This paper investigates the algebraic properties of Cameron--Walker graphs, showing that their associated edge ideals have a specific invariant value and satisfy a key equality relating regularity, dimension, and depth.
Contribution
It establishes that all Cameron--Walker graphs have an $a$-invariant of zero and identifies conditions under which their edge ideals satisfy a fundamental algebraic equality.
Findings
01
All Cameron--Walker graphs have $a$-invariant equal to zero.
02
A class of Cameron--Walker graphs satisfies $s - r = d - e$.
03
The paper links graph structure to algebraic invariants of their edge ideals.
Abstract
Let S be the polynomial ring over a field K and IâS a homogeneous ideal. Let h(S/I,λ) be the h-polynomial of S/I and s=degh(S/I,λ) the degree of h(S/I,λ). It follows that the inequality sârâ€dâe, where r=reg(S/I), d=dimS/I and e=depthS/I, is satisfied and, in addition, the equality sâr=dâe holds if and only if S/I has a unique extremal Betti number. We are interested in finding a natural class of finite simple graphs G for which S/I(G), where I(G) is the edge ideal of G, satisfies sâr=dâe. Let a(S/I(G)) denote the a-invariant of S/I, i.e., a(S/I(G))=sâd. One has a(S/I(G))â€0. In the present paper, by showing the fundamental fact that every Cameron--Walker graph G satisfies a(S/I(G))=0, a class of Cameron--Walker graphs G for which S/I(G)âŠ
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
Regularity and a-invariant of CameronâWalker graphs
Takayuki Hibi, Kyouko Kimura, Kazunori Matsuda and Akiyoshi Tsuchiya
Takayuki Hibi,
Department of Pure and Applied Mathematics,
Graduate School of Information Science and Technology,
Osaka University, Suita, Osaka 565-0871, Japan
Akiyoshi Tsuchiya,
Department of Pure and Applied Mathematics,
Graduate School of Information Science and Technology,
Osaka University, Suita, Osaka 565-0871, Japan
Let S be the polynomial ring over a field K and IâS a homogeneous ideal.
Let h(S/I,λ) be the h-polynomial of S/I and s=degh(S/I,λ) the degree of h(S/I,λ).
It follows that the inequality sârâ€dâe,
where r=reg(S/I), d=dimS/I and e=depthS/I,
is satisfied and, in addition, the equality sâr=dâe holds
if and only if S/I has a unique extremal Betti number.
We are interested in finding a natural class of finite simple graphs G for which S/I(G),
where I(G) is the edge ideal of G, satisfies sâr=dâe.
Let a(S/I(G)) denote the a-invariant of S/I,
i.e., a(S/I(G))=sâd. One has a(S/I(G))â€0.
In the present paper, by showing the fundamental fact that every CameronâWalker graph G satisfies a(S/I(G))=0,
a classification of CameronâWalker graphs G for which S/I(G) satisfies sâr=dâe will be exhibited.
In the current trends on combinatorial and computational commutative algebra,
the study on regularity of edge ideals of finite simple graphs becomes
fashionable and many papers including
[3, 8, 9, 17, 21]
have been published.
In the present paper we are interested in the regularity and the h-polynomials of edge ideals.
Let S=K[x1â,âŠ,xnâ] denote the polynomial ring in
n variables over a field K with each degxiâ=1 and
IâS a homogeneous ideal of S with dimS/I=d.
The Hilbert series H(S/I, λ) of S/I is of the form
[TABLE]
where each hiââZ ([5, Proposition 4.4.1]).
We say that
[TABLE]
with hsâî =0 is the h-polynomial of S/I.
We call the difference
degh(S/I, λ)âdimS/I
the a-invariant ([5, Definition 4.4.4]) of S/I and denote it by a(S/I).
It is known that a(S/I)â€0 if I is a squarefree monomial ideal.
Let
[TABLE]
be the minimal graded free resolution of S/I over S, where p is the projective dimension of S/I.
The (CastelnuovoâMumfordâ) regularity of S/I is
[TABLE]
The inequality
[TABLE]
is well known ([20, Corollary B.4.1]) and its proof is easy.
In fact, since [5, Lemma 4.1.13] says that
[TABLE]
it follows that degh(S/I, λ)â€p+reg(S/I)ân+dimS/I. Furthermore, since
nâp=depth(S/I) by AuslanderâBuchsbaum Theorem,
the inequality (0.1) follows.
In addition, the equality
[TABLE]
holds if and only if ÎČp,p+reg(S/I)â(S/I)î =0, in other words,
if and only if
S/I has a unique extremal Betti number ([10, Definition 4.3.13]).
In particular, the equality (â) holds
if S/I is CohenâMacaulay by [2, Lemma 3]
or I has a pure resolution ([5, p. 153]).
Let G be a finite simple graph (i.e. a graph with no loop and no multiple edge) on the vertex set
V(G)={x1â,x2â,âŠ,xnâ} and its edge set E(G).
Set S=K[V(G)].
The edge ideal of G is
[TABLE]
It is natural to ask for which graph G, its edge ideal I(G) satisfies
a(S/I(G))=0 or the equality (â).
In the present paper we focus on CameronâWalker graphs.
Let us recall the definition of a CameronâWalker graph.
Let im(G) (resp. m(G)) denote the induced matching number
(resp. matching number) of G,
see [11, p.258].
Then
for any finite simple graph G, one has
[TABLE]
by virtue of [9, Theorem 6.7] and [15, Lemma 2.2].
Cameron and Walker [6, Theorem 1] (see also [11, Remark 0.1])
characterized a finite connected simple graph G satisfying im(G)=m(G).
A CameronâWalker graphG is a graph satisfying im(G)=m(G)
which is neither a star graph nor a star triangle;
see Section 1
for more detail.
In [11, 19], CameronâWalker graphs have been studied from a viewpoint of commutative algebra.
In the present paper, we first prove a(S/I(G))=0
for every CameronâWalker graph G (Theorem 1.1)
in Section 1.
We next give a classification of CameronâWalker graphs G
whose edge ideal I(G) satisfies the equality (â)
(Theorem 2.2)
in Section 2.
We also provide some classes of graphs
other than CameronâWalker graphs satisfying
(â) (Proposition 2.10).
In general, there is no relationship between the degree
of the h-polynomial and the regularity even for edge ideals;
see [13].
However we prove in Section 3 that
for a CameronâWalker graph G,
the inequality degh(S/I(G),λ)â„reg(S/I(G)) holds.
Moreover we characterize the CameronâWalker graphs G which satisfy
the equality (Theorem 3.1).
1. a-invariant of CameronâWalker graphs
In this section, we show
Theorem 1.1**.**
Let G be a CameronâWalker graph.
Then a(K[V(G)]/I(G))=0.
a star graph, i.e. a graph joining some paths of length 1 at one common vertex (see Figure 2);
âą
a star triangle, i.e. a graph joining some triangles at one common vertex (see Figure 3);
âą
a connected finite graph consisting of a connected bipartite graph with vertex partition
{v1â,âŠ,vmâ}âȘ{w1â,âŠ,wnâ} such that there is at least one leaf edge
attached to each vertex viâ and that there may be possibly some pendant triangles
attached to each vertex wjâ.
Here a leaf edge is an edge meeting a vertex of degree 1 and a pendant triangle is a triangle
whose two vertices have degree 2 and the rest vertex has degree more than 2.
We say that a finite connected simple graph G is CameronâWalker
if im(G)=m(G) and if G is neither a star graph nor a star triangle.
Remark 1.2**.**
One can consider a star graph G with âŁV(G)âŁâ„3
as a CameronâWalker graph consisting of bipartite graph K1,1â
with some leaf edges and without pendant triangle.
Hence claims for CameronâWalker graph in the below are also true
for such a star graph.
Note that for a CameronâWalker graph G,
the regularity of K[V(G)]/I(G) is equal to im(G) (equivalently, m(G)).
Let G be a CameronâWalker graph.
In what follows we use the following labeling on vertices of G;
see Figure 1:
[TABLE]
where
{v1â,âŠ,vmâ}âȘ{w1â,âŠ,wnâ} is a vertex partition
of a connected bipartite subgraph of G,
xk(i)â (i=1,âŠ,m; k=1,âŠ,siâ) is a vertex such that
{viâ,xk(i)â} is a leaf edge,
and yâ,1(j)â,yâ,2(j)â
(j=1,âŠ,n; â=1,âŠ,tjâ)
are vertices which together with wjâ form a pendant triangle.
Note that siââ„1 and tjââ„0.
Let G be a CameronâWalker graph as in Figure 1.
Then
[TABLE]
Before giving a proof of Proposition 1.3,
several lemmata will be prepared.
Let IâS be a monomial ideal of S and let x be a variable of S
which appears in some monomial belonging to the unique minimal system of
monomial generators of I.
Then, by the additivity of Hilbert series on the exact sequence
0âS/I:(x)(â1) â x âS/IâS/I+(x)â0,
one has
Lemma 1.4**.**
[TABLE]
Let G be a finite simple graph on the vertex set V(G)={x1â,âŠ,xnâ} with the edge set E(G).
For WâV(G), the induced subgraphGWâ is the subgraph of G
such that V(GWâ)=W and E(GWâ)={{xiâ,xjâ}âE(G):xiâ,xjââW}.
For xvââV(G), let NGâ(xvâ) denote the neighborhood of xvâ
and let NGâ[xvâ]=NGâ(xvâ)âȘ{xvâ}.
Then I(G)+(xvâ)=(xvâ)+I(GV(G)â{xvâ}â) and
I(G):(xvâ)=(xiâ:xiââNGâ(xvâ))+I(GV(G)âNGâ[xvâ]â).
Hence
By using Lemma 1.5 again,
one has the Hilbert series of K[V(G)]/I(G)
when G is a star graph or a star triangle.
For sâ„1, we denote by Gsstar(xvâ)â,
the star graph joining s paths of length 1 at the common vertex xvâ;
see Figure 2.
Lemma 1.7**.**
Let sâ„1 be an integer. Then
[TABLE]
In particular,
[TABLE]
For tâ„1, we denote by Gtâł(xvâ)â,
the star triangle joining t triangles at the common vertex xvâ;
see Figure 3.
Lemma 1.8**.**
Let tâ„1 be an integer. Then
[TABLE]
In particular,
[TABLE]
and dimK[V(Gtâł(xvâ)â)]/I(Gtâł(xvâ)â)=t.
Let S1â and S2â be
polynomial rings over a field K.
Let I1â be a nonzero homogeneous ideal of S1â and I2â that of S2â.
Write S for S1ââKâS2â
and regard I1â+I2â as homogeneous ideals of S. Then
[TABLE]
In particular,
[TABLE]
[TABLE]
Let G be a disconnected graph whose connected components are
G1â,âŠ,Grâ.
Then I(G)=âi=1râI(Giâ).
Thus, by virtue of Lemma 1.9, one has
Lemma 1.10**.**
Under the notation as above,
[TABLE]
[TABLE]
here we regard K[V(Giâ)]/I(Giâ) as a 1-dimensional polynomial ring
if Giâ is an isolated vertex.
Now we are in the position to prove Proposition 1.3.
Let G be a CameronâWalker graph as in Figure 1.
We prove the equality (1.1)
by using induction on m+n.
First, we assume that m+n=2.
Then m=n=1.
If t1â=0, then G=Gs1â+1star(v1â)â.
Hence the equality (1.1) follows by Lemma 1.7.
Next assume t1â>0 .
We will show
[TABLE]
Note that
âą
GV(G)â{v1â}â consists of s1â isolated vertices
and a star triangle Gt1ââł(w1â)â;
âą
GV(G)âNGâ[v1â]â consists of t1â star graphs
G1star(y1,1(1)â)â,âŠ,G1star(yt1â,1(1)â)â;
Therefore one has degh(K[V(G)]/I(G), λ)=dimK[V(G)]/I(G)=s1â+t1â, as desired.
Next, we assume that m+n>2.
(First Step.) Let m=1 and n>1.
Suppose that there exists 1â€ââ€n such that tââ=0.
We may assume â=n.
Then we will show
[TABLE]
Since tnâ=0, {v1â,wnâ} is a leaf edge.
Hence we can regard G as a CameronâWalker graph such that its bipartite part is
the star graph Gnâ1star(v1â)â and
the vertex v1â has s1â+1 leaf edges.
Thus, by induction hypothesis, one has
[TABLE]
as desired.
Next, suppose that tjâ>0 for all 1â€jâ€n.
We will show
[TABLE]
Note that
âą
GV(G)â{v1â}â consists of s1â isolated vertices and
n star triangles Gt1ââł(w1â)â,âŠ,Gtnââł(wnâ)â,
âą
GV(G)âNGâ[v1â]â consists of âj=1nâtjâ star graphs
G1star(yâ,1(k)â)â for 1â€kâ€n and 1â€ââ€tkâ;
GV(G)â{w1â}â consists of m+t1â star graphs Gs1âstar(v1â)â,âŠ,Gsmâstar(vmâ)â
and G1star(y1,1(1)â)â,âŠ,G1star(yt1â,1(1)â)â,
âą
GV(G)âNGâ[w1â]â consists of âi=1mâsiâ isolated vertices;
Hence degh(K[V(G)]/I(G), λ)=âi=1mâsiâ+t1â+max{t1â,1}ât1â=âi=1mâsiâ+max{t1â,1}.
Therefore, one has
[TABLE]
as desired.
(Third Step.) Let m>1 and n>1.
Suppose that there exists 1â€ââ€n
such that {vmâ,wââ} is a leaf edge.
We may assume â=n.
Then tnâ=0.
We will show
[TABLE]
Note that we can regard G as a CameronâWalker graph such that
its bipartite part has bipartition
{v1â,âŠ,vmâ}âȘ{w1â,âŠ,wnâ1â},
the vertex viâ has siâ leaf edges for all 1â€iâ€mâ1 and
the vertex vmâ has smâ+1 leaf edges.
Thus, by induction hypothesis, one has
[TABLE]
as desired.
Next, suppose that {vmâ,wââ} is not a leaf edge
for all 1â€ââ€n.
Then GV(G)â{vmâ}â consists of
We give an example after the proof; see Example 1.11.
Note that each graph of type (a2) can be considered as
a CameronâWalker induced subgraph.
Also note that each induced star graph Gsiâstar(viâ)â
(resp. induced pendant triangle Gtjââł(wjâ)â)
appears in (a2) or (a4) (resp. (a3) or (a4)) as a (sub)graph.
Hence by virtue of Lemmata 1.7, 1.8, 1.10 and induction hypothesis, one has
[TABLE]
and
[TABLE]
On the other hand, GV(G)âNGâ[vmâ]â consists of
Note that each induced star graph Gsiâstar(viâ)â
appears in (b1) or (b3) as a (sub)graph.
Also note that the star graphs
G1star(yâ,1(j)â)â, 1â€ââ€tjâ
of type (b2) are the edges of the pendant triangle
Gtjââł(wjâ)â and the total contributions of these graphs
to the degree of h-polynomial and the dimension are both tjâ.
Hence, by virtue of Lemmata 1.7, 1.10 and induction hypothesis, it follows that
We give an example of CameronâWalker graph with m>1 and n>1
which would be helpful to understand (Third Step.) of the proof of
Proposition 1.3.
Example 1.11**.**
Let G be the following CameronâWalker graph:
G=vmâ=v3â
Then the induced subgraph GV(G)â{vmâ}â is as follows.
GV(G)â{vmâ}â=
Also the induced subgraph GV(G)âNGâ[vmâ]â is as follows.
GV(G)âNGâ[vmâ]â=
2. CameronâWalker graphs with the equality (â)
As noted in Introduction, for an arbitrary finite simple graph G,
one has
[TABLE]
where we set S=K[V(G)].
Then it is natural to ask for which graph G satisfies the equality:
[TABLE]
Recall that the equality (â) holds if and only if
S/I(G) has a unique extremal Betti number.
Hence when I(G) has a pure resolution
([5, p. 153]), the equality (â) holds.
Moreover by ([2, Lemma 3]),
it follows that the equality (â) holds
if S/I(G) is CohenâMacaulay.
In this section, we give a classification of CameronâWalker graphs G
with the equality (â).
Throughout this section, let G be a CameronâWalker graph
whose labeling of vertices is as in Figure 1.
By Theorem 1.1, the equality (â) holds if and only if
depth(S/I(G))=reg(S/I(G)).
Both of these invariants have combinatorial explanations.
The regularity is equal to the induced matching number
(or the matching number) of G:
reg(S/I(G))=âj=1nâtjâ+m.
In order to state about the depth, we need some definitions.
For a subset AâV(G),
we set NGâ(A)=âvâAâNGâ(v)âA.
A subset AâV(G) is said to be independent if
{xiâ,xjâ}â/E(G) for any xiâ,xjââA.
We denote by i(G), the minimum cardinality of independent sets A
with AâȘNGâ(A)=V(G).
Then depth(S/I(G))=i(G);
see [11, Corollary 3.7].
We have the following estimation for i(G).
Lemma 2.1**.**
Let G be a CameronâWalker graph
whose labeling of vertices is as in Figure 1.
Then
[TABLE]
Moreover if the bipartite part of G is the complete bipartite graph,
then
[TABLE]
Proof.
The upper bound is clear. We prove the lower bound.
If viâî âAbipâ, then x1(i)â,âŠ,xsiâ(i)ââAâČ;
âą
If wjâî âAbipâ, then yâ,1(j)ââAâČ
or yâ,2(j)ââAâČ for all 1â€ââ€tjâ.
Hence one has
[TABLE]
Thus i(G)â„m+âŁ{j:tjâ>0}âŁ.
When the bipartite part of G is the complete bipartite graph,
one has either Abipââ{v1â,âŠ,vmâ} or
Abipââ{w1â,âŠ,wnâ}.
For the former case, since siââ„1 for all i, it follows that
âŁAâŁâ„âj=1nâtjâ+m.
For the latter case, one has
âŁAâŁâ„âi=1mâsiâ+n
because wjââAbipâ if tjâ=0.
It then follows that
[TABLE]
Combining this with the upper bound, one has the equality.
By virtue of this lemma, we can give a classification of CameronâWalker graphs G
satisfying the equality (â).
Theorem 2.2**.**
Let G be a CameronâWalker graph
whose labeling of vertices is as in Figure 1 and
Gbipâ the bipartite part of G.
Then S/I(G) satisfies the equality (â) if and only if
[TABLE]
holds for all Vâ{v1â,âŠ,vmâ}.
Proof.
Assume that there exists a subset Vâ{v1â,âŠ,vmâ} satisfying
[TABLE]
Let
[TABLE]
[TABLE]
Then A is an independent set with AâȘNGâ(A)=V(G) and
When we use Theorem 2.2, we only need to check the inequality (2.1) for Vâ{v1â,âŠ,vmâ} with NGbipââ(wjâ)âV for some 1â€jâ€n.
Indeed, let V be a subset of {v1â,âŠ,vmâ} such that NGbipââ(wjâ)î âV for all 1â€jâ€n.
Then the inequality (2.1) for V is
â1â€iâ€m,viââVâsiââ„âŁVâŁ,
which always holds since siââ„1 for all 1â€iâ€m.
2. (2)
Considering the inequality (2.1) for V={v1â,âŠ,vmâ},
it follows that âi=1mâsiâ+nâ„âj=1nâtjâ+m
holds if S/I(G) satisfies the equality (â).
Let G be a CameronâWalker graph
whose labeling of vertices is as in Figure 1.
Suppose that tjââ€1 for all 1â€jâ€n.
Then S/I(G) satisfies the equality (â).
Remark 2.5**.**
Let G be a CameronâWalker graph whose labeling of vertices is
as in Figure 1.
Then S/I(G) is Cohen-Macaulay if and only if
siâ=1 for all 1â€iâ€m
and tjâ=1 for all 1â€jâ€n ([11, Theorem 1.3]).
Hence the class of graphs in Corollary 2.4 contains all CohenâMacaulay
CameronâWalker graphs.
Let G be a CameronâWalker graph
whose bipartite part is the complete bipartite graph.
We label the vertices of G as in Figure 1.
Then S/I(G) satisfies the equality (â) if and only if
âi=1mâsiâ+nâ„âj=1nâtjâ+m.
Proof.
Since NGbipââ(wjâ)={v1â,âŠ,vmâ} for all 1â€jâ€n, the claim follows from Theorem 2.2 and Remark 2.3.
In general, one has dimS/I(G)â„depth(S/I(G)).
Then it is natural to ask the following
Question 2.7**.**
Given arbitrary integers d,e with dâ„eâ„1,
are there a CameronâWalker graph G
satisfying dimS/I(G)=d and depth(S/I(G))=e?
As an application of Corollary 2.4,
we give a complete answer for Question 2.7.
We first note about the depth.
Proposition 2.8**.**
Let G be a CameronâWalker graph. Then depthS/I(G)â„2.
Moreover depthS/I(G)=2 if and only if G can be considered as
one of the following CameronâWalker graphs:
(e1)
m=2* and tjâ=0 for all 1â€jâ€n;*
2. (e2)
m=n=1* and t1â=1;*
3. (e3)
m=n=1, t1ââ„2, and s1â=1.
Here, we use labeling of vertices of G
as in Figure 1.
Proof.
Assume that G is a CameronâWalker graph with depth(S/I(G))=1.
By Lemma 2.1, one has m=1 and tjâ=0 for all
1â€jâ€n.
Then G is a star graph but this is a contradiction
since star graphs are not CameronâWalker by definition.
Next assume that G is a CameronâWalker graph with depth(S/I(G))=2.
By Lemma 2.1, one has
âą
m=2 and tjâ=0 for all 1â€jâ€n, or
âą
m=1 and tjâ=0 except for one j.
We consider the case m=1. Since G is not a star graph, there exists
just one j with tjâî =0, say j=1.
When nâ„2, since m=1 and tjâ=0 for 2â€jâ€n,
G can be considered as a CameronâWalker graph whose bipartite subgraph is
of type (1,1) such that v1â has s1â+(nâ1) leaf edges and w1â has
one pendant triangle. Thus we may assume n=1.
If t1ââ„2, then i(G)=depthS/I(G)=2 implies that s1â=1.
Hence the assertion follows.
The converse is easy.
Since any CameronâWalker graph G satisfies depthS/I(G)â„2,
we only consider the case eâ„2 in Question 2.7.
By virtue of Corollary 2.4, we can give a CameronâWalker graph
G satisfying the properties in Question 2.7 with the equality
(â).
Corollary 2.9**.**
Given arbitrary integers d,e with dâ„eâ„2,
there exists a CameronâWalker graph G with the equality (â)
satisfying dimS/I(G)=d and depth(S/I(G))=e.
Proof.
We use the labeling of vertices of a CameronâWalker graph
as in Figure 1.
â** The case d>e:**
Let G be the CameronâWalker graph with m=e, n=1,
s1â=âŻ=seâ1â=1, seâ=dâe, and t1â=0.
Then dim(S/I(G))=âi=1eâsiâ+max{t1â,1}=d.
Also, A:={v1â,âŠ,veâ}
is an independent set of V(G) with AâȘNGâ(A)=V(G)
which gives i(G). Thus one has depthS/I(G)=i(G)=âŁAâŁ=e.
â** The case d=e:**
Let G be the CameronâWalker graph with m=dâ1, n=1,
s1â=âŻ=sdâ1â=1, and t1â=1.
Then dim(S/I(G))=âi=1dâ1âsiâ+max{t1â,1}=d.
Also, A:={x1(1)â,âŠ,xdâ1(1)â}âȘ{w1â}
is an independent set of V(G) with AâȘNGâ(A)=V(G)
which gives i(G). Thus one has depthS/I(G)=i(G)=âŁAâŁ=e.
Finally of the section, we provide some classes of graphs G
which satisfy the equality (â)
other than CameronâWalker graphs.
Proposition 2.10**.**
Let G be the one of the following graph.
Then the equality (â) satisfies:
(1)
The star graph Gsstar(xvâ)â(sâ„1).
2. (2)
The path graph Pnâ(nâ„2).
3. (3)
The n-cycle Cnâ(nâ„3).
4. (4)
The graph Gsâ on {x1â,âŠ,xs+4â} where sâ„1
which consists of the star graph Gsstar(xs+3â)â
on {x1â,âŠ,xsâ}âȘ{xs+3â}
and P4â on {xs+1â,âŠ,bxs+4â}; see Figure 7.
Before proving Proposition 2.10,
we recall some facts on invariants of an edge ideal.
For a finite simple graph G,
the dimension dimS/I(G) is equal to the maximum cardinality
of independent sets of G.
In particular, one has
dimS/I(Pnâ)=ân/2â and
dimS/I(Cnâ)=â(nâ1)/2â.
We also recall the non-vanishing theorem of Betti numbers of edge ideals.
Let G be a finite simple graph.
Suppose that there exists a set of star subgraphs
{B1â,âŠ,Bââ} (ââ„1) of G
satisfying the following conditions:
Recall that the equality (â) is satisfied if and only if
(p,p+r)-th Betti number does not vanish where
p is the projective dimension and r is the regularity.
(1)
Since Gsstar(xvâ)â has no cycle,
one has reg(S/I(Gsstar(xvâ)â))=im(G)=1 by [22].
Also, it is easy to see from Lemma 2.11 that
projdim(S/I(Gsstar(xvâ)â))=s,
and ÎČs,s+1â(S/I(Gsstar(xvâ)â))î =0.
2. (2)
Let V(Pnâ)={x1â,x2â,âŠxnâ}
and E(Pnâ)={{x1â,x2â},{x2â,x3â},âŠ,{xnâ1â,xnâ}}.
It follows from [18, Lemma 2.8] that
depth(S/I(Pnâ))=ân/3â.
Hence by AuslanderâBuchsbaum Theorem, one has
**The case n=3â or n=3â+1 : **
Then p=2â and r=â.
For 1â€kâ€â, let Bkâ be the induced subgraph of Pnâ
on {x3(kâ1)+1â,x3(kâ1)+2â,x3kâ}.
Then Bkâ is the star subgraph
G2star(x3(kâ1)+2â)â.
Take ekâ:={x3(kâ1)+1â,x3(kâ1)+2â}âE(Bkâ).
Then {e1â,âŠ,eââ} forms an induced matching of Pnâ.
Thus Lemma 2.11 says that
ÎČp,p+râ(S/I(Pnâ))=ÎČ2â,2â+ââ(S/I(Pnâ))î =0.
âą
**The case n=3â+2 : **
Then p=2â+1 and r=â+1.
For 1â€kâ€â, let Bkâ be the induced subgraph of Pnâ
on {x3(kâ1)+1â,x3(kâ1)+2â,x3kâ}.
Then Bkâ is the star subgraph
G2star(x3(kâ1)+2â)â.
Also let Bâ+1â be the induced subgraph of Pnâ
on {x3â+1â,x3â+2â},
which is the star subgraph
G1star(x3â+2â)â.
Take ekâ:={x3(kâ1)+1â,x3(kâ1)+2â}âE(Bkâ)
for k=1,âŠ,â,â+1.
Then {e1â,âŠ,eââ,eâ+1â}
forms an induced matching of Pnâ.
Thus Lemma 2.11 says that
ÎČp,p+râ(S/I(Pnâ))=ÎČ2â+1,(2â+1)+â+1â(S/I(Pnâ))î =0.
3. (3)
Let V(Cnâ)={x1â,x2â,âŠxnâ}
and E(Cnâ)={{x1â,x2â},âŠ,{xnâ1â,xnâ},{x1â,xnâ}}.
It follows from [7, p. 117] that
Then we can prove the case where n=3â.
In this case,
p=2â and r=â.
For 1â€kâ€â, let Bkâ be the induced subgraph of Cnâ
on {x3(kâ1)+1â,x3(kâ1)+2â,x3kâ}.
Then Bkâ is the star subgraph
G2star(x3(kâ1)+2â)â.
Take ekâ:={x3(kâ1)+1â,x3(kâ1)+2â}âE(Bkâ).
Then {e1â,âŠ,eââ} forms an induced matching of Cnâ.
Thus Lemma 2.11 says that
ÎČp,p+râ(S/I(Cnâ))=ÎČ2â,2â+ââ(S/I(Cnâ))î =0.
Hence S/I(Cnâ) satisfies the equality (â).
For the cases n=3â+1,3â+2, we compute all invariants
appearing in the equality (â).
We have already known the dimension, the depth, and the regularity.
In order to compute degh(S/I(Cnâ),λ),
consider the short exact sequence
[TABLE]
Since I(Cnâ)+(xnâ)=(xnâ)+I(Pnâ1â), we have
[TABLE]
Also since I(Cnâ):(xnâ)=(x1â,xnâ1â)+(x2âx3â,âŠ,xnâ3âxnâ2â), we have
**The case n=3â+1 : **
Then reg(S/I(Cnâ))=depth(S/I(Cnâ))=â
and dimS/I(Cnâ)=â3â/2â.
Moreover, since
[TABLE]
and
[TABLE]
one has degh(S/I(Cnâ),λ)=â3â/2â.
Hence S/I(Cnâ) satisfies the equality (â).
âą
**The case n=3â+2 : **
Then reg(S/I(Cnâ))=depth(S/I(Cnâ))=â+1
and dimS/I(Cnâ)=â(3â+1)/2â.
Moreover, since
[TABLE]
and
[TABLE]
one has degh(S/I(Cnâ),λ)=â(3â+1)/2â.
Hence S/I(Cnâ) satisfies the equality (â).
4. (4)
Since Gsâ has no cycle,
one has reg(S/I(Gsâ))=im(G)=1 by [22].
Also it is easy to see from Lemma 2.11 that
projdim(S/I(Gsâ))=s+2,
and ÎČs+2,(s+2)+1â(S/I(Gsâ))î =0.
Remark 2.12**.**
The graph Gsâ in Proposition 2.10 (as well as P3â+1â)
is an example of a graph
satisfying (â) with degh(S/I(Gsâ),λ)<dimS/I(Gsâ)(=s+2)
because reg(S/I(Gsâ))=1<2=(s+4)âprojdim(S/I(Gsâ))=depth(S/I(Gsâ)).
Note that CameronâWalker graphs G satisfies
degh(S/I(G),λ)=dimS/I(G).
3. Other properties on CameronâWalker graphs
In this section, we provide some properties on a CameronâWalker graph
derived from the results of previous sections.
Let G be a finite simple graph and S=K[V(G)].
Suppose that S/I(G) is CohenâMacaulay.
Then the equalities (â) and dimS/I(G)=depth(S/I(G)) hold.
Hence one has degh(S/I(G),λ)=reg(S/I(G)).
Nevertheless, degh(S/I(G),λ)=reg(S/I(G)) does not imply that
S/I(G) is CohenâMacaulay, see [12, Example 3.2].
Moreover, in general, there is no relationship
between the regularity and the degree of the h-polynomial.
Actually, [13] proved that for given integers r,sâ„1,
there exists a finite simple graph G such that
reg(S/I(G))=r and degh(S/I(G),λ)=s.
However, we can derive from Proposition 1.3 the relation between
reg(S/I(G)) and degh(S/I(G),λ) when G is CameronâWalker.
Moreover we provide a complete classification of CameronâWalker graphs G
with degh(S/I(G),λ)=reg(S/I(G)).
Theorem 3.1**.**
Let G be a CameronâWalker graph whose labeling of vertices is
as in Figure 1.
Then we have degh(S/I(G),λ)â„reg(S/I(G)).
Moreover the equality degh(S/I(G),λ)=reg(S/I(G))
holds if and only if siâ=1 for all 1â€iâ€m and
tjââ„1 for all 1â€jâ€n.
Proof.
We first note that reg(S/I(G))=âj=1nâtjâ+m.
Combining this with Proposition 1.3, one has
[TABLE]
Note that each summands of right hand-side is non-negative.
Then the desired assertion follows.
Let G be a CameronâWalker graph.
Combining the inequality
[TABLE]
with Theorem 1.1, Theorem 3.1, and Proposition 2.8,
one has
[TABLE]
Then it is natural to ask the following
Question 3.2**.**
Given arbitrary integers d,r,e with dâ„râ„eâ„2,
is there a CameronâWalker graph G
satisfying
[TABLE]
We have already investigated CameronâWalker graphs G
with depthS/I(G)=2 in Proposition 2.8.
Their invariants are as follows:
Therefore we have the following answer for Question 3.2
when e=2.
Corollary 3.3**.**
Let d,r,e be integers with dâ„râ„e=2.
Then there exists a CameronâWalker graph G satisfying
(ââ) if and only if r=2 or r=d.
When eâ„3, we have the following answer for Question 3.2.
Theorem 3.4**.**
Given arbitrary integers d,r,e with dâ„râ„eâ„3,
there exists a CameronâWalker graph G satisfying
dimS/I(G)=degh(S/I(G),λ)=d, reg(S/I(G))=r,
and depth(S/I(G))=e.
Proof.
We use the labeling of vertices of a CameronâWalker graph
as in Figure 1.
Set Vbipâ={v1â,âŠ,vmâ,w1â,âŠ,wnâ}.
â** The case d>r:**
Let G be the CameronâWalker graph with m=eâ1, n=2,
s1â=âŻ=seâ2â=1, seâ1â=dâr, t1â=râe+1, and
t2â=0 such that
Then it is easy to see that
dim(S/I(G))=degh(S/I(G),λ)=d and reg(S/I(G))=r.
Also, A:={v2â,âŠ,veâ1â}âȘ{x1(1)â,w1â}
is an independent set of V(G) with AâȘNGâ(A)=V(G)
which gives i(G). Thus one has depthS/I(G)=i(G)=âŁAâŁ=e.
â** The case d=r:**
Let G be the CameronâWalker graph with m=eâ1, n=1,
s1â=âŻ=seâ1â=1, and t1â=dâe+1.
Then it is easy to see that
dim(S/I(G))=degh(S/I(G),λ)=reg(S/I(G))=d.
Also A:={x1(1)â,âŠ,xeâ1(1)â}âȘ{w1â}
is an independent set of V(G) with AâȘNGâ(A)=V(G)
which gives i(G). Thus one has depthS/I(G)=i(G)=âŁAâŁ=e.
Acknowledgment.
The authors were partially supported by JSPS KAKENHI 26220701, 15K17507,
17K14165 and 16J01549.
Bibliography22
The reference list from the paper itself. Each links out to its DOI / PubMed record.
1[1] S. Beyarslan, H. T. HĂ and T. N. Trung, Regularity of powers of forests and cycles, J. Algebraic Combin. 42 (2015), 1077â1095.
2[2] M. Bigdeli and J. Herzog, Betti diagrams with special shape, Homological and Computational Methods in Commutative Algebra , Springer I Nd AM Series 20 (2017), 33â52.
3[3] T. BiyikoÄlu and Y. Civan, Vertex decomposable graphs, codismantlability, CohenâMacaulayness and Castelnuovo-Mumford regularity, Electron. J. Combin. 21 (2014), 17pp.
4[4] R. R. Bouchat, Free resolutions of some edge ideals of simple graphs, J. Commut. Algebra 2 (2010), 1â35.
5[5] W. Bruns and J. Herzog, Cohen-Macaulay Rings, Revised ED., Cambridge Stud. Adv. Math., vol. 39 , Cambridge University Press, Cambridge, 1998.
6[6] K. Cameron and T. Walker, The graphs with maximum induced matching and maximum matching the same size, Discrete Math. 299 (2005), 49â55.
7[7] M. CimpoeaĆ, On the Stanley depth of edge ideals of line and cyclic graphs, Rom. J. Math. Comput. Sci. 6 (2016), 116â120.
8[8] H. Dao, C. Huneke and J. Schweig, Bounds on the regularity and projective dimension of ideals associated to graphs, J. Algebraic Combin. 38 (2013), 37â55.