This action might not be possible to undo. Are you sure you want to continue?
Vol. 18 (2008) 1118 – 1144
1016443X/08/04111827
DOI 10.1007/s0003900806872
ONLINE FIRST: December 2008
c Birkh¨auser Verlag, Basel 2008
GAFA Geometric And Functional Analysis
LOG CANONICAL THRESHOLDS OF DEL PEZZO
SURFACES
Ivan Cheltsov
Dedicated to Yuri Manin on his seventieth birthday
Abstract. We study global log canonical thresholds of del Pezzo surfaces.
All varieties are assumed to be deﬁned over C.
1 Introduction.
The multiplicity of a nonzero polynomial φ ∈ C[z
1
, · · · , z
n
] at the origin
O ∈ C
n
is the nonnegative integer m such that φ ∈ m
m
\ m
m+1
, where m is
the maximal ideal of polynomials vanishing at the point O in C[z
1
, · · · , z
n
].
It can be deﬁned by derivatives, because the equality
m = min
m ∈ N ∪
¸
0
¸
∂
m
φ(z
1
, . . . , z
n
)
∂
m
1
z
1
∂
m
2
z
2
· · · ∂
mn
z
n
(O) = 0
.
holds. We have a similar invariant that is deﬁned by integrations. This
invariant is given by
c
0
(φ) = sup
c ∈Q
the function
1
φ
c
is locally L
2
near the point O∈C
n
¸
,
and c
0
(φ) is called the log canonical threshold of φ at the point O. The
number c
0
(φ) appears in many places. (The number c
0
(φ) is also called the
complex singularity exponent of φ (see [K]).) For instance, it is known that
c
0
(φ) is the same as the absolute value of the largest root of the Bernstein–
Sato polynomial of φ (see [K]).
Even though the log canonical threshold was known implicitly, it was
formally introduced in the paper [S] as follows. Let X be a variety with
Keywords and phrases: del Pezzo surface, log canonical threshold, αinvariant of
Tian, birational automorphisms
AMS Mathematics Subject Classiﬁcation: 14J45, 14J26, 14E07, 32Q20, 32Q10
Vol. 18, 2008 LOG CANONICAL THRESHOLDS OF DEL PEZZO SURFACES 1119
log terminal singularities, let Z ⊆ X be a closed subvariety, and let D be
an eﬀective QCartier Qdivisor on X. Then the number
lct
Z
(X, D) = sup
¸
λ ∈ Q  the log pair (X, λD) is log canonical along Z
¸
is said to be the log canonical threshold of D along Z. The number
lct
Z
(X, D) is known to be positive and rational. Moreover, if X = C
n
and D = (φ = 0), then the equality
lct
O
(X, D) = c
0
(φ)
holds (see [K]). For the case Z = X we use the notation lct(X, D) instead
of lct
X
(X, D). Then
lct(X, D) = inf
¸
lct
P
(X, D)  P ∈ X
¸
= sup
¸
λ ∈ Q  the log pair (X, λD) is log canonical
¸
.
Even though several methods have been invented in order to compute
log canonical thresholds, it is not easy to compute them in general. How
ever, the log canonical thresholds play a signiﬁcant role in the study of
birational geometry showing many interesting properties (see [K], [P]).
Thus far the log canonical threshold has a local character. In this paper
we wish to develop its global analogue for Fano varieties. We shall see it is
useful to consider the smallest of the log canonical thresholds of eﬀective
Qdivisors numerically equivalent to an anticanonical divisor.
Let X be a Fano variety with log terminal singularities, and G be a
ﬁnite subgroup in Aut(X).
definition 1.1. We deﬁne the global Ginvariant log canonical threshold
of X by the number
lct(X, G) = inf
¸
lct(X, D)
the eﬀective Qdivisor D
is Ginvariant and D ≡ −K
X
¸
.
We put lct(X) = lct(X, G) if the group G is trivial. Note that it follows
from Deﬁnition 1.1 that
lct(X, G) =
sup
λ ∈ Q
the log pair (X, λD) has log canonical singularities
for every Ginvariant eﬀective Qdivisor D ≡ −K
X
¸
0 .
Example 1.2. It follows from Proposition 16.9 in [Ket al.] that
lct(P(1, 1, n)) = 1/(2 +n) for n ∈ N.
For a given Fano variety, it is usually very hard to compute its global log
canonical threshold explicitly (see [C2]). For instance, the papers [H1] and
1120 I. CHELTSOV GAFA
[H2] show that the global log canonical threshold of a rational homogeneous
space of Picard rank 1 and Fano index r is 1/r.
Example 1.3. Let X be a smooth hypersurface in P
n
of degree n ≥ 3.
Then
lct
X
1 −1
n
due to [C1]. It is clear that the inequality lct(X) = 1 − 1/n holds if the
hypersurface X contains a cone of dimension n − 2. But the paper [Pu]
shows that lct(X) = 1 if X is general and n 6.
Global log canonical thresholds of Fano varieties play an important role
in geometry. (It follows from [CS, Append. A] that global log canonical
thresholds of Fano varieties are algebraic counterparts of αinvariants in
troduced in [T1].)
Example 1.4. Let X be a general wellformed quasismooth hypersurface
in P(1, a
1
, . . . , a
4
) of degree
¸
4
i=1
a
i
with terminal singularities such that
−K
3
X
1. Then lct(X) = 1 by [C2], which implies that
Bir(X ×· · · ×X
. .. .
m times
) =
m
¸
i=1
Bir(X), Aut(X ×· · · ×X
. .. .
m times
)
,
the variety X×· · · ×X is not rational and not birational to a conic bundle
(see [C2]).
One of the most interesting applications of global log canonical thresh
olds of Fano varieties is the following result proved in [DK] (see also [N],
[T1] and [CS]).
Theorem 1.5. Let X be a Fano variety with quotient singularities, and
let G be a ﬁnite subgroup on Aut(X) such that the inequality
lct(X, G) >
dim(X)
dim(X) + 1
holds. Then X has a Ginvariant orbifold K¨ahler–Einstein metric.
The following conjecture is inspired by [T3, Question 1].
Conjecture 1.6. For a given Fano variety X with log terminal singu
larities and ﬁnite subgroup G ⊂ Aut(X), the equality
lct(X, G) = lct(X, D)
holds for some Ginvariant eﬀective Qdivisor D on the variety X such that
D ≡ −K
X
.
The main purpose of this paper is to prove the following result.
Theorem 1.7. Let X be a smooth del Pezzo surface. Then
Vol. 18, 2008 LOG CANONICAL THRESHOLDS OF DEL PEZZO SURFACES 1121
lct
X
=
1 when K
2
X
= 1 and  −K
X
 has no cuspidal curves,
5/6 when K
2
X
= 1 and  −K
X
 has a cuspidal curve,
5/6 when K
2
X
= 2 and  −K
X
 has no tacnodal curves,
3/4 when K
2
X
= 2 and  −K
X
 has a tacnodal curve,
3/4 when X is a cubic surface in P
3
without Eckardt points,
2/3 when K
2
X
= 4 or X is a cubic surface in P
3
with an Eckardt point,
1/2 when X
∼
= P
1
×P
1
or K
2
X
∈ {5, 6},
1/3 in the remaining cases.
Taking the paper [P] and Theorem 1.7 into consideration, we see that
the assertion of Conjecture 1.6 holds for smooth del Pezzo surfaces with
trivial group action. Also, in this paper, we prove the following result.
Theorem 1.8. Let X be a del Pezzo surface with ordinary double points
such that K
2
X
= 1. Then
lct
X
=
1 when  −K
X
 does not have cuspidal curves,
3/4 when  −K
X
 has a cuspidal curve C such that
Sing(C) ⊆ Sing(X),
5/6 in the remaining cases.
We see that Theorems 1.5 and 1.8 imply the existence of an orbifold
K¨ahler–Einstein metric on every del Pezzo surface of degree 1 that has at
most ordinary double points. (The problem of the existence of a K¨ahler–
Einstein metric on smooth del Pezzo surfaces is solved in [T2].)
Further we will study global Ginvariant log canonical thresholds of
some smooth del Pezzo surfaces admitting an action of a ﬁnite group G.
Let us consider two examples.
Example 1.9. The simple group PGL(2, F
7
) is a group of automorphisms
of the quartic
x
3
y +y
3
z +z
3
x = 0 ⊂ P
2
∼
= Proj
C[x, y, z]
,
which induces PGL(2, F
7
) ⊂ Aut(P
2
). Then lct(P
2
, PGL(2, F
7
)) = 4/3 by
Lemma 5.1.
Example 1.10. Let X be a del Pezzo surface with ordinary double points
that is given by
4
¸
i=0
x
2
i
=
4
¸
i=0
λ
i
x
2
i
= 0 ⊆ P
4
∼
= Proj
C[x
0
, . . . , x
4
]
,
where λ
1
, . . . , λ
4
∈ C. Then lct(X, Z
4
2
) = 1 by Lemma 5.1.
1122 I. CHELTSOV GAFA
There is a crucial diﬀerence between the two and higherdimensional
cases: in the latter case, we usually assume that G is trivial. For surfaces,
it is not so, and applications are more special.
Example 1.11. Let X be a smooth cubic surface in P
3
that is given by
the equation
x
2
y +xz
2
+zt
2
+tx
2
= 0 ⊂ P
3
∼
= Proj
C[x, y, z, t]
,
and let X
′
be a smooth del Pezzo surface such that K
2
X
′
= 5. Then
Aut(X)
∼
= Aut(X
′
)
∼
= S
5
(see [DoI]). It follows from Lemma 5.1 and Ex
ample 5.5 that lct(X, S
5
) = lct(X
′
, S
5
) = 2. There is a classical embedding
A
5
⊂ Aut(P
1
) such that the induced embeddings Aut(P
1
×X) ⊃ A
5
×S
5
⊂
Aut(P
1
×X
′
) induce the embeddings
A
5
×S
5
∼
= Ω ⊂ Bir(P
3
) ⊃ Γ
∼
= A
5
×S
5
,
respectively. Then Ω and Γ are not conjugated in Bir(P
3
) by Lemma 6.2
and Theorem 6.4.
We would like to thank H. Braden, J.P. Demailly, I. Dolgachev,
J. Koll´ar, J. Park and Yu. Prokhorov for useful comments. We would like
to give a special thanks to J. Koll´ar for pointing out a gap in the old version
of Theorem 6.4. We would like to thank the referee for helpful comments.
2 Basic Tools
Let S be a surface with canonical singularities, and D be an eﬀective Q
divisor on it.
Remark 2.1. Let B be an eﬀective Qdivisor on S such that (S, B) is log
canonical. Then
S,
1
1 −α
(D −αB)
is not log canonical if (S, D) is not log canonical, where α ∈ Q such that
0 α < 1.
Let LCS(S, D) S be a subset such that P ∈ LCS(S, D) if and only if
(S, D) is not log terminal at the point P. The set LCS(S, D) is called the
locus of log canonical singularities.
Lemma 2.2. Suppose that −(K
S
+D) is ample. Then the set LCS(S, D)
is connected.
Proof. See Theorem 17.4 in [Ket al.].
Let P be a smooth point of the surface S. Suppose that (S, D) is not
log canonical at P.
Vol. 18, 2008 LOG CANONICAL THRESHOLDS OF DEL PEZZO SURFACES 1123
Remark 2.3. The inequality mult
P
(D) > 1 holds (see [K]).
Let C be an irreducible curve on the surface S. Put D = mC + Ω,
where m is a nonnegative rational number, and Ω is an eﬀective Qdivisor
such that C ⊆ Supp(Ω).
Remark 2.4. Suppose that C ⊆ LCS(S, D). Then m 1 (see [K]).
Suppose that the inequality m 1 holds and P ∈ C.
Lemma 2.5. Suppose that C is smooth at P. Then C · Ω > 1.
Proof. See Theorem 17.6 in [Ket al.].
Let π:
¯
S → S be a birational morphism, and
¯
D is a proper transform
of D via π. Then
K¯
S
+
¯
D +
r
¸
i=1
a
i
E
i
≡ π
∗
(K
S
+D) ,
where E
i
is a πexceptional curve, and a
i
is a rational number.
Remark 2.6. The log pair (S, D) is log canonical if and only if
(
¯
S,
¯
D +
¸
r
i=1
a
i
E
i
) is log canonical.
Suppose that π is a blow up of the point P. Then r = 1 and π(E
1
) = P.
The log pair
¯
S,
¯
D + (mult
P
(D) −1)E
1
is not log canonical at some point
¯
P ∈ E
1
by Remark 2.6. But a
1
=
mult
P
(D) −1 > 0.
Corollary 2.7. The inequality mult ¯
P
(
¯
D) + mult
P
(D) > 2 holds.
Most of the described results are valid in much more general settings
(see [Ket al.] and [K]).
3 Smooth surfaces.
In this section we prove Theorem 1.7. Let X be a smooth del Pezzo surface.
Putting
1124 I. CHELTSOV GAFA
ω =
1/3 when X
∼
= F
1
or K
2
X
∈
¸
7, 9
¸
,
1/2 when X
∼
= P
1
×P
1
or K
2
X
∈
¸
5, 6
¸
,
2/3 when K
2
X
= 4 or X is a cubic surface in P
3
with an Eckardt point,
3/4 when X is a cubic surface in P
3
without Eckardt points,
3/4 when K
2
X
= 2 and  −K
X
 has a tacnodal curve,
5/6 when K
2
X
= 2 and  −K
X
 has no tacnodal curves,
5/6 when K
2
X
= 1 and  −K
X
 has a cuspidal curve,
1 when K
2
X
= 1 and  −K
X
 has no cuspidal curves,
we see that we must show that lct(X) = ω to prove Theorem 1.7. But
lct(X) ω by [P].
Suppose that the inequality lct(X) < ω holds. To prove Theorem 1.7,
we must show that this assumption leads to a contradiction. There is an
eﬀective Qdivisor D on the surface X such that the equivalence D ≡ −K
X
holds, and (X, ωD) is not log canonical at some point P ∈ X.
Lemma 3.1. The inequality K
2
X
= 1 holds.
Proof. Suppose that K
2
X
= 1. Take C ∈  −K
X
 such that P ∈ C. Then C
is an irreducible curve, and (X, ωC) is log canonical. We may assume that
C ⊆ Supp(D) by Remark 2.1. Then
1 = C · D mult
P
(D) > 1/ω 1,
which is a contradiction. The obtained contradiction completes the proof.
Lemma 3.2. The inequality K
2
X
7 holds.
Proof. The equalities lct(P
2
) = 1/3 and lct(P
1
× P
1
) = 1/2 follow from
Remarks 2.1 and 2.3, which implies that we may assume that X = F
1
to
complete the proof. Then ω = 1/3.
Let L and C be irreducible curves on X such that L
2
= 0 and C
2
= −1.
Then
−K
X
≡ 2C + 3L,
and the singularities of the log pair (X, ω(2C + 3L)) are log canonical.
It follows from Remark 2.3 that L ⊆ Supp(D), because L · D = 2.
Therefore, we may assume that C ⊆ Supp(D) by Remark 2.1. Let Z be a
general curve in C +L such that P ∈ Z. Then
3 = Z · D mult
P
(D) > 1
ω = 3 ,
which is a contradiction. The contradiction obtained completes the proof.
Vol. 18, 2008 LOG CANONICAL THRESHOLDS OF DEL PEZZO SURFACES 1125
Lemma 3.3. The inequality K
2
X
4 holds.
Proof. Suppose that K
2
X
5. Then there is a birational morphism
π: X → S such that
• The morphism π is an isomorphism in a neighborhood of P;
• Either S
∼
= F
1
or S
∼
= P
1
×P
1
or S
∼
= P
2
,
and we may assume that S
∼
= P
1
× P
1
whenever K
2
X
6. Then the log
pair (S, ωπ(D)) is not log canonical at π(P). But π(D) ≡ −K
S
, which is
impossible by Lemma 3.2.
Lemma 3.4. The inequality K
2
X
= 4 holds.
Proof. Suppose that K
2
X
= 4. Then X is an intersection of two quadrics
in P
4
, and
D =
r
¸
i=1
a
i
C
i
≡ −K
X
,
where C
i
is an irreducible curve on the surface X, and 0 a
i
∈ Q.
The equality ω = 2/3 holds. Suppose that a
k
> 1/ω = 3/2. Then
4 = −K
X
· D =
r
¸
i=1
a
i
deg(C
i
) a
k
deg(C
k
) >
3 deg(C
k
)
2
,
which implies that deg(C
k
) 2. Let Z be an irreducible curve on X such
that C
k
+ Z is cut out by a general hyperplane section of X ⊂ P
4
that
passes through C
k
. Then
3 4 −deg(C
k
) = Z · D =
r
¸
i=1
a
i
(Z · C
i
) a
k
(Z · C
k
) = 2a
k
> 3 ,
which is a contradiction. Therefore, we see that ωa
i
1 for every i =
1, . . . , r.
There is λ ∈ Q such that 0 < λ < ω = 2/3 and (X, λD) is not log
canonical at P. Then
LCS(X, λD) = {P}
by Lemma 2.2. But there is a birational morphism π: X →P
2
such that π
is an isomorphism in a neighborhood of the point P. Then π(D) ≡ −λK
P
2.
Let L be a general line on P
2
. Then
π(P) ∪ L ⊆ LCS
P
2
, π(D) +L
,
which is impossible by Lemma 2.2. The obtained contradiction completes
the proof.
1126 I. CHELTSOV GAFA
Let π: U → X be a blow up of the point P, and E be the exceptional
curve of π. Then
¯
D ≡ π
∗
(−K
X
) −mult
P
(D)E ,
where
¯
D is the proper transform of D on the surface U. It follows from
Remark 2.6 that
U, ω
¯
D +ω(mult
P
(D) −1)E
is not log canonical at some point Q ∈ E. Then mult
Q
(
¯
D) + mult
P
(D) >
2/ω by Corollary 2.7.
Lemma 3.5. The inequality K
2
X
= 2 holds.
Proof. Suppose that K
2
X
= 2. There is a double cover ψ: X →P
2
such that
ψ is branched over a smooth quartic curve C ⊂ P
2
. Then either ψ(P) ∈ C
or ψ(P) ∈ C.
Suppose that ψ(P) ∈ C. There is a curve L ∈  − K
X
 that is singular
at P, and we may assume that at least one irreducible component of the
curve L is not contained in the support of the divisor D by Remark 2.1,
because (X, ωL) is log canonical (see [P]). Then
2 = L · D mult
P
(D) mult
P
(L) 2/ω > 2
in the case when L is irreducible. So, we must have L = L
1
+L
2
, where L
1
and L
2
are irreducible smooth curves such that L
1
·L
2
=2 and L
2
1
=L
2
2
= −1.
Without loss of generality, we may assume that L ⊂ Supp(D). Put D =
mL
1
+ Ω, where 0 m ∈ Q, and Ω is an eﬀective Qdivisor such that
L
1
⊆ Supp(Ω). Then
m+ 1 < 2m+ Ω · L
2
= D · L
2
= 1 ,
which is a contradiction. Therefore, we see that ψ(P) ∈ C.
In particular, the log pair (X, ωD) is log canonical outside of ﬁnitely
many points.
There is a unique curve Z ∈  −K
X
 such that P ∈ Z and Q ∈
¯
Z, where
¯
Z is the proper transform of the curve Z on the surface U. Then Z consists
of at most two components.
Suppose that Z is irreducible. We may assume Z ⊆ Supp(D). Hence,
we have
2 −mult
P
(D) =
¯
Z ·
¯
D mult
Q
(
¯
D) > 2/ω −mult
P
(D)
which is a contradiction. So, we must have Z = Z
1
+Z
2
, where Z
1
and Z
1
are irreducible smooth curves such that Z
1
· Z
2
= 2 and Z
2
1
= Z
2
2
= −1.
We may assume that P ∈ Z
1
and P ∈ Z
2
.
Vol. 18, 2008 LOG CANONICAL THRESHOLDS OF DEL PEZZO SURFACES 1127
It is easy to see that the log pair (X, ωZ
1
+ωZ
2
) is log canonical. Thus,
we may assume that either Z
1
⊆ Supp(D) or Z
2
⊆ Supp(D) by Remark 2.1.
But
1 = Z
1
· D mult
P
(D) 1/ω > 1 ,
which implies that Z
2
⊆ Supp(D). Then Z
1
⊆ Supp(D). Put D = ¯ mZ
1
+Υ,
where 0 < ¯ m ∈ Q, and Υ is an eﬀective Qdivisor on the surface X such
that Z
1
⊆ Supp(Υ). Then
2 ¯ m 2 ¯ m+ Υ· Z
2
= D · Z
2
= 1 ,
which gives ¯ m 1/2. But Q ∈
¯
Z
1
, where
¯
Z it the proper transform of Z
1
on the surface U. Then
2−mult
P
(D)1−mult
P
(D)+2 ¯ m=
¯
Z
1
·
¯
Υ > 2/ω−mult
P
(D)>2−mult
P
(D)
by Lemma 2.5. The obtained contradiction completes the proof.
It follows from Lemmas 3.2, 3.3, 3.4, 3.1, 3.5 that X is a smooth cubic
surface in P
3
.
Lemma 3.6. The cubic surface X does not have Eckardt points.
Proof. There is a birational morphism π: X → S such that
• The morphism π is an isomorphism in a neighborhood of the point P;
• The surface S is a smooth del Pezzo surface and K
2
S
= 4.
Suppose that X has an Eckardt point. (A point of a cubic surface is
an Eckardt point if the cubic contains 3 lines passing through this point.)
Then π(D) ≡ −K
S
and (S, ωπ(D)) is not log canonical at the point π(P),
which is impossible by Lemma 3.4.
Therefore, we see that ω = 3/4 and mult
P
(D) > 4/3 by Remark 2.3.
Lemma 3.7. The log pair (X, ωD) is log canonical on X \ P.
Proof. Arguing as in the proof of Lemma 3.4, we see that the locus
LCS(X, ωD) contains ﬁnitely many points. Then the log pair (X, ωD)
is even log terminal on X \ P by Lemma 2.2.
Let T be the unique hyperplane section of X that is singular at P. We
may assume that the support of the divisor D does not contain at least
one irreducible component of the curve T, because (S, ωT) is log canonical
(see [P]). The following cases are possible:
• The curve T is irreducible and U is a del Pezzo surface;
• The curve T is a union of a line and an irreducible conic intersecting
at P;
1128 I. CHELTSOV GAFA
• The curve T consists of 3 lines such that one of them does not pass
through P;
where T is reduced and −K
U
is nef and big. We exclude these cases one
by one.
Lemma 3.8. The curve T is reducible.
Proof. Suppose that T is irreducible. There is a double cover ψ: U → P
2
branched over a quartic curve. Let τ ∈ Aut(U) be an involution induced
by ψ. (The involution τ induces an involution in Bir(X) that is called the
Geiser involution.) It follows from [M] that τ(
¯
T) = E and
τ
∗
π
∗
(−K
X
)
≡ π
∗
(−2K
X
) −3E .
Let
¯
T be the proper transform of T on the surface U. Suppose that
Q ∈
¯
T. Then
3 −2 mult
P
(D) =
¯
T ·
¯
D mult
Q
(
¯
T) mult
Q
(
¯
D)
> mult
Q
(
¯
T)
8/3 −mult
P
(D)
8/3 −mult
P
(D) ,
which implies that mult
P
(D) 1/3. But mult
P
(D) > 4/3. Thus, we see
that Q ∈
¯
T.
Put
˘
Q = π◦τ(Q). Let H be the hyperplane section of X that is singular
at
˘
Q. Then T = H, because P =
˘
Q and T is smooth outside of the point P.
Then P ∈ H, because otherwise
3 = H · T mult
P
(H) mult
P
(T) + mult
˘
Q
(H) mult
˘
Q
(T) 4 .
Let
¯
H be the proper transform of H on the surface U. Put
¯
R = τ(
¯
H)
and R = π(
¯
R). Then
¯
R ≡ π
∗
(−2K
X
) −3E ,
and the curve
¯
R must be singular at the point Q.
Suppose that R is irreducible. The singularities of the log pair
X,
3
8
R
are log canonical, which implies that we may assume that R ⊆ Supp(D)
by Remark 2.1. Then
6 −3 mult
P
(D) =
¯
R ·
¯
D mult
Q
(
¯
R) mult
Q
(
¯
D) > 2
8/3 −mult
P
(D)
,
which implies that mult
P
(D) < 2/3. But mult
P
(D) > 4/3. The curve R
must be reducible
The curves R and H are reducible. So, there is a line L ⊂ X such that
P ∈ L ∋
˘
Q.
Let
¯
L be the proper transform of L on the surface U. Put
¯
Z = τ(
¯
L).
Then
¯
L · E = 0 and
¯
L ·
¯
T =
¯
L · π
∗
(−K
X
) = 1 ,
Vol. 18, 2008 LOG CANONICAL THRESHOLDS OF DEL PEZZO SURFACES 1129
which implies that
¯
Z · E = 1 and
¯
Z · π
∗
(−K
X
) = 2. We have Q ∈
¯
Z. Then
2 −mult
P
(D) =
¯
Z ·
¯
D mult
Q
(
¯
D) > 8/3 −mult
P
(D) > 2 −mult
P
(D)
in the case when
¯
Z ⊆ Supp(
¯
D). Hence, we see that
¯
Z ⊆ Supp(
¯
D).
Put Z = π(
¯
Z). Then Z is a conic and P ∈ Z. Let F be a line on X such
that F + Z is cut out by a hyperplane passing through Z. Then P ∈ F,
because T = F +Z.
Put D = ǫZ + Υ, where ǫ is a positive rational number, and Υ is an
eﬀective Qdivisor whose support does not contain the conic Z. We may
assume that F ⊆ Supp(Υ) by Remark 2.1. Then
1 = F · D = 2ǫ +F · Υ 2ǫ ,
which implies that ǫ 1/2. Let
¯
Υ be the proper transform of Υ on the
surface U. Then
2 −mult
P
(D) +ǫ =
¯
Z ·
¯
Υ > 8/3 −mult
P
(D)
by Lemma 2.5, which implies that ǫ > 2/3. But ǫ 1/2.
Therefore, there is a line L
1
⊂ X such that P ∈ L
1
.
Lemma 3.9. There is a line L
2
⊂ X such that L
1
= L
2
and P ∈ L
2
.
Proof. Suppose that there is no line L
2
⊂ X such that L
1
= L
2
and P ∈ L
2
.
Then T = L
1
+C, where C is an irreducible conic that passes through the
point P.
Let
¯
L
1
and
¯
C be the proper transforms of L
1
and C on the surface U,
respectively. Then
¯
L
2
1
= −2 , −K
U
·
¯
L
1
= 0 ,
¯
C
2
= −1 , −K
U
·
¯
C = 1 ,
but the divisor −K
U
is nef and big. There is a commutative diagram
U
π
ζ
//
W
ψ
X
ρ
// _ _ _ _ _ _
P
2
where ζ is the contraction of the curve
¯
L
1
to an ordinary double point, ψ is
a double cover branched over a quartic curve, and ρ is the projection from
the point P.
Let τ be the biregular involution of U induced by ψ. Then τ(E) =
¯
C
and
τ
∗
(
¯
L
1
) ≡
¯
L
1
, τ
∗
(E) ≡
¯
C , τ
∗
π
∗
(−K
X
)
≡ π
∗
(−2K
X
) −3E −
¯
L
1
.
1130 I. CHELTSOV GAFA
Note that we assumed earlier that the support of the divisor D does
not contain at least one irreducible component of the curve T. Then either
L
1
⊆ Supp(D) or C ⊆ Supp(D). But
¯
L
1
·
¯
D = 1 −mult
P
(D) < 0 ,
which implies that C ⊆ Supp(D) ⊇ L
1
. Put D = mL
1
+ Ω, where m is
a positive rational number, and Ω is an eﬀective Qdivisor whose support
does not contain the line L
1
. Then
m
¯
L
1
+
¯
Ω ≡ π
∗
(−K
X
) −
m+ mult
P
(Ω)
E ≡ π
∗
(−K
X
) −mult
P
(D)E ,
where
¯
Ω is the proper transform of Ω on the surface U. We have
0
¯
C ·
¯
Ω = 2 −mult
P
(Ω) + 2m < 2/3 −m,
which implies that m < 2/3. Then mult
P
(D) = mult
P
(Ω) + m, which
implies that
mult
Q
(
¯
Ω) > 8/3 −mult
P
(Ω) −m
1 + mult
Q
(
¯
L
1
)
. (3.10)
Suppose that Q ∈
¯
L
1
. Then it follows from Lemma 2.5 that
1 −mult
P
(Ω) +m =
¯
L
1
·
¯
Ω > 8/3 −mult
P
(Ω) −m,
which implies that m > 5/6. But m < 2/3. Hence, we see that Q ∈
¯
L
1
.
Suppose that Q ∈
¯
C. Then it follows from the inequality 3.10 that
2 −mult
P
(Ω) −2m =
¯
C ·
¯
Ω > 8/3 −mult
P
(Ω) −m,
which implies that m < 0. Hence, we see that Q ∈
¯
C.
We have τ(E) =
¯
C. Let H be the hyperplane section of the cubic
surface X that is singular at the point π ◦τ(Q) ∈ C. Then P ∈ H, because
C is smooth.
Let
¯
H be the proper transform of H on the surface U. Put
¯
R = τ(
¯
H)
and R = π(
¯
R). Then
¯
R ≡ π
∗
(−2K
X
) −3E −
¯
L
1
,
and the curve
¯
R is singular at the point Q by construction.
Suppose that R is irreducible. Then R+L
1
≡ −2K
X
, but
X,
3
8
(R+L
1
)
is log canonical, which implies that we may assume that R ⊆ Supp(D) by
Remark 2.1. The inequality 3.10 gives
5−2
m+mult
P
(Ω)
−m =
¯
R·
¯
Ω 2 mult
Q
(
¯
Ω) > 2
8/3−m−mult
P
(Ω)
,
which implies that m < 0. Hence, there is a line L ⊂ X such that P ∈ L
and π ◦ τ(Q) ∈ L.
Let
¯
L be the proper transform of the line L on the surface U. Then
¯
L ·
¯
C =
¯
L · π
∗
(−K
X
) = 1 and
¯
L · E =
¯
L ·
¯
L
1
= 0 ,
Vol. 18, 2008 LOG CANONICAL THRESHOLDS OF DEL PEZZO SURFACES 1131
but τ preserves the intersection form. Put
¯
Z = τ(
¯
L). Then
¯
Z · E = 1,
¯
Z ·
¯
L
1
= 0,
¯
Z · π
∗
(−K
X
) = 2.
Suppose that the support of
¯
Ω does not contain
¯
Z. Then the inequality
(3.10) implies that
2 −m−mult
P
(Ω) =
¯
Z ·
¯
Ω > 8/3 −m−mult
P
(Ω) ,
which is impossible. Thus, the support of
¯
Ω must contain the curve
¯
Z.
Put Z = π(
¯
Z). Then Z is a conic that passes through the point P. The
line L is the line on X such that the curve L+Z is cut out by a hyperplane
passing through Z. We have P ∈ F. Put
D = ǫZ +mL
1
+ Υ,
where ǫ is a positive rational number, and Υ is an eﬀective Qdivisor on
the surface X such that the support of the divisor Υ does not contain the
curves Z and L
1
.
We may assume that L ⊆ Supp(Υ), because (X, ω(L+Z)) is log canon
ical. Then
1 = L · D = 2ǫ +mL · L
1
+L · Υ = 2ǫ +L · Υ 2ǫ ,
which implies that ǫ 1/2. But
¯
Z ∩
¯
L
1
= ∅. Then it follows from
Lemma 2.5 that
2 −mult
P
(D) +ǫ =
¯
Z ·
¯
Υ > 8/3 −mult
P
(D) ,
where
¯
Υ is a proper transform of Υ on the surface U. We deduce that
ǫ > 2/3. But ǫ 1/2.
We have T = L
1
+L
2
+L
3
, where L
3
is a line such that P ∈ L
3
. Then
¯
L
2
1
=
¯
L
2
2
= −2 , E ·
¯
L
1
= E ·
¯
L
2
= −K
U
·
¯
L
3
= 1 ,
−K
U
·
¯
L
1
= −K
U
·
¯
L
2
= E ·
¯
L
3
= 0 ,
¯
L
2
3
= −1 ,
where
¯
L
i
is the proper transform of L
i
on the surface U. There is a com
mutative diagram
U
π
ζ
//
W
ψ
X
ρ
// _ _ _ _ _ _
P
2
,
where ζ is the contraction of the curves
¯
L
1
and
¯
L
2
to ordinary double
points, ψ is a double cover branched over a quartic curve, and ρ is the
projection from the point P.
Let τ be the biregular involution of the surface U induced by ψ. Then
τ
∗
π
∗
(−K
X
)
≡ π
∗
(−2K
X
) −3E −
¯
L
1
−
¯
L
2
,
1132 I. CHELTSOV GAFA
and τ(
¯
L
1
) =
¯
L
1
, τ(
¯
L
2
) =
¯
L
2
, τ(
¯
L
3
) = E. Recall that mult
P
(D) > 4/3 by
Remark 2.3.
We assume that T ⊆ Supp(D). Then Supp(D) does not contain one of
L
1
, L
2
, L
3
. But
¯
L
1
·
¯
D =
¯
L
2
·
¯
D = 1 −mult
P
(D) < 0 ,
which implies that L
2
⊆ Supp(D) ⊇ L
2
and L
3
⊆ Supp(D). Put
D = m
1
L
1
+m
2
L
2
+ Ω,
where 0 < m
i
∈ Q, and Ω is an eﬀective Qdivisor such that L
2
⊆
Supp(Ω) ⊇ L
2
.
The inequality m
1
+m
2
1 holds, because 1 −m
1
−m
2
= L
3
· Ω 0.
Let
¯
Ω be the proper transform of Ω on the surface U. Then
m
1
¯
L
1
+m
2
¯
L
2
+
¯
Ω ≡ π
∗
(−K
X
) −
m
1
+m
2
+ mult
P
(Ω)
E ,
where m
1
+m
2
+ mult
P
(Ω) = mult
P
(D). The latter equality implies that
mult
Q
(
¯
Ω) > 8/3 −mult
P
(Ω) −m
1
1 + mult
Q
(
¯
L
1
)
−m
1
1 + mult
Q
(
¯
L
2
)
. (3.11)
Lemma 3.12. The curves
¯
L
1
and
¯
L
2
do not contain the point Q.
Proof. Suppose that Q ∈
¯
L
1
∪
¯
L
2
. Without loss of generality we may
assume that Q ∈
¯
L
1
. Then
1 −mult
P
(Ω) −m
2
+m
1
=
¯
L
1
·
¯
Ω > 8/3 −mult
P
(Ω) −m
1
−m
2
by Lemma 2.5. We have m
1
> 5/6. Then
1 −m
1
+m
2
= Ω · L
2
> 4/3 −m
1
−m
2
,
which implies the inequality m
2
> 1/6. The latter contradicts the inequal
ity m
1
+m
2
1.
Therefore, the point π ◦τ(Q) is contained in the line L
3
, but π ◦τ(Q) ∈
L
1
∪ L
2
.
Lemma 3.13. The line L
3
is the only line on X that passes through the
point π ◦ τ(Q).
Proof. Suppose that there is a line L⊂X such that L=L
3
and π ◦ τ(Q) ∈L.
Then
¯
L ·
¯
L
1
=
¯
L ·
¯
L
2
=
¯
L · E = 0 ,
¯
L · π
∗
(−K
X
) =
¯
L ·
¯
L
3
= 1 ,
where
¯
L is the proper transform of the line L on the surface U.
The involution τ preserves the intersection form. Put
¯
Z = τ(
¯
L) and
Z = π(
¯
Z). Then
¯
Z · E = 1 ,
¯
Z ·
¯
L
3
= 0 ,
¯
Z · π
∗
(−K
X
) = 2 ,
which implies that the curve π(
¯
Z) is a conic passing through the point P.
Vol. 18, 2008 LOG CANONICAL THRESHOLDS OF DEL PEZZO SURFACES 1133
The support of the divisor Ω contains the conic Z, because otherwise
2 −m
1
−m
2
−mult
P
(Ω) =
¯
Z ·
¯
Ω > 8/3 −m
1
−m
2
−mult
P
(Ω) ,
which is impossible. Put D = ǫZ +m
1
L
1
+m
2
L
2
+Υ, where ǫ is a positive
rational number, and Υ is an eﬀective Qdivisor on X whose support does
not contain Z, L
1
, L
2
.
The line L is the line on the surface X such that the curve L+Z is cut
out by a hyperplane that passes through the conic Z. We may assume that
the support of Υ does not contain the line L by Remark 2.1, because the
log pair (X, ω(L +Z)) is log canonical. Then
1 = L · D = 2ǫ +m
1
L · L
1
+m
2
L · L
2
+L · Υ = 2ǫ +L · Υ 2ǫ ,
which implies that ǫ 1/2. But Q ∈
¯
L
1
and Q ∈
¯
L
2
by Lemma 3.12.
Thus, the log pair
U, ǫ
¯
Z +ω
¯
Υ + (ω mult
P
(D) −1)E
is not log canonical at the point Q, where
¯
Υ is a proper transform of Υ on
the surface U. Then
2 −mult
P
(D) +ǫ = 2 −mult
P
(D) +ǫ −m
1
¯
L
1
·
¯
Z −m
2
¯
L
2
·
¯
Z
=
¯
Z ·
¯
Υ > 8/3 −mult
P
(D)
by Lemma 2.5, which implies that ǫ > 2/3. But ǫ 1/2.
Let C ⊂ X be a conic such that C + L
3
is cut out by the hyperplane
tangent to X at π ◦ τ(Q), and let
¯
C be the proper transform of C on the
surface U. Put
¯
Z = τ(
¯
C) and Z = π(
¯
Z). Then
¯
Z ≡ π
∗
(−2K
X
) −4E −
¯
L
1
−
¯
L
2
,
and Z is singular at P. We have
¯
Z · E = 2 and
¯
Z ·
¯
L
1
=
¯
Z ·
¯
L
2
= 0, because
C ∩ L
1
= C ∩ L
2
= ∅.
Lemma 3.14. The support of the divisor D contains Z.
Proof. Suppose that Z ⊆ Supp(D). Then it follows from Corollary 2.7 that
4 −2 mult
P
(D) =
¯
Z ·
¯
D mult
Q
(
¯
D) > 8/3 −mult
P
(D) ,
which implies that mult
P
(D) < 4/3. But mult
P
(D) > 4/3.
Put D = ǫZ + m
1
L
1
+ m
2
L
2
+ Υ, where 0 < ǫ ∈ Q, and Υ is an
eﬀective Qdivisor whose support does not contain the curves Z, L
1
, L
2
.
Then L
1
+L
2
+Z ≡ −2K
X
and
D · L
1
= m
2
−m
1
+ 2ǫ +L
1
· Υ = D · L
2
= m
1
−m
2
+ 2ǫ +L
2
· Υ = 1 ,
which implies that ǫ 1/2. Let
¯
Υ be a proper transform of Υ on the
surface U. Then
4 −2 mult
P
(D) =
¯
Z ·
¯
Υ > 8/3 −mult
P
(D)
by Lemma 2.5, which implies that mult
P
(D) < 4/3. But mult
P
(D) > 4/3.
1134 I. CHELTSOV GAFA
The contradiction obtained completes the proof Theorem 1.7.
4 Singular Surfaces
Let X be a del Pezzo surface with Du Val singularities such that K
2
X
= 1,
and singularities of the surface X consist of ﬁnitely many points of type A
1
or A
2
. Put
ω =
1 when  −K
X
 does not have cuspidal curves,
2/3 when  −K
X
 has a cuspidal curve C such that
Sing(C) is a point of type A
2
,
5/6 when  −K
X
 has cuspidal curves, but their cusps
are not contained in Sing(S),
3/4 in the remaining cases.
Lemma 4.1. The equality lct(X) = ω holds.
Proof. Taking into a consideration curves in −K
X
, we see that lct(X) ω.
Thus, to conclude the proof, we may assume that lct(X) < ω. Then there
is an eﬀective Qdivisor D on the surface X such that D ≡ −K
X
, but
(X, λD) is not log terminal and for some ω > λ ∈ Q.
Suppose that LCS(X, λD) is not zerodimensional. There is an irre
ducible curve C such that
D = mC + Ω
where 1 < 1/λ m ∈ Q, and Ω is an eﬀective Qdivisor such that C ⊆
Supp(Ω). Then
1 = H · D = mH · C +H · Ω > m > 1 ,
where H is a general curve in the pencil −K
X
. Thus, the locus LCS(X, λD)
is zerodimensional.
It follows from Lemma 2.2 that the locus LCS(X, λD) consists of a
single point P ∈ X.
Let Z be the curve in  −K
X
 such that P ∈ Z. Arguing as in the proof
of Lemma 3.1, we see that we may assume that P ∈ Sing(X).
We may assume that Z ⊆ Supp(D), because (X, ωZ) is log canonical,
and Z is irreducible.
Suppose that P is a point of type A
1
. Let π: U → X be a blow up of
the point P. Then
¯
D ≡ π
∗
(−K
X
) −aE,
¯
Z ≡ π
∗
(−K
X
) −E,
Vol. 18, 2008 LOG CANONICAL THRESHOLDS OF DEL PEZZO SURFACES 1135
where
¯
D and
¯
Z are proper transforms of D and Z on the surface U, respec
tively, E is the exceptional curve of π, and a is a positive rational number.
Then a 1/2, because 1 −2a =
¯
Z ·
¯
D 0.
The log pair (U, λ
¯
D+λaE) is not log terminal at some point Q ∈ E by
Remark 2.6. Then
1 2a = E ·
¯
D > 1/λ > 1
by Lemma 2.5, which is a contradiction. Thus, the point P is a singular
point of type A
2
.
There is a birational morphism ζ : W → X such that ζ contracts two
irreducible smooth rational curves E
1
and E
2
to the point P, the morphism
ζ induces an isomorphism
W \ (E
1
∪ E
2
)
∼
= X \ P ,
and W is smooth along E
1
and E
2
. Then E
2
1
= E
2
2
= −2 and E
1
· E
2
= 1.
But
`
D ≡ ζ
∗
(−K
X
) −a
1
E
1
−a
2
E
2
,
`
Z ≡ ζ
∗
(−K
X
) −E
1
−E
2
E ,
where
`
D and
`
Z are proper transforms of D and Z on the surface W,
respectively, and 0 a
i
∈ Q.
The inequalities
`
Z ·
`
D 0, E
1
·
`
D 0, E
1
·
`
D 0 imply that
a
1
+a
2
1 , 2a
1
a
2
, 2a
2
a
1
,
respectively. Thus, we see that a
1
2/3 and a
2
2/3. But the equivalence
K
W
+λ
`
D +λa
1
E
1
+λa
2
E
2
≡ ζ
∗
(K
X
+λD)
implies the existence of a point O∈E
1
∪E
2
such that (W, λ
`
D+λa
1
E
1
+λa
2
E
2
)
is not log terminal at the point O (see Remark 2.6). Without loss of gen
erality, we may assume that O ∈ E
1
.
Suppose that O ∈ E
2
. Then (W, λ
`
D+E
1
) is not log terminal at Q. We
have
2a
1
−a
2
= E
1
·
`
D > 1/λ > 1 ,
by Lemma 2.5, which implies that a
1
> 2/3, because 2a
2
a
1
. But
a
1
2/3.
Thus, we see that O = E
1
∩ E
2
. Then
2a
1
−a
2
= E
1
·
`
D 1/λ −a
2
> 1 −a
2
,
2a
2
−a
1
= E
1
·
`
D 1/λ −a
1
> 1 −a
1
,
by Lemma 2.5, which implies that a
1
> 1/2 and a
2
> 1/2. But a
1
+a
2
1.
The assertion of Theorem 1.8 follows from Lemma 4.1.
1136 I. CHELTSOV GAFA
5 Invariant Thresholds
Let X is a smooth del Pezzo surface, let H be a Cartier divisor on X, let
G be a ﬁnite subgroup in Aut(X) such that the Ginvariant subgroup of
the group Pic(X) is ZH, and
• let r be the biggest natural number such that −K
X
∼ rH,
• let k be the smallest natural number such that k = Σ, where Σ ⊂ X
is a Gorbit,
• let m be the smallest natural number such that there is a Ginvariant
divisor in mH.
It follows from Deﬁnition 1.1 that lct(X, G) m/r.
Lemma 5.1. Suppose that h
0
(X, O
X
((m−r)H)) < k. Then lct(X, G) =
m/r.
Proof. We suppose that lct(X, G) < m/r. Then there is an eﬀective G
invariant Qdivisor D on the surface X such that LCS(X, λD) = ∅ and
D ≡ −K
X
, where 0 < λ ∈ Q such that λ < m/r.
It follows from the Nadel vanishing theorem (see [L, Th. 9.4.8]) that the
sequence
H
0
X, O
X
((m−r)H)
−→ H
0
O
L
⊗O
X
((m−r)H)
−→ 0 (5.2)
is exact, where J(λD) is the multiplier ideal sheaf of λD, and L is the
corresponding subscheme.
Suppose that L is zerodimensional. Then the exact sequence (5.2)
implies that
k > h
0
X, O
X
((m−r)H)
h
0
O
L
⊗O
X
((m−r)H)
= h
0
(O
L
)
 Supp(L) k ,
because the subscheme L is Ginvariant. Hence, the subscheme L is not
zerodimensional.
Thus, there is a Ginvariant reduced curve C on the surface X such that
λD = µC + Ω,
where µ 1, and Ω is an eﬀective onecycle on the surface X, whose
support does not contain any component of the curve C. Then C ∼ lH for
some natural number l. We have l m. But
m > λr µl l m,
because the Ginvariant subgroup of the group Pic(X) is generated by the
divisor H.
Let us show how to apply Lemma 5.1.
Vol. 18, 2008 LOG CANONICAL THRESHOLDS OF DEL PEZZO SURFACES 1137
Example 5.3. Suppose that K
2
X
= 5 and k = 1. Then X has 6 curves
E
1
, . . . , E
6
such that
6
¸
i=1
E
i
∼ −K
X
and E
2
i
= −1. The divisor
¸
6
i=1
E
i
is Ginvariant. Then lct(X, G) = 1 by
Lemma 5.1.
Example 5.4. Suppose that X = P
2
and G = A
5
such that the sub
group G leaves invariant a smooth conic on P
2
. Then lct(X, G) = 2/3 by
Lemma 5.1, because r = 3, k = 6, m = 2.
Example 5.5. Suppose that K
2
X
= 6 and G = Aut(X)
∼
= S
5
(see [RS]).
Then r = 1 and k > 6, because the stabilizer of every point induces a
faithful twodimensional linear representation in its tangent space. Then
lct(X, G) = 2 by Lemma 5.1, because m = 2 (see [RS]).
Even if h
0
(X, O
X
((m − r)H)) k, we still may be able to show that
lct(X, G) = m/r.
Lemma 5.6. Suppose that X be the cubic surface in P
3
that is given by
the equation
x
3
+y
3
+z
3
+t
3
= 0 ⊂ P
3
∼
= Proj
C[x, y, z, t]
,
and G = Aut(X). Then lct(X, G) = 4.
Proof. We have r = 1 and G
∼
= Z
3
3
⋊ S
4
(see [DoI]). Then it is easy to
check that m = 4 and k = 18, which implies that we are unable to apply
Lemma 5.1 to deduce the equality lct(X, G) = 4.
Suppose that lct(X, G) < 4. Then there is an eﬀective Ginvariant Q
divisor D on the cubic surface X such that LCS(X, λD) = ∅ and D ≡
−K
X
, where 0 < λ ∈ Q such that λ < 4.
Arguing as in the proof of Lemma 5.1, we see that the locus LCS(X, λD)
consists of 18 points, because every Gorbit containing at most 20 points
must consist of 18 points. Then
LCS(X, λD) = {O
1
, . . . , O
18
} ,
where O
1
, . . . , O
18
are all Eckardt points of the surface X (see [DoI]).
Let R be a curve on the surface X that is cut out by xyzt = 0. Then
R is Ginvariant, and the log pair (X, R) is log canonical. We may assume
that R ⊆ Supp(D) by Remark 2.1. Then
12 = R· D
18
¸
i=1
mult
O
i
(R) mult
O
i
(D) =
18
¸
i=1
2 mult
O
i
(D) 36 mult
O
i
(D) ,
which implies that mult
O
i
(D) 1/3.
1138 I. CHELTSOV GAFA
Let π: U → X be a blow up of the points O
1
, . . . , O
18
. Then
K
U
+ 4
¯
D +
18
¸
i=1
4 mult
O
i
(D) −1
E
i
≡ π
∗
(K
X
+ 4D) ,
where E
i
is the πexceptional curve such that π(E
i
) = O
i
, and
¯
D is the
proper transform of D on the surface U. Then there is Q
i
∈ E
i
such that
mult
Q
i
(
¯
D) > 1/2 −mult
O
i
(D) for i = 1, . . . , 18.
Let Σ be the Gorbit of the point Q
i
. Then Σ ∩ E
i
= Q
i
, because the
representation induced by the action of the stabilizer of O
i
on its tangent
space is irreducible. We have
mult
O
i
(D) = E
i
·
¯
D > Σ ∩ E
i

1/2 −mult
O
i
(D)
,
which implies that Σ ∩ E
i
 = 1, because mult
O
i
(D) 1/3.
Lemma 5.7. Suppose that K
2
X
= 5 and G = A
5
. Then lct(X, G) = 2.
Proof. The surface X is embedded in P
5
by the linear system −K
X
, and
X contains 10 lines, which we denote as L
1
, . . . , L
10
. Then r = 1 and
Aut(X)
∼
= S
5
(see [RS]).
The divisor
¸
10
i=1
L
i
∼ −2K
X
is S
5
invariant, which implies that
lct(X, G) 2.
The surface X can be obtained as a blow up π: X → P
2
of the four
points
P
1
= (1 : −1 : −1) , P
2
= (−1 : 1 : −1) ,
P
3
= (−1 : −1 : 1) , P
4
= (1 : 1 : 1) ,
of the plane P
2
. Let W be the curve in P
2
that is given by the equation
x
6
+y
6
+z
6
+(x
2
+y
2
+z
2
)(x
4
+y
4
+z
4
) = 12x
2
y
2
z
2
⊂ P
2
∼
= Proj
C[x, y, z]
,
and Z be its proper transform on X. Then Z is S
5
invariant (see [IK]) and
Z ∼ −2K
X
.
The curves Z and
¸
10
i=1
L
i
are the only S
5
invariant curves in  −2K
X
.
Let P be the pencil generated by Z and
¸
10
i=1
L
i
. It follows from [E]
that P is A
5
invariant, and there are exactly 5 singular curves in P, which
can be described in the following way:
• the curve
¸
10
i=1
L
i
;
• two irreducible rational curves R
1
and R
2
that have 6 nodes;
• two ﬁbers F
1
and F
2
each consisting of 5 smooth rational curves.
We have m = 2 and k = 6 by [RS]. The smallest Gorbit are Sing(R
1
)
and Sing(R
2
) (see [IK]).
Vol. 18, 2008 LOG CANONICAL THRESHOLDS OF DEL PEZZO SURFACES 1139
Suppose that lct(X, G) < 2. Then there is an eﬀective Ginvariant
Qdivisor D on the quintic surface X such that LCS(X, λD) = ∅ and
D ≡ −K
X
, where 0 < λ ∈ Q such that λ < 2.
We may assume that the support of D does not contain R
1
and R
2
due to Remark 2.1, because both log pairs (X, R
1
) and (X, R
2
) are log
canonical. Now arguing as in the proof of Lemma 5.1, we see that either
LCS(X, λD) = Sing(R
1
) or LCS(X, λD) = Sing(R
2
).
Without loss of generality we may assume that the locus LCS(X, λD)
consists of the singular points of the curve R
1
. Denote them as O
1
, . . . , O
6
.
Then mult
O
i
(D) 5/6, because
10 = R
1
· D
6
¸
i=1
mult
O
i
(D) mult
O
i
(R
1
) 12 mult
O
i
(D) .
Let π: U → X be a blow up of the points O
1
, . . . , O
6
. Then
K
U
+ 2
¯
D +
6
¸
i=1
2 mult
O
i
(D) −1
E
i
≡ π
∗
(K
X
+ 2D) ,
where E
i
is the πexceptional curve such that π(E
i
) = O
i
, and
¯
D is the
proper transform of D on the surface U. Then mult
Q
i
(
¯
D) > 1−mult
O
i
(D)
for some point Q
i
∈ E
i
, where i = 1, . . . , 6.
Let Σ be the Gorbit of the point Q
i
. Then Σ ∩ E
i
 2, because the
stabilizer of O
i
acts faithfully on its tangent space. We have Σ ∩ E
i
 = 2,
because mult
O
i
(D) 5/6 and
mult
O
i
(D) = E
i
·
¯
D > Σ ∩ E
i

1 −mult
O
i
(D)
.
Let
¯
R
1
be the proper transform of the curve R
1
on the surface U. Then
Σ =
¯
R
1
¸
(E
1
∪ E
2
∪ E
3
∪ E
4
∪ E
5
∪ E
5
) ,
because the orbit of length 2 of the action on E
i
of the stabilizer of O
i
is
unique. We have
12
1 −mult
O
i
(D)
= 10 −2
6
¸
i=1
mult
O
i
(D) =
¯
R
1
·
¯
D 2
6
¸
i=1
mult
Q
i
(
¯
D)
> 12
1 −mult
O
i
(D)
,
which is a contradiction.
Lemma 5.8. Suppose that K
2
X
= 5 and G = Z
5
. Then lct(X, G) = 4/5
holds.
1140 I. CHELTSOV GAFA
Proof. It is well known that the group G ﬁxes exactly two points of the
surfaces X (see [RS]), which we denote as O
1
and O
2
. There are ﬁve conics
Z
1
, . . . , Z
5
⊂ X that passes through O
1
, and the divisor
¸
5
i=1
Z
i
∼ −2K
X
is Ginvariant, which implies that lct(X, G) 4/5.
Suppose that lct(X, G) < 4/5. Then there is an eﬀective Ginvariant
Qdivisor D on the quintic surface X such that LCS(X, λD) = ∅ and
D ≡ −K
X
, where 0 < λ ∈ Q such that λ < 4/5.
The proof of Lemma 5.1 implies that LCS(X,λD) ={O
1
} or LCS(X,λD)
= {O
1
}.
Without loss of generality, we may assume that LCS(X, λD) = {O
1
},
and we may assume that the support of the divisor D does not contain the
conics Z
1
, . . . , Z
5
by Remark 2.1. Then
2 = Z
1
· D mult
O
1
(D) .
Let π: U → X be a blow up of the point O
1
, and E be the πexceptional
curve. Then
mult
Q
(
¯
D) 2/λ −mult
O
1
(D) > 5/2 −mult
O
1
(D)
for some point Q ∈ E by Corollary 2.7, where
¯
D is the proper transform
of D on the surface U.
The point Q must be Ginvariant, because otherwise
mult
O
1
(D) = E ·
¯
D > 5
5/2 −mult
O
1
(D)
,
which is impossible, because mult
O
1
(D) 2.
Let
¯
Z
i
be the proper transform of the conic Z
i
on the surface U. Then
Q ∈ ∪
5
i=1
¯
Z
i
, and there is a birational morphism φ: U → P
2
that contracts
the curves
¯
Z
1
, . . . ,
¯
Z
5
.
The curve φ(E) is a conic that contains φ(
¯
Z
1
), . . . , φ(
¯
Z
5
). Let T
i
be the
proper transform on the surface U of the line in P
2
that passes through the
points φ(Q) and φ(
¯
Z
i
). The log pair
X,
λ
3
5
¸
i=1
π(T
i
)
has log terminal singularities, and
¸
5
i=1
π(T
i
) ≡ 3D. Thus, we may as
sume that the support of the divisor
¯
D does not contain any of the curves
T
1
, · · · , T
5
due to Remark 2.1. Then
3 −mult
O
1
(D) T
i
·
¯
D mult
Q
(
¯
D) ,
which implies that mult
O
1
(D) + mult
Q
(
¯
D) 3.
Let ξ : V → U be a blow up of the point Q, and F be the ξexceptional
divisor. Then
Vol. 18, 2008 LOG CANONICAL THRESHOLDS OF DEL PEZZO SURFACES 1141
K
W
+λ
`
D +
λmult
O
1
(D) −1
`
E +
λmult
O
1
(D) +λmult
Q
(
¯
D) −2
F
≡ (π ◦ ξ)
∗
(K
X
+λD) ,
where
`
D and
`
E are proper transforms of D and E on the surface V , re
spectively. The log pair
W, λ
`
D + (λmult
O
1
(D) −1)
`
E + (λmult
O
1
(D) +λmult
Q
(
¯
D) −2)F
is not log terminal at some point P ∈ F by Remark 2.6, because mult
O
1
(D)
2.
Suppose that P ∈
`
E. Let
`
T be the proper transform on V of the line
on P
2
that is tangent to the conic φ(E) at the point φ(Q). Then P ∈
`
T,
which implies that
5 −2 mult
O
1
(D) −mult
Q
(
¯
D) =
`
T ·
`
D mult
P
(
`
D)
> 5 −2 mult
O
1
(D) −mult
Q
(
¯
D) ,
because we may assume that
`
T ⊆ Supp(
`
D) by Remark 2.1. Hence, we have
P ∈
`
E.
The log pair (W, λ
`
D + (λmult
O
1
(D) + λmult
Q
(
¯
D) − 2)F) is not log
terminal at P. But
λ
`
D +
λmult
O
1
(D) +λmult
Q
(
¯
D) −2
F
is an eﬀective divisor, because mult
Q
(
¯
D) 2/λ −mult
O
1
(D). Then
mult
P
(
`
D) 3/λ−mult
O
1
(D)−mult
Q
(
¯
D) > 15/4−mult
O
1
(D)−mult
Q
(
¯
D) .
Let
`
T
i
be the proper transform of T
i
on the surface V . Suppose that
P ∈
`
T
k
. Then
3 −mult
O
1
(D) −mult
Q
(
¯
D) =
`
T
k
·
`
D > 15/4 −mult
O
1
(D) −mult
Q
(
¯
D) ,
which is a contradiction. Thus, we see that P ∈ ∪
5
i=1
`
T
i
.
Let M be an irreducible curve on V such that P ∈ M, the curve φ◦ξ(M)
is a line that passes through the point φ(Q). Then π◦ξ(M) has an ordinary
double point at O
1
, and π ◦ ξ(M) ≡ −K
X
, because P ∈ ∪
5
i=1
`
T
i
. We may
assume that M ⊆ Supp(
`
D) by Remark 2.1. Then
5 −2 mult
O
1
(D) −mult
Q
(
¯
D) = M ·
`
D > 15/4 −mult
O
1
(D) −mult
Q
(
¯
D) ,
which implies that mult
O
1
(D) 5/4. But mult
O
1
(D) > 5/4.
We did not prove that groups in Example 5.5 and Lemmata 5.6, 5.7 and
5.8 act on X in such a way that the Ginvariant subgroup in Pic(X) is Z.
But the latter is well known (see [DoI]).
1142 I. CHELTSOV GAFA
6 Direct Products
Let X be an arbitrary smooth Fano variety, and let G be a ﬁnite subgroup
in Aut(X) such that the Ginvariant subgroup of the group Pic(X) is Z.
definition 6.1. The variety X is said to be Gbirationally superrigid
if for every Ginvariant linear system M on the variety X that does not
have any ﬁxed components, the singularities of the log pair (X, λM) are
canonical, where λ ∈ Q such that λ > 0 and K
X
+λM≡ 0.
The following result is well known (see [M], [DoI]).
Lemma 6.2. Suppose that X is a smooth del Pezzo surface such that
Σ K
2
X
for any Gorbit Σ ⊂ X. Then X is Gbirationally superrigid.
Proof. Suppose that the surface X is not Gbirationally superrigid. Then
there is a Ginvariant linear system M on the surface X such that M does
not have ﬁxed curves, but (X, λM) is not canonical at some point O ∈ X,
where λ ∈ Q such that λ > 0 and K
X
+λM≡ 0.
Let Σ be the Gorbit of the point O. Then mult
P
(M) > 1/λ for every
point P ∈ Σ. Then
K
2
X
/λ
2
= M
1
· M
2
¸
P∈Σ
mult
2
P
(M) > Σ/λ
2
K
2
X
/λ
2
,
where M
1
and M
2
are suﬃciently general curves in M.
Example 6.3. Let X be a smooth del Pezzo surface such that K
2
X
= 5.
Then Aut(X)
∼
= S
5
, and the proof of Lemma 5.7 implies that the surface
X is A
5
birationally superrigid by Lemma 6.2.
Let X
i
be a smooth G
i
birationally superrigid Fano variety, where G
i
is a an arbitrary ﬁnite subgroup of Aut(X
i
) such that the G
i
invariant
subgroup of Pic(X
i
) is Z, and i = 1, . . . , r.
Theorem 6.4. Suppose that lct(X
i
, G
i
) 1 for every i = 1, . . . , r. Then
• there is no G
1
×· · · ×G
r
equivariant birational map
ρ: X
1
×· · · ×X
r
P
n
;
• every G
1
×· · · ×G
r
equivariant birational automorphism of
X
1
×· · · ×X
r
is biregular;
• for any G
1
×· · ·×G
r
equivariant dominant map ρ : X
1
×· · · ×X
r
Y ,
whose general ﬁber is rationally connected, there a commutative
Vol. 18, 2008 LOG CANONICAL THRESHOLDS OF DEL PEZZO SURFACES 1143
diagram
X
1
×· · · ×X
r
π
ρ
++
W
W
W
W
W
W
W
W
W
W
W
W
W
X
i
1
×· · · ×X
i
k
ξ
// _ _ _ _ _ _ _ _ _ _ _ _
Y
where ξ is a birational map, π is a natural projection, and {i
1
, . . . , i
k
}
⊆ {1, . . . , r}.
Proof. The required assertion follows from the proof of Theorem 1 in [Pu].
Example 6.5. The simple group A
6
is a group of automorphisms of the
sextic
10x
3
y
3
+9zx
5
+9zy
5
+27z
6
= 45x
2
y
2
z
2
+135xyz
4
⊂ P
2
∼
= Proj
C[x, y, z]
and there is an embedding A
6
⊂ Aut(P
2
) such that lct(P
2
, A
6
) = 2 by
Lemma 5.1 (see [Cr]), and A
6
×A
6
acts naturally on P
2
×P
2
. There is an
induced embedding A
6
×A
6
∼
= Ω ⊂ Bir(P
4
) such that Ω is not conjugated
to a subgroup of Aut(P
4
) by Lemma 6.2 and Theorem 6.4.
References
[C1] I. Cheltsov, Log canonical thresholds on hypersurfaces, Sbornik: Math
ematics 192 (2001), 1241–1257.
[C2] I. Cheltsov, Fano varieties with many selfmaps, Advances in Mathemat
ics 217 (2008), 97–124.
[CS] I. Cheltsov, S. Shramov, Loc canonical thresholds of smooth Fano
threefolds, with appendix by J.P. Demailly, Russian Math. Surveys
63:5 (2008), 73–180.
[Cr] S. Crass, Solving the sextic by iteration: a study in complex geometry
and dynamics, Experimental Mathematics 8 (1999), 209–240.
[DK] J.P. Demailly, J. Koll´ ar, Semicontinuity of complex singularity ex
ponents and K¨ahler–Einstein metrics on Fano orbifolds, Annales Scien
tiﬁques de l’
´
Ecole Normale Sup´erieure 34 (2001), 525–556.
[DoI] I. Dolgachev, V. Iskovskikh, Finite subgroups of the plane Cremona
group, arXiv:math.AG/0610595 (2006).
[E] W. Edge, A pencil of fournodal plane sextics, Mathematical Proceedings
of the Cambridge Philosophical Society 89 (1981), 413–421.
[H1] J.M. Hwang, Log canonical thresholds of divisors on Grassmannians,
Mathematische Annalen 334 (2006), 413–418.
[H2] J.M. Hwang, Log canonical thresholds of divisors on Fano manifolds of
Picard rank 1, Compositio Mathematica 143 (2007), 89–94.
1144 I. CHELTSOV GAFA
[IK] N. Inoue, F. Kato, On the geometry of Wiman’s sextic, Journal of
Mathematics of Kyoto University 45 (2005), 743–757.
[K] J. Koll´ ar, Singularities of pairs, Proceedings of Symposia in Pure Math
ematics 62 (1997), 221–287.
[Ket al.] J. Koll´ ar et al., Flips and abundance for algebraic threefolds,
Ast´erisque 211 (1992).
[L] R. Lazarsfeld, Positivity in Algebraic Geometry II, SpringerVerlag,
Berlin, 2004.
[M] Yu. Manin, Rational surfaces over perfect ﬁelds, II, Mathematics of the
USSR, Sbornik 1 (1967), 141–168.
[N] A. Nadel, Multiplier ideal sheaves and K¨ahler–Einstein metrics of positive
scalar curvature, Annals of Mathematics 132 (1990), 549–596.
[P] J. Park, Birational maps of del Pezzo ﬁbrations, Journal fur die Reine
und Angewandte Mathematik 538 (2001), 213–221.
[Pu] A. Pukhlikov, Birational geometry of Fano direct products, Izvestiya:
Mathematics 69 (2005), 1225–1255.
[RS] J. Rauschning, P. Slodowy, An aspect of icosahedral symmetry, Cana
dian Mathematical Bulletin 45 (2005), 686–696.
[S] V. Shokurov, Threedimensional log perestroikas, Izvestiya: Mathemat
ics 56:1 (1992), 105–203.
[T1] G. Tian, On K¨ahler–Einstein metrics on certain K¨ahler manifolds with
c
1
(M) > 0, Inventiones Mathematicae 89 (1987), 225–246.
[T2] G. Tian, On Calabi’s conjecture for complex surfaces with positive ﬁrst
Chern class, Inventiones Mathematicae 101 (1990), 101–172.
[T3] G. Tian, On a set of polarized K¨ahler metrics on algebraic manifolds,
Journal of Diﬀerential Geometry 32 (1990), 99–130.
Ivan Cheltsov, School of Mathematics, University of Edinburgh, Edinburgh EH9
3JZ, UK, I.Cheltsov@ed.ac.uk
Received: May 2007
Revision: October 2007
Accepted: April 2008
Vol. 18, 2008
LOG CANONICAL THRESHOLDS OF DEL PEZZO SURFACES
1119
log terminal singularities, let Z ⊆ X be a closed subvariety, and let D be an eﬀective QCartier Qdivisor on X. Then the number lctZ (X, D) = sup λ ∈ Q  the log pair (X, λD) is log canonical along Z is said to be the log canonical threshold of D along Z. The number lctZ (X, D) is known to be positive and rational. Moreover, if X = Cn and D = (φ = 0), then the equality lctO (X, D) = c0 (φ) holds (see [K]). For the case Z = X we use the notation lct(X, D) instead of lctX (X, D). Then lct(X, D) = inf lctP (X, D)  P ∈ X = sup λ ∈ Q  the log pair (X, λD) is log canonical . Even though several methods have been invented in order to compute log canonical thresholds, it is not easy to compute them in general. However, the log canonical thresholds play a signiﬁcant role in the study of birational geometry showing many interesting properties (see [K], [P]). Thus far the log canonical threshold has a local character. In this paper we wish to develop its global analogue for Fano varieties. We shall see it is useful to consider the smallest of the log canonical thresholds of eﬀective Qdivisors numerically equivalent to an anticanonical divisor. Let X be a Fano variety with log terminal singularities, and G be a ﬁnite subgroup in Aut(X). definition 1.1. We deﬁne the global Ginvariant log canonical threshold of X by the number lct(X, G) = inf lct(X, D) the eﬀective Qdivisor D is Ginvariant and D ≡ −KX . We put lct(X) = lct(X, G) if the group G is trivial. Note that it follows from Deﬁnition 1.1 that lct(X, G) = sup λ ∈ Q the log pair (X, λD) has log canonical singularities for every Ginvariant eﬀective Qdivisor D ≡ −KX 0.
Example 1.2. It follows from Proposition 16.9 in [K et al.] that lct(P(1, 1, n)) = 1/(2 + n) for n ∈ N. For a given Fano variety, it is usually very hard to compute its global log canonical threshold explicitly (see [C2]). For instance, the papers [H1] and
6. D) holds for some Ginvariant eﬀective Qdivisor D on the variety X such that D ≡ −KX .5. lct X 1−1 n due to [C1]. a4 ) of degree 4 ai with terminal singularities such that i=1 3 −KX 1. Let X be a general wellformed quasismooth hypersurface in P(1. (It follows from [CS. Let X be a smooth del Pezzo surface. The main purpose of this paper is to prove the following result. . One of the most interesting applications of global log canonical thresholds of Fano varieties is the following result proved in [DK] (see also [N]. Global log canonical thresholds of Fano varieties play an important role in geometry. [T1] and [CS]). Theorem 1. A] that global log canonical thresholds of Fano varieties are algebraic counterparts of αinvariants introduced in [T1]. . . Append. Then X has a Ginvariant orbifold K¨hler–Einstein metric.4.7. Then . G) > dim(X) + 1 holds. which implies that m Bir(X × · · · × X ) = m times i=1 Bir(X). Conjecture 1. a The following conjecture is inspired by [T3. Example 1. It is clear that the inequality lct(X) = 1 − 1/n holds if the hypersurface X contains a cone of dimension n − 2. Let X be a Fano variety with quotient singularities. and let G be a ﬁnite subgroup on Aut(X) such that the inequality dim(X) lct(X. a1 . For a given Fano variety X with log terminal singularities and ﬁnite subgroup G ⊂ Aut(X).3. the equality lct(X. CHELTSOV GAFA [H2] show that the global log canonical threshold of a rational homogeneous space of Picard rank 1 and Fano index r is 1/r. . Aut(X × · · · × X ) .) Example 1. Theorem 1. G) = lct(X. Then lct(X) = 1 by [C2].1120 I. But the paper [Pu] shows that lct(X) = 1 if X is general and n 6. Then Let X be a smooth hypersurface in Pn of degree n ≥ 3. m times the variety X × · · · × X is not rational and not birational to a conic bundle (see [C2]). Question 1].
) Further we will study global Ginvariant log canonical thresholds of some smooth del Pezzo surfaces admitting an action of a ﬁnite group G. F7 ) ⊂ Aut(P2 ). X X 3/4 when K 2 = 2 and  − KX  has a tacnodal curve. Then lct(X. in this paper. 18.1. 2 . . 3/4 when  − K  has a cuspidal curve C such that X lct X = Sing(C) ⊆ Sing(X). y. = where λ1 . Then 1 when  − KX  does not have cuspidal curves. Example 1. 2 1 when KX = 1 and  − KX  has no cuspidal curves.9. = X 1/3 in the remaining cases. 5/6 when K 2 = 1 and  − KX  has a cuspidal curve. . 2008 LOG CANONICAL THRESHOLDS OF DEL PEZZO SURFACES 1121 Taking the paper [P] and Theorem 1. x4 ] . F7 ) is a group of automorphisms of the quartic x3 y + y 3 z + z 3 x = 0 ⊂ P2 ∼ Proj C[x. Let X be a del Pezzo surface with ordinary double points that is given by 4 4 x2 i i=0 = i=0 2 λi xi = 0 ⊆ P4 ∼ Proj C[x0 .Vol. X 5/6 when K 2 = 2 and  − K  has no tacnodal curves. we prove the following result. . Z4 ) = 1 by Lemma 5. λ4 ∈ C. The simple group PGL(2. .6 holds for smooth del Pezzo surfaces with trivial group action. . 6}. (The problem of the existence of a K¨hler– a Einstein metric on smooth del Pezzo surfaces is solved in [T2].8. 5/6 in the remaining cases. Example 1. F7 )) = 4/3 by Lemma 5.10. Also.5 and 1. X lct X = 3/4 when X is a cubic surface in P3 without Eckardt points.8 imply the existence of an orbifold K¨hler–Einstein metric on every del Pezzo surface of degree 1 that has at a most ordinary double points. we see that the assertion of Conjecture 1. PGL(2. = which induces PGL(2. . We see that Theorems 1.7 into consideration. 1/2 when X ∼ P1 × P1 or K 2 ∈ {5. 2/3 when K 2 = 4 or X is a cubic surface in P3 X with an Eckardt point.1. Let X be a del Pezzo surface with ordinary double points 2 such that KX = 1. . Then lct(P2 . Let us consider two examples. z] . Theorem 1. .
D) is not log terminal at the point P . Let X be a smooth cubic surface in P3 that is given by the equation x2 y + xz 2 + zt2 + tx2 = 0 ⊂ P3 ∼ Proj C[x. Koll´r for pointing out a gap in the old version a of Theorem 6.1. where α ∈ Q such that 0 α < 1. = ′ be a smooth del Pezzo surface such that K 2 = 5. Example 1.].1 and Ex= = ample 5. It follows from Lemma 5. Suppose that −(KS + D) is ample.2. = = respectively.4. J. S5 ) = 2. Then 1 (D − αB) S. D) if and only if (S. We would like to thank the referee for helpful comments. Then and let X X′ Aut(X) ∼ Aut(X ′ ) ∼ S5 (see [DoI]).5 that lct(X. D) is connected.11.P. D) is called the locus of log canonical singularities. D) is not log canonical at P . Let B be an eﬀective Qdivisor on S such that (S. D) S be a subset such that P ∈ LCS(S. Koll´r. and D be an eﬀective Qdivisor on it. I.4 in [K et al. Let P be a smooth point of the surface S. Lemma 2. Park and Yu.4. Prokhorov for useful comments. t] . J. We would like a to give a special thanks to J. we usually assume that G is trivial. Proof. CHELTSOV GAFA There is a crucial diﬀerence between the two and higherdimensional cases: in the latter case. y. Then the set LCS(S. 1−α is not log canonical if (S. The set LCS(S. See Theorem 17. For surfaces. z. Demailly.1122 I. Then Ω and Γ are not conjugated in Bir(P3 ) by Lemma 6. . There is a classical embedding A5 ⊂ Aut(P1 ) such that the induced embeddings Aut(P1 × X) ⊃ A5 × S5 ⊂ Aut(P1 × X ′ ) induce the embeddings A5 × S5 ∼ Ω ⊂ Bir(P3 ) ⊃ Γ ∼ A5 × S5 . J. Dolgachev.2 and Theorem 6. Let LCS(S. S5 ) = lct(X ′ . Braden. Suppose that (S. D) is not log canonical. it is not so. Remark 2. We would like to thank H. and applications are more special. B) is log canonical. 2 Basic Tools Let S be a surface with canonical singularities.
and ai is a rational number. Corollary 2. The inequality multP (D) > 1 holds (see [K]). Suppose that C is smooth at P . ¯ ¯ Most of the described results are valid in much more general settings (see [K et al. ¯ ¯ Let π : S → S be a birational morphism. Proof.] and [K]).6. 3 Smooth surfaces. Remark 2. D + (multP (D) − 1)E1 ¯ is not log canonical at some point P ∈ E1 by Remark 2. Then m Suppose that the inequality m 1 holds and P ∈ C. and Ω is an eﬀective Qdivisor such that C ⊆ Supp(Ω). Suppose that C ⊆ LCS(S.Vol. See Theorem 17. 1 (see [K]).5. Lemma 2. Remark 2. i=1 Suppose that π is a blow up of the point P . The log pair (S.6. In this section we prove Theorem 1. D) is log canonical if and only if ¯ ¯ (S. The inequality multP (D) + multP (D) > 2 holds. Put D = mC + Ω. Then r ¯ KS + D + ¯ i=1 ai Ei ≡ π ∗ (KS + D) . Then r = 1 and π(E1 ) = P .3. Let X be a smooth del Pezzo surface.7. and D is a proper transform of D via π.6 in [K et al. D + r ai Ei ) is log canonical. Let C be an irreducible curve on the surface S. 2008 LOG CANONICAL THRESHOLDS OF DEL PEZZO SURFACES 1123 Remark 2. 18. Putting .4. where Ei is a πexceptional curve. But a1 = multP (D) − 1 > 0.]. The log pair ¯ ¯ S.7. Then C · Ω > 1. where m is a nonnegative rational number. D).
But lct(X) ω by [P]. Then 1=C ·D multP (D) > 1/ω 7 holds.3 that L ⊆ Supp(D). 5/6 when K 2 = 2 and  − KX  has no tacnodal curves. Then 3=Z·D multP (D) > 1 ω = 3 . The equalities lct(P2 ) = 1/3 and lct(P1 × P1 ) = 1/2 follow from Remarks 2. and (X. 3/4 when X is a cubic surface in P3 without Eckardt points. and (X. Then ω = 1/3. Suppose that KX = 1. which implies that we may assume that X = F1 to complete the proof.1. we must show that this assumption leads to a contradiction.3. ωD) is not log canonical at some point P ∈ X. Then −KX ≡ 2C + 3L . ω(2C + 3L)) are log canonical. 2 Lemma 3. Take C ∈  − KX  such that P ∈ C. we see that we must show that lct(X) = ω to prove Theorem 1. 2 Proof.1124 I. 9 .7. It follows from Remark 2. The contradiction obtained completes the proof. and the singularities of the log pair (X. which is a contradiction. The obtained contradiction completes the proof. 6 .1. Let L and C be irreducible curves on X such that L2 = 0 and C 2 = −1.1 and 2. Let Z be a general curve in C + L such that P ∈ Z. The inequality KX Proof.2. 1. ω= 2 3/4 when KX = 2 and  − KX  has a tacnodal curve. There is an eﬀective Qdivisor D on the surface X such that the equivalence D ≡ −KX holds. The inequality KX = 1 holds. = X 2 2/3 when KX = 4 or X is a cubic surface in P3 with an Eckardt point. CHELTSOV GAFA 2 1/3 when X ∼ F1 or KX ∈ 7.7. X 5/6 when K 2 = 1 and  − K  has a cuspidal curve. 2 Lemma 3. Suppose that the inequality lct(X) < ω holds. which is a contradiction. Then C is an irreducible curve. ωC) is log canonical. We may assume that C ⊆ Supp(D) by Remark 2. = 1/2 when X ∼ P1 × P1 or K 2 ∈ 5. . To prove Theorem 1. we may assume that C ⊆ Supp(D) by Remark 2. because L · D = 2.1. X X 2 1 when KX = 1 and  − KX  has no cuspidal curves. Therefore.
Then the log = X pair (S. 18. r. Then there is a birational morphism 5. Suppose that KX = 4.2. we see that ωai 1 for every i = 1. . λD) = {P } by Lemma 2. Let L be a general line on P2 . π(D) + L . Then π(D) ≡ −λKP2 . λD) is not log canonical at P . • Either S ∼ F1 or S ∼ P1 × P1 or S ∼ P2 . There is λ ∈ Q such that 0 < λ < ω = 2/3 and (X. Then X is an intersection of two quadrics 4 . Suppose that KX π : X → S such that 4 holds. Then LCS(X. • The morphism π is an isomorphism in a neighborhood of P . Then r 4 = −KX · D = i=1 ai deg(Ci ) ak deg(Ck ) > 3 deg(Ck ) . Suppose that ak > 1/ω = 3/2.2. But π(D) ≡ −KS . Let Z be an irreducible curve on X such that Ck + Z is cut out by a general hyperplane section of X ⊂ P4 that passes through Ck . which is impossible by Lemma 3. But there is a birational morphism π : X → P2 such that π is an isomorphism in a neighborhood of the point P . where Ci is an irreducible curve on the surface X. 2 Lemma 3. The inequality KX = 4 holds. = = = 1 × P1 whenever K 2 and we may assume that S ∼ P 6. The equality ω = 2/3 holds.3. which is a contradiction. 2008 LOG CANONICAL THRESHOLDS OF DEL PEZZO SURFACES 1125 2 Lemma 3. 2 which implies that deg(Ck ) 2. . The obtained contradiction completes the proof. . and 0 ai ∈ Q. Then r 3 4 − deg(Ck ) = Z · D = i=1 ai (Z · Ci ) ak (Z · Ck ) = 2ak > 3 . . Then π(P ) ∪ L ⊆ LCS P2 . The inequality KX 2 Proof. Therefore.Vol. 2 Proof. and in P r D= i=1 ai Ci ≡ −KX . which is impossible by Lemma 2. ωπ(D)) is not log canonical at π(P ).2.4. .
2 Lemma 3. The inequality KX = 2 holds. where L1 and L2 are irreducible smooth curves such that L1 ·L2 = 2 and L2 = L2 = −1. we may assume that L ⊂ Supp(D).7. Suppose that ψ(P ) ∈ C. Put D = mL1 + Ω. Then either ψ(P ) ∈ C or ψ(P ) ∈ C. ωL) is log canonical (see [P]). CHELTSOV GAFA Let π : U → X be a blow up of the point P .1.6 that ¯ U. the log pair (X. ωD) is log canonical outside of ﬁnitely many points. Then multQ (D) + multP (D) > 2/ω by Corollary 2. Therefore. In particular. There is a double cover ψ : X → P2 such that ψ is branched over a smooth quartic curve C ⊂ P2 . There is a curve L ∈  − KX  that is singular at P . where ¯ Z is the proper transform of the curve Z on the surface U . 1 2 Without loss of generality.5. It follows from Remark 2. Then m + 1 < 2m + Ω · L2 = D · L2 = 1 . because (X. ¯ There is a unique curve Z ∈  − KX  such that P ∈ Z and Q ∈ Z. where Z1 and Z1 2 2 are irreducible smooth curves such that Z1 · Z2 = 2 and Z1 = Z2 = −1. Suppose that Z is irreducible. . we must have L = L1 + L2 . Hence. ω D + ω(multP (D) − 1)E ¯ is not log canonical at some point Q ∈ E. ¯ where D is the proper transform of D on the surface U . We may assume that P ∈ Z1 and P ∈ Z2 . and Ω is an eﬀective Qdivisor such that L1 ⊆ Supp(Ω). we see that ψ(P ) ∈ C. we must have Z = Z1 + Z2 . Then ¯ D ≡ π ∗ (−KX ) − multP (D)E . we have ¯ ¯ ¯ 2 − multP (D) = Z · D multQ (D) > 2/ω − multP (D) which is a contradiction. and E be the exceptional curve of π. So. Then Z consists of at most two components. Then 2=L·D multP (D) multP (L) 2/ω > 2 in the case when L is irreducible. where 0 m ∈ Q. Suppose that KX = 2. So.1126 I. and we may assume that at least one irreducible component of the curve L is not contained in the support of the divisor D by Remark 2. We may assume Z ⊆ Supp(D). 2 Proof. which is a contradiction.
3. . • The curve T is a union of a line and an irreducible conic intersecting at P .Vol.2. which implies that Z2 ⊆ Supp(D).2. 18. Proof. we see that ω = 3/4 and multP (D) > 4/3 by Remark 2.6. Lemma 3. ωD) is even log terminal on X \ P by Lemma 2. ωD) is log canonical on X \ P . We may assume that the support of the divisor D does not contain at least one irreducible component of the curve T .7. which is impossible by Lemma 3. because (S. Let T be the unique hyperplane section of X that is singular at P .3. The cubic surface X does not have Eckardt points. But Q ∈ Z1 . There is a birational morphism π : X → S such that • The morphism π is an isomorphism in a neighborhood of the point P . Thus. The obtained contradiction completes the proof. 3. where Z it the proper transform of Z1 ¯ on the surface U . The log pair (X.3. Suppose that X has an Eckardt point.4. ¯ ¯ ¯ which gives m 1/2. But 1 = Z1 · D multP (D) 1/ω > 1 . ωZ1 + ωZ2 ) is log canonical.) Then π(D) ≡ −KS and (S. ¯ where 0 < m ∈ Q. 2 • The surface S is a smooth del Pezzo surface and KS = 4.1. we may assume that either Z1 ⊆ Supp(D) or Z2 ⊆ Supp(D) by Remark 2. ωT ) is log canonical (see [P]).5. Proof. Therefore.1. Arguing as in the proof of Lemma 3. ωD) contains ﬁnitely many points. Put D = mZ1 +Υ. 3. The following cases are possible: • The curve T is irreducible and U is a del Pezzo surface.5 that X is a smooth cubic surface in P3 . Then the log pair (X. (A point of a cubic surface is an Eckardt point if the cubic contains 3 lines passing through this point. and Υ is an eﬀective Qdivisor on the surface X such ¯ that Z1 ⊆ Supp(Υ). Lemma 3. Then 2−multP (D) 1−multP (D)+2m=Z1 · Υ > 2/ω −multP (D)>2−multP (D) ¯ ¯ ¯ 2m ¯ by Lemma 2.4. ωπ(D)) is not log canonical at the point π(P ).4. 2008 LOG CANONICAL THRESHOLDS OF DEL PEZZO SURFACES 1127 It is easy to see that the log pair (X. Then 2m + Υ · Z2 = D · Z2 = 1 . Then Z1 ⊆ Supp(D). It follows from Lemmas 3. we see that the locus LCS(X. 3.
But multP (D) > 4/3. Let H be the hyperplane section of X that is singular Put Q ˘ Then T = H. ˘ at Q. ¯ and the curve R must be singular at the point Q. We exclude these cases one by one. ¯ be the proper transform of L on the surface U . we see ¯ that Q ∈ T . ˘ ˘ ¯ be the proper transform of H on the surface U .1. Let τ ∈ Aut(U ) be an involution induced by ψ. Suppose that T is irreducible. which implies that multP (D) 1/3. The curve R must be reducible The curves R and H are reducible. Then ¯ ¯ ¯ ¯ 6 − 3 multP (D) = R · D multQ (R) multQ (D) > 2 8/3 − multP (D) . But multP (D) > 4/3. Suppose that ¯ Q ∈ T . Put R = τ (H) ¯ ¯ Let H ¯ and R = π(R). Proof.8. CHELTSOV GAFA • The curve T consists of 3 lines such that one of them does not pass through P . ˘ = π ◦τ (Q). Thus. 3 Suppose that R is irreducible. there is a line L ⊂ X such that ˘ P ∈ L ∋ Q. 8 R are log canonical. ¯ Let T be the proper transform of T on the surface U . (The involution τ induces an involution in Bir(X) that is called the ¯ Geiser involution. So.) It follows from [M] that τ (T ) = E and τ ∗ π ∗ (−KX ) ≡ π ∗ (−2KX ) − 3E . which implies that multP (D) < 2/3. The curve T is reducible. . Put Z = τ (L). Then ¯ ¯ ¯ ¯ 3 − 2 multP (D) = T · D multQ (T ) multQ (D) ¯ > multQ (T ) 8/3 − multP (D) 8/3 − multP (D) . There is a double cover ψ : U → P2 branched over a quartic curve. ¯ ¯ Let L ¯ Then L · E = 0 and ¯ ¯ ¯ L · T = L · π ∗ (−KX ) = 1 . which implies that we may assume that R ⊆ Supp(D) by Remark 2. Lemma 3. Then P ∈ H. because P = Q and T is smooth outside of the point P . where T is reduced and −KU is nef and big. Then ¯ R ≡ π ∗ (−2KX ) − 3E . because otherwise 3 = H · T multP (H) multP (T ) + multQ (H) multQ (T ) 4 . The singularities of the log pair X.1128 I.
Let Υ be the proper transform of Υ on the surface U . there is a line L1 ⊂ X such that P ∈ L1 . τ ∗ (E) ≡ C . that F + Z is cut out by a hyperplane passing through Z. . Then P ∈ F . where C is an irreducible conic that passes through the point P . where ǫ is a positive rational number. 18. ψ is a double cover branched over a quartic curve.Vol. Then τ (E) = C and ¯ ¯ ¯ ¯ τ ∗ (L1 ) ≡ L1 . and Υ is an eﬀective Qdivisor whose support does not contain the conic Z. Then ¯ ¯ 2 − multP (D) + ǫ = Z · Υ > 8/3 − multP (D) by Lemma 2. but the divisor −KU is nef and big. because T = F + Z. Therefore. ¯ Then Z is a conic and P ∈ Z. Suppose that there is no line L2 ⊂ X such that L1 = L2 and P ∈ L2 . There is a line L2 ⊂ X such that L1 = L2 and P ∈ L2 .1. and ρ is the projection from the point P . We have Q ∈ Z. Then 1 = F · D = 2ǫ + F · Υ 2ǫ . ¯ which implies that ǫ 1/2. Proof. which implies that ǫ > 2/3.5. Then T = L1 + C. τ ∗ π ∗ (−KX ) ≡ π ∗ (−2KX ) − 3E − L1 . There is a commutative diagram U π ζ /W ψ X _ _ _ ρ_ _ _/ P2 ¯ where ζ is the contraction of the curve L1 to an ordinary double point. We may assume that F ⊆ Supp(Υ) by Remark 2. we see that Z ⊆ Supp(D). Lemma 3. −KU · C = 1 .9. respectively. Then ¯ ¯ ¯ 2 − multP (D) = Z · D multQ (D) > 8/3 − multP (D) > 2 − multP (D) ¯ ¯ ¯ ¯ in the case when Z ⊆ Supp(D). ¯ Let τ be the biregular involution of U induced by ψ. Let F be a line on X such Put Z = π(Z). Hence. Put D = ǫZ + Υ. But ǫ 1/2. C 2 = −1 . ¯ ¯ Let L1 and C be the proper transforms of L1 and C on the surface U . 2008 LOG CANONICAL THRESHOLDS OF DEL PEZZO SURFACES 1129 ¯ ¯ ¯ which implies that Z · E = 1 and Z · π ∗ (−KX ) = 2. −KU · L1 = 0 . Then ¯1 ¯ ¯ ¯ L2 = −2 .
Then ¯ ¯ mL1 + Ω ≡ π ∗ (−KX ) − m + multP (Ω) E ≡ π ∗ (−KX ) − multP (D)E . ¯ ¯ ¯ Let H be the proper transform of H on the surface U . Then P ∈ H. ¯ where Ω is the proper transform of Ω on the surface U . 8 (R+L1 ) is log canonical. . where m is a positive rational number. which implies that ¯ ¯ multQ (Ω) > 8/3 − multP (Ω) − m 1 + multQ (L1 ) . and Ω is an eﬀective Qdivisor whose support does not contain the line L1 . we see that Q ∈ C.10 that ¯ ¯ 2 − multP (Ω) − 2m = C · Ω > 8/3 − multP (Ω) − m . Put D = mL1 + Ω. because C is smooth. ¯ Let L be the proper transform of the line L on the surface U . Then ¯ ¯ ¯ ¯ ¯ ¯ L · C = L · π ∗ (−KX ) = 1 and L · E = L · L1 = 0 . But m < 2/3. Then multP (D) = multP (Ω) + m. Then it follows from Lemma 2. Let H be the hyperplane section of the cubic surface X that is singular at the point π ◦ τ (Q) ∈ C. 3 Suppose that R is irreducible. Put R = τ (H) ¯ Then and R = π(R). We have ¯ ¯ 0 C · Ω = 2 − multP (Ω) + 2m < 2/3 − m . Then either L1 ⊆ Supp(D) or C ⊆ Supp(D). ¯ ¯ R ≡ π ∗ (−2KX ) − 3E − L1 . Hence.10 gives ¯ ¯ ¯ 5 − 2 m + multP (Ω) − m = R · Ω 2 multQ (Ω) > 2 8/3 − m − multP (Ω) . Then it follows from the inequality 3. Then R+L1 ≡ −2KX . ¯ Suppose that Q ∈ C. Hence.1. CHELTSOV GAFA Note that we assumed earlier that the support of the divisor D does not contain at least one irreducible component of the curve T . we see that Q ∈ L1 . but X. there is a line L ⊂ X such that P ∈ L and π ◦ τ (Q) ∈ L. (3.10) ¯ Suppose that Q ∈ L1 . But ¯ ¯ L1 · D = 1 − multP (D) < 0 . ¯ which implies that m > 5/6.1130 I. which implies that C ⊆ Supp(D) ⊇ L1 . ¯ and the curve R is singular at the point Q by construction. The inequality 3. which implies that m < 0. ¯ We have τ (E) = C. which implies that m < 2/3. ¯ which implies that m < 0.5 that ¯ ¯ 1 − multP (Ω) + m = L1 · Ω > 8/3 − multP (Ω) − m . Hence. which implies that we may assume that R ⊆ Supp(D) by Remark 2.
and ρ is the projection from the point P . . E · L1 = E · L2 = −KU · L3 = 1 . But Z ∩ L1 = ∅. ¯ where Υ is a proper transform of Υ on the surface U . Put D = ǫZ + mL1 + Υ . Thus. Let τ be the biregular involution of the surface U induced by ψ. We may assume that L ⊆ Supp(Υ). ¯ Then Z is a conic that passes through the point P .10) implies that ¯ ¯ 2 − m − multP (Ω) = Z · Ω > 8/3 − m − multP (Ω) . because (X. ¯ where Li is the proper transform of Li on the surface U . 1 2 ¯ ¯ ¯ −KU · L1 = −KU · L2 = E · L3 = 0 . There is a commutative diagram U π ζ /W ψ X _ _ _ ρ_ _ _/ P2 . where ǫ is a positive rational number. We have P ∈ F . 2008 LOG CANONICAL THRESHOLDS OF DEL PEZZO SURFACES 1131 ¯ ¯ ¯ but τ preserves the intersection form. ¯ · L1 = 0.5 that ¯ ¯ 2 − multP (D) + ǫ = Z · Υ > 8/3 − multP (D) . But ǫ 1/2. ψ is a double cover branched over a quartic curve. The Put Z = π(Z). Then ¯ ¯ τ ∗ π ∗ (−KX ) ≡ π ∗ (−2KX ) − 3E − L1 − L2 . ω(L + Z)) is log canonical. Then 1 = L · D = 2ǫ + mL · L1 + L · Υ = 2ǫ + L · Υ 2ǫ . Then Z · E = 1. ¯ ¯ which is impossible. ¯ ¯ Z ¯ ¯ Suppose that the support of Ω does not contain Z. ¯ ¯ which implies that ǫ 1/2. Put Z = τ (L). where L3 is a line such that P ∈ L3 . Then ¯ ¯ ¯ ¯ ¯ L2 = L2 = −2 . 18. Z · π ∗ (−KX ) = 2. We deduce that ǫ > 2/3. ¯3 L2 = −1 . and Υ is an eﬀective Qdivisor on the surface X such that the support of the divisor Υ does not contain the curves Z and L1 . We have T = L1 + L2 + L3 . Then the inequality (3. ¯ ¯ where ζ is the contraction of the curves L1 and L2 to ordinary double points. Then it follows from Lemma 2. the support of Ω must contain the curve Z.Vol. line L is the line on X such that the curve L + Z is cut out by a hyperplane passing through Z.
But ¯ ¯ ¯ ¯ L1 · D = L2 · D = 1 − multP (D) < 0 . Without loss of generality we may ¯ 1 . The latter equality implies that ¯ ¯ multQ (Ω) > 8/3 − multP (Ω) − m1 1 + multQ (L1 ) ¯ − m1 1 + multQ (L2 ) . ¯ ¯ ¯ ¯ Z · E = 1 .12.11) ¯ ¯ Lemma 3. We have m1 > 5/6. but π ◦ τ (Q) ∈ L1 ∪ L2 . because 1 − m1 − m2 = L3 · Ω 0. . Suppose that Q ∈ L1 ∪ L2 . where 0 < mi ∈ Q. and Ω is an eﬀective Qdivisor such that L2 ⊆ Supp(Ω) ⊇ L2 . Then ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ L · L1 = L · L2 = L · E = 0 . Z · L3 = 0 .5. ¯ which implies that the curve π(Z) is a conic passing through the point P . L · π ∗ (−KX ) = L · L3 = 1 . which implies that L2 ⊆ Supp(D) ⊇ L2 and L3 ⊆ Supp(D). Put D = m 1 L1 + m 2 L2 + Ω . Then 1 − m1 + m2 = Ω · L2 > 4/3 − m1 − m2 . Proof. ¯ is the proper transform of the line L on the surface U .13. Then ¯ ¯ ¯ m1 L1 + m2 L2 + Ω ≡ π ∗ (−KX ) − m1 + m2 + multP (Ω) E .3. Therefore. L2 . Lemma 3. ¯ ¯ Proof. where m1 + m2 + multP (Ω) = multP (D). The line L3 is the only line on X that passes through the point π ◦ τ (Q). We assume that T ⊆ Supp(D). The latter contradicts the inequality m1 + m2 1. The curves L1 and L2 do not contain the point Q. ¯ Let Ω be the proper transform of Ω on the surface U . the point π ◦ τ (Q) is contained in the line L3 . Then Supp(D) does not contain one of L1 . Put Z = τ (L) and ¯ Then Z = π(Z). (3. where L ¯ ¯ The involution τ preserves the intersection form. CHELTSOV GAFA ¯ ¯ ¯ ¯ ¯ and τ (L1 ) = L1 . Z · π ∗ (−KX ) = 2 . The inequality m1 + m2 1 holds. Suppose that there is a line L ⊂ X such that L = L3 and π ◦ τ (Q) ∈ L. Then assume that Q ∈ L ¯ ¯ 1 − multP (Ω) − m2 + m1 = L1 · Ω > 8/3 − multP (Ω) − m1 − m2 by Lemma 2. τ (L2 ) = L2 . τ (L3 ) = E. L3 . which implies the inequality m2 > 1/6. Recall that multP (D) > 4/3 by Remark 2.1132 I.
and Υ is an eﬀective Qdivisor whose support does not contain the curves Z. Then L1 + L2 + Z ≡ −2KX and D · L1 = m2 − m1 + 2ǫ + L1 · Υ = D · L2 = m1 − m2 + 2ǫ + L2 · Υ = 1 . Lemma 3. where ǫ is a positive rational number. Then 1 = L · D = 2ǫ + m1 L · L1 + m2 L · L2 + L · Υ = 2ǫ + L · Υ 2ǫ .5.7 that ¯ ¯ ¯ 4 − 2 multP (D) = Z · D multQ (D) > 8/3 − multP (D) . the log pair ¯ ¯ U. 2008 LOG CANONICAL THRESHOLDS OF DEL PEZZO SURFACES 1133 The support of the divisor Ω contains the conic Z. L1 . Let C ⊂ X be a conic such that C + L3 is cut out by the hyperplane ¯ tangent to X at π ◦ τ (Q). where 0 < ǫ ∈ Q. We have Z · E = 2 and Z · L1 = Z · L2 = 0. Let Υ be a proper transform of Υ on the surface U . which implies that multP (D) < 4/3. because otherwise ¯ ¯ 2 − m1 − m2 − multP (Ω) = Z · Ω > 8/3 − m1 − m2 − multP (Ω) . ǫZ + ω Υ + (ω multP (D) − 1)E ¯ is not log canonical at the point Q. Then ¯ ¯ surface U . Then ¯ ¯ 4 − 2 multP (D) = Z · Υ > 8/3 − multP (D) by Lemma 2. Proof.1. and Υ is an eﬀective Qdivisor on X whose support does not contain Z. ¯ ¯ ¯ ¯ ¯ and Z is singular at P . But Q ∈ L1 and Q ∈ L2 by Lemma 3. L1 . L2 . The line L is the line on the surface X such that the curve L + Z is cut out by a hyperplane that passes through the conic Z. ¯ which implies that ǫ 1/2. Then ¯ ¯ ¯ ¯ 2 − multP (D) + ǫ = 2 − multP (D) + ǫ − m1 L1 · Z − m2 L2 · Z ¯ ¯ = Z · Υ > 8/3 − multP (D) by Lemma 2. ω(L + Z)) is log canonical. ¯ ¯ which implies that ǫ 1/2. L2 . 18.14. Put Z ¯ ¯ ¯ Z ≡ π ∗ (−2KX ) − 4E − L1 − L2 .5. But multP (D) > 4/3. . Put D = ǫZ + m1 L1 + m2 L2 + Υ. and let C be the proper transform of C on the ¯ = τ (C) and Z = π(Z). The support of the divisor D contains Z. Put D = ǫZ + m1 L1 + m2 L2 + Υ. because C ∩ L1 = C ∩ L2 = ∅. We may assume that the support of Υ does not contain the line L by Remark 2.12. which is impossible. which implies that ǫ > 2/3. Thus. Then it follows from Corollary 2. Suppose that Z ⊆ Supp(D).Vol. because the log pair (X. which implies that multP (D) < 4/3. where Υ is a proper transform of Υ on the surface U . But multP (D) > 4/3. But ǫ 1/2.
and Z is irreducible. Thus.2 that the locus LCS(X.1134 I. Arguing as in the proof of Lemma 3. but (X. Let Z be the curve in  − KX  such that P ∈ Z. . Thus. ω= 5/6 when  − KX  has cuspidal curves. Let π : U → X be a blow up of the point P . There is an irreducible curve C such that D = mC + Ω where 1 < 1/λ Supp(Ω). to conclude the proof. Then m ∈ Q. We may assume that Z ⊆ Supp(D).7. 2/3 when  − K  has a cuspidal curve C such that X Sing(C) is a point of type A2 . Suppose that LCS(X. and singularities of the surface X consist of ﬁnitely many points of type A1 or A2 . ¯ Z ≡ π ∗ (−KX ) − E . because (X. 4 Singular Surfaces 2 Let X be a del Pezzo surface with Du Val singularities such that KX = 1. Suppose that P is a point of type A1 . λD) is zerodimensional.1. λD) consists of a single point P ∈ X. It follows from Lemma 2. λD) is not log terminal and for some ω > λ ∈ Q. Taking into a consideration curves in −KX .1. Lemma 4. and Ω is an eﬀective Qdivisor such that C ⊆ 1 = H · D = mH · C + H · Ω > m > 1 . Then ¯ D ≡ π ∗ (−KX ) − aE . λD) is not zerodimensional. we see that lct(X) ω. Then there is an eﬀective Qdivisor D on the surface X such that D ≡ −KX . Put 1 when  − KX  does not have cuspidal curves. where H is a general curve in the pencil −KX . we see that we may assume that P ∈ Sing(X). we may assume that lct(X) < ω. Proof. but their cusps are not contained in Sing(S). the locus LCS(X. CHELTSOV GAFA The contradiction obtained completes the proof Theorem 1. The equality lct(X) = ω holds. 3/4 in the remaining cases. ωZ) is log canonical.
Then ¯ 1 2a = E · D > 1/λ > 1 by Lemma 2. ` 2a2 − a1 = E1 · D 1/λ − a1 > 1 − a1 . ` Z ≡ ζ ∗ (−KX ) − E1 − E2 E . because 2a2 a1 2/3. ¯ ¯ Then a 1/2. which is a contradiction. 2008 LOG CANONICAL THRESHOLDS OF DEL PEZZO SURFACES 1135 ¯ ¯ where D and Z are proper transforms of D and Z on the surface U .1.Vol.6. ` ` ` ` The inequalities Z · D 0. which implies that a1 > 1/2 and a2 > 1/2. E1 · D 0. we see that O = E1 ∩ E2 . Then ` 2a1 − a2 = E1 · D 1/λ − a2 > 1 − a2 . ` Suppose that O ∈ E2 . and 0 ai ∈ Q. We have ` 2a1 − a2 = E1 · D > 1/λ > 1 . .5. λD + λaE) is not log terminal at some point Q ∈ E by Remark 2. 2 2 and W is smooth along E1 and E2 . λD + E1 ) is not log terminal at Q.6). respectively. respectively. Without loss of generality. we see that a1 2/3 and a2 2/3. because 1 − 2a = Z · D 0. Thus. λD+λa1 E1 +λa2 E2 ) is not log terminal at the point O (see Remark 2. 1. E is the exceptional curve of π. the morphism ζ induces an isomorphism ∼ W \ (E1 ∪ E2 ) = X \ P . But ` D ≡ ζ ∗ (−KX ) − a1 E1 − a2 E2 .8 follows from Lemma 4. Thus. 2a1 a2 .5. and a is a positive rational number. Then (W. by Lemma 2. 18.5. 2a2 a1 . which implies that a1 > 2/3. respectively. ¯ The log pair (U. But the equivalence ` KW + λD + λa1 E1 + λa2 E2 ≡ ζ ∗ (KX + λD) ` implies the existence of a point O ∈ E1 ∪ E2 such that (W. But by Lemma 2. the point P is a singular point of type A2 . a1 . ` ` where D and Z are proper transforms of D and Z on the surface W . Then E1 = E2 = −2 and E1 · E2 = 1. But a1 +a2 The assertion of Theorem 1. we may assume that O ∈ E1 . There is a birational morphism ζ : W → X such that ζ contracts two irreducible smooth rational curves E1 and E2 to the point P . E1 · D 0 imply that a1 + a2 1. Thus.
8]) that the sequence H 0 X. Then lct(X. OX ((m − r)H) h0 OL ⊗ OX ((m − r)H) = h0 (OL )  Supp(L) k. • let k be the smallest natural number such that k = Σ. Thus. 9. Suppose that L is zerodimensional.2) implies that k > h0 X.1. and Ω is an eﬀective onecycle on the surface X. Suppose that h0 (X. It follows from the Nadel vanishing theorem (see [L. OX ((m − r)H)) < k. and L is the corresponding subscheme. Then C ∼ lH for some natural number l. where 0 < λ ∈ Q such that λ < m/r. where Σ ⊂ X is a Gorbit. where J (λD) is the multiplier ideal sheaf of λD.1 that lct(X. But m > λr µl l m. Lemma 5.1136 I. • let m be the smallest natural number such that there is a Ginvariant divisor in mH. . whose support does not contain any component of the curve C. the subscheme L is not zerodimensional. We have l m. where µ 1. Then the exact sequence (5. We suppose that lct(X. and • let r be the biggest natural number such that −KX ∼ rH. G) = m/r. λD) = ∅ and D ≡ −KX . G) m/r. because the Ginvariant subgroup of the group Pic(X) is generated by the divisor H. It follows from Deﬁnition 1. let G be a ﬁnite subgroup in Aut(X) such that the Ginvariant subgroup of the group Pic(X) is ZH. Proof. let H be a Cartier divisor on X. Th.2) is exact.1. because the subscheme L is Ginvariant. G) < m/r. Let us show how to apply Lemma 5. CHELTSOV GAFA 5 Invariant Thresholds Let X is a smooth del Pezzo surface.4. OX ((m − r)H) −→ H 0 OL ⊗ OX ((m − r)H) −→ 0 (5. Hence. there is a Ginvariant reduced curve C on the surface X such that λD = µC + Ω . Then there is an eﬀective Ginvariant Qdivisor D on the surface X such that LCS(X.
3.Vol. Then lct(X. z.1. Lemma 5. . O18 } . 2008 LOG CANONICAL THRESHOLDS OF DEL PEZZO SURFACES 1137 2 Example 5.1 to deduce the equality lct(X. . where O1 . k = 6. G) = 1 by Example 5. The divisor Lemma 5.1. Suppose that X = P2 and G = A5 such that the subgroup G leaves invariant a smooth conic on P2 . . G) < 4. λD) consists of 18 points. . y. Then 18 18 12 = R · D i=1 multOi (R) multOi (D) = i=1 2 multOi (D) 36 multOi (D) . E6 such that 6 2 and Ei = −1. Then there is an eﬀective Ginvariant Qdivisor D on the cubic surface X such that LCS(X. G) = 2/3 by Lemma 5.1. Suppose that KX = 5 and k = 1. Suppose that X be the cubic surface in P3 that is given by the equation x3 + y 3 + z 3 + t3 = 0 ⊂ P3 ∼ Proj C[x. Suppose that lct(X. which implies that multOi (D) 1/3.4. where 0 < λ ∈ Q such that λ < 4. Then lct(X.1. Then LCS(X. λD) = {O1 . t] . . Then lct(X. . Example 5. We have r = 1 and G ∼ Z3 ⋊ S4 (see [DoI]). Then X has 6 curves E1 .1. because r = 3.5. OX ((m − r)H)) k. Then lct(X. = X Then r = 1 and k > 6. we still may be able to show that lct(X. Even if h0 (X. Arguing as in the proof of Lemma 5. . Let R be a curve on the surface X that is cut out by xyzt = 0. Then it is easy to = 3 check that m = 4 and k = 18. 18. G) = 4.6. because every Gorbit containing at most 20 points must consist of 18 points. and the log pair (X. G) = m/r. because the stabilizer of every point induces a faithful twodimensional linear representation in its tangent space. . R) is log canonical. . m = 2. λD) = ∅ and D ≡ −KX . Then R is Ginvariant. = and G = Aut(X). We may assume that R ⊆ Supp(D) by Remark 2. . Suppose that K 2 = 6 and G = Aut(X) ∼ S5 (see [RS]). . we see that the locus LCS(X. which implies that we are unable to apply Lemma 5. because m = 2 (see [RS]). O18 are all Eckardt points of the surface X (see [DoI]). . Ei ∼ −KX i=1 6 i=1 Ei is Ginvariant. Proof. G) = 2 by Lemma 5. . G) = 4.
i=1 10 Let P be the pencil generated by Z and i=1 Li . . . y. . i i which implies that Σ ∩ Ei  = 1. P3 = (−1 : −1 : 1) . The surface X is embedded in P5 by the linear system −KX . L10 . . G) = 2. The surface X can be obtained as a blow up π : X → P2 of the four points P1 = (1 : −1 : −1) . which we denote as L1 . which can be described in the following way: • the curve 10 Li . Then r = 1 and Aut(X) ∼ S5 (see [RS]). = 10 The divisor i=1 Li ∼ −2KX is S5 invariant. because the representation induced by the action of the stabilizer of Oi on its tangent space is irreducible. and there are exactly 5 singular curves in P. and D is the proper transform of D on the surface U . 2 Lemma 5. . . . Then 18 ¯ KU + 4D + i=1 4 multOi (D) − 1 Ei ≡ π ∗ (KX + 4D) . P2 = (−1 : 1 : −1) . CHELTSOV GAFA Let π : U → X be a blow up of the points O1 . i=1 • two irreducible rational curves R1 and R2 that have 6 nodes. • two ﬁbers F1 and F2 each consisting of 5 smooth rational curves. We have ¯ multO (D) = Ei · D > Σ ∩ Ei  1/2 − multO (D) . . ¯ where Ei is the πexceptional curve such that π(Ei ) = Oi .7. Then there is Qi ∈ Ei such that ¯ multQi (D) > 1/2 − multOi (D) for i = 1. of the plane P2 . and X contains 10 lines. We have m = 2 and k = 6 by [RS]. Let W be the curve in P2 that is given by the equation x6 +y 6 +z 6 +(x2 +y 2 +z 2 )(x4 +y 4 +z 4 ) = 12x2 y 2 z 2 ⊂ P2 ∼ Proj C[x. Proof. = and Z be its proper transform on X. The curves Z and 10 Li are the only S5 invariant curves in  − 2KX . The smallest Gorbit are Sing(R1 ) and Sing(R2 ) (see [IK]).1138 I. It follows from [E] that P is A5 invariant. which implies that lct(X. Then Z is S5 invariant (see [IK]) and Z ∼ −2KX . Suppose that KX = 5 and G = A5 . P4 = (1 : 1 : 1) . . Then Σ ∩ Ei = Qi . . G) 2. . Let Σ be the Gorbit of the point Qi . . 18. because multOi (D) 1/3. z] . . O18 . Then lct(X.
. R2 ) are log canonical. 6. . Without loss of generality we may assume that the locus LCS(X. . Lemma 5. .Vol. . . because the orbit of length 2 of the action on Ei of the stabilizer of Oi is unique. We may assume that the support of D does not contain R1 and R2 due to Remark 2. where i = 1. λD) = Sing(R2 ). because both log pairs (X. we see that either LCS(X. We have Σ ∩ Ei  = 2.1. We have 6 6 12 1 − multOi (D) = 10 − 2 i=1 ¯ ¯ multOi (D) = R1 · D 2 i=1 ¯ multQi (D) > 12 1 − multOi (D) . . where 0 < λ ∈ Q such that λ < 2. because the stabilizer of Oi acts faithfully on its tangent space. Now arguing as in the proof of Lemma 5. i i ¯ Let R1 be the proper transform of the curve R1 on the surface U . Denote them as O1 . Then ¯ Σ = R1 (E1 ∪ E2 ∪ E3 ∪ E4 ∪ E5 ∪ E5 ) . which is a contradiction. G) < 2. O6 . . 2008 LOG CANONICAL THRESHOLDS OF DEL PEZZO SURFACES 1139 Suppose that lct(X. λD) consists of the singular points of the curve R1 . because 6 10 = R1 · D i=1 multOi (D) multOi (R1 ) 12 multOi (D) . G) = 4/5 .8. ¯ where Ei is the πexceptional curve such that π(Ei ) = Oi . . λD) = ∅ and D ≡ −KX . 18. Then lct(X. Then multQi (D) > 1 − multOi (D) for some point Qi ∈ Ei . . holds. 2 Suppose that KX = 5 and G = Z5 . because multOi (D) 5/6 and ¯ multO (D) = Ei · D > Σ ∩ Ei  1 − multO (D) .1. Then Σ ∩ Ei  2. R1 ) and (X. Then multOi (D) 5/6. λD) = Sing(R1 ) or LCS(X. . Then 6 ¯ KU + 2D + i=1 2 multOi (D) − 1 Ei ≡ π ∗ (KX + 2D) . and D is the ¯ proper transform of D on the surface U . . Then there is an eﬀective Ginvariant Qdivisor D on the quintic surface X such that LCS(X. O6 . Let π : U → X be a blow up of the points O1 . Let Σ be the Gorbit of the point Qi .
. The point Q must be Ginvariant. Without loss of generality. Z5 . . ¯ which implies that multO1 (D) + multQ (D) 3.1. Then ¯ ¯ 3 − multO1 (D) Ti · D multQ (D) . G) 4/5.1. . G) < 4/5. Then . which is impossible. Thus. λD) = ∅ and D ≡ −KX . There are ﬁve conics Z1 . . The proof of Lemma 5. which we denote as O1 and O2 . and 5 π(Ti ) ≡ 3D. we may assume that LCS(X. Then ¯ multQ (D) 2/λ − multO1 (D) > 5/2 − multO1 (D) ¯ for some point Q ∈ E by Corollary 2. Z5 by Remark 2.λD) = {O1 } or LCS(X. λD) = {O1 }. CHELTSOV GAFA Proof. . and F be the ξexceptional divisor. T5 due to Remark 2. where 0 < λ ∈ Q such that λ < 4/5. ¯ ¯ The curve φ(E) is a conic that contains φ(Z1 ). . ¯ Let Zi be the proper transform of the conic Zi on the surface U .λD) = {O1 }. and the divisor 5 Zi ∼ −2KX i=1 is Ginvariant. Then 5 Z . and E be the πexceptional curve.1140 I. Let ξ : V → U be a blow up of the point Q. . . .7. and we may assume that the support of the divisor D does not contain the conics Z1 . . . we may asi=1 ¯ sume that the support of the divisor D does not contain any of the curves T1 . · · · . It is well known that the group G ﬁxes exactly two points of the surfaces X (see [RS]). because otherwise ¯ multO1 (D) = E · D > 5 5/2 − multO1 (D) . Then 2 = Z1 · D multO1 (D) . . The log pair X. Let Ti be the 2 that passes through the proper transform on the surface U of the line in P ¯ points φ(Q) and φ(Zi ). . which implies that lct(X. Then there is an eﬀective Ginvariant Qdivisor D on the quintic surface X such that LCS(X. φ(Z5 ). . because multO1 (D) 2. .1 implies that LCS(X. . Z5 ⊂ X that passes through O1 . Let π : U → X be a blow up of the point O1 . λ 3 5 π(Ti ) i=1 has log terminal singularities. and there is a birational morphism φ : U → P2 that contracts Q ∈ ∪i=1 ¯i ¯ ¯ the curves Z1 . Suppose that lct(X. where D is the proper transform of D on the surface U .
Then ¯ ` ` ¯ 3 − multO (D) − multQ (D) = Tk · D > 15/4 − multO (D) − multQ (D) . because multO1 (D) 2.1. Hence. λD + (λ multO1 (D) + λ multQ (D) − 2)F ) is not log terminal at P .6.7 and 5. because multQ (D) 2/λ − multO1 (D). we see that P ∈ Let M be an irreducible curve on V such that P ∈ M . 18. the curve φ◦ξ(M ) is a line that passes through the point φ(Q). Let T be the proper transform on V of the line 2 that is tangent to the conic φ(E) at the point φ(Q). ` ` Suppose that P ∈ E. The log pair ` ` ¯ W.1. Thus. 5. ` ¯ The log pair (W. Then π ◦ξ(M ) has an ordinary ` double point at O1 . and π ◦ ξ(M ) ≡ −KX . 1 1 which is a contradiction. 2008 LOG CANONICAL THRESHOLDS OF DEL PEZZO SURFACES 1141 ` ` ¯ KW + λD + λ multO1 (D) − 1 E + λ multO1 (D) + λ multQ (D) − 2 F ≡ (π ◦ ξ)∗ (KX + λD) . i=1 which implies that multO1 (D) 5/4. respectively. But the latter is well known (see [DoI]). We may i=1 ` by Remark 2. ` ` because we may assume that T ⊆ Supp(D) by Remark 2. ` Let Ti be the proper transform of Ti on the surface V . ` ` where D and E are proper transforms of D and E on the surface V . we have ` P ∈ E. Then P ∈ T . Then assume that M ⊆ Supp(D) ¯ ` ¯ 5 − 2 multO (D) − multQ (D) = M · D > 15/4 − multO (D) − multQ (D) . But ` ¯ λD + λ multO1 (D) + λ multQ (D) − 2 F ¯ is an eﬀective divisor. Then ` multP (D) ¯ ¯ 3/λ−multO1 (D)−multQ (D) > 15/4−multO1 (D)−multQ (D) .6.Vol. 1 1 ` ∪5 Ti . We did not prove that groups in Example 5. λD + (λ multO1 (D) − 1)E + (λ multO1 (D) + λ multQ (D) − 2)F is not log terminal at some point P ∈ F by Remark 2.8 act on X in such a way that the Ginvariant subgroup in Pic(X) is Z. . ` on P which implies that ¯ ` ` ` 5 − 2 multO (D) − multQ (D) = T · D multP (D) 1 ¯ > 5 − 2 multO1 (D) − multQ (D) . Suppose that ` P ∈ Tk . But multO1 (D) > 5/4.5 and Lemmata 5. because P ∈ ∪5 Ti .
Lemma 6. Suppose that the surface X is not Gbirationally superrigid. Proof. Then multP (M) > 1/λ for every point P ∈ Σ. and the proof of Lemma 5. Let X be a smooth del Pezzo surface such that KX = 5. ∼ S5 . • every G1 × · · · × Gr equivariant birational automorphism of X1 × · · · × Xr is biregular. .7 implies that the surface Then Aut(X) = X is A5 birationally superrigid by Lemma 6. . where λ ∈ Q such that λ > 0 and KX + λM ≡ 0. The variety X is said to be Gbirationally superrigid if for every Ginvariant linear system M on the variety X that does not have any ﬁxed components. . Then 2 KX /λ2 = M1 · M2 P ∈Σ mult2 (M) > Σ/λ2 P 2 KX /λ2 . Let Xi be a smooth Gi birationally superrigid Fano variety. where M1 and M2 are suﬃciently general curves in M.4. r. The following result is well known (see [M]. . Then there is a Ginvariant linear system M on the surface X such that M does not have ﬁxed curves. Let Σ be the Gorbit of the point O. λM) is not canonical at some point O ∈ X. . Gi ) 1 for every i = 1. and i = 1.2. • for any G1 ×· · ·×Gr equivariant dominant map ρ : X1 × · · · ×Xr Y. where λ ∈ Q such that λ > 0 and KX + λM ≡ 0.1. Suppose that X is a smooth del Pezzo surface such that Σ 2 KX for any Gorbit Σ ⊂ X. Then • there is no G1 × · · · × Gr equivariant birational map ρ : X1 × · · · × Xr Pn . r. λM) are canonical. definition 6. .2. [DoI]).3. there a commutative . whose general ﬁber is rationally connected.1142 I. . . Suppose that lct(Xi . CHELTSOV GAFA 6 Direct Products Let X be an arbitrary smooth Fano variety. 2 Example 6. the singularities of the log pair (X. Theorem 6. but (X. where Gi is a an arbitrary ﬁnite subgroup of Aut(Xi ) such that the Gi invariant subgroup of Pic(Xi ) is Z. and let G be a ﬁnite subgroup in Aut(X) such that the Ginvariant subgroup of the group Pic(X) is Z. Then X is Gbirationally superrigid.
Shramov. Hwang. J. with appendix by J. S. There is an induced embedding A6 × A6 ∼ Ω ⊂ Bir(P4 ) such that Ω is not conjugated = to a subgroup of Aut(P4 ) by Lemma 6.P. References [C1] [C2] [CS] I. A6 ) = 2 by Lemma 5.2 and Theorem 6. Edge. . W. ik } ⊆ {1. Compositio Mathematica 143 (2007). Cheltsov. Log canonical thresholds on hypersurfaces. .P. Demailly. . Fano varieties with many selfmaps. Iskovskikh.M. Finite subgroups of the plane Cremona group. Mathematical Proceedings of the Cambridge Philosophical Society 89 (1981). Solving the sextic by iteration: a study in complex geometry and dynamics. Cheltsov. z] = and there is an embedding A6 ⊂ Aut(P2 ) such that lct(P2 . J. 73–180.1 (see [Cr]). . Cheltsov. y.5. Proof. Mathematische Annalen 334 (2006). [Cr] [DK] [DoI] [E] [H1] [H2] . Example 6. Log canonical thresholds of divisors on Grassmannians. Semicontinuity of complex singularity exponents and K¨hler–Einstein metrics on Fano orbifolds. Crass. 413–421.AG/0610595 (2006). Annales Sciena ´ tiﬁques de l’Ecole Normale Sup´rieure 34 (2001). . e I. Dolgachev. 1241–1257.M. and A6 × A6 acts naturally on P2 × P2 . Demailly. V. I. . arXiv:math. 413–418.Vol. Kollar. 525–556.4. 18. Log canonical thresholds of divisors on Fano manifolds of Picard rank 1. I. The required assertion follows from the proof of Theorem 1 in [Pu]. 97–124. and {i1 . 2008 LOG CANONICAL THRESHOLDS OF DEL PEZZO SURFACES 1143 diagram X1 × · · · × Xr W W W W W ρ W W W W W W W W _ _ _ _ _ _ _ _ _ _ _ _+/ Y Xi1 × · · · × Xik ξ π where ξ is a birational map. Experimental Mathematics 8 (1999). A pencil of fournodal plane sextics. ´ J. J. r}. S. Loc canonical thresholds of smooth Fano threefolds. . Hwang. 209–240. π is a natural projection. 89–94. . Russian Math. Surveys 63:5 (2008). Sbornik: Mathematics 192 (2001). Advances in Mathematics 217 (2008). The simple group A6 is a group of automorphisms of the sextic 10x3 y 3 + 9zx5 + 9zy 5 + 27z 6 = 45x2 y 2 z 2 + 135xyz 4 ⊂ P2 ∼ Proj C[x.
686–696. School of Mathematics. Lazarsfeld. Annals of Mathematics 132 (1990). Positivity in Algebraic Geometry II. [IK] Ivan Cheltsov. Mathematics of the USSR.ac. Threedimensional log perestroikas. Journal fur die Reine [P] und Angewandte Mathematik 538 (2001). Journal of Mathematics of Kyoto University 45 (2005). J. J. 549–596. Birational geometry of Fano direct products. Rational surfaces over perfect ﬁelds. ´ [K et al.] J. On a set of polarized K¨hler metrics on algebraic manifolds. [T1] G. ´ [K] J. [S] V. Manin. Berlin. Kollar et al. Shokurov. I. Sbornik 1 (1967). e [L] R. Tian. Pukhlikov. On Calabi’s conjecture for complex surfaces with positive ﬁrst Chern class. 221–287.uk Received: May 2007 Revision: October 2007 Accepted: April 2008 . Tian. 105–203. Nadel. 1225–1255. [T2] G. Kato.. Izvestiya: Mathematics 69 (2005). Proceedings of Symposia in Pure Mathematics 62 (1997). An aspect of icosahedral symmetry. CHELTSOV GAFA N. Inventiones Mathematicae 101 (1990). [Pu] A. Singularities of pairs. Edinburgh EH9 3JZ. Inoue. Kollar. Izvestiya: Mathematics 56:1 (1992). [T3] G. 225–246. P. F. Slodowy. Multiplier ideal sheaves and K¨hler–Einstein metrics of positive a scalar curvature. 141–168. UK. 213–221. [N] A.Cheltsov@ed. Birational maps of del Pezzo ﬁbrations. 99–130. Inventiones Mathematicae 89 (1987). Tian. On the geometry of Wiman’s sextic. [M] Yu. II. 2004. a Journal of Diﬀerential Geometry 32 (1990). 743–757. University of Edinburgh. SpringerVerlag.1144 I. Rauschning. Ast´risque 211 (1992). Cana[RS] dian Mathematical Bulletin 45 (2005). On K¨hler–Einstein metrics on certain K¨hler manifolds with a a c1 (M ) > 0. Flips and abundance for algebraic threefolds. Park. 101–172.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue listening from where you left off, or restart the preview.