\(G\)-Fano threefolds, I


Abstract

We classify Fano threefolds with only terminal singularities whose canonical class is Cartier and divisible by \(2\), and satisfying an additional assumption that the \(G\)-invariant part of the Weil divisor class group is of rank \(1\) with respect to an action of some group \(G\). In particular, we find a lot of examples of Fano \(3\)-folds with “many” symmetries.

1 Introduction.↩︎

2. Let \(Y\) be an algebraic variety \(X\) over a field \(\Bbbk\) and let \(G\) be a group. Following works of Yu. I. Manin [1] we say that \(X\) is a \(G\)-variety if the group \(G\) acts on \(\bar X:=X\otimes_\Bbbk \bar \Bbbk\), where \(\bar \Bbbk\) is the algebraic closure of \(\Bbbk\). Moreover, we assume that \(X\), \(G\) and \(\Bbbk\) satisfy one of the following two conditions.

(a) Algebraic case. \(G\) is the Galois group of \(\bar \Bbbk\) over \(\Bbbk\) acting on \(\bar X\) through the second factor. The action of \(G\) on \(X\) is trivial.

(b) Geometric case. The field \(\Bbbk\) is algebraically closed, \(G\) is a finite group, and the action of \(G\) on \(X\) is given by a homomorphism \(G\to \operatorname{Aut}_{\Bbbk}(X)\).

A \(G\)-morphism (resp. rational \(G\)-map) of \(G\)-varieties is a \(\Bbbk\)-morphism (resp. \(\Bbbk\)-rational map) commuting with the action of \(G\) in the geometric case. A projective \(G\)-morphism \(f: X\to Z\) of normal \(G\)-varieties is called \(G\)-Mori fiber space if \(X\) has at worst terminal \(G\mathbb{Q}\)-factorial singularities (see 12), \(f_*\mathscr{O}_X=\mathscr{O}_Z\), the relative invariant Picard group \(\operatorname{Pic}(X/Z)^G\) is of rank \(1\), and the anticanonical divisor \(-K_X\) is \(f\)-ample. In the particular case where \(\dim Z=0\), \(X\) is called a \(G\mathbb{Q}\)-Fano variety. If furthermore the canonical divisor is Cartier, then we say that \(X\) is \(G\)-Fano variety.

Throughout this paper we assume that the ground field has characteristic \(0\). The following is an easy consequence of the Minimal Model Program [2] (cf. [3]).

1. Proposition. Let \(V\) be a \(G\)-variety of dimension \(\le 3\). The following are equivalent:

  1. \(\kappa(V)=-\infty\),

  2. \(V\) is geometrically uniruled,

  3. \(V\) is \(G\)-birationally isomorphic to a variety \(X\) having a structure of \(G\)-Mori fiber space.

Birational classification of \(G\)-surfaces is developed very well [1], [4]. In this and subsequent papers we consider \(G\)-Mori fiber spaces \(X\to Z\), where \(\dim X=3\) and \(Z\) is a point, i.e. the case of \(G\mathbb{Q}\)-Fano threefolds.

3. Let \(X\) be a \(G\)-Fano threefold. It is well-known that \(\operatorname{Pic}(X)\) is a finitely generated torsion free abelian group (see, e.g. [5]). Consider the following composed object: \[\operatorname{V}(X)=\bigl(\operatorname{Cl}(X),\operatorname{Pic}(X),K_X,(\phantom{v},\phantom{v},\phantom{v})\bigr),\] where \(\operatorname{Pic}(X)\) is regarded as a sublattice of \(\operatorname{Cl}(X)\), \(K_X\in \operatorname{Pic}(X)\) is the canonical class of \(X\), and \((\phantom{v},\phantom{v},\phantom{v})\) is the intersection form \(\operatorname{Pic}(X)\times \operatorname{Pic}(X)\times\operatorname{Cl}(X)\to \mathbb{Z}\). Since the singularities of \(X\) are isolated cDV [6], \(\operatorname{Pic}(X)\) is a primitive sublattice in \(\operatorname{Cl}(X)\), i.e. the quotient \(\operatorname{Cl}(X)/\operatorname{Pic}(X)\) is torsion free [7]. Moreover, since \(\rho(X)^G=1\), we have \[\text{\operatorname{Cl}(X)^G is a subgroup of rank 1 containing K_X.}\]

4. In this paper we give a classification of one class of Gorenstein \(G\)-Fano threefolds. More precisely, we consider Fano threefolds such that \(-K_X=2S\) for some ample Cartier divisor \(S\). Then \(X\) is called a del Pezzo threefold (see 14). Smooth del Pezzo threefolds were classified by Iskovskikh [8], see also [9], [5]. Singular ones were discussed from different points of view in many works [10], [11], [12], [13], [14]. We are interested basically in group actions on terminal del Pezzo threefolds \(X\) and the structure of the lattice \(\operatorname{Cl}(X)\).

5. Let \(S\) be a smooth del Pezzo surface of degree \(d:=K_S^2\). Then we have \(\operatorname{Pic}(S) = \mathbb{Z}^{10-d}\). Define \[\Delta:=\{\alpha \in \operatorname{Pic}(S) \mid \alpha^2=-2,\quad \alpha\cdot K_S=0\}.\] Then \(\Delta\) is a root system in \((K_S)^{\perp}\otimes \mathbb{R}\). Depending on \(d\), \(\Delta\) is of the following type ([15]):

d 1 2 3 4 5 6 7 \(8'\) \(8''\)
\(\Delta\) \(E_8\) \(E_7\) \(E_6\) \(D_5\) \(A_4\) \(A_1\times A_2\) \(-\) \(A_1\) \(-\)

where \(8'\) (resp. \(8''\)) corresponds to \(\mathbb{P}^1\times\mathbb{P}^1\) (resp. Hirzebruch surface \(\mathbb{F}_1\)).

6. Now let \(X\) be a del Pezzo threefold. Let \(S\in |{-}\frac{1}{2}K_X|\) be a smooth member [11] and let \(\iota: S \hookrightarrow X\) be the natural embedding. Then \(S\) is a del Pezzo surface of degree \(d=-\frac{1}{8} K_X^3\). It is easy to show that the restriction map \(\iota^*: \operatorname{Cl}(X)\to \operatorname{Pic}(S)\) is injective and its cokernel is torsion free (see 27). Define the following subsets in \(\Delta\subset\operatorname{Pic}(S)\): \[\begin{array}{lll} \Delta'&:=&\{\alpha \in \operatorname{Pic}(S) \mid\alpha^2=-2,\;\alpha\cdot K_S= \alpha\cdot \iota^*\operatorname{Cl}(X)=0\}, \\[5pt] \Delta''&:=&\{\alpha \in \iota^*\operatorname{Cl}(X) \mid\alpha^2=-2,\;\alpha\cdot K_S=0 \}. \end{array}\] In other words, \[\Delta'=\Delta\cap (\iota^*\operatorname{Cl}(X))^{\perp}, \qquad \Delta''=\Delta\cap \iota^*\operatorname{Cl}(X).\] If \(\Delta'\) (resp. \(\Delta''\)) is non-empty, then it is a root subsystem in \(\Delta\). Assume that \(X\) is a \(G\)-variety. Then the group \(G\) naturally acts on \(\iota^*\operatorname{Cl}(X)\) and \(\Delta''\) preserving the class of \(K_S\) and the intersection form. Our classification of \(G\)-del Pezzo threefolds is by types of root systems \(\Delta'\) and \(\Delta''\).

7. Theorem.Let \(X\) be a \(G\)-del Pezzo threefold and let \(\operatorname{d}(X):=-\frac{1}{8}K_X^3\). There are the following possibilities:

\(\operatorname{r}\) \(X\) \(\bar X\) \(Z\) \(\Delta'\) \(\Delta''\) \(\operatorname{p}\) \(\operatorname{s}\)
\(\operatorname{r}\) \(X\) \(\bar X\) \(Z\) \(\Delta'\) \(\Delta''\) \(\operatorname{p}\) \(\operatorname{s}\)
\(\operatorname{d}(X)=1\)
\(1\) \(V_{1}\) \(-\) \(\operatorname{pt}\) \(E_8\) \(-\) \(0\) \(21-\operatorname{h}\)
\(2\) [rkCl612-deg1-quadric] \(-\) \(\mathbb{P}^{1}\) \(D_7\) \(-\) \(0\) \(22-\operatorname{h}\)
\(2\) [rkCl612-P1-bundle-degree611] \(-\) \(\mathbb{P}^{2}\) \(A_7\) \(-\) \(0\) \(22\)
\(2\) [rkCl612-deg1-birational] \(V_{2}\) \(\operatorname{pt}\) \(E_7\) \(A_{1}\) \(2\) \(22-\operatorname{h}\), \(\operatorname{h}\le10\)
\(3\) [rkCl612-deg2-quadric] \(\mathbb{P}^{1}\) \(D_6\) \(2A_1\) \(4\) \(23-\operatorname{h}\)
\(3\) [rkCl612-P1-bundle-degree612] \(\mathbb{P}^{2}\) \(A_6\) \(A_{1}\) \(2\) \(23\)
\(3\) \(V_{3}\) \(\operatorname{pt}\) \(E_6\) \(A_2\) \(6\) \(23-\operatorname{h}\)
\(4\) [rkCl612-P1-bundle-degree613] \(\mathbb{P}^{2}\) \(A_5\) \(A_1\times A_2\) \(8\) \(24\)
\(4\) \(V_{4}\) \(\operatorname{pt}\) \(D_5\) \(A_3\) \(12\) \(24-\operatorname{h}\), \(\operatorname{h}\le 2\)
\(5\) [rkCl612-deg4-quadric] \(\mathbb{P}^{1}\) \(D_4\) \(D_4\) \(24\) \(25-\operatorname{h}\)
\(5\) \(V_{5}\) \(\operatorname{pt}\) \(A_4\) \(A_4\) \(20\) \(25\)
\(6\) [rkCl612-degree615] \(\mathbb{P}^{2}\) \(A_3\) \(D_5\) \(40\) \(26\)
\(7\) 62 \(V_{6}\) \(\mathbb{P}^{2}\) \(A_2\) \(E_6\) \(72\) \(27\)
\(8\) 60 \(\mathbb{P}^{3}\) \(\operatorname{pt}\) \(A_{1}\) \(E_{7}\) \(126\) \(28\)
\(\operatorname{d}(X)=2\)
\(1\) \(V_{2}\) \(-\) \(\operatorname{pt}\) \(E_7\) \(-\) \(0\) \(10-\operatorname{h}\)
\(2\) [rkCl612-deg2-quadric] \(-\) \(\mathbb{P}^{1}\) \(D_6\) \(A_{1}\) \(0\) \(11-\operatorname{h}\)
\(2\) [rkCl612-P1-bundle-degree612] \(-\) \(\mathbb{P}^{2}\) \(A_6\) \(-\) \(0\) \(11\)
\(2\) [rkCl612-deg2-birational] \(V_{3}\) \(\operatorname{pt}\) \(E_6\) \(-\) \(2\) \(11-\operatorname{h}\), \(\operatorname{h}\le 5\)
\(3\) 31[primitive-rho613-d612] \(-\) \((\mathbb{P}^1)^2\) \(A_5\) \(A_2\) \(0\) \(12\)
\(3\) [rkCl612-P1-bundle-degree613] \(\mathbb{P}^{2}\) \(A_5\) \(A_{1}\) \(2\) \(12\)
\(3\) \(V_{4}\) \(\operatorname{pt}\) \(D_5\) \(A_{1}\) \(4\) \(12-\operatorname{h}\), \(\operatorname{h}\le 2\)
\(4\) [rkCl612-deg4-quadric] \(\mathbb{P}^{1}\) \(D_4\) \(3A_1\) \(8\) \(13-\operatorname{h}\)
\(4\) \(V_{5}\) \(\operatorname{pt}\) \(A_4\) \(A_2\) \(6\) \(13\)
\(5\) [rkCl612-degree615] \(\mathbb{P}^{2}\) \(A_3\) \(A_1\times A_3\) \(12\) \(14\)
\(6\) 62 \(V_{6}\) \(\mathbb{P}^{2}\) \(A_2\) \(A_5\) \(20\) \(15\)
\(7\) 58 \(\mathbb{P}^{3}\) \(\operatorname{pt}\) \(A_{1}\) \(D_{6}\) \(32\) \(16\)
\(\operatorname{d}(X)=3\)
\(1\) \(V_{3}\) \(-\) \(\operatorname{pt}\) \(E_6\) \(-\) \(0\) \(\le 5\)
\(2\) [rkCl612-P1-bundle-degree613] \(-\) \(\mathbb{P}^{2}\) \(A_5\) \(A_{1}\) \(0\) \(6\)
\(3\) 76 [rkCl612-deg4-quadric] \(\mathbb{P}^{1}\) \(D_4\) \(-\) \(3\) \(6\le \operatorname{s}\le 7\)
\(5\) 62 \(V_{6}\) \(\mathbb{P}^{2}\) \(A_2\) \(2A_2\) \(9\) \(9\)
\(6\) 56 \(\mathbb{P}^{3}\) \(\operatorname{pt}\) \(A_{1}\) \(A_{5}\) \(15\) \(10\)
\(\operatorname{d}(X)=4\)
\(1\) \(V_{4}\) \(-\) \(\operatorname{pt}\) \(D_5\) \(-\) \(0\) \(\le 2\)
\(2\) [rkCl612-deg4-quadric] \(-\) \(\mathbb{P}^{1}\) \(D_4\) \(-\) \(0\) \(1\le \operatorname{s}\le 3\)
\(3\) 31[primitive-rho613-d614] \(-\) \((\mathbb{P}^1)^2\) \(A_3\) \(2A_1\) \(0\) \(4\)
\(4\) 62 \(V_{6}\) \(\mathbb{P}^{2}\) \(A_2\) \(2A_1\) \(4\) \(5\)
\(5\) 55 \(\mathbb{P}^{3}\) \(\operatorname{pt}\) \(A_{1}\) \(A_{1}\times A_{3}\) \(8\) \(6\)
\(\operatorname{d}(X)=5\)
\(1\) \(V_{5}\) \(-\) \(\operatorname{pt}\) \(A_4\) \(-\) \(0\) \(0\)
\(\operatorname{d}(X)=6\)
\(2\) \(V_{6}\) \(-\) \(\mathbb{P}^{2}\) \(A_2\) \(A_{1}\) \(0\) \(0\)
\(3\) \((\mathbb{P}^{1})^{3}\) \(-\) \(\operatorname{pt}\) \(A_{1}\) \(A_2\) \(0\) \(0\)
\(\operatorname{d}(X)=8\)
\(1\) \(\mathbb{P}^{3}\) \(-\) \(\operatorname{pt}\) \(-\) \(-\) \(0\) \(0\)

Here \(\bar X/Z\) is a primitive birational model of \(X\) (see Theorem 25) and \(\operatorname{h}:=h^{1,2}(\hat{X})\), where \(\hat{X}\) is the standard resolution of \(X\). For compactness, we denote \((\mathbb{P}^1)^k:=\underbrace{\mathbb{P}^1\times \cdots \times \mathbb{P}^1}_k\). For other notation we refer to 10.

8. Remark. For \(\operatorname{d}(X)\le 2\) any del Pezzo threefold automatically has \(G\)-structure (see Remark 16). So, in this case, [class-d611-factorial][class-maximal-rk-d612] is a complete list of del Pezzo threefolds with \(\operatorname{d}(X)\le 2\).

9. Remark. Singular three-dimensional cubics (without group action) whose singularities are only nodes and their small resolutions were classified in [16]. There is the following correspondence between our list and the classification in [16]: 1 \(\longleftrightarrow\) J15, [class-cubic-subSegre] \(\longleftrightarrow\) J14, [class-cubic-empty-delta] \(\longleftrightarrow\) J11, [class-cubic-P1-bundle]\(\longleftrightarrow\) J9, [class-cubic-factorial]\(\longleftrightarrow\) J1–J5.

We hope that our result can be useful for applications to the classification of finite subgroups of the Cremona group \(\operatorname{Cr}_3(\Bbbk)\) [3], [17], and also the the birational classification of rational algebraic threefolds over non-closed fields (cf. [1]).

Acknowledgments. This paper is an extended version of my talk given at “Workshop on Moduli and Birational Geometry” at POSTECH, Korea in August, 2010. I would like to thank POSTECH for the support and hospitality. I also would like to acknowledge discussions that I have on this subject with I. Arzhantsev, I. Cheltsov, A. Kuznetsov, K. Shramov and D. Timashev. Thanks are also due to the referee for suggesting several improvements.

2 Preliminaries.↩︎

10. Notation. We work over an algebraically closed field of characteristic \(0\). Throughout this paper \(X\) denotes a del Pezzo threefold with at worst terminal Gorenstein singularities. Thus we can write \(-K_X=2S\), where \(S=S_X\) is a ample Cartier divisor of \(S\) defined up to linear equivalence. Everywhere below we use the following notation:

  • \(\rho=\rho(X):=\operatorname{rk}\operatorname{Pic}(X)\);

  • \(\operatorname{Cl}(X)\) is the Weil divisor class group;

  • \(\operatorname{r}=\operatorname{r}(X):=\operatorname{rk}\operatorname{Cl}(X)\);

  • \(\operatorname{d}=\operatorname{d}(X):=S^3=-K_X^3/8\), the degree of \(X\);

  • \(\operatorname{p}=\operatorname{p}(X)\) is the number of planes on \(X\);

  • \(\operatorname{s}=\operatorname{s}(X)\) is the number of singular points of \(X\) under an additional assumption that \(X\) has at worst nodes;

  • \(V_6\subset \mathbb{P}^7\) is a smooth del Pezzo threefold with \(\operatorname{d}(X)=6\) and \(\rho=2\), see Theorem 15;

  • \(V_5\subset \mathbb{P}^6\) is a smooth del Pezzo threefold of degree \(5\) (see [5]);

  • \(V_d\), for \(d=1,\dots, 4\), is a del Pezzo threefold of degree \(d\) with terminal factorial singularities (see Theorem 15).

12. Terminal singularities (see [6]). Let \((X, P)\) be a germ of a three-dimensional terminal singularity. Then \((X, P)\) is isolated, i.e, \(\operatorname{Sing}(X)=\{P\}\). If additionally \((X, P)\) is Gorenstein, then \((X, P)\) is analytically isomorphic to a hypersurface singularity of multiplicity \(2\).

Let \(X\) be a threefold with Gorenstein terminal singularities. Then any Weil \(\mathbb{Q}\)-Cartier divisor is Cartier (see e.g. [7]). Equivalently, \(\operatorname{Pic}(X)\) is a primitive sublattice in \(\operatorname{Cl}(X)\).

Let \(X\) be a \(G\)-variety. We say that \(X\) has only \(G\mathbb{Q}\)-factorial singularities if any \(G\)-invariant Weil divisor is \(\mathbb{Q}\)-Cartier.

11. Theorem-Definition ([7]). Let \(X\) be a threefold with terminal singularities. Then there exists a projective birational morphism \(\xi: \hat{X}\to X\) such that

  1. \(\hat{X}\) is normal and has only terminal \(\mathbb{Q}\)-factorial singularities;

  2. \(\xi\) is a crepant morphism, that is, \(K_{\hat{X}}=\xi^*K_X\);

  3. \(\xi\) is small, that is, its exceptional locus does not contain any divisors.

Such a morphism is called \(\mathbb{Q}\)-factorialization* of \(X\). Any two \(\mathbb{Q}\)-factorializations of \(X\) are connected by a sequence of flops.*

13. Theorem [18]. Let \(X\) be a rationally connected threefold with terminal factorial singularities. Assume that \(-K_X=2S\) for some divisor \(S\) and \(\rho(X)>1\). Let \(f: X\to Z\) be an extremal \(K_X\)-negative contraction. Then one of the following holds:

  1. \(Z\simeq \mathbb{P}^1\) and \(f\) is a quadric bundle, i.e. there is an embedding \(X\hookrightarrow \mathbb{P}({\mathscr{E}})\), where \({\mathscr{E}}\) is a rank \(4\) vector bundle on \(Z\), so that each fiber of \(f\) is a quadric in the fiber of \(\mathbb{P}({\mathscr{E}})/Z\);

  2. \(X\) is smooth, \(Z\) is a smooth rational surface, and \(X=\mathbb{P}({\mathscr{E}})\), where \({\mathscr{E}}\) is a rank \(2\) vector bundle on \(Z\);

  3. \(Z\) is a threefold with terminal factorial singularities and \(f\) is blowup of a smooth point on \(Z\).

3 Generalities on del Pezzo threefolds↩︎

14. Definition. Let \(X\) be a projective variety \(X\) with at worst terminal Gorenstein singularities.1 We say that \(X\) is a del Pezzo threefold (resp. weak del Pezzo threefold) if its anti-canonical class \(-K_X\) is divisible by \(2\) and is ample (resp. nef and big).

Note that if \(X\) is a Fano threefold with at worst terminal Gorenstein singularities such that \(-K_X\) is divisible by some positive integer \(q\), then \(q\le 4\). Moreover, \(q=4\) iff \(X\simeq \mathbb{P}^3\) and \(q=3\) iff \(X\) is a quadric in \(\mathbb{P}^4\) (see e.g. [5]). Thus, for a del Pezzo threefold \(X\not \simeq \mathbb{P}^3\), the divisor \(-\frac{1}{2} K_X\) is a primitive element of the lattice \(\operatorname{Cl}(X)\).

15. Theorem ([8], [11], [9], [10], [12]). Let \(X\) be a del Pezzo threefold and let \(S=-\frac{1}{2}K_X\).

  1. \(\dim |S|=\operatorname{d}(X)+1\) and \(\operatorname{d}(X)\le 8\).

  2. The linear system \(|S|\) is base point free (resp. very ample) for \(\operatorname{d}(X)\ge 2\) (resp. \(\operatorname{d}(X)\ge 3\)). If \(\operatorname{d}(X)\ge 4\), then the image of \(X_{\operatorname{d}(X)}\subset \mathbb{P}^{\operatorname{d}(X)+1}\) of \(X\) under the embedding given by \(|S|\) is an intersection of quadrics.

  3. If \(\operatorname{d}(X)=1\), then the linear system \(|S|\) has a unique base point which is a smooth point of \(X\). In this case \(|S|\) defines a rational map \(X \dashrightarrow \mathbb{P}^{2}\) whose general fiber is an elliptic curve. The variety \(X\) is isomorphic to a hypersurface of degree \(6\) in the weighted projective space \(\mathbb{P}(1^3,2,3)\).

  4. If \(\operatorname{d}(X)=2\), then \(|S|\) defines a double cover \(X\to \mathbb{P}^3\) whose branch locus \(B\subset \mathbb{P}^3\) is a surface of degree \(4\) with at worst isolated singularities. The variety \(X\) is isomorphic to a hypersurface of degree \(4\) in \(\mathbb{P}(1^{4},2)\).

  5. If \(\operatorname{d}(X)=3\), then \(X\) is isomorphic to a cubic in \(\mathbb{P}^{4}\).

  6. If \(\operatorname{d}(X)=4\), then \(X\) is isomorphic to a complete intersection of two quadrics in \(\mathbb{P}^{5}\).

  7. If \(\operatorname{d}(X)= 5\), then \(\rho(X)=1\) [19]. A smooth variety of this type is unique up to isomorphism and isomorphic to a section of the Grassmann variety \(\operatorname{Gr}(2,5)\subset \mathbb{P}(\wedge^2 \mathbb{C}^5)\) by a subspace of codimension \(3\).

  8. If \(\operatorname{d}(X)=6\), then \(\rho(X)=3\) or \(2\) [19]. In both cases a smooth variety of this type is unique up to isomorphism and either \(X\simeq \mathbb{P}^1\times \mathbb{P}^1\times \mathbb{P}^1\) or \(X\) is isomorphic to a divisor of bidegree \((1,1)\) in \(\mathbb{P}^2\times \mathbb{P}^2\).

  9. If \(\operatorname{d}(X)= 7\), then \(X\simeq \mathbb{P}(\mathscr{O}_{\mathbb{P}^2}\oplus \mathscr{O}_{\mathbb{P}^2}(1))\).

  10. If \(\operatorname{d}(X)= 8\), then \(X\simeq \mathbb{P}^3\).

The del Pezzo threefolds with \(\operatorname{d}(X)=1\) and \(\operatorname{d}(X)=2\) have their names: double Veronese cone and quartic double solid, respectively.

16. Remark. Let \(X\) be a del Pezzo threefold of degree \(1\) (resp. \(2\)). Then there is a double cover \(\varphi: X\to \mathbb{P}(1^{3},2)\) (resp. \(\varphi: X\to \mathbb{P}^3\)). The corresponding natural Galois involution \(X\to X\) is called Bertini (resp. Geiser) involution. Therefore, any del Pezzo threefold \(X\) with \(\operatorname{d}(X)\le 2\) is a \(G\)-del Pezzo (in the geometric sense).

17. Corollary. Let \(X\) be a del Pezzo threefold. If \(X\) is factorial and singular, then \(\rho(X)=1\).

Outline of the proof. By Theorem 15 we have to check only the case \(\operatorname{d}(X)=6\). Alternatively, one can use Theorem 13 (see the arxiv version). ◻

18. Proposition (see [5]). Let \(X\) be a smooth del Pezzo threefold. Then the Hodge number \(h^{1,2}(X)\) is given by the following table:

\(\operatorname{d}(X)\) \(1\) \(2\) \(3\) \(4\) \(5\) \(6\) \(7\) \(8\)
\(h^{1,2}(X)\) \(21\) \(10\) \(5\) \(2\) \(0\) \(0\) \(0\) \(0\)

23. Definition. Let \(X\) be a weak del Pezzo threefold and let \(S=-\frac{1}{2}K_X\). An irreducible surface \(\Pi\subset X\) is called a plane if \(S^2\cdot \Pi=1\) and, in case \(\operatorname{d}(X)=1\), the base point of \(|S|\) does not lie on \(\Pi\).

19. Lemma. Let \(X\) be a del Pezzo threefold. If \(\Pi\subset X\) is a plane, then \(\Pi\simeq \mathbb{P}^2\) and \(\mathscr{O}_{\Pi}(S)=\mathscr{O}_{\mathbb{P}^2}(1)\).

Proof. The statement is obvious if \(\operatorname{d}(X)\ge 3\) because the divisor \(S\) is very ample in this case. If \(\operatorname{d}(X)=2\), then \(|S|\) defines a double cover \(\varphi:X\to \mathbb{P}^3\) so that \(\varphi(\Pi)\) is a projective plane on \(\mathbb{P}^3\). Thus \(\varphi|_{\Pi}:\pi \to \varphi(\Pi)\simeq \mathbb{P}^2\) is a finite birational morphism, so it is an isomorphism. Finally if \(\operatorname{d}(X)=1\), then \(|S|\) defines a rational map \(\varphi:X\dashrightarrow \mathbb{P}^2\) so that its restriction to \(\Pi\) is a morphism which must be finite and birational. As above we get \(\Pi\simeq \mathbb{P}^2\). ◻

20. Lemma.If \(\Pi\subset X\) is a plane, then there is a \(\mathbb{Q}\)-factorialization \(\xi:\hat{X}\to X\) such that for the proper transform \(\hat{\Pi}\) of \(\Pi\) we have \(\hat{\Pi}\simeq \mathbb{P}^2\) and \(\mathscr{O}_{\hat{\Pi}}(\hat{\Pi})\simeq \mathscr{O}_{\mathbb{P}^2}(-1)\). Therefore, \(\hat{\Pi}\) is contractible, i.e. there is a birational contraction \(\hat{X}\to X'\) of \(\hat{\Pi}\) to a smooth point. Conversely, if \(\xi:\hat{X}\to X\) is a \(\mathbb{Q}\)-factorialization and \(\hat{\Pi}\subset \hat{X}\) is an irreducible surface such that \(\hat{\Pi}\simeq \mathbb{P}^2\) and \(\mathscr{O}_{\hat{\Pi}}(\hat{\Pi})\simeq \mathscr{O}_{\mathbb{P}^2}(-1)\), then \(f(\hat{\Pi})\) is a plane on \(X\).

Proof. Let \(\Pi\subset X\) be a plane. Take a \(\mathbb{Q}\)-factorialization \(\xi:\hat{X}\to X\) so that \(\hat{\Pi}\) is \(f\)-nef. One can do it by performing flops over \(X\). Assume that \(\hat{\Pi}\) is nef. Then by the base point free theorem the linear system \(|n\hat{\Pi}|\) is base point free for \(n\gg 0\). Hence \(|n\Pi|\) has no fixed components. Since \(X\) has at worst isolated singularities, by adjunction we have \[K_{\Pi}=(-2S+\Pi)|_{\Pi} \ge -2S|_{\Pi},\] a contradiction.

Thus \(\hat{\Pi}\) is not nef. Then there is a \(K_{\hat{X}}\)-negative extremal ray \(R\) such that \(\hat{\Pi}\cdot R<0\). Since \(K_{\hat{X}}\) is divisible by \(2\), from the classification of extremal rays (Theorem 13) we see that \(\hat{\Pi}\) is contractible to a smooth point, \(\hat{\Pi}\simeq \mathbb{P}^2\) and \(\mathscr{O}_{\hat{\Pi}}(\hat{\Pi})\simeq \mathscr{O}_{\mathbb{P}^2}(-1)\). The converse statement is obvious. ◻

21. Lemma.Let \(X\) be a del Pezzo threefold and let \(S\in |{-}\frac{1}{2} K_X|\) be a smooth member. Let \(l\in \operatorname{Pic}(S)\) be an element such that \(l^2=l\cdot K_S=-1\) (the class of a line \(L\subset S\)). Assume that \(l\in \iota^*\operatorname{Cl}(X)\), where \(\iota: S\hookrightarrow X\) is the embedding. Then there exists a unique plane \(\Pi\subset X\) such that \(\iota^*\Pi=l\) (i.e., \(\Pi\cap S=L\)).

Proof. Denote by \(\Pi\) any divisor whose class coincides with \(\iota^*l\). Let \(\xi:\hat{X}\to X\) be a \(\mathbb{Q}\)-factorialization as in Lemma 20, let \(\hat{S}:=\xi^{-1}(S)\), and let \(\hat{\Pi}\) be the proper transform of \(\Pi\). By Shokurov’s adjunction theorem the pair \((\hat{X},\hat{\Pi})\) is purely log terminal (PLT). Hence, by the Kawamata-Viehweg vanishing [20] \[H^1 (\hat{X}, \mathscr{O}_{\hat{X}}(\hat{\Pi} -\hat{S})) = H^1 (\hat{X}, \mathscr{O}_{\hat{X}}(\hat{S}+K_{\hat{X}}+\hat{\Pi})) =0.\] Then one can see from the exact sequence \[0 \longrightarrow \mathscr{O}_{\hat{X}}(\hat{\Pi} -\hat{S}) \longrightarrow \mathscr{O}_{\hat{X}}(\hat{\Pi}) \longrightarrow \mathscr{O}_{\hat{S}}(\iota^*l) \longrightarrow 0\] that \(H^0(\hat{X}, \mathscr{O}_{\hat{X}}(\hat{\Pi}))\neq 0\), so we may assume that both \(\hat{\Pi}\) and \(\Pi\) are effective. Since \(S^2\cdot \Pi=1\), \(\Pi\) is a plane. Finally, if there is another plane \(\Pi'\) such that \(\iota^*\Pi'=l\), then \(\Pi\sim \Pi'\) and \(\mathscr{O}_{\hat{\Pi}}(\hat{\Pi})= \mathscr{O}_{\hat{\Pi}}(\hat{\Pi}')\) is positive, a contradiction. ◻

22. Definition. We say that a del Pezzo threefold \(X\) is imprimitive if it contains at least one plane. Otherwise we say that \(X\) is primitive.

The following two theorems are easy consequences of [13].

24. Theorem. Let \(X\) be a primitive weak del Pezzo threefold with at worst terminal Gorenstein singularities. Let \(\xi : \hat{X}\to X\) be a \(\mathbb{Q}\)-factorialization. Then there exists a \(K_{\hat{X}}\)-negative Mori contraction \(f:\hat{X}\to Z\) such that one of the following holds:

  1. \(Z\) is a point, \(\rho(\hat{X})=1\), \(X\) is factorial, and \(\xi\) is an isomorphism;

  2. \(Z\simeq \mathbb{P}^2\), \(\rho(\hat{X})=2\), and \(f\) is a \(\mathbb{P}^1\)-bundle, i.e. \(\hat{X}\) is smooth and \(\hat{X}=\mathbb{P}_{\mathbb{P}^2}({\mathscr{E}})\), where \({\mathscr{E}}\) is a rank-\(2\) vector bundle on \(\mathbb{P}^2\);

  3. \(Z\simeq \mathbb{P}^1\), \(\rho(\hat{X})=2\), and \(f\) is a quadric bundle, i.e. any fiber of \(f\) is an irreducible quadric in \(\mathbb{P}^3\);

  4. \(Z\simeq \mathbb{P}^1\times \mathbb{P}^1\), \(\rho(\hat{X})=3\), and \(f\) is a \(\mathbb{P}^1\)-bundle.

Proof. Almost all the statements are proved in [13]. We have to show only that \(Z\not \simeq \mathbb{F}_2\). Indeed, if \(\dim Z=2\), then for a general member \(\bar S\in |{-}\frac{1}{2} K_{\bar X}|\), the restriction \(f|_{\bar S}: \bar S\to Z\) is birational. Hence \(Z\) is a del Pezzo surface. ◻

25. Theorem. Let \(X\) be an imprimitive del Pezzo threefold with at worst terminal Gorenstein singularities. Then there exists a diagram \[\label{eq-construction} \xymatrix{ & \hat{X} \ar[r]^{\sigma}\ar[ld]_{\xi} & \bar{X} \ar[rd]^{f} \\ X&&&Z }\qquad{(1)}\] where

  1. \(\xi\) is a \(\mathbb{Q}\)-factorialization;

  2. \(\hat{X}\) is an weak del Pezzo threefold with at worst terminal factorial singularities;

  3. \(\sigma\) is a blowup in smooth distinct points \(P_1,\dots,P_n\in \bar X\);

  4. \(\operatorname{d}({X})=\operatorname{d}(\hat{X})=\operatorname{d}(\bar{X})+n\);

  5. \(\bar X\) is a primitive weak del Pezzo threefold with \(\rho(\bar X)\le 2\), thus \(\bar X\) is described by (i)-(iii) of Theorem 24.

26. Corollary. Let \(X\) be a del Pezzo threefold. Then \(\operatorname{r}(X)+\operatorname{d}(X)\le 9\).

Proof. We have \(9\ge \rho(\bar X)+\operatorname{d}(\bar X)=\rho(\hat{X})+\operatorname{d}(\hat{X})=\operatorname{r}(X)+\operatorname{d}(X).\) ◻

27. Corollary. Let \(X\) be a weak del Pezzo threefold and let \(S\in |{-}\frac{1}{2} K_X|\) be a smooth element. Then the restriction map \(\operatorname{Cl}(X)\to \operatorname{Pic}(S)\) is injective and its cokernel is torsion free.

Proof. Clearly the assertion is invariant under taking small modifications. In view of construction ?? , it is sufficient to prove that the restriction map \(\operatorname{Cl}(\bar X)\to \operatorname{Pic}(\bar S)\) is injective and its cokernel is torsion free, where \(\bar S=\sigma(S)\). Thus we may assume that \(X\) is a primitive factorial weak del Pezzo threefold. The assertion is obvious if \(\rho(X)=1\). Assume that \(\rho(X)=2\). Then \(\rho(Z)=1\). Let \(\Theta\) be the ample generator of \(\operatorname{Pic}(Z)\). The group \(\operatorname{Cl}(X)\) is generated by \(f^* \Theta\) and the class of \(S\). Recall that \(Z\) is either \(\mathbb{P}^1\) or \(\mathbb{P}^2\). Hence \(f^*\Theta|_S\) is either a conic or the pull-back of a line on \(\mathbb{P}^2\), respectively. It is easy to see that \(f^*\Theta|_S\) and \(-K_S\sim S|_S\) generate a rank 2 primitive sublattice in \(\operatorname{Pic}(S)\). The case \(\rho(X)=3\) can be treated similarly. ◻

4 Primitive del Pezzo threefolds with \(\operatorname{r}(X)=3\)↩︎

28. Lemma.Let \(X\) be a primitive del Pezzo threefold with \(\operatorname{r}(X)=3\) and let \(\mathcal{F}=|F|\) be a complete one-dimensional linear system (pencil) of Weil divisors without fixed components. There is a small \(\mathbb{Q}\)-factorialization \(\xi: \hat{X}\to X\) such that the proper transform \(\hat{\mathcal{F}}\) of \(\mathcal{F}\) on \(\hat{X}\) is base point free and defines a fibration \(f: \hat{X}\to \mathbb{P}^1\). Moreover, \(f\) factors through a (not unique) \(\mathbb{P}^1\)-bundle contraction \[\label{equation-primitive-r613-fibration} f: \hat{X} \overset{f_1}{\longrightarrow} \mathbb{P}^1\times \mathbb{P}^1 \overset{f_2}{\longrightarrow} \mathbb{P}^1\qquad{(2)}\]

Proof. Take a \(\mathbb{Q}\)-factorialization \(\xi: \hat{X}\to X\) so that \(\hat{\mathcal{F}}\) is \(\xi\)-nef (one can get it by performing flops over \(X\)). Then \(\hat{\mathcal{F}}\) is nef. Indeed, otherwise there is a \(K_{\hat{X}}\)-negative extremal ray \(R\) such that \(\hat{\mathcal{F}}\cdot R<0\). Since \(\hat{\mathcal{F}}\) has no fixed components, \(R\) must define a flipping contraction. On the other hand, \(K_X\) is Cartier, a contradiction [2]. Thus \(\hat{\mathcal{F}}\) is nef. Then \(\hat{\mathcal{F}}\) defines a contraction to a (rational) curve by the base point free theorem. Further, since \(\operatorname{r}(X)=3\), we have \(\rho(\hat{X})=3\). Running the MMP over \(\mathbb{P}^1\) we obtain \(f_1\). ◻

29. Remark-definition. In notation of ?? , another ruling on \(\mathbb{P}^1\times \mathbb{P}^1\) defines another pencil \(\mathcal{F}'\) on \(X\). In this situation, we say that pencils \(\mathcal{F}\) and \(\mathcal{F}'\) are conjugate. Thus there is one-to-one correspondence between

  1. the set of pairs of conjugate pencils \(\mathcal{F}\), \(\mathcal{F}'\) and

  2. the set of \(\mathbb{Q}\)-factorializations \(X'\to X\) together with a structure of \(\mathbb{P}^1\)-bundle \(f' : \hat{X'}\to \mathbb{P}^1\times \mathbb{P}^1\).

30. Corollary. The cone of effective divisors \(\overline{\operatorname{NE}}^1(X)\) is generated by classes of pencils \(\mathcal{F}\) as in Lemma 28.

Proof. Let \(\xi :\hat{X}\to X\) be a small \(\mathbb{Q}\)-factorialization. There are natural identifications \(\operatorname{Cl}(X)= \operatorname{Cl}(\hat{X})\) and \(\overline{\operatorname{NE}}^1(X)=\overline{\operatorname{NE}}^1(\hat{X})\). The variety \(\hat{X}\) is a Mori dream space [21]. Hence \(\overline{\operatorname{NE}}^1(\hat{X})\) is a polyhedral cone generated by a finite number of effective divisors \(D_i\). Running \(D_i\)-MMP on \(\hat{X}\), after a number of flops, we get a \(\mathbb{P}^1\)-bundle over \(\mathbb{P}^1\times \mathbb{P}^1\) (because \(X\) is primitive). This shows that \(D_i\) must coincide with some \(\mathcal{F}\). ◻

31. Theorem. Let \(X\) be a primitive del Pezzo threefold with \(\operatorname{r}(X)=3\). Let \(\{\mathcal{F}_i\}\) be the set of all pencils as in Lemma 28. Then there are the following possibilities for \(\{\mathcal{F}_i\}\), where we draw the graph for \(\{\mathcal{F}_i\}\) so that every two elements are connected by an edge if and only if they are conjugate.

  1. \(\operatorname{d}(X)=2\)

    \[\xygraph{ !{<0cm,0cm>;<2.8cm,0cm>:<0cm,1cm>::} !{(0,3) }*+{\overset{S-\mathcal{F}_1+\mathcal{F}_2}\bullet}="S-F1F2" !{(1,2) }*+{\overset{2S-\mathcal{F}_1}\bullet}="2SF1" !{(-1,2) }*+{\overset{\mathcal{F}_2}\bullet}="F2" !{(1,1) }*+{\underset{S-\mathcal{F}_2}\bullet}="2SF2" !{(-1,1) }*+{\underset{\mathcal{F}_1}\bullet}="F1" !{(0,0) }*+{\underset{S+\mathcal{F}_1-\mathcal{F}_2}\bullet}="SF1-F2" "S-F1F2"-"F2" "S-F1F2"-"2SF1" "2SF1"-"2SF2" "F2"-"F1" "2SF2"-"SF1-F2" "F1"-"SF1-F2" }\]

  2. \(\operatorname{d}(X)=4\) \[\xygraph{ !{<0cm,0cm>;<2.5cm,0cm>:<0cm,1cm>::} !{(0,0) }*+{\underset{S-\mathcal{F}_2}\bullet}="SF2" !{(1,0) }*+{\underset{S-\mathcal{F}_1}\bullet}="SF1" !{(0,1) }*+{\overset{\mathcal{F}_1}\bullet}="F1" !{(1,1) }*+{\overset{\mathcal{F}_2}\bullet}="F2" "SF2"-"SF1" "SF2"-"F1" "F2"-"SF1" "F2"-"F1" }\]

  3. \(\operatorname{d}(X)=6\) and \(X\simeq \mathbb{P}^1\times \mathbb{P}^1\times\mathbb{P}^1\) \[\xygraph{ !{<0cm,0cm>;<2cm,0cm>:<0cm,1cm>::} !{(0,0) }*+{\underset{\mathcal{F}_1}\bullet}="a" !{(1,1) }*+{\overset{S-\mathcal{F}_1-\mathcal{F}_2}\bullet}="b" !{(2,0) }*+{\underset{\mathcal{F}_2}\bullet}="c" "a"-"b" "a"-"c" "b"-"c" }\]

Proof. Let \(\mathcal{F}_1\) and \(\mathcal{F}_2\) be two conjugate pencils and let \(\xi :\hat{X}\to X\) be the corresponding small \(\mathbb{Q}\)-factorialization. Clearly, we have \[\mathcal{F}_1^2\equiv \mathcal{F}_2^2\equiv 0,\; \mathcal{F}_1\cdot \mathcal{F}_2\cdot S=1, \;S^2\cdot \mathcal{F}_1=S^2\cdot \mathcal{F}_2=2.\] For any \(j\), write \(\mathcal{F}_j\sim aS+b_1\mathcal{F}_1+b_2\mathcal{F}_2\), where \(a\ge 0\). Then \[\label{eq-primitive-rk613} \begin{array}{ll} 0=\mathcal{F}_j^2\cdot S&=a^2d+4a(b_1+b_2)+2b_1b_2, \\[5pt] 2=\mathcal{F}_j\cdot S^2&=ad+2(b_1+b_2), \end{array}\tag{1}\] where \(d:=\operatorname{d}(X)\). Therefore, \[\begin{array}{lll} b_1+b_2&=&\frac{1}{2}(2- ad), \\[5pt] b_1b_2&=&\frac{1}{2} a(ad-4). \end{array}\] Since this system has an integer solution in \(b_1\), \(b_2\), the discriminant \[\frac{1}{4}(2- ad)^2-2 a(ad-4)=\frac{1}{4}\bigl(4- a (8 - d) (a d - 4)\bigr)\] must be a square and \(ad\) must be even. Assuming \(a>0\) (i.e. \(\mathcal{F}_j\neq \mathcal{F}_1\), \(\mathcal{F}_2\)), we get \(ad=8\), \(6\), \(4\), or \(2\). Hence, up to permutation of \(b_1\) and \(b_2\), there are the following solutions with \(a>0\):

  1. \(d=1\),\((a, b_1, b_2)=\) \((4, -1, 0)\), \((4, 0, -1)\);

  2. \(d=2\),\((a, b_1, b_2)=\) \((1, -1, 1)\), \((1, 1, -1)\), \(( 2, -1, 0)\), \((2, 0, -1)\);

  3. \(d=4\),\((a, b_1, b_2)=\) \(( 1, -1, 0)\), \((1, 0, -1)\);

  4. \(d=6\),\((a, b_1, b_2)=\) \((1, -1, -1)\).

Note that if \(\mathcal{F}_j\) and \(\mathcal{F}_k\) are conjugate, then \(\mathcal{F}_j\cdot \mathcal{F}_k\cdot S=1\). From this one can see that for each \(\mathcal{F}_j\) there are exactly two divisors in \(\{\mathcal{F}_i\}\) that conjugate to \(\mathcal{F}_j\). Moreover, if \(d\neq 1\), then conjugacy relations are given by graphs in Theorem 31 [primitive-rho613-d612], [primitive-rho613-d614], [primitive-rho613-d616]. In the case \(d=1\) we get the following (disconnected) graph: \[\xygraph{ !{<0cm,0cm>;<3cm,0cm>:<0cm,1cm>::} !{(0,0) }*+{\overset{\mathcal{F}_1}\bullet}="F1" !{(1,0) }*+{\overset{\mathcal{F}_2}\bullet}="F2" !{<0cm,0cm>;<3cm,0cm>:<0cm,1cm>::} !{(2,0) }*+{\overset{4S-\mathcal{F}_1}\bullet}="4SF1" !{(3,0) }*+{\overset{4S-\mathcal{F}_2}\bullet}="4SF2" "4SF1"-"4SF2" "F1"-"F2" }\]

Hence there are only two extremal \(K_{\hat{X}}\)-negative contractions on \(\hat{X}\). On the other hand, the cone \(\overline{\operatorname{NE}}^1(\hat{X})\) has at least three extremal rays, a contradiction. ◻

32. Remark. Let \(X\) be a primitive del Pezzo threefold with \(\operatorname{r}(X)=3\) and \(\operatorname{d}(X)=2\) or \(4\). Let \(\xi : \hat{X}\to X\) be a \(\mathbb{Q}\)-factorialization. Then \(\hat{X}\simeq \mathbb{P}({\mathscr{E}})\), where \({\mathscr{E}}\) is a stable rank two vector bundle on \(Z=\mathbb{P}^1\times \mathbb{P}^1\) with \(c_1({\mathscr{E}})=0\), \(c_2({\mathscr{E}})=6-\operatorname{d}(X)\).

33. Example. If \(\operatorname{d}(X)=4\), an example of such \({\mathscr{E}}\) can be obtained as a restriction of the null-correlation bundle \(\mathscr N\) from \(\mathbb{P}^3\) to \(Z\), where \(Z\subset\mathbb{P}^3\) is the Segre embedding. Recall that the null-correlation bundle is defined by the exact sequence \[0\longrightarrow \mathscr{O}_{\mathbb{P}^3}\longrightarrow \Omega_{\mathbb{P}^3}(2) \longrightarrow\mathscr N(1) \longrightarrow 0.\] Its projectivization \(Y:=\mathbb{P}(\mathscr N)\) is a Fano fourfold of index \(2\) [22]. This \(Y\) has also a structure of \(\mathbb{P}^1\)-bundle over a smooth three-dimensional quadric. Let \(\hat{X}=\mathbb{P}({\mathscr{E}})=\pi^{-1}(Z)\), where \(\pi :Y\to \mathbb{P}^3\) is the natural projection. Then \(\hat{X}\) is a weak del Pezzo threefold of type 31[primitive-rho613-d614].

Examples of del Pezzo threefolds of type 31 [primitive-rho613-d612] can be constructed similarly by restricting to \(Z\subset \mathbb{P}^3\) rank two stable vector bundles \(\mathscr F\) with \(c_1=0\), \(c_2=2\) [23].

Another way to show existence of del Pezzo threefolds of types 31[primitive-rho613-d614] and [primitive-rho613-d612] is by writing down explicit equations:

34. Example. Let \(X\subset \mathbb{P}^5\) is given by the equations \[\begin{cases} x_1x_3-x_2x_4+a_{3,4}x_3x_5 +a_{3,6}x_3x_6+a_{4,5}x_4x_5+a_{4,6}x_4x_6=0 \\ x_1x_5-x_2x_6+b_{3,4}x_3x_5 +b_{3,6}x_3x_6+b_{4,5}x_4x_5+b_{4,6}x_4x_6=0 \end{cases}\] where \(a_{i,j}\), \(b_{i,j}\) are sufficiently general constants. Then \(X\) is a del Pezzo threefold having exactly \(4\) nodes. By Corollary 93 \(\operatorname{r}(X)\ge 3\). On the other hand, by results of 55 and §8 below \(\operatorname{r}(X)=3\). Finally, two quadrics \(x_5=x_6=x_1x_3-x_2x_4=0\) and \(x_3=x_4=x_1x_5-x_2x_6=0\) determine two conjugate pencils. Therefore, \(X\) is of type 31[primitive-rho613-d614].

5 Del Pezzo threefolds with \(\operatorname{r}(X)=2\)↩︎

The results of this section are contained in [14]. We give a short self-contained proof for the convenience of the reader.

35. Let \(X\) be a del Pezzo threefold with \(\operatorname{r}(X)=2\). There exists the following diagram: \[\xymatrix{ &\hat{X}\ar[dl]_{f}\ar[dr]_{\xi}\ar@{-->}[rr] &&\hat{X}^+\ar[dl]^{\xi^{+}}\ar[dr]^{f^+} \\ Z&&X&&Z^+ }\] where \(\xi\), \(\xi^+\) are small \(\mathbb{Q}\)-factorializations, \(\hat{X} \dashrightarrow \hat{X}^+\) is a flop, and \(f\), \(f^+\) are \(K\)-negative extremal contractions. We may assume that \(\dim Z\ge \dim Z^+\). Let \(S=-\frac{1}{2} K_X\) and let \(\hat{S}=h^*S\). Let \(M\) (resp. \(M^+\)) be the ample generator of \(\operatorname{Pic}(Z)\) (resp. \(\operatorname{Pic}(Z^+)\)). Put \(L:=f^*M\) and \(L^+:=f^{{+}*}M^+\). Let \(L'\) be the proper transform of \(L^+\) on \(\hat{X}\). If \(f\) is birational, then \(E\subset \hat{X}\) denotes the \(f\)-exceptional divisor. Similarly, if \(f^+\) is birational, then \(E'\subset \hat{X}\) is the proper transform of \(f^+\)-exceptional divisor.

36. Remark. If in the above notation \(\operatorname{d}(X)\le 2\), then by 16 there is a natural (Bertini or Geiser) involution \(\tau: X\to X\). In this case, \(\hat{X}^+\simeq \hat{X}\), \(\xi^+=\tau\mathbin{\scriptstyle{\circ}}\xi\), \(Z\simeq Z^+\), and \(f^+\) has the same type as \(f\).

The following theorem was proved (in much stronger form) in [14]. A short simple proof can be found in the arxiv version of the paper.

37. Theorem. In the above notation there are the following possibilities.

\(f\) \(f^+\) \(\operatorname{d}\) \(\operatorname{Pic}(\hat{X})\) \(\operatorname{s}\)
\(f\) \(f^+\) \(\operatorname{d}\) \(\operatorname{Pic}(\hat{X})\) \(\operatorname{s}\)
\(\mathbb{P}^1\)-bundle \(\mathbb{P}^1\)-bundle \(1\) \(L+L'\sim 6\hat{S}\) \(22\)
\(2\) \(L+L'\sim 3\hat{S}\) \(11\)
\(3\) \(L+L'\sim 2\hat{S}\) \(6\)
\(6\) \(L+L'\sim \hat{S}\) \(0\)
\(\mathbb{P}^1\)-bundle quadric bundle \(5\) \(L+L'\sim \hat{S}\) \(1\)
quadric bundle quadric bundle \(1\) \(L+L'\sim 4\hat{S}\) \(\le 22\)
\(2\) \(L+L'\sim 2\hat{S}\) \(\le 11\)
\(4\) \(L+L'\sim \hat{S}\) \(\le 3\)
birational \(\mathbb{P}^1\)-bundle \(4\) \(E+L'\sim \hat{S}\) \(3\)
\(7\) \(E+2L'\sim \hat{S}\) \(0\)
birational quadric bundle \(3\) \(E+L'\sim \hat{S}\) \(4\), \(5\), \(6\)
birational birational \(1\) \(E+E'\sim 2\hat{S}\) \(12\le \operatorname{s}\le 22\)
\(2\) \(E+E'\sim \hat{S}\) \(6\le \operatorname{s}\le 11\)

Here in the 5th column we indicate relations between \(L\), \(L'\), \(E\), and \(E'\) in \(\operatorname{Pic}(\hat{X})\).

38. Corollary. Let \(X\) be a del Pezzo threefold with \(\operatorname{r}(X)=1\). Assume that \(X\) is singular. Then \(\operatorname{d}(X)\le 4\). If \(\operatorname{d}(X)= 4\), then every singular point \(P\in X\) is rs-nondegenerate (see 85). Moreover, \(\sum_P \lambda(X,P)\le 2\).

Proof. Let \(P\in X\) be a general point. Let \(\sigma: \tilde{X} \to X\) be the blowup of \(P\), let \(E:=\sigma^{-1}(P)\), and let \(\tilde{S}\) be the proper transform of \(S=-\frac{1}{2}K_X\). Write \(-K_{\tilde{X}}=2\sigma^*S-2E=2\tilde{S}\). Since the linear system \(|\tilde{S}|\) is base point free and big, \(\tilde{X}\) is a weak del Pezzo threefold with at worst factorial terminal singularities, \(\rho(\tilde{X})=2\), and \(\operatorname{d}(\tilde{X})=\operatorname{d}(X)-1\). If \(\operatorname{d}(X)\ge 5\), then by Theorem 37 we have only one possibility [rkCl612-deg4-birational]. But then both \(\tilde{X}\) and \(X\) are smooth. If \(\operatorname{d}(X)=4\), then we have case [rkCl612-cubic-plane]. In this case any singularity \(\tilde{P}\in \tilde{X}\) is analytically isomorphic to the hypersurface singularity given by \(x_1x_2+x_3^2+x_4^{n}=0\). Then \(\lambda(\tilde{X},\tilde{P})=\nu(\tilde{X},\tilde{P})=\lfloor n/2\rfloor\). The last inequality follows by Proposition 91. ◻

39. By [14] all the cases in the table do occur2. Below we give explicit examples of some del Pezzo threefolds with \(\operatorname{r}(X)=2\).

Case [rkCl612-P1-bundle-degree613].↩︎

\(X=X_3\subset \mathbb{P}^4\) is given by an equation of the form \[(x_1x_4-x_2x_3)\ell_1 +(x_2^2-x_1x_3)\ell_2 +(x_3^2-x_2x_4)\ell_3=0,\] where \(\ell_i(x_1,\dots, x_5)\) are linear forms.

Case [rkCl612-degree615]↩︎

(cf. 54 and 68.) Let \(Y\) be the blowup of \(\mathbb{P}^1\times \mathbb{P}^2\) along a smooth curve \(C\) of bidegree \((2,1)\). Then \(Y\) is a Fano threefold with \(-K_Y^3=38\) and \(\rho(Y)=3\) [24]. Let \(S\subset \mathbb{P}^1\times \mathbb{P}^2\) be a (unique) effective divisor of bidegree \((0,1)\) containing \(C\) and let \(\tilde{S}\) be the proper transform of \(S\) on \(Y\). Then \(\tilde{S}\simeq S\simeq \mathbb{P}^1\times \mathbb{P}^1\) and \(\mathscr{O}_{\tilde{S}}(\tilde{S})\) is of type \((-1,-1)\). Therefore, there exists a contraction \(\varphi: Y\to X\), where \(\varphi(\tilde{S})\) is a node. Here \(X\) is a quintic del Pezzo threefold as in [rkCl612-degree615].

Case [rkCl612-deg2-quadric].↩︎

\(X\subset \mathbb{P}(1^4,2)\) is given by the equation \[x_5^2=(x_1x_2-x_3x_4)^2+ (x_1x_2-x_3x_4) q_1(x_1,\dots,x_4)+q_2(x_1,\dots,x_4)^2,\] where \(q_1\) and \(q_2\) are general quadratic forms.

Case [rkCl612-deg4-quadric].↩︎

\(X\subset \mathbb{P}^5\) is given by the equations \[x_1x_2+x_3x_4+x_5^2+x_6l_1(x_1,\dots x_6)= x_1x_3+x_6l_2(x_1,\dots x_6)=0,\] where \(l_i\) are linear forms. It is easy to see that \(X\) contains two singular quadrics given by \(x_6=x_1=x_3x_4+x_5^2=0\) and \(x_6=x_3=x_1x_2+x_5^2=0\). They generate two pencils. Hence \(X\) is of type [rkCl612-deg4-quadric]. For a general choice of \(l_i\) the variety \(X\) has exactly one node.

Case [rkCl612-deg4-birational].↩︎

\(X\subset \mathbb{P}^5\) is given by the equations \[x_3x_4-x_5^2+x_6l_1(x_1,\dots x_6)= x_1x_4-x_2x_5+x_6 l_2(x_1,\dots x_6)=0,\] where \(l_i\) are general linear forms. Then \(X\) contains the plane \(\Pi:=\{x_4=x_5=x_6=0\}\). The singular locus of \(X\) consists of three nodes contained in \(\Pi\) and given there by \(x_3=l_1=0\) and \(x_2=x_3l_2-x_1l_1=0\).

Case [rkCl612-cubic-plane].↩︎

Let \(X\subset \mathbb{P}^4\) be given by the following equation: \[x_1u(x_1,x_2,x_3,x_4,x_5)+x_2v(x_1,x_2,x_3,x_4,x_5)=0,\] where \(u\) and \(v\) are quadratic forms. This cubic contains the plane \(\Pi:=\{x_1=x_2=0\}\) and, for general \(u\) and \(v\), the singular locus consists of four nodes. The projection from \(\Pi\) gives us a quadric bundle structure on \(\hat{X}\) (which is the blowup of \(\Pi\)).

Case [rkCl612-deg1-birational].↩︎

\(X\subset \mathbb{P}(1^3,2,3)\) is given by the equation \[x_5^2=x_4^3+ x_4^2\phi_2 + x_4\phi_4 + \phi_3^2,\] where \(\phi_i(x_1,x_2,x_3)\) are general homogeneous forms of degree \(i\).

Case 2.↩︎

\(X\subset \mathbb{P}(1^4,2)\) is given by the equation \[x_5^2=x_1\phi_3(x_1,\dots,x_4)+q(x_1,\dots,x_4)^2,\] where \(\phi_3\) and \(q\) are general homogeneous forms of degree \(3\) and \(2\), respectively.

6 Root systems↩︎

40. Let \(X\) be a del Pezzo threefold of degree \(d=\operatorname{d}(X)\). In this section we study the image of the restriction map \(\iota^*: \operatorname{Cl}(X) \to \operatorname{Pic}(S)\), where \(S\in |{-}\frac{1}{2}K_X|\) is a smooth member contained in the smooth locus of \(X\) and \(\iota: S\hookrightarrow X\) is an embedding. Define \(\Delta\) and \(\Delta'\) as in 6. If \(X\) is imprimitive, we apply construction ?? with all corresponding notation. In the primitive case, to unify notation, we put \(\sigma=\operatorname{id}\).

Note that \(S\) does not pass through singular points of \(X\). Thus we may identify \(S\) and \(\hat{S}=\xi^{-1}(S)\). Let \(\bar S:= \sigma(S)\). Then \(\bar S\) is a smooth del Pezzo surface, \(\bar S\in |{-}\frac{1}{2} K_{\bar X}|\) and \(\sigma_S: S\to \bar S\) is a blowup of \(\operatorname{r}(X)-\operatorname{r}(\bar X)\) distinct points. Define \(\bar \Delta\) and \(\bar \Delta'\) for \(\bar S\) as in 6.

41. Theorem. (i) In the above notation the image \(\iota^*\operatorname{Cl}(X)\) is the orthogonal complement to \(\Delta'\). In particular, \[\label{equation-roots-rank-Delta} \operatorname{rk}\Delta'+\operatorname{rk}\operatorname{Cl}(X)+\operatorname{d}(X)=10.\qquad{(3)}\]

(ii) We have \(\Delta'=\sigma_S^* \bar \Delta'\).

(iii) According to possibilities for \(Z\) we have the following cases:

  1. If \(Z\) is a point (i.e. \(\rho(\bar X)=1\)), then \(\bar \Delta'=\bar \Delta\). Here \(\bar \Delta'\) is of type \(E_8\), \(E_7\), \(E_6\), \(D_5\), \(A_4\), \(A_1\) in cases \(\operatorname{d}(\bar X)=1\), \(2\), \(3\), \(4\), \(5\), and \(8\), respectively.

  2. If \(Z\simeq \mathbb{P}^2\), then \(\bar \Delta'=\{\alpha \in \bar \Delta \mid \alpha\cdot f^* K_Z= 0\}\). Here \(\bar \Delta'\) is of type \(A_{m}\) (recall that \(\operatorname{d}(\bar X)=1\), \(2\), \(3\), \(5\), or \(6\)).

  3. If \(Z\simeq \mathbb{P}^1\), then \(\bar \Delta'=\{\alpha \in \bar \Delta \mid \alpha\cdot C= 0\}\), where \(C\) is a conic on \(\bar S\). Here \(\bar \Delta'\) is of type \(D_m\) (recall that \(\operatorname{d}(\bar X)=1\), \(2\), or \(4\)).3

  4. If \(X\simeq (\mathbb{P}^1)^3\), then \(\Delta'\) is the subsystem \(A_1\) in \(\Delta\simeq A_1\times A_2\).

  5. If \(X\) is of type 31[primitive-rho613-d612] or [primitive-rho613-d614], then \(\Delta'\) is of type \(A_5\) or \(A_3\), respectively.

Proof.

45. Assume that \(X\) is primitive. Then \(\hat{X}=\bar X\) and \(\sigma=\operatorname{id}\). All the statements are obvious if \(\operatorname{r}(X)=1\). We assume that \(\operatorname{r}(X)\ge2\). Let \(f: \hat{X}\to Z\) be an extremal \(K_{\hat{X}}\)-negative contraction. Let \(S\in |{-}\frac{1}{2}K_{\hat{X}}|\) be a smooth member. Denote by \(\delta: S\to Z\) the restriction of \(f\) to \(S\). Since \(f:\hat{X}\to Z\) is an extremal contraction, the image of \(\iota^*: \operatorname{Pic}(\hat{X}) \to \operatorname{Pic}(S)\) is generated by \(\delta^*\operatorname{Pic}(Z)\) and \(-K_{S}=-\frac{1}{2}K_{\hat{X}}|_{S}\). Clearly, \(f: S\to Z\) is surjective. Fix a standard basis in \(\operatorname{Pic}(S)\) [25]: \[\mathbf{h},\quad\mathbf{e}_1,\dots,\mathbf{e}_n,\] where \(n=9-d\) and \[\mathbf{h}^2=1, \quad \mathbf{e}_i^2=-1,\quad \mathbf{e}_i\cdot \mathbf{e}_j=0 \quad \text{for i\neq j.}\] Since \(\iota^*\operatorname{Pic}(\hat{X})\) is generated by \(\delta^*\operatorname{Pic}(Z)\) and \(-K_S\), we have \[\Delta'=\{\alpha\in \Delta \mid \alpha \cdot \delta^*\operatorname{Pic}(Z)=0\}.\]

42. Case \(Z\simeq \mathbb{P}^2\) and \(f\) is a \(\mathbb{P}^1\)-bundle. Then \(f: S\to \mathbb{P}^2\) is the blowup of \(n=9-d\) points and we can choose the basis \(\mathbf{h}\), \(\mathbf{e}_1,\dots,\mathbf{e}_n\) so that \(\mathbf{h}=f^*\mathscr{O}_{\mathbb{P}^2}(1)\) and \(\mathbf{e}_1,\dots,\mathbf{e}_n\) are \(f\)-exceptional. In this case, \(\iota^*\operatorname{Pic}(\hat{X})\) is generated by \(\mathbf{h}\) and \(K_S\). Hence \(\Delta'=\{\alpha\in \Delta \mid \alpha \cdot \mathbf{h}=0\}\). Then \(\Delta'=\{\mathbf{e}_i-\mathbf{e}_j \mid i\neq j\}\) is a root subsystem of rank \(n-1\) generated by \(\mathbf{e}_1-\mathbf{e}_2\),…,\(\mathbf{e}_{n-1}-\mathbf{e}_n\). Thus \(\Delta'\) is of type \(A_{n-1}\).

43. Case \(Z\simeq \mathbb{P}^1\), i.e. \(f\) is a quadric bundle. Then \(n\ge 4\) and \(\delta: S\to \mathbb{P}^1\) is a conic bundle. Let \(C\) be a fiber. By changing the basis \(\mathbf{h}\), \(\mathbf{e}_1,\dots,\mathbf{e}_n\) we may assume that \(C\sim \mathbf{h}-\mathbf{e}_1\). Then \(\Delta'=\{\alpha\in \Delta \mid \alpha\cdot C=0\}\), i.e. \(\Delta'\) consists of the following elements:

  • \(\mathbf{e}_i-\mathbf{e}_j\),\(i,j>1\), \(i\neq j\).

  • \(\pm(\mathbf{h}-\mathbf{e}_1-\mathbf{e}_i-\mathbf{e}_j)\),\(i,j>1\), \(i\neq j\).

Simple roots can be taken as follows: \[\mathbf{h}-\mathbf{e}_1-\mathbf{e}_2-\mathbf{e}_3, \quad \mathbf{e}_2-\mathbf{e}_3,\dots, \mathbf{e}_{n-1}-\mathbf{e}_n.\] Hence \(\Delta'\) is of type \(D_{n-1}\) if \(n\ge 5\) and \(A_3\) if \(n=4\).

44. Case \(Z\simeq \mathbb{P}^1\times \mathbb{P}^1\) and \(f\) is a \(\mathbb{P}^1\)-bundle. Let \(\ell_i:=F_i|_S\). Then we may assume that \(\ell_1\sim \mathbf{h}-\mathbf{e}_1\), \(\ell_2\sim \mathbf{h}-\mathbf{e}_2\). \(\Delta'\) consists of the following elements:

  • \(\mathbf{e}_i-\mathbf{e}_j\),\(i,j>2\), \(i\neq j\).

  • \(\pm (\mathbf{h}-\mathbf{e}_1-\mathbf{e}_2-\mathbf{e}_i)\),\(i>2\).

Simple roots can be taken as follows: \[\mathbf{h}-\mathbf{e}_1-\mathbf{e}_2-\mathbf{e}_3,\quad \mathbf{e}_3-\mathbf{e}_4,\dots, \mathbf{e}_{n-1}-\mathbf{e}_n.\] Thus \(\Delta'\) is of type \(A_{n-2}\).

This proves our theorem in the case where \(X\) is primitive.

46. Now consider the case where \(X\) is imprimitive. Obviously, the statement of (iii) follows from 45. There is a birational contraction \(\sigma :\hat{X}\to \bar X\), where \(\bar X\) is primitive and \(\sigma\) is a composition of blowups of smooth points. Let \(l:=\operatorname{r}(X)-\operatorname{r}(\bar X)\), let \(E_1,\dots, E_l\) be \(\sigma\)-exceptional divisors, and let \(\mathbf{e}_i=E_i\cap S\) for \(i=1,\dots, l\). By the above, the statement of our theorem holds for \(\bar X\) with root system \(\bar \Delta'\subset \bar \Delta\subset \operatorname{Pic}(\bar S)\). We have a commutative diagram \[\xymatrix{ \operatorname{Pic}(\bar S)\ar@{^(->}[r]^{\sigma^*_S} & \operatorname{Pic}(S) &\simeq&\operatorname{Pic}(\bar S)\oplus \sum_{i=1}^l \mathbf{e}_i\cdot \mathbb{Z} \\ \operatorname{Cl}(\bar X)\ar@{^(->}[r]^{\sigma^*}\ar[u]^{\bar \iota^*} &\operatorname{Cl}(X)\ar[u]^{\iota^*} &\simeq&\operatorname{Cl}(\bar X)\ar@<35pt>[u]^{\bar \iota^*}\oplus \sum_{i=1}^l E_i\cdot \mathbb{Z}\ar@{=}@<-35pt>[u] }\] Now it is easy to see that \(\iota^*\operatorname{Cl}(X)^\perp\subset \sigma_S^*\operatorname{Pic}(\bar S)\). Therefore, \[\sigma_S^*\bar \Delta'\subset \Delta\cap \iota^*\operatorname{Cl}(X)^{\perp} \subset \Delta\cap \sigma_S^*\operatorname{Pic}(\bar S).\] On the other hand, \(\sigma_S^*\bar \Delta'\supset \Delta\cap \sigma_S^*\operatorname{Pic}(\bar S)\). Hence, \(\sigma_S^*\bar \Delta'=\Delta\cap \iota^*\operatorname{Cl}(X)^{\perp}\). This proves (ii). As a consequence we have that the left hand side of ?? is preserved under birational contractions \(\sigma\). By 45 the equality ?? holds for primitive del Pezzo threefolds. Thus ?? holds for imprimitive ones as well. This proves (i).

 ◻

7 Del Pezzo threefolds with maximal \(\operatorname{r}(X)\)↩︎

Recall that \(\operatorname{r}(X)+\operatorname{d}(X)\le 9\) by Corollary 26. In this section we study del Pezzo threefolds with \(\operatorname{r}(X)+\operatorname{d}(X)=9\).

We say that points \(P_1,\dots, P_n\in \mathbb{P}^3\) are in general position if no three of them lie on one line and no four of them lie on one plane.

47. Theorem. Let \(X\) be a del Pezzo threefold with \(\operatorname{r}(X)+\operatorname{d}(X)=9\). Assume that \(X\not \simeq \mathbb{P}^1\times\mathbb{P}^1\times\mathbb{P}^1\). Then

  1. \(X\) can be obtained by applying construction ?? to \(\mathbb{P}^3\simeq V_8\subset \mathbb{P}^9\) where \(\sigma\) is the blowup of \(n:=\operatorname{r}(X)-1\) points \(P_1,\dots, P_n\in V_8\) in general position.

  2. Singular points of \(X\) are images of proper transforms of

    1. lines passing through \(P_i\) and \(P_j\), \(i\neq j\),

    2. twisted cubics passing through six distinct points \(P_{i_1},\dots, P_{i_6}\) (see Claim 49 below).

  3. If all the singularities of \(X\) are nodes, then \(\operatorname{s}(X)=28\), \(16\), \(10\), \(6\), \(3\), \(1\) in cases \(\operatorname{d}(X)=1\), \(2\), \(3\), \(4\), \(5\), \(6\), respectively.

  4. If \(\operatorname{d}(X)\ge 2\), then all the singularities of \(X\) are nodes.

Conversely, assume that \(X\) is a del Pezzo threefold whose singularities are at worst nodes and assume that \(\operatorname{s}(X)=28\), \(16\), \(10\), \(6\), \(3\), \(1\) in cases \(\operatorname{d}(X)=1\), \(2\), \(3\), \(4\), \(5\), \(6\), respectively. Then \(\operatorname{d}(X)+\operatorname{r}(X)=9\).

Note that in the case \(\operatorname{d}(X)=1\) the statement of [main-theorem-maximal-2] is wrong: one can easily construct \(X\) having only 27 singular points, where one of them is not a node.

48. Corollary. Let \(X\) be a del Pezzo threefold with \(\operatorname{r}(X)+\operatorname{d}(X)=9\). If \(\operatorname{d}(X)\ge 3\) and \(\operatorname{d}(X)\neq 6\), then \(X\) is unique up to isomorphism. If \(\operatorname{d}(X)=2\) (resp. \(\operatorname{d}(X)=1\)), then \(X\) belongs to a \(3\)-dimensional (resp. \(6\)-dimensional) family. There are exactly two isomorphism classes of Pezzo threefolds with \(\operatorname{d}(X)=6\), \(\operatorname{r}(X)=3\).

Proof. [main-theorem-maximal-0] If \(X\) is primitive, then either \(X\simeq \mathbb{P}^1\times\mathbb{P}^1\times\mathbb{P}^1\) or \(X\simeq \mathbb{P}^3\) by Theorems 15, 31, and 37. Thus we assume that \(X\) is imprimitive and \(\operatorname{d}(X)\le 7\). We use notation of Theorem 25. Run construction ?? in such a way that \(n\) is maximal possible. On the last step we get a primitive weak del Pezzo threefold \(\bar X\) with \(\rho(\bar X)=9-\operatorname{d}(\bar X)\). Moreover, if \(\rho(\bar X)=3\), then \(n=0\), \(\rho(\hat{X})=\operatorname{r}(X)=3\), and \(\operatorname{d}(X)=6\). By Theorem 31 we have \(X\simeq \mathbb{P}^1\times\mathbb{P}^1\times\mathbb{P}^1\). On the other hand, by Theorem 37 \(\rho(\bar X)\neq 2\). Hence \(\rho(\bar X)=1\), \(\operatorname{d}(\bar X)=8\), and then \(\bar X\simeq \mathbb{P}^3\).

It remains to show that the centers \(P_1,\dots,P_n\) of the blowup \(\hat{X}\to \bar X\simeq \mathbb{P}^3\) are in general position. Indeed, if distinct points \(P_i\), \(P_j\), \(P_k\) lie on a line \(L\subset \mathbb{P}^3\), then for its proper transform \(L'\) on \(\hat{X}\) we have \(-K_{\hat{X}}\cdot L'=-K_{\mathbb{P}^3}\cdot L-3\cdot 2<0\), a contradiction. Similarly, if four distinct points \(P_i\), \(P_j\), \(P_k\), \(P_l\) lie on a plane \(D\subset \mathbb{P}^3\), then then for its proper transform \(\hat{D}\) on \(\hat{X}\) we have \(K_{\hat{X}}^2\cdot \hat{D}=K_{\mathbb{P}^3}^2\cdot D-4\cdot 4=0\). Hence \(\hat{D}\) is contracted by the anticanonical map, a contradiction. This proves [main-theorem-maximal-0].

[main-theorem-maximal-1] Let \(P\in X\) be a singular point. Then \(\xi^{-1}(P)\) is a curve. Let \(\hat{C}\subset \xi^{-1}(P)\) be a component and let \(\bar C:=\sigma(\hat{C})\subset \bar X\). There are two members \(\hat{S}',\, \hat{S}''\in |{-}\frac{1}{2}K_{\hat{X}_{0}}|\) such that \(C \subsetneq\hat{S}'\cap\hat{S}''\). Then \(\bar C \subsetneq\bar S'\cap\bar S''\), where \(\bar S'\), \(\bar S''\subset \hat{X}_{0}=\mathbb{P}^3\) are proper transforms of \(\hat{S}'\) and \(\hat{S}''\). Therefore, \(\deg \bar C\le 3\) and \(\bar C\) is not a plane cubic. If \(\deg \bar C=2\), then \(\bar C\) is a conic and it must contain four distinct points from \(P_1,\dots, P_n\). This contradicts our assumption that \(P_1,\dots, P_n\) are in general position. Therefore, \(\bar C\) is either a line or a twisted cubic. This proves [main-theorem-maximal-1].

[main-theorem-maximal-3] follows by Corollary 93.

[main-theorem-maximal-2] If \(\operatorname{d}(X)\ge 3\), then \(X\) is unique up to isomorphism and the statement [main-theorem-maximal-2] can be checked directly (see 53-56 below). Let \(\operatorname{d}(X)=2\) the \(\xi\)-exceptional set consists of proper transforms of lines \(L_{i,j}\) passing through pairs of distinct points \(P_i\), \(P_j\) and one twisted cubic \(C\) passing through \(P_1,\dots, P_6\). Moreover, the lines \(L_{i,j}\) meet \(C\) transversely. By blowing the points \(P_1,\dots, P_6\) up we get these curves disjointed. Thus \(\xi\) is a small resolution whose exceptional set is a disjointed union of \(16\) smooth rational curves.

The last assertion follows by Corollary 93. ◻

49. Claim. Let \(P_1,\dots, P_6\in \mathbb{P}^3\) be a points in general position. Then there exists a twisted cubic curve \(C=C_3\subset \mathbb{P}^3\) containing \(P_1,\dots, P_6\). Such a curve is unique.

Proof. It is easy and left to the reader. ◻

By Theorem 41 we have the following.

50. Corollary. Let \(X\) be a del Pezzo threefold with \(\operatorname{r}(X)=9-\operatorname{d}(X)\) and \(\operatorname{d}(X)\le 5\). Then the image of \(\iota^*: \operatorname{Cl}(X)\to \operatorname{Pic}(S)\) is a sublattice orthogonal to some root \(\alpha\in \Delta\), i.e. \(\Delta'=\{\pm \alpha\}\). Moreover, \(\Delta''\) is of type \(E_7\), \(D_6\), \(A_5\), \(A_1\times A_3\), \(A_2\) in cases \(\operatorname{d}(X)=1\), \(2\), \(3\), \(4\), \(5\), respectively.

51. Corollary. Let \(X\) be a del Pezzo threefold with \(\operatorname{r}(X)=9-\operatorname{d}(X)\) and \(\operatorname{d}(X)\le 4\).

  1. If \(\operatorname{d}(X)\neq 2\), then the image of the natural map \(G\to \operatorname{Aut}(\Delta'')\) is contained in the Weyl group \(\operatorname{W}(\Delta'')\).

  2. If \(\operatorname{d}(X)\le 3\) and \(\Bbbk\) is algebraically closed (i.e. we are in the geometric case), then the map \(G\to \operatorname{Aut}(\Delta'')\) is an embedding.

Proof. (i) Similar to [15]. If \(\operatorname{d}(X)=1\), then \(\Delta''\) is of type \(E_7\) and \(\operatorname{Aut}(\Delta'')=\operatorname{W}(\Delta'')\) [26]. For \(\operatorname{d}(X)=3\) the group \(\operatorname{Aut}(\Delta'')\) is a direct product of \(\operatorname{W}(\Delta'')\) and \(\pm \operatorname{id}\). If the image of \(G\) is not contained in \(\operatorname{W}(\Delta'')\), then the element \(\tau:=-\operatorname{id}\) can be expressed as \(gw\), where \(g\in G\) and \(w\in \operatorname{W}(\Delta'')\). Note that any reflection \(s\in \operatorname{W}(\Delta'')\) can be extended to an element \(\operatorname{Aut}(\iota^*\operatorname{Cl}(X))\). Hence, the action of \(\tau\) can be extended to an action on \(\iota^*\operatorname{Cl}(X)\) so that \(\tau(K_S)=g w(K_S)=K_S\). Let \(E\) be a plane on \(X\) and let \(\mathbf{e}\) be the class \(\iota^*(E)\). Then \[\textstyle \tau(\mathbf{e})= \tau \bigl(\frac{1}{d} K_S+\mathbf{e}\bigr) -\frac{1}{d} \tau(K_S)=-\bigl(\frac{1}{d} K_S+\mathbf{e}\bigr)-\frac{1}{d} K_S= -\frac{2}{d} K_S-\mathbf{e}.\] In particular, \(2/d\) must be integral, a contradiction.

(ii) Let \(G_0\) be the kernel of the map \(G\to \operatorname{Aut}(\Delta'')\). Then \(G_0\) acts trivially on \(\operatorname{Cl}(X)\). In particular, the diagram ?? is \(G_0\)-equivariant. Thus \(G_0\) acts on \(\bar X=\mathbb{P}^3\) so that there are \(\ge 5\) fixed points in general position, the images of \(\sigma\)-exceptional divisors. Then \(G_0\) must be trivial. ◻

52. Theorem. Let \(X\) be a del Pezzo threefold with \(\operatorname{r}(X)+\operatorname{d}(X)=9\). Assume that \(X\not \simeq \mathbb{P}^1\times \mathbb{P}^1\times \mathbb{P}^1\). Let \(\Pi\subset X\) is a plane, let \(\hat{\Pi}\subset \hat{X}\) be its proper transform, and let \(\bar \Pi =\sigma(\hat{\Pi})\subset \bar X=\mathbb{P}^3\). Then \(\bar \Pi\) is of one of the following types:

  1. \(\bar \Pi\) is one of the points \(P_i\), \(\hat{\Pi}\) is \(\sigma\)-exceptional;

  2. \(\bar \Pi\) is a plane passing through three of the points \(P_i\);

  3. \(\bar \Pi\) is quadratic cone passing through six of the points \(P_i\) so that one of them is the vertex of the cone;

  4. (only for \(\operatorname{d}(X)=1\)) \(\bar \Pi\) is cubic surface passing through all the points \(P_i\) so that four of them are double points;

  5. (only for \(\operatorname{d}(X)=1\)) \(\bar \Pi\) is quartic surface passing through all the points \(P_i\) so that all of them are double points and one of them is a triple point.

The number of planes on \(X\) is given by the following table:

\(\operatorname{d}(X)\) \(7\) \(6\) \(5\) \(4\) \(3\) \(2\) \(1\)
\(\operatorname{p}(X)\) \(1\) \(2\) \(4\) \(8\) \(15\) \(32\) \(126\)

Proof. It is easy to see that all the subvarieties \(\Pi\) described in (i)-(v) are planes. So the number of planes is at least the number indicated in the table. On the other hand, for any plane \(\Pi\subset X\), the intersection \(\Pi \cap S\) is a line whose class in \(\operatorname{Pic}(S)\) is orthogonal to the root \(\alpha\in \operatorname{Pic}(S)\) (see Corollary 50). Define \[\mathcal{E}:=\{ e\in \operatorname{Pic}(S) \mid e^2=K_S\cdot e=-1,\, e\cdot \Delta'=0\}.\] Thus the number of planes is at most \(|\mathcal{E}|\).

Let \(\mathbf{h}\), \(\mathbf{e}_1,\dots,\mathbf{e}_{9-d}\) be a standard basis of \(\operatorname{Pic}(S)\). Since cases \(n\le 3\) are trivial, we may assume that \(n\ge 4\). Then the Weil group \(\operatorname{W}(\Delta)\) transitively acts on \(\Delta\) [25] and we can take it so that \(\alpha=\mathbf{e}_1-\mathbf{e}_2\). Now it is easy to compute \(\mathcal{E}\) (cf. [25]). For example, for \(d=6\) we have \(\mathcal{E}=\{\mathbf{e}_3,\, \mathbf{h}-\mathbf{e}_1-\mathbf{e}_2\}\), and for \(d=5\) we have \(\mathcal{E}=\{\mathbf{e}_4,\mathbf{e}_4, \mathbf{h}-\mathbf{e}_1-\mathbf{e}_2, \mathbf{h}-\mathbf{e}_3-\mathbf{e}_4\}\). Other cases are similar. For \(d=1\) we also can observe that \(\mathcal{E}=\Delta''+K_S\) and apply Corollary 50. ◻

Below we describe del Pezzo threefolds \(X\) with \(\operatorname{r}(X)+\operatorname{d}(X)=9\) explicitly and give examples. These threefolds were studied extensively in classical literature (see, e.g., [27]). We assume that \(X\) is singular (otherwise \(X\simeq \mathbb{P}^3\), \(V_7\), or \(\mathbb{P}^1\times \mathbb{P}^1\times \mathbb{P}^1\)).

53. Sextic del Pezzo threefold.Let \(X\subset \mathbb{P}^2\times \mathbb{P}^2\) be given by the equation \(x_1y_1+x_2y_2=0\). Then \(X\) is a del Pezzo threefold with \(\operatorname{d}(X)=6\) and \(\operatorname{r}(X)=3\). The singular locus consists of one node.

54. Quintic del Pezzo threefold (cf. [28]).Let \(X\subset \operatorname{Gr}(2,5)\) be an intersection of three general Schubert subvarieties of codimension one. Then \(X\) is a del Pezzo threefold with \(\operatorname{d}(X)=5\) and \(\operatorname{r}(X)=4\). The singular locus consists of three nodes.

55. Quartic del Pezzo threefold. Let \(X\subset\mathbb{P}^5\) be an intersection of two quadrics having \(6\) isolated singular points. Then in some coordinate system \(X\) can be given by the equations \[\label{equation-maximal-deg614} x_1^2-x_2^2=x_3^2-x_4^2=x_5^2-x_6^2.\tag{2}\] In [27] this variety is called the tetrahedral quartic threefold. By Corollary 93 \(\operatorname{r}(X)=5\). The variety \(X\) contains \(8\) planes \[\Pi_{\epsilon_1,\epsilon_2,\epsilon_3}=\{ x_1+ \epsilon_1 x_2=x_3 +\epsilon_2 x_4=x_5+\epsilon_3 x_6=0\},\] where \(\epsilon_i=\pm 1\). Clearly, \[\dim \Pi_{\epsilon_1,\epsilon_2,\epsilon_3}\cap \Pi_{\epsilon_1',\epsilon_2',\epsilon_3'}= -1+\frac{1}{2}\sum |\epsilon_i+\epsilon_i'|.\] Therefore, for each plane \(\Pi=\Pi_{\epsilon_1,\epsilon_2,\epsilon_3}\) there is exactly \(3\) planes \(\Pi'\) such that \(\Pi\cap \Pi'\) is a point and exactly \(3\) planes \(\Pi'\) such that \(\Pi\cap \Pi'\) is a line. Note that there are two \(4\)-tuples of planes such that planes in each tuple meet each other only by subsets of dimension \(\le 0\): \[\{\Pi_{+++},\, \Pi_{+--},\, \Pi_{-+-},\, \Pi_{--+}\},\quad \{\Pi_{---},\, \Pi_{-++},\, \Pi_{+-+},\, \Pi_{++-}\}.\] The involution \[\tau: (x_1,x_2,x_3,x_4,x_5,x_6) \longmapsto (x_1,-x_2,x_3,-x_4,x_5,-x_6)\] interchanges these \(4\)-tuples. Hence \(\tau\) induces a birational (cubo-cubic) involution on \(\mathbb{P}^3\). In [29] it is denoted by \(T_{\mathrm{tet}}\). Note however that \(\operatorname{Cl}(X)^{\tau}\not \simeq \mathbb{Z}\), i.e. \(X\) is not \(\tau\)-minimal. \(X\) is minimal with respect to the whole automorphism group.

56. Segre cubic. If \(\operatorname{d}(X)=3\), then \(X\) is can be given by \[\label{equation-Segre-cubic} X=X_3^{\mathrm s}=\left\{\sum_{i=1}^6 x_i= \sum_{i=1}^6 x_i^3=0\right\}\subset \mathbb{P}^4\subset \mathbb{P}^5.\tag{3}\] This cubic satisfies many remarkable properties (see [27]) and is called the Segre cubic. For example, any cubic hypersurface in \(\mathbb{P}^4\) has at most ten isolated singular points, this bound is sharp and achieved exactly for the Segre cubic (up to projective isomorphism). The symmetric group \(\mathfrak{S}_6\) acts on \(X_3^{\mathrm s}\) in the standard way. Moreover, by Corollaries 50 and 51 we see that \(\operatorname{Aut}(X_3^{\mathrm s})=\mathfrak{S}_6\), so the natural map \(\operatorname{Aut}(X_3^{\mathrm s})\to \operatorname{W}(\Delta'')\) is an isomorphism.

58. Quartic double solid. Let \(X\) be a del Pezzo threefold of degree \(2\). Let \(\phi: X \to \mathbb{P}^3\) be the half-anticanonical map. Then \(\phi\) is a double cover whose branch locus \(B\subset \mathbb{P}^3\) is a quartic having \(16\) singular points. It is well-known that such a quartic must be a Kummer surface, so the singularities of \(B\) and \(X\) are at worst nodes [30], [31] (see also [32]). The singular points of \(X\) correspond to \(15\) lines \(L_{ij}\) passing through pairs of points \(P_i\), \(P_j\) and one twisted cubic passing through all points \(P_1,\dots,P_6\). The threefold \(X\) contains \(32\) planes [27]. For each such a plane \(\Pi\) the image \(\pi(\Pi)\) is a plane touching \(B\) along a conic.

57. Example. Let \(B\subset \mathbb{P}^3\) be a surface given by the equation \(x_0^4+x_1^4+x_2^4+x_3^4-4x_0x_1x_2x_3=0\). Then the singular locus of \(S\) consists of \(16\) isolated points which are simple nodes. A double cover \(X\to \mathbb{P}^3\) branched along \(B\) is a del Pezzo threefold with \(\operatorname{d}(X)=2\) and \(\operatorname{r}(X)=7\).

60. Double Veronese cone. Recall that \(X\simeq X_6\subset \mathbb{P}(1^3,2,3)\). The projection from \((0,0,0,0,1)\) indices a double cover \(X\to \mathbb{P}(1^3,2)\) with branch divisor \(B=B_6\subset \mathbb{P}(1^3,2)\). Assume for simplicity that the singularities of \(X\) are at worst nodes. Then \(B\) is a surface having exactly \(28\) points of type \(A_1\). Conversely if \(B\subset \mathbb{P}(1^3,2)\) is a surface of degree 6 whose singularities are exactly \(28\) points of type \(A_1\), then the double cover of \(\mathbb{P}(1^3,2)\) branched at \(B\) is a del Pezzo threefold with \(\operatorname{d}(X)=1\) and \(\operatorname{r}(X)=8\). We refer to [33] for more detailed treatment and more references.

59. Example. Let \(C\subset \mathbb{P}^2\) is given by the equation \(f=x_1^4+x_2^4+x_3^4\). Then the dual curve \(C^*\) is given by \(f^*=(x_1^4+x_2^4+x_3^4)^3-27x_1^4x_2^4x_3^4\). It is easy to check that the discriminant of the polynomial \(h(t)=t^3-(x_1^4+x_2^4+x_3^4)t+2x_1^2x_2^2x_3^2\) is equal to \(4f^*\). The last polynomial defines a surface \(B\subset\mathbb{P}(1^3,2)\) of degree \(6\) having 28 singular points.

61. Corollary. Let \(X\) be a del Pezzo threefold such that \(\operatorname{d}(X)+\operatorname{r}(X)=9\) and \(\operatorname{d}(X)\neq 5\), \(6\), \(7\). Then \(X\) is a \(G\)-del Pezzo threefold with respect to some (geometric) group \(G\).

8 Del Pezzo threefolds with \(\operatorname{r}(X)=8-\operatorname{d}(X)\)↩︎

Let, as above, \(V_6\subset \mathbb{P}^7\) be a smooth del Pezzo threefold of degree \(6\) and let \(f_i: V_6\to \mathbb{P}^2\), \(i=1\), \(2\) be \(\mathbb{P}^1\)-bundles. We say that points \(P_1,\dots, P_n\in V_6\) are in general position if so are the points \(f_i(P_1),\dots, f_i(P_n)\in \mathbb{P}^2\) for \(i=1\) and \(2\).

62. Theorem. Let \(X\) be a del Pezzo threefold with \(\operatorname{r}(X)+\operatorname{d}(X)=8\). Then

  1. \(X\) can be obtained by applying construction ?? to \(V_6\subset \mathbb{P}^7\) where \(\sigma\) is the blowup of from \(n:=6-\operatorname{d}(X)\) points \(P_1,\dots, P_n\in V_6\) in general position.

  2. Singular points of \(X\) are images of proper transforms of

    1. curves of bidegree \((0,1)\) and \((1,0)\) passing through one of the points \(P_i\);

    2. curves of bidegree \((1,1)\) passing through two of the points \(P_i\);

    3. curves of bidegree \((2,2)\) passing through four of the points \(P_i\);

    4. (only for \(\operatorname{d}(X)=1\)) curves of bidegree \((2,3)\) and \((3,2)\) passing through all the points \(P_i\).

  3. If all the singularities of \(X\) are nodes, then \(\operatorname{s}(X)=27\), \(15\), \(9\), \(5\), \(2\), \(0\) in cases \(\operatorname{d}(X)=1\), \(2\), \(3\), \(4\), \(5\), \(6\), respectively.

  4. If \(\operatorname{d}(X)\ge 2\), then all the singularities of \(X\) are nodes.

Conversely, assume that \(X\) is a del Pezzo threefold whose singularities are at worst nodes and assume that \(\operatorname{s}(X)=27\), \(15\), \(9\), \(5\), \(2\), \(0\) in cases \(\operatorname{d}(X)=1\), \(2\), \(3\), \(4\), \(5\), \(6\), respectively. Then \(\operatorname{d}(X)+\operatorname{r}(X)=8\).

Proof. Run construction ?? so that \(n\) is maximal possible. On the last step we get a primitive weak del Pezzo threefold \(\bar X\) with \(\rho(\bar X)=8-\operatorname{d}(\bar X)<8\). Moreover, if \(\rho(\bar X)=3\), then \(n=0\), \(\rho(\hat{X})=\operatorname{r}(X)=3\), and \(\operatorname{d}(X)=5\). This is impossible by Theorem 31. Therefore, \(\rho(\bar X)=2\) and \(\operatorname{d}(\bar X)=6\). By Theorem 37 we have only one possibility: \(\bar X\simeq V_6\). ◻

63. Corollary. Let \(X\) be a del Pezzo threefold with \(\operatorname{r}(X)+\operatorname{d}(X)=8\). If \(\operatorname{d}(X)\ge 5\), then \(X\) is unique up to isomorphism. There are exactly two isomorphism classes of del Pezzo threefolds with \(\operatorname{d}(X)=\operatorname{r}(X)=4\).

Proof. Indeed, in the case \(\operatorname{d}(X)=4\) two non-isomorphic del Pezzo threefolds \(X\) are obtained by blowing up a couple of points corresponding to flags \((L_1,P_1)\), \((L_2,P_2)\in \operatorname{F}(\mathbb{P}^2)=V_6\) such that either \(L_1\cap L_2\neq P_i\) or \(L_1\cap L_2= P_i\). ◻

Similar to Theorem 52 one can prove the following.

64. Theorem. Let \(X\) be a del Pezzo threefold with \(\operatorname{r}(X)+\operatorname{d}(X)=8\). Let \(\Pi\subset X\) is a plane, let \(\hat{\Pi}\subset \hat{X}\) be its proper transform, and let \(\bar \Pi =\sigma(\hat{\Pi})\subset \bar X=V_6\). Then \(\bar \Pi\) is of one of the following types:

  1. \(\bar \Pi\) is one of the points \(P_i\), \(\hat{\Pi}\) is \(\sigma\)-exceptional;

  2. \(f_j(\bar \Pi)\) is a line for \(j=1\) or \(2\), and \(\bar \Pi\) contains two of the points \(P_i\);

  3. \(\bar \Pi\) is an element of \(|{-}\frac{1}{2} K_{V_6}|\) passing through four of the points \(P_i\) so that one of them is a double point;

  4. (only for \(\operatorname{d}(X)=1\)) \(f_j(\bar \Pi)\) is a conic for \(j=1\) or \(2\), and \(\bar \Pi\) contains all the points \(P_i\);

  5. (only for \(\operatorname{d}(X)=1\)) \(\bar \Pi\) is an element of \(|{-}K_{V_6}|\) passing through all of the points \(P_i\) so that all of them are double points and one of them is triple;

  6. (only for \(\operatorname{d}(X)=1\)) \(\bar \Pi\) is an element of \(|{-}K_{V_6}-f_j^*\mathscr{O}_{\mathbb{P}^2}(1)|\), where \(j=1\) or \(2\), passing through all of the points \(P_i\) so that three of them are double points.

The number of planes on \(X\) is given by the following table:

\(\operatorname{d}(X)\) \(6\) \(5\) \(4\) \(3\) \(2\) \(1\)
\(\operatorname{p}(X)\) \(0\) \(1\) \(4\) \(9\) \(20\) \(72\)

65. Corollary. Let \(X\) be a del Pezzo threefold with \(\operatorname{r}(X)=8-\operatorname{d}(X)\) and \(\operatorname{d}(X)\le 5\). Then in some standard basis of \(\operatorname{Pic}(S)\) the image of \(\iota^*: \operatorname{Cl}(X)\to \operatorname{Pic}(S)\) is a sublattice orthogonal to roots \(\mathbf{e}_1-\mathbf{e}_2, \, \mathbf{e}_2-\mathbf{e}_3\in \Delta\), i.e. \(\Delta'=\{\pm \mathbf{e}_1\mp\mathbf{e}_2,\, \pm\mathbf{e}_2\mp\mathbf{e}_3, \, \pm (\mathbf{h}-\mathbf{e}_1-\mathbf{e}_2-\mathbf{e}_3) \}\). Moreover, \(\Delta''\) is of type \(E_6\), \(A_5\), \(2A_2\), \(2A_1\) in cases \(\operatorname{d}(X)=1\), \(2\), \(3\), \(4\), respectively.

Similar to Corollary 51 (ii) we have.

66. Corollary. Let \(X\) be a del Pezzo threefold with \(\operatorname{r}(X)=8-\operatorname{d}(X)\) and \(\operatorname{d}(X)\le 2\). If \(\Bbbk\) is algebraically closed (i.e. we are in the geometric case), then the map \(G\to \operatorname{Aut}(\Delta'')\) is an embedding.

Now we give some examples.

68. Quintic del Pezzo threefold (cf. [28]).Let \(X\subset \operatorname{Gr}(2,5)\) be an intersection of two general Schubert subvarieties of codimension one and one general hyperplane section. Then \(X\) is a del Pezzo threefold with \(\operatorname{d}(X)=5\) and \(\operatorname{r}(X)=3\). The singular locus consists of two nodes.

67. Corollary (cf. [28], [10]). Let \(X\) be a del Pezzo threefold of degree \(5\). Then the singularities of \(X\) are at worst nodes and one of the following holds:

  1. \(X\simeq V_5\), a smooth del Pezzo quintic threefold;

  2. \(\operatorname{s}(X)=1\), \(\operatorname{r}(X)=2\), \(\operatorname{p}(X)=0\), and \(X\) is of type 37, Case [rkCl612-degree615];

  3. \(\operatorname{s}(X)=2\), \(\operatorname{r}(X)=3\), \(\operatorname{p}(X)=1\), and \(X\) is of type 68;

  4. \(\operatorname{s}(X)=3\), \(\operatorname{r}(X)=4\), \(\operatorname{p}(X)=4\), and \(X\) is of type 54.

Proof. Assertions (iii) or (iv) follows by the results of this and previous sections. If \(\operatorname{r}(X)=2\), then we have case (ii) by Theorem 37. Finally, if \(X\) is factorial, then it is smooth by Corollary 38. ◻

69. Quartic del Pezzo threefold.Let \(X\subset \mathbb{P}^5\) be given by the equations \[x_1^2+x_1x_3+x_2x_5=x_1x_3+x_3^2+ x_4x_6=0.\] Then \(X\) is a del Pezzo threefold of degree \(4\) containing exactly \(5\) nodes. By Corollary 93 \(\operatorname{r}(X)\ge 4\). On the other hand, \(X\) is not of type 55 because \(\operatorname{s}(X)<6\). Hence \(\operatorname{r}(X)=4\).

70. Cubic hypersurface. Let \(X\subset \mathbb{P}^4\) be given by the equation \[x_1x_2\ell(x_1\dots,x_5)+(x_3x_4+x_1x_2)x_5=0,\] where \(\ell\) is a general linear form. Then \(X\) is a cubic del Pezzo threefold with \(\operatorname{s}(X)=9\), \(\operatorname{r}(X)=5\), and \(\operatorname{p}(X)=9\) (cf. [16]).

71. Quartic double solid. Let \(Y\) be a hypersurface in \(\mathbb{P}^4\) given by \(\{s_1=4s_4-s_2^2=0\}\subset \mathbb{P}^5\), where \(s_k=\sum x_i^k\). This famous hypersurface is called Igusa quartic. The singular locus of \(Y\) consists of \(15\) lines. Consider a general hyperplane section \(B:=Y\cap \mathbb{P}^3\). Then \(B\) is a quartic having \(15\) nodes (cf. [32]). Let \(X\to \mathbb{P}^3\) be a double cover with branch divisor \(B\). Then \(X\) is a del Pezzo threefold of degree \(2\) with \(\operatorname{s}(X)=15\) and \(\operatorname{r}(X)=6\).

9 \(G\)-del Pezzo threefolds↩︎

72. In this section we prove Theorem 7. We use notation of 40. Furthermore we assume that \(X\) is a \(G\)-del Pezzo threefold. Thus \(\operatorname{Cl}(X)^G\simeq \mathbb{Z}\). By Theorem 37 we may assume that \(\operatorname{r}(X)\ge 3\).

73. Lemma. In the above notation, if \(\operatorname{d}(X)\ge 5\), then \(X\simeq \mathbb{P}^1\times \mathbb{P}^1\times \mathbb{P}^1\).

Proof. Assume that \(X\not \simeq \mathbb{P}^1\times \mathbb{P}^1\times \mathbb{P}^1\). Then \(X\) is singular and \(\operatorname{d}(X)=6\) or \(5\) by Theorems 15.

Consider the case \(\operatorname{d}(X)=6\). Since \(\operatorname{r}(X)\ge 3\), our \(X\) is described in 53. Then \(X\) contains exactly two planes \(\Pi_1\), \(\Pi_2\) and the divisor \(\Pi_1+\Pi_2\) is \(G\)-invariant. Hence \(\Pi_1+\Pi_2\sim a S\) for some positive integer \(a\). Comparing degrees we get \(2=6a\), a contradiction.

Now let \(\operatorname{d}(X)=5\). By Lemma 17 we may assume that \(X\) is not factorial. In this situation, \(X\) is imprimitive. The same arguments as above show that the number of planes on \(X\) in any \(G\)-orbit must be divisible by \(5\). This contradicts Corollary 67. ◻

75. From now on we assume that \(\operatorname{d}(X)\le 4\). By Theorem 31 we may assume that \(X\) is imprimitive. Let \(S\in |{-}\frac{1}{2} K_X|\) be a general member. Let \(n:=\operatorname{rk}\Delta=9-\operatorname{d}(X)\).

74. Lemma.If in the above notation \(\operatorname{d}(X)\le 4\), then \(X\) contains at least two planes \(\Pi_1\), \(\Pi_2\) such that \(\dim \Pi_1\cap \Pi_2\le 0\).

Proof. Since \(X\) is imprimitive, it contains at least one plane \(\Pi_1\). Let \(\Pi_1,\dots,\Pi_l\) be its orbit. Since \(\operatorname{Cl}(X)^G=\mathbb{Z}\cdot S\), \(k\ge 4\). If \(\dim \Pi_i\cap \Pi_j\ge 1\) for all \(i\), \(j\), the linear span of \(\Pi_1,\dots,\Pi_k\) is three-dimensional and so \(X\) cannot be an intersection of quadrics. ◻

First we consider the case \(\Delta''=\emptyset\).

76. Proposition. If in the above notation \(\Delta''=\emptyset\), then \(\operatorname{d}(X)=3\), \(\operatorname{r}(X)=3\), \(\operatorname{p}(X)=3\), and \(X\) is a projection of a del Pezzo threefold \(Y=Y_4\subset \mathbb{P}^5\) of type [rkCl612-deg4-quadric] from a point. If moreover the singularities of \(X\) are at worst nodes, then by [16], \(X\) is of type J11 or J12, and \(6\le \operatorname{s}(X)\le 7\).

Proof. Let \(\mathbf{e}_{n-m+1},\dots, \mathbf{e}_n\) correspond to blowups \(\sigma\). If \(m>1\), then \(\mathbf{e}_{n-1}-\mathbf{e}_n\in \Delta''\). Thus, \(m=1\), \(\operatorname{d}(\bar X)=\operatorname{d}(X)+1\), and we may assume that every two planes on \(X\) meet each other by a subset of dimension \(1\). Therefore, \(\operatorname{r}(\bar X)=2\) and \(\operatorname{r}(X)=3\). By Lemma 74 \(\operatorname{d}(X)\le 3\). Therefore, \(\operatorname{d}(\bar X)\le 4\), and \(8\ge n\ge 6\).

Consider the case where \(f:\bar X\to Z=\mathbb{P}^2\) is a \(\mathbb{P}^1\)-bundle. Then by Theorem 41 we may assume that the vectors \(\mathbf{e}_1-\mathbf{e}_2\), …, \(\mathbf{e}_{n-2}-\mathbf{e}_{n-1}\) form a basis of \(\Delta'\). We have:

  • \(n=6\), \(\operatorname{d}(X)=3\) \(\Longrightarrow\) \(2\mathbf{h}-\sum \mathbf{e}_i\in \Delta''\),

  • \(n=7\), \(\operatorname{d}(X)=2\) \(\Longrightarrow\) \(2\mathbf{h}+\mathbf{e}_7-\sum \mathbf{e}_i\in \Delta''\),

  • \(n=8\), \(\operatorname{d}(X)=1\) \(\Longrightarrow\) \(3\mathbf{h}-\mathbf{e}_8-\sum \mathbf{e}_i\in \Delta''\).

Thus, in all cases we have \(\Delta''\neq \emptyset\), a contradiction.

Consider the case where \(f:\bar X\to Z=\mathbb{P}^1\) is a quadric bundle. Then \(\operatorname{d}(X)=\operatorname{d}(\bar X)-1=1\) or \(3\) by Theorem 37. Again by Theorem 41 we may assume that vectors \[\mathbf{h}-\mathbf{e}_1-\mathbf{e}_2-\mathbf{e}_3,\quad \mathbf{e}_2-\mathbf{e}_3, \dots,\mathbf{e}_{n-2}-\mathbf{e}_{n-1}\] form a basis of \(\Delta'\). If \(n=8\), then \(3\mathbf{h}-\mathbf{e}_8-\sum \mathbf{e}_i\in \Delta''\), a contradiction. Therefore, \(\operatorname{d}(X)=3\) and \(\bar X\) is of type [rkCl612-deg4-quadric]. Thus \(X\) is a cubic in \(\mathbb{P}^4\). Since for any two planes \(\Pi_i,\, \Pi_j\subset X\) we have \(\dim \Pi_i\cap \Pi_j\ge 1\), all the planes on \(X\) are contained in one hyperplane. Hence \(\operatorname{p}(X)=3\). By Proposition 91\(\operatorname{s}(X)\le 7-h^{1,2}(\hat{X})\). If the singularities of \(X\) are at worst nodes, then \(X\) is of type J11 or J12 by [16]. ◻

77. Example. Consider the cubic \(X\subset \mathbb{P}^4\) given by the equation \[x_1x_2x_3+x_0\left(\lambda x_0^2+x_1^2+x_2^2+x_3^2+x_4^2\right)=0.\] Then \(X\) has \(6\) (resp. \(7\)) nodes if \(\lambda\neq 0\) (resp. \(\lambda= 0\)). It is easy to see that \(X\) contains at least \(3\) planes, so \(\operatorname{r}(X)\ge 3\). By [16] we have \(\operatorname{r}(X)= 3\) and \(\operatorname{p}(X)=3\). The symmetric group \(\mathfrak{S}_3\) acts on \(X\) by permutations of \(x_1\), \(x_2\), \(x_3\) so that \(X\) is a \(G\)-Fano threefold.

78. Now we assume that \(\Delta''\neq \emptyset\). Then \(\Delta''\) is a \(G\)-invariant root subsystem in \(\Delta\). By the results of §7 and §8 we may assume that \(\operatorname{r}(X)\le 7-\operatorname{d}(X)\). Further, by Lemma 74 \(\operatorname{d}(X)\le 3\).

82. Consider the case \(\operatorname{d}(X)=3\). There are only the following possibilities:

79. \(\operatorname{d}(\bar X)=4\), \(\operatorname{r}(\bar X)=2\), \(\bar X\) is of type [rkCl612-deg4-quadric], \(\operatorname{r}(X)=3\). Then \(\Delta'\) is described in 43: it is of type \(D_4\) and generated by \(\mathbf{h}-\mathbf{e}_1-\mathbf{e}_2-\mathbf{e}_3\), \(\mathbf{e}_2-\mathbf{e}_3\),\(\mathbf{e}_3-\mathbf{e}_4\),\(\mathbf{e}_4-\mathbf{e}_5\). Any root \(\alpha\in \Delta\) has the form \(\alpha= \pm (\mathbf{e}_i- \mathbf{e}_j)\), \(\pm (\mathbf{h}-\mathbf{e}_i-\mathbf{e}_j-\mathbf{e}_k)\) or \(\pm (2\mathbf{h}-\sum \mathbf{e}_i)\) (see e.g., [15]). Since \(\iota^*\operatorname{Cl}(X)=\Delta'^{\perp}\), we get \(\Delta''=\emptyset\), a contradiction.

80. \(\operatorname{d}(\bar X)=5\), \(\operatorname{r}(\bar X)=1\), \(\bar X=V_5\), \(\operatorname{r}(X)=3\). Similarly, \(\Delta'\) is of type \(A_4\) and generated by \(\mathbf{h}-\mathbf{e}_1-\mathbf{e}_2-\mathbf{e}_3\), \(\mathbf{e}_1-\mathbf{e}_2\),\(\mathbf{e}_2-\mathbf{e}_3\),\(\mathbf{e}_3-\mathbf{e}_4\). In this case, \(\Delta''=\{\pm (\mathbf{e}_5-\mathbf{e}_6)\}\). It is easy to see that the group \(G\) permutes elements \(\mathbf{e}_5, \,\mathbf{e}_6\in \iota^*\operatorname{Cl}(X)\). But then the class of \(\mathbf{e}_5+\mathbf{e}_6\) must be \(G\)-invariant, so it is proportional to \(-K_S\), a contradiction.

81. \(\operatorname{d}(\bar X)=5\), \(\operatorname{r}(\bar X)=2\), \(\bar X\) is of type [rkCl612-degree615], \(\operatorname{r}(X)=4\). Similarly, \(\Delta'\) is of type \(A_3\) and generated by \(\mathbf{e}_1-\mathbf{e}_2\),\(\mathbf{e}_2-\mathbf{e}_3\),\(\mathbf{e}_3-\mathbf{e}_4\). Then \(\Delta''=\{\pm (2\mathbf{h}-\sum \mathbf{e}_i),\;\pm (\mathbf{e}_5-\mathbf{e}_6)\}\). There is a unique element (class of a line on \(S\)) \(\mathbf{x}\in (\Delta'+\Delta'')^{\perp}\) such that \(\mathbf{x}^2=K_X\cdot \mathbf{x}=-1\): \[\mathbf{x}=\mathbf{h}-\mathbf{e}_5-\mathbf{e}_6.\] But then \(x\in \iota^*\operatorname{Cl}(X)\) and \(x\) must be \(G\)-invariant, a contradiction.

83. Finally we consider cases \(\operatorname{d}(X)\le 2\). According to Remark 16 any del Pezzo threefold with \(\operatorname{d}(X)\le 2\) is automatically \(G\)-del Pezzo. Thus all the possibilities for \(\bar X\) with \(2\le \operatorname{d}(\bar X)\le 5\) and \(\operatorname{r}(\bar X)\le 2\) do occur (recall that \(3\le \operatorname{r}(X)\le 7-\operatorname{d}(X)\)):

  • \(\bar X=V_5\) \(\Longrightarrow\) \(\operatorname{d}(X)\le 2\), \(\Delta'\simeq A_4\);

  • \(\bar X=V_4\) \(\Longrightarrow\) \(\operatorname{d}(X)\le 2\), \(\Delta'\simeq D_5\);

  • \(\bar X=V_3\) \(\Longrightarrow\) \(\operatorname{d}(X)\le 2\), \(\Delta'\simeq E_6\);

  • \(\bar X\) is of type [rkCl612-P1-bundle-degree612] \(\Longrightarrow\) \(\operatorname{d}(X)=1\), \(\Delta'\simeq A_6\);

  • \(\bar X\) is of type [rkCl612-P1-bundle-degree613] \(\Longrightarrow\) \(\operatorname{d}(X)\le 2\), \(\Delta'\simeq A_5\);

  • \(\bar X\) is of type [rkCl612-degree615] \(\Longrightarrow\) \(\operatorname{d}(X)\le 2\), \(\Delta'\simeq A_3\);

  • \(\bar X\) is of type [rkCl612-deg2-quadric] \(\Longrightarrow\) \(\operatorname{d}(X)=1\), \(\Delta'\simeq D_6\);

  • \(\bar X\) is of type [rkCl612-deg4-quadric] \(\Longrightarrow\) \(\operatorname{d}(X)\le 2\), \(\Delta'\simeq D_4\).

The number of planes can be found by using Lemma 21 and direct computations.

84. Example. Let \(X\subset \mathbb{P}(1^4,2)\) is given by the equation \[y^2= x_1x_2x_3x_4+\lambda (x_1^2+x_2^2+x_3^2+x_4^2)^2\] where \(\lambda\) is a constant. Then \(X\) has exactly \(12\) nodes and contains \(8\) planes. By Corollary 93 \(\operatorname{r}(X)\ge 3\). Further, by our classification \(X\) is of type [class-2-d4].

More examples of del Pezzo threefolds with \(\operatorname{d}(X)=2\) can be constructed similarly by writing down explicit equations (cf. [32]).

10 Appendix: number of singular points of Fano threefolds↩︎

85. Definition. Let \(V\ni P\) be a threefold terminal Gorenstein (=isolated cDV) singularity. We say that \(V\ni P\) is r-nondegenerate (resolution nondegenerate) if there is a resolution \[\sigma:V_m \overset{\sigma_m} {\longrightarrow} \cdots \overset{\sigma_2} {\longrightarrow} V_1\overset{\sigma_1} {\longrightarrow} V=V_0,\] where each \(\sigma_i\) is a blowup of a singular point \(P_{i-1}\in V_{i-1}\). Such a resolution \(\sigma\) is called standard. In this situation, all varieties \(V_i\) also have only isolated cDV singularities. If furthermore each \(\sigma_i\)-exceptional divisor \(E_i\subset V_i\) is irreducible, then we say that \(V\ni P\) is rs-nondegenerate (strongly resolution nondegenerate).

Denote \(\lambda(V,P):=m\) and let \(\nu(V,P)\) be the number of \(\sigma\)-exceptional divisors. Thus \(\lambda(V,P)\le \nu(V,P)\) and the equality holds if and only if \(V\ni P\) is rs-nondegenerate.

86. Remark. Let \(V\ni P\) be a threefold terminal Gorenstein point and let \(\sigma_1: V_1\to V\) be the blowup of \(P\). Since \(V\ni P\) is a hypersurface singularity, we have an (analytic) embedding \(V_1\subset \tilde{\mathbb{C}}^4\), where \(\tilde{\sigma}_1:\tilde{\mathbb{C}}^4\to \mathbb{C}^4\) is the blowup of the origin. Let \(D:=\tilde{\sigma}_1^{-1}(P)\) be the exceptional divisor. Then \(D\simeq \mathbb{P}^3\). Since \(V\ni P\) is a singularity of multiplicity \(2\), the intersection \(D\cap V_1\) is a quadric in \(\mathbb{P}^3\) (possibly reducible or non-reduced). If \(D\cap V_1\) irreducible, then \(V_1\) is either smooth or has (a unique) terminal singularity. Moreover, the above arguments show that \(2\lambda(V,P)\ge \nu(V,P)\).

87. Proposition. Let \((V\ni 0)\subset \mathbb{C}^4\) be a singularity given by \(t^2=\phi(x,y,z)\), where \(\phi=0\) is an equation of a Du Val singularity. Then \(V\ni 0\) is r-nondegenerate. Moreover, if \(\phi=0\) defines a singularity of type \(A_n\), then \(V\ni 0\) is rs-nondegenerate.

Proof. Direct computation. ◻

88. Corollary. Let \(X\) be a del Pezzo threefold with \(\operatorname{d}(X)\le 2\). Assume that the branch divisor \(B\) of the double cover \(\varphi: X\to \mathbb{P}(1^{3},2)\) (resp. \(\varphi: X\to \mathbb{P}^3\)) has only Du Val singularities (see 16). Then the singularities of \(X\) are r-nondegenerate. If moreover \(B\) has only singularities of type \(A\), then the singularities of \(X\) are rs-nondegenerate.

89. Let \(W\) be a smooth projective fourfold and let \(V\subset W\) be an effective divisor. Define \[\beta(W,V):=c_3(W)\cdot V-c_2(W)\cdot V^2+c_1( W)\cdot V^3- V^4.\] If \(V\) is smooth then \(\beta(W,V)\) coincides with \(\deg c_3(V)=\operatorname{Eu}(V)\), the topological Euler number.

90. Lemma. In the above notation let \(P\in V\) be a singular point, let \(\sigma: \tilde{W}\to W\) be the blowup of \(P\), and let \(\tilde{V}\subset \tilde{W}\) be the proper transform of \(V\). Then \(\beta(\tilde{W},\tilde{V})=\beta(W,V)+4\).

Proof. Direct computation. See the arxiv version of the paper. ◻

91. Proposition. Let \(X\) be a Gorenstein Fano threefold whose singularities are r-nondegenerate terminal points. Assume that

  1. \(X\) can be embedded to a smooth fourfold so that a general member \(X'\in |X|\) is smooth.

Then \[\begin{gather} {\sum_{P\in X}}'\lambda(X,P)\le \sum_{P\in X} \bigl(2\lambda(X,P)-\nu(X,P)\bigr)= \\ = \operatorname{r}(X)-\rho(X)+h^{1,2}(X')-h^{1,2}(\hat{X}), \end{gather}\] where \(\hat{X}\to X\) is the standard resolution and the first sum runs through all rs-nondegenerate points \(P\in X\).

Proof. Put \(\lambda:=\sum_{P\in X} \lambda(X,P)\). Thus \[\begin{gather} 2+2\rho(\hat{X})-2h^{1,2}(\hat{X})=\operatorname{Eu}(\hat{X})=\beta(\hat{Y},\hat{X})=\beta(\hat{Y},\hat{X})+4\lambda= \\ =\operatorname{Eu}(X')+4\lambda=2+2\rho(X')-2h^{1,2}(X')+4\lambda. \end{gather}\] Since \(\rho(\hat{X})= \operatorname{r}(X)+\sum \nu(X,P)\), this gives the desired inequality. ◻

92. Remark. The condition (*) is automatically satisfied if \(X\) is a del Pezzo threefold (see Theorem 15).

93. Corollary. In the notation of 91 assume additionally that the singularities are rs-nondegenerate. Then \[|\operatorname{Sing}(X)| \le \operatorname{r}(X)-\rho(X)+h^{1,2}(X')-h^{1,2}(\hat{X}).\] The equality holds, if all the singularities are nodes.

11 Concluding remarks and open questions↩︎

We would like to propose the following open questions.

94. Give a complete birational classification of del Pezzo threefolds over \(\mathbb{C}\). Non-trivial cases only are factorial del Pezzo threefolds of degree \(\le 3\). All other cases can be reduced to the above ones by using construction ?? (or birationally trivial). It is well-known that a three-dimensional cubic hypersurface with at worst cDV singularities is rational if and only if it is singular [34]. A general smooth (and, in some cases, factorial) del Pezzo threefold of degree \(\le 2\) is not rational [35], [36], [37].

96. Give a complete birational classification of del Pezzo threefolds over algebraically non-closed fields. Here is one example.

95. Theorem. Let \(X\) be a smooth del Pezzo threefold of degree \(5\) over a field \(\Bbbk\). Then \(X\) is \(\Bbbk\)-rational.

Proof. Denote \(\bar X:=X\otimes \bar \Bbbk\). Let \(\Gamma:=\Gamma(X)\) be the Hilbert scheme parameterizing the family of lines on \(X\). It is known that \(\bar \Gamma:=\Gamma\otimes \bar \Bbbk\simeq \mathbb{P}^2_{\bar \Bbbk}\) (see [8], [38]). Moreover, lines with normal bundle \(N_{l/X}\simeq \mathscr{O}(-1)\oplus \mathscr{O}(1)\) are parametrized by some conic \(C\subset \Gamma\) [38]. The conic \(C\) contains a point of degree \(\le 2\). Therefore, there is a line \(\ell \subset \Gamma\) defined over \(\Bbbk\). Let \(H_{\ell}\) be the union of all lines \(L\subset X\) whose class is contained in \(\ell\subset \Gamma\). Then \(H_{\ell}\) in an element of \(|{-}\frac{1}{2} K_X|\) defined over \(\Bbbk\) [8]. In particular, \(\operatorname{Pic}(X)=\mathbb{Z}\cdot \frac{1}{2} K_X\) and the linear system \(|{-}\frac{1}{2} K_X|\) defines an embedding \(X\subset \mathbb{P}^6_{\Bbbk}\). A general pencil of hyperplane sections defines a structure of del Pezzo fibration of degree \(5\) on \(X\). By [15] the variety \(X\) is \(\Bbbk\)-rational. ◻

97. Describe automorphism groups of del Pezzo threefolds over an algebraically closed fields. Which of them are birationally rigid (cf. [39], [40])? These questions are very useful for applications to the classification of finite subgroups of Cremona group \(\operatorname{Cr}_3(\Bbbk)\) [3], [17].

References↩︎

[1]
Yu. I. Manin. Rational surfaces over perfect fields. II. Mat. Sb. (N.S.), 72 (114):161–192, 1967.
[2]
S. Mori. Flip theorem and the existence of minimal models for \(3\)-folds. J. Amer. Math. Soc., 1(1):117–253, 1988.
[3]
Y. Prokhorov. Simple finite subgroups of the Cremona group of rank 3. J. Algebraic Geom.(to appear) ArXiv e-prints 0908.0678.
[4]
V. A. Iskovskikh. Minimal models of rational surfaces over arbitrary fields. Math. USSR-Izv., 14(1):17–39, 1980.
[5]
V. A. Iskovskikh and Yu. Prokhorov. Fano varieties. Algebraic geometry. V. Encyclopaedia Math. Sci.47 Springer, Berlin, 1999.
[6]
M. Reid. Young person’s guide to canonical singularities. In Algebraic geometry, Bowdoin, 1985 (Brunswick, Maine, 1985), Proc. Sympos. Pure Math.46, 345–414. Amer. Math. Soc., Providence, RI, 1987.
[7]
Y. Kawamata. Crepant blowing-up of \(3\)-dimensional canonical singularities and its application to degenerations of surfaces. Ann. of Math. (2), 127(1):93–163, 1988.
[8]
V. A. Iskovskikh. J. Sov. Math., 13:745–814, 1980.
[9]
T. Fujita. On the structure of polarized manifolds with total deficiency one. I-III. J. Math. Soc. Japan, 32, 33, 36:709–725, 415–434, 75–89, 1980, 1981, 1984.
[10]
T. Fujita. Projective varieties of \(\Delta\)-genus one. In Algebraic and topological theories (Kinosaki, 1984), 149–175. Kinokuniya, Tokyo, 1986.
[11]
K.-H. Shin. -dimensional Fano varieties with canonical singularities. Tokyo J. Math., 12(2):375–385, 1989.
[12]
T. Fujita. On singular del Pezzo varieties. In Algebraic geometry (L’Aquila, 1988) Lecture Notes in Math.1417, 117–128. Springer, Berlin, 1990.
[13]
C. Casagrande, P. Jahnke, and I. Radloff. Int. J. Math., 19(2):173–191, 2008.
[14]
P. Jahnke and Th. Peternell. Adv. Geom., 8(3):387–411, 2008.
[15]
Yu. I. Manin. Cubic forms. Algebra, geometry, arithmetic. North-Holland Publishing Co., Amsterdam, II edition, 1986.
[16]
H. Finkelnberg and J. Werner. Small resolutions of nodal cubic threefolds. Indag. Math., 51(2):185–198.
[17]
Y. Prokhorov. p-elementary subgroups of the Cremona group of rank 3. In C. Faber, G. van der Geer, and E. Looijenga, editors, Classification of Algebraic Varieties, Schiermonnikoog, 2009, 327–338. European Math. Soc., 2010.
[18]
S. Cutkosky. Elementary contractions of Gorenstein threefolds. Math. Ann., 280(3):521–525, 1988.
[19]
Y. Namikawa. Smoothing Fano \(3\)-folds. J. Alg. Geom., 6(2):307–324, 1997.
[20]
S. Fukuda. A base point free theorem of Reid type. J. Math. Sci. Univ. Tokyo, 4(3):621–625, 1997.
[21]
Y. Hu and S. Keel. Mich. Math. J., 48:331–348, 2000.
[22]
M. Szurek and J. Wiśniewski. Fano bundles over \(\textbf{P}\sp 3\) and \(Q\sb 3\). Pacific J. Math., 141(1):197–208, 1990.
[23]
R. Hartshorne. Stable vector bundles of rank \(2\) on \(\textbf{P}^{3}\). Math. Ann., 238(3):229–280, 1978.
[24]
S. Mori and S. Mukai. Classification of Fano \(3\)-folds with \(B\sb{2}\geq 2\). Manuscripta Math., 36(2):147–162, 1981/82. Erratum: Manuscripta Math. 110 (2003), 407.
[25]
I. Dolgachev. Topics in classical algebraic geometry. manuscript.
[26]
J.-P. Serre. Complex semisimple Lie algebras. Springer-Verlag, New York, 1987.
[27]
J. G. Semple and L. Roth. Introduction to algebraic geometry. Oxford Science Publications. The Clarendon Press Oxford University Press, New York, 1985. Reprint of the 1949 original.
[28]
J. A. Todd. Proc. London Manth. Soc., 30:513–550, 1930. Ser. 2.
[29]
H. Hudson. Cremona transformations in plane and space., 1927.
[30]
R. W. H. T. Hudson. Kummer’s quartic surface., 1905.
[31]
V.V. Nikulin. Math. USSR, Izv., 9:261–275, 1975.
[32]
C. M. Jessop. Quartic surfaces with singular points., 1916.
[33]
I. Dolgachev and D. Ortland. Point sets in projective spaces and theta functions. Astérisque, (165):210 pp. (1989), 1988.
[34]
H. Clemens and Ph. Griffiths. The intermediate Jacobian of the cubic threefold. Ann. of Math. (2), 95:281–356, 1972.
[35]
M. Artin and D. Mumford. Some elementary examples of unirational varieties which are not rational. Proc. London Math. Soc. (3), 25:75–95, 1972.
[36]
A. Beauville. Variétés de Prym et jacobiennes intermédiaires. Ann. Sci. École Norm. Sup. (4), 10(3):309–391, 1977.
[37]
A. N. Tyurin. The intermediate Jacobian of three-dimensional varieties. In Current problems in mathematics12, 5–57. VINITI, Moscow, 1979.
[38]
M. Furushima and N. Nakayama. The family of lines on the Fano threefold \(V\sb 5\). Nagoya Math. J., 116:111–122, 1989.
[39]
I. Cheltsov and C. Shramov. Five embeddings of one simple group. ArXiv e-prints, 2009.
[40]
I. Cheltsov and C. Shramov. Three embeddings of the Klein simple group into the Cremona group of rank three. ArXiv e-prints, 2010.

  1. In papers [12], [13] authors considered del Pezzo varieties whose singularities are more general than terminal.↩︎

  2. There is a typographical error in [14]: the case \(c_2(\mathcal{F})=6\) occurs.↩︎

  3. Cases (b) and (c) overlap for \(X\) with \(\operatorname{d}(X)=5\).↩︎