math0604193/qd4.tex
1: %%% Local Variables: 
2: %%% mode: latex
3: %%% TeX-master: t
4: %%% End: 
5: 
6: 
7: \documentclass[12pt,english]{smfart}
8: \usepackage{smfenum}
9: \usepackage{bull}
10: \usepackage{amscd}
11: \usepackage{amssymb}
12: \usepackage{epsfig}
13: \usepackage{graphicx}
14: \usepackage[all]{xy}
15: 
16: 
17: \textheight 20cm \textwidth 13cm
18: \advance\headheight 2pt
19: \calclayout
20: 
21: \usepackage[T1]{fontenc}
22: \usepackage{ae}
23: 
24: \renewcommand\AA{\mathbb{A}}
25: \newcommand\QQ{\mathbb{Q}}
26: \newcommand\CC{\mathbb{C}}
27: \newcommand\NN{\mathcal{N}}
28: \newcommand\RR{\mathbb{R}}
29: \newcommand\R{\mathcal{R}}
30: \newcommand\TT{\mathcal{T}}
31: \newcommand\ZZ{\mathbb{Z}}
32: \newcommand\ZZp{\ZZ_{>0}}
33: \newcommand\PP{\mathbb{P}}
34: \newcommand\xx{\mathbf{x}}
35: \newcommand\dd{\,\mathrm{d}}
36: \newcommand\Pone{{\PP^1}}
37: \newcommand\Ptwo{{\PP^2}}
38: \newcommand\Pthree{{\PP^3}}
39: \newcommand\Pfour{{\PP^4}}
40: \newcommand{\sbase}[6]{\eta^{({#1},{#2},{#3},{#4},{#5},{#6})}}
41: \newcommand{\congr}[3]{{#1} \equiv {#2}\ (\mathrm{mod}\ {#3})}
42: \newcommand\ep{\epsilon}
43: \newcommand\lle{{\ll_\ep}}
44: \newcommand\NU{{N(\So,-K_S,B)}}
45: \newcommand\Res{\mathrm{Res}}
46: \newcommand\prim{\mathrm{prim}}
47: \newcommand{\Aone}{{\mathbf A}_1}
48: \newcommand{\Atwo}{{\mathbf A}_2}
49: \newcommand{\Athree}{{\mathbf A}_3}
50: \newcommand{\Afour}{{\mathbf A}_4}
51: \newcommand{\Afive}{{\mathbf A}_5}
52: \newcommand{\Dfour}{{\mathbf D}_4}
53: \newcommand{\Dfive}{{\mathbf D}_5}
54: \newcommand{\Esix}{{\mathbf E}_6}
55: \newcommand\aone{\alpha_1}
56: \newcommand\atwo{\alpha_2}
57: \newcommand\athr{\alpha_3}
58: \newcommand{\Fone}{\eta_2}
59: \newcommand{\Ftwo}{\eta_3\eta_5^2}
60: \newcommand{\Fthr}{\eta_4\eta_6^2}
61: \newcommand{\Tone}{\aone^2\Fone}
62: \newcommand{\Ttwo}{\atwo\Ftwo}
63: \newcommand{\Tthr}{\athr\Fthr}
64: \newcommand{\Cone}{\eta_1\eta_3\eta_4\eta_5\eta_6}
65: \newcommand{\Ctwo}{\eta_1\eta_2\eta_3\eta_4\eta_6}
66: \newcommand{\Cthr}{\eta_1\eta_2\eta_3\eta_4\eta_5}
67: \newcommand{\tS}{{\widetilde S}}
68: \newcommand{\So}{{S^\circ}}
69: \newcommand{\anti}{-K_\tS}
70: \renewcommand{\le}{\leqslant}
71: \renewcommand{\ge}{\geqslant}
72: \newcommand{\ee}{\boldsymbol{\eta}}
73: \renewcommand{\aa}{\boldsymbol{\alpha}}
74: \newcommand{\ai}{\alpha_i}
75: \newcommand{\nd}{\nmid}
76: \newcommand{\restr}{\text{restriction}}
77: \newcommand{\autom}{\text{automatically}}
78: \newcommand{\allow}{\text{allowed}}
79: \newcommand{\qa}[1]{q_{#1,1}}
80: \newcommand{\qb}[1]{q_{#1,2}}
81: \newcommand{\qc}[1]{q_{#1,3}}
82: 
83: \DeclareMathOperator{\NS}{NS}
84: \DeclareMathOperator{\Pic}{Pic}
85: \DeclareMathOperator{\Spec}{Spec}
86: \DeclareMathOperator{\rk}{rk}
87: \DeclareMathOperator{\additiv}{a}
88: \newcommand\Ga{\mathbb{G}_{\additiv}}
89: \DeclareMathOperator{\multiplikativ}{m}
90: \newcommand\Gm{\mathbb{G}_{\multiplikativ}}
91: \newcommand\Gam{\Ga \times \Gm}
92: \newcommand\rto{\dashrightarrow}
93: 
94: \newtheorem{theorem}{Theorem}
95: \newtheorem{lemma}[theorem]{Lemma}
96: \newtheorem{proposition}[theorem]{Proposition}
97: \newtheorem{conj}[theorem]{Conjecture}
98: \theoremstyle{definition}
99: \newtheorem{remark}[theorem]{Remark}
100: \newtheorem{example}[theorem]{Example}
101: 
102: \numberwithin{equation}{section}
103: 
104: \begin{document}
105: 
106: \title[Torsors and rational points]{Universal torsors over Del Pezzo surfaces and rational points}
107: \author{Ulrich Derenthal}
108: \address{Mathematisches Institut\\Universit\"at G\"ottingen\\
109:   Bunsenstr. 3-5\\37073 G\"ottingen\\Germany}
110: \email{derentha@math.uni-goettingen.de}
111: \author{Yuri Tschinkel}
112: \address{Mathematisches Institut\\Universit\"at G\"ottingen\\
113:   Bunsenstr. 3-5\\37073 G\"ottingen\\Germany}
114: \curraddr{Courant Institute of Mathematical Sciences\\New York University\\
115:   251 Mercer St.\\New York, NY 10012}
116: \email{yuri@uni-math.gwdg.de\\tschinkel@cims.nyu.edu}
117: \date{April 8, 2006}
118: 
119: \begin{abstract}
120:   We discuss Manin's conjecture concerning the distribution of
121:   rational points of bounded height on Del Pezzo surfaces, and its
122:   refinement by Peyre, and explain applications of universal torsors to
123:   counting problems.  To illustrate the method, we provide a proof of
124:   Manin's conjecture for the unique split singular quartic Del Pezzo
125:   surface with a singularity of type $\Dfour$.
126: \end{abstract}
127: 
128: \maketitle
129: 
130: \tableofcontents
131: 
132: \section{Introduction}
133: 
134: Let $f\in \ZZ[x_0,\ldots, x_n]$ be a non-singular form of degree $d$. By the
135: circle method,
136: \[N(f,B):= \#\{ \xx \in \ZZ^{n+1}/\pm \mid \max_j(|x_j|)\le B\} \sim
137: c\cdot B^{n+1-d}\] (where $\xx \in \ZZ^{n+1}/\pm$ means that we
138: identify $\xx$ with $-\xx = (-x_0, \dots, -x_n)$) with $c\in
139: \RR_{>0}$, provided that $n \ge 2^d\cdot(d-1)$, and $f(\xx) = 0$ has
140: solutions over all completions of $\QQ$ (see \cite{MR0150129}).  Let
141: $X=X_f\subset \PP^n$ be a smooth hypersurface over $\QQ$, given by
142: $f(\xx)=0$.  It follows that
143: \begin{equation}\label{eqn:linear}
144: N(X,-K_X,B)\:=\#\{ \xx\in X(\QQ) \mid H_{-K_X}(\xx)\le B\} \sim C\cdot B,
145: \end{equation}
146: as $B\to \infty$. Here $X(\QQ)$ is the set of rational points on $X$,
147: represented by primitive vectors $\xx \in (\ZZ^{n+1}_\prim \setminus
148: 0) / \pm$ (i.e., $\xx = (x_0, \dots, x_n)$ is identified with $-\xx$,
149: and there is no prime dividing all coordinates $x_0, \dots, x_n$), and
150: \begin{equation}\label{eqn:height}
151: H_{-K_X}(\xx):=
152: \max_j(|x_j|)^{n+1-d}, \,\, \text{ for }\,\, \xx=(x_0,\dots, x_n) \in
153: (\ZZ^{n+1}_\prim\setminus 0)/\pm.
154: \end{equation}
155: is the \emph{anticanonical height} of a primitive representative.
156: 
157: 
158: In 1989 Manin initiated a program towards understanding connections
159: between certain geometric invariants of algebraic varieties over
160: number fields and their arithmetic properties, in particular, the
161: distribution of rational points of bounded height, see
162: \cite{MR89m:11060} and \cite{MR1032922}.  The main goal is an
163: extension of the asymptotic formula \eqref{eqn:linear} to other
164: algebraic varieties of \emph{small} degree, called Fano varieties,
165: which are not necessarily isomorphic to hypersurfaces in projective
166: space.  
167: 
168: It became apparent, that in general, to obtain a geometric
169: interpretation of asymptotic results, it may be necessary to restrict
170: to appropriate Zariski open subsets of $X$. Otherwise, the number of
171: rational points on a Zariski closed subset of lower dimension may
172: dominate the total number of rational points; e.g., this phenomenon occurs for the surface \eqref{eq:surface} below where we will restrict
173: to the complement of its lines. Furthermore, we may need to allow
174: finite field extensions: while $X(\QQ)$ might be empty, $X(k)$ could
175: still contain infinitely many points for some number field $k$.
176: 
177: Of particular interest are Del Pezzo surfaces (cf. \cite{MR833513}),
178: e.g., cubic surfaces $S_3\subset \PP^3$ or degree 4 surfaces
179: $S_4:=Q_1\cap Q_2\subset \Pfour$, where $Q_1, Q_2$ are \emph{quadrics}
180: (defined by homogeneous equations of degree 2 in $x_0, \dots, x_4$).
181: Geometrically, smooth Del Pezzo surfaces are obtained by blowing up
182: $\le 8$ \emph{general points}\footnote{no three points on a line, no
183: six points on a curve of degree 2, no eight points with one of them
184: singular on a curve of degree 3} in $\Ptwo$. Blowing up is a standard
185: procedure in algebraic geometry (cf. \cite[Section
186: I.4]{MR0463157}). The blow-up $\pi: S' \to S$ of a surface $S$ at a
187: point $p$ replaces $p$ by a curve $E$ in a particular way. We have $S
188: \setminus\{p\} \cong S' \setminus E$, so $S$ and $S'$ are
189: \emph{birational}.  In our situation, this shows that Del Pezzo
190: surfaces are birational to $\Ptwo$, provided the ground field is
191: algebraically closed.
192: 
193: We can think of \emph{divisors} on blow-ups $S$ of $\Ptwo$ as formal
194: sums of curves on $S$. Considering divisors up to a certain
195: equivalence relation (see \cite[Section II.6]{MR0463157}) leads to the
196: \emph{Picard group} $\Pic(S)$ of divisor classes on $S$. 
197: 
198: For two curves on $S$ which intersect transversally, their
199: \emph{intersection number} is the number of intersection points. As
200: explained in \cite[Section V.1]{MR0463157}, this can be extended to
201: arbitrary divisor classes, defining the non-degenerate
202: \emph{intersection form} $(\cdot, \cdot)$ on $\Pic(S)$. In particular,
203: this defines the \emph{self intersection number} $(E,E)$ of (the class
204: of) a curve $E$. Of special interest are irreducible curves for which
205: this number is negative. We call them \emph{exceptional curves}. For
206: smooth Del Pezzo surfaces of degree 3 and 4, the exceptional curves
207: are exactly the lines (in the standard embedding considered above),
208: having self intersection number $-1$.
209: 
210: The singular Del Pezzo surfaces are obtained as follows: we blow up
211: $\Ptwo$ in special configurations of points (e.g., three points on a
212: line). This results in a smooth surface $\tS$ containing exceptional
213: curves with self intersection number $-2$ (called $(-2)$-curves; we do
214: not permit to blow up points on $(-2)$-curves subsequently).
215: Contracting the $(-2)$-curves gives a singular Del Pezzo surface $S$
216: whose \emph{minimal desingularization} is $\tS$. For the
217: surface~\eqref{eq:surface} below, more details can be found in
218: Section~\ref{sec:geometry}.
219: 
220: For number fields, we say that a Del Pezzo surface is split if all of
221: the exceptional curves are defined over that ground field;
222: there exist \emph{non-split} forms, some of which are not birational
223: to $\Ptwo$ over that ground field.
224: 
225: From now on, we work over $\QQ$. Manin's conjecture
226: in the special case of Del Pezzo surfaces can be formulated as follows.
227: 
228: \begin{conj}\label{conj:manin}
229:   Let $S$ be a Del Pezzo surface with at most rational double
230:   points\footnote{``mild'' singularities which can be resolved by
231:     blow-ups to a curve whose irreducible components are isomorphic to
232:     $\Pone$} over $\QQ$. Then there exists a subset $\So \subset S$
233:   which is dense and open in the Zariski topology such that
234:   \begin{equation}\label{eqn:asym}
235:     N(\So, -K_S,B)\sim c_{S,H}\cdot B(\log B)^{r-1},
236:   \end{equation}
237:   as $B\rightarrow \infty$, where $r$ is the rank of the
238:   Picard group of the minimal desingularization $\tS$ of $S$,
239:   over $\QQ$.
240: \end{conj}
241: 
242: The constant $c_{S,H}$ has been defined by Peyre \cite{MR1340296}; it
243: should be non-zero if $S(\QQ)\neq \emptyset$.  Note that a line
244: defined over $\QQ$ on a Del Pezzo surface such as $S_3$ or $S_4$
245: contributes $\sim c\cdot B^2$ rational points to the counting function
246: (for some positive constant $c$). Thus it is expected that $\So$ is
247: the complement to all lines defined over $\QQ$ (exceptional curves).
248: 
249: 
250: Table~\ref{tab:overview} gives an overview of current results towards
251: Conjecture~\ref{conj:manin} for Del Pezzo surfaces. In Column~4
252: (``type of result''), ``asymptotic'' means that the analog of
253: \eqref{eqn:asym} is established, including the predicted value of the
254: constant; ``bounds'' means that only upper and lower bounds of the
255: expected order of magnitude $B(\log B)^{r-1}$ with unknown constants
256: are proved.
257: 
258: The paper \cite{MR1620682} contains a proof of Manin's conjecture for
259: toric Fano varieties, including all smooth Del Pezzo surfaces of
260: degree $\ge 6$ and the $3\Atwo$ cubic surface\footnote{Singular
261:   Del Pezzo surfaces $S$ will be labeled by the type (in the
262:   ADE-classification) and number of their singularities. The
263:   corresponding Dynkin diagram describes the number and intersection
264:   behaviour of the $(-2)$-curves on $\tS$.}.  This result also covers:
265: \begin{itemize}
266: \item all singular surfaces of degree $\ge 7$ (i.e., $\Aone$ in degree
267:   7 and 8),
268: \item $\Aone$, $2\Aone$, $\Atwo+\Aone$ in degree 6,
269: \item $2\Aone$, $\Atwo+\Aone$ in degree 5,
270: \item $4\Aone$, $\Atwo+2\Aone$, $\Athree+2\Aone$ in degree 4.
271: \end{itemize}
272: 
273: \begin{table}[htbp]\centering
274: \begin{tabular}[ht]{|c|c|c|c|c|}
275:   \hline
276:   degree & singularities & (non-)split & type of result & reference\\
277:   \hline\hline
278:   $\ge 6$ & -- & split & asymptotic & \cite{MR1620682}\\
279:   \hline
280:   5 & -- & split & asymptotic & \cite{MR1909606}\\
281:   5 & -- & non-split & asymptotic & \cite{MR2099200}\\
282:   \hline
283:   4 & $\Dfive$ & split & asymptotic & 
284:   \cite{MR1906155}, \cite{math.NT/0412086}\\
285:   4 & $\Dfour$ & non-split & asymptotic & \cite{math.NT/0502510}\\
286:   4 & $\Dfour$ & split & asymptotic & this paper\\
287:   4 & $3\Aone$ & split & bounds & \cite{math.NT/0511041}\\
288:   \hline
289:   3 & $3\Atwo$ & split & asymptotic & 
290:   \cite{MR1620682}, \cite{MR2000b:11074}, \dots\\
291:   3 & $4\Aone$ & split & bounds & \cite{MR2075628}\\
292:   3 & $\Dfour$ & split & bounds & \cite{math.NT/0404245}\\
293:   3 & $\Esix$ & split & asymptotic & 
294:   \cite{math.NT/0504016}, \cite{math.NT/0509370}\\
295:   \hline
296: \end{tabular}
297: \smallskip
298: \caption{Results for Del Pezzo surfaces}
299: \label{tab:overview}
300: \end{table}
301: 
302: Figure \ref{fig:cayley} shows all points of height $\le 50$ on the
303: Cayley cubic surface (Example~\ref{ex:cayley}), which has four
304: singularities of type $\Aone$ and was considered in \cite{MR2075628}.
305: In Figure \ref{fig:e6}, we see points of height $\le 1000$ on the
306: $\Esix$ cubic surface (\cite{math.NT/0504016} and
307: \cite{math.NT/0509370}).
308: 
309: 
310: \begin{figure}[ht]
311:   \centering
312:   \includegraphics[width=11cm]{qd4_cayley}
313:   \caption{Points of height $\le 50$ on the Cayley cubic surface
314:     $x_0x_1x_2+x_0x_1x_3+x_0x_2x_3+x_1x_2x_3=0$.}
315:   \label{fig:cayley}
316: \end{figure}
317: 
318: \begin{figure}[ht]
319:   \centering
320:   \includegraphics[width=11cm]{qd4_e6}
321:   \caption{Points of height $\le 1000$ on the $\Esix$ singular cubic 
322:     surface $x_1x_2^2+x_2x_0^2+x_3^3=0$ with $x_0,x_2 > 0$.}
323:   \label{fig:e6}
324: \end{figure}
325: 
326: 
327: The proofs of Manin's conjecture proceed either via the height zeta
328: function
329: \[Z(s):=\sum_{\xx\in X^{\circ}(\QQ)} H_{-K_X}(\xx)^{-s},\]
330: whose analytic properties are related to the asymptotic
331: \eqref{eqn:asym} by Tauberian theorems, or via the lifting of the
332: counting problem to the \emph{universal torsor} -- an auxiliary
333: variety parametrizing rational points. The torsor approach has been
334: developed by Colliot-Th\'el\`ene and Sansuc in the context of the
335: Brauer-Manin obstruction \cite{MR89f:11082} and applied to Manin's
336: conjecture by Peyre \cite{MR1679842} and Salberger \cite{MR1679841}.
337: 
338: 
339: 
340: In the simplest case of hypersurfaces $X=X_f\subset \PP^n$ over $\QQ$,
341: with $n\ge 4$, this is exactly the passage from rational vectors
342: $\xx=(x_0,\dots, x_n)$, modulo the diagonal action of $\QQ^*$, to
343: primitive lattice points $(\ZZ_\prim^{n+1}\setminus 0)/\pm$. Geometrically,
344: we have
345: \[\begin{CD}
346:   \AA^{n+1} \setminus 0 @>\Gm>> \PP^n
347: \end{CD}
348: \qquad \text{and} \qquad 
349: \begin{CD}
350:   \TT_X @>\Gm>> X.
351: \end{CD}\] Here, $\TT_X$ is the hypersurface in $\AA^{n+1} \setminus
352: 0$ defined by the form $f$, the 1-dimensional torus $\Gm$ is
353: interpreted as the N\'eron-Severi torus $T_{\NS}$, i.e., an algebraic
354: torus whose characters $\mathfrak X^*(T_{\NS})$ are isomorphic to the
355: Picard group (lattice) of $\PP^n$, resp.  $X$, and the map is the
356: natural quotient by its (diagonal) action.  Rational points on the
357: base are lifted to integral points on the torsor, modulo the action of
358: the group of units $T_{\NS}(\ZZ) =\{\pm 1\}$. The height inequality on
359: the base $H(\xx)\le B$ translates into the usual height inequality on
360: the torsor \eqref{eqn:height}.
361: 
362: In general, a torsor under an algebraic torus $T$ is determined by a
363: homomorpism $\chi\,:\, \mathfrak X^*(T)\rightarrow \Pic(X)$ to the
364: Picard group of the underlying variety $X$; the term \emph{universal}
365: is applied when $\chi$ is an isomorphism.
366: 
367: However, for hypersurfaces in $\Pthree$, or more generally for
368: complete intersection surfaces (i.e., $S$ is the intersection of $k$
369: hypersurfaces in $\PP^{k+2}$), the Picard group may have higher rank.
370: For example, for split smooth cubic surfaces $S=S_3\subset \Pthree$
371: the rank is 7, so that the dimension of the corresponding universal
372: torsor $\TT_S$ is 9; for quartic Del Pezzo surfaces these are 6 and 8,
373: respectively.
374: 
375: It is expected that the passage to universal torsors, which can be
376: considered as natural \emph{descent varieties}, will facilitate the
377: proof of Manin's conjecture (Conjecture~\ref{conj:manin}), at least
378: for Del Pezzo surfaces.  Rational points on $S$ are lifted to certain
379: integral points on $\TT_S$, modulo the action of $T_{\NS}(\ZZ)= (\pm
380: 1)^{r}$, where $r$ is the rank of $\Pic(S)$, and the height inequality
381: on $S$ translates into appropriate inequalities on $\TT_S$.  This
382: explains the interest in the projective geometry of torsors, and
383: expecially, in their equations.  The explicit determination of these
384: equations is an interesting algebro-geometric problem, involving tools
385: from invariant theory and toric geometry.
386: 
387: In this note, we illustrate the torsor approach to asymptotics of
388: rational points in the case of a particular singular surface $S\subset \Pfour$ of
389: degree~4 given by:
390: \begin{equation}\label{eq:surface} 
391: x_0x_3-x_1x_4 = x_0x_1+x_1x_3+x_2^2 = 0.
392: \end{equation}
393: This is a split Del Pezzo surface, with a singularity of
394: type~$\Dfour$.
395: 
396: \begin{theorem}\label{thm:main}
397:   The number of $\QQ$-rational points of anticanonical height bounded
398:   by $B$ on the complement $\So$ of the $\QQ$-rational lines on $S$ as
399:   in \eqref{eq:surface} satisfies
400:   \[\NU = c_{S,H}\cdot B\cdot Q(\log B)+O(B (\log B)^3)
401:   \qquad\text{as $B \to \infty$},\]
402:   where $Q$ is a monic polynomial of degree $5$, and 
403:   \[c_{S,H}= \frac{1}{34560} \cdot \omega_\infty \cdot \prod_p
404: (1-1/p)^6(1+6/p+1/p^2)\] is the constant predicted by Peyre
405: \cite{MR1340296}, with $p$ running through all primes and
406:  \begin{equation*}
407:     \omega_\infty = 3\int\int\int_{\{(t,u,v)\in \RR^3 \mid 0\le v \le 1,
408:       |tv^2|,|v^2u|,|v(tv+u^2)|,|t(tv+u^2)|\le 1\}}1 \dd t \dd u \dd v.\
409:   \end{equation*}
410: \end{theorem}
411: 
412: 
413: In \cite{math.NT/0502510}, Manin's conjecture is
414: proved for a non-split surface with a singularity of the same type.
415: However, these results do not follow from each other.
416: 
417: 
418: 
419: In Section \ref{sec:geometry}, we collect some facts about the
420: geometric structure of $S$. In Section~\ref{sec:conformity}, we
421: calculate the expected value of $c_{S,H}$ and show that Theorem
422: \ref{thm:main} agrees with Manin's conjecture.
423: 
424: In our case, the universal torsor is an affine hypersurface.  In
425: Section~\ref{sec:torsor}, we calculate its equation, stressing the
426: relation with the geometry of $S$.  We make explicit the coprimality
427: and the height conditions.  The method is more systematic than the
428: derivation of torsor equations in \cite{math.NT/0412086} and
429: \cite{math.NT/0509370}, and should bootstrap to more complicated
430: cases, e.g., other split Del Pezzo surfaces.
431: 
432: Note that our method gives coprimality conditions which are different
433: from the ones in \cite{math.NT/0412086} and \cite{math.NT/0509370},
434: but which are in a certain sense more natural: they are related to the
435: set of points on $\TT_S$ which are \emph{stable} with respect to the
436: action of the N\'eron-Severi torus (in the sense of geometric
437: invariant theory, c.f., \cite{MR2004511} and
438: \cite{MR2001i:14059}). Our conditions involve only coprimality of
439: certain pairs of variables, while the other method produces a mix of
440: square-free variables and coprimalities.
441: 
442: 
443: In Section \ref{sec:summations}, we estimate the number of integral
444: points on the universal torsor by iterating summations over the torsor
445: variables and using results of elementary analytic number theory.
446: Finally we arrive at Lemma \ref{lem:sum_eta}, which is very similar in
447: appearance to \cite[Lemma 10]{math.NT/0412086} and \cite[Lemma
448: 12]{math.NT/0504016}.  In Section \ref{sec:proof} we use familiar
449: methods of height zeta functions to derive the exact asymptotic.  We
450: isolate the expected constant $c_{S,H}$ and finish the proof of
451: Theorem \ref{thm:main}.  In Section~\ref{sec:torsorexamples} we write down
452: examples of universal torsors for other Del Pezzo surfaces and discuss
453: their geometry.
454: 
455: \
456: 
457: \noindent{\bf Acknowledgment.}
458: Part of this work was done while the authors were visiting the CRM at
459: the Universit\'e de Montr\'eal during the special year on \emph{Analysis in
460: Number Theory}. We are grateful for the invitation and ideal working
461: conditions.
462: 
463: \section{Geometric background}\label{sec:geometry}
464: 
465: In this section, we collect some geometric facts concerning the
466: surface $S$. We show that Manin's conjecture for $S$ is not a special
467: case of available more general results for Del Pezzo surfaces.
468: 
469: \begin{lemma}\label{lem:geometry}
470:   The surface $S$ has the following properties:
471:   \begin{enumerate}
472:   \item \label{item:geometry_singularity} It has exactly one
473:     singularity of type $\Dfour$ at $q = (0:0:0:0:1)$.
474:   \item \label{item:lines} $S$ contains exactly two lines: \[E_5 =
475:     \{x_0=x_1=x_2=0\}\,\,\text{ and }\,\, E_6 = \{x_1=x_2=x_3=0\},\]
476:     which intersect in $q$.
477:   \item \label{item:geometry_projection} The projection from the line
478:     $E_5$ is a birational map 
479:     \[\begin{array}{cccc}
480:       \phi: & S & \rto & \Ptwo\\
481:       & \xx & \mapsto & (x_0:x_2:x_1)
482:     \end{array}\]
483:     which is defined outside $E_5$. It restricts to an
484:     isomorphism between 
485:     \[\So = S \setminus (E_5 \cup E_6) = \{\xx \in S
486:     \mid x_1 \ne 0\}\,\text{ and }\, \AA^2 \cong \{(t:u:v) \mid v
487:     \ne 0\} \subset \Ptwo,\] whose inverse is the restriction of
488:     \[\begin{array}{cccl}
489:       \psi : & \Ptwo & \rto & S,\\
490:       & (t:u:v) & \mapsto & (tv^2:v^3:v^2u:-v(tv+u^2):-t(tv+u^2))
491:     \end{array}\]
492:     Similar results hold for the projection from $E_6$.
493:   \item \label{item:geometry_dynkin} The process of resolving the
494:     singularity $q$ gives four exceptional curves $E_1, \dots, E_4$
495:     and produces the minimal desingularization $\tS$, which is also
496:     the blow-up of $\Ptwo$ in five points.
497:   \end{enumerate}
498: \end{lemma}
499: 
500: \begin{proof}
501: Direct computations.
502: \end{proof}
503: 
504: It will be important to know the details of the sequence of five
505: blow-ups of $\Ptwo$ giving $\tS$ as in Lemma
506: \ref{lem:geometry}(\ref{item:geometry_dynkin}):
507: 
508: In order to describe the points in $\Ptwo$, we need the lines 
509: \[E_3 = \{v=0\},\qquad A_1 = \{u=0\},\qquad A_2 = \{t=0\}\]
510: and the curve $A_3 = \{tv+u^2=0\}$.
511: 
512: \begin{lemma}\label{lem:blow-ups}
513:   The following five blow-ups of $\Ptwo$ result in $\tS$:
514:   \begin{itemize}
515:   \item Blow up the intersection of $E_3, A_1, A_3$, giving $E_2$.
516:   \item Blow up the intersection of $E_2, E_3, A_3$, giving $E_1$.
517:   \item Blow up the intersection of $E_1$ and $A_3$, giving $E_4$.
518:   \item Blow up the intersection of $E_4$ and $A_3$, giving $E_6$.
519:   \item Blow up the intersection of $E_3$ and $A_2$, giving $E_5$.
520:   \end{itemize}
521:   Here, the order of the first four blow-ups is fixed, and the fifth
522:   blow-up can be done at any time.
523:   
524:   The Dynkin diagram in Figure \ref{fig:dynkin} describes the final
525:   configuration of divisors $E_1, \dots, E_6, A_1, A_2, A_3$.  Here,
526:   $A_1, A_2, A_3$ intersect at one point.
527: \end{lemma}
528: 
529: \begin{figure}[ht]
530:   \centering
531:   \[\xymatrix{A_2  \ar@{-}[r]\ar@{-}[dr] \ar@{-}[dd]& E_5 \ar@{-}[r] 
532:     & E_3 \ar@{-}[dr]\\
533:     & A_1 \ar@{-}[r] & E_2 \ar@{-}[r] & E_1 \\
534:     A_3 \ar@{-}[r] \ar@{-}[ur] & E_6 \ar@{-}[r] & E_4 \ar@{-}[ur]}\]
535:   \caption{Extended Dynkin diagram}
536:   \label{fig:dynkin}
537: \end{figure}
538: 
539: 
540: 
541: The quartic Del Pezzo surface with a singularity of type $\Dfour$ is
542: not toric, and Manin's conjecture does not follow from the results of
543: \cite{MR1620682}.  The $\Dfive$ example of \cite{math.NT/0412086} is
544: an equivariant compactification of $\Ga^2$ (i.e., $S$ has a Zariski open
545: subset isomorphic to $\AA^2$, and the obvious action of $\Ga^2$ on
546: this open subset extends to $S$), and thus a special case of
547: \cite{MR1906155}.
548: 
549: \begin{lemma}\label{lem:compact}
550:   The quartic Del Pezzo surface with a singularity of type $\Dfour$ is a
551:   compactification of $\AA^2$, but not an equivariant compactification of
552:   $\Ga^2$.
553: \end{lemma}
554: 
555: \begin{proof}
556:   We follow the strategy of \cite[Remark 3.3]{MR2029868}.
557:   
558:   Consider the maps $\phi, \psi$ as in Lemma
559:   \ref{lem:geometry}(\ref{item:geometry_projection}). As $\psi$
560:   restricts to an isomorphism between $\AA^2$ and the open set $\So
561:   \subset S$, the surface $S$ is a compactification of $\AA^2$.
562:   
563:   If $S$ were an equivariant compactification of $\Ga^2$ then the
564:   projection $\phi$ from $E_5$ would be a $\Ga^2$-equivariant map,
565:   giving a $\Ga^2$-action on $\Ptwo$.  The line $\{v = 0\}$ would be
566:   invariant under this action. The only such action is the standard
567:   translation action
568:   \[\begin{array}{cccc}
569:     \tau:&\Ptwo &\to &\Ptwo,\\
570:     &(t:u:v) &\mapsto &(t+\alpha v: u+\beta v: v).
571:   \end{array}\]
572:   However, this action does not leave the linear series
573:   \[(tv^2:v^3:v^2u:-v(tv+u^2):-t(tv+u^2))\] invariant, 
574:   which can be seen after calculating
575:   \begin{equation*}
576:     \begin{split}
577:       t(tv+u^2) \mapsto &(t+\alpha v)((t+\alpha v)v+(u+\beta v)^2)\\
578:       = &t(tv+u^2) + 2\beta tuv + (\beta^2+\alpha) tv^2 + \alpha
579:       v(tv+u^2)\\ &+ 2\alpha\beta v^2u +
580:       (\alpha\beta^2+\alpha^2) v^3,
581:     \end{split}
582:   \end{equation*}
583:   since the term $tuv$ does not appear in the original linear series.
584: \end{proof}
585: 
586: \section{Manin's conjecture}\label{sec:conformity}
587: 
588: 
589: 
590: 
591: \begin{lemma}\label{lem:conjecture}
592: Let $S$  be the surface \eqref{eq:surface}. Manin's conjecture
593: for $S$ states that the number of rational points of 
594: height $\le B$ outside the two lines is given by 
595: \[\NU \sim c_{S,H}\cdot B(\log B)^5,\]
596:   where $c_{S,H} = \alpha(S)\cdot \beta(S)\cdot \omega_H(S)$ with 
597:   \begin{equation*}
598:     \begin{split}
599:       \alpha(S) &= (5!\cdot 4\cdot 2\cdot 3 \cdot 3 \cdot 2 \cdot
600:       2)^{-1} = (34560)^{-1}\\
601:       \beta(S) &= 1\\
602:       \omega_H(S) &= \omega_\infty \cdot \prod_p (1-1/p)^6(1+6/p+1/p^2)
603:     \end{split}
604:   \end{equation*}
605:   and
606:   \begin{equation*}
607:     \omega_\infty = 3\int\int\int_{\{(t,u,v)\in \RR^3 \mid 0\le v \le 1,
608:       |tv^2|,|v^2u|,|v(tv+u^2)|,|t(tv+u^2)|\le 1\}}1 \dd t \dd u \dd v\
609:   \end{equation*}
610: \end{lemma}
611: 
612: \begin{proof}
613:   Since $S$ is split over $\QQ$, we have $\rk(\Pic(\tS)) = 6$, and the
614:   expected exponent of $\log B$ is $5$.  Further, $\beta(S) = 1$. The
615:   computation of $c_{S,H}$ is done on the desingularization $\tS$. For
616:   the computation of $\alpha(S)$, observe that the effective cone of
617:   $\tS$ in $\Pic(\tS)$ is simplicial (it is generated by the
618:   exceptional curves $E_1, \dots, E_6$, and their number equals the
619:   rank of $\Pic(\tS)$), and
620:   \[\anti = 4E_1+2E_2+3E_3+3E_4+2E_5+2E_6.\]
621:   The calculation is analog to \cite[Lemma 2]{math.NT/0504016} (see
622:   \cite{del_pezzo_alpha} for its calculation in general).  The
623:   constant $\omega_H(S)$ is computed as in
624:   \cite[Lemma~1]{math.NT/0412086} and \cite[Lemma~2]{math.NT/0504016}.
625: \end{proof}
626: 
627: \section{The universal torsor}\label{sec:torsor}
628: 
629: As explained above, the problem of counting rational points of bounded
630: height on the surface $S$ translates into a counting problem for
631: certain integral points on the universal torsor, subject to
632: coprimality and height inequalities.  In the first part of this
633: section, we describe these conditions in detail.  They are obtained by
634: a process of introducing new variables which are the greatest common
635: divisors of other variables.  Geometrically, this corresponds to the
636: realization of $\tS$ as a blow-up of $\Ptwo$ in five points.
637: 
638: In the second part, we prove our claims.
639: 
640: The universal torsor $\TT_S$ of $S$ is an open subset of the
641: hypersurface in $\AA^9 = \Spec \ZZ[\eta_1, \dots,
642: \eta_6,\alpha_1,\alpha_2,\alpha_3]$ defined by the equation
643: \begin{equation}\label{eq:torsor}
644: T(\ee,\aa) = \Tone + \Ttwo + \Tthr = 0.
645: \end{equation}
646: The projection $\Psi:\TT_S\to S$ is defined by 
647: \begin{equation}\label{eq:projection}
648:   (\Psi^*(x_i)) = (\sbase 2 1 2 1 2 0 \atwo, \sbase 4 2 3 3 2 2,
649:   \sbase 3 2 2 2 1 1 \aone, \sbase 2 1 1 2 0 2 \athr, \atwo\athr),
650: \end{equation}
651: where we use the notation $\sbase {n_1}{n_2}{n_3}{n_4}{n_5}{n_6} =
652: \eta_1^{n_1}\eta_2^{n_2}\eta_3^{n_3}\eta_4^{n_4}\eta_5^{n_5}\eta_6^{n_6}$.
653: 
654: The coprimality conditions can be derived from the extended Dynkin
655: diagram (see Figure~\ref{fig:dynkin}). Two variables are allowed to
656: have a common factor if and only if the corresponding divisors ($E_i$
657: for $\eta_i$ and $A_i$ for $\ai$) intersect (i.e., are connected by an
658: edge in the diagram). Furthermore, $\gcd(\alpha_1,\alpha_2,\alpha_3) >
659: 1$ is allowed (corresponding to the fact that $A_1, A_2, A_3$
660: intersect in one point).
661: 
662: We will show below that there is a bijection between rational points
663: on $\So \subset S$ and integral points on an open subset of $\TT_S$,
664: subject to these coprimality conditions.
665: 
666: We will later refer to
667: \begin{align}
668:   \label{eq:coprim_eta}&\text{coprimality between $\eta_i$ as in Figure~\ref{fig:dynkin},}\\
669:   \label{eq:coprim_a1}&\gcd(\aone,\Cone)=1,\\
670:   \label{eq:coprim_a2}&\gcd(\atwo,\Ctwo)=1,\\
671:   \label{eq:coprim_a3}&\gcd(\athr,\Cthr)=1.
672: \end{align}
673: 
674: To count the number of $\xx \in S(\QQ)$ such that $H(\xx) \le B$, we
675: must lift this condition to the universal torsor, i.e.,
676: $H(\Psi(\ee,\aa)) \le B$. This is the same as \[|\sbase 2 1 2 1 2 0
677: \atwo| \le B, \qquad\dots,\qquad |\atwo\athr|\le B,\] using the five
678: monomials occuring in \eqref{eq:projection}. These have no common
679: factors, provided the coprimality conditions are fulfilled (direct
680: verification).
681: 
682: It will be useful to write the height conditions as follows: let
683: \[X_0 = \left(\frac{\sbase 4 2 3 3 2 2}{B}\right)^{1/3},\, X_1 = (B\sbase {-1}{-2}0 0 1
684: 1)^{1/3},\, X_2 = (B\sbase 2 1 0 3 {-2} 4)^{1/3}.\]
685: Then 
686: \begin{align}
687:   \label{eq:height_eta}&|X_0^3| \le 1\\
688:   \label{eq:height_a1}&|X_0^2 (\alpha_1/X_1)| \le 1\\
689:   \label{eq:height_a2}
690:   \begin{split}
691:     &|X_0^2 (\alpha_2/X_2)| \le 1, \qquad
692:     |X_0(X_0(\alpha_2/X_2)+(\alpha_1/X_1)^2)|\le 1, \qquad\\
693:     &|(\alpha_2/X_2)(X_0(\alpha_2/X_2)+(\alpha_1/X_1)^2)| \le 1
694: \end{split}
695: \end{align}
696: are equivalent to the five height conditions. Here we have used the
697: torsor equation to eliminate $\athr$ because in our counting argument
698: we will also use that $\athr$ is determined by the other variables.
699: 
700: We now prove the above claims.
701: 
702: \begin{lemma}\label{lem:bijection}
703:   The map $\Psi$ gives a bijection between the set of points $\xx$ of
704:   $\So(\QQ)$ such that $H(\xx) \le B$ and the set 
705:   \[\TT_1 := \Bigg\{(\ee,\aa)\in \ZZp^6\times\ZZ^3 \Bigg|
706:   \begin{aligned}&\text{equation} ~\eqref{eq:torsor}, \\ 
707:                  &\text{coprimality} ~ 
708:       \eqref{eq:coprim_eta}, \eqref{eq:coprim_a1},
709:       \eqref{eq:coprim_a2}, \eqref{eq:coprim_a3},\\
710:     &\text{inequalities \eqref{eq:height_eta}, \eqref{eq:height_a1},
711:       \eqref{eq:coprim_a3} hold}
712: \end{aligned}
713: \Bigg\}\]
714: \end{lemma}
715: 
716: \begin{proof}
717:   The map $\psi$ of
718:   Lemma~\ref{lem:geometry}(\ref{item:geometry_projection}) induces a
719:   bijection \[\psi_0: (\eta_3, \aone, \atwo) \mapsto (\eta_3^2\atwo,
720:   \eta_3^3, \eta_3^2\aone, \eta_3\athr, \atwo\athr),\] where $\athr
721:   :=-(\eta_3\atwo+\aone^2)$, i.e., \[T_0 := \aone^2+\eta_3\atwo+\athr =
722:   0,\] between
723:   \[\{(\eta_3,\aone, \atwo) \in \ZZp \times \ZZ^2 \mid \gcd(\eta_3,
724:   \aone, \atwo) = 1\} 
725: \,\, \text{ and }\,\, \So(\QQ) \subset S(\QQ).
726: \]
727:   The height function on $\So(\QQ)$ is given by\[H(\psi_0(\eta_3, \aone,
728:   \atwo)) = \frac{\max(|\eta_3^2\atwo|, |\eta_3^3|, |\eta_3^2\aone|,
729:     |\eta_3\athr|, |\atwo\athr|)}{\gcd(\eta_3^2\atwo, \eta_3^3,
730:     \eta_3^2\aone, \eta_3\athr, \atwo\athr)}.\]
731:   
732:   The derivation of the torsor equation from the map $\psi_0$ together
733:   with the coprimality conditions and the lifted height function is
734:   parallel to the blow-up process described in
735:   Lemma~\ref{lem:blow-ups}. More precisely, each line $E_3, A_1, A_2$
736:   in $\Ptwo$ corresponds to a coordinate function $\eta_3, \aone,
737:   \atwo$ vanishing in one of the lines; the blow-up of the
738:   intersection of two divisors gives an exceptional curve $E_i$,
739:   corresponding to the introduction of a new variable $\eta_i$ as the
740:   greatest common divisor of two old variables. Two divisors are
741:   disjoint if and only if the corresponding variables are coprime.
742:   This is summarized in Table~\ref{tab:dictionary}.
743:   \begin{table}[htbp]\centering
744:     \begin{tabular}{|l||l|}
745:       \hline
746:       Variables, Equations & Geometry \\
747:       \hline\hline
748:       variables & divisors\\
749:       initial variables & coordinate lines\\
750:       $\eta_3,\aone, \atwo$ & $E_3,A_1, A_2$\\
751:       \hline
752:       taking $\gcd$ of two variables & blowing up intersection of divisors\\
753:       new $\gcd$-variable & exceptional curve\\
754:       $\eta_2, \eta_1, \eta_4, \eta_6, \eta_5$ & $E_2, E_1, E_4, E_6, E_5$\\
755:       \hline
756:       extra variable & extra curve\\
757:       $\athr$ & $A_3$\\
758:       starting relation & starting description\\
759:       $\athr=-(\eta_3\atwo+\aone^2)$ & $A_3 = \{\eta_3\atwo+\aone^2=0\}$\\
760:       final relation & torsor equation\\
761:       $\Tthr = -(\Ttwo+\Tone)$ & $\Tone+\Ttwo+\Tthr=0$\\
762:       \hline
763:     \end{tabular}
764:     \smallskip
765:     \caption{Dictionary between $\gcd$-process and blow-ups}
766:     \label{tab:dictionary}
767:   \end{table}
768:   
769:   This plan will now be implemented in five steps; at each step, the
770:   map \[\psi_i: \ZZp^{i+1} \times \ZZ^3 \to \So(\QQ)\] gives a bijection
771:   between:
772:   \begin{itemize}
773:   \item the set of all $(\eta_j, \aone, \atwo, \athr) \in \ZZp^{i+}
774:     \times \ZZ^3$ satisfying certain coprimality conditions (described
775:     by the extended Dynkin diagram corresponding to the $i$-th blow-up
776:     of Lemma~\ref{lem:blow-ups}), an equation $T_i$,
777:     \[H(\psi_i(\eta_j, \alpha_j)) = \frac{\max_k(|\psi_i(\eta_j,
778:       \alpha_j)_k|)}{\gcd(\psi_i(\eta_j, \alpha_j)_k)} \le B,\]
779:   \item the set of all $\xx \in \So(\QQ)$ with $H(\xx) \le B$.
780:   \end{itemize}
781:   The steps are as follows:
782:   \begin{enumerate}
783:   \item Let $\eta_2 := \gcd(\eta_3, \aone) \in \ZZp$. Then \[\eta_3 =
784:     \eta_2\eta_3',\qquad \aone = \eta_2\aone',\qquad\text{ with
785:       $\gcd(\eta_3', \aone') = 1$.}\] Since $\eta_2 \mid \athr$, we
786:     can write $\athr = \eta_2\athr'$.  Then $\athr' =
787:     -(\eta_3'\atwo+\eta_2\aone'^2)$. After renaming the variables, we
788:     have \[T_1 = \eta_2\aone^2 + \eta_3\atwo + \athr = 0\] and
789:     \[\psi_1: (\eta_2, \eta_3, \aone, \atwo, \athr) \mapsto 
790:     (\eta_2\eta_3^2\atwo: \eta_2^2\eta_3^3: \eta_2^2\eta_3^2\aone:
791:     \eta_2\eta_3\athr: \atwo\athr).\] Here, we have eliminated the
792:     common factor $\eta_2$ which occured in all five components of the
793:     image. Below, we repeat the corresponding transformation at each
794:     step.
795:   \item Let $\eta_1 := \gcd(\eta_2, \eta_3) \in \ZZp$. Then \[\eta_2 =
796:     \eta_1\eta_2', \qquad \eta_3 = \eta_1\eta_3',\qquad \text{with
797:       $\gcd(\eta_2', \eta_3')=1$.}\] As $\eta_1 \mid \athr$, we write
798:     $\athr = \eta_1\athr'$, and we obtain: \[T_2 = \eta_2\aone^2 +
799:     \eta_3\atwo + \athr = 0\] and
800:     \begin{multline*}
801:     \psi_2: (\eta_1, \eta_2, \eta_3, \aone, \atwo, \athr) 
802:     \mapsto \\
803:     (\eta_1^2\eta_2\eta_3^2\atwo : \eta_1^4\eta_2^2\eta_3^3 :
804:     \eta_1^3\eta_2^2\eta_3^2\aone : \eta_1^2\eta_2\eta_3\athr :
805:     \atwo\athr). 
806:     \end{multline*}
807:   \item Let $\eta_4 := \gcd(\eta_1, \athr) \in \ZZp$. Then \[\eta_1 =
808:     \eta_4\eta_1',\qquad \athr = \eta_4\athr',\qquad \text{with
809:       $\gcd(\eta_1', \athr')=1$.}\] We get after removing $'$ again:
810:     \[T_3 = \eta_2\aone^2 + \eta_3\atwo + \eta_4\athr = 0\] and
811:     \begin{multline*}
812:       \psi_3 : (\eta_1, \eta_2, \eta_3, \eta_4, \aone, \atwo, \athr)
813:       \mapsto\\
814:       (\eta_1^2\eta_2\eta_3^2\eta_4\atwo :
815:       \eta_1^4\eta_2^2\eta_3^3\eta_4^3 :
816:       \eta_1^3\eta_2^2\eta_3^2\eta_4^2\aone :
817:       \eta_1^2\eta_2\eta_3\eta_4^2\athr : \atwo\athr).
818:     \end{multline*}
819:   \item Let $\eta_6 := \gcd(\eta_4, \athr) \in \ZZp$. Then \[\eta_4 =
820:     \eta_6\eta_4',\qquad \athr = \eta_6\athr', \qquad\text{with
821:       $\gcd(\eta_4', \athr') = 1$.}\] We obtain \[T_4 = \eta_2\aone^2 +
822:     \eta_3\atwo + \eta_4\eta_6^2\athr = 0\] and
823:     \begin{multline*}
824:       \psi_4 :(\eta_1, \eta_2, \eta_3, \eta_4, \eta_6, 
825:       \aone, \atwo, \athr) \mapsto \\
826:       (\eta_1^2\eta_2\eta_3^2\eta_4\atwo :
827:       \eta_1^4\eta_2^2\eta_3^3\eta_4^3\eta_6^2 :
828:       \eta_1^3\eta_2^2\eta_3^2\eta_4^2\eta_6\aone :
829:       \eta_1^2\eta_2\eta_3\eta_4^2\eta_6^2\athr : \atwo\athr).
830:     \end{multline*}
831:   \item The final step is $\eta_5 := \gcd(\eta_3, \atwo) \in \ZZp$, we 
832:     could have done it earlier (just as the blow-up of
833:     the intersection of $E_3, A_2$ in Lemma \eqref{eq:projection}). Then
834:     \[\eta_3 = \eta_5\eta_3',\qquad \atwo = \eta_5\atwo',\qquad\text{with
835:       $\gcd(\eta_3', \atwo')=1.$}\]  We get \[T_5 = \eta_2\aone^2 +
836:     \eta_3\eta_5\atwo + \eta_4\eta_6^2\athr = 0\] and
837:     \begin{multline*}
838:       \psi_5 :(\eta_1, \eta_2, \eta_3, \eta_4, \eta_5, \eta_6, \aone,
839:       \atwo, \athr) \mapsto\\
840:       (\eta_1^2\eta_2\eta_3^2\eta_4\eta_5^2\atwo :
841:       \eta_1^4\eta_2^2\eta_3^3\eta_4^3\eta_5^2\eta_6^2 :
842:       \eta_1^3\eta_2^2\eta_3^2\eta_4^2\eta_5\eta_6\aone :
843:       \eta_1^2\eta_2\eta_3\eta_4^2\eta_6^2\athr : \atwo\athr)
844:     \end{multline*}
845:   \end{enumerate}
846:   We observe that at each stage 
847:   the coprimality conditions
848:   correspond to intersection properties of the respective divisors. 
849:   The final result is summarized in
850:   Figure~\ref{fig:dynkin}, which encodes data from \eqref{eq:coprim_eta},
851:   \eqref{eq:coprim_a1}, \eqref{eq:coprim_a2}, \eqref{eq:coprim_a3}.
852:   
853:   Note that $\psi_5$ is  $\Psi$ from \eqref{eq:projection}.
854:   As mentioned above, $\gcd(\psi_5(\eta_j, \alpha_j)_k)$ 
855:   (over all five components of the image) is trivial by the coprimality
856:   conditions of Figure~\ref{fig:dynkin}. Therefore, 
857:   $H(\psi_5(\ee, \aa)) \le B$ is equivalent to \eqref{eq:height_eta},
858:   \eqref{eq:height_a1}, \eqref{eq:height_a2}.
859:   
860:   Finally, $T_5$ is the torsor equation $T$ \eqref{eq:torsor}.
861: \end{proof}
862: 
863: \section{Summations}\label{sec:summations}
864: 
865: In the first step, we estimate the number of $(\alpha_1, \alpha_2,
866: \alpha_3) \in \ZZ^3$ which fulfill the torsor equation $T$
867: \eqref{eq:torsor} and the height and coprimality conditions. For fixed
868: $(\alpha_1, \alpha_2)$, the torsor equation $T$ has a
869: solution $\alpha_3$ if and only if the congruence
870: \[\congr{\Tone+\Ttwo}0\Fthr\] holds and the conditions on the height and
871: coprimalities are fulfilled.
872: 
873: We have already written the height conditions so that they do not
874: depend on $\athr$. For the coprimality, we must ensure that
875: \eqref{eq:coprim_a2} and \eqref{eq:coprim_a3} are fulfilled.
876: 
877: As $\gcd(\Ftwo, \Fthr)=1$, we can find the multiplicative inverse
878: $c_1$ of $\Ftwo$ modulo $\Fthr$, so that
879: \begin{equation}
880: \label{eq:c1c2}
881: c_1\Ftwo = 1+c_2\Fthr
882: \end{equation}
883: for a suitable $c_2$.
884: Choosing 
885: \begin{align}
886:   \label{eq:c1c3a2}\atwo &= c_3\Fthr - c_1\Tone,\\
887:   \label{eq:c2c3a3}\athr &= c_2\Tone-c_3\Ftwo
888: \end{align}
889: gives a solution of \eqref{eq:torsor} for any $c_3 \in \ZZ$.
890: 
891: Without the coprimality conditions, the number of pairs
892: $(\alpha_2,\alpha_3)$ satisfying $T$ and \eqref{eq:height_a2} would
893: differ at most by $O(1)$ from $1/\Fthr$ of the length of the interval
894: described by \eqref{eq:height_a2}.  However, the coprimality
895: conditions \eqref{eq:coprim_a2} and \eqref{eq:coprim_a3} impose
896: further restrictions on the choice of $c_3$.  A slight complication
897: arises from the fact that because of $T$, some of the conditions are
898: fulfilled automatically once $\ee, \aone$ satisfy
899: \eqref{eq:coprim_eta} and \eqref{eq:coprim_a1}.
900: 
901: Conditions~\eqref{eq:coprim_eta} imply that the possibilities for a
902: prime $p$ to divide more than one of the $\eta_i$ are very limited. We
903: distinguish twelve cases, listed in Column~2 of
904: Table~\ref{tab:coprim_alpha}.
905: 
906: \begin{table}[ht]
907:   \centering
908:   \[\begin{array}{|c|c||c|c|c|}
909:     \hline
910:     \text{case} & p\mid \dots & p\mid \alpha_1 & p\mid \alpha_2 & 
911:     p\mid \alpha_3\\
912:     \hline\hline
913:     0 & - & \allow & \allow & \allow\\
914:     i & \eta_1 & \restr & \restr & \restr\\
915:     ii & \eta_2 & \allow & \restr & \autom\\
916:     iii & \eta_3 & \restr & \restr & \autom\\
917:     iv & \eta_4 & \restr & \autom & \restr\\
918:     v & \eta_5 & \restr & \allow & \autom\\
919:     vi & \eta_6 & \restr & \autom & \allow\\
920:     \hline
921:     vii & \eta_1,\eta_2 & \restr & \restr & \autom\\
922:     viii & \eta_1,\eta_3 & \restr & \restr & \autom\\
923:     ix & \eta_1,\eta_4 & \restr & \autom & \restr\\
924:     x & \eta_3,\eta_5 & \restr & \restr & \autom\\
925:     xi & \eta_4,\eta_6 & \restr & \autom & \restr\\
926:     \hline
927:   \end{array}\]
928:   \smallskip
929:   \caption{Coprimality conditions}
930:   \label{tab:coprim_alpha}
931: \end{table}
932: 
933: In Columns 4 and 5, we have denoted the relevant information for the
934: divisibility of $\atwo, \athr$ by primes $p$ which are divisors of the
935: $\eta_i$ in Column 2, but of no other $\eta_j$:
936: \begin{itemize}
937: \item ``allowed'' means that $\ai$ may be divisible by $p$.
938: \item ``automatically'' means that the conditions on the $\eta_i$ and
939:   the other $\alpha_j$ imply that $p \nd \ai$.  These
940:   two cases do not impose conditions on $c_3$ modulo $p$.
941: \item ``restriction'' means that $c_3$ is not allowed to be in a
942:   certain congruence class modulo $p$ in order to fulfill the
943:   condition that $p$ must not divide $\ai$.
944: \end{itemize}
945: 
946: The information in the table is derived as follows:
947: \begin{itemize}
948: \item If $p\mid \eta_3$, then $p \nd c_2$ from \eqref{eq:c1c2}, and $p
949:   \nd \alpha_1\eta_2$ because of \eqref{eq:coprim_eta},
950:   \eqref{eq:coprim_a1}, so by \eqref{eq:c2c3a3}, $p \nd \athr$
951:   independently of the choice of $c_3$. Since $p \nd \Fthr$, we see
952:   from \eqref{eq:c1c3a2} that $p \mid \atwo$ for one in $p$ subsequent
953:   choices of $c_3$ which we must therefore exclude. This explains
954:   cases $iii$ and $viii$.
955: \item In case $vii$, the same is true for $\atwo$. More precisely, we
956:   see that we must exclude $\congr{c_3}0 p$. By \eqref{eq:c2c3a3}, $p
957:   \nd c_3$ implies that $p\nd \athr$, so we do not need another
958:   condition on $c_3$.
959: \item In case $i$, we see that $p\mid \atwo$ for one in $p$ subsequent
960:   choices of $c_3$, and the same holds for $\athr$. However, in this
961:   case, $p$ cannot divide $\atwo, \athr$ for the same choice of $c_3$,
962:   as we can see by considering $T$: since $p \nd \Tone$, it is
963:   impossible that $p\mid \atwo,\athr$. Therefore, we must exclude two
964:   out of $p$ subsequent choices of $p$ in order to fulfill $p \nd
965:   \atwo,\athr$.
966: \item In the other cases, the arguments are similar.
967: \end{itemize}
968: 
969: The number of $(\atwo, \athr) \in \ZZ^2$ subject to $T$,
970: \eqref{eq:coprim_a2}, \eqref{eq:coprim_a3}, \eqref{eq:height_a2}
971: equals the number of $c_3$ such that $\atwo, \athr$ as in
972: \eqref{eq:c1c3a2}, \eqref{eq:c2c3a3} satisfy these conditions. This
973: can be estimated as $1/\Fthr$ of the interval described by
974: \eqref{eq:height_a2}, multiplied by a product of local factors whose
975: value can be read off from Columns 2, 4, 5 of
976: Table~\ref{tab:coprim_alpha}: the divisibility properties of $\eta_i$
977: by $p$ determine whether zero, one or two out of $p$ subsequent values
978: of $c_3$ have to be excluded. Different primes can be considered
979: separately, and we define
980: \[\vartheta_{1,p} := 
981: \begin{cases}
982:   1-2/p, &\text{case $i$,} \\
983:   1-1/p, &\text{cases $ii-iv, vi-xi$,}\\
984:   1, &\text{case $0,v$.}
985: \end{cases}
986: \]
987: Let \[\vartheta_1(\ee) = \prod_p \vartheta_{1,p}\] be the product of
988: these local factors, and
989: \begin{equation}\label{eq:g_1}
990: g_1(u,v) = \int_{\{t \in \RR \mid |tv^2|,|t(tv+u^2)|,|v(tv+u^2)|\le 1\}} 
991: 1 \dd t.
992: \end{equation}
993: Let $\omega(n)$ denote the number of primes dividing $n$.
994: 
995: \begin{lemma}\label{lem:sum_a2a3}
996:   For fixed $(\ee, \aone) \in \ZZp^6 \times \ZZ$ as in
997:   \eqref{eq:coprim_eta}, \eqref{eq:coprim_a1}, \eqref{eq:height_eta},
998:   \eqref{eq:height_a1}, the number of $(\atwo,\athr) \in \ZZ^2$
999:   satisfying $T$, \eqref{eq:coprim_a2}, \eqref{eq:coprim_a3},
1000:   \eqref{eq:height_a2} is \[\NN_1(\ee,\aone) =
1001:   \frac{\vartheta_1(\ee)X_2}{\Fthr} g_1(\aone/X_1,X_0) +
1002:   O(2^{\omega(\Ctwo)}).\] The sum of error terms for all possible
1003:   values of $(\ee, \aone)$ is $\ll B(\log B)^3$.
1004: \end{lemma}
1005: 
1006: \begin{proof}
1007:   The number of $c_3$ such that the resulting $\atwo, \athr$ satisfy
1008:   \eqref{eq:height_a2} differs from
1009:   $\frac{X_2}{\Fthr}g_1(\aone/X_1,X_0)$ by at most $O(1)$.
1010:   
1011:   Each $\vartheta_{1,p} \ne 1$ corresponds to a congruence condition
1012:   on $c_3$ imposed by one of the cases $i-iv,vi-xi$. For each
1013:   congruence condition, the actual ratio of allowed $c_3$ can differ
1014:   at most by $O(1)$ from the $\vartheta_{1,p}$. The total number of
1015:   these primes $p$ is \[\omega(\Ctwo) \ll 2^{\omega(\Ctwo)},\] which
1016:   is independent of $\eta_5$ since any prime dividing only $\eta_5$
1017:   contributes a trivial factor (see case $v$).
1018:   
1019:   Using the estimate \eqref{eq:height_a1} for $\aone$ in the first
1020:   step and ignoring \eqref{eq:coprim_eta} \eqref{eq:coprim_a1}, which
1021:   can only increase the error term, we obtain:
1022:   \begin{equation*}
1023:       \sum_{\ee}\sum_{\aone}
1024:       2^{\omega(\Ctwo)} \le \sum_{\ee} \frac {B\cdot 2^{\omega(\Ctwo)}}
1025:       {\sbase 3 2 2 2 1 1} \ll B(\log B)^3.
1026:   \end{equation*}
1027:   Here, we use $2^{\omega(n)} \lle n^\ep$ for the summations over
1028:   $\eta_1, \eta_2, \eta_3, \eta_4$. For $\eta_6$, we employ
1029:   \[\sum_{n\le x} 2^{\omega(n)} \ll x (\log x)\] together with partial
1030:   summation, contributing a factor $(\log B)^2$, while the summation
1031:   over $\eta_5$ gives another factor $\log B$.
1032: \end{proof}
1033: 
1034: Next, we sum over all $\alpha_1$ subject to the coprimality condition
1035: \eqref{eq:coprim_a1} and the height condition \eqref{eq:height_a1}.
1036: Let
1037: \begin{equation}\label{eq:g_2}
1038: g_2(v) = \int_{\{u \in \RR \mid |v^2u| \le 1\}} g_1(u,v) \dd u
1039: \end{equation}
1040: Similar to our discussion for $\atwo, \athr$, the number of possible
1041: values for $\aone$ as in \eqref{eq:height_a1}, while ignoring
1042: \eqref{eq:coprim_a1} for the moment, is $X_1g_2(X_0) + O(1)$.
1043: 
1044: None of the coprimality conditions are fulfilled automatically, and
1045: only common factors with $\eta_2$ are allowed (see Column 3 of
1046: Table~\ref{tab:coprim_alpha}). Therefore, each prime factor of $\Cone$
1047: reduces the number of allowed $\aone$ by a factor of $\vartheta_{2,p}
1048: = 1-1/p$ with an error of at most $O(1)$. For all other primes $p$,
1049: let $\vartheta_{2,p} = 1$, and let
1050: \[\vartheta_2(\ee) = \prod_p \vartheta_{2,p}\qquad\text{and}\qquad 
1051: \vartheta(\ee) =
1052: \begin{cases}
1053:   \vartheta_1(\ee)\cdot\vartheta_2(\ee), 
1054:   &\text{\eqref{eq:coprim_eta} holds}\\
1055:   0, &\text{otherwise.}
1056: \end{cases}\]
1057: 
1058: \begin{lemma}\label{lem:sum_a1}
1059:   For fixed $\eta \in \ZZp^6$ as in \eqref{eq:coprim_eta},
1060:   \eqref{eq:height_eta}, the sum of $\NN_1(\ee, \aone)$ over all
1061:   $\aone \in \ZZ$ satisfying \eqref{eq:coprim_a1}, \eqref{eq:height_a1}
1062:   is
1063:   \[\NN_2(\ee) := \frac{\vartheta(\ee)X_1X_2}{\Fthr} g_2(X_0) 
1064:   + \R_2(\ee),\] where the sum of error terms $\R_2(\ee)$ over all
1065:   possible $\ee$ is $\ll B\log B$.
1066: \end{lemma}
1067: 
1068: \begin{proof}
1069:   Let \[\NN(b_1,b_2) = \vartheta_1(\ee) \cdot \#\{\aone \in [b_1,b_2]
1070:   \mid \gcd(\aone,\Cone)=1\}.\] Using M\"obius inversion, this is
1071:   estimated as
1072:   \[\NN(b_1,b_2) = \vartheta_1(\ee) \cdot \vartheta_2(\ee) \cdot (b_2-b_1) + 
1073:   \R(b_1,b_2)\] with $\R(b_1,b_2) = O(2^{\omega(\Cone)})$. By partial
1074:   summation, 
1075:   \[\NN_2(\ee) =
1076:   \frac{\vartheta(\ee)X_1X_2}{\Fthr} g_2(X_0) + \R_2(\ee)\] with
1077:   \[\R_2(\ee) = \frac{-X_2}{\Fthr}\int_{\{u \mid |X_0^2u| \le 1\}}
1078:   (D_1g_1)(u,X_0)\R(-X_1/X_0^2,X_1u) \dd u\] where $D_1g_1$ is the
1079:   partial derivative of $g_1$ with respect to the first variable.
1080:   Using the above bound for $\R(b_1,b_2)$, we obtain:
1081:   \[\R_2(\ee) \ll \frac{X_2}{\Fthr}2^{\omega(\Cone)}.\]
1082:   Summing this over all $\ee$ as in \eqref{eq:height_eta} while
1083:   ignoring \eqref{eq:coprim_eta} which can only enlarge the sum, we
1084:   obtain:
1085:   \[\sum_{\ee} \R_2(\ee) \ll \sum_{\ee} \frac{X_2\cdot
1086:     2^{\omega(\Cone)}}{\Fthr X_0^2} = \sum_{\ee} \frac{B\cdot
1087:     2^{\omega(\Cone)}}{\sbase 2 1 2 2 2 2} \ll B\log B\]
1088:   In the first step, we use $X_0 \le 1$.
1089: \end{proof}
1090: 
1091: Let \[\Delta(n) = B^{-2/3}\!\!\!\sum_{\eta_i, \sbase 4 2 3 3 2 2 = n}
1092: \frac{\vartheta(\ee) X_1X_2}{\Fthr} = \!\!\!\sum_{\eta_i,
1093:   \sbase 4 2 3 3 2 2 = n}
1094: \frac{\vartheta(\ee)(\sbase 4 2 3 3 2
1095:   2)^{1/3}}{\sbase 1 1 1 1 1 1}.\]
1096: 
1097: In view of Lemma~\ref{lem:bijection}, the number of rational points of
1098: bounded height on $\So$ can be estimated by summing the result of
1099: Lemma~\ref{lem:sum_a1} over all suitable $\ee$. The error term is the
1100: combination of the error terms in Lemmas~\ref{lem:sum_a2a3} and
1101: \ref{lem:sum_a1}.
1102: 
1103: \begin{lemma}\label{lem:sum_eta}
1104:   We have
1105:   \[\NU = B^{2/3}\sum_{n \le B} \Delta(n)g_2((n/B)^{1/3}) + O(B (\log B)^3).\]
1106: \end{lemma}
1107: 
1108: \section{Completion of the proof}\label{sec:proof}
1109: 
1110: We need an estimate for \[M(t) := \sum_{n \le t} \Delta(t).\] Consider
1111: the Dirichlet series $F(s) := \sum_{n=1}^\infty \Delta(n)n^{-s}$.
1112: Using \[F(s+1/3) = \sum_{\ee}
1113: \frac{\vartheta(\ee)}{\eta_1^{4s+1}\eta_2^{2s+1}\eta_3^{3s+1}
1114:   \eta_4^{3s+1}\eta_5^{2s+1}\eta_6^{2s+1}},\] we write $F(s+1/3) =
1115: \prod_p F_p(s+1/3)$ as its Euler product. To obtain $F_p(s+1/3)$ for a
1116: prime $p$, we need to restrict this sum to the terms in which all
1117: $\eta_i$ are powers of $p$. Note that $\vartheta(\ee)$ is non-zero if
1118: and only if the divisibility of $\eta_i$ by $p$ falls into one of the
1119: twelve cases described in Table~\ref{tab:coprim_alpha}. The value of
1120: $\vartheta(\ee)$ only depends on these cases.
1121: 
1122: Writing $F_p(s+1/3) = \sum_{i=1}^{11}F_{p,i}(s+1/3)$, we have for example:
1123: \begin{align*}
1124:   F_{p,0}(s+1/3) &= 1,\\
1125:   F_{p,1}(s+1/3) &= \sum_{j=1}^\infty
1126:   \frac{(1-1/p)(1-2/p)}{p^{j(4s+1)}}
1127:   = \frac{(1-1/p)(1-2/p)}{p^{4s+1}-1},\\
1128:   F_{p,7}(s+1/3) &= \sum_{j,k=1}^\infty
1129:   \frac{(1-1/p)^2}{p^{j(4s+1)}p^{k(2s+1)}} =
1130:   \frac{(1-1/p)^2}{(p^{4s+1}-1)(p^{2s+1}-1)}.
1131: \end{align*}
1132: The other cases are similiar, giving 
1133: \begin{equation*}
1134:   \begin{split}
1135:     F_p(s+1/3) = &1+\frac{1-1/p}{p^{4s+1}-1}\bigg((1-2/p) +
1136:     \frac{1-1/p}{p^{2s+1}-1} + 2\frac{1-1/p}{p^{3s+1}-1}\bigg)\\
1137:     &+ \frac{1-1/p}{p^{2s+1}-1} + 2\frac{(1-1/p)^2}{p^{3s+1}-1} +
1138:     2\frac{1-1/p}{p^{2s+1}-1} + 2\frac{(1-1/p)^2}{(p^{2s+1}-1)^2}.
1139:   \end{split}
1140: \end{equation*}
1141: 
1142: Defining \[E(s) := \zeta(4s+1)\zeta(3s+1)^2\zeta(2s+1)^3
1143: \qquad\text{and}\qquad G(s) := F(s+1/3)/E(s),\] we see as in
1144: \cite{math.NT/0504016} that the residue of $F(s)t^s/s$ at $s = 1/3$ is
1145: \[\Res(t) = \frac{3 G(0)t^{1/3}Q_1(\log t)}{5! \cdot 4\cdot 2\cdot 3
1146:   \cdot 3 \cdot 2 \cdot 2}\] for a monic $Q_1\in \mathbb R[x]$ of degree
1147: $5$. By Lemma~\ref{lem:conjecture},
1148: $
1149: \alpha(S) = \frac{1}{5!\cdot 4\cdot 2\cdot 3 \cdot 3 \cdot 2
1150: \cdot 2}.
1151: $ 
1152:  By a Tauberian argument
1153: as in \cite[Lemma 13]{math.NT/0504016}:
1154: \begin{lemma}
1155:   $M(t) = \Res(t)+O(t^{1/3-\delta})$ for some $\delta > 0$.
1156: \end{lemma}
1157: 
1158: By partial summation,
1159: \[\sum_{n\le B}\Delta(n)g_2((n/B)^{1/3}) =\alpha(S)
1160: G(0) B^{1/3}Q(\log B) \cdot 3\int_0^1 g_2(v) \dd v +
1161: O(B^{\frac{1}{3}-\delta})\] for a monic polynomial $Q$ of degree 5.
1162: We identify $\omega_H(S)$ from 
1163: \[G(0) = \prod_p \left(1-\frac 1 p\right)^6 \left(1+\frac 6 p + \frac 1
1164:   {p^2}\right),\,\, \text{ and } \omega_\infty = 3 \int_0^1 g_2(v) \dd v. \] 
1165: Together with Lemma~\ref{lem:sum_eta}, this completes the proof of
1166: Theorem~\ref{thm:main}.
1167: 
1168: \section{Equations of universal torsors}\label{sec:torsorexamples}
1169: 
1170: The simplest universal torsors are those which can be realized
1171: as Zariski open subsets of the affine space. 
1172: This happens if and only if the Del Pezzo surface is
1173: toric. 
1174: 
1175: \begin{example}
1176:   There are 20 types of singular Del Pezzo surfaces of degree $d \ge
1177:   3$ whose universal torsor is an open subset of a hypersurface in
1178:   $\AA^{13-d}$. For one example of each type\footnote{For the cubic
1179:     $\Dfour$ case, the universal torsor of a different example is
1180:     calculated in \cite[Section 4]{MR2029868}.}, the equation defining
1181:   the universal torsor is listed in the following table. More details
1182:   can be found in \cite{math.AG/0604194}.
1183:  \[\begin{array}[ht]{|c|c|c|l|}
1184:     \hline
1185:     \text{degree} & \text{singularities} & \text{\# of lines} & 
1186:     \text{defining equation}\\
1187:     \hline\hline
1188:     6 & \Aone & 3 & \eta_2\alpha_1+\eta_3\alpha_2+\eta_4\alpha_3 \\
1189:     6 & \Atwo & 2 & \eta_2\alpha_1^2 +\eta_3\alpha_2+ \eta_4\alpha_3\\
1190:     \hline
1191:     5 & \Aone & 7 & \eta_2\eta_6+\eta_3\eta_7+\eta_4\eta_8 \\
1192:     5 & \Atwo & 4 & 
1193:     \eta_2\eta_5^2\eta_6 + \eta_3\alpha_1 + \eta_4\alpha_2 \\
1194:     5 & \Athree & 2 &
1195:     \eta_1\alpha_1^2 + \eta_3\eta_4^2\alpha_2 + \eta_5\alpha_3 \\
1196:     5 & \Afour & 1 &
1197:     \eta_1^2\eta_2\alpha_1^3 + \eta_4\alpha_2^2 + \eta_5\alpha_3 \\
1198:     \hline
1199:     4 & 3\Aone & 6 & 
1200:     \eta_4\eta_5 + \eta_1\eta_6\eta_7 + \eta_8\eta_9 \\
1201:     4 & \Atwo+\Aone & 6 & 
1202:     \eta_5\eta_7 + \eta_1\eta_3\eta_9^2 + \eta_6\eta_8 \\
1203:     4 & \Athree & 5 & 
1204:     \eta_5\alpha + \eta_1\eta_4^2\eta_7 + \eta_3\eta_6^2\eta_8 \\
1205:     4 & \Athree+\Aone & 3 & 
1206:     \eta_6\alpha_2+\eta_7\alpha_1+\eta_1\eta_3\eta_4^2\eta_5^3\\
1207:     4 & \Afour & 3 &
1208:     \eta_5\alpha_1 + \eta_1\alpha_2^2 + \eta_3\eta_4^2\eta_6^3\eta_7 \\
1209:     4 & \Dfour & 2 & 
1210:     \eta_3\eta_5^2\alpha_2 + \eta_4\eta_6^2\alpha_3 + \eta_2\alpha_1^2\\
1211:     4 & \Dfive & 1 & 
1212:     \eta_3\alpha_1^2 + \eta_2\eta_6^2\alpha_3 + \eta_4\eta_5^2\alpha_2^3\\
1213:     \hline
1214:     3  & \Dfour & 6 & 
1215:     \eta_2\eta_5^2\eta_8 + \eta_3\eta_6^2\eta_9 + \eta_4\eta_7^2\eta_{10}\\
1216:     3 & \Athree + 2\Aone & 5 & 
1217:     \eta_4\eta_6^2\eta_{10} + \eta_1\eta_2\eta_7^2 + \eta_8\eta_9 \\
1218:     3 & 2\Atwo+\Aone & 5 & 
1219:     \eta_3\eta_5\eta_7^2 + \eta_1\eta_6\eta_8 + \eta_9\eta_{10} \\
1220:     3 & \Afour+\Aone & 4 & 
1221:     \eta_1\eta_5\eta_8^2 + \eta_3\eta_4^2\eta_6^3\eta_9 + \eta_7\alpha \\
1222:     3 & \Dfive & 3 & \eta_2\eta_6^2\alpha_2 + 
1223:     \eta_4\eta_5^2\eta_7^3\eta_8 + \eta_3\alpha_1^2 \\
1224:     3 & \Afive+\Aone & 2 & \eta_1^3\eta_2^2\eta_3\eta_7^4\eta_8 + 
1225:     \eta_5\alpha_1^2 + \eta_6\alpha_2 \\
1226:     3 & \Esix & 1 & \eta_4^2\eta_5\eta_7^3\alpha_3 + 
1227:     \eta_2\alpha_2^2 + \eta_1^2\eta_3\alpha_1^3\\
1228:     \hline
1229:   \end{array}\]
1230: \end{example}
1231: 
1232: \begin{example}[Cubic surface with $\Aone+\Athree$ singularities]
1233:   This surface has 7 lines, 4 additional variables correspond to 
1234:   exceptional curves of the desingularization. Its 9-dimensional
1235:   universal torsor is a Zariski open subset of a complete intersection in 
1236:   \[\AA^{11}=\Spec\ZZ[\eta_0, \dots, \eta_3, \mu_0, \dots, \mu_6]\]
1237:   given by
1238:   \[\eta_1\eta_2\mu_1\mu_2+\mu_4\mu_6+\mu_3\mu_5=0\qquad\text{and}\qquad
1239:   \eta_0\eta_1\mu_2^2+\eta_3\mu_5\mu_6+\mu_0\mu_1=0.\]
1240:   See \cite{cox_singular} for more details.
1241: \end{example}
1242: 
1243: There are examples of universal torsors which are not
1244: complete intersections, but have still been successfully used
1245: in the context of Manin's conjecture:
1246: 
1247: \begin{example}[Cayley cubic]\label{ex:cayley}
1248:   The Cayley cubic
1249:   surface \[x_0x_1x_2+x_0x_1x_3+x_0x_2x_3+x_1x_2x_3=0\]
1250:   (Figure~\ref{fig:cayley}) is a split singular cubic surface with
1251:   four singularities $q_1, \dots, q_4$ of type $\Aone$ and nine lines.
1252:   It is the blow-up of $\Ptwo$ in the 6 intersection points of 4 lines
1253:   in general position.
1254:   The universal torsor is an open subvariety of the variety in
1255:   \[\AA^{13} =
1256:   \Spec \ZZ[v_{12},v_{13},v_{14},y_1,y_2,y_3,y_4,z_{12},z_{13},z_{14},
1257:   z_{23},z_{24},z_{25}]\] defined by six equations of the form
1258:   \[z_{ik}z_{il}y_j+z_{jk}z_{jl}y_i = z_{ij}v_{ij}\]
1259:   and three equations of the form
1260:   \[v_{ij}v_{ik} = z_{il}^2y_jy_k-z_{jk}^2y_iy_l,\] where $\{i,j,k,l\}
1261:   = \{1,2,3,4\}$ and \[z_{ij}=z_{ji}, \qquad v_{ij}=v_{ji},
1262:   \qquad\text{and}\qquad v_{ij}=-v_{kl}.\] See \cite{cox_singular} for
1263:   a proof. The variables $y_i$ correspond to the four exceptional
1264:   curves $E_i$ obtained by blowing up $q_i$, $z_{ij}$ correspond to
1265:   the six lines $m_{ij}$ through two of the singularities, and
1266:   $v_{ij}$ correspond to the other three lines $\ell_{ij}$. The first
1267:   six equations can be interpreted in connection with the projection
1268:   from $m_{ij}$, and the other three equations are connected to the
1269:   projection from $\ell_{ij}$.
1270:   
1271:   Upper and lower bounds of the expected order of magnitude have been
1272:   established in \cite{MR2075628}.
1273: \end{example}
1274: 
1275: 
1276: \begin{example}[Smooth degree 5 Del Pezzo surface]
1277:   The blow-up of $\Ptwo$ in \[(1:0:0), \qquad (0:1:0), \qquad 
1278:   (0:0:1), \qquad (1:1:1)\] is a split smooth Del Pezzo surface of degree 5.
1279:   Its universal torsor is an open subset of the variety defined by the
1280:   following five equations in ten variables:
1281:   \begin{align*}
1282:     \lambda_{12}\eta_2 - \lambda_{13}\eta_3 + \lambda_{14}\eta_4 &= 0\\
1283:     \lambda_{12}\eta_1 - \lambda_{23}\eta_3 + \lambda_{24}\eta_4 &= 0\\
1284:     \lambda_{13}\eta_1 - \lambda_{23}\eta_2 + \lambda_{34}\eta_4 &= 0\\
1285:     \lambda_{14}\eta_1 - \lambda_{24}\eta_2 + \lambda_{34}\eta_3 &= 0\\
1286:     \lambda_{12}\lambda_{34} - \lambda_{13}\lambda_{24} +
1287:     \lambda_{14}\lambda_{23} &= 0
1288:   \end{align*}
1289:   The asymptotic formula (\ref{eqn:asym}) has been established in
1290:   \cite{MR1909606}.
1291: \end{example}
1292: 
1293: To illustrate some of the difficulties in proving Conjecture
1294: \ref{conj:manin} for a smooth split cubic surface, we now write down
1295: equations for its universal torsor (up to radical).
1296:   
1297: \begin{example}[Smooth cubic surfaces]
1298:   Let $S$ be the blow-up of $\Ptwo$ in \[(1:0:0), \,\, (0:1:0), \,\,
1299:   (0:0:1),\, \, (1:1:1), \,\, (1:a:b),\, \, (1:c:d),\] in general
1300:   position.  Conjecturally, the universal torsor is an open subset of
1301:   the intersection of 81 quadrics in 27-dimensional space $\Spec
1302:   \ZZ[\eta_i, \mu_{i,j}, \lambda_i]$, where
1303:   \begin{itemize}
1304:   \item $\eta_1, \dots, \eta_6$ correspond to the preimages of the points,
1305:   \item $\mu_{i,j}$ ($i < j \in \{1,\dots, 6\}$) correspond to the 15 lines
1306:     $m_{i,j}$ through two of the points,
1307:   \item $\lambda_1, \dots, \lambda_6$ correspond to the conics $Q_i$
1308:     through five of the six points,
1309:   \end{itemize}
1310:   and relations arise from conic bundle structures on $S$. Batyrev and Popov
1311:   proved that the above variables are indeed generators and that the relations
1312:   give the universal torsor, up to radical \cite{MR2029863}. 
1313: 
1314:   We now write down these equations explicitly (see
1315:   \cite{math.AG/0603111} for more details).  The 81 defining quadrics
1316:   occur in sets of three. These 27 triples correspond to projections
1317:   from the 27 lines on $S$. We use
1318:   \[E:=(b-1)(c-1)-(a-1)(d-1)\,\, \text{ and }\,\, F:=bc-ad\]
1319:   to simplify the equations.
1320: \begin{align*}
1321: \qa{Q_1}&=-\eta_2\mu_{1,2}-\eta_3\mu_{1,3}+\eta_4\mu_{1,4}\\
1322: \qb{Q_1}&=-a\eta_2\mu_{1,2}-b\eta_3\mu_{1,3}+\eta_5\mu_{1,5}\\
1323: \qb{Q_1}&=-c\eta_2\mu_{1,2}-d\eta_3\mu_{1,3}+\eta_6\mu_{1,6}
1324: \displaybreak[0]\\[\baselineskip]
1325: \qa{Q_2}&=\eta_1\mu_{1,2}-\eta_3\mu_{2,3}+\eta_4\mu_{2,4}\\
1326: \qb{Q_2}&=\eta_1\mu_{1,2}-b\eta_3\mu_{2,3}+\eta_5\mu_{2,5}\\
1327: \qc{Q_2}&=\eta_1\mu_{1,2}-d\eta_3\mu_{2,3}+\eta_6\mu_{2,6}
1328: \displaybreak[0]\\[\baselineskip]
1329: \qa{Q_3}&=\eta_1\mu_{1,3}+\eta_2\mu_{2,3}+\eta_4\mu_{3,4}\\
1330: \qb{Q_3}&=\eta_1\mu_{1,3}+a\eta_2\mu_{2,3}+\eta_5\mu_{3,5}\\
1331: \qc{Q_3}&=\eta_1\mu_{1,3}+c\eta_2\mu_{2,3}+\eta_6\mu_{3,6}
1332: \displaybreak[0]\\[\baselineskip]
1333: \qa{Q_4}&=\eta_1\mu_{1,4}+\eta_2\mu_{2,4}+\eta_3\mu_{3,4}\\
1334: \qb{Q_4}&=(1-b)\eta_1\mu_{1,4}+(a-b)\eta_2\mu_{2,4}+\eta_5\mu_{4,5}\\
1335: \qc{Q_4}&=(1-d)\eta_1\mu_{1,4}+(c-d)\eta_2\mu_{2,4}+\eta_6\mu_{4,6}
1336: \displaybreak[0]\\[\baselineskip]
1337: \qa{Q_5}&=1/b\eta_1\mu_{1,5}+a/b\eta_2\mu_{2,5}+\eta_3\mu_{3,5}\\
1338: \qb{Q_5}&=(1-b)/b\eta_1\mu_{1,5}+(a-b)/b\eta_2\mu_{2,5}+\eta_4\mu_{4,5}\\
1339: \qc{Q_5}&=(b-d)/b\eta_1\mu_{1,5}+F/b\eta_2\mu_{2,5}+\eta_6\mu_{5,6}
1340: \displaybreak[0]\\[\baselineskip]
1341: \qa{Q_6}&=1/d\eta_1\mu_{1,6}+c/d\eta_2\mu_{2,6}+\eta_3\mu_{3,6}\\
1342: \qb{Q_6}&=(1-d)/d\eta_1\mu_{1,6}+(c-d)/d\eta_2\mu_{2,6}+\eta_4\mu_{4,6}\\
1343: \qc{Q_6}&=(b-d)/d\eta_1\mu_{1,6}+F/d\eta_2\mu_{2,6}+\eta_5\mu_{5,6}
1344: \displaybreak[0]\\[\baselineskip]
1345: \qa{m_{1,2}}&=\mu_{4,5}\mu_{3,6}-\mu_{3,5}\mu_{4,6}+\mu_{3,4}\mu_{5,6}\\
1346: \qb{m_{1,2}}&=(b-d)\mu_{3,5}\mu_{4,6}+(d-1)\mu_{3,4}\mu_{5,6}+\eta_2\lambda_1\\
1347: \qc{m_{1,2}}&=F\mu_{3,5}\mu_{4,6}+a(d-c)\mu_{3,4}\mu_{5,6}+\eta_1\lambda_2
1348: \displaybreak[0]\\[\baselineskip]
1349: \qa{m_{1,3}}&=\mu_{4,5}\mu_{2,6}-\mu_{2,5}\mu_{4,6}+\mu_{2,4}\mu_{5,6}\\
1350: \qb{m_{1,3}}&=(c-a)\mu_{2,5}\mu_{4,6}+(1-c)\mu_{2,4}\mu_{5,6}+\eta_3\lambda_1\\
1351: \qc{m_{1,3}}&=-F\mu_{2,5}\mu_{4,6}+b(c-d)\mu_{2,4}\mu_{5,6}+\eta_1\lambda_3
1352: \displaybreak[0]\\[\baselineskip]
1353: \qa{m_{2,3}}&=\mu_{4,5}\mu_{1,6}-\mu_{1,5}\mu_{4,6}+\mu_{1,4}\mu_{5,6}\\
1354: \qb{m_{2,3}}&=(a-c)\mu_{1,5}\mu_{4,6}+a(c-1)\mu_{1,4}\mu_{5,6}+\eta_3\lambda_2\\
1355: \qc{m_{2,3}}&=(b-d)\mu_{1,5}\mu_{4,6}+b(d-1)\mu_{1,4}\mu_{5,6}+\eta_2\lambda_3
1356: \displaybreak[0]\\[\baselineskip]
1357: \qa{m_{1,4}}&=\mu_{3,5}\mu_{2,6}-\mu_{2,5}\mu_{3,6}+\mu_{2,3}\mu_{5,6}\\
1358: \qb{m_{1,4}}&=-E\mu_{2,5}\mu_{3,6}+(b-1)(c-1)\mu_{2,3}\mu_{5,6}+\eta_4\lambda_1\\
1359: \qc{m_{1,4}}&=-F\mu_{2,5}\mu_{3,6}+bc\mu_{2,3}\mu_{5,6}+\eta_1\lambda_4
1360: \displaybreak[0]\\[\baselineskip]
1361: \qa{m_{2,4}}&=\mu_{3,5}\mu_{1,6}-\mu_{1,5}\mu_{3,6}+\mu_{1,3}\mu_{5,6}\\
1362: \qb{m_{2,4}}&=E\mu_{1,5}\mu_{3,6}+(a-b)(c-1)\mu_{1,3}\mu_{5,6}+\eta_4\lambda_2\\
1363: \qc{m_{2,4}}&=(b-d)\mu_{1,5}\mu_{3,6}-b\mu_{1,3}\mu_{5,6}+\eta_2\lambda_4
1364: \displaybreak[0]\\[\baselineskip]
1365: \qa{m_{3,4}}&=\mu_{2,5}\mu_{1,6}-\mu_{1,5}\mu_{2,6}+\mu_{1,2}\mu_{5,6}\\
1366: \qb{m_{3,4}}&=-E\mu_{1,5}\mu_{2,6}+(a-b)(1-d)\mu_{1,2}\mu_{5,6}+\eta_4\lambda_3\\
1367: \qc{m_{3,4}}&=(c-a)\mu_{1,5}\mu_{2,6}+a\mu_{1,2}\mu_{5,6}+\eta_3\lambda_4
1368: \displaybreak[0]\\[\baselineskip]
1369: \qa{m_{1,5}}&=\mu_{3,4}\mu_{2,6}-\mu_{2,4}\mu_{3,6}+\mu_{2,3}\mu_{4,6}\\
1370: \qb{m_{1,5}}&=-E\mu_{2,4}\mu_{3,6}+(a-c)(1-b)\mu_{2,3}\mu_{4,6}+\eta_5\lambda_1\\
1371: \qc{m_{1,5}}&=(d-c)\mu_{2,4}\mu_{3,6}+c\mu_{2,3}\mu_{4,6}+\eta_1\lambda_5
1372: \displaybreak[0]\\[\baselineskip]
1373: \qa{m_{2,5}}&=\mu_{3,4}\mu_{1,6}-\mu_{1,4}\mu_{3,6}+\mu_{1,3}\mu_{4,6}\\
1374: \qb{m_{2,5}}&=aE\mu_{1,4}\mu_{3,6}+(a-b)(c-a)\mu_{1,3}\mu_{4,6}+\eta_5\lambda_2\\
1375: \qc{m_{2,5}}&=(1-d)\mu_{1,4}\mu_{3,6}-\mu_{1,3}\mu_{4,6}+\eta_2\lambda_5
1376: \displaybreak[0]\\[\baselineskip]
1377: \qa{m_{3,5}}&=\mu_{2,4}\mu_{1,6}-\mu_{1,4}\mu_{2,6}+\mu_{1,2}\mu_{4,6}\\
1378: \qb{m_{3,5}}&=-bE\mu_{1,4}\mu_{2,6}+(a-b)(b-d)\mu_{1,2}\mu_{4,6}+\eta_5\lambda_3\\
1379: \qc{m_{3,5}}&=(c-1)\mu_{1,4}\mu_{2,6}+\mu_{1,2}\mu_{4,6}+\eta_3\lambda_5
1380: \displaybreak[0]\\[\baselineskip]
1381: \qa{m_{4,5}}&=\mu_{2,3}\mu_{1,6}-\mu_{1,3}\mu_{2,6}+\mu_{1,2}\mu_{3,6}\\
1382: \qb{m_{4,5}}&=b(c-a)\mu_{1,3}\mu_{2,6}+a(b-d)\mu_{1,2}\mu_{3,6}+\eta_5\lambda_4\\
1383: \qc{m_{4,5}}&=(c-1)\mu_{1,3}\mu_{2,6}+(1-d)\mu_{1,2}\mu_{3,6}+\eta_4\lambda_5
1384: \displaybreak[0]\\[\baselineskip]
1385: \qa{m_{1,6}}&=\mu_{3,4}\mu_{2,5}-\mu_{2,4}\mu_{3,5}+\mu_{2,3}\mu_{4,5}\\
1386: \qb{m_{1,6}}&=-E\mu_{2,4}\mu_{3,5}+(a-c)(1-d)\mu_{2,3}\mu_{4,5}+\eta_6\lambda_1\\
1387: \qc{m_{1,6}}&=(b-a)\mu_{2,4}\mu_{3,5}+a\mu_{2,3}\mu_{4,5}+\eta_1\lambda_6
1388: \displaybreak[0]\\[\baselineskip]
1389: \qa{m_{2,6}}&=\mu_{3,4}\mu_{1,5}-\mu_{1,4}\mu_{3,5}+\mu_{1,3}\mu_{4,5}\\
1390: \qb{m_{2,6}}&=cE\mu_{1,4}\mu_{3,5}+(a-c)(d-c)\mu_{1,3}\mu_{4,5}+\eta_6\lambda_2\\
1391: \qc{m_{2,6}}&=(1-b)\mu_{1,4}\mu_{3,5}-\mu_{1,3}\mu_{4,5}+\eta_2\lambda_6
1392: \displaybreak[0]\\[\baselineskip]
1393: \qa{m_{3,6}}&=\mu_{2,4}\mu_{1,5}-\mu_{1,4}\mu_{2,5}+\mu_{1,2}\mu_{4,5}\\
1394: \qb{m_{3,6}}&=-dE\mu_{1,4}\mu_{2,5}+(d-b)(d-c)\mu_{1,2}\mu_{4,5}+\eta_6\lambda_3\\
1395: \qc{m_{3,6}}&=(a-1)\mu_{1,4}\mu_{2,5}+\mu_{1,2}\mu_{4,5}+\eta_3\lambda_6
1396: \displaybreak[0]\\[\baselineskip]
1397: \qa{m_{4,6}}&=\mu_{2,3}\mu_{1,5}-\mu_{1,3}\mu_{2,5}+\mu_{1,2}\mu_{3,5}\\
1398: \qb{m_{4,6}}&=d(c-a)\mu_{1,3}\mu_{2,5}+c(b-d)\mu_{1,2}\mu_{3,5}+\eta_6\lambda_4\\
1399: \qc{m_{4,6}}&=(a-1)\mu_{1,3}\mu_{2,5}+(1-b)\mu_{1,2}\mu_{3,5}+\eta_4\lambda_6
1400: \displaybreak[0]\\[\baselineskip]
1401: \qa{m_{5,6}}&=\mu_{2,3}\mu_{1,4}-\mu_{1,3}\mu_{2,4}+\mu_{1,2}\mu_{3,4}\\
1402: \qb{m_{5,6}}&=d(c-1)\mu_{1,3}\mu_{2,4}+c(1-d)\mu_{1,2}\mu_{3,4}+\eta_6\lambda_5\\
1403: \qc{m_{5,6}}&=b(a-1)\mu_{1,3}\mu_{2,4}+a(1-b)\mu_{1,2}\mu_{3,4}+\eta_5\lambda_6
1404: \displaybreak[0]\\[\baselineskip]
1405: \qa{E_1}&=(d-b)/E\mu_{1,2}\lambda_2+(c-a)/E\mu_{1,3}\lambda_3+\mu_{1,4}\lambda_4\\
1406: \qb{E_1}&=(d-1)/E\mu_{1,2}\lambda_2+(c-1)/E\mu_{1,3}\lambda_3+\mu_{1,5}\lambda_5\\
1407: \qc{E_1}&=(b-1)/E\mu_{1,2}\lambda_2+(a-1)/E\mu_{1,3}\lambda_3+\mu_{1,6}\lambda_6
1408: \displaybreak[0]\\[\baselineskip]
1409: \qa{E_2}&=F/E\mu_{1,2}\lambda_1+(c-a)/E\mu_{2,3}\lambda_3+\mu_{2,4}\lambda_4\\
1410: \qb{E_2}&=(c-d)/E\mu_{1,2}\lambda_1+(c-1)/E\mu_{2,3}\lambda_3+\mu_{2,5}\lambda_5\\
1411: \qc{E_2}&=(a-b)/E\mu_{1,2}\lambda_1+(a-1)/E\mu_{2,3}\lambda_3+\mu_{2,6}\lambda_6
1412: \displaybreak[0]\\[\baselineskip]
1413: \qa{E_3}&=F/E\mu_{1,3}\lambda_1+(b-d)/E\mu_{2,3}\lambda_2+\mu_{3,4}\lambda_4\\
1414: \qb{E_3}&=(c-d)/E\mu_{1,3}\lambda_1+(1-d)/E\mu_{2,3}\lambda_2+\mu_{3,5}\lambda_5\\
1415: \qc{E_3}&=(a-b)/E\mu_{1,3}\lambda_1+(1-b)/E\mu_{2,3}\lambda_2+\mu_{3,6}\lambda_6
1416: \displaybreak[0]\\[\baselineskip]
1417: \qa{E_4}&=F/(a-c)\mu_{1,4}\lambda_1+(b-d)/(a-c)\mu_{2,4}\lambda_2+\mu_{3,4}\lambda_3\\
1418: \qb{E_4}&=c/(a-c)\mu_{1,4}\lambda_1+1/(a-c)\mu_{2,4}\lambda_2+\mu_{4,5}\lambda_5\\
1419: \qc{E_4}&=a/(a-c)\mu_{1,4}\lambda_1+1/(a-c)\mu_{2,4}\lambda_2+\mu_{4,6}\lambda_6
1420: \displaybreak[0]\\[\baselineskip]
1421: \qa{E_5}&=(d-c)/(c-1)\mu_{1,5}\lambda_1+(d-1)/(c-1)\mu_{2,5}\lambda_2+\mu_{3,5}\lambda_3\\
1422: \qb{E_5}&=-c/(c-1)\mu_{1,5}\lambda_1-1/(c-1)\mu_{2,5}\lambda_2+\mu_{4,5}\lambda_4\\
1423: \qc{E_5}&=-1/(c-1)\mu_{1,5}\lambda_1-1/(c-1)\mu_{2,5}\lambda_2+\mu_{5,6}\lambda_6
1424: \displaybreak[0]\\[\baselineskip]
1425: \qa{E_6}&=(b-a)/(a-1)\mu_{1,6}\lambda_1+(b-1)/(a-1)\mu_{2,6}\lambda_2+\mu_{3,6}\lambda_3\\
1426: \qb{E_6}&=-a/(a-1)\mu_{1,6}\lambda_1-1/(a-1)\mu_{2,6}\lambda_2+\mu_{4,6}\lambda_4\\
1427: \qc{E_6}&=-1/(a-1)\mu_{1,6}\lambda_1-1/(a-1)\mu_{2,6}\lambda_2+\mu_{5,6}\lambda_5
1428: \end{align*}
1429: \end{example}
1430: 
1431: 
1432: \
1433: 
1434: In general, the dimension $k$ of the ambient space $\AA^k$ of the
1435: universal torsor is at least as large as the number of lines on the
1436: surface plus the number of exceptional curves of its
1437: desingularization, while the dimension of the universal torsor only
1438: depends on the degree of the surface, so that the number of equations
1439: must grow with $k$.
1440: 
1441: Heuristically, the complexity of universal torsors should be dictated
1442: by the following considerations:
1443: \begin{itemize}
1444: \item The dimension of the universal torsor of split Del Pezzo
1445:   surfaces $S$ is $12-d$, where $d$ is the degree of $S$.
1446: \item For smooth Del Pezzo surfaces, the number of lines is bigger in
1447:   smaller degrees (e.g., 10 lines in degree 5, and 27 lines in degree
1448:   3).
1449: \item Singular surfaces have less lines than smooth surfaces.
1450: \item The number of lines is higher in cases with ``few mild''
1451:   singularities (e.g., for cubics: $\Aone$ with 21 lines, $\Atwo$ with
1452:   15 lines), while it is low for ``bad'' singularities (e.g., 1 for
1453:   the $\Esix$ cubic, 2 for the $\Afive+\Aone$ cubic).
1454: \end{itemize}
1455: Therefore, we expect universal torsors over surfaces which have
1456: low degree, are smooth or have mild singularities to be more complex
1457: than torsors over surfaces in large degree, or with complicated
1458: singularities.
1459: 
1460: \bibliographystyle{alpha}
1461: 
1462: \bibliography{qd4}
1463: 
1464: \end{document}
1465: 
1466: