Download as pdf or txt
Download as pdf or txt
You are on page 1of 21

Digital Object Identier (DOI) 10.

1007/s002090100374
Math. Z. 242, 687707 (2002)
#-Minimal models of uniruled 3-folds
Massimiliano Mella
Dipartimento di Matematica, Universit` a di Ferrara, 44100 Ferrara, Italia
(e-mail: mll@unife.it)
Received: 3 March 2000 / in nal form: 5 September 2001 /
Published online: 1 February 2002 c Springer-Verlag 2002
Mathematics Subject Classication (1991):14J30, 14N05
1. Introduction
The aim of minimal model theory is to choose, inside of a birational class of
varieties, a simple element. This program has been fullled in dimension
2 by the Italian school of the beginning of the century and a decade ago
in dimension 3. After its discovery, the theory of (1)-curves has been
used thoroughly to study algebraic surfaces. Unfortunately in the threefold
case the program and the simple output objects are not easily handled.
It is difcult to use them as a tool to understand the geometry of three
dimensional varieties. Here, after [Re], we rephrase the standard minimal
model program for uniruled varieties, using a polarizing divisor. As in [Re],
it will be called #-minimal model. We will be able, under strong assumption
on the variety studied, to govern the programand understand its output. Even
if quite restrictive, the assumption needed are very geometric in nature. This
allows to apply the #-program in various concrete situations. Indeed one
purpose of this note is to give a generalized and unied treatment of various
results on uniruled varieties, [Al] [CF] [Io].
We rst state the #-program and observe some of its natural properties.
This program is governed by a movable linear system1. The crucial obser-
vation is the following. If the generic element H 1 is a smooth surface
of negative Kodaira dimension, then the #-program is well understood in a
neighborhood of H. This allows to study both the steps of the program and
the nal output for specic families of 3-folds. For Fano varieties with bad
688 M. Mella
anticanonical sections we improve previous results of Alexeev, [Al], Theo-
rem4.1. We describe the birational type of 3-folds T containing a big system
of uniruled surfaces Corollary 5.5. Finally we are able to list the possible #-
models of small degree threefolds embedded in projective spaces. This is the
higher dimensional counterpart of the classical result for surfaces of reduced
degree 1, Theorem5.8. In the appendix we collect some observations and
examples about uniruled varieties embedded of small degree in projective
spaces. This part is based on Castelnuovo bound and its generalization due
to Harris, [Ha].
This note is a revised version of a preprint that circulated in 1996, inspired
by a long stay at Warwick University and by suggestions of M. Reid. I also
beneted a lot from many conversations with A. Corti and I would like to
thank V. Alexeev for pointing me out the papers of CampanaFlenner, [CF],
and Sano, [Sa]. The referees comment on the rst version of this paper
made it clearer and, I hope, easier to read. I would like to thank both Trento
University for supporting me along the rst part of this research and Centro
Nazionale delle Ricerche (grant n. 203.01.66).
2. Notations and preliminaries
All the varieties are dened over C and our notations are compatible with
[KMM]. Let X be a variety we will denote with NE(X) N
1
(X) the
closure of the cone of effective cycles inside the vector space of 1-cycles
modulo numerical equivalence. By means of the intersection product to any
Q-Cartier divisor Dwe associate the hyperplane D

= [Z] : DZ = 0
N
1
(X). We are interested in the part of the cone where K
X
is positive.
If X is terminal this part is locally polyhedral and the rays delimiting it are
called extremal rays.
A surjective morphism f : X Y with connected bers between nor-
mal varieties is calleda Mori ber spaceif K
X
is f-ample, rkPic(X/Y )=
1 and dimX > dimY . Given a projective morphism f : X Y and
A, B Div(X) Qthen A is f-numerically equivalent to B (A
f
B)
if A C = B C for any curve contracted by f. A is f-linearly equivalent
to B ( A
f
B ) if AB f

M, for some line bundle M Pic(Y ), we


will suppress the subscript when no confusion is likely to arise.
2.1 Avariety X is called uniruled if it admits a generically nite dominant
rational map p : Y P
1
>
X. By means of MiyaokaMori characteriza-
tion of uniruled varieties, [MM], the minimal model program and Miyaoka
characterization of minimal 3-folds, [Mi], a threefold is uniruled if and only
if its Kodaira dimension is negative. A uniruled 3-fold is always birational
to a Mori ber space.
# -Minimal models of uniruled 3-folds 689
Let 1be a linear system of Weil divisors, not necessarily complete, on a
variety X. H 1a general member of this linear system. We will say that
1 is movable if h
0
(X, nH) > 1 for some n > 0. Observe that the push
forward of a movable 1 is again movable. In other words the elements of
1 cannot be contracted by a birational projective morphism.
2.2 The following is a particular case of [BS, Prop. 1.4]. Let X be a terminal
3-fold and f : X W a Mori ber space with generic ber F P
r
.
Assume there exists anf-ample line bundle H Pic(X) withH
|F
O(1).
Then X and W are smooth and (X, H) = (P(c), O(1)), with c =

H a
rk(r + 1) vector bundle on W.
2.3 We want here to recall some basic properties of terminal Q-factorial 3-
folds. We will use them throughout the paper without explicit mentioning,
as a reference see for instance [Me]. Let X be a Q-factorial terminal 3-
fold and x Sing(X) a singular point. Then X has isolated singularities
and any irreducible surface S trough x is singular at x. Moreover if we let
i(X) = inft N : tK
X
Pic(X) then for any integral divisor A X
is i(X)A Pic(X).
3. #-Minimal models
Let T be a terminal Q-factorial uniruled 3-fold. As observed in 2.1, there
exists a birational modication : T
>
T
1
>
. . .
>
T
k
to a Mori space
T
k
. is obtainedas a chainof birational modications associatedtoextremal
rays on T, T
1
, . We are looking for a way to choose these extremal rays
in a natural way. To do this we will borrow an idea of Reid, [Re, (2.3)].
Fix a movable linear system1. We will choose an extremal ray, say R
+
[z],
having the smallest value 1 z.
Denition 3.1 Let T be a terminal Q-factorial uniruled 3-fold and 1 a
movable linear system with generic element H 1 on T. Assume that H
is nef, by abuse of language we will say that 1 is nef, then
=
H
= (T, 1)=: sup mQ[H+mK
T
is an effective Q-divisor0,
is the threshold of the pair (T, 1), see [Re, (2.1)].
Theorem 3.2 There exists a pair (T
#
, 1
#
), with generic element H
#

1
#
such that:
i) there is a birational map : T
>
T
#
with

1 = 1
#
;
ii) : T
#
W is a Mori space;
iii) (T, 1)K
T
# +H
#

O
T
#.
690 M. Mella
Denition 3.3 Accordingto[Re], (T
#
,1
#
) will be calleda #-Minimal Model
of the pair (T, 1).
Proof of the Theorem. We will proceed in an inductive way. Let (T
0
, 1
0
) =
(T, 1). H
0
is nef by hypothesis and T
0
is uniruled. Therefore to (T
0
, H
0
)
are naturally associated:
the nef value t
0
= supm Q[mK
T
0
+H
0
is nef ,
a rational map
0
: T
0
>
T
1
, which is either an extremal contraction,
or a ip, of an extremal ray in the face spanned by t
0
K
T
0
+1
0
,
a movable linear system 1
1
:=
0
1
0
on T
1
.
One inductively denes
i
: T
i
>
T
i+1
and (T
i+1
, 1
i+1
) as follows. First
note that t
i1
K
T
i
+H
i
is nef. Let = supd Q[dK
T
i
+(t
i1
K
T
i
+H
i
)
is nef and dene t
i
:= + t
i1
. The second step is to prove that there
exists an extremal ray [C
i
] NE(T
i
) in the face spanned by t
i
K
T
i
+H
i
. By
construction t
i
K
T
i
+H
i
= K
T
i
+(t
i1
K
T
i
+H
i
). Furthermore t
i1
K
T
i
+
H
i
is nef and (t
i
+)K
T
i
+H
i
is not nef. Then there exists a curve C
i
, with
K
T
i
C
i
< 0, in the face spanned by t
i
K
T
i
+H
i
.
Let us dene
i
: T
i
>
T
i+1
the birational modication associated to
the extremal ray [C
i
] NE(T
i
) and 1
i+1
:=
i
1
i
.
The inductive process is therefore composed by divisorial contractions
and ips. Since T
0
is uniruled after nitely many steps we reach a Mori ber
space : T
k
W.
Claim 3.4. (T
k
, 1
k
) satises iii).
Proof of the claim. Let
i
: T
i
>
T
i+1
a birational modication of the
inductive procedure. By Kleimans criteria the cone of nef divisors is the
closure of the ample cone. The cone of nef divisors is therefore contained in
the closure of the cone of effective divisors. So that we always have t
i

H
i
.
Let us observe that the threshold is preserved after any single birational
modication. On one hand
H
i+1

H
i
because the push forward of an
effective divisor is effective. On the other hand for any
i

i
K
T
i+1
= K
T
i
E,
for some 0 (if
i
is a ip then

i
is the trace morphism and = 0).
That is to say

i
(rK
T
i+1
+H
i+1
) = rK
T
i
+H
i
(r t
i
)E,
thus
H
i

H
i+1
. Let =
H
i
and :=
k
: T
k
W the Mori space
ending the process. Since is a bration, then (t
k
+)K
T
k
+H
k
cannot be
effective, therefore t
k
= . .
Let (T
#
, 1
#
) = (T
k
, 1
k
) to conclude. .
# -Minimal models of uniruled 3-folds 691
Remark 3.5 Along the proof of Claim 3.4 we proved that the threshold is
preserved by #-MMP modications. Note that 1
#
is relatively nef. Further-
more if the rational map dened by [mH[ is birational then 1
#
is relatively
ample.
In general the #-Minimal Model is not uniquely determined. This prob-
lem arise when two extremal rays, spanned by the same nef divisor have
not disjoint exceptional loci. For example let T = E F
1
and 1 =

A,
where E is a smooth curve of genus g > 0, A Pic(E) a movable di-
visor and : T E the natural projection. Then there are two extremal
rays spanned by H itself, one of divisorial type and the other of ber type.
The order in which the rays are contracted determines the #-minimal model
(either a P
1
-bundle or a P
2
-bundle).
As mentioned in the introduction we are not able to handle the whole #-
program. We will restrict ourselves to study the following special situation.
Let (T, 1) a pair with
H
< 1. If there is a smooth surface S 1 it is
possible to describe in detail the #-programin a neighborhood of the surface
S, see also [CF, 2].
Proposition 3.6 Let
i
: T
i
>
T
i+1
a birational modication in the #-
program relative to (T, 1), with
H
< 1. Assume that S 1
i
is a smooth
surface. Then
i
(S) = S is a smooth surface and
i|S
: S S is either
an isomorphism or the contraction of a disjoint union of (-1)-curves.
Proof. S is smooth and T
i
is terminal Q-factorial therefore SSing(T
i
) =
. In particular H
i
is a Cartier divisor.
Claim 3.7. If
i
is a ip then S is disjoint from the ipping curves
Proof of the claim. Let C be a curve ipped by
i
then K
T
i
C > 1,
[Mo2], and C is not on the smooth locus of T
i
. So that C , S and S C
is an integer. By denition (K
T
i
+H
i
) C 0 and by hypothesis < 1.
Therefore S C = 0 and the claim is proved. .
Case 3.8 (
i
contracts a divisor E onto a curve) The generic ber F of
i
is out of Sing(T
i
) and K
T
i
F = 1. Since (K
T
i
+ H
i
) F 0, then
H
i
F = 0 and S E is the disjoint union of (-1)-curves.
Case 3.9 (
i
is a divisorial contraction to a point) We can assume that
H
i
is
i
ample, let E the exceptional divisor and B = B
i
= S E.
Moreover H
i|E
ample and therefore let E the exceptional divisor and B =
B
i
= S E. Moreover H
i|E
ample and therefore B is connected. Let

S
:=
i|S
: S S, then by adjunction formula K
S
B
i
< 0. Furthermore

S
is birational therefore B is a (-1)-curve. Then E is smooth along B
692 M. Mella
and E B = 1. By construction K
T
B > 1, therefore K
E
B =
K
T
B E B 3.
Let :

E E a relatively minimal resolution of E and L =

(E
|E
).
Then L

B = 1. Let [

B]

i
b
i
[C
i
] for some [C
i
] NE(

E). Then
3 K

B =

i
b
i
K

E
C
i
and
1 =

i
b
i
L C
i
.
L is nef and K

E
is -nef. Moreover L C = 0 if and only if (C) = pt.
Therefore there is one i such that K

E
C
i
3, and

E P
2
, see for
instance [CKM]. Furthermore L is ample and K
E
B = K

E
B = 3.
That is K
T
i
B = 2 and (H
i|E
)
2
= H
i
B = 1. By relative spannedness,
[AW], H
i|E
is spanned and thus h
0
(E, H
i|E
) 3. This is enough to prove
that (E, H
i|E
) = 0 and (E, E
|E
) (P
2
, O(1)). .
Using the Proposition 3.6 we can control #-minimal model and its output.
Corollary 3.10 Let T be a terminal Q-factorial uniruled 3-fold and 1 a
movable and nef linear system. Set (T
#
, 1
#
) a #-minimal model of (T, 1).
Assume that
H
< 1 and 1 is base point free. Then H
#
Pic(T
#
), 1
#
has at most base points and H
#
is smooth.
Proof. By Bertini TheoremH is smooth therefore we can apply Proposition
3.6 in an inductive way up to reach a model (T
#
, 1
#
). .
We will need the following relative version of Corollary 3.10.
Denition 3.11 Let T be a 3-fold and 1 a movable linear system, with
dim1 1. Assume that H = M+F, where /is a movable linear system
without xed component and F is the xed component. Apair (T
1
, 1
1
) will
be called a log minimal resolution of the pair (T, 1) if there is a morphism
: T
1
T, with the following properties:
T
1
is terminal Q-factorial

1

M = H
1
, where H
1
is a Cartier divisor, dimBsl(1
1
) 0
a general element H
1
1
1
is a minimal resolution of a general element
M /.
Corollary 3.12 For any pair (T, 1) of an irreducible Q-factorial 3-fold T
and a movable linear system 1, with dim1 1, there exists a log minimal
resolution.
# -Minimal models of uniruled 3-folds 693
Remark 3.13 If T is an irreducible uniruled 3-fold and 1 a movable linear
system of Weil divisors, we will call #-Minimal Model of the pair (T, 1), a
#-Minimal Model of a log minimal resolution. Note that this is well dened
only up to birational equivalence.
Using Corollary 3.12 we can describe a pair (T, 1), with dim1 1, in
terms of a birational Mori ber space. A #-Minimal Model of a log minimal
resolution. It is useful to have a way to ensure that the resulting model is
effectively different from the starting one. The following Lemma is in this
direction and will be used in the next section to study Q-Fano 3-folds.
Lemma 3.14 Let (T
0
, 1
0
) be a pair consisting of a uniruled 3-fold and a
movable linear systemwith dim1
0
1. Let : (T
1
, 1
1
) (T
0
, 1
0
) a log
minimal resolution and (T
2
, 1
2
) a #-minimal model. Let H
i
1
i
generic
elements. Assume that H
0
is Cartier and K
H
0
, K
H
2
are ample. Then
K
2
H
0
K
2
H
2
. Furthermore if H
0
is normal and singular then K
2
H
0
< K
2
H
2
.
Proof. K
H
2
is ample. Then
H
1
< 1 and by construction there are two
well dened morphisms : H
1
H
2
and : H
1
H
0
. Moreover is
induced by a log minimal resolution : T
1
T
0
. If we dene the a
i
s by
K
H
1
= (K
T
1
+H
1
)
|H
1
=

(K
T
0
+H
0
)

a
i
E
i

|H
1
=

K
H
0

a
i
E
i
;
then a
i
0.
By Corollary 3.10 is a morphism between two smooth surfaces, thus
K
H
1
=

K
H
2
+

b
i
E
i
,
with b
i
0, where E
i
are either or exceptional.
This yields to
K
2
H
0
= K
2
H
2
+ 2

K
H
2

(a
i
+b
i
)E
i

(a
i
+b
i
)E
i

(a
i
+b
i
)E
i

.
By construction

(a
i
+b
i
)E
i

(a
j
+b
j
)E
j
= (

K
H
0

K
H
2
) (a
j
+b
j
)E
j
If

E
j
= 0 then (

(a
i
+ b
i
)E
i
) (a
j
+ b
j
)E
j
= (

(a
i
+ b
i
)E
i
)

K
H
0
0; while if

E
j
= 0 then (

(a
i
+ b
i
)E
i
) (a
j
+ b
j
)E
j
=

K
H
2
(a
j
+b
j
)E
j
. So in any case
K
2
H
0
= K
2
H
2
+

(a
i
+b
i
)E
i

K
H
0
+

(a
i
+b
i
)E
i

K
H
2
,
694 M. Mella
where in the right hand side the last two terms are non positive.
To prove the last assertion, let us start with a simple but useful observa-
tion. There are not birational contractions, that is morphisms with connected
bers, from a del Pezzo surface to a normal singular surface. In particu-
lar H
1
is not a del Pezzo surface. That is there exists the class of a cycle
[C
1
] NE(H
1
) such that K
H
1
C
1
0. On the other hand H
2
is Fano
thus is not an isomorphism. There exists at least a (-1)-curve, say E
0
,
contracted by . But is a log minimal resolution therefore

(E
0
) ,= 0.
Thus (

(a
i
+b
i
)E
i
)

K
H
0
< 0, and K
2
H
2
> K
2
H
0
. .
Let us derive another technical result, which will be useful in the ap-
pendix.
Lemma 3.15 Let
i
: T
i
>
T
i+1
a step of the #-programof a pair (X, 1).
Assume that H
i
is irreducible and let B T
i
any curve not contained in
the exceptional locus of
i
. Then H
i
B H
i+1

i
(B).
Proof. If
i
is a morphism projection formula
H
i+1

i
B =

i
H
i+1
B = (H
i
+E) B,
allows to conclude.
Assume now that
i
is the ip of the curve C. Let us consider the fol-
lowing diagram
Z
p
~~}
}
}
}
}
}
}
}
q
""
D
D
D
D
D
D
D
D
T
i
f

A
A
A
A
A
A
A
A

i
// _ _ _ _ _ _ _
T
i+1
g
||z
z
z
z
z
z
z
z
W
where f and g are the contractions of [C] and [C
+
], respectively, and p, q
is a resolution of
i
. Since H
i
is f-nef, there exists an M Pic(W) such
that m(H
i
+f

M) is spanned for m 0. Let H


Z
:= p

(m(H
i
+f

M))
then H
Z
= q
1

(m(H
i+1
+ g

M)). Furthermore q

(m(H
i+1
+ g

M)) =
H
Z
+

a
i
E
i
, with a
i
0. Therefore again by projection formula H
i
B
H
i+1

i
B. .
4. General elephants of Q-Fano 3-folds
In this section we will apply the #-program to study Q-Fano 3-folds with
bad anticanonical class. Let T be a Q-Fano 3-fold and assume that the
general element of [K
T
[, has worse than Du Val singularities. According to
# -Minimal models of uniruled 3-folds 695
the general elephant conjecture varieties of this kind shouldnt exist. Alexeev
proved that such a T, if one further assume that dim
|K
T
|
(T) = 3, is
birational to a Q-Fano 3-fold on which a general anticanonical element has
only Du Val singularities, [Al, Th 4.3]. Furthermore he proved that any
Q-Fano 3-fold admits a Gorenstein model, at the expense of introducing
canonical singularities, [Al, Th 4.8]. As observed by Corti, [Co2, 1.19], one
should expect to have a Gorenstein terminal model for a Q-Fano 3-fold with
big anticanonical system. Using #-Minimal Model Program we will show
that a Q-Fano 3-fold T with big anticanonical system and whose general
anticanonical element has worse than Du Val singularities, is birational to
a smooth Fano 3-fold. In such a way we meet Cortis expectations for this
particular class, conjecturally empty, of Q-Fano 3-folds.
Theorem 4.1 Let T be a Q-Fano such that dim
|K
T
|
(T) = 3 and the
general element in [ K
T
[ has worse than Du Val singularities. Then T is
birational to a smooth Fano 3-fold T
#
of Fano index 2.
Proof. Let 1
0
= [ K
T
[ and : (T
1
, 1
1
) (T, 1
0
) a log minimal
resolution of (T, 1
0
). A general element in 1
0
has worse than Du Val
singularities. Then the canonical class of H
1
, K
H
1
= K + H
1
, is negative
on innitely many curves dense in T
1
. In particular
H
1
< 1.
Let : T
1
>
T
2
the result of the #-program applied to (T
1
, 1
1
).
: T
2
W the ber type contraction associated to an extremal ray, say
[Z] NE(T
2
). By Corollary 3.10, 1
2
Pic(T
2
) and a general element
in 1
2
is smooth. Furthermore since dim
H
0
(T) = 3 then 1
2
is relatively
ample and
K
T
2
Z =
1

H Z < 1.
There are the following possibilities:
dimW = 1. The general ber F of is either P
2
or Q
2
and T
2
is
birational to W P
2
, see for instance the appendix in [Co1]. T has
rational singularities and h
1
(T, O
T
) = 0 then W P
1
. In particular T
2
is rational.
dimW = 2. : T
2
W is a conic bundle with a section. So that T
is birational to P
1
W. Again h
1
(T, O
T
) = 0, and via Castelnuovo
rationality criteria, W and henceforth T are rational.
dimW = 0. Then (T
2
) = 1 and 1
2
is ample. The general element
H
2
is a smooth del Pezzo surface and 1
2
has at most base points. By
assumptions dim(1
2
) h
0
(T, K
X
) 4 so that h
0
(H
2
, H
2|H
2
) 3.
By general properties of del Pezzo surfaces [H
2
[ is spanned by global
sections.
If T
2
is smooth or dimW > 0, we have nished. Assume T
2
is a singular
Q-Fano. Let x Sing(T
2
) and /
2
= [H
2
J
x
[. [H
2
[ is ample and spanned
696 M. Mella
therefore /
2
has only base points. In particular the general element is a
singular normal irreducible surface. Let (T
3
, /
3
) a log minimal resolution
of (T
2
, /
2
), and (T
4
, /
4
) a #-model. With
4
: T
4
W
2
the Mori ber
space associated. Observe that also in this case
L
3
< 1.
We have now to analyze the morphism
4
as before. Assume that /
4
is

4
-trivial. Since dim(/
2
) 2 and the generic element is irreducible then
/
2
is not composite with a pencil and dimW
2
= 2. Let
2
: T
2
>
V
2
the map dened by /
2
with dimV
2
= 2. Let =
|H
2
|
: T
2
P
n
, the
morphism dened by sections of [H
2
[ and l a curve contracted by
2
. Any
hyperplane passing trough (x) and any other point of (l) must contain
(l). That is (l) is a line trough (x) P
n
. Let us come back to the
morphism
4
. The generic ber f
0
of
4
, is contained in L
4
. The #-program
is an isomorphismin a neighborhood of f
0
. So that the strict transformof H
2
is a birational section and we conclude as above that the 3-fold is rational.
Assume that /
4
is
4
-ample. If dimW
2
> 0 we conclude that T
4
is
rational. If dimW
2
= 0 then /
4
is ample and a generic element L
4
[/
4
[ is
a smooth del Pezzo surface. Furthermore by Lemma 3.14 K
2
L
4
> K
2
H
2
1.
In particular [/
4
[ is spanned and since it is also ample then h
0
(T
4
, L
4
) 4.
We can therefore iterate our argument and, by Lemma 3.14, it must stop
after at most 7 steps, since K
2
L
k
9.
5. 3-folds with a big uniruled system
Denition 5.1 Let T be a terminal Q-factorial 3-fold and 1 a movable
linear system. We will say that (T, 1) is a pair with a big uniruled system
if H 1 is nef and big and H is a smooth surface of negative Kodaira
dimension.
Lemma 5.2 Let (T, 1) be a pair with a big uniruled system. Then T is
uniruled and (T, 1) < 1.
Proof. Let f H be a generic rational curve such that K
H
f < 0. H is
nef and big therefore K
T
f < 0 and by [MM] T is uniruled. For m 0
let m(K
T
+H) = H+B, with f , B. H is nef and m(K
T
+H) f < 0
then either B is not effective or < 0. Thus in any case
H
< 1. .
Our rst aim is to list the #-Minimal Models of pairs with a big uniruled
system.
Theorem 5.3 Let (T, 1) be a pair with a big uniruled system. Then
(T
#
,1
#
) is one of the following:
i) a Q-Fano 3-fold of index 1/ > 1, with K
T
# 1/H
#
and
|H
#
|
birational, the complete classication is given in [CF] and [Sa]:
# -Minimal models of uniruled 3-folds 697
(P(1, 1, 2, 3), O(6))
(X
6
P(1, 1, 2, 3, a), X
6
x
4
= 0), with 3 a 5
(X
6
P(1, 1, 2, 2, 3), X
6
x
3
= 0)
(X
6
P(1, 1, 1, 2, 3), X
6
x
0
= 0)
(P(1, 1, 1, 2), O(4))
(X
4
P(1, 1, 1, 1, 2), X
4
x
0
= 0)
(X
4
P(1, 1, 1, 2, a), X
4
x
4
= 0), with 2 a 3
(P
3
, O(a)), with a 3, (Q
3
, O(b)), with b 2
(X
3
P(1, 1, 1, 1, 2), X
3
x
4
= 0), (X
3
P
4
, O(1))
(X
2,2
P
5
, O(1))
a linear section of the Grassmann variety parametrising lines in
P
4
, embedded in P
9
by Plucker coordinates
(P(1, 1, 1, 2), O(2)), the cone over the Veronese surface
ii) a bundle over a smooth curve with generic ber (F, H
#
|F
) (P
2
, O(2))
and with at most nitely many bers (G, H
#
|G
) (S
4
, O(1)), where S
4
is the cone over the normal quartic curve and the vertex sits over an
hyper-quotient singularity of type 1/2(1, 1, 1) with f = xy z
2
+t
k
,
for k 1, [YPG],
iii) a quadric bundle with at most cA
1
singularities of type f = x
2
+y
2
+
z
2
+t
k
, for k 2, and H
#
|F
O(1),
iv) (P(E), O(1)) where E is a rk 3 vector bundle over a smooth curve,
v) (P(E), O(1)) where E is a rk 2 vector bundle over a surface of negative
Kodaira dimension.
Before giving the proof let us make a few comments about the output.
Remark 5.4 The singular varieties at the point ii) are canonically birational,
to a smooth projective bundle. In particular there exist singular terminal Mori
ber spaces with generic ber (P
2
, O(2)) and containing a smooth ruled
surface. This corrects an error in [Me, Prop 3.7] and [CF, Prop 3.4], where
such varieties have not been detected. To have an example it is enough to take
a smooth projective bundle over a curve, say X = P
2
P
1
, blowup a conic
C which sits in a ber G and contract the strict transform of G. In this way
we produce a Veronese cone singularity. The smooth surface is a generic
element of the strict transform of divisors in [O
X
(2, 1) J
C
[. Going on
blowing up hyperplane sections and contracting down we get hyper-quotient
index two singularities. Indeed along the proof we will prove that reversing
this process we obtain a canonical desingularization of the singular points
that can appear in such a situation. The same is true also for singular quadric
bundles at point iii). The varieties at point v) are birational to a projective
bundle over either P
2
or a ruled surface, but not in a canonical way. Indeed
this is achieved following the #-program on the base surface by means of
links on the 3-fold T
#
.
698 M. Mella
Theorem 5.3 and Remark 5.4 allow to extend the result of [CF] to the
following class of 3-folds.
Corollary 5.5 Let T be an irreducible threefold and 1a linear systemwith
generic element H. Assume that one of the following is satised:
(T, 1) is a pair with a big uniruled system
dim
H
(T) = 3 and the Kodaira dimension of a resolution of H is
negative.
Then T is birational to one of the following:
H P
1
,
a terminal sestic in either P(1, 1, 1, 2, 3) or P(1, 1, 2, 2, 3),
a terminal quartic in P(1, 1, 1, 1, 2),
a terminal cubic in P
4
.
Proof of the Corollary. Let (T
1
, 1
1
) a log resolution of (T, 1). Under both
assumptions (T
1
1
1
) is a pair with a big uniruled system. Then a #-model
of (T
1
, 1
1
) is in the list of Theorem 5.3. We can now argue exactly as in
[CF, Rem. 1.6]. .
Proof of the Theorem. By Sect. 3 we have a Mori ber space : T
#
W
with K
T
# +H
#

0 and H
#
is smooth.
Case 5.6 (Fano case) Assume that K
T
# H
#
and rkPic(T
#
) = 1, then
H
#
is ample and generated by global sections outside a nite set of smooth
points. H
#
is smooth and K
H
# = (K
T
# + H
#
)
|H
# = (1 1/)H
#
|H
#
therefore H
#
is a del Pezzo surface. The classication of these 3-folds is
known [CF] and [Sa].
Case 5.7 (Fiber type case) Let F be a generic smooth ber. Then (K
T
# +
H
#
)
|F
O
F
thus 1/H
#
|F
= K
F
.
If dimW = 1 then either F P
2
and = 1/3, 2/3 or F Q
2
and = 1/2. Furthermore all bers are irreducible and reduced and the
morphism is at. Furthermore the delta genus of any ber is dened and
it is semi-continuous. This simple observation allows to give a biregular
description of these varieties in the following way.
5.7.1 (F P
2
and = 1/3) Then T
#
P(

H
#
) by (2.2).
5.7.2 (F P
2
and = 2/3) We will prove that the only possible bers
are P
2
or S
4
. The latter with the vertex sitting on a terminal hyper-quotient
singularity of index 2 and type 1/2(1, 1, 1) f = xy z
2
+ t
k
, with
k 1. Incidentally observe that (K
T
# +H
#
)
|F
O
F
(1). Thus T
#
has
# -Minimal models of uniruled 3-folds 699
a birational section and it is birational to a projective bundle. By (2.2) all the
bers on the Gorenstein locus are P
2
. Assume that there is a singular non
Gorenstein point x and a singular ber G x. By Fujita classication and
atness, the only possible singular ber is the cone over the normal quartic
curve. In particular [H
#
[ is relatively very ample. Using this remark we will
be able to describe a birational map fromT
#
to a projective bundle over W
factored by elementary links in the Sarkisov category, [Co1].
Let : T
#
P
N
an embedding given by the sections of / = [H
#
+

M[, for some very ample M on W. Let : Y T


#
the blowup of (x)
with exceptional divisor E =

e
i
E
i
. Let L
1
and L
2
two generic element
of [/ J
x
[. Then

L
i
= L
iY
+E and

G = G
Y
+

i
E
i
,
for integers
i
e
i
. Since / is very ample then L
1Y
L
2Y
G
Y
= 0 and
L
1Y
L
2Y
E
i
is the degree of E
i
as subvariety of P
N1
. Therefore by
projection formula
(5.1) 4 = L
1
L
2
G =

i
deg(E
i
).
G
Y
F
4
and Gis embedded in P
5
by . Let E
0
an irreducible component
of the exceptional divisor that contains the rational normal curve of degree
4. Then degE
0
> 2,
i
= e
i
= 1 and G
Y |E
L
Y |E
is ample. So that E
is irreducible and reduced of degree 4. It is now immediate to observe that
E is either the Veronese surface or the cone over a rational normal quartic
curve. Let f be a ber of the ruling on G
Y
then G
Y
f = 1. Moreover by
adjunction formula
K
Y
f = 1.
This is enough to prove that G
Y
Y is contractible to a 3-fold T
1
. Moreover
T
1
is smooth along the exceptional locus of the contraction. Iterating this
process we get a resolution of the singularities of T
#
. Keep in mind that
(T
#
) P
N
and we are making smooth blow ups of points in the ambient
variety. After nitely many steps we have the following picture
Y

~~}
}
}
}
}
}
}
}

@
@
@
@
@
@
@
@
. . .
~~}
}
}
}
}
}
}
}
}
. . .

A
A
A
A
A
A
A
A
A Y

~~}
}
}
}
}
}
}
}

A
A
A
A
A
A
A
A
T
#
T
1
T
k T

in particular T

is a smooth projective bundle over W and T


k
has a quotient
singularity of type 1/2(1, 1, 1).
We have to describe an analytic neighborhood of the point x T
#
.
By results of Pinkham, [Pi] and Koll` arShepherd-Barron, [KSB, pag 313-
316], the versal deformation of the singularity in S
4
the versal deformation
700 M. Mella
of the singularity in S
4
has only one smooth component of dimension 1,
with generic ber a Veronese surface. Locally this family is the quotient of
C
4
/(xy z
2
+ t) by a cyclic group of order 2 acting on (x, y, z, t) with
weights (1, 1, 1, 0). The local equation of T
#
can be obtained by this one
by a base change t t
k
.
5.7.3 (F Q
2
and = 1/2) Then by Fujita list the only possible singu-
lar ber G is a quadric cone. In particular [H
#
[ is relatively very ample.
Repeating the same argument of the quartic cone we exhibit a canonical
birational map to a smooth quadric bundle over W, factored by elemen-
tary links in the Sarkisov Category. Then argue by [KM, Prop 4.10.1] and
[Cu]. To derive that the possible singularities of T
#
are cA
1
points of the
type k[x, y, z, t]/(x
2
+y
2
+z
2
+t
k
). Another approach is to use the versal
deformation of the cone singularity. That is given by x
2
+y
2
+z
2
+t.
If dimW = 2 the only possibility is F P
1
and = 1/2.
5.7.4 (F P
1
and = 1/2) . Then H
#
|F
= O(1) and by (2.2) W is
smooth and T
#
P(

H
#
). Let S in H
#
a smooth section then
|S
:
S W is a birational morphism, therefore W is either ruled or the blow
up of a ruled surface.
To have a better description of these conic bundles consider the #-
minimal model of (W, G = (
|S
)

(H
|S
)), [Re]. Let (W
#
, G
#
) the #-
Minimal Model of (W, G) and H the push forward of

H on W
#
. Then
W
#
is either P
2
or a ruled surface and (T
#
, H
#
) is birational to (P(H),
O(1))
T
#

// _ _ _
P(H)

W
//
W
#
.
It is possible to factor the birational map (T
#
, H
#
)
>
(P(H), O(1)) by
elementary links in the Sarkisov category. The links are described in the
following way. Let B W a (-1)-curve then D =

(B) F
a
, for
some a 0. If a = 0 then the link is simply the blow down of D. If
a > 0 then let C
0
the exceptional section of the rational scroll D. Then
N
C
0
/T
O(1) O(a). It is possible to antiip it (or op if a = 1)
to get the following
T

// _ _ _

T

//
T

W
//
W
#
.
where is the weighted blow up with weights (1, 1, a). .
# -Minimal models of uniruled 3-folds 701
There exists a natural geometric interpretation of the conditions imposed
in Theorem 5.3.
Theorem 5.8 Let T
d
P
n
be a degree d non degenerate 3-fold. Suppose
that d < 2n 4, then any #-Minimal Model (T
#
, 1
#
) of (T
d
, O(1)) is in
the list of Theorem 5.3.
Remark 5.9 This result is an answer to a question of Mumford, [Mu, 2.15
pg 66], about reduced degree varieties. The reduced degree of a variety
T
d
P
n
of dimension k and degree d is by denition
rd :=
d
n + 1 k
.
We could rephrase Theorem 5.8 saying that 3-folds of reduced degree < 2
have a #-minimal model in the list of Theorem 5.3. If we forget about the
polarization and consider only the 3-fold T
#
then the birational type of this
class of 3-folds is described in Corollary 5.5. Observe that the 3-folds listed
admit an embedding satisfying the numerical criteria.
Theorem 5.8 can be interpreted as the three dimensional generalization
of the following classical result for surfaces, [GH, Proposition at page 525].
Let S P
n
be a non degenerate irreducible surface of degree d n1
then S is either a rational scroll or the Veronese surface.
In the threefold case we loose the biregular character of the classication
and smoothness. Furthermore the varieties listed can have arbitrarily high
irregularity. Consider for example P
2
C, with C a curve of genus g 2
embedded by (O(1), /). Where / is a non special very ample divisor of
degree g+3. Even the irregularity of the base of conic bundles is unbounded.
As the followingexample shows. Let T = F
1
C, where F
1
is the blowupof
P
2
and C is a genus g curve. Consider the linear system1 = [(C
0
+2f, /)[,
with deg/ > 10g 4. Then
1
2
K
T
+1
is nef and denes a conic bundle structure of T onto P
1
C. It is a simple
calculation then to observe that the embedding dened by 1 satises the
degree assumption. This is not surprising indeed. The surface classication
is essentially based on the fact that the unique uniruled curve of degree d n
is the rational normal curve. While uniruled surfaces of degree d < 2n 2
have arbitrary irregularity.
Proof of the Theorem. Let : X T a resolution of singularities and
1 =

O(1). By Lemma A.2 in the appendix (K


X
+ H) H
2
< 0. So
that by adjunction formula, H 1is a smooth surface of negative Kodaira
dimension and (X, 1) is a pair with a big uniruled system. .
702 M. Mella
Consider the proof of Proposition 3.6. If the nef value of the pair (T
i
, 1
i
)
is positive the birational modication associated is the blow down of (P
2
,
O(1)). If we restrict to study ample linear systems 1 it is possible to
generalize all #-Minimal Model Theory to arbitrary dimensions.
Corollary 5.10 Let X
d
P
n
a non degenerate k-fold with k > 3 and only
Q-factorial terminal singularities. Assume that d < 2(n k) 2 then a
#-minimal model (X
#
, H
#
) of (X, O(1)), in adjunction theory language
(X
#
, H
#
) is the rst reduction, is one of the following:
i) a Q-Fano n-fold of Fano index 1/ > k 2, with K
T
# 1/H
#
and
|H
#
|
birational, the complete classication is given in [Fu] if X
#
is Gorenstein and in [CF] and [Sa] in the non-Gorenstein case.
ii) a projective bundle over a smooth curve with bers (F, H
#
|F
) (P
k1
,
O(1)), or a quadric bundle with at most cA
1
singularities, with H
#
|F

O(1);
iii) (P(E), O(1)) where E is a rk(k 1) ample vector bundle either on
P
2
or on a ruled surface.
Proof. Let us argue as in Theorem 5.8. By Lemma A.2 the threshold is<
1/(k 2). By adjunction theory on terminal varieties, [Me], after nitely
many blow downs of (P
k1
, O(1)), we nd a Mori space structure. If
X
#
is Fano and Gorenstein then rkPic(X
#
) = 1. Moreover the index is
an integer dimX1, so that Fujita, [Fu], classication applies. All other
cases are well described in [CF], [Sa] and [Me]. .
Remark 5.11 An equivalent statement was proved by Ionescu for smooth
varieties in [Io]. As in the threefold case let me stress that the irregularity is
unbounded. X
d
is terminal Q-factorial. Under the minimal model conjec-
ture, Corollary 5.10 furnishes the classication of #-models of irreducible
non degenerate k-folds embedded in P
n
of degree d < 2(n k) 2.
Appendix
In this appendix we sum up some observations about small degree varieties.
Let us consider T
k
d
P
n
, a non degenerate irreducible k-fold. It is quite
natural to predict that for small degree d, with respect to n, T must be
uniruled. Our rst aim is to nd the region of the (n, d) plane banned to non
uniruled varieties.
Example A.1 Let E be an elliptic curve and F
k+1
P
k
an hypersurface of
degree k +1. Consider T = EF
k+1
together with the natural projections
p
1
: T E and p
2
: T F
k+1
. Let O
T
(L
e
) = p

1
O
E
(e) p

2
O
F
k+1
(1),
then O
T
(L
e
) is very ample for any e 3; let
e
: T P
n
, the embedding
# -Minimal models of uniruled 3-folds 703
associated to [O
T
(L
e
)[, then n = h
0
(E, O
E
(e)) h
0
(F
4
, O
F
4
(1)) 1 =
(k + 1)e 1 while the degree d = O
T
(L
e
)
k
= k(k + 1)e. In this case
(T) = 0 and d = k(n + 1).
The main message hidden in this example is the following. It is not
possible to take arbitrarily high Veronese embeddings of a xed immersion
to construct varieties sitting in the region in which we are interested in. We
need to evaluate the negativity of the canonical class on hyperplane sections.
Lemma A.2 Let T
k
d
P
n
an irreducible non degenerate k-fold. Let a =
mint N[d < t(n k) + 2, and degenerate k-fold. Let a = mint
N[d < t(n k) +2, and assume that a k. Let : X T a resolution
of singularities and H
i
[

O(1)[, for i = 1, . . . , k1 generic hyperplane


sections. Let C =
k1
1
H
i
, then (K
X
+ (k a)H) C < 0.
Proof. Let S :=
k2
1
H
i
a smooth surface containing C. From genus for-
mula on the surface S
(A.1) g(C) = 1 +
1
2
[d(k 1 (k a)) + ((k a)H +K
X
) C].
We want to bound the genus from above. To do this we will will use Castel-
nuovo inequality, [GH, pag 251]. Let h = h
0
(C, H) = n k + 2, then the
genus of C is bounded by the following sum
g(C) (d h + 1) + (d 2h + 3) + (d 3h + 5) +. . . ,
where we sum only positive terms. Therefore in our hypothesis we have
either
g(C) (a 1)d
a(a 1)
2
(n k) a + 1,
or a(n k) (2a 1) < d < a(n k) + 2 and
g(C) ad
a(a + 1)
2
(n k) a.
So that combining with equation (A.1) we obtain the desired inequalities
(A.2) ((k a)H +K
X
)) C (a 1)(d a(n k) 2),
respectively
((k a)H +K
X
)) C (a + 1)(d a(n k) 2).
.
Theorem A.3 Let T
k
d
P
n
. Assume that d < k(n k) + 2, then T is
uniruled.
704 M. Mella
Proof. Let : X T a resolution of singularities. Fix H
i
[

O(1)[, for
i = 1, . . . , k 1, and C = H
i
. By MiyaokaMori criterion, [MM], it will
be enough to prove that K
X
C < 0. This is the content of Lemma A.2 for
a k. .
The following example settles completely the question. There exist k-
folds X of degree d with (X) = 0 and d = k(n k) + 2.
Example A.4 ([Ha]) Let us consider a (k+1)-fold V , scroll over a rational
normal, curve embedded of degree n k in P
n
and a generic element
T [(k +1)O(1) J
nk2
F
[, where F is a ber of the scroll. Then for the
generic V , [Ha], T is spanned, thus for the generic element K
T
O
T
and degT = k(n k) + 2.
This example is particularly interesting. It is the example of a non unir-
uled variety for which the reduced degree is less then the dimension. In the
curve and surface case this is impossible but starting from 3-folds on there
are plenty of non uniruled variety, even of general type, which have this
property.
Let us now come back to 3-folds. the next example gives a family of
uniruled 3-folds for which d = 3n 8 and n is arbitrarily large. Keep in
mind that by Lemma A.2 this family is just on the edge of the uniruled sector.
Example A.5 Let T
4
P
4
, Z = H
1
H
2
, with H
i
[O
T
(1)[, a smooth
hyperplane curve and M
a
= [O
T
(a) J
a1
Z
[. Let us consider : X =
Bl
Z
(T) T and H
a

O
T
(a) (a 1)E the strict transform of M
a
.
Then rkPic(X) = 2 and X admits a bration f : X P
1
in K3 surfaces
(the bers are the strict transform of surfaces in [O
T
(1) Z[). Therefore
the only extremal ray is the one of the blowing up. H
a
f

O
P
1(a 1) +

O
T
(1), thus it is very ample, let
a
: X Y
a
P
n
the morphism
dened by the sections of [H
a
[.
Let d = H
3
a
the degree of Y
a
P
n
then
Claim A.6 d = 3n 8.
Proof of the claim. By construction E
3
= 8 and E
2

O
T
(1) = 4.
The degree of Y
a
is
d = H
3
a
= 4a
3
+ 8(a 1)
3
12a(a 1)
2
= 12a 8.
To calculate n let H
a
= f

O(a) + E, tensoring with H


a
the structure
sequence of f

O(a) we get
0 O
X
(E) O
X
(H
a
) O
f

O(a)
(E) 0,
E K
X
is nef and big therefore
0 H
0
(X, E) H
0
(X, H
a
)
a
H
0
(F, O
F
(1)) 0.
Finally this yields n = h
0
(X, H
a
) 1 = ah
0
(F, O
F
(1)) = 4a. .
# -Minimal models of uniruled 3-folds 705
Let us further note that there is only one extremal ray on T
a
. Then the
#-minimal program of the pair (Y
a
, O(1)) consists just in the contraction
of E. The #-minimal model of (Y
a
, O(1)) is (T
4
, O(a)). In such a way we
have constructed varieties with Fano #-minimal model, embedded inside
our area with n arbitrarily large.
Remark A.7 It is important to stress the following. There are only nitely
many deformation types of Q-Fano 3-folds, [Ka2]. Hence only nitely many
complete linear systems of bounded degree on a Q-Fano 3-fold. The above
example shows that, choosing a non complete linear system, one can get
Q-Fano #-models with arbitrarily high codimension.
Denition A.8 ([Ko]) Let T be a uniruled variety and H an ample line
bundle. We say that T is uniruled of H-degree at most d if there is a covering
family of rational curves C

such that C

H d.
Let us consider a 3-fold T
d
P
n
, with d < 3(n 3) + 2. By Lemma
A.2 T is uniruled. Our aim is to bound its O(1)-degree of uniruling.
Example A.9 Let T = P
1
F
4
, where F
4
P
3
is a quartic surface.
Consider the line bundle L
a
of bidegree (a, 1) on the product T. Then
L
3
= 12a and h
0
(T, O
T
(L)) = 4(a + 1). L
a
is very ample for any a 1
and embeds T P
n
as a 3-fold of degree d = 3n 9. Furthermore the
L
a
-degree of uniruling is a.
Example A.9 shows that there is no hope to bound the uniruled degree
of 3-folds embedded in our area. On the other hand the behavior of uniruled
degree is not completely uncontrolled.
Theorem A.10 Let T
d
P
n
and assume that d < 3n 7. If d < 2n 4
then T
d
is uniruled of O(1)-degree at most 5. Assume that 2n 4 d <
3(n 3) + 2. Let
=
d
3(n 3) + 2
,
Then T is uniruled of O(1)-degree at most
3
(1 )
.
Proof. Let (T
#
, H
#
) a #-model of (T, O(1)) and the threshold. By a
result of Kawamata, [Ka1], a dense subset of T
#
can be covered by rational
curves B such that K
T
# B 6. Therefore H
#
B 6. Let B
T
the
strict transform of B on T. Then by Lemma 3.15 we get
H B
T
H
#
B 6.
706 M. Mella
It is therefore enough to get a bound on the threshold depending on d.
If d < 2n 4 we already know that < 1 thus T is covered by rational
curves of degree at most 5.
Assume that 2n 4 d < 3n 8. Then by equation (A.2) with
a = k = 3 in Lemma A.2
(A.3) K
T
H
2
2( 1)(3(n 3) + 2).
By denition of threshold if (rK
T
+ H) H
2
< 0 then r > , therefore
substituting Equation (A.3) we obtain the required upper bound for the
threshold

(3(n 3) + 2)
2(1 )(3(n 3) + 2)
,
which allows to conclude.
Remark A.11 The bound in Lemma A.10 is not sharp. In fact we are estimat-
ing the threshold on K
T
H
2
and we are using Kawamata result. Nonetheless
it is interesting to observe that the degree of the ruling is bounded by a func-
tion depending only on , that is on the ratio d/n. Similar results are true
for irreducible varieties of arbitrary dimension, under the Minimal Model
conjecture.
References
[Al] Alexeev,V: General elephants of Q-Fano 3-folds, Comp.Math 91, 91116 (1994)
[AW] Andreatta,M. Wisniewski,J: A note on non vanishing and its applications, Duke
Math. J. 72 (1993) 739755
[BS] Beltrametti,M. Sommese,A.J: On the adjunction theoretic classication of polar-
ized varieties, J. reine angew. Math. 427 (1992), 157192
[CF] Campana,F. Flenner, H: Projective threefolds containing a smooth rational surface
with ample normal bundle, J. reine angew. Math 440 (1993), 7798
[CKM] Clemens, H. Koll ar, J. Mori, S: Higher-dimensional complex geometry, Asterisque
166 (1988).
[Co1] Corti, A. Factoring birational maps of threefolds after Sarkisov, J. Alg. Geo. 4
(1995), 223254
[Co2] Corti, A. Del Pezzo surfaces over Dedekind schemes, Ann. of Math. 144 (1996),
641683
[Cu] Cutkowski, S: Elementary contractions of Gorenstein threefolds, Math. Ann. 280
(1988), 521525
[Fu] Fujita, T: Classication theories of polarized varieties, London Math. Soc. Lecture
Note Series 155 Cambridge University press (1990)
[GH] Grifths, P.: Harris, J. Priciples of algebraic geometry, John Wiley & sons (1978)
[Ha] Harris, J: A bound on the geometric genus of projective varieties, Ann. Scuola
Norm. Sup. Pisa Cl. Sci. (4) 8 (1981), 3568
[Io] Ionescu, P: On varieties whose degree is small with respect to codimension. Math.
Ann. 271 (1985), 339348
# -Minimal models of uniruled 3-folds 707
[Is] V.A. Iskovskikh: Fano 3-folds I, II, Math. USSR Izv. 11, 485527 (1977), 12
469506 (1978)
[Ka1] Kawamata, Y: On the length of an extremal rational curve, Inv. Math. 105, 609611
(1991)
[Ka2] Kawamata, Y: Boundedness of Q-Fano threefolds Proceedings of the International
Conference on Algebra, Part 3 (Novosibirsk, 1989), Contemp. Math. 131, Part 3,
A. M. S. 439445 (1992)
[KMM] Kawamata, Y. Matsuda, K. Matsuki, K.: Introduction to the Minimal Model Pro-
gram in Algebraic Geometry, Sendai, Adv. Studies in Pure Math. 10, Kinokuniya-
North-Holland, 283360 (1987)
[Ko] Koll ar, J: Rational Curves on Algebraic Varieties, Ergebnisse der Math. 32, (1996),
Springer
[KM] Koll ar, J. Mori, S: Classication of three-dimensional ips. J. of the A.M.S.
51992(), 533703
[KSB] Koll ar, J. Shepherd-Barron, N: Threefolds and deformations of surface singulari-
ties. Inv. Math. 91 299338 (1988)
[Me] Mella, M: AdjunctionTheoryonterminal varieties. ProceedingComplexAnalysis
and Geometry editors V. Ancona, E. Ballico, A. Silva: Pitman Research Notes in
Mathematics 366 15316 (1997)
[Mi] Miyaoka, Y: On the Kodaira dimension of minimal threefolds. Math. Ann. 281,
325332 (1988)
[MM] Miyaoka, Y., Mori, S: A numerical criterion for uniruledness. Ann. of Math. 124,
6589 (1986)
[Mo1] Mori, S: Threefold whose canonical bundle are not numerically effective. Ann.
Math. 124, 133176 (1982)
[Mo2] Mori, S: Flip Theorem and the existence of minimal models for 3-folds. Journal
AMS 1, 117253 (1988)
[Mu] Mumford, D: Stability of projective varieties. Ens. Math. 23, 39110 (1977)
[Pi] Pinkham, H: Deformation of algebraic varieties with G
m
-action. Asterisque 20
(1974)
[Re] Reid, M: Surfaces of small degree. Math. Ann. 275, 7180 (1986)
[Sa] Sano, T: Classication of Q-Fano d-folds of index greater than d 2. Nagoya
Math. J. 142 133143 (1996)
[U2] Koll ar et al: Flip and abundance for algebraic threefolds. Asterisque 211 1992
[YPG] Reid, M: Young Persons Guide to canonical singularities. Algebraic Geometry
Bowdoin Part I, Proc. of Symp. of Pure Math. 46, 345414 (1987)

You might also like