, , March 6 in 2005

Let

K

be a field and let

S = K [X 1, . . ., X n]

be a polynomial ring with the standard grading. For a graded ideal

I \subset S

we set

R = S / I

. We denote by

x_{i}

the image of

X_{i}

R

for

i = 1, \dots, n

and set

m = (x_{1}, \dots, x_{n})

, the unique graded maximal ideal.

Also

H_{m}^{i} (R)

denotes the local cohomology module of

R

with regard

m

. A residue class ring

R

is called a generalized Cohen-Macaulay ring (generalized CM ring), or FLC (Finite Length Cohomology) ring, if

H_{m}^{i} (R)

has finite length for

i \neq dim R

. In this case, we will call the ideal

I \subset S

a generalized CM ideal.

As defining ideals of algebraic sets, we can find many examples of generalized CM ideals such as homogeneous coordinate rings of non-singular projective varieties.

For monomial ideals, which are not directly related to algebraic sets, the notions of generalized CM rings and Buchsbaum rings [9] coincide in the square-free case and the combinatorial characterization of generalied CM square-free monomial ideals (Stanley-Reisner ideals) has been given in terms Buchsbaum simplicial complexes [7, 8, 9] . However, as far as the author is concerned, the case of non-square-free monomial ideals has not been studied very much, and the aim of this paper is to give combinatorial characterizations of generalized CM monomial ideals, which are not always square-free.

We first give a generalization of Hochster's formula on local cohomologies for square-free monomial ideals [6] to monomial ideals that are not necessarily square-free (Theorem 1 ). From this formula, we can easily deduce several already known and probably new facts on vanishing degrees of local cohomologies. In particular, the vanishing degrees of generalized CM monomial ideals (Proposition 1 ). This result allows us to deduce combinatorial characterizations of generalized CM monomial ideals in terms of the exponents of variables in the monomial generators (Theorem 2 , Corollary 7 and Theorem 3 ). On the other hand, thanks to the generalized Hochster's formula we can compare local cohomologies for

I

and its radical

\sqrt{I}

(Proposition 3 and 4 ), which, together with the combinatorial characterization of generalized CM property, suggests a method to construct generalized CM monomial ideals from Buchsbaum Stanley-Reisner ideals. Namely, by changing a square-free generator

X_{i_{1}} \dots X_{i_{ℓ}}

of a Buchsbaum Stanley-Reisner ideal

J

to a monomial

X_{i_{1}}^{a_{i_{1}}} \dots X_{i_{ℓ}}^{a_{ℓ}}

(a_{j} \in N, j = 1, \dots, ℓ)

, we make a generator of a generalized CM monomial ideal

I

with

\sqrt{I} = J

, and the combinatorial characterization shows a right choice of the exponents

a_{j}

One way of the construction is changing all the occurrences of the variable

X_{i}

in the minimal set of generators to

X_{i}^{a_{i}}

with a fixed exponent

a_{i}

i = 1, \dots, n

(Example 1 ). In some specific case, we can show, using our combinatorial characterization, that this is the only way of construction (Example 2 ).

For a finite set

S

we denote by

| S |

the cardinarity of

S

, and, for sets

A

and

B

A \subset B

means that

A

is a subset of

B

, which may be equal to

A

. The author thanks Jürgen Herzog for valuable discussions and detailed comments on the early version of the paper.

1 Local cohomologies of monomial ideals

1.1 Generalized Hochster's Formula

In this subsection, we give a natural extension of Hochster's formula on local cohomologies of Stanley-Reisiner ideals to monomial ideals. The proof goes along almost the same line as that for Stanley-Reisner ideals given, for example, in [3] chapter 5.3. But we will give a full detail for the readers' convenience.

Let

I \subset S

be a monomial ideal, which is not necessarily square-free. Then we have

H_{m}^{i} (R) \sim = H^{i} (C^{∙})

where

C^{∙}

is the

\overset{ˇ}{C}

ech complex defined as follows:

C^{∙} : 0 - \to C^{0} - \to C^{1} - \to \dots - \to C^{n} - \to 0, C^{t} = \oplus_{1 \leq i_{1} < \dots < i_{t} \leq n} R_{x_{i_{1}} \dots x_{i_{t}}} .

and the differential

C^{t} - \to C^{t + 1}

of this complex is induced by

(- 1)^{s} n a t : R_{x_{i_{1}} \dots x_{i_{t}}} - \to R_{x_{j_{1}} \dots x_{j_{t + 1}}} with {i_{1}, \dots, i_{t}} = {j_{1}, \dots, {\hat{h}}_{s}, \dots, j_{t + 1}}

where

n a t

is the natural homomorphism to localized rings and

R_{x_{i_{1}} \dots x_{i_{t}}}

, for example, denotes localization by

x_{i_{1}}, \dots, x_{i_{t}}

We can consider a

Z^{n}

-grading to

H_{m}^{i} (R)

C^{∙}

and

R_{x_{i_{1}} \dots x_{i_{t}}}

induced by the multi grading of

S

. See for example [3] for more detailed information about this complex.

Now we will consider the degree

a

subcomplex

C_{a}^{∙}

C^{∙}

for any

a \in Z^{n}

. Before that we will prepare the notation. For a monomial ideal

I \subset S

, we denote by

G (I)

the minimal set of monomial generators. Let

u = X_{1}^{a_{1}} \dots X_{n}^{a_{n}}

be a monimial with

a_{i} \geq 0

for all

i

, then we define

ν_{j} (u) = a_{j}

for

j = 1, \dots, n

, and

supp (u) = {i | a_{i} \neq 0}

. We set

G_{a} = {i | a_{i} < 0}

and

H_{a} = {i | a_{i} > 0}

for

a \in Z^{n}

Lemma 1. Let

x = x_{i_{1}} \dots x_{i_{r}}

with

i_{1} < \dots < i_{r}

and set

F = supp (x)

. For all

a \in Z^{n}

we have

{dim}_{K} (R_{x})_{a} \leq 1

and the following are equivalent

$(i)$ $(R_{x})_{a} \sim = K$
$(i i)$ $F \supset G_{a}$ and for all $u \in G (I)$ there exists $j / \in F$ such that $ν_{j} (u) > a_{j} \geq 0$ .

Notice that the condition

a_{i} \geq 0

(i i)

is redundant because this follows from the condition

F \supset G_{a}

. But it is written for the readers' convenience.

Proof. The proof of

{dim}_{K} (R_{x})_{a} \leq 1

is verbatim the same as that of Lemma 5.3.6 (a) in [3] . Now we assume

(i)

, i.e.,

(R_{x})_{a} \neq 0

. This is equivalent to the condition that there exists a monomial

σ \in R

and

ℓ \in N

such that

$(a)$ $x^{m} σ \neq 0$ for all $m \in N$ , and
$(b)$ $deg \frac{σ}{x^{ℓ}} = a$ ,

where

deg

denotes the multidegree. We know from

(b)

that we have

F \supset G_{a}

because a negative degree

a_{i} (< 0)

a

must come from the denominator of the fraction

σ / x^{ℓ}

and

F = supp (x^{ℓ})

. Now we know that

(a)

is equivalent to the following condition:

for all

u \in G (I)

and for all

m \in N

we have

u |̸ (X_{i_{1}}^{m} \dots X_{i_{r}}^{m}) (X_{1}^{b_{1}} \dots X_{n}^{b_{n}})

where we set

σ = x_{1}^{b_{1}} \dots x_{n}^{b_{n}}

with some integers

b_{j} \geq 0

j = 1, \dots, n

. Namely, for all

u \in G (I)

there exists

i / \in F

such that

ν_{i} (u) > b_{i}

. Furthermore, we know from the condition

F \supset G_{a}

that we have

a_{i} = b_{i}

for

i / \in F

since by

(b)

non-negative degrees in

a

must come from

σ

. Consequently we obtain

(i i)

Now we show the converse. Assume that we have

(i i)

. Set

τ = \prod_{i \in H_{a}} x_{i}^{a_{i}}

and

ρ = \prod_{i \in G_{a}} x_{i}^{- a_{i}}

. Then since

F \supset G_{a}

there exists

ℓ \in N

and a monomial

σ

R

such that

\begin{matrix} x^{ℓ} = ρ σ \end{matrix}

(1)

Now we show that

\frac{σ τ}{x^{ℓ}} \neq 0

R_{x}

\frac{σ τ}{x^{ℓ}} \neq 0

is equivalent to the condition that

x^{m} (σ τ) \neq 0

for all

m \in N

. As in the above discussion, this is equivalent to the condition

\begin{matrix} for all u \in G (I) there exists i / \in F such that ν_{i} (u) > b_{i} \end{matrix}

(2)

where we set

σ τ = x_{1}^{b_{1}} \dots x_{n}^{b_{n}}

for some integers

b_{j} \geq 0

j = 1, \dots, n

. But by ( 1 ) we have

i / \in supp (σ)

for

i / \in F

, so that

b_{i} = ν_{i} (τ) = a_{i} (> 0)

(i.e.,

i \in H_{a}

) or

a_{i} = b_{i} = 0

(i.e.,

i / \in H_{a} \cup G_{a}

). Hence we can replace “

ν_{i} (u) > b_{i}

” in ( 2 ) by “

ν_{i} (u) > a_{i} \geq 0

” and then ( 2 ) is assured by the assumption. Thus we have

\frac{σ τ}{x^{ℓ}} \neq 0

R_{x}

. Therefore

deg \frac{σ τ}{x^{ℓ}} = deg \frac{σ τ}{ρ σ} = deg \prod_{i \in H_{a} \cup G_{a}} x_{i}^{a_{i}} = deg x^{a} = a

as requied. □

Let

a \in Z^{n}

. By Lemma 1 we see that

(C^{i})_{a}

has a basis

{b_{F} : \begin{matrix} F \supset G_{a}, | F | = i, \\ and for all u \in G (I) there exists j / ∈ F such that ν_{j} (u) > a_{j} \geq 0 \end{matrix}} .

Restricting the differentation of

C^{∙}

to the

a

th graded piece, we obtain a complex

(C^{∙})_{a}

of finite dimensional

K

-vector spaces with differentation

\partial : (C^{i})_{a} - \to (C^{i + 1})_{a}

given by

\partial (b_{F}) = \sum (- 1)^{σ (F, F^{'})} b_{F^{'}}

where the sum is taken over all

F^{'}

such that

F^{'} \supset F

with

| F^{'} | = i + 1

and for all

u \in G (I)

there exists

j / \in F^{'}

such that

ν_{j} (u) > a_{j} \geq 0

Also we define

σ (F, F^{'}) = s

F^{'} = {j_{0}, \dots, j_{i}}

and

F = {j_{0}, \dots, {\hat{j}}_{s}, \dots, j_{i}}

. Then we describe the

a

th component of the local cohomology in terms of this subcomplex:

H_{m}^{i} (R)_{a} \sim = H^{i} (C^{∙}) = H^{i} (C_{a}^{∙})

Now we fix our notation on simplicial complex. A simplicial complex

Δ

on a finite set

[n] = {1, \dots, n}

is a collection of subsets of

[n]

such that

F \in Δ

whenever

F \subset G

for some

G \in Δ

. Notice that, for the convenience in the later discussions, we do not assume the condition that

{i} \in Δ

for

i = 1, \dots, n

. We define

dim F = i

| F | = i + 1

and

dim Δ = max {dim F | F \in Δ}

, which will be called the dimension of

Δ

. If we assume a linear order on

[n]

, say

1 < 2 < \dots < n

, then we will call

Δ

oriented, and in this case we always denote an element

F = {i_{1}, \dots, i_{k}} \in Δ

with the orderd sequence

i_{1} < \dots < i_{k}

. For a given oriented simplicial complex of dimension

d - 1

, we denote by

C (Δ)

the augumented oriented chain complex of

Δ

C (Δ) : 0 - \to C_{d - 1} - \to \partial C_{d - 2} - \to \dots - \to C_{0} - \to \partial C_{- 1} - \to 0

where

C_{i} = \oplus_{F \in Δ, dim F = i} Z F and \partial F = \sum_{j = 0}^{i} (- 1)^{j} F_{j}

for all

F \in Δ

. Here we define

F_{j} = {i_{0}, \dots, {\hat{i}}_{j}, \dots, i_{k}}

for

F = {i_{0}, \dots, i_{k}}

. Now for an abelian group

G

, we define the

i

th reduced simplicial homology

{\tilde{H}}_{i} (Δ; G)

Δ

to be the

i

th homology of the complex

C (Δ) \otimes G

for all

i

. Also we define the

i

th reduced simplicial cohomology

{\tilde{H}}^{i} (Δ; G)

Δ

to be the

i

th cohomology of the dual chain complex

{Hom}_{Z} (C (Δ), G)

for all

i

. Notice that we have

{\tilde{H}}_{- 1} (Δ; G) = {\begin{matrix} G & if Δ = {\emptyset} \\ 0 & otherwise \end{matrix},

and if

Δ = \emptyset

then

dim Δ = - 1

and

{\tilde{H}}_{i} (Δ; G) = 0

for all

i

Now we will establish an isomorphism between the complex

(C^{∙})_{a}

a \in Z^{n}

, and a dual chain complex. For any

a \in Z^{n}

, we define a simplicial complex

Δ_{a} = {F - G_{a} | \begin{matrix} F \supset G_{a}, and \\ for all u \in G (I) there exists j / ∈ F such that ν_{j} (u) > a_{j} \geq 0 \end{matrix}} .

Notice that we may have

Δ_{a} = \emptyset

for some

a \in Z^{n}

Lemma 2. For all

a \in Z^{n}

there exists an isomorphism of complexes

α^{∙} : (C^{∙})_{a} - \to {Hom}_{Z} (C (Δ_{a}) [- j - 1], K) j = | G_{a} |

where

C (Δ_{a}) [- j - 1]

means shifting the degree of

C (Δ_{a})

- j - 1

Proof. The assignment $F \mapsto F - G_{a}$ induces an isomorphism $α^{∙} : (C^{∙})_{a} - \to {Hom}_{Z} (C (Δ_{a}) [- j - 1], K)$ of $K$ -vector spaces such that $b_{F} \mapsto φ_{F - G_{a}}$ , where $φ_{F^{'}} (F^{”}) = {\begin{matrix} 1 & if F^{'} = F^{”} \\ 0 & otherwise. \end{matrix}$ That this is a homomorphism of complexes can be checked in a straightforward way. □

Now we come to the main theorem in this section.

Theorem 1. Let

I \subset S = K [X 1, . . ., X n]

be a monomial ideal. Then the multigraded Hilbert series of the local cohomology modules of

R = S / I

with respect to the

Z^{n}

-grading is given by

Hilb (H_{m}^{i} (R), t) = \sum_{F \in Δ} \sum {dim}_{K} {\tilde{H}}_{i - | F | - 1} (Δ_{a}; K) t^{a}

where

t = t_{1} \dots t_{n}

, the second sum runs over

a \in Z^{n}

such that

G_{a} = F

and

a_{j} \leq ρ_{j} - 1

j = 1, \dots, n

, with

ρ_{j} = max {ν_{j} (u) | u \in G (I)}

for

j = 1, \dots, n

, and

Δ

is the simplicial complex corresponding to the Stanley-Reisner ideal

\sqrt{I}

Proof. By Lemma 2 and universal coefficient theorem for simplicial (co)homology, we have

\begin{matrix} Hilb (H_{m}^{i} (R), t) & = & \sum_{a \in Z^{n}} {dim}_{K} H_{m}^{i} (R)_{a} t^{a} = \sum_{a \in Z^{n}} {dim}_{K} H^{i} (C_{a}^{∙}) t^{a} \end{matrix}

\begin{matrix} = & \sum_{a \in Z^{n}} {dim}_{K} {\tilde{H}}_{i - | G_{a} | - 1} (Δ_{a}; K) t^{a} . \end{matrix}

\begin{matrix}  \end{matrix}

It is clear from the definition that

Δ_{a} = \emptyset

if for all

j / \in G_{a}

we have

a_{j} \geq ρ_{j}

Moreover for all

a \in Z^{n}

with

a_{j} \geq ρ_{j}

for at least one index

j / \in G_{a}

we have

{dim}_{K} {\tilde{H}}_{i - | G_{a} | - 1} (Δ_{a}; K) = 0

. To prove this fact we can assume without loss of generality that

a_{1} \geq ρ_{1}

and that

Δ_{a} \neq \emptyset

. Then we have

1 / \in G_{a}

, and, for all

σ = (L - G_{a}) \in Δ_{a}

with

L \supset G_{a}

and

1 / \in σ

, we have

σ \cup {1} \in Δ_{a}

. In fact, since we have

ν_{1} (u) \leq a_{1}

for all

u \in G (I)

the existence of

k / \in L

with

ν_{k} (u) > a_{k}

implies

k / \in L \cup {1}

. Consequently we know that

Δ_{a}

is a cone by the vertex

{1}

so that, as is well known, we have

{\tilde{H}}_{i - | G_{a} | - 1} (Δ_{a}; K) = 0

for all

i

as required. Thus we obtain

\begin{matrix} Hilb (H_{m}^{i} (R), t) & = & \sum_{\begin{matrix} a \in Z^{n} \\ a_{j} \leq ρ_{j} - 1 \\ j = 1, \dots, n \end{matrix}} {dim}_{K} {\tilde{H}}_{i - | G_{a} | - 1} (Δ_{a}; K) t^{a} . \end{matrix}

\begin{matrix}  \end{matrix}

Now if

Δ_{a} \neq \emptyset

, we must have

(G_{a} - G_{a} =) \emptyset \in Δ_{a}

, i.e., for all

u \in G (I)

there exists

j / \in G_{a}

such that

ν_{j} (u) > a_{j} \geq 0

, and this implies that

G_{a} ⊅ supp (u)

for all

u \in G (I)

, namely

G_{a}

is not a non-face of

Δ

, i.e.,

G_{a} \in Δ

. Thus we finally obtain the required formula. □

The original Hochster's formula is a special case of Theorem 1 .

Corollary 1 (Hochster). Let

Δ

be a simplicial complex and let

K [Δ]

be the Stanley-Reisner ring corresponding to

Δ

. Then we have

Hilb (H_{m}^{i} (K [Δ]), t) = \sum_{F \in Δ} {dim}_{K} {\tilde{H}}_{i - | F | - 1} ({lk}_{Δ} F; K) \prod_{j \in F} \frac{t_{j}^{- 1}}{1 - t_{j}^{- 1}},

where

{lk}_{Δ} F = {G | F \cup G \in Δ, F \cap G = \emptyset}

Proof. By Theorem 1 we have

Hilb (H_{m}^{i} (R), t) = \sum_{F \in Δ} \sum_{\begin{matrix} a \in Z_{-}^{n} \\ G_{a} = F \end{matrix}} {dim}_{K} {\tilde{H}}_{i - | F | - 1} (Δ_{a}; K) t^{a}

where

Z_{-}^{n} = {a \in Z^{n} | a_{j} \leq 0 for j = 1, \dots, n}

and

\begin{matrix} Δ_{a} & = & {F - G_{a} | \begin{matrix} F \supset G_{a}, and for all u \in G (I) there exists j / ∈ F \\ such that j \in supp (u) and j / ∈ H_{a} \cup G_{a} \end{matrix}} . \end{matrix}

\begin{matrix} = & {F - G_{a} | F \supset G_{a}, and for all u \in G (I) we have H_{a} \cup F ⊅ supp (u)} . \end{matrix}

\begin{matrix} = & {L | L \cap G_{a} = \emptyset, L \cup G_{a} \cup H_{a} \in Δ} = {lk}_{st H_{a}} G_{a} . \end{matrix}

Then the rest of the proof is exactly as in Theorem 5.3.8 [3] . □

1.2 Vanishing degrees of local cohomolgies

In this subsection, we give some easy consequences of Theorem 1 . We define

a_{i} (R) = max {j | H_{m}^{i} (R)_{j} \neq 0}

H_{m}^{i} (R) \neq 0

and

a_{i} (R) = - \infty

H_{m}^{i} (R) = 0

Similarly, we define and

b_{i} (R) = inf {j | H_{m}^{i} (R)_{j} \neq 0}

H_{m}^{i} (R) \neq 0

and

b_{i} (R) = + \infty

H_{m}^{i} (R) = 0

Recall that

ρ_{j} = max {ν_{j} (u) | u \in G (I)}

for

j = 1, \dots, n

Corollary 2. Let

I \subset S = K [X 1, . . ., X n]

be a monomial ideal. Then

a_{i} (R) \leq \sum_{j = 1}^{n} ρ_{j} - n

for all

i

Proof. By Theorem 1 , the terms in $Hilb (H_{m}^{i} (R), t)$ with the highest total degree are at most ${dim}_{K} {\tilde{H}}_{i - | F | - 1} (Δ_{a}; K) t^{a}$ with $a_{j} = ρ_{j} - 1$ for $j = 1, \dots, n$ . Thus the total degree is at most $\sum_{j} ρ_{j} - n$ . □

From Corollary 2 , we can recover the following well known result.

Corollary 3. Let

I \subset S

be a generalized CM Stanley-Reisner ideal.

Then

a_{i} (R) \leq 0

for all

i

Proof. If $I$ is square-free, then $ρ_{j} \leq 1$ for $j = 1, \dots, n$ . □

For a Stanley-Reisner generalized CM ideal

I \subset S

with

dim R = d

, it is well known that it is Buchsbaum and

b_{i} (R) \geq 0

for all

i (\neq d)

. The following theorem extends this result to monomial ideals in general.

Proposition 1. Let

I \subset S = K [X 1, . . ., X n]

be a monomial ideal. Then following are equivalent:

$(i)$ $ℓ (H_{m}^{i} (S / I)) < \infty$
$(i i)$ $H_{m}^{i} (S / I)_{a} = 0$ for all $a \in Z^{n}$ with $G_{a} \neq \emptyset$ , in particular $b_{i} (S / I) \geq 0$
$(i i i)$ ${\tilde{H}}_{i - | G_{a} | - 1} (Δ_{a}; K) = 0$ for all $a \in Z^{n}$ with $a_{j} \leq ρ_{j} - 1$ $(j = 1, \dots, n)$ and $\emptyset \neq G_{a} \in Δ$ .

Proof. The equivalence of $(i i)$ and $(i i i)$ are immediate from Theorem 1 . We will prove the equivalence of $(i)$ and $(i i i)$ . Assume that $ℓ (H_{m}^{i} (S / I)) < \infty$ . Assume also that there exists $a \in Z^{n}$ such that $a_{j} \leq ρ_{j} - 1$ $(j = 1, \dots, n)$ , $\emptyset \neq G_{a} \in Δ$ and ${\tilde{H}}_{i - | G_{a} | - 1} (Δ_{a}; K) \neq 0$ . Now observe that by the definition of $Δ_{a}$ , the condition is independent of the values $a_{j}$ for $j \in G_{a}$ . This means that the total degree $j = \sum_{k = 1}^{n} a_{k}$ can be any negative integer so that $H_{m}^{i} (R)$ is not of finite length, which contradicts the assumption. Thus we must have ${\tilde{H}}_{i - | G_{a} | - 1} (Δ_{a}; K) = 0$ for all such $a \in Z^{n}$ . The converse implication is straightforward. □

Corollary 4. Let

I \subset S = K [X 1, . . ., X n]

be a generalized CM monomial ideal with

dim R = d (> 0)

. Then

b_{i} (R) \geq 0

for all

i (\neq d)

For a generalized CM ring

R

, there exists an integer

k \in Z

k \geq 1

, such that

m^{k} H_{m}^{i} (R) = 0

for

i \neq dim R

. If this condition holds, we will also call

R

, or

I \subset S

k

-Buchsbaum. An ideal

I

is generelized CM if and only if it is

k

-Buchsbaum for some

k

. If

I

k

-Buchsbaum but not

(k - 1)

-Buchsbaum, then we will call

I

strict

k

-Buchsbaum.

Proposition 2. Let

I \subset S = K [X 1, . . ., X n]

be a generalized CM monomial ideal. Then

R = S / I

(\sum_{j = 1}^{n} ρ_{j} - n + 1)

-Buchsbaum.

Proof. $R$ is $max {a_{i} (R) - b_{i} (R) + 1 | i \neq d}$ -Buchsbaum. Then the required result follows immediately from Corollary 2 and Corollary 4 . □

From Proposition 2 , we immediately know that a Stanley-Reisner ideal is 1-Buchsbaum if it is generalized CM, which is a weaker version of the well-known result that a generalized CM Stanley-Reisner ideal is Buchsbaum. The bound of

k

-Buchsbaumness given in Proposition 2 is best possible. In fact, we can construct strict

(\sum_{j = 1}^{n} ρ_{j} - n + 1)

-Buchsbaum ideals as in the following example.

Example 1. Let

I \subset S

be a Stanley-Reisner Buchsbaum ideal. Notice that such ideals can be constructed with the method presented in [1] and

H_{m}^{i} (S / I)

(

i \neq dim R

) is a

K

-vector space for

i \neq dim R

Now consider a

K

-homomorphism

φ : S - \to S, X_{i} \mapsto X_{i}^{a_{i}} (i = 1, \dots, n)

where

a = (a_{1}, \dots, a_{n}) \in Z^{n}

with

a_{i} \geq 1

for

i = 1, \dots, n

. We define

φ (M) = M {\otimes_{S}}^{φ} S

for a

S

-module, where a left-right

S

-module

^{φ} S

is equal to

S

as a set, it is a right

S

-module in the ordinary sense and its left

S

-module structure is determined by

φ

. Then we have

(1) $φ (S / I) = S / φ (I) S$ ,
(2) $φ$ is an exact functor.

Thus, for

i \neq dim R

, we have

H_{m}^{i} (S / φ (I) S) \sim = φ (H_{m}^{i} (S / I))

and since

H_{m}^{i} (S / I)

is a direct sum of

S / m

H_{m}^{i} (S / φ (I) S)

is a direct sum of

S / (X_{1}^{a_{1}}, \dots, X_{n}^{a_{n}})

Then we know that

m^{k} H_{m}^{i} (S / I) = 0

but

m^{k - 1} H_{m}^{i} (S / I) \neq 0

with

k = \sum_{j = 1}^{n} ρ_{j} - n + 1 = \sum_{j = 1}^{n} a_{j} - n + 1

Remark 1. Bresinsky and Hoa gave a bound for

k

-Buchsbaumness for ideals generated by monomials and binomials

(

Theorem 4.5 [2]

)

. For monomial ideals, our bound is stronger than that of Bresinsky and Hoa. Also, according to K. Yanagawa, Proposition 2 can also be deduced from his theory of square-free modules [10] .

Recall that Castelnuovo-Mumford regularity of the ring

R

is defined by

reg (R) = max {i + j | H_{m}^{i} (R)_{j} \neq 0} .

Let

r = reg (R)

. Then we have

H_{m}^{i} (R)_{j} = 0

for

j > r - i

. Then we have

Corollary 5. Let

I \subset S

be a generalized CM monomial ideal with

d = dim R

and

r = reg (R)

. Then

H_{m}^{i} (R) = 0

for

r + 1 \leq i < d

. In particular, if

I

has

q

-linear resolution, we have

H_{m}^{i} (R) = 0

for

q \leq i < d

Proof. First part is clear from Corollary 4 . If $R$ has $q$ -linear resolution, we have $reg (R) = q - 1$ . Thus the last part also follows immediately. □

2 Generalized Cohen-Macaulay monomial ideals

2.1 FLC property

In this subsection, we give a combinatorial characterization of FLC (finite length cohomology) property for monomial ideals, as an application of Theorem 1 . We prepare some notations. Let

I \subset S = K [X 1, . . ., X n]

be a monomial ideal. If

X_{i}^{a} \in G (I)

for some

1 \leq i \leq n

and

a \in N

, we easily know that

a

must be

ρ_{i} = max {ν_{i} (u) | u \in G (I)}

. Then, by changing the name of the variables if necessary, we can write without loss of generality that

G (I) = {X_{m + 1}^{ρ_{m + 1}}, \dots, X_{n}^{ρ_{n}}} \cup G_{0} (I)

with

m \leq n

, where

G_{0} (I) = {u \in G (I) | | supp (u) | \geq 2}

. We denote by

Δ

the simplicial complex corresponding to a square-free monomial ideal

\sqrt{I}

, which is a complex over the vertex set

[m] = {1, \dots, m}

. We regard

Z^{n}

as a partially ordered set by defining

a \leq b

a, b \in Z^{n}

, to be

a_{i} \leq b_{i}

for

i = 1, \dots, n

. We denote

ρ - 1 = (ρ_{1} - 1, \dots, ρ_{n} - 1) \in Z^{n}

. For

a \in Z^{n}

and a monomial

u \in S

, we define

L (a, u) = {i \in [n] | ν_{i} (u) > a_{i}}

. Also for

a \in Z^{n}

with

a \leq ρ - 1

and

σ \subset [n]

, we define

a (σ) \in Z^{n}

as follows:

a (σ)_{i} = {\begin{matrix} a_{i} & if i / ∈ σ \\ ρ_{i} & if i \in σ \end{matrix}

We abbreviate

a ({j})

a (j)

for

j \in Z

Now we prove

Theorem 2. Let

I \subset S = K [X 1, . . ., X n]

be a monomial ideal. If

ℓ (H_{m}^{i} (S / I)) < \infty

(i > 0)

then, for all

(i - 1)

-face

σ \in Δ

and for all

a \in Z^{n}

such that

$(a)$ $0 \leq a \leq ρ - 1$ ,
$(b)$ $L (a (σ), u) \neq \emptyset$ for all $u \in G_{0} (I)$ , and
$(c)$ $a (σ)$ is maximal with the properties $(a)$ and $(b)$ for a fixed $σ$ ,

we have the following: there exists

ℓ \in [m] \ σ

such that

(i) $a_{ℓ} = ρ_{ℓ} - 1$ , and
(ii) for all $u \in G_{0} (I)$ with $ν_{ℓ} (u) = ρ_{ℓ}$ we have $L (a (σ \cup {ℓ}), u) \neq \emptyset$ .

Proof. Assume that

ℓ (H_{m}^{i} (S / I)) < \infty

. Then, by Proposition 1 , we have

{\tilde{H}}_{- 1} (Δ_{a}; K) = 0

for all

a \in Z^{n}

with

a \leq ρ - 1

G_{a} \in Δ

and

| G_{a} | = i (> 0)

. This implies

Δ_{a} \neq \emptyset

Now for such

a \in Z^{n}

we set

σ = G_{a}

. Notice that

σ \subset [m]

since

σ \in Δ

. We also notice that, as far as the complex

Δ_{a}

is concerned, the values

a_{i}

for

i \in σ = G_{a}

are irrelevant. Thus we will change the values

a_{i}

(i \in σ)

and assume that

0 \leq a \leq ρ - 1

Notice that

Δ_{a}

, for the new

a

, is the same as before we change the values

a_{i}

for

i \in σ

We know that

Δ_{a} = \emptyset

is equivalent to the condition that there exists

u \in G (I)

such that

L (a (σ), u) = \emptyset

. Now we assume that

L (a (σ), u) \neq \emptyset

for all

u \in G (I)

, namely

Δ_{a} \neq \emptyset

The condition

Δ_{a} \neq {\emptyset}

is equivalent to the condition that

{ℓ} \in Δ_{a}

for some

ℓ \in [n]

, i.e., there exists

ℓ \in [n] \ σ

such that for all

u \in G (I)

we can find

k \in [n] \ (σ \cup {ℓ})

satisfying

ν_{k} (u) > a_{k}

. Namely,

\begin{matrix} there exists ℓ \in [n] \ σ such that L (a (σ \cup {ℓ}), u) \neq \emptyset for all u \in G (I) . \end{matrix}

(3)

Under this condition we have, for any

b \in Z^{n}

with

0 \leq b \leq ρ - 1

and

b (σ) \leq a (σ)

L̷ (b (σ \cup {ℓ}), u) \supseteq L (a (σ \cup {ℓ}), u) \neq \emptyset

for all

u \in G (I)

. Thus we can assume that

a (σ)

is maximal satisfying the condition that

L (a (σ), u) \neq \emptyset

for all

u \in G (I)

and

0 \leq a \leq ρ - 1

. Also, since

σ \subset [m]

and

a \leq ρ - 1

, we have

L (a (σ), X_{i}^{ρ_{i}}) = {i} \neq \emptyset

for all

m + 1 \leq i \leq n

. Hence we can replace '

u \in G (I)

' by '

u \in G_{0} (I)

' in the maximality condition for

a (σ)

. Now we have only to show that the condition ( 3 ) is equivalent to

(i)

and

(i i)

in the statement.

Since we have

L (a (σ \cup {ℓ}), X_{j}^{ρ_{j}}) = \emptyset

for all

m + 1 \leq j \leq n

, we can only find the index

ℓ

as in 3

()

[m] \ σ

. Now for

ℓ \in [m] \ σ

, the existence of

k \in [n] \ (σ \cup {ℓ})

satisfying

ν_{k} (u) > a_{k}

is always assured for every

u \in {X_{m + 1}^{ρ_{m + 1}}, \dots, X_{n}^{ρ_{n}}}

. Moreover, if

ℓ / \in supp (u)

for

u \in G_{0} (I)

, then

L (a (σ \cup {ℓ}), u) = L (a (σ), u)

and this is

\neq \emptyset

since

Δ_{a} \neq \emptyset

. Thus 3

()

is equivalent to the existence of

ℓ \in [m] \ σ

such that

\begin{matrix} L (a (σ \cup {ℓ}), u) \neq \emptyset for u \in G_{0} (I) with ℓ \in supp (u) . \end{matrix}

(4)

Assume that

a_{ℓ} < ρ_{ℓ} - 1

and set

e \in Z^{n}

e_{i} = 0

for

i \neq ℓ

and

e_{ℓ} = ρ_{ℓ} - 1 - a_{ℓ}

Then by the maximality of

a (σ)

there exists

u \in G_{0} (u)

such that

\emptyset = L (a (σ) + e, u) \supset L (a (σ \cup ℓ), u)

, which contradicts the condition 3

()

. Thus we must have

a_{ℓ} = ρ_{ℓ} - 1

for

ℓ

as in 3

()

. If

u \in G_{0} (I)

is such that

0 < ν_{ℓ} (u) < ρ_{ℓ}

, then

L (a (σ) \cup {ℓ}, u) = L (a (σ), u) \neq \emptyset

by assumption on

a (σ)

. Thus we can replace '

u \in G_{0} (I)

with

ℓ \in supp (u)

' in the condition ( 4 ) by '

u \in G_{0} (I)

with

ν_{ℓ} (u) = ρ_{ℓ}

Consequently we know that 4

()

is equivalent to

(i)

and

(i i)

. □

From Theorem 2 , we can recover a weaker version of the well-known result as follows.

Corollary 6. If

I \subset S

is a generalized CM Stanley-Reisner ideal, i.e., Buchsbaum ideal, then

Δ

is pure, namely, every facet has the same dimension.

Proof. Let $I \subset S$ be a generalized CM Stanley-Reisner ideal. Then by Theorem 2 we know that for every $0 < i < dim S / I$ and for every $(i - 1)$ -face $σ \in Δ$ there exists $ℓ$ with $1 \leq ℓ \leq n$ such that $σ \cup {ℓ} \in Δ$ . From this we immediately know that $Δ$ is pure. □

2.2 generalized CM monomial ideals of $dim \leq 3$

dim R \leq 1

I

is always (generalized) CM. For

d i m R = 2, 3

, we can give combinatorial characterizations of generalized CM monomial ideals as follows. First we give the

dim 2

case.

Corollary 7. A monomial ideal

I \subset S

is generalized CM with

dim S / I = 2

if and only if

$(i)$ $dim Δ = 1$ , and
$(i i)$ for all j ∈ [ m ] and for all a ∈ Z n such that ( a ) 0 ≤ a ≤ ρ − 1 , ( b ) L ( a ( j ) , u ) ≠ ∅ for all u ∈ G 0 ( I ) , and ( c ) a ( j ) is maximal with the properties ( a ) and ( b ) for a fixed j , we have the following: there exists ℓ ∈ [ m ] \ { j } such that
- (i) $a_{ℓ} = ρ_{ℓ} - 1$ , and
- (ii) for all $u \in G_{0} (I)$ with $ν_{ℓ} (u) = ρ_{ℓ}$ we have $L (a ({j, ℓ}), u) \neq \emptyset$ .

Proof. As is well-known, $dim S / I = dim S / \sqrt{I} = 2$ if and only if $dim Δ = 1$ . Then $S / I$ is generalized CM if and only if $ℓ (H_{m}^{i} (S / I)) < \infty$ for $i = 0, 1$ . $H_{m}^{0} (S / I)$ is always of finite length and $H_{m}^{1} (S / I)$ is of finite length if and only if ${\tilde{H}}_{- | G_{a} |} (Δ_{a}; K) = 0$ for all $a \in Z^{n}$ with $a \leq ρ - 1$ and $\emptyset \neq G_{a} \in Δ$ by Proposition 1 . If $| G_{a} | \geq 2$ , we always have ${\tilde{H}}_{- | G_{a} |} (Δ_{a}; K) = 0$ . Now let $G_{a} = {j}$ . Since $G_{a} \in Δ$ we must have $j \in [m]$ .
Also ${\tilde{H}}_{- 1} (Δ_{a}; K) = 0$ if and only if $Δ_{a} \neq \emptyset$ , which is equivalent to the condition in the statement by the proof of Theorem 2 . □

dim 3

case, we need to give a combinatorial criterion for connectedness of simplicial complexes. Notice that a simplicial complex

Δ

over the vertex set

[m]

is not connected if and only if there exists non-empty disjoint subsets

P, Q \subset [m]

such that

P \cup Q = [m]

and for all

p \in P

and all

q \in Q

there is no 1-face

{p, q} \in Δ

Lemma 3. Assume that

G_{a} = {j}

j \in [m]

, and

Δ_{a} \neq \emptyset

. Then the set of vertices of

Δ_{a}

{ℓ \in [m] \ {j} | for all u \in G_{0} (I) we have L (a ({ℓ, j}), u) \neq \emptyset}

Proof. The

0

th skeleton of

Δ_{a}

\begin{matrix} {{ℓ} | ℓ \neq j, for all u \in G (I) there exists k / \in {ℓ, j} such that ν_{k} (u) > a_{k}} \end{matrix}

\begin{matrix} = & {{ℓ} | ℓ \in [m] \ {j}, for all u \in G_{0} (I) there exists k / \in {ℓ, j} such that ν_{k} (u) > a_{k}} \end{matrix}

\begin{matrix} = & {{ℓ} | ℓ \in [m] \ {j}, for all u \in G_{0} (I) we have L (a ({ℓ, j}), u) \neq \emptyset}, \end{matrix}

where the first equation is because if

m + 1 \leq ℓ \leq n

there is no

k / \in {ℓ, j}

such that

ν_{k} (X_{ℓ}^{ρ_{ℓ}}) > a_{k}

and if

ℓ \in [m]

we always have the index

k / \in {ℓ, j}

such that

ν_{k} (X_{i}^{ρ_{i}}) > a_{i}

for

i = m + 1, \dots, n

, which is actually

k = i

. □

Now we show a combinatorial characterization of

dim 3

generalized CM monomial ideals.

Theorem 3. A monomial ideal

I \subset S

is generalized CM with

dim S / I = 3

if and only if

$(i)$ $dim Δ = 2$ , and
$(i i)$ for all j ∈ [ m ] and for all a ∈ Z n such that ( a ) 0 ≤ a ≤ ρ − 1 , ( b ) L ( a ( j ) , u ) ≠ ∅ for all u ∈ G 0 ( I ) , and ( c ) a ( j ) is maximal with the properties ( a ) and ( b ) for a fixed j , we have
- $(1)$ there exists ℓ ∈ [ m ] \ { j } such that
  - 1. $a_{ℓ} = ρ_{ℓ} - 1$ , and
  - 2. for all $u \in G_{0} (I)$ with $ν_{ℓ} (u) = ρ_{ℓ}$ we have $L (a ({j, ℓ}), u) \neq \emptyset$
- $(2)$ there are no non-empty disjoint subsets P , Q ⊂ [ m ] satisfying the following property:
  - 1. $P \cup Q = [m] \ {j} - L_{a}$ where $L_{a} = {ℓ | L (a (j), u) = {ℓ} for some u \in G_{0} (U)}$ , and
  - 2. for all $x \in P$ and all $y \in Q$ there exists $u \in G_{0} (I)$ such that $L (a (j), u) = {x, y}$
$(i i i)$ for all 1-face σ = { i , j } ∈ Δ and for all a ∈ Z n such that ( a ) 0 ≤ a ≤ ρ − 1 , ( b ) L ( a ( { i , j } ) , u ) ≠ ∅ for all u ∈ G 0 ( I ) , and ( c ) a ( { i , j } ) is maximal with the properties ( a ) and ( b ) for a fixed { i , j } , we have the following: there exists ℓ ∈ [ m ] \ { i , j } such that
- 1. $a_{ℓ} = ρ_{ℓ} - 1$ , and
- 2. for all $u \in G_{0} (I)$ with $ν_{ℓ} (u) = ρ_{ℓ}$ we have $L (a ({i, j, ℓ}), u) \neq \emptyset$ .

Proof.

dim S / I = dim S / \sqrt{I} = 3

if and only if

dim Δ = 2

. Now assume that

dim S / I = 3

. Then

S / I

is generalized CM if and only if

ℓ (H_{m}^{1} (S / I)) < \infty

and

ℓ (H_{m}^{2} (S / I)) < \infty

, which is equivalent to

\begin{matrix} {\tilde{H}}_{- | G_{a} |} (Δ_{a}; K) = 0 \end{matrix}

(5)

and

\begin{matrix} {\tilde{H}}_{1 - | G_{a} |} (Δ_{a}; K) = 0 \end{matrix}

(6)

for all

a \in Z^{n}

with

a \leq ρ - 1

and

\emptyset \neq G_{a} \in Δ

by Proposition 1 . The condition 5

()

is equivalent to

(i i) (1)

by Corollary 7 and the condition 6

()

is equivanent to

\begin{matrix} {\tilde{H}}_{0} (Δ_{a}; K) = 0 for all a \in Z^{n} with a \leq ρ - 1 and G_{a} = {j} \in [m], \end{matrix}

(7)

and

\begin{matrix} {\tilde{H}}_{- 1} (Δ_{a}; K) = 0 for all a \in Z^{n} with a \leq ρ - 1 and G_{a} = {i, j} \in [m] \end{matrix}

(8)

since

{\tilde{H}}_{- k} (Δ_{a}; K) = 0

for

k \geq 2

. The condition 7

()

exactly means the connectedness of the simplicial complex

Δ_{a}

. Let

V_{a}

be the set of vertices of

Δ_{a}

. By what we noticed just before Lemma 3 , this is equivalent to the condition that there exist disjoint no non-empty subsets

P, Q \subset V_{a}

such that

P \cup Q = V_{a}

and for all

x \in P

and all

y \in Q

we have

{x, y} / \in Δ_{a}

. By Lemma 3 we have

V_{a} = [m] \ {j} - L_{a}

where

L_{a} = {ℓ | ℓ \neq j, L (a ({j, ℓ}), u) = \emptyset for some u \in G_{0} (I)}

. Since we have

L (a (j), u) \neq \emptyset

L (a ({j, ℓ}), u) = \emptyset

implies

L (a (j), u) = {ℓ}

. Thus

L_{a} = {ℓ | L (a (j), u) = {ℓ} for some u \in G_{0} (I)}

. The condition

{x, y} / \in Δ_{a}

is equivalent to the condition that

L (a ({x, y, j}), u) = \emptyset

for some

u \in G (I)

. This can also be refined to the condition that

L (a (j), u) = {x, y}

for some

u \in G_{0} (I)

. In fact, first of all we have

L (a ({x, y, j}), X_{i}^{ρ_{i}}) = {i} \neq \emptyset

, for all

i = m + 1, \dots, n

, since

x, y, j \in [m]

and

a_{i} \leq ρ_{i} - 1

. Thus we can replace '

u \in G (I)

' in the above condition by '

u \in G_{0} (I)

'. Also, since

x, y \in V_{a}

L (a ({x, j}), u) \neq \emptyset

and

L (a ({y, j}), u) \neq \emptyset

for all

u \in G_{0} (I)

. Thus

L (a ({x, y, j}), u) = \emptyset

is equivalent to

L (a (j), u) = {x, y}

as required. This is the condition

(i i) (2) .

Now we will show that if

b \in Z^{n}

is such that

b \leq ρ - 1

G_{b} = {j}

L (b (j), u) \neq \emptyset

for all

u \in G_{0} (I)

and

b (j) \leq a (j)

, then

Δ_{b}

is also connected. We prove the contrapositon: if

Δ_{b}

is disjoint then

Δ_{a}

is disjoint too. Assume that there exist disjoint non-empty subsets

P, Q \subset V_{b}

such that

P \cup Q = V_{b}

and for all

x \in P

and all

y \in Q

we have

L (b (j), u) = {x, y}

for some

u \in G_{0} (I)

. First of all, for

u \in G_{0} (I)

we have

L (b ({j, ℓ}), u) \supseteq L (a {j, ℓ}, u)

so that

L_{b} \subseteq L_{a}

and thus

V_{a} \subseteq V_{b}

. Now for all

x \in P \cap V_{a}

and

y \in Q \cap V_{a}

L (a (j), u) \subset L (b (j), u) = {x, y}

for some

u \in G_{0} (I)

Also since

x, y \in V_{a}

we must have

L (a ({j, x}), u) \neq \emptyset

and

L (a ({j, y}), u) \neq \emptyset

. Then we know that we must have

L (a (j), u) = {x, y}

. Thus, by setting

P^{'} = P \cap V_{a}

and

Q^{'} = Q \cap V_{a}

, we obtain the non-empty disjoint subsets

P^{'}, Q^{'} \subset V_{a}

showing the disjointness of

Δ_{a}

. Consequently, we can assume

a (j)

to be maximal as in the statement

(i i) (c)

. Finally, by the proof of Theorem 2 , we know that the condition 8

()

is equivalent to

(i i i)

. □

Remark 2. Unfortunately we do not know a good combinatorial characterization for

{\tilde{H}}_{j} (Δ_{a}; K) = 0

for

j \geq 1

, which is needed to obtain similar results to Theorem 3 for

dim R \geq 4

2.3 Construction from Buchsbaum Stanley-Reisner ideals

In this subsection, we compare local cohomologies of monomial ideals

I \subset S

and

\sqrt{I}

. It is well known that

H_{m}^{i} (S / \sqrt{I})_{a} = 0

for all

a \in Z^{n}

with

H_{a} \neq \emptyset

, which is an immediate consequence from the original version of Hochster's formula for Stanley-Reisner ideals. On the other hand, we may have

H_{m}^{i} (S / I)_{a} \neq 0

for such

a \in Z^{n}

. But for multi-degrees

a \in Z^{n}

with

H_{a} = \emptyset

, we have an isomorphism.

Proposition 3 (Herzog-Takayama-Terai [4] ). Let

I \subset S

be a monomial ideal. Then we have the following isomorphisms of

K

-vector spaces

H_{m}^{i} (S / I)_{a} \sim = H_{m}^{i} (S / \sqrt{I})_{a}

for all

a \in Z^{n}

with

H_{a} = \emptyset

Proposition 4 (Herzog-Takayama-Terai [4] ). Let

I \subset S

be a monomial ideal. Then

ℓ (H_{m}^{i} (S / I)) < \infty

implies

ℓ (H_{m}^{i} (S / \sqrt{I})) < \infty

. In particular, if

I

is generalized CM, then

\sqrt{I}

is also generalized CM (Buchsbaum).

Proof. We will give here a new proof, which is different from that in [4] . Assume that $ℓ (H_{m}^{i} (S / I)) < \infty$ . Then by Proposition 1 $H_{m}^{i} (S / I)_{a} = 0$ for all $a \in Z^{n}$ with $G_{a} \neq \emptyset$ . Thus if $H_{m}^{i} (S / I)_{a} \neq 0$ and $H_{a} = \emptyset$ , we must have $a = (0, \dots, 0)$ . Now by Proposition 3 we have $H_{m}^{i} (S / \sqrt{I})_{a} \sim = H_{m}^{i} (S / I)_{a}$ for all $a \in Z^{n}$ with $H_{a} = \emptyset$ , and this is non-zero if and only if $a = (0, \dots, 0)$ . Thus we have $ℓ (H_{m}^{i} (S / \sqrt{I})) < \infty$ . □

Corollary 8. Let

I \subset S

be a generalized CM monimial ideal and assume that

\sqrt{I}

is not Cohen-Macaulay (but generalized CM by Proposition 4 ).

Then

I

is not Cohen-Macaulay.

Proof. Assume that $I$ is Cohen-Macaulay. Then, by Proposition 3 and the comment before Proposition 3 , we have $H_{m}^{i} (S / \sqrt{I})_{a} = 0$ for all $i < dim S / \sqrt{I}$ and for all $a \in Z^{n}$ , namely $H_{m}^{i} (S / \sqrt{I}) = 0$ for all $i < dim S / \sqrt{I}$ and $\sqrt{I}$ is Cohen-Macaulay. □

These results suggests a method for constructing (non-CM) generalized CM monomial ideals from Buchsbaum Stanley-Reisner ideals: given a Buchsbaum Stanley-Reisner ideal

J

, make monomials

X_{j_{1}}^{e_{1}} \dots X_{j_{p}}^{e_{p}}

(e_{i} \geq 1, i = 1, \dots, n)

, for each generator

X_{j_{1}} \dots X_{j_{p}} \in G (J)

. In general, one can make more than one monomial generators from a single square-free generator. If we choose suitable exponents

e_{i}

, the ideal generated by the monomials is (non-CM) generalized CM. Theorem 2 , Corollary 7 and Theorem 3 give the criteria for suitable exponents.

Example 2. Let

J = (X_{1}, \dots, X_{n}) (X_{n + 1}, \dots, X_{2 n}) \subset S = K [X 1, . . ., X 2 n]

(n \geq 2)

. It is easy to check that

S / I

is Buchsbaum of dimension

n

and depth

1

. Let

I = (X_{i}^{α_{i j}} X_{j}^{β_{i j}} | 1 \leq i \leq n, n + 1 \leq j \leq 2 n)

for some

α_{i j}, β_{i j} \in N

Then

I

is generalized CM ideal if and only if

α_{i, n + 1} = \dots = α_{i, 2 n}

for all

1 \leq i \leq n

and

β_{1, j} = \dots = β_{n, j}

for all

n + 1 \leq j \leq 2 n

, namely

I

is an image of Frobenius map in the sense of Example 1 .

Remark 3. Notice that if

n = 1

then both

J

and

I = (X_{1}^{α_{11}} X_{2}^{β_{11}})

are Cohen-Macaulay for all

α_{11}, β_{11} \in N

Proof of Example 2 . Assume in the following that

I

is generalized CM. Then

I

must satisfy the conditions in Theorem 2 , in particular the condition for

ℓ (H_{m}^{n - 1} (S / I)) < \infty

Now we set

ρ_{i} = max {α_{i j} | n + 1 \leq j \leq 2 n}

and

ɛ_{i} = min {α_{i j} | n + 1 \leq j \leq 2 n}

for

1 \leq i \leq n

and

ρ_{j} = max {β_{i j} | 1 \leq i \leq n}

and

ɛ_{j} = min {β_{i j} | 1 \leq i \leq n}

for

n + 1 \leq j \leq 2 n

. Notice that

ρ_{k} (\geq 1)

and

ɛ_{k} (\geq 1)

denote the maximal and the minimal exponents of the variable

X_{k}

k = 1, \dots, 2 n

. Let

Δ

be the simplicial complex corresponding to

J

, which is the disjoint union of two

(n - 1)

-simplices over the vertex set

[n]

and

[2 n] \ [n]

. Then an

(n - 2)

-face

σ \in Δ

is either

σ = {1, \dots, n} \ {k}

for

k = 1, \dots, n

σ = {n + 1, \dots, 2 n} \ {k}

for

k = n + 1, \dots, 2 n

We know that the condition for

a \in Z^{2 n}

such that

0 \leq a \leq ρ - 1

to be

L (a (σ), u) \neq \emptyset

for all

u \in G (I) = G_{0} (I)

is as follows:

Case (1)

σ = {1, \dots, n} \ {k}

for

1 \leq k \leq n

Since

\begin{matrix} L (a (σ), X_{i}^{α_{i j}} X_{j}^{β_{i j}}) \end{matrix}

\begin{matrix} = & {\begin{matrix} {k, j} & if i = k, a_{k} < α_{k j}, a_{j} < β_{k j} \\ {j} & if i = k, a_{k} \geq α_{k j}, a_{j} < β_{k j}, or if i \neq k, a_{j} < β_{i j} \\ {k} & if i = k, a_{k} < α_{k j}, a_{j} \geq β_{k j} \\ \emptyset & otherwise, \end{matrix} \end{matrix}

we must have

(1) $a_{j} < min {β_{i j} | 1 \leq i \leq n, i \neq k}$ for all $n + 1 \leq j \leq 2 n$ , and
(2) for every $n + 1 \leq j \leq 2 n$ we have at least one of the followings: (a) $a_{k} < α_{k j}$ , (b) $a_{j} < β_{k j}$ .

Case (2)

σ = {n + 1, \dots, 2 n} \ {k}

for

n + 1 \leq k \leq 2 n

Since

\begin{matrix} L (a (σ), X_{i}^{α_{i j}} X_{j}^{β_{i j}}) \end{matrix}

\begin{matrix} = & {\begin{matrix} {k, i} & if j = k, a_{k} < β_{i k}, a_{i} < α_{i k} \\ {i} & if j = k, a_{k} \geq β_{i k}, a_{i} < α_{i k}, or if j \neq k, a_{i} < α_{i j} \\ {k} & if j = k, a_{k} < β_{i k}, a_{i} \geq α_{i k} \\ \emptyset & otherwise, \end{matrix} \end{matrix}

we must have

(1) $a_{i} < min {α_{i j} | n + 1 \leq j \leq 2 n, j \neq k}$ for all $1 \leq i \leq n$ , and
(2) for every $1 \leq i \leq n$ we have at least one of the followings: (a) $a_{k} < β_{i k}$ , (b) $a_{i} < α_{i k}$ .

According to Theorem 2 , for a maximal

a (σ)

, where

a

and

σ

are as above, there must exist an index

ℓ \in [2 n] \ σ

such that

(i)

a_{ℓ} = ρ_{ℓ} - 1

and

(i i)^{'}

for every

u = X_{i}^{α_{i j}} X_{j}^{β_{i j}} \in G (I)

with

ν_{ℓ} (u) = ρ_{ℓ}

we have

L (a (σ \cup {ℓ}), u) \neq \emptyset

. Moreover, by the proof of the Theorem 2 we know that

(i i)^{'}

can be replaced by

(i i) for every u \in G (I) with ℓ \in supp (u) we have L (a (σ \cup {ℓ}), u) \neq \emptyset .

We now consider the condition for

a

α_{i j}

and

β_{i j}

satisfying

(i)

and

(i i)

Case (3) $1 \leq ℓ \leq n$ with $ℓ / \in σ$ By $(i i)$ we must have $\emptyset \neq L (a (σ \cup {ℓ}), X_{ℓ}^{α_{ℓ j}} X_{j}^{β_{ℓ j}}) (\subset {j})$ for all $n + 1 \leq j \leq 2 n$ . This holds if and only if $j / \in σ$ and $a_{j} < β_{ℓ, j}$ .
Case (4) $n + 1 \leq ℓ \leq 2 n$ with $ℓ / \in σ$ Similarly, we have and $a_{i} < α_{i, ℓ}$ for any $1 \leq i \leq n$ with $i / \in σ$ .

Now in case (1), an

ℓ \in [2 n] \ σ

satisfying

(i)

and

(i i)

must be

ℓ = k

(1 \leq k \leq n)

ℓ \in {n + 1, \dots, 2 n}

. Assume that

ℓ \in {n + 1, \dots, 2 n}

. Then the condition of case (1) must imply the condition of case (4). Since

\begin{matrix} max {β_{i ℓ} | 1 \leq i \leq n, i \neq k} - 1 & \leq & max {β_{i ℓ} | 1 \leq i \leq n} - 1 = ρ_{ℓ} - 1 = a_{ℓ} \end{matrix}

\begin{matrix} < & min {β_{i ℓ} | 1 \leq i \leq n, i \neq k}, \end{matrix}

we have

max {β_{i ℓ} | 1 \leq i \leq n, i \neq k} = min {β_{i ℓ} | 1 \leq i \leq n, i \neq k}

so that

β_{i ℓ}

is constant for all

0 \leq i \leq n

with

i \neq k

. Now if

β_{k ℓ} > β_{i ℓ}

for some, and equivalently all,

i (\neq k)

, then we have

ρ_{ℓ} = max {β_{i ℓ} | 1 \leq i \leq n} = β_{k ℓ}

so that

β_{k ℓ} - 1 = ρ_{ℓ} - 1 = a_{ℓ} < min {β_{i ℓ} | 1 \leq i \leq n, i \neq k} = β_{i ℓ}

. Then we have

β_{k ℓ} \leq β_{i ℓ}

, a contradiction. Thus we know

β_{k ℓ} \leq β_{i ℓ}

for all

i \neq k

, and the index

i

as in the condition of case (4) can be at least

i = 1, \dots, n

with

i \neq k

but this contradicts the condition

i / \in σ = {1, 2, \dots, n} \ {k}

. Consequently, an

ℓ \in [2 n] \ σ

satisfying

(i)

and

(i i)

cannot be from

{n + 1, \dots, 2 n}

and we must have

ℓ = k

with

1 \leq k \leq n

Now the condition of Case (1) must imply the condition of Case (3). Comparing Case (1) 1 with the condition of Case (3), we know that we must have

\begin{matrix} min {β_{i j} | 1 \leq i \leq n, i \neq k} \leq β_{k j} \end{matrix}

(9)

for every

j

with

n + 1 \leq j \leq 2 n

. Now we consider similarly with the Case (2) and obtain the condition

\begin{matrix} min {α_{i j} | n + 1 \leq j \leq 2 n, j \neq k} \leq α_{i k} \end{matrix}

(10)

for every

i

with

1 \leq i \leq n

. Finally, the condition 9

()

for

k = 1, \dots, n

together with the condition 10

()

for

k = n + 1, \dots, 2 n

entails

β_{i j}

are constant for all

1 \leq i \leq n

and

α_{i j}

are constant for all

n + 1 \leq j \leq 2 n

, i.e.,

I

is obtained from

J

by Frobenius transformation in the sense of Example 1 . □

Example 3. If we allow to make more than two monomial generators from a single square-free generator, we can construct more generalized CM monomial ideals from the same Stanley-Reisner ideals as in Example 2 , For example, from

J_{1} = (X_{1}, X_{2}) (X_{3}, X_{4}) \subset K [X_{1}, X_{2}, X_{3}, X_{4}]

we make

I_{1} = (X_{1} X_{3}, X_{1}^{2} X_{4}, X_{1} X_{4}^{2}, X_{2}^{2} X_{3}, X_{2} X_{3}^{2}, X_{2} X_{4}) .

Also from

J_{2} = (X_{1}, X_{2}, X_{3}) (X_{4}, X_{5}, X_{6}) \subset K [X_{1}, \dots, X_{6}]

we make

I_{2} = (X_{1}^{3} X_{4}, X_{1} X_{4}^{5}, X_{1} X_{5}, X_{1} X_{6}, X_{2} X_{4}, X_{2} X_{5}, X_{2} X_{6}, X_{3} X_{4}, X_{3} X_{5}, X_{3} X_{6}) .

I_{1}

and

I_{2}

are both generalized CM, but for example

I_{3} = (X_{1}^{3} X_{4}, X_{1}^{2} X_{4}^{2}, X_{1} X_{4}^{3}, X_{1} X_{5}, X_{1} X_{6}, X_{2} X_{4}, X_{2} X_{5}, X_{2} X_{6}, X_{3} X_{4}, X_{3} X_{5}, X_{3} X_{6})

is not generalized CM.

References

A. Björner and T. Hibi, Betti numbers of Buchsbaum complexes, Math. Scand. 67 (1990) no. 2, 193–196.
H. Bresinsky and L. T. Hoa, The $k$ -Buchsbaum property for some polynomial ideals, J. Math. Kyoto Univ. 43-4 (2003), 699–717.
W. Bruns and J. Herzog, Cohen-Macaulay rings, Cambridge studies in advanced mathematics 39, Cambridge, 1993.
J. Herzog, Y. Takayama and N. Terai, On the radical of a monomial ideals, preprint, (2004).
J. Herzog, D. Popescu and Marius Vladoiu, On the Ext-modules of ideals of Borel-type, Contemp. Math. 331 (2003), 171-186.
M. Hochster, Cohen-Macaulay rings, combinatorics, and simplicial complexes. In B. R. McDonald and R. A. Morris (eds.), Ring theory II. Lect. Notes in Pure and Appl. Math. 26, M. Dekker, (1977) 171–223.
P. Schenzel, On the number of faces of simplicial complexes and the purity of Frobenius, Math. Z., 178, (1981), 125–142.
R. P. Stanley, Combinatorics and commutative algebra, Birkhäuser, second edition, 1996.
J. Stückrad and W. Vogel, Buchsbaum rings and applications, Springer, 1986.
K. Yanagawa, Alexander duality for Stanely-Reisner rings and squarefree $N^{n}$ -graded modules, J. Algebra 225, No. 2, (2000), 630-645.

Yukihide Takayama, Department of Mathematical Sciences, Ritsumeikan University, 1-1-1 Nojihigashi, Kusatsu, Shiga 525-8577, Japan E-mail address : takayama@se.ritsumei.ac.jp