(1)}\tau_1,
1139: \xi^{\alpha(4)-\alpha(2)}\tau_2,
1140: \xi^{\alpha(4)-\alpha(\ell)}\tau_{\ell}, 
1141: \xi^{\alpha(4)-2\alpha(1)}\tau_1^2\right>\\
1142: \Gamma(A_5)&=&\left<\xi^{\alpha(5)},\xi^{\alpha(5)-\alpha(1)}\tau_1,
1143: \xi^{\alpha(5)-\alpha(2)}\tau_2,
1144: \xi^{\alpha(5)-\alpha(\ell)}\tau_{\ell}, 
1145: \xi^{\alpha(5)-2\alpha(1)}\tau_1^2, \right. \\
1146: & &\quad \quad \left. \xi^{\alpha(5)-\alpha(1)-\alpha(2)}\tau_1\tau_2 \right>\\
1147: \Gamma(A_6)&=&\left<\xi^{\alpha(6)},\xi^{\alpha(6)-\alpha(1)}\tau_1,
1148: \xi^{\alpha(6)-\alpha(2)}\tau_2,
1149: \xi^{\alpha(6)-\alpha(\ell)}\tau_{\ell}, 
1150: \xi^{\alpha(6)-2\alpha(1)}\tau_1^2, \right. \\
1151: & & \quad \quad \left.
1152: \xi^{\alpha(6)-\alpha(1)-\alpha(2)}\tau_1\tau_2,
1153: \xi^{\alpha(6)-2\alpha(2)}\tau^2_2,
1154: \xi^{\alpha(6)-3\alpha(1)}\tau_1^3\right>
1155: \end{eqnarray*}
1156: Equation~\eqref{eqn:main} gives the relation
1157: $$
1158: \tau_{\ell}\xi^{2\alpha(\ell)}+\tau^2_2\xi^{3\alpha(\ell)-2\alpha(2)}+
1159: \tau_1^3\xi^{3\alpha(\ell)-3\alpha(1)}=0.
1160: $$
1161: Dividing by a suitable monomial $\xi^{\beta}$, we obtain
1162: $$
1163: \tau_{\ell}\xi_{\ell}^3\xi_4^2\xi_5 + \tau_2^2  \xi_2 + 
1164: \tau_1^3 \xi_1^2 \xi_3=0,
1165: $$
1166: a dependence relation in $\Gamma(A_6)$.
1167: This is the only such relation:
1168: Any other relation,
1169: after multiplying through by $\xi^{\beta}$, yields a cubic 
1170: form vanishing on $S\subset \bP^3$,
1171: but equation ~\eqref{eqn:main} is the only such form.
1172: It follows that the sections given above for $A_1,\ldots,A_5$
1173: form bases for $\Gamma(A_1),\ldots,\Gamma(A_5)$.  
1174: 
1175: Since 
1176: $$A_3 \prec A_4 \prec A_5 \prec A_6 \prec 2A_{\ell}$$
1177: we have
1178: \begin{eqnarray*}
1179: \Gamma(A_3)&\hookrightarrow &\Gamma(A_4) \hookrightarrow
1180: \Gamma(A_5) \hookrightarrow \Gamma(A_6) \\
1181: & \hookrightarrow &\Gamma(2A_{\ell})=
1182: \left<w^2,wx,wy,x^2,xy,xz,y^2,yz,z^2 \right>
1183: \end{eqnarray*}
1184: and identifications
1185: \begin{eqnarray*}
1186: \Gamma(A_3)&=&\left<y^2,yz,
1187: wy,z^2 \right>\\
1188: \Gamma(A_4)&=&\left<y^2,yz,
1189: wy,xy,z^2\right>\\
1190: \Gamma(A_5)&=&\left<y^2,yz,wy,xy,z^2,wz\right> \\
1191: \Gamma(A_6)&=&\left<y^2,yz,wy,xy,z^2,wz,w^2\right>. 
1192: \end{eqnarray*}
1193: The sections of $A_3$ induce a morphism
1194: $$\phi_3:\tilde{S} \ra  \bP^3$$
1195: onto a quadric surface with a single ordinary double
1196: point.  The sections of $A_4$ induce
1197: a morphism 
1198: $$\phi_4:\tilde{S} \ra \bP^4$$
1199: with image a quartic Del Pezzo surface with
1200: a rational double point of type $\mathbf{D}_5$.  The sections of
1201: $A_5$ induce a morphism
1202: $$\phi_5:\tilde{S} \ra \bP^5$$ 
1203: with image a quintic Del Pezzo surface with a rational double point
1204: of type $\mathbf{A}_4$.  The sections of $A_6$
1205: induce a morphism 
1206: $$\phi_6:\tilde{S} \ra \bP^6$$
1207: with image a sextic Del Pezzo surface with two rational double points,
1208: of types $\mathbf{A}_1$ and $\mathbf{A}_2$.  
1209: 
1210: We summarize this analysis in the following proposition
1211: \begin{prop}\label{prop:semiampsect}
1212: Every section of $A_j, j=1,2,3,\ell,4,5,6,$
1213: can be expressed as a polynomial in 
1214: $\xi_1,\ldots,\xi_6,\xi_{\ell},\tau_1,\tau_2,\tau_6$.  The only
1215: dependence relation among these is
1216: $$
1217: \tau_{\ell}\xi_{\ell}^3\xi_4^2\xi_5 + \tau_2^2  \xi_2 + 
1218: \tau_1^3 \xi_1^2 \xi_3=0
1219: $$
1220: in $\Gamma(A_6)$.  Each $A_j$ is globally generated and induces
1221: a morphism 
1222: $$\phi_j:\tS \ra \bP^{\chi-1}, \quad \chi=\chi(\cO_{\tS}(A_j)).$$
1223: \end{prop}
1224: 
1225: 
1226: \
1227: 
1228: 
1229: 
1230: The remainder of this section is devoted to proving the following:
1231: \begin{theo}
1232: \label{theo:main}
1233: The homomorphism 
1234: $$
1235: \varrho\,:\, 
1236: k[\xi_1,...,\xi_6,\xi_{\ell}, \tau_1, \tau_2,\tau_\ell] /
1237: \langle \tau_{\ell}\xi_{\ell}^3\xi_4^2\xi_5 + \tau_2^2  \xi_2 + 
1238: \tau_1^3 \xi_1^2 \xi_3\rangle \ra \Cox(\tS)
1239: $$
1240: is an isomorphism. 
1241: \end{theo}
1242: 
1243: If $\varrho$ were not injective, its kernel would have
1244: nontrivial elements in degree $\nu=dA_{\ell}$, for some $d$
1245: sufficiently large.  These translate
1246: into homogeneous polynomials of degree $d$ vanishing on
1247: $S\subset \bP^3$.  All such polynomials are multiples of the
1248: cubic form defining $S$, which itself is a multiple of the relation
1249: we already have. 
1250: 
1251: It remains to show that $\varrho$ is surjective.  By
1252: Proposition~\ref{prop:effcone}, Lemma~\ref{lemm:fin} 
1253: and the analysis of the sections of the $A_i$,
1254: it suffices to prove:
1255: \begin{prop}\label{prop:generate}
1256: $\varrho$ is surjective in degrees corresponding to nef
1257: divisor classes of $\tS$.
1258: \end{prop}
1259: \begin{lem}\label{lem:embed}
1260: For any positive integers 
1261: $c_1,c_2,c_3,c_{\ell},c_4,c_5,c_6$, the image of 
1262: $$\Gamma(A_1)^{\otimes c_1} \otimes \ldots \otimes
1263:   \Gamma(A_{\ell})^{\otimes c_{\ell}} \otimes
1264: \ldots  \otimes  \Gamma(A_6)^{\otimes c_6}
1265: \longrightarrow \Gamma(c_1A_1+\ldots+c_6A_6)$$
1266: is a linear series embedding $\tS$.
1267: \end{lem}
1268: \begin{proof}
1269: Proposition~\ref{prop:semiampsect} says that each $A_j$ is globally
1270: generated, so if the image of 
1271: $$\Gamma(A_1) \otimes \ldots \otimes \Gamma(A_6)
1272: \longrightarrow \Gamma(A_1+\ldots+A_6)$$
1273: embed $\tS$ then the general result follows.
1274: We use the standard criterion:
1275: a linear series gives an embedding iff any length-two subscheme 
1276: $\Sigma \subset \tS$ imposes two independent conditions on the
1277: linear series.  
1278: 
1279: First, suppose the support of $\Sigma$ is not contained in the exceptional
1280: locus of $\phi_{\ell}:\tS \ra S$, i.e., the curves $F_1,F_2,F_3,F_4,F_5,F_6$.
1281: Then $\phi_{\ell}$ maps $\Sigma$ to a subscheme of length two, which
1282: imposes independent conditions on $\Gamma(A_{\ell})$,
1283: and thus independent conditions on the linear series in question.  
1284: Second, suppose that $\Sigma \subset F_j$ for some $j$ (resp.
1285: $\Sigma \subset \ell$).  Since $A_j \cdot F_j=1$ (resp. 
1286: $A_{\ell}\cdot \ell=1$), $\phi_j$ maps $F_j$ (resp. $\ell$) 
1287: isomorphically onto a line.  It follows that $\Sigma$
1288: imposes independent conditions on $\Gamma(A_j)$.  
1289: Third, suppose that $\Sigma$ is reduced with support in $F_i$ and $F_j$,
1290: but is not contained in either $F_i$ or $F_j$.  Consider the
1291: chain of rational curves containing $F_i$ and $F_j$ (see 
1292: Figure~\ref{E6figure}.)  There exists a curve $F_k$
1293: in this chain so that $\phi_k(F_i)\neq \phi_k(F_j)$, so
1294: $\Sigma$ imposes independent conditions on $\Gamma(A_k)$.  
1295: Fourth, suppose that $\Sigma$ is nonreduced and supported
1296: in $F_j$ but not contained in any $F_i$ or $\ell$.  The
1297: morphism $\phi_j$ ramifies at points where $F_j$ meets
1298: one of the other exceptional curves, and the kernel of the
1299: tangent morphism $d\phi_j$ consists of the tangent vectors
1300: to the curves contracted by $\phi_j$.   It follows
1301: that $\phi_j(\Sigma)$ has length two and imposes independent
1302: conditions on $\Gamma(A_j)$.  
1303: \end{proof}
1304: 
1305: The polynomial ring 
1306: $$k[\xi_1,\ldots,\xi_6,\xi_{\ell},\tau_1,\tau_2,\tau_{\ell}]$$
1307: is graded by the N\'eron-Severi group of $\tS$
1308: $$\deg(\xi_j)=F_j, j=1,\ldots,6 \ \deg(\xi_{\ell})=\ell \ 
1309: \deg(\tau_j)=A_j, j=1,2,\ell.$$
1310: This gives an action of the N\'eron-Severi torus
1311: $T(\tS)$ on the corresponding affine space $\bA^{10}$.  
1312: 
1313: We consider the projective toric varieties that arise as 
1314: quotients of $\bA^{10}$ by $T(\tS)$.  As sketched in
1315: \S \ref{sect:toric}, 
1316: these varieties have the one-skeleton
1317: $$ x_1=(0,1,2),x_2=(1,1,3),x_3=(1,2,4), x_{\ell}=(2,3,3),x_4=(2,3,4) $$
1318: $$ x_5=(2,3,5), x_6=(2,3,6), t_1=(-1,0,0), t_2=(0,-1,0), 
1319: t_{\ell}=(0,0,-1) $$
1320: where the $x_j$ correspond to the $\xi_j$ and the $t_j$ correspond
1321: to the $\tau_j$.  
1322: 
1323: \begin{lem} \label{lem:computemoving}
1324: Let $X$ be a toric threefold with one-skeleton $\{x_1,\ldots,t_{\ell}\}$
1325: and divisor class-group $\cX^*(T(\tS))=\rN_1(\tS)$.  
1326: Then 
1327: $$\Move(X)=\Cone(A_1,\ldots,A_6,A_{\ell}).$$
1328: \end{lem}
1329: \begin{proof}
1330: Proposition~\ref{prop:toricmove} reduces this to computing the
1331: intersection of the cones generated by subsets of
1332: $$\{F_1,\ldots,F_6,\ell,A_1,A_2,A_{\ell}\}$$
1333: with nine elements.  Since $A_1,A_2,A_{\ell}$ are effective
1334: combinations of the classes $F_1,\ldots,F_6,$ and $\ell$, it suffices to
1335: compute
1336: $$
1337: \Cone(F_1,\ldots,\hat{\ell},\ldots,A_{\ell}) \bigcap
1338: \left(\bigcap_{i=1,\ldots,6} \Cone(F_1,\ldots,\hat{F_i},\ldots,
1339: A_{\ell})  \right).
1340: $$
1341: 
1342: This intersection obviously contains $A_1,A_2,$ and $A_{\ell}$,
1343: and it is a straightforward computation to show that it also 
1344: contains $A_3,A_4,A_5,A_6$.  For the reverse inclusion, 
1345: suppose that $D$ is contained in the intersection.
1346: Considering $D$ as a divisor
1347: on $\tS$, we see that 
1348: $$D\cdot F_1,\ldots,D\cdot F_6,D\cdot \ell$$
1349: are all nonnegative.  Thus $D$ is an effective sum of $A_j$
1350: by Proposition \ref{prop:effcone}.  
1351: \end{proof}
1352: 
1353: Combining Lemmas~\ref{lem:computemoving} and \ref{lem:embed}
1354: with Propositions~\ref{prop:whenprojective} and 
1355: \ref{prop:semiampsect}, we obtain the following
1356: \begin{prop}\label{prop:toricembedding}
1357: Let $\nu$ be an ample divisor on $\tS$.  Then there exists a
1358: projective toric variety $Y_{\nu}$ with one-skeleton
1359: $\{x_1,\ldots,t_{\ell}\}$ and polarization $\nu$, and an embedding
1360: $\tS \hookrightarrow Y_{\nu}$ with the following properties:
1361: \begin{enumerate}
1362: \item{the divisor class group of $Y_{\nu}$
1363: is isomorphic to the divisor class group of $\tS$
1364: so that the moving cone of $Y_{\nu}$
1365: is identified with the nef cone of $\tS$;}
1366: \item{
1367: the equation for $\tS$ in the Cox ring of $Y_{\nu}$ is 
1368: $$
1369: \tau_{\ell}\xi_{\ell}^3\xi_4^2\xi_5 + \tau_2^2  \xi_2 +
1370: \tau_1^3 \xi_1^2 \xi_3=0
1371: $$
1372: and $[\tS]=A_6$ in the divisor class group of
1373: $Y_{\nu}$;}
1374: \item{$\Cox(Y_{\nu})=k[\xi_1,\ldots,\tau_{\ell}]$
1375: and is mapped isomorphically to the image of the homomorphism $\varrho$.}
1376: \end{enumerate}
1377: \end{prop}
1378: 
1379: For each toric variety $Y_{\nu}$, we can consider the
1380: exact sequence of sheaves
1381: $$0 \ra I_{\tS} \ra \cO_{Y_{\nu}}
1382: \ra \cO_{\tS} \ra 0,$$
1383: where $I_{\tS}\simeq \cO_{Y_{\nu}}(-A_6)$ is the ideal sheaf
1384: of $\tS$.  Given an element $\theta$ in the divisor class group 
1385: of $Y_{\nu}$, we can twist to obtain
1386: $$0 \ra I_{\tS}(\theta) \ra \cO_{Y_{\nu}}(\theta)
1387: \ra \cO_{\tS}(\theta) \ra 0.$$
1388: We should make precise what we mean by the twist
1389: $\cF(\theta)$ of a coherent sheaf
1390: $\cF$ on $Y_{\nu}$:  Realize $\cF$
1391: as the sheafification of a graded module $F$ over $\Cox(Y_{\nu})$
1392: (which exists by \cite{Mus} Theorem 1.1, \cite{Cox} Proposition 3.1), 
1393: shift $F$ by $\theta$,
1394: and then resheafify the shifted module to obtain $\cF(\theta)$.  
1395: Twisting respects exact sequences \cite{Cox} 3.1.  
1396: 
1397: 
1398: The anticanonical divisor of a toric variety is the 
1399: sum of the invariant divisors \cite{fulton} p. 89, so 
1400: $$-K_{Y_{\nu}}=F_1+\ldots +F_6+\ell+A_1+A_2+A_{\ell}=A_{\ell}+A_6$$
1401: and we can rewrite our exact sequence as
1402: $$0 \ra \cO_{Y_{\nu}}(K_{Y_{\nu}}+A_{\ell}+\theta) \ra \cO_{Y_{\nu}}(\theta)
1403: \ra \cO_{\tS}(\theta) \ra 0.$$
1404: 
1405: Suppose that $\theta$ corresponds to a nef class on $\tS$;  we shall
1406: prove that $\varrho$ is surjective in degree $\theta$,
1407: thus proving Proposition~\ref{prop:generate} and
1408: Theorem~\ref{theo:main}.  Since
1409: $$\Gamma(\cO_{Y_{\nu}}(\theta))\simeq 
1410: k[\xi_1,\ldots,\xi_6,\xi_{\ell},\tau_1,\tau_2,\tau_{\ell}]_{\theta}$$
1411: it suffices to show that 
1412: $$H^1(\cO_{Y_{\nu}}(K_{Y_{\nu}}+A_{\ell}+\theta))=0.$$
1413: We apply Proposition~\ref{prop:goodmodels}, with $\nu_0=A_{\ell}+\theta$,
1414: to get a {\em simplicial} toric variety $Y_{\nu}$ on which
1415: $A_{\ell}+\theta$ is nef.  
1416: As $A_{\ell}$ is in the interior
1417: of the effective cone of $Y_{\nu}$, $A_{\ell}+\theta$ is also big.  
1418: Note that $Y_{\nu}$ has finite-quotient singularities, which are
1419: log terminal \cite{KM} \S 5.2.  The desired vanishing follows
1420: from Theorem 2.17 of \cite{Ko}.  Alternately, we could apply
1421: Theorem 0.1 of \cite{Mus}, which applies in arbitrary characteristic
1422: and obviates the need to pass to a simplicial model.
1423: 
1424: 
1425: 
1426: 
1427: 
1428: 
1429: 
1430: 
1431: 
1432: 
1433: 
1434: 
1435: 
1436: 
1437: 
1438: 
1439: 
1440: 
1441: 
1442: 
1443: 
1444: 
1445: 
1446: 
1447: 
1448: 
1449: 
1450: 
1451: 
1452: 
1453: 
1454: 
1455: \
1456: 
1457: 
1458: 
1459: 
1460: 
1461: 
1462: 
1463: \section{$\Dfour$ cubic surface} 
1464: \label{sect:d4}
1465: 
1466: The strategy of the previous section can applied to other
1467: surfaces as well.  Here we illustrate it in the case
1468: of a cubic surface given by the homogeneous equation
1469: $$
1470: S=\{(x_1,x_2,x_3,w)\, :\,  w(x_1+x_2+x_3)^2 = x_1x_2x_3 \} \subset \bP^3.
1471: $$
1472: We summarize its properties: 
1473: \begin{enumerate}
1474: \item $S$ has a single singularity at the point $p=(0,0,0,1)$ of type $\Dfour$.
1475: \item $S$ contains 6 lines with the equations
1476: $$
1477: \begin{array}{cc}
1478: \ell_1':= \{ w=x_1=0\}  &  m_1':= \{ x_1=x_2+x_3=0\} \\ 
1479: \ell_2':= \{ w=x_2=0\}  &  m_2':= \{ x_2=x_1+x_3=0\}  \\
1480: \ell_3':= \{ w=x_3=0\}  &  m_3':= \{x_3=x_1+x_2=0\} 
1481: \end{array}
1482: $$
1483: \item $S$ is the closure of the image of $\bP^2$ under the linear series 
1484: $$
1485: x_1=u_1(u_1+u_2+u_3)^2, \,\, 
1486: x_2=u_2(u_1+u_2+u_3)^2, \,\, 
1487: x_3=u_3(u_1+u_2+u_3)^2,
1488: $$
1489: $$
1490: w=u_1u_2u_3,  
1491: $$
1492: where $\langle u_1,u_2,u_3\rangle =\Gamma(\bP^2,\cO_{\bP^2}(1))$. 
1493: \end{enumerate}
1494: 
1495: \begin{rem}
1496: There are two isomorphism classes of cubic surfaces with a $\Dfour$ singularity
1497: \cite{bruce-wall} Lemma 4.  The other class is
1498: $$S_0=\{(x_1,x_2,x_3,w): w(x_1+x_2+x_3)^2=x_1x_2(-x_1-x_2) \};$$
1499: it is obtained from $S$ by substituting
1500: $$(w,x_1,x_2,x_3)\mapsto (t^{-2}w,x_1,x_2,
1501:  tx_3+(t-1)x_1+(t-1)x_2)$$
1502: and letting $t\ra 0$ in the resulting equation.
1503: 
1504: We can distinguish $S$ and $S_0$ geometrically:
1505: In $S$, the three lines not containing $p$ do not share a common point.
1506: In $S_0$, the analogous lines
1507: $$\{w=x_1=0 \}, \{w=x_2=0 \}, \{w=x_1+x_2=0 \} \subset S_0
1508: $$
1509: are coincident at $w=x_1=x_2=0$.
1510: \end{rem}
1511: 
1512: Let $\beta:\tS\ra S$ denote the minimal desingularization of $S$
1513: and 
1514: $$
1515: \ell_1,\ell_2,\ell_3, m_1, m_2,m_3
1516: $$ 
1517: the strict transforms of the lines. 
1518: The rational map $S\dashrightarrow \bP^2$ induces a morphism 
1519: $\tS\ra \bP^2$
1520: and let $L$ denote the pullback of the hyperplane class. 
1521: Let $E_0,E_1,E_2,E_3$ be the exceptional divisors of $\beta$, ordered 
1522: so that we have the following intersection matrix: 
1523: 
1524: \begin{equation}
1525: \label{Dmatrix}
1526: \begin{array}{c|ccccccc}
1527:  & L & E_1 & E_2 & E_3 & m_1 & m_2 & m_3 \\
1528: \hline
1529: L   & 1 & 0 & 0 & 0 & 0 & 0 & 0 \\
1530: E_1 & 0 & -2 & 0 & 0  & 1 & 0 & 0   \\
1531: E_2 & 0 & 0 & -2 & 0 & 0 & 1 & 0  \\
1532: E_3 & 0 & 0 & 0 & -2  & 0 & 0  & 1   \\
1533: m_1 & 0 & 1 & 0 & 0 & -1 & 0 & 0   \\
1534: m_2 & 0 & 0 & 1 & 0 & 0 & -1 & 0    \\
1535: m_3 & 0 & 0 & 0 & 1 & 0 & 0 & -1  
1536: \end{array}.
1537: \end{equation}
1538: This is a rank seven unimodular matrix;  since the Picard group 
1539: of $\tS$ has rank seven, it is generated by 
1540: $L,E_1,E_2,E_3,m_1,m_2,m_3$. 
1541: In particular, we have
1542: $$
1543: E_0=L-(E_1+E_2+E_3 +m_1 + m_2+m_3) \,\,\, \text{ and } \,\,\,  
1544: \ell_j=L-E_j-2m_j. 
1545: $$
1546: The anticanonical class is given by 
1547: $$
1548: -K_{\tS}= 3L -(E_1+E_2+E_3) -2(m_1+m_2+m_3) = \ell_1+\ell_2+\ell_3.
1549: $$
1550: 
1551: \begin{prop}
1552: The effective cone $\NE(\tS)$ is generated by 
1553: $$
1554: \Xi:=\{ E_0, E_1, E_2, E_3, m_j, \ell_j\}.   
1555: $$
1556: \end{prop}
1557:  
1558: \begin{proof}
1559: Each effective divisor $D$ can be expressed as a sum 
1560: $$
1561: D=M_{\Xi}+b_{E_0} E_0 + b_{E_1}E_1 + \ldots + b_{\ell_{3}}\ell_3,
1562: $$
1563: with nonnegative coefficients, where $M_{\Xi}$ intersects each of the
1564: elements in $\Xi$ nonnegatively and thus is in the dual cone
1565: to $\Cone(\Xi)$.  Direct computation shows that
1566: the dual to $\Cone(\Xi)$ has generators
1567: $$
1568: L,  L-E_i-m_i, 2L-E_i -2m_i, 2L-E_i- E_j -2m_i-2m_j,
1569: $$
1570: $$ 
1571: 2L-E_i-E_j-m_i-2m_j.
1572: $$
1573: Each of these is contained in $\Cone(\Xi)$:
1574: $$
1575: \begin{array}{rcl}
1576: L                   & =&\ell_i+E_i+2m_i,\\
1577: 2L-E_i-2m_i         & =&2\ell_i+E_i+2m_i,\\
1578: 2L-E_i-E_j-m_i-2m_j & =&\ell_i+\ell_j+m_i,\\
1579: L-E_i-m_i           &= &\ell_i+m_i, \\
1580: 2L-E_i-E_j-2m_i-2m_j&= &\ell_i+\ell_j.
1581: \end{array}
1582: $$
1583: It follows that $M_{\Xi}$ and $D$ are 
1584: sums of elements in $\Xi$ with nonnegative coefficients. 
1585: \end{proof}
1586: 
1587: Each of the divisors $m_i,\ell_i$ and $E_i$ has a distinguished
1588: nonzero section (up to a constant), denoted $\mu_i, \la_i$ and $\eta_i$, 
1589: respectively.  We have
1590: $$
1591: \{ \la_i\eta_i\mu_i^2, \eta_0\eta_1\eta_2\eta_3\mu_1\mu_2\mu_3\}
1592: \subset \Gamma(L),
1593: $$
1594: and we may identify 
1595: $$
1596: u_i= \la_i\eta_i\mu_i^2\,\,\, \text{  and }\,\,\,  
1597: u_1+u_2+u_3=\eta_0\eta_1\eta_2\eta_3\mu_1\mu_2\mu_3
1598: $$
1599: after suitably normalizing the $\mu_i,\la_i,$ and $\eta_i$.  
1600: The dependence relation among the sections in $\Gamma(L)$ translates 
1601: into 
1602: \begin{equation}
1603: \label{eqn:la}
1604: \la_1\eta_1\mu_1^2+ \la_2\eta_2\mu_2^2+\la_3\eta_3\mu_3^2 = 
1605: \eta_0\eta_1\eta_2\eta_3\mu_1\mu_2\mu_3.
1606: \end{equation}
1607: An argument similar to the one given at the end of Section~\ref{sect:e6}
1608: proves that the natural homomorphism 
1609: $$
1610: k[\eta_0, ..., \eta_3, \mu_i,\la_i ]/  
1611: \langle \la_1\eta_1\mu_1^2+ \la_2\eta_2\mu_2^2+\la_3\eta_3\mu_3^2 - 
1612: \eta_0\eta_1\eta_2\eta_3\mu_1\mu_2\mu_3 \rangle \ra \Cox(\tS)
1613: $$
1614: is an isomorphism. 
1615: 
1616: \
1617: 
1618: 
1619: The cubic surface $S$ admits an $\mathfrak S_3$-action on the coordinates
1620: $x_1,x_2,x_3$. In  particular, it admits nonsplit forms
1621: over nonclosed  ground fields.  
1622: They can be expressed as follows: 
1623: Let $K/k$ be a cubic extension with Galois closure $E/k$.
1624: Fix a basis 
1625: $\{\gamma,\gamma',\gamma''\}$ for $K$ over $k$
1626: so that elements $Y\in K$ can be represented as
1627: $$
1628: Y=y\gamma+ y'\gamma'+y''\gamma''
1629: $$
1630: with $y,y',y''\in k$.  Choose
1631: $\sigma\in \Gal(E/k)$ so that $\sigma$ and $\sigma^2$ are coset
1632: representatives $\Gal(E/k)$ modulo $\Gal(E/K)$.  Then
1633: $$
1634: w\cdot \Tr_{K/k}(Y)^2=\rN_{K/k}(Y)
1635: $$
1636: is isomorphic, over $E$, to $S$:
1637: $$
1638: \begin{array}{ccl}
1639: x_1 & = & Y= y\gamma +y'\gamma'+y''\gamma'' \\
1640: x_2 & = & \sigma(Y)= 
1641: y\sigma(\gamma) +y'\sigma(\gamma')+y''\sigma(\gamma'') \\
1642: x_3 & = & \sigma^2(Y)=
1643: y\sigma^2(\gamma) +y'\sigma^2(\gamma')+y''\sigma^2(\gamma'') 
1644: \end{array}.
1645: $$
1646: 
1647: Assigning elements $U,V,W\in K$ to 
1648: $\eta_1, \mu_1$ and $\la_1$, respectively, 
1649: the torsor equation \eqref{eqn:la} takes the form
1650: $$
1651: \Tr_{K/k}(UV^2W)=\eta_0\rN_{K/k}(UV).
1652: $$
1653: 
1654: 
1655: \bibliographystyle{smfplain}
1656: \bibliography{e6}
1657: 
1658: \end{document}
1659: 
1660: 
1661: 
1662: 
1663: 
1664: 
1665: 
1666: 
1667: 
1668: 
1669: 
1670: