This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Powers of the edge ideals and matchings in hypergraphs

Fahimeh Khosh-Ahang Ghasr Department of Mathematics, School of Science, Ilam University, P.O.Box 69315-516, Ilam, Iran. f.khoshahang@ilam.ac.ir and fahime-khosh@yahoo.com
Abstract.

In this work, some combinatorial lower bound for regularity of powers of the edge ideal of a uniform hypergarph is gained. A family of hypergraphs whose regularity of edge ideal attains this bound and has a significant difference from the lower bounds heretofore obtained have also been introduced.

Key words and phrases:
edge ideal of hypergraph, graded Betti numbers, matching numbers, power of a monomial ideal, regularity.
2010 Mathematics Subject Classification:
Primary 13F20, 05E40; Secondary 05C65.

1. Introduction

In what follows t,dt,d\in\mathbb{N}, d2d\geq 2, R=K[x1,,xn]R=K[x_{1},\dots,x_{n}] is the polynomial ring over the field KK and S1,,SmS_{1},\dots,S_{m} are monomials in RR. For our convenience, sometimes we use the same notation SkS_{k} for both of the monomial SkS_{k} and {xi:xiSk}\{x_{i}\ :\ x_{i}\mid S_{k}\}. By identifying the vertex xix_{i} and the variable xix_{i}, \mathcal{H} stands for a simple hypergraph with V()={x1,,xn}V(\mathcal{H})=\{x_{1},\dots,x_{n}\} and ()={S1,,Sm}\mathcal{E}(\mathcal{H})=\{S_{1},\dots,S_{m}\}. Recall a hypergraph is dd-uniform if all of its edges have the same cardinality dd and

I=Sk: 1km,I=\langle S_{k}\ :\ 1\leq k\leq m\rangle,

is the edge ideal of \mathcal{H}. So there is a correspondence between dd-uniform hypergraphs and square-free monomial ideals in degree dd (see [18]).

For a minimal graded free resolution

jR(j)βi,jjR(j)β1,jRR/I0,\cdots\longrightarrow\oplus_{j}R(-j)^{\beta_{i,j}}\longrightarrow\cdots\longrightarrow\oplus_{j}R(-j)^{\beta_{1,j}}\longrightarrow R\longrightarrow R/I\longrightarrow 0,

of R/IR/I, βi,j(R/I)\beta_{i,j}(R/I) is called the (i,j)(i,j)th graded Betti number of R/IR/I. Also, recall that the regularity of R/IR/I is defined as

reg(R/I)=max{ji:βi,j(R/I)0}.\mathrm{reg}(R/I)=\max\{j-i\ :\ \beta_{i,j}(R/I)\neq 0\}.

In the last few decades, studying the regularity of square-free monomial ideals because of their connections to other fields such as combinatorics, algebraic topology and computational algebra, has been extensively expanded. In particular, many researchers in combinatorial commutative algebra are interested in computing or bounding the regularity of the edge ideal of graphs and hypergraphs and their powers (see [1] and [9] for some surveys in this context).

In this paper, we firstly investigate free resolutions of powers of a square-free monomial ideal supported on a simplicial complex. Then we use the gained results in Section 2 for a simplicial complex introduced in [8] for powers of the edge ideal of a hypergraph. These yield to characterize or bound the graded Betti numbers and regularity of R/ItR/I^{t} in Section 3. For instance in Corollary 3.5 a combinatorial characterization is gained for vanishing of βi,2di(R/I2)\beta_{i,2di}(R/I^{2}) in terms of the number of matchings of size ii in a dd-uniform hypergraph \mathcal{H}. The needed definitions of some kinds of matchings are presented in Definition 3.1. Moreover the main result of this note is the following result:

Theorem A.
  1. (1)

    Suppose that \mathcal{H} is a hypergraph which has a self semi-induced matching, say 𝒮={S1,,Si}\mathcal{S}=\{S_{1},\dots,S_{i}\}, of type (i,j)(i,j). Then for each \ell with 1i1\leq\ell\leq i we have βi,|S|(t1)+j(R/It)0\beta_{i,|S_{\ell}|(t-1)+j}(R/I^{t})\neq 0 and so

    reg(R/It)|S|(t1)+(|1kiSk|i).\mathrm{reg}(R/I^{t})\geq|S_{\ell}|(t-1)+(|\bigcup_{1\leq k\leq i}S_{k}|-i).
  2. (2)

    Let ss be the number of all self semi-induced matchings in \mathcal{H} of type (i,j)(i,j). Then

    βi,j(R/I)s,\beta_{i,j}(R/I)\geq s,

    and if moreover \mathcal{H} is dd-uniform, then for all integers t>1t>1

    βi,d(t1)+j(R/It)si.\beta_{i,d(t-1)+j}(R/I^{t})\geq si.
  3. (3)

    Suppose that \mathcal{H} is a dd-uniform hypergraph. Then for all tt\in\mathbb{N}

    d(t1)+(d1)(i.m)d(t1)+(s.s.i.m)reg(R/It).d(t-1)+(d-1)(i.m)_{\mathcal{H}}\leq d(t-1)+(s.s.i.m)^{\prime}_{\mathcal{H}}\leq\mathrm{reg}(R/I^{t}).

The above result is indeed a generalization of some parts of [10, Theorem 6.5], [12, Lemma 2.2 and Proposition 2.5], [16, Theorems 3.5(1), 3.7(1) and Corollary 3.6] for arbitrary powers of the edge ideal of a hypergraph. Also, it gives a lower bound better than [3, Theorem 3.7(1) and Corollary 3.8(1)] and [17, Corollary 3.9] for the regularity of powers of the edge ideals of uniform hypergraphs, since there are examples of dd-uniform hypergraphs achieving our bound such that (s.s.i.m)(s.s.i.m)^{\prime}_{\mathcal{H}} is reasonably larger that (d1)(i.m)(d-1)(i.m)_{\mathcal{H}} as presented in Example 3.8.

2. Preliminaries

In this section, for each 1km1\leq k\leq m,

Sk=1inxiai,k,ak=(a1,k,,an,k){0,1}n,1inai,k=d.S_{k}=\prod_{1\leq i\leq n}x_{i}^{a_{i,k}},\textbf{a}_{k}=(a_{1,k},\dots,a_{n,k})\in\{0,1\}^{n},\sum_{1\leq i\leq n}a_{i,k}=d.

Hence II is a square-free monomial ideal of RR generated in degree dd. Now assume that

m,t={b:b=b1e1++bmem=(b1,,bm){0,,t}m,1mb=t}.\mathcal{B}_{m,t}=\{\textbf{b}\ :\ \textbf{b}=b_{1}\textbf{e}_{1}+\dots+b_{m}\textbf{e}_{m}=(b_{1},\dots,b_{m})\in\{0,\dots,t\}^{m},\sum_{1\leq\ell\leq m}b_{\ell}=t\}.

Hereafter we order the elements of m,t\mathcal{B}_{m,t} in such a way that if b=(b1,,bm)\textbf{b}=(b_{1},\dots,b_{m}) and b=(b1,,bm)\textbf{b}^{\prime}=(b^{\prime}_{1},\dots,b^{\prime}_{m}) are two elements of m,t\mathcal{B}_{m,t}, then b<b\textbf{b}<\textbf{b}^{\prime} if and only if there exists an integer 1km1\leq k\leq m such that for all 1k11\leq\ell\leq k-1 we have b=bb_{\ell}=b^{\prime}_{\ell} and bk>bkb_{k}>b^{\prime}_{k}. If we set Sb=S1b1SmbmS^{\textbf{b}}=S_{1}^{b_{1}}\dots S_{m}^{b_{m}} for all b=(b1,,bm)m,t\textbf{b}=(b_{1},\dots,b_{m})\in\mathcal{B}_{m,t}, then It=Sb:bm,tI^{t}=\langle S^{\textbf{b}}\ :\ \textbf{b}\in\mathcal{B}_{m,t}\rangle. But note that {Sb:bm,t}\{S^{\textbf{b}}\ :\ \textbf{b}\in\mathcal{B}_{m,t}\} is not necessarily a minimal set of generators of ItI^{t}. So, ItI^{t} is minimally generated by pm,tp_{m,t} elements, when pm,t|m,t|=(t+m1m1)p_{m,t}\leq|\mathcal{B}_{m,t}|=\left(\begin{array}[]{c}t+m-1\\ m-1\end{array}\right). In this section we are going to study some simplicial resolutions of R/ItR/I^{t}. To this aim we first need some notation.

Notation 2.1.

For given II and tt with described notions, suppose that AA is the n×mn\times m matrix with columns a1,,am\textbf{a}_{1},\dots,\textbf{a}_{m}, and BB is the m×pm,tm\times p_{m,t} matrix with columns b1,,bpm,t\textbf{b}_{1},\dots,\textbf{b}_{p_{m,t}}. Also for all 1,,i{1,,pm,t}\ell_{1},\dots,\ell_{i}\in\{1,\dots,p_{m,t}\}, B[1,,i]B[\ell_{1},\dots,\ell_{i}] is the m×im\times i submatrix of BB with columns b1,,bi\textbf{b}_{\ell_{1}},\dots,\textbf{b}_{\ell_{i}} and B[1,,k^,,i]B[\ell_{1},\dots,\widehat{\ell_{k}},\dots,\ell_{i}] is the m×(i1)m\times(i-1) submatrix of BB with columns b1,,bk1,bk+1,,bi\textbf{b}_{\ell_{1}},\dots,\textbf{b}_{\ell_{k-1}},\textbf{b}_{\ell_{k+1}},\dots,\textbf{b}_{\ell_{i}} (of course after ordering 1,,i\ell_{1},\dots,\ell_{i}). Moreover for each real matrix C=(ci,j)p×qC=(c_{i,j})_{p\times q}, we assign the vector

Max(C)=(max{c11,,c1q},,max{cp1,,cpq}).\mathrm{Max}(C)=\left(\max\{c_{11},\dots,c_{1q}\},\dots,\max\{c_{p1},\dots,c_{pq}\}\right).

Also recall that for each real vector c=(c1,,cn)\textbf{c}=(c_{1},\dots,c_{n}), the sum of c is Sum(c)=c1++cn\mathrm{Sum}(\textbf{c})=c_{1}+\dots+c_{n}.

Suppose that Δ\Delta is a simplicial complex whose vertices are labelled by the minimal monomial generators of II and whose faces are labelled by the least common multiple of the vertices of that face. Then we say that Δ\Delta supports a free resolution FF_{\bullet} of R/IR/I or FF_{\bullet} is supported on Δ\Delta if the labelling of the vertices of Δ\Delta satisfies certain properties and the simplicial chain complex of Δ\Delta can be homogenized using the monomial labels on the faces. A well-known simplicial complex which supports a free resolution of R/IR/I is Taylor complex of II which is a simplex on mm vertices labelled by the minimal monomial generators of II and is denoted by Taylor(I)\mathrm{Taylor}(I).

In view of [4], if Δ\Delta supports a free resolution FF_{\bullet} of ItI^{t}, then Δ\Delta is a subcomplex of Taylor(It)\mathrm{Taylor}(I^{t}) and FF_{\bullet} can be structured as follows.

F:0FdimΔ+1Fi+1i+1FiiFi1F0R/It0,F_{\bullet}:0\rightarrow F_{\mathrm{dim}\Delta+1}\rightarrow\dots\rightarrow F_{i+1}\stackrel{{\scriptstyle\partial_{i+1}}}{{\longrightarrow}}F_{i}\stackrel{{\scriptstyle\partial_{i}}}{{\longrightarrow}}F_{i-1}\rightarrow\dots\rightarrow F_{0}\rightarrow R/I^{t}\rightarrow 0,

where F0=RF_{0}=R and FiF_{i} is the free RR-module whose free generators are eτe_{\tau}s for all faces τ\tau of Δ\Delta of dimension i1i-1. If the vertices of τ\tau is labelled by the monomials Sb1,,SbiS^{\textbf{b}_{\ell_{1}}},\dots,S^{\textbf{b}_{\ell_{i}}}, then we sometimes denote eτe_{\tau} by e1,,ie_{\ell_{1},\dots,\ell_{i}}, where 11<<ipm,t1\leq\ell_{1}<\dots<\ell_{i}\leq p_{m,t}. Also for all 1idimΔ+11\leq i\leq\mathrm{dim}\Delta+1 and e1,,iFie_{\ell_{1},\dots,\ell_{i}}\in F_{i},

i(e1,,i)=1ki(1)kμke1,,k^,,i,\partial_{i}(e_{\ell_{1},\dots,\ell_{i}})=\sum_{1\leq k\leq i}(-1)^{k}\mu_{k}e_{\ell_{1},\dots,\widehat{\ell_{k}},\dots,\ell_{i}},

where

μk=lcm(Sb1,,Sbi)lcm(Sb1,,Sbk^,,Sbi).\mu_{k}=\frac{\mathrm{lcm}(S^{\textbf{b}_{\ell_{1}}},\dots,S^{\textbf{b}_{\ell_{i}}})}{\mathrm{lcm}(S^{\textbf{b}_{\ell_{1}}},\dots,\widehat{S^{\textbf{b}_{\ell_{k}}}},\dots,S^{\textbf{b}_{\ell_{i}}})}.

Note that for each 1ji1\leq j\leq i, if bj=(b1,j,,bm,j)\textbf{b}_{\ell_{j}}=(b_{1,\ell_{j}},\dots,b_{m,\ell_{j}}), then we have

Sbj\displaystyle S^{\textbf{b}_{\ell_{j}}} =S1b1,jSmbm,j=(1inxiai,1)b1,j(1inxiai,m)bm,j\displaystyle=S_{1}^{b_{1,\ell_{j}}}\dots S_{m}^{b_{m,\ell_{j}}}=\left(\prod_{1\leq i\leq n}x_{i}^{a_{i,1}}\right)^{b_{1,\ell_{j}}}\dots\left(\prod_{1\leq i\leq n}x_{i}^{a_{i,m}}\right)^{b_{m,\ell_{j}}}
=1inxiai,1b1,j++ai,mbm,j=1inxici,j,\displaystyle=\prod_{1\leq i\leq n}x_{i}^{a_{i,1}b_{1,\ell_{j}}+\dots+a_{i,m}b_{m,\ell_{j}}}=\prod_{1\leq i\leq n}x_{i}^{c_{i,\ell_{j}}},

where AB=(cr,s)n×pm,tAB=(c_{r,s})_{n\times p_{m,t}}. Now assume that Max(AB[1,,i])=(c1,,cn)\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i}])=(c_{1},\dots,c_{n}) and

Max(AB[1,,k^,,i])=(d1,k,,dn,k).\mathrm{Max}(AB[\ell_{1},\dots,\widehat{\ell_{k}},\dots,\ell_{i}])=(d_{1,k},\dots,d_{n,k}).

Then we have μk=1inxicidi,k\mu_{k}=\prod_{1\leq i\leq n}x_{i}^{c_{i}-d_{i,k}}. By considering the degree of e1,,ie_{\ell_{1},\dots,\ell_{i}} as

deg(e1,,i)=deg(lcm(Sb1,,Sbi))=deg(x1c1xncn)=Sum(Max(AB[1,,i])),\mathrm{deg}(e_{\ell_{1},\dots,\ell_{i}})=\mathrm{deg}(\mathrm{lcm}(S^{\textbf{b}_{\ell_{1}}},\dots,S^{\textbf{b}_{\ell_{i}}}))=\mathrm{deg}(x_{1}^{c_{1}}\dots x_{n}^{c_{n}})=\mathrm{Sum}(\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i}])),

FF_{\bullet} is a graded free resolution of R/ItR/I^{t} which is not necessarily minimal as mentioned before, but we may use it for computing the graded Betti numbers βi,j(R/It)\beta_{i,j}(R/I^{t}) as follows.

βi,j(R/It)\displaystyle\beta_{i,j}(R/I^{t}) =dimK(ToriR(R/It,K))j\displaystyle=\mathrm{dim}_{K}(\mathrm{Tor}_{i}^{R}(R/I^{t},K))_{j} (2.1)
=dimK(Hi(FRR/x1,,xn))j\displaystyle=\mathrm{dim}_{K}(H_{i}(F_{\bullet}\otimes_{R}R/\langle x_{1},\dots,x_{n}\rangle))_{j}
=dimK(Ker¯i/Im¯i+1)j.\displaystyle=\mathrm{dim}_{K}(\mathrm{Ker}\overline{\partial}_{i}/\mathrm{Im}\overline{\partial}_{i+1})_{j}.

After tensoring FF_{\bullet} with R/x1,,xnR/\langle x_{1},\dots,x_{n}\rangle, for each eτe_{\tau} where τ={Sb1,,Sbi}Δ\tau=\{S^{\textbf{b}_{\ell_{1}}},\dots,S^{\textbf{b}_{\ell_{i}}}\}\in\Delta, we have

¯i(e1,,i¯)=1ki,Max(AB[1,,i])=Max(AB[1,,k^,,i])(1)ke1,,k^,,i¯,\overline{\partial}_{i}(\overline{e_{\ell_{1},\dots,\ell_{i}}})=\sum_{1\leq k\leq i,\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i}])=\mathrm{Max}(AB[\ell_{1},\dots,\widehat{\ell_{k}},\dots,\ell_{i}])}(-1)^{k}\overline{e_{\ell_{1},\dots,\widehat{\ell_{k}},\dots,\ell_{i}}}, (2.2)

where for each 0idimΔ+10\leq i\leq\mathrm{dim}\Delta+1 and each member uFiu\in F_{i}, u¯\overline{u} is the natural image of uu in Fi¯=FiRR/x1,,xn\overline{F_{i}}=F_{i}\otimes_{R}R/\langle x_{1},\dots,x_{n}\rangle and ¯i=iRidR/x1,,xn\overline{\partial}_{i}=\partial_{i}\otimes_{R}\mathrm{id}_{R/\langle x_{1},\dots,x_{n}\rangle}.

Afterwards suppose that Δ\Delta is a simplicial complex supporting a free resolution of R/ItR/I^{t}.

Remark 2.2.

In view of Equation (2.2), the following statements hold:

  1. (1)

    e1,,i¯Ker¯i\overline{e_{\ell_{1},\dots,\ell_{i}}}\in\mathrm{Ker}\overline{\partial}_{i} if and only if {Sb1,,Sbi}Δ\{S^{\textbf{b}_{\ell_{1}}},\dots,S^{\textbf{b}_{\ell_{i}}}\}\in\Delta and for all 1ki1\leq k\leq i we have

    Max(AB[1,,i])Max(AB[1,,k^,,i]).\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i}])\neq\mathrm{Max}(AB[\ell_{1},\dots,\widehat{\ell_{k}},\dots,\ell_{i}]).
  2. (2)

    If e1,,i¯Im¯i+1\overline{e_{\ell_{1},\dots,\ell_{i}}}\in\mathrm{Im}\overline{\partial}_{i+1}, then there exists a face {Sb1,,Sbi+1}\{S^{\textbf{b}_{\ell_{1}}},\dots,S^{\textbf{b}_{\ell_{i+1}}}\} of Δ\Delta of dimension ii such that Max(AB[1,,i+1])=Max(AB[1,,i])\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i+1}])=\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i}]).

  3. (3)

    If there exists a face {Sb1,,Sbi+1}\{S^{\textbf{b}_{\ell_{1}}},\dots,S^{\textbf{b}_{\ell_{i+1}}}\} of Δ\Delta of dimension ii such that

    Max(AB[1,,i+1])=Max(AB[1,,i]),\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i+1}])=\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i}]),

    and for all 1ki1\leq k\leq i, we have Max(AB[1,,k^,,i+1])Max(AB[1,,i])\mathrm{Max}(AB[\ell_{1},\dots,\widehat{\ell_{k}},\dots,\ell_{i+1}])\neq\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i}]), then e1,,i¯Im¯i+1\overline{e_{\ell_{1},\dots,\ell_{i}}}\in\mathrm{Im}\overline{\partial}_{i+1}.

  4. (4)

    It can be easily seen that m,1={e1,,em}\mathcal{B}_{m,1}=\{\textbf{e}_{1},\dots,\textbf{e}_{m}\} and so B=ImB=I_{m} where t=1t=1. Hence AB=A=(ai,j)n×mAB=A=(a_{i,j})_{n\times m} is a matrix with entries 0 or 11, in which the iith row associates to the variable xix_{i} and the jjth column associates to the generator SjS_{j} of II. In fact ai,j=1a_{i,j}=1 if xiSjx_{i}\mid S_{j} and else ai,j=0a_{i,j}=0. So, in each column there exist exactly dd entries 11. Also for each ii, the iith row has at least one 11, which lies in columns jj with xiSjx_{i}\mid S_{j}. ThereforeMax(AB)=(1,1,,1)\mathrm{Max}(AB)=(1,1,\dots,1) and Max(AB[1,,i])=(c1,,cn)\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i}])=(c_{1},\dots,c_{n}), where cj=1c_{j}=1 if xj1riSrx_{j}\in\bigcup_{1\leq r\leq i}S_{\ell_{r}}, else cj=0c_{j}=0. This shows that for each 1ki1\leq k\leq i,

    Max(AB[1,,i])Max(AB[1,,k^,,i]),\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i}])\neq\mathrm{Max}(AB[\ell_{1},\dots,\widehat{\ell_{k}},\dots,\ell_{i}]),

    if and only if Sk1ri,rkSrS_{\ell_{k}}\nsubseteq\bigcup_{1\leq r\leq i,r\neq k}S_{\ell_{r}}.

The above remarks illustrate that one can evaluate the graded Betti numbers of the edge ideal of a hypergraph by interaction of its edges as you see for instance in [10, Theorem 6.5], [12, Lemma 2.2] and [16, Section 3]. Note that all of these results just use Taylor resolution of ItI^{t} (when t=1t=1) which has easier structure as you may see above.

One can naturally generalize [16, Lemma 3.2] and obtain the following result, which is needed in the next section.

Lemma 2.3.

Let i,ji,j be integers. Set

i,j={e:e=e1,,i¯Ker¯iIm¯i+1,Sum(Max(AB[1,,i]))=j}.\mathcal{L}_{i,j}=\{e\ :\ e=\overline{e_{\ell_{1},\dots,\ell_{i}}}\in\mathrm{Ker}\overline{\partial}_{i}\setminus\mathrm{Im}\overline{\partial}_{i+1},\mathrm{Sum}(\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i}]))=j\}.
  1. (1)

    If for all faces {Sb1,,Sbi}\{S^{\textbf{b}_{\ell_{1}}},\dots,S^{\textbf{b}_{\ell_{i}}}\} of Δ\Delta with Sum(Max(AB[1,,i]))=j\mathrm{Sum}(\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i}]))=j we have Max(AB[1,,i])Max(AB[1,,k^,,i]),\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i}])\neq\mathrm{Max}(AB[\ell_{1},\dots,\widehat{\ell_{k}},\dots,\ell_{i}]), for all 1ki1\leq k\leq i, then βi,j(R/It)|i,j|\beta_{i,j}(R/I^{t})\leq|\mathcal{L}_{i,j}|.

  2. (2)

    If for all faces {Sb1,,Sbi+1}\{S^{\textbf{b}_{\ell_{1}}},\dots,S^{\textbf{b}_{\ell_{i+1}}}\} of Δ\Delta with Sum(Max(AB[1,,i]))=j\mathrm{Sum}(\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i}]))=j and Max(AB[1,,i+1])=Max(AB[1,,i])\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i+1}])=\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i}]), we have

    Max(AB[1,,k^,,i+1])Max(AB[1,,i]),\mathrm{Max}(AB[\ell_{1},\dots,\widehat{\ell_{k}},\dots,\ell_{i+1}])\neq\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i}]),

    for all 1ki1\leq k\leq i, then βi,j(R/It)|i,j|\beta_{i,j}(R/I^{t})\geq|\mathcal{L}_{i,j}|.

3. Powers of the edge ideal of a hypergraph

To start, we recall the following definitions.

Definition 3.1.

(See [5] and [16, Definitions 2.1 and Notation 2.3].) Let 𝒮={S1,,Si}\mathcal{S}=\{S_{1},\dots,S_{i}\} be a family of edges of \mathcal{H}. We define the type of 𝒮\mathcal{S} as (i,j)(i,j), where ii is the cardinality of 𝒮\mathcal{S} and j=|1iS|j=|\bigcup_{1\leq\ell\leq i}S_{\ell}|.

  1. (1)

    𝒮\mathcal{S} is called a matching in \mathcal{H} if for each ,\ell,\ell^{\prime} with 1<i1\leq\ell<\ell^{\prime}\leq i, SS=S_{\ell}\cap S_{\ell^{\prime}}=\emptyset.

  2. (2)

    𝒮\mathcal{S} is called a self matching in \mathcal{H} if for all kk with 1ki1\leq k\leq i, Sk1i,kSS_{k}\nsubseteq\bigcup_{1\leq\ell\leq i,\ell\neq k}S_{\ell}.

  3. (3)

    𝒮\mathcal{S} is called a semi-induced matching in \mathcal{H} if for each S(){S1,,Si}S\in\mathcal{E}(\mathcal{H})\setminus\{S_{1},\dots,S_{i}\}, S1iSS\nsubseteq\bigcup_{1\leq\ell\leq i}S_{\ell}.

  4. (4)

    𝒮\mathcal{S} is called a self semi-induced matching in \mathcal{H} if 𝒮\mathcal{S} is both self matching and semi-induced matching.

  5. (5)

    𝒮\mathcal{S} is called an induced matching in \mathcal{H} if 𝒮\mathcal{S} is both matching and semi-induced matching.

  6. (6)

    We use the following notions:

    m=max{i:there is a matching of size i in };\displaystyle m_{\mathcal{H}}=\max\{i\ :\ \textrm{there is a matching of size }i\textrm{ in }\mathcal{H}\};
    (i.m)=max{i:there is an induced matching of size i in };\displaystyle(i.m)_{\mathcal{H}}=\max\{i\ :\ \textrm{there is an induced matching of size }i\textrm{ in }\mathcal{H}\};
    (i.m)=max{ji:there is an induced matching of type (i,j) in };\displaystyle(i.m)^{\prime}_{\mathcal{H}}=\max\{j-i\ :\ \textrm{there is an induced matching of type }(i,j)\textrm{ in }\mathcal{H}\};
    (s.s.i.m)=max{i:there is a self semi-induced matching of size i in };\displaystyle(s.s.i.m)_{\mathcal{H}}=\max\{i\ :\ \textrm{there is a self semi-induced matching of size }i\textrm{ in }\mathcal{H}\};
    (s.s.i.m)=max{ji:there is a self semi-induced matching of type (i,j) in };\displaystyle(s.s.i.m)^{\prime}_{\mathcal{H}}=\max\{j-i\ :\ \textrm{there is a self semi-induced matching of type }(i,j)\textrm{ in }\mathcal{H}\};
    (s.i.m)=max{ji:there is a semi-induced matching of type (i,j) in }.\displaystyle(s.i.m)^{\prime}_{\mathcal{H}}=\max\{j-i\ :\ \textrm{there is a semi-induced matching of type }(i,j)\textrm{ in }\mathcal{H}\}.

The following observations may be helpful in the sequel.

Remark 3.2.

Suppose that dd is the maximum size of edges in \mathcal{H}.

  1. (1)

    The following inequalities are straightforward:

    (i.m)min{m,(s.s.i.m)},(i.m)(s.s.i.m)(s.i.m).(i.m)_{\mathcal{H}}\leq\min\{m_{\mathcal{H}},(s.s.i.m)_{\mathcal{H}}\},\ \ (i.m)^{\prime}_{\mathcal{H}}\leq(s.s.i.m)^{\prime}_{\mathcal{H}}\leq(s.i.m)^{\prime}_{\mathcal{H}}.

    Also, if 𝒮={S1,,Si}\mathcal{S}=\{S_{1},\dots,S_{i}\} is an induced matching of type (i,j)(i,j) in \mathcal{H}, then

    ji=|1iS|i=1i|S|i(d1)i(d1)(i.m).j-i=|\bigcup_{1\leq\ell\leq i}S_{\ell}|-i=\sum_{1\leq\ell\leq i}|S_{\ell}|-i\leq(d-1)i\leq(d-1)(i.m)_{\mathcal{H}}.

    Therefore (i.m)(d1)(i.m)(i.m)^{\prime}_{\mathcal{H}}\leq(d-1)(i.m)_{\mathcal{H}}. Furthermore the equality holds when there exists an induced matching of maximum size containing dd-sets. Hence in a dd-uniform hypergraph

    (d1)(i.m)=(i.m)(s.s.i.m)(s.i.m).(d-1)(i.m)_{\mathcal{H}}=(i.m)^{\prime}_{\mathcal{H}}\leq(s.s.i.m)^{\prime}_{\mathcal{H}}\leq(s.i.m)^{\prime}_{\mathcal{H}}.

    Thus in view of [15, Proposition 2.7] for any dd-uniform simple hypergraph \mathcal{H} such that for each pair of distinct edges EE and EE^{\prime}, EEE\cap E^{\prime}\neq\emptyset implies |EE|=d1|E\cap E^{\prime}|=d-1 (in particular for simple graphs) we have

    (d1)(i.m)=(i.m)=(s.s.i.m)=(s.i.m).(d-1)(i.m)_{\mathcal{H}}=(i.m)^{\prime}_{\mathcal{H}}=(s.s.i.m)^{\prime}_{\mathcal{H}}=(s.i.m)^{\prime}_{\mathcal{H}}.
  2. (2)

    The matrix A=(ai,j)n×mA=(a_{i,j})_{n\times m}, defined in Notation 2.1, is the well-known incidence matrix of \mathcal{H}.

  3. (3)

    B=(bi,j)m×pm,tB=(b_{i,j})_{m\times p_{m,t}}, defined in Notation 2.1, is a matrix in which the rows are labelled by SiS_{i}s and the columns are labelled by elements of m,t\mathcal{B}_{m,t}. So, AB=(ci,j)n×pm,tAB=(c_{i,j})_{n\times p_{m,t}}, where ci,j=1km,xiSkbk,jc_{i,j}=\sum_{1\leq k\leq m,x_{i}\in S_{k}}b_{k,j}.

Lemma 3.3.

Suppose that kk\in\mathbb{N} and 𝒮={S1,,Si}\mathcal{S}=\{S_{1},\dots,S_{i}\} is a self semi-induced matching in \mathcal{H}. Then for each (not necessarily distinct integers) 1u1,,uki1\leq u_{1},\dots,u_{k}\leq i, Su1SukS_{u_{1}}\dots S_{u_{k}} is a minimal monomial generator of IkI^{k}.

Proof.

Assume on the contrary that Su1SukS_{u_{1}}\dots S_{u_{k}} is not a minimal generator of IkI^{k} for some 1u1,,uki1\leq u_{1},\dots,u_{k}\leq i. Then there exist (not necessarily distinct) elements r1,,rk{1,,m}r_{1},\dots,r_{k}\in\{1,\dots,m\} such that Sr1SrkSu1SukS_{r_{1}}\dots S_{r_{k}}\mid S_{u_{1}}\dots S_{u_{k}} and {r1,,rk}{u1,,uk}\{r_{1},\dots,r_{k}\}\neq\{u_{1},\dots,u_{k}\}. By omitting equal terms from both sides of the division, we have SrSuj1SujsS_{r}\mid S_{u_{j_{1}}}\dots S_{u_{j_{s}}} for somer{r1,,rk}{uj1,,ujs}r\in\{r_{1},\dots,r_{k}\}\setminus\{u_{j_{1}},\dots,u_{j_{s}}\} where {uj1,,ujs}{u1,,uk}\{u_{j_{1}},\dots,u_{j_{s}}\}\subseteq\{u_{1},\dots,u_{k}\}. This meansSrSuj1SujsS_{r}\subseteq S_{u_{j_{1}}}\cup\dots\cup S_{u_{j_{s}}}. This contradicts to 𝒮\mathcal{S} is a self semi-induced matching. ∎

Now we need some definitions from [7] and [8]. In [8] a simplicial complex 𝕃mt\mathbb{L}_{m}^{t} is defined and it is shown that it supports a free resolution of ItI^{t}. In [7], the case t=2t=2 is investigated individually. So, for our next results of this section, we refer the reader to Definitions 3.1 and 3.4 in [7] and Definitions 4.2 and 5.1 and Proposition 4.3 in [8]. Hereinafter we apply the notions in Section 2 for the simplicial complex 𝕃mt\mathbb{L}_{m}^{t}.

Proposition 3.4.

Suppose that \mathcal{H} is a dd-uniform hypergraph, i>1,j=2dii>1,j=2di and

i,j={e:e=e1,,i¯Ker¯iIm¯i+1,Sum(Max(AB[1,,i]))=j}.\mathcal{L}_{i,j}=\{e\ :\ e=\overline{e_{\ell_{1},\dots,\ell_{i}}}\in\mathrm{Ker}\overline{\partial}_{i}\setminus\mathrm{Im}\overline{\partial}_{i+1},\mathrm{Sum}(\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i}]))=j\}.

Then βi,j(R/I2)=|i,j|\beta_{i,j}(R/I^{2})=|\mathcal{L}_{i,j}|.

Proof.

In view of Lemma 2.3, it is enough to show that the assumptions of Parts 1 and 2 in Lemma 2.3 hold, when t=2t=2, i>1i>1 and j=2dij=2di. To this end, suppose that τ={Sb1,,Sbi}\tau=\{S^{\textbf{b}_{\ell_{1}}},\dots,S^{\textbf{b}_{\ell_{i}}}\} is a face of 𝕃2(I)\mathbb{L}^{2}(I). By means of [7, Definition 3.1], one may assume that τ\tau has one of the following forms:

τ={S12,S1S2,,S1Si} or τ={S1S1,,SiSi},\tau=\{S_{1}^{2},S_{1}S_{2},\dots,S_{1}S_{i}\}\text{ or }\tau=\{S_{\ell_{1}}S_{\ell^{\prime}_{1}},\dots,S_{\ell_{i}}S_{\ell^{\prime}_{i}}\},

when k<k\ell_{k}<\ell^{\prime}_{k} for all 1ki1\leq k\leq i. Now in each case we investigate the degree of e1,,ie_{\ell_{1},\dots,\ell_{i}}. Suppose that Max(AB[1,,i])=(ck)\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i}])=(c_{k}). In view of Remark 3.2(3), in the first case, ck=2c_{k}=2 if xkS1x_{k}\in S_{1}, ck=1c_{k}=1 if xk2iSS1x_{k}\in\bigcup_{2\leq\ell\leq i}S_{\ell}\setminus S_{1} and else ck=0c_{k}=0. So

deg(e1,,i)\displaystyle\mathrm{deg}(e_{\ell_{1},\dots,\ell_{i}}) =ck\displaystyle=\sum c_{k}
=2d+|2iSS1|\displaystyle=2d+|\bigcup_{2\leq\ell\leq i}S_{\ell}\setminus S_{1}|
2d+(i1)d\displaystyle\leq 2d+(i-1)d
=(i+1)d<2di.\displaystyle=(i+1)d<2di.

Hence τ\tau should be in the form of the second one. Now, in the second case ck=2c_{k}=2 if xk1ri(SrSr)x_{k}\in\bigcup_{1\leq r\leq i}(S_{\ell_{r}}\cap S_{\ell^{\prime}_{r}}), ck=1c_{k}=1 if xk1ri((SrSr)(SrSr))x_{k}\in\bigcup_{1\leq r\leq i}((S_{\ell_{r}}\cup S_{\ell^{\prime}_{r}})\setminus(S_{\ell_{r}}\cap S_{\ell^{\prime}_{r}})) and else ck=0c_{k}=0. Hence deg(e1,,i)=ck2di,\mathrm{deg}(e_{\ell_{1},\dots,\ell_{i}})=\sum c_{k}\leq 2di, and the equality holds when (SkSk)(S_{\ell_{k}}\cup S_{\ell^{\prime}_{k}})s are disjoint for all 1ki1\leq k\leq i. So, for all 1ki1\leq k\leq i, the rows associated to the vertices in SkSkS_{\ell_{k}}\cup S_{\ell^{\prime}_{k}} will be zero in Max(AB[1,,k^,,i])\mathrm{Max}(AB[\ell_{1},\dots,\widehat{\ell_{k}},\dots,\ell_{i}]), while they have entries 1 or 2 in Max(AB[1,,i])\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i}]). Thus Max(AB[1,,k^,,i])Max(AB[1,,i])\mathrm{Max}(AB[\ell_{1},\dots,\widehat{\ell_{k}},\dots,\ell_{i}])\neq\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i}]), for all 1ki1\leq k\leq i. This shows that the assumption of Lemma 2.3(1) holds and so βi,j(R/I2)|i,j|\beta_{i,j}(R/I^{2})\leq|\mathcal{L}_{i,j}|.

Now suppose that τ={Sb1,,Sbi+1}\tau=\{S^{\textbf{b}_{\ell_{1}}},\dots,S^{\textbf{b}_{\ell_{i+1}}}\} is a face of 𝕃2(I)\mathbb{L}^{2}(I) with

Sum(Max(AB[1,,i]))=2di,Max(AB[1,,i+1])=Max(AB[1,,i]).\mathrm{Sum}(\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i}]))=2di,\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i+1}])=\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i}]).

By means of the above explanation, we should have τ={S1S1,,Si+1Si+1},\tau=\{S_{\ell_{1}}S_{\ell^{\prime}_{1}},\dots,S_{\ell_{i+1}}S_{\ell^{\prime}_{i+1}}\}, when k<k\ell_{k}<\ell^{\prime}_{k} for all 1ki+11\leq k\leq i+1 and (SkSk)(S_{\ell_{k}}\cup S_{\ell^{\prime}_{k}})s are disjoint for all 1ki1\leq k\leq i and Si+1Si+11ri(SrSr)S_{\ell_{i+1}}\cup S_{\ell^{\prime}_{i+1}}\subseteq\bigcup_{1\leq r\leq i}(S_{\ell_{r}}\cup S_{\ell^{\prime}_{r}}). Suppose that 1ki1\leq k\leq i. Then in view of the structure of 𝕃2(I)\mathbb{L}^{2}(I), SkSkSi+1Si+1S_{\ell_{k}}S_{\ell^{\prime}_{k}}\nmid S_{\ell_{i+1}}S_{\ell^{\prime}_{i+1}}. So there exists a variable xrkx_{r_{k}} such that

\bullet either xrkSkSkx_{r_{k}}\in S_{\ell_{k}}\cup S_{\ell^{\prime}_{k}} but xrkSi+1Si+1x_{r_{k}}\not\in S_{\ell_{i+1}}\cup S_{\ell^{\prime}_{i+1}},

\bullet or xrkSkSkx_{r_{k}}\in S_{\ell_{k}}\cap S_{\ell^{\prime}_{k}} but xrkSi+1Si+1x_{r_{k}}\not\in S_{\ell_{i+1}}\cap S_{\ell^{\prime}_{i+1}}.

Hence

\bullet either the rkr_{k}th row in Max(AB[1,,i])\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i}]) is one or two, while the rkr_{k}th row in

Max(AB[1,,k^,,i+1]),\mathrm{Max}(AB[\ell_{1},\dots,\widehat{\ell_{k}},\dots,\ell_{i+1}]),

is zero,

\bullet or the rkr_{k}th row in Max(AB[1,,i])\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i}]) is two, while the rkr_{k}th row in

Max(AB[1,,k^,,i+1]),\mathrm{Max}(AB[\ell_{1},\dots,\widehat{\ell_{k}},\dots,\ell_{i+1}]),

is one.

Therefore in each case we have

Max(AB[1,,k^,,i+1])Max(AB[1,,i]).\mathrm{Max}(AB[\ell_{1},\dots,\widehat{\ell_{k}},\dots,\ell_{i+1}])\neq\mathrm{Max}(AB[\ell_{1},\dots,\ell_{i}]).

Thus by Lemma 2.3(2), βi,j(R/I2)|i,j|\beta_{i,j}(R/I^{2})\geq|\mathcal{L}_{i,j}|. These complete the proof. ∎

One can use Proposition 3.4 for vanishing of the special graded Betti numbers of the second power of the edge ideal as follows.

Corollary 3.5.

Suppose that \mathcal{H} is a dd-uniform hypergraph and i>1i>1 such that βi,2di(R/I2)0\beta_{i,2di}(R/I^{2})\neq 0. Then \mathcal{H} should have 2i2^{i} matchings of size ii.

Proof.

If βi,2di(R/I2)0\beta_{i,2di}(R/I^{2})\neq 0, then in view of Proposition 3.4, i,2di\mathcal{L}_{i,2di}\neq\emptyset. So, there exists an element eτe_{\tau} in i,2di\mathcal{L}_{i,2di} of degree 2di2di. As one can see in the proof of Proposition 3.4, τ\tau should be in the form of {S1S1,,SiSi}\{S_{\ell_{1}}S_{\ell^{\prime}_{1}},\dots,S_{\ell_{i}}S_{\ell^{\prime}_{i}}\}, when k<k\ell_{k}<\ell^{\prime}_{k} and (SkSk)(S_{\ell_{k}}\cup S_{\ell^{\prime}_{k}})s are disjoint for all 1ki1\leq k\leq i. Hence a set consisting precisely one edge from each set {Sk,Sk}\{S_{\ell_{k}},S_{\ell^{\prime}_{k}}\} for 1ki1\leq k\leq i, will be a matching of size ii in \mathcal{H}. Clearly, there exist 2i2^{i} such sets. ∎

Now we prove Theorem A.

Proof of Theorem A.
  1. (1)

    Suppose that 𝒮={S1,,Si}\mathcal{S}=\{S_{1},\dots,S_{i}\} is a self semi-induced matching of type (i,j)(i,j) in \mathcal{H}. Then for each 1i1\leq\ell\leq i, set τ={St1Sj:1ji}\tau_{\ell}=\{S_{\ell}^{t-1}S_{j}:1\leq j\leq i\}. (Note that if t=1t=1, then τ1==τi\tau_{1}=\dots=\tau_{i}.) By means of Lemma 3.3, for each 1i1\leq\ell\leq i, τ\tau_{\ell} is a face of 𝕃t(I)\mathbb{L}^{t}(I). We show that

    eτ¯(Ker¯iIm¯i+1)|S|(t1)+j.\overline{e_{\tau_{\ell}}}\in(\mathrm{Ker}\overline{\partial}_{i}\setminus\mathrm{Im}\overline{\partial}_{i+1})_{|S_{\ell}|(t-1)+j}.

    Firstly note that

    deg(eτ)\displaystyle\mathrm{deg}(e_{\tau_{\ell}}) =deg(St1lcm(S1,,Si))\displaystyle=\mathrm{deg}(S_{\ell}^{t-1}\mathrm{lcm}(S_{1},\dots,S_{i}))
    =deg(St1)+deg(lcm(S1,,Si))\displaystyle=\mathrm{deg}(S_{\ell}^{t-1})+\mathrm{deg}(\mathrm{lcm}(S_{1},\dots,S_{i}))
    =|S|(t1)+|1kiSk|=|S|(t1)+j.\displaystyle=|S_{\ell}|(t-1)+|\bigcup_{1\leq k\leq i}S_{k}|=|S_{\ell}|(t-1)+j.

    Without loss of generality we may assume that =1\ell=1 and for 1ri1\leq r\leq i, brm,t\textbf{b}_{r}\in\mathcal{B}_{m,t} is the vector associated to the rrth element of τ1\tau_{1}. Now, in view of Remark 2.2(1), to prove eτ1¯Ker¯i\overline{e_{\tau_{1}}}\in\mathrm{Ker}\overline{\partial}_{i} it is enough to show that for all 1ki1\leq k\leq i,

    Max(AB[1,,i])Max(AB[1,,k^,,i]).\mathrm{Max}(AB[1,\dots,i])\neq\mathrm{Max}(AB[1,\dots,\widehat{k},\dots,i]).

    Suppose that 2ki2\leq k\leq i (resp. k=1k=1). Since 𝒮\mathcal{S} is a self-matching, there is a vertex xukSk1i,kSx_{u_{k}}\in S_{k}\setminus\bigcup_{1\leq\ell\leq i,\ell\neq k}S_{\ell}. Now, in view of Remark 3.2(3), if AB[1,,i]=(cr,r)n×iAB[1,\dots,i]=(c_{r,r^{\prime}})_{n\times i}, then cuk,k=1c_{u_{k},k}=1 (resp. ck,k=tc_{\ell_{k},k}=t) and the other entries in the uku_{k}th row are zero (resp. t1t-1). So the uku_{k}th component in Max(AB[1,,i])\mathrm{Max}(AB[1,\dots,i]) is one (resp. tt), while the uku_{k}th component in Max(AB[1,,k^,,i])\mathrm{Max}(AB[1,\dots,\widehat{k},\dots,i]) is zero (resp. t1t-1).

    To prove the last part of our claim, suppose in contrary that e1,,i¯Im¯i+1\overline{e_{1,\dots,i}}\in\mathrm{Im}\overline{\partial}_{i+1}. Then by Remark 2.2(2), there should exist a face τ\tau^{\prime} containing τ1\tau_{1} such that

    Max(AB[1,,i])=Max(AB[1,,i+1]).\mathrm{Max}(AB[1,\dots,i])=\mathrm{Max}(AB[1,\dots,i+1]).

    Note that since τ\tau^{\prime} is a face containing S1tS_{1}^{t}, we should have

    τ={S1t,S1t1S2,,S1t1Si,S1t1Si+1},\tau^{\prime}=\{S_{1}^{t},S_{1}^{t-1}S_{2},\dots,S_{1}^{t-1}S_{i},S_{1}^{t-1}S_{i+1}\},

    by renumbering the edges if it is required. Hence eτ=e1,,i+1e_{\tau^{\prime}}=e_{1,\dots,i+1}, where bi+1\textbf{b}_{i+1} is a vector whose the first component is t1t-1 and the (i+1)(i+1)th component is one and other components are zero. Now, since 𝒮\mathcal{S} is a semi-induced matching, Si+11iSS_{i+1}\nsubseteq\bigcup_{1\leq\ell\leq i}S_{\ell}. Thus similar argument to above paragraph ensures the contradiction Max(AB[1,,i])Max(AB[1,,i+1])\mathrm{Max}(AB[1,\dots,i])\neq\mathrm{Max}(AB[1,\dots,i+1]).

  2. (2)

    The proof of Part 1 implies 2.

  3. (3)

    immediately follows from Remark 3.2(1) and Part 1.

In view of Remark 3.2(1) and [1, Example 5.2], the equalities in Theorem A(3) holds for many classes of hypergraphs and there exists examples illustrating the strictness of the bound.

The following corollary, which is an immediate consequence of [15, Theorem 3.6], Theorem A(3) and Remark 3.2(1), can regain Theorem 2.4 in [13]. Recall that a chain in \mathcal{H} is a sequence v0,S1,v1,,Sk,vkv_{0},S_{1},v_{1},\dots,S_{k},v_{k} of vertices and edges in \mathcal{H}, where viSiv_{i}\in S_{i} for 1ik,viSi+11\leq i\leq k,v_{i}\in S_{i+1} for 0ik10\leq i\leq k-1. When k>1k>1, we say that \mathcal{H} is 𝒞k\mathcal{C}_{k}-free if it doesn’t contain any chain v0,S1,v1,,Sk,v0v_{0},S_{1},v_{1},\dots,S_{k},v_{0} with k>1k>1 and distinct viv_{i}s and EiE_{i}s.

Corollary 3.6.

Let \mathcal{H} be a (C2,C5)(C_{2},C_{5})-free vertex decomposable hypergraph. Then

(s.s.i.m)reg(R/I)(s.i.m).(s.s.i.m)^{\prime}_{\mathcal{H}}\leq\mathrm{reg}(R/I)\leq(s.i.m)^{\prime}_{\mathcal{H}}.

So for any C5C_{5}-free vertex decomposable graph we regain

reg(R/I(G))=(i.m)G.\mathrm{reg}(R/I(G))=(i.m)_{G}.

Note that the following example illustrates that the equality in Corollary 3.6 doesn’t always hold.

Example 3.7.

Assume that \mathcal{H} is a 33-uniform (C2,C5)(C_{2},C_{5})-free vertex decomposable simple hypergraph with vertex set {x1,,x9}\{x_{1},\dots,x_{9}\} and edge set

{{x1,x2,x3},{x4,x5,x6},{x7,x8,x9},{x1,x4,x7}}.\{\{x_{1},x_{2},x_{3}\},\{x_{4},x_{5},x_{6}\},\{x_{7},x_{8},x_{9}\},\{x_{1},x_{4},x_{7}\}\}.

Then one can see that

(s.s.i.m)=4reg(R/I)=(s.i.m)=5.(s.s.i.m)^{\prime}_{\mathcal{H}}=4\leq\mathrm{reg}(R/I)=(s.i.m)^{\prime}_{\mathcal{H}}=5.

The next example gives a class of hypergraphs achieving the equality in Theorem A(3). This example demonstrates the advantage of our lower bound over those previously found.

Example 3.8.

Assume that \mathcal{H} is a dd-uniform simple hypergraph with

V()={x1,,xk}1is1jdk{xi,j},V(\mathcal{H})=\{x_{1},\dots,x_{k}\}\cup\bigcup_{1\leq i\leq s}\bigcup_{1\leq j\leq d-k}\{x_{i,j}\},

and edge set

()={Ei={x1,,xk,xi,1,,xi,dk}:1is},\mathcal{E}(\mathcal{H})=\{E_{i}=\{x_{1},\dots,x_{k},x_{i,1},\dots,x_{i,d-k}\}:1\leq i\leq s\},

where kk is an integer with 1kd11\leq k\leq d-1. One can check that (s.s.i.m)=s(dk1)+k(s.s.i.m)^{\prime}_{\mathcal{H}}=s(d-k-1)+k, since ()\mathcal{E}(\mathcal{H}) forms the desired self semi induced matching in \mathcal{H}. Also if we set u=x1xku=x_{1}\dots x_{k} and ui=1jdkxi,ju_{i}=\prod_{1\leq j\leq d-k}x_{i,j} for every integer ii with 1is1\leq i\leq s, then I=uu1,,usI=u\langle u_{1},\dots,u_{s}\rangle. Thus since u1,,us\langle u_{1},\dots,u_{s}\rangle is a complete intersection, in view of [6, Lemma 4.4], for each tt\in\mathbb{N} we have

reg(R/It)\displaystyle\mathrm{reg}(R/I^{t}) =reg(It)1\displaystyle=\mathrm{reg}(I^{t})-1
=reg(utu1,,ust)1\displaystyle=\mathrm{reg}(u^{t}\langle u_{1},\dots,u_{s}\rangle^{t})-1
=deg(ut)+reg(u1,,ust)1\displaystyle=\deg(u^{t})+\mathrm{reg}(\langle u_{1},\dots,u_{s}\rangle^{t})-1
=kt+(dk)t+(dk1)(s1)1\displaystyle=kt+(d-k)t+(d-k-1)(s-1)-1
=d(t1)+s(dk1)+k\displaystyle=d(t-1)+s(d-k-1)+k
=d(t1)+(s.s.i.m).\displaystyle=d(t-1)+(s.s.i.m)^{\prime}_{\mathcal{H}}.

Also this is obvious that for large values of ss, this bound is reasonably greater than the bound d(t1)+(d1)(i.m)d(t-1)+(d-1)(i.m)_{\mathcal{H}} given in [3, Corollary 3.8].

Although there is no general upper bound known for regularity of powers of the edge ideals of arbitrary hypergraphs, but there is a good one for graphs in terms of matching number ([2]). So using Macaulay 2 ([14]) for some examples of hypergraphs and some positive integers tt and also thanks to the general upper bound for graphs in [2, Theorem 3.4] yields to the following question.

Question 3.9.

Suppose that tt\in\mathbb{N}. Then

  • 1.

    Is the following inequality holds for each dd-uniform hypergraph \mathcal{H} in which every two intersecting edges have exactly d1d-1 common vertices?

    reg(R/It)d(t1)+m(d1).\mathrm{reg}(R/I^{t})\leq d(t-1)+m_{\mathcal{H}}(d-1).
  • 2.

    For which hypergraphs we have reg(R/I()t)=d(t1)+(s.s.i.m)\mathrm{reg}(R/I(\mathcal{H})^{t})=d(t-1)+(s.s.i.m)^{\prime}_{\mathcal{H}}?

Note that the condition of Question 3.9(1) is necessary. For instance by example 3.8 for 33-uniform hypergraph \mathcal{H} with V()={x1,,x5}V(\mathcal{H})=\{x_{1},\dots,x_{5}\} and ()={{x1,x2,x3},{x3,x4,x5}}\mathcal{E}(\mathcal{H})=\{\{x_{1},x_{2},x_{3}\},\{x_{3},x_{4},x_{5}\}\} we see that reg(R/I(H))=3\mathrm{reg}(R/I(H))=3, reg(R/I()3)=9\mathrm{reg}(R/I(\mathcal{H})^{3})=9 and m=1m_{\mathcal{H}}=1 and so d(t1)+m(d1)d(t-1)+m_{\mathcal{H}}(d-1) can not be an upper bound even for the first power and dd-uniform well-known tree hypergraphs.

Acknowledgement. The author is deeply grateful to Sara Faridi for drawing her attention to [7] and [8]. Also she would like to thank the referee for his/her valuable comments which substantially improved the quality of the paper.

References

  • [1] Banerjee, A.; Beyarslan, S. K.; Ha, H. T., Regularity of edge ideals and their powers. Advances in algebra, 17–52, Springer Proc. Math. Stat., 277, Springer, Cham, 2019.
  • [2] Banerjee, A.; Beyarslan, S. K.; Ha, H. T., Regularity of powers of edge ideals: from local properties to global bounds. Algebr. Comb. 3 (2020), no. 4, 839–854.
  • [3] Banerjee, A.; Chakraborty, B.; Das, K. K.; Mandal, M.; Selvaraja, S., Regularity of powers of squarefree monomial ideals. J. Pure Appl. Algebra 226 (2022), no. 2, Paper No. 106807, 12 pp.
  • [4] Bayer, D.; Peeva, I.; Sturmfels, B. Monomial resolutions. Math. Res. Lett. 5 (1998), no. 1-2, 31–46.
  • [5] Berge, C. Graphs and hypergraphs. Translated from the French by Edward Minieka. North-Holland Mathematical Library, Vol. 6. North-Holland Publishing Co., Amsterdam-London; American Elsevier Publishing Co., Inc., New York, 1973. xiv+528 pp.
  • [6] Beyarslan, S.; Ha, H. T.; Trung, T. N., Regularity of powers of forests and cycles. Journal of Algebraic Combinatorics, 42 (2015), no. 4, 1077-1095.
  • [7] Cooper, S. M.; El Khoury, S.; Faridi, S.; Mayes-Tang, S.; Morey, S.; Sega, L. M.; Spiroff, S. Simplicial resolutions for the second power of square-free monomial ideals. arXiv:2103.04074v2 .
  • [8] Cooper, S. M.; El Khoury, S.; Faridi, S.; Mayes-Tang, S.; Morey, S.; Sega, L. M.; Spiroff, S. Simplicial resolutions of powers of square-free monomial ideals. arXiv:2204.03136v1 .
  • [9] Ha, H. T., Regularity of squarefree monomial ideals. Connections between algebra, combinatorics, and geometry, 251–276, Springer Proc. Math. Stat., 76, Springer, New York, 2014.
  • [10] Ha, H. T.; Van Tuyl, A. Monomial ideals, edge ideals of hypergraphs, and their graded Betti numbers. J. Algebraic Combin. 27 (2008), no. 2, 215–245.
  • [11] Results on…
  • [12] Katzman, M. Characteristic-independence of Betti numbers of graph ideals. J. Combin. Theory Ser. A 113 (2006), no. 3, 435–454.
  • [13] Khosh-Ahang, F.; Moradi, S. Regularity and projective dimension of the edge ideal of C5C_{5}-free vertex decomposable graphs. Proc. Amer. Math. Soc. 142 (2014), no. 5, 1567–1576.
  • [14] Mcaulay2 Team, Macaulay2: a software system devoted to supporting research in algebraic geometry and commutative algebra. Available at https://macaulay2.com/TryItOut/
  • [15] Moradi, S.; Khosh-Ahang, F. Matchings in hypergraphs and Castelnuovo-Mumford regularity. Publ. Math. Debrecen 91/3–4 (2017) 427–439.
  • [16] Moradi, S.; Khosh-Ahang, F. Some interpretations for algebraic invariants of edge ideals of hypergraphs via combinatorial invariants. Comm. in Algebra, 48 (2020), no. 6, 2699–2712.
  • [17] Morey, S.; Villarreal, R. H. Edge ideals: algebraic and combinatorial properties. Progress in commutative algebra 1, 85–126, de Gruyter, Berlin, 2012.
  • [18] Villarreal, R. H. Cohen-Macaulay graphs. Manuscripta Math. 66 (1990) 277–293.