This paper investigates the algebraic properties of binomial edge ideals of graphs, providing formulas for Betti numbers, depth, and Cohen-Macaulay defect, and constructing graphs with specific algebraic invariants.
Contribution
It offers explicit computations of Betti numbers for binomial edge ideals of cone graphs and constructs graphs with prescribed Cohen-Macaulay defect and extremal Betti numbers.
Findings
01
Betti numbers of $J_G$ expressed in terms of $J_H$
02
Depth of binomial edge ideals of join graphs determined
03
Existence of graphs with given regularity and extremal Betti number count
Abstract
Let G be a simple graph on the vertex set [n] and JG be the corresponding binomial edge ideal. Let G=v∗H be the cone of v on H. In this article, we compute all the Betti numbers of JG in terms of Betti number of JH and as a consequence, we get the Betti diagram of wheel graph. Also, we study Cohen-Macaulay defect of S/JG in terms of Cohen-Macaulay defect of SH/JH and using this we construct a graph with Cohen-Macaulay defect q for any q≥1. We obtain the depth of binomial edge ideal of join of graphs. Also, we prove that for any pair (r,b) of positive integers with 1≤b<r, there exists a connected graph G such that reg(S/JG)=r and the number of extremal Betti number of S/JG is b.
\beta_{i,i+j}^{S}\left(\frac{S}{J_{G}}\right)=\left\{\begin{array}[]{ll}i\left({\sf k}_{i}(H)+{\sf k}_{i+1}(H)\right),&\text{ if }j=1\\
\beta_{i,i+2}^{S_{H}}\left(\frac{S_{H}}{J_{H}}\right)+2\beta_{i-1,i+1}^{S_{H}}\left(\frac{S_{H}}{J_{H}}\right)+\beta_{i-2,i}^{S_{H}}\left(\frac{S_{H}}{J_{H}}\right)\\
+(i-1)\binom{n+1}{i+1}-(i-1){\sf k}_{i}(H)-(i-1){\sf k}_{i+1}(H),&\text{ if }j=2\\
\beta_{i,i+j}^{S_{H}}\left(\frac{S_{H}}{J_{H}}\right)+2\beta_{i-1,i-1+j}^{S_{H}}\left(\frac{S_{H}}{J_{H}}\right)+\beta_{i-2,i-2+j}^{S_{H}}\left(\frac{S_{H}}{J_{H}}\right),&\text{ if }j\geq 3,\end{array}\right.
\beta_{i,i+j}^{S}\left(\frac{S}{J_{G}}\right)=\left\{\begin{array}[]{ll}i\left({\sf k}_{i}(H)+{\sf k}_{i+1}(H)\right),&\text{ if }j=1\\
\beta_{i,i+2}^{S_{H}}\left(\frac{S_{H}}{J_{H}}\right)+2\beta_{i-1,i+1}^{S_{H}}\left(\frac{S_{H}}{J_{H}}\right)+\beta_{i-2,i}^{S_{H}}\left(\frac{S_{H}}{J_{H}}\right)\\
+(i-1)\binom{n+1}{i+1}-(i-1){\sf k}_{i}(H)-(i-1){\sf k}_{i+1}(H),&\text{ if }j=2\\
\beta_{i,i+j}^{S_{H}}\left(\frac{S_{H}}{J_{H}}\right)+2\beta_{i-1,i-1+j}^{S_{H}}\left(\frac{S_{H}}{J_{H}}\right)+\beta_{i-2,i-2+j}^{S_{H}}\left(\frac{S_{H}}{J_{H}}\right),&\text{ if }j\geq 3,\end{array}\right.
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Full text
DEPTH AND EXTREMAL BETTI NUMBER OF BINOMIAL EDGE IDEALS
Department of Mathematics, Indian Institute of Technology
Madras, Chennai, INDIA - 600036
Abstract.
Let G be a simple graph on the vertex set [n] and JG be the corresponding binomial edge ideal.
Let G=v∗H be the cone of v on H. In this article, we compute all the Betti numbers of JG in terms of Betti numbers of
JH and as a consequence, we get the Betti diagram of wheel graph. Also, we study Cohen-Macaulay defect of S/JG
in terms of Cohen-Macaulay defect of SH/JH and using this we construct a graph with Cohen-Macaulay defect q for any q≥1. We obtain the depth of binomial edge ideal of join of graphs. Also, we prove that for
any pair (r,b) of positive integers with 1≤b<r, there exists a connected graph G such that reg(S/JG)=r
and the number of extremal Betti numbers of S/JG is b.
Key words and phrases:
Binomial edge ideal, Castelnuovo-Mumford regularity, Join of graphs, Depth, Extremal Betti number
Let R=K[x1,…,xm] be the polynomial ring over an arbitrary
field K and M be a finitely generated graded R-module.
Let
[TABLE]
be the minimal graded free resolution of M, where
R(−j) is the free R-module of rank 1 generated in degree j. The number βi,i+jR(M) is the
(i,i+j)-th graded Betti number of M. The
projective dimension and Castelnuovo-Mumford regularity are two invariants associated with
M that can be read off from the minimal free resolution. The
Castelnuovo-Mumford regularity of M, denoted by reg(M), is defined as
[TABLE]
and the projective dimension of M, denoted by pdR(M), is defined as
[TABLE]
A nonzero graded Betti number βi,i+jR(M) is called an extremal Betti number, if βr,r+sR(M)=0
for all pairs (r,s)=(i,j) with r≥i and s≥j. Observe that the extremal Betti number is unique if and only if
βp,p+rR(M)=0, where p=pdR(M) and r=reg(M).
Let G be a simple graph on V(G)={1,2,…,n}
and edge set E(G). Let S=K[x1,…,xn,y1,…,yn] be the polynomial ring over an arbitrary
field K. The ideal JG generated by the binomials xiyj−xjyi, where i<j and {i,j}∈E(G), is known as the
binomial edge ideal of G. The notion of binomial edge ideal
was introduced by Herzog et.al. in [5] and independently by Ohtani in [19]. Algebraic properties and
invariants of binomial edge ideals have been studied by many authors, see [4, 16, 21]. In particular, establishing a relationship between Castelnuovo-Mumford regularity (simply regularity), projective dimension, Hilbert series of binomial edge ideals and combinatorial invariants associated with graphs is an active area of
research, see [1, 11, 13, 18, 20]. In general, the
algebraic invariants such as regularity and depth of JG are hard
to compute. There are bounds known for the regularity and depth of
binomial edge ideals, see [1, 17]. The maximal possible depth
of binomial edge ideal of a connected graph on n vertices is
n+1 (see [1, Theorems 3.19, 3.20]). Also, if G is a
connected graph on n vertices such that S/JG is Cohen-Macaulay,
then 0ptS(S/JG)=n+1. In [2], de Alba and Hoang
studied the depth of some subclass of closed graphs. However not much
more is known about the depth of binomial edge ideal. For an ideal I⊂S, the Cohen-Macaulay defect of S/I is defined to be
cmdef(S/I):=dim(S/I)−0ptS(S/I). We study the depth and
Cohen-Macaulay defect of S/JG, where G is a cone of v on a
graph H, denoted by v∗H (for definition see section 3). We show that the depth remains invariant under the process
of taking cone on connected graph, (Theorem 3.4). As a
consequence, we prove that for any positive integer q, there exists
a graph having Cohen-Macaulay defect equal to q, (Corollary
3.6). We also compute the depth of S/Jv∗H, when H
is a disconnected graph, (Theorem 3.9).
Another homological invariant that helps in understanding more about
its structure is the Betti number. There have been few attempts in
computing the Betti numbers of binomial edge ideals, for example,
Zafar and Zahid for cycles, [24], Schenzel and Zafar for complete
bipartite graphs, [23], Jayanthan et al. for trees and
unicyclic graphs [12]. Extremal Betti numbers of binomial edge
ideals of closed graphs were studied by de Alba and Hoang in
[2]. In [7], Herzog and Rinaldo
studied extremal Betti number of binomial edge ideal of block graphs.
We compute all the Betti numbers of cone of a graph, (Theorem
3.10). As a consequence, we obtain the Betti numbers of
binomial edge ideal of wheel graph, (Corollary 3.11).
We then consider a more general form of cone, namely the join
product of two arbitrary graphs. Given two graphs
G1 and G2, it is interesting to understand the properties of
G1∗G2 (for definition see section 4) in terms of the corresponding
properties of G1 and G2. In [22], Kiani and Saeedi Madani
studied the regularity of JG1∗G2. We computed the Hilbert
series of binomial edge ideal of G1∗G2 in terms of the Hilbert
series of JG1 and JG2, [15]. In this article, we
study the depth of S/JG1∗G2, (Theorems 4.1,
4.3, 4.4). As a consequence, we obtain the
depth of complete multipartite graphs, (Corollary
4.5).
Recently, researchers are trying to construct graphs such that their
corresponding edge ideals satisfy certain algebraic properties. For a
given pair of positive integers (r,s), Hibi and Matsuda in
[10] showed the existence of monomial ideal Ir,s such
that reg(S/Ir,s)=r and the degree of h-polynomial of
S/Ir,s is s. In [8], Hibi et al. constructed a
graph G such that for 1≤b≤r, the regularity of the
monomial edge ideal of G is r and the number of its extremal Betti
numbers is b. Given a pair (r,s) with 1≤r≤s, Hibi and
Matsuda constructed a graph G such that reg(S/JG)=r and the
degree of h-polynomial of S/JG is s, [9].
In this article, we construct a graph G such that reg(S/JG)=r
and the number of extremal Betti numbers of S/JG is b, for 1≤b<r (Theorem 5.4).
2. Preliminaries
In this section, we recall some notation and fundamental results on
graphs and their corresponding binomial edge ideals.
Let G be a finite simple graph with vertex set V(G) and edge set
E(G). For A⊆V(G), G[A] denotes the induced
subgraph of G on the vertex set A, that is, for i,j∈A, {i,j}∈E(G[A]) if and only if {i,j}∈E(G).
For a vertex v, G∖v denotes the induced subgraph of G
on the vertex set V(G)∖{v}. A vertex v∈V(G) is
said to be a cut vertex if G∖v has more components than G. We say that G is kvertex-connected if k<n and for every
A⊂[n] with ∣A∣<k, the induced graph G[Aˉ] is connected, where Aˉ=[n]∖A. The vertex connectivity of a connected graph G, denoted by κ(G), is defined as the maximum positive integer k such that G is k vertex-connected.
A subset U of V(G) is said to be a
clique if G[U] is a complete graph. We denote the number of cliques of cardinality i in G by ki(G). A vertex v is said to be a simplicial vertex if it belongs to exactly one maximal clique. For a vertex v, NG(v)={u∈V(G)\leavevmode:\leavevmode{u,v}∈E(G)} denotes neighborhood of v and Gv is the
graph on the vertex set V(G) and edge set E(Gv)=E(G)∪{{u,w}:u,w∈NG(v)}. A component of G is said to be a nontrivial component if it has atleast one edge.
For T⊂[n], let Tˉ=[n]∖T and cT
denote the number of connected components of G[Tˉ]. Let G1,⋯,GcT be the connected
components of G[Tˉ]. For each i, let Gi~ denote the complete graph on V(Gi) and
PT(G)=(i∈T∪{xi,yi},JG1~,⋯,JG~cT).
It was shown by Herzog et al. that JG=T⊆[n]∩PT(G), [5, Theorem 3.2].
For each i∈T, if i is a cut vertex of the graph G[Tˉ∪{i}],
then we say that T has the cut point property. Set C(G)={∅}∪{T:Thas the cut point property}. It follows from [5, Corollary 3.9]
that T∈C(G) if and only if PT(G) is a minimal prime of JG. It follows from the Auslander-Buchsbaum formula that 0ptS(S/JG)=2n−pdS(S/JG).
The following basic property of depth is used repeatedly in this
article.
Lemma 2.1**.**
Let S be a standard graded polynomial ring and M,N and P be finitely generated graded S-modules.
If 0→MfNgP→0 is a
short exact sequence with f,g
graded homomorphisms of degree zero, then
(i)
0ptS(M)≥min{0ptS(N),0ptS(P)+1},**
2. (ii)
0ptS(M)=0ptS(P)+1* if 0ptS(N)>0ptS(P),*
3. (iii)
0ptS(M)=0ptS(N)* if 0ptS(N)<0ptS(P).*
3. Binomial edge ideal of cone of a graph
In this section, we study the binomial edge ideal of cone of a graph.
Let H be a graph on the vertex set [n]. The cone of v on H, denoted by v∗H, is the graph with
the vertex set V(v∗H)=V(H)⊔{v} and edge set E(v∗H)=E(H)⊔{{v,u}∣u∈V(H)}. From now, we assume
that H is not a complete graph. Set G=v∗H, SH=K[xi,yi:i∈V(H)] and S=SH[xv,yv]. First, we
recall a lemma due to Ohtani which is useful in this section.
Lemma 3.1**.**
([19, Lemma 4.8])
Let G be a graph on V(G) and v∈V(G) such that v is not a simplicial vertex. Then JG=(JG∖v+(xv,yv))∩JGv.
One can see that if G=v∗H, then Gv=Kn+1, Gv∖v=Kn and G∖v=H. Therefore, (xv,yv)+JG∖v+JGv=(xv,yv)+JKn. Thus, by Lemma 3.1, we have the following short exact sequence:
[TABLE]
Remark 3.2**.**
It follows from [3, Theorem 1.1] that if G is a complete graph on [n], then S/JG is Cohen-Macaulay of dimension n+1.
If G is a connected graph which is not a complete graph, then κ(G)≥1. Therefore, by [1, Theorems 3.19, 3.20], we get
pdS(S/JG)≥n−2+κ(G)≥n−1. Thus, for any connected graph G, pdS(S/JG)≥n−1 and hence, by Auslander-Buchsbaum formula, 0ptS(S/JG)≤n+1.
We proceed to prove the following lemma which plays an important role.
Lemma 3.3**.**
Let G be a connected graph on the vertex set [n]. Let p=pdS(S/JG). Then βp,p+1S(S/JG)=0 if and
only if G is a complete graph. Moreover, if βi,i+2S(S/JG) is an extremal Betti number, then i=p.
Proof.
By Remark 3.2, p≥n−1.
It follows from [6, Corollary 4.3] that βp,p+1S(S/JG)=pkp+1(G). Therefore, βp,p+1S(S/JG)=0
if and only if G is a complete graph. Now, if possible assume that i<p. Since βi,i+2S(S/JG) is an extremal Betti
number, βp,p+jS(S/JG)=0 for j≥2, which implies that βp,p+1S(S/JG) must be an extremal Betti number. Thus,
G is a complete graph which contradicts [21, Theorem 2.1], as reg(S/JG)≥2. Hence, the assertion follows.
∎
Let M be a finite graded S-module. The Cohen-Macaulay defect, denoted by cmdef(M), is defined by dim(M)−0ptS(M).
A graded S-module M is said to be almost Cohen-Macaulay if
cmdef(M)=1. A graph G is said to be (almost) Cohen-Macaulay if S/JG is
(almost) Cohen-Macaulay.
First, we recall some basic facts about Betti numbers and minimal free resolution. Let R=K[x1,…,xm], R′=K[xm+1,…,xn] and T=K[x1,…,xn] be
polynomial rings. Let I⊆R and J⊆R′ be homogeneous ideals. Then minimal free resolution of T/(I+J) is tensor product of minimal free resolutions of R/I and R′/J. Also, for all i,j,
[TABLE]
Now, we construct almost Cohen-Macaulay graphs.
Theorem 3.4**.**
Let H be a connected graph on the vertex set [n] and G=v∗H be the cone of v on H. Then, 0ptS(S/JG)=0ptSH(SH/JH).
In particular, if H is Cohen-Macaulay, then G is almost Cohen-Macaulay.
Proof.
Assume that 0ptSH(SH/JH)=n+1. Therefore, pdSH(SH/JH)=n−1 and pdS(S/((xv,yv)+JH))=n+1. Also, we have
pdS(S/((xv,yv)+JKn))=n+1 and pdS(S/JKn+1)=n. Since, H is a connected graph, by Lemma 3.3, there exists a j≥2 such that
βn−1,n−1+jSH(SH/JH)=0. Consider, the long exact sequence of Tor corresponding to (1),
[TABLE]
Since βn+1,n+1+jS(S/((xv,yv)+JH))=0, βn+1,n+1+jS(S/JG)=0. Therefore, pdS(S/JG)≥n+1 and
hence by Auslander-Buchsbaum formula, 0ptS(S/JG)≤n+1. Now
using Lemma 2.1 on the short exact sequence (1), we get that
[TABLE]
Hence, 0ptS(S/JG)=n+1. If 0ptSH(SH/JH)<n+1, then by Lemma 2.1, 0ptS(S/JG)=0ptSH(SH/JH). Now, if H is
Cohen-Macaulay, then 0ptS(S/JG)=n+1. It follows from [15, Theorem 4.6] that dim(S/JG)=n+2. Hence, G is an almost Cohen-Macaulay.
∎
Theorem 3.5**.**
Let H be a connected graph on the vertex set [n] and G=v∗H be the cone of v on H. If dim(SH/JH)≥n+2,
then cmdef(S/JG)=cmdef(SH/JH) and otherwise cmdef(S/JG)=cmdef(SH/JH)+1.
Proof.
It follows from [15, Theorem 4.6] that if dim(SH/JH)≥n+2, then dim(S/JG)=dim(SH/JH). Thus, by
Theorem 3.4, cmdef(S/JG)=cmdef(SH/JH). Now, if dim(SH/JH)=n+1, then again by [15, Theorem 4.6], dim(S/JG)=n+2 and hence cmdef(S/JG)=cmdef(SH/JH)+1. ∎
We now show that one can construct graphs with as large Cohen-Macaulay
defect as one wants.
Corollary 3.6**.**
Let H be a connected graph on [n] and q be a positive
integer. If G=Kq∗H, then 0ptS(S/JG)=0ptSH(SH/JH).
In particular, if H is Cohen-Macaulay, then cmdef(S/JG)=q.
Proof.
Let v1,…,vq be vertices of Kq. Observe that Kq∗H=v1∗(⋯∗(vq∗H)⋯). By recursively applying
Theorem 3.4, 0ptS(S/JG)=0ptSH(SH/JH). Now, if H is Cohen-Macaulay, then 0ptS(S/JG)=n+1 and
it follows from [15, Theorem 4.12] that dim(S/JG)=n+q+1. Hence, the assertion follows.
∎
Let G=Kq∗H, then by [15, Theorem 4.12] and Corollary 3.6, if dim(SH/JH)≥n+q+1,
then cmdef(S/JG)=cmdef(SH/JH) otherwise cmdef(S/JG)=n+q+1−dim(SH/JH)+cmdef(SH/JH).
To compute the depth formula for cone of a disconnected graph, we need the following lemma.
Lemma 3.7**.**
Let G be a disconnected graph on the vertex set [n]. Assume that G has atleast two nontrivial components.
Let p=pdS(S/JG). Then βp,p+1S(S/JG)=0. Moreover, if βi,i+2S(S/JG) is an extremal Betti number, then i=p.
Proof.
Let H1,…,Hq be nontrivial connected components of G with q≥2. By Remark 3.2, for each i∈[q],
pdSHi(SHi/JHi)≥∣V(Hi)∣−1, where SHi=K[xv,yv:v∈V(Hi)]. Let m=∑i=1q∣V(Hi)∣. Thus,
p≥m−q. It follows from [6, Corollary 4.3], that βp,p+1S(S/JG)=pkp+1(G). If possible, βp,p+1S(S/JG)=0,
then G has an induced clique of size atleast m−q+1, which is a contradiction. Now, if possible assume that i<p which implies that βp,p+1S(S/JG)
is an extremal Betti number, which is a contradiction as βp,p+1S(S/JG)=0. Hence, the assertion follows.
∎
Remark 3.8**.**
Let G be a disconnected graph on [n]. If 0ptS(S/JG)=n+1, then either G has atleast two nontrivial components or G
has exactly one nontrivial component which is not a complete graph. Moreover, βn−1,nS(S/JG)=0.
We now compute the depth formula for cone of a disconnected graph.
Theorem 3.9**.**
Let G=v∗H, where H is a disconnected graph on [n]. Then
[TABLE]
Proof.
If 0ptSH(S/JH)<n+1, then by using Lemma 2.1 in the short exact sequence (1), we have 0ptS(S/JG)=0ptSH(SH/JH).
Also, if 0ptSH(SH/JH)>n+1, then by virtue of Lemma 2.10ptS(S/JG)=n+2. Now, assume that 0ptSH(SH/JH)=n+1.
Observe that pdSH(SH/JH)=n−1, pdS(S/((xv,yv)+JH))=n+1 and pdS(S/((xv,yv)+JKn))=n+1. By Remark 3.8,
there exists j≥2 such that βn−1,n−1+jSH(SH/JH)=0. Now consider, the long exact sequence of Tor corresponding to (1),
[TABLE]
Since, βn+1,n+1+jS(S/((xv,yv)+JH))=0 and hence βn+1,n+1+jS(S/JG)=0. Therefore, pdS(S/JG)≥n+1 and hence
0ptS(S/JG)≤n+1. Using Lemma 2.1, we have 0ptS(S/JG)≥n+1 and this completes the proof.
∎
Also, if G=Kq∗H, where H is a disconnected graph, then by Theorems 3.4, 3.9, 0ptS(S/JG)=min{0ptSH(SH/JH),n+2}.
Now we compute the Betti numbers of S/Jv∗H in terms of the Betti numbers of SH/JH.
Theorem 3.10**.**
Let H be a graph on the vertex set [n]. Let G=v∗H be the cone of v on H. Then, for i,j,
[TABLE]
where βi−2,i−2+jS(JGS)=0 and βi−1,i−1+jS(JGS)=0, if i−2<0 and i−1<0 respectively.
Proof.
It follows from [6, Corollary 4.3] that βi,i+1S(S/JG)=iki+1(G).
Let U be a clique in G on (i+1)-vertices. Then either v∈U or v∈/U. If v∈/U, then U is a clique in H on (i+1)-vertices.
If v∈U, then U∖{v} is a clique in H on i-vertices. Therefore, ki+1(G)=ki(H)+ki+1(H) and hence
βi,i+1S(S/JG)=i(ki(H)+ki+1(H)).
Now, consider the long exact sequence of Tor modules corresponding to the short exact sequence (1):
[TABLE]
For j=2, the above long exact sequence of Tor gives us
Now let j≥3. Since, reg(S/((xv,yv)+JKn))=reg(S/JKn+1)=1,
[TABLE]
Then for j≥3, Tori,i+jS(JGS,K)≃Tori,i+jS((xv,yv)+JHS,K) and hence by virtue of (2), we have
[TABLE]
which proves our result.
∎
Let G=Kq∗H be the join of a complete graph and H. Then by using the above theorem recursively, one can compute all the Betti numbers of S/JG.
Now, we compute the Betti diagram of the wheel graph. The wheel graph, denoted by Wn, is the cone of v on Cn, n≥4.
Corollary 3.11**.**
Let Wn=v∗Cn be the wheel graph with n≥4. Then reg(S/JWn)=n−2, pdS(S/JWn)=n+2 and the Betti diagram of S/JWn looks like the following:
βi,2i=βi+2,2i+2=(in),βi+1,2i+1=2(in), if i=3,…,n−3,**
βi,i+n−2=(n+1−i)(i−2n+2)+2(i−3n+1), if i=2,…,n−3,**
βn−2,2n−4=(2n)+3(6n+2)+2(6n+1), βn−1,2n−3=2(3n+1)+2(4n)+4(5n+1), βn,2n−2=(2n−1)−1+(4n+2)+2(4n+1), βn+1,2n−1=2(2n−1)−2+2(3n) and βn+2,2n=(2n−1)−1.
Proof.
The assertion follows from [24, Corollary 16] and Theorem 3.10.
∎
Now, we study the position of extremal Betti number of S/JG in terms of the position of extremal Betti number of SH/JH.
Proposition 3.12**.**
Let H be a connected graph on the vertex set [n]. Let G=v∗H be the cone of v on H. If βi,i+jSH(SH/JH) is an
extremal Betti number, then βi+2,i+2+jS(S/JG) is an extremal Betti number and both are equal. Moreover, if βk,k+lS(S/JG) is
an extremal Betti number, then βk−2,k+l−2SH(SH/JH) is an extremal Betti number and βk,k+lS(S/JG)=βk−2,k+l−2SH(SH/JH).
Proof.
Let βi,i+jSH(SH/JH) be an extremal Betti number of SH/JH. Since H is not a complete graph, by Lemma 3.3,
j≥2. If j≥3, then by Theorem 3.10, βi+2,i+2+jS(S/JG)=βi,i+jSH(SH/JH) and for any pair (r,s) with r≥i+2,
s≥j and (r,s)=(i+2,j), βr,r+sS(S/JG)=0. Let p=pdSH(SH/JH). If j=2, then by Lemma 3.3, βp,p+2SH(SH/JH)
is an extremal Betti number. Therefore it follows from Theorem 3.10 that
[TABLE]
By Remark 3.2, p≥n−1, therefore, βp+2,p+4S(S/JG)=βp,p+2SH(SH/JH). Now, let βk,k+lS(S/JG) is an extremal Betti
number. Therefore, by Lemma 3.3, l≥2. Assume that l≥3. If possible, βk−2,k−2+lSH(SH/JH) is not an extremal Betti number.
Thus, there exists r≥k−2 and s≥l such that (r,s)=(k−2,l) and βr,r+sSH(SH/JH)=0. Therefore, by virtue of Theorem 3.10,
βr+2,r+2+sS(S/JG)=0 which is a contradiction. Hence, βk−2,k−2+lSH(SH/JH) is an extremal Betti number and by Theorem 3.10,
βk,k+lS(S/JG)=βk−2,k−2+lSH(SH/JH). Now, if l=2, then by Lemma 3.3, k=pdS(S/JG). It follows from Theorem 3.4
that k=pdSH(SH/JH)+2≥n+1. Therefore, by Theorem 3.10,
[TABLE]
Hence, the assertion follows.
∎
Let H be a connected graph on [n]. Also, let G=Kq∗H. Then by using Proposition 3.12, we conclude that SH/JH admits unique extremal
Betti number if and only if S/JG admits unique extremal Betti number. In particular, if βp,p+rSH(SH/JH) is an extremal Betti number, then
βp+2q,p+2q+rS(S/JG) is an extremal and βp,p+rSH(SH/JH)=βp+2q,p+2q+rS(S/JG).
4. Depth of join of graphs
In this section, we compute the depth of binomial edge ideal of join of two graphs. Let G1 and G2 be graphs on [n1] and [n2], respectively
with n1,n2≥2. We assume that both G1 and G2 are not complete. The join of G1 and G2, denoted by G1∗G2 is
the graph with vertex set [n1]⊔[n2] and the edge set
E(G1∗G2)=E(G1)∪E(G2)∪{{i,j}:i∈[n1],j∈[n2]}. Let G=G1∗G2. It follows from
[14, Propositions 4.1, 4.5, 4.14] that if PT(G) is a minimal prime of JG for some T⊆[n1]⊔[n2], then either T=∅ or
[n1]⊆T or [n2]⊆T. Therefore, by virtue of [5, Theorem 3.2, Corollary 3.9], we have
[TABLE]
Set Q1=(xi,yi:i∈[n2])+JG1, Q2=(xi,yi:i∈[n1])+JG2 and Q3=P∅(G)∩Q2. One can see that
Q2+P∅(G)=(xi,yi:i∈[n1])+JKn2 and Q1+Q3=(xi,yi:i∈[n2])+JKn1. Thus, we have the following short exact sequences:
[TABLE]
and
[TABLE]
Let Si=K[xj,yj:j∈[ni]] for i=1,2. Observe that 0ptS(S/Q1)=0ptS1(S1/JG1), 0ptS(S/(Q2+P∅(G)))=n2+1, 0ptS(S/(Q1+Q3))=n1+1 and
0ptS(S/Q2)=0ptS2(S2/JG2). Thus, using Lemma 2.1 in short exact sequence (3),
First, we give exact formula for depth of binomial edge ideal of join of two connected graphs.
Theorem 4.1**.**
Let G=G1∗G2 be join of G1 and G2, where G1 and G2 be two connected graphs on vertex sets [n1] and [n2] respectively.
Then
[TABLE]
Proof.
First we prove that 0ptS(S/Q3)=0ptS2(S2/JG2). If 0ptS2(S2/JG2)<n2+1, it follows from short exact sequence (3) and Lemma 2.1
that 0ptS(S/Q3)=0ptS2(S2/JG2). Now, we assume that 0ptS2(S2/JG2)=n2+1, by Auslander-Buchsbaum formula,
pdS2(S2/JG2)=n2−1. Since, G2 is not a complete graph, by Lemma 3.3, there exists j≥2 such that
βn2−1,n2−1+jS2(S2/JG2)=0 which implies that βp,p+jS(S/Q2)=0, where p=2n1+n2−1. Note that pdS(S/P∅(G))=n1+n2−1 and
pdS(S/Q2)=2n1+n2−1=pdS(S/((xi,yi:i∈[n1])+JKn2)). Now consider the long exact
sequence of Tor in homological degree p corresponding to the short exact sequence (3)
[TABLE]
Since j≥2, βp,p+jS(S/(P∅(G)+Q2))=0, which further implies that βp,p+jS(S/Q3)=0. Thus,
pdS(S/Q3)≥p and hence 0ptS(S/Q3)≤2n1+2n2−p=n2+1. Hence, by (5),
0ptS(S/Q3)=n2+1=0ptS2(S2/JG2).
Now, if min{0ptSi(Si/JGi):i=1,2}<n1+1, then using Lemma 2.1 in short exact sequence (4), we get the desired
result. Otherwise, by Remark 3.2, we have 0ptS1(S1/JG1)=n1+1=min{0ptSi(Si/JGi):i=1,2} and therefore
pdS1(S1/JG1)=n1−1.
Since, G1 is not a complete graph, by Lemma 3.3, there exists l≥2 such that βn1−1,n1−1+lS1(S1/JG1)=0
which further implies that βn1+2n2−1,n1+2n2−1+lS(S/Q1)=0. Note that, pdS(S/Q1)=n1+2n2−1=pdS(S/((xi,yi:i∈[n2])+JKn1)).
The long exact sequence of Tor in homological degree q=n1+2n2−1 and graded degree q+l corresponding to (4) is
[TABLE]
Since, βq,q+lS(S/Q1)=0, we have βq,q+lS(S/JG)=0. Therefore, pdS(S/JG)≥q and hence 0ptS(S/JG)≤2n1+2n2−q=n1+1.
It follows from (6) that 0ptS(S/JG)≥n1+1. Hence, the desired result follows. ∎
We now illustrate our result by the following example. A block of a graph is a maximal nontrivial connected subgraph with no cut vertex. A connected graph is said to be a block graph if every block of that graph is a complete graph.
Example 4.2**.**
If G1 be a connected block graph and G2=Cn2 with n1≥n2≥4, then by virtue of [3, Theorem 1.1] 0ptS1(S1/JG1)=n1+1.
By [24, Corollary 16] that 0ptS2(S2/JG2)=n2. Hence, 0ptS(S/JG1∗G2)=0ptS2(S2/JG2)=n2.
Now, we move on to study the join of a connected graph and a disconnected graph.
Theorem 4.3**.**
Let G1 be a connected graph on the vertex set [n1] and G2 be a disconnected graph on the vertex set [n2]. Then
[TABLE]
Proof.
We claim that 0ptS(S/Q3)=min{0ptS2(S2/JG2),n2+2}. First assume that n2+1<0ptS2(S2/JG2). Therefore, the
claim follows from the short exact sequence (3) and Lemma 2.1. If n2+1>0ptS2(S2/JG2), it follows from short exact sequence
(3) that 0ptS(S/Q3)=0ptS2(S2/JG2). Now, we assume that 0ptS2(S2/JG2)=n2+1 which implies that
pdS2(S2/JG2)=n2−1. Since, G2 is a disconnected graph, by Remark 3.8, there exists j≥2 such that
βn2−1,n2−1+jS2(S2/JG2)=0 which further implies that βp,p+jS(S/Q2)=0, where p=2n1+n2−1. Note that
pdS(S/P∅(G))=n1+n2−1, pdS(S/Q2)=2n1+n2−1=pdS(S/((xi,yi:i∈[n1])+JKn2)). Now, consider the long exact sequence of Tor (7).
Since βp,p+jS(S/Q2)=0, we have that βp,p+jS(S/Q3)=0. Thus, pdS(S/Q3)≥p and hence 0ptS(S/Q3)≤2n1+2n2−p=n2+1.
Therefore, by (5), 0ptS(S/Q3)=n2+1=0ptS2(S2/JG2). Hence, we have
[TABLE]
Now, if min{0ptS1(S1/JG1),0ptS2(S2/JG2),n2+2}<n1+1, then by applying Lemma 2.1 in short exact sequence (4),
we get the desired result. Otherwise, we have min{0ptS1(S1/JG1),0ptS2(S2/JG2),n2+2}≥n1+1 and hence by Remark 3.2,
0ptS1(S1/JG1)=n1+1. Therefore pdS1(S1/JG1)=n1−1.
Since, G1 is not a complete graph, by Lemma 3.3, there exists l≥2 such that βn1−1,n1−1+lS1(S1/JG1)=0 which
implies that βq,q+lS(S/Q1)=0, where q=n1+2n2−1. Note that pdS(S/Q1)=q=pdS(S/((xi,yi:i∈[n2])+JKn1)).
Consider, the long exact sequence of Tor (8).
Since, βq,q+lS(S/Q1)=0, we have βq,q+lS(S/JG)=0. Therefore, pdS(S/JG)≥q and hence 0ptS(S/JG)≤2n1+2n2−q=n1+1.
It follows from (6) that 0ptS(S/JG)≥n1+1. Hence, the assertion follows.
∎
We now compute the depth of the binomial edge ideal of join of two disconnected graphs.
Theorem 4.4**.**
Let G=G1∗G2 be the join of G1 and G2, where G1 and G2 are disconnected graphs on [n1] and [n2] respectively. Assume that n2≥n1. Then
Now, if min{0ptS1(S1/JG1),0ptS2(S2/JG2),n2+2}<n1+1, then using Lemma 2.1 in short exact sequence (4), we get the desired result.
If min{0ptS1(S1/JG1),0ptS2(S2/JG2),n2+2}=n1+1, then either 0ptS1(S1/JG1)=n1+1 or 0ptS(S/Q3)=n1+1. Now, if 0ptS1(S1/JG1)=n1+1,
then by Auslander-Buchsbaum formula, pdS1(S1/JG1)=n1−1. Therefore, by virtue of Remark 3.8 there exists j≥2 such that
βn1−1,n1−1+jS1(S1/JG1)=0 which implies that βq,q+jS(S/Q1)=0, where q=n1+2n2−1. Note that pdS(S/Q1)=q=pdS(S/((xi,yi:i∈[n2])+JKn1)).
Consider, the long exact sequence of Tor (8) in graded degree q+j.
Since, βq,q+jS(S/Q1)=0, we have βq,q+jS(S/JG)=0. Therefore, pdS(S/JG)≥q and hence 0ptS(S/JG)≤2n1+2n2−q=n1+1. Now, the assertion
follows from (6). Assume now that 0ptS(S/Q3)=n1+1. Since, n1≤n2, 0ptS(S/Q3)=n1+1=0ptS2(S2/JG2)=0ptS(S/Q2). Note that pdS(S/Q3)=q=pdS(S/Q2).
Since, G2 is a disconnected graph and 0ptS2(S2/JG2)=n1+1≤n2+1, either G2 has atleast two nontrivial components or G2 has one nontrivial component which is not complete. In first case, by Lemma 3.7, there exists j≥2 such that β2n2−n1−1,2n2−n1−1+jS2(S2/JG2)=0 which
further implies that βq,q+jS(S/Q2)=0. If G2 has exactly one nontrivial component say H, then pdS2(S2/JG2)=pdSH(SH/JH)=2n2−n1−1, where SH=K[xj,yj:j∈V(H)]. Now, by Lemma 3.3, there exists j≥2 such that β2n2−n1−1,2n2−n1−1+jS2(S2/JG2)=0 which
further implies that βq,q+jS(S/Q2)=0. The long exact sequence of Tor corresponding to (3) in homological degree q and graded degree q+j is
[TABLE]
Therefore, βq,q+jS(S/Q3)=0. Thus, it follows from (8) that βq,q+jS(S/JG)=0. Therefore, pdS(S/JG)≥q and hence,
0ptS(S/JG)≤2n1+2n2−q=n1+1. Now, along with (6), we get the assertion.
Also, if min{0ptS1(S1/JG1),0ptS2(S2/JG2),n2+2}>n1+1, then again using Lemma 2.1 in the short exact sequence (4), 0ptS(S/JG)=n1+2. Hence, the desired result follows.
∎
As an immediate consequence, we obtain the depth of complete multipartite graph.
Corollary 4.5**.**
Let G=Kn1,⋯,nk be a complete multipartite graph with 2≤n1≤⋯≤nk. Then 0ptS(S/JG)=n1+2.
5. Construction of Graph
In this section, we construct a graph G such that reg(S/JG)=r and the number of extremal Betti numbers of S/JG is b,
where 1≤b<r. We now set some notation for the rest of this section. Let G1 and G2 be two connected graphs which are
not complete on the vertex sets [n1] and [n2], respectively. Let pi=pdSi(Si/JGi) and ri=reg(Si/JGi) for i=1,2.
By Remark 3.2, pi≥ni−1, for i=1,2.
Lemma 5.1**.**
Let G1 and G2 be graphs on [n1] and [n2], respectively. Let G=G1∗G2. If reg(S/JG)=2, then S/JG admits unique extremal Betti number.
We consider the long exact sequence of Tor corresponding to the exact sequence (3)
[TABLE]
where Q2=(xi,yi:i∈[n1])+JG2 and Q3=P∅(G)∩Q2.
Now, we prove that extremal Betti numbers of S/Q3 and S/Q2 coincide in terms of position and value.
Lemma 5.2**.**
Let G=G1∗G2 be the join graph on [n1]⊔[n2]. If βk,k+lS2(JG2S2) is an extremal
Betti number, then βk+2n1,k+2n1+lS(Q3S) is an extremal Betti number. Moreover, extremal Betti number
of S/Q3 is of the form βk+2n1,k+2n1+lS(Q3S), where βk,k+lS2(JG2S2) is an extremal Betti number.
Proof.
It follows from proof of Theorem 4.1 that 0ptS(S/Q3)=0ptS2(S2/JG2). Thus, pdS(S/Q3)=p2+2n1. Since, pdS(S/P∅(G))=n1+n2−1
and pdS(S/((xi,yi:i∈[n1])+JKn2))=2n1+n2−1, by Lemma 3.3,
[TABLE]
Also, Torp2+2n1+1,p2+2n1+1S(S/((xi,yi:i∈[n1])+JKn2))=0. By considering
the long exact sequence of Tor (9) in homological degree p2+2n1 and graded degree p2+2n1+1, we get, βp2+2n1,p2+2n1+1S(Q3S)=0. Now, let
βk,k+lS2(JG2S2) is an extremal Betti number. By virtue of Lemma 3.3, l≥2 and therefore the long exact sequence of
Tor (9) in homological degree p=k+2n1 and graded degree p+l is
[TABLE]
Now if l>2, then we have
[TABLE]
Thus, βp,p+lS(Q3S)=0. Let (s,t)=(p,l) with s≥p,t≥l. Taking homological degree s≥p and graded degree s+t≥p+l in (9), we have
[TABLE]
Note that
s−2n1≥k,t≥l and (s−2n1,t)=(k,l). Therefore,
[TABLE]
Hence,
βp,p+lS(Q3S) is an extremal Betti number.
Now we assume that l=2. It follows from Lemma 3.3 that k=p2. Note that
pdS(S/((xi,yi:i∈[n1])+JKn2))=2n1+n2−1≤p2+2n1. Now, consider the long exact sequence of Tor (9) in homological degree
p=p2+2n1, Torp2+2n1,p2+2n1+jS(Q3S,K)≃Torp2,p2+jS2(JG2S2,K), for j≥2.
Since, βp2,p2+2S2(JG2S2) is an extremal Betti number, βp,p+2S(Q3S) is an extremal Betti number.
Now, let βi,i+jS(Q3S) be an extremal Betti number. Consider, the long exact sequence of Tor (9) in homological degree i and graded degree i+j.
If j>2, then
[TABLE]
Therefore, βi−2n1,i−2n1+jS2(JG2S2)=0. Now, if for some s≥i−2n1,t≥j with (s,t)=(i−2n1,j), βs,s+tS2(JG2S2)=0.
Then, βs+2n1,s+2n1+tS(Q3S)=0 and s+2n1≥i,t≥j and (s+2n1,t)=(i,j) which is a contradiction. Hence, βi−2n1,i−2n1+jS2(JG2S2)
is an extremal Betti number. If j=2, then by Lemma 3.3, i=p2+2n1. For any l≥2, we have Torp2+2n1,p2+2n1+lS(Q3S,K)≃Torp2,p2+lS2(JG2S2,K).
Therefore, βp2,p2+2S2(JG2S2) is an extremal Betti number. This completes the proof.
∎
It follows from above theorem that reg(S/Q3)=reg(S/Q2)=reg(S2/JG2).
Assume that 2≤r1≤r2 and p1≤p2. Since, G1 is a connected graph with r1≥2, by Lemma 3.3,
if βk,k+lS1(JG1S1) is an extremal Betti number, then l≥2.
We now consider long exact sequence of Tor in homological degree k and graded degree k+l≥k+2 corresponding to the exact sequence (4),
[TABLE]
where Q1=(xi,yi:i∈[n2])+JG1.
Lemma 5.3**.**
Let G=G1∗G2 be the join graph on [n1]⊔[n2]. Suppose 0ptS1(S1/JG1)<0ptS2(S2/JG2) i.e., p2+2n1<p1+2n2. If βk,k+lS(S/JG) is an extremal
Betti number, then
[TABLE]
Proof.
Note that pdS(S/((xi,yi:i∈[n2])+JKn1))=2n2+n1−1<2n2+p1+1, by Remark 3.2. Therefore,
Torp1+2n2+1,p1+2n2+1+jS((xi,yi:i∈[n2])+JKn1S,K)=0, for j≥1. If l=2, by Lemma 3.3, k=pdS(S/JG)=p1+2n2. Also if l>2, then
By [21, Theorem 2.1], reg(S/JKn)=1 and hence S/JKn admits unique extremal Betti number. So for r=b=1, consider G=Kn.
Now assume that r≥2. It follows from Betti diagram and Lemmas 3.3, 3.7 that b≤r−1.
Theorem 5.4**.**
Let r and b be two positive integers with 1≤b≤r−1. Then there exists a graph G=Gr,b such that reg(S/JG)=r and
the number of extremal Betti numbers of S/JG is b.
Proof.
Take G=Gr,b=Pr−b+2∗⋯∗Pr+1. Note that n=∣V(G)∣=br−2b(b−3) and by recursively applying Theorem 4.1, 0ptS(S/JG)=r−b+3. Now, by Auslander-Buchsbaum
formula, p=pdS(S/JG)=(2b−1)r−(b−1)(b−3). It follows from [22, Theorem 2.1] that reg(S/JG)=r. We now prove that extremal Betti numbers of S/JG are precisely
βp−i,p+r−b+1S(JGS)=1, for 0≤i≤b−1.
We proceed by induction on b. If b=1, then G=Gr,1=Pr+1. By [3, Corollary 1.2], JG is complete intersection ideal, S/JG has a unique
extremal Betti number, βr,2rS(JGS)=1. Now, assume that b>1 and extremal Betti numbers of S2/JG2 are precisely
βp2−i,p2+r−b+2S2(JG2S2) for 0≤i≤b−2, where G2=Gr,b−1=Pr−b+3∗⋯∗Pr+1, S2=K[xj,yj:j∈V(G2)] and p2=pdS2(S2/JG2).
Observe that n2=∣V(G2)∣=(b−1)r−2(b−1)(b−4). Also, by Theorem 4.1, 0ptS2(S2/JG2)=r−b+4. Thus, by Auslander-Buchsbaum formula, p2=2n2−r+b−4=(2b−3)r−(b−2)(b−4).
Set G1=Pr−b+2, n1=∣V(G1)∣=r−b+2 and S1=K[xj,yj:j∈V(G1)]. By [3, Corollary 1.2], p1=pdS1(S1/JG1)=r−b+1=r1=reg(S1/JG1). Note that G=Gr,b=G1∗G2.
Since JG1 is complete intersection ideal, βp1,p1+r−b+1S1(JG1S1)=1 is the extremal Betti number of S1/JG1.
Therefore, βp1+2n2,p1+2n2+r−b+1S(Q1S) is unique extremal Betti number of S/Q1. Note that p1+2n2=(2b−1)r−(b−1)(b−3)=p.
Let βk,k+lS(S/JG) is an extremal Betti number. Now, by Lemma 5.3, we have
[TABLE]
It follows from Lemma 5.2 that extremal Betti numbers of S/Q3 are βp2+2n1−i,p2+2n1+r−b+2S(Q3S),
for 0≤i≤b−2. Note that p2+2n1=p2+2(r−b+2)=p−1. Therefore, extremal Betti numbers of S/Q3 are
βp−i−1,p+r−b+1S(Q3S)=1, for 0≤i≤b−2. Since b≤r−1, r−b+1≥2. So, for j≥r−b+2≥3 and 1≤k≤p,
[TABLE]
Therefore, βp−i−1,p+r−b+1S(JGS)=1, for 0≤i≤b−2 are extremal Betti numbers of S/JG. Now, it remains to prove that βp,p+r−b+1S(JGS) is an extremal Betti number.
Taking long exact sequence of Tor (9) in homological degree p and graded degree p+j=p+r−b+1≥p+2, we have
Torp,p+r−b+1S(JGS)≃Torp,p+r−b+1S(Q1S)≃Torp1,p1+r−b+1S1(JG1S1).
Therefore, βp,p+r−b+1S(JGS)=1 is also an extremal Betti number. Hence the number of extremal Betti numbers of S/JG is b and the extremal Betti
number are of the form βp−i,p+r−b+1S(JGS)=1, for 0≤i≤b−1.
∎
Observe that the projective dimension of S/JGr,b is very large. Hence the following question arises:
Question 5.5**.**
Does there exist a graph G such that the projective dimension is bounded by a linear function of b and r, where r=reg(S/JG) and b is the number of extremal Betti numbers of S/JG?
Acknowledgement: The authors are grateful to their advisor A. V. Jayanthan for
constant support, valuable ideas and suggestions. The first author thanks the National Board
for Higher Mathematics, India for the financial support. The second author thanks
University Grants Commission, Government of India for the financial support.
Bibliography24
The reference list from the paper itself. Each links out to its DOI / PubMed record.
1[1] Arindam Banerjee and Luis Núñez Betancourt. Graph connectivity and binomial edge ideals. Proc. Amer. Math. Soc. , 145(2):487–499, 2017.
2[2] Hernán de Alba and Do Trong Hoang. On the extremal Betti numbers of the binomial edge ideal of closed graphs. Math. Nachr. , 291(1):28–40, 2018.
3[3] Viviana Ene, Jürgen Herzog, and Takayuki Hibi. Cohen-Macaulay binomial edge ideals. Nagoya Math. J. , 204:57–68, 2011.
4[4] Viviana Ene and Andrei Zarojanu. On the regularity of binomial edge ideals. Math. Nachr. , 288(1):19–24, 2015.
5[5] Jürgen Herzog, Takayuki Hibi, Freyja Hreinsdóttir, Thomas Kahle, and Johannes Rauh. Binomial edge ideals and conditional independence statements. Adv. in Appl. Math. , 45(3):317–333, 2010.
6[6] Jürgen Herzog, Dariush Kiani, and Sara Saeedi Madani. The linear strand of determinantal facet ideals. Michigan Math. J. , 66(1):107–123, 2017.
7[7] Jürgen Herzog and Giancarlo Rinaldo. On the extremal Betti numbers of binomial edge ideals of block graphs. Electron. J. Combin. , 25(1):Paper 1.63, 10, 2018.
8[8] Takayuki Hibi, Kyouko Kimura, and Kazunori Matsuda. Extremal Betti numbers of edge ideals. Arch. Math. (Basel) , 113(2):149–155, 2019.