betaFeedback

WorksMathematics

Associated primes and v-numbers of integral closures of powers of edge ideals

Abstract

Let I=I(G)I = I(G) be the edge ideal of a finite simple graph GG. We investigate the relationship between the associated primes and vv-numbers of powers ItI^t and their integral closures It‾\overline{I^t}. For the third power, we establish the associated prime containment Ass⁡(S/I3‾)⊆Ass⁡(S/I3)\operatorname{Ass}(S/\overline{I^3}) \subseteq \operatorname{Ass}(S/I^3), which yields v(I3)≤v(I3‾)v(I^3) \le v(\overline{I^3}). In contrast, for t≥4t \ge 4, we show that Ass⁡(S/It‾)\operatorname{Ass}(S/\overline{I^t}) and Ass⁡(S/It)\operatorname{Ass}(S/I^t) are in general incomparable. Furthermore, for every fixed t≥3t \ge 3 and r≥4r\ge 4, we construct families of graphs for which the discrepancies v(It‾)−v(It)v(\overline{I^t}) - v(I^t) and v(Ir)−v(Ir‾)v(I^r)-v(\overline{I^r}) are arbitrarily large.

Introduction

Let S=k[x1,…,xn]S=k[x_1,\ldots,x_n] be a standard graded polynomial ring over a field kk, and let I⊆SI\subseteq S be a nonzero homogeneous ideal. For an associated prime P∈Ass⁡(S/I)P\in\operatorname{Ass}(S/I), the local vv-number of II at PP is defined by

vP(I)=min⁡{d≥0∣there exists f∈Sd such that (I:f)=P}.v_P(I)=\min\{d\geq0\mid\text{there exists }f\in S_d\text{ such that }(I:f)=P\}.

The vv-number of II, introduced by Cooper et al. [4] in connection with the minimum distance function of projective Reed–Muller-type codes, is given by

v(I)=min⁡P∈Ass⁡(S/I)vP(I).v(I)=\min_{P\in\operatorname{Ass}(S/I)}v_P(I).

The vv-number of edge ideals was studied by Jaramillo and Villarreal [10]. Conca [3] and Ficarra and Sgroi [7] established the eventual linearity of the vv-numbers of ordinary powers of homogeneous ideals. More recently, Ficarra and Sgroi [8] studied the vv-function of Noetherian graded filtrations and proved its eventual quasi-linearity, including for integral closure filtrations.

Vanmathi A and Sarkar [17] investigated the vv-numbers of integral closures of powers of monomial ideals. Biswas, Mandal, and Phukan [1] showed that the vv-numbers of a monomial ideal and its integral closure need not be comparable. They also proved that if GG is a connected graph, then v(I(G)t)=v(I(G)t‾)v(I(G)^t)=v(\overline{I(G)^t}) for t≥1+∣E(G)∣t\geq1+|E(G)|, and obtained an eventual comparison for disconnected graphs. In this paper, we investigate the relationship between v(I(G)t)v(I(G)^t) and v(I(G)t‾)v(\overline{I(G)^t}) at individual powers. Since I(G)t=I(G)t‾I(G)^t=\overline{I(G)^t} for t=1,2t=1,2, the third power is the first instance where the two ideals, and consequently their associated primes and vv-numbers, can diverge.

On the associated prime side, Trung [16] studied the stability of the associated primes of integral closures of powers of monomial ideals. For edge ideals, Martínez-Bernal, Morey, and Villarreal [12] showed that the eventual stable sets of associated primes of ordinary powers and their integral closures coincide, while Mau and Trung [13] obtained bounds for the stability indices of associated primes of integral closures. In contrast to the combinatorial description of associated primes of ordinary powers given by Lam and Trung [11], a general description of Ass⁡(S/I(G)t‾)\operatorname{Ass}(S/\overline{I(G)^t}) for arbitrary graphs and powers is not available.

Because the vv-number is defined in terms of associated primes, we first investigate their relationship for ordinary powers and their integral closures. We also compare the first powers at which the homogeneous maximal ideal becomes an associated prime. Our main contributions are organized as follows:

  1. Local equality at minimal primes. For every minimal prime P∈Min⁡(S/I(G))P \in\operatorname{Min}(S/I(G)) and every t≥1t \ge1, we prove that the local vv-numbers coincide:

vP(I(G)t)=vP(I(G)t‾).v_P(I(G)^t) = v_P(\overline{I(G)^t}).
  1. Containment and inequality for the third power. For t=3t = 3, we prove that

Ass⁡(S/I(G)3‾)⊆Ass⁡(S/I(G)3)\operatorname{Ass}(S/\overline{I(G)^3}) \subseteq\operatorname{Ass}(S/I(G)^3)

and

v(I(G)3)≤v(I(G)3‾).v(I(G)^3) \le v(\overline{I(G)^3}).
  1. Incomparability for higher powers. For every fixed integer t≥4t \ge4, we show that Ass⁡(S/I(G)t‾)\operatorname{Ass}(S/\overline{I(G)^t}) and Ass⁡(S/I(G)t)\operatorname{Ass}(S/I(G)^t) are, in general, incomparable under set inclusion. Moreover, the first power at which the homogeneous maximal ideal becomes associated to an integral closure can be arbitrarily smaller than the corresponding first power for ordinary powers.

  1. Arbitrarily large gaps for higher powers. For every fixed integer t≥3t \ge3 and any integer r≥1r \ge1, we construct families of graphs for which the discrepancy

v(I(G)t‾)−v(I(G)t)≥r−1.v(\overline{I(G)^t}) - v(I(G)^t) \ge r - 1.

Furthermore, for t≥4t \ge4, we show that the reverse difference v(I(G)t)−v(I(G)t‾)v(I(G)^t) - v(\overline{I(G)^t}) can also be made arbitrarily large.

Acknowledgment. Chau gratefully acknowledges that this research was supported in part by the Pacific Institute for the Mathematical Sciences and a grant from the Simons Foundation International [SFI-MPS-T-Institutes-00020822, OY]. Hà is partially supported by a Simons Foundation grant. u is partially supported by the National Foundation for Science and Technology Development under grant number Nafosted/101/10104/2026/1279.

AI Disclosure. Theorem 4.5 was proved by ChatGPT-6.1 Sol in the case t=4t = 4 and generalized by the authors.

Preliminaries

We collect the graph-theoretic notation and the criterion for membership in the integral closure that will be used throughout the paper.

Definition 2.1. Let GG be a simple graph with vertex set V(G)={1,…,n}V(G) = \{1,\ldots,n\} and edge set E(G)E(G).

  1. A simple graph HH is a subgraph of GG if V(H)⊆V(G)V(H) \subseteq V(G) and E(H)⊆E(G)E(H) \subseteq E(G). It is an induced subgraph of GG if E(H)={{u,v}∈E(G)∣u,v∈V(H)}E(H) = \{\{u,v\} \in E(G) \mid u,v \in V(H)\}.

  2. For a subset U⊆V(G)U \subseteq V(G), we denote by G[U]G[U] and G−UG-U the induced subgraphs of GG on UU and on V(G)∖UV(G) \setminus U, respectively.

  3. For a vertex v∈V(G)v \in V(G), NG(v)N_G(v) denotes the set of neighbors of vv. For a subset U⊆V(G)U \subseteq V(G), NG(U)=⋃u∈UNG(u)N_G(U) = \bigcup_{u \in U} N_G(u) denotes the neighborhood of UU, and NG[U]=NG(U)∪UN_G[U] = N_G(U) \cup U denotes the closed neighborhood of UU.

  4. A subset X⊆V(G)X \subseteq V(G) is called an independent set if no two vertices in XX are adjacent in GG. We denote by α(G)\alpha(G) the maximum size of an independent set of GG.

  5. A subset C⊆V(G)C \subseteq V(G) is called a vertex cover (or simply a cover) of GG if it meets every edge of GG; that is, C∩e≠∅C \cap e \ne\varnothing for all e∈E(G)e \in E(G).

  6. A graph GG is called bipartite if its vertex set can be partitioned as V(G)=U∪VV(G) = U \cup V with U∩V=∅U \cap V = \varnothing such that every edge in E(G)E(G) has one endpoint in UU and the other in VV.

(7) A graph GG is called strongly non-bipartite if each of its connected components is non-bipartite.

(8) An edge ee of GG is called a bridge if its removal increases the number of connected components by one.

(9) A matching in GG is a set of pairwise non-adjacent edges. The matching number of GG, denoted by mat⁡(G)\operatorname{mat}(G), is the size of a maximum matching of GG. A perfect matching is a matching that covers every vertex of GG.

(10) A graph GG is called factor-critical if G−vG-v has a perfect matching for every v∈V(G)v \in V(G).

Throughout this paper, we denote by S=k[x1,…,xn]S = \mathbb{k}[x_1,\ldots,x_n] the polynomial ring over a field k\mathbb{k}, and by m=(x1,…,xn)\mathfrak{m}=(x_1,\ldots,x_n) its maximal homogeneous ideal. For a nonzero monomial f∈Sf \in S, its support, denoted by supp⁡(f)\operatorname{supp}(f), is the set of all variables in SS that divide ff. For each i∈[n]i \in[n], we denote by deg⁡i(f)\operatorname{deg}_i(f) the exponent of xix_i in ff, and by deg⁡(f)\operatorname{deg}(f) the total degree of ff. For any subset X⊆[n]X \subseteq[n], we define

deg⁡X(f)=∑i∈Xdeg⁡i(f).\operatorname{deg}_X(f)=\sum_{i\in X}\operatorname{deg}_i(f).

Definition 2.2. Let GG be a simple graph on V(G)=[n]V(G)=[n] and let a=(a1,…,an)∈Nn\mathbf{a}=(a_1,\ldots,a_n)\in\mathbb{N}^n be a vector of positive integers. The parallelization of GG with respect to a\mathbf{a}, denoted by GaG_{\mathbf{a}}, is the graph obtained by replicating each vertex i∈V(G)i\in V(G) into aia_i copies, denoted i1,…,iaii_1,\ldots,i_{a_i}, such that {ik,jℓ}∈E(Ga)\{i_k,j_\ell\}\in E(G_{\mathbf{a}}) if and only if {i,j}∈E(G)\{i,j\}\in E(G).

For the subsequent arguments, we use the fractional matching number with capacity and its dual formulation.

Definition 2.3. Let GG be a simple graph with V(G)={v1,…,vn}V(G)=\{v_1,\ldots,v_n\}, and let a∈Nn\mathbf{a}\in\mathbb{N}^n be an exponent vector. The fractional matching number of GG with capacity a\mathbf{a} is defined as

νG∗(a)=max⁡{∑e∈E(G)ze | ze≥0 for all e∈E(G), and ∑e∋vze≤av for all v∈V(G)}.\nu_G^*(\mathbf{a})=\max\left\{\sum_{e\in E(G)}z_e\ \middle|\ z_e\geq0\text{ for all }e\in E(G),\text{ and }\sum_{e\ni v}z_e\leq a_v\text{ for all }v\in V(G)\right\}.

We have the following membership criterion in terms of νG∗(a)\nu_G^*(\mathbf{a}), see [9 Proposition 1.1].

Lemma 2.4. Let GG be a simple graph and let a∈Nn\mathbf{a}\in\mathbb{N}^n. Then xa∈I(G)tx^{\mathbf{a}}\in I(G)^t if and only if νG∗(a)≥t\nu_G^*(\mathbf{a})\geq t.

Let QQ be the covering polyhedron of GG, namely

Q={y∈R≥0∣V(G)∣ | yu+yv≥1 for every edge {u,v}∈E(G)}.Q=\left\{\mathbf{y}\in\mathbb{R}_{\geq0}^{|V(G)|}\ \middle|\ y_u+y_v\geq1\text{ for every edge }\{u,v\}\in E(G)\right\}.

By linear programming duality, νG∗(a)\nu_G^*(\mathbf{a}) can also be expressed as the optimal value of the dual problem:

νG∗(a)=min⁡{∑v∈V(G)avyv | y∈Q}.\nu_G^*(\mathbf{a})=\min\left\{\sum_{v\in V(G)}a_vy_v\ \middle|\ \mathbf{y}\in Q\right\}.

Associated primes of ordinary powers and their integral closures

In this section, we investigate associated primes of powers of an edge ideal and their integral closures. Our first result establishes a containment of associated primes for the third power.

Theorem 3.1. Let I=I(G)I=I(G) be the edge ideal of a simple graph GG. If ff is a monomial such that (I3‾:f)=P(\overline{I^3}:f)=P is a prime ideal, then (I3:f)=P(I^3:f)=P. In particular,

Ass⁡(S/I3‾)⊆Ass⁡(S/I3)andν(I3)≤ν(I3‾).\operatorname{Ass}(S/\overline{I^3})\subseteq\operatorname{Ass}(S/I^3)\quad\text{and}\quad\nu(I^3)\leq\nu(\overline{I^3}).

Proof. By [14 Theorem 2.10],

I3‾=I3+(mCmD∣C,D are exceptional pairs of triangles of G)\overline{I^3}=I^3+\left(m_Cm_D\mid C,D\text{ are exceptional pairs of triangles of }G\right)

where mC=∏xv∈E(C)xvm_C=\prod_{xv\in E(C)}xv, and C,DC,D form an exceptional pair if NG[C]∩D=∅N_G[C]\cap D=\varnothing. Set L=I3‾L=\overline{I^3}, and suppose that (L:f)=P(L:f)=P is a monomial prime ideal. Since I3⊆LI^3\subseteq L, we have (I3:f)⊆P(I^3:f)\subseteq P.

Assume for contradiction that there exists a variable x∈Px\in P such that xf∉I3xf\notin I^3. Then xf∈L∖I3xf\in L\setminus I^3, so xfxf is divisible by mCmDm_Cm_D for some exceptional pair of triangles C,DC,D. Write C={x,y,z}C=\{x,y,z\} and D={a,b,c}D=\{a,b,c\}. Since f∉Lf\notin L, the square-free monomial mCmDm_Cm_D cannot divide ff. Thus x∤fx\nmid f and yzabc∣fyzabc\mid f. Furthermore, y2∤fy^2\nmid f and z2∤fz^2\nmid f: otherwise xy,yz,abxy,yz,ab or xz,yz,abxz,yz,ab, respectively, would be three edges whose product divides xfxf.

We claim that yf∉I3yf\notin I^3. Otherwise, since f∉I3f\notin I^3 and yy occurs to exponent one in ff, any product of three edges dividing yfyf must use both copies of yy. At most one of these edges can be yzyz, since zz also occurs to exponent one in ff. As x∤fx\nmid f and no vertex of DD is adjacent to yy, at least one of the two edges incident to yy must be yuyu for some u∉C∪Du\notin C\cup D. But then the product of the three edges yu,xz,abyu,xz,ab divides xfxf, a contradiction. By symmetry, zf∉I3zf\notin I^3.

The monomials yfyf and zfzf have the same support as ff. Every generator mCmDm_Cm_D in the above description of LL is square-free. Therefore, if an exceptional generator divided yfyf or zfzf, it would also divide ff, which is impossible. We conclude that yf,zf∉Lyf,zf\notin L, and hence y,z∉Py,z\notin P. But PP contains II, so it is a vertex cover prime of GG. This contradicts yz∈Iyz\in I.

Thus P⊆(I3:f)P\subseteq(I^3:f), and equality follows. Finally, choose ff with (L:f)=P(L:f)=P and deg⁡(f)=vP(L)\deg(f)=v_P(L). We obtain vP(I3)≤vP(L)v_P(I^3)\leq v_P(L) for every P∈Ass⁡(S/L)P\in\operatorname{Ass}(S/L), which gives the asserted containment and inequality. □

We next give a criterion for the homogeneous maximal ideal to be an associated prime of the integral closure of a power. We first show that every monomial witness for the maximal ideal has a fixed degree.

Lemma 3.2. Let GG be a simple graph and I=I(G)I=I(G). Suppose that f=xaf=x^{\mathbf a} is a monomial such that (It‾:f)=m(\overline{I^t}:f)=\mathfrak m. Then deg⁡(f)=2t−1\deg(f)=2t-1.

Proof. By Lemma 2.4,

νG∗(a)<tandνG∗(a+ev)≥tfor every v∈V(G).\nu_G^*(\mathbf a)<t\quad\text{and}\quad\nu_G^*(\mathbf a+\mathbf e_v)\geq t\quad\text{for every }v\in V(G).

Let y\mathbf y be an optimal extreme point of the covering polyhedron QQ for νG∗(a)\nu_G^*(\mathbf a). We claim that yv>0y_v>0 for all v∈V(G)v\in V(G). Indeed, if yv=0y_v=0, then

νG∗(a+ev)≤(a+ev)⋅y=a⋅y=νG∗(a)<t,\nu_G^*(\mathbf a+\mathbf e_v)\leq(\mathbf a+\mathbf e_v)\cdot\mathbf y=\mathbf a\cdot\mathbf y=\nu_G^*(\mathbf a)<t,

a contradiction.

Let HH be the spanning subgraph of GG whose edges are those {u,v}\{u,v\} satisfying yu+yv=1y_u+y_v=1. Every connected component of HH must contain an odd cycle. Otherwise, a bipartite component, including the possibility of an isolated vertex, would allow us to perturb y\mathbf y by small positive and negative multiples of the vector taking the values 11 and −1-1 on its two parts and zero elsewhere. Since all coordinates of y\mathbf y are positive, the nonnegativity inequalities remain satisfied; the tight edge inequalities are unchanged, and the remaining inequalities remain satisfied for sufficiently small perturbations. This contradicts the fact that y\mathbf y is an extreme point of QQ.

On each connected component of HH, the tight equalities, together with an odd cycle, force yv=1/2y_v=1/2 at every vertex. Hence y=(1/2,…,1/2)\mathbf y=(1/2,\ldots,1/2), and therefore

νG∗(a)=deg⁡(f)2<t.\nu_G^*(\mathbf a)=\frac{\deg(f)}{2}<t.

On the other hand, for any vertex vv, the same feasible dual vector gives

t≤νG∗(a+ev)≤deg⁡(f)+12.t \le\nu_G^*(\mathbf{a}+\mathbf{e}_v) \le\frac{\deg(f)+1}{2}.

Since deg⁡(f)\deg(f) is an integer, these inequalities imply deg⁡(f)=2t−1\deg(f)=2t-1.

The following criterion strengthens the necessary condition in Lemma 3.2 and its converse. For a subset U⊆V(G)U \subseteq V(G), we write a(U)=∑u∈Uau\mathbf{a}(U)=\sum_{u\in U}a_u.

Lemma 3.3. Let GG be a simple graph, let I=I(G)I=I(G), and let f=xaf=x^{\mathbf{a}} be a monomial. Then (It‾:f)=m(\overline{I^t}:f)=\mathfrak{m} if and only if deg⁡(f)=2t−1\deg(f)=2t-1 and

a(NG(U))≥a(U)+1\mathbf{a}(N_G(U)) \ge\mathbf{a}(U)+1

for every nonempty independent set U⊆V(G)U\subseteq V(G).

Proof. Suppose that (It‾:f)=m(\overline{I^t}:f)=\mathfrak{m}. By Lemma 3.2, deg⁡(f)=2t−1\deg(f)=2t-1. Let UU be a nonempty independent set and choose v∈Uv\in U. Since xvf∈It‾x_vf\in\overline{I^t}, there exists a fractional matching with capacity b=a+ev\mathbf{b}=\mathbf{a}+\mathbf{e}_v and total weight at least tt. As ∑u∈V(G)bu=2t\sum_{u\in V(G)}b_u=2t, its total weight must be exactly tt, and all vertex capacities are saturated. Since no edge has both endpoints in UU, all capacity used at vertices of UU must be matched with vertices of NG(U)N_G(U). Thus

a(NG(U))=b(NG(U))≥b(U)=a(U)+1.\mathbf{a}(N_G(U))=\mathbf{b}(N_G(U))\ge\mathbf{b}(U)=\mathbf{a}(U)+1.

Conversely, assume that deg⁡(f)=2t−1\deg(f)=2t-1 and that the stated inequalities hold. In particular, f∉It‾f\notin\overline{I^t} by degree. Fix v∈V(G)v\in V(G) and put b=a+ev\mathbf{b}=\mathbf{a}+\mathbf{e}_v. By Lemma 2.4, it is enough to show that νG∗(b)≥t\nu_G^*(\mathbf{b})\ge t.

The assumed inequalities imply that GG has no isolated vertices. Let y\mathbf{y} be an extreme point of QQ. By [5 Lemma 4.5], every coordinate of y\mathbf{y} belongs to {0,1/2,1}\{0,1/2,1\}. Set

Z={u:yu=0},M={u:yu=1/2},U={u:yu=1}.Z=\{u:y_u=0\},\qquad M=\{u:y_u=1/2\},\qquad U=\{u:y_u=1\}.

The set ZZ is independent and NG(Z)⊆UN_G(Z)\subseteq U. Since deg⁡(f)=2t−1\deg(f)=2t-1, we have

b⋅y=2t−1+a(U)−a(Z)2+yv.\mathbf{b}\mathbin{\cdot}\mathbf{y}=\frac{2t-1+\mathbf{a}(U)-\mathbf{a}(Z)}{2}+y_v.

If Z=∅Z=\varnothing, then yv≥1/2y_v\ge1/2 and b⋅y≥t\mathbf{b}\mathbin{\cdot}\mathbf{y}\ge t. If Z≠∅Z\ne\varnothing, the hypothesis gives a(U)≥a(NG(Z))≥a(Z)+1\mathbf{a}(U)\ge\mathbf{a}(N_G(Z))\ge\mathbf{a}(Z)+1, and again b⋅y≥t\mathbf{b}\mathbin{\cdot}\mathbf{y}\ge t. Thus the dual objective is at least tt at every extreme point of QQ. By linear programming duality, νG∗(b)≥t\nu_G^*(\mathbf{b})\ge t.

It follows that xvf∈It‾x_vf\in\overline{I^t} for every vertex vv. Since f∉It‾f\notin\overline{I^t}, we have (It‾:f)=m(\overline{I^t}:f)=\mathfrak{m}, as desired.

As an immediate consequence, we recover the following sufficient condition, which will also be used in the next section.

Lemma 3.4. Let GG be a graph on an odd number NN of vertices and assume that

∣NG(U)∣>∣U∣|N_G(U)|>|U|

for every nonempty independent set U⊆V(G)U\subseteq V(G). Let I=I(G)I=I(G), let m\mathfrak{m} be the homogeneous maximal ideal of SS, and set t=(N+1)/2t=(N+1)/2. Then

m∈Ass⁡(S/It‾).\mathfrak{m}\in\operatorname{Ass}(S/\overline{I^t}).

Proof. Take f=∏v∈V(G)xvf=\prod_{v\in V(G)}x_v. Then deg⁡(f)=N=2t−1\deg(f)=N=2t-1, and the inequalities in Lemma 3.3 are precisely the assumed strict inequalities. Thus (It‾:f)=m(\overline{I^t}:f)=\mathfrak{m}.

We recall the generalized ear decompositions and invariants of Lam and Trung [11], which will be used to compare the associated primes of ordinary powers.

Definition 3.5. A generalized ear decomposition of a graph GG is a sequence E\mathcal{E} of walks in GG such that the first walk is closed, the endpoints of each subsequent walk belong to earlier walks, and every vertex of GG appears exactly once as an inner point of E\mathcal{E}. A vertex is an inner point if it is neither the last vertex of the first walk nor an endpoint of any subsequent walk; we count the first vertex of the first walk as an inner point. A generalized ear decomposition is odd-beginning if the first walk in each connected component is an odd cycle.

Definition 3.6. For a strongly non-bipartite graph GG, let φ∗(G)\varphi^{*}(G) be the minimum number of even walks in odd-beginning generalized ear decompositions of its connected components, and set

μ∗(G)=φ∗(G)+∣V(G)∣−c(G)2,\mu^{*}(G) = \frac{\varphi^{*}(G) + |V(G)| - c(G)}{2},

where c(G)c(G) denotes the number of connected components of GG.

We will use the criterion [11 Theorem 3.6]: m∈Ass⁡(S/I(G)t)\mathfrak{m} \in\operatorname{Ass}(S/I(G)^{t}) if and only if there exists a dominating set U⊆V(G)U \subseteq V(G) such that G[U]G[U] is strongly non-bipartite and μ∗(G[U])<t\mu^{*}(G[U]) < t. We will also use [11 Lemma 3.9], which gives μ∗(H)=μ∗(H−w)+1\mu^{*}(H) = \mu^{*}(H-w) + 1 when HH is strongly non-bipartite and ww is a leaf.

Lemma 3.7. Let GG be a connected graph and let ee be a bridge of GG. Then every generalized ear decomposition of GG contains the closed walk of length two along ee.

Proof. Let AA and BB be the components of G−eG-e, and suppose that the first walk of a generalized ear decomposition lies in AA. This is automatic unless the first walk is the closed walk of length two along ee, in which case there is nothing to prove. Let WW be the first walk using ee. Before WW, all vertices that have appeared lie in AA, so both endpoints of WW lie in AA. The walk WW must cross ee into BB and return along ee. Since its inner vertices are new and distinct, it can contain no additional vertices between the two crossings, and it must be the closed walk of length two along ee. The same argument applies if the first walk lies in BB. □\square

We can now construct examples for which an integral closure has the maximal ideal as an associated prime while the corresponding ordinary power does not.

Theorem 3.8. For every integer t≥4t \ge4, there exists a simple graph GG such that

m∈Ass⁡(S/I(G)t‾)butm∉Ass⁡(S/I(G)t).\mathfrak{m} \in\operatorname{Ass}(S/\overline{I(G)^{t}}) \quad\text{but} \quad\mathfrak{m} \notin\operatorname{Ass}(S/I(G)^{t}).

Proof. Fix t≥4t \ge4 and set k=t−3k = t - 3. Let C3C_{3} and C2k+1C_{2k+1} be disjoint odd cycles, and let zz be a vertex joined to one vertex of each cycle. Denote by BB the graph on C3∪C2k+1∪{z}C_{3} \cup C_{2k+1} \cup\{z\}. Let GG be obtained from BB by attaching one leaf to every vertex of BB. Thus ∣V(B)∣=2t−1|V(B)| = 2t - 1. Consider the monomial

f=∏v∈V(B)xv.f = \prod_{v \in V(B)} x_v.

Since deg⁡(f)=2t−1\deg(f) = 2t - 1, we have f∉I(G)t‾f \notin\overline{I(G)^{t}}. We claim that xwf∈I(G)t‾x_w f \in\overline{I(G)^{t}} for every vertex ww of GG. If ww belongs to one of the odd cycles, we may use the two edges of that cycle incident to ww, together with a perfect matching of the remaining vertices of that cycle. The other odd cycle, together with zz, has a perfect matching. These edges give tt edge factors dividing xwfx_w f. If ww is the leaf attached to a vertex vv of either cycle, then B−vB-v has a perfect matching of size t−1t-1. Together with the edge wvwv, this gives tt edge factors dividing xwfx_w f. If w=zw=z, we use the two edges joining zz to the two cycles and perfect matchings of the even paths obtained by deleting their neighbors from the cycles. Again, the product of these tt edges divides xzfx_zf. Finally, if ww is the leaf attached to zz, assign fractional matching weight 11 to the edge wzwz and weight 1/21/2 to every edge of the two odd cycles. This is a fractional matching with capacity given by the exponent vector of xwfx_wf and total weight

1+32+2k+12=k+3=t.1+\frac{3}{2}+\frac{2k+1}{2}=k+3=t.

By Lemma 2.4, xwf∈I(G)t‾x_wf\in\overline{I(G)^t} in this last case as well. Hence (I(G)t‾:f)=m(\overline{I(G)^t}:f)=\mathfrak{m}.

It remains to prove that m∉Ass⁡(S/I(G)t)\mathfrak{m}\notin\operatorname{Ass}(S/I(G)^t). Let UU be a dominating set of GG such that G[U]G[U] is strongly non-bipartite. We claim that V(B)⊆UV(B)\subseteq U. Indeed, if v∈V(B)∖Uv\in V(B)\setminus U, its attached leaf must belong to UU in order to be dominated. That leaf would then be an isolated vertex of G[U]G[U], a contradiction.

Thus G[U]G[U] is obtained from BB by adding some of its leaves. By [11 Lemma 3.9], μ∗(G[U])≥μ∗(B)\mu^*(G[U])\geq\mu^*(B). The two edges joining zz to the cycles are bridges. By Lemma 3.7, every generalized ear decomposition of BB has at least two even walks. Conversely, begin with one odd cycle, add the two bridge edges as closed walks of length two, and then add the other odd cycle. This gives an odd-beginning generalized ear decomposition with exactly two even walks. Therefore φ∗(B)=2\varphi^*(B)=2, and

μ∗(B)=2+(2t−1)−12=t.\mu^*(B)=\frac{2+(2t-1)-1}{2}=t.

Consequently, μ∗(G[U])≥t\mu^*(G[U])\geq t for every such dominating set UU. By [11 Theorem 3.6], m∉Ass⁡(S/I(G)t)\mathfrak{m}\notin\operatorname{Ass}(S/I(G)^t). □

We now compare the first powers for which the maximal ideal becomes an associated prime. Set

ιord(G)=min⁡{t≥1∣m∈Ass⁡(S/I(G)t)}andιint(G)=min⁡{t≥1∣m∈Ass⁡(S/I(G)t‾)}.\iota_{\mathrm{ord}}(G)=\min\{t\geq1\mid\mathfrak{m}\in\operatorname{Ass}(S/I(G)^t)\} \quad\text{and}\quad \iota_{\mathrm{int}}(G)=\min\{t\geq1\mid\mathfrak{m}\in\operatorname{Ass}(S/\overline{I(G)^t})\}.

For an integer L≥1L\geq1, let GLG_L be the graph obtained from the path

v0−v1−⋯−vLv_0-v_1-\cdots-v_L

by completing v0v_0 and vLv_L to two triangles. Thus we add vertices a0,b0,aL,bLa_0,b_0,a_L,b_L and the six edges

v0a0, v0b0, a0b0, vLaL, vLbL, aLbL.v_0a_0,\ v_0b_0,\ a_0b_0,\ v_La_L,\ v_Lb_L,\ a_Lb_L.

The graph has L+5L+5 vertices.

Figure 1. The graph GLG_L: two triangles joined by a path of length LL.

We first verify the strict Hall condition on GLG_L.

Lemma 3.9. For every L≥1L \ge1 and every nonempty independent set U⊆V(GL)U \subseteq V(G_L),

∣NGL(U)∣>∣U∣.|N_{G_L}(U)| > |U|.

Moreover, if v0∈Uv_0 \in U, then ∣NGL(U)∣≥∣U∣+2|N_{G_L}(U)| \ge|U| + 2.

Proof. Let P=v0−v1−⋯−vLP = v_0-v_1-\cdots-v_L, and put A=U∩V(P)A = U \cap V(P). An independent set A={vi1,…,viq}A = \{v_{i_1},\ldots,v_{i_q}\} of PP, with i1<⋯<iqi_1 < \cdots< i_q, has at least q−1q-1 neighbors in PP, by choosing one vertex strictly between each two consecutive vertices of AA. If AA avoids both endpoints, it has at least q+1q+1 neighbors in PP. If AA contains exactly one endpoint, it has at least qq neighbors in PP.

Let ee be the number of vertices of UU among {a0,b0,aL,bL}\{a_0,b_0,a_L,b_L\} and let r=∣A∩{v0,vL}∣r = |A \cap\{v_0,v_L\}|. Since UU is independent, at most one of a0,b0a_0,b_0 and at most one of aL,bLa_L,b_L belongs to UU, and none of these outer vertices belongs to UU when the corresponding path endpoint is in UU.

If A≠∅A \ne\varnothing and r=0r=0, the selected outer vertices contribute ee distinct neighbors outside PP, so

∣NGL(U)∣≥∣NP(A)∣+e≥∣A∣+e+1=∣U∣+1.|N_{G_L}(U)| \ge|N_P(A)| + e \ge|A| + e + 1 = |U| + 1.

If r≥1r \ge1, each selected path endpoint contributes two neighbors outside PP, and each selected outer vertex contributes its partner in its triangle. Thus

∣NGL(U)∣≥∣NP(A)∣+2r+e≥∣A∣−1+2r+e≥∣U∣+1.|N_{G_L}(U)| \ge|N_P(A)| + 2r + e \ge|A| - 1 + 2r + e \ge|U| + 1.

If r=1r=1, the stronger bound ∣NP(A)∣≥∣A∣|N_P(A)| \ge|A| gives ∣NGL(U)∣≥∣U∣+2|N_{G_L}(U)| \ge|U| + 2. If r=2r=2, the preceding bound already gives ∣NGL(U)∣≥∣U∣+3|N_{G_L}(U)| \ge|U| + 3. This proves the additional assertion when v0∈Uv_0 \in U. Finally, if A=∅A = \varnothing, the selected outer vertices have disjoint neighborhoods of size two, giving ∣NGL(U)∣≥2∣U∣>∣U∣|N_{G_L}(U)| \ge2|U| > |U|. □

Lemma 3.10. For every L≥3L \ge3, φ∗(GL)=L\varphi^{*}(G_L) = L.

Proof. Every edge of the connecting path is a bridge, so Lemma 3.7 shows that an odd-beginning generalized ear decomposition of GLG_L has at least LL even walks. Conversely, begin with the triangle v0,a0,b0,v0v_0,a_0,b_0,v_0, add the closed walks vi−1,vi,vi−1v_{i-1},v_i,v_{i-1} for i=1,…,Li=1,\ldots,L, and finish with the triangle vL,aL,bL,vLv_L,a_L,b_L,v_L. This is an odd-beginning generalized ear decomposition with exactly LL even walks. □

Lemma 3.11. For every L≥3L \ge3, ιord⁡(GL)=L\iota_{\operatorname{ord}}(G_L) = L.

Proof. By [11 Theorem 3.6], it suffices to prove that

min⁡{μ∗(GL[U])∣U⊆V(GL), U is dominating, GL[U] is strongly non-bipartite}=L−1.\min\{\mu^{*}(G_L[U]) \mid U \subseteq V(G_L),\ U\text{ is dominating},\ G_L[U]\text{ is strongly non-bipartite}\} = L-1.

Let UU be such a dominating set and put H=GL[U]H = G_L[U]. Since the only odd cycles of GLG_L are its two endpoint triangles, HH has at most two connected components. By repeated application of [11 Lemma 3.9], a triangle with an attached path of length pp has

μ∗=p+1.(1)\mu^{*} = p+1. \tag*{(1)}

If HH has two connected components, both endpoint triangles are contained in UU. The vertices of UU on the connecting path form a prefix v0,…,viv_0,\ldots,v_i and a suffix vj,…,vLv_j,\ldots,v_L. The omitted middle block has length r=j−i−1r=j-i-1, where 1≤r≤21 \le r \le2 by domination. Hence

μ∗(H)=(i+1)+(L−j+1)=L−r+1≥L−1.\mu^{*}(H) = (i+1) + (L-j+1) = L-r+1 \ge L-1.

The value L−1L-1 is attained by omitting two consecutive internal vertices of the connecting path.

If HH is connected and contains only one endpoint triangle, it must contain the entire connecting path and possibly one further vertex of the opposite endpoint triangle. It is therefore a triangle with a path of length LL or L+1L+1 attached, so μ∗(H)≥L+1\mu^*(H) \ge L+1. Finally, if HH is connected and contains both endpoint triangles, then H=GLH=G_L, and Lemma 3.10 gives

μ∗(H)=L+(L+5)−12=L+2.\mu^*(H)=\frac{L+(L+5)-1}{2}=L+2.

Taking the minimum gives the claimed formula. □

The integral closure first appearance index can also be determined exactly.

Lemma 3.12. For every integer L≥3L \ge3,

ιint(GL)=⌈L+62⌉.\iota_{\mathrm{int}}(G_L)=\left\lceil\frac{L+6}{2}\right\rceil.

Proof. First suppose that LL is even. By Lemmas 3.9 and 3.4, the product of all L+5L+5 variables is a witness for the maximal ideal of I(GL)(L+6)/2I(G_L)^{(L+6)/2}. If LL is odd, take the same product multiplied by xv0x_{v_0}. The additional assertion of Lemma 3.9 shows that its exponent vector satisfies the inequalities of Lemma 3.3, and its total degree is L+6L+6. It is therefore a witness for I(GL)(L+7)/2‾\overline{I(G_L)^{(L+7)/2}}. In both cases we obtain

ιint(GL)≤⌈L+62⌉.\iota_{\mathrm{int}}(G_L)\le\left\lceil\frac{L+6}{2}\right\rceil.

To prove the reverse inequality, let f=xaf=x^a be a monomial such that (I(GL)s‾:f)=m(\overline{I(G_L)^s}:f)=\mathfrak m, and set U=supp⁡(f)U=\operatorname{supp}(f). By Lemma 3.3, UU is a dominating set. Moreover, each connected component of GL[U]G_L[U] is non-bipartite. Indeed, if such a component had bipartition A∪BA\cup B, both parts would be independent, and all neighbors of AA and BB outside this component would have exponent zero. The inequalities in Lemma 3.3 would therefore give a(B)≥a(A)+1a(B)\ge a(A)+1 and a(A)≥a(B)+1a(A)\ge a(B)+1, a contradiction. The same argument with a singleton rules out an isolated vertex.

We need a bound for the weight of a component WW of GL[U]G_L[U] consisting of a triangle with an attached path u0,u1,…,upu_0,u_1,\ldots,u_p, where u0u_0 is a vertex of the triangle. We claim that

a(W)≥2p+3.a(W)\ge2p+3.

This is clear if p=0p=0. Suppose p≥1p\ge1. For 0≤i≤p−10\le i\le p-1, the set {ui+1,ui+3,…}∩{u0,…,up}\{u_{i+1},u_{i+3},\ldots\}\cap\{u_0,\ldots,u_p\} is independent. Its neighbors of positive exponent are precisely the alternate vertices ui,ui+2,…u_i,u_{i+2},\ldots. Thus Lemma 3.3 gives

bi:=aui−aui+1+⋯+(−1)p−iaup≥1(0≤i≤p−1).b_i:=a_{u_i}-a_{u_{i+1}}+\cdots+(-1)^{p-i}a_{u_p}\ge1\qquad(0\le i\le p-1).

Since bp=aup≥1b_p=a_{u_p}\ge1, it follows that aui=bi+bi+1≥2a_{u_i}=b_i+b_{i+1}\ge2 for 0≤i<p0\le i<p. The other two vertices of the triangle have positive exponents. This proves a(W)≥2p+3a(W)\ge2p+3.

Since GLG_L has only the two endpoint triangles, the strongly non-bipartite graph GL[U]G_L[U] has at most two connected components. If it is connected and contains both triangles, then U=V(GL)U=V(G_L), so deg⁡(f)≥L+5\deg(f)\ge L+5. If it is connected and contains only one triangle, it is a triangle with an attached path of length p=Lp=L or L+1L+1. Hence deg⁡(f)≥2L+3≥L+5\deg(f)\ge2L+3\ge L+5.

Finally, suppose that GL[U]G_L[U] has two components. Both endpoint triangles then belong to UU, and the vertices omitted from the connecting path form a block of length r≤2r\le2 by domination. If the two attached paths have lengths pp and qq, then p+q=L−rp+q=L-r. Applying the preceding estimate to the two components yields

deg⁡(f)≥(2p+3)+(2q+3)=2L−2r+6≥2L+2≥L+5,\deg(f)\ge(2p+3)+(2q+3)=2L-2r+6\ge2L+2\ge L+5,

where the last inequality holds for L≥3L\ge3.

Thus deg⁡(f)≥L+5\deg(f)\ge L+5 in all cases. Lemma 3.2 gives 2s−1=deg⁡(f)2s-1=\deg(f), so s≥⌈(L+6)/2⌉s\ge\left\lceil(L+6)/2\right\rceil. This proves the formula. □

Theorem 3.13. For every integer r≥1r \ge1, there exists a graph GG such that

ιord(G)−ιint(G)=r.\iota_{\mathrm{ord}}(G)-\iota_{\mathrm{int}}(G)=r.

Proof. Take L=2r+6L=2r+6 and let G=GLG=G_L. By Lemma 3.11, ιord(G)=L\iota_{\mathrm{ord}}(G)=L, while Lemma 3.12 gives ιint(G)=(L+6)/2\iota_{\mathrm{int}}(G)=(L+6)/2. Hence

ιord(G)−ιint(G)=L−L+62=L2−3=r\iota_{\mathrm{ord}}(G)-\iota_{\mathrm{int}}(G)=L-\frac{L+6}{2}=\frac{L}{2}-3=r

as claimed. □

Remark 3.14. Martínez-Bernal, Morey, and Villarreal showed that the ordinary and integral closure associated prime filtrations have the same eventual stable set, while their Example 3.9 shows that the two stability indices may differ [12]. Theorem 3.13 gives a stronger first appearance phenomenon: the maximal ideal may occur in an integral closure arbitrarily earlier than in the corresponding ordinary powers.

The reverse difference can also be made arbitrarily large. We first compute both indices for disjoint unions of triangles.

Lemma 3.15. Let H=qK3H=qK_3 be the disjoint union of q≥1q \ge1 triangles. Then

ιord(H)=q+1andιint(H)={(3q+1)/2if q is odd,(3q+4)/2if q is even.\iota_{\mathrm{ord}}(H)=q+1 \quad\text{and}\quad\iota_{\mathrm{int}}(H)= \begin{cases} (3q+1)/2 & \text{if }q\text{ is odd},\\ (3q+4)/2 & \text{if }q\text{ is even}. \end{cases}

Proof. Let FF be the product of all variables of HH. Each triangle contributes one edge factor to FF, whereas multiplying FF by any variable allows two edge factors in the corresponding triangle. Thus (I(H)q+1:F)=m(I(H)^{q+1}:F)=\mathfrak{m} and ιord(H)≤q+1\iota_{\mathrm{ord}}(H)\le q+1.

Conversely, suppose that (I(H)s:xa)=m(I(H)^s:x^{\mathbf a})=\mathfrak{m}. The matching capacity of a\mathbf a is s−1s-1, and it is the sum of the matching capacities on the qq triangles. Each triangle must have positive matching capacity: if its support contained at most one vertex, multiplying by a variable already in that support (or by any variable if the support is empty) would not increase its matching capacity, contradicting the colon condition. Hence s−1≥qs-1\ge q, and ιord(H)=q+1\iota_{\mathrm{ord}}(H)=q+1.

Now suppose that (I(H)‾s:xa)=m(\overline{I(H)}^s:x^{\mathbf a})=\mathfrak{m}. For a triangle with vertices u,v,wu,v,w, Lemma 3.3 gives

av+aw≥au+1,au+aw≥av+1,au+av≥aw+1.a_v+a_w\ge a_u+1,\qquad a_u+a_w\ge a_v+1,\qquad a_u+a_v\ge a_w+1.

These inequalities force all three exponents to be positive. Indeed, if au=0a_u=0, the last two inequalities would give simultaneously aw≥av+1a_w\ge a_v+1 and av≥aw+1a_v\ge a_w+1. Thus deg⁡(xa)≥3q\deg(x^{\mathbf a})\ge3q. By Lemma 3.2, this degree is odd. When qq is odd, it follows that 2s−1≥3q2s-1\ge3q. When qq is even, no triangle can have total weight four: up to permutation, the only positive integer triple with sum four is (2,1,1)(2,1,1), which violates one of the displayed inequalities. Consequently, the smallest possible odd total degree is 3q+33q+3.

These lower bounds are attained. If qq is odd, take all exponents equal to one. If qq is even, take exponents (2,2,2)(2,2,2) on one triangle and (1,1,1)(1,1,1) on all the others. In each case the inequalities of Lemma 3.3 hold on every triangle, hence on their disjoint union. The stated values of ιint(H)\iota_{\mathrm{int}}(H) follow. □

Theorem 3.16. For every integer r≥1r \ge1, there exists a graph GG such that

ιint(G)−ιord(G)=r.\iota_{\mathrm{int}}(G)-\iota_{\mathrm{ord}}(G)=r.

Proof. Take G=(2r+1)K3G=(2r+1)K_3. By Lemma 3.15,

ιint(G)=3r+2andιord(G)=2r+2.\iota_{\mathrm{int}}(G)=3r+2 \quad\text{and}\quad\iota_{\mathrm{ord}}(G)=2r+2.

Their difference is rr, as required. □

In particular, neither of the two associated prime containments holds for all graphs when t≥4t \ge4 is fixed. Indeed, Theorem 3.8 supplies one direction. For the other, take G=(t−1)K3G = (t - 1)K_{3}. Then ιord⁡(G)=t\iota_{\operatorname{ord}}(G) = t, while Lemma 3.15 gives ιint⁡(G)>t\iota_{\operatorname{int}}(G) > t. Thus m∈Ass⁡(S/I(G)t)\mathfrak{m} \in\operatorname{Ass}(S/I(G)^{t}) but m∉Ass⁡(S/I(G)t‾)\mathfrak{m} \notin\operatorname{Ass}(S/\overline{I(G)^{t}}).

Comparison of v-numbers

This section is devoted to the study of vv-numbers of powers of an edge ideal and their integral closures. We start with a result for general squarefree monomial ideals.

Lemma 4.1. Let II be a squarefree monomial ideal and let PP be a minimal prime of II. Then, for every t≥1t \ge1,

vP(It‾)≤vP(It).v_{P}(\overline{I^{t}}) \le v_{P}(I^{t}).

Proof. If I=0I = 0, the assertion is immediate. We may therefore assume that I≠0I \ne0. Let ff be a monomial such that (It:f)=P(I^{t} : f) = P. Since II is squarefree and PP is a minimal prime, ISP=PSPIS_{P} = PS_{P}. Localizing the colon equality at PP gives (PtSP:f)=PSP(P^{t}S_{P} : f) = PS_{P}, which implies that deg⁡P(f)=t−1\deg_{P}(f) = t - 1.

Let Q∈Min⁡(S/I)Q \in\operatorname{Min}(S/I) with Q≠PQ \ne P. Choose a variable xi∈P∖Qx_{i} \in P \setminus Q. Since xif∈It⊆Qtx_{i}f \in I^{t} \subseteq Q^{t} and QtQ^{t} is QQ-primary, we have f∈Qtf \in Q^{t}. It follows that

(I(t):f)=⋂Q∈Min⁡(S/I)(Qt:f)=(Pt:f)=P.(I^{(t)} : f) = \bigcap_{Q \in\operatorname{Min}(S/I)}(Q^{t} : f) = (P^{t} : f) = P.

Since It⊆It‾⊆I(t)I^{t} \subseteq\overline{I^{t}} \subseteq I^{(t)}, we obtain

P=(It:f)⊆(It‾:f)⊆(I(t):f)=P.P = (I^{t} : f) \subseteq(\overline{I^{t}} : f) \subseteq(I^{(t)} : f) = P.

Thus (It‾:f)=P(\overline{I^{t}} : f) = P, proving the assertion. □

When I=I(G)I = I(G) is the edge ideal of a graph, we have the following stronger result.

Theorem 4.2. Let GG be a simple graph and let I=I(G)I = I(G) be its edge ideal. Then, for every t≥1t \ge1, every minimal prime P∈Min⁡(S/I)P \in\operatorname{Min}(S/I), and every monomial f∈Sf \in S, we have

(It:f)=P⟺(It‾:f)=P.(I^{t} : f) = P \quad\Longleftrightarrow\quad(\overline{I^{t}} : f) = P.

In particular,

vP(It‾)=vP(It).v_{P}(\overline{I^{t}}) = v_{P}(I^{t}).

To establish Theorem 4.2, we need the following lemma.

Lemma 4.3. Let I=I(G)I = I(G) be the edge ideal of a simple graph GG, and let CC be a vertex cover of GG. Let u=xau = x^{a} be a monomial of SS such that deg⁡C(u)=t\deg_{C}(u) = t and u∈It‾u \in\overline{I^{t}}. Then u∈Itu \in I^{t}.

Proof. Since u=xa∈It‾u = x^{a} \in\overline{I^{t}}, there exist positive rational numbers c1,…,ckc_{1}, \ldots, c_{k} and minimal generators xb1,…,xbkx^{b_{1}}, \ldots, x^{b_{k}} of ItI^{t} such that ∑i=1kci=1\sum_{i=1}^{k}c_{i}=1 and a≥∑i=1kcibia \ge\sum_{i=1}^{k}c_{i}b_{i}. Because deg⁡C(u)=t\deg_{C}(u) = t and deg⁡C(xbi)≥t\deg_{C}(x^{b_{i}}) \ge t for every ii, it follows that deg⁡C(xbi)=t\deg_{C}(x^{b_{i}}) = t for all ii.

Thus each xbix^{b_{i}} is a product of tt edges having exactly one endpoint in CC. Let JJ be the edge ideal of the bipartite graph consisting of the edges of GG between CC and V(G)∖CV(G) \setminus C. Then xbi∈Jtx^{b_{i}} \in J^{t} for all ii, and hence u∈Jt‾u \in\overline{J^{t}}. Since JJ is the edge ideal of a bipartite graph, it is normal (see [15]), i.e., Jt‾=Jt\overline{J^{t}} = J^{t}. Therefore u∈Jt⊆Itu \in J^{t} \subseteq I^{t}.

Proof of Theorem 4.2. If I=0I = 0, the assertion is immediate. We may therefore assume that I≠0I \ne0. Let CC be the minimal vertex cover corresponding to PP. We first observe that if ff is a monomial such that (J:f)=P(J:f)=P, where JJ is either ItI^t or It‾\overline{I^t}, then

deg⁡C(f)=t−1.\deg_C(f)=t-1.

Indeed, J⊆PtJ \subseteq P^t, as PtP^t is integrally closed. Since xif∈Jx_i f \in J for every xi∈Px_i \in P, we have deg⁡C(f)≥t−1\deg_C(f) \ge t-1. Suppose that deg⁡C(f)≥t\deg_C(f) \ge t. Choose tt variables, counted with multiplicity, from CC whose product divides ff. Because CC is a minimal vertex cover, each of these variables has a neighbor outside CC. Let gg be the product of these neighbors, counted with multiplicity. Then g∉Pg \notin P and gf∈It⊆Jgf \in I^t \subseteq J, contradicting (J:f)=P(J:f)=P. This proves the observation.

Now suppose that (It:f)=P(I^t:f)=P. Since It⊆It‾⊆PtI^t \subseteq\overline{I^t} \subseteq P^t and deg⁡C(f)=t−1\deg_C(f)=t-1, we have

P=(It:f)⊆(It‾:f)⊆(Pt:f)=P.P=(I^t:f)\subseteq(\overline{I^t}:f)\subseteq(P^t:f)=P.

Thus (It‾:f)=P(\overline{I^t}:f)=P.

Conversely, suppose that (It‾:f)=P(\overline{I^t}:f)=P. For each variable xi∈Px_i \in P, we have xif∈It‾x_i f \in\overline{I^t} and deg⁡C(xif)=t\deg_C(x_i f)=t. Lemma 4.3 gives xif∈Itx_i f \in I^t. Consequently,

P⊆(It:f)⊆(It‾:f)=P,P\subseteq(I^t:f)\subseteq(\overline{I^t}:f)=P,

so (It:f)=P(I^t:f)=P. The equality of the local vv-numbers follows.

We now show that the difference v(I(G)t‾)−v(I(G)t)v(\overline{I(G)^t})-v(I(G)^t) can be arbitrarily large for any fixed integer t≥3t \ge3, while the difference v(I(G)t)−v(I(G)t‾)v(I(G)^t)-v(\overline{I(G)^t}) can grow linearly in tt.

Theorem 4.4. For any fixed integer t≥3t \ge3 and any positive integer rr, there exists a graph GG such that

v(I(G)t‾)−v(I(G)t)≥r−1.v(\overline{I(G)^t})-v(I(G)^t)\ge r-1.

Proof. Let HrH_r be the graph consisting of a base triangle abcabc together with rr triangles attached to each of its vertices. Define Gt,rG_{t,r} to be the disjoint union of t−1t-1 copies of HrH_r, and set I=I(Gt,r)I=I(G_{t,r}).

Let ff be the product of all variables corresponding to the t−1t-1 base triangles in Gt,rG_{t,r}. Then deg⁡(f)=3(t−1)\deg(f)=3(t-1) and f∉Itf\notin I^t, since each base triangle contributes at most one edge factor to ff. However, xvf∈Itx_vf\in I^t for every variable xvx_v. Indeed, in the component containing vv, the monomial xvfx_vf admits two edge factors, while the remaining t−2t-2 base triangles each supply one edge factor. Thus (It:f)=m(I^t:f)=\mathfrak{m}, giving

v(It)≤3(t−1).v(I^t)\le3(t-1).

We also have f∈It‾f\in\overline{I^t}. Indeed, assigning weight 1/21/2 to each edge of every base triangle gives a fractional matching with capacities given by the exponent vector of ff and total weight 3(t−1)/2≥t3(t-1)/2\ge t. The assertion follows from Lemma 2.4.

Let g=xbg=x^b be a monomial such that (It‾:g)=P(\overline{I^t}:g)=P is an associated prime of It‾\overline{I^t}. Since g∉It‾g\notin\overline{I^t} while f∈It‾f\in\overline{I^t}, we have f∤gf\nmid g. Thus at least one variable of a base triangle does not divide gg.

Let abcabc be a base triangle and suppose that a∤ga\nmid g. Consider any of the triangles auiviau_iv_i attached to aa. If neither uiu_i nor viv_i divides gg, then the capacities of a,ui,via,u_i,v_i are all zero. Since uiu_i and viv_i have no neighbors outside {a,ui,vi}\{a,u_i,v_i\}, adding either xuix_{u_i} or xvix_{v_i} cannot increase the fractional matching number. By Lemma 2.4, neither xuigx_{u_i}g nor xvigx_{v_i}g belongs to It‾\overline{I^t}. Thus xui,xvi∉Px_{u_i},x_{v_i}\notin P, contradicting xuixvi∈I⊆Px_{u_i}x_{v_i}\in I\subseteq P. Therefore at least one of ui,viu_i,v_i divides gg.

Let k≥1k\ge1 be the number of variables of the base triangles that do not divide gg. Each such missing variable forces at least one variable from each of its rr attached triangles to divide gg. These attached triangles have distinct vertices outside the base triangles. Consequently,

deg⁡(g)≥3(t−1)−k+kr=3(t−1)+k(r−1)≥3(t−1)+r−1.\deg(g)\ge3(t-1)-k+kr=3(t-1)+k(r-1)\ge3(t-1)+r-1.

It follows that v(It‾)≥3(t−1)+r−1v(\overline{I^t}) \ge3(t-1)+r-1, and hence

v(It‾)−v(It)≥r−1v(\overline{I^t})-v(I^t)\ge r-1

as desired. □\square

Theorem 4.5. For any fixed integer t≥4t\ge4 and any positive integer rr, there exists a graph GG such that

v(I(G)t)−v(I(G)t‾)≥r.v(I(G)^t)-v(\overline{I(G)^t})\ge r.

Proof. For ease of reading, we divide the proof into several steps.

Step 1: Graph Construction (Gk)(G_k).

  • The Core Graph (G)(G): The graph GG consists of a triangle with vertices {a,b,c}\{a,b,c\}, a (2t−5)(2t-5)-cycle with vertices {d1,d2,…,d2t−5}\{d_1,d_2,\ldots,d_{2t-5}\}, and an additional vertex zz adjacent to both aa and d1d_1. Let Vcore={a,b,c,d1,…,d2t−5,z}V_{\mathrm{core}}=\{a,b,c,d_1,\ldots,d_{2t-5},z\} denote the set of these core vertices.

  • The Attached Paths (Gk)(G_k): For each vertex x∈Vcorex\in V_{\mathrm{core}}, we attach kk paths of length 2. That is, for each index i∈{1,…,k}i\in\{1,\ldots,k\}, we introduce two new vertices, xi1x_{i1} and xi2x_{i2}, along with the edges {x,xi1}\{x,x_{i1}\} and {xi1,xi2}\{x_{i1},x_{i2}\}.

Step 2: The vv-number of the Integral Closure. By [14 Theorem 2.10], we have I(Gk)t‾=I(Gk)t+(abcd1⋯d2t−5)I(G)\overline{I(G_k)^t}=I(G_k)^t+(abcd_1\cdots d_{2t-5})I(G). It then follows that

I(Gk)t‾:(abcd1⋯d2t−5z)=(x,xi1∣x∈Vcore, i∈[k]).\overline{I(G_k)^t}:(abcd_1\cdots d_{2t-5}z)=(x,x_{i1}\mid x\in V_{\mathrm{core}},\ i\in[k]).

Hence, v(I(Gk)t‾)≤2t−1v(\overline{I(G_k)^t})\le2t-1. Moreover, as v(I(Gk)t‾)≥2t−1v(\overline{I(G_k)^t})\ge2t-1, we conclude that

v(I(Gk)t‾)=2t−1.v(\overline{I(G_k)^t})=2t-1.

Step 3: Lower Bound Setup via the Set Ω\Omega. To establish a lower bound for v(I(Gk)t)v(I(G_k)^t), let FF be a monomial such that v(I(Gk)t)=deg⁡Fv(I(G_k)^t)=\deg F and (I(Gk)t:F)(I(G_k)^t:F) is a prime ideal. We define the subset of core vertices

Ω={x∈Vcore∣x∤F}.\Omega=\{x\in V_{\mathrm{core}}\mid x\nmid F\}.

For any vertex x∈Ωx\in\Omega and for each i∈[k]i\in[k], at least one of xi1x_{i1} or xi2x_{i2} must divide FF. Indeed, if neither xi1x_{i1} nor xi2x_{i2} divides FF, then xi1,xi2∉(I(Gk)t:F)x_{i1},x_{i2}\notin(I(G_k)^t:F), while xi1xi2∈(I(Gk)t:F)x_{i1}x_{i2}\in(I(G_k)^t:F), leading to a contradiction since (I(Gk)t:F)(I(G_k)^t:F) is a prime ideal.

Since each vertex in Ω\Omega forces at least kk distinct variables from the attached paths to divide FF, while the remaining 2t−1−∣Ω∣2t-1-|\Omega| core vertices also divide FF, we obtain the degree bound

deg⁡F≥k∣Ω∣+(2t−1−∣Ω∣)=(k−1)∣Ω∣+2t−1.\deg F\ge k|\Omega|+(2t-1-|\Omega|)=(k-1)|\Omega|+2t-1.

Hence, if Ω≠∅\Omega\ne\varnothing, then ∣Ω∣≥1|\Omega|\ge1, which yields

v(I(Gk)t)≥(k−1)∣Ω∣+2t−1≥k−1+2t−1.v(I(G_k)^t)\ge(k-1)|\Omega|+2t-1\ge k-1+2t-1.

Step 4: The Case Where Ω=∅\Omega=\varnothing. Suppose Ω=∅\Omega=\varnothing. Then every core vertex x∈Vcorex\in V_{\mathrm{core}} divides FF, so the core monomial

f=abcd1⋯d2t−5zf=abcd_1\cdots d_{2t-5}z

divides FF. Observe that fx∈I(Gk)tfx\in I(G_k)^t for all x∈Vcorex\in V_{\mathrm{core}}, as well as for x=xi1x=x_{i1} with x∈Vcore∖{z}x\in V_{\mathrm{core}}\setminus\{z\} and i∈[k]i\in[k]. Thus, the remaining variables in F/fF/f can only come from {xi2∣x∈Vcore, i∈[k]}\{x_{i2}\mid x\in V_{\mathrm{core}},\ i\in[k]\} and {zi1∣i∈[k]}\{z_{i1}\mid i\in[k]\}.

Moreover, for each i∈[k]i\in[k], at least one of zi1z_{i1} or zi2z_{i2} must divide FF. Indeed, if neither divides FF, then zi1,zi2∉(I(Gk)t:F)z_{i1},z_{i2}\notin(I(G_k)^t:F) while zi1zi2∈(I(Gk)t:F)z_{i1}z_{i2}\in(I(G_k)^t:F), a contradiction. Hence,

v(I(Gk)t)=deg⁡F≥2t−1+k.v(I(G_k)^t)=\deg F\ge2t-1+k.

Combining both cases, we deduce that v(I(Gk)t)≥2t−1+k−1v(I(G_k)^t) \ge2t-1+k-1. Subtracting v(I(Gk)t‾)=2t−1v(\overline{I(G_k)^t})=2t-1 gives

v(I(Gk)t)−v(I(Gk)t‾)≥k−1.v(I(G_k)^t)-v(\overline{I(G_k)^t}) \ge k-1.

By setting k=r+1k=r+1, we achieve the desired bound of rr. □

References

Department of Mathematics, University of Manitoba, 420 Machray Hall, 186 Dysart Road, Winnipeg, MB R3T 2N2 Canada

Email address: [email protected]

Tulane University, Mathematics Department, 6823 St. Charles Avenue, New Orleans, LA 70118, USA

Email address: [email protected]

Institute of Mathematics, VAST, 18 Hoang Quoc Viet, Hanoi, Vietnam

Email address: [email protected]

References

  1. [1]P. Biswas, M. Mandal, and P. Phukan, A comparison of the v-number of a monomial ideal and its integral closure, Preprint (2026), arXiv:2609.05044.arxiv.org/abs/2609.05044
  2. [3]A. Conca, A note on the v-invariant, Proc. Amer. Math. Soc. 152 (2024), 2349–2351.arxiv.org/abs/2401.00022
  3. [4]S. M. Cooper, A. Seceleanu, S. O. Tohaneanu, M. Vaz Pinto and R. H. Villarreal, Generalized minimum distance functions and algebraic invariants of Geramita ideals, Adv. in Appl. Math. 112 (2020), 101940.arxiv.org/abs/1812.06529
  4. [5]L. X. Dung, T. T. Hien, H. D. Nguyen, and T. N. Trung, Regularity and Koszul property of symbolic powers of monomial ideals, Math. Z. 298 (2021), 1487–1522.DOI
  5. [7]A. Ficarra and E. Sgroi, Asymptotic behaviour of the v-number of homogeneous ideals, J. Algebra 704 (2026), 273–297.DOI
  6. [8]A. Ficarra and E. Sgroi, Asymptotic behaviour of integer programming and the v-function of a graded filtration, J. J. Algebra Appl. (published online 2025), article 2650236, doi:10.1142/S0219498826502361.DOI
  7. [9]H. T. Hà and N. V. Trung, Membership Criteria and Containments of Powers of Monomial Ideals, Acta Math Vietnam 44 (2019), 117–139.DOI
  8. [10]D. Jaramillo and R. H. Villarreal, The v-number of edge ideals, J. Combin. Theory Ser. A 177 (2021), 105310.arxiv.org/abs/2001.06861
  9. [11]H. M. Lam and N. V. Trung, Associated primes of powers of edge ideals and ear decompositions of graphs, Trans. Amer. Math. Soc. 372 (2019), 3211–3236.DOI
  10. [12]J. Martínez-Bernal, S. Morey, and R. H. Villarreal, Associated primes of powers of edge ideals, Collect. Math. 63 (2012), 361–374.
  11. [13]D. H. Mau and T. N. Trung, Stability of associated primes and depth of integral closures of powers of edge ideals, Preprint (2021), arXiv:2108.01830.arxiv.org/abs/2108.01830
  12. [14]N. C. Minh and T. Vu, Integral closure of powers of edge ideals and their regularity, Journal of Algebra 609 (2022), 120–144.DOI
  13. [15]A. Simis, W. V. Vasconcelos, and R. H. Villarreal, On the ideal theory of graphs, J. Algebra 167 (1994), 389–416.DOI
  14. [16]T. N. Trung, Stability of associated primes of integral closures of monomial ideals, J. Combin. Theory Ser. A 116 (2009), 44–54.DOI
  15. [17]Vanmathi A and P. Sarkar, v-numbers of integral closure filtrations of monomial ideals, Preprint (2025), arXiv:2506.09051.arxiv.org/abs/2506.09051

Paper details

vlnc:W_uypt8nqd
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

GPT-6.1 Sol (bare)

View statement
Looking for related papers…
Contents