Professional Documents
Culture Documents
Abstract. We prove that the Cox ring of the blowing-up of a minimal toric
surface of Picard rank two is finitely generated. As part of our proof of this
result we provide a necessary and sufficient condition for finite generation of
Cox rings of normal projective Q-factorial surfaces.
Introduction
Given a complete toric variety P and a general point e ∈ P, it is an open problem
to decide when the Cox ring of the blowing-up Ble P is finitely generated. For
example when P is a (fake) weighted projective plane, the problem of deciding
non finite generation of the Cox ring can be related to the Nagata conjecture for
plane curves [4]. More recently examples of surfaces Ble P with non-polyhedral
pseudoeffective cone have been given in [3] and used to show that the pseudoeffective
cone of M 0,n is not polyhedral for n ≥ 10. In this paper we focus on minimal
toric surfaces, that is toric surfaces which do not contain curves with negative
self-intersection. These are quotients, by a cyclic subgroup of the big torus, of
either P2 or of P1 × P1 . In the first case (fake weighted projective planes) there
are several known such surfaces P such that the Cox ring of Ble P is not finitely
generated [4, 7–10]. Recently in [11] the authors provided examples where the
effective cone of Ble P is open on one side. Here our contribution is a necessary
and sufficient condition for the Cox ring to be generated in multiplicity one (see
Definition 3.1), generalizing [12, Thm. 1.1]. Our main result concerns the second
case where we prove the following.
Theorem 1. Let P be a minimal toric surface of Picard rank two. Then the Cox
ring of Ble P is finitely generated.
Our strategy for proving this result is the following. First of all we characterize
the strict transforms of one-parameter subgroups that are negative curves. Then
we show that these classes, together with the exceptional divisor and the pullback
of two classes of self-intersection 0 generate the effective cone. Finally we provide a
necessary and sufficient condition for the finite generation of the Cox ring of surfaces
with a polyhedral effective cone (Proposition 1.3). In Theorem 4.4 we show that
Eff(Ble P) is generated in multiplicity one but we don’t know if the same is true
for the Cox ring. In Theorem 3.5 we provide a necessary condition for the Cox
1. Preliminaries
In this section we prove a criterion for finite generation of the effective cone of a
normal projective Q-factorial surface and a criterion for finite generation of the Cox
ring. Both results will be used in Section 4 to prove our main result: Theorem 1. In
what follows we will adopt the following notation: given a finite set of Weil divisors
classes S ∶= {D1 , . . . , Dn } on a projective variety X with finitely generated divisor
class group Cl(X) we will denote by Cone(S) the cone of ClR (X) generated by the
classes D1 , . . . , Dn . We will also denote by
QX ⊆ ClR (X)
the positive light cone of X, i.e. the set of classes D ∈ ClR (X) such that D2 ≥ 0
and D ⋅ H ≥ 0, for some ample class H.
Proposition 1.1. Let X be a normal projective Q-factorial surface with Picard
rank 3. If there exists a set S ∶= {D1 , . . . , Dn } of classes of irreducible curves on X
such that dim(Cone(S)) = 3 and the intersection form is negative semi-definite on
Cone(Dn , D1 ) and Cone(Di , Di+1 ), for 1 ≤ i ≤ n − 1, then Eff(X) = Cone(S).
Proof. Since Di2 ≤ 0 for any i, there exists a subset I ⊆ {1, . . . , n} such that ∣I∣ ≥
3 and the extremal rays of the cone Cone(S) are {Di ∶ i ∈ I}. Moreover the
intersection form is negative semi-definite on the facets of this cone, so that either
QX ⊆ Cone(S), or Cone(S) does not intersect the interior of QX . We claim that
the latter is not possible. Indeed, given an extremal ray Di of Cone(S), by the
Hodge index theorem we have that Di⊥ intersects the interior of QX . This implies
that Di ⋅ Dj < 0 for any other extremal ray Dj of Cone(S), a contradiction.
ON BLOWING UP MINIMAL TORIC SURFACES 3
that is the stable base locus of D is contained in the union of all curves of S. If
D is big, then B(D) ⊆ D⊥ by [18, Thm. 1.1] and the second assumption on S
implies that no curve of S ∩ D⊥ can be in B(D). If D is not big, then 0 = D2 =
D ⋅ ∑[C]∈S mC C = ∑[C]∈S mC (D ⋅ C) implies that B(D) ⊆ D⊥ and again we conclude
that no curve, whose class is in S ∩ D⊥ , can be in B(D). Finally we deduce that
B(D) = ∅, that is D is semiample, by [16, Rem. 2.1.32].
We prove (ii) ⇒ (i). Let S be the set of degrees of a minimal generating
subset of the Cox ring. Observe that all the element of S are classes of irreducible
curves of X. The first condition of Definition 1.2 holds because QX ⊆ Eff(X) =
Eff(X) = Cone(S), where the inclusion is by Riemann-Roch, the second equality is
by hypothesis and the first equality is because Eff(X) is polyhedral. The second
condition holds because
D ∈ Nef(X) = Mov(X) = ⋂ Cone(S ∖ {C}) ⊆ ⋂ Cone(S ∖ {C}),
C∈S C∈S∩D ⊥
1The argument given for Picard rank at least three works for Picard rank two as well.
4 A. LAFACE AND L. UGAGLIA
Remark 1.4. The above proposition provides a criterion for finite generation of
the Cox ring but it does not give its generators. For instance, let us consider
X to be a del Pezzo surface of degree 1. Since X is a Mori dream space, the
rays of Eff(X) are generated by classes of negative curves, which can only be
(−1)-curves. Therefore we can write Eff(X) = Cone(S) ⊇ QX , where S ∶= {[C] ∶
C is a (−1)-curve}. The extremal rays of the nef cone of X are generated by either
pullbacks of lines (contracting 8 disjoint (−1)-curves) or by conic bundles. Thus
S is a pseudo generating set. On the other hand the Cox ring of X is generated
by homogeneous elements whose degrees are either classes of curves in S or the
anticanonical class −KX .
Corollary 1.5. Let X be a normal projective Q-factorial surface with divisor class
group Cl(X) of rank 2. Then the following are equivalent:
(i) the Cox ring of X is finitely generated;
(ii) there are irreducible curves C1 , D1 , C2 , D2 ⊆ X such that Ci ⋅ Di = 0 for
i = 1, 2.
Proof. We prove (i) ⇒ (ii). If the Cox ring of X is finitely generated then Eff(X)
is polyhedral, so that its two extremal rays are generated by classes of irreducible
curves C1 and C2 . The extremal rays of Nef(X) lie on Ci⊥ . If Ri ∈ Ci⊥ is a nef
divisor, then it is semiample by the hypothesis. Thus, up to replace Ri with a
positive multiple, we can assume that the linear system ∣Ri ∣ defines a morphism
φ∶ X → Y with connected fibers onto either a curve or a surface [16, Thm. 2.1.27].
An irreducible component Di of a general fiber of φ is disjoint from Ci .
We prove (ii) ⇒ (i). Let S ∶= {C1 , D1 , C2 , D2 }. The equation Ci ⋅ Di = 0 implies
that one of Ci or Di has self-intersection ≤ 0. Thus the first hypothesis on S in
Proposition 1.3 is satisfied. If Ci2 < 0, then [Di ] ∈ Cone({Di }) ⊆ Cone(S ∖ {Ci }). If
Ci2 = 0, then Di2 = 0 and the two classes [Ci ], [Di ] span the same extremal ray R
of Nef(X). In this case we have [Ci ], [Di ] ∈ R ⊆ Cone(S ∖ {Ci }) ∩ Cone(S ∖ {Di }).
□
2. Toric surfaces
In this section we are going to recall some definitions and to prove some facts
about abelian monoids, toric varieties, lattice ideals and toric surfaces.
2.1. Abelian monoids. Recall that an abelian monoid is a pair (S, +) consisting
of a set S and a binary operation S × S → S, which is commutative, associative and
admits a neutral element 0 ∈ S. The abelian monoid S is cancellative if a + c = b + c
implies a = b, where a, b, c ∈ S, while S is finitely generated if there is a surjective
homomorphism of monoids Nr → S.
Definition 2.1. Let S be an abelian monoid. An ideal I ⊆ S of S is a subset such
that i + s ∈ I for any i ∈ I and s ∈ S. Given a subset U ⊆ S, the ideal generated by
U is
⟨U ⟩ ∶= {u + s ∶ u ∈ U and s ∈ S}.
A basis of an ideal I ⊆ S is a subset B ⊆ S such that ⟨B⟩ = I. One says that B is a
minimal basis for I if B is a basis but no proper subset of B is. If the basis B is
finite, we say that the ideal ⟨B⟩ is finitely generated.
Proposition 2.2. Let S be a finitely generated abelian monoid. Then any minimal
basis of an ideal of S is finite.
ON BLOWING UP MINIMAL TORIC SURFACES 5
Proposition 2.5. If P is a projective toric surface then OPP = Ie and thus Σnef (P) =
Σop
nef (P).
Theorem 2.11. Let P be a Q-factorial projective toric variety. Then the following
hold:
(i) wd(P) is finite;
(ii) wd is constant on the relative interior of cones of Σop
Nef (P).
Proof. We prove (i). Observe that the monoid CP is finitely generated since a
generating set consists of classes of torus invariant irreducible curves. The fiber of
the map
φ∶ wd(P) → OPP , v ↦ [Cv ]
over [Cv ] consists of {v, −v}, in particular the map has finite fibers. Let B ⊆ OPP
be a minimal basis. Since by Proposition 2.2 B is finite, in order to conclude it is
ON BLOWING UP MINIMAL TORIC SURFACES 7
ideal is
+ −
IL ∶= ⟨xm − xm ∶ m ∈ L⟩.
It is known [19, Thm. 3.7] that a minimal binomial basis for a lattice ideal is
essentially unique (up to signs) if the lattice ideal is not a complete intersection.
Given a projective toric variety P ∶= PΣ , with fan Σ ⊆ NQ , let v1 , . . . , vr ∈ N be the
primitive generators of the one-dimensional cones of Σ. Denote by PΣ∗ ∶ Zr → Zn
the homomorphism defined by ei ↦ vi for 1 ≤ i ≤ r and let LΣ ⊆ Zr be the image
of the dual homomorphism PΣ . In what follows we will denote simply by IP the
lattice ideal of the lattice LΣ .
Definition 2.14. Let P ∶= PΣ be a projective toric surface, with fan Σ ⊆ NQ and
let BΣ be a minimal basis of the lattice ideal IP . We define
+ −
Lib(P) ∶= {v ∈ N ∶ m ∈ v ⊥ and xm − xm ∈ BΣ }.
Remark 2.15. Observe that Lib(P) is uniquely defined only if the lattice ideal
is not a complete intersection. For example if P is the projective plane this is not
the case. In general a basis of a lattice L ⊆ Zr does not give a basis BL of the
lattice ideal IP because a saturation step with respect to the product x1 ⋯xr of all
variables is needed [17, Lem. 7.6]. On the other hand if Φ ∶ Qr ⊕ Qr → Qr is the
8 A. LAFACE AND L. UGAGLIA
difference map (a, b) ↦ a − b, then one can show that BL is a subset of the Hilbert
basis B of the cone
σL ∶= Φ−1 (LQ ) ∩ Q2n
≥0 .
To prove this it is sufficient to show that, given IB ∶= ⟨xa − xb ∶ (a, b) ∈ B⟩, the
equality IP = IB holds. If (a, b) ∈ B then m ∶= a − b ∈ L and both a and b have
non-negative entries, so that xa − xb ∈ IP . This proves the “⊇” inclusion. On the
other hand, if xp −xq ∈ IP then (p, q) ∈ σL , so that (p, q) = ∑(a,b)∈B c(a,b) (a, b), where
each c(a,b) is a non-negative integer. Thus in the quotient ring k[x1 , . . . , xn ]/IB we
have the following
x̄p = ∏ x̄ (a,b) = ∏ x̄ (a,b) = x̄ ,
ac bc q
(a,b)∈B (a,b)∈B
p q
which implies x − x ∈ IB , proving the “⊆” inclusion.
Theorem 2.16. Given a projective toric surface P ∶= PΣ , the following inclusions
hold
Neg(P) ⊆ wd(P) ⊆ Lib(P).
Proof. We prove the first inclusion. Let us fix a lattice direction v ∈ Neg(P), i.e.
a v ∈ N such that the class of the curve C̃v has negative self-intersection on the
blowing up Ble P. The class of C̃v spans an extremal ray of the effective cone
Eff(Ble P). Let us consider the facet of the nef cone Nef(Ble P) orthogonal to C̃v
and let us fix a divisor D whose class is in the relative interior of this facet. By
construction D ⋅ C̃v = 0 and D ⋅ C̃w > 0 for any w ≠ v. Given an ample divisor A
on Ble P and a sufficiently small ε ∈ Q>0 , the divisor D + εA is ample and min{(D +
εA) ⋅ C̃u ∶ u ∈ N } is attained at C̃v . Thus the pushforward of D + εA on X is an
ample divisor H such that lw(∆H ) = lwv (∆H ).
We prove the second inclusion. Let {f1 , . . . , fs } be a minimal binomial basis
of the lattice ideal and let ±v1 , . . . , ±vs ∈ N be the corresponding directions. Let
v ∈ wd(P) and let xa − xb be a binomial for Cv . Then xa − xb = ∑si=1 gi fi , where the
gi are homogeneous polynomials. Thus Cv is linearly equivalent to Cvi + E, where
E is the effective divisor defined by gi . Let D be an ample divisor such that v is
a width direction of the polytope ∆D . Then lwv (∆D ) = D ⋅ Cv = D ⋅ Cvi + D ⋅ E =
lwvi (∆D ) + D ⋅ E implies D ⋅ E = 0, so that E = 0. Thus v = vi ∈ Lib(P).
□
Remark 2.17. By [19, Thm. 3.7], if the lattice ideal IP is not a complete intersec-
tion, then we have the equality Lib(P) = Hbs(P)± . In particular for every direction
v ∈ wd(P) we have that both v and −v belong to the Hilbert basis of a cone of Σ.
Both inclusions in Theorem 2.16 can be strict as shown in the following example.
Example 2.18. Let P ∶= PΣ be the toric surface whose fan Σ ⊆ Q2 is given in the
first of the following pictures.
ON BLOWING UP MINIMAL TORIC SURFACES 9
The lattice ideal IΣ is generated by the three binomials x2 x3 −x34 , x1 x2 x4 −x23 , x1 x22 −
x3 x24 which corresponds to the three directions in Lib(P) = {±(0, 1), ±(1, −1), ±(1, 0)}.
The grading matrix for the Cox ring of P is
0 5 4 3
[ ].
1 3 3 2
The nef cone is Nef(P) = Cone([3, 2], [4, 3]) and ΣNef (P) has two maximal cones:
σ1 ∶= Cone([3, 2], [15, 11]) and σ2 ∶= Cone([15, 11], [4, 3]). The class [15, 11] ∈
σ1 ∩ σ2 corresponds to the torus-invariant Weil divisor D ∶= 3D3 + D4 . The divisor
3D is Cartier and its Riemann-Roch polytope
∆3D = Polytope((0, 0), (3, 6), (6, 3), (−1, 0))
has width 6, attained along the directions ±(0, 1), ±(1, −1). Polytopes corresponding
to Cartier classes in the interior of σ1 or of σ2 admit just one of these two width
directions. It follows that wd(P) = {±(0, 1), ±(1, −1)}. Finally the curve C̃v , with
v = (0, 1), is the only one with negative self-intersection (C̃v2 = −2/5), so that
Neg(P) = {±(0, 1)}.
3. Lattice ideals
As before, let P ∶= PΣ be a projective toric variety with fan Σ ⊆ Qn . Denote by
R ∶= K[x1 , . . . , xr ] the Cox ring of P. The Cox ring R(Ble P) of Ble P is isomorphic
to a subalgebra of R[t, t−1 ], where the extra variable t represents the exceptional
divisor (see [13, Prop. 5.2]).
Definition 3.1. We say that R(Ble P) ⊆ R[t, t−1 ] is generated in multiplicity m
if it admits a set of homogeneous generators of the form f t−i with 0 ≤ i ≤ m and
f ∈ R.
In this section we study the problem of determining when the Cox ring of the
blowing-up Ble P is generated in multiplicity one, in terms of the lattice ideal IP .
We begin with a result holding for projective toric varieties and then in the rest of
the section we will restrict to the case of toric surfaces.
Proposition 3.2. Let P be an n-dimensional projective toric variety, let R ∶= R(P)
and let IP = ⟨f1 , . . . , fn ⟩ ⊆ R be a complete intersection. Then
R(Ble P) ≃ R[s1 , . . . , sn , t]/⟨fi − tsi ∶ 1 ≤ i ≤ n⟩.
In particular the Cox ring is generated in multiplicity 1.
Proof. Let S ∶= R[s1 , . . . , sn , t] and let J ∶= ⟨fi − tsi ∶ 1 ≤ i ≤ n⟩ ⊆ S. We have
dim(J + ⟨t⟩) = dim S/(J + ⟨t⟩) = dim R[s1 , . . . , sn ]/IP = dim R.
On the other hand, for any generator xi ∈ R we have
dim(J + ⟨t, xi ⟩) = dim S/(J + ⟨t, xi ⟩) = dim R[s1 , . . . , sn ]/(IP + ⟨xi ⟩) < dim R,
where the last inequality is due to the fact that the lattice ideal IP is a graded-
prime ideal. Then, by [13, Prop. 5.1], the Cox ring of Ble P is isomorphic to
R[s1 , . . . , sn , t]/J sat , where J sat is the saturation J∶ ⟨t⟩∞ . To conclude we need to
show that J = J sat . Let us suppose by contradiction that there exists a g ∈ J sat ∖ J.
If we denote by g̃ the evaluation of g at t = 0, we have
dim S/(J sat + ⟨t⟩) ≤ dim S/(J + ⟨g, t⟩) = dim R[s1 , . . . , sn ]/(IP + ⟨g̃⟩) = dim R − 1.
10 A. LAFACE AND L. UGAGLIA
Since dim R[s1 , . . . , sn , t]/J sat = dim R[s1 , . . . , sn , t±1 ]/J sat = dim R[t±1 ] = dim R+1,
we get a contradiction.
□
3.1. Toric surfaces. In what follows “lattice ideal” will mean lattice ideal of a
projective toric surface. Let us recall the following result (see for instance [14,
Theorem 1] and [19, Proposition 4.1]) about lattice ideals in 3 variables.
Lemma 3.3. Any lattice ideal IP ⊆ k[x1 , x2 , x3 ] is Cohen-Macaulay. Moreover IP
is not a complete intersection if and only if (modulo reordering the indexes) it is
generated by the 2 × 2 minors of a matrix
xα
1
1
xα
2
2
xα
3
3
(3.1) A=( β3 β1 β2 ) ,
x3 x1 x2
with αi , βi > 0 for any i = 1, 2, 3.
When there are only 3 variables, or equivalently the fan Σ has 3 rays, we have
the following characterization of toric surfaces P = PΣ such that the Cox ring of
Ble P is generated in multiplicity 1 (see also [13, Thm. 1.1]).
Proposition 3.4. Let L be a lattice in Z3 associated to the rays of a complete fan
Σ, and let IP ⊆ k[x1 , x2 , x3 ] be the corresponding lattice ideal. The following are
equivalent.
(i) The Cox ring of Ble P is generated in multiplicity 1.
(ii) IP ⊆/ ⟨x1 , x2 , x3 ⟩2 .
(iii) IP is a complete intersection.
Proof. The implication (ii) ⇒ (iii) follows from Lemma 3.3. We prove (iii) ⇒ (ii).
Let us suppose that IP ⊆ ⟨x1 , x2 , x3 ⟩2 . Since IP is a lattice ideal, it does not contain
linear components of codimension 2, and hence, up to reordering the variables, IP
is generated by f1 = xa1 1 −xa2 2 xa3 3 and f2 = xb22 −xb33 , where a1 , b2 , b3 > 1. This implies
that the lattice L is generated by the rows of the following matrix
a −a2 −a3
( 1 ),
0 b2 −b3
whose columns are the rays of the fan Σ. Since the first column is not primitive we
get a contradiction.
The implication (iii) ⇒ (i) is given by Lemma 3.2, so that we conclude by
showing that (i) ⇒ (iii). If the Cox ring of Ble P is generated in multiplicity 1,
by Corollary 1.5 there there exist two one-parameter subgroups Cv and Cw such
that C̃v ⋅ C̃w = 0. This implies that Cv ⋅ Cw = 1, so that Cv and Cw meet only at e,
transversely. Therefore det(v, w) = 1 and we can assume without loss of generality
that v = (1, 0) and w = (0, 1). Let us denote by ρ1 , ρ2 and ρ3 the primitive vectors
generating the 1-dimensional rays of Σ. Since no invariant point of P is contained
in the intersection Cv ∩ Cw , one of the two directions, say v, must lie on a 1-
dimensional ray of Σ, so that we can assume that ρ1 = v. The remaining two rays
of Σ must lie on the second and third quadrant respectively (one of them can lie
on the vertical axis). By [19, Lemma 3.1] we conclude that the lattice ideal IP is a
complete intersection. □
When the number of variables is bigger, we have the following weaker result.
ON BLOWING UP MINIMAL TORIC SURFACES 11
Theorem 3.5. If the Cox ring of Ble P is generated in multiplicity 1, then for any
three indexes i, j, k ∈ {1, . . . , n}, IP ⊆/ ⟨xi , xj , xk ⟩2 .
Proof. Let us suppose by contradiction that there are three indexes, say 1, 2, 3,
such that IP ⊆ ⟨x1 , x2 , x3 ⟩2 , and let us denote by ρ1 , . . . , ρn the primitive vectors
generating the 1-dimensional rays of Σ. Let us fix a basis B = {f1 , . . . , fr } for the
lattice ideal IP , and let us consider the ideal
I ′ ∶= ⟨fi′ ∶ 1 ≤ i ≤ r⟩ ⊆ C[x1 , x2 , x3 ],
where we set fi′ ∶= f (x1 , x2 , x3 , 1, . . . , 1). Let Σ′ be the fan whose rays are generated
by ρ1 , ρ2 and ρ3 , let L′ ⊆ Z3 be the corresponding lattice and let us denote by
P′ the toric surface PΣ′ . We claim that Σ′ is complete. First of all observe that
the map (C∗ )3 → (C∗ )2 , given by the matrix of L′ , factorizes through the map
(C∗ )n → (C∗ )2 , given by L. This implies that IP′ equals the saturation of I ′ with
respect to x1 x2 x3 . Since I ′ ⊆ ⟨x1 , x2 , x3 ⟩2 , the monomials of each binomial in the
saturation of I ′ is divisible by either x1 , x2 or x3 . This gives the claim. We can
then apply Proposition 3.4 to IP′ and deduce that the Cox ring of Ble P′ is not
generated in multiplicity 1. We conclude observing that the contraction of the rays
{ρi ∶ i = 4, . . . , n} induces a surjective morphism π ∶ Ble P → Ble P′ . □
Indeed, in all the examples we checked with the help of Magma [2] we have found
that also the other implication holds, so that we formulate the following.
Conjecture 3.6. Given a be a projective toric surface P the following are equiva-
lent:
(i) the Cox ring of Ble P is generated in multiplicity 1;
(ii) for any three indexes i, j, k, IP ⊆/ ⟨xi , xj , xk ⟩2 .
We have already seen that the above conjecture holds if there are only 3 variables
(Proposition 3.4) and we are going to show that it also holds if the ideal IP is Cohen-
Macaulay, i.e. it has at most 3 minimal generators.
Theorem 3.7. Conjecture 3.6 holds true if IP is Cohen-Macaulay.
Proof. By Theorem 3.5 we only need to prove the implication (ii) ⇒ (i) of Con-
jecture 3.6. If IP has 2 generators, then it is a complete intersection, so that the
Cox ring of Ble P is generated in multiplicity 1 by Proposition 3.2. Let us suppose
that there exist three indexes, say 1, 2, 3, such that IP ⊆ ⟨x1 , x2 , x3 ⟩2 . Reasoning as
in the proof of Proposition 3.4 we would have that the generator of a ray of Σ is
not primitive, a contradiction.
Let us suppose now that IP has 3 minimal generators, so that, by Hilbert–Burch
Theorem [6, Thm. 3.2], IP = ⟨f1 , f2 , f3 ⟩, where fi are the maximal minors of a
2 × 3 matrix A = (ai,j ), whose entries are monomials in x1 , . . . , xn . In this case
we have that J sat + ⟨t⟩ ⊇ ⟨f1 , f2 , f3 , h1 , h2 ⟩, where h1 = a1,1 s1 + a1,2 s2 + a1,3 s3 and
h2 = a2,1 s1 + a2,2 s2 + a2,3 s3 . Let us fix a variable xi , with 1 ≤ i ≤ n. Since xi appears
in at least one of f1 , f2 , f3 , we have that the vanishing of xi implies the vanishing
of at least another variable xj . Let us distinguish two cases.
i) The vanishing of xi and xj forces a third variable xk to be zero. By the
assumption IP ⊆/ ⟨xi , xj , xk ⟩2 we deduce that there exists an entry of A that
does not vanish for xi = xj = xk = 0. In particular, at least one of h1 , h2
does not vanish identically, so that dim(J sat + ⟨t, xi ⟩) ≤ n − 1.
12 A. LAFACE AND L. UGAGLIA
ii) The vanishing xi = xj = 0 does not imply the vanishing of a third variable.
In this case there is at least one of the binomials, say f1 , whose monomials
do not contain xi and xj , so that it survives when we set xi = xj = 0.
Moreover, in each row of A there is at least one entry that does not vanish,
which implies that also h1 and h2 do not vanish identically. This implies
again that dim(J sat + ⟨t, xi ⟩) ≤ n − 1
Since the above reasoning holds for any index 1 ≤ i ≤ n, we conclude that dim(J sat +
⟨t, Πn1 xi ⟩) ≤ n − 1, so that the Cox ring is generated in multiplicity 1.
□
⎪ −1 ⎪ −1
⎧
⎪ a = 0 ⎧
⎪ b = 1
⎪ ⎪
(4.4) ⎨ a0 = 1 ⎨ b0 = 0
⎪
⎪ ai = ci ai−1 + ai−2
⎪ ⎪
⎪ bi = ci bi−1 + bi−2 , 1 ≤ i ≤ n.
⎪
⎩ ⎩
If we set vi ∶= (ai , bi ) ∈ N , for 0 ≤ i ≤ n, we have that vi lies in the cone ⟨(1, 0), (p, q)⟩
if i is even, while it lies in ⟨(p, q), (−1, 0)⟩ if i is odd. Let us denote simply by Ci
the class of C̃vi in Ble P. From (4.3) we have that
Ci = bi H1 + βi H2 − E, where βi ∶= (−1)i (ai q − bi p).
Lemma 4.2. With the notation above, the following hold:
(i) β0 = q, β1 = p and βi = βi−2 − ci βi−1 for 2 ≤ i ≤ n.
(ii) βi bi+1 + bi βi+1 = q, for any 0 ≤ i ≤ n − 1.
(iii) βi−1 bi+1 − βi+1 bi−1 = ci+1 q, for any 1 ≤ i ≤ n − 1.
(iv) βi bi + βi+1 bi+1 < q, for any 0 ≤ i ≤ n − 1.
ON BLOWING UP MINIMAL TORIC SURFACES 13
Proof. We only prove (iv), since (i), (ii) and (iii) easily follow by induction. By (ii)
and (i) we can write
q − βi bi − βi+1 bi+1 = βi bi+1 + bi βi+1 − βi bi − βi+1 bi+1 = (βi+1 − βi )(bi − bi+1 ).
Since we are supposing 2p < q we have that c1 = 0 and c2 > 1, so that {b0 , . . . , bn }
is strictly increasing. In a similar way, by (i) one can deduce that {β0 , . . . , βn } is
strictly decreasing. The statement follows. □
Remark 4.3. Using the intersection matrix (4.2) and Lemma 4.2(ii) we have that
for any 0 ≤ i ≤ n − 1,
βi bi+1 + βi+1 bi
Ci ⋅ Ci+1 = − 1 = 0.
q
4.1. Effective cones. We are now going to prove that the effective cone Eff(Ble P)
is generated in multiplicity 1. In order to do this, we are going to use Proposition 1.1
with the set consisting of E and all the classes Ci having negative self intersection.
Theorem 4.4. With the notation above,
Eff(Ble P) = cone({E} ∪ {Ci ∣ 0 ≤ i ≤ n and 2βi bi < q}),
in particular it is generated in multiplicity 1.
Proof. Let us consider the ordered set U ∶= {E} ∪ {Ci ∣ 0 ≤ i ≤ n and 2βi bi < q}.
Since Ci2 = 2βi bi /q − 1, every Ci ∈ U is the class of a negative curve. Therefore,
by Proposition 1.1, in order to prove our result it is enough to show that the
intersection matrix is negative semidefinite on every cone generated by a pair of
consecutive classes in U.
First of all, C0 = qH2 − E belongs to U. The matrix of the intersection form on
−1 1
Cone(E, C0 ) is ( ), which is negative semidefinite, and the same holds for
1 −1
Cone(Cn , E).
Let us now fix any index 0 ≤ j ≤ n, for which 2βj bj < q, so that Cj ∈ U. If we
have that also 2βj+1 bj+1 < q, then the next class in U is Cj+1 . By Remark 4.3,
the matrix of the intersection form on Cone(Cj , Cj+1 ) is diagonal and hence it
is negative definite. If otherwise 2βj+1 bj+1 ≥ q, by Lemma 4.2(iv) we must have
2βj+2 bj+2 < q, i.e. the next class in U is Cj+2 . The matrix of the intersection form
on Cone(Cj , Cj+2 ) is
2β b
j j βj+2 bj +βj bj+2
⎛ q
−1 q
− 1⎞
β b +β b 2βj+2 bj+2 .
⎝ j+2 j q j j+2 − 1 q
− 1 ⎠
Its determinant can be written as
(βj − βj+2 )(bj+2 − bj ) (bj+2 βj − βj+2 bj )2 c2j+2
2 − = (2βj+1 bj+1 − q)
q q2 q
where the equality follows from Lemma 4.2(iii) and (4.4). Since we are supposing
that 2βj+1 bj+1 − q ≥ 0, the above determinant is non-negative, which implies that
the intersection form is negative semi-definite.
We conclude that the ordered set U satisfy the hypotheses of Proposition 1.1 so
that its classes are the extremal rays of Eff(Ble P). □
A consequence of the above theorem is the following characterisation of the
directions in Neg(P) and wd(P) in the cases we are considering.
14 A. LAFACE AND L. UGAGLIA
Remark 4.6. The above result implies that if Ci2 never vanishes, then wd(P) =
Neg(P) = {vi ∶ Ci2 < 0}. We are now going to see that the condition Cj2 = 0 for
some 0 ≤ j ≤ n turns out to be quite strong. Indeed we know that it is equivalent to
2bj βj = q, so that, by definition of βj we can write pbj = qaj + (−1)j βj . Therefore
qaj + (−1)j βj βj (2aj bj + (−1)j )
(4.5) p= = ,
bj bj
so that bj divides βj , but since we are supposing gcd(p, q) = 1, it must be βj = bj .
In particular, since {bi } is strictly increasing and {βi } is strictly decreasing, there
can be at most one index j such that bj = βj , i.e. there is at most one curve Cj
such that Cj2 = 0. Moreover, from bj = βj we deduce that q = 2b2j and from (4.5)
that p = 2aj bj + (−1)j .
On the other hand, let us fix two positive integers r, s with gcd(r, s) = 1 and
2s ≤ r, and set q = 2r2 and p = 2rs ± 1. Under these hypotheses, by [20, Lemma]
we have that s/r is a convergent for the continued fraction of p/q, i.e. there exists
an index j such that aj = s and bj = r. This implies that βj = r, so that the curve
Cj = βj H1 + bj H2 − E satisfies Cj2 = 0.
We also remark that the matrix of the intersection form on any facet F of
Eff(Ble P) is negative definite, unless F = Cone(E, eHi − E), for i = 1, 2, or F =
Cone(Cj−1 , Cj+1 ), where j is such that Cj2 = 0. Therefore the light cone Q is
tangent to Eff(Ble P) either at the 2 points H1 , H2 or at the 3 points H1 , H2 , Cj ,
if Cj2 = 0.
Example 4.7. Let us fix r = 11, s = 3 and consider the toric surface X with fan
generated by the rays (±1, 0), (±p, ±q), where p = 2rs − 1 = 65 and q = 2r2 = 242.
Let us denote by c = [0, 3, 1, 2, 1, 1, 1, 1, 3] the continued fraction of p/q, so that
b = [0, 1, 3, 4, 11, 15, 26, 41, 67, 242] and β = [242, 65, 47, 18, 11, 7, 4, 3, 1, 0] give rise to
the classes Ci = βi H1 + bi H2 − E, for i = 0, . . . , 9. We have that for any i ≠ 2, 4, 7,
2bi βi < q while 2b2 β2 > q, 2b7 β7 > q, and β4 = b4 = 11, so that 2b4 β4 = q. Therefore
the light cone is tangent to Eff(Ble P) at the points corresponding to H1 , H2 and
C4 = 11H1 + 11H2 − E (see Figure 1).
Example 4.8. Let us consider the case in which (p, q) = (φn−1 , φn+1 ), where we
denote by φn the n-th Fibonacci number2, and n ≥ 5. The continued fraction of
p/q is c = [0, 2, 1, . . . , 1, 2], with ∣c∣ = n − 1. Therefore b0 = 0, bi = φi+1 for any
1 ≤ i ≤ n − 2, and bn−1 = φn+1 . Analogously β0 = φn+1 , βi = φn−i for any 1 ≤ i ≤ n − 2
and βn−1 = 0. We claim that 2βi bi < q for any 0 ≤ i ≤ n − 1. Indeed, if i = 0 or n − 1
we have 2βi bi = 0, while if 1 ≤ i ≤ n − 2 we can write
q − 2βi bi = φn+1 − 2φn−i φi+1
= φi+1 φn+1−i + φi φn−i − 2φn−i φi+1
= φi+1 φn−1−i − φi−1 φn−i
= φi φn−i−1 − φi−1 φn−i−2 > 0,
where the second equality follows from [23, Eqn. (8)]. Therefore, by Proposition 4.4,
all the classes βi H1 +bi H2 −E, for i ∈ {0, . . . , n−1} are negative curves, and, together
with E, they generate the extremal rays of the effective cone.
Figure 1. Eff(Ble P)
By Theorem 4.4, the facets of Eff(Ble P) are Cone(E, C0 ), Cone(Cn , E) and, for any
i such that 2βi bi < q, we have either the facet Cone(Ci , Ci+1 ) (if also 2βi+1 bi+1 < q)
or Cone(Ci , Ci+2 ) (if 2βi+1 bi+1 ≥ q). Therefore we have to consider 3 different cases.
H2
E C0 Ci Ci+2 Ci Ci+1
H1 Ci+1 R
Cn+2 Ci−1 Ci+2
(i) The ray of Nef(Ble P) orthogonal to the facet Cone(E, C0 ) = Cone(E, qH2 − E)
is generated by H2 . Since the latter lies in the relative interior of the facet and
belongs to the set S, it lies in the 2 cones generated by S ∖{E} and by S ∖{qH2 −E}
respectively, so that (4.6) holds. We can reason in a similar way for the facet
Cone(Cn , E) = Cone(qH1 − E, E).
ON BLOWING UP MINIMAL TORIC SURFACES 17
(ii) By Remark 4.3, the ray of Nef(Ble P) orthogonal to the facet Cone(Ci , Ci+2 ) is
generated by Ci+1 . Since the latter belongs to the set S, condition (4.6) holds.
(iii) Finally, let us consider a facet Cone(Ci , Ci+1 ). Again by Remark 4.3 we have
that Cone(Ci−1 , Ci+1 ) is orthogonal to Ci and Cone(Ci , Ci+2 ) is orthogonal to Ci+1 .
Therefore, the ray R orthogonal to the facet Cone(Ci , Ci+1 ) is the intersection
Cone(Ci−1 , Ci+1 ) ∩ Cone(Ci , Ci+2 ). We conclude that (4.6) holds also for this facet.
□
Remark 4.9. The above proof relies on the results of Lemma 1.3. As we already
pointed out in Remark 1.4, this allows to show that the Cox ring of Ble P is finitely
generated, but it does not give any clue on its generators. Therefore, even if we
know that the effective cone is generated in multiplicity 1, we can not deduce that
the same holds for the Cox ring. Nevertheless, since the first row of the matrix (4.1)
corresponds to the monomial x1 xp2 − x3 xp4 in the lattice ideal IP , we can conclude
that for any three indexes i, j, k ∈ {1, 2, 3, 4}, IP ⊆/ ⟨xi , xj , xk ⟩2 . Therefore we have
the following.
Corollary 4.10. Let P be a minimal toric surface of Picard rank two. If Conjec-
ture 3.6 holds, then the Cox ring Cox(Ble P) is generated in multiplicity 1.
5. A Magma library
In this section we present a library of Magma functions [2] that we used in
this paper and we explain some of them in one example. The library is freely
downloadable from this web site:
https://github.com/alaface/Blowing-up-toric-surfaces
The main functions of this library are briefly described here.
● TestMultOne. It takes as input a list of primitive lattice vectors of Z2
generating a fan Σ of a complete toric surface P. It returns true if the
Cox ring of Ble P is generated in multiplicity one.
● TestComp. It takes as input a list of primitive lattice vectors of Z2 gener-
ating a fan Σ of a complete toric surface P. It returns true if the lattice
ideal of P is not contained in any square of ideals generated by subsets of
three variables.
● GensUpTo. It takes as input a list of primitive lattice vectors of Z2 gener-
ating a fan Σ of a complete toric surface P and an integer m. It returns
the divisor classes of generators of the Cox ring of Ble P up to multiplicity
m.
● IsMDS. It takes as input a list of primitive lattice vectors of Z2 generating
a fan Σ of a complete toric surface P and an integer m. It returns true if
the Cox ring of Ble P admits a pseudo generating set in multiplicity up to
m.
Example 5.1. Let P ∶= PΣ be the toric surface whose fan is generated by all the
primitive lattice vectors of Z2 with coordinates having absolute value at most 2.
We display the rays in the following picture.
18 A. LAFACE AND L. UGAGLIA
x6 x4
x7 x5 x3
x8 x2
x9 x1
x10 x16
x11 x13 x15
x12 x14
Using TestComp we found that all the triples {i, j, k} such that IP ⊆ ⟨xi , xj , xk ⟩2
are {2, 6, 12}, {2, 7, 12}, {2, 8, 12}, {4, 10, 14}, {4, 10, 15}, {4, 10, 16}, {6, 10, 16},
{6, 11, 16}, {6, 12, 16}, {2, 8, 14}, {3, 8, 14}, {4, 8, 14}. The maximal subsets of rays
which do not contain these triples are the following
{1, 3, 4, 5, 7, 9, 11, 12, 13, 14, 15, 16} {1, 2, 3, 4, 5, 6, 7, 9, 13, 14, 15, 16}
{1, 3, 5, 7, 8, 9, 10, 11, 12, 13, 15, 16} {1, 2, 3, 4, 5, 7, 9, 11, 13, 14, 15, 16}
{1, 3, 5, 7, 9, 10, 11, 12, 13, 14, 15, 16} {1, 2, 3, 4, 5, 7, 8, 9, 11, 13, 15, 16}
{1, 2, 3, 5, 6, 7, 8, 9, 10, 11, 13, 15} {1, 2, 3, 5, 6, 7, 9, 10, 11, 13, 14, 15}
{1, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15} {1, 2, 3, 5, 9, 10, 11, 12, 13, 14, 15, 16}
{1, 3, 4, 5, 7, 8, 9, 11, 12, 13, 15, 16} {1, 2, 3, 4, 5, 9, 10, 11, 12, 13}
{1, 3, 4, 5, 6, 7, 8, 9, 11, 12, 13, 15} {1, 2, 3, 4, 5, 6, 7, 8, 9, 13, 15, 16}
{1, 5, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16} {1, 5, 6, 7, 8, 9, 13, 14, 15, 16}
{1, 3, 5, 6, 7, 9, 10, 11, 12, 13, 14, 15} {1, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13}
{1, 2, 3, 4, 5, 9, 11, 12, 13, 14, 15, 16} {1, 3, 5, 6, 7, 8, 9, 10, 11, 12, 13, 15}
{1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 13} {1, 2, 3, 5, 7, 9, 10, 11, 13, 14, 15, 16}
{1, 2, 3, 5, 7, 8, 9, 10, 11, 13, 15, 16} {1, 2, 3, 4, 5, 6, 7, 9, 11, 13, 14, 15}
{1, 2, 3, 4, 5, 6, 7, 8, 9, 11, 13, 15} {1, 3, 4, 5, 6, 7, 9, 11, 12, 13, 14, 15}
Using TestMultOne we proved that the Cox ring of Ble P′ is generated in multi-
plicity one for each toric surface P′ whose rays of the fan are given by one of the
above subsets. The same conclusion holds for any toric surface dominated by any
such P′ , i.e. whose fan consists of a subset of those rays. In particular Conjec-
ture 3.6 holds for all these toric surfaces. Using GensUpTo we found four generators
in multiplicity 2 for the Cox ring of Ble P′ , where P′ is the toric surface defined by
the rays {2, 4, 6, 8, 10, 12, 14, 16}. Finally, using IsMDS we checked that the Cox ring
of Ble P′ is finitely generated.
References
[1] Ivan Arzhantsev, Ulrich Derenthal, Jürgen Hausen, and Antonio Laface. Cox rings. Cam-
bridge Studies in Advanced Mathematics, vol. 144. Cambridge University Press, Cambridge,
2015.
[2] Wieb Bosma, John Cannon, and Catherine Playoust. The Magma algebra system. I. The user
language, pages 235–265. 1997.
[3] Ana-Maria Castravet, Antonio Laface, Jenia Tevelev, and Luca Ugaglia. Blown-up toric sur-
faces with non-polyhedral effective cone. Journal für die reine und angewandte Mathematik
(Crelles Journal) 2023 (800):1–44, 2023.
[4] Steven Dale Cutkosky and Kazuhiko Kurano. Asymptotic regularity of powers of ideals of
points in a weighted projective plane. Kyoto J. Math. 51 (1):25–45, 2011.
[5] Jan Draisma, Tyrrell B. McAllister, and Benjamin Nill. Lattice-width directions and
Minkowski’s 3d -theorem. SIAM J. Discrete Math. 26 (3):1104–1107, 2012.
[6] David Eisenbud. The geometry of syzygies. Graduate Texts in Mathematics, vol. 229.
Springer-Verlag, New York, 2005. A second course in commutative algebra and algebraic
geometry.
ON BLOWING UP MINIMAL TORIC SURFACES 19
[7] José Luis González and Kalle Karu. Some non-finitely generated Cox rings. Compos. Math.
152 (5):984–996, 2016.
[8] Javier González Anaya, José Luis González, and Kalle Karu. Constructing non-Mori dream
spaces from negative curves. J. Algebra 539:118–137, 2019.
[9] . On a family of negative curves. J. Pure Appl. Algebra 223 (11):4871–4887, 2019.
[10] Javier González-Anaya, José Luis González, and Kalle Karu. Curves generating extremal rays
in blowups of weighted projective planes. J. Lond. Math. Soc. (2) 104 (3):1342–1362, 2021.
[11] Javier González Anaya, José Luis González, and Kalle Karu. Nonexistence of negative curves.
Int. Math. Res. Not. IMRN 16:14368–14400, 2023.
[12] Juergen Hausen, Simon Keicher, and Antonio Laface. On blowing up the weighted projective
plane. Mathematische Zeitschrift 290:1339–1358, 2016.
[13] Jürgen Hausen, Simon Keicher, and Antonio Laface. Computing Cox rings. Math. Comp. 85
(297):467–502, 2016.
[14] Jürgen Kraft. Singularity of monomial curves in A3 and Gorenstein monomial curves in A4 .
Canad. J. Math. 37 (5):872–892, 1985.
[15] Antonio Laface and Luca Ugaglia. Effective cone of the blow-up of the symmetric product of
a curve. arXiv:2210.11829, 2022.
[16] Robert Lazarsfeld. Positivity in algebraic geometry. I. Ergebnisse der Mathematik und ihrer
Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathematics [Results in Mathematics
and Related Areas. 3rd Series. A Series of Modern Surveys in Mathematics], vol. 48. Springer-
Verlag, Berlin, 2004. Classical setting: line bundles and linear series.
[17] Ezra Miller and Bernd Sturmfels. Combinatorial commutative algebra. Graduate Texts in
Mathematics, vol. 227. Springer-Verlag, New York, 2005.
[18] Michael Nakamaye. Stable base loci of linear series. Math. Ann. 318 (4):837–847, 2000.
[19] Irena Peeva and Bernd Sturmfels. Syzygies of codimension 2 lattice ideals. Math. Z. 229
(1):163–194, 1998.
[20] Attila Pethő. Simple continued fractions for the Fredholm numbers. J. Number Theory 14
(2):232–236, 1982.
[21] Stanley Rabinowitz. A census of convex lattice polygons with at most one interior lattice
point. Ars Combin. 28:83–96, 1989.
[22] J. C. Rosales and P. A. Garcı́a-Sánchez. Finitely generated commutative monoids. Nova Sci-
ence Publishers, Inc., Commack, NY, 1999.
[23] S. Vajda. Fibonacci & Lucas numbers, and the golden section. Ellis Horwood Series: Math-
ematics and its Applications. Ellis Horwood Ltd., Chichester; Halsted Press [John Wiley &
Sons, Inc.], New York, 1989. Theory and applications, With chapter XII by B. W. Conolly.