, , , Mar.10, 2005

Throughout this paper, let

S = k [X_{1}, \dots, X_{n}]

be a homogeneous polynomial ring over a field

k

with

deg X_{i} = 1

. For a simplicial complex

Δ

on vertex set

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

(note that

{i} \in Δ

for all

i

k [Δ] = k [X_{1}, \dots, X_{n}] / I_{Δ}

is called the Stanley–Reisner ring of

Δ

, where

I_{Δ}

is an ideal generated by all square-free monomials

X_{i_{1}} \dots X_{i_{p}}

such that

{i_{1}, \dots, i_{p}} / \in Δ

. The ring

A = k [Δ]

is a homogeneous reduced ring with the unique homogeneous maximal ideal

m = (X_{1}, \dots, X_{n}) k [Δ]

and the Krull dimension

d = dim Δ + 1

. Let

e (A)

denote the multiplicity

e_{0} (m A_{m}, A_{m})

A

, which is equal to the number of facets (i.e., maximal faces)

F

Δ

with

dim F = d - 1

. Also, we frequently call it the multiplicity of

Δ

Note that

Δ

is called pure if all facets of

Δ

have the same dimension. See [1, 9] for more details.

Take a graded minimal free resolution of a homogeneous

k

-algebra

A = S / I

over

S :

0 \to \oplus_{j \in Z} S (- j)^{β_{p, j} (A)} \overset{φ_{p}}{⟶} \dots \overset{φ_{2}}{⟶} \oplus_{j \in Z} S (- j)^{β_{1, j} (A)} \overset{φ_{1}}{⟶} S \to A \to 0 .

Then the initial degree

i n d e g A

(resp. the relation type

r t (A)

) of

A

is defined by

i n d e g A = min {j \in Z : β_{1, j} (A) \neq 0}

(resp.

r t (A) = max {j \in Z : β_{1, j} (A) \neq 0}

Also,

r e g A = max {j - i \in Z : β_{i, j} (A) \neq 0}

is called the Castelnuovo–Mumford regularity of

A

. It is easy to see that

r e g A \geq i n d e g A - 1

, and

A

has linear resolution if equality holds.

The main purpose of this paper is to prove the following theorems

:

Theorem 2.1 . Let

A = k [Δ]

be a Stanley–Reisner ring of Krull dimension

d \geq 2

Put

c o d i m A = c

. If

e (A) \geq (\begin{matrix} n \\ c \end{matrix}) - c

, then

A

is Cohen–Macaulay. Theorem 3.1 . Let

A = k [Δ]

be a Stanley–Reisner ring of Krull dimension

d \geq 2

Put

c o d i m A = c

. Suppose that

Δ

is pure

(

i.e.,

A

is equidimensional

)

. If

e (A) \geq (\begin{matrix} n \\ c \end{matrix}) - 2 c + 1

, then

A

is Cohen–Macaulay. It is easy to prove the above theorems in the case of

d = 2

. When

d = 2

A

is Cohen–Macaulay if and only if

Δ

is connected. In fact, a disconnected graph has at most

(n - 12) (= (n 2) - (n - 2) - 1)

edges. This shows that Theorem 2.1 is true in this case. Similarly, a disconnected graph without an isolated point has at most

(n - 22) + 1 (= (n 2) - 2 (n - 2))

edges. Indeed, such a graph is contained in a disjoint union of an

(n - i)

-complete graph and an

i

-complete graph for some

2 \leq i \leq n - 2

When

i = 2

, the number of edges of the above union is just

(n - 22) + 1

. Thus we also get Theorem 3.1 in this case.

The case

i n d e g A = d

and

c \geq 2

is essential in the above two theorems. In order to prove Theorems 2.1 and 3.1 in this case, we consider their Alexander dual versions

:

Theorem 2.7 . Let

A = k [Δ]

be a Stanley–Reisner ring of Krull dimension

d \geq 2

Suppose that

i n d e g A = d

. If

e (A) \leq d

, then

A

has

d

-linear resolution. In particular,

r t (A) = d

Theorem 3.3 . Let

A = k [Δ]

be a Stanley–Reisner ring of Krull dimension

d \geq 2

Suppose that

i n d e g A = r t (A) = d

. If

e (A) \leq 2 d - 1

, then

A

has

d

-linear resolution.

In particular,

a (A) < 0

. For a Stanley–Reisner ring

A

with

i n d e g A = dim A = d

, it has

d

-linear resolution if and only if

a (R) < 0

. Thus the assertion of Theorem 3.3 could be seen as an analogy of the following: Let

R

be a homogeneous integral domain over an algebraically closed field of characteristic

0

. If

e (R) \leq 2 dim R - 1

and

c o d i m R \geq 2

, then

a (R) < 0

In the last section, we will provide several examples related to the above results.

2 Complexes $Δ$ with $e (k [Δ]) \geq (n c) - c$

In this section, we use the following notation. Let

Δ

be a simplicial complex on

V = [n]

, and let

A = k [Δ] = S / I_{Δ}

be the Stanley–Reisner ring of

Δ

. Put

d = dim A

, and

c = c o d i m A = n - d

. Note that

([n] d)

denotes the family of all

d

-subsets of

[n]

The main purpose of this section is to prove the following theorem.

Theorem 2.1. Let

A = k [Δ]

be a Stanley–Reisner ring of Krull dimension

d \geq 2

. If

e (A) \geq (\begin{matrix} n \\ c \end{matrix}) - c

, then

A

is Cohen–Macaulay.

Let us begin the proof of this theorem with the following lemmas.

Lemma 2.2. If

e (A) \geq (\begin{matrix} n \\ c \end{matrix}) - c

, then

i n d e g A \geq d

Proof. Suppose that $i n d e g A < d$ . Take a squarefree monomial $M \in I_{Δ}$ with $deg M = d - 1$ . Then $(n - d + 1)$ distinct squarefree monomials appear among $X_{1} M, \dots, X_{n} M$ ; say $M_{1}, \dots, M_{c + 1}$ . Let $F_{i} \in ([n] d)$ corresponding to $M_{i}$ , respectively.
Then since no $F_{i}$ is contained in $Δ$ we have $e (A) \leq (\begin{matrix} n \\ d \end{matrix}) - (c + 1) .$ This contradicts the assumption. □

Lemma 2.3. Under the above notation, the following conditions are equivalent

:

$(1)$ $i n d e g A = d + 1$ .
$(2)$ $e (A) = (\begin{matrix} n \\ d \end{matrix})$ .
$(3)$ $I_{Δ} = (X_{i_{1}} \dots X_{i_{d + 1}} : 1 \leq i_{1} < \dots < i_{d + 1} \leq n)$ .
$(4)$ $A$ has $(d + 1)$ -linear resolution.

When this is the case,

A

is Cohen–Macaulay with

r t (A) = d + 1

Proof. See, e.g., [11,Proposition1.2] . □

Therefore we may assume that

i n d e g A = d

to prove Theorem 2.1 .

Lemma 2.4. Suppose

n = d + 1

. If

e (A) \geq d

, then

A

is a hypersurface.

Proof. Suppose that $A$ is not a hypersurface. Then we can write $I_{Δ} = X_{i_{1}} \dots X_{i_{p}} J$ for some monomial ideal $J (\neq R)$ with $h e i g h t J \geq 2$ since $h e i g h t I_{Δ} = 1$ . In particular, $A$ is not Cohen–Macaulay. Thus $i n d e g A \leq d$ by Lemma 2.3 . Then $e (A) = p \leq d - 1$ . This contradicts the assumption. □

Thus we may also assume that

c = c o d i m A \geq 2

. then let

Δ^{*}

be the Alexander dual of

Δ

Δ^{*} = {F \in 2^{V} : V \ F / \in Δ} .

Then

Δ^{*}

is a simplicial complex on the same vertex set

V

Δ

for which the following properties are satisfied

:

Proposition 2.5. Under the above notation, we have

$(1)$ $i n d e g k [Δ^{*}] + dim k [Δ] = n$ .
$(2)$ $r t (k [Δ^{*}]) = b i g h t I_{Δ}$ , where $b i g h t I = max {h e i g h t p : p is a minimal prime divisor of I} .$ In particular, $Δ$ is pure if and only if $r t (k [Δ^{*}]) = i n d e g k [Δ^{*}]$ .
$(3)$ $β_{0, q^{*}} (I_{Δ^{*}}) = e (k [Δ])$ , where $q^{*} = i n d e g k [Δ^{*}]$ .
$(4)$ $(Δ^{*})^{*} = Δ$ .

Also, the following theorem is fundamental. See [3] for more details.

Theorem 2.6 (Eagon–Reiner [3] ).

k [Δ]

is Cohen–Macaulay if and only if

k [Δ^{*}]

has linear resolution.

We want to reduce Theorem 2.1 to its Alexander dual version. Let

Δ^{*}

be the Alexander dual of

Δ

. Then

i n d e g k [Δ^{*}] = n - dim k [Δ] = c

and

dim k [Δ^{*}] = n - i n d e g k [Δ] = n - d = c

. Also, since

i n d e g k [Δ^{*}] = dim k [Δ] = c

, we have

e (k [Δ^{*}]) = (\begin{matrix} n \\ c \end{matrix}) - β_{0, c} (I_{Δ^{*}}) = (\begin{matrix} n \\ c \end{matrix}) - e (A) \leq c

Therefore, it is enough to prove the following theorem.

Theorem 2.7 (Alexander dual version of Theorem 2.1 ). Let

A = k [Δ]

be a Stanley–Reisner ring of Krull dimension

d \geq 2

. Suppose that

i n d e g A = d

. If

e (A) \leq d

, then

A

has

d

-linear resolution. In particular,

r t (A) = d

Proof. (1) Put $a (A) = sup {p \in Z : [H_{m}^{d} (A)]_{p} \neq 0}$ , the $a$ -invariant of $A$ . From the assumption we obtain that $a (A) + d \leq e (A) - 1 \leq d - 1,$ where the first inequality follows from e.g. [6,Lemma3.1] . Hence $a (A) < 0$ .
On the other hand, we have that $[H_{m}^{i} (A)]_{j} = 0$ for all $i$ and $j \geq 1$ since $A$ is a Stanley–Reisner ring. Then $r e g A = inf {p \in Z : [H_{m}^{i} (A)]_{j} = 0 for all i + j > p} \leq d - 1 = i n d e g A - 1 .$ This means that $A$ has $d$ -linear resolution, as required. □

Now let us discuss a generalization of Theorem 2.7 . Let

A = S / I

be an arbitrary homogeneous reduced

k

-algebra over a field

k

of characteristic

p > 0

. The ring

A

is called

F

-pure if the Frobenius map

F : A \to A (a \mapsto a^{p})

is pure. It is known that a Stanley–Reisner ring is

F

-pure, and that if

A

F

-pure then

[H_{m}^{i} (A)]_{j} = 0

for all

j \geq 1

. Thus the proof of Theorem 2.7 involves that of the following proposition.

Proposition 2.8. Let

A = S / I

be a homogeneous F-pure

k

-algebra. Put

dim A = i n d e g A = d \geq 2

. If

e (A) \leq d

, then

A

has

d

-linear resolution. In particular,

r t (A) = d

and

a (A) < 0

3 Complexes $Δ$ with $e (k [Δ]) \geq (n c) - 2 c + 1$

We use the same notation as in the previous section. For a face

G

Δ

and

v \in V

, we put

\begin{matrix} Δ_{V \ {v}} & = & {F \in Δ : v / \in F}, \end{matrix}

\begin{matrix} s t a r_{Δ} G & = & {F \in Δ : F \cup G \in Δ}, \end{matrix}

\begin{matrix} l i n k_{Δ} G & = & {F \in Δ : F \cup G \in Δ, F \cap G = \emptyset} . \end{matrix}

The main purpose of this section is to prove the following theorem.

Theorem 3.1. Let

A = k [Δ]

be a Stanley–Reisner ring of Krull dimension

d \geq 2

. Put

c = c o d i m A

. Suppose that

Δ

is pure. If

e (A) \geq (\begin{matrix} n \\ c \end{matrix}) - 2 c + 1

, then

A

is Cohen–Macaulay.

Now suppose that

c = 1

(resp.

i n d e g A \geq d + 1

). Then the assertion follows from Lemma 2.4 (resp. Lemma 2.3 ). Thus we may assume that

c \geq 2

and

q = i n d e g A \leq d

. The following lemma corresponds to Lemma 2.2 .

Lemma 3.2. If

e (k [Δ]) \geq (n c) - 2 c + 1

, then

i n d e g k [Δ] \geq d - 1

, i.e.,

(1)

i n d e g k [Δ] = d

(2)

i n d e g k [Δ] = d - 1

Proof. Suppose that $i n d e g k [Δ] < d - 1$ . Take a squarefree monomial $M \in I_{Δ}$ with $deg M = d - 2$ . Then there are $(n - d + 22)$ squarefree monomials in degree $d$ in $I_{Δ}$ . Note $(n - d + 22) = (c + 22) \geq 2 c$ . Hence $e (k [Δ]) \leq (n c) - 2 c .$ This contradicts the assumption. □

First, we consider the Alexander dual version of Theorem 3.1 in the case of

i n d e g k [Δ] = d

. Namely, we will prove the following theorem.

Theorem 3.3 (Alexander dual version of Theorem 3.1 , Case

(1)

). Let

A = k [Δ]

be a Stanley–Reisner ring of Krull dimension

d \geq 2

. Suppose that

i n d e g A = r t (A) = d

. If

e (A) \leq 2 d - 1

, then

A

has

d

-linear resolution.

In particular,

a (A) < 0

The proof of the above theorem can be reduced to that of the following theorem, which is a key result in this paper.

Theorem 3.4. Let

A = k [Δ]

be a Stanley–Reisner ring of Krull dimension

d \geq 2

. Suppose that

r t (A) \leq d

. If

e (A) \leq 2 d - 1

, then

r e g A \leq d - 1

, equivalently,

{\tilde{H}}_{d - 1} (Δ) = 0

Proof. Put $e = e (A)$ . Let $Δ^{'}$ be the subcomplex that is spanned by all facets of dimension $d - 1$ . Replacing $Δ$ with $Δ^{'}$ , we may assume that $Δ$ is pure.
We use induction on $d = dim A \geq 2$ . First suppose $d = 2$ . The assumption shows that $Δ$ does not contain the boundary complex of a triangle. Hence ${\tilde{H}}_{1} (Δ) = 0$ since $e (A) \leq 3$ .
Next suppose that $d \geq 3$ , and that the assertion holds for any complex the dimension of which is less than $d - 1$ . Assume that ${\tilde{H}}_{d - 1} (Δ) \neq 0$ . Take one $Δ$ whose multiplicity is minimal among the multiplicities of those complexes. Then $Δ$ does not contain any free face (see [7] ). That is, every face that is not a facet is contained in at least two facets. Indeed, suppose that $Δ$ contains a free face (say, $G$ ) and put $Δ^{'} = Δ \ {F \in Δ : F \supseteq G}$ . Then since $G$ is a free face of $Δ$ , $Δ^{'}$ is homotopy equivalent to $Δ$ and $e (k [Δ^{'}]) = e (k [Δ]) - 1$ . In particular, ${\tilde{H}}_{d - 1} (Δ^{'}) \sim = {\tilde{H}}_{d - 1} (Δ) \neq 0$ . This contradicts the minimality of $e (k [Δ])$ .
First consider the case of $r t (A) = d$ . Take a generator $X_{i_{1}} \dots X_{i_{d}}$ of $I_{Δ}$ . For every $j = 1, \dots, d$ , each $G_{j} = {i_{1}, \dots, \hat{i_{j}}, \dots, i_{d}}$ is contained in at least two facets as mentioned above. Then $e (A) \geq 2 d$ since those facets are different from each other. This is a contradiction.
Next we consider the case of $r t (A) < d$ . Take a Mayer–Vietoris sequence with respect to $Δ = Δ_{V \ {n}} \cup s t a r_{Δ} {n}$ as follows $:$ ${\tilde{H}}_{d - 1} (Δ_{V \ {n}}) \oplus {\tilde{H}}_{d - 1} (s t a r_{Δ} {n}) ⟶ {\tilde{H}}_{d - 1} (Δ) ⟶ {\tilde{H}}_{d - 2} (l i n k_{Δ} {n}) .$ The minimality of $e (k [Δ_{V \ {n}}])$ yields that ${\tilde{H}}_{d - 1} (Δ_{V \ {n}}) = 0$ since $e (k [Δ_{V \ {n}}]) < e (k [Δ])$ . On the other hand, it is known that ${\tilde{H}}_{i} (s t a r_{Δ} {n}) = 0$ for all $i$ . Hence ${\tilde{H}}_{d - 1} (Δ) ↪ {\tilde{H}}_{d - 2} (l i n k_{Δ} {n})$ . In particular, ${\tilde{H}}_{d - 2} (l i n k_{Δ} {n}) \neq 0$ .
Set $Δ^{'} = l i n k_{Δ} {n}$ . Then $Δ^{'}$ is a complex on $V \ {n}$ such that $dim k [Δ^{'}] = d - 1$ and $r t (k [Δ^{'}]) \leq r t (k [Δ]) \leq d - 1$ . In order to apply the induction hypothesis to $Δ^{'}$ , we want to see that $e (k [Δ^{'}]) \leq 2 d - 3$ . In order to do that, we consider $e (k [Δ_{V \ {n}}])$ .
As $Δ \neq s t a r_{Δ} {n}$ , one can take $F = {i_{1}, \dots, i_{p}, n} / \in Δ$ for some $p \leq d - 2$ such that $X_{i_{1}} \dots X_{i_{p}} X_{n}$ is a generator of $I_{Δ}$ . Then $G : = {i_{1}, \dots, i_{p}} \in Δ$ , but it is not a facet of $Δ$ . Thus it is contained in at least two facets of $Δ$ , each of which does not contain $n$ . Hence $e (k [Δ_{V \ {n}}]) \geq 2$ . Thus we get $e (k [Δ^{'}]) = e (k [s t a r_{Δ} {n}]) = e (k [Δ]) - e (k [Δ_{V \ {n}}]) \leq 2 d - 3 .$ By induction hypothesis, we have ${\tilde{H}}_{d - 2} (l i n k_{Δ} {n}) = 0$ . This is a contradiction.
□

Next, we consider the Alexander dual version of Theorem 3.1 in the case of

i n d e g k [Δ] = d - 1

. Namely, we must prove the following proposition.

Proposition 3.5 (Alexander dual version of Theorem 3.1 , Case

(2)

). Let

A = k [Δ]

be a Stanley–Reisner ring of Krull dimension

d \geq 2

Suppose that

i n d e g A = r t (A) = d - 1

. If

μ (I_{Δ}) \geq (\begin{matrix} n \\ d - 1 \end{matrix}) - 2 d + 3

, then

A

has

(d - 1)

-linear resolution with

e (A) = 1

Proof. First we show that $e (A) = 1$ . Now suppose that $e (A) \geq 2$ . Then there exist at least two facets $F_{1}$ and $F_{2}$ with $# (F_{1}) = # F (F_{2}) = d$ . This implies that $f_{d - 2} (Δ) \geq 2 d - 1$ . However, by the assumption, we have $f_{d - 2} (Δ) = (n d - 1) - β_{0, d - 1} (I_{Δ}) = (n d - 1) - μ (I_{Δ}) \leq 2 d - 3 .$ This is a contradiction. Hence we get $e (A) = 1$ .
In order to prove that $A$ has $(d - 1)$ -linear resolution, it is enough to show that $β_{i, j} (A) = 0$ for all $i \geq c$ and $j \geq i + d - 1$ by [8,Theorem5.2] . Also, it suffices to show that ${\tilde{H}}_{d - 1} (Δ) = {\tilde{H}}_{d - 2} (Δ) = {\tilde{H}}_{d - 2} (Δ_{W}) = 0$ for all subsets $W \subset V$ with $# (W) = n - 1$ by virtue of Hochster's formula on the Betti numbers: $β_{i, j} (A) = \sum_{W \subseteq V # (W) = j} {dim}_{k} {\tilde{H}}_{j - i - 1} (Δ_{W}; k) .$

Claim 1. ${\tilde{H}}_{d - 1} (Δ) = {\tilde{H}}_{d - 2} (Δ) = 0$ .

Since $r t (A) \leq d - 1 \leq d$ and $e (A) = 1 \leq 2 d - 1$ , we have ${\tilde{H}}_{d - 1} (Δ) = 0$ by Theorem 3.4 . Now let $F = {1, 2, \dots, d}$ be the unique facet with $# (F) = d$ .
Consider a simplicial subcomplex $Δ^{'} : = Δ \ {F, G}$ where $G = {1, 2, \dots, d - 1}$ . Then $dim k [Δ^{'}] = d - 1$ and $e (k [Δ^{'}]) \leq 2 d - 4 \leq 2 (d - 1) - 1$ . Also, since $r t (k [Δ^{'}]) \leq r t (k [Δ]) \leq d - 1$ , applying Theorem 3.4 to $Δ^{'}$ , we obtain that ${\tilde{H}}_{d - 2} (Δ) \sim = {\tilde{H}}_{d - 2} (Δ^{'}) = 0$ , as required.

Claim 2. ${\tilde{H}}_{d - 2} (Δ_{W}) = 0$ for all subsets $W \subset V$ with $# (W) = n - 1$ .

Let $W$ be a subset of $V$ such that $# (W) = n - 1$ . Put ${a} = V \ W$ . If $a$ is not contained in $F$ , then ${\tilde{H}}_{d - 2} (Δ_{W}) = 0$ by the similar argument as in the proof of the previous claim. So we may assume that $a \in F$ . Then $dim k [Δ_{W}] = d - 1$ and $e (k [Δ_{W}]) \leq (d - 3) + 1 = d - 2 \leq 2 (d - 1) - 1$ . Also, since $r t (k [Δ_{W}]) \leq d - 1$ , we have ${\tilde{H}}_{d - 2} (Δ_{W}) = 0$ by Theorem 3.4 again.
Hence $k [Δ]$ has $(d - 1)$ -linear resolution, as required. □

Example 3.6. Let

ρ

d

be an integers with

0 \leq ρ \leq d - 3

. Let

Δ

be a simplicial complex on

V = [n]

spanned by

F = {1, 2, \dots, d}

, any distinct

ρ

elements from

([n] d - 1) \ ([d] d - 1)

and all elements of

([n] d - 2)

. Then

dim k [Δ] = d

i n d e g k [Δ] = r t (k [Δ]) = d - 1

. Also, we have

μ (I_{Δ}) = β_{0, d - 1} (I_{Δ}) = (n d - 1) - ρ - d \geq (n d - 1) - 2 d + 3 .

Hence

Δ

satisfies the assumption of the above proposition.

On the other hand, we have no results for F-pure

k

-algebras corresponding to Theorem 3.3 . But we remark the following.

Remark 3.7. As mentioned in the introduction, if

A

is a homogeneous integral domain over an algebraically closed field of

c h a r k = 0

with

c o d i m A \geq 2

and

e (A) \leq 2 d - 1

then one has

a (A) < 0

. In fact, it is known that an inequality

a (A) + d \leq ⌈ \frac{e (A) - 1}{c o d i m A} ⌉

holds; see e.g., the remark after Theorem

3.2

in [6] . Moreover, Professor Chikashi Miyazaki told us that this inequality is also true in positive characteristic.

Question 3.8. Let

A = k [A_{1}]

be a homogeneous F-pure, equidimensional

k

-algebra. Put

dim A = i n d e g A = d \geq 2

. If

e (A) \leq 2 d - 1

, then does

a (A) < 0

hold?

4 Buchsbaumness

A Stanley–Reisner ring

A = k [Δ]

is Buchsbaum if and only if

Δ

is pure and

k [l i n k_{Δ} {i}]

is Cohen–Macaulay for every

i \in [n]

. As an application of Theorem 3.1 , we can provide sufficient conditions for

k [Δ]

to be Buchsbaum.

Proposition 4.1. Let

A = k [Δ]

be a Stanley–Reisner ring of Krull dimension

d \geq 3

. Suppose that

Δ

is pure,

i n d e g A = d

and

e (A) \geq (n c) - 2 c

Then

$(1)$ $e (k [l i n k_{Δ} {i}]) \geq (n - 1 c) - 2 c$ for all $i$ .
$(2)$ If $h e i g h t [I_{Δ}]_{d} S \geq 2$ , then $A$ is Buchsbaum.
$(3)$ If $r t (A) = d$ , then $A$ is Buchsbaum.

Proof. We may assume that $c \geq 2$ , $e (A) = (n c) - 2 c$ , and that $Δ \neq s t a r_{Δ} {i}$ for every $i \in [n]$ . Put $Γ_{i} = l i n k_{Δ} {i}$ for each $i \in [n]$ .
(1) We first show the following claim.
Claim: $e (A) \leq (n d) - {(n - 1 d - 1) - e (k [Γ_{i}])}$ for every $i \in [n]$ . Also, equality holds if and only if $i \in F$ holds for all $F \in ([n] d) \ Δ$ .
Put $W_{i} = {F \in ([n] d) : i \in F / \in s t a r_{Δ} {i}}$ . Then $# (W_{i}) \leq # (\cup_{i = 1}^{n} W_{i})$ implies that $(n - 1 d - 1) - e (k [Γ_{j}]) \leq (n d) - e (A),$ as required. Also, equality holds if and only if $W_{i} = \cup_{i = 1}^{n} W_{i}$ , that is, $i \in F$ holds for all $F \in ([n] d) \ Δ$ .
Now suppose that $e (k [Γ_{i}]) \leq (n - 1 d - 1) - 2 c - 1$ for some $i \in [n]$ . Then the claim implies that $e (A) \leq (n d) - 2 c - 1$ , which contradicts the assumption. Thus we get (1).
(2) Suppose that $h e i g h t [I_{Δ}]_{d} S \geq 2$ . Then there is no element $i \in [n]$ for which $i \in F$ holds for all $F \in ([n] d) \ Δ$ . Thus the claim yields that $(n d) - 2 c = e (A) \leq (n d) - [(n - 1 d - 1) - e (k [Γ_{i}])] - 1,$ that is, $e (k [Γ_{i}]) \geq (n - 1 d - 1) - 2 c + 1$ for every $i \in [n]$ . Also, we note that $Γ_{i}$ is pure and $i n d e g k [Γ_{i}] = dim k [Γ_{i}] = d - 1$ . Applying Theorem 3.1 to $k [Γ_{i}]$ , we obtain that $k [Γ_{i}]$ is Cohen–Macaulay. Therefore $A$ is Buchsbaum since $Δ$ is pure.
(3) Now suppose that $A$ is not Buchsbaum. Then since $h e i g h t [I_{Δ}]_{d} S = 1$ , one can take $i \in [n]$ for which $i \in F$ holds for all $F \in ([n] d) \ Δ$ . We may assume $i = n$ .
Then ${1, \dots, \hat{i}, \dots, d + 1} \in Δ$ for all $i \in [d + 1]$ because $n - 1 \geq d + 1$ . This means that $X_{1} \dots X_{d + 1}$ is a generator of $I_{Δ}$ ; thus $r t (A) = d + 1$ . □

5 Examples

Throughout this section, let

c

d

be given integers with

c, d \geq 2

. Set

n = c + d

Example 5.1. Put

F_{i, j} = {1, 2, \dots, \hat{i}, \dots, d, j}

for each

i = 1, \dots, d; j = d + 1, \dots, n

. For a given integers

e

with

1 \leq e \leq c d

, we choose

e

faces (say,

F_{1}, \dots, F_{e}

) from

{F_{i, j} : 1 \leq i \leq d, d + 1 \leq j \leq n}

, which is a simplicial join of

2^{[d]} \ {[d]}

and

c

points.

Let

Δ

be a simplicial complex spanned by

F_{1}, \dots, F_{e}

and all elements of

([n] d - 1)

. Then

k [Δ]

is a

d

-dimensional Stanley–Reisner ring with

i n d e g k [Δ] = r t (k [Δ]) = d

and

e (k [Δ]) = e

In particular, when

e \leq 2 d - 1

k [Δ]

has

d

-linear resolution by Theorem 3.3 . Thus Also, the Alexander dual complexes of them provide examples satisfying hypothesis of Theorem 3.1 .

The following example shows that the assumption “

e (A) \leq 2 d - 1

” is optimal in Theorem 3.3 .

Example 5.2. There exists a complex

Δ

V = [n]

(

n = d + 2

) for which

k [Δ]

does not have

d

-linear resolution with

dim k [Δ] = i n d e g k [Δ] = r t (k [Δ]) = d

and

e (k [Δ]) = 2 d

In fact, put

n = d + 2

. Let

Δ_{0}

be a complex on

V = [n]

such that

k [Δ_{0}]

is a complete intersection defined by

(X_{1} \dots X_{d}, X_{d + 1} X_{d + 2})

. Also, let

Δ

be a complex on

V

that is spanned by all facets of

Δ_{0}

and all elements of

([n] d - 1)

:

I_{Δ} = (X_{1} \dots X_{d}) S + (X_{i_{1}} \dots X_{i_{d - 2}} X_{d + 1} X_{d + 2} : 1 \leq i_{1} < \dots < i_{d - 2} \leq d) S .

Then

{\tilde{H}}_{d - 1} (k [Δ]) \sim = {\tilde{H}}_{d - 1} (k [Δ_{0}]) \neq 0

since

a (k [Δ_{0}]) = 0

. Hence

k [Δ]

does not have linear resolution.

Remark 5.3. The above example is obtained by considering the case

c = 2, e = 2 d

in Example 5.1

The next example shows that the assumption “

r t (A) = d

” is not superfluous in Theorem 3.3 .

Example 5.4. Suppose that

d + 1 \leq e \leq (n d) - 1

. There exists a simplicial complex

Δ

V = [n]

such that

dim k [Δ] = i n d e g k [Δ] = d

r t (k [Δ]) = d + 1

and

e (k [Δ]) = e

. In particular,

k [Δ]

does not have

d

-linear resolution.

In fact, put

ℱ = ([n] d) \ ([d + 1] d)

. Let

Δ_{0}

be a simplicial complex on

V

such that

I_{Δ_{0}} = (X_{1} \dots X_{d} X_{d + 1}) S + (X_{i_{1}} \dots X_{i_{d}} : {i_{1}, \dots, i_{d}} \in ℱ) S .

Then

dim k [Δ_{0}] = i n d e g k [Δ_{0}] = d

r t (k [Δ_{0}]) = d + 1

, and

e (k [Δ_{0}]) = d + 1

For a given integer

e

which satisfies above condition, one obtains the required simplicial complex by adding any

(e - d - 1)

distinct

d

-subsets of

2^{[n]}

that is not contained in

([d + 1] d)

Δ_{0}

Remark 5.5. Now let

Δ

be a simplicial complex on

V = [n]

. Set

A = k [Δ]

. Suppose that

dim A = i n d e g A = d \geq 2

. Then one can easily see that

d \leq r t (A) \leq d + 1

;

r t (A) = d

(

resp.

d + 1

)

1 \leq e (A) \leq d

(

resp.

e (A) = (n d)

)

. So we put

f (n, d) = min {m \in Z : \begin{matrix} r t k [Δ] = d + 1 for all (d - 1) -dimensional \\ complexes Δ on V with i n d e g k [Δ] = d \\ and e (k [Δ]) \geq m \end{matrix}}

Then

f (n, d) \geq c d + 1

by Example 5.1

. From the definition of

f (n, d)

, one can easily see that there exists a simplicial complex

Δ

which satisfies

r t (k [Δ]) = d

and

e (k [Δ]) = e

for each

e

with

d + 1 \leq e \leq f (n, d) - 1

. On the other hand, by virtue of Example 5.4

, one can also find a simplicial complex

Δ

which satisfies

r t (k [Δ]) = d + 1

and

e (k [Δ]) = e

for each

e

with

d + 1 \leq e \leq (n d) - 1

It seems to be difficult to determine

f (n, d)

in general. Let

T (n, p, k)

be the so-called Turan number. Then we have

f (n, d) = (n d) - T (n, d + 1, d) .

In particular, we get

\begin{matrix} f (n, 2) = {\begin{matrix} \frac{n^{2}}{4} + 1, & if n is even; \\ \frac{n^{2} - 1}{4} + 1, & otherwise \end{matrix} \end{matrix}

(5.1)

by Turan's theorem

(

e.g., [2,Theorem7.1.1]

)

However, no formula is known for

T (n, 4, 3)

; see [5,pp.1320] .

In the rest of this section, we show that the purity of

Δ

is very strong condition in Theorem 3.3 .

Proposition 5.6. Then the following conditions are equivalent

:

$(1)$ There exists a $d$ -dimensional Stanley–Reisner ring $k [Δ]$ such that $Δ$ is pure, $i n d e g k [Δ] = d$ and $e (k [Δ]) = e \leq 2 d - 1$ .
$(2)$ $n = d + 2$ , $d \leq 5$ and $(d, e)$ is one of the following pairs $:$ $(2, 2), (2, 3), (3, 4), (3, 5), (4, 6), (4, 7), (5, 9) .$

To prove the proposition, we need the following lemma.

Lemma 5.7. Let

A = k [Δ]

be a

d

-dimensional Stanley–Reisner ring which is not a hypersurface. Suppose that

Δ

is pure and

i n d e g A = d \geq 3

Then there exists a vertex

i \in [n]

such that

e (k [Δ_{V \ {i}}]) \geq 2

Proof. Note that $n \geq d + 2$ by the assumption. Put $e = e (A)$ . Suppose that $e (k [Δ_{V \ {i}}]) = 1$ for all $i$ . Then since there exist $(e - 1)$ facets containing $i$ for each $i \in [n]$ , we have $(d + 2) (e - 1) \leq n (e - 1) \leq d e;$ hence $e \leq \frac{d + 2}{2}$ .
On the other hand, by counting the number of subfacets (i.e., the maximal faces among all faces except facets) of $Δ$ we get $d e \geq (n d - 1)$ since $i n d e g A = d$ and $Δ$ is pure. It follows from these inequalities that $\frac{d (d + 2)}{2} \geq d e \geq (n d - 1) \geq (d + 2 d - 1) = (d + 23) .$ Hence $d \leq 2$ . This is a contradiction. □

Proof of Proposition 5.6 . We first show $(1) ⟹ (2)$ . Let $A = k [Δ]$ be a $d$ -dimensional Stanley–Reisner ring for which $Δ$ is pure, $i n d e g A = d$ , and $e = e (A) \leq 2 d - 1$ . We may assume that $d \geq 3$ . Since $Δ$ is pure, any subfacet is contained in some $d$ -subset of $Δ$ . By counting the number of subfacets that contain $n$ , we obtain that $(n - 1 d - 2) \leq (e - e (k [Δ_{V \ {n}}]) (d - 1) \leq (e - 2) (d - 1),$ where the last inequality follows from Lemma 5.7 .
Now let us see that $n = d + 2$ . Suppose that $n \geq d + 3$ . Then we get $(d + 24) \leq (n - 1 d - 2) \leq (e - 2) (d - 1) \leq (2 d - 3) (d - 1)$ by the assumption. This implies that $d \leq 4$ .
First we consider the case of $d = 4$ . Then $n = d + 3 = 7$ , $e = 2 d - 1 = 7$ . Let ${F_{1}, \dots, F_{7}}$ be the set of facets of $Δ$ . Since $e (k [Δ_{V \ {7}}]) = 2$ , we may assume that $7 \in F$ if and only if $1 \leq i \leq 5$ . Note that $F_{i}$ contains only one subfacet that does not contain $7$ for each $1 \leq i \leq 5$ . On the other hand, one can find at most $4 \times 2$ subfacets as faces of $F_{6}$ or $F_{7}$ . Therefore the total number of subfacets that do not contain $7$ is at most $13$ . However the number of all subfacets which do not contain $7$ is $(7 - 14 - 1) = 20$ since $i n d e g A = 4$ . This is a contradiction.
By the similar observation as in the case of $d = 4$ , one can prove that the case of $d = 3$ does not occur. Therefore we conclude that $n = d + 2$ .
Under the assumption that $n = d + 2$ , let us determine $(d, e)$ . Let $Δ^{*}$ be the Alexander dual of $Δ$ and put $R = k [Δ^{*}]$ . Then $R$ is a two-dimensional Stanley–Reisner ring with $i n d e g R = 2$ . Also, $r t (R) = i n d e g R = 2$ since $Δ$ is pure. Thus by virtue of Turan's theorem (see Eq. 5.1 ), we have $(d + 22) - e = e (R) \leq f (d + 2, 2) - 1 = ⌊ \frac{(d + 2)^{2}}{4} ⌋,$ where $⌊ a ⌋$ denotes the maximum integer that does not exceed $a$ . Namely, we have $2 d - 1 \geq e \geq ⌊ \frac{(d + 1)^{2}}{4} ⌋ .$ It immediately follows from here that $(d, e)$ is one of the pairs listed above.
Conversely, in order to prove $(2) ⟹ (1)$ , it is enough to find $(n, e^{'})$ -graphs (i.e., $1$ -dimensional simplicial complexes $Γ$ on $[n]$ with $e^{'}$ edges) which does not contain any triangle for each $(n, e^{'}) = (4, 4)$ , $(4, 3)$ , $(5, 6)$ , $(5, 5)$ , $(6, 9)$ , $(6, 8)$ , $(7, 12)$ . Those complexes will be given in the following example. □

Example 5.8. There exists a

1

-dimensional simplicial connected complex

Γ

[n]

with with

e (k [Γ]) = e^{'}

and

r t (k [Γ]) = 2

for each

(n, e^{'}) = (4, 4)

(4, 3)

(5, 6)

(5, 5)

(6, 9)

(6, 8)

(7, 12)

. Put

\begin{matrix} S_{4, 4} & = & {[12], [14], [23], [34]}, \end{matrix}

\begin{matrix} S_{4, 3} & = & {[12], [23], [34]}, \end{matrix}

\begin{matrix} S_{5, 6} & = & {[12], [14], [23], [25], [34], [45]}, \end{matrix}

\begin{matrix} S_{5, 5} & = & {[12], [14], [23], [34], [45]}, \end{matrix}

\begin{matrix} S_{6, 9} & = & {[14], [15], [16], [24], [25], [26], [34], [35], [36]}, \end{matrix}

\begin{matrix} S_{6, 8} & = & {[12], [14], [23], [25], [34], [36], [45], [56]}, \end{matrix}

\begin{matrix} S_{7, 12} & = & {[15], [16], [17], [25], [26], [27], [35], [36], [37], [45], [46], [47]]}, \end{matrix}

where

[i_{1} i_{2} \dots i_{p}]

means

{i_{1}, i_{2}, \dots, i_{p}}

Let

Γ_{n, e^{'}}

be a simplicial complex spanned by

S_{n, e^{'}}

. Then

k [Γ_{n, e^{'}}]

is a two-dimensional Cohen–Macaulay Stanley–Reisner ring with

e (k [Γ]) = e^{'}

and

r t (k [Γ]) = 2

. Note that when

e^{'} = f (n, 2) - 1

Γ_{n, e^{'}}

is the so-called Turan graph

T^{2} (n)

, that is, it is the unique complete bipartite graph on

[n]

whose two partition sets differ in size by at most

1

Let

Δ_{d, e}

be the Alexander dual complex of

Γ_{n, e^{'}}

where

d = n - 2

and

e = (d + 22) - e^{'}

. Namely,

Δ_{d, e}

is the complex spanned by

T_{d, e}

, respectively

:

\begin{matrix} T_{2, 2} & = & {[13], [24]}, \end{matrix}

\begin{matrix} T_{2, 3} & = & {[13], [23], [24]}, \end{matrix}

\begin{matrix} T_{3, 4} & = & {[124], [135], [234], [245]}, \end{matrix}

\begin{matrix} T_{3, 5} & = & {[124], [134], [135], [234], [245]}, \end{matrix}

\begin{matrix} T_{4, 6} & = & {[1234], [2345], [3456], [4561], [5612], [6123]}, \end{matrix}

\begin{matrix} T_{4, 7} & = & {[1235], [1246], [1345], [1356], [2345], [2346], [2456]}, \end{matrix}

\begin{matrix} T_{5, 9} & = & {[12345], [12346], [12347], [12567], [13567], [14567], [23567], \end{matrix}

\begin{matrix} [24567], [34567]} . \end{matrix}

Then

A = k [Δ_{d, e}]

is a

d

-dimensional equidimensional Stanley–Reisner ring with

i n d e g A = d

and

e (A) = e

for each

(d, e) = (2, 2)

(2, 3)

(3, 4)

(3, 5)

(4, 6)

(4, 7)

(5, 9)

Corollary 5.9. Let

A = k [Δ]

be a

d

-dimensional Buchsbaum Stanley–Reisner ring which is not a hypersurface. Suppose that

i n d e g A = d \geq 3

. Then

d = 3

and

Δ

is isomorphic to a simplicial complex spanned by

{[124], [134], [135], [235], [245]}

Proof. Since $A$ is Buchsbaum and $i n d e g A = d$ we have $e = e (A) \geq \frac{c + d}{d} (c + d - 2 d - 2)$ by [11,Proposition2.1] . Also, $n = d + 2$ by Proposition 5.6 since $Δ$ is pure. Thus $2 d - 1 \geq e \geq \frac{d + 2}{d} (d d - 2) = \frac{(d + 2) (d - 1)}{2} .$ This implies that $d \leq 3$ , and thus $d = 3$ and $e = 5$ . Then one can easily see that $Δ$ is isomorphic to the complex spanned by ${[124], [134], [135], [235], [245]}$ , which is the Alexander dual complex of a $1$ -dimensional connected complex spanned by ${[12], [23], [34], [45]}$ . □

References

W. Bruns and J. Herzog, Cohen–Macaulay Rings, Cambridge University Press, Cambridge, New York, Sydney, 1993.
R. Diestel, Graph theory. Second edition Graduate Texts in Mathematics, 173, Springer-Verlag, New York, 2000.
J. A. Eagon and V. Reiner, Resolutions of Stanley–Reisner rings and Alexander duality, J. Pure and Applied Algebra 130 (1998), 265–275.
D. Eisenbud and S. Goto, Linear free resolutions and minimal multiplicity, J. Algebra 88 (1984), 89–133.
P. Frankl, Extremal set systems, Handbook of combinatorics, Vol. 1, 2, 1293–1329, Elsevier, Amsterdam, 1995.
L. T. Hoa and C. Miyazaki, Bounds on Castelnuovo–Mumford regularity for generalized Cohen–Macaulay graded rings, Math. Ann. 301 (1995), 587–598.
J. F. P. Hudson, Piecewise Linear Topology, University of Chicago Lecture Notes prepared with the assistance of J. L. Shaneson and J. Lees, W.A.Benjamin, Inc., New York–Amsterdam 1969.
P. Schenzel, Applications of Koszul homology to numbers of generators and syzygies, J. Pure and Applied Algebra. 114 (1997), 287–303.
R. P. Stanley, Combinatorics and Commutative Algebra, Second Edition, Birkhaeuser, Boston/Basel/Stuttgart, 1996.
J. Stückrad and W. Vogel, Buchsbaum Rings and Applications, Springer–Verlag, Berlin, Heidelberg, New York, 1986.
N. Terai and K. Yoshida, Buchsbaum Stanley–Reisner rings with minimal multiplicity, (to appear in Proc. Amer. Math. Soc.).
W. V. Vasconcelos, Cohomological degrees of graded modules, in Six Lectures on Commutative Algebra (eds. J.Elias et.al), Birkhäuser, 1998.

Department of Mathematics, Faculty of Culture and Education, Saga University, Saga 840–8502, Japan E-mail address : terai@cc.saga-u.ac.jp Graduate School of Mathematics, Nagoya University, Nagoya 464–8602, Japan E-mail address : yoshida@math.nagoya-u.ac.jp

Mar.10, 2005

Naoki Terai

Ken-ichi Yoshida