1: %%%%%%%%%%%%%%%%%%%%%%
2: %%% Fano 3-folds, K3 surfaces and graded rings
3: %%% Selma Alt{\i}nok, Gavin Brown and Miles Reid
4: %%% Singapore International Symposium in Topology and Geometry
5: %%% (NUS, 2001), Edited by: A. J. Berrick, M. C. Leung and X. W. Xu
6: %%% Contemp. Math. AMS, 2002, to appear, preprint currently 39 pp.
7: %%% ** requires latest version of amsproc.cls **
8: %%%%%%%%%%%%%%%%%%%%%%
9:
10: % \documentclass{conm-p-l}
11: \documentclass{amsproc}
12: % \usepackage{amsmath,amsfonts,amssymb}
13: \usepackage{amssymb}
14:
15: \newcommand{\Alt}[2]{Alt{\i}nok$_#1(#2)$}
16: \newcommand{\rd}[1]{\lfloor #1\rfloor}
17: \newcommand{\half}{\frac12}
18: \newcommand{\recip}[1]{\frac1{#1}}
19: \newcommand{\Span}[1]{\left<#1\right>}
20: \newcommand{\fie}{\varphi}
21: \newcommand{\la}{\lambda}
22: \newcommand{\si}{\sigma}
23: \newcommand{\ep}{\varepsilon}
24: \newcommand{\Ga}{\Gamma}
25: \newcommand{\La}{\Lambda}
26: \newcommand{\Si}{\Sigma}
27: \newcommand{\bmu}{\mathbf{\mu}}
28: \newcommand{\LK}{\La_{\mathrm{K3}}} % the K3 lattice
29: \newcommand{\C}{\mathbb C}
30: \newcommand{\FF}{\mathbb F}
31: \newcommand{\PP}{\mathbb P}
32: \newcommand{\Q}{\mathbb Q}
33: \newcommand{\Z}{\mathbb Z}
34: \newcommand{\Oh}{\mathcal O}
35: \newcommand{\sB}{\mathcal B}
36: \newcommand{\sC}{\mathcal C}
37: \newcommand{\sE}{\mathcal E}
38: \newcommand{\sL}{\mathcal L}
39: \newcommand{\broken}{\dasharrow}
40: % \newcommand{\ot}{\leftarrow}
41: \newcommand{\divides}{\mid}
42: \newcommand{\iso}{\cong}
43: \newcommand{\into}{\hookrightarrow}
44: \newcommand{\onto}{\twoheadrightarrow}
45: % \newcommand{\osum}{\bigoplus}
46: \DeclareMathOperator{\codim}{codim}
47: \DeclareMathOperator{\divi}{div}
48: \DeclareMathOperator{\hcf}{hcf}
49: \DeclareMathOperator{\rank}{rank}
50: \DeclareMathOperator{\wt}{wt}
51: \DeclareMathOperator{\Cl}{Cl}
52: \DeclareMathOperator{\Grass}{Grass}
53: \DeclareMathOperator{\Pf}{Pf}
54: \DeclareMathOperator{\Pic}{Pic}
55: \DeclareMathOperator{\Proj}{Proj}
56: \DeclareMathOperator{\Spec}{Spec}
57: \DeclareMathOperator{\OGr}{OGr}
58: \DeclareMathOperator{\wGr}{wGr}
59:
60: \newcommand{\Cbar}{\overline C}
61: \newcommand{\FFbar}{\overline{\mathbb F}}
62: \newcommand{\wave}{\widetilde}
63: \newcommand{\Case}[1]{\paragraph{\sc Case #1}}
64: \newcommand{\rest}[1]{{}_{{\textstyle|}#1}}
65:
66: \theoremstyle{plain}
67: \newtheorem{thm}[subsection]{Theorem}
68: \newtheorem{thm-dfn}[subsection]{Theorem-Definition}
69:
70: \theoremstyle{definition}
71: \newtheorem{dfn}[subsection]{Definition}
72: \newtheorem{exa}[subsection]{Example}
73: \newtheorem{exc}[subsection]{Exercise}
74: % \newtheorem{prb}[subsection]{Problem}
75:
76: \theoremstyle{remark}
77: \newtheorem{rmk}[subsection]{Remark}
78: \newtheorem{rmks}[subsection]{Remarks}
79:
80: % \newenvironment{pf}{\paragraph{Proof}}{\par\medskip}
81:
82: \numberwithin{equation}{subsection}
83: \numberwithin{figure}{subsection}
84:
85: \title{Fano 3-folds, K3 surfaces and graded rings}
86: \author{Selma Alt{\i}nok}
87: \address{Selma Alt{\i}nok, Adnan Menderes University, Art and Science
88: Faculty, Department of Mathematics, Aydin 09010, Turkey}
89: \email{saltinok43@hotmail.com}
90: \author{Gavin Brown}
91: \address{Gavin Brown, Math Inst., Univ.\ of Warwick, Coventry CV4 7AL,
92: England}
93: \email{gavinb@maths.warwick.ac.uk}
94: \urladdr{www.maths.warwick.ac.uk/$\!\scriptstyle\sim$gavinb}
95: \author{Miles Reid}
96: \address{Miles Reid, Math Inst., Univ.\ of Warwick, Coventry CV4 7AL,
97: England}
98: \email{miles@maths.warwick.ac.uk}
99: \urladdr{www.maths.warwick.ac.uk/$\!\scriptstyle\sim$miles}
100: \date{Feb 2002}
101: \keywords{Gorenstein ring, weighted projective space, Hilbert series}
102: \subjclass[2000]{Primary: 14J28, 14J30, 14Q15, 16E65}
103: \begin{document}
104:
105: \begin{abstract}
106: Explicit birational geometry of 3-folds represents a second phase of
107: Mori theory, going beyond the foundational work of the 1980s. This
108: paper is a tutorial and colloquial introduction to the explicit
109: classification of Fano \hbox{3-folds} (also known by the older name
110: $\Q$-Fano 3-folds), a subject that we hope is nearing completion.
111: With the intention of remaining accessible to beginners in algebraic
112: geometry, we include examples of elementary calculations of graded
113: rings over curves and K3 surfaces. For us, K3 surfaces have at worst Du Val
114: singularities and are polarised by an ample Weil divisor (you might
115: prefer to call these $\Q$-K3 surfaces); they occur as the general
116: elephant of a Fano 3-fold, but are also interesting in their own right.
117: A second section of the paper runs briefly through the classical theory
118: of nonsingular Fano 3-folds and Mukai's extension to indecomposable
119: Gorenstein Fano \hbox{3-folds}. Ideas sketched out by Takagi at the
120: Singapore conference reduce the study of $\Q$-Fano 3-folds with $g\ge2$
121: (and a suitable assumption on the general elephant) to indecomposable
122: Gorenstein Fano 3-folds together with unprojection data.
123:
124: Much of the information about the anticanonical ring of a Fano
125: \hbox{3-fold} or K3 surface is contained in its Hilbert series. The
126: Hilbert function is determined by orbifold Riemann--Roch (the
127: Lefschetz formula of Atiyah, Singer and Segal, see Reid \cite{YPG});
128: using this, we can treat the Hilbert series as a simple collation of
129: the genus and a basket of cyclic quotient singularities. Many hundreds
130: of families of K3s and Fano 3-folds are known, among them a large
131: number with $g\le0$, and Takagi's methods do not apply to these.
132: However, in many cases, the Hilbert series already gives firm
133: indications of how to construct the variety by biregular or birational
134: methods. A final section of the paper introduces the K3 database in
135: Magma, that manipulates these huge lists without effort.
136: \end{abstract}
137:
138: \maketitle
139: \setcounter{tocdepth}{1}
140: \tableofcontents
141:
142: \section{Introduction}
143: \subsection{The graded ring $R(X,A)$ of a polarised variety}
144: Let $X$ be an irreducible projective variety over $\C$, and $A$ an
145: ample divisor on $X$ (see below for more explanation). For $n\ge0$, we
146: write
147: \[
148: H^0(X,nA) = H^0(X,\Oh_X(nA)) = \sL(nA) =
149: \bigl\{ f \in \C(X) \bigm| \divi f + nA \ge 0 \bigr \}
150: \]
151: for the {\em Riemann--Roch space} (RR space) of $nA$. That is,
152: $H^0(X,nA)$ is the finite dimensional vector space of rational (or
153: meromorphic) functions $f\in\C(X)$ with divisor of poles $\le nA$. Our
154: basic construction is the graded ring
155: \begin{equation}
156: R(X,A) = \bigoplus_{n\ge0} H^0(X,nA),
157: \label{eq!RX}
158: \end{equation}
159: where the product is simply multiplication of rational functions
160: \[
161: H^0(X,nA) \times H^0(X,mA) \to H^0(X,(n+m)A) \quad\hbox{by}\quad
162: (f,g) \mapsto fg.
163: \]
164: This just says that if $f,g$ are rational functions with poles $\le
165: nA,mA$ then $fg$ has poles $\le (n+m)A$.
166:
167: Our special interest is the case when the ring $R(X,A)$ can be
168: described by explicit generators and relations, for example, as a
169: polynomial ring or a polynomial ring divided by a principal ideal
170: (geometrically, a hyper\-surface). When this is possible, it
171: corresponds to embedding $X$ in projective space and determining the
172: defining equations of the image variety. It frequently happens in
173: higher dimensions that the generators $x_i$ of $R(X,A)$ have different
174: weights, so we have to work with weighted projective spaces ({\em
175: w.p.s.})\ and weighted homogeneous ideals. Dolgachev \cite{D} and
176: Fletcher \cite{Fl} are useful as general references on w.p.s.\ and their
177: complete intersections ({\em c.i.}).
178:
179: \begin{rmks}
180: Although we intend to allow $X$ to have mild singularities, please think
181: of it in the first instance as nonsingular. The point of the RR spaces
182: $H^0(X,nA)=\sL(nA)$ is this: because $X$ is projective, the only globally
183: defined regular functions (holomorphic functions) on it are the constants. By
184: allowing poles along a divisor $nA$, we get a finite dimensional space of
185: rational (or meromorphic) functions $H^0(X,nA)$; as we discuss below, the RR
186: theorem predicts $\dim H^0(X,nA)$ in good cases. Choosing a basis
187: $x_0,x_1,\dots,x_k\in H^0(X,nA)$ defines a rational map
188: \[
189: \fie_{nA}\colon X \broken \PP^k \quad\hbox{by}\quad
190: P \mapsto (x_0(P):\cdots:x_k(P));
191: \]
192: there are straightforward criteria that determine when $\fie_{nA}$ is a
193: morphism $X\to\PP^k$ or an embedding $X\into\PP^k$. The divisor $nA$ is
194: {\em very ample} if it defines an embedding $\fie_{nA}\colon X\into \PP^k$,
195: and $A$ is {\em ample} if $nA$ is very ample for some $n$.
196: \end{rmks}
197:
198: \subsection{The RR theorem for a curve}
199: Please skip this stuff if you already know it. If not, we strongly advise
200: you to make an effort to {\em commit it to memory}, since it is one of the
201: central points of algebraic geometry. A nonsingular projective curve $C$ (or
202: compact Riemann surface) has a {\em genus} $g$ that can be defined in several
203: alternative ways:
204: \begin{enumerate}
205: \renewcommand{\labelenumi}{(\arabic{enumi})}
206: \item $C$ is homeomorphic to the traditional picture of a surface with $g$
207: holes.
208: \item $C$ has Euler number $e(C)=2-2g$.
209: \item The tangent bundle to $C$ has degree $2-2g$, or equivalently, the
210: canonical divisor class $K_C$ has degree $2g-2$. Here $K_C$ is defined as the
211: class of the divisor of a rational (or meromorphic) differential.
212: \item $\dim H^0(C,\Oh_C(K_C))=\dim \sL(K_C)=g$. In words, $g$ equals
213: the dimension of the vector space of global regular (or holomorphic)
214: differentials on $C$.
215: \item etc.; for example, you can define $g$ as the integer appearing in the
216: RR theorem for curves (\ref{eq!RRC}).
217: \end{enumerate}
218:
219: \begin{thm}[RR for curves]
220: Let $D=\sum n_iP_i$ be a divisor on $C$. Then
221: \begin{equation}
222: \dim H^0(D)-\dim H^0(K_C-D)=1-g+\deg D,
223: \label{eq!RRC}
224: \end{equation}
225: where $\deg D=\sum n_i$. (Recall that $H^0(D)=\sL(D)$ and
226: $H^0(K_C-D)=\sL(K_C-D)$, and in this language the formula is
227: \[
228: \dim \sL(D)-\dim \sL(K_C-D)=1-g+\deg D.)
229: \]
230:
231: When $\deg D>2g-2$, we get $H^0(K_C-D)=0$, so (\ref{eq!RRC}) simplifies
232: to a formula for $\dim H^0(D)$ in terms of topological invariants of
233: $C$ and $D$.
234: \end{thm}
235:
236: \begin{exa} \label{exa!g=3}
237: As a baby case of our graded ring methods, we compute the canonical ring
238: $R(C,K_C)$ of a curve of genus~3, and conclude that $C$ is a plane quartic
239: curve under suitable extra assumptions.
240:
241: Take $C$ to be a nonsingular projective curve of genus~3, with polarising
242: divisor $K_C$. The starting point is that RR gives $\dim H^0(nK_C)$:
243: \begin{equation}
244: \dim H^0(nK_C)\ =\
245: \begin{cases}
246: 1 & \hbox{if } n=0, \\
247: g & \hbox{if } n=1, \\
248: (2n-1)(g-1) & \hbox{if } n\ge2.
249: \end{cases}
250: \label{eq!g=3}
251: \end{equation}
252: The last number comes from (\ref{eq!RRC}), with the r.h.s.\ equal to
253: $1-g+n(2g-2)$.
254:
255: Now $\dim H^0(C,K_C)=3$. As we said above, choosing a basis $x_1,x_2,x_3$
256: of $H^0(C,K_C)$ defines a rational map $\fie_{K_C}\colon C\broken\PP^2$.
257: A traditional and easy argument based on RR applied to the divisors $K_C-P$
258: and $K_C-P-Q$ proves that $\fie_{K_C}$ is a morphism, either defining an
259: isomorphism $C\iso C_4\subset\PP^2$ of $C$ with a plane quartic curve $C_4$,
260: or a generically 2-to-1 cover $C\to Q_2\subset\PP^2$ over a conic.
261:
262: We argue somewhat more algebraically on the ring
263: \[
264: R(C,K_C)=\bigoplus_{n\ge0}H^0(C,nK_C)
265: \]
266: using (\ref{eq!g=3}); in degree two, $\dim H^0(C,2K_C)=6$. But we already
267: know 6 elements of $H^0(C,2K_C)$, namely the 6 quadratic monomials
268: \[
269: S^2x_i=\bigl\{x_1^2,x_1x_2,x_2^2,x_1x_3,x_2x_3,x_3^2\bigr\}.
270: \]
271: We assume that these monomials are linearly independent, so form a basis
272: of $H^0(C,2K_C)$. Likewise, in degree 3, $\dim H^0(C,3K_C)=10$, and there
273: are $10=\binom{5}{2}$ cubic monomials $S^3x_i$, so we assume that they
274: are linearly independent, and again form a basis. In degree 4, however,
275: we necessarily find a relation, since $\dim H^0(C,4K_C)=14$, but there are
276: $15=\binom{6}{2}$ quartic monomials $S^4x_i$.
277:
278: The conclusion is the prediction that the graded ring $R(C,K_C)$ has the
279: simplest form $R(C,K_C)=\C[x_1,x_2,x_3]/(f_4)$; and the corresponding
280: map $\fie_{K_C}\colon C\to\PP^2$ is an embedding with image
281: $C_4$ given by $f_4(x_1,x_2,x_3)=0$.
282: \end{exa}
283:
284: \begin{rmk}
285: The assumption that the quadratic monomials $S^2x_i$ are linearly
286: independent is of course the hyperelliptic dichotomy. In the
287: non\-hyper\-elliptic case, the image $\fie_{K_C}(C)$ cannot be contained in a
288: curve of degree $\le3$ union a finite set, so that once the relation $f_4$
289: has been detected, one sees that the ring homomorphism
290: \[
291: \C[x_1,x_2,x_3]/(f_4) \to R(C,K_C)
292: \]
293: must be injective. Then it is surjective, because in degree $n$ the
294: quotient ring has dimension $\binom{n+4}{2}-\binom{n}{2}=\dim H^0(C,nK_C)$.
295:
296: In the hyperelliptic case, the canonical ring is
297: \[
298: \C[x_1,x_2,x_3,y]/(Q_2,F_4),
299: \]
300: where $Q_2(x)=0$ is the equation of the image conic, $y$ the new generator
301: in degree 2 required to compensate for this relation, and $F_4:y^2=f_4(x)$.
302: Then $C\to Q_2$ is the double cover ramified in the 8 points $Q_2=f_4=0$.
303:
304: We can even consider a degenerating family with quadratic relation
305: $\la y=Q_2(x)$ that consists of a nonhyperelliptic curve if $\la\ne0$ and a
306: hyperelliptic curve if $\la=0$.
307: \end{rmk}
308:
309: \begin{exa}
310: Consider now a nonhyperelliptic curve $C$ of genus~6. Its canonical
311: embedding $\fie\colon C_{10}\into\PP^5$ has image of codimension~4. As
312: opposed to the hypersurface case, we do not have any surefire way of
313: predicting the equations of a variety of codimension~$\ge4$. This is a
314: major preoccupation of the rest of the paper. In this case, we happen to
315: be lucky, but it still involves a case division into Brill--Noether
316: special and Brill--Noether general:
317:
318: \Case1 $C$ is trigonal or isomorphic to a plane quintic. Trigonal means
319: that $C$ has a linear system $g^1_3$; equivalently, $C$ can be
320: represented as a triple cover $C\to\PP^1$. It is well known that then
321: $\fie_{K_C}(C)$ is contained in a quartic surface scroll (typically,
322: $\PP^1\times\PP^1$ embedded by $\Oh(1,2)$); its equations are the 6
323: quadrics defining the scroll and 3 cubics defining $C$ inside the
324: scroll. The case of a plane quintic is similar, with $\fie_{K_C}(C)$
325: contained in the Veronese image of $\PP^2$, and defined by the 6
326: quadratic equations of the Veronese surface and 3 cubics defining $C$
327: in it.
328:
329: \Case2 $C$ is not trigonal and not isomorphic to a plane quintic. Then
330: the canonical image $\fie_{K_C}(C)$ is a c.i.\ in $\Grass(2,5)$:
331: \[
332: C\ =\ \Grass(2,5) \cap Q_2 \cap H_1 \cap \cdots \cap H_4 \subset\PP^9.
333: \]
334: \end{exa}
335:
336: \begin{rmks} \label{rmk!Pf}
337: Here $\Grass(2,5)$ is the Grassmann variety of 2-dimensional vector
338: subspaces of $\C^5$ in its Pl\"ucker embedding in
339: $\PP^9=\PP(\bigwedge^2\C^5)$. Equivalently, it is the determinantal
340: variety defined by the $4\times4$ diagonal Pfaffians $\Pf_{ij.kl}$ of
341: the $5\times5$ skew matrix
342: \[
343: \begin{pmatrix}
344: x_{12} & x_{13} & x_{14} & x_{15} \\
345: & x_{23} & x_{24} & x_{25} \\
346: && x_{34} & x_{35} \\
347: &&& x_{45} \\
348: \end{pmatrix},
349: \]
350: where $\pm\Pf_{ij.kl}=x_{ij}x_{kl}-x_{ik}x_{jl}+x_{il}x_{jk}$ for
351: $\{i,j,k,l\} \subset\{1,2,3,4,5\}$. (We only write the upper triangular
352: elements. For the whole $5\times5$ matrix, just write zeros in the
353: 5 diagonal entries and the skew elements $-x_{ji}$.)
354:
355: Although Case~2 is a 19th century result, the direct statement and proof
356: is due to Mukai (\cite{Mu1}, Section~5), and is a first substantial case
357: of his general program. He proves that there exists a unique rank~2
358: stable vector bundle $E$ on $C$ with $\det E=K_C$ and $H^0(C,E)=\C^5$;
359: moreover, $E$ is generated by its $H^0$, and $\bigwedge^2 H^0(C,E)\onto
360: H^0(C,\bigwedge^2E)$. By the universal mapping property of the Grassmann
361: variety, $E$ and its sections define a morphism $\psi_E\colon
362: C\to\Grass(2,5)$, making $\fie_{K_C}$ a linear section of the Pl\"ucker
363: embedding:
364: \[
365: \renewcommand{\arraystretch}{1.6}
366: \begin{array}{ccc}
367: C & \into & \PP^5 \\
368: \big\downarrow && \bigcap \\
369: \Grass(2,5) &\into& \PP^9
370: \end{array}
371: \]
372: Since $g=6\ge2\cdot3$, standard Brill--Noether arguments give the
373: existence of a decomposition $K_C=\xi+\eta$ with divisors $\xi$ and
374: $\eta$ such that $H^0(C,\xi)=2$ and $H^0(C,\eta)=3$. RR (\ref{eq!RRC})
375: gives $\deg\xi=4$, and the case assumptions imply that $|\xi|$ is a free
376: $g^1_4$ and $|\eta|$ a free $g^2_6$. For every such decomposition
377: $K_C=\xi+\eta$, the bundle $E$ appears as the unique extension $\xi\into
378: E\onto\eta$ for which the boundary map $H^0(\eta)\to H^1(\xi)$ is zero.
379: \end{rmks}
380:
381: \begin{exa} \label{exa!1/2}
382: As in Example~\ref{exa!g=3}, let $C$ be a nonsingular quartic curve;
383: however, instead of $K_C$, we choose a point $P\in C$ and consider the
384: {\em fractional divisor} $A=K_C+\half P$. The RR space of $nA$ is given
385: by the same formula:
386: \[
387: H^0(C,nA)=\sL(nA)=\bigl\{f \in\C(C) \bigm| \divi f+nA\ge0 \bigr\}.
388: \]
389: The novel point is that a rational function cannot have a pole of
390: fractional order. Thus although the definition allows $f$ a pole of
391: order $n/2$ at $P$, in practice this restricts it to have pole of
392: order $\rd{\frac n2}$.
393:
394: The first time this has any effect is when $n=2$, when
395: \[
396: H^0(C,2A)=H^0(C,2K_C+P)
397: \]
398: has dimension 1 bigger than $H^0(C,2K_C)$. Thus the ring $R(C,A)$
399: needs a new generator $y$ in degree~2. In
400: explicit terms, if $P=(1,0,0)\in C\subset\PP^2$ then the equation of
401: $C$ is $f_4=x_2A+x_3B=0$ where $A,B$ are cubics in $x_1,x_2,x_3$,
402: and $y=B/x_2=-A/x_3$ is a rational section of $\Oh_C(2)$ with pole
403: at $P$. It follows that
404: \[
405: R(C,A)=\C[x_1,x_2,x_3,y]/(yx_2-B,yx_3+A).
406: \]
407: \end{exa}
408:
409: \begin{rmks} \label{rmk!orb}
410: The ring $\C[x_1,x_2,x_3,y]/(yx_2-B,yx_3+A)$ constructed in
411: Example~\ref{exa!1/2} corresponds to embedding $C\into\PP(1,1,1,2)$ as
412: a $(3,3)$ c.i. Here the w.p.s.\ $\PP(1,1,1,2)$ is the cone over the
413: Veronese surface, and $C$ passes through the cone point $(0,0,0,1)$ as a
414: nonsingular branch, with the equations $yx_2=B,yx_3=-A$ determining
415: $x_2,x_3$ as implicit functions.
416:
417: The fractional divisor $K_C+\frac{r-1}{r}P$ is the {\em orbifold
418: canonical class} of $C$, where $P$ is viewed as an orbifold point of
419: order~$r$. The ring $R(C,A)$ is again Gorenstein (see Watanabe
420: \cite{W}); we treat the case $K_C+\frac{2}{3}P$ in
421: Example~\ref{exa!2/3}. More generally, we could take any {\em basket}
422: of orbifold points $P_i$ of order $r_i$ on $C$ and consider the
423: orbifold canonical class $A=K_C+\sum\frac{r_i-1}{r_i}P_i$. The affine
424: cone $\Spec R(C,A)$ corresponds to a $\C^*$ fibration over $C$ with
425: isotropy $\bmu_{r_i}$ over $P_i$, the $\C^*$ analog of a Seifert fibre
426: space. The fact that $R(C,A)$ is a Gorenstein ring means that there is
427: a category of orbifold sheaves on $C$ (more precisely, modules over its
428: affine cone $\sC C=\Spec R(C,A)$) having a nice form of Serre
429: duality.
430:
431: A curve with an orbifold point is a substantial first case of the cyclic
432: quotient singularities needed in higher dimension. For example, our
433: construction of $C_{3,3}\subset\PP(1,1,1,2)$ extends in the obvious way
434: to a K3 surface $S_{3,3}\subset\PP(1^4,2)$, Fano 3-fold
435: $V_{3,3}\subset\PP(1^5,2)$, etc.
436: \end{rmks}
437:
438: \section{Classic Fano 3-folds}
439: The subject starts in the 1930s, when Fano studies projective 3-folds
440: having canonical curve sections. His definition is projective:
441: $V=V^3_{2g-2}\subset\PP^{g+1}$ should have canonical curves as its
442: codimension~2 linear sections:
443: \[
444: V\cap H_1\cap H_2 \ = \ C_{2g-2}\subset\PP^{g-1},
445: \]
446: but this is more-or-less equivalent to assuming that $-K_V$ is very ample
447: (see below). Iskovskikh modernised Fano's treatment in the 1970s. His
448: starting point is a nonsingular 3-fold $V$ with $-K_V$ ample. He analyses
449: the rather few exceptions to $-K_V$ very ample, and corrects and reworks
450: Fano's classification. {From} the 1980s onwards Mukai discovers a new
451: interpretation of Fano and Iskovskikh's results in terms of linear
452: sections of projective homogeneous varieties, and generalises them to
453: indecomposable Gorenstein Fano 3-folds (see below for explanation).
454:
455: \begin{dfn}
456: A {\em nonsingular Fano $3$-fold\/} is a nonsingular irreducible
457: projective variety of dimension $3$ with ample $-K_V$. Likewise, a {\em
458: Gorenstein Fano $3$-fold\/} is a normal irreducible projective $3$-fold
459: $V$ with at worst canonical Gorenstein singularities and $-K_V$ ample.
460: The {\em genus} of a Fano 3-fold is the integer $g$ defined by $\dim
461: H^0(V,-K_V)=g+2$. In the nonsingular case, we deduce that
462: $2g-2=({-}K_V)^3$.
463:
464: We say that $V$ is {\em prime} if $\Cl V=\Z\cdot({-}K_V)$ (here $\Cl
465: V$ is the Weil divisor class group); and $V$ is {\em indecomposable} if
466: there does not exist any decomposition $-K_V=A+B$ where $A,B$ are Weil
467: divisors such that $|A|,|B|$ are nontrivial linear systems.
468: \end{dfn}
469:
470: The point of these definitions is to exclude easy cases such as $\PP^3$
471: (for which $-K=\Oh(4)$) and quadric or cubic hypersurfaces $Q_2,
472: F_3\subset\PP^4$ (for which $-K=\Oh(3)$ or $\Oh(2)$ respectively), that
473: can be handled by simpler methods. The prime condition is Fano and
474: Iskovskikh's assumption that $V$ is factorial and has rank
475: $\rho(V)=\rank\Pic V=1$ and index~1; indecomposable is Mukai's
476: generalisation. Nonsingular Fano 3-folds with $\rho\ge2$ were treated
477: in detail by Mori and Mukai \cite{MM1}--\cite{MM2}.
478:
479: \begin{exa}\label{exa!V4}
480: A quartic $3$-fold containing a plane $\Pi\subset V_4\subset\PP^4$ is
481: indecomposable. If $\Pi$ is the $x_3,x_4,x_5$ coordinate plane defined
482: by $x_1=x_2=0$ then $V$ is defined by an equation $x_1A+x_2B=0$, and
483: this is singular at the points $\Pi\cap(A=B=0)$ (in general 9 ordinary
484: double points).
485: \end{exa}
486:
487: \subsection{General theory} \label{ssec!ele}
488: An {\em elephant} of $V$ is a surface $S\in|{-}K_V|$ with at worst Du
489: Val singularities (rational double points). If $V$ is a Gorenstein Fano
490: 3-fold, an elephant is known to exist by a theorem of Shokurov and Reid
491: \cite{R1}. It is a K3 surface: in fact $K_S=(K_V+S)\rest S=0$ and
492: $H^1(S,\Oh_S)=0$ by Kodaira vanishing. Morever, $|{-}K_V|\rest S$ is an
493: ample complete linear system of Cartier divisors on $S$.
494:
495: With a couple of exceptional cases that are easily classified and that
496: we pass over, it follows that $|{-}K_V|$ is very ample. Taking another
497: elephant $S'\in|{-}K_V|$ and setting $C=S\cap S'$ gives the diagram
498: \[
499: \renewcommand{\arraystretch}{1.3}
500: \begin{array}{ccll}
501: V&\into&\PP^{g+1} & \hbox{anticanonical embedding $\fie_{{-}K_V}$,}\\
502: \bigcup&&\bigcup \\
503: S&\into&\PP^g & \hbox{embedding by $|{-}K_V|\rest S$,} \\
504: \bigcup&&\bigcup \\
505: C&\into&\PP^{g-1} & \hbox{canonical embedding $\fie_{K_C}$.}
506: \end{array}
507: \]
508: Special linear systems on $C$ interpreted in terms of geometric RR
509: give rise to geometric properties of $V$. The hyperelliptic case was
510: already passed over in what we said above; the case of $C$ trigonal
511: leads to $V$ a hypersurface in a scroll, which contradicts $V$
512: indecomposable for $g\ge5$. In the same way, other Brill--Noether
513: special linear systems on $C$ such as $g^2_5$ (when $C$ is a plane
514: quintic) usually contradict $V$ indecomposable.
515:
516: \subsection{Fano 3-folds, specific theory}\label{ssec:specific}
517: None of the above is specific to 3-folds. Canonical curves and K3 surfaces
518: continue to exist for all $g$, so the main result for 3-folds is thus quite
519: remarkable.
520:
521: \begin{thm}[Fano, Iskovskikh, Mukai] \label{th!clF}
522: Indecomposable Gorenstein Fano \hbox{$3$-folds} of genus $g$
523: exist if and only if $2\le g\le10$ or $g=12$. If $g\le5$, the
524: anticanonical ring $R(V,{-}K_V)$ is a hypersurface or complete
525: intersection in projective space or w.p.s. If $g=6,\dots,10$ then $V$ is
526: a complete intersection in a homogeneous projective variety. (There is
527: also an analogous structure result for
528: $g=12$.)
529: \end{thm}
530:
531: For example,
532: \[
533: \begin{array}{rcl}
534: g=2 & \implies & V=Q_6\subset\PP(1,1,1,1,3), \\[6pt]
535: g=4 & \implies & V=Q_2\cap F_3\subset\PP^5.
536: \end{array}
537: \]
538: As a typical case of the homogeneous projective varieties $G/P$,
539: \[
540: \begin{array}{rcl}
541: g=7 & \implies & V=\Si\cap H_1\cap \cdots\cap H_7,
542: \end{array}
543: \]
544: where $\Si=\OGr^{10}(5,10)$ is the 10-dimensional spinor variety or
545: orthogonal Grassmann variety, that is, the subset of $\Grass(5,10)$
546: consisting of maximal isotropic vector subspaces of the standard inner
547: product $\left(\begin{smallmatrix} 0 & I \\ I & 0
548: \end{smallmatrix}\right)$ in its spinor embedding $\Si\subset\PP^{15}$
549: (see Mukai \cite{Mu2}).
550:
551: As an illustration, we sketch the proof in the style of Mukai that
552: there does not exist any Fano 3-fold $V=V_{20}\subset\PP^{12}$ of genus
553: 11. Write $\si\colon V_1\to V$ for the blowup of a point $P\in V$, and
554: $E\subset V_1$ for the exceptional divisor, with $E\iso\PP^2$ and
555: $\Oh_E(-E)\iso\Oh_{\PP^2}(1)$. Then $K_{V_1}=\si^*K_V+2E$. It follows
556: that the anticanonical linear system $|{-}K_{V_1}|=|\si^*({-}K_V)-2E|$
557: is the birational transform of the linear system
558: $|m_P^2\cdot\Oh_V({-}K_V)|$ of hyperplane sections of $V$ containing
559: the tangent plane $T_PV$. We assume that $P$ does not lie on any line
560: of $V$. Then since $V$ is an intersection of quadrics, we deduce that
561: $\fie_{{-}K_{V_1}}\colon V_1\to\PP^8$ is a morphism, birational because
562: $V$ is indecomposable, with image $V'$ a Gorenstein Fano 3-fold of
563: genus~7. Now on the one hand, $V'$ is indecomposable, so is a linear
564: section of $\OGr^{10}(5,10)$ by the result for $g=7$. On the other
565: hand, it contains the image of $E$, a Veronese surface.
566:
567: Finally, arguing on the geometry of $\OGr^{10}(5,10)\subset\PP^{15}$ and
568: on its universal bundle, we can deduce that the only Veronese surfaces $E$
569: it contains span a \hbox{4-plane} $\PP^4\subset\OGr$; this contradicts the
570: construction of $V'$ as a linear section of $\OGr$ containing $E$.
571:
572: Iskovskikh's proof (deriving from Fano) proceeds by the projection from
573: a general line $L\subset V$; it involves proving that lines exist, and
574: various generality statements about its projection and double
575: projection. Mukai's proof is thus a considerable simplification even in
576: the nonsingular case.
577:
578: \section{$\Q$-Fano 3-folds}
579: \subsection{The Mori category}
580: The minimal model program for 3-folds (usually called MMP or Mori theory)
581: was developed by Mori and others from the late 1970s. As in the theory
582: of surfaces, to get one step closer to a minimal model, one makes a
583: birational contraction, for example, contracting a copy of $\PP^2$ on
584: which the canonical class is negative. However, these contractions lead to
585: singularities, so that Mori theory only works in a suitable category of
586: singular varieties. This leads to the following definitions:
587: \begin{dfn}
588: The {\em Mori category} consists of projective varieties with at worst
589: \hbox{$\Q$-factorial} terminal singularities, where
590: \begin{enumerate}
591: \renewcommand{\labelenumi}{(\alph{enumi})}
592: \item A variety $V$ has {\em terminal singularities} if it is normal,
593: $rK_V$ is a Cartier divisor for some $r>0$, and any resolution of
594: singularities $f\colon Y\to V$ with divisorial exceptional locus
595: $\bigcup E_i$ satisfies
596: \[
597: K_Y = f^*K_V + \sum a_iE_i \quad\hbox{with all $a_i>0$.}
598: \]
599: For example, the cone over the Veronese surface (that is, the quotient
600: singularity $\half(1,1,1)$) is terminal: it is resolved by the blowup
601: $Y\to V$ of the origin, which introduces the exceptional divisor
602: $E\iso\PP^2$ with $\Oh_E(-E)\iso\Oh_{\PP^2}(2)$ and $K_Y{}\rest E\iso
603: \Oh_{\PP^2}(-1)$, so that $K_Y=f^*K_X+\half E$.
604:
605: \item $V$ is {\em $\Q$-factorial\/} if every Weil divisor $D$ on $V$
606: has a multiple $rD$ that is Cartier. You should think of this as an
607: analog in algebraic geometry of the condition that $V$ is a
608: $\Q$-homology manifold: a codimension~1 sub\-variety has a dual
609: cohomology class in $H^2(V,\Q)$. For example, the quartic hypersurface
610: of Example~\ref{exa!V4} is not $\Q$-factorial, since it has double
611: points on $\Pi$ at which no multiple of $\Pi$ is Cartier.
612: \end{enumerate}
613: \end{dfn}
614:
615: \subsection{Terminal singularities}
616: Terminal singularities are necessary for Mori theory: the Mori category
617: is closed under contractions and flips, and is the smallest such category.
618:
619: Terminal singularities were classified by Mori and Reid. They are made
620: up of the following ingredients (for more details, see Reid \cite{YPG}):
621: \begin{enumerate}
622: \renewcommand{\labelenumi}{(\arabic{enumi})}
623: \item Mild isolated hypersurface singularities, for example, double
624: points such as $(xy=f(z,t))\subset\C^4_{x,y,z,t}$. While this is a
625: reasonably concrete class of singularities, already the special case
626: $xy=f(z,t)$ is fairly infinite, containing all isolated plane curve
627: singularities.
628: \item The cyclic quotient singularities $\recip{r}(1,a,r-a)$. The
629: notation means the quotient $V=\C^3/(\Z/r)$, where the cyclic group
630: $\Z/r$ acts by
631: \[
632: (x,y,z)\mapsto (\ep x,\ep^ay,\ep^{r-a}z).
633: \]
634: Setting $x=0$ leads to the cyclic quotient singularity
635: $\recip{r}(a,r-a)$, which is the Du Val singularity A$_{r-1}$
636: \[
637: S:(uv=w^r)\subset \C^3_{u,v,w}.
638: \]
639: Thus $(x=0)$ defines a local elephant $S\in|{-}K_V|$.
640: \item Some combinations of the above two types, typically the
641: main Type~A family $\bigl(xy=f(z^r,t)\bigr)/\recip{r}(a,r-a,1,0)$.
642: \end{enumerate}
643:
644: \begin{rmk} \label{rmk!a?b}
645: Whenever we say that a point $P\in V$ is a terminal quotient
646: singularity of type $\recip{r}(1,a,r-a)$, we always assume that $0<a<r$
647: and $a$ is coprime to $r$. The local class group of $P\in V$ is $\Z/r$,
648: with chosen generator $\Oh_V(-K_V)=\Oh(1)$, a {\em polarisation} of
649: the singularity. The generators of the graded ring $R(V,A)$ that
650: serve as orbifold local coordinates then have weights $1,a,r-a$ mod
651: $r$; compare Corti, Pukhlikov and Reid \cite{CPR}, 3.4.6.
652:
653: The quotient singularity $\recip{r}(a,r-a)$ is the Du Val singularity
654: A$_{r-1}$, so of course does not depend on $a$ up to analytic
655: isomorphism; but we work here with K3 surfaces with a chosen
656: polarisation $\Oh(1)=\Oh(D)$, and the weights mod $r$ of the orbifold
657: local coordinates are uniquely determined.
658:
659: Write $b$ for the inverse of $a$ mod $r$ (that is $ab\equiv1$ mod $r$).
660: Choosing $\ep'=\ep^a$ as a new basis for the group of $r$th roots of
661: unity would put the Du Val singularity in the standard A$_{r-1}$ form
662: $\recip{r}(1,r-1)$ and the terminal 3-fold point in the form
663: $\recip{r}(b,1,r-1)$; and $\Oh_S(D)$ is locally isomorphic to the
664: $\ep'{}^b$ eigensheaf. The birational transform of the general divisor
665: $D$ meets the $b$th curve in the resolution (see
666: Figure~\ref{fig!star}). The quantity $b$ also appears in the orbifold
667: RR formulas of \cite{YPG}, Theorem~10.2 (see Theorem~\ref{th!Hi}
668: below), for essentially the same reason. Muddling up $a$ and $b$ in
669: calculations is a common error, but you find out soon enough when your
670: plurigenera turn out to be fractional; see Exercise~\ref{exc!ele}, (3)
671: for a practical instance.
672: \end{rmk}
673:
674: \begin{dfn}
675: A {\em Fano $3$-fold} is a variety $V$ for which $-K_V$ is ample. We
676: usually add conditions to this. A {\em Mori Fano $3$-fold} $V$ is a Fano
677: 3-fold in the Mori category and with $\rank\Pic V=1$. As one of the
678: possible end products of a MMP, these 3-folds are among the basic
679: building blocks of Mori theory. $V$ is {\em prime} if it is in the Mori
680: category and $\Cl V=\Z\cdot(-K_V)$.
681:
682: As in the Gorenstein case, the {\em anticanonical ring} of a Fano
683: 3-fold $V$ is the graded ring
684: \begin{equation}
685: R(V,-K_V)=\bigoplus_{n\ge0} H^0(V,-nK_V)).
686: \label{eq!RV}
687: \end{equation}
688: An {\em elephant} of $V$ is a general divisor $S\in|{-}K_V|$, just as
689: in \ref{ssec!ele}. If $S$ exists, it is no longer a Cartier divisor at
690: singularities of $V$ of index $r>1$. The graded rings of $V$ and $S$
691: are related by
692: \begin{equation}
693: R(S,A) = R(V,-K_V)/(x_0),
694: \label{eq!RS}
695: \end{equation}
696: where $x_0\in H^0(V,{-}K_V)$ is the equation of $S\in|{-}K_V|$. That is,
697: $R(S,A)$ is a quotient of $R(V,-K_V)$ by a principal ideal generated by
698: an element of degree~1. When two rings are related in this way, many
699: basic algebraic properties of one can be inferred from the other;
700: compare Exercise~\ref{exc!ele} and \ref{sssec!F3}. This is called the
701: {\em hyperplane section principle.} The most useful case is when there
702: is an elephant $S$ with at worst Du Val singularities; then $S$ is a K3
703: surface (this is proved as in \ref{ssec!ele}) polarised by the Weil
704: divisor $A=-K_V{}\rest S$. Then via the graded rings, many properties
705: of the Fano 3-fold $V$ can be read from those of $S$.
706: \end{dfn}
707:
708: \begin{rmk}
709: As we see in \ref{sec!g<0} below, there are cases when
710: $|{-}K_V|=\emptyset$, that is, no elephant exists; or when
711: $h^0(-K_V)=1$, and the single surface $S=|{-}K_V|$ happens to have
712: slightly worse than Du Val singularities, so there is no K3 elephant.
713: Nevertheless, the motivation arising from K3 surfaces is our main
714: guiding principle for the study of Fano 3-folds, even in cases such as
715: these when it is logically inapplicable. Compare Exercise~\ref{exc!ele}.
716: \end{rmk}
717:
718: \begin{exa} \label{exa!2/3}
719: Let $x_1,\dots,x_5,y$ be coordinates on $\PP(1^5,2)$. We start from a
720: $(3,3)$ c.i.\ $V_{3,3}\subset\PP^5(1^5,2)$ as described at the end of
721: Remark~\ref{rmk!orb}, and assume that $V$ contains the weighted
722: projective plane
723: \[
724: \Pi=\PP(1,1,2) \quad\hbox{given by}\quad (x_1=x_2=x_3=0),
725: \]
726: and is otherwise general. Since $\Pi$ is a c.i., the two cubic equations
727: of $V$ are of the form
728: \begin{equation}
729: V: \begin{pmatrix}
730: a_1&a_2&a_3 \\
731: b_1&b_2&b_3
732: \end{pmatrix}
733: \begin{pmatrix} x_1 \\ x_2 \\ x_3
734: \end{pmatrix} =0,
735: \label{eq!3}
736: \end{equation}
737: where $a_i,b_i$ are forms of degree 2 in $x_1,\dots,x_5,y$. We show
738: that
739: $\Pi$ can be contracted to a quotient singularity of type
740: $\recip{3}(1,1,2)$ by a rational map $V\broken W$, where
741: $W\subset\PP(1^5,2,3)$ is a Mori Fano 3-fold. The new coordinate $z$
742: will be a rational section of $\Oh_V(-3K_V)$ with pole along $\Pi$.
743: Thinking of (\ref{eq!3}) as 2 simultaneous equations for $x_1,x_2,x_3$
744: with coefficients $a_i,b_i$ and solving them by Cramer's rule gives:
745: \begin{equation}
746: \begin{matrix}
747: zx_1&=&a_2b_3-a_3b_2, \\
748: zx_2&=&a_3b_1-a_1b_3, \\
749: zx_3&=&a_1b_2-a_2b_1. \\
750: \end{matrix}
751: \label{eq!4}
752: \end{equation}
753: The ``constant of proportionality''
754: \[
755: z=\frac{a_2b_3-a_2b_3}{x_1}=
756: \frac{a_3b_1-a_3b_1}{x_2}=\frac{a_1b_2-a_1b_2}{x_3}
757: \]
758: is a well defined rational form of degree~3 on $V$ with pole along $\Pi$.
759:
760: Now consider the rational map
761: \[
762: \fie\colon V\broken W\subset\PP(1^5,2,3) \quad
763: \hbox{given by} \quad (x_1,\dots,x_5,y,z);
764: \]
765: the image $W$ is given by the 5 equations (\ref{eq!3}--\ref{eq!4}).
766: $\fie$ is an
767: morphism wherever $\Pi$ is a Cartier divisor on $V$, contracts $\Pi$
768: to the point $(0,\dots,0,1)$, and is an iso\-morphism outside $\Pi$. The
769: point $(0,\dots,0,1)$ is the $\recip{3}(1,1,2)$ singularity
770: $\C^3_{x_4,x_5,y}/(\Z/3)$, because when $z=1$, (\ref{eq!4}) gives
771: $x_1,x_2,x_3$ as implicit functions. $V$ is quasi-smooth at points of
772: $\Pi$ where the matrix in (\ref{eq!3}) has rank~2, and $\Pi$ is a
773: Cartier divisor there. We can assume that this matrix has rank $\ge1$
774: everywhere on $\Pi$, hence the blowup of $\Pi$ is a small morphism. It
775: makes $\Pi$ a Cartier divisor so that $\fie$ is a morphism on the
776: blowup.
777:
778: Notice that the graded ring $R(V,-K_V+\recip{3}\Pi)=R(W,-K_W)$ is generated
779: by $(x_1,\dots,x_5,y,z)$ with relations (\ref{eq!3}--\ref{eq!4}). These can
780: be put together as the five $4\times4$ Pfaffians of the skew matrix
781: \[
782: \begin{pmatrix}
783: z&a_1&a_2&a_3 \\
784: &b_1&b_2&b_3 \\
785: &&x_3&-x_2 \\
786: &&&x_1 \\
787: \end{pmatrix} \quad\hbox{of degrees}\quad
788: \begin{pmatrix}
789: 3&2&2&2 \\
790: &2&2&2 \\
791: &&1&1 \\
792: &&&1 \\
793: \end{pmatrix}.
794: \]
795: See Remark~\ref{rmk!Pf} for our conventions on Pfaffians.
796:
797: At the end of Remark~\ref{rmk!orb}, we noted that the Fano 3-fold
798: $V_{3,3}\subset\PP(1^5,2)$ has as its linear section the graded ring
799: $R(C,K_C+\half P)$ corresponding to the orbifold canonical class of a
800: curve of genus~3 together with an orbifold point $P$ of order $r=2$. The
801: construction of this example restricted to $C$ is the graded ring
802: $R(C,K_C+\frac23P)$ corresponding to the orbifold canonical class
803: $K_C+\frac23P$ (compare Example~\ref{exa!1/2}).
804: \end{exa}
805:
806: \section{Numerical data and Hilbert series} \label{sec!Hi}
807: \subsection{The aim}
808: Several hundred families of K3 surfaces $S$ and Mori Fano \hbox{3-folds}
809: $V$ are known. We now develop methods to guarantee gratification from
810: this cornu\-copia. The cases treated in Examples~\ref{exa!1/2}
811: and~\ref{exa!2/3} are pretty tame, and more complicated things like
812: Example~\ref{exa!44} below are more typical. Since we study a polarised
813: variety $X,A$ in terms of the graded ring $R(X,A)$ of (\ref{eq!RX}),
814: the natural numerical invariants of $X$ to take is the {\em list of
815: ingredients} that go into its Hilbert series $P_X(t)$. We explain this
816: in Theorem-Definition~\ref{th!Hi}, but first we work through an example.
817:
818: \begin{exa} \label{exa!44}
819: Consider the weighted hypersurfaces:
820: \begin{equation}
821: S_{44} \subset \PP(4,5,13,22) \quad\hbox{and}\quad
822: V_{44} \subset \PP(1,4,5,13,22);
823: \end{equation}
824: this is Fletcher \cite{Fl}, List~13.3, No.~91, Reid1(91) in the Magma
825: database. We sketch how to calculate with these hypersurfaces. For
826: more details, see \cite{Fl}, Section~13 (compare Dolgachev \cite{D}).
827: Write $x_0,\dots,x_4$ or $x,y,z,t,u$ for variables of weights
828: $a_0,\dots,a_4=1,4,5,13,22$, and let $f_{44}(x,y,z,t,u)$ be a general
829: weighted polynomial of degree~44 in them. Set
830: \begin{itemize}
831: \renewcommand{\labelitemi}{}
832: \item $R[V]=\C[x,y,z,t,u]/\bigl(f_{44}(x,y,z,t,u)\bigr)$,
833: \item $\sC V=\Spec R[V]:\bigl(f_{44}=0\bigr) \subset \C^5$, and
834: \item $V=\Proj R[V]:\bigl(f_{44}=0\bigr) \subset \PP(1,4,5,13,22)$.
835: \end{itemize}
836: Then $\sC V$ is the affine cone over $V$, and $V=\Proj R[V]$ the quotient
837: of $\sC V$ by the $\C^*$ action $x_i\mapsto \la^{a_i}x_i$ for $\la\in\C^*$
838: with the given weights $a_i$. By a combination of Bertini's theorem at
839: general points and explicit calculations at the coordinate strata, one
840: checks that $\sC V$ is nonsingular outside the origin for general
841: $f_{44}$. We say that $V$ is {\em quasi-smooth}. Then the only
842: singularities of $V\subset\PP(1,4,5,13,22)$ arise from the fixed
843: points of the $\C^*$ action; these are the points of the $x_i$-axis,
844: fixed by the cyclic group $\bmu_{a_i}\iso\Z/(a_i)$ of $a_i$th roots of
845: 1, and the points of the $y,u$-plane, which are fixed by
846: $\bmu_2=\{\pm1\}$, because $2=\hcf(4,22)$. Thus in general $V$ has the
847: following cyclic quotient singularities:
848: \begin{equation}
849: \begin{cases}
850: \half(1,1,1) & \hbox{at $(f=0)$ on the $yu$ line (one point);} \\
851: \recip{5}(1,3,2) & \hbox{at the $z$ point $(0,0,1,0,0)$;} \\
852: \recip{13}(1,4,9) & \hbox{at the $t$ point $(0,0,0,1,0)$.} \\
853: \end{cases}
854: \label{eq!44}
855: \end{equation}
856: For example, along the $t$ axis of $\sC V$, we can assume that $f_{44}$
857: has the monomial $zt^3$, which means that $\frac{\partial f}{\partial
858: z}\ne0$ when $t\ne0$, so that $z$ is an implicit function of the other
859: variables. It follows that $\sC V$ is nonsingular there and that $V$ has
860: a quotient singularity of type $\recip{13}(1,4,22)=\recip{13}(1,4,9)$
861: at $(0,0,0,1,0)$. At each singular point, $\Oh_V(1)=\Oh_V(-K_V)$ is a
862: generator of the local class group.
863: \end{exa}
864:
865: \begin{dfn}[Hilbert series]
866: Let $R=\bigoplus_{n\ge0} R_n$ be a graded ring. We assume that $R$ is
867: generated by finitely many elements $x_i$ of positive degree over $R_0=\C$.
868: Its Hilbert series $P(t)$ is defined by setting
869: \[
870: P_n=\dim_\C R_n \quad\hbox{and} \quad P(t)=\sum_{n\ge0} P_nt^n.
871: \]
872: In cases of interest, $P_n$ is given by a formula of orbifold RR type
873: (see \cite{YPG}, Chapter~3), possibly with some corrections for low
874: values of $n$ when cohomology is still present.
875: \end{dfn}
876:
877: \begin{exc}
878: Use (\ref{eq!g=3}) to show that the canonical ring $R(C,K_C)$ of a curve
879: $C$ of genus $g$ has Hilbert series
880: \[
881: P(t) \ =\ \frac{1+(g-2)t+(g-2)t^2+t^3}{(1-t)^2}\,.
882: \]
883:
884: Now write $A=K_C+\sum\frac{r-1}{r}P$ for the orbifold canonical class
885: of Remark~\ref{rmk!orb}; the sum takes place over a basket
886: $\sB=\{P,\frac{r-1}{r}\}$ of orbifold points $P$ of order $r$. In
887: degree $n$, only the rounded down integral divisor
888: $\rd{nA}=nK_C+\sum\rd{\frac{n(r-1)}r}P$ moves. Deduce that the Hilbert
889: series of the orbifold canonical ring $R(C,A)$ is given by
890: \begin{align}
891: P(t) & = \sum_\sB h^0\Bigl(nK_C+\sum\rd{\frac{n(r-1)}r}P\Bigr)t^n
892: \notag \\
893: & = \frac{1+(g-2)t+(g-2)t^2+t^3}{(1-t)^2}+
894: \sum_\sB\frac{t(t+\cdots+t^{r-1})}{(1-t)(1-t^{r})}
895: \label{eq!orbC1} \\
896: & = \frac{1-(g-1)t-t^2}{1-t}+\frac{t}{(1-t)^2}\deg A
897: -\sum_\sB\frac{1}{(1-t^r)}\sum_{i=1}^{r-1}
898: \textstyle{\frac{r-i}{r}}t^i. \notag
899: % \label{eq!orbC2}
900: \end{align}
901: The first expression (\ref{eq!orbC1}) shows how much the integral
902: divisor $\sum\rd{\frac{n(r-1)P}r}$ contributes to each $H^0(C,nA)$. The
903: second expression is a transparent case of an orbifold RR formula: we
904: write $\deg A=\frac1r\deg(rA)$ where $rA$ is a Cartier divisor ($A$
905: itself does not make sense as a sheaf); in each term, the effect of the
906: denominator $1-t^r$ is to multiply by $1+t^r+t^{2r}+\cdots$, so that
907: the fractional part $\{\frac{i(r-1)}{r}\}=\frac{r-i}{r}$ discarded in
908: the rounddown is repeated periodically with period $r$.
909:
910: This formula for orbifold curves can serve as a model for the formulas
911: of \cite{YPG}, Chapter~3. In fact, a result of Becky Leng's thesis
912: \cite{Leng} derives the orbifold RR formula for 3-folds (\cite{CPR},
913: Theorem~10.2) via a reduction to the curve case.
914: \end{exc}
915:
916: \subsection{Hilbert series for K3 surfaces and Fano 3-folds}
917: The analogous formulas for the Hilbert series of the graded ring over a
918: K3 surface $S$ or a Fano 3-fold $V$ are contained in Alt{\i}nok \cite{A}
919: and \cite{A1}. The proofs are based on the results of \cite{YPG},
920: Chapter~3.
921:
922: \begin{thm-dfn}\label{th!Hi}
923: Let $S$ be a K3 surface with Du Val singularities, and $D$ a Weil
924: divisor. The Hilbert series $P_S(t)=\sum_{n\ge0}h^0(S,nD)t^n$ is given
925: by
926: \begin{equation}
927: P_S(t)\ =\ \frac{1+t}{1-t}+\frac{t(1+t)}{(1-t)^3}\,\frac{D^2}2
928: -\sum_\sB \recip{(1-t^r)}\sum\limits_{i=1}^{r-1}
929: \frac{\overline{bi}(r-\overline{bi})t^i}{2r}\,,
930: \label{eq!K3-s}
931: \end{equation}
932: where the sum takes place over a {\em basket} $\sB=\{\recip{r}(a,-a)\}$
933: of cyclic quotient singularities. In each term, $b$ is the inverse
934: of $a$ mod $r$, as in Remark~\ref{rmk!a?b}, and $\overline{bi}$ denotes
935: the minimal nonnegative residue mod $r$.
936:
937: We define the {\em genus} $g=g(S,D)$ by the formula:
938: \[
939: P_1(S,D)=h^0(\Oh_S(D))=g+1.
940: \]
941: The {\em numerical data} of $S,D$ are the genus $g$ and the basket
942: $\sB=\{\recip{r}(a,-a)\}$. Calculating $P_1(S,D)$ as the coefficient of
943: $t$ in $(\ref{eq!K3-s})$ gives a formula for $D^2$ in terms of $g$ and
944: $\sB$:
945: \[
946: D^2 = 2g-2 + \sum_\sB \frac{b(r-b)}{r}\,.
947: \]
948:
949: Let $V$ be a Fano $3$-fold and write $A=-K_V$. Then its anticanonical
950: ring has Hilbert series
951: \begin{align}
952: P_V(t)&\ =\ \sum_{n\ge0}h^0(V,nA)t^n \notag \\
953: &\ =\ \frac{1+t}{(1-t)^2}+\frac{t(1+t)}{(1-t)^4}\,\frac{A^3}2
954: -\sum_\sB \recip{(1-t)(1-t^r)}
955: \sum\limits_{i=1}^{r-1}
956: \frac{\overline{bi}(r-\overline{bi})t^i}{2r}\,.
957: \label{eq!Fano_3-f}
958: \end{align}
959: where the sum takes place over a basket $\sB=\{\recip{r}(1,a,r-a)\}$ of
960: terminal quotient singularities. We define the {\em genus} $g=g(V)$ by
961: \[
962: P_1(V,A)=h^0(V,A)=h^0(\Oh_V(-K_V))=g+2.
963: \]
964: Then, as in the K3 case, the {\em numerical data} of $V$ are the genus
965: $g$ and the basket $\sB$; calculating $P_1$ from $(\ref{eq!Fano_3-f})$
966: gives essentially the same formula as above:
967: \[
968: A^3 = 2g-2 + \sum_\sB \frac{b(r-b)}{r}\,.
969: \]
970: \end{thm-dfn}
971:
972: \begin{exc} \label{exc!ele}
973: \begin{enumerate}
974: \renewcommand{\labelenumi}{(\arabic{enumi})}
975: \item If a Fano 3-fold $V$ has a K3 elephant $S\in|{-}K_V|$, the
976: numerical data of $V$ and $S$ are related in the obvious way, and
977: (\ref{eq!K3-s}) and (\ref{eq!Fano_3-f}) only differ by the extra factor
978: $(1-t)$ in the denominator. Use the hyperplane section principle
979: (\ref{eq!RS}) to derive (\ref{eq!Fano_3-f}) from (\ref{eq!K3-s}).
980:
981: \item In the same way, if $S,C$ is a polarised K3 surface having only
982: singularities of type $\recip{r}(1,r-1)$ (with $C$ passing through each
983: singular point as a nonsingular curve in the $1$ eigenspace), derive
984: (\ref{eq!K3-s}) from the orbifold canonical curve formula
985: (\ref{eq!orbC1}). Singularities of type $\recip{r}(1,r-1)$ account for
986: the large majority of all cyclic quotient singularities in K3 baskets.
987:
988: \item The varieties in Example~\ref{exa!44} have $g=0$ and the basket
989: (\ref{eq!44}). Check that
990: \[
991: D^2=-4+\half+\frac{2\cdot3}6+\frac{3\cdot10}{13}=
992: \frac{44}{4\cdot5\cdot13\cdot22}=\recip{130}
993: \]
994: (note the shift from $a=4$ in $\recip{13}(4,9)$ to its inverse $b=10$
995: in the fractional contribution $\frac{3\cdot10}{13}$). Check that
996: $P_S(t)$ given by (\ref{eq!K3-s}) satisfies
997: \[
998: (1-t^4)(1-t^5)(1-t^{13})(1-t^{22})P_S(t)=1-t^{44}.
999: \]
1000: \end{enumerate}
1001: \end{exc}
1002:
1003: \subsection{Numerical data of K3s and Fano 3-folds}\label{sec!g<0}
1004: \subsubsection{Invariants and inequalities}\label{sssec!ineq}
1005: As we saw in Theorem-Definition~\ref{th!Hi}, the numerical data of a
1006: K3 surface or a Fano 3-fold consist of an integer genus $g$ and a
1007: basket $\sB$ of fractional expressions. For a K3 surface in
1008: characteristic~0 we have
1009: \begin{equation}
1010: g\ge-1, \quad 2g-2+\sum\frac{b(r-b)}r>0\quad\hbox{and}\quad
1011: \sum_\sB (r-1)\le 19.
1012: \label{ineq!K3}
1013: \end{equation}
1014: The last inequality comes from the fact that the singularities of the
1015: basket contribute $\sum(r-1)$ exceptional $-2$-curves to $\Pic S$, that
1016: form a negative definite set; in characteristic $p$, the result would
1017: be $\sum(r-1)\le21$.
1018:
1019: There are approximately 6,640 possible baskets satisfying
1020: (\ref{ineq!K3}), and countably many possible values of $g$ for each
1021: basket. A small number of extreme cases are excluded because the
1022: polarisation and the basket cannot fit inside the K3 lattice $\LK$ (see
1023: \ref{sssec!K3L} below); for example, a K3 surface cannot contain
1024: $\ge17$ disjoint \hbox{$-2$-curves}. However, apart from these, all other $g$
1025: and $\sB$ really occur. Most give rise to graded rings of high
1026: codimension, and our methods based on graded ring are no longer
1027: appropriate for studying them; the same applies to canonical curves or
1028: K3 surfaces of genus $g\gg0$, for which explicit equations and graded
1029: ring methods give little information, and other ideas are needed, such
1030: as the period space and Torelli for K3 surfaces.
1031:
1032: For a Fano 3-fold $V$ we have
1033: \begin{equation}
1034: g\ge-2, \quad 2g-2+\sum\frac{b(r-b)}r>0\quad\hbox{and}\quad
1035: \sum_\sB \left(r-\recip{r}\right)<24.
1036: \label{ineq!Fa}
1037: \end{equation}
1038: The last inequality comes from \cite{YPG}, Corollary~10.3, the orbifold
1039: analog of the RR formula $\recip{24}c_1c_2=\chi(\Oh_V)=1$; compare also
1040: Kawamata \cite{Ka1}, 2.2 and \cite{Ka2}. In fact the argument of
1041: \cite{Ka2} using Bogomolov stability gives
1042: \[
1043: \sum \sB\left(1-\recip{r}\right)<24-8(-K_V)^3,
1044: \]
1045: a slightly stronger inequality.
1046:
1047: Note that (\ref{ineq!Fa}) is stronger than (\ref{ineq!K3}) if the number
1048: of singularities is large; for example, (\ref{ineq!Fa}) allows only $15$
1049: singularities. On the other hand, (\ref{ineq!Fa}) is weaker if there are
1050: few singularities -- for example, it would allow a cyclic singularity of
1051: index~24, or a singularity of index 22 plus one of index 2. Since we
1052: expect there to be rather few families of Fano 3-folds with $h^0(-K_V)=0$,
1053: and those with $h^0(-K_V)>0$ to have a K3 elephant after deformation, it
1054: seems unlikely that any of these extra Fano 3-fold cases really occur.
1055:
1056: \subsubsection{Negative genus}\label{ssec!uh}
1057: On a K3 surface, $g=-1$ means that the polarising divisor $D$ is
1058: ineffective; or we can say that a curve of genus $-1$ does not exist
1059: (pretty reasonable, uh?). A Fano 3-fold with $g=-2$ has
1060: $|{-}K_V|=\emptyset$; we can think of $g=-2$ as saying that two things
1061: do not exist: $V$ does not have a canonical curve section $S_1\cap S_2$,
1062: and moreover, it does not even have an elephant $S\in|{-}K_V|$. Only
1063: three or four families of Fano 3-folds with $h^0(-K_V)=0$ are known;
1064: the simplest of these, and the first to be discovered, is the
1065: codimension~2 weighted c.i.\ $V_{12,14}\subset \PP(2,3,4,5,6,7)$ due to
1066: Fletcher (see \cite{Fl}, List~16.7, No.~60 and compare Reid \cite{Ki},
1067: Example~9.14).
1068:
1069: \subsubsection{$\Q$-smoothing and the general elephant}
1070: \label{sssec!GE}
1071: A Mori Fano 3-fold $V$ is allowed to have general terminal
1072: singularities. Or it might happen that $|{-}K_V|\ne\emptyset$, but every
1073: $S\in|{-}K_V|$ has an essential singularity (worse than Du Val); this
1074: is not very surprising if $P_1(V)=1$. For example, one can construct a
1075: surface $S_{14}\subset\PP(2,2,3,7)$ having an elliptic singularity,
1076: but contained in a quasi-smooth $V_{14}\subset\PP(1,2,2,3,7)$.
1077: (Explicit construction: start from the scroll $\FF_3$, with fibre $A$
1078: and negative section $B$; then $S_{14}$ is the double cover ramified in
1079: $B+C_1+C_2$ where $C_1\in|3A+B|$ is the general hyperplane section of
1080: the cone $\FFbar_3$, and $C_2\in|7A+2B|$ has a tacnode on $C_1$, giving
1081: $C_1+C_2$ an infinitely near triple point.)
1082:
1083: It is known that a terminal 3-fold singularity has a {\em
1084: $\Q$-smoothing}, that is, a small deformation with only cyclic
1085: quotient terminal singularities (see \cite{YPG}, 6.4 for explicit
1086: equations). This is the idea behind the basket appearing in
1087: (\ref{eq!Fano_3-f}), and one step in its proof (\cite{YPG}, proof of
1088: Theorem~10.2). It seems reasonable to conjecture that every prime Fano
1089: 3-fold $V$ also has a $\Q$-smoothing; compare Namikawa \cite{N} and
1090: Minagawa \cite{Mi1}, \cite{Mi2} for partial results. One might hope to
1091: prove this in terms of deformation theory and Hodge theoretic
1092: consequences of $-K_V$ ample, by analogy with work on smoothing
1093: Calabi--Yau 3-folds (see Gross \cite{G} and Namikawa and Steenbrink
1094: \cite{NS}). Moreover, if $g\ge-1$, we also conjecture that $V$
1095: together with its elephant $S\in|{-}K_V|$ has a {\em simultaneous
1096: $\Q$-smoothing}, that is, a small deformation such that the pair
1097: $S\subset V$ has only standard cyclic singularities
1098: $\recip{r}(a,r-a)\subset\recip{r}(1,a,r-a)$. If in addition $g\ge0$ and
1099: the basket of $V$ consists only of $\recip{r}(1,1,r-1)$, then after a
1100: small deformation, $V$ would have a curve section $C=S_1\cap S_2$ that
1101: is an orbifold canonical curve as in Remark~\ref{rmk!orb}.
1102:
1103: \subsubsection{Numerical data and the K3 lattice}\label{sssec!K3L}
1104: The numerical data of a polarised K3 surface $S$ corresponds to a
1105: sublattice
1106: \[
1107: L\Bigl(g,\sB=\Bigl\{\recip{r}(a,r-a)\Bigr\}\Bigr)\subset\LK
1108: \]
1109: of the standard K3 lattice. We assume that, after a
1110: \hbox{$\Q$-smoothing}, the singularities of $S$ are exactly the cyclic
1111: quotient singularities of the basket. Let $f\colon T\to S$ be the
1112: minimal resolution of singularities, and consider the sublattice
1113: $L\subset\Pic T$ generated by the exceptional $-2$-curves of $f$
1114: together with $f^*(ND)$, where $N$ is the global index of $D$. The
1115: lattice $L=L(g,\sB)$ is based by the {\em quasistellar graph}
1116: $\Ga(g,\sB)$ of Figure~\ref{fig!star}.
1117: \begin{figure}[ht]
1118: \begin{picture}(200,70)(25,6)
1119: \put(140,30){\circle{12}}
1120: \put(152,27){$2g-2$}
1121: \put(140,36){\line(0,1){20}}
1122: \put(58,56){\circle*{7.5}} \put(55,66){$b$}
1123: \put(40,66){$\recip{r}\Bigl($}
1124: \put(59,56){\line(1,0){21}}
1125: \put(85,56){\circle*{7.5}} \put(80,66){$2b$}
1126: \put(94,56){\dots}
1127: \put(115,56){\circle*{7.5}} \put(101,66){\dots}
1128: \put(117,56){\line(1,0){21}}
1129: \put(140,56){\circle*{7.5}} \put(122,66){$(r-b)b$}
1130: \put(141,56){\line(1,0){21}}
1131: \put(165,56){\circle*{7.5}}
1132: \put(174,56){\dots}
1133: \put(195,56){\circle*{7.5}} \put(180,66){$r-b$}
1134: \put(206,66){$\Bigr)$}
1135: \put(114,30){\line(1,0){20}}
1136: \put(113,30){\circle*{7.5}}
1137: \put(122,10){\line(1,1){15}}
1138: \put(121,8){\circle*{7.5}}
1139: \put(89,8){\dots}
1140: \put(107,8){\line(1,0){28}}
1141: \put(141,8){\dots}
1142: \end{picture}
1143: \caption{The quasistellar graph $\Ga(g,\sB)$.}
1144: \label{fig!star}
1145: \end{figure}
1146: The central vertex is a divisor $B$ of self-intersection $B^2=2g-2$
1147: (ineffective if $g=-1$, that is, $B^2=-4$); for each term
1148: $\recip{r}(a,r-a)\in\sB$, it meets the $b$th curve in a chain of $r-1$
1149: exceptional $-2$-curves, where $ab\equiv1$ mod $r$. (Compare Belcastro
1150: \cite{Be} for the lattice of the famous 95 families of K3
1151: hypersurfaces.)
1152:
1153: The polarising $\Q$-divisor $D=f^*D=B+\sum m_iE_i$ has exceptional
1154: curves in each chain weighted by the monotone sequence of arithmetic
1155: progressions
1156: \[
1157: \recip{r}\Bigr(
1158: b,2b,\dots,b(r-b-1),b(r-b),(b-1)(r-b),\dots,2(r-b),r-b \Bigl),
1159: \]
1160: giving
1161: \[
1162: D\cdot E_i=0, \quad\hbox{and}\quad
1163: D^2=D\cdot B=2g-2+\sum\frac{b(r-b)}{r}\,.
1164: \]
1165:
1166: The lattice $L=L(g,\sB)$ has signature $(+1,-\sum(r-1))$ and
1167: discriminant $(\prod_\sB r)D^2$. Although $L$ is determined up to
1168: isomorphism by the numerical data $g$ and $\sB$, its embedding into the
1169: standard K3 lattice
1170: \[
1171: L\into H^2(T,\Z)=\LK
1172: \]
1173: is not in general completely determined by $g$ and $\sB$, and is an
1174: additional topological invariant of $S,D$. For example, if $S$ is
1175: nonsingular then $L=\Z\cdot D=\Span{2g-2}$ is the lattice with one
1176: generator of square length $2g-2$, but it can happen that $D$ is
1177: divisible in $\Pic S$ by some integer $n$ with $n^2\divides g-1$, and
1178: then $L\subset\LK$ is not a primitive sublattice. If $L$ has high rank,
1179: it can have several inequivalent primitive embeddings $L\into\LK$,
1180: having nonisomorphic orthogonal complements.
1181:
1182: The Hilbert series of a Fano 3-fold $V$ is controlled by the same
1183: combinatorics: the lattice $L(g,\sB)$ is still there (even if $g=-2$).
1184: For simplicity, assume that $V$ has only terminal quotient
1185: singularities $\recip{r}(1,a,r-a)$. Make the economic resolution $Y\to
1186: V$ of these singularities by successive $(1,a,r-a)$ weighted blowups.
1187: These are the Kawamata blowups of Corti, Pukhlikov and Reid
1188: \cite{CPR}, 3.4.2 (see also \cite{YPG}, 5.7; but note that the
1189: chain of Kawamata blowups is a unique ``best'' choice of economic
1190: resolution that came on line after \cite{YPG}). The lattice $L(g,\sB)$
1191: is the Picard lattice $\Pic Y$, with bilinear product
1192: \[
1193: (D_1,D_2)\mapsto (-K_Y)\cdot D_1\cdot D_2.
1194: \]
1195: If $g\ge-1$ and we assume that $V$ (or its deformation) has a K3
1196: elephant, then $L(g,\sB)$ is a sublattice of $\LK$. If $g=-2$, so that
1197: $B^2=-6$, we have to imagine that we are looking at a nonexistent K3
1198: surface having a polarising divisor $D$ with $h^0(D)=-1$. In this case,
1199: there is no obvious a priori relation between $L(g,\sB)$ and $\LK$;
1200: however, as we said in \ref{sssec!ineq}, there are probably only a few
1201: cases, and it is not a substantial restriction for a lattice in this
1202: range to be a sublattice of $\LK$, so who knows?
1203:
1204: \section{From K3s to Fano 3-folds} \label{sec!proj}
1205: Just as the general theory of~\ref{ssec!ele}, what we said in
1206: Section~\ref{sec!Hi} applies to K3 surfaces and to Fano 3-folds alike.
1207: We said in \ref{sssec!ineq} that K3 surfaces go on for ever and for
1208: ever, with many baskets continuing to exist for infinitely many values
1209: of $g$. However, as in the more specific 3-fold theory of
1210: \ref{ssec:specific}, there are results of Kawamata
1211: \cite{Ka1}--\cite{Ka2} saying that Fano 3-folds are bounded. By
1212: analogy with Theorem~\ref{th!clF}, we guess that only a couple of
1213: thousand families exist, and we eventually aspire to a precise
1214: classification. Most Fano \hbox{3-folds} have $H^0(-K_V)\ne0$, and, as
1215: a first attempt, we can take those having a K3 elephant as typical,
1216: and allow ourselves extra generality assumptions on $V$ such as
1217: $\Q$-smoothing, Brill--Noether general behaviour of linear systems
1218: (nonhyperelliptic, etc.).
1219:
1220: \subsection{Takagi's results} \label{sssec!F3}
1221: In his lecture at the Singapore conference \cite{T1}, Takagi sketched
1222: a preliminary classification of prime $\Q$-Fano \hbox{3-folds} of genus
1223: $g\ge2$ having at worst terminal quotient singularities and a K3
1224: elephant. The assumption that $|{-}K_V|$ contains a K3 is a strong
1225: condition: it guarantees that there is a resolution on which $-K_{V'}$
1226: is nef. In this case, the anticanonical system $|{-}K_{V'}|$ is a big
1227: linear system of K3 surfaces, and defines a birational model of $V$ as
1228: an indecomposable Gorenstein Fano 3-fold $X$ (Fano would appreciate
1229: this picture at once!). Note that the anticanonical system $|{-}K_V|$
1230: itself (but not $|{-}2K_V|$, $|{-}3K_V|$, etc.)\ is invariant under
1231: passing to the economic resolution of points of index $r\ge2$. Mukai's
1232: Theorem~\ref{th!clF} applies to $X$, thus reducing the study of
1233: \hbox{$\Q$-Fano} \hbox{3-folds} to problems in projective geo\-metry
1234: concerning sections of homogeneous spaces with special configurations
1235: of planes and singularities. In particular, Takagi proves that if $V$
1236: has at least one terminal quotient singularity then $g\le8$; the cases
1237: $g=7$ and $g=8$ is accessible since $X=\fie_{-K_V}(V)$ is a linear
1238: section of a Grassmann or orthogonal Grassmann variety. Takagi's thesis
1239: \cite{T} also settled all cases with only $\half(1,1,1)$ singularities.
1240:
1241: \subsection{Low codimension} \label{sssec!lowc}
1242: We study Fano 3-folds in terms of the (pluri-) anticanonical ring
1243: \[
1244: R=R(V,-K_V)=\C[x_0,\dots,x_N]/I_V,
1245: \]
1246: where $x_0,\dots,x_N$ (more usually $x,y,\dots,u$, etc.)\ are
1247: homogeneous generators of weights $a_0,\dots,a_N$ and $I_V$ is the
1248: ideal of all relations. Since $-K_V$ is ample, $\Proj R$ is the
1249: anticanonical model of $V$ in w.p.s.:
1250: \[
1251: V=\Proj R\subset\PP^N(a_0,a_1,\dots,a_N).
1252: \]
1253: Here we take the complete anticanonical ring $R(V,-K_V)$, so that
1254: $V\subset\PP^N$ is projectively normal; it is known that $V$ is
1255: projectively Gorenstein, that is, $R$ is a Gorenstein ring (see for
1256: example Goto and Watanabe \cite{GW}).
1257:
1258: The {\em codimension} of $R$ means the codimension $N-3$ of the
1259: anticanonical ring $R(V,-K_V)$. It is a measure of the difficulty of
1260: studying $V$ directly. In codimension~2, a Gorenstein ring is a c.i.\
1261: (Serre), and in codimension~3, it is given by the $2k\times2k$ diagonal
1262: Pfaffians of a skew $(2k+1)\times(2k+1)$ matrix (Buchsbaum and Eisenbud);
1263: only $5\times5$ occurs here, with the single exception of the classic
1264: case $Q_1\cap Q_2\cap Q_3\subset\PP^6$.
1265:
1266: The cases when $V$ is a hypersurface or codimension~2 c.i.\ are settled
1267: in Fletcher \cite{Fl}, and the codimension~3 cases in Alt{\i}nok
1268: \cite{A}. There are 95 families of Fano 3-fold hypersurfaces, 84
1269: codimension~2 c.i.s (including Fletcher's example mentioned in
1270: \ref{ssec!uh} with $H^0(-K_V)=0$), and 70 families of codimension~3
1271: Pfaffians. These varieties are given by explicit equations, that we
1272: can write down without reference to K3s, and depend on an irreducible
1273: parameter space. Their construction is unobstructed, and every family
1274: of K3 surfaces extends to a family of Fano 3-folds.
1275:
1276: \subsection{Weighted Grassmannian} \label{ssec!wGr}
1277: In codimension~3, the five Pfaffians are a weighted form of the equations
1278: of the Grassmannian $\Grass(2,5)$ (see \ref{rmk!Pf}). Thus a
1279: codimension~3 K3 surface $S\subset\PP(a_0,\dots,a_5)$ has Hilbert
1280: series
1281: \[
1282: 1-\sum t^{b_i} + \sum t^{k-b_j} - t^k,
1283: \]
1284: where $k=\sum_0^5 a_i$ and $2k=\sum_1^5 b_i$, so that the skew matrix
1285: in \ref{rmk!Pf} has entries of weight $\wt x_{ij}=k-b_i-b_j$ (compare
1286: Reid \cite{Ki}, Example~3.8). We can view the $x_{ij}$ as coordinates
1287: on $\bigwedge^2U\otimes L$, where $U$ is a weighted $\C^5$ with weights
1288: $c-b_i$, and $L=\C$ with weight $k-2c$ (for some convenient $c$); then
1289: the equations of $S$ are the pullback of the weighted Grassmann
1290: $\wGr\subset\PP(\bigwedge^2U)$ (tensors of rank~2) by a weighted
1291: projective map $\PP(a_0,\dots,a_5)\broken \PP(\bigwedge^2U)$. In the
1292: spirit of Mukai's work discussed in \ref{ssec:specific}, the interesting
1293: question is to construct directly an exceptional (orbi-) vector bundle
1294: $\sE$ on $S$ whose Serre module $H^0_*(\sE)$ is generated by 5 sections
1295: $u_i\in\sE(c-b_i)$, giving the embedding
1296: $S\into\wGr\subset\PP(\bigwedge^2U)$.
1297:
1298: The point here is to get away from the algebraic treatment of the
1299: equations of $S$ to a more geometric understanding of what they mean;
1300: that is, eventually, to replace the appeal to Buchsbaum and Eisenbud's
1301: algebraic result on codimension~3 Gorenstein rings by arguments in
1302: the style of Mukai in terms of exceptional vector bundles on the K3
1303: section. There are preliminary results in this direction due to Corti
1304: and Reid \cite{CR}. This raises important unsolved problems in higher
1305: codimension: weighted versions of Mukai's symmetric spaces make sense,
1306: but seem to give few examples of varieties of small coindex.
1307:
1308: \subsection{Gorenstein unprojection}
1309: In codimension~$\ge4$, there is no analogous structure theorem for
1310: Gorenstein rings; our current strategy for K3s and Fano \hbox{3-folds}
1311: in codimension~4, 5, etc., is based (in most cases) on {\em Gorenstein
1312: projection}. A model case is the projection $S_d\broken S_{d-1}$ of a
1313: del Pezzo surface $S_d\subset\PP^d$ from $P\in S_d$ (for simplicity,
1314: think of $d\ge4$). Here both $S_d$ and $S_{d-1}$ are anticanonical: the
1315: blowup $S'\to S_d$ of $P$ has discrepancy~1, so that the anticanonical
1316: map $S'\to S_{d-1}$ given by $|{-}K_{S_{d-1}}|=|{-}K_{S_d}-E|$ is the
1317: same thing as the linear projection from $P$.
1318:
1319: Although the study of a Fano 3-fold $V\subset\PP^N(a_0,a_1,\dots,a_N)$
1320: is a biregular problem in the first instance, we can use birational
1321: methods to attack it. Fano's linear projection of
1322: $V=V_{2g-2}\subset\PP^{g+1}$ with $g\ge7$ from a line can also be
1323: interpreted as making a blowup $V_1\to V$, then recalculating the
1324: anticanonical ring $R(V_1,-K_{V_1})$ as a subring of $R(V,-K_V)$, thus
1325: deducing a birational map $V\broken V'=\Proj R(V_1,-K_{V_1})$. In
1326: Fano's study, $V$ is the unknown, and is hard to work with because it
1327: has high codimension, whereas its projection $V'$ has smaller
1328: codimension, so it, together with the exceptional divisor of the
1329: projection $F\subset V'$, may be more tractable.
1330:
1331: For us, the key point is that the birational relation between $V$ and
1332: $V'$ (geo\-metrically a projection) can be handled in terms of
1333: inclusions
1334: \[
1335: R(V',-K_{V'})\subset R(V,-K_V)
1336: \]
1337: between Gorenstein rings. This area has recently been clarified by
1338: Papadakis and Reid's treatment of the inverse birational map
1339: $V'\broken V$ as {\em Gorenstein unprojection} or ``constructing
1340: big Gorenstein rings from small ones'' (see \cite{PR} and \cite{Ki},
1341: and compare Kustin and Miller \cite{KM}); in \cite{PR}, the inverse is
1342: constructed in terms of the adjunction formula for the
1343: Grothendieck--Serre dualising sheaf.
1344:
1345: \subsection{Type~I unprojection} \label{ssec:TI}
1346: We discuss here the most straightforward case of projection,
1347: corresponding to Kustin and Miller unprojection, which already applies
1348: to the majority of K3s in codimension $\ge4$. Let
1349: \[
1350: S\subset\PP(a_1,\dots,a_N) \quad\text{or}\quad
1351: V\subset\PP(1,a_1,\dots,a_N)
1352: \]
1353: be a K3 surface or Fano 3-fold. We only treat the Fano case here,
1354: leaving the reader to make the obvious modifications in the K3 case.
1355:
1356: A coordinate point $P_k=(0,\dots,1,\dots,0)\in V$ is a {\em Type~I
1357: centre} if $P_k\in V$ is a terminal quotient singularity
1358: $\recip{r}(1,a,r-a)$, where $\wt x_k=r$, and global coordinates of the
1359: projective space $x_0,x_i,x_j$ of weight equal to $1,a,r-a$ provide
1360: local orbifold coordinates at $P_k$. Let $V_1\to V$ be the Kawamata
1361: blowup, that is, the weighted blowup of $P_k$ with weights $1,a,r-a$;
1362: then the exceptional locus $E\iso\PP(1,a,r-a)$ has minimal discrepancy
1363: $\recip{r}$, and, following \cite{CPR}, we write
1364: \[
1365: A=-K_V \quad\hbox{and}\quad B=-K_{V_1}=A-\recip{r}E.
1366: \]
1367: Now because $\wt(x_0,x_i,x_j)=(1,a,r-a)$, and $x_0,x_i,x_j$ vanish
1368: along $E$ with multi\-plicity $\recip{r},\frac{a}{r},\frac{r-a}{r}$
1369: (see \cite{CPR}, Proposition~3.4.6), they belong to the subring
1370: $R(V_1,B)$. Thus $R(V_1,B)$ is the subring
1371: $k[x_0,\dots,\widehat{x_k},\dots,x_N]\subset R(V,A)$ obtained by
1372: eliminating $x_k$ only. Then
1373: \[
1374: V'=\Proj R(V_1,B)\subset\PP(1,a_1,\dots,\widehat{a_k},\dots,a_N)
1375: \]
1376: is a weak Fano 3-fold containing the plane $\Pi=\PP(1,a,r-a)$. Be
1377: warned that $V'$ is not a Mori Fano 3-fold, since the Weil divisor
1378: $\Pi$ is not in $\Z\cdot(-K_{V'})$; one usually expects $V'$ to be the
1379: {\em midpoint} of a Sarkisov link, compare \cite{CPR}, 4.1, (3).
1380:
1381: A simple example of a Type~I unprojection was treated in
1382: Example~\ref{exa!2/3}. Type~I projections include the construction of
1383: the 64 quadratic involutions of \cite{CPR}, 4.4 (or rather, of the
1384: first half $X\broken Z$, up to the midpoint of the link).
1385:
1386: \begin{exc} \label{ex!t1}
1387: In the numerical data $g,\sB$ for a K3 surface
1388: (Section~\ref{sec!Hi}), replace an element $\recip{r}(a,r-a)$ of $\sB$
1389: by two elements $\recip{a}(r,-r)$ and $\recip{r-a}(r,-r)$, and assume
1390: that $D^2$ remains positive. Study how this numerical projection
1391: affects the rhs of (\ref{eq!K3-s}): show that it subtracts
1392: $\frac{t^r}{(1-t^r)(1-t^a)(1-t^{r-a})}$ from the Hilbert series
1393: $P_{S,D}(t)$, and reduces $D^2$ by $\recip{ra(r-a)}$. Compare
1394: \cite{PR}, Exercise~2.7, and see Example~\ref{exa!dbeg1},
1395: page~\pageref{Altinok3(63)} for a numerical instance.
1396: \end{exc}
1397:
1398: The effect of a Type~I projection as in Exercise~\ref{ex!t1} on the
1399: numerics of Section~\ref{sec!Hi} gives the set of numerical data of K3
1400: surfaces the structure of a directed graph, with comparatively few
1401: connected components. With few exceptions, families of K3 surfaces in
1402: codimension~$\ge2$ have projections to smaller codimension, most
1403: commonly of Type~I. Of the 142 codimension~4 K3s in the K3 database,
1404: 116 have a numerical Type~I centre. All but 2 of the remaining cases
1405: are covered by the higher types of projection discussed in Reid
1406: \cite{Ki}, Section~9 and \cite{T4}. These higher projections, and the
1407: small core of exceptions not admitting any projections, are interesting
1408: and demand further study; we suspect that these more complicated K3s
1409: are unlikely to extend to Fano 3-folds.
1410:
1411: \subsection{Tom and Jerry unprojections to codimension 4} \label{T&J}
1412: When applicable, a Type~I projection as described in \ref{ssec:TI}
1413: reduces the study of a Fano $V$ in w.p.s.\ $\PP^N$ to a variety $V'$ in
1414: w.p.s.\ $\PP^{N-1}$, but specialised to contain a weighted projective
1415: subspace:
1416: \begin{equation}
1417: \PP(1,a,r-a)\subset V'\subset \PP^{N-1}.
1418: \label{eq!t&j}
1419: \end{equation}
1420: Thus Fano \hbox{3-folds} whose numerical data admits a codimension~4
1421: candidate can often be studied via projections. However, setting up the
1422: unprojection data (\ref{eq!t&j}) is still a difficult problem;
1423: complicated features of $V$, such as obstructed equations or reducible
1424: moduli spaces, must be faithfully reproduced in the unprojection data.
1425:
1426: In (\ref{eq!t&j}), the w.p.s.\ $\PP(1,a,r-a)$ is a codimension~4 c.i., and
1427: $V'$ is a codimension~3 variety given by the Pfaffians of a $5\times5$
1428: skew matrix. Thus to obtain codimension~4 Fanos with this type of
1429: projection, the problem is how to put a codimension~4 c.i.\ inside a
1430: $5\times5$ Pfaffian. Similarly for K3s.
1431:
1432: There are two different solutions to this problem, called {\em Tom} and
1433: {\em Jerry}, treated in detail in Papadakis \cite{P}--\cite{P1}
1434: (see also
1435: \cite{Ki}, Examples~6.4 and 6.8, and Section~8). In these two cases, we
1436: specialise the skew $5\times5$ matrix defining $V'$ to
1437: \[
1438: M_{\mathrm{Tom}}=
1439: \left(
1440: \begin{array}{c@{\enspace}c@{\enspace}ccc}
1441: x_{12} && x_{13} & x_{14} & x_{15} \\[2pt]
1442: \cline{2-5}
1443: &\vline&a_{23}&a_{24}&a_{25} \\
1444: &\vline&&a_{34}&a_{35} \\
1445: &\vline&&&a_{45}
1446: \end{array}
1447: \right)
1448: \quad\hbox{or}\quad
1449: M_{\mathrm{Jerry}}=
1450: \left(
1451: \begin{array}{cccc}
1452: a_{12}&a_{13} & a_{14} & a_{15} \\
1453: &a_{23}&a_{24}&a_{25} \\[2pt]
1454: \cline{2-4}
1455: &&x_{34}&x_{35} \\
1456: &&&x_{45}
1457: \end{array}
1458: \right)
1459: \]
1460: where the 6 entries $a_{ij}$ in the bottom left $4\times4$ block of
1461: $M_{\mathrm{Tom}}$ specialise to lie in a codimension~4 complete
1462: intersection ideal $(y_1,\dots,y_4)$ (and ditto for the 7 entries
1463: $a_{ij}$ in the first two rows and columns of $M_{\mathrm{Jerry}}$).
1464: In either case, the theoretical construction of \cite{PR} or \cite{KM}
1465: gives the unprojection $V\subset\PP^N$ and its anticanonical ring, and
1466: Papadakis \cite{P1}, Section~5 gives an explicit presentation of the
1467: ring.
1468:
1469: Tom and Jerry occur in hundreds of constructions of Gorenstein
1470: codimension~4 rings with $9\times16$ presentation, and seem to be
1471: related to the respective cones over $\PP^2\times\PP^2$ and over
1472: $\PP^1\times\PP^1\times\PP^1$ and their weighted homogeneous
1473: deformations; but exactly what this means remains to be elucidated. See
1474: \cite{Ki}, Section~8 for a more detailed discussion.
1475:
1476: To deal with our codimension~4 K3s and Fanos, we have to determine
1477: whether Tom and Jerry matrixes can be set up to give $V$ having only
1478: Mori category singularities. There are a few hundred problems here, and
1479: we have only just started working systematically on the Fanos. For
1480: each case where the numerical data of the K3 admits a Type~I
1481: projection, we can ask for a Tom or Jerry matrix -- from the
1482: experience of Alt{\i}nok's thesis \cite{A} and \cite{CPR}, we suspect
1483: that in each case at least one of Tom or Jerry exists, and that when
1484: both exist, the two families give isomorphic polarised K3s, so do not
1485: correspond to different irreducible components of moduli.
1486:
1487: When we ask the same question for Fanos 3-folds, one of several things
1488: may happen: either of the Tom and Jerry matrixes could exist to give a
1489: Mori Fano \hbox{3-fold}, or either could fail. We suspect that each of
1490: these possibilities happens in many numerical cases. When both Tom and
1491: Jerry exist, the resulting unprojected varieties may give essentially
1492: different different Fano 3-folds: see \cite{Ki}, Examples~6.4 and~6.8,
1493: based on Takagi's thesis, where Tom and Jerry unprojections with the
1494: same numerical data correspond to Fano 3-folds that are not biregular,
1495: and not obviously birational. Since we have several hundred cases to
1496: settle, and much of the calculation comes down to checking that
1497: monomials of suitable degree exist, we hope that most of the
1498: calculation can eventually be entrusted to the computer.
1499:
1500: \section{The K3 database in Magma}
1501: The program outlined in Sections~\ref{sec!Hi}--\ref{sec!proj}
1502: envisaged listing the many thousand possible values $g,\sB$ for the
1503: numerical data of families of K3s or Fano \hbox{3-folds}. For each
1504: $g,\sB$ in the list, we calculate the Hilbert series and try to deduce
1505: a plausible shape for the anticanonical ring and its possible
1506: pro\-jections. In some cases, we can establish the existence of a
1507: quasi-smooth surface, extension to Fano 3-fold, and connected
1508: components of moduli -- these are all questions that become nontrivial
1509: when $R(S,D)$ has codimension $\ge4$. These calculations are akin to
1510: traditional work on embedding curves and surfaces in Italian algebraic
1511: geometry, although rational contributions from quotient singularities
1512: add a certain spice, and unlimited possibilities for errors of
1513: arithmetic. With patience, any particular calculation can be done by
1514: hand, but automation has obvious advantages when doing these
1515: calculations on an industrial scale.
1516:
1517: Several items of this program have already been carried out for
1518: polarised K3 surfaces. Hilbert series methods apply to many other
1519: situations in algebraic geo\-metry involving graded rings or modules;
1520: current work in progress by students at Warwick includes Suzuki \cite{S}
1521: on Fano 3-folds of Fano index $f\ge2$, Keenan's project \cite{Ke} on
1522: subcanonical curves and Buckley's study \cite{Bu} of polarised
1523: Calabi--Yau \hbox{3-folds} with strictly canonical singularities.
1524: Nonetheless, rings over K3 surfaces have several advantages that make
1525: them an ideal target for a computer study: the numerical data and the
1526: RR statement are simple to state, there is no cohomology, and most K3
1527: surfaces have projections to smaller codimension. Moreover, work over
1528: the last 20 years (Fletcher \cite{Fl}, Alt{\i}nok \cite{A} and Corti,
1529: Pukhlikov and Reid \cite{CPR}) already provides us with several
1530: hundred worked out examples of what to expect.
1531:
1532: We have programmed these calculations as routines in John Cannon's
1533: computer algebra system Magma \cite{Ma}. The results of all the
1534: calculations in codimension $\le4$ are assembled as a database
1535: containing 391 families of K3 surfaces. We describe here a little of
1536: what our code does and how to use it. Although more sophisticated
1537: upgrades are projected (and implemented in part, see the Graded Ring
1538: Database Website \cite{Br1}), we restrict ourselves to the first
1539: working version of our routines, in the widely available export~2.8 of
1540: Magma \cite{Ma}. Please see \cite{Br1} for more recent work; this
1541: includes a prototype database of K3s up to codimension~10, and will
1542: document upgrades as they come on line.
1543:
1544: \begin{exa}[making a single surface]\label{exa!dbeg1}
1545: Suppose we want to study a polarised K3 surface $S,D$ with
1546: ineffective $D$ (that is, $g=-1$, see \ref{ssec!uh}), and having basket
1547: of singularities $\sB=\{\recip{2}(1,1),\recip{5}(1,4),
1548: \recip{13}(3,10)\}$. As usual, {\tt >} at the beginning of a line is
1549: the Magma prompt. Anything after this prompt is user input, while
1550: anything else is Magma output (occasionally subjected to a little
1551: editing for legibility). We have formatted the Magma input and output
1552: along the lines of our mathematical writings, and you should not have
1553: too much trouble interpreting the language.
1554: \begin{verbatim}
1555: > Q := Rationals();
1556: > R<t>:=PolynomialRing(Q); // Omit these lines at your peril.
1557: > B := [ [2,1], [5,1], [13,3] ];
1558: > S := K3Surface(-1,B); // We input genus -1 and Basket B.
1559: > S;
1560: Codimension 4 K3 surface with data
1561: Weights: [ 3, 4, 5, 6, 7, 10, 13 ]
1562: Numerator: t^48 - t^36 - t^35 - t^34 - t^33 - t^32 - t^31 - t^30
1563: + t^27 + 2*t^26 + 2*t^25 + 2*t^24 + 2*t^23 + 2*t^22 + t^21
1564: - t^18 - t^17 - t^16 - t^15 - t^14 - t^13 - t^12 + 1
1565: Basket: [ 2, 1 ], [ 5, 1 ], [ 13, 3 ]
1566: \end{verbatim}
1567: The computer has calculated the Hilbert series, and, after
1568: experimenting with a number of possibilities, has found
1569: \[
1570: (1-t^3)(1-t^4)(1-t^5)(1-t^6)(1-t^7)(1-t^{10})(1-t^{13})
1571: \]
1572: as a plausible denominator, corresponding to generators for the ring
1573: $R(S,D)$, or an embedding of $S$ in w.p.s.\
1574: $\PP=\PP^6(3,4,5,6,7,10,13)$. The printout lists the conclusion: a
1575: candidate family of codimension~4 K3 surfaces $S\subset\PP$ with the
1576: stated Hilbert numerator. We should be clear that in general
1577: \begin{quote}
1578: {\em there is no a priori guarantee that this or any other candidate
1579: surface proposed by our Magma functions exists as a subvariety in the
1580: indicated w.p.s.\ with the indicated properties.}
1581: \end{quote}
1582: After all, at this stage the computer has done nothing more than a
1583: formal game with denominators for Hilbert series. Having said that,
1584: assume for a moment that $S\subset\PP^6(3,4,5,6,7,10,13)$ really exists,
1585: polarised by the divisorial sheaf $\Oh_S(D)=\Oh_S(1)$. Clearly $D$ is
1586: not an effective divisor, since no $x_i$ has weight $a_i=1$, but its
1587: multiples do exist: in particular, $S$ has a single effective divisor
1588: in $|3D|$, $|4D|$, $|5D|$, a pencil $|6D|$, etc.
1589:
1590: The {\tt Numerator} is a polynomial $p=P(t)\prod(1 - t^{a_i})$, where
1591: $P(t)$ is the Hilbert series of $R(S,D)$
1592: \[
1593: P_S(t)=\sum_{n\ge0} h^0(S,nD)t^n = \frac{p}{\prod(1 - t^{a_i})}\,.
1594: \]
1595: Here the product is taken over the weights $a_i$ of the w.p.s. At first
1596: sight, the 7 negative terms $-t^{12}-t^{13}-\cdots$ in the numerator
1597: output by Magma might suggest that $S\subset\PP$ could have 7 equations
1598: in degrees 12, 13, 14, 15, 16, 17, 18. Gorenstein rings with
1599: $7\times12$ resolution certainly exist, so this is not completely
1600: excluded, but this structure appears very rarely (compare Papadakis and
1601: Reid \cite{PR}, 2.8); perhaps this candidate appears more exotic than it
1602: actually is. The $9\times 16$ format is ubiquitous, and experience
1603: suggests that this is really a $9\times 16$ resolution in disguise.
1604:
1605: How can we prove that this surface actually exists? The method of
1606: Alt{\i}nok's thesis \cite{A} is to look for a codimension~3 K3
1607: surface that is already known to exist, and could be a projection of
1608: the desired surface. Then we could go on to show that the
1609: codimension~3 surface can be made to contain an {\em unprojection
1610: divisor\/} as in \ref{ssec:TI} that can be contracted or {\em
1611: unprojected\/}, to give the codimension~4 surface. This second stage
1612: is the hard part, and we pass over it for the present (see \cite{PR},
1613: \cite{P1} and \cite{Ki}), although in time the computer will have
1614: something useful to add here too. One point is that as we write there
1615: are many kinds of projection for which the unprojection is still
1616: something of a mystery (see
1617: \cite{Ki}, Section~9). That is, the {\em theory} of unprojection is not
1618: quite complete enough yet to handle all the cases we might need.
1619:
1620: However, the K3 database is very good at finding projections of the
1621: types we already understand, such as the Type~I projections of
1622: Section~\ref{sec!proj}. These occur when we project from a cyclic
1623: quotient singularity $\recip{r}(a,r-a)$ whose local orbifold
1624: coordinates come from generators of $R(S,D)$ of weight $a,r-a$; a
1625: singularity with this property is a {\em Type~I centre}. In our
1626: example, we can see a possible Type~I centre, the singularity
1627: $\recip{13}(3,10)$. The point is that the local weights 3 and 10, a
1628: priori only defined mod~13, can arise as the global generators
1629: corresponding to the weights 3 and 10 of the w.p.s. All the other
1630: singularities fail this test, since the w.p.s.\ has no weight $1$.
1631: \begin{verbatim}
1632: > DB := K3Database("t"); /* Loads the DB, requiring "t" as
1633: variable in Hilbert series. This takes several seconds. */
1634: > Centres(~S,DB); // Searches DB for projection centres of S.
1635: > S;
1636: Codimension 4 K3 surface with data
1637: Weights: [ 3, 4, 5, 6, 7, 10, 13 ]
1638: Numerator: t^48 - t^36 - t^35 - t^34 - t^33 - t^32 - t^31 - t^30
1639: + t^27 + 2*t^26 + 2*t^25 + 2*t^24 + 2*t^23 + 2*t^22 + t^21
1640: - t^18 - t^17 - t^16 - t^15 - t^14 - t^13 - t^12 + 1
1641: Basket: [ 2, 1 ], [ 5, 1 ], [ 13, 3 ]
1642: Centre 1: [ 5, 1, 4 ] has Type 2 projection to 10 in codim 1
1643: Centre 2: [ 13, 3, 10 ] has Type 1 projection to 42 in codim 3
1644: \end{verbatim}
1645: This says that the singularity $\recip{13}(3,10)$ is a Type~I centre
1646: as expected. It has also found a Type~II centre, that we pass over for
1647: the moment. But it does more: Magma applies the calculus of
1648: Exercise~\ref{ex!t1} to predict the weights and basket of the image of
1649: a Type~I projection, and searches the database for surfaces with the
1650: right properties, finding the K3 surface numbered 42 in the database
1651: as a plausible image of the projection from this centre. (The internal
1652: numbering of items in the database is arbitrary, and may differ from
1653: session to session.) We ask what it is:
1654: \label{Altinok3(63)}
1655: \begin{verbatim}
1656: > K3Surface(DB,42);
1657: Codimension 3 K3 surface, number 42, Altinok3(63), with data
1658: Weights: [ 3, 4, 5, 6, 7, 10 ]
1659: Numerator: -t^35 + t^23 + t^22 + t^21 + t^20 + t^19
1660: - t^16 - t^15 - t^14 - t^13 - t^12 + 1
1661: Basket: [ 2, 1 ], [ 3, 1 ], [ 5, 1 ], [ 10, 3 ]
1662: \end{verbatim}
1663: Now if we know that the image codimension~3 surface
1664: $S'\subset\PP(3,4,5,6,7,10)$ exists, and contains the stratum
1665: $\PP(3,10)\subset\PP$ as unprojection divisor, then it has the
1666: required geometric properties to be unprojected, and we conclude from
1667: \cite{PR} that the original surface exists. Indeed, since we know a
1668: lot about unprojection, we can even write down the equations of the
1669: codimension~4 surface $S$ based on those of the codimension~3 surface
1670: $S'$ and its unprojection divisor $\PP(3,10)$. (Since $\PP(3,10)$ is
1671: the c.i.\ defined by $x_4=x_5=x_6=x_7=0$, and the unprojection
1672: variable $s$ has weight $13$, we verify that $S$ also needs equations
1673: in degree $19,20$, so our intuition that $S$ has a $9\times16$
1674: resolution was right.)
1675:
1676: The basic existence part is easy. The structure theorem for Gorenstein
1677: rings of codimension $\le3$ makes it easy for us to figure out
1678: equations. Alt{\i}nok goes further to check that a surface containing
1679: the unprojection divisor can be made. Using Alt{\i}nok's numbering in
1680: \cite{A1}, Tables~5.1 and~5.2, the codimension~3 surface is
1681: \Alt{3}{63}, as the database tells us, while the codimension~4 surface
1682: is \Alt{4}{104}.
1683:
1684: Of course, the projection calculus is easy to do by hand: we could
1685: simply have searched the lists of \cite{Fl} and \cite{A} to find this
1686: surface (as we always did in the dark ages before the great Graded Ring
1687: Database Revolution). For that matter, we could have just called for a
1688: surface having the desired basket and again found this one without
1689: trouble. At this stage, we have not gained much by having the results
1690: of these calculations in a database. The next section contains a
1691: calculation for which the database is essential.
1692: \end{exa}
1693:
1694: \begin{exa}[searching the database]
1695: As databases go, our current database of 391 K3s is tiny. Even so,
1696: its key advantage over a printed list is that we can search it quickly
1697: and accurately. Suppose, for example, that we are interested in finding
1698: out whether our lists contain a K3 having a singularity of index~17.
1699: In this context, 17 is a large number, taking the sublattice
1700: $L(g,\sB)\subset\Pic S$ close to the maximum that can live in the K3
1701: lattice $\LK$ (see \ref{sssec!K3L}). To do this search in Magma, we
1702: first summon the database as usual:
1703: \begin{verbatim}
1704: > DB := K3Database("t");
1705: > #DB; // # returns the number of elements of a list.
1706: 391 \end{verbatim}
1707: Singularities are denoted {\tt p = [r,a]}, with $a$ coprime to the
1708: index {\tt r = p[1]} and $a<r/2$. So we must search the database for
1709: surfaces having a singularity $[r,a]\in\sB$ having $r=17$.
1710: \begin{verbatim}
1711: > surfaces := [ S : S in DB | &or[ p[1] eq 17 : p in Basket(S) ] ];
1712: > #surfaces;
1713: 2
1714: \end{verbatim}
1715: Note that {\tt \&or} taken over a list of Boolean values is equivalent
1716: to ``there exists $p$ in $\sB$ with $p[1]$ equals 17''. The last line
1717: says that there are just two surfaces in DB with a singularity of
1718: index~17. We ask to see them.
1719: \begin{verbatim}
1720: > for S in surfaces do print S;
1721: > end for;
1722: Codimension 2 K3 surface, number 1, Fletcher2(82), with data
1723: Weights: [ 3, 4, 7, 10, 17 ]
1724: Numerator: t^41 - t^21 - t^20 + 1
1725: Basket: [ 2, 1 ], [ 17, 7 ]
1726: Codimension 4 K3 surface, number 250, Altinok4(79), with data
1727: Weights: [ 2, 3, 5, 5, 7, 12, 17 ]
1728: Numerator: t^51 - t^41 - t^39 - t^37 - t^36
1729: + t^29 + t^27 + t^26 + t^25 + t^24 + t^22
1730: - t^15 - t^14 - t^12 - t^10 + 1
1731: Basket: [ 17, 5 ]
1732: \end{verbatim}
1733:
1734: Again, we want to know that these surfaces exist. The codimension~2
1735: surface is Fletcher \cite{Fl}, List~13.8, no.~82, and it is easy to
1736: write down its equations.
1737:
1738: For the codimension~4 surface, the technique of the previous example
1739: locates an image of projection from the Type~I centre
1740: $\recip{17}(5,12)$ in the database, and we can construct $S$ by
1741: unprojection as before. This codimension~4 example $S$, called
1742: \Alt{4}{79} is interesting in that it has exactly the same Hilbert
1743: series as the codimension~4 c.i.\ $T_{10,12,14,15}$; however, none of
1744: the equations of the c.i.\ $T$ can involve the last variable $u_{17}$,
1745: so that $T$ is a kind of weighted cone over a curve. Deformations of
1746: polarised K3 surfaces are unobstructed, as are c.i.s, so that having
1747: proved that $S$ exists, we have found two different irreducible
1748: components of the Hilbert scheme, one containing $S$, the other
1749: containing $T$.
1750: \end{exa}
1751:
1752: \subsection{K3 database functions in Magma}
1753:
1754: As with much computer algebra, just a few internal functions do the
1755: work, and a lot of the rest is cosmetic renaming to simplify the
1756: user's access to these. We give a brief and possibly inadequate
1757: description of some core functions to give some idea of what is
1758: happening inside the computer; these remarks can also serve as a first
1759: introduction to the code for anybody wishing to modify it for use in
1760: other contexts.
1761:
1762: \subsubsection{Listing all baskets} The function
1763: \begin{verbatim}
1764: > Baskets(n);
1765: \end{verbatim}
1766: returns a list of all possible baskets $\sB=\{\recip{r}(a,r-a)\}$ for
1767: K3 surfaces of singular rank $\sum(r-1)<n$. For example, the following
1768: commands generates a list of all baskets with $\sum(r-1) = 11$, and
1769: asks for the 44th basket in the list.
1770: \begin{verbatim}
1771: > BB, gg := Baskets(12); /* BB is the list of baskets,
1772: gg the parallel list of minimal genera */
1773: Actual number: 329 ... Checking degrees ...
1774: > BB11 := [ B : B in BB | not (B eq [])
1775: and &+[p[1]-1 : p in B] eq 11 ]; // &+ is sum over set.
1776: > #BB11;
1777: 109
1778: > b := BB11[44];
1779: > b, gg[Index(BB,b)];
1780: [ [ 4, 1 ], [ 5, 2 ], [ 5, 2 ] ]
1781: 0
1782: > Degree(-1,b);
1783: -17/20
1784: > Degree(0,b);
1785: 23/20
1786: \end{verbatim}
1787:
1788: Recall from \ref{sssec!ineq} that the possible baskets $\sB$ on a K3
1789: surface are limited by two inequalities on the pairs $r,a$. The first
1790: is given by the rank of the Picard group: $\sum(r-1)<n=20$. The
1791: argument $n$ of {\tt Baskets(n)} is the bound (we eventually set
1792: $n=20$ for complete lists). The other inequality, saying that the
1793: polarised surface $S,D$ has degree $D^2>0$, involves the genus $g$ of
1794: $S$. If $g=-1$ or $0$, a basket may give $D^2\le0$. The second return
1795: value of {\tt Baskets} (called $gg$ in the first line above) is a
1796: parallel sequence of minimal genera that give $D^2>0$. In the above
1797: example, we checked that $g=-1$ is illegal for the 44th basket,
1798: because it gives $D^2=-17/20$.
1799:
1800: The numerical data $g,\sB$ with $D^2>0$ determines a Hilbert series
1801: $P(t)=\sum P_nt^n$. If it actually corresponds to a K3 surface,
1802: then vanishing implies that $P_n\ge0$ for all $n\ge1$. However, this
1803: does not follow from the inequality $D^2>0$. Indeed, there exist
1804: numerical data $g,\sB$ with a negative coefficient $P_n$; curiously,
1805: there are just three of these eccentrics (compare \cite{CPR}, 7.9).
1806: The function {\tt Baskets} also checks that $P_n\ge0$ for the first 20
1807: coefficients before confirming the minimum genus $g$. When $n=20$, the
1808: result is all 6640 possible baskets, together with, for each, the
1809: minimum genus that makes the Hilbert series positive.
1810:
1811: \subsubsection{{\tt HilbertSeries(g,B)}, etc.} The Magma function
1812: \begin{verbatim}
1813: > HilbertSeries(g,B);
1814: \end{verbatim}
1815: has the formula (\ref{eq!K3-s}) built in, and simply evaluates it at
1816: the data $g,\sB$ as a rational function in $t$. For example,
1817: \begin{verbatim}
1818: > R<t>:=RationalFunctionField(Q); // Require "t" as variable.
1819: > P:=HilbertSeries(-1,[[ 3, 1 ], [ 4, 1 ], [ 11, 2 ]]);
1820: > P*(1-t^2)*(1-t^3)*(1-t^4)*(1-t^11);
1821: -t^20 - t^15 - t^13 - t^11 + t^9 + t^7 + t^5 + 1
1822: \end{verbatim}
1823: suggests the candidate surface $S\subset\PP(2,3,4,5,7,9,11)$ with the
1824: stated $g$ and $\sB$. (To make this work for other types of graded
1825: rings, you need to figure out what Hilbert series you want to use, and
1826: program it in as a substitute for our formula (\ref{eq!K3-s}).)
1827:
1828: In the above example, we put in the denominator by hand, and this
1829: turned out to be a lucky guess. The heart of the whole package is a
1830: suite of functions to make a reasonable analysis of this Hilbert
1831: series, using a few tricks based on the experience of Alt{\i}nok's
1832: thesis \cite{A}. One expects generators corresponding to positive
1833: terms early on in the Hilbert series, but this ceases to be logically
1834: reliable once relations appear. For each singularity $\recip{r}(a,r-a)$
1835: in the basket, there must be generators of weight divisible by $r$ to
1836: cancel the periodicity in the Hilbert series, and a generator of
1837: weight $\equiv a$ and $-a$ to provide local orbifold coordinates at
1838: the singularity. We usually expect a generator in each degree $r$ as
1839: in the above example, but sometimes a GCD of two different $r$ can
1840: account for two at one go.
1841:
1842: Deploying these tricks is something of an art, and there is no point
1843: in being precise here about how we do it. Experience suggests that
1844: they need to be combined in different ways in different situations (we
1845: find different and better ways of putting in the generators
1846: systematically each time we extend the database). But however we use
1847: the numerical tricks on the Hilbert series, there will usually be work
1848: left to do. Particular features of the geometry may demand extra
1849: generators not predicted by the numerical data. For K3 surfaces in
1850: higher codimension, the existence of particular linear systems often
1851: forces us to include extra generators. One reason for stopping at
1852: codimension~4 with the current version of the Magma database is that
1853: there is little extra work in that case.
1854:
1855: \subsubsection{{\tt MakeK3Database(BB,gg,cmax)}}
1856:
1857: This function takes as its arguments a list of baskets $BB$ with list
1858: of minimal genus $gg$, and an integer {\tt cmax}. It takes each basket
1859: and genus pair in turn and applies any analysis of the Hilbert series
1860: that is implemented. If the codimension gets larger than {\tt cmax},
1861: the result is discarded (in the current implementation --- this is
1862: handled in a more useful way in the proto\-type future implementation).
1863: Once done, the result is wrapped in cosmetics and available for computer
1864: study.
1865:
1866: Of course, as we learn more tricks for analysing Hilbert series, we can
1867: run them through the database, modifying candidate surfaces as we go.
1868: But we believe that the current Magma database of 391 K3 surfaces is
1869: reliable and complete. (For that matter, we are fairly confident of the
1870: codimension~5 list in \cite{Br1}, which has $N_5=162$ elements.)
1871:
1872: \subsubsection{{\tt Centres($\sim$DB)}, etc.}
1873:
1874: There are several functions used to study the database by hand. The
1875: procedure
1876: \begin{verbatim}
1877: > DB := K3Database("t");
1878: > Centres(~DB); // Takes some minutes.
1879: \end{verbatim}
1880: performs the projection calculus of Exercise~\ref{ex!t1} to compute all
1881: possible projections between the surfaces of the database {\tt DB}; the
1882: tilde indicates that the command is {\em procedural\/}, that is, it
1883: actually modifies {\tt DB} by writing in the centres it computes (the
1884: next version of the database will have the centres ready for use on
1885: loading). Once the centres are in, we can look at the projections from
1886: any surface. For example, we can find the surface $S$ of \ref{exa!dbeg1}
1887: in the database and then look for all possible iterated projections
1888: from $S$:
1889: \begin{verbatim}
1890: > S := K3SurfaceFromWeights(DB,[3,4,5,6,7,10,13]);
1891: > pc := ProjectionChains(S,DB); #pc;
1892: 4
1893: > pc[1];
1894: [
1895: [ [ 254, 4 ], [ 5, 1, 4 ], [ 10 ] ],
1896: [ [ 10, 1 ], [ 13, 3, 10 ], [ 40 ] ],
1897: [ [ 40, 1 ], [ 10, 3, 7 ], [ 79 ] ],
1898: [ [ 79, 1 ] ]
1899: ]
1900: \end{verbatim}
1901: This output is opaque (Leitmotif: please buy the upgrade, with its
1902: many major enhancements), but you can probably see what is going on.
1903: There are 4 chains of projections starting with $S$, and {\tt pc[1]}
1904: only asks for the first. If we denote the $n$th surface in the database
1905: by $S_n$, then $S=S_{254}$ and the first chain of projections, returned
1906: by {\tt pc[1]}, is
1907: \[
1908: S\broken S_{10}\broken S_{40}\broken S_{79}.
1909: \]
1910: The output records the codimension of each surface. In this case the
1911: final three surfaces are all codimension 1: the projections between
1912: them are the quadratic involutions of \cite{CPR}. The second column
1913: gives the type of centre {\tt [r,a,r-a]} of each projection.
1914:
1915: The fourth chain of projections is
1916: \begin{verbatim}
1917: > pc[4];
1918: [
1919: [ [ 254, 4 ], [ 13, 3, 10 ], [ 42 ] ],
1920: [ [ 42, 3 ], [ 10, 3, 7 ], [ 81 ] ],
1921: [ [ 81, 2 ], [ 7, 3, 4 ], [ 107 ] ],
1922: [ [ 107, 1 ] ]
1923: ]
1924: \end{verbatim}
1925: It consists purely of Type~I projections. In terms of homogeneous
1926: coordinate rings, at each stage a single variable (of weight 13, 10, 7
1927: respectively, the index of the corresponding centre) is eliminated from
1928: the ring. A basic exercise in Magma makes this very clear:
1929: \begin{verbatim}
1930: > [ Weights(DB[i]) : i in [254,42,81,107] ];
1931: [
1932: [ 3, 4, 5, 6, 7, 10, 13 ],
1933: [ 3, 4, 5, 6, 7, 10 ],
1934: [ 3, 4, 5, 6, 7 ],
1935: [ 3, 4, 5, 6 ]
1936: ]
1937: \end{verbatim}
1938: terminating with the surface $S_{107}$ which is
1939: \begin{verbatim}
1940: > DB[107];
1941: Codimension 1 K3 surface, number 107, Reid1(39), with data
1942: Weights: [ 3, 4, 5, 6 ]
1943: Numerator: -t^18 + 1
1944: Basket: [ 2, 1 ], [ 3, 1 ], [ 3, 1 ], [ 3, 1 ], [ 4, 1 ], [ 5, 1 ]
1945: \end{verbatim}
1946: \begin{exc} Write down a hypersurface $S_{18}\subset\PP(3,4,5,6)$
1947: that contains $\PP(3,4)$ as the unprojection divisor of the final
1948: projection, and singularities on it to provide the inverses of
1949: the successive unprojections.
1950: \end{exc}
1951:
1952: \begin{thebibliography}{CPR}
1953:
1954: \bibitem[A]{A} S. Alt{\i}nok, Graded rings corresponding to polarised
1955: K3 surfaces and $\Q$-Fano 3-folds, Univ. of Warwick PhD thesis, Sep.
1956: 1998, $93+\mathrm{vii~pp.}$, get from
1957: www.maths.warwick.ac.uk/\linebreak[2]$\!\sim$miles/doctors/Selma
1958:
1959: \bibitem[A1]{A1} S. Alt{\i}nok, Hilbert series and applications to
1960: graded rings, submitted
1961:
1962: \bibitem[Be]{Be} Sarah-Marie Belcastro, Picard lattices of families of
1963: K3 surfaces, to appear in Communications in Algebra, preprint
1964: math.AG/9809008, 21 pp.
1965:
1966: \bibitem[Br]{Br} Gavin Brown, Datagraphs in algebraic geometry,
1967: submitted to Proceedings of SNSC '01 (Linz), Ed. F. Winkler
1968:
1969: \bibitem[GRDW]{Br1} Gavin Brown, Magma algebraic geometry database
1970: website, see
1971: www.maths.\linebreak[2]warwick.\linebreak[2]ac.uk/$\!\sim$gavinb/grdb.html
1972:
1973: \bibitem[Bu]{Bu} Anita Buckley, Graded rings over polarised
1974: Calabi--Yau 3-folds, work in progress
1975:
1976: \bibitem[CPR]{CPR} A. Corti, A. Pukhlikov and M. Reid, Birationally
1977: rigid Fano hypersurfaces, in Explicit birational geometry of 3-folds,
1978: A. Corti and M. Reid (eds.), CUP 2000, 175--258
1979:
1980: \bibitem[CR]{CR} A. Corti and M. Reid, Weighted Grassmannians, first
1981: draft submitted to Proc. Francia memorial conference (Genova, 2001),
1982: currently 12~pp.
1983:
1984: \bibitem[D]{D} I. Dolgachev, Weighted projective varieties, in Group
1985: actions and vector fields (Vancouver, B.C., 1981), LNM{\bf956},
1986: pp.~34--71
1987:
1988: \bibitem[Fl]{Fl} A.R. Iano-Fletcher, Working with weighted complete
1989: intersections, in Explicit birational geometry of 3-folds, CUP 2000,
1990: pp.~101--173
1991:
1992: \bibitem[GW]{GW} S. Goto and K-i Watanabe, On graded rings. I, J. Math.
1993: Soc. Japan {\bf30} (1978) 179--213
1994:
1995: \bibitem[G]{G} M. Gross, Deforming Calabi-Yau threefolds, Math. Ann.
1996: {\bf308} (1997) 187--220
1997:
1998: \bibitem[I1]{I1} V. A. Iskovskikh, Fano threefolds. I, Izv. Akad. Nauk
1999: SSSR Ser. Mat. {\bf41} (1977) 516--562 = Math. USSR--Izv. {\bf11}
2000: (1977) 485--527
2001:
2002: \bibitem[I2]{I2} V. A. Iskovskikh, Fano threefolds. II, Izv. Akad. Nauk
2003: SSSR Ser. Mat. {\bf42} (1978) 506--549 = Math. USSR--Izv. {\bf12} (1978)
2004: 469--506
2005:
2006: \bibitem[Ka1]{Ka1} KAWAMATA Yujiro, On the plurigenera of minimal
2007: algebraic 3-folds with $K\approx0$, Math. Ann. {\bf275} (1986) 539--546
2008:
2009: \bibitem[Ka2]{Ka2} KAWAMATA Yujiro, Boundedness of $\Q$-Fano
2010: threefolds, in Proc. Internat. Conference on Algebra (Novosibirsk,
2011: 1989), Contemp. Math. {\bf131}, AMS, 1992, Part~3, pp.~439--445
2012:
2013: \bibitem[Ke]{Ke} Adam Keenan, Riemann--Roch, subcanonical divisors on
2014: curves and graded rings, work in progress
2015:
2016: \bibitem[KM]{KM} A. Kustin and M. Miller, Constructing big Gorenstein
2017: ideals from small ones, J. Algebra {\bf85} (1983) 303--322
2018:
2019: \bibitem[Leng]{Leng} R. Leng, McKay correspondence and equivariant
2020: Riemann--Roch, Warwick PhD thesis in preparation
2021:
2022: \bibitem[Ma]{Ma} Magma (John Cannon's computer algebra system): W.
2023: Bosma, J. Cannon and C.~Playoust, The Magma algebra system I: The user
2024: language, J. Symb. Comp. {\bf24} (1997) 235--265. See also
2025: www.maths.\linebreak[2]usyd.edu.au:8000/u/magma
2026:
2027: \bibitem[Mi1]{Mi1} MINAGAWA Tatsuhiro, Deformations of $\Q$-Calabi--Yau
2028: 3-folds and $\Q$-Fano 3-folds of Fano index~1, math.AG/9905106, J.
2029: Math. Sci. Univ. Tokyo Tokyo {\bf6} (1999) 397-414
2030:
2031: \bibitem[Mi2]{Mi2} MINAGAWA Tatsuhiro, Deformations of weak Fano
2032: 3-folds with only terminal singularities, math.AG/9905107, Osaka J.
2033: Math. {\bf38} (2001), to appear
2034:
2035: \bibitem[MM1]{MM1} MORI Shigefumi and MUKAI Shigeru, Classification of
2036: Fano 3-folds with $B_2\ge2$, Manuscripta Math. {\bf36} (1981/82)
2037: 147--162
2038:
2039: \bibitem[MM2]{MM2} MORI Shigefumi and MUKAI Shigeru, On Fano $3$-folds
2040: with $B_2\ge2$, in Algebraic varieties and analytic varieties
2041: (Tokyo, 1981), Adv. Stud. Pure Math. {\bf1}, Kinokuniya and
2042: North-Holland, 1983, pp.~101--129
2043:
2044: \bibitem[Mu1]{Mu1} MUKAI Shigeru, Curves and Grassmannians, in
2045: Algebraic geometry and related topics (Inchon, 1992), Internat. Press,
2046: Cambridge MA 1993, pp.~19--40
2047:
2048: \bibitem[Mu2]{Mu2} Mukai Shigeru, Curves and symmetric spaces. I,
2049: Amer. J. Math. {\bf117} (1995) 1627--1644
2050:
2051: \bibitem[N]{N} NAMIKAWA Yoshinori, Smoothing Fano $3$-folds, J. alg
2052: geom {\bf6} (1997) 307--324
2053:
2054: \bibitem[NS]{NS} NAMIKAWA Yoshinori and J.H.M. Steenbrink, Global
2055: smoothing of Calabi-Yau threefolds, Invent. Math. {\bf122} (1995)
2056: 403--419
2057:
2058: \bibitem[P]{P} Stavros Papadakis, Gorenstein rings and Kustin--Miller
2059: unprojection, Univ. of Warwick PhD thesis, Aug 2001, vi + 72~pp., get
2060: from
2061: www.maths.warwick.ac.uk/$\!\sim$miles/doctors/\linebreak[2]Stavros
2062:
2063: \bibitem[P1]{P1}
2064: Stavros Papadakis, Kustin--Miller unprojection {\em with} complexes,
2065: submitted, math.AG/\linebreak[2]0111195, 23 pp.
2066:
2067: \bibitem[PR]{PR} Stavros Papadakis and Miles Reid, Kustin--Miller
2068: unprojection without complexes, J. algebraic geometry (to appear),
2069: preprint math.AG/0011094, 18~pp.
2070:
2071: \bibitem[YPG]{YPG} M. Reid, Young person's guide to canonical
2072: singularities, in Algebraic geometry (Bowdoin, 1985), Proc. Sympos. Pure
2073: Math. {\bf46} Part~1, AMS 1987, pp.~345--414,
2074:
2075: \bibitem[R1]{R1} M. Reid, Projective morphisms according to Kawamata,
2076: Warwick preprint, 1983, unpublished, get from
2077: www.maths.warwick.\linebreak[2]ac.uk/$\!\sim$miles/3folds
2078:
2079: \bibitem[Ki]{Ki} M. Reid, Graded rings and birational geometry, in
2080: Proc. of algebraic geo\-metry symposium (Kinosaki, Oct 2000), K. Ohno
2081: (Ed.), 1--72, get from www.maths.warwick.ac.uk/\linebreak[2]$\!\sim$miles/3folds
2082:
2083: \bibitem[S]{S} K. Suzuki, Univ. of Tokyo PhD thesis, in preparation
2084:
2085: \bibitem[T4]{T4} M. Reid, Examples of Type~IV unprojection, preprint
2086: math.AG/\linebreak[2]0108037, 16~pp.
2087:
2088: \bibitem[T]{T} TAKAGI Hiromichi, On the classification of $\Q$-Fano
2089: 3-folds of Gorenstein index~2. I, II, RIMS preprint 1305, Nov 2000,
2090: 66~pp.
2091:
2092: \bibitem[T1]{T1} TAKAGI Hiromichi, work in progress
2093:
2094: \bibitem[W]{W} WATANABE Kei-ichi, Some remarks concerning Demazure's
2095: construction of normal graded rings, Nagoya Math. J. {\bf 83} (1981)
2096: 203--211
2097:
2098: \end{thebibliography}
2099:
2100: \end{document}
2101:
2102: scrap
2103:
2104: 0. Biblio info:
2105:
2106: Selma Alt{\i}nok, Gavin Brown and Miles Reid, Fano 3-folds, K3 surfaces
2107: and graded rings, in Singapore International Symposium in Topology and
2108: Geometry (NUS, 2001), Edited by: A. J. Berrick, M. C. Leung and X. W.
2109: Xu, to appear Contemp. Math. AMS, 2002, preprint currently 28 pp.
2110:
2111: 1. Referee's instructions:
2112:
2113: a. The authors could also mention in Section 2.2 the result in
2114: Mori-Mukai's paper classifying Fano manifolds of dimension 3 with
2115: Picard number at least 2. DONE
2116:
2117: b. For readers' convenience (without having to refer to other
2118: papers) authors may mention in Section 4.2 how type D or E
2119: singularities are deformed (or replaced) by type A singularities to
2120: form the basket B. NOTED in footnote. The referee has missed a small
2121: point -- already the type A hyperquotient singularities have to be
2122: deformed to cyclic quotients
2123:
2124: c. Dolgachev's article on weighted projective spaces may be added as a
2125: reference. DONE
2126:
2127: