.
We conclude that in the case
$d=3$
, for computing local cohomology of the modules
${R}^{D}$
it suffices to check the disconnected subsets
$\Pi \subset \Sigma $
. In general, this may not be so simple, as we will see in the following examples. The first example shows that even for topologically nontrivial
$\Pi $
, in general, the chamber
${C}_{\Pi}^{ss}$
must not be bounded.
Example 6.8:
Let
$\sigma $
be spanned over
${\mathbb{R}}_{\ge 0}$
by the primitive vectors
${n}_{1}=(1,0,0,0)$
,
${n}_{2}=(0,1,0,0)$
,
${n}_{3}=(1,1,1,0)$
,
${n}_{4}=(0,0,1,0)$
,
${n}_{5}=(0,0,0,1)$
. This is the cone from the previous example extended by one ray in fourdimensional direction. We choose
$D=k{D}_{2}$
for some
$k>0$
, and it is straightforward to see that, because
${n}_{5}$
is orthogonal to the other
${n}_{i}$
, the number of chambers is double the number of chambers of the previous example. Moreover, it is easy to see that this time there are no bounded chambers, but still, for
$\Pi :=\{{\rho}_{1},{\rho}_{4}\}$
, we have that
$\Pi \cap \Xi $
consists of two points, and hence
$dim{\stackrel{~}{H}}^{0}(\Pi \cap \Xi ,k)=1$
The next example shows that the contractibility of
$\Pi $
does not imply that the recession cone
${\sigma}_{\Pi}$
is strictly convex. Moreover, the example shows that the strict convexity of
${\sigma}_{\Pi}$
can not depend on the combinatorics of
$\Pi $
in a simple way, but also depends on the concrete embedding of the cone
$\sigma $
in
${N}_{\mathbb{R}}$
.
Example 6.9:
Consider the fourdimensional cone
$\sigma $
spanned by
${n}_{1}=(0,0,0,1)$
,
${n}_{2}=(1,0,0,1)$
,
${n}_{3}=(0,1,0,1)$
,
${n}_{4}=(0,0,1,1)$
,
${n}_{5}=(1,1,0,1)$
,
${n}_{6}=(1,0,1,1)$
,
${n}_{7}=(0,1,1,1)$
,
${n}_{8}=(1,1,1,1)$
, i.e.
$\sigma $
is spanned over the threedimensional unit cube shifted to the hyperplane
${x}_{4}=1$
. Set
$\Pi :=\{{\rho}_{1},{\rho}_{3},{\rho}_{4},{\rho}_{6}\}$
. Then
$\Pi \cap \Xi $
is contractible and we have
${\stackrel{~}{H}}^{i}(\Pi \cap \Xi ,k)=0$
for all
$i$
. But, we have
${n}_{3}+{n}_{6}={n}_{2}+{n}_{7}$
, so the cones
${\sigma}_{\Pi}^{1}$
and
${\sigma}_{\Pi}^{2}$
intersect, and the recession cone
${\stackrel{\u02c7}{\sigma}}_{\Pi}$
is of dimension smaller than
$4$
.
Now consider the cone
${\sigma}^{\prime}$
which is spanned by the same
${n}_{1},...,{n}_{8}$
as
$\sigma $
, except that
${n}_{4}$
and
${n}_{6}$
are replaced by
${n}_{4}^{\prime}=(0,1,1,1)$
and
${n}_{6}^{\prime}=(1,1,1,1)$
.
${\sigma}^{\prime}$
is combinatorially equivalent to
$\sigma $
, but by straightforward computation one finds that
$({\sigma}^{\prime}{)}^{1}$
and
$({\sigma}^{\prime}{)}^{2}$
do not intersect, and thus
${\stackrel{\u02c7}{\sigma}}_{\Pi}$
is a
$d$
dimensional recession cone of
${C}_{\Pi}$
and thus
${C}_{\Pi}^{ss}$
is nonempty for every module
${R}^{D}$
over
$k\left[{\sigma}_{M}^{\prime}\right]$
.
7 Maximal CohenMacaulay Modules of Rank One
By the results of section 5 , the problem of classifying maximal CohenMacaulay modules (MCMs) of rank one now has essentially become a problem of integer programming.
To see this more clearly, let us reformulate the results for this case. For
${R}^{D}$
being an MCM is equivalent to that all local cohomology modules
${H}_{\mathfrak{m}}^{i}{R}^{D}$
vanish for
$i<d$
, where
$\mathfrak{m}$
is the maximal homogeneous ideal of
$k\left[\sigma M\right]$
. This in particular is equivalent to the vanishing of the cohomology groups
${\stackrel{~}{H}}^{i2}({\Sigma}_{m}\cap \Xi ,k)$
for every
$i<d$
and every
$m\in M$
.
Now, as we have seen in example
6.7 , not every
$\Pi $
such that
${C}_{\Pi}^{ss}$
is nonempty in
${M}_{\mathbb{R}}$
equals
${\Sigma}_{m}$
for some
$m\in M$
, i.e. not every
${C}_{\Pi}^{ss}$
has nonempty intersection with
$M$
although it is realized in the complement of the arrangement
${\cup}_{\rho \in \Sigma}{H}_{\rho}$
. So let us state the MCMcondition for
${R}^{D}$
as a theorem:
Theorem 7.1:
${R}^{D}$
is an MCM if and only if for every
$\Pi \subset \Sigma $
one of the following two conditions holds:

(i)
${\stackrel{~}{H}}^{i2}(\Pi \cap \Xi ,k)=0$
for all
$i<d$
,

(ii)
${\stackrel{~}{H}}^{i2}(\Pi \cap \Xi ,k)\ne 0$
for some
$i<d$
and the chamber
${C}_{\Pi}^{ss}$
is either empty or has empty intersection with
$M$
.
We also state another, equivalent formulation:
Theorem 7.2:
${R}^{D}$
is an MCM if and only if for every
$\Pi \subset \Sigma $
one of the following two conditions holds:

(i)
${\stackrel{~}{H}}^{i2}(\Pi \cap \Xi ,k)=0$
for all
$i<d$
,

(ii)
${\stackrel{~}{H}}^{i2}(\Pi \cap \Xi ,k)\ne 0$
for some
$i<d$
and the system of inequalities
$$\begin{array}{cc}\langle m,n\left(\rho \right)\rangle & <{n}_{\rho}\text{for}\rho \in \Pi \end{array}$$  
$$\begin{array}{cc}\langle m,n\left(\rho \right)\rangle & \ge {n}_{\rho}\text{for}\rho \in \Sigma \backslash \Pi \end{array}$$  
$$\begin{array}{}\end{array}$$  
has no integral solution.
One can relate the classification problem for MCMs of rank one to the problem of understanding hyperplane arrangements in
${M}_{\mathbb{R}}$
induced by the hyperplanes
${H}_{\rho}$
, which are shifts of hyperplanes
${\rho}^{\perp}$
corresponding to some cone
$\sigma \in {N}_{\mathbb{R}}$
(such that, in particular, the hyperplanes
${H}_{\rho}$
are rational). If one fixes the combinatorial type of the hyperplane arrangement
${\cup}_{\rho \in \Sigma}{H}_{\rho}$
, say, its matroid of flats, then, in how many ways can this hyperplane arrangement be realized by shifting hyperplanes
${H}_{\rho}$
, while keeping the combinatorial type, such that the cells
${C}_{\Pi}^{ss}$
with some nonvanishing cohomology group do not intersect
$M$
?
8 Singularity Sets
In order to actually proof that some module
${R}^{D}$
is an MCM, one effectively has to check the inequalities of theorem 7.2 for nearly all possible sets
$\Pi \subset \Sigma $
which in general is a quite expensive task. In practice, however, it might be a better strategy to check that some given
${R}^{D}$
is not an MCM. In the rest of this paper we will collect some general results which can be helpful for this purpose.
We introduce the notion of
singularity sets; for the general theory of singularity sets and their relation to local cohomology we refer to the book [
ST71]
. For a variety
$X$
over some algebraically closed field
$k$
and some coherent sheaf
$\mathcal{\mathcal{F}}$
, the singularity sets of
$\mathcal{\mathcal{F}}$
are defined for integers
$i\ge 0$
as
$${S}_{i}(\mathcal{\mathcal{F}}):=\{x\in X{depth}_{{\mathcal{O}}_{X,x}}{\mathcal{\mathcal{F}}}_{x}\le i\},$$
i.e. the set of points
$x$
in
$X$
such that the depth of the stalk
${\mathcal{\mathcal{F}}}_{x}$
does not exceed
$i$
.
The sets
${S}_{i}(\mathcal{\mathcal{F}})$
are closed subsets of
$X$
and every coherent sheaf defines a filtration of
$X$
by closed subsets
$\varnothing \subset {S}_{0}(\mathcal{\mathcal{F}})\subset \cdots \subset {S}_{i}(\mathcal{\mathcal{F}})\subset \cdots =X$
. This filtration of course becomes stationary for
$i\ge dimX$
with
${S}_{dimX}(\mathcal{\mathcal{F}})=X$
. We are only interested in the situation where
$X={U}_{\sigma}$
is an affine toric variety and
$\mathcal{\mathcal{F}}=\mathcal{O}\left(D\right)$
, i.e. the sheafification of the module
${R}^{D}$
over
${U}_{\sigma}$
. Because
$D$
is
$T$
invariant, the depth of
$\mathcal{O}(D{)}_{x}$
remains constant over every orbit
$orb\left(\tau \right)\subset {U}_{\sigma}$
. For
$\tau \prec \sigma $
, he restriction
$\Gamma \left({U}_{\sigma},\mathcal{O}\left(D\right)\right)\to \Gamma \left({U}_{\tau},\mathcal{O}\left(D\right)\right)$
corresponds to the localization
${R}^{D}\to {R}_{\chi \left({m}_{\tau}\right)}^{D}$
, where
${m}_{\tau}$
is a lattice element from the relative interior of the cone
${\tau}^{\perp}\cap \stackrel{\u02c7}{\sigma}$
; in particular,
$k\left[{\tau}_{M}\right]=k[\sigma M{]}_{\chi \left({m}_{\tau}\right)}$
.
Denote
${\tau}_{{M}_{\tau}}:={\tau}_{M}\cap {M}_{\tau}$
, then the semigroup
${\tau}_{M}$
splits into a cartesian product
${\tau}_{M}={\tau}_{M}^{\perp}\times {\tau}_{{M}_{\tau}}$
. Correspondingly, the affine toric variety
${U}_{\tau}$
splits into the cartesian product
${T}^{\tau}\times {U}_{\tau}^{\prime}$
, where
${U}_{\tau}^{\prime}=speck\left[{\tau}_{{M}_{\tau}}\right]$
is an affine toric variety of dimension
$dcodim\tau $
. The corresponding projection
$p:{U}_{\tau}\twoheadrightarrow {U}_{\tau}^{\prime}$
is a flat morphism. The following is a wellknown fact on equivariant sheaves or
$M$
graded modules, respectively, which we present without proof.
Proposition 8.1:
Every
$T$
equivariant coherent sheaf
$\mathcal{\mathcal{E}}$
over
${U}_{\tau}$
is isomorphic to
${p}^{*}{\mathcal{\mathcal{E}}}^{\prime}$
for some
${T}^{\sigma}$
equivariant sheaf
${\mathcal{\mathcal{E}}}^{\prime}$
over
${U}_{\tau}^{\prime}$
. Equivalently, every finitely generated
$M$
graded
$k\left[{\tau}_{M}\right]$
module
$E$
is isomorphic to
${E}^{\prime}{\otimes}_{k\left[{\tau}_{M}^{\prime}\right]}k\left[{\tau}_{M}\right]$
for some
${M}_{\tau}$
graded
$k\left[{\tau}_{M}^{\prime}\right]$
module
${E}^{\prime}$
.
In particular,
$\mathcal{O}\left(D\right)\sim ={p}^{*}\mathcal{O}\left({D}^{\prime}\right)$
, where
${D}^{\prime}={\sum}_{\rho \in \tau \left(1\right)}{n}_{\rho}{D}_{\rho}$
is a
${T}^{\sigma}$
invariant divisor on
${U}_{\tau}^{\prime}$
and
$\mathcal{O}\left({D}^{\prime}\right)$
is the sheafification of the
$M$
graded
$k\left[{\tau}_{{M}_{\tau}}\right]$
module
${R}^{{D}^{\prime}}$
. We obtain:
Lemma 8.2:
For every
$x\in {U}_{\tau}$
, we have
${depth}_{{\mathcal{O}}_{{U}_{\tau},x}}\mathcal{O}(D{)}_{x}={depth}_{{\mathcal{O}}_{{U}_{\tau}^{\prime},p\left(x\right)}}\mathcal{O}({D}^{\prime}{)}_{p\left(x\right)}+codim\tau $
.

Proof.
As the morphism
$p$
is flat and thus local, we can apply [Mat89] , Thm. 23.3 and obtain
${depth}_{k[{\tau}_{M}{]}_{x}}{R}_{x}^{D}={depth}_{k[{\tau}_{M}{]}_{x}}{R}_{p\left(x\right)}^{{D}^{\prime}}{\otimes}_{k[{\tau}_{{M}_{\tau}}{]}_{p\left(x\right)}}k[{\tau}_{M}{]}_{x}={depth}_{k[{\tau}_{{M}_{\tau}}{]}_{p\left(x\right)}}{R}_{p\left(x\right)}^{{D}^{\prime}}+{depth}_{k[{\tau}_{M}{]}_{x}}{\left(k\left[{\tau}_{M}\right]/{\mathfrak{p}}_{\tau}\right)}_{x}$
for every point
$x\in {U}_{\tau}$
, where here
${\mathfrak{p}}_{\tau}$
denotes the maximal homogeneous ideal of
$k\left[{\tau}_{M}\right]$
.
With help of this lemma, we set:
Definition 8.3:
Let
$i\ge 0$
, then we set
${S}_{i}:=\{\tau \prec \sigma {depth}_{k[\sigma M{]}_{x}}{R}_{x}^{D}\le icodim\tau \text{for some point}x\in orb\left(\tau \right)\text{}\}.$
In this definition, we have omitted any explicit reference to
$D$
for clearer notation.
Note that
${S}_{i}$
is starclosed, and by the discussion above,
${S}_{i}\left(\mathcal{O}\left(D\right)\right)$
is equal to
${\cup}_{\tau \in {S}_{i}}orb\left(\tau \right)$
for all
$i\ge 0$
. Now observe:
Lemma 8.4:
${R}^{D}$
is MCM if and only if
${S}_{i}=\varnothing $
for
$0\le i<d$
.
Denote
${\Xi}^{\tau}:=\{\Pi \prec \tau (1\left)\right{\tau}_{\Pi}/\in star\left(\tau \right)\}$
, where we consider
${\Xi}^{\tau}$
as a subcomplex of the simplex of
$\tau $
. The following lemma is immediate:
Lemma 8.5:
Let
$\tau \prec \sigma $
and
$\Pi \subset \tau \left(1\right)$
, then
$\Pi \cap {\Xi}^{\tau}=\Pi \cap {\Xi}_{\tau}$
.
For any subset
$\Pi \subset \tau \left(1\right)$
, the splitting
$M\sim ={\tau}_{M}^{\perp}\times {M}_{\tau}$
is compatible with linear inequalities
$\langle m,n\left(\rho \right)\rangle <{n}_{\rho}$
, for
$\rho \in \Pi $
, respectively
$\langle m,n\left(\rho \right)\rangle \ge {n}_{\rho}$
for
$\rho \in \tau \left(1\right)\backslash \Pi $
, in the sense that some
$m\in M$
fulfills these inequalities if and only if every
${m}^{\prime}$
with
${m}^{\prime}m\in {\tau}_{M}^{\perp}$
fulfills these inequalities. The following is the main result of this section:
Theorem 8.6:
$\tau \in {S}_{k}$
if and only if
${\stackrel{~}{H}}^{i2}(\Pi \cap {\Xi}_{\tau},k)\ne 0$
for some
$i\le kcodim\tau $
and some subset
$\Pi \subset \tau \left(1\right)$
and there exists an integral solution to the system of inequalities
$$\begin{array}{cc}\langle m,n\left(\rho \right)\rangle & <{n}_{\rho}\text{for}\rho \in \Pi \end{array}$$  
$$\begin{array}{cc}\langle m,n\left(\rho \right)\rangle & \ge {n}_{\rho}\text{for}\rho \in \tau \left(1\right)\backslash \Pi \end{array}$$  
$$\begin{array}{}\end{array}$$  
in
$M$
or equivalently, in
${M}_{\tau}$
.

Proof.
We have
$\tau \in {S}_{k}$
if and only if
${H}_{{\mathfrak{p}}_{\tau}}^{i}{R}^{{D}^{\prime}}\ne 0$
for some
$i\le kcodim\tau $
, where
${\mathfrak{p}}_{\tau}$
is the maximal homogeneous ideal of
$k\left[{\tau}_{{M}_{\tau}}\right]$
. This in turn is equivalent to that there exists some
$m\in {M}_{\tau}$
and some
$i\le kcodim\tau $
such that
${\stackrel{~}{H}}^{i}({\Sigma}_{m}^{\tau}\cap {\Xi}^{\tau},k)\ne 0$
, where
${\Sigma}_{m}^{\tau}=\{\rho \in \tau (1\left)\right\langle m,n\left(\rho \right)\rangle <{n}_{\rho}\}$
. Moreover, by lemma 8.5 ,
${\Sigma}_{m}^{\tau}\cap {\Xi}^{\tau}={\Sigma}_{m}^{\tau}\cap {\Xi}_{\tau}$
.
The theorem can help to reduce the number of inequalities one has to check in order to determine the sets
${S}_{i}$
. However, in the case
$d=3$
, this is not of much help.
Proposition 8.7 ([ST71] , Corollary 1.21):
Let
$X$
be an irreducible variety of dimension
$d$
and let
$\mathcal{\mathcal{F}}$
be a coherent sheaf on
$X$
, then
$\mathcal{\mathcal{F}}$
is reflexive if and only if
$dim{S}_{i}(\mathcal{\mathcal{F}})\le i2$
for all
$i<d$
.
For
$d=3$
this implies that
${S}_{2}$
is either
$\left\{\sigma \right\}$
or empty. References

W. Bruns and J. Gubeladze. Semigroup algebras and discrete geometry. In Séminaires et Congrès, volume 6, pages 43–127. Soc. Math. France, Paris, 2002.

W. Bruns and J. Gubeladze. Divisorial Linear Algebra of Normal Semigroup Rings. Algebras and Representation Theory, 6:139–168, 2003.

A. Björner, M. Las Vergnas, B. Sturmfels, N. White, and G. Ziegler. Oriented matroids, volume 46 of Encyclopedia of Mathematics and its Applications. Cambridge University Press, 1993.

W. Fulton. Introduction to Toric Varieties. Princeton University Press, 1993.

S. Goto and K. Watanabe. On graded rings. II:
${\mathbb{Z}}^{n}$
graded rings. Tokyo J. Math., 1:237–261, 1978.

D. Helm and E. Miller. Bass numbers of semigroupgraded local cohomology. Pacific J. of Math., 209(1):41–66, 2003.

D. Helm and E. Miller. Algorithms for graded injective resolutions and local cohomology over semigroup rings. 2004. To appear.

H. Matsumura. Commutative Ring Theory, volume 8 of Cambridge Studies in Advanced Mathematics. Cambridge University Press, 1989.

M. Mustata. Local Cohomology at Monomial Ideals. J. Symb. Comput., 29(4–5):709–720, 2000.

T. Oda. Convex Bodies and Algebraic Geometry, volume 15 of Ergebnisse der Mathematik und ihrer Grenzgebiete. Springer, 1988.

M. Perling. Resolutions and Moduli for Equivariant Sheaves over Toric Varieties. PhD thesis, Universität Kaiserslautern, 2003.

M. Perling. Graded Rings and Equivariant Sheaves on Toric Varieties. Mathematische Nachrichten, 263–264:181–197, 2004.

M. Perling. Resolutions for Equivariant Sheaves over Toric Varieties. ICTP preprint no. IC/2004/59, 2004. math.AG/0503501.

M. Reid. Canonical 3folds. In A. Beauville, editor, Géométrie Algébrique, Angers 1979, pages 273–310. Sijthoff & Noordhoff, Alphen aan den Rijn, The Netherlands, 1980.

Y.T. Siu and G. Trautmann. GapSheaves and Extension of Coherent Analytic Subsheaves. Number 172 in Lecture Notes in Mathematics. SpringerVerlag, 1971.

N. V. Trung and L. T. Hoa. Affine semigroups and CohenMacaulay rings generated by monomials. Trans. AMS, 298(1):145–167, 1986.

G. Ziegler. Lectures on Polytopes, volume 152 of Graduate Texts in Mathematics. Springer, 1995.