We provide explicit formulas for the Betti numbers of all powers of Ferrers ideals and edge ideals of threshold graphs.
Introduction
Let λ=(λ1,…,λn) be a partition, meaning that λ1≥λ2≥⋯≥λn≥1, and let m=λ1. Let R=k[x1,…,xn,y1,…,ym] be a polynomial ring over the field k. The Ferrers ideal associated with the partition λ=(λ1,…,λn) is given by
Iλ=(xiyj∣1≤i≤n,1≤j≤λi).
For each partition λ, Corso and Nagel [4] introduced and studied the Ferrers ideal Iλ and described its minimal free resolution. In this work, we provide a formula for the Betti numbers of all powers of Iλ. Note that there are very few classes of ideals for which the Betti numbers of all powers are known, including maximal ideals, path ideals of paths [1], and certain path ideals of cycles [2]. To achieve this, we provide an ordering of the minimal generators of Iλt and show that this ordering yields a linear quotient, from which the formula for the Betti numbers follows. We now introduce these concepts in more detail to state our main results.
Each finitely generated graded R-module, and in particular R/Iλ, admits a minimal graded free resolution
where R(−j) denotes the free R-module obtained by shifting the degrees of R by j. The number βi,j(R/Iλ) is the (i,j)-th graded Betti number of R/Iλ. Herzog and Takayama [7] introduced the notion of ideals having linear quotients and proved that, in this case, the resolution can be constructed using iterated mapping cones. In particular, if all generators have the same degree, the ideal has a linear free resolution. We now describe the set of minimal generators of Iλt.
For each t≥1, let
C(λ,t)={(i1,…,it;j1,…,jt)∣1≤i1≤⋯≤it≤n,j1≥j2≥⋯≥jt,ja≤λiafor all a=1,…,t}.
Then we have:
Theorem 1.1. The minimal generators of Iλt are given by xi1⋯xityj1⋯yjt for (i1,…,it;j1,…,jt)∈C(λ,t). Furthermore, when these generators are ordered lexicographically with x1>x2>⋯>xn>y1>⋯>ym, Iλt has a linear quotient with respect to this order.
From this, we deduce the following formula for the Betti numbers:
Theorem 1.2. Let λ=(λ1,…,λn) be a partition with λ1=m. Then for all p, we have
The linear quotient property for powers of Ferrers graphs under lexicographic order extends to the class of threshold graphs. This allows us to deduce the Betti numbers of powers of edge ideals of threshold graphs. The remainder of this paper is devoted to proving these formulas and discussing several of their applications.
Linear quotients and Betti numbers
First, we establish the description of the minimal generating set.
Lemma 2.1. Let λ=(λ1,…,λn) be a partition with λ1=m. Then Iλt is minimally generated by xi1⋯xityj1⋯yjt for (i1,…,it;j1,…,jt)∈C(λ,t).
Proof. We have xiayja∈Iλ for all a=1,…,t; hence, u=xi1⋯xityj1⋯yjt∈Iλt. Conversely, assume that f=(xα1yβ1)⋯(xαtyβt)∈Iλt is a generator. We may assume without loss of generality that α1≤⋯≤αt. We can repeatedly apply the exchange relation to obtain the canonical representation. Indeed, suppose that we have two factors xiya and xkyb with i≤k and a<b. Then, by definition, xiyb∈Iλ and xkya∈Iλ. Thus, we can perform the exchange
(xiya)(xkyb)=(xiyb)(xkya).
Applying this operation finitely many times rearranges the y-indices into non-increasing order without altering the x-indices, which completes the proof. □
Definition 2.2. Let I=(u1,…,us) be a monomial ideal with ordered minimal generators u1,…,us. We say that I has linear quotients with respect to this order if, for every j=2,…,s, the colon ideal
(u1,…,uj−1):uj
is generated by variables.
Before proving the result on linear quotients, let us consider an example for clarity.
Example 2.3. Let λ=(3,2). Then the minimal generators of Iλ2 are ordered lexicographically as follows:
With this ordering, we can verify the linear quotients directly; for instance,
(x12y12,x12y1y2):(x12y1y3)=(y1,y2).
Note that while the canonical tuples (i1,…,it;j1,…,jt)∈C(λ,t) write the y-indices in non-increasing order (j1≥⋯≥jt), writing the monomial factors with y-indices appearing in non-decreasing order aligns naturally with the lexicographic ordering.
Lemma 2.4. Let u=xi1⋯xityj1⋯yjt be a minimal generator of Iλt corresponding to the tuple (i1,…,it;j1,…,jt)∈C(λ,t). Let Qu denote the ideal generated by all minimal generators preceding u in the lexicographic order. Setting r=i1 and c=j1, we have
Qu:u=(x1,…,xr−1,y1,…,yc−1).
Proof. Let r=i1 and c=j1. First, we prove the inclusion (x1,…,xr−1,y1,…,yc−1)⊆Qu:u. For any a<r, replacing xr with xa yields a valid minimal generator v in Iλt (after sorting the indices into C(λ,t)) that strictly precedes u in the lexicographic order, meaning xa∈Qu:u. Similarly, for any b<c, replacing yc with yb yields a valid minimal generator v preceding u, so yb∈Qu:u.
Conversely, to show that Qu:u⊆(x1,…,xr−1,y1,…,yc−1), recall the variable ordering x1>⋯>xn>y1>⋯>ym. In the lexicographic order, any preceding generator v<lexu must introduce a strictly larger variable at the earliest possible position where v and u differ. Since u has leading variable components determined by i1=r and j1=c, any variable dividing v/gcd(u,v) that places v before u must belong to the set {x1,…,xr−1} or {y1,…,yc−1}. Consequently, every generator of Qu:u is divided by at least one variable from this set, establishing the reverse inclusion and completing the proof. □
Proof of Theorem 1.1. The conclusion follows directly from Lemma 2.1 and Lemma 2.4. □
Proof of Theorem 1.2. The assertion follows by induction from Lemma 2.4 together with [11 Theorem 2.6]. □
Example 2.5. Let λ=(3,2) and t=4, so that R=k[x1,x2,y1,y2,y3]. Each tuple in C(λ,4) can be written uniquely as
Here, a counts the occurrences of 1 among the i-indices, while q and r count the occurrences of 3 and 2 among the j-indices. Therefore,
∣C(λ,4)∣=a=0∑4q=0∑a(5−q)=55.
The choices of (a,q,r) can be grouped according to the value of i4+j1−2 as follows:
Parameter choices
Number of tuples
i4+j1−2
0≤a≤3, q=0, r=0
4
1
0≤a≤3, q=0, 1≤r≤4
16
2
1≤q≤a≤3, 0≤r≤4−q
20
3
a=4, q=0, r=0
1
0
a=4, q=0, 1≤r≤4
4
1
a=4, 1≤q≤4, 0≤r≤4−q
10
2
Table 1.
Thus, the exponent values 0, 1, 2, 3 occur respectively 1, 8, 26, 20 times. By Theorem 1.2,
βpR(Iλ4)=(p0)+8(p1)+26(p2)+20(p3).
Evaluating this expression gives β0=55, β1=120, β2=86, β3=20, and βp=0 for p≥4. Hence, the Betti sequence is (55,120,86,20).
Remark 2.6. It is well known that Iλt has linear resolutions for all t≥1 (see, e.g., [8 Theorem 3.2]). The depth of powers of R/Iλt is also well known (see, e.g., [9 Proposition 2.18]). Basser, Diethorn, Miranda, and Stinson-Maas [3 Theorem 3.11] show that powers of edge ideals of cochordal graphs have linear quotients. Theorem 1.1 provides a specific ordering that yields linear quotients for Iλt, resulting in a more explicit formula for the Betti numbers.
Betti numbers of powers of threshold graphs
Threshold graphs are a special class of cochordal graphs, where Ferrers graphs correspond precisely to bipartite threshold graphs. They have numerous applications (see, e.g., [10]). Corso and Nagel [5] showed that each threshold graph can be obtained by specializing a Ferrers graph. More precisely, by [5 Corollary 4.6], every threshold graph can be characterized by a partition λ=(λ1,…,λn) with λ1≥λ2≥⋯≥λn>n where {i,j} with i<j is an edge of the threshold graph Tλ if and only if j≤λi. Let S=k[z1,…,zm] where m=λ1. The corresponding edge ideal of Tλ is given by
I(Tλ)=(zizj∣i<j≤λi).
For a monomial u=z1a1⋯ztat with at>0, we denote by last(u)=t the largest index of a variable dividing u.
Theorem 3.1. The minimal generators of I(Tλ)t, ordered lexicographically with z1>z2>⋯>zm, form a linear quotient order for I(Tλ)t. Furthermore, for a minimal generator u=z1a1⋯zmam of I(Tλ)t, we have
Qu:u=(zr∣ar<tandr<last(u)),
where Qu is the ideal generated by all minimal generators preceding u in this order.
Proof. Let q=last(u) and let L=(zr∣ar<tandr<q). First, we claim that L⊆Qu:u. Assume that r<q and ar<t. Choose a factorization of u into a product of t edges, u=e1⋯et. Since q∣u, there exists an edge ej such that zq divides ej. Hence, we may write ej=zpzq with p<q. Since ej∈I(Tλ), we have q≤λp. There are two cases:
Case 1:p=r. We claim that zpzr∈I(Tλ). Indeed, if p<r, then p<r<q≤λp, hence zpzr∈I(Tλ). If r<p, then p<q≤λp≤λr, hence zpzr∈I(Tλ). In either case, we have w=e1⋯ej−1(zpzr)ej+1⋯et∈I(Tλ)t. Furthermore, w/gcd(w,u)=zr. Thus, zr∈Qu:u.
Case 2:p=r. Since ar<t, there exists an edge ei such that zr does not divide ei. Write ei=zazb. Since q=last(u), we have a,b≤q. We claim that zrza∈I(Tλ). Indeed, since zrzq∈I(Tλ), we have r<q≤λr. Now, if r<a, then r<a<q≤λr, hence zrza∈I(Tλ). If a<r, then a<r<q≤λr≤λa, hence zrza∈I(Tλ). Similarly, zrzb∈I(Tλ). Replacing eiej by (zrza)(zrzb) yields a new element w∈I(Tλ)t such that w/gcd(w,u)=zr, which implies zr∈Qu:u.
Now, let v be any generator preceding u in the lexicographic order. Write v=z1b1⋯zmbm. Let r be the smallest index such that br=ar. By definition of the lexicographic order, we have br>ar and bi=ai for all i<r. Since v is a minimal generator of I(Tλ)t, we must have br≤t, which implies ar<br≤t. Furthermore, deg(v)=deg(u), so there must exist an index j>r such that bj<aj. In particular, r<j≤last(u)=q. Consequently, v/gcd(u,v) is divisible by zr, showing that every generator of Qu:u is divisible by a variable zr with r<q and ar<t. The conclusion follows. □
From Theorem 3.1, we see that to find the Betti numbers of the powers I(Tλ)t, we need to determine formulas for two key invariants. First, let
Genst(λ,q)={u∣uis a minimal generator of I(Tλ)twith last(u)=q}
and
Genst∗(λ,q)={u=z1a1⋯zqaq∈Genst(λ,q)∣ai=tfor some i<q}.
We then define the cardinalities:
at(λ;q)=∣Genst(λ,q)∣,
bt(λ;q)=∣Genst∗(λ,q)∣.
Theorem 3.2. Let Tλ be the threshold graph corresponding to a partition λ with λ1≥⋯≥λn≥n. Then
Proof. By Theorem 3.1 and [11 Theorem 2.6], we have
βp(I(Tλ)t)=u∈G(Iλt)∑(p∣C(u)∣),
where C(u) is the set of variables in the quotient ideal Qu:u. For any minimal generator u with last(u)=q, the size of C(u) satisfies
∣C(u)∣={q−2q−1if u∈Genst∗(λ,q),otherwise.
The conclusion follows. □
Corollary 3.3. Let Tλ be the threshold graph corresponding to a partition λ with λ1=m≥⋯≥λn≥n. Then for all t≥2,
pd(I(Tλ)t)={m−1m−2if n≥2,if n=1.
Proof. If n=1, then I(Tλ)=z1(z2,…,zm), and its powers have projective dimension m−2. We may therefore assume that n≥2 and t≥2, which implies λ1=m>n≥2. By the Betti number formula in Theorem 3.2, it suffices to show that there exists a minimal generator u of I(Tλ)t such that ∣C(u)∣=m−1. To this end, consider the monomial u=(z1zm)(z2z3)t−1. Then u∈Genst(λ,m) but u∈/Genst∗(λ,m). Consequently, ∣C(u)∣=m−1. The conclusion follows. □
Lemma 3.4. Let ρλ(q)=∣{1≤r<q∣λr≥q}∣. Then
bt(λ;q)=ρλ(q)(t−1q+t−3).
Proof. Let u=z1a1⋯zqaq be a minimal generator of I(Tλ)t such that ar=t for some index r<q. This condition implies that the edge zrzq belongs to I(Tλ), which is equivalent to λr≥q. Under this condition, we have zrzi∈I(Tλ) for all i∈{1,…,q}∖{r}.
There are ρλ(q) possible choices for the index r<q such that λr≥q. For a fixed r, the condition ar=t means that zrt accounts for t factors of u. The remaining part is a monomial of degree t in the variables {z1,…,zq}∖{zr}. The number of such monomials is given by (t−1q+t−3). Summing over all ρλ(q) possible choices for r, we obtain
bt(λ;q)=ρλ(q)(t−1q+t−3).
:::`
This completes the proof. □
To compute at(λ,q), we introduce the following notation. For a partition λ and an integer q, we define the truncation partition λ[q] to be the partition given by λi[q]=min{q,λi}, and we let λ denote the partition (λ2−1,…,λn−1). For any r,s≥0, we set
A(λ;r,s)=dimk(I(Tλ)t)2r+s.
In particular, A(λ;t,0)=μ(I(Tλ)t), which is the number of minimal generators of I(Tλ)t.
First, we have the following lemma:
Lemma 3.5.With the notation above, we have
at(λ,q)=A(λ[q];t,0)−A(λ[q−1];t,0).
Proof. Any generator of Iλ[q]t that does not belong to Iλ[q−1]t is divisible by zq. The conclusion then follows directly from the definitions. □
Proof. When 2≤q≤n+1, the truncation λ[q] gives rise to a complete graph on q vertices. Thus, at(λ;q) equals the number of tuples of nonnegative integers (a1,…,aq) satisfying ai≤t for all i, aq≥1, and ∑i=1qai=2t.
Setting bq=aq−1, this condition corresponds to tuples of nonnegative integers (a1,…,aq−1,bq) such that ∑i=1q−1ai+bq=2t−1, with the constraints ai≤t for 1≤i≤q−1 and bq≤t−1.
Without these upper-bound constraints, the number of such tuples—equal to the number of degree 2t−1 monomials in q variables—is given by the first term, (q−12t+q−2).
If any of the variables ai (1≤i≤q−1) violates the condition by satisfying ai≥t+1, we can factor out zit+1, leaving a monomial of degree t−2 in q variables. Summing over the q−1 choices for such an index yields the second term, (q−1)(q−1t+q−3).
Similarly, if the boundary variable satisfies bq≥t, the remaining terms form a monomial of degree t−1 in q variables, giving the third term, (q−1t+q−2).
Since no two upper-bound violations can occur simultaneously, we obtain the desired formula. □
Lemma 3.7.Assume that λ1=⋯=λa=m>λa+1 with a≥2. Then
Proof. For simplicity of notation, let J=Iλ and K=Iλˉ. Then we have J=z1(z2,…,zm)+K. Consequently, the t-th power of J admits the expansion
Jt=r=0∑tz1t−r(z2,…,zm)t−rKr.
For any monomial u∈(Jt)2t+s, we can write u=z1dv for some integer d≥0. When d≤t−1, setting r=t−d, the remaining part v belongs to (z2,…,zm)t−rKr, which corresponds to the graded component of Kr in degree 2t+s−d=t+s+r=2r+(t+s−r). This yields the contribution A(λˉ;r,t+s−r).
When d≥t, the monomials arise entirely from the term (z2,…,zm)t. The number of such monomials equals the number of monomials of degree t through t+s in a polynomial ring in m−1 variables. The total number of such monomials is given by
d=t∑t+s(m−2d+m−2)=(m−1m+t+s−1)−(m−1m+t−2).
Combining these two cases yields the desired formula. The conclusion follows. □
Lemma 3.8. Assume that λ1=m>λ2, and let e=m−λ2. Then
Proof. As in the proof of Lemma 3.7, the case where d≥t yields the standard binomial difference term. When d<t, we must account for the graded components of Kr within the polynomial ring in variables z2,…,zm. Since K is supported only on the variables z2,…,zλ2, the remaining e=m−λ2 variables lie outside the support of K. The inner sum over v correctly accounts for the contribution of these e additional variables. □
Corollary 3.9. Let Km denote the complete graph on m vertices. Then for all 0≤p≤m−1, we have
we can express Ep(k) as a telescoping difference Ep(k)=Fp(k+1)−Fp(k), where
Fp(k)=(k+1)(kt+k−1)(pk).
Summing over k from 1 to m−1, the terms telescope, allowing us to deduce the final formula. The conclusion follows.
Remark 3.10.Theorem 3.2 and Lemmas 3.4–3.8 allow us to compute the Betti numbers of arbitrary threshold graphs. Dung and Vu [6] associated a type sequence to any chordal graph and gave a formula for the Betti numbers of its edge ideal in terms of this type sequence. However, we note that the type sequence alone is not sufficient to determine the Betti numbers of powers of its edge ideal. For example, K2,2 and a triangle with a pendant edge both have the same type sequence (2,2), but their second powers do not share the same Betti numbers.
We do not know whether the linear quotient property with respect to the lexicographic ordering holds for powers of edge ideals of arbitrary specializations. Nonetheless, we can show that the projective dimension of powers of any arbitrary specialization is equal to m if it contains a triangle. Note that if the specialization is bipartite, then its edge ideal is isomorphic to a Ferrers ideal.
Proposition 3.11. Let λ=(λ1,…,λn) be a partition and μ=(μ1,…,μn) satisfy μ1≤⋯≤μn<λn. Let I=(zizj∣1≤i≤n,μi<j≤λi) be the edge ideal of the specialization graph Gλ,μ. Assume that I contains a triangle, or equivalently, μ1<n. Then pd(S/It)=m for all t≥2, where m=λ1 denotes the number of vertices.
Proof. First, we show that I is non-bipartite if and only if μ1<n. Indeed, if μ1≥n, then the underlying graph Gλ,μ is bipartite with bipartition {1,…,n}∪{μ1+1,…,λn}. Conversely, if μ1<n, setting j=μ1+1 and k=λn, we see that {1,j,k} forms a triangle.
Next, we claim that the triangle {1,j,k} dominates Gλ,μ. To see this, let v be any vertex:
If v>μ1, then {1,v} is an edge by definition.
If v≤μ1, then μv≤μn<λn=k≤λv, which implies that {v,k} is an edge.
Since Gλ,μ contains a dominating triangle, it follows that pd(S/It)=m for all t≥2. The conclusion follows. □
Remark 3.12. The conclusion of Proposition 3.11 does not hold for arbitrary cochordal graphs containing a triangle. Indeed, let
be the edge ideal of the whisker graph on the complete graph K4. Then pd(S/I2)=7<8.
Acknowledgment. Thanh Vu is partially supported by the Vietnam National Foundation for Science and Technology Development under the grant number NAFOS-TED/101/10104/2026/1279.
References
Faculty of Mathematics and Informatics, Hanoi University of Science and Technology, 1 Dai Co Viet, Bach Mai, Hanoi, Vietnam. Email address: hoang.dotr [email protected]
Institute of Mathematics, VAST, 18 Hoang Quoc Viet, Hanoi, Vietnam Email address:[email protected]
References
[1][BCV1] S. Balanescu, M. Cimpoeas, and T. Vu, Betti numbers of powers of path ideals of cycles, Journal of Algebraic Combinatorics 61 (2025), 44. 1arxiv.org/abs/2404.17880
[2][BCV2] S. Balanescu, M. Cimpoeas, and T. Vu, Graded Betti numbers of powers of path ideals of paths, Journal of Commutative Algebra 18 (2026), 129–138. 1arxiv.org/abs/2405.04747
[3][BDMS] E. Basser, R. Diethorn, R. Miranda, and M. Stinson-Maas, Powers of edge ideals with linear quotients, arXiv:2412.03468 4arxiv.org/abs/2412.03468
[4][CN1] A. Corso and U. Nagel, Monomial and toric ideals associated to Ferrers graphs, Trans. Am. Math. Soc. 361 (2009), 1371–1395. 1DOI
[5][CN2] A. Corso and U. Nagel, Specializations of Ferrers ideals, J Algebra. Comb. 28 (2008), 425–437. 4arxiv.org/abs/math/0703695
[6][DV] L. X. Dung and T. Vu, Cochordal zero divisor graphs and Betti numbers of their edge ideals, Comm. Alg. 54 (2026), 736–744. 9arxiv.org/abs/2411.05251
[7][HT] J. Herzog and Y. Takayama, Resolutions by mapping cones, Homology Homotopy Appl. 4 (2002), 277–294. 1DOI
[8][HHZ] J. Herzog, T. Hibi and Zheng, Monomial ideals whose powers have a linear resolution, Math. Scand. 95 (2004), 23–32. 4DOI
[9][HV] D. T. Hoang and T. Vu, Depth and regularity of tableau ideals, Adv. Appl. Math. 169 (2025), 102913. 4arxiv.org/abs/2311.07778
[10][MP] N. V. R. Mahadev and U. N. Peled, Threshold Graphs and Related Topics, Annals Discrete Mathematics, vol. 56. North-Holland, Amsterdam (1995) 4
[11][SV] L. Sharifan and M. Varbaro, Graded Betti numbers of ideals with linear quotients, Le Matematische 68 (2008), 257–265. 3, 6arxiv.org/abs/0901.3088