THE KOSZULNESS OF NUMERICAL SEMIGROUP RINGS OF

MAXIMAL EMBEDDING DIMENSION

DO VAN KIEN AND NGUYEN HONG NGOC

Abstract. In this paper, we show that the associated graded rings of numerical semi-
group rings of maximal embedding dimension is Cohen-Macaulay. As a consequence, we
conclude that numerical semigroup rings of maximal embedding dimension are Koszul.

1. Introduction
Let k be a field and (R, m) be a standard graded k-algebra with the graded maximal
ideal m. We say that R is Koszul (or a Koszul algebra) if the residue field k = R/m has a
linear free resolution over R of the form
(*) · · · → R(−i)βi → · · · → R(−2)β2 → R(−1)β1 → R → k.
This means that the matrices describing the differentials of k have non-zero entries only
linear forms. Ksozul algebras were originally introduced by Priddy (see [?]) in his study of
homological properties of graded algebras. We refer to the survey articles/books ([?, ?, ?])
for more details. There are important relations between the Koszulness and the structure
of the non commutative algebra Ext∗R (k, k), i.e., the Yoneda-Hopf algebra of k. Among
other things, Koszul algebras are also important because they give an interesting class of
quadratic algebras with rational Poincaré series.
Note that if R is a standard graded k-algebra, then there is a presentation R ∼ =
k[X1 , . . . , Xn ]/I, where k[X1 , . . . , Xn ] is a polynomial ring. We call I the defining ideal
of R. Then the condition that k has a linear resolution over R is equivalent to the fact
that the Betti graded numbers βijR (k) = 0 for all j ̸= i. In particular, β2j R
(k) = 0 for all
j ̸= 2. This implies that whenever R is Koszul then I is quadratics, that is generated
by homogeneous polynomials of degree 2. The converse does not hold in general. For
instance, the ring R = k[X, Y, Z, U ]/(X 2 , Y 2 , Z 2 , U 2 , XY + XZ + XU ) has β34R
(k) = 2 ̸= 0
2 2 2 2
which implies that R is not Koszul, while I = (X , Y , Z , U , XY + XZ + XU ) is qua-
dratic. Nevertheless it is also well-known that (see [?, Theorem 34.12]) if I has a quadratic
Gröbner basis then R is Koszul.
This paper is organized as follows. In Section 2, we will begin with some preliminaries
on numerical semigroups of maximal embedding dimension that we need in the paper.
Section 3 consists of a survey on Koszul filtration, an effective tool to prove an algebra to
Key words and phrases. Cohen-Macaulay ring, Koszul ring, Numerical semigroup ring, associated
graded ring.
1
2

be Koszul. In the last section we give the main results of the paper. More precisely, we
show that the associated graded rings of numerical semigroup rings of maximal embedding
dimension is Cohen-Macaulay. As an application, we prove that numerical semigroup rings
of maximal embedding dimension are Koszul. We also provide a few examples to illustrate
this result.

2. Preliminaries

2.1. Numerical semigroups of maximal embedding dimension.

Let a1 , a2 , · · · , an be a sequence of positive integers such that GCD(a1 , a2 , . . . , an ) = 1.
Let ( n  )
X 
H = ⟨a1 , a2 , . . . , an ⟩ = ci ai  0 ≤ ci ∈ Z for all 1 ≤ i ≤ n .


i=1
Then H is a submonoid of the additive monoid N and H has finite complement in N. We
call H the numerical semigroup generated by the integers ai ’s. The system {a1 , a2 , · · · , an }
is said to be minimal if for every 1 ≤ i ≤ n, ai can be not written as a combination of
{a1 , . . . , an } \ {ai } with integer coefficients.
In what follows, let H = ⟨a1 , a2 , . . . , an ⟩ denote a numerical semigroup minimally gen-
erated by n elements. There are a few invariants of H as follows:

Definition 2.1. (1) The embedding dimension of H denoted by edim(H) is the car-
dinality of the minimal set of generators of H, i.e., edim(H) = n.
(2) The multiplicity of H denoted by e(H) is the smallest non zero element of H, i.e.,
e(H) = a1 .
(3) The Frobenius number of H, denoted by F(S) to be max (Z \ S).
(4) The set of gaps of H, denoted by G(S) to be N \ H.
(5) The genus of H, denoted by g(S) to be the cardinality of G(S).

There is an interesting relation between the embedding dimension and the multiplicity
that edim(H) ≤ e(H) (see [?, Proposition 2.10]). The semigroup H that attains this
bound is called maximal embedding dimension.
Let 0 ̸= a ∈ H. We define the Apéry set of H with respect to a, denoted by Ap(H, a),
to be the set Ap(H, a) = {h ∈ H | h − a ∈ / H}. We summarize several characterizations
of numerical semigroup of maximal embedding dimension in the following.

Proposition 2.2 (see [?]). The following conditions are equivalent.

1) H has maximal embedding dimension.
2) For all x, y ∈ H such that x ≥ y ≥ e(H) then x + y − e(H) ∈ H \ {0}.
3) Ap(H, a1 ) = {0, a2 , . . . , an }.
n
1 X a1 − 1
4) g(S) = ai − .
a1 i=2 2
 3

2.2. Numerical semigroup rings and their associated graded rings. Let H =
⟨a1 , . . . , an ⟩ be a numerical semigroup of embedding dimension n. Let k be a field and
R = k[[H]] = k[[ta1 , . . . , tan ]] is a subring of the formal power series ring k[[t]]. Then R
is a Noetherian local domain of dimension 1 with maximal ideal m = (ta1 , . . . , tan ). We
call R the semigroup ring associated to H. A classical result connecting properties of the
semigroup to properties of the ring due to Kunz [?] states that R is Gorenstein if and only
if H is symmetric. Others have linked properties of the semigroup to properties of the
L i i+1
associated graded ring of R, grm (R) = m /m which is a standard graded ring of di-
i≥0
mension 1. Let xi denote the image of tai in m/m2 for all i. Then grm (R) ∼ = k[x1 , . . . , xn ]
where a ”monomial” x1 . . . xn is nonzero if and only if i=1 ci = maxdeg( ni=1 ci ai ).
c1 cn
Pn P
Pn Pn
If xc11 . . . xcnn ̸= 0 then xc11 . . . xcnn = xd11 . . . xdnn if and only if i=1 ci = di and
Pn Pn Pn Pn i=1
i=1 ci ai = i=1 di ai . Here, for h ∈ H, maxdeg(h) = max{ i=1 ci | h = i=1 ci ai }.
Garcı́a showed the following.

Theorem 2.3 ([?], Theorem 7). Let R = k[[ta1 , . . . , tan ]]. Then grm (R) is Cohen-
Macaulay if and only if x1 is a non-zero-divisor in grm (R).

Several authors tried to find some classes of semigroup rings such that their asso-
ciated graded rings are Cohen–Macaulay or at least have non-decreasing Hilbert func-
tion. D’Anna, Micale and Sammartano in [?, ?] characterized when grm (R) is Buchs-
baum/complete intersection. For the case when the embedding dimension is 3, Robbiano
and Valla [?] gave necessary and sufficient conditions on the generators of the defining
ideal for the associated graded ring to be a complete intersection and for it to be Cohen-
Macaulay.
We note that R and its associated graded ring have the same residue field. We say
that R is Koszul if grm (R) is a Koszul algebra, that is, the residue field has linear free
resolution over grm (R).

3. Koszul filtrations
A useful tool to attack the Koszulness is Koszul filtration. It is a possible way to prove
that a certain algebra is Koszul is to show that it admits such a filtration. We recall the
notion of Koszul filtrations introduced by A.Conca, N.V.Trung and G.Valla in [?].

Definition 3.1. Let G be a standard graded ring. A family F of ideals of G is said to

be a Koszul filtration of G if:
1) Every ideal I ∈ F is generated by linear forms.
2) The ideal (0) and the maximal graded ideal m of G belong to F.
3) For every I ∈ F different from (0), there exists J ∈ F such that J ⊆ I, I/J is
cyclic and J : I ∈ F.
4

In [?], it is proved that all the ideals belonging to such a filtration have a linear free
resolution over G and in particular, since the graded maximal ideal m ∈ F, G will be a
Koszul algebra.
Using the Koszul filtration we have the following interesting fact.

Lemma 3.2. Let I be a monomial ideal generated by quadrics in a polynomial ring S =

k[X1 , X2 , . . . , Xn ]. Then the ring G = S/I is Koszul.

Proof. Denote xi is the image of Xi in G for all 1 ≤ i ≤ n. Let F is the set of all ideals
in G generated by variables and let M1 , M2 , . . . , Ms be a minimal system of monomial
generators for I. Note that for any ideal J ⊆ R generated by variables and any xi ∈ / J,
the colon ideal J : xi is equal to J + (xj such that Xj divides some Mℓ ). The ideal J : xi
is therefore generated by variables and it belongs to F. The family F is a Koszul filtration
because any ideal in F can be filtered simply by dropping one variable by its minimal
generators. By the maximal ideal (x1 , x2 , . . . , xn ) ∈ F has a linear free resolution and
consequently G is Koszul. □

4. Main results
Throughout this section, let H = ⟨a1 , a2 , . . . , an ⟩ be the numerical semigroup with
a1 < a2 < . . . < an . Let k be an infinite field. We set R = k[[H]] = k[[ta1 , ta2 , . . . , tan ]]
is the numerical semigroup ring associated to H over k, where t is an indeterminate. It
is clear that R is a Cohen-Macaulay local ring of dimension 1 with the maximal ideal
m = (ta1 , ta2 , . . . , tan ). Note that e(H) and edim(H) correspond to the multiplicity and
embedding dimension of R, respectively. Therefore if H has maximal embedding dimen-
sion then so is R.
Let T = k[[X1 , X2 , . . . , Xn ]] be the formal power series and S = k[X1 , X2 , . . . , Xn ] the
polynomial ring over k. Let φ : T → R denote the homomorphism of k-algebras defined
by φ(Xi ) = tai for all 1 ≤ i ≤ n. Let I = Ker φ be the defining ideal of R. It is known
that I ⊆ (X1 , . . . , Xn )2 and I is generated by the binomials ni=1 Xiαi − ni=1 Xiβi with
Q Q

αi , βi ≥ 0 and ni=1 ai αi =
P Pn L i i+1
i=1 ai βi . Let G = grm (R) = m /m the associated
i≥0
graded ring of R with respect to the maximal ideal m. Then G is a standard graded
ring with the i-th graded component Gi = mi /mi+1 . Let n := (X1 , X2 , . . . , Xn ) denote
the maximal ideal of T . Then φ induces an epimorphism φ̄ : grn (T ) → grm (R). But
L i i+1 ∼
grn (T ) = n /n = k[X1 , X2 , . . . , Xn ] = S. Hence there is an epimorphism from S on
i≥0
G. Denote I ∗ = (f ∗ | f ∈ I \ {0}) the ideal generated by initial forms of I, where f ∗ is the
homogeneous component of f of least degree. Then G ∼ = S/I ∗ = k[X1 , X2 , . . . , Xn ]/I ∗ .
So we can identity Gi = (Si + I ∗ )/I ∗ for all i ≥ 0 so that G = S/I ∗ . The maximal graded
ideal of G is M = n/I ∗ . We call I ∗ is the defining ideal of G. We begin with the following
which implies that G is Cohen-Macaulay, provided R has maximal embedding dimension.
 5

Note that this result is also found in [?, Theorem 2], but we provide a more elementary
proof.
Proposition 4.1. Suppose that R has maximal embedding dimension. Then its associate
graded ring G is Cohen-Macaulay.
Proof. Firstly, we show that m2 = ta1 m. The inclusion m2 ⊇ ta1 m is clear. For converse,
take any tai +aj ∈ m2 , 1 ≤ i, j ≤ n. Clearly, tai +aj ∈ ta1 m once either ai = a1 or aj = a1 .
If ai ̸= a1 and aj ̸= a1 then by Proposition ??(2) we have ai + aj − a1 ∈ H \ {0}. This
yields that tai +aj −a1 ∈ m, whence tai +aj ∈ ta1 m. Thus m2 = ta1 m and hence ms+1 = ta1 ms
for all s ≥ 1.
Denote xi the images of tai in mi /mi+1 , for all i ≥ 0. One has x1 = ta1 + m2 . Suppose
that x1 .(g + mj+1 ) = 0 in G with g ∈ mj for some j ≥ 0. Then (ta1 + m2 )(g + mj+1 ) = 0
in G. This is equivalent to ta1 g ∈ mj+2 . Because mj+2 = ta1 mj+1 , g ∈ mj+1 . This follows
that g + mj+1 = 0 in G. So, x1 is non-zero divisor in G. Hence thanks to Theorem ??, G
is a Cohen-Macaulay ring. □
Lemma 4.2. There is a homogeneous element x ∈ G1 such that (0 :G x) has the finite
length, i.e., ℓG (0 : x) < ∞.
Proof. We consider an ascending sequence of ideals in G
0 :G M ⊆ 0 :G M2 ⊆ · · · .
Since G is Notherian, there exists n0 ≥ 0 such that 0 :G Mn0 = 0 :G Mn0 +1 = · · · .
This implies that L = 0 :G Mn0 , where L = i≥0 (0 :G Mi ). Hence Mn0 L = 0. This
S

yields ℓG (L) < ∞. We set Ass(G/L) = P1 , ..., Pr the set of associated prime ideals of
G/L. If M ∈ G/L then, by the definition of associated prime ideals, there is an element
x+L ∈ G/L, x ∈ / L such that M = 0 : (x+L). Hence xM ⊆ L which implies xMn0 +1 = 0
(because Mn0 L = 0). It follows that x ∈ 0 : Mn0 +1 = L which is a contradiction. This
T
says that M = ̸ Pi for all i = 1, . . . , r. We see that Pi G1 is a proper k-vector space
T
of G1 . Because if otherwise, Pi G1 = G1 for some 1 ≤ i ≤ r. Then Pi ⊇ G1 . In
particular, (Xj + I ∗ )/I ∗ ∈ Pi for all 1 ≤ j ≤ n, whence Pi = M. This is impossible. So,
G1 \ (Pi ∩ G1 ) ̸= ∅ for all 1 ≤ i ≤ r. On the other hand, by the assumption |k| = ∞,
G1 \ ri=1 (Pi ∩ G1 ) ̸= ∅. This implies that there is an element x ∈ G1 , x ∈
S
/ Pi for all
1 ≤ i ≤ r. Since x ∈ / P for all P ∈ Ass(G/L), x is a non-zero divisor on G/L. Now we
take any f ∈ 0 :G x. Then x(f + L) = xf + L = L. Because x is a non-zero divisor on
G/L, we get f + L = L which implies f ∈ L. So, 0 :G x ⊆ L. But ℓG (L) < ∞, we get
that ℓG (0 :G x) < ∞ as desired. □
With the element x as in Lemma ??, we set Ḡ = G/xG.
Proposition 4.3. Suppose R has maximal embedding dimension. Then x is regular on
G and e(G) = e(Ḡ).
6

Proof. We consider two the following exact sequences

G x
0→ (−1) →
− G → Ḡ → 0,
0 :G x

G G
0 → 0 :G x −
→→ Ḡ → .
0 :G x
q(t)
Denote HG (t) the Hilbert series of G. Since dim G = dim R = 1, HG (t) = for some
1−t
q(t) ∈ Q[t] (see [?, Theorem 16.7]), with q(1) = e(G). Using [?, Theorem 16.1], we get

HḠ (t) = HG (t) − H G

(−1) (t)
0:G x

= HG (t) − tH G (t)
0:G x

= HG (t) − t(HH (t) − H0:G x (t))

= (1 − t)HG (t) + tH0:G x (t)
= q(t) + tH0:G x (t). (∗)

Since R has maximal embedding dimension, by Lemma ??, G is Cohen-Macaulay of

dimension 1. This implies H0M (G) = 0, i.e., i≥0 (0 :G Mi ) = 0, or equivalently, L = 0. It
S

implies that 0 :G x = 0 because 0 :G x ⊆ L. This concludes that x is regular on G and Ḡ

is Artinian. Hence , by the equality (∗), we get

e(Ḡ) = HḠ (1) = q(1) = e(G).

We now show that if R has maximal embedding dimension then it is Koszul.

Theorem 4.4. Suppose that H has maximal embedding dimension. Then the semigroup
ring R = [[H]] is Koszul.

Proof. With the element x as in Lemma ??, we set Ḡ = G/xG. By the definition, we
need to show the associated graded ring G of R is a Koszul algebra. Note that G and
Ḡ or both are Koszul or both are not Koszul (see [?, Exercise 34.14]). So, it is sufficient
to show that Ḡ is Koszul. Since x is a homogeneous element of degree 1 in G = S/I ∗ ,
x has the form x = f + I ∗ , where f ∈ S1 . Without loss of generality, we may assume
f = X1 + X2 + · · · + Xs for some 1 ≤ s ≤ n. Then

k[X1 , . . . , Xn ]/I ∗ ∼ k[X1 , . . . , Xn ]

Ḡ = G/xG = ∗
= .
(X1 + · · · + Xs + I ) (X1 + · · · + Xs ) + I ∗
 7

By dividing polynomials in I ∗ by X1 we can write (X1 +· · ·+Xs )+I ∗ = (X1 +· · ·+Xs )+J
with J ⊆ k[X2 , . . . , Xn ]. Then one has
k[X1 , . . . , Xn ]
Ḡ ∼
=
(X1 + · · · + Xs ) + J
∼ k[X1 , . . . , Xn ]/(X1 + · · · + Xs )
=
((X1 + · · · + Xs ) + J)/(X1 + · · · + Xs )
∼
= k[X2 , . . . , Xn ]/J.
The last equality follows from k[X1 , . . . , Xn ]/(X1 + · · · + Xs ) ∼
= k[X2 , . . . , Xn ] and (X1 +
· · · + Xs ) ∩ J = 0.
Thus it is enough to prove that k[X2 , . . . , Xn ]/J is Koszul. Indeed, we have
e(Ḡ) = ℓ(Ḡ) = dimk [X2 , . . . , Xn ]/J
= dimk k[X2 , . . . , Xn ]/(X2 , . . . , Xn )2 − dimk (X2 , . . . , Xn )2 /J
= n − dimk (X1 , . . . , Xn )2 /J.
On the other hand,
n/I ∗
e(G) ≥ edim(G) = dimk ∗ 2
= dimk n/n2 ( because I ∗ ⊆ n2 )
(n/I )
= dimk k[X1 , . . . , Xn ]/(X1 , . . . , Xn )2 − dimk k[X1 , . . . , Xn ]/(X1 , . . . , Xn )
= n + 1 − n = n.
Since H has maximal embedding dimension, so is R. By Proposition ??, we have e(Ḡ) =
n − dimk (X1 , . . . , Xn )2 /J = e(G) ≥ n which implies that (X1 , . . . , Xn )2 /J = 0. Hence
J = (X1 , . . . , Xn )2 . Finally, it is easy to see that k[X2 , . . . , Xn ]/(X1 , . . . , Xn )2 is Koszul
because it has a filtration
F = {0, (X2 ), (X2 , X3 ), . . . , (X2 , X3 , . . . , Xn )}
in the sense of Definition ?? (also by Lemma ??). Thus Ḡ and hence G is Koszul. □
We close this article with the the following examples. Here we use Macaulay2 (see [?])
to compute Betti diagrams.

Example 4.5. Let H = ⟨4, 11, 14, 17⟩ and R = k[H]. Let G is the associated
graded ring of R. Since R has maximal embedding dimension, G is Koszul. The
fact that in this example the defining ideal of G is I ∗ = (u2 , zu, yu, z 2 , yz, y 2 ), that is,
G∼ = k[x, y, z, u]/(u2 , zu, yu, z 2 , yz, y 2 ). We have the Betti diagram of G:
0 1 2 3 4 5
0 1 4 12 36 108 324
This follows that βijG (k) = 0 for all i ̸= j.

Let us give the following two examples. In which the semigroup ring R has no maximal
embedding dimension while one of them is Koszul and the other one is not Koszul.
8

Example 4.6. Let H = ⟨8, 10, 11, 12⟩ and R = k[H]. Let G is the associated graded
ring of R. In this example the defining ideal of G is I ∗ = (u2 , z 2 − yu, y 2 − xu), that is,
G∼ = k[x, y, z, u]/(u2 , z 2 − yu, y 2 − xu). The Betti diagram of G is given:
0 1 2 3 4 5
0 1 4 9 16 25 36
G
This follows that βij (k) = 0 for all i ̸= j which implies that G also is Koszul. Note that in
this example R has no maximal embedding dimension because e(R) = 8 > edim(R) = 4.
Example 4.7. Let H = ⟨12, 14, 15, 16, 18, 19⟩ and R = k[H]. Let G is the associated
graded ring of R. In this example the defining ideal of G is I ∗ = (w2 , v 2 , uv − zw, zv −
yw, u2 − yv, zu − xw, yu − xv, z 2 − xv, y 2 − xu). In this case the Betti diagram of G is as
follows:
0 1 2 3 4 5
0 1 6 24 84 276 877
1 0 0 0 0 1 14
G
We see that β45 (k) = 1 ̸= 0 which implies that G is not Koszul. Note that in this example
R has no maximal embedding dimension because e(R) = 12 > edim(R) = 4.

References
[1] S. S. Abhyankar, Local rings of high embedding dimension, Amer. J. Math., 89(1967), 1073–
1077.
[2] L. L. Avramov and D. Eisenbud, Regularity of modules over a Koszul algebra, J. Algebra,
153(1992), 85–90.
[3] L. L. Avramov and I. Peeva, Finite regularity and Koszul algebras, Amer. J. Math., 123(2001),
275–281.
[4] J. C. Rosales and Garcı́a-Sánchez, , Numerical Semigroups, volume 20, Springer Science &
Business Media, 2009.
[5] A. Conca, E. De Negri, and M. E. Rossi, Koszul Algebras and Regularity, Commutative algebra,
285–315, Springer, New York, 2013.
[6] A. Conca, N. V. Trung, G. Valla, Koszul property for points in projective spaces, 1998, Mathe-
matica Scandinavica.
[7] R. Fröberg, Koszul algebras in Advances in Commutative Ring Theory (Fez, 1997), Lecture
Notes Pure Appl. Math., 205(1999), Dekker, New York, 337–350.
[8] A. Garcia, Cohen-Macaulayness of the associated graded of a semigroup rings, Comm. Algebra,
10 (1982), no. 4, 393–415.
[9] J. Herzog and D. I. Stamate, Quadratic numerical semigroups and the Koszul property, Kyoto
J. Math., 57 (2017), no. 3, 585–612.
[10] E. Kunz, The value-semigroup of a one-dimensional Gorenstein ring, Proc. Amer. Math. Soc.,
25(1970), 748–751.
[11] Macaulay2, a software system for research in algebraic geometry. Available at
http://www.math.uiuc.edu/Macaulay2/
[12] I. Peeva, Graded syzygies, Springer-Verlag London, 2011.
[13] S. B. Priddy, Koszul resolutions, Trans. Amer. Math. Soc., 152(1970), 39–60.
[14] L. Robbiano and G. Valla, On the equations defining tangent cones, Math. Proc. Cambridge
Philos. Soc., 88(1980), no 2, 28l–297.
[15] V. A. Sapko, Associated graded rings of numerical semigroup rings, Comm. Algebra, 29 (2001),
no. 10, 4759–4773.
[16] J. D. Sally, On the associated graded ring of a local Cohen-Macaulay ring, J. Math. Kyoto Univ.,
17 (1977), 19–21.
 9

[17] M. D’Anna, V. Micale, A. Sammartano, On the associated graded ring of a semigroup ring, J.
Commut. Algebra, 3(2011), no. 2, 147–168.
[18] M. D’Anna, V. Micale, A. Sammartano, When the associated graded ring of a semigroup ring
is complete intersection, J. Pure Appl. Algebra, 217(2013), no. 6, 1007–1017.

Department of Mathematics, Hanoi Pedagogical University 2, Vinh Phuc, Vietnam

Email address: dovankien@hpu2.edu.vn, @gmail.com