Vol. 27 (2010), 35–80
35
The gonality conjecture for curves
on toric surfaces with two P
1
ﬁbrations
Ryo Kawaguchi
(Received March 25, 2010; Revised December 3, 2010)
Abstract
The gonality is an important invariant for the study of linear systems on a
curve. Although it is not so easy to determine the gonality of a given curve, the
gonality conjecture posed by Green and Lazarsfeld predicts that the gonality could
be read oﬀ from the minimal resolution of any one line bundle of suﬃciently large
degree. In this paper, we consider this conjecture for a curve embedded in a toric
surface which has two P
1
ﬁbrations by toric morphisms, and prove it aﬃrmatively
under several conditions.
1. Introduction
As mentioned in Abstract, the aim of this paper is to show the gonality con
jecture (Conjecture 1.2) under certain conditions. First of all, in this section, we
would like to roughly review the background and preliminary results for gonality
and the gonality conjecture. Our main theorem will be stated at the end of this
section.
In this paper, a curve will always mean a nonsingular irreducible complex
projective curve unless otherwise stated. We denote by g
1
k
a 1dimensional linear
system of degree k on a curve. For a curve C, the gonality is deﬁned to be the
minimal degree of surjective morphisms from C to P
1
:
gon(C) = min{degf  f : C → P
1
surjective morphism} = min{k  C has g
1
k
}.
It is known that the gonality of a nonsingular plane curve of degree d (≥ 2) is
equal to d − 1 ([11]). Coppens and Kato generalized this result to the case of
singular plane curves in [2]. They computed the gonality of the normalization of
a plane curve with double points under several numerical conditions on its degree
and the number of singular points. Recently, more general cases were investigated
by Ohkouchi and Sakai ([13]). Besides, in [9], Martens determined the gonality
of a nonsingular curve lying on a Hirzebruch surface.
On the other hand, there exists a close relation between the theory of syzy
gies and geometric properties of a projective variety. In particular, the gonality
2000 Mathematics Subject Classiﬁcation. Primary 14H51; Secondary 14M25.
Key words and phrases. The gonality conjecture, toric surfaces
36 R. Kawaguchi
conjecture (Conjecture 1.2) suggested by Green and Lazarsfeld in [7] predicted
the interaction between the gonality of a curve and the syzygies deﬁned by a
line bundle on the curve. For a projective variety X, a line bundle L on X and
a coherent sheaf F on X, we denote by K
p,q
(X, F, L) the Koszul cohomology,
which is introduced by Green in [5] as the cohomology of the Koszul complex
p+1
_
H
0
(X, L) ⊗H
0
(X, F ⊗(q −1)L) →
p
_
H
0
(X, L) ⊗H
0
(X, F ⊗qL)
→
p−1
_
H
0
(X, L) ⊗H
0
(X, F ⊗(q + 1)L).
For simplicity, if F = O
X
, we suppress it and write K
p,q
(X, L). It is known
that there exists the following relation between the Koszul cohomology and the
syzygies :
Theorem 1.1 ([5]). Let X and L be as above. Denote by S the symmetric alge
bra of H
0
(X, L), and consider the minimal free resolution of a graded Smodule
as
· · · →
q≥q
2
S(−q) ⊗M
2,q
→
q≥q
1
S(−q) ⊗M
1,q
→
q≥q
0
S(−q) ⊗M
0,q
→
∞
q=0
H
0
(X, qL) → 0.
Then K
p,q
(X, L) is isomorphic to M
p,p+q
as a complex vector space.
Conjecture 1.2 ([7]). Let C be a curve of genus g and L a line bundle on C
of suﬃciently large degree compared to 2g. Then K
p,1
(C, L) = 0 for any integer
p ≥ h
0
(C, L) −gon(C).
This conjecture means that we can read oﬀ the gonality of a curve from the
minimal resolution of any one line bundle of suﬃciently large degree. For the
cases where gon(C) = 1, 2, Green has shown this conjecture aﬃrmatively in [5].
The case where gon(C) = 3 has been done by Ehbauer ([4]). As for curves on a
Hirzebruch surface Σ
a
, Aprodu showed it, and computed their gonality indepen
dently of Martens’ result ([1]). In his proof, the following Theorem1.4 played an
essential role.
Deﬁnition 1.3. Let L be a line bundle on a curve C, and q a nonnegative inte
ger. We say that L satisﬁes the property (M
q
) if K
p,1
(C, L) = 0 for any integer
p ≥ h
0
(C, L) −q −1.
Theorem 1.4 ([1]). Let C be an irrational curve which has a g
1
k
. If there is a
nonspecial and globally generated line bundle L on C with the property (M
k−1
),
then C is kgonal, and the gonality conjecture is true for C.
THE GONALITY CONJECTURE FOR CURVES ON TORIC SURFACES WITH TWO P
1
FIBRATIONS 37
Recall that a toric surface has a ﬁnite number of P
1
ﬁbrations by toric mor
phisms, and Hirzebruch surfaces are the simplest examples of toric surfaces which
have one or two such P
1
ﬁbrations. In the wake of Aprodu’s result, we proved the
gonality conjecture for curves lying on a toric surface with a unique P
1
ﬁbration
([8]). This is an extension of Aprodu’s result for Σ
a
with a ≥ 1. Therefore,
it is a valid question to ask whether we can generalize Aprodu’s result for the
case a = 0, that is, whether the gonality conjecture is true for the cruves on a
toric surface with two P
1
ﬁbrations. In this paper, we shall show the following
theorem:
Theorem 1.5 (Main theorem). Let S be a compact nonsingular toric sur
face which has two P
1
ﬁbrations f and f
by toric morphisms, and C an irra
tional curve on S. If 2 ≤ degf
C
< degf

C
and C − F is base point free
(where F denotes the ﬁber of f), then the gonality conjecture is true for C and
gon(C) = degf
C
.
In Section 2, we introduce toric surfaces which are the main stage of our
study, and collect basic facts about them. In Section 3, we will see the main idea
to prove the main theorem. Although a key point of the proof is the existence
of a certain divisor called an auxiliary divisor, we do not mention its concrete
construction at all. Instead, we spend Section4 and 5 to construct an auxiliary
divisor. In these two sections, we will treat essentially diﬀerent cases. We close
this section with some basics of Koszul cohomology, which are essential for our
study.
Theorem 1.6 ([1]). Let X be a nonsingular projective variety, L a line bundle
on X and Y ∈ L an irreducible divisor on X. If the irregularity of X is zero,
then K
p,1
(X, L) K
p,1
(Y, L
Y
) for any integer p.
Theorem 1.7 ([5]). Let L be a line bundle on a curve C and put m = dimϕ
L
(C).
Then L satisﬁes (M
m−1
).
Theorem 1.8 ([1]). Let C be an irrational curve, L a nonspecial and globally
generated line bundle on C, and q a nonnegative integer such that L satisﬁes
(M
q
). Then, for any eﬀective divisor D on C, also L + D satisﬁes (M
q
).
2. Summary of toric surfaces
In this section, we brieﬂy review the theory of toric surfaces. It has the close
connection with the geometry of convex polytopes in the real aﬃne space. Many
basic properties of toric varieties and divisors on them can be interpreted in terms
of the elementary convex geometry.
We will henceforth assume that a surface is always compact and nonsingular.
38 R. Kawaguchi
At ﬁrst, it is a basic fact that a toric surface is obtained from the projective plane
or a Hirzebruch surface by a ﬁnite succession of blowingups with Tﬁxed points
(i.e. points which are invariant with respect to the action on a toric surface by the
algebraic torus T) as centers. A composite of such blowingups is called a toric
morphism. For a toric surface (except for the projective plane), a P
1
ﬁbration
obtained by composing a toric morphism and a ruling of the Hirzebruch surface
is called a toric ruling. Naturally, a toric surface has a ﬁnite number of toric
rulings.
In the present paper, we consider a toric surface S with two toric rulings. In
this case, the fan Δ
S
associated to S has two lines passing through the origin.
For each halfline (which is called a cone) in Δ
S
, a lattice point on it is called a
primitive element if it is closest to the origin. We denote by D
1
the Tinvariant
divisor whose associated cone has a primitive element (0, 1) and number other
cones in Δ
S
clockwise. In particular, we set D
d
0
if its associated cone has a
primitive element (0, −1) (see Fig. 1). We denote by σ(D
i
) the cone associated
σ(D
1
)
σ(D
2
)
σ(D
3
)
σ(D
d
)
σ(D
d
0
)
Figure 1
to Tinvariant divisor D
i
and by (x
i
, y
i
) the primitive element of σ(D
i
). We
set D
0
= D
d
and D
d+1
= D
1
formally. Then the selfintersection numbers of
Tinvariant divisors are computed by the following formula.
Theorem 2.1 ([12]). For any integer i with 1 ≤ i ≤ d, two equalities x
i
D
2
i
=
−x
i−1
−x
i+1
and y
i
D
2
i
= −y
i−1
−y
i+1
hold.
The Picard group of S is generated (not freely) by the classes of D
1
, . . . , D
d
.
For instance, the canonical divisor of S has the relation K
S
∼ −
d
i=1
D
i
. Be
sides, the ﬁbers of two toric rulings can be written as
F
1
∼
d
0
−1
i=2
x
i
D
i
∼ −
d
i=d
0
+1
x
i
D
i
F
2
∼ D
d
0
∼
d
0
−2
i=1
y
i
D
i
+
d
i=d
0
+2
y
i
D
i
.
THE GONALITY CONJECTURE FOR CURVES ON TORIC SURFACES WITH TWO P
1
FIBRATIONS 39
We denote by f
1
(resp. f
2
) the toric ruling of S whose ﬁber is F
1
(resp. F
2
).
Next we collect several basic properties of divisors on a toric surface. In the
remaining part of this section, let D be a divisor on S. In the case where the
complete linear system D is base point free, we have the following two results.
Theorem 2.2 ([12]). If D is base point free, then h
i
(S, D) = 0 for any positive
integer i.
Theorem 2.3 ([10]). The complete linear system D is base point free if and
only if D has a nonnegative intersection number with every Tinvariant divisor
on S.
Let
D
be a lattice polytope associated to D =
d
i=1
n
i
D
i
, that is, we deﬁne
D
= {(z, w) ∈ R
2
 x
i
z + y
i
w ≤ n
i
for 1 ≤ i ≤ d }.
Although this deﬁnition depends on the description of the linear equivalence class
of D, diﬀerences of the description cause only parallel translations of
D
. Hence
the shape of the lattice polytope is determined uniquely. We write the lines which
form the boundaries of
D
as
l
i
(D) = {(z, w) ∈ R
2
 x
i
z + y
i
w = n
i
}.
We can read oﬀ the dimension of cohomology group of D from the number of
lattice points contained in
D
:
Theorem 2.4 ([12]). The equation h
0
(S, D) = (
D
∩ Z
2
) holds. In particular,
for a curve C on S, its genus is equal to the number of the lattice points contained
in the interior of
C
.
Proposition 2.5. Let C be a curve on S and assume that C.F
1
, C.F
2
≥ 2. If
either C −F
1
 or C −F
2
 is base point free, then C is irrational.
Proof. We prove the case where BsC − F
1
 = ∅. By Theorem2.4, it
is suﬃcient to verify that there exists at least one lattice point in the inte
rior of
C
. We ﬁrst note that our assumption and Theorem2.3 imply that
(C − F
1
).D
1
= C.D
1
− 1 ≥ 0. Considering the construction of
C
, in the
case where C.D
1
≥ 2, we see that the lattice point l
1
(C) ∩ l
2
(C) + (−1, −1)
is contained in the interior of
C
. In the case where C.D
1
= 1, we put
a = min{i ≥ 2  C.D
i
≥ 1} and b = max{i ≤ d  C.D
i
≥ 1}. The inequal
ity
C.F
2
= 1 +
d
0
−2
i=2
y
i
C.D
i
+
d
i=d
0
+2
y
i
C.D
i
≥ 2
implies that either a ≤ d
0
−2 or b ≥ d
0
+ 2 holds. We see that the lattice point
40 R. Kawaguchi
l
1
(C) ∩l
2
(C) +(0, −1) (resp. l
1
(C) ∩l
d
(C) +(0, −1)) is contained in the interior
of
C
if a ≤ d
0
−2 (resp. b ≥ d
0
+ 2).
3. Outline of the proof
In order to prove Theorem1.5, we need to introduce a certain divisor de
ﬁned with respect to a given curve, which is called an auxiliary divisor. Its
construction is, however, very complicated. Hence, in this section, we admit the
existence of auxiliary divisors satisfying Assertion 3.1–3.3 below, and will prove
Theorem1.5. We postpone their precise construction to Section 4 (the case where
degf
1

C
≤ degf
2

C
) and Section5 (the case where degf
1

C
≥ degf
2

C
).
Let S be a toric surface as in Section2, and C a curve on S. We denote by
F and F
the ﬁbers of two toric rulings of S, and put k = C.F and k
= C.F
.
Assertion 3.1. Assume that k
≥ k ≥ 2 and C − F is base point free. Then
there exists an eﬀective divisor I
C
satisfying the following properties (i)–(vi) :
(i) I
C
.F = 1,
(ii) H
1
(S, −I
C
) = 0,
(iii) The complete linear systems C −I
C
 and C −I
C
−F are base point free,
(iv) (C −I
C
)
2
≥ 1,
(v) (C −I
C
)
C
is nonspecial,
(vi) K
p,1
(S, C −I
C
) = 0 holds for any integer p ≥ h
0
(S, C −I
C
) −2.
We call I
C
the auxiliary divisor of C. By the properties (iii) and (iv), we can
take a nonsingular irreducible curve C
1
∈ C −I
C
. If C
j
.F
≥ C
j
.F ≥ 2, we can
take inductively
I
j
: the auxiliary divisor of C
j
,
C
j+1
: a nonsingular irreducible curve in C
j
−I
j

for a positive integer j. We put C
0
= C and I
0
= I
C
. In this section, we admit
the following two assertions.
Assertion 3.2. Assume that k
> k ≥ 2 and C − F is base point free. There
exists a positive integer m
0
≤ k such that C
j
.F
= C
j
.F for j = 1, . . . , m
0
− 1,
and moreover, the inequality C
m
0
.F
> C
m
0
.F holds if m
0
≤ k −1.
Assertion 3.3. Assume that k
> k ≥ 2 and C − F is base point free, and let
m
0
be a positive integer in Assertion 3.2. Then the inequality
h
0
(S, C
j
) −h
0
(S, C
j
−I
j−1
) ≥ C
j
.F + 2
THE GONALITY CONJECTURE FOR CURVES ON TORIC SURFACES WITH TWO P
1
FIBRATIONS 41
holds for any integer j with 1 ≤ j ≤ min{m
0
, k −1}. Besides, if k
= k = 2, then
this inequality holds for j = 1.
We have the following lemma, which can be shown by the same argument as
in the proof of Lemma 2.3 in [8].
Lemma 3.4. The isomorphism K
p,1
(S, C −I
C
) K
p,1
(C, (C −I
C
)
C
) holds for
any integer p ≥ h
0
(S, C −2I
C
) + 1.
Theorem1.5 follows immediately from the following proposition and Theo
rem1.4.
Proposition 3.5. Assume that C is irrational, k
> k ≥ 2 and C − F is base
point free. If Assertion 3.1–3.3 hold, then O
C
(C) satisﬁes the property (M
k−1
).
Proof. We ﬁrst show the case where k = 2. By Assertion 3.1 and
Lemma 3.4, we see that K
p,1
(C, C
1

C
) = 0 for any integer p ≥ max{h
0
(S, C
1
) −
2, h
0
(S, C
1
−I
0
)+1}. Hence, by combining this fact with Assertion 3.3, we see that
K
p,1
(C, C
1

C
) = 0 for any integer p ≥ h
0
(S, C
1
) −2. The short exact sequence of
sheaves 0 → O
S
(−I
0
) → O
S
(C
1
) → O
C
(C
1
) → 0 induces the cohomology long
exact sequence
0 → H
0
(S, −I
0
) → H
0
(S, C
1
) → H
0
(C, C
1

C
) → H
1
(S, −I
0
) → · · · .
Since H
0
(S, −I
0
) = H
1
(S, −I
0
) = 0, we have h
0
(S, C
1
) = h
0
(C, C
1

C
). In sum,
we can conclude that K
p,1
(C, C
1

C
) = 0 for any integer p ≥ h
0
(C, C
1

C
) −2, that
is, O
C
(C
1
) satisﬁes (M
1
). Note that O
C
(C
1
) is nonspecial and globally gener
ated by Assertion 3.1. Hence Theorem1.8 implies that also O
C
(C) satisﬁes (M
1
).
Here we note that in fact this lemma holds in the case where k
= k = 2 also.
Indeed, for such case, we can develop the same argument as above by using the
latter part of Assertion 3.3.
We next consider the case where k ≥ 3 under the assumption that our propo
sition is valid for a curve C
on S if C
.F ≤ k − 1, C
.F
> C
.F ≥ 2 and
BsC
− F = ∅. For C, let m
0
be a positive integer as in Assertion 3.2. If
m
0
≤ k −2, then 2 ≤ C
m
0
.F ≤ k −1. Since C
m
0
−F is base point free by As
sertion 3.1, we see that O
C
m
0
(C
m
0
) satisﬁes (M
C
m
0
.F−1
) by the hypothesis of our
induction. Consider the case where m
0
≥ k −1. We have C
k−2
.F
= C
k−2
.F = 2
by the property of m
0
. Since C
k−2
− F is base point free by Assertion 3.1, we
see that O
C
k−2
(C
k−2
) satisﬁes (M
1
). In sum, in any case, we can conclude that
there is an integer m ≤ k −2 such that O
C
m
(C
m
) satisﬁes (M
C
m
.F−1
). Note that
C
0
, . . . , C
m
are irrational by Proposition 2.5 and (iii) in Assertion 3.1.
By Theorem1.6, we have K
p,1
(C
m
, C
m

C
m
) K
p,1
(S, C
m
) for any integer p.
On the other hand, the short exact sequence of sheaves 0 → O
S
→ O
S
(C
m
) →
O
C
m
(C
m
) → 0 induces the cohomology long exact sequence
42 R. Kawaguchi
0 → H
0
(S, O
S
) → H
0
(S, C
m
) → H
0
(C
m
, C
m

C
m
) → H
1
(S, O
S
) → · · · . (1)
Since H
0
(S, O
S
) = C and H
1
(S, O
S
) = 0, we obtain h
0
(C
m
, C
m

C
m
) =
h
0
(S, C
m
) − 1. Hence the vanishing K
p,1
(S, C
m
) = 0 holds for any integer
p ≥ h
0
(S, C
m
) − C
m
.F − 1. Then by the same argument as in the case where
k = 2, one can show that O
C
m−1
(C
m−1
) satisﬁes (M
C
m−1
.F−1
). We can induc
tively verify that O
C
j
(C
j
) satisﬁes (M
C
j
.F−1
) for integers 0 ≤ j ≤ m− 2. The
case where j = 0 is the statement of our proposition.
4. The case where degf
1

C
≤ degf
2

C
Let S be a toric surface deﬁned in Section2, and C a curve on S. In this
section, we consider the case where degf
1

C
≤ degf
2

C
, that is, C.F
1
≤ C.F
2
. We
put F = F
1
and F
= F
2
. The aim of this section is to construct a divisor which
satisﬁes Assertion 3.1–3.3.
4.1 Division of the primitive element
First of all, we remark a basic fact about primitive elements of cones.
Fact 4.1. Let i be an integer with 2 ≤ i ≤ d
0
− 1, and deﬁne c = min{c
∈ Z 
c
x
i
−y
i
≥ 0}. Then there is an integer r with 2 ≤ r ≤ i such that (x
r
, y
r
) = (1, c).
In particular, if x
i
≥ 2, then r ≤ i−1 and there exists a pair of integers (s, t) with
r ≤ s < i < t ≤ d
0
−1 such that (x
i
, y
i
) = (x
s
, y
s
) +(x
t
, y
t
) and x
t
y
s
−y
t
x
s
= 1.
Remark 4.2. Let i be an integer with 2 ≤ i ≤ d
0
− 1. If x
i
≥ 2, then d
0
≥ 5,
3 ≤ i ≤ d
0
−2, and moreover, the equality c = min{c
∈ Z  c
x
s
−y
s
≥ 0} holds
in Fact 4.1.
The following lemma gives us more detailed properties of the above division.
Lemma 4.3. For an integer i with 3 ≤ i ≤ d
0
−2 such that x
i
≥ 2, the pair (s, t)
in Fact 4.1 is uniquely determined. If x
s
≥ 2, then we can divide (x
s
, y
s
) into the
sum of primitive elements as
(x
s
, y
s
) = (x
u
, y
u
) + (x
v
, y
v
) (r ≤ u < s < v ≤ d
0
−1, x
v
y
u
−y
v
x
u
= 1)
by Fact 4.1. Then the inequality v ≥ t holds.
Proof. Assume that r ≤ s < i < t ≤ d
0
−1, r ≤ s
< i < t
≤ d
0
−1,
(x
i
, y
i
) = (x
s
, y
s
) + (x
t
, y
t
) = (x
s
, y
s
) + (x
t
, y
t
)
and x
t
y
s
−y
t
x
s
= x
t
y
s
−y
t
x
s
= 1. We put x = x
t
−x
t
and y = y
t
−y
t
. Since
x
t
y
i
−y
t
x
i
= x
t
y
i
−y
t
x
i
= 1, we have x
i
y −y
i
x = 0. Hence there exists an inte
ger n such that x = nx
i
and y = ny
i
. We thus have x
i
= x
s
+x
t
= x
s
+x
t
−nx
i
,
THE GONALITY CONJECTURE FOR CURVES ON TORIC SURFACES WITH TWO P
1
FIBRATIONS 43
which implies n ≥ 0. Similarly, the equation x
i
= x
s
+x
t
= x
s
+nx
i
+x
t
implies
n ≤ 0. Hence we have n = 0, that is, (x
s
, y
s
) = (x
s
, y
s
) and (x
t
, y
t
) = (x
t
, y
t
).
Next we will show the latter part of the lemma. We put x
= x
t
− x
v
and
y
= y
t
−y
v
. Since x
y
s
−y
x
s
= 0, there exists an integer m such that x
= mx
s
and y
= my
s
. Then the equation x
s
= x
u
+ x
v
= x
u
+ x
t
−mx
s
implies m ≥ 0.
Hence we have x
v
y
t
−y
v
x
t
= x
v
y
−y
v
x
= m ≥ 0, which means v ≥ t.
4.2 The auxiliary divisor
Let S, C, F and F
be as above, and put k = C.F and k
= C.F
. If k ≥ 1,
we deﬁne
λ = max{i ≤ d
0
−1  C.D
i
≥ 1},
μ = min{i ≥ d
0
+ 1  C.D
i
≥ 1},
c
0
= min{c ∈ Z  cx
λ
−y
λ
≥ 0},
e
0
= −max{e ∈ Z  ex
μ
−y
μ
≥ 0}.
If x
λ
≥ 2, then by Fact 4.1 and Lemma 4.3, we can divide primitive elements
repeatedly as
(x
λ
, y
λ
) = (x
s
1
, y
s
1
) + n
α
1
(x
α
1
, y
α
1
), x
α
1
y
s
1
−y
α
1
x
s
1
= 1 (s
1
< λ < α
1
< d
0
−1) ,
(x
s
1
, y
s
1
) = (x
s
2
, y
s
2
) + n
α
2
(x
α
2
, y
α
2
), x
α
2
y
s
2
−y
α
2
x
s
2
= 1 (s
2
< s
1
, α
1
< α
2
< d
0
−1) ,
.
.
.
(x
s
a
0
−1
, y
s
a
0
−1
) = (1, c
0
) + n
α
a
0
(x
α
a
0
, y
α
a
0
), x
α
a
0
c
0
−y
α
a
0
= 1 (α
a
0
−1
< α
a
0
≤ d
0
−1) .
On the other hand, in the case where x
λ
= 1, it is obvious that the equation
(x
λ
, y
λ
) = (1, c
0
) holds. Consequently, we obtain the unique division
(x
λ
, y
λ
) = (1, c
0
) + n
α
1
(x
α
1
, y
α
1
) +· · · + n
α
a
0
(x
α
a
0
, y
α
a
0
), (2)
where λ < α
1
< · · · < α
a
0
≤ d
0
− 1. Considering the deﬁnition of the division
and Fact 4.1, one can obtain the equality
x
α
a
_
c
0
+
a
0
j=a+1
n
α
j
y
α
j
_
−y
α
a
_
1 +
a
0
j=a+1
n
α
j
x
α
j
_
= 1 (3)
for each integer 1 ≤ a ≤ a
0
.
Similarly, we can divide
(x
μ
, y
μ
) = (−1, e
0
) + n
β
1
(x
β
1
, y
β
1
) +· · · + n
β
b
0
(x
β
b
0
, y
β
b
0
), (4)
44 R. Kawaguchi
where d
0
+ 1 ≤ β
1
< · · · < β
b
0
< μ. The equality
_
−1 +
b−1
j=1
m
β
j
x
β
j
_
y
β
b
−
_
e
0
+
b−1
j=1
m
β
j
y
β
j
_
x
β
b
= 1 (5)
holds for each integer 1 ≤ b ≤ b
0
.
Deﬁnition 4.4. Let C be a curve on S such that k ≥ 1, and set n
i
= 0 for
integers λ + 1 ≤ i ≤ d
0
except for i = α
1
, . . . , α
a
0
and m
i
= 0 for integers
d
0
+ 1 ≤ i ≤ μ − 1 except for i = β
1
, . . . , β
b
0
. We deﬁne the auxiliary divisor
I
C
=
d
i=1
p
i
D
i
of C as follows :
p
i
=
⎧
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎩
0 (1 ≤ i ≤ λ, μ ≤ i ≤ d),
x
i
_
c
0
+
d
0
−1
j=i+1
n
j
y
j
_
+y
i
_
−1−
d
0
−1
j=i+1
n
j
x
j
_
(λ + 1 ≤ i ≤ d
0
),
x
i
_
−e
0
−
i−1
j=d
0
+1
m
j
y
j
_
+y
i
_
−1+
i−1
j=d
0
+1
m
j
x
j
_
(d
0
+ 1 ≤ i ≤ μ −1).
We note that (3), (5) and an easy computation imply that p
i
= 1 for
i = α
1
, . . . , α
a
0
, d
0
, β
1
, . . . , β
b
0
. Besides, it is obvious that I
C
.F = 1 by deﬁnition.
The rest of this subsection is devoted to verify that I
C
satisﬁes Assertion 3.1.
Proposition 4.5. The divisor I
C
is eﬀective.
Proof. Let i
1
be an integer with λ+1 ≤ i
1
≤ d
0
−1. Since x
λ
y
i
−y
λ
x
i
< 0
for any integer i with λ + 1 ≤ i ≤ d
0
−1, we have
x
λ
_
c
0
+
d
0
−1
i=i
1
+1
n
i
y
i
_
−y
λ
_
1+
d
0
−1
i=i
1
+1
n
i
x
i
_
=x
λ
_
c
0
+
d
0
−1
i=λ+1
n
i
y
i
_
−y
λ
_
1+
d
0
−1
i=λ+1
n
i
x
i
_
−
i
1
i=λ+1
n
i
(x
λ
y
i
−y
λ
x
i
) ≥ x
λ
y
λ
−y
λ
x
λ
=0.
Hence we have the inequalities
y
i
1
x
i
1
<
y
λ
x
λ
≤
c
0
+
d
0
−1
i=i
1
+1
n
i
y
i
1 +
d
0
−1
i=i
1
+1
n
i
x
i
,
which implies that p
i
1
> 0. Let i
2
be an integer with d
0
+ 1 ≤ i
2
≤ μ −1. Then
the inequality
x
μ
_
e
0
+
i
2
−1
i=d
0
+1
m
i
y
i
_
+y
μ
_
1 −
i
2
−1
i=d
0
+1
m
i
x
i
_
=x
μ
y
μ
−y
μ
x
μ
−
μ−1
i=i
2
m
i
(x
μ
y
i
−y
μ
x
i
) ≤ 0
THE GONALITY CONJECTURE FOR CURVES ON TORIC SURFACES WITH TWO P
1
FIBRATIONS 45
implies that
y
i
2
x
i
2
>
y
μ
x
μ
≥
−e
0
−
i
2
−1
i=d
0
+1
m
i
y
i
1 −
i
2
−1
i=d
0
+1
m
i
x
i
.
We thus obtain p
i
2
> 0.
Proposition 4.6. The complete linear system C −I
C
 is base point free.
Proof. We put H = C −I
C
. By Theorem2.3, it is suﬃcient to verify that
H.D
i
≥ 0 for each integer 1 ≤ i ≤ d. For an integer i with 1 ≤ i ≤ λ − 1 or
μ + 1 ≤ i ≤ d, since I
C
.D
i
= 0, we have H.D
i
= C.D
i
≥ 0. For λ, we have
I
C
.D
λ
= p
λ+1
= x
λ+1
_
c
0
+
d
0
−1
i=λ+2
n
i
y
i
_
+ y
λ+1
_
−1 −
d
0
−1
i=λ+2
n
i
x
i
_
= x
λ+1
_
c
0
+
d
0
−1
i=λ+1
n
i
y
i
_
+y
λ+1
_
−1 −
d
0
−1
i=λ+1
n
i
x
i
_
= x
λ+1
y
λ
−y
λ+1
x
λ
= 1
and H.D
λ
= C.D
λ
−1 ≥ 0.
For the case where i = d
0
, since (c
0
− 1)x
λ
0
< y
λ
0
, (e
0
− 1)x
μ
0
< −y
μ
0
and
D
2
d
0
= 0, easy computations give
C.D
d
0
= C.
_
d
0
−2
i=1
y
i
D
i
+
d
i=d
0
+2
y
i
D
i
_
(6)
≥ C.(D
1
+ y
λ
0
D
λ
0
+ y
μ
0
D
μ
0
) ≥ C.D
1
+ c
0
+ e
0
,
I
C
.D
d
0
=
⎧
⎪
⎪
⎨
⎪
⎪
⎩
c
0
+ e
0
(λ + 2 ≤ d
0
≤ μ −2),
c
0
(λ + 2 ≤ d
0
= μ −1),
e
0
(λ + 1 = d
0
≤ μ −2),
0 (λ + 1 = d
0
= μ −1).
Hence we have H.D
d
0
≥ 0.
We next consider the case where i = λ + 1. Since the case where i = d
0
was
already checked, we assume that λ + 1 < d
0
. By Theorem2.1, we have
I
C
.D
λ+1
= p
λ+1
D
2
λ+1
+ p
λ+2
(7)
= −(x
λ
+ x
λ+2
)
_
c
0
+
d
0
−1
i=λ+2
n
i
y
i
_
+ (y
λ
+ y
λ+2
)
_
1 +
d
0
−1
i=λ+2
n
i
x
i
_
+x
λ+2
_
c
0
+
d
0
−1
i=λ+3
n
i
y
i
_
−y
λ+2
_
1 +
d
0
−1
i=λ+3
n
i
x
i
_
= −x
λ
_
c
0
+
d
0
−1
i=λ+1
n
i
y
i
_
+ y
λ
_
1 +
d
0
−1
i=λ+1
n
i
x
i
_
46 R. Kawaguchi
+n
λ+1
(x
λ
y
λ+1
−y
λ
x
λ+1
)
= −n
λ+1
.
Namely, we have H.D
λ+1
≥ 0. For an integer i with λ + 2 ≤ i ≤ d
0
−1, we have
I
C
.D
i
= −n
i
by a computation similar to (7).
Let us consider the case where i = μ. If μ ≥ d
0
+ 2, then we have
I
C
.D
μ
= p
μ−1
= x
μ−1
_
−e
0
−
μ−2
j=d
0
+1
m
j
y
j
_
+ y
μ−1
_
−1 +
μ−2
j=d
0
+1
m
j
x
j
_
= −x
μ−1
y
μ
+ y
μ−1
x
μ
= 1.
If μ = d
0
+1, then we have I
C
.D
μ
= p
d
0
= 1. Hence we have H.D
μ
= C.D
μ
−1 ≥
0.
Consider the case where i = d
0
+1 under the assumption that d
0
+1 = μ. If
μ ≥ d
0
+ 3, then
I
C
.D
d
0
+1
= 1 −e
0
x
d
0
+1
D
2
d
0
+1
+ x
d
0
+2
(−e
0
−m
d
0
+1
y
d
0
+1
) + y
d
0
+2
(−1 + m
d
0
+1
x
d
0
+1
)
= 1 + e
0
x
d
0
−m
d
0
+1
−y
d
0
+2
= −m
d
0
+1
.
If μ = d
0
+2, then we have e
0
= 1 and m
d
0
+1
= −x
d
0
+2
−1 by deﬁnition. We thus
have I
C
.D
d
0
+1
= p
d
0
+ p
d
0
+1
D
2
d
0
+1
= 1 −x
d
0
+1
D
2
d
0
+1
= −m
d
0
+1
. Hence, in any
case, the inequality H.D
d
0
+1
≥ 0 holds. Let i be an integer with d
0
+2 ≤ i ≤ μ−2.
Then we have I
C
.D
i
= −m
i
and H.D
i
≥ 0 by computing similarly to (7).
Lastly, let us check the case where i = μ − 1. Since the cases of d
0
and
d
0
+ 1 were already checked, we assume that μ − 1 ≥ d
0
+ 2. Then we have
I
C
.D
μ−1
= −m
μ−1
by computing. Namely, we have H.D
μ−1
≥ 0.
We list the intersection numbers of I
C
and the Tinvariant divisors for the
later use.
I
C
.D
i
=
⎧
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎩
−n
i
(i = α
1
, . . . , α
a
0
),
1 (i = λ, μ),
−m
i
(i = β
1
, . . . , β
b
0
),
c
0
+ e
0
(i = d
0
),
0 (otherwise).
(8)
Proposition 4.7. If C − F is base point free, then also C − I
C
− F is base
point free.
Proof. Since
(C −I
C
−F).D
i
=
_
(C −I
C
).D
i
−1 (i = 1, d
0
),
(C −I
C
).D
i
≥ 0 (otherwise),
THE GONALITY CONJECTURE FOR CURVES ON TORIC SURFACES WITH TWO P
1
FIBRATIONS 47
it is suﬃcient to verify that (C − I
C
).D
1
and (C − I
C
).D
d
0
are positive. Since
(C − F).D
1
= C.D
1
− 1 ≥ 0 and I
C
.D
1
= 0, we deduce that (C − I
C
).D
1
is
positive. Besides, by an easy computation, we have
(C −I
C
).D
d
0
= (C −I
C
).
_
d
0
−1
i=1
y
i
D
i
+
d
i=d
0
+1
y
i
D
i
_
≥ (C −I
C
).D
1
≥ 1.
We next check the property (ii) in Assertion 3.1.
Proposition 4.8. The vanishing H
1
(S, −I
C
) = 0 holds.
Proof. By RiemannRoch theorem and Theorem2.4, we have
h
1
(S, −I
C
) = h
0
(S, −I
C
) + h
0
(S, K
S
+ I
C
) −
1
2
I
C
.(K
S
+ I
C
) −1
= −
1
2
I
C
.(K
S
+ I
C
) −1.
Recall the remark after Deﬁnition 4.4 and (8). Then
I
C
.(K
S
+ I
C
)
= I
C
.
_
−D
λ
+
a
0
j=1
(p
α
j
−1)D
α
j
+
b
0
j=1
(p
β
j
−1)D
β
j
−D
μ
+ (p
d
0
−1)D
d
0
_
= I
C
.(−D
λ
−D
μ
) = −2.
Proposition 4.9. Assume that k ≥ 2 and C − F is base point free. Then the
divisor C −I
C
satisﬁes the properties (iv)–(vi) in Assertion 3.1.
Proof. (iv) We put H = C − I
C
and write H ∼
d−1
i=2
h
i
D
i
. As we saw
in the Proposition 4.6, H is base point free. By computing (using Theorem2.1
and 2.3), we see that h
i
is nonnegative for each integer 2 ≤ i ≤ d −1. We have
h
d
0
= H.F = k −1 ≥ 1. Hence by (6) and (8), we have
H
2
≥ (C −I
C
).D
d
0
≥ C.D
d
0
−c
0
−e
0
≥ C.D
1
= (C −F).D
1
+ F.D
1
≥ 1.
(v) The short exact sequence of sheaves 0 → O
S
(−I
C
) → O
S
(H) → O
C
(H) → 0
induces the cohomology long exact sequence
· · · → H
1
(S, H) → H
1
(C, H
C
) → H
2
(S, −I
C
) → · · · .
Since H is base point free, we have H
1
(S, H) = 0 by Theorem2.2. On the other
hand, since p
1
= 0 and p
d
0
= 1, we have H
2
(S, −I
C
) = H
0
(S, K
S
+ I
C
) = 0 by
Theorem2.4. We thus obtain H
1
(C, H
C
) = 0.
(vi) We can take a nonsingular irreducible curve C
1
∈ H by (iii), (iv) and
Bertini’s theorem. Then Theorem1.7 shows that K
p,1
(C
1
, C
1

C
1
) = 0 for any
48 R. Kawaguchi
integer p ≥ h
0
(C
1
, C
1

C
1
) − 1. Considering the cohomology long exact sequence
similar to (1), we obtain h
0
(C
1
, C
1

C
1
) = h
0
(S, C
1
) − 1. Thus we see that
K
p,1
(C
1
, C
1

C
1
) = 0 for any integer p ≥ h
0
(S, C
1
) − 2. On the other hand,
by Theorem1.6, K
p,1
(C
1
, C
1

C
1
) K
p,1
(S, C
1
) holds for any integer p.
The following lemma is necessary to prove Assertion 3.3.
Lemma 4.10. Under the same assumption as in Proposition4.9, we can take a
nonsingular irreducible curve C
1
∈ C −I
C
. The divisor (C −2I
C
)
C
1
is nonspe
cial.
This follows from the inequality
deg(C
1
−I
C
)
C
1
−2g
1
= C
1
.(−I
C
−K
S
) −2 ≥ C
1
.D
1
−2 ≥ −1,
where g
1
denotes the genus of C
1
. Here we used the equalities
⎧
⎨
⎩
p
i
= 0 (1 ≤ i ≤ λ, μ ≤ i ≤ d),
p
i
= 1 (i = α
1
, · · · , α
a
0
, d
0
, β
1
, . . . , β
b
0
),
C
1
.D
i
= 0 (otherwise),
which are obtained by (8) and the mention after Deﬁnition 4.4.
4.3 Proof of Assertion 3.2 and 3.3
In order to prove Assertion 3.2 and 3.3 for I
C
in Deﬁnition4.4, in this sub
section, we consider the operation to take auxiliary divisors repeatedly.
Let D be a divisor on S such that D.D
1
≥ 1. For an integer y with
0 ≤ y <
d
0
−1
i=1
y
i
D.D
i
, we deﬁne
i(D, y) = max
_
j ≥ 1
¸
¸
¸ y <
d
0
−1
i=j
y
i
D.D
i
_
,
x(D, y) =
d
0
−1
i=i(D,y)+1
x
i
D.D
i
+
x
i(D,y)
y
i(D,y)
_
y −
d
0
−1
i=i(D,y)+1
y
i
D.D
i
_
.
Let C be a curve on S such as at the beginning of this section, and k, k
, λ and
μ the integers as in the previous subsection.
Remark 4.11. By deﬁnition, we have x(C, 0) = 0 if y
ν
≥ 1. Besides,
x(C, y) = x
λ
y/y
λ
if i(C, y) = λ.
Lemma 4.12. Assume that k ≥ 1 and C.D
1
≥ 1. Let y be an integer with
c
0
≤ y <
d
0
−1
i=2
y
i
C.D
i
, and deﬁne p = min{n ∈ Z  n > x(C − I
C
, y − c
0
)}.
Then
x(C −I
C
, y −c
0
) + 1 ≤ x(C, y) < p + 1.
THE GONALITY CONJECTURE FOR CURVES ON TORIC SURFACES WITH TWO P
1
FIBRATIONS 49
Proof. (i) We ﬁrst consider the case where y ≥ y
λ
. By deﬁnition,
i(C, y) ≤ λ and
d
0
−1
i=i(C,y)+1
y
i
C.D
i
≤ y <
d
0
−1
i=i(C,y)
y
i
C.D
i
. (9)
By computing, we have
d
0
−1
i=i(C,y)
y
i
(C −I
C
).D
i
=
d
0
−1
i=i(C,y)
y
i
C.D
i
−y
λ
I
C
.D
λ
−
a
0
j=1
y
α
j
I
C
.D
α
j
=
d
0
−1
i=i(C,y)
y
i
C.D
i
−y
λ
+
a
0
j=1
n
α
j
y
α
j
=
d
0
−1
i=i(C,y)
y
i
C.D
i
−c
0
,
d
0
−1
i=i(C,y)+1
y
i
(C −I
C
).D
i
=
⎧
⎪
⎨
⎪
⎩
y
λ
−c
0
(i(C, y) = λ),
d
0
−1
i=i(C,y)+1
y
i
C.D
i
−c
0
(i(C, y) ≤ λ −1).
Hence, by (9) and the assumption y ≥ y
λ
, we have
d
0
−1
i=i(C,y)+1
y
i
(C −I
C
).D
i
≤ y −c
0
<
d
0
−1
i=i(C,y)
y
i
(C −I
C
).D
i
,
which means that i(C −I
C
, y −c
0
) = i(C, y). If i(C, y) = λ, then
x(C −I
C
, y −c
0
) = −
d
0
−1
i=λ+1
x
i
I
C
.D
i
+
x
λ
y
λ
_
y −c
0
+
d
0
−1
i=λ+1
y
i
I
C
.D
i
_
= x
λ
−1 +
x
λ
y
λ
y −x
λ
= x(C, y) −1.
If i(C, y) ≤ λ −1, then
x(C −I
C
, y −c
0
) =
d
0
−1
i=i(C,y)+1
x
i
C.D
i
−x
λ
+
a
0
j=1
n
α
j
x
α
j
+
x
i(C,y)
y
i(C,y)
_
y −c
0
−
d
0
−1
i=i(C,y)+1
y
i
C.D
i
+ y
λ
−
a
0
j=1
n
α
j
y
α
j
_
=
d
0
−1
i=i(C,y)+1
x
i
C.D
i
−1
+
x
i(C,y)
y
i(C,y)
_
y −
d
0
−1
i=i(C,y)+1
y
i
C.D
i
_
= x(C, y) −1.
50 R. Kawaguchi
We obtain x(C −I
C
, y −c
0
) = x(C, y) −1 in both cases. Therefore, the claim is
trivial.
(ii) We next consider the case where y < y
λ
. Note that y
λ
≥ 1 and i(C, y) = λ
in this case. Since
y −c
0
< y
λ
−c
0
=
a
0
j=1
n
α
j
y
α
j
= −
d
0
−1
i=α
1
y
i
I
C
.D
i
=
d
0
−1
i=α
1
y
i
(C −I
C
).D
i
,
there exists an integer a with 1 ≤ a ≤ a
0
such that i(C −I
C
, y −c
0
) = α
a
. Then
x(C −I
C
, y −c
0
) =
a
0
j=a+1
n
α
j
x
α
j
+
x
α
a
y
α
a
_
y −c
0
−
a
0
j=a+1
n
α
j
y
α
j
_
= x
λ
−1 −
a−1
j=1
n
α
j
x
α
j
+
x
α
a
y
α
a
_
y −y
λ
+
a−1
j=1
n
α
j
y
α
j
_
< x
λ
−1 −
x
λ
y
λ
a−1
j=1
n
α
j
y
α
j
+
x
λ
y
λ
_
y −y
λ
+
a−1
j=1
n
α
j
y
α
j
_
=
x
λ
y
λ
y −1 = x(C, y) −1.
We next show the inequality x(C, y) < p + 1. Recall the notation s
i
which ap
peared at the beginning of Subsection 4.2. Then y
s
a
≤ y < y
s
a−1
and
x(C −I
C
, y −c
0
) =
a
0
j=a+1
n
α
j
x
α
j
+
x
α
a
y
α
a
_
y −c
0
−
a
0
j=a+1
n
α
j
y
α
j
_
= x
s
a
−1 +
x
α
a
y
α
a
(y −y
s
a
).
Suppose that p + 1 = x(C, y) = x
λ
y/y
λ
. Then we have y = 0 by the inequality
y < y
λ
and the fact that x
λ
and y
λ
are relatively prime. It follows that p is
negative, a contradiction. Suppose that p +1 < x
λ
y/y
λ
. Since x
λ
/y
λ
< x
α
a
/y
α
a
,
we have
(p + 1)y
α
a
−yx
α
a
≤ −1 = −x
α
a
_
c
0
+
d
0
−1
i=α
a
+1
n
i
y
i
_
+ y
α
a
_
1 +
d
0
−1
i=α
a
+1
n
i
x
i
_
p ≤
d
0
−1
i=α
a
+1
n
i
x
i
+
x
α
a
y
α
a
_
y −c
0
+
d
0
−1
i=α
a
+1
n
i
y
i
_
=
d
0
−1
i=α
a
+1
x
i
(C −I
C
).D
i
+
x
α
a
y
α
a
_
y −c
0
+
d
0
−1
i=α
a
+1
y
i
(C −I
C
).D
i
_
= x(C −I
C
, y −c
0
).
THE GONALITY CONJECTURE FOR CURVES ON TORIC SURFACES WITH TWO P
1
FIBRATIONS 51
This contradicts the deﬁnition of p. Hence we can conclude that p + 1 >
x(C, y).
We deﬁne i
(C, y
) and x
(C, y
) in a similar way to i(C, y) and x(C, y),
respectively. Concretely, in the case where C.D
1
≥ 1, for an integer y
with
0 ≤ y
<
d+1
i=d
0
+1
y
i
C.D
i
, we deﬁne
i
(C, y
) = min
_
j ≤ d + 1
¸
¸
¸ y
<
j
i=d
0
+1
y
i
C.D
i
_
,
x
(C, y
) =
i
(C,y
)−1
i=d
0
+1
x
i
C.D
i
+
x
i
(C,y
)
y
i
(C,y
)
_
y
−
i
(C,y
)−1
i=d
0
+1
y
i
C.D
i
_
.
Then one can obtain the following lemma by an argument similar to that in the
proof of Lemma 4.12.
Lemma 4.13. Assume that k ≥ 1 and C.D
1
≥ 1. Let y
be an integer with
v
0
≤ y
<
d+1
i=d
0
+2
y
i
C.D
i
, and deﬁne p
= max{n ∈ Z  n < x
(C−I
C
, y
−e
0
)}.
Then
p
−1 < x
(C, y
) ≤ x
(C −I
C
, y
−e
0
) −1.
Let us consider the situation that we take auxiliary divisors repeatedly. From
the convex geometrical view point, this means that we reduce the size of the lat
tice polytope step by step. We put C
0
= C, and deﬁne inductively
I
j−1
: the auxiliary divisor of C
j−1
,
C
j
: a nonsingular irreducible curve in C
j−1
−I
j−1
,
λ
j
= max{i ≤ d
0
−1  C
j
.D
i
≥ 1},
μ
j
= min{i ≥ d
0
+ 1  C
j
.D
i
≥ 1},
c
j
= min{c ∈ Z  cx
λ
j
−y
λ
j
≥ 0},
e
j
= −max{e ∈ Z  ex
μ
j
−y
μ
j
≥ 0}
for an integer 1 ≤ j ≤ k.
Lemma 4.14. Assume that k ≥ 1 and C.D
1
≥ 1. Let y and j be inte
gers with y ≤
d
0
−1
i=2
y
i
C
0
.D
i
and 1 ≤ j ≤ k. If y ≥ c
0
+ · · · + c
j−1
, then
x(C
j
, y −c
0
−· · · −c
j−1
) + j ≤ x(C
0
, y) and there is no integer in the halfopen
interval
_
x(C
j
, y −c
0
−· · · −c
j−1
) + j, x(C
0
, y)
¸
.
Proof. The inequality x(C
j
, y − c
0
− · · · − c
j−1
) + j ≤ x(C
0
, y) follows
immediately from Lemma 4.12. Let j
be an integer with 0 ≤ j
≤ j −1. We have
d
0
−1
i=2
y
i
C
j
.D
i
=
d
0
−1
i=2
y
i
C
0
.D
i
−c
0
−· · · −c
j
−1
52 R. Kawaguchi
by computing, where we deﬁne c
0
+· · · +c
j
−1
= 0 in the case where j
= 0. We
thus have c
j
≤ y−c
0
−· · · −c
j
−1
≤
d
0
−1
i=2
y
i
C
j
.D
i
. Then by Lemma 4.12, there
is no integer in the halfopen interval
_
x(C
j
+1
, y−c
0
−· · · −c
j
)+j
+1, x(C
j
, y−
c
0
−· · · −c
j
−1
) + j
¸
. Hence the statement of the lemma is obvious.
A similar lemma holds for the opposite side of the lattice polytope.
Lemma 4.15. Assume that k ≥ 1 and C.D
1
≥ 1. Let y
and j be integers
with y
≤
d+1
i=d
0
+2
y
i
C
0
.D
i
and 1 ≤ j ≤ k. If y
≥ e
0
+ · · · + e
j−1
, then
x
(C
j
, y
−e
0
−· · · −e
j−1
) −j ≥ x
(C
0
, y
) and there is no integer in the halfopen
interval
_
x
(C
0
, y
), x
(C
j
, y
−e
0
−· · · −e
j−1
) −j
_
.
Lemma 4.16. The equality c
0
+· · · + c
x
λ
0
−1
= y
λ
0
holds.
Proof. Suppose c
0
+· · · + c
x
λ
0
−1
≤ y
λ
0
−1. Then we have
x(C
0
, c
0
+· · · + c
x
λ
0
−1
) =
x
λ
0
y
λ
0
(c
0
+· · · + c
x
λ
0
−1
)
≤
x
λ
0
y
λ
0
(y
λ
0
−1) < x
λ
0
≤ x(C
x
λ
0
, 0) + x
λ
0
,
which contradicts Lemma 4.14. Suppose c
0
+ · · · + c
x
λ
0
−1
≥ y
λ
0
+ 1 and put
s = max{i ≤ x
λ
0
−1  c
i
≥ 1}. Then we have
x(C
s
, c
s
−1) + s =
x
λ
s
y
λ
s
(c
s
−1) + s < s + 1,
x(C
0
, c
0
+· · · + c
s
−1) = x(C
0
, c
0
+· · · + c
x
λ
0
−1
−1) ≥ x(C
0
, y
λ
0
)
= x
λ
0
≥ s + 1.
They contradict Lemma 4.14. Hence we can conclude that c
0
+ · · · + c
x
λ
0
−1
=
y
λ
0
.
Lemma 4.17. If c
0
≥ 1 and x
λ
0
≥ 2, then c
0
− 1 ≤ y
λ
j
/x
λ
j
≤ y
λ
0
/x
λ
0
for
integers 1 ≤ j ≤ x
λ
0
− 1, especially the equality y
λ
j
/x
λ
j
= c
0
− 1 holds for
j = x
λ
0
−1.
Proof. Let j be an integer with 1 ≤ j ≤ x
λ
0
−1
− 1. We ﬁrst show the
inequality y
λ
j
/x
λ
j
≥ c
0
−1.
(i) We consider the case where x
λ
j
≥ 2. Since x
j
and y
j
are relatively prime, we
obtain c
j
x
λ
j
> y
λ
j
in this case. Hence
x(C
j
, y
λ
j
−c
j
) + j =
x
λ
j
y
λ
j
(y
λ
j
−c
j
) + j = x
λ
j
−
x
λ
j
y
λ
j
c
j
+ j < x
λ
j
−1 + j.
On the other hand, we have
THE GONALITY CONJECTURE FOR CURVES ON TORIC SURFACES WITH TWO P
1
FIBRATIONS 53
x(C
0
, c
0
+· · · + c
j−1
+ y
λ
j
−c
j
) =
x
λ
0
y
λ
0
(c
0
+· · · + c
j−1
+ y
λ
j
) −
x
λ
0
y
λ
0
c
j
= x(C
0
, c
0
+· · · + c
j−1
+ y
λ
j
) −
x
λ
0
y
λ
0
c
j
≥ x(C
j
, y
λ
j
) + j −
x
λ
0
y
λ
0
c
j
= x
λ
j
+ j −
x
λ
0
y
λ
0
c
j
.
Since (x(C
j
, y
λ
j
−c
j
) +j, x(C
0
, c
0
+· · · +c
j−1
+y
λ
j
−c
j
)] ∩Z = ∅ by Lemma 4.14,
we have c
0
≤ c
j
and y
λ
j
/x
λ
j
> c
j
−1 ≥ c
0
−1.
(ii) We consider the case where x
λ
j
= 1. Note that y
λ
j
= c
j
. We deﬁne
s = max{i ≤ j −1  c
i
≥ 1}. Then we have
x(C
s
, c
s
−1) + s =
x
λ
s
y
λ
s
(c
s
−1) + s < s + 1 ≤ j.
On the other hand, we have
x(C
0
, c
0
+· · · + c
s
−1) = x(C
0
, c
0
+· · · + c
j
−(c
j
+ 1))
≥ x(C
j
, c
j
) + j −
x
λ
0
y
λ
0
(c
j
+ 1) ≥ j + 1 −
x
λ
0
y
λ
0
(c
j
+ 1).
Then by Lemma 4.14, we have c
0
−1 ≤ c
j
= y
λ
j
/x
λ
j
.
We next show the inequality y
λ
j
/x
λ
j
≤ y
λ
0
/x
λ
0
. If y
λ
j
/x
λ
j
= 0, then the
inequality is obviously holds. Hence we consider the case where y
λ
j
/x
λ
j
> 0.
Suppose y
λ
j
/x
λ
j
> y
λ
0
/x
λ
0
. Then we have
x
_
C
0
, c
0
+· · · + c
j−1
+
y
λ
0
x
λ
0
_
= x(C
0
, c
0
+· · · + c
j−1
) + 1
≥ x(C
j
, 0) + j + 1 = j + 1,
x
_
C
j
,
y
λ
0
x
λ
0
_
+ j =
x
λ
j
y
λ
j
·
y
λ
0
x
λ
0
+ j < j + 1.
They contradict Lemma 4.14.
Lastly, we show y
λ
a
/x
λ
a
= c
0
− 1, where we put a = x
λ
0
− 1 for simplicity.
Because of the above arguments, it is suﬃcient to verify that y
λ
a
/x
λ
a
≤ c
0
− 1.
Since this is obvious in the case where y
λ
a
/x
λ
a
= 0, we assume y
λ
a
/x
λ
a
> 0.
Then we have x(C
a
, 0) + a = a = x
λ
0
−1. On the other hand, we have
x(C
0
, c
0
+· · · + c
a−1
) = x(C
0
, c
0
+· · · + c
a
) −
x
λ
0
y
λ
0
c
a
= x(C
0
, y
λ
0
) −
x
λ
0
y
λ
0
c
a
= x
λ
0
−
x
λ
0
y
λ
0
c
a
.
54 R. Kawaguchi
Note that c
0
x
λ
0
> y
λ
0
since x
λ
0
≥ 2. Then by Lemma 4.14, we have
x
λ
0
−x
λ
0
c
a
/y
λ
0
≥ x
λ
0
−1, which implies that c
0
> y
λ
0
/x
λ
0
≥ c
a
≥ y
λ
a
/x
λ
a
.
Assertion 3.2 follows immediately from the following proposition. Needless
to say, the integer m
0
in the following proposition coincides with that in Asser
tion3.2.
Proposition 4.18. Assume that k
> k ≥ 2 and C−F is base point free. Then
there exists a positive integer m
0
≤ k such that C
j
.F
= C
j
.F for j = 1, . . . , m
0
−1
and C
m
0
.F
> C
m
0
.F.
Proof. (i) We ﬁrst consider the case where c
0
≥ 2. Note that x
i
y
λ
0
≤ y
i
x
λ
0
for integers 2 ≤ i ≤ d
0
−1 if C.D
i
≥ 1. Then we have
C
1
.F = k −1 =
λ
0
i=2
x
i
C.D
i
−1
=
λ
0
−1
i=2
x
i
C.D
i
+ x
λ
0
(C.D
λ
0
−1) + x
λ
0
−1,
C
1
.F
= k
−c
0
−e
0
=
λ
0
i=1
y
i
C.D
i
+
d
i=μ
0
y
i
C.D
i
−c
0
−e
0
≥ C.D
1
+
y
λ
0
x
λ
0
λ
0
−1
i=2
x
i
C.D
i
+ y
λ
0
C.D
λ
0
+ y
μ
0
−c
0
−e
0
> 1 + (c
0
−1)
λ
0
−1
i=2
x
i
C.D
i
+ (c
0
−1)x
λ
0
C.D
λ
0
−c
0
= (c
0
−1)
λ
0
−1
i=2
x
i
C.D
i
+ (c
0
−1)x
λ
0
(C.D
λ
0
−1) + (c
0
−1)(x
λ
0
−1).
Hence we have C
1
.F
> C
1
.F. Namely, the lemma is valid for m
0
= 1 in this
case.
(ii) Similarly, in the case where e
0
≥ 2, one can show that the lemma is valid for
m
0
= 1.
From now on, we assume that c
0
≤ 1 and e
0
≤ 1.
(iii) Assume that c
0
+ e
0
≤ 1 or k
≥ k + 2. Then, since
C
1
.F
−C
1
.F = k
−I
0
.D
d
0
−k + 1 = k
−k −c
0
−e
0
+ 1 ≥ 1,
the lemma is valid for m
0
= 1.
(iv) For the remaining case where c
0
= e
0
= 1 and k
= k + 1, it is suﬃcient to
verify the following claim:
ClaimA: Two inequalities 2 ≤ x
λ
0
≤ k, −k ≤ x
μ
0
≤ −2 hold and there
THE GONALITY CONJECTURE FOR CURVES ON TORIC SURFACES WITH TWO P
1
FIBRATIONS 55
exists an integer m
0
with 2 ≤ m
0
≤ min{x
λ
0
, −x
μ
0
} such that
I
j
.D
d
0
=
_
1 (1 ≤ j ≤ m
0
−2),
0 (j = m
0
−1).
First we mention the inequalities k =
d
0
−1
i=2
x
i
C.D
i
≥ x
λ
0
and k =
−
d
i=d
0
+1
x
i
C.D
i
≥ −x
μ
0
. By the deﬁnition of λ
0
, x
i
y
λ
0
≤ y
i
x
λ
0
holds if
2 ≤ i ≤ d
0
− 1 and C.D
i
≥ 1. Similarly, x
i
y
μ
0
≥ y
i
x
μ
0
holds if d
0
+ 1 ≤ i ≤ d
and C.D
i
≥ 1. Hence we obtain
k
= C.D
1
+
d
0
−1
i=2
y
i
C.D
i
+
d
i=d
0
+1
y
i
C.D
i
≥ 1 +
y
λ
0
x
λ
0
d
0
−1
i=2
x
i
C.D
i
+
y
μ
0
x
μ
0
d
i=d
0
+1
x
i
C.D
i
= 1 +
_
y
λ
0
x
λ
0
−
y
μ
0
x
μ
0
_
k.
It follows that
y
λ
0
x
λ
0
−
y
μ
0
x
μ
0
≤ 1. (10)
Since c
0
= e
0
= 1, we have y
λ
0
/x
λ
0
< 1 and y
μ
0
/x
μ
0
> −1, which mean that
x
λ
0
≥ 2 and x
μ
0
≤ −2.
From now on, we assume that x
λ
0
≤ −x
μ
0
. The case where x
λ
0
≥ −x
μ
0
can
be shown in a similar way. Let j be an integer with 1 ≤ j ≤ x
λ
0
− 1 such that
I
1
.D
d
0
= · · · = I
j−1
.D
d
0
= 1. We verify that I
j
.D
d
0
≤ 1, especially I
j
.D
d
0
= 0 if
j = x
λ
0
−1. By Lemma 4.14 and 4.15, we have
x(C
0
, c
0
+· · · + c
j−1
) =
x
λ
0
y
λ
0
(c
0
+· · · + c
j−1
) < x(C
j
, 0) + j + 1, (11)
x
(C
0
, e
0
+· · · + e
j−1
) =
x
μ
0
y
μ
0
(e
0
+· · · + e
j−1
) > x
(C
j
, 0) −j −1.
On the other hand, it follows from Lemma 4.17 that c
j
, e
j
≤ 1. If c
j
= e
j
= 1,
then we have x(C
j
, 0) = x
(C
j
, 0) = 0 and
y
λ
0
x
λ
0
−
y
μ
0
x
μ
0
>
c
0
+· · · + c
j−1
+ e
0
+· · · + e
j−1
j + 1
=
c
0
+ e
0
+ I
1
.D
d
0
+· · · + I
j−1
.D
d
0
j + 1
= 1,
a contradiction. Hence we obtain I
j
.D
d
0
= c
j
+ e
j
≤ 1.
In the case where j = x
λ
0
−1, we have c
j
= 0 by Lemma 4.17 and
c
0
+· · · + c
j−1
j + 1
=
y
λ
0
x
λ
0
56 R. Kawaguchi
by Lemma 4.16. If e
j
= 1, then we have x
(C
j
, 0) = 0. Hence, by noting (11), we
obtain a contradiction
y
λ
0
x
λ
0
−
y
μ
0
x
μ
0
>
c
0
+· · · + c
j−1
+ e
0
+· · · + e
j−1
j + 1
= 1.
Therefore, we can conclude e
j
= 0, that is, I
j
.D
d
0
= 0 in this case.
Let us show one more lemma needed in the proof of Assertion 3.3.
Lemma 4.19. Assume that k
> k ≥ 2 and C − F is base point free, and let
m
0
be a positive integer in Proposition4.18. If c
0
= e
0
= 1, then either of the
inequalities x
λ
j−1
≥ 2 or x
μ
j−1
≤ −2 holds for each integer 1 ≤ j ≤ m
0
−1.
Proof. Note that, in this case, the integer m
0
is equal to that in ClaimA
in the proof of Proposition 4.18. The case where j = 1 is contained in ClaimA.
Hence let j be an integer with 2 ≤ j ≤ m
0
−1, and suppose x
λ
j−1
= −x
μ
j−1
= 1.
Since I
j−1
.D
d
0
= c
j−1
+e
j−1
= 1, either c
j−1
or e
j−1
is equal to one. If c
j−1
= 1,
then we have y
λ
0
/x
λ
0
≥ y
λ
j−1
/x
λ
j−1
= 1 by Lemma 4.17. Hence (10) implies
that y
μ
0
/x
μ
0
= 0. This contradicts the fact that e
0
= 1. We obtain a similar
contradiction in the case where e
j−1
= 1. Therefore, two equalities x
λ
j−1
= 1
and x
μ
j−1
= −1 do not occur at the same time.
Finally, we prove Assertion 3.3.
Proposition 4.20. Assume that k
> k ≥ 2 and C−F is base point free, and let
m
0
be a positive integer in Proposition4.18. Then the inequality in Assertion 3.3
holds for each integer 1 ≤ j ≤ m
0
. Besides, if k
= k = 2, then this inequality
holds for j = 1.
Proof. Considering the cohomology long exact sequence similar to (1), we
obtain h
0
(S, C
j
) = h
0
(C
j
, C
j

C
j
)+1. On the other hand, the short exact sequence
0 → O
S
(−I
j−1
) → O
S
(C
j
− I
j−1
) → O
C
j
(C
j
− I
j−1
) → 0 and Proposition 4.8
induce the cohomology exact sequence
0 → H
0
(S, C
j
−I
j−1
) → H
0
(C
j
, (C
j
−I
j−1
)
C
j
) → 0. (12)
Hence it is suﬃcient to verify the inequality
h
0
(C
j
, C
j

C
j
) −h
0
(C
j
, (C
j
−I
j−1
)
C
j
) ≥ C
j
.F + 1. (13)
The divisor (C
j
−I
j−1
)
C
j
is nonspecial by Lemma 4.10. By RiemannRoch the
orem, we have
degC
j

C
j
−2g
j
= −C
j
.K
S
−2 ≥ C
j
.D
1
−2 ≥ −1, (14)
where g
j
denotes the genus of C
j
. Hence also C
j

C
j
is nonspecial. Thus (13) is
THE GONALITY CONJECTURE FOR CURVES ON TORIC SURFACES WITH TWO P
1
FIBRATIONS 57
equivalent to the inequality
C
j
.I
j−1
≥ C
j
.F + 1. (15)
In the case where c
0
= 1 or e
0
= 1 or k
≥ k + 2, m
0
is equal to one by (i)–(iii)
in the proof of Proposition 4.18. Since C
1
.I
0
≥ C
1
.D
d
0
= C
1
.F
≥ C
1
.F + 1 by
the deﬁnition of m
0
, (15) holds in this case.
Let us consider the case where c
0
= 1, e
0
= 1 and k
= k + 1. In the case
where 1 ≤ j ≤ m
0
−1, we have x
λ
j−1
≥ 2 or x
μ
j−1
≤ −2 by Lemma 4.19. Hence
if we divide (x
λ
j−1
, y
λ
j−1
) and (x
μ
j−1
, y
μ
j−1
) as (2) and (4), then either α
a
0
or
β
b
0
exists. If α
a
0
exists, then by (8), we have
C
j
.I
j−1
≥ C
j
.(D
α
a
0
+ D
d
0
) = −I
j−1
.D
α
a
0
+ C
0
.D
d
0
−(I
0
+· · · + I
j−1
).D
d
0
= n
α
a
0
+ C
0
.F
−j −1 ≥ C
0
.F −j + n
α
a
0
= C
j
.F + n
α
a
0
.
A similar computation is carried out in the case where β
b
0
exists to verify
the inequality C
j
.I
j−1
≥ C
j
.F + m
β
b
0
. In the case where j = m
0
, we have
C
m
0
.I
m
0
−1
≥ C
m
0
.D
d
0
= C
m
0
.F
≥ C
m
0
.F + 1.
Lastly, let us show (15) for j = 1 under the assumption that k
= k = 2.
Namely, we will show the inequality C
1
.I
0
≥ 2. If x
λ
0
≥ 2 or x
μ
0
≤ −2, then this
can be proved by the same argument as above. Assume x
λ
0
= 1 and x
μ
0
= −1.
Since k =
d
0
−1
i=2
x
i
C.D
i
≥ 2, C.D
λ
0
≥ 2 or there exists an integer i with
2 ≤ i ≤ d
0
−1 such that i = λ
0
and C.D
i
≥ 1. We thus have
1 = k
−1 ≥
d
0
−1
i=2
y
i
C.D
i
≥
y
λ
0
x
λ
0
d
0
−1
i=2
x
i
C.D
i
> y
λ
0
.
Hence y
λ
0
must be zero. Similarly, one can obtain y
μ
0
= 0. We thus have
I
0
.D
d
0
= c
0
+ e
0
= 0 and C
1
.I
0
≥ C
1
.D
d
0
= C.D
d
0
= k
= 2.
5. The case where degf
1

C
≥ degf
2

C
In this section, we consider the case where degf
1

C
≥ degf
2

C
, that is,
C.F
1
≥ C.F
2
. We put F = F
2
, F
= F
1
, k = C.F and k
= C.F
. In order
to simplify the argument, we renumber the Tinvariant divisors. Concretely, we
denote by D
1
the Tinvariant divisor whose associated cone has the primitive
element (−1, 0), and number other cones in Δ
S
clockwise. In particular, we set
D
d
0
if its primitive element is (0, 1) (see Fig. 2). Note that, in this case, the ﬁber
F and F
are written as
F ∼ D
d
∼
d−2
i=2
y
i
D
i
58 R. Kawaguchi
F
∼ −
d
0
−1
i=1
x
i
D
i
∼
d−1
i=d
0
+1
x
i
D
i
.
σ(D
1
)
σ(D
2
)
σ(D
d
0
)
σ(D
d−1
)
σ(D
d
)
Figure 2
If k
≥ k ≥ 2, we deﬁne
ν = max{i ≤ d
0
 C.D
i
≥ 1},
ξ = min{i ≥ d
0
 C.D
i
≥ 1},
c
0
= −max{c ∈ Z  x
ν
−cy
ν
≥ 0},
e
0
= min{e ∈ Z  x
ξ
−ey
ξ
≤ 0},
f
0
= min{c
0
, e
0
}.
If ν = 1 or ξ = d −1, then C becomes a curve considered in Section 4. We thus
assume that ν ≥ 2 and ξ ≤ d − 2 in this section. Note that the ﬁve statements
c
0
= 0, e
0
= 0, f
0
= 0, ν = d
0
and ξ = d
0
are equivalent. By Fact 4.1 and the
argument in Subsection 4.1, we obtain the unique divisions
(x
ν
, y
ν
) = (−c
0
, 1) + n
γ
1
(x
γ
1
, y
γ
1
) +· · · + n
γ
s
0
(x
γ
s
0
, y
γ
s
0
) (16)
(ν < γ
1
< · · · < γ
s
0
≤ d
0
),
(x
ξ
, y
ξ
) = (e
0
, 1) + m
δ
1
(x
δ
1
, y
δ
1
) +· · · + m
δ
t
0
(x
δ
t
0
, y
δ
t
0
)
(d
0
≤ δ
1
< · · · < δ
t
0
< ξ),
where n
γ
i
and m
δ
j
are positive integers. We note that
x
γ
s
_
1 +
s
0
j=s+1
n
γ
j
y
γ
j
_
−y
γ
s
_
−c
0
+
s
0
j=s+1
n
γ
j
x
γ
j
_
= 1 (17)
for each integer 1 ≤ s ≤ s
0
and
_
e
0
+
t−1
j=1
m
δ
j
x
δ
j
_
y
δ
t
−
_
1 +
t−1
j=1
m
δ
j
y
δ
j
_
x
δ
t
= 1 (18)
THE GONALITY CONJECTURE FOR CURVES ON TORIC SURFACES WITH TWO P
1
FIBRATIONS 59
for each integer 1 ≤ t ≤ t
0
. There exists an integer ι such that (x
ι
, y
ι
) =
(e
0
− c
0
, 1). Obviously, we see that ι = d
0
(resp. ν + 1 ≤ ι ≤ ξ − 1) if f
0
= 0
(resp. f
0
≥ 1).
5.1 The auxiliary divisor
In this subsection, we keep the above notation. Let C be a curve on S such
that k ≤ k
, ν ≥ 2 and ξ ≤ d −2.
Deﬁnition 5.1. Let C be a curve on S such that k ≥ 2, and set n
i
= 0 for inte
gers ν+1 ≤ i ≤ d
0
except for i = γ
1
, . . . , γ
s
0
and m
i
= 0 for integers d
0
≤ i ≤ ξ−1
except for i = δ
1
, . . . , δ
t
0
. We deﬁne the auxiliary divisor I
C
=
d
i=1
q
i
D
i
+ F
of C as follows :
(i) The case where ι ≤ d
0
(that is, c
0
≥ e
0
).
q
i
=
⎧
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎩
x
i
_
1 +
d
0
j=i+1
n
j
y
j
_
+ y
i
_
c
0
−
d
0
j=i+1
n
j
x
j
_
(ν + 1 ≤ i ≤ ι −1),
e
0
y
i
(ι ≤ i ≤ d
0
),
x
i
_
−1 −
i−1
j=d
0
m
j
y
j
_
+ y
i
_
e
0
+
i−1
j=d
0
m
j
x
j
_
(d
0
+ 1 ≤ i ≤ ξ −1),
0 (otherwise).
(ii) The case where ι ≥ d
0
+ 1 (that is, c
0
< e
0
).
q
i
=
⎧
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎩
x
i
_
1 +
d
0
j=i+1
n
j
y
j
_
+ y
i
_
c
0
−
d
0
j=i+1
n
j
x
j
_
(ν + 1 ≤ i ≤ d
0
),
c
0
y
i
(d
0
+ 1 ≤ i ≤ ι),
x
i
_
−1 −
i−1
j=d
0
m
j
y
j
_
+ y
i
_
e
0
+
i−1
j=d
0
m
j
x
j
_
(ι + 1 ≤ i ≤ ξ −1),
0 (otherwise).
It is obvious that I
C
.F = 1 by deﬁnition. We remark that I
C
= F
if f
0
= 0.
Besides, by (17) and (18), we have q
i
= 1 for i = γ
1
, . . . , γ
i
0
, δ
1
, . . . , δ
j
0
.
Lemma 5.2. If c
0
≥ e
0
≥ 1, then ι ≥ γ
s
0
.
Proof. If ι < γ
s
0
, then we have x
γ
s
0
/y
γ
s
0
> x
ι
/y
ι
= e
0
− c
0
. By (17),
the inequality y
γ
s
0
e
0
< x
γ
s
0
+ y
γ
s
0
c
0
= 1 holds. This implies that y
γ
s
0
= 0, a
contradiction.
Corollary 5.3. If c
0
≥ e
0
≥ 1, then n
i
= 0 for any integer ι + 1 ≤ i ≤ d
0
.
Lemma 5.4. Assume that f
0
≥ 1. If ι = ν + 1 (resp. ι = ξ − 1), then e
0
= 1
(resp. c
0
= 1).
Proof. We note that c
0
and e
0
are positive by assumption. Assume that
60 R. Kawaguchi
ι = ν + 1. Then, since x
ι
y
ν
−y
ι
x
ν
= 1, we have
e
0
= c
0
+
x
ν
y
ν
+
1
y
ν
< c
0
+ (−c
0
+ 1) +
1
y
ν
= 1 +
1
y
ν
≤ 2.
Similarly, if ι = ξ −1, then the equality x
ξ
y
ι
−y
ξ
x
ι
= 1 implies that
c
0
= e
0
−
x
ξ
y
ξ
+
1
y
ξ
< e
0
−(e
0
−1) +
1
y
ξ
= 1 +
1
y
ξ
≤ 2.
There is no essential diﬀerence between the cases where c
0
≥ e
0
and c
0
≤ e
0
.
Hence we will prove all the remaining propositions and lemmas in this subsection
only for the former case.
Proposition 5.5. The divisor I
C
is eﬀective.
This lemma can be shown by a computation similar to that in the proof of
Proposition 4.5. The following Proposition 5.6 and 5.8 correspond to the property
(iii) in Assertion 3.1.
Proposition 5.6. Assume that k ≥ 2. The complete linear system C − I
C
 is
base point free.
Proof. If f
0
= 0, then C.D
d
0
≥ 1 and I
C
= F
. Hence our lemma is
clear in this case. We thus consider the case where f
0
≥ 1, that is, c
0
and e
0
are positive and ξ ≥ d
0
+ 1. We will compute the intersection number I
C
.D
i
only for i = d
0
, ι. In the other cases, we can compute similarly to the proof of
Proposition 4.6 to verify that (C −I
C
).D
i
is nonnegative.
In the case where i = d
0
, we divide the situation into six cases as follows :
(i) ν + 1 = ι = d
0
= ξ −1,
(ii) ν + 2 ≤ ι = d
0
= ξ −1,
(iii) ι + 1 ≤ d
0
= ξ −1,
(iv) ν + 1 = ι = d
0
≤ ξ −2,
(v) ν + 2 ≤ ι = d
0
≤ ξ −2,
(vi) ι + 1 ≤ d
0
≤ ξ −2.
Lemma 5.4 and the assumption c
0
≥ e
0
implies that e
0
= 1 in the case (iv), and
c
0
= e
0
= 1 in the cases (i) and (ii). By deﬁnition, n
d
0
= y
d
0
−1
−1 if ν = d
0
−1,
and m
d
0
= y
d
0
+1
−1 if ξ = d
0
+ 1.
(i) Since I
C
= D
d
0
+F
, we have I
C
.D
d
0
= −y
d
0
−1
−y
d
0
+1
+1 = −n
d
0
−m
d
0
−1.
(ii) Since I
C
= (−1 −n
d
0
+ y
d
0
−1
)D
d
0
−1
+ D
d
0
+ F
, we have
I
C
.D
d
0
= −1 −n
d
0
+ y
d
0
−1
−y
d
0
−1
−y
d
0
+1
+ 1 = −n
d
0
−m
d
0
−1.
THE GONALITY CONJECTURE FOR CURVES ON TORIC SURFACES WITH TWO P
1
FIBRATIONS 61
(iii) In this case, we have e
0
= 1 by deﬁnition. Besides, γ
s
0
≤ d
0
−1 by Lemma 5.2,
which implies that n
d
0
= 0. Hence I
C
.D
d
0
= (y
d
0
−1
D
d
0
−1
+ D
d
0
+ F
).D
d
0
=
−y
d
0
+1
+ 1 = −n
d
0
−m
d
0
.
(iv) Since I
C
= D
d
0
+ (−1 −m
d
0
+ y
d
0
+1
)D
d
0
+1
+ F
, we have
I
C
.D
d
0
= −y
d
0
−1
−y
d
0
+1
−1−m
d
0
+y
d
0
+1
+1 = −y
d
0
−1
−m
d
0
= −n
d
0
−m
d
0
−1.
(v) Note that c
0
= e
0
. Then, since q
d
0
−1
= −1 − n
d
0
+ y
d
0
−1
, q
d
0
= e
0
and
q
d
0
+1
= −1 −m
d
0
+ e
0
y
d
0
+1
, we have
I
C
.D
d
0
= −1 −n
d
0
+ y
d
0
−1
−e
0
(y
d
0
−1
+ y
d
0
+1
) −1 −m
d
0
+ e
0
y
d
0
+1
+ 1
= −n
d
0
−m
d
0
−1.
(vi) In this case, we have n
d
0
= 0 by the same argument as in (iii). Hence
I
C
.D
d
0
= e
0
y
d
0
−1
−e
0
(y
d
0
−1
+ y
d
0
+1
) −1 −m
d
0
+ e
0
y
d
0
+1
+ 1 = −n
d
0
−m
d
0
.
We next compute I
C
.D
ι
under the assumption ι ≤ d
0
−1. If ι ≥ ν + 2, then
we have
I
C
.D
ι
= x
ι−1
(1 + n
ι
y
ι
) + y
ι−1
(c
0
−n
ι
x
ι
) −e
0
(y
ι−1
+ y
ι+1
) + e
0
y
ι+1
= x
ι−1
+ n
ι
(x
ι−1
x
ι
−y
ι−1
x
ι
) + (c
0
−e
0
)y
ι−1
= y
ι
x
ι−1
−x
ι
y
ι−1
−n
ι
= −n
ι
−1.
In the case where ι = ν + 1, since e
0
= 1 by Lemma 5.4, we have I
C
.D
ι
=
−(y
ι−1
+ y
ι+1
) + y
ι+1
= −y
ν
. If y
ν
= 1, then we have n
ι
= 0, that is,
I
C
.D
ι
= −n
ι
− 1. If y
ν
≥ 2, then since γ
s
0
≤ ν + 1 by Lemma 5.2, we have
γ
s
0
= ν + 1 and s
0
= 1. Hence I
C
.D
ι
= −1 −n
γ
1
y
γ
1
= −n
ι
−1.
List 5.7. We list the intersection numbers of I
C
and the Tinvariant divisors on
S.
(A) In the case where f
0
= 0,
I
C
.D
i
=
_
1 (i = d
0
, d),
0 (otherwise).
(B) In the case where f
0
≥ 1 and ι = d
0
,
I
C
.D
i
=
⎧
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎩
−n
d
0
−m
d
0
−1 (i = d
0
),
−n
i
(i = d
0
, i = γ
1
, . . . , γ
s
0
),
1 (i = ν, ξ, d),
−m
i
(i = d
0
, i = δ
1
, . . . , δ
t
0
),
0 (otherwise).
(C) In the case where f
0
≥ 1 and ι ≤ d
0
−1,
62 R. Kawaguchi
I
C
.D
i
=
⎧
⎪
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎪
⎩
−n
d
0
−m
d
0
(i = d
0
),
−n
i
−1 (i = ι),
−n
i
(i = ι, i = γ
1
, . . . , γ
s
0
),
1 (i = ν, ξ, d),
−m
i
(i = d
0
, i = δ
1
, . . . , δ
t
0
),
0 (otherwise).
Proposition 5.8. If C − F is base point free, then also C − I
C
− F is base
point free.
We can easily show the above proposition by using List 5.7. We next check
the property (ii) in Assertion 3.1.
Proposition 5.9. The vanishing H
1
(S, −I
C
) = 0 holds.
Proof. We write the linear equivalence class of F
as F
∼ −
d
0
−1
i=1
x
i
D
i
.
Clearly the coeﬃcients of D
1
and D
d−1
in I
C
are one and zero, respectively. It
follows from Theorem2.4 that H
0
(S, K
S
+ I
C
) = 0. Hence, by the same compu
tation as that in Proposition 4.8, it is enough to show I
C
.(K
S
+ I
C
) = −2.
In the case (A) in List 5.7, since I
C
= F
, we have I
C
.(K
S
+ I
C
) = F
.K
S
=
F
.(−D
d
0
− D
d
) = −2. Recall that q
i
= 1 for i = γ
1
, . . . , γ
s
0
, δ
1
, . . . , δ
t
0
. In the
case (B), we have
I
C
.(K
S
+ I
C
) = I
C
.((q
ν
−1)D
ν
+ (q
d
0
−1)D
d
0
+ (q
ξ
−1)D
ξ
+ (q
d
−1)D
d
+ F
)
= −1 + (q
d
0
−1)(−n
d
0
−m
d
0
−1) −1 −1 + q
d
0
= (e
0
−1)(−n
d
0
−m
d
0
) −2.
This value is clearly equal to minus two if e
0
= 1. In the case where e
0
≥ 2, since
n
d
0
= m
d
0
= 0 by deﬁnition, we have I
C
.(K
S
+I
C
) = −2. Lastly, we consider the
case (C). By computing, we have I
C
.(K
S
+I
C
) = (e
0
−1)(−n
d
0
−m
d
0
−n
ι
) −2.
This value is equal to minus two if e
0
= 1. Consider the case where e
0
≥ 2. Note
that m
d
0
= 0 in this case. Let us show that n
ι
= n
d
0
= 0. This is obvious if
y
ν
= 1 by deﬁnition. We assume y
ν
≥ 2, and prove the inequality ι > γ
s
0
. The
inequality ι ≥ γ
s
0
follows from Lemma 5.2. Suppose that x
γ
s
0
≥ −c
0
+ 2. Then
by (17), we have the inequality 1 = x
γ
s
0
+c
0
y
γ
s
0
≥ −c
0
+2+c
0
y
γ
s
0
, which implies
a contradiction c
0
(y
γ
s
0
−1) ≤ −1. Hence we have x
ι
= e
0
−c
0
≥ 2 − c
0
> x
γ
s
0
.
It follows that n
ι
= n
d
0
= 0 and I
C
.(K
S
+ I
C
) = −2.
Proposition 5.10. Assume that k ≥ 2 and C −F is base point free. Then the
divisor C −I
C
satisﬁes the properties (iv)–(vi) in Assertion 3.1.
Proof. (iv) We put H = C − I
C
. Since H has no base points, we can
write H ∼
d−1
i=2
h
i
D
i
with nonnegative integers h
i
. Then we have h
d−1
=
H.D
d
= C.D
d
−I
C
.D
d
= C.F −1 ≥ 1. By assumption, we have (C −F).D
d−1
=
THE GONALITY CONJECTURE FOR CURVES ON TORIC SURFACES WITH TWO P
1
FIBRATIONS 63
C.D
d−1
−1 ≥ 0. It follows that H
2
≥ H.D
d−1
= C.D
d−1
−I
C
.D
d−1
≥ 1. Recall
that H
2
(S, −I
C
) = 0 as mentioned at the beginning of the proof of Proposi
tion5.9. Then (v) and (vi) can be shown by the same argument as that in the
proof of Proposition 4.9.
We next aim to show the lemma similar to Lemma 4.10. Although it is ideal
that (C −2I
C
)
C
1
is nonspecial for a curve C
1
∈ C −I
C
 in any case, in fact, we
need a certain condition (see Lemma 5.12).
Lemma 5.11. Assume that k ≥ 2, f
0
≥ 1 and C−F is base point free. Besides,
we assume that C satisﬁes neither of the following conditions (a) nor (b) :
(a) y
ν
= 1, C.D
ν
= 1 and C.D
i
= 0 for any integer 2 ≤ i ≤ ν −1.
(b) y
ξ
= 1, C.D
ξ
= 1 and C.D
i
= 0 for any integer ξ ≤ i ≤ d −2.
Then there exists an eﬀective divisor E on S such that C−2I
C
−E is base point
free, h
0
(S, C −2I
C
−E) = h
0
(S, C −2I
C
) and E.(E + K
S
) ≤ 2E.(C −2I
C
).
Proof. We ﬁrst aim to prove the existence of an eﬀective divisor E
1
satis
fying the following properties (i)–(iii) :
(i) (C −2I
C
−E
1
).D
i
≥ 0 for 1 ≤ i ≤ d
0
and i = d,
(ii) h
0
(S, C −2I
C
−E
1
) = h
0
(S, C −2I
C
),
(iii) E
1
.(E
1
+ K
S
) ≤ 2E
1
.(C −2I
C
).
If C.D
ν
≥ 2, the zero divisor satisﬁes (i)–(iii). We thus assume that C.D
ν
= 1,
and deﬁne
ν
= max{i ≤ ν −1  C.D
i
≥ 1},
γ
= min{i ≥ ν + 1  (C −2I
C
).D
i
≥ 1} =
_
γ
1
(y
ν
≥ 2),
ι (y
ν
= 1).
By the information of the intersection numbers
(C −2I
C
).D
i
⎧
⎪
⎪
⎨
⎪
⎪
⎩
≥ 1 (i = ν
),
= −1 (i = ν),
≥ 1 (i = γ
),
= 0 (ν
+ 1 ≤ i ≤ ν −1, ν + 1 ≤ i ≤ γ
−1),
we can see the partial shape of
C−2I
C
around the line l
ν
(C − 2I
C
) (see Fig. 3
(1)).
Let us verify the following two claims (see Fig. 3 (2)) :
Claim1 : The intersection point P
0
= l
ν
−1
(C − 2I
C
) ∩ l
ν
(C − 2I
C
) lies on the
halfline L
1
= {R −a(y
γ
, −x
γ
)  a ≥ 0}.
Claim2 : The intersection point Q
0
= l
γ
(C − 2I
C
) ∩ l
γ
+1
(C − 2I
C
) lies on the
halfline L
2
= {R + a(y
ν
, −x
ν
)  a ≥ 0}.
If we put P = (p
1
, p
2
), then the Xcoordinate of R is greater than or equal to
p
1
−y
ν
(see Fig. 4). Since the Xcoordinate of P
0
is less than or equal to p
1
−y
ν
,
64 R. Kawaguchi
l
ν
(C −2I
C
)
l
ν
(C −2I
C
)
l
γ
(C −2I
C
)
Q = l
ν
(C −2I
C
) ∩ l
γ
(C −2I
C
)
P = l
ν
(C −2I
C
) ∩ l
ν
(C −2I
C
)
R = l
ν
(C −2I
C
) ∩ l
γ
(C −2I
C
)
L
1
L
2
l
ν −1
(C −2I
C
)
C−2I
C
l
γ +1
(C −2I
C
)
(1) (2)
Figure 3
L
1
l
ν −1
(C −2I
C
)
P
P
0
R
−x
ν
y
ν
Figure 4
Claim1 is true if y
ν
< y
ν
. Let us consider the case where y
ν
> y
ν
. It is suﬃcient
to show the inequality
x
γ
(y
ν
−y
ν
) ≤ y
γ
(x
ν
−x
ν
). (19)
In the case where y
ν
= 1, we have x
ν
= −c
0
and γ
= ι. On the other hand,
ν
≥ 2 since C does not satisﬁes the condition (a). We thus have
y
γ
(x
ν
−x
ν
) −x
γ
(y
ν
−y
ν
) = e
0
(y
ν
−1) −x
ν
−c
0
y
ν
> 0.
Assume that y
ν
≥ 2. Since γ
= γ
1
and x
γ
1
y
ν
−y
γ
1
x
ν
= 1 by (17), the inequality
(19) is obvious. We next show Claim2. Since
Q
0
= Q + (C −2I
C
).D
γ
(y
γ
, −x
γ
) = Q−2I
C
.D
γ
(y
γ
, −x
γ
),
it is suﬃcient for Claim2 to verify the inequality −2y
γ
I
C
.D
γ
≥ y
ν
. This is
obvious if y
ν
= 1. We thus consider the case where y
ν
≥ 2. Note that γ
= γ
1
in
this case. If s
0
= 1, then since (x
ν
, y
ν
) = (−c
0
, 1) + n
γ
1
(x
γ
1
, y
γ
1
), we have
THE GONALITY CONJECTURE FOR CURVES ON TORIC SURFACES WITH TWO P
1
FIBRATIONS 65
y
ν
= 1 + n
γ
1
y
γ
1
≤ 2n
γ
1
y
γ
1
≤ −2y
γ
1
I
C
.D
γ
1
.
If s
0
≥ 2, then we have x
γ
2
(y
ν
−n
γ
1
y
γ
1
) −y
γ
2
(x
ν
− n
γ
1
x
γ
1
) = 1 by (17). Hence
we can write (x
γ
1
, y
γ
1
) = ε(x
γ
2
, y
γ
2
) +ζ(x
ν
−n
γ
1
x
γ
1
, y
ν
−n
γ
1
y
γ
1
) with integers ε
and ζ. Since x
γ
2
y
γ
1
−y
γ
2
x
γ
1
= x
γ
1
y
ν
−y
γ
1
x
ν
= 1, we have ε = ζ = 1. It follows
that (n
γ
1
+1)y
γ
1
> y
ν
. Since the inequality I
C
.D
γ
1
≤ −n
γ
1
holds by List 5.7, we
have −2y
γ
1
I
C
.D
γ
1
> y
ν
. Hence Claim2 is true. Besides, we remark that
−I
C
.D
γ
≥ y
ν
(20)
if γ
= ι. This is obvious in the case where y
ν
= 1. If y
ν
≥ 2, then we have
γ
1
= ι, that is, y
γ
1
= 1. Hence the above two inequalities I
C
.D
γ
1
≤ −n
γ
1
and
(n
γ
1
+ 1)y
γ
1
> y
ν
implies that −I
C
.D
γ
≥ y
ν
.
We denote by Q
γ
= (z
γ
, w
γ
) the lattice point in l
γ
(C − 2I
C
) ∩
C−2I
C
which is closest to L
1
. Note that the line l
i
(C − 2I
C
) passes through the point
P (resp. Q) for any integer ν
+1 ≤ i ≤ ν −1 (resp. ν +1 ≤ i ≤ γ
−1). Thus we
can inductively deﬁne positive integers τ
i
for ν
+ 1 ≤ i ≤ γ
− 1 as follows (see
Fig. 5) :
τ
i
: the positive integer such that the line l
i
(C−2I
C
−τ
i
D
i
) passes through Q
i+1
,
Q
i
=(z
i
, w
i
): the lattice point in l
i
(C−2I
C
−τ
i
D
i
)∩
C−2I
C
which is closest to L
1
.
By deﬁnition, we have
l
γ
(C −2I
C
)
Q
γ
Q
γ −1
Q
γ −2
L
1
Figure 5
τ
i
=
_
x
i
(z
0
−z
i
+ y
ν
) + y
i
(w
0
−w
i
−x
ν
) (ν
+ 1 ≤ i ≤ ν −1),
x
i
(z
0
−z
i
) + y
i
(w
0
−w
i
) (ν ≤ i ≤ γ
−1),
(21)
where we set Q = (z
0
, w
0
). We put E
1
=
γ
−1
i=ν
+1
τ
i
D
i
. By deﬁnition, it
is obvious that E
1
satisﬁes the property (i). Besides, for an integer i with
ν
+ 1 ≤ i ≤ γ
− 1, the lattice points contained in the domain surrounded by
l
i−1
(C−2I
C
−E
1
), l
i
(C−2I
C
−E
1
) and L
1
must lie on the line l
i
(C−2I
C
−E
1
).
This means that E
1
satisﬁes the property (ii). For the later use, we note that
66 R. Kawaguchi
τ
γ
−1
≤ −I
C
.D
γ
(22)
holds if γ
= ι. Indeed, if we suppose that γ
= ι and τ
γ
−1
> −I
C
.D
γ
,
then the point Q
= Q + y
ν
(y
γ
, −x
γ
) lies on L
2
since the Xcoordinate of R
(resp. Q
) is less (resp. greater) than or equal to that of P. Namely, Q
lies on
l
γ
(C −2I
C
) ∩
C−2I
C
. On the other hand, by (20), we deduce that Q
is nearer
to L
1
than Q
γ
= Q + τ
γ
−1
(y
γ
, −x
γ
), which contradicts the deﬁnition of Q
γ
.
Let us show that E
1
satisﬁes (iii). By computing, we have (C − 2I
C
).E
1
=
τ
ν
(C − 2I
C
).D
ν
= −τ
ν
. Hence it is suﬃcient to verify E
1
.(E
1
+ K
S
) ≤ −2τ
ν
.
We ﬁrst consider the case where y
ν
≥ 2. Since γ
= γ
1
in this case, we have
x
γ
y
ν
− y
γ
x
ν
= 1 by (17). Hence the lattice point P − (y
ν
, −x
ν
) lies on the
halfline L
1
. We thus see that Q
ν
+1
= P −(y
ν
, −x
ν
) by the property (ii), which
implies that
τ
ν
+1
= x
ν
+1
(z
0
−z
ν
+1
+ y
ν
) + y
ν
+1
(w
0
−w
ν
+1
−x
ν
)
= x
ν
+1
y
ν
+ y
ν
+1
(−x
ν
) = 1.
On the other hand, we can write
Q
γ
= Q + s(y
γ
, −x
γ
),
Q
γ
−1
= Q
γ
+ t(y
γ
−1
, −x
γ
−1
),
Q
ν
= Q
γ
+ (a, b)
with integers s, t, a, and b. We note that ax
ν
+by
ν
≥ 0. Then, by (21), we have
τ
γ
−1
= x
γ
−1
(z
0
−z
γ
−1
) + y
γ
−1
(w
0
−w
γ
−1
)
= x
γ
−1
(−sy
γ
−ty
γ
−1
) + y
γ
−1
(sx
γ
+ tx
γ
−1
) = s,
τ
ν
= x
ν
(z
0
−z
ν
) + y
ν
(w
0
−w
ν
) = x
ν
(−sy
γ
−a) + y
ν
(sx
γ
−b)
= τ
γ
−1
−(ax
ν
+ by
ν
) ≤ τ
γ
−1
.
In sum, we obtain
E
1
.(E
1
+ K
S
) = E
1
.
_
−D
ν
+
γ
1
−1
i=ν
+1
(τ
i
−1)D
i
−D
γ
1
_
≤ E
1
.(−D
ν
+ (τ
ν
−1)D
ν
−D
γ
1
)
= −τ
ν
+1
+ (τ
ν
−1)E
1
.D
ν
−τ
γ
1
−1
≤ −τ
ν
+1
−τ
ν
+ 1 −τ
γ
1
−1
≤ −2τ
ν
.
Let us consider the case where y
ν
= 1. We note that x
ν
= −c
0
and γ
= ι
in this case. If (C − 2I
C
− E
1
).D
i
= 0 for integers ν
+ 1 ≤ i ≤ γ
− 1, then
we have Q
γ
= Q
ν
−1
. Namely, R must be a lattice point. Then, since the
Xcoordinate of R is greater than (resp. less than or equal to) that of Q (resp.
THE GONALITY CONJECTURE FOR CURVES ON TORIC SURFACES WITH TWO P
1
FIBRATIONS 67
P), we have ν
= 1, which contradicts the condition (a). Hence we can take an
integer κ = max{n  ν
+1 ≤ n ≤ γ
−1, (C −2I
C
−E
1
).D
n
≥ 1}. If κ > ν, then
by computing, we have
τ
κ
= x
κ
(−y
γ
) + y
κ
x
γ
= −x
κ
+ (e
0
−c
0
)y
κ
,
τ
ν
≤ x
ν
(−y
γ
+ y
κ
) + y
ν
(x
γ
−x
κ
) = −c
0
y
κ
+ e
0
−x
κ
≤ τ
κ
.
We thus have
E
1
.(E
1
+ K
S
) = E
1
.
_
−D
ν
+
γ
−1
i=ν
+1
(τ
i
−1)D
i
−D
γ
_
(23)
≤ −E
1
.D
ν
+ (τ
κ
−1)E
1
.D
κ
+ (τ
ν
−1)E
1
.D
ν
−E
1
.D
γ
≤ −τ
ν
+1
−τ
κ
−τ
ν
+ 2 −τ
γ
−1
≤ −τ
κ
−τ
ν
≤ −2τ
ν
.
In the case where κ = ν, we have E
1
.D
ν
≤ −2 and
E
1
.(E
1
+K
S
)≤−E
1
.D
ν
+(τ
ν
−1)E
1
.D
ν
−E
1
.D
γ
≤−τ
ν
+1
−2τ
ν
+2−τ
γ
−1
≤−2τ
ν
.
In the case where κ < ν, we have
τ
κ
= x
κ
(y
ν
−y
γ
) + y
κ
(−x
ν
+ x
γ
) = e
0
y
κ
,
τ
ν
≤ x
ν
(−y
γ
) + y
ν
x
γ
= e
0
≤ τ
κ
.
Hence the inequality (23) holds in this case also. Therefore, E
1
satisﬁes the
property (iii).
We deﬁne
δ
= max{i ≤ ξ −1  (C −2I
C
).D
i
≥ 1} =
_
δ
t
0
(y
ξ
≥ 2),
ι (y
ξ
= 1),
ξ
= min{i ≥ ξ + 1  C.D
i
≥ 1}.
Then, by a similar way to that in the case of E
1
, we can construct the eﬀective
divisor E
2
=
ξ
−1
i=δ
+1
ω
i
D
i
satisfying the following properties (i)
–(iii)
:
(i)
(C −2I
C
−E
2
).D
i
≥ 0 for d
0
≤ i ≤ d,
(ii)
h
0
(S, C −2I
C
−E
2
) = h
0
(S, C −2I
C
),
(iii)
E
2
.(E
2
+ K
S
) ≤ 2E
2
.(C −2I
C
).
Note that E
2
= 0 in the case where C.D
ξ
≥ 2. In sum, we obtain the eﬀective
divisors
E
1
=
⎧
⎪
⎨
⎪
⎩
0 (C.D
ν
≥ 2),
γ
−1
i=ν
+1
τ
i
D
i
(C.D
ν
= 1),
γ
=
_
γ
1
(y
ν
≥ 2),
ι (y
ν
= 1),
68 R. Kawaguchi
E
2
=
⎧
⎪
⎨
⎪
⎩
0 (C.D
ξ
≥ 2),
ξ
−1
i=δ
+1
ω
i
D
i
(C.D
ξ
= 1),
δ
=
_
δ
t
0
(y
ξ
≥ 2),
ι (y
ξ
= 1)
which satisfy the properties (i)–(iii) and (i)
–(iii)
.
We put E = E
1
+ E
2
. Let us check that (C − 2I
C
− E).D
i
≥ 0 holds for
any integer 1 ≤ i ≤ d. This is obvious if either E
1
or E
2
is equal to zero.
Hence we consider the case where C.D
ν
= C.D
ξ
= 1. By noting γ
≤ ι ≤ δ
,
(C−2I
C
−E).D
i
is nonnegative if i = ι. Since (C−2I
C
−E).D
ι
is nonnegative
if γ
< ι or δ
> ι, we consider the case where γ
= δ
= ι. Then we have
(C −2I
C
−E).D
ι
≥ −2I
C
.D
ι
−τ
γ
−1
−ω
δ
+1
.
Similarly to (22), one can show the inequality ω
δ
+1
≤ −I
C
.D
δ
in the case where
δ
= ι. We thus have (C − 2I
C
− E).D
ι
≥ 0. In sum, we can conclude that
BsC −2I
C
−E = ∅. Since
C−2I
C
\
C−2I
C
−E
=
C−2I
C
\ (
C−2I
C
−E
1
∩
C−2I
C
−E
1
)
= (
C−2I
C
\
C−2I
C
−E
1
) ∪ (
C−2I
C
\
C−2I
C
−E
2
),
the properties (ii) and (ii)
imply that (
C−2I
C
\
C−2I
C
−E
) ∩ Z = ∅. Namely,
we obtain h
0
(S, C −2I
C
−E) = h
0
(S, C −2I
C
). Lastly, we have
E.(E + K
S
) −2E.(C −2I
C
)
= E
1
.(E
1
+ K
S
) −2E
1
.(C −2I
C
) + E
2
.(E
2
+ K
S
) −2E
2
.(C −2I
C
) + 2E
1
.E
2
≤ 2E
1
.E
2
= 0
by the properties (iii) and (iii)
.
Lemma 5.12. Assume that k ≥ 2 and C − F is base point free. By Proposi
tion5.10, we can take a nonsingular irreducible curve C
1
∈ C − I
C
. If f
0
≥ 1
and C satisﬁes neither of the conditions (a) nor (b) in Lemma 5.11, then the
divisor (C
1
−I
C
)
C
1
is nonspecial.
Proof. Since C−F
1
 has no base points, we have (C−F
1
).D
1
= C.D
1
−1 ≥
0. Hence, by RiemannRoch theorem and List 5.7, we have
degC
1

C
1
−2g
1
= −C
1
.K
S
−2 ≥ C
1
.D
1
−2 = C.D
1
−2 ≥ −1,
where g
1
denotes the genus of C
1
. Hence also C
1

C
1
is nonspecial.
Assume that C satisﬁes neither of (a) nor (b). Since H
1
(S, −I
C
) =
H
2
(S, −I
C
) = 0, the cohomology exact sequence similar to (12) implies that
h
1
(C
1
, (C − 2I
C
)
C
1
) = h
1
(S, C − 2I
C
). We take the eﬀective divisor E as in
THE GONALITY CONJECTURE FOR CURVES ON TORIC SURFACES WITH TWO P
1
FIBRATIONS 69
Lemma 5.11. Then we have
h
1
(S, C −2I
C
)
= h
0
(S, C −2I
C
) +
1
2
(C −2I
C
).(K
S
−C + 2I
C
) −1
= h
0
(S, C −2I
C
−E) +
1
2
(C −2I
C
−E).(K
S
−C + 2I
C
+ E) −1
+
1
2
E.(E + K
S
) −E.(C −2I
C
)
= h
1
(S, C −2I
C
−E) +
1
2
E.(E + K
S
) −E.(C −2I
C
) ≤ 0.
We need Lemma 5.12 in the proof of Assertion 3.3. Hence we must show
Assertion 3.3 independently for the cases excluded in Lemma 5.12.
Lemma 5.13. Assume that k ≥ 2, f
0
≥ 1 and C−F is base point free. If either
of the conditions (a) or (b) in Lemma 5.11 holds, then h
0
(S, C −I
C
) −h
0
(S, C −
2I
C
) ≥ (C −I
C
).F + 2.
1
C.F
C.F
C
f
0
1
C−I
C
C.F −f
0
C−2I
C
1
P
Figure 6
Proof. We prove the case of (a). By Theorem2.4, the statement of the
lemma is equivalent to the inequality (
C−I
C
∩Z
2
) −(
C−2I
C
∩Z
2
) ≥ C.F +1.
Comparing the two lattice polytope
C
and
C−I
C
, by the deﬁnition of I
C
, we
see that the horizontal distance between l
i
(C) and l
i
(C−I
C
) (similarly, l
i
(C−I
C
)
and l
i
(C − 2I
C
)) is at least one for each integer d
0
+ 1 ≤ i ≤ d − 1 (see Fig. 6).
Indeed, if we write the linear equivalence class of F
as F
∼
d−1
d
0
+1
x
i
D
i
, then
the coeﬃcient of D
i
in the linear equivalence class of I
C
is at least x
i
for each
integer d
0
+ 1 ≤ i ≤ d − 1. By noting the existence of the point P, we see that
there exist at least C.F
− f
0
+ 2 lattice points in
C−I
C
\
C−2I
C
. Hence the
lemma is clear if f
0
= 1. Let us consider the case where f
0
≥ 2. In this case, we
have
C.F =
d−2
i=2
y
i
C.D
i
= 1 +
d−2
i=d
0
+1
y
i
C.D
i
,
70 R. Kawaguchi
C.F
=
d−1
i=d
0
+1
x
i
C.D
i
≥ C.D
d−1
+
x
ξ
y
ξ
d−2
i=d
0
+1
y
i
C.D
i
> 1 + (e
0
−1)(C.F −1)
≥ 1 + (f
0
−1)(C.F −1) = C.F + (f
0
−2)(C.F −1) ≥ C.F + f
0
−2.
Hence the lemma is true in this case also.
Lemma 5.14. Assume that k ≥ 2 and f
0
= 0. Then h
0
(S, C −I
C
) −h
0
(S, C −
2I
C
) ≥ (C −I
C
).F + 2.
Proof. Since I
C
= F
in this case, the vertical distance between l
i
(C)
and l
i
(C − I
C
) (similarly, l
i
(C − I
C
) and l
i
(C − 2I
C
)) is just one for each inte
ger d
0
+ 1 ≤ i ≤ d − 1. Hence, by a similar argument to that in the proof of
Lemma 5.13, we have (
C−I
C
∩ Z
2
) −(
C−2I
C
∩ Z
2
) ≥ C.F + 1.
5.2 Proof of Assertion 3.2 and 3.3
In this subsection, we will prove Assertion 3.2 and 3.3. Let D be a divisor on
S such that D.D
1
≥ 1. For an integer x with 0 ≤ x ≤ D.F
, we deﬁne
i(D, x) = max
_
j ≥ 1
¸
¸
¸ x < −
d
0
−1
i=j
x
i
D.D
i
_
,
y(D, x) =
d
0
−1
i=i(D,x)+1
y
i
D.D
i
−
y
i(D,x)
x
i(D,x)
_
x +
d
0
−1
i=i(D,x)+1
x
i
D.D
i
_
.
Let C be a curve on S such as at the beginning of this section and I
C
the aux
iliary divisor of C. In this subsection, we write the linear equivalence class of F
as F
∼
d−1
d
0
+1
x
i
D
i
. Then I
C
is written as
I
C
=
d
0
i=0
q
i
D
i
+
d−1
d
0
+1
(q
i
+ x
i
)D
i
,
where we formally set q
0
= q
d
and D
0
= D
d
.
Remark 5.15. By deﬁnition, we have y(C, 0) = 0. Besides, y(C, x) = −y
ν
x/x
ν
if i(C, x) = ν.
Lemma 5.16. Assume that k ≥ 2 and C.D
1
≥ 1. Let x be an integer with
f
0
≤ x ≤ C.F
, and deﬁne q = min{n ∈ Z  n > y(C −I
C
, x −f
0
)}. Then
y(C −I
C
, x −f
0
) + n
d
0
≤ y(C, x) < q + n
d
0
(c
0
> e
0
),
y(C −I
C
, x −f
0
) + n
d
0
+ 1 ≤ y(C, x) < q + n
d
0
+ 1 (c
0
< e
0
).
Proof. We prove only the former case. The latter case can be proved
by a similar procedure. Since c
0
> e
0
, we have ν, ι ≤ d
0
− 1. Let us compute
d
0
−1
i=i(C,x)
x
i
I
C
.D
i
. Recall (C) in List 5.7. If ι = γ
s
0
, then by noting x
γ
s
0
= e
0
−c
0
,
THE GONALITY CONJECTURE FOR CURVES ON TORIC SURFACES WITH TWO P
1
FIBRATIONS 71
we have
d
0
−1
i=i(C,x)
x
i
I
C
.D
i
= x
ν
−n
γ
1
x
γ
1
−· · · −n
γ
s
0
−1
x
γ
s
0
−1
−(n
γ
s
0
+ 1)x
γ
s
0
= −c
0
−x
γ
s
0
= −e
0
.
If ι ≥ γ
s
0
+ 1, then by noting n
ι
= 0, we have
d
0
−1
i=i(C,x)
x
i
I
C
.D
i
= x
ν
−n
γ
1
x
γ
1
−· · · −n
γ
s
0
x
γ
s
0
−x
ι
= −c
0
−x
ι
= −e
0
.
We thus obtain
d
0
−1
i=i(C,x)+1
x
i
I
C
.D
i
=
_
−e
0
−x
ν
(i(C, x) = ν),
−e
0
(i(C, x) ≤ ν −1).
(24)
Similarly, we can obtain
d
0
−1
i=i(C,x)+1
y
i
I
C
.D
i
=
_
n
d
0
−y
ν
(i(C, x) = ν),
n
d
0
(i(C, x) ≤ ν −1).
(i) Consider the case where x ≥ −x
ν
. If i(C, x) = ν, then we have
y(C −I
C
, x −e
0
) = −
d
0
−1
i=ν+1
y
i
I
C
.D
i
+
y
ν
x
ν
_
x −e
0
−
d
0
−1
i=ν+1
x
i
I
C
.D
i
_
= y
ν
−n
d
0
−
y
ν
x
ν
x −y
ν
= y(C, x) −n
d
0
.
On the other hand, if i(C, x) ≤ ν −1, then we have
−
d
0
−1
i=i(C,x)+1
x
i
C.D
i
≤ x < −
d
0
−1
i=i(C,x)
x
i
C.D
i
−
d
0
−1
i=i(C,x)+1
x
i
(C −I
C
).D
i
≤ x −e
0
< −
d
0
−1
i=i(C,x)
x
i
(C −I
C
).D
i
by (24). This means that i(C −I
C
, x −e
0
) = i(C, x). Hence
y(C −I
C
, x −e
0
)
=
d
0
−1
i=i(C,x)+1
y
i
(C −I
C
).D
i
−
y
i(C,x)
x
i(C,x)
_
x −e
0
+
d
0
−1
i=i(C,x)+1
x
i
(C −I
C
).D
i
_
=
d
0
−1
i=i(C,x)+1
y
i
C.D
i
−n
d
0
−
y
i(C,x)
x
i(C,x)
_
x +
d
0
−1
i=i(C,x)+1
x
i
C.D
i
_
= y(C, x) −n
d
0
.
72 R. Kawaguchi
Consequently, our lemma is true in any case.
(ii) We next consider the case where x < −x
ν
. Note that i(C, x) = ν and
y(C, x) = −y
ν
x/x
ν
in this case. Since C.D
i
= 0 for i(C, x) + 1 ≤ i ≤ d
0
− 1, we
have
x −e
0
< −x
ν
−e
0
=
d
0
−1
i=i(C,x)+1
x
i
I
C
.D
i
= −
d
0
−1
i=i(C,x)+1
x
i
(C −I
C
).D
i
by (24). Hence there exists an integer s with 1 ≤ s ≤ s
0
such that
i(C −I
C
, x −e
0
) = γ
s
. Then
y(C −I
C
, x −e
0
) =
d
0
−1
i=γ
s
+1
y
i
(C −I
C
).D
i
−
y
γ
s
x
γ
s
_
x −e
0
+
d
0
−1
i=γ
s
+1
x
i
(C −I
C
).D
i
_
=
d
0
−1
i=γ
s
y
i
(C −I
C
).D
i
−
y
γ
s
x
γ
s
_
x −e
0
+
d
0
−1
i=γ
s
x
i
(C −I
C
).D
i
_
≤
d
0
−1
i=γ
s
y
i
(C −I
C
).D
i
−
y
ν
x
ν
_
x −e
0
+
d
0
−1
i=γ
s
x
i
(C −I
C
).D
i
_
= −
d
0
−1
i=γ
s
y
i
I
C
.D
i
−
y
ν
x
ν
_
x −e
0
−
d
0
−1
i=γ
s
x
i
I
C
.D
i
_
= −n
d
0
+ y
ν
+
γ
s
−1
i=ν+1
y
i
I
C
.D
i
−
y
ν
x
ν
_
x + x
ν
+
γ
s
−1
i=ν+1
x
i
I
C
.D
i
_
= −n
d
0
−
y
ν
x
ν
x +
γ
s
−1
i=ν+1
_
y
i
−
y
ν
x
ν
x
i
_
I
C
.D
i
≤ −n
d
0
−
y
ν
x
ν
x = y(C, x) −n
d
0
.
We next show the inequality y(C, x) < q +n
d
0
. Suppose that q +n
d
0
= y(C, x) =
−y
ν
x/x
ν
. Note that q + n
d
0
is the integer. Then we have x = 0 by the in
equality x < −x
ν
and the fact that x
ν
and y
ν
are relatively prime. This implies
that q is nonpositive, a contradiction. Suppose that q + n
d
0
< −y
ν
x/x
ν
. Since
y
ν
/x
ν
> y
γ
s
/x
γ
s
, we have
−(q + n
d
0
)x
γ
s
−xy
γ
s
≤ −1 = −x
γ
s
_
1 +
d
0
i=γ
s
+1
n
i
y
i
_
+ y
γ
s
_
−c
0
+
d
0
i=γ
s
+1
n
i
x
i
_
THE GONALITY CONJECTURE FOR CURVES ON TORIC SURFACES WITH TWO P
1
FIBRATIONS 73
q ≤ 1 +
d
0
i=γ
s
+1
n
i
y
i
−n
d
0
−
y
γ
s
x
γ
s
_
x −c
0
+
d
0
i=γ
s
+1
n
i
x
i
_
= 1 +
d
0
−1
i=γ
s
+1
n
i
y
i
−
y
γ
s
x
γ
s
_
x −c
0
+
d
0
−1
i=γ
s
+1
n
i
x
i
_
· · · (∗).
If ι ≥ γ
s
+ 1, then by List 5.7, we have
(∗) = −
d
0
−1
i=γ
s
+1
y
i
I
C
.D
i
−
y
γ
s
x
γ
s
_
x −e
0
−
d
0
−1
i=γ
s
+1
x
i
I
C
.D
i
_
=
d
0
−1
i=γ
s
+1
y
i
(C −I
C
).D
i
−
y
γ
s
x
γ
s
_
x −e
0
+
d
0
−1
i=γ
s
+1
x
i
(C −I
C
).D
i
_
= x(C −I
C
, x −e
0
).
If ι = γ
s
, then since (C −I
C
).D
i
= 0 for γ
s
+ 1 ≤ i ≤ d
0
−1, we have
(∗) = 1 −
y
γ
s
x
γ
s
(x −c
0
) = −
y
γ
s
x
γ
s
(x −e
0
)
=
d
0
−1
i=γ
s
+1
y
i
(C −I
C
).D
i
−
y
γ
s
x
γ
s
_
x −e
0
+
d
0
−1
i=γ
s
+1
x
i
(C −I
C
).D
i
_
= x(C −I
C
, x −e
0
).
Obviously, they contradict the deﬁnition of q. Consequently, we can conclude
that y(C, x) < q + n
d
0
.
We deﬁne i
(D, x
) and y
(D, x
) in a way similar to i(D, x) and y(D, x),
respectively. Concretely, in the case where D.D
d−1
≥ 1, for an integer x
with
0 ≤ x
≤ D.F
, we deﬁne
i
(D, x
) = min
_
j ≤ d −1
¸
¸
¸ x
<
j
i=d
0
+1
x
i
D.D
i
_
,
y
(D, x
) =
i
(D,x
)−1
i=d
0
+1
y
i
D.D
i
+
y
i
(D,x
)
x
i
(D,x
)
_
x
−
i
(D,x
)−1
i=d
0
+1
x
i
D.D
i
_
.
Then we can obtain the following lemma by an argument similar to that in the
proof of Lemma 5.16.
Lemma 5.17. Assume that k ≥ 2 and C.D
d−1
≥ 1. Let x
be an integer with
f
0
≤ x
≤ C.F
, and deﬁne q
= min{n ∈ Z  n > y
(C −I
C
, x
−f
0
)}. Then
y
(C −I
C
, x
−f
0
) + m
d
0
+ 1 ≤ y
(C, x
) < q
+ m
d
0
+ 1 (c
0
> e
0
),
y
(C −I
C
, x
−f
0
) + m
d
0
≤ y
(C, x
) < q
+ m
d
0
(c
0
< e
0
).
74 R. Kawaguchi
Similar to Section4, we next consider the operation to take auxiliary divisors
repeatedly. We put C
0
= C and I
0
= I
C
, and take a nonsingular irreducible
curve C
1
∈ C
0
−I
0
. For an integer n ≥ 2, if C
j
.F ≥ 2, C
j
.F
≥ 2 and
ν
j
= max{i ≤ d
0
 C
j
.D
i
≥ 1} ≥ 2,
ξ
j
= min{i ≥ d
0
 C
j
.D
i
≥ 1} ≤ d −2
for any integer j with 1 ≤ j ≤ n −1, then we can take the auxiliary divisor I
n−1
of C
n−1
and a nonsingular irreducible curve C
n
∈ C
n−1
−I
n−1
 inductively. Note
that n ≤ C.F −1 by the condition C
n−1
.F ≥ 2. We deﬁne
c
j
= −max{c ∈ Z  x
ν
j
−cy
ν
j
≥ 0},
e
j
= min{e ∈ Z  x
ξ
j
−ey
ξ
j
≤ 0},
f
j
= min{c
j
, e
j
},
α
j
: the cardinality of the set {l  0 ≤ l ≤ j, c
l
< e
l
}
for an integer j with 0 ≤ j ≤ n − 1. We divide each (x
ν
j
, y
ν
j
) and (x
ξ
j
, y
ξ
j
) in
ways similar to (16) as
(x
ν
j
, y
ν
j
) = (−c
j
, 1) +
d
0
i=ν
j
+1
n
j
i
(x
i
, y
i
),
(x
ξ
j
, y
ξ
j
) = (e
j
, 1) +
ξ
j
−1
i=d
0
m
j
i
(x
i
, y
i
).
In the following Lemma 5.18–5.20, we assume that we can take nonsingular irre
ducible curves C
1
, . . . , C
n
in the above way.
Lemma 5.18. Assume that C.D
1
≥ 1. Let x be an integer with x ≤
−
d
0
−1
i=1
x
i
C
0
.D
i
. If x ≥ f
0
+ · · · + f
n−1
and C
j
.D
d
0
= 0 for j = 0, . . . , n,
then y(C
n
, x −f
0
−· · · −f
n−1
) + α
n−1
≤ y(C
0
, x) and there is no integer in the
halfopen interval
_
y(C
n
, x −f
0
−· · · −f
n−1
) + α
n−1
, y(C
0
, x)
¸
.
Proof. Let j be an integer with 0 ≤ j ≤ n − 1. By assumption, we have
c
j
= e
j
and n
j
d
0
= m
j
d
0
= 0. Besides, by computing, we have
d
0
−1
i=1
x
i
C
j
.D
i
=
d
0
−1
i=1
x
i
C
0
.D
i
+ f
0
+· · · + f
j−1
,
where we deﬁne f
0
+ · · · + f
j−1
= 0 in the case where j = 0. We thus have
f
j
≤ x−f
0
−· · ·−f
j−1
≤ −
d
0
−1
i=1
x
i
C
j
.D
i
. In the case where c
j
> e
j
, Lemma 5.16
implies that y(C
j+1
, x−f
0
−· · ·−f
j
) ≤ y(C
j
, x−f
0
−· · ·−f
j−1
) and there exists no
integer in the halfopen interval
_
y(C
j+1
, x−f
0
−· · ·−f
j
), y(C
j
, x−f
0
−· · ·−f
j−1
)
¸
.
THE GONALITY CONJECTURE FOR CURVES ON TORIC SURFACES WITH TWO P
1
FIBRATIONS 75
Similarly, if c
j
< e
j
, then there exists no integer in the halfopen interval
_
y(C
j+1
, x −f
0
−· · · −f
j
) +1, y(C
j
, x −f
0
−· · · −f
j−1
)
¸
. Therefore, our lemma
is clear.
A similar lemma holds for the opposite side of the lattice polytope.
Lemma 5.19. Assume that C.D
d−1
≥ 1. Let x
be an integer with x
≤
d−1
i=d
0
+1
x
i
C
0
.D
i
. If x
≥ f
0
+ · · · + f
n−1
and C
j
.D
d
0
= 0 for j = 0, . . . , n,
then y
(C
n
, x
−f
0
−· · · −f
n−1
) + n −α
n−1
≤ y
(C
0
, x
) and there is no integer
in the halfopen interval
_
y
(C
n
, x
−f
0
−· · · −f
n−1
) + n −α
n−1
, y
(C
0
, x
)
¸
.
Lemma 5.20. Assume that k
− k = f
0
− 1 ≥ 1 and C − F is base point free.
If n ≤ k −2 and f
1
= · · · = f
n−1
= 1, then ν
n
≥ 2 and ξ
n
≤ d −2.
Proof. An easy computation shows that C
n−1
.F
= C.F
−f
0
−· · ·−f
n−2
=
C.F − n + 1 = C
n−1
.F. Suppose that ν
n
= 1. Since C
n
.D
i
= 0 for any inte
ger 2 ≤ i ≤ d
0
, by List 5.7, C
n−1
must satisfy the properties c
n−1
< e
n−1
,
y
ν
n−1
= 1, C
n−1
.D
ν
n−1
= 1 and C
n−1
.D
i
= 0 for any integer 2 ≤ i ≤ ν
n−1
− 1.
Since e
n−1
≥ 2, by the same computation as that at the last of the proof of
Lemma 5.13, we have C
n−1
.F
> C
n−1
.F, a contradiction. One can induce the
same contradiction under the assumption that ξ
n
= d −1.
Let us consider Assertion 3.2 with respect to a special case :
Lemma 5.21. Assume k
= k+1 ≥ 3, f
0
= 2 and C−F is base point free. Then
x
ν
0
≤ −2, x
ξ
0
≥ 2 and there exists an integer m with 2 ≤ m ≤ min{−x
ν
0
, x
ξ
0
}
such that f
1
= · · · = f
m−2
= 1 and f
m−1
= 0.
Proof. Since f
0
= 2, we have C.D
d
0
= 0, x
ν
0
< −y
ν
0
and x
ξ
0
> y
ξ
0
. It
follows that x
ν
0
≤ −2 and x
ξ
0
≥ 2. As we saw in the proof of Proposition 5.10
and Lemma 5.12, C.D
1
and C.D
d−1
are positive. By the inequality
k =
d−2
i=2
y
i
C.D
i
(25)
=
d
0
−1
i=2
y
i
C.D
i
+
d−2
i=d
0
+1
y
i
C.D
i
≤
y
ν
0
x
ν
0
d
0
−1
i=2
x
i
C.D
i
+
y
ξ
0
x
ξ
0
d−2
i=d
0
+1
x
i
C.D
i
=
y
ν
0
x
ν
0
(−C.F
+ C.D
1
) +
y
ξ
0
x
ξ
0
(C.F
−C.D
d−1
) ≤
_
−
y
ν
0
x
ν
0
+
y
ξ
0
x
ξ
0
_
(k
−1),
we obtain
−
y
ν
0
x
ν
0
+
y
ξ
0
x
ξ
0
≥ 1. (26)
Consider the case where min{−x
ν
0
, x
ξ
0
} = 2. If x
ν
0
= −2, we have
y
ν
0
= 1, that is, c
0
= 2 by the assumption f
0
= 2. Hence, by (26), we have
76 R. Kawaguchi
1/2 ≤ y
ξ
0
/x
ξ
0
< 1/(e
0
−1), that is, e
0
= 2. Similarly, one can show c
0
= e
0
= 2
in the case where x
ξ
0
= 2. Therefore, by (B) in List 5.7, we have C
1
.D
d
0
≥ 1 in
this case. Hence the lemma is valid for m = 2.
We assume that m
0
= min{−x
ν
0
, x
ξ
0
} ≥ 3. Let us consider the following
operation:
We can take C
1
. →
⎧
⎪
⎨
⎪
⎩
f
1
= 0. → We can ﬁnish our proof.
f
1
= 1. → We can take C
2
by Lemma 5.20 if k ≥ 3. −−
f
1
≥ 2.
→
⎧
⎪
⎨
⎪
⎩
f
2
= 0. → We can ﬁnish our proof.
f
2
= 1. → We can take C
2
by Lemma 5.20 if k ≥ 4. −−
f
2
≥ 2.
→ · · · .
Then, since k = k
−1 ≥ −x
1
C.D
1
−x
ν
0
C.D
ν
0
−1 ≥ m
0
, it is suﬃcient to check
that
(i) f
j
≤ 1 if f
1
= · · · = f
j−1
= 1 for an integer j with 1 ≤ j ≤ m
0
−1,
(ii) especially f
m
0
−1
= 0 if f
1
= · · · = f
m
0
−2
= 1.
(i) Let j be an integer with 1 ≤ j ≤ m
0
− 1 such that f
1
= · · · = f
j−1
= 1. By
Lemma 5.20, we can take the auxiliary divisor I
j−1
of C
j−1
and a nonsingular
irreducible curve C
j
∈ C
j−1
− I
j−1
. Suppose that f
j
≥ 2. Since x
ν
j
≤ −2
and x
ξ
j
≥ 2, we have y(C
j
, 1) = −y
ν
j
/x
ν
j
and y
(C
j
, 1) = y
ξ
j
/x
ξ
j
. Hence, by
Lemma 5.18 and 5.19,
y(C
0
, j + 2) = −
y
ν
0
x
ν
0
(j + 2) < min{l ∈ Z  l > y(C
j
, 1) + α
j−1
}
= min
_
l ∈ Z
¸
¸
¸ l > −
y
ν
j
x
ν
j
+ α
j−1
_
= α
j−1
+ 1,
y
(C
0
, j + 2) =
y
ξ
0
x
ξ
0
(j + 2) < min{l ∈ Z  l > y
(C
j
, 1) + j −α
j−1
}
= min
_
l ∈ Z
¸
¸
¸ l >
y
ξ
j
x
ξ
j
+ j −α
j−1
_
= j −α
j−1
+ 1.
These inequalities imply that −y
ν
0
/x
ν
0
+ y
ξ
0
/x
ξ
0
< 1, a contradiction. We thus
obtain f
j
≤ 1.
(ii) Considering y(C
0
, m
0
− 1), y(C
m
0
−2
, 0), y
(C
0
, m
0
− 1) and y
(C
m
0
−2
, 0),
Lemma 5.18 and 5.19 implies that
_
α
m
0
−3
, −
y
ν
0
x
ν
0
(m
0
−1)
_
∩ Z = ∅, (27)
_
m
0
−2 −α
m
0
−3
,
y
ξ
0
x
ξ
0
(m
0
−1)
_
∩ Z = ∅.
THE GONALITY CONJECTURE FOR CURVES ON TORIC SURFACES WITH TWO P
1
FIBRATIONS 77
Note that x
ν
0
< −y
ν
0
and x
ξ
0
> y
ξ
0
by the assumption f
0
= 2. In the case where
m
0
= −x
ν
0
, we obtain α
m
0
−3
= y
ν
0
−1 by the ﬁrst equality of (27). Since
(y(C
m
0
−2
, 1) + α
m
0
−3
, y(C
0
, m
0
)] ∩ Z =
_
−
y
ν
m
0
−2
x
ν
m
0
−2
+ y
ν
0
−1, y
ν
0
_
∩ Z = ∅,
we have y
ν
m
0
−2
/x
ν
m
0
−2
= −1, that is, c
m
0
−2
= 1. On the other hand, since
(y
(C
m
0
−2
, 1) + m
0
−2 −α
m
0
−3
, y
(C
0
, m
0
)] ∩ Z
=
_
y
ξ
m
0
−2
x
ξ
m
0
−2
−x
ν
0
−y
ν
0
−1, −
y
ξ
0
x
ξ
0
x
ν
0
_
∩ Z = ∅,
we have y
ξ
m
0
−2
/x
ξ
m
0
−2
≥ 1. Indeed, if y
ξ
m
0
−2
/x
ξ
m
0
−2
< 1, then we have
−y
ξ
0
x
ν
0
/x
ξ
0
< −x
ν
0
−y
ν
0
, which contradicts (26). Hence we obtain e
m
0
−2
= 1.
Therefore, we can conclude that C
m
0
−1
.D
d
0
≥ 1 by (B) in List 5.7, which means
that f
m
0
−1
= 0. In the case where m
0
= x
ξ
0
, we obtain α
m
0
−3
= x
ξ
0
− y
ξ
0
− 1
by the second equality of (27). Since
(y(C
m
0
−2
, 1)+α
m
0
−3
, y(C
0
, m
0
)]∩Z =
_
−
y
ν
m
0
−2
x
ν
m
0
−2
+x
ξ
0
−y
ξ
0
−1, −
y
ν
0
x
ν
0
x
ξ
0
_
∩Z = ∅,
we have y
ν
m
0
−2
/x
ν
m
0
−2
≥ −1. Indeed, if y
ν
m
0
−2
/x
ν
m
0
−2
< −1, then we have
−y
ν
0
x
ξ
0
/x
ν
0
< x
ξ
0
− y
ξ
0
, which contradicts (26). We thus have c
m
0
−2
= 1. On
the other hand, since
(y
(C
m
0
−2
, 1)+m
0
−2−α
m
0
−3
, y
(C
0
, m
0
)] ∩Z =
_
y
ξ
m
0
−2
x
ξ
m
0
−2
+y
ξ
0
−1, y
ξ
0
_
∩Z = ∅,
we have y
ξ
m
0
−2
/x
ξ
m
0
−2
= 1, that is, e
m
0
−2
= 1. Consequently, we have f
m
0
−1
= 0
in this case also.
In keeping with the Lemma 5.21, let us show Assertion 3.2.
Proposition 5.22. For the curve C, Assertion 3.2 is true.
Proof. First, if k = 2 and C
1
.F
> C
1
.F (resp. C
1
.F
= C
1
.F), then the
statement is obviously true for m
0
= 1 (resp. m
0
= 2). Hence we assume that
k ≥ 3.
Consider the case where f
0
≤ 1. Since
C
1
.F
−C
1
.F = C.F
−I
0
.F
−C.F + I
0
.F = k
−f
0
−k + 1 ≥ 2 −f
0
≥ 1,
the lemma is valid for m
0
= 1 in this case.
Let us consider the case where f
0
= 2. If k
≥ k + 2, then we have
C
1
.F
≥ C
1
.F + 1, and can ﬁnish the proof. On the other hand, in the case
where k
= k + 1, the lemma is valid for m
0
= m, where m is the integer in
78 R. Kawaguchi
Lemma 5.21. Indeed, this is clear in the case where C
m−1
.F = 1. If C
m−1
.F ≥ 2,
then we can take the auxiliary divisor I
m−1
of C
m−1
and a nonsingular irre
ducible curve C
m
∈ C
m−1
− I
m−1
 by Lemma 5.20. Since f
m−1
= 0, we obtain
C
m
.F
≥ C
m
.F + 1.
From now on, we assume that f
0
≥ 3. Since x
ν
0
< −2y
ν
0
and x
ξ
0
> 2y
ξ
0
,
we have k
≥ k +2 by the same computation as (25). We put a =
d
0
−1
i=2
y
i
C.D
i
and b =
d−2
i=d
0
+1
y
i
C.D
i
.
(i) Consider the case where f
0
≥ 3 and k
= k + 2. Then we have
a ≤
y
ν
0
x
ν
0
d
0
−1
i=2
x
i
C.D
i
< −
1
c
0
−1
d
0
−1
i=2
x
i
C.D
i
=
k
−C.D
1
c
0
−1
=
k + 2 −C.D
1
c
0
−1
,
b ≤
y
ξ
0
x
ξ
0
d−2
i=d
0
+1
x
i
C.D
i
<
1
e
0
−1
d
0
−1
i=2
x
i
C.D
i
=
k
−C.D
d−1
e
0
−1
=
k + 2 −C.D
d−1
e
0
−1
.
Here we note that C.D
1
= 1. Indeed, if C.D
1
≥ 2 and k is even (resp. odd), then
we have a < k/2 and b ≤ k/2 (resp. a ≤ (k − 1)/2 and b < (k + 1)/2), which
contradicts the equality k = a + b. Similarly, one can obtain C.D
d−1
= 1. If
c
0
≥ 4, then we have
a
⎧
⎨
⎩
≤ 1 (k = 3, 4),
<
k + 1
3
(k ≥ 5),
which contradicts that k = a + b and b ≤ (k + 1)/2. We thus have c
0
= 3.
Similarly, one can obtain e
0
= 3. In sum, we have f
1
= 0 by (B) in List 5.7, and
moreover
C
1
.F
= C.F
−I
0
.F
= C.F + 2 −f
0
= C.F −1 = C
1
.F.
Since C
1
.F = k−1 ≥ 2, we can take the auxiliary divisor I
1
of C
1
and a nonsingu
lar irreducible curve C
2
∈ C
1
−I
1
 by Lemma 5.20. Then we obtain C
2
.F
> C
2
.F
by f
1
= 0. Therefore, the lemma is valid for m
0
= 2 in this case.
(ii) Consider the case where f
0
≥ 3 and k
≥ k + 3. If f
0
= 3, then we have
C
1
.F
= C.F
−3 ≥ C.F = C
1
.F + 1.
Assume that f
0
≥ 4 and a ≥ b. Then we have
C
1
.F = C.F −1 = a + b −1 ≤ 2a −1
= 2
ν
0
−1
i=2
y
i
C.D
i
+ 2y
ν
0
(C.D
ν
0
−1) + 2y
ν
0
−1,
C
1
.F
= C.F
−f
0
THE GONALITY CONJECTURE FOR CURVES ON TORIC SURFACES WITH TWO P
1
FIBRATIONS 79
= −
d
0
−1
i=1
x
i
C.D
i
−f
0
≥ C.D
1
−
x
ν
0
y
ν
0
a −f
0
> 1 + (f
0
−1)a −f
0
= (f
0
−1)
ν
0
−1
i=2
y
i
C.D
i
+ (f
0
−1)y
ν
0
(C.D
ν
0
−1) + (f
0
−1)(y
ν
0
−1).
We thus obtain C
1
.F
> C
1
.F if y
ν
0
≥ 2. If y
ν
0
= 1, then by noting k ≥ 3, we have
the inequality
ν
0
i=2
y
i
C.D
i
≥ 2. Hence, in this case, either
ν
0
−1
i=2
y
i
C.D
i
≥ 1
or C.D
ν
0
≥ 2 holds, which implies that C
1
.F
> C
1
.F. Similarly, one can obtain
the same inequality in the case where f
0
≥ 4 and a ≤ b. Therefore, the lemma is
valid for m
0
= 1 in this case.
Proposition 5.23. For the curve C, Assertion 3.3 is true.
Proof. Let j be an integer with 1 ≤ j ≤ min{m
0
, k−1}. If f
j−1
= 0 (resp.
f
j−1
≥ 1 and C
j−1
satisﬁes (a) or (b) in Lemma 5.11), then the statement follows
from Lemma 5.14 (resp. Lemma 5.13).
Consider the case where f
j−1
≥ 1 and C
j−1
satisﬁes neither (a) nor (b).
Note that H
1
(S, −I
j−1
) = 0 by Proposition 5.9 and (C
j
−I
j−1
)
C
j
is nonspecial
by Lemma 5.12. Besides, a similar computation to (14) shows that also C
j

C
j
is nonspecial. Hence we see that the statement of our lemma is equivalent to
the inequality C
j
.I
j−1
≥ C
j
.F + 1 by the same argument as that in the proof of
Proposition 4.20. Let ι
j−1
be an integer such that (x
ι
j−1
, y
ι
j−1
) = (e
j−1
−c
j−1
, 1).
For the auxiliary divisor I
j−1
=
d
i=1
q
i
D
i
+F
, we have q
ι
j−1
= f
j−1
≥ 1. Since
C
j
.D
ι
j−1
≥ 1 by List 5.7, we have
C
j
.I
j−1
≥ C
j
.(q
ι
j−1
D
ι
j−1
+ F
) ≥ C
j
.F
+ 1 ≥ C
j
.F + 1.
The above argument all goes through for j = 1 if C.F
= C.F = 2.
References
[ 1 ] M. Aprodu, On the vanishing of higher syzygies of curves, Math. Z. 241 (2002), 1–15.
[ 2 ] M. Coppens and T. Kato, The gonality of smooth curves with plane models,
Manuscripta Math. 70 (1990), 5–25.
[ 3 ] M. Coppens and T. Kato, Correction to: “The gonality of smooth curves with plane
models”, Manuscripta Math. 71 (1991), 337–338.
[ 4 ] S. Ehbauer, Syzygies of points in projective space and applications, Zerodimensional
schemes, Proceedings to the International conference (Ravello, 1992), de Gruyter, Berlin
(1994), 145–170.
[ 5 ] M. Green, Koszul cohomology and the geometry of projective varieties, J. Diﬀ. Geom.
19 (1984), 125–171.
[ 6 ] M. Green and R. Lazarsfeld, The nonvanishing of certain Koszul cohomology groups,
J. Diﬀ. Geom. 19 (1984), 168–170. (Appendix to [5])
[ 7 ] M. Green and R. Lazarsfeld, On the projective normality of complete linear series on
80 R. Kawaguchi
an algebraic curve, Invent. Math. 83 (1986), 73–90.
[ 8 ] R. Kawaguchi, The gonality conjecture for curves on certain toric surfaces, Osaka J.
Math. 45 (2008), 113126.
[ 9 ] G. Martens, The gonality of curves on a Hirzebruch surface, Arch. Math. 64 (1996),
349–452.
[10] M. Mustat¸ˇ a, Vanishing Theorems on toric varieties, Tohoku Math. J. 54 (2002), 451–
470.
[11] M. Namba, Geometry of projective algebraic curves, Monographs and Textbooks in Pure
and Appl. Math. 88, Marcel Dekker, New York, 1984.
[12] T. Oda, Convex bodies and algebraic geometry, SpringerVerlag, Berlin, 1988.
[13] M. Ohkouchi and F. Sakai, The gonality of singular plane curves, Tokyo J. Math. 27
(2004), 137–147.
Department of Mathematics
Graduate School of Science
Osaka University
Toyonaka, Osaka 560–0043, Japan
email: kawaguchir@cr.math.sci.osakau.ac.jp