, ,

The first author was partially supported by NSF grant DMS-0140172.

<ph f="cmbx">On the </ph> <math xmlns="http://www.w3.org/1998/Math/MathML"> <mi>K</mi> </math> <ph f="cmbx">-theory of smooth toric DM stacks</ph>

Lev A. Borisov and R. Paul Horja

Department of Mathematics, University of Wisconsin, Madison, WI, 53706, USA, borisov@math.wisc.edu, The Fields Institute, 222 College St, Toronto, Ontario, M5T 3J1, Canada, horja@fields.utoronto.ca.

Abstract. We explicitly calculate the Grothendieck $K$ -theory ring of a smooth toric Deligne-Mumford stack and define an analog of the Chern character. In addition, we calculate $K$ -theory pushforwards and pullbacks for weighted blowups of reduced smooth toric DM stacks.

1 Introduction

In this paper we calculate the Grothendieck

K

-theory rings of coherent sheaves on smooth toric Deligne-Mumford stacks. These stacks, which generalize the notion of a smooth toric variety, have been defined in [BCS] .

We are mostly interested in the reduced case, which is characterized by the condition that there is an open substack which is a subscheme.

However, since our technique is applicable to non-reduced stacks as well, we extend the result to the general case in a separate section. We find that the Grothendieck group of a smooth toric Deligne-Mumford stack

P_{Σ}

is generated by classes of invertible sheaves, and we find the generators of the ideal of relations satisfied by these sheaves. In the reduced case

K_{0} (P_{Σ})

is generated by the classes

R_{i}

of the invertible sheaves

ℒ_{i}

which correspond to the one-dimensional cones of the fan

Σ

. These sheaves generalize the sheaves

O (D_{i})

for codimension one strata

D_{i}

in a smooth toric variety. Moreover, we find the generators of the ideal of relations among

R_{i}^{\pm 1}

Theorem 4.10 . Let

B

be the quotient of the Laurent polynomial ring

Z [x_{1}, x_{1}^{- 1}, \dots, x_{n}, x_{n}^{- 1}]

by the ideal generated by the relations

$∙$ $\prod_{i = 1}^{n} x_{i}^{f (v_{i})} = 1$ for any linear function $f : N \to Z$ ,
$∙$ $\prod_{i \in I} (1 - x_{i}) = 0$ for any set $I \subseteq [1, \dots, n]$ such that $v_{i}, i \in I$ are not contained in any cone of $Σ$ .

Then the map

ρ : B \to K_{0} (P_{Σ})

which sends

x_{i}

R_{i}

is an isomorphism.

We also show that the

K

-theory with complex coefficients is a semilocal Artinian

C

-algebra and explicitly describe its local components. We produce a vector space isomorphism between the

K

-theory with complex coefficients and a combinatorially defined ring dubbed SR-cohomology, which we call combinatorial Chern character. It generalizes the usual Chern character in the case of projective toric varieties and is expected to coincide with the equivariant Chern character of [AR] (see also the very recent preprint [JKK] ) in the projective DM stack case.

The paper is structured as follows. In Section 2 we recall the definition and basic properties of smooth toric DM stacks. We restrict our attention to the reduced case, which makes the Gale duality construction of [BCS] significantly easier to describe. In Section 3 we introduce the SR-cohomology in the reduced case and describe its decomposition into sectors. In Section 4 we calculate the

K

-theory of the reduced smooth toric DM stack. The key idea is to use the homogenization trick of [E] to resolve any coherent sheaf on

P_{Σ}

by direct sums of invertible sheaves.

Section 5 defines the combinatorial Chern character, and Section 6 extends the results of Sections 4 and 5 to the nonreduced case. Sections 7 , 8 and 9 describe the

K

-theory pullbacks and pushforwards for special classes of morphisms between reduced smooth toric DM stacks. These results will be used in a subsequent paper [BH] on homological mirror symmetry and the GKZ hypergeometric system of partial differential equations.

There is some overlap between the results of this paper and those of the recent preprint [Ba] , where the equivariant K-theory of smooth toric varieties is studied with the help of the Merkurjev's spectral sequence.

2 Review of reduced smooth toric DM stacks

In this section we will briefly review the definitions of toric Deligne-Mumford stacks, as developed in [BCS] . We are specifically interested in the reduced case, which simplifies the construction significantly.

Let

N

be a free abelian group, and let

Σ

be a simplicial fan in

N

See [F] for the definition of a fan. A stacky fan

Σ

is defined as the data

(Σ, {v_{i}})

where

{v_{i}}

is a collection of lattice points, one for each one-dimensional cone

C_{i} \in Σ

. We require

v_{i} \in C_{i} \cap (N - {0})

, but

v_{i}

may or may not be the minimum lattice point on

C_{i}

. The paper [BCS] makes an additional technical assumption

\begin{matrix} elements v_{i} span  a  finite  index  subgroup  of N . \end{matrix}

(1)

Remark 2.1. It is likely that this assumption 1 is just an artifact of the technique of [BCS] . In general, one can always look at the preimage

N_{f}

N

of the torsion part of

N / S p a n (v_{i}, i = 1, \dots, n)

. The quotient

N / N_{f}

splits off (noncanonically) as a free direct summand. It is reasonable to expect that the DM stack that corresponds to

Σ

is the product of

(C^{*})^{r k (N / N_{f})}

and the toric DM stack of [BCS] constructed for

Σ

N_{f}

instead of

N

. However, the functoriality of this construction is a bit unclear, in view of the fact that the splitting is non–canonical.

The collection

{v_{i}}, i = 1, \dots, n

gives a map

φ : Z^{n} \to N

with finite cokernel. We dualize to get an injective map

φ^{\lor} : N^{\lor} \to (Z^{n})^{\lor} .

The Gale dual of

φ

(see [BCS] ) is simply the map

(Z^{n})^{\lor} \to C o k e r (φ^{\lor}) .

Denote by

G

the algebraic group

H o m (C o k e r (φ)^{\lor}, C^{*}) .

The exact sequence

\begin{matrix} 0 \to N^{\lor} \to (Z^{n})^{\lor} \to C o k e r (φ^{\lor}) \to 0 \end{matrix}

(2)

leads to the exact sequence

0 \to G \to (C^{*})^{n} \to H o m (N^{\lor}, C^{*}) .

Hence,

G

can be thought of as a subgroup of

(C^{*})^{n}

. Its

C

-points can be thought of as collections of

n

nonzero complex numbers

λ_{i}, i = 1, \dots, n

which satisfy the condition

\begin{matrix} ^{n} \prod_{i = 1} λ_{i}^{f (v_{i})} = 1 \end{matrix}

(3)

for all linear functions

f : N \to Z

Consider the subset

Z

C^{n}

that consists of all the points

z = (z_{1}, \dots, z_{n})

such that the set of

v_{i}

for the zero coordinates of

z

is contained in a cone of

Σ

. Then the toric DM stack

P_{Σ}

that corresponds to the stacky fan

Σ = (Σ, {v_{i}})

is defined as the stack quotient

[Z / G]

where

Z

and

G

are endowed with the natural reduced scheme structures.

The action of

(λ_{i})

(z_{i})

is given by

(λ_{i} z_{i})

. It has been shown in [BCS] that

P_{Σ}

is a Deligne-Mumford stack whose moduli space is the simplicial toric variety

P_{Σ}

To a cone in

C

one can associate a closed substack of

P_{Σ}

by looking at the quotient of

N

by the sublattice spanned by

v_{i} \in C

. The new fan is defined as the image of the link of

C

Σ

, and the new

v_{j}

are the images of the old

v_{j}

from the link. Unfortunately, this procedure may introduce torsion into

N

, so the resulting substack is not reduced.

Moreover, in order for the construction of [BCS] to work, the images of

v_{j}

from the link of

C

need to generate a finite index subgroup in

N / S p a n (v_{i} \in C)

. The easiest way to assure this is by imposing a stronger technical condition

\begin{matrix} every  cone  of Σ is  contained  in  a   cone  of  dimension  rk N, \end{matrix}

(4)

see [BCS] . This also assures that

P_{Σ}

is covered by open substacks of the form

C^{r k N} / G_{i}

where

G_{i}

are finite abelian groups. These open substacks can be indexed by the cones in

Σ

of maximum dimension. On the other hand, one can definitely talk about closed toric subvarieties of the coarse moduli space

P_{Σ}

P_{Σ}

without the additional assumption 4 . Throughout this paper we will only use the assumption 1 .

3 Orbifold cohomology and SR-cohomology of reduced smooth toric DM stacks

In this section we introduce some combinatorial invariants of toric DM stacks which we call SR-cohomology rings. They coincide with the orbifold cohomology rings in the projective case but are generally different, even for smooth toric varieties.

A natural combinatorial invariant of a fan is the partial semigroup ring

Z [N, Σ]

which is defined as a free abelian group with the basis

{[w], w \in N \cap Σ}

indexed by the lattice elements inside the support of the fan and multiplication

[w_{1}] * [w_{2}] = {\begin{matrix} [w_{1} + w_{2}], & i f t h e r e e x i s t s C \in Σ, C ∋ w_{1}, w_{2} \\ 0, & o t h e r w i s e . \end{matrix}

The rings

Q [N, Σ]

and

C [N, Σ]

are defined analogously. For a stacky fan

Σ

, these rings are given additional structure of graded rings. The grading can take nonnegative rational values. It is defined by

deg ([w]) = \sum_{i} α_{i}

where

w = \sum_{i} α_{i} v_{i}

is the unique way of writing

w

as a rational linear combination of

v_{i}

that lie in the minimum cone of

Σ

that contains it.

To any stacky fan

Σ

one can associate a SR-cohomology ring, to be denoted by

H_{S R} (P_{Σ}, C)

Definition 3.1. Pick a basis

{f_{1}, \dots, f_{r k N}}

H o m (N, Z)

. Each

f_{i}

gives rise to a degree one element

t_{i} : = \sum_{j = 1}^{n} f_{i} (v_{j}) [v_{j}]

C [N, Σ]

. Then

H_{S R} (P_{Σ}, C) : = C [N, Σ] / 〈 t_{1}, \dots, t_{r k N} 〉 .

The rings

H_{S R} (P_{Σ}, Q)

and

H_{S R} (P_{Σ}, Z)

are defined analogously.

It is clear that the above defined SR-cohomology rings are independent of the choice of the basis

{f_{i}}

. If

P_{Σ}

is a smooth projective toric variety, then its combinatorial cohomology rings are isomorphic to its usual cohomology rings. Indeed, this is precisely the Stanley-Reisner presentation of the cohomology ring of a smooth projective toric variety.

This is the motivation behind our notation

H_{S R}

. We call our rings SR-cohomology rings, as opposed to Stanley-Reisner cohomology rings, to avoid confusion with the term Stanley-Reisner ring that is already used in the literature.

Remark 3.2. We abuse notation slightly and still use

H_{S R} (P_{Σ}, *)

even if

Σ

does not satisfy 1 . Hopefully, one will eventually define toric stacks without this restriction.

The main technical result of [BCS] connects the SR-cohomology of a smooth toric DM stack with its orbifold cohomology, see [CR] and [AGV] . However, one needs an important additional assumption that the coarse moduli space

P_{Σ}

is projective.

Theorem 3.3. ([BCS] ) For any

P_{Σ}

such that

P_{Σ}

is projective, there holds

H_{o r b} (P_{Σ}, Q) \sim = H_{S R} (P_{Σ}, Q) .

Remark 3.4. In general, the SR-cohomology and orbifold or usual cohomology rings are different. The simplest example is given by the fan

Σ

which is the union of the first and the third quadrants in

Z^{2}

. The elements

v_{i}

are

(\pm 1, 0)

and

(0, \pm 1)

The corresponding variety is

P^{1} \times P^{1} - {(0, \infty), (\infty, 0)}

. The SR-cohomology is three-dimensional, with dimension

2

graded component of degree one. It is unclear what geometrically defined cohomology theory on

P^{1} \times P^{1} - {(0, \infty), (\infty, 0)}

can produce such a ring.

Despite Remark 3.4 , the SR-cohomology ring

H_{S R} (P_{Σ}, C)

is very well behaved.

Proposition 3.5. The SR-cohomology ring

H_{S R} (P_{Σ}, C)

of any (reduced) stacky fan

Σ

is a finite dimensional complex vector space. It is a local Artinian graded

C

-algebra with the maximum ideal given by the span of the elements of positive degree.

Proof. Since $H_{S R} (P_{Σ}, C)$ is nonnegatively graded and its degree zero part is isomorphic to $C$ , it is enough to show that it is an Artinian ring.
Hence, let us study $C$ -algebra homomorphisms $φ : H_{S R} (P_{Σ}, C) \to C$ .
In view of relations in $C [N, Σ]$ , the values of $φ ([w])$ are nonzero only for $w$ in some cone $C$ of $Σ$ . Then the relations $t_{i}$ show that these values must be zero for $w \in C$ that are integer linear combinations of $v_{j} \in C$ .
Finally, for any other nonzero $w \in C$ , some positive multiple of it is an integer linear combination of $v_{j} \in C$ . This leads to $φ ([w])^{l} = 0$ , so $φ ([w]) = 0$ for all nonzero $n$ . This shows that the only maximum ideal in $H_{S R} (P_{Σ}, C)$ is the span of the elements of positive degree. □

In the case of projective

P_{Σ}

, there is a natural decomposition of the orbifold and SR-cohomology into direct sum of sectors. Each sector corresponds to the usual cohomology of some closed toric subvariety of

P_{Σ}

. We observe that this decomposition still occurs for the SR-cohomology, even without the projectivity assumption. Moreover, we will not even assume 1 .

Definition 3.6. The untwisted sector of SR-cohomology is defined as the subring of

H_{S R} (P_{Σ}, C)

generated by the images of

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

Remark 3.7. It is easy to see that the choice of

v_{i}

in the corresponding one-dimesnional cone does not change the untwisted sector much, if one works with rational or complex coefficients. Indeed, one can simply rescale the corresponding variables. Consequently, we can talk about the SR-cohomology of the toric variety

P_{Σ}

, in analogy with the projective case. We can define it as the untwisted sector of the SR-cohomology of the stack

P_{Σ}

obtained by picking the

v_{i}

to be the smallest lattice points on the one-dimensional cones of

Σ

. We denote these rings by

H_{S R} (P_{Σ}, Q o r C)

For a cone

C \in Σ

, let

B o x (C)

be the set of elements

v

N

which are linear combinations with rational coefficients in the range

[0, 1)

of elements

v_{i} \in C

. Let

B o x (Σ)

be the union of

B o x (C)

for all

C \in Σ

Proposition 3.8. As a module over the untwisted sector, the ring

H_{S R} (P_{Σ}, C)

is a direct sum of the modules

H_{v}

generated by the images in

H_{S R} (P_{Σ}, C)

of the elements

[v] \in C [N, Σ]

for all

v \in B o x (Σ)

Proof. The statement follows from the analogous result for $C [N, Σ]$ , where it is obvious. □

In the projective case, each of the modules

H_{v}

is isomorphic to the cohomology of the toric subvariety of

P_{Σ}

that corresponds to the minimum cone of

Σ

that contains

v

. Remarkably, this is true in general if one works in SR-cohomology with complex or rational coefficients.

Proposition 3.9. The module

H_{v}

is isomorphic to the SR-cohomology with complex coefficients of the (closed) toric subvariety of

P_{Σ}

that corresponds to the minimum cone of

Σ

that contains

v

Proof. The proof is analogous to that of [BCS,Proposition5.2] , but we briefly sketch it here for the benefit of the reader. Let N 1 denote the quotient of N by the subgroup generated by v i ∈ C , and let N ′ = N 1 / t o r s i o n be its torsion-free part. Let Σ ′ in N ′ be the image of the link of C in Σ and let v i ′ be the images of v i from the link. The SR-cohomology of the closed toric subvariety that corresponds to C is isomorphic to the untwisted sector in the SR-cohomology H S R ( P Σ ′ , C ) of Σ ′ .
Observe that H v is a quotient of the submodule M v of C [ N , Σ ] which is supported on the star of C . Moreover, it is generated by the monomial of the form [ n + v ] where n is a lattice point in the star of C in Σ which is an integer linear combination of the v j in the minimum cone that contains it. This identifies H v with the quotient of the polynomial ring in the variables D 1 , … , D k that correspond to one-dimensional faces of C and cones in its link by the ideal with generators
- $∙$ $\prod_{i \in I} D_{i}$ , if no cone $C^{'} \supseteq C$ contains all $v_{i}, i \in I$ ,
- $∙$ $\sum_{i, v_{i} \in s t a r (C)} f (v_{i}) D_{i}$ for any linear map $f : N \to Z$ .
On the other hand, the untwisted sector of H S R ( P Σ ′ , C ) is isomorphic to the quotient of the polynomial ring in the variables D 1 ′ , … , D k ′ ′ which correspond to one-dimensional faces of cones in the link of C (hence not in C ), by the ideal generated by
- $∙$ $\prod_{i \in I} D_{i}^{'}$ , if no cone $C^{'} \supseteq C$ contains all $v_{i}, i \in I$ ,
- $∙$ $\sum_{i, v_{i} \in l i n k (C)} f^{'} (ρ (v_{i})) D_{i}^{'}$ for any linear map $f^{'} : N^{'} \to Z$ .
Here ρ : N → N ′ is the projection.
To connect these two spaces, observe that linear functions on N ′ lift to linear functions on N . Pick a complementary sublattice in the lattice of linear functions on N and pick its basis f 1 , … , f dim C . These f i provide relations on D i that allow one to express D i , v i ∈ C , in terms of D i , v i ∈ l i n k ( C ) , in the relations for H v . The remaining relations are precisely those of the untwisted sector of H S R ( P Σ ′ , C ) . Moreover, this isomorphism is independent of the choice of the complementary lattice or its basis. □

Remark 3.10. Propositions 3.8 and 3.9 still hold for SR-cohomology with rational coefficients. The proofs are unchanged.

Remark 3.11. Suppose that the fan

Σ

is projective, or that it is a subdivision of a cone. Under these assumptions, if one picks a basis

{f_{j}}

N^{\lor}

, then the corresponding elements

\sum_{i = 1}^{n} f (v_{i}) [v_{i}]

form a regular sequence in

Q [N]^{Σ}

. As a result, the graded dimension of

H_{S R} (Σ, C)

, defined as

\sum_{d \in Q} t^{d} {dim}_{C} H_{S R} (P_{Σ}, C)_{deg = d}

, equals

g r . d i m H_{S R} (P_{Σ}, C) = (1 - t)^{r k N} \sum_{n \in N \cup Σ} t^{deg (n)} .

The proof of this regularity is sketched in [BCS] for the projective case and in [BM] for the cone case. In general, the graded dimension of

H_{S R} (Σ, C)

is the sum over sectors of the graded dimensions of the SR-cohomology of the sector, but it lacks such a nice combinatorial formula.

4 $K$ -theory of reduced toric DM stacks

The goal of this section is to prove a combinatorial description of the

K

-theory of reduced toric DM stack which is analogous to the Stanley-Reisner presentation of the cohomology of a smooth toric variety.

Analogous statements for smooth toric varieties are contained in [Ba] .

The resulting ring is then compared to the SR-cohomology of the stack.

We use the notations from the previous sections. We do not make any assumptions on the stacky fan apart from 1 .

Definition 4.1. Let

X

be a smooth Deligne-Mumford stack.

Define the (Grothendieck)

K

-theory group

K_{0} (X)

to be the quotient of the free abelian group generated by coherent sheaves

ℱ

X

by the relations

[ℱ_{1}] - [ℱ_{2}] + [ℱ_{3}]

for all exact sequences

0 \to ℱ_{1} \to ℱ_{2} \to ℱ_{3} \to 0

Remark 4.2. The

K

-theory of

X

admits a product structure by

[ℱ_{1}] * [ℱ_{2}] = \sum_{i = 0}^{dim X} (- 1)^{i} [T o r^{i} (ℱ_{1}, ℱ_{2})] .

The image of the structure sheaf

O

plays the role of the identity.

We recall that the category of coherent sheaves on

[Z / G]

is equivalent to that of

G

-linearized coherent sheaves on

Z

, see [V,Example7.21] . We will always implicitly use this equivalence.

Definition 4.3. For each

i

from

1

n

we define a

G

-linearized invertible sheaf

ℒ_{i}

Z

as follows. As a sheaf, it will be isomorphic to the structure sheaf

O_{Z}

. For

g = (λ_{1}, \dots, λ_{n}) \in G

the isomorphism

O_{Z} \to g^{*} O_{Z} = O_{Z}

sends

1

λ_{i}

. Here we have used the canonical isomorphism

g^{*} O_{Z} = O_{Z}

. We define

R_{i}

to be the image of

ℒ_{i}

in the

K

-theory of

P_{Σ}

Remark 4.4. The sheaf

ℒ_{i}

has a

G

-invariant global section given by the restriction to

Z

of the

i

-th coordinate function

z_{i}

C^{n}

. Indeed,

g^{*} z_{i} = λ_{i} z_{i}

, so

z_{i}

is compatible with the isomorphisms of the above definition. In the case of a smooth toric variety,

ℒ_{i}

corresponds to

O (D_{i})

where

D_{i}

is the divisor of the

i

-th dimension one cone of

Σ

Proposition 4.5. The elements

R_{i}

K_{0} (P_{Σ})

satisfy the following relations.

$∙$ $\prod_{i = 1}^{n} R_{i}^{f (v_{i})} = 1$ for any linear function $f : N \to Z$ ,
$∙$ $\prod_{i \in I} (1 - R_{i}) = 0$ for any set $I \subseteq [1, \dots, n]$ such that the $v_{i}, i \in I$ are not contained in any cone of $Σ$ .

Proof. The relations 3 and the definition of $ℒ_{i}$ show that $\otimes_{i} ℒ_{i}^{f (v_{i})}$ are in fact trivially linearized on $Z$ for every linear function $f : N \to Z$ .
Passing to $K$ -theory, we get the first set of relations on $R_{i}$ .
To get the second set of relations, consider the Koszul complex of the set of elements $z_{i}, i \in I$ in the ring $A = C [z_{1}, \dots, z_{n}]$ . It gives an exact sequence $\dots \to \oplus_{J \subseteq I, | J | = k} A \to \oplus_{J \subseteq I, | J | = k - 1} A \to \dots$ $\dots \to \oplus_{j \in I} A \to A \to A / 〈 z_{i}, i \in I 〉 \to 0$ of $A$ -modules. We can pass to the associated sheaves on $C^{n}$ and then restrict them to $Z$ to get $0 \to \dots \to \oplus_{J \subseteq I, | J | = k} O_{Z} \to \oplus_{J \subseteq I, | J | = k - 1} O_{C^{n}} \to \dots \to 0$ where we have used the condition on $I$ to see that $A / 〈 z_{i}, i \in I 〉$ is supported outside of $Z$ . The maps between the copies of $O_{Z}$ for $J_{1}$ and $J_{2}$ are zero unless $J_{2} \subset J_{1}$ and are $\pm z_{J_{1} - J_{2}}$ otherwise. We can make this into a complex of $G$ -linearized sheaves $0 \dots \to \oplus_{J \subseteq I, | J | = k} \otimes_{j \in J} ℒ_{j}^{- 1} \to \oplus_{J \subseteq I, | J | = k - 1} \otimes_{j \in J} ℒ_{j}^{- 1} \to \dots \to 0$ by using the $G$ -equivariant maps $z_{j} : ℒ_{j}^{- 1} \to O_{Z}$ which are twists by $ℒ_{j}^{- 1}$ of the maps of Remark 4.4 .
The alternating sum of any long exact sequence of sheaves is zero in the $K$ -theory, which implies the second set of relations, after multiplying by the invertible element $\prod_{i \in I} R_{i}$ . □

Theorem 4.6. The elements

R_{i}

generate the

K_{0}

-theory of the reduced toric DM stack

P_{Σ}

Proof. Consider a

G

-linearized coherent sheaf

ℱ

on the open subset

Z

C^{n}

. Let us denote the embedding of

Z

into

C^{n}

i

. We observe that

i^{*} i_{*} ℱ \to ℱ

is an isomorphism, simply because

i

is an open embedding.

Since

C^{n}

is affine,

i_{*} ℱ

is the sheaf associated to

C [z_{1}, \dots, z_{n}]

-module

M = H^{0} (Z, ℱ)

Lemma 4.7. In the notations above,

M

is a finitely generated module.

Proof of Lemma 4.7 . We know by [H,ExerciseII.5.15] that

ℱ

is a restriction to

Z

of some coherent sheaf on

C^{n}

. This coherent sheaf can be resolved by free sheaves on

C^{n}

. When we restrict to

Z

we see that

ℱ

is resolved by direct sums of

O_{Z}

. Hence, it is enough to see that all the cohomology groups of

O_{Z}

are finitely generated modules over

C [z_{1}, \dots, z_{n}]

. By using the

(C^{*})^{n}

-action, we see that

H^{0} (Z, O_{Z})

is generated by some monomials

\prod_{j} z_{j}^{r_{j}}

, if one thinks of it as the subspace of the quotient field of

C [z_{1}, \dots, z_{n}]

. We easily see that

C^{n} - Z

is a union of coordinate subspaces of codimension at least two, which implies that

r_{i}

have to be nonnegative. Then we see that

H^{0} (Z, O_{Z}) = C [z_{1}, \dots, z_{n}]

To calculate the higher cohomology

H^{*} (Z, O_{Z})

groups, we can cover

Z

by open affine subsets

U_{C}

of the form

U_{C} = {(z_{1}, \dots, z_{n}), z_{i} \neq 0 f o r v_{i} \notin C}

where

C

runs over the set of nonzero cones of

Σ

. The sections of

O_{Z}

U_{C}

are the linear span of monomials

\prod_{j} z_{j}^{r_{j}}

such that the set of

i

with

r_{i} < 0

is contained in the set of

i

with

v_{i} \notin C

. Various intersections

U_{C_{1} C_{2} \dots C_{k}}

U_{C}

give the spans of monomials with the condition that for each

j

with

r_{j} < 0

the lattice element

v_{j}

is not contained in

\cap_{l} C_{l}

. We use the cover by sets of type

U_{C}

to calculate the cohomology of

O_{Z}

as Čech cohomology. For a given

(r_{1}, \dots, r_{n})

, the cohomology of

O_{Z}

that corresponds to this

Z^{n}

-grading is given by the reduced homology of the following simplicial complex. Its set of vertices is the set

Σ_{+}

of all cones of positive dimension in

Σ

, and its maximum simplices are the complements in

Σ_{+}

of the one-element subsets

{C}

that correspond to cones that do not contain any

v_{i}

with

r_{i} < 0

. So we have a simplicial complex which is a union of a collection of complements of one-dimensional subsets. If not all the elements of

Σ_{+}

are in this collection, then the resulting complex is a cone, since adding such an element has no effect on whether a subset of

Σ_{+}

belongs to the complex. Hence, it has trivial reduced homology. If all elements of

Σ_{+}

are a part of the collection, then there can be no

v_{i}

with

r_{i} < 0

. The resulting complex is a sphere, and we have a one-dimensional top reduced homology that corresponds to the monomial in

H^{0} (Z, O_{Z})

This shows that the higher cohomology

H^{> 0} (Z, O_{Z})

groups vanish, which finishes the proof of the lemma.

□

Proof of Theorem 4.6 continues. The

G

-linearization on

ℱ

gives rise to a

G

-action on

M

, which is compatible with the

G

-action on

A = C [z_{1}, \dots, z_{n}]

, in the sense that

g (r m) = g (r) g (m)

for all

g \in G

r \in A

and

m \in M

. Moreover, we claim that

M

is generated by a finite set of eigenelements

m_{j}

with

g (m_{j}) = χ_{j} (g) m_{j}

for some character

χ_{j} : G \to C^{*}

. In view of the structure of

G

, all of its algebraic finite-dimensional actions are diagonalizable. As a consequence, a module

M

is generated by a finite set of eigenelements if and only if

\begin{matrix} \begin{matrix} for any m \in M the linear span of g m, g \in G \\ is finite-dimensional, and the action of G on it is algebraic. \end{matrix} \end{matrix}

(5)

It is clear that this finiteness condition 5 for a

G

-equivariant Noetherian

A

-module implies 5 for all equivariant submodules and quotients.

Recall (see [MF] ) that the

G

-linearization of

ℱ

Z

is given by an isomorphism

φ : μ^{*} ℱ \to π_{2}^{*} ℱ

G \times Z

where

μ

is the multiplication and

π_{2}

is the second projection. Denote by

A_{G}

the ring of regular functions on

G

. We will use the fact that

H^{0} (Z, O_{Z}) = H^{0} (C^{n}, O_{C^{n}}) = A

from the proof of Lemma 4.7 . The isomorphism

φ

induces a map

φ_{g l o b a l} : M \otimes_{A} (A_{G} \otimes_{C} A) \to M \otimes_{C} A_{G}

where the tensor multiplication on the left is via the map

A \to A_{G} \times A

induced by the action of

G

Z

. Indeed, the left hand side maps to the global sections of

μ^{*} (ℱ)

, whereas the right hand side consists of the global sections of

π_{2}^{*} ℱ

. This map

φ_{g l o b a l}

encodes the action of

G

M

by mapping

m \otimes (f \otimes a) \mapsto \sum_{k} m_{k} \otimes f_{k}

such that for any

g \in G

there holds

a g (m) f (g) = \sum_{k} f_{k} (g) m_{k}

. By taking

a = f = 1

we see that there is a finite sum

\sum_{k} m_{k} \otimes f_{k}

such that

g (m) = \sum_{k} f_{k} (g) m_{k}

for all

g \in G

. In particular, the span of

g m, g \in G

is finite-dimensional. By picking a basis of it, and applying the above, we see that the action of

G

is algebraic, which shows 5 .

Consequently, there is a presentation

F_{1} \to F_{0} \to M \to 0

where

F_{i}

is a direct sum of rank one

C [z_{1}, \dots, z_{n}]

-modules generated by eigenelements of

G

. Indeed, the kernel of

F_{0} \to M

also satisfies 5 .

Hence, it is generated by a finite number of eigenelements, which allows one to construct

F_{1}

. We will now use the homogenization trick of [E,Corollary19.8] . Namely, consider the ring

C [z_{0}, \dots, z_{n}]

and extend the

G

-action on it by

g z_{0} = z_{0}

for all

g \in G

. The map

F_{1} \to F_{0}

is given by a matrix

S

of polynomials in

z_{1}, \dots, z_{n}

. Define the matrix

\tilde{S}

of homogeneous polynomials of some large fixed degree

d

by multiplying each monomial in

S

by an appropriate power of

z_{0}

. We will then have a presentation of some module

\tilde{M}

over

C [z_{0}, \dots, z_{n}]

{\tilde{F}}_{1} \to {\tilde{F}}_{0} \to \tilde{M} \to 0 .

Here

{\tilde{F}}_{i}

are direct sums of rank one

G

-equivariant

C [z_{0}, \dots, z_{n}]

-modules generated by homogeneous eigenelements. As in [E,Corollary19.8] , we observe that

C [z_{1}, \dots, z_{n}] = C [z_{0}, \dots, z_{n}] / (1 - z_{0})

and

\tilde{M} / (1 - z_{0}) \tilde{M} = M

. Given a

G

-equivariant homogeneous module

\tilde{F}

over

C [z_{0}, \dots, z_{n}]

, we can look at the finite-dimensional vector space

\tilde{F} / < z_{0}, \dots, z_{n} > \tilde{F} .

This vector space inherits the

G

-action. It also inherits the the grading, which is preserved by the

G

-action. As a result, this vector space has a basis of homogeneous eigenelements. Each such eigenelement can be lifted to an element

r_{j}

\tilde{F}

. By looking at the (finite-dimensional)

G

-span of

r_{j}

\tilde{F}

, we can modify

r_{j}

to be itself a homogeneous eigenelement.

The elements

r_{j}

generate

\tilde{F}

by the graded Nakayama lemma.

We apply this procedure to the kernel

K

{\tilde{F}}_{1} \to {\tilde{F}}_{0}

, and continue on to build a resolution of

\tilde{M}

. Since the end of this resolution is the minimal graded resolution of the homogenous module

K

, it terminates.

We then get a free resolution

0 \to {\tilde{F}}_{l} \to \dots \to {\tilde{F}}_{0} \to \tilde{M} \to 0

\tilde{M}

such that each

{\tilde{F}}_{i}

is freely generated by homogeneous eigenelements

r_{i j}

and the maps are

G

-equivariant and are compatible with the grading. We mod out this resolution by

1 - z_{0}

to get a resolution of

M

. The exactness follows from

T o r_{C [z_{0}, \dots, z_{n}]}^{> 0} (M, C [z_{1}, \dots, z_{n}]) = 0

, as in [E,Corollary19.8] . We thus get a

G

-equivariant resolution of

M

by direct sums of free modules of rank one generated by eigenelements. We claim that each such module can be identified with the global sections of a tensor product of powers of line bundles

ℒ_{i}

. Indeed, one simply needs to show that every character of

G

is a restriction of a character of

(C^{*})^{n}

, which follows from the exact sequence 2 .

We pass to the corresponding exact sequence of

G

-linearized sheaves on

C^{n}

and then restrict it to

Z

to see that

ℱ

is a linear combination of products of

R_{i}

. □

Corollary 4.8. The bounded derived category of the category of coherent sheaves on

P_{Σ}

is generated by the invertible sheaves

\otimes_{i} ℒ_{i}^{r_{i}}

Proof. The argument of Theorem 4.6 shows that every sheaf on $P_{Σ}$ admits a free resolution by direct sums of invertible sheaves of the above type. □

Remark 4.9. We refer the reader to [K2] for much stronger results concerning these derived categories.

Theorem 4.10. Let

B

be the quotient of the Laurent polynomial ring

Z [x_{1}, x_{1}^{- 1}, \dots, x_{n}, x_{n}^{- 1}]

by the ideal generated by the relations

$∙$ $\prod_{i = 1}^{n} x_{i}^{f (v_{i})} = 1$ for any linear function $f : N \to Z$ ,
$∙$ $\prod_{i \in I} (1 - x_{i}) = 0$ for any set $I \subseteq [1, \dots, n]$ such that $v_{i}, i \in I$ are not contained in any cone of $Σ$ .

Then the map

ρ : B \to K_{0} (P_{Σ})

which sends

x_{i}

R_{i}

is an isomorphism.

Proof. Theorem 4.6 shows that

ρ

is surjective. It is therefore sufficient to show its injectivity. We will define a map

ρ_{1} : K_{0} (P_{Σ}) \to B

and prove that

ρ_{1} \circ ρ = i d

For any

G

-linearized sheaf

ℱ

Z

, consider the

G

-equivariant module

M = H^{0} (Z, ℱ)

over

A = C [z_{1}, \dots, z_{n}]

. Consider the

A

-module

C

with trivial

G

-action. For each

i

from

0

n

the finite-dimensional vector space

T o r_{A}^{i} (M, C)

is acted upon by

G

. It is a direct sum over the characters of

χ : G \to C^{*}

of the eigenspaces

V_{i, χ}

. The group

H o m (G, C^{*})

is a quotient of

Z^{n}

N^{\lor}

. Because of the first set of relations on

x_{i}

, every character

χ

gives a well-defined monomial

x_{χ} \in B

. We then define

ρ_{1} : ℱ \mapsto \sum_{χ} \sum_{i = 0}^{n} (- 1)^{i} {dim}_{C} (V_{i, χ}) x_{χ} .

We would like to show that

ρ_{1}

passes down to a well-defined map on

K

-theory, i.e. it is additive on short exact sequences. If

\begin{matrix} 0 \to ℱ_{1} \to ℱ_{2} \to ℱ_{3} \to 0 \end{matrix}

(6)

is an exact sequence of

G

-linearized sheaves on

Z

with

G

-equivariant morphisms, then the sequence of

A

-modules

0 \to M_{1} \to M_{2} \to M_{3}

is only exact on the left. We complete it to a long exact sequence

0 \to M_{1} \to M_{2} \to M_{3} \to M_{4} \to 0 .

This long exact sequence of modules splits into short exact sequences, which, in turn, give long exact sequences of

T o r^{i} (*, C)

. This shows that

ρ_{1} (M_{1}) + ρ_{1} (M_{3}) - ρ_{1} (M_{2}) = \sum_{χ} \sum_{i = 0}^{n} (- 1)^{i} {dim}_{C} (T o r^{i} (M_{4}, C)_{χ}) R_{χ} .

Because of 6 ,

M_{4}

is associated to the

G

-equivariant sheaf

ℱ_{4}

which is supported on

C^{n} - Z

. We will show that

\begin{matrix} \sum_{χ} \sum_{i = 0}^{n} (- 1)^{i} {dim}_{C} (T o r^{i} (M_{4}, C)_{χ}) x_{χ} = 0 \end{matrix}

(7)

B

by Noetherian induction on

M_{4}

. Since the above element is additive on short exact sequences, it is enough to find a

G

-equivariant submodule

M

M_{4}

which satisfies 7 . Pick an associated prime

p

M_{4}

and consider an element

m \in M_{4}

such that

A n n (m) = p

. In general, we can not expect

m

to be an eigenelement, nor can we expect

p

to be

G

-invariant. However, let us look at the vector space

V

which is the linear span of

g m, g \in G

. Observe that for each

g \in G

, the annihilator

g p

g m

is also an associated prime of

M_{4}

. Since

G

may not be connected, it could conceivably permute the associated primes.

Consider the ideal

I \subset A

given by

I = ⋂_{g \in G} g p .

It is clear that

I

G

-invariant and that it annihilates

V

. Hence the submodule

M

M_{4}

generated by

V

is an

A / I

-module.

Since

M_{4}

is supported on

C^{n} - Z

, each of its associated primes contains a prime ideal of

A

that corresponds to an irreducible component of

C^{n} - Z

. More specifically, this is an ideal

J

which is generated by

z_{i}

with indices

i

, such that

v_{i}

do not lie in a cone

C

Σ

. Since

p \supseteq J

, we get

I \supseteq J

, and

M

is also an

A / J

-module. We now repeat the argument of Theorem 4.6 to resolve

M

by direct sums of rank one free

G

-equivariant

A / J

-modules generated by eigenelements. In view of the long exact sequences of

T o r

, it suffices to show that 7 holds true for such rank one

A / J

-module. But this is precisely a relation from the second set of relations on

x_{i}

, times a monomial in

x_{i}

to account for possible character of the action of

G

on the generator.

It remains to show that

ρ_{1} \circ ρ

is the identity. Since

B

is additively generated by monomials

\prod_{i} x_{i}^{r_{i}}

, it is enough to check this for a monomial.

Observe that the global sections of

\otimes_{i} ℒ_{i}^{r_{i}}

form a free module

A_{χ}

over

A

of rank one. Hence

T o r^{> 0} (A_{χ}, C)

are zero and

T o r^{0} (A_{χ}, C) = C

, with character the

χ

that corresponds to

\prod_{i} R_{i}^{r_{i}}

. This finishes the proof. □

5 Combinatorial Chern character

In this section we study the

K

-theory of the reduced toric DM stack

P_{Σ}

in more detail. More specifically, we show that

K_{0} (P_{Σ}, C) : = K_{0} (P_{Σ}) \otimes_{Z} C

is isomorphic as a vector space to

H_{S R} (P_{Σ}, C)

By Theorem 4.10 , we have that

K_{0} (P_{Σ}, C) \sim = B_{C} : = B \otimes_{Z} C

. Its maximum ideals correspond to points

(y_{1}, \dots, y_{n}) \in C^{n}

that satisfy

\prod_{i = 1}^{n} y_{i}^{f (v_{i})} = 1

and

\prod_{i \in I} (1 - y_{i}) = 0

, for

f

and

I

in the definition of

B

Lemma 5.1. The ring

B_{C}

is Artinian. Its maximum ideals are in one-to-one correspondence with the elements of

B o x (Σ)

as follows. A point

v = \sum_{v_{i} \in C} α_{i} v_{i}

corresponds to the

n

-tuple

(y_{1}, \dots, y_{n}) \in C^{n}

with

y_{i} = e^{2 π i α_{i}}

for

v_{i} \in C

and

y_{i} = 1

otherwise.

Proof. We need to solve for

(y_{1}, \dots, y_{n}) \in C^{n}

that satisfy

\prod_{i = 1}^{n} y_{i}^{f (v_{i})} = 1

and

\prod_{i \in I} (1 - y_{i}) = 0

as in the definition of

B

. Because of the second set of equations,

y_{i}

are equal to

1

for all

v_{i}

outside some cone

C \in Σ

We can assume that

C

is generated precisely by

v_{i}

for indices

i

with

y_{i} \neq 1

. To simplify notations, let us assume that these

v_{i}

are

v_{1}, \dots, v_{k}

for some

k \leq r k N

The first set of relations now reads

\begin{matrix} ^{k} \prod_{i = 1} y_{i}^{f (v_{i})} = 1 \end{matrix}

(8)

for any linear function

f : N \to Z

. Let

N_{1}

be the intersection of

N

and the rational span of

v_{1}, \dots, v_{k}

. It is enough to look at

f : N_{1} \to Z

By looking at some

f_{i}

which is zero for

j \in [1, \dots, k] - {i}

, we conclude that

y_{i}

is a root of

1

. We introduce

α_{i} \in [0, 1)

, such that

y_{i} = e^{2 π i α_{i}}

Then the relations 8 amount to

\sum_{i} f (v_{i}) α_{i} \in Z

for all

f \in N_{1}^{\lor}

. This is true if and only if

v = \sum_{i = 1}^{k} α_{i} v_{i} \in N_{1}

. Hence, the solutions to 8 are in one-to-one correspondence with elements of

B o x (C)

. The condition

y_{i} \neq 1

for

v_{i} \in C

assures that

v

does not lie in

B o x (C_{1})

for any proper face

C_{1}

C

Since we are looking at all possible cones

C

here, the description of maximum ideals follows. Finally, the ring

B_{C}

is Artinian, since it is Noetherian of Krull dimension zero. □

Since

B_{C}

is Artinian, it is a direct sum of Artinian local rings obtained by localizing at maximum ideals, which we denote by

(B_{C})_{v}

. We have

B_{C} = \oplus_{v \in B o x (Σ)} (B_{C})_{v}

The next lemma describes the structure of

(B_{C})_{v}

Lemma 5.2. Let

C

be the minimum cone of

Σ

that contains

v

. Then the ring

(B_{C})_{v}

is isomorphic as a

C

-algebra to the SR-cohomology with complex coefficients of the (closed) subvariety of

P_{Σ}

that corresponds to

C

Proof. To simplify notations, we assume that v = ∑ i = 1 k α i v i with α i ∈ ( 0 , 1 ) . We will also index the rest of v i in such a way that v k + 1 , … , v l are contained in some cone C 1 ⊃ C , and v l + 1 , … , v n are not.
We can localize first and then apply our relations. In fact, since B C is Artinian, we may assume to be working in the quotient of the power series ring in x i − y i by a sufficiently high power of the maximum ideal.
This makes x i − 1 nilpotent in ( B C ) v for i > k and it makes x i − e 2 π i α i nilpotent for 1 ≤ i ≤ k . We define z i = log ( x i ) : = ∑ m > 0 1 m ( x i − 1 ) m for i > k and z i = log ( x i e − 2 π i α i ) : = ∑ m > 0 1 m ( x i e − 2 π i α i − 1 ) m for i = 1 , … , k . The elements z i are also nilpotent and we can assume to be working in the quotient B 1 of C [ [ z 1 , … , z n ] ] by a sufficiently high power of the maximum ideal.
We further observe that z j = 0 in ( B C ) v for j > l . Indeed, we have ( x j − 1 ) k ∏ i = 1 ( x i − 1 ) = 0 which translates into ( e z j − 1 ) k ∏ i = 1 ( e 2 π i α i e z i − 1 ) = 0 . Since α i ∈ ( 0 , 1 ) , we see that ( e 2 π i α i e z i − 1 ) is invertible in B 1 , so e z j − 1 = 0 . This gives z j ( 1 + 1 2 z j + . . . ) = 0 , which leads to z j = 0 . As a result, we may just ignore z j for j > l in our calculations and work in the quotient B 2 of C [ [ z 1 , … , z l ] ] by a sufficiently high power of the maximum ideal.
The relations ∏ i ∈ I ( x i − 1 ) = 0 are now only nontrivial for I ⊆ [ 1 , … , l ] . Then every such set can be enlarged by adding all of [ 1 , … , i ] .
Consider the quotient fan Σ C in N / N 1 which is made from the images of the cones that contain C . Since ( x i − 1 ) is invertible for i ≤ k , the relations ∏ i ∈ I ( x i − 1 ) = 0 become relations of the form ∏ i ∈ I 1 ( e z i − 1 ) for I 1 ⊆ [ k + 1 , … , l ] such that z i , i ∈ I 1 are not contained in any cone of Σ C . As before, this is equivalent to ∏ i ∈ I 1 z i = 0 for these I 1 ⊆ [ k + 1 , … , l ] . This completely describes the second set of relations on z i .
We now need to rewrite the first set of relations in terms of z i . Let f : N → Z be a linear function. Then we have k ∏ i = 1 ( e 2 π i α i ) f ( v i ) k + l ∏ i = 1 e z i f ( v i ) − 1 = 0 . Since f ( v ) ∈ Z , this becomes simply e ∑ i = 1 k + l z i f ( v i ) − 1 = 0 , which is further equivalent to ∑ i = 1 k + l z i f ( v i ) = 0 . We can now ignore the integrality condition on f and simply look at all linear functions f : N → Z . Consider the subspace V of linear functions on N that satisfy f ( v i ) = 0 for all i ≤ k , and let V 1 be a complement of it. We can use elements of V 1 to express z 1 , … , z k as linear combinations of z k + 1 , … , z l . This will allow us to write ( B C ) v as the quotient of the ring B 3 = C [ z k + 1 , … , z l ] by a high power of a maximum ideal, and by the relations
- $∙$ $\prod_{i \in I_{1}} z_{i}$ , for $I_{1}$ such that $z_{i}, i \in I_{1}$ do not lie in a cone of $Σ_{C}$ ,
- $∙$ $\sum_{i = 1}^{k + l} z_{i} f (v_{i})$ for any linear function $f : N / N_{1} \to Q$ .
This is immediately recognized as the SR-cohomology ring of the toric subvariety of P Σ that corresponds to the cone C . □

Theorem 5.3. There is a natural vector space isomorphism between

K_{0} (P_{Σ}, C)

and

H_{S R} (P_{Σ}, C)

Proof. Follows directly from Theorem 4.10 , Lemmas 5.1 and 5.2 and Propositions 3.8 and 3.9 . □

Remark 5.4. We call the map of Theorem 5.3 combinatorial Chern character. While it is an isomorphism of vector spaces, we stress that this map is not a ring homomorphism, since

B_{C}

is semilocal and

H_{S R} (P_{Σ}, C)

is local. One motivation behind our construction is that it generalizes the Chern character for projective toric varieties (which, however, is a ring isomorphism).

Remark 5.5. If

P_{Σ}

is projective, one can use the isomorphism of SRand orbifold cohomology to construct a map

K_{0} (Σ, C) \to H_{o r b} (P_{Σ}, C)

. We suspect that this map is the Chern character map of [AR] , which also motivated our terminology. But since our technique is quite different, we found it hard to make the connection explicit.

Remark 5.6. It is an interesting question as to under what conditions on the fan the

K

-theory is torsion-free. We do not know the answer to this, even in the case of smooth toric varieties.

Remark 5.7. The additive isomorphism between

K_{0} (P_{Σ}, C)

and the SR-cohomology

H_{S R} (N, Σ, C)

indicates that the

K

-theory of a reduced toric DM stack possesses an alternative product structure that makes it into a local ring. One wonders what the geometric meaning of this structure may be. A related open question is whether there is a structure like this for the

K

-theory of any DM stack, not necessarily a toric one.

6 The $K$ -theory of nonreduced toric DM stacks

In this section we extend the calculation of

K

-theory of toric stacks to the nonreduced case. Since this is not the main focus of this paper, we only sketch the changes necessary to extend the results.

The principal feature of the nonreduced case is that

N

is now just a finitely generated abelian group, i.e. it is allowed to have torsion. A fan in

N

is a pullback of a fan in

N / t o r s i o n

. A stacky fan is defined by a choice of a nonzero element

v_{i}

in each of the one-dimensional cones of

Σ

. We still have a map

Z^{n} \to N

with finite coindex. This allows one to define a Gale dual

(Z^{n})^{\lor} \to N^{'}

which is the analog of

(Z^{n})^{\lor} \to C o k e r φ

of Section 2 , but we no longer have the surjectivity. We refer the reader to [BCS] for the details.

Consequently, the group

G = H o m (N^{'}, C)

is no longer a subgroup of

(C^{*})^{n}

, but rather maps to it with a finite kernel.

Otherwise, the definition of the open subset

Z \subseteq C^{n}

is the same as in the reduced case. This allows one to define line bundles

ℒ_{i}

P_{Σ}

Unfortunately, they will no longer generate the

K

-theory. The problem is that in the proof of Theorem 4.6 we have used that every character of

G

lifts to a character of

(C^{*})^{n}

. This is no longer the case. However, we can still look at the ring

B

which is the quotient of the character ring of

G

by the relations

\prod_{i \in I} (x_{i} - 1) = 0

for all

I \subseteq [1, \dots, n]

, such that

v_{i}, i \in I

are not contained in any cone of

Σ

. Here

x_{i}

correspond to

ℒ_{i}

as in nonreduced case. With this modification, the proofs of both Theorems 4.6 and 4.10 are extended to the nonreduced case without any major changes to show that

K_{0} (P_{Σ}) \sim = B

The description of the maximum ideals of

B_{C}

from Lemma 5.1 still holds in the nonreduced case, but the proof is a bit more complicated, especially is one does not assume the condition 4 . Specifically, these ideals correspond to elements of

G

whose action on

C^{n}

has eigenvalues one outside of the set of indices

i

for

v_{i}

in some cone

C \in Σ

. So we have a morphism

ψ : N^{'} \to C^{*}

such that the composition

(Z^{n})^{\lor} \to N^{'} \to C^{*}

takes value one on the basis elements that correspond to

v_{i}

outside of

C

. The value of

ψ

on any element of

N^{'}

is a root of one. Indeed,

ψ \in G

fixes a point in

Z

, and it has been shown in [BCS] that all isotropy subgroups of

G

-action are finite. Consequently, we can think of

ψ

as a map

N^{'} \to Q / Z

. Conversely, every such

ψ

gives rise to a group element with the above eigenvalue properties, and hence to a local subring of

B_{C}

We split

(Z^{n})^{\lor}

into

(Z^{dim C})^{\lor} \oplus (Z^{n - dim C})^{\lor}

according to whether the corresponding basis elements lie in

C

. Consider the short exact sequence of complexes

\begin{matrix} 0 \to & (Z^{n - dim C})^{\lor} & \to & (Z^{dim C})^{\lor} \oplus (Z^{n - dim C})^{\lor} & \to & (Z^{dim C})^{\lor} & \to 0 \end{matrix}

\begin{matrix} ↓ & ↓ & ↓ \end{matrix}

\begin{matrix} 0 \to & N^{'} & \to & N^{'} & \to & 0 & \to 0 \end{matrix}

where we assume that the vertical lines are extended by zeroes in both directions. Gale duality with torsion (see [BCS] ) implies that

N

can be canonically identified with the

H^{1}

of the derived

H o m (*, Z)

of the middle complex. The above exact sequence shows that

N / S p a n (v_{i} \in C)

H^{1}

of the derived

H o m ((Z^{n - dim C})^{\lor} \to N^{'}, Z)

We observe that

ψ : N^{'} \to Q / Z

constructed earlier can be identified with elements of

H^{0}

of derived

H o m ((Z^{n - dim C})^{\lor} \to N^{'}, Q / Z)

. The short exact sequence

0 \to Z \to Q \to Q / Z \to 0

yields an exact sequence

H^{0} (H o m ((Z^{n - dim C})^{\lor} \to N^{'}, Z)) \to H^{1} (H o m ((Z^{n - dim C})^{\lor} \to N^{'}, Z)) \to

\to H^{1} (H o m ((Z^{n - dim C})^{\lor} \to N^{'}, Q)) .

Since the last term of the above sequence is torsion free, the group

H^{0} (H o m ((Z^{n - dim C})^{\lor} \to N^{'}, Z))

is precisely the torsion subgroup of

N / S p a n (v_{i} \in C)

. Finally, elements of this torsion subgroup are in one-to-one correspondence with elements of

B o x (C)

. We leave to the reader to check that these identifications are compatible with embeddings

B o x (C_{1}) \subseteq B o x (C_{2})

for

C_{1} \subseteq C_{2}

and that they coincide with the construction of Section 5 in the reduced case. The SR-cohomology of

P_{Σ}

is again given as the quotient of

C [N, Σ]

by the linear relations

\sum_{i} f (v_{i}) [v_{i}]

. Note that it is no longer a local ring, because of the graded zero part is the group ring of the torsion subgroup of

N

. SR-cohomology again splits into the untwisted and twisted sectors according to elements of

B o x (Σ)

, and Propositions 3.8 and 3.9 still hold.

The combinatorial Chern character map of Theorem 5.3 still exists in the nonreduced case. It sends the local subring of the Artinian ring

B_{C}

which corresponds to the point

v \in B o x (Σ)

to the SR-cohomology of the corresponding twisted sector. More specifically, let

v

be this point and let

C

be the minimum cone of

Σ

that contains it. The map from

G

G^{'}

is a finite unramified cover, so the local ring summand for

B_{C}

is isomorphic to the local ring of a semi-local ring

B_{C}^{'}

which is obtained by cutting the character ring of

G^{'}

by the same set of relations. Then the calculation of Lemma 5.2 shows that

(B_{C})_{v} \sim = (B_{C}^{'})_{v}

is isomorphic to the SR-cohomology of the twisted sector of the reduction of

P_{Σ}

that corresponds to the reduction of

v

modulo torsion. However, this reduction does not change the coarse moduli space of the twisted sector, which finishes the construction of the isomorphism. The details are left to the reader.

7 Birational morphisms of reduced toric DM stacks

The statements of this section are implicit in [K1] , but we adjust the exposition for our notations.

Let

N

be a lattice and let

Σ = (Σ, {v_{i}})

be a stacky fan in

N

We will assume that every cone of

Σ

is contained in a cone of

Σ

of dimension

r k N

. As before, let

n

be the number of

v_{i}

and let

Z \subset C^{n}

consist of the points

z = (z_{1}, \dots, z_{n})

such that the set of

v_{i}

for the zero coordinates of

z

is contained in a cone of

Σ

. Let

G

be the subgroup of

(C^{*})^{n}

described by 3 . The toric DM stack

P_{Σ}

is defined as the stack quotient

[Z / G]

where

Z

and

G

are endowed with the natural reduced scheme structures and the action of

(λ_{i})

(z_{i})

is given by

(λ_{i} z_{i})

Let

Σ^{'} = (Σ^{'}, {v_{j}^{'}})

be another stacky fan in the same lattice

N

. Let us further assume that every cone of

Σ^{'}

is contained in a cone of

Σ

. In particular, for every

j

there is a unique way of writing

v_{j}^{'}

in the form

v_{j}^{'} = \sum_{i} α_{i, j} v_{i}, α_{i, j} \in Q

under the assumption that

α_{i, j}

are zero unless

v_{i}

lies in the minimum cone of

Σ

that contains

v_{j}^{'}

. Moreover, let us assume that all

α_{i, j}

above are integer.

Definition 7.1. Under the above assumptions, there is a morphism

P_{Σ^{'}} \to P_{Σ}

defined as follows. Define a homomorphism

(C^{*})^{n^{'}} \to (C^{*})^{n}

which sends

(λ_{j}^{'})

(\prod_{j} (λ_{j}^{'})^{α_{i, j}})

. It is easy to see that this homomorphism sends points of

G^{'}

to points of

G

. We also consider the map from

C^{n^{'}}

C^{n}

which sends

(z_{j}^{'})

(\prod_{j} (z_{j}^{'})^{α_{i, j}})

This map sends points in

Z^{'}

to points in

Z

. Indeed, if the

C^{'}

is a cone of

Σ^{'}

that contains all

v_{j}^{'}

for which

z_{j}^{'} = 0

, then

\prod_{j} (z_{j}^{'})^{α_{i, j}}

is nonzero unless

v_{i}

lies in the minimum cone of

Σ

that contains

C^{'}

. The map

Z^{'} \to Z

is compatible with the map of groups

G^{'} \to G

and their actions. This gives a map of quotient stacks, since one has the map between the corresponding groupoids.

Remark 7.2. We call the toric morphisms of Definition 7.1 birational, because they induce isomorphisms on the big strata

(C^{*})^{r k N}

in both stacks.

8 $K$ -theory pullbacks for birational morphisms

Let

μ : P_{Σ^{'}, {v_{j}^{'}}} \to P_{Σ, {v_{i}}}

be a toric birational morphism of toric DM stacks. Let

R_{i}

be defined as the elements of

K_{0} (P_{Σ, {v_{i}}}, Q)

that correspond to the invertible sheaves that correspond to

v_{i}

, and similarly for

R_{j}^{'}

. There is a pullback map

μ^{*} : K_{0} (P_{Σ, {v_{i}}}, Q) \to K_{0} (P_{Σ^{'}, {v_{j}^{'}}}, Q)

defined by pulling back the coherent sheaves. As we saw earlier in Theorem 4.10 , the elements

\prod_{i} R_{i}^{r_{i}}

span

K_{0} (P_{Σ, {v_{i}}}, Q)

Proposition 8.1. The pullback map

μ^{*}

is given by

μ^{*} \prod_{i} R_{i}^{r_{i}} = \prod_{j} (R_{j}^{'})^{\sum_{i} α_{i, j} r_{i}}

Proof. The category of coherent sheaves on $P_{Σ, {v_{i}}}$ is equivalent to that of $G$ -linearized coherent sheaves on $Z$ , see [V,Example7.21] . Under this equivalence, $\prod_{i} R_{i}^{r_{i}}$ corresponds to the trivial sheaf on $Z$ , linearized by $O_{Z} \to g^{*} O_{Z} = O_{Z}$ which send $1$ to $\prod_{i} (λ_{i})^{r_{i}}$ . These isomorphisms pull back to the isomorphisms of $O_{Z^{'}}$ that, for given $(λ_{1}^{'}, \dots, λ_{n^{'}}^{'})$ , send $1$ to $\prod_{i, j} ((λ_{j}^{'})^{α_{i, j}})^{r_{i}}$ on $Z^{'}$ . Since the power of $λ_{j}^{'}$ is $\sum_{i} α_{i, j} r_{i}$ , the result follows. □

Remark 8.2. It is an amusing, though unnecessary, combinatorial exercise to see that the relations among

\prod_{i} R_{i}^{r_{i}}

get mapped in the ideal of the relations among

\prod_{j} (R_{j}^{'})^{r_{j}^{'}}

9 Weighted blowups and pushforward formulas

Let

(Σ, {v_{i}})

be a stacky fan in lattice

N

. Let

C

be a cone in

Σ

of dimension

d > 1

and let

{v_{1}, \dots v_{d}}

be the set of

v_{i} \in C

. Let

h_{1}, \dots, h_{d}

be some positive integers. Consider

v_{0}^{'} : = \sum_{i = 1}^{d} h_{i} v_{i} .

The cone

C

decomposes into a union of

d

cones of dimension

d

which are given by positive linear combinations of

v_{0}^{'}

and all but one of

v_{i}, 1 \leq i \leq d

We define a stacky fan

(Σ^{'}, {v_{j}^{'}})

as follows. We add an extra ray which corresponds to

v_{0}^{'}

and keep the rest of

v_{i}

unchanged. Thus

n^{'} = n + 1

and

v_{i}^{'} = v_{i}

for

i > 0

. Every cone

\hat{C}

Σ

that does not contain

C

is still a cone of

Σ^{'}

. Every cone

\hat{C} \supseteq C

is subdivided into

d

cones

C_{1}^{'}, \dots, C_{d}^{'}

according to the above decomposition of

C

Our definitions assure that there is a birational morphism

μ : P_{Σ^{'}, {v_{j}^{'}}} \to P_{Σ, {v_{i}}}

which we call the weighted blowdown morphism. We will also denote by

π

the corresponding morphism

Z^{'} \to Z

. Our goal is to calculate the

K

-theory pushforward of

μ

. It is best described in terms of generating functions.

Theorem 9.1. Let

R = R_{0}^{'}

be the

K

-theory class of the invertible sheaf that corresponds to the extra ray

v_{0}^{'}

. There holds

μ_{*} (\frac{1}{1 - R^{- 1} t}) = \frac{1}{1 - t} - \frac{t}{1 - t} \prod_{i} \frac{1 - R_{i}^{- 1}}{1 - R_{i}^{- 1} t^{h_{i}}}

which should be interpreted as an identity of formal power series in

t

with values in

K

-theory.

Proof. We will again identify the abelian categories of sheaves of $P_{Σ^{'}, {v_{j}^{'}}}$ and $P_{Σ, {v_{i}}}$ with the categories of $G^{'}$ and $G$ -linearized sheaves on $Z^{'}$ and $Z$ respectively. The $K$ -theory pushforward of a sheaf $F^{'}$ is defined as the $K$ -theory image of the alternating sum of the higher direct images of $F^{'}$ under $μ$ . In order to describe the higher direct images of $μ$ , we need to describe the direct image functor for $μ$ in terms of $π : Z^{'} \to Z$ . We first observe that the kernel $H$ of $G^{'} \to G$ is isomorphic to $C^{*}$ .
Indeed, the map comes from $(C^{*})^{n + 1} \to (C^{*})^{n}$ defined by $(λ_{0}^{'}, λ_{1}^{'}, \dots, λ_{n}^{'}) \mapsto (λ_{1}^{'} (λ_{0}^{'})^{h_{1}}, \dots, λ_{d}^{'} (λ_{0}^{'})^{h_{d}}, λ_{d + 1}^{'}, \dots, λ_{n}^{'}) .$ The kernel is given by $λ_{0}^{'} = λ$ , $λ_{i}^{'} = λ^{- h_{i}}$ for $1 \leq i \leq d$ , $λ_{> d}^{'} = 1$ , which clearly lies in $G^{'}$ . Moreover, the map $G^{'} \to G$ is split surjective.
Indeed, given $(λ_{i}) \in G$ , the collection $(1, λ_{1}, \dots, λ_{n})$ will lie in $G^{'}$ , which produces the splitting. We will only need surjectivity for our arguments.
For every $G^{'}$ -linearized coherent sheaf $ℱ^{'}$ on $Z^{'}$ , its pushforward $ℱ = π_{*} ℱ^{'}$ on $Z$ inherits the $G^{'}$ -linearization under the action of $G^{'}$ on $Z$ induced from $G^{'} \to G$ . Its subsheaf $ℱ^{H}$ of $H$ -invariants is therefore given the structure of a $G$ -linearized sheaf on $Z$ . The $K$ -theory pushforward of a sheaf on $P_{Σ^{'}, {v_{j}^{'}}}$ that corresponds to $ℱ^{'}$ is given by $ℱ^{H}$ . Indeed, consider the following commutative diagram
$\begin{matrix} Z^{'} & \overset{π}{\to} & Z \end{matrix}$

$\begin{matrix} q^{'} ↓ & ↓ q \end{matrix}$

$\begin{matrix} [Z^{'} / G^{'}] & \overset{μ}{\to} & [Z / G] \end{matrix}$
The sheaf on $[Z^{'} / G^{'}]$ that corresponds to $ℱ^{'}$ is the subsheaf of $G^{'}$ -invariants of $q_{*}^{'} ℱ^{'}$ . Consequently, its direct image in $[Z / G]$ is the subsheaf of $G^{'}$ -invariants of $q_{*} π_{*} ℱ^{'} = q_{*} ℱ$ . The subgroup $H$ of $G^{'}$ acts trivially on $Z$ , so the $G^{'}$ -invariants of $q_{*} ℱ$ are the $G$ -invariants of $q_{*} ℱ^{H}$ . This corresponds to the $G$ -equivariant sheaf $ℱ^{H}$ on $Z$ .
We have thus described the direct image functor in terms of the composition of the direct image functor $π_{*}$ from $C o h_{G^{'}} (Z^{'})$ to $C o h_{G^{'}} (Z)$ and the functor of $H$ -invariants from $C o h_{G^{'}} (Z)$ to $C o h_{G} (Z)$ . The latter is exact, therefore, the higher direct images of $μ$ are given by $(R^{k} π_{*} (ℱ^{'}))^{H}$ .
In order to prove the theorem, we need to calculate the higher direct images of the $G^{'}$ -linearized invertible sheaf $ℒ$ on $Z^{'}$ with the linearization that sends $1$ to $(λ_{0}^{'})^{- l}$ for some integer $l \geq 0$ . We claim that $(R^{> 0} π_{*} ℒ)^{H} = 0 .$ It is sufficient to check the statement at a fiber. Let $z = (z_{1}, \dots, z_{n})$ be a point in $Z$ . The fiber of $π$ consists of points $(z_{0}^{'}, z_{1}^{'}, \dots, z_{n}^{'}) \in Z^{'}$ such that $z = (z_{1}^{'} (z_{0}^{'})^{h_{1}}, \dots, z_{d}^{'} (z_{0}^{'})^{h_{d}}, z_{d + 1}^{'}, \dots, z_{n}^{'}) .$ Hence if one or more of $z_{1}, \dots, z_{d}$ are nonzero, then the fiber is isomorphic to $C^{*}$ , which is an orbit of $H$ . The structure sheaf of $C^{*}$ has no higher cohomology, so $R^{> 0} π_{*} ℒ$ will have zero fibers at such $z$ , and $R^{> 0} π_{*} ℒ$ are supported over $z_{1} = \dots = z_{d} = 0$ . For these $z$ , the fiber is described by $z_{0}^{'} = 0$ , $z_{i}^{'} = z_{i}$ for $i > d$ , and $(z_{1}^{'}, \dots, z_{d}^{'}) \neq (0, \dots, 0)$ . The fiber is isomorphic to $C^{d} - 0$ , with the group $H$ acting by multiplications of the $i$ -th coordinate by $λ^{- h_{i}}$ . The cohomology of the structure sheaf $O$ on $C^{d} - 0$ occurs at $H^{0}$ and $H^{d - 1}$ only, as can be easily calculated by the Čech complex for the covering by the open sets $(z_{i}^{'} \neq 0)$ . Moreover, there is a natural isomorphism $H^{d - 1} (C^{d} - 0, O) \sim = (\prod_{i} (z_{i}^{'})^{- 1}) C [(z_{1}^{'})^{- 1}, \dots, (z_{d}^{'})^{- 1}] .$ This means that the action of $λ \in H$ multiplies monomial generators of $H^{d - 1} (C^{d} - 0, O)$ by $λ^{- \sum_{i} h_{i} s_{i}}$ for some $s_{i} < 0$ . Consequently, it multiplies generators of $H^{d - 1} (C^{d} - 0, ℒ)$ by $λ^{l - \sum_{i} h_{i} s_{i}}$ , which shows that the space of $H$ -invariants is zero. Of course, the above is basically a calculation of cohomology of a line bundle on a weighted projective space.
As a result, to calculate the $K$ -theory pushforward of $R^{- l}$ for $l \geq 0$ it is enough to calculate the direct image of the corresponding invertible sheaf. In the notations of the preceding paragraph, $ℒ$ is naturally embedded into $O$ as an ideal sheaf of $(z_{0}^{'})^{l}$ . As a result, $(π_{*} ℒ)^{H}$ is embedded into $(π_{*} O)^{H}$ . We first show that the latter is isomorphic to $O_{Z}$ , which is just the statement that $μ_{*} O = O$ for a birational morphism.
The isomorphism will be glued from the isomorphisms on $(C^{*})^{n}$ -invariant affine subsets $U_{σ} \subseteq Z$ which correspond to the maximum-dimensional cones $σ$ of $Σ$ . The subset $U_{σ}$ is defined by the condition that $z_{i} \neq 0$ for all $v_{i} \notin σ$ and is isomorphic to $(C^{*})^{n - r k N} \times C^{r k N}$ . If $σ ⊉ C$ , then the preimage $π^{- 1} U_{σ}$ in $Z^{'}$ is given by the conditions $z_{0}^{'} \neq 0$ and $z_{i}^{'} \neq 0$ for $i > 0$ and $v_{i} \notin σ$ . Then $π^{- 1} U_{σ}$ is isomorphic to $(C^{*})^{n + 1 - r k N} \times C^{r k N}$ . The sections of $π_{*} O_{Z^{'}}$ on $U_{σ}$ are spanned by the monomials $(z_{0}^{'})^{s_{0}} \prod_{i} (z_{i}^{'})^{s_{i}}$ with $s_{i} \geq 0$ for all $v_{i} \in σ$ . The $H$ -invariant sections in addition satisfy $s_{0} = \sum_{i = 1}^{d} h_{i} s_{i}$ . The isomorphism from $O_{U_{σ}}$ to $(π_{*} O_{Z^{'}})^{H}$ is constructed by sending $\prod_{i = 1}^{n} z_{i}^{s_{i}}$ to $(z_{0}^{'})^{\sum h_{i} s_{i}} \prod_{i = 1}^{n} (z_{i}^{'})^{s_{i}}$ . In the case when $σ \supseteq C$ , the preimage $π^{- 1} U_{σ}$ is given by the conditions $z_{i}^{'} \neq 0$ for $v_{i} \notin σ$ , and $(z_{1}^{'}, \dots, z_{d}^{'}) \neq 0$ . Consequently, it is isomorphic to $C \times (C^{d} - 0) \times (C^{*})^{n - r k N} \times C^{r k N - d}$ . Because $d \geq 2$ , the sections of $O$ on this space ignore the deletion of $0$ . The sections of $π_{*} O_{Z^{'}}$ are over $U_{σ}$ are therefore spanned by monomials $(z_{0}^{'})^{s_{0}} \prod_{i} (z_{i}^{'})^{s_{i}}$ with $s_{0} \geq 0$ and $s_{i} \geq 0$ for $v_{i} \in σ$ . The $H$ -invariance condition simply expresses $s_{0}$ in terms of other $s_{i}$ but imposes no further restrictions on $s_{1}, \dots, s_{n}$ . Consequently, we again have an isomorphism between $O_{Z}$ and $(π_{*} O_{Z^{'}})^{H}$ . It is clear that these isomorphisms are compatible on the intersections and hence glue together to show $(π_{*} O_{Z^{'}})^{H} \sim = O_{Z}$ .
Obviously, the sheaf $O_{Z}$ is a restriction of $O_{C^{n}}$ via the open embedding $Z \subseteq C^{n}$ . We claim that $(π_{*} ℒ)^{H}$ is also a restriction of an ideal sheaf $ℐ$ from $C^{n}$ . Namely, consider the ideal sheaf $ℐ$ on $C^{N}$ which corresponds to the submodule over $C [z_{1}, \dots, z_{n}]$ which is the span of monomials $\prod_{i = 1}^{n} z_{i}^{s_{i}}$ with $\sum_{i = 1}^{d} s_{i} h_{i} \geq l .$ To show that it restricts to $(π_{*} ℒ)^{H}$ on $Z$ , it is again enough to calculate the sections over the open subsets $U_{σ}$ . For $σ ⊉ C$ , we have $O_{Z} |_{U_{σ}} \sim = ℐ |_{U_{σ}}$ . For $σ \supseteq C$ , the condition on $(z_{0}^{'})^{s_{0}} \prod_{i} (z_{i}^{'})^{s_{i}}$ to lie in the ideal of $(z_{0}^{'})^{l}$ translates into $s_{0} \geq l$ . The invariants of that come from the monomials $\prod_{i} z_{i}^{s_{i}}$ with $\sum_{i = 1}^{d} s_{i} h_{i} \geq l$ , as claimed.
To calculate the pushforward of $ℒ$ we now simply need to calculate the free graded resolution of the ideal $I$ of the polynomial ring $C [z_{1}, \dots, z_{n}]$ which is the span of the monomials with the condition $\prod_{i} z_{i}^{s_{i}}$ with $\sum_{i = 1}^{d} s_{i} h_{i} \geq l$ . Clearly, the variables $z_{d + 1}, \dots, z_{n}$ can be ignored. If $0 \to F^{d} \to F^{d - 1} \to \dots \to F^{0} \to I \to 0$ is such a free resolution, then the alternating sum of the $Z^{d}$ -graded dimensions of $F^{k}$ is the $Z^{d}$ -graded dimension of $I$ , and the same is true for their generating functions. The generating function of a copy $A$ of $C [z_{1}, \dots, z_{n}]$ with the grading shifted so that the multidegree of $1$ is $(r_{1}, \dots, r_{d})$ is $\sum_{d e g \in Z^{d}} {dim}_{C} (A_{d e g}) t^{d e g} =^{d} \prod_{i = 1} \frac{t_{i}^{r_{i}}}{1 - t_{i}} .$ On the the other hand, such a module $A$ gives rise to the invertible sheaf on $Z$ that gives $\prod_{i = 1}^{d} R_{i}^{- r_{i}}$ in $K$ -theory of $P_{Σ, {v_{i}}}$ . So the $K$ -theory pushforward of $R^{- l}$ is given by $μ_{*} R^{- l} =^{d} \prod_{i = 1} (1 - R_{i}^{- 1}) \sum_{(s_{1}, \dots, s_{d}) \in Z_{\geq 0}^{d}, h_{1} s_{1} + \dots + h_{d} s_{d} \geq l} R_{1}^{- s_{1}} \cdot \dots \cdot R_{d}^{- s_{d}}$ where the sum should be interpreted as a formal power series in $R_{i}^{- 1}$ which actually gives a polynomial after being multiplied by $\prod (1 - R_{i}^{- 1})$ .
It remains to observe that if we look at the above in terms of the generating functions for all $l \geq 0$ , then $\sum_{l \geq 0} t^{l} μ_{*} R^{- l} =^{d} \prod_{i = 1} (1 - R_{i}^{- 1}) \sum_{l \geq 0} t^{l} {\sum_{(s_{1}, \dots, s_{d}) \in Z_{\geq 0}^{d}, \sum_{i = 1}^{d} h_{i} s_{i} \geq l}}^{d} \prod_{i = 1} R_{i}^{- s_{i}}$ $=^{d} \prod_{i = 1} (1 - R_{i}^{- 1}) {\sum_{(s_{1}, \dots, s_{d}) \in Z_{\geq 0}^{d}}}^{d} \prod_{i = 1} R_{i}^{- s_{i}} \sum_{0 \leq l \leq \sum_{i = 1}^{d} h_{i} s_{i}} t^{l}$ $=^{d} \prod_{i = 1} (1 - R_{i}^{- 1}) \sum_{(s_{1}, \dots, s_{d}) \in Z_{\geq 0}^{d}} {\frac{1 - t^{1 + \sum_{i} h_{i} s_{i}}}{1 - t}}^{d} \prod_{i = 1} R_{i}^{- s_{i}}$ $= \frac{1}{1 - t} - {\frac{t}{1 - t}}^{d} \prod_{i = 1} (1 - R_{i}^{- 1}) \sum_{(s_{1}, \dots, s_{d}) \in Z_{\geq 0}^{d}} {t^{\sum_{i} h_{i} s_{i}}}^{d} \prod_{i = 1} R_{i}^{- s_{i}}$ $= \frac{1}{1 - t} - {\frac{t}{1 - t}}^{d} \prod_{i = 1} \frac{1 - R_{i}^{- 1}}{1 - R_{i}^{- 1} t^{h_{i}}} .$ We remark that the above calculations should be interpreted as calculations in formal power series in $t$ and $R_{i}^{- 1}$ with only finitely many terms at any given degree, so convergence is never an issue. This gives the desired formula for $μ_{*} \frac{1}{1 - R^{- 1} t}$ . □

Remark 9.2. The above theorem allows one to calculate the pushforward of any element of

K

-theory in view of the description of the pullback in Proposition 8.1 and the pull-push formula. Indeed, every element of

K_{0} (P_{Σ^{'}, {v_{i}^{'}}})

can be written as a polynomial in

R

and

R^{- 1}

with coefficients in

μ^{*} K_{0} (P_{Σ, {v_{i}}})

Since

R

is quasi-unipotent, it can be expressed in terms of negative powers of

R

So far we have considered the weighted blowups of a stratum of codimension

d > 1

. While blowups in codimension one are isomorphisms in the smooth variety case, this is no longer true for weighted blowups of stacks. Namely, let

(Σ, {v_{i}})

be a stacky fan in

N

. Let

C

be a dimension one cone of

Σ

and let

v_{1}

be the chosen lattice point on

C

For a positive integer

k

consider the stacky fan

(Σ^{'}, {v_{i}^{'}})

defined by

Σ^{'} = Σ

v_{1}^{'} = k v_{1}

v_{i}^{'} = v_{i}, i > 1

. There is a birational morphism

μ : P_{Σ^{'}, {v_{i}^{'}}} \to P_{Σ, {v_{i}}}

which we again call weighted blowdown morphism. To complete the discussion, we calculate the corresponding pushforward in

K

-theory.

Since the pullback is given by

μ^{*} R_{1} = R_{1}^{k}

μ^{*} R_{i} = R_{i}^{'}

for

i > 1

, it is enough to calculate the pushforward of

R_{1}^{m}

for

m = 1, \dots, k

Proposition 9.3. For

1 \leq m \leq k

we have

μ_{*} (R_{1}^{'})^{- m} = R_{1}^{- 1}

Proof. The map $G^{'} \to G$ is given by $(λ_{1}^{'}, λ_{2}^{'}, \dots, λ_{n}^{'}) \to (λ_{k}^{'}, λ_{2}^{'}, \dots, λ_{n}^{'})$ and the map $π : Z^{'} \to Z$ is given by $π : (z_{1}, z_{2}, \dots, z_{n}) \mapsto (z_{1}^{k}, z_{2}, \dots, z_{n}) .$ We denote by $H$ the kernel of $G^{'} \to G$ . Since $π$ is finite, its higher direct images vanish, and we only need to calculate $(π_{*} ℒ)^{H}$ for the sheaf $ℒ$ which is isomorphic to $O_{Z^{'}}$ with linearization given by multiplication by $(λ_{1}^{'})^{- m}$ . We can think of $ℒ$ as an ideal sheaf of $O_{Z^{'}}$ generated by $(z_{1}^{'})^{m}$ .
Similar to the proof of Theorem 4.10 , $(π_{*} ℒ)^{H}$ is an ideal sheaf in $O_{Z}$ . It is induced from a sheaf on $C^{n} \supset Z$ . A monomial $\prod_{i} z_{i}^{s_{i}}$ lies in the corresponding ideal of $C [z_{1}, \dots, z_{n}]$ if and only if $s_{1} k \geq m$ . Since $1 \leq m \leq k$ , this is equivalent to $s_{1} \geq 1$ . This ideal sheaf is then identified with $ℒ_{1}^{- 1}$ . □

We rewrite the result of Proposition 9.3 to resemble that of Theorem 9.1 . We denote

R = R_{1}^{'}

Corollary 9.4. For the blowdown of codimension one,

μ_{*} (\frac{1}{1 - R^{- 1} t}) = \frac{1}{1 - t} - \frac{t}{(1 - t)} \frac{(1 - R_{1}^{- 1})}{(1 - R_{1}^{- 1} t^{k})} .

Proof. $μ_{*} (\frac{1}{1 - R^{- 1} t}) = 1 + \sum_{l > 0} μ_{*} R^{- l} t^{l} = 1 + \sum_{l \geq 0} \sum_{m = 1}^{k} μ_{*} R^{- l k - m} t^{l k + m}$ $= 1 + \sum_{l \geq 0} \sum_{m = 1}^{k} μ_{*} (R^{- m} μ^{*} R_{1}^{- l}) t^{l k + m} = 1 + \sum_{l \geq 0} \sum_{m = 1}^{k} R_{1}^{- l - 1} t^{l k + m}$ $= 1 + \frac{R_{1}^{- 1} (t - t^{k + 1})}{(1 - R_{1}^{- 1} t^{k}) (1 - t)} = \frac{1}{1 - t} - \frac{t}{(1 - t)} \frac{(1 - R_{1}^{- 1})}{(1 - R_{1}^{- 1} t^{k})} .$ □

References

D. Abramovich, T. Graber, A. Vistoli, Algebraic orbifold quantum products, Orbifolds in mathematics and physics (Madison, WI, 2001), 1-24, Contemp. Math., 310, Amer. Math. Soc., Providence, RI, 2002.
A. Adem, Y. Ruan, Twisted orbifold $K$ -theory, Comm. Math. Phys. 237 (2003), no. 3, 533–556.
S. Baggio, Equivariant K-Theory of Smooth Toric Varieties, preprint math.AG/0410378.
L. Borisov, L. Chen, G. Smith, The orbifold Chow ring of toric Deligne-Mumford stacks, preprint math.AG/0309229, to appear in J. Amer. Math. Soc.
L. Borisov, R. P. Horja, GKZ systems, toric DM stacks and mirror symmetry, in preparation.
L. Borisov, A. Mavlyutov, String cohomology of Calabi-Yau hypersurfaces via mirror symmetry, Adv. Math. 180 (2003), no. 1, 355–390.
W. Chen, Y. Ruan, Orbifold Gromov-Witten theory, Orbifolds in mathematics and physics (Madison, WI, 2001), Contemporary Mathematics 310, American Mathematical Society, Providence, RI, 2002, 25–85.
D. Eisenbud, Commutative algebra with a view toward algebraic geometry, Graduate Texts in Mathematics, 150. Springer-Verlag, New York, 1995.
W. Fulton, Introduction to toric varieties, Annals of Mathematics Studies, 131. The William H. Roever Lectures in Geometry. Princeton University Press, Princeton, NJ, 1993.
R. Hartshorne, Algebraic geometry, Graduate Texts in Mathematics, No. 52. Springer-Verlag, New York-Heidelberg, 1977.
T. J. Jarvis, R. Kaufmann, T. Kimura, Stringy K-theory and the Chern character, preprint math.AG/0502280.
Y. Kawamata, Log Crepant Birational Maps and Derived Categories, preprint math.AG/0311139.
Y. Kawamata, Derived Categories of Toric Varieties, preprint math.AG/0503102.
G. Mumford, J. Fogarty, Geometric invariant theory. Second edition, Ergebnisse der Mathematik und ihrer Grenzgebiete [Results in Mathematics and Related Areas], 34. Springer-Verlag, Berlin, 1982.
A. Vistoli, Intersection theory on algebraic stacks and on their moduli spaces, Invent. Math. 97 (1989), no. 3, 613–670.

Department of Mathematics, University of Wisconsin, Madison, WI, 53706, USA, borisov@math.wisc.edu, The Fields Institute, 222 College St, Toronto, Ontario, M5T 3J1, Canada, horja@fields.utoronto.ca.