Abstract
Let C be a smooth projective curve in \({\mathbb {P}}^1 \times {\mathbb {P}}^1\) of genus \(g\ne 4\), and assume that it is birationally equivalent to a curve defined by a Laurent polynomial that is non-degenerate with respect to its Newton polygon \(\Delta \). Then we show that the convex hull \(\Delta ^{(1)}\) of the interior lattice points of \(\Delta \) is a standard rectangle, up to a unimodular transformation. Our main auxiliary result, which we believe to be interesting in its own right, is that the first scrollar Betti numbers of \(\Delta \)-non-degenerate curves are encoded in the combinatorics of \(\Delta ^{(1)}\), if \(\Delta \) satisfies some mild conditions.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Let \(f\in k[x^{\pm 1},y^{\pm 1}]\) be an irreducible Laurent polynomial over an algebraically closed field k of characteristic zero and let U(f) be the curve it defines in the two-dimensional torus \({\mathbb {T}}^2=(k^*)^2\). The Newton polygon \(\Delta =\Delta (f)\) of f is the convex hull in \({\mathbb {R}}^2\) of all the exponent vectors in \({\mathbb {Z}}^2\) of the monomials that appear in f with a non-zero coefficient. We will always assume that \(\Delta \) is two-dimensional. We say that f is non-degenerate with respect to its Newton polygon \(\Delta \) (or more briefly, f is \(\Delta \)-non-degenerate) if and only if for each face \(\tau \subset \Delta \) (including \(\tau =\Delta \)) the system
does not have any solutions in \({\mathbb {T}}^2\). Here, \(f_\tau \) is obtained from f by only considering the terms that are supported on \(\tau \). This condition is generically satisfied. Consider the map
The Zariski closure of its full image \(\varphi _\Delta ({\mathbb {T}}^2)\) is a toric surface \(\text {Tor}(\Delta )\), while the Zariski closure of \(\varphi _\Delta (U(f))\) is a hyperplane section \(C_f\) of \(\text {Tor}(\Delta )\), which is smooth if f is non-degenerate. We will denote the projective coordinates of \({\mathbb {P}}^{\sharp (\Delta \cap {\mathbb {Z}}^2)-1}\) by \(X_{i,j}\) where (i, j) runs over \(\Delta \cap {\mathbb {Z}}^2\).
We say that a smooth curve C is \(\Delta \)-non-degenerate if and only if it is birationally equivalent to U(f) for a \(\Delta \)-non-degenerate Laurent polynomial f. Note that if C is moreover projective, then it is isomorphic to \(C_f\). If C is \(\Delta \)-non-degenerate, then a lot of its geometric properties are encoded in the combinatorics of the lattice polygon \(\Delta \). For instance, its geometric genus g(C) equals the number of interior lattice points of \(\Delta \) [8]. Similar interpretations were recently provided for the gonality [3, 7], the Clifford index and dimension [3, 7], the scrollar invariants associated to a gonality pencil [3] and Schreyer’s tetragonal invariants [5].
Given this long list, the following question (initiated in [5]) naturally arises: to what extent can we recover \(\Delta \) from the geometry of a \(\Delta \)-non-degenerate curve? At least, we have to allow two relaxations to this question. First, we can only expect to find back the polygon \(\Delta \) up to a unimodular transformation, i.e. an affine map of the form
with \(A\in \text {GL}_2({\mathbb {Z}})\) and \(B\in {\mathbb {Z}}^2\), since these maps correspond to automorphisms of \({\mathbb {T}}^2\). Secondly, we can (usually) only hope to recover the convex hull of the interior lattice points of \(\Delta \), denoted by \(\Delta ^{(1)}\) (see [5] for an easy example demonstrating the need for this relaxation). In fact, all the aforementioned combinatorial interpretations are in terms of the combinatorics of \(\Delta ^{(1)}\) rather than \(\Delta \) (e.g. \(g(C)=\sharp (\Delta ^{(1)}\cap {\mathbb {Z}}^2)\)).
Given a \(\Delta \)-non-degenerate curve C, we say that the Newton polygon \(\Delta \) is intrinsic for C if and only if for all \(\Delta '\)-non-degenerate curves \(C'\) that are birationally equivalent to C, we have that \(\Delta ^{(1)}\cong \Delta '^{(1)}\). Hereby, we use \(\cong \) to denote the unimodular equivalence relation. Before stating some intrinsicness results, we give notations for some special lattice polygons:
The Newton polygon is intrinsic for all rational (\(\Delta ^{(1)}=\emptyset \)), hyperelliptic (\(\Delta ^{(1)}\) is one-dimensional, and therefore determined by the genus) and trigonal curves of genus at least 5 (\(\Delta ^{(1)}\) has lattice width 1, and is determined by the Maroni invariants). However, there are trigonal curves of genus 4 for which \(\Delta \) is not intrinsic: there exist curves which are non-degenerate with respect to polygons \(\Delta \) and \(\Delta '\), with \(\Delta ^{(1)}=\Upsilon \) and \(\Delta '^{(1)}=\square _{1,1}\). Intrinsicness of the Newton polygon for tetragonal curves was studied in [5]: the Newton polygon \(\Delta \) is intrinsic if \(g(C)\bmod 4\in \{2,3\}\), but it might occasionally be not intrinsic if \(g(C)\bmod 4\in \{0,1\}\). From [3], it follows that non-degenerate smooth plane curves of degree \(d\ge 3\) (\(\Delta ^{(1)}\cong (d-3)\Sigma \)) and curves with Clifford dimension 3 (\(\Delta ^{(1)}\cong 2\Upsilon \)) have an intrinsic Newton polygon. Moreover, a partial result was given for non-degenerate curves on Hirzebruch surfaces \({\mathcal {H}}_n\): the value n is intrinsic.
In this paper, we examine intrinsicness of \(\Delta \) for curves on \({\mathbb {P}}^1\times {\mathbb {P}}^1\). Namely, we will show that a \(\Delta \)-non-degenerate curve C of genus \(g\ne 4\) can be embedded in \({\mathbb {P}}^1\times {\mathbb {P}}^1\) (if and) only if \(\Delta ^{(1)}=\emptyset \) or \(\Delta ^{(1)}\cong \square _{a,b}\) for \(a,b\in {\mathbb {Z}}_{\ge 0}\) satisfying \(g=(a+1)(b+1)\); see Theorem 18 in Sect. 3. In order to prove this result, we give a combinatorial interpretation for the first scrollar Betti numbers of \(\Delta \)-non-degenerate curves with respect to a gonality pencil, as soon as \(\Delta \) satisfies some mild conditions (see Sect. 2).
Notations Let \({\mathbb {P}}^N\) be a projective space with coordinates \((X_0:\ldots :X_N)\). For each projective variety \(V\subset {\mathbb {P}}^N\), we write \({\mathcal {I}}(V)\subset k[X_0,\ldots ,X_N]\) to indicate the homogeneous ideal of V and \({\mathcal {I}}_d(V)\subset {\mathcal {I}}(V)\) to indicate its homogeneous degree d piece. If \(J\subset k[X_0,\ldots ,X_N]\) is a homogeneous ideal, then \({\mathcal {Z}}(J)\subset {\mathbb {P}}^N\) is the zero locus of the polynomials in J.
2 First scrollar Betti numbers
2.1 Definition
We start by recalling the definition and some properties of rational normal scrolls.
Let \(n \in {\mathbb {Z}}_{\ge 2}\) and let \({\mathcal {E}}={\mathcal {O}}(e_1)\oplus \cdots \oplus {\mathcal {O}}(e_n)\) be a locally free sheaf of rank n on \({\mathbb {P}}^1\). Denote by \(\pi :{\mathbb {P}}({\mathcal {E}})\rightarrow {\mathbb {P}}^1\) the corresponding \({\mathbb {P}}^{n-1}\)-bundle. We assume that \(0\le e_1\le e_2\le \cdots \le e_n\) and that \(e_1 + e_2 + \dots + e_n \ge 2\). Set \(N=e_1+e_2+\cdots +e_n+n-1\). Then the image \(S=S(e_1,\ldots ,e_n)\) of the induced morphism
when composed with an isomorphism \({\mathbb {P}}H^0({\mathbb {P}}({\mathcal {E}}),{\mathcal {O}}_{{\mathbb {P}}({\mathcal {E}})}(1)) \rightarrow {\mathbb {P}}^N\), is called a rational normal scroll of type \((e_1, \dots , e_n)\). Up to automorphisms of \({\mathbb {P}}^N\), rational normal scrolls are fully characterized by their type.
They can also be described in a geometric way: consider linearly independent projective subspaces \({\mathbb {P}}_1^{e_1},\ldots ,{\mathbb {P}}_n^{e_n}\subset {\mathbb {P}}^N\) of dimensions \(e_1,\ldots ,e_n\), so their span is the whole projective space \({\mathbb {P}}^N\). For each \(i\in \{1,\ldots ,n\}\), fix a rational normal curve in \({\mathbb {P}}_i\) of degree \(e_i\) parametrized by a Veronese map \(\nu _i:{\mathbb {P}}^1\rightarrow {\mathbb {P}}_i^{e_i}\). Then
is a rational normal scroll of type \((e_1,\ldots ,e_n)\).
The scroll is smooth if and only if \(e_1>0\). In this case, \(\mu : {\mathbb {P}}({\mathcal {E}}) \rightarrow S\) is an isomorphism. If \(0=e_1=\cdots =e_\ell <e_{\ell +1}\) with \(1 \le \ell <n\), then the scroll is a cone with an \((\ell -1)\)-dimensional vertex. In this case \(\mu : {\mathbb {P}}({\mathcal {E}}) \rightarrow S\) is a resolution of singularities and
is an isomorphism for all integers \(\lambda >0\).
The Picard group of \({\mathbb {P}}({\mathcal {E}})\) is freely generated by the class H of a hyperplane section (more precisely, the class corresponding to \(\mu ^*{\mathcal {O}}_{{\mathbb {P}}^N}(1)\)) and the class R of a fiber of \(\pi \); i.e.
We have the following intersection products:
(where \(R^2 = 0\) means that any appearance of \(R^2\) annihilates the product). If we denote the class which corresponds to \(\mu _\lambda ^*{\mathcal {O}}_{{\mathbb {P}}^{N+n\lambda }}(1)\) by \(H'\), we obtain the equality \(H'=H+\lambda R\) in \(\text {Pic}({\mathbb {P}}({\mathcal {E}}))\).
Let C / k be a smooth projective curve of genus g and gonality \(\gamma \ge 4\). Assume that C is canonically embedded in \({\mathbb {P}}^{g-1}\) and fix a gonality pencil \(g^1_\gamma \) on C. By [6, Thm. 2],
is a \((\gamma -1)\)-dimensional rational normal scroll containing C. If S is of type \((e_1,\ldots ,e_{\gamma -1})\), the numbers \(e_1,\ldots ,e_{\gamma -1}\) are called the scrollar invariants of C with respect to \(g_{\gamma }^1\). Using the Riemann-Roch theorem, one can see that \(e_{\gamma -1}\le \frac{2g-2}{\gamma }\).
The following theorem extends a result from [9] on tetragonal and pentagonal curves to arbitrary curves.
Theorem 1
Let C be a canonically embedded smooth projective curve of genus g and gonality \(\gamma \ge 4\). If \(g^1_\gamma \) is a gonality pencil on C, let \(S\subset {\mathbb {P}}^{g-1}\) be the rational normal scroll swept out by \(g^1_\gamma \) and let \(C'\) be the strict transform of C under the resolution \(\mu :{\mathbb {P}}({\mathcal {E}}) \rightarrow S\). Then there exist effective divisors \(D_1,\ldots ,D_{(\gamma ^2-3\gamma )/2}\) on \({\mathbb {P}}({\mathcal {E}})\) along with integers \(b_1, \dots , b_{(\gamma ^2-3\gamma )/2}\), such that \(D_\ell \sim 2H-b_\ell R\) for all \(\ell \) and \(C'\) is the (scheme-theoretical) intersection of the \(D_\ell \)’s. Moreover, the multiset \(\{b_1,\ldots ,b_{(\gamma ^2-3\gamma )/2}\}\) does not depend on the choice of the \(D_\ell \)’s, and
Proof
Define \(\beta _i=\frac{i(\gamma -2-i)}{\gamma -1}{\gamma \atopwithdelims ()i+1}\) and note that \(\beta _1 = (\gamma ^2-3\gamma )/2\). The existence follows from [9, Cor. 4.4] and its proof, where the \(D_\ell \)’s come from an exact sequence of \({\mathcal {O}}_{{\mathbb {P}}({\mathcal {E}})}\)-modules
Tensoring (1) with \({\mathcal {O}}_{{\mathbb {P}}({\mathcal {E}})}(2H + bR)\) for a sufficiently large integer b and computing the Euler characteristics of the terms in the resulting exact sequence, one can show that
see [1, Prop. 2.9].
We are left with showing the independence of the multiset \(\{b_1,\ldots ,b_{(\gamma ^2-3\gamma )/2}\}\). Herefore, consider the exact sequence
If \(\pi :{\mathbb {P}}({\mathcal {E}})\rightarrow {\mathbb {P}}^1\) is the \({\mathbb {P}}^{\gamma -2}\)-bundle and \(\xi \) is the generic point of \({\mathbb {P}}^1\), then
where \({\mathcal {S}}=k(\xi )[x_0,\ldots ,x_{\gamma -2}]\). Applying \(\cdot \,\otimes _{{\mathbb {P}}^1} k(\xi )\) to (2) yields an exact sequence of graded \({\mathcal {S}}\)-modules, that can be extended to a minimal free resolution
of the coordinate ring \({\mathcal {S}}_{C'}\) of \(C'\) over \(k(\xi )\) (see [9, Lemma 4.2] and [2, Step A of Thm. 2.1]). As explained in [9, proof of Thm. 3.2] and [2, proof of Step B of Thm. 2.1], this resolution can be lifted to a minimal free resolution of \({\mathcal {O}}_{{\mathbb {P}}({\mathcal {E}})}\)-modules extending (2). This resolution is unique up to isomorphism by [9, Thm. 3.2] or [2, Thm. 1.3], which implies the independence. \(\square \)
We call the invariants \(b_1,\ldots ,b_{(\gamma ^2-3\gamma )/2}\) the first scrollar Betti numbers of C with respect to \(g^1_\gamma \). The main goal of this section is to give a combinatorial interpretation for these invariants for non-degenerate curves.
In [5], we already treated the case of tetragonal \(\Delta \)-non-degenerate curves: the first scrollar Betti numbers are given by
These numbers are independent from the choice of the gonality pencil. This will no longer be true for non-degenerate curves of higher gonality.
2.2 Scrollar invariants for non-degenerate curves
Let f be a \(\Delta \)-non-degenerate Laurent polynomial and consider the corresponding smooth curve \(C_f\subset \text {Tor}(\Delta )\subset {\mathbb {P}}^N\) with \(N=\sharp (\Delta \cap {\mathbb {Z}}^2)-1\). Assume that the polygon \(\Delta ^{(1)}\) is two-dimensional.
By [8], \(C_f\) is a non-rational and non-hyperelliptic curve and there exists a canonical divisor \(K_\Delta \) on \(C_f\) such that
(where x, y are functions on \(C_f\) through \(\varphi _\Delta \)). In particular, the curve \(C_f\) has genus \(g=\sharp (\Delta ^{(1)}\cap {\mathbb {Z}}^2)\ge 3\); see [3] for more details. Moreover, the Zariski closure \(C=C_f^{can}\) of the image of U(f) under
is a canonical model for \(C_f\). We end up with the inclusions
where T is a toric surface since \(\Delta ^{(1)}\) is two-dimensional.
A lattice direction is a primitive integer vector \(v=(a,b)\in {\mathbb {Z}}^2\). The width \(w(\Delta ,v)\) of \(\Delta \) with respect to a lattice direction v is the smallest integer \(\ell \) such that \(\Delta \) is contained in the strip \(k\le aY-bX\le k+\ell \) of \({\mathbb {R}}^2\) for some \(k\in {\mathbb {Z}}\). The lattice width is defined as \(\text {lw}(\Delta )=\min _v w(\Delta ,v)\). Lattice directions v that attain the minimum are called lattice-width directions.
In [3], we gave a combinatorial interpretation for the gonality \(\gamma \) of \(C=C_f^{can}\) (or \(C_f\)) in terms of the lattice width of \(\Delta \):
where we use our assumption that \(\Delta ^{(1)}\) is two-dimensional. From now on, we make the stronger assumption that \(\gamma =\text {lw}(\Delta )\ge 4\), and that \(\Delta ^{(1)}\) is not equivalent with \(k\Sigma \) for any k or \(\Upsilon \), hence \(\Delta \not \cong d\Sigma \) and \(\Delta \not \cong 2\Upsilon \). Then each lattice-width direction \(v=(a,b)\) gives rise to a rational map
of degree equal to the gonality \(\gamma \). We call the corresponding linear pencil \(g_\gamma ^1\) of C a combinatorial gonality pencil. If \(\Delta \) is sufficiently big (for a precise statement, see [3, Corollary 6.3]), each gonality pencil on C is combinatorial.
Fix a lattice-width direction v of \(\Delta \). After applying a suitable unimodular transformation \(\chi \), we may assume that \(v=(1,0)\) and that \(\Delta \) is contained in the horizontal strip \(0\le Y\le \gamma \) in \({\mathbb {R}}^2\). So, the gonality map \(C\dashrightarrow {\mathbb {P}}^1\) associated to v is the vertical projection to the x-axis. Write
for all \(j\in \{1,\ldots ,\gamma -1\}\). By [3, Theorem 9.1], the scrollar invariants \(e_1,\ldots ,e_{\gamma -1}\) of C with respect to \(g_\gamma ^1\) are equal to \(E_j := i^{(+)}(j)-i^{(-)}(j)\) for \(j\in \{1,\dots ,\gamma -1\}\) (up to order). In fact, a Zariski dense part of the scroll S is parametrized by
Note that \(T=\text {Tor}(\Delta ^{(1)})\subset S\) since the map \(\varphi _{\Delta ^{(1)}}\) can be obtained from the above parametrization by restricting to \(a_j=x^{i^{(-)}(j)}y^j\), so we get the inclusions
If S is singular, then \(\mu : S'=S(e_1+\lambda ,\ldots ,e_{\gamma -1}+\lambda )\cong {\mathbb {P}}({\mathcal {E}}) \rightarrow S\) is a resolution of singularities for each integer \(\lambda >0\) (hereby, we slightly abuse notation: \(\mu \) is the map \(\mu \circ \mu _\lambda ^{-1}\) using the notations in Sect. 2.1). Let \(C'\) and \(T'\) be the strict transforms of respectively C and T under \(\mu \). For each lattice polygon \(\Gamma \subset {\mathbb {R}}^2\), write \(\Gamma [\lambda ]\) to denote the Minkowski sum of \(\Gamma \) and \([(0,0),(\lambda ,0)]\subset {\mathbb {R}}^2\). In other words, \(\Gamma [\lambda ]\) is obtained from \(\Gamma \) by stretching it out in the horizontal direction over a distance \(\lambda \). Using this notation, one can see that \(T'=\text {Tor}(\Delta ^{(1)}[\lambda ])=\text {Tor}(\Delta [\lambda ]^{(1)})\). We end up with the inclusions
2.3 First scrollar Betti numbers of toric surfaces
Let C be a \(\Delta \)-non-degenerate curve and fix a combinatorial gonality pencil \(g^1_\gamma \) on C, corresponding to a lattice direction v. We work under the following assumptions:
-
(i)
\(\Delta ^{(1)}\) is not equivalent with \(k\Sigma \) for any k or \(\Upsilon \), and \(\gamma =\text {lw}(\Delta )\ge 4\),
-
(ii)
\(v=(1,0)\) and \(\Delta \) is contained in the horizontal strip \(0\le Y\le \gamma \), so that \(g^1_\gamma \) corresponds to the vertical projection,
-
(iii)
the curve C is canonically embedded, so that we obtain the sequence of inclusions \(C \subset T \subset S \subset {\mathbb {P}}^{g-1}\) from (4).
Recall that the scrollar invariants \(e_1, \dots , e_{\gamma - 1}\) of C with respect to \(g^1_\gamma \) match with \(E_1,\ldots ,E_{\gamma -1}\) (up to order). Consider \(\mu : {\mathbb {P}}({\mathcal {E}}) \rightarrow S\) and let \(T'\subset {\mathbb {P}}({\mathcal {E}})\) be the strict transform of \(T=\text {Tor}(\Delta ^{(1)})\) under \(\mu \), as in Sect. 2.2. If \(\Delta \) satisfies the condition \({\mathcal {P}}_1(v)\) defined below (see Definition 4), we will provide effective divisors \(D_1,\ldots ,D_{{\gamma -2 \atopwithdelims ()2}}\) on \({\mathbb {P}}({\mathcal {E}})\) along with integers \(b_1, \dots , b_{{\gamma -2 \atopwithdelims ()2}}\), such that the following three conditions are satisfied:
-
\(T'\) is the (scheme-theoretical) intersection of the \(D_\ell \)’s,
-
\(D_\ell \sim 2H-b_\ell R\) for all \(\ell \) (where H is the hyperplane class and R is the class of a fibre in \(\text {Pic}({\mathbb {P}}({\mathcal {E}}))\)), and,
-
\(\sum _{\ell =1}^{{\gamma -2 \atopwithdelims ()2}} b_\ell \, = (\gamma -4)g-(\gamma ^2-3\gamma )+\sharp (\partial \Delta ^{(1)}\cap {\mathbb {Z}}^2)\).
In what follows, we will also assume that \(e_1 > 0\), so that \({\mathbb {P}}({\mathcal {E}})\cong S\). This condition is not essential (see Remark 9), but it allows us to work with the inclusion \(T\subset S\) rather than \(T'\subset {\mathbb {P}}({\mathcal {E}})\). For convenience, we will use the notation \(D_{j_1,j_2}\) for the the divisors, where \(j_1,j_2\in \{1,\ldots ,\gamma -1\}\) such that \(j_2-j_1\ge 2\), and denote the corresponding invariants by \(B_{j_1,j_2}\). Below, we will first introduce divisors \(Y_{j_1,j_2,r}\) of S. Afterwards (see Definition 6), we will define the divisors \(D_{j_1,j_2}\) by means of the divisors \(Y_{j_1,j_2,r}\).
For each \(j_1,j_2\in \{1,\ldots ,\gamma -1\}\) such that \(j_2-j_1\ge 2\) and \(1\le r\le \frac{j_2-j_1}{2}\), let \(Y_{j_1,j_2,r}\subset S\) be the subvariety defined by the binomials of \({\mathcal {I}}_2(\text {Tor}(\Delta ^{(1)}))\) having the form
One can see that \(Y_{j_1,j_2,r}\) is a \((\gamma -2)\)-dimensional toric variety \(\text {Tor}(\Omega _{j_1,j_2,r})\), where \(\Omega _{j_1,j_2,r}\subset {\mathbb {R}}^{\gamma -2}\) is a full-dimensional lattice polytope (see Example 3 for a tangible instance). The (Euclidean) volume of this polytope equals
where \(\epsilon _{j_1,j_2,r}\) is defined as \(\epsilon ^{(-)}_{j_1,j_2,r}+\epsilon ^{(+)}_{j_1,j_2,r}\), with
and
In the above equalities for \(\epsilon ^{(-)}_{j_1,j_2,r}\) and \(\epsilon ^{(+)}_{j_1,j_2,r}\), we use the following result.
Lemma 2
The inequalities
and
hold for all \(j_1,j_2\in \{1,\ldots ,\gamma -1\}\) such that \(j_2-j_1\ge 2\) and \(1\le r\le \frac{j_2-j_1}{2}\).
Proof
We only show the first inequality; the second one follows by symmetry. Consider the line segment \(L=[(i^{(-)}(j_1),j_1),(i^{(-)}(j_2),j_2)]\), and let \((i',j_1+r)\) and \((i'',j_2-r)\) be the intersection points of L with the horizontal lines at heights \(j_1+r\) and \(j_2-r\). Note that L is contained in the interior of \(\Delta \) and that \(i'+i''=i^{(-)}(j_1)+i^{(-)}(j_2)\). If \(i^{(-)}(j_1+r)+i^{(-)}(j_2-r)\ge i^{(-)}(j_1)+i^{(-)}(j_2)+2=i'+i''+2\), then \(i'\le i^{(-)}(j_1+r)-1\) or \(i''\le i^{(-)}(j_2-r)-1\), so \((i^{(-)}(j_1+r)-1,j_1+r)\) or \((i^{(-)}(j_2-r)-1,j_2-r)\) is a lattice point lying in the interior of \(\Delta \). This is in contradiction with the definition of \(i^{(-)}(\cdot )\). \(\square \)
Example 3
Assume that \(\Delta = \Delta (f)\) is as in Fig. 1 (here \(\gamma = 5\)). Appropriate instances of \(\Omega _{j_1,j_2,r}\) can be realized as in Fig. 2. Here, \(\epsilon _{1,3,1}=1\) (since \(\epsilon ^{(+)}_{1,3,1}=1\)), \(\epsilon _{1,4,1}=0\) and \(\epsilon _{2,4,1}=1\) (since \(\epsilon ^{(-)}_{2,4,1}=1\)).
The intersection of \(Y_{j_1,j_2,r}\) with a typical fiber of \(S\rightarrow {\mathbb {P}}^1\) is a quadratic hypersurface, hence there is a \(B_{j_1,j_2,r}\in {\mathbb {Z}}\) such that \(Y_{j_1,j_2,r}\sim 2H-B_{j_1,j_2,r}R\). Taking the intersection product of the latter equation with \(H^{\gamma -2}\), we get
but \(\deg Y_{j_1,j_2,r}=(\gamma -2)!\cdot \text {Vol}(\Omega _{j_1,j_2,r})\), so \(B_{j_1,j_2,r}\) equals \(E_{j_1}+E_{j_2}-\epsilon _{j_1,j_2,r}\).
Write
and
Definition 4
We say that \(\Delta \) satisfies condition \({\mathcal {P}}_1(v)\) if and only if there are no integers \(j_1,j_2\in \{1,\ldots ,\gamma -1\}\) with \(j_2-j_1\ge 2\) such that \({\mathcal {S}}^{(-)}_{j_1,j_2}\) and \({\mathcal {S}}^{(+)}_{j_1,j_2}\) are non-empty and disjoint.
In other words, the condition \({\mathcal {P}}_1(v)\) means that for each pair of integers \(j_1,j_2\in \{1,\ldots ,\gamma -1\}\) with \(j_2-j_1\ge 2\) either at least one of the sets \({\mathcal {S}}^{(-)}_{j_1,j_2}\), \({\mathcal {S}}^{(+)}_{j_1,j_2}\) is empty, or there is a common \(r\in \left\{ 1,\ldots ,\left\lfloor \frac{j_2-j_1}{2}\right\rfloor \right\} \) for which \(\epsilon ^{(-)}_{j_1,j_2,r}=\epsilon ^{(+)}_{j_1,j_2,r}=0\). There is a useful criterion to check whether \({\mathcal {S}}^{(-)}_{j_1,j_2}\) is empty or not (and analogously for \({\mathcal {S}}^{(+)}_{j_1,j_2}\)): \({\mathcal {S}}^{(-)}_{j_1,j_2}=\emptyset \) if and only if all the lattice points \((i^{(-)}(j),j)\) with \(j_1<j<j_2\) lie strictly right from the line segment \(L=\left[ (i^{(-)}(j_1),j_1),(i^{(-)}(j_2),j_2)\right] \).
In the above definition, we also allow the lattice direction v to be different from (1, 0): in that case, first take a unimodular transformation \(\chi \) such that \(\chi (v)=(1,0)\) and that \(\chi (\Delta )\) is contained in the horizontal strip \(0\le Y\le \gamma \), and replace \(\Delta \) by \(\chi (\Delta )\) while checking the condition. The definition is independent of the particular choice of the unimodular transformation \(\chi \).
In fact, in some of the examples below, the lattice direction v is (1, 0), but \(\Delta \) is contained in a horizontal strip of the form \(k\le Y\le k+\gamma \) with \(k\ne 0\). In that case, we do not really need to apply any unimodular transformation \(\chi \) first: we can define the sets \({\mathcal {S}}^{(-)}_{j_1,j_2}\) and \({\mathcal {S}}^{(+)}_{j_1,j_2}\) for \(j_1,j_2\in \{k+1,\ldots ,k+\gamma -1\}\).
Example 5
Assume that a part of \(\Delta ^{(1)}\) looks as in Fig. 3 (for some large enough n). In Table 1, the sets \({\mathcal {S}}^{(-)}_{j_1,j_2}\) and \({\mathcal {S}}^{(+)}_{j_1,j_2}\) are given for all couples \((j_1,j_2)\) with \(j_1+j_2=15\) in this part of the polytope \(\Delta ^{(1)}\). We conclude that \(\Delta \) does not satisfy condition \({\mathcal {P}}_1(v)\) (consider \(j_1=0\) and \(j_2=15\)).
For all polygons \(\Delta \) that satisfy condition \({\mathcal {P}}_1(v)\), we give a recipe to construct the divisors \(D_{j_1,j_2}\) in terms of the subvarieties \(Y_{j_1,j_2,r}\).
Definition 6
Assume that the lattice polygon \(\Delta \) satisfies condition \({\mathcal {P}}_1(v)\).
-
If \({\mathcal {S}}^{(-)}_{j_1,j_2}\cap {\mathcal {S}}^{(+)}_{j_1,j_2}\ne \emptyset \), we define \(D_{j_1,j_2}\) as \(Y_{j_1,j_2,r}\) with \(r\in {\mathcal {S}}^{(-)}_{j_1,j_2}\cap {\mathcal {S}}^{(+)}_{j_1,j_2}\) minimal. Set \(\epsilon _{j_1,j_2}=\epsilon _{j_1,j_2,r}=0\).
-
If \({\mathcal {S}}^{(-)}_{j_1,j_2}=\emptyset \) and \({\mathcal {S}}^{(+)}_{j_1,j_2}\ne \emptyset \) or vice versa, take \(r\in {\mathcal {S}}^{(-)}_{j_1,j_2}\cup {\mathcal {S}}^{(+)}_{j_1,j_2}\) minimal, define \(D_{j_1,j_2}=Y_{j_1,j_2,r}\) and set \(\epsilon _{j_1,j_2}=\epsilon _{j_1,j_2,r}=1\).
-
If \({\mathcal {S}}^{(-)}_{j_1,j_2}={\mathcal {S}}^{(+)}_{j_1,j_2}=\emptyset \), define \(D_{j_1,j_2}=Y_{j_1,j_2,1}\) and set \(\epsilon _{j_1,j_2}=\epsilon _{j_1,j_2,1}=2\).
Remark 7
In Definition 6, the divisor \(D_{j_1,j_2}\) is always of the form \(Y_{j_1,j_2,r}\) and r is chosen such that \(\epsilon _{j_1,j_2,r}\) is minimal, or equivalently, \(B_{j_1,j_2,r}\) is maximal. Moreover, if \(D_{j_1,j_2}=Y_{j_1,j_2,r}\) and if we define \(\epsilon _{j_1,j_2}^{(-)}=\epsilon _{j_1,j_2,r}^{(-)}\) and \(\epsilon _{j_1,j_2}^{(+)}=\epsilon _{j_1,j_2,r}^{(+)}\), then
Here, it is crucial that \(\Delta \) satisfies condition \({\mathcal {P}}_1(v)\): if \({\mathcal {S}}^{(-)}_{j_1,j_2}\) and \({\mathcal {S}}^{(+)}_{j_1,j_2}\) were non-empty and disjoint, then \(\min _r \epsilon _{j_1,j_2,r} = 1\) (take \(r\in {\mathcal {S}}^{(-)}_{j_1,j_2}\cup {\mathcal {S}}^{(+)}_{j_1,j_2}\)), but \(\min _s \epsilon _{j_1,j_2,s}^{(-)}=\min _t \epsilon _{j_1,j_2,t}^{(+)}=0\).
If we set \(B_{j_1,j_2}=E_{j_1}+E_{j_2}-\epsilon _{j_1,j_2}\), we have that \(D_{j_1,j_2}\sim 2H-B_{j_1,j_2}R\) and
Example 8
If \(\partial \Delta ^{(1)}\) meets each horizontal line of height \(j\in \{2,\ldots ,\gamma -2\}\) in two lattice points, we have \(\epsilon _{j_1,j_2,r}=0\) and \({\mathcal {S}}^{(-)}_{j_1,j_2}={\mathcal {S}}^{(+)}_{j_1,j_2}=\left\{ 1,\ldots ,\left\lfloor \frac{j_2-j_1}{2}\right\rfloor \right\} \) for all \(j_1,j_2,r\). Hence, \(\Delta \) satisfies condition \({\mathcal {P}}_1(v)\). Moreover, \(\epsilon _{j_1,j_2}=0\) and \(D_{j_1,j_2}=Y_{j_1,j_2,1}\) for all \(j_1,j_2\). In this case,
and \(\sum \epsilon _{j_1,j_2}=0\), so
Remark 9
If S is singular, let \(\lambda >0\) be an integer and consider the inclusions from (5). Note that \(\Delta [\lambda ]\) satisfies condition \({\mathcal {P}}_1(v)\) if and only if \(\Delta \) satisfies condition \({\mathcal {P}}_1(v)\). We can define the subvarieties \(Y_{j_1,j_2,r}\) and \(D_{j_1,j_2}\) of \(S'\) in the same way as we did before (using \(\Delta [\lambda ]\) instead of \(\Delta \)). Since \(H'=H+\lambda R\), we get that
and \(D_{j_1,j_2}\sim 2H-B_{j_1,j_2}R\).
We are now able to state and prove the main result of this subsection.
Theorem 10
If \(\Delta \) satisfies condition \({\mathcal {P}}_1(v)\), there exist \({\gamma -2 \atopwithdelims ()2}\) effective divisors \(D_{j_1,j_2}\) on \({\mathbb {P}}({\mathcal {E}})\) (with \(j_1,j_2\in \{1,\ldots ,\gamma -1\}\) and \(j_2-j_1\ge 2\)) such that
-
\(T'\) is the (scheme-theoretical) intersection of the divisors \(D_{j_1,j_2}\),
-
\(D_{j_1,j_2}\sim 2H-B_{j_1,j_2}R\) for all \(j_1,j_2\), where \(B_{j_1,j_2} = E_{j_1} + E_{j_2} - \epsilon _{j_1,j_2}\), and,
-
\(\sum _{j_2-j_1\ge 2}\,B_{j_1,j_2} = (\gamma -4)g-(\gamma ^2-3\gamma ) +\sharp (\partial \Delta ^{(1)}\cap {\mathbb {Z}}^2)\).
Proof
By Remark 9, we may assume that S is smooth, hence \({\mathbb {P}}({\mathcal {E}})\cong S\). We need to prove that \({\mathcal {I}}(\text {Tor}(\Delta ^{(1)}))={\mathcal {I}}(\bigcap D_{j_1,j_2})\), where the inclusion \({\mathcal {I}}(\bigcap D_{j_1,j_2})\subset {\mathcal {I}}(\text {Tor}(\Delta ^{(1)}))\) is trivial. Pick an arbitrary quadratic binomial
These binomials generate the ideal, so we only need to show that \(f\in {\mathcal {I}}(\bigcap D_{j_1,j_2})\). Note that \(j_1+j_2=j_3+j_4\), so we may assume that \(j_1\le j_3\le j_4\le j_2\). Moreover, if \(j_1=j_3\) and \(j_4=j_2\), we get that \(f\in {\mathcal {I}}(S)\subset {\mathcal {I}}(\bigcap D_{j_1,j_2})\). So we may even assume that \(j_1<j_3\).
Take r such that \(D_{j_1,j_2}=Y_{j_1,j_2,r}\). We claim that
If \(\epsilon _{j_1,j_2}^{(-)}=\epsilon _{j_1,j_2,r}^{(-)}=0\), this follows from
If \(\epsilon _{j_1,j_2}^{(-)}=\epsilon _{j_1,j_2,r}^{(-)}=1\), we have that \(\epsilon _{j_1,j_2,j_3-j_1}^{(-)}=1\) by (6) (since \(\Delta \) satisfies condition \({\mathcal {P}}_1(v)\)), hence
where we use Lemma 2. Analogously, we can show that \(I\le i^{(+)}(j_1+r)+i^{(+)}(j_2-r)\).
The above claim implies that we can find integers \(i'_1,i'_2\) such that \(i'_1+i'_2=I\), \(i^{(-)}(j_1+r)\le i'_1 \le i^{(+)}(j_1+r)\), \(i^{(-)}(j_2-r)\le i'_2 \le i^{(+)}(j_2-r)\), hence
So we may replace the term \(X_{i_1,j_1}X_{i_2,j_2}\) in f by \(X_{i'_1,j_1+r}X_{i'_2,j_2-r}\) (and in particular, \(j_1\) by \(j_1+r\) and \(j_2\) by \(j_2-r\)). Continuing in this way, we will eventually get that \(j_1=j_3\) and \(j_4=j_2\), hence \(f\in {\mathcal {I}}(S)\). This will happen after a finite number of steps since the maximum of \(j_2-j_1\) and \(j_4-j_3\) decreases after each step.
We are left with proving the formula for the sum of the \(B_{j_1,j_2}\)’s. By Remark 7 and the elaboration of Example 8, it suffices to show that the sum of the \(\epsilon _{j_1,j_2}\) counts the number of times that \(\partial \Delta ^{(1)}\) intersects the horizontal lines of height \(2,\ldots ,\gamma -2\) in a non-lattice point. Let \(A^{(-)}\) be the set of couples \((j_1,j_2)\) such that \(j_1,j_2\in \{1,\ldots ,\gamma -1\}\), \(j_2-j_1\ge 2\) and \({\mathcal {S}}_{j_1,j_2}^{(-)}=\emptyset \) (or equivalently, the line segment \(L=\left[ (i^{(-)}(j_1),j_1),(i^{(-)}(j_2),j_2)\right] \) passes left from all the lattice points \((i^{(-)}(j'),j')\) with \(j_1<j'<j_2\)). Let \(B^{(-)}\) be the set of integers \(j\in \{1,\ldots ,\gamma -1\}\) such that \((i^{(-)}(j),j)\not \in \partial \Delta ^{(1)}\). We claim that the sets \(A^{(-)}\) and \(B^{(-)}\) have the same cardinality. We will do this by giving a concrete bijection between these sets. Analogously, we can define the sets \(A^{(+)}\) and \(B^{(+)}\), and prove that they have the same number of elements. The theorem follows directly, since \(\sharp (A^{(-)}\cup A^{(+)})=\sum _{j_1,j_2}\,\epsilon _{j_1,j_2}\) by (6) and \(\sharp (B^{(-)}\cup B^{(+)})\) is the number of non-lattice point intersections.
If \((j_1,j_2)\in A^{(-)}\), then the line segment \(L=[(i^{(-)}(j_1),j_1),(i^{(-)}(j_2),j_2)]\) will pass at the left hand side of the lattice points \((i^{(-)}(j),j)\) with \(j_1<j<j_2\). For precisely one of these lattice points, the horizontal distance to L will be equal to the minimal value \(\frac{1}{j_2-j_1}\). Consider the map
sending the couple \((j_1,j_2)\) to the value of j of that lattice point; see below for an example. On the other hand, if \(j\in B^{(-)}\), thus \((i^{(-)}(j),j)\not \in \partial \Delta ^{(1)}\), then there should be lattice points \((i^{(-)}(j_1),j_1)\) and \((i^{(-)}(j_2),j_2)\) with \(j_1<j<j_2\) such that \(L=[(i^{(-)}(j_1),j_1),(i^{(-)}(j_2),j_2)]\) passes left from \((i^{(-)}(j),j)\). If we take a couple \((j_1,j_2)\) that satisfies this property and has a minimal value for \(j_2-j_1\), then \((j_1,j_2)\in A^{(-)}\). Indeed, if \((i^{(-)}(j'),j')\) with \(j_1<j'<j_2\) lies on or left from L, then either \((j_1,j')\) or \((j',j_2)\) would also satisfy the condition and would have a smaller value for the difference of the heights. Now let’s show that the couple \((j_1,j_2)\) is unique. If not, there exists another couple \((j_1',j_2')\in A^{(-)}\) with \(j_1'<j<j_2'\) such that \(L'=[(i^{(-)}(j_1'),j_1'),(i^{(-)}(j_2'),j_2')]\) passes left from \((i^{(-)}(j),j)\) with \(j_2'-j_1'=j_2-j_1\). We may assume that \(j_1'<j_1<j<j_2'<j_2\). Then L passes left from \((i^{(-)}(j_2'),j_2')\) and \(L'\) passes left from \((i^{(-)}(j_1),j_1)\), so \(L''=[(i^{(-)}(j_1'),j_1'),(i^{(-)}(j_2),j_2)]\) passes left from all the lattice points \((i^{(-)}(j'),j')\) with \(j_1'<j'<j_2\) (see Fig. 4). Let’s denote the horizontal distance from the line segment \(L''\) to the lattice point \((i^{(-)}(j'),j')\) by \(d(j')\). Using Lemma 2, we obtain that \(d(j_1'+r)+d(j_2-r)=1\) for all \(1\le r\le \frac{j_2-j_1'}{2}\). For precisely one integer \(j_1'<j'<j_2\), the distance \(d(j')\) is equal to the minimal value \(\frac{1}{j_2-j_1'}\). On the other hand, except for \(j'=j_1\) or \(j'=j_2'\), the distance \(d(j')\) has to be at least \(\frac{1}{j_2-j_1}=\frac{1}{j_2'-j_1'}\), since \((i^{(-)}(j'),j')\) lies strictly right from L or \(L'\). So we may assume that \(j'=j_1\) (the case \(j'=j_2'\) is analogous), hence \(d(j_1)=\frac{1}{j_2-j_1'}\) and \(d(j_2')=1-\frac{1}{j_2-j_1'}\) (using \(r=j_1-j_1'\) above). It follows that the horizontal distance to \(L''\) from the point on \(L'\) on height \(j_1\) is equal to
so \(L'\) does not pass left from \((i^{(-)}(j_1),j_1)\), a contradiction. In conclusion we can consider the map
sending j to the unique such couple \((j_1,j_2)\).
The maps \(\alpha ^{(-)}\) and \(\beta ^{(-)}\) are inverse of each other. For instance, to prove that the map \(\alpha ^{(-)}\circ \beta ^{(-)}\) is the identity map on \(B^{(-)}\), consider \(j\in B^{(-)}\) and write \(\beta ^{(-)}(j)=(j_1,j_2)\). If \(\alpha ^{(-)}(j_1,j_2)=j'\ne j\), then the horizontal distance from \((i^{(-)}(j),j)\) to \(L=[(i^{(-)}(j_1),j_1),(i^{(-)}(j_2),j_2)]\) is of the form \(\frac{d}{j_2-j_1}\) with \(1<d<j_2-j_1\). But then either \(L'=[(i^{(-)}(j'),j'),(i^{(-)}(j_2),j_2)]\) (the case \(j'<j\)) or \(L''=[(i^{(-)}(j_1),j_1),(i^{(-)}(j'),j')]\) (the case \(j'>j\)) passes left from \((i^{(-)}(j),j)\). This is in contradiction with \(\beta ^{(-)}(j)=(j_1,j_2)\), since \(j_2-j'\) and \(j'-j_1\) are both strictly smaller than \(j_2-j_1\). We leave the proof of the equality \(\beta ^{(-)}\circ \alpha ^{(-)}=\text {Id}_{A^{(-)}}\) as an exercise. \(\square \)
Example 11
Consider a polygon \(\Delta \) of which a part of the boundary of \(\Delta ^{(1)}\) is as in Fig. 5 (the (i, j)-coordinates are translated a bit). For this horizontal slice of the polygon,
The map \(\alpha ^{(-)}\) is defined as follows:
One can show that the \(D_{j_1,j_2}\)’s in Theorem 10 can be used to resolve \({\mathcal {O}}_{T'}\) as an \({\mathcal {O}}_{{\mathbb {P}}({\mathcal {E}})}\)-module, following Schreyer [9]. For this one needs that the fibers of \(\pi |_{T'}:T'\rightarrow {\mathbb {P}}^1\) have constant Betti numbers and that the corresponding resolutions are pure, but this can be verified. So it is justified to call the \(B_{j_1,j_2}\)’s the first scrollar Betti numbers of the toric surface \(\text {Tor}(\Delta ^{(1)})\), even though we will not push this discussion further.
2.4 First scrollar Betti numbers of non-degenerate curves relative to the toric surface
We will use the same set-up and assumptions as in the beginning of Sect. 2.3. The assumption (i) implies that \(\Delta ^{(2)}\ne \emptyset \). Moreover, we also use the notations appearing in the inclusions (5), so \(C'\) and \(T'\) are the strict transforms of the canonically embedded \(\Delta \)-non-degenerate curve C and the toric surface T under the resolution \(\mu :S'\cong {\mathbb {P}}({\mathcal {E}}) \rightarrow S\). In this section, we will present divisors on \(S'\) that scheme-theoretically cut out \(C'\) from \(T'\).
Herefore, we rely on the following construction from [4]. Write
and consider \(w\in \Delta ^{(2)}\cap {\mathbb {Z}}^2\). For each \((i,j)\in \Delta \cap {\mathbb {Z}}^2\), there exist \(u_{i,j},v_{i,j}\in \Delta ^{(1)}\cap {\mathbb {Z}}^2\) such that \((i,j) - w = (u_{i,j} - w) + (v_{i,j} - w ).\) Hereby, we use that \(\Delta +\Delta ^{(2)}\subset 2\Delta ^{(1)}\) and that the polygon \(\Delta ^{(1)}\) is normal. Then the quadrics
where w ranges over \(\Delta ^{(2)}\cap {\mathbb {Z}}^2\), scheme-theoretically cut out C from T.
In order to create the divisors \(D_\ell \), we will need an extra condition on \(\Delta \), which garantees that we can choose the lattice points \(u_{i,j},v_{i,j}\) in a particular way.
Definition 12
We say that \(\Delta \) satisfies condition \({\mathcal {P}}_2(v)\) if for each lattice point (i, j) of \(\Delta \) and each horizontal line L, there exist two (not necessarily distinct) horizontal lines \(M_1,M_2\), such that for all \(w \in L \cap \Delta ^{(2)} \cap {\mathbb {Z}}^2\), there exist \(u_{i,j} \in M_1 \cap \Delta ^{(1)} \cap {\mathbb {Z}}^2\) and \(v_{i,j} \in M_2 \cap \Delta ^{(1)} \cap {\mathbb {Z}}^2\) (dependent on (i, j) and w) such that
Remark 13
Write
for all \(j\in \{2,\ldots ,\gamma -2\}\). An equivalent definition is as follows: \(\Delta \) satisfies condition \({\mathcal {P}}_2(v)\) if and only if for all \((i,j)\in \Delta \) and for all \(j'\in \{2,\ldots ,\gamma -2\}\), there exist \(j_1,j_2\in \{1,\ldots ,\gamma -1\}\) such that \(j_1+j_2=j+j'\) and
This condition is obviously satisfied for \((i,j) \in \Delta ^{(1)}\) (take \(j_1=j\) and \(j_2=j'\)). Moreover, the condition also holds if (i, j) lies on the interior of a horizontal edge (i.e. the top or bottom edge) of \(\Delta \). Indeed, assume for instance that (i, j) lies in the interior of the top edge \([(i^-,j),(i^+,j)]\) of \(\Delta \). We have that
Hereby, the first inequality follows by replacing L in the proof of Lemma 2 by the half-closed line segment \([(i^{(--)}(j'),j'),(i^-,j)[\). Analogously, we get that
so (7) follows for \(j_1=j'+1\) and \(j_2=j-1\).
Although at first sight the condition \({\mathcal {P}}_2(v)\) might seem strong, it is not so easy to cook up instances of lattice polygons \(\Delta \) for which the condition is not satisfied. The smallest example we have found is a polygon with 46 interior lattice points and lattice width 10.
Example 14
Let \(\Delta \) be as in Fig. 6 (the dashed line indicates \(\Delta ^{(1)}\)). We claim that \(\Delta \) does not satisfy condition \({\mathcal {P}}_2(v)\). Indeed, take the top vertex \((i,j)=(4,10)\) of \(\Delta \) and the horizontal line L at height 6. For the point \(w\in L \cap \Delta ^{(2)} \cap {\mathbb {Z}}^2\), consider the bold-marked lattice points (3, 6) and (6, 6) on L. In both cases, there is a unique decomposition of \((i,j) - w\):
So one sees that it is impossible to take the \(u_{i,j}\)’s and/or the \(v_{i,j}\)’s on the same line, which proves the claim.
Theorem 15
If \(\Delta \) satisfies condition \({\mathcal {P}}_2(v)\), then there exist \(\gamma -3\) effective divisors \(D_\ell \) on \({\mathbb {P}}({\mathcal {E}})\) (with \(2\le \ell \le \gamma -2\)) such that
-
\(C'\) is the (scheme-theoretical) intersection of \(T'\) and the divisors \(D_\ell \),
-
\(D_\ell \sim 2H-B_\ell R\) for all \(\ell \), where
$$\begin{aligned} B_\ell =i^{(++)}(\ell )-i^{(--)}(\ell )=-1+\sharp \{(i,j)\in \Delta ^{(2)}\cap {\mathbb {Z}}^2\,|\,j=\ell \}, \end{aligned}$$so
$$\begin{aligned} \sum _{2\le \ell \le \gamma -2}\,B_\ell =\sharp (\Delta ^{(2)}\cap {\mathbb {Z}}^2)-(\gamma -3). \end{aligned}$$
Proof
The formula for the sum \(\sum _\ell \,B_\ell \) is easily verified, so we focus on the other assertions. Take \(\lambda \ge 0\) so that \(S'=S(e_1+\lambda ,\ldots ,e_{\gamma -1}+\lambda )\) is smooth (and isomorphic to \({\mathbb {P}}({\mathcal {E}})\)) and define \(\Delta '=\Delta [\lambda ]\). We are going to use the inclusions
from (5), where \(T'=\text {Tor}(\Delta '^{(1)})\). Write \((X_{i,j})_{(i,j)\in \Delta '^{(1)}\cap {\mathbb {Z}}^2}\) for the projective coordinates on \({\mathbb {P}}^{g-1+\lambda (\gamma -1)}\).
Let \(\ell \in \{2,\ldots ,\gamma -2\}\) and denote the lattice points of \(\Delta [2\lambda ]^{(2)}\) of height \(\ell \) by \(w_0,\ldots ,w_{B_\ell +2\lambda }\). If \(w\in \{w_0,\ldots ,w_{B_\ell +2\lambda }\}\) and \((i,j)\in \Delta \), then we claim that we can find \(u_{i,j}, v_{i,j}\in \Delta '^{(1)}\) such that \((i,j)-w=(u_{i,j}-w)+(v_{i,j}-w)\), in such a way that their second coordinates are independent from w. Indeed, since \(\Delta \) satisfies condition \({\mathcal {P}}_2(v)\), there exist \(j_1,j_2\in \{1,\ldots ,\gamma -1\}\) such that \(j_1+j_2=j+\ell \) and
hence
This implies that we can take \(u_{i,j}\) and \(v_{i,j}\) with second coordinates \(j_1\) and \(j_2\). Define
A consequence of the choice of \(u_{i,j}, v_{i,j}\) is that
(rather than just \({\mathcal {I}}(T')\)) for all \(r\in \{0,\ldots ,B_\ell +2\lambda -1\}\) and \(s\in \{0,\ldots ,B_\ell +\lambda -1\}\). Since
is a local parameter for the \((\gamma -2)\)-plane \(R_{(0:1)} = \pi ^{-1}(0:1) \subset S'\), it follows that the \(R_{(0:1)}\)-orders of
increase by 1 at each step. For a similar reason, with \(R_{(1:0)} = \pi ^{-1}(1:0)\subset S'\), the \(R_{(1:0)}\)-orders of (8) decrease by 1 at each step. We conclude that there exists an effective divisor \(D_\ell \) such that for all \(i\in \{0,\dots ,B_\ell +2\lambda \}\) we have
on \(S'\). The divisor \(D_\ell \) is in fact the divisor of \(S'\) cut out by the quadrics in (8). Using (9) and Remark 9, we get that
so it is sufficient to show that the quadrics \(Q_w\) (where w ranges over \(\Delta [2\lambda ]^{(2)}\cap {\mathbb {Z}}^2\)) cut out \(C'\) from \(T'\). If \(\lambda =0\), this follows from [4, Theorem 3.3].
Before we prove this, we need to introduce one more notion: for each lattice polygon \(\Gamma \) with two-dimensional \(\Gamma ^{(1)}\), write \(\Gamma ^{max}\) to denote the largest lattice polygon with interior lattice polygon equal to \(\Gamma ^{(1)}\), so \(\Gamma ^{max}\supset \Gamma \). The polygon \(\Gamma ^{max}\) can be constructed as follows. Let \(v_1,\ldots ,v_r\) be the primitive inward pointing normal vectors of the edges of \(\Gamma ^{(1)}\) and write \(\Gamma ^{(1)}\) as an intersection \(\cap _{t=1}^r\,H_t\) of half-planes
(where \(\langle \cdot ,\cdot \rangle \) denotes the standard inner product and \(a_t\in {\mathbb {Z}}\)). Then
We will use the following two properties (see [3, Section 2] for other properties of \(\Gamma ^{max}\)):
-
If \(\Gamma ^{(2)}\ne \emptyset \), then \(2\Gamma ^{(1)}=\Gamma ^{(2)}+\Gamma ^{max}\), since both lattice polygons are defined by the half-planes \(2H_t=\{P\in {\mathbb {R}}^2\,|\,\langle P,v_t\rangle \ge -2a_t\}\).
-
If \(\Phi _1,\Phi _2\) are lattice polygons such that \(\Phi _1 + \Gamma \subset \Phi _2 + \Gamma ^{max}\), then \(\Phi _1\subset \Phi _2\) if \(\Phi _2\) satisfies the following condition: it is the intersection of half-planes \(H'_t\) with \(H'_t\) of the form
$$\begin{aligned} \left\{ P\in {\mathbb {R}}^2\,|\,\langle P,v_t\rangle \ge b_t\right\} \end{aligned}$$for some \(b_t\in {\mathbb {Z}}\) (hence parallel to \(H_t\)).
Indeed, if \(\Phi _1\not \subset \Phi _2\), take a lattice point P in \(\Phi _1\setminus \Phi _2\). Then \(\langle P,v_t\rangle <b_t\) for some value of t. Take \(Q\in \Gamma \) with \(\langle Q,v_t\rangle =a_t-1\) (this is always possible). We have that \(P+Q\in \Phi _1+\Gamma \) and \(\langle P+Q,v_t\rangle <a_t+b_t-1\), but \(\Phi _2 + \Gamma ^{max}\) is the intersection of the half-planes \(H''_t=\{P\in {\mathbb {R}}^2\,|\,\langle P,v_t\rangle \ge a_t+b_t-1\}\), so \(P+Q\not \in \Phi _2 + \Gamma ^{max}\), a contradiction.
Now take \(F\in {\mathcal {I}}(C')\) homogeneous of degree d and let
be the ring morphism that maps \(X_{i,j}\) to \(x^i y^j\). Since \(\xi (F)(x,y)=0\) for all \((x,y)\in {\mathbb {T}}^2\) with \(f(x,y)=0\) (and f is irreducible), the Laurent polynomial \(\xi (F)\) has to be of the form cf for some \(c\in k[x^{\pm 1},y^{\pm 1}]\). The Newton polygon of cf is equal to \(\Delta (c)+\Delta \), while the Newton polygon \(\Delta (\xi (F))\) is contained in
(here, we use the first property of maximal polygons with \(\Gamma =\Delta '\)). So we obtain that
Now we can use the second property of maximal polygons with \(\Phi _1=\Delta (c)\), \(\Phi _2=(d-2)\Delta '^{(1)}+\Delta [2\lambda ]^{(2)}\) and \(\Gamma =\Delta \). Note that \(\Phi _2\) might have a horizontal (top or bottom) edge while \(\Delta ^{(1)}\) has not, but this is not an issue (since \(\Delta ^{(1)}\not \cong k\Sigma \)). It follows that
So we can write
for polynomials \(g_{i,j}\in k[x,y]\) with \(\Delta (g_{i,j})\subset (d-2)\Delta '^{(1)}\). For all lattice points \(w=(i,j)\in \Delta [2\lambda ]^{(2)}\cap {\mathbb {Z}}^2\), there is a homogeneous polynomial \(G_{i,j}\in k[X_{i,j}]_{(i,j)\in \Delta '^{(1)}\cap {\mathbb {Z}}^2}\) such that \(\xi (G_{i,j})=g_{i,j}\). On the other hand, \(\xi (Q_w)=x^i y^j f\), hence
So \(F-\sum _{w=(i,j)}\, G_{i,j}Q_w\) belongs to the kernel of the map \(\xi \), which implies that it is contained in \({\mathcal {I}}_d(T')\), which is what we wanted to prove. \(\square \)
2.5 First scrollar Betti numbers for non-degenerate curves
We are ready to prove the main result of this section, by combining the results from Sects. 2.3 and 2.4.
Theorem 16
Let \(\Delta \) be a lattice polygon with \(\text {lw}(\Delta )\ge 4\) such that \(\Delta ^{(1)}\not \cong \Upsilon \) and \(\Delta ^{(1)}\not \cong k\Sigma \) for any integer k. Assume that \(\Delta \) satisfies the conditions \({\mathcal {P}}_1(v)\) and \({\mathcal {P}}_2(v)\), where v is a lattice-width direction. Let C be a \(\Delta \)-non-degenerate curve and let \(g_\gamma ^1\) be the combinatorial gonality pencil on C corresponding to v (with \(\gamma =\text {lw}(\Delta )\)). Then the first scrollar Betti numbers of C with respect to \(g_\gamma ^1\) are given by
Proof
We use the notations and set-up from Sect. 2.3. Theorems 10 and 15 imply that there exist divisors \(D_\ell \sim 2H-B_\ell R\) on \({\mathbb {P}}({\mathcal {E}})\), with \(\ell \in \{2,\ldots ,\gamma -2\}\), and divisors \(D_{j_1,j_2}\sim 2H-B_{j_1,j_2}R\) on \({\mathbb {P}}({\mathcal {E}})\), with \(j_1,j_2\in \{1,\ldots ,\gamma -1\}\) and \(j_2-j_1\ge 2\), such that \(C'\) is the scheme-theoretical intersection of these divisors. So we can use Theorem 1 to conclude the proof. Note that indeed
as announced in the theorem. \(\square \)
We believe that the above theorem is of independent interest. For instance it is not well-understood which sets of (first) scrollar Betti numbers are possible for canonical curves of a given genus and gonality, and our result can be used to prove certain existence results. It has been conjectured that “most” curves have so-called balanced (first) scrollar Betti numbers, meaning that \(\max |b_i-b_j|\le 1\), see [1] and the references therein. Non-degenerate curves are typically highly non-balanced, since one expects the \(B_{j_1,j_2}\)’s to be about twice the \(B_\ell \)’s.
Example 17
Consider the following lattice polygons \(\Delta _1\) and \(\Delta _2\) of lattice width 7 (and lattice-width direction \(v=(1,0)\)), which only differ from each other at the right hand side.
The polygon \(\Delta _1\) does not satisfy all the combinatorial constraints of Theorem 16, since condition \({\mathcal {P}}_2(v)\) does not hold: \({\mathcal {S}}_{2,6}^{(-)}=\{1\}\) and \({\mathcal {S}}_{2,6}^{(+)}=\{2\}\). Although we have not pursued this, we believe that the conditions \({\mathcal {P}}_1(v)\) and \({\mathcal {P}}_2(v)\) are always fulfilled if \(\gamma <7\).
On the other hand, the polygon \(\Delta _2\) meets all the conditions from the statement, and so we can apply Theorem 16. The first scrollar Betti numbers of a \(\Delta _2\)-degenerate curve are as follows:
The sum of these numbers is 108, which agrees with \((\gamma -3)(g-\gamma -1)\) for \(g=35\) and \(\gamma =7\).
3 Intrinsicness on \({\mathbb {P}}^1\times {\mathbb {P}}^1\)
Theorem 18
Let \(f \in k[x^{\pm 1},y^{\pm 1}]\) be non-degenerate with respect to its (two-dimensional) Newton polygon \(\Delta = \Delta (f)\), and assume that \(\Delta \not \cong 2\Upsilon \). Then U(f) is birationally equivalent to a smooth projective genus g curve in \({\mathbb {P}}^1 \times {\mathbb {P}}^1\) if and only if \(\Delta ^{(1)} = \emptyset \) or \(\Delta ^{(1)} \cong \square _{a,b}\) for some integers \(a\ge b \ge 0\), necessarily satisfying \(g=(a+1)(b+1)\).
Proof
We may assume that U(f) is neither rational, nor elliptic or hyperelliptic (and hence that \(\Delta ^{(1)}\) is two-dimensional) because such curves admit smooth complete models in \({\mathbb {P}}^1 \times {\mathbb {P}}^1\). So for the ‘if’ part we can assume that \(b \ge 1\). But then \(\text {Tor}(\Delta ^{(1)}) \cong {\mathbb {P}}^1 \times {\mathbb {P}}^1\), and the statement follows using the canonical embedding (3).
The real deal is the ‘only if’ part. At least, if a curve C / k is birationally equivalent to a (non-rational, non-elliptic, non-hyperelliptic) smooth projective curve in \({\mathbb {P}}^1 \times {\mathbb {P}}^1\), then it is \(\Delta '\)-non-degenerate with \(\Delta ' = [-1, a+1] \times [-1, b+1]\) for \(a \ge b \ge 1\): this follows from the material in [3, Section 4] (one can use an automorphism of \({\mathbb {P}}^1 \times {\mathbb {P}}^1\) to ensure appropriate behavior with respect to the toric boundary). Note that \(\Delta '^{(1)} = \square _{a,b}\). The geometric genus of C equals \(g=(a+1)(b+1)\) by [8] and its gonality equals \(\gamma = b + 2\) by [3, Cor. 6.2]. We observe that
-
g is composite.
-
C has Clifford dimension equal to 1 by [3, Theorem 8.1].
-
the scrollar invariants of C (with respect to any gonality pencil) are all equal to \(g/(\gamma - 1) - 1\). Indeed, by [3, Theorem 6.1], every gonality pencil is computed by projecting along some lattice-width direction v. If \(a > b\), then the only pair of lattice-width directions is \(\pm (1, 0)\) and from [3, Theorem 9.1], we find that the corresponding scrollar invariants are \(a, a, \dots , a\). If \(a = b\), we also have the pair \(\pm (0,1)\), giving rise to the same scrollar invariants.
-
if \(\gamma \ge 4\), then the first scrollar Betti numbers (with respect to any gonality pencil) take exactly two distinct values: \(2g/(\gamma - 1) - 2\) and \(g/(\gamma - 1) - 3\). Indeed, \(\Delta '\) satisfies condition \({\mathcal {P}}_1(v)\) (see Example 8), but also condition \({\mathcal {P}}_2(v)\): take \((j_1,j_2)=(j,\ell )\) if \(j\in \{-1,\ldots ,b+1\}\), \((j_1,j_2)=(j+1,\ell -1)\) if \(j=-1\) and \((j_1,j_2)=(j-1,\ell +1)\) if \(j=b+1\). By Theorem 16 we find that these numbers are \(2a, 2a, \dots , 2a, a-2,a-2, \dots , a-2\).
A first consequence is that U(f) admits a combinatorial gonality pencil. Indeed, \(\Delta \) cannot be of the form \(2\Upsilon \) (excluded in the statement of the theorem), nor of the form \(d\Sigma \) for some \(d \ge 2\): the cases \(d =2\) and \(d=3\) correspond to rational and elliptic curves (excluded at the beginning of this proof), the case \(d=4\) corresponds to curves of genus 3 (not composite), and the cases where \(d \ge 5\) correspond to curves of Clifford dimension 2.
Without loss of generality we may then assume that \(v=(1,0)\) and \(\Delta \subset \{ \, (i,j)\in {\mathbb {R}}^2\,|\, 0\le j\le \gamma \, \}\), so that our gonality pencil corresponds to the vertical projection. By [3, Theorem 9.1], the numbers \(E_\ell = -1+\sharp \{(i,j)\in \Delta ^{(1)}\cap {\mathbb {Z}}^2\,|\,j=\ell \}\) (for \(\ell = 1, \dots , \gamma - 1\)) are the corresponding scrollar invariants. Hence the \(E_\ell \)’s must all be equal to \(E:=g/(\gamma - 1) - 1\ge 1\).
This already puts severe restrictions on the possible shapes of \(\Delta ^{(1)}\). By horizontally shifting and skewing we may assume that the lattice points at height \(j=1\) are \((0,1), \dots , (E, 1)\) and that the lattice points at height \(j=2\) are \((0,2), \dots , (E,2)\). If \(\gamma = 3\), it follows that \(\Delta ^{(1)}\) has the desired rectangular shape, so we may suppose that \(\gamma \ge 4\). Then by horizontally flipping if needed, we can assume that the lattice points at height \(j=3\) are \((i,3), \dots , (E + i, 3)\) for some \(i \ge 0\). Now \(i \ge 2\) is impossible, for this would introduce a new lattice point at height \(j=2\); thus \(i = 0\) or \(i=1\). Continuing this type of reasoning, we obtain that the lattice points of \(\Delta ^{(1)}\) can be seen as a pile of n blocks of respectively \(m_1, \dots , m_n\) sheets, where each block is shifted to the right over a distance 1 when compared to its predecessor (Fig. 7).
We need to show that \(n = 1\), because then \(\Delta ^{(1)}\) has the desired rectangular shape. We will first prove that \(\Delta \) statisfies condition \({\mathcal {P}}_1(v)\). Since \(i^{(+)}(j)-i^{(-)}(j)=E\) for each value of j, the inequality
holds (so never \(\epsilon ^{(-)}_{j_1,j_2,r}=\epsilon ^{(+)}_{j_1,j_2,r}=1\)) for all \(j_1,j_2\in \{2,\ldots ,\gamma -2\}\) with \(j_2-j_1\ge 2\) and \(r\in \left\{ 1,\ldots ,\left\lfloor \frac{j_2-j_1}{2}\right\rfloor \right\} \). This implies that
Now assume that \({\mathcal {S}}^{(-)}_{j_1,j_2}\) and \({\mathcal {S}}^{(+)}_{j_1,j_2}\) are non-empty and disjoint. In this case, we can take \(r,s\in \left\{ 1,\ldots ,\left\lfloor \frac{j_2-j_1}{2}\right\rfloor \right\} \) such that
If \(r<s\), we get that
and
Subtracting E from both sides of the latter equation yields
so
which is in contradiction with Lemma 2. A similar contradiction can be obtained if \(r>s\).
Now let’s prove that \(\Delta \) also satisfies property \({\mathcal {P}}_2(v)\), where we assume that \(n\ge 2\). By Remark 13 and a symmetry consideration (rotation over \(180^\circ \)), it suffices to check the condition for lattice points (i, j) that lie on the left side of the boundary of \(\Delta \) (and even of \(\Delta ^{max}\)). Take \(\ell \in \{2,\ldots ,\gamma -2\}\), \(w=(i^{(--)}(\ell ),\ell )\) and \(u_{i,j}=(i_1,j_1),v_{i,j}=(i_2,j_2)\in \Delta ^{(1)}\) such that \((i,j)-w=(u_{i,j}-w)+(v_{i,j}-w)\), hence \(j_1+j_2=j+\ell \). It is sufficient to prove that
First assume that \(|j-\ell |>|j_2-j_1|\). If \(j\in \{1,\ldots ,\gamma -1\}\), then
where we use Lemma 2 for the last inequality. Since \(E\ge 1\), the desired inequality (10) follows. We still need to check (10) for points \((i,j)\in \partial \Delta \) with \(j=0\) and \(j=\gamma \), in particular \(i=-1\) resp. \(i=n-1\) because we can assume that (i, j) lies on the left side of the boundary of \(\Delta ^{max}\).
-
If \((i,j)=(-1,0)\), the line segment L between \((i+E,j)=(E-1,0)\) and \(w'=(i^{(++)}(\ell ),\ell )\in \Delta ^{(2)}\) intersects the horizontal lines of heights \(j_1\) and \(j_2\) in points that belong to \(\Delta ^{(1)}\). Using a similar argument as in the proof of Lemma 2, we obtain that \((i+E)+i^{(++)}(\ell )\le i^{(+)}(j_1)+i^{(+)}(j_2)+1\), which gives us (10) using \(E\ge 1\).
-
Analogously, we can handle the case \((i,j)=(n-1,\gamma )\): the line segment L between \((i+E,j)=(n+E-1,\gamma )\) and \(w'\) will intersect the horizontal lines of heights \(j_1\) and \(j_2\) in points that are contained in \(\Delta ^{(1)}\) and (10) follows.
If \(|j-\ell |=|j_2-j_1|\), we may assume that \(j_1=j\in \{1,\ldots ,\gamma -1\}\) and \(j_2=\ell \in \{2,\ldots ,\gamma -2\}\). But then the inequalities \(i\le i^{(-)}(j_1)\le i^{(+)}(j_1)\) and \(i^{(++)}(\ell )\le i^{(+)}(j_2)\) yield (10).
We still have to consider the case where \(|j-\ell |<|j_2-j_1|\), which implies that \(j\in \{1,\ldots ,\gamma -1\}\). By Lemma 2, we have that
hence
Since both conditions \({\mathcal {P}}_1(v)\) and \({\mathcal {P}}_2(v)\) hold for \(\Delta \), we can apply Theorem 16. If \(n \ge 2\), then there is at least one first scrollar Betti number having value \(E - 1 = g/(\gamma - 1) - 2\), for instance \(B_2\). This is distinct from both \(2g/(\gamma - 1) - 2\) and \(g/(\gamma - 1) - 3\): contradiction. Therefore \(n=1\), i.e. \(\Delta ^{(1)}\) has the requested rectangular shape. \(\square \)
4 Open questions
Here are two interesting open questions related to this paper:
-
1.
In Sect. 2, we gave a combinatorial interpretation for the first scrollar Betti numbers of \(\Delta \)-non-degenerate curves C in terms of the combinatorics of \(\Delta \), in case \(\Delta \) satisfies the condition \({\mathcal {P}}_1(v)\) (see Definition 4) and \({\mathcal {P}}_2(v)\) (see Definition 12). Can this be generalized to all polygons \(\Delta \)? There seems to be no geometric reason why this wouldn’t be the case, but we did not succeed to get rid of the conditions.
-
2.
In Theorem 18 of Sect. 3, we showed that non-degenerate curves on \({\mathbb {P}}^1\times {\mathbb {P}}^1\) have an intrinsic Newton polygon (at least, if \(g\ne 4\)). Can this be generalized to \(\Delta \)-non-degenerate curves on Hirzebruch surfaces \({\mathcal {H}}_n\)? In this case, we expect \(\Delta ^{(1)} = \emptyset \) or \(\Delta ^{(1)} \cong \text {conv}\{(0,0),(a+nb,0),(a,b),(0,b)\}\) for some integers \(a,b,n \ge 0\) .
References
Bopp, C., Hoff, M.: Resolutions of general canonical curves on rational normal scrolls. Arch. Math. 105, 239–249 (2015)
Casnati, G., Ekedahl, T.: Covers of algebraic varieties I. A general structure theorem for covers of degree 3, 4 and enriques surfaces. J. Algebraic Geom. 5, 439–460 (1996)
Castryck, Wouter., Cools, Filip.: Linear pencils encoded in the Newton polygon, to appear in International Mathematics Research Notices (arXiv:1402.4651)
Castryck, W., Cools, F.: A minimal set of generators for the canonical ideal of a non-degenerate curve. J. Aust. Math. Soc. 98, 311–323 (2015)
Castryck, W., Cools, F.: A combinatorial interpretation for Schreyer’s tetragonal invariants. Doc. Math. 20, 927–942 (2015)
Eisenbud, D., Harris, J.: On varieties of minimal degree (a centennial account). In: Proceedings in Symposia in Pure Mathematics, vol. 46, pp. 3–13 (1987)
Kawaguchi, R.: The gonality and the Clifford index of curves on a toric surface. J. Algebra 449, 660–686 (2016)
Khovanskii, A.G.: Newton polyhedra and toroidal varieties. Funct. Anal. Appl. 11, 289–296 (1977)
Schreyer, F.-O.: Syzygies of canonical curves and special linear series. Math. Ann. 275, 105–137 (1986)
Acknowledgements
We would like to thank Christian Bopp, Marc Coppens and Jeroen Demeyer for some interesting discussions. This research was supported by Research Project G093913N of the Research Foundation—Flanders (FWO), by the European Community’s Seventh Framework Programme (FP7/2007-2013) with ERC Grant Agreement 615722 MOTMELSUM, and by the Labex CEMPI (ANR-11-LABX-0007-01).
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Castryck, W., Cools, F. Intrinsicness of the Newton polygon for smooth curves on \({\mathbb {P}}^1\times {\mathbb {P}}^1\) . Rev Mat Complut 30, 233–258 (2017). https://doi.org/10.1007/s13163-017-0224-7
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s13163-017-0224-7