betaFeedback

WorksMathematics

A quotient formula for the Behrend function

Abstract

The Behrend function is essential in enumerative geometry. However, computing it, even in zero dimension, has been a highly subtle and mysterious matter since its introduction in the 2000s. We present a quotient formula for the Behrend function of any closed rr-dimensional affine scheme Xr⊂ANX^r \subset \mathbb{A}^N, evaluated at a singular point, where 0≤r≤N−10\leq r \leq N-1. In particular, if the scheme is monomial, we associate parallelepipeds to it and interpret the Behrend function as simple volume information about them. The results here, in particular, recover the recent results of Graffeo-Ricolfi for the zero-dimensional case in 2D in \cite{Graffeo-Ricolfi}, and also the more recent result of Polini-Sammartan-Ulrich on the zero-dimensional case in \cite{PSU26}.

Introduction

The Behrend function (introduced by Kai Behrend in [4]) plays a central role in enumerative geometry; in particular, the Behrend function of zero-dimensional subschemes is crucial in counting invariants in enumerative theories with symmetric perfect obstruction theories ([5, 6]). However, this function is infamous for having unpredictable behaviour; hence, it has been very hard to compute it, even in low dimensions. We provide a practical and elementary way to calculate the Behrend function of any affine scheme at a singular point; the case becomes more interesting when it comes to monomial (affine) schemes1, which come with a combinatorial intuition.

The case of zero-dimensional subschemes in A2\mathbb{A}^2 was first considered in [11] by Graffeo and Ricolfi using toric geometry, which covers many interesting non-trivial classes and confirms the difficulties of the Behrend function calculation. Also, more recently, the case of zero-dimensional subschemes in AN\mathbb{A}^N was considered in [21]. For more information on the applications of the Behrend function in enumerative geometry, we also refer to [11]. For the application of the Behrend function in Donaldson-Thomas theory, we refer to [17].

Before we state the main results, we give some quick key definitions (more details will be in Section 1.1).

Definition 0.1 (Normal cone and special fibre). Let II be an rr-dimensional2 ideal in A=C[x1,…,xN]A = \mathbb{C}[x_1,\ldots,x_N] and let X:=Spec⁡(A/I)X := \operatorname{Spec}(A/I) be singular at the origin o∈X\mathbf{o} \in X. We define the geometric normal cone of II to be N(I):=Spec⁡(gr⁡(I))=Spec⁡(⨁i=0∞IiIi+1)\mathcal{N}(I) := \operatorname{Spec}(\operatorname{gr}(I)) = \operatorname{Spec}\left(\bigoplus_{i=0}^{\infty}\frac{I^i}{I^{i+1}}\right). Note that Proj⁡(⨁i=0∞mimi+1)\operatorname{Proj}\left(\bigoplus_{i=0}^{\infty}\frac{\mathfrak{m}^i}{\mathfrak{m}^{i+1}}\right) defines the exceptional locus of Bl⁡o(X)\operatorname{Bl}_{\mathbf{o}}(X).

We, respectively, denote by Ni\mathcal{N}_i and Ni,j\mathcal{N}_{i,j}, the maximum-dimensional irreducible components of N(I)\mathcal{N}(I), for II, respectively, zero-dimensional or positive-dimensional. We denote by mult⁡(I)\operatorname{mult}(I) the normalised leading coefficient of the Hilbert polynomial of A/IA/I.

Also, we define the special fibre of II, S(I)S(I), to be the reduction3 of N(I)\mathcal{N}(I).4

We use the term degree for either ideal/quotient rings or their corresponding affine or (quasi-)projective schemes.5

Definition 0.2 (Behrend function, see definitions 1.2, 1.3, 1.4). Let X=Spec⁡(C[x1,…,xN]/I)X = \operatorname{Spec}(\mathbb{C}[x_1,\ldots,x_N]/I) be an affine scheme. We define the Behrend function on XX as νX:=Eu⁡(cX)\nu_X := \operatorname{Eu}(c_X), where

cX=cX/AN:=∑Ni,j⊂N(I)(−1)dim⁡π(Ni,j)l(Ni,j)[π(Ni,j)]c_X=c_{X/A^N}:=\sum_{\mathcal{N}_{i,j}\subset\mathcal{N}(I)}(-1)^{\dim\pi(\mathcal{N}_{i,j})}l(\mathcal{N}_{i,j})[\pi(\mathcal{N}_{i,j})]

where l(Ni,j)l(\mathcal{N}_{i,j}) is the length of N(I)\mathcal{N}(I) at the generic point of Ni,j\mathcal{N}_{i,j}, and Eu⁡\operatorname{Eu} is a function from the integral algebraic cycles on XX to constructible Z\mathbb{Z}-valued functions on XX, and π:N(I)→X\pi:\mathcal{N}(I)\to X is the canonical projection map.

Main results

We prove the following main results.

Setup. Let N(I)\mathcal{N}(I) be the geometric normal cone of an rr-dimensional ideal II in the polynomial ring A=C[x1,…,xN]A=\mathbb{C}[x_1,\ldots,x_N], as above. As in the proof of Theorem 2.1, we see that we can write the normal cone as Spec⁡(C[x1,…,xN,w1,…,wl]L)\operatorname{Spec}\left(\frac{\mathbb{C}[x_1,\ldots,x_N,w_1,\ldots,w_l]}{L}\right), for an ideal LL. We can write L=⋂i,jQi,jL=\bigcap_{i,j}Q_{i,j}, where {Qi,j}\{Q_{i,j}\} is its pi,j\mathfrak{p}_{i,j}-primary decomposition. Let mult⁡i,j\operatorname{mult}_{i,j} be the degree of Qi,jQ_{i,j}. Let Di,jD_{i,j} be the degree of pi,j\mathfrak{p}_{i,j}, for 0≤j≤r0\leq j\leq r, and 1≤i≤nj(I)1\leq i\leq n_j(I), where ni(I)n_i(I) is the number of the irreducible components of N(I)\mathcal{N}(I) of the maximal dimension, dominating an ii-dimensional component of Spec⁡(A/I)\operatorname{Spec}(A/I).

In addition, we set n(I)=∑i=0rni(I)n(I)=\sum_{i=0}^{r}n_i(I), which is the number of irreducible components of the maximal dimension of N(I)\mathcal{N}(I).

In this article, for simplicity, we assume that the origin o∈X\mathbf{o}\in X is a singular point of X=Spec⁡(A/I)X=\operatorname{Spec}(A/I), but the results can be considered for any singular point. We are interested in evaluating νX\nu_X at the origin o\mathbf{o}. When dim⁡X=0\dim X=0, we can simply use νX\nu_X for νX(o)\nu_X(\mathbf{o}).

Theorem A. (Quotient formula; see Theorem 2.1). For an rr-dimensional ideal II in A=C[x1,…,xN]A=\mathbb{C}[x_1,\ldots,x_N], suppose X=Spec⁡(A/I)X=\operatorname{Spec}(A/I) has a singularity at the origin o\mathbf{o}. Then we have

vX(o)=∑j=0r∑i=1nj(I)(−1)jmult⁡i,jDi,jEu⁡[π(Ni,j)],(1)v_X(\mathbf{o})=\sum_{j=0}^{r}\sum_{i=1}^{n_j(I)}(-1)^j\frac{\operatorname{mult}_{i,j}}{D_{i,j}}\operatorname{Eu}\left[\pi(\mathcal{N}_{i,j})\right], \tag*{(1)}

where Ni,j\mathcal{N}_{i,j} are the maximal-dimensional irreducible components of N(I)\mathcal{N}(I), and π:N(I)→X\pi:\mathcal{N}(I)\to X is the canonical projection, and Eu⁡\operatorname{Eu} is the local Euler obstruction.

An interesting combinatorial interpretation of this quotient formula for monomial ideals is described below. For a monomial ideal II in NN variables, we associate the convex hull, Conv⁡(I)\operatorname{Conv}(I), as the convex hull of the exponents of the monomial terms of II in RN\mathbb{R}^{N}.

Theorem B (A combinatorial interpretation for monomial ideals; see Theorem 5.2). Let 0≤j≤r0\leq j\leq r and 1≤i≤nj(I)1\leq i\leq n_j(I). Let II be an rr-dimensional monomial ideal in A=C[x1,…,xN]A=\mathbb{C}[x_1,\ldots,x_N], and let o\mathbf{o} be the origin as a singular point of X=Spec⁡(A/I)X=\operatorname{Spec}(A/I). Then

vX(o)=∑j=0r∑i=1nj(I)(−1)jgcd⁡k(vol⁡(parallelepipeds over Ti,jk))Eu⁡[π(Ni,j)],(2)v_X(\mathbf{o})=\sum_{j=0}^{r}\sum_{i=1}^{n_j(I)}(-1)^j\gcd_k\left(\operatorname{vol}(\text{parallelepipeds over }T_{i,j}^{k})\right)\operatorname{Eu}\left[\pi(\mathcal{N}_{i,j})\right], \tag*{(2)}

where {Ti,jk}k\{T_{i,j}^{k}\}_k is a triangulation of Pi,jP_{i,j}, where the latter is the projection of Conv⁡(I)\operatorname{Conv}(I) onto the (N−j)(N-j)-coordinate plane for every fixed jj. Also, Ni,j\mathcal{N}_{i,j} are the maximal-dimensional irreducible components of N(I)\mathcal{N}(I), and π:N(I)→X\pi:\mathcal{N}(I)\to X is the canonical projection, and Eu⁡\operatorname{Eu} is the local Euler obstruction. Also, Pi,jP_{i,j} corresponds to an irreducible component of the normal cone with maximal dimension, dominating a jj-coordinate plane component of Spec⁡(A/I)\operatorname{Spec}(A/I) in AN\mathbb{A}^{N}. The parallelepiped over Ti,jkT_{i,j}^{k} is a parallelepiped produced by the vectors from the origin on RN\mathbb{R}^{N} to the vertices of Ti,jkT_{i,j}^{k} (see Figure 1 and Figure 2).

A simplex and the parallelepiped generated by vectors from the origin to its vertices

Figure 1. Let the simplex Ti,jkT_{i,j}^{k} correspond to exponents associated to vertices AA, BB and CC. The parallelepiped over the simplex ABCABC is the parallelepiped generated over the vectors from the origin of RN\mathbb{R}^{N} to the vertices AA, BB and CC.

Irreducible components of the normal cone of maximal dimension

Figure 2. Irreducible components of the normal cone of maximal dimension are pictured with different colours. We have also picture the triangulation corresponding to each irreducible components. The numbers attached to each component are the corresponding gcd’s.

Remark 0.3. Note that since in the statement of Theorem B, we assumed that X=Spec⁡(A/I)X=\operatorname{Spec}(A/I) has the origin as its singularity, the components of XX correspond to ii-coordinate planes (i.e., ii-dimensional coordinate planes). Therefore, the last sentence of the theorem makes sense.

Remark 0.4. Note that in many interesting cases, for instance when π(Ni,j)\pi(\mathcal{N}_{i,j}) is smooth at the singularity point o∈X\mathbf{o}\in X, or it is a single point, by Lemma 1.5, we have both formulas in Theorem Theorem A and Theorem B without the contribution the local Euler obstruction. Also, when r=0r=0, we always have Eu⁡=1\operatorname{Eu}=1.

To clarify How Theorem A and Theorem B work, we give an explicit example.

Example 0.5. Let I=(x3,y4,z6,xy,xz,y2z2,yz4,y3z)I=(x^{3},y^{4},z^{6},xy,xz,y^{2}z^{2},yz^{4},y^{3}z) be a zero-dimensional ideal and X=Spec⁡(A/I)X=\operatorname{Spec}(A/I), which contains the origin as its singularity. We can see that there are three irreducible components of the normal cone N(I)\mathcal{N}(I) of the maximal dimension. From Theorem A, we have

νX=31+123+122=3+4+6=13.\nu_X=\frac{3}{1}+\frac{12}{3}+\frac{12}{2}=3+4+6=13.

Since II is a monomial ideal, alternatively from Theorem B, we have

νX=gcd⁡(3)+gcd⁡(4,4,4)+gcd⁡(6,6)=3+4+6=13.\nu_X=\gcd(3)+\gcd(4,4,4)+\gcd(6,6)=3+4+6=13.

See Figure 2.

An even simpler interpretation of the Behrend function for monomial schemes is the following.

Theorem C (see Theorem A.1). Let II be a monomial ideal in A=C[x1,x2,…,xN]A=\mathbb{C}[x_{1},x_{2},\ldots,x_{N}] that is singular at the origin6^{6}, and let X=Spec⁡(A/I)X=\operatorname{Spec}(A/I). Then, for any polygon Pi,jP_{i,j} in the lower boundary of the convex hull of II, we have

νX,i,j(o)=∣di,j∣,\nu_{X,i,j}(\mathbf{o})=\lvert d_{i,j}\rvert,

where νX,i,j(o)\nu_{X,i,j}(\mathbf{o}) is the contribution of the Behrend function evaluated at the origin (considering with a suitable sign), and dd is the constant term of the primitive equation $a_{1}x_{1}+a_{2}x_{2}+\cdots+

aNxN+di,j=0a_Nx_N+d_{i,j}=0 of the plane defined by Pi,jP_{i,j}. Then vX(o)v_X(o) can be computed as a signed sum of vX,i,j(o)v_{X,i,j}(o)s, weighted with the corresponding Eu⁡[π(Ni,j)]\operatorname{Eu}[\pi(N_{i,j})].

To prove Theorem B, we need the following theorem, which will be proved inductively.

Theorem D (Normal cone/convex hull correspondence; see Theorem 4.4). Let II be an rr-dimensional monomial ideal. The maximal dimensional irreducible components of the normal cone of II correspond to the polytope decomposition of the lower boundary of the convex hull of II in each (N−j)(N-j)-coordinate plane for 0≤j≤r0 \le j \le r.

From Theorem B and the minimisation procedure of a monomial ideal (we denote the minimisation of II by Imin⁡I_{\min}), we can characterise zero-dimensional monomial ideals with a maximal Behrend number, i.e., when the Behrend number is equal to the multiplicity:

Corollary 0.6 (Characterisation of zero-dimensional monomial ideals with maximal Behrend number; see Proposition 6.4). A zero-dimensional monomial ideal II in C[x1,…,xN]\mathbb{C}[x_1,\ldots,x_N] has maximal Behrend number if and only if the (N−1)(N-1)-boundary of the minimisation Imin⁡I_{\min} only consists of (N−1)(N-1)-simplices and the removed monomials in the process of minimisation lie strictly above the lower boundary of the convex hull of II.

Note that this corollary can be generalised to higher dimensions, but we skip the statement. Our procedure shares the spirit of the following combinatorial result, which will be proved in the appendix. More precisely, we have a generalised version of the celebrated Pick’s Theorem that relaxes the condition that the vertices be integral for certain types of plane polygons.

Theorem E (Generalised Pick’s formula; see Theorem A.7). Let PP be a plane convex polygon with some non-lattice vertices so that it can be integrally and generically embedded in R3\mathbb{R}^3, i.e., there is an embedding of PP into R3\mathbb{R}^3 with all vertices lattice points so that it is not perpendicular to any of the three coordinate planes. Let I~(P)\widetilde{I}(P) and B~(P)\widetilde{B}(P) be the number of internal lattice points and the number of lattice points on the boundary of the embedded polygon in R3\mathbb{R}^3, respectively. Then

area⁡(P)=(B~(P)2+I~(P)−1)(a2+b2+c2),\operatorname{area}(P)=\left(\frac{\widetilde{B}(P)}{2}+\widetilde{I}(P)-1\right)\left(\sqrt{a^2+b^2+c^2}\right),

where ax+by+cz+d=0ax+by+cz+d=0 is the equation of the plane defined by the embedded polygon in R3\mathbb{R}^3, (a,b,c)(a,b,c) is primitive, and d≠0d\ne0.

Remark 0.7. A generalised version for d=0d=0 is given in [23].

Richard Thomas raised the following question, but we do not know the answer, and it could be an interesting future direction.

Question. What is the enumerative interpretation of the Behrend function of ideals that do not define a critical locus of a function?

Previous work. In [11], the authors used toric geometry to compute the Behrend number of fat points in A2\mathbb{A}^2. Our approach recovers their formulas and extends them to higher dimensions using combinatorial methods. Regarding the constancy of the Behrend function on the Hilbert scheme of points in 3D, we refer to [15]. Also, there is a more recent result, [21], on the Behrend number of fat points in AN\mathbb{A}^N using Rees valuations.

Remark 0.8. This work was developed completely independently of and concurrently with [21], which appeared on arXiv in August 2026. While both studies independently derive the formulas for the zero-dimensional case, the present work specifically highlights the arbitrary-dimensional case. Also, while both studies reach similar conclusions for the zero-dimensional case, the different languages and tools used confirm the independence of the results. After conversations with one of the authors, we noticed the following overlaps in the zero-dimensional case: Restricting the formulas for the general case and the monomial case (in Theorem A and Theorem B, respectively) to dimension zero are equivalent to the formulas in [21]. Moreover, the formula for the monomial case in terms of the constant term of the defining hyperplanes in Theorem C, restricted to dimension zero, is equivalent to the results in [21].

Convention. By a polytope, we mean a convex bounded polytope. Furthermore, we always consider ideals with a minimal number of generators. By a lattice point, we mean a point with integer coordinates. Also, a lattice polytope is a polytope such that all its vertices are lattice points. For simplicity, in the polynomial ring with two variables we use x,yx,y instead of x1,x2x_1,x_2, and in the polynomial ring with three variables we use x,y,zx,y,z instead of x1,x2,x3x_1,x_2,x_3. A generator of a monomial ideal in C[x1,…,xN]\mathbb{C}[x_1,\ldots,x_N] is a mixed term or a mixed generator, if it has more than one variable xix_i. For example x2yx^2y is a mixed generator of the ideal (x3,z3,x2y)(x^3,z^3,x^2y) in C[x,y,z]\mathbb{C}[x,y,z]. We use the convex hull of an ideal and the convex hull of the normal cone of the ideal interchangeably. We denote the origin of CN\mathbb{C}^N by o\mathbf{o}. The proofs are considered for o\mathbf{o}, but it can be replaced by any singular point of XX. When XX is zero-dimensional, we simply denote vX(o)v_X(\mathbf{o}) by vXv_X. As mentioned above, the term degree will be used for either ideal/quotient rings or their corresponding affine or (quasi-)projective schemes.

Notation. For an rr-dimensional ideal II in A=C[x1,…,xN]A=\mathbb{C}[x_1,\ldots,x_N] and X=Spec⁡(A/I)X=\operatorname{Spec}(A/I) with the origin o\mathbf{o} as its singular point, we have the following notations for some numbers/notions associated with II (or XX). For the definition of degree, see Definition 1.1.

N(I)\mathcal{N}(I) The normal cone of II (see Definition 0.1).

S(I)\mathcal{S}(I) The special fibre of II as the reduction of N(I)\mathcal{N}(I) (see Definition 0.1).

n(I)=∑j=0rnj(I)n(I)=\sum_{j=0}^{r}n_j(I) Number of the maximal dimensional irreducible components of N(I)\mathcal{N}(I) or S(I)\mathcal{S}(I) (see definition 0.1), where nj(I)n_j(I) is the number of the irreducible components of N(I)\mathcal{N}(I) of the maximum dimension, dominating a jj-dimensional component of Spec⁡(A/I)\operatorname{Spec}(A/I).

Ni\mathcal{N}_i or Ni,j\mathcal{N}_{i,j} Components of N(I)\mathcal{N}(I), for II zero-dimensional or higher dimensional, respectively (see Definition 0.1).

mult⁡(I)\operatorname{mult}(I) The normalised leading coefficient of the Hilbert polynomial of A/IA/I.

mult⁡i,j\operatorname{mult}_{i,j} The degree of the pi,jp_{i,j}-primary decomposition component Qi,jQ_{i,j} of the ideal of the normal cone N(I)\mathcal{N}(I). When II is zero-dimensional, we simply denote this by mult⁡i\operatorname{mult}_i.

Di,jD_{i,j} The degree of pi,j=Qi,jp_{i,j}=\sqrt{Q_{i,j}}, where Qi,jQ_{i,j} are primary decomposition components of the ideal of the normal cone N(I)\mathcal{N}(I) for 0≤j≤r0\leq j\leq r, and 1≤i≤nj(I)1\leq i\leq n_j(I). When II is zero-dimensional, we simply denote this by DiD_i.

l(Ni,j)l(\mathcal{N}_{i,j}) The length of N(I)\mathcal{N}(I) at the generic point of Ni,j\mathcal{N}_{i,j} (see Definition 1.3).

cXc_X Integral cycle on XX (see Definition 1.3).

Eu⁡\operatorname{Eu} Local Euler obstruction (see Definition 1.4).

vX(0)v_X(\mathbf{0}) Behrend function of XX at the origin, when XX is positive dimensional. (see Definition 1.2 and Definition 1.4).

vXv_X Behrend function (Behrend number) of XX, when XX is zero-dimensional. (see Definition 1.2 and Definition 1.4).

Acknowledgements. The author thanks Michele Graffeo, Joachim Jelisiejew, and Andrea Ricolfi, for several discussions at the beginning of the project in 2021/2022. The author also thanks Hal Schenck for several helpful discussions in 2024. The author is grateful to Mark Gross for his support over the years. The author benefited from discussions with David Eisenbud, Barbara Fantechi, Rahul Pandharipande, Francisco Santos, Johannes Schmitt, Mike Stillman, Richard Thomas, Ajith Urundolil-Kumaran, Ravi Vakil, Andrzej Weber and Volkmar Welker. We thank Yunfeng Jiang, Andrea Ricolfi, Alessio Sammartano and Richard Thomas for comments on the final draft. This material is partially based upon work supported by the NSF under Grant No. DMS-1928930, while the author was in residence at SLMath (formerly known as MSRI) in Berkeley during the Fall 2022. The author was supported by the ERC Advanced Grant MSAG, the UKRI grant No. EP/X032779/1, and was partly in residence at ETH Zürich. The author thanks the Isaac Newton Institute for providing the space to discuss with the members of the program “New Equivariant Methods in Algebraic and Differential Geometry”. The rough ideas of the general subject were initiated when the author was partially supported by EPSRC grants EP/T015896/1 and EP/W522648/1. The author also acknowledges using GeoGebra [13] and Macaulay2 [12]. The codes in Appendix § 9 are duo to Mike Stillman, while visiting Cambridge in 2023. While searching for relevant recent results, we noticed a recent preprint, [21], on the zero-dimensional case; we acknowledge that our work is independent of it.

Definitions and Commutative algebra background

In this section, we give some basic definitions related to the original definition of the Behrend function. We also review some definitions and facts in commutative algebra that we need in the following sections. We first define the Behrend function as in [4 Definition 3.6, and Introduction].

Definition 1.1 (Degree of an ideal or a quotient ring). Throughout the paper, by degree of an ideal or a quotient ring, we mean the factorial of the (Krull) dimension of the ideal or the quotient ring, times the leading coefficient of the corresponding Hilbert polynomial.

We also use the term ‘degree’ for the corresponding affine or (quasi-)projective scheme.

Behrend function

We have the following original key definition.fire

Definition 1.2 (Behrend function, [4]). Let XX be a scheme over complex numbers. The Behrend function on XX is defined to be a constructible function

νX:X→Z,\nu_X:X\to\mathbb{Z},

so that if the scheme is proper and embeddable, then for any symmetric obstruction theory on it with associated virtual count deg⁡([X]vir):=∫[X]vir1\deg([X]^{\mathrm{vir}}):=\int_{[X]^{\mathrm{vir}}}1, we have

∫[X]vir1=χ(X,νX),(3)\int_{[X]^{\mathrm{vir}}}1=\chi(X,\nu_X), \tag*{(3)}

where [X]vir[X]^{\mathrm{vir}} is the virtual fundamental class, and χ(X,νX)\chi(X,\nu_X) is the weighted Euler characteristic.

To give a more precise definition of the Behrend function, we need to define the zero locus cycle associated with XX. We use the version in [4 Section 1.1].

Definition 1.3 (Distinguished cycle, [4]). For an ideal II in A=C[x1,…,xN]A = \mathbb{C}[x_1,\ldots,x_N], let N(I)\mathcal{N}(I) be the normal cone of II, with irreducible components of maximal dimension Ni,j\mathcal{N}_{i,j}, and let X=Spec⁡(A/I)X = \operatorname{Spec}(A/I). Let ℓ(Ni,j)\ell(\mathcal{N}_{i,j}) be the length of the normal cone at the generic point of Ni,j\mathcal{N}_{i,j}. Then, there is a canonical projection π:N(I)→X\pi:\mathcal{N}(I)\to X. Now define

cX=cX/AN:=∑Ni,j⊂N(I)(−1)dim⁡π(Ni,j)ℓ(Ni,j)[π(Ni,j)].(4)c_X=c_{X/A^N}:=\sum_{\mathcal{N}_{i,j}\subset\mathcal{N}(I)}(-1)^{\dim\pi(\mathcal{N}_{i,j})}\ell(\mathcal{N}_{i,j})[\pi(\mathcal{N}_{i,j})]. \tag*{(4)}

to be the unique integral cycle on XX (or the distinguished cycle on XX).

We also need the notion of local Euler obstruction as originally defined in [19 Section 3]. We use the version defined in [4 Section 1.2].

Definition 1.4 (Local Euler obstruction, [4, 16]). Let XX be as in Definition 1.3. We define

Eu⁡:Z∗(X)→Cons⁡(X)\operatorname{Eu}:Z_*(X)\to\operatorname{Cons}(X)
{integral algebraic cycles on X}↦{constructible Z-valued functions on X}.\{\text{integral algebraic cycles on }X\}\mapsto\{\text{constructible }\mathbb{Z}\text{-valued functions on }X\}.

More precisely, let X↪ANX\hookrightarrow A^N be an embedding and dim⁡X=r\dim X=r. Let (x1,…,xN)(x_1,\ldots,x_N) be local coordinate at p∈Xp\in X. Also, define X∘X^\circ to be the smooth locus in XX and s:X∘→Gr⁡r(TAN)s:X^\circ\to\operatorname{Gr}_r(TA^N) be a section defined by s(x)=Tx(X∘)s(x)=T_x(X^\circ), where Gr⁡r(TAN)\operatorname{Gr}_r(TA^N) is the Grassmannian bundle of rr dimensional subspaces of TANTA^N over ANA^N. Then, we can define X^\widehat{X} to be the closure of the image of ss. Let α:X^→X\alpha:\widehat{X}\to X, defined by restriction. Also, define the function ∥X∥\lVert X\rVert to be x1x‾1+⋯+xNx‾N\sqrt{x_1\overline{x}_1+\cdots+x_N\overline{x}_N}; so, the differential d∥X∥d\lVert X\rVert is a section of T∗ANT^*A^N. Let tt be the pullback of d∥X∥d\lVert X\rVert under the embedding. Define Bε:={x:∥x∥≤ε}B_\varepsilon:=\{x:\lVert x\rVert\le\varepsilon\} and Sε:={x:∥x∥=ε}S_\varepsilon:=\{x:\lVert x\rVert=\varepsilon\}. We can then define

Eu⁡(X)(p):=⟨Eu⁡(T∗X,t),O⟩,\operatorname{Eu}(X)(p):=\langle\operatorname{Eu}(T^*X,t),\mathcal{O}\rangle,

where O∈H2r(α−1Bε,α−1Sε;Z)\mathcal{O}\in H_{2r}(\alpha^{-1}B_\varepsilon,\alpha^{-1}S_\varepsilon;\mathbb{Z}) and Eu⁡(T∗X,t)\operatorname{Eu}(T^*X,t) denotes the obstruction to extending tt as a non-zero section of T∗XT^*X from α−1Sε\alpha^{-1}S_\varepsilon to α−1Bε\alpha^{-1}B_\varepsilon.

Now, the Behrend function of XX as above (Definition 1.3) can be defined as (See [4 Definition 1.4])

vX:=Eu⁡(cX).(5)v_X:=\operatorname{Eu}(c_X). \tag*{(5)}

Applying Eu⁡\operatorname{Eu} to 1.2, we get

vX=∑Ni,j⊂N(I)(−1)dim⁡π(Ni,j)ℓ(Ni,j)Eu⁡([π(Ni,j)]).(6)v_X=\sum_{\mathcal{N}_{i,j}\subset\mathcal{N}(I)}(-1)^{\dim\pi(\mathcal{N}_{i,j})}\ell(\mathcal{N}_{i,j})\operatorname{Eu}([\pi(\mathcal{N}_{i,j})]). \tag*{(6)}

We have the following properties for.

Lemma 1.5. [16, Definition 2.4],[19, Section 3], [22] We have

  • Eu⁡(X)(p)=1\operatorname{Eu}(X)(p)=1 if ZZ is nonsingular at pp.

  • Eu⁡(X)(p)=0\operatorname{Eu}(X)(p)=0 if p∉Zp\notin Z.

  • Eu⁡(X1×X2)(p1×p2)=Eu⁡(X1)(p1)⋅Eu⁡(X2)(p2)\operatorname{Eu}(X_1\times X_2)(p_1\times p_2)=\operatorname{Eu}(X_1)(p_1)\cdot\operatorname{Eu}(X_2)(p_2).

  • If XX or Deligne-Mumford stack which is reducible at the point pp, and let XiX_i are the irreducible components, then Eu⁡(X)(p)=∑Eu⁡(Xi)(p)\operatorname{Eu}(X)(p)=\sum\operatorname{Eu}(X_i)(p).

We give another definition for future reference.

Definition 1.6 (Dominating the support). We say that a maximum-dimensional irreducible component Ni,j\mathcal{N}_{i,j} of N(I)\mathcal{N}(I) dominates the support π(Ni,j)\pi(\mathcal{N}_{i,j}) in XX.

Remark 1.7. Note that on a smooth point of XX, the value of νX\nu_X equals (−1)dim⁡X(-1)^{\dim X}. Thus, the value of the Behrend function is interesting only on the singular points of XX, which is the subject of this article.

If XX is a critical locus, its Behrend function has a particular enumerative geometric interpretation as in (3), and, in theory, can be computed as follows.

Lemma 1.8. [4 Section 1.2] If XX is a critical locus, i.e., if XX is the zero locus of dfdf, for some regular function ff on AN\mathbb{A}^{N}, then for any C\mathbb{C}-valued point p∈Xp \in X, we have

νX(p)=(−1)N(1−χ(Fp)),(7)\nu_X(p)=(-1)^N(1-\chi(F_p)), \tag*{(7)}

where FpF_p the Milnor fibre of ff at pp.

Proof. See [20 Corollary 2.4 (iii)].

We devote the rest of the section to the commutative algebraic definitions.

Associated primes and primary decomposition

Definition 1.9. [1 Chapter 4] Let RR be a ring. An ideal qq in RR is primary if it is not equal to the whole ring and if xy∈qxy \in q, then either x∈qx \in q or ym∈qy^m \in q, for some m∈Z≥1m \in\mathbb{Z}_{\geq1}. If p=q\mathfrak{p}=\sqrt{q}, then qq is called p\mathfrak{p}-primary.

Definition 1.10. [9 Section 3.1] Let RR be a ring and MM be an RR-module. A prime ideal p\mathfrak{p} in RR is called associated to MM if it is the annihilator of some x∈Mx \in M.

The set of all prime ideals associated with MM is denoted by Ass⁡R(M)\operatorname{Ass}_R(M) or Ass⁡(M)\operatorname{Ass}(M).

Lemma 1.11. [9 Theorem 3.1] Let RR be a Noetherian ring and M≠0M \ne0 be a finitely generated RR-module. Then, Ass⁡(M)\operatorname{Ass}(M) is a finite non-empty set of primes, each containing the annihilator of MM. The set Ass⁡(M)\operatorname{Ass}(M) includes all primes minimal among primes containing ann⁡(M)\operatorname{ann}(M).

If RR is a commutative ring, the minimal elements in Ass⁡R(M)\operatorname{Ass}_R(M) are called isolated primes, and the non-minimal ones are called embedded primes.

Noether normalization

The following lemma, which is called Noether normalization lemma, is a version of it that we need later:

Lemma 1.12. [24 Lemma 10.115.4] For a field kk, and an mm-dimensional ideal II, let S=k[x1,…,xN]IS = \frac{k[x_1,\ldots,x_N]}{I}. Then, there exist a polynomial ring k[y1,…,ym]k[y_1,\ldots,y_m] and a finite injective homomorphism k[y1,…,ym]↪Sk[y_1,\ldots,y_m] \hookrightarrow S.

Normalization and convex hull

Definition 1.13. Let II be a monomial ideal in C[x1,…,xN]\mathbb{C}[x_1,\ldots,x_N]. An element b∈C[x1,…,xN]b \in\mathbb{C}[x_1,\ldots,x_N] is integral if there are ci∈Iic_i \in I^i such that

bl+c1bl−1+…+cl=0.b^l+c_1b^{l-1}+\ldots+c_l=0.

The integral closure of a monomial ideal II is denoted by I‾\overline{I} and consists of all the integral elements as above. A monomial ideal II is integrally closed if I‾=I\overline{I}=I.

Definition 1.14. We define the geometric normal cone of an ideal II in C[x1,…,xN]\mathbb{C}[x_1,\ldots,x_N] to be

Spec⁡(gr⁡(I))=Spec⁡(⨁i=0∞IiIi+1).\operatorname{Spec}(\operatorname{gr}(I))=\operatorname{Spec}\left(\bigoplus_{i=0}^{\infty}\frac{I^i}{I^{i+1}}\right).

Note that we may use the word normal cone for either Spec⁡(⨁i=0∞IiIi+1)\operatorname{Spec}\left(\bigoplus_{i=0}^{\infty}\frac{I^i}{I^{i+1}}\right) or ⨁i=0∞IiIi+1\bigoplus_{i=0}^{\infty}\frac{I^i}{I^{i+1}}, interchangeably.

Note that minimal primes of J(N(I))\mathcal{J}(\mathcal{N}(I)), the defining ideal of N(I)\mathcal{N}(I), are equivalent to the maximal dimensional components of the normal cone N(I)\mathcal{N}(I).

Lemma 1.15. [9 Exercise 4.23], [14, Proposition 1.4.6] Let II be a monomial ideal in C[x1,…,xN]\mathbb{C}[x_1,\ldots,x_N], and Γ\Gamma be the set of exponents of monomials in II. Let Λ\Lambda be the convex hull of Γ\Gamma, and Λ∗\Lambda^* be the set of integral points in Λ\Lambda. Then, the integral closure of II is the ideal generated by Λ∗\Lambda^*.

Corollary 1.16. Let II be a monomial integrally closed ideal in C[x1,…,xN]\mathbb{C}[x_1,\ldots,x_N]. Then, the restriction of II to any non-empty subset of variables {x1,…,xN}\{x_1,\ldots,x_N\} is still integrally closed.

Proof. By Lemma 1.15, there is a one-to-one correspondence between the normal cone of an ideal and the convex hull of the exponents. As the restriction of the convex hull to any non-empty subset of variables {x1,…,xN}\{x_1,\ldots,x_N\} is still a convex hull, the same lemma implies the result.

The quotient formula: proof of the abstract version

In this section, we give a simple proof of Theorem A (see Theorem 2.1 below) using algebraic arguments. Then, in the following sections, we will see a geometric interpretation of this abstract version of the quotient formula.

Theorem 2.1. Let II be an ideal in A=C[x1,…,xN]A=\mathbb{C}[x_1,\ldots,x_N] with ll minimum number of generators, and let X=Spec⁡(A/I)X=\operatorname{Spec}(A/I). Then, for any rr-dimensional ideal II in C[x1,…,xN]\mathbb{C}[x_1,\ldots,x_N], and oo the origin as a singular point of X=Spec⁡(A/I)X=\operatorname{Spec}(A/I), we have

νX(o)=∑j=0r∑i=1nj(I)(−1)jmult⁡i,jDi,jEu⁡([π(Ni,j)]).\nu_X(o)=\sum_{j=0}^{r}\sum_{i=1}^{n_j(I)}(-1)^j\frac{\operatorname{mult}_{i,j}}{D_{i,j}}\operatorname{Eu}\left([\pi(\mathcal{N}_{i,j})]\right).

where mult⁡i,j\operatorname{mult}_{i,j} is the degree of the pi,j\mathfrak{p}_{i,j}-primary decomposition component of maximal dimension, Qi,jQ_{i,j}, of N(I)\mathcal{N}(I), and Di,jD_{i,j} is the degree of pi,j\mathfrak{p}_{i,j}, for 0≤j≤r0\leq j\leq r, and 1≤i≤nj(I)1\leq i\leq n_j(I). Also, Ni,j\mathcal{N}_{i,j} are the maximal-dimensional irreducible components of N(I)\mathcal{N}(I), and π:N(I)→X\pi:\mathcal{N}(I)\to X is the canonical projection, and Eu⁡\operatorname{Eu} is the local Euler obstruction. Also, n(I)=∑j=0rnj(I)n(I)=\sum_{j=0}^{r}n_j(I) is the number of the irreducible components of the maximal dimension of N(I)\mathcal{N}(I), and nj(I)n_j(I) is the number of the irreducible components of the maximal dimension of N(I)\mathcal{N}(I), dominating a jj-dimensional component of Spec⁡(A/I)\operatorname{Spec}(A/I) in AN\mathbb{A}^N.

Proof. Let R=⨁i=0∞IiIi+1R=\bigoplus_{i=0}^{\infty}\frac{I^i}{I^{i+1}}, and N(I)=Spec⁡(R)\mathcal{N}(I)=\operatorname{Spec}(R). We can write RR as C[x1,…,xN,w1,…,wl]L\frac{\mathbb{C}[x_1,\ldots,x_N,w_1,\ldots,w_l]}{L}, where L=I+JL=I+J, and JJ is an ideal in C[x1,…,xN,w1,…,wl]\mathbb{C}[x_1,\ldots,x_N,w_1,\ldots,w_l], and ll is the (minimum) number of generators of II. We write L=⋂i,jQi,jL=\bigcap_{i,j}Q_{i,j}, where {Qi,j}\{Q_{i,j}\} is its pi,j\mathfrak{p}_{i,j}-primary decomposition.

Let {Ni,j}0≤j≤r1≤i≤nj(I)\{\mathcal{N}_{i,j}\}_{\substack{0\leq j\leq r\\1\leq i\leq n_j(I)}} be the set of maximum-dimensional irreducible components of N(I)\mathcal{N}(I), corresponding to Qi,jQ_{i,j}. Think of S(I)\mathcal{S}(I) as the reduction of N(I)\mathcal{N}(I), and let {SNi,j}0≤j≤r1≤i≤nj(I)\{\mathcal{S}_{\mathcal{N}_{i,j}}\}_{\substack{0\leq j\leq r\\1\leq i\leq n_j(I)}} be the set of corresponding irreducible components of S(I)\mathcal{S}(I). Since Qi,j=pi,j\sqrt{Q_{i,j}}=\mathfrak{p}_{i,j}, the components SNi,j\mathcal{S}_{\mathcal{N}_{i,j}} correspond to pi,j\mathfrak{p}_{i,j}. Since dim⁡(L)=N\dim(L)=N, from Noether normalization lemma (Lemma 1.12), there is a finite injective map

ι:C[y1,…,yN]↪R.(8)\iota:\mathbb{C}[y_1,\ldots,y_N]\hookrightarrow R. \tag*{(8)}

For F=C[y1,…,yN]F=\mathbb{C}[y_1,\ldots,y_N], ι\iota induces a free injection

F−1ι:C(y1,…,yN)↪F−1R,(9)F^{-1}\iota:\mathbb{C}(y_1,\ldots,y_N)\hookrightarrow F^{-1}R, \tag*{(9)}

where C(y1,…,yN)\mathbb{C}(y_1,\ldots,y_N) is the function field of C[y1,…,yN]\mathbb{C}[y_1,\ldots,y_N]. Taking Spec of the finite map ι\iota reverses it to

ϕ:N(I)⟶AN,\phi:\mathcal{N}(I)\longrightarrow\mathbb{A}^N,

and taking Spec of F−1ιF^{-1}\iota gives

Spec⁡(F−1R)⟶Spec⁡(C(y1,…,yN)).\operatorname{Spec}(F^{-1}R)\longrightarrow\operatorname{Spec}(\mathbb{C}(y_1,\ldots,y_N)).

Hence, the generic fibre of ϕ\phi is given by Spec⁡(F−1R)\operatorname{Spec}(F^{-1}R), which has irreducible components corresponding to maximum-dimensional (=N=dim⁡(R)=N=\dim(R)) irreducible components of N(I)\mathcal{N}(I), i.e., Ni,j\mathcal{N}_{i,j}, for 0≤j≤r0\leq j\leq r and 1≤i≤nj(I)1\leq i\leq n_j(I). Therefore, the rank of the free map F−1ιF^{-1}\iota is mult⁡(I)=∑0≤j≤r1≤i≤nj(I)mult⁡i,j\operatorname{mult}(I)=\sum_{\substack{0\leq j\leq r\\1\leq i\leq n_j(I)}}\operatorname{mult}_{i,j}.

Now, let Ri,j=Rpi,j=ONi,j,N(I)R_{i,j}=R_{p_{i,j}}=\mathcal{O}_{\mathcal{N}_{i,j},\mathcal{N}(I)} be the local ring of RR at the defining ideal of Ni,j\mathcal{N}_{i,j}; hence, F−1RF^{-1}R can be written as ∏0≤j≤r1≤i≤nj(I)Ri,j\prod_{\substack{0\leq j\leq r\\1\leq i\leq n_j(I)}}R_{i,j}. Therefore, by taking the degree of the map F−1ιF^{-1}\iota, we can write the multiplicity, mult⁡(I)\operatorname{mult}(I), as ∑0≤j≤r1≤i≤nj(I)dim⁡C(y1,…,yN)Ri,j\sum_{\substack{0\leq j\leq r\\1\leq i\leq n_j(I)}}\dim_{\mathbb{C}(y_1,\ldots,y_N)}R_{i,j}.

On the other hand, S(I)\mathcal{S}(I) can be written as ∏0≤j≤r1≤i≤nj(I)K(Ri,j)\prod_{\substack{0\leq j\leq r\\1\leq i\leq n_j(I)}}K(R_{i,j}), where K(Ri,j)K(R_{i,j}) is the residue field of Ri,jR_{i,j} and is a finite extension over C(y1,…,yN)\mathbb{C}(y_1,\ldots,y_N).

Since C(y1,…,yN)\mathbb{C}(y_1,\ldots,y_N) is not necessarily algebraically closed, it is not necessarily equal to its finite extension. The difference measures the difference between the multiplicity and the Behrend function as below.

Similarly as the case of multiplicity, the degree of S(I)\mathcal{S}(I) can be written as

∑0≤j≤r1≤i≤nj(I)dim⁡C(y1,…,yN)K(Ri,j).\sum_{\substack{0\leq j\leq r\\1\leq i\leq n_j(I)}}\dim_{\mathbb{C}(y_1,\ldots,y_N)}K(R_{i,j}).

Note that by taking the residue field of ι\iota, we obtain

deg⁡(pi,j)=dim⁡C(y1,…,yN)K(Ri,j).(10)\deg(p_{i,j})=\dim_{\mathbb{C}(y_1,\ldots,y_N)}K(R_{i,j}). \tag*{(10)}

Note that we can write Qi,jRi,j=pi,jni,jRi,jQ_{i,j}R_{i,j}=p_{i,j}^{n_{i,j}}R_{i,j}, for some integers ni,jn_{i,j}. From this, we have deg⁡(Qi,j)=deg⁡(pi,j)length⁡(Ri,jQi,jRi,j)\deg(Q_{i,j})=\deg(p_{i,j})\operatorname{length}\left(\frac{R_{i,j}}{Q_{i,j}R_{i,j}}\right). Thus, deg⁡(pi,j)∣deg⁡(Qi,j)\deg(p_{i,j})\mid\deg(Q_{i,j}).

Recall that, by our definition, mult⁡i,j\operatorname{mult}_{i,j} is the degree of Ni,j\mathcal{N}_{i,j}, and Di,jD_{i,j} is the degree of SNi,j\mathcal{S}_{\mathcal{N}_{i,j}}; i.e., mult⁡i,j\operatorname{mult}_{i,j} is the contribution of mult⁡(I)\operatorname{mult}(I) to Ni,j\mathcal{N}_{i,j}, and Di,jD_{i,j} is the contribution of the degree of S(I)\mathcal{S}(I) to SNi,j\mathcal{S}_{\mathcal{N}_{i,j}}. Now, from (10), for each fixed i,ji,j, for the component SNi,j⊂Ni,j\mathcal{S}_{\mathcal{N}_{i,j}}\subset\mathcal{N}_{i,j}, we have

mult⁡i,jDi,j=mult⁡∣Ni,jD∣SNi,j=deg⁡Ni,jdeg⁡SNi,j=deg⁡(Qi,j)deg⁡(pi,j)=dim⁡C(y1,…,yN)Ri,jdim⁡C(y1,…,yN)K(Ri,j)=dim⁡K(Ri,j)Ri,j=deg⁡ONi,j,N(I)=l(Ni,j),(11)\begin{aligned} \frac{\operatorname{mult}_{i,j}}{D_{i,j}} = \frac{\operatorname{mult}|_{\mathcal{N}_{i,j}}}{D|_{\mathcal{S}_{\mathcal{N}_{i,j}}}} = \frac{\deg\mathcal{N}_{i,j}}{\deg\mathcal{S}_{\mathcal{N}_{i,j}}} = \frac{\deg(Q_{i,j})}{\deg(p_{i,j})} \\ &= \frac{\dim_{\mathbb{C}(y_1,\ldots,y_N)}R_{i,j}}{\dim_{\mathbb{C}(y_1,\ldots,y_N)}K(R_{i,j})} = \dim_{K(R_{i,j})}R_{i,j} = \deg\mathcal{O}_{\mathcal{N}_{i,j},\mathcal{N}(I)} = l(\mathcal{N}_{i,j}), \tag*{(11)} \end{aligned}

where l(Ni,j)l(\mathcal{N}_{i,j}) is the length of N(I)\mathcal{N}(I) at (the generic point of) Ni,j\mathcal{N}_{i,j}.

Restricting to r=0r=0, we have n0(I)=n(I)n_0(I)=n(I). Also, in this case, we have the notation: mult⁡i:=mult⁡i,0\operatorname{mult}_i:=\operatorname{mult}_{i,0}, Di:=Di,0D_i:=D_{i,0}, Ri:=Ri,0R_i:=R_{i,0}, Qi:=Qi,0Q_i:=Q_{i,0}, pi:=pi,0p_i:=p_{i,0}, Ni:=Ni,0\mathcal{N}_i:=\mathcal{N}_{i,0}, and SNi:=SNi,0\mathcal{S}\mathcal{N}_i:=\mathcal{S}\mathcal{N}_{i,0}. Then, (11) becomes

dim⁡K(Ri)Ri=deg⁡ONi,N(I)=ℓ(Ni)=vX,i(12)\dim_{K(R_i)}R_i=\deg\mathcal{O}_{\mathcal{N}_i,\mathcal{N}(I)}=\ell(\mathcal{N}_i)=v_{X,i} \tag*{(12)}

where vX,iv_{X,i} is the contribution of vXv_X to each irreducible component of the normal cone, and the last equality comes from the definition of the Behrend number for the zero-dimensional case (see in (4) and (5)). Hence, for zero-dimensional II (for which, the contribution of Eu⁡\operatorname{Eu} is 1 by Lemma 1.5), from (12), we have

∑i=1n(I)mult⁡iDi=∑i=1n(I)vi(I)=vX.(13)\sum_{i=1}^{n(I)}\frac{\operatorname{mult}_i}{D_i}=\sum_{i=1}^{n(I)}v_i(I)=v_X. \tag*{(13)}

Now, let r>0r>0, and X=Spec⁡(A/I)X=\operatorname{Spec}(A/I) with a singular point at the origin o∈X\mathbf{o}\in X. Following Definition 1.3, let us consider the canonical projection π:N(I)→X\pi:\mathcal{N}(I)\to X. By definition 1.6, any maximum-dimensional irreducible component of the normal cone, Ni,j⊂N(I)\mathcal{N}_{i,j}\subset\mathcal{N}(I), dominates its corresponding ii-dimensional support π(Ni,j)⊂X\pi(\mathcal{N}_{i,j})\subset X, for each 0≤i≤r0\leq i\leq r and fixed 1≤j≤ni(I)1\leq j\leq n_i(I). Note that since we assume that the origin is a singular point for XX, the components of XX are coordinate planes of possibly different dimensions (as in Remark 0.3). Hence, the general alternating sum in the formula comes from the formula for the cycle on XX introduced in Definition 1.3 (also, see [22 Section 2.2]), and also (11) and (13) above. This means that we have

∑j=0r∑i=1nj(I)(−1)jmult⁡i,jDi,jEu⁡(π(Ni,j))=∑j=0r∑i=1nj(I)(−1)jℓ(Ni,j)Eu⁡(π(Ni,j))=vX(o),\sum_{j=0}^{r}\sum_{i=1}^{n_j(I)}(-1)^j\frac{\operatorname{mult}_{i,j}}{D_{i,j}}\operatorname{Eu}\left(\pi(\mathcal{N}_{i,j})\right)=\sum_{j=0}^{r}\sum_{i=1}^{n_j(I)}(-1)^j\ell(\mathcal{N}_{i,j})\operatorname{Eu}\left(\pi(\mathcal{N}_{i,j})\right)=v_X(\mathbf{o}),

as claimed.

Corollary 2.2 (Upper bound on the Behrend number). For any zero-dimensional ideal II in A=C[x1,…,xN]A=\mathbb{C}[x_1,\ldots,x_N] and X=Spec⁡(A/I)X=\operatorname{Spec}(A/I), we have

vX≤mult⁡(I).v_X\leq\operatorname{mult}(I).

Proof. Note that mult⁡(I)=∑i=1n(I)mult⁡i(I)\operatorname{mult}(I)=\sum_{i=1}^{n(I)}\operatorname{mult}_i(I). So by Theorem 2.1(iii), we obtain the result.

Minimisation

In this section, we introduce the notion of minimisation and apply it to give a more practical interpretation of the quotient formula for monomial ideals. This will also characterise monomial ideals for which the Behrend number is maximal (see Proposition 6.4).

Convention. In this section, we assume that the ideals have a minimum number of generators.

Definition 3.1 (Minimal ideal). We say that a monomial ideal II with a minimum number of generators in C[x1,x2,…,xN]\mathbb{C}[x_1,x_2,\ldots,x_N] is minimal if the removal of any mixed generator from it changes the multiplicity, that is, if I=(x1m1,x2m2,…,xNmN,F1,F2,…,Fk)I=(x_1^{m_1},x_2^{m_2},\ldots,x_N^{m_N},F_1,F_2,\ldots,F_k), where mim_i could possibly be zero (if II is of a dimension greater than zero), and FiF_i are mixed monomial terms. Then,

mult⁡(x1m1,x2m2,…,xNmN,F1,F2,…,Fk)≠mult⁡(x1m1,x2m2,…,xNmN,F1,…,Fi^,…,Fk),\operatorname{mult}(x_1^{m_1},x_2^{m_2},\ldots,x_N^{m_N},F_1,F_2,\ldots,F_k)\ne\operatorname{mult}(x_1^{m_1},x_2^{m_2},\ldots,x_N^{m_N},F_1,\ldots,\widehat{F_i},\ldots,F_k),

for any omitted mixed term F^i\widehat{F}_i. The minimisation of II, which is indicated by Imin⁡I_{\min}, is a minimal ideal with a minimal number of generators such that the set of its generators is contained in the set of the (minimal) generators of II such that mult⁡(I)=mult⁡(Imin⁡)\operatorname{mult}(I)=\operatorname{mult}(I_{\min}).

Equivalently, an ideal II is minimal, if and only if I=Imin⁡I=I_{\min}.

The following lemma is a consequence of the definition:

Lemma 3.2. Let I=(Imin⁡,H1,H2,…,Hl)I=(I_{\min},H_1,H_2,\ldots,H_l). Then

mult⁡(I)=mult⁡(Imin⁡,H1,H2,…,Hl−1)=mult⁡(Imin⁡,H1,H2,…,Hl−2)=…=mult⁡(Imin⁡).\operatorname{mult}(I)=\operatorname{mult}(I_{\min},H_1,H_2,\ldots,H_{l-1})=\operatorname{mult}(I_{\min},H_1,H_2,\ldots,H_{l-2})=\ldots=\operatorname{mult}(I_{\min}).

Now, we list some properties of minimisation in the following proposition, which are direct consequences of the combinatorial interpretation of minimisation in the following sections:

Proposition 3.3. Let II be a monomial ideal in C[x1,x2,…,xN]\mathbb{C}[x_1,x_2,\ldots,x_N].

  • Minimisation preserves normalization: normalization of I=I= normalization of Imin⁡I_{\min}.

  • Normalization preserves minimisation: Imin⁡=(normalization of I)min⁡I_{\min}=(\text{normalization of }I)_{\min}.

  • Minimisation preserves the number of maximal dimensional irreducible components of the normal cone of the ideal: n(I)=n(Imin⁡)n(I)=n(I_{\min}).

An algorithm for minimisation

Let I=(x1m1,x2m2,…,xNmN,N1,N2,…,Ns,F1,F2,…,Fh)I=(x_1^{m_1},x_2^{m_2},\ldots,x_N^{m_N},N_1,N_2,\ldots,N_s,F_1,F_2,\ldots,F_h) be an ideal minimally generated by monomials in C[x1,x2,…,xN]\mathbb{C}[x_1,x_2,\ldots,x_N], such that Fi,NiF_i,N_i are mixed terms and mim_i possibly being zero for some (or all) ii. Also, suppose that NiN_i is contained in the (minimal set of generators of the) normalization of II for 1≤i≤s1\leq i\leq s, and FjF_j is not in the (minimal set of generators of the) normalization of II for 1≤j≤h1\leq j\leq h. Then, we have the following algorithm to find Imin⁡I_{\min}:

  1. Remove all FjF_j's for 1≤j≤h1\leq j\leq h, so that we are left with (x1m1,x2m2,…,xNmN,N1,N2,…,Ns)(x_1^{m_1},x_2^{m_2},\ldots,x_N^{m_N},N_1,N_2,\ldots,N_s),

  2. Omit each NiN_i one by one from II and check if NiN_i appears in the (minimal set of generators of the) normalization of (x1m1,x2m2,…,xNmN,N1,N2,…,Ni^,…,Ns)(x_1^{m_1},x_2^{m_2},\ldots,x_N^{m_N},N_1,N_2,\ldots,\widehat{N_i},\ldots,N_s). (a) if NjN_j appears in the (minimal set of generators of the) normalization of

(x1m1,x2m2,…,xNmN,N1,N2,…,Nj^,…,Ns),(x_1^{m_1},x_2^{m_2},\ldots,x_N^{m_N},N_1,N_2,\ldots,\widehat{N_j},\ldots,N_s),

then mark NjN_j to remove it after completing step (2) for each NiN_i. (b) if NjN_j does not appear in the (minimal set of generators of the) normalization of (x1m1,x2m2,…,xNmN,N1,N2,…,Nj^,…,Ns)(x_1^{m_1},x_2^{m_2},\ldots,x_N^{m_N},N_1,N_2,\ldots,\widehat{N_j},\ldots,N_s), then mark NjN_j to keep it after completing step (2) for each NiN_i. We rename such NjN_j to Nj′N'_j.

  1. The ideal generated by x1m1,x2m2,…,xNmNx_1^{m_1},x_2^{m_2},\ldots,x_N^{m_N} and the monomials in (b) is the (minimal set of generators of the) minimisation of II; i.e., Imin⁡=(x1m1,x2m2,…,xNmN,N1′,N2′,…,Nt′)I_{\min}=(x_1^{m_1},x_2^{m_2},\ldots,x_N^{m_N},N'_1,N'_2,\ldots,N'_t), where t≤st\leq s.

Remark 3.4. Note that in step (2), ordering does not matter. We remove one NiN_i at a time, check the process for it, and then return it and repeat the process for another NiN_i.

Let us look at an example and apply the algorithm above:

Example 3.5. Let

I=(x200,y200,x2y20,x6y19,x11y17,x13y12,x15y11,x16y7,x21y5,x22y4,x24y3,x35y2)I=(x^{200},y^{200},x^2y^{20},x^6y^{19},x^{11}y^{17},x^{13}y^{12},x^{15}y^{11},x^{16}y^7,x^{21}y^5,x^{22}y^4,x^{24}y^3,x^{35}y^2)

be an ideal in C[x,y]\mathbb{C}[x,y]. One can easily check that

(F1,F2,…,F5)=(x6y19,x11y17,x13y12,x15y11,x21y5),(F_1,F_2,\ldots,F_5)=(x^6y^{19},x^{11}y^{17},x^{13}y^{12},x^{15}y^{11},x^{21}y^5),

and

(N1,N2,…,N5)=(x2y20,x16y7,x22y4,x24y3,x35y2),(N_1,N_2,\ldots,N_5)=(x^2y^{20},x^{16}y^7,x^{22}y^4,x^{24}y^3,x^{35}y^2),
(N1′,N2′,…,N4′)=(x2y20,x16y7,x24y3,x35y2).(N'_1,N'_2,\ldots,N'_4)=(x^{2}y^{20},x^{16}y^7,x^{24}y^3,x^{35}y^2).

Therefore, we have Imin⁡=(x200,y200,x2y20,x16y7,x24y3,x35y2)I_{\min}=(x^{200},y^{200},x^2y^{20},x^{16}y^7,x^{24}y^3,x^{35}y^2).

Combinatorial interpretation: background

This section provides a combinatorial background for giving a geometric interpretation of the quotient formula for monomial ideals. The first step is to correspond the monomial terms of a monomial ideal in C[x1,…,xN]\mathbb{C}[x_1,\ldots,x_N] to points in ZN\mathbb{Z}^N; i.e., we have

monomial term x1m1x2m2…xNmN⟷ point (m1,m2,…,mN) in ZN=lattice point,\text{monomial term }x_1^{m_1}x_2^{m_2}\ldots x_N^{m_N}\longleftrightarrow\text{ point }(m_1,m_2,\ldots,m_N)\text{ in }\mathbb{Z}^N=\text{lattice point},

where some mim_i can possibly be zero.

A combinatorial interpretation of the minimisation: determination of vertices in (N−1)(N-1)-boundary

We make a combinatorial sense of the abstract notion of minimisation introduced previously.

Definition 4.1 ((N−1N-1)-boundary). For a minimal ideal with a minimal number of generators

(x1m1,…,xNmN,x1m11x2m21…xNmN1,…,x1m1kx2m2k…xNmNk)(x_1^{m_1},\ldots,x_N^{m_N},x_1^{m_{11}}x_2^{m_{21}}\ldots x_N^{m_{N1}},\ldots,x_1^{m_{1k}}x_2^{m_{2k}}\ldots x_N^{m_{Nk}})

in C[x1,…,xN]\mathbb{C}[x_1,\ldots,x_N], where some mim_i can possibly be zero, the (N−1)(N-1)-boundary is the lower boundary of the unique convex hull generated by mim_i or mijm_{ij}, for 1≤i≤N1\le i\le N and 1≤j≤k1\le j\le k.

The existence and uniqueness of such a decomposition passing through any given finite set of points (after omitting unsuitable points) in RN\mathbb{R}^N is a well-known combinatorial fact; for example, see [8 Corollary 1.7]. The following proposition summarises this for zero-dimensional monomial ideals:

Proposition 4.2. Let II be a zero-dimensional monomial ideal in C[x1,…,xN]\mathbb{C}[x_1,\ldots,x_N], and let P1=(p11,p12,…,p1N),P2=(p21,p22,…,p2N),…,Pk=(pk1,pk2,…,pkN)P_1=(p_{11},p_{12},\ldots,p_{1N}),P_2=(p_{21},p_{22},\ldots,p_{2N}),\ldots,P_k=(p_{k1},p_{k2},\ldots,p_{kN}) be the corresponding points (vectors) in RN\mathbb{R}^N associated with the exponents of the generators of II. Then, for some j≤Nj\le N, there exists a convex (j−1)(j-1)-boundary in (R≥0)j(\mathbb{R}_{\ge0})^j with the maximum number of (j−1)(j-1)-polytopes (as lower facets of the corresponding convex hull) so that:

  • the vertices of the (j−1)(j-1)-polytopes cover all PiP_i’s

  • the vectors corresponding to the vertices of each (j−1)(j-1)-simplex generate an jj dimensional parallelepiped.

The following corollary shows that minimisation provides a way to present exactly the vertices of the (N−1)(N-1)-boundary:

Corollary 4.3. Minimisation removes the non-vertices monomials in the corresponding (N−1)(N-1)-boundary; i.e., minimal ideals only consist of vertices.

Proof. Without loss of generality, we consider the zero-dimensional monomial ideals in 2 variables. The proof for the general case is similar.

Let I=(A,B,C,D,E)⊂C[x,y]I=(A,B,C,D,E)\subset\mathbb{C}[x,y], be a monomial ideal, where B=xm1,A=ym2B=x^{m_1},A=y^{m_2}, as in Figure 3. First, we note that if a monomial generator is not among the minimal generators of the normalization (which is the case for the mixed monomial term EE in Figure 3), then it will not appear on the boundary of the convex hull of the ideal; therefore, if the generators omitted in step (1) of the minimisation are not among vertices. For step (2), if we temporarily ignore a generator gg that is among the minimal generators of normalization but is not in the normalization of the rest of the generators, then that generator must be a vertex (this is the case for vertex CC in Figure 3(ii)).

Figure 3. The minimisation of I=(A,B,C,D,E)I=(A,B,C,D,E) is Imin⁡=(A,B,C)I_{\min}=(A,B,C).

Finally, if the monomial is not a vertex but is on the lower boundary of the convex hull, then by definition it will be omitted in the minimisation process (This is the case for the mixed monomial generator DD in Figure 3(iii)).

As another corollary, we prove Theorem D:

Theorem 4.4. Let II be an rr-dimensional monomial ideal with singularity at the origin. The maximal dimensional irreducible components of the normal cone of II correspond to the polytope decomposition of the lower boundary of the convex hull of II in each (N−j)(N-j)-coordinate plane for 0≤j≤r0 \le j \le r.

Proof. From [10 Section 3.3], we know that the rays of the associated fan to the normal cone correspond to the torus invariant exceptional divisors. Then, from [26 Corollary 2.53], these rays correspond to the maximal dimensional components of the normal cone. So, the way to construct a fan corresponding to the maximum-dimensional irreducible components of the normal cone is as follows. First, we start with the components dominating the origin, i.e., the ones embedded in the whole NN-space and part of the lower boundary of the convex hull there, and consider their corresponding (N−1)(N-1)-hyperplanes containing those components. Next, for the components dominating the axes we need to project the lower boundary of the convex hull onto the (N−1)(N-1)-coordinate planes, look at the restricted convex hull there and consider the (N−1)(N-1)-hyperplanes containing the components corresponding to the lower boundary of the convex hull there there. We continue this process and produce (N−1)(N-1)-hyperplanes containing the components corresponding to the lower boundary of the convex hull in coordinate planes of lower dimensions until we reach to the (N−r)(N-r)-coordinate plane(s). Note that if r=N−1r=N-1, then the corresponding (N−1)(N-1)-hyperplanes for the last step are parallel to some coordinate planes. See Figure 4 for N=2N=2 and r=N−1=1r=N-1=1, in which the hyperplane (line) associated to

Figure 4. (i) The maximum-dimensional components of the normal cone associated to I=(x5,x2y,xy2)I=(x^5,x^2y,xy^2). (ii) The fan generated by the orthogonal rays to the maximum-dimensional components. See the proof of Theorem 4.4

::: the projection onto the xx-axis is the line x=1x=1, which is parallel to the yy-axis (as a coordinate plane).

After constructing all the hyperplanes, the orthogonal vectors to them (which determine a fan) determine the divisors representing the maximal dimension irreducible components of the normal cone. This also aligns with [11 Theorem 5.13] for N=2N=2 and r=0r=0.

Now, we present a 3D examples for a better visualization.

Example 4.5. Let I=(xy,xz,yz)I=(xy,xz,yz) be an ideal in C[x,y,z]\mathbb{C}[x,y,z]. In Figure 5, apart from the hyperplane containing the triangle of the generators, there are three non-coordinate hyperplanes. Hence, the fan of the normal cone in R3\mathbb{R}^3 has four rays corresponding to four maximal dimension irreducible components of the normal cone. Also, see Example 6.7.

The lower boundary of the convex hull of the normal cone of $I=(xy,xz,yz)$

Figure 5. The lower boundary of the convex hull of the normal cone of I=(xy,xz,yz)I=(xy,xz,yz) has four component corresponding to the illustrated four non-coordinate planes. See Example 4.5

A combinatorial version of the quotient formula

Volume interpretation

In this sub-section, we interpret the ingredients of the quotient formula in terms of the information about the corresponding parallelepipeds. Note that using this interpretation, we can directly compute the Behrend number without using the quotient formula. First, we have the following straightforward lemma.

Lemma 5.1. Let RR be a Noetherian ring and let {p1,…,pl}\{p_1,\ldots,p_l\} be minimal primes in Ass⁡(R)\operatorname{Ass}(R). Let QiQ_i be a pip_i-primary ideal corresponding to pip_i with maximal dimension. Then, we have

length⁡(RQi)=length⁡(F−1RQi),\operatorname{length}\left(\frac{R}{Q_i}\right)=\operatorname{length}\left(\frac{F^{-1}R}{Q_i}\right),

where F=k[y1,…,ym]∖0F=k[y_1,\ldots,y_m]\setminus0, where k[y1,…,ym]k[y_1,\ldots,y_m] is the polynomial ring in the Noether normalization lemma (Lemma 1.12).

Theorem 5.2. Let II be a monomial ideal of dimension rr in A=C[x1,…,xN]A=\mathbb{C}[x_1,\ldots,x_N], and let pp be the origin as a singular point of X=Spec⁡(A/I)X=\operatorname{Spec}(A/I). Then

vX(o)=∑j=1ni(I)∑i=0r(−1)igcd⁡(vol⁡(parallelepipeds over Ti,jk)k)Eu⁡([π(Ni,j)])v_X(o)=\sum_{j=1}^{n_i(I)}\sum_{i=0}^{r}(-1)^i\operatorname{gcd}\left(\operatorname{vol}\left(\text{parallelepipeds over }T_{i,j}^{k}\right)_k\right)\operatorname{Eu}\left([\pi(\mathcal{N}_{i,j})]\right)

where {Ti,jk}k\{T_{i,j}^{k}\}_k is a triangulation of Pi,jP_{i,j}, where the latter is the projection of Conv⁡(I)\operatorname{Conv}(I) onto the (N−j)(N-j)-coordinate plane for every fixed jj. Also, Ni,j\mathcal{N}_{i,j} are the maximal-dimensional irreducible components of N(I)\mathcal{N}(I), and π:N(I)→X\pi:\mathcal{N}(I)\to X is the canonical projection, and Eu⁡\operatorname{Eu} is the local Euler obstruction. Also, Pi,jP_{i,j} corresponds to an irreducible component of the normal cone with maximal dimension, dominating a jj-coordinate plane component of Spec⁡(A/I)\operatorname{Spec}(A/I) in AN\mathbb{A}^{N}.

In particular, if II is zero-dimensional, we have

vX=∑j=1n(I)vj(I)=∑j=1n(I)gcd⁡(vol⁡(parallelepipeds over Tjk)k)Eu⁡([π(Nj)])v_X=\sum_{j=1}^{n(I)}v_j(I)=\sum_{j=1}^{n(I)}\operatorname{gcd}\left(\operatorname{vol}\left(\text{parallelepipeds over }T_j^k\right)_k\right)\operatorname{Eu}\left([\pi(\mathcal{N}_j)]\right)
mult⁡(I)=∑j=1n(I)mult⁡j(I)=∑j=1n(I)vol⁡(parallelepiped over Pj),\operatorname{mult}(I)=\sum_{j=1}^{n(I)}\operatorname{mult}_j(I)=\sum_{j=1}^{n(I)}\operatorname{vol}(\text{parallelepiped over }P_j),
Dj=(number of (N−1)-simplices with normalised volume equal to the gcd that can be accommodated in Pj),D_j=(\text{number of }(N-1)\text{-simplices with normalised volume equal to the gcd that can be accommodated in }P_j),

where PjP_j is a (N−1)(N-1)-polytope corresponding to each irreducible component, and Nj\mathcal{N}_j are the maximal-dimensional irreducible components of N(I)\mathcal{N}(I).

Proof. From the proof of Theorem 2.1 and Theorem 4.4, the irreducible components of maximal dimension of the normal cone of II are given by the maximum-dimensional components of the component Ni,j\mathcal{N}_{i,j} corresponding to the pip_i-primary decomposition of the defining ideal of N(I)\mathcal{N}(I). To compute this using Macaulay2, see Appendix § 9. These components dominate the ii-coordinate plane(s), which are projected onto the (N−i)(N-i)-coordinate planes for 0≤i≤r0 \le i \le r.

Let r=0r=0. We follow the notation in the proof of Theorem 2.1. For volume interpretation, we correspond each monomial term x1a1x2a2…xNaNx_1^{a_1}x_2^{a_2}\ldots x_N^{a_N} in II to a point (a1,a2,…,aN)(a_1,a_2,\ldots,a_N) in RN\mathbb{R}^N (which can be thought of as a point in ZN\mathbb{Z}^N). Note that for a monomial ideal II, the number n(I)n(I) is the number of components (polytopes) in the lower boundary of the convex hull Conv⁡(I)\operatorname{Conv}(I). The defining hyperplanes of the components of the lower boundary of the convex hull can be seen as orthogonal hyperplanes to the rays of associated toric fan, and each ray corresponds to an exceptional divisor of Bl⁡o(X)\operatorname{Bl}_{o}(X), where oo is the origin as a singularity point of XX. So, by Theorem 4.4, there is a one-to-one correspondence between the irreducible component of the normal cone corresponding to RjR_j and the lower boundary of the irreducible components of the convex hull of II.

Following the notations in the proof of Theorem 2.1 and Lemma 5.1, we have

dim⁡K(Rj)Rj=length⁡(Rj)=length⁡(F−1RQj)=length⁡(RQj).\dim_{K(R_j)} R_j=\operatorname{length}(R_j)=\operatorname{length}\left(\frac{F^{-1}R}{Q_j}\right)=\operatorname{length}\left(\frac{R}{Q_j}\right).

where QjQ_j is a pjp_j-primary, so that {pj}\{p_j\} are minimal primes in Ass⁡(F−1R)\operatorname{Ass}(F^{-1}R).

Let σ\sigma be an embedding from RR into some Rm\mathbb{R}^m, where m≥Nm\ge N. Note that Rm\mathbb{R}^m contains the lattice Zm\mathbb{Z}^m, which in turn contains the sub-lattice ZN\mathbb{Z}^N, since m≥Nm\ge N. Now, since the maximal dimension components of the normal cone correspond to QjQ_j, then σ\sigma maps QjQ_j to a an ideal lattice σ(Qj)=Λj\sigma(Q_j)=\Lambda_j.7 Hence, σ\sigma is an additive isomorphism on QjQ_j. Since RR is NN-dimensional, Λj\Lambda_j is a sub-lattice of ZN\mathbb{Z}^N (which, in turn, is a sub-lattice of Zm\mathbb{Z}^m) generated by the points in ZN\mathbb{Z}^N corresponding to the lower boundary component of the convex hull of II associated to QjQ_j.

Therefore, using Theorem 4.4, under σ\sigma, the restriction of vXv_X to the jj-th component of the normal cone, vX,jv_{X,j}, is

vX,j=length⁡(RQj)=length⁡(the maximum component of the normal cone corresponding to Qj)=[ZN:Λj].\begin{aligned} v_{X,j}&=\operatorname{length}\left(\frac{R}{Q_j}\right)\\ &=\operatorname{length}(\text{the maximum component of the normal cone corresponding to }Q_j)\\ &=[\mathbb{Z}^N:\Lambda_j]. \end{aligned}

as QjQ_j is pjp_j-primary, for pjp_j minimal (isolated) prime.

The key point here is that [ZN:Λj][\mathbb{Z}^N:\Lambda_j] is the gcd of all possible (normalised) volumes of NN-simplices associated to RjR_j, considering the associated points plus the origin (by [7 Lemma 2.4]; also it can be obtained from the proof of [25 Theorem 2.2]). Hence, they are volumes of corresponding parallelepipeds.

For mult⁡(I)\operatorname{mult}(I), first note that if Λ\Lambda is a sub-lattice of ZN\mathbb{Z}^N, and Γ\Gamma is a sub-lattice of Λ\Lambda, then [ZN:Λ]∣[ZN:Γ][\mathbb{Z}^N:\Lambda]\mid[\mathbb{Z}^N:\Gamma] (by a straightforward argument in [7 Lemma 2.3(1)]). Now, let Λi\Lambda_i be as above and ⨁j=1riΛji\bigoplus_{j=1}^{r_i}\Lambda_j^i be a sub-lattice of Λi\Lambda_i generated by vertices of any fixed triangulation of PiP_i’s into (N−1)(N-1)-simplices. By a similar argument as above, and from the proof of Theorem 2.1, we have

mult⁡i(I)=dim⁡C(y1,…,yN)Ri=[ZN:⨁j=1riΛj′]=∑j=1ri[ZN:Λj′]=∑j=1rivol⁡(the j-th parallelepiped generated by each (N−1)-subsimplex in any fixed triangulation of Pi)\begin{aligned} \operatorname{mult}_{i}(I)=\dim_{\mathbb{C}(y_{1},\ldots,y_{N})}R_{i}=\left[\mathbb{Z}^{N}:\bigoplus_{j=1}^{r_{i}}\Lambda'_{j}\right]=\sum_{j=1}^{r_{i}}\left[\mathbb{Z}^{N}:\Lambda'_{j}\right] \\ &=\sum_{j=1}^{r_{i}}\operatorname{vol}(\text{the }j\text{-th parallelepiped generated by each }(N-1)\text{-subsimplex in any fixed triangulation of }P_{i}) \end{aligned}

As mult⁡(I)=∑i=1n(I)mult⁡i(I)\operatorname{mult}(I)=\sum_{i=1}^{n(I)}\operatorname{mult}_{i}(I), we have the result.

Finally, for the degree of the special fiber, we note that

mult⁡i(I)=[ZN:⨁j=1riΛj′]=[ZN:Λi][Λi:⨁j=1riΛj′]=vi(I)[Λi:⨁j=1riΛj′].\operatorname{mult}_{i}(I)=\left[\mathbb{Z}^{N}:\bigoplus_{j=1}^{r_{i}}\Lambda'_{j}\right]=\left[\mathbb{Z}^{N}:\Lambda_{i}\right]\left[\Lambda_{i}:\bigoplus_{j=1}^{r_{i}}\Lambda'_{j}\right]=v_{i}(I)\left[\Lambda_{i}:\bigoplus_{j=1}^{r_{i}}\Lambda'_{j}\right].

Therefore

∑j=1rivol⁡(the j-th parallelepiped generated by each (N−1)-subsimplex in any triangulation of Pi)=(gcd⁡(vol⁡(parallelepipeds generated by each N possible points in Pi)))([Λi:⨁j=1riΛj′]),\begin{aligned} \sum_{j=1}^{r_{i}}\operatorname{vol}(\text{the }j\text{-th parallelepiped generated by each }(N-1)\text{-subsimplex in any triangulation of }P_{i}) \\ &=(\gcd(\operatorname{vol}(\text{parallelepipeds generated by each }N\text{ possible points in }P_{i})))([\Lambda_{i}:\bigoplus_{j=1}^{r_{i}}\Lambda'_{j}]), \end{aligned}

hence

Di=[Λi:⨁j=1riΛj′]=(number of (N−1)-subsimplices with normalised volume equal to the gcd that can be accommodated in Pi),D_{i}=[\Lambda_{i}:\bigoplus_{j=1}^{r_{i}}\Lambda'_{j}]=(\text{number of }(N-1)\text{-subsimplices with normalised volume equal to the gcd that can be accommodated in }P_{i}),

For r>0r>0, to evaluate the Behrend function at the origin, we need to consider the alternating sum based on the dimension of the maximal dimensional irreducible component of the normal cone to obtain the formula. Using Corollary 1.16 and also from the proof of Theorem 2.1, and the projection π:N(I)→X\pi:\mathcal{N}(I)\to X, for each 0≤i≤r0\leq i\leq r and fixed jj, we know that Ni,j\mathcal{N}_{i,j} dominates the corresponding ii-dimensional support π(Ni,j)\pi(\mathcal{N}_{i,j}) in XX. For monomial II, this means that the support can be viewed as the projection of N(I)\mathcal{N}(I) onto the corresponding (N−i)(N-i)-dimensional complement coordinate plane. Hence, the length is basically the (gcd of the) volume(s) of the parallelepiped(s). This gives us the desired formula. □

Remark 5.3. One can use the existing algorithms for the convex hulls in higher dimensions (for example, see [2] and [3]).

Examples

In this section, we provide examples of different flavours and calculate the Behrend function using our main results. Note that in all the examples, we have we always have Eu⁡=1\operatorname{Eu}=1. So, the contribution of the local Euler obstruction will not appear in the formulas.

Zero-dimensional examples

Let us look at some examples and directly compute the Behrend number:

Example 6.1. Let I=(x10,y6,z8,x4y,y3z,x2z2,xz5)I=(x^{10},y^{6},z^{8},x^{4}y,y^{3}z,x^{2}z^{2},xz^{5}) be a zero-dimensional ideal in A=C[x,y,z]A=\mathbb{C}[x,y,z] and let X=Spec⁡(A/I)X=\operatorname{Spec}(A/I). From Theorem 5.2, we have vX=20+26+24+gcd⁡(24,24)=94v_{X}=20+26+24+\gcd(24,24)=94.

Example 6.2. Let

I=(x21,y8,z9,x3y3z,xy2z2,xy4z,xz3,xy6,x7y3,x11y,x9z,x5y4,x7yz,x9y2)I=(x^{21},y^{8},z^{9},x^{3}y^{3}z,xy^{2}z^{2},xy^{4}z,xz^{3},xy^{6},x^{7}y^{3},x^{11}y,x^{9}z,x^{5}y^{4},x^{7}yz,x^{9}y^{2})
Polytope diagram with vertices A, B, D, E, F, G, and H

Figure 6. The monomial term H=xz5H=xz^{5} is removed in the minimisation process, but it plays a role here as it is contained in the lower boundary of the convex hull, and it gives two polytopes, each with an associated parallelepiped of volume 24.

be a zero-dimensional ideal in A=C[x,y,z]A=\mathbb{C}[x,y,z] and let X=Spec⁡(A/I)X=\operatorname{Spec}(A/I). By Theorem 2.1, we have

vX=721+243+22117+211=72+8+13+21=114.v_X=\frac{72}{1}+\frac{24}{3}+\frac{221}{17}+\frac{21}{1}=72+8+13+21=114.

More directly, from Theorem 5.2, we have

vX=72+gcd⁡(8,8,8)+gcd⁡(13,13,…)+21=114.v_X=72+\gcd(8,8,8)+\gcd(13,13,\ldots)+21=114.

See Figure 7.

Polytope diagram with vertices A through N

Figure 7. The monomial terms A,B,G,K,L,F,JA,B,G,K,L,F,J are removed in the minimisation, but they play a role in the Behrend function calculation as they are on the lower boundary of the convex hull. See Example 6.2.

Example 6.3. Let

I=(y17,xy14,x2y12,x3y10,x5y6,x7y4,x9y3,x11y2,x14y,x16)I=\left(y^{17},xy^{14},x^{2}y^{12},x^{3}y^{10},x^{5}y^{6},x^{7}y^{4},x^{9}y^{3},x^{11}y^{2},x^{14}y,x^{16}\right)

be an ideal in AA and let X=Spec⁡(A/I)X = \operatorname{Spec}(A/I). Then, we have

Imin⁡=(y17,xy14,x5y6,x7y4,x11y2,x16).I_{\min}=(y^{17},xy^{14},x^{5}y^{6},x^{7}y^{4},x^{11}y^{2},x^{16}).

As mentioned above, we can compute the Behrend number as follows.

vX=∑j=1n(I)=5mult⁡jDj=171+644+221+302+321=102.v_X=\sum_{j=1}^{n(I)=5}\frac{\operatorname{mult}_j}{D_j}=\frac{17}{1}+\frac{64}{4}+\frac{22}{1}+\frac{30}{2}+\frac{32}{1}=102.

See Figure 8.

A colored L-shaped region with diagonal segments and grid squares

Figure 8. The colored L-shape area indicates mult⁡i\operatorname{mult}_i for one of the corresponding components. The length of each diagonal over the smallest possible length piece of the diagonal is DiD_i (hence in the pictured component, we have mult⁡i=64\operatorname{mult}_i=64 = the number of coloured squares, and Di=4D_i=4). See Example 6.3.

Maximal Behrend number

We characterise the zero-dimensional monomial ideals with the maximum possible Behrend number (i.e., zero-dimensional ideals II in AA with vX=mult⁡(I)v_X=\operatorname{mult}(I), where X=Spec⁡(A/I)X=\operatorname{Spec}(A/I)):

Proposition 6.4. A zero-dimensional monomial ideal II in C[x1,…,xN]\mathbb{C}[x_1,\ldots,x_N] has a maximal Behrend number if and only if the (N−1)(N-1)-boundary of Imin⁡I_{\min} consists only of (N−1)(N-1)-simplices and the possible removed monomials in the minimisation process lie strictly above the (N−1)(N-1)-boundary.

Proof. Assuming the conditions in the statement and from the proof of Theorem 5.2, we observe that there is only one option for the vol⁡\operatorname{vol} associated with each PiP_i, which is the volume of the parallelepiped over PiP_i itself. So, gcd⁡\gcd is trivial and maximal, which in turn gives the maximal Behrend number.

Example 6.5. Let

I=(x11,y13,z10,x2z8,x3z7,y5z5,xy6,x2yz4,x5y4z3)I=(x^{11},y^{13},z^{10},x^{2}z^{8},x^{3}z^{7},y^{5}z^{5},xy^{6},x^{2}yz^{4},x^{5}y^{4}z^{3})

be a zero-dimensional ideal in A=C[x,y,z]A=\mathbb{C}[x,y,z] and X=Spec⁡(A/I)X=\operatorname{Spec}(A/I). One can easily check that the associated 2-polytope (or polygon) decomposition only consists of triangles, and the monomials x2z8x^{2}z^{8}, y7z5y^{7}z^{5} and x5y4z4x^{5}y^{4}z^{4}, which are removed in the minimisation process, lie strictly above the lower boundary of the associated convex hull. Therefore, by Proposition 6.4, II has maximal Behrend number (νX=mult⁡(I)=264+28+30+110+130=562\nu_X=\operatorname{mult}(I)=264+28+30+110+130=562). See Figure 9.

Convex hull with red points corresponding to removed monomials

Figure 9. The red points corresponding to the removed monomials are strictly above the surface (lower boundary of the convex hull)

Higher dimensional examples

We see a couple of examples of the Behrend function for the higher-dimensional monomial affine subschemes of A2A^{2} and A3A^{3}, respectively.

Example 6.6. Let I=(xy3,x2y2)I=(xy^{3},x^{2}y^{2}) be a one-dimensional ideal in A=C[x,y]A=\mathbb{C}[x,y], and X=Spec⁡(A/I)X=\operatorname{Spec}(A/I). Then, by Theorem 5.2, the Behrend function at the origin o\mathbf{o} (as a singularity of XX) is

νX(o)=4−1−2=1.\nu_X(\mathbf{o})=4-1-2=1.

See Figure 10.

Figure 10. (1) The convex hull for I=(xy3,x2y2)I=(xy^{3},x^{2}y^{2}). (2) The volume of the two-dimensional parallelogram is 4. (3) The projection on the xx and yy axes gives the lengths 1 and 2, respectively.

Example 6.7. Let f(x,y,z)=xaybzcf(x,y,z)=x^{a}y^{b}z^{c}, and X=Crit⁡(f)X=\operatorname{Crit}(f). Then

νX(o)=−1.\nu_X(\mathbf{o})=-1.

This comes from Theorem 5.2:

vX(0)=∣a−1bcab−1cabc−1∣−(∣ab−1a−1b∣+∣bc−1b−1c∣+∣ac−1a−1c∣)+((a−1)+(b−1)+(c−1))=(a+b+c−1)−(2a+2b+2c−3)+((a−1)+(b−1)+(c−1))=−1.\begin{aligned} v_X(\mathbf{0})= \left|\begin{matrix}a-1&b&c\\a&b-1&c\\a&b&c-1\end{matrix}\right|-\left(\left|\begin{matrix}a&b-1\\a-1&b\end{matrix}\right|+\left|\begin{matrix}b&c-1\\b-1&c\end{matrix}\right|+\left|\begin{matrix}a&c-1\\a-1&c\end{matrix}\right|\right)+\left((a-1)+(b-1)+(c-1)\right) \\ &=(a+b+c-1)-(2a+2b+2c-3)+((a-1)+(b-1)+(c-1)) \\ &=-1. \end{aligned}

This can be generalised to any number of variables NN, with f(x1,…,xN)=x1a1⋯xNaNf(x_1,\ldots,x_N)=x_1^{a_1}\cdots x_N^{a_N}, for which we have vCrit⁡(f)(0)=(−1)Nv_{\operatorname{Crit}(f)}(0)=(-1)^N. A particular example of N=3N=3 and a=b=c=1a=b=c=1 is discussed in [22 Example 2.5]. This can also be checked by Lemma 1.8.

A non-monomial example

We provide an example of a non-monomial zero-dimensional ideal in two variables that also fits for using results in [11].

Example 6.8. Let

I=(x4y+x3y2,x5−x3y2,xy5+x3y2+x2y3,x2y4,x3y3,y7−x3y2−x2y3)I=(x^4y+ x^3y^2,x^5-x^3y^2,xy^5+x^3y^2+x^2y^3,x^2y^4,x^3y^3,y^7-x^3y^2-x^2y^3)

be a 0-dimensional ideal in A=C[x,y]A=\mathbb{C}[x,y] and X=Spec⁡(A/I)X=\operatorname{Spec}(A/I). Then, we can use the M2 codes in Appendix § 9 to calculate the ingredients of the quotient formula and calculate the Behrend function

vX=102+71+81+61=26.v_X=\frac{10}{2}+\frac{7}{1}+\frac{8}{1}+\frac{6}{1}=26.

To double-check our answer, as this is a 0-dimensional ideal in two variables for which also the conditions of [11 Theorem 3.17 (2)] hold, we can write I=K1K2I=K_1K_2, where K1=((x+y)+(x,y))((x+y)+(x,y)2)K_1=((x+y)+(x,y))((x+y)+(x,y)^2) and K2=((x+y2)+(x,y))((x+y2)+(x,y)2)((x+y2)+(x,y)3)K_2=((x+y^2)+(x,y))((x+y^2)+(x,y)^2)((x+y^2)+(x,y)^3) and conclude that vX=5+14+2×2×3−1×(2+3)=26v_X=5+14+2\times2\times3-1\times(2+3)=26, by [11 Theorem 3.17 (2)].

Why does minimisation preserve multiplicity?

As we saw in this section, the multiplicity is the sum of the volumes of parallelepipeds spanned by (sub)-simplices of each (N−1)(N-1)-boundary associated with each monomial ideal. Without loss of generality, we answer the question for N=2N=2. To answer the question, we need to check it for each multi⁡i\operatorname{multi}_i, i.e., for the parallelogram generated by vectors (pi−1,qi−1)(p_{i-1},q_{i-1}) and (pi,qi)(p_i,q_i), for each ii. For the sake of simplicity, we restrict to two dimensions, and the argument could be easily generalised to higher dimensions. Let II be an ideal in C[x,y]\mathbb{C}[x,y], so that the parallelogram produced by two of the successive generators xpi−1yqi−1,xpiyqix^{p_{i-1}}y^{q_{i-1}},x^{p_i}y^{q_i} of Imin⁡I_{\min} is as in Figure 11.

Parallelogram generated by $(p_{i-1},q_{i-1})$ and $(p_i,q_i)$

Figure 11.

Now, assume that there is a removed generator xαyβx^\alpha y^\beta on the smaller diagonal of the parallelogram, which is removed in the process of minimisation. Associated to such a point, there are two smaller parallelograms as in Figure 12.

Two smaller parallelograms generated by $(p_{i-1},q_{i-1})$, $(\alpha,\beta)$, and $(p_i,q_i)$

Figure 12.

We want to show that the area of the parallelogram generated by the vectors (pi−1,qi−1)(p_{i-1},q_{i-1}) and (pi,qi)(p_i,q_i) is equal to the sum of the areas of the parallelograms generated by (pi−1,qi−1)(p_{i-1},q_{i-1}) and (α,β)(\alpha,\beta), on the one hand, and (α,β)(\alpha,\beta) and (pi,qi)(p_i,q_i), on the other hand. We denote the area of the parallelogram generated by vectors vv and ww by A(v,w)A(v,w). First note that the line from the point (pi−1,qi−1)(p_{i-1},q_{i-1}) to the point (pi,qi)(p_i,q_i) is given by the following equation in the xyx y plane:

y−qi−1=qi−qi−1pi−pi−1(x−pi−1).y-q_{i-1}=\frac{q_i-q_{i-1}}{p_i-p_{i-1}}(x-p_{i-1}).

As the point (α,β)(\alpha,\beta) is on this line, we have

α=β(pi−pi−1)−qi−1(pi−pi−1)+pi−1(qi−qi−1)qi−qi−1.\alpha=\frac{\beta(p_i-p_{i-1})-q_{i-1}(p_i-p_{i-1})+p_{i-1}(q_i-q_{i-1})}{q_i-q_{i-1}}.

Therefore, we have

A((pi−1,qi−1),(α,β))+A((α,β),(pi,qi))=(αqi−1−pi−1β)+(piβ−αqi)=α(qi−1−qi)+β(pi−pi−1)=−β(pi−pi−1)+qi−1(pi−pi−1)−pi−1(qi−qi−1))+β(pi−pi−1)=qi−1pi−qipi−1=A((pi−1,qi−1),(pi,qi)),\begin{aligned} A((p_{i-1},q_{i-1}),(\alpha,\beta))+A((\alpha,\beta),(p_i,q_i))&=(\alpha q_{i-1}-p_{i-1}\beta)+(p_i\beta-\alpha q_i)\\ &=\alpha(q_{i-1}-q_i)+\beta(p_i-p_{i-1})\\ &=-\beta(p_i-p_{i-1})+q_{i-1}(p_i-p_{i-1})\\ &\quad-p_{i-1}(q_i-q_{i-1}))+\beta(p_i-p_{i-1})\\ &=q_{i-1}p_i-q_i p_{i-1}\\ &=A((p_{i-1},q_{i-1}),(p_i,q_i)), \end{aligned}

as desired.

Appendix A. A generalised Pick’s Theorem

The argument in this section can be written without any mention of the Behrend function. However, as we use a similar combinatorial spirit as of the Behrend function for monomial ideals, we put the result here.

Behrend function in terms of the equations of the lower boundary of the convex hull in three dimensions

We have the following lemma to recover the contribution of the Behrend function on each component from the equation of the plane containing the polygon.

Theorem A.1. Let II be a monomial ideal in A=C[x1,x2,…,xN]A=\mathbb{C}[x_1,x_2,\ldots,x_N] that is singular at the origin, and let X=Spec⁡(A/I)X=\operatorname{Spec}(A/I). Then, for any polygon Pi,jP_{i,j} in the lower boundary of the convex hull of II, we have

vX,i,j(0)=∣d∣,v_{X,i,j}(\mathbf{0})=|d|,

where vX,i,jv_{X,i,j} is the contribution of the Behrend function evaluated at the origin (considering with a suitable sign), and dd is the constant term of the primitive equation of the plane defined by Pi,jP_{i,j}.

Proof. Let a1x1+a2x2+…+aNxN+d=0a_1x_1+a_2x_2+\ldots+a_Nx_N+d=0 be the primitive8 equation of the plane defined by Pi,jP_{i,j}. We only need to restrict to the case when the polygon Pi,jP_{i,j} is just a (N−1)(N-1)-simplex, and the only lattice points on it are just the vertices. From Theorem 5.2, we have

vX,i,j(0)=vol⁡(parallelepiped spanned by the vertices of Pi,j).v_{X,i,j}(\mathbf{0})=\operatorname{vol}(\text{parallelepiped spanned by the vertices of }P_{i,j}).

If A1,A2,…,ANA_1,A_2,\ldots,A_N are the vertices (as in the Figure 13 for N=3N=3 and setting A1=A,A2=BA_1=A,A_2=B and A3=CA_3=C), then we have

A triangular polygon in three-dimensional coordinates with vectors and axes

Figure 13.

vX,i,j(0)=vol⁡(parallelepiped spanned by A1,A2,…,AN)=N!vol⁡(N-simplex spanned by A1,A2,…,AN and the origin)=N!(1Nvol⁡(Pi,j)h)=(N−1)!vol⁡(Pi,j)h,\begin{aligned} v_{X,i,j}(\mathbf{0})&=\operatorname{vol}(\text{parallelepiped spanned by }A_1,A_2,\ldots,A_N)\\ &=N!\operatorname{vol}(N\text{-simplex spanned by }A_1,A_2,\ldots,A_N\text{ and the origin})\\ &=N!\left(\frac{1}{N}\operatorname{vol}(P_{i,j})h\right)=(N-1)!\operatorname{vol}(P_{i,j})h, \end{aligned}

where hh is the distance of the origin from Pi,jP_{i,j}, which is equal to

∣d∣a12+…+aN2.\frac{|d|}{\sqrt{a_1^2+\ldots+a_N^2}}.

On the other hand, vol⁡(Pi,j)=1(N−1)!∣n⃗∣\operatorname{vol}(P_{i,j})=\frac{1}{(N-1)!}|\vec{n}|, where n⃗\vec{n} is the normal vector to Pi,jP_{i,j}. It is straightforward to check that by our assumption, we have ∣n⃗∣=a12+a22+…+aN2|\vec{n}| = \sqrt{a_1^2 + a_2^2 + \ldots+ a_N^2}, in this case. Hence, we have

νX,i,j=(N−1)!vol⁡(Pi,j)h=(N−1)!(1(N−1)!∣n⃗∣)(∣d∣a12+a22+…+aN2)=∣d∣.\nu_{X,i,j} = (N-1)!\operatorname{vol}(P_{i,j})h = (N-1)!\left(\frac{1}{(N-1)!}|\vec{n}|\right)\left(\frac{|d|}{\sqrt{a_1^2+a_2^2+\ldots+a_N^2}}\right)=|d|.

Example A.2. Let I=(x3,y3,z2,xy,xz,yz)I=(x^3,y^3,z^2,xy,xz,yz) be a zero-dimensional ideal in C[x,y,z]\mathbb{C}[x,y,z]. Using Theorem A.1, we can see that the defining equation of the three planes containing the irreducible component of the normal cone of II are x+y+z−2=0x+y+z-2=0, 2x+y+2z−3=02x+y+2z-3=0 and 2x+y+2z−3=02x+y+2z-3=0. Hence, the Behrend number is 2+3+3=82+3+3=8. See Figure 14. Note that this is the example in [11 Section 7] where the authors introduce to show the difficulties in three dimensions.

Figure 14. (1) The irreducible components of the normal cone of I=(x3,y3,z2,xy,xz,yz)I=(x^3,y^3,z^2,xy,xz,yz). (2) The corresponding defining planes.

Pick’s Theorem for embedded polygons

We prove a generalised version of Pick’s theorem for embedded polygons in R3\mathbb{R}^3. The classical Pick’s theorem states the area of a plane polygon in terms of the number of the lattice points inside and on the boundary of the polygon. First, we need the following definition and lemma.

Definition A.3 (Fine triangulation). A fine triangulation for a polygon, is an integral triangulation (i.e., the vertices of all the triangles are lattice points) with the maximum number of triangles.

Definition A.4 (Empty simplex/empty triangle). An empty NN-simplex is a simplex with lattice points only on the vertices. An empty 22-simplex is an empty triangle.

Lemma A.5. Let PP be a polygon in the primitive plane ax+by+cz+d=0ax+by+cz+d=0, and let I(P),B(P)I(P), B(P) be the number of internal lattice points and the number of lattice points on the boundary of PP, respectively. Then

the number of empty triangles in a fine integral triangulation of P=B(P)+2I(P)−2P = B(P)+2I(P)-2.

Proof. Without loss of generality, and up to transformations, we can assume that c=1,a,b,d=0c=1,a,b,d=0. We proceed by induction on the number of lattice points which are not vertices. The base case is when PP a empty triangle, for which we have I(P)=0,B(P)=3I(P)=0,B(P)=3, hence B(P)+2I(P)−2=1B(P)+2I(P)-2=1, which is obviously the number of empty triangles in PP. Now, let PP be a polygon with at least one non-vertex lattice point, and let P′P' be a polygon purely inside PP with B′(P′)B'(P') lattice points on its boundary and I′(P′)I'(P') internal lattice points, and we can assume that the statement holds for P′P' and we want to show that B(P)+2I(P)−2B(P)+2I(P)-2 is also the number of empty triangles in PP. See Figure 15Figure 1.

Figure 15. (1) Proof by induction. (2) Any lattice point pp between PP and P′P' adds two additional empty triangles.

If there is no lattice point between PP and P′P', then there are B(P)+B′(P′)B(P)+B'(P') empty triangles between PP and P′P': set m1=min⁡{B(P),B′(P′)}m_1=\min\{B(P),B'(P')\} and m2=max⁡{B(P),B′(P′)}m_2=\max\{B(P),B'(P')\}. Then there are m1m_1 (potentially, degenerate or concave) quadrilaterals with two sides of it being a side of PP and a side of P′P'. This gives 2m12m_1 empty triangles. On the other hand, for each remaining m2−m1m_2-m_1 side, there exist a unique triangle. Totally this gives 2m1+(m2−m1)=m1+m2=B(P)+B′(P′)2m_1+(m_2-m_1)=m_1+m_2=B(P)+B'(P') empty triangles between PP and P′P'. Now, note that each lattice points pp between PP and P′P' adds two more empty triangles to B(P)+B′(P′)B(P)+B'(P') (see Figure 15Figure 2); i.e., there are totally B(P)+B′(P′)+2(I(P)−I′(P′)−B′(P′))B(P)+B'(P')+2(I(P)-I'(P')-B'(P')) empty triangles between PP and P′P'. Combining this with the induction hypothesis (the number of empty triangles in P′P' is B′(P′)+2I′(P′)−2B'(P')+2I'(P')-2), we conclude that there are B′(P′)+2I′(P′)−2+B(P)+B′(P′)+2(I(P)−I′(P′)−B′(P′))=B'(P')+2I'(P')-2+B(P)+B'(P')+2(I(P)-I'(P')-B'(P'))=

2I(P)+B(P)−2=2I(P)+B(P)-2= empty triangles in PP, as claimed.

Note that the result holds for every plane in R3\mathbb{R}^3, since linear transformation of planes in R3\mathbb{R}^3 preserves the lattice points and empty triangles.

Example A.6. Figure 15 shows two examples related to Lemma A.5, for B(P)=10B(P)=10, I(P)=34I(P)=34 (left) and B(P)=5B(P)=5, I(P)=6I(P)=6 (right).

Now, we can prove a generalised version of Pick’s Theorem:

Theorem A.7. Let PP be a plane convex polygon with some non-lattice vertices so that it can be integrally and generically embedded in R3\mathbb{R}^{3}, i.e., there is an embedding of PP into R3\mathbb{R}^{3} with all vertices lattice points so that it is not perpendicular to any of the three coordinate planes. Let I^(P)\widehat{I}(P) and B^(P)\widehat{B}(P) be the number of internal lattice points and the number of lattice points on the boundary of the embedded polygon in R3\mathbb{R}^{3}, respectively. Then

area⁡(P)=(B^(P)2+I^(P)−1)(a2+b2+c2)\operatorname{area}(P)=\left(\frac{\widehat{B}(P)}{2}+\widehat{I}(P)-1\right)\left(\sqrt{a^{2}+b^{2}+c^{2}}\right)

where ax+by+cz+d=0ax+by+cz+d=0 is the primitive equation of the plane defined by the embedded polygon in R3\mathbb{R}^{3}, and (a,b,c)(a,b,c) is primitive, and d≠0d\ne0.

Proof. From the proof of Theorem A.1 for N=3N=3, the volume of each empty triangle in PP with primitive equation ax+by+cz+d=0ax+by+cz+d=0 is a2+b2+c2/2\sqrt{a^{2}+b^{2}+c^{2}}/2, and from Lemma A.5, the number of such empty triangles in PP is B^(P)+2I^(P)−2\widehat{B}(P)+2\widehat{I}(P)-2. Combining these implies the claim.

Macaulay2 codes

The codes in this appendix are due to Mike Stillman in April 2023. The codes compute the ingredients of the quotient formula for a given ideal, which defines our affine scheme. Here, we consider the example of the one-dimensional subscheme of A3\mathbb{A}^{3} defined by the ideal I=(xy,yz,xz)I=(xy,yz,xz). Here, L is the same as in the proof of Theorem 2.1, and Qs is the same as Qi,jQ_{i,j} or QjQ_j, and Ps is the same as pi,jp_{i,j} or pjp_j (as in the proof of Theorem 2.1).

R=QQ[x,y,z];
I=ideal (x*y,y*z,x*z);
J=trim I;
R1 = normalCone J;
S = first flattenRing R1;
Rsg = (coefficientRing S)[gens S];
L = sub(trim ideal S, Rsg);
Qs = primaryDecomposition L;
maxdim = max (Qs/dim);
Qs = select(Qs, q -> dim q == maxdim);
Ps = Qs/radical;
Qs/degree
Ps/degree

References

Centre for Mathematical Sciences, University of Cambridge, Wilberforce Road, CB3 0WA, Cambridge, United Kingdom

Email address: [email protected]

References

  1. [1]M. F. Atiyah and I. G. MacDonald, Introduction to commutative algebra, Addison-Wesley Publishing Company, 1969. 2. D. Avis, D. Bremner, and R. Seidel, How good are convex hull algorithms?, Computational Geometry: Theory and Applications 7 (1997), no. 5-6, 265–301.
  2. [3]C. B. Barber, D. P. Dobkin, and H. Huhdanpaa, The quickhull algorithm for convex hulls, ACM Transactions on Mathematical Software 22 (1996), no. 4, 469–483.DOI
  3. [4]K. Behrend, Donaldson–Thomas type invariants via microlocal geometry, Ann. of Math. 2 (2009), no. 170, 1307–1338.DOI
  4. [7]M. Blanco and F. Santos, Non-spanning lattice 3-polytopes, Journal of Combinatorial Theory, Series A 161 (2019), 112–133.DOI
  5. [8]W. Bruns and J. Gubeladze, Polytopes, rings, and K-theory, Springer, 2009.
  6. [9]D. Eisenbud, Commutative algebra, Graduate Texts in Mathematics, vol. 150, Springer-Verlag, 1995.
  7. [10]W. Fulton, Introduction to toric varieties. (AM-131), Princeton University Press, 1993.
  8. [11]M. Graffeo and A. T. Ricolfi, On the Behrend function and the blowup of some fat points, Advances in Mathematics 415 (2023).arxiv.org/abs/2202.06904
  9. [12]D. Grayson and M. Stillman, Macaulay2, a software system for research in algebraic geometry, http://www.math.uiuc.edu/Macaulay2/.
  10. [13]M. Hohenwarter, M. Borcherds, G. Ancsin, B. Bencze, M. Blossier, J. Élías, K. Frank, L. Gál, A. Hofstätter, F. Jordan, Z. Konečný, Z. Kovács, E. Lettner, S. Lizelfelner, B. Parisse, C. Solyom-Gecse, C. Stadlbauer, and M. Tomaschko, GeoGebra 5.0.507.0, October 2018, http://www.geogebra.org.
  11. [15]J. Jelsiejew, M. Kool, and R. F. Schiemann, Behrend’s function is not constant on Hilbⁿ(C³), Geometry & Topology 29 (2025), no. 8, 4469–4476.
  12. [16]Y. Jiang, Note on macpherson’s local euler obstruction, Mich. Math. J. 687 (2019), no. 2, 227–250.arxiv.org/abs/1412.3720
  13. [17]D. Joyce and Y. Song, A theory of generalized Donaldson-Thomas invariants, Mem. Amer. Math. Soc. 217 (2012), no. 1020, iv+199.arxiv.org/abs/0810.5645
  14. [18]V. Lyubashevsky, C. Peikert, and O. Regev, On Ideal Lattices and Learning with Errors over Rings, Advances in Cryptology – EUROCRYPT 2010. Lecture Notes in Computer Science 6110 (2010).
  15. [19]R. MacPherson, Chern class for singular algebraic varieties, Ann. Math. 100 (1974), no. 2, 423–432.DOI
  16. [20]A. Parusiński and P. Pragacz, Characteristic classes of hypersurfaces and characteristic cycles, J. Algebraic Geom. 10 (2001), no. 1, 63–79.DOI
  17. [21]C. Polini, A. Sammartano, and B. Ulrich, Behrend function and blowup algebras, arXiv:2608.04713 (2026).arxiv.org/abs/2608.04713
  18. [22]A. T. Ricolfi, A sign that used to annoy me, and still does, J. Geom. Phys. 125 (2023).arxiv.org/abs/2306.08457
  19. [23]L. Si, Pick’s Theorem in Two-Dimensional Subspace of ℝ³, The Scientific World Journal (2015).DOI
  20. [24]The Stacks project authors, The stacks project, https://stacks.math.columbia.edu, 2024.
  21. [25]R. P. Stanley, A zonotope associated with graphical degree sequences, Applied Geometry and Discrete Math. DIMACS Series in Discrete Mathematics and Theoretical Computer Science 4 (1991), 555–570.DOI
  22. [26]S. Telen, Introduction to toric geometry, arXiv:2203.01690 (2022).arxiv.org/abs/2203.01690

Paper details

vlnc:W_5k2waus8
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents